write.c 40 KB

12345678910111213141516171819202122232425262728293031323334353637383940414243444546474849505152535455565758596061626364656667686970717273747576777879808182838485868788899091929394959697989910010110210310410510610710810911011111211311411511611711811912012112212312412512612712812913013113213313413513613713813914014114214314414514614714814915015115215315415515615715815916016116216316416516616716816917017117217317417517617717817918018118218318418518618718818919019119219319419519619719819920020120220320420520620720820921021121221321421521621721821922022122222322422522622722822923023123223323423523623723823924024124224324424524624724824925025125225325425525625725825926026126226326426526626726826927027127227327427527627727827928028128228328428528628728828929029129229329429529629729829930030130230330430530630730830931031131231331431531631731831932032132232332432532632732832933033133233333433533633733833934034134234334434534634734834935035135235335435535635735835936036136236336436536636736836937037137237337437537637737837938038138238338438538638738838939039139239339439539639739839940040140240340440540640740840941041141241341441541641741841942042142242342442542642742842943043143243343443543643743843944044144244344444544644744844945045145245345445545645745845946046146246346446546646746846947047147247347447547647747847948048148248348448548648748848949049149249349449549649749849950050150250350450550650750850951051151251351451551651751851952052152252352452552652752852953053153253353453553653753853954054154254354454554654754854955055155255355455555655755855956056156256356456556656756856957057157257357457557657757857958058158258358458558658758858959059159259359459559659759859960060160260360460560660760860961061161261361461561661761861962062162262362462562662762862963063163263363463563663763863964064164264364464564664764864965065165265365465565665765865966066166266366466566666766866967067167267367467567667767867968068168268368468568668768868969069169269369469569669769869970070170270370470570670770870971071171271371471571671771871972072172272372472572672772872973073173273373473573673773873974074174274374474574674774874975075175275375475575675775875976076176276376476576676776876977077177277377477577677777877978078178278378478578678778878979079179279379479579679779879980080180280380480580680780880981081181281381481581681781881982082182282382482582682782882983083183283383483583683783883984084184284384484584684784884985085185285385485585685785885986086186286386486586686786886987087187287387487587687787887988088188288388488588688788888989089189289389489589689789889990090190290390490590690790890991091191291391491591691791891992092192292392492592692792892993093193293393493593693793893994094194294394494594694794894995095195295395495595695795895996096196296396496596696796896997097197297397497597697797897998098198298398498598698798898999099199299399499599699799899910001001100210031004100510061007100810091010101110121013101410151016101710181019102010211022102310241025102610271028102910301031103210331034103510361037103810391040104110421043104410451046104710481049105010511052105310541055105610571058105910601061106210631064106510661067106810691070107110721073107410751076107710781079108010811082108310841085108610871088108910901091109210931094109510961097109810991100110111021103110411051106110711081109111011111112111311141115111611171118111911201121112211231124112511261127112811291130113111321133113411351136113711381139114011411142114311441145114611471148114911501151115211531154115511561157115811591160116111621163116411651166116711681169117011711172117311741175117611771178117911801181118211831184118511861187118811891190119111921193119411951196119711981199120012011202120312041205120612071208120912101211121212131214121512161217121812191220122112221223122412251226122712281229123012311232123312341235123612371238123912401241124212431244124512461247124812491250125112521253125412551256125712581259126012611262126312641265126612671268126912701271127212731274127512761277127812791280128112821283128412851286128712881289129012911292129312941295129612971298129913001301130213031304130513061307130813091310131113121313131413151316131713181319132013211322132313241325132613271328132913301331133213331334133513361337133813391340134113421343134413451346134713481349135013511352135313541355135613571358135913601361136213631364136513661367136813691370137113721373137413751376137713781379138013811382138313841385138613871388138913901391139213931394139513961397139813991400140114021403140414051406140714081409141014111412141314141415141614171418141914201421142214231424142514261427142814291430143114321433143414351436143714381439144014411442144314441445144614471448144914501451145214531454145514561457145814591460146114621463146414651466146714681469147014711472147314741475147614771478147914801481148214831484148514861487148814891490149114921493149414951496149714981499150015011502150315041505150615071508150915101511151215131514151515161517151815191520152115221523152415251526152715281529153015311532153315341535153615371538153915401541154215431544154515461547154815491550155115521553155415551556155715581559156015611562156315641565156615671568156915701571157215731574157515761577157815791580158115821583158415851586158715881589159015911592159315941595159615971598159916001601
  1. /*
  2. * linux/fs/nfs/write.c
  3. *
  4. * Write file data over NFS.
  5. *
  6. * Copyright (C) 1996, 1997, Olaf Kirch <okir@monad.swb.de>
  7. */
  8. #include <linux/types.h>
  9. #include <linux/slab.h>
  10. #include <linux/mm.h>
  11. #include <linux/pagemap.h>
  12. #include <linux/file.h>
  13. #include <linux/writeback.h>
  14. #include <linux/swap.h>
  15. #include <linux/sunrpc/clnt.h>
  16. #include <linux/nfs_fs.h>
  17. #include <linux/nfs_mount.h>
  18. #include <linux/nfs_page.h>
  19. #include <linux/backing-dev.h>
  20. #include <asm/uaccess.h>
  21. #include <linux/smp_lock.h>
  22. #include "delegation.h"
  23. #include "internal.h"
  24. #include "iostat.h"
  25. #define NFSDBG_FACILITY NFSDBG_PAGECACHE
  26. #define MIN_POOL_WRITE (32)
  27. #define MIN_POOL_COMMIT (4)
  28. /*
  29. * Local function declarations
  30. */
  31. static struct nfs_page * nfs_update_request(struct nfs_open_context*,
  32. struct page *,
  33. unsigned int, unsigned int);
  34. static long nfs_flush_mapping(struct address_space *mapping, struct writeback_control *wbc, int how);
  35. static const struct rpc_call_ops nfs_write_partial_ops;
  36. static const struct rpc_call_ops nfs_write_full_ops;
  37. static const struct rpc_call_ops nfs_commit_ops;
  38. static struct kmem_cache *nfs_wdata_cachep;
  39. static mempool_t *nfs_wdata_mempool;
  40. static mempool_t *nfs_commit_mempool;
  41. struct nfs_write_data *nfs_commit_alloc(void)
  42. {
  43. struct nfs_write_data *p = mempool_alloc(nfs_commit_mempool, GFP_NOFS);
  44. if (p) {
  45. memset(p, 0, sizeof(*p));
  46. INIT_LIST_HEAD(&p->pages);
  47. }
  48. return p;
  49. }
  50. void nfs_commit_rcu_free(struct rcu_head *head)
  51. {
  52. struct nfs_write_data *p = container_of(head, struct nfs_write_data, task.u.tk_rcu);
  53. if (p && (p->pagevec != &p->page_array[0]))
  54. kfree(p->pagevec);
  55. mempool_free(p, nfs_commit_mempool);
  56. }
  57. void nfs_commit_free(struct nfs_write_data *wdata)
  58. {
  59. call_rcu_bh(&wdata->task.u.tk_rcu, nfs_commit_rcu_free);
  60. }
  61. struct nfs_write_data *nfs_writedata_alloc(size_t len)
  62. {
  63. unsigned int pagecount = (len + PAGE_SIZE - 1) >> PAGE_SHIFT;
  64. struct nfs_write_data *p = mempool_alloc(nfs_wdata_mempool, GFP_NOFS);
  65. if (p) {
  66. memset(p, 0, sizeof(*p));
  67. INIT_LIST_HEAD(&p->pages);
  68. p->npages = pagecount;
  69. if (pagecount <= ARRAY_SIZE(p->page_array))
  70. p->pagevec = p->page_array;
  71. else {
  72. p->pagevec = kcalloc(pagecount, sizeof(struct page *), GFP_NOFS);
  73. if (!p->pagevec) {
  74. mempool_free(p, nfs_wdata_mempool);
  75. p = NULL;
  76. }
  77. }
  78. }
  79. return p;
  80. }
  81. static void nfs_writedata_rcu_free(struct rcu_head *head)
  82. {
  83. struct nfs_write_data *p = container_of(head, struct nfs_write_data, task.u.tk_rcu);
  84. if (p && (p->pagevec != &p->page_array[0]))
  85. kfree(p->pagevec);
  86. mempool_free(p, nfs_wdata_mempool);
  87. }
  88. static void nfs_writedata_free(struct nfs_write_data *wdata)
  89. {
  90. call_rcu_bh(&wdata->task.u.tk_rcu, nfs_writedata_rcu_free);
  91. }
  92. void nfs_writedata_release(void *wdata)
  93. {
  94. nfs_writedata_free(wdata);
  95. }
  96. static struct nfs_page *nfs_page_find_request_locked(struct page *page)
  97. {
  98. struct nfs_page *req = NULL;
  99. if (PagePrivate(page)) {
  100. req = (struct nfs_page *)page_private(page);
  101. if (req != NULL)
  102. atomic_inc(&req->wb_count);
  103. }
  104. return req;
  105. }
  106. static struct nfs_page *nfs_page_find_request(struct page *page)
  107. {
  108. struct nfs_page *req = NULL;
  109. spinlock_t *req_lock = &NFS_I(page->mapping->host)->req_lock;
  110. spin_lock(req_lock);
  111. req = nfs_page_find_request_locked(page);
  112. spin_unlock(req_lock);
  113. return req;
  114. }
  115. /* Adjust the file length if we're writing beyond the end */
  116. static void nfs_grow_file(struct page *page, unsigned int offset, unsigned int count)
  117. {
  118. struct inode *inode = page->mapping->host;
  119. loff_t end, i_size = i_size_read(inode);
  120. unsigned long end_index = (i_size - 1) >> PAGE_CACHE_SHIFT;
  121. if (i_size > 0 && page->index < end_index)
  122. return;
  123. end = ((loff_t)page->index << PAGE_CACHE_SHIFT) + ((loff_t)offset+count);
  124. if (i_size >= end)
  125. return;
  126. nfs_inc_stats(inode, NFSIOS_EXTENDWRITE);
  127. i_size_write(inode, end);
  128. }
  129. /* A writeback failed: mark the page as bad, and invalidate the page cache */
  130. static void nfs_set_pageerror(struct page *page)
  131. {
  132. SetPageError(page);
  133. nfs_zap_mapping(page->mapping->host, page->mapping);
  134. }
  135. /* We can set the PG_uptodate flag if we see that a write request
  136. * covers the full page.
  137. */
  138. static void nfs_mark_uptodate(struct page *page, unsigned int base, unsigned int count)
  139. {
  140. if (PageUptodate(page))
  141. return;
  142. if (base != 0)
  143. return;
  144. if (count != nfs_page_length(page))
  145. return;
  146. if (count != PAGE_CACHE_SIZE)
  147. memclear_highpage_flush(page, count, PAGE_CACHE_SIZE - count);
  148. SetPageUptodate(page);
  149. }
  150. static int nfs_writepage_setup(struct nfs_open_context *ctx, struct page *page,
  151. unsigned int offset, unsigned int count)
  152. {
  153. struct nfs_page *req;
  154. int ret;
  155. for (;;) {
  156. req = nfs_update_request(ctx, page, offset, count);
  157. if (!IS_ERR(req))
  158. break;
  159. ret = PTR_ERR(req);
  160. if (ret != -EBUSY)
  161. return ret;
  162. ret = nfs_wb_page(page->mapping->host, page);
  163. if (ret != 0)
  164. return ret;
  165. }
  166. /* Update file length */
  167. nfs_grow_file(page, offset, count);
  168. /* Set the PG_uptodate flag? */
  169. nfs_mark_uptodate(page, offset, count);
  170. nfs_unlock_request(req);
  171. return 0;
  172. }
  173. static int wb_priority(struct writeback_control *wbc)
  174. {
  175. if (wbc->for_reclaim)
  176. return FLUSH_HIGHPRI;
  177. if (wbc->for_kupdate)
  178. return FLUSH_LOWPRI;
  179. return 0;
  180. }
  181. /*
  182. * NFS congestion control
  183. */
  184. int nfs_congestion_kb;
  185. #define NFS_CONGESTION_ON_THRESH (nfs_congestion_kb >> (PAGE_SHIFT-10))
  186. #define NFS_CONGESTION_OFF_THRESH \
  187. (NFS_CONGESTION_ON_THRESH - (NFS_CONGESTION_ON_THRESH >> 2))
  188. static int nfs_set_page_writeback(struct page *page)
  189. {
  190. int ret = test_set_page_writeback(page);
  191. if (!ret) {
  192. struct inode *inode = page->mapping->host;
  193. struct nfs_server *nfss = NFS_SERVER(inode);
  194. if (atomic_inc_return(&nfss->writeback) >
  195. NFS_CONGESTION_ON_THRESH)
  196. set_bdi_congested(&nfss->backing_dev_info, WRITE);
  197. }
  198. return ret;
  199. }
  200. static void nfs_end_page_writeback(struct page *page)
  201. {
  202. struct inode *inode = page->mapping->host;
  203. struct nfs_server *nfss = NFS_SERVER(inode);
  204. end_page_writeback(page);
  205. if (atomic_dec_return(&nfss->writeback) < NFS_CONGESTION_OFF_THRESH) {
  206. clear_bdi_congested(&nfss->backing_dev_info, WRITE);
  207. congestion_end(WRITE);
  208. }
  209. }
  210. /*
  211. * Find an associated nfs write request, and prepare to flush it out
  212. * Returns 1 if there was no write request, or if the request was
  213. * already tagged by nfs_set_page_dirty.Returns 0 if the request
  214. * was not tagged.
  215. * May also return an error if the user signalled nfs_wait_on_request().
  216. */
  217. static int nfs_page_mark_flush(struct page *page)
  218. {
  219. struct nfs_page *req;
  220. struct nfs_inode *nfsi = NFS_I(page->mapping->host);
  221. spinlock_t *req_lock = &nfsi->req_lock;
  222. int ret;
  223. spin_lock(req_lock);
  224. for(;;) {
  225. req = nfs_page_find_request_locked(page);
  226. if (req == NULL) {
  227. spin_unlock(req_lock);
  228. return 1;
  229. }
  230. if (nfs_lock_request_dontget(req))
  231. break;
  232. /* Note: If we hold the page lock, as is the case in nfs_writepage,
  233. * then the call to nfs_lock_request_dontget() will always
  234. * succeed provided that someone hasn't already marked the
  235. * request as dirty (in which case we don't care).
  236. */
  237. spin_unlock(req_lock);
  238. ret = nfs_wait_on_request(req);
  239. nfs_release_request(req);
  240. if (ret != 0)
  241. return ret;
  242. spin_lock(req_lock);
  243. }
  244. if (test_bit(PG_NEED_COMMIT, &req->wb_flags)) {
  245. /* This request is marked for commit */
  246. spin_unlock(req_lock);
  247. nfs_unlock_request(req);
  248. return 1;
  249. }
  250. if (nfs_set_page_writeback(page) == 0) {
  251. nfs_list_remove_request(req);
  252. /* add the request to the inode's dirty list. */
  253. radix_tree_tag_set(&nfsi->nfs_page_tree,
  254. req->wb_index, NFS_PAGE_TAG_DIRTY);
  255. nfs_list_add_request(req, &nfsi->dirty);
  256. nfsi->ndirty++;
  257. spin_unlock(req_lock);
  258. __mark_inode_dirty(page->mapping->host, I_DIRTY_PAGES);
  259. } else
  260. spin_unlock(req_lock);
  261. ret = test_bit(PG_NEED_FLUSH, &req->wb_flags);
  262. nfs_unlock_request(req);
  263. return ret;
  264. }
  265. /*
  266. * Write an mmapped page to the server.
  267. */
  268. static int nfs_writepage_locked(struct page *page, struct writeback_control *wbc)
  269. {
  270. struct nfs_open_context *ctx;
  271. struct inode *inode = page->mapping->host;
  272. unsigned offset;
  273. int err;
  274. nfs_inc_stats(inode, NFSIOS_VFSWRITEPAGE);
  275. nfs_add_stats(inode, NFSIOS_WRITEPAGES, 1);
  276. err = nfs_page_mark_flush(page);
  277. if (err <= 0)
  278. goto out;
  279. err = 0;
  280. offset = nfs_page_length(page);
  281. if (!offset)
  282. goto out;
  283. ctx = nfs_find_open_context(inode, NULL, FMODE_WRITE);
  284. if (ctx == NULL) {
  285. err = -EBADF;
  286. goto out;
  287. }
  288. err = nfs_writepage_setup(ctx, page, 0, offset);
  289. put_nfs_open_context(ctx);
  290. if (err != 0)
  291. goto out;
  292. err = nfs_page_mark_flush(page);
  293. if (err > 0)
  294. err = 0;
  295. out:
  296. if (!wbc->for_writepages)
  297. nfs_flush_mapping(page->mapping, wbc, FLUSH_STABLE|wb_priority(wbc));
  298. return err;
  299. }
  300. int nfs_writepage(struct page *page, struct writeback_control *wbc)
  301. {
  302. int err;
  303. err = nfs_writepage_locked(page, wbc);
  304. unlock_page(page);
  305. return err;
  306. }
  307. int nfs_writepages(struct address_space *mapping, struct writeback_control *wbc)
  308. {
  309. struct inode *inode = mapping->host;
  310. int err;
  311. nfs_inc_stats(inode, NFSIOS_VFSWRITEPAGES);
  312. err = generic_writepages(mapping, wbc);
  313. if (err)
  314. return err;
  315. err = nfs_flush_mapping(mapping, wbc, wb_priority(wbc));
  316. if (err < 0)
  317. goto out;
  318. nfs_add_stats(inode, NFSIOS_WRITEPAGES, err);
  319. err = 0;
  320. out:
  321. return err;
  322. }
  323. /*
  324. * Insert a write request into an inode
  325. */
  326. static int nfs_inode_add_request(struct inode *inode, struct nfs_page *req)
  327. {
  328. struct nfs_inode *nfsi = NFS_I(inode);
  329. int error;
  330. error = radix_tree_insert(&nfsi->nfs_page_tree, req->wb_index, req);
  331. BUG_ON(error == -EEXIST);
  332. if (error)
  333. return error;
  334. if (!nfsi->npages) {
  335. igrab(inode);
  336. nfs_begin_data_update(inode);
  337. if (nfs_have_delegation(inode, FMODE_WRITE))
  338. nfsi->change_attr++;
  339. }
  340. SetPagePrivate(req->wb_page);
  341. set_page_private(req->wb_page, (unsigned long)req);
  342. if (PageDirty(req->wb_page))
  343. set_bit(PG_NEED_FLUSH, &req->wb_flags);
  344. nfsi->npages++;
  345. atomic_inc(&req->wb_count);
  346. return 0;
  347. }
  348. /*
  349. * Remove a write request from an inode
  350. */
  351. static void nfs_inode_remove_request(struct nfs_page *req)
  352. {
  353. struct inode *inode = req->wb_context->dentry->d_inode;
  354. struct nfs_inode *nfsi = NFS_I(inode);
  355. BUG_ON (!NFS_WBACK_BUSY(req));
  356. spin_lock(&nfsi->req_lock);
  357. set_page_private(req->wb_page, 0);
  358. ClearPagePrivate(req->wb_page);
  359. radix_tree_delete(&nfsi->nfs_page_tree, req->wb_index);
  360. if (test_and_clear_bit(PG_NEED_FLUSH, &req->wb_flags))
  361. __set_page_dirty_nobuffers(req->wb_page);
  362. nfsi->npages--;
  363. if (!nfsi->npages) {
  364. spin_unlock(&nfsi->req_lock);
  365. nfs_end_data_update(inode);
  366. iput(inode);
  367. } else
  368. spin_unlock(&nfsi->req_lock);
  369. nfs_clear_request(req);
  370. nfs_release_request(req);
  371. }
  372. static void
  373. nfs_redirty_request(struct nfs_page *req)
  374. {
  375. __set_page_dirty_nobuffers(req->wb_page);
  376. }
  377. /*
  378. * Check if a request is dirty
  379. */
  380. static inline int
  381. nfs_dirty_request(struct nfs_page *req)
  382. {
  383. struct page *page = req->wb_page;
  384. if (page == NULL || test_bit(PG_NEED_COMMIT, &req->wb_flags))
  385. return 0;
  386. return !PageWriteback(req->wb_page);
  387. }
  388. #if defined(CONFIG_NFS_V3) || defined(CONFIG_NFS_V4)
  389. /*
  390. * Add a request to the inode's commit list.
  391. */
  392. static void
  393. nfs_mark_request_commit(struct nfs_page *req)
  394. {
  395. struct inode *inode = req->wb_context->dentry->d_inode;
  396. struct nfs_inode *nfsi = NFS_I(inode);
  397. spin_lock(&nfsi->req_lock);
  398. nfs_list_add_request(req, &nfsi->commit);
  399. nfsi->ncommit++;
  400. set_bit(PG_NEED_COMMIT, &(req)->wb_flags);
  401. spin_unlock(&nfsi->req_lock);
  402. inc_zone_page_state(req->wb_page, NR_UNSTABLE_NFS);
  403. __mark_inode_dirty(inode, I_DIRTY_DATASYNC);
  404. }
  405. static inline
  406. int nfs_write_need_commit(struct nfs_write_data *data)
  407. {
  408. return data->verf.committed != NFS_FILE_SYNC;
  409. }
  410. static inline
  411. int nfs_reschedule_unstable_write(struct nfs_page *req)
  412. {
  413. if (test_bit(PG_NEED_COMMIT, &req->wb_flags)) {
  414. nfs_mark_request_commit(req);
  415. return 1;
  416. }
  417. if (test_and_clear_bit(PG_NEED_RESCHED, &req->wb_flags)) {
  418. nfs_redirty_request(req);
  419. return 1;
  420. }
  421. return 0;
  422. }
  423. #else
  424. static inline void
  425. nfs_mark_request_commit(struct nfs_page *req)
  426. {
  427. }
  428. static inline
  429. int nfs_write_need_commit(struct nfs_write_data *data)
  430. {
  431. return 0;
  432. }
  433. static inline
  434. int nfs_reschedule_unstable_write(struct nfs_page *req)
  435. {
  436. return 0;
  437. }
  438. #endif
  439. /*
  440. * Wait for a request to complete.
  441. *
  442. * Interruptible by signals only if mounted with intr flag.
  443. */
  444. static int nfs_wait_on_requests_locked(struct inode *inode, unsigned long idx_start, unsigned int npages)
  445. {
  446. struct nfs_inode *nfsi = NFS_I(inode);
  447. struct nfs_page *req;
  448. unsigned long idx_end, next;
  449. unsigned int res = 0;
  450. int error;
  451. if (npages == 0)
  452. idx_end = ~0;
  453. else
  454. idx_end = idx_start + npages - 1;
  455. next = idx_start;
  456. while (radix_tree_gang_lookup_tag(&nfsi->nfs_page_tree, (void **)&req, next, 1, NFS_PAGE_TAG_WRITEBACK)) {
  457. if (req->wb_index > idx_end)
  458. break;
  459. next = req->wb_index + 1;
  460. BUG_ON(!NFS_WBACK_BUSY(req));
  461. atomic_inc(&req->wb_count);
  462. spin_unlock(&nfsi->req_lock);
  463. error = nfs_wait_on_request(req);
  464. nfs_release_request(req);
  465. spin_lock(&nfsi->req_lock);
  466. if (error < 0)
  467. return error;
  468. res++;
  469. }
  470. return res;
  471. }
  472. static void nfs_cancel_dirty_list(struct list_head *head)
  473. {
  474. struct nfs_page *req;
  475. while(!list_empty(head)) {
  476. req = nfs_list_entry(head->next);
  477. nfs_list_remove_request(req);
  478. nfs_end_page_writeback(req->wb_page);
  479. nfs_inode_remove_request(req);
  480. nfs_clear_page_writeback(req);
  481. }
  482. }
  483. static void nfs_cancel_commit_list(struct list_head *head)
  484. {
  485. struct nfs_page *req;
  486. while(!list_empty(head)) {
  487. req = nfs_list_entry(head->next);
  488. dec_zone_page_state(req->wb_page, NR_UNSTABLE_NFS);
  489. nfs_list_remove_request(req);
  490. clear_bit(PG_NEED_COMMIT, &(req)->wb_flags);
  491. nfs_inode_remove_request(req);
  492. nfs_unlock_request(req);
  493. }
  494. }
  495. #if defined(CONFIG_NFS_V3) || defined(CONFIG_NFS_V4)
  496. /*
  497. * nfs_scan_commit - Scan an inode for commit requests
  498. * @inode: NFS inode to scan
  499. * @dst: destination list
  500. * @idx_start: lower bound of page->index to scan.
  501. * @npages: idx_start + npages sets the upper bound to scan.
  502. *
  503. * Moves requests from the inode's 'commit' request list.
  504. * The requests are *not* checked to ensure that they form a contiguous set.
  505. */
  506. static int
  507. nfs_scan_commit(struct inode *inode, struct list_head *dst, unsigned long idx_start, unsigned int npages)
  508. {
  509. struct nfs_inode *nfsi = NFS_I(inode);
  510. int res = 0;
  511. if (nfsi->ncommit != 0) {
  512. res = nfs_scan_list(nfsi, &nfsi->commit, dst, idx_start, npages);
  513. nfsi->ncommit -= res;
  514. if ((nfsi->ncommit == 0) != list_empty(&nfsi->commit))
  515. printk(KERN_ERR "NFS: desynchronized value of nfs_i.ncommit.\n");
  516. }
  517. return res;
  518. }
  519. #else
  520. static inline int nfs_scan_commit(struct inode *inode, struct list_head *dst, unsigned long idx_start, unsigned int npages)
  521. {
  522. return 0;
  523. }
  524. #endif
  525. static int nfs_wait_on_write_congestion(struct address_space *mapping)
  526. {
  527. struct inode *inode = mapping->host;
  528. struct backing_dev_info *bdi = mapping->backing_dev_info;
  529. int ret = 0;
  530. might_sleep();
  531. if (!bdi_write_congested(bdi))
  532. return 0;
  533. nfs_inc_stats(inode, NFSIOS_CONGESTIONWAIT);
  534. do {
  535. struct rpc_clnt *clnt = NFS_CLIENT(inode);
  536. sigset_t oldset;
  537. rpc_clnt_sigmask(clnt, &oldset);
  538. ret = congestion_wait_interruptible(WRITE, HZ/10);
  539. rpc_clnt_sigunmask(clnt, &oldset);
  540. if (ret == -ERESTARTSYS)
  541. break;
  542. ret = 0;
  543. } while (bdi_write_congested(bdi));
  544. return ret;
  545. }
  546. /*
  547. * Try to update any existing write request, or create one if there is none.
  548. * In order to match, the request's credentials must match those of
  549. * the calling process.
  550. *
  551. * Note: Should always be called with the Page Lock held!
  552. */
  553. static struct nfs_page * nfs_update_request(struct nfs_open_context* ctx,
  554. struct page *page, unsigned int offset, unsigned int bytes)
  555. {
  556. struct address_space *mapping = page->mapping;
  557. struct inode *inode = mapping->host;
  558. struct nfs_inode *nfsi = NFS_I(inode);
  559. struct nfs_page *req, *new = NULL;
  560. unsigned long rqend, end;
  561. end = offset + bytes;
  562. if (nfs_wait_on_write_congestion(mapping))
  563. return ERR_PTR(-ERESTARTSYS);
  564. for (;;) {
  565. /* Loop over all inode entries and see if we find
  566. * A request for the page we wish to update
  567. */
  568. spin_lock(&nfsi->req_lock);
  569. req = nfs_page_find_request_locked(page);
  570. if (req) {
  571. if (!nfs_lock_request_dontget(req)) {
  572. int error;
  573. spin_unlock(&nfsi->req_lock);
  574. error = nfs_wait_on_request(req);
  575. nfs_release_request(req);
  576. if (error < 0) {
  577. if (new)
  578. nfs_release_request(new);
  579. return ERR_PTR(error);
  580. }
  581. continue;
  582. }
  583. spin_unlock(&nfsi->req_lock);
  584. if (new)
  585. nfs_release_request(new);
  586. break;
  587. }
  588. if (new) {
  589. int error;
  590. nfs_lock_request_dontget(new);
  591. error = nfs_inode_add_request(inode, new);
  592. if (error) {
  593. spin_unlock(&nfsi->req_lock);
  594. nfs_unlock_request(new);
  595. return ERR_PTR(error);
  596. }
  597. spin_unlock(&nfsi->req_lock);
  598. return new;
  599. }
  600. spin_unlock(&nfsi->req_lock);
  601. new = nfs_create_request(ctx, inode, page, offset, bytes);
  602. if (IS_ERR(new))
  603. return new;
  604. }
  605. /* We have a request for our page.
  606. * If the creds don't match, or the
  607. * page addresses don't match,
  608. * tell the caller to wait on the conflicting
  609. * request.
  610. */
  611. rqend = req->wb_offset + req->wb_bytes;
  612. if (req->wb_context != ctx
  613. || req->wb_page != page
  614. || !nfs_dirty_request(req)
  615. || offset > rqend || end < req->wb_offset) {
  616. nfs_unlock_request(req);
  617. return ERR_PTR(-EBUSY);
  618. }
  619. /* Okay, the request matches. Update the region */
  620. if (offset < req->wb_offset) {
  621. req->wb_offset = offset;
  622. req->wb_pgbase = offset;
  623. req->wb_bytes = rqend - req->wb_offset;
  624. }
  625. if (end > rqend)
  626. req->wb_bytes = end - req->wb_offset;
  627. return req;
  628. }
  629. int nfs_flush_incompatible(struct file *file, struct page *page)
  630. {
  631. struct nfs_open_context *ctx = (struct nfs_open_context *)file->private_data;
  632. struct nfs_page *req;
  633. int do_flush, status;
  634. /*
  635. * Look for a request corresponding to this page. If there
  636. * is one, and it belongs to another file, we flush it out
  637. * before we try to copy anything into the page. Do this
  638. * due to the lack of an ACCESS-type call in NFSv2.
  639. * Also do the same if we find a request from an existing
  640. * dropped page.
  641. */
  642. do {
  643. req = nfs_page_find_request(page);
  644. if (req == NULL)
  645. return 0;
  646. do_flush = req->wb_page != page || req->wb_context != ctx
  647. || !nfs_dirty_request(req);
  648. nfs_release_request(req);
  649. if (!do_flush)
  650. return 0;
  651. status = nfs_wb_page(page->mapping->host, page);
  652. } while (status == 0);
  653. return status;
  654. }
  655. /*
  656. * Update and possibly write a cached page of an NFS file.
  657. *
  658. * XXX: Keep an eye on generic_file_read to make sure it doesn't do bad
  659. * things with a page scheduled for an RPC call (e.g. invalidate it).
  660. */
  661. int nfs_updatepage(struct file *file, struct page *page,
  662. unsigned int offset, unsigned int count)
  663. {
  664. struct nfs_open_context *ctx = (struct nfs_open_context *)file->private_data;
  665. struct inode *inode = page->mapping->host;
  666. int status = 0;
  667. nfs_inc_stats(inode, NFSIOS_VFSUPDATEPAGE);
  668. dprintk("NFS: nfs_updatepage(%s/%s %d@%Ld)\n",
  669. file->f_path.dentry->d_parent->d_name.name,
  670. file->f_path.dentry->d_name.name, count,
  671. (long long)(page_offset(page) +offset));
  672. /* If we're not using byte range locks, and we know the page
  673. * is entirely in cache, it may be more efficient to avoid
  674. * fragmenting write requests.
  675. */
  676. if (PageUptodate(page) && inode->i_flock == NULL && !(file->f_mode & O_SYNC)) {
  677. count = max(count + offset, nfs_page_length(page));
  678. offset = 0;
  679. }
  680. status = nfs_writepage_setup(ctx, page, offset, count);
  681. __set_page_dirty_nobuffers(page);
  682. dprintk("NFS: nfs_updatepage returns %d (isize %Ld)\n",
  683. status, (long long)i_size_read(inode));
  684. if (status < 0)
  685. nfs_set_pageerror(page);
  686. return status;
  687. }
  688. static void nfs_writepage_release(struct nfs_page *req)
  689. {
  690. if (PageError(req->wb_page) || !nfs_reschedule_unstable_write(req)) {
  691. nfs_end_page_writeback(req->wb_page);
  692. nfs_inode_remove_request(req);
  693. } else
  694. nfs_end_page_writeback(req->wb_page);
  695. nfs_clear_page_writeback(req);
  696. }
  697. static inline int flush_task_priority(int how)
  698. {
  699. switch (how & (FLUSH_HIGHPRI|FLUSH_LOWPRI)) {
  700. case FLUSH_HIGHPRI:
  701. return RPC_PRIORITY_HIGH;
  702. case FLUSH_LOWPRI:
  703. return RPC_PRIORITY_LOW;
  704. }
  705. return RPC_PRIORITY_NORMAL;
  706. }
  707. /*
  708. * Set up the argument/result storage required for the RPC call.
  709. */
  710. static void nfs_write_rpcsetup(struct nfs_page *req,
  711. struct nfs_write_data *data,
  712. const struct rpc_call_ops *call_ops,
  713. unsigned int count, unsigned int offset,
  714. int how)
  715. {
  716. struct inode *inode;
  717. int flags;
  718. /* Set up the RPC argument and reply structs
  719. * NB: take care not to mess about with data->commit et al. */
  720. data->req = req;
  721. data->inode = inode = req->wb_context->dentry->d_inode;
  722. data->cred = req->wb_context->cred;
  723. data->args.fh = NFS_FH(inode);
  724. data->args.offset = req_offset(req) + offset;
  725. data->args.pgbase = req->wb_pgbase + offset;
  726. data->args.pages = data->pagevec;
  727. data->args.count = count;
  728. data->args.context = req->wb_context;
  729. data->res.fattr = &data->fattr;
  730. data->res.count = count;
  731. data->res.verf = &data->verf;
  732. nfs_fattr_init(&data->fattr);
  733. /* Set up the initial task struct. */
  734. flags = (how & FLUSH_SYNC) ? 0 : RPC_TASK_ASYNC;
  735. rpc_init_task(&data->task, NFS_CLIENT(inode), flags, call_ops, data);
  736. NFS_PROTO(inode)->write_setup(data, how);
  737. data->task.tk_priority = flush_task_priority(how);
  738. data->task.tk_cookie = (unsigned long)inode;
  739. dprintk("NFS: %5u initiated write call "
  740. "(req %s/%Ld, %u bytes @ offset %Lu)\n",
  741. data->task.tk_pid,
  742. inode->i_sb->s_id,
  743. (long long)NFS_FILEID(inode),
  744. count,
  745. (unsigned long long)data->args.offset);
  746. }
  747. static void nfs_execute_write(struct nfs_write_data *data)
  748. {
  749. struct rpc_clnt *clnt = NFS_CLIENT(data->inode);
  750. sigset_t oldset;
  751. rpc_clnt_sigmask(clnt, &oldset);
  752. rpc_execute(&data->task);
  753. rpc_clnt_sigunmask(clnt, &oldset);
  754. }
  755. /*
  756. * Generate multiple small requests to write out a single
  757. * contiguous dirty area on one page.
  758. */
  759. static int nfs_flush_multi(struct inode *inode, struct list_head *head, int how)
  760. {
  761. struct nfs_page *req = nfs_list_entry(head->next);
  762. struct page *page = req->wb_page;
  763. struct nfs_write_data *data;
  764. size_t wsize = NFS_SERVER(inode)->wsize, nbytes;
  765. unsigned int offset;
  766. int requests = 0;
  767. LIST_HEAD(list);
  768. nfs_list_remove_request(req);
  769. nbytes = req->wb_bytes;
  770. do {
  771. size_t len = min(nbytes, wsize);
  772. data = nfs_writedata_alloc(len);
  773. if (!data)
  774. goto out_bad;
  775. list_add(&data->pages, &list);
  776. requests++;
  777. nbytes -= len;
  778. } while (nbytes != 0);
  779. atomic_set(&req->wb_complete, requests);
  780. ClearPageError(page);
  781. offset = 0;
  782. nbytes = req->wb_bytes;
  783. do {
  784. data = list_entry(list.next, struct nfs_write_data, pages);
  785. list_del_init(&data->pages);
  786. data->pagevec[0] = page;
  787. if (nbytes > wsize) {
  788. nfs_write_rpcsetup(req, data, &nfs_write_partial_ops,
  789. wsize, offset, how);
  790. offset += wsize;
  791. nbytes -= wsize;
  792. } else {
  793. nfs_write_rpcsetup(req, data, &nfs_write_partial_ops,
  794. nbytes, offset, how);
  795. nbytes = 0;
  796. }
  797. nfs_execute_write(data);
  798. } while (nbytes != 0);
  799. return 0;
  800. out_bad:
  801. while (!list_empty(&list)) {
  802. data = list_entry(list.next, struct nfs_write_data, pages);
  803. list_del(&data->pages);
  804. nfs_writedata_release(data);
  805. }
  806. nfs_redirty_request(req);
  807. nfs_end_page_writeback(req->wb_page);
  808. nfs_clear_page_writeback(req);
  809. return -ENOMEM;
  810. }
  811. /*
  812. * Create an RPC task for the given write request and kick it.
  813. * The page must have been locked by the caller.
  814. *
  815. * It may happen that the page we're passed is not marked dirty.
  816. * This is the case if nfs_updatepage detects a conflicting request
  817. * that has been written but not committed.
  818. */
  819. static int nfs_flush_one(struct inode *inode, struct list_head *head, int how)
  820. {
  821. struct nfs_page *req;
  822. struct page **pages;
  823. struct nfs_write_data *data;
  824. unsigned int count;
  825. data = nfs_writedata_alloc(NFS_SERVER(inode)->wsize);
  826. if (!data)
  827. goto out_bad;
  828. pages = data->pagevec;
  829. count = 0;
  830. while (!list_empty(head)) {
  831. req = nfs_list_entry(head->next);
  832. nfs_list_remove_request(req);
  833. nfs_list_add_request(req, &data->pages);
  834. ClearPageError(req->wb_page);
  835. *pages++ = req->wb_page;
  836. count += req->wb_bytes;
  837. }
  838. req = nfs_list_entry(data->pages.next);
  839. /* Set up the argument struct */
  840. nfs_write_rpcsetup(req, data, &nfs_write_full_ops, count, 0, how);
  841. nfs_execute_write(data);
  842. return 0;
  843. out_bad:
  844. while (!list_empty(head)) {
  845. struct nfs_page *req = nfs_list_entry(head->next);
  846. nfs_list_remove_request(req);
  847. nfs_redirty_request(req);
  848. nfs_end_page_writeback(req->wb_page);
  849. nfs_clear_page_writeback(req);
  850. }
  851. return -ENOMEM;
  852. }
  853. static int nfs_flush_list(struct inode *inode, struct list_head *head, int npages, int how)
  854. {
  855. LIST_HEAD(one_request);
  856. int (*flush_one)(struct inode *, struct list_head *, int);
  857. struct nfs_page *req;
  858. int wpages = NFS_SERVER(inode)->wpages;
  859. int wsize = NFS_SERVER(inode)->wsize;
  860. int error;
  861. flush_one = nfs_flush_one;
  862. if (wsize < PAGE_CACHE_SIZE)
  863. flush_one = nfs_flush_multi;
  864. /* For single writes, FLUSH_STABLE is more efficient */
  865. if (npages <= wpages && npages == NFS_I(inode)->npages
  866. && nfs_list_entry(head->next)->wb_bytes <= wsize)
  867. how |= FLUSH_STABLE;
  868. do {
  869. nfs_coalesce_requests(head, &one_request, wpages);
  870. req = nfs_list_entry(one_request.next);
  871. error = flush_one(inode, &one_request, how);
  872. if (error < 0)
  873. goto out_err;
  874. } while (!list_empty(head));
  875. return 0;
  876. out_err:
  877. while (!list_empty(head)) {
  878. req = nfs_list_entry(head->next);
  879. nfs_list_remove_request(req);
  880. nfs_redirty_request(req);
  881. nfs_end_page_writeback(req->wb_page);
  882. nfs_clear_page_writeback(req);
  883. }
  884. return error;
  885. }
  886. /*
  887. * Handle a write reply that flushed part of a page.
  888. */
  889. static void nfs_writeback_done_partial(struct rpc_task *task, void *calldata)
  890. {
  891. struct nfs_write_data *data = calldata;
  892. struct nfs_page *req = data->req;
  893. struct page *page = req->wb_page;
  894. dprintk("NFS: write (%s/%Ld %d@%Ld)",
  895. req->wb_context->dentry->d_inode->i_sb->s_id,
  896. (long long)NFS_FILEID(req->wb_context->dentry->d_inode),
  897. req->wb_bytes,
  898. (long long)req_offset(req));
  899. if (nfs_writeback_done(task, data) != 0)
  900. return;
  901. if (task->tk_status < 0) {
  902. nfs_set_pageerror(page);
  903. req->wb_context->error = task->tk_status;
  904. dprintk(", error = %d\n", task->tk_status);
  905. goto out;
  906. }
  907. if (nfs_write_need_commit(data)) {
  908. spinlock_t *req_lock = &NFS_I(page->mapping->host)->req_lock;
  909. spin_lock(req_lock);
  910. if (test_bit(PG_NEED_RESCHED, &req->wb_flags)) {
  911. /* Do nothing we need to resend the writes */
  912. } else if (!test_and_set_bit(PG_NEED_COMMIT, &req->wb_flags)) {
  913. memcpy(&req->wb_verf, &data->verf, sizeof(req->wb_verf));
  914. dprintk(" defer commit\n");
  915. } else if (memcmp(&req->wb_verf, &data->verf, sizeof(req->wb_verf))) {
  916. set_bit(PG_NEED_RESCHED, &req->wb_flags);
  917. clear_bit(PG_NEED_COMMIT, &req->wb_flags);
  918. dprintk(" server reboot detected\n");
  919. }
  920. spin_unlock(req_lock);
  921. } else
  922. dprintk(" OK\n");
  923. out:
  924. if (atomic_dec_and_test(&req->wb_complete))
  925. nfs_writepage_release(req);
  926. }
  927. static const struct rpc_call_ops nfs_write_partial_ops = {
  928. .rpc_call_done = nfs_writeback_done_partial,
  929. .rpc_release = nfs_writedata_release,
  930. };
  931. /*
  932. * Handle a write reply that flushes a whole page.
  933. *
  934. * FIXME: There is an inherent race with invalidate_inode_pages and
  935. * writebacks since the page->count is kept > 1 for as long
  936. * as the page has a write request pending.
  937. */
  938. static void nfs_writeback_done_full(struct rpc_task *task, void *calldata)
  939. {
  940. struct nfs_write_data *data = calldata;
  941. struct nfs_page *req;
  942. struct page *page;
  943. if (nfs_writeback_done(task, data) != 0)
  944. return;
  945. /* Update attributes as result of writeback. */
  946. while (!list_empty(&data->pages)) {
  947. req = nfs_list_entry(data->pages.next);
  948. nfs_list_remove_request(req);
  949. page = req->wb_page;
  950. dprintk("NFS: write (%s/%Ld %d@%Ld)",
  951. req->wb_context->dentry->d_inode->i_sb->s_id,
  952. (long long)NFS_FILEID(req->wb_context->dentry->d_inode),
  953. req->wb_bytes,
  954. (long long)req_offset(req));
  955. if (task->tk_status < 0) {
  956. nfs_set_pageerror(page);
  957. req->wb_context->error = task->tk_status;
  958. dprintk(", error = %d\n", task->tk_status);
  959. goto remove_request;
  960. }
  961. if (nfs_write_need_commit(data)) {
  962. memcpy(&req->wb_verf, &data->verf, sizeof(req->wb_verf));
  963. nfs_mark_request_commit(req);
  964. nfs_end_page_writeback(page);
  965. dprintk(" marked for commit\n");
  966. goto next;
  967. }
  968. dprintk(" OK\n");
  969. remove_request:
  970. nfs_end_page_writeback(page);
  971. nfs_inode_remove_request(req);
  972. next:
  973. nfs_clear_page_writeback(req);
  974. }
  975. }
  976. static const struct rpc_call_ops nfs_write_full_ops = {
  977. .rpc_call_done = nfs_writeback_done_full,
  978. .rpc_release = nfs_writedata_release,
  979. };
  980. /*
  981. * This function is called when the WRITE call is complete.
  982. */
  983. int nfs_writeback_done(struct rpc_task *task, struct nfs_write_data *data)
  984. {
  985. struct nfs_writeargs *argp = &data->args;
  986. struct nfs_writeres *resp = &data->res;
  987. int status;
  988. dprintk("NFS: %5u nfs_writeback_done (status %d)\n",
  989. task->tk_pid, task->tk_status);
  990. /*
  991. * ->write_done will attempt to use post-op attributes to detect
  992. * conflicting writes by other clients. A strict interpretation
  993. * of close-to-open would allow us to continue caching even if
  994. * another writer had changed the file, but some applications
  995. * depend on tighter cache coherency when writing.
  996. */
  997. status = NFS_PROTO(data->inode)->write_done(task, data);
  998. if (status != 0)
  999. return status;
  1000. nfs_add_stats(data->inode, NFSIOS_SERVERWRITTENBYTES, resp->count);
  1001. #if defined(CONFIG_NFS_V3) || defined(CONFIG_NFS_V4)
  1002. if (resp->verf->committed < argp->stable && task->tk_status >= 0) {
  1003. /* We tried a write call, but the server did not
  1004. * commit data to stable storage even though we
  1005. * requested it.
  1006. * Note: There is a known bug in Tru64 < 5.0 in which
  1007. * the server reports NFS_DATA_SYNC, but performs
  1008. * NFS_FILE_SYNC. We therefore implement this checking
  1009. * as a dprintk() in order to avoid filling syslog.
  1010. */
  1011. static unsigned long complain;
  1012. if (time_before(complain, jiffies)) {
  1013. dprintk("NFS: faulty NFS server %s:"
  1014. " (committed = %d) != (stable = %d)\n",
  1015. NFS_SERVER(data->inode)->nfs_client->cl_hostname,
  1016. resp->verf->committed, argp->stable);
  1017. complain = jiffies + 300 * HZ;
  1018. }
  1019. }
  1020. #endif
  1021. /* Is this a short write? */
  1022. if (task->tk_status >= 0 && resp->count < argp->count) {
  1023. static unsigned long complain;
  1024. nfs_inc_stats(data->inode, NFSIOS_SHORTWRITE);
  1025. /* Has the server at least made some progress? */
  1026. if (resp->count != 0) {
  1027. /* Was this an NFSv2 write or an NFSv3 stable write? */
  1028. if (resp->verf->committed != NFS_UNSTABLE) {
  1029. /* Resend from where the server left off */
  1030. argp->offset += resp->count;
  1031. argp->pgbase += resp->count;
  1032. argp->count -= resp->count;
  1033. } else {
  1034. /* Resend as a stable write in order to avoid
  1035. * headaches in the case of a server crash.
  1036. */
  1037. argp->stable = NFS_FILE_SYNC;
  1038. }
  1039. rpc_restart_call(task);
  1040. return -EAGAIN;
  1041. }
  1042. if (time_before(complain, jiffies)) {
  1043. printk(KERN_WARNING
  1044. "NFS: Server wrote zero bytes, expected %u.\n",
  1045. argp->count);
  1046. complain = jiffies + 300 * HZ;
  1047. }
  1048. /* Can't do anything about it except throw an error. */
  1049. task->tk_status = -EIO;
  1050. }
  1051. return 0;
  1052. }
  1053. #if defined(CONFIG_NFS_V3) || defined(CONFIG_NFS_V4)
  1054. void nfs_commit_release(void *wdata)
  1055. {
  1056. nfs_commit_free(wdata);
  1057. }
  1058. /*
  1059. * Set up the argument/result storage required for the RPC call.
  1060. */
  1061. static void nfs_commit_rpcsetup(struct list_head *head,
  1062. struct nfs_write_data *data,
  1063. int how)
  1064. {
  1065. struct nfs_page *first;
  1066. struct inode *inode;
  1067. int flags;
  1068. /* Set up the RPC argument and reply structs
  1069. * NB: take care not to mess about with data->commit et al. */
  1070. list_splice_init(head, &data->pages);
  1071. first = nfs_list_entry(data->pages.next);
  1072. inode = first->wb_context->dentry->d_inode;
  1073. data->inode = inode;
  1074. data->cred = first->wb_context->cred;
  1075. data->args.fh = NFS_FH(data->inode);
  1076. /* Note: we always request a commit of the entire inode */
  1077. data->args.offset = 0;
  1078. data->args.count = 0;
  1079. data->res.count = 0;
  1080. data->res.fattr = &data->fattr;
  1081. data->res.verf = &data->verf;
  1082. nfs_fattr_init(&data->fattr);
  1083. /* Set up the initial task struct. */
  1084. flags = (how & FLUSH_SYNC) ? 0 : RPC_TASK_ASYNC;
  1085. rpc_init_task(&data->task, NFS_CLIENT(inode), flags, &nfs_commit_ops, data);
  1086. NFS_PROTO(inode)->commit_setup(data, how);
  1087. data->task.tk_priority = flush_task_priority(how);
  1088. data->task.tk_cookie = (unsigned long)inode;
  1089. dprintk("NFS: %5u initiated commit call\n", data->task.tk_pid);
  1090. }
  1091. /*
  1092. * Commit dirty pages
  1093. */
  1094. static int
  1095. nfs_commit_list(struct inode *inode, struct list_head *head, int how)
  1096. {
  1097. struct nfs_write_data *data;
  1098. struct nfs_page *req;
  1099. data = nfs_commit_alloc();
  1100. if (!data)
  1101. goto out_bad;
  1102. /* Set up the argument struct */
  1103. nfs_commit_rpcsetup(head, data, how);
  1104. nfs_execute_write(data);
  1105. return 0;
  1106. out_bad:
  1107. while (!list_empty(head)) {
  1108. req = nfs_list_entry(head->next);
  1109. nfs_list_remove_request(req);
  1110. nfs_mark_request_commit(req);
  1111. dec_zone_page_state(req->wb_page, NR_UNSTABLE_NFS);
  1112. nfs_clear_page_writeback(req);
  1113. }
  1114. return -ENOMEM;
  1115. }
  1116. /*
  1117. * COMMIT call returned
  1118. */
  1119. static void nfs_commit_done(struct rpc_task *task, void *calldata)
  1120. {
  1121. struct nfs_write_data *data = calldata;
  1122. struct nfs_page *req;
  1123. dprintk("NFS: %5u nfs_commit_done (status %d)\n",
  1124. task->tk_pid, task->tk_status);
  1125. /* Call the NFS version-specific code */
  1126. if (NFS_PROTO(data->inode)->commit_done(task, data) != 0)
  1127. return;
  1128. while (!list_empty(&data->pages)) {
  1129. req = nfs_list_entry(data->pages.next);
  1130. nfs_list_remove_request(req);
  1131. clear_bit(PG_NEED_COMMIT, &(req)->wb_flags);
  1132. dec_zone_page_state(req->wb_page, NR_UNSTABLE_NFS);
  1133. dprintk("NFS: commit (%s/%Ld %d@%Ld)",
  1134. req->wb_context->dentry->d_inode->i_sb->s_id,
  1135. (long long)NFS_FILEID(req->wb_context->dentry->d_inode),
  1136. req->wb_bytes,
  1137. (long long)req_offset(req));
  1138. if (task->tk_status < 0) {
  1139. req->wb_context->error = task->tk_status;
  1140. nfs_inode_remove_request(req);
  1141. dprintk(", error = %d\n", task->tk_status);
  1142. goto next;
  1143. }
  1144. /* Okay, COMMIT succeeded, apparently. Check the verifier
  1145. * returned by the server against all stored verfs. */
  1146. if (!memcmp(req->wb_verf.verifier, data->verf.verifier, sizeof(data->verf.verifier))) {
  1147. /* We have a match */
  1148. nfs_inode_remove_request(req);
  1149. dprintk(" OK\n");
  1150. goto next;
  1151. }
  1152. /* We have a mismatch. Write the page again */
  1153. dprintk(" mismatch\n");
  1154. nfs_redirty_request(req);
  1155. next:
  1156. nfs_clear_page_writeback(req);
  1157. }
  1158. }
  1159. static const struct rpc_call_ops nfs_commit_ops = {
  1160. .rpc_call_done = nfs_commit_done,
  1161. .rpc_release = nfs_commit_release,
  1162. };
  1163. #else
  1164. static inline int nfs_commit_list(struct inode *inode, struct list_head *head, int how)
  1165. {
  1166. return 0;
  1167. }
  1168. #endif
  1169. static long nfs_flush_mapping(struct address_space *mapping, struct writeback_control *wbc, int how)
  1170. {
  1171. struct nfs_inode *nfsi = NFS_I(mapping->host);
  1172. LIST_HEAD(head);
  1173. long res;
  1174. spin_lock(&nfsi->req_lock);
  1175. res = nfs_scan_dirty(mapping, wbc, &head);
  1176. spin_unlock(&nfsi->req_lock);
  1177. if (res) {
  1178. int error = nfs_flush_list(mapping->host, &head, res, how);
  1179. if (error < 0)
  1180. return error;
  1181. }
  1182. return res;
  1183. }
  1184. #if defined(CONFIG_NFS_V3) || defined(CONFIG_NFS_V4)
  1185. int nfs_commit_inode(struct inode *inode, int how)
  1186. {
  1187. struct nfs_inode *nfsi = NFS_I(inode);
  1188. LIST_HEAD(head);
  1189. int res;
  1190. spin_lock(&nfsi->req_lock);
  1191. res = nfs_scan_commit(inode, &head, 0, 0);
  1192. spin_unlock(&nfsi->req_lock);
  1193. if (res) {
  1194. int error = nfs_commit_list(inode, &head, how);
  1195. if (error < 0)
  1196. return error;
  1197. }
  1198. return res;
  1199. }
  1200. #endif
  1201. long nfs_sync_mapping_wait(struct address_space *mapping, struct writeback_control *wbc, int how)
  1202. {
  1203. struct inode *inode = mapping->host;
  1204. struct nfs_inode *nfsi = NFS_I(inode);
  1205. unsigned long idx_start, idx_end;
  1206. unsigned int npages = 0;
  1207. LIST_HEAD(head);
  1208. int nocommit = how & FLUSH_NOCOMMIT;
  1209. long pages, ret;
  1210. /* FIXME */
  1211. if (wbc->range_cyclic)
  1212. idx_start = 0;
  1213. else {
  1214. idx_start = wbc->range_start >> PAGE_CACHE_SHIFT;
  1215. idx_end = wbc->range_end >> PAGE_CACHE_SHIFT;
  1216. if (idx_end > idx_start) {
  1217. unsigned long l_npages = 1 + idx_end - idx_start;
  1218. npages = l_npages;
  1219. if (sizeof(npages) != sizeof(l_npages) &&
  1220. (unsigned long)npages != l_npages)
  1221. npages = 0;
  1222. }
  1223. }
  1224. how &= ~FLUSH_NOCOMMIT;
  1225. spin_lock(&nfsi->req_lock);
  1226. do {
  1227. wbc->pages_skipped = 0;
  1228. ret = nfs_wait_on_requests_locked(inode, idx_start, npages);
  1229. if (ret != 0)
  1230. continue;
  1231. pages = nfs_scan_dirty(mapping, wbc, &head);
  1232. if (pages != 0) {
  1233. spin_unlock(&nfsi->req_lock);
  1234. if (how & FLUSH_INVALIDATE) {
  1235. nfs_cancel_dirty_list(&head);
  1236. ret = pages;
  1237. } else
  1238. ret = nfs_flush_list(inode, &head, pages, how);
  1239. spin_lock(&nfsi->req_lock);
  1240. continue;
  1241. }
  1242. if (wbc->pages_skipped != 0)
  1243. continue;
  1244. if (nocommit)
  1245. break;
  1246. pages = nfs_scan_commit(inode, &head, idx_start, npages);
  1247. if (pages == 0) {
  1248. if (wbc->pages_skipped != 0)
  1249. continue;
  1250. break;
  1251. }
  1252. if (how & FLUSH_INVALIDATE) {
  1253. spin_unlock(&nfsi->req_lock);
  1254. nfs_cancel_commit_list(&head);
  1255. ret = pages;
  1256. spin_lock(&nfsi->req_lock);
  1257. continue;
  1258. }
  1259. pages += nfs_scan_commit(inode, &head, 0, 0);
  1260. spin_unlock(&nfsi->req_lock);
  1261. ret = nfs_commit_list(inode, &head, how);
  1262. spin_lock(&nfsi->req_lock);
  1263. } while (ret >= 0);
  1264. spin_unlock(&nfsi->req_lock);
  1265. return ret;
  1266. }
  1267. /*
  1268. * flush the inode to disk.
  1269. */
  1270. int nfs_wb_all(struct inode *inode)
  1271. {
  1272. struct address_space *mapping = inode->i_mapping;
  1273. struct writeback_control wbc = {
  1274. .bdi = mapping->backing_dev_info,
  1275. .sync_mode = WB_SYNC_ALL,
  1276. .nr_to_write = LONG_MAX,
  1277. .for_writepages = 1,
  1278. .range_cyclic = 1,
  1279. };
  1280. int ret;
  1281. ret = generic_writepages(mapping, &wbc);
  1282. if (ret < 0)
  1283. goto out;
  1284. ret = nfs_sync_mapping_wait(mapping, &wbc, 0);
  1285. if (ret >= 0)
  1286. return 0;
  1287. out:
  1288. __mark_inode_dirty(mapping->host, I_DIRTY_PAGES);
  1289. return ret;
  1290. }
  1291. int nfs_sync_mapping_range(struct address_space *mapping, loff_t range_start, loff_t range_end, int how)
  1292. {
  1293. struct writeback_control wbc = {
  1294. .bdi = mapping->backing_dev_info,
  1295. .sync_mode = WB_SYNC_ALL,
  1296. .nr_to_write = LONG_MAX,
  1297. .range_start = range_start,
  1298. .range_end = range_end,
  1299. .for_writepages = 1,
  1300. };
  1301. int ret;
  1302. if (!(how & FLUSH_NOWRITEPAGE)) {
  1303. ret = generic_writepages(mapping, &wbc);
  1304. if (ret < 0)
  1305. goto out;
  1306. }
  1307. ret = nfs_sync_mapping_wait(mapping, &wbc, how);
  1308. if (ret >= 0)
  1309. return 0;
  1310. out:
  1311. __mark_inode_dirty(mapping->host, I_DIRTY_PAGES);
  1312. return ret;
  1313. }
  1314. int nfs_wb_page_priority(struct inode *inode, struct page *page, int how)
  1315. {
  1316. loff_t range_start = page_offset(page);
  1317. loff_t range_end = range_start + (loff_t)(PAGE_CACHE_SIZE - 1);
  1318. struct writeback_control wbc = {
  1319. .bdi = page->mapping->backing_dev_info,
  1320. .sync_mode = WB_SYNC_ALL,
  1321. .nr_to_write = LONG_MAX,
  1322. .range_start = range_start,
  1323. .range_end = range_end,
  1324. };
  1325. int ret;
  1326. BUG_ON(!PageLocked(page));
  1327. if (!(how & FLUSH_NOWRITEPAGE) && clear_page_dirty_for_io(page)) {
  1328. ret = nfs_writepage_locked(page, &wbc);
  1329. if (ret < 0)
  1330. goto out;
  1331. }
  1332. if (!PagePrivate(page))
  1333. return 0;
  1334. ret = nfs_sync_mapping_wait(page->mapping, &wbc, how);
  1335. if (ret >= 0)
  1336. return 0;
  1337. out:
  1338. __mark_inode_dirty(inode, I_DIRTY_PAGES);
  1339. return ret;
  1340. }
  1341. /*
  1342. * Write back all requests on one page - we do this before reading it.
  1343. */
  1344. int nfs_wb_page(struct inode *inode, struct page* page)
  1345. {
  1346. return nfs_wb_page_priority(inode, page, FLUSH_STABLE);
  1347. }
  1348. int nfs_set_page_dirty(struct page *page)
  1349. {
  1350. spinlock_t *req_lock = &NFS_I(page->mapping->host)->req_lock;
  1351. struct nfs_page *req;
  1352. int ret;
  1353. spin_lock(req_lock);
  1354. req = nfs_page_find_request_locked(page);
  1355. if (req != NULL) {
  1356. /* Mark any existing write requests for flushing */
  1357. ret = !test_and_set_bit(PG_NEED_FLUSH, &req->wb_flags);
  1358. spin_unlock(req_lock);
  1359. nfs_release_request(req);
  1360. return ret;
  1361. }
  1362. ret = __set_page_dirty_nobuffers(page);
  1363. spin_unlock(req_lock);
  1364. return ret;
  1365. }
  1366. int __init nfs_init_writepagecache(void)
  1367. {
  1368. nfs_wdata_cachep = kmem_cache_create("nfs_write_data",
  1369. sizeof(struct nfs_write_data),
  1370. 0, SLAB_HWCACHE_ALIGN,
  1371. NULL, NULL);
  1372. if (nfs_wdata_cachep == NULL)
  1373. return -ENOMEM;
  1374. nfs_wdata_mempool = mempool_create_slab_pool(MIN_POOL_WRITE,
  1375. nfs_wdata_cachep);
  1376. if (nfs_wdata_mempool == NULL)
  1377. return -ENOMEM;
  1378. nfs_commit_mempool = mempool_create_slab_pool(MIN_POOL_COMMIT,
  1379. nfs_wdata_cachep);
  1380. if (nfs_commit_mempool == NULL)
  1381. return -ENOMEM;
  1382. /*
  1383. * NFS congestion size, scale with available memory.
  1384. *
  1385. * 64MB: 8192k
  1386. * 128MB: 11585k
  1387. * 256MB: 16384k
  1388. * 512MB: 23170k
  1389. * 1GB: 32768k
  1390. * 2GB: 46340k
  1391. * 4GB: 65536k
  1392. * 8GB: 92681k
  1393. * 16GB: 131072k
  1394. *
  1395. * This allows larger machines to have larger/more transfers.
  1396. * Limit the default to 256M
  1397. */
  1398. nfs_congestion_kb = (16*int_sqrt(totalram_pages)) << (PAGE_SHIFT-10);
  1399. if (nfs_congestion_kb > 256*1024)
  1400. nfs_congestion_kb = 256*1024;
  1401. return 0;
  1402. }
  1403. void nfs_destroy_writepagecache(void)
  1404. {
  1405. mempool_destroy(nfs_commit_mempool);
  1406. mempool_destroy(nfs_wdata_mempool);
  1407. kmem_cache_destroy(nfs_wdata_cachep);
  1408. }