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
  20 /* Max limits for throttle policy */
  21 #define THROTL_IOPS_MAX         UINT_MAX
  22
  23 /* CFQ specific, out here for blkcg->cfq_weight */
  24 #define CFQ_WEIGHT_MIN          10
  25 #define CFQ_WEIGHT_MAX          1000
  26 #define CFQ_WEIGHT_DEFAULT      500
  27
  28 #ifdef CONFIG_BLK_CGROUP
  29
  30 enum blkg_rwstat_type {
  31         BLKG_RWSTAT_READ,
  32         BLKG_RWSTAT_WRITE,
  33         BLKG_RWSTAT_SYNC,
  34         BLKG_RWSTAT_ASYNC,
  35
  36         BLKG_RWSTAT_NR,
  37         BLKG_RWSTAT_TOTAL = BLKG_RWSTAT_NR,
  38 };
  39
  40 struct blkcg {
  41         struct cgroup_subsys_state      css;
  42         spinlock_t                      lock;
  43         struct hlist_head               blkg_list;
  44
  45         /* for policies to test whether associated blkcg has changed */
  46         uint64_t                        id;
  47
  48         /* TODO: per-policy storage in blkcg */
  49         unsigned int                    cfq_weight;     /* belongs to cfq */
  50 };
  51
  52 struct blkg_stat {
  53         struct u64_stats_sync           syncp;
  54         uint64_t                        cnt;
  55 };
  56
  57 struct blkg_rwstat {
  58         struct u64_stats_sync           syncp;
  59         uint64_t                        cnt[BLKG_RWSTAT_NR];
  60 };
  61
  62 /*
  63  * A blkcg_gq (blkg) is association between a block cgroup (blkcg) and a
  64  * request_queue (q).  This is used by blkcg policies which need to track
  65  * information per blkcg - q pair.
  66  *
  67  * There can be multiple active blkcg policies and each has its private
  68  * data on each blkg, the size of which is determined by
  69  * blkcg_policy->pd_size.  blkcg core allocates and frees such areas
  70  * together with blkg and invokes pd_init/exit_fn() methods.
  71  *
  72  * Such private data must embed struct blkg_policy_data (pd) at the
  73  * beginning and pd_size can't be smaller than pd.
  74  */
  75 struct blkg_policy_data {
  76         /* the blkg this per-policy data belongs to */
  77         struct blkcg_gq                 *blkg;
  78
  79         /* used during policy activation */
  80         struct list_head                alloc_node;
  81 };
  82
  83 /* association between a blk cgroup and a request queue */
  84 struct blkcg_gq {
  85         /* Pointer to the associated request_queue */
  86         struct request_queue            *q;
  87         struct list_head                q_node;
  88         struct hlist_node               blkcg_node;
  89         struct blkcg                    *blkcg;
  90         /* reference count */
  91         int                             refcnt;
  92
  93         struct blkg_policy_data         *pd[BLKCG_MAX_POLS];
  94
  95         struct rcu_head                 rcu_head;
  96 };
  97
  98 typedef void (blkcg_pol_init_pd_fn)(struct blkcg_gq *blkg);
  99 typedef void (blkcg_pol_exit_pd_fn)(struct blkcg_gq *blkg);
 100 typedef void (blkcg_pol_reset_pd_stats_fn)(struct blkcg_gq *blkg);
 101
 102 struct blkcg_policy {
 103         int                             plid;
 104         /* policy specific private data size */
 105         size_t                          pd_size;
 106         /* cgroup files for the policy */
 107         struct cftype                   *cftypes;
 108
 109         /* operations */
 110         blkcg_pol_init_pd_fn            *pd_init_fn;
 111         blkcg_pol_exit_pd_fn            *pd_exit_fn;
 112         blkcg_pol_reset_pd_stats_fn     *pd_reset_stats_fn;
 113 };
 114
 115 extern struct blkcg blkcg_root;
 116
 117 struct blkcg *cgroup_to_blkcg(struct cgroup *cgroup);
 118 struct blkcg *bio_blkcg(struct bio *bio);
 119 struct blkcg_gq *blkg_lookup(struct blkcg *blkcg, struct request_queue *q);
 120 struct blkcg_gq *blkg_lookup_create(struct blkcg *blkcg,
 121                                     struct request_queue *q);
 122 int blkcg_init_queue(struct request_queue *q);
 123 void blkcg_drain_queue(struct request_queue *q);
 124 void blkcg_exit_queue(struct request_queue *q);
 125
 126 /* Blkio controller policy registration */
 127 int blkcg_policy_register(struct blkcg_policy *pol);
 128 void blkcg_policy_unregister(struct blkcg_policy *pol);
 129 int blkcg_activate_policy(struct request_queue *q,
 130                           const struct blkcg_policy *pol);
 131 void blkcg_deactivate_policy(struct request_queue *q,
 132                              const struct blkcg_policy *pol);
 133
 134 void blkcg_print_blkgs(struct seq_file *sf, struct blkcg *blkcg,
 135                        u64 (*prfill)(struct seq_file *,
 136                                      struct blkg_policy_data *, int),
 137                        const struct blkcg_policy *pol, int data,
 138                        bool show_total);
 139 u64 __blkg_prfill_u64(struct seq_file *sf, struct blkg_policy_data *pd, u64 v);
 140 u64 __blkg_prfill_rwstat(struct seq_file *sf, struct blkg_policy_data *pd,
 141                          const struct blkg_rwstat *rwstat);
 142 u64 blkg_prfill_stat(struct seq_file *sf, struct blkg_policy_data *pd, int off);
 143 u64 blkg_prfill_rwstat(struct seq_file *sf, struct blkg_policy_data *pd,
 144                        int off);
 145
 146 struct blkg_conf_ctx {
 147         struct gendisk                  *disk;
 148         struct blkcg_gq                 *blkg;
 149         u64                             v;
 150 };
 151
 152 int blkg_conf_prep(struct blkcg *blkcg, const struct blkcg_policy *pol,
 153                    const char *input, struct blkg_conf_ctx *ctx);
 154 void blkg_conf_finish(struct blkg_conf_ctx *ctx);
 155
 156
 157 /**
 158  * blkg_to_pdata - get policy private data
 159  * @blkg: blkg of interest
 160  * @pol: policy of interest
 161  *
 162  * Return pointer to private data associated with the @blkg-@pol pair.
 163  */
 164 static inline struct blkg_policy_data *blkg_to_pd(struct blkcg_gq *blkg,
 165                                                   struct blkcg_policy *pol)
 166 {
 167         return blkg ? blkg->pd[pol->plid] : NULL;
 168 }
 169
 170 /**
 171  * pdata_to_blkg - get blkg associated with policy private data
 172  * @pd: policy private data of interest
 173  *
 174  * @pd is policy private data.  Determine the blkg it's associated with.
 175  */
 176 static inline struct blkcg_gq *pd_to_blkg(struct blkg_policy_data *pd)
 177 {
 178         return pd ? pd->blkg : NULL;
 179 }
 180
 181 /**
 182  * blkg_path - format cgroup path of blkg
 183  * @blkg: blkg of interest
 184  * @buf: target buffer
 185  * @buflen: target buffer length
 186  *
 187  * Format the path of the cgroup of @blkg into @buf.
 188  */
 189 static inline int blkg_path(struct blkcg_gq *blkg, char *buf, int buflen)
 190 {
 191         int ret;
 192
 193         rcu_read_lock();
 194         ret = cgroup_path(blkg->blkcg->css.cgroup, buf, buflen);
 195         rcu_read_unlock();
 196         if (ret)
 197                 strncpy(buf, "<unavailable>", buflen);
 198         return ret;
 199 }
 200
 201 /**
 202  * blkg_get - get a blkg reference
 203  * @blkg: blkg to get
 204  *
 205  * The caller should be holding queue_lock and an existing reference.
 206  */
 207 static inline void blkg_get(struct blkcg_gq *blkg)
 208 {
 209         lockdep_assert_held(blkg->q->queue_lock);
 210         WARN_ON_ONCE(!blkg->refcnt);
 211         blkg->refcnt++;
 212 }
 213
 214 void __blkg_release(struct blkcg_gq *blkg);
 215
 216 /**
 217  * blkg_put - put a blkg reference
 218  * @blkg: blkg to put
 219  *
 220  * The caller should be holding queue_lock.
 221  */
 222 static inline void blkg_put(struct blkcg_gq *blkg)
 223 {
 224         lockdep_assert_held(blkg->q->queue_lock);
 225         WARN_ON_ONCE(blkg->refcnt <= 0);
 226         if (!--blkg->refcnt)
 227                 __blkg_release(blkg);
 228 }
 229
 230 /**
 231  * blkg_stat_add - add a value to a blkg_stat
 232  * @stat: target blkg_stat
 233  * @val: value to add
 234  *
 235  * Add @val to @stat.  The caller is responsible for synchronizing calls to
 236  * this function.
 237  */
 238 static inline void blkg_stat_add(struct blkg_stat *stat, uint64_t val)
 239 {
 240         u64_stats_update_begin(&stat->syncp);
 241         stat->cnt += val;
 242         u64_stats_update_end(&stat->syncp);
 243 }
 244
 245 /**
 246  * blkg_stat_read - read the current value of a blkg_stat
 247  * @stat: blkg_stat to read
 248  *
 249  * Read the current value of @stat.  This function can be called without
 250  * synchroniztion and takes care of u64 atomicity.
 251  */
 252 static inline uint64_t blkg_stat_read(struct blkg_stat *stat)
 253 {
 254         unsigned int start;
 255         uint64_t v;
 256
 257         do {
 258                 start = u64_stats_fetch_begin(&stat->syncp);
 259                 v = stat->cnt;
 260         } while (u64_stats_fetch_retry(&stat->syncp, start));
 261
 262         return v;
 263 }
 264
 265 /**
 266  * blkg_stat_reset - reset a blkg_stat
 267  * @stat: blkg_stat to reset
 268  */
 269 static inline void blkg_stat_reset(struct blkg_stat *stat)
 270 {
 271         stat->cnt = 0;
 272 }
 273
 274 /**
 275  * blkg_rwstat_add - add a value to a blkg_rwstat
 276  * @rwstat: target blkg_rwstat
 277  * @rw: mask of REQ_{WRITE|SYNC}
 278  * @val: value to add
 279  *
 280  * Add @val to @rwstat.  The counters are chosen according to @rw.  The
 281  * caller is responsible for synchronizing calls to this function.
 282  */
 283 static inline void blkg_rwstat_add(struct blkg_rwstat *rwstat,
 284                                    int rw, uint64_t val)
 285 {
 286         u64_stats_update_begin(&rwstat->syncp);
 287
 288         if (rw & REQ_WRITE)
 289                 rwstat->cnt[BLKG_RWSTAT_WRITE] += val;
 290         else
 291                 rwstat->cnt[BLKG_RWSTAT_READ] += val;
 292         if (rw & REQ_SYNC)
 293                 rwstat->cnt[BLKG_RWSTAT_SYNC] += val;
 294         else
 295                 rwstat->cnt[BLKG_RWSTAT_ASYNC] += val;
 296
 297         u64_stats_update_end(&rwstat->syncp);
 298 }
 299
 300 /**
 301  * blkg_rwstat_read - read the current values of a blkg_rwstat
 302  * @rwstat: blkg_rwstat to read
 303  *
 304  * Read the current snapshot of @rwstat and return it as the return value.
 305  * This function can be called without synchronization and takes care of
 306  * u64 atomicity.
 307  */
 308 static inline struct blkg_rwstat blkg_rwstat_read(struct blkg_rwstat *rwstat)
 309 {
 310         unsigned int start;
 311         struct blkg_rwstat tmp;
 312
 313         do {
 314                 start = u64_stats_fetch_begin(&rwstat->syncp);
 315                 tmp = *rwstat;
 316         } while (u64_stats_fetch_retry(&rwstat->syncp, start));
 317
 318         return tmp;
 319 }
 320
 321 /**
 322  * blkg_rwstat_sum - read the total count of a blkg_rwstat
 323  * @rwstat: blkg_rwstat to read
 324  *
 325  * Return the total count of @rwstat regardless of the IO direction.  This
 326  * function can be called without synchronization and takes care of u64
 327  * atomicity.
 328  */
 329 static inline uint64_t blkg_rwstat_sum(struct blkg_rwstat *rwstat)
 330 {
 331         struct blkg_rwstat tmp = blkg_rwstat_read(rwstat);
 332
 333         return tmp.cnt[BLKG_RWSTAT_READ] + tmp.cnt[BLKG_RWSTAT_WRITE];
 334 }
 335
 336 /**
 337  * blkg_rwstat_reset - reset a blkg_rwstat
 338  * @rwstat: blkg_rwstat to reset
 339  */
 340 static inline void blkg_rwstat_reset(struct blkg_rwstat *rwstat)
 341 {
 342         memset(rwstat->cnt, 0, sizeof(rwstat->cnt));
 343 }
 344
 345 #else   /* CONFIG_BLK_CGROUP */
 346
 347 struct cgroup;
 348
 349 struct blkg_policy_data {
 350 };
 351
 352 struct blkcg_gq {
 353 };
 354
 355 struct blkcg_policy {
 356 };
 357
 358 static inline struct blkcg *cgroup_to_blkcg(struct cgroup *cgroup) { return NULL; }
 359 static inline struct blkcg *bio_blkcg(struct bio *bio) { return NULL; }
 360 static inline struct blkcg_gq *blkg_lookup(struct blkcg *blkcg, void *key) { return NULL; }
 361 static inline int blkcg_init_queue(struct request_queue *q) { return 0; }
 362 static inline void blkcg_drain_queue(struct request_queue *q) { }
 363 static inline void blkcg_exit_queue(struct request_queue *q) { }
 364 static inline int blkcg_policy_register(struct blkcg_policy *pol) { return 0; }
 365 static inline void blkcg_policy_unregister(struct blkcg_policy *pol) { }
 366 static inline int blkcg_activate_policy(struct request_queue *q,
 367                                         const struct blkcg_policy *pol) { return 0; }
 368 static inline void blkcg_deactivate_policy(struct request_queue *q,
 369                                            const struct blkcg_policy *pol) { }
 370
 371 static inline struct blkg_policy_data *blkg_to_pd(struct blkcg_gq *blkg,
 372                                                   struct blkcg_policy *pol) { return NULL; }
 373 static inline struct blkcg_gq *pd_to_blkg(struct blkg_policy_data *pd) { return NULL; }
 374 static inline char *blkg_path(struct blkcg_gq *blkg) { return NULL; }
 375 static inline void blkg_get(struct blkcg_gq *blkg) { }
 376 static inline void blkg_put(struct blkcg_gq *blkg) { }
 377
 378 #endif  /* CONFIG_BLK_CGROUP */
 379 #endif  /* _BLK_CGROUP_H */