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 struct backing_dev_info default_backing_dev_info = {
  18         .name           = "default",
  19         .ra_pages       = VM_MAX_READAHEAD * 1024 / PAGE_CACHE_SIZE,
  20         .state          = 0,
  21         .capabilities   = BDI_CAP_MAP_COPY,
  22 };
  23 EXPORT_SYMBOL_GPL(default_backing_dev_info);
  24
  25 struct backing_dev_info noop_backing_dev_info = {
  26         .name           = "noop",
  27         .capabilities   = BDI_CAP_NO_ACCT_AND_WRITEBACK,
  28 };
  29 EXPORT_SYMBOL_GPL(noop_backing_dev_info);
  30
  31 static struct class *bdi_class;
  32
  33 /*
  34  * bdi_lock protects updates to bdi_list. bdi_list has RCU reader side
  35  * locking.
  36  */
  37 DEFINE_SPINLOCK(bdi_lock);
  38 LIST_HEAD(bdi_list);
  39
  40 /* bdi_wq serves all asynchronous writeback tasks */
  41 struct workqueue_struct *bdi_wq;
  42
  43 static void bdi_lock_two(struct bdi_writeback *wb1, struct bdi_writeback *wb2)
  44 {
  45         if (wb1 < wb2) {
  46                 spin_lock(&wb1->list_lock);
  47                 spin_lock_nested(&wb2->list_lock, 1);
  48         } else {
  49                 spin_lock(&wb2->list_lock);
  50                 spin_lock_nested(&wb1->list_lock, 1);
  51         }
  52 }
  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_dirty_time;
  73         struct inode *inode;
  74
  75         nr_dirty = nr_io = nr_more_io = nr_dirty_time = 0;
  76         spin_lock(&wb->list_lock);
  77         list_for_each_entry(inode, &wb->b_dirty, i_wb_list)
  78                 nr_dirty++;
  79         list_for_each_entry(inode, &wb->b_io, i_wb_list)
  80                 nr_io++;
  81         list_for_each_entry(inode, &wb->b_more_io, i_wb_list)
  82                 nr_more_io++;
  83         list_for_each_entry(inode, &wb->b_dirty_time, i_wb_list)
  84                 if (inode->i_state & I_DIRTY_TIME)
  85                         nr_dirty_time++;
  86         spin_unlock(&wb->list_lock);
  87
  88         global_dirty_limits(&background_thresh, &dirty_thresh);
  89         bdi_thresh = bdi_dirty_limit(bdi, dirty_thresh);
  90
  91 #define K(x) ((x) << (PAGE_SHIFT - 10))
  92         seq_printf(m,
  93                    "BdiWriteback:       %10lu kB\n"
  94                    "BdiReclaimable:     %10lu kB\n"
  95                    "BdiDirtyThresh:     %10lu kB\n"
  96                    "DirtyThresh:        %10lu kB\n"
  97                    "BackgroundThresh:   %10lu kB\n"
  98                    "BdiDirtied:         %10lu kB\n"
  99                    "BdiWritten:         %10lu kB\n"
 100                    "BdiWriteBandwidth:  %10lu kBps\n"
 101                    "b_dirty:            %10lu\n"
 102                    "b_io:               %10lu\n"
 103                    "b_more_io:          %10lu\n"
 104                    "b_dirty_time:       %10lu\n"
 105                    "bdi_list:           %10u\n"
 106                    "state:              %10lx\n",
 107                    (unsigned long) K(bdi_stat(bdi, BDI_WRITEBACK)),
 108                    (unsigned long) K(bdi_stat(bdi, BDI_RECLAIMABLE)),
 109                    K(bdi_thresh),
 110                    K(dirty_thresh),
 111                    K(background_thresh),
 112                    (unsigned long) K(bdi_stat(bdi, BDI_DIRTIED)),
 113                    (unsigned long) K(bdi_stat(bdi, BDI_WRITTEN)),
 114                    (unsigned long) K(bdi->write_bandwidth),
 115                    nr_dirty,
 116                    nr_io,
 117                    nr_more_io,
 118                    nr_dirty_time,
 119                    !list_empty(&bdi->bdi_list), bdi->state);
 120 #undef K
 121
 122         return 0;
 123 }
 124
 125 static int bdi_debug_stats_open(struct inode *inode, struct file *file)
 126 {
 127         return single_open(file, bdi_debug_stats_show, inode->i_private);
 128 }
 129
 130 static const struct file_operations bdi_debug_stats_fops = {
 131         .open           = bdi_debug_stats_open,
 132         .read           = seq_read,
 133         .llseek         = seq_lseek,
 134         .release        = single_release,
 135 };
 136
 137 static void bdi_debug_register(struct backing_dev_info *bdi, const char *name)
 138 {
 139         bdi->debug_dir = debugfs_create_dir(name, bdi_debug_root);
 140         bdi->debug_stats = debugfs_create_file("stats", 0444, bdi->debug_dir,
 141                                                bdi, &bdi_debug_stats_fops);
 142 }
 143
 144 static void bdi_debug_unregister(struct backing_dev_info *bdi)
 145 {
 146         debugfs_remove(bdi->debug_stats);
 147         debugfs_remove(bdi->debug_dir);
 148 }
 149 #else
 150 static inline void bdi_debug_init(void)
 151 {
 152 }
 153 static inline void bdi_debug_register(struct backing_dev_info *bdi,
 154                                       const char *name)
 155 {
 156 }
 157 static inline void bdi_debug_unregister(struct backing_dev_info *bdi)
 158 {
 159 }
 160 #endif
 161
 162 static ssize_t read_ahead_kb_store(struct device *dev,
 163                                   struct device_attribute *attr,
 164                                   const char *buf, size_t count)
 165 {
 166         struct backing_dev_info *bdi = dev_get_drvdata(dev);
 167         unsigned long read_ahead_kb;
 168         ssize_t ret;
 169
 170         ret = kstrtoul(buf, 10, &read_ahead_kb);
 171         if (ret < 0)
 172                 return ret;
 173
 174         bdi->ra_pages = read_ahead_kb >> (PAGE_SHIFT - 10);
 175
 176         return count;
 177 }
 178
 179 #define K(pages) ((pages) << (PAGE_SHIFT - 10))
 180
 181 #define BDI_SHOW(name, expr)                                            \
 182 static ssize_t name##_show(struct device *dev,                          \
 183                            struct device_attribute *attr, char *page)   \
 184 {                                                                       \
 185         struct backing_dev_info *bdi = dev_get_drvdata(dev);            \
 186                                                                         \
 187         return snprintf(page, PAGE_SIZE-1, "%lld\n", (long long)expr);  \
 188 }                                                                       \
 189 static DEVICE_ATTR_RW(name);
 190
 191 BDI_SHOW(read_ahead_kb, K(bdi->ra_pages))
 192
 193 static ssize_t min_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         unsigned int ratio;
 198         ssize_t ret;
 199
 200         ret = kstrtouint(buf, 10, &ratio);
 201         if (ret < 0)
 202                 return ret;
 203
 204         ret = bdi_set_min_ratio(bdi, ratio);
 205         if (!ret)
 206                 ret = count;
 207
 208         return ret;
 209 }
 210 BDI_SHOW(min_ratio, bdi->min_ratio)
 211
 212 static ssize_t max_ratio_store(struct device *dev,
 213                 struct device_attribute *attr, const char *buf, size_t count)
 214 {
 215         struct backing_dev_info *bdi = dev_get_drvdata(dev);
 216         unsigned int ratio;
 217         ssize_t ret;
 218
 219         ret = kstrtouint(buf, 10, &ratio);
 220         if (ret < 0)
 221                 return ret;
 222
 223         ret = bdi_set_max_ratio(bdi, ratio);
 224         if (!ret)
 225                 ret = count;
 226
 227         return ret;
 228 }
 229 BDI_SHOW(max_ratio, bdi->max_ratio)
 230
 231 static ssize_t stable_pages_required_show(struct device *dev,
 232                                           struct device_attribute *attr,
 233                                           char *page)
 234 {
 235         struct backing_dev_info *bdi = dev_get_drvdata(dev);
 236
 237         return snprintf(page, PAGE_SIZE-1, "%d\n",
 238                         bdi_cap_stable_pages_required(bdi) ? 1 : 0);
 239 }
 240 static DEVICE_ATTR_RO(stable_pages_required);
 241
 242 static struct attribute *bdi_dev_attrs[] = {
 243         &dev_attr_read_ahead_kb.attr,
 244         &dev_attr_min_ratio.attr,
 245         &dev_attr_max_ratio.attr,
 246         &dev_attr_stable_pages_required.attr,
 247         NULL,
 248 };
 249 ATTRIBUTE_GROUPS(bdi_dev);
 250
 251 static __init int bdi_class_init(void)
 252 {
 253         bdi_class = class_create(THIS_MODULE, "bdi");
 254         if (IS_ERR(bdi_class))
 255                 return PTR_ERR(bdi_class);
 256
 257         bdi_class->dev_groups = bdi_dev_groups;
 258         bdi_debug_init();
 259         return 0;
 260 }
 261 postcore_initcall(bdi_class_init);
 262
 263 static int __init default_bdi_init(void)
 264 {
 265         int err;
 266
 267         bdi_wq = alloc_workqueue("writeback", WQ_MEM_RECLAIM | WQ_FREEZABLE |
 268                                               WQ_UNBOUND | WQ_SYSFS, 0);
 269         if (!bdi_wq)
 270                 return -ENOMEM;
 271
 272         err = bdi_init(&default_backing_dev_info);
 273         if (!err)
 274                 bdi_register(&default_backing_dev_info, NULL, "default");
 275         err = bdi_init(&noop_backing_dev_info);
 276
 277         return err;
 278 }
 279 subsys_initcall(default_bdi_init);
 280
 281 int bdi_has_dirty_io(struct backing_dev_info *bdi)
 282 {
 283         return wb_has_dirty_io(&bdi->wb);
 284 }
 285
 286 /*
 287  * This function is used when the first inode for this bdi is marked dirty. It
 288  * wakes-up the corresponding bdi thread which should then take care of the
 289  * periodic background write-out of dirty inodes. Since the write-out would
 290  * starts only 'dirty_writeback_interval' centisecs from now anyway, we just
 291  * set up a timer which wakes the bdi thread up later.
 292  *
 293  * Note, we wouldn't bother setting up the timer, but this function is on the
 294  * fast-path (used by '__mark_inode_dirty()'), so we save few context switches
 295  * by delaying the wake-up.
 296  *
 297  * We have to be careful not to postpone flush work if it is scheduled for
 298  * earlier. Thus we use queue_delayed_work().
 299  */
 300 void bdi_wakeup_thread_delayed(struct backing_dev_info *bdi)
 301 {
 302         unsigned long timeout;
 303
 304         timeout = msecs_to_jiffies(dirty_writeback_interval * 10);
 305         spin_lock_bh(&bdi->wb_lock);
 306         if (test_bit(BDI_registered, &bdi->state))
 307                 queue_delayed_work(bdi_wq, &bdi->wb.dwork, timeout);
 308         spin_unlock_bh(&bdi->wb_lock);
 309 }
 310
 311 /*
 312  * Remove bdi from bdi_list, and ensure that it is no longer visible
 313  */
 314 static void bdi_remove_from_list(struct backing_dev_info *bdi)
 315 {
 316         spin_lock_bh(&bdi_lock);
 317         list_del_rcu(&bdi->bdi_list);
 318         spin_unlock_bh(&bdi_lock);
 319
 320         synchronize_rcu_expedited();
 321 }
 322
 323 int bdi_register(struct backing_dev_info *bdi, struct device *parent,
 324                 const char *fmt, ...)
 325 {
 326         va_list args;
 327         struct device *dev;
 328
 329         if (bdi->dev)   /* The driver needs to use separate queues per device */
 330                 return 0;
 331
 332         va_start(args, fmt);
 333         dev = device_create_vargs(bdi_class, parent, MKDEV(0, 0), bdi, fmt, args);
 334         va_end(args);
 335         if (IS_ERR(dev))
 336                 return PTR_ERR(dev);
 337
 338         bdi->dev = dev;
 339
 340         bdi_debug_register(bdi, dev_name(dev));
 341         set_bit(BDI_registered, &bdi->state);
 342
 343         spin_lock_bh(&bdi_lock);
 344         list_add_tail_rcu(&bdi->bdi_list, &bdi_list);
 345         spin_unlock_bh(&bdi_lock);
 346
 347         trace_writeback_bdi_register(bdi);
 348         return 0;
 349 }
 350 EXPORT_SYMBOL(bdi_register);
 351
 352 int bdi_register_dev(struct backing_dev_info *bdi, dev_t dev)
 353 {
 354         return bdi_register(bdi, NULL, "%u:%u", MAJOR(dev), MINOR(dev));
 355 }
 356 EXPORT_SYMBOL(bdi_register_dev);
 357
 358 /*
 359  * Remove bdi from the global list and shutdown any threads we have running
 360  */
 361 static void bdi_wb_shutdown(struct backing_dev_info *bdi)
 362 {
 363         if (!bdi_cap_writeback_dirty(bdi))
 364                 return;
 365
 366         /*
 367          * Make sure nobody finds us on the bdi_list anymore
 368          */
 369         bdi_remove_from_list(bdi);
 370
 371         /* Make sure nobody queues further work */
 372         spin_lock_bh(&bdi->wb_lock);
 373         clear_bit(BDI_registered, &bdi->state);
 374         spin_unlock_bh(&bdi->wb_lock);
 375
 376         /*
 377          * Drain work list and shutdown the delayed_work.  At this point,
 378          * @bdi->bdi_list is empty telling bdi_Writeback_workfn() that @bdi
 379          * is dying and its work_list needs to be drained no matter what.
 380          */
 381         mod_delayed_work(bdi_wq, &bdi->wb.dwork, 0);
 382         flush_delayed_work(&bdi->wb.dwork);
 383         WARN_ON(!list_empty(&bdi->work_list));
 384         WARN_ON(delayed_work_pending(&bdi->wb.dwork));
 385 }
 386
 387 /*
 388  * This bdi is going away now, make sure that no super_blocks point to it
 389  */
 390 static void bdi_prune_sb(struct backing_dev_info *bdi)
 391 {
 392         struct super_block *sb;
 393
 394         spin_lock(&sb_lock);
 395         list_for_each_entry(sb, &super_blocks, s_list) {
 396                 if (sb->s_bdi == bdi)
 397                         sb->s_bdi = &default_backing_dev_info;
 398         }
 399         spin_unlock(&sb_lock);
 400 }
 401
 402 void bdi_unregister(struct backing_dev_info *bdi)
 403 {
 404         if (bdi->dev) {
 405                 bdi_set_min_ratio(bdi, 0);
 406                 trace_writeback_bdi_unregister(bdi);
 407                 bdi_prune_sb(bdi);
 408
 409                 bdi_wb_shutdown(bdi);
 410                 bdi_debug_unregister(bdi);
 411                 device_unregister(bdi->dev);
 412                 bdi->dev = NULL;
 413         }
 414 }
 415 EXPORT_SYMBOL(bdi_unregister);
 416
 417 static void bdi_wb_init(struct bdi_writeback *wb, struct backing_dev_info *bdi)
 418 {
 419         memset(wb, 0, sizeof(*wb));
 420
 421         wb->bdi = bdi;
 422         wb->last_old_flush = jiffies;
 423         INIT_LIST_HEAD(&wb->b_dirty);
 424         INIT_LIST_HEAD(&wb->b_io);
 425         INIT_LIST_HEAD(&wb->b_more_io);
 426         INIT_LIST_HEAD(&wb->b_dirty_time);
 427         spin_lock_init(&wb->list_lock);
 428         INIT_DELAYED_WORK(&wb->dwork, bdi_writeback_workfn);
 429 }
 430
 431 /*
 432  * Initial write bandwidth: 100 MB/s
 433  */
 434 #define INIT_BW         (100 << (20 - PAGE_SHIFT))
 435
 436 int bdi_init(struct backing_dev_info *bdi)
 437 {
 438         int i, err;
 439
 440         bdi->dev = NULL;
 441
 442         bdi->min_ratio = 0;
 443         bdi->max_ratio = 100;
 444         bdi->max_prop_frac = FPROP_FRAC_BASE;
 445         spin_lock_init(&bdi->wb_lock);
 446         INIT_LIST_HEAD(&bdi->bdi_list);
 447         INIT_LIST_HEAD(&bdi->work_list);
 448
 449         bdi_wb_init(&bdi->wb, bdi);
 450
 451         for (i = 0; i < NR_BDI_STAT_ITEMS; i++) {
 452                 err = percpu_counter_init(&bdi->bdi_stat[i], 0, GFP_KERNEL);
 453                 if (err)
 454                         goto err;
 455         }
 456
 457         bdi->dirty_exceeded = 0;
 458
 459         bdi->bw_time_stamp = jiffies;
 460         bdi->written_stamp = 0;
 461
 462         bdi->balanced_dirty_ratelimit = INIT_BW;
 463         bdi->dirty_ratelimit = INIT_BW;
 464         bdi->write_bandwidth = INIT_BW;
 465         bdi->avg_write_bandwidth = INIT_BW;
 466
 467         err = fprop_local_init_percpu(&bdi->completions, GFP_KERNEL);
 468
 469         if (err) {
 470 err:
 471                 while (i--)
 472                         percpu_counter_destroy(&bdi->bdi_stat[i]);
 473         }
 474
 475         return err;
 476 }
 477 EXPORT_SYMBOL(bdi_init);
 478
 479 void bdi_destroy(struct backing_dev_info *bdi)
 480 {
 481         int i;
 482
 483         /*
 484          * Splice our entries to the default_backing_dev_info.  This
 485          * condition shouldn't happen.  @wb must be empty at this point and
 486          * dirty inodes on it might cause other issues.  This workaround is
 487          * added by ce5f8e779519 ("writeback: splice dirty inode entries to
 488          * default bdi on bdi_destroy()") without root-causing the issue.
 489          *
 490          * http://lkml.kernel.org/g/1253038617-30204-11-git-send-email-jens.axboe@oracle.com
 491          * http://thread.gmane.org/gmane.linux.file-systems/35341/focus=35350
 492          *
 493          * We should probably add WARN_ON() to find out whether it still
 494          * happens and track it down if so.
 495          */
 496         if (bdi_has_dirty_io(bdi)) {
 497                 struct bdi_writeback *dst = &default_backing_dev_info.wb;
 498
 499                 bdi_lock_two(&bdi->wb, dst);
 500                 list_splice(&bdi->wb.b_dirty, &dst->b_dirty);
 501                 list_splice(&bdi->wb.b_io, &dst->b_io);
 502                 list_splice(&bdi->wb.b_more_io, &dst->b_more_io);
 503                 spin_unlock(&bdi->wb.list_lock);
 504                 spin_unlock(&dst->list_lock);
 505         }
 506
 507         bdi_unregister(bdi);
 508
 509         WARN_ON(delayed_work_pending(&bdi->wb.dwork));
 510
 511         for (i = 0; i < NR_BDI_STAT_ITEMS; i++)
 512                 percpu_counter_destroy(&bdi->bdi_stat[i]);
 513
 514         fprop_local_destroy_percpu(&bdi->completions);
 515 }
 516 EXPORT_SYMBOL(bdi_destroy);
 517
 518 /*
 519  * For use from filesystems to quickly init and register a bdi associated
 520  * with dirty writeback
 521  */
 522 int bdi_setup_and_register(struct backing_dev_info *bdi, char *name,
 523                            unsigned int cap)
 524 {
 525         int err;
 526
 527         bdi->name = name;
 528         bdi->capabilities = cap;
 529         err = bdi_init(bdi);
 530         if (err)
 531                 return err;
 532
 533         err = bdi_register(bdi, NULL, "%.28s-%ld", name,
 534                            atomic_long_inc_return(&bdi_seq));
 535         if (err) {
 536                 bdi_destroy(bdi);
 537                 return err;
 538         }
 539
 540         return 0;
 541 }
 542 EXPORT_SYMBOL(bdi_setup_and_register);
 543
 544 static wait_queue_head_t congestion_wqh[2] = {
 545                 __WAIT_QUEUE_HEAD_INITIALIZER(congestion_wqh[0]),
 546                 __WAIT_QUEUE_HEAD_INITIALIZER(congestion_wqh[1])
 547         };
 548 static atomic_t nr_bdi_congested[2];
 549
 550 void clear_bdi_congested(struct backing_dev_info *bdi, int sync)
 551 {
 552         enum bdi_state bit;
 553         wait_queue_head_t *wqh = &congestion_wqh[sync];
 554
 555         bit = sync ? BDI_sync_congested : BDI_async_congested;
 556         if (test_and_clear_bit(bit, &bdi->state))
 557                 atomic_dec(&nr_bdi_congested[sync]);
 558         smp_mb__after_atomic();
 559         if (waitqueue_active(wqh))
 560                 wake_up(wqh);
 561 }
 562 EXPORT_SYMBOL(clear_bdi_congested);
 563
 564 void set_bdi_congested(struct backing_dev_info *bdi, int sync)
 565 {
 566         enum bdi_state bit;
 567
 568         bit = sync ? BDI_sync_congested : BDI_async_congested;
 569         if (!test_and_set_bit(bit, &bdi->state))
 570                 atomic_inc(&nr_bdi_congested[sync]);
 571 }
 572 EXPORT_SYMBOL(set_bdi_congested);
 573
 574 /**
 575  * congestion_wait - wait for a backing_dev to become uncongested
 576  * @sync: SYNC or ASYNC IO
 577  * @timeout: timeout in jiffies
 578  *
 579  * Waits for up to @timeout jiffies for a backing_dev (any backing_dev) to exit
 580  * write congestion.  If no backing_devs are congested then just wait for the
 581  * next write to be completed.
 582  */
 583 long congestion_wait(int sync, long timeout)
 584 {
 585         long ret;
 586         unsigned long start = jiffies;
 587         DEFINE_WAIT(wait);
 588         wait_queue_head_t *wqh = &congestion_wqh[sync];
 589
 590         prepare_to_wait(wqh, &wait, TASK_UNINTERRUPTIBLE);
 591         ret = io_schedule_timeout(timeout);
 592         finish_wait(wqh, &wait);
 593
 594         trace_writeback_congestion_wait(jiffies_to_usecs(timeout),
 595                                         jiffies_to_usecs(jiffies - start));
 596
 597         return ret;
 598 }
 599 EXPORT_SYMBOL(congestion_wait);
 600
 601 /**
 602  * wait_iff_congested - Conditionally wait for a backing_dev to become uncongested or a zone to complete writes
 603  * @zone: A zone to check if it is heavily congested
 604  * @sync: SYNC or ASYNC IO
 605  * @timeout: timeout in jiffies
 606  *
 607  * In the event of a congested backing_dev (any backing_dev) and the given
 608  * @zone has experienced recent congestion, this waits for up to @timeout
 609  * jiffies for either a BDI to exit congestion of the given @sync queue
 610  * or a write to complete.
 611  *
 612  * In the absence of zone congestion, cond_resched() is called to yield
 613  * the processor if necessary but otherwise does not sleep.
 614  *
 615  * The return value is 0 if the sleep is for the full timeout. Otherwise,
 616  * it is the number of jiffies that were still remaining when the function
 617  * returned. return_value == timeout implies the function did not sleep.
 618  */
 619 long wait_iff_congested(struct zone *zone, int sync, long timeout)
 620 {
 621         long ret;
 622         unsigned long start = jiffies;
 623         DEFINE_WAIT(wait);
 624         wait_queue_head_t *wqh = &congestion_wqh[sync];
 625
 626         /*
 627          * If there is no congestion, or heavy congestion is not being
 628          * encountered in the current zone, yield if necessary instead
 629          * of sleeping on the congestion queue
 630          */
 631         if (atomic_read(&nr_bdi_congested[sync]) == 0 ||
 632             !test_bit(ZONE_CONGESTED, &zone->flags)) {
 633                 cond_resched();
 634
 635                 /* In case we scheduled, work out time remaining */
 636                 ret = timeout - (jiffies - start);
 637                 if (ret < 0)
 638                         ret = 0;
 639
 640                 goto out;
 641         }
 642
 643         /* Sleep until uncongested or a write happens */
 644         prepare_to_wait(wqh, &wait, TASK_UNINTERRUPTIBLE);
 645         ret = io_schedule_timeout(timeout);
 646         finish_wait(wqh, &wait);
 647
 648 out:
 649         trace_writeback_wait_iff_congested(jiffies_to_usecs(timeout),
 650                                         jiffies_to_usecs(jiffies - start));
 651
 652         return ret;
 653 }
 654 EXPORT_SYMBOL(wait_iff_congested);
 655
 656 int pdflush_proc_obsolete(struct ctl_table *table, int write,
 657                         void __user *buffer, size_t *lenp, loff_t *ppos)
 658 {
 659         char kbuf[] = "0\n";
 660
 661         if (*ppos || *lenp < sizeof(kbuf)) {
 662                 *lenp = 0;
 663                 return 0;
 664         }
 665
 666         if (copy_to_user(buffer, kbuf, sizeof(kbuf)))
 667                 return -EFAULT;
 668         printk_once(KERN_WARNING "%s exported in /proc is scheduled for removal\n",
 669                         table->procname);
 670
 671         *lenp = 2;
 672         *ppos += *lenp;
 673         return 2;
 674 }