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.h>
19 #include <linux/percpu_counter.h>
20 #include <linux/u64_stats_sync.h>
21 #include <linux/seq_file.h>
22 #include <linux/radix-tree.h>
23 #include <linux/blkdev.h>
24 #include <linux/atomic.h>
25 #include <linux/kthread.h>
27 #include <linux/blk-mq.h>
29 /* percpu_counter batch for blkg_[rw]stats, per-cpu drift doesn't matter */
30 #define BLKG_STAT_CPU_BATCH (INT_MAX / 2)
32 /* Max limits for throttle policy */
33 #define THROTL_IOPS_MAX UINT_MAX
35 #ifdef CONFIG_BLK_CGROUP
37 enum blkg_iostat_type {
48 struct cgroup_subsys_state css;
50 refcount_t online_pin;
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;
65 u64 bytes[BLKG_IOSTAT_NR];
66 u64 ios[BLKG_IOSTAT_NR];
69 struct blkg_iostat_set {
70 struct u64_stats_sync sync;
71 struct blkg_iostat cur;
72 struct blkg_iostat last;
76 * A blkcg_gq (blkg) is association between a block cgroup (blkcg) and a
77 * request_queue (q). This is used by blkcg policies which need to track
78 * information per blkcg - q pair.
80 * There can be multiple active blkcg policies and each blkg:policy pair is
81 * represented by a blkg_policy_data which is allocated and freed by each
82 * policy's pd_alloc/free_fn() methods. A policy can allocate private data
83 * area by allocating larger data structure which embeds blkg_policy_data
86 struct blkg_policy_data {
87 /* the blkg and policy id this per-policy data belongs to */
88 struct blkcg_gq *blkg;
93 * Policies that need to keep per-blkcg data which is independent from any
94 * request_queue associated to it should implement cpd_alloc/free_fn()
95 * methods. A policy can allocate private data area by allocating larger
96 * data structure which embeds blkcg_policy_data at the beginning.
97 * cpd_init() is invoked to let each policy handle per-blkcg data.
99 struct blkcg_policy_data {
100 /* the blkcg and policy id this per-policy data belongs to */
105 /* association between a blk cgroup and a request queue */
107 /* Pointer to the associated request_queue */
108 struct request_queue *q;
109 struct list_head q_node;
110 struct hlist_node blkcg_node;
113 /* all non-root blkcg_gq's are guaranteed to have access to parent */
114 struct blkcg_gq *parent;
116 /* reference count */
117 struct percpu_ref refcnt;
119 /* is this blkg online? protected by both blkcg and q locks */
122 struct blkg_iostat_set __percpu *iostat_cpu;
123 struct blkg_iostat_set iostat;
125 struct blkg_policy_data *pd[BLKCG_MAX_POLS];
127 spinlock_t async_bio_lock;
128 struct bio_list async_bios;
129 struct work_struct async_bio_work;
132 atomic64_t delay_nsec;
133 atomic64_t delay_start;
137 struct rcu_head rcu_head;
140 typedef struct blkcg_policy_data *(blkcg_pol_alloc_cpd_fn)(gfp_t gfp);
141 typedef void (blkcg_pol_init_cpd_fn)(struct blkcg_policy_data *cpd);
142 typedef void (blkcg_pol_free_cpd_fn)(struct blkcg_policy_data *cpd);
143 typedef void (blkcg_pol_bind_cpd_fn)(struct blkcg_policy_data *cpd);
144 typedef struct blkg_policy_data *(blkcg_pol_alloc_pd_fn)(gfp_t gfp,
145 struct request_queue *q, struct blkcg *blkcg);
146 typedef void (blkcg_pol_init_pd_fn)(struct blkg_policy_data *pd);
147 typedef void (blkcg_pol_online_pd_fn)(struct blkg_policy_data *pd);
148 typedef void (blkcg_pol_offline_pd_fn)(struct blkg_policy_data *pd);
149 typedef void (blkcg_pol_free_pd_fn)(struct blkg_policy_data *pd);
150 typedef void (blkcg_pol_reset_pd_stats_fn)(struct blkg_policy_data *pd);
151 typedef size_t (blkcg_pol_stat_pd_fn)(struct blkg_policy_data *pd, char *buf,
154 struct blkcg_policy {
156 /* cgroup files for the policy */
157 struct cftype *dfl_cftypes;
158 struct cftype *legacy_cftypes;
161 blkcg_pol_alloc_cpd_fn *cpd_alloc_fn;
162 blkcg_pol_init_cpd_fn *cpd_init_fn;
163 blkcg_pol_free_cpd_fn *cpd_free_fn;
164 blkcg_pol_bind_cpd_fn *cpd_bind_fn;
166 blkcg_pol_alloc_pd_fn *pd_alloc_fn;
167 blkcg_pol_init_pd_fn *pd_init_fn;
168 blkcg_pol_online_pd_fn *pd_online_fn;
169 blkcg_pol_offline_pd_fn *pd_offline_fn;
170 blkcg_pol_free_pd_fn *pd_free_fn;
171 blkcg_pol_reset_pd_stats_fn *pd_reset_stats_fn;
172 blkcg_pol_stat_pd_fn *pd_stat_fn;
175 extern struct blkcg blkcg_root;
176 extern struct cgroup_subsys_state * const blkcg_root_css;
177 extern bool blkcg_debug_stats;
179 struct blkcg_gq *blkg_lookup_slowpath(struct blkcg *blkcg,
180 struct request_queue *q, bool update_hint);
181 int blkcg_init_queue(struct request_queue *q);
182 void blkcg_exit_queue(struct request_queue *q);
184 /* Blkio controller policy registration */
185 int blkcg_policy_register(struct blkcg_policy *pol);
186 void blkcg_policy_unregister(struct blkcg_policy *pol);
187 int blkcg_activate_policy(struct request_queue *q,
188 const struct blkcg_policy *pol);
189 void blkcg_deactivate_policy(struct request_queue *q,
190 const struct blkcg_policy *pol);
192 const char *blkg_dev_name(struct blkcg_gq *blkg);
193 void blkcg_print_blkgs(struct seq_file *sf, struct blkcg *blkcg,
194 u64 (*prfill)(struct seq_file *,
195 struct blkg_policy_data *, int),
196 const struct blkcg_policy *pol, int data,
198 u64 __blkg_prfill_u64(struct seq_file *sf, struct blkg_policy_data *pd, u64 v);
200 struct blkg_conf_ctx {
201 struct gendisk *disk;
202 struct blkcg_gq *blkg;
206 struct gendisk *blkcg_conf_get_disk(char **inputp);
207 int blkg_conf_prep(struct blkcg *blkcg, const struct blkcg_policy *pol,
208 char *input, struct blkg_conf_ctx *ctx);
209 void blkg_conf_finish(struct blkg_conf_ctx *ctx);
212 * blkcg_css - find the current css
214 * Find the css associated with either the kthread or the current task.
215 * This may return a dying css, so it is up to the caller to use tryget logic
216 * to confirm it is alive and well.
218 static inline struct cgroup_subsys_state *blkcg_css(void)
220 struct cgroup_subsys_state *css;
222 css = kthread_blkcg();
225 return task_css(current, io_cgrp_id);
228 static inline struct blkcg *css_to_blkcg(struct cgroup_subsys_state *css)
230 return css ? container_of(css, struct blkcg, css) : NULL;
234 * __bio_blkcg - internal, inconsistent version to get blkcg
237 * This function is inconsistent and consequently is dangerous to use. The
238 * first part of the function returns a blkcg where a reference is owned by the
239 * bio. This means it does not need to be rcu protected as it cannot go away
240 * with the bio owning a reference to it. However, the latter potentially gets
241 * it from task_css(). This can race against task migration and the cgroup
242 * dying. It is also semantically different as it must be called rcu protected
243 * and is susceptible to failure when trying to get a reference to it.
244 * Therefore, it is not ok to assume that *_get() will always succeed on the
245 * blkcg returned here.
247 static inline struct blkcg *__bio_blkcg(struct bio *bio)
249 if (bio && bio->bi_blkg)
250 return bio->bi_blkg->blkcg;
251 return css_to_blkcg(blkcg_css());
255 * bio_blkcg - grab the blkcg associated with a bio
258 * This returns the blkcg associated with a bio, %NULL if not associated.
259 * Callers are expected to either handle %NULL or know association has been
260 * done prior to calling this.
262 static inline struct blkcg *bio_blkcg(struct bio *bio)
264 if (bio && bio->bi_blkg)
265 return bio->bi_blkg->blkcg;
269 static inline bool blk_cgroup_congested(void)
271 struct cgroup_subsys_state *css;
275 css = kthread_blkcg();
277 css = task_css(current, io_cgrp_id);
279 if (atomic_read(&css->cgroup->congestion_count)) {
290 * bio_issue_as_root_blkg - see if this bio needs to be issued as root blkg
291 * @return: true if this bio needs to be submitted with the root blkg context.
293 * In order to avoid priority inversions we sometimes need to issue a bio as if
294 * it were attached to the root blkg, and then backcharge to the actual owning
295 * blkg. The idea is we do bio_blkcg() to look up the actual context for the
296 * bio and attach the appropriate blkg to the bio. Then we call this helper and
297 * if it is true run with the root blkg for that queue and then do any
298 * backcharging to the originating cgroup once the io is complete.
300 static inline bool bio_issue_as_root_blkg(struct bio *bio)
302 return (bio->bi_opf & (REQ_META | REQ_SWAP)) != 0;
306 * blkcg_parent - get the parent of a blkcg
307 * @blkcg: blkcg of interest
309 * Return the parent blkcg of @blkcg. Can be called anytime.
311 static inline struct blkcg *blkcg_parent(struct blkcg *blkcg)
313 return css_to_blkcg(blkcg->css.parent);
317 * __blkg_lookup - internal version of blkg_lookup()
318 * @blkcg: blkcg of interest
319 * @q: request_queue of interest
320 * @update_hint: whether to update lookup hint with the result or not
322 * This is internal version and shouldn't be used by policy
323 * implementations. Looks up blkgs for the @blkcg - @q pair regardless of
324 * @q's bypass state. If @update_hint is %true, the caller should be
325 * holding @q->queue_lock and lookup hint is updated on success.
327 static inline struct blkcg_gq *__blkg_lookup(struct blkcg *blkcg,
328 struct request_queue *q,
331 struct blkcg_gq *blkg;
333 if (blkcg == &blkcg_root)
336 blkg = rcu_dereference(blkcg->blkg_hint);
337 if (blkg && blkg->q == q)
340 return blkg_lookup_slowpath(blkcg, q, update_hint);
344 * blkg_lookup - lookup blkg for the specified blkcg - q pair
345 * @blkcg: blkcg of interest
346 * @q: request_queue of interest
348 * Lookup blkg for the @blkcg - @q pair. This function should be called
349 * under RCU read lock.
351 static inline struct blkcg_gq *blkg_lookup(struct blkcg *blkcg,
352 struct request_queue *q)
354 WARN_ON_ONCE(!rcu_read_lock_held());
355 return __blkg_lookup(blkcg, q, false);
359 * blk_queue_root_blkg - return blkg for the (blkcg_root, @q) pair
360 * @q: request_queue of interest
362 * Lookup blkg for @q at the root level. See also blkg_lookup().
364 static inline struct blkcg_gq *blk_queue_root_blkg(struct request_queue *q)
370 * blkg_to_pdata - get policy private data
371 * @blkg: blkg of interest
372 * @pol: policy of interest
374 * Return pointer to private data associated with the @blkg-@pol pair.
376 static inline struct blkg_policy_data *blkg_to_pd(struct blkcg_gq *blkg,
377 struct blkcg_policy *pol)
379 return blkg ? blkg->pd[pol->plid] : NULL;
382 static inline struct blkcg_policy_data *blkcg_to_cpd(struct blkcg *blkcg,
383 struct blkcg_policy *pol)
385 return blkcg ? blkcg->cpd[pol->plid] : NULL;
389 * pdata_to_blkg - get blkg associated with policy private data
390 * @pd: policy private data of interest
392 * @pd is policy private data. Determine the blkg it's associated with.
394 static inline struct blkcg_gq *pd_to_blkg(struct blkg_policy_data *pd)
396 return pd ? pd->blkg : NULL;
399 static inline struct blkcg *cpd_to_blkcg(struct blkcg_policy_data *cpd)
401 return cpd ? cpd->blkcg : NULL;
404 extern void blkcg_destroy_blkgs(struct blkcg *blkcg);
407 * blkcg_pin_online - pin online state
408 * @blkcg: blkcg of interest
410 * While pinned, a blkcg is kept online. This is primarily used to
411 * impedance-match blkg and cgwb lifetimes so that blkg doesn't go offline
412 * while an associated cgwb is still active.
414 static inline void blkcg_pin_online(struct blkcg *blkcg)
416 refcount_inc(&blkcg->online_pin);
420 * blkcg_unpin_online - unpin online state
421 * @blkcg: blkcg of interest
423 * This is primarily used to impedance-match blkg and cgwb lifetimes so
424 * that blkg doesn't go offline while an associated cgwb is still active.
425 * When this count goes to zero, all active cgwbs have finished so the
426 * blkcg can continue destruction by calling blkcg_destroy_blkgs().
428 static inline void blkcg_unpin_online(struct blkcg *blkcg)
431 if (!refcount_dec_and_test(&blkcg->online_pin))
433 blkcg_destroy_blkgs(blkcg);
434 blkcg = blkcg_parent(blkcg);
439 * blkg_path - format cgroup path of blkg
440 * @blkg: blkg of interest
441 * @buf: target buffer
442 * @buflen: target buffer length
444 * Format the path of the cgroup of @blkg into @buf.
446 static inline int blkg_path(struct blkcg_gq *blkg, char *buf, int buflen)
448 return cgroup_path(blkg->blkcg->css.cgroup, buf, buflen);
452 * blkg_get - get a blkg reference
455 * The caller should be holding an existing reference.
457 static inline void blkg_get(struct blkcg_gq *blkg)
459 percpu_ref_get(&blkg->refcnt);
463 * blkg_tryget - try and get a blkg reference
466 * This is for use when doing an RCU lookup of the blkg. We may be in the midst
467 * of freeing this blkg, so we can only use it if the refcnt is not zero.
469 static inline bool blkg_tryget(struct blkcg_gq *blkg)
471 return blkg && percpu_ref_tryget(&blkg->refcnt);
475 * blkg_put - put a blkg reference
478 static inline void blkg_put(struct blkcg_gq *blkg)
480 percpu_ref_put(&blkg->refcnt);
484 * blkg_for_each_descendant_pre - pre-order walk of a blkg's descendants
485 * @d_blkg: loop cursor pointing to the current descendant
486 * @pos_css: used for iteration
487 * @p_blkg: target blkg to walk descendants of
489 * Walk @c_blkg through the descendants of @p_blkg. Must be used with RCU
490 * read locked. If called under either blkcg or queue lock, the iteration
491 * is guaranteed to include all and only online blkgs. The caller may
492 * update @pos_css by calling css_rightmost_descendant() to skip subtree.
493 * @p_blkg is included in the iteration and the first node to be visited.
495 #define blkg_for_each_descendant_pre(d_blkg, pos_css, p_blkg) \
496 css_for_each_descendant_pre((pos_css), &(p_blkg)->blkcg->css) \
497 if (((d_blkg) = __blkg_lookup(css_to_blkcg(pos_css), \
498 (p_blkg)->q, false)))
501 * blkg_for_each_descendant_post - post-order walk of a blkg's descendants
502 * @d_blkg: loop cursor pointing to the current descendant
503 * @pos_css: used for iteration
504 * @p_blkg: target blkg to walk descendants of
506 * Similar to blkg_for_each_descendant_pre() but performs post-order
507 * traversal instead. Synchronization rules are the same. @p_blkg is
508 * included in the iteration and the last node to be visited.
510 #define blkg_for_each_descendant_post(d_blkg, pos_css, p_blkg) \
511 css_for_each_descendant_post((pos_css), &(p_blkg)->blkcg->css) \
512 if (((d_blkg) = __blkg_lookup(css_to_blkcg(pos_css), \
513 (p_blkg)->q, false)))
515 bool __blkcg_punt_bio_submit(struct bio *bio);
517 static inline bool blkcg_punt_bio_submit(struct bio *bio)
519 if (bio->bi_opf & REQ_CGROUP_PUNT)
520 return __blkcg_punt_bio_submit(bio);
525 static inline void blkcg_bio_issue_init(struct bio *bio)
527 bio_issue_init(&bio->bi_issue, bio_sectors(bio));
530 static inline void blkcg_use_delay(struct blkcg_gq *blkg)
532 if (WARN_ON_ONCE(atomic_read(&blkg->use_delay) < 0))
534 if (atomic_add_return(1, &blkg->use_delay) == 1)
535 atomic_inc(&blkg->blkcg->css.cgroup->congestion_count);
538 static inline int blkcg_unuse_delay(struct blkcg_gq *blkg)
540 int old = atomic_read(&blkg->use_delay);
542 if (WARN_ON_ONCE(old < 0))
548 * We do this song and dance because we can race with somebody else
549 * adding or removing delay. If we just did an atomic_dec we'd end up
550 * negative and we'd already be in trouble. We need to subtract 1 and
551 * then check to see if we were the last delay so we can drop the
552 * congestion count on the cgroup.
555 int cur = atomic_cmpxchg(&blkg->use_delay, old, old - 1);
564 atomic_dec(&blkg->blkcg->css.cgroup->congestion_count);
569 * blkcg_set_delay - Enable allocator delay mechanism with the specified delay amount
571 * @delay: delay duration in nsecs
573 * When enabled with this function, the delay is not decayed and must be
574 * explicitly cleared with blkcg_clear_delay(). Must not be mixed with
575 * blkcg_[un]use_delay() and blkcg_add_delay() usages.
577 static inline void blkcg_set_delay(struct blkcg_gq *blkg, u64 delay)
579 int old = atomic_read(&blkg->use_delay);
581 /* We only want 1 person setting the congestion count for this blkg. */
582 if (!old && atomic_cmpxchg(&blkg->use_delay, old, -1) == old)
583 atomic_inc(&blkg->blkcg->css.cgroup->congestion_count);
585 atomic64_set(&blkg->delay_nsec, delay);
589 * blkcg_clear_delay - Disable allocator delay mechanism
592 * Disable use_delay mechanism. See blkcg_set_delay().
594 static inline void blkcg_clear_delay(struct blkcg_gq *blkg)
596 int old = atomic_read(&blkg->use_delay);
598 /* We only want 1 person clearing the congestion count for this blkg. */
599 if (old && atomic_cmpxchg(&blkg->use_delay, old, 0) == old)
600 atomic_dec(&blkg->blkcg->css.cgroup->congestion_count);
604 * blk_cgroup_mergeable - Determine whether to allow or disallow merges
605 * @rq: request to merge into
608 * @bio and @rq should belong to the same cgroup and their issue_as_root should
609 * match. The latter is necessary as we don't want to throttle e.g. a metadata
610 * update because it happens to be next to a regular IO.
612 static inline bool blk_cgroup_mergeable(struct request *rq, struct bio *bio)
614 return rq->bio->bi_blkg == bio->bi_blkg &&
615 bio_issue_as_root_blkg(rq->bio) == bio_issue_as_root_blkg(bio);
618 void blk_cgroup_bio_start(struct bio *bio);
619 void blkcg_add_delay(struct blkcg_gq *blkg, u64 now, u64 delta);
620 void blkcg_schedule_throttle(struct request_queue *q, bool use_memdelay);
621 void blkcg_maybe_throttle_current(void);
622 #else /* CONFIG_BLK_CGROUP */
627 struct blkg_policy_data {
630 struct blkcg_policy_data {
636 struct blkcg_policy {
639 #define blkcg_root_css ((struct cgroup_subsys_state *)ERR_PTR(-EINVAL))
641 static inline void blkcg_maybe_throttle_current(void) { }
642 static inline bool blk_cgroup_congested(void) { return false; }
646 static inline void blkcg_schedule_throttle(struct request_queue *q, bool use_memdelay) { }
648 static inline struct blkcg_gq *blkg_lookup(struct blkcg *blkcg, void *key) { return NULL; }
649 static inline struct blkcg_gq *blk_queue_root_blkg(struct request_queue *q)
651 static inline int blkcg_init_queue(struct request_queue *q) { return 0; }
652 static inline void blkcg_exit_queue(struct request_queue *q) { }
653 static inline int blkcg_policy_register(struct blkcg_policy *pol) { return 0; }
654 static inline void blkcg_policy_unregister(struct blkcg_policy *pol) { }
655 static inline int blkcg_activate_policy(struct request_queue *q,
656 const struct blkcg_policy *pol) { return 0; }
657 static inline void blkcg_deactivate_policy(struct request_queue *q,
658 const struct blkcg_policy *pol) { }
660 static inline struct blkcg *__bio_blkcg(struct bio *bio) { return NULL; }
661 static inline struct blkcg *bio_blkcg(struct bio *bio) { return NULL; }
663 static inline struct blkg_policy_data *blkg_to_pd(struct blkcg_gq *blkg,
664 struct blkcg_policy *pol) { return NULL; }
665 static inline struct blkcg_gq *pd_to_blkg(struct blkg_policy_data *pd) { return NULL; }
666 static inline char *blkg_path(struct blkcg_gq *blkg) { return NULL; }
667 static inline void blkg_get(struct blkcg_gq *blkg) { }
668 static inline void blkg_put(struct blkcg_gq *blkg) { }
670 static inline bool blkcg_punt_bio_submit(struct bio *bio) { return false; }
671 static inline void blkcg_bio_issue_init(struct bio *bio) { }
672 static inline void blk_cgroup_bio_start(struct bio *bio) { }
673 static inline bool blk_cgroup_mergeable(struct request *rq, struct bio *bio) { return true; }
675 #define blk_queue_for_each_rl(rl, q) \
676 for ((rl) = &(q)->root_rl; (rl); (rl) = NULL)
678 #endif /* CONFIG_BLOCK */
679 #endif /* CONFIG_BLK_CGROUP */
680 #endif /* _BLK_CGROUP_H */