inode.c 98 KB

123456789101112131415161718192021222324252627282930313233343536373839404142434445464748495051525354555657585960616263646566676869707172737475767778798081828384858687888990919293949596979899100101102103104105106107108109110111112113114115116117118119120121122123124125126127128129130131132133134135136137138139140141142143144145146147148149150151152153154155156157158159160161162163164165166167168169170171172173174175176177178179180181182183184185186187188189190191192193194195196197198199200201202203204205206207208209210211212213214215216217218219220221222223224225226227228229230231232233234235236237238239240241242243244245246247248249250251252253254255256257258259260261262263264265266267268269270271272273274275276277278279280281282283284285286287288289290291292293294295296297298299300301302303304305306307308309310311312313314315316317318319320321322323324325326327328329330331332333334335336337338339340341342343344345346347348349350351352353354355356357358359360361362363364365366367368369370371372373374375376377378379380381382383384385386387388389390391392393394395396397398399400401402403404405406407408409410411412413414415416417418419420421422423424425426427428429430431432433434435436437438439440441442443444445446447448449450451452453454455456457458459460461462463464465466467468469470471472473474475476477478479480481482483484485486487488489490491492493494495496497498499500501502503504505506507508509510511512513514515516517518519520521522523524525526527528529530531532533534535536537538539540541542543544545546547548549550551552553554555556557558559560561562563564565566567568569570571572573574575576577578579580581582583584585586587588589590591592593594595596597598599600601602603604605606607608609610611612613614615616617618619620621622623624625626627628629630631632633634635636637638639640641642643644645646647648649650651652653654655656657658659660661662663664665666667668669670671672673674675676677678679680681682683684685686687688689690691692693694695696697698699700701702703704705706707708709710711712713714715716717718719720721722723724725726727728729730731732733734735736737738739740741742743744745746747748749750751752753754755756757758759760761762763764765766767768769770771772773774775776777778779780781782783784785786787788789790791792793794795796797798799800801802803804805806807808809810811812813814815816817818819820821822823824825826827828829830831832833834835836837838839840841842843844845846847848849850851852853854855856857858859860861862863864865866867868869870871872873874875876877878879880881882883884885886887888889890891892893894895896897898899900901902903904905906907908909910911912913914915916917918919920921922923924925926927928929930931932933934935936937938939940941942943944945946947948949950951952953954955956957958959960961962963964965966967968969970971972973974975976977978979980981982983984985986987988989990991992993994995996997998999100010011002100310041005100610071008100910101011101210131014101510161017101810191020102110221023102410251026102710281029103010311032103310341035103610371038103910401041104210431044104510461047104810491050105110521053105410551056105710581059106010611062106310641065106610671068106910701071107210731074107510761077107810791080108110821083108410851086108710881089109010911092109310941095109610971098109911001101110211031104110511061107110811091110111111121113111411151116111711181119112011211122112311241125112611271128112911301131113211331134113511361137113811391140114111421143114411451146114711481149115011511152115311541155115611571158115911601161116211631164116511661167116811691170117111721173117411751176117711781179118011811182118311841185118611871188118911901191119211931194119511961197119811991200120112021203120412051206120712081209121012111212121312141215121612171218121912201221122212231224122512261227122812291230123112321233123412351236123712381239124012411242124312441245124612471248124912501251125212531254125512561257125812591260126112621263126412651266126712681269127012711272127312741275127612771278127912801281128212831284128512861287128812891290129112921293129412951296129712981299130013011302130313041305130613071308130913101311131213131314131513161317131813191320132113221323132413251326132713281329133013311332133313341335133613371338133913401341134213431344134513461347134813491350135113521353135413551356135713581359136013611362136313641365136613671368136913701371137213731374137513761377137813791380138113821383138413851386138713881389139013911392139313941395139613971398139914001401140214031404140514061407140814091410141114121413141414151416141714181419142014211422142314241425142614271428142914301431143214331434143514361437143814391440144114421443144414451446144714481449145014511452145314541455145614571458145914601461146214631464146514661467146814691470147114721473147414751476147714781479148014811482148314841485148614871488148914901491149214931494149514961497149814991500150115021503150415051506150715081509151015111512151315141515151615171518151915201521152215231524152515261527152815291530153115321533153415351536153715381539154015411542154315441545154615471548154915501551155215531554155515561557155815591560156115621563156415651566156715681569157015711572157315741575157615771578157915801581158215831584158515861587158815891590159115921593159415951596159715981599160016011602160316041605160616071608160916101611161216131614161516161617161816191620162116221623162416251626162716281629163016311632163316341635163616371638163916401641164216431644164516461647164816491650165116521653165416551656165716581659166016611662166316641665166616671668166916701671167216731674167516761677167816791680168116821683168416851686168716881689169016911692169316941695169616971698169917001701170217031704170517061707170817091710171117121713171417151716171717181719172017211722172317241725172617271728172917301731173217331734173517361737173817391740174117421743174417451746174717481749175017511752175317541755175617571758175917601761176217631764176517661767176817691770177117721773177417751776177717781779178017811782178317841785178617871788178917901791179217931794179517961797179817991800180118021803180418051806180718081809181018111812181318141815181618171818181918201821182218231824182518261827182818291830183118321833183418351836183718381839184018411842184318441845184618471848184918501851185218531854185518561857185818591860186118621863186418651866186718681869187018711872187318741875187618771878187918801881188218831884188518861887188818891890189118921893189418951896189718981899190019011902190319041905190619071908190919101911191219131914191519161917191819191920192119221923192419251926192719281929193019311932193319341935193619371938193919401941194219431944194519461947194819491950195119521953195419551956195719581959196019611962196319641965196619671968196919701971197219731974197519761977197819791980198119821983198419851986198719881989199019911992199319941995199619971998199920002001200220032004200520062007200820092010201120122013201420152016201720182019202020212022202320242025202620272028202920302031203220332034203520362037203820392040204120422043204420452046204720482049205020512052205320542055205620572058205920602061206220632064206520662067206820692070207120722073207420752076207720782079208020812082208320842085208620872088208920902091209220932094209520962097209820992100210121022103210421052106210721082109211021112112211321142115211621172118211921202121212221232124212521262127212821292130213121322133213421352136213721382139214021412142214321442145214621472148214921502151215221532154215521562157215821592160216121622163216421652166216721682169217021712172217321742175217621772178217921802181218221832184218521862187218821892190219121922193219421952196219721982199220022012202220322042205220622072208220922102211221222132214221522162217221822192220222122222223222422252226222722282229223022312232223322342235223622372238223922402241224222432244224522462247224822492250225122522253225422552256225722582259226022612262226322642265226622672268226922702271227222732274227522762277227822792280228122822283228422852286228722882289229022912292229322942295229622972298229923002301230223032304230523062307230823092310231123122313231423152316231723182319232023212322232323242325232623272328232923302331233223332334233523362337233823392340234123422343234423452346234723482349235023512352235323542355235623572358235923602361236223632364236523662367236823692370237123722373237423752376237723782379238023812382238323842385238623872388238923902391239223932394239523962397239823992400240124022403240424052406240724082409241024112412241324142415241624172418241924202421242224232424242524262427242824292430243124322433243424352436243724382439244024412442244324442445244624472448244924502451245224532454245524562457245824592460246124622463246424652466246724682469247024712472247324742475247624772478247924802481248224832484248524862487248824892490249124922493249424952496249724982499250025012502250325042505250625072508250925102511251225132514251525162517251825192520252125222523252425252526252725282529253025312532253325342535253625372538253925402541254225432544254525462547254825492550255125522553255425552556255725582559256025612562256325642565256625672568256925702571257225732574257525762577257825792580258125822583258425852586258725882589259025912592259325942595259625972598259926002601260226032604260526062607260826092610261126122613261426152616261726182619262026212622262326242625262626272628262926302631263226332634263526362637263826392640264126422643264426452646264726482649265026512652265326542655265626572658265926602661266226632664266526662667266826692670267126722673267426752676267726782679268026812682268326842685268626872688268926902691269226932694269526962697269826992700270127022703270427052706270727082709271027112712271327142715271627172718271927202721272227232724272527262727272827292730273127322733273427352736273727382739274027412742274327442745274627472748274927502751275227532754275527562757275827592760276127622763276427652766276727682769277027712772277327742775277627772778277927802781278227832784278527862787278827892790279127922793279427952796279727982799280028012802280328042805280628072808280928102811281228132814281528162817281828192820282128222823282428252826282728282829283028312832283328342835283628372838283928402841284228432844284528462847284828492850285128522853285428552856285728582859286028612862286328642865286628672868286928702871287228732874287528762877287828792880288128822883288428852886288728882889289028912892289328942895289628972898289929002901290229032904290529062907290829092910291129122913291429152916291729182919292029212922292329242925292629272928292929302931293229332934293529362937293829392940294129422943294429452946294729482949295029512952295329542955295629572958295929602961296229632964296529662967296829692970297129722973297429752976297729782979298029812982298329842985298629872988298929902991299229932994299529962997299829993000300130023003300430053006300730083009301030113012301330143015301630173018301930203021302230233024302530263027302830293030303130323033303430353036303730383039304030413042304330443045304630473048304930503051305230533054305530563057305830593060306130623063306430653066306730683069307030713072307330743075307630773078307930803081308230833084308530863087308830893090309130923093309430953096309730983099310031013102310331043105310631073108310931103111311231133114311531163117311831193120312131223123312431253126312731283129313031313132313331343135313631373138313931403141314231433144314531463147314831493150315131523153315431553156315731583159316031613162316331643165316631673168316931703171317231733174317531763177317831793180318131823183318431853186318731883189319031913192319331943195319631973198319932003201320232033204320532063207320832093210321132123213321432153216321732183219322032213222322332243225322632273228322932303231323232333234323532363237323832393240324132423243324432453246324732483249325032513252325332543255325632573258325932603261326232633264326532663267326832693270327132723273327432753276327732783279328032813282328332843285328632873288328932903291329232933294329532963297329832993300330133023303330433053306330733083309331033113312331333143315331633173318331933203321332233233324332533263327332833293330333133323333333433353336333733383339334033413342334333443345334633473348334933503351335233533354335533563357335833593360336133623363336433653366336733683369337033713372337333743375
  1. /*
  2. * linux/fs/ext4/inode.c
  3. *
  4. * Copyright (C) 1992, 1993, 1994, 1995
  5. * Remy Card (card@masi.ibp.fr)
  6. * Laboratoire MASI - Institut Blaise Pascal
  7. * Universite Pierre et Marie Curie (Paris VI)
  8. *
  9. * from
  10. *
  11. * linux/fs/minix/inode.c
  12. *
  13. * Copyright (C) 1991, 1992 Linus Torvalds
  14. *
  15. * Goal-directed block allocation by Stephen Tweedie
  16. * (sct@redhat.com), 1993, 1998
  17. * Big-endian to little-endian byte-swapping/bitmaps by
  18. * David S. Miller (davem@caip.rutgers.edu), 1995
  19. * 64-bit file support on 64-bit platforms by Jakub Jelinek
  20. * (jj@sunsite.ms.mff.cuni.cz)
  21. *
  22. * Assorted race fixes, rewrite of ext4_get_block() by Al Viro, 2000
  23. */
  24. #include <linux/module.h>
  25. #include <linux/fs.h>
  26. #include <linux/time.h>
  27. #include <linux/ext4_jbd2.h>
  28. #include <linux/jbd2.h>
  29. #include <linux/highuid.h>
  30. #include <linux/pagemap.h>
  31. #include <linux/quotaops.h>
  32. #include <linux/string.h>
  33. #include <linux/buffer_head.h>
  34. #include <linux/writeback.h>
  35. #include <linux/mpage.h>
  36. #include <linux/uio.h>
  37. #include <linux/bio.h>
  38. #include "xattr.h"
  39. #include "acl.h"
  40. /*
  41. * Test whether an inode is a fast symlink.
  42. */
  43. static int ext4_inode_is_fast_symlink(struct inode *inode)
  44. {
  45. int ea_blocks = EXT4_I(inode)->i_file_acl ?
  46. (inode->i_sb->s_blocksize >> 9) : 0;
  47. return (S_ISLNK(inode->i_mode) && inode->i_blocks - ea_blocks == 0);
  48. }
  49. /*
  50. * The ext4 forget function must perform a revoke if we are freeing data
  51. * which has been journaled. Metadata (eg. indirect blocks) must be
  52. * revoked in all cases.
  53. *
  54. * "bh" may be NULL: a metadata block may have been freed from memory
  55. * but there may still be a record of it in the journal, and that record
  56. * still needs to be revoked.
  57. */
  58. int ext4_forget(handle_t *handle, int is_metadata, struct inode *inode,
  59. struct buffer_head *bh, ext4_fsblk_t blocknr)
  60. {
  61. int err;
  62. might_sleep();
  63. BUFFER_TRACE(bh, "enter");
  64. jbd_debug(4, "forgetting bh %p: is_metadata = %d, mode %o, "
  65. "data mode %lx\n",
  66. bh, is_metadata, inode->i_mode,
  67. test_opt(inode->i_sb, DATA_FLAGS));
  68. /* Never use the revoke function if we are doing full data
  69. * journaling: there is no need to, and a V1 superblock won't
  70. * support it. Otherwise, only skip the revoke on un-journaled
  71. * data blocks. */
  72. if (test_opt(inode->i_sb, DATA_FLAGS) == EXT4_MOUNT_JOURNAL_DATA ||
  73. (!is_metadata && !ext4_should_journal_data(inode))) {
  74. if (bh) {
  75. BUFFER_TRACE(bh, "call jbd2_journal_forget");
  76. return ext4_journal_forget(handle, bh);
  77. }
  78. return 0;
  79. }
  80. /*
  81. * data!=journal && (is_metadata || should_journal_data(inode))
  82. */
  83. BUFFER_TRACE(bh, "call ext4_journal_revoke");
  84. err = ext4_journal_revoke(handle, blocknr, bh);
  85. if (err)
  86. ext4_abort(inode->i_sb, __FUNCTION__,
  87. "error %d when attempting revoke", err);
  88. BUFFER_TRACE(bh, "exit");
  89. return err;
  90. }
  91. /*
  92. * Work out how many blocks we need to proceed with the next chunk of a
  93. * truncate transaction.
  94. */
  95. static unsigned long blocks_for_truncate(struct inode *inode)
  96. {
  97. ext4_lblk_t needed;
  98. needed = inode->i_blocks >> (inode->i_sb->s_blocksize_bits - 9);
  99. /* Give ourselves just enough room to cope with inodes in which
  100. * i_blocks is corrupt: we've seen disk corruptions in the past
  101. * which resulted in random data in an inode which looked enough
  102. * like a regular file for ext4 to try to delete it. Things
  103. * will go a bit crazy if that happens, but at least we should
  104. * try not to panic the whole kernel. */
  105. if (needed < 2)
  106. needed = 2;
  107. /* But we need to bound the transaction so we don't overflow the
  108. * journal. */
  109. if (needed > EXT4_MAX_TRANS_DATA)
  110. needed = EXT4_MAX_TRANS_DATA;
  111. return EXT4_DATA_TRANS_BLOCKS(inode->i_sb) + needed;
  112. }
  113. /*
  114. * Truncate transactions can be complex and absolutely huge. So we need to
  115. * be able to restart the transaction at a conventient checkpoint to make
  116. * sure we don't overflow the journal.
  117. *
  118. * start_transaction gets us a new handle for a truncate transaction,
  119. * and extend_transaction tries to extend the existing one a bit. If
  120. * extend fails, we need to propagate the failure up and restart the
  121. * transaction in the top-level truncate loop. --sct
  122. */
  123. static handle_t *start_transaction(struct inode *inode)
  124. {
  125. handle_t *result;
  126. result = ext4_journal_start(inode, blocks_for_truncate(inode));
  127. if (!IS_ERR(result))
  128. return result;
  129. ext4_std_error(inode->i_sb, PTR_ERR(result));
  130. return result;
  131. }
  132. /*
  133. * Try to extend this transaction for the purposes of truncation.
  134. *
  135. * Returns 0 if we managed to create more room. If we can't create more
  136. * room, and the transaction must be restarted we return 1.
  137. */
  138. static int try_to_extend_transaction(handle_t *handle, struct inode *inode)
  139. {
  140. if (handle->h_buffer_credits > EXT4_RESERVE_TRANS_BLOCKS)
  141. return 0;
  142. if (!ext4_journal_extend(handle, blocks_for_truncate(inode)))
  143. return 0;
  144. return 1;
  145. }
  146. /*
  147. * Restart the transaction associated with *handle. This does a commit,
  148. * so before we call here everything must be consistently dirtied against
  149. * this transaction.
  150. */
  151. static int ext4_journal_test_restart(handle_t *handle, struct inode *inode)
  152. {
  153. jbd_debug(2, "restarting handle %p\n", handle);
  154. return ext4_journal_restart(handle, blocks_for_truncate(inode));
  155. }
  156. /*
  157. * Called at the last iput() if i_nlink is zero.
  158. */
  159. void ext4_delete_inode (struct inode * inode)
  160. {
  161. handle_t *handle;
  162. truncate_inode_pages(&inode->i_data, 0);
  163. if (is_bad_inode(inode))
  164. goto no_delete;
  165. handle = start_transaction(inode);
  166. if (IS_ERR(handle)) {
  167. /*
  168. * If we're going to skip the normal cleanup, we still need to
  169. * make sure that the in-core orphan linked list is properly
  170. * cleaned up.
  171. */
  172. ext4_orphan_del(NULL, inode);
  173. goto no_delete;
  174. }
  175. if (IS_SYNC(inode))
  176. handle->h_sync = 1;
  177. inode->i_size = 0;
  178. if (inode->i_blocks)
  179. ext4_truncate(inode);
  180. /*
  181. * Kill off the orphan record which ext4_truncate created.
  182. * AKPM: I think this can be inside the above `if'.
  183. * Note that ext4_orphan_del() has to be able to cope with the
  184. * deletion of a non-existent orphan - this is because we don't
  185. * know if ext4_truncate() actually created an orphan record.
  186. * (Well, we could do this if we need to, but heck - it works)
  187. */
  188. ext4_orphan_del(handle, inode);
  189. EXT4_I(inode)->i_dtime = get_seconds();
  190. /*
  191. * One subtle ordering requirement: if anything has gone wrong
  192. * (transaction abort, IO errors, whatever), then we can still
  193. * do these next steps (the fs will already have been marked as
  194. * having errors), but we can't free the inode if the mark_dirty
  195. * fails.
  196. */
  197. if (ext4_mark_inode_dirty(handle, inode))
  198. /* If that failed, just do the required in-core inode clear. */
  199. clear_inode(inode);
  200. else
  201. ext4_free_inode(handle, inode);
  202. ext4_journal_stop(handle);
  203. return;
  204. no_delete:
  205. clear_inode(inode); /* We must guarantee clearing of inode... */
  206. }
  207. typedef struct {
  208. __le32 *p;
  209. __le32 key;
  210. struct buffer_head *bh;
  211. } Indirect;
  212. static inline void add_chain(Indirect *p, struct buffer_head *bh, __le32 *v)
  213. {
  214. p->key = *(p->p = v);
  215. p->bh = bh;
  216. }
  217. static int verify_chain(Indirect *from, Indirect *to)
  218. {
  219. while (from <= to && from->key == *from->p)
  220. from++;
  221. return (from > to);
  222. }
  223. /**
  224. * ext4_block_to_path - parse the block number into array of offsets
  225. * @inode: inode in question (we are only interested in its superblock)
  226. * @i_block: block number to be parsed
  227. * @offsets: array to store the offsets in
  228. * @boundary: set this non-zero if the referred-to block is likely to be
  229. * followed (on disk) by an indirect block.
  230. *
  231. * To store the locations of file's data ext4 uses a data structure common
  232. * for UNIX filesystems - tree of pointers anchored in the inode, with
  233. * data blocks at leaves and indirect blocks in intermediate nodes.
  234. * This function translates the block number into path in that tree -
  235. * return value is the path length and @offsets[n] is the offset of
  236. * pointer to (n+1)th node in the nth one. If @block is out of range
  237. * (negative or too large) warning is printed and zero returned.
  238. *
  239. * Note: function doesn't find node addresses, so no IO is needed. All
  240. * we need to know is the capacity of indirect blocks (taken from the
  241. * inode->i_sb).
  242. */
  243. /*
  244. * Portability note: the last comparison (check that we fit into triple
  245. * indirect block) is spelled differently, because otherwise on an
  246. * architecture with 32-bit longs and 8Kb pages we might get into trouble
  247. * if our filesystem had 8Kb blocks. We might use long long, but that would
  248. * kill us on x86. Oh, well, at least the sign propagation does not matter -
  249. * i_block would have to be negative in the very beginning, so we would not
  250. * get there at all.
  251. */
  252. static int ext4_block_to_path(struct inode *inode,
  253. ext4_lblk_t i_block,
  254. ext4_lblk_t offsets[4], int *boundary)
  255. {
  256. int ptrs = EXT4_ADDR_PER_BLOCK(inode->i_sb);
  257. int ptrs_bits = EXT4_ADDR_PER_BLOCK_BITS(inode->i_sb);
  258. const long direct_blocks = EXT4_NDIR_BLOCKS,
  259. indirect_blocks = ptrs,
  260. double_blocks = (1 << (ptrs_bits * 2));
  261. int n = 0;
  262. int final = 0;
  263. if (i_block < 0) {
  264. ext4_warning (inode->i_sb, "ext4_block_to_path", "block < 0");
  265. } else if (i_block < direct_blocks) {
  266. offsets[n++] = i_block;
  267. final = direct_blocks;
  268. } else if ( (i_block -= direct_blocks) < indirect_blocks) {
  269. offsets[n++] = EXT4_IND_BLOCK;
  270. offsets[n++] = i_block;
  271. final = ptrs;
  272. } else if ((i_block -= indirect_blocks) < double_blocks) {
  273. offsets[n++] = EXT4_DIND_BLOCK;
  274. offsets[n++] = i_block >> ptrs_bits;
  275. offsets[n++] = i_block & (ptrs - 1);
  276. final = ptrs;
  277. } else if (((i_block -= double_blocks) >> (ptrs_bits * 2)) < ptrs) {
  278. offsets[n++] = EXT4_TIND_BLOCK;
  279. offsets[n++] = i_block >> (ptrs_bits * 2);
  280. offsets[n++] = (i_block >> ptrs_bits) & (ptrs - 1);
  281. offsets[n++] = i_block & (ptrs - 1);
  282. final = ptrs;
  283. } else {
  284. ext4_warning(inode->i_sb, "ext4_block_to_path", "block > big");
  285. }
  286. if (boundary)
  287. *boundary = final - 1 - (i_block & (ptrs - 1));
  288. return n;
  289. }
  290. /**
  291. * ext4_get_branch - read the chain of indirect blocks leading to data
  292. * @inode: inode in question
  293. * @depth: depth of the chain (1 - direct pointer, etc.)
  294. * @offsets: offsets of pointers in inode/indirect blocks
  295. * @chain: place to store the result
  296. * @err: here we store the error value
  297. *
  298. * Function fills the array of triples <key, p, bh> and returns %NULL
  299. * if everything went OK or the pointer to the last filled triple
  300. * (incomplete one) otherwise. Upon the return chain[i].key contains
  301. * the number of (i+1)-th block in the chain (as it is stored in memory,
  302. * i.e. little-endian 32-bit), chain[i].p contains the address of that
  303. * number (it points into struct inode for i==0 and into the bh->b_data
  304. * for i>0) and chain[i].bh points to the buffer_head of i-th indirect
  305. * block for i>0 and NULL for i==0. In other words, it holds the block
  306. * numbers of the chain, addresses they were taken from (and where we can
  307. * verify that chain did not change) and buffer_heads hosting these
  308. * numbers.
  309. *
  310. * Function stops when it stumbles upon zero pointer (absent block)
  311. * (pointer to last triple returned, *@err == 0)
  312. * or when it gets an IO error reading an indirect block
  313. * (ditto, *@err == -EIO)
  314. * or when it notices that chain had been changed while it was reading
  315. * (ditto, *@err == -EAGAIN)
  316. * or when it reads all @depth-1 indirect blocks successfully and finds
  317. * the whole chain, all way to the data (returns %NULL, *err == 0).
  318. */
  319. static Indirect *ext4_get_branch(struct inode *inode, int depth,
  320. ext4_lblk_t *offsets,
  321. Indirect chain[4], int *err)
  322. {
  323. struct super_block *sb = inode->i_sb;
  324. Indirect *p = chain;
  325. struct buffer_head *bh;
  326. *err = 0;
  327. /* i_data is not going away, no lock needed */
  328. add_chain (chain, NULL, EXT4_I(inode)->i_data + *offsets);
  329. if (!p->key)
  330. goto no_block;
  331. while (--depth) {
  332. bh = sb_bread(sb, le32_to_cpu(p->key));
  333. if (!bh)
  334. goto failure;
  335. /* Reader: pointers */
  336. if (!verify_chain(chain, p))
  337. goto changed;
  338. add_chain(++p, bh, (__le32*)bh->b_data + *++offsets);
  339. /* Reader: end */
  340. if (!p->key)
  341. goto no_block;
  342. }
  343. return NULL;
  344. changed:
  345. brelse(bh);
  346. *err = -EAGAIN;
  347. goto no_block;
  348. failure:
  349. *err = -EIO;
  350. no_block:
  351. return p;
  352. }
  353. /**
  354. * ext4_find_near - find a place for allocation with sufficient locality
  355. * @inode: owner
  356. * @ind: descriptor of indirect block.
  357. *
  358. * This function returns the prefered place for block allocation.
  359. * It is used when heuristic for sequential allocation fails.
  360. * Rules are:
  361. * + if there is a block to the left of our position - allocate near it.
  362. * + if pointer will live in indirect block - allocate near that block.
  363. * + if pointer will live in inode - allocate in the same
  364. * cylinder group.
  365. *
  366. * In the latter case we colour the starting block by the callers PID to
  367. * prevent it from clashing with concurrent allocations for a different inode
  368. * in the same block group. The PID is used here so that functionally related
  369. * files will be close-by on-disk.
  370. *
  371. * Caller must make sure that @ind is valid and will stay that way.
  372. */
  373. static ext4_fsblk_t ext4_find_near(struct inode *inode, Indirect *ind)
  374. {
  375. struct ext4_inode_info *ei = EXT4_I(inode);
  376. __le32 *start = ind->bh ? (__le32*) ind->bh->b_data : ei->i_data;
  377. __le32 *p;
  378. ext4_fsblk_t bg_start;
  379. ext4_grpblk_t colour;
  380. /* Try to find previous block */
  381. for (p = ind->p - 1; p >= start; p--) {
  382. if (*p)
  383. return le32_to_cpu(*p);
  384. }
  385. /* No such thing, so let's try location of indirect block */
  386. if (ind->bh)
  387. return ind->bh->b_blocknr;
  388. /*
  389. * It is going to be referred to from the inode itself? OK, just put it
  390. * into the same cylinder group then.
  391. */
  392. bg_start = ext4_group_first_block_no(inode->i_sb, ei->i_block_group);
  393. colour = (current->pid % 16) *
  394. (EXT4_BLOCKS_PER_GROUP(inode->i_sb) / 16);
  395. return bg_start + colour;
  396. }
  397. /**
  398. * ext4_find_goal - find a prefered place for allocation.
  399. * @inode: owner
  400. * @block: block we want
  401. * @chain: chain of indirect blocks
  402. * @partial: pointer to the last triple within a chain
  403. * @goal: place to store the result.
  404. *
  405. * Normally this function find the prefered place for block allocation,
  406. * stores it in *@goal and returns zero.
  407. */
  408. static ext4_fsblk_t ext4_find_goal(struct inode *inode, ext4_lblk_t block,
  409. Indirect chain[4], Indirect *partial)
  410. {
  411. struct ext4_block_alloc_info *block_i;
  412. block_i = EXT4_I(inode)->i_block_alloc_info;
  413. /*
  414. * try the heuristic for sequential allocation,
  415. * failing that at least try to get decent locality.
  416. */
  417. if (block_i && (block == block_i->last_alloc_logical_block + 1)
  418. && (block_i->last_alloc_physical_block != 0)) {
  419. return block_i->last_alloc_physical_block + 1;
  420. }
  421. return ext4_find_near(inode, partial);
  422. }
  423. /**
  424. * ext4_blks_to_allocate: Look up the block map and count the number
  425. * of direct blocks need to be allocated for the given branch.
  426. *
  427. * @branch: chain of indirect blocks
  428. * @k: number of blocks need for indirect blocks
  429. * @blks: number of data blocks to be mapped.
  430. * @blocks_to_boundary: the offset in the indirect block
  431. *
  432. * return the total number of blocks to be allocate, including the
  433. * direct and indirect blocks.
  434. */
  435. static int ext4_blks_to_allocate(Indirect *branch, int k, unsigned long blks,
  436. int blocks_to_boundary)
  437. {
  438. unsigned long count = 0;
  439. /*
  440. * Simple case, [t,d]Indirect block(s) has not allocated yet
  441. * then it's clear blocks on that path have not allocated
  442. */
  443. if (k > 0) {
  444. /* right now we don't handle cross boundary allocation */
  445. if (blks < blocks_to_boundary + 1)
  446. count += blks;
  447. else
  448. count += blocks_to_boundary + 1;
  449. return count;
  450. }
  451. count++;
  452. while (count < blks && count <= blocks_to_boundary &&
  453. le32_to_cpu(*(branch[0].p + count)) == 0) {
  454. count++;
  455. }
  456. return count;
  457. }
  458. /**
  459. * ext4_alloc_blocks: multiple allocate blocks needed for a branch
  460. * @indirect_blks: the number of blocks need to allocate for indirect
  461. * blocks
  462. *
  463. * @new_blocks: on return it will store the new block numbers for
  464. * the indirect blocks(if needed) and the first direct block,
  465. * @blks: on return it will store the total number of allocated
  466. * direct blocks
  467. */
  468. static int ext4_alloc_blocks(handle_t *handle, struct inode *inode,
  469. ext4_fsblk_t goal, int indirect_blks, int blks,
  470. ext4_fsblk_t new_blocks[4], int *err)
  471. {
  472. int target, i;
  473. unsigned long count = 0;
  474. int index = 0;
  475. ext4_fsblk_t current_block = 0;
  476. int ret = 0;
  477. /*
  478. * Here we try to allocate the requested multiple blocks at once,
  479. * on a best-effort basis.
  480. * To build a branch, we should allocate blocks for
  481. * the indirect blocks(if not allocated yet), and at least
  482. * the first direct block of this branch. That's the
  483. * minimum number of blocks need to allocate(required)
  484. */
  485. target = blks + indirect_blks;
  486. while (1) {
  487. count = target;
  488. /* allocating blocks for indirect blocks and direct blocks */
  489. current_block = ext4_new_blocks(handle,inode,goal,&count,err);
  490. if (*err)
  491. goto failed_out;
  492. target -= count;
  493. /* allocate blocks for indirect blocks */
  494. while (index < indirect_blks && count) {
  495. new_blocks[index++] = current_block++;
  496. count--;
  497. }
  498. if (count > 0)
  499. break;
  500. }
  501. /* save the new block number for the first direct block */
  502. new_blocks[index] = current_block;
  503. /* total number of blocks allocated for direct blocks */
  504. ret = count;
  505. *err = 0;
  506. return ret;
  507. failed_out:
  508. for (i = 0; i <index; i++)
  509. ext4_free_blocks(handle, inode, new_blocks[i], 1);
  510. return ret;
  511. }
  512. /**
  513. * ext4_alloc_branch - allocate and set up a chain of blocks.
  514. * @inode: owner
  515. * @indirect_blks: number of allocated indirect blocks
  516. * @blks: number of allocated direct blocks
  517. * @offsets: offsets (in the blocks) to store the pointers to next.
  518. * @branch: place to store the chain in.
  519. *
  520. * This function allocates blocks, zeroes out all but the last one,
  521. * links them into chain and (if we are synchronous) writes them to disk.
  522. * In other words, it prepares a branch that can be spliced onto the
  523. * inode. It stores the information about that chain in the branch[], in
  524. * the same format as ext4_get_branch() would do. We are calling it after
  525. * we had read the existing part of chain and partial points to the last
  526. * triple of that (one with zero ->key). Upon the exit we have the same
  527. * picture as after the successful ext4_get_block(), except that in one
  528. * place chain is disconnected - *branch->p is still zero (we did not
  529. * set the last link), but branch->key contains the number that should
  530. * be placed into *branch->p to fill that gap.
  531. *
  532. * If allocation fails we free all blocks we've allocated (and forget
  533. * their buffer_heads) and return the error value the from failed
  534. * ext4_alloc_block() (normally -ENOSPC). Otherwise we set the chain
  535. * as described above and return 0.
  536. */
  537. static int ext4_alloc_branch(handle_t *handle, struct inode *inode,
  538. int indirect_blks, int *blks, ext4_fsblk_t goal,
  539. ext4_lblk_t *offsets, Indirect *branch)
  540. {
  541. int blocksize = inode->i_sb->s_blocksize;
  542. int i, n = 0;
  543. int err = 0;
  544. struct buffer_head *bh;
  545. int num;
  546. ext4_fsblk_t new_blocks[4];
  547. ext4_fsblk_t current_block;
  548. num = ext4_alloc_blocks(handle, inode, goal, indirect_blks,
  549. *blks, new_blocks, &err);
  550. if (err)
  551. return err;
  552. branch[0].key = cpu_to_le32(new_blocks[0]);
  553. /*
  554. * metadata blocks and data blocks are allocated.
  555. */
  556. for (n = 1; n <= indirect_blks; n++) {
  557. /*
  558. * Get buffer_head for parent block, zero it out
  559. * and set the pointer to new one, then send
  560. * parent to disk.
  561. */
  562. bh = sb_getblk(inode->i_sb, new_blocks[n-1]);
  563. branch[n].bh = bh;
  564. lock_buffer(bh);
  565. BUFFER_TRACE(bh, "call get_create_access");
  566. err = ext4_journal_get_create_access(handle, bh);
  567. if (err) {
  568. unlock_buffer(bh);
  569. brelse(bh);
  570. goto failed;
  571. }
  572. memset(bh->b_data, 0, blocksize);
  573. branch[n].p = (__le32 *) bh->b_data + offsets[n];
  574. branch[n].key = cpu_to_le32(new_blocks[n]);
  575. *branch[n].p = branch[n].key;
  576. if ( n == indirect_blks) {
  577. current_block = new_blocks[n];
  578. /*
  579. * End of chain, update the last new metablock of
  580. * the chain to point to the new allocated
  581. * data blocks numbers
  582. */
  583. for (i=1; i < num; i++)
  584. *(branch[n].p + i) = cpu_to_le32(++current_block);
  585. }
  586. BUFFER_TRACE(bh, "marking uptodate");
  587. set_buffer_uptodate(bh);
  588. unlock_buffer(bh);
  589. BUFFER_TRACE(bh, "call ext4_journal_dirty_metadata");
  590. err = ext4_journal_dirty_metadata(handle, bh);
  591. if (err)
  592. goto failed;
  593. }
  594. *blks = num;
  595. return err;
  596. failed:
  597. /* Allocation failed, free what we already allocated */
  598. for (i = 1; i <= n ; i++) {
  599. BUFFER_TRACE(branch[i].bh, "call jbd2_journal_forget");
  600. ext4_journal_forget(handle, branch[i].bh);
  601. }
  602. for (i = 0; i <indirect_blks; i++)
  603. ext4_free_blocks(handle, inode, new_blocks[i], 1);
  604. ext4_free_blocks(handle, inode, new_blocks[i], num);
  605. return err;
  606. }
  607. /**
  608. * ext4_splice_branch - splice the allocated branch onto inode.
  609. * @inode: owner
  610. * @block: (logical) number of block we are adding
  611. * @chain: chain of indirect blocks (with a missing link - see
  612. * ext4_alloc_branch)
  613. * @where: location of missing link
  614. * @num: number of indirect blocks we are adding
  615. * @blks: number of direct blocks we are adding
  616. *
  617. * This function fills the missing link and does all housekeeping needed in
  618. * inode (->i_blocks, etc.). In case of success we end up with the full
  619. * chain to new block and return 0.
  620. */
  621. static int ext4_splice_branch(handle_t *handle, struct inode *inode,
  622. ext4_lblk_t block, Indirect *where, int num, int blks)
  623. {
  624. int i;
  625. int err = 0;
  626. struct ext4_block_alloc_info *block_i;
  627. ext4_fsblk_t current_block;
  628. block_i = EXT4_I(inode)->i_block_alloc_info;
  629. /*
  630. * If we're splicing into a [td]indirect block (as opposed to the
  631. * inode) then we need to get write access to the [td]indirect block
  632. * before the splice.
  633. */
  634. if (where->bh) {
  635. BUFFER_TRACE(where->bh, "get_write_access");
  636. err = ext4_journal_get_write_access(handle, where->bh);
  637. if (err)
  638. goto err_out;
  639. }
  640. /* That's it */
  641. *where->p = where->key;
  642. /*
  643. * Update the host buffer_head or inode to point to more just allocated
  644. * direct blocks blocks
  645. */
  646. if (num == 0 && blks > 1) {
  647. current_block = le32_to_cpu(where->key) + 1;
  648. for (i = 1; i < blks; i++)
  649. *(where->p + i ) = cpu_to_le32(current_block++);
  650. }
  651. /*
  652. * update the most recently allocated logical & physical block
  653. * in i_block_alloc_info, to assist find the proper goal block for next
  654. * allocation
  655. */
  656. if (block_i) {
  657. block_i->last_alloc_logical_block = block + blks - 1;
  658. block_i->last_alloc_physical_block =
  659. le32_to_cpu(where[num].key) + blks - 1;
  660. }
  661. /* We are done with atomic stuff, now do the rest of housekeeping */
  662. inode->i_ctime = ext4_current_time(inode);
  663. ext4_mark_inode_dirty(handle, inode);
  664. /* had we spliced it onto indirect block? */
  665. if (where->bh) {
  666. /*
  667. * If we spliced it onto an indirect block, we haven't
  668. * altered the inode. Note however that if it is being spliced
  669. * onto an indirect block at the very end of the file (the
  670. * file is growing) then we *will* alter the inode to reflect
  671. * the new i_size. But that is not done here - it is done in
  672. * generic_commit_write->__mark_inode_dirty->ext4_dirty_inode.
  673. */
  674. jbd_debug(5, "splicing indirect only\n");
  675. BUFFER_TRACE(where->bh, "call ext4_journal_dirty_metadata");
  676. err = ext4_journal_dirty_metadata(handle, where->bh);
  677. if (err)
  678. goto err_out;
  679. } else {
  680. /*
  681. * OK, we spliced it into the inode itself on a direct block.
  682. * Inode was dirtied above.
  683. */
  684. jbd_debug(5, "splicing direct\n");
  685. }
  686. return err;
  687. err_out:
  688. for (i = 1; i <= num; i++) {
  689. BUFFER_TRACE(where[i].bh, "call jbd2_journal_forget");
  690. ext4_journal_forget(handle, where[i].bh);
  691. ext4_free_blocks(handle,inode,le32_to_cpu(where[i-1].key),1);
  692. }
  693. ext4_free_blocks(handle, inode, le32_to_cpu(where[num].key), blks);
  694. return err;
  695. }
  696. /*
  697. * Allocation strategy is simple: if we have to allocate something, we will
  698. * have to go the whole way to leaf. So let's do it before attaching anything
  699. * to tree, set linkage between the newborn blocks, write them if sync is
  700. * required, recheck the path, free and repeat if check fails, otherwise
  701. * set the last missing link (that will protect us from any truncate-generated
  702. * removals - all blocks on the path are immune now) and possibly force the
  703. * write on the parent block.
  704. * That has a nice additional property: no special recovery from the failed
  705. * allocations is needed - we simply release blocks and do not touch anything
  706. * reachable from inode.
  707. *
  708. * `handle' can be NULL if create == 0.
  709. *
  710. * The BKL may not be held on entry here. Be sure to take it early.
  711. * return > 0, # of blocks mapped or allocated.
  712. * return = 0, if plain lookup failed.
  713. * return < 0, error case.
  714. */
  715. int ext4_get_blocks_handle(handle_t *handle, struct inode *inode,
  716. ext4_lblk_t iblock, unsigned long maxblocks,
  717. struct buffer_head *bh_result,
  718. int create, int extend_disksize)
  719. {
  720. int err = -EIO;
  721. ext4_lblk_t offsets[4];
  722. Indirect chain[4];
  723. Indirect *partial;
  724. ext4_fsblk_t goal;
  725. int indirect_blks;
  726. int blocks_to_boundary = 0;
  727. int depth;
  728. struct ext4_inode_info *ei = EXT4_I(inode);
  729. int count = 0;
  730. ext4_fsblk_t first_block = 0;
  731. J_ASSERT(!(EXT4_I(inode)->i_flags & EXT4_EXTENTS_FL));
  732. J_ASSERT(handle != NULL || create == 0);
  733. depth = ext4_block_to_path(inode, iblock, offsets,
  734. &blocks_to_boundary);
  735. if (depth == 0)
  736. goto out;
  737. partial = ext4_get_branch(inode, depth, offsets, chain, &err);
  738. /* Simplest case - block found, no allocation needed */
  739. if (!partial) {
  740. first_block = le32_to_cpu(chain[depth - 1].key);
  741. clear_buffer_new(bh_result);
  742. count++;
  743. /*map more blocks*/
  744. while (count < maxblocks && count <= blocks_to_boundary) {
  745. ext4_fsblk_t blk;
  746. if (!verify_chain(chain, partial)) {
  747. /*
  748. * Indirect block might be removed by
  749. * truncate while we were reading it.
  750. * Handling of that case: forget what we've
  751. * got now. Flag the err as EAGAIN, so it
  752. * will reread.
  753. */
  754. err = -EAGAIN;
  755. count = 0;
  756. break;
  757. }
  758. blk = le32_to_cpu(*(chain[depth-1].p + count));
  759. if (blk == first_block + count)
  760. count++;
  761. else
  762. break;
  763. }
  764. if (err != -EAGAIN)
  765. goto got_it;
  766. }
  767. /* Next simple case - plain lookup or failed read of indirect block */
  768. if (!create || err == -EIO)
  769. goto cleanup;
  770. mutex_lock(&ei->truncate_mutex);
  771. /*
  772. * If the indirect block is missing while we are reading
  773. * the chain(ext4_get_branch() returns -EAGAIN err), or
  774. * if the chain has been changed after we grab the semaphore,
  775. * (either because another process truncated this branch, or
  776. * another get_block allocated this branch) re-grab the chain to see if
  777. * the request block has been allocated or not.
  778. *
  779. * Since we already block the truncate/other get_block
  780. * at this point, we will have the current copy of the chain when we
  781. * splice the branch into the tree.
  782. */
  783. if (err == -EAGAIN || !verify_chain(chain, partial)) {
  784. while (partial > chain) {
  785. brelse(partial->bh);
  786. partial--;
  787. }
  788. partial = ext4_get_branch(inode, depth, offsets, chain, &err);
  789. if (!partial) {
  790. count++;
  791. mutex_unlock(&ei->truncate_mutex);
  792. if (err)
  793. goto cleanup;
  794. clear_buffer_new(bh_result);
  795. goto got_it;
  796. }
  797. }
  798. /*
  799. * Okay, we need to do block allocation. Lazily initialize the block
  800. * allocation info here if necessary
  801. */
  802. if (S_ISREG(inode->i_mode) && (!ei->i_block_alloc_info))
  803. ext4_init_block_alloc_info(inode);
  804. goal = ext4_find_goal(inode, iblock, chain, partial);
  805. /* the number of blocks need to allocate for [d,t]indirect blocks */
  806. indirect_blks = (chain + depth) - partial - 1;
  807. /*
  808. * Next look up the indirect map to count the totoal number of
  809. * direct blocks to allocate for this branch.
  810. */
  811. count = ext4_blks_to_allocate(partial, indirect_blks,
  812. maxblocks, blocks_to_boundary);
  813. /*
  814. * Block out ext4_truncate while we alter the tree
  815. */
  816. err = ext4_alloc_branch(handle, inode, indirect_blks, &count, goal,
  817. offsets + (partial - chain), partial);
  818. /*
  819. * The ext4_splice_branch call will free and forget any buffers
  820. * on the new chain if there is a failure, but that risks using
  821. * up transaction credits, especially for bitmaps where the
  822. * credits cannot be returned. Can we handle this somehow? We
  823. * may need to return -EAGAIN upwards in the worst case. --sct
  824. */
  825. if (!err)
  826. err = ext4_splice_branch(handle, inode, iblock,
  827. partial, indirect_blks, count);
  828. /*
  829. * i_disksize growing is protected by truncate_mutex. Don't forget to
  830. * protect it if you're about to implement concurrent
  831. * ext4_get_block() -bzzz
  832. */
  833. if (!err && extend_disksize && inode->i_size > ei->i_disksize)
  834. ei->i_disksize = inode->i_size;
  835. mutex_unlock(&ei->truncate_mutex);
  836. if (err)
  837. goto cleanup;
  838. set_buffer_new(bh_result);
  839. got_it:
  840. map_bh(bh_result, inode->i_sb, le32_to_cpu(chain[depth-1].key));
  841. if (count > blocks_to_boundary)
  842. set_buffer_boundary(bh_result);
  843. err = count;
  844. /* Clean up and exit */
  845. partial = chain + depth - 1; /* the whole chain */
  846. cleanup:
  847. while (partial > chain) {
  848. BUFFER_TRACE(partial->bh, "call brelse");
  849. brelse(partial->bh);
  850. partial--;
  851. }
  852. BUFFER_TRACE(bh_result, "returned");
  853. out:
  854. return err;
  855. }
  856. #define DIO_CREDITS (EXT4_RESERVE_TRANS_BLOCKS + 32)
  857. static int ext4_get_block(struct inode *inode, sector_t iblock,
  858. struct buffer_head *bh_result, int create)
  859. {
  860. handle_t *handle = ext4_journal_current_handle();
  861. int ret = 0;
  862. unsigned max_blocks = bh_result->b_size >> inode->i_blkbits;
  863. if (!create)
  864. goto get_block; /* A read */
  865. if (max_blocks == 1)
  866. goto get_block; /* A single block get */
  867. if (handle->h_transaction->t_state == T_LOCKED) {
  868. /*
  869. * Huge direct-io writes can hold off commits for long
  870. * periods of time. Let this commit run.
  871. */
  872. ext4_journal_stop(handle);
  873. handle = ext4_journal_start(inode, DIO_CREDITS);
  874. if (IS_ERR(handle))
  875. ret = PTR_ERR(handle);
  876. goto get_block;
  877. }
  878. if (handle->h_buffer_credits <= EXT4_RESERVE_TRANS_BLOCKS) {
  879. /*
  880. * Getting low on buffer credits...
  881. */
  882. ret = ext4_journal_extend(handle, DIO_CREDITS);
  883. if (ret > 0) {
  884. /*
  885. * Couldn't extend the transaction. Start a new one.
  886. */
  887. ret = ext4_journal_restart(handle, DIO_CREDITS);
  888. }
  889. }
  890. get_block:
  891. if (ret == 0) {
  892. ret = ext4_get_blocks_wrap(handle, inode, iblock,
  893. max_blocks, bh_result, create, 0);
  894. if (ret > 0) {
  895. bh_result->b_size = (ret << inode->i_blkbits);
  896. ret = 0;
  897. }
  898. }
  899. return ret;
  900. }
  901. /*
  902. * `handle' can be NULL if create is zero
  903. */
  904. struct buffer_head *ext4_getblk(handle_t *handle, struct inode *inode,
  905. ext4_lblk_t block, int create, int *errp)
  906. {
  907. struct buffer_head dummy;
  908. int fatal = 0, err;
  909. J_ASSERT(handle != NULL || create == 0);
  910. dummy.b_state = 0;
  911. dummy.b_blocknr = -1000;
  912. buffer_trace_init(&dummy.b_history);
  913. err = ext4_get_blocks_wrap(handle, inode, block, 1,
  914. &dummy, create, 1);
  915. /*
  916. * ext4_get_blocks_handle() returns number of blocks
  917. * mapped. 0 in case of a HOLE.
  918. */
  919. if (err > 0) {
  920. if (err > 1)
  921. WARN_ON(1);
  922. err = 0;
  923. }
  924. *errp = err;
  925. if (!err && buffer_mapped(&dummy)) {
  926. struct buffer_head *bh;
  927. bh = sb_getblk(inode->i_sb, dummy.b_blocknr);
  928. if (!bh) {
  929. *errp = -EIO;
  930. goto err;
  931. }
  932. if (buffer_new(&dummy)) {
  933. J_ASSERT(create != 0);
  934. J_ASSERT(handle != NULL);
  935. /*
  936. * Now that we do not always journal data, we should
  937. * keep in mind whether this should always journal the
  938. * new buffer as metadata. For now, regular file
  939. * writes use ext4_get_block instead, so it's not a
  940. * problem.
  941. */
  942. lock_buffer(bh);
  943. BUFFER_TRACE(bh, "call get_create_access");
  944. fatal = ext4_journal_get_create_access(handle, bh);
  945. if (!fatal && !buffer_uptodate(bh)) {
  946. memset(bh->b_data,0,inode->i_sb->s_blocksize);
  947. set_buffer_uptodate(bh);
  948. }
  949. unlock_buffer(bh);
  950. BUFFER_TRACE(bh, "call ext4_journal_dirty_metadata");
  951. err = ext4_journal_dirty_metadata(handle, bh);
  952. if (!fatal)
  953. fatal = err;
  954. } else {
  955. BUFFER_TRACE(bh, "not a new buffer");
  956. }
  957. if (fatal) {
  958. *errp = fatal;
  959. brelse(bh);
  960. bh = NULL;
  961. }
  962. return bh;
  963. }
  964. err:
  965. return NULL;
  966. }
  967. struct buffer_head *ext4_bread(handle_t *handle, struct inode *inode,
  968. ext4_lblk_t block, int create, int *err)
  969. {
  970. struct buffer_head * bh;
  971. bh = ext4_getblk(handle, inode, block, create, err);
  972. if (!bh)
  973. return bh;
  974. if (buffer_uptodate(bh))
  975. return bh;
  976. ll_rw_block(READ_META, 1, &bh);
  977. wait_on_buffer(bh);
  978. if (buffer_uptodate(bh))
  979. return bh;
  980. put_bh(bh);
  981. *err = -EIO;
  982. return NULL;
  983. }
  984. static int walk_page_buffers( handle_t *handle,
  985. struct buffer_head *head,
  986. unsigned from,
  987. unsigned to,
  988. int *partial,
  989. int (*fn)( handle_t *handle,
  990. struct buffer_head *bh))
  991. {
  992. struct buffer_head *bh;
  993. unsigned block_start, block_end;
  994. unsigned blocksize = head->b_size;
  995. int err, ret = 0;
  996. struct buffer_head *next;
  997. for ( bh = head, block_start = 0;
  998. ret == 0 && (bh != head || !block_start);
  999. block_start = block_end, bh = next)
  1000. {
  1001. next = bh->b_this_page;
  1002. block_end = block_start + blocksize;
  1003. if (block_end <= from || block_start >= to) {
  1004. if (partial && !buffer_uptodate(bh))
  1005. *partial = 1;
  1006. continue;
  1007. }
  1008. err = (*fn)(handle, bh);
  1009. if (!ret)
  1010. ret = err;
  1011. }
  1012. return ret;
  1013. }
  1014. /*
  1015. * To preserve ordering, it is essential that the hole instantiation and
  1016. * the data write be encapsulated in a single transaction. We cannot
  1017. * close off a transaction and start a new one between the ext4_get_block()
  1018. * and the commit_write(). So doing the jbd2_journal_start at the start of
  1019. * prepare_write() is the right place.
  1020. *
  1021. * Also, this function can nest inside ext4_writepage() ->
  1022. * block_write_full_page(). In that case, we *know* that ext4_writepage()
  1023. * has generated enough buffer credits to do the whole page. So we won't
  1024. * block on the journal in that case, which is good, because the caller may
  1025. * be PF_MEMALLOC.
  1026. *
  1027. * By accident, ext4 can be reentered when a transaction is open via
  1028. * quota file writes. If we were to commit the transaction while thus
  1029. * reentered, there can be a deadlock - we would be holding a quota
  1030. * lock, and the commit would never complete if another thread had a
  1031. * transaction open and was blocking on the quota lock - a ranking
  1032. * violation.
  1033. *
  1034. * So what we do is to rely on the fact that jbd2_journal_stop/journal_start
  1035. * will _not_ run commit under these circumstances because handle->h_ref
  1036. * is elevated. We'll still have enough credits for the tiny quotafile
  1037. * write.
  1038. */
  1039. static int do_journal_get_write_access(handle_t *handle,
  1040. struct buffer_head *bh)
  1041. {
  1042. if (!buffer_mapped(bh) || buffer_freed(bh))
  1043. return 0;
  1044. return ext4_journal_get_write_access(handle, bh);
  1045. }
  1046. static int ext4_write_begin(struct file *file, struct address_space *mapping,
  1047. loff_t pos, unsigned len, unsigned flags,
  1048. struct page **pagep, void **fsdata)
  1049. {
  1050. struct inode *inode = mapping->host;
  1051. int ret, needed_blocks = ext4_writepage_trans_blocks(inode);
  1052. handle_t *handle;
  1053. int retries = 0;
  1054. struct page *page;
  1055. pgoff_t index;
  1056. unsigned from, to;
  1057. index = pos >> PAGE_CACHE_SHIFT;
  1058. from = pos & (PAGE_CACHE_SIZE - 1);
  1059. to = from + len;
  1060. retry:
  1061. page = __grab_cache_page(mapping, index);
  1062. if (!page)
  1063. return -ENOMEM;
  1064. *pagep = page;
  1065. handle = ext4_journal_start(inode, needed_blocks);
  1066. if (IS_ERR(handle)) {
  1067. unlock_page(page);
  1068. page_cache_release(page);
  1069. ret = PTR_ERR(handle);
  1070. goto out;
  1071. }
  1072. ret = block_write_begin(file, mapping, pos, len, flags, pagep, fsdata,
  1073. ext4_get_block);
  1074. if (!ret && ext4_should_journal_data(inode)) {
  1075. ret = walk_page_buffers(handle, page_buffers(page),
  1076. from, to, NULL, do_journal_get_write_access);
  1077. }
  1078. if (ret) {
  1079. ext4_journal_stop(handle);
  1080. unlock_page(page);
  1081. page_cache_release(page);
  1082. }
  1083. if (ret == -ENOSPC && ext4_should_retry_alloc(inode->i_sb, &retries))
  1084. goto retry;
  1085. out:
  1086. return ret;
  1087. }
  1088. int ext4_journal_dirty_data(handle_t *handle, struct buffer_head *bh)
  1089. {
  1090. int err = jbd2_journal_dirty_data(handle, bh);
  1091. if (err)
  1092. ext4_journal_abort_handle(__FUNCTION__, __FUNCTION__,
  1093. bh, handle, err);
  1094. return err;
  1095. }
  1096. /* For write_end() in data=journal mode */
  1097. static int write_end_fn(handle_t *handle, struct buffer_head *bh)
  1098. {
  1099. if (!buffer_mapped(bh) || buffer_freed(bh))
  1100. return 0;
  1101. set_buffer_uptodate(bh);
  1102. return ext4_journal_dirty_metadata(handle, bh);
  1103. }
  1104. /*
  1105. * Generic write_end handler for ordered and writeback ext4 journal modes.
  1106. * We can't use generic_write_end, because that unlocks the page and we need to
  1107. * unlock the page after ext4_journal_stop, but ext4_journal_stop must run
  1108. * after block_write_end.
  1109. */
  1110. static int ext4_generic_write_end(struct file *file,
  1111. struct address_space *mapping,
  1112. loff_t pos, unsigned len, unsigned copied,
  1113. struct page *page, void *fsdata)
  1114. {
  1115. struct inode *inode = file->f_mapping->host;
  1116. copied = block_write_end(file, mapping, pos, len, copied, page, fsdata);
  1117. if (pos+copied > inode->i_size) {
  1118. i_size_write(inode, pos+copied);
  1119. mark_inode_dirty(inode);
  1120. }
  1121. return copied;
  1122. }
  1123. /*
  1124. * We need to pick up the new inode size which generic_commit_write gave us
  1125. * `file' can be NULL - eg, when called from page_symlink().
  1126. *
  1127. * ext4 never places buffers on inode->i_mapping->private_list. metadata
  1128. * buffers are managed internally.
  1129. */
  1130. static int ext4_ordered_write_end(struct file *file,
  1131. struct address_space *mapping,
  1132. loff_t pos, unsigned len, unsigned copied,
  1133. struct page *page, void *fsdata)
  1134. {
  1135. handle_t *handle = ext4_journal_current_handle();
  1136. struct inode *inode = file->f_mapping->host;
  1137. unsigned from, to;
  1138. int ret = 0, ret2;
  1139. from = pos & (PAGE_CACHE_SIZE - 1);
  1140. to = from + len;
  1141. ret = walk_page_buffers(handle, page_buffers(page),
  1142. from, to, NULL, ext4_journal_dirty_data);
  1143. if (ret == 0) {
  1144. /*
  1145. * generic_write_end() will run mark_inode_dirty() if i_size
  1146. * changes. So let's piggyback the i_disksize mark_inode_dirty
  1147. * into that.
  1148. */
  1149. loff_t new_i_size;
  1150. new_i_size = pos + copied;
  1151. if (new_i_size > EXT4_I(inode)->i_disksize)
  1152. EXT4_I(inode)->i_disksize = new_i_size;
  1153. copied = ext4_generic_write_end(file, mapping, pos, len, copied,
  1154. page, fsdata);
  1155. if (copied < 0)
  1156. ret = copied;
  1157. }
  1158. ret2 = ext4_journal_stop(handle);
  1159. if (!ret)
  1160. ret = ret2;
  1161. unlock_page(page);
  1162. page_cache_release(page);
  1163. return ret ? ret : copied;
  1164. }
  1165. static int ext4_writeback_write_end(struct file *file,
  1166. struct address_space *mapping,
  1167. loff_t pos, unsigned len, unsigned copied,
  1168. struct page *page, void *fsdata)
  1169. {
  1170. handle_t *handle = ext4_journal_current_handle();
  1171. struct inode *inode = file->f_mapping->host;
  1172. int ret = 0, ret2;
  1173. loff_t new_i_size;
  1174. new_i_size = pos + copied;
  1175. if (new_i_size > EXT4_I(inode)->i_disksize)
  1176. EXT4_I(inode)->i_disksize = new_i_size;
  1177. copied = ext4_generic_write_end(file, mapping, pos, len, copied,
  1178. page, fsdata);
  1179. if (copied < 0)
  1180. ret = copied;
  1181. ret2 = ext4_journal_stop(handle);
  1182. if (!ret)
  1183. ret = ret2;
  1184. unlock_page(page);
  1185. page_cache_release(page);
  1186. return ret ? ret : copied;
  1187. }
  1188. static int ext4_journalled_write_end(struct file *file,
  1189. struct address_space *mapping,
  1190. loff_t pos, unsigned len, unsigned copied,
  1191. struct page *page, void *fsdata)
  1192. {
  1193. handle_t *handle = ext4_journal_current_handle();
  1194. struct inode *inode = mapping->host;
  1195. int ret = 0, ret2;
  1196. int partial = 0;
  1197. unsigned from, to;
  1198. from = pos & (PAGE_CACHE_SIZE - 1);
  1199. to = from + len;
  1200. if (copied < len) {
  1201. if (!PageUptodate(page))
  1202. copied = 0;
  1203. page_zero_new_buffers(page, from+copied, to);
  1204. }
  1205. ret = walk_page_buffers(handle, page_buffers(page), from,
  1206. to, &partial, write_end_fn);
  1207. if (!partial)
  1208. SetPageUptodate(page);
  1209. if (pos+copied > inode->i_size)
  1210. i_size_write(inode, pos+copied);
  1211. EXT4_I(inode)->i_state |= EXT4_STATE_JDATA;
  1212. if (inode->i_size > EXT4_I(inode)->i_disksize) {
  1213. EXT4_I(inode)->i_disksize = inode->i_size;
  1214. ret2 = ext4_mark_inode_dirty(handle, inode);
  1215. if (!ret)
  1216. ret = ret2;
  1217. }
  1218. ret2 = ext4_journal_stop(handle);
  1219. if (!ret)
  1220. ret = ret2;
  1221. unlock_page(page);
  1222. page_cache_release(page);
  1223. return ret ? ret : copied;
  1224. }
  1225. /*
  1226. * bmap() is special. It gets used by applications such as lilo and by
  1227. * the swapper to find the on-disk block of a specific piece of data.
  1228. *
  1229. * Naturally, this is dangerous if the block concerned is still in the
  1230. * journal. If somebody makes a swapfile on an ext4 data-journaling
  1231. * filesystem and enables swap, then they may get a nasty shock when the
  1232. * data getting swapped to that swapfile suddenly gets overwritten by
  1233. * the original zero's written out previously to the journal and
  1234. * awaiting writeback in the kernel's buffer cache.
  1235. *
  1236. * So, if we see any bmap calls here on a modified, data-journaled file,
  1237. * take extra steps to flush any blocks which might be in the cache.
  1238. */
  1239. static sector_t ext4_bmap(struct address_space *mapping, sector_t block)
  1240. {
  1241. struct inode *inode = mapping->host;
  1242. journal_t *journal;
  1243. int err;
  1244. if (EXT4_I(inode)->i_state & EXT4_STATE_JDATA) {
  1245. /*
  1246. * This is a REALLY heavyweight approach, but the use of
  1247. * bmap on dirty files is expected to be extremely rare:
  1248. * only if we run lilo or swapon on a freshly made file
  1249. * do we expect this to happen.
  1250. *
  1251. * (bmap requires CAP_SYS_RAWIO so this does not
  1252. * represent an unprivileged user DOS attack --- we'd be
  1253. * in trouble if mortal users could trigger this path at
  1254. * will.)
  1255. *
  1256. * NB. EXT4_STATE_JDATA is not set on files other than
  1257. * regular files. If somebody wants to bmap a directory
  1258. * or symlink and gets confused because the buffer
  1259. * hasn't yet been flushed to disk, they deserve
  1260. * everything they get.
  1261. */
  1262. EXT4_I(inode)->i_state &= ~EXT4_STATE_JDATA;
  1263. journal = EXT4_JOURNAL(inode);
  1264. jbd2_journal_lock_updates(journal);
  1265. err = jbd2_journal_flush(journal);
  1266. jbd2_journal_unlock_updates(journal);
  1267. if (err)
  1268. return 0;
  1269. }
  1270. return generic_block_bmap(mapping,block,ext4_get_block);
  1271. }
  1272. static int bget_one(handle_t *handle, struct buffer_head *bh)
  1273. {
  1274. get_bh(bh);
  1275. return 0;
  1276. }
  1277. static int bput_one(handle_t *handle, struct buffer_head *bh)
  1278. {
  1279. put_bh(bh);
  1280. return 0;
  1281. }
  1282. static int jbd2_journal_dirty_data_fn(handle_t *handle, struct buffer_head *bh)
  1283. {
  1284. if (buffer_mapped(bh))
  1285. return ext4_journal_dirty_data(handle, bh);
  1286. return 0;
  1287. }
  1288. /*
  1289. * Note that we always start a transaction even if we're not journalling
  1290. * data. This is to preserve ordering: any hole instantiation within
  1291. * __block_write_full_page -> ext4_get_block() should be journalled
  1292. * along with the data so we don't crash and then get metadata which
  1293. * refers to old data.
  1294. *
  1295. * In all journalling modes block_write_full_page() will start the I/O.
  1296. *
  1297. * Problem:
  1298. *
  1299. * ext4_writepage() -> kmalloc() -> __alloc_pages() -> page_launder() ->
  1300. * ext4_writepage()
  1301. *
  1302. * Similar for:
  1303. *
  1304. * ext4_file_write() -> generic_file_write() -> __alloc_pages() -> ...
  1305. *
  1306. * Same applies to ext4_get_block(). We will deadlock on various things like
  1307. * lock_journal and i_truncate_mutex.
  1308. *
  1309. * Setting PF_MEMALLOC here doesn't work - too many internal memory
  1310. * allocations fail.
  1311. *
  1312. * 16May01: If we're reentered then journal_current_handle() will be
  1313. * non-zero. We simply *return*.
  1314. *
  1315. * 1 July 2001: @@@ FIXME:
  1316. * In journalled data mode, a data buffer may be metadata against the
  1317. * current transaction. But the same file is part of a shared mapping
  1318. * and someone does a writepage() on it.
  1319. *
  1320. * We will move the buffer onto the async_data list, but *after* it has
  1321. * been dirtied. So there's a small window where we have dirty data on
  1322. * BJ_Metadata.
  1323. *
  1324. * Note that this only applies to the last partial page in the file. The
  1325. * bit which block_write_full_page() uses prepare/commit for. (That's
  1326. * broken code anyway: it's wrong for msync()).
  1327. *
  1328. * It's a rare case: affects the final partial page, for journalled data
  1329. * where the file is subject to bith write() and writepage() in the same
  1330. * transction. To fix it we'll need a custom block_write_full_page().
  1331. * We'll probably need that anyway for journalling writepage() output.
  1332. *
  1333. * We don't honour synchronous mounts for writepage(). That would be
  1334. * disastrous. Any write() or metadata operation will sync the fs for
  1335. * us.
  1336. *
  1337. * AKPM2: if all the page's buffers are mapped to disk and !data=journal,
  1338. * we don't need to open a transaction here.
  1339. */
  1340. static int ext4_ordered_writepage(struct page *page,
  1341. struct writeback_control *wbc)
  1342. {
  1343. struct inode *inode = page->mapping->host;
  1344. struct buffer_head *page_bufs;
  1345. handle_t *handle = NULL;
  1346. int ret = 0;
  1347. int err;
  1348. J_ASSERT(PageLocked(page));
  1349. /*
  1350. * We give up here if we're reentered, because it might be for a
  1351. * different filesystem.
  1352. */
  1353. if (ext4_journal_current_handle())
  1354. goto out_fail;
  1355. handle = ext4_journal_start(inode, ext4_writepage_trans_blocks(inode));
  1356. if (IS_ERR(handle)) {
  1357. ret = PTR_ERR(handle);
  1358. goto out_fail;
  1359. }
  1360. if (!page_has_buffers(page)) {
  1361. create_empty_buffers(page, inode->i_sb->s_blocksize,
  1362. (1 << BH_Dirty)|(1 << BH_Uptodate));
  1363. }
  1364. page_bufs = page_buffers(page);
  1365. walk_page_buffers(handle, page_bufs, 0,
  1366. PAGE_CACHE_SIZE, NULL, bget_one);
  1367. ret = block_write_full_page(page, ext4_get_block, wbc);
  1368. /*
  1369. * The page can become unlocked at any point now, and
  1370. * truncate can then come in and change things. So we
  1371. * can't touch *page from now on. But *page_bufs is
  1372. * safe due to elevated refcount.
  1373. */
  1374. /*
  1375. * And attach them to the current transaction. But only if
  1376. * block_write_full_page() succeeded. Otherwise they are unmapped,
  1377. * and generally junk.
  1378. */
  1379. if (ret == 0) {
  1380. err = walk_page_buffers(handle, page_bufs, 0, PAGE_CACHE_SIZE,
  1381. NULL, jbd2_journal_dirty_data_fn);
  1382. if (!ret)
  1383. ret = err;
  1384. }
  1385. walk_page_buffers(handle, page_bufs, 0,
  1386. PAGE_CACHE_SIZE, NULL, bput_one);
  1387. err = ext4_journal_stop(handle);
  1388. if (!ret)
  1389. ret = err;
  1390. return ret;
  1391. out_fail:
  1392. redirty_page_for_writepage(wbc, page);
  1393. unlock_page(page);
  1394. return ret;
  1395. }
  1396. static int ext4_writeback_writepage(struct page *page,
  1397. struct writeback_control *wbc)
  1398. {
  1399. struct inode *inode = page->mapping->host;
  1400. handle_t *handle = NULL;
  1401. int ret = 0;
  1402. int err;
  1403. if (ext4_journal_current_handle())
  1404. goto out_fail;
  1405. handle = ext4_journal_start(inode, ext4_writepage_trans_blocks(inode));
  1406. if (IS_ERR(handle)) {
  1407. ret = PTR_ERR(handle);
  1408. goto out_fail;
  1409. }
  1410. if (test_opt(inode->i_sb, NOBH) && ext4_should_writeback_data(inode))
  1411. ret = nobh_writepage(page, ext4_get_block, wbc);
  1412. else
  1413. ret = block_write_full_page(page, ext4_get_block, wbc);
  1414. err = ext4_journal_stop(handle);
  1415. if (!ret)
  1416. ret = err;
  1417. return ret;
  1418. out_fail:
  1419. redirty_page_for_writepage(wbc, page);
  1420. unlock_page(page);
  1421. return ret;
  1422. }
  1423. static int ext4_journalled_writepage(struct page *page,
  1424. struct writeback_control *wbc)
  1425. {
  1426. struct inode *inode = page->mapping->host;
  1427. handle_t *handle = NULL;
  1428. int ret = 0;
  1429. int err;
  1430. if (ext4_journal_current_handle())
  1431. goto no_write;
  1432. handle = ext4_journal_start(inode, ext4_writepage_trans_blocks(inode));
  1433. if (IS_ERR(handle)) {
  1434. ret = PTR_ERR(handle);
  1435. goto no_write;
  1436. }
  1437. if (!page_has_buffers(page) || PageChecked(page)) {
  1438. /*
  1439. * It's mmapped pagecache. Add buffers and journal it. There
  1440. * doesn't seem much point in redirtying the page here.
  1441. */
  1442. ClearPageChecked(page);
  1443. ret = block_prepare_write(page, 0, PAGE_CACHE_SIZE,
  1444. ext4_get_block);
  1445. if (ret != 0) {
  1446. ext4_journal_stop(handle);
  1447. goto out_unlock;
  1448. }
  1449. ret = walk_page_buffers(handle, page_buffers(page), 0,
  1450. PAGE_CACHE_SIZE, NULL, do_journal_get_write_access);
  1451. err = walk_page_buffers(handle, page_buffers(page), 0,
  1452. PAGE_CACHE_SIZE, NULL, write_end_fn);
  1453. if (ret == 0)
  1454. ret = err;
  1455. EXT4_I(inode)->i_state |= EXT4_STATE_JDATA;
  1456. unlock_page(page);
  1457. } else {
  1458. /*
  1459. * It may be a page full of checkpoint-mode buffers. We don't
  1460. * really know unless we go poke around in the buffer_heads.
  1461. * But block_write_full_page will do the right thing.
  1462. */
  1463. ret = block_write_full_page(page, ext4_get_block, wbc);
  1464. }
  1465. err = ext4_journal_stop(handle);
  1466. if (!ret)
  1467. ret = err;
  1468. out:
  1469. return ret;
  1470. no_write:
  1471. redirty_page_for_writepage(wbc, page);
  1472. out_unlock:
  1473. unlock_page(page);
  1474. goto out;
  1475. }
  1476. static int ext4_readpage(struct file *file, struct page *page)
  1477. {
  1478. return mpage_readpage(page, ext4_get_block);
  1479. }
  1480. static int
  1481. ext4_readpages(struct file *file, struct address_space *mapping,
  1482. struct list_head *pages, unsigned nr_pages)
  1483. {
  1484. return mpage_readpages(mapping, pages, nr_pages, ext4_get_block);
  1485. }
  1486. static void ext4_invalidatepage(struct page *page, unsigned long offset)
  1487. {
  1488. journal_t *journal = EXT4_JOURNAL(page->mapping->host);
  1489. /*
  1490. * If it's a full truncate we just forget about the pending dirtying
  1491. */
  1492. if (offset == 0)
  1493. ClearPageChecked(page);
  1494. jbd2_journal_invalidatepage(journal, page, offset);
  1495. }
  1496. static int ext4_releasepage(struct page *page, gfp_t wait)
  1497. {
  1498. journal_t *journal = EXT4_JOURNAL(page->mapping->host);
  1499. WARN_ON(PageChecked(page));
  1500. if (!page_has_buffers(page))
  1501. return 0;
  1502. return jbd2_journal_try_to_free_buffers(journal, page, wait);
  1503. }
  1504. /*
  1505. * If the O_DIRECT write will extend the file then add this inode to the
  1506. * orphan list. So recovery will truncate it back to the original size
  1507. * if the machine crashes during the write.
  1508. *
  1509. * If the O_DIRECT write is intantiating holes inside i_size and the machine
  1510. * crashes then stale disk data _may_ be exposed inside the file.
  1511. */
  1512. static ssize_t ext4_direct_IO(int rw, struct kiocb *iocb,
  1513. const struct iovec *iov, loff_t offset,
  1514. unsigned long nr_segs)
  1515. {
  1516. struct file *file = iocb->ki_filp;
  1517. struct inode *inode = file->f_mapping->host;
  1518. struct ext4_inode_info *ei = EXT4_I(inode);
  1519. handle_t *handle = NULL;
  1520. ssize_t ret;
  1521. int orphan = 0;
  1522. size_t count = iov_length(iov, nr_segs);
  1523. if (rw == WRITE) {
  1524. loff_t final_size = offset + count;
  1525. handle = ext4_journal_start(inode, DIO_CREDITS);
  1526. if (IS_ERR(handle)) {
  1527. ret = PTR_ERR(handle);
  1528. goto out;
  1529. }
  1530. if (final_size > inode->i_size) {
  1531. ret = ext4_orphan_add(handle, inode);
  1532. if (ret)
  1533. goto out_stop;
  1534. orphan = 1;
  1535. ei->i_disksize = inode->i_size;
  1536. }
  1537. }
  1538. ret = blockdev_direct_IO(rw, iocb, inode, inode->i_sb->s_bdev, iov,
  1539. offset, nr_segs,
  1540. ext4_get_block, NULL);
  1541. /*
  1542. * Reacquire the handle: ext4_get_block() can restart the transaction
  1543. */
  1544. handle = ext4_journal_current_handle();
  1545. out_stop:
  1546. if (handle) {
  1547. int err;
  1548. if (orphan && inode->i_nlink)
  1549. ext4_orphan_del(handle, inode);
  1550. if (orphan && ret > 0) {
  1551. loff_t end = offset + ret;
  1552. if (end > inode->i_size) {
  1553. ei->i_disksize = end;
  1554. i_size_write(inode, end);
  1555. /*
  1556. * We're going to return a positive `ret'
  1557. * here due to non-zero-length I/O, so there's
  1558. * no way of reporting error returns from
  1559. * ext4_mark_inode_dirty() to userspace. So
  1560. * ignore it.
  1561. */
  1562. ext4_mark_inode_dirty(handle, inode);
  1563. }
  1564. }
  1565. err = ext4_journal_stop(handle);
  1566. if (ret == 0)
  1567. ret = err;
  1568. }
  1569. out:
  1570. return ret;
  1571. }
  1572. /*
  1573. * Pages can be marked dirty completely asynchronously from ext4's journalling
  1574. * activity. By filemap_sync_pte(), try_to_unmap_one(), etc. We cannot do
  1575. * much here because ->set_page_dirty is called under VFS locks. The page is
  1576. * not necessarily locked.
  1577. *
  1578. * We cannot just dirty the page and leave attached buffers clean, because the
  1579. * buffers' dirty state is "definitive". We cannot just set the buffers dirty
  1580. * or jbddirty because all the journalling code will explode.
  1581. *
  1582. * So what we do is to mark the page "pending dirty" and next time writepage
  1583. * is called, propagate that into the buffers appropriately.
  1584. */
  1585. static int ext4_journalled_set_page_dirty(struct page *page)
  1586. {
  1587. SetPageChecked(page);
  1588. return __set_page_dirty_nobuffers(page);
  1589. }
  1590. static const struct address_space_operations ext4_ordered_aops = {
  1591. .readpage = ext4_readpage,
  1592. .readpages = ext4_readpages,
  1593. .writepage = ext4_ordered_writepage,
  1594. .sync_page = block_sync_page,
  1595. .write_begin = ext4_write_begin,
  1596. .write_end = ext4_ordered_write_end,
  1597. .bmap = ext4_bmap,
  1598. .invalidatepage = ext4_invalidatepage,
  1599. .releasepage = ext4_releasepage,
  1600. .direct_IO = ext4_direct_IO,
  1601. .migratepage = buffer_migrate_page,
  1602. };
  1603. static const struct address_space_operations ext4_writeback_aops = {
  1604. .readpage = ext4_readpage,
  1605. .readpages = ext4_readpages,
  1606. .writepage = ext4_writeback_writepage,
  1607. .sync_page = block_sync_page,
  1608. .write_begin = ext4_write_begin,
  1609. .write_end = ext4_writeback_write_end,
  1610. .bmap = ext4_bmap,
  1611. .invalidatepage = ext4_invalidatepage,
  1612. .releasepage = ext4_releasepage,
  1613. .direct_IO = ext4_direct_IO,
  1614. .migratepage = buffer_migrate_page,
  1615. };
  1616. static const struct address_space_operations ext4_journalled_aops = {
  1617. .readpage = ext4_readpage,
  1618. .readpages = ext4_readpages,
  1619. .writepage = ext4_journalled_writepage,
  1620. .sync_page = block_sync_page,
  1621. .write_begin = ext4_write_begin,
  1622. .write_end = ext4_journalled_write_end,
  1623. .set_page_dirty = ext4_journalled_set_page_dirty,
  1624. .bmap = ext4_bmap,
  1625. .invalidatepage = ext4_invalidatepage,
  1626. .releasepage = ext4_releasepage,
  1627. };
  1628. void ext4_set_aops(struct inode *inode)
  1629. {
  1630. if (ext4_should_order_data(inode))
  1631. inode->i_mapping->a_ops = &ext4_ordered_aops;
  1632. else if (ext4_should_writeback_data(inode))
  1633. inode->i_mapping->a_ops = &ext4_writeback_aops;
  1634. else
  1635. inode->i_mapping->a_ops = &ext4_journalled_aops;
  1636. }
  1637. /*
  1638. * ext4_block_truncate_page() zeroes out a mapping from file offset `from'
  1639. * up to the end of the block which corresponds to `from'.
  1640. * This required during truncate. We need to physically zero the tail end
  1641. * of that block so it doesn't yield old data if the file is later grown.
  1642. */
  1643. int ext4_block_truncate_page(handle_t *handle, struct page *page,
  1644. struct address_space *mapping, loff_t from)
  1645. {
  1646. ext4_fsblk_t index = from >> PAGE_CACHE_SHIFT;
  1647. unsigned offset = from & (PAGE_CACHE_SIZE-1);
  1648. unsigned blocksize, length, pos;
  1649. ext4_lblk_t iblock;
  1650. struct inode *inode = mapping->host;
  1651. struct buffer_head *bh;
  1652. int err = 0;
  1653. blocksize = inode->i_sb->s_blocksize;
  1654. length = blocksize - (offset & (blocksize - 1));
  1655. iblock = index << (PAGE_CACHE_SHIFT - inode->i_sb->s_blocksize_bits);
  1656. /*
  1657. * For "nobh" option, we can only work if we don't need to
  1658. * read-in the page - otherwise we create buffers to do the IO.
  1659. */
  1660. if (!page_has_buffers(page) && test_opt(inode->i_sb, NOBH) &&
  1661. ext4_should_writeback_data(inode) && PageUptodate(page)) {
  1662. zero_user_page(page, offset, length, KM_USER0);
  1663. set_page_dirty(page);
  1664. goto unlock;
  1665. }
  1666. if (!page_has_buffers(page))
  1667. create_empty_buffers(page, blocksize, 0);
  1668. /* Find the buffer that contains "offset" */
  1669. bh = page_buffers(page);
  1670. pos = blocksize;
  1671. while (offset >= pos) {
  1672. bh = bh->b_this_page;
  1673. iblock++;
  1674. pos += blocksize;
  1675. }
  1676. err = 0;
  1677. if (buffer_freed(bh)) {
  1678. BUFFER_TRACE(bh, "freed: skip");
  1679. goto unlock;
  1680. }
  1681. if (!buffer_mapped(bh)) {
  1682. BUFFER_TRACE(bh, "unmapped");
  1683. ext4_get_block(inode, iblock, bh, 0);
  1684. /* unmapped? It's a hole - nothing to do */
  1685. if (!buffer_mapped(bh)) {
  1686. BUFFER_TRACE(bh, "still unmapped");
  1687. goto unlock;
  1688. }
  1689. }
  1690. /* Ok, it's mapped. Make sure it's up-to-date */
  1691. if (PageUptodate(page))
  1692. set_buffer_uptodate(bh);
  1693. if (!buffer_uptodate(bh)) {
  1694. err = -EIO;
  1695. ll_rw_block(READ, 1, &bh);
  1696. wait_on_buffer(bh);
  1697. /* Uhhuh. Read error. Complain and punt. */
  1698. if (!buffer_uptodate(bh))
  1699. goto unlock;
  1700. }
  1701. if (ext4_should_journal_data(inode)) {
  1702. BUFFER_TRACE(bh, "get write access");
  1703. err = ext4_journal_get_write_access(handle, bh);
  1704. if (err)
  1705. goto unlock;
  1706. }
  1707. zero_user_page(page, offset, length, KM_USER0);
  1708. BUFFER_TRACE(bh, "zeroed end of block");
  1709. err = 0;
  1710. if (ext4_should_journal_data(inode)) {
  1711. err = ext4_journal_dirty_metadata(handle, bh);
  1712. } else {
  1713. if (ext4_should_order_data(inode))
  1714. err = ext4_journal_dirty_data(handle, bh);
  1715. mark_buffer_dirty(bh);
  1716. }
  1717. unlock:
  1718. unlock_page(page);
  1719. page_cache_release(page);
  1720. return err;
  1721. }
  1722. /*
  1723. * Probably it should be a library function... search for first non-zero word
  1724. * or memcmp with zero_page, whatever is better for particular architecture.
  1725. * Linus?
  1726. */
  1727. static inline int all_zeroes(__le32 *p, __le32 *q)
  1728. {
  1729. while (p < q)
  1730. if (*p++)
  1731. return 0;
  1732. return 1;
  1733. }
  1734. /**
  1735. * ext4_find_shared - find the indirect blocks for partial truncation.
  1736. * @inode: inode in question
  1737. * @depth: depth of the affected branch
  1738. * @offsets: offsets of pointers in that branch (see ext4_block_to_path)
  1739. * @chain: place to store the pointers to partial indirect blocks
  1740. * @top: place to the (detached) top of branch
  1741. *
  1742. * This is a helper function used by ext4_truncate().
  1743. *
  1744. * When we do truncate() we may have to clean the ends of several
  1745. * indirect blocks but leave the blocks themselves alive. Block is
  1746. * partially truncated if some data below the new i_size is refered
  1747. * from it (and it is on the path to the first completely truncated
  1748. * data block, indeed). We have to free the top of that path along
  1749. * with everything to the right of the path. Since no allocation
  1750. * past the truncation point is possible until ext4_truncate()
  1751. * finishes, we may safely do the latter, but top of branch may
  1752. * require special attention - pageout below the truncation point
  1753. * might try to populate it.
  1754. *
  1755. * We atomically detach the top of branch from the tree, store the
  1756. * block number of its root in *@top, pointers to buffer_heads of
  1757. * partially truncated blocks - in @chain[].bh and pointers to
  1758. * their last elements that should not be removed - in
  1759. * @chain[].p. Return value is the pointer to last filled element
  1760. * of @chain.
  1761. *
  1762. * The work left to caller to do the actual freeing of subtrees:
  1763. * a) free the subtree starting from *@top
  1764. * b) free the subtrees whose roots are stored in
  1765. * (@chain[i].p+1 .. end of @chain[i].bh->b_data)
  1766. * c) free the subtrees growing from the inode past the @chain[0].
  1767. * (no partially truncated stuff there). */
  1768. static Indirect *ext4_find_shared(struct inode *inode, int depth,
  1769. ext4_lblk_t offsets[4], Indirect chain[4], __le32 *top)
  1770. {
  1771. Indirect *partial, *p;
  1772. int k, err;
  1773. *top = 0;
  1774. /* Make k index the deepest non-null offest + 1 */
  1775. for (k = depth; k > 1 && !offsets[k-1]; k--)
  1776. ;
  1777. partial = ext4_get_branch(inode, k, offsets, chain, &err);
  1778. /* Writer: pointers */
  1779. if (!partial)
  1780. partial = chain + k-1;
  1781. /*
  1782. * If the branch acquired continuation since we've looked at it -
  1783. * fine, it should all survive and (new) top doesn't belong to us.
  1784. */
  1785. if (!partial->key && *partial->p)
  1786. /* Writer: end */
  1787. goto no_top;
  1788. for (p=partial; p>chain && all_zeroes((__le32*)p->bh->b_data,p->p); p--)
  1789. ;
  1790. /*
  1791. * OK, we've found the last block that must survive. The rest of our
  1792. * branch should be detached before unlocking. However, if that rest
  1793. * of branch is all ours and does not grow immediately from the inode
  1794. * it's easier to cheat and just decrement partial->p.
  1795. */
  1796. if (p == chain + k - 1 && p > chain) {
  1797. p->p--;
  1798. } else {
  1799. *top = *p->p;
  1800. /* Nope, don't do this in ext4. Must leave the tree intact */
  1801. #if 0
  1802. *p->p = 0;
  1803. #endif
  1804. }
  1805. /* Writer: end */
  1806. while(partial > p) {
  1807. brelse(partial->bh);
  1808. partial--;
  1809. }
  1810. no_top:
  1811. return partial;
  1812. }
  1813. /*
  1814. * Zero a number of block pointers in either an inode or an indirect block.
  1815. * If we restart the transaction we must again get write access to the
  1816. * indirect block for further modification.
  1817. *
  1818. * We release `count' blocks on disk, but (last - first) may be greater
  1819. * than `count' because there can be holes in there.
  1820. */
  1821. static void ext4_clear_blocks(handle_t *handle, struct inode *inode,
  1822. struct buffer_head *bh, ext4_fsblk_t block_to_free,
  1823. unsigned long count, __le32 *first, __le32 *last)
  1824. {
  1825. __le32 *p;
  1826. if (try_to_extend_transaction(handle, inode)) {
  1827. if (bh) {
  1828. BUFFER_TRACE(bh, "call ext4_journal_dirty_metadata");
  1829. ext4_journal_dirty_metadata(handle, bh);
  1830. }
  1831. ext4_mark_inode_dirty(handle, inode);
  1832. ext4_journal_test_restart(handle, inode);
  1833. if (bh) {
  1834. BUFFER_TRACE(bh, "retaking write access");
  1835. ext4_journal_get_write_access(handle, bh);
  1836. }
  1837. }
  1838. /*
  1839. * Any buffers which are on the journal will be in memory. We find
  1840. * them on the hash table so jbd2_journal_revoke() will run jbd2_journal_forget()
  1841. * on them. We've already detached each block from the file, so
  1842. * bforget() in jbd2_journal_forget() should be safe.
  1843. *
  1844. * AKPM: turn on bforget in jbd2_journal_forget()!!!
  1845. */
  1846. for (p = first; p < last; p++) {
  1847. u32 nr = le32_to_cpu(*p);
  1848. if (nr) {
  1849. struct buffer_head *bh;
  1850. *p = 0;
  1851. bh = sb_find_get_block(inode->i_sb, nr);
  1852. ext4_forget(handle, 0, inode, bh, nr);
  1853. }
  1854. }
  1855. ext4_free_blocks(handle, inode, block_to_free, count);
  1856. }
  1857. /**
  1858. * ext4_free_data - free a list of data blocks
  1859. * @handle: handle for this transaction
  1860. * @inode: inode we are dealing with
  1861. * @this_bh: indirect buffer_head which contains *@first and *@last
  1862. * @first: array of block numbers
  1863. * @last: points immediately past the end of array
  1864. *
  1865. * We are freeing all blocks refered from that array (numbers are stored as
  1866. * little-endian 32-bit) and updating @inode->i_blocks appropriately.
  1867. *
  1868. * We accumulate contiguous runs of blocks to free. Conveniently, if these
  1869. * blocks are contiguous then releasing them at one time will only affect one
  1870. * or two bitmap blocks (+ group descriptor(s) and superblock) and we won't
  1871. * actually use a lot of journal space.
  1872. *
  1873. * @this_bh will be %NULL if @first and @last point into the inode's direct
  1874. * block pointers.
  1875. */
  1876. static void ext4_free_data(handle_t *handle, struct inode *inode,
  1877. struct buffer_head *this_bh,
  1878. __le32 *first, __le32 *last)
  1879. {
  1880. ext4_fsblk_t block_to_free = 0; /* Starting block # of a run */
  1881. unsigned long count = 0; /* Number of blocks in the run */
  1882. __le32 *block_to_free_p = NULL; /* Pointer into inode/ind
  1883. corresponding to
  1884. block_to_free */
  1885. ext4_fsblk_t nr; /* Current block # */
  1886. __le32 *p; /* Pointer into inode/ind
  1887. for current block */
  1888. int err;
  1889. if (this_bh) { /* For indirect block */
  1890. BUFFER_TRACE(this_bh, "get_write_access");
  1891. err = ext4_journal_get_write_access(handle, this_bh);
  1892. /* Important: if we can't update the indirect pointers
  1893. * to the blocks, we can't free them. */
  1894. if (err)
  1895. return;
  1896. }
  1897. for (p = first; p < last; p++) {
  1898. nr = le32_to_cpu(*p);
  1899. if (nr) {
  1900. /* accumulate blocks to free if they're contiguous */
  1901. if (count == 0) {
  1902. block_to_free = nr;
  1903. block_to_free_p = p;
  1904. count = 1;
  1905. } else if (nr == block_to_free + count) {
  1906. count++;
  1907. } else {
  1908. ext4_clear_blocks(handle, inode, this_bh,
  1909. block_to_free,
  1910. count, block_to_free_p, p);
  1911. block_to_free = nr;
  1912. block_to_free_p = p;
  1913. count = 1;
  1914. }
  1915. }
  1916. }
  1917. if (count > 0)
  1918. ext4_clear_blocks(handle, inode, this_bh, block_to_free,
  1919. count, block_to_free_p, p);
  1920. if (this_bh) {
  1921. BUFFER_TRACE(this_bh, "call ext4_journal_dirty_metadata");
  1922. ext4_journal_dirty_metadata(handle, this_bh);
  1923. }
  1924. }
  1925. /**
  1926. * ext4_free_branches - free an array of branches
  1927. * @handle: JBD handle for this transaction
  1928. * @inode: inode we are dealing with
  1929. * @parent_bh: the buffer_head which contains *@first and *@last
  1930. * @first: array of block numbers
  1931. * @last: pointer immediately past the end of array
  1932. * @depth: depth of the branches to free
  1933. *
  1934. * We are freeing all blocks refered from these branches (numbers are
  1935. * stored as little-endian 32-bit) and updating @inode->i_blocks
  1936. * appropriately.
  1937. */
  1938. static void ext4_free_branches(handle_t *handle, struct inode *inode,
  1939. struct buffer_head *parent_bh,
  1940. __le32 *first, __le32 *last, int depth)
  1941. {
  1942. ext4_fsblk_t nr;
  1943. __le32 *p;
  1944. if (is_handle_aborted(handle))
  1945. return;
  1946. if (depth--) {
  1947. struct buffer_head *bh;
  1948. int addr_per_block = EXT4_ADDR_PER_BLOCK(inode->i_sb);
  1949. p = last;
  1950. while (--p >= first) {
  1951. nr = le32_to_cpu(*p);
  1952. if (!nr)
  1953. continue; /* A hole */
  1954. /* Go read the buffer for the next level down */
  1955. bh = sb_bread(inode->i_sb, nr);
  1956. /*
  1957. * A read failure? Report error and clear slot
  1958. * (should be rare).
  1959. */
  1960. if (!bh) {
  1961. ext4_error(inode->i_sb, "ext4_free_branches",
  1962. "Read failure, inode=%lu, block=%llu",
  1963. inode->i_ino, nr);
  1964. continue;
  1965. }
  1966. /* This zaps the entire block. Bottom up. */
  1967. BUFFER_TRACE(bh, "free child branches");
  1968. ext4_free_branches(handle, inode, bh,
  1969. (__le32*)bh->b_data,
  1970. (__le32*)bh->b_data + addr_per_block,
  1971. depth);
  1972. /*
  1973. * We've probably journalled the indirect block several
  1974. * times during the truncate. But it's no longer
  1975. * needed and we now drop it from the transaction via
  1976. * jbd2_journal_revoke().
  1977. *
  1978. * That's easy if it's exclusively part of this
  1979. * transaction. But if it's part of the committing
  1980. * transaction then jbd2_journal_forget() will simply
  1981. * brelse() it. That means that if the underlying
  1982. * block is reallocated in ext4_get_block(),
  1983. * unmap_underlying_metadata() will find this block
  1984. * and will try to get rid of it. damn, damn.
  1985. *
  1986. * If this block has already been committed to the
  1987. * journal, a revoke record will be written. And
  1988. * revoke records must be emitted *before* clearing
  1989. * this block's bit in the bitmaps.
  1990. */
  1991. ext4_forget(handle, 1, inode, bh, bh->b_blocknr);
  1992. /*
  1993. * Everything below this this pointer has been
  1994. * released. Now let this top-of-subtree go.
  1995. *
  1996. * We want the freeing of this indirect block to be
  1997. * atomic in the journal with the updating of the
  1998. * bitmap block which owns it. So make some room in
  1999. * the journal.
  2000. *
  2001. * We zero the parent pointer *after* freeing its
  2002. * pointee in the bitmaps, so if extend_transaction()
  2003. * for some reason fails to put the bitmap changes and
  2004. * the release into the same transaction, recovery
  2005. * will merely complain about releasing a free block,
  2006. * rather than leaking blocks.
  2007. */
  2008. if (is_handle_aborted(handle))
  2009. return;
  2010. if (try_to_extend_transaction(handle, inode)) {
  2011. ext4_mark_inode_dirty(handle, inode);
  2012. ext4_journal_test_restart(handle, inode);
  2013. }
  2014. ext4_free_blocks(handle, inode, nr, 1);
  2015. if (parent_bh) {
  2016. /*
  2017. * The block which we have just freed is
  2018. * pointed to by an indirect block: journal it
  2019. */
  2020. BUFFER_TRACE(parent_bh, "get_write_access");
  2021. if (!ext4_journal_get_write_access(handle,
  2022. parent_bh)){
  2023. *p = 0;
  2024. BUFFER_TRACE(parent_bh,
  2025. "call ext4_journal_dirty_metadata");
  2026. ext4_journal_dirty_metadata(handle,
  2027. parent_bh);
  2028. }
  2029. }
  2030. }
  2031. } else {
  2032. /* We have reached the bottom of the tree. */
  2033. BUFFER_TRACE(parent_bh, "free data blocks");
  2034. ext4_free_data(handle, inode, parent_bh, first, last);
  2035. }
  2036. }
  2037. /*
  2038. * ext4_truncate()
  2039. *
  2040. * We block out ext4_get_block() block instantiations across the entire
  2041. * transaction, and VFS/VM ensures that ext4_truncate() cannot run
  2042. * simultaneously on behalf of the same inode.
  2043. *
  2044. * As we work through the truncate and commmit bits of it to the journal there
  2045. * is one core, guiding principle: the file's tree must always be consistent on
  2046. * disk. We must be able to restart the truncate after a crash.
  2047. *
  2048. * The file's tree may be transiently inconsistent in memory (although it
  2049. * probably isn't), but whenever we close off and commit a journal transaction,
  2050. * the contents of (the filesystem + the journal) must be consistent and
  2051. * restartable. It's pretty simple, really: bottom up, right to left (although
  2052. * left-to-right works OK too).
  2053. *
  2054. * Note that at recovery time, journal replay occurs *before* the restart of
  2055. * truncate against the orphan inode list.
  2056. *
  2057. * The committed inode has the new, desired i_size (which is the same as
  2058. * i_disksize in this case). After a crash, ext4_orphan_cleanup() will see
  2059. * that this inode's truncate did not complete and it will again call
  2060. * ext4_truncate() to have another go. So there will be instantiated blocks
  2061. * to the right of the truncation point in a crashed ext4 filesystem. But
  2062. * that's fine - as long as they are linked from the inode, the post-crash
  2063. * ext4_truncate() run will find them and release them.
  2064. */
  2065. void ext4_truncate(struct inode *inode)
  2066. {
  2067. handle_t *handle;
  2068. struct ext4_inode_info *ei = EXT4_I(inode);
  2069. __le32 *i_data = ei->i_data;
  2070. int addr_per_block = EXT4_ADDR_PER_BLOCK(inode->i_sb);
  2071. struct address_space *mapping = inode->i_mapping;
  2072. ext4_lblk_t offsets[4];
  2073. Indirect chain[4];
  2074. Indirect *partial;
  2075. __le32 nr = 0;
  2076. int n;
  2077. ext4_lblk_t last_block;
  2078. unsigned blocksize = inode->i_sb->s_blocksize;
  2079. struct page *page;
  2080. if (!(S_ISREG(inode->i_mode) || S_ISDIR(inode->i_mode) ||
  2081. S_ISLNK(inode->i_mode)))
  2082. return;
  2083. if (ext4_inode_is_fast_symlink(inode))
  2084. return;
  2085. if (IS_APPEND(inode) || IS_IMMUTABLE(inode))
  2086. return;
  2087. /*
  2088. * We have to lock the EOF page here, because lock_page() nests
  2089. * outside jbd2_journal_start().
  2090. */
  2091. if ((inode->i_size & (blocksize - 1)) == 0) {
  2092. /* Block boundary? Nothing to do */
  2093. page = NULL;
  2094. } else {
  2095. page = grab_cache_page(mapping,
  2096. inode->i_size >> PAGE_CACHE_SHIFT);
  2097. if (!page)
  2098. return;
  2099. }
  2100. if (EXT4_I(inode)->i_flags & EXT4_EXTENTS_FL)
  2101. return ext4_ext_truncate(inode, page);
  2102. handle = start_transaction(inode);
  2103. if (IS_ERR(handle)) {
  2104. if (page) {
  2105. clear_highpage(page);
  2106. flush_dcache_page(page);
  2107. unlock_page(page);
  2108. page_cache_release(page);
  2109. }
  2110. return; /* AKPM: return what? */
  2111. }
  2112. last_block = (inode->i_size + blocksize-1)
  2113. >> EXT4_BLOCK_SIZE_BITS(inode->i_sb);
  2114. if (page)
  2115. ext4_block_truncate_page(handle, page, mapping, inode->i_size);
  2116. n = ext4_block_to_path(inode, last_block, offsets, NULL);
  2117. if (n == 0)
  2118. goto out_stop; /* error */
  2119. /*
  2120. * OK. This truncate is going to happen. We add the inode to the
  2121. * orphan list, so that if this truncate spans multiple transactions,
  2122. * and we crash, we will resume the truncate when the filesystem
  2123. * recovers. It also marks the inode dirty, to catch the new size.
  2124. *
  2125. * Implication: the file must always be in a sane, consistent
  2126. * truncatable state while each transaction commits.
  2127. */
  2128. if (ext4_orphan_add(handle, inode))
  2129. goto out_stop;
  2130. /*
  2131. * The orphan list entry will now protect us from any crash which
  2132. * occurs before the truncate completes, so it is now safe to propagate
  2133. * the new, shorter inode size (held for now in i_size) into the
  2134. * on-disk inode. We do this via i_disksize, which is the value which
  2135. * ext4 *really* writes onto the disk inode.
  2136. */
  2137. ei->i_disksize = inode->i_size;
  2138. /*
  2139. * From here we block out all ext4_get_block() callers who want to
  2140. * modify the block allocation tree.
  2141. */
  2142. mutex_lock(&ei->truncate_mutex);
  2143. if (n == 1) { /* direct blocks */
  2144. ext4_free_data(handle, inode, NULL, i_data+offsets[0],
  2145. i_data + EXT4_NDIR_BLOCKS);
  2146. goto do_indirects;
  2147. }
  2148. partial = ext4_find_shared(inode, n, offsets, chain, &nr);
  2149. /* Kill the top of shared branch (not detached) */
  2150. if (nr) {
  2151. if (partial == chain) {
  2152. /* Shared branch grows from the inode */
  2153. ext4_free_branches(handle, inode, NULL,
  2154. &nr, &nr+1, (chain+n-1) - partial);
  2155. *partial->p = 0;
  2156. /*
  2157. * We mark the inode dirty prior to restart,
  2158. * and prior to stop. No need for it here.
  2159. */
  2160. } else {
  2161. /* Shared branch grows from an indirect block */
  2162. BUFFER_TRACE(partial->bh, "get_write_access");
  2163. ext4_free_branches(handle, inode, partial->bh,
  2164. partial->p,
  2165. partial->p+1, (chain+n-1) - partial);
  2166. }
  2167. }
  2168. /* Clear the ends of indirect blocks on the shared branch */
  2169. while (partial > chain) {
  2170. ext4_free_branches(handle, inode, partial->bh, partial->p + 1,
  2171. (__le32*)partial->bh->b_data+addr_per_block,
  2172. (chain+n-1) - partial);
  2173. BUFFER_TRACE(partial->bh, "call brelse");
  2174. brelse (partial->bh);
  2175. partial--;
  2176. }
  2177. do_indirects:
  2178. /* Kill the remaining (whole) subtrees */
  2179. switch (offsets[0]) {
  2180. default:
  2181. nr = i_data[EXT4_IND_BLOCK];
  2182. if (nr) {
  2183. ext4_free_branches(handle, inode, NULL, &nr, &nr+1, 1);
  2184. i_data[EXT4_IND_BLOCK] = 0;
  2185. }
  2186. case EXT4_IND_BLOCK:
  2187. nr = i_data[EXT4_DIND_BLOCK];
  2188. if (nr) {
  2189. ext4_free_branches(handle, inode, NULL, &nr, &nr+1, 2);
  2190. i_data[EXT4_DIND_BLOCK] = 0;
  2191. }
  2192. case EXT4_DIND_BLOCK:
  2193. nr = i_data[EXT4_TIND_BLOCK];
  2194. if (nr) {
  2195. ext4_free_branches(handle, inode, NULL, &nr, &nr+1, 3);
  2196. i_data[EXT4_TIND_BLOCK] = 0;
  2197. }
  2198. case EXT4_TIND_BLOCK:
  2199. ;
  2200. }
  2201. ext4_discard_reservation(inode);
  2202. mutex_unlock(&ei->truncate_mutex);
  2203. inode->i_mtime = inode->i_ctime = ext4_current_time(inode);
  2204. ext4_mark_inode_dirty(handle, inode);
  2205. /*
  2206. * In a multi-transaction truncate, we only make the final transaction
  2207. * synchronous
  2208. */
  2209. if (IS_SYNC(inode))
  2210. handle->h_sync = 1;
  2211. out_stop:
  2212. /*
  2213. * If this was a simple ftruncate(), and the file will remain alive
  2214. * then we need to clear up the orphan record which we created above.
  2215. * However, if this was a real unlink then we were called by
  2216. * ext4_delete_inode(), and we allow that function to clean up the
  2217. * orphan info for us.
  2218. */
  2219. if (inode->i_nlink)
  2220. ext4_orphan_del(handle, inode);
  2221. ext4_journal_stop(handle);
  2222. }
  2223. static ext4_fsblk_t ext4_get_inode_block(struct super_block *sb,
  2224. unsigned long ino, struct ext4_iloc *iloc)
  2225. {
  2226. unsigned long desc, group_desc;
  2227. ext4_group_t block_group;
  2228. unsigned long offset;
  2229. ext4_fsblk_t block;
  2230. struct buffer_head *bh;
  2231. struct ext4_group_desc * gdp;
  2232. if (!ext4_valid_inum(sb, ino)) {
  2233. /*
  2234. * This error is already checked for in namei.c unless we are
  2235. * looking at an NFS filehandle, in which case no error
  2236. * report is needed
  2237. */
  2238. return 0;
  2239. }
  2240. block_group = (ino - 1) / EXT4_INODES_PER_GROUP(sb);
  2241. if (block_group >= EXT4_SB(sb)->s_groups_count) {
  2242. ext4_error(sb,"ext4_get_inode_block","group >= groups count");
  2243. return 0;
  2244. }
  2245. smp_rmb();
  2246. group_desc = block_group >> EXT4_DESC_PER_BLOCK_BITS(sb);
  2247. desc = block_group & (EXT4_DESC_PER_BLOCK(sb) - 1);
  2248. bh = EXT4_SB(sb)->s_group_desc[group_desc];
  2249. if (!bh) {
  2250. ext4_error (sb, "ext4_get_inode_block",
  2251. "Descriptor not loaded");
  2252. return 0;
  2253. }
  2254. gdp = (struct ext4_group_desc *)((__u8 *)bh->b_data +
  2255. desc * EXT4_DESC_SIZE(sb));
  2256. /*
  2257. * Figure out the offset within the block group inode table
  2258. */
  2259. offset = ((ino - 1) % EXT4_INODES_PER_GROUP(sb)) *
  2260. EXT4_INODE_SIZE(sb);
  2261. block = ext4_inode_table(sb, gdp) +
  2262. (offset >> EXT4_BLOCK_SIZE_BITS(sb));
  2263. iloc->block_group = block_group;
  2264. iloc->offset = offset & (EXT4_BLOCK_SIZE(sb) - 1);
  2265. return block;
  2266. }
  2267. /*
  2268. * ext4_get_inode_loc returns with an extra refcount against the inode's
  2269. * underlying buffer_head on success. If 'in_mem' is true, we have all
  2270. * data in memory that is needed to recreate the on-disk version of this
  2271. * inode.
  2272. */
  2273. static int __ext4_get_inode_loc(struct inode *inode,
  2274. struct ext4_iloc *iloc, int in_mem)
  2275. {
  2276. ext4_fsblk_t block;
  2277. struct buffer_head *bh;
  2278. block = ext4_get_inode_block(inode->i_sb, inode->i_ino, iloc);
  2279. if (!block)
  2280. return -EIO;
  2281. bh = sb_getblk(inode->i_sb, block);
  2282. if (!bh) {
  2283. ext4_error (inode->i_sb, "ext4_get_inode_loc",
  2284. "unable to read inode block - "
  2285. "inode=%lu, block=%llu",
  2286. inode->i_ino, block);
  2287. return -EIO;
  2288. }
  2289. if (!buffer_uptodate(bh)) {
  2290. lock_buffer(bh);
  2291. if (buffer_uptodate(bh)) {
  2292. /* someone brought it uptodate while we waited */
  2293. unlock_buffer(bh);
  2294. goto has_buffer;
  2295. }
  2296. /*
  2297. * If we have all information of the inode in memory and this
  2298. * is the only valid inode in the block, we need not read the
  2299. * block.
  2300. */
  2301. if (in_mem) {
  2302. struct buffer_head *bitmap_bh;
  2303. struct ext4_group_desc *desc;
  2304. int inodes_per_buffer;
  2305. int inode_offset, i;
  2306. ext4_group_t block_group;
  2307. int start;
  2308. block_group = (inode->i_ino - 1) /
  2309. EXT4_INODES_PER_GROUP(inode->i_sb);
  2310. inodes_per_buffer = bh->b_size /
  2311. EXT4_INODE_SIZE(inode->i_sb);
  2312. inode_offset = ((inode->i_ino - 1) %
  2313. EXT4_INODES_PER_GROUP(inode->i_sb));
  2314. start = inode_offset & ~(inodes_per_buffer - 1);
  2315. /* Is the inode bitmap in cache? */
  2316. desc = ext4_get_group_desc(inode->i_sb,
  2317. block_group, NULL);
  2318. if (!desc)
  2319. goto make_io;
  2320. bitmap_bh = sb_getblk(inode->i_sb,
  2321. ext4_inode_bitmap(inode->i_sb, desc));
  2322. if (!bitmap_bh)
  2323. goto make_io;
  2324. /*
  2325. * If the inode bitmap isn't in cache then the
  2326. * optimisation may end up performing two reads instead
  2327. * of one, so skip it.
  2328. */
  2329. if (!buffer_uptodate(bitmap_bh)) {
  2330. brelse(bitmap_bh);
  2331. goto make_io;
  2332. }
  2333. for (i = start; i < start + inodes_per_buffer; i++) {
  2334. if (i == inode_offset)
  2335. continue;
  2336. if (ext4_test_bit(i, bitmap_bh->b_data))
  2337. break;
  2338. }
  2339. brelse(bitmap_bh);
  2340. if (i == start + inodes_per_buffer) {
  2341. /* all other inodes are free, so skip I/O */
  2342. memset(bh->b_data, 0, bh->b_size);
  2343. set_buffer_uptodate(bh);
  2344. unlock_buffer(bh);
  2345. goto has_buffer;
  2346. }
  2347. }
  2348. make_io:
  2349. /*
  2350. * There are other valid inodes in the buffer, this inode
  2351. * has in-inode xattrs, or we don't have this inode in memory.
  2352. * Read the block from disk.
  2353. */
  2354. get_bh(bh);
  2355. bh->b_end_io = end_buffer_read_sync;
  2356. submit_bh(READ_META, bh);
  2357. wait_on_buffer(bh);
  2358. if (!buffer_uptodate(bh)) {
  2359. ext4_error(inode->i_sb, "ext4_get_inode_loc",
  2360. "unable to read inode block - "
  2361. "inode=%lu, block=%llu",
  2362. inode->i_ino, block);
  2363. brelse(bh);
  2364. return -EIO;
  2365. }
  2366. }
  2367. has_buffer:
  2368. iloc->bh = bh;
  2369. return 0;
  2370. }
  2371. int ext4_get_inode_loc(struct inode *inode, struct ext4_iloc *iloc)
  2372. {
  2373. /* We have all inode data except xattrs in memory here. */
  2374. return __ext4_get_inode_loc(inode, iloc,
  2375. !(EXT4_I(inode)->i_state & EXT4_STATE_XATTR));
  2376. }
  2377. void ext4_set_inode_flags(struct inode *inode)
  2378. {
  2379. unsigned int flags = EXT4_I(inode)->i_flags;
  2380. inode->i_flags &= ~(S_SYNC|S_APPEND|S_IMMUTABLE|S_NOATIME|S_DIRSYNC);
  2381. if (flags & EXT4_SYNC_FL)
  2382. inode->i_flags |= S_SYNC;
  2383. if (flags & EXT4_APPEND_FL)
  2384. inode->i_flags |= S_APPEND;
  2385. if (flags & EXT4_IMMUTABLE_FL)
  2386. inode->i_flags |= S_IMMUTABLE;
  2387. if (flags & EXT4_NOATIME_FL)
  2388. inode->i_flags |= S_NOATIME;
  2389. if (flags & EXT4_DIRSYNC_FL)
  2390. inode->i_flags |= S_DIRSYNC;
  2391. }
  2392. /* Propagate flags from i_flags to EXT4_I(inode)->i_flags */
  2393. void ext4_get_inode_flags(struct ext4_inode_info *ei)
  2394. {
  2395. unsigned int flags = ei->vfs_inode.i_flags;
  2396. ei->i_flags &= ~(EXT4_SYNC_FL|EXT4_APPEND_FL|
  2397. EXT4_IMMUTABLE_FL|EXT4_NOATIME_FL|EXT4_DIRSYNC_FL);
  2398. if (flags & S_SYNC)
  2399. ei->i_flags |= EXT4_SYNC_FL;
  2400. if (flags & S_APPEND)
  2401. ei->i_flags |= EXT4_APPEND_FL;
  2402. if (flags & S_IMMUTABLE)
  2403. ei->i_flags |= EXT4_IMMUTABLE_FL;
  2404. if (flags & S_NOATIME)
  2405. ei->i_flags |= EXT4_NOATIME_FL;
  2406. if (flags & S_DIRSYNC)
  2407. ei->i_flags |= EXT4_DIRSYNC_FL;
  2408. }
  2409. void ext4_read_inode(struct inode * inode)
  2410. {
  2411. struct ext4_iloc iloc;
  2412. struct ext4_inode *raw_inode;
  2413. struct ext4_inode_info *ei = EXT4_I(inode);
  2414. struct buffer_head *bh;
  2415. int block;
  2416. #ifdef CONFIG_EXT4DEV_FS_POSIX_ACL
  2417. ei->i_acl = EXT4_ACL_NOT_CACHED;
  2418. ei->i_default_acl = EXT4_ACL_NOT_CACHED;
  2419. #endif
  2420. ei->i_block_alloc_info = NULL;
  2421. if (__ext4_get_inode_loc(inode, &iloc, 0))
  2422. goto bad_inode;
  2423. bh = iloc.bh;
  2424. raw_inode = ext4_raw_inode(&iloc);
  2425. inode->i_mode = le16_to_cpu(raw_inode->i_mode);
  2426. inode->i_uid = (uid_t)le16_to_cpu(raw_inode->i_uid_low);
  2427. inode->i_gid = (gid_t)le16_to_cpu(raw_inode->i_gid_low);
  2428. if(!(test_opt (inode->i_sb, NO_UID32))) {
  2429. inode->i_uid |= le16_to_cpu(raw_inode->i_uid_high) << 16;
  2430. inode->i_gid |= le16_to_cpu(raw_inode->i_gid_high) << 16;
  2431. }
  2432. inode->i_nlink = le16_to_cpu(raw_inode->i_links_count);
  2433. inode->i_size = le32_to_cpu(raw_inode->i_size);
  2434. ei->i_state = 0;
  2435. ei->i_dir_start_lookup = 0;
  2436. ei->i_dtime = le32_to_cpu(raw_inode->i_dtime);
  2437. /* We now have enough fields to check if the inode was active or not.
  2438. * This is needed because nfsd might try to access dead inodes
  2439. * the test is that same one that e2fsck uses
  2440. * NeilBrown 1999oct15
  2441. */
  2442. if (inode->i_nlink == 0) {
  2443. if (inode->i_mode == 0 ||
  2444. !(EXT4_SB(inode->i_sb)->s_mount_state & EXT4_ORPHAN_FS)) {
  2445. /* this inode is deleted */
  2446. brelse (bh);
  2447. goto bad_inode;
  2448. }
  2449. /* The only unlinked inodes we let through here have
  2450. * valid i_mode and are being read by the orphan
  2451. * recovery code: that's fine, we're about to complete
  2452. * the process of deleting those. */
  2453. }
  2454. inode->i_blocks = le32_to_cpu(raw_inode->i_blocks);
  2455. ei->i_flags = le32_to_cpu(raw_inode->i_flags);
  2456. ei->i_file_acl = le32_to_cpu(raw_inode->i_file_acl);
  2457. if (EXT4_SB(inode->i_sb)->s_es->s_creator_os !=
  2458. cpu_to_le32(EXT4_OS_HURD))
  2459. ei->i_file_acl |=
  2460. ((__u64)le16_to_cpu(raw_inode->i_file_acl_high)) << 32;
  2461. if (!S_ISREG(inode->i_mode)) {
  2462. ei->i_dir_acl = le32_to_cpu(raw_inode->i_dir_acl);
  2463. } else {
  2464. inode->i_size |=
  2465. ((__u64)le32_to_cpu(raw_inode->i_size_high)) << 32;
  2466. }
  2467. ei->i_disksize = inode->i_size;
  2468. inode->i_generation = le32_to_cpu(raw_inode->i_generation);
  2469. ei->i_block_group = iloc.block_group;
  2470. /*
  2471. * NOTE! The in-memory inode i_data array is in little-endian order
  2472. * even on big-endian machines: we do NOT byteswap the block numbers!
  2473. */
  2474. for (block = 0; block < EXT4_N_BLOCKS; block++)
  2475. ei->i_data[block] = raw_inode->i_block[block];
  2476. INIT_LIST_HEAD(&ei->i_orphan);
  2477. if (inode->i_ino >= EXT4_FIRST_INO(inode->i_sb) + 1 &&
  2478. EXT4_INODE_SIZE(inode->i_sb) > EXT4_GOOD_OLD_INODE_SIZE) {
  2479. /*
  2480. * When mke2fs creates big inodes it does not zero out
  2481. * the unused bytes above EXT4_GOOD_OLD_INODE_SIZE,
  2482. * so ignore those first few inodes.
  2483. */
  2484. ei->i_extra_isize = le16_to_cpu(raw_inode->i_extra_isize);
  2485. if (EXT4_GOOD_OLD_INODE_SIZE + ei->i_extra_isize >
  2486. EXT4_INODE_SIZE(inode->i_sb)) {
  2487. brelse (bh);
  2488. goto bad_inode;
  2489. }
  2490. if (ei->i_extra_isize == 0) {
  2491. /* The extra space is currently unused. Use it. */
  2492. ei->i_extra_isize = sizeof(struct ext4_inode) -
  2493. EXT4_GOOD_OLD_INODE_SIZE;
  2494. } else {
  2495. __le32 *magic = (void *)raw_inode +
  2496. EXT4_GOOD_OLD_INODE_SIZE +
  2497. ei->i_extra_isize;
  2498. if (*magic == cpu_to_le32(EXT4_XATTR_MAGIC))
  2499. ei->i_state |= EXT4_STATE_XATTR;
  2500. }
  2501. } else
  2502. ei->i_extra_isize = 0;
  2503. EXT4_INODE_GET_XTIME(i_ctime, inode, raw_inode);
  2504. EXT4_INODE_GET_XTIME(i_mtime, inode, raw_inode);
  2505. EXT4_INODE_GET_XTIME(i_atime, inode, raw_inode);
  2506. EXT4_EINODE_GET_XTIME(i_crtime, ei, raw_inode);
  2507. if (S_ISREG(inode->i_mode)) {
  2508. inode->i_op = &ext4_file_inode_operations;
  2509. inode->i_fop = &ext4_file_operations;
  2510. ext4_set_aops(inode);
  2511. } else if (S_ISDIR(inode->i_mode)) {
  2512. inode->i_op = &ext4_dir_inode_operations;
  2513. inode->i_fop = &ext4_dir_operations;
  2514. } else if (S_ISLNK(inode->i_mode)) {
  2515. if (ext4_inode_is_fast_symlink(inode))
  2516. inode->i_op = &ext4_fast_symlink_inode_operations;
  2517. else {
  2518. inode->i_op = &ext4_symlink_inode_operations;
  2519. ext4_set_aops(inode);
  2520. }
  2521. } else {
  2522. inode->i_op = &ext4_special_inode_operations;
  2523. if (raw_inode->i_block[0])
  2524. init_special_inode(inode, inode->i_mode,
  2525. old_decode_dev(le32_to_cpu(raw_inode->i_block[0])));
  2526. else
  2527. init_special_inode(inode, inode->i_mode,
  2528. new_decode_dev(le32_to_cpu(raw_inode->i_block[1])));
  2529. }
  2530. brelse (iloc.bh);
  2531. ext4_set_inode_flags(inode);
  2532. return;
  2533. bad_inode:
  2534. make_bad_inode(inode);
  2535. return;
  2536. }
  2537. /*
  2538. * Post the struct inode info into an on-disk inode location in the
  2539. * buffer-cache. This gobbles the caller's reference to the
  2540. * buffer_head in the inode location struct.
  2541. *
  2542. * The caller must have write access to iloc->bh.
  2543. */
  2544. static int ext4_do_update_inode(handle_t *handle,
  2545. struct inode *inode,
  2546. struct ext4_iloc *iloc)
  2547. {
  2548. struct ext4_inode *raw_inode = ext4_raw_inode(iloc);
  2549. struct ext4_inode_info *ei = EXT4_I(inode);
  2550. struct buffer_head *bh = iloc->bh;
  2551. int err = 0, rc, block;
  2552. /* For fields not not tracking in the in-memory inode,
  2553. * initialise them to zero for new inodes. */
  2554. if (ei->i_state & EXT4_STATE_NEW)
  2555. memset(raw_inode, 0, EXT4_SB(inode->i_sb)->s_inode_size);
  2556. ext4_get_inode_flags(ei);
  2557. raw_inode->i_mode = cpu_to_le16(inode->i_mode);
  2558. if(!(test_opt(inode->i_sb, NO_UID32))) {
  2559. raw_inode->i_uid_low = cpu_to_le16(low_16_bits(inode->i_uid));
  2560. raw_inode->i_gid_low = cpu_to_le16(low_16_bits(inode->i_gid));
  2561. /*
  2562. * Fix up interoperability with old kernels. Otherwise, old inodes get
  2563. * re-used with the upper 16 bits of the uid/gid intact
  2564. */
  2565. if(!ei->i_dtime) {
  2566. raw_inode->i_uid_high =
  2567. cpu_to_le16(high_16_bits(inode->i_uid));
  2568. raw_inode->i_gid_high =
  2569. cpu_to_le16(high_16_bits(inode->i_gid));
  2570. } else {
  2571. raw_inode->i_uid_high = 0;
  2572. raw_inode->i_gid_high = 0;
  2573. }
  2574. } else {
  2575. raw_inode->i_uid_low =
  2576. cpu_to_le16(fs_high2lowuid(inode->i_uid));
  2577. raw_inode->i_gid_low =
  2578. cpu_to_le16(fs_high2lowgid(inode->i_gid));
  2579. raw_inode->i_uid_high = 0;
  2580. raw_inode->i_gid_high = 0;
  2581. }
  2582. raw_inode->i_links_count = cpu_to_le16(inode->i_nlink);
  2583. raw_inode->i_size = cpu_to_le32(ei->i_disksize);
  2584. EXT4_INODE_SET_XTIME(i_ctime, inode, raw_inode);
  2585. EXT4_INODE_SET_XTIME(i_mtime, inode, raw_inode);
  2586. EXT4_INODE_SET_XTIME(i_atime, inode, raw_inode);
  2587. EXT4_EINODE_SET_XTIME(i_crtime, ei, raw_inode);
  2588. raw_inode->i_blocks = cpu_to_le32(inode->i_blocks);
  2589. raw_inode->i_dtime = cpu_to_le32(ei->i_dtime);
  2590. raw_inode->i_flags = cpu_to_le32(ei->i_flags);
  2591. if (EXT4_SB(inode->i_sb)->s_es->s_creator_os !=
  2592. cpu_to_le32(EXT4_OS_HURD))
  2593. raw_inode->i_file_acl_high =
  2594. cpu_to_le16(ei->i_file_acl >> 32);
  2595. raw_inode->i_file_acl = cpu_to_le32(ei->i_file_acl);
  2596. if (!S_ISREG(inode->i_mode)) {
  2597. raw_inode->i_dir_acl = cpu_to_le32(ei->i_dir_acl);
  2598. } else {
  2599. raw_inode->i_size_high =
  2600. cpu_to_le32(ei->i_disksize >> 32);
  2601. if (ei->i_disksize > 0x7fffffffULL) {
  2602. struct super_block *sb = inode->i_sb;
  2603. if (!EXT4_HAS_RO_COMPAT_FEATURE(sb,
  2604. EXT4_FEATURE_RO_COMPAT_LARGE_FILE) ||
  2605. EXT4_SB(sb)->s_es->s_rev_level ==
  2606. cpu_to_le32(EXT4_GOOD_OLD_REV)) {
  2607. /* If this is the first large file
  2608. * created, add a flag to the superblock.
  2609. */
  2610. err = ext4_journal_get_write_access(handle,
  2611. EXT4_SB(sb)->s_sbh);
  2612. if (err)
  2613. goto out_brelse;
  2614. ext4_update_dynamic_rev(sb);
  2615. EXT4_SET_RO_COMPAT_FEATURE(sb,
  2616. EXT4_FEATURE_RO_COMPAT_LARGE_FILE);
  2617. sb->s_dirt = 1;
  2618. handle->h_sync = 1;
  2619. err = ext4_journal_dirty_metadata(handle,
  2620. EXT4_SB(sb)->s_sbh);
  2621. }
  2622. }
  2623. }
  2624. raw_inode->i_generation = cpu_to_le32(inode->i_generation);
  2625. if (S_ISCHR(inode->i_mode) || S_ISBLK(inode->i_mode)) {
  2626. if (old_valid_dev(inode->i_rdev)) {
  2627. raw_inode->i_block[0] =
  2628. cpu_to_le32(old_encode_dev(inode->i_rdev));
  2629. raw_inode->i_block[1] = 0;
  2630. } else {
  2631. raw_inode->i_block[0] = 0;
  2632. raw_inode->i_block[1] =
  2633. cpu_to_le32(new_encode_dev(inode->i_rdev));
  2634. raw_inode->i_block[2] = 0;
  2635. }
  2636. } else for (block = 0; block < EXT4_N_BLOCKS; block++)
  2637. raw_inode->i_block[block] = ei->i_data[block];
  2638. if (ei->i_extra_isize)
  2639. raw_inode->i_extra_isize = cpu_to_le16(ei->i_extra_isize);
  2640. BUFFER_TRACE(bh, "call ext4_journal_dirty_metadata");
  2641. rc = ext4_journal_dirty_metadata(handle, bh);
  2642. if (!err)
  2643. err = rc;
  2644. ei->i_state &= ~EXT4_STATE_NEW;
  2645. out_brelse:
  2646. brelse (bh);
  2647. ext4_std_error(inode->i_sb, err);
  2648. return err;
  2649. }
  2650. /*
  2651. * ext4_write_inode()
  2652. *
  2653. * We are called from a few places:
  2654. *
  2655. * - Within generic_file_write() for O_SYNC files.
  2656. * Here, there will be no transaction running. We wait for any running
  2657. * trasnaction to commit.
  2658. *
  2659. * - Within sys_sync(), kupdate and such.
  2660. * We wait on commit, if tol to.
  2661. *
  2662. * - Within prune_icache() (PF_MEMALLOC == true)
  2663. * Here we simply return. We can't afford to block kswapd on the
  2664. * journal commit.
  2665. *
  2666. * In all cases it is actually safe for us to return without doing anything,
  2667. * because the inode has been copied into a raw inode buffer in
  2668. * ext4_mark_inode_dirty(). This is a correctness thing for O_SYNC and for
  2669. * knfsd.
  2670. *
  2671. * Note that we are absolutely dependent upon all inode dirtiers doing the
  2672. * right thing: they *must* call mark_inode_dirty() after dirtying info in
  2673. * which we are interested.
  2674. *
  2675. * It would be a bug for them to not do this. The code:
  2676. *
  2677. * mark_inode_dirty(inode)
  2678. * stuff();
  2679. * inode->i_size = expr;
  2680. *
  2681. * is in error because a kswapd-driven write_inode() could occur while
  2682. * `stuff()' is running, and the new i_size will be lost. Plus the inode
  2683. * will no longer be on the superblock's dirty inode list.
  2684. */
  2685. int ext4_write_inode(struct inode *inode, int wait)
  2686. {
  2687. if (current->flags & PF_MEMALLOC)
  2688. return 0;
  2689. if (ext4_journal_current_handle()) {
  2690. jbd_debug(1, "called recursively, non-PF_MEMALLOC!\n");
  2691. dump_stack();
  2692. return -EIO;
  2693. }
  2694. if (!wait)
  2695. return 0;
  2696. return ext4_force_commit(inode->i_sb);
  2697. }
  2698. /*
  2699. * ext4_setattr()
  2700. *
  2701. * Called from notify_change.
  2702. *
  2703. * We want to trap VFS attempts to truncate the file as soon as
  2704. * possible. In particular, we want to make sure that when the VFS
  2705. * shrinks i_size, we put the inode on the orphan list and modify
  2706. * i_disksize immediately, so that during the subsequent flushing of
  2707. * dirty pages and freeing of disk blocks, we can guarantee that any
  2708. * commit will leave the blocks being flushed in an unused state on
  2709. * disk. (On recovery, the inode will get truncated and the blocks will
  2710. * be freed, so we have a strong guarantee that no future commit will
  2711. * leave these blocks visible to the user.)
  2712. *
  2713. * Called with inode->sem down.
  2714. */
  2715. int ext4_setattr(struct dentry *dentry, struct iattr *attr)
  2716. {
  2717. struct inode *inode = dentry->d_inode;
  2718. int error, rc = 0;
  2719. const unsigned int ia_valid = attr->ia_valid;
  2720. error = inode_change_ok(inode, attr);
  2721. if (error)
  2722. return error;
  2723. if ((ia_valid & ATTR_UID && attr->ia_uid != inode->i_uid) ||
  2724. (ia_valid & ATTR_GID && attr->ia_gid != inode->i_gid)) {
  2725. handle_t *handle;
  2726. /* (user+group)*(old+new) structure, inode write (sb,
  2727. * inode block, ? - but truncate inode update has it) */
  2728. handle = ext4_journal_start(inode, 2*(EXT4_QUOTA_INIT_BLOCKS(inode->i_sb)+
  2729. EXT4_QUOTA_DEL_BLOCKS(inode->i_sb))+3);
  2730. if (IS_ERR(handle)) {
  2731. error = PTR_ERR(handle);
  2732. goto err_out;
  2733. }
  2734. error = DQUOT_TRANSFER(inode, attr) ? -EDQUOT : 0;
  2735. if (error) {
  2736. ext4_journal_stop(handle);
  2737. return error;
  2738. }
  2739. /* Update corresponding info in inode so that everything is in
  2740. * one transaction */
  2741. if (attr->ia_valid & ATTR_UID)
  2742. inode->i_uid = attr->ia_uid;
  2743. if (attr->ia_valid & ATTR_GID)
  2744. inode->i_gid = attr->ia_gid;
  2745. error = ext4_mark_inode_dirty(handle, inode);
  2746. ext4_journal_stop(handle);
  2747. }
  2748. if (S_ISREG(inode->i_mode) &&
  2749. attr->ia_valid & ATTR_SIZE && attr->ia_size < inode->i_size) {
  2750. handle_t *handle;
  2751. handle = ext4_journal_start(inode, 3);
  2752. if (IS_ERR(handle)) {
  2753. error = PTR_ERR(handle);
  2754. goto err_out;
  2755. }
  2756. error = ext4_orphan_add(handle, inode);
  2757. EXT4_I(inode)->i_disksize = attr->ia_size;
  2758. rc = ext4_mark_inode_dirty(handle, inode);
  2759. if (!error)
  2760. error = rc;
  2761. ext4_journal_stop(handle);
  2762. }
  2763. rc = inode_setattr(inode, attr);
  2764. /* If inode_setattr's call to ext4_truncate failed to get a
  2765. * transaction handle at all, we need to clean up the in-core
  2766. * orphan list manually. */
  2767. if (inode->i_nlink)
  2768. ext4_orphan_del(NULL, inode);
  2769. if (!rc && (ia_valid & ATTR_MODE))
  2770. rc = ext4_acl_chmod(inode);
  2771. err_out:
  2772. ext4_std_error(inode->i_sb, error);
  2773. if (!error)
  2774. error = rc;
  2775. return error;
  2776. }
  2777. /*
  2778. * How many blocks doth make a writepage()?
  2779. *
  2780. * With N blocks per page, it may be:
  2781. * N data blocks
  2782. * 2 indirect block
  2783. * 2 dindirect
  2784. * 1 tindirect
  2785. * N+5 bitmap blocks (from the above)
  2786. * N+5 group descriptor summary blocks
  2787. * 1 inode block
  2788. * 1 superblock.
  2789. * 2 * EXT4_SINGLEDATA_TRANS_BLOCKS for the quote files
  2790. *
  2791. * 3 * (N + 5) + 2 + 2 * EXT4_SINGLEDATA_TRANS_BLOCKS
  2792. *
  2793. * With ordered or writeback data it's the same, less the N data blocks.
  2794. *
  2795. * If the inode's direct blocks can hold an integral number of pages then a
  2796. * page cannot straddle two indirect blocks, and we can only touch one indirect
  2797. * and dindirect block, and the "5" above becomes "3".
  2798. *
  2799. * This still overestimates under most circumstances. If we were to pass the
  2800. * start and end offsets in here as well we could do block_to_path() on each
  2801. * block and work out the exact number of indirects which are touched. Pah.
  2802. */
  2803. int ext4_writepage_trans_blocks(struct inode *inode)
  2804. {
  2805. int bpp = ext4_journal_blocks_per_page(inode);
  2806. int indirects = (EXT4_NDIR_BLOCKS % bpp) ? 5 : 3;
  2807. int ret;
  2808. if (EXT4_I(inode)->i_flags & EXT4_EXTENTS_FL)
  2809. return ext4_ext_writepage_trans_blocks(inode, bpp);
  2810. if (ext4_should_journal_data(inode))
  2811. ret = 3 * (bpp + indirects) + 2;
  2812. else
  2813. ret = 2 * (bpp + indirects) + 2;
  2814. #ifdef CONFIG_QUOTA
  2815. /* We know that structure was already allocated during DQUOT_INIT so
  2816. * we will be updating only the data blocks + inodes */
  2817. ret += 2*EXT4_QUOTA_TRANS_BLOCKS(inode->i_sb);
  2818. #endif
  2819. return ret;
  2820. }
  2821. /*
  2822. * The caller must have previously called ext4_reserve_inode_write().
  2823. * Give this, we know that the caller already has write access to iloc->bh.
  2824. */
  2825. int ext4_mark_iloc_dirty(handle_t *handle,
  2826. struct inode *inode, struct ext4_iloc *iloc)
  2827. {
  2828. int err = 0;
  2829. /* the do_update_inode consumes one bh->b_count */
  2830. get_bh(iloc->bh);
  2831. /* ext4_do_update_inode() does jbd2_journal_dirty_metadata */
  2832. err = ext4_do_update_inode(handle, inode, iloc);
  2833. put_bh(iloc->bh);
  2834. return err;
  2835. }
  2836. /*
  2837. * On success, We end up with an outstanding reference count against
  2838. * iloc->bh. This _must_ be cleaned up later.
  2839. */
  2840. int
  2841. ext4_reserve_inode_write(handle_t *handle, struct inode *inode,
  2842. struct ext4_iloc *iloc)
  2843. {
  2844. int err = 0;
  2845. if (handle) {
  2846. err = ext4_get_inode_loc(inode, iloc);
  2847. if (!err) {
  2848. BUFFER_TRACE(iloc->bh, "get_write_access");
  2849. err = ext4_journal_get_write_access(handle, iloc->bh);
  2850. if (err) {
  2851. brelse(iloc->bh);
  2852. iloc->bh = NULL;
  2853. }
  2854. }
  2855. }
  2856. ext4_std_error(inode->i_sb, err);
  2857. return err;
  2858. }
  2859. /*
  2860. * Expand an inode by new_extra_isize bytes.
  2861. * Returns 0 on success or negative error number on failure.
  2862. */
  2863. int ext4_expand_extra_isize(struct inode *inode, unsigned int new_extra_isize,
  2864. struct ext4_iloc iloc, handle_t *handle)
  2865. {
  2866. struct ext4_inode *raw_inode;
  2867. struct ext4_xattr_ibody_header *header;
  2868. struct ext4_xattr_entry *entry;
  2869. if (EXT4_I(inode)->i_extra_isize >= new_extra_isize)
  2870. return 0;
  2871. raw_inode = ext4_raw_inode(&iloc);
  2872. header = IHDR(inode, raw_inode);
  2873. entry = IFIRST(header);
  2874. /* No extended attributes present */
  2875. if (!(EXT4_I(inode)->i_state & EXT4_STATE_XATTR) ||
  2876. header->h_magic != cpu_to_le32(EXT4_XATTR_MAGIC)) {
  2877. memset((void *)raw_inode + EXT4_GOOD_OLD_INODE_SIZE, 0,
  2878. new_extra_isize);
  2879. EXT4_I(inode)->i_extra_isize = new_extra_isize;
  2880. return 0;
  2881. }
  2882. /* try to expand with EAs present */
  2883. return ext4_expand_extra_isize_ea(inode, new_extra_isize,
  2884. raw_inode, handle);
  2885. }
  2886. /*
  2887. * What we do here is to mark the in-core inode as clean with respect to inode
  2888. * dirtiness (it may still be data-dirty).
  2889. * This means that the in-core inode may be reaped by prune_icache
  2890. * without having to perform any I/O. This is a very good thing,
  2891. * because *any* task may call prune_icache - even ones which
  2892. * have a transaction open against a different journal.
  2893. *
  2894. * Is this cheating? Not really. Sure, we haven't written the
  2895. * inode out, but prune_icache isn't a user-visible syncing function.
  2896. * Whenever the user wants stuff synced (sys_sync, sys_msync, sys_fsync)
  2897. * we start and wait on commits.
  2898. *
  2899. * Is this efficient/effective? Well, we're being nice to the system
  2900. * by cleaning up our inodes proactively so they can be reaped
  2901. * without I/O. But we are potentially leaving up to five seconds'
  2902. * worth of inodes floating about which prune_icache wants us to
  2903. * write out. One way to fix that would be to get prune_icache()
  2904. * to do a write_super() to free up some memory. It has the desired
  2905. * effect.
  2906. */
  2907. int ext4_mark_inode_dirty(handle_t *handle, struct inode *inode)
  2908. {
  2909. struct ext4_iloc iloc;
  2910. struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
  2911. static unsigned int mnt_count;
  2912. int err, ret;
  2913. might_sleep();
  2914. err = ext4_reserve_inode_write(handle, inode, &iloc);
  2915. if (EXT4_I(inode)->i_extra_isize < sbi->s_want_extra_isize &&
  2916. !(EXT4_I(inode)->i_state & EXT4_STATE_NO_EXPAND)) {
  2917. /*
  2918. * We need extra buffer credits since we may write into EA block
  2919. * with this same handle. If journal_extend fails, then it will
  2920. * only result in a minor loss of functionality for that inode.
  2921. * If this is felt to be critical, then e2fsck should be run to
  2922. * force a large enough s_min_extra_isize.
  2923. */
  2924. if ((jbd2_journal_extend(handle,
  2925. EXT4_DATA_TRANS_BLOCKS(inode->i_sb))) == 0) {
  2926. ret = ext4_expand_extra_isize(inode,
  2927. sbi->s_want_extra_isize,
  2928. iloc, handle);
  2929. if (ret) {
  2930. EXT4_I(inode)->i_state |= EXT4_STATE_NO_EXPAND;
  2931. if (mnt_count !=
  2932. le16_to_cpu(sbi->s_es->s_mnt_count)) {
  2933. ext4_warning(inode->i_sb, __FUNCTION__,
  2934. "Unable to expand inode %lu. Delete"
  2935. " some EAs or run e2fsck.",
  2936. inode->i_ino);
  2937. mnt_count =
  2938. le16_to_cpu(sbi->s_es->s_mnt_count);
  2939. }
  2940. }
  2941. }
  2942. }
  2943. if (!err)
  2944. err = ext4_mark_iloc_dirty(handle, inode, &iloc);
  2945. return err;
  2946. }
  2947. /*
  2948. * ext4_dirty_inode() is called from __mark_inode_dirty()
  2949. *
  2950. * We're really interested in the case where a file is being extended.
  2951. * i_size has been changed by generic_commit_write() and we thus need
  2952. * to include the updated inode in the current transaction.
  2953. *
  2954. * Also, DQUOT_ALLOC_SPACE() will always dirty the inode when blocks
  2955. * are allocated to the file.
  2956. *
  2957. * If the inode is marked synchronous, we don't honour that here - doing
  2958. * so would cause a commit on atime updates, which we don't bother doing.
  2959. * We handle synchronous inodes at the highest possible level.
  2960. */
  2961. void ext4_dirty_inode(struct inode *inode)
  2962. {
  2963. handle_t *current_handle = ext4_journal_current_handle();
  2964. handle_t *handle;
  2965. handle = ext4_journal_start(inode, 2);
  2966. if (IS_ERR(handle))
  2967. goto out;
  2968. if (current_handle &&
  2969. current_handle->h_transaction != handle->h_transaction) {
  2970. /* This task has a transaction open against a different fs */
  2971. printk(KERN_EMERG "%s: transactions do not match!\n",
  2972. __FUNCTION__);
  2973. } else {
  2974. jbd_debug(5, "marking dirty. outer handle=%p\n",
  2975. current_handle);
  2976. ext4_mark_inode_dirty(handle, inode);
  2977. }
  2978. ext4_journal_stop(handle);
  2979. out:
  2980. return;
  2981. }
  2982. #if 0
  2983. /*
  2984. * Bind an inode's backing buffer_head into this transaction, to prevent
  2985. * it from being flushed to disk early. Unlike
  2986. * ext4_reserve_inode_write, this leaves behind no bh reference and
  2987. * returns no iloc structure, so the caller needs to repeat the iloc
  2988. * lookup to mark the inode dirty later.
  2989. */
  2990. static int ext4_pin_inode(handle_t *handle, struct inode *inode)
  2991. {
  2992. struct ext4_iloc iloc;
  2993. int err = 0;
  2994. if (handle) {
  2995. err = ext4_get_inode_loc(inode, &iloc);
  2996. if (!err) {
  2997. BUFFER_TRACE(iloc.bh, "get_write_access");
  2998. err = jbd2_journal_get_write_access(handle, iloc.bh);
  2999. if (!err)
  3000. err = ext4_journal_dirty_metadata(handle,
  3001. iloc.bh);
  3002. brelse(iloc.bh);
  3003. }
  3004. }
  3005. ext4_std_error(inode->i_sb, err);
  3006. return err;
  3007. }
  3008. #endif
  3009. int ext4_change_inode_journal_flag(struct inode *inode, int val)
  3010. {
  3011. journal_t *journal;
  3012. handle_t *handle;
  3013. int err;
  3014. /*
  3015. * We have to be very careful here: changing a data block's
  3016. * journaling status dynamically is dangerous. If we write a
  3017. * data block to the journal, change the status and then delete
  3018. * that block, we risk forgetting to revoke the old log record
  3019. * from the journal and so a subsequent replay can corrupt data.
  3020. * So, first we make sure that the journal is empty and that
  3021. * nobody is changing anything.
  3022. */
  3023. journal = EXT4_JOURNAL(inode);
  3024. if (is_journal_aborted(journal))
  3025. return -EROFS;
  3026. jbd2_journal_lock_updates(journal);
  3027. jbd2_journal_flush(journal);
  3028. /*
  3029. * OK, there are no updates running now, and all cached data is
  3030. * synced to disk. We are now in a completely consistent state
  3031. * which doesn't have anything in the journal, and we know that
  3032. * no filesystem updates are running, so it is safe to modify
  3033. * the inode's in-core data-journaling state flag now.
  3034. */
  3035. if (val)
  3036. EXT4_I(inode)->i_flags |= EXT4_JOURNAL_DATA_FL;
  3037. else
  3038. EXT4_I(inode)->i_flags &= ~EXT4_JOURNAL_DATA_FL;
  3039. ext4_set_aops(inode);
  3040. jbd2_journal_unlock_updates(journal);
  3041. /* Finally we can mark the inode as dirty. */
  3042. handle = ext4_journal_start(inode, 1);
  3043. if (IS_ERR(handle))
  3044. return PTR_ERR(handle);
  3045. err = ext4_mark_inode_dirty(handle, inode);
  3046. handle->h_sync = 1;
  3047. ext4_journal_stop(handle);
  3048. ext4_std_error(inode->i_sb, err);
  3049. return err;
  3050. }