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 *css_to_blkcg(struct cgroup_subsys_state *css)
 183 {
 184         return css ? container_of(css, struct blkcg, css) : NULL;
 185 }
 186
 187 static inline struct blkcg *cgroup_to_blkcg(struct cgroup *cgroup)
 188 {
 189         return css_to_blkcg(cgroup_css(cgroup, blkio_subsys_id));
 190 }
 191
 192 static inline struct blkcg *task_blkcg(struct task_struct *tsk)
 193 {
 194         return css_to_blkcg(task_css(tsk, blkio_subsys_id));
 195 }
 196
 197 static inline struct blkcg *bio_blkcg(struct bio *bio)
 198 {
 199         if (bio && bio->bi_css)
 200                 return css_to_blkcg(bio->bi_css);
 201         return task_blkcg(current);
 202 }
 203
 204 /**
 205  * blkcg_parent - get the parent of a blkcg
 206  * @blkcg: blkcg of interest
 207  *
 208  * Return the parent blkcg of @blkcg.  Can be called anytime.
 209  */
 210 static inline struct blkcg *blkcg_parent(struct blkcg *blkcg)
 211 {
 212         struct cgroup *pcg = blkcg->css.cgroup->parent;
 213
 214         return pcg ? cgroup_to_blkcg(pcg) : NULL;
 215 }
 216
 217 /**
 218  * blkg_to_pdata - get policy private data
 219  * @blkg: blkg of interest
 220  * @pol: policy of interest
 221  *
 222  * Return pointer to private data associated with the @blkg-@pol pair.
 223  */
 224 static inline struct blkg_policy_data *blkg_to_pd(struct blkcg_gq *blkg,
 225                                                   struct blkcg_policy *pol)
 226 {
 227         return blkg ? blkg->pd[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 /**
 242  * blkg_path - format cgroup path of blkg
 243  * @blkg: blkg of interest
 244  * @buf: target buffer
 245  * @buflen: target buffer length
 246  *
 247  * Format the path of the cgroup of @blkg into @buf.
 248  */
 249 static inline int blkg_path(struct blkcg_gq *blkg, char *buf, int buflen)
 250 {
 251         int ret;
 252
 253         ret = cgroup_path(blkg->blkcg->css.cgroup, buf, buflen);
 254         if (ret)
 255                 strncpy(buf, "<unavailable>", buflen);
 256         return ret;
 257 }
 258
 259 /**
 260  * blkg_get - get a blkg reference
 261  * @blkg: blkg to get
 262  *
 263  * The caller should be holding queue_lock and an existing reference.
 264  */
 265 static inline void blkg_get(struct blkcg_gq *blkg)
 266 {
 267         lockdep_assert_held(blkg->q->queue_lock);
 268         WARN_ON_ONCE(!blkg->refcnt);
 269         blkg->refcnt++;
 270 }
 271
 272 void __blkg_release_rcu(struct rcu_head *rcu);
 273
 274 /**
 275  * blkg_put - put a blkg reference
 276  * @blkg: blkg to put
 277  *
 278  * The caller should be holding queue_lock.
 279  */
 280 static inline void blkg_put(struct blkcg_gq *blkg)
 281 {
 282         lockdep_assert_held(blkg->q->queue_lock);
 283         WARN_ON_ONCE(blkg->refcnt <= 0);
 284         if (!--blkg->refcnt)
 285                 call_rcu(&blkg->rcu_head, __blkg_release_rcu);
 286 }
 287
 288 struct blkcg_gq *__blkg_lookup(struct blkcg *blkcg, struct request_queue *q,
 289                                bool update_hint);
 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_cgrp: 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_cgrp by calling cgroup_rightmost_descendant() to skip
 301  * subtree.
 302  */
 303 #define blkg_for_each_descendant_pre(d_blkg, pos_cgrp, p_blkg)          \
 304         cgroup_for_each_descendant_pre((pos_cgrp), (p_blkg)->blkcg->css.cgroup) \
 305                 if (((d_blkg) = __blkg_lookup(cgroup_to_blkcg(pos_cgrp), \
 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_cgrp: 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.
 316  */
 317 #define blkg_for_each_descendant_post(d_blkg, pos_cgrp, p_blkg)         \
 318         cgroup_for_each_descendant_post((pos_cgrp), (p_blkg)->blkcg->css.cgroup) \
 319                 if (((d_blkg) = __blkg_lookup(cgroup_to_blkcg(pos_cgrp), \
 320                                               (p_blkg)->q, false)))
 321
 322 /**
 323  * blk_get_rl - get request_list to use
 324  * @q: request_queue of interest
 325  * @bio: bio which will be attached to the allocated request (may be %NULL)
 326  *
 327  * The caller wants to allocate a request from @q to use for @bio.  Find
 328  * the request_list to use and obtain a reference on it.  Should be called
 329  * under queue_lock.  This function is guaranteed to return non-%NULL
 330  * request_list.
 331  */
 332 static inline struct request_list *blk_get_rl(struct request_queue *q,
 333                                               struct bio *bio)
 334 {
 335         struct blkcg *blkcg;
 336         struct blkcg_gq *blkg;
 337
 338         rcu_read_lock();
 339
 340         blkcg = bio_blkcg(bio);
 341
 342         /* bypass blkg lookup and use @q->root_rl directly for root */
 343         if (blkcg == &blkcg_root)
 344                 goto root_rl;
 345
 346         /*
 347          * Try to use blkg->rl.  blkg lookup may fail under memory pressure
 348          * or if either the blkcg or queue is going away.  Fall back to
 349          * root_rl in such cases.
 350          */
 351         blkg = blkg_lookup_create(blkcg, q);
 352         if (unlikely(IS_ERR(blkg)))
 353                 goto root_rl;
 354
 355         blkg_get(blkg);
 356         rcu_read_unlock();
 357         return &blkg->rl;
 358 root_rl:
 359         rcu_read_unlock();
 360         return &q->root_rl;
 361 }
 362
 363 /**
 364  * blk_put_rl - put request_list
 365  * @rl: request_list to put
 366  *
 367  * Put the reference acquired by blk_get_rl().  Should be called under
 368  * queue_lock.
 369  */
 370 static inline void blk_put_rl(struct request_list *rl)
 371 {
 372         /* root_rl may not have blkg set */
 373         if (rl->blkg && rl->blkg->blkcg != &blkcg_root)
 374                 blkg_put(rl->blkg);
 375 }
 376
 377 /**
 378  * blk_rq_set_rl - associate a request with a request_list
 379  * @rq: request of interest
 380  * @rl: target request_list
 381  *
 382  * Associate @rq with @rl so that accounting and freeing can know the
 383  * request_list @rq came from.
 384  */
 385 static inline void blk_rq_set_rl(struct request *rq, struct request_list *rl)
 386 {
 387         rq->rl = rl;
 388 }
 389
 390 /**
 391  * blk_rq_rl - return the request_list a request came from
 392  * @rq: request of interest
 393  *
 394  * Return the request_list @rq is allocated from.
 395  */
 396 static inline struct request_list *blk_rq_rl(struct request *rq)
 397 {
 398         return rq->rl;
 399 }
 400
 401 struct request_list *__blk_queue_next_rl(struct request_list *rl,
 402                                          struct request_queue *q);
 403 /**
 404  * blk_queue_for_each_rl - iterate through all request_lists of a request_queue
 405  *
 406  * Should be used under queue_lock.
 407  */
 408 #define blk_queue_for_each_rl(rl, q)    \
 409         for ((rl) = &(q)->root_rl; (rl); (rl) = __blk_queue_next_rl((rl), (q)))
 410
 411 /**
 412  * blkg_stat_add - add a value to a blkg_stat
 413  * @stat: target blkg_stat
 414  * @val: value to add
 415  *
 416  * Add @val to @stat.  The caller is responsible for synchronizing calls to
 417  * this function.
 418  */
 419 static inline void blkg_stat_add(struct blkg_stat *stat, uint64_t val)
 420 {
 421         u64_stats_update_begin(&stat->syncp);
 422         stat->cnt += val;
 423         u64_stats_update_end(&stat->syncp);
 424 }
 425
 426 /**
 427  * blkg_stat_read - read the current value of a blkg_stat
 428  * @stat: blkg_stat to read
 429  *
 430  * Read the current value of @stat.  This function can be called without
 431  * synchroniztion and takes care of u64 atomicity.
 432  */
 433 static inline uint64_t blkg_stat_read(struct blkg_stat *stat)
 434 {
 435         unsigned int start;
 436         uint64_t v;
 437
 438         do {
 439                 start = u64_stats_fetch_begin(&stat->syncp);
 440                 v = stat->cnt;
 441         } while (u64_stats_fetch_retry(&stat->syncp, start));
 442
 443         return v;
 444 }
 445
 446 /**
 447  * blkg_stat_reset - reset a blkg_stat
 448  * @stat: blkg_stat to reset
 449  */
 450 static inline void blkg_stat_reset(struct blkg_stat *stat)
 451 {
 452         stat->cnt = 0;
 453 }
 454
 455 /**
 456  * blkg_stat_merge - merge a blkg_stat into another
 457  * @to: the destination blkg_stat
 458  * @from: the source
 459  *
 460  * Add @from's count to @to.
 461  */
 462 static inline void blkg_stat_merge(struct blkg_stat *to, struct blkg_stat *from)
 463 {
 464         blkg_stat_add(to, blkg_stat_read(from));
 465 }
 466
 467 /**
 468  * blkg_rwstat_add - add a value to a blkg_rwstat
 469  * @rwstat: target blkg_rwstat
 470  * @rw: mask of REQ_{WRITE|SYNC}
 471  * @val: value to add
 472  *
 473  * Add @val to @rwstat.  The counters are chosen according to @rw.  The
 474  * caller is responsible for synchronizing calls to this function.
 475  */
 476 static inline void blkg_rwstat_add(struct blkg_rwstat *rwstat,
 477                                    int rw, uint64_t val)
 478 {
 479         u64_stats_update_begin(&rwstat->syncp);
 480
 481         if (rw & REQ_WRITE)
 482                 rwstat->cnt[BLKG_RWSTAT_WRITE] += val;
 483         else
 484                 rwstat->cnt[BLKG_RWSTAT_READ] += val;
 485         if (rw & REQ_SYNC)
 486                 rwstat->cnt[BLKG_RWSTAT_SYNC] += val;
 487         else
 488                 rwstat->cnt[BLKG_RWSTAT_ASYNC] += val;
 489
 490         u64_stats_update_end(&rwstat->syncp);
 491 }
 492
 493 /**
 494  * blkg_rwstat_read - read the current values of a blkg_rwstat
 495  * @rwstat: blkg_rwstat to read
 496  *
 497  * Read the current snapshot of @rwstat and return it as the return value.
 498  * This function can be called without synchronization and takes care of
 499  * u64 atomicity.
 500  */
 501 static inline struct blkg_rwstat blkg_rwstat_read(struct blkg_rwstat *rwstat)
 502 {
 503         unsigned int start;
 504         struct blkg_rwstat tmp;
 505
 506         do {
 507                 start = u64_stats_fetch_begin(&rwstat->syncp);
 508                 tmp = *rwstat;
 509         } while (u64_stats_fetch_retry(&rwstat->syncp, start));
 510
 511         return tmp;
 512 }
 513
 514 /**
 515  * blkg_rwstat_total - read the total count of a blkg_rwstat
 516  * @rwstat: blkg_rwstat to read
 517  *
 518  * Return the total count of @rwstat regardless of the IO direction.  This
 519  * function can be called without synchronization and takes care of u64
 520  * atomicity.
 521  */
 522 static inline uint64_t blkg_rwstat_total(struct blkg_rwstat *rwstat)
 523 {
 524         struct blkg_rwstat tmp = blkg_rwstat_read(rwstat);
 525
 526         return tmp.cnt[BLKG_RWSTAT_READ] + tmp.cnt[BLKG_RWSTAT_WRITE];
 527 }
 528
 529 /**
 530  * blkg_rwstat_reset - reset a blkg_rwstat
 531  * @rwstat: blkg_rwstat to reset
 532  */
 533 static inline void blkg_rwstat_reset(struct blkg_rwstat *rwstat)
 534 {
 535         memset(rwstat->cnt, 0, sizeof(rwstat->cnt));
 536 }
 537
 538 /**
 539  * blkg_rwstat_merge - merge a blkg_rwstat into another
 540  * @to: the destination blkg_rwstat
 541  * @from: the source
 542  *
 543  * Add @from's counts to @to.
 544  */
 545 static inline void blkg_rwstat_merge(struct blkg_rwstat *to,
 546                                      struct blkg_rwstat *from)
 547 {
 548         struct blkg_rwstat v = blkg_rwstat_read(from);
 549         int i;
 550
 551         u64_stats_update_begin(&to->syncp);
 552         for (i = 0; i < BLKG_RWSTAT_NR; i++)
 553                 to->cnt[i] += v.cnt[i];
 554         u64_stats_update_end(&to->syncp);
 555 }
 556
 557 #else   /* CONFIG_BLK_CGROUP */
 558
 559 struct cgroup;
 560 struct blkcg;
 561
 562 struct blkg_policy_data {
 563 };
 564
 565 struct blkcg_gq {
 566 };
 567
 568 struct blkcg_policy {
 569 };
 570
 571 static inline struct blkcg_gq *blkg_lookup(struct blkcg *blkcg, void *key) { return NULL; }
 572 static inline int blkcg_init_queue(struct request_queue *q) { return 0; }
 573 static inline void blkcg_drain_queue(struct request_queue *q) { }
 574 static inline void blkcg_exit_queue(struct request_queue *q) { }
 575 static inline int blkcg_policy_register(struct blkcg_policy *pol) { return 0; }
 576 static inline void blkcg_policy_unregister(struct blkcg_policy *pol) { }
 577 static inline int blkcg_activate_policy(struct request_queue *q,
 578                                         const struct blkcg_policy *pol) { return 0; }
 579 static inline void blkcg_deactivate_policy(struct request_queue *q,
 580                                            const struct blkcg_policy *pol) { }
 581
 582 static inline struct blkcg *cgroup_to_blkcg(struct cgroup *cgroup) { return NULL; }
 583 static inline struct blkcg *bio_blkcg(struct bio *bio) { return NULL; }
 584
 585 static inline struct blkg_policy_data *blkg_to_pd(struct blkcg_gq *blkg,
 586                                                   struct blkcg_policy *pol) { return NULL; }
 587 static inline struct blkcg_gq *pd_to_blkg(struct blkg_policy_data *pd) { return NULL; }
 588 static inline char *blkg_path(struct blkcg_gq *blkg) { return NULL; }
 589 static inline void blkg_get(struct blkcg_gq *blkg) { }
 590 static inline void blkg_put(struct blkcg_gq *blkg) { }
 591
 592 static inline struct request_list *blk_get_rl(struct request_queue *q,
 593                                               struct bio *bio) { return &q->root_rl; }
 594 static inline void blk_put_rl(struct request_list *rl) { }
 595 static inline void blk_rq_set_rl(struct request *rq, struct request_list *rl) { }
 596 static inline struct request_list *blk_rq_rl(struct request *rq) { return &rq->q->root_rl; }
 597
 598 #define blk_queue_for_each_rl(rl, q)    \
 599         for ((rl) = &(q)->root_rl; (rl); (rl) = NULL)
 600
 601 #endif  /* CONFIG_BLK_CGROUP */
 602 #endif  /* _BLK_CGROUP_H */