blob: c1052ee79f01096d6dfe23f8d1021c80216b3ad5 [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. */
Andrew Mortone236a162006-01-18 17:42:26 -080049static int dirty_exceeded __cacheline_aligned_in_smp; /* Dirty mem may be over limit */
Linus Torvalds1da177e2005-04-16 15:20:36 -070050
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 */
Bart Samwelf6ef9432006-03-24 03:15:48 -080078int dirty_writeback_interval = 5 * HZ;
Linus Torvalds1da177e2005-04-16 15:20:36 -070079
80/*
81 * The longest number of centiseconds for which data is allowed to remain dirty
82 */
Bart Samwelf6ef9432006-03-24 03:15:48 -080083int dirty_expire_interval = 30 * HZ;
Linus Torvalds1da177e2005-04-16 15:20:36 -070084
85/*
86 * Flag that makes the machine dump writes/reads and block dirtyings.
87 */
88int block_dump;
89
90/*
Bart Samweled5b43f2006-03-24 03:15:49 -080091 * Flag that puts the machine in "laptop mode". Doubles as a timeout in jiffies:
92 * a full sync is triggered after this time elapses without any disk activity.
Linus Torvalds1da177e2005-04-16 15:20:36 -070093 */
94int laptop_mode;
95
96EXPORT_SYMBOL(laptop_mode);
97
98/* End of sysctl-exported parameters */
99
100
101static void background_writeout(unsigned long _min_pages);
102
103struct writeback_state
104{
105 unsigned long nr_dirty;
106 unsigned long nr_unstable;
107 unsigned long nr_mapped;
108 unsigned long nr_writeback;
109};
110
111static void get_writeback_state(struct writeback_state *wbs)
112{
113 wbs->nr_dirty = read_page_state(nr_dirty);
114 wbs->nr_unstable = read_page_state(nr_unstable);
115 wbs->nr_mapped = read_page_state(nr_mapped);
116 wbs->nr_writeback = read_page_state(nr_writeback);
117}
118
119/*
120 * Work out the current dirty-memory clamping and background writeout
121 * thresholds.
122 *
123 * The main aim here is to lower them aggressively if there is a lot of mapped
124 * memory around. To avoid stressing page reclaim with lots of unreclaimable
125 * pages. It is better to clamp down on writers than to start swapping, and
126 * performing lots of scanning.
127 *
128 * We only allow 1/2 of the currently-unmapped memory to be dirtied.
129 *
130 * We don't permit the clamping level to fall below 5% - that is getting rather
131 * excessive.
132 *
133 * We make sure that the background writeout level is below the adjusted
134 * clamping level.
135 */
136static void
137get_dirty_limits(struct writeback_state *wbs, long *pbackground, long *pdirty,
138 struct address_space *mapping)
139{
140 int background_ratio; /* Percentages */
141 int dirty_ratio;
142 int unmapped_ratio;
143 long background;
144 long dirty;
145 unsigned long available_memory = total_pages;
146 struct task_struct *tsk;
147
148 get_writeback_state(wbs);
149
150#ifdef CONFIG_HIGHMEM
151 /*
152 * If this mapping can only allocate from low memory,
153 * we exclude high memory from our count.
154 */
155 if (mapping && !(mapping_gfp_mask(mapping) & __GFP_HIGHMEM))
156 available_memory -= totalhigh_pages;
157#endif
158
159
160 unmapped_ratio = 100 - (wbs->nr_mapped * 100) / total_pages;
161
162 dirty_ratio = vm_dirty_ratio;
163 if (dirty_ratio > unmapped_ratio / 2)
164 dirty_ratio = unmapped_ratio / 2;
165
166 if (dirty_ratio < 5)
167 dirty_ratio = 5;
168
169 background_ratio = dirty_background_ratio;
170 if (background_ratio >= dirty_ratio)
171 background_ratio = dirty_ratio / 2;
172
173 background = (background_ratio * available_memory) / 100;
174 dirty = (dirty_ratio * available_memory) / 100;
175 tsk = current;
176 if (tsk->flags & PF_LESS_THROTTLE || rt_task(tsk)) {
177 background += background / 4;
178 dirty += dirty / 4;
179 }
180 *pbackground = background;
181 *pdirty = dirty;
182}
183
184/*
185 * balance_dirty_pages() must be called by processes which are generating dirty
186 * data. It looks at the number of dirty pages in the machine and will force
187 * the caller to perform writeback if the system is over `vm_dirty_ratio'.
188 * If we're over `background_thresh' then pdflush is woken to perform some
189 * writeout.
190 */
191static void balance_dirty_pages(struct address_space *mapping)
192{
193 struct writeback_state wbs;
194 long nr_reclaimable;
195 long background_thresh;
196 long dirty_thresh;
197 unsigned long pages_written = 0;
198 unsigned long write_chunk = sync_writeback_pages();
199
200 struct backing_dev_info *bdi = mapping->backing_dev_info;
201
202 for (;;) {
203 struct writeback_control wbc = {
204 .bdi = bdi,
205 .sync_mode = WB_SYNC_NONE,
206 .older_than_this = NULL,
207 .nr_to_write = write_chunk,
208 };
209
210 get_dirty_limits(&wbs, &background_thresh,
211 &dirty_thresh, mapping);
212 nr_reclaimable = wbs.nr_dirty + wbs.nr_unstable;
213 if (nr_reclaimable + wbs.nr_writeback <= dirty_thresh)
214 break;
215
Andrew Mortone236a162006-01-18 17:42:26 -0800216 if (!dirty_exceeded)
217 dirty_exceeded = 1;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700218
219 /* Note: nr_reclaimable denotes nr_dirty + nr_unstable.
220 * Unstable writes are a feature of certain networked
221 * filesystems (i.e. NFS) in which data may have been
222 * written to the server's write cache, but has not yet
223 * been flushed to permanent storage.
224 */
225 if (nr_reclaimable) {
226 writeback_inodes(&wbc);
227 get_dirty_limits(&wbs, &background_thresh,
228 &dirty_thresh, mapping);
229 nr_reclaimable = wbs.nr_dirty + wbs.nr_unstable;
230 if (nr_reclaimable + wbs.nr_writeback <= dirty_thresh)
231 break;
232 pages_written += write_chunk - wbc.nr_to_write;
233 if (pages_written >= write_chunk)
234 break; /* We've done our duty */
235 }
236 blk_congestion_wait(WRITE, HZ/10);
237 }
238
Andrew Mortone236a162006-01-18 17:42:26 -0800239 if (nr_reclaimable + wbs.nr_writeback <= dirty_thresh && dirty_exceeded)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700240 dirty_exceeded = 0;
241
242 if (writeback_in_progress(bdi))
243 return; /* pdflush is already working this queue */
244
245 /*
246 * In laptop mode, we wait until hitting the higher threshold before
247 * starting background writeout, and then write out all the way down
248 * to the lower threshold. So slow writers cause minimal disk activity.
249 *
250 * In normal mode, we start background writeout at the lower
251 * background_thresh, to keep the amount of dirty memory low.
252 */
253 if ((laptop_mode && pages_written) ||
254 (!laptop_mode && (nr_reclaimable > background_thresh)))
255 pdflush_operation(background_writeout, 0);
256}
257
258/**
259 * balance_dirty_pages_ratelimited - balance dirty memory state
Martin Waitz67be2dd2005-05-01 08:59:26 -0700260 * @mapping: address_space which was dirtied
Linus Torvalds1da177e2005-04-16 15:20:36 -0700261 *
262 * Processes which are dirtying memory should call in here once for each page
263 * which was newly dirtied. The function will periodically check the system's
264 * dirty state and will initiate writeback if needed.
265 *
266 * On really big machines, get_writeback_state is expensive, so try to avoid
267 * calling it too often (ratelimiting). But once we're over the dirty memory
268 * limit we decrease the ratelimiting by a lot, to prevent individual processes
269 * from overshooting the limit by (ratelimit_pages) each.
270 */
271void balance_dirty_pages_ratelimited(struct address_space *mapping)
272{
273 static DEFINE_PER_CPU(int, ratelimits) = 0;
274 long ratelimit;
275
276 ratelimit = ratelimit_pages;
277 if (dirty_exceeded)
278 ratelimit = 8;
279
280 /*
281 * Check the rate limiting. Also, we do not want to throttle real-time
282 * tasks in balance_dirty_pages(). Period.
283 */
284 if (get_cpu_var(ratelimits)++ >= ratelimit) {
285 __get_cpu_var(ratelimits) = 0;
286 put_cpu_var(ratelimits);
287 balance_dirty_pages(mapping);
288 return;
289 }
290 put_cpu_var(ratelimits);
291}
292EXPORT_SYMBOL(balance_dirty_pages_ratelimited);
293
294void throttle_vm_writeout(void)
295{
296 struct writeback_state wbs;
297 long background_thresh;
298 long dirty_thresh;
299
300 for ( ; ; ) {
301 get_dirty_limits(&wbs, &background_thresh, &dirty_thresh, NULL);
302
303 /*
304 * Boost the allowable dirty threshold a bit for page
305 * allocators so they don't get DoS'ed by heavy writers
306 */
307 dirty_thresh += dirty_thresh / 10; /* wheeee... */
308
309 if (wbs.nr_unstable + wbs.nr_writeback <= dirty_thresh)
310 break;
311 blk_congestion_wait(WRITE, HZ/10);
312 }
313}
314
315
316/*
317 * writeback at least _min_pages, and keep writing until the amount of dirty
318 * memory is less than the background threshold, or until we're all clean.
319 */
320static void background_writeout(unsigned long _min_pages)
321{
322 long min_pages = _min_pages;
323 struct writeback_control wbc = {
324 .bdi = NULL,
325 .sync_mode = WB_SYNC_NONE,
326 .older_than_this = NULL,
327 .nr_to_write = 0,
328 .nonblocking = 1,
329 };
330
331 for ( ; ; ) {
332 struct writeback_state wbs;
333 long background_thresh;
334 long dirty_thresh;
335
336 get_dirty_limits(&wbs, &background_thresh, &dirty_thresh, NULL);
337 if (wbs.nr_dirty + wbs.nr_unstable < background_thresh
338 && min_pages <= 0)
339 break;
340 wbc.encountered_congestion = 0;
341 wbc.nr_to_write = MAX_WRITEBACK_PAGES;
342 wbc.pages_skipped = 0;
343 writeback_inodes(&wbc);
344 min_pages -= MAX_WRITEBACK_PAGES - wbc.nr_to_write;
345 if (wbc.nr_to_write > 0 || wbc.pages_skipped > 0) {
346 /* Wrote less than expected */
347 blk_congestion_wait(WRITE, HZ/10);
348 if (!wbc.encountered_congestion)
349 break;
350 }
351 }
352}
353
354/*
355 * Start writeback of `nr_pages' pages. If `nr_pages' is zero, write back
356 * the whole world. Returns 0 if a pdflush thread was dispatched. Returns
357 * -1 if all pdflush threads were busy.
358 */
Pekka J Enberg687a21c2005-06-28 20:44:55 -0700359int wakeup_pdflush(long nr_pages)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700360{
361 if (nr_pages == 0) {
362 struct writeback_state wbs;
363
364 get_writeback_state(&wbs);
365 nr_pages = wbs.nr_dirty + wbs.nr_unstable;
366 }
367 return pdflush_operation(background_writeout, nr_pages);
368}
369
370static void wb_timer_fn(unsigned long unused);
371static void laptop_timer_fn(unsigned long unused);
372
Ingo Molnar8d06afa2005-09-09 13:10:40 -0700373static DEFINE_TIMER(wb_timer, wb_timer_fn, 0, 0);
374static DEFINE_TIMER(laptop_mode_wb_timer, laptop_timer_fn, 0, 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700375
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 *
Bart Samwelf6ef9432006-03-24 03:15:48 -0800384 * Try to run once per dirty_writeback_interval. But if a writeback event
385 * takes longer than a dirty_writeback_interval interval, then leave a
Linus Torvalds1da177e2005-04-16 15:20:36 -0700386 * 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);
Bart Samwelf6ef9432006-03-24 03:15:48 -0800410 oldest_jif = jiffies - dirty_expire_interval;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700411 start_jif = jiffies;
Bart Samwelf6ef9432006-03-24 03:15:48 -0800412 next_jif = start_jif + dirty_writeback_interval;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700413 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;
Bart Samwelf6ef9432006-03-24 03:15:48 -0800429 if (dirty_writeback_interval)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700430 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{
Bart Samwelf6ef9432006-03-24 03:15:48 -0800439 proc_dointvec_userhz_jiffies(table, write, file, buffer, length, ppos);
440 if (dirty_writeback_interval) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700441 mod_timer(&wb_timer,
Bart Samwelf6ef9432006-03-24 03:15:48 -0800442 jiffies + dirty_writeback_interval);
443 } else {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700444 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{
Bart Samweled5b43f2006-03-24 03:15:49 -0800472 mod_timer(&laptop_mode_wb_timer, jiffies + laptop_mode);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700473}
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 }
Bart Samwelf6ef9432006-03-24 03:15:48 -0800548 mod_timer(&wb_timer, jiffies + dirty_writeback_interval);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700549 set_ratelimit();
550 register_cpu_notifier(&ratelimit_nb);
551}
552
553int do_writepages(struct address_space *mapping, struct writeback_control *wbc)
554{
Andrew Morton22905f72005-11-16 15:07:01 -0800555 int ret;
556
Linus Torvalds1da177e2005-04-16 15:20:36 -0700557 if (wbc->nr_to_write <= 0)
558 return 0;
Andrew Morton22905f72005-11-16 15:07:01 -0800559 wbc->for_writepages = 1;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700560 if (mapping->a_ops->writepages)
Andrew Morton22905f72005-11-16 15:07:01 -0800561 ret = mapping->a_ops->writepages(mapping, wbc);
562 else
563 ret = generic_writepages(mapping, wbc);
564 wbc->for_writepages = 0;
565 return ret;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700566}
567
568/**
569 * write_one_page - write out a single page and optionally wait on I/O
570 *
Martin Waitz67be2dd2005-05-01 08:59:26 -0700571 * @page: the page to write
572 * @wait: if true, wait on writeout
Linus Torvalds1da177e2005-04-16 15:20:36 -0700573 *
574 * The page must be locked by the caller and will be unlocked upon return.
575 *
576 * write_one_page() returns a negative error code if I/O failed.
577 */
578int write_one_page(struct page *page, int wait)
579{
580 struct address_space *mapping = page->mapping;
581 int ret = 0;
582 struct writeback_control wbc = {
583 .sync_mode = WB_SYNC_ALL,
584 .nr_to_write = 1,
585 };
586
587 BUG_ON(!PageLocked(page));
588
589 if (wait)
590 wait_on_page_writeback(page);
591
592 if (clear_page_dirty_for_io(page)) {
593 page_cache_get(page);
594 ret = mapping->a_ops->writepage(page, &wbc);
595 if (ret == 0 && wait) {
596 wait_on_page_writeback(page);
597 if (PageError(page))
598 ret = -EIO;
599 }
600 page_cache_release(page);
601 } else {
602 unlock_page(page);
603 }
604 return ret;
605}
606EXPORT_SYMBOL(write_one_page);
607
608/*
609 * For address_spaces which do not use buffers. Just tag the page as dirty in
610 * its radix tree.
611 *
612 * This is also used when a single buffer is being dirtied: we want to set the
613 * page dirty in that case, but not all the buffers. This is a "bottom-up"
614 * dirtying, whereas __set_page_dirty_buffers() is a "top-down" dirtying.
615 *
616 * Most callers have locked the page, which pins the address_space in memory.
617 * But zap_pte_range() does not lock the page, however in that case the
618 * mapping is pinned by the vma's ->vm_file reference.
619 *
620 * We take care to handle the case where the page was truncated from the
621 * mapping by re-checking page_mapping() insode tree_lock.
622 */
623int __set_page_dirty_nobuffers(struct page *page)
624{
625 int ret = 0;
626
627 if (!TestSetPageDirty(page)) {
628 struct address_space *mapping = page_mapping(page);
629 struct address_space *mapping2;
630
631 if (mapping) {
632 write_lock_irq(&mapping->tree_lock);
633 mapping2 = page_mapping(page);
634 if (mapping2) { /* Race with truncate? */
635 BUG_ON(mapping2 != mapping);
636 if (mapping_cap_account_dirty(mapping))
637 inc_page_state(nr_dirty);
638 radix_tree_tag_set(&mapping->page_tree,
639 page_index(page), PAGECACHE_TAG_DIRTY);
640 }
641 write_unlock_irq(&mapping->tree_lock);
642 if (mapping->host) {
643 /* !PageAnon && !swapper_space */
644 __mark_inode_dirty(mapping->host,
645 I_DIRTY_PAGES);
646 }
647 }
648 }
649 return ret;
650}
651EXPORT_SYMBOL(__set_page_dirty_nobuffers);
652
653/*
654 * When a writepage implementation decides that it doesn't want to write this
655 * page for some reason, it should redirty the locked page via
656 * redirty_page_for_writepage() and it should then unlock the page and return 0
657 */
658int redirty_page_for_writepage(struct writeback_control *wbc, struct page *page)
659{
660 wbc->pages_skipped++;
661 return __set_page_dirty_nobuffers(page);
662}
663EXPORT_SYMBOL(redirty_page_for_writepage);
664
665/*
666 * If the mapping doesn't provide a set_page_dirty a_op, then
667 * just fall through and assume that it wants buffer_heads.
668 */
669int fastcall set_page_dirty(struct page *page)
670{
671 struct address_space *mapping = page_mapping(page);
672
673 if (likely(mapping)) {
674 int (*spd)(struct page *) = mapping->a_ops->set_page_dirty;
675 if (spd)
676 return (*spd)(page);
677 return __set_page_dirty_buffers(page);
678 }
679 if (!PageDirty(page))
680 SetPageDirty(page);
681 return 0;
682}
683EXPORT_SYMBOL(set_page_dirty);
684
685/*
686 * set_page_dirty() is racy if the caller has no reference against
687 * page->mapping->host, and if the page is unlocked. This is because another
688 * CPU could truncate the page off the mapping and then free the mapping.
689 *
690 * Usually, the page _is_ locked, or the caller is a user-space process which
691 * holds a reference on the inode by having an open file.
692 *
693 * In other cases, the page should be locked before running set_page_dirty().
694 */
695int set_page_dirty_lock(struct page *page)
696{
697 int ret;
698
699 lock_page(page);
700 ret = set_page_dirty(page);
701 unlock_page(page);
702 return ret;
703}
704EXPORT_SYMBOL(set_page_dirty_lock);
705
706/*
707 * Clear a page's dirty flag, while caring for dirty memory accounting.
708 * Returns true if the page was previously dirty.
709 */
710int test_clear_page_dirty(struct page *page)
711{
712 struct address_space *mapping = page_mapping(page);
713 unsigned long flags;
714
715 if (mapping) {
716 write_lock_irqsave(&mapping->tree_lock, flags);
717 if (TestClearPageDirty(page)) {
718 radix_tree_tag_clear(&mapping->page_tree,
719 page_index(page),
720 PAGECACHE_TAG_DIRTY);
721 write_unlock_irqrestore(&mapping->tree_lock, flags);
722 if (mapping_cap_account_dirty(mapping))
723 dec_page_state(nr_dirty);
724 return 1;
725 }
726 write_unlock_irqrestore(&mapping->tree_lock, flags);
727 return 0;
728 }
729 return TestClearPageDirty(page);
730}
731EXPORT_SYMBOL(test_clear_page_dirty);
732
733/*
734 * Clear a page's dirty flag, while caring for dirty memory accounting.
735 * Returns true if the page was previously dirty.
736 *
737 * This is for preparing to put the page under writeout. We leave the page
738 * tagged as dirty in the radix tree so that a concurrent write-for-sync
739 * can discover it via a PAGECACHE_TAG_DIRTY walk. The ->writepage
740 * implementation will run either set_page_writeback() or set_page_dirty(),
741 * at which stage we bring the page's dirty flag and radix-tree dirty tag
742 * back into sync.
743 *
744 * This incoherency between the page's dirty flag and radix-tree tag is
745 * unfortunate, but it only exists while the page is locked.
746 */
747int clear_page_dirty_for_io(struct page *page)
748{
749 struct address_space *mapping = page_mapping(page);
750
751 if (mapping) {
752 if (TestClearPageDirty(page)) {
753 if (mapping_cap_account_dirty(mapping))
754 dec_page_state(nr_dirty);
755 return 1;
756 }
757 return 0;
758 }
759 return TestClearPageDirty(page);
760}
Hans Reiser58bb01a2005-11-18 01:10:53 -0800761EXPORT_SYMBOL(clear_page_dirty_for_io);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700762
763int test_clear_page_writeback(struct page *page)
764{
765 struct address_space *mapping = page_mapping(page);
766 int ret;
767
768 if (mapping) {
769 unsigned long flags;
770
771 write_lock_irqsave(&mapping->tree_lock, flags);
772 ret = TestClearPageWriteback(page);
773 if (ret)
774 radix_tree_tag_clear(&mapping->page_tree,
775 page_index(page),
776 PAGECACHE_TAG_WRITEBACK);
777 write_unlock_irqrestore(&mapping->tree_lock, flags);
778 } else {
779 ret = TestClearPageWriteback(page);
780 }
781 return ret;
782}
783
784int test_set_page_writeback(struct page *page)
785{
786 struct address_space *mapping = page_mapping(page);
787 int ret;
788
789 if (mapping) {
790 unsigned long flags;
791
792 write_lock_irqsave(&mapping->tree_lock, flags);
793 ret = TestSetPageWriteback(page);
794 if (!ret)
795 radix_tree_tag_set(&mapping->page_tree,
796 page_index(page),
797 PAGECACHE_TAG_WRITEBACK);
798 if (!PageDirty(page))
799 radix_tree_tag_clear(&mapping->page_tree,
800 page_index(page),
801 PAGECACHE_TAG_DIRTY);
802 write_unlock_irqrestore(&mapping->tree_lock, flags);
803 } else {
804 ret = TestSetPageWriteback(page);
805 }
806 return ret;
807
808}
809EXPORT_SYMBOL(test_set_page_writeback);
810
811/*
812 * Return true if any of the pages in the mapping are marged with the
813 * passed tag.
814 */
815int mapping_tagged(struct address_space *mapping, int tag)
816{
817 unsigned long flags;
818 int ret;
819
820 read_lock_irqsave(&mapping->tree_lock, flags);
821 ret = radix_tree_tagged(&mapping->page_tree, tag);
822 read_unlock_irqrestore(&mapping->tree_lock, flags);
823 return ret;
824}
825EXPORT_SYMBOL(mapping_tagged);