1 /* SPDX-License-Identifier: GPL-2.0 */ 2 #ifndef _BLK_CGROUP_PRIVATE_H 3 #define _BLK_CGROUP_PRIVATE_H 4 /* 5 * block cgroup private header 6 * 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/blk-mq.h> 19 20 /* percpu_counter batch for blkg_[rw]stats, per-cpu drift doesn't matter */ 21 #define BLKG_STAT_CPU_BATCH (INT_MAX / 2) 22 23 #ifdef CONFIG_BLK_CGROUP 24 25 /* 26 * A blkcg_gq (blkg) is association between a block cgroup (blkcg) and a 27 * request_queue (q). This is used by blkcg policies which need to track 28 * information per blkcg - q pair. 29 * 30 * There can be multiple active blkcg policies and each blkg:policy pair is 31 * represented by a blkg_policy_data which is allocated and freed by each 32 * policy's pd_alloc/free_fn() methods. A policy can allocate private data 33 * area by allocating larger data structure which embeds blkg_policy_data 34 * at the beginning. 35 */ 36 struct blkg_policy_data { 37 /* the blkg and policy id this per-policy data belongs to */ 38 struct blkcg_gq *blkg; 39 int plid; 40 }; 41 42 /* 43 * Policies that need to keep per-blkcg data which is independent from any 44 * request_queue associated to it should implement cpd_alloc/free_fn() 45 * methods. A policy can allocate private data area by allocating larger 46 * data structure which embeds blkcg_policy_data at the beginning. 47 * cpd_init() is invoked to let each policy handle per-blkcg data. 48 */ 49 struct blkcg_policy_data { 50 /* the blkcg and policy id this per-policy data belongs to */ 51 struct blkcg *blkcg; 52 int plid; 53 }; 54 55 typedef struct blkcg_policy_data *(blkcg_pol_alloc_cpd_fn)(gfp_t gfp); 56 typedef void (blkcg_pol_init_cpd_fn)(struct blkcg_policy_data *cpd); 57 typedef void (blkcg_pol_free_cpd_fn)(struct blkcg_policy_data *cpd); 58 typedef void (blkcg_pol_bind_cpd_fn)(struct blkcg_policy_data *cpd); 59 typedef struct blkg_policy_data *(blkcg_pol_alloc_pd_fn)(gfp_t gfp, 60 struct request_queue *q, struct blkcg *blkcg); 61 typedef void (blkcg_pol_init_pd_fn)(struct blkg_policy_data *pd); 62 typedef void (blkcg_pol_online_pd_fn)(struct blkg_policy_data *pd); 63 typedef void (blkcg_pol_offline_pd_fn)(struct blkg_policy_data *pd); 64 typedef void (blkcg_pol_free_pd_fn)(struct blkg_policy_data *pd); 65 typedef void (blkcg_pol_reset_pd_stats_fn)(struct blkg_policy_data *pd); 66 typedef bool (blkcg_pol_stat_pd_fn)(struct blkg_policy_data *pd, 67 struct seq_file *s); 68 69 struct blkcg_policy { 70 int plid; 71 /* cgroup files for the policy */ 72 struct cftype *dfl_cftypes; 73 struct cftype *legacy_cftypes; 74 75 /* operations */ 76 blkcg_pol_alloc_cpd_fn *cpd_alloc_fn; 77 blkcg_pol_init_cpd_fn *cpd_init_fn; 78 blkcg_pol_free_cpd_fn *cpd_free_fn; 79 blkcg_pol_bind_cpd_fn *cpd_bind_fn; 80 81 blkcg_pol_alloc_pd_fn *pd_alloc_fn; 82 blkcg_pol_init_pd_fn *pd_init_fn; 83 blkcg_pol_online_pd_fn *pd_online_fn; 84 blkcg_pol_offline_pd_fn *pd_offline_fn; 85 blkcg_pol_free_pd_fn *pd_free_fn; 86 blkcg_pol_reset_pd_stats_fn *pd_reset_stats_fn; 87 blkcg_pol_stat_pd_fn *pd_stat_fn; 88 }; 89 90 extern struct blkcg blkcg_root; 91 extern bool blkcg_debug_stats; 92 93 struct blkcg_gq *blkg_lookup_slowpath(struct blkcg *blkcg, 94 struct request_queue *q, bool update_hint); 95 int blkcg_init_queue(struct request_queue *q); 96 void blkcg_exit_queue(struct request_queue *q); 97 98 /* Blkio controller policy registration */ 99 int blkcg_policy_register(struct blkcg_policy *pol); 100 void blkcg_policy_unregister(struct blkcg_policy *pol); 101 int blkcg_activate_policy(struct request_queue *q, 102 const struct blkcg_policy *pol); 103 void blkcg_deactivate_policy(struct request_queue *q, 104 const struct blkcg_policy *pol); 105 106 const char *blkg_dev_name(struct blkcg_gq *blkg); 107 void blkcg_print_blkgs(struct seq_file *sf, struct blkcg *blkcg, 108 u64 (*prfill)(struct seq_file *, 109 struct blkg_policy_data *, int), 110 const struct blkcg_policy *pol, int data, 111 bool show_total); 112 u64 __blkg_prfill_u64(struct seq_file *sf, struct blkg_policy_data *pd, u64 v); 113 114 struct blkg_conf_ctx { 115 struct block_device *bdev; 116 struct blkcg_gq *blkg; 117 char *body; 118 }; 119 120 struct block_device *blkcg_conf_open_bdev(char **inputp); 121 int blkg_conf_prep(struct blkcg *blkcg, const struct blkcg_policy *pol, 122 char *input, struct blkg_conf_ctx *ctx); 123 void blkg_conf_finish(struct blkg_conf_ctx *ctx); 124 125 /** 126 * blkcg_css - find the current css 127 * 128 * Find the css associated with either the kthread or the current task. 129 * This may return a dying css, so it is up to the caller to use tryget logic 130 * to confirm it is alive and well. 131 */ 132 static inline struct cgroup_subsys_state *blkcg_css(void) 133 { 134 struct cgroup_subsys_state *css; 135 136 css = kthread_blkcg(); 137 if (css) 138 return css; 139 return task_css(current, io_cgrp_id); 140 } 141 142 /** 143 * bio_issue_as_root_blkg - see if this bio needs to be issued as root blkg 144 * @return: true if this bio needs to be submitted with the root blkg context. 145 * 146 * In order to avoid priority inversions we sometimes need to issue a bio as if 147 * it were attached to the root blkg, and then backcharge to the actual owning 148 * blkg. The idea is we do bio_blkcg() to look up the actual context for the 149 * bio and attach the appropriate blkg to the bio. Then we call this helper and 150 * if it is true run with the root blkg for that queue and then do any 151 * backcharging to the originating cgroup once the io is complete. 152 */ 153 static inline bool bio_issue_as_root_blkg(struct bio *bio) 154 { 155 return (bio->bi_opf & (REQ_META | REQ_SWAP)) != 0; 156 } 157 158 /** 159 * __blkg_lookup - internal version of blkg_lookup() 160 * @blkcg: blkcg of interest 161 * @q: request_queue of interest 162 * @update_hint: whether to update lookup hint with the result or not 163 * 164 * This is internal version and shouldn't be used by policy 165 * implementations. Looks up blkgs for the @blkcg - @q pair regardless of 166 * @q's bypass state. If @update_hint is %true, the caller should be 167 * holding @q->queue_lock and lookup hint is updated on success. 168 */ 169 static inline struct blkcg_gq *__blkg_lookup(struct blkcg *blkcg, 170 struct request_queue *q, 171 bool update_hint) 172 { 173 struct blkcg_gq *blkg; 174 175 if (blkcg == &blkcg_root) 176 return q->root_blkg; 177 178 blkg = rcu_dereference(blkcg->blkg_hint); 179 if (blkg && blkg->q == q) 180 return blkg; 181 182 return blkg_lookup_slowpath(blkcg, q, update_hint); 183 } 184 185 /** 186 * blkg_lookup - lookup blkg for the specified blkcg - q pair 187 * @blkcg: blkcg of interest 188 * @q: request_queue of interest 189 * 190 * Lookup blkg for the @blkcg - @q pair. This function should be called 191 * under RCU read lock. 192 */ 193 static inline struct blkcg_gq *blkg_lookup(struct blkcg *blkcg, 194 struct request_queue *q) 195 { 196 WARN_ON_ONCE(!rcu_read_lock_held()); 197 return __blkg_lookup(blkcg, q, false); 198 } 199 200 /** 201 * blk_queue_root_blkg - return blkg for the (blkcg_root, @q) pair 202 * @q: request_queue of interest 203 * 204 * Lookup blkg for @q at the root level. See also blkg_lookup(). 205 */ 206 static inline struct blkcg_gq *blk_queue_root_blkg(struct request_queue *q) 207 { 208 return q->root_blkg; 209 } 210 211 /** 212 * blkg_to_pdata - get policy private data 213 * @blkg: blkg of interest 214 * @pol: policy of interest 215 * 216 * Return pointer to private data associated with the @blkg-@pol pair. 217 */ 218 static inline struct blkg_policy_data *blkg_to_pd(struct blkcg_gq *blkg, 219 struct blkcg_policy *pol) 220 { 221 return blkg ? blkg->pd[pol->plid] : NULL; 222 } 223 224 static inline struct blkcg_policy_data *blkcg_to_cpd(struct blkcg *blkcg, 225 struct blkcg_policy *pol) 226 { 227 return blkcg ? blkcg->cpd[pol->plid] : NULL; 228 } 229 230 /** 231 * pdata_to_blkg - get blkg associated with policy private data 232 * @pd: policy private data of interest 233 * 234 * @pd is policy private data. Determine the blkg it's associated with. 235 */ 236 static inline struct blkcg_gq *pd_to_blkg(struct blkg_policy_data *pd) 237 { 238 return pd ? pd->blkg : NULL; 239 } 240 241 static inline struct blkcg *cpd_to_blkcg(struct blkcg_policy_data *cpd) 242 { 243 return cpd ? cpd->blkcg : NULL; 244 } 245 246 /** 247 * blkg_path - format cgroup path of blkg 248 * @blkg: blkg of interest 249 * @buf: target buffer 250 * @buflen: target buffer length 251 * 252 * Format the path of the cgroup of @blkg into @buf. 253 */ 254 static inline int blkg_path(struct blkcg_gq *blkg, char *buf, int buflen) 255 { 256 return cgroup_path(blkg->blkcg->css.cgroup, buf, buflen); 257 } 258 259 /** 260 * blkg_get - get a blkg reference 261 * @blkg: blkg to get 262 * 263 * The caller should be holding an existing reference. 264 */ 265 static inline void blkg_get(struct blkcg_gq *blkg) 266 { 267 percpu_ref_get(&blkg->refcnt); 268 } 269 270 /** 271 * blkg_tryget - try and get a blkg reference 272 * @blkg: blkg to get 273 * 274 * This is for use when doing an RCU lookup of the blkg. We may be in the midst 275 * of freeing this blkg, so we can only use it if the refcnt is not zero. 276 */ 277 static inline bool blkg_tryget(struct blkcg_gq *blkg) 278 { 279 return blkg && percpu_ref_tryget(&blkg->refcnt); 280 } 281 282 /** 283 * blkg_put - put a blkg reference 284 * @blkg: blkg to put 285 */ 286 static inline void blkg_put(struct blkcg_gq *blkg) 287 { 288 percpu_ref_put(&blkg->refcnt); 289 } 290 291 /** 292 * blkg_for_each_descendant_pre - pre-order walk of a blkg's descendants 293 * @d_blkg: loop cursor pointing to the current descendant 294 * @pos_css: used for iteration 295 * @p_blkg: target blkg to walk descendants of 296 * 297 * Walk @c_blkg through the descendants of @p_blkg. Must be used with RCU 298 * read locked. If called under either blkcg or queue lock, the iteration 299 * is guaranteed to include all and only online blkgs. The caller may 300 * update @pos_css by calling css_rightmost_descendant() to skip subtree. 301 * @p_blkg is included in the iteration and the first node to be visited. 302 */ 303 #define blkg_for_each_descendant_pre(d_blkg, pos_css, p_blkg) \ 304 css_for_each_descendant_pre((pos_css), &(p_blkg)->blkcg->css) \ 305 if (((d_blkg) = __blkg_lookup(css_to_blkcg(pos_css), \ 306 (p_blkg)->q, false))) 307 308 /** 309 * blkg_for_each_descendant_post - post-order walk of a blkg's descendants 310 * @d_blkg: loop cursor pointing to the current descendant 311 * @pos_css: used for iteration 312 * @p_blkg: target blkg to walk descendants of 313 * 314 * Similar to blkg_for_each_descendant_pre() but performs post-order 315 * traversal instead. Synchronization rules are the same. @p_blkg is 316 * included in the iteration and the last node to be visited. 317 */ 318 #define blkg_for_each_descendant_post(d_blkg, pos_css, p_blkg) \ 319 css_for_each_descendant_post((pos_css), &(p_blkg)->blkcg->css) \ 320 if (((d_blkg) = __blkg_lookup(css_to_blkcg(pos_css), \ 321 (p_blkg)->q, false))) 322 323 bool __blkcg_punt_bio_submit(struct bio *bio); 324 325 static inline bool blkcg_punt_bio_submit(struct bio *bio) 326 { 327 if (bio->bi_opf & REQ_CGROUP_PUNT) 328 return __blkcg_punt_bio_submit(bio); 329 else 330 return false; 331 } 332 333 static inline void blkcg_bio_issue_init(struct bio *bio) 334 { 335 bio_issue_init(&bio->bi_issue, bio_sectors(bio)); 336 } 337 338 static inline void blkcg_use_delay(struct blkcg_gq *blkg) 339 { 340 if (WARN_ON_ONCE(atomic_read(&blkg->use_delay) < 0)) 341 return; 342 if (atomic_add_return(1, &blkg->use_delay) == 1) 343 atomic_inc(&blkg->blkcg->css.cgroup->congestion_count); 344 } 345 346 static inline int blkcg_unuse_delay(struct blkcg_gq *blkg) 347 { 348 int old = atomic_read(&blkg->use_delay); 349 350 if (WARN_ON_ONCE(old < 0)) 351 return 0; 352 if (old == 0) 353 return 0; 354 355 /* 356 * We do this song and dance because we can race with somebody else 357 * adding or removing delay. If we just did an atomic_dec we'd end up 358 * negative and we'd already be in trouble. We need to subtract 1 and 359 * then check to see if we were the last delay so we can drop the 360 * congestion count on the cgroup. 361 */ 362 while (old) { 363 int cur = atomic_cmpxchg(&blkg->use_delay, old, old - 1); 364 if (cur == old) 365 break; 366 old = cur; 367 } 368 369 if (old == 0) 370 return 0; 371 if (old == 1) 372 atomic_dec(&blkg->blkcg->css.cgroup->congestion_count); 373 return 1; 374 } 375 376 /** 377 * blkcg_set_delay - Enable allocator delay mechanism with the specified delay amount 378 * @blkg: target blkg 379 * @delay: delay duration in nsecs 380 * 381 * When enabled with this function, the delay is not decayed and must be 382 * explicitly cleared with blkcg_clear_delay(). Must not be mixed with 383 * blkcg_[un]use_delay() and blkcg_add_delay() usages. 384 */ 385 static inline void blkcg_set_delay(struct blkcg_gq *blkg, u64 delay) 386 { 387 int old = atomic_read(&blkg->use_delay); 388 389 /* We only want 1 person setting the congestion count for this blkg. */ 390 if (!old && atomic_cmpxchg(&blkg->use_delay, old, -1) == old) 391 atomic_inc(&blkg->blkcg->css.cgroup->congestion_count); 392 393 atomic64_set(&blkg->delay_nsec, delay); 394 } 395 396 /** 397 * blkcg_clear_delay - Disable allocator delay mechanism 398 * @blkg: target blkg 399 * 400 * Disable use_delay mechanism. See blkcg_set_delay(). 401 */ 402 static inline void blkcg_clear_delay(struct blkcg_gq *blkg) 403 { 404 int old = atomic_read(&blkg->use_delay); 405 406 /* We only want 1 person clearing the congestion count for this blkg. */ 407 if (old && atomic_cmpxchg(&blkg->use_delay, old, 0) == old) 408 atomic_dec(&blkg->blkcg->css.cgroup->congestion_count); 409 } 410 411 /** 412 * blk_cgroup_mergeable - Determine whether to allow or disallow merges 413 * @rq: request to merge into 414 * @bio: bio to merge 415 * 416 * @bio and @rq should belong to the same cgroup and their issue_as_root should 417 * match. The latter is necessary as we don't want to throttle e.g. a metadata 418 * update because it happens to be next to a regular IO. 419 */ 420 static inline bool blk_cgroup_mergeable(struct request *rq, struct bio *bio) 421 { 422 return rq->bio->bi_blkg == bio->bi_blkg && 423 bio_issue_as_root_blkg(rq->bio) == bio_issue_as_root_blkg(bio); 424 } 425 426 void blk_cgroup_bio_start(struct bio *bio); 427 void blkcg_add_delay(struct blkcg_gq *blkg, u64 now, u64 delta); 428 #else /* CONFIG_BLK_CGROUP */ 429 430 struct blkg_policy_data { 431 }; 432 433 struct blkcg_policy_data { 434 }; 435 436 struct blkcg_policy { 437 }; 438 439 #ifdef CONFIG_BLOCK 440 441 static inline struct blkcg_gq *blkg_lookup(struct blkcg *blkcg, void *key) { return NULL; } 442 static inline struct blkcg_gq *blk_queue_root_blkg(struct request_queue *q) 443 { return NULL; } 444 static inline int blkcg_init_queue(struct request_queue *q) { return 0; } 445 static inline void blkcg_exit_queue(struct request_queue *q) { } 446 static inline int blkcg_policy_register(struct blkcg_policy *pol) { return 0; } 447 static inline void blkcg_policy_unregister(struct blkcg_policy *pol) { } 448 static inline int blkcg_activate_policy(struct request_queue *q, 449 const struct blkcg_policy *pol) { return 0; } 450 static inline void blkcg_deactivate_policy(struct request_queue *q, 451 const struct blkcg_policy *pol) { } 452 453 static inline struct blkg_policy_data *blkg_to_pd(struct blkcg_gq *blkg, 454 struct blkcg_policy *pol) { return NULL; } 455 static inline struct blkcg_gq *pd_to_blkg(struct blkg_policy_data *pd) { return NULL; } 456 static inline char *blkg_path(struct blkcg_gq *blkg) { return NULL; } 457 static inline void blkg_get(struct blkcg_gq *blkg) { } 458 static inline void blkg_put(struct blkcg_gq *blkg) { } 459 460 static inline bool blkcg_punt_bio_submit(struct bio *bio) { return false; } 461 static inline void blkcg_bio_issue_init(struct bio *bio) { } 462 static inline void blk_cgroup_bio_start(struct bio *bio) { } 463 static inline bool blk_cgroup_mergeable(struct request *rq, struct bio *bio) { return true; } 464 465 #define blk_queue_for_each_rl(rl, q) \ 466 for ((rl) = &(q)->root_rl; (rl); (rl) = NULL) 467 468 #endif /* CONFIG_BLOCK */ 469 #endif /* CONFIG_BLK_CGROUP */ 470 471 #endif /* _BLK_CGROUP_PRIVATE_H */ 472