block/blk-cgroup.h

   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
  22 /* Max limits for throttle policy */
  23 #define THROTL_IOPS_MAX         UINT_MAX
  24
  25 /* CFQ specific, out here for blkcg->cfq_weight */
  26 #define CFQ_WEIGHT_MIN          10
  27 #define CFQ_WEIGHT_MAX          1000
  28 #define CFQ_WEIGHT_DEFAULT      500
  29
  30 #ifdef CONFIG_BLK_CGROUP
  31
  32 enum blkg_rwstat_type {
  33         BLKG_RWSTAT_READ,
  34         BLKG_RWSTAT_WRITE,
  35         BLKG_RWSTAT_SYNC,
  36         BLKG_RWSTAT_ASYNC,
  37
  38         BLKG_RWSTAT_NR,
  39         BLKG_RWSTAT_TOTAL = BLKG_RWSTAT_NR,
  40 };
  41
  42 struct blkcg_gq;
  43
  44 struct blkcg {
  45         struct cgroup_subsys_state      css;
  46         spinlock_t                      lock;
  47
  48         struct radix_tree_root          blkg_tree;
  49         struct blkcg_gq                 *blkg_hint;
  50         struct hlist_head               blkg_list;
  51
  52         /* for policies to test whether associated blkcg has changed */
  53         uint64_t                        id;
  54
  55         /* TODO: per-policy storage in blkcg */
  56         unsigned int                    cfq_weight;     /* belongs to cfq */
  57         unsigned int                    cfq_leaf_weight;
  58 };
  59
  60 struct blkg_stat {
  61         struct u64_stats_sync           syncp;
  62         uint64_t                        cnt;
  63 };
  64
  65 struct blkg_rwstat {
  66         struct u64_stats_sync           syncp;
  67         uint64_t                        cnt[BLKG_RWSTAT_NR];
  68 };
  69
  70 /*
  71  * A blkcg_gq (blkg) is association between a block cgroup (blkcg) and a
  72  * request_queue (q).  This is used by blkcg policies which need to track
  73  * information per blkcg - q pair.
  74  *
  75  * There can be multiple active blkcg policies and each has its private
  76  * data on each blkg, the size of which is determined by
  77  * blkcg_policy->pd_size.  blkcg core allocates and frees such areas
  78  * together with blkg and invokes pd_init/exit_fn() methods.
  79  *
  80  * Such private data must embed struct blkg_policy_data (pd) at the
  81  * beginning and pd_size can't be smaller than pd.
  82  */
  83 struct blkg_policy_data {
  84         /* the blkg and policy id this per-policy data belongs to */
  85         struct blkcg_gq                 *blkg;
  86         int                             plid;
  87
  88         /* used during policy activation */
  89         struct list_head                alloc_node;
  90 };
  91
  92 /* association between a blk cgroup and a request queue */
  93 struct blkcg_gq {
  94         /* Pointer to the associated request_queue */
  95         struct request_queue            *q;
  96         struct list_head                q_node;
  97         struct hlist_node               blkcg_node;
  98         struct blkcg                    *blkcg;
  99
 100         /* all non-root blkcg_gq's are guaranteed to have access to parent */
 101         struct blkcg_gq                 *parent;
 102
 103         /* request allocation list for this blkcg-q pair */
 104         struct request_list             rl;
 105
 106         /* reference count */
 107         int                             refcnt;
 108
 109         /* is this blkg online? protected by both blkcg and q locks */
 110         bool                            online;
 111
 112         struct blkg_policy_data         *pd[BLKCG_MAX_POLS];
 113
 114         struct rcu_head                 rcu_head;
 115 };
 116
 117 typedef void (blkcg_pol_init_pd_fn)(struct blkcg_gq *blkg);
 118 typedef void (blkcg_pol_online_pd_fn)(struct blkcg_gq *blkg);
 119 typedef void (blkcg_pol_offline_pd_fn)(struct blkcg_gq *blkg);
 120 typedef void (blkcg_pol_exit_pd_fn)(struct blkcg_gq *blkg);
 121 typedef void (blkcg_pol_reset_pd_stats_fn)(struct blkcg_gq *blkg);
 122
 123 struct blkcg_policy {
 124         int                             plid;
 125         /* policy specific private data size */
 126         size_t                          pd_size;
 127         /* cgroup files for the policy */
 128         struct cftype                   *cftypes;
 129
 130         /* operations */
 131         blkcg_pol_init_pd_fn            *pd_init_fn;
 132         blkcg_pol_online_pd_fn          *pd_online_fn;
 133         blkcg_pol_offline_pd_fn         *pd_offline_fn;
 134         blkcg_pol_exit_pd_fn            *pd_exit_fn;
 135         blkcg_pol_reset_pd_stats_fn     *pd_reset_stats_fn;
 136 };
 137
 138 extern struct blkcg blkcg_root;
 139
 140 struct blkcg_gq *blkg_lookup(struct blkcg *blkcg, struct request_queue *q);
 141 struct blkcg_gq *blkg_lookup_create(struct blkcg *blkcg,
 142                                     struct request_queue *q);
 143 int blkcg_init_queue(struct request_queue *q);
 144 void blkcg_drain_queue(struct request_queue *q);
 145 void blkcg_exit_queue(struct request_queue *q);
 146
 147 /* Blkio controller policy registration */
 148 int blkcg_policy_register(struct blkcg_policy *pol);
 149 void blkcg_policy_unregister(struct blkcg_policy *pol);
 150 int blkcg_activate_policy(struct request_queue *q,
 151                           const struct blkcg_policy *pol);
 152 void blkcg_deactivate_policy(struct request_queue *q,
 153                              const struct blkcg_policy *pol);
 154
 155 void blkcg_print_blkgs(struct seq_file *sf, struct blkcg *blkcg,
 156                        u64 (*prfill)(struct seq_file *,
 157                                      struct blkg_policy_data *, int),
 158                        const struct blkcg_policy *pol, int data,
 159                        bool show_total);
 160 u64 __blkg_prfill_u64(struct seq_file *sf, struct blkg_policy_data *pd, u64 v);
 161 u64 __blkg_prfill_rwstat(struct seq_file *sf, struct blkg_policy_data *pd,
 162                          const struct blkg_rwstat *rwstat);
 163 u64 blkg_prfill_stat(struct seq_file *sf, struct blkg_policy_data *pd, int off);
 164 u64 blkg_prfill_rwstat(struct seq_file *sf, struct blkg_policy_data *pd,
 165                        int off);
 166
 167 u64 blkg_stat_recursive_sum(struct blkg_policy_data *pd, int off);
 168 struct blkg_rwstat blkg_rwstat_recursive_sum(struct blkg_policy_data *pd,
 169                                              int off);
 170
 171 struct blkg_conf_ctx {
 172         struct gendisk                  *disk;
 173         struct blkcg_gq                 *blkg;
 174         u64                             v;
 175 };
 176
 177 int blkg_conf_prep(struct blkcg *blkcg, const struct blkcg_policy *pol,
 178                    const char *input, struct blkg_conf_ctx *ctx);
 179 void blkg_conf_finish(struct blkg_conf_ctx *ctx);
 180
 181
 182 static inline struct blkcg *cgroup_to_blkcg(struct cgroup *cgroup)
 183 {
 184         return container_of(cgroup_subsys_state(cgroup, blkio_subsys_id),
 185                             struct blkcg, css);
 186 }
 187
 188 static inline struct blkcg *task_blkcg(struct task_struct *tsk)
 189 {
 190         return container_of(task_subsys_state(tsk, blkio_subsys_id),
 191                             struct blkcg, css);
 192 }
 193
 194 static inline struct blkcg *bio_blkcg(struct bio *bio)
 195 {
 196         if (bio && bio->bi_css)
 197                 return container_of(bio->bi_css, struct blkcg, css);
 198         return task_blkcg(current);
 199 }
 200
 201 /**
 202  * blkcg_parent - get the parent of a blkcg
 203  * @blkcg: blkcg of interest
 204  *
 205  * Return the parent blkcg of @blkcg.  Can be called anytime.
 206  */
 207 static inline struct blkcg *blkcg_parent(struct blkcg *blkcg)
 208 {
 209         struct cgroup *pcg = blkcg->css.cgroup->parent;
 210
 211         return pcg ? cgroup_to_blkcg(pcg) : NULL;
 212 }
 213
 214 /**
 215  * blkg_to_pdata - get policy private data
 216  * @blkg: blkg of interest
 217  * @pol: policy of interest
 218  *
 219  * Return pointer to private data associated with the @blkg-@pol pair.
 220  */
 221 static inline struct blkg_policy_data *blkg_to_pd(struct blkcg_gq *blkg,
 222                                                   struct blkcg_policy *pol)
 223 {
 224         return blkg ? blkg->pd[pol->plid] : NULL;
 225 }
 226
 227 /**
 228  * pdata_to_blkg - get blkg associated with policy private data
 229  * @pd: policy private data of interest
 230  *
 231  * @pd is policy private data.  Determine the blkg it's associated with.
 232  */
 233 static inline struct blkcg_gq *pd_to_blkg(struct blkg_policy_data *pd)
 234 {
 235         return pd ? pd->blkg : NULL;
 236 }
 237
 238 /**
 239  * blkg_path - format cgroup path of blkg
 240  * @blkg: blkg of interest
 241  * @buf: target buffer
 242  * @buflen: target buffer length
 243  *
 244  * Format the path of the cgroup of @blkg into @buf.
 245  */
 246 static inline int blkg_path(struct blkcg_gq *blkg, char *buf, int buflen)
 247 {
 248         int ret;
 249
 250         ret = cgroup_path(blkg->blkcg->css.cgroup, buf, buflen);
 251         if (ret)
 252                 strncpy(buf, "<unavailable>", buflen);
 253         return ret;
 254 }
 255
 256 /**
 257  * blkg_get - get a blkg reference
 258  * @blkg: blkg to get
 259  *
 260  * The caller should be holding queue_lock and an existing reference.
 261  */
 262 static inline void blkg_get(struct blkcg_gq *blkg)
 263 {
 264         lockdep_assert_held(blkg->q->queue_lock);
 265         WARN_ON_ONCE(!blkg->refcnt);
 266         blkg->refcnt++;
 267 }
 268
 269 void __blkg_release_rcu(struct rcu_head *rcu);
 270
 271 /**
 272  * blkg_put - put a blkg reference
 273  * @blkg: blkg to put
 274  *
 275  * The caller should be holding queue_lock.
 276  */
 277 static inline void blkg_put(struct blkcg_gq *blkg)
 278 {
 279         lockdep_assert_held(blkg->q->queue_lock);
 280         WARN_ON_ONCE(blkg->refcnt <= 0);
 281         if (!--blkg->refcnt)
 282                 call_rcu(&blkg->rcu_head, __blkg_release_rcu);
 283 }
 284
 285 struct blkcg_gq *__blkg_lookup(struct blkcg *blkcg, struct request_queue *q,
 286                                bool update_hint);
 287
 288 /**
 289  * blkg_for_each_descendant_pre - pre-order walk of a blkg's descendants
 290  * @d_blkg: loop cursor pointing to the current descendant
 291  * @pos_cgrp: used for iteration
 292  * @p_blkg: target blkg to walk descendants of
 293  *
 294  * Walk @c_blkg through the descendants of @p_blkg.  Must be used with RCU
 295  * read locked.  If called under either blkcg or queue lock, the iteration
 296  * is guaranteed to include all and only online blkgs.  The caller may
 297  * update @pos_cgrp by calling cgroup_rightmost_descendant() to skip
 298  * subtree.
 299  */
 300 #define blkg_for_each_descendant_pre(d_blkg, pos_cgrp, p_blkg)          \
 301         cgroup_for_each_descendant_pre((pos_cgrp), (p_blkg)->blkcg->css.cgroup) \
 302                 if (((d_blkg) = __blkg_lookup(cgroup_to_blkcg(pos_cgrp), \
 303                                               (p_blkg)->q, false)))
 304
 305 /**
 306  * blkg_for_each_descendant_post - post-order walk of a blkg's descendants
 307  * @d_blkg: loop cursor pointing to the current descendant
 308  * @pos_cgrp: used for iteration
 309  * @p_blkg: target blkg to walk descendants of
 310  *
 311  * Similar to blkg_for_each_descendant_pre() but performs post-order
 312  * traversal instead.  Synchronization rules are the same.
 313  */
 314 #define blkg_for_each_descendant_post(d_blkg, pos_cgrp, p_blkg)         \
 315         cgroup_for_each_descendant_post((pos_cgrp), (p_blkg)->blkcg->css.cgroup) \
 316                 if (((d_blkg) = __blkg_lookup(cgroup_to_blkcg(pos_cgrp), \
 317                                               (p_blkg)->q, false)))
 318
 319 /**
 320  * blk_get_rl - get request_list to use
 321  * @q: request_queue of interest
 322  * @bio: bio which will be attached to the allocated request (may be %NULL)
 323  *
 324  * The caller wants to allocate a request from @q to use for @bio.  Find
 325  * the request_list to use and obtain a reference on it.  Should be called
 326  * under queue_lock.  This function is guaranteed to return non-%NULL
 327  * request_list.
 328  */
 329 static inline struct request_list *blk_get_rl(struct request_queue *q,
 330                                               struct bio *bio)
 331 {
 332         struct blkcg *blkcg;
 333         struct blkcg_gq *blkg;
 334
 335         rcu_read_lock();
 336
 337         blkcg = bio_blkcg(bio);
 338
 339         /* bypass blkg lookup and use @q->root_rl directly for root */
 340         if (blkcg == &blkcg_root)
 341                 goto root_rl;
 342
 343         /*
 344          * Try to use blkg->rl.  blkg lookup may fail under memory pressure
 345          * or if either the blkcg or queue is going away.  Fall back to
 346          * root_rl in such cases.
 347          */
 348         blkg = blkg_lookup_create(blkcg, q);
 349         if (unlikely(IS_ERR(blkg)))
 350                 goto root_rl;
 351
 352         blkg_get(blkg);
 353         rcu_read_unlock();
 354         return &blkg->rl;
 355 root_rl:
 356         rcu_read_unlock();
 357         return &q->root_rl;
 358 }
 359
 360 /**
 361  * blk_put_rl - put request_list
 362  * @rl: request_list to put
 363  *
 364  * Put the reference acquired by blk_get_rl().  Should be called under
 365  * queue_lock.
 366  */
 367 static inline void blk_put_rl(struct request_list *rl)
 368 {
 369         /* root_rl may not have blkg set */
 370         if (rl->blkg && rl->blkg->blkcg != &blkcg_root)
 371                 blkg_put(rl->blkg);
 372 }
 373
 374 /**
 375  * blk_rq_set_rl - associate a request with a request_list
 376  * @rq: request of interest
 377  * @rl: target request_list
 378  *
 379  * Associate @rq with @rl so that accounting and freeing can know the
 380  * request_list @rq came from.
 381  */
 382 static inline void blk_rq_set_rl(struct request *rq, struct request_list *rl)
 383 {
 384         rq->rl = rl;
 385 }
 386
 387 /**
 388  * blk_rq_rl - return the request_list a request came from
 389  * @rq: request of interest
 390  *
 391  * Return the request_list @rq is allocated from.
 392  */
 393 static inline struct request_list *blk_rq_rl(struct request *rq)
 394 {
 395         return rq->rl;
 396 }
 397
 398 struct request_list *__blk_queue_next_rl(struct request_list *rl,
 399                                          struct request_queue *q);
 400 /**
 401  * blk_queue_for_each_rl - iterate through all request_lists of a request_queue
 402  *
 403  * Should be used under queue_lock.
 404  */
 405 #define blk_queue_for_each_rl(rl, q)    \
 406         for ((rl) = &(q)->root_rl; (rl); (rl) = __blk_queue_next_rl((rl), (q)))
 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  */
 416 static 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  */
 430 static 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(&stat->syncp);
 437                 v = stat->cnt;
 438         } while (u64_stats_fetch_retry(&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  */
 447 static 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  */
 459 static 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
 464 /**
 465  * blkg_rwstat_add - add a value to a blkg_rwstat
 466  * @rwstat: target blkg_rwstat
 467  * @rw: mask of REQ_{WRITE|SYNC}
 468  * @val: value to add
 469  *
 470  * Add @val to @rwstat.  The counters are chosen according to @rw.  The
 471  * caller is responsible for synchronizing calls to this function.
 472  */
 473 static inline void blkg_rwstat_add(struct blkg_rwstat *rwstat,
 474                                    int rw, uint64_t val)
 475 {
 476         u64_stats_update_begin(&rwstat->syncp);
 477
 478         if (rw & REQ_WRITE)
 479                 rwstat->cnt[BLKG_RWSTAT_WRITE] += val;
 480         else
 481                 rwstat->cnt[BLKG_RWSTAT_READ] += val;
 482         if (rw & REQ_SYNC)
 483                 rwstat->cnt[BLKG_RWSTAT_SYNC] += val;
 484         else
 485                 rwstat->cnt[BLKG_RWSTAT_ASYNC] += val;
 486
 487         u64_stats_update_end(&rwstat->syncp);
 488 }
 489
 490 /**
 491  * blkg_rwstat_read - read the current values of a blkg_rwstat
 492  * @rwstat: blkg_rwstat to read
 493  *
 494  * Read the current snapshot of @rwstat and return it as the return value.
 495  * This function can be called without synchronization and takes care of
 496  * u64 atomicity.
 497  */
 498 static inline struct blkg_rwstat blkg_rwstat_read(struct blkg_rwstat *rwstat)
 499 {
 500         unsigned int start;
 501         struct blkg_rwstat tmp;
 502
 503         do {
 504                 start = u64_stats_fetch_begin(&rwstat->syncp);
 505                 tmp = *rwstat;
 506         } while (u64_stats_fetch_retry(&rwstat->syncp, start));
 507
 508         return tmp;
 509 }
 510
 511 /**
 512  * blkg_rwstat_total - read the total count of a blkg_rwstat
 513  * @rwstat: blkg_rwstat to read
 514  *
 515  * Return the total count of @rwstat regardless of the IO direction.  This
 516  * function can be called without synchronization and takes care of u64
 517  * atomicity.
 518  */
 519 static inline uint64_t blkg_rwstat_total(struct blkg_rwstat *rwstat)
 520 {
 521         struct blkg_rwstat tmp = blkg_rwstat_read(rwstat);
 522
 523         return tmp.cnt[BLKG_RWSTAT_READ] + tmp.cnt[BLKG_RWSTAT_WRITE];
 524 }
 525
 526 /**
 527  * blkg_rwstat_reset - reset a blkg_rwstat
 528  * @rwstat: blkg_rwstat to reset
 529  */
 530 static inline void blkg_rwstat_reset(struct blkg_rwstat *rwstat)
 531 {
 532         memset(rwstat->cnt, 0, sizeof(rwstat->cnt));
 533 }
 534
 535 /**
 536  * blkg_rwstat_merge - merge a blkg_rwstat into another
 537  * @to: the destination blkg_rwstat
 538  * @from: the source
 539  *
 540  * Add @from's counts to @to.
 541  */
 542 static inline void blkg_rwstat_merge(struct blkg_rwstat *to,
 543                                      struct blkg_rwstat *from)
 544 {
 545         struct blkg_rwstat v = blkg_rwstat_read(from);
 546         int i;
 547
 548         u64_stats_update_begin(&to->syncp);
 549         for (i = 0; i < BLKG_RWSTAT_NR; i++)
 550                 to->cnt[i] += v.cnt[i];
 551         u64_stats_update_end(&to->syncp);
 552 }
 553
 554 #else   /* CONFIG_BLK_CGROUP */
 555
 556 struct cgroup;
 557 struct blkcg;
 558
 559 struct blkg_policy_data {
 560 };
 561
 562 struct blkcg_gq {
 563 };
 564
 565 struct blkcg_policy {
 566 };
 567
 568 static inline struct blkcg_gq *blkg_lookup(struct blkcg *blkcg, void *key) { return NULL; }
 569 static inline int blkcg_init_queue(struct request_queue *q) { return 0; }
 570 static inline void blkcg_drain_queue(struct request_queue *q) { }
 571 static inline void blkcg_exit_queue(struct request_queue *q) { }
 572 static inline int blkcg_policy_register(struct blkcg_policy *pol) { return 0; }
 573 static inline void blkcg_policy_unregister(struct blkcg_policy *pol) { }
 574 static inline int blkcg_activate_policy(struct request_queue *q,
 575                                         const struct blkcg_policy *pol) { return 0; }
 576 static inline void blkcg_deactivate_policy(struct request_queue *q,
 577                                            const struct blkcg_policy *pol) { }
 578
 579 static inline struct blkcg *cgroup_to_blkcg(struct cgroup *cgroup) { return NULL; }
 580 static inline struct blkcg *bio_blkcg(struct bio *bio) { return NULL; }
 581
 582 static inline struct blkg_policy_data *blkg_to_pd(struct blkcg_gq *blkg,
 583                                                   struct blkcg_policy *pol) { return NULL; }
 584 static inline struct blkcg_gq *pd_to_blkg(struct blkg_policy_data *pd) { return NULL; }
 585 static inline char *blkg_path(struct blkcg_gq *blkg) { return NULL; }
 586 static inline void blkg_get(struct blkcg_gq *blkg) { }
 587 static inline void blkg_put(struct blkcg_gq *blkg) { }
 588
 589 static inline struct request_list *blk_get_rl(struct request_queue *q,
 590                                               struct bio *bio) { return &q->root_rl; }
 591 static inline void blk_put_rl(struct request_list *rl) { }
 592 static inline void blk_rq_set_rl(struct request *rq, struct request_list *rl) { }
 593 static inline struct request_list *blk_rq_rl(struct request *rq) { return &rq->q->root_rl; }
 594
 595 #define blk_queue_for_each_rl(rl, q)    \
 596         for ((rl) = &(q)->root_rl; (rl); (rl) = NULL)
 597
 598 #endif  /* CONFIG_BLK_CGROUP */
 599 #endif  /* _BLK_CGROUP_H */