xfs_log_recover.c 130 KB

123456789101112131415161718192021222324252627282930313233343536373839404142434445464748495051525354555657585960616263646566676869707172737475767778798081828384858687888990919293949596979899100101102103104105106107108109110111112113114115116117118119120121122123124125126127128129130131132133134135136137138139140141142143144145146147148149150151152153154155156157158159160161162163164165166167168169170171172173174175176177178179180181182183184185186187188189190191192193194195196197198199200201202203204205206207208209210211212213214215216217218219220221222223224225226227228229230231232233234235236237238239240241242243244245246247248249250251252253254255256257258259260261262263264265266267268269270271272273274275276277278279280281282283284285286287288289290291292293294295296297298299300301302303304305306307308309310311312313314315316317318319320321322323324325326327328329330331332333334335336337338339340341342343344345346347348349350351352353354355356357358359360361362363364365366367368369370371372373374375376377378379380381382383384385386387388389390391392393394395396397398399400401402403404405406407408409410411412413414415416417418419420421422423424425426427428429430431432433434435436437438439440441442443444445446447448449450451452453454455456457458459460461462463464465466467468469470471472473474475476477478479480481482483484485486487488489490491492493494495496497498499500501502503504505506507508509510511512513514515516517518519520521522523524525526527528529530531532533534535536537538539540541542543544545546547548549550551552553554555556557558559560561562563564565566567568569570571572573574575576577578579580581582583584585586587588589590591592593594595596597598599600601602603604605606607608609610611612613614615616617618619620621622623624625626627628629630631632633634635636637638639640641642643644645646647648649650651652653654655656657658659660661662663664665666667668669670671672673674675676677678679680681682683684685686687688689690691692693694695696697698699700701702703704705706707708709710711712713714715716717718719720721722723724725726727728729730731732733734735736737738739740741742743744745746747748749750751752753754755756757758759760761762763764765766767768769770771772773774775776777778779780781782783784785786787788789790791792793794795796797798799800801802803804805806807808809810811812813814815816817818819820821822823824825826827828829830831832833834835836837838839840841842843844845846847848849850851852853854855856857858859860861862863864865866867868869870871872873874875876877878879880881882883884885886887888889890891892893894895896897898899900901902903904905906907908909910911912913914915916917918919920921922923924925926927928929930931932933934935936937938939940941942943944945946947948949950951952953954955956957958959960961962963964965966967968969970971972973974975976977978979980981982983984985986987988989990991992993994995996997998999100010011002100310041005100610071008100910101011101210131014101510161017101810191020102110221023102410251026102710281029103010311032103310341035103610371038103910401041104210431044104510461047104810491050105110521053105410551056105710581059106010611062106310641065106610671068106910701071107210731074107510761077107810791080108110821083108410851086108710881089109010911092109310941095109610971098109911001101110211031104110511061107110811091110111111121113111411151116111711181119112011211122112311241125112611271128112911301131113211331134113511361137113811391140114111421143114411451146114711481149115011511152115311541155115611571158115911601161116211631164116511661167116811691170117111721173117411751176117711781179118011811182118311841185118611871188118911901191119211931194119511961197119811991200120112021203120412051206120712081209121012111212121312141215121612171218121912201221122212231224122512261227122812291230123112321233123412351236123712381239124012411242124312441245124612471248124912501251125212531254125512561257125812591260126112621263126412651266126712681269127012711272127312741275127612771278127912801281128212831284128512861287128812891290129112921293129412951296129712981299130013011302130313041305130613071308130913101311131213131314131513161317131813191320132113221323132413251326132713281329133013311332133313341335133613371338133913401341134213431344134513461347134813491350135113521353135413551356135713581359136013611362136313641365136613671368136913701371137213731374137513761377137813791380138113821383138413851386138713881389139013911392139313941395139613971398139914001401140214031404140514061407140814091410141114121413141414151416141714181419142014211422142314241425142614271428142914301431143214331434143514361437143814391440144114421443144414451446144714481449145014511452145314541455145614571458145914601461146214631464146514661467146814691470147114721473147414751476147714781479148014811482148314841485148614871488148914901491149214931494149514961497149814991500150115021503150415051506150715081509151015111512151315141515151615171518151915201521152215231524152515261527152815291530153115321533153415351536153715381539154015411542154315441545154615471548154915501551155215531554155515561557155815591560156115621563156415651566156715681569157015711572157315741575157615771578157915801581158215831584158515861587158815891590159115921593159415951596159715981599160016011602160316041605160616071608160916101611161216131614161516161617161816191620162116221623162416251626162716281629163016311632163316341635163616371638163916401641164216431644164516461647164816491650165116521653165416551656165716581659166016611662166316641665166616671668166916701671167216731674167516761677167816791680168116821683168416851686168716881689169016911692169316941695169616971698169917001701170217031704170517061707170817091710171117121713171417151716171717181719172017211722172317241725172617271728172917301731173217331734173517361737173817391740174117421743174417451746174717481749175017511752175317541755175617571758175917601761176217631764176517661767176817691770177117721773177417751776177717781779178017811782178317841785178617871788178917901791179217931794179517961797179817991800180118021803180418051806180718081809181018111812181318141815181618171818181918201821182218231824182518261827182818291830183118321833183418351836183718381839184018411842184318441845184618471848184918501851185218531854185518561857185818591860186118621863186418651866186718681869187018711872187318741875187618771878187918801881188218831884188518861887188818891890189118921893189418951896189718981899190019011902190319041905190619071908190919101911191219131914191519161917191819191920192119221923192419251926192719281929193019311932193319341935193619371938193919401941194219431944194519461947194819491950195119521953195419551956195719581959196019611962196319641965196619671968196919701971197219731974197519761977197819791980198119821983198419851986198719881989199019911992199319941995199619971998199920002001200220032004200520062007200820092010201120122013201420152016201720182019202020212022202320242025202620272028202920302031203220332034203520362037203820392040204120422043204420452046204720482049205020512052205320542055205620572058205920602061206220632064206520662067206820692070207120722073207420752076207720782079208020812082208320842085208620872088208920902091209220932094209520962097209820992100210121022103210421052106210721082109211021112112211321142115211621172118211921202121212221232124212521262127212821292130213121322133213421352136213721382139214021412142214321442145214621472148214921502151215221532154215521562157215821592160216121622163216421652166216721682169217021712172217321742175217621772178217921802181218221832184218521862187218821892190219121922193219421952196219721982199220022012202220322042205220622072208220922102211221222132214221522162217221822192220222122222223222422252226222722282229223022312232223322342235223622372238223922402241224222432244224522462247224822492250225122522253225422552256225722582259226022612262226322642265226622672268226922702271227222732274227522762277227822792280228122822283228422852286228722882289229022912292229322942295229622972298229923002301230223032304230523062307230823092310231123122313231423152316231723182319232023212322232323242325232623272328232923302331233223332334233523362337233823392340234123422343234423452346234723482349235023512352235323542355235623572358235923602361236223632364236523662367236823692370237123722373237423752376237723782379238023812382238323842385238623872388238923902391239223932394239523962397239823992400240124022403240424052406240724082409241024112412241324142415241624172418241924202421242224232424242524262427242824292430243124322433243424352436243724382439244024412442244324442445244624472448244924502451245224532454245524562457245824592460246124622463246424652466246724682469247024712472247324742475247624772478247924802481248224832484248524862487248824892490249124922493249424952496249724982499250025012502250325042505250625072508250925102511251225132514251525162517251825192520252125222523252425252526252725282529253025312532253325342535253625372538253925402541254225432544254525462547254825492550255125522553255425552556255725582559256025612562256325642565256625672568256925702571257225732574257525762577257825792580258125822583258425852586258725882589259025912592259325942595259625972598259926002601260226032604260526062607260826092610261126122613261426152616261726182619262026212622262326242625262626272628262926302631263226332634263526362637263826392640264126422643264426452646264726482649265026512652265326542655265626572658265926602661266226632664266526662667266826692670267126722673267426752676267726782679268026812682268326842685268626872688268926902691269226932694269526962697269826992700270127022703270427052706270727082709271027112712271327142715271627172718271927202721272227232724272527262727272827292730273127322733273427352736273727382739274027412742274327442745274627472748274927502751275227532754275527562757275827592760276127622763276427652766276727682769277027712772277327742775277627772778277927802781278227832784278527862787278827892790279127922793279427952796279727982799280028012802280328042805280628072808280928102811281228132814281528162817281828192820282128222823282428252826282728282829283028312832283328342835283628372838283928402841284228432844284528462847284828492850285128522853285428552856285728582859286028612862286328642865286628672868286928702871287228732874287528762877287828792880288128822883288428852886288728882889289028912892289328942895289628972898289929002901290229032904290529062907290829092910291129122913291429152916291729182919292029212922292329242925292629272928292929302931293229332934293529362937293829392940294129422943294429452946294729482949295029512952295329542955295629572958295929602961296229632964296529662967296829692970297129722973297429752976297729782979298029812982298329842985298629872988298929902991299229932994299529962997299829993000300130023003300430053006300730083009301030113012301330143015301630173018301930203021302230233024302530263027302830293030303130323033303430353036303730383039304030413042304330443045304630473048304930503051305230533054305530563057305830593060306130623063306430653066306730683069307030713072307330743075307630773078307930803081308230833084308530863087308830893090309130923093309430953096309730983099310031013102310331043105310631073108310931103111311231133114311531163117311831193120312131223123312431253126312731283129313031313132313331343135313631373138313931403141314231433144314531463147314831493150315131523153315431553156315731583159316031613162316331643165316631673168316931703171317231733174317531763177317831793180318131823183318431853186318731883189319031913192319331943195319631973198319932003201320232033204320532063207320832093210321132123213321432153216321732183219322032213222322332243225322632273228322932303231323232333234323532363237323832393240324132423243324432453246324732483249325032513252325332543255325632573258325932603261326232633264326532663267326832693270327132723273327432753276327732783279328032813282328332843285328632873288328932903291329232933294329532963297329832993300330133023303330433053306330733083309331033113312331333143315331633173318331933203321332233233324332533263327332833293330333133323333333433353336333733383339334033413342334333443345334633473348334933503351335233533354335533563357335833593360336133623363336433653366336733683369337033713372337333743375337633773378337933803381338233833384338533863387338833893390339133923393339433953396339733983399340034013402340334043405340634073408340934103411341234133414341534163417341834193420342134223423342434253426342734283429343034313432343334343435343634373438343934403441344234433444344534463447344834493450345134523453345434553456345734583459346034613462346334643465346634673468346934703471347234733474347534763477347834793480348134823483348434853486348734883489349034913492349334943495349634973498349935003501350235033504350535063507350835093510351135123513351435153516351735183519352035213522352335243525352635273528352935303531353235333534353535363537353835393540354135423543354435453546354735483549355035513552355335543555355635573558355935603561356235633564356535663567356835693570357135723573357435753576357735783579358035813582358335843585358635873588358935903591359235933594359535963597359835993600360136023603360436053606360736083609361036113612361336143615361636173618361936203621362236233624362536263627362836293630363136323633363436353636363736383639364036413642364336443645364636473648364936503651365236533654365536563657365836593660366136623663366436653666366736683669367036713672367336743675367636773678367936803681368236833684368536863687368836893690369136923693369436953696369736983699370037013702370337043705370637073708370937103711371237133714371537163717371837193720372137223723372437253726372737283729373037313732373337343735373637373738373937403741374237433744374537463747374837493750375137523753375437553756375737583759376037613762376337643765376637673768376937703771377237733774377537763777377837793780378137823783378437853786378737883789379037913792379337943795379637973798379938003801380238033804380538063807380838093810381138123813381438153816381738183819382038213822382338243825382638273828382938303831383238333834383538363837383838393840384138423843384438453846384738483849385038513852385338543855385638573858385938603861386238633864386538663867386838693870387138723873387438753876387738783879388038813882388338843885388638873888388938903891389238933894389538963897389838993900390139023903390439053906390739083909391039113912391339143915391639173918391939203921392239233924392539263927392839293930393139323933393439353936393739383939394039413942394339443945394639473948394939503951395239533954395539563957395839593960396139623963396439653966396739683969397039713972397339743975397639773978397939803981398239833984398539863987398839893990399139923993399439953996399739983999400040014002400340044005400640074008400940104011401240134014401540164017401840194020402140224023402440254026402740284029403040314032403340344035403640374038403940404041404240434044404540464047404840494050405140524053405440554056405740584059406040614062406340644065406640674068406940704071407240734074407540764077407840794080408140824083408440854086408740884089409040914092409340944095409640974098409941004101410241034104410541064107410841094110411141124113411441154116411741184119412041214122412341244125412641274128412941304131413241334134413541364137413841394140414141424143414441454146414741484149415041514152415341544155415641574158415941604161416241634164416541664167416841694170417141724173417441754176417741784179418041814182418341844185418641874188418941904191419241934194419541964197419841994200420142024203420442054206420742084209421042114212421342144215421642174218421942204221422242234224422542264227422842294230423142324233423442354236423742384239424042414242424342444245424642474248424942504251425242534254425542564257425842594260426142624263426442654266426742684269427042714272427342744275427642774278427942804281428242834284428542864287428842894290429142924293429442954296429742984299430043014302430343044305430643074308430943104311431243134314431543164317431843194320432143224323432443254326432743284329433043314332433343344335433643374338433943404341434243434344434543464347434843494350435143524353435443554356435743584359436043614362436343644365436643674368436943704371437243734374437543764377437843794380438143824383438443854386438743884389439043914392439343944395439643974398439944004401440244034404440544064407440844094410441144124413441444154416441744184419442044214422442344244425442644274428442944304431443244334434443544364437443844394440444144424443444444454446444744484449445044514452445344544455445644574458445944604461446244634464446544664467446844694470447144724473447444754476447744784479448044814482448344844485448644874488448944904491449244934494449544964497449844994500450145024503450445054506450745084509451045114512451345144515451645174518451945204521452245234524452545264527452845294530453145324533453445354536453745384539454045414542454345444545454645474548454945504551455245534554455545564557455845594560456145624563456445654566456745684569457045714572457345744575457645774578457945804581458245834584458545864587458845894590459145924593459445954596459745984599460046014602460346044605460646074608460946104611461246134614461546164617461846194620462146224623462446254626462746284629463046314632463346344635463646374638463946404641464246434644464546464647464846494650465146524653465446554656465746584659466046614662466346644665466646674668466946704671467246734674467546764677467846794680468146824683468446854686468746884689469046914692469346944695469646974698469947004701470247034704470547064707470847094710471147124713471447154716471747184719472047214722
  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_format.h"
  21. #include "xfs_bit.h"
  22. #include "xfs_log.h"
  23. #include "xfs_inum.h"
  24. #include "xfs_trans.h"
  25. #include "xfs_sb.h"
  26. #include "xfs_ag.h"
  27. #include "xfs_mount.h"
  28. #include "xfs_error.h"
  29. #include "xfs_bmap_btree.h"
  30. #include "xfs_alloc_btree.h"
  31. #include "xfs_ialloc_btree.h"
  32. #include "xfs_btree.h"
  33. #include "xfs_dinode.h"
  34. #include "xfs_inode.h"
  35. #include "xfs_inode_item.h"
  36. #include "xfs_alloc.h"
  37. #include "xfs_ialloc.h"
  38. #include "xfs_log_priv.h"
  39. #include "xfs_buf_item.h"
  40. #include "xfs_log_recover.h"
  41. #include "xfs_extfree_item.h"
  42. #include "xfs_trans_priv.h"
  43. #include "xfs_quota.h"
  44. #include "xfs_cksum.h"
  45. #include "xfs_trace.h"
  46. #include "xfs_icache.h"
  47. #include "xfs_icreate_item.h"
  48. /* Need all the magic numbers and buffer ops structures from these headers */
  49. #include "xfs_symlink.h"
  50. #include "xfs_da_btree.h"
  51. #include "xfs_dir2_format.h"
  52. #include "xfs_dir2.h"
  53. #include "xfs_attr_leaf.h"
  54. #include "xfs_attr_remote.h"
  55. #define BLK_AVG(blk1, blk2) ((blk1+blk2) >> 1)
  56. STATIC int
  57. xlog_find_zeroed(
  58. struct xlog *,
  59. xfs_daddr_t *);
  60. STATIC int
  61. xlog_clear_stale_blocks(
  62. struct xlog *,
  63. xfs_lsn_t);
  64. #if defined(DEBUG)
  65. STATIC void
  66. xlog_recover_check_summary(
  67. struct xlog *);
  68. #else
  69. #define xlog_recover_check_summary(log)
  70. #endif
  71. /*
  72. * This structure is used during recovery to record the buf log items which
  73. * have been canceled and should not be replayed.
  74. */
  75. struct xfs_buf_cancel {
  76. xfs_daddr_t bc_blkno;
  77. uint bc_len;
  78. int bc_refcount;
  79. struct list_head bc_list;
  80. };
  81. /*
  82. * Sector aligned buffer routines for buffer create/read/write/access
  83. */
  84. /*
  85. * Verify the given count of basic blocks is valid number of blocks
  86. * to specify for an operation involving the given XFS log buffer.
  87. * Returns nonzero if the count is valid, 0 otherwise.
  88. */
  89. static inline int
  90. xlog_buf_bbcount_valid(
  91. struct xlog *log,
  92. int bbcount)
  93. {
  94. return bbcount > 0 && bbcount <= log->l_logBBsize;
  95. }
  96. /*
  97. * Allocate a buffer to hold log data. The buffer needs to be able
  98. * to map to a range of nbblks basic blocks at any valid (basic
  99. * block) offset within the log.
  100. */
  101. STATIC xfs_buf_t *
  102. xlog_get_bp(
  103. struct xlog *log,
  104. int nbblks)
  105. {
  106. struct xfs_buf *bp;
  107. if (!xlog_buf_bbcount_valid(log, nbblks)) {
  108. xfs_warn(log->l_mp, "Invalid block length (0x%x) for buffer",
  109. nbblks);
  110. XFS_ERROR_REPORT(__func__, XFS_ERRLEVEL_HIGH, log->l_mp);
  111. return NULL;
  112. }
  113. /*
  114. * We do log I/O in units of log sectors (a power-of-2
  115. * multiple of the basic block size), so we round up the
  116. * requested size to accommodate the basic blocks required
  117. * for complete log sectors.
  118. *
  119. * In addition, the buffer may be used for a non-sector-
  120. * aligned block offset, in which case an I/O of the
  121. * requested size could extend beyond the end of the
  122. * buffer. If the requested size is only 1 basic block it
  123. * will never straddle a sector boundary, so this won't be
  124. * an issue. Nor will this be a problem if the log I/O is
  125. * done in basic blocks (sector size 1). But otherwise we
  126. * extend the buffer by one extra log sector to ensure
  127. * there's space to accommodate this possibility.
  128. */
  129. if (nbblks > 1 && log->l_sectBBsize > 1)
  130. nbblks += log->l_sectBBsize;
  131. nbblks = round_up(nbblks, log->l_sectBBsize);
  132. bp = xfs_buf_get_uncached(log->l_mp->m_logdev_targp, nbblks, 0);
  133. if (bp)
  134. xfs_buf_unlock(bp);
  135. return bp;
  136. }
  137. STATIC void
  138. xlog_put_bp(
  139. xfs_buf_t *bp)
  140. {
  141. xfs_buf_free(bp);
  142. }
  143. /*
  144. * Return the address of the start of the given block number's data
  145. * in a log buffer. The buffer covers a log sector-aligned region.
  146. */
  147. STATIC xfs_caddr_t
  148. xlog_align(
  149. struct xlog *log,
  150. xfs_daddr_t blk_no,
  151. int nbblks,
  152. struct xfs_buf *bp)
  153. {
  154. xfs_daddr_t offset = blk_no & ((xfs_daddr_t)log->l_sectBBsize - 1);
  155. ASSERT(offset + nbblks <= bp->b_length);
  156. return bp->b_addr + BBTOB(offset);
  157. }
  158. /*
  159. * nbblks should be uint, but oh well. Just want to catch that 32-bit length.
  160. */
  161. STATIC int
  162. xlog_bread_noalign(
  163. struct xlog *log,
  164. xfs_daddr_t blk_no,
  165. int nbblks,
  166. struct xfs_buf *bp)
  167. {
  168. int error;
  169. if (!xlog_buf_bbcount_valid(log, nbblks)) {
  170. xfs_warn(log->l_mp, "Invalid block length (0x%x) for buffer",
  171. nbblks);
  172. XFS_ERROR_REPORT(__func__, XFS_ERRLEVEL_HIGH, log->l_mp);
  173. return EFSCORRUPTED;
  174. }
  175. blk_no = round_down(blk_no, log->l_sectBBsize);
  176. nbblks = round_up(nbblks, log->l_sectBBsize);
  177. ASSERT(nbblks > 0);
  178. ASSERT(nbblks <= bp->b_length);
  179. XFS_BUF_SET_ADDR(bp, log->l_logBBstart + blk_no);
  180. XFS_BUF_READ(bp);
  181. bp->b_io_length = nbblks;
  182. bp->b_error = 0;
  183. xfsbdstrat(log->l_mp, bp);
  184. error = xfs_buf_iowait(bp);
  185. if (error)
  186. xfs_buf_ioerror_alert(bp, __func__);
  187. return error;
  188. }
  189. STATIC int
  190. xlog_bread(
  191. struct xlog *log,
  192. xfs_daddr_t blk_no,
  193. int nbblks,
  194. struct xfs_buf *bp,
  195. xfs_caddr_t *offset)
  196. {
  197. int error;
  198. error = xlog_bread_noalign(log, blk_no, nbblks, bp);
  199. if (error)
  200. return error;
  201. *offset = xlog_align(log, blk_no, nbblks, bp);
  202. return 0;
  203. }
  204. /*
  205. * Read at an offset into the buffer. Returns with the buffer in it's original
  206. * state regardless of the result of the read.
  207. */
  208. STATIC int
  209. xlog_bread_offset(
  210. struct xlog *log,
  211. xfs_daddr_t blk_no, /* block to read from */
  212. int nbblks, /* blocks to read */
  213. struct xfs_buf *bp,
  214. xfs_caddr_t offset)
  215. {
  216. xfs_caddr_t orig_offset = bp->b_addr;
  217. int orig_len = BBTOB(bp->b_length);
  218. int error, error2;
  219. error = xfs_buf_associate_memory(bp, offset, BBTOB(nbblks));
  220. if (error)
  221. return error;
  222. error = xlog_bread_noalign(log, blk_no, nbblks, bp);
  223. /* must reset buffer pointer even on error */
  224. error2 = xfs_buf_associate_memory(bp, orig_offset, orig_len);
  225. if (error)
  226. return error;
  227. return error2;
  228. }
  229. /*
  230. * Write out the buffer at the given block for the given number of blocks.
  231. * The buffer is kept locked across the write and is returned locked.
  232. * This can only be used for synchronous log writes.
  233. */
  234. STATIC int
  235. xlog_bwrite(
  236. struct xlog *log,
  237. xfs_daddr_t blk_no,
  238. int nbblks,
  239. struct xfs_buf *bp)
  240. {
  241. int error;
  242. if (!xlog_buf_bbcount_valid(log, nbblks)) {
  243. xfs_warn(log->l_mp, "Invalid block length (0x%x) for buffer",
  244. nbblks);
  245. XFS_ERROR_REPORT(__func__, XFS_ERRLEVEL_HIGH, log->l_mp);
  246. return EFSCORRUPTED;
  247. }
  248. blk_no = round_down(blk_no, log->l_sectBBsize);
  249. nbblks = round_up(nbblks, log->l_sectBBsize);
  250. ASSERT(nbblks > 0);
  251. ASSERT(nbblks <= bp->b_length);
  252. XFS_BUF_SET_ADDR(bp, log->l_logBBstart + blk_no);
  253. XFS_BUF_ZEROFLAGS(bp);
  254. xfs_buf_hold(bp);
  255. xfs_buf_lock(bp);
  256. bp->b_io_length = nbblks;
  257. bp->b_error = 0;
  258. error = xfs_bwrite(bp);
  259. if (error)
  260. xfs_buf_ioerror_alert(bp, __func__);
  261. xfs_buf_relse(bp);
  262. return error;
  263. }
  264. #ifdef DEBUG
  265. /*
  266. * dump debug superblock and log record information
  267. */
  268. STATIC void
  269. xlog_header_check_dump(
  270. xfs_mount_t *mp,
  271. xlog_rec_header_t *head)
  272. {
  273. xfs_debug(mp, "%s: SB : uuid = %pU, fmt = %d\n",
  274. __func__, &mp->m_sb.sb_uuid, XLOG_FMT);
  275. xfs_debug(mp, " log : uuid = %pU, fmt = %d\n",
  276. &head->h_fs_uuid, be32_to_cpu(head->h_fmt));
  277. }
  278. #else
  279. #define xlog_header_check_dump(mp, head)
  280. #endif
  281. /*
  282. * check log record header for recovery
  283. */
  284. STATIC int
  285. xlog_header_check_recover(
  286. xfs_mount_t *mp,
  287. xlog_rec_header_t *head)
  288. {
  289. ASSERT(head->h_magicno == cpu_to_be32(XLOG_HEADER_MAGIC_NUM));
  290. /*
  291. * IRIX doesn't write the h_fmt field and leaves it zeroed
  292. * (XLOG_FMT_UNKNOWN). This stops us from trying to recover
  293. * a dirty log created in IRIX.
  294. */
  295. if (unlikely(head->h_fmt != cpu_to_be32(XLOG_FMT))) {
  296. xfs_warn(mp,
  297. "dirty log written in incompatible format - can't recover");
  298. xlog_header_check_dump(mp, head);
  299. XFS_ERROR_REPORT("xlog_header_check_recover(1)",
  300. XFS_ERRLEVEL_HIGH, mp);
  301. return XFS_ERROR(EFSCORRUPTED);
  302. } else if (unlikely(!uuid_equal(&mp->m_sb.sb_uuid, &head->h_fs_uuid))) {
  303. xfs_warn(mp,
  304. "dirty log entry has mismatched uuid - can't recover");
  305. xlog_header_check_dump(mp, head);
  306. XFS_ERROR_REPORT("xlog_header_check_recover(2)",
  307. XFS_ERRLEVEL_HIGH, mp);
  308. return XFS_ERROR(EFSCORRUPTED);
  309. }
  310. return 0;
  311. }
  312. /*
  313. * read the head block of the log and check the header
  314. */
  315. STATIC int
  316. xlog_header_check_mount(
  317. xfs_mount_t *mp,
  318. xlog_rec_header_t *head)
  319. {
  320. ASSERT(head->h_magicno == cpu_to_be32(XLOG_HEADER_MAGIC_NUM));
  321. if (uuid_is_nil(&head->h_fs_uuid)) {
  322. /*
  323. * IRIX doesn't write the h_fs_uuid or h_fmt fields. If
  324. * h_fs_uuid is nil, we assume this log was last mounted
  325. * by IRIX and continue.
  326. */
  327. xfs_warn(mp, "nil uuid in log - IRIX style log");
  328. } else if (unlikely(!uuid_equal(&mp->m_sb.sb_uuid, &head->h_fs_uuid))) {
  329. xfs_warn(mp, "log has mismatched uuid - can't recover");
  330. xlog_header_check_dump(mp, head);
  331. XFS_ERROR_REPORT("xlog_header_check_mount",
  332. XFS_ERRLEVEL_HIGH, mp);
  333. return XFS_ERROR(EFSCORRUPTED);
  334. }
  335. return 0;
  336. }
  337. STATIC void
  338. xlog_recover_iodone(
  339. struct xfs_buf *bp)
  340. {
  341. if (bp->b_error) {
  342. /*
  343. * We're not going to bother about retrying
  344. * this during recovery. One strike!
  345. */
  346. xfs_buf_ioerror_alert(bp, __func__);
  347. xfs_force_shutdown(bp->b_target->bt_mount,
  348. SHUTDOWN_META_IO_ERROR);
  349. }
  350. bp->b_iodone = NULL;
  351. xfs_buf_ioend(bp, 0);
  352. }
  353. /*
  354. * This routine finds (to an approximation) the first block in the physical
  355. * log which contains the given cycle. It uses a binary search algorithm.
  356. * Note that the algorithm can not be perfect because the disk will not
  357. * necessarily be perfect.
  358. */
  359. STATIC int
  360. xlog_find_cycle_start(
  361. struct xlog *log,
  362. struct xfs_buf *bp,
  363. xfs_daddr_t first_blk,
  364. xfs_daddr_t *last_blk,
  365. uint cycle)
  366. {
  367. xfs_caddr_t offset;
  368. xfs_daddr_t mid_blk;
  369. xfs_daddr_t end_blk;
  370. uint mid_cycle;
  371. int error;
  372. end_blk = *last_blk;
  373. mid_blk = BLK_AVG(first_blk, end_blk);
  374. while (mid_blk != first_blk && mid_blk != end_blk) {
  375. error = xlog_bread(log, mid_blk, 1, bp, &offset);
  376. if (error)
  377. return error;
  378. mid_cycle = xlog_get_cycle(offset);
  379. if (mid_cycle == cycle)
  380. end_blk = mid_blk; /* last_half_cycle == mid_cycle */
  381. else
  382. first_blk = mid_blk; /* first_half_cycle == mid_cycle */
  383. mid_blk = BLK_AVG(first_blk, end_blk);
  384. }
  385. ASSERT((mid_blk == first_blk && mid_blk+1 == end_blk) ||
  386. (mid_blk == end_blk && mid_blk-1 == first_blk));
  387. *last_blk = end_blk;
  388. return 0;
  389. }
  390. /*
  391. * Check that a range of blocks does not contain stop_on_cycle_no.
  392. * Fill in *new_blk with the block offset where such a block is
  393. * found, or with -1 (an invalid block number) if there is no such
  394. * block in the range. The scan needs to occur from front to back
  395. * and the pointer into the region must be updated since a later
  396. * routine will need to perform another test.
  397. */
  398. STATIC int
  399. xlog_find_verify_cycle(
  400. struct xlog *log,
  401. xfs_daddr_t start_blk,
  402. int nbblks,
  403. uint stop_on_cycle_no,
  404. xfs_daddr_t *new_blk)
  405. {
  406. xfs_daddr_t i, j;
  407. uint cycle;
  408. xfs_buf_t *bp;
  409. xfs_daddr_t bufblks;
  410. xfs_caddr_t buf = NULL;
  411. int error = 0;
  412. /*
  413. * Greedily allocate a buffer big enough to handle the full
  414. * range of basic blocks we'll be examining. If that fails,
  415. * try a smaller size. We need to be able to read at least
  416. * a log sector, or we're out of luck.
  417. */
  418. bufblks = 1 << ffs(nbblks);
  419. while (bufblks > log->l_logBBsize)
  420. bufblks >>= 1;
  421. while (!(bp = xlog_get_bp(log, bufblks))) {
  422. bufblks >>= 1;
  423. if (bufblks < log->l_sectBBsize)
  424. return ENOMEM;
  425. }
  426. for (i = start_blk; i < start_blk + nbblks; i += bufblks) {
  427. int bcount;
  428. bcount = min(bufblks, (start_blk + nbblks - i));
  429. error = xlog_bread(log, i, bcount, bp, &buf);
  430. if (error)
  431. goto out;
  432. for (j = 0; j < bcount; j++) {
  433. cycle = xlog_get_cycle(buf);
  434. if (cycle == stop_on_cycle_no) {
  435. *new_blk = i+j;
  436. goto out;
  437. }
  438. buf += BBSIZE;
  439. }
  440. }
  441. *new_blk = -1;
  442. out:
  443. xlog_put_bp(bp);
  444. return error;
  445. }
  446. /*
  447. * Potentially backup over partial log record write.
  448. *
  449. * In the typical case, last_blk is the number of the block directly after
  450. * a good log record. Therefore, we subtract one to get the block number
  451. * of the last block in the given buffer. extra_bblks contains the number
  452. * of blocks we would have read on a previous read. This happens when the
  453. * last log record is split over the end of the physical log.
  454. *
  455. * extra_bblks is the number of blocks potentially verified on a previous
  456. * call to this routine.
  457. */
  458. STATIC int
  459. xlog_find_verify_log_record(
  460. struct xlog *log,
  461. xfs_daddr_t start_blk,
  462. xfs_daddr_t *last_blk,
  463. int extra_bblks)
  464. {
  465. xfs_daddr_t i;
  466. xfs_buf_t *bp;
  467. xfs_caddr_t offset = NULL;
  468. xlog_rec_header_t *head = NULL;
  469. int error = 0;
  470. int smallmem = 0;
  471. int num_blks = *last_blk - start_blk;
  472. int xhdrs;
  473. ASSERT(start_blk != 0 || *last_blk != start_blk);
  474. if (!(bp = xlog_get_bp(log, num_blks))) {
  475. if (!(bp = xlog_get_bp(log, 1)))
  476. return ENOMEM;
  477. smallmem = 1;
  478. } else {
  479. error = xlog_bread(log, start_blk, num_blks, bp, &offset);
  480. if (error)
  481. goto out;
  482. offset += ((num_blks - 1) << BBSHIFT);
  483. }
  484. for (i = (*last_blk) - 1; i >= 0; i--) {
  485. if (i < start_blk) {
  486. /* valid log record not found */
  487. xfs_warn(log->l_mp,
  488. "Log inconsistent (didn't find previous header)");
  489. ASSERT(0);
  490. error = XFS_ERROR(EIO);
  491. goto out;
  492. }
  493. if (smallmem) {
  494. error = xlog_bread(log, i, 1, bp, &offset);
  495. if (error)
  496. goto out;
  497. }
  498. head = (xlog_rec_header_t *)offset;
  499. if (head->h_magicno == cpu_to_be32(XLOG_HEADER_MAGIC_NUM))
  500. break;
  501. if (!smallmem)
  502. offset -= BBSIZE;
  503. }
  504. /*
  505. * We hit the beginning of the physical log & still no header. Return
  506. * to caller. If caller can handle a return of -1, then this routine
  507. * will be called again for the end of the physical log.
  508. */
  509. if (i == -1) {
  510. error = -1;
  511. goto out;
  512. }
  513. /*
  514. * We have the final block of the good log (the first block
  515. * of the log record _before_ the head. So we check the uuid.
  516. */
  517. if ((error = xlog_header_check_mount(log->l_mp, head)))
  518. goto out;
  519. /*
  520. * We may have found a log record header before we expected one.
  521. * last_blk will be the 1st block # with a given cycle #. We may end
  522. * up reading an entire log record. In this case, we don't want to
  523. * reset last_blk. Only when last_blk points in the middle of a log
  524. * record do we update last_blk.
  525. */
  526. if (xfs_sb_version_haslogv2(&log->l_mp->m_sb)) {
  527. uint h_size = be32_to_cpu(head->h_size);
  528. xhdrs = h_size / XLOG_HEADER_CYCLE_SIZE;
  529. if (h_size % XLOG_HEADER_CYCLE_SIZE)
  530. xhdrs++;
  531. } else {
  532. xhdrs = 1;
  533. }
  534. if (*last_blk - i + extra_bblks !=
  535. BTOBB(be32_to_cpu(head->h_len)) + xhdrs)
  536. *last_blk = i;
  537. out:
  538. xlog_put_bp(bp);
  539. return error;
  540. }
  541. /*
  542. * Head is defined to be the point of the log where the next log write
  543. * could go. This means that incomplete LR writes at the end are
  544. * eliminated when calculating the head. We aren't guaranteed that previous
  545. * LR have complete transactions. We only know that a cycle number of
  546. * current cycle number -1 won't be present in the log if we start writing
  547. * from our current block number.
  548. *
  549. * last_blk contains the block number of the first block with a given
  550. * cycle number.
  551. *
  552. * Return: zero if normal, non-zero if error.
  553. */
  554. STATIC int
  555. xlog_find_head(
  556. struct xlog *log,
  557. xfs_daddr_t *return_head_blk)
  558. {
  559. xfs_buf_t *bp;
  560. xfs_caddr_t offset;
  561. xfs_daddr_t new_blk, first_blk, start_blk, last_blk, head_blk;
  562. int num_scan_bblks;
  563. uint first_half_cycle, last_half_cycle;
  564. uint stop_on_cycle;
  565. int error, log_bbnum = log->l_logBBsize;
  566. /* Is the end of the log device zeroed? */
  567. if ((error = xlog_find_zeroed(log, &first_blk)) == -1) {
  568. *return_head_blk = first_blk;
  569. /* Is the whole lot zeroed? */
  570. if (!first_blk) {
  571. /* Linux XFS shouldn't generate totally zeroed logs -
  572. * mkfs etc write a dummy unmount record to a fresh
  573. * log so we can store the uuid in there
  574. */
  575. xfs_warn(log->l_mp, "totally zeroed log");
  576. }
  577. return 0;
  578. } else if (error) {
  579. xfs_warn(log->l_mp, "empty log check failed");
  580. return error;
  581. }
  582. first_blk = 0; /* get cycle # of 1st block */
  583. bp = xlog_get_bp(log, 1);
  584. if (!bp)
  585. return ENOMEM;
  586. error = xlog_bread(log, 0, 1, bp, &offset);
  587. if (error)
  588. goto bp_err;
  589. first_half_cycle = xlog_get_cycle(offset);
  590. last_blk = head_blk = log_bbnum - 1; /* get cycle # of last block */
  591. error = xlog_bread(log, last_blk, 1, bp, &offset);
  592. if (error)
  593. goto bp_err;
  594. last_half_cycle = xlog_get_cycle(offset);
  595. ASSERT(last_half_cycle != 0);
  596. /*
  597. * If the 1st half cycle number is equal to the last half cycle number,
  598. * then the entire log is stamped with the same cycle number. In this
  599. * case, head_blk can't be set to zero (which makes sense). The below
  600. * math doesn't work out properly with head_blk equal to zero. Instead,
  601. * we set it to log_bbnum which is an invalid block number, but this
  602. * value makes the math correct. If head_blk doesn't changed through
  603. * all the tests below, *head_blk is set to zero at the very end rather
  604. * than log_bbnum. In a sense, log_bbnum and zero are the same block
  605. * in a circular file.
  606. */
  607. if (first_half_cycle == last_half_cycle) {
  608. /*
  609. * In this case we believe that the entire log should have
  610. * cycle number last_half_cycle. We need to scan backwards
  611. * from the end verifying that there are no holes still
  612. * containing last_half_cycle - 1. If we find such a hole,
  613. * then the start of that hole will be the new head. The
  614. * simple case looks like
  615. * x | x ... | x - 1 | x
  616. * Another case that fits this picture would be
  617. * x | x + 1 | x ... | x
  618. * In this case the head really is somewhere at the end of the
  619. * log, as one of the latest writes at the beginning was
  620. * incomplete.
  621. * One more case is
  622. * x | x + 1 | x ... | x - 1 | x
  623. * This is really the combination of the above two cases, and
  624. * the head has to end up at the start of the x-1 hole at the
  625. * end of the log.
  626. *
  627. * In the 256k log case, we will read from the beginning to the
  628. * end of the log and search for cycle numbers equal to x-1.
  629. * We don't worry about the x+1 blocks that we encounter,
  630. * because we know that they cannot be the head since the log
  631. * started with x.
  632. */
  633. head_blk = log_bbnum;
  634. stop_on_cycle = last_half_cycle - 1;
  635. } else {
  636. /*
  637. * In this case we want to find the first block with cycle
  638. * number matching last_half_cycle. We expect the log to be
  639. * some variation on
  640. * x + 1 ... | x ... | x
  641. * The first block with cycle number x (last_half_cycle) will
  642. * be where the new head belongs. First we do a binary search
  643. * for the first occurrence of last_half_cycle. The binary
  644. * search may not be totally accurate, so then we scan back
  645. * from there looking for occurrences of last_half_cycle before
  646. * us. If that backwards scan wraps around the beginning of
  647. * the log, then we look for occurrences of last_half_cycle - 1
  648. * at the end of the log. The cases we're looking for look
  649. * like
  650. * v binary search stopped here
  651. * x + 1 ... | x | x + 1 | x ... | x
  652. * ^ but we want to locate this spot
  653. * or
  654. * <---------> less than scan distance
  655. * x + 1 ... | x ... | x - 1 | x
  656. * ^ we want to locate this spot
  657. */
  658. stop_on_cycle = last_half_cycle;
  659. if ((error = xlog_find_cycle_start(log, bp, first_blk,
  660. &head_blk, last_half_cycle)))
  661. goto bp_err;
  662. }
  663. /*
  664. * Now validate the answer. Scan back some number of maximum possible
  665. * blocks and make sure each one has the expected cycle number. The
  666. * maximum is determined by the total possible amount of buffering
  667. * in the in-core log. The following number can be made tighter if
  668. * we actually look at the block size of the filesystem.
  669. */
  670. num_scan_bblks = XLOG_TOTAL_REC_SHIFT(log);
  671. if (head_blk >= num_scan_bblks) {
  672. /*
  673. * We are guaranteed that the entire check can be performed
  674. * in one buffer.
  675. */
  676. start_blk = head_blk - num_scan_bblks;
  677. if ((error = xlog_find_verify_cycle(log,
  678. start_blk, num_scan_bblks,
  679. stop_on_cycle, &new_blk)))
  680. goto bp_err;
  681. if (new_blk != -1)
  682. head_blk = new_blk;
  683. } else { /* need to read 2 parts of log */
  684. /*
  685. * We are going to scan backwards in the log in two parts.
  686. * First we scan the physical end of the log. In this part
  687. * of the log, we are looking for blocks with cycle number
  688. * last_half_cycle - 1.
  689. * If we find one, then we know that the log starts there, as
  690. * we've found a hole that didn't get written in going around
  691. * the end of the physical log. The simple case for this is
  692. * x + 1 ... | x ... | x - 1 | x
  693. * <---------> less than scan distance
  694. * If all of the blocks at the end of the log have cycle number
  695. * last_half_cycle, then we check the blocks at the start of
  696. * the log looking for occurrences of last_half_cycle. If we
  697. * find one, then our current estimate for the location of the
  698. * first occurrence of last_half_cycle is wrong and we move
  699. * back to the hole we've found. This case looks like
  700. * x + 1 ... | x | x + 1 | x ...
  701. * ^ binary search stopped here
  702. * Another case we need to handle that only occurs in 256k
  703. * logs is
  704. * x + 1 ... | x ... | x+1 | x ...
  705. * ^ binary search stops here
  706. * In a 256k log, the scan at the end of the log will see the
  707. * x + 1 blocks. We need to skip past those since that is
  708. * certainly not the head of the log. By searching for
  709. * last_half_cycle-1 we accomplish that.
  710. */
  711. ASSERT(head_blk <= INT_MAX &&
  712. (xfs_daddr_t) num_scan_bblks >= head_blk);
  713. start_blk = log_bbnum - (num_scan_bblks - head_blk);
  714. if ((error = xlog_find_verify_cycle(log, start_blk,
  715. num_scan_bblks - (int)head_blk,
  716. (stop_on_cycle - 1), &new_blk)))
  717. goto bp_err;
  718. if (new_blk != -1) {
  719. head_blk = new_blk;
  720. goto validate_head;
  721. }
  722. /*
  723. * Scan beginning of log now. The last part of the physical
  724. * log is good. This scan needs to verify that it doesn't find
  725. * the last_half_cycle.
  726. */
  727. start_blk = 0;
  728. ASSERT(head_blk <= INT_MAX);
  729. if ((error = xlog_find_verify_cycle(log,
  730. start_blk, (int)head_blk,
  731. stop_on_cycle, &new_blk)))
  732. goto bp_err;
  733. if (new_blk != -1)
  734. head_blk = new_blk;
  735. }
  736. validate_head:
  737. /*
  738. * Now we need to make sure head_blk is not pointing to a block in
  739. * the middle of a log record.
  740. */
  741. num_scan_bblks = XLOG_REC_SHIFT(log);
  742. if (head_blk >= num_scan_bblks) {
  743. start_blk = head_blk - num_scan_bblks; /* don't read head_blk */
  744. /* start ptr at last block ptr before head_blk */
  745. if ((error = xlog_find_verify_log_record(log, start_blk,
  746. &head_blk, 0)) == -1) {
  747. error = XFS_ERROR(EIO);
  748. goto bp_err;
  749. } else if (error)
  750. goto bp_err;
  751. } else {
  752. start_blk = 0;
  753. ASSERT(head_blk <= INT_MAX);
  754. if ((error = xlog_find_verify_log_record(log, start_blk,
  755. &head_blk, 0)) == -1) {
  756. /* We hit the beginning of the log during our search */
  757. start_blk = log_bbnum - (num_scan_bblks - head_blk);
  758. new_blk = log_bbnum;
  759. ASSERT(start_blk <= INT_MAX &&
  760. (xfs_daddr_t) log_bbnum-start_blk >= 0);
  761. ASSERT(head_blk <= INT_MAX);
  762. if ((error = xlog_find_verify_log_record(log,
  763. start_blk, &new_blk,
  764. (int)head_blk)) == -1) {
  765. error = XFS_ERROR(EIO);
  766. goto bp_err;
  767. } else if (error)
  768. goto bp_err;
  769. if (new_blk != log_bbnum)
  770. head_blk = new_blk;
  771. } else if (error)
  772. goto bp_err;
  773. }
  774. xlog_put_bp(bp);
  775. if (head_blk == log_bbnum)
  776. *return_head_blk = 0;
  777. else
  778. *return_head_blk = head_blk;
  779. /*
  780. * When returning here, we have a good block number. Bad block
  781. * means that during a previous crash, we didn't have a clean break
  782. * from cycle number N to cycle number N-1. In this case, we need
  783. * to find the first block with cycle number N-1.
  784. */
  785. return 0;
  786. bp_err:
  787. xlog_put_bp(bp);
  788. if (error)
  789. xfs_warn(log->l_mp, "failed to find log head");
  790. return error;
  791. }
  792. /*
  793. * Find the sync block number or the tail of the log.
  794. *
  795. * This will be the block number of the last record to have its
  796. * associated buffers synced to disk. Every log record header has
  797. * a sync lsn embedded in it. LSNs hold block numbers, so it is easy
  798. * to get a sync block number. The only concern is to figure out which
  799. * log record header to believe.
  800. *
  801. * The following algorithm uses the log record header with the largest
  802. * lsn. The entire log record does not need to be valid. We only care
  803. * that the header is valid.
  804. *
  805. * We could speed up search by using current head_blk buffer, but it is not
  806. * available.
  807. */
  808. STATIC int
  809. xlog_find_tail(
  810. struct xlog *log,
  811. xfs_daddr_t *head_blk,
  812. xfs_daddr_t *tail_blk)
  813. {
  814. xlog_rec_header_t *rhead;
  815. xlog_op_header_t *op_head;
  816. xfs_caddr_t offset = NULL;
  817. xfs_buf_t *bp;
  818. int error, i, found;
  819. xfs_daddr_t umount_data_blk;
  820. xfs_daddr_t after_umount_blk;
  821. xfs_lsn_t tail_lsn;
  822. int hblks;
  823. found = 0;
  824. /*
  825. * Find previous log record
  826. */
  827. if ((error = xlog_find_head(log, head_blk)))
  828. return error;
  829. bp = xlog_get_bp(log, 1);
  830. if (!bp)
  831. return ENOMEM;
  832. if (*head_blk == 0) { /* special case */
  833. error = xlog_bread(log, 0, 1, bp, &offset);
  834. if (error)
  835. goto done;
  836. if (xlog_get_cycle(offset) == 0) {
  837. *tail_blk = 0;
  838. /* leave all other log inited values alone */
  839. goto done;
  840. }
  841. }
  842. /*
  843. * Search backwards looking for log record header block
  844. */
  845. ASSERT(*head_blk < INT_MAX);
  846. for (i = (int)(*head_blk) - 1; i >= 0; i--) {
  847. error = xlog_bread(log, i, 1, bp, &offset);
  848. if (error)
  849. goto done;
  850. if (*(__be32 *)offset == cpu_to_be32(XLOG_HEADER_MAGIC_NUM)) {
  851. found = 1;
  852. break;
  853. }
  854. }
  855. /*
  856. * If we haven't found the log record header block, start looking
  857. * again from the end of the physical log. XXXmiken: There should be
  858. * a check here to make sure we didn't search more than N blocks in
  859. * the previous code.
  860. */
  861. if (!found) {
  862. for (i = log->l_logBBsize - 1; i >= (int)(*head_blk); i--) {
  863. error = xlog_bread(log, i, 1, bp, &offset);
  864. if (error)
  865. goto done;
  866. if (*(__be32 *)offset ==
  867. cpu_to_be32(XLOG_HEADER_MAGIC_NUM)) {
  868. found = 2;
  869. break;
  870. }
  871. }
  872. }
  873. if (!found) {
  874. xfs_warn(log->l_mp, "%s: couldn't find sync record", __func__);
  875. xlog_put_bp(bp);
  876. ASSERT(0);
  877. return XFS_ERROR(EIO);
  878. }
  879. /* find blk_no of tail of log */
  880. rhead = (xlog_rec_header_t *)offset;
  881. *tail_blk = BLOCK_LSN(be64_to_cpu(rhead->h_tail_lsn));
  882. /*
  883. * Reset log values according to the state of the log when we
  884. * crashed. In the case where head_blk == 0, we bump curr_cycle
  885. * one because the next write starts a new cycle rather than
  886. * continuing the cycle of the last good log record. At this
  887. * point we have guaranteed that all partial log records have been
  888. * accounted for. Therefore, we know that the last good log record
  889. * written was complete and ended exactly on the end boundary
  890. * of the physical log.
  891. */
  892. log->l_prev_block = i;
  893. log->l_curr_block = (int)*head_blk;
  894. log->l_curr_cycle = be32_to_cpu(rhead->h_cycle);
  895. if (found == 2)
  896. log->l_curr_cycle++;
  897. atomic64_set(&log->l_tail_lsn, be64_to_cpu(rhead->h_tail_lsn));
  898. atomic64_set(&log->l_last_sync_lsn, be64_to_cpu(rhead->h_lsn));
  899. xlog_assign_grant_head(&log->l_reserve_head.grant, log->l_curr_cycle,
  900. BBTOB(log->l_curr_block));
  901. xlog_assign_grant_head(&log->l_write_head.grant, log->l_curr_cycle,
  902. BBTOB(log->l_curr_block));
  903. /*
  904. * Look for unmount record. If we find it, then we know there
  905. * was a clean unmount. Since 'i' could be the last block in
  906. * the physical log, we convert to a log block before comparing
  907. * to the head_blk.
  908. *
  909. * Save the current tail lsn to use to pass to
  910. * xlog_clear_stale_blocks() below. We won't want to clear the
  911. * unmount record if there is one, so we pass the lsn of the
  912. * unmount record rather than the block after it.
  913. */
  914. if (xfs_sb_version_haslogv2(&log->l_mp->m_sb)) {
  915. int h_size = be32_to_cpu(rhead->h_size);
  916. int h_version = be32_to_cpu(rhead->h_version);
  917. if ((h_version & XLOG_VERSION_2) &&
  918. (h_size > XLOG_HEADER_CYCLE_SIZE)) {
  919. hblks = h_size / XLOG_HEADER_CYCLE_SIZE;
  920. if (h_size % XLOG_HEADER_CYCLE_SIZE)
  921. hblks++;
  922. } else {
  923. hblks = 1;
  924. }
  925. } else {
  926. hblks = 1;
  927. }
  928. after_umount_blk = (i + hblks + (int)
  929. BTOBB(be32_to_cpu(rhead->h_len))) % log->l_logBBsize;
  930. tail_lsn = atomic64_read(&log->l_tail_lsn);
  931. if (*head_blk == after_umount_blk &&
  932. be32_to_cpu(rhead->h_num_logops) == 1) {
  933. umount_data_blk = (i + hblks) % log->l_logBBsize;
  934. error = xlog_bread(log, umount_data_blk, 1, bp, &offset);
  935. if (error)
  936. goto done;
  937. op_head = (xlog_op_header_t *)offset;
  938. if (op_head->oh_flags & XLOG_UNMOUNT_TRANS) {
  939. /*
  940. * Set tail and last sync so that newly written
  941. * log records will point recovery to after the
  942. * current unmount record.
  943. */
  944. xlog_assign_atomic_lsn(&log->l_tail_lsn,
  945. log->l_curr_cycle, after_umount_blk);
  946. xlog_assign_atomic_lsn(&log->l_last_sync_lsn,
  947. log->l_curr_cycle, after_umount_blk);
  948. *tail_blk = after_umount_blk;
  949. /*
  950. * Note that the unmount was clean. If the unmount
  951. * was not clean, we need to know this to rebuild the
  952. * superblock counters from the perag headers if we
  953. * have a filesystem using non-persistent counters.
  954. */
  955. log->l_mp->m_flags |= XFS_MOUNT_WAS_CLEAN;
  956. }
  957. }
  958. /*
  959. * Make sure that there are no blocks in front of the head
  960. * with the same cycle number as the head. This can happen
  961. * because we allow multiple outstanding log writes concurrently,
  962. * and the later writes might make it out before earlier ones.
  963. *
  964. * We use the lsn from before modifying it so that we'll never
  965. * overwrite the unmount record after a clean unmount.
  966. *
  967. * Do this only if we are going to recover the filesystem
  968. *
  969. * NOTE: This used to say "if (!readonly)"
  970. * However on Linux, we can & do recover a read-only filesystem.
  971. * We only skip recovery if NORECOVERY is specified on mount,
  972. * in which case we would not be here.
  973. *
  974. * But... if the -device- itself is readonly, just skip this.
  975. * We can't recover this device anyway, so it won't matter.
  976. */
  977. if (!xfs_readonly_buftarg(log->l_mp->m_logdev_targp))
  978. error = xlog_clear_stale_blocks(log, tail_lsn);
  979. done:
  980. xlog_put_bp(bp);
  981. if (error)
  982. xfs_warn(log->l_mp, "failed to locate log tail");
  983. return error;
  984. }
  985. /*
  986. * Is the log zeroed at all?
  987. *
  988. * The last binary search should be changed to perform an X block read
  989. * once X becomes small enough. You can then search linearly through
  990. * the X blocks. This will cut down on the number of reads we need to do.
  991. *
  992. * If the log is partially zeroed, this routine will pass back the blkno
  993. * of the first block with cycle number 0. It won't have a complete LR
  994. * preceding it.
  995. *
  996. * Return:
  997. * 0 => the log is completely written to
  998. * -1 => use *blk_no as the first block of the log
  999. * >0 => error has occurred
  1000. */
  1001. STATIC int
  1002. xlog_find_zeroed(
  1003. struct xlog *log,
  1004. xfs_daddr_t *blk_no)
  1005. {
  1006. xfs_buf_t *bp;
  1007. xfs_caddr_t offset;
  1008. uint first_cycle, last_cycle;
  1009. xfs_daddr_t new_blk, last_blk, start_blk;
  1010. xfs_daddr_t num_scan_bblks;
  1011. int error, log_bbnum = log->l_logBBsize;
  1012. *blk_no = 0;
  1013. /* check totally zeroed log */
  1014. bp = xlog_get_bp(log, 1);
  1015. if (!bp)
  1016. return ENOMEM;
  1017. error = xlog_bread(log, 0, 1, bp, &offset);
  1018. if (error)
  1019. goto bp_err;
  1020. first_cycle = xlog_get_cycle(offset);
  1021. if (first_cycle == 0) { /* completely zeroed log */
  1022. *blk_no = 0;
  1023. xlog_put_bp(bp);
  1024. return -1;
  1025. }
  1026. /* check partially zeroed log */
  1027. error = xlog_bread(log, log_bbnum-1, 1, bp, &offset);
  1028. if (error)
  1029. goto bp_err;
  1030. last_cycle = xlog_get_cycle(offset);
  1031. if (last_cycle != 0) { /* log completely written to */
  1032. xlog_put_bp(bp);
  1033. return 0;
  1034. } else if (first_cycle != 1) {
  1035. /*
  1036. * If the cycle of the last block is zero, the cycle of
  1037. * the first block must be 1. If it's not, maybe we're
  1038. * not looking at a log... Bail out.
  1039. */
  1040. xfs_warn(log->l_mp,
  1041. "Log inconsistent or not a log (last==0, first!=1)");
  1042. error = XFS_ERROR(EINVAL);
  1043. goto bp_err;
  1044. }
  1045. /* we have a partially zeroed log */
  1046. last_blk = log_bbnum-1;
  1047. if ((error = xlog_find_cycle_start(log, bp, 0, &last_blk, 0)))
  1048. goto bp_err;
  1049. /*
  1050. * Validate the answer. Because there is no way to guarantee that
  1051. * the entire log is made up of log records which are the same size,
  1052. * we scan over the defined maximum blocks. At this point, the maximum
  1053. * is not chosen to mean anything special. XXXmiken
  1054. */
  1055. num_scan_bblks = XLOG_TOTAL_REC_SHIFT(log);
  1056. ASSERT(num_scan_bblks <= INT_MAX);
  1057. if (last_blk < num_scan_bblks)
  1058. num_scan_bblks = last_blk;
  1059. start_blk = last_blk - num_scan_bblks;
  1060. /*
  1061. * We search for any instances of cycle number 0 that occur before
  1062. * our current estimate of the head. What we're trying to detect is
  1063. * 1 ... | 0 | 1 | 0...
  1064. * ^ binary search ends here
  1065. */
  1066. if ((error = xlog_find_verify_cycle(log, start_blk,
  1067. (int)num_scan_bblks, 0, &new_blk)))
  1068. goto bp_err;
  1069. if (new_blk != -1)
  1070. last_blk = new_blk;
  1071. /*
  1072. * Potentially backup over partial log record write. We don't need
  1073. * to search the end of the log because we know it is zero.
  1074. */
  1075. if ((error = xlog_find_verify_log_record(log, start_blk,
  1076. &last_blk, 0)) == -1) {
  1077. error = XFS_ERROR(EIO);
  1078. goto bp_err;
  1079. } else if (error)
  1080. goto bp_err;
  1081. *blk_no = last_blk;
  1082. bp_err:
  1083. xlog_put_bp(bp);
  1084. if (error)
  1085. return error;
  1086. return -1;
  1087. }
  1088. /*
  1089. * These are simple subroutines used by xlog_clear_stale_blocks() below
  1090. * to initialize a buffer full of empty log record headers and write
  1091. * them into the log.
  1092. */
  1093. STATIC void
  1094. xlog_add_record(
  1095. struct xlog *log,
  1096. xfs_caddr_t buf,
  1097. int cycle,
  1098. int block,
  1099. int tail_cycle,
  1100. int tail_block)
  1101. {
  1102. xlog_rec_header_t *recp = (xlog_rec_header_t *)buf;
  1103. memset(buf, 0, BBSIZE);
  1104. recp->h_magicno = cpu_to_be32(XLOG_HEADER_MAGIC_NUM);
  1105. recp->h_cycle = cpu_to_be32(cycle);
  1106. recp->h_version = cpu_to_be32(
  1107. xfs_sb_version_haslogv2(&log->l_mp->m_sb) ? 2 : 1);
  1108. recp->h_lsn = cpu_to_be64(xlog_assign_lsn(cycle, block));
  1109. recp->h_tail_lsn = cpu_to_be64(xlog_assign_lsn(tail_cycle, tail_block));
  1110. recp->h_fmt = cpu_to_be32(XLOG_FMT);
  1111. memcpy(&recp->h_fs_uuid, &log->l_mp->m_sb.sb_uuid, sizeof(uuid_t));
  1112. }
  1113. STATIC int
  1114. xlog_write_log_records(
  1115. struct xlog *log,
  1116. int cycle,
  1117. int start_block,
  1118. int blocks,
  1119. int tail_cycle,
  1120. int tail_block)
  1121. {
  1122. xfs_caddr_t offset;
  1123. xfs_buf_t *bp;
  1124. int balign, ealign;
  1125. int sectbb = log->l_sectBBsize;
  1126. int end_block = start_block + blocks;
  1127. int bufblks;
  1128. int error = 0;
  1129. int i, j = 0;
  1130. /*
  1131. * Greedily allocate a buffer big enough to handle the full
  1132. * range of basic blocks to be written. If that fails, try
  1133. * a smaller size. We need to be able to write at least a
  1134. * log sector, or we're out of luck.
  1135. */
  1136. bufblks = 1 << ffs(blocks);
  1137. while (bufblks > log->l_logBBsize)
  1138. bufblks >>= 1;
  1139. while (!(bp = xlog_get_bp(log, bufblks))) {
  1140. bufblks >>= 1;
  1141. if (bufblks < sectbb)
  1142. return ENOMEM;
  1143. }
  1144. /* We may need to do a read at the start to fill in part of
  1145. * the buffer in the starting sector not covered by the first
  1146. * write below.
  1147. */
  1148. balign = round_down(start_block, sectbb);
  1149. if (balign != start_block) {
  1150. error = xlog_bread_noalign(log, start_block, 1, bp);
  1151. if (error)
  1152. goto out_put_bp;
  1153. j = start_block - balign;
  1154. }
  1155. for (i = start_block; i < end_block; i += bufblks) {
  1156. int bcount, endcount;
  1157. bcount = min(bufblks, end_block - start_block);
  1158. endcount = bcount - j;
  1159. /* We may need to do a read at the end to fill in part of
  1160. * the buffer in the final sector not covered by the write.
  1161. * If this is the same sector as the above read, skip it.
  1162. */
  1163. ealign = round_down(end_block, sectbb);
  1164. if (j == 0 && (start_block + endcount > ealign)) {
  1165. offset = bp->b_addr + BBTOB(ealign - start_block);
  1166. error = xlog_bread_offset(log, ealign, sectbb,
  1167. bp, offset);
  1168. if (error)
  1169. break;
  1170. }
  1171. offset = xlog_align(log, start_block, endcount, bp);
  1172. for (; j < endcount; j++) {
  1173. xlog_add_record(log, offset, cycle, i+j,
  1174. tail_cycle, tail_block);
  1175. offset += BBSIZE;
  1176. }
  1177. error = xlog_bwrite(log, start_block, endcount, bp);
  1178. if (error)
  1179. break;
  1180. start_block += endcount;
  1181. j = 0;
  1182. }
  1183. out_put_bp:
  1184. xlog_put_bp(bp);
  1185. return error;
  1186. }
  1187. /*
  1188. * This routine is called to blow away any incomplete log writes out
  1189. * in front of the log head. We do this so that we won't become confused
  1190. * if we come up, write only a little bit more, and then crash again.
  1191. * If we leave the partial log records out there, this situation could
  1192. * cause us to think those partial writes are valid blocks since they
  1193. * have the current cycle number. We get rid of them by overwriting them
  1194. * with empty log records with the old cycle number rather than the
  1195. * current one.
  1196. *
  1197. * The tail lsn is passed in rather than taken from
  1198. * the log so that we will not write over the unmount record after a
  1199. * clean unmount in a 512 block log. Doing so would leave the log without
  1200. * any valid log records in it until a new one was written. If we crashed
  1201. * during that time we would not be able to recover.
  1202. */
  1203. STATIC int
  1204. xlog_clear_stale_blocks(
  1205. struct xlog *log,
  1206. xfs_lsn_t tail_lsn)
  1207. {
  1208. int tail_cycle, head_cycle;
  1209. int tail_block, head_block;
  1210. int tail_distance, max_distance;
  1211. int distance;
  1212. int error;
  1213. tail_cycle = CYCLE_LSN(tail_lsn);
  1214. tail_block = BLOCK_LSN(tail_lsn);
  1215. head_cycle = log->l_curr_cycle;
  1216. head_block = log->l_curr_block;
  1217. /*
  1218. * Figure out the distance between the new head of the log
  1219. * and the tail. We want to write over any blocks beyond the
  1220. * head that we may have written just before the crash, but
  1221. * we don't want to overwrite the tail of the log.
  1222. */
  1223. if (head_cycle == tail_cycle) {
  1224. /*
  1225. * The tail is behind the head in the physical log,
  1226. * so the distance from the head to the tail is the
  1227. * distance from the head to the end of the log plus
  1228. * the distance from the beginning of the log to the
  1229. * tail.
  1230. */
  1231. if (unlikely(head_block < tail_block || head_block >= log->l_logBBsize)) {
  1232. XFS_ERROR_REPORT("xlog_clear_stale_blocks(1)",
  1233. XFS_ERRLEVEL_LOW, log->l_mp);
  1234. return XFS_ERROR(EFSCORRUPTED);
  1235. }
  1236. tail_distance = tail_block + (log->l_logBBsize - head_block);
  1237. } else {
  1238. /*
  1239. * The head is behind the tail in the physical log,
  1240. * so the distance from the head to the tail is just
  1241. * the tail block minus the head block.
  1242. */
  1243. if (unlikely(head_block >= tail_block || head_cycle != (tail_cycle + 1))){
  1244. XFS_ERROR_REPORT("xlog_clear_stale_blocks(2)",
  1245. XFS_ERRLEVEL_LOW, log->l_mp);
  1246. return XFS_ERROR(EFSCORRUPTED);
  1247. }
  1248. tail_distance = tail_block - head_block;
  1249. }
  1250. /*
  1251. * If the head is right up against the tail, we can't clear
  1252. * anything.
  1253. */
  1254. if (tail_distance <= 0) {
  1255. ASSERT(tail_distance == 0);
  1256. return 0;
  1257. }
  1258. max_distance = XLOG_TOTAL_REC_SHIFT(log);
  1259. /*
  1260. * Take the smaller of the maximum amount of outstanding I/O
  1261. * we could have and the distance to the tail to clear out.
  1262. * We take the smaller so that we don't overwrite the tail and
  1263. * we don't waste all day writing from the head to the tail
  1264. * for no reason.
  1265. */
  1266. max_distance = MIN(max_distance, tail_distance);
  1267. if ((head_block + max_distance) <= log->l_logBBsize) {
  1268. /*
  1269. * We can stomp all the blocks we need to without
  1270. * wrapping around the end of the log. Just do it
  1271. * in a single write. Use the cycle number of the
  1272. * current cycle minus one so that the log will look like:
  1273. * n ... | n - 1 ...
  1274. */
  1275. error = xlog_write_log_records(log, (head_cycle - 1),
  1276. head_block, max_distance, tail_cycle,
  1277. tail_block);
  1278. if (error)
  1279. return error;
  1280. } else {
  1281. /*
  1282. * We need to wrap around the end of the physical log in
  1283. * order to clear all the blocks. Do it in two separate
  1284. * I/Os. The first write should be from the head to the
  1285. * end of the physical log, and it should use the current
  1286. * cycle number minus one just like above.
  1287. */
  1288. distance = log->l_logBBsize - head_block;
  1289. error = xlog_write_log_records(log, (head_cycle - 1),
  1290. head_block, distance, tail_cycle,
  1291. tail_block);
  1292. if (error)
  1293. return error;
  1294. /*
  1295. * Now write the blocks at the start of the physical log.
  1296. * This writes the remainder of the blocks we want to clear.
  1297. * It uses the current cycle number since we're now on the
  1298. * same cycle as the head so that we get:
  1299. * n ... n ... | n - 1 ...
  1300. * ^^^^^ blocks we're writing
  1301. */
  1302. distance = max_distance - (log->l_logBBsize - head_block);
  1303. error = xlog_write_log_records(log, head_cycle, 0, distance,
  1304. tail_cycle, tail_block);
  1305. if (error)
  1306. return error;
  1307. }
  1308. return 0;
  1309. }
  1310. /******************************************************************************
  1311. *
  1312. * Log recover routines
  1313. *
  1314. ******************************************************************************
  1315. */
  1316. STATIC xlog_recover_t *
  1317. xlog_recover_find_tid(
  1318. struct hlist_head *head,
  1319. xlog_tid_t tid)
  1320. {
  1321. xlog_recover_t *trans;
  1322. hlist_for_each_entry(trans, head, r_list) {
  1323. if (trans->r_log_tid == tid)
  1324. return trans;
  1325. }
  1326. return NULL;
  1327. }
  1328. STATIC void
  1329. xlog_recover_new_tid(
  1330. struct hlist_head *head,
  1331. xlog_tid_t tid,
  1332. xfs_lsn_t lsn)
  1333. {
  1334. xlog_recover_t *trans;
  1335. trans = kmem_zalloc(sizeof(xlog_recover_t), KM_SLEEP);
  1336. trans->r_log_tid = tid;
  1337. trans->r_lsn = lsn;
  1338. INIT_LIST_HEAD(&trans->r_itemq);
  1339. INIT_HLIST_NODE(&trans->r_list);
  1340. hlist_add_head(&trans->r_list, head);
  1341. }
  1342. STATIC void
  1343. xlog_recover_add_item(
  1344. struct list_head *head)
  1345. {
  1346. xlog_recover_item_t *item;
  1347. item = kmem_zalloc(sizeof(xlog_recover_item_t), KM_SLEEP);
  1348. INIT_LIST_HEAD(&item->ri_list);
  1349. list_add_tail(&item->ri_list, head);
  1350. }
  1351. STATIC int
  1352. xlog_recover_add_to_cont_trans(
  1353. struct xlog *log,
  1354. struct xlog_recover *trans,
  1355. xfs_caddr_t dp,
  1356. int len)
  1357. {
  1358. xlog_recover_item_t *item;
  1359. xfs_caddr_t ptr, old_ptr;
  1360. int old_len;
  1361. if (list_empty(&trans->r_itemq)) {
  1362. /* finish copying rest of trans header */
  1363. xlog_recover_add_item(&trans->r_itemq);
  1364. ptr = (xfs_caddr_t) &trans->r_theader +
  1365. sizeof(xfs_trans_header_t) - len;
  1366. memcpy(ptr, dp, len); /* d, s, l */
  1367. return 0;
  1368. }
  1369. /* take the tail entry */
  1370. item = list_entry(trans->r_itemq.prev, xlog_recover_item_t, ri_list);
  1371. old_ptr = item->ri_buf[item->ri_cnt-1].i_addr;
  1372. old_len = item->ri_buf[item->ri_cnt-1].i_len;
  1373. ptr = kmem_realloc(old_ptr, len+old_len, old_len, KM_SLEEP);
  1374. memcpy(&ptr[old_len], dp, len); /* d, s, l */
  1375. item->ri_buf[item->ri_cnt-1].i_len += len;
  1376. item->ri_buf[item->ri_cnt-1].i_addr = ptr;
  1377. trace_xfs_log_recover_item_add_cont(log, trans, item, 0);
  1378. return 0;
  1379. }
  1380. /*
  1381. * The next region to add is the start of a new region. It could be
  1382. * a whole region or it could be the first part of a new region. Because
  1383. * of this, the assumption here is that the type and size fields of all
  1384. * format structures fit into the first 32 bits of the structure.
  1385. *
  1386. * This works because all regions must be 32 bit aligned. Therefore, we
  1387. * either have both fields or we have neither field. In the case we have
  1388. * neither field, the data part of the region is zero length. We only have
  1389. * a log_op_header and can throw away the header since a new one will appear
  1390. * later. If we have at least 4 bytes, then we can determine how many regions
  1391. * will appear in the current log item.
  1392. */
  1393. STATIC int
  1394. xlog_recover_add_to_trans(
  1395. struct xlog *log,
  1396. struct xlog_recover *trans,
  1397. xfs_caddr_t dp,
  1398. int len)
  1399. {
  1400. xfs_inode_log_format_t *in_f; /* any will do */
  1401. xlog_recover_item_t *item;
  1402. xfs_caddr_t ptr;
  1403. if (!len)
  1404. return 0;
  1405. if (list_empty(&trans->r_itemq)) {
  1406. /* we need to catch log corruptions here */
  1407. if (*(uint *)dp != XFS_TRANS_HEADER_MAGIC) {
  1408. xfs_warn(log->l_mp, "%s: bad header magic number",
  1409. __func__);
  1410. ASSERT(0);
  1411. return XFS_ERROR(EIO);
  1412. }
  1413. if (len == sizeof(xfs_trans_header_t))
  1414. xlog_recover_add_item(&trans->r_itemq);
  1415. memcpy(&trans->r_theader, dp, len); /* d, s, l */
  1416. return 0;
  1417. }
  1418. ptr = kmem_alloc(len, KM_SLEEP);
  1419. memcpy(ptr, dp, len);
  1420. in_f = (xfs_inode_log_format_t *)ptr;
  1421. /* take the tail entry */
  1422. item = list_entry(trans->r_itemq.prev, xlog_recover_item_t, ri_list);
  1423. if (item->ri_total != 0 &&
  1424. item->ri_total == item->ri_cnt) {
  1425. /* tail item is in use, get a new one */
  1426. xlog_recover_add_item(&trans->r_itemq);
  1427. item = list_entry(trans->r_itemq.prev,
  1428. xlog_recover_item_t, ri_list);
  1429. }
  1430. if (item->ri_total == 0) { /* first region to be added */
  1431. if (in_f->ilf_size == 0 ||
  1432. in_f->ilf_size > XLOG_MAX_REGIONS_IN_ITEM) {
  1433. xfs_warn(log->l_mp,
  1434. "bad number of regions (%d) in inode log format",
  1435. in_f->ilf_size);
  1436. ASSERT(0);
  1437. free(ptr);
  1438. return XFS_ERROR(EIO);
  1439. }
  1440. item->ri_total = in_f->ilf_size;
  1441. item->ri_buf =
  1442. kmem_zalloc(item->ri_total * sizeof(xfs_log_iovec_t),
  1443. KM_SLEEP);
  1444. }
  1445. ASSERT(item->ri_total > item->ri_cnt);
  1446. /* Description region is ri_buf[0] */
  1447. item->ri_buf[item->ri_cnt].i_addr = ptr;
  1448. item->ri_buf[item->ri_cnt].i_len = len;
  1449. item->ri_cnt++;
  1450. trace_xfs_log_recover_item_add(log, trans, item, 0);
  1451. return 0;
  1452. }
  1453. /*
  1454. * Sort the log items in the transaction.
  1455. *
  1456. * The ordering constraints are defined by the inode allocation and unlink
  1457. * behaviour. The rules are:
  1458. *
  1459. * 1. Every item is only logged once in a given transaction. Hence it
  1460. * represents the last logged state of the item. Hence ordering is
  1461. * dependent on the order in which operations need to be performed so
  1462. * required initial conditions are always met.
  1463. *
  1464. * 2. Cancelled buffers are recorded in pass 1 in a separate table and
  1465. * there's nothing to replay from them so we can simply cull them
  1466. * from the transaction. However, we can't do that until after we've
  1467. * replayed all the other items because they may be dependent on the
  1468. * cancelled buffer and replaying the cancelled buffer can remove it
  1469. * form the cancelled buffer table. Hence they have tobe done last.
  1470. *
  1471. * 3. Inode allocation buffers must be replayed before inode items that
  1472. * read the buffer and replay changes into it. For filesystems using the
  1473. * ICREATE transactions, this means XFS_LI_ICREATE objects need to get
  1474. * treated the same as inode allocation buffers as they create and
  1475. * initialise the buffers directly.
  1476. *
  1477. * 4. Inode unlink buffers must be replayed after inode items are replayed.
  1478. * This ensures that inodes are completely flushed to the inode buffer
  1479. * in a "free" state before we remove the unlinked inode list pointer.
  1480. *
  1481. * Hence the ordering needs to be inode allocation buffers first, inode items
  1482. * second, inode unlink buffers third and cancelled buffers last.
  1483. *
  1484. * But there's a problem with that - we can't tell an inode allocation buffer
  1485. * apart from a regular buffer, so we can't separate them. We can, however,
  1486. * tell an inode unlink buffer from the others, and so we can separate them out
  1487. * from all the other buffers and move them to last.
  1488. *
  1489. * Hence, 4 lists, in order from head to tail:
  1490. * - buffer_list for all buffers except cancelled/inode unlink buffers
  1491. * - item_list for all non-buffer items
  1492. * - inode_buffer_list for inode unlink buffers
  1493. * - cancel_list for the cancelled buffers
  1494. *
  1495. * Note that we add objects to the tail of the lists so that first-to-last
  1496. * ordering is preserved within the lists. Adding objects to the head of the
  1497. * list means when we traverse from the head we walk them in last-to-first
  1498. * order. For cancelled buffers and inode unlink buffers this doesn't matter,
  1499. * but for all other items there may be specific ordering that we need to
  1500. * preserve.
  1501. */
  1502. STATIC int
  1503. xlog_recover_reorder_trans(
  1504. struct xlog *log,
  1505. struct xlog_recover *trans,
  1506. int pass)
  1507. {
  1508. xlog_recover_item_t *item, *n;
  1509. LIST_HEAD(sort_list);
  1510. LIST_HEAD(cancel_list);
  1511. LIST_HEAD(buffer_list);
  1512. LIST_HEAD(inode_buffer_list);
  1513. LIST_HEAD(inode_list);
  1514. list_splice_init(&trans->r_itemq, &sort_list);
  1515. list_for_each_entry_safe(item, n, &sort_list, ri_list) {
  1516. xfs_buf_log_format_t *buf_f = item->ri_buf[0].i_addr;
  1517. switch (ITEM_TYPE(item)) {
  1518. case XFS_LI_ICREATE:
  1519. list_move_tail(&item->ri_list, &buffer_list);
  1520. break;
  1521. case XFS_LI_BUF:
  1522. if (buf_f->blf_flags & XFS_BLF_CANCEL) {
  1523. trace_xfs_log_recover_item_reorder_head(log,
  1524. trans, item, pass);
  1525. list_move(&item->ri_list, &cancel_list);
  1526. break;
  1527. }
  1528. if (buf_f->blf_flags & XFS_BLF_INODE_BUF) {
  1529. list_move(&item->ri_list, &inode_buffer_list);
  1530. break;
  1531. }
  1532. list_move_tail(&item->ri_list, &buffer_list);
  1533. break;
  1534. case XFS_LI_INODE:
  1535. case XFS_LI_DQUOT:
  1536. case XFS_LI_QUOTAOFF:
  1537. case XFS_LI_EFD:
  1538. case XFS_LI_EFI:
  1539. trace_xfs_log_recover_item_reorder_tail(log,
  1540. trans, item, pass);
  1541. list_move_tail(&item->ri_list, &inode_list);
  1542. break;
  1543. default:
  1544. xfs_warn(log->l_mp,
  1545. "%s: unrecognized type of log operation",
  1546. __func__);
  1547. ASSERT(0);
  1548. return XFS_ERROR(EIO);
  1549. }
  1550. }
  1551. ASSERT(list_empty(&sort_list));
  1552. if (!list_empty(&buffer_list))
  1553. list_splice(&buffer_list, &trans->r_itemq);
  1554. if (!list_empty(&inode_list))
  1555. list_splice_tail(&inode_list, &trans->r_itemq);
  1556. if (!list_empty(&inode_buffer_list))
  1557. list_splice_tail(&inode_buffer_list, &trans->r_itemq);
  1558. if (!list_empty(&cancel_list))
  1559. list_splice_tail(&cancel_list, &trans->r_itemq);
  1560. return 0;
  1561. }
  1562. /*
  1563. * Build up the table of buf cancel records so that we don't replay
  1564. * cancelled data in the second pass. For buffer records that are
  1565. * not cancel records, there is nothing to do here so we just return.
  1566. *
  1567. * If we get a cancel record which is already in the table, this indicates
  1568. * that the buffer was cancelled multiple times. In order to ensure
  1569. * that during pass 2 we keep the record in the table until we reach its
  1570. * last occurrence in the log, we keep a reference count in the cancel
  1571. * record in the table to tell us how many times we expect to see this
  1572. * record during the second pass.
  1573. */
  1574. STATIC int
  1575. xlog_recover_buffer_pass1(
  1576. struct xlog *log,
  1577. struct xlog_recover_item *item)
  1578. {
  1579. xfs_buf_log_format_t *buf_f = item->ri_buf[0].i_addr;
  1580. struct list_head *bucket;
  1581. struct xfs_buf_cancel *bcp;
  1582. /*
  1583. * If this isn't a cancel buffer item, then just return.
  1584. */
  1585. if (!(buf_f->blf_flags & XFS_BLF_CANCEL)) {
  1586. trace_xfs_log_recover_buf_not_cancel(log, buf_f);
  1587. return 0;
  1588. }
  1589. /*
  1590. * Insert an xfs_buf_cancel record into the hash table of them.
  1591. * If there is already an identical record, bump its reference count.
  1592. */
  1593. bucket = XLOG_BUF_CANCEL_BUCKET(log, buf_f->blf_blkno);
  1594. list_for_each_entry(bcp, bucket, bc_list) {
  1595. if (bcp->bc_blkno == buf_f->blf_blkno &&
  1596. bcp->bc_len == buf_f->blf_len) {
  1597. bcp->bc_refcount++;
  1598. trace_xfs_log_recover_buf_cancel_ref_inc(log, buf_f);
  1599. return 0;
  1600. }
  1601. }
  1602. bcp = kmem_alloc(sizeof(struct xfs_buf_cancel), KM_SLEEP);
  1603. bcp->bc_blkno = buf_f->blf_blkno;
  1604. bcp->bc_len = buf_f->blf_len;
  1605. bcp->bc_refcount = 1;
  1606. list_add_tail(&bcp->bc_list, bucket);
  1607. trace_xfs_log_recover_buf_cancel_add(log, buf_f);
  1608. return 0;
  1609. }
  1610. /*
  1611. * Check to see whether the buffer being recovered has a corresponding
  1612. * entry in the buffer cancel record table. If it is, return the cancel
  1613. * buffer structure to the caller.
  1614. */
  1615. STATIC struct xfs_buf_cancel *
  1616. xlog_peek_buffer_cancelled(
  1617. struct xlog *log,
  1618. xfs_daddr_t blkno,
  1619. uint len,
  1620. ushort flags)
  1621. {
  1622. struct list_head *bucket;
  1623. struct xfs_buf_cancel *bcp;
  1624. if (!log->l_buf_cancel_table) {
  1625. /* empty table means no cancelled buffers in the log */
  1626. ASSERT(!(flags & XFS_BLF_CANCEL));
  1627. return NULL;
  1628. }
  1629. bucket = XLOG_BUF_CANCEL_BUCKET(log, blkno);
  1630. list_for_each_entry(bcp, bucket, bc_list) {
  1631. if (bcp->bc_blkno == blkno && bcp->bc_len == len)
  1632. return bcp;
  1633. }
  1634. /*
  1635. * We didn't find a corresponding entry in the table, so return 0 so
  1636. * that the buffer is NOT cancelled.
  1637. */
  1638. ASSERT(!(flags & XFS_BLF_CANCEL));
  1639. return NULL;
  1640. }
  1641. /*
  1642. * If the buffer is being cancelled then return 1 so that it will be cancelled,
  1643. * otherwise return 0. If the buffer is actually a buffer cancel item
  1644. * (XFS_BLF_CANCEL is set), then decrement the refcount on the entry in the
  1645. * table and remove it from the table if this is the last reference.
  1646. *
  1647. * We remove the cancel record from the table when we encounter its last
  1648. * occurrence in the log so that if the same buffer is re-used again after its
  1649. * last cancellation we actually replay the changes made at that point.
  1650. */
  1651. STATIC int
  1652. xlog_check_buffer_cancelled(
  1653. struct xlog *log,
  1654. xfs_daddr_t blkno,
  1655. uint len,
  1656. ushort flags)
  1657. {
  1658. struct xfs_buf_cancel *bcp;
  1659. bcp = xlog_peek_buffer_cancelled(log, blkno, len, flags);
  1660. if (!bcp)
  1661. return 0;
  1662. /*
  1663. * We've go a match, so return 1 so that the recovery of this buffer
  1664. * is cancelled. If this buffer is actually a buffer cancel log
  1665. * item, then decrement the refcount on the one in the table and
  1666. * remove it if this is the last reference.
  1667. */
  1668. if (flags & XFS_BLF_CANCEL) {
  1669. if (--bcp->bc_refcount == 0) {
  1670. list_del(&bcp->bc_list);
  1671. kmem_free(bcp);
  1672. }
  1673. }
  1674. return 1;
  1675. }
  1676. /*
  1677. * Perform recovery for a buffer full of inodes. In these buffers, the only
  1678. * data which should be recovered is that which corresponds to the
  1679. * di_next_unlinked pointers in the on disk inode structures. The rest of the
  1680. * data for the inodes is always logged through the inodes themselves rather
  1681. * than the inode buffer and is recovered in xlog_recover_inode_pass2().
  1682. *
  1683. * The only time when buffers full of inodes are fully recovered is when the
  1684. * buffer is full of newly allocated inodes. In this case the buffer will
  1685. * not be marked as an inode buffer and so will be sent to
  1686. * xlog_recover_do_reg_buffer() below during recovery.
  1687. */
  1688. STATIC int
  1689. xlog_recover_do_inode_buffer(
  1690. struct xfs_mount *mp,
  1691. xlog_recover_item_t *item,
  1692. struct xfs_buf *bp,
  1693. xfs_buf_log_format_t *buf_f)
  1694. {
  1695. int i;
  1696. int item_index = 0;
  1697. int bit = 0;
  1698. int nbits = 0;
  1699. int reg_buf_offset = 0;
  1700. int reg_buf_bytes = 0;
  1701. int next_unlinked_offset;
  1702. int inodes_per_buf;
  1703. xfs_agino_t *logged_nextp;
  1704. xfs_agino_t *buffer_nextp;
  1705. trace_xfs_log_recover_buf_inode_buf(mp->m_log, buf_f);
  1706. /*
  1707. * Post recovery validation only works properly on CRC enabled
  1708. * filesystems.
  1709. */
  1710. if (xfs_sb_version_hascrc(&mp->m_sb))
  1711. bp->b_ops = &xfs_inode_buf_ops;
  1712. inodes_per_buf = BBTOB(bp->b_io_length) >> mp->m_sb.sb_inodelog;
  1713. for (i = 0; i < inodes_per_buf; i++) {
  1714. next_unlinked_offset = (i * mp->m_sb.sb_inodesize) +
  1715. offsetof(xfs_dinode_t, di_next_unlinked);
  1716. while (next_unlinked_offset >=
  1717. (reg_buf_offset + reg_buf_bytes)) {
  1718. /*
  1719. * The next di_next_unlinked field is beyond
  1720. * the current logged region. Find the next
  1721. * logged region that contains or is beyond
  1722. * the current di_next_unlinked field.
  1723. */
  1724. bit += nbits;
  1725. bit = xfs_next_bit(buf_f->blf_data_map,
  1726. buf_f->blf_map_size, bit);
  1727. /*
  1728. * If there are no more logged regions in the
  1729. * buffer, then we're done.
  1730. */
  1731. if (bit == -1)
  1732. return 0;
  1733. nbits = xfs_contig_bits(buf_f->blf_data_map,
  1734. buf_f->blf_map_size, bit);
  1735. ASSERT(nbits > 0);
  1736. reg_buf_offset = bit << XFS_BLF_SHIFT;
  1737. reg_buf_bytes = nbits << XFS_BLF_SHIFT;
  1738. item_index++;
  1739. }
  1740. /*
  1741. * If the current logged region starts after the current
  1742. * di_next_unlinked field, then move on to the next
  1743. * di_next_unlinked field.
  1744. */
  1745. if (next_unlinked_offset < reg_buf_offset)
  1746. continue;
  1747. ASSERT(item->ri_buf[item_index].i_addr != NULL);
  1748. ASSERT((item->ri_buf[item_index].i_len % XFS_BLF_CHUNK) == 0);
  1749. ASSERT((reg_buf_offset + reg_buf_bytes) <=
  1750. BBTOB(bp->b_io_length));
  1751. /*
  1752. * The current logged region contains a copy of the
  1753. * current di_next_unlinked field. Extract its value
  1754. * and copy it to the buffer copy.
  1755. */
  1756. logged_nextp = item->ri_buf[item_index].i_addr +
  1757. next_unlinked_offset - reg_buf_offset;
  1758. if (unlikely(*logged_nextp == 0)) {
  1759. xfs_alert(mp,
  1760. "Bad inode buffer log record (ptr = 0x%p, bp = 0x%p). "
  1761. "Trying to replay bad (0) inode di_next_unlinked field.",
  1762. item, bp);
  1763. XFS_ERROR_REPORT("xlog_recover_do_inode_buf",
  1764. XFS_ERRLEVEL_LOW, mp);
  1765. return XFS_ERROR(EFSCORRUPTED);
  1766. }
  1767. buffer_nextp = (xfs_agino_t *)xfs_buf_offset(bp,
  1768. next_unlinked_offset);
  1769. *buffer_nextp = *logged_nextp;
  1770. /*
  1771. * If necessary, recalculate the CRC in the on-disk inode. We
  1772. * have to leave the inode in a consistent state for whoever
  1773. * reads it next....
  1774. */
  1775. xfs_dinode_calc_crc(mp, (struct xfs_dinode *)
  1776. xfs_buf_offset(bp, i * mp->m_sb.sb_inodesize));
  1777. }
  1778. return 0;
  1779. }
  1780. /*
  1781. * V5 filesystems know the age of the buffer on disk being recovered. We can
  1782. * have newer objects on disk than we are replaying, and so for these cases we
  1783. * don't want to replay the current change as that will make the buffer contents
  1784. * temporarily invalid on disk.
  1785. *
  1786. * The magic number might not match the buffer type we are going to recover
  1787. * (e.g. reallocated blocks), so we ignore the xfs_buf_log_format flags. Hence
  1788. * extract the LSN of the existing object in the buffer based on it's current
  1789. * magic number. If we don't recognise the magic number in the buffer, then
  1790. * return a LSN of -1 so that the caller knows it was an unrecognised block and
  1791. * so can recover the buffer.
  1792. *
  1793. * Note: we cannot rely solely on magic number matches to determine that the
  1794. * buffer has a valid LSN - we also need to verify that it belongs to this
  1795. * filesystem, so we need to extract the object's LSN and compare it to that
  1796. * which we read from the superblock. If the UUIDs don't match, then we've got a
  1797. * stale metadata block from an old filesystem instance that we need to recover
  1798. * over the top of.
  1799. */
  1800. static xfs_lsn_t
  1801. xlog_recover_get_buf_lsn(
  1802. struct xfs_mount *mp,
  1803. struct xfs_buf *bp)
  1804. {
  1805. __uint32_t magic32;
  1806. __uint16_t magic16;
  1807. __uint16_t magicda;
  1808. void *blk = bp->b_addr;
  1809. uuid_t *uuid;
  1810. xfs_lsn_t lsn = -1;
  1811. /* v4 filesystems always recover immediately */
  1812. if (!xfs_sb_version_hascrc(&mp->m_sb))
  1813. goto recover_immediately;
  1814. magic32 = be32_to_cpu(*(__be32 *)blk);
  1815. switch (magic32) {
  1816. case XFS_ABTB_CRC_MAGIC:
  1817. case XFS_ABTC_CRC_MAGIC:
  1818. case XFS_ABTB_MAGIC:
  1819. case XFS_ABTC_MAGIC:
  1820. case XFS_IBT_CRC_MAGIC:
  1821. case XFS_IBT_MAGIC: {
  1822. struct xfs_btree_block *btb = blk;
  1823. lsn = be64_to_cpu(btb->bb_u.s.bb_lsn);
  1824. uuid = &btb->bb_u.s.bb_uuid;
  1825. break;
  1826. }
  1827. case XFS_BMAP_CRC_MAGIC:
  1828. case XFS_BMAP_MAGIC: {
  1829. struct xfs_btree_block *btb = blk;
  1830. lsn = be64_to_cpu(btb->bb_u.l.bb_lsn);
  1831. uuid = &btb->bb_u.l.bb_uuid;
  1832. break;
  1833. }
  1834. case XFS_AGF_MAGIC:
  1835. lsn = be64_to_cpu(((struct xfs_agf *)blk)->agf_lsn);
  1836. uuid = &((struct xfs_agf *)blk)->agf_uuid;
  1837. break;
  1838. case XFS_AGFL_MAGIC:
  1839. lsn = be64_to_cpu(((struct xfs_agfl *)blk)->agfl_lsn);
  1840. uuid = &((struct xfs_agfl *)blk)->agfl_uuid;
  1841. break;
  1842. case XFS_AGI_MAGIC:
  1843. lsn = be64_to_cpu(((struct xfs_agi *)blk)->agi_lsn);
  1844. uuid = &((struct xfs_agi *)blk)->agi_uuid;
  1845. break;
  1846. case XFS_SYMLINK_MAGIC:
  1847. lsn = be64_to_cpu(((struct xfs_dsymlink_hdr *)blk)->sl_lsn);
  1848. uuid = &((struct xfs_dsymlink_hdr *)blk)->sl_uuid;
  1849. break;
  1850. case XFS_DIR3_BLOCK_MAGIC:
  1851. case XFS_DIR3_DATA_MAGIC:
  1852. case XFS_DIR3_FREE_MAGIC:
  1853. lsn = be64_to_cpu(((struct xfs_dir3_blk_hdr *)blk)->lsn);
  1854. uuid = &((struct xfs_dir3_blk_hdr *)blk)->uuid;
  1855. break;
  1856. case XFS_ATTR3_RMT_MAGIC:
  1857. lsn = be64_to_cpu(((struct xfs_attr3_rmt_hdr *)blk)->rm_lsn);
  1858. uuid = &((struct xfs_attr3_rmt_hdr *)blk)->rm_uuid;
  1859. break;
  1860. case XFS_SB_MAGIC:
  1861. lsn = be64_to_cpu(((struct xfs_dsb *)blk)->sb_lsn);
  1862. uuid = &((struct xfs_dsb *)blk)->sb_uuid;
  1863. break;
  1864. default:
  1865. break;
  1866. }
  1867. if (lsn != (xfs_lsn_t)-1) {
  1868. if (!uuid_equal(&mp->m_sb.sb_uuid, uuid))
  1869. goto recover_immediately;
  1870. return lsn;
  1871. }
  1872. magicda = be16_to_cpu(((struct xfs_da_blkinfo *)blk)->magic);
  1873. switch (magicda) {
  1874. case XFS_DIR3_LEAF1_MAGIC:
  1875. case XFS_DIR3_LEAFN_MAGIC:
  1876. case XFS_DA3_NODE_MAGIC:
  1877. lsn = be64_to_cpu(((struct xfs_da3_blkinfo *)blk)->lsn);
  1878. uuid = &((struct xfs_da3_blkinfo *)blk)->uuid;
  1879. break;
  1880. default:
  1881. break;
  1882. }
  1883. if (lsn != (xfs_lsn_t)-1) {
  1884. if (!uuid_equal(&mp->m_sb.sb_uuid, uuid))
  1885. goto recover_immediately;
  1886. return lsn;
  1887. }
  1888. /*
  1889. * We do individual object checks on dquot and inode buffers as they
  1890. * have their own individual LSN records. Also, we could have a stale
  1891. * buffer here, so we have to at least recognise these buffer types.
  1892. *
  1893. * A notd complexity here is inode unlinked list processing - it logs
  1894. * the inode directly in the buffer, but we don't know which inodes have
  1895. * been modified, and there is no global buffer LSN. Hence we need to
  1896. * recover all inode buffer types immediately. This problem will be
  1897. * fixed by logical logging of the unlinked list modifications.
  1898. */
  1899. magic16 = be16_to_cpu(*(__be16 *)blk);
  1900. switch (magic16) {
  1901. case XFS_DQUOT_MAGIC:
  1902. case XFS_DINODE_MAGIC:
  1903. goto recover_immediately;
  1904. default:
  1905. break;
  1906. }
  1907. /* unknown buffer contents, recover immediately */
  1908. recover_immediately:
  1909. return (xfs_lsn_t)-1;
  1910. }
  1911. /*
  1912. * Validate the recovered buffer is of the correct type and attach the
  1913. * appropriate buffer operations to them for writeback. Magic numbers are in a
  1914. * few places:
  1915. * the first 16 bits of the buffer (inode buffer, dquot buffer),
  1916. * the first 32 bits of the buffer (most blocks),
  1917. * inside a struct xfs_da_blkinfo at the start of the buffer.
  1918. */
  1919. static void
  1920. xlog_recover_validate_buf_type(
  1921. struct xfs_mount *mp,
  1922. struct xfs_buf *bp,
  1923. xfs_buf_log_format_t *buf_f)
  1924. {
  1925. struct xfs_da_blkinfo *info = bp->b_addr;
  1926. __uint32_t magic32;
  1927. __uint16_t magic16;
  1928. __uint16_t magicda;
  1929. magic32 = be32_to_cpu(*(__be32 *)bp->b_addr);
  1930. magic16 = be16_to_cpu(*(__be16*)bp->b_addr);
  1931. magicda = be16_to_cpu(info->magic);
  1932. switch (xfs_blft_from_flags(buf_f)) {
  1933. case XFS_BLFT_BTREE_BUF:
  1934. switch (magic32) {
  1935. case XFS_ABTB_CRC_MAGIC:
  1936. case XFS_ABTC_CRC_MAGIC:
  1937. case XFS_ABTB_MAGIC:
  1938. case XFS_ABTC_MAGIC:
  1939. bp->b_ops = &xfs_allocbt_buf_ops;
  1940. break;
  1941. case XFS_IBT_CRC_MAGIC:
  1942. case XFS_IBT_MAGIC:
  1943. bp->b_ops = &xfs_inobt_buf_ops;
  1944. break;
  1945. case XFS_BMAP_CRC_MAGIC:
  1946. case XFS_BMAP_MAGIC:
  1947. bp->b_ops = &xfs_bmbt_buf_ops;
  1948. break;
  1949. default:
  1950. xfs_warn(mp, "Bad btree block magic!");
  1951. ASSERT(0);
  1952. break;
  1953. }
  1954. break;
  1955. case XFS_BLFT_AGF_BUF:
  1956. if (magic32 != XFS_AGF_MAGIC) {
  1957. xfs_warn(mp, "Bad AGF block magic!");
  1958. ASSERT(0);
  1959. break;
  1960. }
  1961. bp->b_ops = &xfs_agf_buf_ops;
  1962. break;
  1963. case XFS_BLFT_AGFL_BUF:
  1964. if (!xfs_sb_version_hascrc(&mp->m_sb))
  1965. break;
  1966. if (magic32 != XFS_AGFL_MAGIC) {
  1967. xfs_warn(mp, "Bad AGFL block magic!");
  1968. ASSERT(0);
  1969. break;
  1970. }
  1971. bp->b_ops = &xfs_agfl_buf_ops;
  1972. break;
  1973. case XFS_BLFT_AGI_BUF:
  1974. if (magic32 != XFS_AGI_MAGIC) {
  1975. xfs_warn(mp, "Bad AGI block magic!");
  1976. ASSERT(0);
  1977. break;
  1978. }
  1979. bp->b_ops = &xfs_agi_buf_ops;
  1980. break;
  1981. case XFS_BLFT_UDQUOT_BUF:
  1982. case XFS_BLFT_PDQUOT_BUF:
  1983. case XFS_BLFT_GDQUOT_BUF:
  1984. #ifdef CONFIG_XFS_QUOTA
  1985. if (magic16 != XFS_DQUOT_MAGIC) {
  1986. xfs_warn(mp, "Bad DQUOT block magic!");
  1987. ASSERT(0);
  1988. break;
  1989. }
  1990. bp->b_ops = &xfs_dquot_buf_ops;
  1991. #else
  1992. xfs_alert(mp,
  1993. "Trying to recover dquots without QUOTA support built in!");
  1994. ASSERT(0);
  1995. #endif
  1996. break;
  1997. case XFS_BLFT_DINO_BUF:
  1998. /*
  1999. * we get here with inode allocation buffers, not buffers that
  2000. * track unlinked list changes.
  2001. */
  2002. if (magic16 != XFS_DINODE_MAGIC) {
  2003. xfs_warn(mp, "Bad INODE block magic!");
  2004. ASSERT(0);
  2005. break;
  2006. }
  2007. bp->b_ops = &xfs_inode_buf_ops;
  2008. break;
  2009. case XFS_BLFT_SYMLINK_BUF:
  2010. if (magic32 != XFS_SYMLINK_MAGIC) {
  2011. xfs_warn(mp, "Bad symlink block magic!");
  2012. ASSERT(0);
  2013. break;
  2014. }
  2015. bp->b_ops = &xfs_symlink_buf_ops;
  2016. break;
  2017. case XFS_BLFT_DIR_BLOCK_BUF:
  2018. if (magic32 != XFS_DIR2_BLOCK_MAGIC &&
  2019. magic32 != XFS_DIR3_BLOCK_MAGIC) {
  2020. xfs_warn(mp, "Bad dir block magic!");
  2021. ASSERT(0);
  2022. break;
  2023. }
  2024. bp->b_ops = &xfs_dir3_block_buf_ops;
  2025. break;
  2026. case XFS_BLFT_DIR_DATA_BUF:
  2027. if (magic32 != XFS_DIR2_DATA_MAGIC &&
  2028. magic32 != XFS_DIR3_DATA_MAGIC) {
  2029. xfs_warn(mp, "Bad dir data magic!");
  2030. ASSERT(0);
  2031. break;
  2032. }
  2033. bp->b_ops = &xfs_dir3_data_buf_ops;
  2034. break;
  2035. case XFS_BLFT_DIR_FREE_BUF:
  2036. if (magic32 != XFS_DIR2_FREE_MAGIC &&
  2037. magic32 != XFS_DIR3_FREE_MAGIC) {
  2038. xfs_warn(mp, "Bad dir3 free magic!");
  2039. ASSERT(0);
  2040. break;
  2041. }
  2042. bp->b_ops = &xfs_dir3_free_buf_ops;
  2043. break;
  2044. case XFS_BLFT_DIR_LEAF1_BUF:
  2045. if (magicda != XFS_DIR2_LEAF1_MAGIC &&
  2046. magicda != XFS_DIR3_LEAF1_MAGIC) {
  2047. xfs_warn(mp, "Bad dir leaf1 magic!");
  2048. ASSERT(0);
  2049. break;
  2050. }
  2051. bp->b_ops = &xfs_dir3_leaf1_buf_ops;
  2052. break;
  2053. case XFS_BLFT_DIR_LEAFN_BUF:
  2054. if (magicda != XFS_DIR2_LEAFN_MAGIC &&
  2055. magicda != XFS_DIR3_LEAFN_MAGIC) {
  2056. xfs_warn(mp, "Bad dir leafn magic!");
  2057. ASSERT(0);
  2058. break;
  2059. }
  2060. bp->b_ops = &xfs_dir3_leafn_buf_ops;
  2061. break;
  2062. case XFS_BLFT_DA_NODE_BUF:
  2063. if (magicda != XFS_DA_NODE_MAGIC &&
  2064. magicda != XFS_DA3_NODE_MAGIC) {
  2065. xfs_warn(mp, "Bad da node magic!");
  2066. ASSERT(0);
  2067. break;
  2068. }
  2069. bp->b_ops = &xfs_da3_node_buf_ops;
  2070. break;
  2071. case XFS_BLFT_ATTR_LEAF_BUF:
  2072. if (magicda != XFS_ATTR_LEAF_MAGIC &&
  2073. magicda != XFS_ATTR3_LEAF_MAGIC) {
  2074. xfs_warn(mp, "Bad attr leaf magic!");
  2075. ASSERT(0);
  2076. break;
  2077. }
  2078. bp->b_ops = &xfs_attr3_leaf_buf_ops;
  2079. break;
  2080. case XFS_BLFT_ATTR_RMT_BUF:
  2081. if (!xfs_sb_version_hascrc(&mp->m_sb))
  2082. break;
  2083. if (magic32 != XFS_ATTR3_RMT_MAGIC) {
  2084. xfs_warn(mp, "Bad attr remote magic!");
  2085. ASSERT(0);
  2086. break;
  2087. }
  2088. bp->b_ops = &xfs_attr3_rmt_buf_ops;
  2089. break;
  2090. case XFS_BLFT_SB_BUF:
  2091. if (magic32 != XFS_SB_MAGIC) {
  2092. xfs_warn(mp, "Bad SB block magic!");
  2093. ASSERT(0);
  2094. break;
  2095. }
  2096. bp->b_ops = &xfs_sb_buf_ops;
  2097. break;
  2098. default:
  2099. xfs_warn(mp, "Unknown buffer type %d!",
  2100. xfs_blft_from_flags(buf_f));
  2101. break;
  2102. }
  2103. }
  2104. /*
  2105. * Perform a 'normal' buffer recovery. Each logged region of the
  2106. * buffer should be copied over the corresponding region in the
  2107. * given buffer. The bitmap in the buf log format structure indicates
  2108. * where to place the logged data.
  2109. */
  2110. STATIC void
  2111. xlog_recover_do_reg_buffer(
  2112. struct xfs_mount *mp,
  2113. xlog_recover_item_t *item,
  2114. struct xfs_buf *bp,
  2115. xfs_buf_log_format_t *buf_f)
  2116. {
  2117. int i;
  2118. int bit;
  2119. int nbits;
  2120. int error;
  2121. trace_xfs_log_recover_buf_reg_buf(mp->m_log, buf_f);
  2122. bit = 0;
  2123. i = 1; /* 0 is the buf format structure */
  2124. while (1) {
  2125. bit = xfs_next_bit(buf_f->blf_data_map,
  2126. buf_f->blf_map_size, bit);
  2127. if (bit == -1)
  2128. break;
  2129. nbits = xfs_contig_bits(buf_f->blf_data_map,
  2130. buf_f->blf_map_size, bit);
  2131. ASSERT(nbits > 0);
  2132. ASSERT(item->ri_buf[i].i_addr != NULL);
  2133. ASSERT(item->ri_buf[i].i_len % XFS_BLF_CHUNK == 0);
  2134. ASSERT(BBTOB(bp->b_io_length) >=
  2135. ((uint)bit << XFS_BLF_SHIFT) + (nbits << XFS_BLF_SHIFT));
  2136. /*
  2137. * The dirty regions logged in the buffer, even though
  2138. * contiguous, may span multiple chunks. This is because the
  2139. * dirty region may span a physical page boundary in a buffer
  2140. * and hence be split into two separate vectors for writing into
  2141. * the log. Hence we need to trim nbits back to the length of
  2142. * the current region being copied out of the log.
  2143. */
  2144. if (item->ri_buf[i].i_len < (nbits << XFS_BLF_SHIFT))
  2145. nbits = item->ri_buf[i].i_len >> XFS_BLF_SHIFT;
  2146. /*
  2147. * Do a sanity check if this is a dquot buffer. Just checking
  2148. * the first dquot in the buffer should do. XXXThis is
  2149. * probably a good thing to do for other buf types also.
  2150. */
  2151. error = 0;
  2152. if (buf_f->blf_flags &
  2153. (XFS_BLF_UDQUOT_BUF|XFS_BLF_PDQUOT_BUF|XFS_BLF_GDQUOT_BUF)) {
  2154. if (item->ri_buf[i].i_addr == NULL) {
  2155. xfs_alert(mp,
  2156. "XFS: NULL dquot in %s.", __func__);
  2157. goto next;
  2158. }
  2159. if (item->ri_buf[i].i_len < sizeof(xfs_disk_dquot_t)) {
  2160. xfs_alert(mp,
  2161. "XFS: dquot too small (%d) in %s.",
  2162. item->ri_buf[i].i_len, __func__);
  2163. goto next;
  2164. }
  2165. error = xfs_qm_dqcheck(mp, item->ri_buf[i].i_addr,
  2166. -1, 0, XFS_QMOPT_DOWARN,
  2167. "dquot_buf_recover");
  2168. if (error)
  2169. goto next;
  2170. }
  2171. memcpy(xfs_buf_offset(bp,
  2172. (uint)bit << XFS_BLF_SHIFT), /* dest */
  2173. item->ri_buf[i].i_addr, /* source */
  2174. nbits<<XFS_BLF_SHIFT); /* length */
  2175. next:
  2176. i++;
  2177. bit += nbits;
  2178. }
  2179. /* Shouldn't be any more regions */
  2180. ASSERT(i == item->ri_total);
  2181. /*
  2182. * We can only do post recovery validation on items on CRC enabled
  2183. * fielsystems as we need to know when the buffer was written to be able
  2184. * to determine if we should have replayed the item. If we replay old
  2185. * metadata over a newer buffer, then it will enter a temporarily
  2186. * inconsistent state resulting in verification failures. Hence for now
  2187. * just avoid the verification stage for non-crc filesystems
  2188. */
  2189. if (xfs_sb_version_hascrc(&mp->m_sb))
  2190. xlog_recover_validate_buf_type(mp, bp, buf_f);
  2191. }
  2192. /*
  2193. * Do some primitive error checking on ondisk dquot data structures.
  2194. */
  2195. int
  2196. xfs_qm_dqcheck(
  2197. struct xfs_mount *mp,
  2198. xfs_disk_dquot_t *ddq,
  2199. xfs_dqid_t id,
  2200. uint type, /* used only when IO_dorepair is true */
  2201. uint flags,
  2202. char *str)
  2203. {
  2204. xfs_dqblk_t *d = (xfs_dqblk_t *)ddq;
  2205. int errs = 0;
  2206. /*
  2207. * We can encounter an uninitialized dquot buffer for 2 reasons:
  2208. * 1. If we crash while deleting the quotainode(s), and those blks got
  2209. * used for user data. This is because we take the path of regular
  2210. * file deletion; however, the size field of quotainodes is never
  2211. * updated, so all the tricks that we play in itruncate_finish
  2212. * don't quite matter.
  2213. *
  2214. * 2. We don't play the quota buffers when there's a quotaoff logitem.
  2215. * But the allocation will be replayed so we'll end up with an
  2216. * uninitialized quota block.
  2217. *
  2218. * This is all fine; things are still consistent, and we haven't lost
  2219. * any quota information. Just don't complain about bad dquot blks.
  2220. */
  2221. if (ddq->d_magic != cpu_to_be16(XFS_DQUOT_MAGIC)) {
  2222. if (flags & XFS_QMOPT_DOWARN)
  2223. xfs_alert(mp,
  2224. "%s : XFS dquot ID 0x%x, magic 0x%x != 0x%x",
  2225. str, id, be16_to_cpu(ddq->d_magic), XFS_DQUOT_MAGIC);
  2226. errs++;
  2227. }
  2228. if (ddq->d_version != XFS_DQUOT_VERSION) {
  2229. if (flags & XFS_QMOPT_DOWARN)
  2230. xfs_alert(mp,
  2231. "%s : XFS dquot ID 0x%x, version 0x%x != 0x%x",
  2232. str, id, ddq->d_version, XFS_DQUOT_VERSION);
  2233. errs++;
  2234. }
  2235. if (ddq->d_flags != XFS_DQ_USER &&
  2236. ddq->d_flags != XFS_DQ_PROJ &&
  2237. ddq->d_flags != XFS_DQ_GROUP) {
  2238. if (flags & XFS_QMOPT_DOWARN)
  2239. xfs_alert(mp,
  2240. "%s : XFS dquot ID 0x%x, unknown flags 0x%x",
  2241. str, id, ddq->d_flags);
  2242. errs++;
  2243. }
  2244. if (id != -1 && id != be32_to_cpu(ddq->d_id)) {
  2245. if (flags & XFS_QMOPT_DOWARN)
  2246. xfs_alert(mp,
  2247. "%s : ondisk-dquot 0x%p, ID mismatch: "
  2248. "0x%x expected, found id 0x%x",
  2249. str, ddq, id, be32_to_cpu(ddq->d_id));
  2250. errs++;
  2251. }
  2252. if (!errs && ddq->d_id) {
  2253. if (ddq->d_blk_softlimit &&
  2254. be64_to_cpu(ddq->d_bcount) >
  2255. be64_to_cpu(ddq->d_blk_softlimit)) {
  2256. if (!ddq->d_btimer) {
  2257. if (flags & XFS_QMOPT_DOWARN)
  2258. xfs_alert(mp,
  2259. "%s : Dquot ID 0x%x (0x%p) BLK TIMER NOT STARTED",
  2260. str, (int)be32_to_cpu(ddq->d_id), ddq);
  2261. errs++;
  2262. }
  2263. }
  2264. if (ddq->d_ino_softlimit &&
  2265. be64_to_cpu(ddq->d_icount) >
  2266. be64_to_cpu(ddq->d_ino_softlimit)) {
  2267. if (!ddq->d_itimer) {
  2268. if (flags & XFS_QMOPT_DOWARN)
  2269. xfs_alert(mp,
  2270. "%s : Dquot ID 0x%x (0x%p) INODE TIMER NOT STARTED",
  2271. str, (int)be32_to_cpu(ddq->d_id), ddq);
  2272. errs++;
  2273. }
  2274. }
  2275. if (ddq->d_rtb_softlimit &&
  2276. be64_to_cpu(ddq->d_rtbcount) >
  2277. be64_to_cpu(ddq->d_rtb_softlimit)) {
  2278. if (!ddq->d_rtbtimer) {
  2279. if (flags & XFS_QMOPT_DOWARN)
  2280. xfs_alert(mp,
  2281. "%s : Dquot ID 0x%x (0x%p) RTBLK TIMER NOT STARTED",
  2282. str, (int)be32_to_cpu(ddq->d_id), ddq);
  2283. errs++;
  2284. }
  2285. }
  2286. }
  2287. if (!errs || !(flags & XFS_QMOPT_DQREPAIR))
  2288. return errs;
  2289. if (flags & XFS_QMOPT_DOWARN)
  2290. xfs_notice(mp, "Re-initializing dquot ID 0x%x", id);
  2291. /*
  2292. * Typically, a repair is only requested by quotacheck.
  2293. */
  2294. ASSERT(id != -1);
  2295. ASSERT(flags & XFS_QMOPT_DQREPAIR);
  2296. memset(d, 0, sizeof(xfs_dqblk_t));
  2297. d->dd_diskdq.d_magic = cpu_to_be16(XFS_DQUOT_MAGIC);
  2298. d->dd_diskdq.d_version = XFS_DQUOT_VERSION;
  2299. d->dd_diskdq.d_flags = type;
  2300. d->dd_diskdq.d_id = cpu_to_be32(id);
  2301. if (xfs_sb_version_hascrc(&mp->m_sb)) {
  2302. uuid_copy(&d->dd_uuid, &mp->m_sb.sb_uuid);
  2303. xfs_update_cksum((char *)d, sizeof(struct xfs_dqblk),
  2304. XFS_DQUOT_CRC_OFF);
  2305. }
  2306. return errs;
  2307. }
  2308. /*
  2309. * Perform a dquot buffer recovery.
  2310. * Simple algorithm: if we have found a QUOTAOFF log item of the same type
  2311. * (ie. USR or GRP), then just toss this buffer away; don't recover it.
  2312. * Else, treat it as a regular buffer and do recovery.
  2313. */
  2314. STATIC void
  2315. xlog_recover_do_dquot_buffer(
  2316. struct xfs_mount *mp,
  2317. struct xlog *log,
  2318. struct xlog_recover_item *item,
  2319. struct xfs_buf *bp,
  2320. struct xfs_buf_log_format *buf_f)
  2321. {
  2322. uint type;
  2323. trace_xfs_log_recover_buf_dquot_buf(log, buf_f);
  2324. /*
  2325. * Filesystems are required to send in quota flags at mount time.
  2326. */
  2327. if (mp->m_qflags == 0) {
  2328. return;
  2329. }
  2330. type = 0;
  2331. if (buf_f->blf_flags & XFS_BLF_UDQUOT_BUF)
  2332. type |= XFS_DQ_USER;
  2333. if (buf_f->blf_flags & XFS_BLF_PDQUOT_BUF)
  2334. type |= XFS_DQ_PROJ;
  2335. if (buf_f->blf_flags & XFS_BLF_GDQUOT_BUF)
  2336. type |= XFS_DQ_GROUP;
  2337. /*
  2338. * This type of quotas was turned off, so ignore this buffer
  2339. */
  2340. if (log->l_quotaoffs_flag & type)
  2341. return;
  2342. xlog_recover_do_reg_buffer(mp, item, bp, buf_f);
  2343. }
  2344. /*
  2345. * This routine replays a modification made to a buffer at runtime.
  2346. * There are actually two types of buffer, regular and inode, which
  2347. * are handled differently. Inode buffers are handled differently
  2348. * in that we only recover a specific set of data from them, namely
  2349. * the inode di_next_unlinked fields. This is because all other inode
  2350. * data is actually logged via inode records and any data we replay
  2351. * here which overlaps that may be stale.
  2352. *
  2353. * When meta-data buffers are freed at run time we log a buffer item
  2354. * with the XFS_BLF_CANCEL bit set to indicate that previous copies
  2355. * of the buffer in the log should not be replayed at recovery time.
  2356. * This is so that if the blocks covered by the buffer are reused for
  2357. * file data before we crash we don't end up replaying old, freed
  2358. * meta-data into a user's file.
  2359. *
  2360. * To handle the cancellation of buffer log items, we make two passes
  2361. * over the log during recovery. During the first we build a table of
  2362. * those buffers which have been cancelled, and during the second we
  2363. * only replay those buffers which do not have corresponding cancel
  2364. * records in the table. See xlog_recover_buffer_pass[1,2] above
  2365. * for more details on the implementation of the table of cancel records.
  2366. */
  2367. STATIC int
  2368. xlog_recover_buffer_pass2(
  2369. struct xlog *log,
  2370. struct list_head *buffer_list,
  2371. struct xlog_recover_item *item,
  2372. xfs_lsn_t current_lsn)
  2373. {
  2374. xfs_buf_log_format_t *buf_f = item->ri_buf[0].i_addr;
  2375. xfs_mount_t *mp = log->l_mp;
  2376. xfs_buf_t *bp;
  2377. int error;
  2378. uint buf_flags;
  2379. xfs_lsn_t lsn;
  2380. /*
  2381. * In this pass we only want to recover all the buffers which have
  2382. * not been cancelled and are not cancellation buffers themselves.
  2383. */
  2384. if (xlog_check_buffer_cancelled(log, buf_f->blf_blkno,
  2385. buf_f->blf_len, buf_f->blf_flags)) {
  2386. trace_xfs_log_recover_buf_cancel(log, buf_f);
  2387. return 0;
  2388. }
  2389. trace_xfs_log_recover_buf_recover(log, buf_f);
  2390. buf_flags = 0;
  2391. if (buf_f->blf_flags & XFS_BLF_INODE_BUF)
  2392. buf_flags |= XBF_UNMAPPED;
  2393. bp = xfs_buf_read(mp->m_ddev_targp, buf_f->blf_blkno, buf_f->blf_len,
  2394. buf_flags, NULL);
  2395. if (!bp)
  2396. return XFS_ERROR(ENOMEM);
  2397. error = bp->b_error;
  2398. if (error) {
  2399. xfs_buf_ioerror_alert(bp, "xlog_recover_do..(read#1)");
  2400. goto out_release;
  2401. }
  2402. /*
  2403. * recover the buffer only if we get an LSN from it and it's less than
  2404. * the lsn of the transaction we are replaying.
  2405. */
  2406. lsn = xlog_recover_get_buf_lsn(mp, bp);
  2407. if (lsn && lsn != -1 && XFS_LSN_CMP(lsn, current_lsn) >= 0)
  2408. goto out_release;
  2409. if (buf_f->blf_flags & XFS_BLF_INODE_BUF) {
  2410. error = xlog_recover_do_inode_buffer(mp, item, bp, buf_f);
  2411. } else if (buf_f->blf_flags &
  2412. (XFS_BLF_UDQUOT_BUF|XFS_BLF_PDQUOT_BUF|XFS_BLF_GDQUOT_BUF)) {
  2413. xlog_recover_do_dquot_buffer(mp, log, item, bp, buf_f);
  2414. } else {
  2415. xlog_recover_do_reg_buffer(mp, item, bp, buf_f);
  2416. }
  2417. if (error)
  2418. goto out_release;
  2419. /*
  2420. * Perform delayed write on the buffer. Asynchronous writes will be
  2421. * slower when taking into account all the buffers to be flushed.
  2422. *
  2423. * Also make sure that only inode buffers with good sizes stay in
  2424. * the buffer cache. The kernel moves inodes in buffers of 1 block
  2425. * or XFS_INODE_CLUSTER_SIZE bytes, whichever is bigger. The inode
  2426. * buffers in the log can be a different size if the log was generated
  2427. * by an older kernel using unclustered inode buffers or a newer kernel
  2428. * running with a different inode cluster size. Regardless, if the
  2429. * the inode buffer size isn't MAX(blocksize, XFS_INODE_CLUSTER_SIZE)
  2430. * for *our* value of XFS_INODE_CLUSTER_SIZE, then we need to keep
  2431. * the buffer out of the buffer cache so that the buffer won't
  2432. * overlap with future reads of those inodes.
  2433. */
  2434. if (XFS_DINODE_MAGIC ==
  2435. be16_to_cpu(*((__be16 *)xfs_buf_offset(bp, 0))) &&
  2436. (BBTOB(bp->b_io_length) != MAX(log->l_mp->m_sb.sb_blocksize,
  2437. (__uint32_t)XFS_INODE_CLUSTER_SIZE(log->l_mp)))) {
  2438. xfs_buf_stale(bp);
  2439. error = xfs_bwrite(bp);
  2440. } else {
  2441. ASSERT(bp->b_target->bt_mount == mp);
  2442. bp->b_iodone = xlog_recover_iodone;
  2443. xfs_buf_delwri_queue(bp, buffer_list);
  2444. }
  2445. out_release:
  2446. xfs_buf_relse(bp);
  2447. return error;
  2448. }
  2449. /*
  2450. * Inode fork owner changes
  2451. *
  2452. * If we have been told that we have to reparent the inode fork, it's because an
  2453. * extent swap operation on a CRC enabled filesystem has been done and we are
  2454. * replaying it. We need to walk the BMBT of the appropriate fork and change the
  2455. * owners of it.
  2456. *
  2457. * The complexity here is that we don't have an inode context to work with, so
  2458. * after we've replayed the inode we need to instantiate one. This is where the
  2459. * fun begins.
  2460. *
  2461. * We are in the middle of log recovery, so we can't run transactions. That
  2462. * means we cannot use cache coherent inode instantiation via xfs_iget(), as
  2463. * that will result in the corresponding iput() running the inode through
  2464. * xfs_inactive(). If we've just replayed an inode core that changes the link
  2465. * count to zero (i.e. it's been unlinked), then xfs_inactive() will run
  2466. * transactions (bad!).
  2467. *
  2468. * So, to avoid this, we instantiate an inode directly from the inode core we've
  2469. * just recovered. We have the buffer still locked, and all we really need to
  2470. * instantiate is the inode core and the forks being modified. We can do this
  2471. * manually, then run the inode btree owner change, and then tear down the
  2472. * xfs_inode without having to run any transactions at all.
  2473. *
  2474. * Also, because we don't have a transaction context available here but need to
  2475. * gather all the buffers we modify for writeback so we pass the buffer_list
  2476. * instead for the operation to use.
  2477. */
  2478. STATIC int
  2479. xfs_recover_inode_owner_change(
  2480. struct xfs_mount *mp,
  2481. struct xfs_dinode *dip,
  2482. struct xfs_inode_log_format *in_f,
  2483. struct list_head *buffer_list)
  2484. {
  2485. struct xfs_inode *ip;
  2486. int error;
  2487. ASSERT(in_f->ilf_fields & (XFS_ILOG_DOWNER|XFS_ILOG_AOWNER));
  2488. ip = xfs_inode_alloc(mp, in_f->ilf_ino);
  2489. if (!ip)
  2490. return ENOMEM;
  2491. /* instantiate the inode */
  2492. xfs_dinode_from_disk(&ip->i_d, dip);
  2493. ASSERT(ip->i_d.di_version >= 3);
  2494. error = xfs_iformat_fork(ip, dip);
  2495. if (error)
  2496. goto out_free_ip;
  2497. if (in_f->ilf_fields & XFS_ILOG_DOWNER) {
  2498. ASSERT(in_f->ilf_fields & XFS_ILOG_DBROOT);
  2499. error = xfs_bmbt_change_owner(NULL, ip, XFS_DATA_FORK,
  2500. ip->i_ino, buffer_list);
  2501. if (error)
  2502. goto out_free_ip;
  2503. }
  2504. if (in_f->ilf_fields & XFS_ILOG_AOWNER) {
  2505. ASSERT(in_f->ilf_fields & XFS_ILOG_ABROOT);
  2506. error = xfs_bmbt_change_owner(NULL, ip, XFS_ATTR_FORK,
  2507. ip->i_ino, buffer_list);
  2508. if (error)
  2509. goto out_free_ip;
  2510. }
  2511. out_free_ip:
  2512. xfs_inode_free(ip);
  2513. return error;
  2514. }
  2515. STATIC int
  2516. xlog_recover_inode_pass2(
  2517. struct xlog *log,
  2518. struct list_head *buffer_list,
  2519. struct xlog_recover_item *item,
  2520. xfs_lsn_t current_lsn)
  2521. {
  2522. xfs_inode_log_format_t *in_f;
  2523. xfs_mount_t *mp = log->l_mp;
  2524. xfs_buf_t *bp;
  2525. xfs_dinode_t *dip;
  2526. int len;
  2527. xfs_caddr_t src;
  2528. xfs_caddr_t dest;
  2529. int error;
  2530. int attr_index;
  2531. uint fields;
  2532. xfs_icdinode_t *dicp;
  2533. uint isize;
  2534. int need_free = 0;
  2535. if (item->ri_buf[0].i_len == sizeof(xfs_inode_log_format_t)) {
  2536. in_f = item->ri_buf[0].i_addr;
  2537. } else {
  2538. in_f = kmem_alloc(sizeof(xfs_inode_log_format_t), KM_SLEEP);
  2539. need_free = 1;
  2540. error = xfs_inode_item_format_convert(&item->ri_buf[0], in_f);
  2541. if (error)
  2542. goto error;
  2543. }
  2544. /*
  2545. * Inode buffers can be freed, look out for it,
  2546. * and do not replay the inode.
  2547. */
  2548. if (xlog_check_buffer_cancelled(log, in_f->ilf_blkno,
  2549. in_f->ilf_len, 0)) {
  2550. error = 0;
  2551. trace_xfs_log_recover_inode_cancel(log, in_f);
  2552. goto error;
  2553. }
  2554. trace_xfs_log_recover_inode_recover(log, in_f);
  2555. bp = xfs_buf_read(mp->m_ddev_targp, in_f->ilf_blkno, in_f->ilf_len, 0,
  2556. &xfs_inode_buf_ops);
  2557. if (!bp) {
  2558. error = ENOMEM;
  2559. goto error;
  2560. }
  2561. error = bp->b_error;
  2562. if (error) {
  2563. xfs_buf_ioerror_alert(bp, "xlog_recover_do..(read#2)");
  2564. goto out_release;
  2565. }
  2566. ASSERT(in_f->ilf_fields & XFS_ILOG_CORE);
  2567. dip = (xfs_dinode_t *)xfs_buf_offset(bp, in_f->ilf_boffset);
  2568. /*
  2569. * Make sure the place we're flushing out to really looks
  2570. * like an inode!
  2571. */
  2572. if (unlikely(dip->di_magic != cpu_to_be16(XFS_DINODE_MAGIC))) {
  2573. xfs_alert(mp,
  2574. "%s: Bad inode magic number, dip = 0x%p, dino bp = 0x%p, ino = %Ld",
  2575. __func__, dip, bp, in_f->ilf_ino);
  2576. XFS_ERROR_REPORT("xlog_recover_inode_pass2(1)",
  2577. XFS_ERRLEVEL_LOW, mp);
  2578. error = EFSCORRUPTED;
  2579. goto out_release;
  2580. }
  2581. dicp = item->ri_buf[1].i_addr;
  2582. if (unlikely(dicp->di_magic != XFS_DINODE_MAGIC)) {
  2583. xfs_alert(mp,
  2584. "%s: Bad inode log record, rec ptr 0x%p, ino %Ld",
  2585. __func__, item, in_f->ilf_ino);
  2586. XFS_ERROR_REPORT("xlog_recover_inode_pass2(2)",
  2587. XFS_ERRLEVEL_LOW, mp);
  2588. error = EFSCORRUPTED;
  2589. goto out_release;
  2590. }
  2591. /*
  2592. * If the inode has an LSN in it, recover the inode only if it's less
  2593. * than the lsn of the transaction we are replaying. Note: we still
  2594. * need to replay an owner change even though the inode is more recent
  2595. * than the transaction as there is no guarantee that all the btree
  2596. * blocks are more recent than this transaction, too.
  2597. */
  2598. if (dip->di_version >= 3) {
  2599. xfs_lsn_t lsn = be64_to_cpu(dip->di_lsn);
  2600. if (lsn && lsn != -1 && XFS_LSN_CMP(lsn, current_lsn) >= 0) {
  2601. trace_xfs_log_recover_inode_skip(log, in_f);
  2602. error = 0;
  2603. goto out_owner_change;
  2604. }
  2605. }
  2606. /*
  2607. * di_flushiter is only valid for v1/2 inodes. All changes for v3 inodes
  2608. * are transactional and if ordering is necessary we can determine that
  2609. * more accurately by the LSN field in the V3 inode core. Don't trust
  2610. * the inode versions we might be changing them here - use the
  2611. * superblock flag to determine whether we need to look at di_flushiter
  2612. * to skip replay when the on disk inode is newer than the log one
  2613. */
  2614. if (!xfs_sb_version_hascrc(&mp->m_sb) &&
  2615. dicp->di_flushiter < be16_to_cpu(dip->di_flushiter)) {
  2616. /*
  2617. * Deal with the wrap case, DI_MAX_FLUSH is less
  2618. * than smaller numbers
  2619. */
  2620. if (be16_to_cpu(dip->di_flushiter) == DI_MAX_FLUSH &&
  2621. dicp->di_flushiter < (DI_MAX_FLUSH >> 1)) {
  2622. /* do nothing */
  2623. } else {
  2624. trace_xfs_log_recover_inode_skip(log, in_f);
  2625. error = 0;
  2626. goto out_release;
  2627. }
  2628. }
  2629. /* Take the opportunity to reset the flush iteration count */
  2630. dicp->di_flushiter = 0;
  2631. if (unlikely(S_ISREG(dicp->di_mode))) {
  2632. if ((dicp->di_format != XFS_DINODE_FMT_EXTENTS) &&
  2633. (dicp->di_format != XFS_DINODE_FMT_BTREE)) {
  2634. XFS_CORRUPTION_ERROR("xlog_recover_inode_pass2(3)",
  2635. XFS_ERRLEVEL_LOW, mp, dicp);
  2636. xfs_alert(mp,
  2637. "%s: Bad regular inode log record, rec ptr 0x%p, "
  2638. "ino ptr = 0x%p, ino bp = 0x%p, ino %Ld",
  2639. __func__, item, dip, bp, in_f->ilf_ino);
  2640. error = EFSCORRUPTED;
  2641. goto out_release;
  2642. }
  2643. } else if (unlikely(S_ISDIR(dicp->di_mode))) {
  2644. if ((dicp->di_format != XFS_DINODE_FMT_EXTENTS) &&
  2645. (dicp->di_format != XFS_DINODE_FMT_BTREE) &&
  2646. (dicp->di_format != XFS_DINODE_FMT_LOCAL)) {
  2647. XFS_CORRUPTION_ERROR("xlog_recover_inode_pass2(4)",
  2648. XFS_ERRLEVEL_LOW, mp, dicp);
  2649. xfs_alert(mp,
  2650. "%s: Bad dir inode log record, rec ptr 0x%p, "
  2651. "ino ptr = 0x%p, ino bp = 0x%p, ino %Ld",
  2652. __func__, item, dip, bp, in_f->ilf_ino);
  2653. error = EFSCORRUPTED;
  2654. goto out_release;
  2655. }
  2656. }
  2657. if (unlikely(dicp->di_nextents + dicp->di_anextents > dicp->di_nblocks)){
  2658. XFS_CORRUPTION_ERROR("xlog_recover_inode_pass2(5)",
  2659. XFS_ERRLEVEL_LOW, mp, dicp);
  2660. xfs_alert(mp,
  2661. "%s: Bad inode log record, rec ptr 0x%p, dino ptr 0x%p, "
  2662. "dino bp 0x%p, ino %Ld, total extents = %d, nblocks = %Ld",
  2663. __func__, item, dip, bp, in_f->ilf_ino,
  2664. dicp->di_nextents + dicp->di_anextents,
  2665. dicp->di_nblocks);
  2666. error = EFSCORRUPTED;
  2667. goto out_release;
  2668. }
  2669. if (unlikely(dicp->di_forkoff > mp->m_sb.sb_inodesize)) {
  2670. XFS_CORRUPTION_ERROR("xlog_recover_inode_pass2(6)",
  2671. XFS_ERRLEVEL_LOW, mp, dicp);
  2672. xfs_alert(mp,
  2673. "%s: Bad inode log record, rec ptr 0x%p, dino ptr 0x%p, "
  2674. "dino bp 0x%p, ino %Ld, forkoff 0x%x", __func__,
  2675. item, dip, bp, in_f->ilf_ino, dicp->di_forkoff);
  2676. error = EFSCORRUPTED;
  2677. goto out_release;
  2678. }
  2679. isize = xfs_icdinode_size(dicp->di_version);
  2680. if (unlikely(item->ri_buf[1].i_len > isize)) {
  2681. XFS_CORRUPTION_ERROR("xlog_recover_inode_pass2(7)",
  2682. XFS_ERRLEVEL_LOW, mp, dicp);
  2683. xfs_alert(mp,
  2684. "%s: Bad inode log record length %d, rec ptr 0x%p",
  2685. __func__, item->ri_buf[1].i_len, item);
  2686. error = EFSCORRUPTED;
  2687. goto out_release;
  2688. }
  2689. /* The core is in in-core format */
  2690. xfs_dinode_to_disk(dip, dicp);
  2691. /* the rest is in on-disk format */
  2692. if (item->ri_buf[1].i_len > isize) {
  2693. memcpy((char *)dip + isize,
  2694. item->ri_buf[1].i_addr + isize,
  2695. item->ri_buf[1].i_len - isize);
  2696. }
  2697. fields = in_f->ilf_fields;
  2698. switch (fields & (XFS_ILOG_DEV | XFS_ILOG_UUID)) {
  2699. case XFS_ILOG_DEV:
  2700. xfs_dinode_put_rdev(dip, in_f->ilf_u.ilfu_rdev);
  2701. break;
  2702. case XFS_ILOG_UUID:
  2703. memcpy(XFS_DFORK_DPTR(dip),
  2704. &in_f->ilf_u.ilfu_uuid,
  2705. sizeof(uuid_t));
  2706. break;
  2707. }
  2708. if (in_f->ilf_size == 2)
  2709. goto out_owner_change;
  2710. len = item->ri_buf[2].i_len;
  2711. src = item->ri_buf[2].i_addr;
  2712. ASSERT(in_f->ilf_size <= 4);
  2713. ASSERT((in_f->ilf_size == 3) || (fields & XFS_ILOG_AFORK));
  2714. ASSERT(!(fields & XFS_ILOG_DFORK) ||
  2715. (len == in_f->ilf_dsize));
  2716. switch (fields & XFS_ILOG_DFORK) {
  2717. case XFS_ILOG_DDATA:
  2718. case XFS_ILOG_DEXT:
  2719. memcpy(XFS_DFORK_DPTR(dip), src, len);
  2720. break;
  2721. case XFS_ILOG_DBROOT:
  2722. xfs_bmbt_to_bmdr(mp, (struct xfs_btree_block *)src, len,
  2723. (xfs_bmdr_block_t *)XFS_DFORK_DPTR(dip),
  2724. XFS_DFORK_DSIZE(dip, mp));
  2725. break;
  2726. default:
  2727. /*
  2728. * There are no data fork flags set.
  2729. */
  2730. ASSERT((fields & XFS_ILOG_DFORK) == 0);
  2731. break;
  2732. }
  2733. /*
  2734. * If we logged any attribute data, recover it. There may or
  2735. * may not have been any other non-core data logged in this
  2736. * transaction.
  2737. */
  2738. if (in_f->ilf_fields & XFS_ILOG_AFORK) {
  2739. if (in_f->ilf_fields & XFS_ILOG_DFORK) {
  2740. attr_index = 3;
  2741. } else {
  2742. attr_index = 2;
  2743. }
  2744. len = item->ri_buf[attr_index].i_len;
  2745. src = item->ri_buf[attr_index].i_addr;
  2746. ASSERT(len == in_f->ilf_asize);
  2747. switch (in_f->ilf_fields & XFS_ILOG_AFORK) {
  2748. case XFS_ILOG_ADATA:
  2749. case XFS_ILOG_AEXT:
  2750. dest = XFS_DFORK_APTR(dip);
  2751. ASSERT(len <= XFS_DFORK_ASIZE(dip, mp));
  2752. memcpy(dest, src, len);
  2753. break;
  2754. case XFS_ILOG_ABROOT:
  2755. dest = XFS_DFORK_APTR(dip);
  2756. xfs_bmbt_to_bmdr(mp, (struct xfs_btree_block *)src,
  2757. len, (xfs_bmdr_block_t*)dest,
  2758. XFS_DFORK_ASIZE(dip, mp));
  2759. break;
  2760. default:
  2761. xfs_warn(log->l_mp, "%s: Invalid flag", __func__);
  2762. ASSERT(0);
  2763. error = EIO;
  2764. goto out_release;
  2765. }
  2766. }
  2767. out_owner_change:
  2768. if (in_f->ilf_fields & (XFS_ILOG_DOWNER|XFS_ILOG_AOWNER))
  2769. error = xfs_recover_inode_owner_change(mp, dip, in_f,
  2770. buffer_list);
  2771. /* re-generate the checksum. */
  2772. xfs_dinode_calc_crc(log->l_mp, dip);
  2773. ASSERT(bp->b_target->bt_mount == mp);
  2774. bp->b_iodone = xlog_recover_iodone;
  2775. xfs_buf_delwri_queue(bp, buffer_list);
  2776. out_release:
  2777. xfs_buf_relse(bp);
  2778. error:
  2779. if (need_free)
  2780. kmem_free(in_f);
  2781. return XFS_ERROR(error);
  2782. }
  2783. /*
  2784. * Recover QUOTAOFF records. We simply make a note of it in the xlog
  2785. * structure, so that we know not to do any dquot item or dquot buffer recovery,
  2786. * of that type.
  2787. */
  2788. STATIC int
  2789. xlog_recover_quotaoff_pass1(
  2790. struct xlog *log,
  2791. struct xlog_recover_item *item)
  2792. {
  2793. xfs_qoff_logformat_t *qoff_f = item->ri_buf[0].i_addr;
  2794. ASSERT(qoff_f);
  2795. /*
  2796. * The logitem format's flag tells us if this was user quotaoff,
  2797. * group/project quotaoff or both.
  2798. */
  2799. if (qoff_f->qf_flags & XFS_UQUOTA_ACCT)
  2800. log->l_quotaoffs_flag |= XFS_DQ_USER;
  2801. if (qoff_f->qf_flags & XFS_PQUOTA_ACCT)
  2802. log->l_quotaoffs_flag |= XFS_DQ_PROJ;
  2803. if (qoff_f->qf_flags & XFS_GQUOTA_ACCT)
  2804. log->l_quotaoffs_flag |= XFS_DQ_GROUP;
  2805. return (0);
  2806. }
  2807. /*
  2808. * Recover a dquot record
  2809. */
  2810. STATIC int
  2811. xlog_recover_dquot_pass2(
  2812. struct xlog *log,
  2813. struct list_head *buffer_list,
  2814. struct xlog_recover_item *item,
  2815. xfs_lsn_t current_lsn)
  2816. {
  2817. xfs_mount_t *mp = log->l_mp;
  2818. xfs_buf_t *bp;
  2819. struct xfs_disk_dquot *ddq, *recddq;
  2820. int error;
  2821. xfs_dq_logformat_t *dq_f;
  2822. uint type;
  2823. /*
  2824. * Filesystems are required to send in quota flags at mount time.
  2825. */
  2826. if (mp->m_qflags == 0)
  2827. return (0);
  2828. recddq = item->ri_buf[1].i_addr;
  2829. if (recddq == NULL) {
  2830. xfs_alert(log->l_mp, "NULL dquot in %s.", __func__);
  2831. return XFS_ERROR(EIO);
  2832. }
  2833. if (item->ri_buf[1].i_len < sizeof(xfs_disk_dquot_t)) {
  2834. xfs_alert(log->l_mp, "dquot too small (%d) in %s.",
  2835. item->ri_buf[1].i_len, __func__);
  2836. return XFS_ERROR(EIO);
  2837. }
  2838. /*
  2839. * This type of quotas was turned off, so ignore this record.
  2840. */
  2841. type = recddq->d_flags & (XFS_DQ_USER | XFS_DQ_PROJ | XFS_DQ_GROUP);
  2842. ASSERT(type);
  2843. if (log->l_quotaoffs_flag & type)
  2844. return (0);
  2845. /*
  2846. * At this point we know that quota was _not_ turned off.
  2847. * Since the mount flags are not indicating to us otherwise, this
  2848. * must mean that quota is on, and the dquot needs to be replayed.
  2849. * Remember that we may not have fully recovered the superblock yet,
  2850. * so we can't do the usual trick of looking at the SB quota bits.
  2851. *
  2852. * The other possibility, of course, is that the quota subsystem was
  2853. * removed since the last mount - ENOSYS.
  2854. */
  2855. dq_f = item->ri_buf[0].i_addr;
  2856. ASSERT(dq_f);
  2857. error = xfs_qm_dqcheck(mp, recddq, dq_f->qlf_id, 0, XFS_QMOPT_DOWARN,
  2858. "xlog_recover_dquot_pass2 (log copy)");
  2859. if (error)
  2860. return XFS_ERROR(EIO);
  2861. ASSERT(dq_f->qlf_len == 1);
  2862. error = xfs_trans_read_buf(mp, NULL, mp->m_ddev_targp, dq_f->qlf_blkno,
  2863. XFS_FSB_TO_BB(mp, dq_f->qlf_len), 0, &bp,
  2864. NULL);
  2865. if (error)
  2866. return error;
  2867. ASSERT(bp);
  2868. ddq = (xfs_disk_dquot_t *)xfs_buf_offset(bp, dq_f->qlf_boffset);
  2869. /*
  2870. * At least the magic num portion should be on disk because this
  2871. * was among a chunk of dquots created earlier, and we did some
  2872. * minimal initialization then.
  2873. */
  2874. error = xfs_qm_dqcheck(mp, ddq, dq_f->qlf_id, 0, XFS_QMOPT_DOWARN,
  2875. "xlog_recover_dquot_pass2");
  2876. if (error) {
  2877. xfs_buf_relse(bp);
  2878. return XFS_ERROR(EIO);
  2879. }
  2880. /*
  2881. * If the dquot has an LSN in it, recover the dquot only if it's less
  2882. * than the lsn of the transaction we are replaying.
  2883. */
  2884. if (xfs_sb_version_hascrc(&mp->m_sb)) {
  2885. struct xfs_dqblk *dqb = (struct xfs_dqblk *)ddq;
  2886. xfs_lsn_t lsn = be64_to_cpu(dqb->dd_lsn);
  2887. if (lsn && lsn != -1 && XFS_LSN_CMP(lsn, current_lsn) >= 0) {
  2888. goto out_release;
  2889. }
  2890. }
  2891. memcpy(ddq, recddq, item->ri_buf[1].i_len);
  2892. if (xfs_sb_version_hascrc(&mp->m_sb)) {
  2893. xfs_update_cksum((char *)ddq, sizeof(struct xfs_dqblk),
  2894. XFS_DQUOT_CRC_OFF);
  2895. }
  2896. ASSERT(dq_f->qlf_size == 2);
  2897. ASSERT(bp->b_target->bt_mount == mp);
  2898. bp->b_iodone = xlog_recover_iodone;
  2899. xfs_buf_delwri_queue(bp, buffer_list);
  2900. out_release:
  2901. xfs_buf_relse(bp);
  2902. return 0;
  2903. }
  2904. /*
  2905. * This routine is called to create an in-core extent free intent
  2906. * item from the efi format structure which was logged on disk.
  2907. * It allocates an in-core efi, copies the extents from the format
  2908. * structure into it, and adds the efi to the AIL with the given
  2909. * LSN.
  2910. */
  2911. STATIC int
  2912. xlog_recover_efi_pass2(
  2913. struct xlog *log,
  2914. struct xlog_recover_item *item,
  2915. xfs_lsn_t lsn)
  2916. {
  2917. int error;
  2918. xfs_mount_t *mp = log->l_mp;
  2919. xfs_efi_log_item_t *efip;
  2920. xfs_efi_log_format_t *efi_formatp;
  2921. efi_formatp = item->ri_buf[0].i_addr;
  2922. efip = xfs_efi_init(mp, efi_formatp->efi_nextents);
  2923. if ((error = xfs_efi_copy_format(&(item->ri_buf[0]),
  2924. &(efip->efi_format)))) {
  2925. xfs_efi_item_free(efip);
  2926. return error;
  2927. }
  2928. atomic_set(&efip->efi_next_extent, efi_formatp->efi_nextents);
  2929. spin_lock(&log->l_ailp->xa_lock);
  2930. /*
  2931. * xfs_trans_ail_update() drops the AIL lock.
  2932. */
  2933. xfs_trans_ail_update(log->l_ailp, &efip->efi_item, lsn);
  2934. return 0;
  2935. }
  2936. /*
  2937. * This routine is called when an efd format structure is found in
  2938. * a committed transaction in the log. It's purpose is to cancel
  2939. * the corresponding efi if it was still in the log. To do this
  2940. * it searches the AIL for the efi with an id equal to that in the
  2941. * efd format structure. If we find it, we remove the efi from the
  2942. * AIL and free it.
  2943. */
  2944. STATIC int
  2945. xlog_recover_efd_pass2(
  2946. struct xlog *log,
  2947. struct xlog_recover_item *item)
  2948. {
  2949. xfs_efd_log_format_t *efd_formatp;
  2950. xfs_efi_log_item_t *efip = NULL;
  2951. xfs_log_item_t *lip;
  2952. __uint64_t efi_id;
  2953. struct xfs_ail_cursor cur;
  2954. struct xfs_ail *ailp = log->l_ailp;
  2955. efd_formatp = item->ri_buf[0].i_addr;
  2956. ASSERT((item->ri_buf[0].i_len == (sizeof(xfs_efd_log_format_32_t) +
  2957. ((efd_formatp->efd_nextents - 1) * sizeof(xfs_extent_32_t)))) ||
  2958. (item->ri_buf[0].i_len == (sizeof(xfs_efd_log_format_64_t) +
  2959. ((efd_formatp->efd_nextents - 1) * sizeof(xfs_extent_64_t)))));
  2960. efi_id = efd_formatp->efd_efi_id;
  2961. /*
  2962. * Search for the efi with the id in the efd format structure
  2963. * in the AIL.
  2964. */
  2965. spin_lock(&ailp->xa_lock);
  2966. lip = xfs_trans_ail_cursor_first(ailp, &cur, 0);
  2967. while (lip != NULL) {
  2968. if (lip->li_type == XFS_LI_EFI) {
  2969. efip = (xfs_efi_log_item_t *)lip;
  2970. if (efip->efi_format.efi_id == efi_id) {
  2971. /*
  2972. * xfs_trans_ail_delete() drops the
  2973. * AIL lock.
  2974. */
  2975. xfs_trans_ail_delete(ailp, lip,
  2976. SHUTDOWN_CORRUPT_INCORE);
  2977. xfs_efi_item_free(efip);
  2978. spin_lock(&ailp->xa_lock);
  2979. break;
  2980. }
  2981. }
  2982. lip = xfs_trans_ail_cursor_next(ailp, &cur);
  2983. }
  2984. xfs_trans_ail_cursor_done(ailp, &cur);
  2985. spin_unlock(&ailp->xa_lock);
  2986. return 0;
  2987. }
  2988. /*
  2989. * This routine is called when an inode create format structure is found in a
  2990. * committed transaction in the log. It's purpose is to initialise the inodes
  2991. * being allocated on disk. This requires us to get inode cluster buffers that
  2992. * match the range to be intialised, stamped with inode templates and written
  2993. * by delayed write so that subsequent modifications will hit the cached buffer
  2994. * and only need writing out at the end of recovery.
  2995. */
  2996. STATIC int
  2997. xlog_recover_do_icreate_pass2(
  2998. struct xlog *log,
  2999. struct list_head *buffer_list,
  3000. xlog_recover_item_t *item)
  3001. {
  3002. struct xfs_mount *mp = log->l_mp;
  3003. struct xfs_icreate_log *icl;
  3004. xfs_agnumber_t agno;
  3005. xfs_agblock_t agbno;
  3006. unsigned int count;
  3007. unsigned int isize;
  3008. xfs_agblock_t length;
  3009. icl = (struct xfs_icreate_log *)item->ri_buf[0].i_addr;
  3010. if (icl->icl_type != XFS_LI_ICREATE) {
  3011. xfs_warn(log->l_mp, "xlog_recover_do_icreate_trans: bad type");
  3012. return EINVAL;
  3013. }
  3014. if (icl->icl_size != 1) {
  3015. xfs_warn(log->l_mp, "xlog_recover_do_icreate_trans: bad icl size");
  3016. return EINVAL;
  3017. }
  3018. agno = be32_to_cpu(icl->icl_ag);
  3019. if (agno >= mp->m_sb.sb_agcount) {
  3020. xfs_warn(log->l_mp, "xlog_recover_do_icreate_trans: bad agno");
  3021. return EINVAL;
  3022. }
  3023. agbno = be32_to_cpu(icl->icl_agbno);
  3024. if (!agbno || agbno == NULLAGBLOCK || agbno >= mp->m_sb.sb_agblocks) {
  3025. xfs_warn(log->l_mp, "xlog_recover_do_icreate_trans: bad agbno");
  3026. return EINVAL;
  3027. }
  3028. isize = be32_to_cpu(icl->icl_isize);
  3029. if (isize != mp->m_sb.sb_inodesize) {
  3030. xfs_warn(log->l_mp, "xlog_recover_do_icreate_trans: bad isize");
  3031. return EINVAL;
  3032. }
  3033. count = be32_to_cpu(icl->icl_count);
  3034. if (!count) {
  3035. xfs_warn(log->l_mp, "xlog_recover_do_icreate_trans: bad count");
  3036. return EINVAL;
  3037. }
  3038. length = be32_to_cpu(icl->icl_length);
  3039. if (!length || length >= mp->m_sb.sb_agblocks) {
  3040. xfs_warn(log->l_mp, "xlog_recover_do_icreate_trans: bad length");
  3041. return EINVAL;
  3042. }
  3043. /* existing allocation is fixed value */
  3044. ASSERT(count == XFS_IALLOC_INODES(mp));
  3045. ASSERT(length == XFS_IALLOC_BLOCKS(mp));
  3046. if (count != XFS_IALLOC_INODES(mp) ||
  3047. length != XFS_IALLOC_BLOCKS(mp)) {
  3048. xfs_warn(log->l_mp, "xlog_recover_do_icreate_trans: bad count 2");
  3049. return EINVAL;
  3050. }
  3051. /*
  3052. * Inode buffers can be freed. Do not replay the inode initialisation as
  3053. * we could be overwriting something written after this inode buffer was
  3054. * cancelled.
  3055. *
  3056. * XXX: we need to iterate all buffers and only init those that are not
  3057. * cancelled. I think that a more fine grained factoring of
  3058. * xfs_ialloc_inode_init may be appropriate here to enable this to be
  3059. * done easily.
  3060. */
  3061. if (xlog_check_buffer_cancelled(log,
  3062. XFS_AGB_TO_DADDR(mp, agno, agbno), length, 0))
  3063. return 0;
  3064. xfs_ialloc_inode_init(mp, NULL, buffer_list, agno, agbno, length,
  3065. be32_to_cpu(icl->icl_gen));
  3066. return 0;
  3067. }
  3068. /*
  3069. * Free up any resources allocated by the transaction
  3070. *
  3071. * Remember that EFIs, EFDs, and IUNLINKs are handled later.
  3072. */
  3073. STATIC void
  3074. xlog_recover_free_trans(
  3075. struct xlog_recover *trans)
  3076. {
  3077. xlog_recover_item_t *item, *n;
  3078. int i;
  3079. list_for_each_entry_safe(item, n, &trans->r_itemq, ri_list) {
  3080. /* Free the regions in the item. */
  3081. list_del(&item->ri_list);
  3082. for (i = 0; i < item->ri_cnt; i++)
  3083. kmem_free(item->ri_buf[i].i_addr);
  3084. /* Free the item itself */
  3085. kmem_free(item->ri_buf);
  3086. kmem_free(item);
  3087. }
  3088. /* Free the transaction recover structure */
  3089. kmem_free(trans);
  3090. }
  3091. STATIC void
  3092. xlog_recover_buffer_ra_pass2(
  3093. struct xlog *log,
  3094. struct xlog_recover_item *item)
  3095. {
  3096. struct xfs_buf_log_format *buf_f = item->ri_buf[0].i_addr;
  3097. struct xfs_mount *mp = log->l_mp;
  3098. if (xlog_peek_buffer_cancelled(log, buf_f->blf_blkno,
  3099. buf_f->blf_len, buf_f->blf_flags)) {
  3100. return;
  3101. }
  3102. xfs_buf_readahead(mp->m_ddev_targp, buf_f->blf_blkno,
  3103. buf_f->blf_len, NULL);
  3104. }
  3105. STATIC void
  3106. xlog_recover_inode_ra_pass2(
  3107. struct xlog *log,
  3108. struct xlog_recover_item *item)
  3109. {
  3110. struct xfs_inode_log_format ilf_buf;
  3111. struct xfs_inode_log_format *ilfp;
  3112. struct xfs_mount *mp = log->l_mp;
  3113. int error;
  3114. if (item->ri_buf[0].i_len == sizeof(struct xfs_inode_log_format)) {
  3115. ilfp = item->ri_buf[0].i_addr;
  3116. } else {
  3117. ilfp = &ilf_buf;
  3118. memset(ilfp, 0, sizeof(*ilfp));
  3119. error = xfs_inode_item_format_convert(&item->ri_buf[0], ilfp);
  3120. if (error)
  3121. return;
  3122. }
  3123. if (xlog_peek_buffer_cancelled(log, ilfp->ilf_blkno, ilfp->ilf_len, 0))
  3124. return;
  3125. xfs_buf_readahead(mp->m_ddev_targp, ilfp->ilf_blkno,
  3126. ilfp->ilf_len, &xfs_inode_buf_ra_ops);
  3127. }
  3128. STATIC void
  3129. xlog_recover_dquot_ra_pass2(
  3130. struct xlog *log,
  3131. struct xlog_recover_item *item)
  3132. {
  3133. struct xfs_mount *mp = log->l_mp;
  3134. struct xfs_disk_dquot *recddq;
  3135. struct xfs_dq_logformat *dq_f;
  3136. uint type;
  3137. if (mp->m_qflags == 0)
  3138. return;
  3139. recddq = item->ri_buf[1].i_addr;
  3140. if (recddq == NULL)
  3141. return;
  3142. if (item->ri_buf[1].i_len < sizeof(struct xfs_disk_dquot))
  3143. return;
  3144. type = recddq->d_flags & (XFS_DQ_USER | XFS_DQ_PROJ | XFS_DQ_GROUP);
  3145. ASSERT(type);
  3146. if (log->l_quotaoffs_flag & type)
  3147. return;
  3148. dq_f = item->ri_buf[0].i_addr;
  3149. ASSERT(dq_f);
  3150. ASSERT(dq_f->qlf_len == 1);
  3151. xfs_buf_readahead(mp->m_ddev_targp, dq_f->qlf_blkno,
  3152. XFS_FSB_TO_BB(mp, dq_f->qlf_len), NULL);
  3153. }
  3154. STATIC void
  3155. xlog_recover_ra_pass2(
  3156. struct xlog *log,
  3157. struct xlog_recover_item *item)
  3158. {
  3159. switch (ITEM_TYPE(item)) {
  3160. case XFS_LI_BUF:
  3161. xlog_recover_buffer_ra_pass2(log, item);
  3162. break;
  3163. case XFS_LI_INODE:
  3164. xlog_recover_inode_ra_pass2(log, item);
  3165. break;
  3166. case XFS_LI_DQUOT:
  3167. xlog_recover_dquot_ra_pass2(log, item);
  3168. break;
  3169. case XFS_LI_EFI:
  3170. case XFS_LI_EFD:
  3171. case XFS_LI_QUOTAOFF:
  3172. default:
  3173. break;
  3174. }
  3175. }
  3176. STATIC int
  3177. xlog_recover_commit_pass1(
  3178. struct xlog *log,
  3179. struct xlog_recover *trans,
  3180. struct xlog_recover_item *item)
  3181. {
  3182. trace_xfs_log_recover_item_recover(log, trans, item, XLOG_RECOVER_PASS1);
  3183. switch (ITEM_TYPE(item)) {
  3184. case XFS_LI_BUF:
  3185. return xlog_recover_buffer_pass1(log, item);
  3186. case XFS_LI_QUOTAOFF:
  3187. return xlog_recover_quotaoff_pass1(log, item);
  3188. case XFS_LI_INODE:
  3189. case XFS_LI_EFI:
  3190. case XFS_LI_EFD:
  3191. case XFS_LI_DQUOT:
  3192. case XFS_LI_ICREATE:
  3193. /* nothing to do in pass 1 */
  3194. return 0;
  3195. default:
  3196. xfs_warn(log->l_mp, "%s: invalid item type (%d)",
  3197. __func__, ITEM_TYPE(item));
  3198. ASSERT(0);
  3199. return XFS_ERROR(EIO);
  3200. }
  3201. }
  3202. STATIC int
  3203. xlog_recover_commit_pass2(
  3204. struct xlog *log,
  3205. struct xlog_recover *trans,
  3206. struct list_head *buffer_list,
  3207. struct xlog_recover_item *item)
  3208. {
  3209. trace_xfs_log_recover_item_recover(log, trans, item, XLOG_RECOVER_PASS2);
  3210. switch (ITEM_TYPE(item)) {
  3211. case XFS_LI_BUF:
  3212. return xlog_recover_buffer_pass2(log, buffer_list, item,
  3213. trans->r_lsn);
  3214. case XFS_LI_INODE:
  3215. return xlog_recover_inode_pass2(log, buffer_list, item,
  3216. trans->r_lsn);
  3217. case XFS_LI_EFI:
  3218. return xlog_recover_efi_pass2(log, item, trans->r_lsn);
  3219. case XFS_LI_EFD:
  3220. return xlog_recover_efd_pass2(log, item);
  3221. case XFS_LI_DQUOT:
  3222. return xlog_recover_dquot_pass2(log, buffer_list, item,
  3223. trans->r_lsn);
  3224. case XFS_LI_ICREATE:
  3225. return xlog_recover_do_icreate_pass2(log, buffer_list, item);
  3226. case XFS_LI_QUOTAOFF:
  3227. /* nothing to do in pass2 */
  3228. return 0;
  3229. default:
  3230. xfs_warn(log->l_mp, "%s: invalid item type (%d)",
  3231. __func__, ITEM_TYPE(item));
  3232. ASSERT(0);
  3233. return XFS_ERROR(EIO);
  3234. }
  3235. }
  3236. STATIC int
  3237. xlog_recover_items_pass2(
  3238. struct xlog *log,
  3239. struct xlog_recover *trans,
  3240. struct list_head *buffer_list,
  3241. struct list_head *item_list)
  3242. {
  3243. struct xlog_recover_item *item;
  3244. int error = 0;
  3245. list_for_each_entry(item, item_list, ri_list) {
  3246. error = xlog_recover_commit_pass2(log, trans,
  3247. buffer_list, item);
  3248. if (error)
  3249. return error;
  3250. }
  3251. return error;
  3252. }
  3253. /*
  3254. * Perform the transaction.
  3255. *
  3256. * If the transaction modifies a buffer or inode, do it now. Otherwise,
  3257. * EFIs and EFDs get queued up by adding entries into the AIL for them.
  3258. */
  3259. STATIC int
  3260. xlog_recover_commit_trans(
  3261. struct xlog *log,
  3262. struct xlog_recover *trans,
  3263. int pass)
  3264. {
  3265. int error = 0;
  3266. int error2;
  3267. int items_queued = 0;
  3268. struct xlog_recover_item *item;
  3269. struct xlog_recover_item *next;
  3270. LIST_HEAD (buffer_list);
  3271. LIST_HEAD (ra_list);
  3272. LIST_HEAD (done_list);
  3273. #define XLOG_RECOVER_COMMIT_QUEUE_MAX 100
  3274. hlist_del(&trans->r_list);
  3275. error = xlog_recover_reorder_trans(log, trans, pass);
  3276. if (error)
  3277. return error;
  3278. list_for_each_entry_safe(item, next, &trans->r_itemq, ri_list) {
  3279. switch (pass) {
  3280. case XLOG_RECOVER_PASS1:
  3281. error = xlog_recover_commit_pass1(log, trans, item);
  3282. break;
  3283. case XLOG_RECOVER_PASS2:
  3284. xlog_recover_ra_pass2(log, item);
  3285. list_move_tail(&item->ri_list, &ra_list);
  3286. items_queued++;
  3287. if (items_queued >= XLOG_RECOVER_COMMIT_QUEUE_MAX) {
  3288. error = xlog_recover_items_pass2(log, trans,
  3289. &buffer_list, &ra_list);
  3290. list_splice_tail_init(&ra_list, &done_list);
  3291. items_queued = 0;
  3292. }
  3293. break;
  3294. default:
  3295. ASSERT(0);
  3296. }
  3297. if (error)
  3298. goto out;
  3299. }
  3300. out:
  3301. if (!list_empty(&ra_list)) {
  3302. if (!error)
  3303. error = xlog_recover_items_pass2(log, trans,
  3304. &buffer_list, &ra_list);
  3305. list_splice_tail_init(&ra_list, &done_list);
  3306. }
  3307. if (!list_empty(&done_list))
  3308. list_splice_init(&done_list, &trans->r_itemq);
  3309. xlog_recover_free_trans(trans);
  3310. error2 = xfs_buf_delwri_submit(&buffer_list);
  3311. return error ? error : error2;
  3312. }
  3313. STATIC int
  3314. xlog_recover_unmount_trans(
  3315. struct xlog *log,
  3316. struct xlog_recover *trans)
  3317. {
  3318. /* Do nothing now */
  3319. xfs_warn(log->l_mp, "%s: Unmount LR", __func__);
  3320. return 0;
  3321. }
  3322. /*
  3323. * There are two valid states of the r_state field. 0 indicates that the
  3324. * transaction structure is in a normal state. We have either seen the
  3325. * start of the transaction or the last operation we added was not a partial
  3326. * operation. If the last operation we added to the transaction was a
  3327. * partial operation, we need to mark r_state with XLOG_WAS_CONT_TRANS.
  3328. *
  3329. * NOTE: skip LRs with 0 data length.
  3330. */
  3331. STATIC int
  3332. xlog_recover_process_data(
  3333. struct xlog *log,
  3334. struct hlist_head rhash[],
  3335. struct xlog_rec_header *rhead,
  3336. xfs_caddr_t dp,
  3337. int pass)
  3338. {
  3339. xfs_caddr_t lp;
  3340. int num_logops;
  3341. xlog_op_header_t *ohead;
  3342. xlog_recover_t *trans;
  3343. xlog_tid_t tid;
  3344. int error;
  3345. unsigned long hash;
  3346. uint flags;
  3347. lp = dp + be32_to_cpu(rhead->h_len);
  3348. num_logops = be32_to_cpu(rhead->h_num_logops);
  3349. /* check the log format matches our own - else we can't recover */
  3350. if (xlog_header_check_recover(log->l_mp, rhead))
  3351. return (XFS_ERROR(EIO));
  3352. while ((dp < lp) && num_logops) {
  3353. ASSERT(dp + sizeof(xlog_op_header_t) <= lp);
  3354. ohead = (xlog_op_header_t *)dp;
  3355. dp += sizeof(xlog_op_header_t);
  3356. if (ohead->oh_clientid != XFS_TRANSACTION &&
  3357. ohead->oh_clientid != XFS_LOG) {
  3358. xfs_warn(log->l_mp, "%s: bad clientid 0x%x",
  3359. __func__, ohead->oh_clientid);
  3360. ASSERT(0);
  3361. return (XFS_ERROR(EIO));
  3362. }
  3363. tid = be32_to_cpu(ohead->oh_tid);
  3364. hash = XLOG_RHASH(tid);
  3365. trans = xlog_recover_find_tid(&rhash[hash], tid);
  3366. if (trans == NULL) { /* not found; add new tid */
  3367. if (ohead->oh_flags & XLOG_START_TRANS)
  3368. xlog_recover_new_tid(&rhash[hash], tid,
  3369. be64_to_cpu(rhead->h_lsn));
  3370. } else {
  3371. if (dp + be32_to_cpu(ohead->oh_len) > lp) {
  3372. xfs_warn(log->l_mp, "%s: bad length 0x%x",
  3373. __func__, be32_to_cpu(ohead->oh_len));
  3374. WARN_ON(1);
  3375. return (XFS_ERROR(EIO));
  3376. }
  3377. flags = ohead->oh_flags & ~XLOG_END_TRANS;
  3378. if (flags & XLOG_WAS_CONT_TRANS)
  3379. flags &= ~XLOG_CONTINUE_TRANS;
  3380. switch (flags) {
  3381. case XLOG_COMMIT_TRANS:
  3382. error = xlog_recover_commit_trans(log,
  3383. trans, pass);
  3384. break;
  3385. case XLOG_UNMOUNT_TRANS:
  3386. error = xlog_recover_unmount_trans(log, trans);
  3387. break;
  3388. case XLOG_WAS_CONT_TRANS:
  3389. error = xlog_recover_add_to_cont_trans(log,
  3390. trans, dp,
  3391. be32_to_cpu(ohead->oh_len));
  3392. break;
  3393. case XLOG_START_TRANS:
  3394. xfs_warn(log->l_mp, "%s: bad transaction",
  3395. __func__);
  3396. ASSERT(0);
  3397. error = XFS_ERROR(EIO);
  3398. break;
  3399. case 0:
  3400. case XLOG_CONTINUE_TRANS:
  3401. error = xlog_recover_add_to_trans(log, trans,
  3402. dp, be32_to_cpu(ohead->oh_len));
  3403. break;
  3404. default:
  3405. xfs_warn(log->l_mp, "%s: bad flag 0x%x",
  3406. __func__, flags);
  3407. ASSERT(0);
  3408. error = XFS_ERROR(EIO);
  3409. break;
  3410. }
  3411. if (error)
  3412. return error;
  3413. }
  3414. dp += be32_to_cpu(ohead->oh_len);
  3415. num_logops--;
  3416. }
  3417. return 0;
  3418. }
  3419. /*
  3420. * Process an extent free intent item that was recovered from
  3421. * the log. We need to free the extents that it describes.
  3422. */
  3423. STATIC int
  3424. xlog_recover_process_efi(
  3425. xfs_mount_t *mp,
  3426. xfs_efi_log_item_t *efip)
  3427. {
  3428. xfs_efd_log_item_t *efdp;
  3429. xfs_trans_t *tp;
  3430. int i;
  3431. int error = 0;
  3432. xfs_extent_t *extp;
  3433. xfs_fsblock_t startblock_fsb;
  3434. ASSERT(!test_bit(XFS_EFI_RECOVERED, &efip->efi_flags));
  3435. /*
  3436. * First check the validity of the extents described by the
  3437. * EFI. If any are bad, then assume that all are bad and
  3438. * just toss the EFI.
  3439. */
  3440. for (i = 0; i < efip->efi_format.efi_nextents; i++) {
  3441. extp = &(efip->efi_format.efi_extents[i]);
  3442. startblock_fsb = XFS_BB_TO_FSB(mp,
  3443. XFS_FSB_TO_DADDR(mp, extp->ext_start));
  3444. if ((startblock_fsb == 0) ||
  3445. (extp->ext_len == 0) ||
  3446. (startblock_fsb >= mp->m_sb.sb_dblocks) ||
  3447. (extp->ext_len >= mp->m_sb.sb_agblocks)) {
  3448. /*
  3449. * This will pull the EFI from the AIL and
  3450. * free the memory associated with it.
  3451. */
  3452. set_bit(XFS_EFI_RECOVERED, &efip->efi_flags);
  3453. xfs_efi_release(efip, efip->efi_format.efi_nextents);
  3454. return XFS_ERROR(EIO);
  3455. }
  3456. }
  3457. tp = xfs_trans_alloc(mp, 0);
  3458. error = xfs_trans_reserve(tp, &M_RES(mp)->tr_itruncate, 0, 0);
  3459. if (error)
  3460. goto abort_error;
  3461. efdp = xfs_trans_get_efd(tp, efip, efip->efi_format.efi_nextents);
  3462. for (i = 0; i < efip->efi_format.efi_nextents; i++) {
  3463. extp = &(efip->efi_format.efi_extents[i]);
  3464. error = xfs_free_extent(tp, extp->ext_start, extp->ext_len);
  3465. if (error)
  3466. goto abort_error;
  3467. xfs_trans_log_efd_extent(tp, efdp, extp->ext_start,
  3468. extp->ext_len);
  3469. }
  3470. set_bit(XFS_EFI_RECOVERED, &efip->efi_flags);
  3471. error = xfs_trans_commit(tp, 0);
  3472. return error;
  3473. abort_error:
  3474. xfs_trans_cancel(tp, XFS_TRANS_ABORT);
  3475. return error;
  3476. }
  3477. /*
  3478. * When this is called, all of the EFIs which did not have
  3479. * corresponding EFDs should be in the AIL. What we do now
  3480. * is free the extents associated with each one.
  3481. *
  3482. * Since we process the EFIs in normal transactions, they
  3483. * will be removed at some point after the commit. This prevents
  3484. * us from just walking down the list processing each one.
  3485. * We'll use a flag in the EFI to skip those that we've already
  3486. * processed and use the AIL iteration mechanism's generation
  3487. * count to try to speed this up at least a bit.
  3488. *
  3489. * When we start, we know that the EFIs are the only things in
  3490. * the AIL. As we process them, however, other items are added
  3491. * to the AIL. Since everything added to the AIL must come after
  3492. * everything already in the AIL, we stop processing as soon as
  3493. * we see something other than an EFI in the AIL.
  3494. */
  3495. STATIC int
  3496. xlog_recover_process_efis(
  3497. struct xlog *log)
  3498. {
  3499. xfs_log_item_t *lip;
  3500. xfs_efi_log_item_t *efip;
  3501. int error = 0;
  3502. struct xfs_ail_cursor cur;
  3503. struct xfs_ail *ailp;
  3504. ailp = log->l_ailp;
  3505. spin_lock(&ailp->xa_lock);
  3506. lip = xfs_trans_ail_cursor_first(ailp, &cur, 0);
  3507. while (lip != NULL) {
  3508. /*
  3509. * We're done when we see something other than an EFI.
  3510. * There should be no EFIs left in the AIL now.
  3511. */
  3512. if (lip->li_type != XFS_LI_EFI) {
  3513. #ifdef DEBUG
  3514. for (; lip; lip = xfs_trans_ail_cursor_next(ailp, &cur))
  3515. ASSERT(lip->li_type != XFS_LI_EFI);
  3516. #endif
  3517. break;
  3518. }
  3519. /*
  3520. * Skip EFIs that we've already processed.
  3521. */
  3522. efip = (xfs_efi_log_item_t *)lip;
  3523. if (test_bit(XFS_EFI_RECOVERED, &efip->efi_flags)) {
  3524. lip = xfs_trans_ail_cursor_next(ailp, &cur);
  3525. continue;
  3526. }
  3527. spin_unlock(&ailp->xa_lock);
  3528. error = xlog_recover_process_efi(log->l_mp, efip);
  3529. spin_lock(&ailp->xa_lock);
  3530. if (error)
  3531. goto out;
  3532. lip = xfs_trans_ail_cursor_next(ailp, &cur);
  3533. }
  3534. out:
  3535. xfs_trans_ail_cursor_done(ailp, &cur);
  3536. spin_unlock(&ailp->xa_lock);
  3537. return error;
  3538. }
  3539. /*
  3540. * This routine performs a transaction to null out a bad inode pointer
  3541. * in an agi unlinked inode hash bucket.
  3542. */
  3543. STATIC void
  3544. xlog_recover_clear_agi_bucket(
  3545. xfs_mount_t *mp,
  3546. xfs_agnumber_t agno,
  3547. int bucket)
  3548. {
  3549. xfs_trans_t *tp;
  3550. xfs_agi_t *agi;
  3551. xfs_buf_t *agibp;
  3552. int offset;
  3553. int error;
  3554. tp = xfs_trans_alloc(mp, XFS_TRANS_CLEAR_AGI_BUCKET);
  3555. error = xfs_trans_reserve(tp, &M_RES(mp)->tr_clearagi, 0, 0);
  3556. if (error)
  3557. goto out_abort;
  3558. error = xfs_read_agi(mp, tp, agno, &agibp);
  3559. if (error)
  3560. goto out_abort;
  3561. agi = XFS_BUF_TO_AGI(agibp);
  3562. agi->agi_unlinked[bucket] = cpu_to_be32(NULLAGINO);
  3563. offset = offsetof(xfs_agi_t, agi_unlinked) +
  3564. (sizeof(xfs_agino_t) * bucket);
  3565. xfs_trans_log_buf(tp, agibp, offset,
  3566. (offset + sizeof(xfs_agino_t) - 1));
  3567. error = xfs_trans_commit(tp, 0);
  3568. if (error)
  3569. goto out_error;
  3570. return;
  3571. out_abort:
  3572. xfs_trans_cancel(tp, XFS_TRANS_ABORT);
  3573. out_error:
  3574. xfs_warn(mp, "%s: failed to clear agi %d. Continuing.", __func__, agno);
  3575. return;
  3576. }
  3577. STATIC xfs_agino_t
  3578. xlog_recover_process_one_iunlink(
  3579. struct xfs_mount *mp,
  3580. xfs_agnumber_t agno,
  3581. xfs_agino_t agino,
  3582. int bucket)
  3583. {
  3584. struct xfs_buf *ibp;
  3585. struct xfs_dinode *dip;
  3586. struct xfs_inode *ip;
  3587. xfs_ino_t ino;
  3588. int error;
  3589. ino = XFS_AGINO_TO_INO(mp, agno, agino);
  3590. error = xfs_iget(mp, NULL, ino, 0, 0, &ip);
  3591. if (error)
  3592. goto fail;
  3593. /*
  3594. * Get the on disk inode to find the next inode in the bucket.
  3595. */
  3596. error = xfs_imap_to_bp(mp, NULL, &ip->i_imap, &dip, &ibp, 0, 0);
  3597. if (error)
  3598. goto fail_iput;
  3599. ASSERT(ip->i_d.di_nlink == 0);
  3600. ASSERT(ip->i_d.di_mode != 0);
  3601. /* setup for the next pass */
  3602. agino = be32_to_cpu(dip->di_next_unlinked);
  3603. xfs_buf_relse(ibp);
  3604. /*
  3605. * Prevent any DMAPI event from being sent when the reference on
  3606. * the inode is dropped.
  3607. */
  3608. ip->i_d.di_dmevmask = 0;
  3609. IRELE(ip);
  3610. return agino;
  3611. fail_iput:
  3612. IRELE(ip);
  3613. fail:
  3614. /*
  3615. * We can't read in the inode this bucket points to, or this inode
  3616. * is messed up. Just ditch this bucket of inodes. We will lose
  3617. * some inodes and space, but at least we won't hang.
  3618. *
  3619. * Call xlog_recover_clear_agi_bucket() to perform a transaction to
  3620. * clear the inode pointer in the bucket.
  3621. */
  3622. xlog_recover_clear_agi_bucket(mp, agno, bucket);
  3623. return NULLAGINO;
  3624. }
  3625. /*
  3626. * xlog_iunlink_recover
  3627. *
  3628. * This is called during recovery to process any inodes which
  3629. * we unlinked but not freed when the system crashed. These
  3630. * inodes will be on the lists in the AGI blocks. What we do
  3631. * here is scan all the AGIs and fully truncate and free any
  3632. * inodes found on the lists. Each inode is removed from the
  3633. * lists when it has been fully truncated and is freed. The
  3634. * freeing of the inode and its removal from the list must be
  3635. * atomic.
  3636. */
  3637. STATIC void
  3638. xlog_recover_process_iunlinks(
  3639. struct xlog *log)
  3640. {
  3641. xfs_mount_t *mp;
  3642. xfs_agnumber_t agno;
  3643. xfs_agi_t *agi;
  3644. xfs_buf_t *agibp;
  3645. xfs_agino_t agino;
  3646. int bucket;
  3647. int error;
  3648. uint mp_dmevmask;
  3649. mp = log->l_mp;
  3650. /*
  3651. * Prevent any DMAPI event from being sent while in this function.
  3652. */
  3653. mp_dmevmask = mp->m_dmevmask;
  3654. mp->m_dmevmask = 0;
  3655. for (agno = 0; agno < mp->m_sb.sb_agcount; agno++) {
  3656. /*
  3657. * Find the agi for this ag.
  3658. */
  3659. error = xfs_read_agi(mp, NULL, agno, &agibp);
  3660. if (error) {
  3661. /*
  3662. * AGI is b0rked. Don't process it.
  3663. *
  3664. * We should probably mark the filesystem as corrupt
  3665. * after we've recovered all the ag's we can....
  3666. */
  3667. continue;
  3668. }
  3669. /*
  3670. * Unlock the buffer so that it can be acquired in the normal
  3671. * course of the transaction to truncate and free each inode.
  3672. * Because we are not racing with anyone else here for the AGI
  3673. * buffer, we don't even need to hold it locked to read the
  3674. * initial unlinked bucket entries out of the buffer. We keep
  3675. * buffer reference though, so that it stays pinned in memory
  3676. * while we need the buffer.
  3677. */
  3678. agi = XFS_BUF_TO_AGI(agibp);
  3679. xfs_buf_unlock(agibp);
  3680. for (bucket = 0; bucket < XFS_AGI_UNLINKED_BUCKETS; bucket++) {
  3681. agino = be32_to_cpu(agi->agi_unlinked[bucket]);
  3682. while (agino != NULLAGINO) {
  3683. agino = xlog_recover_process_one_iunlink(mp,
  3684. agno, agino, bucket);
  3685. }
  3686. }
  3687. xfs_buf_rele(agibp);
  3688. }
  3689. mp->m_dmevmask = mp_dmevmask;
  3690. }
  3691. /*
  3692. * Upack the log buffer data and crc check it. If the check fails, issue a
  3693. * warning if and only if the CRC in the header is non-zero. This makes the
  3694. * check an advisory warning, and the zero CRC check will prevent failure
  3695. * warnings from being emitted when upgrading the kernel from one that does not
  3696. * add CRCs by default.
  3697. *
  3698. * When filesystems are CRC enabled, this CRC mismatch becomes a fatal log
  3699. * corruption failure
  3700. */
  3701. STATIC int
  3702. xlog_unpack_data_crc(
  3703. struct xlog_rec_header *rhead,
  3704. xfs_caddr_t dp,
  3705. struct xlog *log)
  3706. {
  3707. __le32 crc;
  3708. crc = xlog_cksum(log, rhead, dp, be32_to_cpu(rhead->h_len));
  3709. if (crc != rhead->h_crc) {
  3710. if (rhead->h_crc || xfs_sb_version_hascrc(&log->l_mp->m_sb)) {
  3711. xfs_alert(log->l_mp,
  3712. "log record CRC mismatch: found 0x%x, expected 0x%x.\n",
  3713. le32_to_cpu(rhead->h_crc),
  3714. le32_to_cpu(crc));
  3715. xfs_hex_dump(dp, 32);
  3716. }
  3717. /*
  3718. * If we've detected a log record corruption, then we can't
  3719. * recover past this point. Abort recovery if we are enforcing
  3720. * CRC protection by punting an error back up the stack.
  3721. */
  3722. if (xfs_sb_version_hascrc(&log->l_mp->m_sb))
  3723. return EFSCORRUPTED;
  3724. }
  3725. return 0;
  3726. }
  3727. STATIC int
  3728. xlog_unpack_data(
  3729. struct xlog_rec_header *rhead,
  3730. xfs_caddr_t dp,
  3731. struct xlog *log)
  3732. {
  3733. int i, j, k;
  3734. int error;
  3735. error = xlog_unpack_data_crc(rhead, dp, log);
  3736. if (error)
  3737. return error;
  3738. for (i = 0; i < BTOBB(be32_to_cpu(rhead->h_len)) &&
  3739. i < (XLOG_HEADER_CYCLE_SIZE / BBSIZE); i++) {
  3740. *(__be32 *)dp = *(__be32 *)&rhead->h_cycle_data[i];
  3741. dp += BBSIZE;
  3742. }
  3743. if (xfs_sb_version_haslogv2(&log->l_mp->m_sb)) {
  3744. xlog_in_core_2_t *xhdr = (xlog_in_core_2_t *)rhead;
  3745. for ( ; i < BTOBB(be32_to_cpu(rhead->h_len)); i++) {
  3746. j = i / (XLOG_HEADER_CYCLE_SIZE / BBSIZE);
  3747. k = i % (XLOG_HEADER_CYCLE_SIZE / BBSIZE);
  3748. *(__be32 *)dp = xhdr[j].hic_xheader.xh_cycle_data[k];
  3749. dp += BBSIZE;
  3750. }
  3751. }
  3752. return 0;
  3753. }
  3754. STATIC int
  3755. xlog_valid_rec_header(
  3756. struct xlog *log,
  3757. struct xlog_rec_header *rhead,
  3758. xfs_daddr_t blkno)
  3759. {
  3760. int hlen;
  3761. if (unlikely(rhead->h_magicno != cpu_to_be32(XLOG_HEADER_MAGIC_NUM))) {
  3762. XFS_ERROR_REPORT("xlog_valid_rec_header(1)",
  3763. XFS_ERRLEVEL_LOW, log->l_mp);
  3764. return XFS_ERROR(EFSCORRUPTED);
  3765. }
  3766. if (unlikely(
  3767. (!rhead->h_version ||
  3768. (be32_to_cpu(rhead->h_version) & (~XLOG_VERSION_OKBITS))))) {
  3769. xfs_warn(log->l_mp, "%s: unrecognised log version (%d).",
  3770. __func__, be32_to_cpu(rhead->h_version));
  3771. return XFS_ERROR(EIO);
  3772. }
  3773. /* LR body must have data or it wouldn't have been written */
  3774. hlen = be32_to_cpu(rhead->h_len);
  3775. if (unlikely( hlen <= 0 || hlen > INT_MAX )) {
  3776. XFS_ERROR_REPORT("xlog_valid_rec_header(2)",
  3777. XFS_ERRLEVEL_LOW, log->l_mp);
  3778. return XFS_ERROR(EFSCORRUPTED);
  3779. }
  3780. if (unlikely( blkno > log->l_logBBsize || blkno > INT_MAX )) {
  3781. XFS_ERROR_REPORT("xlog_valid_rec_header(3)",
  3782. XFS_ERRLEVEL_LOW, log->l_mp);
  3783. return XFS_ERROR(EFSCORRUPTED);
  3784. }
  3785. return 0;
  3786. }
  3787. /*
  3788. * Read the log from tail to head and process the log records found.
  3789. * Handle the two cases where the tail and head are in the same cycle
  3790. * and where the active portion of the log wraps around the end of
  3791. * the physical log separately. The pass parameter is passed through
  3792. * to the routines called to process the data and is not looked at
  3793. * here.
  3794. */
  3795. STATIC int
  3796. xlog_do_recovery_pass(
  3797. struct xlog *log,
  3798. xfs_daddr_t head_blk,
  3799. xfs_daddr_t tail_blk,
  3800. int pass)
  3801. {
  3802. xlog_rec_header_t *rhead;
  3803. xfs_daddr_t blk_no;
  3804. xfs_caddr_t offset;
  3805. xfs_buf_t *hbp, *dbp;
  3806. int error = 0, h_size;
  3807. int bblks, split_bblks;
  3808. int hblks, split_hblks, wrapped_hblks;
  3809. struct hlist_head rhash[XLOG_RHASH_SIZE];
  3810. ASSERT(head_blk != tail_blk);
  3811. /*
  3812. * Read the header of the tail block and get the iclog buffer size from
  3813. * h_size. Use this to tell how many sectors make up the log header.
  3814. */
  3815. if (xfs_sb_version_haslogv2(&log->l_mp->m_sb)) {
  3816. /*
  3817. * When using variable length iclogs, read first sector of
  3818. * iclog header and extract the header size from it. Get a
  3819. * new hbp that is the correct size.
  3820. */
  3821. hbp = xlog_get_bp(log, 1);
  3822. if (!hbp)
  3823. return ENOMEM;
  3824. error = xlog_bread(log, tail_blk, 1, hbp, &offset);
  3825. if (error)
  3826. goto bread_err1;
  3827. rhead = (xlog_rec_header_t *)offset;
  3828. error = xlog_valid_rec_header(log, rhead, tail_blk);
  3829. if (error)
  3830. goto bread_err1;
  3831. h_size = be32_to_cpu(rhead->h_size);
  3832. if ((be32_to_cpu(rhead->h_version) & XLOG_VERSION_2) &&
  3833. (h_size > XLOG_HEADER_CYCLE_SIZE)) {
  3834. hblks = h_size / XLOG_HEADER_CYCLE_SIZE;
  3835. if (h_size % XLOG_HEADER_CYCLE_SIZE)
  3836. hblks++;
  3837. xlog_put_bp(hbp);
  3838. hbp = xlog_get_bp(log, hblks);
  3839. } else {
  3840. hblks = 1;
  3841. }
  3842. } else {
  3843. ASSERT(log->l_sectBBsize == 1);
  3844. hblks = 1;
  3845. hbp = xlog_get_bp(log, 1);
  3846. h_size = XLOG_BIG_RECORD_BSIZE;
  3847. }
  3848. if (!hbp)
  3849. return ENOMEM;
  3850. dbp = xlog_get_bp(log, BTOBB(h_size));
  3851. if (!dbp) {
  3852. xlog_put_bp(hbp);
  3853. return ENOMEM;
  3854. }
  3855. memset(rhash, 0, sizeof(rhash));
  3856. if (tail_blk <= head_blk) {
  3857. for (blk_no = tail_blk; blk_no < head_blk; ) {
  3858. error = xlog_bread(log, blk_no, hblks, hbp, &offset);
  3859. if (error)
  3860. goto bread_err2;
  3861. rhead = (xlog_rec_header_t *)offset;
  3862. error = xlog_valid_rec_header(log, rhead, blk_no);
  3863. if (error)
  3864. goto bread_err2;
  3865. /* blocks in data section */
  3866. bblks = (int)BTOBB(be32_to_cpu(rhead->h_len));
  3867. error = xlog_bread(log, blk_no + hblks, bblks, dbp,
  3868. &offset);
  3869. if (error)
  3870. goto bread_err2;
  3871. error = xlog_unpack_data(rhead, offset, log);
  3872. if (error)
  3873. goto bread_err2;
  3874. error = xlog_recover_process_data(log,
  3875. rhash, rhead, offset, pass);
  3876. if (error)
  3877. goto bread_err2;
  3878. blk_no += bblks + hblks;
  3879. }
  3880. } else {
  3881. /*
  3882. * Perform recovery around the end of the physical log.
  3883. * When the head is not on the same cycle number as the tail,
  3884. * we can't do a sequential recovery as above.
  3885. */
  3886. blk_no = tail_blk;
  3887. while (blk_no < log->l_logBBsize) {
  3888. /*
  3889. * Check for header wrapping around physical end-of-log
  3890. */
  3891. offset = hbp->b_addr;
  3892. split_hblks = 0;
  3893. wrapped_hblks = 0;
  3894. if (blk_no + hblks <= log->l_logBBsize) {
  3895. /* Read header in one read */
  3896. error = xlog_bread(log, blk_no, hblks, hbp,
  3897. &offset);
  3898. if (error)
  3899. goto bread_err2;
  3900. } else {
  3901. /* This LR is split across physical log end */
  3902. if (blk_no != log->l_logBBsize) {
  3903. /* some data before physical log end */
  3904. ASSERT(blk_no <= INT_MAX);
  3905. split_hblks = log->l_logBBsize - (int)blk_no;
  3906. ASSERT(split_hblks > 0);
  3907. error = xlog_bread(log, blk_no,
  3908. split_hblks, hbp,
  3909. &offset);
  3910. if (error)
  3911. goto bread_err2;
  3912. }
  3913. /*
  3914. * Note: this black magic still works with
  3915. * large sector sizes (non-512) only because:
  3916. * - we increased the buffer size originally
  3917. * by 1 sector giving us enough extra space
  3918. * for the second read;
  3919. * - the log start is guaranteed to be sector
  3920. * aligned;
  3921. * - we read the log end (LR header start)
  3922. * _first_, then the log start (LR header end)
  3923. * - order is important.
  3924. */
  3925. wrapped_hblks = hblks - split_hblks;
  3926. error = xlog_bread_offset(log, 0,
  3927. wrapped_hblks, hbp,
  3928. offset + BBTOB(split_hblks));
  3929. if (error)
  3930. goto bread_err2;
  3931. }
  3932. rhead = (xlog_rec_header_t *)offset;
  3933. error = xlog_valid_rec_header(log, rhead,
  3934. split_hblks ? blk_no : 0);
  3935. if (error)
  3936. goto bread_err2;
  3937. bblks = (int)BTOBB(be32_to_cpu(rhead->h_len));
  3938. blk_no += hblks;
  3939. /* Read in data for log record */
  3940. if (blk_no + bblks <= log->l_logBBsize) {
  3941. error = xlog_bread(log, blk_no, bblks, dbp,
  3942. &offset);
  3943. if (error)
  3944. goto bread_err2;
  3945. } else {
  3946. /* This log record is split across the
  3947. * physical end of log */
  3948. offset = dbp->b_addr;
  3949. split_bblks = 0;
  3950. if (blk_no != log->l_logBBsize) {
  3951. /* some data is before the physical
  3952. * end of log */
  3953. ASSERT(!wrapped_hblks);
  3954. ASSERT(blk_no <= INT_MAX);
  3955. split_bblks =
  3956. log->l_logBBsize - (int)blk_no;
  3957. ASSERT(split_bblks > 0);
  3958. error = xlog_bread(log, blk_no,
  3959. split_bblks, dbp,
  3960. &offset);
  3961. if (error)
  3962. goto bread_err2;
  3963. }
  3964. /*
  3965. * Note: this black magic still works with
  3966. * large sector sizes (non-512) only because:
  3967. * - we increased the buffer size originally
  3968. * by 1 sector giving us enough extra space
  3969. * for the second read;
  3970. * - the log start is guaranteed to be sector
  3971. * aligned;
  3972. * - we read the log end (LR header start)
  3973. * _first_, then the log start (LR header end)
  3974. * - order is important.
  3975. */
  3976. error = xlog_bread_offset(log, 0,
  3977. bblks - split_bblks, dbp,
  3978. offset + BBTOB(split_bblks));
  3979. if (error)
  3980. goto bread_err2;
  3981. }
  3982. error = xlog_unpack_data(rhead, offset, log);
  3983. if (error)
  3984. goto bread_err2;
  3985. error = xlog_recover_process_data(log, rhash,
  3986. rhead, offset, pass);
  3987. if (error)
  3988. goto bread_err2;
  3989. blk_no += bblks;
  3990. }
  3991. ASSERT(blk_no >= log->l_logBBsize);
  3992. blk_no -= log->l_logBBsize;
  3993. /* read first part of physical log */
  3994. while (blk_no < head_blk) {
  3995. error = xlog_bread(log, blk_no, hblks, hbp, &offset);
  3996. if (error)
  3997. goto bread_err2;
  3998. rhead = (xlog_rec_header_t *)offset;
  3999. error = xlog_valid_rec_header(log, rhead, blk_no);
  4000. if (error)
  4001. goto bread_err2;
  4002. bblks = (int)BTOBB(be32_to_cpu(rhead->h_len));
  4003. error = xlog_bread(log, blk_no+hblks, bblks, dbp,
  4004. &offset);
  4005. if (error)
  4006. goto bread_err2;
  4007. error = xlog_unpack_data(rhead, offset, log);
  4008. if (error)
  4009. goto bread_err2;
  4010. error = xlog_recover_process_data(log, rhash,
  4011. rhead, offset, pass);
  4012. if (error)
  4013. goto bread_err2;
  4014. blk_no += bblks + hblks;
  4015. }
  4016. }
  4017. bread_err2:
  4018. xlog_put_bp(dbp);
  4019. bread_err1:
  4020. xlog_put_bp(hbp);
  4021. return error;
  4022. }
  4023. /*
  4024. * Do the recovery of the log. We actually do this in two phases.
  4025. * The two passes are necessary in order to implement the function
  4026. * of cancelling a record written into the log. The first pass
  4027. * determines those things which have been cancelled, and the
  4028. * second pass replays log items normally except for those which
  4029. * have been cancelled. The handling of the replay and cancellations
  4030. * takes place in the log item type specific routines.
  4031. *
  4032. * The table of items which have cancel records in the log is allocated
  4033. * and freed at this level, since only here do we know when all of
  4034. * the log recovery has been completed.
  4035. */
  4036. STATIC int
  4037. xlog_do_log_recovery(
  4038. struct xlog *log,
  4039. xfs_daddr_t head_blk,
  4040. xfs_daddr_t tail_blk)
  4041. {
  4042. int error, i;
  4043. ASSERT(head_blk != tail_blk);
  4044. /*
  4045. * First do a pass to find all of the cancelled buf log items.
  4046. * Store them in the buf_cancel_table for use in the second pass.
  4047. */
  4048. log->l_buf_cancel_table = kmem_zalloc(XLOG_BC_TABLE_SIZE *
  4049. sizeof(struct list_head),
  4050. KM_SLEEP);
  4051. for (i = 0; i < XLOG_BC_TABLE_SIZE; i++)
  4052. INIT_LIST_HEAD(&log->l_buf_cancel_table[i]);
  4053. error = xlog_do_recovery_pass(log, head_blk, tail_blk,
  4054. XLOG_RECOVER_PASS1);
  4055. if (error != 0) {
  4056. kmem_free(log->l_buf_cancel_table);
  4057. log->l_buf_cancel_table = NULL;
  4058. return error;
  4059. }
  4060. /*
  4061. * Then do a second pass to actually recover the items in the log.
  4062. * When it is complete free the table of buf cancel items.
  4063. */
  4064. error = xlog_do_recovery_pass(log, head_blk, tail_blk,
  4065. XLOG_RECOVER_PASS2);
  4066. #ifdef DEBUG
  4067. if (!error) {
  4068. int i;
  4069. for (i = 0; i < XLOG_BC_TABLE_SIZE; i++)
  4070. ASSERT(list_empty(&log->l_buf_cancel_table[i]));
  4071. }
  4072. #endif /* DEBUG */
  4073. kmem_free(log->l_buf_cancel_table);
  4074. log->l_buf_cancel_table = NULL;
  4075. return error;
  4076. }
  4077. /*
  4078. * Do the actual recovery
  4079. */
  4080. STATIC int
  4081. xlog_do_recover(
  4082. struct xlog *log,
  4083. xfs_daddr_t head_blk,
  4084. xfs_daddr_t tail_blk)
  4085. {
  4086. int error;
  4087. xfs_buf_t *bp;
  4088. xfs_sb_t *sbp;
  4089. /*
  4090. * First replay the images in the log.
  4091. */
  4092. error = xlog_do_log_recovery(log, head_blk, tail_blk);
  4093. if (error)
  4094. return error;
  4095. /*
  4096. * If IO errors happened during recovery, bail out.
  4097. */
  4098. if (XFS_FORCED_SHUTDOWN(log->l_mp)) {
  4099. return (EIO);
  4100. }
  4101. /*
  4102. * We now update the tail_lsn since much of the recovery has completed
  4103. * and there may be space available to use. If there were no extent
  4104. * or iunlinks, we can free up the entire log and set the tail_lsn to
  4105. * be the last_sync_lsn. This was set in xlog_find_tail to be the
  4106. * lsn of the last known good LR on disk. If there are extent frees
  4107. * or iunlinks they will have some entries in the AIL; so we look at
  4108. * the AIL to determine how to set the tail_lsn.
  4109. */
  4110. xlog_assign_tail_lsn(log->l_mp);
  4111. /*
  4112. * Now that we've finished replaying all buffer and inode
  4113. * updates, re-read in the superblock and reverify it.
  4114. */
  4115. bp = xfs_getsb(log->l_mp, 0);
  4116. XFS_BUF_UNDONE(bp);
  4117. ASSERT(!(XFS_BUF_ISWRITE(bp)));
  4118. XFS_BUF_READ(bp);
  4119. XFS_BUF_UNASYNC(bp);
  4120. bp->b_ops = &xfs_sb_buf_ops;
  4121. xfsbdstrat(log->l_mp, bp);
  4122. error = xfs_buf_iowait(bp);
  4123. if (error) {
  4124. xfs_buf_ioerror_alert(bp, __func__);
  4125. ASSERT(0);
  4126. xfs_buf_relse(bp);
  4127. return error;
  4128. }
  4129. /* Convert superblock from on-disk format */
  4130. sbp = &log->l_mp->m_sb;
  4131. xfs_sb_from_disk(sbp, XFS_BUF_TO_SBP(bp));
  4132. ASSERT(sbp->sb_magicnum == XFS_SB_MAGIC);
  4133. ASSERT(xfs_sb_good_version(sbp));
  4134. xfs_buf_relse(bp);
  4135. /* We've re-read the superblock so re-initialize per-cpu counters */
  4136. xfs_icsb_reinit_counters(log->l_mp);
  4137. xlog_recover_check_summary(log);
  4138. /* Normal transactions can now occur */
  4139. log->l_flags &= ~XLOG_ACTIVE_RECOVERY;
  4140. return 0;
  4141. }
  4142. /*
  4143. * Perform recovery and re-initialize some log variables in xlog_find_tail.
  4144. *
  4145. * Return error or zero.
  4146. */
  4147. int
  4148. xlog_recover(
  4149. struct xlog *log)
  4150. {
  4151. xfs_daddr_t head_blk, tail_blk;
  4152. int error;
  4153. /* find the tail of the log */
  4154. if ((error = xlog_find_tail(log, &head_blk, &tail_blk)))
  4155. return error;
  4156. if (tail_blk != head_blk) {
  4157. /* There used to be a comment here:
  4158. *
  4159. * disallow recovery on read-only mounts. note -- mount
  4160. * checks for ENOSPC and turns it into an intelligent
  4161. * error message.
  4162. * ...but this is no longer true. Now, unless you specify
  4163. * NORECOVERY (in which case this function would never be
  4164. * called), we just go ahead and recover. We do this all
  4165. * under the vfs layer, so we can get away with it unless
  4166. * the device itself is read-only, in which case we fail.
  4167. */
  4168. if ((error = xfs_dev_is_read_only(log->l_mp, "recovery"))) {
  4169. return error;
  4170. }
  4171. /*
  4172. * Version 5 superblock log feature mask validation. We know the
  4173. * log is dirty so check if there are any unknown log features
  4174. * in what we need to recover. If there are unknown features
  4175. * (e.g. unsupported transactions, then simply reject the
  4176. * attempt at recovery before touching anything.
  4177. */
  4178. if (XFS_SB_VERSION_NUM(&log->l_mp->m_sb) == XFS_SB_VERSION_5 &&
  4179. xfs_sb_has_incompat_log_feature(&log->l_mp->m_sb,
  4180. XFS_SB_FEAT_INCOMPAT_LOG_UNKNOWN)) {
  4181. xfs_warn(log->l_mp,
  4182. "Superblock has unknown incompatible log features (0x%x) enabled.\n"
  4183. "The log can not be fully and/or safely recovered by this kernel.\n"
  4184. "Please recover the log on a kernel that supports the unknown features.",
  4185. (log->l_mp->m_sb.sb_features_log_incompat &
  4186. XFS_SB_FEAT_INCOMPAT_LOG_UNKNOWN));
  4187. return EINVAL;
  4188. }
  4189. xfs_notice(log->l_mp, "Starting recovery (logdev: %s)",
  4190. log->l_mp->m_logname ? log->l_mp->m_logname
  4191. : "internal");
  4192. error = xlog_do_recover(log, head_blk, tail_blk);
  4193. log->l_flags |= XLOG_RECOVERY_NEEDED;
  4194. }
  4195. return error;
  4196. }
  4197. /*
  4198. * In the first part of recovery we replay inodes and buffers and build
  4199. * up the list of extent free items which need to be processed. Here
  4200. * we process the extent free items and clean up the on disk unlinked
  4201. * inode lists. This is separated from the first part of recovery so
  4202. * that the root and real-time bitmap inodes can be read in from disk in
  4203. * between the two stages. This is necessary so that we can free space
  4204. * in the real-time portion of the file system.
  4205. */
  4206. int
  4207. xlog_recover_finish(
  4208. struct xlog *log)
  4209. {
  4210. /*
  4211. * Now we're ready to do the transactions needed for the
  4212. * rest of recovery. Start with completing all the extent
  4213. * free intent records and then process the unlinked inode
  4214. * lists. At this point, we essentially run in normal mode
  4215. * except that we're still performing recovery actions
  4216. * rather than accepting new requests.
  4217. */
  4218. if (log->l_flags & XLOG_RECOVERY_NEEDED) {
  4219. int error;
  4220. error = xlog_recover_process_efis(log);
  4221. if (error) {
  4222. xfs_alert(log->l_mp, "Failed to recover EFIs");
  4223. return error;
  4224. }
  4225. /*
  4226. * Sync the log to get all the EFIs out of the AIL.
  4227. * This isn't absolutely necessary, but it helps in
  4228. * case the unlink transactions would have problems
  4229. * pushing the EFIs out of the way.
  4230. */
  4231. xfs_log_force(log->l_mp, XFS_LOG_SYNC);
  4232. xlog_recover_process_iunlinks(log);
  4233. xlog_recover_check_summary(log);
  4234. xfs_notice(log->l_mp, "Ending recovery (logdev: %s)",
  4235. log->l_mp->m_logname ? log->l_mp->m_logname
  4236. : "internal");
  4237. log->l_flags &= ~XLOG_RECOVERY_NEEDED;
  4238. } else {
  4239. xfs_info(log->l_mp, "Ending clean mount");
  4240. }
  4241. return 0;
  4242. }
  4243. #if defined(DEBUG)
  4244. /*
  4245. * Read all of the agf and agi counters and check that they
  4246. * are consistent with the superblock counters.
  4247. */
  4248. void
  4249. xlog_recover_check_summary(
  4250. struct xlog *log)
  4251. {
  4252. xfs_mount_t *mp;
  4253. xfs_agf_t *agfp;
  4254. xfs_buf_t *agfbp;
  4255. xfs_buf_t *agibp;
  4256. xfs_agnumber_t agno;
  4257. __uint64_t freeblks;
  4258. __uint64_t itotal;
  4259. __uint64_t ifree;
  4260. int error;
  4261. mp = log->l_mp;
  4262. freeblks = 0LL;
  4263. itotal = 0LL;
  4264. ifree = 0LL;
  4265. for (agno = 0; agno < mp->m_sb.sb_agcount; agno++) {
  4266. error = xfs_read_agf(mp, NULL, agno, 0, &agfbp);
  4267. if (error) {
  4268. xfs_alert(mp, "%s agf read failed agno %d error %d",
  4269. __func__, agno, error);
  4270. } else {
  4271. agfp = XFS_BUF_TO_AGF(agfbp);
  4272. freeblks += be32_to_cpu(agfp->agf_freeblks) +
  4273. be32_to_cpu(agfp->agf_flcount);
  4274. xfs_buf_relse(agfbp);
  4275. }
  4276. error = xfs_read_agi(mp, NULL, agno, &agibp);
  4277. if (error) {
  4278. xfs_alert(mp, "%s agi read failed agno %d error %d",
  4279. __func__, agno, error);
  4280. } else {
  4281. struct xfs_agi *agi = XFS_BUF_TO_AGI(agibp);
  4282. itotal += be32_to_cpu(agi->agi_count);
  4283. ifree += be32_to_cpu(agi->agi_freecount);
  4284. xfs_buf_relse(agibp);
  4285. }
  4286. }
  4287. }
  4288. #endif /* DEBUG */