blob: 81627ebcd313fcd5ee3545a0996d2802fab36085 [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 */
Wu Fengguang1b5e62b2009-03-23 08:57:38 +080069int dirty_background_ratio = 10;
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 */
Wu Fengguang1b5e62b2009-03-23 08:57:38 +080086int vm_dirty_ratio = 20;
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/*
Alexey Dobriyan704503d2009-03-31 15:23:18 -070095 * The interval between `kupdate'-style writebacks
Linus Torvalds1da177e2005-04-16 15:20:36 -070096 */
Toshiyuki Okajima22ef37e2009-05-16 22:56:28 -070097unsigned int dirty_writeback_interval = 5 * 100; /* centiseconds */
Linus Torvalds1da177e2005-04-16 15:20:36 -070098
99/*
Alexey Dobriyan704503d2009-03-31 15:23:18 -0700100 * The longest time for which data is allowed to remain dirty
Linus Torvalds1da177e2005-04-16 15:20:36 -0700101 */
Toshiyuki Okajima22ef37e2009-05-16 22:56:28 -0700102unsigned int dirty_expire_interval = 30 * 100; /* centiseconds */
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
Nick Piggin1cf6e7d2009-02-18 14:48:18 -0800243void task_dirty_inc(struct task_struct *tsk)
Peter Zijlstra3e26c142007-10-16 23:25:50 -0700244{
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 */
H Hartley Sweetendcf975d2009-06-16 15:31:44 -0700268static void clip_bdi_dirty_limit(struct backing_dev_info *bdi,
269 unsigned long dirty, unsigned long *pbdi_dirty)
Peter Zijlstra04fbfdc2007-10-16 23:25:50 -0700270{
H Hartley Sweetendcf975d2009-06-16 15:31:44 -0700271 unsigned long avail_dirty;
Peter Zijlstra04fbfdc2007-10-16 23:25:50 -0700272
H Hartley Sweetendcf975d2009-06-16 15:31:44 -0700273 avail_dirty = global_page_state(NR_FILE_DIRTY) +
Peter Zijlstra04fbfdc2007-10-16 23:25:50 -0700274 global_page_state(NR_WRITEBACK) +
Miklos Szeredifc3ba692008-04-30 00:54:38 -0700275 global_page_state(NR_UNSTABLE_NFS) +
H Hartley Sweetendcf975d2009-06-16 15:31:44 -0700276 global_page_state(NR_WRITEBACK_TEMP);
Peter Zijlstra04fbfdc2007-10-16 23:25:50 -0700277
H Hartley Sweetendcf975d2009-06-16 15:31:44 -0700278 if (avail_dirty < dirty)
279 avail_dirty = dirty - avail_dirty;
280 else
Peter Zijlstra04fbfdc2007-10-16 23:25:50 -0700281 avail_dirty = 0;
282
283 avail_dirty += bdi_stat(bdi, BDI_RECLAIMABLE) +
284 bdi_stat(bdi, BDI_WRITEBACK);
285
286 *pbdi_dirty = min(*pbdi_dirty, avail_dirty);
287}
288
Peter Zijlstra3e26c142007-10-16 23:25:50 -0700289static inline void task_dirties_fraction(struct task_struct *tsk,
290 long *numerator, long *denominator)
291{
292 prop_fraction_single(&vm_dirties, &tsk->dirties,
293 numerator, denominator);
294}
295
296/*
297 * scale the dirty limit
298 *
299 * task specific dirty limit:
300 *
301 * dirty -= (dirty/8) * p_{t}
302 */
H Hartley Sweetendcf975d2009-06-16 15:31:44 -0700303static void task_dirty_limit(struct task_struct *tsk, unsigned long *pdirty)
Peter Zijlstra3e26c142007-10-16 23:25:50 -0700304{
305 long numerator, denominator;
H Hartley Sweetendcf975d2009-06-16 15:31:44 -0700306 unsigned long dirty = *pdirty;
Peter Zijlstra3e26c142007-10-16 23:25:50 -0700307 u64 inv = dirty >> 3;
308
309 task_dirties_fraction(tsk, &numerator, &denominator);
310 inv *= numerator;
311 do_div(inv, denominator);
312
313 dirty -= inv;
314 if (dirty < *pdirty/2)
315 dirty = *pdirty/2;
316
317 *pdirty = dirty;
318}
319
Peter Zijlstra04fbfdc2007-10-16 23:25:50 -0700320/*
Peter Zijlstra189d3c42008-04-30 00:54:35 -0700321 *
322 */
323static DEFINE_SPINLOCK(bdi_lock);
324static unsigned int bdi_min_ratio;
325
326int bdi_set_min_ratio(struct backing_dev_info *bdi, unsigned int min_ratio)
327{
328 int ret = 0;
329 unsigned long flags;
330
331 spin_lock_irqsave(&bdi_lock, flags);
Peter Zijlstraa42dde02008-04-30 00:54:36 -0700332 if (min_ratio > bdi->max_ratio) {
Peter Zijlstra189d3c42008-04-30 00:54:35 -0700333 ret = -EINVAL;
Peter Zijlstraa42dde02008-04-30 00:54:36 -0700334 } else {
335 min_ratio -= bdi->min_ratio;
336 if (bdi_min_ratio + min_ratio < 100) {
337 bdi_min_ratio += min_ratio;
338 bdi->min_ratio += min_ratio;
339 } else {
340 ret = -EINVAL;
341 }
342 }
Peter Zijlstra189d3c42008-04-30 00:54:35 -0700343 spin_unlock_irqrestore(&bdi_lock, flags);
344
345 return ret;
346}
347
Peter Zijlstraa42dde02008-04-30 00:54:36 -0700348int bdi_set_max_ratio(struct backing_dev_info *bdi, unsigned max_ratio)
349{
350 unsigned long flags;
351 int ret = 0;
352
353 if (max_ratio > 100)
354 return -EINVAL;
355
356 spin_lock_irqsave(&bdi_lock, flags);
357 if (bdi->min_ratio > max_ratio) {
358 ret = -EINVAL;
359 } else {
360 bdi->max_ratio = max_ratio;
361 bdi->max_prop_frac = (PROP_FRAC_BASE * max_ratio) / 100;
362 }
363 spin_unlock_irqrestore(&bdi_lock, flags);
364
365 return ret;
366}
367EXPORT_SYMBOL(bdi_set_max_ratio);
368
Peter Zijlstra189d3c42008-04-30 00:54:35 -0700369/*
Linus Torvalds1da177e2005-04-16 15:20:36 -0700370 * Work out the current dirty-memory clamping and background writeout
371 * thresholds.
372 *
373 * The main aim here is to lower them aggressively if there is a lot of mapped
374 * memory around. To avoid stressing page reclaim with lots of unreclaimable
375 * pages. It is better to clamp down on writers than to start swapping, and
376 * performing lots of scanning.
377 *
378 * We only allow 1/2 of the currently-unmapped memory to be dirtied.
379 *
380 * We don't permit the clamping level to fall below 5% - that is getting rather
381 * excessive.
382 *
383 * We make sure that the background writeout level is below the adjusted
384 * clamping level.
385 */
Christoph Lameter1b424462007-05-06 14:48:59 -0700386
387static unsigned long highmem_dirtyable_memory(unsigned long total)
388{
389#ifdef CONFIG_HIGHMEM
390 int node;
391 unsigned long x = 0;
392
Lee Schermerhorn37b07e42007-10-16 01:25:39 -0700393 for_each_node_state(node, N_HIGH_MEMORY) {
Christoph Lameter1b424462007-05-06 14:48:59 -0700394 struct zone *z =
395 &NODE_DATA(node)->node_zones[ZONE_HIGHMEM];
396
Rik van Riel4f98a2f2008-10-18 20:26:32 -0700397 x += zone_page_state(z, NR_FREE_PAGES) + zone_lru_pages(z);
Christoph Lameter1b424462007-05-06 14:48:59 -0700398 }
399 /*
400 * Make sure that the number of highmem pages is never larger
401 * than the number of the total dirtyable memory. This can only
402 * occur in very strange VM situations but we want to make sure
403 * that this does not occur.
404 */
405 return min(x, total);
406#else
407 return 0;
408#endif
409}
410
Steven Rostedt3eefae92008-05-12 21:21:04 +0200411/**
412 * determine_dirtyable_memory - amount of memory that may be used
413 *
414 * Returns the numebr of pages that can currently be freed and used
415 * by the kernel for direct mappings.
416 */
417unsigned long determine_dirtyable_memory(void)
Christoph Lameter1b424462007-05-06 14:48:59 -0700418{
419 unsigned long x;
420
Rik van Riel4f98a2f2008-10-18 20:26:32 -0700421 x = global_page_state(NR_FREE_PAGES) + global_lru_pages();
Bron Gondwana195cf452008-02-04 22:29:20 -0800422
423 if (!vm_highmem_is_dirtyable)
424 x -= highmem_dirtyable_memory(x);
425
Christoph Lameter1b424462007-05-06 14:48:59 -0700426 return x + 1; /* Ensure that we never return 0 */
427}
428
Peter Zijlstracf0ca9f2008-04-30 00:54:32 -0700429void
David Rientjes364aeb22009-01-06 14:39:29 -0800430get_dirty_limits(unsigned long *pbackground, unsigned long *pdirty,
431 unsigned long *pbdi_dirty, struct backing_dev_info *bdi)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700432{
David Rientjes364aeb22009-01-06 14:39:29 -0800433 unsigned long background;
434 unsigned long dirty;
Christoph Lameter1b424462007-05-06 14:48:59 -0700435 unsigned long available_memory = determine_dirtyable_memory();
Linus Torvalds1da177e2005-04-16 15:20:36 -0700436 struct task_struct *tsk;
437
David Rientjes2da02992009-01-06 14:39:31 -0800438 if (vm_dirty_bytes)
439 dirty = DIV_ROUND_UP(vm_dirty_bytes, PAGE_SIZE);
440 else {
441 int dirty_ratio;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700442
David Rientjes2da02992009-01-06 14:39:31 -0800443 dirty_ratio = vm_dirty_ratio;
444 if (dirty_ratio < 5)
445 dirty_ratio = 5;
446 dirty = (dirty_ratio * available_memory) / 100;
447 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700448
David Rientjes2da02992009-01-06 14:39:31 -0800449 if (dirty_background_bytes)
450 background = DIV_ROUND_UP(dirty_background_bytes, PAGE_SIZE);
451 else
452 background = (dirty_background_ratio * available_memory) / 100;
453
454 if (background >= dirty)
455 background = dirty / 2;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700456 tsk = current;
457 if (tsk->flags & PF_LESS_THROTTLE || rt_task(tsk)) {
458 background += background / 4;
459 dirty += dirty / 4;
460 }
461 *pbackground = background;
462 *pdirty = dirty;
Peter Zijlstra04fbfdc2007-10-16 23:25:50 -0700463
464 if (bdi) {
Peter Zijlstra189d3c42008-04-30 00:54:35 -0700465 u64 bdi_dirty;
Peter Zijlstra04fbfdc2007-10-16 23:25:50 -0700466 long numerator, denominator;
467
468 /*
469 * Calculate this BDI's share of the dirty ratio.
470 */
471 bdi_writeout_fraction(bdi, &numerator, &denominator);
472
Peter Zijlstra189d3c42008-04-30 00:54:35 -0700473 bdi_dirty = (dirty * (100 - bdi_min_ratio)) / 100;
Peter Zijlstra04fbfdc2007-10-16 23:25:50 -0700474 bdi_dirty *= numerator;
475 do_div(bdi_dirty, denominator);
Peter Zijlstra189d3c42008-04-30 00:54:35 -0700476 bdi_dirty += (dirty * bdi->min_ratio) / 100;
Peter Zijlstraa42dde02008-04-30 00:54:36 -0700477 if (bdi_dirty > (dirty * bdi->max_ratio) / 100)
478 bdi_dirty = dirty * bdi->max_ratio / 100;
Peter Zijlstra04fbfdc2007-10-16 23:25:50 -0700479
480 *pbdi_dirty = bdi_dirty;
481 clip_bdi_dirty_limit(bdi, dirty, pbdi_dirty);
Peter Zijlstra3e26c142007-10-16 23:25:50 -0700482 task_dirty_limit(current, pbdi_dirty);
Peter Zijlstra04fbfdc2007-10-16 23:25:50 -0700483 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700484}
485
486/*
487 * balance_dirty_pages() must be called by processes which are generating dirty
488 * data. It looks at the number of dirty pages in the machine and will force
489 * the caller to perform writeback if the system is over `vm_dirty_ratio'.
490 * If we're over `background_thresh' then pdflush is woken to perform some
491 * writeout.
492 */
493static void balance_dirty_pages(struct address_space *mapping)
494{
Peter Zijlstra5fce25a2007-11-14 16:59:15 -0800495 long nr_reclaimable, bdi_nr_reclaimable;
496 long nr_writeback, bdi_nr_writeback;
David Rientjes364aeb22009-01-06 14:39:29 -0800497 unsigned long background_thresh;
498 unsigned long dirty_thresh;
499 unsigned long bdi_thresh;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700500 unsigned long pages_written = 0;
501 unsigned long write_chunk = sync_writeback_pages();
502
503 struct backing_dev_info *bdi = mapping->backing_dev_info;
504
505 for (;;) {
506 struct writeback_control wbc = {
507 .bdi = bdi,
508 .sync_mode = WB_SYNC_NONE,
509 .older_than_this = NULL,
510 .nr_to_write = write_chunk,
OGAWA Hirofumi111ebb62006-06-23 02:03:26 -0700511 .range_cyclic = 1,
Linus Torvalds1da177e2005-04-16 15:20:36 -0700512 };
513
Peter Zijlstra04fbfdc2007-10-16 23:25:50 -0700514 get_dirty_limits(&background_thresh, &dirty_thresh,
515 &bdi_thresh, bdi);
Peter Zijlstra5fce25a2007-11-14 16:59:15 -0800516
517 nr_reclaimable = global_page_state(NR_FILE_DIRTY) +
518 global_page_state(NR_UNSTABLE_NFS);
519 nr_writeback = global_page_state(NR_WRITEBACK);
520
Peter Zijlstra04fbfdc2007-10-16 23:25:50 -0700521 bdi_nr_reclaimable = bdi_stat(bdi, BDI_RECLAIMABLE);
522 bdi_nr_writeback = bdi_stat(bdi, BDI_WRITEBACK);
Peter Zijlstra5fce25a2007-11-14 16:59:15 -0800523
Peter Zijlstra04fbfdc2007-10-16 23:25:50 -0700524 if (bdi_nr_reclaimable + bdi_nr_writeback <= bdi_thresh)
525 break;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700526
Peter Zijlstra5fce25a2007-11-14 16:59:15 -0800527 /*
528 * Throttle it only when the background writeback cannot
529 * catch-up. This avoids (excessively) small writeouts
530 * when the bdi limits are ramping up.
531 */
532 if (nr_reclaimable + nr_writeback <
533 (background_thresh + dirty_thresh) / 2)
534 break;
535
Peter Zijlstra04fbfdc2007-10-16 23:25:50 -0700536 if (!bdi->dirty_exceeded)
537 bdi->dirty_exceeded = 1;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700538
539 /* Note: nr_reclaimable denotes nr_dirty + nr_unstable.
540 * Unstable writes are a feature of certain networked
541 * filesystems (i.e. NFS) in which data may have been
542 * written to the server's write cache, but has not yet
543 * been flushed to permanent storage.
Richard Kennedyd7831a02009-06-30 11:41:35 -0700544 * Only move pages to writeback if this bdi is over its
545 * threshold otherwise wait until the disk writes catch
546 * up.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700547 */
Richard Kennedyd7831a02009-06-30 11:41:35 -0700548 if (bdi_nr_reclaimable > bdi_thresh) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700549 writeback_inodes(&wbc);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700550 pages_written += write_chunk - wbc.nr_to_write;
Peter Zijlstra04fbfdc2007-10-16 23:25:50 -0700551 get_dirty_limits(&background_thresh, &dirty_thresh,
552 &bdi_thresh, bdi);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700553 }
Peter Zijlstra04fbfdc2007-10-16 23:25:50 -0700554
555 /*
556 * In order to avoid the stacked BDI deadlock we need
557 * to ensure we accurately count the 'dirty' pages when
558 * the threshold is low.
559 *
560 * Otherwise it would be possible to get thresh+n pages
561 * reported dirty, even though there are thresh-m pages
562 * actually dirty; with m+n sitting in the percpu
563 * deltas.
564 */
565 if (bdi_thresh < 2*bdi_stat_error(bdi)) {
566 bdi_nr_reclaimable = bdi_stat_sum(bdi, BDI_RECLAIMABLE);
567 bdi_nr_writeback = bdi_stat_sum(bdi, BDI_WRITEBACK);
568 } else if (bdi_nr_reclaimable) {
569 bdi_nr_reclaimable = bdi_stat(bdi, BDI_RECLAIMABLE);
570 bdi_nr_writeback = bdi_stat(bdi, BDI_WRITEBACK);
571 }
572
573 if (bdi_nr_reclaimable + bdi_nr_writeback <= bdi_thresh)
574 break;
575 if (pages_written >= write_chunk)
576 break; /* We've done our duty */
577
Jens Axboe8aa7e842009-07-09 14:52:32 +0200578 congestion_wait(BLK_RW_ASYNC, HZ/10);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700579 }
580
Peter Zijlstra04fbfdc2007-10-16 23:25:50 -0700581 if (bdi_nr_reclaimable + bdi_nr_writeback < bdi_thresh &&
582 bdi->dirty_exceeded)
583 bdi->dirty_exceeded = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700584
585 if (writeback_in_progress(bdi))
586 return; /* pdflush is already working this queue */
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) ||
Peter Zijlstra04fbfdc2007-10-16 23:25:50 -0700597 (!laptop_mode && (global_page_state(NR_FILE_DIRTY)
598 + global_page_state(NR_UNSTABLE_NFS)
599 > background_thresh)))
Linus Torvalds1da177e2005-04-16 15:20:36 -0700600 pdflush_operation(background_writeout, 0);
601}
602
Peter Zijlstraa200ee12007-10-08 18:54:37 +0200603void set_page_dirty_balance(struct page *page, int page_mkwrite)
Peter Zijlstraedc79b22006-09-25 23:30:58 -0700604{
Peter Zijlstraa200ee12007-10-08 18:54:37 +0200605 if (set_page_dirty(page) || page_mkwrite) {
Peter Zijlstraedc79b22006-09-25 23:30:58 -0700606 struct address_space *mapping = page_mapping(page);
607
608 if (mapping)
609 balance_dirty_pages_ratelimited(mapping);
610 }
611}
612
Linus Torvalds1da177e2005-04-16 15:20:36 -0700613/**
Andrew Mortonfa5a7342006-03-24 03:18:10 -0800614 * balance_dirty_pages_ratelimited_nr - balance dirty memory state
Martin Waitz67be2dd2005-05-01 08:59:26 -0700615 * @mapping: address_space which was dirtied
Martin Waitza5802902006-04-02 13:59:55 +0200616 * @nr_pages_dirtied: number of pages which the caller has just dirtied
Linus Torvalds1da177e2005-04-16 15:20:36 -0700617 *
618 * Processes which are dirtying memory should call in here once for each page
619 * which was newly dirtied. The function will periodically check the system's
620 * dirty state and will initiate writeback if needed.
621 *
622 * On really big machines, get_writeback_state is expensive, so try to avoid
623 * calling it too often (ratelimiting). But once we're over the dirty memory
624 * limit we decrease the ratelimiting by a lot, to prevent individual processes
625 * from overshooting the limit by (ratelimit_pages) each.
626 */
Andrew Mortonfa5a7342006-03-24 03:18:10 -0800627void balance_dirty_pages_ratelimited_nr(struct address_space *mapping,
628 unsigned long nr_pages_dirtied)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700629{
Andrew Mortonfa5a7342006-03-24 03:18:10 -0800630 static DEFINE_PER_CPU(unsigned long, ratelimits) = 0;
631 unsigned long ratelimit;
632 unsigned long *p;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700633
634 ratelimit = ratelimit_pages;
Peter Zijlstra04fbfdc2007-10-16 23:25:50 -0700635 if (mapping->backing_dev_info->dirty_exceeded)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700636 ratelimit = 8;
637
638 /*
639 * Check the rate limiting. Also, we do not want to throttle real-time
640 * tasks in balance_dirty_pages(). Period.
641 */
Andrew Mortonfa5a7342006-03-24 03:18:10 -0800642 preempt_disable();
643 p = &__get_cpu_var(ratelimits);
644 *p += nr_pages_dirtied;
645 if (unlikely(*p >= ratelimit)) {
646 *p = 0;
647 preempt_enable();
Linus Torvalds1da177e2005-04-16 15:20:36 -0700648 balance_dirty_pages(mapping);
649 return;
650 }
Andrew Mortonfa5a7342006-03-24 03:18:10 -0800651 preempt_enable();
Linus Torvalds1da177e2005-04-16 15:20:36 -0700652}
Andrew Mortonfa5a7342006-03-24 03:18:10 -0800653EXPORT_SYMBOL(balance_dirty_pages_ratelimited_nr);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700654
Andrew Morton232ea4d2007-02-28 20:13:21 -0800655void throttle_vm_writeout(gfp_t gfp_mask)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700656{
David Rientjes364aeb22009-01-06 14:39:29 -0800657 unsigned long background_thresh;
658 unsigned long dirty_thresh;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700659
660 for ( ; ; ) {
Peter Zijlstra04fbfdc2007-10-16 23:25:50 -0700661 get_dirty_limits(&background_thresh, &dirty_thresh, NULL, NULL);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700662
663 /*
664 * Boost the allowable dirty threshold a bit for page
665 * allocators so they don't get DoS'ed by heavy writers
666 */
667 dirty_thresh += dirty_thresh / 10; /* wheeee... */
668
Christoph Lameterc24f21b2006-06-30 01:55:42 -0700669 if (global_page_state(NR_UNSTABLE_NFS) +
670 global_page_state(NR_WRITEBACK) <= dirty_thresh)
671 break;
Jens Axboe8aa7e842009-07-09 14:52:32 +0200672 congestion_wait(BLK_RW_ASYNC, HZ/10);
Fengguang Wu369f2382007-10-16 23:30:45 -0700673
674 /*
675 * The caller might hold locks which can prevent IO completion
676 * or progress in the filesystem. So we cannot just sit here
677 * waiting for IO to complete.
678 */
679 if ((gfp_mask & (__GFP_FS|__GFP_IO)) != (__GFP_FS|__GFP_IO))
680 break;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700681 }
682}
683
Linus Torvalds1da177e2005-04-16 15:20:36 -0700684/*
685 * writeback at least _min_pages, and keep writing until the amount of dirty
686 * memory is less than the background threshold, or until we're all clean.
687 */
688static void background_writeout(unsigned long _min_pages)
689{
690 long min_pages = _min_pages;
691 struct writeback_control wbc = {
692 .bdi = NULL,
693 .sync_mode = WB_SYNC_NONE,
694 .older_than_this = NULL,
695 .nr_to_write = 0,
696 .nonblocking = 1,
OGAWA Hirofumi111ebb62006-06-23 02:03:26 -0700697 .range_cyclic = 1,
Linus Torvalds1da177e2005-04-16 15:20:36 -0700698 };
699
700 for ( ; ; ) {
David Rientjes364aeb22009-01-06 14:39:29 -0800701 unsigned long background_thresh;
702 unsigned long dirty_thresh;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700703
Peter Zijlstra04fbfdc2007-10-16 23:25:50 -0700704 get_dirty_limits(&background_thresh, &dirty_thresh, NULL, NULL);
Christoph Lameterc24f21b2006-06-30 01:55:42 -0700705 if (global_page_state(NR_FILE_DIRTY) +
706 global_page_state(NR_UNSTABLE_NFS) < background_thresh
Linus Torvalds1da177e2005-04-16 15:20:36 -0700707 && min_pages <= 0)
708 break;
Fengguang Wu8bc3be22008-02-04 22:29:36 -0800709 wbc.more_io = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700710 wbc.encountered_congestion = 0;
711 wbc.nr_to_write = MAX_WRITEBACK_PAGES;
712 wbc.pages_skipped = 0;
713 writeback_inodes(&wbc);
714 min_pages -= MAX_WRITEBACK_PAGES - wbc.nr_to_write;
715 if (wbc.nr_to_write > 0 || wbc.pages_skipped > 0) {
716 /* Wrote less than expected */
Fengguang Wu8bc3be22008-02-04 22:29:36 -0800717 if (wbc.encountered_congestion || wbc.more_io)
Jens Axboe8aa7e842009-07-09 14:52:32 +0200718 congestion_wait(BLK_RW_ASYNC, HZ/10);
Fengguang Wu8bc3be22008-02-04 22:29:36 -0800719 else
Linus Torvalds1da177e2005-04-16 15:20:36 -0700720 break;
721 }
722 }
723}
724
725/*
726 * Start writeback of `nr_pages' pages. If `nr_pages' is zero, write back
727 * the whole world. Returns 0 if a pdflush thread was dispatched. Returns
728 * -1 if all pdflush threads were busy.
729 */
Pekka J Enberg687a21c2005-06-28 20:44:55 -0700730int wakeup_pdflush(long nr_pages)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700731{
Christoph Lameterc24f21b2006-06-30 01:55:42 -0700732 if (nr_pages == 0)
733 nr_pages = global_page_state(NR_FILE_DIRTY) +
734 global_page_state(NR_UNSTABLE_NFS);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700735 return pdflush_operation(background_writeout, nr_pages);
736}
737
738static void wb_timer_fn(unsigned long unused);
739static void laptop_timer_fn(unsigned long unused);
740
Ingo Molnar8d06afa2005-09-09 13:10:40 -0700741static DEFINE_TIMER(wb_timer, wb_timer_fn, 0, 0);
742static DEFINE_TIMER(laptop_mode_wb_timer, laptop_timer_fn, 0, 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700743
744/*
745 * Periodic writeback of "old" data.
746 *
747 * Define "old": the first time one of an inode's pages is dirtied, we mark the
748 * dirtying-time in the inode's address_space. So this periodic writeback code
749 * just walks the superblock inode list, writing back any inodes which are
750 * older than a specific point in time.
751 *
Bart Samwelf6ef9432006-03-24 03:15:48 -0800752 * Try to run once per dirty_writeback_interval. But if a writeback event
753 * takes longer than a dirty_writeback_interval interval, then leave a
Linus Torvalds1da177e2005-04-16 15:20:36 -0700754 * one-second gap.
755 *
756 * older_than_this takes precedence over nr_to_write. So we'll only write back
757 * all dirty pages if they are all attached to "old" mappings.
758 */
759static void wb_kupdate(unsigned long arg)
760{
761 unsigned long oldest_jif;
762 unsigned long start_jif;
763 unsigned long next_jif;
764 long nr_to_write;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700765 struct writeback_control wbc = {
766 .bdi = NULL,
767 .sync_mode = WB_SYNC_NONE,
768 .older_than_this = &oldest_jif,
769 .nr_to_write = 0,
770 .nonblocking = 1,
771 .for_kupdate = 1,
OGAWA Hirofumi111ebb62006-06-23 02:03:26 -0700772 .range_cyclic = 1,
Linus Torvalds1da177e2005-04-16 15:20:36 -0700773 };
774
775 sync_supers();
776
Toshiyuki Okajima22ef37e2009-05-16 22:56:28 -0700777 oldest_jif = jiffies - msecs_to_jiffies(dirty_expire_interval * 10);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700778 start_jif = jiffies;
Alexey Dobriyan704503d2009-03-31 15:23:18 -0700779 next_jif = start_jif + msecs_to_jiffies(dirty_writeback_interval * 10);
Christoph Lameterc24f21b2006-06-30 01:55:42 -0700780 nr_to_write = global_page_state(NR_FILE_DIRTY) +
781 global_page_state(NR_UNSTABLE_NFS) +
Linus Torvalds1da177e2005-04-16 15:20:36 -0700782 (inodes_stat.nr_inodes - inodes_stat.nr_unused);
783 while (nr_to_write > 0) {
Fengguang Wu8bc3be22008-02-04 22:29:36 -0800784 wbc.more_io = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700785 wbc.encountered_congestion = 0;
786 wbc.nr_to_write = MAX_WRITEBACK_PAGES;
787 writeback_inodes(&wbc);
788 if (wbc.nr_to_write > 0) {
Fengguang Wu8bc3be22008-02-04 22:29:36 -0800789 if (wbc.encountered_congestion || wbc.more_io)
Jens Axboe8aa7e842009-07-09 14:52:32 +0200790 congestion_wait(BLK_RW_ASYNC, HZ/10);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700791 else
792 break; /* All the old data is written */
793 }
794 nr_to_write -= MAX_WRITEBACK_PAGES - wbc.nr_to_write;
795 }
796 if (time_before(next_jif, jiffies + HZ))
797 next_jif = jiffies + HZ;
Bart Samwelf6ef9432006-03-24 03:15:48 -0800798 if (dirty_writeback_interval)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700799 mod_timer(&wb_timer, next_jif);
800}
801
802/*
803 * sysctl handler for /proc/sys/vm/dirty_writeback_centisecs
804 */
805int dirty_writeback_centisecs_handler(ctl_table *table, int write,
Andrew Morton3e733f02007-07-15 23:41:05 -0700806 struct file *file, void __user *buffer, size_t *length, loff_t *ppos)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700807{
Alexey Dobriyan704503d2009-03-31 15:23:18 -0700808 proc_dointvec(table, write, file, buffer, length, ppos);
Andrew Morton3e733f02007-07-15 23:41:05 -0700809 if (dirty_writeback_interval)
Alexey Dobriyan704503d2009-03-31 15:23:18 -0700810 mod_timer(&wb_timer, jiffies +
811 msecs_to_jiffies(dirty_writeback_interval * 10));
Andrew Morton3e733f02007-07-15 23:41:05 -0700812 else
Linus Torvalds1da177e2005-04-16 15:20:36 -0700813 del_timer(&wb_timer);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700814 return 0;
815}
816
817static void wb_timer_fn(unsigned long unused)
818{
819 if (pdflush_operation(wb_kupdate, 0) < 0)
820 mod_timer(&wb_timer, jiffies + HZ); /* delay 1 second */
821}
822
823static void laptop_flush(unsigned long unused)
824{
825 sys_sync();
826}
827
828static void laptop_timer_fn(unsigned long unused)
829{
830 pdflush_operation(laptop_flush, 0);
831}
832
833/*
834 * We've spun up the disk and we're in laptop mode: schedule writeback
835 * of all dirty data a few seconds from now. If the flush is already scheduled
836 * then push it back - the user is still using the disk.
837 */
838void laptop_io_completion(void)
839{
Bart Samweled5b43f2006-03-24 03:15:49 -0800840 mod_timer(&laptop_mode_wb_timer, jiffies + laptop_mode);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700841}
842
843/*
844 * We're in laptop mode and we've just synced. The sync's writes will have
845 * caused another writeback to be scheduled by laptop_io_completion.
846 * Nothing needs to be written back anymore, so we unschedule the writeback.
847 */
848void laptop_sync_completion(void)
849{
850 del_timer(&laptop_mode_wb_timer);
851}
852
853/*
854 * If ratelimit_pages is too high then we can get into dirty-data overload
855 * if a large number of processes all perform writes at the same time.
856 * If it is too low then SMP machines will call the (expensive)
857 * get_writeback_state too often.
858 *
859 * Here we set ratelimit_pages to a level which ensures that when all CPUs are
860 * dirtying in parallel, we cannot go more than 3% (1/32) over the dirty memory
861 * thresholds before writeback cuts in.
862 *
863 * But the limit should not be set too high. Because it also controls the
864 * amount of memory which the balance_dirty_pages() caller has to write back.
865 * If this is too large then the caller will block on the IO queue all the
866 * time. So limit it to four megabytes - the balance_dirty_pages() caller
867 * will write six megabyte chunks, max.
868 */
869
Chandra Seetharaman2d1d43f2006-09-29 02:01:25 -0700870void writeback_set_ratelimit(void)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700871{
Chandra Seetharaman40c99aa2006-09-29 02:01:24 -0700872 ratelimit_pages = vm_total_pages / (num_online_cpus() * 32);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700873 if (ratelimit_pages < 16)
874 ratelimit_pages = 16;
875 if (ratelimit_pages * PAGE_CACHE_SIZE > 4096 * 1024)
876 ratelimit_pages = (4096 * 1024) / PAGE_CACHE_SIZE;
877}
878
Chandra Seetharaman26c21432006-06-27 02:54:10 -0700879static int __cpuinit
Linus Torvalds1da177e2005-04-16 15:20:36 -0700880ratelimit_handler(struct notifier_block *self, unsigned long u, void *v)
881{
Chandra Seetharaman2d1d43f2006-09-29 02:01:25 -0700882 writeback_set_ratelimit();
Paul E. McKenneyaa0f0302007-02-10 01:46:37 -0800883 return NOTIFY_DONE;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700884}
885
Chandra Seetharaman74b85f32006-06-27 02:54:09 -0700886static struct notifier_block __cpuinitdata ratelimit_nb = {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700887 .notifier_call = ratelimit_handler,
888 .next = NULL,
889};
890
891/*
Linus Torvaldsdc6e29d2007-01-29 16:37:38 -0800892 * Called early on to tune the page writeback dirty limits.
893 *
894 * We used to scale dirty pages according to how total memory
895 * related to pages that could be allocated for buffers (by
896 * comparing nr_free_buffer_pages() to vm_total_pages.
897 *
898 * However, that was when we used "dirty_ratio" to scale with
899 * all memory, and we don't do that any more. "dirty_ratio"
900 * is now applied to total non-HIGHPAGE memory (by subtracting
901 * totalhigh_pages from vm_total_pages), and as such we can't
902 * get into the old insane situation any more where we had
903 * large amounts of dirty pages compared to a small amount of
904 * non-HIGHMEM memory.
905 *
906 * But we might still want to scale the dirty_ratio by how
907 * much memory the box has..
Linus Torvalds1da177e2005-04-16 15:20:36 -0700908 */
909void __init page_writeback_init(void)
910{
Peter Zijlstra04fbfdc2007-10-16 23:25:50 -0700911 int shift;
912
Alexey Dobriyan704503d2009-03-31 15:23:18 -0700913 mod_timer(&wb_timer,
914 jiffies + msecs_to_jiffies(dirty_writeback_interval * 10));
Chandra Seetharaman2d1d43f2006-09-29 02:01:25 -0700915 writeback_set_ratelimit();
Linus Torvalds1da177e2005-04-16 15:20:36 -0700916 register_cpu_notifier(&ratelimit_nb);
Peter Zijlstra04fbfdc2007-10-16 23:25:50 -0700917
918 shift = calc_period_shift();
919 prop_descriptor_init(&vm_completions, shift);
Peter Zijlstra3e26c142007-10-16 23:25:50 -0700920 prop_descriptor_init(&vm_dirties, shift);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700921}
922
David Howells811d7362006-08-29 19:06:09 +0100923/**
Miklos Szeredi0ea97182007-05-10 22:22:51 -0700924 * 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 +0100925 * @mapping: address space structure to write
926 * @wbc: subtract the number of written pages from *@wbc->nr_to_write
Miklos Szeredi0ea97182007-05-10 22:22:51 -0700927 * @writepage: function called for each page
928 * @data: data passed to writepage function
David Howells811d7362006-08-29 19:06:09 +0100929 *
Miklos Szeredi0ea97182007-05-10 22:22:51 -0700930 * If a page is already under I/O, write_cache_pages() skips it, even
David Howells811d7362006-08-29 19:06:09 +0100931 * if it's dirty. This is desirable behaviour for memory-cleaning writeback,
932 * but it is INCORRECT for data-integrity system calls such as fsync(). fsync()
933 * and msync() need to guarantee that all the data which was dirty at the time
934 * the call was made get new I/O started against them. If wbc->sync_mode is
935 * WB_SYNC_ALL then we were called for data integrity and we must wait for
936 * existing IO to complete.
David Howells811d7362006-08-29 19:06:09 +0100937 */
Miklos Szeredi0ea97182007-05-10 22:22:51 -0700938int write_cache_pages(struct address_space *mapping,
939 struct writeback_control *wbc, writepage_t writepage,
940 void *data)
David Howells811d7362006-08-29 19:06:09 +0100941{
942 struct backing_dev_info *bdi = mapping->backing_dev_info;
943 int ret = 0;
944 int done = 0;
David Howells811d7362006-08-29 19:06:09 +0100945 struct pagevec pvec;
946 int nr_pages;
Nick Piggin31a12662009-01-06 14:39:04 -0800947 pgoff_t uninitialized_var(writeback_index);
David Howells811d7362006-08-29 19:06:09 +0100948 pgoff_t index;
949 pgoff_t end; /* Inclusive */
Nick Pigginbd19e012009-01-06 14:39:06 -0800950 pgoff_t done_index;
Nick Piggin31a12662009-01-06 14:39:04 -0800951 int cycled;
David Howells811d7362006-08-29 19:06:09 +0100952 int range_whole = 0;
Aneesh Kumar K.V17bc6c32008-10-16 10:09:17 -0400953 long nr_to_write = wbc->nr_to_write;
David Howells811d7362006-08-29 19:06:09 +0100954
955 if (wbc->nonblocking && bdi_write_congested(bdi)) {
956 wbc->encountered_congestion = 1;
957 return 0;
958 }
959
David Howells811d7362006-08-29 19:06:09 +0100960 pagevec_init(&pvec, 0);
961 if (wbc->range_cyclic) {
Nick Piggin31a12662009-01-06 14:39:04 -0800962 writeback_index = mapping->writeback_index; /* prev offset */
963 index = writeback_index;
964 if (index == 0)
965 cycled = 1;
966 else
967 cycled = 0;
David Howells811d7362006-08-29 19:06:09 +0100968 end = -1;
969 } else {
970 index = wbc->range_start >> PAGE_CACHE_SHIFT;
971 end = wbc->range_end >> PAGE_CACHE_SHIFT;
972 if (wbc->range_start == 0 && wbc->range_end == LLONG_MAX)
973 range_whole = 1;
Nick Piggin31a12662009-01-06 14:39:04 -0800974 cycled = 1; /* ignore range_cyclic tests */
David Howells811d7362006-08-29 19:06:09 +0100975 }
976retry:
Nick Pigginbd19e012009-01-06 14:39:06 -0800977 done_index = index;
Nick Piggin5a3d5c92009-01-06 14:39:09 -0800978 while (!done && (index <= end)) {
979 int i;
980
981 nr_pages = pagevec_lookup_tag(&pvec, mapping, &index,
982 PAGECACHE_TAG_DIRTY,
983 min(end - index, (pgoff_t)PAGEVEC_SIZE-1) + 1);
984 if (nr_pages == 0)
985 break;
David Howells811d7362006-08-29 19:06:09 +0100986
David Howells811d7362006-08-29 19:06:09 +0100987 for (i = 0; i < nr_pages; i++) {
988 struct page *page = pvec.pages[i];
989
Nick Piggind5482cd2009-01-06 14:39:11 -0800990 /*
991 * At this point, the page may be truncated or
992 * invalidated (changing page->mapping to NULL), or
993 * even swizzled back from swapper_space to tmpfs file
994 * mapping. However, page->index will not change
995 * because we have a reference on the page.
996 */
997 if (page->index > end) {
998 /*
999 * can't be range_cyclic (1st pass) because
1000 * end == -1 in that case.
1001 */
1002 done = 1;
1003 break;
1004 }
1005
Nick Pigginbd19e012009-01-06 14:39:06 -08001006 done_index = page->index + 1;
1007
David Howells811d7362006-08-29 19:06:09 +01001008 lock_page(page);
1009
Nick Piggin5a3d5c92009-01-06 14:39:09 -08001010 /*
1011 * Page truncated or invalidated. We can freely skip it
1012 * then, even for data integrity operations: the page
1013 * has disappeared concurrently, so there could be no
1014 * real expectation of this data interity operation
1015 * even if there is now a new, dirty page at the same
1016 * pagecache address.
1017 */
David Howells811d7362006-08-29 19:06:09 +01001018 if (unlikely(page->mapping != mapping)) {
Nick Piggin5a3d5c92009-01-06 14:39:09 -08001019continue_unlock:
David Howells811d7362006-08-29 19:06:09 +01001020 unlock_page(page);
1021 continue;
1022 }
1023
Nick Piggin515f4a02009-01-06 14:39:10 -08001024 if (!PageDirty(page)) {
1025 /* someone wrote it for us */
1026 goto continue_unlock;
1027 }
David Howells811d7362006-08-29 19:06:09 +01001028
Nick Piggin515f4a02009-01-06 14:39:10 -08001029 if (PageWriteback(page)) {
1030 if (wbc->sync_mode != WB_SYNC_NONE)
1031 wait_on_page_writeback(page);
1032 else
1033 goto continue_unlock;
1034 }
1035
1036 BUG_ON(PageWriteback(page));
1037 if (!clear_page_dirty_for_io(page))
Nick Piggin5a3d5c92009-01-06 14:39:09 -08001038 goto continue_unlock;
David Howells811d7362006-08-29 19:06:09 +01001039
Miklos Szeredi0ea97182007-05-10 22:22:51 -07001040 ret = (*writepage)(page, wbc, data);
Nick Piggin00266772009-01-06 14:39:06 -08001041 if (unlikely(ret)) {
1042 if (ret == AOP_WRITEPAGE_ACTIVATE) {
1043 unlock_page(page);
1044 ret = 0;
1045 } else {
1046 /*
1047 * done_index is set past this page,
1048 * so media errors will not choke
1049 * background writeout for the entire
1050 * file. This has consequences for
1051 * range_cyclic semantics (ie. it may
1052 * not be suitable for data integrity
1053 * writeout).
1054 */
1055 done = 1;
1056 break;
1057 }
1058 }
David Howells811d7362006-08-29 19:06:09 +01001059
Federico Cuello89e12192009-02-11 13:04:39 -08001060 if (nr_to_write > 0) {
Artem Bityutskiydcf6a792009-02-02 18:33:49 +02001061 nr_to_write--;
Federico Cuello89e12192009-02-11 13:04:39 -08001062 if (nr_to_write == 0 &&
1063 wbc->sync_mode == WB_SYNC_NONE) {
1064 /*
1065 * We stop writing back only if we are
1066 * not doing integrity sync. In case of
1067 * integrity sync we have to keep going
1068 * because someone may be concurrently
1069 * dirtying pages, and we might have
1070 * synced a lot of newly appeared dirty
1071 * pages, but have not synced all of the
1072 * old dirty pages.
1073 */
1074 done = 1;
1075 break;
1076 }
Nick Piggin05fe4782009-01-06 14:39:08 -08001077 }
Artem Bityutskiydcf6a792009-02-02 18:33:49 +02001078
David Howells811d7362006-08-29 19:06:09 +01001079 if (wbc->nonblocking && bdi_write_congested(bdi)) {
1080 wbc->encountered_congestion = 1;
1081 done = 1;
Andrew Morton82fd1a92009-01-06 14:39:11 -08001082 break;
David Howells811d7362006-08-29 19:06:09 +01001083 }
1084 }
1085 pagevec_release(&pvec);
1086 cond_resched();
1087 }
Nick Piggin3a4c6802009-02-12 04:34:23 +01001088 if (!cycled && !done) {
David Howells811d7362006-08-29 19:06:09 +01001089 /*
Nick Piggin31a12662009-01-06 14:39:04 -08001090 * range_cyclic:
David Howells811d7362006-08-29 19:06:09 +01001091 * We hit the last page and there is more work to be done: wrap
1092 * back to the start of the file
1093 */
Nick Piggin31a12662009-01-06 14:39:04 -08001094 cycled = 1;
David Howells811d7362006-08-29 19:06:09 +01001095 index = 0;
Nick Piggin31a12662009-01-06 14:39:04 -08001096 end = writeback_index - 1;
David Howells811d7362006-08-29 19:06:09 +01001097 goto retry;
1098 }
Aneesh Kumar K.V17bc6c32008-10-16 10:09:17 -04001099 if (!wbc->no_nrwrite_index_update) {
1100 if (wbc->range_cyclic || (range_whole && nr_to_write > 0))
Nick Pigginbd19e012009-01-06 14:39:06 -08001101 mapping->writeback_index = done_index;
Aneesh Kumar K.V17bc6c32008-10-16 10:09:17 -04001102 wbc->nr_to_write = nr_to_write;
1103 }
Aneesh Kumar K.V06d6cf62008-07-11 19:27:31 -04001104
David Howells811d7362006-08-29 19:06:09 +01001105 return ret;
1106}
Miklos Szeredi0ea97182007-05-10 22:22:51 -07001107EXPORT_SYMBOL(write_cache_pages);
1108
1109/*
1110 * Function used by generic_writepages to call the real writepage
1111 * function and set the mapping flags on error
1112 */
1113static int __writepage(struct page *page, struct writeback_control *wbc,
1114 void *data)
1115{
1116 struct address_space *mapping = data;
1117 int ret = mapping->a_ops->writepage(page, wbc);
1118 mapping_set_error(mapping, ret);
1119 return ret;
1120}
1121
1122/**
1123 * generic_writepages - walk the list of dirty pages of the given address space and writepage() all of them.
1124 * @mapping: address space structure to write
1125 * @wbc: subtract the number of written pages from *@wbc->nr_to_write
1126 *
1127 * This is a library function, which implements the writepages()
1128 * address_space_operation.
1129 */
1130int generic_writepages(struct address_space *mapping,
1131 struct writeback_control *wbc)
1132{
1133 /* deal with chardevs and other special file */
1134 if (!mapping->a_ops->writepage)
1135 return 0;
1136
1137 return write_cache_pages(mapping, wbc, __writepage, mapping);
1138}
David Howells811d7362006-08-29 19:06:09 +01001139
1140EXPORT_SYMBOL(generic_writepages);
1141
Linus Torvalds1da177e2005-04-16 15:20:36 -07001142int do_writepages(struct address_space *mapping, struct writeback_control *wbc)
1143{
Andrew Morton22905f72005-11-16 15:07:01 -08001144 int ret;
1145
Linus Torvalds1da177e2005-04-16 15:20:36 -07001146 if (wbc->nr_to_write <= 0)
1147 return 0;
Andrew Morton22905f72005-11-16 15:07:01 -08001148 wbc->for_writepages = 1;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001149 if (mapping->a_ops->writepages)
Peter Zijlstrad08b3852006-09-25 23:30:57 -07001150 ret = mapping->a_ops->writepages(mapping, wbc);
Andrew Morton22905f72005-11-16 15:07:01 -08001151 else
1152 ret = generic_writepages(mapping, wbc);
1153 wbc->for_writepages = 0;
1154 return ret;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001155}
1156
1157/**
1158 * write_one_page - write out a single page and optionally wait on I/O
Martin Waitz67be2dd2005-05-01 08:59:26 -07001159 * @page: the page to write
1160 * @wait: if true, wait on writeout
Linus Torvalds1da177e2005-04-16 15:20:36 -07001161 *
1162 * The page must be locked by the caller and will be unlocked upon return.
1163 *
1164 * write_one_page() returns a negative error code if I/O failed.
1165 */
1166int write_one_page(struct page *page, int wait)
1167{
1168 struct address_space *mapping = page->mapping;
1169 int ret = 0;
1170 struct writeback_control wbc = {
1171 .sync_mode = WB_SYNC_ALL,
1172 .nr_to_write = 1,
1173 };
1174
1175 BUG_ON(!PageLocked(page));
1176
1177 if (wait)
1178 wait_on_page_writeback(page);
1179
1180 if (clear_page_dirty_for_io(page)) {
1181 page_cache_get(page);
1182 ret = mapping->a_ops->writepage(page, &wbc);
1183 if (ret == 0 && wait) {
1184 wait_on_page_writeback(page);
1185 if (PageError(page))
1186 ret = -EIO;
1187 }
1188 page_cache_release(page);
1189 } else {
1190 unlock_page(page);
1191 }
1192 return ret;
1193}
1194EXPORT_SYMBOL(write_one_page);
1195
1196/*
Ken Chen76719322007-02-10 01:43:15 -08001197 * For address_spaces which do not use buffers nor write back.
1198 */
1199int __set_page_dirty_no_writeback(struct page *page)
1200{
1201 if (!PageDirty(page))
1202 SetPageDirty(page);
1203 return 0;
1204}
1205
1206/*
Edward Shishkine3a7cca2009-03-31 15:19:39 -07001207 * Helper function for set_page_dirty family.
1208 * NOTE: This relies on being atomic wrt interrupts.
1209 */
1210void account_page_dirtied(struct page *page, struct address_space *mapping)
1211{
1212 if (mapping_cap_account_dirty(mapping)) {
1213 __inc_zone_page_state(page, NR_FILE_DIRTY);
1214 __inc_bdi_stat(mapping->backing_dev_info, BDI_RECLAIMABLE);
1215 task_dirty_inc(current);
1216 task_io_account_write(PAGE_CACHE_SIZE);
1217 }
1218}
1219
1220/*
Linus Torvalds1da177e2005-04-16 15:20:36 -07001221 * For address_spaces which do not use buffers. Just tag the page as dirty in
1222 * its radix tree.
1223 *
1224 * This is also used when a single buffer is being dirtied: we want to set the
1225 * page dirty in that case, but not all the buffers. This is a "bottom-up"
1226 * dirtying, whereas __set_page_dirty_buffers() is a "top-down" dirtying.
1227 *
1228 * Most callers have locked the page, which pins the address_space in memory.
1229 * But zap_pte_range() does not lock the page, however in that case the
1230 * mapping is pinned by the vma's ->vm_file reference.
1231 *
1232 * We take care to handle the case where the page was truncated from the
Simon Arlott183ff222007-10-20 01:27:18 +02001233 * mapping by re-checking page_mapping() inside tree_lock.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001234 */
1235int __set_page_dirty_nobuffers(struct page *page)
1236{
Linus Torvalds1da177e2005-04-16 15:20:36 -07001237 if (!TestSetPageDirty(page)) {
1238 struct address_space *mapping = page_mapping(page);
1239 struct address_space *mapping2;
1240
Andrew Morton8c085402006-12-10 02:19:24 -08001241 if (!mapping)
1242 return 1;
1243
Nick Piggin19fd6232008-07-25 19:45:32 -07001244 spin_lock_irq(&mapping->tree_lock);
Andrew Morton8c085402006-12-10 02:19:24 -08001245 mapping2 = page_mapping(page);
1246 if (mapping2) { /* Race with truncate? */
1247 BUG_ON(mapping2 != mapping);
Nick Piggin787d2212007-07-17 04:03:34 -07001248 WARN_ON_ONCE(!PagePrivate(page) && !PageUptodate(page));
Edward Shishkine3a7cca2009-03-31 15:19:39 -07001249 account_page_dirtied(page, mapping);
Andrew Morton8c085402006-12-10 02:19:24 -08001250 radix_tree_tag_set(&mapping->page_tree,
1251 page_index(page), PAGECACHE_TAG_DIRTY);
1252 }
Nick Piggin19fd6232008-07-25 19:45:32 -07001253 spin_unlock_irq(&mapping->tree_lock);
Andrew Morton8c085402006-12-10 02:19:24 -08001254 if (mapping->host) {
1255 /* !PageAnon && !swapper_space */
1256 __mark_inode_dirty(mapping->host, I_DIRTY_PAGES);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001257 }
Andrew Morton4741c9f2006-03-24 03:18:11 -08001258 return 1;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001259 }
Andrew Morton4741c9f2006-03-24 03:18:11 -08001260 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001261}
1262EXPORT_SYMBOL(__set_page_dirty_nobuffers);
1263
1264/*
1265 * When a writepage implementation decides that it doesn't want to write this
1266 * page for some reason, it should redirty the locked page via
1267 * redirty_page_for_writepage() and it should then unlock the page and return 0
1268 */
1269int redirty_page_for_writepage(struct writeback_control *wbc, struct page *page)
1270{
1271 wbc->pages_skipped++;
1272 return __set_page_dirty_nobuffers(page);
1273}
1274EXPORT_SYMBOL(redirty_page_for_writepage);
1275
1276/*
1277 * If the mapping doesn't provide a set_page_dirty a_op, then
1278 * just fall through and assume that it wants buffer_heads.
1279 */
Nick Piggin1cf6e7d2009-02-18 14:48:18 -08001280int set_page_dirty(struct page *page)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001281{
1282 struct address_space *mapping = page_mapping(page);
1283
1284 if (likely(mapping)) {
1285 int (*spd)(struct page *) = mapping->a_ops->set_page_dirty;
David Howells93614012006-09-30 20:45:40 +02001286#ifdef CONFIG_BLOCK
1287 if (!spd)
1288 spd = __set_page_dirty_buffers;
1289#endif
1290 return (*spd)(page);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001291 }
Andrew Morton4741c9f2006-03-24 03:18:11 -08001292 if (!PageDirty(page)) {
1293 if (!TestSetPageDirty(page))
1294 return 1;
1295 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001296 return 0;
1297}
1298EXPORT_SYMBOL(set_page_dirty);
1299
1300/*
1301 * set_page_dirty() is racy if the caller has no reference against
1302 * page->mapping->host, and if the page is unlocked. This is because another
1303 * CPU could truncate the page off the mapping and then free the mapping.
1304 *
1305 * Usually, the page _is_ locked, or the caller is a user-space process which
1306 * holds a reference on the inode by having an open file.
1307 *
1308 * In other cases, the page should be locked before running set_page_dirty().
1309 */
1310int set_page_dirty_lock(struct page *page)
1311{
1312 int ret;
1313
Nick Piggindb376482006-09-25 23:31:24 -07001314 lock_page_nosync(page);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001315 ret = set_page_dirty(page);
1316 unlock_page(page);
1317 return ret;
1318}
1319EXPORT_SYMBOL(set_page_dirty_lock);
1320
1321/*
Linus Torvalds1da177e2005-04-16 15:20:36 -07001322 * Clear a page's dirty flag, while caring for dirty memory accounting.
1323 * Returns true if the page was previously dirty.
1324 *
1325 * This is for preparing to put the page under writeout. We leave the page
1326 * tagged as dirty in the radix tree so that a concurrent write-for-sync
1327 * can discover it via a PAGECACHE_TAG_DIRTY walk. The ->writepage
1328 * implementation will run either set_page_writeback() or set_page_dirty(),
1329 * at which stage we bring the page's dirty flag and radix-tree dirty tag
1330 * back into sync.
1331 *
1332 * This incoherency between the page's dirty flag and radix-tree tag is
1333 * unfortunate, but it only exists while the page is locked.
1334 */
1335int clear_page_dirty_for_io(struct page *page)
1336{
1337 struct address_space *mapping = page_mapping(page);
1338
Nick Piggin79352892007-07-19 01:47:22 -07001339 BUG_ON(!PageLocked(page));
1340
Fengguang Wufe3cba12007-07-19 01:48:07 -07001341 ClearPageReclaim(page);
Linus Torvalds7658cc22006-12-29 10:00:58 -08001342 if (mapping && mapping_cap_account_dirty(mapping)) {
1343 /*
1344 * Yes, Virginia, this is indeed insane.
1345 *
1346 * We use this sequence to make sure that
1347 * (a) we account for dirty stats properly
1348 * (b) we tell the low-level filesystem to
1349 * mark the whole page dirty if it was
1350 * dirty in a pagetable. Only to then
1351 * (c) clean the page again and return 1 to
1352 * cause the writeback.
1353 *
1354 * This way we avoid all nasty races with the
1355 * dirty bit in multiple places and clearing
1356 * them concurrently from different threads.
1357 *
1358 * Note! Normally the "set_page_dirty(page)"
1359 * has no effect on the actual dirty bit - since
1360 * that will already usually be set. But we
1361 * need the side effects, and it can help us
1362 * avoid races.
1363 *
1364 * We basically use the page "master dirty bit"
1365 * as a serialization point for all the different
1366 * threads doing their things.
Linus Torvalds7658cc22006-12-29 10:00:58 -08001367 */
1368 if (page_mkclean(page))
1369 set_page_dirty(page);
Nick Piggin79352892007-07-19 01:47:22 -07001370 /*
1371 * We carefully synchronise fault handlers against
1372 * installing a dirty pte and marking the page dirty
1373 * at this point. We do this by having them hold the
1374 * page lock at some point after installing their
1375 * pte, but before marking the page dirty.
1376 * Pages are always locked coming in here, so we get
1377 * the desired exclusion. See mm/memory.c:do_wp_page()
1378 * for more comments.
1379 */
Linus Torvalds7658cc22006-12-29 10:00:58 -08001380 if (TestClearPageDirty(page)) {
Andrew Morton8c085402006-12-10 02:19:24 -08001381 dec_zone_page_state(page, NR_FILE_DIRTY);
Peter Zijlstrac9e51e42007-10-16 23:25:47 -07001382 dec_bdi_stat(mapping->backing_dev_info,
1383 BDI_RECLAIMABLE);
Linus Torvalds7658cc22006-12-29 10:00:58 -08001384 return 1;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001385 }
Linus Torvalds7658cc22006-12-29 10:00:58 -08001386 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001387 }
Linus Torvalds7658cc22006-12-29 10:00:58 -08001388 return TestClearPageDirty(page);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001389}
Hans Reiser58bb01a2005-11-18 01:10:53 -08001390EXPORT_SYMBOL(clear_page_dirty_for_io);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001391
1392int test_clear_page_writeback(struct page *page)
1393{
1394 struct address_space *mapping = page_mapping(page);
1395 int ret;
1396
1397 if (mapping) {
Peter Zijlstra69cb51d2007-10-16 23:25:48 -07001398 struct backing_dev_info *bdi = mapping->backing_dev_info;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001399 unsigned long flags;
1400
Nick Piggin19fd6232008-07-25 19:45:32 -07001401 spin_lock_irqsave(&mapping->tree_lock, flags);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001402 ret = TestClearPageWriteback(page);
Peter Zijlstra69cb51d2007-10-16 23:25:48 -07001403 if (ret) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001404 radix_tree_tag_clear(&mapping->page_tree,
1405 page_index(page),
1406 PAGECACHE_TAG_WRITEBACK);
Miklos Szeredie4ad08f2008-04-30 00:54:37 -07001407 if (bdi_cap_account_writeback(bdi)) {
Peter Zijlstra69cb51d2007-10-16 23:25:48 -07001408 __dec_bdi_stat(bdi, BDI_WRITEBACK);
Peter Zijlstra04fbfdc2007-10-16 23:25:50 -07001409 __bdi_writeout_inc(bdi);
1410 }
Peter Zijlstra69cb51d2007-10-16 23:25:48 -07001411 }
Nick Piggin19fd6232008-07-25 19:45:32 -07001412 spin_unlock_irqrestore(&mapping->tree_lock, flags);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001413 } else {
1414 ret = TestClearPageWriteback(page);
1415 }
Andrew Mortond688abf2007-07-19 01:49:17 -07001416 if (ret)
1417 dec_zone_page_state(page, NR_WRITEBACK);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001418 return ret;
1419}
1420
1421int test_set_page_writeback(struct page *page)
1422{
1423 struct address_space *mapping = page_mapping(page);
1424 int ret;
1425
1426 if (mapping) {
Peter Zijlstra69cb51d2007-10-16 23:25:48 -07001427 struct backing_dev_info *bdi = mapping->backing_dev_info;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001428 unsigned long flags;
1429
Nick Piggin19fd6232008-07-25 19:45:32 -07001430 spin_lock_irqsave(&mapping->tree_lock, flags);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001431 ret = TestSetPageWriteback(page);
Peter Zijlstra69cb51d2007-10-16 23:25:48 -07001432 if (!ret) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001433 radix_tree_tag_set(&mapping->page_tree,
1434 page_index(page),
1435 PAGECACHE_TAG_WRITEBACK);
Miklos Szeredie4ad08f2008-04-30 00:54:37 -07001436 if (bdi_cap_account_writeback(bdi))
Peter Zijlstra69cb51d2007-10-16 23:25:48 -07001437 __inc_bdi_stat(bdi, BDI_WRITEBACK);
1438 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001439 if (!PageDirty(page))
1440 radix_tree_tag_clear(&mapping->page_tree,
1441 page_index(page),
1442 PAGECACHE_TAG_DIRTY);
Nick Piggin19fd6232008-07-25 19:45:32 -07001443 spin_unlock_irqrestore(&mapping->tree_lock, flags);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001444 } else {
1445 ret = TestSetPageWriteback(page);
1446 }
Andrew Mortond688abf2007-07-19 01:49:17 -07001447 if (!ret)
1448 inc_zone_page_state(page, NR_WRITEBACK);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001449 return ret;
1450
1451}
1452EXPORT_SYMBOL(test_set_page_writeback);
1453
1454/*
Nick Piggin00128182007-10-16 01:24:40 -07001455 * Return true if any of the pages in the mapping are marked with the
Linus Torvalds1da177e2005-04-16 15:20:36 -07001456 * passed tag.
1457 */
1458int mapping_tagged(struct address_space *mapping, int tag)
1459{
Linus Torvalds1da177e2005-04-16 15:20:36 -07001460 int ret;
Nick Piggin00128182007-10-16 01:24:40 -07001461 rcu_read_lock();
Linus Torvalds1da177e2005-04-16 15:20:36 -07001462 ret = radix_tree_tagged(&mapping->page_tree, tag);
Nick Piggin00128182007-10-16 01:24:40 -07001463 rcu_read_unlock();
Linus Torvalds1da177e2005-04-16 15:20:36 -07001464 return ret;
1465}
1466EXPORT_SYMBOL(mapping_tagged);