aops.c 61 KB

123456789101112131415161718192021222324252627282930313233343536373839404142434445464748495051525354555657585960616263646566676869707172737475767778798081828384858687888990919293949596979899100101102103104105106107108109110111112113114115116117118119120121122123124125126127128129130131132133134135136137138139140141142143144145146147148149150151152153154155156157158159160161162163164165166167168169170171172173174175176177178179180181182183184185186187188189190191192193194195196197198199200201202203204205206207208209210211212213214215216217218219220221222223224225226227228229230231232233234235236237238239240241242243244245246247248249250251252253254255256257258259260261262263264265266267268269270271272273274275276277278279280281282283284285286287288289290291292293294295296297298299300301302303304305306307308309310311312313314315316317318319320321322323324325326327328329330331332333334335336337338339340341342343344345346347348349350351352353354355356357358359360361362363364365366367368369370371372373374375376377378379380381382383384385386387388389390391392393394395396397398399400401402403404405406407408409410411412413414415416417418419420421422423424425426427428429430431432433434435436437438439440441442443444445446447448449450451452453454455456457458459460461462463464465466467468469470471472473474475476477478479480481482483484485486487488489490491492493494495496497498499500501502503504505506507508509510511512513514515516517518519520521522523524525526527528529530531532533534535536537538539540541542543544545546547548549550551552553554555556557558559560561562563564565566567568569570571572573574575576577578579580581582583584585586587588589590591592593594595596597598599600601602603604605606607608609610611612613614615616617618619620621622623624625626627628629630631632633634635636637638639640641642643644645646647648649650651652653654655656657658659660661662663664665666667668669670671672673674675676677678679680681682683684685686687688689690691692693694695696697698699700701702703704705706707708709710711712713714715716717718719720721722723724725726727728729730731732733734735736737738739740741742743744745746747748749750751752753754755756757758759760761762763764765766767768769770771772773774775776777778779780781782783784785786787788789790791792793794795796797798799800801802803804805806807808809810811812813814815816817818819820821822823824825826827828829830831832833834835836837838839840841842843844845846847848849850851852853854855856857858859860861862863864865866867868869870871872873874875876877878879880881882883884885886887888889890891892893894895896897898899900901902903904905906907908909910911912913914915916917918919920921922923924925926927928929930931932933934935936937938939940941942943944945946947948949950951952953954955956957958959960961962963964965966967968969970971972973974975976977978979980981982983984985986987988989990991992993994995996997998999100010011002100310041005100610071008100910101011101210131014101510161017101810191020102110221023102410251026102710281029103010311032103310341035103610371038103910401041104210431044104510461047104810491050105110521053105410551056105710581059106010611062106310641065106610671068106910701071107210731074107510761077107810791080108110821083108410851086108710881089109010911092109310941095109610971098109911001101110211031104110511061107110811091110111111121113111411151116111711181119112011211122112311241125112611271128112911301131113211331134113511361137113811391140114111421143114411451146114711481149115011511152115311541155115611571158115911601161116211631164116511661167116811691170117111721173117411751176117711781179118011811182118311841185118611871188118911901191119211931194119511961197119811991200120112021203120412051206120712081209121012111212121312141215121612171218121912201221122212231224122512261227122812291230123112321233123412351236123712381239124012411242124312441245124612471248124912501251125212531254125512561257125812591260126112621263126412651266126712681269127012711272127312741275127612771278127912801281128212831284128512861287128812891290129112921293129412951296129712981299130013011302130313041305130613071308130913101311131213131314131513161317131813191320132113221323132413251326132713281329133013311332133313341335133613371338133913401341134213431344134513461347134813491350135113521353135413551356135713581359136013611362136313641365136613671368136913701371137213731374137513761377137813791380138113821383138413851386138713881389139013911392139313941395139613971398139914001401140214031404140514061407140814091410141114121413141414151416141714181419142014211422142314241425142614271428142914301431143214331434143514361437143814391440144114421443144414451446144714481449145014511452145314541455145614571458145914601461146214631464146514661467146814691470147114721473147414751476147714781479148014811482148314841485148614871488148914901491149214931494149514961497149814991500150115021503150415051506150715081509151015111512151315141515151615171518151915201521152215231524152515261527152815291530153115321533153415351536153715381539154015411542154315441545154615471548154915501551155215531554155515561557155815591560156115621563156415651566156715681569157015711572157315741575157615771578157915801581158215831584158515861587158815891590159115921593159415951596159715981599160016011602160316041605160616071608160916101611161216131614161516161617161816191620162116221623162416251626162716281629163016311632163316341635163616371638163916401641164216431644164516461647164816491650165116521653165416551656165716581659166016611662166316641665166616671668166916701671167216731674167516761677167816791680168116821683168416851686168716881689169016911692169316941695169616971698169917001701170217031704170517061707170817091710171117121713171417151716171717181719172017211722172317241725172617271728172917301731173217331734173517361737173817391740174117421743174417451746174717481749175017511752175317541755175617571758175917601761176217631764176517661767176817691770177117721773177417751776177717781779178017811782178317841785178617871788178917901791179217931794179517961797179817991800180118021803180418051806180718081809181018111812181318141815181618171818181918201821182218231824182518261827182818291830183118321833183418351836183718381839184018411842184318441845184618471848184918501851185218531854185518561857185818591860186118621863186418651866186718681869187018711872187318741875187618771878187918801881188218831884188518861887188818891890189118921893189418951896189718981899190019011902190319041905190619071908190919101911191219131914191519161917191819191920192119221923192419251926192719281929193019311932193319341935193619371938193919401941194219431944194519461947194819491950195119521953195419551956195719581959196019611962196319641965196619671968196919701971197219731974197519761977197819791980198119821983198419851986198719881989199019911992199319941995199619971998199920002001200220032004200520062007200820092010201120122013201420152016201720182019202020212022202320242025202620272028202920302031203220332034203520362037203820392040204120422043204420452046204720482049205020512052205320542055205620572058205920602061206220632064206520662067206820692070207120722073207420752076207720782079208020812082208320842085208620872088208920902091209220932094209520962097209820992100210121022103210421052106210721082109211021112112211321142115211621172118211921202121212221232124212521262127212821292130213121322133213421352136213721382139214021412142214321442145214621472148214921502151215221532154215521562157215821592160216121622163216421652166216721682169217021712172217321742175217621772178217921802181218221832184218521862187218821892190219121922193219421952196219721982199220022012202220322042205220622072208220922102211221222132214221522162217221822192220222122222223222422252226222722282229223022312232223322342235223622372238223922402241224222432244224522462247224822492250225122522253225422552256225722582259226022612262226322642265226622672268226922702271227222732274227522762277227822792280228122822283228422852286228722882289229022912292229322942295229622972298229923002301230223032304230523062307230823092310231123122313231423152316231723182319232023212322232323242325232623272328232923302331233223332334233523362337233823392340234123422343234423452346234723482349235023512352235323542355235623572358235923602361236223632364236523662367236823692370237123722373237423752376237723782379238023812382238323842385238623872388238923902391239223932394239523962397239823992400240124022403240424052406240724082409241024112412241324142415241624172418241924202421242224232424242524262427242824292430243124322433243424352436243724382439244024412442244324442445244624472448
  1. /* -*- mode: c; c-basic-offset: 8; -*-
  2. * vim: noexpandtab sw=8 ts=8 sts=0:
  3. *
  4. * Copyright (C) 2002, 2004 Oracle. All rights reserved.
  5. *
  6. * This program is free software; you can redistribute it and/or
  7. * modify it under the terms of the GNU General Public
  8. * License as published by the Free Software Foundation; either
  9. * version 2 of the License, or (at your option) any later version.
  10. *
  11. * This program is distributed in the hope that it will be useful,
  12. * but WITHOUT ANY WARRANTY; without even the implied warranty of
  13. * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
  14. * General Public License for more details.
  15. *
  16. * You should have received a copy of the GNU General Public
  17. * License along with this program; if not, write to the
  18. * Free Software Foundation, Inc., 59 Temple Place - Suite 330,
  19. * Boston, MA 021110-1307, USA.
  20. */
  21. #include <linux/fs.h>
  22. #include <linux/slab.h>
  23. #include <linux/highmem.h>
  24. #include <linux/pagemap.h>
  25. #include <asm/byteorder.h>
  26. #include <linux/swap.h>
  27. #include <linux/pipe_fs_i.h>
  28. #include <linux/mpage.h>
  29. #include <linux/quotaops.h>
  30. #include <linux/blkdev.h>
  31. #include <linux/uio.h>
  32. #include <cluster/masklog.h>
  33. #include "ocfs2.h"
  34. #include "alloc.h"
  35. #include "aops.h"
  36. #include "dlmglue.h"
  37. #include "extent_map.h"
  38. #include "file.h"
  39. #include "inode.h"
  40. #include "journal.h"
  41. #include "suballoc.h"
  42. #include "super.h"
  43. #include "symlink.h"
  44. #include "refcounttree.h"
  45. #include "ocfs2_trace.h"
  46. #include "buffer_head_io.h"
  47. #include "dir.h"
  48. #include "namei.h"
  49. #include "sysfile.h"
  50. static int ocfs2_symlink_get_block(struct inode *inode, sector_t iblock,
  51. struct buffer_head *bh_result, int create)
  52. {
  53. int err = -EIO;
  54. int status;
  55. struct ocfs2_dinode *fe = NULL;
  56. struct buffer_head *bh = NULL;
  57. struct buffer_head *buffer_cache_bh = NULL;
  58. struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
  59. void *kaddr;
  60. trace_ocfs2_symlink_get_block(
  61. (unsigned long long)OCFS2_I(inode)->ip_blkno,
  62. (unsigned long long)iblock, bh_result, create);
  63. BUG_ON(ocfs2_inode_is_fast_symlink(inode));
  64. if ((iblock << inode->i_sb->s_blocksize_bits) > PATH_MAX + 1) {
  65. mlog(ML_ERROR, "block offset > PATH_MAX: %llu",
  66. (unsigned long long)iblock);
  67. goto bail;
  68. }
  69. status = ocfs2_read_inode_block(inode, &bh);
  70. if (status < 0) {
  71. mlog_errno(status);
  72. goto bail;
  73. }
  74. fe = (struct ocfs2_dinode *) bh->b_data;
  75. if ((u64)iblock >= ocfs2_clusters_to_blocks(inode->i_sb,
  76. le32_to_cpu(fe->i_clusters))) {
  77. err = -ENOMEM;
  78. mlog(ML_ERROR, "block offset is outside the allocated size: "
  79. "%llu\n", (unsigned long long)iblock);
  80. goto bail;
  81. }
  82. /* We don't use the page cache to create symlink data, so if
  83. * need be, copy it over from the buffer cache. */
  84. if (!buffer_uptodate(bh_result) && ocfs2_inode_is_new(inode)) {
  85. u64 blkno = le64_to_cpu(fe->id2.i_list.l_recs[0].e_blkno) +
  86. iblock;
  87. buffer_cache_bh = sb_getblk(osb->sb, blkno);
  88. if (!buffer_cache_bh) {
  89. err = -ENOMEM;
  90. mlog(ML_ERROR, "couldn't getblock for symlink!\n");
  91. goto bail;
  92. }
  93. /* we haven't locked out transactions, so a commit
  94. * could've happened. Since we've got a reference on
  95. * the bh, even if it commits while we're doing the
  96. * copy, the data is still good. */
  97. if (buffer_jbd(buffer_cache_bh)
  98. && ocfs2_inode_is_new(inode)) {
  99. kaddr = kmap_atomic(bh_result->b_page);
  100. if (!kaddr) {
  101. mlog(ML_ERROR, "couldn't kmap!\n");
  102. goto bail;
  103. }
  104. memcpy(kaddr + (bh_result->b_size * iblock),
  105. buffer_cache_bh->b_data,
  106. bh_result->b_size);
  107. kunmap_atomic(kaddr);
  108. set_buffer_uptodate(bh_result);
  109. }
  110. brelse(buffer_cache_bh);
  111. }
  112. map_bh(bh_result, inode->i_sb,
  113. le64_to_cpu(fe->id2.i_list.l_recs[0].e_blkno) + iblock);
  114. err = 0;
  115. bail:
  116. brelse(bh);
  117. return err;
  118. }
  119. int ocfs2_get_block(struct inode *inode, sector_t iblock,
  120. struct buffer_head *bh_result, int create)
  121. {
  122. int err = 0;
  123. unsigned int ext_flags;
  124. u64 max_blocks = bh_result->b_size >> inode->i_blkbits;
  125. u64 p_blkno, count, past_eof;
  126. struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
  127. trace_ocfs2_get_block((unsigned long long)OCFS2_I(inode)->ip_blkno,
  128. (unsigned long long)iblock, bh_result, create);
  129. if (OCFS2_I(inode)->ip_flags & OCFS2_INODE_SYSTEM_FILE)
  130. mlog(ML_NOTICE, "get_block on system inode 0x%p (%lu)\n",
  131. inode, inode->i_ino);
  132. if (S_ISLNK(inode->i_mode)) {
  133. /* this always does I/O for some reason. */
  134. err = ocfs2_symlink_get_block(inode, iblock, bh_result, create);
  135. goto bail;
  136. }
  137. err = ocfs2_extent_map_get_blocks(inode, iblock, &p_blkno, &count,
  138. &ext_flags);
  139. if (err) {
  140. mlog(ML_ERROR, "Error %d from get_blocks(0x%p, %llu, 1, "
  141. "%llu, NULL)\n", err, inode, (unsigned long long)iblock,
  142. (unsigned long long)p_blkno);
  143. goto bail;
  144. }
  145. if (max_blocks < count)
  146. count = max_blocks;
  147. /*
  148. * ocfs2 never allocates in this function - the only time we
  149. * need to use BH_New is when we're extending i_size on a file
  150. * system which doesn't support holes, in which case BH_New
  151. * allows __block_write_begin() to zero.
  152. *
  153. * If we see this on a sparse file system, then a truncate has
  154. * raced us and removed the cluster. In this case, we clear
  155. * the buffers dirty and uptodate bits and let the buffer code
  156. * ignore it as a hole.
  157. */
  158. if (create && p_blkno == 0 && ocfs2_sparse_alloc(osb)) {
  159. clear_buffer_dirty(bh_result);
  160. clear_buffer_uptodate(bh_result);
  161. goto bail;
  162. }
  163. /* Treat the unwritten extent as a hole for zeroing purposes. */
  164. if (p_blkno && !(ext_flags & OCFS2_EXT_UNWRITTEN))
  165. map_bh(bh_result, inode->i_sb, p_blkno);
  166. bh_result->b_size = count << inode->i_blkbits;
  167. if (!ocfs2_sparse_alloc(osb)) {
  168. if (p_blkno == 0) {
  169. err = -EIO;
  170. mlog(ML_ERROR,
  171. "iblock = %llu p_blkno = %llu blkno=(%llu)\n",
  172. (unsigned long long)iblock,
  173. (unsigned long long)p_blkno,
  174. (unsigned long long)OCFS2_I(inode)->ip_blkno);
  175. mlog(ML_ERROR, "Size %llu, clusters %u\n", (unsigned long long)i_size_read(inode), OCFS2_I(inode)->ip_clusters);
  176. dump_stack();
  177. goto bail;
  178. }
  179. }
  180. past_eof = ocfs2_blocks_for_bytes(inode->i_sb, i_size_read(inode));
  181. trace_ocfs2_get_block_end((unsigned long long)OCFS2_I(inode)->ip_blkno,
  182. (unsigned long long)past_eof);
  183. if (create && (iblock >= past_eof))
  184. set_buffer_new(bh_result);
  185. bail:
  186. if (err < 0)
  187. err = -EIO;
  188. return err;
  189. }
  190. int ocfs2_read_inline_data(struct inode *inode, struct page *page,
  191. struct buffer_head *di_bh)
  192. {
  193. void *kaddr;
  194. loff_t size;
  195. struct ocfs2_dinode *di = (struct ocfs2_dinode *)di_bh->b_data;
  196. if (!(le16_to_cpu(di->i_dyn_features) & OCFS2_INLINE_DATA_FL)) {
  197. ocfs2_error(inode->i_sb, "Inode %llu lost inline data flag",
  198. (unsigned long long)OCFS2_I(inode)->ip_blkno);
  199. return -EROFS;
  200. }
  201. size = i_size_read(inode);
  202. if (size > PAGE_CACHE_SIZE ||
  203. size > ocfs2_max_inline_data_with_xattr(inode->i_sb, di)) {
  204. ocfs2_error(inode->i_sb,
  205. "Inode %llu has with inline data has bad size: %Lu",
  206. (unsigned long long)OCFS2_I(inode)->ip_blkno,
  207. (unsigned long long)size);
  208. return -EROFS;
  209. }
  210. kaddr = kmap_atomic(page);
  211. if (size)
  212. memcpy(kaddr, di->id2.i_data.id_data, size);
  213. /* Clear the remaining part of the page */
  214. memset(kaddr + size, 0, PAGE_CACHE_SIZE - size);
  215. flush_dcache_page(page);
  216. kunmap_atomic(kaddr);
  217. SetPageUptodate(page);
  218. return 0;
  219. }
  220. static int ocfs2_readpage_inline(struct inode *inode, struct page *page)
  221. {
  222. int ret;
  223. struct buffer_head *di_bh = NULL;
  224. BUG_ON(!PageLocked(page));
  225. BUG_ON(!(OCFS2_I(inode)->ip_dyn_features & OCFS2_INLINE_DATA_FL));
  226. ret = ocfs2_read_inode_block(inode, &di_bh);
  227. if (ret) {
  228. mlog_errno(ret);
  229. goto out;
  230. }
  231. ret = ocfs2_read_inline_data(inode, page, di_bh);
  232. out:
  233. unlock_page(page);
  234. brelse(di_bh);
  235. return ret;
  236. }
  237. static int ocfs2_readpage(struct file *file, struct page *page)
  238. {
  239. struct inode *inode = page->mapping->host;
  240. struct ocfs2_inode_info *oi = OCFS2_I(inode);
  241. loff_t start = (loff_t)page->index << PAGE_CACHE_SHIFT;
  242. int ret, unlock = 1;
  243. trace_ocfs2_readpage((unsigned long long)oi->ip_blkno,
  244. (page ? page->index : 0));
  245. ret = ocfs2_inode_lock_with_page(inode, NULL, 0, page);
  246. if (ret != 0) {
  247. if (ret == AOP_TRUNCATED_PAGE)
  248. unlock = 0;
  249. mlog_errno(ret);
  250. goto out;
  251. }
  252. if (down_read_trylock(&oi->ip_alloc_sem) == 0) {
  253. /*
  254. * Unlock the page and cycle ip_alloc_sem so that we don't
  255. * busyloop waiting for ip_alloc_sem to unlock
  256. */
  257. ret = AOP_TRUNCATED_PAGE;
  258. unlock_page(page);
  259. unlock = 0;
  260. down_read(&oi->ip_alloc_sem);
  261. up_read(&oi->ip_alloc_sem);
  262. goto out_inode_unlock;
  263. }
  264. /*
  265. * i_size might have just been updated as we grabed the meta lock. We
  266. * might now be discovering a truncate that hit on another node.
  267. * block_read_full_page->get_block freaks out if it is asked to read
  268. * beyond the end of a file, so we check here. Callers
  269. * (generic_file_read, vm_ops->fault) are clever enough to check i_size
  270. * and notice that the page they just read isn't needed.
  271. *
  272. * XXX sys_readahead() seems to get that wrong?
  273. */
  274. if (start >= i_size_read(inode)) {
  275. zero_user(page, 0, PAGE_SIZE);
  276. SetPageUptodate(page);
  277. ret = 0;
  278. goto out_alloc;
  279. }
  280. if (oi->ip_dyn_features & OCFS2_INLINE_DATA_FL)
  281. ret = ocfs2_readpage_inline(inode, page);
  282. else
  283. ret = block_read_full_page(page, ocfs2_get_block);
  284. unlock = 0;
  285. out_alloc:
  286. up_read(&OCFS2_I(inode)->ip_alloc_sem);
  287. out_inode_unlock:
  288. ocfs2_inode_unlock(inode, 0);
  289. out:
  290. if (unlock)
  291. unlock_page(page);
  292. return ret;
  293. }
  294. /*
  295. * This is used only for read-ahead. Failures or difficult to handle
  296. * situations are safe to ignore.
  297. *
  298. * Right now, we don't bother with BH_Boundary - in-inode extent lists
  299. * are quite large (243 extents on 4k blocks), so most inodes don't
  300. * grow out to a tree. If need be, detecting boundary extents could
  301. * trivially be added in a future version of ocfs2_get_block().
  302. */
  303. static int ocfs2_readpages(struct file *filp, struct address_space *mapping,
  304. struct list_head *pages, unsigned nr_pages)
  305. {
  306. int ret, err = -EIO;
  307. struct inode *inode = mapping->host;
  308. struct ocfs2_inode_info *oi = OCFS2_I(inode);
  309. loff_t start;
  310. struct page *last;
  311. /*
  312. * Use the nonblocking flag for the dlm code to avoid page
  313. * lock inversion, but don't bother with retrying.
  314. */
  315. ret = ocfs2_inode_lock_full(inode, NULL, 0, OCFS2_LOCK_NONBLOCK);
  316. if (ret)
  317. return err;
  318. if (down_read_trylock(&oi->ip_alloc_sem) == 0) {
  319. ocfs2_inode_unlock(inode, 0);
  320. return err;
  321. }
  322. /*
  323. * Don't bother with inline-data. There isn't anything
  324. * to read-ahead in that case anyway...
  325. */
  326. if (oi->ip_dyn_features & OCFS2_INLINE_DATA_FL)
  327. goto out_unlock;
  328. /*
  329. * Check whether a remote node truncated this file - we just
  330. * drop out in that case as it's not worth handling here.
  331. */
  332. last = list_entry(pages->prev, struct page, lru);
  333. start = (loff_t)last->index << PAGE_CACHE_SHIFT;
  334. if (start >= i_size_read(inode))
  335. goto out_unlock;
  336. err = mpage_readpages(mapping, pages, nr_pages, ocfs2_get_block);
  337. out_unlock:
  338. up_read(&oi->ip_alloc_sem);
  339. ocfs2_inode_unlock(inode, 0);
  340. return err;
  341. }
  342. /* Note: Because we don't support holes, our allocation has
  343. * already happened (allocation writes zeros to the file data)
  344. * so we don't have to worry about ordered writes in
  345. * ocfs2_writepage.
  346. *
  347. * ->writepage is called during the process of invalidating the page cache
  348. * during blocked lock processing. It can't block on any cluster locks
  349. * to during block mapping. It's relying on the fact that the block
  350. * mapping can't have disappeared under the dirty pages that it is
  351. * being asked to write back.
  352. */
  353. static int ocfs2_writepage(struct page *page, struct writeback_control *wbc)
  354. {
  355. trace_ocfs2_writepage(
  356. (unsigned long long)OCFS2_I(page->mapping->host)->ip_blkno,
  357. page->index);
  358. return block_write_full_page(page, ocfs2_get_block, wbc);
  359. }
  360. /* Taken from ext3. We don't necessarily need the full blown
  361. * functionality yet, but IMHO it's better to cut and paste the whole
  362. * thing so we can avoid introducing our own bugs (and easily pick up
  363. * their fixes when they happen) --Mark */
  364. int walk_page_buffers( handle_t *handle,
  365. struct buffer_head *head,
  366. unsigned from,
  367. unsigned to,
  368. int *partial,
  369. int (*fn)( handle_t *handle,
  370. struct buffer_head *bh))
  371. {
  372. struct buffer_head *bh;
  373. unsigned block_start, block_end;
  374. unsigned blocksize = head->b_size;
  375. int err, ret = 0;
  376. struct buffer_head *next;
  377. for ( bh = head, block_start = 0;
  378. ret == 0 && (bh != head || !block_start);
  379. block_start = block_end, bh = next)
  380. {
  381. next = bh->b_this_page;
  382. block_end = block_start + blocksize;
  383. if (block_end <= from || block_start >= to) {
  384. if (partial && !buffer_uptodate(bh))
  385. *partial = 1;
  386. continue;
  387. }
  388. err = (*fn)(handle, bh);
  389. if (!ret)
  390. ret = err;
  391. }
  392. return ret;
  393. }
  394. static sector_t ocfs2_bmap(struct address_space *mapping, sector_t block)
  395. {
  396. sector_t status;
  397. u64 p_blkno = 0;
  398. int err = 0;
  399. struct inode *inode = mapping->host;
  400. trace_ocfs2_bmap((unsigned long long)OCFS2_I(inode)->ip_blkno,
  401. (unsigned long long)block);
  402. /* We don't need to lock journal system files, since they aren't
  403. * accessed concurrently from multiple nodes.
  404. */
  405. if (!INODE_JOURNAL(inode)) {
  406. err = ocfs2_inode_lock(inode, NULL, 0);
  407. if (err) {
  408. if (err != -ENOENT)
  409. mlog_errno(err);
  410. goto bail;
  411. }
  412. down_read(&OCFS2_I(inode)->ip_alloc_sem);
  413. }
  414. if (!(OCFS2_I(inode)->ip_dyn_features & OCFS2_INLINE_DATA_FL))
  415. err = ocfs2_extent_map_get_blocks(inode, block, &p_blkno, NULL,
  416. NULL);
  417. if (!INODE_JOURNAL(inode)) {
  418. up_read(&OCFS2_I(inode)->ip_alloc_sem);
  419. ocfs2_inode_unlock(inode, 0);
  420. }
  421. if (err) {
  422. mlog(ML_ERROR, "get_blocks() failed, block = %llu\n",
  423. (unsigned long long)block);
  424. mlog_errno(err);
  425. goto bail;
  426. }
  427. bail:
  428. status = err ? 0 : p_blkno;
  429. return status;
  430. }
  431. /*
  432. * TODO: Make this into a generic get_blocks function.
  433. *
  434. * From do_direct_io in direct-io.c:
  435. * "So what we do is to permit the ->get_blocks function to populate
  436. * bh.b_size with the size of IO which is permitted at this offset and
  437. * this i_blkbits."
  438. *
  439. * This function is called directly from get_more_blocks in direct-io.c.
  440. *
  441. * called like this: dio->get_blocks(dio->inode, fs_startblk,
  442. * fs_count, map_bh, dio->rw == WRITE);
  443. */
  444. static int ocfs2_direct_IO_get_blocks(struct inode *inode, sector_t iblock,
  445. struct buffer_head *bh_result, int create)
  446. {
  447. int ret;
  448. u32 cpos = 0;
  449. int alloc_locked = 0;
  450. u64 p_blkno, inode_blocks, contig_blocks;
  451. unsigned int ext_flags;
  452. unsigned char blocksize_bits = inode->i_sb->s_blocksize_bits;
  453. unsigned long max_blocks = bh_result->b_size >> inode->i_blkbits;
  454. unsigned long len = bh_result->b_size;
  455. unsigned int clusters_to_alloc = 0;
  456. cpos = ocfs2_blocks_to_clusters(inode->i_sb, iblock);
  457. /* This function won't even be called if the request isn't all
  458. * nicely aligned and of the right size, so there's no need
  459. * for us to check any of that. */
  460. inode_blocks = ocfs2_blocks_for_bytes(inode->i_sb, i_size_read(inode));
  461. /* This figures out the size of the next contiguous block, and
  462. * our logical offset */
  463. ret = ocfs2_extent_map_get_blocks(inode, iblock, &p_blkno,
  464. &contig_blocks, &ext_flags);
  465. if (ret) {
  466. mlog(ML_ERROR, "get_blocks() failed iblock=%llu\n",
  467. (unsigned long long)iblock);
  468. ret = -EIO;
  469. goto bail;
  470. }
  471. /* We should already CoW the refcounted extent in case of create. */
  472. BUG_ON(create && (ext_flags & OCFS2_EXT_REFCOUNTED));
  473. /* allocate blocks if no p_blkno is found, and create == 1 */
  474. if (!p_blkno && create) {
  475. ret = ocfs2_inode_lock(inode, NULL, 1);
  476. if (ret < 0) {
  477. mlog_errno(ret);
  478. goto bail;
  479. }
  480. alloc_locked = 1;
  481. /* fill hole, allocate blocks can't be larger than the size
  482. * of the hole */
  483. clusters_to_alloc = ocfs2_clusters_for_bytes(inode->i_sb, len);
  484. if (clusters_to_alloc > contig_blocks)
  485. clusters_to_alloc = contig_blocks;
  486. /* allocate extent and insert them into the extent tree */
  487. ret = ocfs2_extend_allocation(inode, cpos,
  488. clusters_to_alloc, 0);
  489. if (ret < 0) {
  490. mlog_errno(ret);
  491. goto bail;
  492. }
  493. ret = ocfs2_extent_map_get_blocks(inode, iblock, &p_blkno,
  494. &contig_blocks, &ext_flags);
  495. if (ret < 0) {
  496. mlog(ML_ERROR, "get_blocks() failed iblock=%llu\n",
  497. (unsigned long long)iblock);
  498. ret = -EIO;
  499. goto bail;
  500. }
  501. }
  502. /*
  503. * get_more_blocks() expects us to describe a hole by clearing
  504. * the mapped bit on bh_result().
  505. *
  506. * Consider an unwritten extent as a hole.
  507. */
  508. if (p_blkno && !(ext_flags & OCFS2_EXT_UNWRITTEN))
  509. map_bh(bh_result, inode->i_sb, p_blkno);
  510. else
  511. clear_buffer_mapped(bh_result);
  512. /* make sure we don't map more than max_blocks blocks here as
  513. that's all the kernel will handle at this point. */
  514. if (max_blocks < contig_blocks)
  515. contig_blocks = max_blocks;
  516. bh_result->b_size = contig_blocks << blocksize_bits;
  517. bail:
  518. if (alloc_locked)
  519. ocfs2_inode_unlock(inode, 1);
  520. return ret;
  521. }
  522. /*
  523. * ocfs2_dio_end_io is called by the dio core when a dio is finished. We're
  524. * particularly interested in the aio/dio case. We use the rw_lock DLM lock
  525. * to protect io on one node from truncation on another.
  526. */
  527. static void ocfs2_dio_end_io(struct kiocb *iocb,
  528. loff_t offset,
  529. ssize_t bytes,
  530. void *private)
  531. {
  532. struct inode *inode = file_inode(iocb->ki_filp);
  533. int level;
  534. /* this io's submitter should not have unlocked this before we could */
  535. BUG_ON(!ocfs2_iocb_is_rw_locked(iocb));
  536. if (ocfs2_iocb_is_sem_locked(iocb))
  537. ocfs2_iocb_clear_sem_locked(iocb);
  538. if (ocfs2_iocb_is_unaligned_aio(iocb)) {
  539. ocfs2_iocb_clear_unaligned_aio(iocb);
  540. mutex_unlock(&OCFS2_I(inode)->ip_unaligned_aio);
  541. }
  542. ocfs2_iocb_clear_rw_locked(iocb);
  543. level = ocfs2_iocb_rw_locked_level(iocb);
  544. ocfs2_rw_unlock(inode, level);
  545. }
  546. static int ocfs2_releasepage(struct page *page, gfp_t wait)
  547. {
  548. if (!page_has_buffers(page))
  549. return 0;
  550. return try_to_free_buffers(page);
  551. }
  552. static int ocfs2_is_overwrite(struct ocfs2_super *osb,
  553. struct inode *inode, loff_t offset)
  554. {
  555. int ret = 0;
  556. u32 v_cpos = 0;
  557. u32 p_cpos = 0;
  558. unsigned int num_clusters = 0;
  559. unsigned int ext_flags = 0;
  560. v_cpos = ocfs2_bytes_to_clusters(osb->sb, offset);
  561. ret = ocfs2_get_clusters(inode, v_cpos, &p_cpos,
  562. &num_clusters, &ext_flags);
  563. if (ret < 0) {
  564. mlog_errno(ret);
  565. return ret;
  566. }
  567. if (p_cpos && !(ext_flags & OCFS2_EXT_UNWRITTEN))
  568. return 1;
  569. return 0;
  570. }
  571. static int ocfs2_direct_IO_zero_extend(struct ocfs2_super *osb,
  572. struct inode *inode, loff_t offset,
  573. u64 zero_len, int cluster_align)
  574. {
  575. u32 p_cpos = 0;
  576. u32 v_cpos = ocfs2_bytes_to_clusters(osb->sb, i_size_read(inode));
  577. unsigned int num_clusters = 0;
  578. unsigned int ext_flags = 0;
  579. int ret = 0;
  580. if (offset <= i_size_read(inode) || cluster_align)
  581. return 0;
  582. ret = ocfs2_get_clusters(inode, v_cpos, &p_cpos, &num_clusters,
  583. &ext_flags);
  584. if (ret < 0) {
  585. mlog_errno(ret);
  586. return ret;
  587. }
  588. if (p_cpos && !(ext_flags & OCFS2_EXT_UNWRITTEN)) {
  589. u64 s = i_size_read(inode);
  590. sector_t sector = (p_cpos << (osb->s_clustersize_bits - 9)) +
  591. (do_div(s, osb->s_clustersize) >> 9);
  592. ret = blkdev_issue_zeroout(osb->sb->s_bdev, sector,
  593. zero_len >> 9, GFP_NOFS, false);
  594. if (ret < 0)
  595. mlog_errno(ret);
  596. }
  597. return ret;
  598. }
  599. static int ocfs2_direct_IO_extend_no_holes(struct ocfs2_super *osb,
  600. struct inode *inode, loff_t offset)
  601. {
  602. u64 zero_start, zero_len, total_zero_len;
  603. u32 p_cpos = 0, clusters_to_add;
  604. u32 v_cpos = ocfs2_bytes_to_clusters(osb->sb, i_size_read(inode));
  605. unsigned int num_clusters = 0;
  606. unsigned int ext_flags = 0;
  607. u32 size_div, offset_div;
  608. int ret = 0;
  609. {
  610. u64 o = offset;
  611. u64 s = i_size_read(inode);
  612. offset_div = do_div(o, osb->s_clustersize);
  613. size_div = do_div(s, osb->s_clustersize);
  614. }
  615. if (offset <= i_size_read(inode))
  616. return 0;
  617. clusters_to_add = ocfs2_bytes_to_clusters(inode->i_sb, offset) -
  618. ocfs2_bytes_to_clusters(inode->i_sb, i_size_read(inode));
  619. total_zero_len = offset - i_size_read(inode);
  620. if (clusters_to_add)
  621. total_zero_len -= offset_div;
  622. /* Allocate clusters to fill out holes, and this is only needed
  623. * when we add more than one clusters. Otherwise the cluster will
  624. * be allocated during direct IO */
  625. if (clusters_to_add > 1) {
  626. ret = ocfs2_extend_allocation(inode,
  627. OCFS2_I(inode)->ip_clusters,
  628. clusters_to_add - 1, 0);
  629. if (ret) {
  630. mlog_errno(ret);
  631. goto out;
  632. }
  633. }
  634. while (total_zero_len) {
  635. ret = ocfs2_get_clusters(inode, v_cpos, &p_cpos, &num_clusters,
  636. &ext_flags);
  637. if (ret < 0) {
  638. mlog_errno(ret);
  639. goto out;
  640. }
  641. zero_start = ocfs2_clusters_to_bytes(osb->sb, p_cpos) +
  642. size_div;
  643. zero_len = ocfs2_clusters_to_bytes(osb->sb, num_clusters) -
  644. size_div;
  645. zero_len = min(total_zero_len, zero_len);
  646. if (p_cpos && !(ext_flags & OCFS2_EXT_UNWRITTEN)) {
  647. ret = blkdev_issue_zeroout(osb->sb->s_bdev,
  648. zero_start >> 9, zero_len >> 9,
  649. GFP_NOFS, false);
  650. if (ret < 0) {
  651. mlog_errno(ret);
  652. goto out;
  653. }
  654. }
  655. total_zero_len -= zero_len;
  656. v_cpos += ocfs2_bytes_to_clusters(osb->sb, zero_len + size_div);
  657. /* Only at first iteration can be cluster not aligned.
  658. * So set size_div to 0 for the rest */
  659. size_div = 0;
  660. }
  661. out:
  662. return ret;
  663. }
  664. static ssize_t ocfs2_direct_IO_write(struct kiocb *iocb,
  665. struct iov_iter *iter,
  666. loff_t offset)
  667. {
  668. ssize_t ret = 0;
  669. ssize_t written = 0;
  670. bool orphaned = false;
  671. int is_overwrite = 0;
  672. struct file *file = iocb->ki_filp;
  673. struct inode *inode = file_inode(file)->i_mapping->host;
  674. struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
  675. struct buffer_head *di_bh = NULL;
  676. size_t count = iter->count;
  677. journal_t *journal = osb->journal->j_journal;
  678. u64 zero_len_head, zero_len_tail;
  679. int cluster_align_head, cluster_align_tail;
  680. loff_t final_size = offset + count;
  681. int append_write = offset >= i_size_read(inode) ? 1 : 0;
  682. unsigned int num_clusters = 0;
  683. unsigned int ext_flags = 0;
  684. {
  685. u64 o = offset;
  686. u64 s = i_size_read(inode);
  687. zero_len_head = do_div(o, 1 << osb->s_clustersize_bits);
  688. cluster_align_head = !zero_len_head;
  689. zero_len_tail = osb->s_clustersize -
  690. do_div(s, osb->s_clustersize);
  691. if ((offset - i_size_read(inode)) < zero_len_tail)
  692. zero_len_tail = offset - i_size_read(inode);
  693. cluster_align_tail = !zero_len_tail;
  694. }
  695. /*
  696. * when final_size > inode->i_size, inode->i_size will be
  697. * updated after direct write, so add the inode to orphan
  698. * dir first.
  699. */
  700. if (final_size > i_size_read(inode)) {
  701. ret = ocfs2_add_inode_to_orphan(osb, inode);
  702. if (ret < 0) {
  703. mlog_errno(ret);
  704. goto out;
  705. }
  706. orphaned = true;
  707. }
  708. if (append_write) {
  709. ret = ocfs2_inode_lock(inode, NULL, 1);
  710. if (ret < 0) {
  711. mlog_errno(ret);
  712. goto clean_orphan;
  713. }
  714. /* zeroing out the previously allocated cluster tail
  715. * that but not zeroed */
  716. if (ocfs2_sparse_alloc(OCFS2_SB(inode->i_sb)))
  717. ret = ocfs2_direct_IO_zero_extend(osb, inode, offset,
  718. zero_len_tail, cluster_align_tail);
  719. else
  720. ret = ocfs2_direct_IO_extend_no_holes(osb, inode,
  721. offset);
  722. if (ret < 0) {
  723. mlog_errno(ret);
  724. ocfs2_inode_unlock(inode, 1);
  725. goto clean_orphan;
  726. }
  727. is_overwrite = ocfs2_is_overwrite(osb, inode, offset);
  728. if (is_overwrite < 0) {
  729. mlog_errno(is_overwrite);
  730. ocfs2_inode_unlock(inode, 1);
  731. goto clean_orphan;
  732. }
  733. ocfs2_inode_unlock(inode, 1);
  734. }
  735. written = __blockdev_direct_IO(iocb, inode, inode->i_sb->s_bdev, iter,
  736. offset, ocfs2_direct_IO_get_blocks,
  737. ocfs2_dio_end_io, NULL, 0);
  738. if (unlikely(written < 0)) {
  739. loff_t i_size = i_size_read(inode);
  740. if (offset + count > i_size) {
  741. ret = ocfs2_inode_lock(inode, &di_bh, 1);
  742. if (ret < 0) {
  743. mlog_errno(ret);
  744. goto clean_orphan;
  745. }
  746. if (i_size == i_size_read(inode)) {
  747. ret = ocfs2_truncate_file(inode, di_bh,
  748. i_size);
  749. if (ret < 0) {
  750. if (ret != -ENOSPC)
  751. mlog_errno(ret);
  752. ocfs2_inode_unlock(inode, 1);
  753. brelse(di_bh);
  754. goto clean_orphan;
  755. }
  756. }
  757. ocfs2_inode_unlock(inode, 1);
  758. brelse(di_bh);
  759. ret = jbd2_journal_force_commit(journal);
  760. if (ret < 0)
  761. mlog_errno(ret);
  762. }
  763. } else if (written > 0 && append_write && !is_overwrite &&
  764. !cluster_align_head) {
  765. /* zeroing out the allocated cluster head */
  766. u32 p_cpos = 0;
  767. u32 v_cpos = ocfs2_bytes_to_clusters(osb->sb, offset);
  768. ret = ocfs2_inode_lock(inode, NULL, 0);
  769. if (ret < 0) {
  770. mlog_errno(ret);
  771. goto clean_orphan;
  772. }
  773. ret = ocfs2_get_clusters(inode, v_cpos, &p_cpos,
  774. &num_clusters, &ext_flags);
  775. if (ret < 0) {
  776. mlog_errno(ret);
  777. ocfs2_inode_unlock(inode, 0);
  778. goto clean_orphan;
  779. }
  780. BUG_ON(!p_cpos || (ext_flags & OCFS2_EXT_UNWRITTEN));
  781. ret = blkdev_issue_zeroout(osb->sb->s_bdev,
  782. p_cpos << (osb->s_clustersize_bits - 9),
  783. zero_len_head >> 9, GFP_NOFS, false);
  784. if (ret < 0)
  785. mlog_errno(ret);
  786. ocfs2_inode_unlock(inode, 0);
  787. }
  788. clean_orphan:
  789. if (orphaned) {
  790. int tmp_ret;
  791. int update_isize = written > 0 ? 1 : 0;
  792. loff_t end = update_isize ? offset + written : 0;
  793. tmp_ret = ocfs2_del_inode_from_orphan(osb, inode,
  794. update_isize, end);
  795. if (tmp_ret < 0) {
  796. ret = tmp_ret;
  797. goto out;
  798. }
  799. tmp_ret = jbd2_journal_force_commit(journal);
  800. if (tmp_ret < 0) {
  801. ret = tmp_ret;
  802. mlog_errno(tmp_ret);
  803. }
  804. }
  805. out:
  806. if (ret >= 0)
  807. ret = written;
  808. return ret;
  809. }
  810. static ssize_t ocfs2_direct_IO(struct kiocb *iocb, struct iov_iter *iter,
  811. loff_t offset)
  812. {
  813. struct file *file = iocb->ki_filp;
  814. struct inode *inode = file_inode(file)->i_mapping->host;
  815. struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
  816. int full_coherency = !(osb->s_mount_opt &
  817. OCFS2_MOUNT_COHERENCY_BUFFERED);
  818. /*
  819. * Fallback to buffered I/O if we see an inode without
  820. * extents.
  821. */
  822. if (OCFS2_I(inode)->ip_dyn_features & OCFS2_INLINE_DATA_FL)
  823. return 0;
  824. /* Fallback to buffered I/O if we are appending and
  825. * concurrent O_DIRECT writes are allowed.
  826. */
  827. if (i_size_read(inode) <= offset && !full_coherency)
  828. return 0;
  829. if (iov_iter_rw(iter) == READ)
  830. return __blockdev_direct_IO(iocb, inode, inode->i_sb->s_bdev,
  831. iter, offset,
  832. ocfs2_direct_IO_get_blocks,
  833. ocfs2_dio_end_io, NULL, 0);
  834. else
  835. return ocfs2_direct_IO_write(iocb, iter, offset);
  836. }
  837. static void ocfs2_figure_cluster_boundaries(struct ocfs2_super *osb,
  838. u32 cpos,
  839. unsigned int *start,
  840. unsigned int *end)
  841. {
  842. unsigned int cluster_start = 0, cluster_end = PAGE_CACHE_SIZE;
  843. if (unlikely(PAGE_CACHE_SHIFT > osb->s_clustersize_bits)) {
  844. unsigned int cpp;
  845. cpp = 1 << (PAGE_CACHE_SHIFT - osb->s_clustersize_bits);
  846. cluster_start = cpos % cpp;
  847. cluster_start = cluster_start << osb->s_clustersize_bits;
  848. cluster_end = cluster_start + osb->s_clustersize;
  849. }
  850. BUG_ON(cluster_start > PAGE_SIZE);
  851. BUG_ON(cluster_end > PAGE_SIZE);
  852. if (start)
  853. *start = cluster_start;
  854. if (end)
  855. *end = cluster_end;
  856. }
  857. /*
  858. * 'from' and 'to' are the region in the page to avoid zeroing.
  859. *
  860. * If pagesize > clustersize, this function will avoid zeroing outside
  861. * of the cluster boundary.
  862. *
  863. * from == to == 0 is code for "zero the entire cluster region"
  864. */
  865. static void ocfs2_clear_page_regions(struct page *page,
  866. struct ocfs2_super *osb, u32 cpos,
  867. unsigned from, unsigned to)
  868. {
  869. void *kaddr;
  870. unsigned int cluster_start, cluster_end;
  871. ocfs2_figure_cluster_boundaries(osb, cpos, &cluster_start, &cluster_end);
  872. kaddr = kmap_atomic(page);
  873. if (from || to) {
  874. if (from > cluster_start)
  875. memset(kaddr + cluster_start, 0, from - cluster_start);
  876. if (to < cluster_end)
  877. memset(kaddr + to, 0, cluster_end - to);
  878. } else {
  879. memset(kaddr + cluster_start, 0, cluster_end - cluster_start);
  880. }
  881. kunmap_atomic(kaddr);
  882. }
  883. /*
  884. * Nonsparse file systems fully allocate before we get to the write
  885. * code. This prevents ocfs2_write() from tagging the write as an
  886. * allocating one, which means ocfs2_map_page_blocks() might try to
  887. * read-in the blocks at the tail of our file. Avoid reading them by
  888. * testing i_size against each block offset.
  889. */
  890. static int ocfs2_should_read_blk(struct inode *inode, struct page *page,
  891. unsigned int block_start)
  892. {
  893. u64 offset = page_offset(page) + block_start;
  894. if (ocfs2_sparse_alloc(OCFS2_SB(inode->i_sb)))
  895. return 1;
  896. if (i_size_read(inode) > offset)
  897. return 1;
  898. return 0;
  899. }
  900. /*
  901. * Some of this taken from __block_write_begin(). We already have our
  902. * mapping by now though, and the entire write will be allocating or
  903. * it won't, so not much need to use BH_New.
  904. *
  905. * This will also skip zeroing, which is handled externally.
  906. */
  907. int ocfs2_map_page_blocks(struct page *page, u64 *p_blkno,
  908. struct inode *inode, unsigned int from,
  909. unsigned int to, int new)
  910. {
  911. int ret = 0;
  912. struct buffer_head *head, *bh, *wait[2], **wait_bh = wait;
  913. unsigned int block_end, block_start;
  914. unsigned int bsize = 1 << inode->i_blkbits;
  915. if (!page_has_buffers(page))
  916. create_empty_buffers(page, bsize, 0);
  917. head = page_buffers(page);
  918. for (bh = head, block_start = 0; bh != head || !block_start;
  919. bh = bh->b_this_page, block_start += bsize) {
  920. block_end = block_start + bsize;
  921. clear_buffer_new(bh);
  922. /*
  923. * Ignore blocks outside of our i/o range -
  924. * they may belong to unallocated clusters.
  925. */
  926. if (block_start >= to || block_end <= from) {
  927. if (PageUptodate(page))
  928. set_buffer_uptodate(bh);
  929. continue;
  930. }
  931. /*
  932. * For an allocating write with cluster size >= page
  933. * size, we always write the entire page.
  934. */
  935. if (new)
  936. set_buffer_new(bh);
  937. if (!buffer_mapped(bh)) {
  938. map_bh(bh, inode->i_sb, *p_blkno);
  939. unmap_underlying_metadata(bh->b_bdev, bh->b_blocknr);
  940. }
  941. if (PageUptodate(page)) {
  942. if (!buffer_uptodate(bh))
  943. set_buffer_uptodate(bh);
  944. } else if (!buffer_uptodate(bh) && !buffer_delay(bh) &&
  945. !buffer_new(bh) &&
  946. ocfs2_should_read_blk(inode, page, block_start) &&
  947. (block_start < from || block_end > to)) {
  948. ll_rw_block(READ, 1, &bh);
  949. *wait_bh++=bh;
  950. }
  951. *p_blkno = *p_blkno + 1;
  952. }
  953. /*
  954. * If we issued read requests - let them complete.
  955. */
  956. while(wait_bh > wait) {
  957. wait_on_buffer(*--wait_bh);
  958. if (!buffer_uptodate(*wait_bh))
  959. ret = -EIO;
  960. }
  961. if (ret == 0 || !new)
  962. return ret;
  963. /*
  964. * If we get -EIO above, zero out any newly allocated blocks
  965. * to avoid exposing stale data.
  966. */
  967. bh = head;
  968. block_start = 0;
  969. do {
  970. block_end = block_start + bsize;
  971. if (block_end <= from)
  972. goto next_bh;
  973. if (block_start >= to)
  974. break;
  975. zero_user(page, block_start, bh->b_size);
  976. set_buffer_uptodate(bh);
  977. mark_buffer_dirty(bh);
  978. next_bh:
  979. block_start = block_end;
  980. bh = bh->b_this_page;
  981. } while (bh != head);
  982. return ret;
  983. }
  984. #if (PAGE_CACHE_SIZE >= OCFS2_MAX_CLUSTERSIZE)
  985. #define OCFS2_MAX_CTXT_PAGES 1
  986. #else
  987. #define OCFS2_MAX_CTXT_PAGES (OCFS2_MAX_CLUSTERSIZE / PAGE_CACHE_SIZE)
  988. #endif
  989. #define OCFS2_MAX_CLUSTERS_PER_PAGE (PAGE_CACHE_SIZE / OCFS2_MIN_CLUSTERSIZE)
  990. /*
  991. * Describe the state of a single cluster to be written to.
  992. */
  993. struct ocfs2_write_cluster_desc {
  994. u32 c_cpos;
  995. u32 c_phys;
  996. /*
  997. * Give this a unique field because c_phys eventually gets
  998. * filled.
  999. */
  1000. unsigned c_new;
  1001. unsigned c_unwritten;
  1002. unsigned c_needs_zero;
  1003. };
  1004. struct ocfs2_write_ctxt {
  1005. /* Logical cluster position / len of write */
  1006. u32 w_cpos;
  1007. u32 w_clen;
  1008. /* First cluster allocated in a nonsparse extend */
  1009. u32 w_first_new_cpos;
  1010. struct ocfs2_write_cluster_desc w_desc[OCFS2_MAX_CLUSTERS_PER_PAGE];
  1011. /*
  1012. * This is true if page_size > cluster_size.
  1013. *
  1014. * It triggers a set of special cases during write which might
  1015. * have to deal with allocating writes to partial pages.
  1016. */
  1017. unsigned int w_large_pages;
  1018. /*
  1019. * Pages involved in this write.
  1020. *
  1021. * w_target_page is the page being written to by the user.
  1022. *
  1023. * w_pages is an array of pages which always contains
  1024. * w_target_page, and in the case of an allocating write with
  1025. * page_size < cluster size, it will contain zero'd and mapped
  1026. * pages adjacent to w_target_page which need to be written
  1027. * out in so that future reads from that region will get
  1028. * zero's.
  1029. */
  1030. unsigned int w_num_pages;
  1031. struct page *w_pages[OCFS2_MAX_CTXT_PAGES];
  1032. struct page *w_target_page;
  1033. /*
  1034. * w_target_locked is used for page_mkwrite path indicating no unlocking
  1035. * against w_target_page in ocfs2_write_end_nolock.
  1036. */
  1037. unsigned int w_target_locked:1;
  1038. /*
  1039. * ocfs2_write_end() uses this to know what the real range to
  1040. * write in the target should be.
  1041. */
  1042. unsigned int w_target_from;
  1043. unsigned int w_target_to;
  1044. /*
  1045. * We could use journal_current_handle() but this is cleaner,
  1046. * IMHO -Mark
  1047. */
  1048. handle_t *w_handle;
  1049. struct buffer_head *w_di_bh;
  1050. struct ocfs2_cached_dealloc_ctxt w_dealloc;
  1051. };
  1052. void ocfs2_unlock_and_free_pages(struct page **pages, int num_pages)
  1053. {
  1054. int i;
  1055. for(i = 0; i < num_pages; i++) {
  1056. if (pages[i]) {
  1057. unlock_page(pages[i]);
  1058. mark_page_accessed(pages[i]);
  1059. page_cache_release(pages[i]);
  1060. }
  1061. }
  1062. }
  1063. static void ocfs2_unlock_pages(struct ocfs2_write_ctxt *wc)
  1064. {
  1065. int i;
  1066. /*
  1067. * w_target_locked is only set to true in the page_mkwrite() case.
  1068. * The intent is to allow us to lock the target page from write_begin()
  1069. * to write_end(). The caller must hold a ref on w_target_page.
  1070. */
  1071. if (wc->w_target_locked) {
  1072. BUG_ON(!wc->w_target_page);
  1073. for (i = 0; i < wc->w_num_pages; i++) {
  1074. if (wc->w_target_page == wc->w_pages[i]) {
  1075. wc->w_pages[i] = NULL;
  1076. break;
  1077. }
  1078. }
  1079. mark_page_accessed(wc->w_target_page);
  1080. page_cache_release(wc->w_target_page);
  1081. }
  1082. ocfs2_unlock_and_free_pages(wc->w_pages, wc->w_num_pages);
  1083. }
  1084. static void ocfs2_free_write_ctxt(struct ocfs2_write_ctxt *wc)
  1085. {
  1086. ocfs2_unlock_pages(wc);
  1087. brelse(wc->w_di_bh);
  1088. kfree(wc);
  1089. }
  1090. static int ocfs2_alloc_write_ctxt(struct ocfs2_write_ctxt **wcp,
  1091. struct ocfs2_super *osb, loff_t pos,
  1092. unsigned len, struct buffer_head *di_bh)
  1093. {
  1094. u32 cend;
  1095. struct ocfs2_write_ctxt *wc;
  1096. wc = kzalloc(sizeof(struct ocfs2_write_ctxt), GFP_NOFS);
  1097. if (!wc)
  1098. return -ENOMEM;
  1099. wc->w_cpos = pos >> osb->s_clustersize_bits;
  1100. wc->w_first_new_cpos = UINT_MAX;
  1101. cend = (pos + len - 1) >> osb->s_clustersize_bits;
  1102. wc->w_clen = cend - wc->w_cpos + 1;
  1103. get_bh(di_bh);
  1104. wc->w_di_bh = di_bh;
  1105. if (unlikely(PAGE_CACHE_SHIFT > osb->s_clustersize_bits))
  1106. wc->w_large_pages = 1;
  1107. else
  1108. wc->w_large_pages = 0;
  1109. ocfs2_init_dealloc_ctxt(&wc->w_dealloc);
  1110. *wcp = wc;
  1111. return 0;
  1112. }
  1113. /*
  1114. * If a page has any new buffers, zero them out here, and mark them uptodate
  1115. * and dirty so they'll be written out (in order to prevent uninitialised
  1116. * block data from leaking). And clear the new bit.
  1117. */
  1118. static void ocfs2_zero_new_buffers(struct page *page, unsigned from, unsigned to)
  1119. {
  1120. unsigned int block_start, block_end;
  1121. struct buffer_head *head, *bh;
  1122. BUG_ON(!PageLocked(page));
  1123. if (!page_has_buffers(page))
  1124. return;
  1125. bh = head = page_buffers(page);
  1126. block_start = 0;
  1127. do {
  1128. block_end = block_start + bh->b_size;
  1129. if (buffer_new(bh)) {
  1130. if (block_end > from && block_start < to) {
  1131. if (!PageUptodate(page)) {
  1132. unsigned start, end;
  1133. start = max(from, block_start);
  1134. end = min(to, block_end);
  1135. zero_user_segment(page, start, end);
  1136. set_buffer_uptodate(bh);
  1137. }
  1138. clear_buffer_new(bh);
  1139. mark_buffer_dirty(bh);
  1140. }
  1141. }
  1142. block_start = block_end;
  1143. bh = bh->b_this_page;
  1144. } while (bh != head);
  1145. }
  1146. /*
  1147. * Only called when we have a failure during allocating write to write
  1148. * zero's to the newly allocated region.
  1149. */
  1150. static void ocfs2_write_failure(struct inode *inode,
  1151. struct ocfs2_write_ctxt *wc,
  1152. loff_t user_pos, unsigned user_len)
  1153. {
  1154. int i;
  1155. unsigned from = user_pos & (PAGE_CACHE_SIZE - 1),
  1156. to = user_pos + user_len;
  1157. struct page *tmppage;
  1158. ocfs2_zero_new_buffers(wc->w_target_page, from, to);
  1159. for(i = 0; i < wc->w_num_pages; i++) {
  1160. tmppage = wc->w_pages[i];
  1161. if (page_has_buffers(tmppage)) {
  1162. if (ocfs2_should_order_data(inode))
  1163. ocfs2_jbd2_file_inode(wc->w_handle, inode);
  1164. block_commit_write(tmppage, from, to);
  1165. }
  1166. }
  1167. }
  1168. static int ocfs2_prepare_page_for_write(struct inode *inode, u64 *p_blkno,
  1169. struct ocfs2_write_ctxt *wc,
  1170. struct page *page, u32 cpos,
  1171. loff_t user_pos, unsigned user_len,
  1172. int new)
  1173. {
  1174. int ret;
  1175. unsigned int map_from = 0, map_to = 0;
  1176. unsigned int cluster_start, cluster_end;
  1177. unsigned int user_data_from = 0, user_data_to = 0;
  1178. ocfs2_figure_cluster_boundaries(OCFS2_SB(inode->i_sb), cpos,
  1179. &cluster_start, &cluster_end);
  1180. /* treat the write as new if the a hole/lseek spanned across
  1181. * the page boundary.
  1182. */
  1183. new = new | ((i_size_read(inode) <= page_offset(page)) &&
  1184. (page_offset(page) <= user_pos));
  1185. if (page == wc->w_target_page) {
  1186. map_from = user_pos & (PAGE_CACHE_SIZE - 1);
  1187. map_to = map_from + user_len;
  1188. if (new)
  1189. ret = ocfs2_map_page_blocks(page, p_blkno, inode,
  1190. cluster_start, cluster_end,
  1191. new);
  1192. else
  1193. ret = ocfs2_map_page_blocks(page, p_blkno, inode,
  1194. map_from, map_to, new);
  1195. if (ret) {
  1196. mlog_errno(ret);
  1197. goto out;
  1198. }
  1199. user_data_from = map_from;
  1200. user_data_to = map_to;
  1201. if (new) {
  1202. map_from = cluster_start;
  1203. map_to = cluster_end;
  1204. }
  1205. } else {
  1206. /*
  1207. * If we haven't allocated the new page yet, we
  1208. * shouldn't be writing it out without copying user
  1209. * data. This is likely a math error from the caller.
  1210. */
  1211. BUG_ON(!new);
  1212. map_from = cluster_start;
  1213. map_to = cluster_end;
  1214. ret = ocfs2_map_page_blocks(page, p_blkno, inode,
  1215. cluster_start, cluster_end, new);
  1216. if (ret) {
  1217. mlog_errno(ret);
  1218. goto out;
  1219. }
  1220. }
  1221. /*
  1222. * Parts of newly allocated pages need to be zero'd.
  1223. *
  1224. * Above, we have also rewritten 'to' and 'from' - as far as
  1225. * the rest of the function is concerned, the entire cluster
  1226. * range inside of a page needs to be written.
  1227. *
  1228. * We can skip this if the page is up to date - it's already
  1229. * been zero'd from being read in as a hole.
  1230. */
  1231. if (new && !PageUptodate(page))
  1232. ocfs2_clear_page_regions(page, OCFS2_SB(inode->i_sb),
  1233. cpos, user_data_from, user_data_to);
  1234. flush_dcache_page(page);
  1235. out:
  1236. return ret;
  1237. }
  1238. /*
  1239. * This function will only grab one clusters worth of pages.
  1240. */
  1241. static int ocfs2_grab_pages_for_write(struct address_space *mapping,
  1242. struct ocfs2_write_ctxt *wc,
  1243. u32 cpos, loff_t user_pos,
  1244. unsigned user_len, int new,
  1245. struct page *mmap_page)
  1246. {
  1247. int ret = 0, i;
  1248. unsigned long start, target_index, end_index, index;
  1249. struct inode *inode = mapping->host;
  1250. loff_t last_byte;
  1251. target_index = user_pos >> PAGE_CACHE_SHIFT;
  1252. /*
  1253. * Figure out how many pages we'll be manipulating here. For
  1254. * non allocating write, we just change the one
  1255. * page. Otherwise, we'll need a whole clusters worth. If we're
  1256. * writing past i_size, we only need enough pages to cover the
  1257. * last page of the write.
  1258. */
  1259. if (new) {
  1260. wc->w_num_pages = ocfs2_pages_per_cluster(inode->i_sb);
  1261. start = ocfs2_align_clusters_to_page_index(inode->i_sb, cpos);
  1262. /*
  1263. * We need the index *past* the last page we could possibly
  1264. * touch. This is the page past the end of the write or
  1265. * i_size, whichever is greater.
  1266. */
  1267. last_byte = max(user_pos + user_len, i_size_read(inode));
  1268. BUG_ON(last_byte < 1);
  1269. end_index = ((last_byte - 1) >> PAGE_CACHE_SHIFT) + 1;
  1270. if ((start + wc->w_num_pages) > end_index)
  1271. wc->w_num_pages = end_index - start;
  1272. } else {
  1273. wc->w_num_pages = 1;
  1274. start = target_index;
  1275. }
  1276. for(i = 0; i < wc->w_num_pages; i++) {
  1277. index = start + i;
  1278. if (index == target_index && mmap_page) {
  1279. /*
  1280. * ocfs2_pagemkwrite() is a little different
  1281. * and wants us to directly use the page
  1282. * passed in.
  1283. */
  1284. lock_page(mmap_page);
  1285. /* Exit and let the caller retry */
  1286. if (mmap_page->mapping != mapping) {
  1287. WARN_ON(mmap_page->mapping);
  1288. unlock_page(mmap_page);
  1289. ret = -EAGAIN;
  1290. goto out;
  1291. }
  1292. page_cache_get(mmap_page);
  1293. wc->w_pages[i] = mmap_page;
  1294. wc->w_target_locked = true;
  1295. } else {
  1296. wc->w_pages[i] = find_or_create_page(mapping, index,
  1297. GFP_NOFS);
  1298. if (!wc->w_pages[i]) {
  1299. ret = -ENOMEM;
  1300. mlog_errno(ret);
  1301. goto out;
  1302. }
  1303. }
  1304. wait_for_stable_page(wc->w_pages[i]);
  1305. if (index == target_index)
  1306. wc->w_target_page = wc->w_pages[i];
  1307. }
  1308. out:
  1309. if (ret)
  1310. wc->w_target_locked = false;
  1311. return ret;
  1312. }
  1313. /*
  1314. * Prepare a single cluster for write one cluster into the file.
  1315. */
  1316. static int ocfs2_write_cluster(struct address_space *mapping,
  1317. u32 phys, unsigned int unwritten,
  1318. unsigned int should_zero,
  1319. struct ocfs2_alloc_context *data_ac,
  1320. struct ocfs2_alloc_context *meta_ac,
  1321. struct ocfs2_write_ctxt *wc, u32 cpos,
  1322. loff_t user_pos, unsigned user_len)
  1323. {
  1324. int ret, i, new;
  1325. u64 v_blkno, p_blkno;
  1326. struct inode *inode = mapping->host;
  1327. struct ocfs2_extent_tree et;
  1328. new = phys == 0 ? 1 : 0;
  1329. if (new) {
  1330. u32 tmp_pos;
  1331. /*
  1332. * This is safe to call with the page locks - it won't take
  1333. * any additional semaphores or cluster locks.
  1334. */
  1335. tmp_pos = cpos;
  1336. ret = ocfs2_add_inode_data(OCFS2_SB(inode->i_sb), inode,
  1337. &tmp_pos, 1, 0, wc->w_di_bh,
  1338. wc->w_handle, data_ac,
  1339. meta_ac, NULL);
  1340. /*
  1341. * This shouldn't happen because we must have already
  1342. * calculated the correct meta data allocation required. The
  1343. * internal tree allocation code should know how to increase
  1344. * transaction credits itself.
  1345. *
  1346. * If need be, we could handle -EAGAIN for a
  1347. * RESTART_TRANS here.
  1348. */
  1349. mlog_bug_on_msg(ret == -EAGAIN,
  1350. "Inode %llu: EAGAIN return during allocation.\n",
  1351. (unsigned long long)OCFS2_I(inode)->ip_blkno);
  1352. if (ret < 0) {
  1353. mlog_errno(ret);
  1354. goto out;
  1355. }
  1356. } else if (unwritten) {
  1357. ocfs2_init_dinode_extent_tree(&et, INODE_CACHE(inode),
  1358. wc->w_di_bh);
  1359. ret = ocfs2_mark_extent_written(inode, &et,
  1360. wc->w_handle, cpos, 1, phys,
  1361. meta_ac, &wc->w_dealloc);
  1362. if (ret < 0) {
  1363. mlog_errno(ret);
  1364. goto out;
  1365. }
  1366. }
  1367. if (should_zero)
  1368. v_blkno = ocfs2_clusters_to_blocks(inode->i_sb, cpos);
  1369. else
  1370. v_blkno = user_pos >> inode->i_sb->s_blocksize_bits;
  1371. /*
  1372. * The only reason this should fail is due to an inability to
  1373. * find the extent added.
  1374. */
  1375. ret = ocfs2_extent_map_get_blocks(inode, v_blkno, &p_blkno, NULL,
  1376. NULL);
  1377. if (ret < 0) {
  1378. mlog(ML_ERROR, "Get physical blkno failed for inode %llu, "
  1379. "at logical block %llu",
  1380. (unsigned long long)OCFS2_I(inode)->ip_blkno,
  1381. (unsigned long long)v_blkno);
  1382. goto out;
  1383. }
  1384. BUG_ON(p_blkno == 0);
  1385. for(i = 0; i < wc->w_num_pages; i++) {
  1386. int tmpret;
  1387. tmpret = ocfs2_prepare_page_for_write(inode, &p_blkno, wc,
  1388. wc->w_pages[i], cpos,
  1389. user_pos, user_len,
  1390. should_zero);
  1391. if (tmpret) {
  1392. mlog_errno(tmpret);
  1393. if (ret == 0)
  1394. ret = tmpret;
  1395. }
  1396. }
  1397. /*
  1398. * We only have cleanup to do in case of allocating write.
  1399. */
  1400. if (ret && new)
  1401. ocfs2_write_failure(inode, wc, user_pos, user_len);
  1402. out:
  1403. return ret;
  1404. }
  1405. static int ocfs2_write_cluster_by_desc(struct address_space *mapping,
  1406. struct ocfs2_alloc_context *data_ac,
  1407. struct ocfs2_alloc_context *meta_ac,
  1408. struct ocfs2_write_ctxt *wc,
  1409. loff_t pos, unsigned len)
  1410. {
  1411. int ret, i;
  1412. loff_t cluster_off;
  1413. unsigned int local_len = len;
  1414. struct ocfs2_write_cluster_desc *desc;
  1415. struct ocfs2_super *osb = OCFS2_SB(mapping->host->i_sb);
  1416. for (i = 0; i < wc->w_clen; i++) {
  1417. desc = &wc->w_desc[i];
  1418. /*
  1419. * We have to make sure that the total write passed in
  1420. * doesn't extend past a single cluster.
  1421. */
  1422. local_len = len;
  1423. cluster_off = pos & (osb->s_clustersize - 1);
  1424. if ((cluster_off + local_len) > osb->s_clustersize)
  1425. local_len = osb->s_clustersize - cluster_off;
  1426. ret = ocfs2_write_cluster(mapping, desc->c_phys,
  1427. desc->c_unwritten,
  1428. desc->c_needs_zero,
  1429. data_ac, meta_ac,
  1430. wc, desc->c_cpos, pos, local_len);
  1431. if (ret) {
  1432. mlog_errno(ret);
  1433. goto out;
  1434. }
  1435. len -= local_len;
  1436. pos += local_len;
  1437. }
  1438. ret = 0;
  1439. out:
  1440. return ret;
  1441. }
  1442. /*
  1443. * ocfs2_write_end() wants to know which parts of the target page it
  1444. * should complete the write on. It's easiest to compute them ahead of
  1445. * time when a more complete view of the write is available.
  1446. */
  1447. static void ocfs2_set_target_boundaries(struct ocfs2_super *osb,
  1448. struct ocfs2_write_ctxt *wc,
  1449. loff_t pos, unsigned len, int alloc)
  1450. {
  1451. struct ocfs2_write_cluster_desc *desc;
  1452. wc->w_target_from = pos & (PAGE_CACHE_SIZE - 1);
  1453. wc->w_target_to = wc->w_target_from + len;
  1454. if (alloc == 0)
  1455. return;
  1456. /*
  1457. * Allocating write - we may have different boundaries based
  1458. * on page size and cluster size.
  1459. *
  1460. * NOTE: We can no longer compute one value from the other as
  1461. * the actual write length and user provided length may be
  1462. * different.
  1463. */
  1464. if (wc->w_large_pages) {
  1465. /*
  1466. * We only care about the 1st and last cluster within
  1467. * our range and whether they should be zero'd or not. Either
  1468. * value may be extended out to the start/end of a
  1469. * newly allocated cluster.
  1470. */
  1471. desc = &wc->w_desc[0];
  1472. if (desc->c_needs_zero)
  1473. ocfs2_figure_cluster_boundaries(osb,
  1474. desc->c_cpos,
  1475. &wc->w_target_from,
  1476. NULL);
  1477. desc = &wc->w_desc[wc->w_clen - 1];
  1478. if (desc->c_needs_zero)
  1479. ocfs2_figure_cluster_boundaries(osb,
  1480. desc->c_cpos,
  1481. NULL,
  1482. &wc->w_target_to);
  1483. } else {
  1484. wc->w_target_from = 0;
  1485. wc->w_target_to = PAGE_CACHE_SIZE;
  1486. }
  1487. }
  1488. /*
  1489. * Populate each single-cluster write descriptor in the write context
  1490. * with information about the i/o to be done.
  1491. *
  1492. * Returns the number of clusters that will have to be allocated, as
  1493. * well as a worst case estimate of the number of extent records that
  1494. * would have to be created during a write to an unwritten region.
  1495. */
  1496. static int ocfs2_populate_write_desc(struct inode *inode,
  1497. struct ocfs2_write_ctxt *wc,
  1498. unsigned int *clusters_to_alloc,
  1499. unsigned int *extents_to_split)
  1500. {
  1501. int ret;
  1502. struct ocfs2_write_cluster_desc *desc;
  1503. unsigned int num_clusters = 0;
  1504. unsigned int ext_flags = 0;
  1505. u32 phys = 0;
  1506. int i;
  1507. *clusters_to_alloc = 0;
  1508. *extents_to_split = 0;
  1509. for (i = 0; i < wc->w_clen; i++) {
  1510. desc = &wc->w_desc[i];
  1511. desc->c_cpos = wc->w_cpos + i;
  1512. if (num_clusters == 0) {
  1513. /*
  1514. * Need to look up the next extent record.
  1515. */
  1516. ret = ocfs2_get_clusters(inode, desc->c_cpos, &phys,
  1517. &num_clusters, &ext_flags);
  1518. if (ret) {
  1519. mlog_errno(ret);
  1520. goto out;
  1521. }
  1522. /* We should already CoW the refcountd extent. */
  1523. BUG_ON(ext_flags & OCFS2_EXT_REFCOUNTED);
  1524. /*
  1525. * Assume worst case - that we're writing in
  1526. * the middle of the extent.
  1527. *
  1528. * We can assume that the write proceeds from
  1529. * left to right, in which case the extent
  1530. * insert code is smart enough to coalesce the
  1531. * next splits into the previous records created.
  1532. */
  1533. if (ext_flags & OCFS2_EXT_UNWRITTEN)
  1534. *extents_to_split = *extents_to_split + 2;
  1535. } else if (phys) {
  1536. /*
  1537. * Only increment phys if it doesn't describe
  1538. * a hole.
  1539. */
  1540. phys++;
  1541. }
  1542. /*
  1543. * If w_first_new_cpos is < UINT_MAX, we have a non-sparse
  1544. * file that got extended. w_first_new_cpos tells us
  1545. * where the newly allocated clusters are so we can
  1546. * zero them.
  1547. */
  1548. if (desc->c_cpos >= wc->w_first_new_cpos) {
  1549. BUG_ON(phys == 0);
  1550. desc->c_needs_zero = 1;
  1551. }
  1552. desc->c_phys = phys;
  1553. if (phys == 0) {
  1554. desc->c_new = 1;
  1555. desc->c_needs_zero = 1;
  1556. *clusters_to_alloc = *clusters_to_alloc + 1;
  1557. }
  1558. if (ext_flags & OCFS2_EXT_UNWRITTEN) {
  1559. desc->c_unwritten = 1;
  1560. desc->c_needs_zero = 1;
  1561. }
  1562. num_clusters--;
  1563. }
  1564. ret = 0;
  1565. out:
  1566. return ret;
  1567. }
  1568. static int ocfs2_write_begin_inline(struct address_space *mapping,
  1569. struct inode *inode,
  1570. struct ocfs2_write_ctxt *wc)
  1571. {
  1572. int ret;
  1573. struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
  1574. struct page *page;
  1575. handle_t *handle;
  1576. struct ocfs2_dinode *di = (struct ocfs2_dinode *)wc->w_di_bh->b_data;
  1577. handle = ocfs2_start_trans(osb, OCFS2_INODE_UPDATE_CREDITS);
  1578. if (IS_ERR(handle)) {
  1579. ret = PTR_ERR(handle);
  1580. mlog_errno(ret);
  1581. goto out;
  1582. }
  1583. page = find_or_create_page(mapping, 0, GFP_NOFS);
  1584. if (!page) {
  1585. ocfs2_commit_trans(osb, handle);
  1586. ret = -ENOMEM;
  1587. mlog_errno(ret);
  1588. goto out;
  1589. }
  1590. /*
  1591. * If we don't set w_num_pages then this page won't get unlocked
  1592. * and freed on cleanup of the write context.
  1593. */
  1594. wc->w_pages[0] = wc->w_target_page = page;
  1595. wc->w_num_pages = 1;
  1596. ret = ocfs2_journal_access_di(handle, INODE_CACHE(inode), wc->w_di_bh,
  1597. OCFS2_JOURNAL_ACCESS_WRITE);
  1598. if (ret) {
  1599. ocfs2_commit_trans(osb, handle);
  1600. mlog_errno(ret);
  1601. goto out;
  1602. }
  1603. if (!(OCFS2_I(inode)->ip_dyn_features & OCFS2_INLINE_DATA_FL))
  1604. ocfs2_set_inode_data_inline(inode, di);
  1605. if (!PageUptodate(page)) {
  1606. ret = ocfs2_read_inline_data(inode, page, wc->w_di_bh);
  1607. if (ret) {
  1608. ocfs2_commit_trans(osb, handle);
  1609. goto out;
  1610. }
  1611. }
  1612. wc->w_handle = handle;
  1613. out:
  1614. return ret;
  1615. }
  1616. int ocfs2_size_fits_inline_data(struct buffer_head *di_bh, u64 new_size)
  1617. {
  1618. struct ocfs2_dinode *di = (struct ocfs2_dinode *)di_bh->b_data;
  1619. if (new_size <= le16_to_cpu(di->id2.i_data.id_count))
  1620. return 1;
  1621. return 0;
  1622. }
  1623. static int ocfs2_try_to_write_inline_data(struct address_space *mapping,
  1624. struct inode *inode, loff_t pos,
  1625. unsigned len, struct page *mmap_page,
  1626. struct ocfs2_write_ctxt *wc)
  1627. {
  1628. int ret, written = 0;
  1629. loff_t end = pos + len;
  1630. struct ocfs2_inode_info *oi = OCFS2_I(inode);
  1631. struct ocfs2_dinode *di = NULL;
  1632. trace_ocfs2_try_to_write_inline_data((unsigned long long)oi->ip_blkno,
  1633. len, (unsigned long long)pos,
  1634. oi->ip_dyn_features);
  1635. /*
  1636. * Handle inodes which already have inline data 1st.
  1637. */
  1638. if (oi->ip_dyn_features & OCFS2_INLINE_DATA_FL) {
  1639. if (mmap_page == NULL &&
  1640. ocfs2_size_fits_inline_data(wc->w_di_bh, end))
  1641. goto do_inline_write;
  1642. /*
  1643. * The write won't fit - we have to give this inode an
  1644. * inline extent list now.
  1645. */
  1646. ret = ocfs2_convert_inline_data_to_extents(inode, wc->w_di_bh);
  1647. if (ret)
  1648. mlog_errno(ret);
  1649. goto out;
  1650. }
  1651. /*
  1652. * Check whether the inode can accept inline data.
  1653. */
  1654. if (oi->ip_clusters != 0 || i_size_read(inode) != 0)
  1655. return 0;
  1656. /*
  1657. * Check whether the write can fit.
  1658. */
  1659. di = (struct ocfs2_dinode *)wc->w_di_bh->b_data;
  1660. if (mmap_page ||
  1661. end > ocfs2_max_inline_data_with_xattr(inode->i_sb, di))
  1662. return 0;
  1663. do_inline_write:
  1664. ret = ocfs2_write_begin_inline(mapping, inode, wc);
  1665. if (ret) {
  1666. mlog_errno(ret);
  1667. goto out;
  1668. }
  1669. /*
  1670. * This signals to the caller that the data can be written
  1671. * inline.
  1672. */
  1673. written = 1;
  1674. out:
  1675. return written ? written : ret;
  1676. }
  1677. /*
  1678. * This function only does anything for file systems which can't
  1679. * handle sparse files.
  1680. *
  1681. * What we want to do here is fill in any hole between the current end
  1682. * of allocation and the end of our write. That way the rest of the
  1683. * write path can treat it as an non-allocating write, which has no
  1684. * special case code for sparse/nonsparse files.
  1685. */
  1686. static int ocfs2_expand_nonsparse_inode(struct inode *inode,
  1687. struct buffer_head *di_bh,
  1688. loff_t pos, unsigned len,
  1689. struct ocfs2_write_ctxt *wc)
  1690. {
  1691. int ret;
  1692. loff_t newsize = pos + len;
  1693. BUG_ON(ocfs2_sparse_alloc(OCFS2_SB(inode->i_sb)));
  1694. if (newsize <= i_size_read(inode))
  1695. return 0;
  1696. ret = ocfs2_extend_no_holes(inode, di_bh, newsize, pos);
  1697. if (ret)
  1698. mlog_errno(ret);
  1699. wc->w_first_new_cpos =
  1700. ocfs2_clusters_for_bytes(inode->i_sb, i_size_read(inode));
  1701. return ret;
  1702. }
  1703. static int ocfs2_zero_tail(struct inode *inode, struct buffer_head *di_bh,
  1704. loff_t pos)
  1705. {
  1706. int ret = 0;
  1707. BUG_ON(!ocfs2_sparse_alloc(OCFS2_SB(inode->i_sb)));
  1708. if (pos > i_size_read(inode))
  1709. ret = ocfs2_zero_extend(inode, di_bh, pos);
  1710. return ret;
  1711. }
  1712. /*
  1713. * Try to flush truncate logs if we can free enough clusters from it.
  1714. * As for return value, "< 0" means error, "0" no space and "1" means
  1715. * we have freed enough spaces and let the caller try to allocate again.
  1716. */
  1717. static int ocfs2_try_to_free_truncate_log(struct ocfs2_super *osb,
  1718. unsigned int needed)
  1719. {
  1720. tid_t target;
  1721. int ret = 0;
  1722. unsigned int truncated_clusters;
  1723. mutex_lock(&osb->osb_tl_inode->i_mutex);
  1724. truncated_clusters = osb->truncated_clusters;
  1725. mutex_unlock(&osb->osb_tl_inode->i_mutex);
  1726. /*
  1727. * Check whether we can succeed in allocating if we free
  1728. * the truncate log.
  1729. */
  1730. if (truncated_clusters < needed)
  1731. goto out;
  1732. ret = ocfs2_flush_truncate_log(osb);
  1733. if (ret) {
  1734. mlog_errno(ret);
  1735. goto out;
  1736. }
  1737. if (jbd2_journal_start_commit(osb->journal->j_journal, &target)) {
  1738. jbd2_log_wait_commit(osb->journal->j_journal, target);
  1739. ret = 1;
  1740. }
  1741. out:
  1742. return ret;
  1743. }
  1744. int ocfs2_write_begin_nolock(struct file *filp,
  1745. struct address_space *mapping,
  1746. loff_t pos, unsigned len, unsigned flags,
  1747. struct page **pagep, void **fsdata,
  1748. struct buffer_head *di_bh, struct page *mmap_page)
  1749. {
  1750. int ret, cluster_of_pages, credits = OCFS2_INODE_UPDATE_CREDITS;
  1751. unsigned int clusters_to_alloc, extents_to_split, clusters_need = 0;
  1752. struct ocfs2_write_ctxt *wc;
  1753. struct inode *inode = mapping->host;
  1754. struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
  1755. struct ocfs2_dinode *di;
  1756. struct ocfs2_alloc_context *data_ac = NULL;
  1757. struct ocfs2_alloc_context *meta_ac = NULL;
  1758. handle_t *handle;
  1759. struct ocfs2_extent_tree et;
  1760. int try_free = 1, ret1;
  1761. try_again:
  1762. ret = ocfs2_alloc_write_ctxt(&wc, osb, pos, len, di_bh);
  1763. if (ret) {
  1764. mlog_errno(ret);
  1765. return ret;
  1766. }
  1767. if (ocfs2_supports_inline_data(osb)) {
  1768. ret = ocfs2_try_to_write_inline_data(mapping, inode, pos, len,
  1769. mmap_page, wc);
  1770. if (ret == 1) {
  1771. ret = 0;
  1772. goto success;
  1773. }
  1774. if (ret < 0) {
  1775. mlog_errno(ret);
  1776. goto out;
  1777. }
  1778. }
  1779. if (ocfs2_sparse_alloc(osb))
  1780. ret = ocfs2_zero_tail(inode, di_bh, pos);
  1781. else
  1782. ret = ocfs2_expand_nonsparse_inode(inode, di_bh, pos, len,
  1783. wc);
  1784. if (ret) {
  1785. mlog_errno(ret);
  1786. goto out;
  1787. }
  1788. ret = ocfs2_check_range_for_refcount(inode, pos, len);
  1789. if (ret < 0) {
  1790. mlog_errno(ret);
  1791. goto out;
  1792. } else if (ret == 1) {
  1793. clusters_need = wc->w_clen;
  1794. ret = ocfs2_refcount_cow(inode, di_bh,
  1795. wc->w_cpos, wc->w_clen, UINT_MAX);
  1796. if (ret) {
  1797. mlog_errno(ret);
  1798. goto out;
  1799. }
  1800. }
  1801. ret = ocfs2_populate_write_desc(inode, wc, &clusters_to_alloc,
  1802. &extents_to_split);
  1803. if (ret) {
  1804. mlog_errno(ret);
  1805. goto out;
  1806. }
  1807. clusters_need += clusters_to_alloc;
  1808. di = (struct ocfs2_dinode *)wc->w_di_bh->b_data;
  1809. trace_ocfs2_write_begin_nolock(
  1810. (unsigned long long)OCFS2_I(inode)->ip_blkno,
  1811. (long long)i_size_read(inode),
  1812. le32_to_cpu(di->i_clusters),
  1813. pos, len, flags, mmap_page,
  1814. clusters_to_alloc, extents_to_split);
  1815. /*
  1816. * We set w_target_from, w_target_to here so that
  1817. * ocfs2_write_end() knows which range in the target page to
  1818. * write out. An allocation requires that we write the entire
  1819. * cluster range.
  1820. */
  1821. if (clusters_to_alloc || extents_to_split) {
  1822. /*
  1823. * XXX: We are stretching the limits of
  1824. * ocfs2_lock_allocators(). It greatly over-estimates
  1825. * the work to be done.
  1826. */
  1827. ocfs2_init_dinode_extent_tree(&et, INODE_CACHE(inode),
  1828. wc->w_di_bh);
  1829. ret = ocfs2_lock_allocators(inode, &et,
  1830. clusters_to_alloc, extents_to_split,
  1831. &data_ac, &meta_ac);
  1832. if (ret) {
  1833. mlog_errno(ret);
  1834. goto out;
  1835. }
  1836. if (data_ac)
  1837. data_ac->ac_resv = &OCFS2_I(inode)->ip_la_data_resv;
  1838. credits = ocfs2_calc_extend_credits(inode->i_sb,
  1839. &di->id2.i_list);
  1840. }
  1841. /*
  1842. * We have to zero sparse allocated clusters, unwritten extent clusters,
  1843. * and non-sparse clusters we just extended. For non-sparse writes,
  1844. * we know zeros will only be needed in the first and/or last cluster.
  1845. */
  1846. if (clusters_to_alloc || extents_to_split ||
  1847. (wc->w_clen && (wc->w_desc[0].c_needs_zero ||
  1848. wc->w_desc[wc->w_clen - 1].c_needs_zero)))
  1849. cluster_of_pages = 1;
  1850. else
  1851. cluster_of_pages = 0;
  1852. ocfs2_set_target_boundaries(osb, wc, pos, len, cluster_of_pages);
  1853. handle = ocfs2_start_trans(osb, credits);
  1854. if (IS_ERR(handle)) {
  1855. ret = PTR_ERR(handle);
  1856. mlog_errno(ret);
  1857. goto out;
  1858. }
  1859. wc->w_handle = handle;
  1860. if (clusters_to_alloc) {
  1861. ret = dquot_alloc_space_nodirty(inode,
  1862. ocfs2_clusters_to_bytes(osb->sb, clusters_to_alloc));
  1863. if (ret)
  1864. goto out_commit;
  1865. }
  1866. /*
  1867. * We don't want this to fail in ocfs2_write_end(), so do it
  1868. * here.
  1869. */
  1870. ret = ocfs2_journal_access_di(handle, INODE_CACHE(inode), wc->w_di_bh,
  1871. OCFS2_JOURNAL_ACCESS_WRITE);
  1872. if (ret) {
  1873. mlog_errno(ret);
  1874. goto out_quota;
  1875. }
  1876. /*
  1877. * Fill our page array first. That way we've grabbed enough so
  1878. * that we can zero and flush if we error after adding the
  1879. * extent.
  1880. */
  1881. ret = ocfs2_grab_pages_for_write(mapping, wc, wc->w_cpos, pos, len,
  1882. cluster_of_pages, mmap_page);
  1883. if (ret && ret != -EAGAIN) {
  1884. mlog_errno(ret);
  1885. goto out_quota;
  1886. }
  1887. /*
  1888. * ocfs2_grab_pages_for_write() returns -EAGAIN if it could not lock
  1889. * the target page. In this case, we exit with no error and no target
  1890. * page. This will trigger the caller, page_mkwrite(), to re-try
  1891. * the operation.
  1892. */
  1893. if (ret == -EAGAIN) {
  1894. BUG_ON(wc->w_target_page);
  1895. ret = 0;
  1896. goto out_quota;
  1897. }
  1898. ret = ocfs2_write_cluster_by_desc(mapping, data_ac, meta_ac, wc, pos,
  1899. len);
  1900. if (ret) {
  1901. mlog_errno(ret);
  1902. goto out_quota;
  1903. }
  1904. if (data_ac)
  1905. ocfs2_free_alloc_context(data_ac);
  1906. if (meta_ac)
  1907. ocfs2_free_alloc_context(meta_ac);
  1908. success:
  1909. *pagep = wc->w_target_page;
  1910. *fsdata = wc;
  1911. return 0;
  1912. out_quota:
  1913. if (clusters_to_alloc)
  1914. dquot_free_space(inode,
  1915. ocfs2_clusters_to_bytes(osb->sb, clusters_to_alloc));
  1916. out_commit:
  1917. ocfs2_commit_trans(osb, handle);
  1918. out:
  1919. ocfs2_free_write_ctxt(wc);
  1920. if (data_ac) {
  1921. ocfs2_free_alloc_context(data_ac);
  1922. data_ac = NULL;
  1923. }
  1924. if (meta_ac) {
  1925. ocfs2_free_alloc_context(meta_ac);
  1926. meta_ac = NULL;
  1927. }
  1928. if (ret == -ENOSPC && try_free) {
  1929. /*
  1930. * Try to free some truncate log so that we can have enough
  1931. * clusters to allocate.
  1932. */
  1933. try_free = 0;
  1934. ret1 = ocfs2_try_to_free_truncate_log(osb, clusters_need);
  1935. if (ret1 == 1)
  1936. goto try_again;
  1937. if (ret1 < 0)
  1938. mlog_errno(ret1);
  1939. }
  1940. return ret;
  1941. }
  1942. static int ocfs2_write_begin(struct file *file, struct address_space *mapping,
  1943. loff_t pos, unsigned len, unsigned flags,
  1944. struct page **pagep, void **fsdata)
  1945. {
  1946. int ret;
  1947. struct buffer_head *di_bh = NULL;
  1948. struct inode *inode = mapping->host;
  1949. ret = ocfs2_inode_lock(inode, &di_bh, 1);
  1950. if (ret) {
  1951. mlog_errno(ret);
  1952. return ret;
  1953. }
  1954. /*
  1955. * Take alloc sem here to prevent concurrent lookups. That way
  1956. * the mapping, zeroing and tree manipulation within
  1957. * ocfs2_write() will be safe against ->readpage(). This
  1958. * should also serve to lock out allocation from a shared
  1959. * writeable region.
  1960. */
  1961. down_write(&OCFS2_I(inode)->ip_alloc_sem);
  1962. ret = ocfs2_write_begin_nolock(file, mapping, pos, len, flags, pagep,
  1963. fsdata, di_bh, NULL);
  1964. if (ret) {
  1965. mlog_errno(ret);
  1966. goto out_fail;
  1967. }
  1968. brelse(di_bh);
  1969. return 0;
  1970. out_fail:
  1971. up_write(&OCFS2_I(inode)->ip_alloc_sem);
  1972. brelse(di_bh);
  1973. ocfs2_inode_unlock(inode, 1);
  1974. return ret;
  1975. }
  1976. static void ocfs2_write_end_inline(struct inode *inode, loff_t pos,
  1977. unsigned len, unsigned *copied,
  1978. struct ocfs2_dinode *di,
  1979. struct ocfs2_write_ctxt *wc)
  1980. {
  1981. void *kaddr;
  1982. if (unlikely(*copied < len)) {
  1983. if (!PageUptodate(wc->w_target_page)) {
  1984. *copied = 0;
  1985. return;
  1986. }
  1987. }
  1988. kaddr = kmap_atomic(wc->w_target_page);
  1989. memcpy(di->id2.i_data.id_data + pos, kaddr + pos, *copied);
  1990. kunmap_atomic(kaddr);
  1991. trace_ocfs2_write_end_inline(
  1992. (unsigned long long)OCFS2_I(inode)->ip_blkno,
  1993. (unsigned long long)pos, *copied,
  1994. le16_to_cpu(di->id2.i_data.id_count),
  1995. le16_to_cpu(di->i_dyn_features));
  1996. }
  1997. int ocfs2_write_end_nolock(struct address_space *mapping,
  1998. loff_t pos, unsigned len, unsigned copied,
  1999. struct page *page, void *fsdata)
  2000. {
  2001. int i;
  2002. unsigned from, to, start = pos & (PAGE_CACHE_SIZE - 1);
  2003. struct inode *inode = mapping->host;
  2004. struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
  2005. struct ocfs2_write_ctxt *wc = fsdata;
  2006. struct ocfs2_dinode *di = (struct ocfs2_dinode *)wc->w_di_bh->b_data;
  2007. handle_t *handle = wc->w_handle;
  2008. struct page *tmppage;
  2009. if (OCFS2_I(inode)->ip_dyn_features & OCFS2_INLINE_DATA_FL) {
  2010. ocfs2_write_end_inline(inode, pos, len, &copied, di, wc);
  2011. goto out_write_size;
  2012. }
  2013. if (unlikely(copied < len)) {
  2014. if (!PageUptodate(wc->w_target_page))
  2015. copied = 0;
  2016. ocfs2_zero_new_buffers(wc->w_target_page, start+copied,
  2017. start+len);
  2018. }
  2019. flush_dcache_page(wc->w_target_page);
  2020. for(i = 0; i < wc->w_num_pages; i++) {
  2021. tmppage = wc->w_pages[i];
  2022. if (tmppage == wc->w_target_page) {
  2023. from = wc->w_target_from;
  2024. to = wc->w_target_to;
  2025. BUG_ON(from > PAGE_CACHE_SIZE ||
  2026. to > PAGE_CACHE_SIZE ||
  2027. to < from);
  2028. } else {
  2029. /*
  2030. * Pages adjacent to the target (if any) imply
  2031. * a hole-filling write in which case we want
  2032. * to flush their entire range.
  2033. */
  2034. from = 0;
  2035. to = PAGE_CACHE_SIZE;
  2036. }
  2037. if (page_has_buffers(tmppage)) {
  2038. if (ocfs2_should_order_data(inode))
  2039. ocfs2_jbd2_file_inode(wc->w_handle, inode);
  2040. block_commit_write(tmppage, from, to);
  2041. }
  2042. }
  2043. out_write_size:
  2044. pos += copied;
  2045. if (pos > i_size_read(inode)) {
  2046. i_size_write(inode, pos);
  2047. mark_inode_dirty(inode);
  2048. }
  2049. inode->i_blocks = ocfs2_inode_sector_count(inode);
  2050. di->i_size = cpu_to_le64((u64)i_size_read(inode));
  2051. inode->i_mtime = inode->i_ctime = CURRENT_TIME;
  2052. di->i_mtime = di->i_ctime = cpu_to_le64(inode->i_mtime.tv_sec);
  2053. di->i_mtime_nsec = di->i_ctime_nsec = cpu_to_le32(inode->i_mtime.tv_nsec);
  2054. ocfs2_update_inode_fsync_trans(handle, inode, 1);
  2055. ocfs2_journal_dirty(handle, wc->w_di_bh);
  2056. /* unlock pages before dealloc since it needs acquiring j_trans_barrier
  2057. * lock, or it will cause a deadlock since journal commit threads holds
  2058. * this lock and will ask for the page lock when flushing the data.
  2059. * put it here to preserve the unlock order.
  2060. */
  2061. ocfs2_unlock_pages(wc);
  2062. ocfs2_commit_trans(osb, handle);
  2063. ocfs2_run_deallocs(osb, &wc->w_dealloc);
  2064. brelse(wc->w_di_bh);
  2065. kfree(wc);
  2066. return copied;
  2067. }
  2068. static int ocfs2_write_end(struct file *file, struct address_space *mapping,
  2069. loff_t pos, unsigned len, unsigned copied,
  2070. struct page *page, void *fsdata)
  2071. {
  2072. int ret;
  2073. struct inode *inode = mapping->host;
  2074. ret = ocfs2_write_end_nolock(mapping, pos, len, copied, page, fsdata);
  2075. up_write(&OCFS2_I(inode)->ip_alloc_sem);
  2076. ocfs2_inode_unlock(inode, 1);
  2077. return ret;
  2078. }
  2079. const struct address_space_operations ocfs2_aops = {
  2080. .readpage = ocfs2_readpage,
  2081. .readpages = ocfs2_readpages,
  2082. .writepage = ocfs2_writepage,
  2083. .write_begin = ocfs2_write_begin,
  2084. .write_end = ocfs2_write_end,
  2085. .bmap = ocfs2_bmap,
  2086. .direct_IO = ocfs2_direct_IO,
  2087. .invalidatepage = block_invalidatepage,
  2088. .releasepage = ocfs2_releasepage,
  2089. .migratepage = buffer_migrate_page,
  2090. .is_partially_uptodate = block_is_partially_uptodate,
  2091. .error_remove_page = generic_error_remove_page,
  2092. };