Clean up duplicate includes in fs/ecryptfs/
[linux-2.6/linux-acpi-2.6/ibm-acpi-2.6.git] / mm / page-writeback.c
blob33485213158810d5b50879556cb76c268faa4b68
1 /*
2 * mm/page-writeback.c
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
8 * address_space level.
10 * 10Apr2002 akpm@zip.com.au
11 * Initial version
14 #include <linux/kernel.h>
15 #include <linux/module.h>
16 #include <linux/spinlock.h>
17 #include <linux/fs.h>
18 #include <linux/mm.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_LOCK 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 * The generator of dirty data starts writeback at this percentage
74 int vm_dirty_ratio = 10;
77 * The interval between `kupdate'-style writebacks, in jiffies
79 int dirty_writeback_interval = 5 * HZ;
82 * The longest number of jiffies for which data is allowed to remain dirty
84 int dirty_expire_interval = 30 * HZ;
87 * Flag that makes the machine dump writes/reads and block dirtyings.
89 int block_dump;
92 * Flag that puts the machine in "laptop mode". Doubles as a timeout in jiffies:
93 * a full sync is triggered after this time elapses without any disk activity.
95 int laptop_mode;
97 EXPORT_SYMBOL(laptop_mode);
99 /* End of sysctl-exported parameters */
102 static void background_writeout(unsigned long _min_pages);
105 * Scale the writeback cache size proportional to the relative writeout speeds.
107 * We do this by keeping a floating proportion between BDIs, based on page
108 * writeback completions [end_page_writeback()]. Those devices that write out
109 * pages fastest will get the larger share, while the slower will get a smaller
110 * share.
112 * We use page writeout completions because we are interested in getting rid of
113 * dirty pages. Having them written out is the primary goal.
115 * We introduce a concept of time, a period over which we measure these events,
116 * because demand can/will vary over time. The length of this period itself is
117 * measured in page writeback completions.
120 static struct prop_descriptor vm_completions;
121 static struct prop_descriptor vm_dirties;
123 static unsigned long determine_dirtyable_memory(void);
126 * couple the period to the dirty_ratio:
128 * period/2 ~ roundup_pow_of_two(dirty limit)
130 static int calc_period_shift(void)
132 unsigned long dirty_total;
134 dirty_total = (vm_dirty_ratio * determine_dirtyable_memory()) / 100;
135 return 2 + ilog2(dirty_total - 1);
139 * update the period when the dirty ratio changes.
141 int dirty_ratio_handler(struct ctl_table *table, int write,
142 struct file *filp, void __user *buffer, size_t *lenp,
143 loff_t *ppos)
145 int old_ratio = vm_dirty_ratio;
146 int ret = proc_dointvec_minmax(table, write, filp, buffer, lenp, ppos);
147 if (ret == 0 && write && vm_dirty_ratio != old_ratio) {
148 int shift = calc_period_shift();
149 prop_change_shift(&vm_completions, shift);
150 prop_change_shift(&vm_dirties, shift);
152 return ret;
156 * Increment the BDI's writeout completion count and the global writeout
157 * completion count. Called from test_clear_page_writeback().
159 static inline void __bdi_writeout_inc(struct backing_dev_info *bdi)
161 __prop_inc_percpu(&vm_completions, &bdi->completions);
164 static inline void task_dirty_inc(struct task_struct *tsk)
166 prop_inc_single(&vm_dirties, &tsk->dirties);
170 * Obtain an accurate fraction of the BDI's portion.
172 static void bdi_writeout_fraction(struct backing_dev_info *bdi,
173 long *numerator, long *denominator)
175 if (bdi_cap_writeback_dirty(bdi)) {
176 prop_fraction_percpu(&vm_completions, &bdi->completions,
177 numerator, denominator);
178 } else {
179 *numerator = 0;
180 *denominator = 1;
185 * Clip the earned share of dirty pages to that which is actually available.
186 * This avoids exceeding the total dirty_limit when the floating averages
187 * fluctuate too quickly.
189 static void
190 clip_bdi_dirty_limit(struct backing_dev_info *bdi, long dirty, long *pbdi_dirty)
192 long avail_dirty;
194 avail_dirty = dirty -
195 (global_page_state(NR_FILE_DIRTY) +
196 global_page_state(NR_WRITEBACK) +
197 global_page_state(NR_UNSTABLE_NFS));
199 if (avail_dirty < 0)
200 avail_dirty = 0;
202 avail_dirty += bdi_stat(bdi, BDI_RECLAIMABLE) +
203 bdi_stat(bdi, BDI_WRITEBACK);
205 *pbdi_dirty = min(*pbdi_dirty, avail_dirty);
208 static inline void task_dirties_fraction(struct task_struct *tsk,
209 long *numerator, long *denominator)
211 prop_fraction_single(&vm_dirties, &tsk->dirties,
212 numerator, denominator);
216 * scale the dirty limit
218 * task specific dirty limit:
220 * dirty -= (dirty/8) * p_{t}
222 void task_dirty_limit(struct task_struct *tsk, long *pdirty)
224 long numerator, denominator;
225 long dirty = *pdirty;
226 u64 inv = dirty >> 3;
228 task_dirties_fraction(tsk, &numerator, &denominator);
229 inv *= numerator;
230 do_div(inv, denominator);
232 dirty -= inv;
233 if (dirty < *pdirty/2)
234 dirty = *pdirty/2;
236 *pdirty = dirty;
240 * Work out the current dirty-memory clamping and background writeout
241 * thresholds.
243 * The main aim here is to lower them aggressively if there is a lot of mapped
244 * memory around. To avoid stressing page reclaim with lots of unreclaimable
245 * pages. It is better to clamp down on writers than to start swapping, and
246 * performing lots of scanning.
248 * We only allow 1/2 of the currently-unmapped memory to be dirtied.
250 * We don't permit the clamping level to fall below 5% - that is getting rather
251 * excessive.
253 * We make sure that the background writeout level is below the adjusted
254 * clamping level.
257 static unsigned long highmem_dirtyable_memory(unsigned long total)
259 #ifdef CONFIG_HIGHMEM
260 int node;
261 unsigned long x = 0;
263 for_each_node_state(node, N_HIGH_MEMORY) {
264 struct zone *z =
265 &NODE_DATA(node)->node_zones[ZONE_HIGHMEM];
267 x += zone_page_state(z, NR_FREE_PAGES)
268 + zone_page_state(z, NR_INACTIVE)
269 + zone_page_state(z, NR_ACTIVE);
272 * Make sure that the number of highmem pages is never larger
273 * than the number of the total dirtyable memory. This can only
274 * occur in very strange VM situations but we want to make sure
275 * that this does not occur.
277 return min(x, total);
278 #else
279 return 0;
280 #endif
283 static unsigned long determine_dirtyable_memory(void)
285 unsigned long x;
287 x = global_page_state(NR_FREE_PAGES)
288 + global_page_state(NR_INACTIVE)
289 + global_page_state(NR_ACTIVE);
290 x -= highmem_dirtyable_memory(x);
291 return x + 1; /* Ensure that we never return 0 */
294 static void
295 get_dirty_limits(long *pbackground, long *pdirty, long *pbdi_dirty,
296 struct backing_dev_info *bdi)
298 int background_ratio; /* Percentages */
299 int dirty_ratio;
300 int unmapped_ratio;
301 long background;
302 long dirty;
303 unsigned long available_memory = determine_dirtyable_memory();
304 struct task_struct *tsk;
306 unmapped_ratio = 100 - ((global_page_state(NR_FILE_MAPPED) +
307 global_page_state(NR_ANON_PAGES)) * 100) /
308 available_memory;
310 dirty_ratio = vm_dirty_ratio;
311 if (dirty_ratio > unmapped_ratio / 2)
312 dirty_ratio = unmapped_ratio / 2;
314 if (dirty_ratio < 5)
315 dirty_ratio = 5;
317 background_ratio = dirty_background_ratio;
318 if (background_ratio >= dirty_ratio)
319 background_ratio = dirty_ratio / 2;
321 background = (background_ratio * available_memory) / 100;
322 dirty = (dirty_ratio * available_memory) / 100;
323 tsk = current;
324 if (tsk->flags & PF_LESS_THROTTLE || rt_task(tsk)) {
325 background += background / 4;
326 dirty += dirty / 4;
328 *pbackground = background;
329 *pdirty = dirty;
331 if (bdi) {
332 u64 bdi_dirty = dirty;
333 long numerator, denominator;
336 * Calculate this BDI's share of the dirty ratio.
338 bdi_writeout_fraction(bdi, &numerator, &denominator);
340 bdi_dirty *= numerator;
341 do_div(bdi_dirty, denominator);
343 *pbdi_dirty = bdi_dirty;
344 clip_bdi_dirty_limit(bdi, dirty, pbdi_dirty);
345 task_dirty_limit(current, pbdi_dirty);
350 * balance_dirty_pages() must be called by processes which are generating dirty
351 * data. It looks at the number of dirty pages in the machine and will force
352 * the caller to perform writeback if the system is over `vm_dirty_ratio'.
353 * If we're over `background_thresh' then pdflush is woken to perform some
354 * writeout.
356 static void balance_dirty_pages(struct address_space *mapping)
358 long bdi_nr_reclaimable;
359 long bdi_nr_writeback;
360 long background_thresh;
361 long dirty_thresh;
362 long bdi_thresh;
363 unsigned long pages_written = 0;
364 unsigned long write_chunk = sync_writeback_pages();
366 struct backing_dev_info *bdi = mapping->backing_dev_info;
368 for (;;) {
369 struct writeback_control wbc = {
370 .bdi = bdi,
371 .sync_mode = WB_SYNC_NONE,
372 .older_than_this = NULL,
373 .nr_to_write = write_chunk,
374 .range_cyclic = 1,
377 get_dirty_limits(&background_thresh, &dirty_thresh,
378 &bdi_thresh, bdi);
379 bdi_nr_reclaimable = bdi_stat(bdi, BDI_RECLAIMABLE);
380 bdi_nr_writeback = bdi_stat(bdi, BDI_WRITEBACK);
381 if (bdi_nr_reclaimable + bdi_nr_writeback <= bdi_thresh)
382 break;
384 if (!bdi->dirty_exceeded)
385 bdi->dirty_exceeded = 1;
387 /* Note: nr_reclaimable denotes nr_dirty + nr_unstable.
388 * Unstable writes are a feature of certain networked
389 * filesystems (i.e. NFS) in which data may have been
390 * written to the server's write cache, but has not yet
391 * been flushed to permanent storage.
393 if (bdi_nr_reclaimable) {
394 writeback_inodes(&wbc);
395 pages_written += write_chunk - wbc.nr_to_write;
396 get_dirty_limits(&background_thresh, &dirty_thresh,
397 &bdi_thresh, bdi);
401 * In order to avoid the stacked BDI deadlock we need
402 * to ensure we accurately count the 'dirty' pages when
403 * the threshold is low.
405 * Otherwise it would be possible to get thresh+n pages
406 * reported dirty, even though there are thresh-m pages
407 * actually dirty; with m+n sitting in the percpu
408 * deltas.
410 if (bdi_thresh < 2*bdi_stat_error(bdi)) {
411 bdi_nr_reclaimable = bdi_stat_sum(bdi, BDI_RECLAIMABLE);
412 bdi_nr_writeback = bdi_stat_sum(bdi, BDI_WRITEBACK);
413 } else if (bdi_nr_reclaimable) {
414 bdi_nr_reclaimable = bdi_stat(bdi, BDI_RECLAIMABLE);
415 bdi_nr_writeback = bdi_stat(bdi, BDI_WRITEBACK);
418 if (bdi_nr_reclaimable + bdi_nr_writeback <= bdi_thresh)
419 break;
420 if (pages_written >= write_chunk)
421 break; /* We've done our duty */
423 congestion_wait(WRITE, HZ/10);
426 if (bdi_nr_reclaimable + bdi_nr_writeback < bdi_thresh &&
427 bdi->dirty_exceeded)
428 bdi->dirty_exceeded = 0;
430 if (writeback_in_progress(bdi))
431 return; /* pdflush is already working this queue */
434 * In laptop mode, we wait until hitting the higher threshold before
435 * starting background writeout, and then write out all the way down
436 * to the lower threshold. So slow writers cause minimal disk activity.
438 * In normal mode, we start background writeout at the lower
439 * background_thresh, to keep the amount of dirty memory low.
441 if ((laptop_mode && pages_written) ||
442 (!laptop_mode && (global_page_state(NR_FILE_DIRTY)
443 + global_page_state(NR_UNSTABLE_NFS)
444 > background_thresh)))
445 pdflush_operation(background_writeout, 0);
448 void set_page_dirty_balance(struct page *page, int page_mkwrite)
450 if (set_page_dirty(page) || page_mkwrite) {
451 struct address_space *mapping = page_mapping(page);
453 if (mapping)
454 balance_dirty_pages_ratelimited(mapping);
459 * balance_dirty_pages_ratelimited_nr - balance dirty memory state
460 * @mapping: address_space which was dirtied
461 * @nr_pages_dirtied: number of pages which the caller has just dirtied
463 * Processes which are dirtying memory should call in here once for each page
464 * which was newly dirtied. The function will periodically check the system's
465 * dirty state and will initiate writeback if needed.
467 * On really big machines, get_writeback_state is expensive, so try to avoid
468 * calling it too often (ratelimiting). But once we're over the dirty memory
469 * limit we decrease the ratelimiting by a lot, to prevent individual processes
470 * from overshooting the limit by (ratelimit_pages) each.
472 void balance_dirty_pages_ratelimited_nr(struct address_space *mapping,
473 unsigned long nr_pages_dirtied)
475 static DEFINE_PER_CPU(unsigned long, ratelimits) = 0;
476 unsigned long ratelimit;
477 unsigned long *p;
479 ratelimit = ratelimit_pages;
480 if (mapping->backing_dev_info->dirty_exceeded)
481 ratelimit = 8;
484 * Check the rate limiting. Also, we do not want to throttle real-time
485 * tasks in balance_dirty_pages(). Period.
487 preempt_disable();
488 p = &__get_cpu_var(ratelimits);
489 *p += nr_pages_dirtied;
490 if (unlikely(*p >= ratelimit)) {
491 *p = 0;
492 preempt_enable();
493 balance_dirty_pages(mapping);
494 return;
496 preempt_enable();
498 EXPORT_SYMBOL(balance_dirty_pages_ratelimited_nr);
500 void throttle_vm_writeout(gfp_t gfp_mask)
502 long background_thresh;
503 long dirty_thresh;
505 if ((gfp_mask & (__GFP_FS|__GFP_IO)) != (__GFP_FS|__GFP_IO)) {
507 * The caller might hold locks which can prevent IO completion
508 * or progress in the filesystem. So we cannot just sit here
509 * waiting for IO to complete.
511 congestion_wait(WRITE, HZ/10);
512 return;
515 for ( ; ; ) {
516 get_dirty_limits(&background_thresh, &dirty_thresh, NULL, NULL);
519 * Boost the allowable dirty threshold a bit for page
520 * allocators so they don't get DoS'ed by heavy writers
522 dirty_thresh += dirty_thresh / 10; /* wheeee... */
524 if (global_page_state(NR_UNSTABLE_NFS) +
525 global_page_state(NR_WRITEBACK) <= dirty_thresh)
526 break;
527 congestion_wait(WRITE, HZ/10);
532 * writeback at least _min_pages, and keep writing until the amount of dirty
533 * memory is less than the background threshold, or until we're all clean.
535 static void background_writeout(unsigned long _min_pages)
537 long min_pages = _min_pages;
538 struct writeback_control wbc = {
539 .bdi = NULL,
540 .sync_mode = WB_SYNC_NONE,
541 .older_than_this = NULL,
542 .nr_to_write = 0,
543 .nonblocking = 1,
544 .range_cyclic = 1,
547 for ( ; ; ) {
548 long background_thresh;
549 long dirty_thresh;
551 get_dirty_limits(&background_thresh, &dirty_thresh, NULL, NULL);
552 if (global_page_state(NR_FILE_DIRTY) +
553 global_page_state(NR_UNSTABLE_NFS) < background_thresh
554 && min_pages <= 0)
555 break;
556 wbc.encountered_congestion = 0;
557 wbc.nr_to_write = MAX_WRITEBACK_PAGES;
558 wbc.pages_skipped = 0;
559 writeback_inodes(&wbc);
560 min_pages -= MAX_WRITEBACK_PAGES - wbc.nr_to_write;
561 if (wbc.nr_to_write > 0 || wbc.pages_skipped > 0) {
562 /* Wrote less than expected */
563 congestion_wait(WRITE, HZ/10);
564 if (!wbc.encountered_congestion)
565 break;
571 * Start writeback of `nr_pages' pages. If `nr_pages' is zero, write back
572 * the whole world. Returns 0 if a pdflush thread was dispatched. Returns
573 * -1 if all pdflush threads were busy.
575 int wakeup_pdflush(long nr_pages)
577 if (nr_pages == 0)
578 nr_pages = global_page_state(NR_FILE_DIRTY) +
579 global_page_state(NR_UNSTABLE_NFS);
580 return pdflush_operation(background_writeout, nr_pages);
583 static void wb_timer_fn(unsigned long unused);
584 static void laptop_timer_fn(unsigned long unused);
586 static DEFINE_TIMER(wb_timer, wb_timer_fn, 0, 0);
587 static DEFINE_TIMER(laptop_mode_wb_timer, laptop_timer_fn, 0, 0);
590 * Periodic writeback of "old" data.
592 * Define "old": the first time one of an inode's pages is dirtied, we mark the
593 * dirtying-time in the inode's address_space. So this periodic writeback code
594 * just walks the superblock inode list, writing back any inodes which are
595 * older than a specific point in time.
597 * Try to run once per dirty_writeback_interval. But if a writeback event
598 * takes longer than a dirty_writeback_interval interval, then leave a
599 * one-second gap.
601 * older_than_this takes precedence over nr_to_write. So we'll only write back
602 * all dirty pages if they are all attached to "old" mappings.
604 static void wb_kupdate(unsigned long arg)
606 unsigned long oldest_jif;
607 unsigned long start_jif;
608 unsigned long next_jif;
609 long nr_to_write;
610 struct writeback_control wbc = {
611 .bdi = NULL,
612 .sync_mode = WB_SYNC_NONE,
613 .older_than_this = &oldest_jif,
614 .nr_to_write = 0,
615 .nonblocking = 1,
616 .for_kupdate = 1,
617 .range_cyclic = 1,
620 sync_supers();
622 oldest_jif = jiffies - dirty_expire_interval;
623 start_jif = jiffies;
624 next_jif = start_jif + dirty_writeback_interval;
625 nr_to_write = global_page_state(NR_FILE_DIRTY) +
626 global_page_state(NR_UNSTABLE_NFS) +
627 (inodes_stat.nr_inodes - inodes_stat.nr_unused);
628 while (nr_to_write > 0) {
629 wbc.encountered_congestion = 0;
630 wbc.nr_to_write = MAX_WRITEBACK_PAGES;
631 writeback_inodes(&wbc);
632 if (wbc.nr_to_write > 0) {
633 if (wbc.encountered_congestion)
634 congestion_wait(WRITE, HZ/10);
635 else
636 break; /* All the old data is written */
638 nr_to_write -= MAX_WRITEBACK_PAGES - wbc.nr_to_write;
640 if (time_before(next_jif, jiffies + HZ))
641 next_jif = jiffies + HZ;
642 if (dirty_writeback_interval)
643 mod_timer(&wb_timer, next_jif);
647 * sysctl handler for /proc/sys/vm/dirty_writeback_centisecs
649 int dirty_writeback_centisecs_handler(ctl_table *table, int write,
650 struct file *file, void __user *buffer, size_t *length, loff_t *ppos)
652 proc_dointvec_userhz_jiffies(table, write, file, buffer, length, ppos);
653 if (dirty_writeback_interval)
654 mod_timer(&wb_timer, jiffies + dirty_writeback_interval);
655 else
656 del_timer(&wb_timer);
657 return 0;
660 static void wb_timer_fn(unsigned long unused)
662 if (pdflush_operation(wb_kupdate, 0) < 0)
663 mod_timer(&wb_timer, jiffies + HZ); /* delay 1 second */
666 static void laptop_flush(unsigned long unused)
668 sys_sync();
671 static void laptop_timer_fn(unsigned long unused)
673 pdflush_operation(laptop_flush, 0);
677 * We've spun up the disk and we're in laptop mode: schedule writeback
678 * of all dirty data a few seconds from now. If the flush is already scheduled
679 * then push it back - the user is still using the disk.
681 void laptop_io_completion(void)
683 mod_timer(&laptop_mode_wb_timer, jiffies + laptop_mode);
687 * We're in laptop mode and we've just synced. The sync's writes will have
688 * caused another writeback to be scheduled by laptop_io_completion.
689 * Nothing needs to be written back anymore, so we unschedule the writeback.
691 void laptop_sync_completion(void)
693 del_timer(&laptop_mode_wb_timer);
697 * If ratelimit_pages is too high then we can get into dirty-data overload
698 * if a large number of processes all perform writes at the same time.
699 * If it is too low then SMP machines will call the (expensive)
700 * get_writeback_state too often.
702 * Here we set ratelimit_pages to a level which ensures that when all CPUs are
703 * dirtying in parallel, we cannot go more than 3% (1/32) over the dirty memory
704 * thresholds before writeback cuts in.
706 * But the limit should not be set too high. Because it also controls the
707 * amount of memory which the balance_dirty_pages() caller has to write back.
708 * If this is too large then the caller will block on the IO queue all the
709 * time. So limit it to four megabytes - the balance_dirty_pages() caller
710 * will write six megabyte chunks, max.
713 void writeback_set_ratelimit(void)
715 ratelimit_pages = vm_total_pages / (num_online_cpus() * 32);
716 if (ratelimit_pages < 16)
717 ratelimit_pages = 16;
718 if (ratelimit_pages * PAGE_CACHE_SIZE > 4096 * 1024)
719 ratelimit_pages = (4096 * 1024) / PAGE_CACHE_SIZE;
722 static int __cpuinit
723 ratelimit_handler(struct notifier_block *self, unsigned long u, void *v)
725 writeback_set_ratelimit();
726 return NOTIFY_DONE;
729 static struct notifier_block __cpuinitdata ratelimit_nb = {
730 .notifier_call = ratelimit_handler,
731 .next = NULL,
735 * Called early on to tune the page writeback dirty limits.
737 * We used to scale dirty pages according to how total memory
738 * related to pages that could be allocated for buffers (by
739 * comparing nr_free_buffer_pages() to vm_total_pages.
741 * However, that was when we used "dirty_ratio" to scale with
742 * all memory, and we don't do that any more. "dirty_ratio"
743 * is now applied to total non-HIGHPAGE memory (by subtracting
744 * totalhigh_pages from vm_total_pages), and as such we can't
745 * get into the old insane situation any more where we had
746 * large amounts of dirty pages compared to a small amount of
747 * non-HIGHMEM memory.
749 * But we might still want to scale the dirty_ratio by how
750 * much memory the box has..
752 void __init page_writeback_init(void)
754 int shift;
756 mod_timer(&wb_timer, jiffies + dirty_writeback_interval);
757 writeback_set_ratelimit();
758 register_cpu_notifier(&ratelimit_nb);
760 shift = calc_period_shift();
761 prop_descriptor_init(&vm_completions, shift);
762 prop_descriptor_init(&vm_dirties, shift);
766 * write_cache_pages - walk the list of dirty pages of the given address space and write all of them.
767 * @mapping: address space structure to write
768 * @wbc: subtract the number of written pages from *@wbc->nr_to_write
769 * @writepage: function called for each page
770 * @data: data passed to writepage function
772 * If a page is already under I/O, write_cache_pages() skips it, even
773 * if it's dirty. This is desirable behaviour for memory-cleaning writeback,
774 * but it is INCORRECT for data-integrity system calls such as fsync(). fsync()
775 * and msync() need to guarantee that all the data which was dirty at the time
776 * the call was made get new I/O started against them. If wbc->sync_mode is
777 * WB_SYNC_ALL then we were called for data integrity and we must wait for
778 * existing IO to complete.
780 int write_cache_pages(struct address_space *mapping,
781 struct writeback_control *wbc, writepage_t writepage,
782 void *data)
784 struct backing_dev_info *bdi = mapping->backing_dev_info;
785 int ret = 0;
786 int done = 0;
787 struct pagevec pvec;
788 int nr_pages;
789 pgoff_t index;
790 pgoff_t end; /* Inclusive */
791 int scanned = 0;
792 int range_whole = 0;
794 if (wbc->nonblocking && bdi_write_congested(bdi)) {
795 wbc->encountered_congestion = 1;
796 return 0;
799 pagevec_init(&pvec, 0);
800 if (wbc->range_cyclic) {
801 index = mapping->writeback_index; /* Start from prev offset */
802 end = -1;
803 } else {
804 index = wbc->range_start >> PAGE_CACHE_SHIFT;
805 end = wbc->range_end >> PAGE_CACHE_SHIFT;
806 if (wbc->range_start == 0 && wbc->range_end == LLONG_MAX)
807 range_whole = 1;
808 scanned = 1;
810 retry:
811 while (!done && (index <= end) &&
812 (nr_pages = pagevec_lookup_tag(&pvec, mapping, &index,
813 PAGECACHE_TAG_DIRTY,
814 min(end - index, (pgoff_t)PAGEVEC_SIZE-1) + 1))) {
815 unsigned i;
817 scanned = 1;
818 for (i = 0; i < nr_pages; i++) {
819 struct page *page = pvec.pages[i];
822 * At this point we hold neither mapping->tree_lock nor
823 * lock on the page itself: the page may be truncated or
824 * invalidated (changing page->mapping to NULL), or even
825 * swizzled back from swapper_space to tmpfs file
826 * mapping
828 lock_page(page);
830 if (unlikely(page->mapping != mapping)) {
831 unlock_page(page);
832 continue;
835 if (!wbc->range_cyclic && page->index > end) {
836 done = 1;
837 unlock_page(page);
838 continue;
841 if (wbc->sync_mode != WB_SYNC_NONE)
842 wait_on_page_writeback(page);
844 if (PageWriteback(page) ||
845 !clear_page_dirty_for_io(page)) {
846 unlock_page(page);
847 continue;
850 ret = (*writepage)(page, wbc, data);
852 if (unlikely(ret == AOP_WRITEPAGE_ACTIVATE)) {
853 unlock_page(page);
854 ret = 0;
856 if (ret || (--(wbc->nr_to_write) <= 0))
857 done = 1;
858 if (wbc->nonblocking && bdi_write_congested(bdi)) {
859 wbc->encountered_congestion = 1;
860 done = 1;
863 pagevec_release(&pvec);
864 cond_resched();
866 if (!scanned && !done) {
868 * We hit the last page and there is more work to be done: wrap
869 * back to the start of the file
871 scanned = 1;
872 index = 0;
873 goto retry;
875 if (wbc->range_cyclic || (range_whole && wbc->nr_to_write > 0))
876 mapping->writeback_index = index;
877 return ret;
879 EXPORT_SYMBOL(write_cache_pages);
882 * Function used by generic_writepages to call the real writepage
883 * function and set the mapping flags on error
885 static int __writepage(struct page *page, struct writeback_control *wbc,
886 void *data)
888 struct address_space *mapping = data;
889 int ret = mapping->a_ops->writepage(page, wbc);
890 mapping_set_error(mapping, ret);
891 return ret;
895 * generic_writepages - walk the list of dirty pages of the given address space and writepage() all of them.
896 * @mapping: address space structure to write
897 * @wbc: subtract the number of written pages from *@wbc->nr_to_write
899 * This is a library function, which implements the writepages()
900 * address_space_operation.
902 int generic_writepages(struct address_space *mapping,
903 struct writeback_control *wbc)
905 /* deal with chardevs and other special file */
906 if (!mapping->a_ops->writepage)
907 return 0;
909 return write_cache_pages(mapping, wbc, __writepage, mapping);
912 EXPORT_SYMBOL(generic_writepages);
914 int do_writepages(struct address_space *mapping, struct writeback_control *wbc)
916 int ret;
918 if (wbc->nr_to_write <= 0)
919 return 0;
920 wbc->for_writepages = 1;
921 if (mapping->a_ops->writepages)
922 ret = mapping->a_ops->writepages(mapping, wbc);
923 else
924 ret = generic_writepages(mapping, wbc);
925 wbc->for_writepages = 0;
926 return ret;
930 * write_one_page - write out a single page and optionally wait on I/O
931 * @page: the page to write
932 * @wait: if true, wait on writeout
934 * The page must be locked by the caller and will be unlocked upon return.
936 * write_one_page() returns a negative error code if I/O failed.
938 int write_one_page(struct page *page, int wait)
940 struct address_space *mapping = page->mapping;
941 int ret = 0;
942 struct writeback_control wbc = {
943 .sync_mode = WB_SYNC_ALL,
944 .nr_to_write = 1,
947 BUG_ON(!PageLocked(page));
949 if (wait)
950 wait_on_page_writeback(page);
952 if (clear_page_dirty_for_io(page)) {
953 page_cache_get(page);
954 ret = mapping->a_ops->writepage(page, &wbc);
955 if (ret == 0 && wait) {
956 wait_on_page_writeback(page);
957 if (PageError(page))
958 ret = -EIO;
960 page_cache_release(page);
961 } else {
962 unlock_page(page);
964 return ret;
966 EXPORT_SYMBOL(write_one_page);
969 * For address_spaces which do not use buffers nor write back.
971 int __set_page_dirty_no_writeback(struct page *page)
973 if (!PageDirty(page))
974 SetPageDirty(page);
975 return 0;
979 * For address_spaces which do not use buffers. Just tag the page as dirty in
980 * its radix tree.
982 * This is also used when a single buffer is being dirtied: we want to set the
983 * page dirty in that case, but not all the buffers. This is a "bottom-up"
984 * dirtying, whereas __set_page_dirty_buffers() is a "top-down" dirtying.
986 * Most callers have locked the page, which pins the address_space in memory.
987 * But zap_pte_range() does not lock the page, however in that case the
988 * mapping is pinned by the vma's ->vm_file reference.
990 * We take care to handle the case where the page was truncated from the
991 * mapping by re-checking page_mapping() insode tree_lock.
993 int __set_page_dirty_nobuffers(struct page *page)
995 if (!TestSetPageDirty(page)) {
996 struct address_space *mapping = page_mapping(page);
997 struct address_space *mapping2;
999 if (!mapping)
1000 return 1;
1002 write_lock_irq(&mapping->tree_lock);
1003 mapping2 = page_mapping(page);
1004 if (mapping2) { /* Race with truncate? */
1005 BUG_ON(mapping2 != mapping);
1006 WARN_ON_ONCE(!PagePrivate(page) && !PageUptodate(page));
1007 if (mapping_cap_account_dirty(mapping)) {
1008 __inc_zone_page_state(page, NR_FILE_DIRTY);
1009 __inc_bdi_stat(mapping->backing_dev_info,
1010 BDI_RECLAIMABLE);
1011 task_io_account_write(PAGE_CACHE_SIZE);
1013 radix_tree_tag_set(&mapping->page_tree,
1014 page_index(page), PAGECACHE_TAG_DIRTY);
1016 write_unlock_irq(&mapping->tree_lock);
1017 if (mapping->host) {
1018 /* !PageAnon && !swapper_space */
1019 __mark_inode_dirty(mapping->host, I_DIRTY_PAGES);
1021 return 1;
1023 return 0;
1025 EXPORT_SYMBOL(__set_page_dirty_nobuffers);
1028 * When a writepage implementation decides that it doesn't want to write this
1029 * page for some reason, it should redirty the locked page via
1030 * redirty_page_for_writepage() and it should then unlock the page and return 0
1032 int redirty_page_for_writepage(struct writeback_control *wbc, struct page *page)
1034 wbc->pages_skipped++;
1035 return __set_page_dirty_nobuffers(page);
1037 EXPORT_SYMBOL(redirty_page_for_writepage);
1040 * If the mapping doesn't provide a set_page_dirty a_op, then
1041 * just fall through and assume that it wants buffer_heads.
1043 static int __set_page_dirty(struct page *page)
1045 struct address_space *mapping = page_mapping(page);
1047 if (likely(mapping)) {
1048 int (*spd)(struct page *) = mapping->a_ops->set_page_dirty;
1049 #ifdef CONFIG_BLOCK
1050 if (!spd)
1051 spd = __set_page_dirty_buffers;
1052 #endif
1053 return (*spd)(page);
1055 if (!PageDirty(page)) {
1056 if (!TestSetPageDirty(page))
1057 return 1;
1059 return 0;
1062 int fastcall set_page_dirty(struct page *page)
1064 int ret = __set_page_dirty(page);
1065 if (ret)
1066 task_dirty_inc(current);
1067 return ret;
1069 EXPORT_SYMBOL(set_page_dirty);
1072 * set_page_dirty() is racy if the caller has no reference against
1073 * page->mapping->host, and if the page is unlocked. This is because another
1074 * CPU could truncate the page off the mapping and then free the mapping.
1076 * Usually, the page _is_ locked, or the caller is a user-space process which
1077 * holds a reference on the inode by having an open file.
1079 * In other cases, the page should be locked before running set_page_dirty().
1081 int set_page_dirty_lock(struct page *page)
1083 int ret;
1085 lock_page_nosync(page);
1086 ret = set_page_dirty(page);
1087 unlock_page(page);
1088 return ret;
1090 EXPORT_SYMBOL(set_page_dirty_lock);
1093 * Clear a page's dirty flag, while caring for dirty memory accounting.
1094 * Returns true if the page was previously dirty.
1096 * This is for preparing to put the page under writeout. We leave the page
1097 * tagged as dirty in the radix tree so that a concurrent write-for-sync
1098 * can discover it via a PAGECACHE_TAG_DIRTY walk. The ->writepage
1099 * implementation will run either set_page_writeback() or set_page_dirty(),
1100 * at which stage we bring the page's dirty flag and radix-tree dirty tag
1101 * back into sync.
1103 * This incoherency between the page's dirty flag and radix-tree tag is
1104 * unfortunate, but it only exists while the page is locked.
1106 int clear_page_dirty_for_io(struct page *page)
1108 struct address_space *mapping = page_mapping(page);
1110 BUG_ON(!PageLocked(page));
1112 ClearPageReclaim(page);
1113 if (mapping && mapping_cap_account_dirty(mapping)) {
1115 * Yes, Virginia, this is indeed insane.
1117 * We use this sequence to make sure that
1118 * (a) we account for dirty stats properly
1119 * (b) we tell the low-level filesystem to
1120 * mark the whole page dirty if it was
1121 * dirty in a pagetable. Only to then
1122 * (c) clean the page again and return 1 to
1123 * cause the writeback.
1125 * This way we avoid all nasty races with the
1126 * dirty bit in multiple places and clearing
1127 * them concurrently from different threads.
1129 * Note! Normally the "set_page_dirty(page)"
1130 * has no effect on the actual dirty bit - since
1131 * that will already usually be set. But we
1132 * need the side effects, and it can help us
1133 * avoid races.
1135 * We basically use the page "master dirty bit"
1136 * as a serialization point for all the different
1137 * threads doing their things.
1139 if (page_mkclean(page))
1140 set_page_dirty(page);
1142 * We carefully synchronise fault handlers against
1143 * installing a dirty pte and marking the page dirty
1144 * at this point. We do this by having them hold the
1145 * page lock at some point after installing their
1146 * pte, but before marking the page dirty.
1147 * Pages are always locked coming in here, so we get
1148 * the desired exclusion. See mm/memory.c:do_wp_page()
1149 * for more comments.
1151 if (TestClearPageDirty(page)) {
1152 dec_zone_page_state(page, NR_FILE_DIRTY);
1153 dec_bdi_stat(mapping->backing_dev_info,
1154 BDI_RECLAIMABLE);
1155 return 1;
1157 return 0;
1159 return TestClearPageDirty(page);
1161 EXPORT_SYMBOL(clear_page_dirty_for_io);
1163 int test_clear_page_writeback(struct page *page)
1165 struct address_space *mapping = page_mapping(page);
1166 int ret;
1168 if (mapping) {
1169 struct backing_dev_info *bdi = mapping->backing_dev_info;
1170 unsigned long flags;
1172 write_lock_irqsave(&mapping->tree_lock, flags);
1173 ret = TestClearPageWriteback(page);
1174 if (ret) {
1175 radix_tree_tag_clear(&mapping->page_tree,
1176 page_index(page),
1177 PAGECACHE_TAG_WRITEBACK);
1178 if (bdi_cap_writeback_dirty(bdi)) {
1179 __dec_bdi_stat(bdi, BDI_WRITEBACK);
1180 __bdi_writeout_inc(bdi);
1183 write_unlock_irqrestore(&mapping->tree_lock, flags);
1184 } else {
1185 ret = TestClearPageWriteback(page);
1187 if (ret)
1188 dec_zone_page_state(page, NR_WRITEBACK);
1189 return ret;
1192 int test_set_page_writeback(struct page *page)
1194 struct address_space *mapping = page_mapping(page);
1195 int ret;
1197 if (mapping) {
1198 struct backing_dev_info *bdi = mapping->backing_dev_info;
1199 unsigned long flags;
1201 write_lock_irqsave(&mapping->tree_lock, flags);
1202 ret = TestSetPageWriteback(page);
1203 if (!ret) {
1204 radix_tree_tag_set(&mapping->page_tree,
1205 page_index(page),
1206 PAGECACHE_TAG_WRITEBACK);
1207 if (bdi_cap_writeback_dirty(bdi))
1208 __inc_bdi_stat(bdi, BDI_WRITEBACK);
1210 if (!PageDirty(page))
1211 radix_tree_tag_clear(&mapping->page_tree,
1212 page_index(page),
1213 PAGECACHE_TAG_DIRTY);
1214 write_unlock_irqrestore(&mapping->tree_lock, flags);
1215 } else {
1216 ret = TestSetPageWriteback(page);
1218 if (!ret)
1219 inc_zone_page_state(page, NR_WRITEBACK);
1220 return ret;
1223 EXPORT_SYMBOL(test_set_page_writeback);
1226 * Return true if any of the pages in the mapping are marked with the
1227 * passed tag.
1229 int mapping_tagged(struct address_space *mapping, int tag)
1231 int ret;
1232 rcu_read_lock();
1233 ret = radix_tree_tagged(&mapping->page_tree, tag);
1234 rcu_read_unlock();
1235 return ret;
1237 EXPORT_SYMBOL(mapping_tagged);