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