1#ifndef _BLK_CGROUP_H 2#define _BLK_CGROUP_H 3/* 4 * Common Block IO controller cgroup interface 5 * 6 * Based on ideas and code from CFQ, CFS and BFQ: 7 * Copyright (C) 2003 Jens Axboe <axboe@kernel.dk> 8 * 9 * Copyright (C) 2008 Fabio Checconi <fabio@gandalf.sssup.it> 10 * Paolo Valente <paolo.valente@unimore.it> 11 * 12 * Copyright (C) 2009 Vivek Goyal <vgoyal@redhat.com> 13 * Nauman Rafique <nauman@google.com> 14 */ 15 16#include <linux/cgroup.h> 17#include <linux/u64_stats_sync.h> 18#include <linux/seq_file.h> 19#include <linux/radix-tree.h> 20#include <linux/blkdev.h> 21#include <linux/atomic.h> 22 23/* Max limits for throttle policy */ 24#define THROTL_IOPS_MAX UINT_MAX 25 26/* CFQ specific, out here for blkcg->cfq_weight */ 27#define CFQ_WEIGHT_MIN 10 28#define CFQ_WEIGHT_MAX 1000 29#define CFQ_WEIGHT_DEFAULT 500 30 31#ifdef CONFIG_BLK_CGROUP 32 33enum blkg_rwstat_type { 34 BLKG_RWSTAT_READ, 35 BLKG_RWSTAT_WRITE, 36 BLKG_RWSTAT_SYNC, 37 BLKG_RWSTAT_ASYNC, 38 39 BLKG_RWSTAT_NR, 40 BLKG_RWSTAT_TOTAL = BLKG_RWSTAT_NR, 41}; 42 43struct blkcg_gq; 44 45struct blkcg { 46 struct cgroup_subsys_state css; 47 spinlock_t lock; 48 49 struct radix_tree_root blkg_tree; 50 struct blkcg_gq *blkg_hint; 51 struct hlist_head blkg_list; 52 53 /* TODO: per-policy storage in blkcg */ 54 unsigned int cfq_weight; /* belongs to cfq */ 55 unsigned int cfq_leaf_weight; 56}; 57 58struct blkg_stat { 59 struct u64_stats_sync syncp; 60 uint64_t cnt; 61}; 62 63struct blkg_rwstat { 64 struct u64_stats_sync syncp; 65 uint64_t cnt[BLKG_RWSTAT_NR]; 66}; 67 68/* 69 * A blkcg_gq (blkg) is association between a block cgroup (blkcg) and a 70 * request_queue (q). This is used by blkcg policies which need to track 71 * information per blkcg - q pair. 72 * 73 * There can be multiple active blkcg policies and each has its private 74 * data on each blkg, the size of which is determined by 75 * blkcg_policy->pd_size. blkcg core allocates and frees such areas 76 * together with blkg and invokes pd_init/exit_fn() methods. 77 * 78 * Such private data must embed struct blkg_policy_data (pd) at the 79 * beginning and pd_size can't be smaller than pd. 80 */ 81struct blkg_policy_data { 82 /* the blkg and policy id this per-policy data belongs to */ 83 struct blkcg_gq *blkg; 84 int plid; 85 86 /* used during policy activation */ 87 struct list_head alloc_node; 88}; 89 90/* association between a blk cgroup and a request queue */ 91struct blkcg_gq { 92 /* Pointer to the associated request_queue */ 93 struct request_queue *q; 94 struct list_head q_node; 95 struct hlist_node blkcg_node; 96 struct blkcg *blkcg; 97 98 /* all non-root blkcg_gq's are guaranteed to have access to parent */ 99 struct blkcg_gq *parent; 100 101 /* request allocation list for this blkcg-q pair */ 102 struct request_list rl; 103 104 /* reference count */ 105 atomic_t refcnt; 106 107 /* is this blkg online? protected by both blkcg and q locks */ 108 bool online; 109 110 struct blkg_policy_data *pd[BLKCG_MAX_POLS]; 111 112 struct rcu_head rcu_head; 113}; 114 115typedef void (blkcg_pol_init_pd_fn)(struct blkcg_gq *blkg); 116typedef void (blkcg_pol_online_pd_fn)(struct blkcg_gq *blkg); 117typedef void (blkcg_pol_offline_pd_fn)(struct blkcg_gq *blkg); 118typedef void (blkcg_pol_exit_pd_fn)(struct blkcg_gq *blkg); 119typedef void (blkcg_pol_reset_pd_stats_fn)(struct blkcg_gq *blkg); 120 121struct blkcg_policy { 122 int plid; 123 /* policy specific private data size */ 124 size_t pd_size; 125 /* cgroup files for the policy */ 126 struct cftype *cftypes; 127 128 /* operations */ 129 blkcg_pol_init_pd_fn *pd_init_fn; 130 blkcg_pol_online_pd_fn *pd_online_fn; 131 blkcg_pol_offline_pd_fn *pd_offline_fn; 132 blkcg_pol_exit_pd_fn *pd_exit_fn; 133 blkcg_pol_reset_pd_stats_fn *pd_reset_stats_fn; 134}; 135 136extern struct blkcg blkcg_root; 137 138struct blkcg_gq *blkg_lookup(struct blkcg *blkcg, struct request_queue *q); 139struct blkcg_gq *blkg_lookup_create(struct blkcg *blkcg, 140 struct request_queue *q); 141int blkcg_init_queue(struct request_queue *q); 142void blkcg_drain_queue(struct request_queue *q); 143void blkcg_exit_queue(struct request_queue *q); 144 145/* Blkio controller policy registration */ 146int blkcg_policy_register(struct blkcg_policy *pol); 147void blkcg_policy_unregister(struct blkcg_policy *pol); 148int blkcg_activate_policy(struct request_queue *q, 149 const struct blkcg_policy *pol); 150void blkcg_deactivate_policy(struct request_queue *q, 151 const struct blkcg_policy *pol); 152 153void blkcg_print_blkgs(struct seq_file *sf, struct blkcg *blkcg, 154 u64 (*prfill)(struct seq_file *, 155 struct blkg_policy_data *, int), 156 const struct blkcg_policy *pol, int data, 157 bool show_total); 158u64 __blkg_prfill_u64(struct seq_file *sf, struct blkg_policy_data *pd, u64 v); 159u64 __blkg_prfill_rwstat(struct seq_file *sf, struct blkg_policy_data *pd, 160 const struct blkg_rwstat *rwstat); 161u64 blkg_prfill_stat(struct seq_file *sf, struct blkg_policy_data *pd, int off); 162u64 blkg_prfill_rwstat(struct seq_file *sf, struct blkg_policy_data *pd, 163 int off); 164 165u64 blkg_stat_recursive_sum(struct blkg_policy_data *pd, int off); 166struct blkg_rwstat blkg_rwstat_recursive_sum(struct blkg_policy_data *pd, 167 int off); 168 169struct blkg_conf_ctx { 170 struct gendisk *disk; 171 struct blkcg_gq *blkg; 172 u64 v; 173}; 174 175int blkg_conf_prep(struct blkcg *blkcg, const struct blkcg_policy *pol, 176 const char *input, struct blkg_conf_ctx *ctx); 177void blkg_conf_finish(struct blkg_conf_ctx *ctx); 178 179 180static inline struct blkcg *css_to_blkcg(struct cgroup_subsys_state *css) 181{ 182 return css ? container_of(css, struct blkcg, css) : NULL; 183} 184 185static inline struct blkcg *task_blkcg(struct task_struct *tsk) 186{ 187 return css_to_blkcg(task_css(tsk, blkio_cgrp_id)); 188} 189 190static inline struct blkcg *bio_blkcg(struct bio *bio) 191{ 192 if (bio && bio->bi_css) 193 return css_to_blkcg(bio->bi_css); 194 return task_blkcg(current); 195} 196 197/** 198 * blkcg_parent - get the parent of a blkcg 199 * @blkcg: blkcg of interest 200 * 201 * Return the parent blkcg of @blkcg. Can be called anytime. 202 */ 203static inline struct blkcg *blkcg_parent(struct blkcg *blkcg) 204{ 205 return css_to_blkcg(blkcg->css.parent); 206} 207 208/** 209 * blkg_to_pdata - get policy private data 210 * @blkg: blkg of interest 211 * @pol: policy of interest 212 * 213 * Return pointer to private data associated with the @blkg-@pol pair. 214 */ 215static inline struct blkg_policy_data *blkg_to_pd(struct blkcg_gq *blkg, 216 struct blkcg_policy *pol) 217{ 218 return blkg ? blkg->pd[pol->plid] : NULL; 219} 220 221/** 222 * pdata_to_blkg - get blkg associated with policy private data 223 * @pd: policy private data of interest 224 * 225 * @pd is policy private data. Determine the blkg it's associated with. 226 */ 227static inline struct blkcg_gq *pd_to_blkg(struct blkg_policy_data *pd) 228{ 229 return pd ? pd->blkg : NULL; 230} 231 232/** 233 * blkg_path - format cgroup path of blkg 234 * @blkg: blkg of interest 235 * @buf: target buffer 236 * @buflen: target buffer length 237 * 238 * Format the path of the cgroup of @blkg into @buf. 239 */ 240static inline int blkg_path(struct blkcg_gq *blkg, char *buf, int buflen) 241{ 242 char *p; 243 244 p = cgroup_path(blkg->blkcg->css.cgroup, buf, buflen); 245 if (!p) { 246 strncpy(buf, "<unavailable>", buflen); 247 return -ENAMETOOLONG; 248 } 249 250 memmove(buf, p, buf + buflen - p); 251 return 0; 252} 253 254/** 255 * blkg_get - get a blkg reference 256 * @blkg: blkg to get 257 * 258 * The caller should be holding an existing reference. 259 */ 260static inline void blkg_get(struct blkcg_gq *blkg) 261{ 262 WARN_ON_ONCE(atomic_read(&blkg->refcnt) <= 0); 263 atomic_inc(&blkg->refcnt); 264} 265 266void __blkg_release_rcu(struct rcu_head *rcu); 267 268/** 269 * blkg_put - put a blkg reference 270 * @blkg: blkg to put 271 */ 272static inline void blkg_put(struct blkcg_gq *blkg) 273{ 274 WARN_ON_ONCE(atomic_read(&blkg->refcnt) <= 0); 275 if (atomic_dec_and_test(&blkg->refcnt)) 276 call_rcu(&blkg->rcu_head, __blkg_release_rcu); 277} 278 279struct blkcg_gq *__blkg_lookup(struct blkcg *blkcg, struct request_queue *q, 280 bool update_hint); 281 282/** 283 * blkg_for_each_descendant_pre - pre-order walk of a blkg's descendants 284 * @d_blkg: loop cursor pointing to the current descendant 285 * @pos_css: used for iteration 286 * @p_blkg: target blkg to walk descendants of 287 * 288 * Walk @c_blkg through the descendants of @p_blkg. Must be used with RCU 289 * read locked. If called under either blkcg or queue lock, the iteration 290 * is guaranteed to include all and only online blkgs. The caller may 291 * update @pos_css by calling css_rightmost_descendant() to skip subtree. 292 * @p_blkg is included in the iteration and the first node to be visited. 293 */ 294#define blkg_for_each_descendant_pre(d_blkg, pos_css, p_blkg) \ 295 css_for_each_descendant_pre((pos_css), &(p_blkg)->blkcg->css) \ 296 if (((d_blkg) = __blkg_lookup(css_to_blkcg(pos_css), \ 297 (p_blkg)->q, false))) 298 299/** 300 * blkg_for_each_descendant_post - post-order walk of a blkg's descendants 301 * @d_blkg: loop cursor pointing to the current descendant 302 * @pos_css: used for iteration 303 * @p_blkg: target blkg to walk descendants of 304 * 305 * Similar to blkg_for_each_descendant_pre() but performs post-order 306 * traversal instead. Synchronization rules are the same. @p_blkg is 307 * included in the iteration and the last node to be visited. 308 */ 309#define blkg_for_each_descendant_post(d_blkg, pos_css, p_blkg) \ 310 css_for_each_descendant_post((pos_css), &(p_blkg)->blkcg->css) \ 311 if (((d_blkg) = __blkg_lookup(css_to_blkcg(pos_css), \ 312 (p_blkg)->q, false))) 313 314/** 315 * blk_get_rl - get request_list to use 316 * @q: request_queue of interest 317 * @bio: bio which will be attached to the allocated request (may be %NULL) 318 * 319 * The caller wants to allocate a request from @q to use for @bio. Find 320 * the request_list to use and obtain a reference on it. Should be called 321 * under queue_lock. This function is guaranteed to return non-%NULL 322 * request_list. 323 */ 324static inline struct request_list *blk_get_rl(struct request_queue *q, 325 struct bio *bio) 326{ 327 struct blkcg *blkcg; 328 struct blkcg_gq *blkg; 329 330 rcu_read_lock(); 331 332 blkcg = bio_blkcg(bio); 333 334 /* bypass blkg lookup and use @q->root_rl directly for root */ 335 if (blkcg == &blkcg_root) 336 goto root_rl; 337 338 /* 339 * Try to use blkg->rl. blkg lookup may fail under memory pressure 340 * or if either the blkcg or queue is going away. Fall back to 341 * root_rl in such cases. 342 */ 343 blkg = blkg_lookup_create(blkcg, q); 344 if (unlikely(IS_ERR(blkg))) 345 goto root_rl; 346 347 blkg_get(blkg); 348 rcu_read_unlock(); 349 return &blkg->rl; 350root_rl: 351 rcu_read_unlock(); 352 return &q->root_rl; 353} 354 355/** 356 * blk_put_rl - put request_list 357 * @rl: request_list to put 358 * 359 * Put the reference acquired by blk_get_rl(). Should be called under 360 * queue_lock. 361 */ 362static inline void blk_put_rl(struct request_list *rl) 363{ 364 /* root_rl may not have blkg set */ 365 if (rl->blkg && rl->blkg->blkcg != &blkcg_root) 366 blkg_put(rl->blkg); 367} 368 369/** 370 * blk_rq_set_rl - associate a request with a request_list 371 * @rq: request of interest 372 * @rl: target request_list 373 * 374 * Associate @rq with @rl so that accounting and freeing can know the 375 * request_list @rq came from. 376 */ 377static inline void blk_rq_set_rl(struct request *rq, struct request_list *rl) 378{ 379 rq->rl = rl; 380} 381 382/** 383 * blk_rq_rl - return the request_list a request came from 384 * @rq: request of interest 385 * 386 * Return the request_list @rq is allocated from. 387 */ 388static inline struct request_list *blk_rq_rl(struct request *rq) 389{ 390 return rq->rl; 391} 392 393struct request_list *__blk_queue_next_rl(struct request_list *rl, 394 struct request_queue *q); 395/** 396 * blk_queue_for_each_rl - iterate through all request_lists of a request_queue 397 * 398 * Should be used under queue_lock. 399 */ 400#define blk_queue_for_each_rl(rl, q) \ 401 for ((rl) = &(q)->root_rl; (rl); (rl) = __blk_queue_next_rl((rl), (q))) 402 403static inline void blkg_stat_init(struct blkg_stat *stat) 404{ 405 u64_stats_init(&stat->syncp); 406} 407 408/** 409 * blkg_stat_add - add a value to a blkg_stat 410 * @stat: target blkg_stat 411 * @val: value to add 412 * 413 * Add @val to @stat. The caller is responsible for synchronizing calls to 414 * this function. 415 */ 416static inline void blkg_stat_add(struct blkg_stat *stat, uint64_t val) 417{ 418 u64_stats_update_begin(&stat->syncp); 419 stat->cnt += val; 420 u64_stats_update_end(&stat->syncp); 421} 422 423/** 424 * blkg_stat_read - read the current value of a blkg_stat 425 * @stat: blkg_stat to read 426 * 427 * Read the current value of @stat. This function can be called without 428 * synchroniztion and takes care of u64 atomicity. 429 */ 430static inline uint64_t blkg_stat_read(struct blkg_stat *stat) 431{ 432 unsigned int start; 433 uint64_t v; 434 435 do { 436 start = u64_stats_fetch_begin_irq(&stat->syncp); 437 v = stat->cnt; 438 } while (u64_stats_fetch_retry_irq(&stat->syncp, start)); 439 440 return v; 441} 442 443/** 444 * blkg_stat_reset - reset a blkg_stat 445 * @stat: blkg_stat to reset 446 */ 447static inline void blkg_stat_reset(struct blkg_stat *stat) 448{ 449 stat->cnt = 0; 450} 451 452/** 453 * blkg_stat_merge - merge a blkg_stat into another 454 * @to: the destination blkg_stat 455 * @from: the source 456 * 457 * Add @from's count to @to. 458 */ 459static inline void blkg_stat_merge(struct blkg_stat *to, struct blkg_stat *from) 460{ 461 blkg_stat_add(to, blkg_stat_read(from)); 462} 463 464static inline void blkg_rwstat_init(struct blkg_rwstat *rwstat) 465{ 466 u64_stats_init(&rwstat->syncp); 467} 468 469/** 470 * blkg_rwstat_add - add a value to a blkg_rwstat 471 * @rwstat: target blkg_rwstat 472 * @rw: mask of REQ_{WRITE|SYNC} 473 * @val: value to add 474 * 475 * Add @val to @rwstat. The counters are chosen according to @rw. The 476 * caller is responsible for synchronizing calls to this function. 477 */ 478static inline void blkg_rwstat_add(struct blkg_rwstat *rwstat, 479 int rw, uint64_t val) 480{ 481 u64_stats_update_begin(&rwstat->syncp); 482 483 if (rw & REQ_WRITE) 484 rwstat->cnt[BLKG_RWSTAT_WRITE] += val; 485 else 486 rwstat->cnt[BLKG_RWSTAT_READ] += val; 487 if (rw & REQ_SYNC) 488 rwstat->cnt[BLKG_RWSTAT_SYNC] += val; 489 else 490 rwstat->cnt[BLKG_RWSTAT_ASYNC] += val; 491 492 u64_stats_update_end(&rwstat->syncp); 493} 494 495/** 496 * blkg_rwstat_read - read the current values of a blkg_rwstat 497 * @rwstat: blkg_rwstat to read 498 * 499 * Read the current snapshot of @rwstat and return it as the return value. 500 * This function can be called without synchronization and takes care of 501 * u64 atomicity. 502 */ 503static inline struct blkg_rwstat blkg_rwstat_read(struct blkg_rwstat *rwstat) 504{ 505 unsigned int start; 506 struct blkg_rwstat tmp; 507 508 do { 509 start = u64_stats_fetch_begin_irq(&rwstat->syncp); 510 tmp = *rwstat; 511 } while (u64_stats_fetch_retry_irq(&rwstat->syncp, start)); 512 513 return tmp; 514} 515 516/** 517 * blkg_rwstat_total - read the total count of a blkg_rwstat 518 * @rwstat: blkg_rwstat to read 519 * 520 * Return the total count of @rwstat regardless of the IO direction. This 521 * function can be called without synchronization and takes care of u64 522 * atomicity. 523 */ 524static inline uint64_t blkg_rwstat_total(struct blkg_rwstat *rwstat) 525{ 526 struct blkg_rwstat tmp = blkg_rwstat_read(rwstat); 527 528 return tmp.cnt[BLKG_RWSTAT_READ] + tmp.cnt[BLKG_RWSTAT_WRITE]; 529} 530 531/** 532 * blkg_rwstat_reset - reset a blkg_rwstat 533 * @rwstat: blkg_rwstat to reset 534 */ 535static inline void blkg_rwstat_reset(struct blkg_rwstat *rwstat) 536{ 537 memset(rwstat->cnt, 0, sizeof(rwstat->cnt)); 538} 539 540/** 541 * blkg_rwstat_merge - merge a blkg_rwstat into another 542 * @to: the destination blkg_rwstat 543 * @from: the source 544 * 545 * Add @from's counts to @to. 546 */ 547static inline void blkg_rwstat_merge(struct blkg_rwstat *to, 548 struct blkg_rwstat *from) 549{ 550 struct blkg_rwstat v = blkg_rwstat_read(from); 551 int i; 552 553 u64_stats_update_begin(&to->syncp); 554 for (i = 0; i < BLKG_RWSTAT_NR; i++) 555 to->cnt[i] += v.cnt[i]; 556 u64_stats_update_end(&to->syncp); 557} 558 559#else /* CONFIG_BLK_CGROUP */ 560 561struct cgroup; 562struct blkcg; 563 564struct blkg_policy_data { 565}; 566 567struct blkcg_gq { 568}; 569 570struct blkcg_policy { 571}; 572 573static inline struct blkcg_gq *blkg_lookup(struct blkcg *blkcg, void *key) { return NULL; } 574static inline int blkcg_init_queue(struct request_queue *q) { return 0; } 575static inline void blkcg_drain_queue(struct request_queue *q) { } 576static inline void blkcg_exit_queue(struct request_queue *q) { } 577static inline int blkcg_policy_register(struct blkcg_policy *pol) { return 0; } 578static inline void blkcg_policy_unregister(struct blkcg_policy *pol) { } 579static inline int blkcg_activate_policy(struct request_queue *q, 580 const struct blkcg_policy *pol) { return 0; } 581static inline void blkcg_deactivate_policy(struct request_queue *q, 582 const struct blkcg_policy *pol) { } 583 584static inline struct blkcg *bio_blkcg(struct bio *bio) { return NULL; } 585 586static inline struct blkg_policy_data *blkg_to_pd(struct blkcg_gq *blkg, 587 struct blkcg_policy *pol) { return NULL; } 588static inline struct blkcg_gq *pd_to_blkg(struct blkg_policy_data *pd) { return NULL; } 589static inline char *blkg_path(struct blkcg_gq *blkg) { return NULL; } 590static inline void blkg_get(struct blkcg_gq *blkg) { } 591static inline void blkg_put(struct blkcg_gq *blkg) { } 592 593static inline struct request_list *blk_get_rl(struct request_queue *q, 594 struct bio *bio) { return &q->root_rl; } 595static inline void blk_put_rl(struct request_list *rl) { } 596static inline void blk_rq_set_rl(struct request *rq, struct request_list *rl) { } 597static inline struct request_list *blk_rq_rl(struct request *rq) { return &rq->q->root_rl; } 598 599#define blk_queue_for_each_rl(rl, q) \ 600 for ((rl) = &(q)->root_rl; (rl); (rl) = NULL) 601 602#endif /* CONFIG_BLK_CGROUP */ 603#endif /* _BLK_CGROUP_H */ 604