block/blk-cgroup.c

   1 /*
   2  * Common Block IO controller cgroup interface
   3  *
   4  * Based on ideas and code from CFQ, CFS and BFQ:
   5  * Copyright (C) 2003 Jens Axboe <axboe@kernel.dk>
   6  *
   7  * Copyright (C) 2008 Fabio Checconi <fabio@gandalf.sssup.it>
   8  *                    Paolo Valente <paolo.valente@unimore.it>
   9  *
  10  * Copyright (C) 2009 Vivek Goyal <vgoyal@redhat.com>
  11  *                    Nauman Rafique <nauman@google.com>
  12  */
  13 #include <linux/ioprio.h>
  14 #include <linux/seq_file.h>
  15 #include <linux/kdev_t.h>
  16 #include <linux/module.h>
  17 #include <linux/err.h>
  18 #include <linux/blkdev.h>
  19 #include <linux/slab.h>
  20 #include "blk-cgroup.h"
  21 #include <linux/genhd.h>
  22
  23 #define MAX_KEY_LEN 100
  24
  25 static DEFINE_SPINLOCK(blkio_list_lock);
  26 static LIST_HEAD(blkio_list);
  27
  28 struct blkio_cgroup blkio_root_cgroup = { .weight = 2*BLKIO_WEIGHT_DEFAULT };
  29 EXPORT_SYMBOL_GPL(blkio_root_cgroup);
  30
  31 static struct cgroup_subsys_state *blkiocg_create(struct cgroup *);
  32 static int blkiocg_can_attach(struct cgroup *, struct cgroup_taskset *);
  33 static void blkiocg_attach(struct cgroup *, struct cgroup_taskset *);
  34 static void blkiocg_destroy(struct cgroup *);
  35 static int blkiocg_populate(struct cgroup_subsys *, struct cgroup *);
  36
  37 /* for encoding cft->private value on file */
  38 #define BLKIOFILE_PRIVATE(x, val)       (((x) << 16) | (val))
  39 /* What policy owns the file, proportional or throttle */
  40 #define BLKIOFILE_POLICY(val)           (((val) >> 16) & 0xffff)
  41 #define BLKIOFILE_ATTR(val)             ((val) & 0xffff)
  42
  43 struct cgroup_subsys blkio_subsys = {
  44         .name = "blkio",
  45         .create = blkiocg_create,
  46         .can_attach = blkiocg_can_attach,
  47         .attach = blkiocg_attach,
  48         .destroy = blkiocg_destroy,
  49         .populate = blkiocg_populate,
  50 #ifdef CONFIG_BLK_CGROUP
  51         /* note: blkio_subsys_id is otherwise defined in blk-cgroup.h */
  52         .subsys_id = blkio_subsys_id,
  53 #endif
  54         .use_id = 1,
  55         .module = THIS_MODULE,
  56 };
  57 EXPORT_SYMBOL_GPL(blkio_subsys);
  58
  59 static inline void blkio_policy_insert_node(struct blkio_cgroup *blkcg,
  60                                             struct blkio_policy_node *pn)
  61 {
  62         list_add(&pn->node, &blkcg->policy_list);
  63 }
  64
  65 static inline bool cftype_blkg_same_policy(struct cftype *cft,
  66                         struct blkio_group *blkg)
  67 {
  68         enum blkio_policy_id plid = BLKIOFILE_POLICY(cft->private);
  69
  70         if (blkg->plid == plid)
  71                 return 1;
  72
  73         return 0;
  74 }
  75
  76 /* Determines if policy node matches cgroup file being accessed */
  77 static inline bool pn_matches_cftype(struct cftype *cft,
  78                         struct blkio_policy_node *pn)
  79 {
  80         enum blkio_policy_id plid = BLKIOFILE_POLICY(cft->private);
  81         int fileid = BLKIOFILE_ATTR(cft->private);
  82
  83         return (plid == pn->plid && fileid == pn->fileid);
  84 }
  85
  86 /* Must be called with blkcg->lock held */
  87 static inline void blkio_policy_delete_node(struct blkio_policy_node *pn)
  88 {
  89         list_del(&pn->node);
  90 }
  91
  92 /* Must be called with blkcg->lock held */
  93 static struct blkio_policy_node *
  94 blkio_policy_search_node(const struct blkio_cgroup *blkcg, dev_t dev,
  95                 enum blkio_policy_id plid, int fileid)
  96 {
  97         struct blkio_policy_node *pn;
  98
  99         list_for_each_entry(pn, &blkcg->policy_list, node) {
 100                 if (pn->dev == dev && pn->plid == plid && pn->fileid == fileid)
 101                         return pn;
 102         }
 103
 104         return NULL;
 105 }
 106
 107 struct blkio_cgroup *cgroup_to_blkio_cgroup(struct cgroup *cgroup)
 108 {
 109         return container_of(cgroup_subsys_state(cgroup, blkio_subsys_id),
 110                             struct blkio_cgroup, css);
 111 }
 112 EXPORT_SYMBOL_GPL(cgroup_to_blkio_cgroup);
 113
 114 struct blkio_cgroup *task_blkio_cgroup(struct task_struct *tsk)
 115 {
 116         return container_of(task_subsys_state(tsk, blkio_subsys_id),
 117                             struct blkio_cgroup, css);
 118 }
 119 EXPORT_SYMBOL_GPL(task_blkio_cgroup);
 120
 121 static inline void
 122 blkio_update_group_weight(struct blkio_group *blkg, unsigned int weight)
 123 {
 124         struct blkio_policy_type *blkiop;
 125
 126         list_for_each_entry(blkiop, &blkio_list, list) {
 127                 /* If this policy does not own the blkg, do not send updates */
 128                 if (blkiop->plid != blkg->plid)
 129                         continue;
 130                 if (blkiop->ops.blkio_update_group_weight_fn)
 131                         blkiop->ops.blkio_update_group_weight_fn(blkg->key,
 132                                                         blkg, weight);
 133         }
 134 }
 135
 136 static inline void blkio_update_group_bps(struct blkio_group *blkg, u64 bps,
 137                                 int fileid)
 138 {
 139         struct blkio_policy_type *blkiop;
 140
 141         list_for_each_entry(blkiop, &blkio_list, list) {
 142
 143                 /* If this policy does not own the blkg, do not send updates */
 144                 if (blkiop->plid != blkg->plid)
 145                         continue;
 146
 147                 if (fileid == BLKIO_THROTL_read_bps_device
 148                     && blkiop->ops.blkio_update_group_read_bps_fn)
 149                         blkiop->ops.blkio_update_group_read_bps_fn(blkg->key,
 150                                                                 blkg, bps);
 151
 152                 if (fileid == BLKIO_THROTL_write_bps_device
 153                     && blkiop->ops.blkio_update_group_write_bps_fn)
 154                         blkiop->ops.blkio_update_group_write_bps_fn(blkg->key,
 155                                                                 blkg, bps);
 156         }
 157 }
 158
 159 static inline void blkio_update_group_iops(struct blkio_group *blkg,
 160                         unsigned int iops, int fileid)
 161 {
 162         struct blkio_policy_type *blkiop;
 163
 164         list_for_each_entry(blkiop, &blkio_list, list) {
 165
 166                 /* If this policy does not own the blkg, do not send updates */
 167                 if (blkiop->plid != blkg->plid)
 168                         continue;
 169
 170                 if (fileid == BLKIO_THROTL_read_iops_device
 171                     && blkiop->ops.blkio_update_group_read_iops_fn)
 172                         blkiop->ops.blkio_update_group_read_iops_fn(blkg->key,
 173                                                                 blkg, iops);
 174
 175                 if (fileid == BLKIO_THROTL_write_iops_device
 176                     && blkiop->ops.blkio_update_group_write_iops_fn)
 177                         blkiop->ops.blkio_update_group_write_iops_fn(blkg->key,
 178                                                                 blkg,iops);
 179         }
 180 }
 181
 182 /*
 183  * Add to the appropriate stat variable depending on the request type.
 184  * This should be called with the blkg->stats_lock held.
 185  */
 186 static void blkio_add_stat(uint64_t *stat, uint64_t add, bool direction,
 187                                 bool sync)
 188 {
 189         if (direction)
 190                 stat[BLKIO_STAT_WRITE] += add;
 191         else
 192                 stat[BLKIO_STAT_READ] += add;
 193         if (sync)
 194                 stat[BLKIO_STAT_SYNC] += add;
 195         else
 196                 stat[BLKIO_STAT_ASYNC] += add;
 197 }
 198
 199 /*
 200  * Decrements the appropriate stat variable if non-zero depending on the
 201  * request type. Panics on value being zero.
 202  * This should be called with the blkg->stats_lock held.
 203  */
 204 static void blkio_check_and_dec_stat(uint64_t *stat, bool direction, bool sync)
 205 {
 206         if (direction) {
 207                 BUG_ON(stat[BLKIO_STAT_WRITE] == 0);
 208                 stat[BLKIO_STAT_WRITE]--;
 209         } else {
 210                 BUG_ON(stat[BLKIO_STAT_READ] == 0);
 211                 stat[BLKIO_STAT_READ]--;
 212         }
 213         if (sync) {
 214                 BUG_ON(stat[BLKIO_STAT_SYNC] == 0);
 215                 stat[BLKIO_STAT_SYNC]--;
 216         } else {
 217                 BUG_ON(stat[BLKIO_STAT_ASYNC] == 0);
 218                 stat[BLKIO_STAT_ASYNC]--;
 219         }
 220 }
 221
 222 #ifdef CONFIG_DEBUG_BLK_CGROUP
 223 /* This should be called with the blkg->stats_lock held. */
 224 static void blkio_set_start_group_wait_time(struct blkio_group *blkg,
 225                                                 struct blkio_group *curr_blkg)
 226 {
 227         if (blkio_blkg_waiting(&blkg->stats))
 228                 return;
 229         if (blkg == curr_blkg)
 230                 return;
 231         blkg->stats.start_group_wait_time = sched_clock();
 232         blkio_mark_blkg_waiting(&blkg->stats);
 233 }
 234
 235 /* This should be called with the blkg->stats_lock held. */
 236 static void blkio_update_group_wait_time(struct blkio_group_stats *stats)
 237 {
 238         unsigned long long now;
 239
 240         if (!blkio_blkg_waiting(stats))
 241                 return;
 242
 243         now = sched_clock();
 244         if (time_after64(now, stats->start_group_wait_time))
 245                 stats->group_wait_time += now - stats->start_group_wait_time;
 246         blkio_clear_blkg_waiting(stats);
 247 }
 248
 249 /* This should be called with the blkg->stats_lock held. */
 250 static void blkio_end_empty_time(struct blkio_group_stats *stats)
 251 {
 252         unsigned long long now;
 253
 254         if (!blkio_blkg_empty(stats))
 255                 return;
 256
 257         now = sched_clock();
 258         if (time_after64(now, stats->start_empty_time))
 259                 stats->empty_time += now - stats->start_empty_time;
 260         blkio_clear_blkg_empty(stats);
 261 }
 262
 263 void blkiocg_update_set_idle_time_stats(struct blkio_group *blkg)
 264 {
 265         unsigned long flags;
 266
 267         spin_lock_irqsave(&blkg->stats_lock, flags);
 268         BUG_ON(blkio_blkg_idling(&blkg->stats));
 269         blkg->stats.start_idle_time = sched_clock();
 270         blkio_mark_blkg_idling(&blkg->stats);
 271         spin_unlock_irqrestore(&blkg->stats_lock, flags);
 272 }
 273 EXPORT_SYMBOL_GPL(blkiocg_update_set_idle_time_stats);
 274
 275 void blkiocg_update_idle_time_stats(struct blkio_group *blkg)
 276 {
 277         unsigned long flags;
 278         unsigned long long now;
 279         struct blkio_group_stats *stats;
 280
 281         spin_lock_irqsave(&blkg->stats_lock, flags);
 282         stats = &blkg->stats;
 283         if (blkio_blkg_idling(stats)) {
 284                 now = sched_clock();
 285                 if (time_after64(now, stats->start_idle_time))
 286                         stats->idle_time += now - stats->start_idle_time;
 287                 blkio_clear_blkg_idling(stats);
 288         }
 289         spin_unlock_irqrestore(&blkg->stats_lock, flags);
 290 }
 291 EXPORT_SYMBOL_GPL(blkiocg_update_idle_time_stats);
 292
 293 void blkiocg_update_avg_queue_size_stats(struct blkio_group *blkg)
 294 {
 295         unsigned long flags;
 296         struct blkio_group_stats *stats;
 297
 298         spin_lock_irqsave(&blkg->stats_lock, flags);
 299         stats = &blkg->stats;
 300         stats->avg_queue_size_sum +=
 301                         stats->stat_arr[BLKIO_STAT_QUEUED][BLKIO_STAT_READ] +
 302                         stats->stat_arr[BLKIO_STAT_QUEUED][BLKIO_STAT_WRITE];
 303         stats->avg_queue_size_samples++;
 304         blkio_update_group_wait_time(stats);
 305         spin_unlock_irqrestore(&blkg->stats_lock, flags);
 306 }
 307 EXPORT_SYMBOL_GPL(blkiocg_update_avg_queue_size_stats);
 308
 309 void blkiocg_set_start_empty_time(struct blkio_group *blkg)
 310 {
 311         unsigned long flags;
 312         struct blkio_group_stats *stats;
 313
 314         spin_lock_irqsave(&blkg->stats_lock, flags);
 315         stats = &blkg->stats;
 316
 317         if (stats->stat_arr[BLKIO_STAT_QUEUED][BLKIO_STAT_READ] ||
 318                         stats->stat_arr[BLKIO_STAT_QUEUED][BLKIO_STAT_WRITE]) {
 319                 spin_unlock_irqrestore(&blkg->stats_lock, flags);
 320                 return;
 321         }
 322
 323         /*
 324          * group is already marked empty. This can happen if cfqq got new
 325          * request in parent group and moved to this group while being added
 326          * to service tree. Just ignore the event and move on.
 327          */
 328         if(blkio_blkg_empty(stats)) {
 329                 spin_unlock_irqrestore(&blkg->stats_lock, flags);
 330                 return;
 331         }
 332
 333         stats->start_empty_time = sched_clock();
 334         blkio_mark_blkg_empty(stats);
 335         spin_unlock_irqrestore(&blkg->stats_lock, flags);
 336 }
 337 EXPORT_SYMBOL_GPL(blkiocg_set_start_empty_time);
 338
 339 void blkiocg_update_dequeue_stats(struct blkio_group *blkg,
 340                         unsigned long dequeue)
 341 {
 342         blkg->stats.dequeue += dequeue;
 343 }
 344 EXPORT_SYMBOL_GPL(blkiocg_update_dequeue_stats);
 345 #else
 346 static inline void blkio_set_start_group_wait_time(struct blkio_group *blkg,
 347                                         struct blkio_group *curr_blkg) {}
 348 static inline void blkio_end_empty_time(struct blkio_group_stats *stats) {}
 349 #endif
 350
 351 void blkiocg_update_io_add_stats(struct blkio_group *blkg,
 352                         struct blkio_group *curr_blkg, bool direction,
 353                         bool sync)
 354 {
 355         unsigned long flags;
 356
 357         spin_lock_irqsave(&blkg->stats_lock, flags);
 358         blkio_add_stat(blkg->stats.stat_arr[BLKIO_STAT_QUEUED], 1, direction,
 359                         sync);
 360         blkio_end_empty_time(&blkg->stats);
 361         blkio_set_start_group_wait_time(blkg, curr_blkg);
 362         spin_unlock_irqrestore(&blkg->stats_lock, flags);
 363 }
 364 EXPORT_SYMBOL_GPL(blkiocg_update_io_add_stats);
 365
 366 void blkiocg_update_io_remove_stats(struct blkio_group *blkg,
 367                                                 bool direction, bool sync)
 368 {
 369         unsigned long flags;
 370
 371         spin_lock_irqsave(&blkg->stats_lock, flags);
 372         blkio_check_and_dec_stat(blkg->stats.stat_arr[BLKIO_STAT_QUEUED],
 373                                         direction, sync);
 374         spin_unlock_irqrestore(&blkg->stats_lock, flags);
 375 }
 376 EXPORT_SYMBOL_GPL(blkiocg_update_io_remove_stats);
 377
 378 void blkiocg_update_timeslice_used(struct blkio_group *blkg, unsigned long time,
 379                                 unsigned long unaccounted_time)
 380 {
 381         unsigned long flags;
 382
 383         spin_lock_irqsave(&blkg->stats_lock, flags);
 384         blkg->stats.time += time;
 385 #ifdef CONFIG_DEBUG_BLK_CGROUP
 386         blkg->stats.unaccounted_time += unaccounted_time;
 387 #endif
 388         spin_unlock_irqrestore(&blkg->stats_lock, flags);
 389 }
 390 EXPORT_SYMBOL_GPL(blkiocg_update_timeslice_used);
 391
 392 /*
 393  * should be called under rcu read lock or queue lock to make sure blkg pointer
 394  * is valid.
 395  */
 396 void blkiocg_update_dispatch_stats(struct blkio_group *blkg,
 397                                 uint64_t bytes, bool direction, bool sync)
 398 {
 399         struct blkio_group_stats_cpu *stats_cpu;
 400         unsigned long flags;
 401
 402         /*
 403          * Disabling interrupts to provide mutual exclusion between two
 404          * writes on same cpu. It probably is not needed for 64bit. Not
 405          * optimizing that case yet.
 406          */
 407         local_irq_save(flags);
 408
 409         stats_cpu = this_cpu_ptr(blkg->stats_cpu);
 410
 411         u64_stats_update_begin(&stats_cpu->syncp);
 412         stats_cpu->sectors += bytes >> 9;
 413         blkio_add_stat(stats_cpu->stat_arr_cpu[BLKIO_STAT_CPU_SERVICED],
 414                         1, direction, sync);
 415         blkio_add_stat(stats_cpu->stat_arr_cpu[BLKIO_STAT_CPU_SERVICE_BYTES],
 416                         bytes, direction, sync);
 417         u64_stats_update_end(&stats_cpu->syncp);
 418         local_irq_restore(flags);
 419 }
 420 EXPORT_SYMBOL_GPL(blkiocg_update_dispatch_stats);
 421
 422 void blkiocg_update_completion_stats(struct blkio_group *blkg,
 423         uint64_t start_time, uint64_t io_start_time, bool direction, bool sync)
 424 {
 425         struct blkio_group_stats *stats;
 426         unsigned long flags;
 427         unsigned long long now = sched_clock();
 428
 429         spin_lock_irqsave(&blkg->stats_lock, flags);
 430         stats = &blkg->stats;
 431         if (time_after64(now, io_start_time))
 432                 blkio_add_stat(stats->stat_arr[BLKIO_STAT_SERVICE_TIME],
 433                                 now - io_start_time, direction, sync);
 434         if (time_after64(io_start_time, start_time))
 435                 blkio_add_stat(stats->stat_arr[BLKIO_STAT_WAIT_TIME],
 436                                 io_start_time - start_time, direction, sync);
 437         spin_unlock_irqrestore(&blkg->stats_lock, flags);
 438 }
 439 EXPORT_SYMBOL_GPL(blkiocg_update_completion_stats);
 440
 441 /*  Merged stats are per cpu.  */
 442 void blkiocg_update_io_merged_stats(struct blkio_group *blkg, bool direction,
 443                                         bool sync)
 444 {
 445         struct blkio_group_stats_cpu *stats_cpu;
 446         unsigned long flags;
 447
 448         /*
 449          * Disabling interrupts to provide mutual exclusion between two
 450          * writes on same cpu. It probably is not needed for 64bit. Not
 451          * optimizing that case yet.
 452          */
 453         local_irq_save(flags);
 454
 455         stats_cpu = this_cpu_ptr(blkg->stats_cpu);
 456
 457         u64_stats_update_begin(&stats_cpu->syncp);
 458         blkio_add_stat(stats_cpu->stat_arr_cpu[BLKIO_STAT_CPU_MERGED], 1,
 459                                 direction, sync);
 460         u64_stats_update_end(&stats_cpu->syncp);
 461         local_irq_restore(flags);
 462 }
 463 EXPORT_SYMBOL_GPL(blkiocg_update_io_merged_stats);
 464
 465 /*
 466  * This function allocates the per cpu stats for blkio_group. Should be called
 467  * from sleepable context as alloc_per_cpu() requires that.
 468  */
 469 int blkio_alloc_blkg_stats(struct blkio_group *blkg)
 470 {
 471         /* Allocate memory for per cpu stats */
 472         blkg->stats_cpu = alloc_percpu(struct blkio_group_stats_cpu);
 473         if (!blkg->stats_cpu)
 474                 return -ENOMEM;
 475         return 0;
 476 }
 477 EXPORT_SYMBOL_GPL(blkio_alloc_blkg_stats);
 478
 479 void blkiocg_add_blkio_group(struct blkio_cgroup *blkcg,
 480                 struct blkio_group *blkg, void *key, dev_t dev,
 481                 enum blkio_policy_id plid)
 482 {
 483         unsigned long flags;
 484
 485         spin_lock_irqsave(&blkcg->lock, flags);
 486         spin_lock_init(&blkg->stats_lock);
 487         rcu_assign_pointer(blkg->key, key);
 488         blkg->blkcg_id = css_id(&blkcg->css);
 489         hlist_add_head_rcu(&blkg->blkcg_node, &blkcg->blkg_list);
 490         blkg->plid = plid;
 491         spin_unlock_irqrestore(&blkcg->lock, flags);
 492         /* Need to take css reference ? */
 493         cgroup_path(blkcg->css.cgroup, blkg->path, sizeof(blkg->path));
 494         blkg->dev = dev;
 495 }
 496 EXPORT_SYMBOL_GPL(blkiocg_add_blkio_group);
 497
 498 static void __blkiocg_del_blkio_group(struct blkio_group *blkg)
 499 {
 500         hlist_del_init_rcu(&blkg->blkcg_node);
 501         blkg->blkcg_id = 0;
 502 }
 503
 504 /*
 505  * returns 0 if blkio_group was still on cgroup list. Otherwise returns 1
 506  * indicating that blk_group was unhashed by the time we got to it.
 507  */
 508 int blkiocg_del_blkio_group(struct blkio_group *blkg)
 509 {
 510         struct blkio_cgroup *blkcg;
 511         unsigned long flags;
 512         struct cgroup_subsys_state *css;
 513         int ret = 1;
 514
 515         rcu_read_lock();
 516         css = css_lookup(&blkio_subsys, blkg->blkcg_id);
 517         if (css) {
 518                 blkcg = container_of(css, struct blkio_cgroup, css);
 519                 spin_lock_irqsave(&blkcg->lock, flags);
 520                 if (!hlist_unhashed(&blkg->blkcg_node)) {
 521                         __blkiocg_del_blkio_group(blkg);
 522                         ret = 0;
 523                 }
 524                 spin_unlock_irqrestore(&blkcg->lock, flags);
 525         }
 526
 527         rcu_read_unlock();
 528         return ret;
 529 }
 530 EXPORT_SYMBOL_GPL(blkiocg_del_blkio_group);
 531
 532 /* called under rcu_read_lock(). */
 533 struct blkio_group *blkiocg_lookup_group(struct blkio_cgroup *blkcg, void *key)
 534 {
 535         struct blkio_group *blkg;
 536         struct hlist_node *n;
 537         void *__key;
 538
 539         hlist_for_each_entry_rcu(blkg, n, &blkcg->blkg_list, blkcg_node) {
 540                 __key = blkg->key;
 541                 if (__key == key)
 542                         return blkg;
 543         }
 544
 545         return NULL;
 546 }
 547 EXPORT_SYMBOL_GPL(blkiocg_lookup_group);
 548
 549 static void blkio_reset_stats_cpu(struct blkio_group *blkg)
 550 {
 551         struct blkio_group_stats_cpu *stats_cpu;
 552         int i, j, k;
 553         /*
 554          * Note: On 64 bit arch this should not be an issue. This has the
 555          * possibility of returning some inconsistent value on 32bit arch
 556          * as 64bit update on 32bit is non atomic. Taking care of this
 557          * corner case makes code very complicated, like sending IPIs to
 558          * cpus, taking care of stats of offline cpus etc.
 559          *
 560          * reset stats is anyway more of a debug feature and this sounds a
 561          * corner case. So I am not complicating the code yet until and
 562          * unless this becomes a real issue.
 563          */
 564         for_each_possible_cpu(i) {
 565                 stats_cpu = per_cpu_ptr(blkg->stats_cpu, i);
 566                 stats_cpu->sectors = 0;
 567                 for(j = 0; j < BLKIO_STAT_CPU_NR; j++)
 568                         for (k = 0; k < BLKIO_STAT_TOTAL; k++)
 569                                 stats_cpu->stat_arr_cpu[j][k] = 0;
 570         }
 571 }
 572
 573 static int
 574 blkiocg_reset_stats(struct cgroup *cgroup, struct cftype *cftype, u64 val)
 575 {
 576         struct blkio_cgroup *blkcg;
 577         struct blkio_group *blkg;
 578         struct blkio_group_stats *stats;
 579         struct hlist_node *n;
 580         uint64_t queued[BLKIO_STAT_TOTAL];
 581         int i;
 582 #ifdef CONFIG_DEBUG_BLK_CGROUP
 583         bool idling, waiting, empty;
 584         unsigned long long now = sched_clock();
 585 #endif
 586
 587         blkcg = cgroup_to_blkio_cgroup(cgroup);
 588         spin_lock_irq(&blkcg->lock);
 589         hlist_for_each_entry(blkg, n, &blkcg->blkg_list, blkcg_node) {
 590                 spin_lock(&blkg->stats_lock);
 591                 stats = &blkg->stats;
 592 #ifdef CONFIG_DEBUG_BLK_CGROUP
 593                 idling = blkio_blkg_idling(stats);
 594                 waiting = blkio_blkg_waiting(stats);
 595                 empty = blkio_blkg_empty(stats);
 596 #endif
 597                 for (i = 0; i < BLKIO_STAT_TOTAL; i++)
 598                         queued[i] = stats->stat_arr[BLKIO_STAT_QUEUED][i];
 599                 memset(stats, 0, sizeof(struct blkio_group_stats));
 600                 for (i = 0; i < BLKIO_STAT_TOTAL; i++)
 601                         stats->stat_arr[BLKIO_STAT_QUEUED][i] = queued[i];
 602 #ifdef CONFIG_DEBUG_BLK_CGROUP
 603                 if (idling) {
 604                         blkio_mark_blkg_idling(stats);
 605                         stats->start_idle_time = now;
 606                 }
 607                 if (waiting) {
 608                         blkio_mark_blkg_waiting(stats);
 609                         stats->start_group_wait_time = now;
 610                 }
 611                 if (empty) {
 612                         blkio_mark_blkg_empty(stats);
 613                         stats->start_empty_time = now;
 614                 }
 615 #endif
 616                 spin_unlock(&blkg->stats_lock);
 617
 618                 /* Reset Per cpu stats which don't take blkg->stats_lock */
 619                 blkio_reset_stats_cpu(blkg);
 620         }
 621
 622         spin_unlock_irq(&blkcg->lock);
 623         return 0;
 624 }
 625
 626 static void blkio_get_key_name(enum stat_sub_type type, dev_t dev, char *str,
 627                                 int chars_left, bool diskname_only)
 628 {
 629         snprintf(str, chars_left, "%d:%d", MAJOR(dev), MINOR(dev));
 630         chars_left -= strlen(str);
 631         if (chars_left <= 0) {
 632                 printk(KERN_WARNING
 633                         "Possibly incorrect cgroup stat display format");
 634                 return;
 635         }
 636         if (diskname_only)
 637                 return;
 638         switch (type) {
 639         case BLKIO_STAT_READ:
 640                 strlcat(str, " Read", chars_left);
 641                 break;
 642         case BLKIO_STAT_WRITE:
 643                 strlcat(str, " Write", chars_left);
 644                 break;
 645         case BLKIO_STAT_SYNC:
 646                 strlcat(str, " Sync", chars_left);
 647                 break;
 648         case BLKIO_STAT_ASYNC:
 649                 strlcat(str, " Async", chars_left);
 650                 break;
 651         case BLKIO_STAT_TOTAL:
 652                 strlcat(str, " Total", chars_left);
 653                 break;
 654         default:
 655                 strlcat(str, " Invalid", chars_left);
 656         }
 657 }
 658
 659 static uint64_t blkio_fill_stat(char *str, int chars_left, uint64_t val,
 660                                 struct cgroup_map_cb *cb, dev_t dev)
 661 {
 662         blkio_get_key_name(0, dev, str, chars_left, true);
 663         cb->fill(cb, str, val);
 664         return val;
 665 }
 666
 667
 668 static uint64_t blkio_read_stat_cpu(struct blkio_group *blkg,
 669                         enum stat_type_cpu type, enum stat_sub_type sub_type)
 670 {
 671         int cpu;
 672         struct blkio_group_stats_cpu *stats_cpu;
 673         u64 val = 0, tval;
 674
 675         for_each_possible_cpu(cpu) {
 676                 unsigned int start;
 677                 stats_cpu  = per_cpu_ptr(blkg->stats_cpu, cpu);
 678
 679                 do {
 680                         start = u64_stats_fetch_begin(&stats_cpu->syncp);
 681                         if (type == BLKIO_STAT_CPU_SECTORS)
 682                                 tval = stats_cpu->sectors;
 683                         else
 684                                 tval = stats_cpu->stat_arr_cpu[type][sub_type];
 685                 } while(u64_stats_fetch_retry(&stats_cpu->syncp, start));
 686
 687                 val += tval;
 688         }
 689
 690         return val;
 691 }
 692
 693 static uint64_t blkio_get_stat_cpu(struct blkio_group *blkg,
 694                 struct cgroup_map_cb *cb, dev_t dev, enum stat_type_cpu type)
 695 {
 696         uint64_t disk_total, val;
 697         char key_str[MAX_KEY_LEN];
 698         enum stat_sub_type sub_type;
 699
 700         if (type == BLKIO_STAT_CPU_SECTORS) {
 701                 val = blkio_read_stat_cpu(blkg, type, 0);
 702                 return blkio_fill_stat(key_str, MAX_KEY_LEN - 1, val, cb, dev);
 703         }
 704
 705         for (sub_type = BLKIO_STAT_READ; sub_type < BLKIO_STAT_TOTAL;
 706                         sub_type++) {
 707                 blkio_get_key_name(sub_type, dev, key_str, MAX_KEY_LEN, false);
 708                 val = blkio_read_stat_cpu(blkg, type, sub_type);
 709                 cb->fill(cb, key_str, val);
 710         }
 711
 712         disk_total = blkio_read_stat_cpu(blkg, type, BLKIO_STAT_READ) +
 713                         blkio_read_stat_cpu(blkg, type, BLKIO_STAT_WRITE);
 714
 715         blkio_get_key_name(BLKIO_STAT_TOTAL, dev, key_str, MAX_KEY_LEN, false);
 716         cb->fill(cb, key_str, disk_total);
 717         return disk_total;
 718 }
 719
 720 /* This should be called with blkg->stats_lock held */
 721 static uint64_t blkio_get_stat(struct blkio_group *blkg,
 722                 struct cgroup_map_cb *cb, dev_t dev, enum stat_type type)
 723 {
 724         uint64_t disk_total;
 725         char key_str[MAX_KEY_LEN];
 726         enum stat_sub_type sub_type;
 727
 728         if (type == BLKIO_STAT_TIME)
 729                 return blkio_fill_stat(key_str, MAX_KEY_LEN - 1,
 730                                         blkg->stats.time, cb, dev);
 731 #ifdef CONFIG_DEBUG_BLK_CGROUP
 732         if (type == BLKIO_STAT_UNACCOUNTED_TIME)
 733                 return blkio_fill_stat(key_str, MAX_KEY_LEN - 1,
 734                                         blkg->stats.unaccounted_time, cb, dev);
 735         if (type == BLKIO_STAT_AVG_QUEUE_SIZE) {
 736                 uint64_t sum = blkg->stats.avg_queue_size_sum;
 737                 uint64_t samples = blkg->stats.avg_queue_size_samples;
 738                 if (samples)
 739                         do_div(sum, samples);
 740                 else
 741                         sum = 0;
 742                 return blkio_fill_stat(key_str, MAX_KEY_LEN - 1, sum, cb, dev);
 743         }
 744         if (type == BLKIO_STAT_GROUP_WAIT_TIME)
 745                 return blkio_fill_stat(key_str, MAX_KEY_LEN - 1,
 746                                         blkg->stats.group_wait_time, cb, dev);
 747         if (type == BLKIO_STAT_IDLE_TIME)
 748                 return blkio_fill_stat(key_str, MAX_KEY_LEN - 1,
 749                                         blkg->stats.idle_time, cb, dev);
 750         if (type == BLKIO_STAT_EMPTY_TIME)
 751                 return blkio_fill_stat(key_str, MAX_KEY_LEN - 1,
 752                                         blkg->stats.empty_time, cb, dev);
 753         if (type == BLKIO_STAT_DEQUEUE)
 754                 return blkio_fill_stat(key_str, MAX_KEY_LEN - 1,
 755                                         blkg->stats.dequeue, cb, dev);
 756 #endif
 757
 758         for (sub_type = BLKIO_STAT_READ; sub_type < BLKIO_STAT_TOTAL;
 759                         sub_type++) {
 760                 blkio_get_key_name(sub_type, dev, key_str, MAX_KEY_LEN, false);
 761                 cb->fill(cb, key_str, blkg->stats.stat_arr[type][sub_type]);
 762         }
 763         disk_total = blkg->stats.stat_arr[type][BLKIO_STAT_READ] +
 764                         blkg->stats.stat_arr[type][BLKIO_STAT_WRITE];
 765         blkio_get_key_name(BLKIO_STAT_TOTAL, dev, key_str, MAX_KEY_LEN, false);
 766         cb->fill(cb, key_str, disk_total);
 767         return disk_total;
 768 }
 769
 770 static int blkio_policy_parse_and_set(char *buf,
 771         struct blkio_policy_node *newpn, enum blkio_policy_id plid, int fileid)
 772 {
 773         struct gendisk *disk = NULL;
 774         char *s[4], *p, *major_s = NULL, *minor_s = NULL;
 775         unsigned long major, minor;
 776         int i = 0, ret = -EINVAL;
 777         int part;
 778         dev_t dev;
 779         u64 temp;
 780
 781         memset(s, 0, sizeof(s));
 782
 783         while ((p = strsep(&buf, " ")) != NULL) {
 784                 if (!*p)
 785                         continue;
 786
 787                 s[i++] = p;
 788
 789                 /* Prevent from inputing too many things */
 790                 if (i == 3)
 791                         break;
 792         }
 793
 794         if (i != 2)
 795                 goto out;
 796
 797         p = strsep(&s[0], ":");
 798         if (p != NULL)
 799                 major_s = p;
 800         else
 801                 goto out;
 802
 803         minor_s = s[0];
 804         if (!minor_s)
 805                 goto out;
 806
 807         if (strict_strtoul(major_s, 10, &major))
 808                 goto out;
 809
 810         if (strict_strtoul(minor_s, 10, &minor))
 811                 goto out;
 812
 813         dev = MKDEV(major, minor);
 814
 815         if (strict_strtoull(s[1], 10, &temp))
 816                 goto out;
 817
 818         /* For rule removal, do not check for device presence. */
 819         if (temp) {
 820                 disk = get_gendisk(dev, &part);
 821                 if (!disk || part) {
 822                         ret = -ENODEV;
 823                         goto out;
 824                 }
 825         }
 826
 827         newpn->dev = dev;
 828
 829         switch (plid) {
 830         case BLKIO_POLICY_PROP:
 831                 if ((temp < BLKIO_WEIGHT_MIN && temp > 0) ||
 832                      temp > BLKIO_WEIGHT_MAX)
 833                         goto out;
 834
 835                 newpn->plid = plid;
 836                 newpn->fileid = fileid;
 837                 newpn->val.weight = temp;
 838                 break;
 839         case BLKIO_POLICY_THROTL:
 840                 switch(fileid) {
 841                 case BLKIO_THROTL_read_bps_device:
 842                 case BLKIO_THROTL_write_bps_device:
 843                         newpn->plid = plid;
 844                         newpn->fileid = fileid;
 845                         newpn->val.bps = temp;
 846                         break;
 847                 case BLKIO_THROTL_read_iops_device:
 848                 case BLKIO_THROTL_write_iops_device:
 849                         if (temp > THROTL_IOPS_MAX)
 850                                 goto out;
 851
 852                         newpn->plid = plid;
 853                         newpn->fileid = fileid;
 854                         newpn->val.iops = (unsigned int)temp;
 855                         break;
 856                 }
 857                 break;
 858         default:
 859                 BUG();
 860         }
 861         ret = 0;
 862 out:
 863         put_disk(disk);
 864         return ret;
 865 }
 866
 867 unsigned int blkcg_get_weight(struct blkio_cgroup *blkcg,
 868                               dev_t dev)
 869 {
 870         struct blkio_policy_node *pn;
 871         unsigned long flags;
 872         unsigned int weight;
 873
 874         spin_lock_irqsave(&blkcg->lock, flags);
 875
 876         pn = blkio_policy_search_node(blkcg, dev, BLKIO_POLICY_PROP,
 877                                 BLKIO_PROP_weight_device);
 878         if (pn)
 879                 weight = pn->val.weight;
 880         else
 881                 weight = blkcg->weight;
 882
 883         spin_unlock_irqrestore(&blkcg->lock, flags);
 884
 885         return weight;
 886 }
 887 EXPORT_SYMBOL_GPL(blkcg_get_weight);
 888
 889 uint64_t blkcg_get_read_bps(struct blkio_cgroup *blkcg, dev_t dev)
 890 {
 891         struct blkio_policy_node *pn;
 892         unsigned long flags;
 893         uint64_t bps = -1;
 894
 895         spin_lock_irqsave(&blkcg->lock, flags);
 896         pn = blkio_policy_search_node(blkcg, dev, BLKIO_POLICY_THROTL,
 897                                 BLKIO_THROTL_read_bps_device);
 898         if (pn)
 899                 bps = pn->val.bps;
 900         spin_unlock_irqrestore(&blkcg->lock, flags);
 901
 902         return bps;
 903 }
 904
 905 uint64_t blkcg_get_write_bps(struct blkio_cgroup *blkcg, dev_t dev)
 906 {
 907         struct blkio_policy_node *pn;
 908         unsigned long flags;
 909         uint64_t bps = -1;
 910
 911         spin_lock_irqsave(&blkcg->lock, flags);
 912         pn = blkio_policy_search_node(blkcg, dev, BLKIO_POLICY_THROTL,
 913                                 BLKIO_THROTL_write_bps_device);
 914         if (pn)
 915                 bps = pn->val.bps;
 916         spin_unlock_irqrestore(&blkcg->lock, flags);
 917
 918         return bps;
 919 }
 920
 921 unsigned int blkcg_get_read_iops(struct blkio_cgroup *blkcg, dev_t dev)
 922 {
 923         struct blkio_policy_node *pn;
 924         unsigned long flags;
 925         unsigned int iops = -1;
 926
 927         spin_lock_irqsave(&blkcg->lock, flags);
 928         pn = blkio_policy_search_node(blkcg, dev, BLKIO_POLICY_THROTL,
 929                                 BLKIO_THROTL_read_iops_device);
 930         if (pn)
 931                 iops = pn->val.iops;
 932         spin_unlock_irqrestore(&blkcg->lock, flags);
 933
 934         return iops;
 935 }
 936
 937 unsigned int blkcg_get_write_iops(struct blkio_cgroup *blkcg, dev_t dev)
 938 {
 939         struct blkio_policy_node *pn;
 940         unsigned long flags;
 941         unsigned int iops = -1;
 942
 943         spin_lock_irqsave(&blkcg->lock, flags);
 944         pn = blkio_policy_search_node(blkcg, dev, BLKIO_POLICY_THROTL,
 945                                 BLKIO_THROTL_write_iops_device);
 946         if (pn)
 947                 iops = pn->val.iops;
 948         spin_unlock_irqrestore(&blkcg->lock, flags);
 949
 950         return iops;
 951 }
 952
 953 /* Checks whether user asked for deleting a policy rule */
 954 static bool blkio_delete_rule_command(struct blkio_policy_node *pn)
 955 {
 956         switch(pn->plid) {
 957         case BLKIO_POLICY_PROP:
 958                 if (pn->val.weight == 0)
 959                         return 1;
 960                 break;
 961         case BLKIO_POLICY_THROTL:
 962                 switch(pn->fileid) {
 963                 case BLKIO_THROTL_read_bps_device:
 964                 case BLKIO_THROTL_write_bps_device:
 965                         if (pn->val.bps == 0)
 966                                 return 1;
 967                         break;
 968                 case BLKIO_THROTL_read_iops_device:
 969                 case BLKIO_THROTL_write_iops_device:
 970                         if (pn->val.iops == 0)
 971                                 return 1;
 972                 }
 973                 break;
 974         default:
 975                 BUG();
 976         }
 977
 978         return 0;
 979 }
 980
 981 static void blkio_update_policy_rule(struct blkio_policy_node *oldpn,
 982                                         struct blkio_policy_node *newpn)
 983 {
 984         switch(oldpn->plid) {
 985         case BLKIO_POLICY_PROP:
 986                 oldpn->val.weight = newpn->val.weight;
 987                 break;
 988         case BLKIO_POLICY_THROTL:
 989                 switch(newpn->fileid) {
 990                 case BLKIO_THROTL_read_bps_device:
 991                 case BLKIO_THROTL_write_bps_device:
 992                         oldpn->val.bps = newpn->val.bps;
 993                         break;
 994                 case BLKIO_THROTL_read_iops_device:
 995                 case BLKIO_THROTL_write_iops_device:
 996                         oldpn->val.iops = newpn->val.iops;
 997                 }
 998                 break;
 999         default:
1000                 BUG();
1001         }
1002 }
1003
1004 /*
1005  * Some rules/values in blkg have changed. Propagate those to respective
1006  * policies.
1007  */
1008 static void blkio_update_blkg_policy(struct blkio_cgroup *blkcg,
1009                 struct blkio_group *blkg, struct blkio_policy_node *pn)
1010 {
1011         unsigned int weight, iops;
1012         u64 bps;
1013
1014         switch(pn->plid) {
1015         case BLKIO_POLICY_PROP:
1016                 weight = pn->val.weight ? pn->val.weight :
1017                                 blkcg->weight;
1018                 blkio_update_group_weight(blkg, weight);
1019                 break;
1020         case BLKIO_POLICY_THROTL:
1021                 switch(pn->fileid) {
1022                 case BLKIO_THROTL_read_bps_device:
1023                 case BLKIO_THROTL_write_bps_device:
1024                         bps = pn->val.bps ? pn->val.bps : (-1);
1025                         blkio_update_group_bps(blkg, bps, pn->fileid);
1026                         break;
1027                 case BLKIO_THROTL_read_iops_device:
1028                 case BLKIO_THROTL_write_iops_device:
1029                         iops = pn->val.iops ? pn->val.iops : (-1);
1030                         blkio_update_group_iops(blkg, iops, pn->fileid);
1031                         break;
1032                 }
1033                 break;
1034         default:
1035                 BUG();
1036         }
1037 }
1038
1039 /*
1040  * A policy node rule has been updated. Propagate this update to all the
1041  * block groups which might be affected by this update.
1042  */
1043 static void blkio_update_policy_node_blkg(struct blkio_cgroup *blkcg,
1044                                 struct blkio_policy_node *pn)
1045 {
1046         struct blkio_group *blkg;
1047         struct hlist_node *n;
1048
1049         spin_lock(&blkio_list_lock);
1050         spin_lock_irq(&blkcg->lock);
1051
1052         hlist_for_each_entry(blkg, n, &blkcg->blkg_list, blkcg_node) {
1053                 if (pn->dev != blkg->dev || pn->plid != blkg->plid)
1054                         continue;
1055                 blkio_update_blkg_policy(blkcg, blkg, pn);
1056         }
1057
1058         spin_unlock_irq(&blkcg->lock);
1059         spin_unlock(&blkio_list_lock);
1060 }
1061
1062 static int blkiocg_file_write(struct cgroup *cgrp, struct cftype *cft,
1063                                        const char *buffer)
1064 {
1065         int ret = 0;
1066         char *buf;
1067         struct blkio_policy_node *newpn, *pn;
1068         struct blkio_cgroup *blkcg;
1069         int keep_newpn = 0;
1070         enum blkio_policy_id plid = BLKIOFILE_POLICY(cft->private);
1071         int fileid = BLKIOFILE_ATTR(cft->private);
1072
1073         buf = kstrdup(buffer, GFP_KERNEL);
1074         if (!buf)
1075                 return -ENOMEM;
1076
1077         newpn = kzalloc(sizeof(*newpn), GFP_KERNEL);
1078         if (!newpn) {
1079                 ret = -ENOMEM;
1080                 goto free_buf;
1081         }
1082
1083         ret = blkio_policy_parse_and_set(buf, newpn, plid, fileid);
1084         if (ret)
1085                 goto free_newpn;
1086
1087         blkcg = cgroup_to_blkio_cgroup(cgrp);
1088
1089         spin_lock_irq(&blkcg->lock);
1090
1091         pn = blkio_policy_search_node(blkcg, newpn->dev, plid, fileid);
1092         if (!pn) {
1093                 if (!blkio_delete_rule_command(newpn)) {
1094                         blkio_policy_insert_node(blkcg, newpn);
1095                         keep_newpn = 1;
1096                 }
1097                 spin_unlock_irq(&blkcg->lock);
1098                 goto update_io_group;
1099         }
1100
1101         if (blkio_delete_rule_command(newpn)) {
1102                 blkio_policy_delete_node(pn);
1103                 kfree(pn);
1104                 spin_unlock_irq(&blkcg->lock);
1105                 goto update_io_group;
1106         }
1107         spin_unlock_irq(&blkcg->lock);
1108
1109         blkio_update_policy_rule(pn, newpn);
1110
1111 update_io_group:
1112         blkio_update_policy_node_blkg(blkcg, newpn);
1113
1114 free_newpn:
1115         if (!keep_newpn)
1116                 kfree(newpn);
1117 free_buf:
1118         kfree(buf);
1119         return ret;
1120 }
1121
1122 static void
1123 blkio_print_policy_node(struct seq_file *m, struct blkio_policy_node *pn)
1124 {
1125         switch(pn->plid) {
1126                 case BLKIO_POLICY_PROP:
1127                         if (pn->fileid == BLKIO_PROP_weight_device)
1128                                 seq_printf(m, "%u:%u\t%u\n", MAJOR(pn->dev),
1129                                         MINOR(pn->dev), pn->val.weight);
1130                         break;
1131                 case BLKIO_POLICY_THROTL:
1132                         switch(pn->fileid) {
1133                         case BLKIO_THROTL_read_bps_device:
1134                         case BLKIO_THROTL_write_bps_device:
1135                                 seq_printf(m, "%u:%u\t%llu\n", MAJOR(pn->dev),
1136                                         MINOR(pn->dev), pn->val.bps);
1137                                 break;
1138                         case BLKIO_THROTL_read_iops_device:
1139                         case BLKIO_THROTL_write_iops_device:
1140                                 seq_printf(m, "%u:%u\t%u\n", MAJOR(pn->dev),
1141                                         MINOR(pn->dev), pn->val.iops);
1142                                 break;
1143                         }
1144                         break;
1145                 default:
1146                         BUG();
1147         }
1148 }
1149
1150 /* cgroup files which read their data from policy nodes end up here */
1151 static void blkio_read_policy_node_files(struct cftype *cft,
1152                         struct blkio_cgroup *blkcg, struct seq_file *m)
1153 {
1154         struct blkio_policy_node *pn;
1155
1156         if (!list_empty(&blkcg->policy_list)) {
1157                 spin_lock_irq(&blkcg->lock);
1158                 list_for_each_entry(pn, &blkcg->policy_list, node) {
1159                         if (!pn_matches_cftype(cft, pn))
1160                                 continue;
1161                         blkio_print_policy_node(m, pn);
1162                 }
1163                 spin_unlock_irq(&blkcg->lock);
1164         }
1165 }
1166
1167 static int blkiocg_file_read(struct cgroup *cgrp, struct cftype *cft,
1168                                 struct seq_file *m)
1169 {
1170         struct blkio_cgroup *blkcg;
1171         enum blkio_policy_id plid = BLKIOFILE_POLICY(cft->private);
1172         int name = BLKIOFILE_ATTR(cft->private);
1173
1174         blkcg = cgroup_to_blkio_cgroup(cgrp);
1175
1176         switch(plid) {
1177         case BLKIO_POLICY_PROP:
1178                 switch(name) {
1179                 case BLKIO_PROP_weight_device:
1180                         blkio_read_policy_node_files(cft, blkcg, m);
1181                         return 0;
1182                 default:
1183                         BUG();
1184                 }
1185                 break;
1186         case BLKIO_POLICY_THROTL:
1187                 switch(name){
1188                 case BLKIO_THROTL_read_bps_device:
1189                 case BLKIO_THROTL_write_bps_device:
1190                 case BLKIO_THROTL_read_iops_device:
1191                 case BLKIO_THROTL_write_iops_device:
1192                         blkio_read_policy_node_files(cft, blkcg, m);
1193                         return 0;
1194                 default:
1195                         BUG();
1196                 }
1197                 break;
1198         default:
1199                 BUG();
1200         }
1201
1202         return 0;
1203 }
1204
1205 static int blkio_read_blkg_stats(struct blkio_cgroup *blkcg,
1206                 struct cftype *cft, struct cgroup_map_cb *cb,
1207                 enum stat_type type, bool show_total, bool pcpu)
1208 {
1209         struct blkio_group *blkg;
1210         struct hlist_node *n;
1211         uint64_t cgroup_total = 0;
1212
1213         rcu_read_lock();
1214         hlist_for_each_entry_rcu(blkg, n, &blkcg->blkg_list, blkcg_node) {
1215                 if (blkg->dev) {
1216                         if (!cftype_blkg_same_policy(cft, blkg))
1217                                 continue;
1218                         if (pcpu)
1219                                 cgroup_total += blkio_get_stat_cpu(blkg, cb,
1220                                                 blkg->dev, type);
1221                         else {
1222                                 spin_lock_irq(&blkg->stats_lock);
1223                                 cgroup_total += blkio_get_stat(blkg, cb,
1224                                                 blkg->dev, type);
1225                                 spin_unlock_irq(&blkg->stats_lock);
1226                         }
1227                 }
1228         }
1229         if (show_total)
1230                 cb->fill(cb, "Total", cgroup_total);
1231         rcu_read_unlock();
1232         return 0;
1233 }
1234
1235 /* All map kind of cgroup file get serviced by this function */
1236 static int blkiocg_file_read_map(struct cgroup *cgrp, struct cftype *cft,
1237                                 struct cgroup_map_cb *cb)
1238 {
1239         struct blkio_cgroup *blkcg;
1240         enum blkio_policy_id plid = BLKIOFILE_POLICY(cft->private);
1241         int name = BLKIOFILE_ATTR(cft->private);
1242
1243         blkcg = cgroup_to_blkio_cgroup(cgrp);
1244
1245         switch(plid) {
1246         case BLKIO_POLICY_PROP:
1247                 switch(name) {
1248                 case BLKIO_PROP_time:
1249                         return blkio_read_blkg_stats(blkcg, cft, cb,
1250                                                 BLKIO_STAT_TIME, 0, 0);
1251                 case BLKIO_PROP_sectors:
1252                         return blkio_read_blkg_stats(blkcg, cft, cb,
1253                                                 BLKIO_STAT_CPU_SECTORS, 0, 1);
1254                 case BLKIO_PROP_io_service_bytes:
1255                         return blkio_read_blkg_stats(blkcg, cft, cb,
1256                                         BLKIO_STAT_CPU_SERVICE_BYTES, 1, 1);
1257                 case BLKIO_PROP_io_serviced:
1258                         return blkio_read_blkg_stats(blkcg, cft, cb,
1259                                                 BLKIO_STAT_CPU_SERVICED, 1, 1);
1260                 case BLKIO_PROP_io_service_time:
1261                         return blkio_read_blkg_stats(blkcg, cft, cb,
1262                                                 BLKIO_STAT_SERVICE_TIME, 1, 0);
1263                 case BLKIO_PROP_io_wait_time:
1264                         return blkio_read_blkg_stats(blkcg, cft, cb,
1265                                                 BLKIO_STAT_WAIT_TIME, 1, 0);
1266                 case BLKIO_PROP_io_merged:
1267                         return blkio_read_blkg_stats(blkcg, cft, cb,
1268                                                 BLKIO_STAT_CPU_MERGED, 1, 1);
1269                 case BLKIO_PROP_io_queued:
1270                         return blkio_read_blkg_stats(blkcg, cft, cb,
1271                                                 BLKIO_STAT_QUEUED, 1, 0);
1272 #ifdef CONFIG_DEBUG_BLK_CGROUP
1273                 case BLKIO_PROP_unaccounted_time:
1274                         return blkio_read_blkg_stats(blkcg, cft, cb,
1275                                         BLKIO_STAT_UNACCOUNTED_TIME, 0, 0);
1276                 case BLKIO_PROP_dequeue:
1277                         return blkio_read_blkg_stats(blkcg, cft, cb,
1278                                                 BLKIO_STAT_DEQUEUE, 0, 0);
1279                 case BLKIO_PROP_avg_queue_size:
1280                         return blkio_read_blkg_stats(blkcg, cft, cb,
1281                                         BLKIO_STAT_AVG_QUEUE_SIZE, 0, 0);
1282                 case BLKIO_PROP_group_wait_time:
1283                         return blkio_read_blkg_stats(blkcg, cft, cb,
1284                                         BLKIO_STAT_GROUP_WAIT_TIME, 0, 0);
1285                 case BLKIO_PROP_idle_time:
1286                         return blkio_read_blkg_stats(blkcg, cft, cb,
1287                                                 BLKIO_STAT_IDLE_TIME, 0, 0);
1288                 case BLKIO_PROP_empty_time:
1289                         return blkio_read_blkg_stats(blkcg, cft, cb,
1290                                                 BLKIO_STAT_EMPTY_TIME, 0, 0);
1291 #endif
1292                 default:
1293                         BUG();
1294                 }
1295                 break;
1296         case BLKIO_POLICY_THROTL:
1297                 switch(name){
1298                 case BLKIO_THROTL_io_service_bytes:
1299                         return blkio_read_blkg_stats(blkcg, cft, cb,
1300                                                 BLKIO_STAT_CPU_SERVICE_BYTES, 1, 1);
1301                 case BLKIO_THROTL_io_serviced:
1302                         return blkio_read_blkg_stats(blkcg, cft, cb,
1303                                                 BLKIO_STAT_CPU_SERVICED, 1, 1);
1304                 default:
1305                         BUG();
1306                 }
1307                 break;
1308         default:
1309                 BUG();
1310         }
1311
1312         return 0;
1313 }
1314
1315 static int blkio_weight_write(struct blkio_cgroup *blkcg, u64 val)
1316 {
1317         struct blkio_group *blkg;
1318         struct hlist_node *n;
1319         struct blkio_policy_node *pn;
1320
1321         if (val < BLKIO_WEIGHT_MIN || val > BLKIO_WEIGHT_MAX)
1322                 return -EINVAL;
1323
1324         spin_lock(&blkio_list_lock);
1325         spin_lock_irq(&blkcg->lock);
1326         blkcg->weight = (unsigned int)val;
1327
1328         hlist_for_each_entry(blkg, n, &blkcg->blkg_list, blkcg_node) {
1329                 pn = blkio_policy_search_node(blkcg, blkg->dev,
1330                                 BLKIO_POLICY_PROP, BLKIO_PROP_weight_device);
1331                 if (pn)
1332                         continue;
1333
1334                 blkio_update_group_weight(blkg, blkcg->weight);
1335         }
1336         spin_unlock_irq(&blkcg->lock);
1337         spin_unlock(&blkio_list_lock);
1338         return 0;
1339 }
1340
1341 static u64 blkiocg_file_read_u64 (struct cgroup *cgrp, struct cftype *cft) {
1342         struct blkio_cgroup *blkcg;
1343         enum blkio_policy_id plid = BLKIOFILE_POLICY(cft->private);
1344         int name = BLKIOFILE_ATTR(cft->private);
1345
1346         blkcg = cgroup_to_blkio_cgroup(cgrp);
1347
1348         switch(plid) {
1349         case BLKIO_POLICY_PROP:
1350                 switch(name) {
1351                 case BLKIO_PROP_weight:
1352                         return (u64)blkcg->weight;
1353                 }
1354                 break;
1355         default:
1356                 BUG();
1357         }
1358         return 0;
1359 }
1360
1361 static int
1362 blkiocg_file_write_u64(struct cgroup *cgrp, struct cftype *cft, u64 val)
1363 {
1364         struct blkio_cgroup *blkcg;
1365         enum blkio_policy_id plid = BLKIOFILE_POLICY(cft->private);
1366         int name = BLKIOFILE_ATTR(cft->private);
1367
1368         blkcg = cgroup_to_blkio_cgroup(cgrp);
1369
1370         switch(plid) {
1371         case BLKIO_POLICY_PROP:
1372                 switch(name) {
1373                 case BLKIO_PROP_weight:
1374                         return blkio_weight_write(blkcg, val);
1375                 }
1376                 break;
1377         default:
1378                 BUG();
1379         }
1380
1381         return 0;
1382 }
1383
1384 struct cftype blkio_files[] = {
1385         {
1386                 .name = "weight_device",
1387                 .private = BLKIOFILE_PRIVATE(BLKIO_POLICY_PROP,
1388                                 BLKIO_PROP_weight_device),
1389                 .read_seq_string = blkiocg_file_read,
1390                 .write_string = blkiocg_file_write,
1391                 .max_write_len = 256,
1392         },
1393         {
1394                 .name = "weight",
1395                 .private = BLKIOFILE_PRIVATE(BLKIO_POLICY_PROP,
1396                                 BLKIO_PROP_weight),
1397                 .read_u64 = blkiocg_file_read_u64,
1398                 .write_u64 = blkiocg_file_write_u64,
1399         },
1400         {
1401                 .name = "time",
1402                 .private = BLKIOFILE_PRIVATE(BLKIO_POLICY_PROP,
1403                                 BLKIO_PROP_time),
1404                 .read_map = blkiocg_file_read_map,
1405         },
1406         {
1407                 .name = "sectors",
1408                 .private = BLKIOFILE_PRIVATE(BLKIO_POLICY_PROP,
1409                                 BLKIO_PROP_sectors),
1410                 .read_map = blkiocg_file_read_map,
1411         },
1412         {
1413                 .name = "io_service_bytes",
1414                 .private = BLKIOFILE_PRIVATE(BLKIO_POLICY_PROP,
1415                                 BLKIO_PROP_io_service_bytes),
1416                 .read_map = blkiocg_file_read_map,
1417         },
1418         {
1419                 .name = "io_serviced",
1420                 .private = BLKIOFILE_PRIVATE(BLKIO_POLICY_PROP,
1421                                 BLKIO_PROP_io_serviced),
1422                 .read_map = blkiocg_file_read_map,
1423         },
1424         {
1425                 .name = "io_service_time",
1426                 .private = BLKIOFILE_PRIVATE(BLKIO_POLICY_PROP,
1427                                 BLKIO_PROP_io_service_time),
1428                 .read_map = blkiocg_file_read_map,
1429         },
1430         {
1431                 .name = "io_wait_time",
1432                 .private = BLKIOFILE_PRIVATE(BLKIO_POLICY_PROP,
1433                                 BLKIO_PROP_io_wait_time),
1434                 .read_map = blkiocg_file_read_map,
1435         },
1436         {
1437                 .name = "io_merged",
1438                 .private = BLKIOFILE_PRIVATE(BLKIO_POLICY_PROP,
1439                                 BLKIO_PROP_io_merged),
1440                 .read_map = blkiocg_file_read_map,
1441         },
1442         {
1443                 .name = "io_queued",
1444                 .private = BLKIOFILE_PRIVATE(BLKIO_POLICY_PROP,
1445                                 BLKIO_PROP_io_queued),
1446                 .read_map = blkiocg_file_read_map,
1447         },
1448         {
1449                 .name = "reset_stats",
1450                 .write_u64 = blkiocg_reset_stats,
1451         },
1452 #ifdef CONFIG_BLK_DEV_THROTTLING
1453         {
1454                 .name = "throttle.read_bps_device",
1455                 .private = BLKIOFILE_PRIVATE(BLKIO_POLICY_THROTL,
1456                                 BLKIO_THROTL_read_bps_device),
1457                 .read_seq_string = blkiocg_file_read,
1458                 .write_string = blkiocg_file_write,
1459                 .max_write_len = 256,
1460         },
1461
1462         {
1463                 .name = "throttle.write_bps_device",
1464                 .private = BLKIOFILE_PRIVATE(BLKIO_POLICY_THROTL,
1465                                 BLKIO_THROTL_write_bps_device),
1466                 .read_seq_string = blkiocg_file_read,
1467                 .write_string = blkiocg_file_write,
1468                 .max_write_len = 256,
1469         },
1470
1471         {
1472                 .name = "throttle.read_iops_device",
1473                 .private = BLKIOFILE_PRIVATE(BLKIO_POLICY_THROTL,
1474                                 BLKIO_THROTL_read_iops_device),
1475                 .read_seq_string = blkiocg_file_read,
1476                 .write_string = blkiocg_file_write,
1477                 .max_write_len = 256,
1478         },
1479
1480         {
1481                 .name = "throttle.write_iops_device",
1482                 .private = BLKIOFILE_PRIVATE(BLKIO_POLICY_THROTL,
1483                                 BLKIO_THROTL_write_iops_device),
1484                 .read_seq_string = blkiocg_file_read,
1485                 .write_string = blkiocg_file_write,
1486                 .max_write_len = 256,
1487         },
1488         {
1489                 .name = "throttle.io_service_bytes",
1490                 .private = BLKIOFILE_PRIVATE(BLKIO_POLICY_THROTL,
1491                                 BLKIO_THROTL_io_service_bytes),
1492                 .read_map = blkiocg_file_read_map,
1493         },
1494         {
1495                 .name = "throttle.io_serviced",
1496                 .private = BLKIOFILE_PRIVATE(BLKIO_POLICY_THROTL,
1497                                 BLKIO_THROTL_io_serviced),
1498                 .read_map = blkiocg_file_read_map,
1499         },
1500 #endif /* CONFIG_BLK_DEV_THROTTLING */
1501
1502 #ifdef CONFIG_DEBUG_BLK_CGROUP
1503         {
1504                 .name = "avg_queue_size",
1505                 .private = BLKIOFILE_PRIVATE(BLKIO_POLICY_PROP,
1506                                 BLKIO_PROP_avg_queue_size),
1507                 .read_map = blkiocg_file_read_map,
1508         },
1509         {
1510                 .name = "group_wait_time",
1511                 .private = BLKIOFILE_PRIVATE(BLKIO_POLICY_PROP,
1512                                 BLKIO_PROP_group_wait_time),
1513                 .read_map = blkiocg_file_read_map,
1514         },
1515         {
1516                 .name = "idle_time",
1517                 .private = BLKIOFILE_PRIVATE(BLKIO_POLICY_PROP,
1518                                 BLKIO_PROP_idle_time),
1519                 .read_map = blkiocg_file_read_map,
1520         },
1521         {
1522                 .name = "empty_time",
1523                 .private = BLKIOFILE_PRIVATE(BLKIO_POLICY_PROP,
1524                                 BLKIO_PROP_empty_time),
1525                 .read_map = blkiocg_file_read_map,
1526         },
1527         {
1528                 .name = "dequeue",
1529                 .private = BLKIOFILE_PRIVATE(BLKIO_POLICY_PROP,
1530                                 BLKIO_PROP_dequeue),
1531                 .read_map = blkiocg_file_read_map,
1532         },
1533         {
1534                 .name = "unaccounted_time",
1535                 .private = BLKIOFILE_PRIVATE(BLKIO_POLICY_PROP,
1536                                 BLKIO_PROP_unaccounted_time),
1537                 .read_map = blkiocg_file_read_map,
1538         },
1539 #endif
1540 };
1541
1542 static int blkiocg_populate(struct cgroup_subsys *subsys, struct cgroup *cgroup)
1543 {
1544         return cgroup_add_files(cgroup, subsys, blkio_files,
1545                                 ARRAY_SIZE(blkio_files));
1546 }
1547
1548 static void blkiocg_destroy(struct cgroup *cgroup)
1549 {
1550         struct blkio_cgroup *blkcg = cgroup_to_blkio_cgroup(cgroup);
1551         unsigned long flags;
1552         struct blkio_group *blkg;
1553         void *key;
1554         struct blkio_policy_type *blkiop;
1555         struct blkio_policy_node *pn, *pntmp;
1556
1557         rcu_read_lock();
1558         do {
1559                 spin_lock_irqsave(&blkcg->lock, flags);
1560
1561                 if (hlist_empty(&blkcg->blkg_list)) {
1562                         spin_unlock_irqrestore(&blkcg->lock, flags);
1563                         break;
1564                 }
1565
1566                 blkg = hlist_entry(blkcg->blkg_list.first, struct blkio_group,
1567                                         blkcg_node);
1568                 key = rcu_dereference(blkg->key);
1569                 __blkiocg_del_blkio_group(blkg);
1570
1571                 spin_unlock_irqrestore(&blkcg->lock, flags);
1572
1573                 /*
1574                  * This blkio_group is being unlinked as associated cgroup is
1575                  * going away. Let all the IO controlling policies know about
1576                  * this event.
1577                  */
1578                 spin_lock(&blkio_list_lock);
1579                 list_for_each_entry(blkiop, &blkio_list, list) {
1580                         if (blkiop->plid != blkg->plid)
1581                                 continue;
1582                         blkiop->ops.blkio_unlink_group_fn(key, blkg);
1583                 }
1584                 spin_unlock(&blkio_list_lock);
1585         } while (1);
1586
1587         list_for_each_entry_safe(pn, pntmp, &blkcg->policy_list, node) {
1588                 blkio_policy_delete_node(pn);
1589                 kfree(pn);
1590         }
1591
1592         free_css_id(&blkio_subsys, &blkcg->css);
1593         rcu_read_unlock();
1594         if (blkcg != &blkio_root_cgroup)
1595                 kfree(blkcg);
1596 }
1597
1598 static struct cgroup_subsys_state *blkiocg_create(struct cgroup *cgroup)
1599 {
1600         struct blkio_cgroup *blkcg;
1601         struct cgroup *parent = cgroup->parent;
1602
1603         if (!parent) {
1604                 blkcg = &blkio_root_cgroup;
1605                 goto done;
1606         }
1607
1608         blkcg = kzalloc(sizeof(*blkcg), GFP_KERNEL);
1609         if (!blkcg)
1610                 return ERR_PTR(-ENOMEM);
1611
1612         blkcg->weight = BLKIO_WEIGHT_DEFAULT;
1613 done:
1614         spin_lock_init(&blkcg->lock);
1615         INIT_HLIST_HEAD(&blkcg->blkg_list);
1616
1617         INIT_LIST_HEAD(&blkcg->policy_list);
1618         return &blkcg->css;
1619 }
1620
1621 /*
1622  * We cannot support shared io contexts, as we have no mean to support
1623  * two tasks with the same ioc in two different groups without major rework
1624  * of the main cic data structures.  For now we allow a task to change
1625  * its cgroup only if it's the only owner of its ioc.
1626  */
1627 static int blkiocg_can_attach(struct cgroup *cgrp, struct cgroup_taskset *tset)
1628 {
1629         struct task_struct *task;
1630         struct io_context *ioc;
1631         int ret = 0;
1632
1633         /* task_lock() is needed to avoid races with exit_io_context() */
1634         cgroup_taskset_for_each(task, cgrp, tset) {
1635                 task_lock(task);
1636                 ioc = task->io_context;
1637                 if (ioc && atomic_read(&ioc->nr_tasks) > 1)
1638                         ret = -EINVAL;
1639                 task_unlock(task);
1640                 if (ret)
1641                         break;
1642         }
1643         return ret;
1644 }
1645
1646 static void blkiocg_attach(struct cgroup *cgrp, struct cgroup_taskset *tset)
1647 {
1648         struct task_struct *task;
1649         struct io_context *ioc;
1650
1651         cgroup_taskset_for_each(task, cgrp, tset) {
1652                 /* we don't lose anything even if ioc allocation fails */
1653                 ioc = get_task_io_context(task, GFP_ATOMIC, NUMA_NO_NODE);
1654                 if (ioc) {
1655                         ioc_cgroup_changed(ioc);
1656                         put_io_context(ioc);
1657                 }
1658         }
1659 }
1660
1661 void blkio_policy_register(struct blkio_policy_type *blkiop)
1662 {
1663         spin_lock(&blkio_list_lock);
1664         list_add_tail(&blkiop->list, &blkio_list);
1665         spin_unlock(&blkio_list_lock);
1666 }
1667 EXPORT_SYMBOL_GPL(blkio_policy_register);
1668
1669 void blkio_policy_unregister(struct blkio_policy_type *blkiop)
1670 {
1671         spin_lock(&blkio_list_lock);
1672         list_del_init(&blkiop->list);
1673         spin_unlock(&blkio_list_lock);
1674 }
1675 EXPORT_SYMBOL_GPL(blkio_policy_unregister);
1676
1677 static int __init init_cgroup_blkio(void)
1678 {
1679         return cgroup_load_subsys(&blkio_subsys);
1680 }
1681
1682 static void __exit exit_cgroup_blkio(void)
1683 {
1684         cgroup_unload_subsys(&blkio_subsys);
1685 }
1686
1687 module_init(init_cgroup_blkio);
1688 module_exit(exit_cgroup_blkio);
1689 MODULE_LICENSE("GPL");