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 | { | |
247 | if (bdi_cap_writeback_dirty(bdi)) { | |
248 | prop_fraction_percpu(&vm_completions, &bdi->completions, | |
249 | numerator, denominator); | |
250 | } else { | |
251 | *numerator = 0; | |
252 | *denominator = 1; | |
253 | } | |
254 | } | |
255 | ||
256 | /* | |
257 | * Clip the earned share of dirty pages to that which is actually available. | |
258 | * This avoids exceeding the total dirty_limit when the floating averages | |
259 | * fluctuate too quickly. | |
260 | */ | |
dcf975d5 HS |
261 | static void clip_bdi_dirty_limit(struct backing_dev_info *bdi, |
262 | unsigned long dirty, unsigned long *pbdi_dirty) | |
04fbfdc1 | 263 | { |
dcf975d5 | 264 | unsigned long avail_dirty; |
04fbfdc1 | 265 | |
dcf975d5 | 266 | avail_dirty = global_page_state(NR_FILE_DIRTY) + |
04fbfdc1 | 267 | global_page_state(NR_WRITEBACK) + |
fc3ba692 | 268 | global_page_state(NR_UNSTABLE_NFS) + |
dcf975d5 | 269 | global_page_state(NR_WRITEBACK_TEMP); |
04fbfdc1 | 270 | |
dcf975d5 HS |
271 | if (avail_dirty < dirty) |
272 | avail_dirty = dirty - avail_dirty; | |
273 | else | |
04fbfdc1 PZ |
274 | avail_dirty = 0; |
275 | ||
276 | avail_dirty += bdi_stat(bdi, BDI_RECLAIMABLE) + | |
277 | bdi_stat(bdi, BDI_WRITEBACK); | |
278 | ||
279 | *pbdi_dirty = min(*pbdi_dirty, avail_dirty); | |
280 | } | |
281 | ||
3e26c149 PZ |
282 | static inline void task_dirties_fraction(struct task_struct *tsk, |
283 | long *numerator, long *denominator) | |
284 | { | |
285 | prop_fraction_single(&vm_dirties, &tsk->dirties, | |
286 | numerator, denominator); | |
287 | } | |
288 | ||
289 | /* | |
290 | * scale the dirty limit | |
291 | * | |
292 | * task specific dirty limit: | |
293 | * | |
294 | * dirty -= (dirty/8) * p_{t} | |
295 | */ | |
dcf975d5 | 296 | static void task_dirty_limit(struct task_struct *tsk, unsigned long *pdirty) |
3e26c149 PZ |
297 | { |
298 | long numerator, denominator; | |
dcf975d5 | 299 | unsigned long dirty = *pdirty; |
3e26c149 PZ |
300 | u64 inv = dirty >> 3; |
301 | ||
302 | task_dirties_fraction(tsk, &numerator, &denominator); | |
303 | inv *= numerator; | |
304 | do_div(inv, denominator); | |
305 | ||
306 | dirty -= inv; | |
307 | if (dirty < *pdirty/2) | |
308 | dirty = *pdirty/2; | |
309 | ||
310 | *pdirty = dirty; | |
311 | } | |
312 | ||
189d3c4a PZ |
313 | /* |
314 | * | |
315 | */ | |
189d3c4a PZ |
316 | static unsigned int bdi_min_ratio; |
317 | ||
318 | int bdi_set_min_ratio(struct backing_dev_info *bdi, unsigned int min_ratio) | |
319 | { | |
320 | int ret = 0; | |
189d3c4a | 321 | |
cfc4ba53 | 322 | spin_lock_bh(&bdi_lock); |
a42dde04 | 323 | if (min_ratio > bdi->max_ratio) { |
189d3c4a | 324 | ret = -EINVAL; |
a42dde04 PZ |
325 | } else { |
326 | min_ratio -= bdi->min_ratio; | |
327 | if (bdi_min_ratio + min_ratio < 100) { | |
328 | bdi_min_ratio += min_ratio; | |
329 | bdi->min_ratio += min_ratio; | |
330 | } else { | |
331 | ret = -EINVAL; | |
332 | } | |
333 | } | |
cfc4ba53 | 334 | spin_unlock_bh(&bdi_lock); |
a42dde04 PZ |
335 | |
336 | return ret; | |
337 | } | |
338 | ||
339 | int bdi_set_max_ratio(struct backing_dev_info *bdi, unsigned max_ratio) | |
340 | { | |
a42dde04 PZ |
341 | int ret = 0; |
342 | ||
343 | if (max_ratio > 100) | |
344 | return -EINVAL; | |
345 | ||
cfc4ba53 | 346 | spin_lock_bh(&bdi_lock); |
a42dde04 PZ |
347 | if (bdi->min_ratio > max_ratio) { |
348 | ret = -EINVAL; | |
349 | } else { | |
350 | bdi->max_ratio = max_ratio; | |
351 | bdi->max_prop_frac = (PROP_FRAC_BASE * max_ratio) / 100; | |
352 | } | |
cfc4ba53 | 353 | spin_unlock_bh(&bdi_lock); |
189d3c4a PZ |
354 | |
355 | return ret; | |
356 | } | |
a42dde04 | 357 | EXPORT_SYMBOL(bdi_set_max_ratio); |
189d3c4a | 358 | |
1da177e4 LT |
359 | /* |
360 | * Work out the current dirty-memory clamping and background writeout | |
361 | * thresholds. | |
362 | * | |
363 | * The main aim here is to lower them aggressively if there is a lot of mapped | |
364 | * memory around. To avoid stressing page reclaim with lots of unreclaimable | |
365 | * pages. It is better to clamp down on writers than to start swapping, and | |
366 | * performing lots of scanning. | |
367 | * | |
368 | * We only allow 1/2 of the currently-unmapped memory to be dirtied. | |
369 | * | |
370 | * We don't permit the clamping level to fall below 5% - that is getting rather | |
371 | * excessive. | |
372 | * | |
373 | * We make sure that the background writeout level is below the adjusted | |
374 | * clamping level. | |
375 | */ | |
1b424464 CL |
376 | |
377 | static unsigned long highmem_dirtyable_memory(unsigned long total) | |
378 | { | |
379 | #ifdef CONFIG_HIGHMEM | |
380 | int node; | |
381 | unsigned long x = 0; | |
382 | ||
37b07e41 | 383 | for_each_node_state(node, N_HIGH_MEMORY) { |
1b424464 CL |
384 | struct zone *z = |
385 | &NODE_DATA(node)->node_zones[ZONE_HIGHMEM]; | |
386 | ||
adea02a1 WF |
387 | x += zone_page_state(z, NR_FREE_PAGES) + |
388 | zone_reclaimable_pages(z); | |
1b424464 CL |
389 | } |
390 | /* | |
391 | * Make sure that the number of highmem pages is never larger | |
392 | * than the number of the total dirtyable memory. This can only | |
393 | * occur in very strange VM situations but we want to make sure | |
394 | * that this does not occur. | |
395 | */ | |
396 | return min(x, total); | |
397 | #else | |
398 | return 0; | |
399 | #endif | |
400 | } | |
401 | ||
3eefae99 SR |
402 | /** |
403 | * determine_dirtyable_memory - amount of memory that may be used | |
404 | * | |
405 | * Returns the numebr of pages that can currently be freed and used | |
406 | * by the kernel for direct mappings. | |
407 | */ | |
408 | unsigned long determine_dirtyable_memory(void) | |
1b424464 CL |
409 | { |
410 | unsigned long x; | |
411 | ||
adea02a1 | 412 | x = global_page_state(NR_FREE_PAGES) + global_reclaimable_pages(); |
195cf453 BG |
413 | |
414 | if (!vm_highmem_is_dirtyable) | |
415 | x -= highmem_dirtyable_memory(x); | |
416 | ||
1b424464 CL |
417 | return x + 1; /* Ensure that we never return 0 */ |
418 | } | |
419 | ||
cf0ca9fe | 420 | void |
364aeb28 DR |
421 | get_dirty_limits(unsigned long *pbackground, unsigned long *pdirty, |
422 | unsigned long *pbdi_dirty, struct backing_dev_info *bdi) | |
1da177e4 | 423 | { |
364aeb28 DR |
424 | unsigned long background; |
425 | unsigned long dirty; | |
1b424464 | 426 | unsigned long available_memory = determine_dirtyable_memory(); |
1da177e4 LT |
427 | struct task_struct *tsk; |
428 | ||
2da02997 DR |
429 | if (vm_dirty_bytes) |
430 | dirty = DIV_ROUND_UP(vm_dirty_bytes, PAGE_SIZE); | |
431 | else { | |
432 | int dirty_ratio; | |
433 | ||
434 | dirty_ratio = vm_dirty_ratio; | |
435 | if (dirty_ratio < 5) | |
436 | dirty_ratio = 5; | |
437 | dirty = (dirty_ratio * available_memory) / 100; | |
438 | } | |
1da177e4 | 439 | |
2da02997 DR |
440 | if (dirty_background_bytes) |
441 | background = DIV_ROUND_UP(dirty_background_bytes, PAGE_SIZE); | |
442 | else | |
443 | background = (dirty_background_ratio * available_memory) / 100; | |
1da177e4 | 444 | |
2da02997 DR |
445 | if (background >= dirty) |
446 | background = dirty / 2; | |
1da177e4 LT |
447 | tsk = current; |
448 | if (tsk->flags & PF_LESS_THROTTLE || rt_task(tsk)) { | |
449 | background += background / 4; | |
450 | dirty += dirty / 4; | |
451 | } | |
452 | *pbackground = background; | |
453 | *pdirty = dirty; | |
04fbfdc1 PZ |
454 | |
455 | if (bdi) { | |
189d3c4a | 456 | u64 bdi_dirty; |
04fbfdc1 PZ |
457 | long numerator, denominator; |
458 | ||
459 | /* | |
460 | * Calculate this BDI's share of the dirty ratio. | |
461 | */ | |
462 | bdi_writeout_fraction(bdi, &numerator, &denominator); | |
463 | ||
189d3c4a | 464 | bdi_dirty = (dirty * (100 - bdi_min_ratio)) / 100; |
04fbfdc1 PZ |
465 | bdi_dirty *= numerator; |
466 | do_div(bdi_dirty, denominator); | |
189d3c4a | 467 | bdi_dirty += (dirty * bdi->min_ratio) / 100; |
a42dde04 PZ |
468 | if (bdi_dirty > (dirty * bdi->max_ratio) / 100) |
469 | bdi_dirty = dirty * bdi->max_ratio / 100; | |
04fbfdc1 PZ |
470 | |
471 | *pbdi_dirty = bdi_dirty; | |
472 | clip_bdi_dirty_limit(bdi, dirty, pbdi_dirty); | |
3e26c149 | 473 | task_dirty_limit(current, pbdi_dirty); |
04fbfdc1 | 474 | } |
1da177e4 LT |
475 | } |
476 | ||
477 | /* | |
478 | * balance_dirty_pages() must be called by processes which are generating dirty | |
479 | * data. It looks at the number of dirty pages in the machine and will force | |
480 | * the caller to perform writeback if the system is over `vm_dirty_ratio'. | |
5b0830cb JA |
481 | * If we're over `background_thresh' then the writeback threads are woken to |
482 | * perform some writeout. | |
1da177e4 | 483 | */ |
3a2e9a5a WF |
484 | static void balance_dirty_pages(struct address_space *mapping, |
485 | unsigned long write_chunk) | |
1da177e4 | 486 | { |
5fce25a9 PZ |
487 | long nr_reclaimable, bdi_nr_reclaimable; |
488 | long nr_writeback, bdi_nr_writeback; | |
364aeb28 DR |
489 | unsigned long background_thresh; |
490 | unsigned long dirty_thresh; | |
491 | unsigned long bdi_thresh; | |
1da177e4 | 492 | unsigned long pages_written = 0; |
87c6a9b2 | 493 | unsigned long pause = 1; |
1da177e4 LT |
494 | |
495 | struct backing_dev_info *bdi = mapping->backing_dev_info; | |
496 | ||
497 | for (;;) { | |
498 | struct writeback_control wbc = { | |
1da177e4 LT |
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 | */ |
028c2dd1 | 539 | trace_wbc_balance_dirty_start(&wbc, bdi); |
d7831a0b | 540 | if (bdi_nr_reclaimable > bdi_thresh) { |
9c3a8ee8 | 541 | writeback_inodes_wb(&bdi->wb, &wbc); |
1da177e4 | 542 | pages_written += write_chunk - wbc.nr_to_write; |
04fbfdc1 PZ |
543 | get_dirty_limits(&background_thresh, &dirty_thresh, |
544 | &bdi_thresh, bdi); | |
028c2dd1 | 545 | trace_wbc_balance_dirty_written(&wbc, bdi); |
04fbfdc1 PZ |
546 | } |
547 | ||
548 | /* | |
549 | * In order to avoid the stacked BDI deadlock we need | |
550 | * to ensure we accurately count the 'dirty' pages when | |
551 | * the threshold is low. | |
552 | * | |
553 | * Otherwise it would be possible to get thresh+n pages | |
554 | * reported dirty, even though there are thresh-m pages | |
555 | * actually dirty; with m+n sitting in the percpu | |
556 | * deltas. | |
557 | */ | |
558 | if (bdi_thresh < 2*bdi_stat_error(bdi)) { | |
559 | bdi_nr_reclaimable = bdi_stat_sum(bdi, BDI_RECLAIMABLE); | |
560 | bdi_nr_writeback = bdi_stat_sum(bdi, BDI_WRITEBACK); | |
561 | } else if (bdi_nr_reclaimable) { | |
562 | bdi_nr_reclaimable = bdi_stat(bdi, BDI_RECLAIMABLE); | |
563 | bdi_nr_writeback = bdi_stat(bdi, BDI_WRITEBACK); | |
1da177e4 | 564 | } |
04fbfdc1 PZ |
565 | |
566 | if (bdi_nr_reclaimable + bdi_nr_writeback <= bdi_thresh) | |
567 | break; | |
568 | if (pages_written >= write_chunk) | |
569 | break; /* We've done our duty */ | |
570 | ||
028c2dd1 | 571 | trace_wbc_balance_dirty_wait(&wbc, bdi); |
d25105e8 WF |
572 | __set_current_state(TASK_INTERRUPTIBLE); |
573 | io_schedule_timeout(pause); | |
87c6a9b2 JA |
574 | |
575 | /* | |
576 | * Increase the delay for each loop, up to our previous | |
577 | * default of taking a 100ms nap. | |
578 | */ | |
579 | pause <<= 1; | |
580 | if (pause > HZ / 10) | |
581 | pause = HZ / 10; | |
1da177e4 LT |
582 | } |
583 | ||
04fbfdc1 PZ |
584 | if (bdi_nr_reclaimable + bdi_nr_writeback < bdi_thresh && |
585 | bdi->dirty_exceeded) | |
586 | bdi->dirty_exceeded = 0; | |
1da177e4 LT |
587 | |
588 | if (writeback_in_progress(bdi)) | |
5b0830cb | 589 | return; |
1da177e4 LT |
590 | |
591 | /* | |
592 | * In laptop mode, we wait until hitting the higher threshold before | |
593 | * starting background writeout, and then write out all the way down | |
594 | * to the lower threshold. So slow writers cause minimal disk activity. | |
595 | * | |
596 | * In normal mode, we start background writeout at the lower | |
597 | * background_thresh, to keep the amount of dirty memory low. | |
598 | */ | |
599 | if ((laptop_mode && pages_written) || | |
d3ddec76 WF |
600 | (!laptop_mode && ((global_page_state(NR_FILE_DIRTY) |
601 | + global_page_state(NR_UNSTABLE_NFS)) | |
b6e51316 | 602 | > background_thresh))) |
c5444198 | 603 | bdi_start_background_writeback(bdi); |
1da177e4 LT |
604 | } |
605 | ||
a200ee18 | 606 | void set_page_dirty_balance(struct page *page, int page_mkwrite) |
edc79b2a | 607 | { |
a200ee18 | 608 | if (set_page_dirty(page) || page_mkwrite) { |
edc79b2a PZ |
609 | struct address_space *mapping = page_mapping(page); |
610 | ||
611 | if (mapping) | |
612 | balance_dirty_pages_ratelimited(mapping); | |
613 | } | |
614 | } | |
615 | ||
245b2e70 TH |
616 | static DEFINE_PER_CPU(unsigned long, bdp_ratelimits) = 0; |
617 | ||
1da177e4 | 618 | /** |
fa5a734e | 619 | * balance_dirty_pages_ratelimited_nr - balance dirty memory state |
67be2dd1 | 620 | * @mapping: address_space which was dirtied |
a580290c | 621 | * @nr_pages_dirtied: number of pages which the caller has just dirtied |
1da177e4 LT |
622 | * |
623 | * Processes which are dirtying memory should call in here once for each page | |
624 | * which was newly dirtied. The function will periodically check the system's | |
625 | * dirty state and will initiate writeback if needed. | |
626 | * | |
627 | * On really big machines, get_writeback_state is expensive, so try to avoid | |
628 | * calling it too often (ratelimiting). But once we're over the dirty memory | |
629 | * limit we decrease the ratelimiting by a lot, to prevent individual processes | |
630 | * from overshooting the limit by (ratelimit_pages) each. | |
631 | */ | |
fa5a734e AM |
632 | void balance_dirty_pages_ratelimited_nr(struct address_space *mapping, |
633 | unsigned long nr_pages_dirtied) | |
1da177e4 | 634 | { |
fa5a734e AM |
635 | unsigned long ratelimit; |
636 | unsigned long *p; | |
1da177e4 LT |
637 | |
638 | ratelimit = ratelimit_pages; | |
04fbfdc1 | 639 | if (mapping->backing_dev_info->dirty_exceeded) |
1da177e4 LT |
640 | ratelimit = 8; |
641 | ||
642 | /* | |
643 | * Check the rate limiting. Also, we do not want to throttle real-time | |
644 | * tasks in balance_dirty_pages(). Period. | |
645 | */ | |
fa5a734e | 646 | preempt_disable(); |
245b2e70 | 647 | p = &__get_cpu_var(bdp_ratelimits); |
fa5a734e AM |
648 | *p += nr_pages_dirtied; |
649 | if (unlikely(*p >= ratelimit)) { | |
3a2e9a5a | 650 | ratelimit = sync_writeback_pages(*p); |
fa5a734e AM |
651 | *p = 0; |
652 | preempt_enable(); | |
3a2e9a5a | 653 | balance_dirty_pages(mapping, ratelimit); |
1da177e4 LT |
654 | return; |
655 | } | |
fa5a734e | 656 | preempt_enable(); |
1da177e4 | 657 | } |
fa5a734e | 658 | EXPORT_SYMBOL(balance_dirty_pages_ratelimited_nr); |
1da177e4 | 659 | |
232ea4d6 | 660 | void throttle_vm_writeout(gfp_t gfp_mask) |
1da177e4 | 661 | { |
364aeb28 DR |
662 | unsigned long background_thresh; |
663 | unsigned long dirty_thresh; | |
1da177e4 LT |
664 | |
665 | for ( ; ; ) { | |
04fbfdc1 | 666 | get_dirty_limits(&background_thresh, &dirty_thresh, NULL, NULL); |
1da177e4 LT |
667 | |
668 | /* | |
669 | * Boost the allowable dirty threshold a bit for page | |
670 | * allocators so they don't get DoS'ed by heavy writers | |
671 | */ | |
672 | dirty_thresh += dirty_thresh / 10; /* wheeee... */ | |
673 | ||
c24f21bd CL |
674 | if (global_page_state(NR_UNSTABLE_NFS) + |
675 | global_page_state(NR_WRITEBACK) <= dirty_thresh) | |
676 | break; | |
8aa7e847 | 677 | congestion_wait(BLK_RW_ASYNC, HZ/10); |
369f2389 FW |
678 | |
679 | /* | |
680 | * The caller might hold locks which can prevent IO completion | |
681 | * or progress in the filesystem. So we cannot just sit here | |
682 | * waiting for IO to complete. | |
683 | */ | |
684 | if ((gfp_mask & (__GFP_FS|__GFP_IO)) != (__GFP_FS|__GFP_IO)) | |
685 | break; | |
1da177e4 LT |
686 | } |
687 | } | |
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); |
6423104b | 696 | bdi_arm_supers_timer(); |
1da177e4 LT |
697 | return 0; |
698 | } | |
699 | ||
c2c4986e | 700 | #ifdef CONFIG_BLOCK |
31373d09 | 701 | void laptop_mode_timer_fn(unsigned long data) |
1da177e4 | 702 | { |
31373d09 MG |
703 | struct request_queue *q = (struct request_queue *)data; |
704 | int nr_pages = global_page_state(NR_FILE_DIRTY) + | |
705 | global_page_state(NR_UNSTABLE_NFS); | |
1da177e4 | 706 | |
31373d09 MG |
707 | /* |
708 | * We want to write everything out, not just down to the dirty | |
709 | * threshold | |
710 | */ | |
31373d09 | 711 | if (bdi_has_dirty_io(&q->backing_dev_info)) |
c5444198 | 712 | bdi_start_writeback(&q->backing_dev_info, nr_pages); |
1da177e4 LT |
713 | } |
714 | ||
715 | /* | |
716 | * We've spun up the disk and we're in laptop mode: schedule writeback | |
717 | * of all dirty data a few seconds from now. If the flush is already scheduled | |
718 | * then push it back - the user is still using the disk. | |
719 | */ | |
31373d09 | 720 | void laptop_io_completion(struct backing_dev_info *info) |
1da177e4 | 721 | { |
31373d09 | 722 | mod_timer(&info->laptop_mode_wb_timer, jiffies + laptop_mode); |
1da177e4 LT |
723 | } |
724 | ||
725 | /* | |
726 | * We're in laptop mode and we've just synced. The sync's writes will have | |
727 | * caused another writeback to be scheduled by laptop_io_completion. | |
728 | * Nothing needs to be written back anymore, so we unschedule the writeback. | |
729 | */ | |
730 | void laptop_sync_completion(void) | |
731 | { | |
31373d09 MG |
732 | struct backing_dev_info *bdi; |
733 | ||
734 | rcu_read_lock(); | |
735 | ||
736 | list_for_each_entry_rcu(bdi, &bdi_list, bdi_list) | |
737 | del_timer(&bdi->laptop_mode_wb_timer); | |
738 | ||
739 | rcu_read_unlock(); | |
1da177e4 | 740 | } |
c2c4986e | 741 | #endif |
1da177e4 LT |
742 | |
743 | /* | |
744 | * If ratelimit_pages is too high then we can get into dirty-data overload | |
745 | * if a large number of processes all perform writes at the same time. | |
746 | * If it is too low then SMP machines will call the (expensive) | |
747 | * get_writeback_state too often. | |
748 | * | |
749 | * Here we set ratelimit_pages to a level which ensures that when all CPUs are | |
750 | * dirtying in parallel, we cannot go more than 3% (1/32) over the dirty memory | |
751 | * thresholds before writeback cuts in. | |
752 | * | |
753 | * But the limit should not be set too high. Because it also controls the | |
754 | * amount of memory which the balance_dirty_pages() caller has to write back. | |
755 | * If this is too large then the caller will block on the IO queue all the | |
756 | * time. So limit it to four megabytes - the balance_dirty_pages() caller | |
757 | * will write six megabyte chunks, max. | |
758 | */ | |
759 | ||
2d1d43f6 | 760 | void writeback_set_ratelimit(void) |
1da177e4 | 761 | { |
40c99aae | 762 | ratelimit_pages = vm_total_pages / (num_online_cpus() * 32); |
1da177e4 LT |
763 | if (ratelimit_pages < 16) |
764 | ratelimit_pages = 16; | |
765 | if (ratelimit_pages * PAGE_CACHE_SIZE > 4096 * 1024) | |
766 | ratelimit_pages = (4096 * 1024) / PAGE_CACHE_SIZE; | |
767 | } | |
768 | ||
26c2143b | 769 | static int __cpuinit |
1da177e4 LT |
770 | ratelimit_handler(struct notifier_block *self, unsigned long u, void *v) |
771 | { | |
2d1d43f6 | 772 | writeback_set_ratelimit(); |
aa0f0303 | 773 | return NOTIFY_DONE; |
1da177e4 LT |
774 | } |
775 | ||
74b85f37 | 776 | static struct notifier_block __cpuinitdata ratelimit_nb = { |
1da177e4 LT |
777 | .notifier_call = ratelimit_handler, |
778 | .next = NULL, | |
779 | }; | |
780 | ||
781 | /* | |
dc6e29da LT |
782 | * Called early on to tune the page writeback dirty limits. |
783 | * | |
784 | * We used to scale dirty pages according to how total memory | |
785 | * related to pages that could be allocated for buffers (by | |
786 | * comparing nr_free_buffer_pages() to vm_total_pages. | |
787 | * | |
788 | * However, that was when we used "dirty_ratio" to scale with | |
789 | * all memory, and we don't do that any more. "dirty_ratio" | |
790 | * is now applied to total non-HIGHPAGE memory (by subtracting | |
791 | * totalhigh_pages from vm_total_pages), and as such we can't | |
792 | * get into the old insane situation any more where we had | |
793 | * large amounts of dirty pages compared to a small amount of | |
794 | * non-HIGHMEM memory. | |
795 | * | |
796 | * But we might still want to scale the dirty_ratio by how | |
797 | * much memory the box has.. | |
1da177e4 LT |
798 | */ |
799 | void __init page_writeback_init(void) | |
800 | { | |
04fbfdc1 PZ |
801 | int shift; |
802 | ||
2d1d43f6 | 803 | writeback_set_ratelimit(); |
1da177e4 | 804 | register_cpu_notifier(&ratelimit_nb); |
04fbfdc1 PZ |
805 | |
806 | shift = calc_period_shift(); | |
807 | prop_descriptor_init(&vm_completions, shift); | |
3e26c149 | 808 | prop_descriptor_init(&vm_dirties, shift); |
1da177e4 LT |
809 | } |
810 | ||
f446daae JK |
811 | /** |
812 | * tag_pages_for_writeback - tag pages to be written by write_cache_pages | |
813 | * @mapping: address space structure to write | |
814 | * @start: starting page index | |
815 | * @end: ending page index (inclusive) | |
816 | * | |
817 | * This function scans the page range from @start to @end (inclusive) and tags | |
818 | * all pages that have DIRTY tag set with a special TOWRITE tag. The idea is | |
819 | * that write_cache_pages (or whoever calls this function) will then use | |
820 | * TOWRITE tag to identify pages eligible for writeback. This mechanism is | |
821 | * used to avoid livelocking of writeback by a process steadily creating new | |
822 | * dirty pages in the file (thus it is important for this function to be quick | |
823 | * so that it can tag pages faster than a dirtying process can create them). | |
824 | */ | |
825 | /* | |
826 | * We tag pages in batches of WRITEBACK_TAG_BATCH to reduce tree_lock latency. | |
827 | */ | |
f446daae JK |
828 | void tag_pages_for_writeback(struct address_space *mapping, |
829 | pgoff_t start, pgoff_t end) | |
830 | { | |
3c111a07 | 831 | #define WRITEBACK_TAG_BATCH 4096 |
f446daae JK |
832 | unsigned long tagged; |
833 | ||
834 | do { | |
835 | spin_lock_irq(&mapping->tree_lock); | |
836 | tagged = radix_tree_range_tag_if_tagged(&mapping->page_tree, | |
837 | &start, end, WRITEBACK_TAG_BATCH, | |
838 | PAGECACHE_TAG_DIRTY, PAGECACHE_TAG_TOWRITE); | |
839 | spin_unlock_irq(&mapping->tree_lock); | |
840 | WARN_ON_ONCE(tagged > WRITEBACK_TAG_BATCH); | |
841 | cond_resched(); | |
842 | } while (tagged >= WRITEBACK_TAG_BATCH); | |
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 | } |
f446daae JK |
900 | if (wbc->sync_mode == WB_SYNC_ALL) |
901 | tag = PAGECACHE_TAG_TOWRITE; | |
902 | else | |
903 | tag = PAGECACHE_TAG_DIRTY; | |
811d736f | 904 | retry: |
f446daae JK |
905 | if (wbc->sync_mode == WB_SYNC_ALL) |
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 | ||
935 | done_index = page->index + 1; | |
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 | */ | |
985 | done = 1; | |
986 | break; | |
987 | } | |
0b564927 | 988 | } |
00266770 | 989 | |
0b564927 DC |
990 | if (wbc->nr_to_write > 0) { |
991 | if (--wbc->nr_to_write == 0 && | |
89e12190 FC |
992 | wbc->sync_mode == WB_SYNC_NONE) { |
993 | /* | |
994 | * We stop writing back only if we are | |
995 | * not doing integrity sync. In case of | |
996 | * integrity sync we have to keep going | |
997 | * because someone may be concurrently | |
998 | * dirtying pages, and we might have | |
999 | * synced a lot of newly appeared dirty | |
1000 | * pages, but have not synced all of the | |
1001 | * old dirty pages. | |
1002 | */ | |
1003 | done = 1; | |
1004 | break; | |
1005 | } | |
05fe478d | 1006 | } |
811d736f DH |
1007 | } |
1008 | pagevec_release(&pvec); | |
1009 | cond_resched(); | |
1010 | } | |
3a4c6800 | 1011 | if (!cycled && !done) { |
811d736f | 1012 | /* |
31a12666 | 1013 | * range_cyclic: |
811d736f DH |
1014 | * We hit the last page and there is more work to be done: wrap |
1015 | * back to the start of the file | |
1016 | */ | |
31a12666 | 1017 | cycled = 1; |
811d736f | 1018 | index = 0; |
31a12666 | 1019 | end = writeback_index - 1; |
811d736f DH |
1020 | goto retry; |
1021 | } | |
0b564927 DC |
1022 | if (wbc->range_cyclic || (range_whole && wbc->nr_to_write > 0)) |
1023 | mapping->writeback_index = done_index; | |
06d6cf69 | 1024 | |
811d736f DH |
1025 | return ret; |
1026 | } | |
0ea97180 MS |
1027 | EXPORT_SYMBOL(write_cache_pages); |
1028 | ||
1029 | /* | |
1030 | * Function used by generic_writepages to call the real writepage | |
1031 | * function and set the mapping flags on error | |
1032 | */ | |
1033 | static int __writepage(struct page *page, struct writeback_control *wbc, | |
1034 | void *data) | |
1035 | { | |
1036 | struct address_space *mapping = data; | |
1037 | int ret = mapping->a_ops->writepage(page, wbc); | |
1038 | mapping_set_error(mapping, ret); | |
1039 | return ret; | |
1040 | } | |
1041 | ||
1042 | /** | |
1043 | * generic_writepages - walk the list of dirty pages of the given address space and writepage() all of them. | |
1044 | * @mapping: address space structure to write | |
1045 | * @wbc: subtract the number of written pages from *@wbc->nr_to_write | |
1046 | * | |
1047 | * This is a library function, which implements the writepages() | |
1048 | * address_space_operation. | |
1049 | */ | |
1050 | int generic_writepages(struct address_space *mapping, | |
1051 | struct writeback_control *wbc) | |
1052 | { | |
1053 | /* deal with chardevs and other special file */ | |
1054 | if (!mapping->a_ops->writepage) | |
1055 | return 0; | |
1056 | ||
1057 | return write_cache_pages(mapping, wbc, __writepage, mapping); | |
1058 | } | |
811d736f DH |
1059 | |
1060 | EXPORT_SYMBOL(generic_writepages); | |
1061 | ||
1da177e4 LT |
1062 | int do_writepages(struct address_space *mapping, struct writeback_control *wbc) |
1063 | { | |
22905f77 AM |
1064 | int ret; |
1065 | ||
1da177e4 LT |
1066 | if (wbc->nr_to_write <= 0) |
1067 | return 0; | |
1068 | if (mapping->a_ops->writepages) | |
d08b3851 | 1069 | ret = mapping->a_ops->writepages(mapping, wbc); |
22905f77 AM |
1070 | else |
1071 | ret = generic_writepages(mapping, wbc); | |
22905f77 | 1072 | return ret; |
1da177e4 LT |
1073 | } |
1074 | ||
1075 | /** | |
1076 | * write_one_page - write out a single page and optionally wait on I/O | |
67be2dd1 MW |
1077 | * @page: the page to write |
1078 | * @wait: if true, wait on writeout | |
1da177e4 LT |
1079 | * |
1080 | * The page must be locked by the caller and will be unlocked upon return. | |
1081 | * | |
1082 | * write_one_page() returns a negative error code if I/O failed. | |
1083 | */ | |
1084 | int write_one_page(struct page *page, int wait) | |
1085 | { | |
1086 | struct address_space *mapping = page->mapping; | |
1087 | int ret = 0; | |
1088 | struct writeback_control wbc = { | |
1089 | .sync_mode = WB_SYNC_ALL, | |
1090 | .nr_to_write = 1, | |
1091 | }; | |
1092 | ||
1093 | BUG_ON(!PageLocked(page)); | |
1094 | ||
1095 | if (wait) | |
1096 | wait_on_page_writeback(page); | |
1097 | ||
1098 | if (clear_page_dirty_for_io(page)) { | |
1099 | page_cache_get(page); | |
1100 | ret = mapping->a_ops->writepage(page, &wbc); | |
1101 | if (ret == 0 && wait) { | |
1102 | wait_on_page_writeback(page); | |
1103 | if (PageError(page)) | |
1104 | ret = -EIO; | |
1105 | } | |
1106 | page_cache_release(page); | |
1107 | } else { | |
1108 | unlock_page(page); | |
1109 | } | |
1110 | return ret; | |
1111 | } | |
1112 | EXPORT_SYMBOL(write_one_page); | |
1113 | ||
76719325 KC |
1114 | /* |
1115 | * For address_spaces which do not use buffers nor write back. | |
1116 | */ | |
1117 | int __set_page_dirty_no_writeback(struct page *page) | |
1118 | { | |
1119 | if (!PageDirty(page)) | |
1120 | SetPageDirty(page); | |
1121 | return 0; | |
1122 | } | |
1123 | ||
e3a7cca1 ES |
1124 | /* |
1125 | * Helper function for set_page_dirty family. | |
1126 | * NOTE: This relies on being atomic wrt interrupts. | |
1127 | */ | |
1128 | void account_page_dirtied(struct page *page, struct address_space *mapping) | |
1129 | { | |
1130 | if (mapping_cap_account_dirty(mapping)) { | |
1131 | __inc_zone_page_state(page, NR_FILE_DIRTY); | |
1132 | __inc_bdi_stat(mapping->backing_dev_info, BDI_RECLAIMABLE); | |
1133 | task_dirty_inc(current); | |
1134 | task_io_account_write(PAGE_CACHE_SIZE); | |
1135 | } | |
1136 | } | |
1137 | ||
1da177e4 LT |
1138 | /* |
1139 | * For address_spaces which do not use buffers. Just tag the page as dirty in | |
1140 | * its radix tree. | |
1141 | * | |
1142 | * This is also used when a single buffer is being dirtied: we want to set the | |
1143 | * page dirty in that case, but not all the buffers. This is a "bottom-up" | |
1144 | * dirtying, whereas __set_page_dirty_buffers() is a "top-down" dirtying. | |
1145 | * | |
1146 | * Most callers have locked the page, which pins the address_space in memory. | |
1147 | * But zap_pte_range() does not lock the page, however in that case the | |
1148 | * mapping is pinned by the vma's ->vm_file reference. | |
1149 | * | |
1150 | * We take care to handle the case where the page was truncated from the | |
183ff22b | 1151 | * mapping by re-checking page_mapping() inside tree_lock. |
1da177e4 LT |
1152 | */ |
1153 | int __set_page_dirty_nobuffers(struct page *page) | |
1154 | { | |
1da177e4 LT |
1155 | if (!TestSetPageDirty(page)) { |
1156 | struct address_space *mapping = page_mapping(page); | |
1157 | struct address_space *mapping2; | |
1158 | ||
8c08540f AM |
1159 | if (!mapping) |
1160 | return 1; | |
1161 | ||
19fd6231 | 1162 | spin_lock_irq(&mapping->tree_lock); |
8c08540f AM |
1163 | mapping2 = page_mapping(page); |
1164 | if (mapping2) { /* Race with truncate? */ | |
1165 | BUG_ON(mapping2 != mapping); | |
787d2214 | 1166 | WARN_ON_ONCE(!PagePrivate(page) && !PageUptodate(page)); |
e3a7cca1 | 1167 | account_page_dirtied(page, mapping); |
8c08540f AM |
1168 | radix_tree_tag_set(&mapping->page_tree, |
1169 | page_index(page), PAGECACHE_TAG_DIRTY); | |
1170 | } | |
19fd6231 | 1171 | spin_unlock_irq(&mapping->tree_lock); |
8c08540f AM |
1172 | if (mapping->host) { |
1173 | /* !PageAnon && !swapper_space */ | |
1174 | __mark_inode_dirty(mapping->host, I_DIRTY_PAGES); | |
1da177e4 | 1175 | } |
4741c9fd | 1176 | return 1; |
1da177e4 | 1177 | } |
4741c9fd | 1178 | return 0; |
1da177e4 LT |
1179 | } |
1180 | EXPORT_SYMBOL(__set_page_dirty_nobuffers); | |
1181 | ||
1182 | /* | |
1183 | * When a writepage implementation decides that it doesn't want to write this | |
1184 | * page for some reason, it should redirty the locked page via | |
1185 | * redirty_page_for_writepage() and it should then unlock the page and return 0 | |
1186 | */ | |
1187 | int redirty_page_for_writepage(struct writeback_control *wbc, struct page *page) | |
1188 | { | |
1189 | wbc->pages_skipped++; | |
1190 | return __set_page_dirty_nobuffers(page); | |
1191 | } | |
1192 | EXPORT_SYMBOL(redirty_page_for_writepage); | |
1193 | ||
1194 | /* | |
6746aff7 WF |
1195 | * Dirty a page. |
1196 | * | |
1197 | * For pages with a mapping this should be done under the page lock | |
1198 | * for the benefit of asynchronous memory errors who prefer a consistent | |
1199 | * dirty state. This rule can be broken in some special cases, | |
1200 | * but should be better not to. | |
1201 | * | |
1da177e4 LT |
1202 | * If the mapping doesn't provide a set_page_dirty a_op, then |
1203 | * just fall through and assume that it wants buffer_heads. | |
1204 | */ | |
1cf6e7d8 | 1205 | int set_page_dirty(struct page *page) |
1da177e4 LT |
1206 | { |
1207 | struct address_space *mapping = page_mapping(page); | |
1208 | ||
1209 | if (likely(mapping)) { | |
1210 | int (*spd)(struct page *) = mapping->a_ops->set_page_dirty; | |
9361401e DH |
1211 | #ifdef CONFIG_BLOCK |
1212 | if (!spd) | |
1213 | spd = __set_page_dirty_buffers; | |
1214 | #endif | |
1215 | return (*spd)(page); | |
1da177e4 | 1216 | } |
4741c9fd AM |
1217 | if (!PageDirty(page)) { |
1218 | if (!TestSetPageDirty(page)) | |
1219 | return 1; | |
1220 | } | |
1da177e4 LT |
1221 | return 0; |
1222 | } | |
1223 | EXPORT_SYMBOL(set_page_dirty); | |
1224 | ||
1225 | /* | |
1226 | * set_page_dirty() is racy if the caller has no reference against | |
1227 | * page->mapping->host, and if the page is unlocked. This is because another | |
1228 | * CPU could truncate the page off the mapping and then free the mapping. | |
1229 | * | |
1230 | * Usually, the page _is_ locked, or the caller is a user-space process which | |
1231 | * holds a reference on the inode by having an open file. | |
1232 | * | |
1233 | * In other cases, the page should be locked before running set_page_dirty(). | |
1234 | */ | |
1235 | int set_page_dirty_lock(struct page *page) | |
1236 | { | |
1237 | int ret; | |
1238 | ||
db37648c | 1239 | lock_page_nosync(page); |
1da177e4 LT |
1240 | ret = set_page_dirty(page); |
1241 | unlock_page(page); | |
1242 | return ret; | |
1243 | } | |
1244 | EXPORT_SYMBOL(set_page_dirty_lock); | |
1245 | ||
1da177e4 LT |
1246 | /* |
1247 | * Clear a page's dirty flag, while caring for dirty memory accounting. | |
1248 | * Returns true if the page was previously dirty. | |
1249 | * | |
1250 | * This is for preparing to put the page under writeout. We leave the page | |
1251 | * tagged as dirty in the radix tree so that a concurrent write-for-sync | |
1252 | * can discover it via a PAGECACHE_TAG_DIRTY walk. The ->writepage | |
1253 | * implementation will run either set_page_writeback() or set_page_dirty(), | |
1254 | * at which stage we bring the page's dirty flag and radix-tree dirty tag | |
1255 | * back into sync. | |
1256 | * | |
1257 | * This incoherency between the page's dirty flag and radix-tree tag is | |
1258 | * unfortunate, but it only exists while the page is locked. | |
1259 | */ | |
1260 | int clear_page_dirty_for_io(struct page *page) | |
1261 | { | |
1262 | struct address_space *mapping = page_mapping(page); | |
1263 | ||
79352894 NP |
1264 | BUG_ON(!PageLocked(page)); |
1265 | ||
fe3cba17 | 1266 | ClearPageReclaim(page); |
7658cc28 LT |
1267 | if (mapping && mapping_cap_account_dirty(mapping)) { |
1268 | /* | |
1269 | * Yes, Virginia, this is indeed insane. | |
1270 | * | |
1271 | * We use this sequence to make sure that | |
1272 | * (a) we account for dirty stats properly | |
1273 | * (b) we tell the low-level filesystem to | |
1274 | * mark the whole page dirty if it was | |
1275 | * dirty in a pagetable. Only to then | |
1276 | * (c) clean the page again and return 1 to | |
1277 | * cause the writeback. | |
1278 | * | |
1279 | * This way we avoid all nasty races with the | |
1280 | * dirty bit in multiple places and clearing | |
1281 | * them concurrently from different threads. | |
1282 | * | |
1283 | * Note! Normally the "set_page_dirty(page)" | |
1284 | * has no effect on the actual dirty bit - since | |
1285 | * that will already usually be set. But we | |
1286 | * need the side effects, and it can help us | |
1287 | * avoid races. | |
1288 | * | |
1289 | * We basically use the page "master dirty bit" | |
1290 | * as a serialization point for all the different | |
1291 | * threads doing their things. | |
7658cc28 LT |
1292 | */ |
1293 | if (page_mkclean(page)) | |
1294 | set_page_dirty(page); | |
79352894 NP |
1295 | /* |
1296 | * We carefully synchronise fault handlers against | |
1297 | * installing a dirty pte and marking the page dirty | |
1298 | * at this point. We do this by having them hold the | |
1299 | * page lock at some point after installing their | |
1300 | * pte, but before marking the page dirty. | |
1301 | * Pages are always locked coming in here, so we get | |
1302 | * the desired exclusion. See mm/memory.c:do_wp_page() | |
1303 | * for more comments. | |
1304 | */ | |
7658cc28 | 1305 | if (TestClearPageDirty(page)) { |
8c08540f | 1306 | dec_zone_page_state(page, NR_FILE_DIRTY); |
c9e51e41 PZ |
1307 | dec_bdi_stat(mapping->backing_dev_info, |
1308 | BDI_RECLAIMABLE); | |
7658cc28 | 1309 | return 1; |
1da177e4 | 1310 | } |
7658cc28 | 1311 | return 0; |
1da177e4 | 1312 | } |
7658cc28 | 1313 | return TestClearPageDirty(page); |
1da177e4 | 1314 | } |
58bb01a9 | 1315 | EXPORT_SYMBOL(clear_page_dirty_for_io); |
1da177e4 LT |
1316 | |
1317 | int test_clear_page_writeback(struct page *page) | |
1318 | { | |
1319 | struct address_space *mapping = page_mapping(page); | |
1320 | int ret; | |
1321 | ||
1322 | if (mapping) { | |
69cb51d1 | 1323 | struct backing_dev_info *bdi = mapping->backing_dev_info; |
1da177e4 LT |
1324 | unsigned long flags; |
1325 | ||
19fd6231 | 1326 | spin_lock_irqsave(&mapping->tree_lock, flags); |
1da177e4 | 1327 | ret = TestClearPageWriteback(page); |
69cb51d1 | 1328 | if (ret) { |
1da177e4 LT |
1329 | radix_tree_tag_clear(&mapping->page_tree, |
1330 | page_index(page), | |
1331 | PAGECACHE_TAG_WRITEBACK); | |
e4ad08fe | 1332 | if (bdi_cap_account_writeback(bdi)) { |
69cb51d1 | 1333 | __dec_bdi_stat(bdi, BDI_WRITEBACK); |
04fbfdc1 PZ |
1334 | __bdi_writeout_inc(bdi); |
1335 | } | |
69cb51d1 | 1336 | } |
19fd6231 | 1337 | spin_unlock_irqrestore(&mapping->tree_lock, flags); |
1da177e4 LT |
1338 | } else { |
1339 | ret = TestClearPageWriteback(page); | |
1340 | } | |
d688abf5 AM |
1341 | if (ret) |
1342 | dec_zone_page_state(page, NR_WRITEBACK); | |
1da177e4 LT |
1343 | return ret; |
1344 | } | |
1345 | ||
1346 | int test_set_page_writeback(struct page *page) | |
1347 | { | |
1348 | struct address_space *mapping = page_mapping(page); | |
1349 | int ret; | |
1350 | ||
1351 | if (mapping) { | |
69cb51d1 | 1352 | struct backing_dev_info *bdi = mapping->backing_dev_info; |
1da177e4 LT |
1353 | unsigned long flags; |
1354 | ||
19fd6231 | 1355 | spin_lock_irqsave(&mapping->tree_lock, flags); |
1da177e4 | 1356 | ret = TestSetPageWriteback(page); |
69cb51d1 | 1357 | if (!ret) { |
1da177e4 LT |
1358 | radix_tree_tag_set(&mapping->page_tree, |
1359 | page_index(page), | |
1360 | PAGECACHE_TAG_WRITEBACK); | |
e4ad08fe | 1361 | if (bdi_cap_account_writeback(bdi)) |
69cb51d1 PZ |
1362 | __inc_bdi_stat(bdi, BDI_WRITEBACK); |
1363 | } | |
1da177e4 LT |
1364 | if (!PageDirty(page)) |
1365 | radix_tree_tag_clear(&mapping->page_tree, | |
1366 | page_index(page), | |
1367 | PAGECACHE_TAG_DIRTY); | |
f446daae JK |
1368 | radix_tree_tag_clear(&mapping->page_tree, |
1369 | page_index(page), | |
1370 | PAGECACHE_TAG_TOWRITE); | |
19fd6231 | 1371 | spin_unlock_irqrestore(&mapping->tree_lock, flags); |
1da177e4 LT |
1372 | } else { |
1373 | ret = TestSetPageWriteback(page); | |
1374 | } | |
d688abf5 AM |
1375 | if (!ret) |
1376 | inc_zone_page_state(page, NR_WRITEBACK); | |
1da177e4 LT |
1377 | return ret; |
1378 | ||
1379 | } | |
1380 | EXPORT_SYMBOL(test_set_page_writeback); | |
1381 | ||
1382 | /* | |
00128188 | 1383 | * Return true if any of the pages in the mapping are marked with the |
1da177e4 LT |
1384 | * passed tag. |
1385 | */ | |
1386 | int mapping_tagged(struct address_space *mapping, int tag) | |
1387 | { | |
1da177e4 | 1388 | int ret; |
00128188 | 1389 | rcu_read_lock(); |
1da177e4 | 1390 | ret = radix_tree_tagged(&mapping->page_tree, tag); |
00128188 | 1391 | rcu_read_unlock(); |
1da177e4 LT |
1392 | return ret; |
1393 | } | |
1394 | EXPORT_SYMBOL(mapping_tagged); |