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 /* percpu_counter batch for blkg_[rw]stats, per-cpu drift doesn't matter */
26 #define BLKG_STAT_CPU_BATCH (INT_MAX / 2)
28 /* Max limits for throttle policy */
29 #define THROTL_IOPS_MAX UINT_MAX
31 #ifdef CONFIG_BLK_CGROUP
33 enum blkg_rwstat_type {
41 BLKG_RWSTAT_TOTAL = BLKG_RWSTAT_NR,
47 struct cgroup_subsys_state css;
50 struct radix_tree_root blkg_tree;
51 struct blkcg_gq __rcu *blkg_hint;
52 struct hlist_head blkg_list;
54 struct blkcg_policy_data *cpd[BLKCG_MAX_POLS];
56 struct list_head all_blkcgs_node;
57 #ifdef CONFIG_CGROUP_WRITEBACK
58 struct list_head cgwb_list;
59 refcount_t cgwb_refcnt;
64 * blkg_[rw]stat->aux_cnt is excluded for local stats but included for
65 * recursive. Used to carry stats of dead children, and, for blkg_rwstat,
66 * to carry result values from read and sum operations.
69 struct percpu_counter cpu_cnt;
74 struct percpu_counter cpu_cnt[BLKG_RWSTAT_NR];
75 atomic64_t aux_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 /* request allocation list for this blkcg-q pair */
126 struct request_list rl;
128 /* reference count */
131 /* is this blkg online? protected by both blkcg and q locks */
134 struct blkg_rwstat stat_bytes;
135 struct blkg_rwstat stat_ios;
137 struct blkg_policy_data *pd[BLKCG_MAX_POLS];
139 struct rcu_head rcu_head;
142 atomic64_t delay_nsec;
143 atomic64_t delay_start;
148 typedef struct blkcg_policy_data *(blkcg_pol_alloc_cpd_fn)(gfp_t gfp);
149 typedef void (blkcg_pol_init_cpd_fn)(struct blkcg_policy_data *cpd);
150 typedef void (blkcg_pol_free_cpd_fn)(struct blkcg_policy_data *cpd);
151 typedef void (blkcg_pol_bind_cpd_fn)(struct blkcg_policy_data *cpd);
152 typedef struct blkg_policy_data *(blkcg_pol_alloc_pd_fn)(gfp_t gfp, int node);
153 typedef void (blkcg_pol_init_pd_fn)(struct blkg_policy_data *pd);
154 typedef void (blkcg_pol_online_pd_fn)(struct blkg_policy_data *pd);
155 typedef void (blkcg_pol_offline_pd_fn)(struct blkg_policy_data *pd);
156 typedef void (blkcg_pol_free_pd_fn)(struct blkg_policy_data *pd);
157 typedef void (blkcg_pol_reset_pd_stats_fn)(struct blkg_policy_data *pd);
158 typedef size_t (blkcg_pol_stat_pd_fn)(struct blkg_policy_data *pd, char *buf,
161 struct blkcg_policy {
163 /* cgroup files for the policy */
164 struct cftype *dfl_cftypes;
165 struct cftype *legacy_cftypes;
168 blkcg_pol_alloc_cpd_fn *cpd_alloc_fn;
169 blkcg_pol_init_cpd_fn *cpd_init_fn;
170 blkcg_pol_free_cpd_fn *cpd_free_fn;
171 blkcg_pol_bind_cpd_fn *cpd_bind_fn;
173 blkcg_pol_alloc_pd_fn *pd_alloc_fn;
174 blkcg_pol_init_pd_fn *pd_init_fn;
175 blkcg_pol_online_pd_fn *pd_online_fn;
176 blkcg_pol_offline_pd_fn *pd_offline_fn;
177 blkcg_pol_free_pd_fn *pd_free_fn;
178 blkcg_pol_reset_pd_stats_fn *pd_reset_stats_fn;
179 blkcg_pol_stat_pd_fn *pd_stat_fn;
182 extern struct blkcg blkcg_root;
183 extern struct cgroup_subsys_state * const blkcg_root_css;
185 struct blkcg_gq *blkg_lookup_slowpath(struct blkcg *blkcg,
186 struct request_queue *q, bool update_hint);
187 struct blkcg_gq *__blkg_lookup_create(struct blkcg *blkcg,
188 struct request_queue *q);
189 struct blkcg_gq *blkg_lookup_create(struct blkcg *blkcg,
190 struct request_queue *q);
191 int blkcg_init_queue(struct request_queue *q);
192 void blkcg_drain_queue(struct request_queue *q);
193 void blkcg_exit_queue(struct request_queue *q);
195 /* Blkio controller policy registration */
196 int blkcg_policy_register(struct blkcg_policy *pol);
197 void blkcg_policy_unregister(struct blkcg_policy *pol);
198 int blkcg_activate_policy(struct request_queue *q,
199 const struct blkcg_policy *pol);
200 void blkcg_deactivate_policy(struct request_queue *q,
201 const struct blkcg_policy *pol);
203 const char *blkg_dev_name(struct blkcg_gq *blkg);
204 void blkcg_print_blkgs(struct seq_file *sf, struct blkcg *blkcg,
205 u64 (*prfill)(struct seq_file *,
206 struct blkg_policy_data *, int),
207 const struct blkcg_policy *pol, int data,
209 u64 __blkg_prfill_u64(struct seq_file *sf, struct blkg_policy_data *pd, u64 v);
210 u64 __blkg_prfill_rwstat(struct seq_file *sf, struct blkg_policy_data *pd,
211 const struct blkg_rwstat *rwstat);
212 u64 blkg_prfill_stat(struct seq_file *sf, struct blkg_policy_data *pd, int off);
213 u64 blkg_prfill_rwstat(struct seq_file *sf, struct blkg_policy_data *pd,
215 int blkg_print_stat_bytes(struct seq_file *sf, void *v);
216 int blkg_print_stat_ios(struct seq_file *sf, void *v);
217 int blkg_print_stat_bytes_recursive(struct seq_file *sf, void *v);
218 int blkg_print_stat_ios_recursive(struct seq_file *sf, void *v);
220 u64 blkg_stat_recursive_sum(struct blkcg_gq *blkg,
221 struct blkcg_policy *pol, int off);
222 struct blkg_rwstat blkg_rwstat_recursive_sum(struct blkcg_gq *blkg,
223 struct blkcg_policy *pol, int off);
225 struct blkg_conf_ctx {
226 struct gendisk *disk;
227 struct blkcg_gq *blkg;
231 int blkg_conf_prep(struct blkcg *blkcg, const struct blkcg_policy *pol,
232 char *input, struct blkg_conf_ctx *ctx);
233 void blkg_conf_finish(struct blkg_conf_ctx *ctx);
236 * blkcg_css - find the current css
238 * Find the css associated with either the kthread or the current task.
239 * This may return a dying css, so it is up to the caller to use tryget logic
240 * to confirm it is alive and well.
242 static inline struct cgroup_subsys_state *blkcg_css(void)
244 struct cgroup_subsys_state *css;
246 css = kthread_blkcg();
249 return task_css(current, io_cgrp_id);
252 static inline struct blkcg *css_to_blkcg(struct cgroup_subsys_state *css)
254 return css ? container_of(css, struct blkcg, css) : NULL;
258 * __bio_blkcg - internal version of bio_blkcg for bfq and cfq
261 * There is a flaw using this version of the function. In particular, this was
262 * used in a broken paradigm where association was called on the given css. It
263 * is possible though that the returned css from task_css() is in the process
264 * of dying due to migration of the current task. So it is improper to assume
265 * *_get() is going to succeed. Both BFQ and CFQ rely on this logic and will
266 * take additional work to handle more gracefully.
268 static inline struct blkcg *__bio_blkcg(struct bio *bio)
270 if (bio && bio->bi_blkg)
271 return bio->bi_blkg->blkcg;
272 return css_to_blkcg(blkcg_css());
276 * bio_blkcg - grab the blkcg associated with a bio
279 * This returns the blkcg associated with a bio, NULL if not associated.
280 * Callers are expected to either handle NULL or know association has been
281 * done prior to calling this.
283 static inline struct blkcg *bio_blkcg(struct bio *bio)
285 if (bio && bio->bi_blkg)
286 return bio->bi_blkg->blkcg;
290 static inline bool blk_cgroup_congested(void)
292 struct cgroup_subsys_state *css;
296 css = kthread_blkcg();
298 css = task_css(current, io_cgrp_id);
300 if (atomic_read(&css->cgroup->congestion_count)) {
311 * bio_issue_as_root_blkg - see if this bio needs to be issued as root blkg
312 * @return: true if this bio needs to be submitted with the root blkg context.
314 * In order to avoid priority inversions we sometimes need to issue a bio as if
315 * it were attached to the root blkg, and then backcharge to the actual owning
316 * blkg. The idea is we do bio_blkcg() to look up the actual context for the
317 * bio and attach the appropriate blkg to the bio. Then we call this helper and
318 * if it is true run with the root blkg for that queue and then do any
319 * backcharging to the originating cgroup once the io is complete.
321 static inline bool bio_issue_as_root_blkg(struct bio *bio)
323 return (bio->bi_opf & (REQ_META | REQ_SWAP)) != 0;
327 * blkcg_parent - get the parent of a blkcg
328 * @blkcg: blkcg of interest
330 * Return the parent blkcg of @blkcg. Can be called anytime.
332 static inline struct blkcg *blkcg_parent(struct blkcg *blkcg)
334 return css_to_blkcg(blkcg->css.parent);
338 * __blkg_lookup - internal version of blkg_lookup()
339 * @blkcg: blkcg of interest
340 * @q: request_queue of interest
341 * @update_hint: whether to update lookup hint with the result or not
343 * This is internal version and shouldn't be used by policy
344 * implementations. Looks up blkgs for the @blkcg - @q pair regardless of
345 * @q's bypass state. If @update_hint is %true, the caller should be
346 * holding @q->queue_lock and lookup hint is updated on success.
348 static inline struct blkcg_gq *__blkg_lookup(struct blkcg *blkcg,
349 struct request_queue *q,
352 struct blkcg_gq *blkg;
354 if (blkcg == &blkcg_root)
357 blkg = rcu_dereference(blkcg->blkg_hint);
358 if (blkg && blkg->q == q)
361 return blkg_lookup_slowpath(blkcg, q, update_hint);
365 * blkg_lookup - lookup blkg for the specified blkcg - q pair
366 * @blkcg: blkcg of interest
367 * @q: request_queue of interest
369 * Lookup blkg for the @blkcg - @q pair. This function should be called
370 * under RCU read lock and is guaranteed to return %NULL if @q is bypassing
371 * - see blk_queue_bypass_start() for details.
373 static inline struct blkcg_gq *blkg_lookup(struct blkcg *blkcg,
374 struct request_queue *q)
376 WARN_ON_ONCE(!rcu_read_lock_held());
378 if (unlikely(blk_queue_bypass(q)))
380 return __blkg_lookup(blkcg, q, false);
384 * blk_queue_root_blkg - return blkg for the (blkcg_root, @q) pair
385 * @q: request_queue of interest
387 * Lookup blkg for @q at the root level. See also blkg_lookup().
389 static inline struct blkcg_gq *blk_queue_root_blkg(struct request_queue *q)
395 * blkg_to_pdata - get policy private data
396 * @blkg: blkg of interest
397 * @pol: policy of interest
399 * Return pointer to private data associated with the @blkg-@pol pair.
401 static inline struct blkg_policy_data *blkg_to_pd(struct blkcg_gq *blkg,
402 struct blkcg_policy *pol)
404 return blkg ? blkg->pd[pol->plid] : NULL;
407 static inline struct blkcg_policy_data *blkcg_to_cpd(struct blkcg *blkcg,
408 struct blkcg_policy *pol)
410 return blkcg ? blkcg->cpd[pol->plid] : NULL;
414 * pdata_to_blkg - get blkg associated with policy private data
415 * @pd: policy private data of interest
417 * @pd is policy private data. Determine the blkg it's associated with.
419 static inline struct blkcg_gq *pd_to_blkg(struct blkg_policy_data *pd)
421 return pd ? pd->blkg : NULL;
424 static inline struct blkcg *cpd_to_blkcg(struct blkcg_policy_data *cpd)
426 return cpd ? cpd->blkcg : NULL;
429 extern void blkcg_destroy_blkgs(struct blkcg *blkcg);
431 #ifdef CONFIG_CGROUP_WRITEBACK
434 * blkcg_cgwb_get - get a reference for blkcg->cgwb_list
435 * @blkcg: blkcg of interest
437 * This is used to track the number of active wb's related to a blkcg.
439 static inline void blkcg_cgwb_get(struct blkcg *blkcg)
441 refcount_inc(&blkcg->cgwb_refcnt);
445 * blkcg_cgwb_put - put a reference for @blkcg->cgwb_list
446 * @blkcg: blkcg of interest
448 * This is used to track the number of active wb's related to a blkcg.
449 * When this count goes to zero, all active wb has finished so the
450 * blkcg can continue destruction by calling blkcg_destroy_blkgs().
451 * This work may occur in cgwb_release_workfn() on the cgwb_release
454 static inline void blkcg_cgwb_put(struct blkcg *blkcg)
456 if (refcount_dec_and_test(&blkcg->cgwb_refcnt))
457 blkcg_destroy_blkgs(blkcg);
462 static inline void blkcg_cgwb_get(struct blkcg *blkcg) { }
464 static inline void blkcg_cgwb_put(struct blkcg *blkcg)
466 /* wb isn't being accounted, so trigger destruction right away */
467 blkcg_destroy_blkgs(blkcg);
473 * blkg_path - format cgroup path of blkg
474 * @blkg: blkg of interest
475 * @buf: target buffer
476 * @buflen: target buffer length
478 * Format the path of the cgroup of @blkg into @buf.
480 static inline int blkg_path(struct blkcg_gq *blkg, char *buf, int buflen)
482 return cgroup_path(blkg->blkcg->css.cgroup, buf, buflen);
486 * blkg_get - get a blkg reference
489 * The caller should be holding an existing reference.
491 static inline void blkg_get(struct blkcg_gq *blkg)
493 WARN_ON_ONCE(atomic_read(&blkg->refcnt) <= 0);
494 atomic_inc(&blkg->refcnt);
498 * blkg_try_get - try and get a blkg reference
501 * This is for use when doing an RCU lookup of the blkg. We may be in the midst
502 * of freeing this blkg, so we can only use it if the refcnt is not zero.
504 static inline struct blkcg_gq *blkg_try_get(struct blkcg_gq *blkg)
506 if (atomic_inc_not_zero(&blkg->refcnt))
512 * blkg_try_get_closest - try and get a blkg ref on the closet blkg
515 * This walks up the blkg tree to find the closest non-dying blkg and returns
516 * the blkg that it did association with as it may not be the passed in blkg.
518 static inline struct blkcg_gq *blkg_try_get_closest(struct blkcg_gq *blkg)
520 while (!atomic_inc_not_zero(&blkg->refcnt))
526 void __blkg_release_rcu(struct rcu_head *rcu);
529 * blkg_put - put a blkg reference
532 static inline void blkg_put(struct blkcg_gq *blkg)
534 WARN_ON_ONCE(atomic_read(&blkg->refcnt) <= 0);
535 if (atomic_dec_and_test(&blkg->refcnt))
536 call_rcu(&blkg->rcu_head, __blkg_release_rcu);
540 * blkg_for_each_descendant_pre - pre-order walk of a blkg's descendants
541 * @d_blkg: loop cursor pointing to the current descendant
542 * @pos_css: used for iteration
543 * @p_blkg: target blkg to walk descendants of
545 * Walk @c_blkg through the descendants of @p_blkg. Must be used with RCU
546 * read locked. If called under either blkcg or queue lock, the iteration
547 * is guaranteed to include all and only online blkgs. The caller may
548 * update @pos_css by calling css_rightmost_descendant() to skip subtree.
549 * @p_blkg is included in the iteration and the first node to be visited.
551 #define blkg_for_each_descendant_pre(d_blkg, pos_css, p_blkg) \
552 css_for_each_descendant_pre((pos_css), &(p_blkg)->blkcg->css) \
553 if (((d_blkg) = __blkg_lookup(css_to_blkcg(pos_css), \
554 (p_blkg)->q, false)))
557 * blkg_for_each_descendant_post - post-order walk of a blkg's descendants
558 * @d_blkg: loop cursor pointing to the current descendant
559 * @pos_css: used for iteration
560 * @p_blkg: target blkg to walk descendants of
562 * Similar to blkg_for_each_descendant_pre() but performs post-order
563 * traversal instead. Synchronization rules are the same. @p_blkg is
564 * included in the iteration and the last node to be visited.
566 #define blkg_for_each_descendant_post(d_blkg, pos_css, p_blkg) \
567 css_for_each_descendant_post((pos_css), &(p_blkg)->blkcg->css) \
568 if (((d_blkg) = __blkg_lookup(css_to_blkcg(pos_css), \
569 (p_blkg)->q, false)))
572 * blk_get_rl - get request_list to use
573 * @q: request_queue of interest
574 * @bio: bio which will be attached to the allocated request (may be %NULL)
576 * The caller wants to allocate a request from @q to use for @bio. Find
577 * the request_list to use and obtain a reference on it. Should be called
578 * under queue_lock. This function is guaranteed to return non-%NULL
581 static inline struct request_list *blk_get_rl(struct request_queue *q,
585 struct blkcg_gq *blkg;
589 blkcg = bio_blkcg(bio);
591 blkcg = css_to_blkcg(blkcg_css());
593 /* bypass blkg lookup and use @q->root_rl directly for root */
594 if (blkcg == &blkcg_root)
598 * Try to use blkg->rl. blkg lookup may fail under memory pressure
599 * or if either the blkcg or queue is going away. Fall back to
600 * root_rl in such cases.
602 blkg = blkg_lookup(blkcg, q);
606 if (!blkg_try_get(blkg))
616 * blk_put_rl - put request_list
617 * @rl: request_list to put
619 * Put the reference acquired by blk_get_rl(). Should be called under
622 static inline void blk_put_rl(struct request_list *rl)
624 if (rl->blkg->blkcg != &blkcg_root)
629 * blk_rq_set_rl - associate a request with a request_list
630 * @rq: request of interest
631 * @rl: target request_list
633 * Associate @rq with @rl so that accounting and freeing can know the
634 * request_list @rq came from.
636 static inline void blk_rq_set_rl(struct request *rq, struct request_list *rl)
642 * blk_rq_rl - return the request_list a request came from
643 * @rq: request of interest
645 * Return the request_list @rq is allocated from.
647 static inline struct request_list *blk_rq_rl(struct request *rq)
652 struct request_list *__blk_queue_next_rl(struct request_list *rl,
653 struct request_queue *q);
655 * blk_queue_for_each_rl - iterate through all request_lists of a request_queue
657 * Should be used under queue_lock.
659 #define blk_queue_for_each_rl(rl, q) \
660 for ((rl) = &(q)->root_rl; (rl); (rl) = __blk_queue_next_rl((rl), (q)))
662 static inline int blkg_stat_init(struct blkg_stat *stat, gfp_t gfp)
666 ret = percpu_counter_init(&stat->cpu_cnt, 0, gfp);
670 atomic64_set(&stat->aux_cnt, 0);
674 static inline void blkg_stat_exit(struct blkg_stat *stat)
676 percpu_counter_destroy(&stat->cpu_cnt);
680 * blkg_stat_add - add a value to a blkg_stat
681 * @stat: target blkg_stat
684 * Add @val to @stat. The caller must ensure that IRQ on the same CPU
685 * don't re-enter this function for the same counter.
687 static inline void blkg_stat_add(struct blkg_stat *stat, uint64_t val)
689 percpu_counter_add_batch(&stat->cpu_cnt, val, BLKG_STAT_CPU_BATCH);
693 * blkg_stat_read - read the current value of a blkg_stat
694 * @stat: blkg_stat to read
696 static inline uint64_t blkg_stat_read(struct blkg_stat *stat)
698 return percpu_counter_sum_positive(&stat->cpu_cnt);
702 * blkg_stat_reset - reset a blkg_stat
703 * @stat: blkg_stat to reset
705 static inline void blkg_stat_reset(struct blkg_stat *stat)
707 percpu_counter_set(&stat->cpu_cnt, 0);
708 atomic64_set(&stat->aux_cnt, 0);
712 * blkg_stat_add_aux - add a blkg_stat into another's aux count
713 * @to: the destination blkg_stat
716 * Add @from's count including the aux one to @to's aux count.
718 static inline void blkg_stat_add_aux(struct blkg_stat *to,
719 struct blkg_stat *from)
721 atomic64_add(blkg_stat_read(from) + atomic64_read(&from->aux_cnt),
725 static inline int blkg_rwstat_init(struct blkg_rwstat *rwstat, gfp_t gfp)
729 for (i = 0; i < BLKG_RWSTAT_NR; i++) {
730 ret = percpu_counter_init(&rwstat->cpu_cnt[i], 0, gfp);
733 percpu_counter_destroy(&rwstat->cpu_cnt[i]);
736 atomic64_set(&rwstat->aux_cnt[i], 0);
741 static inline void blkg_rwstat_exit(struct blkg_rwstat *rwstat)
745 for (i = 0; i < BLKG_RWSTAT_NR; i++)
746 percpu_counter_destroy(&rwstat->cpu_cnt[i]);
750 * blkg_rwstat_add - add a value to a blkg_rwstat
751 * @rwstat: target blkg_rwstat
752 * @op: REQ_OP and flags
755 * Add @val to @rwstat. The counters are chosen according to @rw. The
756 * caller is responsible for synchronizing calls to this function.
758 static inline void blkg_rwstat_add(struct blkg_rwstat *rwstat,
759 unsigned int op, uint64_t val)
761 struct percpu_counter *cnt;
763 if (op_is_discard(op))
764 cnt = &rwstat->cpu_cnt[BLKG_RWSTAT_DISCARD];
765 else if (op_is_write(op))
766 cnt = &rwstat->cpu_cnt[BLKG_RWSTAT_WRITE];
768 cnt = &rwstat->cpu_cnt[BLKG_RWSTAT_READ];
770 percpu_counter_add_batch(cnt, val, BLKG_STAT_CPU_BATCH);
773 cnt = &rwstat->cpu_cnt[BLKG_RWSTAT_SYNC];
775 cnt = &rwstat->cpu_cnt[BLKG_RWSTAT_ASYNC];
777 percpu_counter_add_batch(cnt, val, BLKG_STAT_CPU_BATCH);
781 * blkg_rwstat_read - read the current values of a blkg_rwstat
782 * @rwstat: blkg_rwstat to read
784 * Read the current snapshot of @rwstat and return it in the aux counts.
786 static inline struct blkg_rwstat blkg_rwstat_read(struct blkg_rwstat *rwstat)
788 struct blkg_rwstat result;
791 for (i = 0; i < BLKG_RWSTAT_NR; i++)
792 atomic64_set(&result.aux_cnt[i],
793 percpu_counter_sum_positive(&rwstat->cpu_cnt[i]));
798 * blkg_rwstat_total - read the total count of a blkg_rwstat
799 * @rwstat: blkg_rwstat to read
801 * Return the total count of @rwstat regardless of the IO direction. This
802 * function can be called without synchronization and takes care of u64
805 static inline uint64_t blkg_rwstat_total(struct blkg_rwstat *rwstat)
807 struct blkg_rwstat tmp = blkg_rwstat_read(rwstat);
809 return atomic64_read(&tmp.aux_cnt[BLKG_RWSTAT_READ]) +
810 atomic64_read(&tmp.aux_cnt[BLKG_RWSTAT_WRITE]);
814 * blkg_rwstat_reset - reset a blkg_rwstat
815 * @rwstat: blkg_rwstat to reset
817 static inline void blkg_rwstat_reset(struct blkg_rwstat *rwstat)
821 for (i = 0; i < BLKG_RWSTAT_NR; i++) {
822 percpu_counter_set(&rwstat->cpu_cnt[i], 0);
823 atomic64_set(&rwstat->aux_cnt[i], 0);
828 * blkg_rwstat_add_aux - add a blkg_rwstat into another's aux count
829 * @to: the destination blkg_rwstat
832 * Add @from's count including the aux one to @to's aux count.
834 static inline void blkg_rwstat_add_aux(struct blkg_rwstat *to,
835 struct blkg_rwstat *from)
837 u64 sum[BLKG_RWSTAT_NR];
840 for (i = 0; i < BLKG_RWSTAT_NR; i++)
841 sum[i] = percpu_counter_sum_positive(&from->cpu_cnt[i]);
843 for (i = 0; i < BLKG_RWSTAT_NR; i++)
844 atomic64_add(sum[i] + atomic64_read(&from->aux_cnt[i]),
848 #ifdef CONFIG_BLK_DEV_THROTTLING
849 extern bool blk_throtl_bio(struct request_queue *q, struct blkcg_gq *blkg,
852 static inline bool blk_throtl_bio(struct request_queue *q, struct blkcg_gq *blkg,
853 struct bio *bio) { return false; }
857 static inline void blkcg_bio_issue_init(struct bio *bio)
859 bio_issue_init(&bio->bi_issue, bio_sectors(bio));
862 static inline bool blkcg_bio_issue_check(struct request_queue *q,
865 struct blkcg_gq *blkg;
870 bio_associate_create_blkg(q, bio);
873 throtl = blk_throtl_bio(q, blkg, bio);
877 * If the bio is flagged with BIO_QUEUE_ENTERED it means this
878 * is a split bio and we would have already accounted for the
881 if (!bio_flagged(bio, BIO_QUEUE_ENTERED))
882 blkg_rwstat_add(&blkg->stat_bytes, bio->bi_opf,
883 bio->bi_iter.bi_size);
884 blkg_rwstat_add(&blkg->stat_ios, bio->bi_opf, 1);
887 blkcg_bio_issue_init(bio);
893 static inline void blkcg_use_delay(struct blkcg_gq *blkg)
895 if (atomic_add_return(1, &blkg->use_delay) == 1)
896 atomic_inc(&blkg->blkcg->css.cgroup->congestion_count);
899 static inline int blkcg_unuse_delay(struct blkcg_gq *blkg)
901 int old = atomic_read(&blkg->use_delay);
907 * We do this song and dance because we can race with somebody else
908 * adding or removing delay. If we just did an atomic_dec we'd end up
909 * negative and we'd already be in trouble. We need to subtract 1 and
910 * then check to see if we were the last delay so we can drop the
911 * congestion count on the cgroup.
914 int cur = atomic_cmpxchg(&blkg->use_delay, old, old - 1);
923 atomic_dec(&blkg->blkcg->css.cgroup->congestion_count);
927 static inline void blkcg_clear_delay(struct blkcg_gq *blkg)
929 int old = atomic_read(&blkg->use_delay);
932 /* We only want 1 person clearing the congestion count for this blkg. */
934 int cur = atomic_cmpxchg(&blkg->use_delay, old, 0);
936 atomic_dec(&blkg->blkcg->css.cgroup->congestion_count);
943 void blkcg_add_delay(struct blkcg_gq *blkg, u64 now, u64 delta);
944 void blkcg_schedule_throttle(struct request_queue *q, bool use_memdelay);
945 void blkcg_maybe_throttle_current(void);
946 #else /* CONFIG_BLK_CGROUP */
951 struct blkg_policy_data {
954 struct blkcg_policy_data {
960 struct blkcg_policy {
963 #define blkcg_root_css ((struct cgroup_subsys_state *)ERR_PTR(-EINVAL))
965 static inline void blkcg_maybe_throttle_current(void) { }
966 static inline bool blk_cgroup_congested(void) { return false; }
970 static inline void blkcg_schedule_throttle(struct request_queue *q, bool use_memdelay) { }
972 static inline struct blkcg_gq *blkg_lookup(struct blkcg *blkcg, void *key) { return NULL; }
973 static inline struct blkcg_gq *blk_queue_root_blkg(struct request_queue *q)
975 static inline int blkcg_init_queue(struct request_queue *q) { return 0; }
976 static inline void blkcg_drain_queue(struct request_queue *q) { }
977 static inline void blkcg_exit_queue(struct request_queue *q) { }
978 static inline int blkcg_policy_register(struct blkcg_policy *pol) { return 0; }
979 static inline void blkcg_policy_unregister(struct blkcg_policy *pol) { }
980 static inline int blkcg_activate_policy(struct request_queue *q,
981 const struct blkcg_policy *pol) { return 0; }
982 static inline void blkcg_deactivate_policy(struct request_queue *q,
983 const struct blkcg_policy *pol) { }
985 static inline struct blkcg *__bio_blkcg(struct bio *bio) { return NULL; }
986 static inline struct blkcg *bio_blkcg(struct bio *bio) { return NULL; }
988 static inline struct blkg_policy_data *blkg_to_pd(struct blkcg_gq *blkg,
989 struct blkcg_policy *pol) { return NULL; }
990 static inline struct blkcg_gq *pd_to_blkg(struct blkg_policy_data *pd) { return NULL; }
991 static inline char *blkg_path(struct blkcg_gq *blkg) { return NULL; }
992 static inline void blkg_get(struct blkcg_gq *blkg) { }
993 static inline void blkg_put(struct blkcg_gq *blkg) { }
995 static inline struct request_list *blk_get_rl(struct request_queue *q,
996 struct bio *bio) { return &q->root_rl; }
997 static inline void blk_put_rl(struct request_list *rl) { }
998 static inline void blk_rq_set_rl(struct request *rq, struct request_list *rl) { }
999 static inline struct request_list *blk_rq_rl(struct request *rq) { return &rq->q->root_rl; }
1001 static inline void blkcg_bio_issue_init(struct bio *bio) { }
1002 static inline bool blkcg_bio_issue_check(struct request_queue *q,
1003 struct bio *bio) { return true; }
1005 #define blk_queue_for_each_rl(rl, q) \
1006 for ((rl) = &(q)->root_rl; (rl); (rl) = NULL)
1008 #endif /* CONFIG_BLOCK */
1009 #endif /* CONFIG_BLK_CGROUP */
1010 #endif /* _BLK_CGROUP_H */