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
 121 struct blkio_group_stats {
 122         /* total disk time and nr sectors dispatched by this group */
 123         uint64_t time;
 124         uint64_t stat_arr[BLKIO_STAT_ARR_NR][BLKIO_STAT_TOTAL];
 125 #ifdef CONFIG_DEBUG_BLK_CGROUP
 126         /* Time not charged to this cgroup */
 127         uint64_t unaccounted_time;
 128
 129         /* Sum of number of IOs queued across all samples */
 130         uint64_t avg_queue_size_sum;
 131         /* Count of samples taken for average */
 132         uint64_t avg_queue_size_samples;
 133         /* How many times this group has been removed from service tree */
 134         unsigned long dequeue;
 135
 136         /* Total time spent waiting for it to be assigned a timeslice. */
 137         uint64_t group_wait_time;
 138
 139         /* Time spent idling for this blkio_group */
 140         uint64_t idle_time;
 141         /*
 142          * Total time when we have requests queued and do not contain the
 143          * current active queue.
 144          */
 145         uint64_t empty_time;
 146
 147         /* fields after this shouldn't be cleared on stat reset */
 148         uint64_t start_group_wait_time;
 149         uint64_t start_idle_time;
 150         uint64_t start_empty_time;
 151         uint16_t flags;
 152 #endif
 153 };
 154
 155 #ifdef CONFIG_DEBUG_BLK_CGROUP
 156 #define BLKG_STATS_DEBUG_CLEAR_START    \
 157         offsetof(struct blkio_group_stats, unaccounted_time)
 158 #define BLKG_STATS_DEBUG_CLEAR_SIZE     \
 159         (offsetof(struct blkio_group_stats, start_group_wait_time) - \
 160          BLKG_STATS_DEBUG_CLEAR_START)
 161 #endif
 162
 163 /* Per cpu blkio group stats */
 164 struct blkio_group_stats_cpu {
 165         uint64_t sectors;
 166         uint64_t stat_arr_cpu[BLKIO_STAT_CPU_NR][BLKIO_STAT_TOTAL];
 167         struct u64_stats_sync syncp;
 168 };
 169
 170 struct blkio_group_conf {
 171         unsigned int weight;
 172         unsigned int iops[2];
 173         u64 bps[2];
 174 };
 175
 176 /* per-blkg per-policy data */
 177 struct blkg_policy_data {
 178         /* the blkg this per-policy data belongs to */
 179         struct blkio_group *blkg;
 180
 181         /* Configuration */
 182         struct blkio_group_conf conf;
 183
 184         struct blkio_group_stats stats;
 185         /* Per cpu stats pointer */
 186         struct blkio_group_stats_cpu __percpu *stats_cpu;
 187
 188         /* pol->pdata_size bytes of private data used by policy impl */
 189         char pdata[] __aligned(__alignof__(unsigned long long));
 190 };
 191
 192 struct blkio_group {
 193         /* Pointer to the associated request_queue */
 194         struct request_queue *q;
 195         struct list_head q_node;
 196         struct hlist_node blkcg_node;
 197         struct blkio_cgroup *blkcg;
 198         /* Store cgroup path */
 199         char path[128];
 200         /* reference count */
 201         int refcnt;
 202
 203         /* Need to serialize the stats in the case of reset/update */
 204         spinlock_t stats_lock;
 205         struct blkg_policy_data *pd[BLKIO_NR_POLICIES];
 206
 207         /* List of blkg waiting for per cpu stats memory to be allocated */
 208         struct list_head alloc_node;
 209         struct rcu_head rcu_head;
 210 };
 211
 212 typedef void (blkio_init_group_fn)(struct blkio_group *blkg);
 213 typedef void (blkio_update_group_weight_fn)(struct request_queue *q,
 214                         struct blkio_group *blkg, unsigned int weight);
 215 typedef void (blkio_update_group_read_bps_fn)(struct request_queue *q,
 216                         struct blkio_group *blkg, u64 read_bps);
 217 typedef void (blkio_update_group_write_bps_fn)(struct request_queue *q,
 218                         struct blkio_group *blkg, u64 write_bps);
 219 typedef void (blkio_update_group_read_iops_fn)(struct request_queue *q,
 220                         struct blkio_group *blkg, unsigned int read_iops);
 221 typedef void (blkio_update_group_write_iops_fn)(struct request_queue *q,
 222                         struct blkio_group *blkg, unsigned int write_iops);
 223
 224 struct blkio_policy_ops {
 225         blkio_init_group_fn *blkio_init_group_fn;
 226         blkio_update_group_weight_fn *blkio_update_group_weight_fn;
 227         blkio_update_group_read_bps_fn *blkio_update_group_read_bps_fn;
 228         blkio_update_group_write_bps_fn *blkio_update_group_write_bps_fn;
 229         blkio_update_group_read_iops_fn *blkio_update_group_read_iops_fn;
 230         blkio_update_group_write_iops_fn *blkio_update_group_write_iops_fn;
 231 };
 232
 233 struct blkio_policy_type {
 234         struct list_head list;
 235         struct blkio_policy_ops ops;
 236         enum blkio_policy_id plid;
 237         size_t pdata_size;              /* policy specific private data size */
 238 };
 239
 240 extern int blkcg_init_queue(struct request_queue *q);
 241 extern void blkcg_drain_queue(struct request_queue *q);
 242 extern void blkcg_exit_queue(struct request_queue *q);
 243
 244 /* Blkio controller policy registration */
 245 extern void blkio_policy_register(struct blkio_policy_type *);
 246 extern void blkio_policy_unregister(struct blkio_policy_type *);
 247 extern void blkg_destroy_all(struct request_queue *q, bool destroy_root);
 248 extern void update_root_blkg_pd(struct request_queue *q,
 249                                 enum blkio_policy_id plid);
 250
 251 /**
 252  * blkg_to_pdata - get policy private data
 253  * @blkg: blkg of interest
 254  * @pol: policy of interest
 255  *
 256  * Return pointer to private data associated with the @blkg-@pol pair.
 257  */
 258 static inline void *blkg_to_pdata(struct blkio_group *blkg,
 259                               struct blkio_policy_type *pol)
 260 {
 261         return blkg ? blkg->pd[pol->plid]->pdata : NULL;
 262 }
 263
 264 /**
 265  * pdata_to_blkg - get blkg associated with policy private data
 266  * @pdata: policy private data of interest
 267  * @pol: policy @pdata is for
 268  *
 269  * @pdata is policy private data for @pol.  Determine the blkg it's
 270  * associated with.
 271  */
 272 static inline struct blkio_group *pdata_to_blkg(void *pdata,
 273                                                 struct blkio_policy_type *pol)
 274 {
 275         if (pdata) {
 276                 struct blkg_policy_data *pd =
 277                         container_of(pdata, struct blkg_policy_data, pdata);
 278                 return pd->blkg;
 279         }
 280         return NULL;
 281 }
 282
 283 static inline char *blkg_path(struct blkio_group *blkg)
 284 {
 285         return blkg->path;
 286 }
 287
 288 /**
 289  * blkg_get - get a blkg reference
 290  * @blkg: blkg to get
 291  *
 292  * The caller should be holding queue_lock and an existing reference.
 293  */
 294 static inline void blkg_get(struct blkio_group *blkg)
 295 {
 296         lockdep_assert_held(blkg->q->queue_lock);
 297         WARN_ON_ONCE(!blkg->refcnt);
 298         blkg->refcnt++;
 299 }
 300
 301 void __blkg_release(struct blkio_group *blkg);
 302
 303 /**
 304  * blkg_put - put a blkg reference
 305  * @blkg: blkg to put
 306  *
 307  * The caller should be holding queue_lock.
 308  */
 309 static inline void blkg_put(struct blkio_group *blkg)
 310 {
 311         lockdep_assert_held(blkg->q->queue_lock);
 312         WARN_ON_ONCE(blkg->refcnt <= 0);
 313         if (!--blkg->refcnt)
 314                 __blkg_release(blkg);
 315 }
 316
 317 #else
 318
 319 struct blkio_group {
 320 };
 321
 322 struct blkio_policy_type {
 323 };
 324
 325 static inline int blkcg_init_queue(struct request_queue *q) { return 0; }
 326 static inline void blkcg_drain_queue(struct request_queue *q) { }
 327 static inline void blkcg_exit_queue(struct request_queue *q) { }
 328 static inline void blkio_policy_register(struct blkio_policy_type *blkiop) { }
 329 static inline void blkio_policy_unregister(struct blkio_policy_type *blkiop) { }
 330 static inline void blkg_destroy_all(struct request_queue *q,
 331                                     bool destory_root) { }
 332 static inline void update_root_blkg_pd(struct request_queue *q,
 333                                        enum blkio_policy_id plid) { }
 334
 335 static inline void *blkg_to_pdata(struct blkio_group *blkg,
 336                                 struct blkio_policy_type *pol) { return NULL; }
 337 static inline struct blkio_group *pdata_to_blkg(void *pdata,
 338                                 struct blkio_policy_type *pol) { return NULL; }
 339 static inline char *blkg_path(struct blkio_group *blkg) { return NULL; }
 340 static inline void blkg_get(struct blkio_group *blkg) { }
 341 static inline void blkg_put(struct blkio_group *blkg) { }
 342
 343 #endif
 344
 345 #define BLKIO_WEIGHT_MIN        10
 346 #define BLKIO_WEIGHT_MAX        1000
 347 #define BLKIO_WEIGHT_DEFAULT    500
 348
 349 #ifdef CONFIG_DEBUG_BLK_CGROUP
 350 void blkiocg_update_avg_queue_size_stats(struct blkio_group *blkg,
 351                                          struct blkio_policy_type *pol);
 352 void blkiocg_update_dequeue_stats(struct blkio_group *blkg,
 353                                   struct blkio_policy_type *pol,
 354                                   unsigned long dequeue);
 355 void blkiocg_update_set_idle_time_stats(struct blkio_group *blkg,
 356                                         struct blkio_policy_type *pol);
 357 void blkiocg_update_idle_time_stats(struct blkio_group *blkg,
 358                                     struct blkio_policy_type *pol);
 359 void blkiocg_set_start_empty_time(struct blkio_group *blkg,
 360                                   struct blkio_policy_type *pol);
 361
 362 #define BLKG_FLAG_FNS(name)                                             \
 363 static inline void blkio_mark_blkg_##name(                              \
 364                 struct blkio_group_stats *stats)                        \
 365 {                                                                       \
 366         stats->flags |= (1 << BLKG_##name);                             \
 367 }                                                                       \
 368 static inline void blkio_clear_blkg_##name(                             \
 369                 struct blkio_group_stats *stats)                        \
 370 {                                                                       \
 371         stats->flags &= ~(1 << BLKG_##name);                            \
 372 }                                                                       \
 373 static inline int blkio_blkg_##name(struct blkio_group_stats *stats)    \
 374 {                                                                       \
 375         return (stats->flags & (1 << BLKG_##name)) != 0;                \
 376 }                                                                       \
 377
 378 BLKG_FLAG_FNS(waiting)
 379 BLKG_FLAG_FNS(idling)
 380 BLKG_FLAG_FNS(empty)
 381 #undef BLKG_FLAG_FNS
 382 #else
 383 static inline void blkiocg_update_avg_queue_size_stats(struct blkio_group *blkg,
 384                         struct blkio_policy_type *pol) { }
 385 static inline void blkiocg_update_dequeue_stats(struct blkio_group *blkg,
 386                         struct blkio_policy_type *pol, unsigned long dequeue) { }
 387 static inline void blkiocg_update_set_idle_time_stats(struct blkio_group *blkg,
 388                         struct blkio_policy_type *pol) { }
 389 static inline void blkiocg_update_idle_time_stats(struct blkio_group *blkg,
 390                         struct blkio_policy_type *pol) { }
 391 static inline void blkiocg_set_start_empty_time(struct blkio_group *blkg,
 392                         struct blkio_policy_type *pol) { }
 393 #endif
 394
 395 #ifdef CONFIG_BLK_CGROUP
 396 extern struct blkio_cgroup blkio_root_cgroup;
 397 extern struct blkio_cgroup *cgroup_to_blkio_cgroup(struct cgroup *cgroup);
 398 extern struct blkio_cgroup *bio_blkio_cgroup(struct bio *bio);
 399 extern struct blkio_group *blkg_lookup(struct blkio_cgroup *blkcg,
 400                                        struct request_queue *q);
 401 struct blkio_group *blkg_lookup_create(struct blkio_cgroup *blkcg,
 402                                        struct request_queue *q,
 403                                        enum blkio_policy_id plid,
 404                                        bool for_root);
 405 void blkiocg_update_timeslice_used(struct blkio_group *blkg,
 406                                    struct blkio_policy_type *pol,
 407                                    unsigned long time,
 408                                    unsigned long unaccounted_time);
 409 void blkiocg_update_dispatch_stats(struct blkio_group *blkg,
 410                                    struct blkio_policy_type *pol,
 411                                    uint64_t bytes, bool direction, bool sync);
 412 void blkiocg_update_completion_stats(struct blkio_group *blkg,
 413                                      struct blkio_policy_type *pol,
 414                                      uint64_t start_time,
 415                                      uint64_t io_start_time, bool direction,
 416                                      bool sync);
 417 void blkiocg_update_io_merged_stats(struct blkio_group *blkg,
 418                                     struct blkio_policy_type *pol,
 419                                     bool direction, bool sync);
 420 void blkiocg_update_io_add_stats(struct blkio_group *blkg,
 421                                  struct blkio_policy_type *pol,
 422                                  struct blkio_group *curr_blkg, bool direction,
 423                                  bool sync);
 424 void blkiocg_update_io_remove_stats(struct blkio_group *blkg,
 425                                     struct blkio_policy_type *pol,
 426                                     bool direction, bool sync);
 427 #else
 428 struct cgroup;
 429 static inline struct blkio_cgroup *
 430 cgroup_to_blkio_cgroup(struct cgroup *cgroup) { return NULL; }
 431 static inline struct blkio_cgroup *
 432 bio_blkio_cgroup(struct bio *bio) { return NULL; }
 433
 434 static inline struct blkio_group *blkg_lookup(struct blkio_cgroup *blkcg,
 435                                               void *key) { return NULL; }
 436 static inline void blkiocg_update_timeslice_used(struct blkio_group *blkg,
 437                         struct blkio_policy_type *pol, unsigned long time,
 438                         unsigned long unaccounted_time) { }
 439 static inline void blkiocg_update_dispatch_stats(struct blkio_group *blkg,
 440                         struct blkio_policy_type *pol, uint64_t bytes,
 441                         bool direction, bool sync) { }
 442 static inline void blkiocg_update_completion_stats(struct blkio_group *blkg,
 443                         struct blkio_policy_type *pol, uint64_t start_time,
 444                         uint64_t io_start_time, bool direction, bool sync) { }
 445 static inline void blkiocg_update_io_merged_stats(struct blkio_group *blkg,
 446                         struct blkio_policy_type *pol, bool direction,
 447                         bool sync) { }
 448 static inline void blkiocg_update_io_add_stats(struct blkio_group *blkg,
 449                         struct blkio_policy_type *pol,
 450                         struct blkio_group *curr_blkg, bool direction,
 451                         bool sync) { }
 452 static inline void blkiocg_update_io_remove_stats(struct blkio_group *blkg,
 453                         struct blkio_policy_type *pol, bool direction,
 454                         bool sync) { }
 455 #endif
 456 #endif /* _BLK_CGROUP_H */