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++)
57 blk_exit_rl(&blkg->rl);
62 * blkg_alloc - allocate a blkg
63 * @blkcg: block cgroup the new blkg is associated with
64 * @q: request_queue the new blkg is associated with
65 * @gfp_mask: allocation mask to use
67 * Allocate a new blkg assocating @blkcg and @q.
69 static struct blkcg_gq *blkg_alloc(struct blkcg *blkcg, struct request_queue *q,
72 struct blkcg_gq *blkg;
75 /* alloc and init base part */
76 blkg = kzalloc_node(sizeof(*blkg), gfp_mask, q->node);
81 INIT_LIST_HEAD(&blkg->q_node);
85 /* root blkg uses @q->root_rl, init rl only for !root blkgs */
86 if (blkcg != &blkcg_root) {
87 if (blk_init_rl(&blkg->rl, q, gfp_mask))
92 for (i = 0; i < BLKCG_MAX_POLS; i++) {
93 struct blkcg_policy *pol = blkcg_policy[i];
94 struct blkg_policy_data *pd;
96 if (!blkcg_policy_enabled(q, pol))
99 /* alloc per-policy data and attach it to blkg */
100 pd = kzalloc_node(pol->pd_size, gfp_mask, q->node);
117 * __blkg_lookup - internal version of blkg_lookup()
118 * @blkcg: blkcg of interest
119 * @q: request_queue of interest
120 * @update_hint: whether to update lookup hint with the result or not
122 * This is internal version and shouldn't be used by policy
123 * implementations. Looks up blkgs for the @blkcg - @q pair regardless of
124 * @q's bypass state. If @update_hint is %true, the caller should be
125 * holding @q->queue_lock and lookup hint is updated on success.
127 struct blkcg_gq *__blkg_lookup(struct blkcg *blkcg, struct request_queue *q,
130 struct blkcg_gq *blkg;
132 blkg = rcu_dereference(blkcg->blkg_hint);
133 if (blkg && blkg->q == q)
137 * Hint didn't match. Look up from the radix tree. Note that the
138 * hint can only be updated under queue_lock as otherwise @blkg
139 * could have already been removed from blkg_tree. The caller is
140 * responsible for grabbing queue_lock if @update_hint.
142 blkg = radix_tree_lookup(&blkcg->blkg_tree, q->id);
143 if (blkg && blkg->q == q) {
145 lockdep_assert_held(q->queue_lock);
146 rcu_assign_pointer(blkcg->blkg_hint, blkg);
155 * blkg_lookup - lookup blkg for the specified blkcg - q pair
156 * @blkcg: blkcg of interest
157 * @q: request_queue of interest
159 * Lookup blkg for the @blkcg - @q pair. This function should be called
160 * under RCU read lock and is guaranteed to return %NULL if @q is bypassing
161 * - see blk_queue_bypass_start() for details.
163 struct blkcg_gq *blkg_lookup(struct blkcg *blkcg, struct request_queue *q)
165 WARN_ON_ONCE(!rcu_read_lock_held());
167 if (unlikely(blk_queue_bypass(q)))
169 return __blkg_lookup(blkcg, q, false);
171 EXPORT_SYMBOL_GPL(blkg_lookup);
174 * If @new_blkg is %NULL, this function tries to allocate a new one as
175 * necessary using %GFP_ATOMIC. @new_blkg is always consumed on return.
177 static struct blkcg_gq *blkg_create(struct blkcg *blkcg,
178 struct request_queue *q,
179 struct blkcg_gq *new_blkg)
181 struct blkcg_gq *blkg;
184 WARN_ON_ONCE(!rcu_read_lock_held());
185 lockdep_assert_held(q->queue_lock);
187 /* blkg holds a reference to blkcg */
188 if (!css_tryget(&blkcg->css)) {
195 new_blkg = blkg_alloc(blkcg, q, GFP_ATOMIC);
196 if (unlikely(!new_blkg)) {
204 if (blkcg_parent(blkcg)) {
205 blkg->parent = __blkg_lookup(blkcg_parent(blkcg), q, false);
206 if (WARN_ON_ONCE(!blkg->parent)) {
210 blkg_get(blkg->parent);
213 /* invoke per-policy init */
214 for (i = 0; i < BLKCG_MAX_POLS; i++) {
215 struct blkcg_policy *pol = blkcg_policy[i];
217 if (blkg->pd[i] && pol->pd_init_fn)
218 pol->pd_init_fn(blkg);
222 spin_lock(&blkcg->lock);
223 ret = radix_tree_insert(&blkcg->blkg_tree, q->id, blkg);
225 hlist_add_head_rcu(&blkg->blkcg_node, &blkcg->blkg_list);
226 list_add(&blkg->q_node, &q->blkg_list);
228 for (i = 0; i < BLKCG_MAX_POLS; i++) {
229 struct blkcg_policy *pol = blkcg_policy[i];
231 if (blkg->pd[i] && pol->pd_online_fn)
232 pol->pd_online_fn(blkg);
236 spin_unlock(&blkcg->lock);
241 /* @blkg failed fully initialized, use the usual release path */
246 css_put(&blkcg->css);
253 * blkg_lookup_create - lookup blkg, try to create one if not there
254 * @blkcg: blkcg of interest
255 * @q: request_queue of interest
257 * Lookup blkg for the @blkcg - @q pair. If it doesn't exist, try to
258 * create one. blkg creation is performed recursively from blkcg_root such
259 * that all non-root blkg's have access to the parent blkg. This function
260 * should be called under RCU read lock and @q->queue_lock.
262 * Returns pointer to the looked up or created blkg on success, ERR_PTR()
263 * value on error. If @q is dead, returns ERR_PTR(-EINVAL). If @q is not
264 * dead and bypassing, returns ERR_PTR(-EBUSY).
266 struct blkcg_gq *blkg_lookup_create(struct blkcg *blkcg,
267 struct request_queue *q)
269 struct blkcg_gq *blkg;
271 WARN_ON_ONCE(!rcu_read_lock_held());
272 lockdep_assert_held(q->queue_lock);
275 * This could be the first entry point of blkcg implementation and
276 * we shouldn't allow anything to go through for a bypassing queue.
278 if (unlikely(blk_queue_bypass(q)))
279 return ERR_PTR(blk_queue_dying(q) ? -EINVAL : -EBUSY);
281 blkg = __blkg_lookup(blkcg, q, true);
286 * Create blkgs walking down from blkcg_root to @blkcg, so that all
287 * non-root blkgs have access to their parents.
290 struct blkcg *pos = blkcg;
291 struct blkcg *parent = blkcg_parent(blkcg);
293 while (parent && !__blkg_lookup(parent, q, false)) {
295 parent = blkcg_parent(parent);
298 blkg = blkg_create(pos, q, NULL);
299 if (pos == blkcg || IS_ERR(blkg))
303 EXPORT_SYMBOL_GPL(blkg_lookup_create);
305 static void blkg_destroy(struct blkcg_gq *blkg)
307 struct blkcg *blkcg = blkg->blkcg;
310 lockdep_assert_held(blkg->q->queue_lock);
311 lockdep_assert_held(&blkcg->lock);
313 /* Something wrong if we are trying to remove same group twice */
314 WARN_ON_ONCE(list_empty(&blkg->q_node));
315 WARN_ON_ONCE(hlist_unhashed(&blkg->blkcg_node));
317 for (i = 0; i < BLKCG_MAX_POLS; i++) {
318 struct blkcg_policy *pol = blkcg_policy[i];
320 if (blkg->pd[i] && pol->pd_offline_fn)
321 pol->pd_offline_fn(blkg);
323 blkg->online = false;
325 radix_tree_delete(&blkcg->blkg_tree, blkg->q->id);
326 list_del_init(&blkg->q_node);
327 hlist_del_init_rcu(&blkg->blkcg_node);
330 * Both setting lookup hint to and clearing it from @blkg are done
331 * under queue_lock. If it's not pointing to @blkg now, it never
332 * will. Hint assignment itself can race safely.
334 if (rcu_dereference_raw(blkcg->blkg_hint) == blkg)
335 rcu_assign_pointer(blkcg->blkg_hint, NULL);
338 * Put the reference taken at the time of creation so that when all
339 * queues are gone, group can be destroyed.
345 * blkg_destroy_all - destroy all blkgs associated with a request_queue
346 * @q: request_queue of interest
348 * Destroy all blkgs associated with @q.
350 static void blkg_destroy_all(struct request_queue *q)
352 struct blkcg_gq *blkg, *n;
354 lockdep_assert_held(q->queue_lock);
356 list_for_each_entry_safe(blkg, n, &q->blkg_list, q_node) {
357 struct blkcg *blkcg = blkg->blkcg;
359 spin_lock(&blkcg->lock);
361 spin_unlock(&blkcg->lock);
365 * root blkg is destroyed. Just clear the pointer since
366 * root_rl does not take reference on root blkg.
369 q->root_rl.blkg = NULL;
372 static void blkg_rcu_free(struct rcu_head *rcu_head)
374 blkg_free(container_of(rcu_head, struct blkcg_gq, rcu_head));
377 void __blkg_release(struct blkcg_gq *blkg)
381 /* tell policies that this one is being freed */
382 for (i = 0; i < BLKCG_MAX_POLS; i++) {
383 struct blkcg_policy *pol = blkcg_policy[i];
385 if (blkg->pd[i] && pol->pd_exit_fn)
386 pol->pd_exit_fn(blkg);
389 /* release the blkcg and parent blkg refs this blkg has been holding */
390 css_put(&blkg->blkcg->css);
392 blkg_put(blkg->parent);
395 * A group is freed in rcu manner. But having an rcu lock does not
396 * mean that one can access all the fields of blkg and assume these
397 * are valid. For example, don't try to follow throtl_data and
398 * request queue links.
400 * Having a reference to blkg under an rcu allows acess to only
401 * values local to groups like group stats and group rate limits
403 call_rcu(&blkg->rcu_head, blkg_rcu_free);
405 EXPORT_SYMBOL_GPL(__blkg_release);
408 * The next function used by blk_queue_for_each_rl(). It's a bit tricky
409 * because the root blkg uses @q->root_rl instead of its own rl.
411 struct request_list *__blk_queue_next_rl(struct request_list *rl,
412 struct request_queue *q)
414 struct list_head *ent;
415 struct blkcg_gq *blkg;
418 * Determine the current blkg list_head. The first entry is
419 * root_rl which is off @q->blkg_list and mapped to the head.
421 if (rl == &q->root_rl) {
423 /* There are no more block groups, hence no request lists */
427 blkg = container_of(rl, struct blkcg_gq, rl);
431 /* walk to the next list_head, skip root blkcg */
433 if (ent == &q->root_blkg->q_node)
435 if (ent == &q->blkg_list)
438 blkg = container_of(ent, struct blkcg_gq, q_node);
442 static int blkcg_reset_stats(struct cgroup *cgroup, struct cftype *cftype,
445 struct blkcg *blkcg = cgroup_to_blkcg(cgroup);
446 struct blkcg_gq *blkg;
449 mutex_lock(&blkcg_pol_mutex);
450 spin_lock_irq(&blkcg->lock);
453 * Note that stat reset is racy - it doesn't synchronize against
454 * stat updates. This is a debug feature which shouldn't exist
455 * anyway. If you get hit by a race, retry.
457 hlist_for_each_entry(blkg, &blkcg->blkg_list, blkcg_node) {
458 for (i = 0; i < BLKCG_MAX_POLS; i++) {
459 struct blkcg_policy *pol = blkcg_policy[i];
461 if (blkcg_policy_enabled(blkg->q, pol) &&
462 pol->pd_reset_stats_fn)
463 pol->pd_reset_stats_fn(blkg);
467 spin_unlock_irq(&blkcg->lock);
468 mutex_unlock(&blkcg_pol_mutex);
472 static const char *blkg_dev_name(struct blkcg_gq *blkg)
474 /* some drivers (floppy) instantiate a queue w/o disk registered */
475 if (blkg->q->backing_dev_info.dev)
476 return dev_name(blkg->q->backing_dev_info.dev);
481 * blkcg_print_blkgs - helper for printing per-blkg data
482 * @sf: seq_file to print to
483 * @blkcg: blkcg of interest
484 * @prfill: fill function to print out a blkg
485 * @pol: policy in question
486 * @data: data to be passed to @prfill
487 * @show_total: to print out sum of prfill return values or not
489 * This function invokes @prfill on each blkg of @blkcg if pd for the
490 * policy specified by @pol exists. @prfill is invoked with @sf, the
491 * policy data and @data and the matching queue lock held. If @show_total
492 * is %true, the sum of the return values from @prfill is printed with
493 * "Total" label at the end.
495 * This is to be used to construct print functions for
496 * cftype->read_seq_string method.
498 void blkcg_print_blkgs(struct seq_file *sf, struct blkcg *blkcg,
499 u64 (*prfill)(struct seq_file *,
500 struct blkg_policy_data *, int),
501 const struct blkcg_policy *pol, int data,
504 struct blkcg_gq *blkg;
508 hlist_for_each_entry_rcu(blkg, &blkcg->blkg_list, blkcg_node) {
509 spin_lock_irq(blkg->q->queue_lock);
510 if (blkcg_policy_enabled(blkg->q, pol))
511 total += prfill(sf, blkg->pd[pol->plid], data);
512 spin_unlock_irq(blkg->q->queue_lock);
517 seq_printf(sf, "Total %llu\n", (unsigned long long)total);
519 EXPORT_SYMBOL_GPL(blkcg_print_blkgs);
522 * __blkg_prfill_u64 - prfill helper for a single u64 value
523 * @sf: seq_file to print to
524 * @pd: policy private data of interest
527 * Print @v to @sf for the device assocaited with @pd.
529 u64 __blkg_prfill_u64(struct seq_file *sf, struct blkg_policy_data *pd, u64 v)
531 const char *dname = blkg_dev_name(pd->blkg);
536 seq_printf(sf, "%s %llu\n", dname, (unsigned long long)v);
539 EXPORT_SYMBOL_GPL(__blkg_prfill_u64);
542 * __blkg_prfill_rwstat - prfill helper for a blkg_rwstat
543 * @sf: seq_file to print to
544 * @pd: policy private data of interest
545 * @rwstat: rwstat to print
547 * Print @rwstat to @sf for the device assocaited with @pd.
549 u64 __blkg_prfill_rwstat(struct seq_file *sf, struct blkg_policy_data *pd,
550 const struct blkg_rwstat *rwstat)
552 static const char *rwstr[] = {
553 [BLKG_RWSTAT_READ] = "Read",
554 [BLKG_RWSTAT_WRITE] = "Write",
555 [BLKG_RWSTAT_SYNC] = "Sync",
556 [BLKG_RWSTAT_ASYNC] = "Async",
558 const char *dname = blkg_dev_name(pd->blkg);
565 for (i = 0; i < BLKG_RWSTAT_NR; i++)
566 seq_printf(sf, "%s %s %llu\n", dname, rwstr[i],
567 (unsigned long long)rwstat->cnt[i]);
569 v = rwstat->cnt[BLKG_RWSTAT_READ] + rwstat->cnt[BLKG_RWSTAT_WRITE];
570 seq_printf(sf, "%s Total %llu\n", dname, (unsigned long long)v);
573 EXPORT_SYMBOL_GPL(__blkg_prfill_rwstat);
576 * blkg_prfill_stat - prfill callback for blkg_stat
577 * @sf: seq_file to print to
578 * @pd: policy private data of interest
579 * @off: offset to the blkg_stat in @pd
581 * prfill callback for printing a blkg_stat.
583 u64 blkg_prfill_stat(struct seq_file *sf, struct blkg_policy_data *pd, int off)
585 return __blkg_prfill_u64(sf, pd, blkg_stat_read((void *)pd + off));
587 EXPORT_SYMBOL_GPL(blkg_prfill_stat);
590 * blkg_prfill_rwstat - prfill callback for blkg_rwstat
591 * @sf: seq_file to print to
592 * @pd: policy private data of interest
593 * @off: offset to the blkg_rwstat in @pd
595 * prfill callback for printing a blkg_rwstat.
597 u64 blkg_prfill_rwstat(struct seq_file *sf, struct blkg_policy_data *pd,
600 struct blkg_rwstat rwstat = blkg_rwstat_read((void *)pd + off);
602 return __blkg_prfill_rwstat(sf, pd, &rwstat);
604 EXPORT_SYMBOL_GPL(blkg_prfill_rwstat);
607 * blkg_stat_recursive_sum - collect hierarchical blkg_stat
608 * @pd: policy private data of interest
609 * @off: offset to the blkg_stat in @pd
611 * Collect the blkg_stat specified by @off from @pd and all its online
612 * descendants and return the sum. The caller must be holding the queue
613 * lock for online tests.
615 u64 blkg_stat_recursive_sum(struct blkg_policy_data *pd, int off)
617 struct blkcg_policy *pol = blkcg_policy[pd->plid];
618 struct blkcg_gq *pos_blkg;
619 struct cgroup *pos_cgrp;
622 lockdep_assert_held(pd->blkg->q->queue_lock);
624 sum = blkg_stat_read((void *)pd + off);
627 blkg_for_each_descendant_pre(pos_blkg, pos_cgrp, pd_to_blkg(pd)) {
628 struct blkg_policy_data *pos_pd = blkg_to_pd(pos_blkg, pol);
629 struct blkg_stat *stat = (void *)pos_pd + off;
631 if (pos_blkg->online)
632 sum += blkg_stat_read(stat);
638 EXPORT_SYMBOL_GPL(blkg_stat_recursive_sum);
641 * blkg_rwstat_recursive_sum - collect hierarchical blkg_rwstat
642 * @pd: policy private data of interest
643 * @off: offset to the blkg_stat in @pd
645 * Collect the blkg_rwstat specified by @off from @pd and all its online
646 * descendants and return the sum. The caller must be holding the queue
647 * lock for online tests.
649 struct blkg_rwstat blkg_rwstat_recursive_sum(struct blkg_policy_data *pd,
652 struct blkcg_policy *pol = blkcg_policy[pd->plid];
653 struct blkcg_gq *pos_blkg;
654 struct cgroup *pos_cgrp;
655 struct blkg_rwstat sum;
658 lockdep_assert_held(pd->blkg->q->queue_lock);
660 sum = blkg_rwstat_read((void *)pd + off);
663 blkg_for_each_descendant_pre(pos_blkg, pos_cgrp, pd_to_blkg(pd)) {
664 struct blkg_policy_data *pos_pd = blkg_to_pd(pos_blkg, pol);
665 struct blkg_rwstat *rwstat = (void *)pos_pd + off;
666 struct blkg_rwstat tmp;
668 if (!pos_blkg->online)
671 tmp = blkg_rwstat_read(rwstat);
673 for (i = 0; i < BLKG_RWSTAT_NR; i++)
674 sum.cnt[i] += tmp.cnt[i];
680 EXPORT_SYMBOL_GPL(blkg_rwstat_recursive_sum);
683 * blkg_conf_prep - parse and prepare for per-blkg config update
684 * @blkcg: target block cgroup
685 * @pol: target policy
686 * @input: input string
687 * @ctx: blkg_conf_ctx to be filled
689 * Parse per-blkg config update from @input and initialize @ctx with the
690 * result. @ctx->blkg points to the blkg to be updated and @ctx->v the new
691 * value. This function returns with RCU read lock and queue lock held and
692 * must be paired with blkg_conf_finish().
694 int blkg_conf_prep(struct blkcg *blkcg, const struct blkcg_policy *pol,
695 const char *input, struct blkg_conf_ctx *ctx)
696 __acquires(rcu) __acquires(disk->queue->queue_lock)
698 struct gendisk *disk;
699 struct blkcg_gq *blkg;
700 unsigned int major, minor;
701 unsigned long long v;
704 if (sscanf(input, "%u:%u %llu", &major, &minor, &v) != 3)
707 disk = get_gendisk(MKDEV(major, minor), &part);
712 spin_lock_irq(disk->queue->queue_lock);
714 if (blkcg_policy_enabled(disk->queue, pol))
715 blkg = blkg_lookup_create(blkcg, disk->queue);
717 blkg = ERR_PTR(-EINVAL);
722 spin_unlock_irq(disk->queue->queue_lock);
725 * If queue was bypassing, we should retry. Do so after a
726 * short msleep(). It isn't strictly necessary but queue
727 * can be bypassing for some time and it's always nice to
728 * avoid busy looping.
732 ret = restart_syscall();
742 EXPORT_SYMBOL_GPL(blkg_conf_prep);
745 * blkg_conf_finish - finish up per-blkg config update
746 * @ctx: blkg_conf_ctx intiailized by blkg_conf_prep()
748 * Finish up after per-blkg config update. This function must be paired
749 * with blkg_conf_prep().
751 void blkg_conf_finish(struct blkg_conf_ctx *ctx)
752 __releases(ctx->disk->queue->queue_lock) __releases(rcu)
754 spin_unlock_irq(ctx->disk->queue->queue_lock);
758 EXPORT_SYMBOL_GPL(blkg_conf_finish);
760 struct cftype blkcg_files[] = {
762 .name = "reset_stats",
763 .write_u64 = blkcg_reset_stats,
769 * blkcg_css_offline - cgroup css_offline callback
770 * @cgroup: cgroup of interest
772 * This function is called when @cgroup is about to go away and responsible
773 * for shooting down all blkgs associated with @cgroup. blkgs should be
774 * removed while holding both q and blkcg locks. As blkcg lock is nested
775 * inside q lock, this function performs reverse double lock dancing.
777 * This is the blkcg counterpart of ioc_release_fn().
779 static void blkcg_css_offline(struct cgroup *cgroup)
781 struct blkcg *blkcg = cgroup_to_blkcg(cgroup);
783 spin_lock_irq(&blkcg->lock);
785 while (!hlist_empty(&blkcg->blkg_list)) {
786 struct blkcg_gq *blkg = hlist_entry(blkcg->blkg_list.first,
787 struct blkcg_gq, blkcg_node);
788 struct request_queue *q = blkg->q;
790 if (spin_trylock(q->queue_lock)) {
792 spin_unlock(q->queue_lock);
794 spin_unlock_irq(&blkcg->lock);
796 spin_lock_irq(&blkcg->lock);
800 spin_unlock_irq(&blkcg->lock);
803 static void blkcg_css_free(struct cgroup *cgroup)
805 struct blkcg *blkcg = cgroup_to_blkcg(cgroup);
807 if (blkcg != &blkcg_root)
811 static struct cgroup_subsys_state *blkcg_css_alloc(struct cgroup *cgroup)
813 static atomic64_t id_seq = ATOMIC64_INIT(0);
815 struct cgroup *parent = cgroup->parent;
822 blkcg = kzalloc(sizeof(*blkcg), GFP_KERNEL);
824 return ERR_PTR(-ENOMEM);
826 blkcg->cfq_weight = CFQ_WEIGHT_DEFAULT;
827 blkcg->cfq_leaf_weight = CFQ_WEIGHT_DEFAULT;
828 blkcg->id = atomic64_inc_return(&id_seq); /* root is 0, start from 1 */
830 spin_lock_init(&blkcg->lock);
831 INIT_RADIX_TREE(&blkcg->blkg_tree, GFP_ATOMIC);
832 INIT_HLIST_HEAD(&blkcg->blkg_list);
838 * blkcg_init_queue - initialize blkcg part of request queue
839 * @q: request_queue to initialize
841 * Called from blk_alloc_queue_node(). Responsible for initializing blkcg
842 * part of new request_queue @q.
845 * 0 on success, -errno on failure.
847 int blkcg_init_queue(struct request_queue *q)
851 return blk_throtl_init(q);
855 * blkcg_drain_queue - drain blkcg part of request_queue
856 * @q: request_queue to drain
858 * Called from blk_drain_queue(). Responsible for draining blkcg part.
860 void blkcg_drain_queue(struct request_queue *q)
862 lockdep_assert_held(q->queue_lock);
868 * blkcg_exit_queue - exit and release blkcg part of request_queue
869 * @q: request_queue being released
871 * Called from blk_release_queue(). Responsible for exiting blkcg part.
873 void blkcg_exit_queue(struct request_queue *q)
875 spin_lock_irq(q->queue_lock);
877 spin_unlock_irq(q->queue_lock);
883 * We cannot support shared io contexts, as we have no mean to support
884 * two tasks with the same ioc in two different groups without major rework
885 * of the main cic data structures. For now we allow a task to change
886 * its cgroup only if it's the only owner of its ioc.
888 static int blkcg_can_attach(struct cgroup *cgrp, struct cgroup_taskset *tset)
890 struct task_struct *task;
891 struct io_context *ioc;
894 /* task_lock() is needed to avoid races with exit_io_context() */
895 cgroup_taskset_for_each(task, cgrp, tset) {
897 ioc = task->io_context;
898 if (ioc && atomic_read(&ioc->nr_tasks) > 1)
907 struct cgroup_subsys blkio_subsys = {
909 .css_alloc = blkcg_css_alloc,
910 .css_offline = blkcg_css_offline,
911 .css_free = blkcg_css_free,
912 .can_attach = blkcg_can_attach,
913 .subsys_id = blkio_subsys_id,
914 .base_cftypes = blkcg_files,
915 .module = THIS_MODULE,
918 * blkio subsystem is utterly broken in terms of hierarchy support.
919 * It treats all cgroups equally regardless of where they're
920 * located in the hierarchy - all cgroups are treated as if they're
921 * right below the root. Fix it and remove the following.
923 .broken_hierarchy = true,
925 EXPORT_SYMBOL_GPL(blkio_subsys);
928 * blkcg_activate_policy - activate a blkcg policy on a request_queue
929 * @q: request_queue of interest
930 * @pol: blkcg policy to activate
932 * Activate @pol on @q. Requires %GFP_KERNEL context. @q goes through
933 * bypass mode to populate its blkgs with policy_data for @pol.
935 * Activation happens with @q bypassed, so nobody would be accessing blkgs
936 * from IO path. Update of each blkg is protected by both queue and blkcg
937 * locks so that holding either lock and testing blkcg_policy_enabled() is
938 * always enough for dereferencing policy data.
940 * The caller is responsible for synchronizing [de]activations and policy
941 * [un]registerations. Returns 0 on success, -errno on failure.
943 int blkcg_activate_policy(struct request_queue *q,
944 const struct blkcg_policy *pol)
947 struct blkcg_gq *blkg, *new_blkg;
948 struct blkg_policy_data *pd, *n;
952 if (blkcg_policy_enabled(q, pol))
955 /* preallocations for root blkg */
956 new_blkg = blkg_alloc(&blkcg_root, q, GFP_KERNEL);
960 blk_queue_bypass_start(q);
962 preloaded = !radix_tree_preload(GFP_KERNEL);
965 * Make sure the root blkg exists and count the existing blkgs. As
966 * @q is bypassing at this point, blkg_lookup_create() can't be
967 * used. Open code it.
969 spin_lock_irq(q->queue_lock);
972 blkg = __blkg_lookup(&blkcg_root, q, false);
976 blkg = blkg_create(&blkcg_root, q, new_blkg);
980 radix_tree_preload_end();
987 q->root_rl.blkg = blkg;
989 list_for_each_entry(blkg, &q->blkg_list, q_node)
992 spin_unlock_irq(q->queue_lock);
994 /* allocate policy_data for all existing blkgs */
996 pd = kzalloc_node(pol->pd_size, GFP_KERNEL, q->node);
1001 list_add_tail(&pd->alloc_node, &pds);
1005 * Install the allocated pds. With @q bypassing, no new blkg
1006 * should have been created while the queue lock was dropped.
1008 spin_lock_irq(q->queue_lock);
1010 list_for_each_entry(blkg, &q->blkg_list, q_node) {
1011 if (WARN_ON(list_empty(&pds))) {
1012 /* umm... this shouldn't happen, just abort */
1016 pd = list_first_entry(&pds, struct blkg_policy_data, alloc_node);
1017 list_del_init(&pd->alloc_node);
1019 /* grab blkcg lock too while installing @pd on @blkg */
1020 spin_lock(&blkg->blkcg->lock);
1022 blkg->pd[pol->plid] = pd;
1024 pd->plid = pol->plid;
1025 pol->pd_init_fn(blkg);
1027 spin_unlock(&blkg->blkcg->lock);
1030 __set_bit(pol->plid, q->blkcg_pols);
1033 spin_unlock_irq(q->queue_lock);
1035 blk_queue_bypass_end(q);
1036 list_for_each_entry_safe(pd, n, &pds, alloc_node)
1040 EXPORT_SYMBOL_GPL(blkcg_activate_policy);
1043 * blkcg_deactivate_policy - deactivate a blkcg policy on a request_queue
1044 * @q: request_queue of interest
1045 * @pol: blkcg policy to deactivate
1047 * Deactivate @pol on @q. Follows the same synchronization rules as
1048 * blkcg_activate_policy().
1050 void blkcg_deactivate_policy(struct request_queue *q,
1051 const struct blkcg_policy *pol)
1053 struct blkcg_gq *blkg;
1055 if (!blkcg_policy_enabled(q, pol))
1058 blk_queue_bypass_start(q);
1059 spin_lock_irq(q->queue_lock);
1061 __clear_bit(pol->plid, q->blkcg_pols);
1063 /* if no policy is left, no need for blkgs - shoot them down */
1064 if (bitmap_empty(q->blkcg_pols, BLKCG_MAX_POLS))
1065 blkg_destroy_all(q);
1067 list_for_each_entry(blkg, &q->blkg_list, q_node) {
1068 /* grab blkcg lock too while removing @pd from @blkg */
1069 spin_lock(&blkg->blkcg->lock);
1071 if (pol->pd_offline_fn)
1072 pol->pd_offline_fn(blkg);
1073 if (pol->pd_exit_fn)
1074 pol->pd_exit_fn(blkg);
1076 kfree(blkg->pd[pol->plid]);
1077 blkg->pd[pol->plid] = NULL;
1079 spin_unlock(&blkg->blkcg->lock);
1082 spin_unlock_irq(q->queue_lock);
1083 blk_queue_bypass_end(q);
1085 EXPORT_SYMBOL_GPL(blkcg_deactivate_policy);
1088 * blkcg_policy_register - register a blkcg policy
1089 * @pol: blkcg policy to register
1091 * Register @pol with blkcg core. Might sleep and @pol may be modified on
1092 * successful registration. Returns 0 on success and -errno on failure.
1094 int blkcg_policy_register(struct blkcg_policy *pol)
1098 if (WARN_ON(pol->pd_size < sizeof(struct blkg_policy_data)))
1101 mutex_lock(&blkcg_pol_mutex);
1103 /* find an empty slot */
1105 for (i = 0; i < BLKCG_MAX_POLS; i++)
1106 if (!blkcg_policy[i])
1108 if (i >= BLKCG_MAX_POLS)
1111 /* register and update blkgs */
1113 blkcg_policy[i] = pol;
1115 /* everything is in place, add intf files for the new policy */
1117 WARN_ON(cgroup_add_cftypes(&blkio_subsys, pol->cftypes));
1120 mutex_unlock(&blkcg_pol_mutex);
1123 EXPORT_SYMBOL_GPL(blkcg_policy_register);
1126 * blkcg_policy_unregister - unregister a blkcg policy
1127 * @pol: blkcg policy to unregister
1129 * Undo blkcg_policy_register(@pol). Might sleep.
1131 void blkcg_policy_unregister(struct blkcg_policy *pol)
1133 mutex_lock(&blkcg_pol_mutex);
1135 if (WARN_ON(blkcg_policy[pol->plid] != pol))
1138 /* kill the intf files first */
1140 cgroup_rm_cftypes(&blkio_subsys, pol->cftypes);
1142 /* unregister and update blkgs */
1143 blkcg_policy[pol->plid] = NULL;
1145 mutex_unlock(&blkcg_pol_mutex);
1147 EXPORT_SYMBOL_GPL(blkcg_policy_unregister);