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