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 /*
 396  * should be called under rcu read lock or queue lock to make sure blkg pointer
 397  * is valid.
 398  */
 399 void blkiocg_update_dispatch_stats(struct blkio_group *blkg,
 400                                 uint64_t bytes, bool direction, bool sync)
 401 {
 402         struct blkio_group_stats_cpu *stats_cpu;
 403         unsigned long flags;
 404
 405         /*
 406          * Disabling interrupts to provide mutual exclusion between two
 407          * writes on same cpu. It probably is not needed for 64bit. Not
 408          * optimizing that case yet.
 409          */
 410         local_irq_save(flags);
 411
 412         stats_cpu = this_cpu_ptr(blkg->stats_cpu);
 413
 414         u64_stats_update_begin(&stats_cpu->syncp);
 415         stats_cpu->sectors += bytes >> 9;
 416         blkio_add_stat(stats_cpu->stat_arr_cpu[BLKIO_STAT_CPU_SERVICED],
 417                         1, direction, sync);
 418         blkio_add_stat(stats_cpu->stat_arr_cpu[BLKIO_STAT_CPU_SERVICE_BYTES],
 419                         bytes, direction, sync);
 420         u64_stats_update_end(&stats_cpu->syncp);
 421         local_irq_restore(flags);
 422 }
 423 EXPORT_SYMBOL_GPL(blkiocg_update_dispatch_stats);
 424
 425 void blkiocg_update_completion_stats(struct blkio_group *blkg,
 426         uint64_t start_time, uint64_t io_start_time, bool direction, bool sync)
 427 {
 428         struct blkio_group_stats *stats;
 429         unsigned long flags;
 430         unsigned long long now = sched_clock();
 431
 432         spin_lock_irqsave(&blkg->stats_lock, flags);
 433         stats = &blkg->stats;
 434         if (time_after64(now, io_start_time))
 435                 blkio_add_stat(stats->stat_arr[BLKIO_STAT_SERVICE_TIME],
 436                                 now - io_start_time, direction, sync);
 437         if (time_after64(io_start_time, start_time))
 438                 blkio_add_stat(stats->stat_arr[BLKIO_STAT_WAIT_TIME],
 439                                 io_start_time - start_time, direction, sync);
 440         spin_unlock_irqrestore(&blkg->stats_lock, flags);
 441 }
 442 EXPORT_SYMBOL_GPL(blkiocg_update_completion_stats);
 443
 444 /*  Merged stats are per cpu.  */
 445 void blkiocg_update_io_merged_stats(struct blkio_group *blkg, bool direction,
 446                                         bool sync)
 447 {
 448         struct blkio_group_stats_cpu *stats_cpu;
 449         unsigned long flags;
 450
 451         /*
 452          * Disabling interrupts to provide mutual exclusion between two
 453          * writes on same cpu. It probably is not needed for 64bit. Not
 454          * optimizing that case yet.
 455          */
 456         local_irq_save(flags);
 457
 458         stats_cpu = this_cpu_ptr(blkg->stats_cpu);
 459
 460         u64_stats_update_begin(&stats_cpu->syncp);
 461         blkio_add_stat(stats_cpu->stat_arr_cpu[BLKIO_STAT_CPU_MERGED], 1,
 462                                 direction, sync);
 463         u64_stats_update_end(&stats_cpu->syncp);
 464         local_irq_restore(flags);
 465 }
 466 EXPORT_SYMBOL_GPL(blkiocg_update_io_merged_stats);
 467
 468 /*
 469  * This function allocates the per cpu stats for blkio_group. Should be called
 470  * from sleepable context as alloc_per_cpu() requires that.
 471  */
 472 int blkio_alloc_blkg_stats(struct blkio_group *blkg)
 473 {
 474         /* Allocate memory for per cpu stats */
 475         blkg->stats_cpu = alloc_percpu(struct blkio_group_stats_cpu);
 476         if (!blkg->stats_cpu)
 477                 return -ENOMEM;
 478         return 0;
 479 }
 480 EXPORT_SYMBOL_GPL(blkio_alloc_blkg_stats);
 481
 482 void blkiocg_add_blkio_group(struct blkio_cgroup *blkcg,
 483                 struct blkio_group *blkg, void *key, dev_t dev,
 484                 enum blkio_policy_id plid)
 485 {
 486         unsigned long flags;
 487
 488         spin_lock_irqsave(&blkcg->lock, flags);
 489         spin_lock_init(&blkg->stats_lock);
 490         rcu_assign_pointer(blkg->key, key);
 491         blkg->blkcg_id = css_id(&blkcg->css);
 492         hlist_add_head_rcu(&blkg->blkcg_node, &blkcg->blkg_list);
 493         blkg->plid = plid;
 494         spin_unlock_irqrestore(&blkcg->lock, flags);
 495         /* Need to take css reference ? */
 496         cgroup_path(blkcg->css.cgroup, blkg->path, sizeof(blkg->path));
 497         blkg->dev = dev;
 498 }
 499 EXPORT_SYMBOL_GPL(blkiocg_add_blkio_group);
 500
 501 static void __blkiocg_del_blkio_group(struct blkio_group *blkg)
 502 {
 503         hlist_del_init_rcu(&blkg->blkcg_node);
 504         blkg->blkcg_id = 0;
 505 }
 506
 507 /*
 508  * returns 0 if blkio_group was still on cgroup list. Otherwise returns 1
 509  * indicating that blk_group was unhashed by the time we got to it.
 510  */
 511 int blkiocg_del_blkio_group(struct blkio_group *blkg)
 512 {
 513         struct blkio_cgroup *blkcg;
 514         unsigned long flags;
 515         struct cgroup_subsys_state *css;
 516         int ret = 1;
 517
 518         rcu_read_lock();
 519         css = css_lookup(&blkio_subsys, blkg->blkcg_id);
 520         if (css) {
 521                 blkcg = container_of(css, struct blkio_cgroup, css);
 522                 spin_lock_irqsave(&blkcg->lock, flags);
 523                 if (!hlist_unhashed(&blkg->blkcg_node)) {
 524                         __blkiocg_del_blkio_group(blkg);
 525                         ret = 0;
 526                 }
 527                 spin_unlock_irqrestore(&blkcg->lock, flags);
 528         }
 529
 530         rcu_read_unlock();
 531         return ret;
 532 }
 533 EXPORT_SYMBOL_GPL(blkiocg_del_blkio_group);
 534
 535 /* called under rcu_read_lock(). */
 536 struct blkio_group *blkiocg_lookup_group(struct blkio_cgroup *blkcg, void *key)
 537 {
 538         struct blkio_group *blkg;
 539         struct hlist_node *n;
 540         void *__key;
 541
 542         hlist_for_each_entry_rcu(blkg, n, &blkcg->blkg_list, blkcg_node) {
 543                 __key = blkg->key;
 544                 if (__key == key)
 545                         return blkg;
 546         }
 547
 548         return NULL;
 549 }
 550 EXPORT_SYMBOL_GPL(blkiocg_lookup_group);
 551
 552 static void blkio_reset_stats_cpu(struct blkio_group *blkg)
 553 {
 554         struct blkio_group_stats_cpu *stats_cpu;
 555         int i, j, k;
 556         /*
 557          * Note: On 64 bit arch this should not be an issue. This has the
 558          * possibility of returning some inconsistent value on 32bit arch
 559          * as 64bit update on 32bit is non atomic. Taking care of this
 560          * corner case makes code very complicated, like sending IPIs to
 561          * cpus, taking care of stats of offline cpus etc.
 562          *
 563          * reset stats is anyway more of a debug feature and this sounds a
 564          * corner case. So I am not complicating the code yet until and
 565          * unless this becomes a real issue.
 566          */
 567         for_each_possible_cpu(i) {
 568                 stats_cpu = per_cpu_ptr(blkg->stats_cpu, i);
 569                 stats_cpu->sectors = 0;
 570                 for(j = 0; j < BLKIO_STAT_CPU_NR; j++)
 571                         for (k = 0; k < BLKIO_STAT_TOTAL; k++)
 572                                 stats_cpu->stat_arr_cpu[j][k] = 0;
 573         }
 574 }
 575
 576 static int
 577 blkiocg_reset_stats(struct cgroup *cgroup, struct cftype *cftype, u64 val)
 578 {
 579         struct blkio_cgroup *blkcg;
 580         struct blkio_group *blkg;
 581         struct blkio_group_stats *stats;
 582         struct hlist_node *n;
 583         uint64_t queued[BLKIO_STAT_TOTAL];
 584         int i;
 585 #ifdef CONFIG_DEBUG_BLK_CGROUP
 586         bool idling, waiting, empty;
 587         unsigned long long now = sched_clock();
 588 #endif
 589
 590         blkcg = cgroup_to_blkio_cgroup(cgroup);
 591         spin_lock_irq(&blkcg->lock);
 592         hlist_for_each_entry(blkg, n, &blkcg->blkg_list, blkcg_node) {
 593                 spin_lock(&blkg->stats_lock);
 594                 stats = &blkg->stats;
 595 #ifdef CONFIG_DEBUG_BLK_CGROUP
 596                 idling = blkio_blkg_idling(stats);
 597                 waiting = blkio_blkg_waiting(stats);
 598                 empty = blkio_blkg_empty(stats);
 599 #endif
 600                 for (i = 0; i < BLKIO_STAT_TOTAL; i++)
 601                         queued[i] = stats->stat_arr[BLKIO_STAT_QUEUED][i];
 602                 memset(stats, 0, sizeof(struct blkio_group_stats));
 603                 for (i = 0; i < BLKIO_STAT_TOTAL; i++)
 604                         stats->stat_arr[BLKIO_STAT_QUEUED][i] = queued[i];
 605 #ifdef CONFIG_DEBUG_BLK_CGROUP
 606                 if (idling) {
 607                         blkio_mark_blkg_idling(stats);
 608                         stats->start_idle_time = now;
 609                 }
 610                 if (waiting) {
 611                         blkio_mark_blkg_waiting(stats);
 612                         stats->start_group_wait_time = now;
 613                 }
 614                 if (empty) {
 615                         blkio_mark_blkg_empty(stats);
 616                         stats->start_empty_time = now;
 617                 }
 618 #endif
 619                 spin_unlock(&blkg->stats_lock);
 620
 621                 /* Reset Per cpu stats which don't take blkg->stats_lock */
 622                 blkio_reset_stats_cpu(blkg);
 623         }
 624
 625         spin_unlock_irq(&blkcg->lock);
 626         return 0;
 627 }
 628
 629 static void blkio_get_key_name(enum stat_sub_type type, dev_t dev, char *str,
 630                                 int chars_left, bool diskname_only)
 631 {
 632         snprintf(str, chars_left, "%d:%d", MAJOR(dev), MINOR(dev));
 633         chars_left -= strlen(str);
 634         if (chars_left <= 0) {
 635                 printk(KERN_WARNING
 636                         "Possibly incorrect cgroup stat display format");
 637                 return;
 638         }
 639         if (diskname_only)
 640                 return;
 641         switch (type) {
 642         case BLKIO_STAT_READ:
 643                 strlcat(str, " Read", chars_left);
 644                 break;
 645         case BLKIO_STAT_WRITE:
 646                 strlcat(str, " Write", chars_left);
 647                 break;
 648         case BLKIO_STAT_SYNC:
 649                 strlcat(str, " Sync", chars_left);
 650                 break;
 651         case BLKIO_STAT_ASYNC:
 652                 strlcat(str, " Async", chars_left);
 653                 break;
 654         case BLKIO_STAT_TOTAL:
 655                 strlcat(str, " Total", chars_left);
 656                 break;
 657         default:
 658                 strlcat(str, " Invalid", chars_left);
 659         }
 660 }
 661
 662 static uint64_t blkio_fill_stat(char *str, int chars_left, uint64_t val,
 663                                 struct cgroup_map_cb *cb, dev_t dev)
 664 {
 665         blkio_get_key_name(0, dev, str, chars_left, true);
 666         cb->fill(cb, str, val);
 667         return val;
 668 }
 669
 670
 671 static uint64_t blkio_read_stat_cpu(struct blkio_group *blkg,
 672                         enum stat_type_cpu type, enum stat_sub_type sub_type)
 673 {
 674         int cpu;
 675         struct blkio_group_stats_cpu *stats_cpu;
 676         u64 val = 0, tval;
 677
 678         for_each_possible_cpu(cpu) {
 679                 unsigned int start;
 680                 stats_cpu  = per_cpu_ptr(blkg->stats_cpu, cpu);
 681
 682                 do {
 683                         start = u64_stats_fetch_begin(&stats_cpu->syncp);
 684                         if (type == BLKIO_STAT_CPU_SECTORS)
 685                                 tval = stats_cpu->sectors;
 686                         else
 687                                 tval = stats_cpu->stat_arr_cpu[type][sub_type];
 688                 } while(u64_stats_fetch_retry(&stats_cpu->syncp, start));
 689
 690                 val += tval;
 691         }
 692
 693         return val;
 694 }
 695
 696 static uint64_t blkio_get_stat_cpu(struct blkio_group *blkg,
 697                 struct cgroup_map_cb *cb, dev_t dev, enum stat_type_cpu type)
 698 {
 699         uint64_t disk_total, val;
 700         char key_str[MAX_KEY_LEN];
 701         enum stat_sub_type sub_type;
 702
 703         if (type == BLKIO_STAT_CPU_SECTORS) {
 704                 val = blkio_read_stat_cpu(blkg, type, 0);
 705                 return blkio_fill_stat(key_str, MAX_KEY_LEN - 1, val, cb, dev);
 706         }
 707
 708         for (sub_type = BLKIO_STAT_READ; sub_type < BLKIO_STAT_TOTAL;
 709                         sub_type++) {
 710                 blkio_get_key_name(sub_type, dev, key_str, MAX_KEY_LEN, false);
 711                 val = blkio_read_stat_cpu(blkg, type, sub_type);
 712                 cb->fill(cb, key_str, val);
 713         }
 714
 715         disk_total = blkio_read_stat_cpu(blkg, type, BLKIO_STAT_READ) +
 716                         blkio_read_stat_cpu(blkg, type, BLKIO_STAT_WRITE);
 717
 718         blkio_get_key_name(BLKIO_STAT_TOTAL, dev, key_str, MAX_KEY_LEN, false);
 719         cb->fill(cb, key_str, disk_total);
 720         return disk_total;
 721 }
 722
 723 /* This should be called with blkg->stats_lock held */
 724 static uint64_t blkio_get_stat(struct blkio_group *blkg,
 725                 struct cgroup_map_cb *cb, dev_t dev, enum stat_type type)
 726 {
 727         uint64_t disk_total;
 728         char key_str[MAX_KEY_LEN];
 729         enum stat_sub_type sub_type;
 730
 731         if (type == BLKIO_STAT_TIME)
 732                 return blkio_fill_stat(key_str, MAX_KEY_LEN - 1,
 733                                         blkg->stats.time, cb, dev);
 734 #ifdef CONFIG_DEBUG_BLK_CGROUP
 735         if (type == BLKIO_STAT_UNACCOUNTED_TIME)
 736                 return blkio_fill_stat(key_str, MAX_KEY_LEN - 1,
 737                                         blkg->stats.unaccounted_time, cb, dev);
 738         if (type == BLKIO_STAT_AVG_QUEUE_SIZE) {
 739                 uint64_t sum = blkg->stats.avg_queue_size_sum;
 740                 uint64_t samples = blkg->stats.avg_queue_size_samples;
 741                 if (samples)
 742                         do_div(sum, samples);
 743                 else
 744                         sum = 0;
 745                 return blkio_fill_stat(key_str, MAX_KEY_LEN - 1, sum, cb, dev);
 746         }
 747         if (type == BLKIO_STAT_GROUP_WAIT_TIME)
 748                 return blkio_fill_stat(key_str, MAX_KEY_LEN - 1,
 749                                         blkg->stats.group_wait_time, cb, dev);
 750         if (type == BLKIO_STAT_IDLE_TIME)
 751                 return blkio_fill_stat(key_str, MAX_KEY_LEN - 1,
 752                                         blkg->stats.idle_time, cb, dev);
 753         if (type == BLKIO_STAT_EMPTY_TIME)
 754                 return blkio_fill_stat(key_str, MAX_KEY_LEN - 1,
 755                                         blkg->stats.empty_time, cb, dev);
 756         if (type == BLKIO_STAT_DEQUEUE)
 757                 return blkio_fill_stat(key_str, MAX_KEY_LEN - 1,
 758                                         blkg->stats.dequeue, cb, dev);
 759 #endif
 760
 761         for (sub_type = BLKIO_STAT_READ; sub_type < BLKIO_STAT_TOTAL;
 762                         sub_type++) {
 763                 blkio_get_key_name(sub_type, dev, key_str, MAX_KEY_LEN, false);
 764                 cb->fill(cb, key_str, blkg->stats.stat_arr[type][sub_type]);
 765         }
 766         disk_total = blkg->stats.stat_arr[type][BLKIO_STAT_READ] +
 767                         blkg->stats.stat_arr[type][BLKIO_STAT_WRITE];
 768         blkio_get_key_name(BLKIO_STAT_TOTAL, dev, key_str, MAX_KEY_LEN, false);
 769         cb->fill(cb, key_str, disk_total);
 770         return disk_total;
 771 }
 772
 773 static int blkio_check_dev_num(dev_t dev)
 774 {
 775         int part = 0;
 776         struct gendisk *disk;
 777
 778         disk = get_gendisk(dev, &part);
 779         if (!disk || part)
 780                 return -ENODEV;
 781
 782         return 0;
 783 }
 784
 785 static int blkio_policy_parse_and_set(char *buf,
 786         struct blkio_policy_node *newpn, enum blkio_policy_id plid, int fileid)
 787 {
 788         char *s[4], *p, *major_s = NULL, *minor_s = NULL;
 789         int ret;
 790         unsigned long major, minor, temp;
 791         int i = 0;
 792         dev_t dev;
 793         u64 bps, iops;
 794
 795         memset(s, 0, sizeof(s));
 796
 797         while ((p = strsep(&buf, " ")) != NULL) {
 798                 if (!*p)
 799                         continue;
 800
 801                 s[i++] = p;
 802
 803                 /* Prevent from inputing too many things */
 804                 if (i == 3)
 805                         break;
 806         }
 807
 808         if (i != 2)
 809                 return -EINVAL;
 810
 811         p = strsep(&s[0], ":");
 812         if (p != NULL)
 813                 major_s = p;
 814         else
 815                 return -EINVAL;
 816
 817         minor_s = s[0];
 818         if (!minor_s)
 819                 return -EINVAL;
 820
 821         ret = strict_strtoul(major_s, 10, &major);
 822         if (ret)
 823                 return -EINVAL;
 824
 825         ret = strict_strtoul(minor_s, 10, &minor);
 826         if (ret)
 827                 return -EINVAL;
 828
 829         dev = MKDEV(major, minor);
 830
 831         ret = blkio_check_dev_num(dev);
 832         if (ret)
 833                 return ret;
 834
 835         newpn->dev = dev;
 836
 837         if (s[1] == NULL)
 838                 return -EINVAL;
 839
 840         switch (plid) {
 841         case BLKIO_POLICY_PROP:
 842                 ret = strict_strtoul(s[1], 10, &temp);
 843                 if (ret || (temp < BLKIO_WEIGHT_MIN && temp > 0) ||
 844                         temp > BLKIO_WEIGHT_MAX)
 845                         return -EINVAL;
 846
 847                 newpn->plid = plid;
 848                 newpn->fileid = fileid;
 849                 newpn->val.weight = temp;
 850                 break;
 851         case BLKIO_POLICY_THROTL:
 852                 switch(fileid) {
 853                 case BLKIO_THROTL_read_bps_device:
 854                 case BLKIO_THROTL_write_bps_device:
 855                         ret = strict_strtoull(s[1], 10, &bps);
 856                         if (ret)
 857                                 return -EINVAL;
 858
 859                         newpn->plid = plid;
 860                         newpn->fileid = fileid;
 861                         newpn->val.bps = bps;
 862                         break;
 863                 case BLKIO_THROTL_read_iops_device:
 864                 case BLKIO_THROTL_write_iops_device:
 865                         ret = strict_strtoull(s[1], 10, &iops);
 866                         if (ret)
 867                                 return -EINVAL;
 868
 869                         if (iops > THROTL_IOPS_MAX)
 870                                 return -EINVAL;
 871
 872                         newpn->plid = plid;
 873                         newpn->fileid = fileid;
 874                         newpn->val.iops = (unsigned int)iops;
 875                         break;
 876                 }
 877                 break;
 878         default:
 879                 BUG();
 880         }
 881
 882         return 0;
 883 }
 884
 885 unsigned int blkcg_get_weight(struct blkio_cgroup *blkcg,
 886                               dev_t dev)
 887 {
 888         struct blkio_policy_node *pn;
 889
 890         pn = blkio_policy_search_node(blkcg, dev, BLKIO_POLICY_PROP,
 891                                 BLKIO_PROP_weight_device);
 892         if (pn)
 893                 return pn->val.weight;
 894         else
 895                 return blkcg->weight;
 896 }
 897 EXPORT_SYMBOL_GPL(blkcg_get_weight);
 898
 899 uint64_t blkcg_get_read_bps(struct blkio_cgroup *blkcg, dev_t dev)
 900 {
 901         struct blkio_policy_node *pn;
 902
 903         pn = blkio_policy_search_node(blkcg, dev, BLKIO_POLICY_THROTL,
 904                                 BLKIO_THROTL_read_bps_device);
 905         if (pn)
 906                 return pn->val.bps;
 907         else
 908                 return -1;
 909 }
 910
 911 uint64_t blkcg_get_write_bps(struct blkio_cgroup *blkcg, dev_t dev)
 912 {
 913         struct blkio_policy_node *pn;
 914         pn = blkio_policy_search_node(blkcg, dev, BLKIO_POLICY_THROTL,
 915                                 BLKIO_THROTL_write_bps_device);
 916         if (pn)
 917                 return pn->val.bps;
 918         else
 919                 return -1;
 920 }
 921
 922 unsigned int blkcg_get_read_iops(struct blkio_cgroup *blkcg, dev_t dev)
 923 {
 924         struct blkio_policy_node *pn;
 925
 926         pn = blkio_policy_search_node(blkcg, dev, BLKIO_POLICY_THROTL,
 927                                 BLKIO_THROTL_read_iops_device);
 928         if (pn)
 929                 return pn->val.iops;
 930         else
 931                 return -1;
 932 }
 933
 934 unsigned int blkcg_get_write_iops(struct blkio_cgroup *blkcg, dev_t dev)
 935 {
 936         struct blkio_policy_node *pn;
 937         pn = blkio_policy_search_node(blkcg, dev, BLKIO_POLICY_THROTL,
 938                                 BLKIO_THROTL_write_iops_device);
 939         if (pn)
 940                 return pn->val.iops;
 941         else
 942                 return -1;
 943 }
 944
 945 /* Checks whether user asked for deleting a policy rule */
 946 static bool blkio_delete_rule_command(struct blkio_policy_node *pn)
 947 {
 948         switch(pn->plid) {
 949         case BLKIO_POLICY_PROP:
 950                 if (pn->val.weight == 0)
 951                         return 1;
 952                 break;
 953         case BLKIO_POLICY_THROTL:
 954                 switch(pn->fileid) {
 955                 case BLKIO_THROTL_read_bps_device:
 956                 case BLKIO_THROTL_write_bps_device:
 957                         if (pn->val.bps == 0)
 958                                 return 1;
 959                         break;
 960                 case BLKIO_THROTL_read_iops_device:
 961                 case BLKIO_THROTL_write_iops_device:
 962                         if (pn->val.iops == 0)
 963                                 return 1;
 964                 }
 965                 break;
 966         default:
 967                 BUG();
 968         }
 969
 970         return 0;
 971 }
 972
 973 static void blkio_update_policy_rule(struct blkio_policy_node *oldpn,
 974                                         struct blkio_policy_node *newpn)
 975 {
 976         switch(oldpn->plid) {
 977         case BLKIO_POLICY_PROP:
 978                 oldpn->val.weight = newpn->val.weight;
 979                 break;
 980         case BLKIO_POLICY_THROTL:
 981                 switch(newpn->fileid) {
 982                 case BLKIO_THROTL_read_bps_device:
 983                 case BLKIO_THROTL_write_bps_device:
 984                         oldpn->val.bps = newpn->val.bps;
 985                         break;
 986                 case BLKIO_THROTL_read_iops_device:
 987                 case BLKIO_THROTL_write_iops_device:
 988                         oldpn->val.iops = newpn->val.iops;
 989                 }
 990                 break;
 991         default:
 992                 BUG();
 993         }
 994 }
 995
 996 /*
 997  * Some rules/values in blkg have changed. Propagate those to respective
 998  * policies.
 999  */
1000 static void blkio_update_blkg_policy(struct blkio_cgroup *blkcg,
1001                 struct blkio_group *blkg, struct blkio_policy_node *pn)
1002 {
1003         unsigned int weight, iops;
1004         u64 bps;
1005
1006         switch(pn->plid) {
1007         case BLKIO_POLICY_PROP:
1008                 weight = pn->val.weight ? pn->val.weight :
1009                                 blkcg->weight;
1010                 blkio_update_group_weight(blkg, weight);
1011                 break;
1012         case BLKIO_POLICY_THROTL:
1013                 switch(pn->fileid) {
1014                 case BLKIO_THROTL_read_bps_device:
1015                 case BLKIO_THROTL_write_bps_device:
1016                         bps = pn->val.bps ? pn->val.bps : (-1);
1017                         blkio_update_group_bps(blkg, bps, pn->fileid);
1018                         break;
1019                 case BLKIO_THROTL_read_iops_device:
1020                 case BLKIO_THROTL_write_iops_device:
1021                         iops = pn->val.iops ? pn->val.iops : (-1);
1022                         blkio_update_group_iops(blkg, iops, pn->fileid);
1023                         break;
1024                 }
1025                 break;
1026         default:
1027                 BUG();
1028         }
1029 }
1030
1031 /*
1032  * A policy node rule has been updated. Propagate this update to all the
1033  * block groups which might be affected by this update.
1034  */
1035 static void blkio_update_policy_node_blkg(struct blkio_cgroup *blkcg,
1036                                 struct blkio_policy_node *pn)
1037 {
1038         struct blkio_group *blkg;
1039         struct hlist_node *n;
1040
1041         spin_lock(&blkio_list_lock);
1042         spin_lock_irq(&blkcg->lock);
1043
1044         hlist_for_each_entry(blkg, n, &blkcg->blkg_list, blkcg_node) {
1045                 if (pn->dev != blkg->dev || pn->plid != blkg->plid)
1046                         continue;
1047                 blkio_update_blkg_policy(blkcg, blkg, pn);
1048         }
1049
1050         spin_unlock_irq(&blkcg->lock);
1051         spin_unlock(&blkio_list_lock);
1052 }
1053
1054 static int blkiocg_file_write(struct cgroup *cgrp, struct cftype *cft,
1055                                        const char *buffer)
1056 {
1057         int ret = 0;
1058         char *buf;
1059         struct blkio_policy_node *newpn, *pn;
1060         struct blkio_cgroup *blkcg;
1061         int keep_newpn = 0;
1062         enum blkio_policy_id plid = BLKIOFILE_POLICY(cft->private);
1063         int fileid = BLKIOFILE_ATTR(cft->private);
1064
1065         buf = kstrdup(buffer, GFP_KERNEL);
1066         if (!buf)
1067                 return -ENOMEM;
1068
1069         newpn = kzalloc(sizeof(*newpn), GFP_KERNEL);
1070         if (!newpn) {
1071                 ret = -ENOMEM;
1072                 goto free_buf;
1073         }
1074
1075         ret = blkio_policy_parse_and_set(buf, newpn, plid, fileid);
1076         if (ret)
1077                 goto free_newpn;
1078
1079         blkcg = cgroup_to_blkio_cgroup(cgrp);
1080
1081         spin_lock_irq(&blkcg->lock);
1082
1083         pn = blkio_policy_search_node(blkcg, newpn->dev, plid, fileid);
1084         if (!pn) {
1085                 if (!blkio_delete_rule_command(newpn)) {
1086                         blkio_policy_insert_node(blkcg, newpn);
1087                         keep_newpn = 1;
1088                 }
1089                 spin_unlock_irq(&blkcg->lock);
1090                 goto update_io_group;
1091         }
1092
1093         if (blkio_delete_rule_command(newpn)) {
1094                 blkio_policy_delete_node(pn);
1095                 spin_unlock_irq(&blkcg->lock);
1096                 goto update_io_group;
1097         }
1098         spin_unlock_irq(&blkcg->lock);
1099
1100         blkio_update_policy_rule(pn, newpn);
1101
1102 update_io_group:
1103         blkio_update_policy_node_blkg(blkcg, newpn);
1104
1105 free_newpn:
1106         if (!keep_newpn)
1107                 kfree(newpn);
1108 free_buf:
1109         kfree(buf);
1110         return ret;
1111 }
1112
1113 static void
1114 blkio_print_policy_node(struct seq_file *m, struct blkio_policy_node *pn)
1115 {
1116         switch(pn->plid) {
1117                 case BLKIO_POLICY_PROP:
1118                         if (pn->fileid == BLKIO_PROP_weight_device)
1119                                 seq_printf(m, "%u:%u\t%u\n", MAJOR(pn->dev),
1120                                         MINOR(pn->dev), pn->val.weight);
1121                         break;
1122                 case BLKIO_POLICY_THROTL:
1123                         switch(pn->fileid) {
1124                         case BLKIO_THROTL_read_bps_device:
1125                         case BLKIO_THROTL_write_bps_device:
1126                                 seq_printf(m, "%u:%u\t%llu\n", MAJOR(pn->dev),
1127                                         MINOR(pn->dev), pn->val.bps);
1128                                 break;
1129                         case BLKIO_THROTL_read_iops_device:
1130                         case BLKIO_THROTL_write_iops_device:
1131                                 seq_printf(m, "%u:%u\t%u\n", MAJOR(pn->dev),
1132                                         MINOR(pn->dev), pn->val.iops);
1133                                 break;
1134                         }
1135                         break;
1136                 default:
1137                         BUG();
1138         }
1139 }
1140
1141 /* cgroup files which read their data from policy nodes end up here */
1142 static void blkio_read_policy_node_files(struct cftype *cft,
1143                         struct blkio_cgroup *blkcg, struct seq_file *m)
1144 {
1145         struct blkio_policy_node *pn;
1146
1147         if (!list_empty(&blkcg->policy_list)) {
1148                 spin_lock_irq(&blkcg->lock);
1149                 list_for_each_entry(pn, &blkcg->policy_list, node) {
1150                         if (!pn_matches_cftype(cft, pn))
1151                                 continue;
1152                         blkio_print_policy_node(m, pn);
1153                 }
1154                 spin_unlock_irq(&blkcg->lock);
1155         }
1156 }
1157
1158 static int blkiocg_file_read(struct cgroup *cgrp, struct cftype *cft,
1159                                 struct seq_file *m)
1160 {
1161         struct blkio_cgroup *blkcg;
1162         enum blkio_policy_id plid = BLKIOFILE_POLICY(cft->private);
1163         int name = BLKIOFILE_ATTR(cft->private);
1164
1165         blkcg = cgroup_to_blkio_cgroup(cgrp);
1166
1167         switch(plid) {
1168         case BLKIO_POLICY_PROP:
1169                 switch(name) {
1170                 case BLKIO_PROP_weight_device:
1171                         blkio_read_policy_node_files(cft, blkcg, m);
1172                         return 0;
1173                 default:
1174                         BUG();
1175                 }
1176                 break;
1177         case BLKIO_POLICY_THROTL:
1178                 switch(name){
1179                 case BLKIO_THROTL_read_bps_device:
1180                 case BLKIO_THROTL_write_bps_device:
1181                 case BLKIO_THROTL_read_iops_device:
1182                 case BLKIO_THROTL_write_iops_device:
1183                         blkio_read_policy_node_files(cft, blkcg, m);
1184                         return 0;
1185                 default:
1186                         BUG();
1187                 }
1188                 break;
1189         default:
1190                 BUG();
1191         }
1192
1193         return 0;
1194 }
1195
1196 static int blkio_read_blkg_stats(struct blkio_cgroup *blkcg,
1197                 struct cftype *cft, struct cgroup_map_cb *cb,
1198                 enum stat_type type, bool show_total, bool pcpu)
1199 {
1200         struct blkio_group *blkg;
1201         struct hlist_node *n;
1202         uint64_t cgroup_total = 0;
1203
1204         rcu_read_lock();
1205         hlist_for_each_entry_rcu(blkg, n, &blkcg->blkg_list, blkcg_node) {
1206                 if (blkg->dev) {
1207                         if (!cftype_blkg_same_policy(cft, blkg))
1208                                 continue;
1209                         if (pcpu)
1210                                 cgroup_total += blkio_get_stat_cpu(blkg, cb,
1211                                                 blkg->dev, type);
1212                         else {
1213                                 spin_lock_irq(&blkg->stats_lock);
1214                                 cgroup_total += blkio_get_stat(blkg, cb,
1215                                                 blkg->dev, type);
1216                                 spin_unlock_irq(&blkg->stats_lock);
1217                         }
1218                 }
1219         }
1220         if (show_total)
1221                 cb->fill(cb, "Total", cgroup_total);
1222         rcu_read_unlock();
1223         return 0;
1224 }
1225
1226 /* All map kind of cgroup file get serviced by this function */
1227 static int blkiocg_file_read_map(struct cgroup *cgrp, struct cftype *cft,
1228                                 struct cgroup_map_cb *cb)
1229 {
1230         struct blkio_cgroup *blkcg;
1231         enum blkio_policy_id plid = BLKIOFILE_POLICY(cft->private);
1232         int name = BLKIOFILE_ATTR(cft->private);
1233
1234         blkcg = cgroup_to_blkio_cgroup(cgrp);
1235
1236         switch(plid) {
1237         case BLKIO_POLICY_PROP:
1238                 switch(name) {
1239                 case BLKIO_PROP_time:
1240                         return blkio_read_blkg_stats(blkcg, cft, cb,
1241                                                 BLKIO_STAT_TIME, 0, 0);
1242                 case BLKIO_PROP_sectors:
1243                         return blkio_read_blkg_stats(blkcg, cft, cb,
1244                                                 BLKIO_STAT_CPU_SECTORS, 0, 1);
1245                 case BLKIO_PROP_io_service_bytes:
1246                         return blkio_read_blkg_stats(blkcg, cft, cb,
1247                                         BLKIO_STAT_CPU_SERVICE_BYTES, 1, 1);
1248                 case BLKIO_PROP_io_serviced:
1249                         return blkio_read_blkg_stats(blkcg, cft, cb,
1250                                                 BLKIO_STAT_CPU_SERVICED, 1, 1);
1251                 case BLKIO_PROP_io_service_time:
1252                         return blkio_read_blkg_stats(blkcg, cft, cb,
1253                                                 BLKIO_STAT_SERVICE_TIME, 1, 0);
1254                 case BLKIO_PROP_io_wait_time:
1255                         return blkio_read_blkg_stats(blkcg, cft, cb,
1256                                                 BLKIO_STAT_WAIT_TIME, 1, 0);
1257                 case BLKIO_PROP_io_merged:
1258                         return blkio_read_blkg_stats(blkcg, cft, cb,
1259                                                 BLKIO_STAT_CPU_MERGED, 1, 1);
1260                 case BLKIO_PROP_io_queued:
1261                         return blkio_read_blkg_stats(blkcg, cft, cb,
1262                                                 BLKIO_STAT_QUEUED, 1, 0);
1263 #ifdef CONFIG_DEBUG_BLK_CGROUP
1264                 case BLKIO_PROP_unaccounted_time:
1265                         return blkio_read_blkg_stats(blkcg, cft, cb,
1266                                         BLKIO_STAT_UNACCOUNTED_TIME, 0, 0);
1267                 case BLKIO_PROP_dequeue:
1268                         return blkio_read_blkg_stats(blkcg, cft, cb,
1269                                                 BLKIO_STAT_DEQUEUE, 0, 0);
1270                 case BLKIO_PROP_avg_queue_size:
1271                         return blkio_read_blkg_stats(blkcg, cft, cb,
1272                                         BLKIO_STAT_AVG_QUEUE_SIZE, 0, 0);
1273                 case BLKIO_PROP_group_wait_time:
1274                         return blkio_read_blkg_stats(blkcg, cft, cb,
1275                                         BLKIO_STAT_GROUP_WAIT_TIME, 0, 0);
1276                 case BLKIO_PROP_idle_time:
1277                         return blkio_read_blkg_stats(blkcg, cft, cb,
1278                                                 BLKIO_STAT_IDLE_TIME, 0, 0);
1279                 case BLKIO_PROP_empty_time:
1280                         return blkio_read_blkg_stats(blkcg, cft, cb,
1281                                                 BLKIO_STAT_EMPTY_TIME, 0, 0);
1282 #endif
1283                 default:
1284                         BUG();
1285                 }
1286                 break;
1287         case BLKIO_POLICY_THROTL:
1288                 switch(name){
1289                 case BLKIO_THROTL_io_service_bytes:
1290                         return blkio_read_blkg_stats(blkcg, cft, cb,
1291                                                 BLKIO_STAT_CPU_SERVICE_BYTES, 1, 1);
1292                 case BLKIO_THROTL_io_serviced:
1293                         return blkio_read_blkg_stats(blkcg, cft, cb,
1294                                                 BLKIO_STAT_CPU_SERVICED, 1, 1);
1295                 default:
1296                         BUG();
1297                 }
1298                 break;
1299         default:
1300                 BUG();
1301         }
1302
1303         return 0;
1304 }
1305
1306 static int blkio_weight_write(struct blkio_cgroup *blkcg, u64 val)
1307 {
1308         struct blkio_group *blkg;
1309         struct hlist_node *n;
1310         struct blkio_policy_node *pn;
1311
1312         if (val < BLKIO_WEIGHT_MIN || val > BLKIO_WEIGHT_MAX)
1313                 return -EINVAL;
1314
1315         spin_lock(&blkio_list_lock);
1316         spin_lock_irq(&blkcg->lock);
1317         blkcg->weight = (unsigned int)val;
1318
1319         hlist_for_each_entry(blkg, n, &blkcg->blkg_list, blkcg_node) {
1320                 pn = blkio_policy_search_node(blkcg, blkg->dev,
1321                                 BLKIO_POLICY_PROP, BLKIO_PROP_weight_device);
1322                 if (pn)
1323                         continue;
1324
1325                 blkio_update_group_weight(blkg, blkcg->weight);
1326         }
1327         spin_unlock_irq(&blkcg->lock);
1328         spin_unlock(&blkio_list_lock);
1329         return 0;
1330 }
1331
1332 static u64 blkiocg_file_read_u64 (struct cgroup *cgrp, struct cftype *cft) {
1333         struct blkio_cgroup *blkcg;
1334         enum blkio_policy_id plid = BLKIOFILE_POLICY(cft->private);
1335         int name = BLKIOFILE_ATTR(cft->private);
1336
1337         blkcg = cgroup_to_blkio_cgroup(cgrp);
1338
1339         switch(plid) {
1340         case BLKIO_POLICY_PROP:
1341                 switch(name) {
1342                 case BLKIO_PROP_weight:
1343                         return (u64)blkcg->weight;
1344                 }
1345                 break;
1346         default:
1347                 BUG();
1348         }
1349         return 0;
1350 }
1351
1352 static int
1353 blkiocg_file_write_u64(struct cgroup *cgrp, struct cftype *cft, u64 val)
1354 {
1355         struct blkio_cgroup *blkcg;
1356         enum blkio_policy_id plid = BLKIOFILE_POLICY(cft->private);
1357         int name = BLKIOFILE_ATTR(cft->private);
1358
1359         blkcg = cgroup_to_blkio_cgroup(cgrp);
1360
1361         switch(plid) {
1362         case BLKIO_POLICY_PROP:
1363                 switch(name) {
1364                 case BLKIO_PROP_weight:
1365                         return blkio_weight_write(blkcg, val);
1366                 }
1367                 break;
1368         default:
1369                 BUG();
1370         }
1371
1372         return 0;
1373 }
1374
1375 struct cftype blkio_files[] = {
1376         {
1377                 .name = "weight_device",
1378                 .private = BLKIOFILE_PRIVATE(BLKIO_POLICY_PROP,
1379                                 BLKIO_PROP_weight_device),
1380                 .read_seq_string = blkiocg_file_read,
1381                 .write_string = blkiocg_file_write,
1382                 .max_write_len = 256,
1383         },
1384         {
1385                 .name = "weight",
1386                 .private = BLKIOFILE_PRIVATE(BLKIO_POLICY_PROP,
1387                                 BLKIO_PROP_weight),
1388                 .read_u64 = blkiocg_file_read_u64,
1389                 .write_u64 = blkiocg_file_write_u64,
1390         },
1391         {
1392                 .name = "time",
1393                 .private = BLKIOFILE_PRIVATE(BLKIO_POLICY_PROP,
1394                                 BLKIO_PROP_time),
1395                 .read_map = blkiocg_file_read_map,
1396         },
1397         {
1398                 .name = "sectors",
1399                 .private = BLKIOFILE_PRIVATE(BLKIO_POLICY_PROP,
1400                                 BLKIO_PROP_sectors),
1401                 .read_map = blkiocg_file_read_map,
1402         },
1403         {
1404                 .name = "io_service_bytes",
1405                 .private = BLKIOFILE_PRIVATE(BLKIO_POLICY_PROP,
1406                                 BLKIO_PROP_io_service_bytes),
1407                 .read_map = blkiocg_file_read_map,
1408         },
1409         {
1410                 .name = "io_serviced",
1411                 .private = BLKIOFILE_PRIVATE(BLKIO_POLICY_PROP,
1412                                 BLKIO_PROP_io_serviced),
1413                 .read_map = blkiocg_file_read_map,
1414         },
1415         {
1416                 .name = "io_service_time",
1417                 .private = BLKIOFILE_PRIVATE(BLKIO_POLICY_PROP,
1418                                 BLKIO_PROP_io_service_time),
1419                 .read_map = blkiocg_file_read_map,
1420         },
1421         {
1422                 .name = "io_wait_time",
1423                 .private = BLKIOFILE_PRIVATE(BLKIO_POLICY_PROP,
1424                                 BLKIO_PROP_io_wait_time),
1425                 .read_map = blkiocg_file_read_map,
1426         },
1427         {
1428                 .name = "io_merged",
1429                 .private = BLKIOFILE_PRIVATE(BLKIO_POLICY_PROP,
1430                                 BLKIO_PROP_io_merged),
1431                 .read_map = blkiocg_file_read_map,
1432         },
1433         {
1434                 .name = "io_queued",
1435                 .private = BLKIOFILE_PRIVATE(BLKIO_POLICY_PROP,
1436                                 BLKIO_PROP_io_queued),
1437                 .read_map = blkiocg_file_read_map,
1438         },
1439         {
1440                 .name = "reset_stats",
1441                 .write_u64 = blkiocg_reset_stats,
1442         },
1443 #ifdef CONFIG_BLK_DEV_THROTTLING
1444         {
1445                 .name = "throttle.read_bps_device",
1446                 .private = BLKIOFILE_PRIVATE(BLKIO_POLICY_THROTL,
1447                                 BLKIO_THROTL_read_bps_device),
1448                 .read_seq_string = blkiocg_file_read,
1449                 .write_string = blkiocg_file_write,
1450                 .max_write_len = 256,
1451         },
1452
1453         {
1454                 .name = "throttle.write_bps_device",
1455                 .private = BLKIOFILE_PRIVATE(BLKIO_POLICY_THROTL,
1456                                 BLKIO_THROTL_write_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.read_iops_device",
1464                 .private = BLKIOFILE_PRIVATE(BLKIO_POLICY_THROTL,
1465                                 BLKIO_THROTL_read_iops_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.write_iops_device",
1473                 .private = BLKIOFILE_PRIVATE(BLKIO_POLICY_THROTL,
1474                                 BLKIO_THROTL_write_iops_device),
1475                 .read_seq_string = blkiocg_file_read,
1476                 .write_string = blkiocg_file_write,
1477                 .max_write_len = 256,
1478         },
1479         {
1480                 .name = "throttle.io_service_bytes",
1481                 .private = BLKIOFILE_PRIVATE(BLKIO_POLICY_THROTL,
1482                                 BLKIO_THROTL_io_service_bytes),
1483                 .read_map = blkiocg_file_read_map,
1484         },
1485         {
1486                 .name = "throttle.io_serviced",
1487                 .private = BLKIOFILE_PRIVATE(BLKIO_POLICY_THROTL,
1488                                 BLKIO_THROTL_io_serviced),
1489                 .read_map = blkiocg_file_read_map,
1490         },
1491 #endif /* CONFIG_BLK_DEV_THROTTLING */
1492
1493 #ifdef CONFIG_DEBUG_BLK_CGROUP
1494         {
1495                 .name = "avg_queue_size",
1496                 .private = BLKIOFILE_PRIVATE(BLKIO_POLICY_PROP,
1497                                 BLKIO_PROP_avg_queue_size),
1498                 .read_map = blkiocg_file_read_map,
1499         },
1500         {
1501                 .name = "group_wait_time",
1502                 .private = BLKIOFILE_PRIVATE(BLKIO_POLICY_PROP,
1503                                 BLKIO_PROP_group_wait_time),
1504                 .read_map = blkiocg_file_read_map,
1505         },
1506         {
1507                 .name = "idle_time",
1508                 .private = BLKIOFILE_PRIVATE(BLKIO_POLICY_PROP,
1509                                 BLKIO_PROP_idle_time),
1510                 .read_map = blkiocg_file_read_map,
1511         },
1512         {
1513                 .name = "empty_time",
1514                 .private = BLKIOFILE_PRIVATE(BLKIO_POLICY_PROP,
1515                                 BLKIO_PROP_empty_time),
1516                 .read_map = blkiocg_file_read_map,
1517         },
1518         {
1519                 .name = "dequeue",
1520                 .private = BLKIOFILE_PRIVATE(BLKIO_POLICY_PROP,
1521                                 BLKIO_PROP_dequeue),
1522                 .read_map = blkiocg_file_read_map,
1523         },
1524         {
1525                 .name = "unaccounted_time",
1526                 .private = BLKIOFILE_PRIVATE(BLKIO_POLICY_PROP,
1527                                 BLKIO_PROP_unaccounted_time),
1528                 .read_map = blkiocg_file_read_map,
1529         },
1530 #endif
1531 };
1532
1533 static int blkiocg_populate(struct cgroup_subsys *subsys, struct cgroup *cgroup)
1534 {
1535         return cgroup_add_files(cgroup, subsys, blkio_files,
1536                                 ARRAY_SIZE(blkio_files));
1537 }
1538
1539 static void blkiocg_destroy(struct cgroup_subsys *subsys, struct cgroup *cgroup)
1540 {
1541         struct blkio_cgroup *blkcg = cgroup_to_blkio_cgroup(cgroup);
1542         unsigned long flags;
1543         struct blkio_group *blkg;
1544         void *key;
1545         struct blkio_policy_type *blkiop;
1546         struct blkio_policy_node *pn, *pntmp;
1547
1548         rcu_read_lock();
1549         do {
1550                 spin_lock_irqsave(&blkcg->lock, flags);
1551
1552                 if (hlist_empty(&blkcg->blkg_list)) {
1553                         spin_unlock_irqrestore(&blkcg->lock, flags);
1554                         break;
1555                 }
1556
1557                 blkg = hlist_entry(blkcg->blkg_list.first, struct blkio_group,
1558                                         blkcg_node);
1559                 key = rcu_dereference(blkg->key);
1560                 __blkiocg_del_blkio_group(blkg);
1561
1562                 spin_unlock_irqrestore(&blkcg->lock, flags);
1563
1564                 /*
1565                  * This blkio_group is being unlinked as associated cgroup is
1566                  * going away. Let all the IO controlling policies know about
1567                  * this event.
1568                  */
1569                 spin_lock(&blkio_list_lock);
1570                 list_for_each_entry(blkiop, &blkio_list, list) {
1571                         if (blkiop->plid != blkg->plid)
1572                                 continue;
1573                         blkiop->ops.blkio_unlink_group_fn(key, blkg);
1574                 }
1575                 spin_unlock(&blkio_list_lock);
1576         } while (1);
1577
1578         list_for_each_entry_safe(pn, pntmp, &blkcg->policy_list, node) {
1579                 blkio_policy_delete_node(pn);
1580                 kfree(pn);
1581         }
1582
1583         free_css_id(&blkio_subsys, &blkcg->css);
1584         rcu_read_unlock();
1585         if (blkcg != &blkio_root_cgroup)
1586                 kfree(blkcg);
1587 }
1588
1589 static struct cgroup_subsys_state *
1590 blkiocg_create(struct cgroup_subsys *subsys, struct cgroup *cgroup)
1591 {
1592         struct blkio_cgroup *blkcg;
1593         struct cgroup *parent = cgroup->parent;
1594
1595         if (!parent) {
1596                 blkcg = &blkio_root_cgroup;
1597                 goto done;
1598         }
1599
1600         blkcg = kzalloc(sizeof(*blkcg), GFP_KERNEL);
1601         if (!blkcg)
1602                 return ERR_PTR(-ENOMEM);
1603
1604         blkcg->weight = BLKIO_WEIGHT_DEFAULT;
1605 done:
1606         spin_lock_init(&blkcg->lock);
1607         INIT_HLIST_HEAD(&blkcg->blkg_list);
1608
1609         INIT_LIST_HEAD(&blkcg->policy_list);
1610         return &blkcg->css;
1611 }
1612
1613 /*
1614  * We cannot support shared io contexts, as we have no mean to support
1615  * two tasks with the same ioc in two different groups without major rework
1616  * of the main cic data structures.  For now we allow a task to change
1617  * its cgroup only if it's the only owner of its ioc.
1618  */
1619 static int blkiocg_can_attach(struct cgroup_subsys *subsys,
1620                                 struct cgroup *cgroup, struct task_struct *tsk,
1621                                 bool threadgroup)
1622 {
1623         struct io_context *ioc;
1624         int ret = 0;
1625
1626         /* task_lock() is needed to avoid races with exit_io_context() */
1627         task_lock(tsk);
1628         ioc = tsk->io_context;
1629         if (ioc && atomic_read(&ioc->nr_tasks) > 1)
1630                 ret = -EINVAL;
1631         task_unlock(tsk);
1632
1633         return ret;
1634 }
1635
1636 static void blkiocg_attach(struct cgroup_subsys *subsys, struct cgroup *cgroup,
1637                                 struct cgroup *prev, struct task_struct *tsk,
1638                                 bool threadgroup)
1639 {
1640         struct io_context *ioc;
1641
1642         task_lock(tsk);
1643         ioc = tsk->io_context;
1644         if (ioc)
1645                 ioc->cgroup_changed = 1;
1646         task_unlock(tsk);
1647 }
1648
1649 void blkio_policy_register(struct blkio_policy_type *blkiop)
1650 {
1651         spin_lock(&blkio_list_lock);
1652         list_add_tail(&blkiop->list, &blkio_list);
1653         spin_unlock(&blkio_list_lock);
1654 }
1655 EXPORT_SYMBOL_GPL(blkio_policy_register);
1656
1657 void blkio_policy_unregister(struct blkio_policy_type *blkiop)
1658 {
1659         spin_lock(&blkio_list_lock);
1660         list_del_init(&blkiop->list);
1661         spin_unlock(&blkio_list_lock);
1662 }
1663 EXPORT_SYMBOL_GPL(blkio_policy_unregister);
1664
1665 static int __init init_cgroup_blkio(void)
1666 {
1667         return cgroup_load_subsys(&blkio_subsys);
1668 }
1669
1670 static void __exit exit_cgroup_blkio(void)
1671 {
1672         cgroup_unload_subsys(&blkio_subsys);
1673 }
1674
1675 module_init(init_cgroup_blkio);
1676 module_exit(exit_cgroup_blkio);
1677 MODULE_LICENSE("GPL");