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
  19 enum blkio_policy_id {
  20         BLKIO_POLICY_PROP = 0,          /* Proportional Bandwidth division */
  21         BLKIO_POLICY_THROTL,            /* Throttling */
  22
  23         BLKIO_NR_POLICIES,
  24 };
  25
  26 /* Max limits for throttle policy */
  27 #define THROTL_IOPS_MAX         UINT_MAX
  28
  29 #ifdef CONFIG_BLK_CGROUP
  30
  31 enum stat_type {
  32         /* Number of IOs merged */
  33         BLKIO_STAT_MERGED,
  34         /* Total time spent (in ns) between request dispatch to the driver and
  35          * request completion for IOs doen by this cgroup. This may not be
  36          * accurate when NCQ is turned on. */
  37         BLKIO_STAT_SERVICE_TIME,
  38         /* Total time spent waiting in scheduler queue in ns */
  39         BLKIO_STAT_WAIT_TIME,
  40         /* Number of IOs queued up */
  41         BLKIO_STAT_QUEUED,
  42
  43         /* All the single valued stats go below this */
  44         BLKIO_STAT_TIME,
  45 #ifdef CONFIG_DEBUG_BLK_CGROUP
  46         /* Time not charged to this cgroup */
  47         BLKIO_STAT_UNACCOUNTED_TIME,
  48         BLKIO_STAT_AVG_QUEUE_SIZE,
  49         BLKIO_STAT_IDLE_TIME,
  50         BLKIO_STAT_EMPTY_TIME,
  51         BLKIO_STAT_GROUP_WAIT_TIME,
  52         BLKIO_STAT_DEQUEUE
  53 #endif
  54 };
  55
  56 /* Types lower than this live in stat_arr and have subtypes */
  57 #define BLKIO_STAT_ARR_NR       (BLKIO_STAT_QUEUED + 1)
  58
  59 /* Per cpu stats */
  60 enum stat_type_cpu {
  61         BLKIO_STAT_CPU_SECTORS,
  62         /* Total bytes transferred */
  63         BLKIO_STAT_CPU_SERVICE_BYTES,
  64         /* Total IOs serviced, post merge */
  65         BLKIO_STAT_CPU_SERVICED,
  66         BLKIO_STAT_CPU_NR
  67 };
  68
  69 enum stat_sub_type {
  70         BLKIO_STAT_READ = 0,
  71         BLKIO_STAT_WRITE,
  72         BLKIO_STAT_SYNC,
  73         BLKIO_STAT_ASYNC,
  74         BLKIO_STAT_TOTAL
  75 };
  76
  77 /* blkg state flags */
  78 enum blkg_state_flags {
  79         BLKG_waiting = 0,
  80         BLKG_idling,
  81         BLKG_empty,
  82 };
  83
  84 /* cgroup files owned by proportional weight policy */
  85 enum blkcg_file_name_prop {
  86         BLKIO_PROP_weight = 1,
  87         BLKIO_PROP_weight_device,
  88         BLKIO_PROP_io_service_bytes,
  89         BLKIO_PROP_io_serviced,
  90         BLKIO_PROP_time,
  91         BLKIO_PROP_sectors,
  92         BLKIO_PROP_unaccounted_time,
  93         BLKIO_PROP_io_service_time,
  94         BLKIO_PROP_io_wait_time,
  95         BLKIO_PROP_io_merged,
  96         BLKIO_PROP_io_queued,
  97         BLKIO_PROP_avg_queue_size,
  98         BLKIO_PROP_group_wait_time,
  99         BLKIO_PROP_idle_time,
 100         BLKIO_PROP_empty_time,
 101         BLKIO_PROP_dequeue,
 102 };
 103
 104 /* cgroup files owned by throttle policy */
 105 enum blkcg_file_name_throtl {
 106         BLKIO_THROTL_read_bps_device,
 107         BLKIO_THROTL_write_bps_device,
 108         BLKIO_THROTL_read_iops_device,
 109         BLKIO_THROTL_write_iops_device,
 110         BLKIO_THROTL_io_service_bytes,
 111         BLKIO_THROTL_io_serviced,
 112 };
 113
 114 struct blkio_cgroup {
 115         struct cgroup_subsys_state css;
 116         unsigned int weight;
 117         spinlock_t lock;
 118         struct hlist_head blkg_list;
 119
 120         /* for policies to test whether associated blkcg has changed */
 121         uint64_t id;
 122 };
 123
 124 struct blkio_group_stats {
 125         struct u64_stats_sync syncp;
 126         /* total disk time and nr sectors dispatched by this group */
 127         uint64_t time;
 128         uint64_t stat_arr[BLKIO_STAT_ARR_NR][BLKIO_STAT_TOTAL];
 129 #ifdef CONFIG_DEBUG_BLK_CGROUP
 130         /* Time not charged to this cgroup */
 131         uint64_t unaccounted_time;
 132
 133         /* Sum of number of IOs queued across all samples */
 134         uint64_t avg_queue_size_sum;
 135         /* Count of samples taken for average */
 136         uint64_t avg_queue_size_samples;
 137         /* How many times this group has been removed from service tree */
 138         unsigned long dequeue;
 139
 140         /* Total time spent waiting for it to be assigned a timeslice. */
 141         uint64_t group_wait_time;
 142
 143         /* Time spent idling for this blkio_group */
 144         uint64_t idle_time;
 145         /*
 146          * Total time when we have requests queued and do not contain the
 147          * current active queue.
 148          */
 149         uint64_t empty_time;
 150
 151         /* fields after this shouldn't be cleared on stat reset */
 152         uint64_t start_group_wait_time;
 153         uint64_t start_idle_time;
 154         uint64_t start_empty_time;
 155         uint16_t flags;
 156 #endif
 157 };
 158
 159 #ifdef CONFIG_DEBUG_BLK_CGROUP
 160 #define BLKG_STATS_DEBUG_CLEAR_START    \
 161         offsetof(struct blkio_group_stats, unaccounted_time)
 162 #define BLKG_STATS_DEBUG_CLEAR_SIZE     \
 163         (offsetof(struct blkio_group_stats, start_group_wait_time) - \
 164          BLKG_STATS_DEBUG_CLEAR_START)
 165 #endif
 166
 167 /* Per cpu blkio group stats */
 168 struct blkio_group_stats_cpu {
 169         uint64_t sectors;
 170         uint64_t stat_arr_cpu[BLKIO_STAT_CPU_NR][BLKIO_STAT_TOTAL];
 171         struct u64_stats_sync syncp;
 172 };
 173
 174 struct blkio_group_conf {
 175         unsigned int weight;
 176         unsigned int iops[2];
 177         u64 bps[2];
 178 };
 179
 180 /* per-blkg per-policy data */
 181 struct blkg_policy_data {
 182         /* the blkg this per-policy data belongs to */
 183         struct blkio_group *blkg;
 184
 185         /* Configuration */
 186         struct blkio_group_conf conf;
 187
 188         struct blkio_group_stats stats;
 189         /* Per cpu stats pointer */
 190         struct blkio_group_stats_cpu __percpu *stats_cpu;
 191
 192         /* pol->pdata_size bytes of private data used by policy impl */
 193         char pdata[] __aligned(__alignof__(unsigned long long));
 194 };
 195
 196 struct blkio_group {
 197         /* Pointer to the associated request_queue */
 198         struct request_queue *q;
 199         struct list_head q_node;
 200         struct hlist_node blkcg_node;
 201         struct blkio_cgroup *blkcg;
 202         /* Store cgroup path */
 203         char path[128];
 204         /* reference count */
 205         int refcnt;
 206
 207         struct blkg_policy_data *pd[BLKIO_NR_POLICIES];
 208
 209         /* List of blkg waiting for per cpu stats memory to be allocated */
 210         struct list_head alloc_node;
 211         struct rcu_head rcu_head;
 212 };
 213
 214 typedef void (blkio_init_group_fn)(struct blkio_group *blkg);
 215 typedef void (blkio_update_group_weight_fn)(struct request_queue *q,
 216                         struct blkio_group *blkg, unsigned int weight);
 217 typedef void (blkio_update_group_read_bps_fn)(struct request_queue *q,
 218                         struct blkio_group *blkg, u64 read_bps);
 219 typedef void (blkio_update_group_write_bps_fn)(struct request_queue *q,
 220                         struct blkio_group *blkg, u64 write_bps);
 221 typedef void (blkio_update_group_read_iops_fn)(struct request_queue *q,
 222                         struct blkio_group *blkg, unsigned int read_iops);
 223 typedef void (blkio_update_group_write_iops_fn)(struct request_queue *q,
 224                         struct blkio_group *blkg, unsigned int write_iops);
 225
 226 struct blkio_policy_ops {
 227         blkio_init_group_fn *blkio_init_group_fn;
 228         blkio_update_group_weight_fn *blkio_update_group_weight_fn;
 229         blkio_update_group_read_bps_fn *blkio_update_group_read_bps_fn;
 230         blkio_update_group_write_bps_fn *blkio_update_group_write_bps_fn;
 231         blkio_update_group_read_iops_fn *blkio_update_group_read_iops_fn;
 232         blkio_update_group_write_iops_fn *blkio_update_group_write_iops_fn;
 233 };
 234
 235 struct blkio_policy_type {
 236         struct list_head list;
 237         struct blkio_policy_ops ops;
 238         enum blkio_policy_id plid;
 239         size_t pdata_size;              /* policy specific private data size */
 240 };
 241
 242 extern int blkcg_init_queue(struct request_queue *q);
 243 extern void blkcg_drain_queue(struct request_queue *q);
 244 extern void blkcg_exit_queue(struct request_queue *q);
 245
 246 /* Blkio controller policy registration */
 247 extern void blkio_policy_register(struct blkio_policy_type *);
 248 extern void blkio_policy_unregister(struct blkio_policy_type *);
 249 extern void blkg_destroy_all(struct request_queue *q, bool destroy_root);
 250 extern void update_root_blkg_pd(struct request_queue *q,
 251                                 enum blkio_policy_id plid);
 252
 253 /**
 254  * blkg_to_pdata - get policy private data
 255  * @blkg: blkg of interest
 256  * @pol: policy of interest
 257  *
 258  * Return pointer to private data associated with the @blkg-@pol pair.
 259  */
 260 static inline void *blkg_to_pdata(struct blkio_group *blkg,
 261                               struct blkio_policy_type *pol)
 262 {
 263         return blkg ? blkg->pd[pol->plid]->pdata : NULL;
 264 }
 265
 266 /**
 267  * pdata_to_blkg - get blkg associated with policy private data
 268  * @pdata: policy private data of interest
 269  * @pol: policy @pdata is for
 270  *
 271  * @pdata is policy private data for @pol.  Determine the blkg it's
 272  * associated with.
 273  */
 274 static inline struct blkio_group *pdata_to_blkg(void *pdata,
 275                                                 struct blkio_policy_type *pol)
 276 {
 277         if (pdata) {
 278                 struct blkg_policy_data *pd =
 279                         container_of(pdata, struct blkg_policy_data, pdata);
 280                 return pd->blkg;
 281         }
 282         return NULL;
 283 }
 284
 285 static inline char *blkg_path(struct blkio_group *blkg)
 286 {
 287         return blkg->path;
 288 }
 289
 290 /**
 291  * blkg_get - get a blkg reference
 292  * @blkg: blkg to get
 293  *
 294  * The caller should be holding queue_lock and an existing reference.
 295  */
 296 static inline void blkg_get(struct blkio_group *blkg)
 297 {
 298         lockdep_assert_held(blkg->q->queue_lock);
 299         WARN_ON_ONCE(!blkg->refcnt);
 300         blkg->refcnt++;
 301 }
 302
 303 void __blkg_release(struct blkio_group *blkg);
 304
 305 /**
 306  * blkg_put - put a blkg reference
 307  * @blkg: blkg to put
 308  *
 309  * The caller should be holding queue_lock.
 310  */
 311 static inline void blkg_put(struct blkio_group *blkg)
 312 {
 313         lockdep_assert_held(blkg->q->queue_lock);
 314         WARN_ON_ONCE(blkg->refcnt <= 0);
 315         if (!--blkg->refcnt)
 316                 __blkg_release(blkg);
 317 }
 318
 319 #else
 320
 321 struct blkio_group {
 322 };
 323
 324 struct blkio_policy_type {
 325 };
 326
 327 static inline int blkcg_init_queue(struct request_queue *q) { return 0; }
 328 static inline void blkcg_drain_queue(struct request_queue *q) { }
 329 static inline void blkcg_exit_queue(struct request_queue *q) { }
 330 static inline void blkio_policy_register(struct blkio_policy_type *blkiop) { }
 331 static inline void blkio_policy_unregister(struct blkio_policy_type *blkiop) { }
 332 static inline void blkg_destroy_all(struct request_queue *q,
 333                                     bool destory_root) { }
 334 static inline void update_root_blkg_pd(struct request_queue *q,
 335                                        enum blkio_policy_id plid) { }
 336
 337 static inline void *blkg_to_pdata(struct blkio_group *blkg,
 338                                 struct blkio_policy_type *pol) { return NULL; }
 339 static inline struct blkio_group *pdata_to_blkg(void *pdata,
 340                                 struct blkio_policy_type *pol) { return NULL; }
 341 static inline char *blkg_path(struct blkio_group *blkg) { return NULL; }
 342 static inline void blkg_get(struct blkio_group *blkg) { }
 343 static inline void blkg_put(struct blkio_group *blkg) { }
 344
 345 #endif
 346
 347 #define BLKIO_WEIGHT_MIN        10
 348 #define BLKIO_WEIGHT_MAX        1000
 349 #define BLKIO_WEIGHT_DEFAULT    500
 350
 351 #ifdef CONFIG_DEBUG_BLK_CGROUP
 352 void blkiocg_update_avg_queue_size_stats(struct blkio_group *blkg,
 353                                          struct blkio_policy_type *pol);
 354 void blkiocg_update_dequeue_stats(struct blkio_group *blkg,
 355                                   struct blkio_policy_type *pol,
 356                                   unsigned long dequeue);
 357 void blkiocg_update_set_idle_time_stats(struct blkio_group *blkg,
 358                                         struct blkio_policy_type *pol);
 359 void blkiocg_update_idle_time_stats(struct blkio_group *blkg,
 360                                     struct blkio_policy_type *pol);
 361 void blkiocg_set_start_empty_time(struct blkio_group *blkg,
 362                                   struct blkio_policy_type *pol);
 363
 364 #define BLKG_FLAG_FNS(name)                                             \
 365 static inline void blkio_mark_blkg_##name(                              \
 366                 struct blkio_group_stats *stats)                        \
 367 {                                                                       \
 368         stats->flags |= (1 << BLKG_##name);                             \
 369 }                                                                       \
 370 static inline void blkio_clear_blkg_##name(                             \
 371                 struct blkio_group_stats *stats)                        \
 372 {                                                                       \
 373         stats->flags &= ~(1 << BLKG_##name);                            \
 374 }                                                                       \
 375 static inline int blkio_blkg_##name(struct blkio_group_stats *stats)    \
 376 {                                                                       \
 377         return (stats->flags & (1 << BLKG_##name)) != 0;                \
 378 }                                                                       \
 379
 380 BLKG_FLAG_FNS(waiting)
 381 BLKG_FLAG_FNS(idling)
 382 BLKG_FLAG_FNS(empty)
 383 #undef BLKG_FLAG_FNS
 384 #else
 385 static inline void blkiocg_update_avg_queue_size_stats(struct blkio_group *blkg,
 386                         struct blkio_policy_type *pol) { }
 387 static inline void blkiocg_update_dequeue_stats(struct blkio_group *blkg,
 388                         struct blkio_policy_type *pol, unsigned long dequeue) { }
 389 static inline void blkiocg_update_set_idle_time_stats(struct blkio_group *blkg,
 390                         struct blkio_policy_type *pol) { }
 391 static inline void blkiocg_update_idle_time_stats(struct blkio_group *blkg,
 392                         struct blkio_policy_type *pol) { }
 393 static inline void blkiocg_set_start_empty_time(struct blkio_group *blkg,
 394                         struct blkio_policy_type *pol) { }
 395 #endif
 396
 397 #ifdef CONFIG_BLK_CGROUP
 398 extern struct blkio_cgroup blkio_root_cgroup;
 399 extern struct blkio_cgroup *cgroup_to_blkio_cgroup(struct cgroup *cgroup);
 400 extern struct blkio_cgroup *bio_blkio_cgroup(struct bio *bio);
 401 extern struct blkio_group *blkg_lookup(struct blkio_cgroup *blkcg,
 402                                        struct request_queue *q);
 403 struct blkio_group *blkg_lookup_create(struct blkio_cgroup *blkcg,
 404                                        struct request_queue *q,
 405                                        enum blkio_policy_id plid,
 406                                        bool for_root);
 407 void blkiocg_update_timeslice_used(struct blkio_group *blkg,
 408                                    struct blkio_policy_type *pol,
 409                                    unsigned long time,
 410                                    unsigned long unaccounted_time);
 411 void blkiocg_update_dispatch_stats(struct blkio_group *blkg,
 412                                    struct blkio_policy_type *pol,
 413                                    uint64_t bytes, bool direction, bool sync);
 414 void blkiocg_update_completion_stats(struct blkio_group *blkg,
 415                                      struct blkio_policy_type *pol,
 416                                      uint64_t start_time,
 417                                      uint64_t io_start_time, bool direction,
 418                                      bool sync);
 419 void blkiocg_update_io_merged_stats(struct blkio_group *blkg,
 420                                     struct blkio_policy_type *pol,
 421                                     bool direction, bool sync);
 422 void blkiocg_update_io_add_stats(struct blkio_group *blkg,
 423                                  struct blkio_policy_type *pol,
 424                                  struct blkio_group *curr_blkg, bool direction,
 425                                  bool sync);
 426 void blkiocg_update_io_remove_stats(struct blkio_group *blkg,
 427                                     struct blkio_policy_type *pol,
 428                                     bool direction, bool sync);
 429 #else
 430 struct cgroup;
 431 static inline struct blkio_cgroup *
 432 cgroup_to_blkio_cgroup(struct cgroup *cgroup) { return NULL; }
 433 static inline struct blkio_cgroup *
 434 bio_blkio_cgroup(struct bio *bio) { return NULL; }
 435
 436 static inline struct blkio_group *blkg_lookup(struct blkio_cgroup *blkcg,
 437                                               void *key) { return NULL; }
 438 static inline void blkiocg_update_timeslice_used(struct blkio_group *blkg,
 439                         struct blkio_policy_type *pol, unsigned long time,
 440                         unsigned long unaccounted_time) { }
 441 static inline void blkiocg_update_dispatch_stats(struct blkio_group *blkg,
 442                         struct blkio_policy_type *pol, uint64_t bytes,
 443                         bool direction, bool sync) { }
 444 static inline void blkiocg_update_completion_stats(struct blkio_group *blkg,
 445                         struct blkio_policy_type *pol, uint64_t start_time,
 446                         uint64_t io_start_time, bool direction, bool sync) { }
 447 static inline void blkiocg_update_io_merged_stats(struct blkio_group *blkg,
 448                         struct blkio_policy_type *pol, bool direction,
 449                         bool sync) { }
 450 static inline void blkiocg_update_io_add_stats(struct blkio_group *blkg,
 451                         struct blkio_policy_type *pol,
 452                         struct blkio_group *curr_blkg, bool direction,
 453                         bool sync) { }
 454 static inline void blkiocg_update_io_remove_stats(struct blkio_group *blkg,
 455                         struct blkio_policy_type *pol, bool direction,
 456                         bool sync) { }
 457 #endif
 458 #endif /* _BLK_CGROUP_H */