Commit | Line | Data |
---|---|---|
1da177e4 | 1 | /* |
f30c2269 | 2 | * mm/page-writeback.c |
1da177e4 LT |
3 | * |
4 | * Copyright (C) 2002, Linus Torvalds. | |
5 | * | |
6 | * Contains functions related to writing back dirty pages at the | |
7 | * address_space level. | |
8 | * | |
9 | * 10Apr2002 akpm@zip.com.au | |
10 | * Initial version | |
11 | */ | |
12 | ||
13 | #include <linux/kernel.h> | |
14 | #include <linux/module.h> | |
15 | #include <linux/spinlock.h> | |
16 | #include <linux/fs.h> | |
17 | #include <linux/mm.h> | |
18 | #include <linux/swap.h> | |
19 | #include <linux/slab.h> | |
20 | #include <linux/pagemap.h> | |
21 | #include <linux/writeback.h> | |
22 | #include <linux/init.h> | |
23 | #include <linux/backing-dev.h> | |
55e829af | 24 | #include <linux/task_io_accounting_ops.h> |
1da177e4 LT |
25 | #include <linux/blkdev.h> |
26 | #include <linux/mpage.h> | |
d08b3851 | 27 | #include <linux/rmap.h> |
1da177e4 LT |
28 | #include <linux/percpu.h> |
29 | #include <linux/notifier.h> | |
30 | #include <linux/smp.h> | |
31 | #include <linux/sysctl.h> | |
32 | #include <linux/cpu.h> | |
33 | #include <linux/syscalls.h> | |
cf9a2ae8 | 34 | #include <linux/buffer_head.h> |
811d736f | 35 | #include <linux/pagevec.h> |
1da177e4 LT |
36 | |
37 | /* | |
38 | * The maximum number of pages to writeout in a single bdflush/kupdate | |
39 | * operation. We do this so we don't hold I_LOCK against an inode for | |
40 | * enormous amounts of time, which would block a userspace task which has | |
41 | * been forced to throttle against that inode. Also, the code reevaluates | |
42 | * the dirty each time it has written this many pages. | |
43 | */ | |
44 | #define MAX_WRITEBACK_PAGES 1024 | |
45 | ||
46 | /* | |
47 | * After a CPU has dirtied this many pages, balance_dirty_pages_ratelimited | |
48 | * will look to see if it needs to force writeback or throttling. | |
49 | */ | |
50 | static long ratelimit_pages = 32; | |
51 | ||
e236a166 | 52 | static int dirty_exceeded __cacheline_aligned_in_smp; /* Dirty mem may be over limit */ |
1da177e4 LT |
53 | |
54 | /* | |
55 | * When balance_dirty_pages decides that the caller needs to perform some | |
56 | * non-background writeback, this is how many pages it will attempt to write. | |
57 | * It should be somewhat larger than RATELIMIT_PAGES to ensure that reasonably | |
58 | * large amounts of I/O are submitted. | |
59 | */ | |
60 | static inline long sync_writeback_pages(void) | |
61 | { | |
62 | return ratelimit_pages + ratelimit_pages / 2; | |
63 | } | |
64 | ||
65 | /* The following parameters are exported via /proc/sys/vm */ | |
66 | ||
67 | /* | |
68 | * Start background writeback (via pdflush) at this percentage | |
69 | */ | |
70 | int dirty_background_ratio = 10; | |
71 | ||
72 | /* | |
73 | * The generator of dirty data starts writeback at this percentage | |
74 | */ | |
75 | int vm_dirty_ratio = 40; | |
76 | ||
77 | /* | |
fd5403c7 | 78 | * The interval between `kupdate'-style writebacks, in jiffies |
1da177e4 | 79 | */ |
f6ef9438 | 80 | int dirty_writeback_interval = 5 * HZ; |
1da177e4 LT |
81 | |
82 | /* | |
fd5403c7 | 83 | * The longest number of jiffies for which data is allowed to remain dirty |
1da177e4 | 84 | */ |
f6ef9438 | 85 | int dirty_expire_interval = 30 * HZ; |
1da177e4 LT |
86 | |
87 | /* | |
88 | * Flag that makes the machine dump writes/reads and block dirtyings. | |
89 | */ | |
90 | int block_dump; | |
91 | ||
92 | /* | |
ed5b43f1 BS |
93 | * Flag that puts the machine in "laptop mode". Doubles as a timeout in jiffies: |
94 | * a full sync is triggered after this time elapses without any disk activity. | |
1da177e4 LT |
95 | */ |
96 | int laptop_mode; | |
97 | ||
98 | EXPORT_SYMBOL(laptop_mode); | |
99 | ||
100 | /* End of sysctl-exported parameters */ | |
101 | ||
102 | ||
103 | static void background_writeout(unsigned long _min_pages); | |
104 | ||
1da177e4 LT |
105 | /* |
106 | * Work out the current dirty-memory clamping and background writeout | |
107 | * thresholds. | |
108 | * | |
109 | * The main aim here is to lower them aggressively if there is a lot of mapped | |
110 | * memory around. To avoid stressing page reclaim with lots of unreclaimable | |
111 | * pages. It is better to clamp down on writers than to start swapping, and | |
112 | * performing lots of scanning. | |
113 | * | |
114 | * We only allow 1/2 of the currently-unmapped memory to be dirtied. | |
115 | * | |
116 | * We don't permit the clamping level to fall below 5% - that is getting rather | |
117 | * excessive. | |
118 | * | |
119 | * We make sure that the background writeout level is below the adjusted | |
120 | * clamping level. | |
121 | */ | |
122 | static void | |
c24f21bd CL |
123 | get_dirty_limits(long *pbackground, long *pdirty, |
124 | struct address_space *mapping) | |
1da177e4 LT |
125 | { |
126 | int background_ratio; /* Percentages */ | |
127 | int dirty_ratio; | |
128 | int unmapped_ratio; | |
129 | long background; | |
130 | long dirty; | |
40c99aae | 131 | unsigned long available_memory = vm_total_pages; |
1da177e4 LT |
132 | struct task_struct *tsk; |
133 | ||
1da177e4 LT |
134 | #ifdef CONFIG_HIGHMEM |
135 | /* | |
136 | * If this mapping can only allocate from low memory, | |
137 | * we exclude high memory from our count. | |
138 | */ | |
139 | if (mapping && !(mapping_gfp_mask(mapping) & __GFP_HIGHMEM)) | |
140 | available_memory -= totalhigh_pages; | |
141 | #endif | |
142 | ||
143 | ||
c24f21bd CL |
144 | unmapped_ratio = 100 - ((global_page_state(NR_FILE_MAPPED) + |
145 | global_page_state(NR_ANON_PAGES)) * 100) / | |
40c99aae | 146 | vm_total_pages; |
1da177e4 LT |
147 | |
148 | dirty_ratio = vm_dirty_ratio; | |
149 | if (dirty_ratio > unmapped_ratio / 2) | |
150 | dirty_ratio = unmapped_ratio / 2; | |
151 | ||
152 | if (dirty_ratio < 5) | |
153 | dirty_ratio = 5; | |
154 | ||
155 | background_ratio = dirty_background_ratio; | |
156 | if (background_ratio >= dirty_ratio) | |
157 | background_ratio = dirty_ratio / 2; | |
158 | ||
159 | background = (background_ratio * available_memory) / 100; | |
160 | dirty = (dirty_ratio * available_memory) / 100; | |
161 | tsk = current; | |
162 | if (tsk->flags & PF_LESS_THROTTLE || rt_task(tsk)) { | |
163 | background += background / 4; | |
164 | dirty += dirty / 4; | |
165 | } | |
166 | *pbackground = background; | |
167 | *pdirty = dirty; | |
168 | } | |
169 | ||
170 | /* | |
171 | * balance_dirty_pages() must be called by processes which are generating dirty | |
172 | * data. It looks at the number of dirty pages in the machine and will force | |
173 | * the caller to perform writeback if the system is over `vm_dirty_ratio'. | |
174 | * If we're over `background_thresh' then pdflush is woken to perform some | |
175 | * writeout. | |
176 | */ | |
177 | static void balance_dirty_pages(struct address_space *mapping) | |
178 | { | |
1da177e4 LT |
179 | long nr_reclaimable; |
180 | long background_thresh; | |
181 | long dirty_thresh; | |
182 | unsigned long pages_written = 0; | |
183 | unsigned long write_chunk = sync_writeback_pages(); | |
184 | ||
185 | struct backing_dev_info *bdi = mapping->backing_dev_info; | |
186 | ||
187 | for (;;) { | |
188 | struct writeback_control wbc = { | |
189 | .bdi = bdi, | |
190 | .sync_mode = WB_SYNC_NONE, | |
191 | .older_than_this = NULL, | |
192 | .nr_to_write = write_chunk, | |
111ebb6e | 193 | .range_cyclic = 1, |
1da177e4 LT |
194 | }; |
195 | ||
c24f21bd CL |
196 | get_dirty_limits(&background_thresh, &dirty_thresh, mapping); |
197 | nr_reclaimable = global_page_state(NR_FILE_DIRTY) + | |
198 | global_page_state(NR_UNSTABLE_NFS); | |
199 | if (nr_reclaimable + global_page_state(NR_WRITEBACK) <= | |
200 | dirty_thresh) | |
201 | break; | |
1da177e4 | 202 | |
e236a166 AM |
203 | if (!dirty_exceeded) |
204 | dirty_exceeded = 1; | |
1da177e4 LT |
205 | |
206 | /* Note: nr_reclaimable denotes nr_dirty + nr_unstable. | |
207 | * Unstable writes are a feature of certain networked | |
208 | * filesystems (i.e. NFS) in which data may have been | |
209 | * written to the server's write cache, but has not yet | |
210 | * been flushed to permanent storage. | |
211 | */ | |
212 | if (nr_reclaimable) { | |
213 | writeback_inodes(&wbc); | |
c24f21bd CL |
214 | get_dirty_limits(&background_thresh, |
215 | &dirty_thresh, mapping); | |
216 | nr_reclaimable = global_page_state(NR_FILE_DIRTY) + | |
217 | global_page_state(NR_UNSTABLE_NFS); | |
218 | if (nr_reclaimable + | |
219 | global_page_state(NR_WRITEBACK) | |
220 | <= dirty_thresh) | |
221 | break; | |
1da177e4 LT |
222 | pages_written += write_chunk - wbc.nr_to_write; |
223 | if (pages_written >= write_chunk) | |
224 | break; /* We've done our duty */ | |
225 | } | |
3fcfab16 | 226 | congestion_wait(WRITE, HZ/10); |
1da177e4 LT |
227 | } |
228 | ||
c24f21bd CL |
229 | if (nr_reclaimable + global_page_state(NR_WRITEBACK) |
230 | <= dirty_thresh && dirty_exceeded) | |
231 | dirty_exceeded = 0; | |
1da177e4 LT |
232 | |
233 | if (writeback_in_progress(bdi)) | |
234 | return; /* pdflush is already working this queue */ | |
235 | ||
236 | /* | |
237 | * In laptop mode, we wait until hitting the higher threshold before | |
238 | * starting background writeout, and then write out all the way down | |
239 | * to the lower threshold. So slow writers cause minimal disk activity. | |
240 | * | |
241 | * In normal mode, we start background writeout at the lower | |
242 | * background_thresh, to keep the amount of dirty memory low. | |
243 | */ | |
244 | if ((laptop_mode && pages_written) || | |
245 | (!laptop_mode && (nr_reclaimable > background_thresh))) | |
246 | pdflush_operation(background_writeout, 0); | |
247 | } | |
248 | ||
edc79b2a PZ |
249 | void set_page_dirty_balance(struct page *page) |
250 | { | |
251 | if (set_page_dirty(page)) { | |
252 | struct address_space *mapping = page_mapping(page); | |
253 | ||
254 | if (mapping) | |
255 | balance_dirty_pages_ratelimited(mapping); | |
256 | } | |
257 | } | |
258 | ||
1da177e4 | 259 | /** |
fa5a734e | 260 | * balance_dirty_pages_ratelimited_nr - balance dirty memory state |
67be2dd1 | 261 | * @mapping: address_space which was dirtied |
a580290c | 262 | * @nr_pages_dirtied: number of pages which the caller has just dirtied |
1da177e4 LT |
263 | * |
264 | * Processes which are dirtying memory should call in here once for each page | |
265 | * which was newly dirtied. The function will periodically check the system's | |
266 | * dirty state and will initiate writeback if needed. | |
267 | * | |
268 | * On really big machines, get_writeback_state is expensive, so try to avoid | |
269 | * calling it too often (ratelimiting). But once we're over the dirty memory | |
270 | * limit we decrease the ratelimiting by a lot, to prevent individual processes | |
271 | * from overshooting the limit by (ratelimit_pages) each. | |
272 | */ | |
fa5a734e AM |
273 | void balance_dirty_pages_ratelimited_nr(struct address_space *mapping, |
274 | unsigned long nr_pages_dirtied) | |
1da177e4 | 275 | { |
fa5a734e AM |
276 | static DEFINE_PER_CPU(unsigned long, ratelimits) = 0; |
277 | unsigned long ratelimit; | |
278 | unsigned long *p; | |
1da177e4 LT |
279 | |
280 | ratelimit = ratelimit_pages; | |
281 | if (dirty_exceeded) | |
282 | ratelimit = 8; | |
283 | ||
284 | /* | |
285 | * Check the rate limiting. Also, we do not want to throttle real-time | |
286 | * tasks in balance_dirty_pages(). Period. | |
287 | */ | |
fa5a734e AM |
288 | preempt_disable(); |
289 | p = &__get_cpu_var(ratelimits); | |
290 | *p += nr_pages_dirtied; | |
291 | if (unlikely(*p >= ratelimit)) { | |
292 | *p = 0; | |
293 | preempt_enable(); | |
1da177e4 LT |
294 | balance_dirty_pages(mapping); |
295 | return; | |
296 | } | |
fa5a734e | 297 | preempt_enable(); |
1da177e4 | 298 | } |
fa5a734e | 299 | EXPORT_SYMBOL(balance_dirty_pages_ratelimited_nr); |
1da177e4 LT |
300 | |
301 | void throttle_vm_writeout(void) | |
302 | { | |
1da177e4 LT |
303 | long background_thresh; |
304 | long dirty_thresh; | |
305 | ||
306 | for ( ; ; ) { | |
c24f21bd | 307 | get_dirty_limits(&background_thresh, &dirty_thresh, NULL); |
1da177e4 LT |
308 | |
309 | /* | |
310 | * Boost the allowable dirty threshold a bit for page | |
311 | * allocators so they don't get DoS'ed by heavy writers | |
312 | */ | |
313 | dirty_thresh += dirty_thresh / 10; /* wheeee... */ | |
314 | ||
c24f21bd CL |
315 | if (global_page_state(NR_UNSTABLE_NFS) + |
316 | global_page_state(NR_WRITEBACK) <= dirty_thresh) | |
317 | break; | |
3fcfab16 | 318 | congestion_wait(WRITE, HZ/10); |
1da177e4 LT |
319 | } |
320 | } | |
321 | ||
322 | ||
323 | /* | |
324 | * writeback at least _min_pages, and keep writing until the amount of dirty | |
325 | * memory is less than the background threshold, or until we're all clean. | |
326 | */ | |
327 | static void background_writeout(unsigned long _min_pages) | |
328 | { | |
329 | long min_pages = _min_pages; | |
330 | struct writeback_control wbc = { | |
331 | .bdi = NULL, | |
332 | .sync_mode = WB_SYNC_NONE, | |
333 | .older_than_this = NULL, | |
334 | .nr_to_write = 0, | |
335 | .nonblocking = 1, | |
111ebb6e | 336 | .range_cyclic = 1, |
1da177e4 LT |
337 | }; |
338 | ||
339 | for ( ; ; ) { | |
1da177e4 LT |
340 | long background_thresh; |
341 | long dirty_thresh; | |
342 | ||
c24f21bd CL |
343 | get_dirty_limits(&background_thresh, &dirty_thresh, NULL); |
344 | if (global_page_state(NR_FILE_DIRTY) + | |
345 | global_page_state(NR_UNSTABLE_NFS) < background_thresh | |
1da177e4 LT |
346 | && min_pages <= 0) |
347 | break; | |
348 | wbc.encountered_congestion = 0; | |
349 | wbc.nr_to_write = MAX_WRITEBACK_PAGES; | |
350 | wbc.pages_skipped = 0; | |
351 | writeback_inodes(&wbc); | |
352 | min_pages -= MAX_WRITEBACK_PAGES - wbc.nr_to_write; | |
353 | if (wbc.nr_to_write > 0 || wbc.pages_skipped > 0) { | |
354 | /* Wrote less than expected */ | |
3fcfab16 | 355 | congestion_wait(WRITE, HZ/10); |
1da177e4 LT |
356 | if (!wbc.encountered_congestion) |
357 | break; | |
358 | } | |
359 | } | |
360 | } | |
361 | ||
362 | /* | |
363 | * Start writeback of `nr_pages' pages. If `nr_pages' is zero, write back | |
364 | * the whole world. Returns 0 if a pdflush thread was dispatched. Returns | |
365 | * -1 if all pdflush threads were busy. | |
366 | */ | |
687a21ce | 367 | int wakeup_pdflush(long nr_pages) |
1da177e4 | 368 | { |
c24f21bd CL |
369 | if (nr_pages == 0) |
370 | nr_pages = global_page_state(NR_FILE_DIRTY) + | |
371 | global_page_state(NR_UNSTABLE_NFS); | |
1da177e4 LT |
372 | return pdflush_operation(background_writeout, nr_pages); |
373 | } | |
374 | ||
375 | static void wb_timer_fn(unsigned long unused); | |
376 | static void laptop_timer_fn(unsigned long unused); | |
377 | ||
8d06afab IM |
378 | static DEFINE_TIMER(wb_timer, wb_timer_fn, 0, 0); |
379 | static DEFINE_TIMER(laptop_mode_wb_timer, laptop_timer_fn, 0, 0); | |
1da177e4 LT |
380 | |
381 | /* | |
382 | * Periodic writeback of "old" data. | |
383 | * | |
384 | * Define "old": the first time one of an inode's pages is dirtied, we mark the | |
385 | * dirtying-time in the inode's address_space. So this periodic writeback code | |
386 | * just walks the superblock inode list, writing back any inodes which are | |
387 | * older than a specific point in time. | |
388 | * | |
f6ef9438 BS |
389 | * Try to run once per dirty_writeback_interval. But if a writeback event |
390 | * takes longer than a dirty_writeback_interval interval, then leave a | |
1da177e4 LT |
391 | * one-second gap. |
392 | * | |
393 | * older_than_this takes precedence over nr_to_write. So we'll only write back | |
394 | * all dirty pages if they are all attached to "old" mappings. | |
395 | */ | |
396 | static void wb_kupdate(unsigned long arg) | |
397 | { | |
398 | unsigned long oldest_jif; | |
399 | unsigned long start_jif; | |
400 | unsigned long next_jif; | |
401 | long nr_to_write; | |
1da177e4 LT |
402 | struct writeback_control wbc = { |
403 | .bdi = NULL, | |
404 | .sync_mode = WB_SYNC_NONE, | |
405 | .older_than_this = &oldest_jif, | |
406 | .nr_to_write = 0, | |
407 | .nonblocking = 1, | |
408 | .for_kupdate = 1, | |
111ebb6e | 409 | .range_cyclic = 1, |
1da177e4 LT |
410 | }; |
411 | ||
412 | sync_supers(); | |
413 | ||
f6ef9438 | 414 | oldest_jif = jiffies - dirty_expire_interval; |
1da177e4 | 415 | start_jif = jiffies; |
f6ef9438 | 416 | next_jif = start_jif + dirty_writeback_interval; |
c24f21bd CL |
417 | nr_to_write = global_page_state(NR_FILE_DIRTY) + |
418 | global_page_state(NR_UNSTABLE_NFS) + | |
1da177e4 LT |
419 | (inodes_stat.nr_inodes - inodes_stat.nr_unused); |
420 | while (nr_to_write > 0) { | |
421 | wbc.encountered_congestion = 0; | |
422 | wbc.nr_to_write = MAX_WRITEBACK_PAGES; | |
423 | writeback_inodes(&wbc); | |
424 | if (wbc.nr_to_write > 0) { | |
425 | if (wbc.encountered_congestion) | |
3fcfab16 | 426 | congestion_wait(WRITE, HZ/10); |
1da177e4 LT |
427 | else |
428 | break; /* All the old data is written */ | |
429 | } | |
430 | nr_to_write -= MAX_WRITEBACK_PAGES - wbc.nr_to_write; | |
431 | } | |
432 | if (time_before(next_jif, jiffies + HZ)) | |
433 | next_jif = jiffies + HZ; | |
f6ef9438 | 434 | if (dirty_writeback_interval) |
1da177e4 LT |
435 | mod_timer(&wb_timer, next_jif); |
436 | } | |
437 | ||
438 | /* | |
439 | * sysctl handler for /proc/sys/vm/dirty_writeback_centisecs | |
440 | */ | |
441 | int dirty_writeback_centisecs_handler(ctl_table *table, int write, | |
442 | struct file *file, void __user *buffer, size_t *length, loff_t *ppos) | |
443 | { | |
f6ef9438 BS |
444 | proc_dointvec_userhz_jiffies(table, write, file, buffer, length, ppos); |
445 | if (dirty_writeback_interval) { | |
1da177e4 | 446 | mod_timer(&wb_timer, |
f6ef9438 BS |
447 | jiffies + dirty_writeback_interval); |
448 | } else { | |
1da177e4 LT |
449 | del_timer(&wb_timer); |
450 | } | |
451 | return 0; | |
452 | } | |
453 | ||
454 | static void wb_timer_fn(unsigned long unused) | |
455 | { | |
456 | if (pdflush_operation(wb_kupdate, 0) < 0) | |
457 | mod_timer(&wb_timer, jiffies + HZ); /* delay 1 second */ | |
458 | } | |
459 | ||
460 | static void laptop_flush(unsigned long unused) | |
461 | { | |
462 | sys_sync(); | |
463 | } | |
464 | ||
465 | static void laptop_timer_fn(unsigned long unused) | |
466 | { | |
467 | pdflush_operation(laptop_flush, 0); | |
468 | } | |
469 | ||
470 | /* | |
471 | * We've spun up the disk and we're in laptop mode: schedule writeback | |
472 | * of all dirty data a few seconds from now. If the flush is already scheduled | |
473 | * then push it back - the user is still using the disk. | |
474 | */ | |
475 | void laptop_io_completion(void) | |
476 | { | |
ed5b43f1 | 477 | mod_timer(&laptop_mode_wb_timer, jiffies + laptop_mode); |
1da177e4 LT |
478 | } |
479 | ||
480 | /* | |
481 | * We're in laptop mode and we've just synced. The sync's writes will have | |
482 | * caused another writeback to be scheduled by laptop_io_completion. | |
483 | * Nothing needs to be written back anymore, so we unschedule the writeback. | |
484 | */ | |
485 | void laptop_sync_completion(void) | |
486 | { | |
487 | del_timer(&laptop_mode_wb_timer); | |
488 | } | |
489 | ||
490 | /* | |
491 | * If ratelimit_pages is too high then we can get into dirty-data overload | |
492 | * if a large number of processes all perform writes at the same time. | |
493 | * If it is too low then SMP machines will call the (expensive) | |
494 | * get_writeback_state too often. | |
495 | * | |
496 | * Here we set ratelimit_pages to a level which ensures that when all CPUs are | |
497 | * dirtying in parallel, we cannot go more than 3% (1/32) over the dirty memory | |
498 | * thresholds before writeback cuts in. | |
499 | * | |
500 | * But the limit should not be set too high. Because it also controls the | |
501 | * amount of memory which the balance_dirty_pages() caller has to write back. | |
502 | * If this is too large then the caller will block on the IO queue all the | |
503 | * time. So limit it to four megabytes - the balance_dirty_pages() caller | |
504 | * will write six megabyte chunks, max. | |
505 | */ | |
506 | ||
2d1d43f6 | 507 | void writeback_set_ratelimit(void) |
1da177e4 | 508 | { |
40c99aae | 509 | ratelimit_pages = vm_total_pages / (num_online_cpus() * 32); |
1da177e4 LT |
510 | if (ratelimit_pages < 16) |
511 | ratelimit_pages = 16; | |
512 | if (ratelimit_pages * PAGE_CACHE_SIZE > 4096 * 1024) | |
513 | ratelimit_pages = (4096 * 1024) / PAGE_CACHE_SIZE; | |
514 | } | |
515 | ||
26c2143b | 516 | static int __cpuinit |
1da177e4 LT |
517 | ratelimit_handler(struct notifier_block *self, unsigned long u, void *v) |
518 | { | |
2d1d43f6 | 519 | writeback_set_ratelimit(); |
1da177e4 LT |
520 | return 0; |
521 | } | |
522 | ||
74b85f37 | 523 | static struct notifier_block __cpuinitdata ratelimit_nb = { |
1da177e4 LT |
524 | .notifier_call = ratelimit_handler, |
525 | .next = NULL, | |
526 | }; | |
527 | ||
528 | /* | |
529 | * If the machine has a large highmem:lowmem ratio then scale back the default | |
530 | * dirty memory thresholds: allowing too much dirty highmem pins an excessive | |
531 | * number of buffer_heads. | |
532 | */ | |
533 | void __init page_writeback_init(void) | |
534 | { | |
535 | long buffer_pages = nr_free_buffer_pages(); | |
536 | long correction; | |
537 | ||
40c99aae | 538 | correction = (100 * 4 * buffer_pages) / vm_total_pages; |
1da177e4 LT |
539 | |
540 | if (correction < 100) { | |
541 | dirty_background_ratio *= correction; | |
542 | dirty_background_ratio /= 100; | |
543 | vm_dirty_ratio *= correction; | |
544 | vm_dirty_ratio /= 100; | |
545 | ||
546 | if (dirty_background_ratio <= 0) | |
547 | dirty_background_ratio = 1; | |
548 | if (vm_dirty_ratio <= 0) | |
549 | vm_dirty_ratio = 1; | |
550 | } | |
f6ef9438 | 551 | mod_timer(&wb_timer, jiffies + dirty_writeback_interval); |
2d1d43f6 | 552 | writeback_set_ratelimit(); |
1da177e4 LT |
553 | register_cpu_notifier(&ratelimit_nb); |
554 | } | |
555 | ||
811d736f DH |
556 | /** |
557 | * generic_writepages - walk the list of dirty pages of the given | |
558 | * address space and writepage() all of them. | |
559 | * | |
560 | * @mapping: address space structure to write | |
561 | * @wbc: subtract the number of written pages from *@wbc->nr_to_write | |
562 | * | |
563 | * This is a library function, which implements the writepages() | |
564 | * address_space_operation. | |
565 | * | |
566 | * If a page is already under I/O, generic_writepages() skips it, even | |
567 | * if it's dirty. This is desirable behaviour for memory-cleaning writeback, | |
568 | * but it is INCORRECT for data-integrity system calls such as fsync(). fsync() | |
569 | * and msync() need to guarantee that all the data which was dirty at the time | |
570 | * the call was made get new I/O started against them. If wbc->sync_mode is | |
571 | * WB_SYNC_ALL then we were called for data integrity and we must wait for | |
572 | * existing IO to complete. | |
573 | * | |
574 | * Derived from mpage_writepages() - if you fix this you should check that | |
575 | * also! | |
576 | */ | |
577 | int generic_writepages(struct address_space *mapping, | |
578 | struct writeback_control *wbc) | |
579 | { | |
580 | struct backing_dev_info *bdi = mapping->backing_dev_info; | |
581 | int ret = 0; | |
582 | int done = 0; | |
583 | int (*writepage)(struct page *page, struct writeback_control *wbc); | |
584 | struct pagevec pvec; | |
585 | int nr_pages; | |
586 | pgoff_t index; | |
587 | pgoff_t end; /* Inclusive */ | |
588 | int scanned = 0; | |
589 | int range_whole = 0; | |
590 | ||
591 | if (wbc->nonblocking && bdi_write_congested(bdi)) { | |
592 | wbc->encountered_congestion = 1; | |
593 | return 0; | |
594 | } | |
595 | ||
596 | writepage = mapping->a_ops->writepage; | |
597 | ||
598 | /* deal with chardevs and other special file */ | |
599 | if (!writepage) | |
600 | return 0; | |
601 | ||
602 | pagevec_init(&pvec, 0); | |
603 | if (wbc->range_cyclic) { | |
604 | index = mapping->writeback_index; /* Start from prev offset */ | |
605 | end = -1; | |
606 | } else { | |
607 | index = wbc->range_start >> PAGE_CACHE_SHIFT; | |
608 | end = wbc->range_end >> PAGE_CACHE_SHIFT; | |
609 | if (wbc->range_start == 0 && wbc->range_end == LLONG_MAX) | |
610 | range_whole = 1; | |
611 | scanned = 1; | |
612 | } | |
613 | retry: | |
614 | while (!done && (index <= end) && | |
615 | (nr_pages = pagevec_lookup_tag(&pvec, mapping, &index, | |
616 | PAGECACHE_TAG_DIRTY, | |
617 | min(end - index, (pgoff_t)PAGEVEC_SIZE-1) + 1))) { | |
618 | unsigned i; | |
619 | ||
620 | scanned = 1; | |
621 | for (i = 0; i < nr_pages; i++) { | |
622 | struct page *page = pvec.pages[i]; | |
623 | ||
624 | /* | |
625 | * At this point we hold neither mapping->tree_lock nor | |
626 | * lock on the page itself: the page may be truncated or | |
627 | * invalidated (changing page->mapping to NULL), or even | |
628 | * swizzled back from swapper_space to tmpfs file | |
629 | * mapping | |
630 | */ | |
631 | lock_page(page); | |
632 | ||
633 | if (unlikely(page->mapping != mapping)) { | |
634 | unlock_page(page); | |
635 | continue; | |
636 | } | |
637 | ||
638 | if (!wbc->range_cyclic && page->index > end) { | |
639 | done = 1; | |
640 | unlock_page(page); | |
641 | continue; | |
642 | } | |
643 | ||
644 | if (wbc->sync_mode != WB_SYNC_NONE) | |
645 | wait_on_page_writeback(page); | |
646 | ||
647 | if (PageWriteback(page) || | |
648 | !clear_page_dirty_for_io(page)) { | |
649 | unlock_page(page); | |
650 | continue; | |
651 | } | |
652 | ||
653 | ret = (*writepage)(page, wbc); | |
654 | if (ret) { | |
655 | if (ret == -ENOSPC) | |
656 | set_bit(AS_ENOSPC, &mapping->flags); | |
657 | else | |
658 | set_bit(AS_EIO, &mapping->flags); | |
659 | } | |
660 | ||
661 | if (unlikely(ret == AOP_WRITEPAGE_ACTIVATE)) | |
662 | unlock_page(page); | |
663 | if (ret || (--(wbc->nr_to_write) <= 0)) | |
664 | done = 1; | |
665 | if (wbc->nonblocking && bdi_write_congested(bdi)) { | |
666 | wbc->encountered_congestion = 1; | |
667 | done = 1; | |
668 | } | |
669 | } | |
670 | pagevec_release(&pvec); | |
671 | cond_resched(); | |
672 | } | |
673 | if (!scanned && !done) { | |
674 | /* | |
675 | * We hit the last page and there is more work to be done: wrap | |
676 | * back to the start of the file | |
677 | */ | |
678 | scanned = 1; | |
679 | index = 0; | |
680 | goto retry; | |
681 | } | |
682 | if (wbc->range_cyclic || (range_whole && wbc->nr_to_write > 0)) | |
683 | mapping->writeback_index = index; | |
684 | return ret; | |
685 | } | |
686 | ||
687 | EXPORT_SYMBOL(generic_writepages); | |
688 | ||
1da177e4 LT |
689 | int do_writepages(struct address_space *mapping, struct writeback_control *wbc) |
690 | { | |
22905f77 AM |
691 | int ret; |
692 | ||
1da177e4 LT |
693 | if (wbc->nr_to_write <= 0) |
694 | return 0; | |
22905f77 | 695 | wbc->for_writepages = 1; |
1da177e4 | 696 | if (mapping->a_ops->writepages) |
d08b3851 | 697 | ret = mapping->a_ops->writepages(mapping, wbc); |
22905f77 AM |
698 | else |
699 | ret = generic_writepages(mapping, wbc); | |
700 | wbc->for_writepages = 0; | |
701 | return ret; | |
1da177e4 LT |
702 | } |
703 | ||
704 | /** | |
705 | * write_one_page - write out a single page and optionally wait on I/O | |
706 | * | |
67be2dd1 MW |
707 | * @page: the page to write |
708 | * @wait: if true, wait on writeout | |
1da177e4 LT |
709 | * |
710 | * The page must be locked by the caller and will be unlocked upon return. | |
711 | * | |
712 | * write_one_page() returns a negative error code if I/O failed. | |
713 | */ | |
714 | int write_one_page(struct page *page, int wait) | |
715 | { | |
716 | struct address_space *mapping = page->mapping; | |
717 | int ret = 0; | |
718 | struct writeback_control wbc = { | |
719 | .sync_mode = WB_SYNC_ALL, | |
720 | .nr_to_write = 1, | |
721 | }; | |
722 | ||
723 | BUG_ON(!PageLocked(page)); | |
724 | ||
725 | if (wait) | |
726 | wait_on_page_writeback(page); | |
727 | ||
728 | if (clear_page_dirty_for_io(page)) { | |
729 | page_cache_get(page); | |
730 | ret = mapping->a_ops->writepage(page, &wbc); | |
731 | if (ret == 0 && wait) { | |
732 | wait_on_page_writeback(page); | |
733 | if (PageError(page)) | |
734 | ret = -EIO; | |
735 | } | |
736 | page_cache_release(page); | |
737 | } else { | |
738 | unlock_page(page); | |
739 | } | |
740 | return ret; | |
741 | } | |
742 | EXPORT_SYMBOL(write_one_page); | |
743 | ||
744 | /* | |
745 | * For address_spaces which do not use buffers. Just tag the page as dirty in | |
746 | * its radix tree. | |
747 | * | |
748 | * This is also used when a single buffer is being dirtied: we want to set the | |
749 | * page dirty in that case, but not all the buffers. This is a "bottom-up" | |
750 | * dirtying, whereas __set_page_dirty_buffers() is a "top-down" dirtying. | |
751 | * | |
752 | * Most callers have locked the page, which pins the address_space in memory. | |
753 | * But zap_pte_range() does not lock the page, however in that case the | |
754 | * mapping is pinned by the vma's ->vm_file reference. | |
755 | * | |
756 | * We take care to handle the case where the page was truncated from the | |
757 | * mapping by re-checking page_mapping() insode tree_lock. | |
758 | */ | |
759 | int __set_page_dirty_nobuffers(struct page *page) | |
760 | { | |
1da177e4 LT |
761 | if (!TestSetPageDirty(page)) { |
762 | struct address_space *mapping = page_mapping(page); | |
763 | struct address_space *mapping2; | |
764 | ||
8c08540f AM |
765 | if (!mapping) |
766 | return 1; | |
767 | ||
768 | write_lock_irq(&mapping->tree_lock); | |
769 | mapping2 = page_mapping(page); | |
770 | if (mapping2) { /* Race with truncate? */ | |
771 | BUG_ON(mapping2 != mapping); | |
55e829af | 772 | if (mapping_cap_account_dirty(mapping)) { |
8c08540f | 773 | __inc_zone_page_state(page, NR_FILE_DIRTY); |
55e829af AM |
774 | task_io_account_write(PAGE_CACHE_SIZE); |
775 | } | |
8c08540f AM |
776 | radix_tree_tag_set(&mapping->page_tree, |
777 | page_index(page), PAGECACHE_TAG_DIRTY); | |
778 | } | |
779 | write_unlock_irq(&mapping->tree_lock); | |
780 | if (mapping->host) { | |
781 | /* !PageAnon && !swapper_space */ | |
782 | __mark_inode_dirty(mapping->host, I_DIRTY_PAGES); | |
1da177e4 | 783 | } |
4741c9fd | 784 | return 1; |
1da177e4 | 785 | } |
4741c9fd | 786 | return 0; |
1da177e4 LT |
787 | } |
788 | EXPORT_SYMBOL(__set_page_dirty_nobuffers); | |
789 | ||
790 | /* | |
791 | * When a writepage implementation decides that it doesn't want to write this | |
792 | * page for some reason, it should redirty the locked page via | |
793 | * redirty_page_for_writepage() and it should then unlock the page and return 0 | |
794 | */ | |
795 | int redirty_page_for_writepage(struct writeback_control *wbc, struct page *page) | |
796 | { | |
797 | wbc->pages_skipped++; | |
798 | return __set_page_dirty_nobuffers(page); | |
799 | } | |
800 | EXPORT_SYMBOL(redirty_page_for_writepage); | |
801 | ||
802 | /* | |
803 | * If the mapping doesn't provide a set_page_dirty a_op, then | |
804 | * just fall through and assume that it wants buffer_heads. | |
805 | */ | |
806 | int fastcall set_page_dirty(struct page *page) | |
807 | { | |
808 | struct address_space *mapping = page_mapping(page); | |
809 | ||
810 | if (likely(mapping)) { | |
811 | int (*spd)(struct page *) = mapping->a_ops->set_page_dirty; | |
9361401e DH |
812 | #ifdef CONFIG_BLOCK |
813 | if (!spd) | |
814 | spd = __set_page_dirty_buffers; | |
815 | #endif | |
816 | return (*spd)(page); | |
1da177e4 | 817 | } |
4741c9fd AM |
818 | if (!PageDirty(page)) { |
819 | if (!TestSetPageDirty(page)) | |
820 | return 1; | |
821 | } | |
1da177e4 LT |
822 | return 0; |
823 | } | |
824 | EXPORT_SYMBOL(set_page_dirty); | |
825 | ||
826 | /* | |
827 | * set_page_dirty() is racy if the caller has no reference against | |
828 | * page->mapping->host, and if the page is unlocked. This is because another | |
829 | * CPU could truncate the page off the mapping and then free the mapping. | |
830 | * | |
831 | * Usually, the page _is_ locked, or the caller is a user-space process which | |
832 | * holds a reference on the inode by having an open file. | |
833 | * | |
834 | * In other cases, the page should be locked before running set_page_dirty(). | |
835 | */ | |
836 | int set_page_dirty_lock(struct page *page) | |
837 | { | |
838 | int ret; | |
839 | ||
db37648c | 840 | lock_page_nosync(page); |
1da177e4 LT |
841 | ret = set_page_dirty(page); |
842 | unlock_page(page); | |
843 | return ret; | |
844 | } | |
845 | EXPORT_SYMBOL(set_page_dirty_lock); | |
846 | ||
847 | /* | |
848 | * Clear a page's dirty flag, while caring for dirty memory accounting. | |
849 | * Returns true if the page was previously dirty. | |
850 | */ | |
851 | int test_clear_page_dirty(struct page *page) | |
852 | { | |
853 | struct address_space *mapping = page_mapping(page); | |
854 | unsigned long flags; | |
855 | ||
8c08540f AM |
856 | if (!mapping) |
857 | return TestClearPageDirty(page); | |
858 | ||
859 | write_lock_irqsave(&mapping->tree_lock, flags); | |
860 | if (TestClearPageDirty(page)) { | |
861 | radix_tree_tag_clear(&mapping->page_tree, | |
862 | page_index(page), PAGECACHE_TAG_DIRTY); | |
1da177e4 | 863 | write_unlock_irqrestore(&mapping->tree_lock, flags); |
8c08540f AM |
864 | /* |
865 | * We can continue to use `mapping' here because the | |
866 | * page is locked, which pins the address_space | |
867 | */ | |
868 | if (mapping_cap_account_dirty(mapping)) { | |
869 | page_mkclean(page); | |
870 | dec_zone_page_state(page, NR_FILE_DIRTY); | |
871 | } | |
872 | return 1; | |
1da177e4 | 873 | } |
8c08540f AM |
874 | write_unlock_irqrestore(&mapping->tree_lock, flags); |
875 | return 0; | |
1da177e4 LT |
876 | } |
877 | EXPORT_SYMBOL(test_clear_page_dirty); | |
878 | ||
879 | /* | |
880 | * Clear a page's dirty flag, while caring for dirty memory accounting. | |
881 | * Returns true if the page was previously dirty. | |
882 | * | |
883 | * This is for preparing to put the page under writeout. We leave the page | |
884 | * tagged as dirty in the radix tree so that a concurrent write-for-sync | |
885 | * can discover it via a PAGECACHE_TAG_DIRTY walk. The ->writepage | |
886 | * implementation will run either set_page_writeback() or set_page_dirty(), | |
887 | * at which stage we bring the page's dirty flag and radix-tree dirty tag | |
888 | * back into sync. | |
889 | * | |
890 | * This incoherency between the page's dirty flag and radix-tree tag is | |
891 | * unfortunate, but it only exists while the page is locked. | |
892 | */ | |
893 | int clear_page_dirty_for_io(struct page *page) | |
894 | { | |
895 | struct address_space *mapping = page_mapping(page); | |
896 | ||
8c08540f AM |
897 | if (!mapping) |
898 | return TestClearPageDirty(page); | |
899 | ||
900 | if (TestClearPageDirty(page)) { | |
901 | if (mapping_cap_account_dirty(mapping)) { | |
902 | page_mkclean(page); | |
903 | dec_zone_page_state(page, NR_FILE_DIRTY); | |
1da177e4 | 904 | } |
8c08540f | 905 | return 1; |
1da177e4 | 906 | } |
8c08540f | 907 | return 0; |
1da177e4 | 908 | } |
58bb01a9 | 909 | EXPORT_SYMBOL(clear_page_dirty_for_io); |
1da177e4 LT |
910 | |
911 | int test_clear_page_writeback(struct page *page) | |
912 | { | |
913 | struct address_space *mapping = page_mapping(page); | |
914 | int ret; | |
915 | ||
916 | if (mapping) { | |
917 | unsigned long flags; | |
918 | ||
919 | write_lock_irqsave(&mapping->tree_lock, flags); | |
920 | ret = TestClearPageWriteback(page); | |
921 | if (ret) | |
922 | radix_tree_tag_clear(&mapping->page_tree, | |
923 | page_index(page), | |
924 | PAGECACHE_TAG_WRITEBACK); | |
925 | write_unlock_irqrestore(&mapping->tree_lock, flags); | |
926 | } else { | |
927 | ret = TestClearPageWriteback(page); | |
928 | } | |
929 | return ret; | |
930 | } | |
931 | ||
932 | int test_set_page_writeback(struct page *page) | |
933 | { | |
934 | struct address_space *mapping = page_mapping(page); | |
935 | int ret; | |
936 | ||
937 | if (mapping) { | |
938 | unsigned long flags; | |
939 | ||
940 | write_lock_irqsave(&mapping->tree_lock, flags); | |
941 | ret = TestSetPageWriteback(page); | |
942 | if (!ret) | |
943 | radix_tree_tag_set(&mapping->page_tree, | |
944 | page_index(page), | |
945 | PAGECACHE_TAG_WRITEBACK); | |
946 | if (!PageDirty(page)) | |
947 | radix_tree_tag_clear(&mapping->page_tree, | |
948 | page_index(page), | |
949 | PAGECACHE_TAG_DIRTY); | |
950 | write_unlock_irqrestore(&mapping->tree_lock, flags); | |
951 | } else { | |
952 | ret = TestSetPageWriteback(page); | |
953 | } | |
954 | return ret; | |
955 | ||
956 | } | |
957 | EXPORT_SYMBOL(test_set_page_writeback); | |
958 | ||
959 | /* | |
960 | * Return true if any of the pages in the mapping are marged with the | |
961 | * passed tag. | |
962 | */ | |
963 | int mapping_tagged(struct address_space *mapping, int tag) | |
964 | { | |
965 | unsigned long flags; | |
966 | int ret; | |
967 | ||
968 | read_lock_irqsave(&mapping->tree_lock, flags); | |
969 | ret = radix_tree_tagged(&mapping->page_tree, tag); | |
970 | read_unlock_irqrestore(&mapping->tree_lock, flags); | |
971 | return ret; | |
972 | } | |
973 | EXPORT_SYMBOL(mapping_tagged); |