]>
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 | * | |
10 | * 10Apr2002 akpm@zip.com.au | |
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 LT |
37 | |
38 | /* | |
39 | * The maximum number of pages to writeout in a single bdflush/kupdate | |
1c0eeaf5 | 40 | * operation. We do this so we don't hold I_SYNC against an inode for |
1da177e4 LT |
41 | * enormous amounts of time, which would block a userspace task which has |
42 | * been forced to throttle against that inode. Also, the code reevaluates | |
43 | * the dirty each time it has written this many pages. | |
44 | */ | |
45 | #define MAX_WRITEBACK_PAGES 1024 | |
46 | ||
47 | /* | |
48 | * After a CPU has dirtied this many pages, balance_dirty_pages_ratelimited | |
49 | * will look to see if it needs to force writeback or throttling. | |
50 | */ | |
51 | static long ratelimit_pages = 32; | |
52 | ||
1da177e4 LT |
53 | /* |
54 | * When balance_dirty_pages decides that the caller needs to perform some | |
55 | * non-background writeback, this is how many pages it will attempt to write. | |
56 | * It should be somewhat larger than RATELIMIT_PAGES to ensure that reasonably | |
57 | * large amounts of I/O are submitted. | |
58 | */ | |
59 | static inline long sync_writeback_pages(void) | |
60 | { | |
61 | return ratelimit_pages + ratelimit_pages / 2; | |
62 | } | |
63 | ||
64 | /* The following parameters are exported via /proc/sys/vm */ | |
65 | ||
66 | /* | |
67 | * Start background writeback (via pdflush) at this percentage | |
68 | */ | |
07db59bd | 69 | int dirty_background_ratio = 5; |
1da177e4 | 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 | */ | |
07db59bd | 80 | int vm_dirty_ratio = 10; |
1da177e4 LT |
81 | |
82 | /* | |
fd5403c7 | 83 | * The interval between `kupdate'-style writebacks, in jiffies |
1da177e4 | 84 | */ |
f6ef9438 | 85 | int dirty_writeback_interval = 5 * HZ; |
1da177e4 LT |
86 | |
87 | /* | |
fd5403c7 | 88 | * The longest number of jiffies for which data is allowed to remain dirty |
1da177e4 | 89 | */ |
f6ef9438 | 90 | int dirty_expire_interval = 30 * HZ; |
1da177e4 LT |
91 | |
92 | /* | |
93 | * Flag that makes the machine dump writes/reads and block dirtyings. | |
94 | */ | |
95 | int block_dump; | |
96 | ||
97 | /* | |
ed5b43f1 BS |
98 | * Flag that puts the machine in "laptop mode". Doubles as a timeout in jiffies: |
99 | * a full sync is triggered after this time elapses without any disk activity. | |
1da177e4 LT |
100 | */ |
101 | int laptop_mode; | |
102 | ||
103 | EXPORT_SYMBOL(laptop_mode); | |
104 | ||
105 | /* End of sysctl-exported parameters */ | |
106 | ||
107 | ||
108 | static void background_writeout(unsigned long _min_pages); | |
109 | ||
04fbfdc1 PZ |
110 | /* |
111 | * Scale the writeback cache size proportional to the relative writeout speeds. | |
112 | * | |
113 | * We do this by keeping a floating proportion between BDIs, based on page | |
114 | * writeback completions [end_page_writeback()]. Those devices that write out | |
115 | * pages fastest will get the larger share, while the slower will get a smaller | |
116 | * share. | |
117 | * | |
118 | * We use page writeout completions because we are interested in getting rid of | |
119 | * dirty pages. Having them written out is the primary goal. | |
120 | * | |
121 | * We introduce a concept of time, a period over which we measure these events, | |
122 | * because demand can/will vary over time. The length of this period itself is | |
123 | * measured in page writeback completions. | |
124 | * | |
125 | */ | |
126 | static struct prop_descriptor vm_completions; | |
3e26c149 | 127 | static struct prop_descriptor vm_dirties; |
04fbfdc1 PZ |
128 | |
129 | static unsigned long determine_dirtyable_memory(void); | |
130 | ||
131 | /* | |
132 | * couple the period to the dirty_ratio: | |
133 | * | |
134 | * period/2 ~ roundup_pow_of_two(dirty limit) | |
135 | */ | |
136 | static int calc_period_shift(void) | |
137 | { | |
138 | unsigned long dirty_total; | |
139 | ||
140 | dirty_total = (vm_dirty_ratio * determine_dirtyable_memory()) / 100; | |
141 | return 2 + ilog2(dirty_total - 1); | |
142 | } | |
143 | ||
144 | /* | |
145 | * update the period when the dirty ratio changes. | |
146 | */ | |
147 | int dirty_ratio_handler(struct ctl_table *table, int write, | |
148 | struct file *filp, void __user *buffer, size_t *lenp, | |
149 | loff_t *ppos) | |
150 | { | |
151 | int old_ratio = vm_dirty_ratio; | |
152 | int ret = proc_dointvec_minmax(table, write, filp, buffer, lenp, ppos); | |
153 | if (ret == 0 && write && vm_dirty_ratio != old_ratio) { | |
154 | int shift = calc_period_shift(); | |
155 | prop_change_shift(&vm_completions, shift); | |
3e26c149 | 156 | prop_change_shift(&vm_dirties, shift); |
04fbfdc1 PZ |
157 | } |
158 | return ret; | |
159 | } | |
160 | ||
161 | /* | |
162 | * Increment the BDI's writeout completion count and the global writeout | |
163 | * completion count. Called from test_clear_page_writeback(). | |
164 | */ | |
165 | static inline void __bdi_writeout_inc(struct backing_dev_info *bdi) | |
166 | { | |
a42dde04 PZ |
167 | __prop_inc_percpu_max(&vm_completions, &bdi->completions, |
168 | bdi->max_prop_frac); | |
04fbfdc1 PZ |
169 | } |
170 | ||
3e26c149 PZ |
171 | static inline void task_dirty_inc(struct task_struct *tsk) |
172 | { | |
173 | prop_inc_single(&vm_dirties, &tsk->dirties); | |
174 | } | |
175 | ||
04fbfdc1 PZ |
176 | /* |
177 | * Obtain an accurate fraction of the BDI's portion. | |
178 | */ | |
179 | static void bdi_writeout_fraction(struct backing_dev_info *bdi, | |
180 | long *numerator, long *denominator) | |
181 | { | |
182 | if (bdi_cap_writeback_dirty(bdi)) { | |
183 | prop_fraction_percpu(&vm_completions, &bdi->completions, | |
184 | numerator, denominator); | |
185 | } else { | |
186 | *numerator = 0; | |
187 | *denominator = 1; | |
188 | } | |
189 | } | |
190 | ||
191 | /* | |
192 | * Clip the earned share of dirty pages to that which is actually available. | |
193 | * This avoids exceeding the total dirty_limit when the floating averages | |
194 | * fluctuate too quickly. | |
195 | */ | |
196 | static void | |
197 | clip_bdi_dirty_limit(struct backing_dev_info *bdi, long dirty, long *pbdi_dirty) | |
198 | { | |
199 | long avail_dirty; | |
200 | ||
201 | avail_dirty = dirty - | |
202 | (global_page_state(NR_FILE_DIRTY) + | |
203 | global_page_state(NR_WRITEBACK) + | |
204 | global_page_state(NR_UNSTABLE_NFS)); | |
205 | ||
206 | if (avail_dirty < 0) | |
207 | avail_dirty = 0; | |
208 | ||
209 | avail_dirty += bdi_stat(bdi, BDI_RECLAIMABLE) + | |
210 | bdi_stat(bdi, BDI_WRITEBACK); | |
211 | ||
212 | *pbdi_dirty = min(*pbdi_dirty, avail_dirty); | |
213 | } | |
214 | ||
3e26c149 PZ |
215 | static inline void task_dirties_fraction(struct task_struct *tsk, |
216 | long *numerator, long *denominator) | |
217 | { | |
218 | prop_fraction_single(&vm_dirties, &tsk->dirties, | |
219 | numerator, denominator); | |
220 | } | |
221 | ||
222 | /* | |
223 | * scale the dirty limit | |
224 | * | |
225 | * task specific dirty limit: | |
226 | * | |
227 | * dirty -= (dirty/8) * p_{t} | |
228 | */ | |
f61eaf9f | 229 | static void task_dirty_limit(struct task_struct *tsk, long *pdirty) |
3e26c149 PZ |
230 | { |
231 | long numerator, denominator; | |
232 | long dirty = *pdirty; | |
233 | u64 inv = dirty >> 3; | |
234 | ||
235 | task_dirties_fraction(tsk, &numerator, &denominator); | |
236 | inv *= numerator; | |
237 | do_div(inv, denominator); | |
238 | ||
239 | dirty -= inv; | |
240 | if (dirty < *pdirty/2) | |
241 | dirty = *pdirty/2; | |
242 | ||
243 | *pdirty = dirty; | |
244 | } | |
245 | ||
189d3c4a PZ |
246 | /* |
247 | * | |
248 | */ | |
249 | static DEFINE_SPINLOCK(bdi_lock); | |
250 | static unsigned int bdi_min_ratio; | |
251 | ||
252 | int bdi_set_min_ratio(struct backing_dev_info *bdi, unsigned int min_ratio) | |
253 | { | |
254 | int ret = 0; | |
255 | unsigned long flags; | |
256 | ||
257 | spin_lock_irqsave(&bdi_lock, flags); | |
a42dde04 | 258 | if (min_ratio > bdi->max_ratio) { |
189d3c4a | 259 | ret = -EINVAL; |
a42dde04 PZ |
260 | } else { |
261 | min_ratio -= bdi->min_ratio; | |
262 | if (bdi_min_ratio + min_ratio < 100) { | |
263 | bdi_min_ratio += min_ratio; | |
264 | bdi->min_ratio += min_ratio; | |
265 | } else { | |
266 | ret = -EINVAL; | |
267 | } | |
268 | } | |
269 | spin_unlock_irqrestore(&bdi_lock, flags); | |
270 | ||
271 | return ret; | |
272 | } | |
273 | ||
274 | int bdi_set_max_ratio(struct backing_dev_info *bdi, unsigned max_ratio) | |
275 | { | |
276 | unsigned long flags; | |
277 | int ret = 0; | |
278 | ||
279 | if (max_ratio > 100) | |
280 | return -EINVAL; | |
281 | ||
282 | spin_lock_irqsave(&bdi_lock, flags); | |
283 | if (bdi->min_ratio > max_ratio) { | |
284 | ret = -EINVAL; | |
285 | } else { | |
286 | bdi->max_ratio = max_ratio; | |
287 | bdi->max_prop_frac = (PROP_FRAC_BASE * max_ratio) / 100; | |
288 | } | |
189d3c4a PZ |
289 | spin_unlock_irqrestore(&bdi_lock, flags); |
290 | ||
291 | return ret; | |
292 | } | |
a42dde04 | 293 | EXPORT_SYMBOL(bdi_set_max_ratio); |
189d3c4a | 294 | |
1da177e4 LT |
295 | /* |
296 | * Work out the current dirty-memory clamping and background writeout | |
297 | * thresholds. | |
298 | * | |
299 | * The main aim here is to lower them aggressively if there is a lot of mapped | |
300 | * memory around. To avoid stressing page reclaim with lots of unreclaimable | |
301 | * pages. It is better to clamp down on writers than to start swapping, and | |
302 | * performing lots of scanning. | |
303 | * | |
304 | * We only allow 1/2 of the currently-unmapped memory to be dirtied. | |
305 | * | |
306 | * We don't permit the clamping level to fall below 5% - that is getting rather | |
307 | * excessive. | |
308 | * | |
309 | * We make sure that the background writeout level is below the adjusted | |
310 | * clamping level. | |
311 | */ | |
1b424464 CL |
312 | |
313 | static unsigned long highmem_dirtyable_memory(unsigned long total) | |
314 | { | |
315 | #ifdef CONFIG_HIGHMEM | |
316 | int node; | |
317 | unsigned long x = 0; | |
318 | ||
37b07e41 | 319 | for_each_node_state(node, N_HIGH_MEMORY) { |
1b424464 CL |
320 | struct zone *z = |
321 | &NODE_DATA(node)->node_zones[ZONE_HIGHMEM]; | |
322 | ||
323 | x += zone_page_state(z, NR_FREE_PAGES) | |
324 | + zone_page_state(z, NR_INACTIVE) | |
325 | + zone_page_state(z, NR_ACTIVE); | |
326 | } | |
327 | /* | |
328 | * Make sure that the number of highmem pages is never larger | |
329 | * than the number of the total dirtyable memory. This can only | |
330 | * occur in very strange VM situations but we want to make sure | |
331 | * that this does not occur. | |
332 | */ | |
333 | return min(x, total); | |
334 | #else | |
335 | return 0; | |
336 | #endif | |
337 | } | |
338 | ||
339 | static unsigned long determine_dirtyable_memory(void) | |
340 | { | |
341 | unsigned long x; | |
342 | ||
343 | x = global_page_state(NR_FREE_PAGES) | |
344 | + global_page_state(NR_INACTIVE) | |
345 | + global_page_state(NR_ACTIVE); | |
195cf453 BG |
346 | |
347 | if (!vm_highmem_is_dirtyable) | |
348 | x -= highmem_dirtyable_memory(x); | |
349 | ||
1b424464 CL |
350 | return x + 1; /* Ensure that we never return 0 */ |
351 | } | |
352 | ||
cf0ca9fe | 353 | void |
04fbfdc1 PZ |
354 | get_dirty_limits(long *pbackground, long *pdirty, long *pbdi_dirty, |
355 | struct backing_dev_info *bdi) | |
1da177e4 LT |
356 | { |
357 | int background_ratio; /* Percentages */ | |
358 | int dirty_ratio; | |
1da177e4 LT |
359 | long background; |
360 | long dirty; | |
1b424464 | 361 | unsigned long available_memory = determine_dirtyable_memory(); |
1da177e4 LT |
362 | struct task_struct *tsk; |
363 | ||
1da177e4 | 364 | dirty_ratio = vm_dirty_ratio; |
1da177e4 LT |
365 | if (dirty_ratio < 5) |
366 | dirty_ratio = 5; | |
367 | ||
368 | background_ratio = dirty_background_ratio; | |
369 | if (background_ratio >= dirty_ratio) | |
370 | background_ratio = dirty_ratio / 2; | |
371 | ||
372 | background = (background_ratio * available_memory) / 100; | |
373 | dirty = (dirty_ratio * available_memory) / 100; | |
374 | tsk = current; | |
375 | if (tsk->flags & PF_LESS_THROTTLE || rt_task(tsk)) { | |
376 | background += background / 4; | |
377 | dirty += dirty / 4; | |
378 | } | |
379 | *pbackground = background; | |
380 | *pdirty = dirty; | |
04fbfdc1 PZ |
381 | |
382 | if (bdi) { | |
189d3c4a | 383 | u64 bdi_dirty; |
04fbfdc1 PZ |
384 | long numerator, denominator; |
385 | ||
386 | /* | |
387 | * Calculate this BDI's share of the dirty ratio. | |
388 | */ | |
389 | bdi_writeout_fraction(bdi, &numerator, &denominator); | |
390 | ||
189d3c4a | 391 | bdi_dirty = (dirty * (100 - bdi_min_ratio)) / 100; |
04fbfdc1 PZ |
392 | bdi_dirty *= numerator; |
393 | do_div(bdi_dirty, denominator); | |
189d3c4a | 394 | bdi_dirty += (dirty * bdi->min_ratio) / 100; |
a42dde04 PZ |
395 | if (bdi_dirty > (dirty * bdi->max_ratio) / 100) |
396 | bdi_dirty = dirty * bdi->max_ratio / 100; | |
04fbfdc1 PZ |
397 | |
398 | *pbdi_dirty = bdi_dirty; | |
399 | clip_bdi_dirty_limit(bdi, dirty, pbdi_dirty); | |
3e26c149 | 400 | task_dirty_limit(current, pbdi_dirty); |
04fbfdc1 | 401 | } |
1da177e4 LT |
402 | } |
403 | ||
404 | /* | |
405 | * balance_dirty_pages() must be called by processes which are generating dirty | |
406 | * data. It looks at the number of dirty pages in the machine and will force | |
407 | * the caller to perform writeback if the system is over `vm_dirty_ratio'. | |
408 | * If we're over `background_thresh' then pdflush is woken to perform some | |
409 | * writeout. | |
410 | */ | |
411 | static void balance_dirty_pages(struct address_space *mapping) | |
412 | { | |
5fce25a9 PZ |
413 | long nr_reclaimable, bdi_nr_reclaimable; |
414 | long nr_writeback, bdi_nr_writeback; | |
1da177e4 LT |
415 | long background_thresh; |
416 | long dirty_thresh; | |
04fbfdc1 | 417 | long bdi_thresh; |
1da177e4 LT |
418 | unsigned long pages_written = 0; |
419 | unsigned long write_chunk = sync_writeback_pages(); | |
420 | ||
421 | struct backing_dev_info *bdi = mapping->backing_dev_info; | |
422 | ||
423 | for (;;) { | |
424 | struct writeback_control wbc = { | |
425 | .bdi = bdi, | |
426 | .sync_mode = WB_SYNC_NONE, | |
427 | .older_than_this = NULL, | |
428 | .nr_to_write = write_chunk, | |
111ebb6e | 429 | .range_cyclic = 1, |
1da177e4 LT |
430 | }; |
431 | ||
04fbfdc1 PZ |
432 | get_dirty_limits(&background_thresh, &dirty_thresh, |
433 | &bdi_thresh, bdi); | |
5fce25a9 PZ |
434 | |
435 | nr_reclaimable = global_page_state(NR_FILE_DIRTY) + | |
436 | global_page_state(NR_UNSTABLE_NFS); | |
437 | nr_writeback = global_page_state(NR_WRITEBACK); | |
438 | ||
04fbfdc1 PZ |
439 | bdi_nr_reclaimable = bdi_stat(bdi, BDI_RECLAIMABLE); |
440 | bdi_nr_writeback = bdi_stat(bdi, BDI_WRITEBACK); | |
5fce25a9 | 441 | |
04fbfdc1 PZ |
442 | if (bdi_nr_reclaimable + bdi_nr_writeback <= bdi_thresh) |
443 | break; | |
1da177e4 | 444 | |
5fce25a9 PZ |
445 | /* |
446 | * Throttle it only when the background writeback cannot | |
447 | * catch-up. This avoids (excessively) small writeouts | |
448 | * when the bdi limits are ramping up. | |
449 | */ | |
450 | if (nr_reclaimable + nr_writeback < | |
451 | (background_thresh + dirty_thresh) / 2) | |
452 | break; | |
453 | ||
04fbfdc1 PZ |
454 | if (!bdi->dirty_exceeded) |
455 | bdi->dirty_exceeded = 1; | |
1da177e4 LT |
456 | |
457 | /* Note: nr_reclaimable denotes nr_dirty + nr_unstable. | |
458 | * Unstable writes are a feature of certain networked | |
459 | * filesystems (i.e. NFS) in which data may have been | |
460 | * written to the server's write cache, but has not yet | |
461 | * been flushed to permanent storage. | |
462 | */ | |
04fbfdc1 | 463 | if (bdi_nr_reclaimable) { |
1da177e4 | 464 | writeback_inodes(&wbc); |
1da177e4 | 465 | pages_written += write_chunk - wbc.nr_to_write; |
04fbfdc1 PZ |
466 | get_dirty_limits(&background_thresh, &dirty_thresh, |
467 | &bdi_thresh, bdi); | |
468 | } | |
469 | ||
470 | /* | |
471 | * In order to avoid the stacked BDI deadlock we need | |
472 | * to ensure we accurately count the 'dirty' pages when | |
473 | * the threshold is low. | |
474 | * | |
475 | * Otherwise it would be possible to get thresh+n pages | |
476 | * reported dirty, even though there are thresh-m pages | |
477 | * actually dirty; with m+n sitting in the percpu | |
478 | * deltas. | |
479 | */ | |
480 | if (bdi_thresh < 2*bdi_stat_error(bdi)) { | |
481 | bdi_nr_reclaimable = bdi_stat_sum(bdi, BDI_RECLAIMABLE); | |
482 | bdi_nr_writeback = bdi_stat_sum(bdi, BDI_WRITEBACK); | |
483 | } else if (bdi_nr_reclaimable) { | |
484 | bdi_nr_reclaimable = bdi_stat(bdi, BDI_RECLAIMABLE); | |
485 | bdi_nr_writeback = bdi_stat(bdi, BDI_WRITEBACK); | |
1da177e4 | 486 | } |
04fbfdc1 PZ |
487 | |
488 | if (bdi_nr_reclaimable + bdi_nr_writeback <= bdi_thresh) | |
489 | break; | |
490 | if (pages_written >= write_chunk) | |
491 | break; /* We've done our duty */ | |
492 | ||
3fcfab16 | 493 | congestion_wait(WRITE, HZ/10); |
1da177e4 LT |
494 | } |
495 | ||
04fbfdc1 PZ |
496 | if (bdi_nr_reclaimable + bdi_nr_writeback < bdi_thresh && |
497 | bdi->dirty_exceeded) | |
498 | bdi->dirty_exceeded = 0; | |
1da177e4 LT |
499 | |
500 | if (writeback_in_progress(bdi)) | |
501 | return; /* pdflush is already working this queue */ | |
502 | ||
503 | /* | |
504 | * In laptop mode, we wait until hitting the higher threshold before | |
505 | * starting background writeout, and then write out all the way down | |
506 | * to the lower threshold. So slow writers cause minimal disk activity. | |
507 | * | |
508 | * In normal mode, we start background writeout at the lower | |
509 | * background_thresh, to keep the amount of dirty memory low. | |
510 | */ | |
511 | if ((laptop_mode && pages_written) || | |
04fbfdc1 PZ |
512 | (!laptop_mode && (global_page_state(NR_FILE_DIRTY) |
513 | + global_page_state(NR_UNSTABLE_NFS) | |
514 | > background_thresh))) | |
1da177e4 LT |
515 | pdflush_operation(background_writeout, 0); |
516 | } | |
517 | ||
a200ee18 | 518 | void set_page_dirty_balance(struct page *page, int page_mkwrite) |
edc79b2a | 519 | { |
a200ee18 | 520 | if (set_page_dirty(page) || page_mkwrite) { |
edc79b2a PZ |
521 | struct address_space *mapping = page_mapping(page); |
522 | ||
523 | if (mapping) | |
524 | balance_dirty_pages_ratelimited(mapping); | |
525 | } | |
526 | } | |
527 | ||
1da177e4 | 528 | /** |
fa5a734e | 529 | * balance_dirty_pages_ratelimited_nr - balance dirty memory state |
67be2dd1 | 530 | * @mapping: address_space which was dirtied |
a580290c | 531 | * @nr_pages_dirtied: number of pages which the caller has just dirtied |
1da177e4 LT |
532 | * |
533 | * Processes which are dirtying memory should call in here once for each page | |
534 | * which was newly dirtied. The function will periodically check the system's | |
535 | * dirty state and will initiate writeback if needed. | |
536 | * | |
537 | * On really big machines, get_writeback_state is expensive, so try to avoid | |
538 | * calling it too often (ratelimiting). But once we're over the dirty memory | |
539 | * limit we decrease the ratelimiting by a lot, to prevent individual processes | |
540 | * from overshooting the limit by (ratelimit_pages) each. | |
541 | */ | |
fa5a734e AM |
542 | void balance_dirty_pages_ratelimited_nr(struct address_space *mapping, |
543 | unsigned long nr_pages_dirtied) | |
1da177e4 | 544 | { |
fa5a734e AM |
545 | static DEFINE_PER_CPU(unsigned long, ratelimits) = 0; |
546 | unsigned long ratelimit; | |
547 | unsigned long *p; | |
1da177e4 LT |
548 | |
549 | ratelimit = ratelimit_pages; | |
04fbfdc1 | 550 | if (mapping->backing_dev_info->dirty_exceeded) |
1da177e4 LT |
551 | ratelimit = 8; |
552 | ||
553 | /* | |
554 | * Check the rate limiting. Also, we do not want to throttle real-time | |
555 | * tasks in balance_dirty_pages(). Period. | |
556 | */ | |
fa5a734e AM |
557 | preempt_disable(); |
558 | p = &__get_cpu_var(ratelimits); | |
559 | *p += nr_pages_dirtied; | |
560 | if (unlikely(*p >= ratelimit)) { | |
561 | *p = 0; | |
562 | preempt_enable(); | |
1da177e4 LT |
563 | balance_dirty_pages(mapping); |
564 | return; | |
565 | } | |
fa5a734e | 566 | preempt_enable(); |
1da177e4 | 567 | } |
fa5a734e | 568 | EXPORT_SYMBOL(balance_dirty_pages_ratelimited_nr); |
1da177e4 | 569 | |
232ea4d6 | 570 | void throttle_vm_writeout(gfp_t gfp_mask) |
1da177e4 | 571 | { |
1da177e4 LT |
572 | long background_thresh; |
573 | long dirty_thresh; | |
574 | ||
575 | for ( ; ; ) { | |
04fbfdc1 | 576 | get_dirty_limits(&background_thresh, &dirty_thresh, NULL, NULL); |
1da177e4 LT |
577 | |
578 | /* | |
579 | * Boost the allowable dirty threshold a bit for page | |
580 | * allocators so they don't get DoS'ed by heavy writers | |
581 | */ | |
582 | dirty_thresh += dirty_thresh / 10; /* wheeee... */ | |
583 | ||
c24f21bd CL |
584 | if (global_page_state(NR_UNSTABLE_NFS) + |
585 | global_page_state(NR_WRITEBACK) <= dirty_thresh) | |
586 | break; | |
3fcfab16 | 587 | congestion_wait(WRITE, HZ/10); |
369f2389 FW |
588 | |
589 | /* | |
590 | * The caller might hold locks which can prevent IO completion | |
591 | * or progress in the filesystem. So we cannot just sit here | |
592 | * waiting for IO to complete. | |
593 | */ | |
594 | if ((gfp_mask & (__GFP_FS|__GFP_IO)) != (__GFP_FS|__GFP_IO)) | |
595 | break; | |
1da177e4 LT |
596 | } |
597 | } | |
598 | ||
1da177e4 LT |
599 | /* |
600 | * writeback at least _min_pages, and keep writing until the amount of dirty | |
601 | * memory is less than the background threshold, or until we're all clean. | |
602 | */ | |
603 | static void background_writeout(unsigned long _min_pages) | |
604 | { | |
605 | long min_pages = _min_pages; | |
606 | struct writeback_control wbc = { | |
607 | .bdi = NULL, | |
608 | .sync_mode = WB_SYNC_NONE, | |
609 | .older_than_this = NULL, | |
610 | .nr_to_write = 0, | |
611 | .nonblocking = 1, | |
111ebb6e | 612 | .range_cyclic = 1, |
1da177e4 LT |
613 | }; |
614 | ||
615 | for ( ; ; ) { | |
1da177e4 LT |
616 | long background_thresh; |
617 | long dirty_thresh; | |
618 | ||
04fbfdc1 | 619 | get_dirty_limits(&background_thresh, &dirty_thresh, NULL, NULL); |
c24f21bd CL |
620 | if (global_page_state(NR_FILE_DIRTY) + |
621 | global_page_state(NR_UNSTABLE_NFS) < background_thresh | |
1da177e4 LT |
622 | && min_pages <= 0) |
623 | break; | |
8bc3be27 | 624 | wbc.more_io = 0; |
1da177e4 LT |
625 | wbc.encountered_congestion = 0; |
626 | wbc.nr_to_write = MAX_WRITEBACK_PAGES; | |
627 | wbc.pages_skipped = 0; | |
628 | writeback_inodes(&wbc); | |
629 | min_pages -= MAX_WRITEBACK_PAGES - wbc.nr_to_write; | |
630 | if (wbc.nr_to_write > 0 || wbc.pages_skipped > 0) { | |
631 | /* Wrote less than expected */ | |
8bc3be27 FW |
632 | if (wbc.encountered_congestion || wbc.more_io) |
633 | congestion_wait(WRITE, HZ/10); | |
634 | else | |
1da177e4 LT |
635 | break; |
636 | } | |
637 | } | |
638 | } | |
639 | ||
640 | /* | |
641 | * Start writeback of `nr_pages' pages. If `nr_pages' is zero, write back | |
642 | * the whole world. Returns 0 if a pdflush thread was dispatched. Returns | |
643 | * -1 if all pdflush threads were busy. | |
644 | */ | |
687a21ce | 645 | int wakeup_pdflush(long nr_pages) |
1da177e4 | 646 | { |
c24f21bd CL |
647 | if (nr_pages == 0) |
648 | nr_pages = global_page_state(NR_FILE_DIRTY) + | |
649 | global_page_state(NR_UNSTABLE_NFS); | |
1da177e4 LT |
650 | return pdflush_operation(background_writeout, nr_pages); |
651 | } | |
652 | ||
653 | static void wb_timer_fn(unsigned long unused); | |
654 | static void laptop_timer_fn(unsigned long unused); | |
655 | ||
8d06afab IM |
656 | static DEFINE_TIMER(wb_timer, wb_timer_fn, 0, 0); |
657 | static DEFINE_TIMER(laptop_mode_wb_timer, laptop_timer_fn, 0, 0); | |
1da177e4 LT |
658 | |
659 | /* | |
660 | * Periodic writeback of "old" data. | |
661 | * | |
662 | * Define "old": the first time one of an inode's pages is dirtied, we mark the | |
663 | * dirtying-time in the inode's address_space. So this periodic writeback code | |
664 | * just walks the superblock inode list, writing back any inodes which are | |
665 | * older than a specific point in time. | |
666 | * | |
f6ef9438 BS |
667 | * Try to run once per dirty_writeback_interval. But if a writeback event |
668 | * takes longer than a dirty_writeback_interval interval, then leave a | |
1da177e4 LT |
669 | * one-second gap. |
670 | * | |
671 | * older_than_this takes precedence over nr_to_write. So we'll only write back | |
672 | * all dirty pages if they are all attached to "old" mappings. | |
673 | */ | |
674 | static void wb_kupdate(unsigned long arg) | |
675 | { | |
676 | unsigned long oldest_jif; | |
677 | unsigned long start_jif; | |
678 | unsigned long next_jif; | |
679 | long nr_to_write; | |
1da177e4 LT |
680 | struct writeback_control wbc = { |
681 | .bdi = NULL, | |
682 | .sync_mode = WB_SYNC_NONE, | |
683 | .older_than_this = &oldest_jif, | |
684 | .nr_to_write = 0, | |
685 | .nonblocking = 1, | |
686 | .for_kupdate = 1, | |
111ebb6e | 687 | .range_cyclic = 1, |
1da177e4 LT |
688 | }; |
689 | ||
690 | sync_supers(); | |
691 | ||
f6ef9438 | 692 | oldest_jif = jiffies - dirty_expire_interval; |
1da177e4 | 693 | start_jif = jiffies; |
f6ef9438 | 694 | next_jif = start_jif + dirty_writeback_interval; |
c24f21bd CL |
695 | nr_to_write = global_page_state(NR_FILE_DIRTY) + |
696 | global_page_state(NR_UNSTABLE_NFS) + | |
1da177e4 LT |
697 | (inodes_stat.nr_inodes - inodes_stat.nr_unused); |
698 | while (nr_to_write > 0) { | |
8bc3be27 | 699 | wbc.more_io = 0; |
1da177e4 LT |
700 | wbc.encountered_congestion = 0; |
701 | wbc.nr_to_write = MAX_WRITEBACK_PAGES; | |
702 | writeback_inodes(&wbc); | |
703 | if (wbc.nr_to_write > 0) { | |
8bc3be27 | 704 | if (wbc.encountered_congestion || wbc.more_io) |
3fcfab16 | 705 | congestion_wait(WRITE, HZ/10); |
1da177e4 LT |
706 | else |
707 | break; /* All the old data is written */ | |
708 | } | |
709 | nr_to_write -= MAX_WRITEBACK_PAGES - wbc.nr_to_write; | |
710 | } | |
711 | if (time_before(next_jif, jiffies + HZ)) | |
712 | next_jif = jiffies + HZ; | |
f6ef9438 | 713 | if (dirty_writeback_interval) |
1da177e4 LT |
714 | mod_timer(&wb_timer, next_jif); |
715 | } | |
716 | ||
717 | /* | |
718 | * sysctl handler for /proc/sys/vm/dirty_writeback_centisecs | |
719 | */ | |
720 | int dirty_writeback_centisecs_handler(ctl_table *table, int write, | |
3e733f07 | 721 | struct file *file, void __user *buffer, size_t *length, loff_t *ppos) |
1da177e4 | 722 | { |
f6ef9438 | 723 | proc_dointvec_userhz_jiffies(table, write, file, buffer, length, ppos); |
3e733f07 AM |
724 | if (dirty_writeback_interval) |
725 | mod_timer(&wb_timer, jiffies + dirty_writeback_interval); | |
726 | else | |
1da177e4 | 727 | del_timer(&wb_timer); |
1da177e4 LT |
728 | return 0; |
729 | } | |
730 | ||
731 | static void wb_timer_fn(unsigned long unused) | |
732 | { | |
733 | if (pdflush_operation(wb_kupdate, 0) < 0) | |
734 | mod_timer(&wb_timer, jiffies + HZ); /* delay 1 second */ | |
735 | } | |
736 | ||
737 | static void laptop_flush(unsigned long unused) | |
738 | { | |
739 | sys_sync(); | |
740 | } | |
741 | ||
742 | static void laptop_timer_fn(unsigned long unused) | |
743 | { | |
744 | pdflush_operation(laptop_flush, 0); | |
745 | } | |
746 | ||
747 | /* | |
748 | * We've spun up the disk and we're in laptop mode: schedule writeback | |
749 | * of all dirty data a few seconds from now. If the flush is already scheduled | |
750 | * then push it back - the user is still using the disk. | |
751 | */ | |
752 | void laptop_io_completion(void) | |
753 | { | |
ed5b43f1 | 754 | mod_timer(&laptop_mode_wb_timer, jiffies + laptop_mode); |
1da177e4 LT |
755 | } |
756 | ||
757 | /* | |
758 | * We're in laptop mode and we've just synced. The sync's writes will have | |
759 | * caused another writeback to be scheduled by laptop_io_completion. | |
760 | * Nothing needs to be written back anymore, so we unschedule the writeback. | |
761 | */ | |
762 | void laptop_sync_completion(void) | |
763 | { | |
764 | del_timer(&laptop_mode_wb_timer); | |
765 | } | |
766 | ||
767 | /* | |
768 | * If ratelimit_pages is too high then we can get into dirty-data overload | |
769 | * if a large number of processes all perform writes at the same time. | |
770 | * If it is too low then SMP machines will call the (expensive) | |
771 | * get_writeback_state too often. | |
772 | * | |
773 | * Here we set ratelimit_pages to a level which ensures that when all CPUs are | |
774 | * dirtying in parallel, we cannot go more than 3% (1/32) over the dirty memory | |
775 | * thresholds before writeback cuts in. | |
776 | * | |
777 | * But the limit should not be set too high. Because it also controls the | |
778 | * amount of memory which the balance_dirty_pages() caller has to write back. | |
779 | * If this is too large then the caller will block on the IO queue all the | |
780 | * time. So limit it to four megabytes - the balance_dirty_pages() caller | |
781 | * will write six megabyte chunks, max. | |
782 | */ | |
783 | ||
2d1d43f6 | 784 | void writeback_set_ratelimit(void) |
1da177e4 | 785 | { |
40c99aae | 786 | ratelimit_pages = vm_total_pages / (num_online_cpus() * 32); |
1da177e4 LT |
787 | if (ratelimit_pages < 16) |
788 | ratelimit_pages = 16; | |
789 | if (ratelimit_pages * PAGE_CACHE_SIZE > 4096 * 1024) | |
790 | ratelimit_pages = (4096 * 1024) / PAGE_CACHE_SIZE; | |
791 | } | |
792 | ||
26c2143b | 793 | static int __cpuinit |
1da177e4 LT |
794 | ratelimit_handler(struct notifier_block *self, unsigned long u, void *v) |
795 | { | |
2d1d43f6 | 796 | writeback_set_ratelimit(); |
aa0f0303 | 797 | return NOTIFY_DONE; |
1da177e4 LT |
798 | } |
799 | ||
74b85f37 | 800 | static struct notifier_block __cpuinitdata ratelimit_nb = { |
1da177e4 LT |
801 | .notifier_call = ratelimit_handler, |
802 | .next = NULL, | |
803 | }; | |
804 | ||
805 | /* | |
dc6e29da LT |
806 | * Called early on to tune the page writeback dirty limits. |
807 | * | |
808 | * We used to scale dirty pages according to how total memory | |
809 | * related to pages that could be allocated for buffers (by | |
810 | * comparing nr_free_buffer_pages() to vm_total_pages. | |
811 | * | |
812 | * However, that was when we used "dirty_ratio" to scale with | |
813 | * all memory, and we don't do that any more. "dirty_ratio" | |
814 | * is now applied to total non-HIGHPAGE memory (by subtracting | |
815 | * totalhigh_pages from vm_total_pages), and as such we can't | |
816 | * get into the old insane situation any more where we had | |
817 | * large amounts of dirty pages compared to a small amount of | |
818 | * non-HIGHMEM memory. | |
819 | * | |
820 | * But we might still want to scale the dirty_ratio by how | |
821 | * much memory the box has.. | |
1da177e4 LT |
822 | */ |
823 | void __init page_writeback_init(void) | |
824 | { | |
04fbfdc1 PZ |
825 | int shift; |
826 | ||
f6ef9438 | 827 | mod_timer(&wb_timer, jiffies + dirty_writeback_interval); |
2d1d43f6 | 828 | writeback_set_ratelimit(); |
1da177e4 | 829 | register_cpu_notifier(&ratelimit_nb); |
04fbfdc1 PZ |
830 | |
831 | shift = calc_period_shift(); | |
832 | prop_descriptor_init(&vm_completions, shift); | |
3e26c149 | 833 | prop_descriptor_init(&vm_dirties, shift); |
1da177e4 LT |
834 | } |
835 | ||
811d736f | 836 | /** |
0ea97180 | 837 | * write_cache_pages - walk the list of dirty pages of the given address space and write all of them. |
811d736f DH |
838 | * @mapping: address space structure to write |
839 | * @wbc: subtract the number of written pages from *@wbc->nr_to_write | |
0ea97180 MS |
840 | * @writepage: function called for each page |
841 | * @data: data passed to writepage function | |
811d736f | 842 | * |
0ea97180 | 843 | * If a page is already under I/O, write_cache_pages() skips it, even |
811d736f DH |
844 | * if it's dirty. This is desirable behaviour for memory-cleaning writeback, |
845 | * but it is INCORRECT for data-integrity system calls such as fsync(). fsync() | |
846 | * and msync() need to guarantee that all the data which was dirty at the time | |
847 | * the call was made get new I/O started against them. If wbc->sync_mode is | |
848 | * WB_SYNC_ALL then we were called for data integrity and we must wait for | |
849 | * existing IO to complete. | |
811d736f | 850 | */ |
0ea97180 MS |
851 | int write_cache_pages(struct address_space *mapping, |
852 | struct writeback_control *wbc, writepage_t writepage, | |
853 | void *data) | |
811d736f DH |
854 | { |
855 | struct backing_dev_info *bdi = mapping->backing_dev_info; | |
856 | int ret = 0; | |
857 | int done = 0; | |
811d736f DH |
858 | struct pagevec pvec; |
859 | int nr_pages; | |
860 | pgoff_t index; | |
861 | pgoff_t end; /* Inclusive */ | |
862 | int scanned = 0; | |
863 | int range_whole = 0; | |
864 | ||
865 | if (wbc->nonblocking && bdi_write_congested(bdi)) { | |
866 | wbc->encountered_congestion = 1; | |
867 | return 0; | |
868 | } | |
869 | ||
811d736f DH |
870 | pagevec_init(&pvec, 0); |
871 | if (wbc->range_cyclic) { | |
872 | index = mapping->writeback_index; /* Start from prev offset */ | |
873 | end = -1; | |
874 | } else { | |
875 | index = wbc->range_start >> PAGE_CACHE_SHIFT; | |
876 | end = wbc->range_end >> PAGE_CACHE_SHIFT; | |
877 | if (wbc->range_start == 0 && wbc->range_end == LLONG_MAX) | |
878 | range_whole = 1; | |
879 | scanned = 1; | |
880 | } | |
881 | retry: | |
882 | while (!done && (index <= end) && | |
883 | (nr_pages = pagevec_lookup_tag(&pvec, mapping, &index, | |
884 | PAGECACHE_TAG_DIRTY, | |
885 | min(end - index, (pgoff_t)PAGEVEC_SIZE-1) + 1))) { | |
886 | unsigned i; | |
887 | ||
888 | scanned = 1; | |
889 | for (i = 0; i < nr_pages; i++) { | |
890 | struct page *page = pvec.pages[i]; | |
891 | ||
892 | /* | |
893 | * At this point we hold neither mapping->tree_lock nor | |
894 | * lock on the page itself: the page may be truncated or | |
895 | * invalidated (changing page->mapping to NULL), or even | |
896 | * swizzled back from swapper_space to tmpfs file | |
897 | * mapping | |
898 | */ | |
899 | lock_page(page); | |
900 | ||
901 | if (unlikely(page->mapping != mapping)) { | |
902 | unlock_page(page); | |
903 | continue; | |
904 | } | |
905 | ||
906 | if (!wbc->range_cyclic && page->index > end) { | |
907 | done = 1; | |
908 | unlock_page(page); | |
909 | continue; | |
910 | } | |
911 | ||
912 | if (wbc->sync_mode != WB_SYNC_NONE) | |
913 | wait_on_page_writeback(page); | |
914 | ||
915 | if (PageWriteback(page) || | |
916 | !clear_page_dirty_for_io(page)) { | |
917 | unlock_page(page); | |
918 | continue; | |
919 | } | |
920 | ||
0ea97180 | 921 | ret = (*writepage)(page, wbc, data); |
811d736f | 922 | |
e4230030 | 923 | if (unlikely(ret == AOP_WRITEPAGE_ACTIVATE)) { |
811d736f | 924 | unlock_page(page); |
e4230030 AM |
925 | ret = 0; |
926 | } | |
811d736f DH |
927 | if (ret || (--(wbc->nr_to_write) <= 0)) |
928 | done = 1; | |
929 | if (wbc->nonblocking && bdi_write_congested(bdi)) { | |
930 | wbc->encountered_congestion = 1; | |
931 | done = 1; | |
932 | } | |
933 | } | |
934 | pagevec_release(&pvec); | |
935 | cond_resched(); | |
936 | } | |
937 | if (!scanned && !done) { | |
938 | /* | |
939 | * We hit the last page and there is more work to be done: wrap | |
940 | * back to the start of the file | |
941 | */ | |
942 | scanned = 1; | |
943 | index = 0; | |
944 | goto retry; | |
945 | } | |
946 | if (wbc->range_cyclic || (range_whole && wbc->nr_to_write > 0)) | |
947 | mapping->writeback_index = index; | |
948 | return ret; | |
949 | } | |
0ea97180 MS |
950 | EXPORT_SYMBOL(write_cache_pages); |
951 | ||
952 | /* | |
953 | * Function used by generic_writepages to call the real writepage | |
954 | * function and set the mapping flags on error | |
955 | */ | |
956 | static int __writepage(struct page *page, struct writeback_control *wbc, | |
957 | void *data) | |
958 | { | |
959 | struct address_space *mapping = data; | |
960 | int ret = mapping->a_ops->writepage(page, wbc); | |
961 | mapping_set_error(mapping, ret); | |
962 | return ret; | |
963 | } | |
964 | ||
965 | /** | |
966 | * generic_writepages - walk the list of dirty pages of the given address space and writepage() all of them. | |
967 | * @mapping: address space structure to write | |
968 | * @wbc: subtract the number of written pages from *@wbc->nr_to_write | |
969 | * | |
970 | * This is a library function, which implements the writepages() | |
971 | * address_space_operation. | |
972 | */ | |
973 | int generic_writepages(struct address_space *mapping, | |
974 | struct writeback_control *wbc) | |
975 | { | |
976 | /* deal with chardevs and other special file */ | |
977 | if (!mapping->a_ops->writepage) | |
978 | return 0; | |
979 | ||
980 | return write_cache_pages(mapping, wbc, __writepage, mapping); | |
981 | } | |
811d736f DH |
982 | |
983 | EXPORT_SYMBOL(generic_writepages); | |
984 | ||
1da177e4 LT |
985 | int do_writepages(struct address_space *mapping, struct writeback_control *wbc) |
986 | { | |
22905f77 AM |
987 | int ret; |
988 | ||
1da177e4 LT |
989 | if (wbc->nr_to_write <= 0) |
990 | return 0; | |
22905f77 | 991 | wbc->for_writepages = 1; |
1da177e4 | 992 | if (mapping->a_ops->writepages) |
d08b3851 | 993 | ret = mapping->a_ops->writepages(mapping, wbc); |
22905f77 AM |
994 | else |
995 | ret = generic_writepages(mapping, wbc); | |
996 | wbc->for_writepages = 0; | |
997 | return ret; | |
1da177e4 LT |
998 | } |
999 | ||
1000 | /** | |
1001 | * write_one_page - write out a single page and optionally wait on I/O | |
67be2dd1 MW |
1002 | * @page: the page to write |
1003 | * @wait: if true, wait on writeout | |
1da177e4 LT |
1004 | * |
1005 | * The page must be locked by the caller and will be unlocked upon return. | |
1006 | * | |
1007 | * write_one_page() returns a negative error code if I/O failed. | |
1008 | */ | |
1009 | int write_one_page(struct page *page, int wait) | |
1010 | { | |
1011 | struct address_space *mapping = page->mapping; | |
1012 | int ret = 0; | |
1013 | struct writeback_control wbc = { | |
1014 | .sync_mode = WB_SYNC_ALL, | |
1015 | .nr_to_write = 1, | |
1016 | }; | |
1017 | ||
1018 | BUG_ON(!PageLocked(page)); | |
1019 | ||
1020 | if (wait) | |
1021 | wait_on_page_writeback(page); | |
1022 | ||
1023 | if (clear_page_dirty_for_io(page)) { | |
1024 | page_cache_get(page); | |
1025 | ret = mapping->a_ops->writepage(page, &wbc); | |
1026 | if (ret == 0 && wait) { | |
1027 | wait_on_page_writeback(page); | |
1028 | if (PageError(page)) | |
1029 | ret = -EIO; | |
1030 | } | |
1031 | page_cache_release(page); | |
1032 | } else { | |
1033 | unlock_page(page); | |
1034 | } | |
1035 | return ret; | |
1036 | } | |
1037 | EXPORT_SYMBOL(write_one_page); | |
1038 | ||
76719325 KC |
1039 | /* |
1040 | * For address_spaces which do not use buffers nor write back. | |
1041 | */ | |
1042 | int __set_page_dirty_no_writeback(struct page *page) | |
1043 | { | |
1044 | if (!PageDirty(page)) | |
1045 | SetPageDirty(page); | |
1046 | return 0; | |
1047 | } | |
1048 | ||
1da177e4 LT |
1049 | /* |
1050 | * For address_spaces which do not use buffers. Just tag the page as dirty in | |
1051 | * its radix tree. | |
1052 | * | |
1053 | * This is also used when a single buffer is being dirtied: we want to set the | |
1054 | * page dirty in that case, but not all the buffers. This is a "bottom-up" | |
1055 | * dirtying, whereas __set_page_dirty_buffers() is a "top-down" dirtying. | |
1056 | * | |
1057 | * Most callers have locked the page, which pins the address_space in memory. | |
1058 | * But zap_pte_range() does not lock the page, however in that case the | |
1059 | * mapping is pinned by the vma's ->vm_file reference. | |
1060 | * | |
1061 | * We take care to handle the case where the page was truncated from the | |
183ff22b | 1062 | * mapping by re-checking page_mapping() inside tree_lock. |
1da177e4 LT |
1063 | */ |
1064 | int __set_page_dirty_nobuffers(struct page *page) | |
1065 | { | |
1da177e4 LT |
1066 | if (!TestSetPageDirty(page)) { |
1067 | struct address_space *mapping = page_mapping(page); | |
1068 | struct address_space *mapping2; | |
1069 | ||
8c08540f AM |
1070 | if (!mapping) |
1071 | return 1; | |
1072 | ||
1073 | write_lock_irq(&mapping->tree_lock); | |
1074 | mapping2 = page_mapping(page); | |
1075 | if (mapping2) { /* Race with truncate? */ | |
1076 | BUG_ON(mapping2 != mapping); | |
787d2214 | 1077 | WARN_ON_ONCE(!PagePrivate(page) && !PageUptodate(page)); |
55e829af | 1078 | if (mapping_cap_account_dirty(mapping)) { |
8c08540f | 1079 | __inc_zone_page_state(page, NR_FILE_DIRTY); |
c9e51e41 PZ |
1080 | __inc_bdi_stat(mapping->backing_dev_info, |
1081 | BDI_RECLAIMABLE); | |
55e829af AM |
1082 | task_io_account_write(PAGE_CACHE_SIZE); |
1083 | } | |
8c08540f AM |
1084 | radix_tree_tag_set(&mapping->page_tree, |
1085 | page_index(page), PAGECACHE_TAG_DIRTY); | |
1086 | } | |
1087 | write_unlock_irq(&mapping->tree_lock); | |
1088 | if (mapping->host) { | |
1089 | /* !PageAnon && !swapper_space */ | |
1090 | __mark_inode_dirty(mapping->host, I_DIRTY_PAGES); | |
1da177e4 | 1091 | } |
4741c9fd | 1092 | return 1; |
1da177e4 | 1093 | } |
4741c9fd | 1094 | return 0; |
1da177e4 LT |
1095 | } |
1096 | EXPORT_SYMBOL(__set_page_dirty_nobuffers); | |
1097 | ||
1098 | /* | |
1099 | * When a writepage implementation decides that it doesn't want to write this | |
1100 | * page for some reason, it should redirty the locked page via | |
1101 | * redirty_page_for_writepage() and it should then unlock the page and return 0 | |
1102 | */ | |
1103 | int redirty_page_for_writepage(struct writeback_control *wbc, struct page *page) | |
1104 | { | |
1105 | wbc->pages_skipped++; | |
1106 | return __set_page_dirty_nobuffers(page); | |
1107 | } | |
1108 | EXPORT_SYMBOL(redirty_page_for_writepage); | |
1109 | ||
1110 | /* | |
1111 | * If the mapping doesn't provide a set_page_dirty a_op, then | |
1112 | * just fall through and assume that it wants buffer_heads. | |
1113 | */ | |
3e26c149 | 1114 | static int __set_page_dirty(struct page *page) |
1da177e4 LT |
1115 | { |
1116 | struct address_space *mapping = page_mapping(page); | |
1117 | ||
1118 | if (likely(mapping)) { | |
1119 | int (*spd)(struct page *) = mapping->a_ops->set_page_dirty; | |
9361401e DH |
1120 | #ifdef CONFIG_BLOCK |
1121 | if (!spd) | |
1122 | spd = __set_page_dirty_buffers; | |
1123 | #endif | |
1124 | return (*spd)(page); | |
1da177e4 | 1125 | } |
4741c9fd AM |
1126 | if (!PageDirty(page)) { |
1127 | if (!TestSetPageDirty(page)) | |
1128 | return 1; | |
1129 | } | |
1da177e4 LT |
1130 | return 0; |
1131 | } | |
3e26c149 | 1132 | |
920c7a5d | 1133 | int set_page_dirty(struct page *page) |
3e26c149 PZ |
1134 | { |
1135 | int ret = __set_page_dirty(page); | |
1136 | if (ret) | |
1137 | task_dirty_inc(current); | |
1138 | return ret; | |
1139 | } | |
1da177e4 LT |
1140 | EXPORT_SYMBOL(set_page_dirty); |
1141 | ||
1142 | /* | |
1143 | * set_page_dirty() is racy if the caller has no reference against | |
1144 | * page->mapping->host, and if the page is unlocked. This is because another | |
1145 | * CPU could truncate the page off the mapping and then free the mapping. | |
1146 | * | |
1147 | * Usually, the page _is_ locked, or the caller is a user-space process which | |
1148 | * holds a reference on the inode by having an open file. | |
1149 | * | |
1150 | * In other cases, the page should be locked before running set_page_dirty(). | |
1151 | */ | |
1152 | int set_page_dirty_lock(struct page *page) | |
1153 | { | |
1154 | int ret; | |
1155 | ||
db37648c | 1156 | lock_page_nosync(page); |
1da177e4 LT |
1157 | ret = set_page_dirty(page); |
1158 | unlock_page(page); | |
1159 | return ret; | |
1160 | } | |
1161 | EXPORT_SYMBOL(set_page_dirty_lock); | |
1162 | ||
1da177e4 LT |
1163 | /* |
1164 | * Clear a page's dirty flag, while caring for dirty memory accounting. | |
1165 | * Returns true if the page was previously dirty. | |
1166 | * | |
1167 | * This is for preparing to put the page under writeout. We leave the page | |
1168 | * tagged as dirty in the radix tree so that a concurrent write-for-sync | |
1169 | * can discover it via a PAGECACHE_TAG_DIRTY walk. The ->writepage | |
1170 | * implementation will run either set_page_writeback() or set_page_dirty(), | |
1171 | * at which stage we bring the page's dirty flag and radix-tree dirty tag | |
1172 | * back into sync. | |
1173 | * | |
1174 | * This incoherency between the page's dirty flag and radix-tree tag is | |
1175 | * unfortunate, but it only exists while the page is locked. | |
1176 | */ | |
1177 | int clear_page_dirty_for_io(struct page *page) | |
1178 | { | |
1179 | struct address_space *mapping = page_mapping(page); | |
1180 | ||
79352894 NP |
1181 | BUG_ON(!PageLocked(page)); |
1182 | ||
fe3cba17 | 1183 | ClearPageReclaim(page); |
7658cc28 LT |
1184 | if (mapping && mapping_cap_account_dirty(mapping)) { |
1185 | /* | |
1186 | * Yes, Virginia, this is indeed insane. | |
1187 | * | |
1188 | * We use this sequence to make sure that | |
1189 | * (a) we account for dirty stats properly | |
1190 | * (b) we tell the low-level filesystem to | |
1191 | * mark the whole page dirty if it was | |
1192 | * dirty in a pagetable. Only to then | |
1193 | * (c) clean the page again and return 1 to | |
1194 | * cause the writeback. | |
1195 | * | |
1196 | * This way we avoid all nasty races with the | |
1197 | * dirty bit in multiple places and clearing | |
1198 | * them concurrently from different threads. | |
1199 | * | |
1200 | * Note! Normally the "set_page_dirty(page)" | |
1201 | * has no effect on the actual dirty bit - since | |
1202 | * that will already usually be set. But we | |
1203 | * need the side effects, and it can help us | |
1204 | * avoid races. | |
1205 | * | |
1206 | * We basically use the page "master dirty bit" | |
1207 | * as a serialization point for all the different | |
1208 | * threads doing their things. | |
7658cc28 LT |
1209 | */ |
1210 | if (page_mkclean(page)) | |
1211 | set_page_dirty(page); | |
79352894 NP |
1212 | /* |
1213 | * We carefully synchronise fault handlers against | |
1214 | * installing a dirty pte and marking the page dirty | |
1215 | * at this point. We do this by having them hold the | |
1216 | * page lock at some point after installing their | |
1217 | * pte, but before marking the page dirty. | |
1218 | * Pages are always locked coming in here, so we get | |
1219 | * the desired exclusion. See mm/memory.c:do_wp_page() | |
1220 | * for more comments. | |
1221 | */ | |
7658cc28 | 1222 | if (TestClearPageDirty(page)) { |
8c08540f | 1223 | dec_zone_page_state(page, NR_FILE_DIRTY); |
c9e51e41 PZ |
1224 | dec_bdi_stat(mapping->backing_dev_info, |
1225 | BDI_RECLAIMABLE); | |
7658cc28 | 1226 | return 1; |
1da177e4 | 1227 | } |
7658cc28 | 1228 | return 0; |
1da177e4 | 1229 | } |
7658cc28 | 1230 | return TestClearPageDirty(page); |
1da177e4 | 1231 | } |
58bb01a9 | 1232 | EXPORT_SYMBOL(clear_page_dirty_for_io); |
1da177e4 LT |
1233 | |
1234 | int test_clear_page_writeback(struct page *page) | |
1235 | { | |
1236 | struct address_space *mapping = page_mapping(page); | |
1237 | int ret; | |
1238 | ||
1239 | if (mapping) { | |
69cb51d1 | 1240 | struct backing_dev_info *bdi = mapping->backing_dev_info; |
1da177e4 LT |
1241 | unsigned long flags; |
1242 | ||
1243 | write_lock_irqsave(&mapping->tree_lock, flags); | |
1244 | ret = TestClearPageWriteback(page); | |
69cb51d1 | 1245 | if (ret) { |
1da177e4 LT |
1246 | radix_tree_tag_clear(&mapping->page_tree, |
1247 | page_index(page), | |
1248 | PAGECACHE_TAG_WRITEBACK); | |
e4ad08fe | 1249 | if (bdi_cap_account_writeback(bdi)) { |
69cb51d1 | 1250 | __dec_bdi_stat(bdi, BDI_WRITEBACK); |
04fbfdc1 PZ |
1251 | __bdi_writeout_inc(bdi); |
1252 | } | |
69cb51d1 | 1253 | } |
1da177e4 LT |
1254 | write_unlock_irqrestore(&mapping->tree_lock, flags); |
1255 | } else { | |
1256 | ret = TestClearPageWriteback(page); | |
1257 | } | |
d688abf5 AM |
1258 | if (ret) |
1259 | dec_zone_page_state(page, NR_WRITEBACK); | |
1da177e4 LT |
1260 | return ret; |
1261 | } | |
1262 | ||
1263 | int test_set_page_writeback(struct page *page) | |
1264 | { | |
1265 | struct address_space *mapping = page_mapping(page); | |
1266 | int ret; | |
1267 | ||
1268 | if (mapping) { | |
69cb51d1 | 1269 | struct backing_dev_info *bdi = mapping->backing_dev_info; |
1da177e4 LT |
1270 | unsigned long flags; |
1271 | ||
1272 | write_lock_irqsave(&mapping->tree_lock, flags); | |
1273 | ret = TestSetPageWriteback(page); | |
69cb51d1 | 1274 | if (!ret) { |
1da177e4 LT |
1275 | radix_tree_tag_set(&mapping->page_tree, |
1276 | page_index(page), | |
1277 | PAGECACHE_TAG_WRITEBACK); | |
e4ad08fe | 1278 | if (bdi_cap_account_writeback(bdi)) |
69cb51d1 PZ |
1279 | __inc_bdi_stat(bdi, BDI_WRITEBACK); |
1280 | } | |
1da177e4 LT |
1281 | if (!PageDirty(page)) |
1282 | radix_tree_tag_clear(&mapping->page_tree, | |
1283 | page_index(page), | |
1284 | PAGECACHE_TAG_DIRTY); | |
1285 | write_unlock_irqrestore(&mapping->tree_lock, flags); | |
1286 | } else { | |
1287 | ret = TestSetPageWriteback(page); | |
1288 | } | |
d688abf5 AM |
1289 | if (!ret) |
1290 | inc_zone_page_state(page, NR_WRITEBACK); | |
1da177e4 LT |
1291 | return ret; |
1292 | ||
1293 | } | |
1294 | EXPORT_SYMBOL(test_set_page_writeback); | |
1295 | ||
1296 | /* | |
00128188 | 1297 | * Return true if any of the pages in the mapping are marked with the |
1da177e4 LT |
1298 | * passed tag. |
1299 | */ | |
1300 | int mapping_tagged(struct address_space *mapping, int tag) | |
1301 | { | |
1da177e4 | 1302 | int ret; |
00128188 | 1303 | rcu_read_lock(); |
1da177e4 | 1304 | ret = radix_tree_tagged(&mapping->page_tree, tag); |
00128188 | 1305 | rcu_read_unlock(); |
1da177e4 LT |
1306 | return ret; |
1307 | } | |
1308 | EXPORT_SYMBOL(mapping_tagged); |