Commit | Line | Data |
---|---|---|
1da177e4 | 1 | /* |
f30c2269 | 2 | * mm/page-writeback.c |
1da177e4 LT |
3 | * |
4 | * Copyright (C) 2002, Linus Torvalds. | |
04fbfdc1 | 5 | * Copyright (C) 2007 Red Hat, Inc., Peter Zijlstra <pzijlstr@redhat.com> |
1da177e4 LT |
6 | * |
7 | * Contains functions related to writing back dirty pages at the | |
8 | * address_space level. | |
9 | * | |
e1f8e874 | 10 | * 10Apr2002 Andrew Morton |
1da177e4 LT |
11 | * Initial version |
12 | */ | |
13 | ||
14 | #include <linux/kernel.h> | |
15 | #include <linux/module.h> | |
16 | #include <linux/spinlock.h> | |
17 | #include <linux/fs.h> | |
18 | #include <linux/mm.h> | |
19 | #include <linux/swap.h> | |
20 | #include <linux/slab.h> | |
21 | #include <linux/pagemap.h> | |
22 | #include <linux/writeback.h> | |
23 | #include <linux/init.h> | |
24 | #include <linux/backing-dev.h> | |
55e829af | 25 | #include <linux/task_io_accounting_ops.h> |
1da177e4 LT |
26 | #include <linux/blkdev.h> |
27 | #include <linux/mpage.h> | |
d08b3851 | 28 | #include <linux/rmap.h> |
1da177e4 LT |
29 | #include <linux/percpu.h> |
30 | #include <linux/notifier.h> | |
31 | #include <linux/smp.h> | |
32 | #include <linux/sysctl.h> | |
33 | #include <linux/cpu.h> | |
34 | #include <linux/syscalls.h> | |
cf9a2ae8 | 35 | #include <linux/buffer_head.h> |
811d736f | 36 | #include <linux/pagevec.h> |
028c2dd1 | 37 | #include <trace/events/writeback.h> |
1da177e4 | 38 | |
1da177e4 LT |
39 | /* |
40 | * 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 | */ | |
43 | static long ratelimit_pages = 32; | |
44 | ||
1da177e4 LT |
45 | /* |
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. | |
3a2e9a5a | 48 | * It should be somewhat larger than dirtied pages to ensure that reasonably |
1da177e4 LT |
49 | * large amounts of I/O are submitted. |
50 | */ | |
3a2e9a5a | 51 | static inline long sync_writeback_pages(unsigned long dirtied) |
1da177e4 | 52 | { |
3a2e9a5a WF |
53 | if (dirtied < ratelimit_pages) |
54 | dirtied = ratelimit_pages; | |
55 | ||
56 | return dirtied + dirtied / 2; | |
1da177e4 LT |
57 | } |
58 | ||
59 | /* The following parameters are exported via /proc/sys/vm */ | |
60 | ||
61 | /* | |
5b0830cb | 62 | * Start background writeback (via writeback threads) at this percentage |
1da177e4 | 63 | */ |
1b5e62b4 | 64 | int dirty_background_ratio = 10; |
1da177e4 | 65 | |
2da02997 DR |
66 | /* |
67 | * dirty_background_bytes starts at 0 (disabled) so that it is a function of | |
68 | * dirty_background_ratio * the amount of dirtyable memory | |
69 | */ | |
70 | unsigned long dirty_background_bytes; | |
71 | ||
195cf453 BG |
72 | /* |
73 | * free highmem will not be subtracted from the total free memory | |
74 | * for calculating free ratios if vm_highmem_is_dirtyable is true | |
75 | */ | |
76 | int vm_highmem_is_dirtyable; | |
77 | ||
1da177e4 LT |
78 | /* |
79 | * The generator of dirty data starts writeback at this percentage | |
80 | */ | |
1b5e62b4 | 81 | int vm_dirty_ratio = 20; |
1da177e4 | 82 | |
2da02997 DR |
83 | /* |
84 | * vm_dirty_bytes starts at 0 (disabled) so that it is a function of | |
85 | * vm_dirty_ratio * the amount of dirtyable memory | |
86 | */ | |
87 | unsigned long vm_dirty_bytes; | |
88 | ||
1da177e4 | 89 | /* |
704503d8 | 90 | * The interval between `kupdate'-style writebacks |
1da177e4 | 91 | */ |
22ef37ee | 92 | unsigned int dirty_writeback_interval = 5 * 100; /* centiseconds */ |
1da177e4 LT |
93 | |
94 | /* | |
704503d8 | 95 | * The longest time for which data is allowed to remain dirty |
1da177e4 | 96 | */ |
22ef37ee | 97 | unsigned int dirty_expire_interval = 30 * 100; /* centiseconds */ |
1da177e4 LT |
98 | |
99 | /* | |
100 | * Flag that makes the machine dump writes/reads and block dirtyings. | |
101 | */ | |
102 | int block_dump; | |
103 | ||
104 | /* | |
ed5b43f1 BS |
105 | * 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. | |
1da177e4 LT |
107 | */ |
108 | int laptop_mode; | |
109 | ||
110 | EXPORT_SYMBOL(laptop_mode); | |
111 | ||
112 | /* End of sysctl-exported parameters */ | |
113 | ||
114 | ||
04fbfdc1 PZ |
115 | /* |
116 | * 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 | */ | |
131 | static struct prop_descriptor vm_completions; | |
3e26c149 | 132 | static struct prop_descriptor vm_dirties; |
04fbfdc1 | 133 | |
04fbfdc1 PZ |
134 | /* |
135 | * couple the period to the dirty_ratio: | |
136 | * | |
137 | * period/2 ~ roundup_pow_of_two(dirty limit) | |
138 | */ | |
139 | static int calc_period_shift(void) | |
140 | { | |
141 | unsigned long dirty_total; | |
142 | ||
2da02997 DR |
143 | 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; | |
04fbfdc1 PZ |
148 | return 2 + ilog2(dirty_total - 1); |
149 | } | |
150 | ||
151 | /* | |
2da02997 | 152 | * update the period when the dirty threshold changes. |
04fbfdc1 | 153 | */ |
2da02997 DR |
154 | static 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 | ||
161 | int dirty_background_ratio_handler(struct ctl_table *table, int write, | |
8d65af78 | 162 | void __user *buffer, size_t *lenp, |
2da02997 DR |
163 | loff_t *ppos) |
164 | { | |
165 | int ret; | |
166 | ||
8d65af78 | 167 | ret = proc_dointvec_minmax(table, write, buffer, lenp, ppos); |
2da02997 DR |
168 | if (ret == 0 && write) |
169 | dirty_background_bytes = 0; | |
170 | return ret; | |
171 | } | |
172 | ||
173 | int dirty_background_bytes_handler(struct ctl_table *table, int write, | |
8d65af78 | 174 | void __user *buffer, size_t *lenp, |
2da02997 DR |
175 | loff_t *ppos) |
176 | { | |
177 | int ret; | |
178 | ||
8d65af78 | 179 | ret = proc_doulongvec_minmax(table, write, buffer, lenp, ppos); |
2da02997 DR |
180 | if (ret == 0 && write) |
181 | dirty_background_ratio = 0; | |
182 | return ret; | |
183 | } | |
184 | ||
04fbfdc1 | 185 | int dirty_ratio_handler(struct ctl_table *table, int write, |
8d65af78 | 186 | void __user *buffer, size_t *lenp, |
04fbfdc1 PZ |
187 | loff_t *ppos) |
188 | { | |
189 | int old_ratio = vm_dirty_ratio; | |
2da02997 DR |
190 | int ret; |
191 | ||
8d65af78 | 192 | ret = proc_dointvec_minmax(table, write, buffer, lenp, ppos); |
04fbfdc1 | 193 | if (ret == 0 && write && vm_dirty_ratio != old_ratio) { |
2da02997 DR |
194 | update_completion_period(); |
195 | vm_dirty_bytes = 0; | |
196 | } | |
197 | return ret; | |
198 | } | |
199 | ||
200 | ||
201 | int dirty_bytes_handler(struct ctl_table *table, int write, | |
8d65af78 | 202 | void __user *buffer, size_t *lenp, |
2da02997 DR |
203 | loff_t *ppos) |
204 | { | |
fc3501d4 | 205 | unsigned long old_bytes = vm_dirty_bytes; |
2da02997 DR |
206 | int ret; |
207 | ||
8d65af78 | 208 | ret = proc_doulongvec_minmax(table, write, buffer, lenp, ppos); |
2da02997 DR |
209 | if (ret == 0 && write && vm_dirty_bytes != old_bytes) { |
210 | update_completion_period(); | |
211 | vm_dirty_ratio = 0; | |
04fbfdc1 PZ |
212 | } |
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 | */ | |
220 | static inline void __bdi_writeout_inc(struct backing_dev_info *bdi) | |
221 | { | |
a42dde04 PZ |
222 | __prop_inc_percpu_max(&vm_completions, &bdi->completions, |
223 | bdi->max_prop_frac); | |
04fbfdc1 PZ |
224 | } |
225 | ||
dd5656e5 MS |
226 | void 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 | } | |
234 | EXPORT_SYMBOL_GPL(bdi_writeout_inc); | |
235 | ||
1cf6e7d8 | 236 | void task_dirty_inc(struct task_struct *tsk) |
3e26c149 PZ |
237 | { |
238 | prop_inc_single(&vm_dirties, &tsk->dirties); | |
239 | } | |
240 | ||
04fbfdc1 PZ |
241 | /* |
242 | * Obtain an accurate fraction of the BDI's portion. | |
243 | */ | |
244 | static void bdi_writeout_fraction(struct backing_dev_info *bdi, | |
245 | long *numerator, long *denominator) | |
246 | { | |
3efaf0fa | 247 | prop_fraction_percpu(&vm_completions, &bdi->completions, |
04fbfdc1 | 248 | numerator, denominator); |
04fbfdc1 PZ |
249 | } |
250 | ||
3e26c149 PZ |
251 | static inline void task_dirties_fraction(struct task_struct *tsk, |
252 | long *numerator, long *denominator) | |
253 | { | |
254 | prop_fraction_single(&vm_dirties, &tsk->dirties, | |
255 | numerator, denominator); | |
256 | } | |
257 | ||
258 | /* | |
1babe183 | 259 | * task_dirty_limit - scale down dirty throttling threshold for one task |
3e26c149 PZ |
260 | * |
261 | * task specific dirty limit: | |
262 | * | |
263 | * dirty -= (dirty/8) * p_{t} | |
1babe183 WF |
264 | * |
265 | * To protect light/slow dirtying tasks from heavier/fast ones, we start | |
266 | * throttling individual tasks before reaching the bdi dirty limit. | |
267 | * Relatively low thresholds will be allocated to heavy dirtiers. So when | |
268 | * dirty pages grow large, heavy dirtiers will be throttled first, which will | |
269 | * effectively curb the growth of dirty pages. Light dirtiers with high enough | |
270 | * dirty threshold may never get throttled. | |
3e26c149 | 271 | */ |
16c4042f WF |
272 | static unsigned long task_dirty_limit(struct task_struct *tsk, |
273 | unsigned long bdi_dirty) | |
3e26c149 PZ |
274 | { |
275 | long numerator, denominator; | |
16c4042f | 276 | unsigned long dirty = bdi_dirty; |
3e26c149 PZ |
277 | u64 inv = dirty >> 3; |
278 | ||
279 | task_dirties_fraction(tsk, &numerator, &denominator); | |
280 | inv *= numerator; | |
281 | do_div(inv, denominator); | |
282 | ||
283 | dirty -= inv; | |
3e26c149 | 284 | |
16c4042f | 285 | return max(dirty, bdi_dirty/2); |
3e26c149 PZ |
286 | } |
287 | ||
189d3c4a PZ |
288 | /* |
289 | * | |
290 | */ | |
189d3c4a PZ |
291 | static unsigned int bdi_min_ratio; |
292 | ||
293 | int bdi_set_min_ratio(struct backing_dev_info *bdi, unsigned int min_ratio) | |
294 | { | |
295 | int ret = 0; | |
189d3c4a | 296 | |
cfc4ba53 | 297 | spin_lock_bh(&bdi_lock); |
a42dde04 | 298 | if (min_ratio > bdi->max_ratio) { |
189d3c4a | 299 | ret = -EINVAL; |
a42dde04 PZ |
300 | } else { |
301 | min_ratio -= bdi->min_ratio; | |
302 | if (bdi_min_ratio + min_ratio < 100) { | |
303 | bdi_min_ratio += min_ratio; | |
304 | bdi->min_ratio += min_ratio; | |
305 | } else { | |
306 | ret = -EINVAL; | |
307 | } | |
308 | } | |
cfc4ba53 | 309 | spin_unlock_bh(&bdi_lock); |
a42dde04 PZ |
310 | |
311 | return ret; | |
312 | } | |
313 | ||
314 | int bdi_set_max_ratio(struct backing_dev_info *bdi, unsigned max_ratio) | |
315 | { | |
a42dde04 PZ |
316 | int ret = 0; |
317 | ||
318 | if (max_ratio > 100) | |
319 | return -EINVAL; | |
320 | ||
cfc4ba53 | 321 | spin_lock_bh(&bdi_lock); |
a42dde04 PZ |
322 | if (bdi->min_ratio > max_ratio) { |
323 | ret = -EINVAL; | |
324 | } else { | |
325 | bdi->max_ratio = max_ratio; | |
326 | bdi->max_prop_frac = (PROP_FRAC_BASE * max_ratio) / 100; | |
327 | } | |
cfc4ba53 | 328 | spin_unlock_bh(&bdi_lock); |
189d3c4a PZ |
329 | |
330 | return ret; | |
331 | } | |
a42dde04 | 332 | EXPORT_SYMBOL(bdi_set_max_ratio); |
189d3c4a | 333 | |
1da177e4 LT |
334 | /* |
335 | * Work out the current dirty-memory clamping and background writeout | |
336 | * thresholds. | |
337 | * | |
338 | * The main aim here is to lower them aggressively if there is a lot of mapped | |
339 | * memory around. To avoid stressing page reclaim with lots of unreclaimable | |
340 | * pages. It is better to clamp down on writers than to start swapping, and | |
341 | * performing lots of scanning. | |
342 | * | |
343 | * We only allow 1/2 of the currently-unmapped memory to be dirtied. | |
344 | * | |
345 | * We don't permit the clamping level to fall below 5% - that is getting rather | |
346 | * excessive. | |
347 | * | |
348 | * We make sure that the background writeout level is below the adjusted | |
349 | * clamping level. | |
350 | */ | |
1b424464 CL |
351 | |
352 | static unsigned long highmem_dirtyable_memory(unsigned long total) | |
353 | { | |
354 | #ifdef CONFIG_HIGHMEM | |
355 | int node; | |
356 | unsigned long x = 0; | |
357 | ||
37b07e41 | 358 | for_each_node_state(node, N_HIGH_MEMORY) { |
1b424464 CL |
359 | struct zone *z = |
360 | &NODE_DATA(node)->node_zones[ZONE_HIGHMEM]; | |
361 | ||
adea02a1 WF |
362 | x += zone_page_state(z, NR_FREE_PAGES) + |
363 | zone_reclaimable_pages(z); | |
1b424464 CL |
364 | } |
365 | /* | |
366 | * Make sure that the number of highmem pages is never larger | |
367 | * than the number of the total dirtyable memory. This can only | |
368 | * occur in very strange VM situations but we want to make sure | |
369 | * that this does not occur. | |
370 | */ | |
371 | return min(x, total); | |
372 | #else | |
373 | return 0; | |
374 | #endif | |
375 | } | |
376 | ||
3eefae99 SR |
377 | /** |
378 | * determine_dirtyable_memory - amount of memory that may be used | |
379 | * | |
380 | * Returns the numebr of pages that can currently be freed and used | |
381 | * by the kernel for direct mappings. | |
382 | */ | |
383 | unsigned long determine_dirtyable_memory(void) | |
1b424464 CL |
384 | { |
385 | unsigned long x; | |
386 | ||
adea02a1 | 387 | x = global_page_state(NR_FREE_PAGES) + global_reclaimable_pages(); |
195cf453 BG |
388 | |
389 | if (!vm_highmem_is_dirtyable) | |
390 | x -= highmem_dirtyable_memory(x); | |
391 | ||
1b424464 CL |
392 | return x + 1; /* Ensure that we never return 0 */ |
393 | } | |
394 | ||
03ab450f | 395 | /* |
1babe183 WF |
396 | * global_dirty_limits - background-writeback and dirty-throttling thresholds |
397 | * | |
398 | * Calculate the dirty thresholds based on sysctl parameters | |
399 | * - vm.dirty_background_ratio or vm.dirty_background_bytes | |
400 | * - vm.dirty_ratio or vm.dirty_bytes | |
401 | * The dirty limits will be lifted by 1/4 for PF_LESS_THROTTLE (ie. nfsd) and | |
ebd1373d | 402 | * real-time tasks. |
1babe183 | 403 | */ |
16c4042f | 404 | void global_dirty_limits(unsigned long *pbackground, unsigned long *pdirty) |
1da177e4 | 405 | { |
364aeb28 DR |
406 | unsigned long background; |
407 | unsigned long dirty; | |
240c879f | 408 | unsigned long uninitialized_var(available_memory); |
1da177e4 LT |
409 | struct task_struct *tsk; |
410 | ||
240c879f MK |
411 | if (!vm_dirty_bytes || !dirty_background_bytes) |
412 | available_memory = determine_dirtyable_memory(); | |
413 | ||
2da02997 DR |
414 | if (vm_dirty_bytes) |
415 | dirty = DIV_ROUND_UP(vm_dirty_bytes, PAGE_SIZE); | |
4cbec4c8 WF |
416 | else |
417 | dirty = (vm_dirty_ratio * available_memory) / 100; | |
1da177e4 | 418 | |
2da02997 DR |
419 | if (dirty_background_bytes) |
420 | background = DIV_ROUND_UP(dirty_background_bytes, PAGE_SIZE); | |
421 | else | |
422 | background = (dirty_background_ratio * available_memory) / 100; | |
1da177e4 | 423 | |
2da02997 DR |
424 | if (background >= dirty) |
425 | background = dirty / 2; | |
1da177e4 LT |
426 | tsk = current; |
427 | if (tsk->flags & PF_LESS_THROTTLE || rt_task(tsk)) { | |
428 | background += background / 4; | |
429 | dirty += dirty / 4; | |
430 | } | |
431 | *pbackground = background; | |
432 | *pdirty = dirty; | |
16c4042f | 433 | } |
04fbfdc1 | 434 | |
6f718656 | 435 | /** |
1babe183 | 436 | * bdi_dirty_limit - @bdi's share of dirty throttling threshold |
6f718656 WF |
437 | * @bdi: the backing_dev_info to query |
438 | * @dirty: global dirty limit in pages | |
439 | * | |
440 | * Returns @bdi's dirty limit in pages. The term "dirty" in the context of | |
441 | * dirty balancing includes all PG_dirty, PG_writeback and NFS unstable pages. | |
442 | * And the "limit" in the name is not seriously taken as hard limit in | |
443 | * balance_dirty_pages(). | |
1babe183 | 444 | * |
6f718656 | 445 | * It allocates high/low dirty limits to fast/slow devices, in order to prevent |
1babe183 WF |
446 | * - starving fast devices |
447 | * - piling up dirty pages (that will take long time to sync) on slow devices | |
448 | * | |
449 | * The bdi's share of dirty limit will be adapting to its throughput and | |
450 | * bounded by the bdi->min_ratio and/or bdi->max_ratio parameters, if set. | |
451 | */ | |
452 | unsigned long bdi_dirty_limit(struct backing_dev_info *bdi, unsigned long dirty) | |
16c4042f WF |
453 | { |
454 | u64 bdi_dirty; | |
455 | long numerator, denominator; | |
04fbfdc1 | 456 | |
16c4042f WF |
457 | /* |
458 | * Calculate this BDI's share of the dirty ratio. | |
459 | */ | |
460 | bdi_writeout_fraction(bdi, &numerator, &denominator); | |
04fbfdc1 | 461 | |
16c4042f WF |
462 | bdi_dirty = (dirty * (100 - bdi_min_ratio)) / 100; |
463 | bdi_dirty *= numerator; | |
464 | do_div(bdi_dirty, denominator); | |
04fbfdc1 | 465 | |
16c4042f WF |
466 | bdi_dirty += (dirty * bdi->min_ratio) / 100; |
467 | if (bdi_dirty > (dirty * bdi->max_ratio) / 100) | |
468 | bdi_dirty = dirty * bdi->max_ratio / 100; | |
469 | ||
470 | return bdi_dirty; | |
1da177e4 LT |
471 | } |
472 | ||
473 | /* | |
474 | * balance_dirty_pages() must be called by processes which are generating dirty | |
475 | * data. It looks at the number of dirty pages in the machine and will force | |
476 | * the caller to perform writeback if the system is over `vm_dirty_ratio'. | |
5b0830cb JA |
477 | * If we're over `background_thresh' then the writeback threads are woken to |
478 | * perform some writeout. | |
1da177e4 | 479 | */ |
3a2e9a5a WF |
480 | static void balance_dirty_pages(struct address_space *mapping, |
481 | unsigned long write_chunk) | |
1da177e4 | 482 | { |
5fce25a9 PZ |
483 | long nr_reclaimable, bdi_nr_reclaimable; |
484 | long nr_writeback, bdi_nr_writeback; | |
364aeb28 DR |
485 | unsigned long background_thresh; |
486 | unsigned long dirty_thresh; | |
487 | unsigned long bdi_thresh; | |
1da177e4 | 488 | unsigned long pages_written = 0; |
87c6a9b2 | 489 | unsigned long pause = 1; |
e50e3720 | 490 | bool dirty_exceeded = false; |
1da177e4 LT |
491 | struct backing_dev_info *bdi = mapping->backing_dev_info; |
492 | ||
3efaf0fa WF |
493 | if (!bdi_cap_account_dirty(bdi)) |
494 | return; | |
495 | ||
1da177e4 LT |
496 | for (;;) { |
497 | struct writeback_control wbc = { | |
1da177e4 LT |
498 | .sync_mode = WB_SYNC_NONE, |
499 | .older_than_this = NULL, | |
500 | .nr_to_write = write_chunk, | |
111ebb6e | 501 | .range_cyclic = 1, |
1da177e4 LT |
502 | }; |
503 | ||
5fce25a9 PZ |
504 | nr_reclaimable = global_page_state(NR_FILE_DIRTY) + |
505 | global_page_state(NR_UNSTABLE_NFS); | |
506 | nr_writeback = global_page_state(NR_WRITEBACK); | |
507 | ||
16c4042f WF |
508 | global_dirty_limits(&background_thresh, &dirty_thresh); |
509 | ||
510 | /* | |
511 | * Throttle it only when the background writeback cannot | |
512 | * catch-up. This avoids (excessively) small writeouts | |
513 | * when the bdi limits are ramping up. | |
514 | */ | |
4cbec4c8 | 515 | if (nr_reclaimable + nr_writeback <= |
16c4042f WF |
516 | (background_thresh + dirty_thresh) / 2) |
517 | break; | |
518 | ||
519 | bdi_thresh = bdi_dirty_limit(bdi, dirty_thresh); | |
520 | bdi_thresh = task_dirty_limit(current, bdi_thresh); | |
521 | ||
e50e3720 WF |
522 | /* |
523 | * In order to avoid the stacked BDI deadlock we need | |
524 | * to ensure we accurately count the 'dirty' pages when | |
525 | * the threshold is low. | |
526 | * | |
527 | * Otherwise it would be possible to get thresh+n pages | |
528 | * reported dirty, even though there are thresh-m pages | |
529 | * actually dirty; with m+n sitting in the percpu | |
530 | * deltas. | |
531 | */ | |
532 | if (bdi_thresh < 2*bdi_stat_error(bdi)) { | |
533 | bdi_nr_reclaimable = bdi_stat_sum(bdi, BDI_RECLAIMABLE); | |
534 | bdi_nr_writeback = bdi_stat_sum(bdi, BDI_WRITEBACK); | |
535 | } else { | |
536 | bdi_nr_reclaimable = bdi_stat(bdi, BDI_RECLAIMABLE); | |
537 | bdi_nr_writeback = bdi_stat(bdi, BDI_WRITEBACK); | |
538 | } | |
5fce25a9 | 539 | |
e50e3720 WF |
540 | /* |
541 | * The bdi thresh is somehow "soft" limit derived from the | |
542 | * global "hard" limit. The former helps to prevent heavy IO | |
543 | * bdi or process from holding back light ones; The latter is | |
544 | * the last resort safeguard. | |
545 | */ | |
546 | dirty_exceeded = | |
4cbec4c8 WF |
547 | (bdi_nr_reclaimable + bdi_nr_writeback > bdi_thresh) |
548 | || (nr_reclaimable + nr_writeback > dirty_thresh); | |
e50e3720 WF |
549 | |
550 | if (!dirty_exceeded) | |
04fbfdc1 | 551 | break; |
1da177e4 | 552 | |
04fbfdc1 PZ |
553 | if (!bdi->dirty_exceeded) |
554 | bdi->dirty_exceeded = 1; | |
1da177e4 LT |
555 | |
556 | /* Note: nr_reclaimable denotes nr_dirty + nr_unstable. | |
557 | * Unstable writes are a feature of certain networked | |
558 | * filesystems (i.e. NFS) in which data may have been | |
559 | * written to the server's write cache, but has not yet | |
560 | * been flushed to permanent storage. | |
d7831a0b RK |
561 | * Only move pages to writeback if this bdi is over its |
562 | * threshold otherwise wait until the disk writes catch | |
563 | * up. | |
1da177e4 | 564 | */ |
028c2dd1 | 565 | trace_wbc_balance_dirty_start(&wbc, bdi); |
d7831a0b | 566 | if (bdi_nr_reclaimable > bdi_thresh) { |
9c3a8ee8 | 567 | writeback_inodes_wb(&bdi->wb, &wbc); |
1da177e4 | 568 | pages_written += write_chunk - wbc.nr_to_write; |
028c2dd1 | 569 | trace_wbc_balance_dirty_written(&wbc, bdi); |
e50e3720 WF |
570 | if (pages_written >= write_chunk) |
571 | break; /* We've done our duty */ | |
04fbfdc1 | 572 | } |
028c2dd1 | 573 | trace_wbc_balance_dirty_wait(&wbc, bdi); |
d153ba64 | 574 | __set_current_state(TASK_UNINTERRUPTIBLE); |
d25105e8 | 575 | io_schedule_timeout(pause); |
87c6a9b2 JA |
576 | |
577 | /* | |
578 | * Increase the delay for each loop, up to our previous | |
579 | * default of taking a 100ms nap. | |
580 | */ | |
581 | pause <<= 1; | |
582 | if (pause > HZ / 10) | |
583 | pause = HZ / 10; | |
1da177e4 LT |
584 | } |
585 | ||
e50e3720 | 586 | if (!dirty_exceeded && bdi->dirty_exceeded) |
04fbfdc1 | 587 | bdi->dirty_exceeded = 0; |
1da177e4 LT |
588 | |
589 | if (writeback_in_progress(bdi)) | |
5b0830cb | 590 | return; |
1da177e4 LT |
591 | |
592 | /* | |
593 | * In laptop mode, we wait until hitting the higher threshold before | |
594 | * starting background writeout, and then write out all the way down | |
595 | * to the lower threshold. So slow writers cause minimal disk activity. | |
596 | * | |
597 | * In normal mode, we start background writeout at the lower | |
598 | * background_thresh, to keep the amount of dirty memory low. | |
599 | */ | |
600 | if ((laptop_mode && pages_written) || | |
e50e3720 | 601 | (!laptop_mode && (nr_reclaimable > background_thresh))) |
c5444198 | 602 | bdi_start_background_writeback(bdi); |
1da177e4 LT |
603 | } |
604 | ||
a200ee18 | 605 | void set_page_dirty_balance(struct page *page, int page_mkwrite) |
edc79b2a | 606 | { |
a200ee18 | 607 | if (set_page_dirty(page) || page_mkwrite) { |
edc79b2a PZ |
608 | struct address_space *mapping = page_mapping(page); |
609 | ||
610 | if (mapping) | |
611 | balance_dirty_pages_ratelimited(mapping); | |
612 | } | |
613 | } | |
614 | ||
245b2e70 TH |
615 | static DEFINE_PER_CPU(unsigned long, bdp_ratelimits) = 0; |
616 | ||
1da177e4 | 617 | /** |
fa5a734e | 618 | * balance_dirty_pages_ratelimited_nr - balance dirty memory state |
67be2dd1 | 619 | * @mapping: address_space which was dirtied |
a580290c | 620 | * @nr_pages_dirtied: number of pages which the caller has just dirtied |
1da177e4 LT |
621 | * |
622 | * Processes which are dirtying memory should call in here once for each page | |
623 | * which was newly dirtied. The function will periodically check the system's | |
624 | * dirty state and will initiate writeback if needed. | |
625 | * | |
626 | * On really big machines, get_writeback_state is expensive, so try to avoid | |
627 | * calling it too often (ratelimiting). But once we're over the dirty memory | |
628 | * limit we decrease the ratelimiting by a lot, to prevent individual processes | |
629 | * from overshooting the limit by (ratelimit_pages) each. | |
630 | */ | |
fa5a734e AM |
631 | void balance_dirty_pages_ratelimited_nr(struct address_space *mapping, |
632 | unsigned long nr_pages_dirtied) | |
1da177e4 | 633 | { |
fa5a734e AM |
634 | unsigned long ratelimit; |
635 | unsigned long *p; | |
1da177e4 LT |
636 | |
637 | ratelimit = ratelimit_pages; | |
04fbfdc1 | 638 | if (mapping->backing_dev_info->dirty_exceeded) |
1da177e4 LT |
639 | ratelimit = 8; |
640 | ||
641 | /* | |
642 | * Check the rate limiting. Also, we do not want to throttle real-time | |
643 | * tasks in balance_dirty_pages(). Period. | |
644 | */ | |
fa5a734e | 645 | preempt_disable(); |
245b2e70 | 646 | p = &__get_cpu_var(bdp_ratelimits); |
fa5a734e AM |
647 | *p += nr_pages_dirtied; |
648 | if (unlikely(*p >= ratelimit)) { | |
3a2e9a5a | 649 | ratelimit = sync_writeback_pages(*p); |
fa5a734e AM |
650 | *p = 0; |
651 | preempt_enable(); | |
3a2e9a5a | 652 | balance_dirty_pages(mapping, ratelimit); |
1da177e4 LT |
653 | return; |
654 | } | |
fa5a734e | 655 | preempt_enable(); |
1da177e4 | 656 | } |
fa5a734e | 657 | EXPORT_SYMBOL(balance_dirty_pages_ratelimited_nr); |
1da177e4 | 658 | |
232ea4d6 | 659 | void throttle_vm_writeout(gfp_t gfp_mask) |
1da177e4 | 660 | { |
364aeb28 DR |
661 | unsigned long background_thresh; |
662 | unsigned long dirty_thresh; | |
1da177e4 LT |
663 | |
664 | for ( ; ; ) { | |
16c4042f | 665 | global_dirty_limits(&background_thresh, &dirty_thresh); |
1da177e4 LT |
666 | |
667 | /* | |
668 | * Boost the allowable dirty threshold a bit for page | |
669 | * allocators so they don't get DoS'ed by heavy writers | |
670 | */ | |
671 | dirty_thresh += dirty_thresh / 10; /* wheeee... */ | |
672 | ||
c24f21bd CL |
673 | if (global_page_state(NR_UNSTABLE_NFS) + |
674 | global_page_state(NR_WRITEBACK) <= dirty_thresh) | |
675 | break; | |
8aa7e847 | 676 | congestion_wait(BLK_RW_ASYNC, HZ/10); |
369f2389 FW |
677 | |
678 | /* | |
679 | * The caller might hold locks which can prevent IO completion | |
680 | * or progress in the filesystem. So we cannot just sit here | |
681 | * waiting for IO to complete. | |
682 | */ | |
683 | if ((gfp_mask & (__GFP_FS|__GFP_IO)) != (__GFP_FS|__GFP_IO)) | |
684 | break; | |
1da177e4 LT |
685 | } |
686 | } | |
687 | ||
1da177e4 LT |
688 | /* |
689 | * sysctl handler for /proc/sys/vm/dirty_writeback_centisecs | |
690 | */ | |
691 | int dirty_writeback_centisecs_handler(ctl_table *table, int write, | |
8d65af78 | 692 | void __user *buffer, size_t *length, loff_t *ppos) |
1da177e4 | 693 | { |
8d65af78 | 694 | proc_dointvec(table, write, buffer, length, ppos); |
6423104b | 695 | bdi_arm_supers_timer(); |
1da177e4 LT |
696 | return 0; |
697 | } | |
698 | ||
c2c4986e | 699 | #ifdef CONFIG_BLOCK |
31373d09 | 700 | void laptop_mode_timer_fn(unsigned long data) |
1da177e4 | 701 | { |
31373d09 MG |
702 | struct request_queue *q = (struct request_queue *)data; |
703 | int nr_pages = global_page_state(NR_FILE_DIRTY) + | |
704 | global_page_state(NR_UNSTABLE_NFS); | |
1da177e4 | 705 | |
31373d09 MG |
706 | /* |
707 | * We want to write everything out, not just down to the dirty | |
708 | * threshold | |
709 | */ | |
31373d09 | 710 | if (bdi_has_dirty_io(&q->backing_dev_info)) |
c5444198 | 711 | bdi_start_writeback(&q->backing_dev_info, nr_pages); |
1da177e4 LT |
712 | } |
713 | ||
714 | /* | |
715 | * We've spun up the disk and we're in laptop mode: schedule writeback | |
716 | * of all dirty data a few seconds from now. If the flush is already scheduled | |
717 | * then push it back - the user is still using the disk. | |
718 | */ | |
31373d09 | 719 | void laptop_io_completion(struct backing_dev_info *info) |
1da177e4 | 720 | { |
31373d09 | 721 | mod_timer(&info->laptop_mode_wb_timer, jiffies + laptop_mode); |
1da177e4 LT |
722 | } |
723 | ||
724 | /* | |
725 | * We're in laptop mode and we've just synced. The sync's writes will have | |
726 | * caused another writeback to be scheduled by laptop_io_completion. | |
727 | * Nothing needs to be written back anymore, so we unschedule the writeback. | |
728 | */ | |
729 | void laptop_sync_completion(void) | |
730 | { | |
31373d09 MG |
731 | struct backing_dev_info *bdi; |
732 | ||
733 | rcu_read_lock(); | |
734 | ||
735 | list_for_each_entry_rcu(bdi, &bdi_list, bdi_list) | |
736 | del_timer(&bdi->laptop_mode_wb_timer); | |
737 | ||
738 | rcu_read_unlock(); | |
1da177e4 | 739 | } |
c2c4986e | 740 | #endif |
1da177e4 LT |
741 | |
742 | /* | |
743 | * If ratelimit_pages is too high then we can get into dirty-data overload | |
744 | * if a large number of processes all perform writes at the same time. | |
745 | * If it is too low then SMP machines will call the (expensive) | |
746 | * get_writeback_state too often. | |
747 | * | |
748 | * Here we set ratelimit_pages to a level which ensures that when all CPUs are | |
749 | * dirtying in parallel, we cannot go more than 3% (1/32) over the dirty memory | |
750 | * thresholds before writeback cuts in. | |
751 | * | |
752 | * But the limit should not be set too high. Because it also controls the | |
753 | * amount of memory which the balance_dirty_pages() caller has to write back. | |
754 | * If this is too large then the caller will block on the IO queue all the | |
755 | * time. So limit it to four megabytes - the balance_dirty_pages() caller | |
756 | * will write six megabyte chunks, max. | |
757 | */ | |
758 | ||
2d1d43f6 | 759 | void writeback_set_ratelimit(void) |
1da177e4 | 760 | { |
40c99aae | 761 | ratelimit_pages = vm_total_pages / (num_online_cpus() * 32); |
1da177e4 LT |
762 | if (ratelimit_pages < 16) |
763 | ratelimit_pages = 16; | |
764 | if (ratelimit_pages * PAGE_CACHE_SIZE > 4096 * 1024) | |
765 | ratelimit_pages = (4096 * 1024) / PAGE_CACHE_SIZE; | |
766 | } | |
767 | ||
26c2143b | 768 | static int __cpuinit |
1da177e4 LT |
769 | ratelimit_handler(struct notifier_block *self, unsigned long u, void *v) |
770 | { | |
2d1d43f6 | 771 | writeback_set_ratelimit(); |
aa0f0303 | 772 | return NOTIFY_DONE; |
1da177e4 LT |
773 | } |
774 | ||
74b85f37 | 775 | static struct notifier_block __cpuinitdata ratelimit_nb = { |
1da177e4 LT |
776 | .notifier_call = ratelimit_handler, |
777 | .next = NULL, | |
778 | }; | |
779 | ||
780 | /* | |
dc6e29da LT |
781 | * Called early on to tune the page writeback dirty limits. |
782 | * | |
783 | * We used to scale dirty pages according to how total memory | |
784 | * related to pages that could be allocated for buffers (by | |
785 | * comparing nr_free_buffer_pages() to vm_total_pages. | |
786 | * | |
787 | * However, that was when we used "dirty_ratio" to scale with | |
788 | * all memory, and we don't do that any more. "dirty_ratio" | |
789 | * is now applied to total non-HIGHPAGE memory (by subtracting | |
790 | * totalhigh_pages from vm_total_pages), and as such we can't | |
791 | * get into the old insane situation any more where we had | |
792 | * large amounts of dirty pages compared to a small amount of | |
793 | * non-HIGHMEM memory. | |
794 | * | |
795 | * But we might still want to scale the dirty_ratio by how | |
796 | * much memory the box has.. | |
1da177e4 LT |
797 | */ |
798 | void __init page_writeback_init(void) | |
799 | { | |
04fbfdc1 PZ |
800 | int shift; |
801 | ||
2d1d43f6 | 802 | writeback_set_ratelimit(); |
1da177e4 | 803 | register_cpu_notifier(&ratelimit_nb); |
04fbfdc1 PZ |
804 | |
805 | shift = calc_period_shift(); | |
806 | prop_descriptor_init(&vm_completions, shift); | |
3e26c149 | 807 | prop_descriptor_init(&vm_dirties, shift); |
1da177e4 LT |
808 | } |
809 | ||
f446daae JK |
810 | /** |
811 | * tag_pages_for_writeback - tag pages to be written by write_cache_pages | |
812 | * @mapping: address space structure to write | |
813 | * @start: starting page index | |
814 | * @end: ending page index (inclusive) | |
815 | * | |
816 | * This function scans the page range from @start to @end (inclusive) and tags | |
817 | * all pages that have DIRTY tag set with a special TOWRITE tag. The idea is | |
818 | * that write_cache_pages (or whoever calls this function) will then use | |
819 | * TOWRITE tag to identify pages eligible for writeback. This mechanism is | |
820 | * used to avoid livelocking of writeback by a process steadily creating new | |
821 | * dirty pages in the file (thus it is important for this function to be quick | |
822 | * so that it can tag pages faster than a dirtying process can create them). | |
823 | */ | |
824 | /* | |
825 | * We tag pages in batches of WRITEBACK_TAG_BATCH to reduce tree_lock latency. | |
826 | */ | |
f446daae JK |
827 | void tag_pages_for_writeback(struct address_space *mapping, |
828 | pgoff_t start, pgoff_t end) | |
829 | { | |
3c111a07 | 830 | #define WRITEBACK_TAG_BATCH 4096 |
f446daae JK |
831 | unsigned long tagged; |
832 | ||
833 | do { | |
834 | spin_lock_irq(&mapping->tree_lock); | |
835 | tagged = radix_tree_range_tag_if_tagged(&mapping->page_tree, | |
836 | &start, end, WRITEBACK_TAG_BATCH, | |
837 | PAGECACHE_TAG_DIRTY, PAGECACHE_TAG_TOWRITE); | |
838 | spin_unlock_irq(&mapping->tree_lock); | |
839 | WARN_ON_ONCE(tagged > WRITEBACK_TAG_BATCH); | |
840 | cond_resched(); | |
d5ed3a4a JK |
841 | /* We check 'start' to handle wrapping when end == ~0UL */ |
842 | } while (tagged >= WRITEBACK_TAG_BATCH && start); | |
f446daae JK |
843 | } |
844 | EXPORT_SYMBOL(tag_pages_for_writeback); | |
845 | ||
811d736f | 846 | /** |
0ea97180 | 847 | * write_cache_pages - walk the list of dirty pages of the given address space and write all of them. |
811d736f DH |
848 | * @mapping: address space structure to write |
849 | * @wbc: subtract the number of written pages from *@wbc->nr_to_write | |
0ea97180 MS |
850 | * @writepage: function called for each page |
851 | * @data: data passed to writepage function | |
811d736f | 852 | * |
0ea97180 | 853 | * If a page is already under I/O, write_cache_pages() skips it, even |
811d736f DH |
854 | * if it's dirty. This is desirable behaviour for memory-cleaning writeback, |
855 | * but it is INCORRECT for data-integrity system calls such as fsync(). fsync() | |
856 | * and msync() need to guarantee that all the data which was dirty at the time | |
857 | * the call was made get new I/O started against them. If wbc->sync_mode is | |
858 | * WB_SYNC_ALL then we were called for data integrity and we must wait for | |
859 | * existing IO to complete. | |
f446daae JK |
860 | * |
861 | * To avoid livelocks (when other process dirties new pages), we first tag | |
862 | * pages which should be written back with TOWRITE tag and only then start | |
863 | * writing them. For data-integrity sync we have to be careful so that we do | |
864 | * not miss some pages (e.g., because some other process has cleared TOWRITE | |
865 | * tag we set). The rule we follow is that TOWRITE tag can be cleared only | |
866 | * by the process clearing the DIRTY tag (and submitting the page for IO). | |
811d736f | 867 | */ |
0ea97180 MS |
868 | int write_cache_pages(struct address_space *mapping, |
869 | struct writeback_control *wbc, writepage_t writepage, | |
870 | void *data) | |
811d736f | 871 | { |
811d736f DH |
872 | int ret = 0; |
873 | int done = 0; | |
811d736f DH |
874 | struct pagevec pvec; |
875 | int nr_pages; | |
31a12666 | 876 | pgoff_t uninitialized_var(writeback_index); |
811d736f DH |
877 | pgoff_t index; |
878 | pgoff_t end; /* Inclusive */ | |
bd19e012 | 879 | pgoff_t done_index; |
31a12666 | 880 | int cycled; |
811d736f | 881 | int range_whole = 0; |
f446daae | 882 | int tag; |
811d736f | 883 | |
811d736f DH |
884 | pagevec_init(&pvec, 0); |
885 | if (wbc->range_cyclic) { | |
31a12666 NP |
886 | writeback_index = mapping->writeback_index; /* prev offset */ |
887 | index = writeback_index; | |
888 | if (index == 0) | |
889 | cycled = 1; | |
890 | else | |
891 | cycled = 0; | |
811d736f DH |
892 | end = -1; |
893 | } else { | |
894 | index = wbc->range_start >> PAGE_CACHE_SHIFT; | |
895 | end = wbc->range_end >> PAGE_CACHE_SHIFT; | |
896 | if (wbc->range_start == 0 && wbc->range_end == LLONG_MAX) | |
897 | range_whole = 1; | |
31a12666 | 898 | cycled = 1; /* ignore range_cyclic tests */ |
811d736f | 899 | } |
6e6938b6 | 900 | if (wbc->sync_mode == WB_SYNC_ALL || wbc->tagged_writepages) |
f446daae JK |
901 | tag = PAGECACHE_TAG_TOWRITE; |
902 | else | |
903 | tag = PAGECACHE_TAG_DIRTY; | |
811d736f | 904 | retry: |
6e6938b6 | 905 | if (wbc->sync_mode == WB_SYNC_ALL || wbc->tagged_writepages) |
f446daae | 906 | tag_pages_for_writeback(mapping, index, end); |
bd19e012 | 907 | done_index = index; |
5a3d5c98 NP |
908 | while (!done && (index <= end)) { |
909 | int i; | |
910 | ||
f446daae | 911 | nr_pages = pagevec_lookup_tag(&pvec, mapping, &index, tag, |
5a3d5c98 NP |
912 | min(end - index, (pgoff_t)PAGEVEC_SIZE-1) + 1); |
913 | if (nr_pages == 0) | |
914 | break; | |
811d736f | 915 | |
811d736f DH |
916 | for (i = 0; i < nr_pages; i++) { |
917 | struct page *page = pvec.pages[i]; | |
918 | ||
919 | /* | |
d5482cdf NP |
920 | * At this point, the page may be truncated or |
921 | * invalidated (changing page->mapping to NULL), or | |
922 | * even swizzled back from swapper_space to tmpfs file | |
923 | * mapping. However, page->index will not change | |
924 | * because we have a reference on the page. | |
811d736f | 925 | */ |
d5482cdf NP |
926 | if (page->index > end) { |
927 | /* | |
928 | * can't be range_cyclic (1st pass) because | |
929 | * end == -1 in that case. | |
930 | */ | |
931 | done = 1; | |
932 | break; | |
933 | } | |
934 | ||
cf15b07c | 935 | done_index = page->index; |
d5482cdf | 936 | |
811d736f DH |
937 | lock_page(page); |
938 | ||
5a3d5c98 NP |
939 | /* |
940 | * Page truncated or invalidated. We can freely skip it | |
941 | * then, even for data integrity operations: the page | |
942 | * has disappeared concurrently, so there could be no | |
943 | * real expectation of this data interity operation | |
944 | * even if there is now a new, dirty page at the same | |
945 | * pagecache address. | |
946 | */ | |
811d736f | 947 | if (unlikely(page->mapping != mapping)) { |
5a3d5c98 | 948 | continue_unlock: |
811d736f DH |
949 | unlock_page(page); |
950 | continue; | |
951 | } | |
952 | ||
515f4a03 NP |
953 | if (!PageDirty(page)) { |
954 | /* someone wrote it for us */ | |
955 | goto continue_unlock; | |
956 | } | |
957 | ||
958 | if (PageWriteback(page)) { | |
959 | if (wbc->sync_mode != WB_SYNC_NONE) | |
960 | wait_on_page_writeback(page); | |
961 | else | |
962 | goto continue_unlock; | |
963 | } | |
811d736f | 964 | |
515f4a03 NP |
965 | BUG_ON(PageWriteback(page)); |
966 | if (!clear_page_dirty_for_io(page)) | |
5a3d5c98 | 967 | goto continue_unlock; |
811d736f | 968 | |
9e094383 | 969 | trace_wbc_writepage(wbc, mapping->backing_dev_info); |
0ea97180 | 970 | ret = (*writepage)(page, wbc, data); |
00266770 NP |
971 | if (unlikely(ret)) { |
972 | if (ret == AOP_WRITEPAGE_ACTIVATE) { | |
973 | unlock_page(page); | |
974 | ret = 0; | |
975 | } else { | |
976 | /* | |
977 | * done_index is set past this page, | |
978 | * so media errors will not choke | |
979 | * background writeout for the entire | |
980 | * file. This has consequences for | |
981 | * range_cyclic semantics (ie. it may | |
982 | * not be suitable for data integrity | |
983 | * writeout). | |
984 | */ | |
cf15b07c | 985 | done_index = page->index + 1; |
00266770 NP |
986 | done = 1; |
987 | break; | |
988 | } | |
0b564927 | 989 | } |
00266770 | 990 | |
546a1924 DC |
991 | /* |
992 | * We stop writing back only if we are not doing | |
993 | * integrity sync. In case of integrity sync we have to | |
994 | * keep going until we have written all the pages | |
995 | * we tagged for writeback prior to entering this loop. | |
996 | */ | |
997 | if (--wbc->nr_to_write <= 0 && | |
998 | wbc->sync_mode == WB_SYNC_NONE) { | |
999 | done = 1; | |
1000 | break; | |
05fe478d | 1001 | } |
811d736f DH |
1002 | } |
1003 | pagevec_release(&pvec); | |
1004 | cond_resched(); | |
1005 | } | |
3a4c6800 | 1006 | if (!cycled && !done) { |
811d736f | 1007 | /* |
31a12666 | 1008 | * range_cyclic: |
811d736f DH |
1009 | * We hit the last page and there is more work to be done: wrap |
1010 | * back to the start of the file | |
1011 | */ | |
31a12666 | 1012 | cycled = 1; |
811d736f | 1013 | index = 0; |
31a12666 | 1014 | end = writeback_index - 1; |
811d736f DH |
1015 | goto retry; |
1016 | } | |
0b564927 DC |
1017 | if (wbc->range_cyclic || (range_whole && wbc->nr_to_write > 0)) |
1018 | mapping->writeback_index = done_index; | |
06d6cf69 | 1019 | |
811d736f DH |
1020 | return ret; |
1021 | } | |
0ea97180 MS |
1022 | EXPORT_SYMBOL(write_cache_pages); |
1023 | ||
1024 | /* | |
1025 | * Function used by generic_writepages to call the real writepage | |
1026 | * function and set the mapping flags on error | |
1027 | */ | |
1028 | static int __writepage(struct page *page, struct writeback_control *wbc, | |
1029 | void *data) | |
1030 | { | |
1031 | struct address_space *mapping = data; | |
1032 | int ret = mapping->a_ops->writepage(page, wbc); | |
1033 | mapping_set_error(mapping, ret); | |
1034 | return ret; | |
1035 | } | |
1036 | ||
1037 | /** | |
1038 | * generic_writepages - walk the list of dirty pages of the given address space and writepage() all of them. | |
1039 | * @mapping: address space structure to write | |
1040 | * @wbc: subtract the number of written pages from *@wbc->nr_to_write | |
1041 | * | |
1042 | * This is a library function, which implements the writepages() | |
1043 | * address_space_operation. | |
1044 | */ | |
1045 | int generic_writepages(struct address_space *mapping, | |
1046 | struct writeback_control *wbc) | |
1047 | { | |
9b6096a6 SL |
1048 | struct blk_plug plug; |
1049 | int ret; | |
1050 | ||
0ea97180 MS |
1051 | /* deal with chardevs and other special file */ |
1052 | if (!mapping->a_ops->writepage) | |
1053 | return 0; | |
1054 | ||
9b6096a6 SL |
1055 | blk_start_plug(&plug); |
1056 | ret = write_cache_pages(mapping, wbc, __writepage, mapping); | |
1057 | blk_finish_plug(&plug); | |
1058 | return ret; | |
0ea97180 | 1059 | } |
811d736f DH |
1060 | |
1061 | EXPORT_SYMBOL(generic_writepages); | |
1062 | ||
1da177e4 LT |
1063 | int do_writepages(struct address_space *mapping, struct writeback_control *wbc) |
1064 | { | |
22905f77 AM |
1065 | int ret; |
1066 | ||
1da177e4 LT |
1067 | if (wbc->nr_to_write <= 0) |
1068 | return 0; | |
1069 | if (mapping->a_ops->writepages) | |
d08b3851 | 1070 | ret = mapping->a_ops->writepages(mapping, wbc); |
22905f77 AM |
1071 | else |
1072 | ret = generic_writepages(mapping, wbc); | |
22905f77 | 1073 | return ret; |
1da177e4 LT |
1074 | } |
1075 | ||
1076 | /** | |
1077 | * write_one_page - write out a single page and optionally wait on I/O | |
67be2dd1 MW |
1078 | * @page: the page to write |
1079 | * @wait: if true, wait on writeout | |
1da177e4 LT |
1080 | * |
1081 | * The page must be locked by the caller and will be unlocked upon return. | |
1082 | * | |
1083 | * write_one_page() returns a negative error code if I/O failed. | |
1084 | */ | |
1085 | int write_one_page(struct page *page, int wait) | |
1086 | { | |
1087 | struct address_space *mapping = page->mapping; | |
1088 | int ret = 0; | |
1089 | struct writeback_control wbc = { | |
1090 | .sync_mode = WB_SYNC_ALL, | |
1091 | .nr_to_write = 1, | |
1092 | }; | |
1093 | ||
1094 | BUG_ON(!PageLocked(page)); | |
1095 | ||
1096 | if (wait) | |
1097 | wait_on_page_writeback(page); | |
1098 | ||
1099 | if (clear_page_dirty_for_io(page)) { | |
1100 | page_cache_get(page); | |
1101 | ret = mapping->a_ops->writepage(page, &wbc); | |
1102 | if (ret == 0 && wait) { | |
1103 | wait_on_page_writeback(page); | |
1104 | if (PageError(page)) | |
1105 | ret = -EIO; | |
1106 | } | |
1107 | page_cache_release(page); | |
1108 | } else { | |
1109 | unlock_page(page); | |
1110 | } | |
1111 | return ret; | |
1112 | } | |
1113 | EXPORT_SYMBOL(write_one_page); | |
1114 | ||
76719325 KC |
1115 | /* |
1116 | * For address_spaces which do not use buffers nor write back. | |
1117 | */ | |
1118 | int __set_page_dirty_no_writeback(struct page *page) | |
1119 | { | |
1120 | if (!PageDirty(page)) | |
c3f0da63 | 1121 | return !TestSetPageDirty(page); |
76719325 KC |
1122 | return 0; |
1123 | } | |
1124 | ||
e3a7cca1 ES |
1125 | /* |
1126 | * Helper function for set_page_dirty family. | |
1127 | * NOTE: This relies on being atomic wrt interrupts. | |
1128 | */ | |
1129 | void account_page_dirtied(struct page *page, struct address_space *mapping) | |
1130 | { | |
1131 | if (mapping_cap_account_dirty(mapping)) { | |
1132 | __inc_zone_page_state(page, NR_FILE_DIRTY); | |
ea941f0e | 1133 | __inc_zone_page_state(page, NR_DIRTIED); |
e3a7cca1 ES |
1134 | __inc_bdi_stat(mapping->backing_dev_info, BDI_RECLAIMABLE); |
1135 | task_dirty_inc(current); | |
1136 | task_io_account_write(PAGE_CACHE_SIZE); | |
1137 | } | |
1138 | } | |
679ceace | 1139 | EXPORT_SYMBOL(account_page_dirtied); |
e3a7cca1 | 1140 | |
f629d1c9 MR |
1141 | /* |
1142 | * Helper function for set_page_writeback family. | |
1143 | * NOTE: Unlike account_page_dirtied this does not rely on being atomic | |
1144 | * wrt interrupts. | |
1145 | */ | |
1146 | void account_page_writeback(struct page *page) | |
1147 | { | |
1148 | inc_zone_page_state(page, NR_WRITEBACK); | |
ea941f0e | 1149 | inc_zone_page_state(page, NR_WRITTEN); |
f629d1c9 MR |
1150 | } |
1151 | EXPORT_SYMBOL(account_page_writeback); | |
1152 | ||
1da177e4 LT |
1153 | /* |
1154 | * For address_spaces which do not use buffers. Just tag the page as dirty in | |
1155 | * its radix tree. | |
1156 | * | |
1157 | * This is also used when a single buffer is being dirtied: we want to set the | |
1158 | * page dirty in that case, but not all the buffers. This is a "bottom-up" | |
1159 | * dirtying, whereas __set_page_dirty_buffers() is a "top-down" dirtying. | |
1160 | * | |
1161 | * Most callers have locked the page, which pins the address_space in memory. | |
1162 | * But zap_pte_range() does not lock the page, however in that case the | |
1163 | * mapping is pinned by the vma's ->vm_file reference. | |
1164 | * | |
1165 | * We take care to handle the case where the page was truncated from the | |
183ff22b | 1166 | * mapping by re-checking page_mapping() inside tree_lock. |
1da177e4 LT |
1167 | */ |
1168 | int __set_page_dirty_nobuffers(struct page *page) | |
1169 | { | |
1da177e4 LT |
1170 | if (!TestSetPageDirty(page)) { |
1171 | struct address_space *mapping = page_mapping(page); | |
1172 | struct address_space *mapping2; | |
1173 | ||
8c08540f AM |
1174 | if (!mapping) |
1175 | return 1; | |
1176 | ||
19fd6231 | 1177 | spin_lock_irq(&mapping->tree_lock); |
8c08540f AM |
1178 | mapping2 = page_mapping(page); |
1179 | if (mapping2) { /* Race with truncate? */ | |
1180 | BUG_ON(mapping2 != mapping); | |
787d2214 | 1181 | WARN_ON_ONCE(!PagePrivate(page) && !PageUptodate(page)); |
e3a7cca1 | 1182 | account_page_dirtied(page, mapping); |
8c08540f AM |
1183 | radix_tree_tag_set(&mapping->page_tree, |
1184 | page_index(page), PAGECACHE_TAG_DIRTY); | |
1185 | } | |
19fd6231 | 1186 | spin_unlock_irq(&mapping->tree_lock); |
8c08540f AM |
1187 | if (mapping->host) { |
1188 | /* !PageAnon && !swapper_space */ | |
1189 | __mark_inode_dirty(mapping->host, I_DIRTY_PAGES); | |
1da177e4 | 1190 | } |
4741c9fd | 1191 | return 1; |
1da177e4 | 1192 | } |
4741c9fd | 1193 | return 0; |
1da177e4 LT |
1194 | } |
1195 | EXPORT_SYMBOL(__set_page_dirty_nobuffers); | |
1196 | ||
1197 | /* | |
1198 | * When a writepage implementation decides that it doesn't want to write this | |
1199 | * page for some reason, it should redirty the locked page via | |
1200 | * redirty_page_for_writepage() and it should then unlock the page and return 0 | |
1201 | */ | |
1202 | int redirty_page_for_writepage(struct writeback_control *wbc, struct page *page) | |
1203 | { | |
1204 | wbc->pages_skipped++; | |
1205 | return __set_page_dirty_nobuffers(page); | |
1206 | } | |
1207 | EXPORT_SYMBOL(redirty_page_for_writepage); | |
1208 | ||
1209 | /* | |
6746aff7 WF |
1210 | * Dirty a page. |
1211 | * | |
1212 | * For pages with a mapping this should be done under the page lock | |
1213 | * for the benefit of asynchronous memory errors who prefer a consistent | |
1214 | * dirty state. This rule can be broken in some special cases, | |
1215 | * but should be better not to. | |
1216 | * | |
1da177e4 LT |
1217 | * If the mapping doesn't provide a set_page_dirty a_op, then |
1218 | * just fall through and assume that it wants buffer_heads. | |
1219 | */ | |
1cf6e7d8 | 1220 | int set_page_dirty(struct page *page) |
1da177e4 LT |
1221 | { |
1222 | struct address_space *mapping = page_mapping(page); | |
1223 | ||
1224 | if (likely(mapping)) { | |
1225 | int (*spd)(struct page *) = mapping->a_ops->set_page_dirty; | |
278df9f4 MK |
1226 | /* |
1227 | * readahead/lru_deactivate_page could remain | |
1228 | * PG_readahead/PG_reclaim due to race with end_page_writeback | |
1229 | * About readahead, if the page is written, the flags would be | |
1230 | * reset. So no problem. | |
1231 | * About lru_deactivate_page, if the page is redirty, the flag | |
1232 | * will be reset. So no problem. but if the page is used by readahead | |
1233 | * it will confuse readahead and make it restart the size rampup | |
1234 | * process. But it's a trivial problem. | |
1235 | */ | |
1236 | ClearPageReclaim(page); | |
9361401e DH |
1237 | #ifdef CONFIG_BLOCK |
1238 | if (!spd) | |
1239 | spd = __set_page_dirty_buffers; | |
1240 | #endif | |
1241 | return (*spd)(page); | |
1da177e4 | 1242 | } |
4741c9fd AM |
1243 | if (!PageDirty(page)) { |
1244 | if (!TestSetPageDirty(page)) | |
1245 | return 1; | |
1246 | } | |
1da177e4 LT |
1247 | return 0; |
1248 | } | |
1249 | EXPORT_SYMBOL(set_page_dirty); | |
1250 | ||
1251 | /* | |
1252 | * set_page_dirty() is racy if the caller has no reference against | |
1253 | * page->mapping->host, and if the page is unlocked. This is because another | |
1254 | * CPU could truncate the page off the mapping and then free the mapping. | |
1255 | * | |
1256 | * Usually, the page _is_ locked, or the caller is a user-space process which | |
1257 | * holds a reference on the inode by having an open file. | |
1258 | * | |
1259 | * In other cases, the page should be locked before running set_page_dirty(). | |
1260 | */ | |
1261 | int set_page_dirty_lock(struct page *page) | |
1262 | { | |
1263 | int ret; | |
1264 | ||
7eaceacc | 1265 | lock_page(page); |
1da177e4 LT |
1266 | ret = set_page_dirty(page); |
1267 | unlock_page(page); | |
1268 | return ret; | |
1269 | } | |
1270 | EXPORT_SYMBOL(set_page_dirty_lock); | |
1271 | ||
1da177e4 LT |
1272 | /* |
1273 | * Clear a page's dirty flag, while caring for dirty memory accounting. | |
1274 | * Returns true if the page was previously dirty. | |
1275 | * | |
1276 | * This is for preparing to put the page under writeout. We leave the page | |
1277 | * tagged as dirty in the radix tree so that a concurrent write-for-sync | |
1278 | * can discover it via a PAGECACHE_TAG_DIRTY walk. The ->writepage | |
1279 | * implementation will run either set_page_writeback() or set_page_dirty(), | |
1280 | * at which stage we bring the page's dirty flag and radix-tree dirty tag | |
1281 | * back into sync. | |
1282 | * | |
1283 | * This incoherency between the page's dirty flag and radix-tree tag is | |
1284 | * unfortunate, but it only exists while the page is locked. | |
1285 | */ | |
1286 | int clear_page_dirty_for_io(struct page *page) | |
1287 | { | |
1288 | struct address_space *mapping = page_mapping(page); | |
1289 | ||
79352894 NP |
1290 | BUG_ON(!PageLocked(page)); |
1291 | ||
7658cc28 LT |
1292 | if (mapping && mapping_cap_account_dirty(mapping)) { |
1293 | /* | |
1294 | * Yes, Virginia, this is indeed insane. | |
1295 | * | |
1296 | * We use this sequence to make sure that | |
1297 | * (a) we account for dirty stats properly | |
1298 | * (b) we tell the low-level filesystem to | |
1299 | * mark the whole page dirty if it was | |
1300 | * dirty in a pagetable. Only to then | |
1301 | * (c) clean the page again and return 1 to | |
1302 | * cause the writeback. | |
1303 | * | |
1304 | * This way we avoid all nasty races with the | |
1305 | * dirty bit in multiple places and clearing | |
1306 | * them concurrently from different threads. | |
1307 | * | |
1308 | * Note! Normally the "set_page_dirty(page)" | |
1309 | * has no effect on the actual dirty bit - since | |
1310 | * that will already usually be set. But we | |
1311 | * need the side effects, and it can help us | |
1312 | * avoid races. | |
1313 | * | |
1314 | * We basically use the page "master dirty bit" | |
1315 | * as a serialization point for all the different | |
1316 | * threads doing their things. | |
7658cc28 LT |
1317 | */ |
1318 | if (page_mkclean(page)) | |
1319 | set_page_dirty(page); | |
79352894 NP |
1320 | /* |
1321 | * We carefully synchronise fault handlers against | |
1322 | * installing a dirty pte and marking the page dirty | |
1323 | * at this point. We do this by having them hold the | |
1324 | * page lock at some point after installing their | |
1325 | * pte, but before marking the page dirty. | |
1326 | * Pages are always locked coming in here, so we get | |
1327 | * the desired exclusion. See mm/memory.c:do_wp_page() | |
1328 | * for more comments. | |
1329 | */ | |
7658cc28 | 1330 | if (TestClearPageDirty(page)) { |
8c08540f | 1331 | dec_zone_page_state(page, NR_FILE_DIRTY); |
c9e51e41 PZ |
1332 | dec_bdi_stat(mapping->backing_dev_info, |
1333 | BDI_RECLAIMABLE); | |
7658cc28 | 1334 | return 1; |
1da177e4 | 1335 | } |
7658cc28 | 1336 | return 0; |
1da177e4 | 1337 | } |
7658cc28 | 1338 | return TestClearPageDirty(page); |
1da177e4 | 1339 | } |
58bb01a9 | 1340 | EXPORT_SYMBOL(clear_page_dirty_for_io); |
1da177e4 LT |
1341 | |
1342 | int test_clear_page_writeback(struct page *page) | |
1343 | { | |
1344 | struct address_space *mapping = page_mapping(page); | |
1345 | int ret; | |
1346 | ||
1347 | if (mapping) { | |
69cb51d1 | 1348 | struct backing_dev_info *bdi = mapping->backing_dev_info; |
1da177e4 LT |
1349 | unsigned long flags; |
1350 | ||
19fd6231 | 1351 | spin_lock_irqsave(&mapping->tree_lock, flags); |
1da177e4 | 1352 | ret = TestClearPageWriteback(page); |
69cb51d1 | 1353 | if (ret) { |
1da177e4 LT |
1354 | radix_tree_tag_clear(&mapping->page_tree, |
1355 | page_index(page), | |
1356 | PAGECACHE_TAG_WRITEBACK); | |
e4ad08fe | 1357 | if (bdi_cap_account_writeback(bdi)) { |
69cb51d1 | 1358 | __dec_bdi_stat(bdi, BDI_WRITEBACK); |
04fbfdc1 PZ |
1359 | __bdi_writeout_inc(bdi); |
1360 | } | |
69cb51d1 | 1361 | } |
19fd6231 | 1362 | spin_unlock_irqrestore(&mapping->tree_lock, flags); |
1da177e4 LT |
1363 | } else { |
1364 | ret = TestClearPageWriteback(page); | |
1365 | } | |
d688abf5 AM |
1366 | if (ret) |
1367 | dec_zone_page_state(page, NR_WRITEBACK); | |
1da177e4 LT |
1368 | return ret; |
1369 | } | |
1370 | ||
1371 | int test_set_page_writeback(struct page *page) | |
1372 | { | |
1373 | struct address_space *mapping = page_mapping(page); | |
1374 | int ret; | |
1375 | ||
1376 | if (mapping) { | |
69cb51d1 | 1377 | struct backing_dev_info *bdi = mapping->backing_dev_info; |
1da177e4 LT |
1378 | unsigned long flags; |
1379 | ||
19fd6231 | 1380 | spin_lock_irqsave(&mapping->tree_lock, flags); |
1da177e4 | 1381 | ret = TestSetPageWriteback(page); |
69cb51d1 | 1382 | if (!ret) { |
1da177e4 LT |
1383 | radix_tree_tag_set(&mapping->page_tree, |
1384 | page_index(page), | |
1385 | PAGECACHE_TAG_WRITEBACK); | |
e4ad08fe | 1386 | if (bdi_cap_account_writeback(bdi)) |
69cb51d1 PZ |
1387 | __inc_bdi_stat(bdi, BDI_WRITEBACK); |
1388 | } | |
1da177e4 LT |
1389 | if (!PageDirty(page)) |
1390 | radix_tree_tag_clear(&mapping->page_tree, | |
1391 | page_index(page), | |
1392 | PAGECACHE_TAG_DIRTY); | |
f446daae JK |
1393 | radix_tree_tag_clear(&mapping->page_tree, |
1394 | page_index(page), | |
1395 | PAGECACHE_TAG_TOWRITE); | |
19fd6231 | 1396 | spin_unlock_irqrestore(&mapping->tree_lock, flags); |
1da177e4 LT |
1397 | } else { |
1398 | ret = TestSetPageWriteback(page); | |
1399 | } | |
d688abf5 | 1400 | if (!ret) |
f629d1c9 | 1401 | account_page_writeback(page); |
1da177e4 LT |
1402 | return ret; |
1403 | ||
1404 | } | |
1405 | EXPORT_SYMBOL(test_set_page_writeback); | |
1406 | ||
1407 | /* | |
00128188 | 1408 | * Return true if any of the pages in the mapping are marked with the |
1da177e4 LT |
1409 | * passed tag. |
1410 | */ | |
1411 | int mapping_tagged(struct address_space *mapping, int tag) | |
1412 | { | |
1da177e4 | 1413 | int ret; |
00128188 | 1414 | rcu_read_lock(); |
1da177e4 | 1415 | ret = radix_tree_tagged(&mapping->page_tree, tag); |
00128188 | 1416 | rcu_read_unlock(); |
1da177e4 LT |
1417 | return ret; |
1418 | } | |
1419 | EXPORT_SYMBOL(mapping_tagged); |