mm/backing-dev.c

   1
   2 #include <linux/wait.h>
   3 #include <linux/backing-dev.h>
   4 #include <linux/kthread.h>
   5 #include <linux/freezer.h>
   6 #include <linux/fs.h>
   7 #include <linux/pagemap.h>
   8 #include <linux/mm.h>
   9 #include <linux/sched.h>
  10 #include <linux/module.h>
  11 #include <linux/writeback.h>
  12 #include <linux/device.h>
  13
  14 static atomic_long_t bdi_seq = ATOMIC_LONG_INIT(0);
  15
  16 void default_unplug_io_fn(struct backing_dev_info *bdi, struct page *page)
  17 {
  18 }
  19 EXPORT_SYMBOL(default_unplug_io_fn);
  20
  21 struct backing_dev_info default_backing_dev_info = {
  22         .name           = "default",
  23         .ra_pages       = VM_MAX_READAHEAD * 1024 / PAGE_CACHE_SIZE,
  24         .state          = 0,
  25         .capabilities   = BDI_CAP_MAP_COPY,
  26         .unplug_io_fn   = default_unplug_io_fn,
  27 };
  28 EXPORT_SYMBOL_GPL(default_backing_dev_info);
  29
  30 struct backing_dev_info noop_backing_dev_info = {
  31         .name           = "noop",
  32 };
  33 EXPORT_SYMBOL_GPL(noop_backing_dev_info);
  34
  35 static struct class *bdi_class;
  36
  37 /*
  38  * bdi_lock protects updates to bdi_list and bdi_pending_list, as well as
  39  * reader side protection for bdi_pending_list. bdi_list has RCU reader side
  40  * locking.
  41  */
  42 DEFINE_SPINLOCK(bdi_lock);
  43 LIST_HEAD(bdi_list);
  44 LIST_HEAD(bdi_pending_list);
  45
  46 static struct task_struct *sync_supers_tsk;
  47 static struct timer_list sync_supers_timer;
  48
  49 static int bdi_sync_supers(void *);
  50 static void sync_supers_timer_fn(unsigned long);
  51
  52 static void bdi_add_default_flusher_task(struct backing_dev_info *bdi);
  53
  54 #ifdef CONFIG_DEBUG_FS
  55 #include <linux/debugfs.h>
  56 #include <linux/seq_file.h>
  57
  58 static struct dentry *bdi_debug_root;
  59
  60 static void bdi_debug_init(void)
  61 {
  62         bdi_debug_root = debugfs_create_dir("bdi", NULL);
  63 }
  64
  65 static int bdi_debug_stats_show(struct seq_file *m, void *v)
  66 {
  67         struct backing_dev_info *bdi = m->private;
  68         struct bdi_writeback *wb = &bdi->wb;
  69         unsigned long background_thresh;
  70         unsigned long dirty_thresh;
  71         unsigned long bdi_thresh;
  72         unsigned long nr_dirty, nr_io, nr_more_io, nr_wb;
  73         struct inode *inode;
  74
  75         nr_wb = nr_dirty = nr_io = nr_more_io = 0;
  76         spin_lock(&inode_lock);
  77         list_for_each_entry(inode, &wb->b_dirty, i_list)
  78                 nr_dirty++;
  79         list_for_each_entry(inode, &wb->b_io, i_list)
  80                 nr_io++;
  81         list_for_each_entry(inode, &wb->b_more_io, i_list)
  82                 nr_more_io++;
  83         spin_unlock(&inode_lock);
  84
  85         get_dirty_limits(&background_thresh, &dirty_thresh, &bdi_thresh, bdi);
  86
  87 #define K(x) ((x) << (PAGE_SHIFT - 10))
  88         seq_printf(m,
  89                    "BdiWriteback:     %8lu kB\n"
  90                    "BdiReclaimable:   %8lu kB\n"
  91                    "BdiDirtyThresh:   %8lu kB\n"
  92                    "DirtyThresh:      %8lu kB\n"
  93                    "BackgroundThresh: %8lu kB\n"
  94                    "b_dirty:          %8lu\n"
  95                    "b_io:             %8lu\n"
  96                    "b_more_io:        %8lu\n"
  97                    "bdi_list:         %8u\n"
  98                    "state:            %8lx\n",
  99                    (unsigned long) K(bdi_stat(bdi, BDI_WRITEBACK)),
 100                    (unsigned long) K(bdi_stat(bdi, BDI_RECLAIMABLE)),
 101                    K(bdi_thresh), K(dirty_thresh),
 102                    K(background_thresh), nr_dirty, nr_io, nr_more_io,
 103                    !list_empty(&bdi->bdi_list), bdi->state);
 104 #undef K
 105
 106         return 0;
 107 }
 108
 109 static int bdi_debug_stats_open(struct inode *inode, struct file *file)
 110 {
 111         return single_open(file, bdi_debug_stats_show, inode->i_private);
 112 }
 113
 114 static const struct file_operations bdi_debug_stats_fops = {
 115         .open           = bdi_debug_stats_open,
 116         .read           = seq_read,
 117         .llseek         = seq_lseek,
 118         .release        = single_release,
 119 };
 120
 121 static void bdi_debug_register(struct backing_dev_info *bdi, const char *name)
 122 {
 123         bdi->debug_dir = debugfs_create_dir(name, bdi_debug_root);
 124         bdi->debug_stats = debugfs_create_file("stats", 0444, bdi->debug_dir,
 125                                                bdi, &bdi_debug_stats_fops);
 126 }
 127
 128 static void bdi_debug_unregister(struct backing_dev_info *bdi)
 129 {
 130         debugfs_remove(bdi->debug_stats);
 131         debugfs_remove(bdi->debug_dir);
 132 }
 133 #else
 134 static inline void bdi_debug_init(void)
 135 {
 136 }
 137 static inline void bdi_debug_register(struct backing_dev_info *bdi,
 138                                       const char *name)
 139 {
 140 }
 141 static inline void bdi_debug_unregister(struct backing_dev_info *bdi)
 142 {
 143 }
 144 #endif
 145
 146 static ssize_t read_ahead_kb_store(struct device *dev,
 147                                   struct device_attribute *attr,
 148                                   const char *buf, size_t count)
 149 {
 150         struct backing_dev_info *bdi = dev_get_drvdata(dev);
 151         char *end;
 152         unsigned long read_ahead_kb;
 153         ssize_t ret = -EINVAL;
 154
 155         read_ahead_kb = simple_strtoul(buf, &end, 10);
 156         if (*buf && (end[0] == '\0' || (end[0] == '\n' && end[1] == '\0'))) {
 157                 bdi->ra_pages = read_ahead_kb >> (PAGE_SHIFT - 10);
 158                 ret = count;
 159         }
 160         return ret;
 161 }
 162
 163 #define K(pages) ((pages) << (PAGE_SHIFT - 10))
 164
 165 #define BDI_SHOW(name, expr)                                            \
 166 static ssize_t name##_show(struct device *dev,                          \
 167                            struct device_attribute *attr, char *page)   \
 168 {                                                                       \
 169         struct backing_dev_info *bdi = dev_get_drvdata(dev);            \
 170                                                                         \
 171         return snprintf(page, PAGE_SIZE-1, "%lld\n", (long long)expr);  \
 172 }
 173
 174 BDI_SHOW(read_ahead_kb, K(bdi->ra_pages))
 175
 176 static ssize_t min_ratio_store(struct device *dev,
 177                 struct device_attribute *attr, const char *buf, size_t count)
 178 {
 179         struct backing_dev_info *bdi = dev_get_drvdata(dev);
 180         char *end;
 181         unsigned int ratio;
 182         ssize_t ret = -EINVAL;
 183
 184         ratio = simple_strtoul(buf, &end, 10);
 185         if (*buf && (end[0] == '\0' || (end[0] == '\n' && end[1] == '\0'))) {
 186                 ret = bdi_set_min_ratio(bdi, ratio);
 187                 if (!ret)
 188                         ret = count;
 189         }
 190         return ret;
 191 }
 192 BDI_SHOW(min_ratio, bdi->min_ratio)
 193
 194 static ssize_t max_ratio_store(struct device *dev,
 195                 struct device_attribute *attr, const char *buf, size_t count)
 196 {
 197         struct backing_dev_info *bdi = dev_get_drvdata(dev);
 198         char *end;
 199         unsigned int ratio;
 200         ssize_t ret = -EINVAL;
 201
 202         ratio = simple_strtoul(buf, &end, 10);
 203         if (*buf && (end[0] == '\0' || (end[0] == '\n' && end[1] == '\0'))) {
 204                 ret = bdi_set_max_ratio(bdi, ratio);
 205                 if (!ret)
 206                         ret = count;
 207         }
 208         return ret;
 209 }
 210 BDI_SHOW(max_ratio, bdi->max_ratio)
 211
 212 #define __ATTR_RW(attr) __ATTR(attr, 0644, attr##_show, attr##_store)
 213
 214 static struct device_attribute bdi_dev_attrs[] = {
 215         __ATTR_RW(read_ahead_kb),
 216         __ATTR_RW(min_ratio),
 217         __ATTR_RW(max_ratio),
 218         __ATTR_NULL,
 219 };
 220
 221 static __init int bdi_class_init(void)
 222 {
 223         bdi_class = class_create(THIS_MODULE, "bdi");
 224         if (IS_ERR(bdi_class))
 225                 return PTR_ERR(bdi_class);
 226
 227         bdi_class->dev_attrs = bdi_dev_attrs;
 228         bdi_debug_init();
 229         return 0;
 230 }
 231 postcore_initcall(bdi_class_init);
 232
 233 static int __init default_bdi_init(void)
 234 {
 235         int err;
 236
 237         sync_supers_tsk = kthread_run(bdi_sync_supers, NULL, "sync_supers");
 238         BUG_ON(IS_ERR(sync_supers_tsk));
 239
 240         init_timer(&sync_supers_timer);
 241         setup_timer(&sync_supers_timer, sync_supers_timer_fn, 0);
 242         bdi_arm_supers_timer();
 243
 244         err = bdi_init(&default_backing_dev_info);
 245         if (!err)
 246                 bdi_register(&default_backing_dev_info, NULL, "default");
 247
 248         return err;
 249 }
 250 subsys_initcall(default_bdi_init);
 251
 252 static void bdi_wb_init(struct bdi_writeback *wb, struct backing_dev_info *bdi)
 253 {
 254         memset(wb, 0, sizeof(*wb));
 255
 256         wb->bdi = bdi;
 257         wb->last_old_flush = jiffies;
 258         INIT_LIST_HEAD(&wb->b_dirty);
 259         INIT_LIST_HEAD(&wb->b_io);
 260         INIT_LIST_HEAD(&wb->b_more_io);
 261 }
 262
 263 int bdi_has_dirty_io(struct backing_dev_info *bdi)
 264 {
 265         return wb_has_dirty_io(&bdi->wb);
 266 }
 267
 268 static void bdi_flush_io(struct backing_dev_info *bdi)
 269 {
 270         struct writeback_control wbc = {
 271                 .sync_mode              = WB_SYNC_NONE,
 272                 .older_than_this        = NULL,
 273                 .range_cyclic           = 1,
 274                 .nr_to_write            = 1024,
 275         };
 276
 277         writeback_inodes_wb(&bdi->wb, &wbc);
 278 }
 279
 280 /*
 281  * kupdated() used to do this. We cannot do it from the bdi_forker_task()
 282  * or we risk deadlocking on ->s_umount. The longer term solution would be
 283  * to implement sync_supers_bdi() or similar and simply do it from the
 284  * bdi writeback tasks individually.
 285  */
 286 static int bdi_sync_supers(void *unused)
 287 {
 288         set_user_nice(current, 0);
 289
 290         while (!kthread_should_stop()) {
 291                 set_current_state(TASK_INTERRUPTIBLE);
 292                 schedule();
 293
 294                 /*
 295                  * Do this periodically, like kupdated() did before.
 296                  */
 297                 sync_supers();
 298         }
 299
 300         return 0;
 301 }
 302
 303 void bdi_arm_supers_timer(void)
 304 {
 305         unsigned long next;
 306
 307         if (!dirty_writeback_interval)
 308                 return;
 309
 310         next = msecs_to_jiffies(dirty_writeback_interval * 10) + jiffies;
 311         mod_timer(&sync_supers_timer, round_jiffies_up(next));
 312 }
 313
 314 static void sync_supers_timer_fn(unsigned long unused)
 315 {
 316         wake_up_process(sync_supers_tsk);
 317         bdi_arm_supers_timer();
 318 }
 319
 320 static int bdi_forker_task(void *ptr)
 321 {
 322         struct bdi_writeback *me = ptr;
 323
 324         current->flags |= PF_FLUSHER | PF_SWAPWRITE;
 325         set_freezable();
 326
 327         /*
 328          * Our parent may run at a different priority, just set us to normal
 329          */
 330         set_user_nice(current, 0);
 331
 332         for (;;) {
 333                 struct backing_dev_info *bdi, *tmp;
 334                 struct bdi_writeback *wb;
 335
 336                 /*
 337                  * Temporary measure, we want to make sure we don't see
 338                  * dirty data on the default backing_dev_info
 339                  */
 340                 if (wb_has_dirty_io(me) || !list_empty(&me->bdi->work_list))
 341                         wb_do_writeback(me, 0);
 342
 343                 spin_lock_bh(&bdi_lock);
 344
 345                 /*
 346                  * Check if any existing bdi's have dirty data without
 347                  * a thread registered. If so, set that up.
 348                  */
 349                 list_for_each_entry_safe(bdi, tmp, &bdi_list, bdi_list) {
 350                         if (bdi->wb.task)
 351                                 continue;
 352                         if (list_empty(&bdi->work_list) &&
 353                             !bdi_has_dirty_io(bdi))
 354                                 continue;
 355
 356                         bdi_add_default_flusher_task(bdi);
 357                 }
 358
 359                 set_current_state(TASK_INTERRUPTIBLE);
 360
 361                 if (list_empty(&bdi_pending_list)) {
 362                         unsigned long wait;
 363
 364                         spin_unlock_bh(&bdi_lock);
 365                         wait = msecs_to_jiffies(dirty_writeback_interval * 10);
 366                         if (wait)
 367                                 schedule_timeout(wait);
 368                         else
 369                                 schedule();
 370                         try_to_freeze();
 371                         continue;
 372                 }
 373
 374                 __set_current_state(TASK_RUNNING);
 375
 376                 /*
 377                  * This is our real job - check for pending entries in
 378                  * bdi_pending_list, and create the tasks that got added
 379                  */
 380                 bdi = list_entry(bdi_pending_list.next, struct backing_dev_info,
 381                                  bdi_list);
 382                 list_del_init(&bdi->bdi_list);
 383                 spin_unlock_bh(&bdi_lock);
 384
 385                 wb = &bdi->wb;
 386                 wb->task = kthread_run(bdi_writeback_thread, wb, "flush-%s",
 387                                         dev_name(bdi->dev));
 388                 /*
 389                  * If task creation fails, then readd the bdi to
 390                  * the pending list and force writeout of the bdi
 391                  * from this forker thread. That will free some memory
 392                  * and we can try again.
 393                  */
 394                 if (IS_ERR(wb->task)) {
 395                         wb->task = NULL;
 396
 397                         /*
 398                          * Add this 'bdi' to the back, so we get
 399                          * a chance to flush other bdi's to free
 400                          * memory.
 401                          */
 402                         spin_lock_bh(&bdi_lock);
 403                         list_add_tail(&bdi->bdi_list, &bdi_pending_list);
 404                         spin_unlock_bh(&bdi_lock);
 405
 406                         bdi_flush_io(bdi);
 407                 }
 408         }
 409
 410         return 0;
 411 }
 412
 413 static void bdi_add_to_pending(struct rcu_head *head)
 414 {
 415         struct backing_dev_info *bdi;
 416
 417         bdi = container_of(head, struct backing_dev_info, rcu_head);
 418         INIT_LIST_HEAD(&bdi->bdi_list);
 419
 420         spin_lock(&bdi_lock);
 421         list_add_tail(&bdi->bdi_list, &bdi_pending_list);
 422         spin_unlock(&bdi_lock);
 423
 424         /*
 425          * We are now on the pending list, wake up bdi_forker_task()
 426          * to finish the job and add us back to the active bdi_list
 427          */
 428         wake_up_process(default_backing_dev_info.wb.task);
 429 }
 430
 431 /*
 432  * Add the default flusher task that gets created for any bdi
 433  * that has dirty data pending writeout
 434  */
 435 void static bdi_add_default_flusher_task(struct backing_dev_info *bdi)
 436 {
 437         if (!bdi_cap_writeback_dirty(bdi))
 438                 return;
 439
 440         if (WARN_ON(!test_bit(BDI_registered, &bdi->state))) {
 441                 printk(KERN_ERR "bdi %p/%s is not registered!\n",
 442                                                         bdi, bdi->name);
 443                 return;
 444         }
 445
 446         /*
 447          * Check with the helper whether to proceed adding a task. Will only
 448          * abort if we two or more simultanous calls to
 449          * bdi_add_default_flusher_task() occured, further additions will block
 450          * waiting for previous additions to finish.
 451          */
 452         if (!test_and_set_bit(BDI_pending, &bdi->state)) {
 453                 list_del_rcu(&bdi->bdi_list);
 454
 455                 /*
 456                  * We must wait for the current RCU period to end before
 457                  * moving to the pending list. So schedule that operation
 458                  * from an RCU callback.
 459                  */
 460                 call_rcu(&bdi->rcu_head, bdi_add_to_pending);
 461         }
 462 }
 463
 464 /*
 465  * Remove bdi from bdi_list, and ensure that it is no longer visible
 466  */
 467 static void bdi_remove_from_list(struct backing_dev_info *bdi)
 468 {
 469         spin_lock_bh(&bdi_lock);
 470         list_del_rcu(&bdi->bdi_list);
 471         spin_unlock_bh(&bdi_lock);
 472
 473         synchronize_rcu();
 474 }
 475
 476 int bdi_register(struct backing_dev_info *bdi, struct device *parent,
 477                 const char *fmt, ...)
 478 {
 479         va_list args;
 480         int ret = 0;
 481         struct device *dev;
 482
 483         if (bdi->dev)   /* The driver needs to use separate queues per device */
 484                 goto exit;
 485
 486         va_start(args, fmt);
 487         dev = device_create_vargs(bdi_class, parent, MKDEV(0, 0), bdi, fmt, args);
 488         va_end(args);
 489         if (IS_ERR(dev)) {
 490                 ret = PTR_ERR(dev);
 491                 goto exit;
 492         }
 493
 494         spin_lock_bh(&bdi_lock);
 495         list_add_tail_rcu(&bdi->bdi_list, &bdi_list);
 496         spin_unlock_bh(&bdi_lock);
 497
 498         bdi->dev = dev;
 499
 500         /*
 501          * Just start the forker thread for our default backing_dev_info,
 502          * and add other bdi's to the list. They will get a thread created
 503          * on-demand when they need it.
 504          */
 505         if (bdi_cap_flush_forker(bdi)) {
 506                 struct bdi_writeback *wb = &bdi->wb;
 507
 508                 wb->task = kthread_run(bdi_forker_task, wb, "bdi-%s",
 509                                                 dev_name(dev));
 510                 if (IS_ERR(wb->task)) {
 511                         wb->task = NULL;
 512                         ret = -ENOMEM;
 513
 514                         bdi_remove_from_list(bdi);
 515                         goto exit;
 516                 }
 517         }
 518
 519         bdi_debug_register(bdi, dev_name(dev));
 520         set_bit(BDI_registered, &bdi->state);
 521 exit:
 522         return ret;
 523 }
 524 EXPORT_SYMBOL(bdi_register);
 525
 526 int bdi_register_dev(struct backing_dev_info *bdi, dev_t dev)
 527 {
 528         return bdi_register(bdi, NULL, "%u:%u", MAJOR(dev), MINOR(dev));
 529 }
 530 EXPORT_SYMBOL(bdi_register_dev);
 531
 532 /*
 533  * Remove bdi from the global list and shutdown any threads we have running
 534  */
 535 static void bdi_wb_shutdown(struct backing_dev_info *bdi)
 536 {
 537         if (!bdi_cap_writeback_dirty(bdi))
 538                 return;
 539
 540         /*
 541          * If setup is pending, wait for that to complete first
 542          */
 543         wait_on_bit(&bdi->state, BDI_pending, bdi_sched_wait,
 544                         TASK_UNINTERRUPTIBLE);
 545
 546         /*
 547          * Make sure nobody finds us on the bdi_list anymore
 548          */
 549         bdi_remove_from_list(bdi);
 550
 551         /*
 552          * Finally, kill the kernel thread. We don't need to be RCU
 553          * safe anymore, since the bdi is gone from visibility. Force
 554          * unfreeze of the thread before calling kthread_stop(), otherwise
 555          * it would never exet if it is currently stuck in the refrigerator.
 556          */
 557         if (bdi->wb.task) {
 558                 thaw_process(bdi->wb.task);
 559                 kthread_stop(bdi->wb.task);
 560         }
 561 }
 562
 563 /*
 564  * This bdi is going away now, make sure that no super_blocks point to it
 565  */
 566 static void bdi_prune_sb(struct backing_dev_info *bdi)
 567 {
 568         struct super_block *sb;
 569
 570         spin_lock(&sb_lock);
 571         list_for_each_entry(sb, &super_blocks, s_list) {
 572                 if (sb->s_bdi == bdi)
 573                         sb->s_bdi = NULL;
 574         }
 575         spin_unlock(&sb_lock);
 576 }
 577
 578 void bdi_unregister(struct backing_dev_info *bdi)
 579 {
 580         if (bdi->dev) {
 581                 bdi_prune_sb(bdi);
 582
 583                 if (!bdi_cap_flush_forker(bdi))
 584                         bdi_wb_shutdown(bdi);
 585                 bdi_debug_unregister(bdi);
 586                 device_unregister(bdi->dev);
 587                 bdi->dev = NULL;
 588         }
 589 }
 590 EXPORT_SYMBOL(bdi_unregister);
 591
 592 int bdi_init(struct backing_dev_info *bdi)
 593 {
 594         int i, err;
 595
 596         bdi->dev = NULL;
 597
 598         bdi->min_ratio = 0;
 599         bdi->max_ratio = 100;
 600         bdi->max_prop_frac = PROP_FRAC_BASE;
 601         spin_lock_init(&bdi->wb_lock);
 602         INIT_RCU_HEAD(&bdi->rcu_head);
 603         INIT_LIST_HEAD(&bdi->bdi_list);
 604         INIT_LIST_HEAD(&bdi->work_list);
 605
 606         bdi_wb_init(&bdi->wb, bdi);
 607
 608         for (i = 0; i < NR_BDI_STAT_ITEMS; i++) {
 609                 err = percpu_counter_init(&bdi->bdi_stat[i], 0);
 610                 if (err)
 611                         goto err;
 612         }
 613
 614         bdi->dirty_exceeded = 0;
 615         err = prop_local_init_percpu(&bdi->completions);
 616
 617         if (err) {
 618 err:
 619                 while (i--)
 620                         percpu_counter_destroy(&bdi->bdi_stat[i]);
 621         }
 622
 623         return err;
 624 }
 625 EXPORT_SYMBOL(bdi_init);
 626
 627 void bdi_destroy(struct backing_dev_info *bdi)
 628 {
 629         int i;
 630
 631         /*
 632          * Splice our entries to the default_backing_dev_info, if this
 633          * bdi disappears
 634          */
 635         if (bdi_has_dirty_io(bdi)) {
 636                 struct bdi_writeback *dst = &default_backing_dev_info.wb;
 637
 638                 spin_lock(&inode_lock);
 639                 list_splice(&bdi->wb.b_dirty, &dst->b_dirty);
 640                 list_splice(&bdi->wb.b_io, &dst->b_io);
 641                 list_splice(&bdi->wb.b_more_io, &dst->b_more_io);
 642                 spin_unlock(&inode_lock);
 643         }
 644
 645         bdi_unregister(bdi);
 646
 647         for (i = 0; i < NR_BDI_STAT_ITEMS; i++)
 648                 percpu_counter_destroy(&bdi->bdi_stat[i]);
 649
 650         prop_local_destroy_percpu(&bdi->completions);
 651 }
 652 EXPORT_SYMBOL(bdi_destroy);
 653
 654 /*
 655  * For use from filesystems to quickly init and register a bdi associated
 656  * with dirty writeback
 657  */
 658 int bdi_setup_and_register(struct backing_dev_info *bdi, char *name,
 659                            unsigned int cap)
 660 {
 661         char tmp[32];
 662         int err;
 663
 664         bdi->name = name;
 665         bdi->capabilities = cap;
 666         err = bdi_init(bdi);
 667         if (err)
 668                 return err;
 669
 670         sprintf(tmp, "%.28s%s", name, "-%d");
 671         err = bdi_register(bdi, NULL, tmp, atomic_long_inc_return(&bdi_seq));
 672         if (err) {
 673                 bdi_destroy(bdi);
 674                 return err;
 675         }
 676
 677         return 0;
 678 }
 679 EXPORT_SYMBOL(bdi_setup_and_register);
 680
 681 static wait_queue_head_t congestion_wqh[2] = {
 682                 __WAIT_QUEUE_HEAD_INITIALIZER(congestion_wqh[0]),
 683                 __WAIT_QUEUE_HEAD_INITIALIZER(congestion_wqh[1])
 684         };
 685
 686 void clear_bdi_congested(struct backing_dev_info *bdi, int sync)
 687 {
 688         enum bdi_state bit;
 689         wait_queue_head_t *wqh = &congestion_wqh[sync];
 690
 691         bit = sync ? BDI_sync_congested : BDI_async_congested;
 692         clear_bit(bit, &bdi->state);
 693         smp_mb__after_clear_bit();
 694         if (waitqueue_active(wqh))
 695                 wake_up(wqh);
 696 }
 697 EXPORT_SYMBOL(clear_bdi_congested);
 698
 699 void set_bdi_congested(struct backing_dev_info *bdi, int sync)
 700 {
 701         enum bdi_state bit;
 702
 703         bit = sync ? BDI_sync_congested : BDI_async_congested;
 704         set_bit(bit, &bdi->state);
 705 }
 706 EXPORT_SYMBOL(set_bdi_congested);
 707
 708 /**
 709  * congestion_wait - wait for a backing_dev to become uncongested
 710  * @sync: SYNC or ASYNC IO
 711  * @timeout: timeout in jiffies
 712  *
 713  * Waits for up to @timeout jiffies for a backing_dev (any backing_dev) to exit
 714  * write congestion.  If no backing_devs are congested then just wait for the
 715  * next write to be completed.
 716  */
 717 long congestion_wait(int sync, long timeout)
 718 {
 719         long ret;
 720         DEFINE_WAIT(wait);
 721         wait_queue_head_t *wqh = &congestion_wqh[sync];
 722
 723         prepare_to_wait(wqh, &wait, TASK_UNINTERRUPTIBLE);
 724         ret = io_schedule_timeout(timeout);
 725         finish_wait(wqh, &wait);
 726         return ret;
 727 }
 728 EXPORT_SYMBOL(congestion_wait);
 729