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:
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>
28 /* percpu_counter batch for blkg_[rw]stats, per-cpu drift doesn't matter */
29 #define BLKG_STAT_CPU_BATCH (INT_MAX / 2)
31 /* Max limits for throttle policy */
32 #define THROTL_IOPS_MAX UINT_MAX
34 #ifdef CONFIG_BLK_CGROUP
36 enum blkg_iostat_type {
47 struct cgroup_subsys_state css;
49 refcount_t online_pin;
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;
64 u64 bytes[BLKG_IOSTAT_NR];
65 u64 ios[BLKG_IOSTAT_NR];
68 struct blkg_iostat_set {
69 struct u64_stats_sync sync;
70 struct blkg_iostat cur;
71 struct blkg_iostat last;
75 * A blkcg_gq (blkg) is association between a block cgroup (blkcg) and a
76 * request_queue (q). This is used by blkcg policies which need to track
77 * information per blkcg - q pair.
79 * There can be multiple active blkcg policies and each blkg:policy pair is
80 * represented by a blkg_policy_data which is allocated and freed by each
81 * policy's pd_alloc/free_fn() methods. A policy can allocate private data
82 * area by allocating larger data structure which embeds blkg_policy_data
85 struct blkg_policy_data {
86 /* the blkg and policy id this per-policy data belongs to */
87 struct blkcg_gq *blkg;
92 * Policies that need to keep per-blkcg data which is independent from any
93 * request_queue associated to it should implement cpd_alloc/free_fn()
94 * methods. A policy can allocate private data area by allocating larger
95 * data structure which embeds blkcg_policy_data at the beginning.
96 * cpd_init() is invoked to let each policy handle per-blkcg data.
98 struct blkcg_policy_data {
99 /* the blkcg and policy id this per-policy data belongs to */
104 /* association between a blk cgroup and a request queue */
106 /* Pointer to the associated request_queue */
107 struct request_queue *q;
108 struct list_head q_node;
109 struct hlist_node blkcg_node;
112 /* all non-root blkcg_gq's are guaranteed to have access to parent */
113 struct blkcg_gq *parent;
115 /* reference count */
116 struct percpu_ref refcnt;
118 /* is this blkg online? protected by both blkcg and q locks */
121 struct blkg_iostat_set __percpu *iostat_cpu;
122 struct blkg_iostat_set iostat;
124 struct blkg_policy_data *pd[BLKCG_MAX_POLS];
126 spinlock_t async_bio_lock;
127 struct bio_list async_bios;
128 struct work_struct async_bio_work;
131 atomic64_t delay_nsec;
132 atomic64_t delay_start;
136 struct rcu_head rcu_head;
139 typedef struct blkcg_policy_data *(blkcg_pol_alloc_cpd_fn)(gfp_t gfp);
140 typedef void (blkcg_pol_init_cpd_fn)(struct blkcg_policy_data *cpd);
141 typedef void (blkcg_pol_free_cpd_fn)(struct blkcg_policy_data *cpd);
142 typedef void (blkcg_pol_bind_cpd_fn)(struct blkcg_policy_data *cpd);
143 typedef struct blkg_policy_data *(blkcg_pol_alloc_pd_fn)(gfp_t gfp,
144 struct request_queue *q, struct blkcg *blkcg);
145 typedef void (blkcg_pol_init_pd_fn)(struct blkg_policy_data *pd);
146 typedef void (blkcg_pol_online_pd_fn)(struct blkg_policy_data *pd);
147 typedef void (blkcg_pol_offline_pd_fn)(struct blkg_policy_data *pd);
148 typedef void (blkcg_pol_free_pd_fn)(struct blkg_policy_data *pd);
149 typedef void (blkcg_pol_reset_pd_stats_fn)(struct blkg_policy_data *pd);
150 typedef size_t (blkcg_pol_stat_pd_fn)(struct blkg_policy_data *pd, char *buf,
153 struct blkcg_policy {
155 /* cgroup files for the policy */
156 struct cftype *dfl_cftypes;
157 struct cftype *legacy_cftypes;
160 blkcg_pol_alloc_cpd_fn *cpd_alloc_fn;
161 blkcg_pol_init_cpd_fn *cpd_init_fn;
162 blkcg_pol_free_cpd_fn *cpd_free_fn;
163 blkcg_pol_bind_cpd_fn *cpd_bind_fn;
165 blkcg_pol_alloc_pd_fn *pd_alloc_fn;
166 blkcg_pol_init_pd_fn *pd_init_fn;
167 blkcg_pol_online_pd_fn *pd_online_fn;
168 blkcg_pol_offline_pd_fn *pd_offline_fn;
169 blkcg_pol_free_pd_fn *pd_free_fn;
170 blkcg_pol_reset_pd_stats_fn *pd_reset_stats_fn;
171 blkcg_pol_stat_pd_fn *pd_stat_fn;
174 extern struct blkcg blkcg_root;
175 extern struct cgroup_subsys_state * const blkcg_root_css;
176 extern bool blkcg_debug_stats;
178 struct blkcg_gq *blkg_lookup_slowpath(struct blkcg *blkcg,
179 struct request_queue *q, bool update_hint);
180 int blkcg_init_queue(struct request_queue *q);
181 void blkcg_exit_queue(struct request_queue *q);
183 /* Blkio controller policy registration */
184 int blkcg_policy_register(struct blkcg_policy *pol);
185 void blkcg_policy_unregister(struct blkcg_policy *pol);
186 int blkcg_activate_policy(struct request_queue *q,
187 const struct blkcg_policy *pol);
188 void blkcg_deactivate_policy(struct request_queue *q,
189 const struct blkcg_policy *pol);
191 const char *blkg_dev_name(struct blkcg_gq *blkg);
192 void blkcg_print_blkgs(struct seq_file *sf, struct blkcg *blkcg,
193 u64 (*prfill)(struct seq_file *,
194 struct blkg_policy_data *, int),
195 const struct blkcg_policy *pol, int data,
197 u64 __blkg_prfill_u64(struct seq_file *sf, struct blkg_policy_data *pd, u64 v);
199 struct blkg_conf_ctx {
200 struct block_device *bdev;
201 struct blkcg_gq *blkg;
205 struct block_device *blkcg_conf_open_bdev(char **inputp);
206 int blkg_conf_prep(struct blkcg *blkcg, const struct blkcg_policy *pol,
207 char *input, struct blkg_conf_ctx *ctx);
208 void blkg_conf_finish(struct blkg_conf_ctx *ctx);
211 * blkcg_css - find the current css
213 * Find the css associated with either the kthread or the current task.
214 * This may return a dying css, so it is up to the caller to use tryget logic
215 * to confirm it is alive and well.
217 static inline struct cgroup_subsys_state *blkcg_css(void)
219 struct cgroup_subsys_state *css;
221 css = kthread_blkcg();
224 return task_css(current, io_cgrp_id);
227 static inline struct blkcg *css_to_blkcg(struct cgroup_subsys_state *css)
229 return css ? container_of(css, struct blkcg, css) : NULL;
233 * __bio_blkcg - internal, inconsistent version to get blkcg
236 * This function is inconsistent and consequently is dangerous to use. The
237 * first part of the function returns a blkcg where a reference is owned by the
238 * bio. This means it does not need to be rcu protected as it cannot go away
239 * with the bio owning a reference to it. However, the latter potentially gets
240 * it from task_css(). This can race against task migration and the cgroup
241 * dying. It is also semantically different as it must be called rcu protected
242 * and is susceptible to failure when trying to get a reference to it.
243 * Therefore, it is not ok to assume that *_get() will always succeed on the
244 * blkcg returned here.
246 static inline struct blkcg *__bio_blkcg(struct bio *bio)
248 if (bio && bio->bi_blkg)
249 return bio->bi_blkg->blkcg;
250 return css_to_blkcg(blkcg_css());
254 * bio_blkcg - grab the blkcg associated with a bio
257 * This returns the blkcg associated with a bio, %NULL if not associated.
258 * Callers are expected to either handle %NULL or know association has been
259 * done prior to calling this.
261 static inline struct blkcg *bio_blkcg(struct bio *bio)
263 if (bio && bio->bi_blkg)
264 return bio->bi_blkg->blkcg;
268 static inline bool blk_cgroup_congested(void)
270 struct cgroup_subsys_state *css;
274 css = kthread_blkcg();
276 css = task_css(current, io_cgrp_id);
278 if (atomic_read(&css->cgroup->congestion_count)) {
289 * bio_issue_as_root_blkg - see if this bio needs to be issued as root blkg
290 * @return: true if this bio needs to be submitted with the root blkg context.
292 * In order to avoid priority inversions we sometimes need to issue a bio as if
293 * it were attached to the root blkg, and then backcharge to the actual owning
294 * blkg. The idea is we do bio_blkcg() to look up the actual context for the
295 * bio and attach the appropriate blkg to the bio. Then we call this helper and
296 * if it is true run with the root blkg for that queue and then do any
297 * backcharging to the originating cgroup once the io is complete.
299 static inline bool bio_issue_as_root_blkg(struct bio *bio)
301 return (bio->bi_opf & (REQ_META | REQ_SWAP)) != 0;
305 * blkcg_parent - get the parent of a blkcg
306 * @blkcg: blkcg of interest
308 * Return the parent blkcg of @blkcg. Can be called anytime.
310 static inline struct blkcg *blkcg_parent(struct blkcg *blkcg)
312 return css_to_blkcg(blkcg->css.parent);
316 * __blkg_lookup - internal version of blkg_lookup()
317 * @blkcg: blkcg of interest
318 * @q: request_queue of interest
319 * @update_hint: whether to update lookup hint with the result or not
321 * This is internal version and shouldn't be used by policy
322 * implementations. Looks up blkgs for the @blkcg - @q pair regardless of
323 * @q's bypass state. If @update_hint is %true, the caller should be
324 * holding @q->queue_lock and lookup hint is updated on success.
326 static inline struct blkcg_gq *__blkg_lookup(struct blkcg *blkcg,
327 struct request_queue *q,
330 struct blkcg_gq *blkg;
332 if (blkcg == &blkcg_root)
335 blkg = rcu_dereference(blkcg->blkg_hint);
336 if (blkg && blkg->q == q)
339 return blkg_lookup_slowpath(blkcg, q, update_hint);
343 * blkg_lookup - lookup blkg for the specified blkcg - q pair
344 * @blkcg: blkcg of interest
345 * @q: request_queue of interest
347 * Lookup blkg for the @blkcg - @q pair. This function should be called
348 * under RCU read lock.
350 static inline struct blkcg_gq *blkg_lookup(struct blkcg *blkcg,
351 struct request_queue *q)
353 WARN_ON_ONCE(!rcu_read_lock_held());
354 return __blkg_lookup(blkcg, q, false);
358 * blk_queue_root_blkg - return blkg for the (blkcg_root, @q) pair
359 * @q: request_queue of interest
361 * Lookup blkg for @q at the root level. See also blkg_lookup().
363 static inline struct blkcg_gq *blk_queue_root_blkg(struct request_queue *q)
369 * blkg_to_pdata - get policy private data
370 * @blkg: blkg of interest
371 * @pol: policy of interest
373 * Return pointer to private data associated with the @blkg-@pol pair.
375 static inline struct blkg_policy_data *blkg_to_pd(struct blkcg_gq *blkg,
376 struct blkcg_policy *pol)
378 return blkg ? blkg->pd[pol->plid] : NULL;
381 static inline struct blkcg_policy_data *blkcg_to_cpd(struct blkcg *blkcg,
382 struct blkcg_policy *pol)
384 return blkcg ? blkcg->cpd[pol->plid] : NULL;
388 * pdata_to_blkg - get blkg associated with policy private data
389 * @pd: policy private data of interest
391 * @pd is policy private data. Determine the blkg it's associated with.
393 static inline struct blkcg_gq *pd_to_blkg(struct blkg_policy_data *pd)
395 return pd ? pd->blkg : NULL;
398 static inline struct blkcg *cpd_to_blkcg(struct blkcg_policy_data *cpd)
400 return cpd ? cpd->blkcg : NULL;
403 extern void blkcg_destroy_blkgs(struct blkcg *blkcg);
406 * blkcg_pin_online - pin online state
407 * @blkcg: blkcg of interest
409 * While pinned, a blkcg is kept online. This is primarily used to
410 * impedance-match blkg and cgwb lifetimes so that blkg doesn't go offline
411 * while an associated cgwb is still active.
413 static inline void blkcg_pin_online(struct blkcg *blkcg)
415 refcount_inc(&blkcg->online_pin);
419 * blkcg_unpin_online - unpin online state
420 * @blkcg: blkcg of interest
422 * This is primarily used to impedance-match blkg and cgwb lifetimes so
423 * that blkg doesn't go offline while an associated cgwb is still active.
424 * When this count goes to zero, all active cgwbs have finished so the
425 * blkcg can continue destruction by calling blkcg_destroy_blkgs().
427 static inline void blkcg_unpin_online(struct blkcg *blkcg)
430 if (!refcount_dec_and_test(&blkcg->online_pin))
432 blkcg_destroy_blkgs(blkcg);
433 blkcg = blkcg_parent(blkcg);
438 * blkg_path - format cgroup path of blkg
439 * @blkg: blkg of interest
440 * @buf: target buffer
441 * @buflen: target buffer length
443 * Format the path of the cgroup of @blkg into @buf.
445 static inline int blkg_path(struct blkcg_gq *blkg, char *buf, int buflen)
447 return cgroup_path(blkg->blkcg->css.cgroup, buf, buflen);
451 * blkg_get - get a blkg reference
454 * The caller should be holding an existing reference.
456 static inline void blkg_get(struct blkcg_gq *blkg)
458 percpu_ref_get(&blkg->refcnt);
462 * blkg_tryget - try and get a blkg reference
465 * This is for use when doing an RCU lookup of the blkg. We may be in the midst
466 * of freeing this blkg, so we can only use it if the refcnt is not zero.
468 static inline bool blkg_tryget(struct blkcg_gq *blkg)
470 return blkg && percpu_ref_tryget(&blkg->refcnt);
474 * blkg_put - put a blkg reference
477 static inline void blkg_put(struct blkcg_gq *blkg)
479 percpu_ref_put(&blkg->refcnt);
483 * blkg_for_each_descendant_pre - pre-order walk of a blkg's descendants
484 * @d_blkg: loop cursor pointing to the current descendant
485 * @pos_css: used for iteration
486 * @p_blkg: target blkg to walk descendants of
488 * Walk @c_blkg through the descendants of @p_blkg. Must be used with RCU
489 * read locked. If called under either blkcg or queue lock, the iteration
490 * is guaranteed to include all and only online blkgs. The caller may
491 * update @pos_css by calling css_rightmost_descendant() to skip subtree.
492 * @p_blkg is included in the iteration and the first node to be visited.
494 #define blkg_for_each_descendant_pre(d_blkg, pos_css, p_blkg) \
495 css_for_each_descendant_pre((pos_css), &(p_blkg)->blkcg->css) \
496 if (((d_blkg) = __blkg_lookup(css_to_blkcg(pos_css), \
497 (p_blkg)->q, false)))
500 * blkg_for_each_descendant_post - post-order walk of a blkg's descendants
501 * @d_blkg: loop cursor pointing to the current descendant
502 * @pos_css: used for iteration
503 * @p_blkg: target blkg to walk descendants of
505 * Similar to blkg_for_each_descendant_pre() but performs post-order
506 * traversal instead. Synchronization rules are the same. @p_blkg is
507 * included in the iteration and the last node to be visited.
509 #define blkg_for_each_descendant_post(d_blkg, pos_css, p_blkg) \
510 css_for_each_descendant_post((pos_css), &(p_blkg)->blkcg->css) \
511 if (((d_blkg) = __blkg_lookup(css_to_blkcg(pos_css), \
512 (p_blkg)->q, false)))
514 bool __blkcg_punt_bio_submit(struct bio *bio);
516 static inline bool blkcg_punt_bio_submit(struct bio *bio)
518 if (bio->bi_opf & REQ_CGROUP_PUNT)
519 return __blkcg_punt_bio_submit(bio);
524 static inline void blkcg_bio_issue_init(struct bio *bio)
526 bio_issue_init(&bio->bi_issue, bio_sectors(bio));
529 static inline void blkcg_use_delay(struct blkcg_gq *blkg)
531 if (WARN_ON_ONCE(atomic_read(&blkg->use_delay) < 0))
533 if (atomic_add_return(1, &blkg->use_delay) == 1)
534 atomic_inc(&blkg->blkcg->css.cgroup->congestion_count);
537 static inline int blkcg_unuse_delay(struct blkcg_gq *blkg)
539 int old = atomic_read(&blkg->use_delay);
541 if (WARN_ON_ONCE(old < 0))
547 * We do this song and dance because we can race with somebody else
548 * adding or removing delay. If we just did an atomic_dec we'd end up
549 * negative and we'd already be in trouble. We need to subtract 1 and
550 * then check to see if we were the last delay so we can drop the
551 * congestion count on the cgroup.
554 int cur = atomic_cmpxchg(&blkg->use_delay, old, old - 1);
563 atomic_dec(&blkg->blkcg->css.cgroup->congestion_count);
568 * blkcg_set_delay - Enable allocator delay mechanism with the specified delay amount
570 * @delay: delay duration in nsecs
572 * When enabled with this function, the delay is not decayed and must be
573 * explicitly cleared with blkcg_clear_delay(). Must not be mixed with
574 * blkcg_[un]use_delay() and blkcg_add_delay() usages.
576 static inline void blkcg_set_delay(struct blkcg_gq *blkg, u64 delay)
578 int old = atomic_read(&blkg->use_delay);
580 /* We only want 1 person setting the congestion count for this blkg. */
581 if (!old && atomic_cmpxchg(&blkg->use_delay, old, -1) == old)
582 atomic_inc(&blkg->blkcg->css.cgroup->congestion_count);
584 atomic64_set(&blkg->delay_nsec, delay);
588 * blkcg_clear_delay - Disable allocator delay mechanism
591 * Disable use_delay mechanism. See blkcg_set_delay().
593 static inline void blkcg_clear_delay(struct blkcg_gq *blkg)
595 int old = atomic_read(&blkg->use_delay);
597 /* We only want 1 person clearing the congestion count for this blkg. */
598 if (old && atomic_cmpxchg(&blkg->use_delay, old, 0) == old)
599 atomic_dec(&blkg->blkcg->css.cgroup->congestion_count);
602 void blk_cgroup_bio_start(struct bio *bio);
603 void blkcg_add_delay(struct blkcg_gq *blkg, u64 now, u64 delta);
604 void blkcg_schedule_throttle(struct request_queue *q, bool use_memdelay);
605 void blkcg_maybe_throttle_current(void);
606 #else /* CONFIG_BLK_CGROUP */
611 struct blkg_policy_data {
614 struct blkcg_policy_data {
620 struct blkcg_policy {
623 #define blkcg_root_css ((struct cgroup_subsys_state *)ERR_PTR(-EINVAL))
625 static inline void blkcg_maybe_throttle_current(void) { }
626 static inline bool blk_cgroup_congested(void) { return false; }
630 static inline void blkcg_schedule_throttle(struct request_queue *q, bool use_memdelay) { }
632 static inline struct blkcg_gq *blkg_lookup(struct blkcg *blkcg, void *key) { return NULL; }
633 static inline struct blkcg_gq *blk_queue_root_blkg(struct request_queue *q)
635 static inline int blkcg_init_queue(struct request_queue *q) { return 0; }
636 static inline void blkcg_exit_queue(struct request_queue *q) { }
637 static inline int blkcg_policy_register(struct blkcg_policy *pol) { return 0; }
638 static inline void blkcg_policy_unregister(struct blkcg_policy *pol) { }
639 static inline int blkcg_activate_policy(struct request_queue *q,
640 const struct blkcg_policy *pol) { return 0; }
641 static inline void blkcg_deactivate_policy(struct request_queue *q,
642 const struct blkcg_policy *pol) { }
644 static inline struct blkcg *__bio_blkcg(struct bio *bio) { return NULL; }
645 static inline struct blkcg *bio_blkcg(struct bio *bio) { return NULL; }
647 static inline struct blkg_policy_data *blkg_to_pd(struct blkcg_gq *blkg,
648 struct blkcg_policy *pol) { return NULL; }
649 static inline struct blkcg_gq *pd_to_blkg(struct blkg_policy_data *pd) { return NULL; }
650 static inline char *blkg_path(struct blkcg_gq *blkg) { return NULL; }
651 static inline void blkg_get(struct blkcg_gq *blkg) { }
652 static inline void blkg_put(struct blkcg_gq *blkg) { }
654 static inline bool blkcg_punt_bio_submit(struct bio *bio) { return false; }
655 static inline void blkcg_bio_issue_init(struct bio *bio) { }
656 static inline void blk_cgroup_bio_start(struct bio *bio) { }
658 #define blk_queue_for_each_rl(rl, q) \
659 for ((rl) = &(q)->root_rl; (rl); (rl) = NULL)
661 #endif /* CONFIG_BLOCK */
662 #endif /* CONFIG_BLK_CGROUP */
663 #endif /* _BLK_CGROUP_H */