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/kdev_t.h>
  15 #include <linux/module.h>
  16 #include <linux/err.h>
  17 #include <linux/blkdev.h>
  18 #include <linux/slab.h>
  19 #include <linux/genhd.h>
  20 #include <linux/delay.h>
  21 #include <linux/atomic.h>
  22 #include "blk-cgroup.h"
  23 #include "blk.h"
  24
  25 #define MAX_KEY_LEN 100
  26
  27 static DEFINE_MUTEX(blkcg_pol_mutex);
  28
  29 struct blkcg blkcg_root = { .cfq_weight = 2 * CFQ_WEIGHT_DEFAULT };
  30 EXPORT_SYMBOL_GPL(blkcg_root);
  31
  32 static struct blkcg_policy *blkcg_policy[BLKCG_MAX_POLS];
  33
  34 static bool blkcg_policy_enabled(struct request_queue *q,
  35                                  const struct blkcg_policy *pol)
  36 {
  37         return pol && test_bit(pol->plid, q->blkcg_pols);
  38 }
  39
  40 /**
  41  * blkg_free - free a blkg
  42  * @blkg: blkg to free
  43  *
  44  * Free @blkg which may be partially allocated.
  45  */
  46 static void blkg_free(struct blkcg_gq *blkg)
  47 {
  48         int i;
  49
  50         if (!blkg)
  51                 return;
  52
  53         for (i = 0; i < BLKCG_MAX_POLS; i++) {
  54                 struct blkcg_policy *pol = blkcg_policy[i];
  55                 struct blkg_policy_data *pd = blkg->pd[i];
  56
  57                 if (!pd)
  58                         continue;
  59
  60                 if (pol && pol->pd_exit_fn)
  61                         pol->pd_exit_fn(blkg);
  62
  63                 kfree(pd);
  64         }
  65
  66         blk_exit_rl(&blkg->rl);
  67         kfree(blkg);
  68 }
  69
  70 /**
  71  * blkg_alloc - allocate a blkg
  72  * @blkcg: block cgroup the new blkg is associated with
  73  * @q: request_queue the new blkg is associated with
  74  * @gfp_mask: allocation mask to use
  75  *
  76  * Allocate a new blkg assocating @blkcg and @q.
  77  */
  78 static struct blkcg_gq *blkg_alloc(struct blkcg *blkcg, struct request_queue *q,
  79                                    gfp_t gfp_mask)
  80 {
  81         struct blkcg_gq *blkg;
  82         int i;
  83
  84         /* alloc and init base part */
  85         blkg = kzalloc_node(sizeof(*blkg), gfp_mask, q->node);
  86         if (!blkg)
  87                 return NULL;
  88
  89         blkg->q = q;
  90         INIT_LIST_HEAD(&blkg->q_node);
  91         blkg->blkcg = blkcg;
  92         blkg->refcnt = 1;
  93
  94         /* root blkg uses @q->root_rl, init rl only for !root blkgs */
  95         if (blkcg != &blkcg_root) {
  96                 if (blk_init_rl(&blkg->rl, q, gfp_mask))
  97                         goto err_free;
  98                 blkg->rl.blkg = blkg;
  99         }
 100
 101         for (i = 0; i < BLKCG_MAX_POLS; i++) {
 102                 struct blkcg_policy *pol = blkcg_policy[i];
 103                 struct blkg_policy_data *pd;
 104
 105                 if (!blkcg_policy_enabled(q, pol))
 106                         continue;
 107
 108                 /* alloc per-policy data and attach it to blkg */
 109                 pd = kzalloc_node(pol->pd_size, gfp_mask, q->node);
 110                 if (!pd)
 111                         goto err_free;
 112
 113                 blkg->pd[i] = pd;
 114                 pd->blkg = blkg;
 115
 116                 /* invoke per-policy init */
 117                 if (blkcg_policy_enabled(blkg->q, pol))
 118                         pol->pd_init_fn(blkg);
 119         }
 120
 121         return blkg;
 122
 123 err_free:
 124         blkg_free(blkg);
 125         return NULL;
 126 }
 127
 128 static struct blkcg_gq *__blkg_lookup(struct blkcg *blkcg,
 129                                       struct request_queue *q)
 130 {
 131         struct blkcg_gq *blkg;
 132
 133         blkg = rcu_dereference(blkcg->blkg_hint);
 134         if (blkg && blkg->q == q)
 135                 return blkg;
 136
 137         /*
 138          * Hint didn't match.  Look up from the radix tree.  Note that we
 139          * may not be holding queue_lock and thus are not sure whether
 140          * @blkg from blkg_tree has already been removed or not, so we
 141          * can't update hint to the lookup result.  Leave it to the caller.
 142          */
 143         blkg = radix_tree_lookup(&blkcg->blkg_tree, q->id);
 144         if (blkg && blkg->q == q)
 145                 return blkg;
 146
 147         return NULL;
 148 }
 149
 150 /**
 151  * blkg_lookup - lookup blkg for the specified blkcg - q pair
 152  * @blkcg: blkcg of interest
 153  * @q: request_queue of interest
 154  *
 155  * Lookup blkg for the @blkcg - @q pair.  This function should be called
 156  * under RCU read lock and is guaranteed to return %NULL if @q is bypassing
 157  * - see blk_queue_bypass_start() for details.
 158  */
 159 struct blkcg_gq *blkg_lookup(struct blkcg *blkcg, struct request_queue *q)
 160 {
 161         WARN_ON_ONCE(!rcu_read_lock_held());
 162
 163         if (unlikely(blk_queue_bypass(q)))
 164                 return NULL;
 165         return __blkg_lookup(blkcg, q);
 166 }
 167 EXPORT_SYMBOL_GPL(blkg_lookup);
 168
 169 /*
 170  * If @new_blkg is %NULL, this function tries to allocate a new one as
 171  * necessary using %GFP_ATOMIC.  @new_blkg is always consumed on return.
 172  */
 173 static struct blkcg_gq *__blkg_lookup_create(struct blkcg *blkcg,
 174                                              struct request_queue *q,
 175                                              struct blkcg_gq *new_blkg)
 176 {
 177         struct blkcg_gq *blkg;
 178         int ret;
 179
 180         WARN_ON_ONCE(!rcu_read_lock_held());
 181         lockdep_assert_held(q->queue_lock);
 182
 183         /* lookup and update hint on success, see __blkg_lookup() for details */
 184         blkg = __blkg_lookup(blkcg, q);
 185         if (blkg) {
 186                 rcu_assign_pointer(blkcg->blkg_hint, blkg);
 187                 goto out_free;
 188         }
 189
 190         /* blkg holds a reference to blkcg */
 191         if (!css_tryget(&blkcg->css)) {
 192                 blkg = ERR_PTR(-EINVAL);
 193                 goto out_free;
 194         }
 195
 196         /* allocate */
 197         if (!new_blkg) {
 198                 new_blkg = blkg_alloc(blkcg, q, GFP_ATOMIC);
 199                 if (unlikely(!new_blkg)) {
 200                         blkg = ERR_PTR(-ENOMEM);
 201                         goto out_put;
 202                 }
 203         }
 204         blkg = new_blkg;
 205
 206         /* insert */
 207         spin_lock(&blkcg->lock);
 208         ret = radix_tree_insert(&blkcg->blkg_tree, q->id, blkg);
 209         if (likely(!ret)) {
 210                 hlist_add_head_rcu(&blkg->blkcg_node, &blkcg->blkg_list);
 211                 list_add(&blkg->q_node, &q->blkg_list);
 212         }
 213         spin_unlock(&blkcg->lock);
 214
 215         if (!ret)
 216                 return blkg;
 217
 218         blkg = ERR_PTR(ret);
 219 out_put:
 220         css_put(&blkcg->css);
 221 out_free:
 222         blkg_free(new_blkg);
 223         return blkg;
 224 }
 225
 226 struct blkcg_gq *blkg_lookup_create(struct blkcg *blkcg,
 227                                     struct request_queue *q)
 228 {
 229         /*
 230          * This could be the first entry point of blkcg implementation and
 231          * we shouldn't allow anything to go through for a bypassing queue.
 232          */
 233         if (unlikely(blk_queue_bypass(q)))
 234                 return ERR_PTR(blk_queue_dead(q) ? -EINVAL : -EBUSY);
 235         return __blkg_lookup_create(blkcg, q, NULL);
 236 }
 237 EXPORT_SYMBOL_GPL(blkg_lookup_create);
 238
 239 static void blkg_destroy(struct blkcg_gq *blkg)
 240 {
 241         struct blkcg *blkcg = blkg->blkcg;
 242
 243         lockdep_assert_held(blkg->q->queue_lock);
 244         lockdep_assert_held(&blkcg->lock);
 245
 246         /* Something wrong if we are trying to remove same group twice */
 247         WARN_ON_ONCE(list_empty(&blkg->q_node));
 248         WARN_ON_ONCE(hlist_unhashed(&blkg->blkcg_node));
 249
 250         radix_tree_delete(&blkcg->blkg_tree, blkg->q->id);
 251         list_del_init(&blkg->q_node);
 252         hlist_del_init_rcu(&blkg->blkcg_node);
 253
 254         /*
 255          * Both setting lookup hint to and clearing it from @blkg are done
 256          * under queue_lock.  If it's not pointing to @blkg now, it never
 257          * will.  Hint assignment itself can race safely.
 258          */
 259         if (rcu_dereference_raw(blkcg->blkg_hint) == blkg)
 260                 rcu_assign_pointer(blkcg->blkg_hint, NULL);
 261
 262         /*
 263          * Put the reference taken at the time of creation so that when all
 264          * queues are gone, group can be destroyed.
 265          */
 266         blkg_put(blkg);
 267 }
 268
 269 /**
 270  * blkg_destroy_all - destroy all blkgs associated with a request_queue
 271  * @q: request_queue of interest
 272  *
 273  * Destroy all blkgs associated with @q.
 274  */
 275 static void blkg_destroy_all(struct request_queue *q)
 276 {
 277         struct blkcg_gq *blkg, *n;
 278
 279         lockdep_assert_held(q->queue_lock);
 280
 281         list_for_each_entry_safe(blkg, n, &q->blkg_list, q_node) {
 282                 struct blkcg *blkcg = blkg->blkcg;
 283
 284                 spin_lock(&blkcg->lock);
 285                 blkg_destroy(blkg);
 286                 spin_unlock(&blkcg->lock);
 287         }
 288 }
 289
 290 static void blkg_rcu_free(struct rcu_head *rcu_head)
 291 {
 292         blkg_free(container_of(rcu_head, struct blkcg_gq, rcu_head));
 293 }
 294
 295 void __blkg_release(struct blkcg_gq *blkg)
 296 {
 297         /* release the extra blkcg reference this blkg has been holding */
 298         css_put(&blkg->blkcg->css);
 299
 300         /*
 301          * A group is freed in rcu manner. But having an rcu lock does not
 302          * mean that one can access all the fields of blkg and assume these
 303          * are valid. For example, don't try to follow throtl_data and
 304          * request queue links.
 305          *
 306          * Having a reference to blkg under an rcu allows acess to only
 307          * values local to groups like group stats and group rate limits
 308          */
 309         call_rcu(&blkg->rcu_head, blkg_rcu_free);
 310 }
 311 EXPORT_SYMBOL_GPL(__blkg_release);
 312
 313 /*
 314  * The next function used by blk_queue_for_each_rl().  It's a bit tricky
 315  * because the root blkg uses @q->root_rl instead of its own rl.
 316  */
 317 struct request_list *__blk_queue_next_rl(struct request_list *rl,
 318                                          struct request_queue *q)
 319 {
 320         struct list_head *ent;
 321         struct blkcg_gq *blkg;
 322
 323         /*
 324          * Determine the current blkg list_head.  The first entry is
 325          * root_rl which is off @q->blkg_list and mapped to the head.
 326          */
 327         if (rl == &q->root_rl) {
 328                 ent = &q->blkg_list;
 329         } else {
 330                 blkg = container_of(rl, struct blkcg_gq, rl);
 331                 ent = &blkg->q_node;
 332         }
 333
 334         /* walk to the next list_head, skip root blkcg */
 335         ent = ent->next;
 336         if (ent == &q->root_blkg->q_node)
 337                 ent = ent->next;
 338         if (ent == &q->blkg_list)
 339                 return NULL;
 340
 341         blkg = container_of(ent, struct blkcg_gq, q_node);
 342         return &blkg->rl;
 343 }
 344
 345 static int blkcg_reset_stats(struct cgroup *cgroup, struct cftype *cftype,
 346                              u64 val)
 347 {
 348         struct blkcg *blkcg = cgroup_to_blkcg(cgroup);
 349         struct blkcg_gq *blkg;
 350         struct hlist_node *n;
 351         int i;
 352
 353         mutex_lock(&blkcg_pol_mutex);
 354         spin_lock_irq(&blkcg->lock);
 355
 356         /*
 357          * Note that stat reset is racy - it doesn't synchronize against
 358          * stat updates.  This is a debug feature which shouldn't exist
 359          * anyway.  If you get hit by a race, retry.
 360          */
 361         hlist_for_each_entry(blkg, n, &blkcg->blkg_list, blkcg_node) {
 362                 for (i = 0; i < BLKCG_MAX_POLS; i++) {
 363                         struct blkcg_policy *pol = blkcg_policy[i];
 364
 365                         if (blkcg_policy_enabled(blkg->q, pol) &&
 366                             pol->pd_reset_stats_fn)
 367                                 pol->pd_reset_stats_fn(blkg);
 368                 }
 369         }
 370
 371         spin_unlock_irq(&blkcg->lock);
 372         mutex_unlock(&blkcg_pol_mutex);
 373         return 0;
 374 }
 375
 376 static const char *blkg_dev_name(struct blkcg_gq *blkg)
 377 {
 378         /* some drivers (floppy) instantiate a queue w/o disk registered */
 379         if (blkg->q->backing_dev_info.dev)
 380                 return dev_name(blkg->q->backing_dev_info.dev);
 381         return NULL;
 382 }
 383
 384 /**
 385  * blkcg_print_blkgs - helper for printing per-blkg data
 386  * @sf: seq_file to print to
 387  * @blkcg: blkcg of interest
 388  * @prfill: fill function to print out a blkg
 389  * @pol: policy in question
 390  * @data: data to be passed to @prfill
 391  * @show_total: to print out sum of prfill return values or not
 392  *
 393  * This function invokes @prfill on each blkg of @blkcg if pd for the
 394  * policy specified by @pol exists.  @prfill is invoked with @sf, the
 395  * policy data and @data.  If @show_total is %true, the sum of the return
 396  * values from @prfill is printed with "Total" label at the end.
 397  *
 398  * This is to be used to construct print functions for
 399  * cftype->read_seq_string method.
 400  */
 401 void blkcg_print_blkgs(struct seq_file *sf, struct blkcg *blkcg,
 402                        u64 (*prfill)(struct seq_file *,
 403                                      struct blkg_policy_data *, int),
 404                        const struct blkcg_policy *pol, int data,
 405                        bool show_total)
 406 {
 407         struct blkcg_gq *blkg;
 408         struct hlist_node *n;
 409         u64 total = 0;
 410
 411         spin_lock_irq(&blkcg->lock);
 412         hlist_for_each_entry(blkg, n, &blkcg->blkg_list, blkcg_node)
 413                 if (blkcg_policy_enabled(blkg->q, pol))
 414                         total += prfill(sf, blkg->pd[pol->plid], data);
 415         spin_unlock_irq(&blkcg->lock);
 416
 417         if (show_total)
 418                 seq_printf(sf, "Total %llu\n", (unsigned long long)total);
 419 }
 420 EXPORT_SYMBOL_GPL(blkcg_print_blkgs);
 421
 422 /**
 423  * __blkg_prfill_u64 - prfill helper for a single u64 value
 424  * @sf: seq_file to print to
 425  * @pd: policy private data of interest
 426  * @v: value to print
 427  *
 428  * Print @v to @sf for the device assocaited with @pd.
 429  */
 430 u64 __blkg_prfill_u64(struct seq_file *sf, struct blkg_policy_data *pd, u64 v)
 431 {
 432         const char *dname = blkg_dev_name(pd->blkg);
 433
 434         if (!dname)
 435                 return 0;
 436
 437         seq_printf(sf, "%s %llu\n", dname, (unsigned long long)v);
 438         return v;
 439 }
 440 EXPORT_SYMBOL_GPL(__blkg_prfill_u64);
 441
 442 /**
 443  * __blkg_prfill_rwstat - prfill helper for a blkg_rwstat
 444  * @sf: seq_file to print to
 445  * @pd: policy private data of interest
 446  * @rwstat: rwstat to print
 447  *
 448  * Print @rwstat to @sf for the device assocaited with @pd.
 449  */
 450 u64 __blkg_prfill_rwstat(struct seq_file *sf, struct blkg_policy_data *pd,
 451                          const struct blkg_rwstat *rwstat)
 452 {
 453         static const char *rwstr[] = {
 454                 [BLKG_RWSTAT_READ]      = "Read",
 455                 [BLKG_RWSTAT_WRITE]     = "Write",
 456                 [BLKG_RWSTAT_SYNC]      = "Sync",
 457                 [BLKG_RWSTAT_ASYNC]     = "Async",
 458         };
 459         const char *dname = blkg_dev_name(pd->blkg);
 460         u64 v;
 461         int i;
 462
 463         if (!dname)
 464                 return 0;
 465
 466         for (i = 0; i < BLKG_RWSTAT_NR; i++)
 467                 seq_printf(sf, "%s %s %llu\n", dname, rwstr[i],
 468                            (unsigned long long)rwstat->cnt[i]);
 469
 470         v = rwstat->cnt[BLKG_RWSTAT_READ] + rwstat->cnt[BLKG_RWSTAT_WRITE];
 471         seq_printf(sf, "%s Total %llu\n", dname, (unsigned long long)v);
 472         return v;
 473 }
 474
 475 /**
 476  * blkg_prfill_stat - prfill callback for blkg_stat
 477  * @sf: seq_file to print to
 478  * @pd: policy private data of interest
 479  * @off: offset to the blkg_stat in @pd
 480  *
 481  * prfill callback for printing a blkg_stat.
 482  */
 483 u64 blkg_prfill_stat(struct seq_file *sf, struct blkg_policy_data *pd, int off)
 484 {
 485         return __blkg_prfill_u64(sf, pd, blkg_stat_read((void *)pd + off));
 486 }
 487 EXPORT_SYMBOL_GPL(blkg_prfill_stat);
 488
 489 /**
 490  * blkg_prfill_rwstat - prfill callback for blkg_rwstat
 491  * @sf: seq_file to print to
 492  * @pd: policy private data of interest
 493  * @off: offset to the blkg_rwstat in @pd
 494  *
 495  * prfill callback for printing a blkg_rwstat.
 496  */
 497 u64 blkg_prfill_rwstat(struct seq_file *sf, struct blkg_policy_data *pd,
 498                        int off)
 499 {
 500         struct blkg_rwstat rwstat = blkg_rwstat_read((void *)pd + off);
 501
 502         return __blkg_prfill_rwstat(sf, pd, &rwstat);
 503 }
 504 EXPORT_SYMBOL_GPL(blkg_prfill_rwstat);
 505
 506 /**
 507  * blkg_conf_prep - parse and prepare for per-blkg config update
 508  * @blkcg: target block cgroup
 509  * @pol: target policy
 510  * @input: input string
 511  * @ctx: blkg_conf_ctx to be filled
 512  *
 513  * Parse per-blkg config update from @input and initialize @ctx with the
 514  * result.  @ctx->blkg points to the blkg to be updated and @ctx->v the new
 515  * value.  This function returns with RCU read lock and queue lock held and
 516  * must be paired with blkg_conf_finish().
 517  */
 518 int blkg_conf_prep(struct blkcg *blkcg, const struct blkcg_policy *pol,
 519                    const char *input, struct blkg_conf_ctx *ctx)
 520         __acquires(rcu) __acquires(disk->queue->queue_lock)
 521 {
 522         struct gendisk *disk;
 523         struct blkcg_gq *blkg;
 524         unsigned int major, minor;
 525         unsigned long long v;
 526         int part, ret;
 527
 528         if (sscanf(input, "%u:%u %llu", &major, &minor, &v) != 3)
 529                 return -EINVAL;
 530
 531         disk = get_gendisk(MKDEV(major, minor), &part);
 532         if (!disk || part)
 533                 return -EINVAL;
 534
 535         rcu_read_lock();
 536         spin_lock_irq(disk->queue->queue_lock);
 537
 538         if (blkcg_policy_enabled(disk->queue, pol))
 539                 blkg = blkg_lookup_create(blkcg, disk->queue);
 540         else
 541                 blkg = ERR_PTR(-EINVAL);
 542
 543         if (IS_ERR(blkg)) {
 544                 ret = PTR_ERR(blkg);
 545                 rcu_read_unlock();
 546                 spin_unlock_irq(disk->queue->queue_lock);
 547                 put_disk(disk);
 548                 /*
 549                  * If queue was bypassing, we should retry.  Do so after a
 550                  * short msleep().  It isn't strictly necessary but queue
 551                  * can be bypassing for some time and it's always nice to
 552                  * avoid busy looping.
 553                  */
 554                 if (ret == -EBUSY) {
 555                         msleep(10);
 556                         ret = restart_syscall();
 557                 }
 558                 return ret;
 559         }
 560
 561         ctx->disk = disk;
 562         ctx->blkg = blkg;
 563         ctx->v = v;
 564         return 0;
 565 }
 566 EXPORT_SYMBOL_GPL(blkg_conf_prep);
 567
 568 /**
 569  * blkg_conf_finish - finish up per-blkg config update
 570  * @ctx: blkg_conf_ctx intiailized by blkg_conf_prep()
 571  *
 572  * Finish up after per-blkg config update.  This function must be paired
 573  * with blkg_conf_prep().
 574  */
 575 void blkg_conf_finish(struct blkg_conf_ctx *ctx)
 576         __releases(ctx->disk->queue->queue_lock) __releases(rcu)
 577 {
 578         spin_unlock_irq(ctx->disk->queue->queue_lock);
 579         rcu_read_unlock();
 580         put_disk(ctx->disk);
 581 }
 582 EXPORT_SYMBOL_GPL(blkg_conf_finish);
 583
 584 struct cftype blkcg_files[] = {
 585         {
 586                 .name = "reset_stats",
 587                 .write_u64 = blkcg_reset_stats,
 588         },
 589         { }     /* terminate */
 590 };
 591
 592 /**
 593  * blkcg_pre_destroy - cgroup pre_destroy callback
 594  * @cgroup: cgroup of interest
 595  *
 596  * This function is called when @cgroup is about to go away and responsible
 597  * for shooting down all blkgs associated with @cgroup.  blkgs should be
 598  * removed while holding both q and blkcg locks.  As blkcg lock is nested
 599  * inside q lock, this function performs reverse double lock dancing.
 600  *
 601  * This is the blkcg counterpart of ioc_release_fn().
 602  */
 603 static int blkcg_pre_destroy(struct cgroup *cgroup)
 604 {
 605         struct blkcg *blkcg = cgroup_to_blkcg(cgroup);
 606
 607         spin_lock_irq(&blkcg->lock);
 608
 609         while (!hlist_empty(&blkcg->blkg_list)) {
 610                 struct blkcg_gq *blkg = hlist_entry(blkcg->blkg_list.first,
 611                                                 struct blkcg_gq, blkcg_node);
 612                 struct request_queue *q = blkg->q;
 613
 614                 if (spin_trylock(q->queue_lock)) {
 615                         blkg_destroy(blkg);
 616                         spin_unlock(q->queue_lock);
 617                 } else {
 618                         spin_unlock_irq(&blkcg->lock);
 619                         cpu_relax();
 620                         spin_lock_irq(&blkcg->lock);
 621                 }
 622         }
 623
 624         spin_unlock_irq(&blkcg->lock);
 625         return 0;
 626 }
 627
 628 static void blkcg_destroy(struct cgroup *cgroup)
 629 {
 630         struct blkcg *blkcg = cgroup_to_blkcg(cgroup);
 631
 632         if (blkcg != &blkcg_root)
 633                 kfree(blkcg);
 634 }
 635
 636 static struct cgroup_subsys_state *blkcg_create(struct cgroup *cgroup)
 637 {
 638         static atomic64_t id_seq = ATOMIC64_INIT(0);
 639         struct blkcg *blkcg;
 640         struct cgroup *parent = cgroup->parent;
 641
 642         if (!parent) {
 643                 blkcg = &blkcg_root;
 644                 goto done;
 645         }
 646
 647         blkcg = kzalloc(sizeof(*blkcg), GFP_KERNEL);
 648         if (!blkcg)
 649                 return ERR_PTR(-ENOMEM);
 650
 651         blkcg->cfq_weight = CFQ_WEIGHT_DEFAULT;
 652         blkcg->id = atomic64_inc_return(&id_seq); /* root is 0, start from 1 */
 653 done:
 654         spin_lock_init(&blkcg->lock);
 655         INIT_RADIX_TREE(&blkcg->blkg_tree, GFP_ATOMIC);
 656         INIT_HLIST_HEAD(&blkcg->blkg_list);
 657
 658         return &blkcg->css;
 659 }
 660
 661 /**
 662  * blkcg_init_queue - initialize blkcg part of request queue
 663  * @q: request_queue to initialize
 664  *
 665  * Called from blk_alloc_queue_node(). Responsible for initializing blkcg
 666  * part of new request_queue @q.
 667  *
 668  * RETURNS:
 669  * 0 on success, -errno on failure.
 670  */
 671 int blkcg_init_queue(struct request_queue *q)
 672 {
 673         might_sleep();
 674
 675         return blk_throtl_init(q);
 676 }
 677
 678 /**
 679  * blkcg_drain_queue - drain blkcg part of request_queue
 680  * @q: request_queue to drain
 681  *
 682  * Called from blk_drain_queue().  Responsible for draining blkcg part.
 683  */
 684 void blkcg_drain_queue(struct request_queue *q)
 685 {
 686         lockdep_assert_held(q->queue_lock);
 687
 688         blk_throtl_drain(q);
 689 }
 690
 691 /**
 692  * blkcg_exit_queue - exit and release blkcg part of request_queue
 693  * @q: request_queue being released
 694  *
 695  * Called from blk_release_queue().  Responsible for exiting blkcg part.
 696  */
 697 void blkcg_exit_queue(struct request_queue *q)
 698 {
 699         spin_lock_irq(q->queue_lock);
 700         blkg_destroy_all(q);
 701         spin_unlock_irq(q->queue_lock);
 702
 703         blk_throtl_exit(q);
 704 }
 705
 706 /*
 707  * We cannot support shared io contexts, as we have no mean to support
 708  * two tasks with the same ioc in two different groups without major rework
 709  * of the main cic data structures.  For now we allow a task to change
 710  * its cgroup only if it's the only owner of its ioc.
 711  */
 712 static int blkcg_can_attach(struct cgroup *cgrp, struct cgroup_taskset *tset)
 713 {
 714         struct task_struct *task;
 715         struct io_context *ioc;
 716         int ret = 0;
 717
 718         /* task_lock() is needed to avoid races with exit_io_context() */
 719         cgroup_taskset_for_each(task, cgrp, tset) {
 720                 task_lock(task);
 721                 ioc = task->io_context;
 722                 if (ioc && atomic_read(&ioc->nr_tasks) > 1)
 723                         ret = -EINVAL;
 724                 task_unlock(task);
 725                 if (ret)
 726                         break;
 727         }
 728         return ret;
 729 }
 730
 731 struct cgroup_subsys blkio_subsys = {
 732         .name = "blkio",
 733         .create = blkcg_create,
 734         .can_attach = blkcg_can_attach,
 735         .pre_destroy = blkcg_pre_destroy,
 736         .destroy = blkcg_destroy,
 737         .subsys_id = blkio_subsys_id,
 738         .base_cftypes = blkcg_files,
 739         .module = THIS_MODULE,
 740 };
 741 EXPORT_SYMBOL_GPL(blkio_subsys);
 742
 743 /**
 744  * blkcg_activate_policy - activate a blkcg policy on a request_queue
 745  * @q: request_queue of interest
 746  * @pol: blkcg policy to activate
 747  *
 748  * Activate @pol on @q.  Requires %GFP_KERNEL context.  @q goes through
 749  * bypass mode to populate its blkgs with policy_data for @pol.
 750  *
 751  * Activation happens with @q bypassed, so nobody would be accessing blkgs
 752  * from IO path.  Update of each blkg is protected by both queue and blkcg
 753  * locks so that holding either lock and testing blkcg_policy_enabled() is
 754  * always enough for dereferencing policy data.
 755  *
 756  * The caller is responsible for synchronizing [de]activations and policy
 757  * [un]registerations.  Returns 0 on success, -errno on failure.
 758  */
 759 int blkcg_activate_policy(struct request_queue *q,
 760                           const struct blkcg_policy *pol)
 761 {
 762         LIST_HEAD(pds);
 763         struct blkcg_gq *blkg;
 764         struct blkg_policy_data *pd, *n;
 765         int cnt = 0, ret;
 766         bool preloaded;
 767
 768         if (blkcg_policy_enabled(q, pol))
 769                 return 0;
 770
 771         /* preallocations for root blkg */
 772         blkg = blkg_alloc(&blkcg_root, q, GFP_KERNEL);
 773         if (!blkg)
 774                 return -ENOMEM;
 775
 776         preloaded = !radix_tree_preload(GFP_KERNEL);
 777
 778         blk_queue_bypass_start(q);
 779
 780         /* make sure the root blkg exists and count the existing blkgs */
 781         spin_lock_irq(q->queue_lock);
 782
 783         rcu_read_lock();
 784         blkg = __blkg_lookup_create(&blkcg_root, q, blkg);
 785         rcu_read_unlock();
 786
 787         if (preloaded)
 788                 radix_tree_preload_end();
 789
 790         if (IS_ERR(blkg)) {
 791                 ret = PTR_ERR(blkg);
 792                 goto out_unlock;
 793         }
 794         q->root_blkg = blkg;
 795         q->root_rl.blkg = blkg;
 796
 797         list_for_each_entry(blkg, &q->blkg_list, q_node)
 798                 cnt++;
 799
 800         spin_unlock_irq(q->queue_lock);
 801
 802         /* allocate policy_data for all existing blkgs */
 803         while (cnt--) {
 804                 pd = kzalloc_node(pol->pd_size, GFP_KERNEL, q->node);
 805                 if (!pd) {
 806                         ret = -ENOMEM;
 807                         goto out_free;
 808                 }
 809                 list_add_tail(&pd->alloc_node, &pds);
 810         }
 811
 812         /*
 813          * Install the allocated pds.  With @q bypassing, no new blkg
 814          * should have been created while the queue lock was dropped.
 815          */
 816         spin_lock_irq(q->queue_lock);
 817
 818         list_for_each_entry(blkg, &q->blkg_list, q_node) {
 819                 if (WARN_ON(list_empty(&pds))) {
 820                         /* umm... this shouldn't happen, just abort */
 821                         ret = -ENOMEM;
 822                         goto out_unlock;
 823                 }
 824                 pd = list_first_entry(&pds, struct blkg_policy_data, alloc_node);
 825                 list_del_init(&pd->alloc_node);
 826
 827                 /* grab blkcg lock too while installing @pd on @blkg */
 828                 spin_lock(&blkg->blkcg->lock);
 829
 830                 blkg->pd[pol->plid] = pd;
 831                 pd->blkg = blkg;
 832                 pol->pd_init_fn(blkg);
 833
 834                 spin_unlock(&blkg->blkcg->lock);
 835         }
 836
 837         __set_bit(pol->plid, q->blkcg_pols);
 838         ret = 0;
 839 out_unlock:
 840         spin_unlock_irq(q->queue_lock);
 841 out_free:
 842         blk_queue_bypass_end(q);
 843         list_for_each_entry_safe(pd, n, &pds, alloc_node)
 844                 kfree(pd);
 845         return ret;
 846 }
 847 EXPORT_SYMBOL_GPL(blkcg_activate_policy);
 848
 849 /**
 850  * blkcg_deactivate_policy - deactivate a blkcg policy on a request_queue
 851  * @q: request_queue of interest
 852  * @pol: blkcg policy to deactivate
 853  *
 854  * Deactivate @pol on @q.  Follows the same synchronization rules as
 855  * blkcg_activate_policy().
 856  */
 857 void blkcg_deactivate_policy(struct request_queue *q,
 858                              const struct blkcg_policy *pol)
 859 {
 860         struct blkcg_gq *blkg;
 861
 862         if (!blkcg_policy_enabled(q, pol))
 863                 return;
 864
 865         blk_queue_bypass_start(q);
 866         spin_lock_irq(q->queue_lock);
 867
 868         __clear_bit(pol->plid, q->blkcg_pols);
 869
 870         /* if no policy is left, no need for blkgs - shoot them down */
 871         if (bitmap_empty(q->blkcg_pols, BLKCG_MAX_POLS))
 872                 blkg_destroy_all(q);
 873
 874         list_for_each_entry(blkg, &q->blkg_list, q_node) {
 875                 /* grab blkcg lock too while removing @pd from @blkg */
 876                 spin_lock(&blkg->blkcg->lock);
 877
 878                 if (pol->pd_exit_fn)
 879                         pol->pd_exit_fn(blkg);
 880
 881                 kfree(blkg->pd[pol->plid]);
 882                 blkg->pd[pol->plid] = NULL;
 883
 884                 spin_unlock(&blkg->blkcg->lock);
 885         }
 886
 887         spin_unlock_irq(q->queue_lock);
 888         blk_queue_bypass_end(q);
 889 }
 890 EXPORT_SYMBOL_GPL(blkcg_deactivate_policy);
 891
 892 /**
 893  * blkcg_policy_register - register a blkcg policy
 894  * @pol: blkcg policy to register
 895  *
 896  * Register @pol with blkcg core.  Might sleep and @pol may be modified on
 897  * successful registration.  Returns 0 on success and -errno on failure.
 898  */
 899 int blkcg_policy_register(struct blkcg_policy *pol)
 900 {
 901         int i, ret;
 902
 903         if (WARN_ON(pol->pd_size < sizeof(struct blkg_policy_data)))
 904                 return -EINVAL;
 905
 906         mutex_lock(&blkcg_pol_mutex);
 907
 908         /* find an empty slot */
 909         ret = -ENOSPC;
 910         for (i = 0; i < BLKCG_MAX_POLS; i++)
 911                 if (!blkcg_policy[i])
 912                         break;
 913         if (i >= BLKCG_MAX_POLS)
 914                 goto out_unlock;
 915
 916         /* register and update blkgs */
 917         pol->plid = i;
 918         blkcg_policy[i] = pol;
 919
 920         /* everything is in place, add intf files for the new policy */
 921         if (pol->cftypes)
 922                 WARN_ON(cgroup_add_cftypes(&blkio_subsys, pol->cftypes));
 923         ret = 0;
 924 out_unlock:
 925         mutex_unlock(&blkcg_pol_mutex);
 926         return ret;
 927 }
 928 EXPORT_SYMBOL_GPL(blkcg_policy_register);
 929
 930 /**
 931  * blkcg_policy_unregister - unregister a blkcg policy
 932  * @pol: blkcg policy to unregister
 933  *
 934  * Undo blkcg_policy_register(@pol).  Might sleep.
 935  */
 936 void blkcg_policy_unregister(struct blkcg_policy *pol)
 937 {
 938         mutex_lock(&blkcg_pol_mutex);
 939
 940         if (WARN_ON(blkcg_policy[pol->plid] != pol))
 941                 goto out_unlock;
 942
 943         /* kill the intf files first */
 944         if (pol->cftypes)
 945                 cgroup_rm_cftypes(&blkio_subsys, pol->cftypes);
 946
 947         /* unregister and update blkgs */
 948         blkcg_policy[pol->plid] = NULL;
 949 out_unlock:
 950         mutex_unlock(&blkcg_pol_mutex);
 951 }
 952 EXPORT_SYMBOL_GPL(blkcg_policy_unregister);