blob: 6ddd6a29c73b200295ab6255fce988d7b74ade3a [file] [log] [blame]
Linus Torvalds1da177e2005-04-16 15:20:36 -07001/*
2 * mm/page-writeback.c.
3 *
4 * Copyright (C) 2002, Linus Torvalds.
5 *
6 * Contains functions related to writing back dirty pages at the
7 * address_space level.
8 *
9 * 10Apr2002 akpm@zip.com.au
10 * Initial version
11 */
12
13#include <linux/kernel.h>
14#include <linux/module.h>
15#include <linux/spinlock.h>
16#include <linux/fs.h>
17#include <linux/mm.h>
18#include <linux/swap.h>
19#include <linux/slab.h>
20#include <linux/pagemap.h>
21#include <linux/writeback.h>
22#include <linux/init.h>
23#include <linux/backing-dev.h>
24#include <linux/blkdev.h>
25#include <linux/mpage.h>
26#include <linux/percpu.h>
27#include <linux/notifier.h>
28#include <linux/smp.h>
29#include <linux/sysctl.h>
30#include <linux/cpu.h>
31#include <linux/syscalls.h>
32
33/*
34 * The maximum number of pages to writeout in a single bdflush/kupdate
35 * operation. We do this so we don't hold I_LOCK against an inode for
36 * enormous amounts of time, which would block a userspace task which has
37 * been forced to throttle against that inode. Also, the code reevaluates
38 * the dirty each time it has written this many pages.
39 */
40#define MAX_WRITEBACK_PAGES 1024
41
42/*
43 * After a CPU has dirtied this many pages, balance_dirty_pages_ratelimited
44 * will look to see if it needs to force writeback or throttling.
45 */
46static long ratelimit_pages = 32;
47
48static long total_pages; /* The total number of pages in the machine. */
49static int dirty_exceeded; /* Dirty mem may be over limit */
50
51/*
52 * When balance_dirty_pages decides that the caller needs to perform some
53 * non-background writeback, this is how many pages it will attempt to write.
54 * It should be somewhat larger than RATELIMIT_PAGES to ensure that reasonably
55 * large amounts of I/O are submitted.
56 */
57static inline long sync_writeback_pages(void)
58{
59 return ratelimit_pages + ratelimit_pages / 2;
60}
61
62/* The following parameters are exported via /proc/sys/vm */
63
64/*
65 * Start background writeback (via pdflush) at this percentage
66 */
67int dirty_background_ratio = 10;
68
69/*
70 * The generator of dirty data starts writeback at this percentage
71 */
72int vm_dirty_ratio = 40;
73
74/*
75 * The interval between `kupdate'-style writebacks, in centiseconds
76 * (hundredths of a second)
77 */
78int dirty_writeback_centisecs = 5 * 100;
79
80/*
81 * The longest number of centiseconds for which data is allowed to remain dirty
82 */
83int dirty_expire_centisecs = 30 * 100;
84
85/*
86 * Flag that makes the machine dump writes/reads and block dirtyings.
87 */
88int block_dump;
89
90/*
91 * Flag that puts the machine in "laptop mode".
92 */
93int laptop_mode;
94
95EXPORT_SYMBOL(laptop_mode);
96
97/* End of sysctl-exported parameters */
98
99
100static void background_writeout(unsigned long _min_pages);
101
102struct writeback_state
103{
104 unsigned long nr_dirty;
105 unsigned long nr_unstable;
106 unsigned long nr_mapped;
107 unsigned long nr_writeback;
108};
109
110static void get_writeback_state(struct writeback_state *wbs)
111{
112 wbs->nr_dirty = read_page_state(nr_dirty);
113 wbs->nr_unstable = read_page_state(nr_unstable);
114 wbs->nr_mapped = read_page_state(nr_mapped);
115 wbs->nr_writeback = read_page_state(nr_writeback);
116}
117
118/*
119 * Work out the current dirty-memory clamping and background writeout
120 * thresholds.
121 *
122 * The main aim here is to lower them aggressively if there is a lot of mapped
123 * memory around. To avoid stressing page reclaim with lots of unreclaimable
124 * pages. It is better to clamp down on writers than to start swapping, and
125 * performing lots of scanning.
126 *
127 * We only allow 1/2 of the currently-unmapped memory to be dirtied.
128 *
129 * We don't permit the clamping level to fall below 5% - that is getting rather
130 * excessive.
131 *
132 * We make sure that the background writeout level is below the adjusted
133 * clamping level.
134 */
135static void
136get_dirty_limits(struct writeback_state *wbs, long *pbackground, long *pdirty,
137 struct address_space *mapping)
138{
139 int background_ratio; /* Percentages */
140 int dirty_ratio;
141 int unmapped_ratio;
142 long background;
143 long dirty;
144 unsigned long available_memory = total_pages;
145 struct task_struct *tsk;
146
147 get_writeback_state(wbs);
148
149#ifdef CONFIG_HIGHMEM
150 /*
151 * If this mapping can only allocate from low memory,
152 * we exclude high memory from our count.
153 */
154 if (mapping && !(mapping_gfp_mask(mapping) & __GFP_HIGHMEM))
155 available_memory -= totalhigh_pages;
156#endif
157
158
159 unmapped_ratio = 100 - (wbs->nr_mapped * 100) / total_pages;
160
161 dirty_ratio = vm_dirty_ratio;
162 if (dirty_ratio > unmapped_ratio / 2)
163 dirty_ratio = unmapped_ratio / 2;
164
165 if (dirty_ratio < 5)
166 dirty_ratio = 5;
167
168 background_ratio = dirty_background_ratio;
169 if (background_ratio >= dirty_ratio)
170 background_ratio = dirty_ratio / 2;
171
172 background = (background_ratio * available_memory) / 100;
173 dirty = (dirty_ratio * available_memory) / 100;
174 tsk = current;
175 if (tsk->flags & PF_LESS_THROTTLE || rt_task(tsk)) {
176 background += background / 4;
177 dirty += dirty / 4;
178 }
179 *pbackground = background;
180 *pdirty = dirty;
181}
182
183/*
184 * balance_dirty_pages() must be called by processes which are generating dirty
185 * data. It looks at the number of dirty pages in the machine and will force
186 * the caller to perform writeback if the system is over `vm_dirty_ratio'.
187 * If we're over `background_thresh' then pdflush is woken to perform some
188 * writeout.
189 */
190static void balance_dirty_pages(struct address_space *mapping)
191{
192 struct writeback_state wbs;
193 long nr_reclaimable;
194 long background_thresh;
195 long dirty_thresh;
196 unsigned long pages_written = 0;
197 unsigned long write_chunk = sync_writeback_pages();
198
199 struct backing_dev_info *bdi = mapping->backing_dev_info;
200
201 for (;;) {
202 struct writeback_control wbc = {
203 .bdi = bdi,
204 .sync_mode = WB_SYNC_NONE,
205 .older_than_this = NULL,
206 .nr_to_write = write_chunk,
207 };
208
209 get_dirty_limits(&wbs, &background_thresh,
210 &dirty_thresh, mapping);
211 nr_reclaimable = wbs.nr_dirty + wbs.nr_unstable;
212 if (nr_reclaimable + wbs.nr_writeback <= dirty_thresh)
213 break;
214
215 dirty_exceeded = 1;
216
217 /* Note: nr_reclaimable denotes nr_dirty + nr_unstable.
218 * Unstable writes are a feature of certain networked
219 * filesystems (i.e. NFS) in which data may have been
220 * written to the server's write cache, but has not yet
221 * been flushed to permanent storage.
222 */
223 if (nr_reclaimable) {
224 writeback_inodes(&wbc);
225 get_dirty_limits(&wbs, &background_thresh,
226 &dirty_thresh, mapping);
227 nr_reclaimable = wbs.nr_dirty + wbs.nr_unstable;
228 if (nr_reclaimable + wbs.nr_writeback <= dirty_thresh)
229 break;
230 pages_written += write_chunk - wbc.nr_to_write;
231 if (pages_written >= write_chunk)
232 break; /* We've done our duty */
233 }
234 blk_congestion_wait(WRITE, HZ/10);
235 }
236
237 if (nr_reclaimable + wbs.nr_writeback <= dirty_thresh)
238 dirty_exceeded = 0;
239
240 if (writeback_in_progress(bdi))
241 return; /* pdflush is already working this queue */
242
243 /*
244 * In laptop mode, we wait until hitting the higher threshold before
245 * starting background writeout, and then write out all the way down
246 * to the lower threshold. So slow writers cause minimal disk activity.
247 *
248 * In normal mode, we start background writeout at the lower
249 * background_thresh, to keep the amount of dirty memory low.
250 */
251 if ((laptop_mode && pages_written) ||
252 (!laptop_mode && (nr_reclaimable > background_thresh)))
253 pdflush_operation(background_writeout, 0);
254}
255
256/**
257 * balance_dirty_pages_ratelimited - balance dirty memory state
258 * @mapping - address_space which was dirtied
259 *
260 * Processes which are dirtying memory should call in here once for each page
261 * which was newly dirtied. The function will periodically check the system's
262 * dirty state and will initiate writeback if needed.
263 *
264 * On really big machines, get_writeback_state is expensive, so try to avoid
265 * calling it too often (ratelimiting). But once we're over the dirty memory
266 * limit we decrease the ratelimiting by a lot, to prevent individual processes
267 * from overshooting the limit by (ratelimit_pages) each.
268 */
269void balance_dirty_pages_ratelimited(struct address_space *mapping)
270{
271 static DEFINE_PER_CPU(int, ratelimits) = 0;
272 long ratelimit;
273
274 ratelimit = ratelimit_pages;
275 if (dirty_exceeded)
276 ratelimit = 8;
277
278 /*
279 * Check the rate limiting. Also, we do not want to throttle real-time
280 * tasks in balance_dirty_pages(). Period.
281 */
282 if (get_cpu_var(ratelimits)++ >= ratelimit) {
283 __get_cpu_var(ratelimits) = 0;
284 put_cpu_var(ratelimits);
285 balance_dirty_pages(mapping);
286 return;
287 }
288 put_cpu_var(ratelimits);
289}
290EXPORT_SYMBOL(balance_dirty_pages_ratelimited);
291
292void throttle_vm_writeout(void)
293{
294 struct writeback_state wbs;
295 long background_thresh;
296 long dirty_thresh;
297
298 for ( ; ; ) {
299 get_dirty_limits(&wbs, &background_thresh, &dirty_thresh, NULL);
300
301 /*
302 * Boost the allowable dirty threshold a bit for page
303 * allocators so they don't get DoS'ed by heavy writers
304 */
305 dirty_thresh += dirty_thresh / 10; /* wheeee... */
306
307 if (wbs.nr_unstable + wbs.nr_writeback <= dirty_thresh)
308 break;
309 blk_congestion_wait(WRITE, HZ/10);
310 }
311}
312
313
314/*
315 * writeback at least _min_pages, and keep writing until the amount of dirty
316 * memory is less than the background threshold, or until we're all clean.
317 */
318static void background_writeout(unsigned long _min_pages)
319{
320 long min_pages = _min_pages;
321 struct writeback_control wbc = {
322 .bdi = NULL,
323 .sync_mode = WB_SYNC_NONE,
324 .older_than_this = NULL,
325 .nr_to_write = 0,
326 .nonblocking = 1,
327 };
328
329 for ( ; ; ) {
330 struct writeback_state wbs;
331 long background_thresh;
332 long dirty_thresh;
333
334 get_dirty_limits(&wbs, &background_thresh, &dirty_thresh, NULL);
335 if (wbs.nr_dirty + wbs.nr_unstable < background_thresh
336 && min_pages <= 0)
337 break;
338 wbc.encountered_congestion = 0;
339 wbc.nr_to_write = MAX_WRITEBACK_PAGES;
340 wbc.pages_skipped = 0;
341 writeback_inodes(&wbc);
342 min_pages -= MAX_WRITEBACK_PAGES - wbc.nr_to_write;
343 if (wbc.nr_to_write > 0 || wbc.pages_skipped > 0) {
344 /* Wrote less than expected */
345 blk_congestion_wait(WRITE, HZ/10);
346 if (!wbc.encountered_congestion)
347 break;
348 }
349 }
350}
351
352/*
353 * Start writeback of `nr_pages' pages. If `nr_pages' is zero, write back
354 * the whole world. Returns 0 if a pdflush thread was dispatched. Returns
355 * -1 if all pdflush threads were busy.
356 */
357int wakeup_bdflush(long nr_pages)
358{
359 if (nr_pages == 0) {
360 struct writeback_state wbs;
361
362 get_writeback_state(&wbs);
363 nr_pages = wbs.nr_dirty + wbs.nr_unstable;
364 }
365 return pdflush_operation(background_writeout, nr_pages);
366}
367
368static void wb_timer_fn(unsigned long unused);
369static void laptop_timer_fn(unsigned long unused);
370
371static struct timer_list wb_timer =
372 TIMER_INITIALIZER(wb_timer_fn, 0, 0);
373static struct timer_list laptop_mode_wb_timer =
374 TIMER_INITIALIZER(laptop_timer_fn, 0, 0);
375
376/*
377 * Periodic writeback of "old" data.
378 *
379 * Define "old": the first time one of an inode's pages is dirtied, we mark the
380 * dirtying-time in the inode's address_space. So this periodic writeback code
381 * just walks the superblock inode list, writing back any inodes which are
382 * older than a specific point in time.
383 *
384 * Try to run once per dirty_writeback_centisecs. But if a writeback event
385 * takes longer than a dirty_writeback_centisecs interval, then leave a
386 * one-second gap.
387 *
388 * older_than_this takes precedence over nr_to_write. So we'll only write back
389 * all dirty pages if they are all attached to "old" mappings.
390 */
391static void wb_kupdate(unsigned long arg)
392{
393 unsigned long oldest_jif;
394 unsigned long start_jif;
395 unsigned long next_jif;
396 long nr_to_write;
397 struct writeback_state wbs;
398 struct writeback_control wbc = {
399 .bdi = NULL,
400 .sync_mode = WB_SYNC_NONE,
401 .older_than_this = &oldest_jif,
402 .nr_to_write = 0,
403 .nonblocking = 1,
404 .for_kupdate = 1,
405 };
406
407 sync_supers();
408
409 get_writeback_state(&wbs);
410 oldest_jif = jiffies - (dirty_expire_centisecs * HZ) / 100;
411 start_jif = jiffies;
412 next_jif = start_jif + (dirty_writeback_centisecs * HZ) / 100;
413 nr_to_write = wbs.nr_dirty + wbs.nr_unstable +
414 (inodes_stat.nr_inodes - inodes_stat.nr_unused);
415 while (nr_to_write > 0) {
416 wbc.encountered_congestion = 0;
417 wbc.nr_to_write = MAX_WRITEBACK_PAGES;
418 writeback_inodes(&wbc);
419 if (wbc.nr_to_write > 0) {
420 if (wbc.encountered_congestion)
421 blk_congestion_wait(WRITE, HZ/10);
422 else
423 break; /* All the old data is written */
424 }
425 nr_to_write -= MAX_WRITEBACK_PAGES - wbc.nr_to_write;
426 }
427 if (time_before(next_jif, jiffies + HZ))
428 next_jif = jiffies + HZ;
429 if (dirty_writeback_centisecs)
430 mod_timer(&wb_timer, next_jif);
431}
432
433/*
434 * sysctl handler for /proc/sys/vm/dirty_writeback_centisecs
435 */
436int dirty_writeback_centisecs_handler(ctl_table *table, int write,
437 struct file *file, void __user *buffer, size_t *length, loff_t *ppos)
438{
439 proc_dointvec(table, write, file, buffer, length, ppos);
440 if (dirty_writeback_centisecs) {
441 mod_timer(&wb_timer,
442 jiffies + (dirty_writeback_centisecs * HZ) / 100);
443 } else {
444 del_timer(&wb_timer);
445 }
446 return 0;
447}
448
449static void wb_timer_fn(unsigned long unused)
450{
451 if (pdflush_operation(wb_kupdate, 0) < 0)
452 mod_timer(&wb_timer, jiffies + HZ); /* delay 1 second */
453}
454
455static void laptop_flush(unsigned long unused)
456{
457 sys_sync();
458}
459
460static void laptop_timer_fn(unsigned long unused)
461{
462 pdflush_operation(laptop_flush, 0);
463}
464
465/*
466 * We've spun up the disk and we're in laptop mode: schedule writeback
467 * of all dirty data a few seconds from now. If the flush is already scheduled
468 * then push it back - the user is still using the disk.
469 */
470void laptop_io_completion(void)
471{
472 mod_timer(&laptop_mode_wb_timer, jiffies + laptop_mode * HZ);
473}
474
475/*
476 * We're in laptop mode and we've just synced. The sync's writes will have
477 * caused another writeback to be scheduled by laptop_io_completion.
478 * Nothing needs to be written back anymore, so we unschedule the writeback.
479 */
480void laptop_sync_completion(void)
481{
482 del_timer(&laptop_mode_wb_timer);
483}
484
485/*
486 * If ratelimit_pages is too high then we can get into dirty-data overload
487 * if a large number of processes all perform writes at the same time.
488 * If it is too low then SMP machines will call the (expensive)
489 * get_writeback_state too often.
490 *
491 * Here we set ratelimit_pages to a level which ensures that when all CPUs are
492 * dirtying in parallel, we cannot go more than 3% (1/32) over the dirty memory
493 * thresholds before writeback cuts in.
494 *
495 * But the limit should not be set too high. Because it also controls the
496 * amount of memory which the balance_dirty_pages() caller has to write back.
497 * If this is too large then the caller will block on the IO queue all the
498 * time. So limit it to four megabytes - the balance_dirty_pages() caller
499 * will write six megabyte chunks, max.
500 */
501
502static void set_ratelimit(void)
503{
504 ratelimit_pages = total_pages / (num_online_cpus() * 32);
505 if (ratelimit_pages < 16)
506 ratelimit_pages = 16;
507 if (ratelimit_pages * PAGE_CACHE_SIZE > 4096 * 1024)
508 ratelimit_pages = (4096 * 1024) / PAGE_CACHE_SIZE;
509}
510
511static int
512ratelimit_handler(struct notifier_block *self, unsigned long u, void *v)
513{
514 set_ratelimit();
515 return 0;
516}
517
518static struct notifier_block ratelimit_nb = {
519 .notifier_call = ratelimit_handler,
520 .next = NULL,
521};
522
523/*
524 * If the machine has a large highmem:lowmem ratio then scale back the default
525 * dirty memory thresholds: allowing too much dirty highmem pins an excessive
526 * number of buffer_heads.
527 */
528void __init page_writeback_init(void)
529{
530 long buffer_pages = nr_free_buffer_pages();
531 long correction;
532
533 total_pages = nr_free_pagecache_pages();
534
535 correction = (100 * 4 * buffer_pages) / total_pages;
536
537 if (correction < 100) {
538 dirty_background_ratio *= correction;
539 dirty_background_ratio /= 100;
540 vm_dirty_ratio *= correction;
541 vm_dirty_ratio /= 100;
542
543 if (dirty_background_ratio <= 0)
544 dirty_background_ratio = 1;
545 if (vm_dirty_ratio <= 0)
546 vm_dirty_ratio = 1;
547 }
548 mod_timer(&wb_timer, jiffies + (dirty_writeback_centisecs * HZ) / 100);
549 set_ratelimit();
550 register_cpu_notifier(&ratelimit_nb);
551}
552
553int do_writepages(struct address_space *mapping, struct writeback_control *wbc)
554{
555 if (wbc->nr_to_write <= 0)
556 return 0;
557 if (mapping->a_ops->writepages)
558 return mapping->a_ops->writepages(mapping, wbc);
559 return generic_writepages(mapping, wbc);
560}
561
562/**
563 * write_one_page - write out a single page and optionally wait on I/O
564 *
565 * @page - the page to write
566 * @wait - if true, wait on writeout
567 *
568 * The page must be locked by the caller and will be unlocked upon return.
569 *
570 * write_one_page() returns a negative error code if I/O failed.
571 */
572int write_one_page(struct page *page, int wait)
573{
574 struct address_space *mapping = page->mapping;
575 int ret = 0;
576 struct writeback_control wbc = {
577 .sync_mode = WB_SYNC_ALL,
578 .nr_to_write = 1,
579 };
580
581 BUG_ON(!PageLocked(page));
582
583 if (wait)
584 wait_on_page_writeback(page);
585
586 if (clear_page_dirty_for_io(page)) {
587 page_cache_get(page);
588 ret = mapping->a_ops->writepage(page, &wbc);
589 if (ret == 0 && wait) {
590 wait_on_page_writeback(page);
591 if (PageError(page))
592 ret = -EIO;
593 }
594 page_cache_release(page);
595 } else {
596 unlock_page(page);
597 }
598 return ret;
599}
600EXPORT_SYMBOL(write_one_page);
601
602/*
603 * For address_spaces which do not use buffers. Just tag the page as dirty in
604 * its radix tree.
605 *
606 * This is also used when a single buffer is being dirtied: we want to set the
607 * page dirty in that case, but not all the buffers. This is a "bottom-up"
608 * dirtying, whereas __set_page_dirty_buffers() is a "top-down" dirtying.
609 *
610 * Most callers have locked the page, which pins the address_space in memory.
611 * But zap_pte_range() does not lock the page, however in that case the
612 * mapping is pinned by the vma's ->vm_file reference.
613 *
614 * We take care to handle the case where the page was truncated from the
615 * mapping by re-checking page_mapping() insode tree_lock.
616 */
617int __set_page_dirty_nobuffers(struct page *page)
618{
619 int ret = 0;
620
621 if (!TestSetPageDirty(page)) {
622 struct address_space *mapping = page_mapping(page);
623 struct address_space *mapping2;
624
625 if (mapping) {
626 write_lock_irq(&mapping->tree_lock);
627 mapping2 = page_mapping(page);
628 if (mapping2) { /* Race with truncate? */
629 BUG_ON(mapping2 != mapping);
630 if (mapping_cap_account_dirty(mapping))
631 inc_page_state(nr_dirty);
632 radix_tree_tag_set(&mapping->page_tree,
633 page_index(page), PAGECACHE_TAG_DIRTY);
634 }
635 write_unlock_irq(&mapping->tree_lock);
636 if (mapping->host) {
637 /* !PageAnon && !swapper_space */
638 __mark_inode_dirty(mapping->host,
639 I_DIRTY_PAGES);
640 }
641 }
642 }
643 return ret;
644}
645EXPORT_SYMBOL(__set_page_dirty_nobuffers);
646
647/*
648 * When a writepage implementation decides that it doesn't want to write this
649 * page for some reason, it should redirty the locked page via
650 * redirty_page_for_writepage() and it should then unlock the page and return 0
651 */
652int redirty_page_for_writepage(struct writeback_control *wbc, struct page *page)
653{
654 wbc->pages_skipped++;
655 return __set_page_dirty_nobuffers(page);
656}
657EXPORT_SYMBOL(redirty_page_for_writepage);
658
659/*
660 * If the mapping doesn't provide a set_page_dirty a_op, then
661 * just fall through and assume that it wants buffer_heads.
662 */
663int fastcall set_page_dirty(struct page *page)
664{
665 struct address_space *mapping = page_mapping(page);
666
667 if (likely(mapping)) {
668 int (*spd)(struct page *) = mapping->a_ops->set_page_dirty;
669 if (spd)
670 return (*spd)(page);
671 return __set_page_dirty_buffers(page);
672 }
673 if (!PageDirty(page))
674 SetPageDirty(page);
675 return 0;
676}
677EXPORT_SYMBOL(set_page_dirty);
678
679/*
680 * set_page_dirty() is racy if the caller has no reference against
681 * page->mapping->host, and if the page is unlocked. This is because another
682 * CPU could truncate the page off the mapping and then free the mapping.
683 *
684 * Usually, the page _is_ locked, or the caller is a user-space process which
685 * holds a reference on the inode by having an open file.
686 *
687 * In other cases, the page should be locked before running set_page_dirty().
688 */
689int set_page_dirty_lock(struct page *page)
690{
691 int ret;
692
693 lock_page(page);
694 ret = set_page_dirty(page);
695 unlock_page(page);
696 return ret;
697}
698EXPORT_SYMBOL(set_page_dirty_lock);
699
700/*
701 * Clear a page's dirty flag, while caring for dirty memory accounting.
702 * Returns true if the page was previously dirty.
703 */
704int test_clear_page_dirty(struct page *page)
705{
706 struct address_space *mapping = page_mapping(page);
707 unsigned long flags;
708
709 if (mapping) {
710 write_lock_irqsave(&mapping->tree_lock, flags);
711 if (TestClearPageDirty(page)) {
712 radix_tree_tag_clear(&mapping->page_tree,
713 page_index(page),
714 PAGECACHE_TAG_DIRTY);
715 write_unlock_irqrestore(&mapping->tree_lock, flags);
716 if (mapping_cap_account_dirty(mapping))
717 dec_page_state(nr_dirty);
718 return 1;
719 }
720 write_unlock_irqrestore(&mapping->tree_lock, flags);
721 return 0;
722 }
723 return TestClearPageDirty(page);
724}
725EXPORT_SYMBOL(test_clear_page_dirty);
726
727/*
728 * Clear a page's dirty flag, while caring for dirty memory accounting.
729 * Returns true if the page was previously dirty.
730 *
731 * This is for preparing to put the page under writeout. We leave the page
732 * tagged as dirty in the radix tree so that a concurrent write-for-sync
733 * can discover it via a PAGECACHE_TAG_DIRTY walk. The ->writepage
734 * implementation will run either set_page_writeback() or set_page_dirty(),
735 * at which stage we bring the page's dirty flag and radix-tree dirty tag
736 * back into sync.
737 *
738 * This incoherency between the page's dirty flag and radix-tree tag is
739 * unfortunate, but it only exists while the page is locked.
740 */
741int clear_page_dirty_for_io(struct page *page)
742{
743 struct address_space *mapping = page_mapping(page);
744
745 if (mapping) {
746 if (TestClearPageDirty(page)) {
747 if (mapping_cap_account_dirty(mapping))
748 dec_page_state(nr_dirty);
749 return 1;
750 }
751 return 0;
752 }
753 return TestClearPageDirty(page);
754}
755EXPORT_SYMBOL(clear_page_dirty_for_io);
756
757int test_clear_page_writeback(struct page *page)
758{
759 struct address_space *mapping = page_mapping(page);
760 int ret;
761
762 if (mapping) {
763 unsigned long flags;
764
765 write_lock_irqsave(&mapping->tree_lock, flags);
766 ret = TestClearPageWriteback(page);
767 if (ret)
768 radix_tree_tag_clear(&mapping->page_tree,
769 page_index(page),
770 PAGECACHE_TAG_WRITEBACK);
771 write_unlock_irqrestore(&mapping->tree_lock, flags);
772 } else {
773 ret = TestClearPageWriteback(page);
774 }
775 return ret;
776}
777
778int test_set_page_writeback(struct page *page)
779{
780 struct address_space *mapping = page_mapping(page);
781 int ret;
782
783 if (mapping) {
784 unsigned long flags;
785
786 write_lock_irqsave(&mapping->tree_lock, flags);
787 ret = TestSetPageWriteback(page);
788 if (!ret)
789 radix_tree_tag_set(&mapping->page_tree,
790 page_index(page),
791 PAGECACHE_TAG_WRITEBACK);
792 if (!PageDirty(page))
793 radix_tree_tag_clear(&mapping->page_tree,
794 page_index(page),
795 PAGECACHE_TAG_DIRTY);
796 write_unlock_irqrestore(&mapping->tree_lock, flags);
797 } else {
798 ret = TestSetPageWriteback(page);
799 }
800 return ret;
801
802}
803EXPORT_SYMBOL(test_set_page_writeback);
804
805/*
806 * Return true if any of the pages in the mapping are marged with the
807 * passed tag.
808 */
809int mapping_tagged(struct address_space *mapping, int tag)
810{
811 unsigned long flags;
812 int ret;
813
814 read_lock_irqsave(&mapping->tree_lock, flags);
815 ret = radix_tree_tagged(&mapping->page_tree, tag);
816 read_unlock_irqrestore(&mapping->tree_lock, flags);
817 return ret;
818}
819EXPORT_SYMBOL(mapping_tagged);