page-writeback.c 28 KB

123456789101112131415161718192021222324252627282930313233343536373839404142434445464748495051525354555657585960616263646566676869707172737475767778798081828384858687888990919293949596979899100101102103104105106107108109110111112113114115116117118119120121122123124125126127128129130131132133134135136137138139140141142143144145146147148149150151152153154155156157158159160161162163164165166167168169170171172173174175176177178179180181182183184185186187188189190191192193194195196197198199200201202203204205206207208209210211212213214215216217218219220221222223224225226227228229230231232233234235236237238239240241242243244245246247248249250251252253254255256257258259260261262263264265266267268269270271272273274275276277278279280281282283284285286287288289290291292293294295296297298299300301302303304305306307308309310311312313314315316317318319320321322323324325326327328329330331332333334335336337338339340341342343344345346347348349350351352353354355356357358359360361362363364365366367368369370371372373374375376377378379380381382383384385386387388389390391392393394395396397398399400401402403404405406407408409410411412413414415416417418419420421422423424425426427428429430431432433434435436437438439440441442443444445446447448449450451452453454455456457458459460461462463464465466467468469470471472473474475476477478479480481482483484485486487488489490491492493494495496497498499500501502503504505506507508509510511512513514515516517518519520521522523524525526527528529530531532533534535536537538539540541542543544545546547548549550551552553554555556557558559560561562563564565566567568569570571572573574575576577578579580581582583584585586587588589590591592593594595596597598599600601602603604605606607608609610611612613614615616617618619620621622623624625626627628629630631632633634635636637638639640641642643644645646647648649650651652653654655656657658659660661662663664665666667668669670671672673674675676677678679680681682683684685686687688689690691692693694695696697698699700701702703704705706707708709710711712713714715716717718719720721722723724725726727728729730731732733734735736737738739740741742743744745746747748749750751752753754755756757758759760761762763764765766767768769770771772773774775776777778779780781782783784785786787788789790791792793794795796797798799800801802803804805806807808809810811812813814815816817818819820821822823824825826827828829830831832833834835836837838839840841842843844845846847848849850851852853854855856857858859860861862863864865866867868869870871872873874875876877878879880881882883884885886887888889890891892893894895896897898899900901902903904905906907908909910911912913914915916917918919920921922923924925926927928929930931932933934935936937938939940941942943944945946947948949950951952953954955956957958959960961962963964965966967968969970971972973974975976977978979980981
  1. /*
  2. * mm/page-writeback.c
  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. #include <linux/kernel.h>
  13. #include <linux/module.h>
  14. #include <linux/spinlock.h>
  15. #include <linux/fs.h>
  16. #include <linux/mm.h>
  17. #include <linux/swap.h>
  18. #include <linux/slab.h>
  19. #include <linux/pagemap.h>
  20. #include <linux/writeback.h>
  21. #include <linux/init.h>
  22. #include <linux/backing-dev.h>
  23. #include <linux/task_io_accounting_ops.h>
  24. #include <linux/blkdev.h>
  25. #include <linux/mpage.h>
  26. #include <linux/rmap.h>
  27. #include <linux/percpu.h>
  28. #include <linux/notifier.h>
  29. #include <linux/smp.h>
  30. #include <linux/sysctl.h>
  31. #include <linux/cpu.h>
  32. #include <linux/syscalls.h>
  33. #include <linux/buffer_head.h>
  34. #include <linux/pagevec.h>
  35. /*
  36. * The maximum number of pages to writeout in a single bdflush/kupdate
  37. * operation. We do this so we don't hold I_LOCK against an inode for
  38. * enormous amounts of time, which would block a userspace task which has
  39. * been forced to throttle against that inode. Also, the code reevaluates
  40. * the dirty each time it has written this many pages.
  41. */
  42. #define MAX_WRITEBACK_PAGES 1024
  43. /*
  44. * After a CPU has dirtied this many pages, balance_dirty_pages_ratelimited
  45. * will look to see if it needs to force writeback or throttling.
  46. */
  47. static long ratelimit_pages = 32;
  48. static int dirty_exceeded __cacheline_aligned_in_smp; /* Dirty mem may be over limit */
  49. /*
  50. * When balance_dirty_pages decides that the caller needs to perform some
  51. * non-background writeback, this is how many pages it will attempt to write.
  52. * It should be somewhat larger than RATELIMIT_PAGES to ensure that reasonably
  53. * large amounts of I/O are submitted.
  54. */
  55. static inline long sync_writeback_pages(void)
  56. {
  57. return ratelimit_pages + ratelimit_pages / 2;
  58. }
  59. /* The following parameters are exported via /proc/sys/vm */
  60. /*
  61. * Start background writeback (via pdflush) at this percentage
  62. */
  63. int dirty_background_ratio = 10;
  64. /*
  65. * The generator of dirty data starts writeback at this percentage
  66. */
  67. int vm_dirty_ratio = 40;
  68. /*
  69. * The interval between `kupdate'-style writebacks, in jiffies
  70. */
  71. int dirty_writeback_interval = 5 * HZ;
  72. /*
  73. * The longest number of jiffies for which data is allowed to remain dirty
  74. */
  75. int dirty_expire_interval = 30 * HZ;
  76. /*
  77. * Flag that makes the machine dump writes/reads and block dirtyings.
  78. */
  79. int block_dump;
  80. /*
  81. * Flag that puts the machine in "laptop mode". Doubles as a timeout in jiffies:
  82. * a full sync is triggered after this time elapses without any disk activity.
  83. */
  84. int laptop_mode;
  85. EXPORT_SYMBOL(laptop_mode);
  86. /* End of sysctl-exported parameters */
  87. static void background_writeout(unsigned long _min_pages);
  88. /*
  89. * Work out the current dirty-memory clamping and background writeout
  90. * thresholds.
  91. *
  92. * The main aim here is to lower them aggressively if there is a lot of mapped
  93. * memory around. To avoid stressing page reclaim with lots of unreclaimable
  94. * pages. It is better to clamp down on writers than to start swapping, and
  95. * performing lots of scanning.
  96. *
  97. * We only allow 1/2 of the currently-unmapped memory to be dirtied.
  98. *
  99. * We don't permit the clamping level to fall below 5% - that is getting rather
  100. * excessive.
  101. *
  102. * We make sure that the background writeout level is below the adjusted
  103. * clamping level.
  104. */
  105. static void
  106. get_dirty_limits(long *pbackground, long *pdirty,
  107. struct address_space *mapping)
  108. {
  109. int background_ratio; /* Percentages */
  110. int dirty_ratio;
  111. int unmapped_ratio;
  112. long background;
  113. long dirty;
  114. unsigned long available_memory = vm_total_pages;
  115. struct task_struct *tsk;
  116. #ifdef CONFIG_HIGHMEM
  117. /*
  118. * We always exclude high memory from our count.
  119. */
  120. available_memory -= totalhigh_pages;
  121. #endif
  122. unmapped_ratio = 100 - ((global_page_state(NR_FILE_MAPPED) +
  123. global_page_state(NR_ANON_PAGES)) * 100) /
  124. vm_total_pages;
  125. dirty_ratio = vm_dirty_ratio;
  126. if (dirty_ratio > unmapped_ratio / 2)
  127. dirty_ratio = unmapped_ratio / 2;
  128. if (dirty_ratio < 5)
  129. dirty_ratio = 5;
  130. background_ratio = dirty_background_ratio;
  131. if (background_ratio >= dirty_ratio)
  132. background_ratio = dirty_ratio / 2;
  133. background = (background_ratio * available_memory) / 100;
  134. dirty = (dirty_ratio * available_memory) / 100;
  135. tsk = current;
  136. if (tsk->flags & PF_LESS_THROTTLE || rt_task(tsk)) {
  137. background += background / 4;
  138. dirty += dirty / 4;
  139. }
  140. *pbackground = background;
  141. *pdirty = dirty;
  142. }
  143. /*
  144. * balance_dirty_pages() must be called by processes which are generating dirty
  145. * data. It looks at the number of dirty pages in the machine and will force
  146. * the caller to perform writeback if the system is over `vm_dirty_ratio'.
  147. * If we're over `background_thresh' then pdflush is woken to perform some
  148. * writeout.
  149. */
  150. static void balance_dirty_pages(struct address_space *mapping)
  151. {
  152. long nr_reclaimable;
  153. long background_thresh;
  154. long dirty_thresh;
  155. unsigned long pages_written = 0;
  156. unsigned long write_chunk = sync_writeback_pages();
  157. struct backing_dev_info *bdi = mapping->backing_dev_info;
  158. for (;;) {
  159. struct writeback_control wbc = {
  160. .bdi = bdi,
  161. .sync_mode = WB_SYNC_NONE,
  162. .older_than_this = NULL,
  163. .nr_to_write = write_chunk,
  164. .range_cyclic = 1,
  165. };
  166. get_dirty_limits(&background_thresh, &dirty_thresh, mapping);
  167. nr_reclaimable = global_page_state(NR_FILE_DIRTY) +
  168. global_page_state(NR_UNSTABLE_NFS);
  169. if (nr_reclaimable + global_page_state(NR_WRITEBACK) <=
  170. dirty_thresh)
  171. break;
  172. if (!dirty_exceeded)
  173. dirty_exceeded = 1;
  174. /* Note: nr_reclaimable denotes nr_dirty + nr_unstable.
  175. * Unstable writes are a feature of certain networked
  176. * filesystems (i.e. NFS) in which data may have been
  177. * written to the server's write cache, but has not yet
  178. * been flushed to permanent storage.
  179. */
  180. if (nr_reclaimable) {
  181. writeback_inodes(&wbc);
  182. get_dirty_limits(&background_thresh,
  183. &dirty_thresh, mapping);
  184. nr_reclaimable = global_page_state(NR_FILE_DIRTY) +
  185. global_page_state(NR_UNSTABLE_NFS);
  186. if (nr_reclaimable +
  187. global_page_state(NR_WRITEBACK)
  188. <= dirty_thresh)
  189. break;
  190. pages_written += write_chunk - wbc.nr_to_write;
  191. if (pages_written >= write_chunk)
  192. break; /* We've done our duty */
  193. }
  194. congestion_wait(WRITE, HZ/10);
  195. }
  196. if (nr_reclaimable + global_page_state(NR_WRITEBACK)
  197. <= dirty_thresh && dirty_exceeded)
  198. dirty_exceeded = 0;
  199. if (writeback_in_progress(bdi))
  200. return; /* pdflush is already working this queue */
  201. /*
  202. * In laptop mode, we wait until hitting the higher threshold before
  203. * starting background writeout, and then write out all the way down
  204. * to the lower threshold. So slow writers cause minimal disk activity.
  205. *
  206. * In normal mode, we start background writeout at the lower
  207. * background_thresh, to keep the amount of dirty memory low.
  208. */
  209. if ((laptop_mode && pages_written) ||
  210. (!laptop_mode && (nr_reclaimable > background_thresh)))
  211. pdflush_operation(background_writeout, 0);
  212. }
  213. void set_page_dirty_balance(struct page *page)
  214. {
  215. if (set_page_dirty(page)) {
  216. struct address_space *mapping = page_mapping(page);
  217. if (mapping)
  218. balance_dirty_pages_ratelimited(mapping);
  219. }
  220. }
  221. /**
  222. * balance_dirty_pages_ratelimited_nr - balance dirty memory state
  223. * @mapping: address_space which was dirtied
  224. * @nr_pages_dirtied: number of pages which the caller has just dirtied
  225. *
  226. * Processes which are dirtying memory should call in here once for each page
  227. * which was newly dirtied. The function will periodically check the system's
  228. * dirty state and will initiate writeback if needed.
  229. *
  230. * On really big machines, get_writeback_state is expensive, so try to avoid
  231. * calling it too often (ratelimiting). But once we're over the dirty memory
  232. * limit we decrease the ratelimiting by a lot, to prevent individual processes
  233. * from overshooting the limit by (ratelimit_pages) each.
  234. */
  235. void balance_dirty_pages_ratelimited_nr(struct address_space *mapping,
  236. unsigned long nr_pages_dirtied)
  237. {
  238. static DEFINE_PER_CPU(unsigned long, ratelimits) = 0;
  239. unsigned long ratelimit;
  240. unsigned long *p;
  241. ratelimit = ratelimit_pages;
  242. if (dirty_exceeded)
  243. ratelimit = 8;
  244. /*
  245. * Check the rate limiting. Also, we do not want to throttle real-time
  246. * tasks in balance_dirty_pages(). Period.
  247. */
  248. preempt_disable();
  249. p = &__get_cpu_var(ratelimits);
  250. *p += nr_pages_dirtied;
  251. if (unlikely(*p >= ratelimit)) {
  252. *p = 0;
  253. preempt_enable();
  254. balance_dirty_pages(mapping);
  255. return;
  256. }
  257. preempt_enable();
  258. }
  259. EXPORT_SYMBOL(balance_dirty_pages_ratelimited_nr);
  260. void throttle_vm_writeout(gfp_t gfp_mask)
  261. {
  262. long background_thresh;
  263. long dirty_thresh;
  264. if ((gfp_mask & (__GFP_FS|__GFP_IO)) != (__GFP_FS|__GFP_IO)) {
  265. /*
  266. * The caller might hold locks which can prevent IO completion
  267. * or progress in the filesystem. So we cannot just sit here
  268. * waiting for IO to complete.
  269. */
  270. congestion_wait(WRITE, HZ/10);
  271. return;
  272. }
  273. for ( ; ; ) {
  274. get_dirty_limits(&background_thresh, &dirty_thresh, NULL);
  275. /*
  276. * Boost the allowable dirty threshold a bit for page
  277. * allocators so they don't get DoS'ed by heavy writers
  278. */
  279. dirty_thresh += dirty_thresh / 10; /* wheeee... */
  280. if (global_page_state(NR_UNSTABLE_NFS) +
  281. global_page_state(NR_WRITEBACK) <= dirty_thresh)
  282. break;
  283. congestion_wait(WRITE, HZ/10);
  284. }
  285. }
  286. /*
  287. * writeback at least _min_pages, and keep writing until the amount of dirty
  288. * memory is less than the background threshold, or until we're all clean.
  289. */
  290. static void background_writeout(unsigned long _min_pages)
  291. {
  292. long min_pages = _min_pages;
  293. struct writeback_control wbc = {
  294. .bdi = NULL,
  295. .sync_mode = WB_SYNC_NONE,
  296. .older_than_this = NULL,
  297. .nr_to_write = 0,
  298. .nonblocking = 1,
  299. .range_cyclic = 1,
  300. };
  301. for ( ; ; ) {
  302. long background_thresh;
  303. long dirty_thresh;
  304. get_dirty_limits(&background_thresh, &dirty_thresh, NULL);
  305. if (global_page_state(NR_FILE_DIRTY) +
  306. global_page_state(NR_UNSTABLE_NFS) < background_thresh
  307. && min_pages <= 0)
  308. break;
  309. wbc.encountered_congestion = 0;
  310. wbc.nr_to_write = MAX_WRITEBACK_PAGES;
  311. wbc.pages_skipped = 0;
  312. writeback_inodes(&wbc);
  313. min_pages -= MAX_WRITEBACK_PAGES - wbc.nr_to_write;
  314. if (wbc.nr_to_write > 0 || wbc.pages_skipped > 0) {
  315. /* Wrote less than expected */
  316. congestion_wait(WRITE, HZ/10);
  317. if (!wbc.encountered_congestion)
  318. break;
  319. }
  320. }
  321. }
  322. /*
  323. * Start writeback of `nr_pages' pages. If `nr_pages' is zero, write back
  324. * the whole world. Returns 0 if a pdflush thread was dispatched. Returns
  325. * -1 if all pdflush threads were busy.
  326. */
  327. int wakeup_pdflush(long nr_pages)
  328. {
  329. if (nr_pages == 0)
  330. nr_pages = global_page_state(NR_FILE_DIRTY) +
  331. global_page_state(NR_UNSTABLE_NFS);
  332. return pdflush_operation(background_writeout, nr_pages);
  333. }
  334. static void wb_timer_fn(unsigned long unused);
  335. static void laptop_timer_fn(unsigned long unused);
  336. static DEFINE_TIMER(wb_timer, wb_timer_fn, 0, 0);
  337. static DEFINE_TIMER(laptop_mode_wb_timer, laptop_timer_fn, 0, 0);
  338. /*
  339. * Periodic writeback of "old" data.
  340. *
  341. * Define "old": the first time one of an inode's pages is dirtied, we mark the
  342. * dirtying-time in the inode's address_space. So this periodic writeback code
  343. * just walks the superblock inode list, writing back any inodes which are
  344. * older than a specific point in time.
  345. *
  346. * Try to run once per dirty_writeback_interval. But if a writeback event
  347. * takes longer than a dirty_writeback_interval interval, then leave a
  348. * one-second gap.
  349. *
  350. * older_than_this takes precedence over nr_to_write. So we'll only write back
  351. * all dirty pages if they are all attached to "old" mappings.
  352. */
  353. static void wb_kupdate(unsigned long arg)
  354. {
  355. unsigned long oldest_jif;
  356. unsigned long start_jif;
  357. unsigned long next_jif;
  358. long nr_to_write;
  359. struct writeback_control wbc = {
  360. .bdi = NULL,
  361. .sync_mode = WB_SYNC_NONE,
  362. .older_than_this = &oldest_jif,
  363. .nr_to_write = 0,
  364. .nonblocking = 1,
  365. .for_kupdate = 1,
  366. .range_cyclic = 1,
  367. };
  368. sync_supers();
  369. oldest_jif = jiffies - dirty_expire_interval;
  370. start_jif = jiffies;
  371. next_jif = start_jif + dirty_writeback_interval;
  372. nr_to_write = global_page_state(NR_FILE_DIRTY) +
  373. global_page_state(NR_UNSTABLE_NFS) +
  374. (inodes_stat.nr_inodes - inodes_stat.nr_unused);
  375. while (nr_to_write > 0) {
  376. wbc.encountered_congestion = 0;
  377. wbc.nr_to_write = MAX_WRITEBACK_PAGES;
  378. writeback_inodes(&wbc);
  379. if (wbc.nr_to_write > 0) {
  380. if (wbc.encountered_congestion)
  381. congestion_wait(WRITE, HZ/10);
  382. else
  383. break; /* All the old data is written */
  384. }
  385. nr_to_write -= MAX_WRITEBACK_PAGES - wbc.nr_to_write;
  386. }
  387. if (time_before(next_jif, jiffies + HZ))
  388. next_jif = jiffies + HZ;
  389. if (dirty_writeback_interval)
  390. mod_timer(&wb_timer, next_jif);
  391. }
  392. /*
  393. * sysctl handler for /proc/sys/vm/dirty_writeback_centisecs
  394. */
  395. int dirty_writeback_centisecs_handler(ctl_table *table, int write,
  396. struct file *file, void __user *buffer, size_t *length, loff_t *ppos)
  397. {
  398. proc_dointvec_userhz_jiffies(table, write, file, buffer, length, ppos);
  399. if (dirty_writeback_interval) {
  400. mod_timer(&wb_timer,
  401. jiffies + dirty_writeback_interval);
  402. } else {
  403. del_timer(&wb_timer);
  404. }
  405. return 0;
  406. }
  407. static void wb_timer_fn(unsigned long unused)
  408. {
  409. if (pdflush_operation(wb_kupdate, 0) < 0)
  410. mod_timer(&wb_timer, jiffies + HZ); /* delay 1 second */
  411. }
  412. static void laptop_flush(unsigned long unused)
  413. {
  414. sys_sync();
  415. }
  416. static void laptop_timer_fn(unsigned long unused)
  417. {
  418. pdflush_operation(laptop_flush, 0);
  419. }
  420. /*
  421. * We've spun up the disk and we're in laptop mode: schedule writeback
  422. * of all dirty data a few seconds from now. If the flush is already scheduled
  423. * then push it back - the user is still using the disk.
  424. */
  425. void laptop_io_completion(void)
  426. {
  427. mod_timer(&laptop_mode_wb_timer, jiffies + laptop_mode);
  428. }
  429. /*
  430. * We're in laptop mode and we've just synced. The sync's writes will have
  431. * caused another writeback to be scheduled by laptop_io_completion.
  432. * Nothing needs to be written back anymore, so we unschedule the writeback.
  433. */
  434. void laptop_sync_completion(void)
  435. {
  436. del_timer(&laptop_mode_wb_timer);
  437. }
  438. /*
  439. * If ratelimit_pages is too high then we can get into dirty-data overload
  440. * if a large number of processes all perform writes at the same time.
  441. * If it is too low then SMP machines will call the (expensive)
  442. * get_writeback_state too often.
  443. *
  444. * Here we set ratelimit_pages to a level which ensures that when all CPUs are
  445. * dirtying in parallel, we cannot go more than 3% (1/32) over the dirty memory
  446. * thresholds before writeback cuts in.
  447. *
  448. * But the limit should not be set too high. Because it also controls the
  449. * amount of memory which the balance_dirty_pages() caller has to write back.
  450. * If this is too large then the caller will block on the IO queue all the
  451. * time. So limit it to four megabytes - the balance_dirty_pages() caller
  452. * will write six megabyte chunks, max.
  453. */
  454. void writeback_set_ratelimit(void)
  455. {
  456. ratelimit_pages = vm_total_pages / (num_online_cpus() * 32);
  457. if (ratelimit_pages < 16)
  458. ratelimit_pages = 16;
  459. if (ratelimit_pages * PAGE_CACHE_SIZE > 4096 * 1024)
  460. ratelimit_pages = (4096 * 1024) / PAGE_CACHE_SIZE;
  461. }
  462. static int __cpuinit
  463. ratelimit_handler(struct notifier_block *self, unsigned long u, void *v)
  464. {
  465. writeback_set_ratelimit();
  466. return NOTIFY_DONE;
  467. }
  468. static struct notifier_block __cpuinitdata ratelimit_nb = {
  469. .notifier_call = ratelimit_handler,
  470. .next = NULL,
  471. };
  472. /*
  473. * Called early on to tune the page writeback dirty limits.
  474. *
  475. * We used to scale dirty pages according to how total memory
  476. * related to pages that could be allocated for buffers (by
  477. * comparing nr_free_buffer_pages() to vm_total_pages.
  478. *
  479. * However, that was when we used "dirty_ratio" to scale with
  480. * all memory, and we don't do that any more. "dirty_ratio"
  481. * is now applied to total non-HIGHPAGE memory (by subtracting
  482. * totalhigh_pages from vm_total_pages), and as such we can't
  483. * get into the old insane situation any more where we had
  484. * large amounts of dirty pages compared to a small amount of
  485. * non-HIGHMEM memory.
  486. *
  487. * But we might still want to scale the dirty_ratio by how
  488. * much memory the box has..
  489. */
  490. void __init page_writeback_init(void)
  491. {
  492. mod_timer(&wb_timer, jiffies + dirty_writeback_interval);
  493. writeback_set_ratelimit();
  494. register_cpu_notifier(&ratelimit_nb);
  495. }
  496. /**
  497. * generic_writepages - walk the list of dirty pages of the given address space and writepage() all of them.
  498. * @mapping: address space structure to write
  499. * @wbc: subtract the number of written pages from *@wbc->nr_to_write
  500. *
  501. * This is a library function, which implements the writepages()
  502. * address_space_operation.
  503. *
  504. * If a page is already under I/O, generic_writepages() skips it, even
  505. * if it's dirty. This is desirable behaviour for memory-cleaning writeback,
  506. * but it is INCORRECT for data-integrity system calls such as fsync(). fsync()
  507. * and msync() need to guarantee that all the data which was dirty at the time
  508. * the call was made get new I/O started against them. If wbc->sync_mode is
  509. * WB_SYNC_ALL then we were called for data integrity and we must wait for
  510. * existing IO to complete.
  511. *
  512. * Derived from mpage_writepages() - if you fix this you should check that
  513. * also!
  514. */
  515. int generic_writepages(struct address_space *mapping,
  516. struct writeback_control *wbc)
  517. {
  518. struct backing_dev_info *bdi = mapping->backing_dev_info;
  519. int ret = 0;
  520. int done = 0;
  521. int (*writepage)(struct page *page, struct writeback_control *wbc);
  522. struct pagevec pvec;
  523. int nr_pages;
  524. pgoff_t index;
  525. pgoff_t end; /* Inclusive */
  526. int scanned = 0;
  527. int range_whole = 0;
  528. if (wbc->nonblocking && bdi_write_congested(bdi)) {
  529. wbc->encountered_congestion = 1;
  530. return 0;
  531. }
  532. writepage = mapping->a_ops->writepage;
  533. /* deal with chardevs and other special file */
  534. if (!writepage)
  535. return 0;
  536. pagevec_init(&pvec, 0);
  537. if (wbc->range_cyclic) {
  538. index = mapping->writeback_index; /* Start from prev offset */
  539. end = -1;
  540. } else {
  541. index = wbc->range_start >> PAGE_CACHE_SHIFT;
  542. end = wbc->range_end >> PAGE_CACHE_SHIFT;
  543. if (wbc->range_start == 0 && wbc->range_end == LLONG_MAX)
  544. range_whole = 1;
  545. scanned = 1;
  546. }
  547. retry:
  548. while (!done && (index <= end) &&
  549. (nr_pages = pagevec_lookup_tag(&pvec, mapping, &index,
  550. PAGECACHE_TAG_DIRTY,
  551. min(end - index, (pgoff_t)PAGEVEC_SIZE-1) + 1))) {
  552. unsigned i;
  553. scanned = 1;
  554. for (i = 0; i < nr_pages; i++) {
  555. struct page *page = pvec.pages[i];
  556. /*
  557. * At this point we hold neither mapping->tree_lock nor
  558. * lock on the page itself: the page may be truncated or
  559. * invalidated (changing page->mapping to NULL), or even
  560. * swizzled back from swapper_space to tmpfs file
  561. * mapping
  562. */
  563. lock_page(page);
  564. if (unlikely(page->mapping != mapping)) {
  565. unlock_page(page);
  566. continue;
  567. }
  568. if (!wbc->range_cyclic && page->index > end) {
  569. done = 1;
  570. unlock_page(page);
  571. continue;
  572. }
  573. if (wbc->sync_mode != WB_SYNC_NONE)
  574. wait_on_page_writeback(page);
  575. if (PageWriteback(page) ||
  576. !clear_page_dirty_for_io(page)) {
  577. unlock_page(page);
  578. continue;
  579. }
  580. ret = (*writepage)(page, wbc);
  581. if (ret) {
  582. if (ret == -ENOSPC)
  583. set_bit(AS_ENOSPC, &mapping->flags);
  584. else
  585. set_bit(AS_EIO, &mapping->flags);
  586. }
  587. if (unlikely(ret == AOP_WRITEPAGE_ACTIVATE))
  588. unlock_page(page);
  589. if (ret || (--(wbc->nr_to_write) <= 0))
  590. done = 1;
  591. if (wbc->nonblocking && bdi_write_congested(bdi)) {
  592. wbc->encountered_congestion = 1;
  593. done = 1;
  594. }
  595. }
  596. pagevec_release(&pvec);
  597. cond_resched();
  598. }
  599. if (!scanned && !done) {
  600. /*
  601. * We hit the last page and there is more work to be done: wrap
  602. * back to the start of the file
  603. */
  604. scanned = 1;
  605. index = 0;
  606. goto retry;
  607. }
  608. if (wbc->range_cyclic || (range_whole && wbc->nr_to_write > 0))
  609. mapping->writeback_index = index;
  610. return ret;
  611. }
  612. EXPORT_SYMBOL(generic_writepages);
  613. int do_writepages(struct address_space *mapping, struct writeback_control *wbc)
  614. {
  615. int ret;
  616. if (wbc->nr_to_write <= 0)
  617. return 0;
  618. wbc->for_writepages = 1;
  619. if (mapping->a_ops->writepages)
  620. ret = mapping->a_ops->writepages(mapping, wbc);
  621. else
  622. ret = generic_writepages(mapping, wbc);
  623. wbc->for_writepages = 0;
  624. return ret;
  625. }
  626. /**
  627. * write_one_page - write out a single page and optionally wait on I/O
  628. * @page: the page to write
  629. * @wait: if true, wait on writeout
  630. *
  631. * The page must be locked by the caller and will be unlocked upon return.
  632. *
  633. * write_one_page() returns a negative error code if I/O failed.
  634. */
  635. int write_one_page(struct page *page, int wait)
  636. {
  637. struct address_space *mapping = page->mapping;
  638. int ret = 0;
  639. struct writeback_control wbc = {
  640. .sync_mode = WB_SYNC_ALL,
  641. .nr_to_write = 1,
  642. };
  643. BUG_ON(!PageLocked(page));
  644. if (wait)
  645. wait_on_page_writeback(page);
  646. if (clear_page_dirty_for_io(page)) {
  647. page_cache_get(page);
  648. ret = mapping->a_ops->writepage(page, &wbc);
  649. if (ret == 0 && wait) {
  650. wait_on_page_writeback(page);
  651. if (PageError(page))
  652. ret = -EIO;
  653. }
  654. page_cache_release(page);
  655. } else {
  656. unlock_page(page);
  657. }
  658. return ret;
  659. }
  660. EXPORT_SYMBOL(write_one_page);
  661. /*
  662. * For address_spaces which do not use buffers nor write back.
  663. */
  664. int __set_page_dirty_no_writeback(struct page *page)
  665. {
  666. if (!PageDirty(page))
  667. SetPageDirty(page);
  668. return 0;
  669. }
  670. /*
  671. * For address_spaces which do not use buffers. Just tag the page as dirty in
  672. * its radix tree.
  673. *
  674. * This is also used when a single buffer is being dirtied: we want to set the
  675. * page dirty in that case, but not all the buffers. This is a "bottom-up"
  676. * dirtying, whereas __set_page_dirty_buffers() is a "top-down" dirtying.
  677. *
  678. * Most callers have locked the page, which pins the address_space in memory.
  679. * But zap_pte_range() does not lock the page, however in that case the
  680. * mapping is pinned by the vma's ->vm_file reference.
  681. *
  682. * We take care to handle the case where the page was truncated from the
  683. * mapping by re-checking page_mapping() insode tree_lock.
  684. */
  685. int __set_page_dirty_nobuffers(struct page *page)
  686. {
  687. if (!TestSetPageDirty(page)) {
  688. struct address_space *mapping = page_mapping(page);
  689. struct address_space *mapping2;
  690. if (!mapping)
  691. return 1;
  692. write_lock_irq(&mapping->tree_lock);
  693. mapping2 = page_mapping(page);
  694. if (mapping2) { /* Race with truncate? */
  695. BUG_ON(mapping2 != mapping);
  696. if (mapping_cap_account_dirty(mapping)) {
  697. __inc_zone_page_state(page, NR_FILE_DIRTY);
  698. task_io_account_write(PAGE_CACHE_SIZE);
  699. }
  700. radix_tree_tag_set(&mapping->page_tree,
  701. page_index(page), PAGECACHE_TAG_DIRTY);
  702. }
  703. write_unlock_irq(&mapping->tree_lock);
  704. if (mapping->host) {
  705. /* !PageAnon && !swapper_space */
  706. __mark_inode_dirty(mapping->host, I_DIRTY_PAGES);
  707. }
  708. return 1;
  709. }
  710. return 0;
  711. }
  712. EXPORT_SYMBOL(__set_page_dirty_nobuffers);
  713. /*
  714. * When a writepage implementation decides that it doesn't want to write this
  715. * page for some reason, it should redirty the locked page via
  716. * redirty_page_for_writepage() and it should then unlock the page and return 0
  717. */
  718. int redirty_page_for_writepage(struct writeback_control *wbc, struct page *page)
  719. {
  720. wbc->pages_skipped++;
  721. return __set_page_dirty_nobuffers(page);
  722. }
  723. EXPORT_SYMBOL(redirty_page_for_writepage);
  724. /*
  725. * If the mapping doesn't provide a set_page_dirty a_op, then
  726. * just fall through and assume that it wants buffer_heads.
  727. */
  728. int fastcall set_page_dirty(struct page *page)
  729. {
  730. struct address_space *mapping = page_mapping(page);
  731. if (likely(mapping)) {
  732. int (*spd)(struct page *) = mapping->a_ops->set_page_dirty;
  733. #ifdef CONFIG_BLOCK
  734. if (!spd)
  735. spd = __set_page_dirty_buffers;
  736. #endif
  737. return (*spd)(page);
  738. }
  739. if (!PageDirty(page)) {
  740. if (!TestSetPageDirty(page))
  741. return 1;
  742. }
  743. return 0;
  744. }
  745. EXPORT_SYMBOL(set_page_dirty);
  746. /*
  747. * set_page_dirty() is racy if the caller has no reference against
  748. * page->mapping->host, and if the page is unlocked. This is because another
  749. * CPU could truncate the page off the mapping and then free the mapping.
  750. *
  751. * Usually, the page _is_ locked, or the caller is a user-space process which
  752. * holds a reference on the inode by having an open file.
  753. *
  754. * In other cases, the page should be locked before running set_page_dirty().
  755. */
  756. int set_page_dirty_lock(struct page *page)
  757. {
  758. int ret;
  759. lock_page_nosync(page);
  760. ret = set_page_dirty(page);
  761. unlock_page(page);
  762. return ret;
  763. }
  764. EXPORT_SYMBOL(set_page_dirty_lock);
  765. /*
  766. * Clear a page's dirty flag, while caring for dirty memory accounting.
  767. * Returns true if the page was previously dirty.
  768. *
  769. * This is for preparing to put the page under writeout. We leave the page
  770. * tagged as dirty in the radix tree so that a concurrent write-for-sync
  771. * can discover it via a PAGECACHE_TAG_DIRTY walk. The ->writepage
  772. * implementation will run either set_page_writeback() or set_page_dirty(),
  773. * at which stage we bring the page's dirty flag and radix-tree dirty tag
  774. * back into sync.
  775. *
  776. * This incoherency between the page's dirty flag and radix-tree tag is
  777. * unfortunate, but it only exists while the page is locked.
  778. */
  779. int clear_page_dirty_for_io(struct page *page)
  780. {
  781. struct address_space *mapping = page_mapping(page);
  782. if (mapping && mapping_cap_account_dirty(mapping)) {
  783. /*
  784. * Yes, Virginia, this is indeed insane.
  785. *
  786. * We use this sequence to make sure that
  787. * (a) we account for dirty stats properly
  788. * (b) we tell the low-level filesystem to
  789. * mark the whole page dirty if it was
  790. * dirty in a pagetable. Only to then
  791. * (c) clean the page again and return 1 to
  792. * cause the writeback.
  793. *
  794. * This way we avoid all nasty races with the
  795. * dirty bit in multiple places and clearing
  796. * them concurrently from different threads.
  797. *
  798. * Note! Normally the "set_page_dirty(page)"
  799. * has no effect on the actual dirty bit - since
  800. * that will already usually be set. But we
  801. * need the side effects, and it can help us
  802. * avoid races.
  803. *
  804. * We basically use the page "master dirty bit"
  805. * as a serialization point for all the different
  806. * threads doing their things.
  807. *
  808. * FIXME! We still have a race here: if somebody
  809. * adds the page back to the page tables in
  810. * between the "page_mkclean()" and the "TestClearPageDirty()",
  811. * we might have it mapped without the dirty bit set.
  812. */
  813. if (page_mkclean(page))
  814. set_page_dirty(page);
  815. if (TestClearPageDirty(page)) {
  816. dec_zone_page_state(page, NR_FILE_DIRTY);
  817. return 1;
  818. }
  819. return 0;
  820. }
  821. return TestClearPageDirty(page);
  822. }
  823. EXPORT_SYMBOL(clear_page_dirty_for_io);
  824. int test_clear_page_writeback(struct page *page)
  825. {
  826. struct address_space *mapping = page_mapping(page);
  827. int ret;
  828. if (mapping) {
  829. unsigned long flags;
  830. write_lock_irqsave(&mapping->tree_lock, flags);
  831. ret = TestClearPageWriteback(page);
  832. if (ret)
  833. radix_tree_tag_clear(&mapping->page_tree,
  834. page_index(page),
  835. PAGECACHE_TAG_WRITEBACK);
  836. write_unlock_irqrestore(&mapping->tree_lock, flags);
  837. } else {
  838. ret = TestClearPageWriteback(page);
  839. }
  840. return ret;
  841. }
  842. int test_set_page_writeback(struct page *page)
  843. {
  844. struct address_space *mapping = page_mapping(page);
  845. int ret;
  846. if (mapping) {
  847. unsigned long flags;
  848. write_lock_irqsave(&mapping->tree_lock, flags);
  849. ret = TestSetPageWriteback(page);
  850. if (!ret)
  851. radix_tree_tag_set(&mapping->page_tree,
  852. page_index(page),
  853. PAGECACHE_TAG_WRITEBACK);
  854. if (!PageDirty(page))
  855. radix_tree_tag_clear(&mapping->page_tree,
  856. page_index(page),
  857. PAGECACHE_TAG_DIRTY);
  858. write_unlock_irqrestore(&mapping->tree_lock, flags);
  859. } else {
  860. ret = TestSetPageWriteback(page);
  861. }
  862. return ret;
  863. }
  864. EXPORT_SYMBOL(test_set_page_writeback);
  865. /*
  866. * Return true if any of the pages in the mapping are marged with the
  867. * passed tag.
  868. */
  869. int mapping_tagged(struct address_space *mapping, int tag)
  870. {
  871. unsigned long flags;
  872. int ret;
  873. read_lock_irqsave(&mapping->tree_lock, flags);
  874. ret = radix_tree_tagged(&mapping->page_tree, tag);
  875. read_unlock_irqrestore(&mapping->tree_lock, flags);
  876. return ret;
  877. }
  878. EXPORT_SYMBOL(mapping_tagged);