blob: b4edfe7ce06c1bf5f6d692235eac4e07f9068490 [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>
Dave Chinner028c2dd2010-07-07 13:24:07 +100037#include <trace/events/writeback.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070038
39/*
Linus Torvalds1da177e2005-04-16 15:20:36 -070040 * After a CPU has dirtied this many pages, balance_dirty_pages_ratelimited
41 * will look to see if it needs to force writeback or throttling.
42 */
43static long ratelimit_pages = 32;
44
Linus Torvalds1da177e2005-04-16 15:20:36 -070045/*
46 * When balance_dirty_pages decides that the caller needs to perform some
47 * non-background writeback, this is how many pages it will attempt to write.
Wu Fengguang3a2e9a52009-09-23 21:56:00 +080048 * It should be somewhat larger than dirtied pages to ensure that reasonably
Linus Torvalds1da177e2005-04-16 15:20:36 -070049 * large amounts of I/O are submitted.
50 */
Wu Fengguang3a2e9a52009-09-23 21:56:00 +080051static inline long sync_writeback_pages(unsigned long dirtied)
Linus Torvalds1da177e2005-04-16 15:20:36 -070052{
Wu Fengguang3a2e9a52009-09-23 21:56:00 +080053 if (dirtied < ratelimit_pages)
54 dirtied = ratelimit_pages;
55
56 return dirtied + dirtied / 2;
Linus Torvalds1da177e2005-04-16 15:20:36 -070057}
58
59/* The following parameters are exported via /proc/sys/vm */
60
61/*
Jens Axboe5b0830c2009-09-23 19:37:09 +020062 * Start background writeback (via writeback threads) at this percentage
Linus Torvalds1da177e2005-04-16 15:20:36 -070063 */
Wu Fengguang1b5e62b2009-03-23 08:57:38 +080064int dirty_background_ratio = 10;
Linus Torvalds1da177e2005-04-16 15:20:36 -070065
66/*
David Rientjes2da02992009-01-06 14:39:31 -080067 * dirty_background_bytes starts at 0 (disabled) so that it is a function of
68 * dirty_background_ratio * the amount of dirtyable memory
69 */
70unsigned long dirty_background_bytes;
71
72/*
Bron Gondwana195cf4532008-02-04 22:29:20 -080073 * free highmem will not be subtracted from the total free memory
74 * for calculating free ratios if vm_highmem_is_dirtyable is true
75 */
76int vm_highmem_is_dirtyable;
77
78/*
Linus Torvalds1da177e2005-04-16 15:20:36 -070079 * The generator of dirty data starts writeback at this percentage
80 */
Wu Fengguang1b5e62b2009-03-23 08:57:38 +080081int vm_dirty_ratio = 20;
Linus Torvalds1da177e2005-04-16 15:20:36 -070082
83/*
David Rientjes2da02992009-01-06 14:39:31 -080084 * vm_dirty_bytes starts at 0 (disabled) so that it is a function of
85 * vm_dirty_ratio * the amount of dirtyable memory
86 */
87unsigned long vm_dirty_bytes;
88
89/*
Alexey Dobriyan704503d2009-03-31 15:23:18 -070090 * The interval between `kupdate'-style writebacks
Linus Torvalds1da177e2005-04-16 15:20:36 -070091 */
Toshiyuki Okajima22ef37e2009-05-16 22:56:28 -070092unsigned int dirty_writeback_interval = 5 * 100; /* centiseconds */
Linus Torvalds1da177e2005-04-16 15:20:36 -070093
94/*
Alexey Dobriyan704503d2009-03-31 15:23:18 -070095 * The longest time for which data is allowed to remain dirty
Linus Torvalds1da177e2005-04-16 15:20:36 -070096 */
Toshiyuki Okajima22ef37e2009-05-16 22:56:28 -070097unsigned int dirty_expire_interval = 30 * 100; /* centiseconds */
Linus Torvalds1da177e2005-04-16 15:20:36 -070098
99/*
100 * Flag that makes the machine dump writes/reads and block dirtyings.
101 */
102int block_dump;
103
104/*
Bart Samweled5b43f2006-03-24 03:15:49 -0800105 * Flag that puts the machine in "laptop mode". Doubles as a timeout in jiffies:
106 * a full sync is triggered after this time elapses without any disk activity.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700107 */
108int laptop_mode;
109
110EXPORT_SYMBOL(laptop_mode);
111
112/* End of sysctl-exported parameters */
113
114
Linus Torvalds1da177e2005-04-16 15:20:36 -0700115/*
Peter Zijlstra04fbfdc2007-10-16 23:25:50 -0700116 * Scale the writeback cache size proportional to the relative writeout speeds.
117 *
118 * We do this by keeping a floating proportion between BDIs, based on page
119 * writeback completions [end_page_writeback()]. Those devices that write out
120 * pages fastest will get the larger share, while the slower will get a smaller
121 * share.
122 *
123 * We use page writeout completions because we are interested in getting rid of
124 * dirty pages. Having them written out is the primary goal.
125 *
126 * We introduce a concept of time, a period over which we measure these events,
127 * because demand can/will vary over time. The length of this period itself is
128 * measured in page writeback completions.
129 *
130 */
131static struct prop_descriptor vm_completions;
Peter Zijlstra3e26c142007-10-16 23:25:50 -0700132static struct prop_descriptor vm_dirties;
Peter Zijlstra04fbfdc2007-10-16 23:25:50 -0700133
Peter Zijlstra04fbfdc2007-10-16 23:25:50 -0700134/*
135 * couple the period to the dirty_ratio:
136 *
137 * period/2 ~ roundup_pow_of_two(dirty limit)
138 */
139static int calc_period_shift(void)
140{
141 unsigned long dirty_total;
142
David Rientjes2da02992009-01-06 14:39:31 -0800143 if (vm_dirty_bytes)
144 dirty_total = vm_dirty_bytes / PAGE_SIZE;
145 else
146 dirty_total = (vm_dirty_ratio * determine_dirtyable_memory()) /
147 100;
Peter Zijlstra04fbfdc2007-10-16 23:25:50 -0700148 return 2 + ilog2(dirty_total - 1);
149}
150
151/*
David Rientjes2da02992009-01-06 14:39:31 -0800152 * update the period when the dirty threshold changes.
Peter Zijlstra04fbfdc2007-10-16 23:25:50 -0700153 */
David Rientjes2da02992009-01-06 14:39:31 -0800154static void update_completion_period(void)
155{
156 int shift = calc_period_shift();
157 prop_change_shift(&vm_completions, shift);
158 prop_change_shift(&vm_dirties, shift);
159}
160
161int dirty_background_ratio_handler(struct ctl_table *table, int write,
Alexey Dobriyan8d65af72009-09-23 15:57:19 -0700162 void __user *buffer, size_t *lenp,
David Rientjes2da02992009-01-06 14:39:31 -0800163 loff_t *ppos)
164{
165 int ret;
166
Alexey Dobriyan8d65af72009-09-23 15:57:19 -0700167 ret = proc_dointvec_minmax(table, write, buffer, lenp, ppos);
David Rientjes2da02992009-01-06 14:39:31 -0800168 if (ret == 0 && write)
169 dirty_background_bytes = 0;
170 return ret;
171}
172
173int dirty_background_bytes_handler(struct ctl_table *table, int write,
Alexey Dobriyan8d65af72009-09-23 15:57:19 -0700174 void __user *buffer, size_t *lenp,
David Rientjes2da02992009-01-06 14:39:31 -0800175 loff_t *ppos)
176{
177 int ret;
178
Alexey Dobriyan8d65af72009-09-23 15:57:19 -0700179 ret = proc_doulongvec_minmax(table, write, buffer, lenp, ppos);
David Rientjes2da02992009-01-06 14:39:31 -0800180 if (ret == 0 && write)
181 dirty_background_ratio = 0;
182 return ret;
183}
184
Peter Zijlstra04fbfdc2007-10-16 23:25:50 -0700185int dirty_ratio_handler(struct ctl_table *table, int write,
Alexey Dobriyan8d65af72009-09-23 15:57:19 -0700186 void __user *buffer, size_t *lenp,
Peter Zijlstra04fbfdc2007-10-16 23:25:50 -0700187 loff_t *ppos)
188{
189 int old_ratio = vm_dirty_ratio;
David Rientjes2da02992009-01-06 14:39:31 -0800190 int ret;
191
Alexey Dobriyan8d65af72009-09-23 15:57:19 -0700192 ret = proc_dointvec_minmax(table, write, buffer, lenp, ppos);
Peter Zijlstra04fbfdc2007-10-16 23:25:50 -0700193 if (ret == 0 && write && vm_dirty_ratio != old_ratio) {
David Rientjes2da02992009-01-06 14:39:31 -0800194 update_completion_period();
195 vm_dirty_bytes = 0;
196 }
197 return ret;
198}
199
200
201int dirty_bytes_handler(struct ctl_table *table, int write,
Alexey Dobriyan8d65af72009-09-23 15:57:19 -0700202 void __user *buffer, size_t *lenp,
David Rientjes2da02992009-01-06 14:39:31 -0800203 loff_t *ppos)
204{
Sven Wegenerfc3501d2009-02-11 13:04:23 -0800205 unsigned long old_bytes = vm_dirty_bytes;
David Rientjes2da02992009-01-06 14:39:31 -0800206 int ret;
207
Alexey Dobriyan8d65af72009-09-23 15:57:19 -0700208 ret = proc_doulongvec_minmax(table, write, buffer, lenp, ppos);
David Rientjes2da02992009-01-06 14:39:31 -0800209 if (ret == 0 && write && vm_dirty_bytes != old_bytes) {
210 update_completion_period();
211 vm_dirty_ratio = 0;
Peter Zijlstra04fbfdc2007-10-16 23:25:50 -0700212 }
213 return ret;
214}
215
216/*
217 * Increment the BDI's writeout completion count and the global writeout
218 * completion count. Called from test_clear_page_writeback().
219 */
220static inline void __bdi_writeout_inc(struct backing_dev_info *bdi)
221{
Peter Zijlstraa42dde02008-04-30 00:54:36 -0700222 __prop_inc_percpu_max(&vm_completions, &bdi->completions,
223 bdi->max_prop_frac);
Peter Zijlstra04fbfdc2007-10-16 23:25:50 -0700224}
225
Miklos Szeredidd5656e2008-04-30 00:54:37 -0700226void bdi_writeout_inc(struct backing_dev_info *bdi)
227{
228 unsigned long flags;
229
230 local_irq_save(flags);
231 __bdi_writeout_inc(bdi);
232 local_irq_restore(flags);
233}
234EXPORT_SYMBOL_GPL(bdi_writeout_inc);
235
Nick Piggin1cf6e7d2009-02-18 14:48:18 -0800236void task_dirty_inc(struct task_struct *tsk)
Peter Zijlstra3e26c142007-10-16 23:25:50 -0700237{
238 prop_inc_single(&vm_dirties, &tsk->dirties);
239}
240
Peter Zijlstra04fbfdc2007-10-16 23:25:50 -0700241/*
242 * Obtain an accurate fraction of the BDI's portion.
243 */
244static void bdi_writeout_fraction(struct backing_dev_info *bdi,
245 long *numerator, long *denominator)
246{
247 if (bdi_cap_writeback_dirty(bdi)) {
248 prop_fraction_percpu(&vm_completions, &bdi->completions,
249 numerator, denominator);
250 } else {
251 *numerator = 0;
252 *denominator = 1;
253 }
254}
255
Peter Zijlstra3e26c142007-10-16 23:25:50 -0700256static inline void task_dirties_fraction(struct task_struct *tsk,
257 long *numerator, long *denominator)
258{
259 prop_fraction_single(&vm_dirties, &tsk->dirties,
260 numerator, denominator);
261}
262
263/*
Wu Fengguang1babe182010-08-11 14:17:40 -0700264 * task_dirty_limit - scale down dirty throttling threshold for one task
Peter Zijlstra3e26c142007-10-16 23:25:50 -0700265 *
266 * task specific dirty limit:
267 *
268 * dirty -= (dirty/8) * p_{t}
Wu Fengguang1babe182010-08-11 14:17:40 -0700269 *
270 * To protect light/slow dirtying tasks from heavier/fast ones, we start
271 * throttling individual tasks before reaching the bdi dirty limit.
272 * Relatively low thresholds will be allocated to heavy dirtiers. So when
273 * dirty pages grow large, heavy dirtiers will be throttled first, which will
274 * effectively curb the growth of dirty pages. Light dirtiers with high enough
275 * dirty threshold may never get throttled.
Peter Zijlstra3e26c142007-10-16 23:25:50 -0700276 */
Wu Fengguang16c40422010-08-11 14:17:39 -0700277static unsigned long task_dirty_limit(struct task_struct *tsk,
278 unsigned long bdi_dirty)
Peter Zijlstra3e26c142007-10-16 23:25:50 -0700279{
280 long numerator, denominator;
Wu Fengguang16c40422010-08-11 14:17:39 -0700281 unsigned long dirty = bdi_dirty;
Peter Zijlstra3e26c142007-10-16 23:25:50 -0700282 u64 inv = dirty >> 3;
283
284 task_dirties_fraction(tsk, &numerator, &denominator);
285 inv *= numerator;
286 do_div(inv, denominator);
287
288 dirty -= inv;
Peter Zijlstra3e26c142007-10-16 23:25:50 -0700289
Wu Fengguang16c40422010-08-11 14:17:39 -0700290 return max(dirty, bdi_dirty/2);
Peter Zijlstra3e26c142007-10-16 23:25:50 -0700291}
292
Peter Zijlstra04fbfdc2007-10-16 23:25:50 -0700293/*
Peter Zijlstra189d3c42008-04-30 00:54:35 -0700294 *
295 */
Peter Zijlstra189d3c42008-04-30 00:54:35 -0700296static unsigned int bdi_min_ratio;
297
298int bdi_set_min_ratio(struct backing_dev_info *bdi, unsigned int min_ratio)
299{
300 int ret = 0;
Peter Zijlstra189d3c42008-04-30 00:54:35 -0700301
Jens Axboecfc4ba52009-09-14 13:12:40 +0200302 spin_lock_bh(&bdi_lock);
Peter Zijlstraa42dde02008-04-30 00:54:36 -0700303 if (min_ratio > bdi->max_ratio) {
Peter Zijlstra189d3c42008-04-30 00:54:35 -0700304 ret = -EINVAL;
Peter Zijlstraa42dde02008-04-30 00:54:36 -0700305 } else {
306 min_ratio -= bdi->min_ratio;
307 if (bdi_min_ratio + min_ratio < 100) {
308 bdi_min_ratio += min_ratio;
309 bdi->min_ratio += min_ratio;
310 } else {
311 ret = -EINVAL;
312 }
313 }
Jens Axboecfc4ba52009-09-14 13:12:40 +0200314 spin_unlock_bh(&bdi_lock);
Peter Zijlstra189d3c42008-04-30 00:54:35 -0700315
316 return ret;
317}
318
Peter Zijlstraa42dde02008-04-30 00:54:36 -0700319int bdi_set_max_ratio(struct backing_dev_info *bdi, unsigned max_ratio)
320{
Peter Zijlstraa42dde02008-04-30 00:54:36 -0700321 int ret = 0;
322
323 if (max_ratio > 100)
324 return -EINVAL;
325
Jens Axboecfc4ba52009-09-14 13:12:40 +0200326 spin_lock_bh(&bdi_lock);
Peter Zijlstraa42dde02008-04-30 00:54:36 -0700327 if (bdi->min_ratio > max_ratio) {
328 ret = -EINVAL;
329 } else {
330 bdi->max_ratio = max_ratio;
331 bdi->max_prop_frac = (PROP_FRAC_BASE * max_ratio) / 100;
332 }
Jens Axboecfc4ba52009-09-14 13:12:40 +0200333 spin_unlock_bh(&bdi_lock);
Peter Zijlstraa42dde02008-04-30 00:54:36 -0700334
335 return ret;
336}
337EXPORT_SYMBOL(bdi_set_max_ratio);
338
Peter Zijlstra189d3c42008-04-30 00:54:35 -0700339/*
Linus Torvalds1da177e2005-04-16 15:20:36 -0700340 * Work out the current dirty-memory clamping and background writeout
341 * thresholds.
342 *
343 * The main aim here is to lower them aggressively if there is a lot of mapped
344 * memory around. To avoid stressing page reclaim with lots of unreclaimable
345 * pages. It is better to clamp down on writers than to start swapping, and
346 * performing lots of scanning.
347 *
348 * We only allow 1/2 of the currently-unmapped memory to be dirtied.
349 *
350 * We don't permit the clamping level to fall below 5% - that is getting rather
351 * excessive.
352 *
353 * We make sure that the background writeout level is below the adjusted
354 * clamping level.
355 */
Christoph Lameter1b424462007-05-06 14:48:59 -0700356
357static unsigned long highmem_dirtyable_memory(unsigned long total)
358{
359#ifdef CONFIG_HIGHMEM
360 int node;
361 unsigned long x = 0;
362
Lee Schermerhorn37b07e42007-10-16 01:25:39 -0700363 for_each_node_state(node, N_HIGH_MEMORY) {
Christoph Lameter1b424462007-05-06 14:48:59 -0700364 struct zone *z =
365 &NODE_DATA(node)->node_zones[ZONE_HIGHMEM];
366
Wu Fengguangadea02a2009-09-21 17:01:42 -0700367 x += zone_page_state(z, NR_FREE_PAGES) +
368 zone_reclaimable_pages(z);
Christoph Lameter1b424462007-05-06 14:48:59 -0700369 }
370 /*
371 * Make sure that the number of highmem pages is never larger
372 * than the number of the total dirtyable memory. This can only
373 * occur in very strange VM situations but we want to make sure
374 * that this does not occur.
375 */
376 return min(x, total);
377#else
378 return 0;
379#endif
380}
381
Steven Rostedt3eefae92008-05-12 21:21:04 +0200382/**
383 * determine_dirtyable_memory - amount of memory that may be used
384 *
385 * Returns the numebr of pages that can currently be freed and used
386 * by the kernel for direct mappings.
387 */
388unsigned long determine_dirtyable_memory(void)
Christoph Lameter1b424462007-05-06 14:48:59 -0700389{
390 unsigned long x;
391
Wu Fengguangadea02a2009-09-21 17:01:42 -0700392 x = global_page_state(NR_FREE_PAGES) + global_reclaimable_pages();
Bron Gondwana195cf4532008-02-04 22:29:20 -0800393
394 if (!vm_highmem_is_dirtyable)
395 x -= highmem_dirtyable_memory(x);
396
Christoph Lameter1b424462007-05-06 14:48:59 -0700397 return x + 1; /* Ensure that we never return 0 */
398}
399
Randy Dunlap03ab4502010-08-14 13:05:17 -0700400/*
Wu Fengguang1babe182010-08-11 14:17:40 -0700401 * global_dirty_limits - background-writeback and dirty-throttling thresholds
402 *
403 * Calculate the dirty thresholds based on sysctl parameters
404 * - vm.dirty_background_ratio or vm.dirty_background_bytes
405 * - vm.dirty_ratio or vm.dirty_bytes
406 * The dirty limits will be lifted by 1/4 for PF_LESS_THROTTLE (ie. nfsd) and
407 * runtime tasks.
408 */
Wu Fengguang16c40422010-08-11 14:17:39 -0700409void global_dirty_limits(unsigned long *pbackground, unsigned long *pdirty)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700410{
David Rientjes364aeb22009-01-06 14:39:29 -0800411 unsigned long background;
412 unsigned long dirty;
Christoph Lameter1b424462007-05-06 14:48:59 -0700413 unsigned long available_memory = determine_dirtyable_memory();
Linus Torvalds1da177e2005-04-16 15:20:36 -0700414 struct task_struct *tsk;
415
David Rientjes2da02992009-01-06 14:39:31 -0800416 if (vm_dirty_bytes)
417 dirty = DIV_ROUND_UP(vm_dirty_bytes, PAGE_SIZE);
Wu Fengguang4cbec4c2010-10-26 14:21:45 -0700418 else
419 dirty = (vm_dirty_ratio * available_memory) / 100;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700420
David Rientjes2da02992009-01-06 14:39:31 -0800421 if (dirty_background_bytes)
422 background = DIV_ROUND_UP(dirty_background_bytes, PAGE_SIZE);
423 else
424 background = (dirty_background_ratio * available_memory) / 100;
425
426 if (background >= dirty)
427 background = dirty / 2;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700428 tsk = current;
429 if (tsk->flags & PF_LESS_THROTTLE || rt_task(tsk)) {
430 background += background / 4;
431 dirty += dirty / 4;
432 }
433 *pbackground = background;
434 *pdirty = dirty;
Wu Fengguang16c40422010-08-11 14:17:39 -0700435}
Peter Zijlstra04fbfdc2007-10-16 23:25:50 -0700436
Randy Dunlap03ab4502010-08-14 13:05:17 -0700437/*
Wu Fengguang1babe182010-08-11 14:17:40 -0700438 * bdi_dirty_limit - @bdi's share of dirty throttling threshold
439 *
440 * Allocate high/low dirty limits to fast/slow devices, in order to prevent
441 * - starving fast devices
442 * - piling up dirty pages (that will take long time to sync) on slow devices
443 *
444 * The bdi's share of dirty limit will be adapting to its throughput and
445 * bounded by the bdi->min_ratio and/or bdi->max_ratio parameters, if set.
446 */
447unsigned long bdi_dirty_limit(struct backing_dev_info *bdi, unsigned long dirty)
Wu Fengguang16c40422010-08-11 14:17:39 -0700448{
449 u64 bdi_dirty;
450 long numerator, denominator;
Peter Zijlstra04fbfdc2007-10-16 23:25:50 -0700451
Wu Fengguang16c40422010-08-11 14:17:39 -0700452 /*
453 * Calculate this BDI's share of the dirty ratio.
454 */
455 bdi_writeout_fraction(bdi, &numerator, &denominator);
Peter Zijlstra04fbfdc2007-10-16 23:25:50 -0700456
Wu Fengguang16c40422010-08-11 14:17:39 -0700457 bdi_dirty = (dirty * (100 - bdi_min_ratio)) / 100;
458 bdi_dirty *= numerator;
459 do_div(bdi_dirty, denominator);
Peter Zijlstra04fbfdc2007-10-16 23:25:50 -0700460
Wu Fengguang16c40422010-08-11 14:17:39 -0700461 bdi_dirty += (dirty * bdi->min_ratio) / 100;
462 if (bdi_dirty > (dirty * bdi->max_ratio) / 100)
463 bdi_dirty = dirty * bdi->max_ratio / 100;
464
465 return bdi_dirty;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700466}
467
468/*
469 * balance_dirty_pages() must be called by processes which are generating dirty
470 * data. It looks at the number of dirty pages in the machine and will force
471 * the caller to perform writeback if the system is over `vm_dirty_ratio'.
Jens Axboe5b0830c2009-09-23 19:37:09 +0200472 * If we're over `background_thresh' then the writeback threads are woken to
473 * perform some writeout.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700474 */
Wu Fengguang3a2e9a52009-09-23 21:56:00 +0800475static void balance_dirty_pages(struct address_space *mapping,
476 unsigned long write_chunk)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700477{
Peter Zijlstra5fce25a2007-11-14 16:59:15 -0800478 long nr_reclaimable, bdi_nr_reclaimable;
479 long nr_writeback, bdi_nr_writeback;
David Rientjes364aeb22009-01-06 14:39:29 -0800480 unsigned long background_thresh;
481 unsigned long dirty_thresh;
482 unsigned long bdi_thresh;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700483 unsigned long pages_written = 0;
Jens Axboe87c6a9b2009-09-17 19:59:14 +0200484 unsigned long pause = 1;
Wu Fengguange50e3722010-08-11 14:17:37 -0700485 bool dirty_exceeded = false;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700486 struct backing_dev_info *bdi = mapping->backing_dev_info;
487
488 for (;;) {
489 struct writeback_control wbc = {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700490 .sync_mode = WB_SYNC_NONE,
491 .older_than_this = NULL,
492 .nr_to_write = write_chunk,
OGAWA Hirofumi111ebb62006-06-23 02:03:26 -0700493 .range_cyclic = 1,
Linus Torvalds1da177e2005-04-16 15:20:36 -0700494 };
495
Peter Zijlstra5fce25a2007-11-14 16:59:15 -0800496 nr_reclaimable = global_page_state(NR_FILE_DIRTY) +
497 global_page_state(NR_UNSTABLE_NFS);
498 nr_writeback = global_page_state(NR_WRITEBACK);
499
Wu Fengguang16c40422010-08-11 14:17:39 -0700500 global_dirty_limits(&background_thresh, &dirty_thresh);
501
502 /*
503 * Throttle it only when the background writeback cannot
504 * catch-up. This avoids (excessively) small writeouts
505 * when the bdi limits are ramping up.
506 */
Wu Fengguang4cbec4c2010-10-26 14:21:45 -0700507 if (nr_reclaimable + nr_writeback <=
Wu Fengguang16c40422010-08-11 14:17:39 -0700508 (background_thresh + dirty_thresh) / 2)
509 break;
510
511 bdi_thresh = bdi_dirty_limit(bdi, dirty_thresh);
512 bdi_thresh = task_dirty_limit(current, bdi_thresh);
513
Wu Fengguange50e3722010-08-11 14:17:37 -0700514 /*
515 * In order to avoid the stacked BDI deadlock we need
516 * to ensure we accurately count the 'dirty' pages when
517 * the threshold is low.
518 *
519 * Otherwise it would be possible to get thresh+n pages
520 * reported dirty, even though there are thresh-m pages
521 * actually dirty; with m+n sitting in the percpu
522 * deltas.
523 */
524 if (bdi_thresh < 2*bdi_stat_error(bdi)) {
525 bdi_nr_reclaimable = bdi_stat_sum(bdi, BDI_RECLAIMABLE);
526 bdi_nr_writeback = bdi_stat_sum(bdi, BDI_WRITEBACK);
527 } else {
528 bdi_nr_reclaimable = bdi_stat(bdi, BDI_RECLAIMABLE);
529 bdi_nr_writeback = bdi_stat(bdi, BDI_WRITEBACK);
530 }
Peter Zijlstra5fce25a2007-11-14 16:59:15 -0800531
Wu Fengguange50e3722010-08-11 14:17:37 -0700532 /*
533 * The bdi thresh is somehow "soft" limit derived from the
534 * global "hard" limit. The former helps to prevent heavy IO
535 * bdi or process from holding back light ones; The latter is
536 * the last resort safeguard.
537 */
538 dirty_exceeded =
Wu Fengguang4cbec4c2010-10-26 14:21:45 -0700539 (bdi_nr_reclaimable + bdi_nr_writeback > bdi_thresh)
540 || (nr_reclaimable + nr_writeback > dirty_thresh);
Wu Fengguange50e3722010-08-11 14:17:37 -0700541
542 if (!dirty_exceeded)
Peter Zijlstra04fbfdc2007-10-16 23:25:50 -0700543 break;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700544
Peter Zijlstra04fbfdc2007-10-16 23:25:50 -0700545 if (!bdi->dirty_exceeded)
546 bdi->dirty_exceeded = 1;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700547
548 /* Note: nr_reclaimable denotes nr_dirty + nr_unstable.
549 * Unstable writes are a feature of certain networked
550 * filesystems (i.e. NFS) in which data may have been
551 * written to the server's write cache, but has not yet
552 * been flushed to permanent storage.
Richard Kennedyd7831a02009-06-30 11:41:35 -0700553 * Only move pages to writeback if this bdi is over its
554 * threshold otherwise wait until the disk writes catch
555 * up.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700556 */
Dave Chinner028c2dd2010-07-07 13:24:07 +1000557 trace_wbc_balance_dirty_start(&wbc, bdi);
Richard Kennedyd7831a02009-06-30 11:41:35 -0700558 if (bdi_nr_reclaimable > bdi_thresh) {
Christoph Hellwig9c3a8ee2010-06-10 12:07:27 +0200559 writeback_inodes_wb(&bdi->wb, &wbc);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700560 pages_written += write_chunk - wbc.nr_to_write;
Dave Chinner028c2dd2010-07-07 13:24:07 +1000561 trace_wbc_balance_dirty_written(&wbc, bdi);
Wu Fengguange50e3722010-08-11 14:17:37 -0700562 if (pages_written >= write_chunk)
563 break; /* We've done our duty */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700564 }
Dave Chinner028c2dd2010-07-07 13:24:07 +1000565 trace_wbc_balance_dirty_wait(&wbc, bdi);
Wu Fengguangd153ba62010-12-21 17:24:21 -0800566 __set_current_state(TASK_UNINTERRUPTIBLE);
Wu Fengguangd25105e2009-10-09 12:40:42 +0200567 io_schedule_timeout(pause);
Jens Axboe87c6a9b2009-09-17 19:59:14 +0200568
569 /*
570 * Increase the delay for each loop, up to our previous
571 * default of taking a 100ms nap.
572 */
573 pause <<= 1;
574 if (pause > HZ / 10)
575 pause = HZ / 10;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700576 }
577
Wu Fengguange50e3722010-08-11 14:17:37 -0700578 if (!dirty_exceeded && bdi->dirty_exceeded)
Peter Zijlstra04fbfdc2007-10-16 23:25:50 -0700579 bdi->dirty_exceeded = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700580
581 if (writeback_in_progress(bdi))
Jens Axboe5b0830c2009-09-23 19:37:09 +0200582 return;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700583
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) ||
Wu Fengguange50e3722010-08-11 14:17:37 -0700593 (!laptop_mode && (nr_reclaimable > background_thresh)))
Christoph Hellwigc5444192010-06-08 18:15:15 +0200594 bdi_start_background_writeback(bdi);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700595}
596
Peter Zijlstraa200ee12007-10-08 18:54:37 +0200597void set_page_dirty_balance(struct page *page, int page_mkwrite)
Peter Zijlstraedc79b22006-09-25 23:30:58 -0700598{
Peter Zijlstraa200ee12007-10-08 18:54:37 +0200599 if (set_page_dirty(page) || page_mkwrite) {
Peter Zijlstraedc79b22006-09-25 23:30:58 -0700600 struct address_space *mapping = page_mapping(page);
601
602 if (mapping)
603 balance_dirty_pages_ratelimited(mapping);
604 }
605}
606
Tejun Heo245b2e72009-06-24 15:13:48 +0900607static DEFINE_PER_CPU(unsigned long, bdp_ratelimits) = 0;
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 unsigned long ratelimit;
627 unsigned long *p;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700628
629 ratelimit = ratelimit_pages;
Peter Zijlstra04fbfdc2007-10-16 23:25:50 -0700630 if (mapping->backing_dev_info->dirty_exceeded)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700631 ratelimit = 8;
632
633 /*
634 * Check the rate limiting. Also, we do not want to throttle real-time
635 * tasks in balance_dirty_pages(). Period.
636 */
Andrew Mortonfa5a7342006-03-24 03:18:10 -0800637 preempt_disable();
Tejun Heo245b2e72009-06-24 15:13:48 +0900638 p = &__get_cpu_var(bdp_ratelimits);
Andrew Mortonfa5a7342006-03-24 03:18:10 -0800639 *p += nr_pages_dirtied;
640 if (unlikely(*p >= ratelimit)) {
Wu Fengguang3a2e9a52009-09-23 21:56:00 +0800641 ratelimit = sync_writeback_pages(*p);
Andrew Mortonfa5a7342006-03-24 03:18:10 -0800642 *p = 0;
643 preempt_enable();
Wu Fengguang3a2e9a52009-09-23 21:56:00 +0800644 balance_dirty_pages(mapping, ratelimit);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700645 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 ( ; ; ) {
Wu Fengguang16c40422010-08-11 14:17:39 -0700657 global_dirty_limits(&background_thresh, &dirty_thresh);
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;
Jens Axboe8aa7e842009-07-09 14:52:32 +0200668 congestion_wait(BLK_RW_ASYNC, 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/*
Linus Torvalds1da177e2005-04-16 15:20:36 -0700681 * sysctl handler for /proc/sys/vm/dirty_writeback_centisecs
682 */
683int dirty_writeback_centisecs_handler(ctl_table *table, int write,
Alexey Dobriyan8d65af72009-09-23 15:57:19 -0700684 void __user *buffer, size_t *length, loff_t *ppos)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700685{
Alexey Dobriyan8d65af72009-09-23 15:57:19 -0700686 proc_dointvec(table, write, buffer, length, ppos);
Jens Axboe64231042010-05-21 20:00:35 +0200687 bdi_arm_supers_timer();
Linus Torvalds1da177e2005-04-16 15:20:36 -0700688 return 0;
689}
690
Jens Axboec2c49862010-05-20 09:18:47 +0200691#ifdef CONFIG_BLOCK
Matthew Garrett31373d02010-04-06 14:25:14 +0200692void laptop_mode_timer_fn(unsigned long data)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700693{
Matthew Garrett31373d02010-04-06 14:25:14 +0200694 struct request_queue *q = (struct request_queue *)data;
695 int nr_pages = global_page_state(NR_FILE_DIRTY) +
696 global_page_state(NR_UNSTABLE_NFS);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700697
Matthew Garrett31373d02010-04-06 14:25:14 +0200698 /*
699 * We want to write everything out, not just down to the dirty
700 * threshold
701 */
Matthew Garrett31373d02010-04-06 14:25:14 +0200702 if (bdi_has_dirty_io(&q->backing_dev_info))
Christoph Hellwigc5444192010-06-08 18:15:15 +0200703 bdi_start_writeback(&q->backing_dev_info, nr_pages);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700704}
705
706/*
707 * We've spun up the disk and we're in laptop mode: schedule writeback
708 * of all dirty data a few seconds from now. If the flush is already scheduled
709 * then push it back - the user is still using the disk.
710 */
Matthew Garrett31373d02010-04-06 14:25:14 +0200711void laptop_io_completion(struct backing_dev_info *info)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700712{
Matthew Garrett31373d02010-04-06 14:25:14 +0200713 mod_timer(&info->laptop_mode_wb_timer, jiffies + laptop_mode);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700714}
715
716/*
717 * We're in laptop mode and we've just synced. The sync's writes will have
718 * caused another writeback to be scheduled by laptop_io_completion.
719 * Nothing needs to be written back anymore, so we unschedule the writeback.
720 */
721void laptop_sync_completion(void)
722{
Matthew Garrett31373d02010-04-06 14:25:14 +0200723 struct backing_dev_info *bdi;
724
725 rcu_read_lock();
726
727 list_for_each_entry_rcu(bdi, &bdi_list, bdi_list)
728 del_timer(&bdi->laptop_mode_wb_timer);
729
730 rcu_read_unlock();
Linus Torvalds1da177e2005-04-16 15:20:36 -0700731}
Jens Axboec2c49862010-05-20 09:18:47 +0200732#endif
Linus Torvalds1da177e2005-04-16 15:20:36 -0700733
734/*
735 * If ratelimit_pages is too high then we can get into dirty-data overload
736 * if a large number of processes all perform writes at the same time.
737 * If it is too low then SMP machines will call the (expensive)
738 * get_writeback_state too often.
739 *
740 * Here we set ratelimit_pages to a level which ensures that when all CPUs are
741 * dirtying in parallel, we cannot go more than 3% (1/32) over the dirty memory
742 * thresholds before writeback cuts in.
743 *
744 * But the limit should not be set too high. Because it also controls the
745 * amount of memory which the balance_dirty_pages() caller has to write back.
746 * If this is too large then the caller will block on the IO queue all the
747 * time. So limit it to four megabytes - the balance_dirty_pages() caller
748 * will write six megabyte chunks, max.
749 */
750
Chandra Seetharaman2d1d43f2006-09-29 02:01:25 -0700751void writeback_set_ratelimit(void)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700752{
Chandra Seetharaman40c99aa2006-09-29 02:01:24 -0700753 ratelimit_pages = vm_total_pages / (num_online_cpus() * 32);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700754 if (ratelimit_pages < 16)
755 ratelimit_pages = 16;
756 if (ratelimit_pages * PAGE_CACHE_SIZE > 4096 * 1024)
757 ratelimit_pages = (4096 * 1024) / PAGE_CACHE_SIZE;
758}
759
Chandra Seetharaman26c21432006-06-27 02:54:10 -0700760static int __cpuinit
Linus Torvalds1da177e2005-04-16 15:20:36 -0700761ratelimit_handler(struct notifier_block *self, unsigned long u, void *v)
762{
Chandra Seetharaman2d1d43f2006-09-29 02:01:25 -0700763 writeback_set_ratelimit();
Paul E. McKenneyaa0f0302007-02-10 01:46:37 -0800764 return NOTIFY_DONE;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700765}
766
Chandra Seetharaman74b85f32006-06-27 02:54:09 -0700767static struct notifier_block __cpuinitdata ratelimit_nb = {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700768 .notifier_call = ratelimit_handler,
769 .next = NULL,
770};
771
772/*
Linus Torvaldsdc6e29d2007-01-29 16:37:38 -0800773 * Called early on to tune the page writeback dirty limits.
774 *
775 * We used to scale dirty pages according to how total memory
776 * related to pages that could be allocated for buffers (by
777 * comparing nr_free_buffer_pages() to vm_total_pages.
778 *
779 * However, that was when we used "dirty_ratio" to scale with
780 * all memory, and we don't do that any more. "dirty_ratio"
781 * is now applied to total non-HIGHPAGE memory (by subtracting
782 * totalhigh_pages from vm_total_pages), and as such we can't
783 * get into the old insane situation any more where we had
784 * large amounts of dirty pages compared to a small amount of
785 * non-HIGHMEM memory.
786 *
787 * But we might still want to scale the dirty_ratio by how
788 * much memory the box has..
Linus Torvalds1da177e2005-04-16 15:20:36 -0700789 */
790void __init page_writeback_init(void)
791{
Peter Zijlstra04fbfdc2007-10-16 23:25:50 -0700792 int shift;
793
Chandra Seetharaman2d1d43f2006-09-29 02:01:25 -0700794 writeback_set_ratelimit();
Linus Torvalds1da177e2005-04-16 15:20:36 -0700795 register_cpu_notifier(&ratelimit_nb);
Peter Zijlstra04fbfdc2007-10-16 23:25:50 -0700796
797 shift = calc_period_shift();
798 prop_descriptor_init(&vm_completions, shift);
Peter Zijlstra3e26c142007-10-16 23:25:50 -0700799 prop_descriptor_init(&vm_dirties, shift);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700800}
801
David Howells811d7362006-08-29 19:06:09 +0100802/**
Jan Karaf446daae2010-08-09 17:19:12 -0700803 * tag_pages_for_writeback - tag pages to be written by write_cache_pages
804 * @mapping: address space structure to write
805 * @start: starting page index
806 * @end: ending page index (inclusive)
807 *
808 * This function scans the page range from @start to @end (inclusive) and tags
809 * all pages that have DIRTY tag set with a special TOWRITE tag. The idea is
810 * that write_cache_pages (or whoever calls this function) will then use
811 * TOWRITE tag to identify pages eligible for writeback. This mechanism is
812 * used to avoid livelocking of writeback by a process steadily creating new
813 * dirty pages in the file (thus it is important for this function to be quick
814 * so that it can tag pages faster than a dirtying process can create them).
815 */
816/*
817 * We tag pages in batches of WRITEBACK_TAG_BATCH to reduce tree_lock latency.
818 */
Jan Karaf446daae2010-08-09 17:19:12 -0700819void tag_pages_for_writeback(struct address_space *mapping,
820 pgoff_t start, pgoff_t end)
821{
Randy Dunlap3c111a02010-08-11 14:17:30 -0700822#define WRITEBACK_TAG_BATCH 4096
Jan Karaf446daae2010-08-09 17:19:12 -0700823 unsigned long tagged;
824
825 do {
826 spin_lock_irq(&mapping->tree_lock);
827 tagged = radix_tree_range_tag_if_tagged(&mapping->page_tree,
828 &start, end, WRITEBACK_TAG_BATCH,
829 PAGECACHE_TAG_DIRTY, PAGECACHE_TAG_TOWRITE);
830 spin_unlock_irq(&mapping->tree_lock);
831 WARN_ON_ONCE(tagged > WRITEBACK_TAG_BATCH);
832 cond_resched();
Jan Karad5ed3a42010-08-19 14:13:33 -0700833 /* We check 'start' to handle wrapping when end == ~0UL */
834 } while (tagged >= WRITEBACK_TAG_BATCH && start);
Jan Karaf446daae2010-08-09 17:19:12 -0700835}
836EXPORT_SYMBOL(tag_pages_for_writeback);
837
838/**
Miklos Szeredi0ea97182007-05-10 22:22:51 -0700839 * 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 +0100840 * @mapping: address space structure to write
841 * @wbc: subtract the number of written pages from *@wbc->nr_to_write
Miklos Szeredi0ea97182007-05-10 22:22:51 -0700842 * @writepage: function called for each page
843 * @data: data passed to writepage function
David Howells811d7362006-08-29 19:06:09 +0100844 *
Miklos Szeredi0ea97182007-05-10 22:22:51 -0700845 * If a page is already under I/O, write_cache_pages() skips it, even
David Howells811d7362006-08-29 19:06:09 +0100846 * if it's dirty. This is desirable behaviour for memory-cleaning writeback,
847 * but it is INCORRECT for data-integrity system calls such as fsync(). fsync()
848 * and msync() need to guarantee that all the data which was dirty at the time
849 * the call was made get new I/O started against them. If wbc->sync_mode is
850 * WB_SYNC_ALL then we were called for data integrity and we must wait for
851 * existing IO to complete.
Jan Karaf446daae2010-08-09 17:19:12 -0700852 *
853 * To avoid livelocks (when other process dirties new pages), we first tag
854 * pages which should be written back with TOWRITE tag and only then start
855 * writing them. For data-integrity sync we have to be careful so that we do
856 * not miss some pages (e.g., because some other process has cleared TOWRITE
857 * tag we set). The rule we follow is that TOWRITE tag can be cleared only
858 * by the process clearing the DIRTY tag (and submitting the page for IO).
David Howells811d7362006-08-29 19:06:09 +0100859 */
Miklos Szeredi0ea97182007-05-10 22:22:51 -0700860int write_cache_pages(struct address_space *mapping,
861 struct writeback_control *wbc, writepage_t writepage,
862 void *data)
David Howells811d7362006-08-29 19:06:09 +0100863{
David Howells811d7362006-08-29 19:06:09 +0100864 int ret = 0;
865 int done = 0;
David Howells811d7362006-08-29 19:06:09 +0100866 struct pagevec pvec;
867 int nr_pages;
Nick Piggin31a12662009-01-06 14:39:04 -0800868 pgoff_t uninitialized_var(writeback_index);
David Howells811d7362006-08-29 19:06:09 +0100869 pgoff_t index;
870 pgoff_t end; /* Inclusive */
Nick Pigginbd19e012009-01-06 14:39:06 -0800871 pgoff_t done_index;
Nick Piggin31a12662009-01-06 14:39:04 -0800872 int cycled;
David Howells811d7362006-08-29 19:06:09 +0100873 int range_whole = 0;
Jan Karaf446daae2010-08-09 17:19:12 -0700874 int tag;
David Howells811d7362006-08-29 19:06:09 +0100875
David Howells811d7362006-08-29 19:06:09 +0100876 pagevec_init(&pvec, 0);
877 if (wbc->range_cyclic) {
Nick Piggin31a12662009-01-06 14:39:04 -0800878 writeback_index = mapping->writeback_index; /* prev offset */
879 index = writeback_index;
880 if (index == 0)
881 cycled = 1;
882 else
883 cycled = 0;
David Howells811d7362006-08-29 19:06:09 +0100884 end = -1;
885 } else {
886 index = wbc->range_start >> PAGE_CACHE_SHIFT;
887 end = wbc->range_end >> PAGE_CACHE_SHIFT;
888 if (wbc->range_start == 0 && wbc->range_end == LLONG_MAX)
889 range_whole = 1;
Nick Piggin31a12662009-01-06 14:39:04 -0800890 cycled = 1; /* ignore range_cyclic tests */
David Howells811d7362006-08-29 19:06:09 +0100891 }
Jan Karaf446daae2010-08-09 17:19:12 -0700892 if (wbc->sync_mode == WB_SYNC_ALL)
893 tag = PAGECACHE_TAG_TOWRITE;
894 else
895 tag = PAGECACHE_TAG_DIRTY;
David Howells811d7362006-08-29 19:06:09 +0100896retry:
Jan Karaf446daae2010-08-09 17:19:12 -0700897 if (wbc->sync_mode == WB_SYNC_ALL)
898 tag_pages_for_writeback(mapping, index, end);
Nick Pigginbd19e012009-01-06 14:39:06 -0800899 done_index = index;
Nick Piggin5a3d5c92009-01-06 14:39:09 -0800900 while (!done && (index <= end)) {
901 int i;
902
Jan Karaf446daae2010-08-09 17:19:12 -0700903 nr_pages = pagevec_lookup_tag(&pvec, mapping, &index, tag,
Nick Piggin5a3d5c92009-01-06 14:39:09 -0800904 min(end - index, (pgoff_t)PAGEVEC_SIZE-1) + 1);
905 if (nr_pages == 0)
906 break;
David Howells811d7362006-08-29 19:06:09 +0100907
David Howells811d7362006-08-29 19:06:09 +0100908 for (i = 0; i < nr_pages; i++) {
909 struct page *page = pvec.pages[i];
910
Nick Piggind5482cd2009-01-06 14:39:11 -0800911 /*
912 * At this point, the page may be truncated or
913 * invalidated (changing page->mapping to NULL), or
914 * even swizzled back from swapper_space to tmpfs file
915 * mapping. However, page->index will not change
916 * because we have a reference on the page.
917 */
918 if (page->index > end) {
919 /*
920 * can't be range_cyclic (1st pass) because
921 * end == -1 in that case.
922 */
923 done = 1;
924 break;
925 }
926
Nick Pigginbd19e012009-01-06 14:39:06 -0800927 done_index = page->index + 1;
928
David Howells811d7362006-08-29 19:06:09 +0100929 lock_page(page);
930
Nick Piggin5a3d5c92009-01-06 14:39:09 -0800931 /*
932 * Page truncated or invalidated. We can freely skip it
933 * then, even for data integrity operations: the page
934 * has disappeared concurrently, so there could be no
935 * real expectation of this data interity operation
936 * even if there is now a new, dirty page at the same
937 * pagecache address.
938 */
David Howells811d7362006-08-29 19:06:09 +0100939 if (unlikely(page->mapping != mapping)) {
Nick Piggin5a3d5c92009-01-06 14:39:09 -0800940continue_unlock:
David Howells811d7362006-08-29 19:06:09 +0100941 unlock_page(page);
942 continue;
943 }
944
Nick Piggin515f4a02009-01-06 14:39:10 -0800945 if (!PageDirty(page)) {
946 /* someone wrote it for us */
947 goto continue_unlock;
948 }
David Howells811d7362006-08-29 19:06:09 +0100949
Nick Piggin515f4a02009-01-06 14:39:10 -0800950 if (PageWriteback(page)) {
951 if (wbc->sync_mode != WB_SYNC_NONE)
952 wait_on_page_writeback(page);
953 else
954 goto continue_unlock;
955 }
956
957 BUG_ON(PageWriteback(page));
958 if (!clear_page_dirty_for_io(page))
Nick Piggin5a3d5c92009-01-06 14:39:09 -0800959 goto continue_unlock;
David Howells811d7362006-08-29 19:06:09 +0100960
Dave Chinner9e094382010-07-07 13:24:08 +1000961 trace_wbc_writepage(wbc, mapping->backing_dev_info);
Miklos Szeredi0ea97182007-05-10 22:22:51 -0700962 ret = (*writepage)(page, wbc, data);
Nick Piggin00266772009-01-06 14:39:06 -0800963 if (unlikely(ret)) {
964 if (ret == AOP_WRITEPAGE_ACTIVATE) {
965 unlock_page(page);
966 ret = 0;
967 } else {
968 /*
969 * done_index is set past this page,
970 * so media errors will not choke
971 * background writeout for the entire
972 * file. This has consequences for
973 * range_cyclic semantics (ie. it may
974 * not be suitable for data integrity
975 * writeout).
976 */
977 done = 1;
978 break;
979 }
Dave Chinner0b564922010-06-09 10:37:18 +1000980 }
David Howells811d7362006-08-29 19:06:09 +0100981
Dave Chinner546a1922010-08-24 11:44:34 +1000982 /*
983 * We stop writing back only if we are not doing
984 * integrity sync. In case of integrity sync we have to
985 * keep going until we have written all the pages
986 * we tagged for writeback prior to entering this loop.
987 */
988 if (--wbc->nr_to_write <= 0 &&
989 wbc->sync_mode == WB_SYNC_NONE) {
990 done = 1;
991 break;
Nick Piggin05fe4782009-01-06 14:39:08 -0800992 }
David Howells811d7362006-08-29 19:06:09 +0100993 }
994 pagevec_release(&pvec);
995 cond_resched();
996 }
Nick Piggin3a4c6802009-02-12 04:34:23 +0100997 if (!cycled && !done) {
David Howells811d7362006-08-29 19:06:09 +0100998 /*
Nick Piggin31a12662009-01-06 14:39:04 -0800999 * range_cyclic:
David Howells811d7362006-08-29 19:06:09 +01001000 * We hit the last page and there is more work to be done: wrap
1001 * back to the start of the file
1002 */
Nick Piggin31a12662009-01-06 14:39:04 -08001003 cycled = 1;
David Howells811d7362006-08-29 19:06:09 +01001004 index = 0;
Nick Piggin31a12662009-01-06 14:39:04 -08001005 end = writeback_index - 1;
David Howells811d7362006-08-29 19:06:09 +01001006 goto retry;
1007 }
Dave Chinner0b564922010-06-09 10:37:18 +10001008 if (wbc->range_cyclic || (range_whole && wbc->nr_to_write > 0))
1009 mapping->writeback_index = done_index;
Aneesh Kumar K.V06d6cf62008-07-11 19:27:31 -04001010
David Howells811d7362006-08-29 19:06:09 +01001011 return ret;
1012}
Miklos Szeredi0ea97182007-05-10 22:22:51 -07001013EXPORT_SYMBOL(write_cache_pages);
1014
1015/*
1016 * Function used by generic_writepages to call the real writepage
1017 * function and set the mapping flags on error
1018 */
1019static int __writepage(struct page *page, struct writeback_control *wbc,
1020 void *data)
1021{
1022 struct address_space *mapping = data;
1023 int ret = mapping->a_ops->writepage(page, wbc);
1024 mapping_set_error(mapping, ret);
1025 return ret;
1026}
1027
1028/**
1029 * generic_writepages - walk the list of dirty pages of the given address space and writepage() all of them.
1030 * @mapping: address space structure to write
1031 * @wbc: subtract the number of written pages from *@wbc->nr_to_write
1032 *
1033 * This is a library function, which implements the writepages()
1034 * address_space_operation.
1035 */
1036int generic_writepages(struct address_space *mapping,
1037 struct writeback_control *wbc)
1038{
1039 /* deal with chardevs and other special file */
1040 if (!mapping->a_ops->writepage)
1041 return 0;
1042
1043 return write_cache_pages(mapping, wbc, __writepage, mapping);
1044}
David Howells811d7362006-08-29 19:06:09 +01001045
1046EXPORT_SYMBOL(generic_writepages);
1047
Linus Torvalds1da177e2005-04-16 15:20:36 -07001048int do_writepages(struct address_space *mapping, struct writeback_control *wbc)
1049{
Andrew Morton22905f72005-11-16 15:07:01 -08001050 int ret;
1051
Linus Torvalds1da177e2005-04-16 15:20:36 -07001052 if (wbc->nr_to_write <= 0)
1053 return 0;
1054 if (mapping->a_ops->writepages)
Peter Zijlstrad08b3852006-09-25 23:30:57 -07001055 ret = mapping->a_ops->writepages(mapping, wbc);
Andrew Morton22905f72005-11-16 15:07:01 -08001056 else
1057 ret = generic_writepages(mapping, wbc);
Andrew Morton22905f72005-11-16 15:07:01 -08001058 return ret;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001059}
1060
1061/**
1062 * write_one_page - write out a single page and optionally wait on I/O
Martin Waitz67be2dd2005-05-01 08:59:26 -07001063 * @page: the page to write
1064 * @wait: if true, wait on writeout
Linus Torvalds1da177e2005-04-16 15:20:36 -07001065 *
1066 * The page must be locked by the caller and will be unlocked upon return.
1067 *
1068 * write_one_page() returns a negative error code if I/O failed.
1069 */
1070int write_one_page(struct page *page, int wait)
1071{
1072 struct address_space *mapping = page->mapping;
1073 int ret = 0;
1074 struct writeback_control wbc = {
1075 .sync_mode = WB_SYNC_ALL,
1076 .nr_to_write = 1,
1077 };
1078
1079 BUG_ON(!PageLocked(page));
1080
1081 if (wait)
1082 wait_on_page_writeback(page);
1083
1084 if (clear_page_dirty_for_io(page)) {
1085 page_cache_get(page);
1086 ret = mapping->a_ops->writepage(page, &wbc);
1087 if (ret == 0 && wait) {
1088 wait_on_page_writeback(page);
1089 if (PageError(page))
1090 ret = -EIO;
1091 }
1092 page_cache_release(page);
1093 } else {
1094 unlock_page(page);
1095 }
1096 return ret;
1097}
1098EXPORT_SYMBOL(write_one_page);
1099
1100/*
Ken Chen76719322007-02-10 01:43:15 -08001101 * For address_spaces which do not use buffers nor write back.
1102 */
1103int __set_page_dirty_no_writeback(struct page *page)
1104{
1105 if (!PageDirty(page))
1106 SetPageDirty(page);
1107 return 0;
1108}
1109
1110/*
Edward Shishkine3a7cca2009-03-31 15:19:39 -07001111 * Helper function for set_page_dirty family.
1112 * NOTE: This relies on being atomic wrt interrupts.
1113 */
1114void account_page_dirtied(struct page *page, struct address_space *mapping)
1115{
1116 if (mapping_cap_account_dirty(mapping)) {
1117 __inc_zone_page_state(page, NR_FILE_DIRTY);
Michael Rubinea941f02010-10-26 14:21:35 -07001118 __inc_zone_page_state(page, NR_DIRTIED);
Edward Shishkine3a7cca2009-03-31 15:19:39 -07001119 __inc_bdi_stat(mapping->backing_dev_info, BDI_RECLAIMABLE);
1120 task_dirty_inc(current);
1121 task_io_account_write(PAGE_CACHE_SIZE);
1122 }
1123}
Michael Rubin679ceac2010-08-20 02:31:26 -07001124EXPORT_SYMBOL(account_page_dirtied);
Edward Shishkine3a7cca2009-03-31 15:19:39 -07001125
1126/*
Michael Rubinf629d1c2010-10-26 14:21:33 -07001127 * Helper function for set_page_writeback family.
1128 * NOTE: Unlike account_page_dirtied this does not rely on being atomic
1129 * wrt interrupts.
1130 */
1131void account_page_writeback(struct page *page)
1132{
1133 inc_zone_page_state(page, NR_WRITEBACK);
Michael Rubinea941f02010-10-26 14:21:35 -07001134 inc_zone_page_state(page, NR_WRITTEN);
Michael Rubinf629d1c2010-10-26 14:21:33 -07001135}
1136EXPORT_SYMBOL(account_page_writeback);
1137
1138/*
Linus Torvalds1da177e2005-04-16 15:20:36 -07001139 * For address_spaces which do not use buffers. Just tag the page as dirty in
1140 * its radix tree.
1141 *
1142 * This is also used when a single buffer is being dirtied: we want to set the
1143 * page dirty in that case, but not all the buffers. This is a "bottom-up"
1144 * dirtying, whereas __set_page_dirty_buffers() is a "top-down" dirtying.
1145 *
1146 * Most callers have locked the page, which pins the address_space in memory.
1147 * But zap_pte_range() does not lock the page, however in that case the
1148 * mapping is pinned by the vma's ->vm_file reference.
1149 *
1150 * We take care to handle the case where the page was truncated from the
Simon Arlott183ff222007-10-20 01:27:18 +02001151 * mapping by re-checking page_mapping() inside tree_lock.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001152 */
1153int __set_page_dirty_nobuffers(struct page *page)
1154{
Linus Torvalds1da177e2005-04-16 15:20:36 -07001155 if (!TestSetPageDirty(page)) {
1156 struct address_space *mapping = page_mapping(page);
1157 struct address_space *mapping2;
1158
Andrew Morton8c085402006-12-10 02:19:24 -08001159 if (!mapping)
1160 return 1;
1161
Nick Piggin19fd6232008-07-25 19:45:32 -07001162 spin_lock_irq(&mapping->tree_lock);
Andrew Morton8c085402006-12-10 02:19:24 -08001163 mapping2 = page_mapping(page);
1164 if (mapping2) { /* Race with truncate? */
1165 BUG_ON(mapping2 != mapping);
Nick Piggin787d2212007-07-17 04:03:34 -07001166 WARN_ON_ONCE(!PagePrivate(page) && !PageUptodate(page));
Edward Shishkine3a7cca2009-03-31 15:19:39 -07001167 account_page_dirtied(page, mapping);
Andrew Morton8c085402006-12-10 02:19:24 -08001168 radix_tree_tag_set(&mapping->page_tree,
1169 page_index(page), PAGECACHE_TAG_DIRTY);
1170 }
Nick Piggin19fd6232008-07-25 19:45:32 -07001171 spin_unlock_irq(&mapping->tree_lock);
Andrew Morton8c085402006-12-10 02:19:24 -08001172 if (mapping->host) {
1173 /* !PageAnon && !swapper_space */
1174 __mark_inode_dirty(mapping->host, I_DIRTY_PAGES);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001175 }
Andrew Morton4741c9f2006-03-24 03:18:11 -08001176 return 1;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001177 }
Andrew Morton4741c9f2006-03-24 03:18:11 -08001178 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001179}
1180EXPORT_SYMBOL(__set_page_dirty_nobuffers);
1181
1182/*
1183 * When a writepage implementation decides that it doesn't want to write this
1184 * page for some reason, it should redirty the locked page via
1185 * redirty_page_for_writepage() and it should then unlock the page and return 0
1186 */
1187int redirty_page_for_writepage(struct writeback_control *wbc, struct page *page)
1188{
1189 wbc->pages_skipped++;
1190 return __set_page_dirty_nobuffers(page);
1191}
1192EXPORT_SYMBOL(redirty_page_for_writepage);
1193
1194/*
Wu Fengguang6746aff2009-09-16 11:50:14 +02001195 * Dirty a page.
1196 *
1197 * For pages with a mapping this should be done under the page lock
1198 * for the benefit of asynchronous memory errors who prefer a consistent
1199 * dirty state. This rule can be broken in some special cases,
1200 * but should be better not to.
1201 *
Linus Torvalds1da177e2005-04-16 15:20:36 -07001202 * If the mapping doesn't provide a set_page_dirty a_op, then
1203 * just fall through and assume that it wants buffer_heads.
1204 */
Nick Piggin1cf6e7d2009-02-18 14:48:18 -08001205int set_page_dirty(struct page *page)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001206{
1207 struct address_space *mapping = page_mapping(page);
1208
1209 if (likely(mapping)) {
1210 int (*spd)(struct page *) = mapping->a_ops->set_page_dirty;
David Howells93614012006-09-30 20:45:40 +02001211#ifdef CONFIG_BLOCK
1212 if (!spd)
1213 spd = __set_page_dirty_buffers;
1214#endif
1215 return (*spd)(page);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001216 }
Andrew Morton4741c9f2006-03-24 03:18:11 -08001217 if (!PageDirty(page)) {
1218 if (!TestSetPageDirty(page))
1219 return 1;
1220 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001221 return 0;
1222}
1223EXPORT_SYMBOL(set_page_dirty);
1224
1225/*
1226 * set_page_dirty() is racy if the caller has no reference against
1227 * page->mapping->host, and if the page is unlocked. This is because another
1228 * CPU could truncate the page off the mapping and then free the mapping.
1229 *
1230 * Usually, the page _is_ locked, or the caller is a user-space process which
1231 * holds a reference on the inode by having an open file.
1232 *
1233 * In other cases, the page should be locked before running set_page_dirty().
1234 */
1235int set_page_dirty_lock(struct page *page)
1236{
1237 int ret;
1238
Nick Piggindb376482006-09-25 23:31:24 -07001239 lock_page_nosync(page);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001240 ret = set_page_dirty(page);
1241 unlock_page(page);
1242 return ret;
1243}
1244EXPORT_SYMBOL(set_page_dirty_lock);
1245
1246/*
Linus Torvalds1da177e2005-04-16 15:20:36 -07001247 * Clear a page's dirty flag, while caring for dirty memory accounting.
1248 * Returns true if the page was previously dirty.
1249 *
1250 * This is for preparing to put the page under writeout. We leave the page
1251 * tagged as dirty in the radix tree so that a concurrent write-for-sync
1252 * can discover it via a PAGECACHE_TAG_DIRTY walk. The ->writepage
1253 * implementation will run either set_page_writeback() or set_page_dirty(),
1254 * at which stage we bring the page's dirty flag and radix-tree dirty tag
1255 * back into sync.
1256 *
1257 * This incoherency between the page's dirty flag and radix-tree tag is
1258 * unfortunate, but it only exists while the page is locked.
1259 */
1260int clear_page_dirty_for_io(struct page *page)
1261{
1262 struct address_space *mapping = page_mapping(page);
1263
Nick Piggin79352892007-07-19 01:47:22 -07001264 BUG_ON(!PageLocked(page));
1265
Fengguang Wufe3cba12007-07-19 01:48:07 -07001266 ClearPageReclaim(page);
Linus Torvalds7658cc22006-12-29 10:00:58 -08001267 if (mapping && mapping_cap_account_dirty(mapping)) {
1268 /*
1269 * Yes, Virginia, this is indeed insane.
1270 *
1271 * We use this sequence to make sure that
1272 * (a) we account for dirty stats properly
1273 * (b) we tell the low-level filesystem to
1274 * mark the whole page dirty if it was
1275 * dirty in a pagetable. Only to then
1276 * (c) clean the page again and return 1 to
1277 * cause the writeback.
1278 *
1279 * This way we avoid all nasty races with the
1280 * dirty bit in multiple places and clearing
1281 * them concurrently from different threads.
1282 *
1283 * Note! Normally the "set_page_dirty(page)"
1284 * has no effect on the actual dirty bit - since
1285 * that will already usually be set. But we
1286 * need the side effects, and it can help us
1287 * avoid races.
1288 *
1289 * We basically use the page "master dirty bit"
1290 * as a serialization point for all the different
1291 * threads doing their things.
Linus Torvalds7658cc22006-12-29 10:00:58 -08001292 */
1293 if (page_mkclean(page))
1294 set_page_dirty(page);
Nick Piggin79352892007-07-19 01:47:22 -07001295 /*
1296 * We carefully synchronise fault handlers against
1297 * installing a dirty pte and marking the page dirty
1298 * at this point. We do this by having them hold the
1299 * page lock at some point after installing their
1300 * pte, but before marking the page dirty.
1301 * Pages are always locked coming in here, so we get
1302 * the desired exclusion. See mm/memory.c:do_wp_page()
1303 * for more comments.
1304 */
Linus Torvalds7658cc22006-12-29 10:00:58 -08001305 if (TestClearPageDirty(page)) {
Andrew Morton8c085402006-12-10 02:19:24 -08001306 dec_zone_page_state(page, NR_FILE_DIRTY);
Peter Zijlstrac9e51e42007-10-16 23:25:47 -07001307 dec_bdi_stat(mapping->backing_dev_info,
1308 BDI_RECLAIMABLE);
Linus Torvalds7658cc22006-12-29 10:00:58 -08001309 return 1;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001310 }
Linus Torvalds7658cc22006-12-29 10:00:58 -08001311 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001312 }
Linus Torvalds7658cc22006-12-29 10:00:58 -08001313 return TestClearPageDirty(page);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001314}
Hans Reiser58bb01a2005-11-18 01:10:53 -08001315EXPORT_SYMBOL(clear_page_dirty_for_io);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001316
1317int test_clear_page_writeback(struct page *page)
1318{
1319 struct address_space *mapping = page_mapping(page);
1320 int ret;
1321
1322 if (mapping) {
Peter Zijlstra69cb51d2007-10-16 23:25:48 -07001323 struct backing_dev_info *bdi = mapping->backing_dev_info;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001324 unsigned long flags;
1325
Nick Piggin19fd6232008-07-25 19:45:32 -07001326 spin_lock_irqsave(&mapping->tree_lock, flags);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001327 ret = TestClearPageWriteback(page);
Peter Zijlstra69cb51d2007-10-16 23:25:48 -07001328 if (ret) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001329 radix_tree_tag_clear(&mapping->page_tree,
1330 page_index(page),
1331 PAGECACHE_TAG_WRITEBACK);
Miklos Szeredie4ad08f2008-04-30 00:54:37 -07001332 if (bdi_cap_account_writeback(bdi)) {
Peter Zijlstra69cb51d2007-10-16 23:25:48 -07001333 __dec_bdi_stat(bdi, BDI_WRITEBACK);
Peter Zijlstra04fbfdc2007-10-16 23:25:50 -07001334 __bdi_writeout_inc(bdi);
1335 }
Peter Zijlstra69cb51d2007-10-16 23:25:48 -07001336 }
Nick Piggin19fd6232008-07-25 19:45:32 -07001337 spin_unlock_irqrestore(&mapping->tree_lock, flags);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001338 } else {
1339 ret = TestClearPageWriteback(page);
1340 }
Andrew Mortond688abf2007-07-19 01:49:17 -07001341 if (ret)
1342 dec_zone_page_state(page, NR_WRITEBACK);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001343 return ret;
1344}
1345
1346int test_set_page_writeback(struct page *page)
1347{
1348 struct address_space *mapping = page_mapping(page);
1349 int ret;
1350
1351 if (mapping) {
Peter Zijlstra69cb51d2007-10-16 23:25:48 -07001352 struct backing_dev_info *bdi = mapping->backing_dev_info;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001353 unsigned long flags;
1354
Nick Piggin19fd6232008-07-25 19:45:32 -07001355 spin_lock_irqsave(&mapping->tree_lock, flags);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001356 ret = TestSetPageWriteback(page);
Peter Zijlstra69cb51d2007-10-16 23:25:48 -07001357 if (!ret) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001358 radix_tree_tag_set(&mapping->page_tree,
1359 page_index(page),
1360 PAGECACHE_TAG_WRITEBACK);
Miklos Szeredie4ad08f2008-04-30 00:54:37 -07001361 if (bdi_cap_account_writeback(bdi))
Peter Zijlstra69cb51d2007-10-16 23:25:48 -07001362 __inc_bdi_stat(bdi, BDI_WRITEBACK);
1363 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001364 if (!PageDirty(page))
1365 radix_tree_tag_clear(&mapping->page_tree,
1366 page_index(page),
1367 PAGECACHE_TAG_DIRTY);
Jan Karaf446daae2010-08-09 17:19:12 -07001368 radix_tree_tag_clear(&mapping->page_tree,
1369 page_index(page),
1370 PAGECACHE_TAG_TOWRITE);
Nick Piggin19fd6232008-07-25 19:45:32 -07001371 spin_unlock_irqrestore(&mapping->tree_lock, flags);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001372 } else {
1373 ret = TestSetPageWriteback(page);
1374 }
Andrew Mortond688abf2007-07-19 01:49:17 -07001375 if (!ret)
Michael Rubinf629d1c2010-10-26 14:21:33 -07001376 account_page_writeback(page);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001377 return ret;
1378
1379}
1380EXPORT_SYMBOL(test_set_page_writeback);
1381
1382/*
Nick Piggin00128182007-10-16 01:24:40 -07001383 * Return true if any of the pages in the mapping are marked with the
Linus Torvalds1da177e2005-04-16 15:20:36 -07001384 * passed tag.
1385 */
1386int mapping_tagged(struct address_space *mapping, int tag)
1387{
Linus Torvalds1da177e2005-04-16 15:20:36 -07001388 int ret;
Nick Piggin00128182007-10-16 01:24:40 -07001389 rcu_read_lock();
Linus Torvalds1da177e2005-04-16 15:20:36 -07001390 ret = radix_tree_tagged(&mapping->page_tree, tag);
Nick Piggin00128182007-10-16 01:24:40 -07001391 rcu_read_unlock();
Linus Torvalds1da177e2005-04-16 15:20:36 -07001392 return ret;
1393}
1394EXPORT_SYMBOL(mapping_tagged);