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