1 /* SPDX-License-Identifier: GPL-2.0 */
5 * Common Block IO controller cgroup interface
7 * Based on ideas and code from CFQ, CFS and BFQ:
8 * Copyright (C) 2003 Jens Axboe <axboe@kernel.dk>
10 * Copyright (C) 2008 Fabio Checconi <fabio@gandalf.sssup.it>
11 * Paolo Valente <paolo.valente@unimore.it>
13 * Copyright (C) 2009 Vivek Goyal <vgoyal@redhat.com>
14 * Nauman Rafique <nauman@google.com>
17 #include <linux/cgroup.h>
18 #include <linux/percpu_counter.h>
19 #include <linux/seq_file.h>
20 #include <linux/radix-tree.h>
21 #include <linux/blkdev.h>
22 #include <linux/atomic.h>
23 #include <linux/kthread.h>
25 #include <linux/blkdev.h>
27 /* percpu_counter batch for blkg_[rw]stats, per-cpu drift doesn't matter */
28 #define BLKG_STAT_CPU_BATCH (INT_MAX / 2)
30 /* Max limits for throttle policy */
31 #define THROTL_IOPS_MAX UINT_MAX
33 #ifdef CONFIG_BLK_CGROUP
35 enum blkg_rwstat_type {
43 BLKG_RWSTAT_TOTAL = BLKG_RWSTAT_NR,
49 struct cgroup_subsys_state css;
52 struct radix_tree_root blkg_tree;
53 struct blkcg_gq __rcu *blkg_hint;
54 struct hlist_head blkg_list;
56 struct blkcg_policy_data *cpd[BLKCG_MAX_POLS];
58 struct list_head all_blkcgs_node;
59 #ifdef CONFIG_CGROUP_WRITEBACK
60 struct list_head cgwb_list;
61 refcount_t cgwb_refcnt;
66 * blkg_[rw]stat->aux_cnt is excluded for local stats but included for
67 * recursive. Used to carry stats of dead children.
70 struct percpu_counter cpu_cnt[BLKG_RWSTAT_NR];
71 atomic64_t aux_cnt[BLKG_RWSTAT_NR];
74 struct blkg_rwstat_sample {
75 u64 cnt[BLKG_RWSTAT_NR];
79 * A blkcg_gq (blkg) is association between a block cgroup (blkcg) and a
80 * request_queue (q). This is used by blkcg policies which need to track
81 * information per blkcg - q pair.
83 * There can be multiple active blkcg policies and each blkg:policy pair is
84 * represented by a blkg_policy_data which is allocated and freed by each
85 * policy's pd_alloc/free_fn() methods. A policy can allocate private data
86 * area by allocating larger data structure which embeds blkg_policy_data
89 struct blkg_policy_data {
90 /* the blkg and policy id this per-policy data belongs to */
91 struct blkcg_gq *blkg;
96 * Policies that need to keep per-blkcg data which is independent from any
97 * request_queue associated to it should implement cpd_alloc/free_fn()
98 * methods. A policy can allocate private data area by allocating larger
99 * data structure which embeds blkcg_policy_data at the beginning.
100 * cpd_init() is invoked to let each policy handle per-blkcg data.
102 struct blkcg_policy_data {
103 /* the blkcg and policy id this per-policy data belongs to */
108 /* association between a blk cgroup and a request queue */
110 /* Pointer to the associated request_queue */
111 struct request_queue *q;
112 struct list_head q_node;
113 struct hlist_node blkcg_node;
117 * Each blkg gets congested separately and the congestion state is
118 * propagated to the matching bdi_writeback_congested.
120 struct bdi_writeback_congested *wb_congested;
122 /* all non-root blkcg_gq's are guaranteed to have access to parent */
123 struct blkcg_gq *parent;
125 /* reference count */
126 struct percpu_ref refcnt;
128 /* is this blkg online? protected by both blkcg and q locks */
131 struct blkg_rwstat stat_bytes;
132 struct blkg_rwstat stat_ios;
134 struct blkg_policy_data *pd[BLKCG_MAX_POLS];
136 spinlock_t async_bio_lock;
137 struct bio_list async_bios;
138 struct work_struct async_bio_work;
141 atomic64_t delay_nsec;
142 atomic64_t delay_start;
146 struct rcu_head rcu_head;
149 typedef struct blkcg_policy_data *(blkcg_pol_alloc_cpd_fn)(gfp_t gfp);
150 typedef void (blkcg_pol_init_cpd_fn)(struct blkcg_policy_data *cpd);
151 typedef void (blkcg_pol_free_cpd_fn)(struct blkcg_policy_data *cpd);
152 typedef void (blkcg_pol_bind_cpd_fn)(struct blkcg_policy_data *cpd);
153 typedef struct blkg_policy_data *(blkcg_pol_alloc_pd_fn)(gfp_t gfp,
154 struct request_queue *q, struct blkcg *blkcg);
155 typedef void (blkcg_pol_init_pd_fn)(struct blkg_policy_data *pd);
156 typedef void (blkcg_pol_online_pd_fn)(struct blkg_policy_data *pd);
157 typedef void (blkcg_pol_offline_pd_fn)(struct blkg_policy_data *pd);
158 typedef void (blkcg_pol_free_pd_fn)(struct blkg_policy_data *pd);
159 typedef void (blkcg_pol_reset_pd_stats_fn)(struct blkg_policy_data *pd);
160 typedef size_t (blkcg_pol_stat_pd_fn)(struct blkg_policy_data *pd, char *buf,
163 struct blkcg_policy {
165 /* cgroup files for the policy */
166 struct cftype *dfl_cftypes;
167 struct cftype *legacy_cftypes;
170 blkcg_pol_alloc_cpd_fn *cpd_alloc_fn;
171 blkcg_pol_init_cpd_fn *cpd_init_fn;
172 blkcg_pol_free_cpd_fn *cpd_free_fn;
173 blkcg_pol_bind_cpd_fn *cpd_bind_fn;
175 blkcg_pol_alloc_pd_fn *pd_alloc_fn;
176 blkcg_pol_init_pd_fn *pd_init_fn;
177 blkcg_pol_online_pd_fn *pd_online_fn;
178 blkcg_pol_offline_pd_fn *pd_offline_fn;
179 blkcg_pol_free_pd_fn *pd_free_fn;
180 blkcg_pol_reset_pd_stats_fn *pd_reset_stats_fn;
181 blkcg_pol_stat_pd_fn *pd_stat_fn;
184 extern struct blkcg blkcg_root;
185 extern struct cgroup_subsys_state * const blkcg_root_css;
186 extern bool blkcg_debug_stats;
188 struct blkcg_gq *blkg_lookup_slowpath(struct blkcg *blkcg,
189 struct request_queue *q, bool update_hint);
190 struct blkcg_gq *__blkg_lookup_create(struct blkcg *blkcg,
191 struct request_queue *q);
192 struct blkcg_gq *blkg_lookup_create(struct blkcg *blkcg,
193 struct request_queue *q);
194 int blkcg_init_queue(struct request_queue *q);
195 void blkcg_drain_queue(struct request_queue *q);
196 void blkcg_exit_queue(struct request_queue *q);
198 /* Blkio controller policy registration */
199 int blkcg_policy_register(struct blkcg_policy *pol);
200 void blkcg_policy_unregister(struct blkcg_policy *pol);
201 int blkcg_activate_policy(struct request_queue *q,
202 const struct blkcg_policy *pol);
203 void blkcg_deactivate_policy(struct request_queue *q,
204 const struct blkcg_policy *pol);
206 static inline u64 blkg_rwstat_read_counter(struct blkg_rwstat *rwstat,
209 return atomic64_read(&rwstat->aux_cnt[idx]) +
210 percpu_counter_sum_positive(&rwstat->cpu_cnt[idx]);
213 const char *blkg_dev_name(struct blkcg_gq *blkg);
214 void blkcg_print_blkgs(struct seq_file *sf, struct blkcg *blkcg,
215 u64 (*prfill)(struct seq_file *,
216 struct blkg_policy_data *, int),
217 const struct blkcg_policy *pol, int data,
219 u64 __blkg_prfill_u64(struct seq_file *sf, struct blkg_policy_data *pd, u64 v);
220 u64 __blkg_prfill_rwstat(struct seq_file *sf, struct blkg_policy_data *pd,
221 const struct blkg_rwstat_sample *rwstat);
222 u64 blkg_prfill_rwstat(struct seq_file *sf, struct blkg_policy_data *pd,
224 int blkg_print_stat_bytes(struct seq_file *sf, void *v);
225 int blkg_print_stat_ios(struct seq_file *sf, void *v);
226 int blkg_print_stat_bytes_recursive(struct seq_file *sf, void *v);
227 int blkg_print_stat_ios_recursive(struct seq_file *sf, void *v);
229 void blkg_rwstat_recursive_sum(struct blkcg_gq *blkg, struct blkcg_policy *pol,
230 int off, struct blkg_rwstat_sample *sum);
232 struct blkg_conf_ctx {
233 struct gendisk *disk;
234 struct blkcg_gq *blkg;
238 struct gendisk *blkcg_conf_get_disk(char **inputp);
239 int blkg_conf_prep(struct blkcg *blkcg, const struct blkcg_policy *pol,
240 char *input, struct blkg_conf_ctx *ctx);
241 void blkg_conf_finish(struct blkg_conf_ctx *ctx);
244 * blkcg_css - find the current css
246 * Find the css associated with either the kthread or the current task.
247 * This may return a dying css, so it is up to the caller to use tryget logic
248 * to confirm it is alive and well.
250 static inline struct cgroup_subsys_state *blkcg_css(void)
252 struct cgroup_subsys_state *css;
254 css = kthread_blkcg();
257 return task_css(current, io_cgrp_id);
260 static inline struct blkcg *css_to_blkcg(struct cgroup_subsys_state *css)
262 return css ? container_of(css, struct blkcg, css) : NULL;
266 * __bio_blkcg - internal, inconsistent version to get blkcg
269 * This function is inconsistent and consequently is dangerous to use. The
270 * first part of the function returns a blkcg where a reference is owned by the
271 * bio. This means it does not need to be rcu protected as it cannot go away
272 * with the bio owning a reference to it. However, the latter potentially gets
273 * it from task_css(). This can race against task migration and the cgroup
274 * dying. It is also semantically different as it must be called rcu protected
275 * and is susceptible to failure when trying to get a reference to it.
276 * Therefore, it is not ok to assume that *_get() will always succeed on the
277 * blkcg returned here.
279 static inline struct blkcg *__bio_blkcg(struct bio *bio)
281 if (bio && bio->bi_blkg)
282 return bio->bi_blkg->blkcg;
283 return css_to_blkcg(blkcg_css());
287 * bio_blkcg - grab the blkcg associated with a bio
290 * This returns the blkcg associated with a bio, %NULL if not associated.
291 * Callers are expected to either handle %NULL or know association has been
292 * done prior to calling this.
294 static inline struct blkcg *bio_blkcg(struct bio *bio)
296 if (bio && bio->bi_blkg)
297 return bio->bi_blkg->blkcg;
301 static inline bool blk_cgroup_congested(void)
303 struct cgroup_subsys_state *css;
307 css = kthread_blkcg();
309 css = task_css(current, io_cgrp_id);
311 if (atomic_read(&css->cgroup->congestion_count)) {
322 * bio_issue_as_root_blkg - see if this bio needs to be issued as root blkg
323 * @return: true if this bio needs to be submitted with the root blkg context.
325 * In order to avoid priority inversions we sometimes need to issue a bio as if
326 * it were attached to the root blkg, and then backcharge to the actual owning
327 * blkg. The idea is we do bio_blkcg() to look up the actual context for the
328 * bio and attach the appropriate blkg to the bio. Then we call this helper and
329 * if it is true run with the root blkg for that queue and then do any
330 * backcharging to the originating cgroup once the io is complete.
332 static inline bool bio_issue_as_root_blkg(struct bio *bio)
334 return (bio->bi_opf & (REQ_META | REQ_SWAP)) != 0;
338 * blkcg_parent - get the parent of a blkcg
339 * @blkcg: blkcg of interest
341 * Return the parent blkcg of @blkcg. Can be called anytime.
343 static inline struct blkcg *blkcg_parent(struct blkcg *blkcg)
345 return css_to_blkcg(blkcg->css.parent);
349 * __blkg_lookup - internal version of blkg_lookup()
350 * @blkcg: blkcg of interest
351 * @q: request_queue of interest
352 * @update_hint: whether to update lookup hint with the result or not
354 * This is internal version and shouldn't be used by policy
355 * implementations. Looks up blkgs for the @blkcg - @q pair regardless of
356 * @q's bypass state. If @update_hint is %true, the caller should be
357 * holding @q->queue_lock and lookup hint is updated on success.
359 static inline struct blkcg_gq *__blkg_lookup(struct blkcg *blkcg,
360 struct request_queue *q,
363 struct blkcg_gq *blkg;
365 if (blkcg == &blkcg_root)
368 blkg = rcu_dereference(blkcg->blkg_hint);
369 if (blkg && blkg->q == q)
372 return blkg_lookup_slowpath(blkcg, q, update_hint);
376 * blkg_lookup - lookup blkg for the specified blkcg - q pair
377 * @blkcg: blkcg of interest
378 * @q: request_queue of interest
380 * Lookup blkg for the @blkcg - @q pair. This function should be called
381 * under RCU read lock.
383 static inline struct blkcg_gq *blkg_lookup(struct blkcg *blkcg,
384 struct request_queue *q)
386 WARN_ON_ONCE(!rcu_read_lock_held());
387 return __blkg_lookup(blkcg, q, false);
391 * blk_queue_root_blkg - return blkg for the (blkcg_root, @q) pair
392 * @q: request_queue of interest
394 * Lookup blkg for @q at the root level. See also blkg_lookup().
396 static inline struct blkcg_gq *blk_queue_root_blkg(struct request_queue *q)
402 * blkg_to_pdata - get policy private data
403 * @blkg: blkg of interest
404 * @pol: policy of interest
406 * Return pointer to private data associated with the @blkg-@pol pair.
408 static inline struct blkg_policy_data *blkg_to_pd(struct blkcg_gq *blkg,
409 struct blkcg_policy *pol)
411 return blkg ? blkg->pd[pol->plid] : NULL;
414 static inline struct blkcg_policy_data *blkcg_to_cpd(struct blkcg *blkcg,
415 struct blkcg_policy *pol)
417 return blkcg ? blkcg->cpd[pol->plid] : NULL;
421 * pdata_to_blkg - get blkg associated with policy private data
422 * @pd: policy private data of interest
424 * @pd is policy private data. Determine the blkg it's associated with.
426 static inline struct blkcg_gq *pd_to_blkg(struct blkg_policy_data *pd)
428 return pd ? pd->blkg : NULL;
431 static inline struct blkcg *cpd_to_blkcg(struct blkcg_policy_data *cpd)
433 return cpd ? cpd->blkcg : NULL;
436 extern void blkcg_destroy_blkgs(struct blkcg *blkcg);
438 #ifdef CONFIG_CGROUP_WRITEBACK
441 * blkcg_cgwb_get - get a reference for blkcg->cgwb_list
442 * @blkcg: blkcg of interest
444 * This is used to track the number of active wb's related to a blkcg.
446 static inline void blkcg_cgwb_get(struct blkcg *blkcg)
448 refcount_inc(&blkcg->cgwb_refcnt);
452 * blkcg_cgwb_put - put a reference for @blkcg->cgwb_list
453 * @blkcg: blkcg of interest
455 * This is used to track the number of active wb's related to a blkcg.
456 * When this count goes to zero, all active wb has finished so the
457 * blkcg can continue destruction by calling blkcg_destroy_blkgs().
458 * This work may occur in cgwb_release_workfn() on the cgwb_release
461 static inline void blkcg_cgwb_put(struct blkcg *blkcg)
463 if (refcount_dec_and_test(&blkcg->cgwb_refcnt))
464 blkcg_destroy_blkgs(blkcg);
469 static inline void blkcg_cgwb_get(struct blkcg *blkcg) { }
471 static inline void blkcg_cgwb_put(struct blkcg *blkcg)
473 /* wb isn't being accounted, so trigger destruction right away */
474 blkcg_destroy_blkgs(blkcg);
480 * blkg_path - format cgroup path of blkg
481 * @blkg: blkg of interest
482 * @buf: target buffer
483 * @buflen: target buffer length
485 * Format the path of the cgroup of @blkg into @buf.
487 static inline int blkg_path(struct blkcg_gq *blkg, char *buf, int buflen)
489 return cgroup_path(blkg->blkcg->css.cgroup, buf, buflen);
493 * blkg_get - get a blkg reference
496 * The caller should be holding an existing reference.
498 static inline void blkg_get(struct blkcg_gq *blkg)
500 percpu_ref_get(&blkg->refcnt);
504 * blkg_tryget - try and get a blkg reference
507 * This is for use when doing an RCU lookup of the blkg. We may be in the midst
508 * of freeing this blkg, so we can only use it if the refcnt is not zero.
510 static inline bool blkg_tryget(struct blkcg_gq *blkg)
512 return blkg && percpu_ref_tryget(&blkg->refcnt);
516 * blkg_tryget_closest - try and get a blkg ref on the closet blkg
519 * This needs to be called rcu protected. As the failure mode here is to walk
520 * up the blkg tree, this ensure that the blkg->parent pointers are always
521 * valid. This returns the blkg that it ended up taking a reference on or %NULL
522 * if no reference was taken.
524 static inline struct blkcg_gq *blkg_tryget_closest(struct blkcg_gq *blkg)
526 struct blkcg_gq *ret_blkg = NULL;
528 WARN_ON_ONCE(!rcu_read_lock_held());
531 if (blkg_tryget(blkg)) {
542 * blkg_put - put a blkg reference
545 static inline void blkg_put(struct blkcg_gq *blkg)
547 percpu_ref_put(&blkg->refcnt);
551 * blkg_for_each_descendant_pre - pre-order walk of a blkg's descendants
552 * @d_blkg: loop cursor pointing to the current descendant
553 * @pos_css: used for iteration
554 * @p_blkg: target blkg to walk descendants of
556 * Walk @c_blkg through the descendants of @p_blkg. Must be used with RCU
557 * read locked. If called under either blkcg or queue lock, the iteration
558 * is guaranteed to include all and only online blkgs. The caller may
559 * update @pos_css by calling css_rightmost_descendant() to skip subtree.
560 * @p_blkg is included in the iteration and the first node to be visited.
562 #define blkg_for_each_descendant_pre(d_blkg, pos_css, p_blkg) \
563 css_for_each_descendant_pre((pos_css), &(p_blkg)->blkcg->css) \
564 if (((d_blkg) = __blkg_lookup(css_to_blkcg(pos_css), \
565 (p_blkg)->q, false)))
568 * blkg_for_each_descendant_post - post-order walk of a blkg's descendants
569 * @d_blkg: loop cursor pointing to the current descendant
570 * @pos_css: used for iteration
571 * @p_blkg: target blkg to walk descendants of
573 * Similar to blkg_for_each_descendant_pre() but performs post-order
574 * traversal instead. Synchronization rules are the same. @p_blkg is
575 * included in the iteration and the last node to be visited.
577 #define blkg_for_each_descendant_post(d_blkg, pos_css, p_blkg) \
578 css_for_each_descendant_post((pos_css), &(p_blkg)->blkcg->css) \
579 if (((d_blkg) = __blkg_lookup(css_to_blkcg(pos_css), \
580 (p_blkg)->q, false)))
582 static inline int blkg_rwstat_init(struct blkg_rwstat *rwstat, gfp_t gfp)
586 for (i = 0; i < BLKG_RWSTAT_NR; i++) {
587 ret = percpu_counter_init(&rwstat->cpu_cnt[i], 0, gfp);
590 percpu_counter_destroy(&rwstat->cpu_cnt[i]);
593 atomic64_set(&rwstat->aux_cnt[i], 0);
598 static inline void blkg_rwstat_exit(struct blkg_rwstat *rwstat)
602 for (i = 0; i < BLKG_RWSTAT_NR; i++)
603 percpu_counter_destroy(&rwstat->cpu_cnt[i]);
607 * blkg_rwstat_add - add a value to a blkg_rwstat
608 * @rwstat: target blkg_rwstat
609 * @op: REQ_OP and flags
612 * Add @val to @rwstat. The counters are chosen according to @rw. The
613 * caller is responsible for synchronizing calls to this function.
615 static inline void blkg_rwstat_add(struct blkg_rwstat *rwstat,
616 unsigned int op, uint64_t val)
618 struct percpu_counter *cnt;
620 if (op_is_discard(op))
621 cnt = &rwstat->cpu_cnt[BLKG_RWSTAT_DISCARD];
622 else if (op_is_write(op))
623 cnt = &rwstat->cpu_cnt[BLKG_RWSTAT_WRITE];
625 cnt = &rwstat->cpu_cnt[BLKG_RWSTAT_READ];
627 percpu_counter_add_batch(cnt, val, BLKG_STAT_CPU_BATCH);
630 cnt = &rwstat->cpu_cnt[BLKG_RWSTAT_SYNC];
632 cnt = &rwstat->cpu_cnt[BLKG_RWSTAT_ASYNC];
634 percpu_counter_add_batch(cnt, val, BLKG_STAT_CPU_BATCH);
638 * blkg_rwstat_read - read the current values of a blkg_rwstat
639 * @rwstat: blkg_rwstat to read
641 * Read the current snapshot of @rwstat and return it in the aux counts.
643 static inline void blkg_rwstat_read(struct blkg_rwstat *rwstat,
644 struct blkg_rwstat_sample *result)
648 for (i = 0; i < BLKG_RWSTAT_NR; i++)
650 percpu_counter_sum_positive(&rwstat->cpu_cnt[i]);
654 * blkg_rwstat_total - read the total count of a blkg_rwstat
655 * @rwstat: blkg_rwstat to read
657 * Return the total count of @rwstat regardless of the IO direction. This
658 * function can be called without synchronization and takes care of u64
661 static inline uint64_t blkg_rwstat_total(struct blkg_rwstat *rwstat)
663 struct blkg_rwstat_sample tmp = { };
665 blkg_rwstat_read(rwstat, &tmp);
666 return tmp.cnt[BLKG_RWSTAT_READ] + tmp.cnt[BLKG_RWSTAT_WRITE];
670 * blkg_rwstat_reset - reset a blkg_rwstat
671 * @rwstat: blkg_rwstat to reset
673 static inline void blkg_rwstat_reset(struct blkg_rwstat *rwstat)
677 for (i = 0; i < BLKG_RWSTAT_NR; i++) {
678 percpu_counter_set(&rwstat->cpu_cnt[i], 0);
679 atomic64_set(&rwstat->aux_cnt[i], 0);
684 * blkg_rwstat_add_aux - add a blkg_rwstat into another's aux count
685 * @to: the destination blkg_rwstat
688 * Add @from's count including the aux one to @to's aux count.
690 static inline void blkg_rwstat_add_aux(struct blkg_rwstat *to,
691 struct blkg_rwstat *from)
693 u64 sum[BLKG_RWSTAT_NR];
696 for (i = 0; i < BLKG_RWSTAT_NR; i++)
697 sum[i] = percpu_counter_sum_positive(&from->cpu_cnt[i]);
699 for (i = 0; i < BLKG_RWSTAT_NR; i++)
700 atomic64_add(sum[i] + atomic64_read(&from->aux_cnt[i]),
704 #ifdef CONFIG_BLK_DEV_THROTTLING
705 extern bool blk_throtl_bio(struct request_queue *q, struct blkcg_gq *blkg,
708 static inline bool blk_throtl_bio(struct request_queue *q, struct blkcg_gq *blkg,
709 struct bio *bio) { return false; }
712 bool __blkcg_punt_bio_submit(struct bio *bio);
714 static inline bool blkcg_punt_bio_submit(struct bio *bio)
716 if (bio->bi_opf & REQ_CGROUP_PUNT)
717 return __blkcg_punt_bio_submit(bio);
722 static inline void blkcg_bio_issue_init(struct bio *bio)
724 bio_issue_init(&bio->bi_issue, bio_sectors(bio));
727 static inline bool blkcg_bio_issue_check(struct request_queue *q,
730 struct blkcg_gq *blkg;
736 char b[BDEVNAME_SIZE];
739 "no blkg associated for bio on block-device: %s\n",
740 bio_devname(bio, b));
741 bio_associate_blkg(bio);
746 throtl = blk_throtl_bio(q, blkg, bio);
750 * If the bio is flagged with BIO_QUEUE_ENTERED it means this
751 * is a split bio and we would have already accounted for the
754 if (!bio_flagged(bio, BIO_QUEUE_ENTERED))
755 blkg_rwstat_add(&blkg->stat_bytes, bio->bi_opf,
756 bio->bi_iter.bi_size);
757 blkg_rwstat_add(&blkg->stat_ios, bio->bi_opf, 1);
760 blkcg_bio_issue_init(bio);
766 static inline void blkcg_use_delay(struct blkcg_gq *blkg)
768 if (atomic_add_return(1, &blkg->use_delay) == 1)
769 atomic_inc(&blkg->blkcg->css.cgroup->congestion_count);
773 * blk_cgroup_mergeable - Determine whether to allow or disallow merges
774 * @rq: request to merge into
777 * @bio and @rq should belong to the same cgroup and their issue_as_root should
778 * match. The latter is necessary as we don't want to throttle e.g. a metadata
779 * update because it happens to be next to a regular IO.
781 static inline bool blk_cgroup_mergeable(struct request *rq, struct bio *bio)
783 return rq->bio->bi_blkg == bio->bi_blkg &&
784 bio_issue_as_root_blkg(rq->bio) == bio_issue_as_root_blkg(bio);
787 static inline int blkcg_unuse_delay(struct blkcg_gq *blkg)
789 int old = atomic_read(&blkg->use_delay);
795 * We do this song and dance because we can race with somebody else
796 * adding or removing delay. If we just did an atomic_dec we'd end up
797 * negative and we'd already be in trouble. We need to subtract 1 and
798 * then check to see if we were the last delay so we can drop the
799 * congestion count on the cgroup.
802 int cur = atomic_cmpxchg(&blkg->use_delay, old, old - 1);
811 atomic_dec(&blkg->blkcg->css.cgroup->congestion_count);
815 static inline void blkcg_clear_delay(struct blkcg_gq *blkg)
817 int old = atomic_read(&blkg->use_delay);
820 /* We only want 1 person clearing the congestion count for this blkg. */
822 int cur = atomic_cmpxchg(&blkg->use_delay, old, 0);
824 atomic_dec(&blkg->blkcg->css.cgroup->congestion_count);
831 void blkcg_add_delay(struct blkcg_gq *blkg, u64 now, u64 delta);
832 void blkcg_schedule_throttle(struct request_queue *q, bool use_memdelay);
833 void blkcg_maybe_throttle_current(void);
834 #else /* CONFIG_BLK_CGROUP */
839 struct blkg_policy_data {
842 struct blkcg_policy_data {
848 struct blkcg_policy {
851 #define blkcg_root_css ((struct cgroup_subsys_state *)ERR_PTR(-EINVAL))
853 static inline void blkcg_maybe_throttle_current(void) { }
854 static inline bool blk_cgroup_congested(void) { return false; }
858 static inline void blkcg_schedule_throttle(struct request_queue *q, bool use_memdelay) { }
860 static inline struct blkcg_gq *blkg_lookup(struct blkcg *blkcg, void *key) { return NULL; }
861 static inline struct blkcg_gq *blk_queue_root_blkg(struct request_queue *q)
863 static inline int blkcg_init_queue(struct request_queue *q) { return 0; }
864 static inline void blkcg_drain_queue(struct request_queue *q) { }
865 static inline void blkcg_exit_queue(struct request_queue *q) { }
866 static inline int blkcg_policy_register(struct blkcg_policy *pol) { return 0; }
867 static inline void blkcg_policy_unregister(struct blkcg_policy *pol) { }
868 static inline int blkcg_activate_policy(struct request_queue *q,
869 const struct blkcg_policy *pol) { return 0; }
870 static inline void blkcg_deactivate_policy(struct request_queue *q,
871 const struct blkcg_policy *pol) { }
873 static inline struct blkcg *__bio_blkcg(struct bio *bio) { return NULL; }
874 static inline struct blkcg *bio_blkcg(struct bio *bio) { return NULL; }
876 static inline struct blkg_policy_data *blkg_to_pd(struct blkcg_gq *blkg,
877 struct blkcg_policy *pol) { return NULL; }
878 static inline struct blkcg_gq *pd_to_blkg(struct blkg_policy_data *pd) { return NULL; }
879 static inline char *blkg_path(struct blkcg_gq *blkg) { return NULL; }
880 static inline void blkg_get(struct blkcg_gq *blkg) { }
881 static inline void blkg_put(struct blkcg_gq *blkg) { }
883 static inline bool blkcg_punt_bio_submit(struct bio *bio) { return false; }
884 static inline void blkcg_bio_issue_init(struct bio *bio) { }
885 static inline bool blkcg_bio_issue_check(struct request_queue *q,
886 struct bio *bio) { return true; }
887 static inline bool blk_cgroup_mergeable(struct request *rq, struct bio *bio) { return true; }
889 #define blk_queue_for_each_rl(rl, q) \
890 for ((rl) = &(q)->root_rl; (rl); (rl) = NULL)
892 #endif /* CONFIG_BLOCK */
893 #endif /* CONFIG_BLK_CGROUP */
894 #endif /* _BLK_CGROUP_H */