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