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 and bdi_pending_list, as well as
  35  * reader side protection for bdi_pending_list. bdi_list has RCU reader side
  36  * locking.
  37  */
  38 DEFINE_SPINLOCK(bdi_lock);
  39 LIST_HEAD(bdi_list);
  40 LIST_HEAD(bdi_pending_list);
  41
  42 static struct task_struct *sync_supers_tsk;
  43 static struct timer_list sync_supers_timer;
  44
  45 static int bdi_sync_supers(void *);
  46 static void sync_supers_timer_fn(unsigned long);
  47
  48 #ifdef CONFIG_DEBUG_FS
  49 #include <linux/debugfs.h>
  50 #include <linux/seq_file.h>
  51
  52 static struct dentry *bdi_debug_root;
  53
  54 static void bdi_debug_init(void)
  55 {
  56         bdi_debug_root = debugfs_create_dir("bdi", NULL);
  57 }
  58
  59 static int bdi_debug_stats_show(struct seq_file *m, void *v)
  60 {
  61         struct backing_dev_info *bdi = m->private;
  62         struct bdi_writeback *wb = &bdi->wb;
  63         unsigned long background_thresh;
  64         unsigned long dirty_thresh;
  65         unsigned long bdi_thresh;
  66         unsigned long nr_dirty, nr_io, nr_more_io, nr_wb;
  67         struct inode *inode;
  68
  69         nr_wb = nr_dirty = nr_io = nr_more_io = 0;
  70         spin_lock(&inode_wb_list_lock);
  71         list_for_each_entry(inode, &wb->b_dirty, i_wb_list)
  72                 nr_dirty++;
  73         list_for_each_entry(inode, &wb->b_io, i_wb_list)
  74                 nr_io++;
  75         list_for_each_entry(inode, &wb->b_more_io, i_wb_list)
  76                 nr_more_io++;
  77         spin_unlock(&inode_wb_list_lock);
  78
  79         global_dirty_limits(&background_thresh, &dirty_thresh);
  80         bdi_thresh = bdi_dirty_limit(bdi, dirty_thresh);
  81
  82 #define K(x) ((x) << (PAGE_SHIFT - 10))
  83         seq_printf(m,
  84                    "BdiWriteback:     %8lu kB\n"
  85                    "BdiReclaimable:   %8lu kB\n"
  86                    "BdiDirtyThresh:   %8lu kB\n"
  87                    "DirtyThresh:      %8lu kB\n"
  88                    "BackgroundThresh: %8lu kB\n"
  89                    "b_dirty:          %8lu\n"
  90                    "b_io:             %8lu\n"
  91                    "b_more_io:        %8lu\n"
  92                    "bdi_list:         %8u\n"
  93                    "state:            %8lx\n",
  94                    (unsigned long) K(bdi_stat(bdi, BDI_WRITEBACK)),
  95                    (unsigned long) K(bdi_stat(bdi, BDI_RECLAIMABLE)),
  96                    K(bdi_thresh), K(dirty_thresh),
  97                    K(background_thresh), nr_dirty, nr_io, nr_more_io,
  98                    !list_empty(&bdi->bdi_list), bdi->state);
  99 #undef K
 100
 101         return 0;
 102 }
 103
 104 static int bdi_debug_stats_open(struct inode *inode, struct file *file)
 105 {
 106         return single_open(file, bdi_debug_stats_show, inode->i_private);
 107 }
 108
 109 static const struct file_operations bdi_debug_stats_fops = {
 110         .open           = bdi_debug_stats_open,
 111         .read           = seq_read,
 112         .llseek         = seq_lseek,
 113         .release        = single_release,
 114 };
 115
 116 static void bdi_debug_register(struct backing_dev_info *bdi, const char *name)
 117 {
 118         bdi->debug_dir = debugfs_create_dir(name, bdi_debug_root);
 119         bdi->debug_stats = debugfs_create_file("stats", 0444, bdi->debug_dir,
 120                                                bdi, &bdi_debug_stats_fops);
 121 }
 122
 123 static void bdi_debug_unregister(struct backing_dev_info *bdi)
 124 {
 125         debugfs_remove(bdi->debug_stats);
 126         debugfs_remove(bdi->debug_dir);
 127 }
 128 #else
 129 static inline void bdi_debug_init(void)
 130 {
 131 }
 132 static inline void bdi_debug_register(struct backing_dev_info *bdi,
 133                                       const char *name)
 134 {
 135 }
 136 static inline void bdi_debug_unregister(struct backing_dev_info *bdi)
 137 {
 138 }
 139 #endif
 140
 141 static ssize_t read_ahead_kb_store(struct device *dev,
 142                                   struct device_attribute *attr,
 143                                   const char *buf, size_t count)
 144 {
 145         struct backing_dev_info *bdi = dev_get_drvdata(dev);
 146         char *end;
 147         unsigned long read_ahead_kb;
 148         ssize_t ret = -EINVAL;
 149
 150         read_ahead_kb = simple_strtoul(buf, &end, 10);
 151         if (*buf && (end[0] == '\0' || (end[0] == '\n' && end[1] == '\0'))) {
 152                 bdi->ra_pages = read_ahead_kb >> (PAGE_SHIFT - 10);
 153                 ret = count;
 154         }
 155         return ret;
 156 }
 157
 158 #define K(pages) ((pages) << (PAGE_SHIFT - 10))
 159
 160 #define BDI_SHOW(name, expr)                                            \
 161 static ssize_t name##_show(struct device *dev,                          \
 162                            struct device_attribute *attr, char *page)   \
 163 {                                                                       \
 164         struct backing_dev_info *bdi = dev_get_drvdata(dev);            \
 165                                                                         \
 166         return snprintf(page, PAGE_SIZE-1, "%lld\n", (long long)expr);  \
 167 }
 168
 169 BDI_SHOW(read_ahead_kb, K(bdi->ra_pages))
 170
 171 static ssize_t min_ratio_store(struct device *dev,
 172                 struct device_attribute *attr, const char *buf, size_t count)
 173 {
 174         struct backing_dev_info *bdi = dev_get_drvdata(dev);
 175         char *end;
 176         unsigned int ratio;
 177         ssize_t ret = -EINVAL;
 178
 179         ratio = simple_strtoul(buf, &end, 10);
 180         if (*buf && (end[0] == '\0' || (end[0] == '\n' && end[1] == '\0'))) {
 181                 ret = bdi_set_min_ratio(bdi, ratio);
 182                 if (!ret)
 183                         ret = count;
 184         }
 185         return ret;
 186 }
 187 BDI_SHOW(min_ratio, bdi->min_ratio)
 188
 189 static ssize_t max_ratio_store(struct device *dev,
 190                 struct device_attribute *attr, const char *buf, size_t count)
 191 {
 192         struct backing_dev_info *bdi = dev_get_drvdata(dev);
 193         char *end;
 194         unsigned int ratio;
 195         ssize_t ret = -EINVAL;
 196
 197         ratio = simple_strtoul(buf, &end, 10);
 198         if (*buf && (end[0] == '\0' || (end[0] == '\n' && end[1] == '\0'))) {
 199                 ret = bdi_set_max_ratio(bdi, ratio);
 200                 if (!ret)
 201                         ret = count;
 202         }
 203         return ret;
 204 }
 205 BDI_SHOW(max_ratio, bdi->max_ratio)
 206
 207 #define __ATTR_RW(attr) __ATTR(attr, 0644, attr##_show, attr##_store)
 208
 209 static struct device_attribute bdi_dev_attrs[] = {
 210         __ATTR_RW(read_ahead_kb),
 211         __ATTR_RW(min_ratio),
 212         __ATTR_RW(max_ratio),
 213         __ATTR_NULL,
 214 };
 215
 216 static __init int bdi_class_init(void)
 217 {
 218         bdi_class = class_create(THIS_MODULE, "bdi");
 219         if (IS_ERR(bdi_class))
 220                 return PTR_ERR(bdi_class);
 221
 222         bdi_class->dev_attrs = bdi_dev_attrs;
 223         bdi_debug_init();
 224         return 0;
 225 }
 226 postcore_initcall(bdi_class_init);
 227
 228 static int __init default_bdi_init(void)
 229 {
 230         int err;
 231
 232         sync_supers_tsk = kthread_run(bdi_sync_supers, NULL, "sync_supers");
 233         BUG_ON(IS_ERR(sync_supers_tsk));
 234
 235         setup_timer(&sync_supers_timer, sync_supers_timer_fn, 0);
 236         bdi_arm_supers_timer();
 237
 238         err = bdi_init(&default_backing_dev_info);
 239         if (!err)
 240                 bdi_register(&default_backing_dev_info, NULL, "default");
 241         err = bdi_init(&noop_backing_dev_info);
 242
 243         return err;
 244 }
 245 subsys_initcall(default_bdi_init);
 246
 247 int bdi_has_dirty_io(struct backing_dev_info *bdi)
 248 {
 249         return wb_has_dirty_io(&bdi->wb);
 250 }
 251
 252 static void bdi_flush_io(struct backing_dev_info *bdi)
 253 {
 254         struct writeback_control wbc = {
 255                 .sync_mode              = WB_SYNC_NONE,
 256                 .older_than_this        = NULL,
 257                 .range_cyclic           = 1,
 258                 .nr_to_write            = 1024,
 259         };
 260
 261         writeback_inodes_wb(&bdi->wb, &wbc);
 262 }
 263
 264 /*
 265  * kupdated() used to do this. We cannot do it from the bdi_forker_thread()
 266  * or we risk deadlocking on ->s_umount. The longer term solution would be
 267  * to implement sync_supers_bdi() or similar and simply do it from the
 268  * bdi writeback thread individually.
 269  */
 270 static int bdi_sync_supers(void *unused)
 271 {
 272         set_user_nice(current, 0);
 273
 274         while (!kthread_should_stop()) {
 275                 set_current_state(TASK_INTERRUPTIBLE);
 276                 schedule();
 277
 278                 /*
 279                  * Do this periodically, like kupdated() did before.
 280                  */
 281                 sync_supers();
 282         }
 283
 284         return 0;
 285 }
 286
 287 void bdi_arm_supers_timer(void)
 288 {
 289         unsigned long next;
 290
 291         if (!dirty_writeback_interval)
 292                 return;
 293
 294         next = msecs_to_jiffies(dirty_writeback_interval * 10) + jiffies;
 295         mod_timer(&sync_supers_timer, round_jiffies_up(next));
 296 }
 297
 298 static void sync_supers_timer_fn(unsigned long unused)
 299 {
 300         wake_up_process(sync_supers_tsk);
 301         bdi_arm_supers_timer();
 302 }
 303
 304 static void wakeup_timer_fn(unsigned long data)
 305 {
 306         struct backing_dev_info *bdi = (struct backing_dev_info *)data;
 307
 308         spin_lock_bh(&bdi->wb_lock);
 309         if (bdi->wb.task) {
 310                 trace_writeback_wake_thread(bdi);
 311                 wake_up_process(bdi->wb.task);
 312         } else {
 313                 /*
 314                  * When bdi tasks are inactive for long time, they are killed.
 315                  * In this case we have to wake-up the forker thread which
 316                  * should create and run the bdi thread.
 317                  */
 318                 trace_writeback_wake_forker_thread(bdi);
 319                 wake_up_process(default_backing_dev_info.wb.task);
 320         }
 321         spin_unlock_bh(&bdi->wb_lock);
 322 }
 323
 324 /*
 325  * This function is used when the first inode for this bdi is marked dirty. It
 326  * wakes-up the corresponding bdi thread which should then take care of the
 327  * periodic background write-out of dirty inodes. Since the write-out would
 328  * starts only 'dirty_writeback_interval' centisecs from now anyway, we just
 329  * set up a timer which wakes the bdi thread up later.
 330  *
 331  * Note, we wouldn't bother setting up the timer, but this function is on the
 332  * fast-path (used by '__mark_inode_dirty()'), so we save few context switches
 333  * by delaying the wake-up.
 334  */
 335 void bdi_wakeup_thread_delayed(struct backing_dev_info *bdi)
 336 {
 337         unsigned long timeout;
 338
 339         timeout = msecs_to_jiffies(dirty_writeback_interval * 10);
 340         mod_timer(&bdi->wb.wakeup_timer, jiffies + timeout);
 341 }
 342
 343 /*
 344  * Calculate the longest interval (jiffies) bdi threads are allowed to be
 345  * inactive.
 346  */
 347 static unsigned long bdi_longest_inactive(void)
 348 {
 349         unsigned long interval;
 350
 351         interval = msecs_to_jiffies(dirty_writeback_interval * 10);
 352         return max(5UL * 60 * HZ, interval);
 353 }
 354
 355 static int bdi_forker_thread(void *ptr)
 356 {
 357         struct bdi_writeback *me = ptr;
 358
 359         current->flags |= PF_SWAPWRITE;
 360         set_freezable();
 361
 362         /*
 363          * Our parent may run at a different priority, just set us to normal
 364          */
 365         set_user_nice(current, 0);
 366
 367         for (;;) {
 368                 struct task_struct *task = NULL;
 369                 struct backing_dev_info *bdi;
 370                 enum {
 371                         NO_ACTION,   /* Nothing to do */
 372                         FORK_THREAD, /* Fork bdi thread */
 373                         KILL_THREAD, /* Kill inactive bdi thread */
 374                 } action = NO_ACTION;
 375
 376                 /*
 377                  * Temporary measure, we want to make sure we don't see
 378                  * dirty data on the default backing_dev_info
 379                  */
 380                 if (wb_has_dirty_io(me) || !list_empty(&me->bdi->work_list)) {
 381                         del_timer(&me->wakeup_timer);
 382                         wb_do_writeback(me, 0);
 383                 }
 384
 385                 spin_lock_bh(&bdi_lock);
 386                 set_current_state(TASK_INTERRUPTIBLE);
 387
 388                 list_for_each_entry(bdi, &bdi_list, bdi_list) {
 389                         bool have_dirty_io;
 390
 391                         if (!bdi_cap_writeback_dirty(bdi) ||
 392                              bdi_cap_flush_forker(bdi))
 393                                 continue;
 394
 395                         WARN(!test_bit(BDI_registered, &bdi->state),
 396                              "bdi %p/%s is not registered!\n", bdi, bdi->name);
 397
 398                         have_dirty_io = !list_empty(&bdi->work_list) ||
 399                                         wb_has_dirty_io(&bdi->wb);
 400
 401                         /*
 402                          * If the bdi has work to do, but the thread does not
 403                          * exist - create it.
 404                          */
 405                         if (!bdi->wb.task && have_dirty_io) {
 406                                 /*
 407                                  * Set the pending bit - if someone will try to
 408                                  * unregister this bdi - it'll wait on this bit.
 409                                  */
 410                                 set_bit(BDI_pending, &bdi->state);
 411                                 action = FORK_THREAD;
 412                                 break;
 413                         }
 414
 415                         spin_lock(&bdi->wb_lock);
 416
 417                         /*
 418                          * If there is no work to do and the bdi thread was
 419                          * inactive long enough - kill it. The wb_lock is taken
 420                          * to make sure no-one adds more work to this bdi and
 421                          * wakes the bdi thread up.
 422                          */
 423                         if (bdi->wb.task && !have_dirty_io &&
 424                             time_after(jiffies, bdi->wb.last_active +
 425                                                 bdi_longest_inactive())) {
 426                                 task = bdi->wb.task;
 427                                 bdi->wb.task = NULL;
 428                                 spin_unlock(&bdi->wb_lock);
 429                                 set_bit(BDI_pending, &bdi->state);
 430                                 action = KILL_THREAD;
 431                                 break;
 432                         }
 433                         spin_unlock(&bdi->wb_lock);
 434                 }
 435                 spin_unlock_bh(&bdi_lock);
 436
 437                 /* Keep working if default bdi still has things to do */
 438                 if (!list_empty(&me->bdi->work_list))
 439                         __set_current_state(TASK_RUNNING);
 440
 441                 switch (action) {
 442                 case FORK_THREAD:
 443                         __set_current_state(TASK_RUNNING);
 444                         task = kthread_create(bdi_writeback_thread, &bdi->wb,
 445                                               "flush-%s", dev_name(bdi->dev));
 446                         if (IS_ERR(task)) {
 447                                 /*
 448                                  * If thread creation fails, force writeout of
 449                                  * the bdi from the thread.
 450                                  */
 451                                 bdi_flush_io(bdi);
 452                         } else {
 453                                 /*
 454                                  * The spinlock makes sure we do not lose
 455                                  * wake-ups when racing with 'bdi_queue_work()'.
 456                                  * And as soon as the bdi thread is visible, we
 457                                  * can start it.
 458                                  */
 459                                 spin_lock_bh(&bdi->wb_lock);
 460                                 bdi->wb.task = task;
 461                                 spin_unlock_bh(&bdi->wb_lock);
 462                                 wake_up_process(task);
 463                         }
 464                         break;
 465
 466                 case KILL_THREAD:
 467                         __set_current_state(TASK_RUNNING);
 468                         kthread_stop(task);
 469                         break;
 470
 471                 case NO_ACTION:
 472                         if (!wb_has_dirty_io(me) || !dirty_writeback_interval)
 473                                 /*
 474                                  * There are no dirty data. The only thing we
 475                                  * should now care about is checking for
 476                                  * inactive bdi threads and killing them. Thus,
 477                                  * let's sleep for longer time, save energy and
 478                                  * be friendly for battery-driven devices.
 479                                  */
 480                                 schedule_timeout(bdi_longest_inactive());
 481                         else
 482                                 schedule_timeout(msecs_to_jiffies(dirty_writeback_interval * 10));
 483                         try_to_freeze();
 484                         /* Back to the main loop */
 485                         continue;
 486                 }
 487
 488                 /*
 489                  * Clear pending bit and wakeup anybody waiting to tear us down.
 490                  */
 491                 clear_bit(BDI_pending, &bdi->state);
 492                 smp_mb__after_clear_bit();
 493                 wake_up_bit(&bdi->state, BDI_pending);
 494         }
 495
 496         return 0;
 497 }
 498
 499 /*
 500  * Remove bdi from bdi_list, and ensure that it is no longer visible
 501  */
 502 static void bdi_remove_from_list(struct backing_dev_info *bdi)
 503 {
 504         spin_lock_bh(&bdi_lock);
 505         list_del_rcu(&bdi->bdi_list);
 506         spin_unlock_bh(&bdi_lock);
 507
 508         synchronize_rcu();
 509 }
 510
 511 int bdi_register(struct backing_dev_info *bdi, struct device *parent,
 512                 const char *fmt, ...)
 513 {
 514         va_list args;
 515         struct device *dev;
 516
 517         if (bdi->dev)   /* The driver needs to use separate queues per device */
 518                 return 0;
 519
 520         va_start(args, fmt);
 521         dev = device_create_vargs(bdi_class, parent, MKDEV(0, 0), bdi, fmt, args);
 522         va_end(args);
 523         if (IS_ERR(dev))
 524                 return PTR_ERR(dev);
 525
 526         bdi->dev = dev;
 527
 528         /*
 529          * Just start the forker thread for our default backing_dev_info,
 530          * and add other bdi's to the list. They will get a thread created
 531          * on-demand when they need it.
 532          */
 533         if (bdi_cap_flush_forker(bdi)) {
 534                 struct bdi_writeback *wb = &bdi->wb;
 535
 536                 wb->task = kthread_run(bdi_forker_thread, wb, "bdi-%s",
 537                                                 dev_name(dev));
 538                 if (IS_ERR(wb->task))
 539                         return PTR_ERR(wb->task);
 540         }
 541
 542         bdi_debug_register(bdi, dev_name(dev));
 543         set_bit(BDI_registered, &bdi->state);
 544
 545         spin_lock_bh(&bdi_lock);
 546         list_add_tail_rcu(&bdi->bdi_list, &bdi_list);
 547         spin_unlock_bh(&bdi_lock);
 548
 549         trace_writeback_bdi_register(bdi);
 550         return 0;
 551 }
 552 EXPORT_SYMBOL(bdi_register);
 553
 554 int bdi_register_dev(struct backing_dev_info *bdi, dev_t dev)
 555 {
 556         return bdi_register(bdi, NULL, "%u:%u", MAJOR(dev), MINOR(dev));
 557 }
 558 EXPORT_SYMBOL(bdi_register_dev);
 559
 560 /*
 561  * Remove bdi from the global list and shutdown any threads we have running
 562  */
 563 static void bdi_wb_shutdown(struct backing_dev_info *bdi)
 564 {
 565         if (!bdi_cap_writeback_dirty(bdi))
 566                 return;
 567
 568         /*
 569          * Make sure nobody finds us on the bdi_list anymore
 570          */
 571         bdi_remove_from_list(bdi);
 572
 573         /*
 574          * If setup is pending, wait for that to complete first
 575          */
 576         wait_on_bit(&bdi->state, BDI_pending, bdi_sched_wait,
 577                         TASK_UNINTERRUPTIBLE);
 578
 579         /*
 580          * Finally, kill the kernel thread. We don't need to be RCU
 581          * safe anymore, since the bdi is gone from visibility. Force
 582          * unfreeze of the thread before calling kthread_stop(), otherwise
 583          * it would never exet if it is currently stuck in the refrigerator.
 584          */
 585         if (bdi->wb.task) {
 586                 thaw_process(bdi->wb.task);
 587                 kthread_stop(bdi->wb.task);
 588         }
 589 }
 590
 591 /*
 592  * This bdi is going away now, make sure that no super_blocks point to it
 593  */
 594 static void bdi_prune_sb(struct backing_dev_info *bdi)
 595 {
 596         struct super_block *sb;
 597
 598         spin_lock(&sb_lock);
 599         list_for_each_entry(sb, &super_blocks, s_list) {
 600                 if (sb->s_bdi == bdi)
 601                         sb->s_bdi = &default_backing_dev_info;
 602         }
 603         spin_unlock(&sb_lock);
 604 }
 605
 606 void bdi_unregister(struct backing_dev_info *bdi)
 607 {
 608         if (bdi->dev) {
 609                 trace_writeback_bdi_unregister(bdi);
 610                 bdi_prune_sb(bdi);
 611                 del_timer_sync(&bdi->wb.wakeup_timer);
 612
 613                 if (!bdi_cap_flush_forker(bdi))
 614                         bdi_wb_shutdown(bdi);
 615                 bdi_debug_unregister(bdi);
 616                 device_unregister(bdi->dev);
 617                 bdi->dev = NULL;
 618         }
 619 }
 620 EXPORT_SYMBOL(bdi_unregister);
 621
 622 static void bdi_wb_init(struct bdi_writeback *wb, struct backing_dev_info *bdi)
 623 {
 624         memset(wb, 0, sizeof(*wb));
 625
 626         wb->bdi = bdi;
 627         wb->last_old_flush = jiffies;
 628         INIT_LIST_HEAD(&wb->b_dirty);
 629         INIT_LIST_HEAD(&wb->b_io);
 630         INIT_LIST_HEAD(&wb->b_more_io);
 631         setup_timer(&wb->wakeup_timer, wakeup_timer_fn, (unsigned long)bdi);
 632 }
 633
 634 int bdi_init(struct backing_dev_info *bdi)
 635 {
 636         int i, err;
 637
 638         bdi->dev = NULL;
 639
 640         bdi->min_ratio = 0;
 641         bdi->max_ratio = 100;
 642         bdi->max_prop_frac = PROP_FRAC_BASE;
 643         spin_lock_init(&bdi->wb_lock);
 644         INIT_LIST_HEAD(&bdi->bdi_list);
 645         INIT_LIST_HEAD(&bdi->work_list);
 646
 647         bdi_wb_init(&bdi->wb, bdi);
 648
 649         for (i = 0; i < NR_BDI_STAT_ITEMS; i++) {
 650                 err = percpu_counter_init(&bdi->bdi_stat[i], 0);
 651                 if (err)
 652                         goto err;
 653         }
 654
 655         bdi->dirty_exceeded = 0;
 656         err = prop_local_init_percpu(&bdi->completions);
 657
 658         if (err) {
 659 err:
 660                 while (i--)
 661                         percpu_counter_destroy(&bdi->bdi_stat[i]);
 662         }
 663
 664         return err;
 665 }
 666 EXPORT_SYMBOL(bdi_init);
 667
 668 void bdi_destroy(struct backing_dev_info *bdi)
 669 {
 670         int i;
 671
 672         /*
 673          * Splice our entries to the default_backing_dev_info, if this
 674          * bdi disappears
 675          */
 676         if (bdi_has_dirty_io(bdi)) {
 677                 struct bdi_writeback *dst = &default_backing_dev_info.wb;
 678
 679                 spin_lock(&inode_wb_list_lock);
 680                 list_splice(&bdi->wb.b_dirty, &dst->b_dirty);
 681                 list_splice(&bdi->wb.b_io, &dst->b_io);
 682                 list_splice(&bdi->wb.b_more_io, &dst->b_more_io);
 683                 spin_unlock(&inode_wb_list_lock);
 684         }
 685
 686         bdi_unregister(bdi);
 687
 688         for (i = 0; i < NR_BDI_STAT_ITEMS; i++)
 689                 percpu_counter_destroy(&bdi->bdi_stat[i]);
 690
 691         prop_local_destroy_percpu(&bdi->completions);
 692 }
 693 EXPORT_SYMBOL(bdi_destroy);
 694
 695 /*
 696  * For use from filesystems to quickly init and register a bdi associated
 697  * with dirty writeback
 698  */
 699 int bdi_setup_and_register(struct backing_dev_info *bdi, char *name,
 700                            unsigned int cap)
 701 {
 702         char tmp[32];
 703         int err;
 704
 705         bdi->name = name;
 706         bdi->capabilities = cap;
 707         err = bdi_init(bdi);
 708         if (err)
 709                 return err;
 710
 711         sprintf(tmp, "%.28s%s", name, "-%d");
 712         err = bdi_register(bdi, NULL, tmp, atomic_long_inc_return(&bdi_seq));
 713         if (err) {
 714                 bdi_destroy(bdi);
 715                 return err;
 716         }
 717
 718         return 0;
 719 }
 720 EXPORT_SYMBOL(bdi_setup_and_register);
 721
 722 static wait_queue_head_t congestion_wqh[2] = {
 723                 __WAIT_QUEUE_HEAD_INITIALIZER(congestion_wqh[0]),
 724                 __WAIT_QUEUE_HEAD_INITIALIZER(congestion_wqh[1])
 725         };
 726 static atomic_t nr_bdi_congested[2];
 727
 728 void clear_bdi_congested(struct backing_dev_info *bdi, int sync)
 729 {
 730         enum bdi_state bit;
 731         wait_queue_head_t *wqh = &congestion_wqh[sync];
 732
 733         bit = sync ? BDI_sync_congested : BDI_async_congested;
 734         if (test_and_clear_bit(bit, &bdi->state))
 735                 atomic_dec(&nr_bdi_congested[sync]);
 736         smp_mb__after_clear_bit();
 737         if (waitqueue_active(wqh))
 738                 wake_up(wqh);
 739 }
 740 EXPORT_SYMBOL(clear_bdi_congested);
 741
 742 void set_bdi_congested(struct backing_dev_info *bdi, int sync)
 743 {
 744         enum bdi_state bit;
 745
 746         bit = sync ? BDI_sync_congested : BDI_async_congested;
 747         if (!test_and_set_bit(bit, &bdi->state))
 748                 atomic_inc(&nr_bdi_congested[sync]);
 749 }
 750 EXPORT_SYMBOL(set_bdi_congested);
 751
 752 /**
 753  * congestion_wait - wait for a backing_dev to become uncongested
 754  * @sync: SYNC or ASYNC IO
 755  * @timeout: timeout in jiffies
 756  *
 757  * Waits for up to @timeout jiffies for a backing_dev (any backing_dev) to exit
 758  * write congestion.  If no backing_devs are congested then just wait for the
 759  * next write to be completed.
 760  */
 761 long congestion_wait(int sync, long timeout)
 762 {
 763         long ret;
 764         unsigned long start = jiffies;
 765         DEFINE_WAIT(wait);
 766         wait_queue_head_t *wqh = &congestion_wqh[sync];
 767
 768         prepare_to_wait(wqh, &wait, TASK_UNINTERRUPTIBLE);
 769         ret = io_schedule_timeout(timeout);
 770         finish_wait(wqh, &wait);
 771
 772         trace_writeback_congestion_wait(jiffies_to_usecs(timeout),
 773                                         jiffies_to_usecs(jiffies - start));
 774
 775         return ret;
 776 }
 777 EXPORT_SYMBOL(congestion_wait);
 778
 779 /**
 780  * wait_iff_congested - Conditionally wait for a backing_dev to become uncongested or a zone to complete writes
 781  * @zone: A zone to check if it is heavily congested
 782  * @sync: SYNC or ASYNC IO
 783  * @timeout: timeout in jiffies
 784  *
 785  * In the event of a congested backing_dev (any backing_dev) and the given
 786  * @zone has experienced recent congestion, this waits for up to @timeout
 787  * jiffies for either a BDI to exit congestion of the given @sync queue
 788  * or a write to complete.
 789  *
 790  * In the absence of zone congestion, cond_resched() is called to yield
 791  * the processor if necessary but otherwise does not sleep.
 792  *
 793  * The return value is 0 if the sleep is for the full timeout. Otherwise,
 794  * it is the number of jiffies that were still remaining when the function
 795  * returned. return_value == timeout implies the function did not sleep.
 796  */
 797 long wait_iff_congested(struct zone *zone, int sync, long timeout)
 798 {
 799         long ret;
 800         unsigned long start = jiffies;
 801         DEFINE_WAIT(wait);
 802         wait_queue_head_t *wqh = &congestion_wqh[sync];
 803
 804         /*
 805          * If there is no congestion, or heavy congestion is not being
 806          * encountered in the current zone, yield if necessary instead
 807          * of sleeping on the congestion queue
 808          */
 809         if (atomic_read(&nr_bdi_congested[sync]) == 0 ||
 810                         !zone_is_reclaim_congested(zone)) {
 811                 cond_resched();
 812
 813                 /* In case we scheduled, work out time remaining */
 814                 ret = timeout - (jiffies - start);
 815                 if (ret < 0)
 816                         ret = 0;
 817
 818                 goto out;
 819         }
 820
 821         /* Sleep until uncongested or a write happens */
 822         prepare_to_wait(wqh, &wait, TASK_UNINTERRUPTIBLE);
 823         ret = io_schedule_timeout(timeout);
 824         finish_wait(wqh, &wait);
 825
 826 out:
 827         trace_writeback_wait_iff_congested(jiffies_to_usecs(timeout),
 828                                         jiffies_to_usecs(jiffies - start));
 829
 830         return ret;
 831 }
 832 EXPORT_SYMBOL(wait_iff_congested);