writeback: balance_dirty_pages(): reduce calls to global_page_state
[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
03ba3782 239 setup_timer(&sync_supers_timer, sync_supers_timer_fn, 0);
6423104b 240 bdi_arm_supers_timer();
03ba3782 241
26160158
JA
242 err = bdi_init(&default_backing_dev_info);
243 if (!err)
244 bdi_register(&default_backing_dev_info, NULL, "default");
245
246 return err;
247}
248subsys_initcall(default_bdi_init);
249
03ba3782
JA
250int bdi_has_dirty_io(struct backing_dev_info *bdi)
251{
252 return wb_has_dirty_io(&bdi->wb);
253}
254
255static void bdi_flush_io(struct backing_dev_info *bdi)
256{
257 struct writeback_control wbc = {
03ba3782
JA
258 .sync_mode = WB_SYNC_NONE,
259 .older_than_this = NULL,
260 .range_cyclic = 1,
261 .nr_to_write = 1024,
262 };
263
9c3a8ee8 264 writeback_inodes_wb(&bdi->wb, &wbc);
03ba3782
JA
265}
266
267/*
6f904ff0 268 * kupdated() used to do this. We cannot do it from the bdi_forker_thread()
03ba3782
JA
269 * or we risk deadlocking on ->s_umount. The longer term solution would be
270 * to implement sync_supers_bdi() or similar and simply do it from the
6f904ff0 271 * bdi writeback thread individually.
03ba3782
JA
272 */
273static int bdi_sync_supers(void *unused)
274{
275 set_user_nice(current, 0);
276
277 while (!kthread_should_stop()) {
278 set_current_state(TASK_INTERRUPTIBLE);
279 schedule();
280
281 /*
282 * Do this periodically, like kupdated() did before.
283 */
284 sync_supers();
285 }
286
287 return 0;
288}
289
6423104b 290void bdi_arm_supers_timer(void)
03ba3782
JA
291{
292 unsigned long next;
293
6423104b
JA
294 if (!dirty_writeback_interval)
295 return;
296
03ba3782
JA
297 next = msecs_to_jiffies(dirty_writeback_interval * 10) + jiffies;
298 mod_timer(&sync_supers_timer, round_jiffies_up(next));
299}
300
301static void sync_supers_timer_fn(unsigned long unused)
302{
303 wake_up_process(sync_supers_tsk);
6423104b 304 bdi_arm_supers_timer();
03ba3782
JA
305}
306
6467716a
AB
307static void wakeup_timer_fn(unsigned long data)
308{
309 struct backing_dev_info *bdi = (struct backing_dev_info *)data;
310
311 spin_lock_bh(&bdi->wb_lock);
312 if (bdi->wb.task) {
60332023 313 trace_writeback_wake_thread(bdi);
6467716a
AB
314 wake_up_process(bdi->wb.task);
315 } else {
316 /*
317 * When bdi tasks are inactive for long time, they are killed.
318 * In this case we have to wake-up the forker thread which
319 * should create and run the bdi thread.
320 */
60332023 321 trace_writeback_wake_forker_thread(bdi);
6467716a
AB
322 wake_up_process(default_backing_dev_info.wb.task);
323 }
324 spin_unlock_bh(&bdi->wb_lock);
325}
326
327/*
328 * This function is used when the first inode for this bdi is marked dirty. It
329 * wakes-up the corresponding bdi thread which should then take care of the
330 * periodic background write-out of dirty inodes. Since the write-out would
331 * starts only 'dirty_writeback_interval' centisecs from now anyway, we just
332 * set up a timer which wakes the bdi thread up later.
333 *
334 * Note, we wouldn't bother setting up the timer, but this function is on the
335 * fast-path (used by '__mark_inode_dirty()'), so we save few context switches
336 * by delaying the wake-up.
337 */
338void bdi_wakeup_thread_delayed(struct backing_dev_info *bdi)
339{
340 unsigned long timeout;
341
342 timeout = msecs_to_jiffies(dirty_writeback_interval * 10);
343 mod_timer(&bdi->wb.wakeup_timer, jiffies + timeout);
344}
345
fff5b85a
AB
346/*
347 * Calculate the longest interval (jiffies) bdi threads are allowed to be
348 * inactive.
349 */
350static unsigned long bdi_longest_inactive(void)
351{
352 unsigned long interval;
353
354 interval = msecs_to_jiffies(dirty_writeback_interval * 10);
355 return max(5UL * 60 * HZ, interval);
356}
357
6f904ff0 358static int bdi_forker_thread(void *ptr)
03ba3782
JA
359{
360 struct bdi_writeback *me = ptr;
361
c1955ce3
CH
362 current->flags |= PF_FLUSHER | PF_SWAPWRITE;
363 set_freezable();
364
365 /*
366 * Our parent may run at a different priority, just set us to normal
367 */
368 set_user_nice(current, 0);
03ba3782
JA
369
370 for (;;) {
fff5b85a 371 struct task_struct *task = NULL;
78c40cb6 372 struct backing_dev_info *bdi;
adf39240
AB
373 enum {
374 NO_ACTION, /* Nothing to do */
375 FORK_THREAD, /* Fork bdi thread */
fff5b85a 376 KILL_THREAD, /* Kill inactive bdi thread */
adf39240 377 } action = NO_ACTION;
03ba3782
JA
378
379 /*
380 * Temporary measure, we want to make sure we don't see
381 * dirty data on the default backing_dev_info
382 */
6467716a
AB
383 if (wb_has_dirty_io(me) || !list_empty(&me->bdi->work_list)) {
384 del_timer(&me->wakeup_timer);
03ba3782 385 wb_do_writeback(me, 0);
6467716a 386 }
03ba3782 387
cfc4ba53 388 spin_lock_bh(&bdi_lock);
c5f7ad23 389 set_current_state(TASK_INTERRUPTIBLE);
03ba3782 390
78c40cb6 391 list_for_each_entry(bdi, &bdi_list, bdi_list) {
adf39240
AB
392 bool have_dirty_io;
393
394 if (!bdi_cap_writeback_dirty(bdi) ||
395 bdi_cap_flush_forker(bdi))
03ba3782
JA
396 continue;
397
080dcec4
AB
398 WARN(!test_bit(BDI_registered, &bdi->state),
399 "bdi %p/%s is not registered!\n", bdi, bdi->name);
400
adf39240
AB
401 have_dirty_io = !list_empty(&bdi->work_list) ||
402 wb_has_dirty_io(&bdi->wb);
78c40cb6
AB
403
404 /*
adf39240
AB
405 * If the bdi has work to do, but the thread does not
406 * exist - create it.
78c40cb6 407 */
adf39240
AB
408 if (!bdi->wb.task && have_dirty_io) {
409 /*
410 * Set the pending bit - if someone will try to
411 * unregister this bdi - it'll wait on this bit.
412 */
413 set_bit(BDI_pending, &bdi->state);
414 action = FORK_THREAD;
415 break;
416 }
fff5b85a 417
6bf05d03
JA
418 spin_lock(&bdi->wb_lock);
419
fff5b85a
AB
420 /*
421 * If there is no work to do and the bdi thread was
422 * inactive long enough - kill it. The wb_lock is taken
423 * to make sure no-one adds more work to this bdi and
424 * wakes the bdi thread up.
425 */
426 if (bdi->wb.task && !have_dirty_io &&
427 time_after(jiffies, bdi->wb.last_active +
428 bdi_longest_inactive())) {
429 task = bdi->wb.task;
430 bdi->wb.task = NULL;
431 spin_unlock(&bdi->wb_lock);
432 set_bit(BDI_pending, &bdi->state);
433 action = KILL_THREAD;
434 break;
435 }
6bf05d03 436 spin_unlock(&bdi->wb_lock);
03ba3782 437 }
080dcec4 438 spin_unlock_bh(&bdi_lock);
03ba3782 439
c4ec7908
AB
440 /* Keep working if default bdi still has things to do */
441 if (!list_empty(&me->bdi->work_list))
442 __set_current_state(TASK_RUNNING);
443
adf39240
AB
444 switch (action) {
445 case FORK_THREAD:
446 __set_current_state(TASK_RUNNING);
447 task = kthread_run(bdi_writeback_thread, &bdi->wb, "flush-%s",
448 dev_name(bdi->dev));
449 if (IS_ERR(task)) {
450 /*
451 * If thread creation fails, force writeout of
452 * the bdi from the thread.
453 */
454 bdi_flush_io(bdi);
fff5b85a
AB
455 } else {
456 /*
457 * The spinlock makes sure we do not lose
458 * wake-ups when racing with 'bdi_queue_work()'.
459 */
6467716a 460 spin_lock_bh(&bdi->wb_lock);
adf39240 461 bdi->wb.task = task;
6467716a 462 spin_unlock_bh(&bdi->wb_lock);
fff5b85a
AB
463 }
464 break;
465
466 case KILL_THREAD:
467 __set_current_state(TASK_RUNNING);
468 kthread_stop(task);
adf39240 469 break;
03ba3782 470
adf39240 471 case NO_ACTION:
253c34e9
AB
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());
6423104b 481 else
253c34e9 482 schedule_timeout(msecs_to_jiffies(dirty_writeback_interval * 10));
03ba3782 483 try_to_freeze();
adf39240 484 /* Back to the main loop */
03ba3782
JA
485 continue;
486 }
fff5b85a
AB
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);
03ba3782
JA
494 }
495
496 return 0;
497}
498
cfc4ba53
JA
499/*
500 * Remove bdi from bdi_list, and ensure that it is no longer visible
501 */
502static 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
cf0ca9fe
PZ
511int bdi_register(struct backing_dev_info *bdi, struct device *parent,
512 const char *fmt, ...)
513{
cf0ca9fe 514 va_list args;
cf0ca9fe
PZ
515 struct device *dev;
516
69fc208b 517 if (bdi->dev) /* The driver needs to use separate queues per device */
c284de61 518 return 0;
f1d0b063 519
cf0ca9fe 520 va_start(args, fmt);
19051c50 521 dev = device_create_vargs(bdi_class, parent, MKDEV(0, 0), bdi, fmt, args);
cf0ca9fe 522 va_end(args);
c284de61
AB
523 if (IS_ERR(dev))
524 return PTR_ERR(dev);
66f3b8e2 525
cf0ca9fe 526 bdi->dev = dev;
cf0ca9fe 527
03ba3782
JA
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
6f904ff0 536 wb->task = kthread_run(bdi_forker_thread, wb, "bdi-%s",
03ba3782 537 dev_name(dev));
c284de61
AB
538 if (IS_ERR(wb->task))
539 return PTR_ERR(wb->task);
03ba3782
JA
540 }
541
542 bdi_debug_register(bdi, dev_name(dev));
500b067c 543 set_bit(BDI_registered, &bdi->state);
c284de61
AB
544
545 spin_lock_bh(&bdi_lock);
546 list_add_tail_rcu(&bdi->bdi_list, &bdi_list);
547 spin_unlock_bh(&bdi_lock);
548
455b2864 549 trace_writeback_bdi_register(bdi);
c284de61 550 return 0;
cf0ca9fe
PZ
551}
552EXPORT_SYMBOL(bdi_register);
553
554int 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}
558EXPORT_SYMBOL(bdi_register_dev);
559
03ba3782
JA
560/*
561 * Remove bdi from the global list and shutdown any threads we have running
562 */
563static void bdi_wb_shutdown(struct backing_dev_info *bdi)
66f3b8e2 564{
03ba3782
JA
565 if (!bdi_cap_writeback_dirty(bdi))
566 return;
567
568 /*
fff5b85a 569 * Make sure nobody finds us on the bdi_list anymore
03ba3782 570 */
fff5b85a 571 bdi_remove_from_list(bdi);
03ba3782
JA
572
573 /*
fff5b85a 574 * If setup is pending, wait for that to complete first
03ba3782 575 */
fff5b85a
AB
576 wait_on_bit(&bdi->state, BDI_pending, bdi_sched_wait,
577 TASK_UNINTERRUPTIBLE);
03ba3782
JA
578
579 /*
c1955ce3 580 * Finally, kill the kernel thread. We don't need to be RCU
c62b17a5
RD
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.
03ba3782 584 */
c1955ce3
CH
585 if (bdi->wb.task) {
586 thaw_process(bdi->wb.task);
587 kthread_stop(bdi->wb.task);
c62b17a5 588 }
66f3b8e2
JA
589}
590
592b09a4
JA
591/*
592 * This bdi is going away now, make sure that no super_blocks point to it
593 */
594static 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 = NULL;
602 }
603 spin_unlock(&sb_lock);
604}
605
cf0ca9fe
PZ
606void bdi_unregister(struct backing_dev_info *bdi)
607{
608 if (bdi->dev) {
455b2864 609 trace_writeback_bdi_unregister(bdi);
8c4db335 610 bdi_prune_sb(bdi);
6467716a 611 del_timer_sync(&bdi->wb.wakeup_timer);
8c4db335 612
03ba3782
JA
613 if (!bdi_cap_flush_forker(bdi))
614 bdi_wb_shutdown(bdi);
76f1418b 615 bdi_debug_unregister(bdi);
cf0ca9fe
PZ
616 device_unregister(bdi->dev);
617 bdi->dev = NULL;
618 }
619}
620EXPORT_SYMBOL(bdi_unregister);
3fcfab16 621
6467716a
AB
622static 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
b2e8fb6e
PZ
634int bdi_init(struct backing_dev_info *bdi)
635{
03ba3782 636 int i, err;
b2e8fb6e 637
cf0ca9fe
PZ
638 bdi->dev = NULL;
639
189d3c4a 640 bdi->min_ratio = 0;
a42dde04
PZ
641 bdi->max_ratio = 100;
642 bdi->max_prop_frac = PROP_FRAC_BASE;
03ba3782 643 spin_lock_init(&bdi->wb_lock);
66f3b8e2 644 INIT_LIST_HEAD(&bdi->bdi_list);
03ba3782
JA
645 INIT_LIST_HEAD(&bdi->work_list);
646
647 bdi_wb_init(&bdi->wb, bdi);
648
b2e8fb6e 649 for (i = 0; i < NR_BDI_STAT_ITEMS; i++) {
ea319518 650 err = percpu_counter_init(&bdi->bdi_stat[i], 0);
04fbfdc1
PZ
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) {
659err:
4b01a0b1 660 while (i--)
04fbfdc1 661 percpu_counter_destroy(&bdi->bdi_stat[i]);
b2e8fb6e
PZ
662 }
663
664 return err;
665}
666EXPORT_SYMBOL(bdi_init);
667
668void bdi_destroy(struct backing_dev_info *bdi)
669{
670 int i;
671
ce5f8e77
JA
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_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_lock);
684 }
66f3b8e2 685
cf0ca9fe
PZ
686 bdi_unregister(bdi);
687
b2e8fb6e
PZ
688 for (i = 0; i < NR_BDI_STAT_ITEMS; i++)
689 percpu_counter_destroy(&bdi->bdi_stat[i]);
04fbfdc1
PZ
690
691 prop_local_destroy_percpu(&bdi->completions);
b2e8fb6e
PZ
692}
693EXPORT_SYMBOL(bdi_destroy);
694
c3c53206
JA
695/*
696 * For use from filesystems to quickly init and register a bdi associated
697 * with dirty writeback
698 */
699int 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}
720EXPORT_SYMBOL(bdi_setup_and_register);
721
3fcfab16
AM
722static 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
1faa16d2 727void clear_bdi_congested(struct backing_dev_info *bdi, int sync)
3fcfab16
AM
728{
729 enum bdi_state bit;
1faa16d2 730 wait_queue_head_t *wqh = &congestion_wqh[sync];
3fcfab16 731
1faa16d2 732 bit = sync ? BDI_sync_congested : BDI_async_congested;
3fcfab16
AM
733 clear_bit(bit, &bdi->state);
734 smp_mb__after_clear_bit();
735 if (waitqueue_active(wqh))
736 wake_up(wqh);
737}
738EXPORT_SYMBOL(clear_bdi_congested);
739
1faa16d2 740void set_bdi_congested(struct backing_dev_info *bdi, int sync)
3fcfab16
AM
741{
742 enum bdi_state bit;
743
1faa16d2 744 bit = sync ? BDI_sync_congested : BDI_async_congested;
3fcfab16
AM
745 set_bit(bit, &bdi->state);
746}
747EXPORT_SYMBOL(set_bdi_congested);
748
749/**
750 * congestion_wait - wait for a backing_dev to become uncongested
8aa7e847 751 * @sync: SYNC or ASYNC IO
3fcfab16
AM
752 * @timeout: timeout in jiffies
753 *
754 * Waits for up to @timeout jiffies for a backing_dev (any backing_dev) to exit
755 * write congestion. If no backing_devs are congested then just wait for the
756 * next write to be completed.
757 */
8aa7e847 758long congestion_wait(int sync, long timeout)
3fcfab16
AM
759{
760 long ret;
761 DEFINE_WAIT(wait);
8aa7e847 762 wait_queue_head_t *wqh = &congestion_wqh[sync];
3fcfab16
AM
763
764 prepare_to_wait(wqh, &wait, TASK_UNINTERRUPTIBLE);
765 ret = io_schedule_timeout(timeout);
766 finish_wait(wqh, &wait);
767 return ret;
768}
769EXPORT_SYMBOL(congestion_wait);
04fbfdc1 770