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;
  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         /*
  76          * inode lock is enough here, the bdi->wb_list is protected by
  77          * RCU on the reader side
  78          */
  79         nr_wb = nr_dirty = nr_io = nr_more_io = 0;
  80         spin_lock(&inode_lock);
  81         list_for_each_entry(wb, &bdi->wb_list, list) {
  82                 nr_wb++;
  83                 list_for_each_entry(inode, &wb->b_dirty, i_list)
  84                         nr_dirty++;
  85                 list_for_each_entry(inode, &wb->b_io, i_list)
  86                         nr_io++;
  87                 list_for_each_entry(inode, &wb->b_more_io, i_list)
  88                         nr_more_io++;
  89         }
  90         spin_unlock(&inode_lock);
  91
  92         get_dirty_limits(&background_thresh, &dirty_thresh, &bdi_thresh, bdi);
  93
  94 #define K(x) ((x) << (PAGE_SHIFT - 10))
  95         seq_printf(m,
  96                    "BdiWriteback:     %8lu kB\n"
  97                    "BdiReclaimable:   %8lu kB\n"
  98                    "BdiDirtyThresh:   %8lu kB\n"
  99                    "DirtyThresh:      %8lu kB\n"
 100                    "BackgroundThresh: %8lu kB\n"
 101                    "WritebackThreads: %8lu\n"
 102                    "b_dirty:          %8lu\n"
 103                    "b_io:             %8lu\n"
 104                    "b_more_io:        %8lu\n"
 105                    "bdi_list:         %8u\n"
 106                    "state:            %8lx\n"
 107                    "wb_mask:          %8lx\n"
 108                    "wb_list:          %8u\n"
 109                    "wb_cnt:           %8u\n",
 110                    (unsigned long) K(bdi_stat(bdi, BDI_WRITEBACK)),
 111                    (unsigned long) K(bdi_stat(bdi, BDI_RECLAIMABLE)),
 112                    K(bdi_thresh), K(dirty_thresh),
 113                    K(background_thresh), nr_wb, nr_dirty, nr_io, nr_more_io,
 114                    !list_empty(&bdi->bdi_list), bdi->state, bdi->wb_mask,
 115                    !list_empty(&bdi->wb_list), bdi->wb_cnt);
 116 #undef K
 117
 118         return 0;
 119 }
 120
 121 static int bdi_debug_stats_open(struct inode *inode, struct file *file)
 122 {
 123         return single_open(file, bdi_debug_stats_show, inode->i_private);
 124 }
 125
 126 static const struct file_operations bdi_debug_stats_fops = {
 127         .open           = bdi_debug_stats_open,
 128         .read           = seq_read,
 129         .llseek         = seq_lseek,
 130         .release        = single_release,
 131 };
 132
 133 static void bdi_debug_register(struct backing_dev_info *bdi, const char *name)
 134 {
 135         bdi->debug_dir = debugfs_create_dir(name, bdi_debug_root);
 136         bdi->debug_stats = debugfs_create_file("stats", 0444, bdi->debug_dir,
 137                                                bdi, &bdi_debug_stats_fops);
 138 }
 139
 140 static void bdi_debug_unregister(struct backing_dev_info *bdi)
 141 {
 142         debugfs_remove(bdi->debug_stats);
 143         debugfs_remove(bdi->debug_dir);
 144 }
 145 #else
 146 static inline void bdi_debug_init(void)
 147 {
 148 }
 149 static inline void bdi_debug_register(struct backing_dev_info *bdi,
 150                                       const char *name)
 151 {
 152 }
 153 static inline void bdi_debug_unregister(struct backing_dev_info *bdi)
 154 {
 155 }
 156 #endif
 157
 158 static ssize_t read_ahead_kb_store(struct device *dev,
 159                                   struct device_attribute *attr,
 160                                   const char *buf, size_t count)
 161 {
 162         struct backing_dev_info *bdi = dev_get_drvdata(dev);
 163         char *end;
 164         unsigned long read_ahead_kb;
 165         ssize_t ret = -EINVAL;
 166
 167         read_ahead_kb = simple_strtoul(buf, &end, 10);
 168         if (*buf && (end[0] == '\0' || (end[0] == '\n' && end[1] == '\0'))) {
 169                 bdi->ra_pages = read_ahead_kb >> (PAGE_SHIFT - 10);
 170                 ret = count;
 171         }
 172         return ret;
 173 }
 174
 175 #define K(pages) ((pages) << (PAGE_SHIFT - 10))
 176
 177 #define BDI_SHOW(name, expr)                                            \
 178 static ssize_t name##_show(struct device *dev,                          \
 179                            struct device_attribute *attr, char *page)   \
 180 {                                                                       \
 181         struct backing_dev_info *bdi = dev_get_drvdata(dev);            \
 182                                                                         \
 183         return snprintf(page, PAGE_SIZE-1, "%lld\n", (long long)expr);  \
 184 }
 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         char *end;
 193         unsigned int ratio;
 194         ssize_t ret = -EINVAL;
 195
 196         ratio = simple_strtoul(buf, &end, 10);
 197         if (*buf && (end[0] == '\0' || (end[0] == '\n' && end[1] == '\0'))) {
 198                 ret = bdi_set_min_ratio(bdi, ratio);
 199                 if (!ret)
 200                         ret = count;
 201         }
 202         return ret;
 203 }
 204 BDI_SHOW(min_ratio, bdi->min_ratio)
 205
 206 static ssize_t max_ratio_store(struct device *dev,
 207                 struct device_attribute *attr, const char *buf, size_t count)
 208 {
 209         struct backing_dev_info *bdi = dev_get_drvdata(dev);
 210         char *end;
 211         unsigned int ratio;
 212         ssize_t ret = -EINVAL;
 213
 214         ratio = simple_strtoul(buf, &end, 10);
 215         if (*buf && (end[0] == '\0' || (end[0] == '\n' && end[1] == '\0'))) {
 216                 ret = bdi_set_max_ratio(bdi, ratio);
 217                 if (!ret)
 218                         ret = count;
 219         }
 220         return ret;
 221 }
 222 BDI_SHOW(max_ratio, bdi->max_ratio)
 223
 224 #define __ATTR_RW(attr) __ATTR(attr, 0644, attr##_show, attr##_store)
 225
 226 static struct device_attribute bdi_dev_attrs[] = {
 227         __ATTR_RW(read_ahead_kb),
 228         __ATTR_RW(min_ratio),
 229         __ATTR_RW(max_ratio),
 230         __ATTR_NULL,
 231 };
 232
 233 static __init int bdi_class_init(void)
 234 {
 235         bdi_class = class_create(THIS_MODULE, "bdi");
 236         if (IS_ERR(bdi_class))
 237                 return PTR_ERR(bdi_class);
 238
 239         bdi_class->dev_attrs = bdi_dev_attrs;
 240         bdi_debug_init();
 241         return 0;
 242 }
 243 postcore_initcall(bdi_class_init);
 244
 245 static int __init default_bdi_init(void)
 246 {
 247         int err;
 248
 249         sync_supers_tsk = kthread_run(bdi_sync_supers, NULL, "sync_supers");
 250         BUG_ON(IS_ERR(sync_supers_tsk));
 251
 252         init_timer(&sync_supers_timer);
 253         setup_timer(&sync_supers_timer, sync_supers_timer_fn, 0);
 254         bdi_arm_supers_timer();
 255
 256         err = bdi_init(&default_backing_dev_info);
 257         if (!err)
 258                 bdi_register(&default_backing_dev_info, NULL, "default");
 259
 260         return err;
 261 }
 262 subsys_initcall(default_bdi_init);
 263
 264 static void bdi_wb_init(struct bdi_writeback *wb, struct backing_dev_info *bdi)
 265 {
 266         memset(wb, 0, sizeof(*wb));
 267
 268         wb->bdi = bdi;
 269         wb->last_old_flush = jiffies;
 270         INIT_LIST_HEAD(&wb->b_dirty);
 271         INIT_LIST_HEAD(&wb->b_io);
 272         INIT_LIST_HEAD(&wb->b_more_io);
 273 }
 274
 275 static void bdi_task_init(struct backing_dev_info *bdi,
 276                           struct bdi_writeback *wb)
 277 {
 278         struct task_struct *tsk = current;
 279
 280         spin_lock(&bdi->wb_lock);
 281         list_add_tail_rcu(&wb->list, &bdi->wb_list);
 282         spin_unlock(&bdi->wb_lock);
 283
 284         tsk->flags |= PF_FLUSHER | PF_SWAPWRITE;
 285         set_freezable();
 286
 287         /*
 288          * Our parent may run at a different priority, just set us to normal
 289          */
 290         set_user_nice(tsk, 0);
 291 }
 292
 293 static int bdi_start_fn(void *ptr)
 294 {
 295         struct bdi_writeback *wb = ptr;
 296         struct backing_dev_info *bdi = wb->bdi;
 297         int ret;
 298
 299         /*
 300          * Add us to the active bdi_list
 301          */
 302         spin_lock_bh(&bdi_lock);
 303         list_add_rcu(&bdi->bdi_list, &bdi_list);
 304         spin_unlock_bh(&bdi_lock);
 305
 306         bdi_task_init(bdi, wb);
 307
 308         /*
 309          * Clear pending bit and wakeup anybody waiting to tear us down
 310          */
 311         clear_bit(BDI_pending, &bdi->state);
 312         smp_mb__after_clear_bit();
 313         wake_up_bit(&bdi->state, BDI_pending);
 314
 315         ret = bdi_writeback_task(wb);
 316
 317         /*
 318          * Remove us from the list
 319          */
 320         spin_lock(&bdi->wb_lock);
 321         list_del_rcu(&wb->list);
 322         spin_unlock(&bdi->wb_lock);
 323
 324         /*
 325          * Flush any work that raced with us exiting. No new work
 326          * will be added, since this bdi isn't discoverable anymore.
 327          */
 328         if (!list_empty(&bdi->work_list))
 329                 wb_do_writeback(wb, 1);
 330
 331         wb->task = NULL;
 332         return ret;
 333 }
 334
 335 int bdi_has_dirty_io(struct backing_dev_info *bdi)
 336 {
 337         return wb_has_dirty_io(&bdi->wb);
 338 }
 339
 340 static void bdi_flush_io(struct backing_dev_info *bdi)
 341 {
 342         struct writeback_control wbc = {
 343                 .bdi                    = bdi,
 344                 .sync_mode              = WB_SYNC_NONE,
 345                 .older_than_this        = NULL,
 346                 .range_cyclic           = 1,
 347                 .nr_to_write            = 1024,
 348         };
 349
 350         writeback_inodes_wbc(&wbc);
 351 }
 352
 353 /*
 354  * kupdated() used to do this. We cannot do it from the bdi_forker_task()
 355  * or we risk deadlocking on ->s_umount. The longer term solution would be
 356  * to implement sync_supers_bdi() or similar and simply do it from the
 357  * bdi writeback tasks individually.
 358  */
 359 static int bdi_sync_supers(void *unused)
 360 {
 361         set_user_nice(current, 0);
 362
 363         while (!kthread_should_stop()) {
 364                 set_current_state(TASK_INTERRUPTIBLE);
 365                 schedule();
 366
 367                 /*
 368                  * Do this periodically, like kupdated() did before.
 369                  */
 370                 sync_supers();
 371         }
 372
 373         return 0;
 374 }
 375
 376 void bdi_arm_supers_timer(void)
 377 {
 378         unsigned long next;
 379
 380         if (!dirty_writeback_interval)
 381                 return;
 382
 383         next = msecs_to_jiffies(dirty_writeback_interval * 10) + jiffies;
 384         mod_timer(&sync_supers_timer, round_jiffies_up(next));
 385 }
 386
 387 static void sync_supers_timer_fn(unsigned long unused)
 388 {
 389         wake_up_process(sync_supers_tsk);
 390         bdi_arm_supers_timer();
 391 }
 392
 393 static int bdi_forker_task(void *ptr)
 394 {
 395         struct bdi_writeback *me = ptr;
 396
 397         bdi_task_init(me->bdi, me);
 398
 399         for (;;) {
 400                 struct backing_dev_info *bdi, *tmp;
 401                 struct bdi_writeback *wb;
 402
 403                 /*
 404                  * Temporary measure, we want to make sure we don't see
 405                  * dirty data on the default backing_dev_info
 406                  */
 407                 if (wb_has_dirty_io(me) || !list_empty(&me->bdi->work_list))
 408                         wb_do_writeback(me, 0);
 409
 410                 spin_lock_bh(&bdi_lock);
 411
 412                 /*
 413                  * Check if any existing bdi's have dirty data without
 414                  * a thread registered. If so, set that up.
 415                  */
 416                 list_for_each_entry_safe(bdi, tmp, &bdi_list, bdi_list) {
 417                         if (bdi->wb.task)
 418                                 continue;
 419                         if (list_empty(&bdi->work_list) &&
 420                             !bdi_has_dirty_io(bdi))
 421                                 continue;
 422
 423                         bdi_add_default_flusher_task(bdi);
 424                 }
 425
 426                 set_current_state(TASK_INTERRUPTIBLE);
 427
 428                 if (list_empty(&bdi_pending_list)) {
 429                         unsigned long wait;
 430
 431                         spin_unlock_bh(&bdi_lock);
 432                         wait = msecs_to_jiffies(dirty_writeback_interval * 10);
 433                         if (wait)
 434                                 schedule_timeout(wait);
 435                         else
 436                                 schedule();
 437                         try_to_freeze();
 438                         continue;
 439                 }
 440
 441                 __set_current_state(TASK_RUNNING);
 442
 443                 /*
 444                  * This is our real job - check for pending entries in
 445                  * bdi_pending_list, and create the tasks that got added
 446                  */
 447                 bdi = list_entry(bdi_pending_list.next, struct backing_dev_info,
 448                                  bdi_list);
 449                 list_del_init(&bdi->bdi_list);
 450                 spin_unlock_bh(&bdi_lock);
 451
 452                 wb = &bdi->wb;
 453                 wb->task = kthread_run(bdi_start_fn, wb, "flush-%s",
 454                                         dev_name(bdi->dev));
 455                 /*
 456                  * If task creation fails, then readd the bdi to
 457                  * the pending list and force writeout of the bdi
 458                  * from this forker thread. That will free some memory
 459                  * and we can try again.
 460                  */
 461                 if (IS_ERR(wb->task)) {
 462                         wb->task = NULL;
 463
 464                         /*
 465                          * Add this 'bdi' to the back, so we get
 466                          * a chance to flush other bdi's to free
 467                          * memory.
 468                          */
 469                         spin_lock_bh(&bdi_lock);
 470                         list_add_tail(&bdi->bdi_list, &bdi_pending_list);
 471                         spin_unlock_bh(&bdi_lock);
 472
 473                         bdi_flush_io(bdi);
 474                 }
 475         }
 476
 477         return 0;
 478 }
 479
 480 static void bdi_add_to_pending(struct rcu_head *head)
 481 {
 482         struct backing_dev_info *bdi;
 483
 484         bdi = container_of(head, struct backing_dev_info, rcu_head);
 485         INIT_LIST_HEAD(&bdi->bdi_list);
 486
 487         spin_lock(&bdi_lock);
 488         list_add_tail(&bdi->bdi_list, &bdi_pending_list);
 489         spin_unlock(&bdi_lock);
 490
 491         /*
 492          * We are now on the pending list, wake up bdi_forker_task()
 493          * to finish the job and add us back to the active bdi_list
 494          */
 495         wake_up_process(default_backing_dev_info.wb.task);
 496 }
 497
 498 /*
 499  * Add the default flusher task that gets created for any bdi
 500  * that has dirty data pending writeout
 501  */
 502 void static bdi_add_default_flusher_task(struct backing_dev_info *bdi)
 503 {
 504         if (!bdi_cap_writeback_dirty(bdi))
 505                 return;
 506
 507         if (WARN_ON(!test_bit(BDI_registered, &bdi->state))) {
 508                 printk(KERN_ERR "bdi %p/%s is not registered!\n",
 509                                                         bdi, bdi->name);
 510                 return;
 511         }
 512
 513         /*
 514          * Check with the helper whether to proceed adding a task. Will only
 515          * abort if we two or more simultanous calls to
 516          * bdi_add_default_flusher_task() occured, further additions will block
 517          * waiting for previous additions to finish.
 518          */
 519         if (!test_and_set_bit(BDI_pending, &bdi->state)) {
 520                 list_del_rcu(&bdi->bdi_list);
 521
 522                 /*
 523                  * We must wait for the current RCU period to end before
 524                  * moving to the pending list. So schedule that operation
 525                  * from an RCU callback.
 526                  */
 527                 call_rcu(&bdi->rcu_head, bdi_add_to_pending);
 528         }
 529 }
 530
 531 /*
 532  * Remove bdi from bdi_list, and ensure that it is no longer visible
 533  */
 534 static void bdi_remove_from_list(struct backing_dev_info *bdi)
 535 {
 536         spin_lock_bh(&bdi_lock);
 537         list_del_rcu(&bdi->bdi_list);
 538         spin_unlock_bh(&bdi_lock);
 539
 540         synchronize_rcu();
 541 }
 542
 543 int bdi_register(struct backing_dev_info *bdi, struct device *parent,
 544                 const char *fmt, ...)
 545 {
 546         va_list args;
 547         int ret = 0;
 548         struct device *dev;
 549
 550         if (bdi->dev)   /* The driver needs to use separate queues per device */
 551                 goto exit;
 552
 553         va_start(args, fmt);
 554         dev = device_create_vargs(bdi_class, parent, MKDEV(0, 0), bdi, fmt, args);
 555         va_end(args);
 556         if (IS_ERR(dev)) {
 557                 ret = PTR_ERR(dev);
 558                 goto exit;
 559         }
 560
 561         spin_lock_bh(&bdi_lock);
 562         list_add_tail_rcu(&bdi->bdi_list, &bdi_list);
 563         spin_unlock_bh(&bdi_lock);
 564
 565         bdi->dev = dev;
 566
 567         /*
 568          * Just start the forker thread for our default backing_dev_info,
 569          * and add other bdi's to the list. They will get a thread created
 570          * on-demand when they need it.
 571          */
 572         if (bdi_cap_flush_forker(bdi)) {
 573                 struct bdi_writeback *wb = &bdi->wb;
 574
 575                 wb->task = kthread_run(bdi_forker_task, wb, "bdi-%s",
 576                                                 dev_name(dev));
 577                 if (IS_ERR(wb->task)) {
 578                         wb->task = NULL;
 579                         ret = -ENOMEM;
 580
 581                         bdi_remove_from_list(bdi);
 582                         goto exit;
 583                 }
 584         }
 585
 586         bdi_debug_register(bdi, dev_name(dev));
 587         set_bit(BDI_registered, &bdi->state);
 588 exit:
 589         return ret;
 590 }
 591 EXPORT_SYMBOL(bdi_register);
 592
 593 int bdi_register_dev(struct backing_dev_info *bdi, dev_t dev)
 594 {
 595         return bdi_register(bdi, NULL, "%u:%u", MAJOR(dev), MINOR(dev));
 596 }
 597 EXPORT_SYMBOL(bdi_register_dev);
 598
 599 /*
 600  * Remove bdi from the global list and shutdown any threads we have running
 601  */
 602 static void bdi_wb_shutdown(struct backing_dev_info *bdi)
 603 {
 604         struct bdi_writeback *wb;
 605
 606         if (!bdi_cap_writeback_dirty(bdi))
 607                 return;
 608
 609         /*
 610          * If setup is pending, wait for that to complete first
 611          */
 612         wait_on_bit(&bdi->state, BDI_pending, bdi_sched_wait,
 613                         TASK_UNINTERRUPTIBLE);
 614
 615         /*
 616          * Make sure nobody finds us on the bdi_list anymore
 617          */
 618         bdi_remove_from_list(bdi);
 619
 620         /*
 621          * Finally, kill the kernel threads. We don't need to be RCU
 622          * safe anymore, since the bdi is gone from visibility. Force
 623          * unfreeze of the thread before calling kthread_stop(), otherwise
 624          * it would never exet if it is currently stuck in the refrigerator.
 625          */
 626         list_for_each_entry(wb, &bdi->wb_list, list) {
 627                 thaw_process(wb->task);
 628                 kthread_stop(wb->task);
 629         }
 630 }
 631
 632 /*
 633  * This bdi is going away now, make sure that no super_blocks point to it
 634  */
 635 static void bdi_prune_sb(struct backing_dev_info *bdi)
 636 {
 637         struct super_block *sb;
 638
 639         spin_lock(&sb_lock);
 640         list_for_each_entry(sb, &super_blocks, s_list) {
 641                 if (sb->s_bdi == bdi)
 642                         sb->s_bdi = NULL;
 643         }
 644         spin_unlock(&sb_lock);
 645 }
 646
 647 void bdi_unregister(struct backing_dev_info *bdi)
 648 {
 649         if (bdi->dev) {
 650                 bdi_prune_sb(bdi);
 651
 652                 if (!bdi_cap_flush_forker(bdi))
 653                         bdi_wb_shutdown(bdi);
 654                 bdi_debug_unregister(bdi);
 655                 device_unregister(bdi->dev);
 656                 bdi->dev = NULL;
 657         }
 658 }
 659 EXPORT_SYMBOL(bdi_unregister);
 660
 661 int bdi_init(struct backing_dev_info *bdi)
 662 {
 663         int i, err;
 664
 665         bdi->dev = NULL;
 666
 667         bdi->min_ratio = 0;
 668         bdi->max_ratio = 100;
 669         bdi->max_prop_frac = PROP_FRAC_BASE;
 670         spin_lock_init(&bdi->wb_lock);
 671         INIT_RCU_HEAD(&bdi->rcu_head);
 672         INIT_LIST_HEAD(&bdi->bdi_list);
 673         INIT_LIST_HEAD(&bdi->wb_list);
 674         INIT_LIST_HEAD(&bdi->work_list);
 675
 676         bdi_wb_init(&bdi->wb, bdi);
 677
 678         /*
 679          * Just one thread support for now, hard code mask and count
 680          */
 681         bdi->wb_mask = 1;
 682         bdi->wb_cnt = 1;
 683
 684         for (i = 0; i < NR_BDI_STAT_ITEMS; i++) {
 685                 err = percpu_counter_init(&bdi->bdi_stat[i], 0);
 686                 if (err)
 687                         goto err;
 688         }
 689
 690         bdi->dirty_exceeded = 0;
 691         err = prop_local_init_percpu(&bdi->completions);
 692
 693         if (err) {
 694 err:
 695                 while (i--)
 696                         percpu_counter_destroy(&bdi->bdi_stat[i]);
 697         }
 698
 699         return err;
 700 }
 701 EXPORT_SYMBOL(bdi_init);
 702
 703 void bdi_destroy(struct backing_dev_info *bdi)
 704 {
 705         int i;
 706
 707         /*
 708          * Splice our entries to the default_backing_dev_info, if this
 709          * bdi disappears
 710          */
 711         if (bdi_has_dirty_io(bdi)) {
 712                 struct bdi_writeback *dst = &default_backing_dev_info.wb;
 713
 714                 spin_lock(&inode_lock);
 715                 list_splice(&bdi->wb.b_dirty, &dst->b_dirty);
 716                 list_splice(&bdi->wb.b_io, &dst->b_io);
 717                 list_splice(&bdi->wb.b_more_io, &dst->b_more_io);
 718                 spin_unlock(&inode_lock);
 719         }
 720
 721         bdi_unregister(bdi);
 722
 723         for (i = 0; i < NR_BDI_STAT_ITEMS; i++)
 724                 percpu_counter_destroy(&bdi->bdi_stat[i]);
 725
 726         prop_local_destroy_percpu(&bdi->completions);
 727 }
 728 EXPORT_SYMBOL(bdi_destroy);
 729
 730 /*
 731  * For use from filesystems to quickly init and register a bdi associated
 732  * with dirty writeback
 733  */
 734 int bdi_setup_and_register(struct backing_dev_info *bdi, char *name,
 735                            unsigned int cap)
 736 {
 737         char tmp[32];
 738         int err;
 739
 740         bdi->name = name;
 741         bdi->capabilities = cap;
 742         err = bdi_init(bdi);
 743         if (err)
 744                 return err;
 745
 746         sprintf(tmp, "%.28s%s", name, "-%d");
 747         err = bdi_register(bdi, NULL, tmp, atomic_long_inc_return(&bdi_seq));
 748         if (err) {
 749                 bdi_destroy(bdi);
 750                 return err;
 751         }
 752
 753         return 0;
 754 }
 755 EXPORT_SYMBOL(bdi_setup_and_register);
 756
 757 static wait_queue_head_t congestion_wqh[2] = {
 758                 __WAIT_QUEUE_HEAD_INITIALIZER(congestion_wqh[0]),
 759                 __WAIT_QUEUE_HEAD_INITIALIZER(congestion_wqh[1])
 760         };
 761
 762 void clear_bdi_congested(struct backing_dev_info *bdi, int sync)
 763 {
 764         enum bdi_state bit;
 765         wait_queue_head_t *wqh = &congestion_wqh[sync];
 766
 767         bit = sync ? BDI_sync_congested : BDI_async_congested;
 768         clear_bit(bit, &bdi->state);
 769         smp_mb__after_clear_bit();
 770         if (waitqueue_active(wqh))
 771                 wake_up(wqh);
 772 }
 773 EXPORT_SYMBOL(clear_bdi_congested);
 774
 775 void set_bdi_congested(struct backing_dev_info *bdi, int sync)
 776 {
 777         enum bdi_state bit;
 778
 779         bit = sync ? BDI_sync_congested : BDI_async_congested;
 780         set_bit(bit, &bdi->state);
 781 }
 782 EXPORT_SYMBOL(set_bdi_congested);
 783
 784 /**
 785  * congestion_wait - wait for a backing_dev to become uncongested
 786  * @sync: SYNC or ASYNC IO
 787  * @timeout: timeout in jiffies
 788  *
 789  * Waits for up to @timeout jiffies for a backing_dev (any backing_dev) to exit
 790  * write congestion.  If no backing_devs are congested then just wait for the
 791  * next write to be completed.
 792  */
 793 long congestion_wait(int sync, long timeout)
 794 {
 795         long ret;
 796         DEFINE_WAIT(wait);
 797         wait_queue_head_t *wqh = &congestion_wqh[sync];
 798
 799         prepare_to_wait(wqh, &wait, TASK_UNINTERRUPTIBLE);
 800         ret = io_schedule_timeout(timeout);
 801         finish_wait(wqh, &wait);
 802         return ret;
 803 }
 804 EXPORT_SYMBOL(congestion_wait);
 805