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>
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 /* reference count */
127 struct percpu_ref refcnt;
129 /* is this blkg online? protected by both blkcg and q locks */
132 struct blkg_rwstat stat_bytes;
133 struct blkg_rwstat stat_ios;
135 struct blkg_policy_data *pd[BLKCG_MAX_POLS];
137 struct rcu_head rcu_head;
140 atomic64_t delay_nsec;
141 atomic64_t delay_start;
146 typedef struct blkcg_policy_data *(blkcg_pol_alloc_cpd_fn)(gfp_t gfp);
147 typedef void (blkcg_pol_init_cpd_fn)(struct blkcg_policy_data *cpd);
148 typedef void (blkcg_pol_free_cpd_fn)(struct blkcg_policy_data *cpd);
149 typedef void (blkcg_pol_bind_cpd_fn)(struct blkcg_policy_data *cpd);
150 typedef struct blkg_policy_data *(blkcg_pol_alloc_pd_fn)(gfp_t gfp, int node);
151 typedef void (blkcg_pol_init_pd_fn)(struct blkg_policy_data *pd);
152 typedef void (blkcg_pol_online_pd_fn)(struct blkg_policy_data *pd);
153 typedef void (blkcg_pol_offline_pd_fn)(struct blkg_policy_data *pd);
154 typedef void (blkcg_pol_free_pd_fn)(struct blkg_policy_data *pd);
155 typedef void (blkcg_pol_reset_pd_stats_fn)(struct blkg_policy_data *pd);
156 typedef size_t (blkcg_pol_stat_pd_fn)(struct blkg_policy_data *pd, char *buf,
159 struct blkcg_policy {
161 /* cgroup files for the policy */
162 struct cftype *dfl_cftypes;
163 struct cftype *legacy_cftypes;
166 blkcg_pol_alloc_cpd_fn *cpd_alloc_fn;
167 blkcg_pol_init_cpd_fn *cpd_init_fn;
168 blkcg_pol_free_cpd_fn *cpd_free_fn;
169 blkcg_pol_bind_cpd_fn *cpd_bind_fn;
171 blkcg_pol_alloc_pd_fn *pd_alloc_fn;
172 blkcg_pol_init_pd_fn *pd_init_fn;
173 blkcg_pol_online_pd_fn *pd_online_fn;
174 blkcg_pol_offline_pd_fn *pd_offline_fn;
175 blkcg_pol_free_pd_fn *pd_free_fn;
176 blkcg_pol_reset_pd_stats_fn *pd_reset_stats_fn;
177 blkcg_pol_stat_pd_fn *pd_stat_fn;
180 extern struct blkcg blkcg_root;
181 extern struct cgroup_subsys_state * const blkcg_root_css;
183 struct blkcg_gq *blkg_lookup_slowpath(struct blkcg *blkcg,
184 struct request_queue *q, bool update_hint);
185 struct blkcg_gq *__blkg_lookup_create(struct blkcg *blkcg,
186 struct request_queue *q);
187 struct blkcg_gq *blkg_lookup_create(struct blkcg *blkcg,
188 struct request_queue *q);
189 int blkcg_init_queue(struct request_queue *q);
190 void blkcg_drain_queue(struct request_queue *q);
191 void blkcg_exit_queue(struct request_queue *q);
193 /* Blkio controller policy registration */
194 int blkcg_policy_register(struct blkcg_policy *pol);
195 void blkcg_policy_unregister(struct blkcg_policy *pol);
196 int blkcg_activate_policy(struct request_queue *q,
197 const struct blkcg_policy *pol);
198 void blkcg_deactivate_policy(struct request_queue *q,
199 const struct blkcg_policy *pol);
201 static inline u64 blkg_rwstat_read_counter(struct blkg_rwstat *rwstat,
204 return atomic64_read(&rwstat->aux_cnt[idx]) +
205 percpu_counter_sum_positive(&rwstat->cpu_cnt[idx]);
208 const char *blkg_dev_name(struct blkcg_gq *blkg);
209 void blkcg_print_blkgs(struct seq_file *sf, struct blkcg *blkcg,
210 u64 (*prfill)(struct seq_file *,
211 struct blkg_policy_data *, int),
212 const struct blkcg_policy *pol, int data,
214 u64 __blkg_prfill_u64(struct seq_file *sf, struct blkg_policy_data *pd, u64 v);
215 u64 __blkg_prfill_rwstat(struct seq_file *sf, struct blkg_policy_data *pd,
216 const struct blkg_rwstat *rwstat);
217 u64 blkg_prfill_stat(struct seq_file *sf, struct blkg_policy_data *pd, int off);
218 u64 blkg_prfill_rwstat(struct seq_file *sf, struct blkg_policy_data *pd,
220 int blkg_print_stat_bytes(struct seq_file *sf, void *v);
221 int blkg_print_stat_ios(struct seq_file *sf, void *v);
222 int blkg_print_stat_bytes_recursive(struct seq_file *sf, void *v);
223 int blkg_print_stat_ios_recursive(struct seq_file *sf, void *v);
225 u64 blkg_stat_recursive_sum(struct blkcg_gq *blkg,
226 struct blkcg_policy *pol, int off);
227 void blkg_rwstat_recursive_sum(struct blkcg_gq *blkg, struct blkcg_policy *pol,
228 int off, struct blkg_rwstat *sum);
230 struct blkg_conf_ctx {
231 struct gendisk *disk;
232 struct blkcg_gq *blkg;
236 int blkg_conf_prep(struct blkcg *blkcg, const struct blkcg_policy *pol,
237 char *input, struct blkg_conf_ctx *ctx);
238 void blkg_conf_finish(struct blkg_conf_ctx *ctx);
241 * blkcg_css - find the current css
243 * Find the css associated with either the kthread or the current task.
244 * This may return a dying css, so it is up to the caller to use tryget logic
245 * to confirm it is alive and well.
247 static inline struct cgroup_subsys_state *blkcg_css(void)
249 struct cgroup_subsys_state *css;
251 css = kthread_blkcg();
254 return task_css(current, io_cgrp_id);
257 static inline struct blkcg *css_to_blkcg(struct cgroup_subsys_state *css)
259 return css ? container_of(css, struct blkcg, css) : NULL;
263 * __bio_blkcg - internal, inconsistent version to get blkcg
266 * This function is inconsistent and consequently is dangerous to use. The
267 * first part of the function returns a blkcg where a reference is owned by the
268 * bio. This means it does not need to be rcu protected as it cannot go away
269 * with the bio owning a reference to it. However, the latter potentially gets
270 * it from task_css(). This can race against task migration and the cgroup
271 * dying. It is also semantically different as it must be called rcu protected
272 * and is susceptible to failure when trying to get a reference to it.
273 * Therefore, it is not ok to assume that *_get() will always succeed on the
274 * blkcg returned here.
276 static inline struct blkcg *__bio_blkcg(struct bio *bio)
278 if (bio && bio->bi_blkg)
279 return bio->bi_blkg->blkcg;
280 return css_to_blkcg(blkcg_css());
284 * bio_blkcg - grab the blkcg associated with a bio
287 * This returns the blkcg associated with a bio, %NULL if not associated.
288 * Callers are expected to either handle %NULL or know association has been
289 * done prior to calling this.
291 static inline struct blkcg *bio_blkcg(struct bio *bio)
293 if (bio && bio->bi_blkg)
294 return bio->bi_blkg->blkcg;
298 static inline bool blk_cgroup_congested(void)
300 struct cgroup_subsys_state *css;
304 css = kthread_blkcg();
306 css = task_css(current, io_cgrp_id);
308 if (atomic_read(&css->cgroup->congestion_count)) {
319 * bio_issue_as_root_blkg - see if this bio needs to be issued as root blkg
320 * @return: true if this bio needs to be submitted with the root blkg context.
322 * In order to avoid priority inversions we sometimes need to issue a bio as if
323 * it were attached to the root blkg, and then backcharge to the actual owning
324 * blkg. The idea is we do bio_blkcg() to look up the actual context for the
325 * bio and attach the appropriate blkg to the bio. Then we call this helper and
326 * if it is true run with the root blkg for that queue and then do any
327 * backcharging to the originating cgroup once the io is complete.
329 static inline bool bio_issue_as_root_blkg(struct bio *bio)
331 return (bio->bi_opf & (REQ_META | REQ_SWAP)) != 0;
335 * blkcg_parent - get the parent of a blkcg
336 * @blkcg: blkcg of interest
338 * Return the parent blkcg of @blkcg. Can be called anytime.
340 static inline struct blkcg *blkcg_parent(struct blkcg *blkcg)
342 return css_to_blkcg(blkcg->css.parent);
346 * __blkg_lookup - internal version of blkg_lookup()
347 * @blkcg: blkcg of interest
348 * @q: request_queue of interest
349 * @update_hint: whether to update lookup hint with the result or not
351 * This is internal version and shouldn't be used by policy
352 * implementations. Looks up blkgs for the @blkcg - @q pair regardless of
353 * @q's bypass state. If @update_hint is %true, the caller should be
354 * holding @q->queue_lock and lookup hint is updated on success.
356 static inline struct blkcg_gq *__blkg_lookup(struct blkcg *blkcg,
357 struct request_queue *q,
360 struct blkcg_gq *blkg;
362 if (blkcg == &blkcg_root)
365 blkg = rcu_dereference(blkcg->blkg_hint);
366 if (blkg && blkg->q == q)
369 return blkg_lookup_slowpath(blkcg, q, update_hint);
373 * blkg_lookup - lookup blkg for the specified blkcg - q pair
374 * @blkcg: blkcg of interest
375 * @q: request_queue of interest
377 * Lookup blkg for the @blkcg - @q pair. This function should be called
378 * under RCU read loc.
380 static inline struct blkcg_gq *blkg_lookup(struct blkcg *blkcg,
381 struct request_queue *q)
383 WARN_ON_ONCE(!rcu_read_lock_held());
384 return __blkg_lookup(blkcg, q, false);
388 * blk_queue_root_blkg - return blkg for the (blkcg_root, @q) pair
389 * @q: request_queue of interest
391 * Lookup blkg for @q at the root level. See also blkg_lookup().
393 static inline struct blkcg_gq *blk_queue_root_blkg(struct request_queue *q)
399 * blkg_to_pdata - get policy private data
400 * @blkg: blkg of interest
401 * @pol: policy of interest
403 * Return pointer to private data associated with the @blkg-@pol pair.
405 static inline struct blkg_policy_data *blkg_to_pd(struct blkcg_gq *blkg,
406 struct blkcg_policy *pol)
408 return blkg ? blkg->pd[pol->plid] : NULL;
411 static inline struct blkcg_policy_data *blkcg_to_cpd(struct blkcg *blkcg,
412 struct blkcg_policy *pol)
414 return blkcg ? blkcg->cpd[pol->plid] : NULL;
418 * pdata_to_blkg - get blkg associated with policy private data
419 * @pd: policy private data of interest
421 * @pd is policy private data. Determine the blkg it's associated with.
423 static inline struct blkcg_gq *pd_to_blkg(struct blkg_policy_data *pd)
425 return pd ? pd->blkg : NULL;
428 static inline struct blkcg *cpd_to_blkcg(struct blkcg_policy_data *cpd)
430 return cpd ? cpd->blkcg : NULL;
433 extern void blkcg_destroy_blkgs(struct blkcg *blkcg);
435 #ifdef CONFIG_CGROUP_WRITEBACK
438 * blkcg_cgwb_get - get a reference for blkcg->cgwb_list
439 * @blkcg: blkcg of interest
441 * This is used to track the number of active wb's related to a blkcg.
443 static inline void blkcg_cgwb_get(struct blkcg *blkcg)
445 refcount_inc(&blkcg->cgwb_refcnt);
449 * blkcg_cgwb_put - put a reference for @blkcg->cgwb_list
450 * @blkcg: blkcg of interest
452 * This is used to track the number of active wb's related to a blkcg.
453 * When this count goes to zero, all active wb has finished so the
454 * blkcg can continue destruction by calling blkcg_destroy_blkgs().
455 * This work may occur in cgwb_release_workfn() on the cgwb_release
458 static inline void blkcg_cgwb_put(struct blkcg *blkcg)
460 if (refcount_dec_and_test(&blkcg->cgwb_refcnt))
461 blkcg_destroy_blkgs(blkcg);
466 static inline void blkcg_cgwb_get(struct blkcg *blkcg) { }
468 static inline void blkcg_cgwb_put(struct blkcg *blkcg)
470 /* wb isn't being accounted, so trigger destruction right away */
471 blkcg_destroy_blkgs(blkcg);
477 * blkg_path - format cgroup path of blkg
478 * @blkg: blkg of interest
479 * @buf: target buffer
480 * @buflen: target buffer length
482 * Format the path of the cgroup of @blkg into @buf.
484 static inline int blkg_path(struct blkcg_gq *blkg, char *buf, int buflen)
486 return cgroup_path(blkg->blkcg->css.cgroup, buf, buflen);
490 * blkg_get - get a blkg reference
493 * The caller should be holding an existing reference.
495 static inline void blkg_get(struct blkcg_gq *blkg)
497 percpu_ref_get(&blkg->refcnt);
501 * blkg_tryget - try and get a blkg reference
504 * This is for use when doing an RCU lookup of the blkg. We may be in the midst
505 * of freeing this blkg, so we can only use it if the refcnt is not zero.
507 static inline bool blkg_tryget(struct blkcg_gq *blkg)
509 return blkg && percpu_ref_tryget(&blkg->refcnt);
513 * blkg_tryget_closest - try and get a blkg ref on the closet blkg
516 * This needs to be called rcu protected. As the failure mode here is to walk
517 * up the blkg tree, this ensure that the blkg->parent pointers are always
518 * valid. This returns the blkg that it ended up taking a reference on or %NULL
519 * if no reference was taken.
521 static inline struct blkcg_gq *blkg_tryget_closest(struct blkcg_gq *blkg)
523 struct blkcg_gq *ret_blkg = NULL;
525 WARN_ON_ONCE(!rcu_read_lock_held());
528 if (blkg_tryget(blkg)) {
539 * blkg_put - put a blkg reference
542 static inline void blkg_put(struct blkcg_gq *blkg)
544 percpu_ref_put(&blkg->refcnt);
548 * blkg_for_each_descendant_pre - pre-order walk of a blkg's descendants
549 * @d_blkg: loop cursor pointing to the current descendant
550 * @pos_css: used for iteration
551 * @p_blkg: target blkg to walk descendants of
553 * Walk @c_blkg through the descendants of @p_blkg. Must be used with RCU
554 * read locked. If called under either blkcg or queue lock, the iteration
555 * is guaranteed to include all and only online blkgs. The caller may
556 * update @pos_css by calling css_rightmost_descendant() to skip subtree.
557 * @p_blkg is included in the iteration and the first node to be visited.
559 #define blkg_for_each_descendant_pre(d_blkg, pos_css, p_blkg) \
560 css_for_each_descendant_pre((pos_css), &(p_blkg)->blkcg->css) \
561 if (((d_blkg) = __blkg_lookup(css_to_blkcg(pos_css), \
562 (p_blkg)->q, false)))
565 * blkg_for_each_descendant_post - post-order walk of a blkg's descendants
566 * @d_blkg: loop cursor pointing to the current descendant
567 * @pos_css: used for iteration
568 * @p_blkg: target blkg to walk descendants of
570 * Similar to blkg_for_each_descendant_pre() but performs post-order
571 * traversal instead. Synchronization rules are the same. @p_blkg is
572 * included in the iteration and the last node to be visited.
574 #define blkg_for_each_descendant_post(d_blkg, pos_css, p_blkg) \
575 css_for_each_descendant_post((pos_css), &(p_blkg)->blkcg->css) \
576 if (((d_blkg) = __blkg_lookup(css_to_blkcg(pos_css), \
577 (p_blkg)->q, false)))
579 static inline int blkg_stat_init(struct blkg_stat *stat, gfp_t gfp)
583 ret = percpu_counter_init(&stat->cpu_cnt, 0, gfp);
587 atomic64_set(&stat->aux_cnt, 0);
591 static inline void blkg_stat_exit(struct blkg_stat *stat)
593 percpu_counter_destroy(&stat->cpu_cnt);
597 * blkg_stat_add - add a value to a blkg_stat
598 * @stat: target blkg_stat
601 * Add @val to @stat. The caller must ensure that IRQ on the same CPU
602 * don't re-enter this function for the same counter.
604 static inline void blkg_stat_add(struct blkg_stat *stat, uint64_t val)
606 percpu_counter_add_batch(&stat->cpu_cnt, val, BLKG_STAT_CPU_BATCH);
610 * blkg_stat_read - read the current value of a blkg_stat
611 * @stat: blkg_stat to read
613 static inline uint64_t blkg_stat_read(struct blkg_stat *stat)
615 return percpu_counter_sum_positive(&stat->cpu_cnt);
619 * blkg_stat_reset - reset a blkg_stat
620 * @stat: blkg_stat to reset
622 static inline void blkg_stat_reset(struct blkg_stat *stat)
624 percpu_counter_set(&stat->cpu_cnt, 0);
625 atomic64_set(&stat->aux_cnt, 0);
629 * blkg_stat_add_aux - add a blkg_stat into another's aux count
630 * @to: the destination blkg_stat
633 * Add @from's count including the aux one to @to's aux count.
635 static inline void blkg_stat_add_aux(struct blkg_stat *to,
636 struct blkg_stat *from)
638 atomic64_add(blkg_stat_read(from) + atomic64_read(&from->aux_cnt),
642 static inline int blkg_rwstat_init(struct blkg_rwstat *rwstat, gfp_t gfp)
646 for (i = 0; i < BLKG_RWSTAT_NR; i++) {
647 ret = percpu_counter_init(&rwstat->cpu_cnt[i], 0, gfp);
650 percpu_counter_destroy(&rwstat->cpu_cnt[i]);
653 atomic64_set(&rwstat->aux_cnt[i], 0);
658 static inline void blkg_rwstat_exit(struct blkg_rwstat *rwstat)
662 for (i = 0; i < BLKG_RWSTAT_NR; i++)
663 percpu_counter_destroy(&rwstat->cpu_cnt[i]);
667 * blkg_rwstat_add - add a value to a blkg_rwstat
668 * @rwstat: target blkg_rwstat
669 * @op: REQ_OP and flags
672 * Add @val to @rwstat. The counters are chosen according to @rw. The
673 * caller is responsible for synchronizing calls to this function.
675 static inline void blkg_rwstat_add(struct blkg_rwstat *rwstat,
676 unsigned int op, uint64_t val)
678 struct percpu_counter *cnt;
680 if (op_is_discard(op))
681 cnt = &rwstat->cpu_cnt[BLKG_RWSTAT_DISCARD];
682 else if (op_is_write(op))
683 cnt = &rwstat->cpu_cnt[BLKG_RWSTAT_WRITE];
685 cnt = &rwstat->cpu_cnt[BLKG_RWSTAT_READ];
687 percpu_counter_add_batch(cnt, val, BLKG_STAT_CPU_BATCH);
690 cnt = &rwstat->cpu_cnt[BLKG_RWSTAT_SYNC];
692 cnt = &rwstat->cpu_cnt[BLKG_RWSTAT_ASYNC];
694 percpu_counter_add_batch(cnt, val, BLKG_STAT_CPU_BATCH);
698 * blkg_rwstat_read - read the current values of a blkg_rwstat
699 * @rwstat: blkg_rwstat to read
701 * Read the current snapshot of @rwstat and return it in the aux counts.
703 static inline void blkg_rwstat_read(struct blkg_rwstat *rwstat,
704 struct blkg_rwstat *result)
708 for (i = 0; i < BLKG_RWSTAT_NR; i++)
709 atomic64_set(&result->aux_cnt[i],
710 percpu_counter_sum_positive(&rwstat->cpu_cnt[i]));
714 * blkg_rwstat_total - read the total count of a blkg_rwstat
715 * @rwstat: blkg_rwstat to read
717 * Return the total count of @rwstat regardless of the IO direction. This
718 * function can be called without synchronization and takes care of u64
721 static inline uint64_t blkg_rwstat_total(struct blkg_rwstat *rwstat)
723 struct blkg_rwstat tmp = { };
725 blkg_rwstat_read(rwstat, &tmp);
726 return atomic64_read(&tmp.aux_cnt[BLKG_RWSTAT_READ]) +
727 atomic64_read(&tmp.aux_cnt[BLKG_RWSTAT_WRITE]);
731 * blkg_rwstat_reset - reset a blkg_rwstat
732 * @rwstat: blkg_rwstat to reset
734 static inline void blkg_rwstat_reset(struct blkg_rwstat *rwstat)
738 for (i = 0; i < BLKG_RWSTAT_NR; i++) {
739 percpu_counter_set(&rwstat->cpu_cnt[i], 0);
740 atomic64_set(&rwstat->aux_cnt[i], 0);
745 * blkg_rwstat_add_aux - add a blkg_rwstat into another's aux count
746 * @to: the destination blkg_rwstat
749 * Add @from's count including the aux one to @to's aux count.
751 static inline void blkg_rwstat_add_aux(struct blkg_rwstat *to,
752 struct blkg_rwstat *from)
754 u64 sum[BLKG_RWSTAT_NR];
757 for (i = 0; i < BLKG_RWSTAT_NR; i++)
758 sum[i] = percpu_counter_sum_positive(&from->cpu_cnt[i]);
760 for (i = 0; i < BLKG_RWSTAT_NR; i++)
761 atomic64_add(sum[i] + atomic64_read(&from->aux_cnt[i]),
765 #ifdef CONFIG_BLK_DEV_THROTTLING
766 extern bool blk_throtl_bio(struct request_queue *q, struct blkcg_gq *blkg,
769 static inline bool blk_throtl_bio(struct request_queue *q, struct blkcg_gq *blkg,
770 struct bio *bio) { return false; }
774 static inline void blkcg_bio_issue_init(struct bio *bio)
776 bio_issue_init(&bio->bi_issue, bio_sectors(bio));
779 static inline bool blkcg_bio_issue_check(struct request_queue *q,
782 struct blkcg_gq *blkg;
788 char b[BDEVNAME_SIZE];
791 "no blkg associated for bio on block-device: %s\n",
792 bio_devname(bio, b));
793 bio_associate_blkg(bio);
798 throtl = blk_throtl_bio(q, blkg, bio);
802 * If the bio is flagged with BIO_QUEUE_ENTERED it means this
803 * is a split bio and we would have already accounted for the
806 if (!bio_flagged(bio, BIO_QUEUE_ENTERED))
807 blkg_rwstat_add(&blkg->stat_bytes, bio->bi_opf,
808 bio->bi_iter.bi_size);
809 blkg_rwstat_add(&blkg->stat_ios, bio->bi_opf, 1);
812 blkcg_bio_issue_init(bio);
818 static inline void blkcg_use_delay(struct blkcg_gq *blkg)
820 if (atomic_add_return(1, &blkg->use_delay) == 1)
821 atomic_inc(&blkg->blkcg->css.cgroup->congestion_count);
824 static inline int blkcg_unuse_delay(struct blkcg_gq *blkg)
826 int old = atomic_read(&blkg->use_delay);
832 * We do this song and dance because we can race with somebody else
833 * adding or removing delay. If we just did an atomic_dec we'd end up
834 * negative and we'd already be in trouble. We need to subtract 1 and
835 * then check to see if we were the last delay so we can drop the
836 * congestion count on the cgroup.
839 int cur = atomic_cmpxchg(&blkg->use_delay, old, old - 1);
848 atomic_dec(&blkg->blkcg->css.cgroup->congestion_count);
852 static inline void blkcg_clear_delay(struct blkcg_gq *blkg)
854 int old = atomic_read(&blkg->use_delay);
857 /* We only want 1 person clearing the congestion count for this blkg. */
859 int cur = atomic_cmpxchg(&blkg->use_delay, old, 0);
861 atomic_dec(&blkg->blkcg->css.cgroup->congestion_count);
868 void blkcg_add_delay(struct blkcg_gq *blkg, u64 now, u64 delta);
869 void blkcg_schedule_throttle(struct request_queue *q, bool use_memdelay);
870 void blkcg_maybe_throttle_current(void);
871 #else /* CONFIG_BLK_CGROUP */
876 struct blkg_policy_data {
879 struct blkcg_policy_data {
885 struct blkcg_policy {
888 #define blkcg_root_css ((struct cgroup_subsys_state *)ERR_PTR(-EINVAL))
890 static inline void blkcg_maybe_throttle_current(void) { }
891 static inline bool blk_cgroup_congested(void) { return false; }
895 static inline void blkcg_schedule_throttle(struct request_queue *q, bool use_memdelay) { }
897 static inline struct blkcg_gq *blkg_lookup(struct blkcg *blkcg, void *key) { return NULL; }
898 static inline struct blkcg_gq *blk_queue_root_blkg(struct request_queue *q)
900 static inline int blkcg_init_queue(struct request_queue *q) { return 0; }
901 static inline void blkcg_drain_queue(struct request_queue *q) { }
902 static inline void blkcg_exit_queue(struct request_queue *q) { }
903 static inline int blkcg_policy_register(struct blkcg_policy *pol) { return 0; }
904 static inline void blkcg_policy_unregister(struct blkcg_policy *pol) { }
905 static inline int blkcg_activate_policy(struct request_queue *q,
906 const struct blkcg_policy *pol) { return 0; }
907 static inline void blkcg_deactivate_policy(struct request_queue *q,
908 const struct blkcg_policy *pol) { }
910 static inline struct blkcg *__bio_blkcg(struct bio *bio) { return NULL; }
911 static inline struct blkcg *bio_blkcg(struct bio *bio) { return NULL; }
913 static inline struct blkg_policy_data *blkg_to_pd(struct blkcg_gq *blkg,
914 struct blkcg_policy *pol) { return NULL; }
915 static inline struct blkcg_gq *pd_to_blkg(struct blkg_policy_data *pd) { return NULL; }
916 static inline char *blkg_path(struct blkcg_gq *blkg) { return NULL; }
917 static inline void blkg_get(struct blkcg_gq *blkg) { }
918 static inline void blkg_put(struct blkcg_gq *blkg) { }
920 static inline void blkcg_bio_issue_init(struct bio *bio) { }
921 static inline bool blkcg_bio_issue_check(struct request_queue *q,
922 struct bio *bio) { return true; }
924 #define blk_queue_for_each_rl(rl, q) \
925 for ((rl) = &(q)->root_rl; (rl); (rl) = NULL)
927 #endif /* CONFIG_BLOCK */
928 #endif /* CONFIG_BLK_CGROUP */
929 #endif /* _BLK_CGROUP_H */