fs-writeback.c 22 KB

123456789101112131415161718192021222324252627282930313233343536373839404142434445464748495051525354555657585960616263646566676869707172737475767778798081828384858687888990919293949596979899100101102103104105106107108109110111112113114115116117118119120121122123124125126127128129130131132133134135136137138139140141142143144145146147148149150151152153154155156157158159160161162163164165166167168169170171172173174175176177178179180181182183184185186187188189190191192193194195196197198199200201202203204205206207208209210211212213214215216217218219220221222223224225226227228229230231232233234235236237238239240241242243244245246247248249250251252253254255256257258259260261262263264265266267268269270271272273274275276277278279280281282283284285286287288289290291292293294295296297298299300301302303304305306307308309310311312313314315316317318319320321322323324325326327328329330331332333334335336337338339340341342343344345346347348349350351352353354355356357358359360361362363364365366367368369370371372373374375376377378379380381382383384385386387388389390391392393394395396397398399400401402403404405406407408409410411412413414415416417418419420421422423424425426427428429430431432433434435436437438439440441442443444445446447448449450451452453454455456457458459460461462463464465466467468469470471472473474475476477478479480481482483484485486487488489490491492493494495496497498499500501502503504505506507508509510511512513514515516517518519520521522523524525526527528529530531532533534535536537538539540541542543544545546547548549550551552553554555556557558559560561562563564565566567568569570571572573574575576577578579580581582583584585586587588589590591592593594595596597598599600601602603604605606607608609610611612613614615616617618619620621622623624625626627628629630631632633634635636637638639640641642643644645646647648649650651652653654655656657658659660661662663664665666667668669670671672673674675676677678679680681682683684685686687688689690691692693694695696697698699700701702703704705706707708709710711712713714715716717718719720721722723724725726727728729730731732733734735736737738739740741742743744745746747748749750751752753754755756757758759760761762763764765
  1. /*
  2. * fs/fs-writeback.c
  3. *
  4. * Copyright (C) 2002, Linus Torvalds.
  5. *
  6. * Contains all the functions related to writing back and waiting
  7. * upon dirty inodes against superblocks, and writing back dirty
  8. * pages against inodes. ie: data writeback. Writeout of the
  9. * inode itself is not handled here.
  10. *
  11. * 10Apr2002 akpm@zip.com.au
  12. * Split out of fs/inode.c
  13. * Additions for address_space-based writeback
  14. */
  15. #include <linux/kernel.h>
  16. #include <linux/module.h>
  17. #include <linux/spinlock.h>
  18. #include <linux/sched.h>
  19. #include <linux/fs.h>
  20. #include <linux/mm.h>
  21. #include <linux/writeback.h>
  22. #include <linux/blkdev.h>
  23. #include <linux/backing-dev.h>
  24. #include <linux/buffer_head.h>
  25. #include "internal.h"
  26. /**
  27. * __mark_inode_dirty - internal function
  28. * @inode: inode to mark
  29. * @flags: what kind of dirty (i.e. I_DIRTY_SYNC)
  30. * Mark an inode as dirty. Callers should use mark_inode_dirty or
  31. * mark_inode_dirty_sync.
  32. *
  33. * Put the inode on the super block's dirty list.
  34. *
  35. * CAREFUL! We mark it dirty unconditionally, but move it onto the
  36. * dirty list only if it is hashed or if it refers to a blockdev.
  37. * If it was not hashed, it will never be added to the dirty list
  38. * even if it is later hashed, as it will have been marked dirty already.
  39. *
  40. * In short, make sure you hash any inodes _before_ you start marking
  41. * them dirty.
  42. *
  43. * This function *must* be atomic for the I_DIRTY_PAGES case -
  44. * set_page_dirty() is called under spinlock in several places.
  45. *
  46. * Note that for blockdevs, inode->dirtied_when represents the dirtying time of
  47. * the block-special inode (/dev/hda1) itself. And the ->dirtied_when field of
  48. * the kernel-internal blockdev inode represents the dirtying time of the
  49. * blockdev's pages. This is why for I_DIRTY_PAGES we always use
  50. * page->mapping->host, so the page-dirtying time is recorded in the internal
  51. * blockdev inode.
  52. */
  53. void __mark_inode_dirty(struct inode *inode, int flags)
  54. {
  55. struct super_block *sb = inode->i_sb;
  56. /*
  57. * Don't do this for I_DIRTY_PAGES - that doesn't actually
  58. * dirty the inode itself
  59. */
  60. if (flags & (I_DIRTY_SYNC | I_DIRTY_DATASYNC)) {
  61. if (sb->s_op->dirty_inode)
  62. sb->s_op->dirty_inode(inode);
  63. }
  64. /*
  65. * make sure that changes are seen by all cpus before we test i_state
  66. * -- mikulas
  67. */
  68. smp_mb();
  69. /* avoid the locking if we can */
  70. if ((inode->i_state & flags) == flags)
  71. return;
  72. if (unlikely(block_dump)) {
  73. struct dentry *dentry = NULL;
  74. const char *name = "?";
  75. if (!list_empty(&inode->i_dentry)) {
  76. dentry = list_entry(inode->i_dentry.next,
  77. struct dentry, d_alias);
  78. if (dentry && dentry->d_name.name)
  79. name = (const char *) dentry->d_name.name;
  80. }
  81. if (inode->i_ino || strcmp(inode->i_sb->s_id, "bdev"))
  82. printk(KERN_DEBUG
  83. "%s(%d): dirtied inode %lu (%s) on %s\n",
  84. current->comm, current->pid, inode->i_ino,
  85. name, inode->i_sb->s_id);
  86. }
  87. spin_lock(&inode_lock);
  88. if ((inode->i_state & flags) != flags) {
  89. const int was_dirty = inode->i_state & I_DIRTY;
  90. inode->i_state |= flags;
  91. /*
  92. * If the inode is locked, just update its dirty state.
  93. * The unlocker will place the inode on the appropriate
  94. * superblock list, based upon its state.
  95. */
  96. if (inode->i_state & I_LOCK)
  97. goto out;
  98. /*
  99. * Only add valid (hashed) inodes to the superblock's
  100. * dirty list. Add blockdev inodes as well.
  101. */
  102. if (!S_ISBLK(inode->i_mode)) {
  103. if (hlist_unhashed(&inode->i_hash))
  104. goto out;
  105. }
  106. if (inode->i_state & (I_FREEING|I_CLEAR))
  107. goto out;
  108. /*
  109. * If the inode was already on s_dirty or s_io, don't
  110. * reposition it (that would break s_dirty time-ordering).
  111. */
  112. if (!was_dirty) {
  113. inode->dirtied_when = jiffies;
  114. list_move(&inode->i_list, &sb->s_dirty);
  115. }
  116. }
  117. out:
  118. spin_unlock(&inode_lock);
  119. }
  120. EXPORT_SYMBOL(__mark_inode_dirty);
  121. static int write_inode(struct inode *inode, int sync)
  122. {
  123. if (inode->i_sb->s_op->write_inode && !is_bad_inode(inode))
  124. return inode->i_sb->s_op->write_inode(inode, sync);
  125. return 0;
  126. }
  127. /*
  128. * Redirty an inode: set its when-it-was dirtied timestamp and move it to the
  129. * furthest end of its superblock's dirty-inode list.
  130. *
  131. * Before stamping the inode's ->dirtied_when, we check to see whether it is
  132. * already the most-recently-dirtied inode on the s_dirty list. If that is
  133. * the case then the inode must have been redirtied while it was being written
  134. * out and we don't reset its dirtied_when.
  135. */
  136. static void redirty_tail(struct inode *inode)
  137. {
  138. struct super_block *sb = inode->i_sb;
  139. if (!list_empty(&sb->s_dirty)) {
  140. struct inode *tail_inode;
  141. tail_inode = list_entry(sb->s_dirty.next, struct inode, i_list);
  142. if (!time_after_eq(inode->dirtied_when,
  143. tail_inode->dirtied_when))
  144. inode->dirtied_when = jiffies;
  145. }
  146. list_move(&inode->i_list, &sb->s_dirty);
  147. }
  148. /*
  149. * Redirty an inode, but mark it as the very next-to-be-written inode on its
  150. * superblock's dirty-inode list.
  151. * We need to preserve s_dirty's reverse-time-orderedness, so we cheat by
  152. * setting this inode's dirtied_when to the same value as that of the inode
  153. * which is presently head-of-list, if present head-of-list is newer than this
  154. * inode. (head-of-list is the least-recently-dirtied inode: the oldest one).
  155. */
  156. static void redirty_head(struct inode *inode)
  157. {
  158. struct super_block *sb = inode->i_sb;
  159. if (!list_empty(&sb->s_dirty)) {
  160. struct inode *head_inode;
  161. head_inode = list_entry(sb->s_dirty.prev, struct inode, i_list);
  162. if (time_after(inode->dirtied_when, head_inode->dirtied_when))
  163. inode->dirtied_when = head_inode->dirtied_when;
  164. }
  165. list_move_tail(&inode->i_list, &sb->s_dirty);
  166. }
  167. /*
  168. * Write a single inode's dirty pages and inode data out to disk.
  169. * If `wait' is set, wait on the writeout.
  170. *
  171. * The whole writeout design is quite complex and fragile. We want to avoid
  172. * starvation of particular inodes when others are being redirtied, prevent
  173. * livelocks, etc.
  174. *
  175. * Called under inode_lock.
  176. */
  177. static int
  178. __sync_single_inode(struct inode *inode, struct writeback_control *wbc)
  179. {
  180. unsigned dirty;
  181. struct address_space *mapping = inode->i_mapping;
  182. int wait = wbc->sync_mode == WB_SYNC_ALL;
  183. int ret;
  184. BUG_ON(inode->i_state & I_LOCK);
  185. /* Set I_LOCK, reset I_DIRTY */
  186. dirty = inode->i_state & I_DIRTY;
  187. inode->i_state |= I_LOCK;
  188. inode->i_state &= ~I_DIRTY;
  189. spin_unlock(&inode_lock);
  190. ret = do_writepages(mapping, wbc);
  191. /* Don't write the inode if only I_DIRTY_PAGES was set */
  192. if (dirty & (I_DIRTY_SYNC | I_DIRTY_DATASYNC)) {
  193. int err = write_inode(inode, wait);
  194. if (ret == 0)
  195. ret = err;
  196. }
  197. if (wait) {
  198. int err = filemap_fdatawait(mapping);
  199. if (ret == 0)
  200. ret = err;
  201. }
  202. spin_lock(&inode_lock);
  203. inode->i_state &= ~I_LOCK;
  204. if (!(inode->i_state & I_FREEING)) {
  205. if (!(inode->i_state & I_DIRTY) &&
  206. mapping_tagged(mapping, PAGECACHE_TAG_DIRTY)) {
  207. /*
  208. * We didn't write back all the pages. nfs_writepages()
  209. * sometimes bales out without doing anything. Redirty
  210. * the inode. It is moved from s_io onto s_dirty.
  211. */
  212. /*
  213. * akpm: if the caller was the kupdate function we put
  214. * this inode at the head of s_dirty so it gets first
  215. * consideration. Otherwise, move it to the tail, for
  216. * the reasons described there. I'm not really sure
  217. * how much sense this makes. Presumably I had a good
  218. * reasons for doing it this way, and I'd rather not
  219. * muck with it at present.
  220. */
  221. if (wbc->for_kupdate) {
  222. /*
  223. * For the kupdate function we leave the inode
  224. * at the head of sb_dirty so it will get more
  225. * writeout as soon as the queue becomes
  226. * uncongested.
  227. */
  228. inode->i_state |= I_DIRTY_PAGES;
  229. redirty_head(inode);
  230. } else {
  231. /*
  232. * Otherwise fully redirty the inode so that
  233. * other inodes on this superblock will get some
  234. * writeout. Otherwise heavy writing to one
  235. * file would indefinitely suspend writeout of
  236. * all the other files.
  237. */
  238. inode->i_state |= I_DIRTY_PAGES;
  239. redirty_tail(inode);
  240. }
  241. } else if (inode->i_state & I_DIRTY) {
  242. /*
  243. * Someone redirtied the inode while were writing back
  244. * the pages.
  245. */
  246. redirty_tail(inode);
  247. } else if (atomic_read(&inode->i_count)) {
  248. /*
  249. * The inode is clean, inuse
  250. */
  251. list_move(&inode->i_list, &inode_in_use);
  252. } else {
  253. /*
  254. * The inode is clean, unused
  255. */
  256. list_move(&inode->i_list, &inode_unused);
  257. }
  258. }
  259. wake_up_inode(inode);
  260. return ret;
  261. }
  262. /*
  263. * Write out an inode's dirty pages. Called under inode_lock. Either the
  264. * caller has ref on the inode (either via __iget or via syscall against an fd)
  265. * or the inode has I_WILL_FREE set (via generic_forget_inode)
  266. */
  267. static int
  268. __writeback_single_inode(struct inode *inode, struct writeback_control *wbc)
  269. {
  270. wait_queue_head_t *wqh;
  271. if (!atomic_read(&inode->i_count))
  272. WARN_ON(!(inode->i_state & (I_WILL_FREE|I_FREEING)));
  273. else
  274. WARN_ON(inode->i_state & I_WILL_FREE);
  275. if ((wbc->sync_mode != WB_SYNC_ALL) && (inode->i_state & I_LOCK)) {
  276. struct address_space *mapping = inode->i_mapping;
  277. int ret;
  278. /*
  279. * We're skipping this inode because it's locked, and we're not
  280. * doing writeback-for-data-integrity. Move it to the head of
  281. * s_dirty so that writeback can proceed with the other inodes
  282. * on s_io. We'll have another go at writing back this inode
  283. * when the s_dirty iodes get moved back onto s_io.
  284. */
  285. redirty_head(inode);
  286. /*
  287. * Even if we don't actually write the inode itself here,
  288. * we can at least start some of the data writeout..
  289. */
  290. spin_unlock(&inode_lock);
  291. ret = do_writepages(mapping, wbc);
  292. spin_lock(&inode_lock);
  293. return ret;
  294. }
  295. /*
  296. * It's a data-integrity sync. We must wait.
  297. */
  298. if (inode->i_state & I_LOCK) {
  299. DEFINE_WAIT_BIT(wq, &inode->i_state, __I_LOCK);
  300. wqh = bit_waitqueue(&inode->i_state, __I_LOCK);
  301. do {
  302. spin_unlock(&inode_lock);
  303. __wait_on_bit(wqh, &wq, inode_wait,
  304. TASK_UNINTERRUPTIBLE);
  305. spin_lock(&inode_lock);
  306. } while (inode->i_state & I_LOCK);
  307. }
  308. return __sync_single_inode(inode, wbc);
  309. }
  310. /*
  311. * Write out a superblock's list of dirty inodes. A wait will be performed
  312. * upon no inodes, all inodes or the final one, depending upon sync_mode.
  313. *
  314. * If older_than_this is non-NULL, then only write out inodes which
  315. * had their first dirtying at a time earlier than *older_than_this.
  316. *
  317. * If we're a pdlfush thread, then implement pdflush collision avoidance
  318. * against the entire list.
  319. *
  320. * WB_SYNC_HOLD is a hack for sys_sync(): reattach the inode to sb->s_dirty so
  321. * that it can be located for waiting on in __writeback_single_inode().
  322. *
  323. * Called under inode_lock.
  324. *
  325. * If `bdi' is non-zero then we're being asked to writeback a specific queue.
  326. * This function assumes that the blockdev superblock's inodes are backed by
  327. * a variety of queues, so all inodes are searched. For other superblocks,
  328. * assume that all inodes are backed by the same queue.
  329. *
  330. * FIXME: this linear search could get expensive with many fileystems. But
  331. * how to fix? We need to go from an address_space to all inodes which share
  332. * a queue with that address_space. (Easy: have a global "dirty superblocks"
  333. * list).
  334. *
  335. * The inodes to be written are parked on sb->s_io. They are moved back onto
  336. * sb->s_dirty as they are selected for writing. This way, none can be missed
  337. * on the writer throttling path, and we get decent balancing between many
  338. * throttled threads: we don't want them all piling up on __wait_on_inode.
  339. */
  340. static void
  341. sync_sb_inodes(struct super_block *sb, struct writeback_control *wbc)
  342. {
  343. const unsigned long start = jiffies; /* livelock avoidance */
  344. if (!wbc->for_kupdate || list_empty(&sb->s_io))
  345. list_splice_init(&sb->s_dirty, &sb->s_io);
  346. while (!list_empty(&sb->s_io)) {
  347. struct inode *inode = list_entry(sb->s_io.prev,
  348. struct inode, i_list);
  349. struct address_space *mapping = inode->i_mapping;
  350. struct backing_dev_info *bdi = mapping->backing_dev_info;
  351. long pages_skipped;
  352. if (!bdi_cap_writeback_dirty(bdi)) {
  353. redirty_tail(inode);
  354. if (sb_is_blkdev_sb(sb)) {
  355. /*
  356. * Dirty memory-backed blockdev: the ramdisk
  357. * driver does this. Skip just this inode
  358. */
  359. continue;
  360. }
  361. /*
  362. * Dirty memory-backed inode against a filesystem other
  363. * than the kernel-internal bdev filesystem. Skip the
  364. * entire superblock.
  365. */
  366. break;
  367. }
  368. if (wbc->nonblocking && bdi_write_congested(bdi)) {
  369. wbc->encountered_congestion = 1;
  370. if (!sb_is_blkdev_sb(sb))
  371. break; /* Skip a congested fs */
  372. redirty_head(inode);
  373. continue; /* Skip a congested blockdev */
  374. }
  375. if (wbc->bdi && bdi != wbc->bdi) {
  376. if (!sb_is_blkdev_sb(sb))
  377. break; /* fs has the wrong queue */
  378. redirty_head(inode);
  379. continue; /* blockdev has wrong queue */
  380. }
  381. /* Was this inode dirtied after sync_sb_inodes was called? */
  382. if (time_after(inode->dirtied_when, start))
  383. break;
  384. /* Was this inode dirtied too recently? */
  385. if (wbc->older_than_this && time_after(inode->dirtied_when,
  386. *wbc->older_than_this))
  387. break;
  388. /* Is another pdflush already flushing this queue? */
  389. if (current_is_pdflush() && !writeback_acquire(bdi))
  390. break;
  391. BUG_ON(inode->i_state & I_FREEING);
  392. __iget(inode);
  393. pages_skipped = wbc->pages_skipped;
  394. __writeback_single_inode(inode, wbc);
  395. if (wbc->sync_mode == WB_SYNC_HOLD) {
  396. inode->dirtied_when = jiffies;
  397. list_move(&inode->i_list, &sb->s_dirty);
  398. }
  399. if (current_is_pdflush())
  400. writeback_release(bdi);
  401. if (wbc->pages_skipped != pages_skipped) {
  402. /*
  403. * writeback is not making progress due to locked
  404. * buffers. Skip this inode for now.
  405. */
  406. redirty_tail(inode);
  407. }
  408. spin_unlock(&inode_lock);
  409. iput(inode);
  410. cond_resched();
  411. spin_lock(&inode_lock);
  412. if (wbc->nr_to_write <= 0)
  413. break;
  414. }
  415. return; /* Leave any unwritten inodes on s_io */
  416. }
  417. /*
  418. * Start writeback of dirty pagecache data against all unlocked inodes.
  419. *
  420. * Note:
  421. * We don't need to grab a reference to superblock here. If it has non-empty
  422. * ->s_dirty it's hadn't been killed yet and kill_super() won't proceed
  423. * past sync_inodes_sb() until both the ->s_dirty and ->s_io lists are
  424. * empty. Since __sync_single_inode() regains inode_lock before it finally moves
  425. * inode from superblock lists we are OK.
  426. *
  427. * If `older_than_this' is non-zero then only flush inodes which have a
  428. * flushtime older than *older_than_this.
  429. *
  430. * If `bdi' is non-zero then we will scan the first inode against each
  431. * superblock until we find the matching ones. One group will be the dirty
  432. * inodes against a filesystem. Then when we hit the dummy blockdev superblock,
  433. * sync_sb_inodes will seekout the blockdev which matches `bdi'. Maybe not
  434. * super-efficient but we're about to do a ton of I/O...
  435. */
  436. void
  437. writeback_inodes(struct writeback_control *wbc)
  438. {
  439. struct super_block *sb;
  440. might_sleep();
  441. spin_lock(&sb_lock);
  442. restart:
  443. sb = sb_entry(super_blocks.prev);
  444. for (; sb != sb_entry(&super_blocks); sb = sb_entry(sb->s_list.prev)) {
  445. if (!list_empty(&sb->s_dirty) || !list_empty(&sb->s_io)) {
  446. /* we're making our own get_super here */
  447. sb->s_count++;
  448. spin_unlock(&sb_lock);
  449. /*
  450. * If we can't get the readlock, there's no sense in
  451. * waiting around, most of the time the FS is going to
  452. * be unmounted by the time it is released.
  453. */
  454. if (down_read_trylock(&sb->s_umount)) {
  455. if (sb->s_root) {
  456. spin_lock(&inode_lock);
  457. sync_sb_inodes(sb, wbc);
  458. spin_unlock(&inode_lock);
  459. }
  460. up_read(&sb->s_umount);
  461. }
  462. spin_lock(&sb_lock);
  463. if (__put_super_and_need_restart(sb))
  464. goto restart;
  465. }
  466. if (wbc->nr_to_write <= 0)
  467. break;
  468. }
  469. spin_unlock(&sb_lock);
  470. }
  471. /*
  472. * writeback and wait upon the filesystem's dirty inodes. The caller will
  473. * do this in two passes - one to write, and one to wait. WB_SYNC_HOLD is
  474. * used to park the written inodes on sb->s_dirty for the wait pass.
  475. *
  476. * A finite limit is set on the number of pages which will be written.
  477. * To prevent infinite livelock of sys_sync().
  478. *
  479. * We add in the number of potentially dirty inodes, because each inode write
  480. * can dirty pagecache in the underlying blockdev.
  481. */
  482. void sync_inodes_sb(struct super_block *sb, int wait)
  483. {
  484. struct writeback_control wbc = {
  485. .sync_mode = wait ? WB_SYNC_ALL : WB_SYNC_HOLD,
  486. .range_start = 0,
  487. .range_end = LLONG_MAX,
  488. };
  489. unsigned long nr_dirty = global_page_state(NR_FILE_DIRTY);
  490. unsigned long nr_unstable = global_page_state(NR_UNSTABLE_NFS);
  491. wbc.nr_to_write = nr_dirty + nr_unstable +
  492. (inodes_stat.nr_inodes - inodes_stat.nr_unused) +
  493. nr_dirty + nr_unstable;
  494. wbc.nr_to_write += wbc.nr_to_write / 2; /* Bit more for luck */
  495. spin_lock(&inode_lock);
  496. sync_sb_inodes(sb, &wbc);
  497. spin_unlock(&inode_lock);
  498. }
  499. /*
  500. * Rather lame livelock avoidance.
  501. */
  502. static void set_sb_syncing(int val)
  503. {
  504. struct super_block *sb;
  505. spin_lock(&sb_lock);
  506. sb = sb_entry(super_blocks.prev);
  507. for (; sb != sb_entry(&super_blocks); sb = sb_entry(sb->s_list.prev)) {
  508. sb->s_syncing = val;
  509. }
  510. spin_unlock(&sb_lock);
  511. }
  512. /**
  513. * sync_inodes - writes all inodes to disk
  514. * @wait: wait for completion
  515. *
  516. * sync_inodes() goes through each super block's dirty inode list, writes the
  517. * inodes out, waits on the writeout and puts the inodes back on the normal
  518. * list.
  519. *
  520. * This is for sys_sync(). fsync_dev() uses the same algorithm. The subtle
  521. * part of the sync functions is that the blockdev "superblock" is processed
  522. * last. This is because the write_inode() function of a typical fs will
  523. * perform no I/O, but will mark buffers in the blockdev mapping as dirty.
  524. * What we want to do is to perform all that dirtying first, and then write
  525. * back all those inode blocks via the blockdev mapping in one sweep. So the
  526. * additional (somewhat redundant) sync_blockdev() calls here are to make
  527. * sure that really happens. Because if we call sync_inodes_sb(wait=1) with
  528. * outstanding dirty inodes, the writeback goes block-at-a-time within the
  529. * filesystem's write_inode(). This is extremely slow.
  530. */
  531. static void __sync_inodes(int wait)
  532. {
  533. struct super_block *sb;
  534. spin_lock(&sb_lock);
  535. restart:
  536. list_for_each_entry(sb, &super_blocks, s_list) {
  537. if (sb->s_syncing)
  538. continue;
  539. sb->s_syncing = 1;
  540. sb->s_count++;
  541. spin_unlock(&sb_lock);
  542. down_read(&sb->s_umount);
  543. if (sb->s_root) {
  544. sync_inodes_sb(sb, wait);
  545. sync_blockdev(sb->s_bdev);
  546. }
  547. up_read(&sb->s_umount);
  548. spin_lock(&sb_lock);
  549. if (__put_super_and_need_restart(sb))
  550. goto restart;
  551. }
  552. spin_unlock(&sb_lock);
  553. }
  554. void sync_inodes(int wait)
  555. {
  556. set_sb_syncing(0);
  557. __sync_inodes(0);
  558. if (wait) {
  559. set_sb_syncing(0);
  560. __sync_inodes(1);
  561. }
  562. }
  563. /**
  564. * write_inode_now - write an inode to disk
  565. * @inode: inode to write to disk
  566. * @sync: whether the write should be synchronous or not
  567. *
  568. * This function commits an inode to disk immediately if it is dirty. This is
  569. * primarily needed by knfsd.
  570. *
  571. * The caller must either have a ref on the inode or must have set I_WILL_FREE.
  572. */
  573. int write_inode_now(struct inode *inode, int sync)
  574. {
  575. int ret;
  576. struct writeback_control wbc = {
  577. .nr_to_write = LONG_MAX,
  578. .sync_mode = WB_SYNC_ALL,
  579. .range_start = 0,
  580. .range_end = LLONG_MAX,
  581. };
  582. if (!mapping_cap_writeback_dirty(inode->i_mapping))
  583. wbc.nr_to_write = 0;
  584. might_sleep();
  585. spin_lock(&inode_lock);
  586. ret = __writeback_single_inode(inode, &wbc);
  587. spin_unlock(&inode_lock);
  588. if (sync)
  589. wait_on_inode(inode);
  590. return ret;
  591. }
  592. EXPORT_SYMBOL(write_inode_now);
  593. /**
  594. * sync_inode - write an inode and its pages to disk.
  595. * @inode: the inode to sync
  596. * @wbc: controls the writeback mode
  597. *
  598. * sync_inode() will write an inode and its pages to disk. It will also
  599. * correctly update the inode on its superblock's dirty inode lists and will
  600. * update inode->i_state.
  601. *
  602. * The caller must have a ref on the inode.
  603. */
  604. int sync_inode(struct inode *inode, struct writeback_control *wbc)
  605. {
  606. int ret;
  607. spin_lock(&inode_lock);
  608. ret = __writeback_single_inode(inode, wbc);
  609. spin_unlock(&inode_lock);
  610. return ret;
  611. }
  612. EXPORT_SYMBOL(sync_inode);
  613. /**
  614. * generic_osync_inode - flush all dirty data for a given inode to disk
  615. * @inode: inode to write
  616. * @mapping: the address_space that should be flushed
  617. * @what: what to write and wait upon
  618. *
  619. * This can be called by file_write functions for files which have the
  620. * O_SYNC flag set, to flush dirty writes to disk.
  621. *
  622. * @what is a bitmask, specifying which part of the inode's data should be
  623. * written and waited upon.
  624. *
  625. * OSYNC_DATA: i_mapping's dirty data
  626. * OSYNC_METADATA: the buffers at i_mapping->private_list
  627. * OSYNC_INODE: the inode itself
  628. */
  629. int generic_osync_inode(struct inode *inode, struct address_space *mapping, int what)
  630. {
  631. int err = 0;
  632. int need_write_inode_now = 0;
  633. int err2;
  634. if (what & OSYNC_DATA)
  635. err = filemap_fdatawrite(mapping);
  636. if (what & (OSYNC_METADATA|OSYNC_DATA)) {
  637. err2 = sync_mapping_buffers(mapping);
  638. if (!err)
  639. err = err2;
  640. }
  641. if (what & OSYNC_DATA) {
  642. err2 = filemap_fdatawait(mapping);
  643. if (!err)
  644. err = err2;
  645. }
  646. spin_lock(&inode_lock);
  647. if ((inode->i_state & I_DIRTY) &&
  648. ((what & OSYNC_INODE) || (inode->i_state & I_DIRTY_DATASYNC)))
  649. need_write_inode_now = 1;
  650. spin_unlock(&inode_lock);
  651. if (need_write_inode_now) {
  652. err2 = write_inode_now(inode, 1);
  653. if (!err)
  654. err = err2;
  655. }
  656. else
  657. wait_on_inode(inode);
  658. return err;
  659. }
  660. EXPORT_SYMBOL(generic_osync_inode);
  661. /**
  662. * writeback_acquire: attempt to get exclusive writeback access to a device
  663. * @bdi: the device's backing_dev_info structure
  664. *
  665. * It is a waste of resources to have more than one pdflush thread blocked on
  666. * a single request queue. Exclusion at the request_queue level is obtained
  667. * via a flag in the request_queue's backing_dev_info.state.
  668. *
  669. * Non-request_queue-backed address_spaces will share default_backing_dev_info,
  670. * unless they implement their own. Which is somewhat inefficient, as this
  671. * may prevent concurrent writeback against multiple devices.
  672. */
  673. int writeback_acquire(struct backing_dev_info *bdi)
  674. {
  675. return !test_and_set_bit(BDI_pdflush, &bdi->state);
  676. }
  677. /**
  678. * writeback_in_progress: determine whether there is writeback in progress
  679. * @bdi: the device's backing_dev_info structure.
  680. *
  681. * Determine whether there is writeback in progress against a backing device.
  682. */
  683. int writeback_in_progress(struct backing_dev_info *bdi)
  684. {
  685. return test_bit(BDI_pdflush, &bdi->state);
  686. }
  687. /**
  688. * writeback_release: relinquish exclusive writeback access against a device.
  689. * @bdi: the device's backing_dev_info structure
  690. */
  691. void writeback_release(struct backing_dev_info *bdi)
  692. {
  693. BUG_ON(!writeback_in_progress(bdi));
  694. clear_bit(BDI_pdflush, &bdi->state);
  695. }