~ [ source navigation ] ~ [ diff markup ] ~ [ identifier search ] ~

TOMOYO Linux Cross Reference
Linux/mm/page-writeback.c

Version: ~ [ linux-5.18-rc6 ] ~ [ linux-5.17.6 ] ~ [ linux-5.16.20 ] ~ [ linux-5.15.38 ] ~ [ linux-5.14.21 ] ~ [ linux-5.13.19 ] ~ [ linux-5.12.19 ] ~ [ linux-5.11.22 ] ~ [ linux-5.10.114 ] ~ [ linux-5.9.16 ] ~ [ linux-5.8.18 ] ~ [ linux-5.7.19 ] ~ [ linux-5.6.19 ] ~ [ linux-5.5.19 ] ~ [ linux-5.4.192 ] ~ [ linux-5.3.18 ] ~ [ linux-5.2.21 ] ~ [ linux-5.1.21 ] ~ [ linux-5.0.21 ] ~ [ linux-4.20.17 ] ~ [ linux-4.19.241 ] ~ [ linux-4.18.20 ] ~ [ linux-4.17.19 ] ~ [ linux-4.16.18 ] ~ [ linux-4.15.18 ] ~ [ linux-4.14.277 ] ~ [ linux-4.13.16 ] ~ [ linux-4.12.14 ] ~ [ linux-4.11.12 ] ~ [ linux-4.10.17 ] ~ [ linux-4.9.312 ] ~ [ linux-4.8.17 ] ~ [ linux-4.7.10 ] ~ [ linux-4.6.7 ] ~ [ linux-4.5.7 ] ~ [ linux-4.4.302 ] ~ [ linux-4.3.6 ] ~ [ linux-4.2.8 ] ~ [ linux-4.1.52 ] ~ [ linux-4.0.9 ] ~ [ linux-3.10.108 ] ~ [ linux-2.6.32.71 ] ~ [ linux-2.6.0 ] ~ [ linux-2.4.37.11 ] ~ [ unix-v6-master ] ~ [ ccs-tools-1.8.9 ] ~ [ policy-sample ] ~
Architecture: ~ [ i386 ] ~ [ alpha ] ~ [ m68k ] ~ [ mips ] ~ [ ppc ] ~ [ sparc ] ~ [ sparc64 ] ~

  1 /*
  2  * mm/page-writeback.c
  3  *
  4  * Copyright (C) 2002, Linus Torvalds.
  5  * Copyright (C) 2007 Red Hat, Inc., Peter Zijlstra <pzijlstr@redhat.com>
  6  *
  7  * Contains functions related to writing back dirty pages at the
  8  * address_space level.
  9  *
 10  * 10Apr2002    Andrew Morton
 11  *              Initial version
 12  */
 13 
 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>
 37 
 38 /*
 39  * After a CPU has dirtied this many pages, balance_dirty_pages_ratelimited
 40  * will look to see if it needs to force writeback or throttling.
 41  */
 42 static long ratelimit_pages = 32;
 43 
 44 /*
 45  * When balance_dirty_pages decides that the caller needs to perform some
 46  * non-background writeback, this is how many pages it will attempt to write.
 47  * It should be somewhat larger than dirtied pages to ensure that reasonably
 48  * large amounts of I/O are submitted.
 49  */
 50 static inline long sync_writeback_pages(unsigned long dirtied)
 51 {
 52         if (dirtied < ratelimit_pages)
 53                 dirtied = ratelimit_pages;
 54 
 55         return dirtied + dirtied / 2;
 56 }
 57 
 58 /* The following parameters are exported via /proc/sys/vm */
 59 
 60 /*
 61  * Start background writeback (via writeback threads) at this percentage
 62  */
 63 int dirty_background_ratio = 10;
 64 
 65 /*
 66  * dirty_background_bytes starts at 0 (disabled) so that it is a function of
 67  * dirty_background_ratio * the amount of dirtyable memory
 68  */
 69 unsigned long dirty_background_bytes;
 70 
 71 /*
 72  * free highmem will not be subtracted from the total free memory
 73  * for calculating free ratios if vm_highmem_is_dirtyable is true
 74  */
 75 int vm_highmem_is_dirtyable;
 76 
 77 /*
 78  * The generator of dirty data starts writeback at this percentage
 79  */
 80 int vm_dirty_ratio = 20;
 81 
 82 /*
 83  * vm_dirty_bytes starts at 0 (disabled) so that it is a function of
 84  * vm_dirty_ratio * the amount of dirtyable memory
 85  */
 86 unsigned long vm_dirty_bytes;
 87 
 88 /*
 89  * The interval between `kupdate'-style writebacks
 90  */
 91 unsigned int dirty_writeback_interval = 5 * 100; /* centiseconds */
 92 
 93 /*
 94  * The longest time for which data is allowed to remain dirty
 95  */
 96 unsigned int dirty_expire_interval = 30 * 100; /* centiseconds */
 97 
 98 /*
 99  * Flag that makes the machine dump writes/reads and block dirtyings.
100  */
101 int block_dump;
102 
103 /*
104  * Flag that puts the machine in "laptop mode". Doubles as a timeout in jiffies:
105  * a full sync is triggered after this time elapses without any disk activity.
106  */
107 int laptop_mode;
108 
109 EXPORT_SYMBOL(laptop_mode);
110 
111 /* End of sysctl-exported parameters */
112 
113 
114 /*
115  * Scale the writeback cache size proportional to the relative writeout speeds.
116  *
117  * We do this by keeping a floating proportion between BDIs, based on page
118  * writeback completions [end_page_writeback()]. Those devices that write out
119  * pages fastest will get the larger share, while the slower will get a smaller
120  * share.
121  *
122  * We use page writeout completions because we are interested in getting rid of
123  * dirty pages. Having them written out is the primary goal.
124  *
125  * We introduce a concept of time, a period over which we measure these events,
126  * because demand can/will vary over time. The length of this period itself is
127  * measured in page writeback completions.
128  *
129  */
130 static struct prop_descriptor vm_completions;
131 static struct prop_descriptor vm_dirties;
132 
133 /*
134  * couple the period to the dirty_ratio:
135  *
136  *   period/2 ~ roundup_pow_of_two(dirty limit)
137  */
138 static int calc_period_shift(void)
139 {
140         unsigned long dirty_total;
141 
142         if (vm_dirty_bytes)
143                 dirty_total = vm_dirty_bytes / PAGE_SIZE;
144         else
145                 dirty_total = (vm_dirty_ratio * determine_dirtyable_memory()) /
146                                 100;
147         return 2 + ilog2(dirty_total - 1);
148 }
149 
150 /*
151  * update the period when the dirty threshold changes.
152  */
153 static void update_completion_period(void)
154 {
155         int shift = calc_period_shift();
156         prop_change_shift(&vm_completions, shift);
157         prop_change_shift(&vm_dirties, shift);
158 }
159 
160 int dirty_background_ratio_handler(struct ctl_table *table, int write,
161                 void __user *buffer, size_t *lenp,
162                 loff_t *ppos)
163 {
164         int ret;
165 
166         ret = proc_dointvec_minmax(table, write, buffer, lenp, ppos);
167         if (ret == 0 && write)
168                 dirty_background_bytes = 0;
169         return ret;
170 }
171 
172 int dirty_background_bytes_handler(struct ctl_table *table, int write,
173                 void __user *buffer, size_t *lenp,
174                 loff_t *ppos)
175 {
176         int ret;
177 
178         ret = proc_doulongvec_minmax(table, write, buffer, lenp, ppos);
179         if (ret == 0 && write)
180                 dirty_background_ratio = 0;
181         return ret;
182 }
183 
184 int dirty_ratio_handler(struct ctl_table *table, int write,
185                 void __user *buffer, size_t *lenp,
186                 loff_t *ppos)
187 {
188         int old_ratio = vm_dirty_ratio;
189         int ret;
190 
191         ret = proc_dointvec_minmax(table, write, buffer, lenp, ppos);
192         if (ret == 0 && write && vm_dirty_ratio != old_ratio) {
193                 update_completion_period();
194                 vm_dirty_bytes = 0;
195         }
196         return ret;
197 }
198 
199 
200 int dirty_bytes_handler(struct ctl_table *table, int write,
201                 void __user *buffer, size_t *lenp,
202                 loff_t *ppos)
203 {
204         unsigned long old_bytes = vm_dirty_bytes;
205         int ret;
206 
207         ret = proc_doulongvec_minmax(table, write, buffer, lenp, ppos);
208         if (ret == 0 && write && vm_dirty_bytes != old_bytes) {
209                 update_completion_period();
210                 vm_dirty_ratio = 0;
211         }
212         return ret;
213 }
214 
215 /*
216  * Increment the BDI's writeout completion count and the global writeout
217  * completion count. Called from test_clear_page_writeback().
218  */
219 static inline void __bdi_writeout_inc(struct backing_dev_info *bdi)
220 {
221         __prop_inc_percpu_max(&vm_completions, &bdi->completions,
222                               bdi->max_prop_frac);
223 }
224 
225 void bdi_writeout_inc(struct backing_dev_info *bdi)
226 {
227         unsigned long flags;
228 
229         local_irq_save(flags);
230         __bdi_writeout_inc(bdi);
231         local_irq_restore(flags);
232 }
233 EXPORT_SYMBOL_GPL(bdi_writeout_inc);
234 
235 void task_dirty_inc(struct task_struct *tsk)
236 {
237         prop_inc_single(&vm_dirties, &tsk->dirties);
238 }
239 
240 /*
241  * Obtain an accurate fraction of the BDI's portion.
242  */
243 static void bdi_writeout_fraction(struct backing_dev_info *bdi,
244                 long *numerator, long *denominator)
245 {
246         if (bdi_cap_writeback_dirty(bdi)) {
247                 prop_fraction_percpu(&vm_completions, &bdi->completions,
248                                 numerator, denominator);
249         } else {
250                 *numerator = 0;
251                 *denominator = 1;
252         }
253 }
254 
255 /*
256  * Clip the earned share of dirty pages to that which is actually available.
257  * This avoids exceeding the total dirty_limit when the floating averages
258  * fluctuate too quickly.
259  */
260 static void clip_bdi_dirty_limit(struct backing_dev_info *bdi,
261                 unsigned long dirty, unsigned long *pbdi_dirty)
262 {
263         unsigned long avail_dirty;
264 
265         avail_dirty = global_page_state(NR_FILE_DIRTY) +
266                  global_page_state(NR_WRITEBACK) +
267                  global_page_state(NR_UNSTABLE_NFS) +
268                  global_page_state(NR_WRITEBACK_TEMP);
269 
270         if (avail_dirty < dirty)
271                 avail_dirty = dirty - avail_dirty;
272         else
273                 avail_dirty = 0;
274 
275         avail_dirty += bdi_stat(bdi, BDI_RECLAIMABLE) +
276                 bdi_stat(bdi, BDI_WRITEBACK);
277 
278         *pbdi_dirty = min(*pbdi_dirty, avail_dirty);
279 }
280 
281 static inline void task_dirties_fraction(struct task_struct *tsk,
282                 long *numerator, long *denominator)
283 {
284         prop_fraction_single(&vm_dirties, &tsk->dirties,
285                                 numerator, denominator);
286 }
287 
288 /*
289  * scale the dirty limit
290  *
291  * task specific dirty limit:
292  *
293  *   dirty -= (dirty/8) * p_{t}
294  */
295 static void task_dirty_limit(struct task_struct *tsk, unsigned long *pdirty)
296 {
297         long numerator, denominator;
298         unsigned long dirty = *pdirty;
299         u64 inv = dirty >> 3;
300 
301         task_dirties_fraction(tsk, &numerator, &denominator);
302         inv *= numerator;
303         do_div(inv, denominator);
304 
305         dirty -= inv;
306         if (dirty < *pdirty/2)
307                 dirty = *pdirty/2;
308 
309         *pdirty = dirty;
310 }
311 
312 /*
313  *
314  */
315 static unsigned int bdi_min_ratio;
316 
317 int bdi_set_min_ratio(struct backing_dev_info *bdi, unsigned int min_ratio)
318 {
319         int ret = 0;
320 
321         spin_lock_bh(&bdi_lock);
322         if (min_ratio > bdi->max_ratio) {
323                 ret = -EINVAL;
324         } else {
325                 min_ratio -= bdi->min_ratio;
326                 if (bdi_min_ratio + min_ratio < 100) {
327                         bdi_min_ratio += min_ratio;
328                         bdi->min_ratio += min_ratio;
329                 } else {
330                         ret = -EINVAL;
331                 }
332         }
333         spin_unlock_bh(&bdi_lock);
334 
335         return ret;
336 }
337 
338 int bdi_set_max_ratio(struct backing_dev_info *bdi, unsigned max_ratio)
339 {
340         int ret = 0;
341 
342         if (max_ratio > 100)
343                 return -EINVAL;
344 
345         spin_lock_bh(&bdi_lock);
346         if (bdi->min_ratio > max_ratio) {
347                 ret = -EINVAL;
348         } else {
349                 bdi->max_ratio = max_ratio;
350                 bdi->max_prop_frac = (PROP_FRAC_BASE * max_ratio) / 100;
351         }
352         spin_unlock_bh(&bdi_lock);
353 
354         return ret;
355 }
356 EXPORT_SYMBOL(bdi_set_max_ratio);
357 
358 /*
359  * Work out the current dirty-memory clamping and background writeout
360  * thresholds.
361  *
362  * The main aim here is to lower them aggressively if there is a lot of mapped
363  * memory around.  To avoid stressing page reclaim with lots of unreclaimable
364  * pages.  It is better to clamp down on writers than to start swapping, and
365  * performing lots of scanning.
366  *
367  * We only allow 1/2 of the currently-unmapped memory to be dirtied.
368  *
369  * We don't permit the clamping level to fall below 5% - that is getting rather
370  * excessive.
371  *
372  * We make sure that the background writeout level is below the adjusted
373  * clamping level.
374  */
375 
376 static unsigned long highmem_dirtyable_memory(unsigned long total)
377 {
378 #ifdef CONFIG_HIGHMEM
379         int node;
380         unsigned long x = 0;
381 
382         for_each_node_state(node, N_HIGH_MEMORY) {
383                 struct zone *z =
384                         &NODE_DATA(node)->node_zones[ZONE_HIGHMEM];
385 
386                 x += zone_page_state(z, NR_FREE_PAGES) +
387                      zone_reclaimable_pages(z);
388         }
389         /*
390          * Make sure that the number of highmem pages is never larger
391          * than the number of the total dirtyable memory. This can only
392          * occur in very strange VM situations but we want to make sure
393          * that this does not occur.
394          */
395         return min(x, total);
396 #else
397         return 0;
398 #endif
399 }
400 
401 /**
402  * determine_dirtyable_memory - amount of memory that may be used
403  *
404  * Returns the numebr of pages that can currently be freed and used
405  * by the kernel for direct mappings.
406  */
407 unsigned long determine_dirtyable_memory(void)
408 {
409         unsigned long x;
410 
411         x = global_page_state(NR_FREE_PAGES) + global_reclaimable_pages();
412 
413         if (!vm_highmem_is_dirtyable)
414                 x -= highmem_dirtyable_memory(x);
415 
416         return x + 1;   /* Ensure that we never return 0 */
417 }
418 
419 void
420 get_dirty_limits(unsigned long *pbackground, unsigned long *pdirty,
421                  unsigned long *pbdi_dirty, struct backing_dev_info *bdi)
422 {
423         unsigned long background;
424         unsigned long dirty;
425         unsigned long available_memory = determine_dirtyable_memory();
426         struct task_struct *tsk;
427 
428         if (vm_dirty_bytes)
429                 dirty = DIV_ROUND_UP(vm_dirty_bytes, PAGE_SIZE);
430         else {
431                 int dirty_ratio;
432 
433                 dirty_ratio = vm_dirty_ratio;
434                 if (dirty_ratio < 5)
435                         dirty_ratio = 5;
436                 dirty = (dirty_ratio * available_memory) / 100;
437         }
438 
439         if (dirty_background_bytes)
440                 background = DIV_ROUND_UP(dirty_background_bytes, PAGE_SIZE);
441         else
442                 background = (dirty_background_ratio * available_memory) / 100;
443 
444         if (background >= dirty)
445                 background = dirty / 2;
446         tsk = current;
447         if (tsk->flags & PF_LESS_THROTTLE || rt_task(tsk)) {
448                 background += background / 4;
449                 dirty += dirty / 4;
450         }
451         *pbackground = background;
452         *pdirty = dirty;
453 
454         if (bdi) {
455                 u64 bdi_dirty;
456                 long numerator, denominator;
457 
458                 /*
459                  * Calculate this BDI's share of the dirty ratio.
460                  */
461                 bdi_writeout_fraction(bdi, &numerator, &denominator);
462 
463                 bdi_dirty = (dirty * (100 - bdi_min_ratio)) / 100;
464                 bdi_dirty *= numerator;
465                 do_div(bdi_dirty, denominator);
466                 bdi_dirty += (dirty * bdi->min_ratio) / 100;
467                 if (bdi_dirty > (dirty * bdi->max_ratio) / 100)
468                         bdi_dirty = dirty * bdi->max_ratio / 100;
469 
470                 *pbdi_dirty = bdi_dirty;
471                 clip_bdi_dirty_limit(bdi, dirty, pbdi_dirty);
472                 task_dirty_limit(current, pbdi_dirty);
473         }
474 }
475 
476 /*
477  * balance_dirty_pages() must be called by processes which are generating dirty
478  * data.  It looks at the number of dirty pages in the machine and will force
479  * the caller to perform writeback if the system is over `vm_dirty_ratio'.
480  * If we're over `background_thresh' then the writeback threads are woken to
481  * perform some writeout.
482  */
483 static void balance_dirty_pages(struct address_space *mapping,
484                                 unsigned long write_chunk)
485 {
486         long nr_reclaimable, bdi_nr_reclaimable;
487         long nr_writeback, bdi_nr_writeback;
488         unsigned long background_thresh;
489         unsigned long dirty_thresh;
490         unsigned long bdi_thresh;
491         unsigned long pages_written = 0;
492         unsigned long pause = 1;
493 
494         struct backing_dev_info *bdi = mapping->backing_dev_info;
495 
496         for (;;) {
497                 struct writeback_control wbc = {
498                         .bdi            = bdi,
499                         .sync_mode      = WB_SYNC_NONE,
500                         .older_than_this = NULL,
501                         .nr_to_write    = write_chunk,
502                         .range_cyclic   = 1,
503                 };
504 
505                 get_dirty_limits(&background_thresh, &dirty_thresh,
506                                 &bdi_thresh, bdi);
507 
508                 nr_reclaimable = global_page_state(NR_FILE_DIRTY) +
509                                         global_page_state(NR_UNSTABLE_NFS);
510                 nr_writeback = global_page_state(NR_WRITEBACK);
511 
512                 bdi_nr_reclaimable = bdi_stat(bdi, BDI_RECLAIMABLE);
513                 bdi_nr_writeback = bdi_stat(bdi, BDI_WRITEBACK);
514 
515                 if (bdi_nr_reclaimable + bdi_nr_writeback <= bdi_thresh)
516                         break;
517 
518                 /*
519                  * Throttle it only when the background writeback cannot
520                  * catch-up. This avoids (excessively) small writeouts
521                  * when the bdi limits are ramping up.
522                  */
523                 if (nr_reclaimable + nr_writeback <
524                                 (background_thresh + dirty_thresh) / 2)
525                         break;
526 
527                 if (!bdi->dirty_exceeded)
528                         bdi->dirty_exceeded = 1;
529 
530                 /* Note: nr_reclaimable denotes nr_dirty + nr_unstable.
531                  * Unstable writes are a feature of certain networked
532                  * filesystems (i.e. NFS) in which data may have been
533                  * written to the server's write cache, but has not yet
534                  * been flushed to permanent storage.
535                  * Only move pages to writeback if this bdi is over its
536                  * threshold otherwise wait until the disk writes catch
537                  * up.
538                  */
539                 if (bdi_nr_reclaimable > bdi_thresh) {
540                         writeback_inodes_wbc(&wbc);
541                         pages_written += write_chunk - wbc.nr_to_write;
542                         get_dirty_limits(&background_thresh, &dirty_thresh,
543                                        &bdi_thresh, bdi);
544                 }
545 
546                 /*
547                  * In order to avoid the stacked BDI deadlock we need
548                  * to ensure we accurately count the 'dirty' pages when
549                  * the threshold is low.
550                  *
551                  * Otherwise it would be possible to get thresh+n pages
552                  * reported dirty, even though there are thresh-m pages
553                  * actually dirty; with m+n sitting in the percpu
554                  * deltas.
555                  */
556                 if (bdi_thresh < 2*bdi_stat_error(bdi)) {
557                         bdi_nr_reclaimable = bdi_stat_sum(bdi, BDI_RECLAIMABLE);
558                         bdi_nr_writeback = bdi_stat_sum(bdi, BDI_WRITEBACK);
559                 } else if (bdi_nr_reclaimable) {
560                         bdi_nr_reclaimable = bdi_stat(bdi, BDI_RECLAIMABLE);
561                         bdi_nr_writeback = bdi_stat(bdi, BDI_WRITEBACK);
562                 }
563 
564                 if (bdi_nr_reclaimable + bdi_nr_writeback <= bdi_thresh)
565                         break;
566                 if (pages_written >= write_chunk)
567                         break;          /* We've done our duty */
568 
569                 __set_current_state(TASK_INTERRUPTIBLE);
570                 io_schedule_timeout(pause);
571 
572                 /*
573                  * Increase the delay for each loop, up to our previous
574                  * default of taking a 100ms nap.
575                  */
576                 pause <<= 1;
577                 if (pause > HZ / 10)
578                         pause = HZ / 10;
579         }
580 
581         if (bdi_nr_reclaimable + bdi_nr_writeback < bdi_thresh &&
582                         bdi->dirty_exceeded)
583                 bdi->dirty_exceeded = 0;
584 
585         if (writeback_in_progress(bdi))
586                 return;
587 
588         /*
589          * In laptop mode, we wait until hitting the higher threshold before
590          * starting background writeout, and then write out all the way down
591          * to the lower threshold.  So slow writers cause minimal disk activity.
592          *
593          * In normal mode, we start background writeout at the lower
594          * background_thresh, to keep the amount of dirty memory low.
595          */
596         if ((laptop_mode && pages_written) ||
597             (!laptop_mode && ((global_page_state(NR_FILE_DIRTY)
598                                + global_page_state(NR_UNSTABLE_NFS))
599                                           > background_thresh)))
600                 bdi_start_writeback(bdi, NULL, 0);
601 }
602 
603 void set_page_dirty_balance(struct page *page, int page_mkwrite)
604 {
605         if (set_page_dirty(page) || page_mkwrite) {
606                 struct address_space *mapping = page_mapping(page);
607 
608                 if (mapping)
609                         balance_dirty_pages_ratelimited(mapping);
610         }
611 }
612 
613 static DEFINE_PER_CPU(unsigned long, bdp_ratelimits) = 0;
614 
615 /**
616  * balance_dirty_pages_ratelimited_nr - balance dirty memory state
617  * @mapping: address_space which was dirtied
618  * @nr_pages_dirtied: number of pages which the caller has just dirtied
619  *
620  * Processes which are dirtying memory should call in here once for each page
621  * which was newly dirtied.  The function will periodically check the system's
622  * dirty state and will initiate writeback if needed.
623  *
624  * On really big machines, get_writeback_state is expensive, so try to avoid
625  * calling it too often (ratelimiting).  But once we're over the dirty memory
626  * limit we decrease the ratelimiting by a lot, to prevent individual processes
627  * from overshooting the limit by (ratelimit_pages) each.
628  */
629 void balance_dirty_pages_ratelimited_nr(struct address_space *mapping,
630                                         unsigned long nr_pages_dirtied)
631 {
632         unsigned long ratelimit;
633         unsigned long *p;
634 
635         ratelimit = ratelimit_pages;
636         if (mapping->backing_dev_info->dirty_exceeded)
637                 ratelimit = 8;
638 
639         /*
640          * Check the rate limiting. Also, we do not want to throttle real-time
641          * tasks in balance_dirty_pages(). Period.
642          */
643         preempt_disable();
644         p =  &__get_cpu_var(bdp_ratelimits);
645         *p += nr_pages_dirtied;
646         if (unlikely(*p >= ratelimit)) {
647                 ratelimit = sync_writeback_pages(*p);
648                 *p = 0;
649                 preempt_enable();
650                 balance_dirty_pages(mapping, ratelimit);
651                 return;
652         }
653         preempt_enable();
654 }
655 EXPORT_SYMBOL(balance_dirty_pages_ratelimited_nr);
656 
657 void throttle_vm_writeout(gfp_t gfp_mask)
658 {
659         unsigned long background_thresh;
660         unsigned long dirty_thresh;
661 
662         for ( ; ; ) {
663                 get_dirty_limits(&background_thresh, &dirty_thresh, NULL, NULL);
664 
665                 /*
666                  * Boost the allowable dirty threshold a bit for page
667                  * allocators so they don't get DoS'ed by heavy writers
668                  */
669                 dirty_thresh += dirty_thresh / 10;      /* wheeee... */
670 
671                 if (global_page_state(NR_UNSTABLE_NFS) +
672                         global_page_state(NR_WRITEBACK) <= dirty_thresh)
673                                 break;
674                 congestion_wait(BLK_RW_ASYNC, HZ/10);
675 
676                 /*
677                  * The caller might hold locks which can prevent IO completion
678                  * or progress in the filesystem.  So we cannot just sit here
679                  * waiting for IO to complete.
680                  */
681                 if ((gfp_mask & (__GFP_FS|__GFP_IO)) != (__GFP_FS|__GFP_IO))
682                         break;
683         }
684 }
685 
686 static void laptop_timer_fn(unsigned long unused);
687 
688 static DEFINE_TIMER(laptop_mode_wb_timer, laptop_timer_fn, 0, 0);
689 
690 /*
691  * sysctl handler for /proc/sys/vm/dirty_writeback_centisecs
692  */
693 int dirty_writeback_centisecs_handler(ctl_table *table, int write,
694         void __user *buffer, size_t *length, loff_t *ppos)
695 {
696         proc_dointvec(table, write, buffer, length, ppos);
697         bdi_arm_supers_timer();
698         return 0;
699 }
700 
701 static void do_laptop_sync(struct work_struct *work)
702 {
703         wakeup_flusher_threads(0);
704         kfree(work);
705 }
706 
707 static void laptop_timer_fn(unsigned long unused)
708 {
709         struct work_struct *work;
710 
711         work = kmalloc(sizeof(*work), GFP_ATOMIC);
712         if (work) {
713                 INIT_WORK(work, do_laptop_sync);
714                 schedule_work(work);
715         }
716 }
717 
718 /*
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.
722  */
723 void laptop_io_completion(void)
724 {
725         mod_timer(&laptop_mode_wb_timer, jiffies + laptop_mode);
726 }
727 
728 /*
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.
732  */
733 void laptop_sync_completion(void)
734 {
735         del_timer(&laptop_mode_wb_timer);
736 }
737 
738 /*
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.
743  *
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.
747  *
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.
753  */
754 
755 void writeback_set_ratelimit(void)
756 {
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;
762 }
763 
764 static int __cpuinit
765 ratelimit_handler(struct notifier_block *self, unsigned long u, void *v)
766 {
767         writeback_set_ratelimit();
768         return NOTIFY_DONE;
769 }
770 
771 static struct notifier_block __cpuinitdata ratelimit_nb = {
772         .notifier_call  = ratelimit_handler,
773         .next           = NULL,
774 };
775 
776 /*
777  * Called early on to tune the page writeback dirty limits.
778  *
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.
782  *
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.
790  *
791  * But we might still want to scale the dirty_ratio by how
792  * much memory the box has..
793  */
794 void __init page_writeback_init(void)
795 {
796         int shift;
797 
798         writeback_set_ratelimit();
799         register_cpu_notifier(&ratelimit_nb);
800 
801         shift = calc_period_shift();
802         prop_descriptor_init(&vm_completions, shift);
803         prop_descriptor_init(&vm_dirties, shift);
804 }
805 
806 /**
807  * write_cache_pages - walk the list of dirty pages of the given address space and write all of them.
808  * @mapping: address space structure to write
809  * @wbc: subtract the number of written pages from *@wbc->nr_to_write
810  * @writepage: function called for each page
811  * @data: data passed to writepage function
812  *
813  * If a page is already under I/O, write_cache_pages() skips it, even
814  * if it's dirty.  This is desirable behaviour for memory-cleaning writeback,
815  * but it is INCORRECT for data-integrity system calls such as fsync().  fsync()
816  * and msync() need to guarantee that all the data which was dirty at the time
817  * the call was made get new I/O started against them.  If wbc->sync_mode is
818  * WB_SYNC_ALL then we were called for data integrity and we must wait for
819  * existing IO to complete.
820  */
821 int write_cache_pages(struct address_space *mapping,
822                       struct writeback_control *wbc, writepage_t writepage,
823                       void *data)
824 {
825         struct backing_dev_info *bdi = mapping->backing_dev_info;
826         int ret = 0;
827         int done = 0;
828         struct pagevec pvec;
829         int nr_pages;
830         pgoff_t uninitialized_var(writeback_index);
831         pgoff_t index;
832         pgoff_t end;            /* Inclusive */
833         pgoff_t done_index;
834         int cycled;
835         int range_whole = 0;
836         long nr_to_write = wbc->nr_to_write;
837 
838         if (wbc->nonblocking && bdi_write_congested(bdi)) {
839                 wbc->encountered_congestion = 1;
840                 return 0;
841         }
842 
843         pagevec_init(&pvec, 0);
844         if (wbc->range_cyclic) {
845                 writeback_index = mapping->writeback_index; /* prev offset */
846                 index = writeback_index;
847                 if (index == 0)
848                         cycled = 1;
849                 else
850                         cycled = 0;
851                 end = -1;
852         } else {
853                 index = wbc->range_start >> PAGE_CACHE_SHIFT;
854                 end = wbc->range_end >> PAGE_CACHE_SHIFT;
855                 if (wbc->range_start == 0 && wbc->range_end == LLONG_MAX)
856                         range_whole = 1;
857                 cycled = 1; /* ignore range_cyclic tests */
858         }
859 retry:
860         done_index = index;
861         while (!done && (index <= end)) {
862                 int i;
863 
864                 nr_pages = pagevec_lookup_tag(&pvec, mapping, &index,
865                               PAGECACHE_TAG_DIRTY,
866                               min(end - index, (pgoff_t)PAGEVEC_SIZE-1) + 1);
867                 if (nr_pages == 0)
868                         break;
869 
870                 for (i = 0; i < nr_pages; i++) {
871                         struct page *page = pvec.pages[i];
872 
873                         /*
874                          * At this point, the page may be truncated or
875                          * invalidated (changing page->mapping to NULL), or
876                          * even swizzled back from swapper_space to tmpfs file
877                          * mapping. However, page->index will not change
878                          * because we have a reference on the page.
879                          */
880                         if (page->index > end) {
881                                 /*
882                                  * can't be range_cyclic (1st pass) because
883                                  * end == -1 in that case.
884                                  */
885                                 done = 1;
886                                 break;
887                         }
888 
889                         done_index = page->index + 1;
890 
891                         lock_page(page);
892 
893                         /*
894                          * Page truncated or invalidated. We can freely skip it
895                          * then, even for data integrity operations: the page
896                          * has disappeared concurrently, so there could be no
897                          * real expectation of this data interity operation
898                          * even if there is now a new, dirty page at the same
899                          * pagecache address.
900                          */
901                         if (unlikely(page->mapping != mapping)) {
902 continue_unlock:
903                                 unlock_page(page);
904                                 continue;
905                         }
906 
907                         if (!PageDirty(page)) {
908                                 /* someone wrote it for us */
909                                 goto continue_unlock;
910                         }
911 
912                         if (PageWriteback(page)) {
913                                 if (wbc->sync_mode != WB_SYNC_NONE)
914                                         wait_on_page_writeback(page);
915                                 else
916                                         goto continue_unlock;
917                         }
918 
919                         BUG_ON(PageWriteback(page));
920                         if (!clear_page_dirty_for_io(page))
921                                 goto continue_unlock;
922 
923                         ret = (*writepage)(page, wbc, data);
924                         if (unlikely(ret)) {
925                                 if (ret == AOP_WRITEPAGE_ACTIVATE) {
926                                         unlock_page(page);
927                                         ret = 0;
928                                 } else {
929                                         /*
930                                          * done_index is set past this page,
931                                          * so media errors will not choke
932                                          * background writeout for the entire
933                                          * file. This has consequences for
934                                          * range_cyclic semantics (ie. it may
935                                          * not be suitable for data integrity
936                                          * writeout).
937                                          */
938                                         done = 1;
939                                         break;
940                                 }
941                         }
942 
943                         if (nr_to_write > 0) {
944                                 nr_to_write--;
945                                 if (nr_to_write == 0 &&
946                                     wbc->sync_mode == WB_SYNC_NONE) {
947                                         /*
948                                          * We stop writing back only if we are
949                                          * not doing integrity sync. In case of
950                                          * integrity sync we have to keep going
951                                          * because someone may be concurrently
952                                          * dirtying pages, and we might have
953                                          * synced a lot of newly appeared dirty
954                                          * pages, but have not synced all of the
955                                          * old dirty pages.
956                                          */
957                                         done = 1;
958                                         break;
959                                 }
960                         }
961 
962                         if (wbc->nonblocking && bdi_write_congested(bdi)) {
963                                 wbc->encountered_congestion = 1;
964                                 done = 1;
965                                 break;
966                         }
967                 }
968                 pagevec_release(&pvec);
969                 cond_resched();
970         }
971         if (!cycled && !done) {
972                 /*
973                  * range_cyclic:
974                  * We hit the last page and there is more work to be done: wrap
975                  * back to the start of the file
976                  */
977                 cycled = 1;
978                 index = 0;
979                 end = writeback_index - 1;
980                 goto retry;
981         }
982         if (!wbc->no_nrwrite_index_update) {
983                 if (wbc->range_cyclic || (range_whole && nr_to_write > 0))
984                         mapping->writeback_index = done_index;
985                 wbc->nr_to_write = nr_to_write;
986         }
987 
988         return ret;
989 }
990 EXPORT_SYMBOL(write_cache_pages);
991 
992 /*
993  * Function used by generic_writepages to call the real writepage
994  * function and set the mapping flags on error
995  */
996 static int __writepage(struct page *page, struct writeback_control *wbc,
997                        void *data)
998 {
999         struct address_space *mapping = data;
1000         int ret = mapping->a_ops->writepage(page, wbc);
1001         mapping_set_error(mapping, ret);
1002         return ret;
1003 }
1004 
1005 /**
1006  * generic_writepages - walk the list of dirty pages of the given address space and writepage() all of them.
1007  * @mapping: address space structure to write
1008  * @wbc: subtract the number of written pages from *@wbc->nr_to_write
1009  *
1010  * This is a library function, which implements the writepages()
1011  * address_space_operation.
1012  */
1013 int generic_writepages(struct address_space *mapping,
1014                        struct writeback_control *wbc)
1015 {
1016         /* deal with chardevs and other special file */
1017         if (!mapping->a_ops->writepage)
1018                 return 0;
1019 
1020         return write_cache_pages(mapping, wbc, __writepage, mapping);
1021 }
1022 
1023 EXPORT_SYMBOL(generic_writepages);
1024 
1025 int do_writepages(struct address_space *mapping, struct writeback_control *wbc)
1026 {
1027         int ret;
1028 
1029         if (wbc->nr_to_write <= 0)
1030                 return 0;
1031         if (mapping->a_ops->writepages)
1032                 ret = mapping->a_ops->writepages(mapping, wbc);
1033         else
1034                 ret = generic_writepages(mapping, wbc);
1035         return ret;
1036 }
1037 
1038 /**
1039  * write_one_page - write out a single page and optionally wait on I/O
1040  * @page: the page to write
1041  * @wait: if true, wait on writeout
1042  *
1043  * The page must be locked by the caller and will be unlocked upon return.
1044  *
1045  * write_one_page() returns a negative error code if I/O failed.
1046  */
1047 int write_one_page(struct page *page, int wait)
1048 {
1049         struct address_space *mapping = page->mapping;
1050         int ret = 0;
1051         struct writeback_control wbc = {
1052                 .sync_mode = WB_SYNC_ALL,
1053                 .nr_to_write = 1,
1054         };
1055 
1056         BUG_ON(!PageLocked(page));
1057 
1058         if (wait)
1059                 wait_on_page_writeback(page);
1060 
1061         if (clear_page_dirty_for_io(page)) {
1062                 page_cache_get(page);
1063                 ret = mapping->a_ops->writepage(page, &wbc);
1064                 if (ret == 0 && wait) {
1065                         wait_on_page_writeback(page);
1066                         if (PageError(page))
1067                                 ret = -EIO;
1068                 }
1069                 page_cache_release(page);
1070         } else {
1071                 unlock_page(page);
1072         }
1073         return ret;
1074 }
1075 EXPORT_SYMBOL(write_one_page);
1076 
1077 /*
1078  * For address_spaces which do not use buffers nor write back.
1079  */
1080 int __set_page_dirty_no_writeback(struct page *page)
1081 {
1082         if (!PageDirty(page))
1083                 SetPageDirty(page);
1084         return 0;
1085 }
1086 
1087 /*
1088  * Helper function for set_page_dirty family.
1089  * NOTE: This relies on being atomic wrt interrupts.
1090  */
1091 void account_page_dirtied(struct page *page, struct address_space *mapping)
1092 {
1093         if (mapping_cap_account_dirty(mapping)) {
1094                 __inc_zone_page_state(page, NR_FILE_DIRTY);
1095                 __inc_bdi_stat(mapping->backing_dev_info, BDI_RECLAIMABLE);
1096                 task_dirty_inc(current);
1097                 task_io_account_write(PAGE_CACHE_SIZE);
1098         }
1099 }
1100 
1101 /*
1102  * For address_spaces which do not use buffers.  Just tag the page as dirty in
1103  * its radix tree.
1104  *
1105  * This is also used when a single buffer is being dirtied: we want to set the
1106  * page dirty in that case, but not all the buffers.  This is a "bottom-up"
1107  * dirtying, whereas __set_page_dirty_buffers() is a "top-down" dirtying.
1108  *
1109  * Most callers have locked the page, which pins the address_space in memory.
1110  * But zap_pte_range() does not lock the page, however in that case the
1111  * mapping is pinned by the vma's ->vm_file reference.
1112  *
1113  * We take care to handle the case where the page was truncated from the
1114  * mapping by re-checking page_mapping() inside tree_lock.
1115  */
1116 int __set_page_dirty_nobuffers(struct page *page)
1117 {
1118         if (!TestSetPageDirty(page)) {
1119                 struct address_space *mapping = page_mapping(page);
1120                 struct address_space *mapping2;
1121 
1122                 if (!mapping)
1123                         return 1;
1124 
1125                 spin_lock_irq(&mapping->tree_lock);
1126                 mapping2 = page_mapping(page);
1127                 if (mapping2) { /* Race with truncate? */
1128                         BUG_ON(mapping2 != mapping);
1129                         WARN_ON_ONCE(!PagePrivate(page) && !PageUptodate(page));
1130                         account_page_dirtied(page, mapping);
1131                         radix_tree_tag_set(&mapping->page_tree,
1132                                 page_index(page), PAGECACHE_TAG_DIRTY);
1133                 }
1134                 spin_unlock_irq(&mapping->tree_lock);
1135                 if (mapping->host) {
1136                         /* !PageAnon && !swapper_space */
1137                         __mark_inode_dirty(mapping->host, I_DIRTY_PAGES);
1138                 }
1139                 return 1;
1140         }
1141         return 0;
1142 }
1143 EXPORT_SYMBOL(__set_page_dirty_nobuffers);
1144 
1145 /*
1146  * When a writepage implementation decides that it doesn't want to write this
1147  * page for some reason, it should redirty the locked page via
1148  * redirty_page_for_writepage() and it should then unlock the page and return 0
1149  */
1150 int redirty_page_for_writepage(struct writeback_control *wbc, struct page *page)
1151 {
1152         wbc->pages_skipped++;
1153         return __set_page_dirty_nobuffers(page);
1154 }
1155 EXPORT_SYMBOL(redirty_page_for_writepage);
1156 
1157 /*
1158  * Dirty a page.
1159  *
1160  * For pages with a mapping this should be done under the page lock
1161  * for the benefit of asynchronous memory errors who prefer a consistent
1162  * dirty state. This rule can be broken in some special cases,
1163  * but should be better not to.
1164  *
1165  * If the mapping doesn't provide a set_page_dirty a_op, then
1166  * just fall through and assume that it wants buffer_heads.
1167  */
1168 int set_page_dirty(struct page *page)
1169 {
1170         struct address_space *mapping = page_mapping(page);
1171 
1172         if (likely(mapping)) {
1173                 int (*spd)(struct page *) = mapping->a_ops->set_page_dirty;
1174 #ifdef CONFIG_BLOCK
1175                 if (!spd)
1176                         spd = __set_page_dirty_buffers;
1177 #endif
1178                 return (*spd)(page);
1179         }
1180         if (!PageDirty(page)) {
1181                 if (!TestSetPageDirty(page))
1182                         return 1;
1183         }
1184         return 0;
1185 }
1186 EXPORT_SYMBOL(set_page_dirty);
1187 
1188 /*
1189  * set_page_dirty() is racy if the caller has no reference against
1190  * page->mapping->host, and if the page is unlocked.  This is because another
1191  * CPU could truncate the page off the mapping and then free the mapping.
1192  *
1193  * Usually, the page _is_ locked, or the caller is a user-space process which
1194  * holds a reference on the inode by having an open file.
1195  *
1196  * In other cases, the page should be locked before running set_page_dirty().
1197  */
1198 int set_page_dirty_lock(struct page *page)
1199 {
1200         int ret;
1201 
1202         lock_page_nosync(page);
1203         ret = set_page_dirty(page);
1204         unlock_page(page);
1205         return ret;
1206 }
1207 EXPORT_SYMBOL(set_page_dirty_lock);
1208 
1209 /*
1210  * Clear a page's dirty flag, while caring for dirty memory accounting.
1211  * Returns true if the page was previously dirty.
1212  *
1213  * This is for preparing to put the page under writeout.  We leave the page
1214  * tagged as dirty in the radix tree so that a concurrent write-for-sync
1215  * can discover it via a PAGECACHE_TAG_DIRTY walk.  The ->writepage
1216  * implementation will run either set_page_writeback() or set_page_dirty(),
1217  * at which stage we bring the page's dirty flag and radix-tree dirty tag
1218  * back into sync.
1219  *
1220  * This incoherency between the page's dirty flag and radix-tree tag is
1221  * unfortunate, but it only exists while the page is locked.
1222  */
1223 int clear_page_dirty_for_io(struct page *page)
1224 {
1225         struct address_space *mapping = page_mapping(page);
1226 
1227         BUG_ON(!PageLocked(page));
1228 
1229         ClearPageReclaim(page);
1230         if (mapping && mapping_cap_account_dirty(mapping)) {
1231                 /*
1232                  * Yes, Virginia, this is indeed insane.
1233                  *
1234                  * We use this sequence to make sure that
1235                  *  (a) we account for dirty stats properly
1236                  *  (b) we tell the low-level filesystem to
1237                  *      mark the whole page dirty if it was
1238                  *      dirty in a pagetable. Only to then
1239                  *  (c) clean the page again and return 1 to
1240                  *      cause the writeback.
1241                  *
1242                  * This way we avoid all nasty races with the
1243                  * dirty bit in multiple places and clearing
1244                  * them concurrently from different threads.
1245                  *
1246                  * Note! Normally the "set_page_dirty(page)"
1247                  * has no effect on the actual dirty bit - since
1248                  * that will already usually be set. But we
1249                  * need the side effects, and it can help us
1250                  * avoid races.
1251                  *
1252                  * We basically use the page "master dirty bit"
1253                  * as a serialization point for all the different
1254                  * threads doing their things.
1255                  */
1256                 if (page_mkclean(page))
1257                         set_page_dirty(page);
1258                 /*
1259                  * We carefully synchronise fault handlers against
1260                  * installing a dirty pte and marking the page dirty
1261                  * at this point. We do this by having them hold the
1262                  * page lock at some point after installing their
1263                  * pte, but before marking the page dirty.
1264                  * Pages are always locked coming in here, so we get
1265                  * the desired exclusion. See mm/memory.c:do_wp_page()
1266                  * for more comments.
1267                  */
1268                 if (TestClearPageDirty(page)) {
1269                         dec_zone_page_state(page, NR_FILE_DIRTY);
1270                         dec_bdi_stat(mapping->backing_dev_info,
1271                                         BDI_RECLAIMABLE);
1272                         return 1;
1273                 }
1274                 return 0;
1275         }
1276         return TestClearPageDirty(page);
1277 }
1278 EXPORT_SYMBOL(clear_page_dirty_for_io);
1279 
1280 int test_clear_page_writeback(struct page *page)
1281 {
1282         struct address_space *mapping = page_mapping(page);
1283         int ret;
1284 
1285         if (mapping) {
1286                 struct backing_dev_info *bdi = mapping->backing_dev_info;
1287                 unsigned long flags;
1288 
1289                 spin_lock_irqsave(&mapping->tree_lock, flags);
1290                 ret = TestClearPageWriteback(page);
1291                 if (ret) {
1292                         radix_tree_tag_clear(&mapping->page_tree,
1293                                                 page_index(page),
1294                                                 PAGECACHE_TAG_WRITEBACK);
1295                         if (bdi_cap_account_writeback(bdi)) {
1296                                 __dec_bdi_stat(bdi, BDI_WRITEBACK);
1297                                 __bdi_writeout_inc(bdi);
1298                         }
1299                 }
1300                 spin_unlock_irqrestore(&mapping->tree_lock, flags);
1301         } else {
1302                 ret = TestClearPageWriteback(page);
1303         }
1304         if (ret)
1305                 dec_zone_page_state(page, NR_WRITEBACK);
1306         return ret;
1307 }
1308 
1309 int test_set_page_writeback(struct page *page)
1310 {
1311         struct address_space *mapping = page_mapping(page);
1312         int ret;
1313 
1314         if (mapping) {
1315                 struct backing_dev_info *bdi = mapping->backing_dev_info;
1316                 unsigned long flags;
1317 
1318                 spin_lock_irqsave(&mapping->tree_lock, flags);
1319                 ret = TestSetPageWriteback(page);
1320                 if (!ret) {
1321                         radix_tree_tag_set(&mapping->page_tree,
1322                                                 page_index(page),
1323                                                 PAGECACHE_TAG_WRITEBACK);
1324                         if (bdi_cap_account_writeback(bdi))
1325                                 __inc_bdi_stat(bdi, BDI_WRITEBACK);
1326                 }
1327                 if (!PageDirty(page))
1328                         radix_tree_tag_clear(&mapping->page_tree,
1329                                                 page_index(page),
1330                                                 PAGECACHE_TAG_DIRTY);
1331                 spin_unlock_irqrestore(&mapping->tree_lock, flags);
1332         } else {
1333                 ret = TestSetPageWriteback(page);
1334         }
1335         if (!ret)
1336                 inc_zone_page_state(page, NR_WRITEBACK);
1337         return ret;
1338 
1339 }
1340 EXPORT_SYMBOL(test_set_page_writeback);
1341 
1342 /*
1343  * Return true if any of the pages in the mapping are marked with the
1344  * passed tag.
1345  */
1346 int mapping_tagged(struct address_space *mapping, int tag)
1347 {
1348         int ret;
1349         rcu_read_lock();
1350         ret = radix_tree_tagged(&mapping->page_tree, tag);
1351         rcu_read_unlock();
1352         return ret;
1353 }
1354 EXPORT_SYMBOL(mapping_tagged);
1355 

~ [ source navigation ] ~ [ diff markup ] ~ [ identifier search ] ~

kernel.org | git.kernel.org | LWN.net | Project Home | Wiki (Japanese) | Wiki (English) | SVN repository | Mail admin

Linux® is a registered trademark of Linus Torvalds in the United States and other countries.
TOMOYO® is a registered trademark of NTT DATA CORPORATION.

osdn.jp