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>
24 #include <linux/blkdev.h>
26 /* percpu_counter batch for blkg_[rw]stats, per-cpu drift doesn't matter */
27 #define BLKG_STAT_CPU_BATCH (INT_MAX / 2)
29 /* Max limits for throttle policy */
30 #define THROTL_IOPS_MAX UINT_MAX
32 #ifdef CONFIG_BLK_CGROUP
34 enum blkg_rwstat_type {
42 BLKG_RWSTAT_TOTAL = BLKG_RWSTAT_NR,
48 struct cgroup_subsys_state css;
51 struct radix_tree_root blkg_tree;
52 struct blkcg_gq __rcu *blkg_hint;
53 struct hlist_head blkg_list;
55 struct blkcg_policy_data *cpd[BLKCG_MAX_POLS];
57 struct list_head all_blkcgs_node;
58 #ifdef CONFIG_CGROUP_WRITEBACK
59 struct list_head cgwb_list;
60 refcount_t cgwb_refcnt;
65 * blkg_[rw]stat->aux_cnt is excluded for local stats but included for
66 * recursive. Used to carry stats of dead children, and, for blkg_rwstat,
67 * to carry result values from read and sum operations.
70 struct percpu_counter cpu_cnt;
75 struct percpu_counter cpu_cnt[BLKG_RWSTAT_NR];
76 atomic64_t aux_cnt[BLKG_RWSTAT_NR];
80 * A blkcg_gq (blkg) is association between a block cgroup (blkcg) and a
81 * request_queue (q). This is used by blkcg policies which need to track
82 * information per blkcg - q pair.
84 * There can be multiple active blkcg policies and each blkg:policy pair is
85 * represented by a blkg_policy_data which is allocated and freed by each
86 * policy's pd_alloc/free_fn() methods. A policy can allocate private data
87 * area by allocating larger data structure which embeds blkg_policy_data
90 struct blkg_policy_data {
91 /* the blkg and policy id this per-policy data belongs to */
92 struct blkcg_gq *blkg;
97 * Policies that need to keep per-blkcg data which is independent from any
98 * request_queue associated to it should implement cpd_alloc/free_fn()
99 * methods. A policy can allocate private data area by allocating larger
100 * data structure which embeds blkcg_policy_data at the beginning.
101 * cpd_init() is invoked to let each policy handle per-blkcg data.
103 struct blkcg_policy_data {
104 /* the blkcg and policy id this per-policy data belongs to */
109 /* association between a blk cgroup and a request queue */
111 /* Pointer to the associated request_queue */
112 struct request_queue *q;
113 struct list_head q_node;
114 struct hlist_node blkcg_node;
118 * Each blkg gets congested separately and the congestion state is
119 * propagated to the matching bdi_writeback_congested.
121 struct bdi_writeback_congested *wb_congested;
123 /* all non-root blkcg_gq's are guaranteed to have access to parent */
124 struct blkcg_gq *parent;
126 /* request allocation list for this blkcg-q pair */
127 struct request_list rl;
129 /* reference count */
132 /* is this blkg online? protected by both blkcg and q locks */
135 struct blkg_rwstat stat_bytes;
136 struct blkg_rwstat stat_ios;
138 struct blkg_policy_data *pd[BLKCG_MAX_POLS];
140 struct rcu_head rcu_head;
143 atomic64_t delay_nsec;
144 atomic64_t delay_start;
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, int node);
154 typedef void (blkcg_pol_init_pd_fn)(struct blkg_policy_data *pd);
155 typedef void (blkcg_pol_online_pd_fn)(struct blkg_policy_data *pd);
156 typedef void (blkcg_pol_offline_pd_fn)(struct blkg_policy_data *pd);
157 typedef void (blkcg_pol_free_pd_fn)(struct blkg_policy_data *pd);
158 typedef void (blkcg_pol_reset_pd_stats_fn)(struct blkg_policy_data *pd);
159 typedef size_t (blkcg_pol_stat_pd_fn)(struct blkg_policy_data *pd, char *buf,
162 struct blkcg_policy {
164 /* cgroup files for the policy */
165 struct cftype *dfl_cftypes;
166 struct cftype *legacy_cftypes;
169 blkcg_pol_alloc_cpd_fn *cpd_alloc_fn;
170 blkcg_pol_init_cpd_fn *cpd_init_fn;
171 blkcg_pol_free_cpd_fn *cpd_free_fn;
172 blkcg_pol_bind_cpd_fn *cpd_bind_fn;
174 blkcg_pol_alloc_pd_fn *pd_alloc_fn;
175 blkcg_pol_init_pd_fn *pd_init_fn;
176 blkcg_pol_online_pd_fn *pd_online_fn;
177 blkcg_pol_offline_pd_fn *pd_offline_fn;
178 blkcg_pol_free_pd_fn *pd_free_fn;
179 blkcg_pol_reset_pd_stats_fn *pd_reset_stats_fn;
180 blkcg_pol_stat_pd_fn *pd_stat_fn;
183 extern struct blkcg blkcg_root;
184 extern struct cgroup_subsys_state * const blkcg_root_css;
186 struct blkcg_gq *blkg_lookup_slowpath(struct blkcg *blkcg,
187 struct request_queue *q, bool update_hint);
188 struct blkcg_gq *blkg_lookup_create(struct blkcg *blkcg,
189 struct request_queue *q);
190 int blkcg_init_queue(struct request_queue *q);
191 void blkcg_drain_queue(struct request_queue *q);
192 void blkcg_exit_queue(struct request_queue *q);
194 /* Blkio controller policy registration */
195 int blkcg_policy_register(struct blkcg_policy *pol);
196 void blkcg_policy_unregister(struct blkcg_policy *pol);
197 int blkcg_activate_policy(struct request_queue *q,
198 const struct blkcg_policy *pol);
199 void blkcg_deactivate_policy(struct request_queue *q,
200 const struct blkcg_policy *pol);
202 const char *blkg_dev_name(struct blkcg_gq *blkg);
203 void blkcg_print_blkgs(struct seq_file *sf, struct blkcg *blkcg,
204 u64 (*prfill)(struct seq_file *,
205 struct blkg_policy_data *, int),
206 const struct blkcg_policy *pol, int data,
208 u64 __blkg_prfill_u64(struct seq_file *sf, struct blkg_policy_data *pd, u64 v);
209 u64 __blkg_prfill_rwstat(struct seq_file *sf, struct blkg_policy_data *pd,
210 const struct blkg_rwstat *rwstat);
211 u64 blkg_prfill_stat(struct seq_file *sf, struct blkg_policy_data *pd, int off);
212 u64 blkg_prfill_rwstat(struct seq_file *sf, struct blkg_policy_data *pd,
214 int blkg_print_stat_bytes(struct seq_file *sf, void *v);
215 int blkg_print_stat_ios(struct seq_file *sf, void *v);
216 int blkg_print_stat_bytes_recursive(struct seq_file *sf, void *v);
217 int blkg_print_stat_ios_recursive(struct seq_file *sf, void *v);
219 u64 blkg_stat_recursive_sum(struct blkcg_gq *blkg,
220 struct blkcg_policy *pol, int off);
221 struct blkg_rwstat blkg_rwstat_recursive_sum(struct blkcg_gq *blkg,
222 struct blkcg_policy *pol, int off);
224 struct blkg_conf_ctx {
225 struct gendisk *disk;
226 struct blkcg_gq *blkg;
230 int blkg_conf_prep(struct blkcg *blkcg, const struct blkcg_policy *pol,
231 char *input, struct blkg_conf_ctx *ctx);
232 void blkg_conf_finish(struct blkg_conf_ctx *ctx);
235 static inline struct blkcg *css_to_blkcg(struct cgroup_subsys_state *css)
237 return css ? container_of(css, struct blkcg, css) : NULL;
240 static inline struct blkcg *bio_blkcg(struct bio *bio)
242 struct cgroup_subsys_state *css;
244 if (bio && bio->bi_css)
245 return css_to_blkcg(bio->bi_css);
246 css = kthread_blkcg();
248 return css_to_blkcg(css);
249 return css_to_blkcg(task_css(current, io_cgrp_id));
252 static inline bool blk_cgroup_congested(void)
254 struct cgroup_subsys_state *css;
258 css = kthread_blkcg();
260 css = task_css(current, io_cgrp_id);
262 if (atomic_read(&css->cgroup->congestion_count)) {
273 * bio_issue_as_root_blkg - see if this bio needs to be issued as root blkg
274 * @return: true if this bio needs to be submitted with the root blkg context.
276 * In order to avoid priority inversions we sometimes need to issue a bio as if
277 * it were attached to the root blkg, and then backcharge to the actual owning
278 * blkg. The idea is we do bio_blkcg() to look up the actual context for the
279 * bio and attach the appropriate blkg to the bio. Then we call this helper and
280 * if it is true run with the root blkg for that queue and then do any
281 * backcharging to the originating cgroup once the io is complete.
283 static inline bool bio_issue_as_root_blkg(struct bio *bio)
285 return (bio->bi_opf & (REQ_META | REQ_SWAP)) != 0;
289 * blkcg_parent - get the parent of a blkcg
290 * @blkcg: blkcg of interest
292 * Return the parent blkcg of @blkcg. Can be called anytime.
294 static inline struct blkcg *blkcg_parent(struct blkcg *blkcg)
296 return css_to_blkcg(blkcg->css.parent);
300 * __blkg_lookup - internal version of blkg_lookup()
301 * @blkcg: blkcg of interest
302 * @q: request_queue of interest
303 * @update_hint: whether to update lookup hint with the result or not
305 * This is internal version and shouldn't be used by policy
306 * implementations. Looks up blkgs for the @blkcg - @q pair regardless of
307 * @q's bypass state. If @update_hint is %true, the caller should be
308 * holding @q->queue_lock and lookup hint is updated on success.
310 static inline struct blkcg_gq *__blkg_lookup(struct blkcg *blkcg,
311 struct request_queue *q,
314 struct blkcg_gq *blkg;
316 if (blkcg == &blkcg_root)
319 blkg = rcu_dereference(blkcg->blkg_hint);
320 if (blkg && blkg->q == q)
323 return blkg_lookup_slowpath(blkcg, q, update_hint);
327 * blkg_lookup - lookup blkg for the specified blkcg - q pair
328 * @blkcg: blkcg of interest
329 * @q: request_queue of interest
331 * Lookup blkg for the @blkcg - @q pair. This function should be called
332 * under RCU read lock and is guaranteed to return %NULL if @q is bypassing
333 * - see blk_queue_bypass_start() for details.
335 static inline struct blkcg_gq *blkg_lookup(struct blkcg *blkcg,
336 struct request_queue *q)
338 WARN_ON_ONCE(!rcu_read_lock_held());
340 if (unlikely(blk_queue_bypass(q)))
342 return __blkg_lookup(blkcg, q, false);
346 * blk_queue_root_blkg - return blkg for the (blkcg_root, @q) pair
347 * @q: request_queue of interest
349 * Lookup blkg for @q at the root level. See also blkg_lookup().
351 static inline struct blkcg_gq *blk_queue_root_blkg(struct request_queue *q)
357 * blkg_to_pdata - get policy private data
358 * @blkg: blkg of interest
359 * @pol: policy of interest
361 * Return pointer to private data associated with the @blkg-@pol pair.
363 static inline struct blkg_policy_data *blkg_to_pd(struct blkcg_gq *blkg,
364 struct blkcg_policy *pol)
366 return blkg ? blkg->pd[pol->plid] : NULL;
369 static inline struct blkcg_policy_data *blkcg_to_cpd(struct blkcg *blkcg,
370 struct blkcg_policy *pol)
372 return blkcg ? blkcg->cpd[pol->plid] : NULL;
376 * pdata_to_blkg - get blkg associated with policy private data
377 * @pd: policy private data of interest
379 * @pd is policy private data. Determine the blkg it's associated with.
381 static inline struct blkcg_gq *pd_to_blkg(struct blkg_policy_data *pd)
383 return pd ? pd->blkg : NULL;
386 static inline struct blkcg *cpd_to_blkcg(struct blkcg_policy_data *cpd)
388 return cpd ? cpd->blkcg : NULL;
391 extern void blkcg_destroy_blkgs(struct blkcg *blkcg);
393 #ifdef CONFIG_CGROUP_WRITEBACK
396 * blkcg_cgwb_get - get a reference for blkcg->cgwb_list
397 * @blkcg: blkcg of interest
399 * This is used to track the number of active wb's related to a blkcg.
401 static inline void blkcg_cgwb_get(struct blkcg *blkcg)
403 refcount_inc(&blkcg->cgwb_refcnt);
407 * blkcg_cgwb_put - put a reference for @blkcg->cgwb_list
408 * @blkcg: blkcg of interest
410 * This is used to track the number of active wb's related to a blkcg.
411 * When this count goes to zero, all active wb has finished so the
412 * blkcg can continue destruction by calling blkcg_destroy_blkgs().
413 * This work may occur in cgwb_release_workfn() on the cgwb_release
416 static inline void blkcg_cgwb_put(struct blkcg *blkcg)
418 if (refcount_dec_and_test(&blkcg->cgwb_refcnt))
419 blkcg_destroy_blkgs(blkcg);
424 static inline void blkcg_cgwb_get(struct blkcg *blkcg) { }
426 static inline void blkcg_cgwb_put(struct blkcg *blkcg)
428 /* wb isn't being accounted, so trigger destruction right away */
429 blkcg_destroy_blkgs(blkcg);
435 * blkg_path - format cgroup path of blkg
436 * @blkg: blkg of interest
437 * @buf: target buffer
438 * @buflen: target buffer length
440 * Format the path of the cgroup of @blkg into @buf.
442 static inline int blkg_path(struct blkcg_gq *blkg, char *buf, int buflen)
444 return cgroup_path(blkg->blkcg->css.cgroup, buf, buflen);
448 * blkg_get - get a blkg reference
451 * The caller should be holding an existing reference.
453 static inline void blkg_get(struct blkcg_gq *blkg)
455 WARN_ON_ONCE(atomic_read(&blkg->refcnt) <= 0);
456 atomic_inc(&blkg->refcnt);
460 * blkg_try_get - try and get a blkg reference
463 * This is for use when doing an RCU lookup of the blkg. We may be in the midst
464 * of freeing this blkg, so we can only use it if the refcnt is not zero.
466 static inline struct blkcg_gq *blkg_try_get(struct blkcg_gq *blkg)
468 if (atomic_inc_not_zero(&blkg->refcnt))
474 void __blkg_release_rcu(struct rcu_head *rcu);
477 * blkg_put - put a blkg reference
480 static inline void blkg_put(struct blkcg_gq *blkg)
482 WARN_ON_ONCE(atomic_read(&blkg->refcnt) <= 0);
483 if (atomic_dec_and_test(&blkg->refcnt))
484 call_rcu(&blkg->rcu_head, __blkg_release_rcu);
488 * blkg_for_each_descendant_pre - pre-order walk of a blkg's descendants
489 * @d_blkg: loop cursor pointing to the current descendant
490 * @pos_css: used for iteration
491 * @p_blkg: target blkg to walk descendants of
493 * Walk @c_blkg through the descendants of @p_blkg. Must be used with RCU
494 * read locked. If called under either blkcg or queue lock, the iteration
495 * is guaranteed to include all and only online blkgs. The caller may
496 * update @pos_css by calling css_rightmost_descendant() to skip subtree.
497 * @p_blkg is included in the iteration and the first node to be visited.
499 #define blkg_for_each_descendant_pre(d_blkg, pos_css, p_blkg) \
500 css_for_each_descendant_pre((pos_css), &(p_blkg)->blkcg->css) \
501 if (((d_blkg) = __blkg_lookup(css_to_blkcg(pos_css), \
502 (p_blkg)->q, false)))
505 * blkg_for_each_descendant_post - post-order walk of a blkg's descendants
506 * @d_blkg: loop cursor pointing to the current descendant
507 * @pos_css: used for iteration
508 * @p_blkg: target blkg to walk descendants of
510 * Similar to blkg_for_each_descendant_pre() but performs post-order
511 * traversal instead. Synchronization rules are the same. @p_blkg is
512 * included in the iteration and the last node to be visited.
514 #define blkg_for_each_descendant_post(d_blkg, pos_css, p_blkg) \
515 css_for_each_descendant_post((pos_css), &(p_blkg)->blkcg->css) \
516 if (((d_blkg) = __blkg_lookup(css_to_blkcg(pos_css), \
517 (p_blkg)->q, false)))
520 * blk_get_rl - get request_list to use
521 * @q: request_queue of interest
522 * @bio: bio which will be attached to the allocated request (may be %NULL)
524 * The caller wants to allocate a request from @q to use for @bio. Find
525 * the request_list to use and obtain a reference on it. Should be called
526 * under queue_lock. This function is guaranteed to return non-%NULL
529 static inline struct request_list *blk_get_rl(struct request_queue *q,
533 struct blkcg_gq *blkg;
537 blkcg = bio_blkcg(bio);
539 /* bypass blkg lookup and use @q->root_rl directly for root */
540 if (blkcg == &blkcg_root)
544 * Try to use blkg->rl. blkg lookup may fail under memory pressure
545 * or if either the blkcg or queue is going away. Fall back to
546 * root_rl in such cases.
548 blkg = blkg_lookup(blkcg, q);
561 * blk_put_rl - put request_list
562 * @rl: request_list to put
564 * Put the reference acquired by blk_get_rl(). Should be called under
567 static inline void blk_put_rl(struct request_list *rl)
569 if (rl->blkg->blkcg != &blkcg_root)
574 * blk_rq_set_rl - associate a request with a request_list
575 * @rq: request of interest
576 * @rl: target request_list
578 * Associate @rq with @rl so that accounting and freeing can know the
579 * request_list @rq came from.
581 static inline void blk_rq_set_rl(struct request *rq, struct request_list *rl)
587 * blk_rq_rl - return the request_list a request came from
588 * @rq: request of interest
590 * Return the request_list @rq is allocated from.
592 static inline struct request_list *blk_rq_rl(struct request *rq)
597 struct request_list *__blk_queue_next_rl(struct request_list *rl,
598 struct request_queue *q);
600 * blk_queue_for_each_rl - iterate through all request_lists of a request_queue
602 * Should be used under queue_lock.
604 #define blk_queue_for_each_rl(rl, q) \
605 for ((rl) = &(q)->root_rl; (rl); (rl) = __blk_queue_next_rl((rl), (q)))
607 static inline int blkg_stat_init(struct blkg_stat *stat, gfp_t gfp)
611 ret = percpu_counter_init(&stat->cpu_cnt, 0, gfp);
615 atomic64_set(&stat->aux_cnt, 0);
619 static inline void blkg_stat_exit(struct blkg_stat *stat)
621 percpu_counter_destroy(&stat->cpu_cnt);
625 * blkg_stat_add - add a value to a blkg_stat
626 * @stat: target blkg_stat
629 * Add @val to @stat. The caller must ensure that IRQ on the same CPU
630 * don't re-enter this function for the same counter.
632 static inline void blkg_stat_add(struct blkg_stat *stat, uint64_t val)
634 percpu_counter_add_batch(&stat->cpu_cnt, val, BLKG_STAT_CPU_BATCH);
638 * blkg_stat_read - read the current value of a blkg_stat
639 * @stat: blkg_stat to read
641 static inline uint64_t blkg_stat_read(struct blkg_stat *stat)
643 return percpu_counter_sum_positive(&stat->cpu_cnt);
647 * blkg_stat_reset - reset a blkg_stat
648 * @stat: blkg_stat to reset
650 static inline void blkg_stat_reset(struct blkg_stat *stat)
652 percpu_counter_set(&stat->cpu_cnt, 0);
653 atomic64_set(&stat->aux_cnt, 0);
657 * blkg_stat_add_aux - add a blkg_stat into another's aux count
658 * @to: the destination blkg_stat
661 * Add @from's count including the aux one to @to's aux count.
663 static inline void blkg_stat_add_aux(struct blkg_stat *to,
664 struct blkg_stat *from)
666 atomic64_add(blkg_stat_read(from) + atomic64_read(&from->aux_cnt),
670 static inline int blkg_rwstat_init(struct blkg_rwstat *rwstat, gfp_t gfp)
674 for (i = 0; i < BLKG_RWSTAT_NR; i++) {
675 ret = percpu_counter_init(&rwstat->cpu_cnt[i], 0, gfp);
678 percpu_counter_destroy(&rwstat->cpu_cnt[i]);
681 atomic64_set(&rwstat->aux_cnt[i], 0);
686 static inline void blkg_rwstat_exit(struct blkg_rwstat *rwstat)
690 for (i = 0; i < BLKG_RWSTAT_NR; i++)
691 percpu_counter_destroy(&rwstat->cpu_cnt[i]);
695 * blkg_rwstat_add - add a value to a blkg_rwstat
696 * @rwstat: target blkg_rwstat
697 * @op: REQ_OP and flags
700 * Add @val to @rwstat. The counters are chosen according to @rw. The
701 * caller is responsible for synchronizing calls to this function.
703 static inline void blkg_rwstat_add(struct blkg_rwstat *rwstat,
704 unsigned int op, uint64_t val)
706 struct percpu_counter *cnt;
708 if (op_is_discard(op))
709 cnt = &rwstat->cpu_cnt[BLKG_RWSTAT_DISCARD];
710 else if (op_is_write(op))
711 cnt = &rwstat->cpu_cnt[BLKG_RWSTAT_WRITE];
713 cnt = &rwstat->cpu_cnt[BLKG_RWSTAT_READ];
715 percpu_counter_add_batch(cnt, val, BLKG_STAT_CPU_BATCH);
718 cnt = &rwstat->cpu_cnt[BLKG_RWSTAT_SYNC];
720 cnt = &rwstat->cpu_cnt[BLKG_RWSTAT_ASYNC];
722 percpu_counter_add_batch(cnt, val, BLKG_STAT_CPU_BATCH);
726 * blkg_rwstat_read - read the current values of a blkg_rwstat
727 * @rwstat: blkg_rwstat to read
729 * Read the current snapshot of @rwstat and return it in the aux counts.
731 static inline struct blkg_rwstat blkg_rwstat_read(struct blkg_rwstat *rwstat)
733 struct blkg_rwstat result;
736 for (i = 0; i < BLKG_RWSTAT_NR; i++)
737 atomic64_set(&result.aux_cnt[i],
738 percpu_counter_sum_positive(&rwstat->cpu_cnt[i]));
743 * blkg_rwstat_total - read the total count of a blkg_rwstat
744 * @rwstat: blkg_rwstat to read
746 * Return the total count of @rwstat regardless of the IO direction. This
747 * function can be called without synchronization and takes care of u64
750 static inline uint64_t blkg_rwstat_total(struct blkg_rwstat *rwstat)
752 struct blkg_rwstat tmp = blkg_rwstat_read(rwstat);
754 return atomic64_read(&tmp.aux_cnt[BLKG_RWSTAT_READ]) +
755 atomic64_read(&tmp.aux_cnt[BLKG_RWSTAT_WRITE]);
759 * blkg_rwstat_reset - reset a blkg_rwstat
760 * @rwstat: blkg_rwstat to reset
762 static inline void blkg_rwstat_reset(struct blkg_rwstat *rwstat)
766 for (i = 0; i < BLKG_RWSTAT_NR; i++) {
767 percpu_counter_set(&rwstat->cpu_cnt[i], 0);
768 atomic64_set(&rwstat->aux_cnt[i], 0);
773 * blkg_rwstat_add_aux - add a blkg_rwstat into another's aux count
774 * @to: the destination blkg_rwstat
777 * Add @from's count including the aux one to @to's aux count.
779 static inline void blkg_rwstat_add_aux(struct blkg_rwstat *to,
780 struct blkg_rwstat *from)
782 u64 sum[BLKG_RWSTAT_NR];
785 for (i = 0; i < BLKG_RWSTAT_NR; i++)
786 sum[i] = percpu_counter_sum_positive(&from->cpu_cnt[i]);
788 for (i = 0; i < BLKG_RWSTAT_NR; i++)
789 atomic64_add(sum[i] + atomic64_read(&from->aux_cnt[i]),
793 #ifdef CONFIG_BLK_DEV_THROTTLING
794 extern bool blk_throtl_bio(struct request_queue *q, struct blkcg_gq *blkg,
797 static inline bool blk_throtl_bio(struct request_queue *q, struct blkcg_gq *blkg,
798 struct bio *bio) { return false; }
801 static inline bool blkcg_bio_issue_check(struct request_queue *q,
805 struct blkcg_gq *blkg;
809 blkcg = bio_blkcg(bio);
811 /* associate blkcg if bio hasn't attached one */
812 bio_associate_blkcg(bio, &blkcg->css);
814 blkg = blkg_lookup(blkcg, q);
815 if (unlikely(!blkg)) {
816 spin_lock_irq(q->queue_lock);
817 blkg = blkg_lookup_create(blkcg, q);
820 spin_unlock_irq(q->queue_lock);
823 throtl = blk_throtl_bio(q, blkg, bio);
826 blkg = blkg ?: q->root_blkg;
828 * If the bio is flagged with BIO_QUEUE_ENTERED it means this
829 * is a split bio and we would have already accounted for the
832 if (!bio_flagged(bio, BIO_QUEUE_ENTERED))
833 blkg_rwstat_add(&blkg->stat_bytes, bio->bi_opf,
834 bio->bi_iter.bi_size);
835 blkg_rwstat_add(&blkg->stat_ios, bio->bi_opf, 1);
842 static inline void blkcg_use_delay(struct blkcg_gq *blkg)
844 if (atomic_add_return(1, &blkg->use_delay) == 1)
845 atomic_inc(&blkg->blkcg->css.cgroup->congestion_count);
849 * blk_cgroup_mergeable - Determine whether to allow or disallow merges
850 * @rq: request to merge into
853 * @bio and @rq should belong to the same cgroup and their issue_as_root should
854 * match. The latter is necessary as we don't want to throttle e.g. a metadata
855 * update because it happens to be next to a regular IO.
857 static inline bool blk_cgroup_mergeable(struct request *rq, struct bio *bio)
859 return rq->bio->bi_blkg == bio->bi_blkg &&
860 bio_issue_as_root_blkg(rq->bio) == bio_issue_as_root_blkg(bio);
863 static inline int blkcg_unuse_delay(struct blkcg_gq *blkg)
865 int old = atomic_read(&blkg->use_delay);
871 * We do this song and dance because we can race with somebody else
872 * adding or removing delay. If we just did an atomic_dec we'd end up
873 * negative and we'd already be in trouble. We need to subtract 1 and
874 * then check to see if we were the last delay so we can drop the
875 * congestion count on the cgroup.
878 int cur = atomic_cmpxchg(&blkg->use_delay, old, old - 1);
887 atomic_dec(&blkg->blkcg->css.cgroup->congestion_count);
891 static inline void blkcg_clear_delay(struct blkcg_gq *blkg)
893 int old = atomic_read(&blkg->use_delay);
896 /* We only want 1 person clearing the congestion count for this blkg. */
898 int cur = atomic_cmpxchg(&blkg->use_delay, old, 0);
900 atomic_dec(&blkg->blkcg->css.cgroup->congestion_count);
907 void blkcg_add_delay(struct blkcg_gq *blkg, u64 now, u64 delta);
908 void blkcg_schedule_throttle(struct request_queue *q, bool use_memdelay);
909 void blkcg_maybe_throttle_current(void);
910 #else /* CONFIG_BLK_CGROUP */
915 struct blkg_policy_data {
918 struct blkcg_policy_data {
924 struct blkcg_policy {
927 #define blkcg_root_css ((struct cgroup_subsys_state *)ERR_PTR(-EINVAL))
929 static inline void blkcg_maybe_throttle_current(void) { }
930 static inline bool blk_cgroup_congested(void) { return false; }
934 static inline void blkcg_schedule_throttle(struct request_queue *q, bool use_memdelay) { }
936 static inline struct blkcg_gq *blkg_lookup(struct blkcg *blkcg, void *key) { return NULL; }
937 static inline struct blkcg_gq *blk_queue_root_blkg(struct request_queue *q)
939 static inline int blkcg_init_queue(struct request_queue *q) { return 0; }
940 static inline void blkcg_drain_queue(struct request_queue *q) { }
941 static inline void blkcg_exit_queue(struct request_queue *q) { }
942 static inline int blkcg_policy_register(struct blkcg_policy *pol) { return 0; }
943 static inline void blkcg_policy_unregister(struct blkcg_policy *pol) { }
944 static inline int blkcg_activate_policy(struct request_queue *q,
945 const struct blkcg_policy *pol) { return 0; }
946 static inline void blkcg_deactivate_policy(struct request_queue *q,
947 const struct blkcg_policy *pol) { }
949 static inline struct blkcg *bio_blkcg(struct bio *bio) { return NULL; }
951 static inline struct blkg_policy_data *blkg_to_pd(struct blkcg_gq *blkg,
952 struct blkcg_policy *pol) { return NULL; }
953 static inline struct blkcg_gq *pd_to_blkg(struct blkg_policy_data *pd) { return NULL; }
954 static inline char *blkg_path(struct blkcg_gq *blkg) { return NULL; }
955 static inline void blkg_get(struct blkcg_gq *blkg) { }
956 static inline void blkg_put(struct blkcg_gq *blkg) { }
958 static inline struct request_list *blk_get_rl(struct request_queue *q,
959 struct bio *bio) { return &q->root_rl; }
960 static inline void blk_put_rl(struct request_list *rl) { }
961 static inline void blk_rq_set_rl(struct request *rq, struct request_list *rl) { }
962 static inline struct request_list *blk_rq_rl(struct request *rq) { return &rq->q->root_rl; }
964 static inline bool blkcg_bio_issue_check(struct request_queue *q,
965 struct bio *bio) { return true; }
966 static inline bool blk_cgroup_mergeable(struct request *rq, struct bio *bio) { return true; }
968 #define blk_queue_for_each_rl(rl, q) \
969 for ((rl) = &(q)->root_rl; (rl); (rl) = NULL)
971 #endif /* CONFIG_BLOCK */
972 #endif /* CONFIG_BLK_CGROUP */
973 #endif /* _BLK_CGROUP_H */