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
34 #define FC_APPID_LEN 129
37 #ifdef CONFIG_BLK_CGROUP
39 enum blkg_iostat_type {
50 struct cgroup_subsys_state css;
52 refcount_t online_pin;
54 struct radix_tree_root blkg_tree;
55 struct blkcg_gq __rcu *blkg_hint;
56 struct hlist_head blkg_list;
58 struct blkcg_policy_data *cpd[BLKCG_MAX_POLS];
60 struct list_head all_blkcgs_node;
61 #ifdef CONFIG_BLK_CGROUP_FC_APPID
62 char fc_app_id[FC_APPID_LEN];
64 #ifdef CONFIG_CGROUP_WRITEBACK
65 struct list_head cgwb_list;
70 u64 bytes[BLKG_IOSTAT_NR];
71 u64 ios[BLKG_IOSTAT_NR];
74 struct blkg_iostat_set {
75 struct u64_stats_sync sync;
76 struct blkg_iostat cur;
77 struct blkg_iostat last;
81 * A blkcg_gq (blkg) is association between a block cgroup (blkcg) and a
82 * request_queue (q). This is used by blkcg policies which need to track
83 * information per blkcg - q pair.
85 * There can be multiple active blkcg policies and each blkg:policy pair is
86 * represented by a blkg_policy_data which is allocated and freed by each
87 * policy's pd_alloc/free_fn() methods. A policy can allocate private data
88 * area by allocating larger data structure which embeds blkg_policy_data
91 struct blkg_policy_data {
92 /* the blkg and policy id this per-policy data belongs to */
93 struct blkcg_gq *blkg;
98 * Policies that need to keep per-blkcg data which is independent from any
99 * request_queue associated to it should implement cpd_alloc/free_fn()
100 * methods. A policy can allocate private data area by allocating larger
101 * data structure which embeds blkcg_policy_data at the beginning.
102 * cpd_init() is invoked to let each policy handle per-blkcg data.
104 struct blkcg_policy_data {
105 /* the blkcg and policy id this per-policy data belongs to */
110 /* association between a blk cgroup and a request queue */
112 /* Pointer to the associated request_queue */
113 struct request_queue *q;
114 struct list_head q_node;
115 struct hlist_node blkcg_node;
118 /* all non-root blkcg_gq's are guaranteed to have access to parent */
119 struct blkcg_gq *parent;
121 /* reference count */
122 struct percpu_ref refcnt;
124 /* is this blkg online? protected by both blkcg and q locks */
127 struct blkg_iostat_set __percpu *iostat_cpu;
128 struct blkg_iostat_set iostat;
130 struct blkg_policy_data *pd[BLKCG_MAX_POLS];
132 spinlock_t async_bio_lock;
133 struct bio_list async_bios;
134 struct work_struct async_bio_work;
137 atomic64_t delay_nsec;
138 atomic64_t delay_start;
142 struct rcu_head rcu_head;
145 typedef struct blkcg_policy_data *(blkcg_pol_alloc_cpd_fn)(gfp_t gfp);
146 typedef void (blkcg_pol_init_cpd_fn)(struct blkcg_policy_data *cpd);
147 typedef void (blkcg_pol_free_cpd_fn)(struct blkcg_policy_data *cpd);
148 typedef void (blkcg_pol_bind_cpd_fn)(struct blkcg_policy_data *cpd);
149 typedef struct blkg_policy_data *(blkcg_pol_alloc_pd_fn)(gfp_t gfp,
150 struct request_queue *q, struct blkcg *blkcg);
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 bool (blkcg_pol_stat_pd_fn)(struct blkg_policy_data *pd,
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;
182 extern bool blkcg_debug_stats;
184 struct blkcg_gq *blkg_lookup_slowpath(struct blkcg *blkcg,
185 struct request_queue *q, bool update_hint);
186 int blkcg_init_queue(struct request_queue *q);
187 void blkcg_exit_queue(struct request_queue *q);
189 /* Blkio controller policy registration */
190 int blkcg_policy_register(struct blkcg_policy *pol);
191 void blkcg_policy_unregister(struct blkcg_policy *pol);
192 int blkcg_activate_policy(struct request_queue *q,
193 const struct blkcg_policy *pol);
194 void blkcg_deactivate_policy(struct request_queue *q,
195 const struct blkcg_policy *pol);
197 const char *blkg_dev_name(struct blkcg_gq *blkg);
198 void blkcg_print_blkgs(struct seq_file *sf, struct blkcg *blkcg,
199 u64 (*prfill)(struct seq_file *,
200 struct blkg_policy_data *, int),
201 const struct blkcg_policy *pol, int data,
203 u64 __blkg_prfill_u64(struct seq_file *sf, struct blkg_policy_data *pd, u64 v);
205 struct blkg_conf_ctx {
206 struct block_device *bdev;
207 struct blkcg_gq *blkg;
211 struct block_device *blkcg_conf_open_bdev(char **inputp);
212 int blkg_conf_prep(struct blkcg *blkcg, const struct blkcg_policy *pol,
213 char *input, struct blkg_conf_ctx *ctx);
214 void blkg_conf_finish(struct blkg_conf_ctx *ctx);
217 * blkcg_css - find the current css
219 * Find the css associated with either the kthread or the current task.
220 * This may return a dying css, so it is up to the caller to use tryget logic
221 * to confirm it is alive and well.
223 static inline struct cgroup_subsys_state *blkcg_css(void)
225 struct cgroup_subsys_state *css;
227 css = kthread_blkcg();
230 return task_css(current, io_cgrp_id);
233 static inline struct blkcg *css_to_blkcg(struct cgroup_subsys_state *css)
235 return css ? container_of(css, struct blkcg, css) : NULL;
239 * __bio_blkcg - internal, inconsistent version to get blkcg
242 * This function is inconsistent and consequently is dangerous to use. The
243 * first part of the function returns a blkcg where a reference is owned by the
244 * bio. This means it does not need to be rcu protected as it cannot go away
245 * with the bio owning a reference to it. However, the latter potentially gets
246 * it from task_css(). This can race against task migration and the cgroup
247 * dying. It is also semantically different as it must be called rcu protected
248 * and is susceptible to failure when trying to get a reference to it.
249 * Therefore, it is not ok to assume that *_get() will always succeed on the
250 * blkcg returned here.
252 static inline struct blkcg *__bio_blkcg(struct bio *bio)
254 if (bio && bio->bi_blkg)
255 return bio->bi_blkg->blkcg;
256 return css_to_blkcg(blkcg_css());
260 * bio_blkcg - grab the blkcg associated with a bio
263 * This returns the blkcg associated with a bio, %NULL if not associated.
264 * Callers are expected to either handle %NULL or know association has been
265 * done prior to calling this.
267 static inline struct blkcg *bio_blkcg(struct bio *bio)
269 if (bio && bio->bi_blkg)
270 return bio->bi_blkg->blkcg;
274 static inline bool blk_cgroup_congested(void)
276 struct cgroup_subsys_state *css;
280 css = kthread_blkcg();
282 css = task_css(current, io_cgrp_id);
284 if (atomic_read(&css->cgroup->congestion_count)) {
295 * bio_issue_as_root_blkg - see if this bio needs to be issued as root blkg
296 * @return: true if this bio needs to be submitted with the root blkg context.
298 * In order to avoid priority inversions we sometimes need to issue a bio as if
299 * it were attached to the root blkg, and then backcharge to the actual owning
300 * blkg. The idea is we do bio_blkcg() to look up the actual context for the
301 * bio and attach the appropriate blkg to the bio. Then we call this helper and
302 * if it is true run with the root blkg for that queue and then do any
303 * backcharging to the originating cgroup once the io is complete.
305 static inline bool bio_issue_as_root_blkg(struct bio *bio)
307 return (bio->bi_opf & (REQ_META | REQ_SWAP)) != 0;
311 * blkcg_parent - get the parent of a blkcg
312 * @blkcg: blkcg of interest
314 * Return the parent blkcg of @blkcg. Can be called anytime.
316 static inline struct blkcg *blkcg_parent(struct blkcg *blkcg)
318 return css_to_blkcg(blkcg->css.parent);
322 * __blkg_lookup - internal version of blkg_lookup()
323 * @blkcg: blkcg of interest
324 * @q: request_queue of interest
325 * @update_hint: whether to update lookup hint with the result or not
327 * This is internal version and shouldn't be used by policy
328 * implementations. Looks up blkgs for the @blkcg - @q pair regardless of
329 * @q's bypass state. If @update_hint is %true, the caller should be
330 * holding @q->queue_lock and lookup hint is updated on success.
332 static inline struct blkcg_gq *__blkg_lookup(struct blkcg *blkcg,
333 struct request_queue *q,
336 struct blkcg_gq *blkg;
338 if (blkcg == &blkcg_root)
341 blkg = rcu_dereference(blkcg->blkg_hint);
342 if (blkg && blkg->q == q)
345 return blkg_lookup_slowpath(blkcg, q, update_hint);
349 * blkg_lookup - lookup blkg for the specified blkcg - q pair
350 * @blkcg: blkcg of interest
351 * @q: request_queue of interest
353 * Lookup blkg for the @blkcg - @q pair. This function should be called
354 * under RCU read lock.
356 static inline struct blkcg_gq *blkg_lookup(struct blkcg *blkcg,
357 struct request_queue *q)
359 WARN_ON_ONCE(!rcu_read_lock_held());
360 return __blkg_lookup(blkcg, q, false);
364 * blk_queue_root_blkg - return blkg for the (blkcg_root, @q) pair
365 * @q: request_queue of interest
367 * Lookup blkg for @q at the root level. See also blkg_lookup().
369 static inline struct blkcg_gq *blk_queue_root_blkg(struct request_queue *q)
375 * blkg_to_pdata - get policy private data
376 * @blkg: blkg of interest
377 * @pol: policy of interest
379 * Return pointer to private data associated with the @blkg-@pol pair.
381 static inline struct blkg_policy_data *blkg_to_pd(struct blkcg_gq *blkg,
382 struct blkcg_policy *pol)
384 return blkg ? blkg->pd[pol->plid] : NULL;
387 static inline struct blkcg_policy_data *blkcg_to_cpd(struct blkcg *blkcg,
388 struct blkcg_policy *pol)
390 return blkcg ? blkcg->cpd[pol->plid] : NULL;
394 * pdata_to_blkg - get blkg associated with policy private data
395 * @pd: policy private data of interest
397 * @pd is policy private data. Determine the blkg it's associated with.
399 static inline struct blkcg_gq *pd_to_blkg(struct blkg_policy_data *pd)
401 return pd ? pd->blkg : NULL;
404 static inline struct blkcg *cpd_to_blkcg(struct blkcg_policy_data *cpd)
406 return cpd ? cpd->blkcg : NULL;
409 extern void blkcg_destroy_blkgs(struct blkcg *blkcg);
412 * blkcg_pin_online - pin online state
413 * @blkcg: blkcg of interest
415 * While pinned, a blkcg is kept online. This is primarily used to
416 * impedance-match blkg and cgwb lifetimes so that blkg doesn't go offline
417 * while an associated cgwb is still active.
419 static inline void blkcg_pin_online(struct blkcg *blkcg)
421 refcount_inc(&blkcg->online_pin);
425 * blkcg_unpin_online - unpin online state
426 * @blkcg: blkcg of interest
428 * This is primarily used to impedance-match blkg and cgwb lifetimes so
429 * that blkg doesn't go offline while an associated cgwb is still active.
430 * When this count goes to zero, all active cgwbs have finished so the
431 * blkcg can continue destruction by calling blkcg_destroy_blkgs().
433 static inline void blkcg_unpin_online(struct blkcg *blkcg)
436 if (!refcount_dec_and_test(&blkcg->online_pin))
438 blkcg_destroy_blkgs(blkcg);
439 blkcg = blkcg_parent(blkcg);
444 * blkg_path - format cgroup path of blkg
445 * @blkg: blkg of interest
446 * @buf: target buffer
447 * @buflen: target buffer length
449 * Format the path of the cgroup of @blkg into @buf.
451 static inline int blkg_path(struct blkcg_gq *blkg, char *buf, int buflen)
453 return cgroup_path(blkg->blkcg->css.cgroup, buf, buflen);
457 * blkg_get - get a blkg reference
460 * The caller should be holding an existing reference.
462 static inline void blkg_get(struct blkcg_gq *blkg)
464 percpu_ref_get(&blkg->refcnt);
468 * blkg_tryget - try and get a blkg reference
471 * This is for use when doing an RCU lookup of the blkg. We may be in the midst
472 * of freeing this blkg, so we can only use it if the refcnt is not zero.
474 static inline bool blkg_tryget(struct blkcg_gq *blkg)
476 return blkg && percpu_ref_tryget(&blkg->refcnt);
480 * blkg_put - put a blkg reference
483 static inline void blkg_put(struct blkcg_gq *blkg)
485 percpu_ref_put(&blkg->refcnt);
489 * blkg_for_each_descendant_pre - pre-order walk of a blkg's descendants
490 * @d_blkg: loop cursor pointing to the current descendant
491 * @pos_css: used for iteration
492 * @p_blkg: target blkg to walk descendants of
494 * Walk @c_blkg through the descendants of @p_blkg. Must be used with RCU
495 * read locked. If called under either blkcg or queue lock, the iteration
496 * is guaranteed to include all and only online blkgs. The caller may
497 * update @pos_css by calling css_rightmost_descendant() to skip subtree.
498 * @p_blkg is included in the iteration and the first node to be visited.
500 #define blkg_for_each_descendant_pre(d_blkg, pos_css, p_blkg) \
501 css_for_each_descendant_pre((pos_css), &(p_blkg)->blkcg->css) \
502 if (((d_blkg) = __blkg_lookup(css_to_blkcg(pos_css), \
503 (p_blkg)->q, false)))
506 * blkg_for_each_descendant_post - post-order walk of a blkg's descendants
507 * @d_blkg: loop cursor pointing to the current descendant
508 * @pos_css: used for iteration
509 * @p_blkg: target blkg to walk descendants of
511 * Similar to blkg_for_each_descendant_pre() but performs post-order
512 * traversal instead. Synchronization rules are the same. @p_blkg is
513 * included in the iteration and the last node to be visited.
515 #define blkg_for_each_descendant_post(d_blkg, pos_css, p_blkg) \
516 css_for_each_descendant_post((pos_css), &(p_blkg)->blkcg->css) \
517 if (((d_blkg) = __blkg_lookup(css_to_blkcg(pos_css), \
518 (p_blkg)->q, false)))
520 bool __blkcg_punt_bio_submit(struct bio *bio);
522 static inline bool blkcg_punt_bio_submit(struct bio *bio)
524 if (bio->bi_opf & REQ_CGROUP_PUNT)
525 return __blkcg_punt_bio_submit(bio);
530 static inline void blkcg_bio_issue_init(struct bio *bio)
532 bio_issue_init(&bio->bi_issue, bio_sectors(bio));
535 static inline void blkcg_use_delay(struct blkcg_gq *blkg)
537 if (WARN_ON_ONCE(atomic_read(&blkg->use_delay) < 0))
539 if (atomic_add_return(1, &blkg->use_delay) == 1)
540 atomic_inc(&blkg->blkcg->css.cgroup->congestion_count);
543 static inline int blkcg_unuse_delay(struct blkcg_gq *blkg)
545 int old = atomic_read(&blkg->use_delay);
547 if (WARN_ON_ONCE(old < 0))
553 * We do this song and dance because we can race with somebody else
554 * adding or removing delay. If we just did an atomic_dec we'd end up
555 * negative and we'd already be in trouble. We need to subtract 1 and
556 * then check to see if we were the last delay so we can drop the
557 * congestion count on the cgroup.
560 int cur = atomic_cmpxchg(&blkg->use_delay, old, old - 1);
569 atomic_dec(&blkg->blkcg->css.cgroup->congestion_count);
574 * blkcg_set_delay - Enable allocator delay mechanism with the specified delay amount
576 * @delay: delay duration in nsecs
578 * When enabled with this function, the delay is not decayed and must be
579 * explicitly cleared with blkcg_clear_delay(). Must not be mixed with
580 * blkcg_[un]use_delay() and blkcg_add_delay() usages.
582 static inline void blkcg_set_delay(struct blkcg_gq *blkg, u64 delay)
584 int old = atomic_read(&blkg->use_delay);
586 /* We only want 1 person setting the congestion count for this blkg. */
587 if (!old && atomic_cmpxchg(&blkg->use_delay, old, -1) == old)
588 atomic_inc(&blkg->blkcg->css.cgroup->congestion_count);
590 atomic64_set(&blkg->delay_nsec, delay);
594 * blkcg_clear_delay - Disable allocator delay mechanism
597 * Disable use_delay mechanism. See blkcg_set_delay().
599 static inline void blkcg_clear_delay(struct blkcg_gq *blkg)
601 int old = atomic_read(&blkg->use_delay);
603 /* We only want 1 person clearing the congestion count for this blkg. */
604 if (old && atomic_cmpxchg(&blkg->use_delay, old, 0) == old)
605 atomic_dec(&blkg->blkcg->css.cgroup->congestion_count);
609 * blk_cgroup_mergeable - Determine whether to allow or disallow merges
610 * @rq: request to merge into
613 * @bio and @rq should belong to the same cgroup and their issue_as_root should
614 * match. The latter is necessary as we don't want to throttle e.g. a metadata
615 * update because it happens to be next to a regular IO.
617 static inline bool blk_cgroup_mergeable(struct request *rq, struct bio *bio)
619 return rq->bio->bi_blkg == bio->bi_blkg &&
620 bio_issue_as_root_blkg(rq->bio) == bio_issue_as_root_blkg(bio);
623 void blk_cgroup_bio_start(struct bio *bio);
624 void blkcg_add_delay(struct blkcg_gq *blkg, u64 now, u64 delta);
625 void blkcg_schedule_throttle(struct request_queue *q, bool use_memdelay);
626 void blkcg_maybe_throttle_current(void);
627 #else /* CONFIG_BLK_CGROUP */
632 struct blkg_policy_data {
635 struct blkcg_policy_data {
641 struct blkcg_policy {
644 #define blkcg_root_css ((struct cgroup_subsys_state *)ERR_PTR(-EINVAL))
646 static inline void blkcg_maybe_throttle_current(void) { }
647 static inline bool blk_cgroup_congested(void) { return false; }
651 static inline void blkcg_schedule_throttle(struct request_queue *q, bool use_memdelay) { }
653 static inline struct blkcg_gq *blkg_lookup(struct blkcg *blkcg, void *key) { return NULL; }
654 static inline struct blkcg_gq *blk_queue_root_blkg(struct request_queue *q)
656 static inline int blkcg_init_queue(struct request_queue *q) { return 0; }
657 static inline void blkcg_exit_queue(struct request_queue *q) { }
658 static inline int blkcg_policy_register(struct blkcg_policy *pol) { return 0; }
659 static inline void blkcg_policy_unregister(struct blkcg_policy *pol) { }
660 static inline int blkcg_activate_policy(struct request_queue *q,
661 const struct blkcg_policy *pol) { return 0; }
662 static inline void blkcg_deactivate_policy(struct request_queue *q,
663 const struct blkcg_policy *pol) { }
665 static inline struct blkcg *__bio_blkcg(struct bio *bio) { return NULL; }
666 static inline struct blkcg *bio_blkcg(struct bio *bio) { return NULL; }
668 static inline struct blkg_policy_data *blkg_to_pd(struct blkcg_gq *blkg,
669 struct blkcg_policy *pol) { return NULL; }
670 static inline struct blkcg_gq *pd_to_blkg(struct blkg_policy_data *pd) { return NULL; }
671 static inline char *blkg_path(struct blkcg_gq *blkg) { return NULL; }
672 static inline void blkg_get(struct blkcg_gq *blkg) { }
673 static inline void blkg_put(struct blkcg_gq *blkg) { }
675 static inline bool blkcg_punt_bio_submit(struct bio *bio) { return false; }
676 static inline void blkcg_bio_issue_init(struct bio *bio) { }
677 static inline void blk_cgroup_bio_start(struct bio *bio) { }
678 static inline bool blk_cgroup_mergeable(struct request *rq, struct bio *bio) { return true; }
680 #define blk_queue_for_each_rl(rl, q) \
681 for ((rl) = &(q)->root_rl; (rl); (rl) = NULL)
683 #endif /* CONFIG_BLOCK */
684 #endif /* CONFIG_BLK_CGROUP */
686 #ifdef CONFIG_BLK_CGROUP_FC_APPID
688 * Sets the fc_app_id field associted to blkcg
689 * @app_id: application identifier
690 * @cgrp_id: cgroup id
691 * @app_id_len: size of application identifier
693 static inline int blkcg_set_fc_appid(char *app_id, u64 cgrp_id, size_t app_id_len)
696 struct cgroup_subsys_state *css;
700 if (app_id_len > FC_APPID_LEN)
703 cgrp = cgroup_get_from_id(cgrp_id);
706 css = cgroup_get_e_css(cgrp, &io_cgrp_subsys);
711 blkcg = css_to_blkcg(css);
713 * There is a slight race condition on setting the appid.
714 * Worst case an I/O may not find the right id.
715 * This is no different from the I/O we let pass while obtaining
716 * the vmid from the fabric.
717 * Adding the overhead of a lock is not necessary.
719 strlcpy(blkcg->fc_app_id, app_id, app_id_len);
727 * blkcg_get_fc_appid - get the fc app identifier associated with a bio
730 * On success return the fc_app_id, on failure return NULL
732 static inline char *blkcg_get_fc_appid(struct bio *bio)
734 if (bio && bio->bi_blkg &&
735 (bio->bi_blkg->blkcg->fc_app_id[0] != '\0'))
736 return bio->bi_blkg->blkcg->fc_app_id;
740 static inline int blkcg_set_fc_appid(char *buf, u64 id, size_t len) { return -EINVAL; }
741 static inline char *blkcg_get_fc_appid(struct bio *bio) { return NULL; }
742 #endif /*CONFIG_BLK_CGROUP_FC_APPID*/
743 #endif /* _BLK_CGROUP_H */