xfs_inode.c 137 KB

123456789101112131415161718192021222324252627282930313233343536373839404142434445464748495051525354555657585960616263646566676869707172737475767778798081828384858687888990919293949596979899100101102103104105106107108109110111112113114115116117118119120121122123124125126127128129130131132133134135136137138139140141142143144145146147148149150151152153154155156157158159160161162163164165166167168169170171172173174175176177178179180181182183184185186187188189190191192193194195196197198199200201202203204205206207208209210211212213214215216217218219220221222223224225226227228229230231232233234235236237238239240241242243244245246247248249250251252253254255256257258259260261262263264265266267268269270271272273274275276277278279280281282283284285286287288289290291292293294295296297298299300301302303304305306307308309310311312313314315316317318319320321322323324325326327328329330331332333334335336337338339340341342343344345346347348349350351352353354355356357358359360361362363364365366367368369370371372373374375376377378379380381382383384385386387388389390391392393394395396397398399400401402403404405406407408409410411412413414415416417418419420421422423424425426427428429430431432433434435436437438439440441442443444445446447448449450451452453454455456457458459460461462463464465466467468469470471472473474475476477478479480481482483484485486487488489490491492493494495496497498499500501502503504505506507508509510511512513514515516517518519520521522523524525526527528529530531532533534535536537538539540541542543544545546547548549550551552553554555556557558559560561562563564565566567568569570571572573574575576577578579580581582583584585586587588589590591592593594595596597598599600601602603604605606607608609610611612613614615616617618619620621622623624625626627628629630631632633634635636637638639640641642643644645646647648649650651652653654655656657658659660661662663664665666667668669670671672673674675676677678679680681682683684685686687688689690691692693694695696697698699700701702703704705706707708709710711712713714715716717718719720721722723724725726727728729730731732733734735736737738739740741742743744745746747748749750751752753754755756757758759760761762763764765766767768769770771772773774775776777778779780781782783784785786787788789790791792793794795796797798799800801802803804805806807808809810811812813814815816817818819820821822823824825826827828829830831832833834835836837838839840841842843844845846847848849850851852853854855856857858859860861862863864865866867868869870871872873874875876877878879880881882883884885886887888889890891892893894895896897898899900901902903904905906907908909910911912913914915916917918919920921922923924925926927928929930931932933934935936937938939940941942943944945946947948949950951952953954955956957958959960961962963964965966967968969970971972973974975976977978979980981982983984985986987988989990991992993994995996997998999100010011002100310041005100610071008100910101011101210131014101510161017101810191020102110221023102410251026102710281029103010311032103310341035103610371038103910401041104210431044104510461047104810491050105110521053105410551056105710581059106010611062106310641065106610671068106910701071107210731074107510761077107810791080108110821083108410851086108710881089109010911092109310941095109610971098109911001101110211031104110511061107110811091110111111121113111411151116111711181119112011211122112311241125112611271128112911301131113211331134113511361137113811391140114111421143114411451146114711481149115011511152115311541155115611571158115911601161116211631164116511661167116811691170117111721173117411751176117711781179118011811182118311841185118611871188118911901191119211931194119511961197119811991200120112021203120412051206120712081209121012111212121312141215121612171218121912201221122212231224122512261227122812291230123112321233123412351236123712381239124012411242124312441245124612471248124912501251125212531254125512561257125812591260126112621263126412651266126712681269127012711272127312741275127612771278127912801281128212831284128512861287128812891290129112921293129412951296129712981299130013011302130313041305130613071308130913101311131213131314131513161317131813191320132113221323132413251326132713281329133013311332133313341335133613371338133913401341134213431344134513461347134813491350135113521353135413551356135713581359136013611362136313641365136613671368136913701371137213731374137513761377137813791380138113821383138413851386138713881389139013911392139313941395139613971398139914001401140214031404140514061407140814091410141114121413141414151416141714181419142014211422142314241425142614271428142914301431143214331434143514361437143814391440144114421443144414451446144714481449145014511452145314541455145614571458145914601461146214631464146514661467146814691470147114721473147414751476147714781479148014811482148314841485148614871488148914901491149214931494149514961497149814991500150115021503150415051506150715081509151015111512151315141515151615171518151915201521152215231524152515261527152815291530153115321533153415351536153715381539154015411542154315441545154615471548154915501551155215531554155515561557155815591560156115621563156415651566156715681569157015711572157315741575157615771578157915801581158215831584158515861587158815891590159115921593159415951596159715981599160016011602160316041605160616071608160916101611161216131614161516161617161816191620162116221623162416251626162716281629163016311632163316341635163616371638163916401641164216431644164516461647164816491650165116521653165416551656165716581659166016611662166316641665166616671668166916701671167216731674167516761677167816791680168116821683168416851686168716881689169016911692169316941695169616971698169917001701170217031704170517061707170817091710171117121713171417151716171717181719172017211722172317241725172617271728172917301731173217331734173517361737173817391740174117421743174417451746174717481749175017511752175317541755175617571758175917601761176217631764176517661767176817691770177117721773177417751776177717781779178017811782178317841785178617871788178917901791179217931794179517961797179817991800180118021803180418051806180718081809181018111812181318141815181618171818181918201821182218231824182518261827182818291830183118321833183418351836183718381839184018411842184318441845184618471848184918501851185218531854185518561857185818591860186118621863186418651866186718681869187018711872187318741875187618771878187918801881188218831884188518861887188818891890189118921893189418951896189718981899190019011902190319041905190619071908190919101911191219131914191519161917191819191920192119221923192419251926192719281929193019311932193319341935193619371938193919401941194219431944194519461947194819491950195119521953195419551956195719581959196019611962196319641965196619671968196919701971197219731974197519761977197819791980198119821983198419851986198719881989199019911992199319941995199619971998199920002001200220032004200520062007200820092010201120122013201420152016201720182019202020212022202320242025202620272028202920302031203220332034203520362037203820392040204120422043204420452046204720482049205020512052205320542055205620572058205920602061206220632064206520662067206820692070207120722073207420752076207720782079208020812082208320842085208620872088208920902091209220932094209520962097209820992100210121022103210421052106210721082109211021112112211321142115211621172118211921202121212221232124212521262127212821292130213121322133213421352136213721382139214021412142214321442145214621472148214921502151215221532154215521562157215821592160216121622163216421652166216721682169217021712172217321742175217621772178217921802181218221832184218521862187218821892190219121922193219421952196219721982199220022012202220322042205220622072208220922102211221222132214221522162217221822192220222122222223222422252226222722282229223022312232223322342235223622372238223922402241224222432244224522462247224822492250225122522253225422552256225722582259226022612262226322642265226622672268226922702271227222732274227522762277227822792280228122822283228422852286228722882289229022912292229322942295229622972298229923002301230223032304230523062307230823092310231123122313231423152316231723182319232023212322232323242325232623272328232923302331233223332334233523362337233823392340234123422343234423452346234723482349235023512352235323542355235623572358235923602361236223632364236523662367236823692370237123722373237423752376237723782379238023812382238323842385238623872388238923902391239223932394239523962397239823992400240124022403240424052406240724082409241024112412241324142415241624172418241924202421242224232424242524262427242824292430243124322433243424352436243724382439244024412442244324442445244624472448244924502451245224532454245524562457245824592460246124622463246424652466246724682469247024712472247324742475247624772478247924802481248224832484248524862487248824892490249124922493249424952496249724982499250025012502250325042505250625072508250925102511251225132514251525162517251825192520252125222523252425252526252725282529253025312532253325342535253625372538253925402541254225432544254525462547254825492550255125522553255425552556255725582559256025612562256325642565256625672568256925702571257225732574257525762577257825792580258125822583258425852586258725882589259025912592259325942595259625972598259926002601260226032604260526062607260826092610261126122613261426152616261726182619262026212622262326242625262626272628262926302631263226332634263526362637263826392640264126422643264426452646264726482649265026512652265326542655265626572658265926602661266226632664266526662667266826692670267126722673267426752676267726782679268026812682268326842685268626872688268926902691269226932694269526962697269826992700270127022703270427052706270727082709271027112712271327142715271627172718271927202721272227232724272527262727272827292730273127322733273427352736273727382739274027412742274327442745274627472748274927502751275227532754275527562757275827592760276127622763276427652766276727682769277027712772277327742775277627772778277927802781278227832784278527862787278827892790279127922793279427952796279727982799280028012802280328042805280628072808280928102811281228132814281528162817281828192820282128222823282428252826282728282829283028312832283328342835283628372838283928402841284228432844284528462847284828492850285128522853285428552856285728582859286028612862286328642865286628672868286928702871287228732874287528762877287828792880288128822883288428852886288728882889289028912892289328942895289628972898289929002901290229032904290529062907290829092910291129122913291429152916291729182919292029212922292329242925292629272928292929302931293229332934293529362937293829392940294129422943294429452946294729482949295029512952295329542955295629572958295929602961296229632964296529662967296829692970297129722973297429752976297729782979298029812982298329842985298629872988298929902991299229932994299529962997299829993000300130023003300430053006300730083009301030113012301330143015301630173018301930203021302230233024302530263027302830293030303130323033303430353036303730383039304030413042304330443045304630473048304930503051305230533054305530563057305830593060306130623063306430653066306730683069307030713072307330743075307630773078307930803081308230833084308530863087308830893090309130923093309430953096309730983099310031013102310331043105310631073108310931103111311231133114311531163117311831193120312131223123312431253126312731283129313031313132313331343135313631373138313931403141314231433144314531463147314831493150315131523153315431553156315731583159316031613162316331643165316631673168316931703171317231733174317531763177317831793180318131823183318431853186318731883189319031913192319331943195319631973198319932003201320232033204320532063207320832093210321132123213321432153216321732183219322032213222322332243225322632273228322932303231323232333234323532363237323832393240324132423243324432453246324732483249325032513252325332543255325632573258325932603261326232633264326532663267326832693270327132723273327432753276327732783279328032813282328332843285328632873288328932903291329232933294329532963297329832993300330133023303330433053306330733083309331033113312331333143315331633173318331933203321332233233324332533263327332833293330333133323333333433353336333733383339334033413342334333443345334633473348334933503351335233533354335533563357335833593360336133623363336433653366336733683369337033713372337333743375337633773378337933803381338233833384338533863387338833893390339133923393339433953396339733983399340034013402340334043405340634073408340934103411341234133414341534163417341834193420342134223423342434253426342734283429343034313432343334343435343634373438343934403441344234433444344534463447344834493450345134523453345434553456345734583459346034613462346334643465346634673468346934703471347234733474347534763477347834793480348134823483348434853486348734883489349034913492349334943495349634973498349935003501350235033504350535063507350835093510351135123513351435153516351735183519352035213522352335243525352635273528352935303531353235333534353535363537353835393540354135423543354435453546354735483549355035513552355335543555355635573558355935603561356235633564356535663567356835693570357135723573357435753576357735783579358035813582358335843585358635873588358935903591359235933594359535963597359835993600360136023603360436053606360736083609361036113612361336143615361636173618361936203621362236233624362536263627362836293630363136323633363436353636363736383639364036413642364336443645364636473648364936503651365236533654365536563657365836593660366136623663366436653666366736683669367036713672367336743675367636773678367936803681368236833684368536863687368836893690369136923693369436953696369736983699370037013702370337043705370637073708370937103711371237133714371537163717371837193720372137223723372437253726372737283729373037313732373337343735373637373738373937403741374237433744374537463747374837493750375137523753375437553756375737583759376037613762376337643765376637673768376937703771377237733774377537763777377837793780378137823783378437853786378737883789379037913792379337943795379637973798379938003801380238033804380538063807380838093810381138123813381438153816381738183819382038213822382338243825382638273828382938303831383238333834383538363837383838393840384138423843384438453846384738483849385038513852385338543855385638573858385938603861386238633864386538663867386838693870387138723873387438753876387738783879388038813882388338843885388638873888388938903891389238933894389538963897389838993900390139023903390439053906390739083909391039113912391339143915391639173918391939203921392239233924392539263927392839293930393139323933393439353936393739383939394039413942394339443945394639473948394939503951395239533954395539563957395839593960396139623963396439653966396739683969397039713972397339743975397639773978397939803981398239833984398539863987398839893990399139923993399439953996399739983999400040014002400340044005400640074008400940104011401240134014401540164017401840194020402140224023402440254026402740284029403040314032403340344035403640374038403940404041404240434044404540464047404840494050405140524053405440554056405740584059406040614062406340644065406640674068406940704071407240734074407540764077407840794080408140824083408440854086408740884089409040914092409340944095409640974098409941004101410241034104410541064107410841094110411141124113411441154116411741184119412041214122412341244125412641274128412941304131413241334134413541364137413841394140414141424143414441454146414741484149415041514152415341544155415641574158415941604161416241634164416541664167416841694170417141724173417441754176417741784179418041814182418341844185418641874188418941904191419241934194419541964197419841994200420142024203420442054206420742084209421042114212421342144215421642174218421942204221422242234224422542264227422842294230423142324233423442354236423742384239424042414242424342444245424642474248424942504251425242534254425542564257425842594260426142624263426442654266426742684269427042714272427342744275427642774278427942804281428242834284428542864287428842894290429142924293429442954296429742984299430043014302430343044305430643074308430943104311431243134314431543164317431843194320432143224323432443254326432743284329433043314332433343344335433643374338433943404341434243434344434543464347434843494350435143524353435443554356435743584359436043614362436343644365436643674368436943704371437243734374437543764377437843794380438143824383438443854386438743884389439043914392439343944395439643974398439944004401440244034404440544064407440844094410441144124413441444154416441744184419442044214422442344244425442644274428442944304431443244334434443544364437443844394440444144424443444444454446444744484449445044514452445344544455445644574458445944604461446244634464446544664467446844694470447144724473447444754476447744784479448044814482448344844485448644874488448944904491449244934494449544964497449844994500450145024503450445054506450745084509451045114512451345144515451645174518451945204521452245234524452545264527452845294530453145324533453445354536453745384539454045414542454345444545454645474548454945504551455245534554455545564557455845594560456145624563456445654566456745684569457045714572457345744575457645774578457945804581458245834584458545864587458845894590459145924593459445954596459745984599460046014602460346044605460646074608460946104611461246134614461546164617461846194620462146224623462446254626462746284629463046314632463346344635463646374638463946404641464246434644464546464647464846494650465146524653465446554656465746584659466046614662466346644665466646674668466946704671467246734674467546764677467846794680468146824683468446854686468746884689469046914692469346944695469646974698469947004701470247034704470547064707470847094710471147124713471447154716471747184719472047214722472347244725472647274728472947304731473247334734473547364737473847394740474147424743474447454746474747484749475047514752475347544755475647574758475947604761
  1. /*
  2. * Copyright (c) 2000-2006 Silicon Graphics, Inc.
  3. * All Rights Reserved.
  4. *
  5. * This program is free software; you can redistribute it and/or
  6. * modify it under the terms of the GNU General Public License as
  7. * published by the Free Software Foundation.
  8. *
  9. * This program is distributed in the hope that it would be useful,
  10. * but WITHOUT ANY WARRANTY; without even the implied warranty of
  11. * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
  12. * GNU General Public License for more details.
  13. *
  14. * You should have received a copy of the GNU General Public License
  15. * along with this program; if not, write the Free Software Foundation,
  16. * Inc., 51 Franklin St, Fifth Floor, Boston, MA 02110-1301 USA
  17. */
  18. #include "xfs.h"
  19. #include "xfs_fs.h"
  20. #include "xfs_types.h"
  21. #include "xfs_bit.h"
  22. #include "xfs_log.h"
  23. #include "xfs_inum.h"
  24. #include "xfs_imap.h"
  25. #include "xfs_trans.h"
  26. #include "xfs_trans_priv.h"
  27. #include "xfs_sb.h"
  28. #include "xfs_ag.h"
  29. #include "xfs_dir2.h"
  30. #include "xfs_dmapi.h"
  31. #include "xfs_mount.h"
  32. #include "xfs_bmap_btree.h"
  33. #include "xfs_alloc_btree.h"
  34. #include "xfs_ialloc_btree.h"
  35. #include "xfs_dir2_sf.h"
  36. #include "xfs_attr_sf.h"
  37. #include "xfs_dinode.h"
  38. #include "xfs_inode.h"
  39. #include "xfs_buf_item.h"
  40. #include "xfs_inode_item.h"
  41. #include "xfs_btree.h"
  42. #include "xfs_alloc.h"
  43. #include "xfs_ialloc.h"
  44. #include "xfs_bmap.h"
  45. #include "xfs_rw.h"
  46. #include "xfs_error.h"
  47. #include "xfs_utils.h"
  48. #include "xfs_dir2_trace.h"
  49. #include "xfs_quota.h"
  50. #include "xfs_acl.h"
  51. kmem_zone_t *xfs_ifork_zone;
  52. kmem_zone_t *xfs_inode_zone;
  53. kmem_zone_t *xfs_chashlist_zone;
  54. /*
  55. * Used in xfs_itruncate(). This is the maximum number of extents
  56. * freed from a file in a single transaction.
  57. */
  58. #define XFS_ITRUNC_MAX_EXTENTS 2
  59. STATIC int xfs_iflush_int(xfs_inode_t *, xfs_buf_t *);
  60. STATIC int xfs_iformat_local(xfs_inode_t *, xfs_dinode_t *, int, int);
  61. STATIC int xfs_iformat_extents(xfs_inode_t *, xfs_dinode_t *, int);
  62. STATIC int xfs_iformat_btree(xfs_inode_t *, xfs_dinode_t *, int);
  63. #ifdef DEBUG
  64. /*
  65. * Make sure that the extents in the given memory buffer
  66. * are valid.
  67. */
  68. STATIC void
  69. xfs_validate_extents(
  70. xfs_ifork_t *ifp,
  71. int nrecs,
  72. int disk,
  73. xfs_exntfmt_t fmt)
  74. {
  75. xfs_bmbt_rec_t *ep;
  76. xfs_bmbt_irec_t irec;
  77. xfs_bmbt_rec_t rec;
  78. int i;
  79. for (i = 0; i < nrecs; i++) {
  80. ep = xfs_iext_get_ext(ifp, i);
  81. rec.l0 = get_unaligned((__uint64_t*)&ep->l0);
  82. rec.l1 = get_unaligned((__uint64_t*)&ep->l1);
  83. if (disk)
  84. xfs_bmbt_disk_get_all(&rec, &irec);
  85. else
  86. xfs_bmbt_get_all(&rec, &irec);
  87. if (fmt == XFS_EXTFMT_NOSTATE)
  88. ASSERT(irec.br_state == XFS_EXT_NORM);
  89. }
  90. }
  91. #else /* DEBUG */
  92. #define xfs_validate_extents(ifp, nrecs, disk, fmt)
  93. #endif /* DEBUG */
  94. /*
  95. * Check that none of the inode's in the buffer have a next
  96. * unlinked field of 0.
  97. */
  98. #if defined(DEBUG)
  99. void
  100. xfs_inobp_check(
  101. xfs_mount_t *mp,
  102. xfs_buf_t *bp)
  103. {
  104. int i;
  105. int j;
  106. xfs_dinode_t *dip;
  107. j = mp->m_inode_cluster_size >> mp->m_sb.sb_inodelog;
  108. for (i = 0; i < j; i++) {
  109. dip = (xfs_dinode_t *)xfs_buf_offset(bp,
  110. i * mp->m_sb.sb_inodesize);
  111. if (!dip->di_next_unlinked) {
  112. xfs_fs_cmn_err(CE_ALERT, mp,
  113. "Detected a bogus zero next_unlinked field in incore inode buffer 0x%p. About to pop an ASSERT.",
  114. bp);
  115. ASSERT(dip->di_next_unlinked);
  116. }
  117. }
  118. }
  119. #endif
  120. /*
  121. * This routine is called to map an inode number within a file
  122. * system to the buffer containing the on-disk version of the
  123. * inode. It returns a pointer to the buffer containing the
  124. * on-disk inode in the bpp parameter, and in the dip parameter
  125. * it returns a pointer to the on-disk inode within that buffer.
  126. *
  127. * If a non-zero error is returned, then the contents of bpp and
  128. * dipp are undefined.
  129. *
  130. * Use xfs_imap() to determine the size and location of the
  131. * buffer to read from disk.
  132. */
  133. STATIC int
  134. xfs_inotobp(
  135. xfs_mount_t *mp,
  136. xfs_trans_t *tp,
  137. xfs_ino_t ino,
  138. xfs_dinode_t **dipp,
  139. xfs_buf_t **bpp,
  140. int *offset)
  141. {
  142. int di_ok;
  143. xfs_imap_t imap;
  144. xfs_buf_t *bp;
  145. int error;
  146. xfs_dinode_t *dip;
  147. /*
  148. * Call the space management code to find the location of the
  149. * inode on disk.
  150. */
  151. imap.im_blkno = 0;
  152. error = xfs_imap(mp, tp, ino, &imap, XFS_IMAP_LOOKUP);
  153. if (error != 0) {
  154. cmn_err(CE_WARN,
  155. "xfs_inotobp: xfs_imap() returned an "
  156. "error %d on %s. Returning error.", error, mp->m_fsname);
  157. return error;
  158. }
  159. /*
  160. * If the inode number maps to a block outside the bounds of the
  161. * file system then return NULL rather than calling read_buf
  162. * and panicing when we get an error from the driver.
  163. */
  164. if ((imap.im_blkno + imap.im_len) >
  165. XFS_FSB_TO_BB(mp, mp->m_sb.sb_dblocks)) {
  166. cmn_err(CE_WARN,
  167. "xfs_inotobp: inode number (%llu + %d) maps to a block outside the bounds "
  168. "of the file system %s. Returning EINVAL.",
  169. (unsigned long long)imap.im_blkno,
  170. imap.im_len, mp->m_fsname);
  171. return XFS_ERROR(EINVAL);
  172. }
  173. /*
  174. * Read in the buffer. If tp is NULL, xfs_trans_read_buf() will
  175. * default to just a read_buf() call.
  176. */
  177. error = xfs_trans_read_buf(mp, tp, mp->m_ddev_targp, imap.im_blkno,
  178. (int)imap.im_len, XFS_BUF_LOCK, &bp);
  179. if (error) {
  180. cmn_err(CE_WARN,
  181. "xfs_inotobp: xfs_trans_read_buf() returned an "
  182. "error %d on %s. Returning error.", error, mp->m_fsname);
  183. return error;
  184. }
  185. dip = (xfs_dinode_t *)xfs_buf_offset(bp, 0);
  186. di_ok =
  187. INT_GET(dip->di_core.di_magic, ARCH_CONVERT) == XFS_DINODE_MAGIC &&
  188. XFS_DINODE_GOOD_VERSION(INT_GET(dip->di_core.di_version, ARCH_CONVERT));
  189. if (unlikely(XFS_TEST_ERROR(!di_ok, mp, XFS_ERRTAG_ITOBP_INOTOBP,
  190. XFS_RANDOM_ITOBP_INOTOBP))) {
  191. XFS_CORRUPTION_ERROR("xfs_inotobp", XFS_ERRLEVEL_LOW, mp, dip);
  192. xfs_trans_brelse(tp, bp);
  193. cmn_err(CE_WARN,
  194. "xfs_inotobp: XFS_TEST_ERROR() returned an "
  195. "error on %s. Returning EFSCORRUPTED.", mp->m_fsname);
  196. return XFS_ERROR(EFSCORRUPTED);
  197. }
  198. xfs_inobp_check(mp, bp);
  199. /*
  200. * Set *dipp to point to the on-disk inode in the buffer.
  201. */
  202. *dipp = (xfs_dinode_t *)xfs_buf_offset(bp, imap.im_boffset);
  203. *bpp = bp;
  204. *offset = imap.im_boffset;
  205. return 0;
  206. }
  207. /*
  208. * This routine is called to map an inode to the buffer containing
  209. * the on-disk version of the inode. It returns a pointer to the
  210. * buffer containing the on-disk inode in the bpp parameter, and in
  211. * the dip parameter it returns a pointer to the on-disk inode within
  212. * that buffer.
  213. *
  214. * If a non-zero error is returned, then the contents of bpp and
  215. * dipp are undefined.
  216. *
  217. * If the inode is new and has not yet been initialized, use xfs_imap()
  218. * to determine the size and location of the buffer to read from disk.
  219. * If the inode has already been mapped to its buffer and read in once,
  220. * then use the mapping information stored in the inode rather than
  221. * calling xfs_imap(). This allows us to avoid the overhead of looking
  222. * at the inode btree for small block file systems (see xfs_dilocate()).
  223. * We can tell whether the inode has been mapped in before by comparing
  224. * its disk block address to 0. Only uninitialized inodes will have
  225. * 0 for the disk block address.
  226. */
  227. int
  228. xfs_itobp(
  229. xfs_mount_t *mp,
  230. xfs_trans_t *tp,
  231. xfs_inode_t *ip,
  232. xfs_dinode_t **dipp,
  233. xfs_buf_t **bpp,
  234. xfs_daddr_t bno,
  235. uint imap_flags)
  236. {
  237. xfs_imap_t imap;
  238. xfs_buf_t *bp;
  239. int error;
  240. int i;
  241. int ni;
  242. if (ip->i_blkno == (xfs_daddr_t)0) {
  243. /*
  244. * Call the space management code to find the location of the
  245. * inode on disk.
  246. */
  247. imap.im_blkno = bno;
  248. if ((error = xfs_imap(mp, tp, ip->i_ino, &imap,
  249. XFS_IMAP_LOOKUP | imap_flags)))
  250. return error;
  251. /*
  252. * If the inode number maps to a block outside the bounds
  253. * of the file system then return NULL rather than calling
  254. * read_buf and panicing when we get an error from the
  255. * driver.
  256. */
  257. if ((imap.im_blkno + imap.im_len) >
  258. XFS_FSB_TO_BB(mp, mp->m_sb.sb_dblocks)) {
  259. #ifdef DEBUG
  260. xfs_fs_cmn_err(CE_ALERT, mp, "xfs_itobp: "
  261. "(imap.im_blkno (0x%llx) "
  262. "+ imap.im_len (0x%llx)) > "
  263. " XFS_FSB_TO_BB(mp, "
  264. "mp->m_sb.sb_dblocks) (0x%llx)",
  265. (unsigned long long) imap.im_blkno,
  266. (unsigned long long) imap.im_len,
  267. XFS_FSB_TO_BB(mp, mp->m_sb.sb_dblocks));
  268. #endif /* DEBUG */
  269. return XFS_ERROR(EINVAL);
  270. }
  271. /*
  272. * Fill in the fields in the inode that will be used to
  273. * map the inode to its buffer from now on.
  274. */
  275. ip->i_blkno = imap.im_blkno;
  276. ip->i_len = imap.im_len;
  277. ip->i_boffset = imap.im_boffset;
  278. } else {
  279. /*
  280. * We've already mapped the inode once, so just use the
  281. * mapping that we saved the first time.
  282. */
  283. imap.im_blkno = ip->i_blkno;
  284. imap.im_len = ip->i_len;
  285. imap.im_boffset = ip->i_boffset;
  286. }
  287. ASSERT(bno == 0 || bno == imap.im_blkno);
  288. /*
  289. * Read in the buffer. If tp is NULL, xfs_trans_read_buf() will
  290. * default to just a read_buf() call.
  291. */
  292. error = xfs_trans_read_buf(mp, tp, mp->m_ddev_targp, imap.im_blkno,
  293. (int)imap.im_len, XFS_BUF_LOCK, &bp);
  294. if (error) {
  295. #ifdef DEBUG
  296. xfs_fs_cmn_err(CE_ALERT, mp, "xfs_itobp: "
  297. "xfs_trans_read_buf() returned error %d, "
  298. "imap.im_blkno 0x%llx, imap.im_len 0x%llx",
  299. error, (unsigned long long) imap.im_blkno,
  300. (unsigned long long) imap.im_len);
  301. #endif /* DEBUG */
  302. return error;
  303. }
  304. /*
  305. * Validate the magic number and version of every inode in the buffer
  306. * (if DEBUG kernel) or the first inode in the buffer, otherwise.
  307. * No validation is done here in userspace (xfs_repair).
  308. */
  309. #if !defined(__KERNEL__)
  310. ni = 0;
  311. #elif defined(DEBUG)
  312. ni = BBTOB(imap.im_len) >> mp->m_sb.sb_inodelog;
  313. #else /* usual case */
  314. ni = 1;
  315. #endif
  316. for (i = 0; i < ni; i++) {
  317. int di_ok;
  318. xfs_dinode_t *dip;
  319. dip = (xfs_dinode_t *)xfs_buf_offset(bp,
  320. (i << mp->m_sb.sb_inodelog));
  321. di_ok = INT_GET(dip->di_core.di_magic, ARCH_CONVERT) == XFS_DINODE_MAGIC &&
  322. XFS_DINODE_GOOD_VERSION(INT_GET(dip->di_core.di_version, ARCH_CONVERT));
  323. if (unlikely(XFS_TEST_ERROR(!di_ok, mp,
  324. XFS_ERRTAG_ITOBP_INOTOBP,
  325. XFS_RANDOM_ITOBP_INOTOBP))) {
  326. if (imap_flags & XFS_IMAP_BULKSTAT) {
  327. xfs_trans_brelse(tp, bp);
  328. return XFS_ERROR(EINVAL);
  329. }
  330. #ifdef DEBUG
  331. cmn_err(CE_ALERT,
  332. "Device %s - bad inode magic/vsn "
  333. "daddr %lld #%d (magic=%x)",
  334. XFS_BUFTARG_NAME(mp->m_ddev_targp),
  335. (unsigned long long)imap.im_blkno, i,
  336. INT_GET(dip->di_core.di_magic, ARCH_CONVERT));
  337. #endif
  338. XFS_CORRUPTION_ERROR("xfs_itobp", XFS_ERRLEVEL_HIGH,
  339. mp, dip);
  340. xfs_trans_brelse(tp, bp);
  341. return XFS_ERROR(EFSCORRUPTED);
  342. }
  343. }
  344. xfs_inobp_check(mp, bp);
  345. /*
  346. * Mark the buffer as an inode buffer now that it looks good
  347. */
  348. XFS_BUF_SET_VTYPE(bp, B_FS_INO);
  349. /*
  350. * Set *dipp to point to the on-disk inode in the buffer.
  351. */
  352. *dipp = (xfs_dinode_t *)xfs_buf_offset(bp, imap.im_boffset);
  353. *bpp = bp;
  354. return 0;
  355. }
  356. /*
  357. * Move inode type and inode format specific information from the
  358. * on-disk inode to the in-core inode. For fifos, devs, and sockets
  359. * this means set if_rdev to the proper value. For files, directories,
  360. * and symlinks this means to bring in the in-line data or extent
  361. * pointers. For a file in B-tree format, only the root is immediately
  362. * brought in-core. The rest will be in-lined in if_extents when it
  363. * is first referenced (see xfs_iread_extents()).
  364. */
  365. STATIC int
  366. xfs_iformat(
  367. xfs_inode_t *ip,
  368. xfs_dinode_t *dip)
  369. {
  370. xfs_attr_shortform_t *atp;
  371. int size;
  372. int error;
  373. xfs_fsize_t di_size;
  374. ip->i_df.if_ext_max =
  375. XFS_IFORK_DSIZE(ip) / (uint)sizeof(xfs_bmbt_rec_t);
  376. error = 0;
  377. if (unlikely(
  378. INT_GET(dip->di_core.di_nextents, ARCH_CONVERT) +
  379. INT_GET(dip->di_core.di_anextents, ARCH_CONVERT) >
  380. INT_GET(dip->di_core.di_nblocks, ARCH_CONVERT))) {
  381. xfs_fs_repair_cmn_err(CE_WARN, ip->i_mount,
  382. "corrupt dinode %Lu, extent total = %d, nblocks = %Lu.",
  383. (unsigned long long)ip->i_ino,
  384. (int)(INT_GET(dip->di_core.di_nextents, ARCH_CONVERT)
  385. + INT_GET(dip->di_core.di_anextents, ARCH_CONVERT)),
  386. (unsigned long long)
  387. INT_GET(dip->di_core.di_nblocks, ARCH_CONVERT));
  388. XFS_CORRUPTION_ERROR("xfs_iformat(1)", XFS_ERRLEVEL_LOW,
  389. ip->i_mount, dip);
  390. return XFS_ERROR(EFSCORRUPTED);
  391. }
  392. if (unlikely(INT_GET(dip->di_core.di_forkoff, ARCH_CONVERT) > ip->i_mount->m_sb.sb_inodesize)) {
  393. xfs_fs_repair_cmn_err(CE_WARN, ip->i_mount,
  394. "corrupt dinode %Lu, forkoff = 0x%x.",
  395. (unsigned long long)ip->i_ino,
  396. (int)(INT_GET(dip->di_core.di_forkoff, ARCH_CONVERT)));
  397. XFS_CORRUPTION_ERROR("xfs_iformat(2)", XFS_ERRLEVEL_LOW,
  398. ip->i_mount, dip);
  399. return XFS_ERROR(EFSCORRUPTED);
  400. }
  401. switch (ip->i_d.di_mode & S_IFMT) {
  402. case S_IFIFO:
  403. case S_IFCHR:
  404. case S_IFBLK:
  405. case S_IFSOCK:
  406. if (unlikely(INT_GET(dip->di_core.di_format, ARCH_CONVERT) != XFS_DINODE_FMT_DEV)) {
  407. XFS_CORRUPTION_ERROR("xfs_iformat(3)", XFS_ERRLEVEL_LOW,
  408. ip->i_mount, dip);
  409. return XFS_ERROR(EFSCORRUPTED);
  410. }
  411. ip->i_d.di_size = 0;
  412. ip->i_df.if_u2.if_rdev = INT_GET(dip->di_u.di_dev, ARCH_CONVERT);
  413. break;
  414. case S_IFREG:
  415. case S_IFLNK:
  416. case S_IFDIR:
  417. switch (INT_GET(dip->di_core.di_format, ARCH_CONVERT)) {
  418. case XFS_DINODE_FMT_LOCAL:
  419. /*
  420. * no local regular files yet
  421. */
  422. if (unlikely((INT_GET(dip->di_core.di_mode, ARCH_CONVERT) & S_IFMT) == S_IFREG)) {
  423. xfs_fs_repair_cmn_err(CE_WARN, ip->i_mount,
  424. "corrupt inode %Lu "
  425. "(local format for regular file).",
  426. (unsigned long long) ip->i_ino);
  427. XFS_CORRUPTION_ERROR("xfs_iformat(4)",
  428. XFS_ERRLEVEL_LOW,
  429. ip->i_mount, dip);
  430. return XFS_ERROR(EFSCORRUPTED);
  431. }
  432. di_size = INT_GET(dip->di_core.di_size, ARCH_CONVERT);
  433. if (unlikely(di_size > XFS_DFORK_DSIZE(dip, ip->i_mount))) {
  434. xfs_fs_repair_cmn_err(CE_WARN, ip->i_mount,
  435. "corrupt inode %Lu "
  436. "(bad size %Ld for local inode).",
  437. (unsigned long long) ip->i_ino,
  438. (long long) di_size);
  439. XFS_CORRUPTION_ERROR("xfs_iformat(5)",
  440. XFS_ERRLEVEL_LOW,
  441. ip->i_mount, dip);
  442. return XFS_ERROR(EFSCORRUPTED);
  443. }
  444. size = (int)di_size;
  445. error = xfs_iformat_local(ip, dip, XFS_DATA_FORK, size);
  446. break;
  447. case XFS_DINODE_FMT_EXTENTS:
  448. error = xfs_iformat_extents(ip, dip, XFS_DATA_FORK);
  449. break;
  450. case XFS_DINODE_FMT_BTREE:
  451. error = xfs_iformat_btree(ip, dip, XFS_DATA_FORK);
  452. break;
  453. default:
  454. XFS_ERROR_REPORT("xfs_iformat(6)", XFS_ERRLEVEL_LOW,
  455. ip->i_mount);
  456. return XFS_ERROR(EFSCORRUPTED);
  457. }
  458. break;
  459. default:
  460. XFS_ERROR_REPORT("xfs_iformat(7)", XFS_ERRLEVEL_LOW, ip->i_mount);
  461. return XFS_ERROR(EFSCORRUPTED);
  462. }
  463. if (error) {
  464. return error;
  465. }
  466. if (!XFS_DFORK_Q(dip))
  467. return 0;
  468. ASSERT(ip->i_afp == NULL);
  469. ip->i_afp = kmem_zone_zalloc(xfs_ifork_zone, KM_SLEEP);
  470. ip->i_afp->if_ext_max =
  471. XFS_IFORK_ASIZE(ip) / (uint)sizeof(xfs_bmbt_rec_t);
  472. switch (INT_GET(dip->di_core.di_aformat, ARCH_CONVERT)) {
  473. case XFS_DINODE_FMT_LOCAL:
  474. atp = (xfs_attr_shortform_t *)XFS_DFORK_APTR(dip);
  475. size = be16_to_cpu(atp->hdr.totsize);
  476. error = xfs_iformat_local(ip, dip, XFS_ATTR_FORK, size);
  477. break;
  478. case XFS_DINODE_FMT_EXTENTS:
  479. error = xfs_iformat_extents(ip, dip, XFS_ATTR_FORK);
  480. break;
  481. case XFS_DINODE_FMT_BTREE:
  482. error = xfs_iformat_btree(ip, dip, XFS_ATTR_FORK);
  483. break;
  484. default:
  485. error = XFS_ERROR(EFSCORRUPTED);
  486. break;
  487. }
  488. if (error) {
  489. kmem_zone_free(xfs_ifork_zone, ip->i_afp);
  490. ip->i_afp = NULL;
  491. xfs_idestroy_fork(ip, XFS_DATA_FORK);
  492. }
  493. return error;
  494. }
  495. /*
  496. * The file is in-lined in the on-disk inode.
  497. * If it fits into if_inline_data, then copy
  498. * it there, otherwise allocate a buffer for it
  499. * and copy the data there. Either way, set
  500. * if_data to point at the data.
  501. * If we allocate a buffer for the data, make
  502. * sure that its size is a multiple of 4 and
  503. * record the real size in i_real_bytes.
  504. */
  505. STATIC int
  506. xfs_iformat_local(
  507. xfs_inode_t *ip,
  508. xfs_dinode_t *dip,
  509. int whichfork,
  510. int size)
  511. {
  512. xfs_ifork_t *ifp;
  513. int real_size;
  514. /*
  515. * If the size is unreasonable, then something
  516. * is wrong and we just bail out rather than crash in
  517. * kmem_alloc() or memcpy() below.
  518. */
  519. if (unlikely(size > XFS_DFORK_SIZE(dip, ip->i_mount, whichfork))) {
  520. xfs_fs_repair_cmn_err(CE_WARN, ip->i_mount,
  521. "corrupt inode %Lu "
  522. "(bad size %d for local fork, size = %d).",
  523. (unsigned long long) ip->i_ino, size,
  524. XFS_DFORK_SIZE(dip, ip->i_mount, whichfork));
  525. XFS_CORRUPTION_ERROR("xfs_iformat_local", XFS_ERRLEVEL_LOW,
  526. ip->i_mount, dip);
  527. return XFS_ERROR(EFSCORRUPTED);
  528. }
  529. ifp = XFS_IFORK_PTR(ip, whichfork);
  530. real_size = 0;
  531. if (size == 0)
  532. ifp->if_u1.if_data = NULL;
  533. else if (size <= sizeof(ifp->if_u2.if_inline_data))
  534. ifp->if_u1.if_data = ifp->if_u2.if_inline_data;
  535. else {
  536. real_size = roundup(size, 4);
  537. ifp->if_u1.if_data = kmem_alloc(real_size, KM_SLEEP);
  538. }
  539. ifp->if_bytes = size;
  540. ifp->if_real_bytes = real_size;
  541. if (size)
  542. memcpy(ifp->if_u1.if_data, XFS_DFORK_PTR(dip, whichfork), size);
  543. ifp->if_flags &= ~XFS_IFEXTENTS;
  544. ifp->if_flags |= XFS_IFINLINE;
  545. return 0;
  546. }
  547. /*
  548. * The file consists of a set of extents all
  549. * of which fit into the on-disk inode.
  550. * If there are few enough extents to fit into
  551. * the if_inline_ext, then copy them there.
  552. * Otherwise allocate a buffer for them and copy
  553. * them into it. Either way, set if_extents
  554. * to point at the extents.
  555. */
  556. STATIC int
  557. xfs_iformat_extents(
  558. xfs_inode_t *ip,
  559. xfs_dinode_t *dip,
  560. int whichfork)
  561. {
  562. xfs_bmbt_rec_t *ep, *dp;
  563. xfs_ifork_t *ifp;
  564. int nex;
  565. int size;
  566. int i;
  567. ifp = XFS_IFORK_PTR(ip, whichfork);
  568. nex = XFS_DFORK_NEXTENTS(dip, whichfork);
  569. size = nex * (uint)sizeof(xfs_bmbt_rec_t);
  570. /*
  571. * If the number of extents is unreasonable, then something
  572. * is wrong and we just bail out rather than crash in
  573. * kmem_alloc() or memcpy() below.
  574. */
  575. if (unlikely(size < 0 || size > XFS_DFORK_SIZE(dip, ip->i_mount, whichfork))) {
  576. xfs_fs_repair_cmn_err(CE_WARN, ip->i_mount,
  577. "corrupt inode %Lu ((a)extents = %d).",
  578. (unsigned long long) ip->i_ino, nex);
  579. XFS_CORRUPTION_ERROR("xfs_iformat_extents(1)", XFS_ERRLEVEL_LOW,
  580. ip->i_mount, dip);
  581. return XFS_ERROR(EFSCORRUPTED);
  582. }
  583. ifp->if_real_bytes = 0;
  584. if (nex == 0)
  585. ifp->if_u1.if_extents = NULL;
  586. else if (nex <= XFS_INLINE_EXTS)
  587. ifp->if_u1.if_extents = ifp->if_u2.if_inline_ext;
  588. else
  589. xfs_iext_add(ifp, 0, nex);
  590. ifp->if_bytes = size;
  591. if (size) {
  592. dp = (xfs_bmbt_rec_t *) XFS_DFORK_PTR(dip, whichfork);
  593. xfs_validate_extents(ifp, nex, 1, XFS_EXTFMT_INODE(ip));
  594. for (i = 0; i < nex; i++, dp++) {
  595. ep = xfs_iext_get_ext(ifp, i);
  596. ep->l0 = INT_GET(get_unaligned((__uint64_t*)&dp->l0),
  597. ARCH_CONVERT);
  598. ep->l1 = INT_GET(get_unaligned((__uint64_t*)&dp->l1),
  599. ARCH_CONVERT);
  600. }
  601. xfs_bmap_trace_exlist("xfs_iformat_extents", ip, nex,
  602. whichfork);
  603. if (whichfork != XFS_DATA_FORK ||
  604. XFS_EXTFMT_INODE(ip) == XFS_EXTFMT_NOSTATE)
  605. if (unlikely(xfs_check_nostate_extents(
  606. ifp, 0, nex))) {
  607. XFS_ERROR_REPORT("xfs_iformat_extents(2)",
  608. XFS_ERRLEVEL_LOW,
  609. ip->i_mount);
  610. return XFS_ERROR(EFSCORRUPTED);
  611. }
  612. }
  613. ifp->if_flags |= XFS_IFEXTENTS;
  614. return 0;
  615. }
  616. /*
  617. * The file has too many extents to fit into
  618. * the inode, so they are in B-tree format.
  619. * Allocate a buffer for the root of the B-tree
  620. * and copy the root into it. The i_extents
  621. * field will remain NULL until all of the
  622. * extents are read in (when they are needed).
  623. */
  624. STATIC int
  625. xfs_iformat_btree(
  626. xfs_inode_t *ip,
  627. xfs_dinode_t *dip,
  628. int whichfork)
  629. {
  630. xfs_bmdr_block_t *dfp;
  631. xfs_ifork_t *ifp;
  632. /* REFERENCED */
  633. int nrecs;
  634. int size;
  635. ifp = XFS_IFORK_PTR(ip, whichfork);
  636. dfp = (xfs_bmdr_block_t *)XFS_DFORK_PTR(dip, whichfork);
  637. size = XFS_BMAP_BROOT_SPACE(dfp);
  638. nrecs = XFS_BMAP_BROOT_NUMRECS(dfp);
  639. /*
  640. * blow out if -- fork has less extents than can fit in
  641. * fork (fork shouldn't be a btree format), root btree
  642. * block has more records than can fit into the fork,
  643. * or the number of extents is greater than the number of
  644. * blocks.
  645. */
  646. if (unlikely(XFS_IFORK_NEXTENTS(ip, whichfork) <= ifp->if_ext_max
  647. || XFS_BMDR_SPACE_CALC(nrecs) >
  648. XFS_DFORK_SIZE(dip, ip->i_mount, whichfork)
  649. || XFS_IFORK_NEXTENTS(ip, whichfork) > ip->i_d.di_nblocks)) {
  650. xfs_fs_repair_cmn_err(CE_WARN, ip->i_mount,
  651. "corrupt inode %Lu (btree).",
  652. (unsigned long long) ip->i_ino);
  653. XFS_ERROR_REPORT("xfs_iformat_btree", XFS_ERRLEVEL_LOW,
  654. ip->i_mount);
  655. return XFS_ERROR(EFSCORRUPTED);
  656. }
  657. ifp->if_broot_bytes = size;
  658. ifp->if_broot = kmem_alloc(size, KM_SLEEP);
  659. ASSERT(ifp->if_broot != NULL);
  660. /*
  661. * Copy and convert from the on-disk structure
  662. * to the in-memory structure.
  663. */
  664. xfs_bmdr_to_bmbt(dfp, XFS_DFORK_SIZE(dip, ip->i_mount, whichfork),
  665. ifp->if_broot, size);
  666. ifp->if_flags &= ~XFS_IFEXTENTS;
  667. ifp->if_flags |= XFS_IFBROOT;
  668. return 0;
  669. }
  670. /*
  671. * xfs_xlate_dinode_core - translate an xfs_inode_core_t between ondisk
  672. * and native format
  673. *
  674. * buf = on-disk representation
  675. * dip = native representation
  676. * dir = direction - +ve -> disk to native
  677. * -ve -> native to disk
  678. */
  679. void
  680. xfs_xlate_dinode_core(
  681. xfs_caddr_t buf,
  682. xfs_dinode_core_t *dip,
  683. int dir)
  684. {
  685. xfs_dinode_core_t *buf_core = (xfs_dinode_core_t *)buf;
  686. xfs_dinode_core_t *mem_core = (xfs_dinode_core_t *)dip;
  687. xfs_arch_t arch = ARCH_CONVERT;
  688. ASSERT(dir);
  689. INT_XLATE(buf_core->di_magic, mem_core->di_magic, dir, arch);
  690. INT_XLATE(buf_core->di_mode, mem_core->di_mode, dir, arch);
  691. INT_XLATE(buf_core->di_version, mem_core->di_version, dir, arch);
  692. INT_XLATE(buf_core->di_format, mem_core->di_format, dir, arch);
  693. INT_XLATE(buf_core->di_onlink, mem_core->di_onlink, dir, arch);
  694. INT_XLATE(buf_core->di_uid, mem_core->di_uid, dir, arch);
  695. INT_XLATE(buf_core->di_gid, mem_core->di_gid, dir, arch);
  696. INT_XLATE(buf_core->di_nlink, mem_core->di_nlink, dir, arch);
  697. INT_XLATE(buf_core->di_projid, mem_core->di_projid, dir, arch);
  698. if (dir > 0) {
  699. memcpy(mem_core->di_pad, buf_core->di_pad,
  700. sizeof(buf_core->di_pad));
  701. } else {
  702. memcpy(buf_core->di_pad, mem_core->di_pad,
  703. sizeof(buf_core->di_pad));
  704. }
  705. INT_XLATE(buf_core->di_flushiter, mem_core->di_flushiter, dir, arch);
  706. INT_XLATE(buf_core->di_atime.t_sec, mem_core->di_atime.t_sec,
  707. dir, arch);
  708. INT_XLATE(buf_core->di_atime.t_nsec, mem_core->di_atime.t_nsec,
  709. dir, arch);
  710. INT_XLATE(buf_core->di_mtime.t_sec, mem_core->di_mtime.t_sec,
  711. dir, arch);
  712. INT_XLATE(buf_core->di_mtime.t_nsec, mem_core->di_mtime.t_nsec,
  713. dir, arch);
  714. INT_XLATE(buf_core->di_ctime.t_sec, mem_core->di_ctime.t_sec,
  715. dir, arch);
  716. INT_XLATE(buf_core->di_ctime.t_nsec, mem_core->di_ctime.t_nsec,
  717. dir, arch);
  718. INT_XLATE(buf_core->di_size, mem_core->di_size, dir, arch);
  719. INT_XLATE(buf_core->di_nblocks, mem_core->di_nblocks, dir, arch);
  720. INT_XLATE(buf_core->di_extsize, mem_core->di_extsize, dir, arch);
  721. INT_XLATE(buf_core->di_nextents, mem_core->di_nextents, dir, arch);
  722. INT_XLATE(buf_core->di_anextents, mem_core->di_anextents, dir, arch);
  723. INT_XLATE(buf_core->di_forkoff, mem_core->di_forkoff, dir, arch);
  724. INT_XLATE(buf_core->di_aformat, mem_core->di_aformat, dir, arch);
  725. INT_XLATE(buf_core->di_dmevmask, mem_core->di_dmevmask, dir, arch);
  726. INT_XLATE(buf_core->di_dmstate, mem_core->di_dmstate, dir, arch);
  727. INT_XLATE(buf_core->di_flags, mem_core->di_flags, dir, arch);
  728. INT_XLATE(buf_core->di_gen, mem_core->di_gen, dir, arch);
  729. }
  730. STATIC uint
  731. _xfs_dic2xflags(
  732. __uint16_t di_flags)
  733. {
  734. uint flags = 0;
  735. if (di_flags & XFS_DIFLAG_ANY) {
  736. if (di_flags & XFS_DIFLAG_REALTIME)
  737. flags |= XFS_XFLAG_REALTIME;
  738. if (di_flags & XFS_DIFLAG_PREALLOC)
  739. flags |= XFS_XFLAG_PREALLOC;
  740. if (di_flags & XFS_DIFLAG_IMMUTABLE)
  741. flags |= XFS_XFLAG_IMMUTABLE;
  742. if (di_flags & XFS_DIFLAG_APPEND)
  743. flags |= XFS_XFLAG_APPEND;
  744. if (di_flags & XFS_DIFLAG_SYNC)
  745. flags |= XFS_XFLAG_SYNC;
  746. if (di_flags & XFS_DIFLAG_NOATIME)
  747. flags |= XFS_XFLAG_NOATIME;
  748. if (di_flags & XFS_DIFLAG_NODUMP)
  749. flags |= XFS_XFLAG_NODUMP;
  750. if (di_flags & XFS_DIFLAG_RTINHERIT)
  751. flags |= XFS_XFLAG_RTINHERIT;
  752. if (di_flags & XFS_DIFLAG_PROJINHERIT)
  753. flags |= XFS_XFLAG_PROJINHERIT;
  754. if (di_flags & XFS_DIFLAG_NOSYMLINKS)
  755. flags |= XFS_XFLAG_NOSYMLINKS;
  756. if (di_flags & XFS_DIFLAG_EXTSIZE)
  757. flags |= XFS_XFLAG_EXTSIZE;
  758. if (di_flags & XFS_DIFLAG_EXTSZINHERIT)
  759. flags |= XFS_XFLAG_EXTSZINHERIT;
  760. if (di_flags & XFS_DIFLAG_NODEFRAG)
  761. flags |= XFS_XFLAG_NODEFRAG;
  762. }
  763. return flags;
  764. }
  765. uint
  766. xfs_ip2xflags(
  767. xfs_inode_t *ip)
  768. {
  769. xfs_dinode_core_t *dic = &ip->i_d;
  770. return _xfs_dic2xflags(dic->di_flags) |
  771. (XFS_CFORK_Q(dic) ? XFS_XFLAG_HASATTR : 0);
  772. }
  773. uint
  774. xfs_dic2xflags(
  775. xfs_dinode_core_t *dic)
  776. {
  777. return _xfs_dic2xflags(INT_GET(dic->di_flags, ARCH_CONVERT)) |
  778. (XFS_CFORK_Q_DISK(dic) ? XFS_XFLAG_HASATTR : 0);
  779. }
  780. /*
  781. * Given a mount structure and an inode number, return a pointer
  782. * to a newly allocated in-core inode corresponding to the given
  783. * inode number.
  784. *
  785. * Initialize the inode's attributes and extent pointers if it
  786. * already has them (it will not if the inode has no links).
  787. */
  788. int
  789. xfs_iread(
  790. xfs_mount_t *mp,
  791. xfs_trans_t *tp,
  792. xfs_ino_t ino,
  793. xfs_inode_t **ipp,
  794. xfs_daddr_t bno,
  795. uint imap_flags)
  796. {
  797. xfs_buf_t *bp;
  798. xfs_dinode_t *dip;
  799. xfs_inode_t *ip;
  800. int error;
  801. ASSERT(xfs_inode_zone != NULL);
  802. ip = kmem_zone_zalloc(xfs_inode_zone, KM_SLEEP);
  803. ip->i_ino = ino;
  804. ip->i_mount = mp;
  805. spin_lock_init(&ip->i_flags_lock);
  806. /*
  807. * Get pointer's to the on-disk inode and the buffer containing it.
  808. * If the inode number refers to a block outside the file system
  809. * then xfs_itobp() will return NULL. In this case we should
  810. * return NULL as well. Set i_blkno to 0 so that xfs_itobp() will
  811. * know that this is a new incore inode.
  812. */
  813. error = xfs_itobp(mp, tp, ip, &dip, &bp, bno, imap_flags);
  814. if (error) {
  815. kmem_zone_free(xfs_inode_zone, ip);
  816. return error;
  817. }
  818. /*
  819. * Initialize inode's trace buffers.
  820. * Do this before xfs_iformat in case it adds entries.
  821. */
  822. #ifdef XFS_BMAP_TRACE
  823. ip->i_xtrace = ktrace_alloc(XFS_BMAP_KTRACE_SIZE, KM_SLEEP);
  824. #endif
  825. #ifdef XFS_BMBT_TRACE
  826. ip->i_btrace = ktrace_alloc(XFS_BMBT_KTRACE_SIZE, KM_SLEEP);
  827. #endif
  828. #ifdef XFS_RW_TRACE
  829. ip->i_rwtrace = ktrace_alloc(XFS_RW_KTRACE_SIZE, KM_SLEEP);
  830. #endif
  831. #ifdef XFS_ILOCK_TRACE
  832. ip->i_lock_trace = ktrace_alloc(XFS_ILOCK_KTRACE_SIZE, KM_SLEEP);
  833. #endif
  834. #ifdef XFS_DIR2_TRACE
  835. ip->i_dir_trace = ktrace_alloc(XFS_DIR2_KTRACE_SIZE, KM_SLEEP);
  836. #endif
  837. /*
  838. * If we got something that isn't an inode it means someone
  839. * (nfs or dmi) has a stale handle.
  840. */
  841. if (INT_GET(dip->di_core.di_magic, ARCH_CONVERT) != XFS_DINODE_MAGIC) {
  842. kmem_zone_free(xfs_inode_zone, ip);
  843. xfs_trans_brelse(tp, bp);
  844. #ifdef DEBUG
  845. xfs_fs_cmn_err(CE_ALERT, mp, "xfs_iread: "
  846. "dip->di_core.di_magic (0x%x) != "
  847. "XFS_DINODE_MAGIC (0x%x)",
  848. INT_GET(dip->di_core.di_magic, ARCH_CONVERT),
  849. XFS_DINODE_MAGIC);
  850. #endif /* DEBUG */
  851. return XFS_ERROR(EINVAL);
  852. }
  853. /*
  854. * If the on-disk inode is already linked to a directory
  855. * entry, copy all of the inode into the in-core inode.
  856. * xfs_iformat() handles copying in the inode format
  857. * specific information.
  858. * Otherwise, just get the truly permanent information.
  859. */
  860. if (dip->di_core.di_mode) {
  861. xfs_xlate_dinode_core((xfs_caddr_t)&dip->di_core,
  862. &(ip->i_d), 1);
  863. error = xfs_iformat(ip, dip);
  864. if (error) {
  865. kmem_zone_free(xfs_inode_zone, ip);
  866. xfs_trans_brelse(tp, bp);
  867. #ifdef DEBUG
  868. xfs_fs_cmn_err(CE_ALERT, mp, "xfs_iread: "
  869. "xfs_iformat() returned error %d",
  870. error);
  871. #endif /* DEBUG */
  872. return error;
  873. }
  874. } else {
  875. ip->i_d.di_magic = INT_GET(dip->di_core.di_magic, ARCH_CONVERT);
  876. ip->i_d.di_version = INT_GET(dip->di_core.di_version, ARCH_CONVERT);
  877. ip->i_d.di_gen = INT_GET(dip->di_core.di_gen, ARCH_CONVERT);
  878. ip->i_d.di_flushiter = INT_GET(dip->di_core.di_flushiter, ARCH_CONVERT);
  879. /*
  880. * Make sure to pull in the mode here as well in
  881. * case the inode is released without being used.
  882. * This ensures that xfs_inactive() will see that
  883. * the inode is already free and not try to mess
  884. * with the uninitialized part of it.
  885. */
  886. ip->i_d.di_mode = 0;
  887. /*
  888. * Initialize the per-fork minima and maxima for a new
  889. * inode here. xfs_iformat will do it for old inodes.
  890. */
  891. ip->i_df.if_ext_max =
  892. XFS_IFORK_DSIZE(ip) / (uint)sizeof(xfs_bmbt_rec_t);
  893. }
  894. INIT_LIST_HEAD(&ip->i_reclaim);
  895. /*
  896. * The inode format changed when we moved the link count and
  897. * made it 32 bits long. If this is an old format inode,
  898. * convert it in memory to look like a new one. If it gets
  899. * flushed to disk we will convert back before flushing or
  900. * logging it. We zero out the new projid field and the old link
  901. * count field. We'll handle clearing the pad field (the remains
  902. * of the old uuid field) when we actually convert the inode to
  903. * the new format. We don't change the version number so that we
  904. * can distinguish this from a real new format inode.
  905. */
  906. if (ip->i_d.di_version == XFS_DINODE_VERSION_1) {
  907. ip->i_d.di_nlink = ip->i_d.di_onlink;
  908. ip->i_d.di_onlink = 0;
  909. ip->i_d.di_projid = 0;
  910. }
  911. ip->i_delayed_blks = 0;
  912. /*
  913. * Mark the buffer containing the inode as something to keep
  914. * around for a while. This helps to keep recently accessed
  915. * meta-data in-core longer.
  916. */
  917. XFS_BUF_SET_REF(bp, XFS_INO_REF);
  918. /*
  919. * Use xfs_trans_brelse() to release the buffer containing the
  920. * on-disk inode, because it was acquired with xfs_trans_read_buf()
  921. * in xfs_itobp() above. If tp is NULL, this is just a normal
  922. * brelse(). If we're within a transaction, then xfs_trans_brelse()
  923. * will only release the buffer if it is not dirty within the
  924. * transaction. It will be OK to release the buffer in this case,
  925. * because inodes on disk are never destroyed and we will be
  926. * locking the new in-core inode before putting it in the hash
  927. * table where other processes can find it. Thus we don't have
  928. * to worry about the inode being changed just because we released
  929. * the buffer.
  930. */
  931. xfs_trans_brelse(tp, bp);
  932. *ipp = ip;
  933. return 0;
  934. }
  935. /*
  936. * Read in extents from a btree-format inode.
  937. * Allocate and fill in if_extents. Real work is done in xfs_bmap.c.
  938. */
  939. int
  940. xfs_iread_extents(
  941. xfs_trans_t *tp,
  942. xfs_inode_t *ip,
  943. int whichfork)
  944. {
  945. int error;
  946. xfs_ifork_t *ifp;
  947. xfs_extnum_t nextents;
  948. size_t size;
  949. if (unlikely(XFS_IFORK_FORMAT(ip, whichfork) != XFS_DINODE_FMT_BTREE)) {
  950. XFS_ERROR_REPORT("xfs_iread_extents", XFS_ERRLEVEL_LOW,
  951. ip->i_mount);
  952. return XFS_ERROR(EFSCORRUPTED);
  953. }
  954. nextents = XFS_IFORK_NEXTENTS(ip, whichfork);
  955. size = nextents * sizeof(xfs_bmbt_rec_t);
  956. ifp = XFS_IFORK_PTR(ip, whichfork);
  957. /*
  958. * We know that the size is valid (it's checked in iformat_btree)
  959. */
  960. ifp->if_lastex = NULLEXTNUM;
  961. ifp->if_bytes = ifp->if_real_bytes = 0;
  962. ifp->if_flags |= XFS_IFEXTENTS;
  963. xfs_iext_add(ifp, 0, nextents);
  964. error = xfs_bmap_read_extents(tp, ip, whichfork);
  965. if (error) {
  966. xfs_iext_destroy(ifp);
  967. ifp->if_flags &= ~XFS_IFEXTENTS;
  968. return error;
  969. }
  970. xfs_validate_extents(ifp, nextents, 0, XFS_EXTFMT_INODE(ip));
  971. return 0;
  972. }
  973. /*
  974. * Allocate an inode on disk and return a copy of its in-core version.
  975. * The in-core inode is locked exclusively. Set mode, nlink, and rdev
  976. * appropriately within the inode. The uid and gid for the inode are
  977. * set according to the contents of the given cred structure.
  978. *
  979. * Use xfs_dialloc() to allocate the on-disk inode. If xfs_dialloc()
  980. * has a free inode available, call xfs_iget()
  981. * to obtain the in-core version of the allocated inode. Finally,
  982. * fill in the inode and log its initial contents. In this case,
  983. * ialloc_context would be set to NULL and call_again set to false.
  984. *
  985. * If xfs_dialloc() does not have an available inode,
  986. * it will replenish its supply by doing an allocation. Since we can
  987. * only do one allocation within a transaction without deadlocks, we
  988. * must commit the current transaction before returning the inode itself.
  989. * In this case, therefore, we will set call_again to true and return.
  990. * The caller should then commit the current transaction, start a new
  991. * transaction, and call xfs_ialloc() again to actually get the inode.
  992. *
  993. * To ensure that some other process does not grab the inode that
  994. * was allocated during the first call to xfs_ialloc(), this routine
  995. * also returns the [locked] bp pointing to the head of the freelist
  996. * as ialloc_context. The caller should hold this buffer across
  997. * the commit and pass it back into this routine on the second call.
  998. */
  999. int
  1000. xfs_ialloc(
  1001. xfs_trans_t *tp,
  1002. xfs_inode_t *pip,
  1003. mode_t mode,
  1004. xfs_nlink_t nlink,
  1005. xfs_dev_t rdev,
  1006. cred_t *cr,
  1007. xfs_prid_t prid,
  1008. int okalloc,
  1009. xfs_buf_t **ialloc_context,
  1010. boolean_t *call_again,
  1011. xfs_inode_t **ipp)
  1012. {
  1013. xfs_ino_t ino;
  1014. xfs_inode_t *ip;
  1015. bhv_vnode_t *vp;
  1016. uint flags;
  1017. int error;
  1018. /*
  1019. * Call the space management code to pick
  1020. * the on-disk inode to be allocated.
  1021. */
  1022. error = xfs_dialloc(tp, pip->i_ino, mode, okalloc,
  1023. ialloc_context, call_again, &ino);
  1024. if (error != 0) {
  1025. return error;
  1026. }
  1027. if (*call_again || ino == NULLFSINO) {
  1028. *ipp = NULL;
  1029. return 0;
  1030. }
  1031. ASSERT(*ialloc_context == NULL);
  1032. /*
  1033. * Get the in-core inode with the lock held exclusively.
  1034. * This is because we're setting fields here we need
  1035. * to prevent others from looking at until we're done.
  1036. */
  1037. error = xfs_trans_iget(tp->t_mountp, tp, ino,
  1038. XFS_IGET_CREATE, XFS_ILOCK_EXCL, &ip);
  1039. if (error != 0) {
  1040. return error;
  1041. }
  1042. ASSERT(ip != NULL);
  1043. vp = XFS_ITOV(ip);
  1044. ip->i_d.di_mode = (__uint16_t)mode;
  1045. ip->i_d.di_onlink = 0;
  1046. ip->i_d.di_nlink = nlink;
  1047. ASSERT(ip->i_d.di_nlink == nlink);
  1048. ip->i_d.di_uid = current_fsuid(cr);
  1049. ip->i_d.di_gid = current_fsgid(cr);
  1050. ip->i_d.di_projid = prid;
  1051. memset(&(ip->i_d.di_pad[0]), 0, sizeof(ip->i_d.di_pad));
  1052. /*
  1053. * If the superblock version is up to where we support new format
  1054. * inodes and this is currently an old format inode, then change
  1055. * the inode version number now. This way we only do the conversion
  1056. * here rather than here and in the flush/logging code.
  1057. */
  1058. if (XFS_SB_VERSION_HASNLINK(&tp->t_mountp->m_sb) &&
  1059. ip->i_d.di_version == XFS_DINODE_VERSION_1) {
  1060. ip->i_d.di_version = XFS_DINODE_VERSION_2;
  1061. /*
  1062. * We've already zeroed the old link count, the projid field,
  1063. * and the pad field.
  1064. */
  1065. }
  1066. /*
  1067. * Project ids won't be stored on disk if we are using a version 1 inode.
  1068. */
  1069. if ( (prid != 0) && (ip->i_d.di_version == XFS_DINODE_VERSION_1))
  1070. xfs_bump_ino_vers2(tp, ip);
  1071. if (XFS_INHERIT_GID(pip, vp->v_vfsp)) {
  1072. ip->i_d.di_gid = pip->i_d.di_gid;
  1073. if ((pip->i_d.di_mode & S_ISGID) && (mode & S_IFMT) == S_IFDIR) {
  1074. ip->i_d.di_mode |= S_ISGID;
  1075. }
  1076. }
  1077. /*
  1078. * If the group ID of the new file does not match the effective group
  1079. * ID or one of the supplementary group IDs, the S_ISGID bit is cleared
  1080. * (and only if the irix_sgid_inherit compatibility variable is set).
  1081. */
  1082. if ((irix_sgid_inherit) &&
  1083. (ip->i_d.di_mode & S_ISGID) &&
  1084. (!in_group_p((gid_t)ip->i_d.di_gid))) {
  1085. ip->i_d.di_mode &= ~S_ISGID;
  1086. }
  1087. ip->i_d.di_size = 0;
  1088. ip->i_d.di_nextents = 0;
  1089. ASSERT(ip->i_d.di_nblocks == 0);
  1090. xfs_ichgtime(ip, XFS_ICHGTIME_CHG|XFS_ICHGTIME_ACC|XFS_ICHGTIME_MOD);
  1091. /*
  1092. * di_gen will have been taken care of in xfs_iread.
  1093. */
  1094. ip->i_d.di_extsize = 0;
  1095. ip->i_d.di_dmevmask = 0;
  1096. ip->i_d.di_dmstate = 0;
  1097. ip->i_d.di_flags = 0;
  1098. flags = XFS_ILOG_CORE;
  1099. switch (mode & S_IFMT) {
  1100. case S_IFIFO:
  1101. case S_IFCHR:
  1102. case S_IFBLK:
  1103. case S_IFSOCK:
  1104. ip->i_d.di_format = XFS_DINODE_FMT_DEV;
  1105. ip->i_df.if_u2.if_rdev = rdev;
  1106. ip->i_df.if_flags = 0;
  1107. flags |= XFS_ILOG_DEV;
  1108. break;
  1109. case S_IFREG:
  1110. case S_IFDIR:
  1111. if (unlikely(pip->i_d.di_flags & XFS_DIFLAG_ANY)) {
  1112. uint di_flags = 0;
  1113. if ((mode & S_IFMT) == S_IFDIR) {
  1114. if (pip->i_d.di_flags & XFS_DIFLAG_RTINHERIT)
  1115. di_flags |= XFS_DIFLAG_RTINHERIT;
  1116. if (pip->i_d.di_flags & XFS_DIFLAG_EXTSZINHERIT) {
  1117. di_flags |= XFS_DIFLAG_EXTSZINHERIT;
  1118. ip->i_d.di_extsize = pip->i_d.di_extsize;
  1119. }
  1120. } else if ((mode & S_IFMT) == S_IFREG) {
  1121. if (pip->i_d.di_flags & XFS_DIFLAG_RTINHERIT) {
  1122. di_flags |= XFS_DIFLAG_REALTIME;
  1123. ip->i_iocore.io_flags |= XFS_IOCORE_RT;
  1124. }
  1125. if (pip->i_d.di_flags & XFS_DIFLAG_EXTSZINHERIT) {
  1126. di_flags |= XFS_DIFLAG_EXTSIZE;
  1127. ip->i_d.di_extsize = pip->i_d.di_extsize;
  1128. }
  1129. }
  1130. if ((pip->i_d.di_flags & XFS_DIFLAG_NOATIME) &&
  1131. xfs_inherit_noatime)
  1132. di_flags |= XFS_DIFLAG_NOATIME;
  1133. if ((pip->i_d.di_flags & XFS_DIFLAG_NODUMP) &&
  1134. xfs_inherit_nodump)
  1135. di_flags |= XFS_DIFLAG_NODUMP;
  1136. if ((pip->i_d.di_flags & XFS_DIFLAG_SYNC) &&
  1137. xfs_inherit_sync)
  1138. di_flags |= XFS_DIFLAG_SYNC;
  1139. if ((pip->i_d.di_flags & XFS_DIFLAG_NOSYMLINKS) &&
  1140. xfs_inherit_nosymlinks)
  1141. di_flags |= XFS_DIFLAG_NOSYMLINKS;
  1142. if (pip->i_d.di_flags & XFS_DIFLAG_PROJINHERIT)
  1143. di_flags |= XFS_DIFLAG_PROJINHERIT;
  1144. if ((pip->i_d.di_flags & XFS_DIFLAG_NODEFRAG) &&
  1145. xfs_inherit_nodefrag)
  1146. di_flags |= XFS_DIFLAG_NODEFRAG;
  1147. ip->i_d.di_flags |= di_flags;
  1148. }
  1149. /* FALLTHROUGH */
  1150. case S_IFLNK:
  1151. ip->i_d.di_format = XFS_DINODE_FMT_EXTENTS;
  1152. ip->i_df.if_flags = XFS_IFEXTENTS;
  1153. ip->i_df.if_bytes = ip->i_df.if_real_bytes = 0;
  1154. ip->i_df.if_u1.if_extents = NULL;
  1155. break;
  1156. default:
  1157. ASSERT(0);
  1158. }
  1159. /*
  1160. * Attribute fork settings for new inode.
  1161. */
  1162. ip->i_d.di_aformat = XFS_DINODE_FMT_EXTENTS;
  1163. ip->i_d.di_anextents = 0;
  1164. /*
  1165. * Log the new values stuffed into the inode.
  1166. */
  1167. xfs_trans_log_inode(tp, ip, flags);
  1168. /* now that we have an i_mode we can setup inode ops and unlock */
  1169. bhv_vfs_init_vnode(XFS_MTOVFS(tp->t_mountp), vp, XFS_ITOBHV(ip), 1);
  1170. *ipp = ip;
  1171. return 0;
  1172. }
  1173. /*
  1174. * Check to make sure that there are no blocks allocated to the
  1175. * file beyond the size of the file. We don't check this for
  1176. * files with fixed size extents or real time extents, but we
  1177. * at least do it for regular files.
  1178. */
  1179. #ifdef DEBUG
  1180. void
  1181. xfs_isize_check(
  1182. xfs_mount_t *mp,
  1183. xfs_inode_t *ip,
  1184. xfs_fsize_t isize)
  1185. {
  1186. xfs_fileoff_t map_first;
  1187. int nimaps;
  1188. xfs_bmbt_irec_t imaps[2];
  1189. if ((ip->i_d.di_mode & S_IFMT) != S_IFREG)
  1190. return;
  1191. if (ip->i_d.di_flags & (XFS_DIFLAG_REALTIME | XFS_DIFLAG_EXTSIZE))
  1192. return;
  1193. nimaps = 2;
  1194. map_first = XFS_B_TO_FSB(mp, (xfs_ufsize_t)isize);
  1195. /*
  1196. * The filesystem could be shutting down, so bmapi may return
  1197. * an error.
  1198. */
  1199. if (xfs_bmapi(NULL, ip, map_first,
  1200. (XFS_B_TO_FSB(mp,
  1201. (xfs_ufsize_t)XFS_MAXIOFFSET(mp)) -
  1202. map_first),
  1203. XFS_BMAPI_ENTIRE, NULL, 0, imaps, &nimaps,
  1204. NULL, NULL))
  1205. return;
  1206. ASSERT(nimaps == 1);
  1207. ASSERT(imaps[0].br_startblock == HOLESTARTBLOCK);
  1208. }
  1209. #endif /* DEBUG */
  1210. /*
  1211. * Calculate the last possible buffered byte in a file. This must
  1212. * include data that was buffered beyond the EOF by the write code.
  1213. * This also needs to deal with overflowing the xfs_fsize_t type
  1214. * which can happen for sizes near the limit.
  1215. *
  1216. * We also need to take into account any blocks beyond the EOF. It
  1217. * may be the case that they were buffered by a write which failed.
  1218. * In that case the pages will still be in memory, but the inode size
  1219. * will never have been updated.
  1220. */
  1221. xfs_fsize_t
  1222. xfs_file_last_byte(
  1223. xfs_inode_t *ip)
  1224. {
  1225. xfs_mount_t *mp;
  1226. xfs_fsize_t last_byte;
  1227. xfs_fileoff_t last_block;
  1228. xfs_fileoff_t size_last_block;
  1229. int error;
  1230. ASSERT(ismrlocked(&(ip->i_iolock), MR_UPDATE | MR_ACCESS));
  1231. mp = ip->i_mount;
  1232. /*
  1233. * Only check for blocks beyond the EOF if the extents have
  1234. * been read in. This eliminates the need for the inode lock,
  1235. * and it also saves us from looking when it really isn't
  1236. * necessary.
  1237. */
  1238. if (ip->i_df.if_flags & XFS_IFEXTENTS) {
  1239. error = xfs_bmap_last_offset(NULL, ip, &last_block,
  1240. XFS_DATA_FORK);
  1241. if (error) {
  1242. last_block = 0;
  1243. }
  1244. } else {
  1245. last_block = 0;
  1246. }
  1247. size_last_block = XFS_B_TO_FSB(mp, (xfs_ufsize_t)ip->i_d.di_size);
  1248. last_block = XFS_FILEOFF_MAX(last_block, size_last_block);
  1249. last_byte = XFS_FSB_TO_B(mp, last_block);
  1250. if (last_byte < 0) {
  1251. return XFS_MAXIOFFSET(mp);
  1252. }
  1253. last_byte += (1 << mp->m_writeio_log);
  1254. if (last_byte < 0) {
  1255. return XFS_MAXIOFFSET(mp);
  1256. }
  1257. return last_byte;
  1258. }
  1259. #if defined(XFS_RW_TRACE)
  1260. STATIC void
  1261. xfs_itrunc_trace(
  1262. int tag,
  1263. xfs_inode_t *ip,
  1264. int flag,
  1265. xfs_fsize_t new_size,
  1266. xfs_off_t toss_start,
  1267. xfs_off_t toss_finish)
  1268. {
  1269. if (ip->i_rwtrace == NULL) {
  1270. return;
  1271. }
  1272. ktrace_enter(ip->i_rwtrace,
  1273. (void*)((long)tag),
  1274. (void*)ip,
  1275. (void*)(unsigned long)((ip->i_d.di_size >> 32) & 0xffffffff),
  1276. (void*)(unsigned long)(ip->i_d.di_size & 0xffffffff),
  1277. (void*)((long)flag),
  1278. (void*)(unsigned long)((new_size >> 32) & 0xffffffff),
  1279. (void*)(unsigned long)(new_size & 0xffffffff),
  1280. (void*)(unsigned long)((toss_start >> 32) & 0xffffffff),
  1281. (void*)(unsigned long)(toss_start & 0xffffffff),
  1282. (void*)(unsigned long)((toss_finish >> 32) & 0xffffffff),
  1283. (void*)(unsigned long)(toss_finish & 0xffffffff),
  1284. (void*)(unsigned long)current_cpu(),
  1285. (void*)(unsigned long)current_pid(),
  1286. (void*)NULL,
  1287. (void*)NULL,
  1288. (void*)NULL);
  1289. }
  1290. #else
  1291. #define xfs_itrunc_trace(tag, ip, flag, new_size, toss_start, toss_finish)
  1292. #endif
  1293. /*
  1294. * Start the truncation of the file to new_size. The new size
  1295. * must be smaller than the current size. This routine will
  1296. * clear the buffer and page caches of file data in the removed
  1297. * range, and xfs_itruncate_finish() will remove the underlying
  1298. * disk blocks.
  1299. *
  1300. * The inode must have its I/O lock locked EXCLUSIVELY, and it
  1301. * must NOT have the inode lock held at all. This is because we're
  1302. * calling into the buffer/page cache code and we can't hold the
  1303. * inode lock when we do so.
  1304. *
  1305. * We need to wait for any direct I/Os in flight to complete before we
  1306. * proceed with the truncate. This is needed to prevent the extents
  1307. * being read or written by the direct I/Os from being removed while the
  1308. * I/O is in flight as there is no other method of synchronising
  1309. * direct I/O with the truncate operation. Also, because we hold
  1310. * the IOLOCK in exclusive mode, we prevent new direct I/Os from being
  1311. * started until the truncate completes and drops the lock. Essentially,
  1312. * the vn_iowait() call forms an I/O barrier that provides strict ordering
  1313. * between direct I/Os and the truncate operation.
  1314. *
  1315. * The flags parameter can have either the value XFS_ITRUNC_DEFINITE
  1316. * or XFS_ITRUNC_MAYBE. The XFS_ITRUNC_MAYBE value should be used
  1317. * in the case that the caller is locking things out of order and
  1318. * may not be able to call xfs_itruncate_finish() with the inode lock
  1319. * held without dropping the I/O lock. If the caller must drop the
  1320. * I/O lock before calling xfs_itruncate_finish(), then xfs_itruncate_start()
  1321. * must be called again with all the same restrictions as the initial
  1322. * call.
  1323. */
  1324. void
  1325. xfs_itruncate_start(
  1326. xfs_inode_t *ip,
  1327. uint flags,
  1328. xfs_fsize_t new_size)
  1329. {
  1330. xfs_fsize_t last_byte;
  1331. xfs_off_t toss_start;
  1332. xfs_mount_t *mp;
  1333. bhv_vnode_t *vp;
  1334. ASSERT(ismrlocked(&ip->i_iolock, MR_UPDATE) != 0);
  1335. ASSERT((new_size == 0) || (new_size <= ip->i_d.di_size));
  1336. ASSERT((flags == XFS_ITRUNC_DEFINITE) ||
  1337. (flags == XFS_ITRUNC_MAYBE));
  1338. mp = ip->i_mount;
  1339. vp = XFS_ITOV(ip);
  1340. vn_iowait(vp); /* wait for the completion of any pending DIOs */
  1341. /*
  1342. * Call toss_pages or flushinval_pages to get rid of pages
  1343. * overlapping the region being removed. We have to use
  1344. * the less efficient flushinval_pages in the case that the
  1345. * caller may not be able to finish the truncate without
  1346. * dropping the inode's I/O lock. Make sure
  1347. * to catch any pages brought in by buffers overlapping
  1348. * the EOF by searching out beyond the isize by our
  1349. * block size. We round new_size up to a block boundary
  1350. * so that we don't toss things on the same block as
  1351. * new_size but before it.
  1352. *
  1353. * Before calling toss_page or flushinval_pages, make sure to
  1354. * call remapf() over the same region if the file is mapped.
  1355. * This frees up mapped file references to the pages in the
  1356. * given range and for the flushinval_pages case it ensures
  1357. * that we get the latest mapped changes flushed out.
  1358. */
  1359. toss_start = XFS_B_TO_FSB(mp, (xfs_ufsize_t)new_size);
  1360. toss_start = XFS_FSB_TO_B(mp, toss_start);
  1361. if (toss_start < 0) {
  1362. /*
  1363. * The place to start tossing is beyond our maximum
  1364. * file size, so there is no way that the data extended
  1365. * out there.
  1366. */
  1367. return;
  1368. }
  1369. last_byte = xfs_file_last_byte(ip);
  1370. xfs_itrunc_trace(XFS_ITRUNC_START, ip, flags, new_size, toss_start,
  1371. last_byte);
  1372. if (last_byte > toss_start) {
  1373. if (flags & XFS_ITRUNC_DEFINITE) {
  1374. bhv_vop_toss_pages(vp, toss_start, -1, FI_REMAPF_LOCKED);
  1375. } else {
  1376. bhv_vop_flushinval_pages(vp, toss_start, -1, FI_REMAPF_LOCKED);
  1377. }
  1378. }
  1379. #ifdef DEBUG
  1380. if (new_size == 0) {
  1381. ASSERT(VN_CACHED(vp) == 0);
  1382. }
  1383. #endif
  1384. }
  1385. /*
  1386. * Shrink the file to the given new_size. The new
  1387. * size must be smaller than the current size.
  1388. * This will free up the underlying blocks
  1389. * in the removed range after a call to xfs_itruncate_start()
  1390. * or xfs_atruncate_start().
  1391. *
  1392. * The transaction passed to this routine must have made
  1393. * a permanent log reservation of at least XFS_ITRUNCATE_LOG_RES.
  1394. * This routine may commit the given transaction and
  1395. * start new ones, so make sure everything involved in
  1396. * the transaction is tidy before calling here.
  1397. * Some transaction will be returned to the caller to be
  1398. * committed. The incoming transaction must already include
  1399. * the inode, and both inode locks must be held exclusively.
  1400. * The inode must also be "held" within the transaction. On
  1401. * return the inode will be "held" within the returned transaction.
  1402. * This routine does NOT require any disk space to be reserved
  1403. * for it within the transaction.
  1404. *
  1405. * The fork parameter must be either xfs_attr_fork or xfs_data_fork,
  1406. * and it indicates the fork which is to be truncated. For the
  1407. * attribute fork we only support truncation to size 0.
  1408. *
  1409. * We use the sync parameter to indicate whether or not the first
  1410. * transaction we perform might have to be synchronous. For the attr fork,
  1411. * it needs to be so if the unlink of the inode is not yet known to be
  1412. * permanent in the log. This keeps us from freeing and reusing the
  1413. * blocks of the attribute fork before the unlink of the inode becomes
  1414. * permanent.
  1415. *
  1416. * For the data fork, we normally have to run synchronously if we're
  1417. * being called out of the inactive path or we're being called
  1418. * out of the create path where we're truncating an existing file.
  1419. * Either way, the truncate needs to be sync so blocks don't reappear
  1420. * in the file with altered data in case of a crash. wsync filesystems
  1421. * can run the first case async because anything that shrinks the inode
  1422. * has to run sync so by the time we're called here from inactive, the
  1423. * inode size is permanently set to 0.
  1424. *
  1425. * Calls from the truncate path always need to be sync unless we're
  1426. * in a wsync filesystem and the file has already been unlinked.
  1427. *
  1428. * The caller is responsible for correctly setting the sync parameter.
  1429. * It gets too hard for us to guess here which path we're being called
  1430. * out of just based on inode state.
  1431. */
  1432. int
  1433. xfs_itruncate_finish(
  1434. xfs_trans_t **tp,
  1435. xfs_inode_t *ip,
  1436. xfs_fsize_t new_size,
  1437. int fork,
  1438. int sync)
  1439. {
  1440. xfs_fsblock_t first_block;
  1441. xfs_fileoff_t first_unmap_block;
  1442. xfs_fileoff_t last_block;
  1443. xfs_filblks_t unmap_len=0;
  1444. xfs_mount_t *mp;
  1445. xfs_trans_t *ntp;
  1446. int done;
  1447. int committed;
  1448. xfs_bmap_free_t free_list;
  1449. int error;
  1450. ASSERT(ismrlocked(&ip->i_iolock, MR_UPDATE) != 0);
  1451. ASSERT(ismrlocked(&ip->i_lock, MR_UPDATE) != 0);
  1452. ASSERT((new_size == 0) || (new_size <= ip->i_d.di_size));
  1453. ASSERT(*tp != NULL);
  1454. ASSERT((*tp)->t_flags & XFS_TRANS_PERM_LOG_RES);
  1455. ASSERT(ip->i_transp == *tp);
  1456. ASSERT(ip->i_itemp != NULL);
  1457. ASSERT(ip->i_itemp->ili_flags & XFS_ILI_HOLD);
  1458. ntp = *tp;
  1459. mp = (ntp)->t_mountp;
  1460. ASSERT(! XFS_NOT_DQATTACHED(mp, ip));
  1461. /*
  1462. * We only support truncating the entire attribute fork.
  1463. */
  1464. if (fork == XFS_ATTR_FORK) {
  1465. new_size = 0LL;
  1466. }
  1467. first_unmap_block = XFS_B_TO_FSB(mp, (xfs_ufsize_t)new_size);
  1468. xfs_itrunc_trace(XFS_ITRUNC_FINISH1, ip, 0, new_size, 0, 0);
  1469. /*
  1470. * The first thing we do is set the size to new_size permanently
  1471. * on disk. This way we don't have to worry about anyone ever
  1472. * being able to look at the data being freed even in the face
  1473. * of a crash. What we're getting around here is the case where
  1474. * we free a block, it is allocated to another file, it is written
  1475. * to, and then we crash. If the new data gets written to the
  1476. * file but the log buffers containing the free and reallocation
  1477. * don't, then we'd end up with garbage in the blocks being freed.
  1478. * As long as we make the new_size permanent before actually
  1479. * freeing any blocks it doesn't matter if they get writtten to.
  1480. *
  1481. * The callers must signal into us whether or not the size
  1482. * setting here must be synchronous. There are a few cases
  1483. * where it doesn't have to be synchronous. Those cases
  1484. * occur if the file is unlinked and we know the unlink is
  1485. * permanent or if the blocks being truncated are guaranteed
  1486. * to be beyond the inode eof (regardless of the link count)
  1487. * and the eof value is permanent. Both of these cases occur
  1488. * only on wsync-mounted filesystems. In those cases, we're
  1489. * guaranteed that no user will ever see the data in the blocks
  1490. * that are being truncated so the truncate can run async.
  1491. * In the free beyond eof case, the file may wind up with
  1492. * more blocks allocated to it than it needs if we crash
  1493. * and that won't get fixed until the next time the file
  1494. * is re-opened and closed but that's ok as that shouldn't
  1495. * be too many blocks.
  1496. *
  1497. * However, we can't just make all wsync xactions run async
  1498. * because there's one call out of the create path that needs
  1499. * to run sync where it's truncating an existing file to size
  1500. * 0 whose size is > 0.
  1501. *
  1502. * It's probably possible to come up with a test in this
  1503. * routine that would correctly distinguish all the above
  1504. * cases from the values of the function parameters and the
  1505. * inode state but for sanity's sake, I've decided to let the
  1506. * layers above just tell us. It's simpler to correctly figure
  1507. * out in the layer above exactly under what conditions we
  1508. * can run async and I think it's easier for others read and
  1509. * follow the logic in case something has to be changed.
  1510. * cscope is your friend -- rcc.
  1511. *
  1512. * The attribute fork is much simpler.
  1513. *
  1514. * For the attribute fork we allow the caller to tell us whether
  1515. * the unlink of the inode that led to this call is yet permanent
  1516. * in the on disk log. If it is not and we will be freeing extents
  1517. * in this inode then we make the first transaction synchronous
  1518. * to make sure that the unlink is permanent by the time we free
  1519. * the blocks.
  1520. */
  1521. if (fork == XFS_DATA_FORK) {
  1522. if (ip->i_d.di_nextents > 0) {
  1523. ip->i_d.di_size = new_size;
  1524. xfs_trans_log_inode(ntp, ip, XFS_ILOG_CORE);
  1525. }
  1526. } else if (sync) {
  1527. ASSERT(!(mp->m_flags & XFS_MOUNT_WSYNC));
  1528. if (ip->i_d.di_anextents > 0)
  1529. xfs_trans_set_sync(ntp);
  1530. }
  1531. ASSERT(fork == XFS_DATA_FORK ||
  1532. (fork == XFS_ATTR_FORK &&
  1533. ((sync && !(mp->m_flags & XFS_MOUNT_WSYNC)) ||
  1534. (sync == 0 && (mp->m_flags & XFS_MOUNT_WSYNC)))));
  1535. /*
  1536. * Since it is possible for space to become allocated beyond
  1537. * the end of the file (in a crash where the space is allocated
  1538. * but the inode size is not yet updated), simply remove any
  1539. * blocks which show up between the new EOF and the maximum
  1540. * possible file size. If the first block to be removed is
  1541. * beyond the maximum file size (ie it is the same as last_block),
  1542. * then there is nothing to do.
  1543. */
  1544. last_block = XFS_B_TO_FSB(mp, (xfs_ufsize_t)XFS_MAXIOFFSET(mp));
  1545. ASSERT(first_unmap_block <= last_block);
  1546. done = 0;
  1547. if (last_block == first_unmap_block) {
  1548. done = 1;
  1549. } else {
  1550. unmap_len = last_block - first_unmap_block + 1;
  1551. }
  1552. while (!done) {
  1553. /*
  1554. * Free up up to XFS_ITRUNC_MAX_EXTENTS. xfs_bunmapi()
  1555. * will tell us whether it freed the entire range or
  1556. * not. If this is a synchronous mount (wsync),
  1557. * then we can tell bunmapi to keep all the
  1558. * transactions asynchronous since the unlink
  1559. * transaction that made this inode inactive has
  1560. * already hit the disk. There's no danger of
  1561. * the freed blocks being reused, there being a
  1562. * crash, and the reused blocks suddenly reappearing
  1563. * in this file with garbage in them once recovery
  1564. * runs.
  1565. */
  1566. XFS_BMAP_INIT(&free_list, &first_block);
  1567. error = XFS_BUNMAPI(mp, ntp, &ip->i_iocore,
  1568. first_unmap_block, unmap_len,
  1569. XFS_BMAPI_AFLAG(fork) |
  1570. (sync ? 0 : XFS_BMAPI_ASYNC),
  1571. XFS_ITRUNC_MAX_EXTENTS,
  1572. &first_block, &free_list,
  1573. NULL, &done);
  1574. if (error) {
  1575. /*
  1576. * If the bunmapi call encounters an error,
  1577. * return to the caller where the transaction
  1578. * can be properly aborted. We just need to
  1579. * make sure we're not holding any resources
  1580. * that we were not when we came in.
  1581. */
  1582. xfs_bmap_cancel(&free_list);
  1583. return error;
  1584. }
  1585. /*
  1586. * Duplicate the transaction that has the permanent
  1587. * reservation and commit the old transaction.
  1588. */
  1589. error = xfs_bmap_finish(tp, &free_list, &committed);
  1590. ntp = *tp;
  1591. if (error) {
  1592. /*
  1593. * If the bmap finish call encounters an error,
  1594. * return to the caller where the transaction
  1595. * can be properly aborted. We just need to
  1596. * make sure we're not holding any resources
  1597. * that we were not when we came in.
  1598. *
  1599. * Aborting from this point might lose some
  1600. * blocks in the file system, but oh well.
  1601. */
  1602. xfs_bmap_cancel(&free_list);
  1603. if (committed) {
  1604. /*
  1605. * If the passed in transaction committed
  1606. * in xfs_bmap_finish(), then we want to
  1607. * add the inode to this one before returning.
  1608. * This keeps things simple for the higher
  1609. * level code, because it always knows that
  1610. * the inode is locked and held in the
  1611. * transaction that returns to it whether
  1612. * errors occur or not. We don't mark the
  1613. * inode dirty so that this transaction can
  1614. * be easily aborted if possible.
  1615. */
  1616. xfs_trans_ijoin(ntp, ip,
  1617. XFS_ILOCK_EXCL | XFS_IOLOCK_EXCL);
  1618. xfs_trans_ihold(ntp, ip);
  1619. }
  1620. return error;
  1621. }
  1622. if (committed) {
  1623. /*
  1624. * The first xact was committed,
  1625. * so add the inode to the new one.
  1626. * Mark it dirty so it will be logged
  1627. * and moved forward in the log as
  1628. * part of every commit.
  1629. */
  1630. xfs_trans_ijoin(ntp, ip,
  1631. XFS_ILOCK_EXCL | XFS_IOLOCK_EXCL);
  1632. xfs_trans_ihold(ntp, ip);
  1633. xfs_trans_log_inode(ntp, ip, XFS_ILOG_CORE);
  1634. }
  1635. ntp = xfs_trans_dup(ntp);
  1636. (void) xfs_trans_commit(*tp, 0, NULL);
  1637. *tp = ntp;
  1638. error = xfs_trans_reserve(ntp, 0, XFS_ITRUNCATE_LOG_RES(mp), 0,
  1639. XFS_TRANS_PERM_LOG_RES,
  1640. XFS_ITRUNCATE_LOG_COUNT);
  1641. /*
  1642. * Add the inode being truncated to the next chained
  1643. * transaction.
  1644. */
  1645. xfs_trans_ijoin(ntp, ip, XFS_ILOCK_EXCL | XFS_IOLOCK_EXCL);
  1646. xfs_trans_ihold(ntp, ip);
  1647. if (error)
  1648. return (error);
  1649. }
  1650. /*
  1651. * Only update the size in the case of the data fork, but
  1652. * always re-log the inode so that our permanent transaction
  1653. * can keep on rolling it forward in the log.
  1654. */
  1655. if (fork == XFS_DATA_FORK) {
  1656. xfs_isize_check(mp, ip, new_size);
  1657. ip->i_d.di_size = new_size;
  1658. }
  1659. xfs_trans_log_inode(ntp, ip, XFS_ILOG_CORE);
  1660. ASSERT((new_size != 0) ||
  1661. (fork == XFS_ATTR_FORK) ||
  1662. (ip->i_delayed_blks == 0));
  1663. ASSERT((new_size != 0) ||
  1664. (fork == XFS_ATTR_FORK) ||
  1665. (ip->i_d.di_nextents == 0));
  1666. xfs_itrunc_trace(XFS_ITRUNC_FINISH2, ip, 0, new_size, 0, 0);
  1667. return 0;
  1668. }
  1669. /*
  1670. * xfs_igrow_start
  1671. *
  1672. * Do the first part of growing a file: zero any data in the last
  1673. * block that is beyond the old EOF. We need to do this before
  1674. * the inode is joined to the transaction to modify the i_size.
  1675. * That way we can drop the inode lock and call into the buffer
  1676. * cache to get the buffer mapping the EOF.
  1677. */
  1678. int
  1679. xfs_igrow_start(
  1680. xfs_inode_t *ip,
  1681. xfs_fsize_t new_size,
  1682. cred_t *credp)
  1683. {
  1684. int error;
  1685. ASSERT(ismrlocked(&(ip->i_lock), MR_UPDATE) != 0);
  1686. ASSERT(ismrlocked(&(ip->i_iolock), MR_UPDATE) != 0);
  1687. ASSERT(new_size > ip->i_d.di_size);
  1688. /*
  1689. * Zero any pages that may have been created by
  1690. * xfs_write_file() beyond the end of the file
  1691. * and any blocks between the old and new file sizes.
  1692. */
  1693. error = xfs_zero_eof(XFS_ITOV(ip), &ip->i_iocore, new_size,
  1694. ip->i_d.di_size);
  1695. return error;
  1696. }
  1697. /*
  1698. * xfs_igrow_finish
  1699. *
  1700. * This routine is called to extend the size of a file.
  1701. * The inode must have both the iolock and the ilock locked
  1702. * for update and it must be a part of the current transaction.
  1703. * The xfs_igrow_start() function must have been called previously.
  1704. * If the change_flag is not zero, the inode change timestamp will
  1705. * be updated.
  1706. */
  1707. void
  1708. xfs_igrow_finish(
  1709. xfs_trans_t *tp,
  1710. xfs_inode_t *ip,
  1711. xfs_fsize_t new_size,
  1712. int change_flag)
  1713. {
  1714. ASSERT(ismrlocked(&(ip->i_lock), MR_UPDATE) != 0);
  1715. ASSERT(ismrlocked(&(ip->i_iolock), MR_UPDATE) != 0);
  1716. ASSERT(ip->i_transp == tp);
  1717. ASSERT(new_size > ip->i_d.di_size);
  1718. /*
  1719. * Update the file size. Update the inode change timestamp
  1720. * if change_flag set.
  1721. */
  1722. ip->i_d.di_size = new_size;
  1723. if (change_flag)
  1724. xfs_ichgtime(ip, XFS_ICHGTIME_CHG);
  1725. xfs_trans_log_inode(tp, ip, XFS_ILOG_CORE);
  1726. }
  1727. /*
  1728. * This is called when the inode's link count goes to 0.
  1729. * We place the on-disk inode on a list in the AGI. It
  1730. * will be pulled from this list when the inode is freed.
  1731. */
  1732. int
  1733. xfs_iunlink(
  1734. xfs_trans_t *tp,
  1735. xfs_inode_t *ip)
  1736. {
  1737. xfs_mount_t *mp;
  1738. xfs_agi_t *agi;
  1739. xfs_dinode_t *dip;
  1740. xfs_buf_t *agibp;
  1741. xfs_buf_t *ibp;
  1742. xfs_agnumber_t agno;
  1743. xfs_daddr_t agdaddr;
  1744. xfs_agino_t agino;
  1745. short bucket_index;
  1746. int offset;
  1747. int error;
  1748. int agi_ok;
  1749. ASSERT(ip->i_d.di_nlink == 0);
  1750. ASSERT(ip->i_d.di_mode != 0);
  1751. ASSERT(ip->i_transp == tp);
  1752. mp = tp->t_mountp;
  1753. agno = XFS_INO_TO_AGNO(mp, ip->i_ino);
  1754. agdaddr = XFS_AG_DADDR(mp, agno, XFS_AGI_DADDR(mp));
  1755. /*
  1756. * Get the agi buffer first. It ensures lock ordering
  1757. * on the list.
  1758. */
  1759. error = xfs_trans_read_buf(mp, tp, mp->m_ddev_targp, agdaddr,
  1760. XFS_FSS_TO_BB(mp, 1), 0, &agibp);
  1761. if (error) {
  1762. return error;
  1763. }
  1764. /*
  1765. * Validate the magic number of the agi block.
  1766. */
  1767. agi = XFS_BUF_TO_AGI(agibp);
  1768. agi_ok =
  1769. be32_to_cpu(agi->agi_magicnum) == XFS_AGI_MAGIC &&
  1770. XFS_AGI_GOOD_VERSION(be32_to_cpu(agi->agi_versionnum));
  1771. if (unlikely(XFS_TEST_ERROR(!agi_ok, mp, XFS_ERRTAG_IUNLINK,
  1772. XFS_RANDOM_IUNLINK))) {
  1773. XFS_CORRUPTION_ERROR("xfs_iunlink", XFS_ERRLEVEL_LOW, mp, agi);
  1774. xfs_trans_brelse(tp, agibp);
  1775. return XFS_ERROR(EFSCORRUPTED);
  1776. }
  1777. /*
  1778. * Get the index into the agi hash table for the
  1779. * list this inode will go on.
  1780. */
  1781. agino = XFS_INO_TO_AGINO(mp, ip->i_ino);
  1782. ASSERT(agino != 0);
  1783. bucket_index = agino % XFS_AGI_UNLINKED_BUCKETS;
  1784. ASSERT(agi->agi_unlinked[bucket_index]);
  1785. ASSERT(be32_to_cpu(agi->agi_unlinked[bucket_index]) != agino);
  1786. if (be32_to_cpu(agi->agi_unlinked[bucket_index]) != NULLAGINO) {
  1787. /*
  1788. * There is already another inode in the bucket we need
  1789. * to add ourselves to. Add us at the front of the list.
  1790. * Here we put the head pointer into our next pointer,
  1791. * and then we fall through to point the head at us.
  1792. */
  1793. error = xfs_itobp(mp, tp, ip, &dip, &ibp, 0, 0);
  1794. if (error) {
  1795. return error;
  1796. }
  1797. ASSERT(INT_GET(dip->di_next_unlinked, ARCH_CONVERT) == NULLAGINO);
  1798. ASSERT(dip->di_next_unlinked);
  1799. /* both on-disk, don't endian flip twice */
  1800. dip->di_next_unlinked = agi->agi_unlinked[bucket_index];
  1801. offset = ip->i_boffset +
  1802. offsetof(xfs_dinode_t, di_next_unlinked);
  1803. xfs_trans_inode_buf(tp, ibp);
  1804. xfs_trans_log_buf(tp, ibp, offset,
  1805. (offset + sizeof(xfs_agino_t) - 1));
  1806. xfs_inobp_check(mp, ibp);
  1807. }
  1808. /*
  1809. * Point the bucket head pointer at the inode being inserted.
  1810. */
  1811. ASSERT(agino != 0);
  1812. agi->agi_unlinked[bucket_index] = cpu_to_be32(agino);
  1813. offset = offsetof(xfs_agi_t, agi_unlinked) +
  1814. (sizeof(xfs_agino_t) * bucket_index);
  1815. xfs_trans_log_buf(tp, agibp, offset,
  1816. (offset + sizeof(xfs_agino_t) - 1));
  1817. return 0;
  1818. }
  1819. /*
  1820. * Pull the on-disk inode from the AGI unlinked list.
  1821. */
  1822. STATIC int
  1823. xfs_iunlink_remove(
  1824. xfs_trans_t *tp,
  1825. xfs_inode_t *ip)
  1826. {
  1827. xfs_ino_t next_ino;
  1828. xfs_mount_t *mp;
  1829. xfs_agi_t *agi;
  1830. xfs_dinode_t *dip;
  1831. xfs_buf_t *agibp;
  1832. xfs_buf_t *ibp;
  1833. xfs_agnumber_t agno;
  1834. xfs_daddr_t agdaddr;
  1835. xfs_agino_t agino;
  1836. xfs_agino_t next_agino;
  1837. xfs_buf_t *last_ibp;
  1838. xfs_dinode_t *last_dip = NULL;
  1839. short bucket_index;
  1840. int offset, last_offset = 0;
  1841. int error;
  1842. int agi_ok;
  1843. /*
  1844. * First pull the on-disk inode from the AGI unlinked list.
  1845. */
  1846. mp = tp->t_mountp;
  1847. agno = XFS_INO_TO_AGNO(mp, ip->i_ino);
  1848. agdaddr = XFS_AG_DADDR(mp, agno, XFS_AGI_DADDR(mp));
  1849. /*
  1850. * Get the agi buffer first. It ensures lock ordering
  1851. * on the list.
  1852. */
  1853. error = xfs_trans_read_buf(mp, tp, mp->m_ddev_targp, agdaddr,
  1854. XFS_FSS_TO_BB(mp, 1), 0, &agibp);
  1855. if (error) {
  1856. cmn_err(CE_WARN,
  1857. "xfs_iunlink_remove: xfs_trans_read_buf() returned an error %d on %s. Returning error.",
  1858. error, mp->m_fsname);
  1859. return error;
  1860. }
  1861. /*
  1862. * Validate the magic number of the agi block.
  1863. */
  1864. agi = XFS_BUF_TO_AGI(agibp);
  1865. agi_ok =
  1866. be32_to_cpu(agi->agi_magicnum) == XFS_AGI_MAGIC &&
  1867. XFS_AGI_GOOD_VERSION(be32_to_cpu(agi->agi_versionnum));
  1868. if (unlikely(XFS_TEST_ERROR(!agi_ok, mp, XFS_ERRTAG_IUNLINK_REMOVE,
  1869. XFS_RANDOM_IUNLINK_REMOVE))) {
  1870. XFS_CORRUPTION_ERROR("xfs_iunlink_remove", XFS_ERRLEVEL_LOW,
  1871. mp, agi);
  1872. xfs_trans_brelse(tp, agibp);
  1873. cmn_err(CE_WARN,
  1874. "xfs_iunlink_remove: XFS_TEST_ERROR() returned an error on %s. Returning EFSCORRUPTED.",
  1875. mp->m_fsname);
  1876. return XFS_ERROR(EFSCORRUPTED);
  1877. }
  1878. /*
  1879. * Get the index into the agi hash table for the
  1880. * list this inode will go on.
  1881. */
  1882. agino = XFS_INO_TO_AGINO(mp, ip->i_ino);
  1883. ASSERT(agino != 0);
  1884. bucket_index = agino % XFS_AGI_UNLINKED_BUCKETS;
  1885. ASSERT(be32_to_cpu(agi->agi_unlinked[bucket_index]) != NULLAGINO);
  1886. ASSERT(agi->agi_unlinked[bucket_index]);
  1887. if (be32_to_cpu(agi->agi_unlinked[bucket_index]) == agino) {
  1888. /*
  1889. * We're at the head of the list. Get the inode's
  1890. * on-disk buffer to see if there is anyone after us
  1891. * on the list. Only modify our next pointer if it
  1892. * is not already NULLAGINO. This saves us the overhead
  1893. * of dealing with the buffer when there is no need to
  1894. * change it.
  1895. */
  1896. error = xfs_itobp(mp, tp, ip, &dip, &ibp, 0, 0);
  1897. if (error) {
  1898. cmn_err(CE_WARN,
  1899. "xfs_iunlink_remove: xfs_itobp() returned an error %d on %s. Returning error.",
  1900. error, mp->m_fsname);
  1901. return error;
  1902. }
  1903. next_agino = INT_GET(dip->di_next_unlinked, ARCH_CONVERT);
  1904. ASSERT(next_agino != 0);
  1905. if (next_agino != NULLAGINO) {
  1906. INT_SET(dip->di_next_unlinked, ARCH_CONVERT, NULLAGINO);
  1907. offset = ip->i_boffset +
  1908. offsetof(xfs_dinode_t, di_next_unlinked);
  1909. xfs_trans_inode_buf(tp, ibp);
  1910. xfs_trans_log_buf(tp, ibp, offset,
  1911. (offset + sizeof(xfs_agino_t) - 1));
  1912. xfs_inobp_check(mp, ibp);
  1913. } else {
  1914. xfs_trans_brelse(tp, ibp);
  1915. }
  1916. /*
  1917. * Point the bucket head pointer at the next inode.
  1918. */
  1919. ASSERT(next_agino != 0);
  1920. ASSERT(next_agino != agino);
  1921. agi->agi_unlinked[bucket_index] = cpu_to_be32(next_agino);
  1922. offset = offsetof(xfs_agi_t, agi_unlinked) +
  1923. (sizeof(xfs_agino_t) * bucket_index);
  1924. xfs_trans_log_buf(tp, agibp, offset,
  1925. (offset + sizeof(xfs_agino_t) - 1));
  1926. } else {
  1927. /*
  1928. * We need to search the list for the inode being freed.
  1929. */
  1930. next_agino = be32_to_cpu(agi->agi_unlinked[bucket_index]);
  1931. last_ibp = NULL;
  1932. while (next_agino != agino) {
  1933. /*
  1934. * If the last inode wasn't the one pointing to
  1935. * us, then release its buffer since we're not
  1936. * going to do anything with it.
  1937. */
  1938. if (last_ibp != NULL) {
  1939. xfs_trans_brelse(tp, last_ibp);
  1940. }
  1941. next_ino = XFS_AGINO_TO_INO(mp, agno, next_agino);
  1942. error = xfs_inotobp(mp, tp, next_ino, &last_dip,
  1943. &last_ibp, &last_offset);
  1944. if (error) {
  1945. cmn_err(CE_WARN,
  1946. "xfs_iunlink_remove: xfs_inotobp() returned an error %d on %s. Returning error.",
  1947. error, mp->m_fsname);
  1948. return error;
  1949. }
  1950. next_agino = INT_GET(last_dip->di_next_unlinked, ARCH_CONVERT);
  1951. ASSERT(next_agino != NULLAGINO);
  1952. ASSERT(next_agino != 0);
  1953. }
  1954. /*
  1955. * Now last_ibp points to the buffer previous to us on
  1956. * the unlinked list. Pull us from the list.
  1957. */
  1958. error = xfs_itobp(mp, tp, ip, &dip, &ibp, 0, 0);
  1959. if (error) {
  1960. cmn_err(CE_WARN,
  1961. "xfs_iunlink_remove: xfs_itobp() returned an error %d on %s. Returning error.",
  1962. error, mp->m_fsname);
  1963. return error;
  1964. }
  1965. next_agino = INT_GET(dip->di_next_unlinked, ARCH_CONVERT);
  1966. ASSERT(next_agino != 0);
  1967. ASSERT(next_agino != agino);
  1968. if (next_agino != NULLAGINO) {
  1969. INT_SET(dip->di_next_unlinked, ARCH_CONVERT, NULLAGINO);
  1970. offset = ip->i_boffset +
  1971. offsetof(xfs_dinode_t, di_next_unlinked);
  1972. xfs_trans_inode_buf(tp, ibp);
  1973. xfs_trans_log_buf(tp, ibp, offset,
  1974. (offset + sizeof(xfs_agino_t) - 1));
  1975. xfs_inobp_check(mp, ibp);
  1976. } else {
  1977. xfs_trans_brelse(tp, ibp);
  1978. }
  1979. /*
  1980. * Point the previous inode on the list to the next inode.
  1981. */
  1982. INT_SET(last_dip->di_next_unlinked, ARCH_CONVERT, next_agino);
  1983. ASSERT(next_agino != 0);
  1984. offset = last_offset + offsetof(xfs_dinode_t, di_next_unlinked);
  1985. xfs_trans_inode_buf(tp, last_ibp);
  1986. xfs_trans_log_buf(tp, last_ibp, offset,
  1987. (offset + sizeof(xfs_agino_t) - 1));
  1988. xfs_inobp_check(mp, last_ibp);
  1989. }
  1990. return 0;
  1991. }
  1992. STATIC_INLINE int xfs_inode_clean(xfs_inode_t *ip)
  1993. {
  1994. return (((ip->i_itemp == NULL) ||
  1995. !(ip->i_itemp->ili_format.ilf_fields & XFS_ILOG_ALL)) &&
  1996. (ip->i_update_core == 0));
  1997. }
  1998. STATIC void
  1999. xfs_ifree_cluster(
  2000. xfs_inode_t *free_ip,
  2001. xfs_trans_t *tp,
  2002. xfs_ino_t inum)
  2003. {
  2004. xfs_mount_t *mp = free_ip->i_mount;
  2005. int blks_per_cluster;
  2006. int nbufs;
  2007. int ninodes;
  2008. int i, j, found, pre_flushed;
  2009. xfs_daddr_t blkno;
  2010. xfs_buf_t *bp;
  2011. xfs_ihash_t *ih;
  2012. xfs_inode_t *ip, **ip_found;
  2013. xfs_inode_log_item_t *iip;
  2014. xfs_log_item_t *lip;
  2015. SPLDECL(s);
  2016. if (mp->m_sb.sb_blocksize >= XFS_INODE_CLUSTER_SIZE(mp)) {
  2017. blks_per_cluster = 1;
  2018. ninodes = mp->m_sb.sb_inopblock;
  2019. nbufs = XFS_IALLOC_BLOCKS(mp);
  2020. } else {
  2021. blks_per_cluster = XFS_INODE_CLUSTER_SIZE(mp) /
  2022. mp->m_sb.sb_blocksize;
  2023. ninodes = blks_per_cluster * mp->m_sb.sb_inopblock;
  2024. nbufs = XFS_IALLOC_BLOCKS(mp) / blks_per_cluster;
  2025. }
  2026. ip_found = kmem_alloc(ninodes * sizeof(xfs_inode_t *), KM_NOFS);
  2027. for (j = 0; j < nbufs; j++, inum += ninodes) {
  2028. blkno = XFS_AGB_TO_DADDR(mp, XFS_INO_TO_AGNO(mp, inum),
  2029. XFS_INO_TO_AGBNO(mp, inum));
  2030. /*
  2031. * Look for each inode in memory and attempt to lock it,
  2032. * we can be racing with flush and tail pushing here.
  2033. * any inode we get the locks on, add to an array of
  2034. * inode items to process later.
  2035. *
  2036. * The get the buffer lock, we could beat a flush
  2037. * or tail pushing thread to the lock here, in which
  2038. * case they will go looking for the inode buffer
  2039. * and fail, we need some other form of interlock
  2040. * here.
  2041. */
  2042. found = 0;
  2043. for (i = 0; i < ninodes; i++) {
  2044. ih = XFS_IHASH(mp, inum + i);
  2045. read_lock(&ih->ih_lock);
  2046. for (ip = ih->ih_next; ip != NULL; ip = ip->i_next) {
  2047. if (ip->i_ino == inum + i)
  2048. break;
  2049. }
  2050. /* Inode not in memory or we found it already,
  2051. * nothing to do
  2052. */
  2053. if (!ip || xfs_iflags_test(ip, XFS_ISTALE)) {
  2054. read_unlock(&ih->ih_lock);
  2055. continue;
  2056. }
  2057. if (xfs_inode_clean(ip)) {
  2058. read_unlock(&ih->ih_lock);
  2059. continue;
  2060. }
  2061. /* If we can get the locks then add it to the
  2062. * list, otherwise by the time we get the bp lock
  2063. * below it will already be attached to the
  2064. * inode buffer.
  2065. */
  2066. /* This inode will already be locked - by us, lets
  2067. * keep it that way.
  2068. */
  2069. if (ip == free_ip) {
  2070. if (xfs_iflock_nowait(ip)) {
  2071. xfs_iflags_set(ip, XFS_ISTALE);
  2072. if (xfs_inode_clean(ip)) {
  2073. xfs_ifunlock(ip);
  2074. } else {
  2075. ip_found[found++] = ip;
  2076. }
  2077. }
  2078. read_unlock(&ih->ih_lock);
  2079. continue;
  2080. }
  2081. if (xfs_ilock_nowait(ip, XFS_ILOCK_EXCL)) {
  2082. if (xfs_iflock_nowait(ip)) {
  2083. xfs_iflags_set(ip, XFS_ISTALE);
  2084. if (xfs_inode_clean(ip)) {
  2085. xfs_ifunlock(ip);
  2086. xfs_iunlock(ip, XFS_ILOCK_EXCL);
  2087. } else {
  2088. ip_found[found++] = ip;
  2089. }
  2090. } else {
  2091. xfs_iunlock(ip, XFS_ILOCK_EXCL);
  2092. }
  2093. }
  2094. read_unlock(&ih->ih_lock);
  2095. }
  2096. bp = xfs_trans_get_buf(tp, mp->m_ddev_targp, blkno,
  2097. mp->m_bsize * blks_per_cluster,
  2098. XFS_BUF_LOCK);
  2099. pre_flushed = 0;
  2100. lip = XFS_BUF_FSPRIVATE(bp, xfs_log_item_t *);
  2101. while (lip) {
  2102. if (lip->li_type == XFS_LI_INODE) {
  2103. iip = (xfs_inode_log_item_t *)lip;
  2104. ASSERT(iip->ili_logged == 1);
  2105. lip->li_cb = (void(*)(xfs_buf_t*,xfs_log_item_t*)) xfs_istale_done;
  2106. AIL_LOCK(mp,s);
  2107. iip->ili_flush_lsn = iip->ili_item.li_lsn;
  2108. AIL_UNLOCK(mp, s);
  2109. xfs_iflags_set(iip->ili_inode, XFS_ISTALE);
  2110. pre_flushed++;
  2111. }
  2112. lip = lip->li_bio_list;
  2113. }
  2114. for (i = 0; i < found; i++) {
  2115. ip = ip_found[i];
  2116. iip = ip->i_itemp;
  2117. if (!iip) {
  2118. ip->i_update_core = 0;
  2119. xfs_ifunlock(ip);
  2120. xfs_iunlock(ip, XFS_ILOCK_EXCL);
  2121. continue;
  2122. }
  2123. iip->ili_last_fields = iip->ili_format.ilf_fields;
  2124. iip->ili_format.ilf_fields = 0;
  2125. iip->ili_logged = 1;
  2126. AIL_LOCK(mp,s);
  2127. iip->ili_flush_lsn = iip->ili_item.li_lsn;
  2128. AIL_UNLOCK(mp, s);
  2129. xfs_buf_attach_iodone(bp,
  2130. (void(*)(xfs_buf_t*,xfs_log_item_t*))
  2131. xfs_istale_done, (xfs_log_item_t *)iip);
  2132. if (ip != free_ip) {
  2133. xfs_iunlock(ip, XFS_ILOCK_EXCL);
  2134. }
  2135. }
  2136. if (found || pre_flushed)
  2137. xfs_trans_stale_inode_buf(tp, bp);
  2138. xfs_trans_binval(tp, bp);
  2139. }
  2140. kmem_free(ip_found, ninodes * sizeof(xfs_inode_t *));
  2141. }
  2142. /*
  2143. * This is called to return an inode to the inode free list.
  2144. * The inode should already be truncated to 0 length and have
  2145. * no pages associated with it. This routine also assumes that
  2146. * the inode is already a part of the transaction.
  2147. *
  2148. * The on-disk copy of the inode will have been added to the list
  2149. * of unlinked inodes in the AGI. We need to remove the inode from
  2150. * that list atomically with respect to freeing it here.
  2151. */
  2152. int
  2153. xfs_ifree(
  2154. xfs_trans_t *tp,
  2155. xfs_inode_t *ip,
  2156. xfs_bmap_free_t *flist)
  2157. {
  2158. int error;
  2159. int delete;
  2160. xfs_ino_t first_ino;
  2161. ASSERT(ismrlocked(&ip->i_lock, MR_UPDATE));
  2162. ASSERT(ip->i_transp == tp);
  2163. ASSERT(ip->i_d.di_nlink == 0);
  2164. ASSERT(ip->i_d.di_nextents == 0);
  2165. ASSERT(ip->i_d.di_anextents == 0);
  2166. ASSERT((ip->i_d.di_size == 0) ||
  2167. ((ip->i_d.di_mode & S_IFMT) != S_IFREG));
  2168. ASSERT(ip->i_d.di_nblocks == 0);
  2169. /*
  2170. * Pull the on-disk inode from the AGI unlinked list.
  2171. */
  2172. error = xfs_iunlink_remove(tp, ip);
  2173. if (error != 0) {
  2174. return error;
  2175. }
  2176. error = xfs_difree(tp, ip->i_ino, flist, &delete, &first_ino);
  2177. if (error != 0) {
  2178. return error;
  2179. }
  2180. ip->i_d.di_mode = 0; /* mark incore inode as free */
  2181. ip->i_d.di_flags = 0;
  2182. ip->i_d.di_dmevmask = 0;
  2183. ip->i_d.di_forkoff = 0; /* mark the attr fork not in use */
  2184. ip->i_df.if_ext_max =
  2185. XFS_IFORK_DSIZE(ip) / (uint)sizeof(xfs_bmbt_rec_t);
  2186. ip->i_d.di_format = XFS_DINODE_FMT_EXTENTS;
  2187. ip->i_d.di_aformat = XFS_DINODE_FMT_EXTENTS;
  2188. /*
  2189. * Bump the generation count so no one will be confused
  2190. * by reincarnations of this inode.
  2191. */
  2192. ip->i_d.di_gen++;
  2193. xfs_trans_log_inode(tp, ip, XFS_ILOG_CORE);
  2194. if (delete) {
  2195. xfs_ifree_cluster(ip, tp, first_ino);
  2196. }
  2197. return 0;
  2198. }
  2199. /*
  2200. * Reallocate the space for if_broot based on the number of records
  2201. * being added or deleted as indicated in rec_diff. Move the records
  2202. * and pointers in if_broot to fit the new size. When shrinking this
  2203. * will eliminate holes between the records and pointers created by
  2204. * the caller. When growing this will create holes to be filled in
  2205. * by the caller.
  2206. *
  2207. * The caller must not request to add more records than would fit in
  2208. * the on-disk inode root. If the if_broot is currently NULL, then
  2209. * if we adding records one will be allocated. The caller must also
  2210. * not request that the number of records go below zero, although
  2211. * it can go to zero.
  2212. *
  2213. * ip -- the inode whose if_broot area is changing
  2214. * ext_diff -- the change in the number of records, positive or negative,
  2215. * requested for the if_broot array.
  2216. */
  2217. void
  2218. xfs_iroot_realloc(
  2219. xfs_inode_t *ip,
  2220. int rec_diff,
  2221. int whichfork)
  2222. {
  2223. int cur_max;
  2224. xfs_ifork_t *ifp;
  2225. xfs_bmbt_block_t *new_broot;
  2226. int new_max;
  2227. size_t new_size;
  2228. char *np;
  2229. char *op;
  2230. /*
  2231. * Handle the degenerate case quietly.
  2232. */
  2233. if (rec_diff == 0) {
  2234. return;
  2235. }
  2236. ifp = XFS_IFORK_PTR(ip, whichfork);
  2237. if (rec_diff > 0) {
  2238. /*
  2239. * If there wasn't any memory allocated before, just
  2240. * allocate it now and get out.
  2241. */
  2242. if (ifp->if_broot_bytes == 0) {
  2243. new_size = (size_t)XFS_BMAP_BROOT_SPACE_CALC(rec_diff);
  2244. ifp->if_broot = (xfs_bmbt_block_t*)kmem_alloc(new_size,
  2245. KM_SLEEP);
  2246. ifp->if_broot_bytes = (int)new_size;
  2247. return;
  2248. }
  2249. /*
  2250. * If there is already an existing if_broot, then we need
  2251. * to realloc() it and shift the pointers to their new
  2252. * location. The records don't change location because
  2253. * they are kept butted up against the btree block header.
  2254. */
  2255. cur_max = XFS_BMAP_BROOT_MAXRECS(ifp->if_broot_bytes);
  2256. new_max = cur_max + rec_diff;
  2257. new_size = (size_t)XFS_BMAP_BROOT_SPACE_CALC(new_max);
  2258. ifp->if_broot = (xfs_bmbt_block_t *)
  2259. kmem_realloc(ifp->if_broot,
  2260. new_size,
  2261. (size_t)XFS_BMAP_BROOT_SPACE_CALC(cur_max), /* old size */
  2262. KM_SLEEP);
  2263. op = (char *)XFS_BMAP_BROOT_PTR_ADDR(ifp->if_broot, 1,
  2264. ifp->if_broot_bytes);
  2265. np = (char *)XFS_BMAP_BROOT_PTR_ADDR(ifp->if_broot, 1,
  2266. (int)new_size);
  2267. ifp->if_broot_bytes = (int)new_size;
  2268. ASSERT(ifp->if_broot_bytes <=
  2269. XFS_IFORK_SIZE(ip, whichfork) + XFS_BROOT_SIZE_ADJ);
  2270. memmove(np, op, cur_max * (uint)sizeof(xfs_dfsbno_t));
  2271. return;
  2272. }
  2273. /*
  2274. * rec_diff is less than 0. In this case, we are shrinking the
  2275. * if_broot buffer. It must already exist. If we go to zero
  2276. * records, just get rid of the root and clear the status bit.
  2277. */
  2278. ASSERT((ifp->if_broot != NULL) && (ifp->if_broot_bytes > 0));
  2279. cur_max = XFS_BMAP_BROOT_MAXRECS(ifp->if_broot_bytes);
  2280. new_max = cur_max + rec_diff;
  2281. ASSERT(new_max >= 0);
  2282. if (new_max > 0)
  2283. new_size = (size_t)XFS_BMAP_BROOT_SPACE_CALC(new_max);
  2284. else
  2285. new_size = 0;
  2286. if (new_size > 0) {
  2287. new_broot = (xfs_bmbt_block_t *)kmem_alloc(new_size, KM_SLEEP);
  2288. /*
  2289. * First copy over the btree block header.
  2290. */
  2291. memcpy(new_broot, ifp->if_broot, sizeof(xfs_bmbt_block_t));
  2292. } else {
  2293. new_broot = NULL;
  2294. ifp->if_flags &= ~XFS_IFBROOT;
  2295. }
  2296. /*
  2297. * Only copy the records and pointers if there are any.
  2298. */
  2299. if (new_max > 0) {
  2300. /*
  2301. * First copy the records.
  2302. */
  2303. op = (char *)XFS_BMAP_BROOT_REC_ADDR(ifp->if_broot, 1,
  2304. ifp->if_broot_bytes);
  2305. np = (char *)XFS_BMAP_BROOT_REC_ADDR(new_broot, 1,
  2306. (int)new_size);
  2307. memcpy(np, op, new_max * (uint)sizeof(xfs_bmbt_rec_t));
  2308. /*
  2309. * Then copy the pointers.
  2310. */
  2311. op = (char *)XFS_BMAP_BROOT_PTR_ADDR(ifp->if_broot, 1,
  2312. ifp->if_broot_bytes);
  2313. np = (char *)XFS_BMAP_BROOT_PTR_ADDR(new_broot, 1,
  2314. (int)new_size);
  2315. memcpy(np, op, new_max * (uint)sizeof(xfs_dfsbno_t));
  2316. }
  2317. kmem_free(ifp->if_broot, ifp->if_broot_bytes);
  2318. ifp->if_broot = new_broot;
  2319. ifp->if_broot_bytes = (int)new_size;
  2320. ASSERT(ifp->if_broot_bytes <=
  2321. XFS_IFORK_SIZE(ip, whichfork) + XFS_BROOT_SIZE_ADJ);
  2322. return;
  2323. }
  2324. /*
  2325. * This is called when the amount of space needed for if_data
  2326. * is increased or decreased. The change in size is indicated by
  2327. * the number of bytes that need to be added or deleted in the
  2328. * byte_diff parameter.
  2329. *
  2330. * If the amount of space needed has decreased below the size of the
  2331. * inline buffer, then switch to using the inline buffer. Otherwise,
  2332. * use kmem_realloc() or kmem_alloc() to adjust the size of the buffer
  2333. * to what is needed.
  2334. *
  2335. * ip -- the inode whose if_data area is changing
  2336. * byte_diff -- the change in the number of bytes, positive or negative,
  2337. * requested for the if_data array.
  2338. */
  2339. void
  2340. xfs_idata_realloc(
  2341. xfs_inode_t *ip,
  2342. int byte_diff,
  2343. int whichfork)
  2344. {
  2345. xfs_ifork_t *ifp;
  2346. int new_size;
  2347. int real_size;
  2348. if (byte_diff == 0) {
  2349. return;
  2350. }
  2351. ifp = XFS_IFORK_PTR(ip, whichfork);
  2352. new_size = (int)ifp->if_bytes + byte_diff;
  2353. ASSERT(new_size >= 0);
  2354. if (new_size == 0) {
  2355. if (ifp->if_u1.if_data != ifp->if_u2.if_inline_data) {
  2356. kmem_free(ifp->if_u1.if_data, ifp->if_real_bytes);
  2357. }
  2358. ifp->if_u1.if_data = NULL;
  2359. real_size = 0;
  2360. } else if (new_size <= sizeof(ifp->if_u2.if_inline_data)) {
  2361. /*
  2362. * If the valid extents/data can fit in if_inline_ext/data,
  2363. * copy them from the malloc'd vector and free it.
  2364. */
  2365. if (ifp->if_u1.if_data == NULL) {
  2366. ifp->if_u1.if_data = ifp->if_u2.if_inline_data;
  2367. } else if (ifp->if_u1.if_data != ifp->if_u2.if_inline_data) {
  2368. ASSERT(ifp->if_real_bytes != 0);
  2369. memcpy(ifp->if_u2.if_inline_data, ifp->if_u1.if_data,
  2370. new_size);
  2371. kmem_free(ifp->if_u1.if_data, ifp->if_real_bytes);
  2372. ifp->if_u1.if_data = ifp->if_u2.if_inline_data;
  2373. }
  2374. real_size = 0;
  2375. } else {
  2376. /*
  2377. * Stuck with malloc/realloc.
  2378. * For inline data, the underlying buffer must be
  2379. * a multiple of 4 bytes in size so that it can be
  2380. * logged and stay on word boundaries. We enforce
  2381. * that here.
  2382. */
  2383. real_size = roundup(new_size, 4);
  2384. if (ifp->if_u1.if_data == NULL) {
  2385. ASSERT(ifp->if_real_bytes == 0);
  2386. ifp->if_u1.if_data = kmem_alloc(real_size, KM_SLEEP);
  2387. } else if (ifp->if_u1.if_data != ifp->if_u2.if_inline_data) {
  2388. /*
  2389. * Only do the realloc if the underlying size
  2390. * is really changing.
  2391. */
  2392. if (ifp->if_real_bytes != real_size) {
  2393. ifp->if_u1.if_data =
  2394. kmem_realloc(ifp->if_u1.if_data,
  2395. real_size,
  2396. ifp->if_real_bytes,
  2397. KM_SLEEP);
  2398. }
  2399. } else {
  2400. ASSERT(ifp->if_real_bytes == 0);
  2401. ifp->if_u1.if_data = kmem_alloc(real_size, KM_SLEEP);
  2402. memcpy(ifp->if_u1.if_data, ifp->if_u2.if_inline_data,
  2403. ifp->if_bytes);
  2404. }
  2405. }
  2406. ifp->if_real_bytes = real_size;
  2407. ifp->if_bytes = new_size;
  2408. ASSERT(ifp->if_bytes <= XFS_IFORK_SIZE(ip, whichfork));
  2409. }
  2410. /*
  2411. * Map inode to disk block and offset.
  2412. *
  2413. * mp -- the mount point structure for the current file system
  2414. * tp -- the current transaction
  2415. * ino -- the inode number of the inode to be located
  2416. * imap -- this structure is filled in with the information necessary
  2417. * to retrieve the given inode from disk
  2418. * flags -- flags to pass to xfs_dilocate indicating whether or not
  2419. * lookups in the inode btree were OK or not
  2420. */
  2421. int
  2422. xfs_imap(
  2423. xfs_mount_t *mp,
  2424. xfs_trans_t *tp,
  2425. xfs_ino_t ino,
  2426. xfs_imap_t *imap,
  2427. uint flags)
  2428. {
  2429. xfs_fsblock_t fsbno;
  2430. int len;
  2431. int off;
  2432. int error;
  2433. fsbno = imap->im_blkno ?
  2434. XFS_DADDR_TO_FSB(mp, imap->im_blkno) : NULLFSBLOCK;
  2435. error = xfs_dilocate(mp, tp, ino, &fsbno, &len, &off, flags);
  2436. if (error != 0) {
  2437. return error;
  2438. }
  2439. imap->im_blkno = XFS_FSB_TO_DADDR(mp, fsbno);
  2440. imap->im_len = XFS_FSB_TO_BB(mp, len);
  2441. imap->im_agblkno = XFS_FSB_TO_AGBNO(mp, fsbno);
  2442. imap->im_ioffset = (ushort)off;
  2443. imap->im_boffset = (ushort)(off << mp->m_sb.sb_inodelog);
  2444. return 0;
  2445. }
  2446. void
  2447. xfs_idestroy_fork(
  2448. xfs_inode_t *ip,
  2449. int whichfork)
  2450. {
  2451. xfs_ifork_t *ifp;
  2452. ifp = XFS_IFORK_PTR(ip, whichfork);
  2453. if (ifp->if_broot != NULL) {
  2454. kmem_free(ifp->if_broot, ifp->if_broot_bytes);
  2455. ifp->if_broot = NULL;
  2456. }
  2457. /*
  2458. * If the format is local, then we can't have an extents
  2459. * array so just look for an inline data array. If we're
  2460. * not local then we may or may not have an extents list,
  2461. * so check and free it up if we do.
  2462. */
  2463. if (XFS_IFORK_FORMAT(ip, whichfork) == XFS_DINODE_FMT_LOCAL) {
  2464. if ((ifp->if_u1.if_data != ifp->if_u2.if_inline_data) &&
  2465. (ifp->if_u1.if_data != NULL)) {
  2466. ASSERT(ifp->if_real_bytes != 0);
  2467. kmem_free(ifp->if_u1.if_data, ifp->if_real_bytes);
  2468. ifp->if_u1.if_data = NULL;
  2469. ifp->if_real_bytes = 0;
  2470. }
  2471. } else if ((ifp->if_flags & XFS_IFEXTENTS) &&
  2472. ((ifp->if_flags & XFS_IFEXTIREC) ||
  2473. ((ifp->if_u1.if_extents != NULL) &&
  2474. (ifp->if_u1.if_extents != ifp->if_u2.if_inline_ext)))) {
  2475. ASSERT(ifp->if_real_bytes != 0);
  2476. xfs_iext_destroy(ifp);
  2477. }
  2478. ASSERT(ifp->if_u1.if_extents == NULL ||
  2479. ifp->if_u1.if_extents == ifp->if_u2.if_inline_ext);
  2480. ASSERT(ifp->if_real_bytes == 0);
  2481. if (whichfork == XFS_ATTR_FORK) {
  2482. kmem_zone_free(xfs_ifork_zone, ip->i_afp);
  2483. ip->i_afp = NULL;
  2484. }
  2485. }
  2486. /*
  2487. * This is called free all the memory associated with an inode.
  2488. * It must free the inode itself and any buffers allocated for
  2489. * if_extents/if_data and if_broot. It must also free the lock
  2490. * associated with the inode.
  2491. */
  2492. void
  2493. xfs_idestroy(
  2494. xfs_inode_t *ip)
  2495. {
  2496. switch (ip->i_d.di_mode & S_IFMT) {
  2497. case S_IFREG:
  2498. case S_IFDIR:
  2499. case S_IFLNK:
  2500. xfs_idestroy_fork(ip, XFS_DATA_FORK);
  2501. break;
  2502. }
  2503. if (ip->i_afp)
  2504. xfs_idestroy_fork(ip, XFS_ATTR_FORK);
  2505. mrfree(&ip->i_lock);
  2506. mrfree(&ip->i_iolock);
  2507. freesema(&ip->i_flock);
  2508. #ifdef XFS_BMAP_TRACE
  2509. ktrace_free(ip->i_xtrace);
  2510. #endif
  2511. #ifdef XFS_BMBT_TRACE
  2512. ktrace_free(ip->i_btrace);
  2513. #endif
  2514. #ifdef XFS_RW_TRACE
  2515. ktrace_free(ip->i_rwtrace);
  2516. #endif
  2517. #ifdef XFS_ILOCK_TRACE
  2518. ktrace_free(ip->i_lock_trace);
  2519. #endif
  2520. #ifdef XFS_DIR2_TRACE
  2521. ktrace_free(ip->i_dir_trace);
  2522. #endif
  2523. if (ip->i_itemp) {
  2524. /*
  2525. * Only if we are shutting down the fs will we see an
  2526. * inode still in the AIL. If it is there, we should remove
  2527. * it to prevent a use-after-free from occurring.
  2528. */
  2529. xfs_mount_t *mp = ip->i_mount;
  2530. xfs_log_item_t *lip = &ip->i_itemp->ili_item;
  2531. int s;
  2532. ASSERT(((lip->li_flags & XFS_LI_IN_AIL) == 0) ||
  2533. XFS_FORCED_SHUTDOWN(ip->i_mount));
  2534. if (lip->li_flags & XFS_LI_IN_AIL) {
  2535. AIL_LOCK(mp, s);
  2536. if (lip->li_flags & XFS_LI_IN_AIL)
  2537. xfs_trans_delete_ail(mp, lip, s);
  2538. else
  2539. AIL_UNLOCK(mp, s);
  2540. }
  2541. xfs_inode_item_destroy(ip);
  2542. }
  2543. kmem_zone_free(xfs_inode_zone, ip);
  2544. }
  2545. /*
  2546. * Increment the pin count of the given buffer.
  2547. * This value is protected by ipinlock spinlock in the mount structure.
  2548. */
  2549. void
  2550. xfs_ipin(
  2551. xfs_inode_t *ip)
  2552. {
  2553. ASSERT(ismrlocked(&ip->i_lock, MR_UPDATE));
  2554. atomic_inc(&ip->i_pincount);
  2555. }
  2556. /*
  2557. * Decrement the pin count of the given inode, and wake up
  2558. * anyone in xfs_iwait_unpin() if the count goes to 0. The
  2559. * inode must have been previously pinned with a call to xfs_ipin().
  2560. */
  2561. void
  2562. xfs_iunpin(
  2563. xfs_inode_t *ip)
  2564. {
  2565. ASSERT(atomic_read(&ip->i_pincount) > 0);
  2566. if (atomic_dec_and_lock(&ip->i_pincount, &ip->i_flags_lock)) {
  2567. /*
  2568. * If the inode is currently being reclaimed, the link between
  2569. * the bhv_vnode and the xfs_inode will be broken after the
  2570. * XFS_IRECLAIM* flag is set. Hence, if these flags are not
  2571. * set, then we can move forward and mark the linux inode dirty
  2572. * knowing that it is still valid as it won't freed until after
  2573. * the bhv_vnode<->xfs_inode link is broken in xfs_reclaim. The
  2574. * i_flags_lock is used to synchronise the setting of the
  2575. * XFS_IRECLAIM* flags and the breaking of the link, and so we
  2576. * can execute atomically w.r.t to reclaim by holding this lock
  2577. * here.
  2578. *
  2579. * However, we still need to issue the unpin wakeup call as the
  2580. * inode reclaim may be blocked waiting for the inode to become
  2581. * unpinned.
  2582. */
  2583. if (!__xfs_iflags_test(ip, XFS_IRECLAIM|XFS_IRECLAIMABLE)) {
  2584. bhv_vnode_t *vp = XFS_ITOV_NULL(ip);
  2585. struct inode *inode = NULL;
  2586. BUG_ON(vp == NULL);
  2587. inode = vn_to_inode(vp);
  2588. BUG_ON(inode->i_state & I_CLEAR);
  2589. /* make sync come back and flush this inode */
  2590. if (!(inode->i_state & (I_NEW|I_FREEING)))
  2591. mark_inode_dirty_sync(inode);
  2592. }
  2593. spin_unlock(&ip->i_flags_lock);
  2594. wake_up(&ip->i_ipin_wait);
  2595. }
  2596. }
  2597. /*
  2598. * This is called to wait for the given inode to be unpinned.
  2599. * It will sleep until this happens. The caller must have the
  2600. * inode locked in at least shared mode so that the buffer cannot
  2601. * be subsequently pinned once someone is waiting for it to be
  2602. * unpinned.
  2603. */
  2604. STATIC void
  2605. xfs_iunpin_wait(
  2606. xfs_inode_t *ip)
  2607. {
  2608. xfs_inode_log_item_t *iip;
  2609. xfs_lsn_t lsn;
  2610. ASSERT(ismrlocked(&ip->i_lock, MR_UPDATE | MR_ACCESS));
  2611. if (atomic_read(&ip->i_pincount) == 0) {
  2612. return;
  2613. }
  2614. iip = ip->i_itemp;
  2615. if (iip && iip->ili_last_lsn) {
  2616. lsn = iip->ili_last_lsn;
  2617. } else {
  2618. lsn = (xfs_lsn_t)0;
  2619. }
  2620. /*
  2621. * Give the log a push so we don't wait here too long.
  2622. */
  2623. xfs_log_force(ip->i_mount, lsn, XFS_LOG_FORCE);
  2624. wait_event(ip->i_ipin_wait, (atomic_read(&ip->i_pincount) == 0));
  2625. }
  2626. /*
  2627. * xfs_iextents_copy()
  2628. *
  2629. * This is called to copy the REAL extents (as opposed to the delayed
  2630. * allocation extents) from the inode into the given buffer. It
  2631. * returns the number of bytes copied into the buffer.
  2632. *
  2633. * If there are no delayed allocation extents, then we can just
  2634. * memcpy() the extents into the buffer. Otherwise, we need to
  2635. * examine each extent in turn and skip those which are delayed.
  2636. */
  2637. int
  2638. xfs_iextents_copy(
  2639. xfs_inode_t *ip,
  2640. xfs_bmbt_rec_t *buffer,
  2641. int whichfork)
  2642. {
  2643. int copied;
  2644. xfs_bmbt_rec_t *dest_ep;
  2645. xfs_bmbt_rec_t *ep;
  2646. #ifdef XFS_BMAP_TRACE
  2647. static char fname[] = "xfs_iextents_copy";
  2648. #endif
  2649. int i;
  2650. xfs_ifork_t *ifp;
  2651. int nrecs;
  2652. xfs_fsblock_t start_block;
  2653. ifp = XFS_IFORK_PTR(ip, whichfork);
  2654. ASSERT(ismrlocked(&ip->i_lock, MR_UPDATE|MR_ACCESS));
  2655. ASSERT(ifp->if_bytes > 0);
  2656. nrecs = ifp->if_bytes / (uint)sizeof(xfs_bmbt_rec_t);
  2657. xfs_bmap_trace_exlist(fname, ip, nrecs, whichfork);
  2658. ASSERT(nrecs > 0);
  2659. /*
  2660. * There are some delayed allocation extents in the
  2661. * inode, so copy the extents one at a time and skip
  2662. * the delayed ones. There must be at least one
  2663. * non-delayed extent.
  2664. */
  2665. dest_ep = buffer;
  2666. copied = 0;
  2667. for (i = 0; i < nrecs; i++) {
  2668. ep = xfs_iext_get_ext(ifp, i);
  2669. start_block = xfs_bmbt_get_startblock(ep);
  2670. if (ISNULLSTARTBLOCK(start_block)) {
  2671. /*
  2672. * It's a delayed allocation extent, so skip it.
  2673. */
  2674. continue;
  2675. }
  2676. /* Translate to on disk format */
  2677. put_unaligned(INT_GET(ep->l0, ARCH_CONVERT),
  2678. (__uint64_t*)&dest_ep->l0);
  2679. put_unaligned(INT_GET(ep->l1, ARCH_CONVERT),
  2680. (__uint64_t*)&dest_ep->l1);
  2681. dest_ep++;
  2682. copied++;
  2683. }
  2684. ASSERT(copied != 0);
  2685. xfs_validate_extents(ifp, copied, 1, XFS_EXTFMT_INODE(ip));
  2686. return (copied * (uint)sizeof(xfs_bmbt_rec_t));
  2687. }
  2688. /*
  2689. * Each of the following cases stores data into the same region
  2690. * of the on-disk inode, so only one of them can be valid at
  2691. * any given time. While it is possible to have conflicting formats
  2692. * and log flags, e.g. having XFS_ILOG_?DATA set when the fork is
  2693. * in EXTENTS format, this can only happen when the fork has
  2694. * changed formats after being modified but before being flushed.
  2695. * In these cases, the format always takes precedence, because the
  2696. * format indicates the current state of the fork.
  2697. */
  2698. /*ARGSUSED*/
  2699. STATIC int
  2700. xfs_iflush_fork(
  2701. xfs_inode_t *ip,
  2702. xfs_dinode_t *dip,
  2703. xfs_inode_log_item_t *iip,
  2704. int whichfork,
  2705. xfs_buf_t *bp)
  2706. {
  2707. char *cp;
  2708. xfs_ifork_t *ifp;
  2709. xfs_mount_t *mp;
  2710. #ifdef XFS_TRANS_DEBUG
  2711. int first;
  2712. #endif
  2713. static const short brootflag[2] =
  2714. { XFS_ILOG_DBROOT, XFS_ILOG_ABROOT };
  2715. static const short dataflag[2] =
  2716. { XFS_ILOG_DDATA, XFS_ILOG_ADATA };
  2717. static const short extflag[2] =
  2718. { XFS_ILOG_DEXT, XFS_ILOG_AEXT };
  2719. if (iip == NULL)
  2720. return 0;
  2721. ifp = XFS_IFORK_PTR(ip, whichfork);
  2722. /*
  2723. * This can happen if we gave up in iformat in an error path,
  2724. * for the attribute fork.
  2725. */
  2726. if (ifp == NULL) {
  2727. ASSERT(whichfork == XFS_ATTR_FORK);
  2728. return 0;
  2729. }
  2730. cp = XFS_DFORK_PTR(dip, whichfork);
  2731. mp = ip->i_mount;
  2732. switch (XFS_IFORK_FORMAT(ip, whichfork)) {
  2733. case XFS_DINODE_FMT_LOCAL:
  2734. if ((iip->ili_format.ilf_fields & dataflag[whichfork]) &&
  2735. (ifp->if_bytes > 0)) {
  2736. ASSERT(ifp->if_u1.if_data != NULL);
  2737. ASSERT(ifp->if_bytes <= XFS_IFORK_SIZE(ip, whichfork));
  2738. memcpy(cp, ifp->if_u1.if_data, ifp->if_bytes);
  2739. }
  2740. break;
  2741. case XFS_DINODE_FMT_EXTENTS:
  2742. ASSERT((ifp->if_flags & XFS_IFEXTENTS) ||
  2743. !(iip->ili_format.ilf_fields & extflag[whichfork]));
  2744. ASSERT((xfs_iext_get_ext(ifp, 0) != NULL) ||
  2745. (ifp->if_bytes == 0));
  2746. ASSERT((xfs_iext_get_ext(ifp, 0) == NULL) ||
  2747. (ifp->if_bytes > 0));
  2748. if ((iip->ili_format.ilf_fields & extflag[whichfork]) &&
  2749. (ifp->if_bytes > 0)) {
  2750. ASSERT(XFS_IFORK_NEXTENTS(ip, whichfork) > 0);
  2751. (void)xfs_iextents_copy(ip, (xfs_bmbt_rec_t *)cp,
  2752. whichfork);
  2753. }
  2754. break;
  2755. case XFS_DINODE_FMT_BTREE:
  2756. if ((iip->ili_format.ilf_fields & brootflag[whichfork]) &&
  2757. (ifp->if_broot_bytes > 0)) {
  2758. ASSERT(ifp->if_broot != NULL);
  2759. ASSERT(ifp->if_broot_bytes <=
  2760. (XFS_IFORK_SIZE(ip, whichfork) +
  2761. XFS_BROOT_SIZE_ADJ));
  2762. xfs_bmbt_to_bmdr(ifp->if_broot, ifp->if_broot_bytes,
  2763. (xfs_bmdr_block_t *)cp,
  2764. XFS_DFORK_SIZE(dip, mp, whichfork));
  2765. }
  2766. break;
  2767. case XFS_DINODE_FMT_DEV:
  2768. if (iip->ili_format.ilf_fields & XFS_ILOG_DEV) {
  2769. ASSERT(whichfork == XFS_DATA_FORK);
  2770. INT_SET(dip->di_u.di_dev, ARCH_CONVERT, ip->i_df.if_u2.if_rdev);
  2771. }
  2772. break;
  2773. case XFS_DINODE_FMT_UUID:
  2774. if (iip->ili_format.ilf_fields & XFS_ILOG_UUID) {
  2775. ASSERT(whichfork == XFS_DATA_FORK);
  2776. memcpy(&dip->di_u.di_muuid, &ip->i_df.if_u2.if_uuid,
  2777. sizeof(uuid_t));
  2778. }
  2779. break;
  2780. default:
  2781. ASSERT(0);
  2782. break;
  2783. }
  2784. return 0;
  2785. }
  2786. /*
  2787. * xfs_iflush() will write a modified inode's changes out to the
  2788. * inode's on disk home. The caller must have the inode lock held
  2789. * in at least shared mode and the inode flush semaphore must be
  2790. * held as well. The inode lock will still be held upon return from
  2791. * the call and the caller is free to unlock it.
  2792. * The inode flush lock will be unlocked when the inode reaches the disk.
  2793. * The flags indicate how the inode's buffer should be written out.
  2794. */
  2795. int
  2796. xfs_iflush(
  2797. xfs_inode_t *ip,
  2798. uint flags)
  2799. {
  2800. xfs_inode_log_item_t *iip;
  2801. xfs_buf_t *bp;
  2802. xfs_dinode_t *dip;
  2803. xfs_mount_t *mp;
  2804. int error;
  2805. /* REFERENCED */
  2806. xfs_chash_t *ch;
  2807. xfs_inode_t *iq;
  2808. int clcount; /* count of inodes clustered */
  2809. int bufwasdelwri;
  2810. enum { INT_DELWRI = (1 << 0), INT_ASYNC = (1 << 1) };
  2811. SPLDECL(s);
  2812. XFS_STATS_INC(xs_iflush_count);
  2813. ASSERT(ismrlocked(&ip->i_lock, MR_UPDATE|MR_ACCESS));
  2814. ASSERT(issemalocked(&(ip->i_flock)));
  2815. ASSERT(ip->i_d.di_format != XFS_DINODE_FMT_BTREE ||
  2816. ip->i_d.di_nextents > ip->i_df.if_ext_max);
  2817. iip = ip->i_itemp;
  2818. mp = ip->i_mount;
  2819. /*
  2820. * If the inode isn't dirty, then just release the inode
  2821. * flush lock and do nothing.
  2822. */
  2823. if ((ip->i_update_core == 0) &&
  2824. ((iip == NULL) || !(iip->ili_format.ilf_fields & XFS_ILOG_ALL))) {
  2825. ASSERT((iip != NULL) ?
  2826. !(iip->ili_item.li_flags & XFS_LI_IN_AIL) : 1);
  2827. xfs_ifunlock(ip);
  2828. return 0;
  2829. }
  2830. /*
  2831. * We can't flush the inode until it is unpinned, so
  2832. * wait for it. We know noone new can pin it, because
  2833. * we are holding the inode lock shared and you need
  2834. * to hold it exclusively to pin the inode.
  2835. */
  2836. xfs_iunpin_wait(ip);
  2837. /*
  2838. * This may have been unpinned because the filesystem is shutting
  2839. * down forcibly. If that's the case we must not write this inode
  2840. * to disk, because the log record didn't make it to disk!
  2841. */
  2842. if (XFS_FORCED_SHUTDOWN(mp)) {
  2843. ip->i_update_core = 0;
  2844. if (iip)
  2845. iip->ili_format.ilf_fields = 0;
  2846. xfs_ifunlock(ip);
  2847. return XFS_ERROR(EIO);
  2848. }
  2849. /*
  2850. * Get the buffer containing the on-disk inode.
  2851. */
  2852. error = xfs_itobp(mp, NULL, ip, &dip, &bp, 0, 0);
  2853. if (error) {
  2854. xfs_ifunlock(ip);
  2855. return error;
  2856. }
  2857. /*
  2858. * Decide how buffer will be flushed out. This is done before
  2859. * the call to xfs_iflush_int because this field is zeroed by it.
  2860. */
  2861. if (iip != NULL && iip->ili_format.ilf_fields != 0) {
  2862. /*
  2863. * Flush out the inode buffer according to the directions
  2864. * of the caller. In the cases where the caller has given
  2865. * us a choice choose the non-delwri case. This is because
  2866. * the inode is in the AIL and we need to get it out soon.
  2867. */
  2868. switch (flags) {
  2869. case XFS_IFLUSH_SYNC:
  2870. case XFS_IFLUSH_DELWRI_ELSE_SYNC:
  2871. flags = 0;
  2872. break;
  2873. case XFS_IFLUSH_ASYNC:
  2874. case XFS_IFLUSH_DELWRI_ELSE_ASYNC:
  2875. flags = INT_ASYNC;
  2876. break;
  2877. case XFS_IFLUSH_DELWRI:
  2878. flags = INT_DELWRI;
  2879. break;
  2880. default:
  2881. ASSERT(0);
  2882. flags = 0;
  2883. break;
  2884. }
  2885. } else {
  2886. switch (flags) {
  2887. case XFS_IFLUSH_DELWRI_ELSE_SYNC:
  2888. case XFS_IFLUSH_DELWRI_ELSE_ASYNC:
  2889. case XFS_IFLUSH_DELWRI:
  2890. flags = INT_DELWRI;
  2891. break;
  2892. case XFS_IFLUSH_ASYNC:
  2893. flags = INT_ASYNC;
  2894. break;
  2895. case XFS_IFLUSH_SYNC:
  2896. flags = 0;
  2897. break;
  2898. default:
  2899. ASSERT(0);
  2900. flags = 0;
  2901. break;
  2902. }
  2903. }
  2904. /*
  2905. * First flush out the inode that xfs_iflush was called with.
  2906. */
  2907. error = xfs_iflush_int(ip, bp);
  2908. if (error) {
  2909. goto corrupt_out;
  2910. }
  2911. /*
  2912. * inode clustering:
  2913. * see if other inodes can be gathered into this write
  2914. */
  2915. ip->i_chash->chl_buf = bp;
  2916. ch = XFS_CHASH(mp, ip->i_blkno);
  2917. s = mutex_spinlock(&ch->ch_lock);
  2918. clcount = 0;
  2919. for (iq = ip->i_cnext; iq != ip; iq = iq->i_cnext) {
  2920. /*
  2921. * Do an un-protected check to see if the inode is dirty and
  2922. * is a candidate for flushing. These checks will be repeated
  2923. * later after the appropriate locks are acquired.
  2924. */
  2925. iip = iq->i_itemp;
  2926. if ((iq->i_update_core == 0) &&
  2927. ((iip == NULL) ||
  2928. !(iip->ili_format.ilf_fields & XFS_ILOG_ALL)) &&
  2929. xfs_ipincount(iq) == 0) {
  2930. continue;
  2931. }
  2932. /*
  2933. * Try to get locks. If any are unavailable,
  2934. * then this inode cannot be flushed and is skipped.
  2935. */
  2936. /* get inode locks (just i_lock) */
  2937. if (xfs_ilock_nowait(iq, XFS_ILOCK_SHARED)) {
  2938. /* get inode flush lock */
  2939. if (xfs_iflock_nowait(iq)) {
  2940. /* check if pinned */
  2941. if (xfs_ipincount(iq) == 0) {
  2942. /* arriving here means that
  2943. * this inode can be flushed.
  2944. * first re-check that it's
  2945. * dirty
  2946. */
  2947. iip = iq->i_itemp;
  2948. if ((iq->i_update_core != 0)||
  2949. ((iip != NULL) &&
  2950. (iip->ili_format.ilf_fields & XFS_ILOG_ALL))) {
  2951. clcount++;
  2952. error = xfs_iflush_int(iq, bp);
  2953. if (error) {
  2954. xfs_iunlock(iq,
  2955. XFS_ILOCK_SHARED);
  2956. goto cluster_corrupt_out;
  2957. }
  2958. } else {
  2959. xfs_ifunlock(iq);
  2960. }
  2961. } else {
  2962. xfs_ifunlock(iq);
  2963. }
  2964. }
  2965. xfs_iunlock(iq, XFS_ILOCK_SHARED);
  2966. }
  2967. }
  2968. mutex_spinunlock(&ch->ch_lock, s);
  2969. if (clcount) {
  2970. XFS_STATS_INC(xs_icluster_flushcnt);
  2971. XFS_STATS_ADD(xs_icluster_flushinode, clcount);
  2972. }
  2973. /*
  2974. * If the buffer is pinned then push on the log so we won't
  2975. * get stuck waiting in the write for too long.
  2976. */
  2977. if (XFS_BUF_ISPINNED(bp)){
  2978. xfs_log_force(mp, (xfs_lsn_t)0, XFS_LOG_FORCE);
  2979. }
  2980. if (flags & INT_DELWRI) {
  2981. xfs_bdwrite(mp, bp);
  2982. } else if (flags & INT_ASYNC) {
  2983. xfs_bawrite(mp, bp);
  2984. } else {
  2985. error = xfs_bwrite(mp, bp);
  2986. }
  2987. return error;
  2988. corrupt_out:
  2989. xfs_buf_relse(bp);
  2990. xfs_force_shutdown(mp, SHUTDOWN_CORRUPT_INCORE);
  2991. xfs_iflush_abort(ip);
  2992. /*
  2993. * Unlocks the flush lock
  2994. */
  2995. return XFS_ERROR(EFSCORRUPTED);
  2996. cluster_corrupt_out:
  2997. /* Corruption detected in the clustering loop. Invalidate the
  2998. * inode buffer and shut down the filesystem.
  2999. */
  3000. mutex_spinunlock(&ch->ch_lock, s);
  3001. /*
  3002. * Clean up the buffer. If it was B_DELWRI, just release it --
  3003. * brelse can handle it with no problems. If not, shut down the
  3004. * filesystem before releasing the buffer.
  3005. */
  3006. if ((bufwasdelwri= XFS_BUF_ISDELAYWRITE(bp))) {
  3007. xfs_buf_relse(bp);
  3008. }
  3009. xfs_force_shutdown(mp, SHUTDOWN_CORRUPT_INCORE);
  3010. if(!bufwasdelwri) {
  3011. /*
  3012. * Just like incore_relse: if we have b_iodone functions,
  3013. * mark the buffer as an error and call them. Otherwise
  3014. * mark it as stale and brelse.
  3015. */
  3016. if (XFS_BUF_IODONE_FUNC(bp)) {
  3017. XFS_BUF_CLR_BDSTRAT_FUNC(bp);
  3018. XFS_BUF_UNDONE(bp);
  3019. XFS_BUF_STALE(bp);
  3020. XFS_BUF_SHUT(bp);
  3021. XFS_BUF_ERROR(bp,EIO);
  3022. xfs_biodone(bp);
  3023. } else {
  3024. XFS_BUF_STALE(bp);
  3025. xfs_buf_relse(bp);
  3026. }
  3027. }
  3028. xfs_iflush_abort(iq);
  3029. /*
  3030. * Unlocks the flush lock
  3031. */
  3032. return XFS_ERROR(EFSCORRUPTED);
  3033. }
  3034. STATIC int
  3035. xfs_iflush_int(
  3036. xfs_inode_t *ip,
  3037. xfs_buf_t *bp)
  3038. {
  3039. xfs_inode_log_item_t *iip;
  3040. xfs_dinode_t *dip;
  3041. xfs_mount_t *mp;
  3042. #ifdef XFS_TRANS_DEBUG
  3043. int first;
  3044. #endif
  3045. SPLDECL(s);
  3046. ASSERT(ismrlocked(&ip->i_lock, MR_UPDATE|MR_ACCESS));
  3047. ASSERT(issemalocked(&(ip->i_flock)));
  3048. ASSERT(ip->i_d.di_format != XFS_DINODE_FMT_BTREE ||
  3049. ip->i_d.di_nextents > ip->i_df.if_ext_max);
  3050. iip = ip->i_itemp;
  3051. mp = ip->i_mount;
  3052. /*
  3053. * If the inode isn't dirty, then just release the inode
  3054. * flush lock and do nothing.
  3055. */
  3056. if ((ip->i_update_core == 0) &&
  3057. ((iip == NULL) || !(iip->ili_format.ilf_fields & XFS_ILOG_ALL))) {
  3058. xfs_ifunlock(ip);
  3059. return 0;
  3060. }
  3061. /* set *dip = inode's place in the buffer */
  3062. dip = (xfs_dinode_t *)xfs_buf_offset(bp, ip->i_boffset);
  3063. /*
  3064. * Clear i_update_core before copying out the data.
  3065. * This is for coordination with our timestamp updates
  3066. * that don't hold the inode lock. They will always
  3067. * update the timestamps BEFORE setting i_update_core,
  3068. * so if we clear i_update_core after they set it we
  3069. * are guaranteed to see their updates to the timestamps.
  3070. * I believe that this depends on strongly ordered memory
  3071. * semantics, but we have that. We use the SYNCHRONIZE
  3072. * macro to make sure that the compiler does not reorder
  3073. * the i_update_core access below the data copy below.
  3074. */
  3075. ip->i_update_core = 0;
  3076. SYNCHRONIZE();
  3077. /*
  3078. * Make sure to get the latest atime from the Linux inode.
  3079. */
  3080. xfs_synchronize_atime(ip);
  3081. if (XFS_TEST_ERROR(INT_GET(dip->di_core.di_magic,ARCH_CONVERT) != XFS_DINODE_MAGIC,
  3082. mp, XFS_ERRTAG_IFLUSH_1, XFS_RANDOM_IFLUSH_1)) {
  3083. xfs_cmn_err(XFS_PTAG_IFLUSH, CE_ALERT, mp,
  3084. "xfs_iflush: Bad inode %Lu magic number 0x%x, ptr 0x%p",
  3085. ip->i_ino, (int) INT_GET(dip->di_core.di_magic, ARCH_CONVERT), dip);
  3086. goto corrupt_out;
  3087. }
  3088. if (XFS_TEST_ERROR(ip->i_d.di_magic != XFS_DINODE_MAGIC,
  3089. mp, XFS_ERRTAG_IFLUSH_2, XFS_RANDOM_IFLUSH_2)) {
  3090. xfs_cmn_err(XFS_PTAG_IFLUSH, CE_ALERT, mp,
  3091. "xfs_iflush: Bad inode %Lu, ptr 0x%p, magic number 0x%x",
  3092. ip->i_ino, ip, ip->i_d.di_magic);
  3093. goto corrupt_out;
  3094. }
  3095. if ((ip->i_d.di_mode & S_IFMT) == S_IFREG) {
  3096. if (XFS_TEST_ERROR(
  3097. (ip->i_d.di_format != XFS_DINODE_FMT_EXTENTS) &&
  3098. (ip->i_d.di_format != XFS_DINODE_FMT_BTREE),
  3099. mp, XFS_ERRTAG_IFLUSH_3, XFS_RANDOM_IFLUSH_3)) {
  3100. xfs_cmn_err(XFS_PTAG_IFLUSH, CE_ALERT, mp,
  3101. "xfs_iflush: Bad regular inode %Lu, ptr 0x%p",
  3102. ip->i_ino, ip);
  3103. goto corrupt_out;
  3104. }
  3105. } else if ((ip->i_d.di_mode & S_IFMT) == S_IFDIR) {
  3106. if (XFS_TEST_ERROR(
  3107. (ip->i_d.di_format != XFS_DINODE_FMT_EXTENTS) &&
  3108. (ip->i_d.di_format != XFS_DINODE_FMT_BTREE) &&
  3109. (ip->i_d.di_format != XFS_DINODE_FMT_LOCAL),
  3110. mp, XFS_ERRTAG_IFLUSH_4, XFS_RANDOM_IFLUSH_4)) {
  3111. xfs_cmn_err(XFS_PTAG_IFLUSH, CE_ALERT, mp,
  3112. "xfs_iflush: Bad directory inode %Lu, ptr 0x%p",
  3113. ip->i_ino, ip);
  3114. goto corrupt_out;
  3115. }
  3116. }
  3117. if (XFS_TEST_ERROR(ip->i_d.di_nextents + ip->i_d.di_anextents >
  3118. ip->i_d.di_nblocks, mp, XFS_ERRTAG_IFLUSH_5,
  3119. XFS_RANDOM_IFLUSH_5)) {
  3120. xfs_cmn_err(XFS_PTAG_IFLUSH, CE_ALERT, mp,
  3121. "xfs_iflush: detected corrupt incore inode %Lu, total extents = %d, nblocks = %Ld, ptr 0x%p",
  3122. ip->i_ino,
  3123. ip->i_d.di_nextents + ip->i_d.di_anextents,
  3124. ip->i_d.di_nblocks,
  3125. ip);
  3126. goto corrupt_out;
  3127. }
  3128. if (XFS_TEST_ERROR(ip->i_d.di_forkoff > mp->m_sb.sb_inodesize,
  3129. mp, XFS_ERRTAG_IFLUSH_6, XFS_RANDOM_IFLUSH_6)) {
  3130. xfs_cmn_err(XFS_PTAG_IFLUSH, CE_ALERT, mp,
  3131. "xfs_iflush: bad inode %Lu, forkoff 0x%x, ptr 0x%p",
  3132. ip->i_ino, ip->i_d.di_forkoff, ip);
  3133. goto corrupt_out;
  3134. }
  3135. /*
  3136. * bump the flush iteration count, used to detect flushes which
  3137. * postdate a log record during recovery.
  3138. */
  3139. ip->i_d.di_flushiter++;
  3140. /*
  3141. * Copy the dirty parts of the inode into the on-disk
  3142. * inode. We always copy out the core of the inode,
  3143. * because if the inode is dirty at all the core must
  3144. * be.
  3145. */
  3146. xfs_xlate_dinode_core((xfs_caddr_t)&(dip->di_core), &(ip->i_d), -1);
  3147. /* Wrap, we never let the log put out DI_MAX_FLUSH */
  3148. if (ip->i_d.di_flushiter == DI_MAX_FLUSH)
  3149. ip->i_d.di_flushiter = 0;
  3150. /*
  3151. * If this is really an old format inode and the superblock version
  3152. * has not been updated to support only new format inodes, then
  3153. * convert back to the old inode format. If the superblock version
  3154. * has been updated, then make the conversion permanent.
  3155. */
  3156. ASSERT(ip->i_d.di_version == XFS_DINODE_VERSION_1 ||
  3157. XFS_SB_VERSION_HASNLINK(&mp->m_sb));
  3158. if (ip->i_d.di_version == XFS_DINODE_VERSION_1) {
  3159. if (!XFS_SB_VERSION_HASNLINK(&mp->m_sb)) {
  3160. /*
  3161. * Convert it back.
  3162. */
  3163. ASSERT(ip->i_d.di_nlink <= XFS_MAXLINK_1);
  3164. INT_SET(dip->di_core.di_onlink, ARCH_CONVERT, ip->i_d.di_nlink);
  3165. } else {
  3166. /*
  3167. * The superblock version has already been bumped,
  3168. * so just make the conversion to the new inode
  3169. * format permanent.
  3170. */
  3171. ip->i_d.di_version = XFS_DINODE_VERSION_2;
  3172. INT_SET(dip->di_core.di_version, ARCH_CONVERT, XFS_DINODE_VERSION_2);
  3173. ip->i_d.di_onlink = 0;
  3174. dip->di_core.di_onlink = 0;
  3175. memset(&(ip->i_d.di_pad[0]), 0, sizeof(ip->i_d.di_pad));
  3176. memset(&(dip->di_core.di_pad[0]), 0,
  3177. sizeof(dip->di_core.di_pad));
  3178. ASSERT(ip->i_d.di_projid == 0);
  3179. }
  3180. }
  3181. if (xfs_iflush_fork(ip, dip, iip, XFS_DATA_FORK, bp) == EFSCORRUPTED) {
  3182. goto corrupt_out;
  3183. }
  3184. if (XFS_IFORK_Q(ip)) {
  3185. /*
  3186. * The only error from xfs_iflush_fork is on the data fork.
  3187. */
  3188. (void) xfs_iflush_fork(ip, dip, iip, XFS_ATTR_FORK, bp);
  3189. }
  3190. xfs_inobp_check(mp, bp);
  3191. /*
  3192. * We've recorded everything logged in the inode, so we'd
  3193. * like to clear the ilf_fields bits so we don't log and
  3194. * flush things unnecessarily. However, we can't stop
  3195. * logging all this information until the data we've copied
  3196. * into the disk buffer is written to disk. If we did we might
  3197. * overwrite the copy of the inode in the log with all the
  3198. * data after re-logging only part of it, and in the face of
  3199. * a crash we wouldn't have all the data we need to recover.
  3200. *
  3201. * What we do is move the bits to the ili_last_fields field.
  3202. * When logging the inode, these bits are moved back to the
  3203. * ilf_fields field. In the xfs_iflush_done() routine we
  3204. * clear ili_last_fields, since we know that the information
  3205. * those bits represent is permanently on disk. As long as
  3206. * the flush completes before the inode is logged again, then
  3207. * both ilf_fields and ili_last_fields will be cleared.
  3208. *
  3209. * We can play with the ilf_fields bits here, because the inode
  3210. * lock must be held exclusively in order to set bits there
  3211. * and the flush lock protects the ili_last_fields bits.
  3212. * Set ili_logged so the flush done
  3213. * routine can tell whether or not to look in the AIL.
  3214. * Also, store the current LSN of the inode so that we can tell
  3215. * whether the item has moved in the AIL from xfs_iflush_done().
  3216. * In order to read the lsn we need the AIL lock, because
  3217. * it is a 64 bit value that cannot be read atomically.
  3218. */
  3219. if (iip != NULL && iip->ili_format.ilf_fields != 0) {
  3220. iip->ili_last_fields = iip->ili_format.ilf_fields;
  3221. iip->ili_format.ilf_fields = 0;
  3222. iip->ili_logged = 1;
  3223. ASSERT(sizeof(xfs_lsn_t) == 8); /* don't lock if it shrinks */
  3224. AIL_LOCK(mp,s);
  3225. iip->ili_flush_lsn = iip->ili_item.li_lsn;
  3226. AIL_UNLOCK(mp, s);
  3227. /*
  3228. * Attach the function xfs_iflush_done to the inode's
  3229. * buffer. This will remove the inode from the AIL
  3230. * and unlock the inode's flush lock when the inode is
  3231. * completely written to disk.
  3232. */
  3233. xfs_buf_attach_iodone(bp, (void(*)(xfs_buf_t*,xfs_log_item_t*))
  3234. xfs_iflush_done, (xfs_log_item_t *)iip);
  3235. ASSERT(XFS_BUF_FSPRIVATE(bp, void *) != NULL);
  3236. ASSERT(XFS_BUF_IODONE_FUNC(bp) != NULL);
  3237. } else {
  3238. /*
  3239. * We're flushing an inode which is not in the AIL and has
  3240. * not been logged but has i_update_core set. For this
  3241. * case we can use a B_DELWRI flush and immediately drop
  3242. * the inode flush lock because we can avoid the whole
  3243. * AIL state thing. It's OK to drop the flush lock now,
  3244. * because we've already locked the buffer and to do anything
  3245. * you really need both.
  3246. */
  3247. if (iip != NULL) {
  3248. ASSERT(iip->ili_logged == 0);
  3249. ASSERT(iip->ili_last_fields == 0);
  3250. ASSERT((iip->ili_item.li_flags & XFS_LI_IN_AIL) == 0);
  3251. }
  3252. xfs_ifunlock(ip);
  3253. }
  3254. return 0;
  3255. corrupt_out:
  3256. return XFS_ERROR(EFSCORRUPTED);
  3257. }
  3258. /*
  3259. * Flush all inactive inodes in mp.
  3260. */
  3261. void
  3262. xfs_iflush_all(
  3263. xfs_mount_t *mp)
  3264. {
  3265. xfs_inode_t *ip;
  3266. bhv_vnode_t *vp;
  3267. again:
  3268. XFS_MOUNT_ILOCK(mp);
  3269. ip = mp->m_inodes;
  3270. if (ip == NULL)
  3271. goto out;
  3272. do {
  3273. /* Make sure we skip markers inserted by sync */
  3274. if (ip->i_mount == NULL) {
  3275. ip = ip->i_mnext;
  3276. continue;
  3277. }
  3278. vp = XFS_ITOV_NULL(ip);
  3279. if (!vp) {
  3280. XFS_MOUNT_IUNLOCK(mp);
  3281. xfs_finish_reclaim(ip, 0, XFS_IFLUSH_ASYNC);
  3282. goto again;
  3283. }
  3284. ASSERT(vn_count(vp) == 0);
  3285. ip = ip->i_mnext;
  3286. } while (ip != mp->m_inodes);
  3287. out:
  3288. XFS_MOUNT_IUNLOCK(mp);
  3289. }
  3290. /*
  3291. * xfs_iaccess: check accessibility of inode for mode.
  3292. */
  3293. int
  3294. xfs_iaccess(
  3295. xfs_inode_t *ip,
  3296. mode_t mode,
  3297. cred_t *cr)
  3298. {
  3299. int error;
  3300. mode_t orgmode = mode;
  3301. struct inode *inode = vn_to_inode(XFS_ITOV(ip));
  3302. if (mode & S_IWUSR) {
  3303. umode_t imode = inode->i_mode;
  3304. if (IS_RDONLY(inode) &&
  3305. (S_ISREG(imode) || S_ISDIR(imode) || S_ISLNK(imode)))
  3306. return XFS_ERROR(EROFS);
  3307. if (IS_IMMUTABLE(inode))
  3308. return XFS_ERROR(EACCES);
  3309. }
  3310. /*
  3311. * If there's an Access Control List it's used instead of
  3312. * the mode bits.
  3313. */
  3314. if ((error = _ACL_XFS_IACCESS(ip, mode, cr)) != -1)
  3315. return error ? XFS_ERROR(error) : 0;
  3316. if (current_fsuid(cr) != ip->i_d.di_uid) {
  3317. mode >>= 3;
  3318. if (!in_group_p((gid_t)ip->i_d.di_gid))
  3319. mode >>= 3;
  3320. }
  3321. /*
  3322. * If the DACs are ok we don't need any capability check.
  3323. */
  3324. if ((ip->i_d.di_mode & mode) == mode)
  3325. return 0;
  3326. /*
  3327. * Read/write DACs are always overridable.
  3328. * Executable DACs are overridable if at least one exec bit is set.
  3329. */
  3330. if (!(orgmode & S_IXUSR) ||
  3331. (inode->i_mode & S_IXUGO) || S_ISDIR(inode->i_mode))
  3332. if (capable_cred(cr, CAP_DAC_OVERRIDE))
  3333. return 0;
  3334. if ((orgmode == S_IRUSR) ||
  3335. (S_ISDIR(inode->i_mode) && (!(orgmode & S_IWUSR)))) {
  3336. if (capable_cred(cr, CAP_DAC_READ_SEARCH))
  3337. return 0;
  3338. #ifdef NOISE
  3339. cmn_err(CE_NOTE, "Ick: mode=%o, orgmode=%o", mode, orgmode);
  3340. #endif /* NOISE */
  3341. return XFS_ERROR(EACCES);
  3342. }
  3343. return XFS_ERROR(EACCES);
  3344. }
  3345. /*
  3346. * xfs_iroundup: round up argument to next power of two
  3347. */
  3348. uint
  3349. xfs_iroundup(
  3350. uint v)
  3351. {
  3352. int i;
  3353. uint m;
  3354. if ((v & (v - 1)) == 0)
  3355. return v;
  3356. ASSERT((v & 0x80000000) == 0);
  3357. if ((v & (v + 1)) == 0)
  3358. return v + 1;
  3359. for (i = 0, m = 1; i < 31; i++, m <<= 1) {
  3360. if (v & m)
  3361. continue;
  3362. v |= m;
  3363. if ((v & (v + 1)) == 0)
  3364. return v + 1;
  3365. }
  3366. ASSERT(0);
  3367. return( 0 );
  3368. }
  3369. #ifdef XFS_ILOCK_TRACE
  3370. ktrace_t *xfs_ilock_trace_buf;
  3371. void
  3372. xfs_ilock_trace(xfs_inode_t *ip, int lock, unsigned int lockflags, inst_t *ra)
  3373. {
  3374. ktrace_enter(ip->i_lock_trace,
  3375. (void *)ip,
  3376. (void *)(unsigned long)lock, /* 1 = LOCK, 3=UNLOCK, etc */
  3377. (void *)(unsigned long)lockflags, /* XFS_ILOCK_EXCL etc */
  3378. (void *)ra, /* caller of ilock */
  3379. (void *)(unsigned long)current_cpu(),
  3380. (void *)(unsigned long)current_pid(),
  3381. NULL,NULL,NULL,NULL,NULL,NULL,NULL,NULL,NULL,NULL);
  3382. }
  3383. #endif
  3384. /*
  3385. * Return a pointer to the extent record at file index idx.
  3386. */
  3387. xfs_bmbt_rec_t *
  3388. xfs_iext_get_ext(
  3389. xfs_ifork_t *ifp, /* inode fork pointer */
  3390. xfs_extnum_t idx) /* index of target extent */
  3391. {
  3392. ASSERT(idx >= 0);
  3393. if ((ifp->if_flags & XFS_IFEXTIREC) && (idx == 0)) {
  3394. return ifp->if_u1.if_ext_irec->er_extbuf;
  3395. } else if (ifp->if_flags & XFS_IFEXTIREC) {
  3396. xfs_ext_irec_t *erp; /* irec pointer */
  3397. int erp_idx = 0; /* irec index */
  3398. xfs_extnum_t page_idx = idx; /* ext index in target list */
  3399. erp = xfs_iext_idx_to_irec(ifp, &page_idx, &erp_idx, 0);
  3400. return &erp->er_extbuf[page_idx];
  3401. } else if (ifp->if_bytes) {
  3402. return &ifp->if_u1.if_extents[idx];
  3403. } else {
  3404. return NULL;
  3405. }
  3406. }
  3407. /*
  3408. * Insert new item(s) into the extent records for incore inode
  3409. * fork 'ifp'. 'count' new items are inserted at index 'idx'.
  3410. */
  3411. void
  3412. xfs_iext_insert(
  3413. xfs_ifork_t *ifp, /* inode fork pointer */
  3414. xfs_extnum_t idx, /* starting index of new items */
  3415. xfs_extnum_t count, /* number of inserted items */
  3416. xfs_bmbt_irec_t *new) /* items to insert */
  3417. {
  3418. xfs_bmbt_rec_t *ep; /* extent record pointer */
  3419. xfs_extnum_t i; /* extent record index */
  3420. ASSERT(ifp->if_flags & XFS_IFEXTENTS);
  3421. xfs_iext_add(ifp, idx, count);
  3422. for (i = idx; i < idx + count; i++, new++) {
  3423. ep = xfs_iext_get_ext(ifp, i);
  3424. xfs_bmbt_set_all(ep, new);
  3425. }
  3426. }
  3427. /*
  3428. * This is called when the amount of space required for incore file
  3429. * extents needs to be increased. The ext_diff parameter stores the
  3430. * number of new extents being added and the idx parameter contains
  3431. * the extent index where the new extents will be added. If the new
  3432. * extents are being appended, then we just need to (re)allocate and
  3433. * initialize the space. Otherwise, if the new extents are being
  3434. * inserted into the middle of the existing entries, a bit more work
  3435. * is required to make room for the new extents to be inserted. The
  3436. * caller is responsible for filling in the new extent entries upon
  3437. * return.
  3438. */
  3439. void
  3440. xfs_iext_add(
  3441. xfs_ifork_t *ifp, /* inode fork pointer */
  3442. xfs_extnum_t idx, /* index to begin adding exts */
  3443. int ext_diff) /* number of extents to add */
  3444. {
  3445. int byte_diff; /* new bytes being added */
  3446. int new_size; /* size of extents after adding */
  3447. xfs_extnum_t nextents; /* number of extents in file */
  3448. nextents = ifp->if_bytes / (uint)sizeof(xfs_bmbt_rec_t);
  3449. ASSERT((idx >= 0) && (idx <= nextents));
  3450. byte_diff = ext_diff * sizeof(xfs_bmbt_rec_t);
  3451. new_size = ifp->if_bytes + byte_diff;
  3452. /*
  3453. * If the new number of extents (nextents + ext_diff)
  3454. * fits inside the inode, then continue to use the inline
  3455. * extent buffer.
  3456. */
  3457. if (nextents + ext_diff <= XFS_INLINE_EXTS) {
  3458. if (idx < nextents) {
  3459. memmove(&ifp->if_u2.if_inline_ext[idx + ext_diff],
  3460. &ifp->if_u2.if_inline_ext[idx],
  3461. (nextents - idx) * sizeof(xfs_bmbt_rec_t));
  3462. memset(&ifp->if_u2.if_inline_ext[idx], 0, byte_diff);
  3463. }
  3464. ifp->if_u1.if_extents = ifp->if_u2.if_inline_ext;
  3465. ifp->if_real_bytes = 0;
  3466. ifp->if_lastex = nextents + ext_diff;
  3467. }
  3468. /*
  3469. * Otherwise use a linear (direct) extent list.
  3470. * If the extents are currently inside the inode,
  3471. * xfs_iext_realloc_direct will switch us from
  3472. * inline to direct extent allocation mode.
  3473. */
  3474. else if (nextents + ext_diff <= XFS_LINEAR_EXTS) {
  3475. xfs_iext_realloc_direct(ifp, new_size);
  3476. if (idx < nextents) {
  3477. memmove(&ifp->if_u1.if_extents[idx + ext_diff],
  3478. &ifp->if_u1.if_extents[idx],
  3479. (nextents - idx) * sizeof(xfs_bmbt_rec_t));
  3480. memset(&ifp->if_u1.if_extents[idx], 0, byte_diff);
  3481. }
  3482. }
  3483. /* Indirection array */
  3484. else {
  3485. xfs_ext_irec_t *erp;
  3486. int erp_idx = 0;
  3487. int page_idx = idx;
  3488. ASSERT(nextents + ext_diff > XFS_LINEAR_EXTS);
  3489. if (ifp->if_flags & XFS_IFEXTIREC) {
  3490. erp = xfs_iext_idx_to_irec(ifp, &page_idx, &erp_idx, 1);
  3491. } else {
  3492. xfs_iext_irec_init(ifp);
  3493. ASSERT(ifp->if_flags & XFS_IFEXTIREC);
  3494. erp = ifp->if_u1.if_ext_irec;
  3495. }
  3496. /* Extents fit in target extent page */
  3497. if (erp && erp->er_extcount + ext_diff <= XFS_LINEAR_EXTS) {
  3498. if (page_idx < erp->er_extcount) {
  3499. memmove(&erp->er_extbuf[page_idx + ext_diff],
  3500. &erp->er_extbuf[page_idx],
  3501. (erp->er_extcount - page_idx) *
  3502. sizeof(xfs_bmbt_rec_t));
  3503. memset(&erp->er_extbuf[page_idx], 0, byte_diff);
  3504. }
  3505. erp->er_extcount += ext_diff;
  3506. xfs_iext_irec_update_extoffs(ifp, erp_idx + 1, ext_diff);
  3507. }
  3508. /* Insert a new extent page */
  3509. else if (erp) {
  3510. xfs_iext_add_indirect_multi(ifp,
  3511. erp_idx, page_idx, ext_diff);
  3512. }
  3513. /*
  3514. * If extent(s) are being appended to the last page in
  3515. * the indirection array and the new extent(s) don't fit
  3516. * in the page, then erp is NULL and erp_idx is set to
  3517. * the next index needed in the indirection array.
  3518. */
  3519. else {
  3520. int count = ext_diff;
  3521. while (count) {
  3522. erp = xfs_iext_irec_new(ifp, erp_idx);
  3523. erp->er_extcount = count;
  3524. count -= MIN(count, (int)XFS_LINEAR_EXTS);
  3525. if (count) {
  3526. erp_idx++;
  3527. }
  3528. }
  3529. }
  3530. }
  3531. ifp->if_bytes = new_size;
  3532. }
  3533. /*
  3534. * This is called when incore extents are being added to the indirection
  3535. * array and the new extents do not fit in the target extent list. The
  3536. * erp_idx parameter contains the irec index for the target extent list
  3537. * in the indirection array, and the idx parameter contains the extent
  3538. * index within the list. The number of extents being added is stored
  3539. * in the count parameter.
  3540. *
  3541. * |-------| |-------|
  3542. * | | | | idx - number of extents before idx
  3543. * | idx | | count |
  3544. * | | | | count - number of extents being inserted at idx
  3545. * |-------| |-------|
  3546. * | count | | nex2 | nex2 - number of extents after idx + count
  3547. * |-------| |-------|
  3548. */
  3549. void
  3550. xfs_iext_add_indirect_multi(
  3551. xfs_ifork_t *ifp, /* inode fork pointer */
  3552. int erp_idx, /* target extent irec index */
  3553. xfs_extnum_t idx, /* index within target list */
  3554. int count) /* new extents being added */
  3555. {
  3556. int byte_diff; /* new bytes being added */
  3557. xfs_ext_irec_t *erp; /* pointer to irec entry */
  3558. xfs_extnum_t ext_diff; /* number of extents to add */
  3559. xfs_extnum_t ext_cnt; /* new extents still needed */
  3560. xfs_extnum_t nex2; /* extents after idx + count */
  3561. xfs_bmbt_rec_t *nex2_ep = NULL; /* temp list for nex2 extents */
  3562. int nlists; /* number of irec's (lists) */
  3563. ASSERT(ifp->if_flags & XFS_IFEXTIREC);
  3564. erp = &ifp->if_u1.if_ext_irec[erp_idx];
  3565. nex2 = erp->er_extcount - idx;
  3566. nlists = ifp->if_real_bytes / XFS_IEXT_BUFSZ;
  3567. /*
  3568. * Save second part of target extent list
  3569. * (all extents past */
  3570. if (nex2) {
  3571. byte_diff = nex2 * sizeof(xfs_bmbt_rec_t);
  3572. nex2_ep = (xfs_bmbt_rec_t *) kmem_alloc(byte_diff, KM_SLEEP);
  3573. memmove(nex2_ep, &erp->er_extbuf[idx], byte_diff);
  3574. erp->er_extcount -= nex2;
  3575. xfs_iext_irec_update_extoffs(ifp, erp_idx + 1, -nex2);
  3576. memset(&erp->er_extbuf[idx], 0, byte_diff);
  3577. }
  3578. /*
  3579. * Add the new extents to the end of the target
  3580. * list, then allocate new irec record(s) and
  3581. * extent buffer(s) as needed to store the rest
  3582. * of the new extents.
  3583. */
  3584. ext_cnt = count;
  3585. ext_diff = MIN(ext_cnt, (int)XFS_LINEAR_EXTS - erp->er_extcount);
  3586. if (ext_diff) {
  3587. erp->er_extcount += ext_diff;
  3588. xfs_iext_irec_update_extoffs(ifp, erp_idx + 1, ext_diff);
  3589. ext_cnt -= ext_diff;
  3590. }
  3591. while (ext_cnt) {
  3592. erp_idx++;
  3593. erp = xfs_iext_irec_new(ifp, erp_idx);
  3594. ext_diff = MIN(ext_cnt, (int)XFS_LINEAR_EXTS);
  3595. erp->er_extcount = ext_diff;
  3596. xfs_iext_irec_update_extoffs(ifp, erp_idx + 1, ext_diff);
  3597. ext_cnt -= ext_diff;
  3598. }
  3599. /* Add nex2 extents back to indirection array */
  3600. if (nex2) {
  3601. xfs_extnum_t ext_avail;
  3602. int i;
  3603. byte_diff = nex2 * sizeof(xfs_bmbt_rec_t);
  3604. ext_avail = XFS_LINEAR_EXTS - erp->er_extcount;
  3605. i = 0;
  3606. /*
  3607. * If nex2 extents fit in the current page, append
  3608. * nex2_ep after the new extents.
  3609. */
  3610. if (nex2 <= ext_avail) {
  3611. i = erp->er_extcount;
  3612. }
  3613. /*
  3614. * Otherwise, check if space is available in the
  3615. * next page.
  3616. */
  3617. else if ((erp_idx < nlists - 1) &&
  3618. (nex2 <= (ext_avail = XFS_LINEAR_EXTS -
  3619. ifp->if_u1.if_ext_irec[erp_idx+1].er_extcount))) {
  3620. erp_idx++;
  3621. erp++;
  3622. /* Create a hole for nex2 extents */
  3623. memmove(&erp->er_extbuf[nex2], erp->er_extbuf,
  3624. erp->er_extcount * sizeof(xfs_bmbt_rec_t));
  3625. }
  3626. /*
  3627. * Final choice, create a new extent page for
  3628. * nex2 extents.
  3629. */
  3630. else {
  3631. erp_idx++;
  3632. erp = xfs_iext_irec_new(ifp, erp_idx);
  3633. }
  3634. memmove(&erp->er_extbuf[i], nex2_ep, byte_diff);
  3635. kmem_free(nex2_ep, byte_diff);
  3636. erp->er_extcount += nex2;
  3637. xfs_iext_irec_update_extoffs(ifp, erp_idx + 1, nex2);
  3638. }
  3639. }
  3640. /*
  3641. * This is called when the amount of space required for incore file
  3642. * extents needs to be decreased. The ext_diff parameter stores the
  3643. * number of extents to be removed and the idx parameter contains
  3644. * the extent index where the extents will be removed from.
  3645. *
  3646. * If the amount of space needed has decreased below the linear
  3647. * limit, XFS_IEXT_BUFSZ, then switch to using the contiguous
  3648. * extent array. Otherwise, use kmem_realloc() to adjust the
  3649. * size to what is needed.
  3650. */
  3651. void
  3652. xfs_iext_remove(
  3653. xfs_ifork_t *ifp, /* inode fork pointer */
  3654. xfs_extnum_t idx, /* index to begin removing exts */
  3655. int ext_diff) /* number of extents to remove */
  3656. {
  3657. xfs_extnum_t nextents; /* number of extents in file */
  3658. int new_size; /* size of extents after removal */
  3659. ASSERT(ext_diff > 0);
  3660. nextents = ifp->if_bytes / (uint)sizeof(xfs_bmbt_rec_t);
  3661. new_size = (nextents - ext_diff) * sizeof(xfs_bmbt_rec_t);
  3662. if (new_size == 0) {
  3663. xfs_iext_destroy(ifp);
  3664. } else if (ifp->if_flags & XFS_IFEXTIREC) {
  3665. xfs_iext_remove_indirect(ifp, idx, ext_diff);
  3666. } else if (ifp->if_real_bytes) {
  3667. xfs_iext_remove_direct(ifp, idx, ext_diff);
  3668. } else {
  3669. xfs_iext_remove_inline(ifp, idx, ext_diff);
  3670. }
  3671. ifp->if_bytes = new_size;
  3672. }
  3673. /*
  3674. * This removes ext_diff extents from the inline buffer, beginning
  3675. * at extent index idx.
  3676. */
  3677. void
  3678. xfs_iext_remove_inline(
  3679. xfs_ifork_t *ifp, /* inode fork pointer */
  3680. xfs_extnum_t idx, /* index to begin removing exts */
  3681. int ext_diff) /* number of extents to remove */
  3682. {
  3683. int nextents; /* number of extents in file */
  3684. ASSERT(!(ifp->if_flags & XFS_IFEXTIREC));
  3685. ASSERT(idx < XFS_INLINE_EXTS);
  3686. nextents = ifp->if_bytes / (uint)sizeof(xfs_bmbt_rec_t);
  3687. ASSERT(((nextents - ext_diff) > 0) &&
  3688. (nextents - ext_diff) < XFS_INLINE_EXTS);
  3689. if (idx + ext_diff < nextents) {
  3690. memmove(&ifp->if_u2.if_inline_ext[idx],
  3691. &ifp->if_u2.if_inline_ext[idx + ext_diff],
  3692. (nextents - (idx + ext_diff)) *
  3693. sizeof(xfs_bmbt_rec_t));
  3694. memset(&ifp->if_u2.if_inline_ext[nextents - ext_diff],
  3695. 0, ext_diff * sizeof(xfs_bmbt_rec_t));
  3696. } else {
  3697. memset(&ifp->if_u2.if_inline_ext[idx], 0,
  3698. ext_diff * sizeof(xfs_bmbt_rec_t));
  3699. }
  3700. }
  3701. /*
  3702. * This removes ext_diff extents from a linear (direct) extent list,
  3703. * beginning at extent index idx. If the extents are being removed
  3704. * from the end of the list (ie. truncate) then we just need to re-
  3705. * allocate the list to remove the extra space. Otherwise, if the
  3706. * extents are being removed from the middle of the existing extent
  3707. * entries, then we first need to move the extent records beginning
  3708. * at idx + ext_diff up in the list to overwrite the records being
  3709. * removed, then remove the extra space via kmem_realloc.
  3710. */
  3711. void
  3712. xfs_iext_remove_direct(
  3713. xfs_ifork_t *ifp, /* inode fork pointer */
  3714. xfs_extnum_t idx, /* index to begin removing exts */
  3715. int ext_diff) /* number of extents to remove */
  3716. {
  3717. xfs_extnum_t nextents; /* number of extents in file */
  3718. int new_size; /* size of extents after removal */
  3719. ASSERT(!(ifp->if_flags & XFS_IFEXTIREC));
  3720. new_size = ifp->if_bytes -
  3721. (ext_diff * sizeof(xfs_bmbt_rec_t));
  3722. nextents = ifp->if_bytes / (uint)sizeof(xfs_bmbt_rec_t);
  3723. if (new_size == 0) {
  3724. xfs_iext_destroy(ifp);
  3725. return;
  3726. }
  3727. /* Move extents up in the list (if needed) */
  3728. if (idx + ext_diff < nextents) {
  3729. memmove(&ifp->if_u1.if_extents[idx],
  3730. &ifp->if_u1.if_extents[idx + ext_diff],
  3731. (nextents - (idx + ext_diff)) *
  3732. sizeof(xfs_bmbt_rec_t));
  3733. }
  3734. memset(&ifp->if_u1.if_extents[nextents - ext_diff],
  3735. 0, ext_diff * sizeof(xfs_bmbt_rec_t));
  3736. /*
  3737. * Reallocate the direct extent list. If the extents
  3738. * will fit inside the inode then xfs_iext_realloc_direct
  3739. * will switch from direct to inline extent allocation
  3740. * mode for us.
  3741. */
  3742. xfs_iext_realloc_direct(ifp, new_size);
  3743. ifp->if_bytes = new_size;
  3744. }
  3745. /*
  3746. * This is called when incore extents are being removed from the
  3747. * indirection array and the extents being removed span multiple extent
  3748. * buffers. The idx parameter contains the file extent index where we
  3749. * want to begin removing extents, and the count parameter contains
  3750. * how many extents need to be removed.
  3751. *
  3752. * |-------| |-------|
  3753. * | nex1 | | | nex1 - number of extents before idx
  3754. * |-------| | count |
  3755. * | | | | count - number of extents being removed at idx
  3756. * | count | |-------|
  3757. * | | | nex2 | nex2 - number of extents after idx + count
  3758. * |-------| |-------|
  3759. */
  3760. void
  3761. xfs_iext_remove_indirect(
  3762. xfs_ifork_t *ifp, /* inode fork pointer */
  3763. xfs_extnum_t idx, /* index to begin removing extents */
  3764. int count) /* number of extents to remove */
  3765. {
  3766. xfs_ext_irec_t *erp; /* indirection array pointer */
  3767. int erp_idx = 0; /* indirection array index */
  3768. xfs_extnum_t ext_cnt; /* extents left to remove */
  3769. xfs_extnum_t ext_diff; /* extents to remove in current list */
  3770. xfs_extnum_t nex1; /* number of extents before idx */
  3771. xfs_extnum_t nex2; /* extents after idx + count */
  3772. int nlists; /* entries in indirection array */
  3773. int page_idx = idx; /* index in target extent list */
  3774. ASSERT(ifp->if_flags & XFS_IFEXTIREC);
  3775. erp = xfs_iext_idx_to_irec(ifp, &page_idx, &erp_idx, 0);
  3776. ASSERT(erp != NULL);
  3777. nlists = ifp->if_real_bytes / XFS_IEXT_BUFSZ;
  3778. nex1 = page_idx;
  3779. ext_cnt = count;
  3780. while (ext_cnt) {
  3781. nex2 = MAX((erp->er_extcount - (nex1 + ext_cnt)), 0);
  3782. ext_diff = MIN(ext_cnt, (erp->er_extcount - nex1));
  3783. /*
  3784. * Check for deletion of entire list;
  3785. * xfs_iext_irec_remove() updates extent offsets.
  3786. */
  3787. if (ext_diff == erp->er_extcount) {
  3788. xfs_iext_irec_remove(ifp, erp_idx);
  3789. ext_cnt -= ext_diff;
  3790. nex1 = 0;
  3791. if (ext_cnt) {
  3792. ASSERT(erp_idx < ifp->if_real_bytes /
  3793. XFS_IEXT_BUFSZ);
  3794. erp = &ifp->if_u1.if_ext_irec[erp_idx];
  3795. nex1 = 0;
  3796. continue;
  3797. } else {
  3798. break;
  3799. }
  3800. }
  3801. /* Move extents up (if needed) */
  3802. if (nex2) {
  3803. memmove(&erp->er_extbuf[nex1],
  3804. &erp->er_extbuf[nex1 + ext_diff],
  3805. nex2 * sizeof(xfs_bmbt_rec_t));
  3806. }
  3807. /* Zero out rest of page */
  3808. memset(&erp->er_extbuf[nex1 + nex2], 0, (XFS_IEXT_BUFSZ -
  3809. ((nex1 + nex2) * sizeof(xfs_bmbt_rec_t))));
  3810. /* Update remaining counters */
  3811. erp->er_extcount -= ext_diff;
  3812. xfs_iext_irec_update_extoffs(ifp, erp_idx + 1, -ext_diff);
  3813. ext_cnt -= ext_diff;
  3814. nex1 = 0;
  3815. erp_idx++;
  3816. erp++;
  3817. }
  3818. ifp->if_bytes -= count * sizeof(xfs_bmbt_rec_t);
  3819. xfs_iext_irec_compact(ifp);
  3820. }
  3821. /*
  3822. * Create, destroy, or resize a linear (direct) block of extents.
  3823. */
  3824. void
  3825. xfs_iext_realloc_direct(
  3826. xfs_ifork_t *ifp, /* inode fork pointer */
  3827. int new_size) /* new size of extents */
  3828. {
  3829. int rnew_size; /* real new size of extents */
  3830. rnew_size = new_size;
  3831. ASSERT(!(ifp->if_flags & XFS_IFEXTIREC) ||
  3832. ((new_size >= 0) && (new_size <= XFS_IEXT_BUFSZ) &&
  3833. (new_size != ifp->if_real_bytes)));
  3834. /* Free extent records */
  3835. if (new_size == 0) {
  3836. xfs_iext_destroy(ifp);
  3837. }
  3838. /* Resize direct extent list and zero any new bytes */
  3839. else if (ifp->if_real_bytes) {
  3840. /* Check if extents will fit inside the inode */
  3841. if (new_size <= XFS_INLINE_EXTS * sizeof(xfs_bmbt_rec_t)) {
  3842. xfs_iext_direct_to_inline(ifp, new_size /
  3843. (uint)sizeof(xfs_bmbt_rec_t));
  3844. ifp->if_bytes = new_size;
  3845. return;
  3846. }
  3847. if ((new_size & (new_size - 1)) != 0) {
  3848. rnew_size = xfs_iroundup(new_size);
  3849. }
  3850. if (rnew_size != ifp->if_real_bytes) {
  3851. ifp->if_u1.if_extents = (xfs_bmbt_rec_t *)
  3852. kmem_realloc(ifp->if_u1.if_extents,
  3853. rnew_size,
  3854. ifp->if_real_bytes,
  3855. KM_SLEEP);
  3856. }
  3857. if (rnew_size > ifp->if_real_bytes) {
  3858. memset(&ifp->if_u1.if_extents[ifp->if_bytes /
  3859. (uint)sizeof(xfs_bmbt_rec_t)], 0,
  3860. rnew_size - ifp->if_real_bytes);
  3861. }
  3862. }
  3863. /*
  3864. * Switch from the inline extent buffer to a direct
  3865. * extent list. Be sure to include the inline extent
  3866. * bytes in new_size.
  3867. */
  3868. else {
  3869. new_size += ifp->if_bytes;
  3870. if ((new_size & (new_size - 1)) != 0) {
  3871. rnew_size = xfs_iroundup(new_size);
  3872. }
  3873. xfs_iext_inline_to_direct(ifp, rnew_size);
  3874. }
  3875. ifp->if_real_bytes = rnew_size;
  3876. ifp->if_bytes = new_size;
  3877. }
  3878. /*
  3879. * Switch from linear (direct) extent records to inline buffer.
  3880. */
  3881. void
  3882. xfs_iext_direct_to_inline(
  3883. xfs_ifork_t *ifp, /* inode fork pointer */
  3884. xfs_extnum_t nextents) /* number of extents in file */
  3885. {
  3886. ASSERT(ifp->if_flags & XFS_IFEXTENTS);
  3887. ASSERT(nextents <= XFS_INLINE_EXTS);
  3888. /*
  3889. * The inline buffer was zeroed when we switched
  3890. * from inline to direct extent allocation mode,
  3891. * so we don't need to clear it here.
  3892. */
  3893. memcpy(ifp->if_u2.if_inline_ext, ifp->if_u1.if_extents,
  3894. nextents * sizeof(xfs_bmbt_rec_t));
  3895. kmem_free(ifp->if_u1.if_extents, ifp->if_real_bytes);
  3896. ifp->if_u1.if_extents = ifp->if_u2.if_inline_ext;
  3897. ifp->if_real_bytes = 0;
  3898. }
  3899. /*
  3900. * Switch from inline buffer to linear (direct) extent records.
  3901. * new_size should already be rounded up to the next power of 2
  3902. * by the caller (when appropriate), so use new_size as it is.
  3903. * However, since new_size may be rounded up, we can't update
  3904. * if_bytes here. It is the caller's responsibility to update
  3905. * if_bytes upon return.
  3906. */
  3907. void
  3908. xfs_iext_inline_to_direct(
  3909. xfs_ifork_t *ifp, /* inode fork pointer */
  3910. int new_size) /* number of extents in file */
  3911. {
  3912. ifp->if_u1.if_extents = (xfs_bmbt_rec_t *)
  3913. kmem_alloc(new_size, KM_SLEEP);
  3914. memset(ifp->if_u1.if_extents, 0, new_size);
  3915. if (ifp->if_bytes) {
  3916. memcpy(ifp->if_u1.if_extents, ifp->if_u2.if_inline_ext,
  3917. ifp->if_bytes);
  3918. memset(ifp->if_u2.if_inline_ext, 0, XFS_INLINE_EXTS *
  3919. sizeof(xfs_bmbt_rec_t));
  3920. }
  3921. ifp->if_real_bytes = new_size;
  3922. }
  3923. /*
  3924. * Resize an extent indirection array to new_size bytes.
  3925. */
  3926. void
  3927. xfs_iext_realloc_indirect(
  3928. xfs_ifork_t *ifp, /* inode fork pointer */
  3929. int new_size) /* new indirection array size */
  3930. {
  3931. int nlists; /* number of irec's (ex lists) */
  3932. int size; /* current indirection array size */
  3933. ASSERT(ifp->if_flags & XFS_IFEXTIREC);
  3934. nlists = ifp->if_real_bytes / XFS_IEXT_BUFSZ;
  3935. size = nlists * sizeof(xfs_ext_irec_t);
  3936. ASSERT(ifp->if_real_bytes);
  3937. ASSERT((new_size >= 0) && (new_size != size));
  3938. if (new_size == 0) {
  3939. xfs_iext_destroy(ifp);
  3940. } else {
  3941. ifp->if_u1.if_ext_irec = (xfs_ext_irec_t *)
  3942. kmem_realloc(ifp->if_u1.if_ext_irec,
  3943. new_size, size, KM_SLEEP);
  3944. }
  3945. }
  3946. /*
  3947. * Switch from indirection array to linear (direct) extent allocations.
  3948. */
  3949. void
  3950. xfs_iext_indirect_to_direct(
  3951. xfs_ifork_t *ifp) /* inode fork pointer */
  3952. {
  3953. xfs_bmbt_rec_t *ep; /* extent record pointer */
  3954. xfs_extnum_t nextents; /* number of extents in file */
  3955. int size; /* size of file extents */
  3956. ASSERT(ifp->if_flags & XFS_IFEXTIREC);
  3957. nextents = ifp->if_bytes / (uint)sizeof(xfs_bmbt_rec_t);
  3958. ASSERT(nextents <= XFS_LINEAR_EXTS);
  3959. size = nextents * sizeof(xfs_bmbt_rec_t);
  3960. xfs_iext_irec_compact_full(ifp);
  3961. ASSERT(ifp->if_real_bytes == XFS_IEXT_BUFSZ);
  3962. ep = ifp->if_u1.if_ext_irec->er_extbuf;
  3963. kmem_free(ifp->if_u1.if_ext_irec, sizeof(xfs_ext_irec_t));
  3964. ifp->if_flags &= ~XFS_IFEXTIREC;
  3965. ifp->if_u1.if_extents = ep;
  3966. ifp->if_bytes = size;
  3967. if (nextents < XFS_LINEAR_EXTS) {
  3968. xfs_iext_realloc_direct(ifp, size);
  3969. }
  3970. }
  3971. /*
  3972. * Free incore file extents.
  3973. */
  3974. void
  3975. xfs_iext_destroy(
  3976. xfs_ifork_t *ifp) /* inode fork pointer */
  3977. {
  3978. if (ifp->if_flags & XFS_IFEXTIREC) {
  3979. int erp_idx;
  3980. int nlists;
  3981. nlists = ifp->if_real_bytes / XFS_IEXT_BUFSZ;
  3982. for (erp_idx = nlists - 1; erp_idx >= 0 ; erp_idx--) {
  3983. xfs_iext_irec_remove(ifp, erp_idx);
  3984. }
  3985. ifp->if_flags &= ~XFS_IFEXTIREC;
  3986. } else if (ifp->if_real_bytes) {
  3987. kmem_free(ifp->if_u1.if_extents, ifp->if_real_bytes);
  3988. } else if (ifp->if_bytes) {
  3989. memset(ifp->if_u2.if_inline_ext, 0, XFS_INLINE_EXTS *
  3990. sizeof(xfs_bmbt_rec_t));
  3991. }
  3992. ifp->if_u1.if_extents = NULL;
  3993. ifp->if_real_bytes = 0;
  3994. ifp->if_bytes = 0;
  3995. }
  3996. /*
  3997. * Return a pointer to the extent record for file system block bno.
  3998. */
  3999. xfs_bmbt_rec_t * /* pointer to found extent record */
  4000. xfs_iext_bno_to_ext(
  4001. xfs_ifork_t *ifp, /* inode fork pointer */
  4002. xfs_fileoff_t bno, /* block number to search for */
  4003. xfs_extnum_t *idxp) /* index of target extent */
  4004. {
  4005. xfs_bmbt_rec_t *base; /* pointer to first extent */
  4006. xfs_filblks_t blockcount = 0; /* number of blocks in extent */
  4007. xfs_bmbt_rec_t *ep = NULL; /* pointer to target extent */
  4008. xfs_ext_irec_t *erp = NULL; /* indirection array pointer */
  4009. int high; /* upper boundary in search */
  4010. xfs_extnum_t idx = 0; /* index of target extent */
  4011. int low; /* lower boundary in search */
  4012. xfs_extnum_t nextents; /* number of file extents */
  4013. xfs_fileoff_t startoff = 0; /* start offset of extent */
  4014. nextents = ifp->if_bytes / (uint)sizeof(xfs_bmbt_rec_t);
  4015. if (nextents == 0) {
  4016. *idxp = 0;
  4017. return NULL;
  4018. }
  4019. low = 0;
  4020. if (ifp->if_flags & XFS_IFEXTIREC) {
  4021. /* Find target extent list */
  4022. int erp_idx = 0;
  4023. erp = xfs_iext_bno_to_irec(ifp, bno, &erp_idx);
  4024. base = erp->er_extbuf;
  4025. high = erp->er_extcount - 1;
  4026. } else {
  4027. base = ifp->if_u1.if_extents;
  4028. high = nextents - 1;
  4029. }
  4030. /* Binary search extent records */
  4031. while (low <= high) {
  4032. idx = (low + high) >> 1;
  4033. ep = base + idx;
  4034. startoff = xfs_bmbt_get_startoff(ep);
  4035. blockcount = xfs_bmbt_get_blockcount(ep);
  4036. if (bno < startoff) {
  4037. high = idx - 1;
  4038. } else if (bno >= startoff + blockcount) {
  4039. low = idx + 1;
  4040. } else {
  4041. /* Convert back to file-based extent index */
  4042. if (ifp->if_flags & XFS_IFEXTIREC) {
  4043. idx += erp->er_extoff;
  4044. }
  4045. *idxp = idx;
  4046. return ep;
  4047. }
  4048. }
  4049. /* Convert back to file-based extent index */
  4050. if (ifp->if_flags & XFS_IFEXTIREC) {
  4051. idx += erp->er_extoff;
  4052. }
  4053. if (bno >= startoff + blockcount) {
  4054. if (++idx == nextents) {
  4055. ep = NULL;
  4056. } else {
  4057. ep = xfs_iext_get_ext(ifp, idx);
  4058. }
  4059. }
  4060. *idxp = idx;
  4061. return ep;
  4062. }
  4063. /*
  4064. * Return a pointer to the indirection array entry containing the
  4065. * extent record for filesystem block bno. Store the index of the
  4066. * target irec in *erp_idxp.
  4067. */
  4068. xfs_ext_irec_t * /* pointer to found extent record */
  4069. xfs_iext_bno_to_irec(
  4070. xfs_ifork_t *ifp, /* inode fork pointer */
  4071. xfs_fileoff_t bno, /* block number to search for */
  4072. int *erp_idxp) /* irec index of target ext list */
  4073. {
  4074. xfs_ext_irec_t *erp = NULL; /* indirection array pointer */
  4075. xfs_ext_irec_t *erp_next; /* next indirection array entry */
  4076. int erp_idx; /* indirection array index */
  4077. int nlists; /* number of extent irec's (lists) */
  4078. int high; /* binary search upper limit */
  4079. int low; /* binary search lower limit */
  4080. ASSERT(ifp->if_flags & XFS_IFEXTIREC);
  4081. nlists = ifp->if_real_bytes / XFS_IEXT_BUFSZ;
  4082. erp_idx = 0;
  4083. low = 0;
  4084. high = nlists - 1;
  4085. while (low <= high) {
  4086. erp_idx = (low + high) >> 1;
  4087. erp = &ifp->if_u1.if_ext_irec[erp_idx];
  4088. erp_next = erp_idx < nlists - 1 ? erp + 1 : NULL;
  4089. if (bno < xfs_bmbt_get_startoff(erp->er_extbuf)) {
  4090. high = erp_idx - 1;
  4091. } else if (erp_next && bno >=
  4092. xfs_bmbt_get_startoff(erp_next->er_extbuf)) {
  4093. low = erp_idx + 1;
  4094. } else {
  4095. break;
  4096. }
  4097. }
  4098. *erp_idxp = erp_idx;
  4099. return erp;
  4100. }
  4101. /*
  4102. * Return a pointer to the indirection array entry containing the
  4103. * extent record at file extent index *idxp. Store the index of the
  4104. * target irec in *erp_idxp and store the page index of the target
  4105. * extent record in *idxp.
  4106. */
  4107. xfs_ext_irec_t *
  4108. xfs_iext_idx_to_irec(
  4109. xfs_ifork_t *ifp, /* inode fork pointer */
  4110. xfs_extnum_t *idxp, /* extent index (file -> page) */
  4111. int *erp_idxp, /* pointer to target irec */
  4112. int realloc) /* new bytes were just added */
  4113. {
  4114. xfs_ext_irec_t *prev; /* pointer to previous irec */
  4115. xfs_ext_irec_t *erp = NULL; /* pointer to current irec */
  4116. int erp_idx; /* indirection array index */
  4117. int nlists; /* number of irec's (ex lists) */
  4118. int high; /* binary search upper limit */
  4119. int low; /* binary search lower limit */
  4120. xfs_extnum_t page_idx = *idxp; /* extent index in target list */
  4121. ASSERT(ifp->if_flags & XFS_IFEXTIREC);
  4122. ASSERT(page_idx >= 0 && page_idx <=
  4123. ifp->if_bytes / (uint)sizeof(xfs_bmbt_rec_t));
  4124. nlists = ifp->if_real_bytes / XFS_IEXT_BUFSZ;
  4125. erp_idx = 0;
  4126. low = 0;
  4127. high = nlists - 1;
  4128. /* Binary search extent irec's */
  4129. while (low <= high) {
  4130. erp_idx = (low + high) >> 1;
  4131. erp = &ifp->if_u1.if_ext_irec[erp_idx];
  4132. prev = erp_idx > 0 ? erp - 1 : NULL;
  4133. if (page_idx < erp->er_extoff || (page_idx == erp->er_extoff &&
  4134. realloc && prev && prev->er_extcount < XFS_LINEAR_EXTS)) {
  4135. high = erp_idx - 1;
  4136. } else if (page_idx > erp->er_extoff + erp->er_extcount ||
  4137. (page_idx == erp->er_extoff + erp->er_extcount &&
  4138. !realloc)) {
  4139. low = erp_idx + 1;
  4140. } else if (page_idx == erp->er_extoff + erp->er_extcount &&
  4141. erp->er_extcount == XFS_LINEAR_EXTS) {
  4142. ASSERT(realloc);
  4143. page_idx = 0;
  4144. erp_idx++;
  4145. erp = erp_idx < nlists ? erp + 1 : NULL;
  4146. break;
  4147. } else {
  4148. page_idx -= erp->er_extoff;
  4149. break;
  4150. }
  4151. }
  4152. *idxp = page_idx;
  4153. *erp_idxp = erp_idx;
  4154. return(erp);
  4155. }
  4156. /*
  4157. * Allocate and initialize an indirection array once the space needed
  4158. * for incore extents increases above XFS_IEXT_BUFSZ.
  4159. */
  4160. void
  4161. xfs_iext_irec_init(
  4162. xfs_ifork_t *ifp) /* inode fork pointer */
  4163. {
  4164. xfs_ext_irec_t *erp; /* indirection array pointer */
  4165. xfs_extnum_t nextents; /* number of extents in file */
  4166. ASSERT(!(ifp->if_flags & XFS_IFEXTIREC));
  4167. nextents = ifp->if_bytes / (uint)sizeof(xfs_bmbt_rec_t);
  4168. ASSERT(nextents <= XFS_LINEAR_EXTS);
  4169. erp = (xfs_ext_irec_t *)
  4170. kmem_alloc(sizeof(xfs_ext_irec_t), KM_SLEEP);
  4171. if (nextents == 0) {
  4172. ifp->if_u1.if_extents = (xfs_bmbt_rec_t *)
  4173. kmem_alloc(XFS_IEXT_BUFSZ, KM_SLEEP);
  4174. } else if (!ifp->if_real_bytes) {
  4175. xfs_iext_inline_to_direct(ifp, XFS_IEXT_BUFSZ);
  4176. } else if (ifp->if_real_bytes < XFS_IEXT_BUFSZ) {
  4177. xfs_iext_realloc_direct(ifp, XFS_IEXT_BUFSZ);
  4178. }
  4179. erp->er_extbuf = ifp->if_u1.if_extents;
  4180. erp->er_extcount = nextents;
  4181. erp->er_extoff = 0;
  4182. ifp->if_flags |= XFS_IFEXTIREC;
  4183. ifp->if_real_bytes = XFS_IEXT_BUFSZ;
  4184. ifp->if_bytes = nextents * sizeof(xfs_bmbt_rec_t);
  4185. ifp->if_u1.if_ext_irec = erp;
  4186. return;
  4187. }
  4188. /*
  4189. * Allocate and initialize a new entry in the indirection array.
  4190. */
  4191. xfs_ext_irec_t *
  4192. xfs_iext_irec_new(
  4193. xfs_ifork_t *ifp, /* inode fork pointer */
  4194. int erp_idx) /* index for new irec */
  4195. {
  4196. xfs_ext_irec_t *erp; /* indirection array pointer */
  4197. int i; /* loop counter */
  4198. int nlists; /* number of irec's (ex lists) */
  4199. ASSERT(ifp->if_flags & XFS_IFEXTIREC);
  4200. nlists = ifp->if_real_bytes / XFS_IEXT_BUFSZ;
  4201. /* Resize indirection array */
  4202. xfs_iext_realloc_indirect(ifp, ++nlists *
  4203. sizeof(xfs_ext_irec_t));
  4204. /*
  4205. * Move records down in the array so the
  4206. * new page can use erp_idx.
  4207. */
  4208. erp = ifp->if_u1.if_ext_irec;
  4209. for (i = nlists - 1; i > erp_idx; i--) {
  4210. memmove(&erp[i], &erp[i-1], sizeof(xfs_ext_irec_t));
  4211. }
  4212. ASSERT(i == erp_idx);
  4213. /* Initialize new extent record */
  4214. erp = ifp->if_u1.if_ext_irec;
  4215. erp[erp_idx].er_extbuf = (xfs_bmbt_rec_t *)
  4216. kmem_alloc(XFS_IEXT_BUFSZ, KM_SLEEP);
  4217. ifp->if_real_bytes = nlists * XFS_IEXT_BUFSZ;
  4218. memset(erp[erp_idx].er_extbuf, 0, XFS_IEXT_BUFSZ);
  4219. erp[erp_idx].er_extcount = 0;
  4220. erp[erp_idx].er_extoff = erp_idx > 0 ?
  4221. erp[erp_idx-1].er_extoff + erp[erp_idx-1].er_extcount : 0;
  4222. return (&erp[erp_idx]);
  4223. }
  4224. /*
  4225. * Remove a record from the indirection array.
  4226. */
  4227. void
  4228. xfs_iext_irec_remove(
  4229. xfs_ifork_t *ifp, /* inode fork pointer */
  4230. int erp_idx) /* irec index to remove */
  4231. {
  4232. xfs_ext_irec_t *erp; /* indirection array pointer */
  4233. int i; /* loop counter */
  4234. int nlists; /* number of irec's (ex lists) */
  4235. ASSERT(ifp->if_flags & XFS_IFEXTIREC);
  4236. nlists = ifp->if_real_bytes / XFS_IEXT_BUFSZ;
  4237. erp = &ifp->if_u1.if_ext_irec[erp_idx];
  4238. if (erp->er_extbuf) {
  4239. xfs_iext_irec_update_extoffs(ifp, erp_idx + 1,
  4240. -erp->er_extcount);
  4241. kmem_free(erp->er_extbuf, XFS_IEXT_BUFSZ);
  4242. }
  4243. /* Compact extent records */
  4244. erp = ifp->if_u1.if_ext_irec;
  4245. for (i = erp_idx; i < nlists - 1; i++) {
  4246. memmove(&erp[i], &erp[i+1], sizeof(xfs_ext_irec_t));
  4247. }
  4248. /*
  4249. * Manually free the last extent record from the indirection
  4250. * array. A call to xfs_iext_realloc_indirect() with a size
  4251. * of zero would result in a call to xfs_iext_destroy() which
  4252. * would in turn call this function again, creating a nasty
  4253. * infinite loop.
  4254. */
  4255. if (--nlists) {
  4256. xfs_iext_realloc_indirect(ifp,
  4257. nlists * sizeof(xfs_ext_irec_t));
  4258. } else {
  4259. kmem_free(ifp->if_u1.if_ext_irec,
  4260. sizeof(xfs_ext_irec_t));
  4261. }
  4262. ifp->if_real_bytes = nlists * XFS_IEXT_BUFSZ;
  4263. }
  4264. /*
  4265. * This is called to clean up large amounts of unused memory allocated
  4266. * by the indirection array. Before compacting anything though, verify
  4267. * that the indirection array is still needed and switch back to the
  4268. * linear extent list (or even the inline buffer) if possible. The
  4269. * compaction policy is as follows:
  4270. *
  4271. * Full Compaction: Extents fit into a single page (or inline buffer)
  4272. * Full Compaction: Extents occupy less than 10% of allocated space
  4273. * Partial Compaction: Extents occupy > 10% and < 50% of allocated space
  4274. * No Compaction: Extents occupy at least 50% of allocated space
  4275. */
  4276. void
  4277. xfs_iext_irec_compact(
  4278. xfs_ifork_t *ifp) /* inode fork pointer */
  4279. {
  4280. xfs_extnum_t nextents; /* number of extents in file */
  4281. int nlists; /* number of irec's (ex lists) */
  4282. ASSERT(ifp->if_flags & XFS_IFEXTIREC);
  4283. nlists = ifp->if_real_bytes / XFS_IEXT_BUFSZ;
  4284. nextents = ifp->if_bytes / (uint)sizeof(xfs_bmbt_rec_t);
  4285. if (nextents == 0) {
  4286. xfs_iext_destroy(ifp);
  4287. } else if (nextents <= XFS_INLINE_EXTS) {
  4288. xfs_iext_indirect_to_direct(ifp);
  4289. xfs_iext_direct_to_inline(ifp, nextents);
  4290. } else if (nextents <= XFS_LINEAR_EXTS) {
  4291. xfs_iext_indirect_to_direct(ifp);
  4292. } else if (nextents < (nlists * XFS_LINEAR_EXTS) >> 3) {
  4293. xfs_iext_irec_compact_full(ifp);
  4294. } else if (nextents < (nlists * XFS_LINEAR_EXTS) >> 1) {
  4295. xfs_iext_irec_compact_pages(ifp);
  4296. }
  4297. }
  4298. /*
  4299. * Combine extents from neighboring extent pages.
  4300. */
  4301. void
  4302. xfs_iext_irec_compact_pages(
  4303. xfs_ifork_t *ifp) /* inode fork pointer */
  4304. {
  4305. xfs_ext_irec_t *erp, *erp_next;/* pointers to irec entries */
  4306. int erp_idx = 0; /* indirection array index */
  4307. int nlists; /* number of irec's (ex lists) */
  4308. ASSERT(ifp->if_flags & XFS_IFEXTIREC);
  4309. nlists = ifp->if_real_bytes / XFS_IEXT_BUFSZ;
  4310. while (erp_idx < nlists - 1) {
  4311. erp = &ifp->if_u1.if_ext_irec[erp_idx];
  4312. erp_next = erp + 1;
  4313. if (erp_next->er_extcount <=
  4314. (XFS_LINEAR_EXTS - erp->er_extcount)) {
  4315. memmove(&erp->er_extbuf[erp->er_extcount],
  4316. erp_next->er_extbuf, erp_next->er_extcount *
  4317. sizeof(xfs_bmbt_rec_t));
  4318. erp->er_extcount += erp_next->er_extcount;
  4319. /*
  4320. * Free page before removing extent record
  4321. * so er_extoffs don't get modified in
  4322. * xfs_iext_irec_remove.
  4323. */
  4324. kmem_free(erp_next->er_extbuf, XFS_IEXT_BUFSZ);
  4325. erp_next->er_extbuf = NULL;
  4326. xfs_iext_irec_remove(ifp, erp_idx + 1);
  4327. nlists = ifp->if_real_bytes / XFS_IEXT_BUFSZ;
  4328. } else {
  4329. erp_idx++;
  4330. }
  4331. }
  4332. }
  4333. /*
  4334. * Fully compact the extent records managed by the indirection array.
  4335. */
  4336. void
  4337. xfs_iext_irec_compact_full(
  4338. xfs_ifork_t *ifp) /* inode fork pointer */
  4339. {
  4340. xfs_bmbt_rec_t *ep, *ep_next; /* extent record pointers */
  4341. xfs_ext_irec_t *erp, *erp_next; /* extent irec pointers */
  4342. int erp_idx = 0; /* extent irec index */
  4343. int ext_avail; /* empty entries in ex list */
  4344. int ext_diff; /* number of exts to add */
  4345. int nlists; /* number of irec's (ex lists) */
  4346. ASSERT(ifp->if_flags & XFS_IFEXTIREC);
  4347. nlists = ifp->if_real_bytes / XFS_IEXT_BUFSZ;
  4348. erp = ifp->if_u1.if_ext_irec;
  4349. ep = &erp->er_extbuf[erp->er_extcount];
  4350. erp_next = erp + 1;
  4351. ep_next = erp_next->er_extbuf;
  4352. while (erp_idx < nlists - 1) {
  4353. ext_avail = XFS_LINEAR_EXTS - erp->er_extcount;
  4354. ext_diff = MIN(ext_avail, erp_next->er_extcount);
  4355. memcpy(ep, ep_next, ext_diff * sizeof(xfs_bmbt_rec_t));
  4356. erp->er_extcount += ext_diff;
  4357. erp_next->er_extcount -= ext_diff;
  4358. /* Remove next page */
  4359. if (erp_next->er_extcount == 0) {
  4360. /*
  4361. * Free page before removing extent record
  4362. * so er_extoffs don't get modified in
  4363. * xfs_iext_irec_remove.
  4364. */
  4365. kmem_free(erp_next->er_extbuf,
  4366. erp_next->er_extcount * sizeof(xfs_bmbt_rec_t));
  4367. erp_next->er_extbuf = NULL;
  4368. xfs_iext_irec_remove(ifp, erp_idx + 1);
  4369. erp = &ifp->if_u1.if_ext_irec[erp_idx];
  4370. nlists = ifp->if_real_bytes / XFS_IEXT_BUFSZ;
  4371. /* Update next page */
  4372. } else {
  4373. /* Move rest of page up to become next new page */
  4374. memmove(erp_next->er_extbuf, ep_next,
  4375. erp_next->er_extcount * sizeof(xfs_bmbt_rec_t));
  4376. ep_next = erp_next->er_extbuf;
  4377. memset(&ep_next[erp_next->er_extcount], 0,
  4378. (XFS_LINEAR_EXTS - erp_next->er_extcount) *
  4379. sizeof(xfs_bmbt_rec_t));
  4380. }
  4381. if (erp->er_extcount == XFS_LINEAR_EXTS) {
  4382. erp_idx++;
  4383. if (erp_idx < nlists)
  4384. erp = &ifp->if_u1.if_ext_irec[erp_idx];
  4385. else
  4386. break;
  4387. }
  4388. ep = &erp->er_extbuf[erp->er_extcount];
  4389. erp_next = erp + 1;
  4390. ep_next = erp_next->er_extbuf;
  4391. }
  4392. }
  4393. /*
  4394. * This is called to update the er_extoff field in the indirection
  4395. * array when extents have been added or removed from one of the
  4396. * extent lists. erp_idx contains the irec index to begin updating
  4397. * at and ext_diff contains the number of extents that were added
  4398. * or removed.
  4399. */
  4400. void
  4401. xfs_iext_irec_update_extoffs(
  4402. xfs_ifork_t *ifp, /* inode fork pointer */
  4403. int erp_idx, /* irec index to update */
  4404. int ext_diff) /* number of new extents */
  4405. {
  4406. int i; /* loop counter */
  4407. int nlists; /* number of irec's (ex lists */
  4408. ASSERT(ifp->if_flags & XFS_IFEXTIREC);
  4409. nlists = ifp->if_real_bytes / XFS_IEXT_BUFSZ;
  4410. for (i = erp_idx; i < nlists; i++) {
  4411. ifp->if_u1.if_ext_irec[i].er_extoff += ext_diff;
  4412. }
  4413. }