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_list:          %8u\n",
 108                    (unsigned long) K(bdi_stat(bdi, BDI_WRITEBACK)),
 109                    (unsigned long) K(bdi_stat(bdi, BDI_RECLAIMABLE)),
 110                    K(bdi_thresh), K(dirty_thresh),
 111                    K(background_thresh), nr_wb, nr_dirty, nr_io, nr_more_io,
 112                    !list_empty(&bdi->bdi_list), bdi->state,
 113                    !list_empty(&bdi->wb_list));
 114 #undef K
 115
 116         return 0;
 117 }
 118
 119 static int bdi_debug_stats_open(struct inode *inode, struct file *file)
 120 {
 121         return single_open(file, bdi_debug_stats_show, inode->i_private);
 122 }
 123
 124 static const struct file_operations bdi_debug_stats_fops = {
 125         .open           = bdi_debug_stats_open,
 126         .read           = seq_read,
 127         .llseek         = seq_lseek,
 128         .release        = single_release,
 129 };
 130
 131 static void bdi_debug_register(struct backing_dev_info *bdi, const char *name)
 132 {
 133         bdi->debug_dir = debugfs_create_dir(name, bdi_debug_root);
 134         bdi->debug_stats = debugfs_create_file("stats", 0444, bdi->debug_dir,
 135                                                bdi, &bdi_debug_stats_fops);
 136 }
 137
 138 static void bdi_debug_unregister(struct backing_dev_info *bdi)
 139 {
 140         debugfs_remove(bdi->debug_stats);
 141         debugfs_remove(bdi->debug_dir);
 142 }
 143 #else
 144 static inline void bdi_debug_init(void)
 145 {
 146 }
 147 static inline void bdi_debug_register(struct backing_dev_info *bdi,
 148                                       const char *name)
 149 {
 150 }
 151 static inline void bdi_debug_unregister(struct backing_dev_info *bdi)
 152 {
 153 }
 154 #endif
 155
 156 static ssize_t read_ahead_kb_store(struct device *dev,
 157                                   struct device_attribute *attr,
 158                                   const char *buf, size_t count)
 159 {
 160         struct backing_dev_info *bdi = dev_get_drvdata(dev);
 161         char *end;
 162         unsigned long read_ahead_kb;
 163         ssize_t ret = -EINVAL;
 164
 165         read_ahead_kb = simple_strtoul(buf, &end, 10);
 166         if (*buf && (end[0] == '\0' || (end[0] == '\n' && end[1] == '\0'))) {
 167                 bdi->ra_pages = read_ahead_kb >> (PAGE_SHIFT - 10);
 168                 ret = count;
 169         }
 170         return ret;
 171 }
 172
 173 #define K(pages) ((pages) << (PAGE_SHIFT - 10))
 174
 175 #define BDI_SHOW(name, expr)                                            \
 176 static ssize_t name##_show(struct device *dev,                          \
 177                            struct device_attribute *attr, char *page)   \
 178 {                                                                       \
 179         struct backing_dev_info *bdi = dev_get_drvdata(dev);            \
 180                                                                         \
 181         return snprintf(page, PAGE_SIZE-1, "%lld\n", (long long)expr);  \
 182 }
 183
 184 BDI_SHOW(read_ahead_kb, K(bdi->ra_pages))
 185
 186 static ssize_t min_ratio_store(struct device *dev,
 187                 struct device_attribute *attr, const char *buf, size_t count)
 188 {
 189         struct backing_dev_info *bdi = dev_get_drvdata(dev);
 190         char *end;
 191         unsigned int ratio;
 192         ssize_t ret = -EINVAL;
 193
 194         ratio = simple_strtoul(buf, &end, 10);
 195         if (*buf && (end[0] == '\0' || (end[0] == '\n' && end[1] == '\0'))) {
 196                 ret = bdi_set_min_ratio(bdi, ratio);
 197                 if (!ret)
 198                         ret = count;
 199         }
 200         return ret;
 201 }
 202 BDI_SHOW(min_ratio, bdi->min_ratio)
 203
 204 static ssize_t max_ratio_store(struct device *dev,
 205                 struct device_attribute *attr, const char *buf, size_t count)
 206 {
 207         struct backing_dev_info *bdi = dev_get_drvdata(dev);
 208         char *end;
 209         unsigned int ratio;
 210         ssize_t ret = -EINVAL;
 211
 212         ratio = simple_strtoul(buf, &end, 10);
 213         if (*buf && (end[0] == '\0' || (end[0] == '\n' && end[1] == '\0'))) {
 214                 ret = bdi_set_max_ratio(bdi, ratio);
 215                 if (!ret)
 216                         ret = count;
 217         }
 218         return ret;
 219 }
 220 BDI_SHOW(max_ratio, bdi->max_ratio)
 221
 222 #define __ATTR_RW(attr) __ATTR(attr, 0644, attr##_show, attr##_store)
 223
 224 static struct device_attribute bdi_dev_attrs[] = {
 225         __ATTR_RW(read_ahead_kb),
 226         __ATTR_RW(min_ratio),
 227         __ATTR_RW(max_ratio),
 228         __ATTR_NULL,
 229 };
 230
 231 static __init int bdi_class_init(void)
 232 {
 233         bdi_class = class_create(THIS_MODULE, "bdi");
 234         if (IS_ERR(bdi_class))
 235                 return PTR_ERR(bdi_class);
 236
 237         bdi_class->dev_attrs = bdi_dev_attrs;
 238         bdi_debug_init();
 239         return 0;
 240 }
 241 postcore_initcall(bdi_class_init);
 242
 243 static int __init default_bdi_init(void)
 244 {
 245         int err;
 246
 247         sync_supers_tsk = kthread_run(bdi_sync_supers, NULL, "sync_supers");
 248         BUG_ON(IS_ERR(sync_supers_tsk));
 249
 250         init_timer(&sync_supers_timer);
 251         setup_timer(&sync_supers_timer, sync_supers_timer_fn, 0);
 252         bdi_arm_supers_timer();
 253
 254         err = bdi_init(&default_backing_dev_info);
 255         if (!err)
 256                 bdi_register(&default_backing_dev_info, NULL, "default");
 257
 258         return err;
 259 }
 260 subsys_initcall(default_bdi_init);
 261
 262 static void bdi_wb_init(struct bdi_writeback *wb, struct backing_dev_info *bdi)
 263 {
 264         memset(wb, 0, sizeof(*wb));
 265
 266         wb->bdi = bdi;
 267         wb->last_old_flush = jiffies;
 268         INIT_LIST_HEAD(&wb->b_dirty);
 269         INIT_LIST_HEAD(&wb->b_io);
 270         INIT_LIST_HEAD(&wb->b_more_io);
 271 }
 272
 273 static void bdi_task_init(struct backing_dev_info *bdi,
 274                           struct bdi_writeback *wb)
 275 {
 276         struct task_struct *tsk = current;
 277
 278         spin_lock(&bdi->wb_lock);
 279         list_add_tail_rcu(&wb->list, &bdi->wb_list);
 280         spin_unlock(&bdi->wb_lock);
 281
 282         tsk->flags |= PF_FLUSHER | PF_SWAPWRITE;
 283         set_freezable();
 284
 285         /*
 286          * Our parent may run at a different priority, just set us to normal
 287          */
 288         set_user_nice(tsk, 0);
 289 }
 290
 291 static int bdi_start_fn(void *ptr)
 292 {
 293         struct bdi_writeback *wb = ptr;
 294         struct backing_dev_info *bdi = wb->bdi;
 295         int ret;
 296
 297         /*
 298          * Add us to the active bdi_list
 299          */
 300         spin_lock_bh(&bdi_lock);
 301         list_add_rcu(&bdi->bdi_list, &bdi_list);
 302         spin_unlock_bh(&bdi_lock);
 303
 304         bdi_task_init(bdi, wb);
 305
 306         /*
 307          * Clear pending bit and wakeup anybody waiting to tear us down
 308          */
 309         clear_bit(BDI_pending, &bdi->state);
 310         smp_mb__after_clear_bit();
 311         wake_up_bit(&bdi->state, BDI_pending);
 312
 313         ret = bdi_writeback_task(wb);
 314
 315         /*
 316          * Remove us from the list
 317          */
 318         spin_lock(&bdi->wb_lock);
 319         list_del_rcu(&wb->list);
 320         spin_unlock(&bdi->wb_lock);
 321
 322         /*
 323          * Flush any work that raced with us exiting. No new work
 324          * will be added, since this bdi isn't discoverable anymore.
 325          */
 326         if (!list_empty(&bdi->work_list))
 327                 wb_do_writeback(wb, 1);
 328
 329         wb->task = NULL;
 330         return ret;
 331 }
 332
 333 int bdi_has_dirty_io(struct backing_dev_info *bdi)
 334 {
 335         return wb_has_dirty_io(&bdi->wb);
 336 }
 337
 338 static void bdi_flush_io(struct backing_dev_info *bdi)
 339 {
 340         struct writeback_control wbc = {
 341                 .sync_mode              = WB_SYNC_NONE,
 342                 .older_than_this        = NULL,
 343                 .range_cyclic           = 1,
 344                 .nr_to_write            = 1024,
 345         };
 346
 347         writeback_inodes_wb(&bdi->wb, &wbc);
 348 }
 349
 350 /*
 351  * kupdated() used to do this. We cannot do it from the bdi_forker_task()
 352  * or we risk deadlocking on ->s_umount. The longer term solution would be
 353  * to implement sync_supers_bdi() or similar and simply do it from the
 354  * bdi writeback tasks individually.
 355  */
 356 static int bdi_sync_supers(void *unused)
 357 {
 358         set_user_nice(current, 0);
 359
 360         while (!kthread_should_stop()) {
 361                 set_current_state(TASK_INTERRUPTIBLE);
 362                 schedule();
 363
 364                 /*
 365                  * Do this periodically, like kupdated() did before.
 366                  */
 367                 sync_supers();
 368         }
 369
 370         return 0;
 371 }
 372
 373 void bdi_arm_supers_timer(void)
 374 {
 375         unsigned long next;
 376
 377         if (!dirty_writeback_interval)
 378                 return;
 379
 380         next = msecs_to_jiffies(dirty_writeback_interval * 10) + jiffies;
 381         mod_timer(&sync_supers_timer, round_jiffies_up(next));
 382 }
 383
 384 static void sync_supers_timer_fn(unsigned long unused)
 385 {
 386         wake_up_process(sync_supers_tsk);
 387         bdi_arm_supers_timer();
 388 }
 389
 390 static int bdi_forker_task(void *ptr)
 391 {
 392         struct bdi_writeback *me = ptr;
 393
 394         bdi_task_init(me->bdi, me);
 395
 396         for (;;) {
 397                 struct backing_dev_info *bdi, *tmp;
 398                 struct bdi_writeback *wb;
 399
 400                 /*
 401                  * Temporary measure, we want to make sure we don't see
 402                  * dirty data on the default backing_dev_info
 403                  */
 404                 if (wb_has_dirty_io(me) || !list_empty(&me->bdi->work_list))
 405                         wb_do_writeback(me, 0);
 406
 407                 spin_lock_bh(&bdi_lock);
 408
 409                 /*
 410                  * Check if any existing bdi's have dirty data without
 411                  * a thread registered. If so, set that up.
 412                  */
 413                 list_for_each_entry_safe(bdi, tmp, &bdi_list, bdi_list) {
 414                         if (bdi->wb.task)
 415                                 continue;
 416                         if (list_empty(&bdi->work_list) &&
 417                             !bdi_has_dirty_io(bdi))
 418                                 continue;
 419
 420                         bdi_add_default_flusher_task(bdi);
 421                 }
 422
 423                 set_current_state(TASK_INTERRUPTIBLE);
 424
 425                 if (list_empty(&bdi_pending_list)) {
 426                         unsigned long wait;
 427
 428                         spin_unlock_bh(&bdi_lock);
 429                         wait = msecs_to_jiffies(dirty_writeback_interval * 10);
 430                         if (wait)
 431                                 schedule_timeout(wait);
 432                         else
 433                                 schedule();
 434                         try_to_freeze();
 435                         continue;
 436                 }
 437
 438                 __set_current_state(TASK_RUNNING);
 439
 440                 /*
 441                  * This is our real job - check for pending entries in
 442                  * bdi_pending_list, and create the tasks that got added
 443                  */
 444                 bdi = list_entry(bdi_pending_list.next, struct backing_dev_info,
 445                                  bdi_list);
 446                 list_del_init(&bdi->bdi_list);
 447                 spin_unlock_bh(&bdi_lock);
 448
 449                 wb = &bdi->wb;
 450                 wb->task = kthread_run(bdi_start_fn, wb, "flush-%s",
 451                                         dev_name(bdi->dev));
 452                 /*
 453                  * If task creation fails, then readd the bdi to
 454                  * the pending list and force writeout of the bdi
 455                  * from this forker thread. That will free some memory
 456                  * and we can try again.
 457                  */
 458                 if (IS_ERR(wb->task)) {
 459                         wb->task = NULL;
 460
 461                         /*
 462                          * Add this 'bdi' to the back, so we get
 463                          * a chance to flush other bdi's to free
 464                          * memory.
 465                          */
 466                         spin_lock_bh(&bdi_lock);
 467                         list_add_tail(&bdi->bdi_list, &bdi_pending_list);
 468                         spin_unlock_bh(&bdi_lock);
 469
 470                         bdi_flush_io(bdi);
 471                 }
 472         }
 473
 474         return 0;
 475 }
 476
 477 static void bdi_add_to_pending(struct rcu_head *head)
 478 {
 479         struct backing_dev_info *bdi;
 480
 481         bdi = container_of(head, struct backing_dev_info, rcu_head);
 482         INIT_LIST_HEAD(&bdi->bdi_list);
 483
 484         spin_lock(&bdi_lock);
 485         list_add_tail(&bdi->bdi_list, &bdi_pending_list);
 486         spin_unlock(&bdi_lock);
 487
 488         /*
 489          * We are now on the pending list, wake up bdi_forker_task()
 490          * to finish the job and add us back to the active bdi_list
 491          */
 492         wake_up_process(default_backing_dev_info.wb.task);
 493 }
 494
 495 /*
 496  * Add the default flusher task that gets created for any bdi
 497  * that has dirty data pending writeout
 498  */
 499 void static bdi_add_default_flusher_task(struct backing_dev_info *bdi)
 500 {
 501         if (!bdi_cap_writeback_dirty(bdi))
 502                 return;
 503
 504         if (WARN_ON(!test_bit(BDI_registered, &bdi->state))) {
 505                 printk(KERN_ERR "bdi %p/%s is not registered!\n",
 506                                                         bdi, bdi->name);
 507                 return;
 508         }
 509
 510         /*
 511          * Check with the helper whether to proceed adding a task. Will only
 512          * abort if we two or more simultanous calls to
 513          * bdi_add_default_flusher_task() occured, further additions will block
 514          * waiting for previous additions to finish.
 515          */
 516         if (!test_and_set_bit(BDI_pending, &bdi->state)) {
 517                 list_del_rcu(&bdi->bdi_list);
 518
 519                 /*
 520                  * We must wait for the current RCU period to end before
 521                  * moving to the pending list. So schedule that operation
 522                  * from an RCU callback.
 523                  */
 524                 call_rcu(&bdi->rcu_head, bdi_add_to_pending);
 525         }
 526 }
 527
 528 /*
 529  * Remove bdi from bdi_list, and ensure that it is no longer visible
 530  */
 531 static void bdi_remove_from_list(struct backing_dev_info *bdi)
 532 {
 533         spin_lock_bh(&bdi_lock);
 534         list_del_rcu(&bdi->bdi_list);
 535         spin_unlock_bh(&bdi_lock);
 536
 537         synchronize_rcu();
 538 }
 539
 540 int bdi_register(struct backing_dev_info *bdi, struct device *parent,
 541                 const char *fmt, ...)
 542 {
 543         va_list args;
 544         int ret = 0;
 545         struct device *dev;
 546
 547         if (bdi->dev)   /* The driver needs to use separate queues per device */
 548                 goto exit;
 549
 550         va_start(args, fmt);
 551         dev = device_create_vargs(bdi_class, parent, MKDEV(0, 0), bdi, fmt, args);
 552         va_end(args);
 553         if (IS_ERR(dev)) {
 554                 ret = PTR_ERR(dev);
 555                 goto exit;
 556         }
 557
 558         spin_lock_bh(&bdi_lock);
 559         list_add_tail_rcu(&bdi->bdi_list, &bdi_list);
 560         spin_unlock_bh(&bdi_lock);
 561
 562         bdi->dev = dev;
 563
 564         /*
 565          * Just start the forker thread for our default backing_dev_info,
 566          * and add other bdi's to the list. They will get a thread created
 567          * on-demand when they need it.
 568          */
 569         if (bdi_cap_flush_forker(bdi)) {
 570                 struct bdi_writeback *wb = &bdi->wb;
 571
 572                 wb->task = kthread_run(bdi_forker_task, wb, "bdi-%s",
 573                                                 dev_name(dev));
 574                 if (IS_ERR(wb->task)) {
 575                         wb->task = NULL;
 576                         ret = -ENOMEM;
 577
 578                         bdi_remove_from_list(bdi);
 579                         goto exit;
 580                 }
 581         }
 582
 583         bdi_debug_register(bdi, dev_name(dev));
 584         set_bit(BDI_registered, &bdi->state);
 585 exit:
 586         return ret;
 587 }
 588 EXPORT_SYMBOL(bdi_register);
 589
 590 int bdi_register_dev(struct backing_dev_info *bdi, dev_t dev)
 591 {
 592         return bdi_register(bdi, NULL, "%u:%u", MAJOR(dev), MINOR(dev));
 593 }
 594 EXPORT_SYMBOL(bdi_register_dev);
 595
 596 /*
 597  * Remove bdi from the global list and shutdown any threads we have running
 598  */
 599 static void bdi_wb_shutdown(struct backing_dev_info *bdi)
 600 {
 601         struct bdi_writeback *wb;
 602
 603         if (!bdi_cap_writeback_dirty(bdi))
 604                 return;
 605
 606         /*
 607          * If setup is pending, wait for that to complete first
 608          */
 609         wait_on_bit(&bdi->state, BDI_pending, bdi_sched_wait,
 610                         TASK_UNINTERRUPTIBLE);
 611
 612         /*
 613          * Make sure nobody finds us on the bdi_list anymore
 614          */
 615         bdi_remove_from_list(bdi);
 616
 617         /*
 618          * Finally, kill the kernel threads. We don't need to be RCU
 619          * safe anymore, since the bdi is gone from visibility. Force
 620          * unfreeze of the thread before calling kthread_stop(), otherwise
 621          * it would never exet if it is currently stuck in the refrigerator.
 622          */
 623         list_for_each_entry(wb, &bdi->wb_list, list) {
 624                 thaw_process(wb->task);
 625                 kthread_stop(wb->task);
 626         }
 627 }
 628
 629 /*
 630  * This bdi is going away now, make sure that no super_blocks point to it
 631  */
 632 static void bdi_prune_sb(struct backing_dev_info *bdi)
 633 {
 634         struct super_block *sb;
 635
 636         spin_lock(&sb_lock);
 637         list_for_each_entry(sb, &super_blocks, s_list) {
 638                 if (sb->s_bdi == bdi)
 639                         sb->s_bdi = NULL;
 640         }
 641         spin_unlock(&sb_lock);
 642 }
 643
 644 void bdi_unregister(struct backing_dev_info *bdi)
 645 {
 646         if (bdi->dev) {
 647                 bdi_prune_sb(bdi);
 648
 649                 if (!bdi_cap_flush_forker(bdi))
 650                         bdi_wb_shutdown(bdi);
 651                 bdi_debug_unregister(bdi);
 652                 device_unregister(bdi->dev);
 653                 bdi->dev = NULL;
 654         }
 655 }
 656 EXPORT_SYMBOL(bdi_unregister);
 657
 658 int bdi_init(struct backing_dev_info *bdi)
 659 {
 660         int i, err;
 661
 662         bdi->dev = NULL;
 663
 664         bdi->min_ratio = 0;
 665         bdi->max_ratio = 100;
 666         bdi->max_prop_frac = PROP_FRAC_BASE;
 667         spin_lock_init(&bdi->wb_lock);
 668         INIT_RCU_HEAD(&bdi->rcu_head);
 669         INIT_LIST_HEAD(&bdi->bdi_list);
 670         INIT_LIST_HEAD(&bdi->wb_list);
 671         INIT_LIST_HEAD(&bdi->work_list);
 672
 673         bdi_wb_init(&bdi->wb, bdi);
 674
 675         for (i = 0; i < NR_BDI_STAT_ITEMS; i++) {
 676                 err = percpu_counter_init(&bdi->bdi_stat[i], 0);
 677                 if (err)
 678                         goto err;
 679         }
 680
 681         bdi->dirty_exceeded = 0;
 682         err = prop_local_init_percpu(&bdi->completions);
 683
 684         if (err) {
 685 err:
 686                 while (i--)
 687                         percpu_counter_destroy(&bdi->bdi_stat[i]);
 688         }
 689
 690         return err;
 691 }
 692 EXPORT_SYMBOL(bdi_init);
 693
 694 void bdi_destroy(struct backing_dev_info *bdi)
 695 {
 696         int i;
 697
 698         /*
 699          * Splice our entries to the default_backing_dev_info, if this
 700          * bdi disappears
 701          */
 702         if (bdi_has_dirty_io(bdi)) {
 703                 struct bdi_writeback *dst = &default_backing_dev_info.wb;
 704
 705                 spin_lock(&inode_lock);
 706                 list_splice(&bdi->wb.b_dirty, &dst->b_dirty);
 707                 list_splice(&bdi->wb.b_io, &dst->b_io);
 708                 list_splice(&bdi->wb.b_more_io, &dst->b_more_io);
 709                 spin_unlock(&inode_lock);
 710         }
 711
 712         bdi_unregister(bdi);
 713
 714         for (i = 0; i < NR_BDI_STAT_ITEMS; i++)
 715                 percpu_counter_destroy(&bdi->bdi_stat[i]);
 716
 717         prop_local_destroy_percpu(&bdi->completions);
 718 }
 719 EXPORT_SYMBOL(bdi_destroy);
 720
 721 /*
 722  * For use from filesystems to quickly init and register a bdi associated
 723  * with dirty writeback
 724  */
 725 int bdi_setup_and_register(struct backing_dev_info *bdi, char *name,
 726                            unsigned int cap)
 727 {
 728         char tmp[32];
 729         int err;
 730
 731         bdi->name = name;
 732         bdi->capabilities = cap;
 733         err = bdi_init(bdi);
 734         if (err)
 735                 return err;
 736
 737         sprintf(tmp, "%.28s%s", name, "-%d");
 738         err = bdi_register(bdi, NULL, tmp, atomic_long_inc_return(&bdi_seq));
 739         if (err) {
 740                 bdi_destroy(bdi);
 741                 return err;
 742         }
 743
 744         return 0;
 745 }
 746 EXPORT_SYMBOL(bdi_setup_and_register);
 747
 748 static wait_queue_head_t congestion_wqh[2] = {
 749                 __WAIT_QUEUE_HEAD_INITIALIZER(congestion_wqh[0]),
 750                 __WAIT_QUEUE_HEAD_INITIALIZER(congestion_wqh[1])
 751         };
 752
 753 void clear_bdi_congested(struct backing_dev_info *bdi, int sync)
 754 {
 755         enum bdi_state bit;
 756         wait_queue_head_t *wqh = &congestion_wqh[sync];
 757
 758         bit = sync ? BDI_sync_congested : BDI_async_congested;
 759         clear_bit(bit, &bdi->state);
 760         smp_mb__after_clear_bit();
 761         if (waitqueue_active(wqh))
 762                 wake_up(wqh);
 763 }
 764 EXPORT_SYMBOL(clear_bdi_congested);
 765
 766 void set_bdi_congested(struct backing_dev_info *bdi, int sync)
 767 {
 768         enum bdi_state bit;
 769
 770         bit = sync ? BDI_sync_congested : BDI_async_congested;
 771         set_bit(bit, &bdi->state);
 772 }
 773 EXPORT_SYMBOL(set_bdi_congested);
 774
 775 /**
 776  * congestion_wait - wait for a backing_dev to become uncongested
 777  * @sync: SYNC or ASYNC IO
 778  * @timeout: timeout in jiffies
 779  *
 780  * Waits for up to @timeout jiffies for a backing_dev (any backing_dev) to exit
 781  * write congestion.  If no backing_devs are congested then just wait for the
 782  * next write to be completed.
 783  */
 784 long congestion_wait(int sync, long timeout)
 785 {
 786         long ret;
 787         DEFINE_WAIT(wait);
 788         wait_queue_head_t *wqh = &congestion_wqh[sync];
 789
 790         prepare_to_wait(wqh, &wait, TASK_UNINTERRUPTIBLE);
 791         ret = io_schedule_timeout(timeout);
 792         finish_wait(wqh, &wait);
 793         return ret;
 794 }
 795 EXPORT_SYMBOL(congestion_wait);
 796