1 /* SPDX-License-Identifier: GPL-2.0 */
2 #ifndef _BLK_CGROUP_PRIVATE_H
3 #define _BLK_CGROUP_PRIVATE_H
5 * block cgroup private header
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/blk-cgroup.h>
18 #include <linux/cgroup.h>
19 #include <linux/kthread.h>
20 #include <linux/blk-mq.h>
21 #include <linux/llist.h>
24 struct blkg_policy_data;
27 /* percpu_counter batch for blkg_[rw]stats, per-cpu drift doesn't matter */
28 #define BLKG_STAT_CPU_BATCH (INT_MAX / 2)
30 #ifdef CONFIG_BLK_CGROUP
32 enum blkg_iostat_type {
41 u64 bytes[BLKG_IOSTAT_NR];
42 u64 ios[BLKG_IOSTAT_NR];
45 struct blkg_iostat_set {
46 struct u64_stats_sync sync;
47 struct blkcg_gq *blkg;
48 struct llist_node lnode;
49 int lqueued; /* queued in llist */
50 struct blkg_iostat cur;
51 struct blkg_iostat last;
54 /* association between a blk cgroup and a request queue */
56 /* Pointer to the associated request_queue */
57 struct request_queue *q;
58 struct list_head q_node;
59 struct hlist_node blkcg_node;
62 /* all non-root blkcg_gq's are guaranteed to have access to parent */
63 struct blkcg_gq *parent;
66 struct percpu_ref refcnt;
68 /* is this blkg online? protected by both blkcg and q locks */
71 struct blkg_iostat_set __percpu *iostat_cpu;
72 struct blkg_iostat_set iostat;
74 struct blkg_policy_data *pd[BLKCG_MAX_POLS];
76 spinlock_t async_bio_lock;
77 struct bio_list async_bios;
79 struct work_struct async_bio_work;
80 struct work_struct free_work;
84 atomic64_t delay_nsec;
85 atomic64_t delay_start;
89 struct rcu_head rcu_head;
93 struct cgroup_subsys_state css;
95 refcount_t online_pin;
97 struct radix_tree_root blkg_tree;
98 struct blkcg_gq __rcu *blkg_hint;
99 struct hlist_head blkg_list;
101 struct blkcg_policy_data *cpd[BLKCG_MAX_POLS];
103 struct list_head all_blkcgs_node;
106 * List of updated percpu blkg_iostat_set's since the last flush.
108 struct llist_head __percpu *lhead;
110 #ifdef CONFIG_BLK_CGROUP_FC_APPID
111 char fc_app_id[FC_APPID_LEN];
113 #ifdef CONFIG_CGROUP_WRITEBACK
114 struct list_head cgwb_list;
118 static inline struct blkcg *css_to_blkcg(struct cgroup_subsys_state *css)
120 return css ? container_of(css, struct blkcg, css) : NULL;
124 * A blkcg_gq (blkg) is association between a block cgroup (blkcg) and a
125 * request_queue (q). This is used by blkcg policies which need to track
126 * information per blkcg - q pair.
128 * There can be multiple active blkcg policies and each blkg:policy pair is
129 * represented by a blkg_policy_data which is allocated and freed by each
130 * policy's pd_alloc/free_fn() methods. A policy can allocate private data
131 * area by allocating larger data structure which embeds blkg_policy_data
134 struct blkg_policy_data {
135 /* the blkg and policy id this per-policy data belongs to */
136 struct blkcg_gq *blkg;
142 * Policies that need to keep per-blkcg data which is independent from any
143 * request_queue associated to it should implement cpd_alloc/free_fn()
144 * methods. A policy can allocate private data area by allocating larger
145 * data structure which embeds blkcg_policy_data at the beginning.
146 * cpd_init() is invoked to let each policy handle per-blkcg data.
148 struct blkcg_policy_data {
149 /* the blkcg and policy id this per-policy data belongs to */
154 typedef struct blkcg_policy_data *(blkcg_pol_alloc_cpd_fn)(gfp_t gfp);
155 typedef void (blkcg_pol_init_cpd_fn)(struct blkcg_policy_data *cpd);
156 typedef void (blkcg_pol_free_cpd_fn)(struct blkcg_policy_data *cpd);
157 typedef void (blkcg_pol_bind_cpd_fn)(struct blkcg_policy_data *cpd);
158 typedef struct blkg_policy_data *(blkcg_pol_alloc_pd_fn)(struct gendisk *disk,
159 struct blkcg *blkcg, gfp_t gfp);
160 typedef void (blkcg_pol_init_pd_fn)(struct blkg_policy_data *pd);
161 typedef void (blkcg_pol_online_pd_fn)(struct blkg_policy_data *pd);
162 typedef void (blkcg_pol_offline_pd_fn)(struct blkg_policy_data *pd);
163 typedef void (blkcg_pol_free_pd_fn)(struct blkg_policy_data *pd);
164 typedef void (blkcg_pol_reset_pd_stats_fn)(struct blkg_policy_data *pd);
165 typedef void (blkcg_pol_stat_pd_fn)(struct blkg_policy_data *pd,
168 struct blkcg_policy {
170 /* cgroup files for the policy */
171 struct cftype *dfl_cftypes;
172 struct cftype *legacy_cftypes;
175 blkcg_pol_alloc_cpd_fn *cpd_alloc_fn;
176 blkcg_pol_init_cpd_fn *cpd_init_fn;
177 blkcg_pol_free_cpd_fn *cpd_free_fn;
178 blkcg_pol_bind_cpd_fn *cpd_bind_fn;
180 blkcg_pol_alloc_pd_fn *pd_alloc_fn;
181 blkcg_pol_init_pd_fn *pd_init_fn;
182 blkcg_pol_online_pd_fn *pd_online_fn;
183 blkcg_pol_offline_pd_fn *pd_offline_fn;
184 blkcg_pol_free_pd_fn *pd_free_fn;
185 blkcg_pol_reset_pd_stats_fn *pd_reset_stats_fn;
186 blkcg_pol_stat_pd_fn *pd_stat_fn;
189 extern struct blkcg blkcg_root;
190 extern bool blkcg_debug_stats;
192 int blkcg_init_disk(struct gendisk *disk);
193 void blkcg_exit_disk(struct gendisk *disk);
195 /* Blkio controller policy registration */
196 int blkcg_policy_register(struct blkcg_policy *pol);
197 void blkcg_policy_unregister(struct blkcg_policy *pol);
198 int blkcg_activate_policy(struct gendisk *disk, const struct blkcg_policy *pol);
199 void blkcg_deactivate_policy(struct gendisk *disk,
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);
210 struct blkg_conf_ctx {
211 struct block_device *bdev;
212 struct blkcg_gq *blkg;
216 struct block_device *blkcg_conf_open_bdev(char **inputp);
217 int blkg_conf_prep(struct blkcg *blkcg, const struct blkcg_policy *pol,
218 char *input, struct blkg_conf_ctx *ctx);
219 void blkg_conf_finish(struct blkg_conf_ctx *ctx);
222 * bio_issue_as_root_blkg - see if this bio needs to be issued as root blkg
223 * @return: true if this bio needs to be submitted with the root blkg context.
225 * In order to avoid priority inversions we sometimes need to issue a bio as if
226 * it were attached to the root blkg, and then backcharge to the actual owning
227 * blkg. The idea is we do bio_blkcg_css() to look up the actual context for
228 * the bio and attach the appropriate blkg to the bio. Then we call this helper
229 * and if it is true run with the root blkg for that queue and then do any
230 * backcharging to the originating cgroup once the io is complete.
232 static inline bool bio_issue_as_root_blkg(struct bio *bio)
234 return (bio->bi_opf & (REQ_META | REQ_SWAP)) != 0;
238 * blkg_lookup - lookup blkg for the specified blkcg - q pair
239 * @blkcg: blkcg of interest
240 * @q: request_queue of interest
242 * Lookup blkg for the @blkcg - @q pair.
244 * Must be called in a RCU critical section.
246 static inline struct blkcg_gq *blkg_lookup(struct blkcg *blkcg,
247 struct request_queue *q)
249 struct blkcg_gq *blkg;
251 WARN_ON_ONCE(!rcu_read_lock_held());
253 if (blkcg == &blkcg_root)
256 blkg = rcu_dereference(blkcg->blkg_hint);
257 if (blkg && blkg->q == q)
260 blkg = radix_tree_lookup(&blkcg->blkg_tree, q->id);
261 if (blkg && blkg->q != q)
267 * blkg_to_pdata - get policy private data
268 * @blkg: blkg of interest
269 * @pol: policy of interest
271 * Return pointer to private data associated with the @blkg-@pol pair.
273 static inline struct blkg_policy_data *blkg_to_pd(struct blkcg_gq *blkg,
274 struct blkcg_policy *pol)
276 return blkg ? blkg->pd[pol->plid] : NULL;
279 static inline struct blkcg_policy_data *blkcg_to_cpd(struct blkcg *blkcg,
280 struct blkcg_policy *pol)
282 return blkcg ? blkcg->cpd[pol->plid] : NULL;
286 * pdata_to_blkg - get blkg associated with policy private data
287 * @pd: policy private data of interest
289 * @pd is policy private data. Determine the blkg it's associated with.
291 static inline struct blkcg_gq *pd_to_blkg(struct blkg_policy_data *pd)
293 return pd ? pd->blkg : NULL;
296 static inline struct blkcg *cpd_to_blkcg(struct blkcg_policy_data *cpd)
298 return cpd ? cpd->blkcg : NULL;
302 * blkg_path - format cgroup path of blkg
303 * @blkg: blkg of interest
304 * @buf: target buffer
305 * @buflen: target buffer length
307 * Format the path of the cgroup of @blkg into @buf.
309 static inline int blkg_path(struct blkcg_gq *blkg, char *buf, int buflen)
311 return cgroup_path(blkg->blkcg->css.cgroup, buf, buflen);
315 * blkg_get - get a blkg reference
318 * The caller should be holding an existing reference.
320 static inline void blkg_get(struct blkcg_gq *blkg)
322 percpu_ref_get(&blkg->refcnt);
326 * blkg_tryget - try and get a blkg reference
329 * This is for use when doing an RCU lookup of the blkg. We may be in the midst
330 * of freeing this blkg, so we can only use it if the refcnt is not zero.
332 static inline bool blkg_tryget(struct blkcg_gq *blkg)
334 return blkg && percpu_ref_tryget(&blkg->refcnt);
338 * blkg_put - put a blkg reference
341 static inline void blkg_put(struct blkcg_gq *blkg)
343 percpu_ref_put(&blkg->refcnt);
347 * blkg_for_each_descendant_pre - pre-order walk of a blkg's descendants
348 * @d_blkg: loop cursor pointing to the current descendant
349 * @pos_css: used for iteration
350 * @p_blkg: target blkg to walk descendants of
352 * Walk @c_blkg through the descendants of @p_blkg. Must be used with RCU
353 * read locked. If called under either blkcg or queue lock, the iteration
354 * is guaranteed to include all and only online blkgs. The caller may
355 * update @pos_css by calling css_rightmost_descendant() to skip subtree.
356 * @p_blkg is included in the iteration and the first node to be visited.
358 #define blkg_for_each_descendant_pre(d_blkg, pos_css, p_blkg) \
359 css_for_each_descendant_pre((pos_css), &(p_blkg)->blkcg->css) \
360 if (((d_blkg) = blkg_lookup(css_to_blkcg(pos_css), \
364 * blkg_for_each_descendant_post - post-order walk of a blkg's descendants
365 * @d_blkg: loop cursor pointing to the current descendant
366 * @pos_css: used for iteration
367 * @p_blkg: target blkg to walk descendants of
369 * Similar to blkg_for_each_descendant_pre() but performs post-order
370 * traversal instead. Synchronization rules are the same. @p_blkg is
371 * included in the iteration and the last node to be visited.
373 #define blkg_for_each_descendant_post(d_blkg, pos_css, p_blkg) \
374 css_for_each_descendant_post((pos_css), &(p_blkg)->blkcg->css) \
375 if (((d_blkg) = blkg_lookup(css_to_blkcg(pos_css), \
378 bool __blkcg_punt_bio_submit(struct bio *bio);
380 static inline bool blkcg_punt_bio_submit(struct bio *bio)
382 if (bio->bi_opf & REQ_CGROUP_PUNT)
383 return __blkcg_punt_bio_submit(bio);
388 static inline void blkcg_bio_issue_init(struct bio *bio)
390 bio_issue_init(&bio->bi_issue, bio_sectors(bio));
393 static inline void blkcg_use_delay(struct blkcg_gq *blkg)
395 if (WARN_ON_ONCE(atomic_read(&blkg->use_delay) < 0))
397 if (atomic_add_return(1, &blkg->use_delay) == 1)
398 atomic_inc(&blkg->blkcg->css.cgroup->congestion_count);
401 static inline int blkcg_unuse_delay(struct blkcg_gq *blkg)
403 int old = atomic_read(&blkg->use_delay);
405 if (WARN_ON_ONCE(old < 0))
411 * We do this song and dance because we can race with somebody else
412 * adding or removing delay. If we just did an atomic_dec we'd end up
413 * negative and we'd already be in trouble. We need to subtract 1 and
414 * then check to see if we were the last delay so we can drop the
415 * congestion count on the cgroup.
417 while (old && !atomic_try_cmpxchg(&blkg->use_delay, &old, old - 1))
423 atomic_dec(&blkg->blkcg->css.cgroup->congestion_count);
428 * blkcg_set_delay - Enable allocator delay mechanism with the specified delay amount
430 * @delay: delay duration in nsecs
432 * When enabled with this function, the delay is not decayed and must be
433 * explicitly cleared with blkcg_clear_delay(). Must not be mixed with
434 * blkcg_[un]use_delay() and blkcg_add_delay() usages.
436 static inline void blkcg_set_delay(struct blkcg_gq *blkg, u64 delay)
438 int old = atomic_read(&blkg->use_delay);
440 /* We only want 1 person setting the congestion count for this blkg. */
441 if (!old && atomic_try_cmpxchg(&blkg->use_delay, &old, -1))
442 atomic_inc(&blkg->blkcg->css.cgroup->congestion_count);
444 atomic64_set(&blkg->delay_nsec, delay);
448 * blkcg_clear_delay - Disable allocator delay mechanism
451 * Disable use_delay mechanism. See blkcg_set_delay().
453 static inline void blkcg_clear_delay(struct blkcg_gq *blkg)
455 int old = atomic_read(&blkg->use_delay);
457 /* We only want 1 person clearing the congestion count for this blkg. */
458 if (old && atomic_try_cmpxchg(&blkg->use_delay, &old, 0))
459 atomic_dec(&blkg->blkcg->css.cgroup->congestion_count);
463 * blk_cgroup_mergeable - Determine whether to allow or disallow merges
464 * @rq: request to merge into
467 * @bio and @rq should belong to the same cgroup and their issue_as_root should
468 * match. The latter is necessary as we don't want to throttle e.g. a metadata
469 * update because it happens to be next to a regular IO.
471 static inline bool blk_cgroup_mergeable(struct request *rq, struct bio *bio)
473 return rq->bio->bi_blkg == bio->bi_blkg &&
474 bio_issue_as_root_blkg(rq->bio) == bio_issue_as_root_blkg(bio);
477 void blk_cgroup_bio_start(struct bio *bio);
478 void blkcg_add_delay(struct blkcg_gq *blkg, u64 now, u64 delta);
479 #else /* CONFIG_BLK_CGROUP */
481 struct blkg_policy_data {
484 struct blkcg_policy_data {
487 struct blkcg_policy {
493 static inline struct blkcg_gq *blkg_lookup(struct blkcg *blkcg, void *key) { return NULL; }
494 static inline int blkcg_init_disk(struct gendisk *disk) { return 0; }
495 static inline void blkcg_exit_disk(struct gendisk *disk) { }
496 static inline int blkcg_policy_register(struct blkcg_policy *pol) { return 0; }
497 static inline void blkcg_policy_unregister(struct blkcg_policy *pol) { }
498 static inline int blkcg_activate_policy(struct gendisk *disk,
499 const struct blkcg_policy *pol) { return 0; }
500 static inline void blkcg_deactivate_policy(struct gendisk *disk,
501 const struct blkcg_policy *pol) { }
503 static inline struct blkg_policy_data *blkg_to_pd(struct blkcg_gq *blkg,
504 struct blkcg_policy *pol) { return NULL; }
505 static inline struct blkcg_gq *pd_to_blkg(struct blkg_policy_data *pd) { return NULL; }
506 static inline char *blkg_path(struct blkcg_gq *blkg) { return NULL; }
507 static inline void blkg_get(struct blkcg_gq *blkg) { }
508 static inline void blkg_put(struct blkcg_gq *blkg) { }
510 static inline bool blkcg_punt_bio_submit(struct bio *bio) { return false; }
511 static inline void blkcg_bio_issue_init(struct bio *bio) { }
512 static inline void blk_cgroup_bio_start(struct bio *bio) { }
513 static inline bool blk_cgroup_mergeable(struct request *rq, struct bio *bio) { return true; }
515 #define blk_queue_for_each_rl(rl, q) \
516 for ((rl) = &(q)->root_rl; (rl); (rl) = NULL)
518 #endif /* CONFIG_BLK_CGROUP */
520 #endif /* _BLK_CGROUP_PRIVATE_H */