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(struct blkcg_gq *blkg);
 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                 __blkg_release(blkg);
 283 }
 284
 285 /**
 286  * blk_get_rl - get request_list to use
 287  * @q: request_queue of interest
 288  * @bio: bio which will be attached to the allocated request (may be %NULL)
 289  *
 290  * The caller wants to allocate a request from @q to use for @bio.  Find
 291  * the request_list to use and obtain a reference on it.  Should be called
 292  * under queue_lock.  This function is guaranteed to return non-%NULL
 293  * request_list.
 294  */
 295 static inline struct request_list *blk_get_rl(struct request_queue *q,
 296                                               struct bio *bio)
 297 {
 298         struct blkcg *blkcg;
 299         struct blkcg_gq *blkg;
 300
 301         rcu_read_lock();
 302
 303         blkcg = bio_blkcg(bio);
 304
 305         /* bypass blkg lookup and use @q->root_rl directly for root */
 306         if (blkcg == &blkcg_root)
 307                 goto root_rl;
 308
 309         /*
 310          * Try to use blkg->rl.  blkg lookup may fail under memory pressure
 311          * or if either the blkcg or queue is going away.  Fall back to
 312          * root_rl in such cases.
 313          */
 314         blkg = blkg_lookup_create(blkcg, q);
 315         if (unlikely(IS_ERR(blkg)))
 316                 goto root_rl;
 317
 318         blkg_get(blkg);
 319         rcu_read_unlock();
 320         return &blkg->rl;
 321 root_rl:
 322         rcu_read_unlock();
 323         return &q->root_rl;
 324 }
 325
 326 /**
 327  * blk_put_rl - put request_list
 328  * @rl: request_list to put
 329  *
 330  * Put the reference acquired by blk_get_rl().  Should be called under
 331  * queue_lock.
 332  */
 333 static inline void blk_put_rl(struct request_list *rl)
 334 {
 335         /* root_rl may not have blkg set */
 336         if (rl->blkg && rl->blkg->blkcg != &blkcg_root)
 337                 blkg_put(rl->blkg);
 338 }
 339
 340 /**
 341  * blk_rq_set_rl - associate a request with a request_list
 342  * @rq: request of interest
 343  * @rl: target request_list
 344  *
 345  * Associate @rq with @rl so that accounting and freeing can know the
 346  * request_list @rq came from.
 347  */
 348 static inline void blk_rq_set_rl(struct request *rq, struct request_list *rl)
 349 {
 350         rq->rl = rl;
 351 }
 352
 353 /**
 354  * blk_rq_rl - return the request_list a request came from
 355  * @rq: request of interest
 356  *
 357  * Return the request_list @rq is allocated from.
 358  */
 359 static inline struct request_list *blk_rq_rl(struct request *rq)
 360 {
 361         return rq->rl;
 362 }
 363
 364 struct request_list *__blk_queue_next_rl(struct request_list *rl,
 365                                          struct request_queue *q);
 366 /**
 367  * blk_queue_for_each_rl - iterate through all request_lists of a request_queue
 368  *
 369  * Should be used under queue_lock.
 370  */
 371 #define blk_queue_for_each_rl(rl, q)    \
 372         for ((rl) = &(q)->root_rl; (rl); (rl) = __blk_queue_next_rl((rl), (q)))
 373
 374 /**
 375  * blkg_stat_add - add a value to a blkg_stat
 376  * @stat: target blkg_stat
 377  * @val: value to add
 378  *
 379  * Add @val to @stat.  The caller is responsible for synchronizing calls to
 380  * this function.
 381  */
 382 static inline void blkg_stat_add(struct blkg_stat *stat, uint64_t val)
 383 {
 384         u64_stats_update_begin(&stat->syncp);
 385         stat->cnt += val;
 386         u64_stats_update_end(&stat->syncp);
 387 }
 388
 389 /**
 390  * blkg_stat_read - read the current value of a blkg_stat
 391  * @stat: blkg_stat to read
 392  *
 393  * Read the current value of @stat.  This function can be called without
 394  * synchroniztion and takes care of u64 atomicity.
 395  */
 396 static inline uint64_t blkg_stat_read(struct blkg_stat *stat)
 397 {
 398         unsigned int start;
 399         uint64_t v;
 400
 401         do {
 402                 start = u64_stats_fetch_begin_bh(&stat->syncp);
 403                 v = stat->cnt;
 404         } while (u64_stats_fetch_retry_bh(&stat->syncp, start));
 405
 406         return v;
 407 }
 408
 409 /**
 410  * blkg_stat_reset - reset a blkg_stat
 411  * @stat: blkg_stat to reset
 412  */
 413 static inline void blkg_stat_reset(struct blkg_stat *stat)
 414 {
 415         stat->cnt = 0;
 416 }
 417
 418 /**
 419  * blkg_stat_merge - merge a blkg_stat into another
 420  * @to: the destination blkg_stat
 421  * @from: the source
 422  *
 423  * Add @from's count to @to.
 424  */
 425 static inline void blkg_stat_merge(struct blkg_stat *to, struct blkg_stat *from)
 426 {
 427         blkg_stat_add(to, blkg_stat_read(from));
 428 }
 429
 430 /**
 431  * blkg_rwstat_add - add a value to a blkg_rwstat
 432  * @rwstat: target blkg_rwstat
 433  * @rw: mask of REQ_{WRITE|SYNC}
 434  * @val: value to add
 435  *
 436  * Add @val to @rwstat.  The counters are chosen according to @rw.  The
 437  * caller is responsible for synchronizing calls to this function.
 438  */
 439 static inline void blkg_rwstat_add(struct blkg_rwstat *rwstat,
 440                                    int rw, uint64_t val)
 441 {
 442         u64_stats_update_begin(&rwstat->syncp);
 443
 444         if (rw & REQ_WRITE)
 445                 rwstat->cnt[BLKG_RWSTAT_WRITE] += val;
 446         else
 447                 rwstat->cnt[BLKG_RWSTAT_READ] += val;
 448         if (rw & REQ_SYNC)
 449                 rwstat->cnt[BLKG_RWSTAT_SYNC] += val;
 450         else
 451                 rwstat->cnt[BLKG_RWSTAT_ASYNC] += val;
 452
 453         u64_stats_update_end(&rwstat->syncp);
 454 }
 455
 456 /**
 457  * blkg_rwstat_read - read the current values of a blkg_rwstat
 458  * @rwstat: blkg_rwstat to read
 459  *
 460  * Read the current snapshot of @rwstat and return it as the return value.
 461  * This function can be called without synchronization and takes care of
 462  * u64 atomicity.
 463  */
 464 static inline struct blkg_rwstat blkg_rwstat_read(struct blkg_rwstat *rwstat)
 465 {
 466         unsigned int start;
 467         struct blkg_rwstat tmp;
 468
 469         do {
 470                 start = u64_stats_fetch_begin_bh(&rwstat->syncp);
 471                 tmp = *rwstat;
 472         } while (u64_stats_fetch_retry_bh(&rwstat->syncp, start));
 473
 474         return tmp;
 475 }
 476
 477 /**
 478  * blkg_rwstat_total - read the total count of a blkg_rwstat
 479  * @rwstat: blkg_rwstat to read
 480  *
 481  * Return the total count of @rwstat regardless of the IO direction.  This
 482  * function can be called without synchronization and takes care of u64
 483  * atomicity.
 484  */
 485 static inline uint64_t blkg_rwstat_total(struct blkg_rwstat *rwstat)
 486 {
 487         struct blkg_rwstat tmp = blkg_rwstat_read(rwstat);
 488
 489         return tmp.cnt[BLKG_RWSTAT_READ] + tmp.cnt[BLKG_RWSTAT_WRITE];
 490 }
 491
 492 /**
 493  * blkg_rwstat_reset - reset a blkg_rwstat
 494  * @rwstat: blkg_rwstat to reset
 495  */
 496 static inline void blkg_rwstat_reset(struct blkg_rwstat *rwstat)
 497 {
 498         memset(rwstat->cnt, 0, sizeof(rwstat->cnt));
 499 }
 500
 501 /**
 502  * blkg_rwstat_merge - merge a blkg_rwstat into another
 503  * @to: the destination blkg_rwstat
 504  * @from: the source
 505  *
 506  * Add @from's counts to @to.
 507  */
 508 static inline void blkg_rwstat_merge(struct blkg_rwstat *to,
 509                                      struct blkg_rwstat *from)
 510 {
 511         struct blkg_rwstat v = blkg_rwstat_read(from);
 512         int i;
 513
 514         u64_stats_update_begin(&to->syncp);
 515         for (i = 0; i < BLKG_RWSTAT_NR; i++)
 516                 to->cnt[i] += v.cnt[i];
 517         u64_stats_update_end(&to->syncp);
 518 }
 519
 520 #else   /* CONFIG_BLK_CGROUP */
 521
 522 struct cgroup;
 523 struct blkcg;
 524
 525 struct blkg_policy_data {
 526 };
 527
 528 struct blkcg_gq {
 529 };
 530
 531 struct blkcg_policy {
 532 };
 533
 534 static inline struct blkcg_gq *blkg_lookup(struct blkcg *blkcg, void *key) { return NULL; }
 535 static inline int blkcg_init_queue(struct request_queue *q) { return 0; }
 536 static inline void blkcg_drain_queue(struct request_queue *q) { }
 537 static inline void blkcg_exit_queue(struct request_queue *q) { }
 538 static inline int blkcg_policy_register(struct blkcg_policy *pol) { return 0; }
 539 static inline void blkcg_policy_unregister(struct blkcg_policy *pol) { }
 540 static inline int blkcg_activate_policy(struct request_queue *q,
 541                                         const struct blkcg_policy *pol) { return 0; }
 542 static inline void blkcg_deactivate_policy(struct request_queue *q,
 543                                            const struct blkcg_policy *pol) { }
 544
 545 static inline struct blkcg *cgroup_to_blkcg(struct cgroup *cgroup) { return NULL; }
 546 static inline struct blkcg *bio_blkcg(struct bio *bio) { return NULL; }
 547
 548 static inline struct blkg_policy_data *blkg_to_pd(struct blkcg_gq *blkg,
 549                                                   struct blkcg_policy *pol) { return NULL; }
 550 static inline struct blkcg_gq *pd_to_blkg(struct blkg_policy_data *pd) { return NULL; }
 551 static inline char *blkg_path(struct blkcg_gq *blkg) { return NULL; }
 552 static inline void blkg_get(struct blkcg_gq *blkg) { }
 553 static inline void blkg_put(struct blkcg_gq *blkg) { }
 554
 555 static inline struct request_list *blk_get_rl(struct request_queue *q,
 556                                               struct bio *bio) { return &q->root_rl; }
 557 static inline void blk_put_rl(struct request_list *rl) { }
 558 static inline void blk_rq_set_rl(struct request *rq, struct request_list *rl) { }
 559 static inline struct request_list *blk_rq_rl(struct request *rq) { return &rq->q->root_rl; }
 560
 561 #define blk_queue_for_each_rl(rl, q)    \
 562         for ((rl) = &(q)->root_rl; (rl); (rl) = NULL)
 563
 564 #endif  /* CONFIG_BLK_CGROUP */
 565 #endif  /* _BLK_CGROUP_H */