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>
23 struct blkg_policy_data;
26 /* percpu_counter batch for blkg_[rw]stats, per-cpu drift doesn't matter */
27 #define BLKG_STAT_CPU_BATCH (INT_MAX / 2)
29 #ifdef CONFIG_BLK_CGROUP
31 enum blkg_iostat_type {
40 u64 bytes[BLKG_IOSTAT_NR];
41 u64 ios[BLKG_IOSTAT_NR];
44 struct blkg_iostat_set {
45 struct u64_stats_sync sync;
46 struct blkg_iostat cur;
47 struct blkg_iostat last;
50 /* association between a blk cgroup and a request queue */
52 /* Pointer to the associated request_queue */
53 struct request_queue *q;
54 struct list_head q_node;
55 struct hlist_node blkcg_node;
58 /* all non-root blkcg_gq's are guaranteed to have access to parent */
59 struct blkcg_gq *parent;
62 struct percpu_ref refcnt;
64 /* is this blkg online? protected by both blkcg and q locks */
67 struct blkg_iostat_set __percpu *iostat_cpu;
68 struct blkg_iostat_set iostat;
70 struct blkg_policy_data *pd[BLKCG_MAX_POLS];
72 spinlock_t async_bio_lock;
73 struct bio_list async_bios;
75 struct work_struct async_bio_work;
76 struct work_struct free_work;
80 atomic64_t delay_nsec;
81 atomic64_t delay_start;
85 struct rcu_head rcu_head;
89 struct cgroup_subsys_state css;
91 refcount_t online_pin;
93 struct radix_tree_root blkg_tree;
94 struct blkcg_gq __rcu *blkg_hint;
95 struct hlist_head blkg_list;
97 struct blkcg_policy_data *cpd[BLKCG_MAX_POLS];
99 struct list_head all_blkcgs_node;
100 #ifdef CONFIG_BLK_CGROUP_FC_APPID
101 char fc_app_id[FC_APPID_LEN];
103 #ifdef CONFIG_CGROUP_WRITEBACK
104 struct list_head cgwb_list;
108 static inline struct blkcg *css_to_blkcg(struct cgroup_subsys_state *css)
110 return css ? container_of(css, struct blkcg, css) : NULL;
114 * A blkcg_gq (blkg) is association between a block cgroup (blkcg) and a
115 * request_queue (q). This is used by blkcg policies which need to track
116 * information per blkcg - q pair.
118 * There can be multiple active blkcg policies and each blkg:policy pair is
119 * represented by a blkg_policy_data which is allocated and freed by each
120 * policy's pd_alloc/free_fn() methods. A policy can allocate private data
121 * area by allocating larger data structure which embeds blkg_policy_data
124 struct blkg_policy_data {
125 /* the blkg and policy id this per-policy data belongs to */
126 struct blkcg_gq *blkg;
131 * Policies that need to keep per-blkcg data which is independent from any
132 * request_queue associated to it should implement cpd_alloc/free_fn()
133 * methods. A policy can allocate private data area by allocating larger
134 * data structure which embeds blkcg_policy_data at the beginning.
135 * cpd_init() is invoked to let each policy handle per-blkcg data.
137 struct blkcg_policy_data {
138 /* the blkcg and policy id this per-policy data belongs to */
143 typedef struct blkcg_policy_data *(blkcg_pol_alloc_cpd_fn)(gfp_t gfp);
144 typedef void (blkcg_pol_init_cpd_fn)(struct blkcg_policy_data *cpd);
145 typedef void (blkcg_pol_free_cpd_fn)(struct blkcg_policy_data *cpd);
146 typedef void (blkcg_pol_bind_cpd_fn)(struct blkcg_policy_data *cpd);
147 typedef struct blkg_policy_data *(blkcg_pol_alloc_pd_fn)(gfp_t gfp,
148 struct request_queue *q, struct blkcg *blkcg);
149 typedef void (blkcg_pol_init_pd_fn)(struct blkg_policy_data *pd);
150 typedef void (blkcg_pol_online_pd_fn)(struct blkg_policy_data *pd);
151 typedef void (blkcg_pol_offline_pd_fn)(struct blkg_policy_data *pd);
152 typedef void (blkcg_pol_free_pd_fn)(struct blkg_policy_data *pd);
153 typedef void (blkcg_pol_reset_pd_stats_fn)(struct blkg_policy_data *pd);
154 typedef void (blkcg_pol_stat_pd_fn)(struct blkg_policy_data *pd,
157 struct blkcg_policy {
159 /* cgroup files for the policy */
160 struct cftype *dfl_cftypes;
161 struct cftype *legacy_cftypes;
164 blkcg_pol_alloc_cpd_fn *cpd_alloc_fn;
165 blkcg_pol_init_cpd_fn *cpd_init_fn;
166 blkcg_pol_free_cpd_fn *cpd_free_fn;
167 blkcg_pol_bind_cpd_fn *cpd_bind_fn;
169 blkcg_pol_alloc_pd_fn *pd_alloc_fn;
170 blkcg_pol_init_pd_fn *pd_init_fn;
171 blkcg_pol_online_pd_fn *pd_online_fn;
172 blkcg_pol_offline_pd_fn *pd_offline_fn;
173 blkcg_pol_free_pd_fn *pd_free_fn;
174 blkcg_pol_reset_pd_stats_fn *pd_reset_stats_fn;
175 blkcg_pol_stat_pd_fn *pd_stat_fn;
178 extern struct blkcg blkcg_root;
179 extern bool blkcg_debug_stats;
181 int blkcg_init_disk(struct gendisk *disk);
182 void blkcg_exit_disk(struct gendisk *disk);
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 block_device *bdev;
202 struct blkcg_gq *blkg;
206 struct block_device *blkcg_conf_open_bdev(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 * bio_issue_as_root_blkg - see if this bio needs to be issued as root blkg
213 * @return: true if this bio needs to be submitted with the root blkg context.
215 * In order to avoid priority inversions we sometimes need to issue a bio as if
216 * it were attached to the root blkg, and then backcharge to the actual owning
217 * blkg. The idea is we do bio_blkcg_css() to look up the actual context for
218 * the bio and attach the appropriate blkg to the bio. Then we call this helper
219 * and if it is true run with the root blkg for that queue and then do any
220 * backcharging to the originating cgroup once the io is complete.
222 static inline bool bio_issue_as_root_blkg(struct bio *bio)
224 return (bio->bi_opf & (REQ_META | REQ_SWAP)) != 0;
228 * blkg_lookup - lookup blkg for the specified blkcg - q pair
229 * @blkcg: blkcg of interest
230 * @q: request_queue of interest
232 * Lookup blkg for the @blkcg - @q pair.
234 * Must be called in a RCU critical section.
236 static inline struct blkcg_gq *blkg_lookup(struct blkcg *blkcg,
237 struct request_queue *q)
239 struct blkcg_gq *blkg;
241 WARN_ON_ONCE(!rcu_read_lock_held());
243 if (blkcg == &blkcg_root)
246 blkg = rcu_dereference(blkcg->blkg_hint);
247 if (blkg && blkg->q == q)
250 blkg = radix_tree_lookup(&blkcg->blkg_tree, q->id);
251 if (blkg && blkg->q != q)
257 * blkg_to_pdata - get policy private data
258 * @blkg: blkg of interest
259 * @pol: policy of interest
261 * Return pointer to private data associated with the @blkg-@pol pair.
263 static inline struct blkg_policy_data *blkg_to_pd(struct blkcg_gq *blkg,
264 struct blkcg_policy *pol)
266 return blkg ? blkg->pd[pol->plid] : NULL;
269 static inline struct blkcg_policy_data *blkcg_to_cpd(struct blkcg *blkcg,
270 struct blkcg_policy *pol)
272 return blkcg ? blkcg->cpd[pol->plid] : NULL;
276 * pdata_to_blkg - get blkg associated with policy private data
277 * @pd: policy private data of interest
279 * @pd is policy private data. Determine the blkg it's associated with.
281 static inline struct blkcg_gq *pd_to_blkg(struct blkg_policy_data *pd)
283 return pd ? pd->blkg : NULL;
286 static inline struct blkcg *cpd_to_blkcg(struct blkcg_policy_data *cpd)
288 return cpd ? cpd->blkcg : NULL;
292 * blkg_path - format cgroup path of blkg
293 * @blkg: blkg of interest
294 * @buf: target buffer
295 * @buflen: target buffer length
297 * Format the path of the cgroup of @blkg into @buf.
299 static inline int blkg_path(struct blkcg_gq *blkg, char *buf, int buflen)
301 return cgroup_path(blkg->blkcg->css.cgroup, buf, buflen);
305 * blkg_get - get a blkg reference
308 * The caller should be holding an existing reference.
310 static inline void blkg_get(struct blkcg_gq *blkg)
312 percpu_ref_get(&blkg->refcnt);
316 * blkg_tryget - try and get a blkg reference
319 * This is for use when doing an RCU lookup of the blkg. We may be in the midst
320 * of freeing this blkg, so we can only use it if the refcnt is not zero.
322 static inline bool blkg_tryget(struct blkcg_gq *blkg)
324 return blkg && percpu_ref_tryget(&blkg->refcnt);
328 * blkg_put - put a blkg reference
331 static inline void blkg_put(struct blkcg_gq *blkg)
333 percpu_ref_put(&blkg->refcnt);
337 * blkg_for_each_descendant_pre - pre-order walk of a blkg's descendants
338 * @d_blkg: loop cursor pointing to the current descendant
339 * @pos_css: used for iteration
340 * @p_blkg: target blkg to walk descendants of
342 * Walk @c_blkg through the descendants of @p_blkg. Must be used with RCU
343 * read locked. If called under either blkcg or queue lock, the iteration
344 * is guaranteed to include all and only online blkgs. The caller may
345 * update @pos_css by calling css_rightmost_descendant() to skip subtree.
346 * @p_blkg is included in the iteration and the first node to be visited.
348 #define blkg_for_each_descendant_pre(d_blkg, pos_css, p_blkg) \
349 css_for_each_descendant_pre((pos_css), &(p_blkg)->blkcg->css) \
350 if (((d_blkg) = blkg_lookup(css_to_blkcg(pos_css), \
354 * blkg_for_each_descendant_post - post-order walk of a blkg's descendants
355 * @d_blkg: loop cursor pointing to the current descendant
356 * @pos_css: used for iteration
357 * @p_blkg: target blkg to walk descendants of
359 * Similar to blkg_for_each_descendant_pre() but performs post-order
360 * traversal instead. Synchronization rules are the same. @p_blkg is
361 * included in the iteration and the last node to be visited.
363 #define blkg_for_each_descendant_post(d_blkg, pos_css, p_blkg) \
364 css_for_each_descendant_post((pos_css), &(p_blkg)->blkcg->css) \
365 if (((d_blkg) = blkg_lookup(css_to_blkcg(pos_css), \
368 bool __blkcg_punt_bio_submit(struct bio *bio);
370 static inline bool blkcg_punt_bio_submit(struct bio *bio)
372 if (bio->bi_opf & REQ_CGROUP_PUNT)
373 return __blkcg_punt_bio_submit(bio);
378 static inline void blkcg_bio_issue_init(struct bio *bio)
380 bio_issue_init(&bio->bi_issue, bio_sectors(bio));
383 static inline void blkcg_use_delay(struct blkcg_gq *blkg)
385 if (WARN_ON_ONCE(atomic_read(&blkg->use_delay) < 0))
387 if (atomic_add_return(1, &blkg->use_delay) == 1)
388 atomic_inc(&blkg->blkcg->css.cgroup->congestion_count);
391 static inline int blkcg_unuse_delay(struct blkcg_gq *blkg)
393 int old = atomic_read(&blkg->use_delay);
395 if (WARN_ON_ONCE(old < 0))
401 * We do this song and dance because we can race with somebody else
402 * adding or removing delay. If we just did an atomic_dec we'd end up
403 * negative and we'd already be in trouble. We need to subtract 1 and
404 * then check to see if we were the last delay so we can drop the
405 * congestion count on the cgroup.
407 while (old && !atomic_try_cmpxchg(&blkg->use_delay, &old, old - 1))
413 atomic_dec(&blkg->blkcg->css.cgroup->congestion_count);
418 * blkcg_set_delay - Enable allocator delay mechanism with the specified delay amount
420 * @delay: delay duration in nsecs
422 * When enabled with this function, the delay is not decayed and must be
423 * explicitly cleared with blkcg_clear_delay(). Must not be mixed with
424 * blkcg_[un]use_delay() and blkcg_add_delay() usages.
426 static inline void blkcg_set_delay(struct blkcg_gq *blkg, u64 delay)
428 int old = atomic_read(&blkg->use_delay);
430 /* We only want 1 person setting the congestion count for this blkg. */
431 if (!old && atomic_try_cmpxchg(&blkg->use_delay, &old, -1))
432 atomic_inc(&blkg->blkcg->css.cgroup->congestion_count);
434 atomic64_set(&blkg->delay_nsec, delay);
438 * blkcg_clear_delay - Disable allocator delay mechanism
441 * Disable use_delay mechanism. See blkcg_set_delay().
443 static inline void blkcg_clear_delay(struct blkcg_gq *blkg)
445 int old = atomic_read(&blkg->use_delay);
447 /* We only want 1 person clearing the congestion count for this blkg. */
448 if (old && atomic_try_cmpxchg(&blkg->use_delay, &old, 0))
449 atomic_dec(&blkg->blkcg->css.cgroup->congestion_count);
453 * blk_cgroup_mergeable - Determine whether to allow or disallow merges
454 * @rq: request to merge into
457 * @bio and @rq should belong to the same cgroup and their issue_as_root should
458 * match. The latter is necessary as we don't want to throttle e.g. a metadata
459 * update because it happens to be next to a regular IO.
461 static inline bool blk_cgroup_mergeable(struct request *rq, struct bio *bio)
463 return rq->bio->bi_blkg == bio->bi_blkg &&
464 bio_issue_as_root_blkg(rq->bio) == bio_issue_as_root_blkg(bio);
467 void blk_cgroup_bio_start(struct bio *bio);
468 void blkcg_add_delay(struct blkcg_gq *blkg, u64 now, u64 delta);
469 #else /* CONFIG_BLK_CGROUP */
471 struct blkg_policy_data {
474 struct blkcg_policy_data {
477 struct blkcg_policy {
483 static inline struct blkcg_gq *blkg_lookup(struct blkcg *blkcg, void *key) { return NULL; }
484 static inline int blkcg_init_disk(struct gendisk *disk) { return 0; }
485 static inline void blkcg_exit_disk(struct gendisk *disk) { }
486 static inline int blkcg_policy_register(struct blkcg_policy *pol) { return 0; }
487 static inline void blkcg_policy_unregister(struct blkcg_policy *pol) { }
488 static inline int blkcg_activate_policy(struct request_queue *q,
489 const struct blkcg_policy *pol) { return 0; }
490 static inline void blkcg_deactivate_policy(struct request_queue *q,
491 const struct blkcg_policy *pol) { }
493 static inline struct blkg_policy_data *blkg_to_pd(struct blkcg_gq *blkg,
494 struct blkcg_policy *pol) { return NULL; }
495 static inline struct blkcg_gq *pd_to_blkg(struct blkg_policy_data *pd) { return NULL; }
496 static inline char *blkg_path(struct blkcg_gq *blkg) { return NULL; }
497 static inline void blkg_get(struct blkcg_gq *blkg) { }
498 static inline void blkg_put(struct blkcg_gq *blkg) { }
500 static inline bool blkcg_punt_bio_submit(struct bio *bio) { return false; }
501 static inline void blkcg_bio_issue_init(struct bio *bio) { }
502 static inline void blk_cgroup_bio_start(struct bio *bio) { }
503 static inline bool blk_cgroup_mergeable(struct request *rq, struct bio *bio) { return true; }
505 #define blk_queue_for_each_rl(rl, q) \
506 for ((rl) = &(q)->root_rl; (rl); (rl) = NULL)
508 #endif /* CONFIG_BLK_CGROUP */
510 #endif /* _BLK_CGROUP_PRIVATE_H */