blob: 6106a5c7ed44d0f8f134a32d4e21834835e7f4c3 [file] [log] [blame]
Linus Torvalds1da177e2005-04-16 15:20:36 -07001/*
Uwe Zeisbergerf30c2262006-10-03 23:01:26 +02002 * mm/page-writeback.c
Linus Torvalds1da177e2005-04-16 15:20:36 -07003 *
4 * Copyright (C) 2002, Linus Torvalds.
Peter Zijlstra04fbfdc2007-10-16 23:25:50 -07005 * Copyright (C) 2007 Red Hat, Inc., Peter Zijlstra <pzijlstr@redhat.com>
Linus Torvalds1da177e2005-04-16 15:20:36 -07006 *
7 * Contains functions related to writing back dirty pages at the
8 * address_space level.
9 *
Francois Camie1f8e872008-10-15 22:01:59 -070010 * 10Apr2002 Andrew Morton
Linus Torvalds1da177e2005-04-16 15:20:36 -070011 * 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>
Andrew Morton55e829a2006-12-10 02:19:27 -080025#include <linux/task_io_accounting_ops.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070026#include <linux/blkdev.h>
27#include <linux/mpage.h>
Peter Zijlstrad08b3852006-09-25 23:30:57 -070028#include <linux/rmap.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070029#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>
David Howellscf9a2ae2006-08-29 19:05:54 +010035#include <linux/buffer_head.h>
David Howells811d7362006-08-29 19:06:09 +010036#include <linux/pagevec.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070037
38/*
39 * The maximum number of pages to writeout in a single bdflush/kupdate
Joern Engel1c0eeaf2007-10-16 23:30:44 -070040 * operation. We do this so we don't hold I_SYNC against an inode for
Linus Torvalds1da177e2005-04-16 15:20:36 -070041 * 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.
44 */
45#define MAX_WRITEBACK_PAGES 1024
46
47/*
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.
50 */
51static long ratelimit_pages = 32;
52
Linus Torvalds1da177e2005-04-16 15:20:36 -070053/*
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.
58 */
59static inline long sync_writeback_pages(void)
60{
61 return ratelimit_pages + ratelimit_pages / 2;
62}
63
64/* The following parameters are exported via /proc/sys/vm */
65
66/*
67 * Start background writeback (via pdflush) at this percentage
68 */
Linus Torvalds07db59b2007-04-27 09:10:47 -070069int dirty_background_ratio = 5;
Linus Torvalds1da177e2005-04-16 15:20:36 -070070
71/*
David Rientjes2da02992009-01-06 14:39:31 -080072 * dirty_background_bytes starts at 0 (disabled) so that it is a function of
73 * dirty_background_ratio * the amount of dirtyable memory
74 */
75unsigned long dirty_background_bytes;
76
77/*
Bron Gondwana195cf452008-02-04 22:29:20 -080078 * free highmem will not be subtracted from the total free memory
79 * for calculating free ratios if vm_highmem_is_dirtyable is true
80 */
81int vm_highmem_is_dirtyable;
82
83/*
Linus Torvalds1da177e2005-04-16 15:20:36 -070084 * The generator of dirty data starts writeback at this percentage
85 */
Linus Torvalds07db59b2007-04-27 09:10:47 -070086int vm_dirty_ratio = 10;
Linus Torvalds1da177e2005-04-16 15:20:36 -070087
88/*
David Rientjes2da02992009-01-06 14:39:31 -080089 * vm_dirty_bytes starts at 0 (disabled) so that it is a function of
90 * vm_dirty_ratio * the amount of dirtyable memory
91 */
92unsigned long vm_dirty_bytes;
93
94/*
Coywolf Qi Huntfd5403c2006-04-10 22:54:35 -070095 * The interval between `kupdate'-style writebacks, in jiffies
Linus Torvalds1da177e2005-04-16 15:20:36 -070096 */
Bart Samwelf6ef9432006-03-24 03:15:48 -080097int dirty_writeback_interval = 5 * HZ;
Linus Torvalds1da177e2005-04-16 15:20:36 -070098
99/*
Coywolf Qi Huntfd5403c2006-04-10 22:54:35 -0700100 * The longest number of jiffies for which data is allowed to remain dirty
Linus Torvalds1da177e2005-04-16 15:20:36 -0700101 */
Bart Samwelf6ef9432006-03-24 03:15:48 -0800102int dirty_expire_interval = 30 * HZ;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700103
104/*
105 * Flag that makes the machine dump writes/reads and block dirtyings.
106 */
107int block_dump;
108
109/*
Bart Samweled5b43f2006-03-24 03:15:49 -0800110 * Flag that puts the machine in "laptop mode". Doubles as a timeout in jiffies:
111 * a full sync is triggered after this time elapses without any disk activity.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700112 */
113int laptop_mode;
114
115EXPORT_SYMBOL(laptop_mode);
116
117/* End of sysctl-exported parameters */
118
119
120static void background_writeout(unsigned long _min_pages);
121
Linus Torvalds1da177e2005-04-16 15:20:36 -0700122/*
Peter Zijlstra04fbfdc2007-10-16 23:25:50 -0700123 * Scale the writeback cache size proportional to the relative writeout speeds.
124 *
125 * We do this by keeping a floating proportion between BDIs, based on page
126 * writeback completions [end_page_writeback()]. Those devices that write out
127 * pages fastest will get the larger share, while the slower will get a smaller
128 * share.
129 *
130 * We use page writeout completions because we are interested in getting rid of
131 * dirty pages. Having them written out is the primary goal.
132 *
133 * We introduce a concept of time, a period over which we measure these events,
134 * because demand can/will vary over time. The length of this period itself is
135 * measured in page writeback completions.
136 *
137 */
138static struct prop_descriptor vm_completions;
Peter Zijlstra3e26c142007-10-16 23:25:50 -0700139static struct prop_descriptor vm_dirties;
Peter Zijlstra04fbfdc2007-10-16 23:25:50 -0700140
Peter Zijlstra04fbfdc2007-10-16 23:25:50 -0700141/*
142 * couple the period to the dirty_ratio:
143 *
144 * period/2 ~ roundup_pow_of_two(dirty limit)
145 */
146static int calc_period_shift(void)
147{
148 unsigned long dirty_total;
149
David Rientjes2da02992009-01-06 14:39:31 -0800150 if (vm_dirty_bytes)
151 dirty_total = vm_dirty_bytes / PAGE_SIZE;
152 else
153 dirty_total = (vm_dirty_ratio * determine_dirtyable_memory()) /
154 100;
Peter Zijlstra04fbfdc2007-10-16 23:25:50 -0700155 return 2 + ilog2(dirty_total - 1);
156}
157
158/*
David Rientjes2da02992009-01-06 14:39:31 -0800159 * update the period when the dirty threshold changes.
Peter Zijlstra04fbfdc2007-10-16 23:25:50 -0700160 */
David Rientjes2da02992009-01-06 14:39:31 -0800161static void update_completion_period(void)
162{
163 int shift = calc_period_shift();
164 prop_change_shift(&vm_completions, shift);
165 prop_change_shift(&vm_dirties, shift);
166}
167
168int dirty_background_ratio_handler(struct ctl_table *table, int write,
169 struct file *filp, void __user *buffer, size_t *lenp,
170 loff_t *ppos)
171{
172 int ret;
173
174 ret = proc_dointvec_minmax(table, write, filp, buffer, lenp, ppos);
175 if (ret == 0 && write)
176 dirty_background_bytes = 0;
177 return ret;
178}
179
180int dirty_background_bytes_handler(struct ctl_table *table, int write,
181 struct file *filp, void __user *buffer, size_t *lenp,
182 loff_t *ppos)
183{
184 int ret;
185
186 ret = proc_doulongvec_minmax(table, write, filp, buffer, lenp, ppos);
187 if (ret == 0 && write)
188 dirty_background_ratio = 0;
189 return ret;
190}
191
Peter Zijlstra04fbfdc2007-10-16 23:25:50 -0700192int dirty_ratio_handler(struct ctl_table *table, int write,
193 struct file *filp, void __user *buffer, size_t *lenp,
194 loff_t *ppos)
195{
196 int old_ratio = vm_dirty_ratio;
David Rientjes2da02992009-01-06 14:39:31 -0800197 int ret;
198
199 ret = proc_dointvec_minmax(table, write, filp, buffer, lenp, ppos);
Peter Zijlstra04fbfdc2007-10-16 23:25:50 -0700200 if (ret == 0 && write && vm_dirty_ratio != old_ratio) {
David Rientjes2da02992009-01-06 14:39:31 -0800201 update_completion_period();
202 vm_dirty_bytes = 0;
203 }
204 return ret;
205}
206
207
208int dirty_bytes_handler(struct ctl_table *table, int write,
209 struct file *filp, void __user *buffer, size_t *lenp,
210 loff_t *ppos)
211{
Sven Wegenerfc3501d2009-02-11 13:04:23 -0800212 unsigned long old_bytes = vm_dirty_bytes;
David Rientjes2da02992009-01-06 14:39:31 -0800213 int ret;
214
215 ret = proc_doulongvec_minmax(table, write, filp, buffer, lenp, ppos);
216 if (ret == 0 && write && vm_dirty_bytes != old_bytes) {
217 update_completion_period();
218 vm_dirty_ratio = 0;
Peter Zijlstra04fbfdc2007-10-16 23:25:50 -0700219 }
220 return ret;
221}
222
223/*
224 * Increment the BDI's writeout completion count and the global writeout
225 * completion count. Called from test_clear_page_writeback().
226 */
227static inline void __bdi_writeout_inc(struct backing_dev_info *bdi)
228{
Peter Zijlstraa42dde02008-04-30 00:54:36 -0700229 __prop_inc_percpu_max(&vm_completions, &bdi->completions,
230 bdi->max_prop_frac);
Peter Zijlstra04fbfdc2007-10-16 23:25:50 -0700231}
232
Miklos Szeredidd5656e2008-04-30 00:54:37 -0700233void bdi_writeout_inc(struct backing_dev_info *bdi)
234{
235 unsigned long flags;
236
237 local_irq_save(flags);
238 __bdi_writeout_inc(bdi);
239 local_irq_restore(flags);
240}
241EXPORT_SYMBOL_GPL(bdi_writeout_inc);
242
Peter Zijlstra3e26c142007-10-16 23:25:50 -0700243static inline void task_dirty_inc(struct task_struct *tsk)
244{
245 prop_inc_single(&vm_dirties, &tsk->dirties);
246}
247
Peter Zijlstra04fbfdc2007-10-16 23:25:50 -0700248/*
249 * Obtain an accurate fraction of the BDI's portion.
250 */
251static void bdi_writeout_fraction(struct backing_dev_info *bdi,
252 long *numerator, long *denominator)
253{
254 if (bdi_cap_writeback_dirty(bdi)) {
255 prop_fraction_percpu(&vm_completions, &bdi->completions,
256 numerator, denominator);
257 } else {
258 *numerator = 0;
259 *denominator = 1;
260 }
261}
262
263/*
264 * Clip the earned share of dirty pages to that which is actually available.
265 * This avoids exceeding the total dirty_limit when the floating averages
266 * fluctuate too quickly.
267 */
268static void
269clip_bdi_dirty_limit(struct backing_dev_info *bdi, long dirty, long *pbdi_dirty)
270{
271 long avail_dirty;
272
273 avail_dirty = dirty -
274 (global_page_state(NR_FILE_DIRTY) +
275 global_page_state(NR_WRITEBACK) +
Miklos Szeredifc3ba692008-04-30 00:54:38 -0700276 global_page_state(NR_UNSTABLE_NFS) +
277 global_page_state(NR_WRITEBACK_TEMP));
Peter Zijlstra04fbfdc2007-10-16 23:25:50 -0700278
279 if (avail_dirty < 0)
280 avail_dirty = 0;
281
282 avail_dirty += bdi_stat(bdi, BDI_RECLAIMABLE) +
283 bdi_stat(bdi, BDI_WRITEBACK);
284
285 *pbdi_dirty = min(*pbdi_dirty, avail_dirty);
286}
287
Peter Zijlstra3e26c142007-10-16 23:25:50 -0700288static inline void task_dirties_fraction(struct task_struct *tsk,
289 long *numerator, long *denominator)
290{
291 prop_fraction_single(&vm_dirties, &tsk->dirties,
292 numerator, denominator);
293}
294
295/*
296 * scale the dirty limit
297 *
298 * task specific dirty limit:
299 *
300 * dirty -= (dirty/8) * p_{t}
301 */
Adrian Bunkf61eaf92008-02-04 22:29:08 -0800302static void task_dirty_limit(struct task_struct *tsk, long *pdirty)
Peter Zijlstra3e26c142007-10-16 23:25:50 -0700303{
304 long numerator, denominator;
305 long dirty = *pdirty;
306 u64 inv = dirty >> 3;
307
308 task_dirties_fraction(tsk, &numerator, &denominator);
309 inv *= numerator;
310 do_div(inv, denominator);
311
312 dirty -= inv;
313 if (dirty < *pdirty/2)
314 dirty = *pdirty/2;
315
316 *pdirty = dirty;
317}
318
Peter Zijlstra04fbfdc2007-10-16 23:25:50 -0700319/*
Peter Zijlstra189d3c42008-04-30 00:54:35 -0700320 *
321 */
322static DEFINE_SPINLOCK(bdi_lock);
323static unsigned int bdi_min_ratio;
324
325int bdi_set_min_ratio(struct backing_dev_info *bdi, unsigned int min_ratio)
326{
327 int ret = 0;
328 unsigned long flags;
329
330 spin_lock_irqsave(&bdi_lock, flags);
Peter Zijlstraa42dde02008-04-30 00:54:36 -0700331 if (min_ratio > bdi->max_ratio) {
Peter Zijlstra189d3c42008-04-30 00:54:35 -0700332 ret = -EINVAL;
Peter Zijlstraa42dde02008-04-30 00:54:36 -0700333 } else {
334 min_ratio -= bdi->min_ratio;
335 if (bdi_min_ratio + min_ratio < 100) {
336 bdi_min_ratio += min_ratio;
337 bdi->min_ratio += min_ratio;
338 } else {
339 ret = -EINVAL;
340 }
341 }
Peter Zijlstra189d3c42008-04-30 00:54:35 -0700342 spin_unlock_irqrestore(&bdi_lock, flags);
343
344 return ret;
345}
346
Peter Zijlstraa42dde02008-04-30 00:54:36 -0700347int bdi_set_max_ratio(struct backing_dev_info *bdi, unsigned max_ratio)
348{
349 unsigned long flags;
350 int ret = 0;
351
352 if (max_ratio > 100)
353 return -EINVAL;
354
355 spin_lock_irqsave(&bdi_lock, flags);
356 if (bdi->min_ratio > max_ratio) {
357 ret = -EINVAL;
358 } else {
359 bdi->max_ratio = max_ratio;
360 bdi->max_prop_frac = (PROP_FRAC_BASE * max_ratio) / 100;
361 }
362 spin_unlock_irqrestore(&bdi_lock, flags);
363
364 return ret;
365}
366EXPORT_SYMBOL(bdi_set_max_ratio);
367
Peter Zijlstra189d3c42008-04-30 00:54:35 -0700368/*
Linus Torvalds1da177e2005-04-16 15:20:36 -0700369 * Work out the current dirty-memory clamping and background writeout
370 * thresholds.
371 *
372 * The main aim here is to lower them aggressively if there is a lot of mapped
373 * memory around. To avoid stressing page reclaim with lots of unreclaimable
374 * pages. It is better to clamp down on writers than to start swapping, and
375 * performing lots of scanning.
376 *
377 * We only allow 1/2 of the currently-unmapped memory to be dirtied.
378 *
379 * We don't permit the clamping level to fall below 5% - that is getting rather
380 * excessive.
381 *
382 * We make sure that the background writeout level is below the adjusted
383 * clamping level.
384 */
Christoph Lameter1b424462007-05-06 14:48:59 -0700385
386static unsigned long highmem_dirtyable_memory(unsigned long total)
387{
388#ifdef CONFIG_HIGHMEM
389 int node;
390 unsigned long x = 0;
391
Lee Schermerhorn37b07e42007-10-16 01:25:39 -0700392 for_each_node_state(node, N_HIGH_MEMORY) {
Christoph Lameter1b424462007-05-06 14:48:59 -0700393 struct zone *z =
394 &NODE_DATA(node)->node_zones[ZONE_HIGHMEM];
395
Rik van Riel4f98a2f2008-10-18 20:26:32 -0700396 x += zone_page_state(z, NR_FREE_PAGES) + zone_lru_pages(z);
Christoph Lameter1b424462007-05-06 14:48:59 -0700397 }
398 /*
399 * Make sure that the number of highmem pages is never larger
400 * than the number of the total dirtyable memory. This can only
401 * occur in very strange VM situations but we want to make sure
402 * that this does not occur.
403 */
404 return min(x, total);
405#else
406 return 0;
407#endif
408}
409
Steven Rostedt3eefae92008-05-12 21:21:04 +0200410/**
411 * determine_dirtyable_memory - amount of memory that may be used
412 *
413 * Returns the numebr of pages that can currently be freed and used
414 * by the kernel for direct mappings.
415 */
416unsigned long determine_dirtyable_memory(void)
Christoph Lameter1b424462007-05-06 14:48:59 -0700417{
418 unsigned long x;
419
Rik van Riel4f98a2f2008-10-18 20:26:32 -0700420 x = global_page_state(NR_FREE_PAGES) + global_lru_pages();
Bron Gondwana195cf452008-02-04 22:29:20 -0800421
422 if (!vm_highmem_is_dirtyable)
423 x -= highmem_dirtyable_memory(x);
424
Christoph Lameter1b424462007-05-06 14:48:59 -0700425 return x + 1; /* Ensure that we never return 0 */
426}
427
Peter Zijlstracf0ca9f2008-04-30 00:54:32 -0700428void
David Rientjes364aeb22009-01-06 14:39:29 -0800429get_dirty_limits(unsigned long *pbackground, unsigned long *pdirty,
430 unsigned long *pbdi_dirty, struct backing_dev_info *bdi)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700431{
David Rientjes364aeb22009-01-06 14:39:29 -0800432 unsigned long background;
433 unsigned long dirty;
Christoph Lameter1b424462007-05-06 14:48:59 -0700434 unsigned long available_memory = determine_dirtyable_memory();
Linus Torvalds1da177e2005-04-16 15:20:36 -0700435 struct task_struct *tsk;
436
David Rientjes2da02992009-01-06 14:39:31 -0800437 if (vm_dirty_bytes)
438 dirty = DIV_ROUND_UP(vm_dirty_bytes, PAGE_SIZE);
439 else {
440 int dirty_ratio;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700441
David Rientjes2da02992009-01-06 14:39:31 -0800442 dirty_ratio = vm_dirty_ratio;
443 if (dirty_ratio < 5)
444 dirty_ratio = 5;
445 dirty = (dirty_ratio * available_memory) / 100;
446 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700447
David Rientjes2da02992009-01-06 14:39:31 -0800448 if (dirty_background_bytes)
449 background = DIV_ROUND_UP(dirty_background_bytes, PAGE_SIZE);
450 else
451 background = (dirty_background_ratio * available_memory) / 100;
452
453 if (background >= dirty)
454 background = dirty / 2;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700455 tsk = current;
456 if (tsk->flags & PF_LESS_THROTTLE || rt_task(tsk)) {
457 background += background / 4;
458 dirty += dirty / 4;
459 }
460 *pbackground = background;
461 *pdirty = dirty;
Peter Zijlstra04fbfdc2007-10-16 23:25:50 -0700462
463 if (bdi) {
Peter Zijlstra189d3c42008-04-30 00:54:35 -0700464 u64 bdi_dirty;
Peter Zijlstra04fbfdc2007-10-16 23:25:50 -0700465 long numerator, denominator;
466
467 /*
468 * Calculate this BDI's share of the dirty ratio.
469 */
470 bdi_writeout_fraction(bdi, &numerator, &denominator);
471
Peter Zijlstra189d3c42008-04-30 00:54:35 -0700472 bdi_dirty = (dirty * (100 - bdi_min_ratio)) / 100;
Peter Zijlstra04fbfdc2007-10-16 23:25:50 -0700473 bdi_dirty *= numerator;
474 do_div(bdi_dirty, denominator);
Peter Zijlstra189d3c42008-04-30 00:54:35 -0700475 bdi_dirty += (dirty * bdi->min_ratio) / 100;
Peter Zijlstraa42dde02008-04-30 00:54:36 -0700476 if (bdi_dirty > (dirty * bdi->max_ratio) / 100)
477 bdi_dirty = dirty * bdi->max_ratio / 100;
Peter Zijlstra04fbfdc2007-10-16 23:25:50 -0700478
479 *pbdi_dirty = bdi_dirty;
480 clip_bdi_dirty_limit(bdi, dirty, pbdi_dirty);
Peter Zijlstra3e26c142007-10-16 23:25:50 -0700481 task_dirty_limit(current, pbdi_dirty);
Peter Zijlstra04fbfdc2007-10-16 23:25:50 -0700482 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700483}
484
485/*
486 * balance_dirty_pages() must be called by processes which are generating dirty
487 * data. It looks at the number of dirty pages in the machine and will force
488 * the caller to perform writeback if the system is over `vm_dirty_ratio'.
489 * If we're over `background_thresh' then pdflush is woken to perform some
490 * writeout.
491 */
492static void balance_dirty_pages(struct address_space *mapping)
493{
Peter Zijlstra5fce25a2007-11-14 16:59:15 -0800494 long nr_reclaimable, bdi_nr_reclaimable;
495 long nr_writeback, bdi_nr_writeback;
David Rientjes364aeb22009-01-06 14:39:29 -0800496 unsigned long background_thresh;
497 unsigned long dirty_thresh;
498 unsigned long bdi_thresh;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700499 unsigned long pages_written = 0;
500 unsigned long write_chunk = sync_writeback_pages();
501
502 struct backing_dev_info *bdi = mapping->backing_dev_info;
503
504 for (;;) {
505 struct writeback_control wbc = {
506 .bdi = bdi,
507 .sync_mode = WB_SYNC_NONE,
508 .older_than_this = NULL,
509 .nr_to_write = write_chunk,
OGAWA Hirofumi111ebb62006-06-23 02:03:26 -0700510 .range_cyclic = 1,
Linus Torvalds1da177e2005-04-16 15:20:36 -0700511 };
512
Peter Zijlstra04fbfdc2007-10-16 23:25:50 -0700513 get_dirty_limits(&background_thresh, &dirty_thresh,
514 &bdi_thresh, bdi);
Peter Zijlstra5fce25a2007-11-14 16:59:15 -0800515
516 nr_reclaimable = global_page_state(NR_FILE_DIRTY) +
517 global_page_state(NR_UNSTABLE_NFS);
518 nr_writeback = global_page_state(NR_WRITEBACK);
519
Peter Zijlstra04fbfdc2007-10-16 23:25:50 -0700520 bdi_nr_reclaimable = bdi_stat(bdi, BDI_RECLAIMABLE);
521 bdi_nr_writeback = bdi_stat(bdi, BDI_WRITEBACK);
Peter Zijlstra5fce25a2007-11-14 16:59:15 -0800522
Peter Zijlstra04fbfdc2007-10-16 23:25:50 -0700523 if (bdi_nr_reclaimable + bdi_nr_writeback <= bdi_thresh)
524 break;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700525
Peter Zijlstra5fce25a2007-11-14 16:59:15 -0800526 /*
527 * Throttle it only when the background writeback cannot
528 * catch-up. This avoids (excessively) small writeouts
529 * when the bdi limits are ramping up.
530 */
531 if (nr_reclaimable + nr_writeback <
532 (background_thresh + dirty_thresh) / 2)
533 break;
534
Peter Zijlstra04fbfdc2007-10-16 23:25:50 -0700535 if (!bdi->dirty_exceeded)
536 bdi->dirty_exceeded = 1;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700537
538 /* Note: nr_reclaimable denotes nr_dirty + nr_unstable.
539 * Unstable writes are a feature of certain networked
540 * filesystems (i.e. NFS) in which data may have been
541 * written to the server's write cache, but has not yet
542 * been flushed to permanent storage.
543 */
Peter Zijlstra04fbfdc2007-10-16 23:25:50 -0700544 if (bdi_nr_reclaimable) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700545 writeback_inodes(&wbc);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700546 pages_written += write_chunk - wbc.nr_to_write;
Peter Zijlstra04fbfdc2007-10-16 23:25:50 -0700547 get_dirty_limits(&background_thresh, &dirty_thresh,
548 &bdi_thresh, bdi);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700549 }
Peter Zijlstra04fbfdc2007-10-16 23:25:50 -0700550
551 /*
552 * In order to avoid the stacked BDI deadlock we need
553 * to ensure we accurately count the 'dirty' pages when
554 * the threshold is low.
555 *
556 * Otherwise it would be possible to get thresh+n pages
557 * reported dirty, even though there are thresh-m pages
558 * actually dirty; with m+n sitting in the percpu
559 * deltas.
560 */
561 if (bdi_thresh < 2*bdi_stat_error(bdi)) {
562 bdi_nr_reclaimable = bdi_stat_sum(bdi, BDI_RECLAIMABLE);
563 bdi_nr_writeback = bdi_stat_sum(bdi, BDI_WRITEBACK);
564 } else if (bdi_nr_reclaimable) {
565 bdi_nr_reclaimable = bdi_stat(bdi, BDI_RECLAIMABLE);
566 bdi_nr_writeback = bdi_stat(bdi, BDI_WRITEBACK);
567 }
568
569 if (bdi_nr_reclaimable + bdi_nr_writeback <= bdi_thresh)
570 break;
571 if (pages_written >= write_chunk)
572 break; /* We've done our duty */
573
Andrew Morton3fcfab12006-10-19 23:28:16 -0700574 congestion_wait(WRITE, HZ/10);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700575 }
576
Peter Zijlstra04fbfdc2007-10-16 23:25:50 -0700577 if (bdi_nr_reclaimable + bdi_nr_writeback < bdi_thresh &&
578 bdi->dirty_exceeded)
579 bdi->dirty_exceeded = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700580
581 if (writeback_in_progress(bdi))
582 return; /* pdflush is already working this queue */
583
584 /*
585 * In laptop mode, we wait until hitting the higher threshold before
586 * starting background writeout, and then write out all the way down
587 * to the lower threshold. So slow writers cause minimal disk activity.
588 *
589 * In normal mode, we start background writeout at the lower
590 * background_thresh, to keep the amount of dirty memory low.
591 */
592 if ((laptop_mode && pages_written) ||
Peter Zijlstra04fbfdc2007-10-16 23:25:50 -0700593 (!laptop_mode && (global_page_state(NR_FILE_DIRTY)
594 + global_page_state(NR_UNSTABLE_NFS)
595 > background_thresh)))
Linus Torvalds1da177e2005-04-16 15:20:36 -0700596 pdflush_operation(background_writeout, 0);
597}
598
Peter Zijlstraa200ee12007-10-08 18:54:37 +0200599void set_page_dirty_balance(struct page *page, int page_mkwrite)
Peter Zijlstraedc79b22006-09-25 23:30:58 -0700600{
Peter Zijlstraa200ee12007-10-08 18:54:37 +0200601 if (set_page_dirty(page) || page_mkwrite) {
Peter Zijlstraedc79b22006-09-25 23:30:58 -0700602 struct address_space *mapping = page_mapping(page);
603
604 if (mapping)
605 balance_dirty_pages_ratelimited(mapping);
606 }
607}
608
Linus Torvalds1da177e2005-04-16 15:20:36 -0700609/**
Andrew Mortonfa5a7342006-03-24 03:18:10 -0800610 * balance_dirty_pages_ratelimited_nr - balance dirty memory state
Martin Waitz67be2dd2005-05-01 08:59:26 -0700611 * @mapping: address_space which was dirtied
Martin Waitza5802902006-04-02 13:59:55 +0200612 * @nr_pages_dirtied: number of pages which the caller has just dirtied
Linus Torvalds1da177e2005-04-16 15:20:36 -0700613 *
614 * Processes which are dirtying memory should call in here once for each page
615 * which was newly dirtied. The function will periodically check the system's
616 * dirty state and will initiate writeback if needed.
617 *
618 * On really big machines, get_writeback_state is expensive, so try to avoid
619 * calling it too often (ratelimiting). But once we're over the dirty memory
620 * limit we decrease the ratelimiting by a lot, to prevent individual processes
621 * from overshooting the limit by (ratelimit_pages) each.
622 */
Andrew Mortonfa5a7342006-03-24 03:18:10 -0800623void balance_dirty_pages_ratelimited_nr(struct address_space *mapping,
624 unsigned long nr_pages_dirtied)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700625{
Andrew Mortonfa5a7342006-03-24 03:18:10 -0800626 static DEFINE_PER_CPU(unsigned long, ratelimits) = 0;
627 unsigned long ratelimit;
628 unsigned long *p;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700629
630 ratelimit = ratelimit_pages;
Peter Zijlstra04fbfdc2007-10-16 23:25:50 -0700631 if (mapping->backing_dev_info->dirty_exceeded)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700632 ratelimit = 8;
633
634 /*
635 * Check the rate limiting. Also, we do not want to throttle real-time
636 * tasks in balance_dirty_pages(). Period.
637 */
Andrew Mortonfa5a7342006-03-24 03:18:10 -0800638 preempt_disable();
639 p = &__get_cpu_var(ratelimits);
640 *p += nr_pages_dirtied;
641 if (unlikely(*p >= ratelimit)) {
642 *p = 0;
643 preempt_enable();
Linus Torvalds1da177e2005-04-16 15:20:36 -0700644 balance_dirty_pages(mapping);
645 return;
646 }
Andrew Mortonfa5a7342006-03-24 03:18:10 -0800647 preempt_enable();
Linus Torvalds1da177e2005-04-16 15:20:36 -0700648}
Andrew Mortonfa5a7342006-03-24 03:18:10 -0800649EXPORT_SYMBOL(balance_dirty_pages_ratelimited_nr);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700650
Andrew Morton232ea4d2007-02-28 20:13:21 -0800651void throttle_vm_writeout(gfp_t gfp_mask)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700652{
David Rientjes364aeb22009-01-06 14:39:29 -0800653 unsigned long background_thresh;
654 unsigned long dirty_thresh;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700655
656 for ( ; ; ) {
Peter Zijlstra04fbfdc2007-10-16 23:25:50 -0700657 get_dirty_limits(&background_thresh, &dirty_thresh, NULL, NULL);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700658
659 /*
660 * Boost the allowable dirty threshold a bit for page
661 * allocators so they don't get DoS'ed by heavy writers
662 */
663 dirty_thresh += dirty_thresh / 10; /* wheeee... */
664
Christoph Lameterc24f21b2006-06-30 01:55:42 -0700665 if (global_page_state(NR_UNSTABLE_NFS) +
666 global_page_state(NR_WRITEBACK) <= dirty_thresh)
667 break;
Andrew Morton3fcfab12006-10-19 23:28:16 -0700668 congestion_wait(WRITE, HZ/10);
Fengguang Wu369f2382007-10-16 23:30:45 -0700669
670 /*
671 * The caller might hold locks which can prevent IO completion
672 * or progress in the filesystem. So we cannot just sit here
673 * waiting for IO to complete.
674 */
675 if ((gfp_mask & (__GFP_FS|__GFP_IO)) != (__GFP_FS|__GFP_IO))
676 break;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700677 }
678}
679
Linus Torvalds1da177e2005-04-16 15:20:36 -0700680/*
681 * writeback at least _min_pages, and keep writing until the amount of dirty
682 * memory is less than the background threshold, or until we're all clean.
683 */
684static void background_writeout(unsigned long _min_pages)
685{
686 long min_pages = _min_pages;
687 struct writeback_control wbc = {
688 .bdi = NULL,
689 .sync_mode = WB_SYNC_NONE,
690 .older_than_this = NULL,
691 .nr_to_write = 0,
692 .nonblocking = 1,
OGAWA Hirofumi111ebb62006-06-23 02:03:26 -0700693 .range_cyclic = 1,
Linus Torvalds1da177e2005-04-16 15:20:36 -0700694 };
695
696 for ( ; ; ) {
David Rientjes364aeb22009-01-06 14:39:29 -0800697 unsigned long background_thresh;
698 unsigned long dirty_thresh;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700699
Peter Zijlstra04fbfdc2007-10-16 23:25:50 -0700700 get_dirty_limits(&background_thresh, &dirty_thresh, NULL, NULL);
Christoph Lameterc24f21b2006-06-30 01:55:42 -0700701 if (global_page_state(NR_FILE_DIRTY) +
702 global_page_state(NR_UNSTABLE_NFS) < background_thresh
Linus Torvalds1da177e2005-04-16 15:20:36 -0700703 && min_pages <= 0)
704 break;
Fengguang Wu8bc3be22008-02-04 22:29:36 -0800705 wbc.more_io = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700706 wbc.encountered_congestion = 0;
707 wbc.nr_to_write = MAX_WRITEBACK_PAGES;
708 wbc.pages_skipped = 0;
709 writeback_inodes(&wbc);
710 min_pages -= MAX_WRITEBACK_PAGES - wbc.nr_to_write;
711 if (wbc.nr_to_write > 0 || wbc.pages_skipped > 0) {
712 /* Wrote less than expected */
Fengguang Wu8bc3be22008-02-04 22:29:36 -0800713 if (wbc.encountered_congestion || wbc.more_io)
714 congestion_wait(WRITE, HZ/10);
715 else
Linus Torvalds1da177e2005-04-16 15:20:36 -0700716 break;
717 }
718 }
719}
720
721/*
722 * Start writeback of `nr_pages' pages. If `nr_pages' is zero, write back
723 * the whole world. Returns 0 if a pdflush thread was dispatched. Returns
724 * -1 if all pdflush threads were busy.
725 */
Pekka J Enberg687a21c2005-06-28 20:44:55 -0700726int wakeup_pdflush(long nr_pages)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700727{
Christoph Lameterc24f21b2006-06-30 01:55:42 -0700728 if (nr_pages == 0)
729 nr_pages = global_page_state(NR_FILE_DIRTY) +
730 global_page_state(NR_UNSTABLE_NFS);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700731 return pdflush_operation(background_writeout, nr_pages);
732}
733
734static void wb_timer_fn(unsigned long unused);
735static void laptop_timer_fn(unsigned long unused);
736
Ingo Molnar8d06afa2005-09-09 13:10:40 -0700737static DEFINE_TIMER(wb_timer, wb_timer_fn, 0, 0);
738static DEFINE_TIMER(laptop_mode_wb_timer, laptop_timer_fn, 0, 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700739
740/*
741 * Periodic writeback of "old" data.
742 *
743 * Define "old": the first time one of an inode's pages is dirtied, we mark the
744 * dirtying-time in the inode's address_space. So this periodic writeback code
745 * just walks the superblock inode list, writing back any inodes which are
746 * older than a specific point in time.
747 *
Bart Samwelf6ef9432006-03-24 03:15:48 -0800748 * Try to run once per dirty_writeback_interval. But if a writeback event
749 * takes longer than a dirty_writeback_interval interval, then leave a
Linus Torvalds1da177e2005-04-16 15:20:36 -0700750 * one-second gap.
751 *
752 * older_than_this takes precedence over nr_to_write. So we'll only write back
753 * all dirty pages if they are all attached to "old" mappings.
754 */
755static void wb_kupdate(unsigned long arg)
756{
757 unsigned long oldest_jif;
758 unsigned long start_jif;
759 unsigned long next_jif;
760 long nr_to_write;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700761 struct writeback_control wbc = {
762 .bdi = NULL,
763 .sync_mode = WB_SYNC_NONE,
764 .older_than_this = &oldest_jif,
765 .nr_to_write = 0,
766 .nonblocking = 1,
767 .for_kupdate = 1,
OGAWA Hirofumi111ebb62006-06-23 02:03:26 -0700768 .range_cyclic = 1,
Linus Torvalds1da177e2005-04-16 15:20:36 -0700769 };
770
771 sync_supers();
772
Bart Samwelf6ef9432006-03-24 03:15:48 -0800773 oldest_jif = jiffies - dirty_expire_interval;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700774 start_jif = jiffies;
Bart Samwelf6ef9432006-03-24 03:15:48 -0800775 next_jif = start_jif + dirty_writeback_interval;
Christoph Lameterc24f21b2006-06-30 01:55:42 -0700776 nr_to_write = global_page_state(NR_FILE_DIRTY) +
777 global_page_state(NR_UNSTABLE_NFS) +
Linus Torvalds1da177e2005-04-16 15:20:36 -0700778 (inodes_stat.nr_inodes - inodes_stat.nr_unused);
779 while (nr_to_write > 0) {
Fengguang Wu8bc3be22008-02-04 22:29:36 -0800780 wbc.more_io = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700781 wbc.encountered_congestion = 0;
782 wbc.nr_to_write = MAX_WRITEBACK_PAGES;
783 writeback_inodes(&wbc);
784 if (wbc.nr_to_write > 0) {
Fengguang Wu8bc3be22008-02-04 22:29:36 -0800785 if (wbc.encountered_congestion || wbc.more_io)
Andrew Morton3fcfab12006-10-19 23:28:16 -0700786 congestion_wait(WRITE, HZ/10);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700787 else
788 break; /* All the old data is written */
789 }
790 nr_to_write -= MAX_WRITEBACK_PAGES - wbc.nr_to_write;
791 }
792 if (time_before(next_jif, jiffies + HZ))
793 next_jif = jiffies + HZ;
Bart Samwelf6ef9432006-03-24 03:15:48 -0800794 if (dirty_writeback_interval)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700795 mod_timer(&wb_timer, next_jif);
796}
797
798/*
799 * sysctl handler for /proc/sys/vm/dirty_writeback_centisecs
800 */
801int dirty_writeback_centisecs_handler(ctl_table *table, int write,
Andrew Morton3e733f02007-07-15 23:41:05 -0700802 struct file *file, void __user *buffer, size_t *length, loff_t *ppos)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700803{
Bart Samwelf6ef9432006-03-24 03:15:48 -0800804 proc_dointvec_userhz_jiffies(table, write, file, buffer, length, ppos);
Andrew Morton3e733f02007-07-15 23:41:05 -0700805 if (dirty_writeback_interval)
806 mod_timer(&wb_timer, jiffies + dirty_writeback_interval);
807 else
Linus Torvalds1da177e2005-04-16 15:20:36 -0700808 del_timer(&wb_timer);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700809 return 0;
810}
811
812static void wb_timer_fn(unsigned long unused)
813{
814 if (pdflush_operation(wb_kupdate, 0) < 0)
815 mod_timer(&wb_timer, jiffies + HZ); /* delay 1 second */
816}
817
818static void laptop_flush(unsigned long unused)
819{
820 sys_sync();
821}
822
823static void laptop_timer_fn(unsigned long unused)
824{
825 pdflush_operation(laptop_flush, 0);
826}
827
828/*
829 * We've spun up the disk and we're in laptop mode: schedule writeback
830 * of all dirty data a few seconds from now. If the flush is already scheduled
831 * then push it back - the user is still using the disk.
832 */
833void laptop_io_completion(void)
834{
Bart Samweled5b43f2006-03-24 03:15:49 -0800835 mod_timer(&laptop_mode_wb_timer, jiffies + laptop_mode);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700836}
837
838/*
839 * We're in laptop mode and we've just synced. The sync's writes will have
840 * caused another writeback to be scheduled by laptop_io_completion.
841 * Nothing needs to be written back anymore, so we unschedule the writeback.
842 */
843void laptop_sync_completion(void)
844{
845 del_timer(&laptop_mode_wb_timer);
846}
847
848/*
849 * If ratelimit_pages is too high then we can get into dirty-data overload
850 * if a large number of processes all perform writes at the same time.
851 * If it is too low then SMP machines will call the (expensive)
852 * get_writeback_state too often.
853 *
854 * Here we set ratelimit_pages to a level which ensures that when all CPUs are
855 * dirtying in parallel, we cannot go more than 3% (1/32) over the dirty memory
856 * thresholds before writeback cuts in.
857 *
858 * But the limit should not be set too high. Because it also controls the
859 * amount of memory which the balance_dirty_pages() caller has to write back.
860 * If this is too large then the caller will block on the IO queue all the
861 * time. So limit it to four megabytes - the balance_dirty_pages() caller
862 * will write six megabyte chunks, max.
863 */
864
Chandra Seetharaman2d1d43f2006-09-29 02:01:25 -0700865void writeback_set_ratelimit(void)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700866{
Chandra Seetharaman40c99aa2006-09-29 02:01:24 -0700867 ratelimit_pages = vm_total_pages / (num_online_cpus() * 32);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700868 if (ratelimit_pages < 16)
869 ratelimit_pages = 16;
870 if (ratelimit_pages * PAGE_CACHE_SIZE > 4096 * 1024)
871 ratelimit_pages = (4096 * 1024) / PAGE_CACHE_SIZE;
872}
873
Chandra Seetharaman26c21432006-06-27 02:54:10 -0700874static int __cpuinit
Linus Torvalds1da177e2005-04-16 15:20:36 -0700875ratelimit_handler(struct notifier_block *self, unsigned long u, void *v)
876{
Chandra Seetharaman2d1d43f2006-09-29 02:01:25 -0700877 writeback_set_ratelimit();
Paul E. McKenneyaa0f0302007-02-10 01:46:37 -0800878 return NOTIFY_DONE;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700879}
880
Chandra Seetharaman74b85f32006-06-27 02:54:09 -0700881static struct notifier_block __cpuinitdata ratelimit_nb = {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700882 .notifier_call = ratelimit_handler,
883 .next = NULL,
884};
885
886/*
Linus Torvaldsdc6e29d2007-01-29 16:37:38 -0800887 * Called early on to tune the page writeback dirty limits.
888 *
889 * We used to scale dirty pages according to how total memory
890 * related to pages that could be allocated for buffers (by
891 * comparing nr_free_buffer_pages() to vm_total_pages.
892 *
893 * However, that was when we used "dirty_ratio" to scale with
894 * all memory, and we don't do that any more. "dirty_ratio"
895 * is now applied to total non-HIGHPAGE memory (by subtracting
896 * totalhigh_pages from vm_total_pages), and as such we can't
897 * get into the old insane situation any more where we had
898 * large amounts of dirty pages compared to a small amount of
899 * non-HIGHMEM memory.
900 *
901 * But we might still want to scale the dirty_ratio by how
902 * much memory the box has..
Linus Torvalds1da177e2005-04-16 15:20:36 -0700903 */
904void __init page_writeback_init(void)
905{
Peter Zijlstra04fbfdc2007-10-16 23:25:50 -0700906 int shift;
907
Bart Samwelf6ef9432006-03-24 03:15:48 -0800908 mod_timer(&wb_timer, jiffies + dirty_writeback_interval);
Chandra Seetharaman2d1d43f2006-09-29 02:01:25 -0700909 writeback_set_ratelimit();
Linus Torvalds1da177e2005-04-16 15:20:36 -0700910 register_cpu_notifier(&ratelimit_nb);
Peter Zijlstra04fbfdc2007-10-16 23:25:50 -0700911
912 shift = calc_period_shift();
913 prop_descriptor_init(&vm_completions, shift);
Peter Zijlstra3e26c142007-10-16 23:25:50 -0700914 prop_descriptor_init(&vm_dirties, shift);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700915}
916
David Howells811d7362006-08-29 19:06:09 +0100917/**
Miklos Szeredi0ea97182007-05-10 22:22:51 -0700918 * write_cache_pages - walk the list of dirty pages of the given address space and write all of them.
David Howells811d7362006-08-29 19:06:09 +0100919 * @mapping: address space structure to write
920 * @wbc: subtract the number of written pages from *@wbc->nr_to_write
Miklos Szeredi0ea97182007-05-10 22:22:51 -0700921 * @writepage: function called for each page
922 * @data: data passed to writepage function
David Howells811d7362006-08-29 19:06:09 +0100923 *
Miklos Szeredi0ea97182007-05-10 22:22:51 -0700924 * If a page is already under I/O, write_cache_pages() skips it, even
David Howells811d7362006-08-29 19:06:09 +0100925 * if it's dirty. This is desirable behaviour for memory-cleaning writeback,
926 * but it is INCORRECT for data-integrity system calls such as fsync(). fsync()
927 * and msync() need to guarantee that all the data which was dirty at the time
928 * the call was made get new I/O started against them. If wbc->sync_mode is
929 * WB_SYNC_ALL then we were called for data integrity and we must wait for
930 * existing IO to complete.
David Howells811d7362006-08-29 19:06:09 +0100931 */
Miklos Szeredi0ea97182007-05-10 22:22:51 -0700932int write_cache_pages(struct address_space *mapping,
933 struct writeback_control *wbc, writepage_t writepage,
934 void *data)
David Howells811d7362006-08-29 19:06:09 +0100935{
936 struct backing_dev_info *bdi = mapping->backing_dev_info;
937 int ret = 0;
938 int done = 0;
David Howells811d7362006-08-29 19:06:09 +0100939 struct pagevec pvec;
940 int nr_pages;
Nick Piggin31a12662009-01-06 14:39:04 -0800941 pgoff_t uninitialized_var(writeback_index);
David Howells811d7362006-08-29 19:06:09 +0100942 pgoff_t index;
943 pgoff_t end; /* Inclusive */
Nick Pigginbd19e012009-01-06 14:39:06 -0800944 pgoff_t done_index;
Nick Piggin31a12662009-01-06 14:39:04 -0800945 int cycled;
David Howells811d7362006-08-29 19:06:09 +0100946 int range_whole = 0;
Aneesh Kumar K.V17bc6c32008-10-16 10:09:17 -0400947 long nr_to_write = wbc->nr_to_write;
David Howells811d7362006-08-29 19:06:09 +0100948
949 if (wbc->nonblocking && bdi_write_congested(bdi)) {
950 wbc->encountered_congestion = 1;
951 return 0;
952 }
953
David Howells811d7362006-08-29 19:06:09 +0100954 pagevec_init(&pvec, 0);
955 if (wbc->range_cyclic) {
Nick Piggin31a12662009-01-06 14:39:04 -0800956 writeback_index = mapping->writeback_index; /* prev offset */
957 index = writeback_index;
958 if (index == 0)
959 cycled = 1;
960 else
961 cycled = 0;
David Howells811d7362006-08-29 19:06:09 +0100962 end = -1;
963 } else {
964 index = wbc->range_start >> PAGE_CACHE_SHIFT;
965 end = wbc->range_end >> PAGE_CACHE_SHIFT;
966 if (wbc->range_start == 0 && wbc->range_end == LLONG_MAX)
967 range_whole = 1;
Nick Piggin31a12662009-01-06 14:39:04 -0800968 cycled = 1; /* ignore range_cyclic tests */
David Howells811d7362006-08-29 19:06:09 +0100969 }
970retry:
Nick Pigginbd19e012009-01-06 14:39:06 -0800971 done_index = index;
Nick Piggin5a3d5c92009-01-06 14:39:09 -0800972 while (!done && (index <= end)) {
973 int i;
974
975 nr_pages = pagevec_lookup_tag(&pvec, mapping, &index,
976 PAGECACHE_TAG_DIRTY,
977 min(end - index, (pgoff_t)PAGEVEC_SIZE-1) + 1);
978 if (nr_pages == 0)
979 break;
David Howells811d7362006-08-29 19:06:09 +0100980
David Howells811d7362006-08-29 19:06:09 +0100981 for (i = 0; i < nr_pages; i++) {
982 struct page *page = pvec.pages[i];
983
Nick Piggind5482cd2009-01-06 14:39:11 -0800984 /*
985 * At this point, the page may be truncated or
986 * invalidated (changing page->mapping to NULL), or
987 * even swizzled back from swapper_space to tmpfs file
988 * mapping. However, page->index will not change
989 * because we have a reference on the page.
990 */
991 if (page->index > end) {
992 /*
993 * can't be range_cyclic (1st pass) because
994 * end == -1 in that case.
995 */
996 done = 1;
997 break;
998 }
999
Nick Pigginbd19e012009-01-06 14:39:06 -08001000 done_index = page->index + 1;
1001
David Howells811d7362006-08-29 19:06:09 +01001002 lock_page(page);
1003
Nick Piggin5a3d5c92009-01-06 14:39:09 -08001004 /*
1005 * Page truncated or invalidated. We can freely skip it
1006 * then, even for data integrity operations: the page
1007 * has disappeared concurrently, so there could be no
1008 * real expectation of this data interity operation
1009 * even if there is now a new, dirty page at the same
1010 * pagecache address.
1011 */
David Howells811d7362006-08-29 19:06:09 +01001012 if (unlikely(page->mapping != mapping)) {
Nick Piggin5a3d5c92009-01-06 14:39:09 -08001013continue_unlock:
David Howells811d7362006-08-29 19:06:09 +01001014 unlock_page(page);
1015 continue;
1016 }
1017
Nick Piggin515f4a02009-01-06 14:39:10 -08001018 if (!PageDirty(page)) {
1019 /* someone wrote it for us */
1020 goto continue_unlock;
1021 }
David Howells811d7362006-08-29 19:06:09 +01001022
Nick Piggin515f4a02009-01-06 14:39:10 -08001023 if (PageWriteback(page)) {
1024 if (wbc->sync_mode != WB_SYNC_NONE)
1025 wait_on_page_writeback(page);
1026 else
1027 goto continue_unlock;
1028 }
1029
1030 BUG_ON(PageWriteback(page));
1031 if (!clear_page_dirty_for_io(page))
Nick Piggin5a3d5c92009-01-06 14:39:09 -08001032 goto continue_unlock;
David Howells811d7362006-08-29 19:06:09 +01001033
Miklos Szeredi0ea97182007-05-10 22:22:51 -07001034 ret = (*writepage)(page, wbc, data);
Nick Piggin00266772009-01-06 14:39:06 -08001035 if (unlikely(ret)) {
1036 if (ret == AOP_WRITEPAGE_ACTIVATE) {
1037 unlock_page(page);
1038 ret = 0;
1039 } else {
1040 /*
1041 * done_index is set past this page,
1042 * so media errors will not choke
1043 * background writeout for the entire
1044 * file. This has consequences for
1045 * range_cyclic semantics (ie. it may
1046 * not be suitable for data integrity
1047 * writeout).
1048 */
1049 done = 1;
1050 break;
1051 }
1052 }
David Howells811d7362006-08-29 19:06:09 +01001053
Federico Cuello89e12192009-02-11 13:04:39 -08001054 if (nr_to_write > 0) {
Artem Bityutskiydcf6a792009-02-02 18:33:49 +02001055 nr_to_write--;
Federico Cuello89e12192009-02-11 13:04:39 -08001056 if (nr_to_write == 0 &&
1057 wbc->sync_mode == WB_SYNC_NONE) {
1058 /*
1059 * We stop writing back only if we are
1060 * not doing integrity sync. In case of
1061 * integrity sync we have to keep going
1062 * because someone may be concurrently
1063 * dirtying pages, and we might have
1064 * synced a lot of newly appeared dirty
1065 * pages, but have not synced all of the
1066 * old dirty pages.
1067 */
1068 done = 1;
1069 break;
1070 }
Nick Piggin05fe4782009-01-06 14:39:08 -08001071 }
Artem Bityutskiydcf6a792009-02-02 18:33:49 +02001072
David Howells811d7362006-08-29 19:06:09 +01001073 if (wbc->nonblocking && bdi_write_congested(bdi)) {
1074 wbc->encountered_congestion = 1;
1075 done = 1;
Andrew Morton82fd1a92009-01-06 14:39:11 -08001076 break;
David Howells811d7362006-08-29 19:06:09 +01001077 }
1078 }
1079 pagevec_release(&pvec);
1080 cond_resched();
1081 }
Nick Piggin31a12662009-01-06 14:39:04 -08001082 if (!cycled) {
David Howells811d7362006-08-29 19:06:09 +01001083 /*
Nick Piggin31a12662009-01-06 14:39:04 -08001084 * range_cyclic:
David Howells811d7362006-08-29 19:06:09 +01001085 * We hit the last page and there is more work to be done: wrap
1086 * back to the start of the file
1087 */
Nick Piggin31a12662009-01-06 14:39:04 -08001088 cycled = 1;
David Howells811d7362006-08-29 19:06:09 +01001089 index = 0;
Nick Piggin31a12662009-01-06 14:39:04 -08001090 end = writeback_index - 1;
David Howells811d7362006-08-29 19:06:09 +01001091 goto retry;
1092 }
Aneesh Kumar K.V17bc6c32008-10-16 10:09:17 -04001093 if (!wbc->no_nrwrite_index_update) {
1094 if (wbc->range_cyclic || (range_whole && nr_to_write > 0))
Nick Pigginbd19e012009-01-06 14:39:06 -08001095 mapping->writeback_index = done_index;
Aneesh Kumar K.V17bc6c32008-10-16 10:09:17 -04001096 wbc->nr_to_write = nr_to_write;
1097 }
Aneesh Kumar K.V06d6cf62008-07-11 19:27:31 -04001098
David Howells811d7362006-08-29 19:06:09 +01001099 return ret;
1100}
Miklos Szeredi0ea97182007-05-10 22:22:51 -07001101EXPORT_SYMBOL(write_cache_pages);
1102
1103/*
1104 * Function used by generic_writepages to call the real writepage
1105 * function and set the mapping flags on error
1106 */
1107static int __writepage(struct page *page, struct writeback_control *wbc,
1108 void *data)
1109{
1110 struct address_space *mapping = data;
1111 int ret = mapping->a_ops->writepage(page, wbc);
1112 mapping_set_error(mapping, ret);
1113 return ret;
1114}
1115
1116/**
1117 * generic_writepages - walk the list of dirty pages of the given address space and writepage() all of them.
1118 * @mapping: address space structure to write
1119 * @wbc: subtract the number of written pages from *@wbc->nr_to_write
1120 *
1121 * This is a library function, which implements the writepages()
1122 * address_space_operation.
1123 */
1124int generic_writepages(struct address_space *mapping,
1125 struct writeback_control *wbc)
1126{
1127 /* deal with chardevs and other special file */
1128 if (!mapping->a_ops->writepage)
1129 return 0;
1130
1131 return write_cache_pages(mapping, wbc, __writepage, mapping);
1132}
David Howells811d7362006-08-29 19:06:09 +01001133
1134EXPORT_SYMBOL(generic_writepages);
1135
Linus Torvalds1da177e2005-04-16 15:20:36 -07001136int do_writepages(struct address_space *mapping, struct writeback_control *wbc)
1137{
Andrew Morton22905f72005-11-16 15:07:01 -08001138 int ret;
1139
Linus Torvalds1da177e2005-04-16 15:20:36 -07001140 if (wbc->nr_to_write <= 0)
1141 return 0;
Andrew Morton22905f72005-11-16 15:07:01 -08001142 wbc->for_writepages = 1;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001143 if (mapping->a_ops->writepages)
Peter Zijlstrad08b3852006-09-25 23:30:57 -07001144 ret = mapping->a_ops->writepages(mapping, wbc);
Andrew Morton22905f72005-11-16 15:07:01 -08001145 else
1146 ret = generic_writepages(mapping, wbc);
1147 wbc->for_writepages = 0;
1148 return ret;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001149}
1150
1151/**
1152 * write_one_page - write out a single page and optionally wait on I/O
Martin Waitz67be2dd2005-05-01 08:59:26 -07001153 * @page: the page to write
1154 * @wait: if true, wait on writeout
Linus Torvalds1da177e2005-04-16 15:20:36 -07001155 *
1156 * The page must be locked by the caller and will be unlocked upon return.
1157 *
1158 * write_one_page() returns a negative error code if I/O failed.
1159 */
1160int write_one_page(struct page *page, int wait)
1161{
1162 struct address_space *mapping = page->mapping;
1163 int ret = 0;
1164 struct writeback_control wbc = {
1165 .sync_mode = WB_SYNC_ALL,
1166 .nr_to_write = 1,
1167 };
1168
1169 BUG_ON(!PageLocked(page));
1170
1171 if (wait)
1172 wait_on_page_writeback(page);
1173
1174 if (clear_page_dirty_for_io(page)) {
1175 page_cache_get(page);
1176 ret = mapping->a_ops->writepage(page, &wbc);
1177 if (ret == 0 && wait) {
1178 wait_on_page_writeback(page);
1179 if (PageError(page))
1180 ret = -EIO;
1181 }
1182 page_cache_release(page);
1183 } else {
1184 unlock_page(page);
1185 }
1186 return ret;
1187}
1188EXPORT_SYMBOL(write_one_page);
1189
1190/*
Ken Chen76719322007-02-10 01:43:15 -08001191 * For address_spaces which do not use buffers nor write back.
1192 */
1193int __set_page_dirty_no_writeback(struct page *page)
1194{
1195 if (!PageDirty(page))
1196 SetPageDirty(page);
1197 return 0;
1198}
1199
1200/*
Linus Torvalds1da177e2005-04-16 15:20:36 -07001201 * For address_spaces which do not use buffers. Just tag the page as dirty in
1202 * its radix tree.
1203 *
1204 * This is also used when a single buffer is being dirtied: we want to set the
1205 * page dirty in that case, but not all the buffers. This is a "bottom-up"
1206 * dirtying, whereas __set_page_dirty_buffers() is a "top-down" dirtying.
1207 *
1208 * Most callers have locked the page, which pins the address_space in memory.
1209 * But zap_pte_range() does not lock the page, however in that case the
1210 * mapping is pinned by the vma's ->vm_file reference.
1211 *
1212 * We take care to handle the case where the page was truncated from the
Simon Arlott183ff222007-10-20 01:27:18 +02001213 * mapping by re-checking page_mapping() inside tree_lock.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001214 */
1215int __set_page_dirty_nobuffers(struct page *page)
1216{
Linus Torvalds1da177e2005-04-16 15:20:36 -07001217 if (!TestSetPageDirty(page)) {
1218 struct address_space *mapping = page_mapping(page);
1219 struct address_space *mapping2;
1220
Andrew Morton8c085402006-12-10 02:19:24 -08001221 if (!mapping)
1222 return 1;
1223
Nick Piggin19fd6232008-07-25 19:45:32 -07001224 spin_lock_irq(&mapping->tree_lock);
Andrew Morton8c085402006-12-10 02:19:24 -08001225 mapping2 = page_mapping(page);
1226 if (mapping2) { /* Race with truncate? */
1227 BUG_ON(mapping2 != mapping);
Nick Piggin787d2212007-07-17 04:03:34 -07001228 WARN_ON_ONCE(!PagePrivate(page) && !PageUptodate(page));
Andrew Morton55e829a2006-12-10 02:19:27 -08001229 if (mapping_cap_account_dirty(mapping)) {
Andrew Morton8c085402006-12-10 02:19:24 -08001230 __inc_zone_page_state(page, NR_FILE_DIRTY);
Peter Zijlstrac9e51e42007-10-16 23:25:47 -07001231 __inc_bdi_stat(mapping->backing_dev_info,
1232 BDI_RECLAIMABLE);
Andrew Morton55e829a2006-12-10 02:19:27 -08001233 task_io_account_write(PAGE_CACHE_SIZE);
1234 }
Andrew Morton8c085402006-12-10 02:19:24 -08001235 radix_tree_tag_set(&mapping->page_tree,
1236 page_index(page), PAGECACHE_TAG_DIRTY);
1237 }
Nick Piggin19fd6232008-07-25 19:45:32 -07001238 spin_unlock_irq(&mapping->tree_lock);
Andrew Morton8c085402006-12-10 02:19:24 -08001239 if (mapping->host) {
1240 /* !PageAnon && !swapper_space */
1241 __mark_inode_dirty(mapping->host, I_DIRTY_PAGES);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001242 }
Andrew Morton4741c9f2006-03-24 03:18:11 -08001243 return 1;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001244 }
Andrew Morton4741c9f2006-03-24 03:18:11 -08001245 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001246}
1247EXPORT_SYMBOL(__set_page_dirty_nobuffers);
1248
1249/*
1250 * When a writepage implementation decides that it doesn't want to write this
1251 * page for some reason, it should redirty the locked page via
1252 * redirty_page_for_writepage() and it should then unlock the page and return 0
1253 */
1254int redirty_page_for_writepage(struct writeback_control *wbc, struct page *page)
1255{
1256 wbc->pages_skipped++;
1257 return __set_page_dirty_nobuffers(page);
1258}
1259EXPORT_SYMBOL(redirty_page_for_writepage);
1260
1261/*
1262 * If the mapping doesn't provide a set_page_dirty a_op, then
1263 * just fall through and assume that it wants buffer_heads.
1264 */
Peter Zijlstra3e26c142007-10-16 23:25:50 -07001265static int __set_page_dirty(struct page *page)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001266{
1267 struct address_space *mapping = page_mapping(page);
1268
1269 if (likely(mapping)) {
1270 int (*spd)(struct page *) = mapping->a_ops->set_page_dirty;
David Howells93614012006-09-30 20:45:40 +02001271#ifdef CONFIG_BLOCK
1272 if (!spd)
1273 spd = __set_page_dirty_buffers;
1274#endif
1275 return (*spd)(page);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001276 }
Andrew Morton4741c9f2006-03-24 03:18:11 -08001277 if (!PageDirty(page)) {
1278 if (!TestSetPageDirty(page))
1279 return 1;
1280 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001281 return 0;
1282}
Peter Zijlstra3e26c142007-10-16 23:25:50 -07001283
Harvey Harrison920c7a52008-02-04 22:29:26 -08001284int set_page_dirty(struct page *page)
Peter Zijlstra3e26c142007-10-16 23:25:50 -07001285{
1286 int ret = __set_page_dirty(page);
1287 if (ret)
1288 task_dirty_inc(current);
1289 return ret;
1290}
Linus Torvalds1da177e2005-04-16 15:20:36 -07001291EXPORT_SYMBOL(set_page_dirty);
1292
1293/*
1294 * set_page_dirty() is racy if the caller has no reference against
1295 * page->mapping->host, and if the page is unlocked. This is because another
1296 * CPU could truncate the page off the mapping and then free the mapping.
1297 *
1298 * Usually, the page _is_ locked, or the caller is a user-space process which
1299 * holds a reference on the inode by having an open file.
1300 *
1301 * In other cases, the page should be locked before running set_page_dirty().
1302 */
1303int set_page_dirty_lock(struct page *page)
1304{
1305 int ret;
1306
Nick Piggindb376482006-09-25 23:31:24 -07001307 lock_page_nosync(page);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001308 ret = set_page_dirty(page);
1309 unlock_page(page);
1310 return ret;
1311}
1312EXPORT_SYMBOL(set_page_dirty_lock);
1313
1314/*
Linus Torvalds1da177e2005-04-16 15:20:36 -07001315 * Clear a page's dirty flag, while caring for dirty memory accounting.
1316 * Returns true if the page was previously dirty.
1317 *
1318 * This is for preparing to put the page under writeout. We leave the page
1319 * tagged as dirty in the radix tree so that a concurrent write-for-sync
1320 * can discover it via a PAGECACHE_TAG_DIRTY walk. The ->writepage
1321 * implementation will run either set_page_writeback() or set_page_dirty(),
1322 * at which stage we bring the page's dirty flag and radix-tree dirty tag
1323 * back into sync.
1324 *
1325 * This incoherency between the page's dirty flag and radix-tree tag is
1326 * unfortunate, but it only exists while the page is locked.
1327 */
1328int clear_page_dirty_for_io(struct page *page)
1329{
1330 struct address_space *mapping = page_mapping(page);
1331
Nick Piggin79352892007-07-19 01:47:22 -07001332 BUG_ON(!PageLocked(page));
1333
Fengguang Wufe3cba12007-07-19 01:48:07 -07001334 ClearPageReclaim(page);
Linus Torvalds7658cc22006-12-29 10:00:58 -08001335 if (mapping && mapping_cap_account_dirty(mapping)) {
1336 /*
1337 * Yes, Virginia, this is indeed insane.
1338 *
1339 * We use this sequence to make sure that
1340 * (a) we account for dirty stats properly
1341 * (b) we tell the low-level filesystem to
1342 * mark the whole page dirty if it was
1343 * dirty in a pagetable. Only to then
1344 * (c) clean the page again and return 1 to
1345 * cause the writeback.
1346 *
1347 * This way we avoid all nasty races with the
1348 * dirty bit in multiple places and clearing
1349 * them concurrently from different threads.
1350 *
1351 * Note! Normally the "set_page_dirty(page)"
1352 * has no effect on the actual dirty bit - since
1353 * that will already usually be set. But we
1354 * need the side effects, and it can help us
1355 * avoid races.
1356 *
1357 * We basically use the page "master dirty bit"
1358 * as a serialization point for all the different
1359 * threads doing their things.
Linus Torvalds7658cc22006-12-29 10:00:58 -08001360 */
1361 if (page_mkclean(page))
1362 set_page_dirty(page);
Nick Piggin79352892007-07-19 01:47:22 -07001363 /*
1364 * We carefully synchronise fault handlers against
1365 * installing a dirty pte and marking the page dirty
1366 * at this point. We do this by having them hold the
1367 * page lock at some point after installing their
1368 * pte, but before marking the page dirty.
1369 * Pages are always locked coming in here, so we get
1370 * the desired exclusion. See mm/memory.c:do_wp_page()
1371 * for more comments.
1372 */
Linus Torvalds7658cc22006-12-29 10:00:58 -08001373 if (TestClearPageDirty(page)) {
Andrew Morton8c085402006-12-10 02:19:24 -08001374 dec_zone_page_state(page, NR_FILE_DIRTY);
Peter Zijlstrac9e51e42007-10-16 23:25:47 -07001375 dec_bdi_stat(mapping->backing_dev_info,
1376 BDI_RECLAIMABLE);
Linus Torvalds7658cc22006-12-29 10:00:58 -08001377 return 1;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001378 }
Linus Torvalds7658cc22006-12-29 10:00:58 -08001379 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001380 }
Linus Torvalds7658cc22006-12-29 10:00:58 -08001381 return TestClearPageDirty(page);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001382}
Hans Reiser58bb01a2005-11-18 01:10:53 -08001383EXPORT_SYMBOL(clear_page_dirty_for_io);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001384
1385int test_clear_page_writeback(struct page *page)
1386{
1387 struct address_space *mapping = page_mapping(page);
1388 int ret;
1389
1390 if (mapping) {
Peter Zijlstra69cb51d2007-10-16 23:25:48 -07001391 struct backing_dev_info *bdi = mapping->backing_dev_info;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001392 unsigned long flags;
1393
Nick Piggin19fd6232008-07-25 19:45:32 -07001394 spin_lock_irqsave(&mapping->tree_lock, flags);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001395 ret = TestClearPageWriteback(page);
Peter Zijlstra69cb51d2007-10-16 23:25:48 -07001396 if (ret) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001397 radix_tree_tag_clear(&mapping->page_tree,
1398 page_index(page),
1399 PAGECACHE_TAG_WRITEBACK);
Miklos Szeredie4ad08f2008-04-30 00:54:37 -07001400 if (bdi_cap_account_writeback(bdi)) {
Peter Zijlstra69cb51d2007-10-16 23:25:48 -07001401 __dec_bdi_stat(bdi, BDI_WRITEBACK);
Peter Zijlstra04fbfdc2007-10-16 23:25:50 -07001402 __bdi_writeout_inc(bdi);
1403 }
Peter Zijlstra69cb51d2007-10-16 23:25:48 -07001404 }
Nick Piggin19fd6232008-07-25 19:45:32 -07001405 spin_unlock_irqrestore(&mapping->tree_lock, flags);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001406 } else {
1407 ret = TestClearPageWriteback(page);
1408 }
Andrew Mortond688abf2007-07-19 01:49:17 -07001409 if (ret)
1410 dec_zone_page_state(page, NR_WRITEBACK);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001411 return ret;
1412}
1413
1414int test_set_page_writeback(struct page *page)
1415{
1416 struct address_space *mapping = page_mapping(page);
1417 int ret;
1418
1419 if (mapping) {
Peter Zijlstra69cb51d2007-10-16 23:25:48 -07001420 struct backing_dev_info *bdi = mapping->backing_dev_info;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001421 unsigned long flags;
1422
Nick Piggin19fd6232008-07-25 19:45:32 -07001423 spin_lock_irqsave(&mapping->tree_lock, flags);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001424 ret = TestSetPageWriteback(page);
Peter Zijlstra69cb51d2007-10-16 23:25:48 -07001425 if (!ret) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001426 radix_tree_tag_set(&mapping->page_tree,
1427 page_index(page),
1428 PAGECACHE_TAG_WRITEBACK);
Miklos Szeredie4ad08f2008-04-30 00:54:37 -07001429 if (bdi_cap_account_writeback(bdi))
Peter Zijlstra69cb51d2007-10-16 23:25:48 -07001430 __inc_bdi_stat(bdi, BDI_WRITEBACK);
1431 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001432 if (!PageDirty(page))
1433 radix_tree_tag_clear(&mapping->page_tree,
1434 page_index(page),
1435 PAGECACHE_TAG_DIRTY);
Nick Piggin19fd6232008-07-25 19:45:32 -07001436 spin_unlock_irqrestore(&mapping->tree_lock, flags);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001437 } else {
1438 ret = TestSetPageWriteback(page);
1439 }
Andrew Mortond688abf2007-07-19 01:49:17 -07001440 if (!ret)
1441 inc_zone_page_state(page, NR_WRITEBACK);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001442 return ret;
1443
1444}
1445EXPORT_SYMBOL(test_set_page_writeback);
1446
1447/*
Nick Piggin00128182007-10-16 01:24:40 -07001448 * Return true if any of the pages in the mapping are marked with the
Linus Torvalds1da177e2005-04-16 15:20:36 -07001449 * passed tag.
1450 */
1451int mapping_tagged(struct address_space *mapping, int tag)
1452{
Linus Torvalds1da177e2005-04-16 15:20:36 -07001453 int ret;
Nick Piggin00128182007-10-16 01:24:40 -07001454 rcu_read_lock();
Linus Torvalds1da177e2005-04-16 15:20:36 -07001455 ret = radix_tree_tagged(&mapping->page_tree, tag);
Nick Piggin00128182007-10-16 01:24:40 -07001456 rcu_read_unlock();
Linus Torvalds1da177e2005-04-16 15:20:36 -07001457 return ret;
1458}
1459EXPORT_SYMBOL(mapping_tagged);