4 * Copyright (C) 2002, Linus Torvalds.
5 * Copyright (C) 2007 Red Hat, Inc., Peter Zijlstra <pzijlstr@redhat.com>
7 * Contains functions related to writing back dirty pages at the
10 * 10Apr2002 akpm@zip.com.au
14 #include <linux/kernel.h>
15 #include <linux/module.h>
16 #include <linux/spinlock.h>
19 #include <linux/swap.h>
20 #include <linux/slab.h>
21 #include <linux/pagemap.h>
22 #include <linux/writeback.h>
23 #include <linux/init.h>
24 #include <linux/backing-dev.h>
25 #include <linux/task_io_accounting_ops.h>
26 #include <linux/blkdev.h>
27 #include <linux/mpage.h>
28 #include <linux/rmap.h>
29 #include <linux/percpu.h>
30 #include <linux/notifier.h>
31 #include <linux/smp.h>
32 #include <linux/sysctl.h>
33 #include <linux/cpu.h>
34 #include <linux/syscalls.h>
35 #include <linux/buffer_head.h>
36 #include <linux/pagevec.h>
39 * The maximum number of pages to writeout in a single bdflush/kupdate
40 * operation. We do this so we don't hold I_SYNC against an inode for
41 * enormous amounts of time, which would block a userspace task which has
42 * been forced to throttle against that inode. Also, the code reevaluates
43 * the dirty each time it has written this many pages.
45 #define MAX_WRITEBACK_PAGES 1024
48 * After a CPU has dirtied this many pages, balance_dirty_pages_ratelimited
49 * will look to see if it needs to force writeback or throttling.
51 static long ratelimit_pages = 32;
54 * When balance_dirty_pages decides that the caller needs to perform some
55 * non-background writeback, this is how many pages it will attempt to write.
56 * It should be somewhat larger than RATELIMIT_PAGES to ensure that reasonably
57 * large amounts of I/O are submitted.
59 static inline long sync_writeback_pages(void)
61 return ratelimit_pages + ratelimit_pages / 2;
64 /* The following parameters are exported via /proc/sys/vm */
67 * Start background writeback (via pdflush) at this percentage
69 int dirty_background_ratio = 5;
72 * free highmem will not be subtracted from the total free memory
73 * for calculating free ratios if vm_highmem_is_dirtyable is true
75 int vm_highmem_is_dirtyable;
78 * The generator of dirty data starts writeback at this percentage
80 int vm_dirty_ratio = 10;
83 * The interval between `kupdate'-style writebacks, in jiffies
85 int dirty_writeback_interval = 5 * HZ;
88 * The longest number of jiffies for which data is allowed to remain dirty
90 int dirty_expire_interval = 30 * HZ;
93 * Flag that makes the machine dump writes/reads and block dirtyings.
98 * Flag that puts the machine in "laptop mode". Doubles as a timeout in jiffies:
99 * a full sync is triggered after this time elapses without any disk activity.
103 EXPORT_SYMBOL(laptop_mode);
105 /* End of sysctl-exported parameters */
108 static void background_writeout(unsigned long _min_pages);
111 * Scale the writeback cache size proportional to the relative writeout speeds.
113 * We do this by keeping a floating proportion between BDIs, based on page
114 * writeback completions [end_page_writeback()]. Those devices that write out
115 * pages fastest will get the larger share, while the slower will get a smaller
118 * We use page writeout completions because we are interested in getting rid of
119 * dirty pages. Having them written out is the primary goal.
121 * We introduce a concept of time, a period over which we measure these events,
122 * because demand can/will vary over time. The length of this period itself is
123 * measured in page writeback completions.
126 static struct prop_descriptor vm_completions;
127 static struct prop_descriptor vm_dirties;
129 static unsigned long determine_dirtyable_memory(void);
132 * couple the period to the dirty_ratio:
134 * period/2 ~ roundup_pow_of_two(dirty limit)
136 static int calc_period_shift(void)
138 unsigned long dirty_total;
140 dirty_total = (vm_dirty_ratio * determine_dirtyable_memory()) / 100;
141 return 2 + ilog2(dirty_total - 1);
145 * update the period when the dirty ratio changes.
147 int dirty_ratio_handler(struct ctl_table *table, int write,
148 struct file *filp, void __user *buffer, size_t *lenp,
151 int old_ratio = vm_dirty_ratio;
152 int ret = proc_dointvec_minmax(table, write, filp, buffer, lenp, ppos);
153 if (ret == 0 && write && vm_dirty_ratio != old_ratio) {
154 int shift = calc_period_shift();
155 prop_change_shift(&vm_completions, shift);
156 prop_change_shift(&vm_dirties, shift);
162 * Increment the BDI's writeout completion count and the global writeout
163 * completion count. Called from test_clear_page_writeback().
165 static inline void __bdi_writeout_inc(struct backing_dev_info *bdi)
167 __prop_inc_percpu(&vm_completions, &bdi->completions);
170 static inline void task_dirty_inc(struct task_struct *tsk)
172 prop_inc_single(&vm_dirties, &tsk->dirties);
176 * Obtain an accurate fraction of the BDI's portion.
178 static void bdi_writeout_fraction(struct backing_dev_info *bdi,
179 long *numerator, long *denominator)
181 if (bdi_cap_writeback_dirty(bdi)) {
182 prop_fraction_percpu(&vm_completions, &bdi->completions,
183 numerator, denominator);
191 * Clip the earned share of dirty pages to that which is actually available.
192 * This avoids exceeding the total dirty_limit when the floating averages
193 * fluctuate too quickly.
196 clip_bdi_dirty_limit(struct backing_dev_info *bdi, long dirty, long *pbdi_dirty)
200 avail_dirty = dirty -
201 (global_page_state(NR_FILE_DIRTY) +
202 global_page_state(NR_WRITEBACK) +
203 global_page_state(NR_UNSTABLE_NFS));
208 avail_dirty += bdi_stat(bdi, BDI_RECLAIMABLE) +
209 bdi_stat(bdi, BDI_WRITEBACK);
211 *pbdi_dirty = min(*pbdi_dirty, avail_dirty);
214 static inline void task_dirties_fraction(struct task_struct *tsk,
215 long *numerator, long *denominator)
217 prop_fraction_single(&vm_dirties, &tsk->dirties,
218 numerator, denominator);
222 * scale the dirty limit
224 * task specific dirty limit:
226 * dirty -= (dirty/8) * p_{t}
228 static void task_dirty_limit(struct task_struct *tsk, long *pdirty)
230 long numerator, denominator;
231 long dirty = *pdirty;
232 u64 inv = dirty >> 3;
234 task_dirties_fraction(tsk, &numerator, &denominator);
236 do_div(inv, denominator);
239 if (dirty < *pdirty/2)
248 static DEFINE_SPINLOCK(bdi_lock);
249 static unsigned int bdi_min_ratio;
251 int bdi_set_min_ratio(struct backing_dev_info *bdi, unsigned int min_ratio)
256 spin_lock_irqsave(&bdi_lock, flags);
257 min_ratio -= bdi->min_ratio;
258 if (bdi_min_ratio + min_ratio < 100) {
259 bdi_min_ratio += min_ratio;
260 bdi->min_ratio += min_ratio;
263 spin_unlock_irqrestore(&bdi_lock, flags);
269 * Work out the current dirty-memory clamping and background writeout
272 * The main aim here is to lower them aggressively if there is a lot of mapped
273 * memory around. To avoid stressing page reclaim with lots of unreclaimable
274 * pages. It is better to clamp down on writers than to start swapping, and
275 * performing lots of scanning.
277 * We only allow 1/2 of the currently-unmapped memory to be dirtied.
279 * We don't permit the clamping level to fall below 5% - that is getting rather
282 * We make sure that the background writeout level is below the adjusted
286 static unsigned long highmem_dirtyable_memory(unsigned long total)
288 #ifdef CONFIG_HIGHMEM
292 for_each_node_state(node, N_HIGH_MEMORY) {
294 &NODE_DATA(node)->node_zones[ZONE_HIGHMEM];
296 x += zone_page_state(z, NR_FREE_PAGES)
297 + zone_page_state(z, NR_INACTIVE)
298 + zone_page_state(z, NR_ACTIVE);
301 * Make sure that the number of highmem pages is never larger
302 * than the number of the total dirtyable memory. This can only
303 * occur in very strange VM situations but we want to make sure
304 * that this does not occur.
306 return min(x, total);
312 static unsigned long determine_dirtyable_memory(void)
316 x = global_page_state(NR_FREE_PAGES)
317 + global_page_state(NR_INACTIVE)
318 + global_page_state(NR_ACTIVE);
320 if (!vm_highmem_is_dirtyable)
321 x -= highmem_dirtyable_memory(x);
323 return x + 1; /* Ensure that we never return 0 */
327 get_dirty_limits(long *pbackground, long *pdirty, long *pbdi_dirty,
328 struct backing_dev_info *bdi)
330 int background_ratio; /* Percentages */
334 unsigned long available_memory = determine_dirtyable_memory();
335 struct task_struct *tsk;
337 dirty_ratio = vm_dirty_ratio;
341 background_ratio = dirty_background_ratio;
342 if (background_ratio >= dirty_ratio)
343 background_ratio = dirty_ratio / 2;
345 background = (background_ratio * available_memory) / 100;
346 dirty = (dirty_ratio * available_memory) / 100;
348 if (tsk->flags & PF_LESS_THROTTLE || rt_task(tsk)) {
349 background += background / 4;
352 *pbackground = background;
357 long numerator, denominator;
360 * Calculate this BDI's share of the dirty ratio.
362 bdi_writeout_fraction(bdi, &numerator, &denominator);
364 bdi_dirty = (dirty * (100 - bdi_min_ratio)) / 100;
365 bdi_dirty *= numerator;
366 do_div(bdi_dirty, denominator);
367 bdi_dirty += (dirty * bdi->min_ratio) / 100;
369 *pbdi_dirty = bdi_dirty;
370 clip_bdi_dirty_limit(bdi, dirty, pbdi_dirty);
371 task_dirty_limit(current, pbdi_dirty);
376 * balance_dirty_pages() must be called by processes which are generating dirty
377 * data. It looks at the number of dirty pages in the machine and will force
378 * the caller to perform writeback if the system is over `vm_dirty_ratio'.
379 * If we're over `background_thresh' then pdflush is woken to perform some
382 static void balance_dirty_pages(struct address_space *mapping)
384 long nr_reclaimable, bdi_nr_reclaimable;
385 long nr_writeback, bdi_nr_writeback;
386 long background_thresh;
389 unsigned long pages_written = 0;
390 unsigned long write_chunk = sync_writeback_pages();
392 struct backing_dev_info *bdi = mapping->backing_dev_info;
395 struct writeback_control wbc = {
397 .sync_mode = WB_SYNC_NONE,
398 .older_than_this = NULL,
399 .nr_to_write = write_chunk,
403 get_dirty_limits(&background_thresh, &dirty_thresh,
406 nr_reclaimable = global_page_state(NR_FILE_DIRTY) +
407 global_page_state(NR_UNSTABLE_NFS);
408 nr_writeback = global_page_state(NR_WRITEBACK);
410 bdi_nr_reclaimable = bdi_stat(bdi, BDI_RECLAIMABLE);
411 bdi_nr_writeback = bdi_stat(bdi, BDI_WRITEBACK);
413 if (bdi_nr_reclaimable + bdi_nr_writeback <= bdi_thresh)
417 * Throttle it only when the background writeback cannot
418 * catch-up. This avoids (excessively) small writeouts
419 * when the bdi limits are ramping up.
421 if (nr_reclaimable + nr_writeback <
422 (background_thresh + dirty_thresh) / 2)
425 if (!bdi->dirty_exceeded)
426 bdi->dirty_exceeded = 1;
428 /* Note: nr_reclaimable denotes nr_dirty + nr_unstable.
429 * Unstable writes are a feature of certain networked
430 * filesystems (i.e. NFS) in which data may have been
431 * written to the server's write cache, but has not yet
432 * been flushed to permanent storage.
434 if (bdi_nr_reclaimable) {
435 writeback_inodes(&wbc);
436 pages_written += write_chunk - wbc.nr_to_write;
437 get_dirty_limits(&background_thresh, &dirty_thresh,
442 * In order to avoid the stacked BDI deadlock we need
443 * to ensure we accurately count the 'dirty' pages when
444 * the threshold is low.
446 * Otherwise it would be possible to get thresh+n pages
447 * reported dirty, even though there are thresh-m pages
448 * actually dirty; with m+n sitting in the percpu
451 if (bdi_thresh < 2*bdi_stat_error(bdi)) {
452 bdi_nr_reclaimable = bdi_stat_sum(bdi, BDI_RECLAIMABLE);
453 bdi_nr_writeback = bdi_stat_sum(bdi, BDI_WRITEBACK);
454 } else if (bdi_nr_reclaimable) {
455 bdi_nr_reclaimable = bdi_stat(bdi, BDI_RECLAIMABLE);
456 bdi_nr_writeback = bdi_stat(bdi, BDI_WRITEBACK);
459 if (bdi_nr_reclaimable + bdi_nr_writeback <= bdi_thresh)
461 if (pages_written >= write_chunk)
462 break; /* We've done our duty */
464 congestion_wait(WRITE, HZ/10);
467 if (bdi_nr_reclaimable + bdi_nr_writeback < bdi_thresh &&
469 bdi->dirty_exceeded = 0;
471 if (writeback_in_progress(bdi))
472 return; /* pdflush is already working this queue */
475 * In laptop mode, we wait until hitting the higher threshold before
476 * starting background writeout, and then write out all the way down
477 * to the lower threshold. So slow writers cause minimal disk activity.
479 * In normal mode, we start background writeout at the lower
480 * background_thresh, to keep the amount of dirty memory low.
482 if ((laptop_mode && pages_written) ||
483 (!laptop_mode && (global_page_state(NR_FILE_DIRTY)
484 + global_page_state(NR_UNSTABLE_NFS)
485 > background_thresh)))
486 pdflush_operation(background_writeout, 0);
489 void set_page_dirty_balance(struct page *page, int page_mkwrite)
491 if (set_page_dirty(page) || page_mkwrite) {
492 struct address_space *mapping = page_mapping(page);
495 balance_dirty_pages_ratelimited(mapping);
500 * balance_dirty_pages_ratelimited_nr - balance dirty memory state
501 * @mapping: address_space which was dirtied
502 * @nr_pages_dirtied: number of pages which the caller has just dirtied
504 * Processes which are dirtying memory should call in here once for each page
505 * which was newly dirtied. The function will periodically check the system's
506 * dirty state and will initiate writeback if needed.
508 * On really big machines, get_writeback_state is expensive, so try to avoid
509 * calling it too often (ratelimiting). But once we're over the dirty memory
510 * limit we decrease the ratelimiting by a lot, to prevent individual processes
511 * from overshooting the limit by (ratelimit_pages) each.
513 void balance_dirty_pages_ratelimited_nr(struct address_space *mapping,
514 unsigned long nr_pages_dirtied)
516 static DEFINE_PER_CPU(unsigned long, ratelimits) = 0;
517 unsigned long ratelimit;
520 ratelimit = ratelimit_pages;
521 if (mapping->backing_dev_info->dirty_exceeded)
525 * Check the rate limiting. Also, we do not want to throttle real-time
526 * tasks in balance_dirty_pages(). Period.
529 p = &__get_cpu_var(ratelimits);
530 *p += nr_pages_dirtied;
531 if (unlikely(*p >= ratelimit)) {
534 balance_dirty_pages(mapping);
539 EXPORT_SYMBOL(balance_dirty_pages_ratelimited_nr);
541 void throttle_vm_writeout(gfp_t gfp_mask)
543 long background_thresh;
547 get_dirty_limits(&background_thresh, &dirty_thresh, NULL, NULL);
550 * Boost the allowable dirty threshold a bit for page
551 * allocators so they don't get DoS'ed by heavy writers
553 dirty_thresh += dirty_thresh / 10; /* wheeee... */
555 if (global_page_state(NR_UNSTABLE_NFS) +
556 global_page_state(NR_WRITEBACK) <= dirty_thresh)
558 congestion_wait(WRITE, HZ/10);
561 * The caller might hold locks which can prevent IO completion
562 * or progress in the filesystem. So we cannot just sit here
563 * waiting for IO to complete.
565 if ((gfp_mask & (__GFP_FS|__GFP_IO)) != (__GFP_FS|__GFP_IO))
571 * writeback at least _min_pages, and keep writing until the amount of dirty
572 * memory is less than the background threshold, or until we're all clean.
574 static void background_writeout(unsigned long _min_pages)
576 long min_pages = _min_pages;
577 struct writeback_control wbc = {
579 .sync_mode = WB_SYNC_NONE,
580 .older_than_this = NULL,
587 long background_thresh;
590 get_dirty_limits(&background_thresh, &dirty_thresh, NULL, NULL);
591 if (global_page_state(NR_FILE_DIRTY) +
592 global_page_state(NR_UNSTABLE_NFS) < background_thresh
596 wbc.encountered_congestion = 0;
597 wbc.nr_to_write = MAX_WRITEBACK_PAGES;
598 wbc.pages_skipped = 0;
599 writeback_inodes(&wbc);
600 min_pages -= MAX_WRITEBACK_PAGES - wbc.nr_to_write;
601 if (wbc.nr_to_write > 0 || wbc.pages_skipped > 0) {
602 /* Wrote less than expected */
603 if (wbc.encountered_congestion || wbc.more_io)
604 congestion_wait(WRITE, HZ/10);
612 * Start writeback of `nr_pages' pages. If `nr_pages' is zero, write back
613 * the whole world. Returns 0 if a pdflush thread was dispatched. Returns
614 * -1 if all pdflush threads were busy.
616 int wakeup_pdflush(long nr_pages)
619 nr_pages = global_page_state(NR_FILE_DIRTY) +
620 global_page_state(NR_UNSTABLE_NFS);
621 return pdflush_operation(background_writeout, nr_pages);
624 static void wb_timer_fn(unsigned long unused);
625 static void laptop_timer_fn(unsigned long unused);
627 static DEFINE_TIMER(wb_timer, wb_timer_fn, 0, 0);
628 static DEFINE_TIMER(laptop_mode_wb_timer, laptop_timer_fn, 0, 0);
631 * Periodic writeback of "old" data.
633 * Define "old": the first time one of an inode's pages is dirtied, we mark the
634 * dirtying-time in the inode's address_space. So this periodic writeback code
635 * just walks the superblock inode list, writing back any inodes which are
636 * older than a specific point in time.
638 * Try to run once per dirty_writeback_interval. But if a writeback event
639 * takes longer than a dirty_writeback_interval interval, then leave a
642 * older_than_this takes precedence over nr_to_write. So we'll only write back
643 * all dirty pages if they are all attached to "old" mappings.
645 static void wb_kupdate(unsigned long arg)
647 unsigned long oldest_jif;
648 unsigned long start_jif;
649 unsigned long next_jif;
651 struct writeback_control wbc = {
653 .sync_mode = WB_SYNC_NONE,
654 .older_than_this = &oldest_jif,
663 oldest_jif = jiffies - dirty_expire_interval;
665 next_jif = start_jif + dirty_writeback_interval;
666 nr_to_write = global_page_state(NR_FILE_DIRTY) +
667 global_page_state(NR_UNSTABLE_NFS) +
668 (inodes_stat.nr_inodes - inodes_stat.nr_unused);
669 while (nr_to_write > 0) {
671 wbc.encountered_congestion = 0;
672 wbc.nr_to_write = MAX_WRITEBACK_PAGES;
673 writeback_inodes(&wbc);
674 if (wbc.nr_to_write > 0) {
675 if (wbc.encountered_congestion || wbc.more_io)
676 congestion_wait(WRITE, HZ/10);
678 break; /* All the old data is written */
680 nr_to_write -= MAX_WRITEBACK_PAGES - wbc.nr_to_write;
682 if (time_before(next_jif, jiffies + HZ))
683 next_jif = jiffies + HZ;
684 if (dirty_writeback_interval)
685 mod_timer(&wb_timer, next_jif);
689 * sysctl handler for /proc/sys/vm/dirty_writeback_centisecs
691 int dirty_writeback_centisecs_handler(ctl_table *table, int write,
692 struct file *file, void __user *buffer, size_t *length, loff_t *ppos)
694 proc_dointvec_userhz_jiffies(table, write, file, buffer, length, ppos);
695 if (dirty_writeback_interval)
696 mod_timer(&wb_timer, jiffies + dirty_writeback_interval);
698 del_timer(&wb_timer);
702 static void wb_timer_fn(unsigned long unused)
704 if (pdflush_operation(wb_kupdate, 0) < 0)
705 mod_timer(&wb_timer, jiffies + HZ); /* delay 1 second */
708 static void laptop_flush(unsigned long unused)
713 static void laptop_timer_fn(unsigned long unused)
715 pdflush_operation(laptop_flush, 0);
719 * We've spun up the disk and we're in laptop mode: schedule writeback
720 * of all dirty data a few seconds from now. If the flush is already scheduled
721 * then push it back - the user is still using the disk.
723 void laptop_io_completion(void)
725 mod_timer(&laptop_mode_wb_timer, jiffies + laptop_mode);
729 * We're in laptop mode and we've just synced. The sync's writes will have
730 * caused another writeback to be scheduled by laptop_io_completion.
731 * Nothing needs to be written back anymore, so we unschedule the writeback.
733 void laptop_sync_completion(void)
735 del_timer(&laptop_mode_wb_timer);
739 * If ratelimit_pages is too high then we can get into dirty-data overload
740 * if a large number of processes all perform writes at the same time.
741 * If it is too low then SMP machines will call the (expensive)
742 * get_writeback_state too often.
744 * Here we set ratelimit_pages to a level which ensures that when all CPUs are
745 * dirtying in parallel, we cannot go more than 3% (1/32) over the dirty memory
746 * thresholds before writeback cuts in.
748 * But the limit should not be set too high. Because it also controls the
749 * amount of memory which the balance_dirty_pages() caller has to write back.
750 * If this is too large then the caller will block on the IO queue all the
751 * time. So limit it to four megabytes - the balance_dirty_pages() caller
752 * will write six megabyte chunks, max.
755 void writeback_set_ratelimit(void)
757 ratelimit_pages = vm_total_pages / (num_online_cpus() * 32);
758 if (ratelimit_pages < 16)
759 ratelimit_pages = 16;
760 if (ratelimit_pages * PAGE_CACHE_SIZE > 4096 * 1024)
761 ratelimit_pages = (4096 * 1024) / PAGE_CACHE_SIZE;
765 ratelimit_handler(struct notifier_block *self, unsigned long u, void *v)
767 writeback_set_ratelimit();
771 static struct notifier_block __cpuinitdata ratelimit_nb = {
772 .notifier_call = ratelimit_handler,
777 * Called early on to tune the page writeback dirty limits.
779 * We used to scale dirty pages according to how total memory
780 * related to pages that could be allocated for buffers (by
781 * comparing nr_free_buffer_pages() to vm_total_pages.
783 * However, that was when we used "dirty_ratio" to scale with
784 * all memory, and we don't do that any more. "dirty_ratio"
785 * is now applied to total non-HIGHPAGE memory (by subtracting
786 * totalhigh_pages from vm_total_pages), and as such we can't
787 * get into the old insane situation any more where we had
788 * large amounts of dirty pages compared to a small amount of
789 * non-HIGHMEM memory.
791 * But we might still want to scale the dirty_ratio by how
792 * much memory the box has..
794 void __init page_writeback_init(void)
798 mod_timer(&wb_timer, jiffies + dirty_writeback_interval);
799 writeback_set_ratelimit();
800 register_cpu_notifier(&ratelimit_nb);
802 shift = calc_period_shift();
803 prop_descriptor_init(&vm_completions, shift);
804 prop_descriptor_init(&vm_dirties, shift);
808 * write_cache_pages - walk the list of dirty pages of the given address space and write all of them.
809 * @mapping: address space structure to write
810 * @wbc: subtract the number of written pages from *@wbc->nr_to_write
811 * @writepage: function called for each page
812 * @data: data passed to writepage function
814 * If a page is already under I/O, write_cache_pages() skips it, even
815 * if it's dirty. This is desirable behaviour for memory-cleaning writeback,
816 * but it is INCORRECT for data-integrity system calls such as fsync(). fsync()
817 * and msync() need to guarantee that all the data which was dirty at the time
818 * the call was made get new I/O started against them. If wbc->sync_mode is
819 * WB_SYNC_ALL then we were called for data integrity and we must wait for
820 * existing IO to complete.
822 int write_cache_pages(struct address_space *mapping,
823 struct writeback_control *wbc, writepage_t writepage,
826 struct backing_dev_info *bdi = mapping->backing_dev_info;
832 pgoff_t end; /* Inclusive */
836 if (wbc->nonblocking && bdi_write_congested(bdi)) {
837 wbc->encountered_congestion = 1;
841 pagevec_init(&pvec, 0);
842 if (wbc->range_cyclic) {
843 index = mapping->writeback_index; /* Start from prev offset */
846 index = wbc->range_start >> PAGE_CACHE_SHIFT;
847 end = wbc->range_end >> PAGE_CACHE_SHIFT;
848 if (wbc->range_start == 0 && wbc->range_end == LLONG_MAX)
853 while (!done && (index <= end) &&
854 (nr_pages = pagevec_lookup_tag(&pvec, mapping, &index,
856 min(end - index, (pgoff_t)PAGEVEC_SIZE-1) + 1))) {
860 for (i = 0; i < nr_pages; i++) {
861 struct page *page = pvec.pages[i];
864 * At this point we hold neither mapping->tree_lock nor
865 * lock on the page itself: the page may be truncated or
866 * invalidated (changing page->mapping to NULL), or even
867 * swizzled back from swapper_space to tmpfs file
872 if (unlikely(page->mapping != mapping)) {
877 if (!wbc->range_cyclic && page->index > end) {
883 if (wbc->sync_mode != WB_SYNC_NONE)
884 wait_on_page_writeback(page);
886 if (PageWriteback(page) ||
887 !clear_page_dirty_for_io(page)) {
892 ret = (*writepage)(page, wbc, data);
894 if (unlikely(ret == AOP_WRITEPAGE_ACTIVATE)) {
898 if (ret || (--(wbc->nr_to_write) <= 0))
900 if (wbc->nonblocking && bdi_write_congested(bdi)) {
901 wbc->encountered_congestion = 1;
905 pagevec_release(&pvec);
908 if (!scanned && !done) {
910 * We hit the last page and there is more work to be done: wrap
911 * back to the start of the file
917 if (wbc->range_cyclic || (range_whole && wbc->nr_to_write > 0))
918 mapping->writeback_index = index;
921 EXPORT_SYMBOL(write_cache_pages);
924 * Function used by generic_writepages to call the real writepage
925 * function and set the mapping flags on error
927 static int __writepage(struct page *page, struct writeback_control *wbc,
930 struct address_space *mapping = data;
931 int ret = mapping->a_ops->writepage(page, wbc);
932 mapping_set_error(mapping, ret);
937 * generic_writepages - walk the list of dirty pages of the given address space and writepage() all of them.
938 * @mapping: address space structure to write
939 * @wbc: subtract the number of written pages from *@wbc->nr_to_write
941 * This is a library function, which implements the writepages()
942 * address_space_operation.
944 int generic_writepages(struct address_space *mapping,
945 struct writeback_control *wbc)
947 /* deal with chardevs and other special file */
948 if (!mapping->a_ops->writepage)
951 return write_cache_pages(mapping, wbc, __writepage, mapping);
954 EXPORT_SYMBOL(generic_writepages);
956 int do_writepages(struct address_space *mapping, struct writeback_control *wbc)
960 if (wbc->nr_to_write <= 0)
962 wbc->for_writepages = 1;
963 if (mapping->a_ops->writepages)
964 ret = mapping->a_ops->writepages(mapping, wbc);
966 ret = generic_writepages(mapping, wbc);
967 wbc->for_writepages = 0;
972 * write_one_page - write out a single page and optionally wait on I/O
973 * @page: the page to write
974 * @wait: if true, wait on writeout
976 * The page must be locked by the caller and will be unlocked upon return.
978 * write_one_page() returns a negative error code if I/O failed.
980 int write_one_page(struct page *page, int wait)
982 struct address_space *mapping = page->mapping;
984 struct writeback_control wbc = {
985 .sync_mode = WB_SYNC_ALL,
989 BUG_ON(!PageLocked(page));
992 wait_on_page_writeback(page);
994 if (clear_page_dirty_for_io(page)) {
995 page_cache_get(page);
996 ret = mapping->a_ops->writepage(page, &wbc);
997 if (ret == 0 && wait) {
998 wait_on_page_writeback(page);
1002 page_cache_release(page);
1008 EXPORT_SYMBOL(write_one_page);
1011 * For address_spaces which do not use buffers nor write back.
1013 int __set_page_dirty_no_writeback(struct page *page)
1015 if (!PageDirty(page))
1021 * For address_spaces which do not use buffers. Just tag the page as dirty in
1024 * This is also used when a single buffer is being dirtied: we want to set the
1025 * page dirty in that case, but not all the buffers. This is a "bottom-up"
1026 * dirtying, whereas __set_page_dirty_buffers() is a "top-down" dirtying.
1028 * Most callers have locked the page, which pins the address_space in memory.
1029 * But zap_pte_range() does not lock the page, however in that case the
1030 * mapping is pinned by the vma's ->vm_file reference.
1032 * We take care to handle the case where the page was truncated from the
1033 * mapping by re-checking page_mapping() inside tree_lock.
1035 int __set_page_dirty_nobuffers(struct page *page)
1037 if (!TestSetPageDirty(page)) {
1038 struct address_space *mapping = page_mapping(page);
1039 struct address_space *mapping2;
1044 write_lock_irq(&mapping->tree_lock);
1045 mapping2 = page_mapping(page);
1046 if (mapping2) { /* Race with truncate? */
1047 BUG_ON(mapping2 != mapping);
1048 WARN_ON_ONCE(!PagePrivate(page) && !PageUptodate(page));
1049 if (mapping_cap_account_dirty(mapping)) {
1050 __inc_zone_page_state(page, NR_FILE_DIRTY);
1051 __inc_bdi_stat(mapping->backing_dev_info,
1053 task_io_account_write(PAGE_CACHE_SIZE);
1055 radix_tree_tag_set(&mapping->page_tree,
1056 page_index(page), PAGECACHE_TAG_DIRTY);
1058 write_unlock_irq(&mapping->tree_lock);
1059 if (mapping->host) {
1060 /* !PageAnon && !swapper_space */
1061 __mark_inode_dirty(mapping->host, I_DIRTY_PAGES);
1067 EXPORT_SYMBOL(__set_page_dirty_nobuffers);
1070 * When a writepage implementation decides that it doesn't want to write this
1071 * page for some reason, it should redirty the locked page via
1072 * redirty_page_for_writepage() and it should then unlock the page and return 0
1074 int redirty_page_for_writepage(struct writeback_control *wbc, struct page *page)
1076 wbc->pages_skipped++;
1077 return __set_page_dirty_nobuffers(page);
1079 EXPORT_SYMBOL(redirty_page_for_writepage);
1082 * If the mapping doesn't provide a set_page_dirty a_op, then
1083 * just fall through and assume that it wants buffer_heads.
1085 static int __set_page_dirty(struct page *page)
1087 struct address_space *mapping = page_mapping(page);
1089 if (likely(mapping)) {
1090 int (*spd)(struct page *) = mapping->a_ops->set_page_dirty;
1093 spd = __set_page_dirty_buffers;
1095 return (*spd)(page);
1097 if (!PageDirty(page)) {
1098 if (!TestSetPageDirty(page))
1104 int set_page_dirty(struct page *page)
1106 int ret = __set_page_dirty(page);
1108 task_dirty_inc(current);
1111 EXPORT_SYMBOL(set_page_dirty);
1114 * set_page_dirty() is racy if the caller has no reference against
1115 * page->mapping->host, and if the page is unlocked. This is because another
1116 * CPU could truncate the page off the mapping and then free the mapping.
1118 * Usually, the page _is_ locked, or the caller is a user-space process which
1119 * holds a reference on the inode by having an open file.
1121 * In other cases, the page should be locked before running set_page_dirty().
1123 int set_page_dirty_lock(struct page *page)
1127 lock_page_nosync(page);
1128 ret = set_page_dirty(page);
1132 EXPORT_SYMBOL(set_page_dirty_lock);
1135 * Clear a page's dirty flag, while caring for dirty memory accounting.
1136 * Returns true if the page was previously dirty.
1138 * This is for preparing to put the page under writeout. We leave the page
1139 * tagged as dirty in the radix tree so that a concurrent write-for-sync
1140 * can discover it via a PAGECACHE_TAG_DIRTY walk. The ->writepage
1141 * implementation will run either set_page_writeback() or set_page_dirty(),
1142 * at which stage we bring the page's dirty flag and radix-tree dirty tag
1145 * This incoherency between the page's dirty flag and radix-tree tag is
1146 * unfortunate, but it only exists while the page is locked.
1148 int clear_page_dirty_for_io(struct page *page)
1150 struct address_space *mapping = page_mapping(page);
1152 BUG_ON(!PageLocked(page));
1154 ClearPageReclaim(page);
1155 if (mapping && mapping_cap_account_dirty(mapping)) {
1157 * Yes, Virginia, this is indeed insane.
1159 * We use this sequence to make sure that
1160 * (a) we account for dirty stats properly
1161 * (b) we tell the low-level filesystem to
1162 * mark the whole page dirty if it was
1163 * dirty in a pagetable. Only to then
1164 * (c) clean the page again and return 1 to
1165 * cause the writeback.
1167 * This way we avoid all nasty races with the
1168 * dirty bit in multiple places and clearing
1169 * them concurrently from different threads.
1171 * Note! Normally the "set_page_dirty(page)"
1172 * has no effect on the actual dirty bit - since
1173 * that will already usually be set. But we
1174 * need the side effects, and it can help us
1177 * We basically use the page "master dirty bit"
1178 * as a serialization point for all the different
1179 * threads doing their things.
1181 if (page_mkclean(page))
1182 set_page_dirty(page);
1184 * We carefully synchronise fault handlers against
1185 * installing a dirty pte and marking the page dirty
1186 * at this point. We do this by having them hold the
1187 * page lock at some point after installing their
1188 * pte, but before marking the page dirty.
1189 * Pages are always locked coming in here, so we get
1190 * the desired exclusion. See mm/memory.c:do_wp_page()
1191 * for more comments.
1193 if (TestClearPageDirty(page)) {
1194 dec_zone_page_state(page, NR_FILE_DIRTY);
1195 dec_bdi_stat(mapping->backing_dev_info,
1201 return TestClearPageDirty(page);
1203 EXPORT_SYMBOL(clear_page_dirty_for_io);
1205 int test_clear_page_writeback(struct page *page)
1207 struct address_space *mapping = page_mapping(page);
1211 struct backing_dev_info *bdi = mapping->backing_dev_info;
1212 unsigned long flags;
1214 write_lock_irqsave(&mapping->tree_lock, flags);
1215 ret = TestClearPageWriteback(page);
1217 radix_tree_tag_clear(&mapping->page_tree,
1219 PAGECACHE_TAG_WRITEBACK);
1220 if (bdi_cap_writeback_dirty(bdi)) {
1221 __dec_bdi_stat(bdi, BDI_WRITEBACK);
1222 __bdi_writeout_inc(bdi);
1225 write_unlock_irqrestore(&mapping->tree_lock, flags);
1227 ret = TestClearPageWriteback(page);
1230 dec_zone_page_state(page, NR_WRITEBACK);
1234 int test_set_page_writeback(struct page *page)
1236 struct address_space *mapping = page_mapping(page);
1240 struct backing_dev_info *bdi = mapping->backing_dev_info;
1241 unsigned long flags;
1243 write_lock_irqsave(&mapping->tree_lock, flags);
1244 ret = TestSetPageWriteback(page);
1246 radix_tree_tag_set(&mapping->page_tree,
1248 PAGECACHE_TAG_WRITEBACK);
1249 if (bdi_cap_writeback_dirty(bdi))
1250 __inc_bdi_stat(bdi, BDI_WRITEBACK);
1252 if (!PageDirty(page))
1253 radix_tree_tag_clear(&mapping->page_tree,
1255 PAGECACHE_TAG_DIRTY);
1256 write_unlock_irqrestore(&mapping->tree_lock, flags);
1258 ret = TestSetPageWriteback(page);
1261 inc_zone_page_state(page, NR_WRITEBACK);
1265 EXPORT_SYMBOL(test_set_page_writeback);
1268 * Return true if any of the pages in the mapping are marked with the
1271 int mapping_tagged(struct address_space *mapping, int tag)
1275 ret = radix_tree_tagged(&mapping->page_tree, tag);
1279 EXPORT_SYMBOL(mapping_tagged);