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> |
1da177e4 | 37 | |
1da177e4 LT |
38 | /* |
39 | * After a CPU has dirtied this many pages, balance_dirty_pages_ratelimited | |
40 | * will look to see if it needs to force writeback or throttling. | |
41 | */ | |
42 | static long ratelimit_pages = 32; | |
43 | ||
1da177e4 LT |
44 | /* |
45 | * When balance_dirty_pages decides that the caller needs to perform some | |
46 | * non-background writeback, this is how many pages it will attempt to write. | |
3a2e9a5a | 47 | * It should be somewhat larger than dirtied pages to ensure that reasonably |
1da177e4 LT |
48 | * large amounts of I/O are submitted. |
49 | */ | |
3a2e9a5a | 50 | static inline long sync_writeback_pages(unsigned long dirtied) |
1da177e4 | 51 | { |
3a2e9a5a WF |
52 | if (dirtied < ratelimit_pages) |
53 | dirtied = ratelimit_pages; | |
54 | ||
55 | return dirtied + dirtied / 2; | |
1da177e4 LT |
56 | } |
57 | ||
58 | /* The following parameters are exported via /proc/sys/vm */ | |
59 | ||
60 | /* | |
5b0830cb | 61 | * Start background writeback (via writeback threads) at this percentage |
1da177e4 | 62 | */ |
1b5e62b4 | 63 | int dirty_background_ratio = 10; |
1da177e4 | 64 | |
2da02997 DR |
65 | /* |
66 | * dirty_background_bytes starts at 0 (disabled) so that it is a function of | |
67 | * dirty_background_ratio * the amount of dirtyable memory | |
68 | */ | |
69 | unsigned long dirty_background_bytes; | |
70 | ||
195cf453 BG |
71 | /* |
72 | * free highmem will not be subtracted from the total free memory | |
73 | * for calculating free ratios if vm_highmem_is_dirtyable is true | |
74 | */ | |
75 | int vm_highmem_is_dirtyable; | |
76 | ||
1da177e4 LT |
77 | /* |
78 | * The generator of dirty data starts writeback at this percentage | |
79 | */ | |
1b5e62b4 | 80 | int vm_dirty_ratio = 20; |
1da177e4 | 81 | |
2da02997 DR |
82 | /* |
83 | * vm_dirty_bytes starts at 0 (disabled) so that it is a function of | |
84 | * vm_dirty_ratio * the amount of dirtyable memory | |
85 | */ | |
86 | unsigned long vm_dirty_bytes; | |
87 | ||
1da177e4 | 88 | /* |
704503d8 | 89 | * The interval between `kupdate'-style writebacks |
1da177e4 | 90 | */ |
22ef37ee | 91 | unsigned int dirty_writeback_interval = 5 * 100; /* centiseconds */ |
1da177e4 LT |
92 | |
93 | /* | |
704503d8 | 94 | * The longest time for which data is allowed to remain dirty |
1da177e4 | 95 | */ |
22ef37ee | 96 | unsigned int dirty_expire_interval = 30 * 100; /* centiseconds */ |
1da177e4 LT |
97 | |
98 | /* | |
99 | * Flag that makes the machine dump writes/reads and block dirtyings. | |
100 | */ | |
101 | int block_dump; | |
102 | ||
103 | /* | |
ed5b43f1 BS |
104 | * Flag that puts the machine in "laptop mode". Doubles as a timeout in jiffies: |
105 | * a full sync is triggered after this time elapses without any disk activity. | |
1da177e4 LT |
106 | */ |
107 | int laptop_mode; | |
108 | ||
109 | EXPORT_SYMBOL(laptop_mode); | |
110 | ||
111 | /* End of sysctl-exported parameters */ | |
112 | ||
113 | ||
04fbfdc1 PZ |
114 | /* |
115 | * Scale the writeback cache size proportional to the relative writeout speeds. | |
116 | * | |
117 | * We do this by keeping a floating proportion between BDIs, based on page | |
118 | * writeback completions [end_page_writeback()]. Those devices that write out | |
119 | * pages fastest will get the larger share, while the slower will get a smaller | |
120 | * share. | |
121 | * | |
122 | * We use page writeout completions because we are interested in getting rid of | |
123 | * dirty pages. Having them written out is the primary goal. | |
124 | * | |
125 | * We introduce a concept of time, a period over which we measure these events, | |
126 | * because demand can/will vary over time. The length of this period itself is | |
127 | * measured in page writeback completions. | |
128 | * | |
129 | */ | |
130 | static struct prop_descriptor vm_completions; | |
3e26c149 | 131 | static struct prop_descriptor vm_dirties; |
04fbfdc1 | 132 | |
04fbfdc1 PZ |
133 | /* |
134 | * couple the period to the dirty_ratio: | |
135 | * | |
136 | * period/2 ~ roundup_pow_of_two(dirty limit) | |
137 | */ | |
138 | static int calc_period_shift(void) | |
139 | { | |
140 | unsigned long dirty_total; | |
141 | ||
2da02997 DR |
142 | if (vm_dirty_bytes) |
143 | dirty_total = vm_dirty_bytes / PAGE_SIZE; | |
144 | else | |
145 | dirty_total = (vm_dirty_ratio * determine_dirtyable_memory()) / | |
146 | 100; | |
04fbfdc1 PZ |
147 | return 2 + ilog2(dirty_total - 1); |
148 | } | |
149 | ||
150 | /* | |
2da02997 | 151 | * update the period when the dirty threshold changes. |
04fbfdc1 | 152 | */ |
2da02997 DR |
153 | static void update_completion_period(void) |
154 | { | |
155 | int shift = calc_period_shift(); | |
156 | prop_change_shift(&vm_completions, shift); | |
157 | prop_change_shift(&vm_dirties, shift); | |
158 | } | |
159 | ||
160 | int dirty_background_ratio_handler(struct ctl_table *table, int write, | |
8d65af78 | 161 | void __user *buffer, size_t *lenp, |
2da02997 DR |
162 | loff_t *ppos) |
163 | { | |
164 | int ret; | |
165 | ||
8d65af78 | 166 | ret = proc_dointvec_minmax(table, write, buffer, lenp, ppos); |
2da02997 DR |
167 | if (ret == 0 && write) |
168 | dirty_background_bytes = 0; | |
169 | return ret; | |
170 | } | |
171 | ||
172 | int dirty_background_bytes_handler(struct ctl_table *table, int write, | |
8d65af78 | 173 | void __user *buffer, size_t *lenp, |
2da02997 DR |
174 | loff_t *ppos) |
175 | { | |
176 | int ret; | |
177 | ||
8d65af78 | 178 | ret = proc_doulongvec_minmax(table, write, buffer, lenp, ppos); |
2da02997 DR |
179 | if (ret == 0 && write) |
180 | dirty_background_ratio = 0; | |
181 | return ret; | |
182 | } | |
183 | ||
04fbfdc1 | 184 | int dirty_ratio_handler(struct ctl_table *table, int write, |
8d65af78 | 185 | void __user *buffer, size_t *lenp, |
04fbfdc1 PZ |
186 | loff_t *ppos) |
187 | { | |
188 | int old_ratio = vm_dirty_ratio; | |
2da02997 DR |
189 | int ret; |
190 | ||
8d65af78 | 191 | ret = proc_dointvec_minmax(table, write, buffer, lenp, ppos); |
04fbfdc1 | 192 | if (ret == 0 && write && vm_dirty_ratio != old_ratio) { |
2da02997 DR |
193 | update_completion_period(); |
194 | vm_dirty_bytes = 0; | |
195 | } | |
196 | return ret; | |
197 | } | |
198 | ||
199 | ||
200 | int dirty_bytes_handler(struct ctl_table *table, int write, | |
8d65af78 | 201 | void __user *buffer, size_t *lenp, |
2da02997 DR |
202 | loff_t *ppos) |
203 | { | |
fc3501d4 | 204 | unsigned long old_bytes = vm_dirty_bytes; |
2da02997 DR |
205 | int ret; |
206 | ||
8d65af78 | 207 | ret = proc_doulongvec_minmax(table, write, buffer, lenp, ppos); |
2da02997 DR |
208 | if (ret == 0 && write && vm_dirty_bytes != old_bytes) { |
209 | update_completion_period(); | |
210 | vm_dirty_ratio = 0; | |
04fbfdc1 PZ |
211 | } |
212 | return ret; | |
213 | } | |
214 | ||
215 | /* | |
216 | * Increment the BDI's writeout completion count and the global writeout | |
217 | * completion count. Called from test_clear_page_writeback(). | |
218 | */ | |
219 | static inline void __bdi_writeout_inc(struct backing_dev_info *bdi) | |
220 | { | |
a42dde04 PZ |
221 | __prop_inc_percpu_max(&vm_completions, &bdi->completions, |
222 | bdi->max_prop_frac); | |
04fbfdc1 PZ |
223 | } |
224 | ||
dd5656e5 MS |
225 | void bdi_writeout_inc(struct backing_dev_info *bdi) |
226 | { | |
227 | unsigned long flags; | |
228 | ||
229 | local_irq_save(flags); | |
230 | __bdi_writeout_inc(bdi); | |
231 | local_irq_restore(flags); | |
232 | } | |
233 | EXPORT_SYMBOL_GPL(bdi_writeout_inc); | |
234 | ||
1cf6e7d8 | 235 | void task_dirty_inc(struct task_struct *tsk) |
3e26c149 PZ |
236 | { |
237 | prop_inc_single(&vm_dirties, &tsk->dirties); | |
238 | } | |
239 | ||
04fbfdc1 PZ |
240 | /* |
241 | * Obtain an accurate fraction of the BDI's portion. | |
242 | */ | |
243 | static void bdi_writeout_fraction(struct backing_dev_info *bdi, | |
244 | long *numerator, long *denominator) | |
245 | { | |
246 | if (bdi_cap_writeback_dirty(bdi)) { | |
247 | prop_fraction_percpu(&vm_completions, &bdi->completions, | |
248 | numerator, denominator); | |
249 | } else { | |
250 | *numerator = 0; | |
251 | *denominator = 1; | |
252 | } | |
253 | } | |
254 | ||
255 | /* | |
256 | * Clip the earned share of dirty pages to that which is actually available. | |
257 | * This avoids exceeding the total dirty_limit when the floating averages | |
258 | * fluctuate too quickly. | |
259 | */ | |
dcf975d5 HS |
260 | static void clip_bdi_dirty_limit(struct backing_dev_info *bdi, |
261 | unsigned long dirty, unsigned long *pbdi_dirty) | |
04fbfdc1 | 262 | { |
dcf975d5 | 263 | unsigned long avail_dirty; |
04fbfdc1 | 264 | |
dcf975d5 | 265 | avail_dirty = global_page_state(NR_FILE_DIRTY) + |
04fbfdc1 | 266 | global_page_state(NR_WRITEBACK) + |
fc3ba692 | 267 | global_page_state(NR_UNSTABLE_NFS) + |
dcf975d5 | 268 | global_page_state(NR_WRITEBACK_TEMP); |
04fbfdc1 | 269 | |
dcf975d5 HS |
270 | if (avail_dirty < dirty) |
271 | avail_dirty = dirty - avail_dirty; | |
272 | else | |
04fbfdc1 PZ |
273 | avail_dirty = 0; |
274 | ||
275 | avail_dirty += bdi_stat(bdi, BDI_RECLAIMABLE) + | |
276 | bdi_stat(bdi, BDI_WRITEBACK); | |
277 | ||
278 | *pbdi_dirty = min(*pbdi_dirty, avail_dirty); | |
279 | } | |
280 | ||
3e26c149 PZ |
281 | static inline void task_dirties_fraction(struct task_struct *tsk, |
282 | long *numerator, long *denominator) | |
283 | { | |
284 | prop_fraction_single(&vm_dirties, &tsk->dirties, | |
285 | numerator, denominator); | |
286 | } | |
287 | ||
288 | /* | |
289 | * scale the dirty limit | |
290 | * | |
291 | * task specific dirty limit: | |
292 | * | |
293 | * dirty -= (dirty/8) * p_{t} | |
294 | */ | |
dcf975d5 | 295 | static void task_dirty_limit(struct task_struct *tsk, unsigned long *pdirty) |
3e26c149 PZ |
296 | { |
297 | long numerator, denominator; | |
dcf975d5 | 298 | unsigned long dirty = *pdirty; |
3e26c149 PZ |
299 | u64 inv = dirty >> 3; |
300 | ||
301 | task_dirties_fraction(tsk, &numerator, &denominator); | |
302 | inv *= numerator; | |
303 | do_div(inv, denominator); | |
304 | ||
305 | dirty -= inv; | |
306 | if (dirty < *pdirty/2) | |
307 | dirty = *pdirty/2; | |
308 | ||
309 | *pdirty = dirty; | |
310 | } | |
311 | ||
189d3c4a PZ |
312 | /* |
313 | * | |
314 | */ | |
189d3c4a PZ |
315 | static unsigned int bdi_min_ratio; |
316 | ||
317 | int bdi_set_min_ratio(struct backing_dev_info *bdi, unsigned int min_ratio) | |
318 | { | |
319 | int ret = 0; | |
189d3c4a | 320 | |
cfc4ba53 | 321 | spin_lock_bh(&bdi_lock); |
a42dde04 | 322 | if (min_ratio > bdi->max_ratio) { |
189d3c4a | 323 | ret = -EINVAL; |
a42dde04 PZ |
324 | } else { |
325 | min_ratio -= bdi->min_ratio; | |
326 | if (bdi_min_ratio + min_ratio < 100) { | |
327 | bdi_min_ratio += min_ratio; | |
328 | bdi->min_ratio += min_ratio; | |
329 | } else { | |
330 | ret = -EINVAL; | |
331 | } | |
332 | } | |
cfc4ba53 | 333 | spin_unlock_bh(&bdi_lock); |
a42dde04 PZ |
334 | |
335 | return ret; | |
336 | } | |
337 | ||
338 | int bdi_set_max_ratio(struct backing_dev_info *bdi, unsigned max_ratio) | |
339 | { | |
a42dde04 PZ |
340 | int ret = 0; |
341 | ||
342 | if (max_ratio > 100) | |
343 | return -EINVAL; | |
344 | ||
cfc4ba53 | 345 | spin_lock_bh(&bdi_lock); |
a42dde04 PZ |
346 | if (bdi->min_ratio > max_ratio) { |
347 | ret = -EINVAL; | |
348 | } else { | |
349 | bdi->max_ratio = max_ratio; | |
350 | bdi->max_prop_frac = (PROP_FRAC_BASE * max_ratio) / 100; | |
351 | } | |
cfc4ba53 | 352 | spin_unlock_bh(&bdi_lock); |
189d3c4a PZ |
353 | |
354 | return ret; | |
355 | } | |
a42dde04 | 356 | EXPORT_SYMBOL(bdi_set_max_ratio); |
189d3c4a | 357 | |
1da177e4 LT |
358 | /* |
359 | * Work out the current dirty-memory clamping and background writeout | |
360 | * thresholds. | |
361 | * | |
362 | * The main aim here is to lower them aggressively if there is a lot of mapped | |
363 | * memory around. To avoid stressing page reclaim with lots of unreclaimable | |
364 | * pages. It is better to clamp down on writers than to start swapping, and | |
365 | * performing lots of scanning. | |
366 | * | |
367 | * We only allow 1/2 of the currently-unmapped memory to be dirtied. | |
368 | * | |
369 | * We don't permit the clamping level to fall below 5% - that is getting rather | |
370 | * excessive. | |
371 | * | |
372 | * We make sure that the background writeout level is below the adjusted | |
373 | * clamping level. | |
374 | */ | |
1b424464 CL |
375 | |
376 | static unsigned long highmem_dirtyable_memory(unsigned long total) | |
377 | { | |
378 | #ifdef CONFIG_HIGHMEM | |
379 | int node; | |
380 | unsigned long x = 0; | |
381 | ||
37b07e41 | 382 | for_each_node_state(node, N_HIGH_MEMORY) { |
1b424464 CL |
383 | struct zone *z = |
384 | &NODE_DATA(node)->node_zones[ZONE_HIGHMEM]; | |
385 | ||
adea02a1 WF |
386 | x += zone_page_state(z, NR_FREE_PAGES) + |
387 | zone_reclaimable_pages(z); | |
1b424464 CL |
388 | } |
389 | /* | |
390 | * Make sure that the number of highmem pages is never larger | |
391 | * than the number of the total dirtyable memory. This can only | |
392 | * occur in very strange VM situations but we want to make sure | |
393 | * that this does not occur. | |
394 | */ | |
395 | return min(x, total); | |
396 | #else | |
397 | return 0; | |
398 | #endif | |
399 | } | |
400 | ||
3eefae99 SR |
401 | /** |
402 | * determine_dirtyable_memory - amount of memory that may be used | |
403 | * | |
404 | * Returns the numebr of pages that can currently be freed and used | |
405 | * by the kernel for direct mappings. | |
406 | */ | |
407 | unsigned long determine_dirtyable_memory(void) | |
1b424464 CL |
408 | { |
409 | unsigned long x; | |
410 | ||
adea02a1 | 411 | x = global_page_state(NR_FREE_PAGES) + global_reclaimable_pages(); |
195cf453 BG |
412 | |
413 | if (!vm_highmem_is_dirtyable) | |
414 | x -= highmem_dirtyable_memory(x); | |
415 | ||
1b424464 CL |
416 | return x + 1; /* Ensure that we never return 0 */ |
417 | } | |
418 | ||
cf0ca9fe | 419 | void |
364aeb28 DR |
420 | get_dirty_limits(unsigned long *pbackground, unsigned long *pdirty, |
421 | unsigned long *pbdi_dirty, struct backing_dev_info *bdi) | |
1da177e4 | 422 | { |
364aeb28 DR |
423 | unsigned long background; |
424 | unsigned long dirty; | |
1b424464 | 425 | unsigned long available_memory = determine_dirtyable_memory(); |
1da177e4 LT |
426 | struct task_struct *tsk; |
427 | ||
2da02997 DR |
428 | if (vm_dirty_bytes) |
429 | dirty = DIV_ROUND_UP(vm_dirty_bytes, PAGE_SIZE); | |
430 | else { | |
431 | int dirty_ratio; | |
432 | ||
433 | dirty_ratio = vm_dirty_ratio; | |
434 | if (dirty_ratio < 5) | |
435 | dirty_ratio = 5; | |
436 | dirty = (dirty_ratio * available_memory) / 100; | |
437 | } | |
1da177e4 | 438 | |
2da02997 DR |
439 | if (dirty_background_bytes) |
440 | background = DIV_ROUND_UP(dirty_background_bytes, PAGE_SIZE); | |
441 | else | |
442 | background = (dirty_background_ratio * available_memory) / 100; | |
1da177e4 | 443 | |
2da02997 DR |
444 | if (background >= dirty) |
445 | background = dirty / 2; | |
1da177e4 LT |
446 | tsk = current; |
447 | if (tsk->flags & PF_LESS_THROTTLE || rt_task(tsk)) { | |
448 | background += background / 4; | |
449 | dirty += dirty / 4; | |
450 | } | |
451 | *pbackground = background; | |
452 | *pdirty = dirty; | |
04fbfdc1 PZ |
453 | |
454 | if (bdi) { | |
189d3c4a | 455 | u64 bdi_dirty; |
04fbfdc1 PZ |
456 | long numerator, denominator; |
457 | ||
458 | /* | |
459 | * Calculate this BDI's share of the dirty ratio. | |
460 | */ | |
461 | bdi_writeout_fraction(bdi, &numerator, &denominator); | |
462 | ||
189d3c4a | 463 | bdi_dirty = (dirty * (100 - bdi_min_ratio)) / 100; |
04fbfdc1 PZ |
464 | bdi_dirty *= numerator; |
465 | do_div(bdi_dirty, denominator); | |
189d3c4a | 466 | bdi_dirty += (dirty * bdi->min_ratio) / 100; |
a42dde04 PZ |
467 | if (bdi_dirty > (dirty * bdi->max_ratio) / 100) |
468 | bdi_dirty = dirty * bdi->max_ratio / 100; | |
04fbfdc1 PZ |
469 | |
470 | *pbdi_dirty = bdi_dirty; | |
471 | clip_bdi_dirty_limit(bdi, dirty, pbdi_dirty); | |
3e26c149 | 472 | task_dirty_limit(current, pbdi_dirty); |
04fbfdc1 | 473 | } |
1da177e4 LT |
474 | } |
475 | ||
476 | /* | |
477 | * balance_dirty_pages() must be called by processes which are generating dirty | |
478 | * data. It looks at the number of dirty pages in the machine and will force | |
479 | * the caller to perform writeback if the system is over `vm_dirty_ratio'. | |
5b0830cb JA |
480 | * If we're over `background_thresh' then the writeback threads are woken to |
481 | * perform some writeout. | |
1da177e4 | 482 | */ |
3a2e9a5a WF |
483 | static void balance_dirty_pages(struct address_space *mapping, |
484 | unsigned long write_chunk) | |
1da177e4 | 485 | { |
5fce25a9 PZ |
486 | long nr_reclaimable, bdi_nr_reclaimable; |
487 | long nr_writeback, bdi_nr_writeback; | |
364aeb28 DR |
488 | unsigned long background_thresh; |
489 | unsigned long dirty_thresh; | |
490 | unsigned long bdi_thresh; | |
1da177e4 | 491 | unsigned long pages_written = 0; |
87c6a9b2 | 492 | unsigned long pause = 1; |
1da177e4 LT |
493 | |
494 | struct backing_dev_info *bdi = mapping->backing_dev_info; | |
495 | ||
496 | for (;;) { | |
497 | struct writeback_control wbc = { | |
498 | .bdi = bdi, | |
499 | .sync_mode = WB_SYNC_NONE, | |
500 | .older_than_this = NULL, | |
501 | .nr_to_write = write_chunk, | |
111ebb6e | 502 | .range_cyclic = 1, |
1da177e4 LT |
503 | }; |
504 | ||
04fbfdc1 PZ |
505 | get_dirty_limits(&background_thresh, &dirty_thresh, |
506 | &bdi_thresh, bdi); | |
5fce25a9 PZ |
507 | |
508 | nr_reclaimable = global_page_state(NR_FILE_DIRTY) + | |
509 | global_page_state(NR_UNSTABLE_NFS); | |
510 | nr_writeback = global_page_state(NR_WRITEBACK); | |
511 | ||
04fbfdc1 PZ |
512 | bdi_nr_reclaimable = bdi_stat(bdi, BDI_RECLAIMABLE); |
513 | bdi_nr_writeback = bdi_stat(bdi, BDI_WRITEBACK); | |
5fce25a9 | 514 | |
04fbfdc1 PZ |
515 | if (bdi_nr_reclaimable + bdi_nr_writeback <= bdi_thresh) |
516 | break; | |
1da177e4 | 517 | |
5fce25a9 PZ |
518 | /* |
519 | * Throttle it only when the background writeback cannot | |
520 | * catch-up. This avoids (excessively) small writeouts | |
521 | * when the bdi limits are ramping up. | |
522 | */ | |
523 | if (nr_reclaimable + nr_writeback < | |
524 | (background_thresh + dirty_thresh) / 2) | |
525 | break; | |
526 | ||
04fbfdc1 PZ |
527 | if (!bdi->dirty_exceeded) |
528 | bdi->dirty_exceeded = 1; | |
1da177e4 LT |
529 | |
530 | /* Note: nr_reclaimable denotes nr_dirty + nr_unstable. | |
531 | * Unstable writes are a feature of certain networked | |
532 | * filesystems (i.e. NFS) in which data may have been | |
533 | * written to the server's write cache, but has not yet | |
534 | * been flushed to permanent storage. | |
d7831a0b RK |
535 | * Only move pages to writeback if this bdi is over its |
536 | * threshold otherwise wait until the disk writes catch | |
537 | * up. | |
1da177e4 | 538 | */ |
d7831a0b | 539 | if (bdi_nr_reclaimable > bdi_thresh) { |
03ba3782 | 540 | writeback_inodes_wbc(&wbc); |
1da177e4 | 541 | pages_written += write_chunk - wbc.nr_to_write; |
04fbfdc1 PZ |
542 | get_dirty_limits(&background_thresh, &dirty_thresh, |
543 | &bdi_thresh, bdi); | |
544 | } | |
545 | ||
546 | /* | |
547 | * In order to avoid the stacked BDI deadlock we need | |
548 | * to ensure we accurately count the 'dirty' pages when | |
549 | * the threshold is low. | |
550 | * | |
551 | * Otherwise it would be possible to get thresh+n pages | |
552 | * reported dirty, even though there are thresh-m pages | |
553 | * actually dirty; with m+n sitting in the percpu | |
554 | * deltas. | |
555 | */ | |
556 | if (bdi_thresh < 2*bdi_stat_error(bdi)) { | |
557 | bdi_nr_reclaimable = bdi_stat_sum(bdi, BDI_RECLAIMABLE); | |
558 | bdi_nr_writeback = bdi_stat_sum(bdi, BDI_WRITEBACK); | |
559 | } else if (bdi_nr_reclaimable) { | |
560 | bdi_nr_reclaimable = bdi_stat(bdi, BDI_RECLAIMABLE); | |
561 | bdi_nr_writeback = bdi_stat(bdi, BDI_WRITEBACK); | |
1da177e4 | 562 | } |
04fbfdc1 PZ |
563 | |
564 | if (bdi_nr_reclaimable + bdi_nr_writeback <= bdi_thresh) | |
565 | break; | |
566 | if (pages_written >= write_chunk) | |
567 | break; /* We've done our duty */ | |
568 | ||
87c6a9b2 JA |
569 | schedule_timeout_interruptible(pause); |
570 | ||
571 | /* | |
572 | * Increase the delay for each loop, up to our previous | |
573 | * default of taking a 100ms nap. | |
574 | */ | |
575 | pause <<= 1; | |
576 | if (pause > HZ / 10) | |
577 | pause = HZ / 10; | |
1da177e4 LT |
578 | } |
579 | ||
04fbfdc1 PZ |
580 | if (bdi_nr_reclaimable + bdi_nr_writeback < bdi_thresh && |
581 | bdi->dirty_exceeded) | |
582 | bdi->dirty_exceeded = 0; | |
1da177e4 LT |
583 | |
584 | if (writeback_in_progress(bdi)) | |
5b0830cb | 585 | return; |
1da177e4 LT |
586 | |
587 | /* | |
588 | * In laptop mode, we wait until hitting the higher threshold before | |
589 | * starting background writeout, and then write out all the way down | |
590 | * to the lower threshold. So slow writers cause minimal disk activity. | |
591 | * | |
592 | * In normal mode, we start background writeout at the lower | |
593 | * background_thresh, to keep the amount of dirty memory low. | |
594 | */ | |
595 | if ((laptop_mode && pages_written) || | |
d3ddec76 WF |
596 | (!laptop_mode && ((global_page_state(NR_FILE_DIRTY) |
597 | + global_page_state(NR_UNSTABLE_NFS)) | |
b6e51316 | 598 | > background_thresh))) |
d3ddec76 | 599 | bdi_start_writeback(bdi, 0); |
1da177e4 LT |
600 | } |
601 | ||
a200ee18 | 602 | void set_page_dirty_balance(struct page *page, int page_mkwrite) |
edc79b2a | 603 | { |
a200ee18 | 604 | if (set_page_dirty(page) || page_mkwrite) { |
edc79b2a PZ |
605 | struct address_space *mapping = page_mapping(page); |
606 | ||
607 | if (mapping) | |
608 | balance_dirty_pages_ratelimited(mapping); | |
609 | } | |
610 | } | |
611 | ||
245b2e70 TH |
612 | static DEFINE_PER_CPU(unsigned long, bdp_ratelimits) = 0; |
613 | ||
1da177e4 | 614 | /** |
fa5a734e | 615 | * balance_dirty_pages_ratelimited_nr - balance dirty memory state |
67be2dd1 | 616 | * @mapping: address_space which was dirtied |
a580290c | 617 | * @nr_pages_dirtied: number of pages which the caller has just dirtied |
1da177e4 LT |
618 | * |
619 | * Processes which are dirtying memory should call in here once for each page | |
620 | * which was newly dirtied. The function will periodically check the system's | |
621 | * dirty state and will initiate writeback if needed. | |
622 | * | |
623 | * On really big machines, get_writeback_state is expensive, so try to avoid | |
624 | * calling it too often (ratelimiting). But once we're over the dirty memory | |
625 | * limit we decrease the ratelimiting by a lot, to prevent individual processes | |
626 | * from overshooting the limit by (ratelimit_pages) each. | |
627 | */ | |
fa5a734e AM |
628 | void balance_dirty_pages_ratelimited_nr(struct address_space *mapping, |
629 | unsigned long nr_pages_dirtied) | |
1da177e4 | 630 | { |
fa5a734e AM |
631 | unsigned long ratelimit; |
632 | unsigned long *p; | |
1da177e4 LT |
633 | |
634 | ratelimit = ratelimit_pages; | |
04fbfdc1 | 635 | if (mapping->backing_dev_info->dirty_exceeded) |
1da177e4 LT |
636 | ratelimit = 8; |
637 | ||
638 | /* | |
639 | * Check the rate limiting. Also, we do not want to throttle real-time | |
640 | * tasks in balance_dirty_pages(). Period. | |
641 | */ | |
fa5a734e | 642 | preempt_disable(); |
245b2e70 | 643 | p = &__get_cpu_var(bdp_ratelimits); |
fa5a734e AM |
644 | *p += nr_pages_dirtied; |
645 | if (unlikely(*p >= ratelimit)) { | |
3a2e9a5a | 646 | ratelimit = sync_writeback_pages(*p); |
fa5a734e AM |
647 | *p = 0; |
648 | preempt_enable(); | |
3a2e9a5a | 649 | balance_dirty_pages(mapping, ratelimit); |
1da177e4 LT |
650 | return; |
651 | } | |
fa5a734e | 652 | preempt_enable(); |
1da177e4 | 653 | } |
fa5a734e | 654 | EXPORT_SYMBOL(balance_dirty_pages_ratelimited_nr); |
1da177e4 | 655 | |
232ea4d6 | 656 | void throttle_vm_writeout(gfp_t gfp_mask) |
1da177e4 | 657 | { |
364aeb28 DR |
658 | unsigned long background_thresh; |
659 | unsigned long dirty_thresh; | |
1da177e4 LT |
660 | |
661 | for ( ; ; ) { | |
04fbfdc1 | 662 | get_dirty_limits(&background_thresh, &dirty_thresh, NULL, NULL); |
1da177e4 LT |
663 | |
664 | /* | |
665 | * Boost the allowable dirty threshold a bit for page | |
666 | * allocators so they don't get DoS'ed by heavy writers | |
667 | */ | |
668 | dirty_thresh += dirty_thresh / 10; /* wheeee... */ | |
669 | ||
c24f21bd CL |
670 | if (global_page_state(NR_UNSTABLE_NFS) + |
671 | global_page_state(NR_WRITEBACK) <= dirty_thresh) | |
672 | break; | |
8aa7e847 | 673 | congestion_wait(BLK_RW_ASYNC, HZ/10); |
369f2389 FW |
674 | |
675 | /* | |
676 | * The caller might hold locks which can prevent IO completion | |
677 | * or progress in the filesystem. So we cannot just sit here | |
678 | * waiting for IO to complete. | |
679 | */ | |
680 | if ((gfp_mask & (__GFP_FS|__GFP_IO)) != (__GFP_FS|__GFP_IO)) | |
681 | break; | |
1da177e4 LT |
682 | } |
683 | } | |
684 | ||
1da177e4 LT |
685 | static void laptop_timer_fn(unsigned long unused); |
686 | ||
8d06afab | 687 | static DEFINE_TIMER(laptop_mode_wb_timer, laptop_timer_fn, 0, 0); |
1da177e4 | 688 | |
1da177e4 LT |
689 | /* |
690 | * sysctl handler for /proc/sys/vm/dirty_writeback_centisecs | |
691 | */ | |
692 | int dirty_writeback_centisecs_handler(ctl_table *table, int write, | |
8d65af78 | 693 | void __user *buffer, size_t *length, loff_t *ppos) |
1da177e4 | 694 | { |
8d65af78 | 695 | proc_dointvec(table, write, buffer, length, ppos); |
1da177e4 LT |
696 | return 0; |
697 | } | |
698 | ||
03ba3782 | 699 | static void do_laptop_sync(struct work_struct *work) |
1da177e4 | 700 | { |
03ba3782 JA |
701 | wakeup_flusher_threads(0); |
702 | kfree(work); | |
1da177e4 LT |
703 | } |
704 | ||
705 | static void laptop_timer_fn(unsigned long unused) | |
706 | { | |
03ba3782 JA |
707 | struct work_struct *work; |
708 | ||
709 | work = kmalloc(sizeof(*work), GFP_ATOMIC); | |
710 | if (work) { | |
711 | INIT_WORK(work, do_laptop_sync); | |
712 | schedule_work(work); | |
713 | } | |
1da177e4 LT |
714 | } |
715 | ||
716 | /* | |
717 | * We've spun up the disk and we're in laptop mode: schedule writeback | |
718 | * of all dirty data a few seconds from now. If the flush is already scheduled | |
719 | * then push it back - the user is still using the disk. | |
720 | */ | |
721 | void laptop_io_completion(void) | |
722 | { | |
ed5b43f1 | 723 | mod_timer(&laptop_mode_wb_timer, jiffies + laptop_mode); |
1da177e4 LT |
724 | } |
725 | ||
726 | /* | |
727 | * We're in laptop mode and we've just synced. The sync's writes will have | |
728 | * caused another writeback to be scheduled by laptop_io_completion. | |
729 | * Nothing needs to be written back anymore, so we unschedule the writeback. | |
730 | */ | |
731 | void laptop_sync_completion(void) | |
732 | { | |
733 | del_timer(&laptop_mode_wb_timer); | |
734 | } | |
735 | ||
736 | /* | |
737 | * If ratelimit_pages is too high then we can get into dirty-data overload | |
738 | * if a large number of processes all perform writes at the same time. | |
739 | * If it is too low then SMP machines will call the (expensive) | |
740 | * get_writeback_state too often. | |
741 | * | |
742 | * Here we set ratelimit_pages to a level which ensures that when all CPUs are | |
743 | * dirtying in parallel, we cannot go more than 3% (1/32) over the dirty memory | |
744 | * thresholds before writeback cuts in. | |
745 | * | |
746 | * But the limit should not be set too high. Because it also controls the | |
747 | * amount of memory which the balance_dirty_pages() caller has to write back. | |
748 | * If this is too large then the caller will block on the IO queue all the | |
749 | * time. So limit it to four megabytes - the balance_dirty_pages() caller | |
750 | * will write six megabyte chunks, max. | |
751 | */ | |
752 | ||
2d1d43f6 | 753 | void writeback_set_ratelimit(void) |
1da177e4 | 754 | { |
40c99aae | 755 | ratelimit_pages = vm_total_pages / (num_online_cpus() * 32); |
1da177e4 LT |
756 | if (ratelimit_pages < 16) |
757 | ratelimit_pages = 16; | |
758 | if (ratelimit_pages * PAGE_CACHE_SIZE > 4096 * 1024) | |
759 | ratelimit_pages = (4096 * 1024) / PAGE_CACHE_SIZE; | |
760 | } | |
761 | ||
26c2143b | 762 | static int __cpuinit |
1da177e4 LT |
763 | ratelimit_handler(struct notifier_block *self, unsigned long u, void *v) |
764 | { | |
2d1d43f6 | 765 | writeback_set_ratelimit(); |
aa0f0303 | 766 | return NOTIFY_DONE; |
1da177e4 LT |
767 | } |
768 | ||
74b85f37 | 769 | static struct notifier_block __cpuinitdata ratelimit_nb = { |
1da177e4 LT |
770 | .notifier_call = ratelimit_handler, |
771 | .next = NULL, | |
772 | }; | |
773 | ||
774 | /* | |
dc6e29da LT |
775 | * Called early on to tune the page writeback dirty limits. |
776 | * | |
777 | * We used to scale dirty pages according to how total memory | |
778 | * related to pages that could be allocated for buffers (by | |
779 | * comparing nr_free_buffer_pages() to vm_total_pages. | |
780 | * | |
781 | * However, that was when we used "dirty_ratio" to scale with | |
782 | * all memory, and we don't do that any more. "dirty_ratio" | |
783 | * is now applied to total non-HIGHPAGE memory (by subtracting | |
784 | * totalhigh_pages from vm_total_pages), and as such we can't | |
785 | * get into the old insane situation any more where we had | |
786 | * large amounts of dirty pages compared to a small amount of | |
787 | * non-HIGHMEM memory. | |
788 | * | |
789 | * But we might still want to scale the dirty_ratio by how | |
790 | * much memory the box has.. | |
1da177e4 LT |
791 | */ |
792 | void __init page_writeback_init(void) | |
793 | { | |
04fbfdc1 PZ |
794 | int shift; |
795 | ||
2d1d43f6 | 796 | writeback_set_ratelimit(); |
1da177e4 | 797 | register_cpu_notifier(&ratelimit_nb); |
04fbfdc1 PZ |
798 | |
799 | shift = calc_period_shift(); | |
800 | prop_descriptor_init(&vm_completions, shift); | |
3e26c149 | 801 | prop_descriptor_init(&vm_dirties, shift); |
1da177e4 LT |
802 | } |
803 | ||
811d736f | 804 | /** |
0ea97180 | 805 | * write_cache_pages - walk the list of dirty pages of the given address space and write all of them. |
811d736f DH |
806 | * @mapping: address space structure to write |
807 | * @wbc: subtract the number of written pages from *@wbc->nr_to_write | |
0ea97180 MS |
808 | * @writepage: function called for each page |
809 | * @data: data passed to writepage function | |
811d736f | 810 | * |
0ea97180 | 811 | * If a page is already under I/O, write_cache_pages() skips it, even |
811d736f DH |
812 | * if it's dirty. This is desirable behaviour for memory-cleaning writeback, |
813 | * but it is INCORRECT for data-integrity system calls such as fsync(). fsync() | |
814 | * and msync() need to guarantee that all the data which was dirty at the time | |
815 | * the call was made get new I/O started against them. If wbc->sync_mode is | |
816 | * WB_SYNC_ALL then we were called for data integrity and we must wait for | |
817 | * existing IO to complete. | |
811d736f | 818 | */ |
0ea97180 MS |
819 | int write_cache_pages(struct address_space *mapping, |
820 | struct writeback_control *wbc, writepage_t writepage, | |
821 | void *data) | |
811d736f DH |
822 | { |
823 | struct backing_dev_info *bdi = mapping->backing_dev_info; | |
824 | int ret = 0; | |
825 | int done = 0; | |
811d736f DH |
826 | struct pagevec pvec; |
827 | int nr_pages; | |
31a12666 | 828 | pgoff_t uninitialized_var(writeback_index); |
811d736f DH |
829 | pgoff_t index; |
830 | pgoff_t end; /* Inclusive */ | |
bd19e012 | 831 | pgoff_t done_index; |
31a12666 | 832 | int cycled; |
811d736f | 833 | int range_whole = 0; |
17bc6c30 | 834 | long nr_to_write = wbc->nr_to_write; |
811d736f DH |
835 | |
836 | if (wbc->nonblocking && bdi_write_congested(bdi)) { | |
837 | wbc->encountered_congestion = 1; | |
838 | return 0; | |
839 | } | |
840 | ||
811d736f DH |
841 | pagevec_init(&pvec, 0); |
842 | if (wbc->range_cyclic) { | |
31a12666 NP |
843 | writeback_index = mapping->writeback_index; /* prev offset */ |
844 | index = writeback_index; | |
845 | if (index == 0) | |
846 | cycled = 1; | |
847 | else | |
848 | cycled = 0; | |
811d736f DH |
849 | end = -1; |
850 | } else { | |
851 | index = wbc->range_start >> PAGE_CACHE_SHIFT; | |
852 | end = wbc->range_end >> PAGE_CACHE_SHIFT; | |
853 | if (wbc->range_start == 0 && wbc->range_end == LLONG_MAX) | |
854 | range_whole = 1; | |
31a12666 | 855 | cycled = 1; /* ignore range_cyclic tests */ |
811d736f DH |
856 | } |
857 | retry: | |
bd19e012 | 858 | done_index = index; |
5a3d5c98 NP |
859 | while (!done && (index <= end)) { |
860 | int i; | |
861 | ||
862 | nr_pages = pagevec_lookup_tag(&pvec, mapping, &index, | |
863 | PAGECACHE_TAG_DIRTY, | |
864 | min(end - index, (pgoff_t)PAGEVEC_SIZE-1) + 1); | |
865 | if (nr_pages == 0) | |
866 | break; | |
811d736f | 867 | |
811d736f DH |
868 | for (i = 0; i < nr_pages; i++) { |
869 | struct page *page = pvec.pages[i]; | |
870 | ||
871 | /* | |
d5482cdf NP |
872 | * At this point, the page may be truncated or |
873 | * invalidated (changing page->mapping to NULL), or | |
874 | * even swizzled back from swapper_space to tmpfs file | |
875 | * mapping. However, page->index will not change | |
876 | * because we have a reference on the page. | |
811d736f | 877 | */ |
d5482cdf NP |
878 | if (page->index > end) { |
879 | /* | |
880 | * can't be range_cyclic (1st pass) because | |
881 | * end == -1 in that case. | |
882 | */ | |
883 | done = 1; | |
884 | break; | |
885 | } | |
886 | ||
887 | done_index = page->index + 1; | |
888 | ||
811d736f DH |
889 | lock_page(page); |
890 | ||
5a3d5c98 NP |
891 | /* |
892 | * Page truncated or invalidated. We can freely skip it | |
893 | * then, even for data integrity operations: the page | |
894 | * has disappeared concurrently, so there could be no | |
895 | * real expectation of this data interity operation | |
896 | * even if there is now a new, dirty page at the same | |
897 | * pagecache address. | |
898 | */ | |
811d736f | 899 | if (unlikely(page->mapping != mapping)) { |
5a3d5c98 | 900 | continue_unlock: |
811d736f DH |
901 | unlock_page(page); |
902 | continue; | |
903 | } | |
904 | ||
515f4a03 NP |
905 | if (!PageDirty(page)) { |
906 | /* someone wrote it for us */ | |
907 | goto continue_unlock; | |
908 | } | |
909 | ||
910 | if (PageWriteback(page)) { | |
911 | if (wbc->sync_mode != WB_SYNC_NONE) | |
912 | wait_on_page_writeback(page); | |
913 | else | |
914 | goto continue_unlock; | |
915 | } | |
811d736f | 916 | |
515f4a03 NP |
917 | BUG_ON(PageWriteback(page)); |
918 | if (!clear_page_dirty_for_io(page)) | |
5a3d5c98 | 919 | goto continue_unlock; |
811d736f | 920 | |
0ea97180 | 921 | ret = (*writepage)(page, wbc, data); |
00266770 NP |
922 | if (unlikely(ret)) { |
923 | if (ret == AOP_WRITEPAGE_ACTIVATE) { | |
924 | unlock_page(page); | |
925 | ret = 0; | |
926 | } else { | |
927 | /* | |
928 | * done_index is set past this page, | |
929 | * so media errors will not choke | |
930 | * background writeout for the entire | |
931 | * file. This has consequences for | |
932 | * range_cyclic semantics (ie. it may | |
933 | * not be suitable for data integrity | |
934 | * writeout). | |
935 | */ | |
936 | done = 1; | |
937 | break; | |
938 | } | |
939 | } | |
940 | ||
89e12190 | 941 | if (nr_to_write > 0) { |
dcf6a79d | 942 | nr_to_write--; |
89e12190 FC |
943 | if (nr_to_write == 0 && |
944 | wbc->sync_mode == WB_SYNC_NONE) { | |
945 | /* | |
946 | * We stop writing back only if we are | |
947 | * not doing integrity sync. In case of | |
948 | * integrity sync we have to keep going | |
949 | * because someone may be concurrently | |
950 | * dirtying pages, and we might have | |
951 | * synced a lot of newly appeared dirty | |
952 | * pages, but have not synced all of the | |
953 | * old dirty pages. | |
954 | */ | |
955 | done = 1; | |
956 | break; | |
957 | } | |
05fe478d | 958 | } |
dcf6a79d | 959 | |
811d736f DH |
960 | if (wbc->nonblocking && bdi_write_congested(bdi)) { |
961 | wbc->encountered_congestion = 1; | |
962 | done = 1; | |
82fd1a9a | 963 | break; |
811d736f DH |
964 | } |
965 | } | |
966 | pagevec_release(&pvec); | |
967 | cond_resched(); | |
968 | } | |
3a4c6800 | 969 | if (!cycled && !done) { |
811d736f | 970 | /* |
31a12666 | 971 | * range_cyclic: |
811d736f DH |
972 | * We hit the last page and there is more work to be done: wrap |
973 | * back to the start of the file | |
974 | */ | |
31a12666 | 975 | cycled = 1; |
811d736f | 976 | index = 0; |
31a12666 | 977 | end = writeback_index - 1; |
811d736f DH |
978 | goto retry; |
979 | } | |
17bc6c30 AK |
980 | if (!wbc->no_nrwrite_index_update) { |
981 | if (wbc->range_cyclic || (range_whole && nr_to_write > 0)) | |
bd19e012 | 982 | mapping->writeback_index = done_index; |
17bc6c30 AK |
983 | wbc->nr_to_write = nr_to_write; |
984 | } | |
06d6cf69 | 985 | |
811d736f DH |
986 | return ret; |
987 | } | |
0ea97180 MS |
988 | EXPORT_SYMBOL(write_cache_pages); |
989 | ||
990 | /* | |
991 | * Function used by generic_writepages to call the real writepage | |
992 | * function and set the mapping flags on error | |
993 | */ | |
994 | static int __writepage(struct page *page, struct writeback_control *wbc, | |
995 | void *data) | |
996 | { | |
997 | struct address_space *mapping = data; | |
998 | int ret = mapping->a_ops->writepage(page, wbc); | |
999 | mapping_set_error(mapping, ret); | |
1000 | return ret; | |
1001 | } | |
1002 | ||
1003 | /** | |
1004 | * generic_writepages - walk the list of dirty pages of the given address space and writepage() all of them. | |
1005 | * @mapping: address space structure to write | |
1006 | * @wbc: subtract the number of written pages from *@wbc->nr_to_write | |
1007 | * | |
1008 | * This is a library function, which implements the writepages() | |
1009 | * address_space_operation. | |
1010 | */ | |
1011 | int generic_writepages(struct address_space *mapping, | |
1012 | struct writeback_control *wbc) | |
1013 | { | |
1014 | /* deal with chardevs and other special file */ | |
1015 | if (!mapping->a_ops->writepage) | |
1016 | return 0; | |
1017 | ||
1018 | return write_cache_pages(mapping, wbc, __writepage, mapping); | |
1019 | } | |
811d736f DH |
1020 | |
1021 | EXPORT_SYMBOL(generic_writepages); | |
1022 | ||
1da177e4 LT |
1023 | int do_writepages(struct address_space *mapping, struct writeback_control *wbc) |
1024 | { | |
22905f77 AM |
1025 | int ret; |
1026 | ||
1da177e4 LT |
1027 | if (wbc->nr_to_write <= 0) |
1028 | return 0; | |
1029 | if (mapping->a_ops->writepages) | |
d08b3851 | 1030 | ret = mapping->a_ops->writepages(mapping, wbc); |
22905f77 AM |
1031 | else |
1032 | ret = generic_writepages(mapping, wbc); | |
22905f77 | 1033 | return ret; |
1da177e4 LT |
1034 | } |
1035 | ||
1036 | /** | |
1037 | * write_one_page - write out a single page and optionally wait on I/O | |
67be2dd1 MW |
1038 | * @page: the page to write |
1039 | * @wait: if true, wait on writeout | |
1da177e4 LT |
1040 | * |
1041 | * The page must be locked by the caller and will be unlocked upon return. | |
1042 | * | |
1043 | * write_one_page() returns a negative error code if I/O failed. | |
1044 | */ | |
1045 | int write_one_page(struct page *page, int wait) | |
1046 | { | |
1047 | struct address_space *mapping = page->mapping; | |
1048 | int ret = 0; | |
1049 | struct writeback_control wbc = { | |
1050 | .sync_mode = WB_SYNC_ALL, | |
1051 | .nr_to_write = 1, | |
1052 | }; | |
1053 | ||
1054 | BUG_ON(!PageLocked(page)); | |
1055 | ||
1056 | if (wait) | |
1057 | wait_on_page_writeback(page); | |
1058 | ||
1059 | if (clear_page_dirty_for_io(page)) { | |
1060 | page_cache_get(page); | |
1061 | ret = mapping->a_ops->writepage(page, &wbc); | |
1062 | if (ret == 0 && wait) { | |
1063 | wait_on_page_writeback(page); | |
1064 | if (PageError(page)) | |
1065 | ret = -EIO; | |
1066 | } | |
1067 | page_cache_release(page); | |
1068 | } else { | |
1069 | unlock_page(page); | |
1070 | } | |
1071 | return ret; | |
1072 | } | |
1073 | EXPORT_SYMBOL(write_one_page); | |
1074 | ||
76719325 KC |
1075 | /* |
1076 | * For address_spaces which do not use buffers nor write back. | |
1077 | */ | |
1078 | int __set_page_dirty_no_writeback(struct page *page) | |
1079 | { | |
1080 | if (!PageDirty(page)) | |
1081 | SetPageDirty(page); | |
1082 | return 0; | |
1083 | } | |
1084 | ||
e3a7cca1 ES |
1085 | /* |
1086 | * Helper function for set_page_dirty family. | |
1087 | * NOTE: This relies on being atomic wrt interrupts. | |
1088 | */ | |
1089 | void account_page_dirtied(struct page *page, struct address_space *mapping) | |
1090 | { | |
1091 | if (mapping_cap_account_dirty(mapping)) { | |
1092 | __inc_zone_page_state(page, NR_FILE_DIRTY); | |
1093 | __inc_bdi_stat(mapping->backing_dev_info, BDI_RECLAIMABLE); | |
1094 | task_dirty_inc(current); | |
1095 | task_io_account_write(PAGE_CACHE_SIZE); | |
1096 | } | |
1097 | } | |
1098 | ||
1da177e4 LT |
1099 | /* |
1100 | * For address_spaces which do not use buffers. Just tag the page as dirty in | |
1101 | * its radix tree. | |
1102 | * | |
1103 | * This is also used when a single buffer is being dirtied: we want to set the | |
1104 | * page dirty in that case, but not all the buffers. This is a "bottom-up" | |
1105 | * dirtying, whereas __set_page_dirty_buffers() is a "top-down" dirtying. | |
1106 | * | |
1107 | * Most callers have locked the page, which pins the address_space in memory. | |
1108 | * But zap_pte_range() does not lock the page, however in that case the | |
1109 | * mapping is pinned by the vma's ->vm_file reference. | |
1110 | * | |
1111 | * We take care to handle the case where the page was truncated from the | |
183ff22b | 1112 | * mapping by re-checking page_mapping() inside tree_lock. |
1da177e4 LT |
1113 | */ |
1114 | int __set_page_dirty_nobuffers(struct page *page) | |
1115 | { | |
1da177e4 LT |
1116 | if (!TestSetPageDirty(page)) { |
1117 | struct address_space *mapping = page_mapping(page); | |
1118 | struct address_space *mapping2; | |
1119 | ||
8c08540f AM |
1120 | if (!mapping) |
1121 | return 1; | |
1122 | ||
19fd6231 | 1123 | spin_lock_irq(&mapping->tree_lock); |
8c08540f AM |
1124 | mapping2 = page_mapping(page); |
1125 | if (mapping2) { /* Race with truncate? */ | |
1126 | BUG_ON(mapping2 != mapping); | |
787d2214 | 1127 | WARN_ON_ONCE(!PagePrivate(page) && !PageUptodate(page)); |
e3a7cca1 | 1128 | account_page_dirtied(page, mapping); |
8c08540f AM |
1129 | radix_tree_tag_set(&mapping->page_tree, |
1130 | page_index(page), PAGECACHE_TAG_DIRTY); | |
1131 | } | |
19fd6231 | 1132 | spin_unlock_irq(&mapping->tree_lock); |
8c08540f AM |
1133 | if (mapping->host) { |
1134 | /* !PageAnon && !swapper_space */ | |
1135 | __mark_inode_dirty(mapping->host, I_DIRTY_PAGES); | |
1da177e4 | 1136 | } |
4741c9fd | 1137 | return 1; |
1da177e4 | 1138 | } |
4741c9fd | 1139 | return 0; |
1da177e4 LT |
1140 | } |
1141 | EXPORT_SYMBOL(__set_page_dirty_nobuffers); | |
1142 | ||
1143 | /* | |
1144 | * When a writepage implementation decides that it doesn't want to write this | |
1145 | * page for some reason, it should redirty the locked page via | |
1146 | * redirty_page_for_writepage() and it should then unlock the page and return 0 | |
1147 | */ | |
1148 | int redirty_page_for_writepage(struct writeback_control *wbc, struct page *page) | |
1149 | { | |
1150 | wbc->pages_skipped++; | |
1151 | return __set_page_dirty_nobuffers(page); | |
1152 | } | |
1153 | EXPORT_SYMBOL(redirty_page_for_writepage); | |
1154 | ||
1155 | /* | |
6746aff7 WF |
1156 | * Dirty a page. |
1157 | * | |
1158 | * For pages with a mapping this should be done under the page lock | |
1159 | * for the benefit of asynchronous memory errors who prefer a consistent | |
1160 | * dirty state. This rule can be broken in some special cases, | |
1161 | * but should be better not to. | |
1162 | * | |
1da177e4 LT |
1163 | * If the mapping doesn't provide a set_page_dirty a_op, then |
1164 | * just fall through and assume that it wants buffer_heads. | |
1165 | */ | |
1cf6e7d8 | 1166 | int set_page_dirty(struct page *page) |
1da177e4 LT |
1167 | { |
1168 | struct address_space *mapping = page_mapping(page); | |
1169 | ||
1170 | if (likely(mapping)) { | |
1171 | int (*spd)(struct page *) = mapping->a_ops->set_page_dirty; | |
9361401e DH |
1172 | #ifdef CONFIG_BLOCK |
1173 | if (!spd) | |
1174 | spd = __set_page_dirty_buffers; | |
1175 | #endif | |
1176 | return (*spd)(page); | |
1da177e4 | 1177 | } |
4741c9fd AM |
1178 | if (!PageDirty(page)) { |
1179 | if (!TestSetPageDirty(page)) | |
1180 | return 1; | |
1181 | } | |
1da177e4 LT |
1182 | return 0; |
1183 | } | |
1184 | EXPORT_SYMBOL(set_page_dirty); | |
1185 | ||
1186 | /* | |
1187 | * set_page_dirty() is racy if the caller has no reference against | |
1188 | * page->mapping->host, and if the page is unlocked. This is because another | |
1189 | * CPU could truncate the page off the mapping and then free the mapping. | |
1190 | * | |
1191 | * Usually, the page _is_ locked, or the caller is a user-space process which | |
1192 | * holds a reference on the inode by having an open file. | |
1193 | * | |
1194 | * In other cases, the page should be locked before running set_page_dirty(). | |
1195 | */ | |
1196 | int set_page_dirty_lock(struct page *page) | |
1197 | { | |
1198 | int ret; | |
1199 | ||
db37648c | 1200 | lock_page_nosync(page); |
1da177e4 LT |
1201 | ret = set_page_dirty(page); |
1202 | unlock_page(page); | |
1203 | return ret; | |
1204 | } | |
1205 | EXPORT_SYMBOL(set_page_dirty_lock); | |
1206 | ||
1da177e4 LT |
1207 | /* |
1208 | * Clear a page's dirty flag, while caring for dirty memory accounting. | |
1209 | * Returns true if the page was previously dirty. | |
1210 | * | |
1211 | * This is for preparing to put the page under writeout. We leave the page | |
1212 | * tagged as dirty in the radix tree so that a concurrent write-for-sync | |
1213 | * can discover it via a PAGECACHE_TAG_DIRTY walk. The ->writepage | |
1214 | * implementation will run either set_page_writeback() or set_page_dirty(), | |
1215 | * at which stage we bring the page's dirty flag and radix-tree dirty tag | |
1216 | * back into sync. | |
1217 | * | |
1218 | * This incoherency between the page's dirty flag and radix-tree tag is | |
1219 | * unfortunate, but it only exists while the page is locked. | |
1220 | */ | |
1221 | int clear_page_dirty_for_io(struct page *page) | |
1222 | { | |
1223 | struct address_space *mapping = page_mapping(page); | |
1224 | ||
79352894 NP |
1225 | BUG_ON(!PageLocked(page)); |
1226 | ||
fe3cba17 | 1227 | ClearPageReclaim(page); |
7658cc28 LT |
1228 | if (mapping && mapping_cap_account_dirty(mapping)) { |
1229 | /* | |
1230 | * Yes, Virginia, this is indeed insane. | |
1231 | * | |
1232 | * We use this sequence to make sure that | |
1233 | * (a) we account for dirty stats properly | |
1234 | * (b) we tell the low-level filesystem to | |
1235 | * mark the whole page dirty if it was | |
1236 | * dirty in a pagetable. Only to then | |
1237 | * (c) clean the page again and return 1 to | |
1238 | * cause the writeback. | |
1239 | * | |
1240 | * This way we avoid all nasty races with the | |
1241 | * dirty bit in multiple places and clearing | |
1242 | * them concurrently from different threads. | |
1243 | * | |
1244 | * Note! Normally the "set_page_dirty(page)" | |
1245 | * has no effect on the actual dirty bit - since | |
1246 | * that will already usually be set. But we | |
1247 | * need the side effects, and it can help us | |
1248 | * avoid races. | |
1249 | * | |
1250 | * We basically use the page "master dirty bit" | |
1251 | * as a serialization point for all the different | |
1252 | * threads doing their things. | |
7658cc28 LT |
1253 | */ |
1254 | if (page_mkclean(page)) | |
1255 | set_page_dirty(page); | |
79352894 NP |
1256 | /* |
1257 | * We carefully synchronise fault handlers against | |
1258 | * installing a dirty pte and marking the page dirty | |
1259 | * at this point. We do this by having them hold the | |
1260 | * page lock at some point after installing their | |
1261 | * pte, but before marking the page dirty. | |
1262 | * Pages are always locked coming in here, so we get | |
1263 | * the desired exclusion. See mm/memory.c:do_wp_page() | |
1264 | * for more comments. | |
1265 | */ | |
7658cc28 | 1266 | if (TestClearPageDirty(page)) { |
8c08540f | 1267 | dec_zone_page_state(page, NR_FILE_DIRTY); |
c9e51e41 PZ |
1268 | dec_bdi_stat(mapping->backing_dev_info, |
1269 | BDI_RECLAIMABLE); | |
7658cc28 | 1270 | return 1; |
1da177e4 | 1271 | } |
7658cc28 | 1272 | return 0; |
1da177e4 | 1273 | } |
7658cc28 | 1274 | return TestClearPageDirty(page); |
1da177e4 | 1275 | } |
58bb01a9 | 1276 | EXPORT_SYMBOL(clear_page_dirty_for_io); |
1da177e4 LT |
1277 | |
1278 | int test_clear_page_writeback(struct page *page) | |
1279 | { | |
1280 | struct address_space *mapping = page_mapping(page); | |
1281 | int ret; | |
1282 | ||
1283 | if (mapping) { | |
69cb51d1 | 1284 | struct backing_dev_info *bdi = mapping->backing_dev_info; |
1da177e4 LT |
1285 | unsigned long flags; |
1286 | ||
19fd6231 | 1287 | spin_lock_irqsave(&mapping->tree_lock, flags); |
1da177e4 | 1288 | ret = TestClearPageWriteback(page); |
69cb51d1 | 1289 | if (ret) { |
1da177e4 LT |
1290 | radix_tree_tag_clear(&mapping->page_tree, |
1291 | page_index(page), | |
1292 | PAGECACHE_TAG_WRITEBACK); | |
e4ad08fe | 1293 | if (bdi_cap_account_writeback(bdi)) { |
69cb51d1 | 1294 | __dec_bdi_stat(bdi, BDI_WRITEBACK); |
04fbfdc1 PZ |
1295 | __bdi_writeout_inc(bdi); |
1296 | } | |
69cb51d1 | 1297 | } |
19fd6231 | 1298 | spin_unlock_irqrestore(&mapping->tree_lock, flags); |
1da177e4 LT |
1299 | } else { |
1300 | ret = TestClearPageWriteback(page); | |
1301 | } | |
d688abf5 AM |
1302 | if (ret) |
1303 | dec_zone_page_state(page, NR_WRITEBACK); | |
1da177e4 LT |
1304 | return ret; |
1305 | } | |
1306 | ||
1307 | int test_set_page_writeback(struct page *page) | |
1308 | { | |
1309 | struct address_space *mapping = page_mapping(page); | |
1310 | int ret; | |
1311 | ||
1312 | if (mapping) { | |
69cb51d1 | 1313 | struct backing_dev_info *bdi = mapping->backing_dev_info; |
1da177e4 LT |
1314 | unsigned long flags; |
1315 | ||
19fd6231 | 1316 | spin_lock_irqsave(&mapping->tree_lock, flags); |
1da177e4 | 1317 | ret = TestSetPageWriteback(page); |
69cb51d1 | 1318 | if (!ret) { |
1da177e4 LT |
1319 | radix_tree_tag_set(&mapping->page_tree, |
1320 | page_index(page), | |
1321 | PAGECACHE_TAG_WRITEBACK); | |
e4ad08fe | 1322 | if (bdi_cap_account_writeback(bdi)) |
69cb51d1 PZ |
1323 | __inc_bdi_stat(bdi, BDI_WRITEBACK); |
1324 | } | |
1da177e4 LT |
1325 | if (!PageDirty(page)) |
1326 | radix_tree_tag_clear(&mapping->page_tree, | |
1327 | page_index(page), | |
1328 | PAGECACHE_TAG_DIRTY); | |
19fd6231 | 1329 | spin_unlock_irqrestore(&mapping->tree_lock, flags); |
1da177e4 LT |
1330 | } else { |
1331 | ret = TestSetPageWriteback(page); | |
1332 | } | |
d688abf5 AM |
1333 | if (!ret) |
1334 | inc_zone_page_state(page, NR_WRITEBACK); | |
1da177e4 LT |
1335 | return ret; |
1336 | ||
1337 | } | |
1338 | EXPORT_SYMBOL(test_set_page_writeback); | |
1339 | ||
1340 | /* | |
00128188 | 1341 | * Return true if any of the pages in the mapping are marked with the |
1da177e4 LT |
1342 | * passed tag. |
1343 | */ | |
1344 | int mapping_tagged(struct address_space *mapping, int tag) | |
1345 | { | |
1da177e4 | 1346 | int ret; |
00128188 | 1347 | rcu_read_lock(); |
1da177e4 | 1348 | ret = radix_tree_tagged(&mapping->page_tree, tag); |
00128188 | 1349 | rcu_read_unlock(); |
1da177e4 LT |
1350 | return ret; |
1351 | } | |
1352 | EXPORT_SYMBOL(mapping_tagged); |