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 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;
  73         struct inode *inode;
  74
  75         nr_dirty = nr_io = nr_more_io = 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         spin_unlock(&wb->list_lock);
  84
  85         global_dirty_limits(&background_thresh, &dirty_thresh);
  86         bdi_thresh = bdi_dirty_limit(bdi, dirty_thresh);
  87
  88 #define K(x) ((x) << (PAGE_SHIFT - 10))
  89         seq_printf(m,
  90                    "BdiWriteback:       %10lu kB\n"
  91                    "BdiReclaimable:     %10lu kB\n"
  92                    "BdiDirtyThresh:     %10lu kB\n"
  93                    "DirtyThresh:        %10lu kB\n"
  94                    "BackgroundThresh:   %10lu kB\n"
  95                    "BdiDirtied:         %10lu kB\n"
  96                    "BdiWritten:         %10lu kB\n"
  97                    "BdiWriteBandwidth:  %10lu kBps\n"
  98                    "b_dirty:            %10lu\n"
  99                    "b_io:               %10lu\n"
 100                    "b_more_io:          %10lu\n"
 101                    "bdi_list:           %10u\n"
 102                    "state:              %10lx\n",
 103                    (unsigned long) K(bdi_stat(bdi, BDI_WRITEBACK)),
 104                    (unsigned long) K(bdi_stat(bdi, BDI_RECLAIMABLE)),
 105                    K(bdi_thresh),
 106                    K(dirty_thresh),
 107                    K(background_thresh),
 108                    (unsigned long) K(bdi_stat(bdi, BDI_DIRTIED)),
 109                    (unsigned long) K(bdi_stat(bdi, BDI_WRITTEN)),
 110                    (unsigned long) K(bdi->write_bandwidth),
 111                    nr_dirty,
 112                    nr_io,
 113                    nr_more_io,
 114                    !list_empty(&bdi->bdi_list), bdi->state);
 115 #undef K
 116
 117         return 0;
 118 }
 119
 120 static int bdi_debug_stats_open(struct inode *inode, struct file *file)
 121 {
 122         return single_open(file, bdi_debug_stats_show, inode->i_private);
 123 }
 124
 125 static const struct file_operations bdi_debug_stats_fops = {
 126         .open           = bdi_debug_stats_open,
 127         .read           = seq_read,
 128         .llseek         = seq_lseek,
 129         .release        = single_release,
 130 };
 131
 132 static void bdi_debug_register(struct backing_dev_info *bdi, const char *name)
 133 {
 134         bdi->debug_dir = debugfs_create_dir(name, bdi_debug_root);
 135         bdi->debug_stats = debugfs_create_file("stats", 0444, bdi->debug_dir,
 136                                                bdi, &bdi_debug_stats_fops);
 137 }
 138
 139 static void bdi_debug_unregister(struct backing_dev_info *bdi)
 140 {
 141         debugfs_remove(bdi->debug_stats);
 142         debugfs_remove(bdi->debug_dir);
 143 }
 144 #else
 145 static inline void bdi_debug_init(void)
 146 {
 147 }
 148 static inline void bdi_debug_register(struct backing_dev_info *bdi,
 149                                       const char *name)
 150 {
 151 }
 152 static inline void bdi_debug_unregister(struct backing_dev_info *bdi)
 153 {
 154 }
 155 #endif
 156
 157 static ssize_t read_ahead_kb_store(struct device *dev,
 158                                   struct device_attribute *attr,
 159                                   const char *buf, size_t count)
 160 {
 161         struct backing_dev_info *bdi = dev_get_drvdata(dev);
 162         unsigned long read_ahead_kb;
 163         ssize_t ret;
 164
 165         ret = kstrtoul(buf, 10, &read_ahead_kb);
 166         if (ret < 0)
 167                 return ret;
 168
 169         bdi->ra_pages = read_ahead_kb >> (PAGE_SHIFT - 10);
 170
 171         return count;
 172 }
 173
 174 #define K(pages) ((pages) << (PAGE_SHIFT - 10))
 175
 176 #define BDI_SHOW(name, expr)                                            \
 177 static ssize_t name##_show(struct device *dev,                          \
 178                            struct device_attribute *attr, char *page)   \
 179 {                                                                       \
 180         struct backing_dev_info *bdi = dev_get_drvdata(dev);            \
 181                                                                         \
 182         return snprintf(page, PAGE_SIZE-1, "%lld\n", (long long)expr);  \
 183 }                                                                       \
 184 static DEVICE_ATTR_RW(name);
 185
 186 BDI_SHOW(read_ahead_kb, K(bdi->ra_pages))
 187
 188 static ssize_t min_ratio_store(struct device *dev,
 189                 struct device_attribute *attr, const char *buf, size_t count)
 190 {
 191         struct backing_dev_info *bdi = dev_get_drvdata(dev);
 192         unsigned int ratio;
 193         ssize_t ret;
 194
 195         ret = kstrtouint(buf, 10, &ratio);
 196         if (ret < 0)
 197                 return ret;
 198
 199         ret = bdi_set_min_ratio(bdi, ratio);
 200         if (!ret)
 201                 ret = count;
 202
 203         return ret;
 204 }
 205 BDI_SHOW(min_ratio, bdi->min_ratio)
 206
 207 static ssize_t max_ratio_store(struct device *dev,
 208                 struct device_attribute *attr, const char *buf, size_t count)
 209 {
 210         struct backing_dev_info *bdi = dev_get_drvdata(dev);
 211         unsigned int ratio;
 212         ssize_t ret;
 213
 214         ret = kstrtouint(buf, 10, &ratio);
 215         if (ret < 0)
 216                 return ret;
 217
 218         ret = bdi_set_max_ratio(bdi, ratio);
 219         if (!ret)
 220                 ret = count;
 221
 222         return ret;
 223 }
 224 BDI_SHOW(max_ratio, bdi->max_ratio)
 225
 226 static ssize_t stable_pages_required_show(struct device *dev,
 227                                           struct device_attribute *attr,
 228                                           char *page)
 229 {
 230         struct backing_dev_info *bdi = dev_get_drvdata(dev);
 231
 232         return snprintf(page, PAGE_SIZE-1, "%d\n",
 233                         bdi_cap_stable_pages_required(bdi) ? 1 : 0);
 234 }
 235 static DEVICE_ATTR_RO(stable_pages_required);
 236
 237 static struct attribute *bdi_dev_attrs[] = {
 238         &dev_attr_read_ahead_kb.attr,
 239         &dev_attr_min_ratio.attr,
 240         &dev_attr_max_ratio.attr,
 241         &dev_attr_stable_pages_required.attr,
 242         NULL,
 243 };
 244 ATTRIBUTE_GROUPS(bdi_dev);
 245
 246 static __init int bdi_class_init(void)
 247 {
 248         bdi_class = class_create(THIS_MODULE, "bdi");
 249         if (IS_ERR(bdi_class))
 250                 return PTR_ERR(bdi_class);
 251
 252         bdi_class->dev_groups = bdi_dev_groups;
 253         bdi_debug_init();
 254         return 0;
 255 }
 256 postcore_initcall(bdi_class_init);
 257
 258 static int __init default_bdi_init(void)
 259 {
 260         int err;
 261
 262         bdi_wq = alloc_workqueue("writeback", WQ_MEM_RECLAIM | WQ_FREEZABLE |
 263                                               WQ_UNBOUND | WQ_SYSFS, 0);
 264         if (!bdi_wq)
 265                 return -ENOMEM;
 266
 267         err = bdi_init(&default_backing_dev_info);
 268         if (!err)
 269                 bdi_register(&default_backing_dev_info, NULL, "default");
 270         err = bdi_init(&noop_backing_dev_info);
 271
 272         return err;
 273 }
 274 subsys_initcall(default_bdi_init);
 275
 276 int bdi_has_dirty_io(struct backing_dev_info *bdi)
 277 {
 278         return wb_has_dirty_io(&bdi->wb);
 279 }
 280
 281 /*
 282  * This function is used when the first inode for this bdi is marked dirty. It
 283  * wakes-up the corresponding bdi thread which should then take care of the
 284  * periodic background write-out of dirty inodes. Since the write-out would
 285  * starts only 'dirty_writeback_interval' centisecs from now anyway, we just
 286  * set up a timer which wakes the bdi thread up later.
 287  *
 288  * Note, we wouldn't bother setting up the timer, but this function is on the
 289  * fast-path (used by '__mark_inode_dirty()'), so we save few context switches
 290  * by delaying the wake-up.
 291  *
 292  * We have to be careful not to postpone flush work if it is scheduled for
 293  * earlier. Thus we use queue_delayed_work().
 294  */
 295 void bdi_wakeup_thread_delayed(struct backing_dev_info *bdi)
 296 {
 297         unsigned long timeout;
 298
 299         timeout = msecs_to_jiffies(dirty_writeback_interval * 10);
 300         spin_lock_bh(&bdi->wb_lock);
 301         if (test_bit(BDI_registered, &bdi->state))
 302                 queue_delayed_work(bdi_wq, &bdi->wb.dwork, timeout);
 303         spin_unlock_bh(&bdi->wb_lock);
 304 }
 305
 306 /*
 307  * Remove bdi from bdi_list, and ensure that it is no longer visible
 308  */
 309 static void bdi_remove_from_list(struct backing_dev_info *bdi)
 310 {
 311         spin_lock_bh(&bdi_lock);
 312         list_del_rcu(&bdi->bdi_list);
 313         spin_unlock_bh(&bdi_lock);
 314
 315         synchronize_rcu_expedited();
 316 }
 317
 318 int bdi_register(struct backing_dev_info *bdi, struct device *parent,
 319                 const char *fmt, ...)
 320 {
 321         va_list args;
 322         struct device *dev;
 323
 324         if (bdi->dev)   /* The driver needs to use separate queues per device */
 325                 return 0;
 326
 327         va_start(args, fmt);
 328         dev = device_create_vargs(bdi_class, parent, MKDEV(0, 0), bdi, fmt, args);
 329         va_end(args);
 330         if (IS_ERR(dev))
 331                 return PTR_ERR(dev);
 332
 333         bdi->dev = dev;
 334
 335         bdi_debug_register(bdi, dev_name(dev));
 336         set_bit(BDI_registered, &bdi->state);
 337
 338         spin_lock_bh(&bdi_lock);
 339         list_add_tail_rcu(&bdi->bdi_list, &bdi_list);
 340         spin_unlock_bh(&bdi_lock);
 341
 342         trace_writeback_bdi_register(bdi);
 343         return 0;
 344 }
 345 EXPORT_SYMBOL(bdi_register);
 346
 347 int bdi_register_dev(struct backing_dev_info *bdi, dev_t dev)
 348 {
 349         return bdi_register(bdi, NULL, "%u:%u", MAJOR(dev), MINOR(dev));
 350 }
 351 EXPORT_SYMBOL(bdi_register_dev);
 352
 353 int bdi_register_owner(struct backing_dev_info *bdi, struct device *owner)
 354 {
 355         int rc;
 356
 357         rc = bdi_register(bdi, NULL, "%u:%u", MAJOR(owner->devt),
 358                         MINOR(owner->devt));
 359         if (rc)
 360                 return rc;
 361         bdi->owner = owner;
 362         get_device(owner);
 363         return 0;
 364 }
 365 EXPORT_SYMBOL(bdi_register_owner);
 366
 367 /*
 368  * Remove bdi from the global list and shutdown any threads we have running
 369  */
 370 static void bdi_wb_shutdown(struct backing_dev_info *bdi)
 371 {
 372         if (!bdi_cap_writeback_dirty(bdi))
 373                 return;
 374
 375         /*
 376          * Make sure nobody finds us on the bdi_list anymore
 377          */
 378         bdi_remove_from_list(bdi);
 379
 380         /* Make sure nobody queues further work */
 381         spin_lock_bh(&bdi->wb_lock);
 382         clear_bit(BDI_registered, &bdi->state);
 383         spin_unlock_bh(&bdi->wb_lock);
 384
 385         /*
 386          * Drain work list and shutdown the delayed_work.  At this point,
 387          * @bdi->bdi_list is empty telling bdi_Writeback_workfn() that @bdi
 388          * is dying and its work_list needs to be drained no matter what.
 389          */
 390         mod_delayed_work(bdi_wq, &bdi->wb.dwork, 0);
 391         flush_delayed_work(&bdi->wb.dwork);
 392         WARN_ON(!list_empty(&bdi->work_list));
 393
 394         /*
 395          * This shouldn't be necessary unless @bdi for some reason has
 396          * unflushed dirty IO after work_list is drained.  Do it anyway
 397          * just in case.
 398          */
 399         cancel_delayed_work_sync(&bdi->wb.dwork);
 400 }
 401
 402 /*
 403  * This bdi is going away now, make sure that no super_blocks point to it
 404  */
 405 static void bdi_prune_sb(struct backing_dev_info *bdi)
 406 {
 407         struct super_block *sb;
 408
 409         spin_lock(&sb_lock);
 410         list_for_each_entry(sb, &super_blocks, s_list) {
 411                 if (sb->s_bdi == bdi)
 412                         sb->s_bdi = &default_backing_dev_info;
 413         }
 414         spin_unlock(&sb_lock);
 415 }
 416
 417 void bdi_unregister(struct backing_dev_info *bdi)
 418 {
 419         struct device *dev = bdi->dev;
 420
 421         if (dev) {
 422                 bdi_set_min_ratio(bdi, 0);
 423                 trace_writeback_bdi_unregister(bdi);
 424                 bdi_prune_sb(bdi);
 425
 426                 bdi_wb_shutdown(bdi);
 427                 bdi_debug_unregister(bdi);
 428
 429                 spin_lock_bh(&bdi->wb_lock);
 430                 bdi->dev = NULL;
 431                 spin_unlock_bh(&bdi->wb_lock);
 432
 433                 device_unregister(dev);
 434         }
 435
 436         if (bdi->owner) {
 437                 put_device(bdi->owner);
 438                 bdi->owner = NULL;
 439         }
 440 }
 441 EXPORT_SYMBOL(bdi_unregister);
 442
 443 static void bdi_wb_init(struct bdi_writeback *wb, struct backing_dev_info *bdi)
 444 {
 445         memset(wb, 0, sizeof(*wb));
 446
 447         wb->bdi = bdi;
 448         wb->last_old_flush = jiffies;
 449         INIT_LIST_HEAD(&wb->b_dirty);
 450         INIT_LIST_HEAD(&wb->b_io);
 451         INIT_LIST_HEAD(&wb->b_more_io);
 452         spin_lock_init(&wb->list_lock);
 453         INIT_DELAYED_WORK(&wb->dwork, bdi_writeback_workfn);
 454 }
 455
 456 /*
 457  * Initial write bandwidth: 100 MB/s
 458  */
 459 #define INIT_BW         (100 << (20 - PAGE_SHIFT))
 460
 461 int bdi_init(struct backing_dev_info *bdi)
 462 {
 463         int i, err;
 464
 465         bdi->dev = NULL;
 466
 467         bdi->min_ratio = 0;
 468         bdi->max_ratio = 100;
 469         bdi->max_prop_frac = FPROP_FRAC_BASE;
 470         spin_lock_init(&bdi->wb_lock);
 471         INIT_LIST_HEAD(&bdi->bdi_list);
 472         INIT_LIST_HEAD(&bdi->work_list);
 473
 474         bdi_wb_init(&bdi->wb, bdi);
 475
 476         for (i = 0; i < NR_BDI_STAT_ITEMS; i++) {
 477                 err = percpu_counter_init(&bdi->bdi_stat[i], 0);
 478                 if (err)
 479                         goto err;
 480         }
 481
 482         bdi->dirty_exceeded = 0;
 483
 484         bdi->bw_time_stamp = jiffies;
 485         bdi->written_stamp = 0;
 486
 487         bdi->balanced_dirty_ratelimit = INIT_BW;
 488         bdi->dirty_ratelimit = INIT_BW;
 489         bdi->write_bandwidth = INIT_BW;
 490         bdi->avg_write_bandwidth = INIT_BW;
 491
 492         err = fprop_local_init_percpu(&bdi->completions);
 493
 494         if (err) {
 495 err:
 496                 while (i--)
 497                         percpu_counter_destroy(&bdi->bdi_stat[i]);
 498         }
 499
 500         return err;
 501 }
 502 EXPORT_SYMBOL(bdi_init);
 503
 504 void bdi_destroy(struct backing_dev_info *bdi)
 505 {
 506         int i;
 507
 508         /*
 509          * Splice our entries to the default_backing_dev_info, if this
 510          * bdi disappears
 511          */
 512         if (bdi_has_dirty_io(bdi)) {
 513                 struct bdi_writeback *dst = &default_backing_dev_info.wb;
 514
 515                 bdi_lock_two(&bdi->wb, dst);
 516                 list_splice(&bdi->wb.b_dirty, &dst->b_dirty);
 517                 list_splice(&bdi->wb.b_io, &dst->b_io);
 518                 list_splice(&bdi->wb.b_more_io, &dst->b_more_io);
 519                 spin_unlock(&bdi->wb.list_lock);
 520                 spin_unlock(&dst->list_lock);
 521         }
 522
 523         bdi_unregister(bdi);
 524
 525         /*
 526          * If bdi_unregister() had already been called earlier, the dwork
 527          * could still be pending because bdi_prune_sb() can race with the
 528          * bdi_wakeup_thread_delayed() calls from __mark_inode_dirty().
 529          */
 530         cancel_delayed_work_sync(&bdi->wb.dwork);
 531
 532         for (i = 0; i < NR_BDI_STAT_ITEMS; i++)
 533                 percpu_counter_destroy(&bdi->bdi_stat[i]);
 534
 535         fprop_local_destroy_percpu(&bdi->completions);
 536 }
 537 EXPORT_SYMBOL(bdi_destroy);
 538
 539 /*
 540  * For use from filesystems to quickly init and register a bdi associated
 541  * with dirty writeback
 542  */
 543 int bdi_setup_and_register(struct backing_dev_info *bdi, char *name,
 544                            unsigned int cap)
 545 {
 546         int err;
 547
 548         bdi->name = name;
 549         bdi->capabilities = cap;
 550         err = bdi_init(bdi);
 551         if (err)
 552                 return err;
 553
 554         err = bdi_register(bdi, NULL, "%.28s-%ld", name,
 555                            atomic_long_inc_return(&bdi_seq));
 556         if (err) {
 557                 bdi_destroy(bdi);
 558                 return err;
 559         }
 560
 561         return 0;
 562 }
 563 EXPORT_SYMBOL(bdi_setup_and_register);
 564
 565 static wait_queue_head_t congestion_wqh[2] = {
 566                 __WAIT_QUEUE_HEAD_INITIALIZER(congestion_wqh[0]),
 567                 __WAIT_QUEUE_HEAD_INITIALIZER(congestion_wqh[1])
 568         };
 569 static atomic_t nr_bdi_congested[2];
 570
 571 void clear_bdi_congested(struct backing_dev_info *bdi, int sync)
 572 {
 573         enum bdi_state bit;
 574         wait_queue_head_t *wqh = &congestion_wqh[sync];
 575
 576         bit = sync ? BDI_sync_congested : BDI_async_congested;
 577         if (test_and_clear_bit(bit, &bdi->state))
 578                 atomic_dec(&nr_bdi_congested[sync]);
 579         smp_mb__after_atomic();
 580         if (waitqueue_active(wqh))
 581                 wake_up(wqh);
 582 }
 583 EXPORT_SYMBOL(clear_bdi_congested);
 584
 585 void set_bdi_congested(struct backing_dev_info *bdi, int sync)
 586 {
 587         enum bdi_state bit;
 588
 589         bit = sync ? BDI_sync_congested : BDI_async_congested;
 590         if (!test_and_set_bit(bit, &bdi->state))
 591                 atomic_inc(&nr_bdi_congested[sync]);
 592 }
 593 EXPORT_SYMBOL(set_bdi_congested);
 594
 595 /**
 596  * congestion_wait - wait for a backing_dev to become uncongested
 597  * @sync: SYNC or ASYNC IO
 598  * @timeout: timeout in jiffies
 599  *
 600  * Waits for up to @timeout jiffies for a backing_dev (any backing_dev) to exit
 601  * write congestion.  If no backing_devs are congested then just wait for the
 602  * next write to be completed.
 603  */
 604 long congestion_wait(int sync, long timeout)
 605 {
 606         long ret;
 607         unsigned long start = jiffies;
 608         DEFINE_WAIT(wait);
 609         wait_queue_head_t *wqh = &congestion_wqh[sync];
 610
 611         prepare_to_wait(wqh, &wait, TASK_UNINTERRUPTIBLE);
 612         ret = io_schedule_timeout(timeout);
 613         finish_wait(wqh, &wait);
 614
 615         trace_writeback_congestion_wait(jiffies_to_usecs(timeout),
 616                                         jiffies_to_usecs(jiffies - start));
 617
 618         return ret;
 619 }
 620 EXPORT_SYMBOL(congestion_wait);
 621
 622 /**
 623  * wait_iff_congested - Conditionally wait for a backing_dev to become uncongested or a zone to complete writes
 624  * @zone: A zone to check if it is heavily congested
 625  * @sync: SYNC or ASYNC IO
 626  * @timeout: timeout in jiffies
 627  *
 628  * In the event of a congested backing_dev (any backing_dev) and the given
 629  * @zone has experienced recent congestion, this waits for up to @timeout
 630  * jiffies for either a BDI to exit congestion of the given @sync queue
 631  * or a write to complete.
 632  *
 633  * In the absence of zone congestion, a short sleep or a cond_resched is
 634  * performed to yield the processor and to allow other subsystems to make
 635  * a forward progress.
 636  *
 637  * The return value is 0 if the sleep is for the full timeout. Otherwise,
 638  * it is the number of jiffies that were still remaining when the function
 639  * returned. return_value == timeout implies the function did not sleep.
 640  */
 641 long wait_iff_congested(struct zone *zone, int sync, long timeout)
 642 {
 643         long ret;
 644         unsigned long start = jiffies;
 645         DEFINE_WAIT(wait);
 646         wait_queue_head_t *wqh = &congestion_wqh[sync];
 647
 648         /*
 649          * If there is no congestion, or heavy congestion is not being
 650          * encountered in the current zone, yield if necessary instead
 651          * of sleeping on the congestion queue
 652          */
 653         if (atomic_read(&nr_bdi_congested[sync]) == 0 ||
 654                         !zone_is_reclaim_congested(zone)) {
 655
 656                 /*
 657                  * Memory allocation/reclaim might be called from a WQ
 658                  * context and the current implementation of the WQ
 659                  * concurrency control doesn't recognize that a particular
 660                  * WQ is congested if the worker thread is looping without
 661                  * ever sleeping. Therefore we have to do a short sleep
 662                  * here rather than calling cond_resched().
 663                  */
 664                 if (current->flags & PF_WQ_WORKER)
 665                         schedule_timeout_uninterruptible(1);
 666                 else
 667                         cond_resched();
 668
 669                 /* In case we scheduled, work out time remaining */
 670                 ret = timeout - (jiffies - start);
 671                 if (ret < 0)
 672                         ret = 0;
 673
 674                 goto out;
 675         }
 676
 677         /* Sleep until uncongested or a write happens */
 678         prepare_to_wait(wqh, &wait, TASK_UNINTERRUPTIBLE);
 679         ret = io_schedule_timeout(timeout);
 680         finish_wait(wqh, &wait);
 681
 682 out:
 683         trace_writeback_wait_iff_congested(jiffies_to_usecs(timeout),
 684                                         jiffies_to_usecs(jiffies - start));
 685
 686         return ret;
 687 }
 688 EXPORT_SYMBOL(wait_iff_congested);
 689
 690 int pdflush_proc_obsolete(struct ctl_table *table, int write,
 691                         void __user *buffer, size_t *lenp, loff_t *ppos)
 692 {
 693         char kbuf[] = "0\n";
 694
 695         if (*ppos || *lenp < sizeof(kbuf)) {
 696                 *lenp = 0;
 697                 return 0;
 698         }
 699
 700         if (copy_to_user(buffer, kbuf, sizeof(kbuf)))
 701                 return -EFAULT;
 702         printk_once(KERN_WARNING "%s exported in /proc is scheduled for removal\n",
 703                         table->procname);
 704
 705         *lenp = 2;
 706         *ppos += *lenp;
 707         return 2;
 708 }