500 lines · c
1/* SPDX-License-Identifier: GPL-2.0 */2#ifndef _BLK_CGROUP_PRIVATE_H3#define _BLK_CGROUP_PRIVATE_H4/*5 * block cgroup private header6 *7 * Based on ideas and code from CFQ, CFS and BFQ:8 * Copyright (C) 2003 Jens Axboe <axboe@kernel.dk>9 *10 * Copyright (C) 2008 Fabio Checconi <fabio@gandalf.sssup.it>11 * Paolo Valente <paolo.valente@unimore.it>12 *13 * Copyright (C) 2009 Vivek Goyal <vgoyal@redhat.com>14 * Nauman Rafique <nauman@google.com>15 */16 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>22#include "blk.h"23 24struct blkcg_gq;25struct blkg_policy_data;26 27 28/* percpu_counter batch for blkg_[rw]stats, per-cpu drift doesn't matter */29#define BLKG_STAT_CPU_BATCH (INT_MAX / 2)30 31#ifdef CONFIG_BLK_CGROUP32 33enum blkg_iostat_type {34 BLKG_IOSTAT_READ,35 BLKG_IOSTAT_WRITE,36 BLKG_IOSTAT_DISCARD,37 38 BLKG_IOSTAT_NR,39};40 41struct blkg_iostat {42 u64 bytes[BLKG_IOSTAT_NR];43 u64 ios[BLKG_IOSTAT_NR];44};45 46struct blkg_iostat_set {47 struct u64_stats_sync sync;48 struct blkcg_gq *blkg;49 struct llist_node lnode;50 int lqueued; /* queued in llist */51 struct blkg_iostat cur;52 struct blkg_iostat last;53};54 55/* association between a blk cgroup and a request queue */56struct blkcg_gq {57 /* Pointer to the associated request_queue */58 struct request_queue *q;59 struct list_head q_node;60 struct hlist_node blkcg_node;61 struct blkcg *blkcg;62 63 /* all non-root blkcg_gq's are guaranteed to have access to parent */64 struct blkcg_gq *parent;65 66 /* reference count */67 struct percpu_ref refcnt;68 69 /* is this blkg online? protected by both blkcg and q locks */70 bool online;71 72 struct blkg_iostat_set __percpu *iostat_cpu;73 struct blkg_iostat_set iostat;74 75 struct blkg_policy_data *pd[BLKCG_MAX_POLS];76#ifdef CONFIG_BLK_CGROUP_PUNT_BIO77 spinlock_t async_bio_lock;78 struct bio_list async_bios;79#endif80 union {81 struct work_struct async_bio_work;82 struct work_struct free_work;83 };84 85 atomic_t use_delay;86 atomic64_t delay_nsec;87 atomic64_t delay_start;88 u64 last_delay;89 int last_use;90 91 struct rcu_head rcu_head;92};93 94struct blkcg {95 struct cgroup_subsys_state css;96 spinlock_t lock;97 refcount_t online_pin;98 /* If there is block congestion on this cgroup. */99 atomic_t congestion_count;100 101 struct radix_tree_root blkg_tree;102 struct blkcg_gq __rcu *blkg_hint;103 struct hlist_head blkg_list;104 105 struct blkcg_policy_data *cpd[BLKCG_MAX_POLS];106 107 struct list_head all_blkcgs_node;108 109 /*110 * List of updated percpu blkg_iostat_set's since the last flush.111 */112 struct llist_head __percpu *lhead;113 114#ifdef CONFIG_BLK_CGROUP_FC_APPID115 char fc_app_id[FC_APPID_LEN];116#endif117#ifdef CONFIG_CGROUP_WRITEBACK118 struct list_head cgwb_list;119#endif120};121 122static inline struct blkcg *css_to_blkcg(struct cgroup_subsys_state *css)123{124 return css ? container_of(css, struct blkcg, css) : NULL;125}126 127/*128 * A blkcg_gq (blkg) is association between a block cgroup (blkcg) and a129 * request_queue (q). This is used by blkcg policies which need to track130 * information per blkcg - q pair.131 *132 * There can be multiple active blkcg policies and each blkg:policy pair is133 * represented by a blkg_policy_data which is allocated and freed by each134 * policy's pd_alloc/free_fn() methods. A policy can allocate private data135 * area by allocating larger data structure which embeds blkg_policy_data136 * at the beginning.137 */138struct blkg_policy_data {139 /* the blkg and policy id this per-policy data belongs to */140 struct blkcg_gq *blkg;141 int plid;142 bool online;143};144 145/*146 * Policies that need to keep per-blkcg data which is independent from any147 * request_queue associated to it should implement cpd_alloc/free_fn()148 * methods. A policy can allocate private data area by allocating larger149 * data structure which embeds blkcg_policy_data at the beginning.150 * cpd_init() is invoked to let each policy handle per-blkcg data.151 */152struct blkcg_policy_data {153 /* the blkcg and policy id this per-policy data belongs to */154 struct blkcg *blkcg;155 int plid;156};157 158typedef struct blkcg_policy_data *(blkcg_pol_alloc_cpd_fn)(gfp_t gfp);159typedef void (blkcg_pol_init_cpd_fn)(struct blkcg_policy_data *cpd);160typedef void (blkcg_pol_free_cpd_fn)(struct blkcg_policy_data *cpd);161typedef void (blkcg_pol_bind_cpd_fn)(struct blkcg_policy_data *cpd);162typedef struct blkg_policy_data *(blkcg_pol_alloc_pd_fn)(struct gendisk *disk,163 struct blkcg *blkcg, gfp_t gfp);164typedef void (blkcg_pol_init_pd_fn)(struct blkg_policy_data *pd);165typedef void (blkcg_pol_online_pd_fn)(struct blkg_policy_data *pd);166typedef void (blkcg_pol_offline_pd_fn)(struct blkg_policy_data *pd);167typedef void (blkcg_pol_free_pd_fn)(struct blkg_policy_data *pd);168typedef void (blkcg_pol_reset_pd_stats_fn)(struct blkg_policy_data *pd);169typedef void (blkcg_pol_stat_pd_fn)(struct blkg_policy_data *pd,170 struct seq_file *s);171 172struct blkcg_policy {173 int plid;174 /* cgroup files for the policy */175 struct cftype *dfl_cftypes;176 struct cftype *legacy_cftypes;177 178 /* operations */179 blkcg_pol_alloc_cpd_fn *cpd_alloc_fn;180 blkcg_pol_free_cpd_fn *cpd_free_fn;181 182 blkcg_pol_alloc_pd_fn *pd_alloc_fn;183 blkcg_pol_init_pd_fn *pd_init_fn;184 blkcg_pol_online_pd_fn *pd_online_fn;185 blkcg_pol_offline_pd_fn *pd_offline_fn;186 blkcg_pol_free_pd_fn *pd_free_fn;187 blkcg_pol_reset_pd_stats_fn *pd_reset_stats_fn;188 blkcg_pol_stat_pd_fn *pd_stat_fn;189};190 191extern struct blkcg blkcg_root;192extern bool blkcg_debug_stats;193 194void blkg_init_queue(struct request_queue *q);195int blkcg_init_disk(struct gendisk *disk);196void blkcg_exit_disk(struct gendisk *disk);197 198/* Blkio controller policy registration */199int blkcg_policy_register(struct blkcg_policy *pol);200void blkcg_policy_unregister(struct blkcg_policy *pol);201int blkcg_activate_policy(struct gendisk *disk, const struct blkcg_policy *pol);202void blkcg_deactivate_policy(struct gendisk *disk,203 const struct blkcg_policy *pol);204 205const char *blkg_dev_name(struct blkcg_gq *blkg);206void blkcg_print_blkgs(struct seq_file *sf, struct blkcg *blkcg,207 u64 (*prfill)(struct seq_file *,208 struct blkg_policy_data *, int),209 const struct blkcg_policy *pol, int data,210 bool show_total);211u64 __blkg_prfill_u64(struct seq_file *sf, struct blkg_policy_data *pd, u64 v);212 213struct blkg_conf_ctx {214 char *input;215 char *body;216 struct block_device *bdev;217 struct blkcg_gq *blkg;218};219 220void blkg_conf_init(struct blkg_conf_ctx *ctx, char *input);221int blkg_conf_open_bdev(struct blkg_conf_ctx *ctx);222int blkg_conf_prep(struct blkcg *blkcg, const struct blkcg_policy *pol,223 struct blkg_conf_ctx *ctx);224void blkg_conf_exit(struct blkg_conf_ctx *ctx);225 226/**227 * bio_issue_as_root_blkg - see if this bio needs to be issued as root blkg228 * @return: true if this bio needs to be submitted with the root blkg context.229 *230 * In order to avoid priority inversions we sometimes need to issue a bio as if231 * it were attached to the root blkg, and then backcharge to the actual owning232 * blkg. The idea is we do bio_blkcg_css() to look up the actual context for233 * the bio and attach the appropriate blkg to the bio. Then we call this helper234 * and if it is true run with the root blkg for that queue and then do any235 * backcharging to the originating cgroup once the io is complete.236 */237static inline bool bio_issue_as_root_blkg(struct bio *bio)238{239 return (bio->bi_opf & (REQ_META | REQ_SWAP)) != 0;240}241 242/**243 * blkg_lookup - lookup blkg for the specified blkcg - q pair244 * @blkcg: blkcg of interest245 * @q: request_queue of interest246 *247 * Lookup blkg for the @blkcg - @q pair.248 249 * Must be called in a RCU critical section.250 */251static inline struct blkcg_gq *blkg_lookup(struct blkcg *blkcg,252 struct request_queue *q)253{254 struct blkcg_gq *blkg;255 256 if (blkcg == &blkcg_root)257 return q->root_blkg;258 259 blkg = rcu_dereference_check(blkcg->blkg_hint,260 lockdep_is_held(&q->queue_lock));261 if (blkg && blkg->q == q)262 return blkg;263 264 blkg = radix_tree_lookup(&blkcg->blkg_tree, q->id);265 if (blkg && blkg->q != q)266 blkg = NULL;267 return blkg;268}269 270/**271 * blkg_to_pdata - get policy private data272 * @blkg: blkg of interest273 * @pol: policy of interest274 *275 * Return pointer to private data associated with the @blkg-@pol pair.276 */277static inline struct blkg_policy_data *blkg_to_pd(struct blkcg_gq *blkg,278 struct blkcg_policy *pol)279{280 return blkg ? blkg->pd[pol->plid] : NULL;281}282 283static inline struct blkcg_policy_data *blkcg_to_cpd(struct blkcg *blkcg,284 struct blkcg_policy *pol)285{286 return blkcg ? blkcg->cpd[pol->plid] : NULL;287}288 289/**290 * pdata_to_blkg - get blkg associated with policy private data291 * @pd: policy private data of interest292 *293 * @pd is policy private data. Determine the blkg it's associated with.294 */295static inline struct blkcg_gq *pd_to_blkg(struct blkg_policy_data *pd)296{297 return pd ? pd->blkg : NULL;298}299 300static inline struct blkcg *cpd_to_blkcg(struct blkcg_policy_data *cpd)301{302 return cpd ? cpd->blkcg : NULL;303}304 305/**306 * blkg_get - get a blkg reference307 * @blkg: blkg to get308 *309 * The caller should be holding an existing reference.310 */311static inline void blkg_get(struct blkcg_gq *blkg)312{313 percpu_ref_get(&blkg->refcnt);314}315 316/**317 * blkg_tryget - try and get a blkg reference318 * @blkg: blkg to get319 *320 * This is for use when doing an RCU lookup of the blkg. We may be in the midst321 * of freeing this blkg, so we can only use it if the refcnt is not zero.322 */323static inline bool blkg_tryget(struct blkcg_gq *blkg)324{325 return blkg && percpu_ref_tryget(&blkg->refcnt);326}327 328/**329 * blkg_put - put a blkg reference330 * @blkg: blkg to put331 */332static inline void blkg_put(struct blkcg_gq *blkg)333{334 percpu_ref_put(&blkg->refcnt);335}336 337/**338 * blkg_for_each_descendant_pre - pre-order walk of a blkg's descendants339 * @d_blkg: loop cursor pointing to the current descendant340 * @pos_css: used for iteration341 * @p_blkg: target blkg to walk descendants of342 *343 * Walk @c_blkg through the descendants of @p_blkg. Must be used with RCU344 * read locked. If called under either blkcg or queue lock, the iteration345 * is guaranteed to include all and only online blkgs. The caller may346 * update @pos_css by calling css_rightmost_descendant() to skip subtree.347 * @p_blkg is included in the iteration and the first node to be visited.348 */349#define blkg_for_each_descendant_pre(d_blkg, pos_css, p_blkg) \350 css_for_each_descendant_pre((pos_css), &(p_blkg)->blkcg->css) \351 if (((d_blkg) = blkg_lookup(css_to_blkcg(pos_css), \352 (p_blkg)->q)))353 354/**355 * blkg_for_each_descendant_post - post-order walk of a blkg's descendants356 * @d_blkg: loop cursor pointing to the current descendant357 * @pos_css: used for iteration358 * @p_blkg: target blkg to walk descendants of359 *360 * Similar to blkg_for_each_descendant_pre() but performs post-order361 * traversal instead. Synchronization rules are the same. @p_blkg is362 * included in the iteration and the last node to be visited.363 */364#define blkg_for_each_descendant_post(d_blkg, pos_css, p_blkg) \365 css_for_each_descendant_post((pos_css), &(p_blkg)->blkcg->css) \366 if (((d_blkg) = blkg_lookup(css_to_blkcg(pos_css), \367 (p_blkg)->q)))368 369static inline void blkcg_bio_issue_init(struct bio *bio)370{371 bio_issue_init(&bio->bi_issue, bio_sectors(bio));372}373 374static inline void blkcg_use_delay(struct blkcg_gq *blkg)375{376 if (WARN_ON_ONCE(atomic_read(&blkg->use_delay) < 0))377 return;378 if (atomic_add_return(1, &blkg->use_delay) == 1)379 atomic_inc(&blkg->blkcg->congestion_count);380}381 382static inline int blkcg_unuse_delay(struct blkcg_gq *blkg)383{384 int old = atomic_read(&blkg->use_delay);385 386 if (WARN_ON_ONCE(old < 0))387 return 0;388 if (old == 0)389 return 0;390 391 /*392 * We do this song and dance because we can race with somebody else393 * adding or removing delay. If we just did an atomic_dec we'd end up394 * negative and we'd already be in trouble. We need to subtract 1 and395 * then check to see if we were the last delay so we can drop the396 * congestion count on the cgroup.397 */398 while (old && !atomic_try_cmpxchg(&blkg->use_delay, &old, old - 1))399 ;400 401 if (old == 0)402 return 0;403 if (old == 1)404 atomic_dec(&blkg->blkcg->congestion_count);405 return 1;406}407 408/**409 * blkcg_set_delay - Enable allocator delay mechanism with the specified delay amount410 * @blkg: target blkg411 * @delay: delay duration in nsecs412 *413 * When enabled with this function, the delay is not decayed and must be414 * explicitly cleared with blkcg_clear_delay(). Must not be mixed with415 * blkcg_[un]use_delay() and blkcg_add_delay() usages.416 */417static inline void blkcg_set_delay(struct blkcg_gq *blkg, u64 delay)418{419 int old = atomic_read(&blkg->use_delay);420 421 /* We only want 1 person setting the congestion count for this blkg. */422 if (!old && atomic_try_cmpxchg(&blkg->use_delay, &old, -1))423 atomic_inc(&blkg->blkcg->congestion_count);424 425 atomic64_set(&blkg->delay_nsec, delay);426}427 428/**429 * blkcg_clear_delay - Disable allocator delay mechanism430 * @blkg: target blkg431 *432 * Disable use_delay mechanism. See blkcg_set_delay().433 */434static inline void blkcg_clear_delay(struct blkcg_gq *blkg)435{436 int old = atomic_read(&blkg->use_delay);437 438 /* We only want 1 person clearing the congestion count for this blkg. */439 if (old && atomic_try_cmpxchg(&blkg->use_delay, &old, 0))440 atomic_dec(&blkg->blkcg->congestion_count);441}442 443/**444 * blk_cgroup_mergeable - Determine whether to allow or disallow merges445 * @rq: request to merge into446 * @bio: bio to merge447 *448 * @bio and @rq should belong to the same cgroup and their issue_as_root should449 * match. The latter is necessary as we don't want to throttle e.g. a metadata450 * update because it happens to be next to a regular IO.451 */452static inline bool blk_cgroup_mergeable(struct request *rq, struct bio *bio)453{454 return rq->bio->bi_blkg == bio->bi_blkg &&455 bio_issue_as_root_blkg(rq->bio) == bio_issue_as_root_blkg(bio);456}457 458void blk_cgroup_bio_start(struct bio *bio);459void blkcg_add_delay(struct blkcg_gq *blkg, u64 now, u64 delta);460#else /* CONFIG_BLK_CGROUP */461 462struct blkg_policy_data {463};464 465struct blkcg_policy_data {466};467 468struct blkcg_policy {469};470 471struct blkcg {472};473 474static inline struct blkcg_gq *blkg_lookup(struct blkcg *blkcg, void *key) { return NULL; }475static inline void blkg_init_queue(struct request_queue *q) { }476static inline int blkcg_init_disk(struct gendisk *disk) { return 0; }477static inline void blkcg_exit_disk(struct gendisk *disk) { }478static inline int blkcg_policy_register(struct blkcg_policy *pol) { return 0; }479static inline void blkcg_policy_unregister(struct blkcg_policy *pol) { }480static inline int blkcg_activate_policy(struct gendisk *disk,481 const struct blkcg_policy *pol) { return 0; }482static inline void blkcg_deactivate_policy(struct gendisk *disk,483 const struct blkcg_policy *pol) { }484 485static inline struct blkg_policy_data *blkg_to_pd(struct blkcg_gq *blkg,486 struct blkcg_policy *pol) { return NULL; }487static inline struct blkcg_gq *pd_to_blkg(struct blkg_policy_data *pd) { return NULL; }488static inline void blkg_get(struct blkcg_gq *blkg) { }489static inline void blkg_put(struct blkcg_gq *blkg) { }490static inline void blkcg_bio_issue_init(struct bio *bio) { }491static inline void blk_cgroup_bio_start(struct bio *bio) { }492static inline bool blk_cgroup_mergeable(struct request *rq, struct bio *bio) { return true; }493 494#define blk_queue_for_each_rl(rl, q) \495 for ((rl) = &(q)->root_rl; (rl); (rl) = NULL)496 497#endif /* CONFIG_BLK_CGROUP */498 499#endif /* _BLK_CGROUP_PRIVATE_H */500