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 struct blkg_conf_ctx {
 168         struct gendisk                  *disk;
 169         struct blkcg_gq                 *blkg;
 170         u64                             v;
 171 };
 172
 173 int blkg_conf_prep(struct blkcg *blkcg, const struct blkcg_policy *pol,
 174                    const char *input, struct blkg_conf_ctx *ctx);
 175 void blkg_conf_finish(struct blkg_conf_ctx *ctx);
 176
 177
 178 static inline struct blkcg *cgroup_to_blkcg(struct cgroup *cgroup)
 179 {
 180         return container_of(cgroup_subsys_state(cgroup, blkio_subsys_id),
 181                             struct blkcg, css);
 182 }
 183
 184 static inline struct blkcg *task_blkcg(struct task_struct *tsk)
 185 {
 186         return container_of(task_subsys_state(tsk, blkio_subsys_id),
 187                             struct blkcg, css);
 188 }
 189
 190 static inline struct blkcg *bio_blkcg(struct bio *bio)
 191 {
 192         if (bio && bio->bi_css)
 193                 return container_of(bio->bi_css, struct blkcg, 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  */
 203 static inline struct blkcg *blkcg_parent(struct blkcg *blkcg)
 204 {
 205         struct cgroup *pcg = blkcg->css.cgroup->parent;
 206
 207         return pcg ? cgroup_to_blkcg(pcg) : NULL;
 208 }
 209
 210 /**
 211  * blkg_to_pdata - get policy private data
 212  * @blkg: blkg of interest
 213  * @pol: policy of interest
 214  *
 215  * Return pointer to private data associated with the @blkg-@pol pair.
 216  */
 217 static inline struct blkg_policy_data *blkg_to_pd(struct blkcg_gq *blkg,
 218                                                   struct blkcg_policy *pol)
 219 {
 220         return blkg ? blkg->pd[pol->plid] : NULL;
 221 }
 222
 223 /**
 224  * pdata_to_blkg - get blkg associated with policy private data
 225  * @pd: policy private data of interest
 226  *
 227  * @pd is policy private data.  Determine the blkg it's associated with.
 228  */
 229 static inline struct blkcg_gq *pd_to_blkg(struct blkg_policy_data *pd)
 230 {
 231         return pd ? pd->blkg : NULL;
 232 }
 233
 234 /**
 235  * blkg_path - format cgroup path of blkg
 236  * @blkg: blkg of interest
 237  * @buf: target buffer
 238  * @buflen: target buffer length
 239  *
 240  * Format the path of the cgroup of @blkg into @buf.
 241  */
 242 static inline int blkg_path(struct blkcg_gq *blkg, char *buf, int buflen)
 243 {
 244         int ret;
 245
 246         rcu_read_lock();
 247         ret = cgroup_path(blkg->blkcg->css.cgroup, buf, buflen);
 248         rcu_read_unlock();
 249         if (ret)
 250                 strncpy(buf, "<unavailable>", buflen);
 251         return ret;
 252 }
 253
 254 /**
 255  * blkg_get - get a blkg reference
 256  * @blkg: blkg to get
 257  *
 258  * The caller should be holding queue_lock and an existing reference.
 259  */
 260 static inline void blkg_get(struct blkcg_gq *blkg)
 261 {
 262         lockdep_assert_held(blkg->q->queue_lock);
 263         WARN_ON_ONCE(!blkg->refcnt);
 264         blkg->refcnt++;
 265 }
 266
 267 void __blkg_release(struct blkcg_gq *blkg);
 268
 269 /**
 270  * blkg_put - put a blkg reference
 271  * @blkg: blkg to put
 272  *
 273  * The caller should be holding queue_lock.
 274  */
 275 static inline void blkg_put(struct blkcg_gq *blkg)
 276 {
 277         lockdep_assert_held(blkg->q->queue_lock);
 278         WARN_ON_ONCE(blkg->refcnt <= 0);
 279         if (!--blkg->refcnt)
 280                 __blkg_release(blkg);
 281 }
 282
 283 /**
 284  * blk_get_rl - get request_list to use
 285  * @q: request_queue of interest
 286  * @bio: bio which will be attached to the allocated request (may be %NULL)
 287  *
 288  * The caller wants to allocate a request from @q to use for @bio.  Find
 289  * the request_list to use and obtain a reference on it.  Should be called
 290  * under queue_lock.  This function is guaranteed to return non-%NULL
 291  * request_list.
 292  */
 293 static inline struct request_list *blk_get_rl(struct request_queue *q,
 294                                               struct bio *bio)
 295 {
 296         struct blkcg *blkcg;
 297         struct blkcg_gq *blkg;
 298
 299         rcu_read_lock();
 300
 301         blkcg = bio_blkcg(bio);
 302
 303         /* bypass blkg lookup and use @q->root_rl directly for root */
 304         if (blkcg == &blkcg_root)
 305                 goto root_rl;
 306
 307         /*
 308          * Try to use blkg->rl.  blkg lookup may fail under memory pressure
 309          * or if either the blkcg or queue is going away.  Fall back to
 310          * root_rl in such cases.
 311          */
 312         blkg = blkg_lookup_create(blkcg, q);
 313         if (unlikely(IS_ERR(blkg)))
 314                 goto root_rl;
 315
 316         blkg_get(blkg);
 317         rcu_read_unlock();
 318         return &blkg->rl;
 319 root_rl:
 320         rcu_read_unlock();
 321         return &q->root_rl;
 322 }
 323
 324 /**
 325  * blk_put_rl - put request_list
 326  * @rl: request_list to put
 327  *
 328  * Put the reference acquired by blk_get_rl().  Should be called under
 329  * queue_lock.
 330  */
 331 static inline void blk_put_rl(struct request_list *rl)
 332 {
 333         /* root_rl may not have blkg set */
 334         if (rl->blkg && rl->blkg->blkcg != &blkcg_root)
 335                 blkg_put(rl->blkg);
 336 }
 337
 338 /**
 339  * blk_rq_set_rl - associate a request with a request_list
 340  * @rq: request of interest
 341  * @rl: target request_list
 342  *
 343  * Associate @rq with @rl so that accounting and freeing can know the
 344  * request_list @rq came from.
 345  */
 346 static inline void blk_rq_set_rl(struct request *rq, struct request_list *rl)
 347 {
 348         rq->rl = rl;
 349 }
 350
 351 /**
 352  * blk_rq_rl - return the request_list a request came from
 353  * @rq: request of interest
 354  *
 355  * Return the request_list @rq is allocated from.
 356  */
 357 static inline struct request_list *blk_rq_rl(struct request *rq)
 358 {
 359         return rq->rl;
 360 }
 361
 362 struct request_list *__blk_queue_next_rl(struct request_list *rl,
 363                                          struct request_queue *q);
 364 /**
 365  * blk_queue_for_each_rl - iterate through all request_lists of a request_queue
 366  *
 367  * Should be used under queue_lock.
 368  */
 369 #define blk_queue_for_each_rl(rl, q)    \
 370         for ((rl) = &(q)->root_rl; (rl); (rl) = __blk_queue_next_rl((rl), (q)))
 371
 372 /**
 373  * blkg_stat_add - add a value to a blkg_stat
 374  * @stat: target blkg_stat
 375  * @val: value to add
 376  *
 377  * Add @val to @stat.  The caller is responsible for synchronizing calls to
 378  * this function.
 379  */
 380 static inline void blkg_stat_add(struct blkg_stat *stat, uint64_t val)
 381 {
 382         u64_stats_update_begin(&stat->syncp);
 383         stat->cnt += val;
 384         u64_stats_update_end(&stat->syncp);
 385 }
 386
 387 /**
 388  * blkg_stat_read - read the current value of a blkg_stat
 389  * @stat: blkg_stat to read
 390  *
 391  * Read the current value of @stat.  This function can be called without
 392  * synchroniztion and takes care of u64 atomicity.
 393  */
 394 static inline uint64_t blkg_stat_read(struct blkg_stat *stat)
 395 {
 396         unsigned int start;
 397         uint64_t v;
 398
 399         do {
 400                 start = u64_stats_fetch_begin(&stat->syncp);
 401                 v = stat->cnt;
 402         } while (u64_stats_fetch_retry(&stat->syncp, start));
 403
 404         return v;
 405 }
 406
 407 /**
 408  * blkg_stat_reset - reset a blkg_stat
 409  * @stat: blkg_stat to reset
 410  */
 411 static inline void blkg_stat_reset(struct blkg_stat *stat)
 412 {
 413         stat->cnt = 0;
 414 }
 415
 416 /**
 417  * blkg_rwstat_add - add a value to a blkg_rwstat
 418  * @rwstat: target blkg_rwstat
 419  * @rw: mask of REQ_{WRITE|SYNC}
 420  * @val: value to add
 421  *
 422  * Add @val to @rwstat.  The counters are chosen according to @rw.  The
 423  * caller is responsible for synchronizing calls to this function.
 424  */
 425 static inline void blkg_rwstat_add(struct blkg_rwstat *rwstat,
 426                                    int rw, uint64_t val)
 427 {
 428         u64_stats_update_begin(&rwstat->syncp);
 429
 430         if (rw & REQ_WRITE)
 431                 rwstat->cnt[BLKG_RWSTAT_WRITE] += val;
 432         else
 433                 rwstat->cnt[BLKG_RWSTAT_READ] += val;
 434         if (rw & REQ_SYNC)
 435                 rwstat->cnt[BLKG_RWSTAT_SYNC] += val;
 436         else
 437                 rwstat->cnt[BLKG_RWSTAT_ASYNC] += val;
 438
 439         u64_stats_update_end(&rwstat->syncp);
 440 }
 441
 442 /**
 443  * blkg_rwstat_read - read the current values of a blkg_rwstat
 444  * @rwstat: blkg_rwstat to read
 445  *
 446  * Read the current snapshot of @rwstat and return it as the return value.
 447  * This function can be called without synchronization and takes care of
 448  * u64 atomicity.
 449  */
 450 static inline struct blkg_rwstat blkg_rwstat_read(struct blkg_rwstat *rwstat)
 451 {
 452         unsigned int start;
 453         struct blkg_rwstat tmp;
 454
 455         do {
 456                 start = u64_stats_fetch_begin(&rwstat->syncp);
 457                 tmp = *rwstat;
 458         } while (u64_stats_fetch_retry(&rwstat->syncp, start));
 459
 460         return tmp;
 461 }
 462
 463 /**
 464  * blkg_rwstat_sum - read the total count of a blkg_rwstat
 465  * @rwstat: blkg_rwstat to read
 466  *
 467  * Return the total count of @rwstat regardless of the IO direction.  This
 468  * function can be called without synchronization and takes care of u64
 469  * atomicity.
 470  */
 471 static inline uint64_t blkg_rwstat_sum(struct blkg_rwstat *rwstat)
 472 {
 473         struct blkg_rwstat tmp = blkg_rwstat_read(rwstat);
 474
 475         return tmp.cnt[BLKG_RWSTAT_READ] + tmp.cnt[BLKG_RWSTAT_WRITE];
 476 }
 477
 478 /**
 479  * blkg_rwstat_reset - reset a blkg_rwstat
 480  * @rwstat: blkg_rwstat to reset
 481  */
 482 static inline void blkg_rwstat_reset(struct blkg_rwstat *rwstat)
 483 {
 484         memset(rwstat->cnt, 0, sizeof(rwstat->cnt));
 485 }
 486
 487 #else   /* CONFIG_BLK_CGROUP */
 488
 489 struct cgroup;
 490 struct blkcg;
 491
 492 struct blkg_policy_data {
 493 };
 494
 495 struct blkcg_gq {
 496 };
 497
 498 struct blkcg_policy {
 499 };
 500
 501 static inline struct blkcg_gq *blkg_lookup(struct blkcg *blkcg, void *key) { return NULL; }
 502 static inline int blkcg_init_queue(struct request_queue *q) { return 0; }
 503 static inline void blkcg_drain_queue(struct request_queue *q) { }
 504 static inline void blkcg_exit_queue(struct request_queue *q) { }
 505 static inline int blkcg_policy_register(struct blkcg_policy *pol) { return 0; }
 506 static inline void blkcg_policy_unregister(struct blkcg_policy *pol) { }
 507 static inline int blkcg_activate_policy(struct request_queue *q,
 508                                         const struct blkcg_policy *pol) { return 0; }
 509 static inline void blkcg_deactivate_policy(struct request_queue *q,
 510                                            const struct blkcg_policy *pol) { }
 511
 512 static inline struct blkcg *cgroup_to_blkcg(struct cgroup *cgroup) { return NULL; }
 513 static inline struct blkcg *bio_blkcg(struct bio *bio) { return NULL; }
 514
 515 static inline struct blkg_policy_data *blkg_to_pd(struct blkcg_gq *blkg,
 516                                                   struct blkcg_policy *pol) { return NULL; }
 517 static inline struct blkcg_gq *pd_to_blkg(struct blkg_policy_data *pd) { return NULL; }
 518 static inline char *blkg_path(struct blkcg_gq *blkg) { return NULL; }
 519 static inline void blkg_get(struct blkcg_gq *blkg) { }
 520 static inline void blkg_put(struct blkcg_gq *blkg) { }
 521
 522 static inline struct request_list *blk_get_rl(struct request_queue *q,
 523                                               struct bio *bio) { return &q->root_rl; }
 524 static inline void blk_put_rl(struct request_list *rl) { }
 525 static inline void blk_rq_set_rl(struct request *rq, struct request_list *rl) { }
 526 static inline struct request_list *blk_rq_rl(struct request *rq) { return &rq->q->root_rl; }
 527
 528 #define blk_queue_for_each_rl(rl, q)    \
 529         for ((rl) = &(q)->root_rl; (rl); (rl) = NULL)
 530
 531 #endif  /* CONFIG_BLK_CGROUP */
 532 #endif  /* _BLK_CGROUP_H */