writeback: prevent unnecessary bdi threads wakeups
[linux-2.6-block.git] / mm / backing-dev.c
CommitLineData
3fcfab16
AM
1
2#include <linux/wait.h>
3#include <linux/backing-dev.h>
03ba3782
JA
4#include <linux/kthread.h>
5#include <linux/freezer.h>
3fcfab16 6#include <linux/fs.h>
26160158 7#include <linux/pagemap.h>
03ba3782 8#include <linux/mm.h>
3fcfab16
AM
9#include <linux/sched.h>
10#include <linux/module.h>
cf0ca9fe
PZ
11#include <linux/writeback.h>
12#include <linux/device.h>
455b2864 13#include <trace/events/writeback.h>
cf0ca9fe 14
c3c53206
JA
15static atomic_long_t bdi_seq = ATOMIC_LONG_INIT(0);
16
26160158
JA
17void default_unplug_io_fn(struct backing_dev_info *bdi, struct page *page)
18{
19}
20EXPORT_SYMBOL(default_unplug_io_fn);
21
22struct backing_dev_info default_backing_dev_info = {
d993831f 23 .name = "default",
26160158
JA
24 .ra_pages = VM_MAX_READAHEAD * 1024 / PAGE_CACHE_SIZE,
25 .state = 0,
26 .capabilities = BDI_CAP_MAP_COPY,
27 .unplug_io_fn = default_unplug_io_fn,
28};
29EXPORT_SYMBOL_GPL(default_backing_dev_info);
cf0ca9fe 30
5129a469
JE
31struct backing_dev_info noop_backing_dev_info = {
32 .name = "noop",
33};
34EXPORT_SYMBOL_GPL(noop_backing_dev_info);
35
cf0ca9fe 36static struct class *bdi_class;
cfc4ba53
JA
37
38/*
39 * bdi_lock protects updates to bdi_list and bdi_pending_list, as well as
40 * reader side protection for bdi_pending_list. bdi_list has RCU reader side
41 * locking.
42 */
03ba3782 43DEFINE_SPINLOCK(bdi_lock);
66f3b8e2 44LIST_HEAD(bdi_list);
03ba3782
JA
45LIST_HEAD(bdi_pending_list);
46
47static struct task_struct *sync_supers_tsk;
48static struct timer_list sync_supers_timer;
49
50static int bdi_sync_supers(void *);
51static void sync_supers_timer_fn(unsigned long);
03ba3782 52
76f1418b
MS
53#ifdef CONFIG_DEBUG_FS
54#include <linux/debugfs.h>
55#include <linux/seq_file.h>
56
57static struct dentry *bdi_debug_root;
58
59static void bdi_debug_init(void)
60{
61 bdi_debug_root = debugfs_create_dir("bdi", NULL);
62}
63
64static int bdi_debug_stats_show(struct seq_file *m, void *v)
65{
66 struct backing_dev_info *bdi = m->private;
c1955ce3 67 struct bdi_writeback *wb = &bdi->wb;
364aeb28
DR
68 unsigned long background_thresh;
69 unsigned long dirty_thresh;
70 unsigned long bdi_thresh;
f09b00d3
JA
71 unsigned long nr_dirty, nr_io, nr_more_io, nr_wb;
72 struct inode *inode;
73
f09b00d3
JA
74 nr_wb = nr_dirty = nr_io = nr_more_io = 0;
75 spin_lock(&inode_lock);
c1955ce3
CH
76 list_for_each_entry(inode, &wb->b_dirty, i_list)
77 nr_dirty++;
78 list_for_each_entry(inode, &wb->b_io, i_list)
79 nr_io++;
80 list_for_each_entry(inode, &wb->b_more_io, i_list)
81 nr_more_io++;
f09b00d3 82 spin_unlock(&inode_lock);
76f1418b
MS
83
84 get_dirty_limits(&background_thresh, &dirty_thresh, &bdi_thresh, bdi);
85
86#define K(x) ((x) << (PAGE_SHIFT - 10))
87 seq_printf(m,
88 "BdiWriteback: %8lu kB\n"
89 "BdiReclaimable: %8lu kB\n"
90 "BdiDirtyThresh: %8lu kB\n"
91 "DirtyThresh: %8lu kB\n"
f09b00d3 92 "BackgroundThresh: %8lu kB\n"
f09b00d3
JA
93 "b_dirty: %8lu\n"
94 "b_io: %8lu\n"
95 "b_more_io: %8lu\n"
96 "bdi_list: %8u\n"
c1955ce3 97 "state: %8lx\n",
76f1418b
MS
98 (unsigned long) K(bdi_stat(bdi, BDI_WRITEBACK)),
99 (unsigned long) K(bdi_stat(bdi, BDI_RECLAIMABLE)),
f09b00d3 100 K(bdi_thresh), K(dirty_thresh),
c1955ce3
CH
101 K(background_thresh), nr_dirty, nr_io, nr_more_io,
102 !list_empty(&bdi->bdi_list), bdi->state);
76f1418b
MS
103#undef K
104
105 return 0;
106}
107
108static int bdi_debug_stats_open(struct inode *inode, struct file *file)
109{
110 return single_open(file, bdi_debug_stats_show, inode->i_private);
111}
112
113static const struct file_operations bdi_debug_stats_fops = {
114 .open = bdi_debug_stats_open,
115 .read = seq_read,
116 .llseek = seq_lseek,
117 .release = single_release,
118};
119
120static void bdi_debug_register(struct backing_dev_info *bdi, const char *name)
121{
122 bdi->debug_dir = debugfs_create_dir(name, bdi_debug_root);
123 bdi->debug_stats = debugfs_create_file("stats", 0444, bdi->debug_dir,
124 bdi, &bdi_debug_stats_fops);
125}
126
127static void bdi_debug_unregister(struct backing_dev_info *bdi)
128{
129 debugfs_remove(bdi->debug_stats);
130 debugfs_remove(bdi->debug_dir);
131}
132#else
133static inline void bdi_debug_init(void)
134{
135}
136static inline void bdi_debug_register(struct backing_dev_info *bdi,
137 const char *name)
138{
139}
140static inline void bdi_debug_unregister(struct backing_dev_info *bdi)
141{
142}
143#endif
144
cf0ca9fe
PZ
145static ssize_t read_ahead_kb_store(struct device *dev,
146 struct device_attribute *attr,
147 const char *buf, size_t count)
148{
149 struct backing_dev_info *bdi = dev_get_drvdata(dev);
150 char *end;
151 unsigned long read_ahead_kb;
152 ssize_t ret = -EINVAL;
153
154 read_ahead_kb = simple_strtoul(buf, &end, 10);
155 if (*buf && (end[0] == '\0' || (end[0] == '\n' && end[1] == '\0'))) {
156 bdi->ra_pages = read_ahead_kb >> (PAGE_SHIFT - 10);
157 ret = count;
158 }
159 return ret;
160}
161
162#define K(pages) ((pages) << (PAGE_SHIFT - 10))
163
164#define BDI_SHOW(name, expr) \
165static ssize_t name##_show(struct device *dev, \
166 struct device_attribute *attr, char *page) \
167{ \
168 struct backing_dev_info *bdi = dev_get_drvdata(dev); \
169 \
170 return snprintf(page, PAGE_SIZE-1, "%lld\n", (long long)expr); \
171}
172
173BDI_SHOW(read_ahead_kb, K(bdi->ra_pages))
174
189d3c4a
PZ
175static ssize_t min_ratio_store(struct device *dev,
176 struct device_attribute *attr, const char *buf, size_t count)
177{
178 struct backing_dev_info *bdi = dev_get_drvdata(dev);
179 char *end;
180 unsigned int ratio;
181 ssize_t ret = -EINVAL;
182
183 ratio = simple_strtoul(buf, &end, 10);
184 if (*buf && (end[0] == '\0' || (end[0] == '\n' && end[1] == '\0'))) {
185 ret = bdi_set_min_ratio(bdi, ratio);
186 if (!ret)
187 ret = count;
188 }
189 return ret;
190}
191BDI_SHOW(min_ratio, bdi->min_ratio)
192
a42dde04
PZ
193static ssize_t max_ratio_store(struct device *dev,
194 struct device_attribute *attr, const char *buf, size_t count)
195{
196 struct backing_dev_info *bdi = dev_get_drvdata(dev);
197 char *end;
198 unsigned int ratio;
199 ssize_t ret = -EINVAL;
200
201 ratio = simple_strtoul(buf, &end, 10);
202 if (*buf && (end[0] == '\0' || (end[0] == '\n' && end[1] == '\0'))) {
203 ret = bdi_set_max_ratio(bdi, ratio);
204 if (!ret)
205 ret = count;
206 }
207 return ret;
208}
209BDI_SHOW(max_ratio, bdi->max_ratio)
210
cf0ca9fe
PZ
211#define __ATTR_RW(attr) __ATTR(attr, 0644, attr##_show, attr##_store)
212
213static struct device_attribute bdi_dev_attrs[] = {
214 __ATTR_RW(read_ahead_kb),
189d3c4a 215 __ATTR_RW(min_ratio),
a42dde04 216 __ATTR_RW(max_ratio),
cf0ca9fe
PZ
217 __ATTR_NULL,
218};
219
220static __init int bdi_class_init(void)
221{
222 bdi_class = class_create(THIS_MODULE, "bdi");
14421453
AB
223 if (IS_ERR(bdi_class))
224 return PTR_ERR(bdi_class);
225
cf0ca9fe 226 bdi_class->dev_attrs = bdi_dev_attrs;
76f1418b 227 bdi_debug_init();
cf0ca9fe
PZ
228 return 0;
229}
76f1418b 230postcore_initcall(bdi_class_init);
cf0ca9fe 231
26160158
JA
232static int __init default_bdi_init(void)
233{
234 int err;
235
03ba3782
JA
236 sync_supers_tsk = kthread_run(bdi_sync_supers, NULL, "sync_supers");
237 BUG_ON(IS_ERR(sync_supers_tsk));
238
239 init_timer(&sync_supers_timer);
240 setup_timer(&sync_supers_timer, sync_supers_timer_fn, 0);
6423104b 241 bdi_arm_supers_timer();
03ba3782 242
26160158
JA
243 err = bdi_init(&default_backing_dev_info);
244 if (!err)
245 bdi_register(&default_backing_dev_info, NULL, "default");
246
247 return err;
248}
249subsys_initcall(default_bdi_init);
250
03ba3782
JA
251static void bdi_wb_init(struct bdi_writeback *wb, struct backing_dev_info *bdi)
252{
253 memset(wb, 0, sizeof(*wb));
254
255 wb->bdi = bdi;
256 wb->last_old_flush = jiffies;
257 INIT_LIST_HEAD(&wb->b_dirty);
258 INIT_LIST_HEAD(&wb->b_io);
259 INIT_LIST_HEAD(&wb->b_more_io);
260}
261
03ba3782
JA
262int bdi_has_dirty_io(struct backing_dev_info *bdi)
263{
264 return wb_has_dirty_io(&bdi->wb);
265}
266
267static void bdi_flush_io(struct backing_dev_info *bdi)
268{
269 struct writeback_control wbc = {
03ba3782
JA
270 .sync_mode = WB_SYNC_NONE,
271 .older_than_this = NULL,
272 .range_cyclic = 1,
273 .nr_to_write = 1024,
274 };
275
9c3a8ee8 276 writeback_inodes_wb(&bdi->wb, &wbc);
03ba3782
JA
277}
278
279/*
6f904ff0 280 * kupdated() used to do this. We cannot do it from the bdi_forker_thread()
03ba3782
JA
281 * or we risk deadlocking on ->s_umount. The longer term solution would be
282 * to implement sync_supers_bdi() or similar and simply do it from the
6f904ff0 283 * bdi writeback thread individually.
03ba3782
JA
284 */
285static int bdi_sync_supers(void *unused)
286{
287 set_user_nice(current, 0);
288
289 while (!kthread_should_stop()) {
290 set_current_state(TASK_INTERRUPTIBLE);
291 schedule();
292
293 /*
294 * Do this periodically, like kupdated() did before.
295 */
296 sync_supers();
297 }
298
299 return 0;
300}
301
6423104b 302void bdi_arm_supers_timer(void)
03ba3782
JA
303{
304 unsigned long next;
305
6423104b
JA
306 if (!dirty_writeback_interval)
307 return;
308
03ba3782
JA
309 next = msecs_to_jiffies(dirty_writeback_interval * 10) + jiffies;
310 mod_timer(&sync_supers_timer, round_jiffies_up(next));
311}
312
313static void sync_supers_timer_fn(unsigned long unused)
314{
315 wake_up_process(sync_supers_tsk);
6423104b 316 bdi_arm_supers_timer();
03ba3782
JA
317}
318
fff5b85a
AB
319/*
320 * Calculate the longest interval (jiffies) bdi threads are allowed to be
321 * inactive.
322 */
323static unsigned long bdi_longest_inactive(void)
324{
325 unsigned long interval;
326
327 interval = msecs_to_jiffies(dirty_writeback_interval * 10);
328 return max(5UL * 60 * HZ, interval);
329}
330
6f904ff0 331static int bdi_forker_thread(void *ptr)
03ba3782
JA
332{
333 struct bdi_writeback *me = ptr;
334
c1955ce3
CH
335 current->flags |= PF_FLUSHER | PF_SWAPWRITE;
336 set_freezable();
337
338 /*
339 * Our parent may run at a different priority, just set us to normal
340 */
341 set_user_nice(current, 0);
03ba3782
JA
342
343 for (;;) {
fff5b85a 344 struct task_struct *task = NULL;
78c40cb6 345 struct backing_dev_info *bdi;
adf39240
AB
346 enum {
347 NO_ACTION, /* Nothing to do */
348 FORK_THREAD, /* Fork bdi thread */
fff5b85a 349 KILL_THREAD, /* Kill inactive bdi thread */
adf39240 350 } action = NO_ACTION;
03ba3782
JA
351
352 /*
353 * Temporary measure, we want to make sure we don't see
354 * dirty data on the default backing_dev_info
355 */
356 if (wb_has_dirty_io(me) || !list_empty(&me->bdi->work_list))
357 wb_do_writeback(me, 0);
358
cfc4ba53 359 spin_lock_bh(&bdi_lock);
c5f7ad23 360 set_current_state(TASK_INTERRUPTIBLE);
03ba3782 361
78c40cb6 362 list_for_each_entry(bdi, &bdi_list, bdi_list) {
adf39240
AB
363 bool have_dirty_io;
364
365 if (!bdi_cap_writeback_dirty(bdi) ||
366 bdi_cap_flush_forker(bdi))
03ba3782
JA
367 continue;
368
080dcec4
AB
369 WARN(!test_bit(BDI_registered, &bdi->state),
370 "bdi %p/%s is not registered!\n", bdi, bdi->name);
371
adf39240
AB
372 have_dirty_io = !list_empty(&bdi->work_list) ||
373 wb_has_dirty_io(&bdi->wb);
78c40cb6
AB
374
375 /*
adf39240
AB
376 * If the bdi has work to do, but the thread does not
377 * exist - create it.
78c40cb6 378 */
adf39240
AB
379 if (!bdi->wb.task && have_dirty_io) {
380 /*
381 * Set the pending bit - if someone will try to
382 * unregister this bdi - it'll wait on this bit.
383 */
384 set_bit(BDI_pending, &bdi->state);
385 action = FORK_THREAD;
386 break;
387 }
fff5b85a
AB
388
389 spin_lock(&bdi->wb_lock);
390 /*
391 * If there is no work to do and the bdi thread was
392 * inactive long enough - kill it. The wb_lock is taken
393 * to make sure no-one adds more work to this bdi and
394 * wakes the bdi thread up.
395 */
396 if (bdi->wb.task && !have_dirty_io &&
397 time_after(jiffies, bdi->wb.last_active +
398 bdi_longest_inactive())) {
399 task = bdi->wb.task;
400 bdi->wb.task = NULL;
401 spin_unlock(&bdi->wb_lock);
402 set_bit(BDI_pending, &bdi->state);
403 action = KILL_THREAD;
404 break;
405 }
406 spin_unlock(&bdi->wb_lock);
03ba3782 407 }
080dcec4 408 spin_unlock_bh(&bdi_lock);
03ba3782 409
c4ec7908
AB
410 /* Keep working if default bdi still has things to do */
411 if (!list_empty(&me->bdi->work_list))
412 __set_current_state(TASK_RUNNING);
413
adf39240
AB
414 switch (action) {
415 case FORK_THREAD:
416 __set_current_state(TASK_RUNNING);
417 task = kthread_run(bdi_writeback_thread, &bdi->wb, "flush-%s",
418 dev_name(bdi->dev));
419 if (IS_ERR(task)) {
420 /*
421 * If thread creation fails, force writeout of
422 * the bdi from the thread.
423 */
424 bdi_flush_io(bdi);
fff5b85a
AB
425 } else {
426 /*
427 * The spinlock makes sure we do not lose
428 * wake-ups when racing with 'bdi_queue_work()'.
429 */
430 spin_lock(&bdi->wb_lock);
adf39240 431 bdi->wb.task = task;
fff5b85a
AB
432 spin_unlock(&bdi->wb_lock);
433 }
434 break;
435
436 case KILL_THREAD:
437 __set_current_state(TASK_RUNNING);
438 kthread_stop(task);
adf39240 439 break;
03ba3782 440
adf39240 441 case NO_ACTION:
253c34e9
AB
442 if (!wb_has_dirty_io(me) || !dirty_writeback_interval)
443 /*
444 * There are no dirty data. The only thing we
445 * should now care about is checking for
446 * inactive bdi threads and killing them. Thus,
447 * let's sleep for longer time, save energy and
448 * be friendly for battery-driven devices.
449 */
450 schedule_timeout(bdi_longest_inactive());
6423104b 451 else
253c34e9 452 schedule_timeout(msecs_to_jiffies(dirty_writeback_interval * 10));
03ba3782 453 try_to_freeze();
adf39240 454 /* Back to the main loop */
03ba3782
JA
455 continue;
456 }
fff5b85a
AB
457
458 /*
459 * Clear pending bit and wakeup anybody waiting to tear us down.
460 */
461 clear_bit(BDI_pending, &bdi->state);
462 smp_mb__after_clear_bit();
463 wake_up_bit(&bdi->state, BDI_pending);
03ba3782
JA
464 }
465
466 return 0;
467}
468
cfc4ba53
JA
469/*
470 * Remove bdi from bdi_list, and ensure that it is no longer visible
471 */
472static void bdi_remove_from_list(struct backing_dev_info *bdi)
473{
474 spin_lock_bh(&bdi_lock);
475 list_del_rcu(&bdi->bdi_list);
476 spin_unlock_bh(&bdi_lock);
477
478 synchronize_rcu();
479}
480
cf0ca9fe
PZ
481int bdi_register(struct backing_dev_info *bdi, struct device *parent,
482 const char *fmt, ...)
483{
cf0ca9fe
PZ
484 va_list args;
485 int ret = 0;
486 struct device *dev;
487
69fc208b 488 if (bdi->dev) /* The driver needs to use separate queues per device */
f1d0b063
KS
489 goto exit;
490
cf0ca9fe 491 va_start(args, fmt);
19051c50 492 dev = device_create_vargs(bdi_class, parent, MKDEV(0, 0), bdi, fmt, args);
cf0ca9fe 493 va_end(args);
cf0ca9fe
PZ
494 if (IS_ERR(dev)) {
495 ret = PTR_ERR(dev);
496 goto exit;
497 }
498
cfc4ba53
JA
499 spin_lock_bh(&bdi_lock);
500 list_add_tail_rcu(&bdi->bdi_list, &bdi_list);
501 spin_unlock_bh(&bdi_lock);
66f3b8e2 502
cf0ca9fe 503 bdi->dev = dev;
cf0ca9fe 504
03ba3782
JA
505 /*
506 * Just start the forker thread for our default backing_dev_info,
507 * and add other bdi's to the list. They will get a thread created
508 * on-demand when they need it.
509 */
510 if (bdi_cap_flush_forker(bdi)) {
511 struct bdi_writeback *wb = &bdi->wb;
512
6f904ff0 513 wb->task = kthread_run(bdi_forker_thread, wb, "bdi-%s",
03ba3782
JA
514 dev_name(dev));
515 if (IS_ERR(wb->task)) {
516 wb->task = NULL;
517 ret = -ENOMEM;
518
cfc4ba53 519 bdi_remove_from_list(bdi);
03ba3782
JA
520 goto exit;
521 }
522 }
523
524 bdi_debug_register(bdi, dev_name(dev));
500b067c 525 set_bit(BDI_registered, &bdi->state);
455b2864 526 trace_writeback_bdi_register(bdi);
cf0ca9fe 527exit:
cf0ca9fe
PZ
528 return ret;
529}
530EXPORT_SYMBOL(bdi_register);
531
532int bdi_register_dev(struct backing_dev_info *bdi, dev_t dev)
533{
534 return bdi_register(bdi, NULL, "%u:%u", MAJOR(dev), MINOR(dev));
535}
536EXPORT_SYMBOL(bdi_register_dev);
537
03ba3782
JA
538/*
539 * Remove bdi from the global list and shutdown any threads we have running
540 */
541static void bdi_wb_shutdown(struct backing_dev_info *bdi)
66f3b8e2 542{
03ba3782
JA
543 if (!bdi_cap_writeback_dirty(bdi))
544 return;
545
546 /*
fff5b85a 547 * Make sure nobody finds us on the bdi_list anymore
03ba3782 548 */
fff5b85a 549 bdi_remove_from_list(bdi);
03ba3782
JA
550
551 /*
fff5b85a 552 * If setup is pending, wait for that to complete first
03ba3782 553 */
fff5b85a
AB
554 wait_on_bit(&bdi->state, BDI_pending, bdi_sched_wait,
555 TASK_UNINTERRUPTIBLE);
03ba3782
JA
556
557 /*
c1955ce3 558 * Finally, kill the kernel thread. We don't need to be RCU
c62b17a5
RD
559 * safe anymore, since the bdi is gone from visibility. Force
560 * unfreeze of the thread before calling kthread_stop(), otherwise
561 * it would never exet if it is currently stuck in the refrigerator.
03ba3782 562 */
c1955ce3
CH
563 if (bdi->wb.task) {
564 thaw_process(bdi->wb.task);
565 kthread_stop(bdi->wb.task);
c62b17a5 566 }
66f3b8e2
JA
567}
568
592b09a4
JA
569/*
570 * This bdi is going away now, make sure that no super_blocks point to it
571 */
572static void bdi_prune_sb(struct backing_dev_info *bdi)
573{
574 struct super_block *sb;
575
576 spin_lock(&sb_lock);
577 list_for_each_entry(sb, &super_blocks, s_list) {
578 if (sb->s_bdi == bdi)
579 sb->s_bdi = NULL;
580 }
581 spin_unlock(&sb_lock);
582}
583
cf0ca9fe
PZ
584void bdi_unregister(struct backing_dev_info *bdi)
585{
586 if (bdi->dev) {
455b2864 587 trace_writeback_bdi_unregister(bdi);
8c4db335
JA
588 bdi_prune_sb(bdi);
589
03ba3782
JA
590 if (!bdi_cap_flush_forker(bdi))
591 bdi_wb_shutdown(bdi);
76f1418b 592 bdi_debug_unregister(bdi);
cf0ca9fe
PZ
593 device_unregister(bdi->dev);
594 bdi->dev = NULL;
595 }
596}
597EXPORT_SYMBOL(bdi_unregister);
3fcfab16 598
b2e8fb6e
PZ
599int bdi_init(struct backing_dev_info *bdi)
600{
03ba3782 601 int i, err;
b2e8fb6e 602
cf0ca9fe
PZ
603 bdi->dev = NULL;
604
189d3c4a 605 bdi->min_ratio = 0;
a42dde04
PZ
606 bdi->max_ratio = 100;
607 bdi->max_prop_frac = PROP_FRAC_BASE;
03ba3782 608 spin_lock_init(&bdi->wb_lock);
66f3b8e2 609 INIT_LIST_HEAD(&bdi->bdi_list);
03ba3782
JA
610 INIT_LIST_HEAD(&bdi->work_list);
611
612 bdi_wb_init(&bdi->wb, bdi);
613
b2e8fb6e 614 for (i = 0; i < NR_BDI_STAT_ITEMS; i++) {
ea319518 615 err = percpu_counter_init(&bdi->bdi_stat[i], 0);
04fbfdc1
PZ
616 if (err)
617 goto err;
618 }
619
620 bdi->dirty_exceeded = 0;
621 err = prop_local_init_percpu(&bdi->completions);
622
623 if (err) {
624err:
4b01a0b1 625 while (i--)
04fbfdc1 626 percpu_counter_destroy(&bdi->bdi_stat[i]);
b2e8fb6e
PZ
627 }
628
629 return err;
630}
631EXPORT_SYMBOL(bdi_init);
632
633void bdi_destroy(struct backing_dev_info *bdi)
634{
635 int i;
636
ce5f8e77
JA
637 /*
638 * Splice our entries to the default_backing_dev_info, if this
639 * bdi disappears
640 */
641 if (bdi_has_dirty_io(bdi)) {
642 struct bdi_writeback *dst = &default_backing_dev_info.wb;
643
644 spin_lock(&inode_lock);
645 list_splice(&bdi->wb.b_dirty, &dst->b_dirty);
646 list_splice(&bdi->wb.b_io, &dst->b_io);
647 list_splice(&bdi->wb.b_more_io, &dst->b_more_io);
648 spin_unlock(&inode_lock);
649 }
66f3b8e2 650
cf0ca9fe
PZ
651 bdi_unregister(bdi);
652
b2e8fb6e
PZ
653 for (i = 0; i < NR_BDI_STAT_ITEMS; i++)
654 percpu_counter_destroy(&bdi->bdi_stat[i]);
04fbfdc1
PZ
655
656 prop_local_destroy_percpu(&bdi->completions);
b2e8fb6e
PZ
657}
658EXPORT_SYMBOL(bdi_destroy);
659
c3c53206
JA
660/*
661 * For use from filesystems to quickly init and register a bdi associated
662 * with dirty writeback
663 */
664int bdi_setup_and_register(struct backing_dev_info *bdi, char *name,
665 unsigned int cap)
666{
667 char tmp[32];
668 int err;
669
670 bdi->name = name;
671 bdi->capabilities = cap;
672 err = bdi_init(bdi);
673 if (err)
674 return err;
675
676 sprintf(tmp, "%.28s%s", name, "-%d");
677 err = bdi_register(bdi, NULL, tmp, atomic_long_inc_return(&bdi_seq));
678 if (err) {
679 bdi_destroy(bdi);
680 return err;
681 }
682
683 return 0;
684}
685EXPORT_SYMBOL(bdi_setup_and_register);
686
3fcfab16
AM
687static wait_queue_head_t congestion_wqh[2] = {
688 __WAIT_QUEUE_HEAD_INITIALIZER(congestion_wqh[0]),
689 __WAIT_QUEUE_HEAD_INITIALIZER(congestion_wqh[1])
690 };
691
1faa16d2 692void clear_bdi_congested(struct backing_dev_info *bdi, int sync)
3fcfab16
AM
693{
694 enum bdi_state bit;
1faa16d2 695 wait_queue_head_t *wqh = &congestion_wqh[sync];
3fcfab16 696
1faa16d2 697 bit = sync ? BDI_sync_congested : BDI_async_congested;
3fcfab16
AM
698 clear_bit(bit, &bdi->state);
699 smp_mb__after_clear_bit();
700 if (waitqueue_active(wqh))
701 wake_up(wqh);
702}
703EXPORT_SYMBOL(clear_bdi_congested);
704
1faa16d2 705void set_bdi_congested(struct backing_dev_info *bdi, int sync)
3fcfab16
AM
706{
707 enum bdi_state bit;
708
1faa16d2 709 bit = sync ? BDI_sync_congested : BDI_async_congested;
3fcfab16
AM
710 set_bit(bit, &bdi->state);
711}
712EXPORT_SYMBOL(set_bdi_congested);
713
714/**
715 * congestion_wait - wait for a backing_dev to become uncongested
8aa7e847 716 * @sync: SYNC or ASYNC IO
3fcfab16
AM
717 * @timeout: timeout in jiffies
718 *
719 * Waits for up to @timeout jiffies for a backing_dev (any backing_dev) to exit
720 * write congestion. If no backing_devs are congested then just wait for the
721 * next write to be completed.
722 */
8aa7e847 723long congestion_wait(int sync, long timeout)
3fcfab16
AM
724{
725 long ret;
726 DEFINE_WAIT(wait);
8aa7e847 727 wait_queue_head_t *wqh = &congestion_wqh[sync];
3fcfab16
AM
728
729 prepare_to_wait(wqh, &wait, TASK_UNINTERRUPTIBLE);
730 ret = io_schedule_timeout(timeout);
731 finish_wait(wqh, &wait);
732 return ret;
733}
734EXPORT_SYMBOL(congestion_wait);
04fbfdc1 735