xfs_log_recover.c 110 KB

123456789101112131415161718192021222324252627282930313233343536373839404142434445464748495051525354555657585960616263646566676869707172737475767778798081828384858687888990919293949596979899100101102103104105106107108109110111112113114115116117118119120121122123124125126127128129130131132133134135136137138139140141142143144145146147148149150151152153154155156157158159160161162163164165166167168169170171172173174175176177178179180181182183184185186187188189190191192193194195196197198199200201202203204205206207208209210211212213214215216217218219220221222223224225226227228229230231232233234235236237238239240241242243244245246247248249250251252253254255256257258259260261262263264265266267268269270271272273274275276277278279280281282283284285286287288289290291292293294295296297298299300301302303304305306307308309310311312313314315316317318319320321322323324325326327328329330331332333334335336337338339340341342343344345346347348349350351352353354355356357358359360361362363364365366367368369370371372373374375376377378379380381382383384385386387388389390391392393394395396397398399400401402403404405406407408409410411412413414415416417418419420421422423424425426427428429430431432433434435436437438439440441442443444445446447448449450451452453454455456457458459460461462463464465466467468469470471472473474475476477478479480481482483484485486487488489490491492493494495496497498499500501502503504505506507508509510511512513514515516517518519520521522523524525526527528529530531532533534535536537538539540541542543544545546547548549550551552553554555556557558559560561562563564565566567568569570571572573574575576577578579580581582583584585586587588589590591592593594595596597598599600601602603604605606607608609610611612613614615616617618619620621622623624625626627628629630631632633634635636637638639640641642643644645646647648649650651652653654655656657658659660661662663664665666667668669670671672673674675676677678679680681682683684685686687688689690691692693694695696697698699700701702703704705706707708709710711712713714715716717718719720721722723724725726727728729730731732733734735736737738739740741742743744745746747748749750751752753754755756757758759760761762763764765766767768769770771772773774775776777778779780781782783784785786787788789790791792793794795796797798799800801802803804805806807808809810811812813814815816817818819820821822823824825826827828829830831832833834835836837838839840841842843844845846847848849850851852853854855856857858859860861862863864865866867868869870871872873874875876877878879880881882883884885886887888889890891892893894895896897898899900901902903904905906907908909910911912913914915916917918919920921922923924925926927928929930931932933934935936937938939940941942943944945946947948949950951952953954955956957958959960961962963964965966967968969970971972973974975976977978979980981982983984985986987988989990991992993994995996997998999100010011002100310041005100610071008100910101011101210131014101510161017101810191020102110221023102410251026102710281029103010311032103310341035103610371038103910401041104210431044104510461047104810491050105110521053105410551056105710581059106010611062106310641065106610671068106910701071107210731074107510761077107810791080108110821083108410851086108710881089109010911092109310941095109610971098109911001101110211031104110511061107110811091110111111121113111411151116111711181119112011211122112311241125112611271128112911301131113211331134113511361137113811391140114111421143114411451146114711481149115011511152115311541155115611571158115911601161116211631164116511661167116811691170117111721173117411751176117711781179118011811182118311841185118611871188118911901191119211931194119511961197119811991200120112021203120412051206120712081209121012111212121312141215121612171218121912201221122212231224122512261227122812291230123112321233123412351236123712381239124012411242124312441245124612471248124912501251125212531254125512561257125812591260126112621263126412651266126712681269127012711272127312741275127612771278127912801281128212831284128512861287128812891290129112921293129412951296129712981299130013011302130313041305130613071308130913101311131213131314131513161317131813191320132113221323132413251326132713281329133013311332133313341335133613371338133913401341134213431344134513461347134813491350135113521353135413551356135713581359136013611362136313641365136613671368136913701371137213731374137513761377137813791380138113821383138413851386138713881389139013911392139313941395139613971398139914001401140214031404140514061407140814091410141114121413141414151416141714181419142014211422142314241425142614271428142914301431143214331434143514361437143814391440144114421443144414451446144714481449145014511452145314541455145614571458145914601461146214631464146514661467146814691470147114721473147414751476147714781479148014811482148314841485148614871488148914901491149214931494149514961497149814991500150115021503150415051506150715081509151015111512151315141515151615171518151915201521152215231524152515261527152815291530153115321533153415351536153715381539154015411542154315441545154615471548154915501551155215531554155515561557155815591560156115621563156415651566156715681569157015711572157315741575157615771578157915801581158215831584158515861587158815891590159115921593159415951596159715981599160016011602160316041605160616071608160916101611161216131614161516161617161816191620162116221623162416251626162716281629163016311632163316341635163616371638163916401641164216431644164516461647164816491650165116521653165416551656165716581659166016611662166316641665166616671668166916701671167216731674167516761677167816791680168116821683168416851686168716881689169016911692169316941695169616971698169917001701170217031704170517061707170817091710171117121713171417151716171717181719172017211722172317241725172617271728172917301731173217331734173517361737173817391740174117421743174417451746174717481749175017511752175317541755175617571758175917601761176217631764176517661767176817691770177117721773177417751776177717781779178017811782178317841785178617871788178917901791179217931794179517961797179817991800180118021803180418051806180718081809181018111812181318141815181618171818181918201821182218231824182518261827182818291830183118321833183418351836183718381839184018411842184318441845184618471848184918501851185218531854185518561857185818591860186118621863186418651866186718681869187018711872187318741875187618771878187918801881188218831884188518861887188818891890189118921893189418951896189718981899190019011902190319041905190619071908190919101911191219131914191519161917191819191920192119221923192419251926192719281929193019311932193319341935193619371938193919401941194219431944194519461947194819491950195119521953195419551956195719581959196019611962196319641965196619671968196919701971197219731974197519761977197819791980198119821983198419851986198719881989199019911992199319941995199619971998199920002001200220032004200520062007200820092010201120122013201420152016201720182019202020212022202320242025202620272028202920302031203220332034203520362037203820392040204120422043204420452046204720482049205020512052205320542055205620572058205920602061206220632064206520662067206820692070207120722073207420752076207720782079208020812082208320842085208620872088208920902091209220932094209520962097209820992100210121022103210421052106210721082109211021112112211321142115211621172118211921202121212221232124212521262127212821292130213121322133213421352136213721382139214021412142214321442145214621472148214921502151215221532154215521562157215821592160216121622163216421652166216721682169217021712172217321742175217621772178217921802181218221832184218521862187218821892190219121922193219421952196219721982199220022012202220322042205220622072208220922102211221222132214221522162217221822192220222122222223222422252226222722282229223022312232223322342235223622372238223922402241224222432244224522462247224822492250225122522253225422552256225722582259226022612262226322642265226622672268226922702271227222732274227522762277227822792280228122822283228422852286228722882289229022912292229322942295229622972298229923002301230223032304230523062307230823092310231123122313231423152316231723182319232023212322232323242325232623272328232923302331233223332334233523362337233823392340234123422343234423452346234723482349235023512352235323542355235623572358235923602361236223632364236523662367236823692370237123722373237423752376237723782379238023812382238323842385238623872388238923902391239223932394239523962397239823992400240124022403240424052406240724082409241024112412241324142415241624172418241924202421242224232424242524262427242824292430243124322433243424352436243724382439244024412442244324442445244624472448244924502451245224532454245524562457245824592460246124622463246424652466246724682469247024712472247324742475247624772478247924802481248224832484248524862487248824892490249124922493249424952496249724982499250025012502250325042505250625072508250925102511251225132514251525162517251825192520252125222523252425252526252725282529253025312532253325342535253625372538253925402541254225432544254525462547254825492550255125522553255425552556255725582559256025612562256325642565256625672568256925702571257225732574257525762577257825792580258125822583258425852586258725882589259025912592259325942595259625972598259926002601260226032604260526062607260826092610261126122613261426152616261726182619262026212622262326242625262626272628262926302631263226332634263526362637263826392640264126422643264426452646264726482649265026512652265326542655265626572658265926602661266226632664266526662667266826692670267126722673267426752676267726782679268026812682268326842685268626872688268926902691269226932694269526962697269826992700270127022703270427052706270727082709271027112712271327142715271627172718271927202721272227232724272527262727272827292730273127322733273427352736273727382739274027412742274327442745274627472748274927502751275227532754275527562757275827592760276127622763276427652766276727682769277027712772277327742775277627772778277927802781278227832784278527862787278827892790279127922793279427952796279727982799280028012802280328042805280628072808280928102811281228132814281528162817281828192820282128222823282428252826282728282829283028312832283328342835283628372838283928402841284228432844284528462847284828492850285128522853285428552856285728582859286028612862286328642865286628672868286928702871287228732874287528762877287828792880288128822883288428852886288728882889289028912892289328942895289628972898289929002901290229032904290529062907290829092910291129122913291429152916291729182919292029212922292329242925292629272928292929302931293229332934293529362937293829392940294129422943294429452946294729482949295029512952295329542955295629572958295929602961296229632964296529662967296829692970297129722973297429752976297729782979298029812982298329842985298629872988298929902991299229932994299529962997299829993000300130023003300430053006300730083009301030113012301330143015301630173018301930203021302230233024302530263027302830293030303130323033303430353036303730383039304030413042304330443045304630473048304930503051305230533054305530563057305830593060306130623063306430653066306730683069307030713072307330743075307630773078307930803081308230833084308530863087308830893090309130923093309430953096309730983099310031013102310331043105310631073108310931103111311231133114311531163117311831193120312131223123312431253126312731283129313031313132313331343135313631373138313931403141314231433144314531463147314831493150315131523153315431553156315731583159316031613162316331643165316631673168316931703171317231733174317531763177317831793180318131823183318431853186318731883189319031913192319331943195319631973198319932003201320232033204320532063207320832093210321132123213321432153216321732183219322032213222322332243225322632273228322932303231323232333234323532363237323832393240324132423243324432453246324732483249325032513252325332543255325632573258325932603261326232633264326532663267326832693270327132723273327432753276327732783279328032813282328332843285328632873288328932903291329232933294329532963297329832993300330133023303330433053306330733083309331033113312331333143315331633173318331933203321332233233324332533263327332833293330333133323333333433353336333733383339334033413342334333443345334633473348334933503351335233533354335533563357335833593360336133623363336433653366336733683369337033713372337333743375337633773378337933803381338233833384338533863387338833893390339133923393339433953396339733983399340034013402340334043405340634073408340934103411341234133414341534163417341834193420342134223423342434253426342734283429343034313432343334343435343634373438343934403441344234433444344534463447344834493450345134523453345434553456345734583459346034613462346334643465346634673468346934703471347234733474347534763477347834793480348134823483348434853486348734883489349034913492349334943495349634973498349935003501350235033504350535063507350835093510351135123513351435153516351735183519352035213522352335243525352635273528352935303531353235333534353535363537353835393540354135423543354435453546354735483549355035513552355335543555355635573558355935603561356235633564356535663567356835693570357135723573357435753576357735783579358035813582358335843585358635873588358935903591359235933594359535963597359835993600360136023603360436053606360736083609361036113612361336143615361636173618361936203621362236233624362536263627362836293630363136323633363436353636363736383639364036413642364336443645364636473648364936503651365236533654365536563657365836593660366136623663366436653666366736683669367036713672367336743675367636773678367936803681368236833684368536863687368836893690369136923693369436953696369736983699370037013702370337043705370637073708370937103711371237133714371537163717371837193720372137223723372437253726372737283729373037313732373337343735373637373738373937403741374237433744374537463747374837493750375137523753375437553756375737583759376037613762376337643765376637673768376937703771377237733774377537763777377837793780378137823783378437853786378737883789379037913792379337943795379637973798379938003801380238033804380538063807380838093810381138123813381438153816381738183819382038213822382338243825382638273828382938303831383238333834383538363837383838393840384138423843384438453846384738483849385038513852385338543855385638573858385938603861386238633864386538663867386838693870387138723873387438753876387738783879388038813882388338843885388638873888388938903891389238933894389538963897389838993900390139023903390439053906390739083909391039113912391339143915391639173918391939203921392239233924392539263927392839293930393139323933393439353936393739383939394039413942394339443945394639473948394939503951395239533954395539563957395839593960396139623963396439653966396739683969397039713972397339743975397639773978397939803981398239833984398539863987398839893990399139923993399439953996399739983999400040014002400340044005400640074008400940104011401240134014401540164017401840194020402140224023402440254026402740284029403040314032403340344035403640374038403940404041404240434044404540464047
  1. /*
  2. * Copyright (c) 2000-2006 Silicon Graphics, Inc.
  3. * All Rights Reserved.
  4. *
  5. * This program is free software; you can redistribute it and/or
  6. * modify it under the terms of the GNU General Public License as
  7. * published by the Free Software Foundation.
  8. *
  9. * This program is distributed in the hope that it would be useful,
  10. * but WITHOUT ANY WARRANTY; without even the implied warranty of
  11. * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
  12. * GNU General Public License for more details.
  13. *
  14. * You should have received a copy of the GNU General Public License
  15. * along with this program; if not, write the Free Software Foundation,
  16. * Inc., 51 Franklin St, Fifth Floor, Boston, MA 02110-1301 USA
  17. */
  18. #include "xfs.h"
  19. #include "xfs_fs.h"
  20. #include "xfs_types.h"
  21. #include "xfs_bit.h"
  22. #include "xfs_log.h"
  23. #include "xfs_inum.h"
  24. #include "xfs_trans.h"
  25. #include "xfs_sb.h"
  26. #include "xfs_ag.h"
  27. #include "xfs_dir2.h"
  28. #include "xfs_dmapi.h"
  29. #include "xfs_mount.h"
  30. #include "xfs_error.h"
  31. #include "xfs_bmap_btree.h"
  32. #include "xfs_alloc_btree.h"
  33. #include "xfs_ialloc_btree.h"
  34. #include "xfs_dir2_sf.h"
  35. #include "xfs_attr_sf.h"
  36. #include "xfs_dinode.h"
  37. #include "xfs_inode.h"
  38. #include "xfs_inode_item.h"
  39. #include "xfs_imap.h"
  40. #include "xfs_alloc.h"
  41. #include "xfs_ialloc.h"
  42. #include "xfs_log_priv.h"
  43. #include "xfs_buf_item.h"
  44. #include "xfs_log_recover.h"
  45. #include "xfs_extfree_item.h"
  46. #include "xfs_trans_priv.h"
  47. #include "xfs_quota.h"
  48. #include "xfs_rw.h"
  49. STATIC int xlog_find_zeroed(xlog_t *, xfs_daddr_t *);
  50. STATIC int xlog_clear_stale_blocks(xlog_t *, xfs_lsn_t);
  51. STATIC void xlog_recover_insert_item_backq(xlog_recover_item_t **q,
  52. xlog_recover_item_t *item);
  53. #if defined(DEBUG)
  54. STATIC void xlog_recover_check_summary(xlog_t *);
  55. STATIC void xlog_recover_check_ail(xfs_mount_t *, xfs_log_item_t *, int);
  56. #else
  57. #define xlog_recover_check_summary(log)
  58. #define xlog_recover_check_ail(mp, lip, gen)
  59. #endif
  60. /*
  61. * Sector aligned buffer routines for buffer create/read/write/access
  62. */
  63. #define XLOG_SECTOR_ROUNDUP_BBCOUNT(log, bbs) \
  64. ( ((log)->l_sectbb_mask && (bbs & (log)->l_sectbb_mask)) ? \
  65. ((bbs + (log)->l_sectbb_mask + 1) & ~(log)->l_sectbb_mask) : (bbs) )
  66. #define XLOG_SECTOR_ROUNDDOWN_BLKNO(log, bno) ((bno) & ~(log)->l_sectbb_mask)
  67. xfs_buf_t *
  68. xlog_get_bp(
  69. xlog_t *log,
  70. int num_bblks)
  71. {
  72. ASSERT(num_bblks > 0);
  73. if (log->l_sectbb_log) {
  74. if (num_bblks > 1)
  75. num_bblks += XLOG_SECTOR_ROUNDUP_BBCOUNT(log, 1);
  76. num_bblks = XLOG_SECTOR_ROUNDUP_BBCOUNT(log, num_bblks);
  77. }
  78. return xfs_buf_get_noaddr(BBTOB(num_bblks), log->l_mp->m_logdev_targp);
  79. }
  80. void
  81. xlog_put_bp(
  82. xfs_buf_t *bp)
  83. {
  84. xfs_buf_free(bp);
  85. }
  86. /*
  87. * nbblks should be uint, but oh well. Just want to catch that 32-bit length.
  88. */
  89. int
  90. xlog_bread(
  91. xlog_t *log,
  92. xfs_daddr_t blk_no,
  93. int nbblks,
  94. xfs_buf_t *bp)
  95. {
  96. int error;
  97. if (log->l_sectbb_log) {
  98. blk_no = XLOG_SECTOR_ROUNDDOWN_BLKNO(log, blk_no);
  99. nbblks = XLOG_SECTOR_ROUNDUP_BBCOUNT(log, nbblks);
  100. }
  101. ASSERT(nbblks > 0);
  102. ASSERT(BBTOB(nbblks) <= XFS_BUF_SIZE(bp));
  103. ASSERT(bp);
  104. XFS_BUF_SET_ADDR(bp, log->l_logBBstart + blk_no);
  105. XFS_BUF_READ(bp);
  106. XFS_BUF_BUSY(bp);
  107. XFS_BUF_SET_COUNT(bp, BBTOB(nbblks));
  108. XFS_BUF_SET_TARGET(bp, log->l_mp->m_logdev_targp);
  109. xfsbdstrat(log->l_mp, bp);
  110. if ((error = xfs_iowait(bp)))
  111. xfs_ioerror_alert("xlog_bread", log->l_mp,
  112. bp, XFS_BUF_ADDR(bp));
  113. return error;
  114. }
  115. /*
  116. * Write out the buffer at the given block for the given number of blocks.
  117. * The buffer is kept locked across the write and is returned locked.
  118. * This can only be used for synchronous log writes.
  119. */
  120. STATIC int
  121. xlog_bwrite(
  122. xlog_t *log,
  123. xfs_daddr_t blk_no,
  124. int nbblks,
  125. xfs_buf_t *bp)
  126. {
  127. int error;
  128. if (log->l_sectbb_log) {
  129. blk_no = XLOG_SECTOR_ROUNDDOWN_BLKNO(log, blk_no);
  130. nbblks = XLOG_SECTOR_ROUNDUP_BBCOUNT(log, nbblks);
  131. }
  132. ASSERT(nbblks > 0);
  133. ASSERT(BBTOB(nbblks) <= XFS_BUF_SIZE(bp));
  134. XFS_BUF_SET_ADDR(bp, log->l_logBBstart + blk_no);
  135. XFS_BUF_ZEROFLAGS(bp);
  136. XFS_BUF_BUSY(bp);
  137. XFS_BUF_HOLD(bp);
  138. XFS_BUF_PSEMA(bp, PRIBIO);
  139. XFS_BUF_SET_COUNT(bp, BBTOB(nbblks));
  140. XFS_BUF_SET_TARGET(bp, log->l_mp->m_logdev_targp);
  141. if ((error = xfs_bwrite(log->l_mp, bp)))
  142. xfs_ioerror_alert("xlog_bwrite", log->l_mp,
  143. bp, XFS_BUF_ADDR(bp));
  144. return error;
  145. }
  146. STATIC xfs_caddr_t
  147. xlog_align(
  148. xlog_t *log,
  149. xfs_daddr_t blk_no,
  150. int nbblks,
  151. xfs_buf_t *bp)
  152. {
  153. xfs_caddr_t ptr;
  154. if (!log->l_sectbb_log)
  155. return XFS_BUF_PTR(bp);
  156. ptr = XFS_BUF_PTR(bp) + BBTOB((int)blk_no & log->l_sectbb_mask);
  157. ASSERT(XFS_BUF_SIZE(bp) >=
  158. BBTOB(nbblks + (blk_no & log->l_sectbb_mask)));
  159. return ptr;
  160. }
  161. #ifdef DEBUG
  162. /*
  163. * dump debug superblock and log record information
  164. */
  165. STATIC void
  166. xlog_header_check_dump(
  167. xfs_mount_t *mp,
  168. xlog_rec_header_t *head)
  169. {
  170. int b;
  171. cmn_err(CE_DEBUG, "%s: SB : uuid = ", __FUNCTION__);
  172. for (b = 0; b < 16; b++)
  173. cmn_err(CE_DEBUG, "%02x", ((uchar_t *)&mp->m_sb.sb_uuid)[b]);
  174. cmn_err(CE_DEBUG, ", fmt = %d\n", XLOG_FMT);
  175. cmn_err(CE_DEBUG, " log : uuid = ");
  176. for (b = 0; b < 16; b++)
  177. cmn_err(CE_DEBUG, "%02x",((uchar_t *)&head->h_fs_uuid)[b]);
  178. cmn_err(CE_DEBUG, ", fmt = %d\n", INT_GET(head->h_fmt, ARCH_CONVERT));
  179. }
  180. #else
  181. #define xlog_header_check_dump(mp, head)
  182. #endif
  183. /*
  184. * check log record header for recovery
  185. */
  186. STATIC int
  187. xlog_header_check_recover(
  188. xfs_mount_t *mp,
  189. xlog_rec_header_t *head)
  190. {
  191. ASSERT(INT_GET(head->h_magicno, ARCH_CONVERT) == XLOG_HEADER_MAGIC_NUM);
  192. /*
  193. * IRIX doesn't write the h_fmt field and leaves it zeroed
  194. * (XLOG_FMT_UNKNOWN). This stops us from trying to recover
  195. * a dirty log created in IRIX.
  196. */
  197. if (unlikely(INT_GET(head->h_fmt, ARCH_CONVERT) != XLOG_FMT)) {
  198. xlog_warn(
  199. "XFS: dirty log written in incompatible format - can't recover");
  200. xlog_header_check_dump(mp, head);
  201. XFS_ERROR_REPORT("xlog_header_check_recover(1)",
  202. XFS_ERRLEVEL_HIGH, mp);
  203. return XFS_ERROR(EFSCORRUPTED);
  204. } else if (unlikely(!uuid_equal(&mp->m_sb.sb_uuid, &head->h_fs_uuid))) {
  205. xlog_warn(
  206. "XFS: dirty log entry has mismatched uuid - can't recover");
  207. xlog_header_check_dump(mp, head);
  208. XFS_ERROR_REPORT("xlog_header_check_recover(2)",
  209. XFS_ERRLEVEL_HIGH, mp);
  210. return XFS_ERROR(EFSCORRUPTED);
  211. }
  212. return 0;
  213. }
  214. /*
  215. * read the head block of the log and check the header
  216. */
  217. STATIC int
  218. xlog_header_check_mount(
  219. xfs_mount_t *mp,
  220. xlog_rec_header_t *head)
  221. {
  222. ASSERT(INT_GET(head->h_magicno, ARCH_CONVERT) == XLOG_HEADER_MAGIC_NUM);
  223. if (uuid_is_nil(&head->h_fs_uuid)) {
  224. /*
  225. * IRIX doesn't write the h_fs_uuid or h_fmt fields. If
  226. * h_fs_uuid is nil, we assume this log was last mounted
  227. * by IRIX and continue.
  228. */
  229. xlog_warn("XFS: nil uuid in log - IRIX style log");
  230. } else if (unlikely(!uuid_equal(&mp->m_sb.sb_uuid, &head->h_fs_uuid))) {
  231. xlog_warn("XFS: log has mismatched uuid - can't recover");
  232. xlog_header_check_dump(mp, head);
  233. XFS_ERROR_REPORT("xlog_header_check_mount",
  234. XFS_ERRLEVEL_HIGH, mp);
  235. return XFS_ERROR(EFSCORRUPTED);
  236. }
  237. return 0;
  238. }
  239. STATIC void
  240. xlog_recover_iodone(
  241. struct xfs_buf *bp)
  242. {
  243. xfs_mount_t *mp;
  244. ASSERT(XFS_BUF_FSPRIVATE(bp, void *));
  245. if (XFS_BUF_GETERROR(bp)) {
  246. /*
  247. * We're not going to bother about retrying
  248. * this during recovery. One strike!
  249. */
  250. mp = XFS_BUF_FSPRIVATE(bp, xfs_mount_t *);
  251. xfs_ioerror_alert("xlog_recover_iodone",
  252. mp, bp, XFS_BUF_ADDR(bp));
  253. xfs_force_shutdown(mp, SHUTDOWN_META_IO_ERROR);
  254. }
  255. XFS_BUF_SET_FSPRIVATE(bp, NULL);
  256. XFS_BUF_CLR_IODONE_FUNC(bp);
  257. xfs_biodone(bp);
  258. }
  259. /*
  260. * This routine finds (to an approximation) the first block in the physical
  261. * log which contains the given cycle. It uses a binary search algorithm.
  262. * Note that the algorithm can not be perfect because the disk will not
  263. * necessarily be perfect.
  264. */
  265. int
  266. xlog_find_cycle_start(
  267. xlog_t *log,
  268. xfs_buf_t *bp,
  269. xfs_daddr_t first_blk,
  270. xfs_daddr_t *last_blk,
  271. uint cycle)
  272. {
  273. xfs_caddr_t offset;
  274. xfs_daddr_t mid_blk;
  275. uint mid_cycle;
  276. int error;
  277. mid_blk = BLK_AVG(first_blk, *last_blk);
  278. while (mid_blk != first_blk && mid_blk != *last_blk) {
  279. if ((error = xlog_bread(log, mid_blk, 1, bp)))
  280. return error;
  281. offset = xlog_align(log, mid_blk, 1, bp);
  282. mid_cycle = GET_CYCLE(offset, ARCH_CONVERT);
  283. if (mid_cycle == cycle) {
  284. *last_blk = mid_blk;
  285. /* last_half_cycle == mid_cycle */
  286. } else {
  287. first_blk = mid_blk;
  288. /* first_half_cycle == mid_cycle */
  289. }
  290. mid_blk = BLK_AVG(first_blk, *last_blk);
  291. }
  292. ASSERT((mid_blk == first_blk && mid_blk+1 == *last_blk) ||
  293. (mid_blk == *last_blk && mid_blk-1 == first_blk));
  294. return 0;
  295. }
  296. /*
  297. * Check that the range of blocks does not contain the cycle number
  298. * given. The scan needs to occur from front to back and the ptr into the
  299. * region must be updated since a later routine will need to perform another
  300. * test. If the region is completely good, we end up returning the same
  301. * last block number.
  302. *
  303. * Set blkno to -1 if we encounter no errors. This is an invalid block number
  304. * since we don't ever expect logs to get this large.
  305. */
  306. STATIC int
  307. xlog_find_verify_cycle(
  308. xlog_t *log,
  309. xfs_daddr_t start_blk,
  310. int nbblks,
  311. uint stop_on_cycle_no,
  312. xfs_daddr_t *new_blk)
  313. {
  314. xfs_daddr_t i, j;
  315. uint cycle;
  316. xfs_buf_t *bp;
  317. xfs_daddr_t bufblks;
  318. xfs_caddr_t buf = NULL;
  319. int error = 0;
  320. bufblks = 1 << ffs(nbblks);
  321. while (!(bp = xlog_get_bp(log, bufblks))) {
  322. /* can't get enough memory to do everything in one big buffer */
  323. bufblks >>= 1;
  324. if (bufblks <= log->l_sectbb_log)
  325. return ENOMEM;
  326. }
  327. for (i = start_blk; i < start_blk + nbblks; i += bufblks) {
  328. int bcount;
  329. bcount = min(bufblks, (start_blk + nbblks - i));
  330. if ((error = xlog_bread(log, i, bcount, bp)))
  331. goto out;
  332. buf = xlog_align(log, i, bcount, bp);
  333. for (j = 0; j < bcount; j++) {
  334. cycle = GET_CYCLE(buf, ARCH_CONVERT);
  335. if (cycle == stop_on_cycle_no) {
  336. *new_blk = i+j;
  337. goto out;
  338. }
  339. buf += BBSIZE;
  340. }
  341. }
  342. *new_blk = -1;
  343. out:
  344. xlog_put_bp(bp);
  345. return error;
  346. }
  347. /*
  348. * Potentially backup over partial log record write.
  349. *
  350. * In the typical case, last_blk is the number of the block directly after
  351. * a good log record. Therefore, we subtract one to get the block number
  352. * of the last block in the given buffer. extra_bblks contains the number
  353. * of blocks we would have read on a previous read. This happens when the
  354. * last log record is split over the end of the physical log.
  355. *
  356. * extra_bblks is the number of blocks potentially verified on a previous
  357. * call to this routine.
  358. */
  359. STATIC int
  360. xlog_find_verify_log_record(
  361. xlog_t *log,
  362. xfs_daddr_t start_blk,
  363. xfs_daddr_t *last_blk,
  364. int extra_bblks)
  365. {
  366. xfs_daddr_t i;
  367. xfs_buf_t *bp;
  368. xfs_caddr_t offset = NULL;
  369. xlog_rec_header_t *head = NULL;
  370. int error = 0;
  371. int smallmem = 0;
  372. int num_blks = *last_blk - start_blk;
  373. int xhdrs;
  374. ASSERT(start_blk != 0 || *last_blk != start_blk);
  375. if (!(bp = xlog_get_bp(log, num_blks))) {
  376. if (!(bp = xlog_get_bp(log, 1)))
  377. return ENOMEM;
  378. smallmem = 1;
  379. } else {
  380. if ((error = xlog_bread(log, start_blk, num_blks, bp)))
  381. goto out;
  382. offset = xlog_align(log, start_blk, num_blks, bp);
  383. offset += ((num_blks - 1) << BBSHIFT);
  384. }
  385. for (i = (*last_blk) - 1; i >= 0; i--) {
  386. if (i < start_blk) {
  387. /* valid log record not found */
  388. xlog_warn(
  389. "XFS: Log inconsistent (didn't find previous header)");
  390. ASSERT(0);
  391. error = XFS_ERROR(EIO);
  392. goto out;
  393. }
  394. if (smallmem) {
  395. if ((error = xlog_bread(log, i, 1, bp)))
  396. goto out;
  397. offset = xlog_align(log, i, 1, bp);
  398. }
  399. head = (xlog_rec_header_t *)offset;
  400. if (XLOG_HEADER_MAGIC_NUM ==
  401. INT_GET(head->h_magicno, ARCH_CONVERT))
  402. break;
  403. if (!smallmem)
  404. offset -= BBSIZE;
  405. }
  406. /*
  407. * We hit the beginning of the physical log & still no header. Return
  408. * to caller. If caller can handle a return of -1, then this routine
  409. * will be called again for the end of the physical log.
  410. */
  411. if (i == -1) {
  412. error = -1;
  413. goto out;
  414. }
  415. /*
  416. * We have the final block of the good log (the first block
  417. * of the log record _before_ the head. So we check the uuid.
  418. */
  419. if ((error = xlog_header_check_mount(log->l_mp, head)))
  420. goto out;
  421. /*
  422. * We may have found a log record header before we expected one.
  423. * last_blk will be the 1st block # with a given cycle #. We may end
  424. * up reading an entire log record. In this case, we don't want to
  425. * reset last_blk. Only when last_blk points in the middle of a log
  426. * record do we update last_blk.
  427. */
  428. if (XFS_SB_VERSION_HASLOGV2(&log->l_mp->m_sb)) {
  429. uint h_size = INT_GET(head->h_size, ARCH_CONVERT);
  430. xhdrs = h_size / XLOG_HEADER_CYCLE_SIZE;
  431. if (h_size % XLOG_HEADER_CYCLE_SIZE)
  432. xhdrs++;
  433. } else {
  434. xhdrs = 1;
  435. }
  436. if (*last_blk - i + extra_bblks
  437. != BTOBB(INT_GET(head->h_len, ARCH_CONVERT)) + xhdrs)
  438. *last_blk = i;
  439. out:
  440. xlog_put_bp(bp);
  441. return error;
  442. }
  443. /*
  444. * Head is defined to be the point of the log where the next log write
  445. * write could go. This means that incomplete LR writes at the end are
  446. * eliminated when calculating the head. We aren't guaranteed that previous
  447. * LR have complete transactions. We only know that a cycle number of
  448. * current cycle number -1 won't be present in the log if we start writing
  449. * from our current block number.
  450. *
  451. * last_blk contains the block number of the first block with a given
  452. * cycle number.
  453. *
  454. * Return: zero if normal, non-zero if error.
  455. */
  456. STATIC int
  457. xlog_find_head(
  458. xlog_t *log,
  459. xfs_daddr_t *return_head_blk)
  460. {
  461. xfs_buf_t *bp;
  462. xfs_caddr_t offset;
  463. xfs_daddr_t new_blk, first_blk, start_blk, last_blk, head_blk;
  464. int num_scan_bblks;
  465. uint first_half_cycle, last_half_cycle;
  466. uint stop_on_cycle;
  467. int error, log_bbnum = log->l_logBBsize;
  468. /* Is the end of the log device zeroed? */
  469. if ((error = xlog_find_zeroed(log, &first_blk)) == -1) {
  470. *return_head_blk = first_blk;
  471. /* Is the whole lot zeroed? */
  472. if (!first_blk) {
  473. /* Linux XFS shouldn't generate totally zeroed logs -
  474. * mkfs etc write a dummy unmount record to a fresh
  475. * log so we can store the uuid in there
  476. */
  477. xlog_warn("XFS: totally zeroed log");
  478. }
  479. return 0;
  480. } else if (error) {
  481. xlog_warn("XFS: empty log check failed");
  482. return error;
  483. }
  484. first_blk = 0; /* get cycle # of 1st block */
  485. bp = xlog_get_bp(log, 1);
  486. if (!bp)
  487. return ENOMEM;
  488. if ((error = xlog_bread(log, 0, 1, bp)))
  489. goto bp_err;
  490. offset = xlog_align(log, 0, 1, bp);
  491. first_half_cycle = GET_CYCLE(offset, ARCH_CONVERT);
  492. last_blk = head_blk = log_bbnum - 1; /* get cycle # of last block */
  493. if ((error = xlog_bread(log, last_blk, 1, bp)))
  494. goto bp_err;
  495. offset = xlog_align(log, last_blk, 1, bp);
  496. last_half_cycle = GET_CYCLE(offset, ARCH_CONVERT);
  497. ASSERT(last_half_cycle != 0);
  498. /*
  499. * If the 1st half cycle number is equal to the last half cycle number,
  500. * then the entire log is stamped with the same cycle number. In this
  501. * case, head_blk can't be set to zero (which makes sense). The below
  502. * math doesn't work out properly with head_blk equal to zero. Instead,
  503. * we set it to log_bbnum which is an invalid block number, but this
  504. * value makes the math correct. If head_blk doesn't changed through
  505. * all the tests below, *head_blk is set to zero at the very end rather
  506. * than log_bbnum. In a sense, log_bbnum and zero are the same block
  507. * in a circular file.
  508. */
  509. if (first_half_cycle == last_half_cycle) {
  510. /*
  511. * In this case we believe that the entire log should have
  512. * cycle number last_half_cycle. We need to scan backwards
  513. * from the end verifying that there are no holes still
  514. * containing last_half_cycle - 1. If we find such a hole,
  515. * then the start of that hole will be the new head. The
  516. * simple case looks like
  517. * x | x ... | x - 1 | x
  518. * Another case that fits this picture would be
  519. * x | x + 1 | x ... | x
  520. * In this case the head really is somewhere at the end of the
  521. * log, as one of the latest writes at the beginning was
  522. * incomplete.
  523. * One more case is
  524. * x | x + 1 | x ... | x - 1 | x
  525. * This is really the combination of the above two cases, and
  526. * the head has to end up at the start of the x-1 hole at the
  527. * end of the log.
  528. *
  529. * In the 256k log case, we will read from the beginning to the
  530. * end of the log and search for cycle numbers equal to x-1.
  531. * We don't worry about the x+1 blocks that we encounter,
  532. * because we know that they cannot be the head since the log
  533. * started with x.
  534. */
  535. head_blk = log_bbnum;
  536. stop_on_cycle = last_half_cycle - 1;
  537. } else {
  538. /*
  539. * In this case we want to find the first block with cycle
  540. * number matching last_half_cycle. We expect the log to be
  541. * some variation on
  542. * x + 1 ... | x ...
  543. * The first block with cycle number x (last_half_cycle) will
  544. * be where the new head belongs. First we do a binary search
  545. * for the first occurrence of last_half_cycle. The binary
  546. * search may not be totally accurate, so then we scan back
  547. * from there looking for occurrences of last_half_cycle before
  548. * us. If that backwards scan wraps around the beginning of
  549. * the log, then we look for occurrences of last_half_cycle - 1
  550. * at the end of the log. The cases we're looking for look
  551. * like
  552. * x + 1 ... | x | x + 1 | x ...
  553. * ^ binary search stopped here
  554. * or
  555. * x + 1 ... | x ... | x - 1 | x
  556. * <---------> less than scan distance
  557. */
  558. stop_on_cycle = last_half_cycle;
  559. if ((error = xlog_find_cycle_start(log, bp, first_blk,
  560. &head_blk, last_half_cycle)))
  561. goto bp_err;
  562. }
  563. /*
  564. * Now validate the answer. Scan back some number of maximum possible
  565. * blocks and make sure each one has the expected cycle number. The
  566. * maximum is determined by the total possible amount of buffering
  567. * in the in-core log. The following number can be made tighter if
  568. * we actually look at the block size of the filesystem.
  569. */
  570. num_scan_bblks = XLOG_TOTAL_REC_SHIFT(log);
  571. if (head_blk >= num_scan_bblks) {
  572. /*
  573. * We are guaranteed that the entire check can be performed
  574. * in one buffer.
  575. */
  576. start_blk = head_blk - num_scan_bblks;
  577. if ((error = xlog_find_verify_cycle(log,
  578. start_blk, num_scan_bblks,
  579. stop_on_cycle, &new_blk)))
  580. goto bp_err;
  581. if (new_blk != -1)
  582. head_blk = new_blk;
  583. } else { /* need to read 2 parts of log */
  584. /*
  585. * We are going to scan backwards in the log in two parts.
  586. * First we scan the physical end of the log. In this part
  587. * of the log, we are looking for blocks with cycle number
  588. * last_half_cycle - 1.
  589. * If we find one, then we know that the log starts there, as
  590. * we've found a hole that didn't get written in going around
  591. * the end of the physical log. The simple case for this is
  592. * x + 1 ... | x ... | x - 1 | x
  593. * <---------> less than scan distance
  594. * If all of the blocks at the end of the log have cycle number
  595. * last_half_cycle, then we check the blocks at the start of
  596. * the log looking for occurrences of last_half_cycle. If we
  597. * find one, then our current estimate for the location of the
  598. * first occurrence of last_half_cycle is wrong and we move
  599. * back to the hole we've found. This case looks like
  600. * x + 1 ... | x | x + 1 | x ...
  601. * ^ binary search stopped here
  602. * Another case we need to handle that only occurs in 256k
  603. * logs is
  604. * x + 1 ... | x ... | x+1 | x ...
  605. * ^ binary search stops here
  606. * In a 256k log, the scan at the end of the log will see the
  607. * x + 1 blocks. We need to skip past those since that is
  608. * certainly not the head of the log. By searching for
  609. * last_half_cycle-1 we accomplish that.
  610. */
  611. start_blk = log_bbnum - num_scan_bblks + head_blk;
  612. ASSERT(head_blk <= INT_MAX &&
  613. (xfs_daddr_t) num_scan_bblks - head_blk >= 0);
  614. if ((error = xlog_find_verify_cycle(log, start_blk,
  615. num_scan_bblks - (int)head_blk,
  616. (stop_on_cycle - 1), &new_blk)))
  617. goto bp_err;
  618. if (new_blk != -1) {
  619. head_blk = new_blk;
  620. goto bad_blk;
  621. }
  622. /*
  623. * Scan beginning of log now. The last part of the physical
  624. * log is good. This scan needs to verify that it doesn't find
  625. * the last_half_cycle.
  626. */
  627. start_blk = 0;
  628. ASSERT(head_blk <= INT_MAX);
  629. if ((error = xlog_find_verify_cycle(log,
  630. start_blk, (int)head_blk,
  631. stop_on_cycle, &new_blk)))
  632. goto bp_err;
  633. if (new_blk != -1)
  634. head_blk = new_blk;
  635. }
  636. bad_blk:
  637. /*
  638. * Now we need to make sure head_blk is not pointing to a block in
  639. * the middle of a log record.
  640. */
  641. num_scan_bblks = XLOG_REC_SHIFT(log);
  642. if (head_blk >= num_scan_bblks) {
  643. start_blk = head_blk - num_scan_bblks; /* don't read head_blk */
  644. /* start ptr at last block ptr before head_blk */
  645. if ((error = xlog_find_verify_log_record(log, start_blk,
  646. &head_blk, 0)) == -1) {
  647. error = XFS_ERROR(EIO);
  648. goto bp_err;
  649. } else if (error)
  650. goto bp_err;
  651. } else {
  652. start_blk = 0;
  653. ASSERT(head_blk <= INT_MAX);
  654. if ((error = xlog_find_verify_log_record(log, start_blk,
  655. &head_blk, 0)) == -1) {
  656. /* We hit the beginning of the log during our search */
  657. start_blk = log_bbnum - num_scan_bblks + head_blk;
  658. new_blk = log_bbnum;
  659. ASSERT(start_blk <= INT_MAX &&
  660. (xfs_daddr_t) log_bbnum-start_blk >= 0);
  661. ASSERT(head_blk <= INT_MAX);
  662. if ((error = xlog_find_verify_log_record(log,
  663. start_blk, &new_blk,
  664. (int)head_blk)) == -1) {
  665. error = XFS_ERROR(EIO);
  666. goto bp_err;
  667. } else if (error)
  668. goto bp_err;
  669. if (new_blk != log_bbnum)
  670. head_blk = new_blk;
  671. } else if (error)
  672. goto bp_err;
  673. }
  674. xlog_put_bp(bp);
  675. if (head_blk == log_bbnum)
  676. *return_head_blk = 0;
  677. else
  678. *return_head_blk = head_blk;
  679. /*
  680. * When returning here, we have a good block number. Bad block
  681. * means that during a previous crash, we didn't have a clean break
  682. * from cycle number N to cycle number N-1. In this case, we need
  683. * to find the first block with cycle number N-1.
  684. */
  685. return 0;
  686. bp_err:
  687. xlog_put_bp(bp);
  688. if (error)
  689. xlog_warn("XFS: failed to find log head");
  690. return error;
  691. }
  692. /*
  693. * Find the sync block number or the tail of the log.
  694. *
  695. * This will be the block number of the last record to have its
  696. * associated buffers synced to disk. Every log record header has
  697. * a sync lsn embedded in it. LSNs hold block numbers, so it is easy
  698. * to get a sync block number. The only concern is to figure out which
  699. * log record header to believe.
  700. *
  701. * The following algorithm uses the log record header with the largest
  702. * lsn. The entire log record does not need to be valid. We only care
  703. * that the header is valid.
  704. *
  705. * We could speed up search by using current head_blk buffer, but it is not
  706. * available.
  707. */
  708. int
  709. xlog_find_tail(
  710. xlog_t *log,
  711. xfs_daddr_t *head_blk,
  712. xfs_daddr_t *tail_blk)
  713. {
  714. xlog_rec_header_t *rhead;
  715. xlog_op_header_t *op_head;
  716. xfs_caddr_t offset = NULL;
  717. xfs_buf_t *bp;
  718. int error, i, found;
  719. xfs_daddr_t umount_data_blk;
  720. xfs_daddr_t after_umount_blk;
  721. xfs_lsn_t tail_lsn;
  722. int hblks;
  723. found = 0;
  724. /*
  725. * Find previous log record
  726. */
  727. if ((error = xlog_find_head(log, head_blk)))
  728. return error;
  729. bp = xlog_get_bp(log, 1);
  730. if (!bp)
  731. return ENOMEM;
  732. if (*head_blk == 0) { /* special case */
  733. if ((error = xlog_bread(log, 0, 1, bp)))
  734. goto bread_err;
  735. offset = xlog_align(log, 0, 1, bp);
  736. if (GET_CYCLE(offset, ARCH_CONVERT) == 0) {
  737. *tail_blk = 0;
  738. /* leave all other log inited values alone */
  739. goto exit;
  740. }
  741. }
  742. /*
  743. * Search backwards looking for log record header block
  744. */
  745. ASSERT(*head_blk < INT_MAX);
  746. for (i = (int)(*head_blk) - 1; i >= 0; i--) {
  747. if ((error = xlog_bread(log, i, 1, bp)))
  748. goto bread_err;
  749. offset = xlog_align(log, i, 1, bp);
  750. if (XLOG_HEADER_MAGIC_NUM ==
  751. INT_GET(*(uint *)offset, ARCH_CONVERT)) {
  752. found = 1;
  753. break;
  754. }
  755. }
  756. /*
  757. * If we haven't found the log record header block, start looking
  758. * again from the end of the physical log. XXXmiken: There should be
  759. * a check here to make sure we didn't search more than N blocks in
  760. * the previous code.
  761. */
  762. if (!found) {
  763. for (i = log->l_logBBsize - 1; i >= (int)(*head_blk); i--) {
  764. if ((error = xlog_bread(log, i, 1, bp)))
  765. goto bread_err;
  766. offset = xlog_align(log, i, 1, bp);
  767. if (XLOG_HEADER_MAGIC_NUM ==
  768. INT_GET(*(uint*)offset, ARCH_CONVERT)) {
  769. found = 2;
  770. break;
  771. }
  772. }
  773. }
  774. if (!found) {
  775. xlog_warn("XFS: xlog_find_tail: couldn't find sync record");
  776. ASSERT(0);
  777. return XFS_ERROR(EIO);
  778. }
  779. /* find blk_no of tail of log */
  780. rhead = (xlog_rec_header_t *)offset;
  781. *tail_blk = BLOCK_LSN(INT_GET(rhead->h_tail_lsn, ARCH_CONVERT));
  782. /*
  783. * Reset log values according to the state of the log when we
  784. * crashed. In the case where head_blk == 0, we bump curr_cycle
  785. * one because the next write starts a new cycle rather than
  786. * continuing the cycle of the last good log record. At this
  787. * point we have guaranteed that all partial log records have been
  788. * accounted for. Therefore, we know that the last good log record
  789. * written was complete and ended exactly on the end boundary
  790. * of the physical log.
  791. */
  792. log->l_prev_block = i;
  793. log->l_curr_block = (int)*head_blk;
  794. log->l_curr_cycle = INT_GET(rhead->h_cycle, ARCH_CONVERT);
  795. if (found == 2)
  796. log->l_curr_cycle++;
  797. log->l_tail_lsn = INT_GET(rhead->h_tail_lsn, ARCH_CONVERT);
  798. log->l_last_sync_lsn = INT_GET(rhead->h_lsn, ARCH_CONVERT);
  799. log->l_grant_reserve_cycle = log->l_curr_cycle;
  800. log->l_grant_reserve_bytes = BBTOB(log->l_curr_block);
  801. log->l_grant_write_cycle = log->l_curr_cycle;
  802. log->l_grant_write_bytes = BBTOB(log->l_curr_block);
  803. /*
  804. * Look for unmount record. If we find it, then we know there
  805. * was a clean unmount. Since 'i' could be the last block in
  806. * the physical log, we convert to a log block before comparing
  807. * to the head_blk.
  808. *
  809. * Save the current tail lsn to use to pass to
  810. * xlog_clear_stale_blocks() below. We won't want to clear the
  811. * unmount record if there is one, so we pass the lsn of the
  812. * unmount record rather than the block after it.
  813. */
  814. if (XFS_SB_VERSION_HASLOGV2(&log->l_mp->m_sb)) {
  815. int h_size = INT_GET(rhead->h_size, ARCH_CONVERT);
  816. int h_version = INT_GET(rhead->h_version, ARCH_CONVERT);
  817. if ((h_version & XLOG_VERSION_2) &&
  818. (h_size > XLOG_HEADER_CYCLE_SIZE)) {
  819. hblks = h_size / XLOG_HEADER_CYCLE_SIZE;
  820. if (h_size % XLOG_HEADER_CYCLE_SIZE)
  821. hblks++;
  822. } else {
  823. hblks = 1;
  824. }
  825. } else {
  826. hblks = 1;
  827. }
  828. after_umount_blk = (i + hblks + (int)
  829. BTOBB(INT_GET(rhead->h_len, ARCH_CONVERT))) % log->l_logBBsize;
  830. tail_lsn = log->l_tail_lsn;
  831. if (*head_blk == after_umount_blk &&
  832. INT_GET(rhead->h_num_logops, ARCH_CONVERT) == 1) {
  833. umount_data_blk = (i + hblks) % log->l_logBBsize;
  834. if ((error = xlog_bread(log, umount_data_blk, 1, bp))) {
  835. goto bread_err;
  836. }
  837. offset = xlog_align(log, umount_data_blk, 1, bp);
  838. op_head = (xlog_op_header_t *)offset;
  839. if (op_head->oh_flags & XLOG_UNMOUNT_TRANS) {
  840. /*
  841. * Set tail and last sync so that newly written
  842. * log records will point recovery to after the
  843. * current unmount record.
  844. */
  845. ASSIGN_ANY_LSN_HOST(log->l_tail_lsn, log->l_curr_cycle,
  846. after_umount_blk);
  847. ASSIGN_ANY_LSN_HOST(log->l_last_sync_lsn, log->l_curr_cycle,
  848. after_umount_blk);
  849. *tail_blk = after_umount_blk;
  850. }
  851. }
  852. /*
  853. * Make sure that there are no blocks in front of the head
  854. * with the same cycle number as the head. This can happen
  855. * because we allow multiple outstanding log writes concurrently,
  856. * and the later writes might make it out before earlier ones.
  857. *
  858. * We use the lsn from before modifying it so that we'll never
  859. * overwrite the unmount record after a clean unmount.
  860. *
  861. * Do this only if we are going to recover the filesystem
  862. *
  863. * NOTE: This used to say "if (!readonly)"
  864. * However on Linux, we can & do recover a read-only filesystem.
  865. * We only skip recovery if NORECOVERY is specified on mount,
  866. * in which case we would not be here.
  867. *
  868. * But... if the -device- itself is readonly, just skip this.
  869. * We can't recover this device anyway, so it won't matter.
  870. */
  871. if (!xfs_readonly_buftarg(log->l_mp->m_logdev_targp)) {
  872. error = xlog_clear_stale_blocks(log, tail_lsn);
  873. }
  874. bread_err:
  875. exit:
  876. xlog_put_bp(bp);
  877. if (error)
  878. xlog_warn("XFS: failed to locate log tail");
  879. return error;
  880. }
  881. /*
  882. * Is the log zeroed at all?
  883. *
  884. * The last binary search should be changed to perform an X block read
  885. * once X becomes small enough. You can then search linearly through
  886. * the X blocks. This will cut down on the number of reads we need to do.
  887. *
  888. * If the log is partially zeroed, this routine will pass back the blkno
  889. * of the first block with cycle number 0. It won't have a complete LR
  890. * preceding it.
  891. *
  892. * Return:
  893. * 0 => the log is completely written to
  894. * -1 => use *blk_no as the first block of the log
  895. * >0 => error has occurred
  896. */
  897. int
  898. xlog_find_zeroed(
  899. xlog_t *log,
  900. xfs_daddr_t *blk_no)
  901. {
  902. xfs_buf_t *bp;
  903. xfs_caddr_t offset;
  904. uint first_cycle, last_cycle;
  905. xfs_daddr_t new_blk, last_blk, start_blk;
  906. xfs_daddr_t num_scan_bblks;
  907. int error, log_bbnum = log->l_logBBsize;
  908. *blk_no = 0;
  909. /* check totally zeroed log */
  910. bp = xlog_get_bp(log, 1);
  911. if (!bp)
  912. return ENOMEM;
  913. if ((error = xlog_bread(log, 0, 1, bp)))
  914. goto bp_err;
  915. offset = xlog_align(log, 0, 1, bp);
  916. first_cycle = GET_CYCLE(offset, ARCH_CONVERT);
  917. if (first_cycle == 0) { /* completely zeroed log */
  918. *blk_no = 0;
  919. xlog_put_bp(bp);
  920. return -1;
  921. }
  922. /* check partially zeroed log */
  923. if ((error = xlog_bread(log, log_bbnum-1, 1, bp)))
  924. goto bp_err;
  925. offset = xlog_align(log, log_bbnum-1, 1, bp);
  926. last_cycle = GET_CYCLE(offset, ARCH_CONVERT);
  927. if (last_cycle != 0) { /* log completely written to */
  928. xlog_put_bp(bp);
  929. return 0;
  930. } else if (first_cycle != 1) {
  931. /*
  932. * If the cycle of the last block is zero, the cycle of
  933. * the first block must be 1. If it's not, maybe we're
  934. * not looking at a log... Bail out.
  935. */
  936. xlog_warn("XFS: Log inconsistent or not a log (last==0, first!=1)");
  937. return XFS_ERROR(EINVAL);
  938. }
  939. /* we have a partially zeroed log */
  940. last_blk = log_bbnum-1;
  941. if ((error = xlog_find_cycle_start(log, bp, 0, &last_blk, 0)))
  942. goto bp_err;
  943. /*
  944. * Validate the answer. Because there is no way to guarantee that
  945. * the entire log is made up of log records which are the same size,
  946. * we scan over the defined maximum blocks. At this point, the maximum
  947. * is not chosen to mean anything special. XXXmiken
  948. */
  949. num_scan_bblks = XLOG_TOTAL_REC_SHIFT(log);
  950. ASSERT(num_scan_bblks <= INT_MAX);
  951. if (last_blk < num_scan_bblks)
  952. num_scan_bblks = last_blk;
  953. start_blk = last_blk - num_scan_bblks;
  954. /*
  955. * We search for any instances of cycle number 0 that occur before
  956. * our current estimate of the head. What we're trying to detect is
  957. * 1 ... | 0 | 1 | 0...
  958. * ^ binary search ends here
  959. */
  960. if ((error = xlog_find_verify_cycle(log, start_blk,
  961. (int)num_scan_bblks, 0, &new_blk)))
  962. goto bp_err;
  963. if (new_blk != -1)
  964. last_blk = new_blk;
  965. /*
  966. * Potentially backup over partial log record write. We don't need
  967. * to search the end of the log because we know it is zero.
  968. */
  969. if ((error = xlog_find_verify_log_record(log, start_blk,
  970. &last_blk, 0)) == -1) {
  971. error = XFS_ERROR(EIO);
  972. goto bp_err;
  973. } else if (error)
  974. goto bp_err;
  975. *blk_no = last_blk;
  976. bp_err:
  977. xlog_put_bp(bp);
  978. if (error)
  979. return error;
  980. return -1;
  981. }
  982. /*
  983. * These are simple subroutines used by xlog_clear_stale_blocks() below
  984. * to initialize a buffer full of empty log record headers and write
  985. * them into the log.
  986. */
  987. STATIC void
  988. xlog_add_record(
  989. xlog_t *log,
  990. xfs_caddr_t buf,
  991. int cycle,
  992. int block,
  993. int tail_cycle,
  994. int tail_block)
  995. {
  996. xlog_rec_header_t *recp = (xlog_rec_header_t *)buf;
  997. memset(buf, 0, BBSIZE);
  998. INT_SET(recp->h_magicno, ARCH_CONVERT, XLOG_HEADER_MAGIC_NUM);
  999. INT_SET(recp->h_cycle, ARCH_CONVERT, cycle);
  1000. INT_SET(recp->h_version, ARCH_CONVERT,
  1001. XFS_SB_VERSION_HASLOGV2(&log->l_mp->m_sb) ? 2 : 1);
  1002. ASSIGN_ANY_LSN_DISK(recp->h_lsn, cycle, block);
  1003. ASSIGN_ANY_LSN_DISK(recp->h_tail_lsn, tail_cycle, tail_block);
  1004. INT_SET(recp->h_fmt, ARCH_CONVERT, XLOG_FMT);
  1005. memcpy(&recp->h_fs_uuid, &log->l_mp->m_sb.sb_uuid, sizeof(uuid_t));
  1006. }
  1007. STATIC int
  1008. xlog_write_log_records(
  1009. xlog_t *log,
  1010. int cycle,
  1011. int start_block,
  1012. int blocks,
  1013. int tail_cycle,
  1014. int tail_block)
  1015. {
  1016. xfs_caddr_t offset;
  1017. xfs_buf_t *bp;
  1018. int balign, ealign;
  1019. int sectbb = XLOG_SECTOR_ROUNDUP_BBCOUNT(log, 1);
  1020. int end_block = start_block + blocks;
  1021. int bufblks;
  1022. int error = 0;
  1023. int i, j = 0;
  1024. bufblks = 1 << ffs(blocks);
  1025. while (!(bp = xlog_get_bp(log, bufblks))) {
  1026. bufblks >>= 1;
  1027. if (bufblks <= log->l_sectbb_log)
  1028. return ENOMEM;
  1029. }
  1030. /* We may need to do a read at the start to fill in part of
  1031. * the buffer in the starting sector not covered by the first
  1032. * write below.
  1033. */
  1034. balign = XLOG_SECTOR_ROUNDDOWN_BLKNO(log, start_block);
  1035. if (balign != start_block) {
  1036. if ((error = xlog_bread(log, start_block, 1, bp))) {
  1037. xlog_put_bp(bp);
  1038. return error;
  1039. }
  1040. j = start_block - balign;
  1041. }
  1042. for (i = start_block; i < end_block; i += bufblks) {
  1043. int bcount, endcount;
  1044. bcount = min(bufblks, end_block - start_block);
  1045. endcount = bcount - j;
  1046. /* We may need to do a read at the end to fill in part of
  1047. * the buffer in the final sector not covered by the write.
  1048. * If this is the same sector as the above read, skip it.
  1049. */
  1050. ealign = XLOG_SECTOR_ROUNDDOWN_BLKNO(log, end_block);
  1051. if (j == 0 && (start_block + endcount > ealign)) {
  1052. offset = XFS_BUF_PTR(bp);
  1053. balign = BBTOB(ealign - start_block);
  1054. XFS_BUF_SET_PTR(bp, offset + balign, BBTOB(sectbb));
  1055. if ((error = xlog_bread(log, ealign, sectbb, bp)))
  1056. break;
  1057. XFS_BUF_SET_PTR(bp, offset, bufblks);
  1058. }
  1059. offset = xlog_align(log, start_block, endcount, bp);
  1060. for (; j < endcount; j++) {
  1061. xlog_add_record(log, offset, cycle, i+j,
  1062. tail_cycle, tail_block);
  1063. offset += BBSIZE;
  1064. }
  1065. error = xlog_bwrite(log, start_block, endcount, bp);
  1066. if (error)
  1067. break;
  1068. start_block += endcount;
  1069. j = 0;
  1070. }
  1071. xlog_put_bp(bp);
  1072. return error;
  1073. }
  1074. /*
  1075. * This routine is called to blow away any incomplete log writes out
  1076. * in front of the log head. We do this so that we won't become confused
  1077. * if we come up, write only a little bit more, and then crash again.
  1078. * If we leave the partial log records out there, this situation could
  1079. * cause us to think those partial writes are valid blocks since they
  1080. * have the current cycle number. We get rid of them by overwriting them
  1081. * with empty log records with the old cycle number rather than the
  1082. * current one.
  1083. *
  1084. * The tail lsn is passed in rather than taken from
  1085. * the log so that we will not write over the unmount record after a
  1086. * clean unmount in a 512 block log. Doing so would leave the log without
  1087. * any valid log records in it until a new one was written. If we crashed
  1088. * during that time we would not be able to recover.
  1089. */
  1090. STATIC int
  1091. xlog_clear_stale_blocks(
  1092. xlog_t *log,
  1093. xfs_lsn_t tail_lsn)
  1094. {
  1095. int tail_cycle, head_cycle;
  1096. int tail_block, head_block;
  1097. int tail_distance, max_distance;
  1098. int distance;
  1099. int error;
  1100. tail_cycle = CYCLE_LSN(tail_lsn);
  1101. tail_block = BLOCK_LSN(tail_lsn);
  1102. head_cycle = log->l_curr_cycle;
  1103. head_block = log->l_curr_block;
  1104. /*
  1105. * Figure out the distance between the new head of the log
  1106. * and the tail. We want to write over any blocks beyond the
  1107. * head that we may have written just before the crash, but
  1108. * we don't want to overwrite the tail of the log.
  1109. */
  1110. if (head_cycle == tail_cycle) {
  1111. /*
  1112. * The tail is behind the head in the physical log,
  1113. * so the distance from the head to the tail is the
  1114. * distance from the head to the end of the log plus
  1115. * the distance from the beginning of the log to the
  1116. * tail.
  1117. */
  1118. if (unlikely(head_block < tail_block || head_block >= log->l_logBBsize)) {
  1119. XFS_ERROR_REPORT("xlog_clear_stale_blocks(1)",
  1120. XFS_ERRLEVEL_LOW, log->l_mp);
  1121. return XFS_ERROR(EFSCORRUPTED);
  1122. }
  1123. tail_distance = tail_block + (log->l_logBBsize - head_block);
  1124. } else {
  1125. /*
  1126. * The head is behind the tail in the physical log,
  1127. * so the distance from the head to the tail is just
  1128. * the tail block minus the head block.
  1129. */
  1130. if (unlikely(head_block >= tail_block || head_cycle != (tail_cycle + 1))){
  1131. XFS_ERROR_REPORT("xlog_clear_stale_blocks(2)",
  1132. XFS_ERRLEVEL_LOW, log->l_mp);
  1133. return XFS_ERROR(EFSCORRUPTED);
  1134. }
  1135. tail_distance = tail_block - head_block;
  1136. }
  1137. /*
  1138. * If the head is right up against the tail, we can't clear
  1139. * anything.
  1140. */
  1141. if (tail_distance <= 0) {
  1142. ASSERT(tail_distance == 0);
  1143. return 0;
  1144. }
  1145. max_distance = XLOG_TOTAL_REC_SHIFT(log);
  1146. /*
  1147. * Take the smaller of the maximum amount of outstanding I/O
  1148. * we could have and the distance to the tail to clear out.
  1149. * We take the smaller so that we don't overwrite the tail and
  1150. * we don't waste all day writing from the head to the tail
  1151. * for no reason.
  1152. */
  1153. max_distance = MIN(max_distance, tail_distance);
  1154. if ((head_block + max_distance) <= log->l_logBBsize) {
  1155. /*
  1156. * We can stomp all the blocks we need to without
  1157. * wrapping around the end of the log. Just do it
  1158. * in a single write. Use the cycle number of the
  1159. * current cycle minus one so that the log will look like:
  1160. * n ... | n - 1 ...
  1161. */
  1162. error = xlog_write_log_records(log, (head_cycle - 1),
  1163. head_block, max_distance, tail_cycle,
  1164. tail_block);
  1165. if (error)
  1166. return error;
  1167. } else {
  1168. /*
  1169. * We need to wrap around the end of the physical log in
  1170. * order to clear all the blocks. Do it in two separate
  1171. * I/Os. The first write should be from the head to the
  1172. * end of the physical log, and it should use the current
  1173. * cycle number minus one just like above.
  1174. */
  1175. distance = log->l_logBBsize - head_block;
  1176. error = xlog_write_log_records(log, (head_cycle - 1),
  1177. head_block, distance, tail_cycle,
  1178. tail_block);
  1179. if (error)
  1180. return error;
  1181. /*
  1182. * Now write the blocks at the start of the physical log.
  1183. * This writes the remainder of the blocks we want to clear.
  1184. * It uses the current cycle number since we're now on the
  1185. * same cycle as the head so that we get:
  1186. * n ... n ... | n - 1 ...
  1187. * ^^^^^ blocks we're writing
  1188. */
  1189. distance = max_distance - (log->l_logBBsize - head_block);
  1190. error = xlog_write_log_records(log, head_cycle, 0, distance,
  1191. tail_cycle, tail_block);
  1192. if (error)
  1193. return error;
  1194. }
  1195. return 0;
  1196. }
  1197. /******************************************************************************
  1198. *
  1199. * Log recover routines
  1200. *
  1201. ******************************************************************************
  1202. */
  1203. STATIC xlog_recover_t *
  1204. xlog_recover_find_tid(
  1205. xlog_recover_t *q,
  1206. xlog_tid_t tid)
  1207. {
  1208. xlog_recover_t *p = q;
  1209. while (p != NULL) {
  1210. if (p->r_log_tid == tid)
  1211. break;
  1212. p = p->r_next;
  1213. }
  1214. return p;
  1215. }
  1216. STATIC void
  1217. xlog_recover_put_hashq(
  1218. xlog_recover_t **q,
  1219. xlog_recover_t *trans)
  1220. {
  1221. trans->r_next = *q;
  1222. *q = trans;
  1223. }
  1224. STATIC void
  1225. xlog_recover_add_item(
  1226. xlog_recover_item_t **itemq)
  1227. {
  1228. xlog_recover_item_t *item;
  1229. item = kmem_zalloc(sizeof(xlog_recover_item_t), KM_SLEEP);
  1230. xlog_recover_insert_item_backq(itemq, item);
  1231. }
  1232. STATIC int
  1233. xlog_recover_add_to_cont_trans(
  1234. xlog_recover_t *trans,
  1235. xfs_caddr_t dp,
  1236. int len)
  1237. {
  1238. xlog_recover_item_t *item;
  1239. xfs_caddr_t ptr, old_ptr;
  1240. int old_len;
  1241. item = trans->r_itemq;
  1242. if (item == 0) {
  1243. /* finish copying rest of trans header */
  1244. xlog_recover_add_item(&trans->r_itemq);
  1245. ptr = (xfs_caddr_t) &trans->r_theader +
  1246. sizeof(xfs_trans_header_t) - len;
  1247. memcpy(ptr, dp, len); /* d, s, l */
  1248. return 0;
  1249. }
  1250. item = item->ri_prev;
  1251. old_ptr = item->ri_buf[item->ri_cnt-1].i_addr;
  1252. old_len = item->ri_buf[item->ri_cnt-1].i_len;
  1253. ptr = kmem_realloc(old_ptr, len+old_len, old_len, 0u);
  1254. memcpy(&ptr[old_len], dp, len); /* d, s, l */
  1255. item->ri_buf[item->ri_cnt-1].i_len += len;
  1256. item->ri_buf[item->ri_cnt-1].i_addr = ptr;
  1257. return 0;
  1258. }
  1259. /*
  1260. * The next region to add is the start of a new region. It could be
  1261. * a whole region or it could be the first part of a new region. Because
  1262. * of this, the assumption here is that the type and size fields of all
  1263. * format structures fit into the first 32 bits of the structure.
  1264. *
  1265. * This works because all regions must be 32 bit aligned. Therefore, we
  1266. * either have both fields or we have neither field. In the case we have
  1267. * neither field, the data part of the region is zero length. We only have
  1268. * a log_op_header and can throw away the header since a new one will appear
  1269. * later. If we have at least 4 bytes, then we can determine how many regions
  1270. * will appear in the current log item.
  1271. */
  1272. STATIC int
  1273. xlog_recover_add_to_trans(
  1274. xlog_recover_t *trans,
  1275. xfs_caddr_t dp,
  1276. int len)
  1277. {
  1278. xfs_inode_log_format_t *in_f; /* any will do */
  1279. xlog_recover_item_t *item;
  1280. xfs_caddr_t ptr;
  1281. if (!len)
  1282. return 0;
  1283. item = trans->r_itemq;
  1284. if (item == 0) {
  1285. ASSERT(*(uint *)dp == XFS_TRANS_HEADER_MAGIC);
  1286. if (len == sizeof(xfs_trans_header_t))
  1287. xlog_recover_add_item(&trans->r_itemq);
  1288. memcpy(&trans->r_theader, dp, len); /* d, s, l */
  1289. return 0;
  1290. }
  1291. ptr = kmem_alloc(len, KM_SLEEP);
  1292. memcpy(ptr, dp, len);
  1293. in_f = (xfs_inode_log_format_t *)ptr;
  1294. if (item->ri_prev->ri_total != 0 &&
  1295. item->ri_prev->ri_total == item->ri_prev->ri_cnt) {
  1296. xlog_recover_add_item(&trans->r_itemq);
  1297. }
  1298. item = trans->r_itemq;
  1299. item = item->ri_prev;
  1300. if (item->ri_total == 0) { /* first region to be added */
  1301. item->ri_total = in_f->ilf_size;
  1302. ASSERT(item->ri_total <= XLOG_MAX_REGIONS_IN_ITEM);
  1303. item->ri_buf = kmem_zalloc((item->ri_total *
  1304. sizeof(xfs_log_iovec_t)), KM_SLEEP);
  1305. }
  1306. ASSERT(item->ri_total > item->ri_cnt);
  1307. /* Description region is ri_buf[0] */
  1308. item->ri_buf[item->ri_cnt].i_addr = ptr;
  1309. item->ri_buf[item->ri_cnt].i_len = len;
  1310. item->ri_cnt++;
  1311. return 0;
  1312. }
  1313. STATIC void
  1314. xlog_recover_new_tid(
  1315. xlog_recover_t **q,
  1316. xlog_tid_t tid,
  1317. xfs_lsn_t lsn)
  1318. {
  1319. xlog_recover_t *trans;
  1320. trans = kmem_zalloc(sizeof(xlog_recover_t), KM_SLEEP);
  1321. trans->r_log_tid = tid;
  1322. trans->r_lsn = lsn;
  1323. xlog_recover_put_hashq(q, trans);
  1324. }
  1325. STATIC int
  1326. xlog_recover_unlink_tid(
  1327. xlog_recover_t **q,
  1328. xlog_recover_t *trans)
  1329. {
  1330. xlog_recover_t *tp;
  1331. int found = 0;
  1332. ASSERT(trans != 0);
  1333. if (trans == *q) {
  1334. *q = (*q)->r_next;
  1335. } else {
  1336. tp = *q;
  1337. while (tp != 0) {
  1338. if (tp->r_next == trans) {
  1339. found = 1;
  1340. break;
  1341. }
  1342. tp = tp->r_next;
  1343. }
  1344. if (!found) {
  1345. xlog_warn(
  1346. "XFS: xlog_recover_unlink_tid: trans not found");
  1347. ASSERT(0);
  1348. return XFS_ERROR(EIO);
  1349. }
  1350. tp->r_next = tp->r_next->r_next;
  1351. }
  1352. return 0;
  1353. }
  1354. STATIC void
  1355. xlog_recover_insert_item_backq(
  1356. xlog_recover_item_t **q,
  1357. xlog_recover_item_t *item)
  1358. {
  1359. if (*q == 0) {
  1360. item->ri_prev = item->ri_next = item;
  1361. *q = item;
  1362. } else {
  1363. item->ri_next = *q;
  1364. item->ri_prev = (*q)->ri_prev;
  1365. (*q)->ri_prev = item;
  1366. item->ri_prev->ri_next = item;
  1367. }
  1368. }
  1369. STATIC void
  1370. xlog_recover_insert_item_frontq(
  1371. xlog_recover_item_t **q,
  1372. xlog_recover_item_t *item)
  1373. {
  1374. xlog_recover_insert_item_backq(q, item);
  1375. *q = item;
  1376. }
  1377. STATIC int
  1378. xlog_recover_reorder_trans(
  1379. xlog_t *log,
  1380. xlog_recover_t *trans)
  1381. {
  1382. xlog_recover_item_t *first_item, *itemq, *itemq_next;
  1383. xfs_buf_log_format_t *buf_f;
  1384. ushort flags = 0;
  1385. first_item = itemq = trans->r_itemq;
  1386. trans->r_itemq = NULL;
  1387. do {
  1388. itemq_next = itemq->ri_next;
  1389. buf_f = (xfs_buf_log_format_t *)itemq->ri_buf[0].i_addr;
  1390. switch (ITEM_TYPE(itemq)) {
  1391. case XFS_LI_BUF:
  1392. flags = buf_f->blf_flags;
  1393. if (!(flags & XFS_BLI_CANCEL)) {
  1394. xlog_recover_insert_item_frontq(&trans->r_itemq,
  1395. itemq);
  1396. break;
  1397. }
  1398. case XFS_LI_INODE:
  1399. case XFS_LI_DQUOT:
  1400. case XFS_LI_QUOTAOFF:
  1401. case XFS_LI_EFD:
  1402. case XFS_LI_EFI:
  1403. xlog_recover_insert_item_backq(&trans->r_itemq, itemq);
  1404. break;
  1405. default:
  1406. xlog_warn(
  1407. "XFS: xlog_recover_reorder_trans: unrecognized type of log operation");
  1408. ASSERT(0);
  1409. return XFS_ERROR(EIO);
  1410. }
  1411. itemq = itemq_next;
  1412. } while (first_item != itemq);
  1413. return 0;
  1414. }
  1415. /*
  1416. * Build up the table of buf cancel records so that we don't replay
  1417. * cancelled data in the second pass. For buffer records that are
  1418. * not cancel records, there is nothing to do here so we just return.
  1419. *
  1420. * If we get a cancel record which is already in the table, this indicates
  1421. * that the buffer was cancelled multiple times. In order to ensure
  1422. * that during pass 2 we keep the record in the table until we reach its
  1423. * last occurrence in the log, we keep a reference count in the cancel
  1424. * record in the table to tell us how many times we expect to see this
  1425. * record during the second pass.
  1426. */
  1427. STATIC void
  1428. xlog_recover_do_buffer_pass1(
  1429. xlog_t *log,
  1430. xfs_buf_log_format_t *buf_f)
  1431. {
  1432. xfs_buf_cancel_t *bcp;
  1433. xfs_buf_cancel_t *nextp;
  1434. xfs_buf_cancel_t *prevp;
  1435. xfs_buf_cancel_t **bucket;
  1436. xfs_daddr_t blkno = 0;
  1437. uint len = 0;
  1438. ushort flags = 0;
  1439. switch (buf_f->blf_type) {
  1440. case XFS_LI_BUF:
  1441. blkno = buf_f->blf_blkno;
  1442. len = buf_f->blf_len;
  1443. flags = buf_f->blf_flags;
  1444. break;
  1445. }
  1446. /*
  1447. * If this isn't a cancel buffer item, then just return.
  1448. */
  1449. if (!(flags & XFS_BLI_CANCEL))
  1450. return;
  1451. /*
  1452. * Insert an xfs_buf_cancel record into the hash table of
  1453. * them. If there is already an identical record, bump
  1454. * its reference count.
  1455. */
  1456. bucket = &log->l_buf_cancel_table[(__uint64_t)blkno %
  1457. XLOG_BC_TABLE_SIZE];
  1458. /*
  1459. * If the hash bucket is empty then just insert a new record into
  1460. * the bucket.
  1461. */
  1462. if (*bucket == NULL) {
  1463. bcp = (xfs_buf_cancel_t *)kmem_alloc(sizeof(xfs_buf_cancel_t),
  1464. KM_SLEEP);
  1465. bcp->bc_blkno = blkno;
  1466. bcp->bc_len = len;
  1467. bcp->bc_refcount = 1;
  1468. bcp->bc_next = NULL;
  1469. *bucket = bcp;
  1470. return;
  1471. }
  1472. /*
  1473. * The hash bucket is not empty, so search for duplicates of our
  1474. * record. If we find one them just bump its refcount. If not
  1475. * then add us at the end of the list.
  1476. */
  1477. prevp = NULL;
  1478. nextp = *bucket;
  1479. while (nextp != NULL) {
  1480. if (nextp->bc_blkno == blkno && nextp->bc_len == len) {
  1481. nextp->bc_refcount++;
  1482. return;
  1483. }
  1484. prevp = nextp;
  1485. nextp = nextp->bc_next;
  1486. }
  1487. ASSERT(prevp != NULL);
  1488. bcp = (xfs_buf_cancel_t *)kmem_alloc(sizeof(xfs_buf_cancel_t),
  1489. KM_SLEEP);
  1490. bcp->bc_blkno = blkno;
  1491. bcp->bc_len = len;
  1492. bcp->bc_refcount = 1;
  1493. bcp->bc_next = NULL;
  1494. prevp->bc_next = bcp;
  1495. }
  1496. /*
  1497. * Check to see whether the buffer being recovered has a corresponding
  1498. * entry in the buffer cancel record table. If it does then return 1
  1499. * so that it will be cancelled, otherwise return 0. If the buffer is
  1500. * actually a buffer cancel item (XFS_BLI_CANCEL is set), then decrement
  1501. * the refcount on the entry in the table and remove it from the table
  1502. * if this is the last reference.
  1503. *
  1504. * We remove the cancel record from the table when we encounter its
  1505. * last occurrence in the log so that if the same buffer is re-used
  1506. * again after its last cancellation we actually replay the changes
  1507. * made at that point.
  1508. */
  1509. STATIC int
  1510. xlog_check_buffer_cancelled(
  1511. xlog_t *log,
  1512. xfs_daddr_t blkno,
  1513. uint len,
  1514. ushort flags)
  1515. {
  1516. xfs_buf_cancel_t *bcp;
  1517. xfs_buf_cancel_t *prevp;
  1518. xfs_buf_cancel_t **bucket;
  1519. if (log->l_buf_cancel_table == NULL) {
  1520. /*
  1521. * There is nothing in the table built in pass one,
  1522. * so this buffer must not be cancelled.
  1523. */
  1524. ASSERT(!(flags & XFS_BLI_CANCEL));
  1525. return 0;
  1526. }
  1527. bucket = &log->l_buf_cancel_table[(__uint64_t)blkno %
  1528. XLOG_BC_TABLE_SIZE];
  1529. bcp = *bucket;
  1530. if (bcp == NULL) {
  1531. /*
  1532. * There is no corresponding entry in the table built
  1533. * in pass one, so this buffer has not been cancelled.
  1534. */
  1535. ASSERT(!(flags & XFS_BLI_CANCEL));
  1536. return 0;
  1537. }
  1538. /*
  1539. * Search for an entry in the buffer cancel table that
  1540. * matches our buffer.
  1541. */
  1542. prevp = NULL;
  1543. while (bcp != NULL) {
  1544. if (bcp->bc_blkno == blkno && bcp->bc_len == len) {
  1545. /*
  1546. * We've go a match, so return 1 so that the
  1547. * recovery of this buffer is cancelled.
  1548. * If this buffer is actually a buffer cancel
  1549. * log item, then decrement the refcount on the
  1550. * one in the table and remove it if this is the
  1551. * last reference.
  1552. */
  1553. if (flags & XFS_BLI_CANCEL) {
  1554. bcp->bc_refcount--;
  1555. if (bcp->bc_refcount == 0) {
  1556. if (prevp == NULL) {
  1557. *bucket = bcp->bc_next;
  1558. } else {
  1559. prevp->bc_next = bcp->bc_next;
  1560. }
  1561. kmem_free(bcp,
  1562. sizeof(xfs_buf_cancel_t));
  1563. }
  1564. }
  1565. return 1;
  1566. }
  1567. prevp = bcp;
  1568. bcp = bcp->bc_next;
  1569. }
  1570. /*
  1571. * We didn't find a corresponding entry in the table, so
  1572. * return 0 so that the buffer is NOT cancelled.
  1573. */
  1574. ASSERT(!(flags & XFS_BLI_CANCEL));
  1575. return 0;
  1576. }
  1577. STATIC int
  1578. xlog_recover_do_buffer_pass2(
  1579. xlog_t *log,
  1580. xfs_buf_log_format_t *buf_f)
  1581. {
  1582. xfs_daddr_t blkno = 0;
  1583. ushort flags = 0;
  1584. uint len = 0;
  1585. switch (buf_f->blf_type) {
  1586. case XFS_LI_BUF:
  1587. blkno = buf_f->blf_blkno;
  1588. flags = buf_f->blf_flags;
  1589. len = buf_f->blf_len;
  1590. break;
  1591. }
  1592. return xlog_check_buffer_cancelled(log, blkno, len, flags);
  1593. }
  1594. /*
  1595. * Perform recovery for a buffer full of inodes. In these buffers,
  1596. * the only data which should be recovered is that which corresponds
  1597. * to the di_next_unlinked pointers in the on disk inode structures.
  1598. * The rest of the data for the inodes is always logged through the
  1599. * inodes themselves rather than the inode buffer and is recovered
  1600. * in xlog_recover_do_inode_trans().
  1601. *
  1602. * The only time when buffers full of inodes are fully recovered is
  1603. * when the buffer is full of newly allocated inodes. In this case
  1604. * the buffer will not be marked as an inode buffer and so will be
  1605. * sent to xlog_recover_do_reg_buffer() below during recovery.
  1606. */
  1607. STATIC int
  1608. xlog_recover_do_inode_buffer(
  1609. xfs_mount_t *mp,
  1610. xlog_recover_item_t *item,
  1611. xfs_buf_t *bp,
  1612. xfs_buf_log_format_t *buf_f)
  1613. {
  1614. int i;
  1615. int item_index;
  1616. int bit;
  1617. int nbits;
  1618. int reg_buf_offset;
  1619. int reg_buf_bytes;
  1620. int next_unlinked_offset;
  1621. int inodes_per_buf;
  1622. xfs_agino_t *logged_nextp;
  1623. xfs_agino_t *buffer_nextp;
  1624. unsigned int *data_map = NULL;
  1625. unsigned int map_size = 0;
  1626. switch (buf_f->blf_type) {
  1627. case XFS_LI_BUF:
  1628. data_map = buf_f->blf_data_map;
  1629. map_size = buf_f->blf_map_size;
  1630. break;
  1631. }
  1632. /*
  1633. * Set the variables corresponding to the current region to
  1634. * 0 so that we'll initialize them on the first pass through
  1635. * the loop.
  1636. */
  1637. reg_buf_offset = 0;
  1638. reg_buf_bytes = 0;
  1639. bit = 0;
  1640. nbits = 0;
  1641. item_index = 0;
  1642. inodes_per_buf = XFS_BUF_COUNT(bp) >> mp->m_sb.sb_inodelog;
  1643. for (i = 0; i < inodes_per_buf; i++) {
  1644. next_unlinked_offset = (i * mp->m_sb.sb_inodesize) +
  1645. offsetof(xfs_dinode_t, di_next_unlinked);
  1646. while (next_unlinked_offset >=
  1647. (reg_buf_offset + reg_buf_bytes)) {
  1648. /*
  1649. * The next di_next_unlinked field is beyond
  1650. * the current logged region. Find the next
  1651. * logged region that contains or is beyond
  1652. * the current di_next_unlinked field.
  1653. */
  1654. bit += nbits;
  1655. bit = xfs_next_bit(data_map, map_size, bit);
  1656. /*
  1657. * If there are no more logged regions in the
  1658. * buffer, then we're done.
  1659. */
  1660. if (bit == -1) {
  1661. return 0;
  1662. }
  1663. nbits = xfs_contig_bits(data_map, map_size,
  1664. bit);
  1665. ASSERT(nbits > 0);
  1666. reg_buf_offset = bit << XFS_BLI_SHIFT;
  1667. reg_buf_bytes = nbits << XFS_BLI_SHIFT;
  1668. item_index++;
  1669. }
  1670. /*
  1671. * If the current logged region starts after the current
  1672. * di_next_unlinked field, then move on to the next
  1673. * di_next_unlinked field.
  1674. */
  1675. if (next_unlinked_offset < reg_buf_offset) {
  1676. continue;
  1677. }
  1678. ASSERT(item->ri_buf[item_index].i_addr != NULL);
  1679. ASSERT((item->ri_buf[item_index].i_len % XFS_BLI_CHUNK) == 0);
  1680. ASSERT((reg_buf_offset + reg_buf_bytes) <= XFS_BUF_COUNT(bp));
  1681. /*
  1682. * The current logged region contains a copy of the
  1683. * current di_next_unlinked field. Extract its value
  1684. * and copy it to the buffer copy.
  1685. */
  1686. logged_nextp = (xfs_agino_t *)
  1687. ((char *)(item->ri_buf[item_index].i_addr) +
  1688. (next_unlinked_offset - reg_buf_offset));
  1689. if (unlikely(*logged_nextp == 0)) {
  1690. xfs_fs_cmn_err(CE_ALERT, mp,
  1691. "bad inode buffer log record (ptr = 0x%p, bp = 0x%p). XFS trying to replay bad (0) inode di_next_unlinked field",
  1692. item, bp);
  1693. XFS_ERROR_REPORT("xlog_recover_do_inode_buf",
  1694. XFS_ERRLEVEL_LOW, mp);
  1695. return XFS_ERROR(EFSCORRUPTED);
  1696. }
  1697. buffer_nextp = (xfs_agino_t *)xfs_buf_offset(bp,
  1698. next_unlinked_offset);
  1699. *buffer_nextp = *logged_nextp;
  1700. }
  1701. return 0;
  1702. }
  1703. /*
  1704. * Perform a 'normal' buffer recovery. Each logged region of the
  1705. * buffer should be copied over the corresponding region in the
  1706. * given buffer. The bitmap in the buf log format structure indicates
  1707. * where to place the logged data.
  1708. */
  1709. /*ARGSUSED*/
  1710. STATIC void
  1711. xlog_recover_do_reg_buffer(
  1712. xfs_mount_t *mp,
  1713. xlog_recover_item_t *item,
  1714. xfs_buf_t *bp,
  1715. xfs_buf_log_format_t *buf_f)
  1716. {
  1717. int i;
  1718. int bit;
  1719. int nbits;
  1720. unsigned int *data_map = NULL;
  1721. unsigned int map_size = 0;
  1722. int error;
  1723. switch (buf_f->blf_type) {
  1724. case XFS_LI_BUF:
  1725. data_map = buf_f->blf_data_map;
  1726. map_size = buf_f->blf_map_size;
  1727. break;
  1728. }
  1729. bit = 0;
  1730. i = 1; /* 0 is the buf format structure */
  1731. while (1) {
  1732. bit = xfs_next_bit(data_map, map_size, bit);
  1733. if (bit == -1)
  1734. break;
  1735. nbits = xfs_contig_bits(data_map, map_size, bit);
  1736. ASSERT(nbits > 0);
  1737. ASSERT(item->ri_buf[i].i_addr != 0);
  1738. ASSERT(item->ri_buf[i].i_len % XFS_BLI_CHUNK == 0);
  1739. ASSERT(XFS_BUF_COUNT(bp) >=
  1740. ((uint)bit << XFS_BLI_SHIFT)+(nbits<<XFS_BLI_SHIFT));
  1741. /*
  1742. * Do a sanity check if this is a dquot buffer. Just checking
  1743. * the first dquot in the buffer should do. XXXThis is
  1744. * probably a good thing to do for other buf types also.
  1745. */
  1746. error = 0;
  1747. if (buf_f->blf_flags &
  1748. (XFS_BLI_UDQUOT_BUF|XFS_BLI_PDQUOT_BUF|XFS_BLI_GDQUOT_BUF)) {
  1749. error = xfs_qm_dqcheck((xfs_disk_dquot_t *)
  1750. item->ri_buf[i].i_addr,
  1751. -1, 0, XFS_QMOPT_DOWARN,
  1752. "dquot_buf_recover");
  1753. }
  1754. if (!error)
  1755. memcpy(xfs_buf_offset(bp,
  1756. (uint)bit << XFS_BLI_SHIFT), /* dest */
  1757. item->ri_buf[i].i_addr, /* source */
  1758. nbits<<XFS_BLI_SHIFT); /* length */
  1759. i++;
  1760. bit += nbits;
  1761. }
  1762. /* Shouldn't be any more regions */
  1763. ASSERT(i == item->ri_total);
  1764. }
  1765. /*
  1766. * Do some primitive error checking on ondisk dquot data structures.
  1767. */
  1768. int
  1769. xfs_qm_dqcheck(
  1770. xfs_disk_dquot_t *ddq,
  1771. xfs_dqid_t id,
  1772. uint type, /* used only when IO_dorepair is true */
  1773. uint flags,
  1774. char *str)
  1775. {
  1776. xfs_dqblk_t *d = (xfs_dqblk_t *)ddq;
  1777. int errs = 0;
  1778. /*
  1779. * We can encounter an uninitialized dquot buffer for 2 reasons:
  1780. * 1. If we crash while deleting the quotainode(s), and those blks got
  1781. * used for user data. This is because we take the path of regular
  1782. * file deletion; however, the size field of quotainodes is never
  1783. * updated, so all the tricks that we play in itruncate_finish
  1784. * don't quite matter.
  1785. *
  1786. * 2. We don't play the quota buffers when there's a quotaoff logitem.
  1787. * But the allocation will be replayed so we'll end up with an
  1788. * uninitialized quota block.
  1789. *
  1790. * This is all fine; things are still consistent, and we haven't lost
  1791. * any quota information. Just don't complain about bad dquot blks.
  1792. */
  1793. if (be16_to_cpu(ddq->d_magic) != XFS_DQUOT_MAGIC) {
  1794. if (flags & XFS_QMOPT_DOWARN)
  1795. cmn_err(CE_ALERT,
  1796. "%s : XFS dquot ID 0x%x, magic 0x%x != 0x%x",
  1797. str, id, be16_to_cpu(ddq->d_magic), XFS_DQUOT_MAGIC);
  1798. errs++;
  1799. }
  1800. if (ddq->d_version != XFS_DQUOT_VERSION) {
  1801. if (flags & XFS_QMOPT_DOWARN)
  1802. cmn_err(CE_ALERT,
  1803. "%s : XFS dquot ID 0x%x, version 0x%x != 0x%x",
  1804. str, id, ddq->d_version, XFS_DQUOT_VERSION);
  1805. errs++;
  1806. }
  1807. if (ddq->d_flags != XFS_DQ_USER &&
  1808. ddq->d_flags != XFS_DQ_PROJ &&
  1809. ddq->d_flags != XFS_DQ_GROUP) {
  1810. if (flags & XFS_QMOPT_DOWARN)
  1811. cmn_err(CE_ALERT,
  1812. "%s : XFS dquot ID 0x%x, unknown flags 0x%x",
  1813. str, id, ddq->d_flags);
  1814. errs++;
  1815. }
  1816. if (id != -1 && id != be32_to_cpu(ddq->d_id)) {
  1817. if (flags & XFS_QMOPT_DOWARN)
  1818. cmn_err(CE_ALERT,
  1819. "%s : ondisk-dquot 0x%p, ID mismatch: "
  1820. "0x%x expected, found id 0x%x",
  1821. str, ddq, id, be32_to_cpu(ddq->d_id));
  1822. errs++;
  1823. }
  1824. if (!errs && ddq->d_id) {
  1825. if (ddq->d_blk_softlimit &&
  1826. be64_to_cpu(ddq->d_bcount) >=
  1827. be64_to_cpu(ddq->d_blk_softlimit)) {
  1828. if (!ddq->d_btimer) {
  1829. if (flags & XFS_QMOPT_DOWARN)
  1830. cmn_err(CE_ALERT,
  1831. "%s : Dquot ID 0x%x (0x%p) "
  1832. "BLK TIMER NOT STARTED",
  1833. str, (int)be32_to_cpu(ddq->d_id), ddq);
  1834. errs++;
  1835. }
  1836. }
  1837. if (ddq->d_ino_softlimit &&
  1838. be64_to_cpu(ddq->d_icount) >=
  1839. be64_to_cpu(ddq->d_ino_softlimit)) {
  1840. if (!ddq->d_itimer) {
  1841. if (flags & XFS_QMOPT_DOWARN)
  1842. cmn_err(CE_ALERT,
  1843. "%s : Dquot ID 0x%x (0x%p) "
  1844. "INODE TIMER NOT STARTED",
  1845. str, (int)be32_to_cpu(ddq->d_id), ddq);
  1846. errs++;
  1847. }
  1848. }
  1849. if (ddq->d_rtb_softlimit &&
  1850. be64_to_cpu(ddq->d_rtbcount) >=
  1851. be64_to_cpu(ddq->d_rtb_softlimit)) {
  1852. if (!ddq->d_rtbtimer) {
  1853. if (flags & XFS_QMOPT_DOWARN)
  1854. cmn_err(CE_ALERT,
  1855. "%s : Dquot ID 0x%x (0x%p) "
  1856. "RTBLK TIMER NOT STARTED",
  1857. str, (int)be32_to_cpu(ddq->d_id), ddq);
  1858. errs++;
  1859. }
  1860. }
  1861. }
  1862. if (!errs || !(flags & XFS_QMOPT_DQREPAIR))
  1863. return errs;
  1864. if (flags & XFS_QMOPT_DOWARN)
  1865. cmn_err(CE_NOTE, "Re-initializing dquot ID 0x%x", id);
  1866. /*
  1867. * Typically, a repair is only requested by quotacheck.
  1868. */
  1869. ASSERT(id != -1);
  1870. ASSERT(flags & XFS_QMOPT_DQREPAIR);
  1871. memset(d, 0, sizeof(xfs_dqblk_t));
  1872. d->dd_diskdq.d_magic = cpu_to_be16(XFS_DQUOT_MAGIC);
  1873. d->dd_diskdq.d_version = XFS_DQUOT_VERSION;
  1874. d->dd_diskdq.d_flags = type;
  1875. d->dd_diskdq.d_id = cpu_to_be32(id);
  1876. return errs;
  1877. }
  1878. /*
  1879. * Perform a dquot buffer recovery.
  1880. * Simple algorithm: if we have found a QUOTAOFF logitem of the same type
  1881. * (ie. USR or GRP), then just toss this buffer away; don't recover it.
  1882. * Else, treat it as a regular buffer and do recovery.
  1883. */
  1884. STATIC void
  1885. xlog_recover_do_dquot_buffer(
  1886. xfs_mount_t *mp,
  1887. xlog_t *log,
  1888. xlog_recover_item_t *item,
  1889. xfs_buf_t *bp,
  1890. xfs_buf_log_format_t *buf_f)
  1891. {
  1892. uint type;
  1893. /*
  1894. * Filesystems are required to send in quota flags at mount time.
  1895. */
  1896. if (mp->m_qflags == 0) {
  1897. return;
  1898. }
  1899. type = 0;
  1900. if (buf_f->blf_flags & XFS_BLI_UDQUOT_BUF)
  1901. type |= XFS_DQ_USER;
  1902. if (buf_f->blf_flags & XFS_BLI_PDQUOT_BUF)
  1903. type |= XFS_DQ_PROJ;
  1904. if (buf_f->blf_flags & XFS_BLI_GDQUOT_BUF)
  1905. type |= XFS_DQ_GROUP;
  1906. /*
  1907. * This type of quotas was turned off, so ignore this buffer
  1908. */
  1909. if (log->l_quotaoffs_flag & type)
  1910. return;
  1911. xlog_recover_do_reg_buffer(mp, item, bp, buf_f);
  1912. }
  1913. /*
  1914. * This routine replays a modification made to a buffer at runtime.
  1915. * There are actually two types of buffer, regular and inode, which
  1916. * are handled differently. Inode buffers are handled differently
  1917. * in that we only recover a specific set of data from them, namely
  1918. * the inode di_next_unlinked fields. This is because all other inode
  1919. * data is actually logged via inode records and any data we replay
  1920. * here which overlaps that may be stale.
  1921. *
  1922. * When meta-data buffers are freed at run time we log a buffer item
  1923. * with the XFS_BLI_CANCEL bit set to indicate that previous copies
  1924. * of the buffer in the log should not be replayed at recovery time.
  1925. * This is so that if the blocks covered by the buffer are reused for
  1926. * file data before we crash we don't end up replaying old, freed
  1927. * meta-data into a user's file.
  1928. *
  1929. * To handle the cancellation of buffer log items, we make two passes
  1930. * over the log during recovery. During the first we build a table of
  1931. * those buffers which have been cancelled, and during the second we
  1932. * only replay those buffers which do not have corresponding cancel
  1933. * records in the table. See xlog_recover_do_buffer_pass[1,2] above
  1934. * for more details on the implementation of the table of cancel records.
  1935. */
  1936. STATIC int
  1937. xlog_recover_do_buffer_trans(
  1938. xlog_t *log,
  1939. xlog_recover_item_t *item,
  1940. int pass)
  1941. {
  1942. xfs_buf_log_format_t *buf_f;
  1943. xfs_mount_t *mp;
  1944. xfs_buf_t *bp;
  1945. int error;
  1946. int cancel;
  1947. xfs_daddr_t blkno;
  1948. int len;
  1949. ushort flags;
  1950. buf_f = (xfs_buf_log_format_t *)item->ri_buf[0].i_addr;
  1951. if (pass == XLOG_RECOVER_PASS1) {
  1952. /*
  1953. * In this pass we're only looking for buf items
  1954. * with the XFS_BLI_CANCEL bit set.
  1955. */
  1956. xlog_recover_do_buffer_pass1(log, buf_f);
  1957. return 0;
  1958. } else {
  1959. /*
  1960. * In this pass we want to recover all the buffers
  1961. * which have not been cancelled and are not
  1962. * cancellation buffers themselves. The routine
  1963. * we call here will tell us whether or not to
  1964. * continue with the replay of this buffer.
  1965. */
  1966. cancel = xlog_recover_do_buffer_pass2(log, buf_f);
  1967. if (cancel) {
  1968. return 0;
  1969. }
  1970. }
  1971. switch (buf_f->blf_type) {
  1972. case XFS_LI_BUF:
  1973. blkno = buf_f->blf_blkno;
  1974. len = buf_f->blf_len;
  1975. flags = buf_f->blf_flags;
  1976. break;
  1977. default:
  1978. xfs_fs_cmn_err(CE_ALERT, log->l_mp,
  1979. "xfs_log_recover: unknown buffer type 0x%x, logdev %s",
  1980. buf_f->blf_type, log->l_mp->m_logname ?
  1981. log->l_mp->m_logname : "internal");
  1982. XFS_ERROR_REPORT("xlog_recover_do_buffer_trans",
  1983. XFS_ERRLEVEL_LOW, log->l_mp);
  1984. return XFS_ERROR(EFSCORRUPTED);
  1985. }
  1986. mp = log->l_mp;
  1987. if (flags & XFS_BLI_INODE_BUF) {
  1988. bp = xfs_buf_read_flags(mp->m_ddev_targp, blkno, len,
  1989. XFS_BUF_LOCK);
  1990. } else {
  1991. bp = xfs_buf_read(mp->m_ddev_targp, blkno, len, 0);
  1992. }
  1993. if (XFS_BUF_ISERROR(bp)) {
  1994. xfs_ioerror_alert("xlog_recover_do..(read#1)", log->l_mp,
  1995. bp, blkno);
  1996. error = XFS_BUF_GETERROR(bp);
  1997. xfs_buf_relse(bp);
  1998. return error;
  1999. }
  2000. error = 0;
  2001. if (flags & XFS_BLI_INODE_BUF) {
  2002. error = xlog_recover_do_inode_buffer(mp, item, bp, buf_f);
  2003. } else if (flags &
  2004. (XFS_BLI_UDQUOT_BUF|XFS_BLI_PDQUOT_BUF|XFS_BLI_GDQUOT_BUF)) {
  2005. xlog_recover_do_dquot_buffer(mp, log, item, bp, buf_f);
  2006. } else {
  2007. xlog_recover_do_reg_buffer(mp, item, bp, buf_f);
  2008. }
  2009. if (error)
  2010. return XFS_ERROR(error);
  2011. /*
  2012. * Perform delayed write on the buffer. Asynchronous writes will be
  2013. * slower when taking into account all the buffers to be flushed.
  2014. *
  2015. * Also make sure that only inode buffers with good sizes stay in
  2016. * the buffer cache. The kernel moves inodes in buffers of 1 block
  2017. * or XFS_INODE_CLUSTER_SIZE bytes, whichever is bigger. The inode
  2018. * buffers in the log can be a different size if the log was generated
  2019. * by an older kernel using unclustered inode buffers or a newer kernel
  2020. * running with a different inode cluster size. Regardless, if the
  2021. * the inode buffer size isn't MAX(blocksize, XFS_INODE_CLUSTER_SIZE)
  2022. * for *our* value of XFS_INODE_CLUSTER_SIZE, then we need to keep
  2023. * the buffer out of the buffer cache so that the buffer won't
  2024. * overlap with future reads of those inodes.
  2025. */
  2026. if (XFS_DINODE_MAGIC ==
  2027. INT_GET(*((__uint16_t *)(xfs_buf_offset(bp, 0))), ARCH_CONVERT) &&
  2028. (XFS_BUF_COUNT(bp) != MAX(log->l_mp->m_sb.sb_blocksize,
  2029. (__uint32_t)XFS_INODE_CLUSTER_SIZE(log->l_mp)))) {
  2030. XFS_BUF_STALE(bp);
  2031. error = xfs_bwrite(mp, bp);
  2032. } else {
  2033. ASSERT(XFS_BUF_FSPRIVATE(bp, void *) == NULL ||
  2034. XFS_BUF_FSPRIVATE(bp, xfs_mount_t *) == mp);
  2035. XFS_BUF_SET_FSPRIVATE(bp, mp);
  2036. XFS_BUF_SET_IODONE_FUNC(bp, xlog_recover_iodone);
  2037. xfs_bdwrite(mp, bp);
  2038. }
  2039. return (error);
  2040. }
  2041. STATIC int
  2042. xlog_recover_do_inode_trans(
  2043. xlog_t *log,
  2044. xlog_recover_item_t *item,
  2045. int pass)
  2046. {
  2047. xfs_inode_log_format_t *in_f;
  2048. xfs_mount_t *mp;
  2049. xfs_buf_t *bp;
  2050. xfs_imap_t imap;
  2051. xfs_dinode_t *dip;
  2052. xfs_ino_t ino;
  2053. int len;
  2054. xfs_caddr_t src;
  2055. xfs_caddr_t dest;
  2056. int error;
  2057. int attr_index;
  2058. uint fields;
  2059. xfs_dinode_core_t *dicp;
  2060. int need_free = 0;
  2061. if (pass == XLOG_RECOVER_PASS1) {
  2062. return 0;
  2063. }
  2064. if (item->ri_buf[0].i_len == sizeof(xfs_inode_log_format_t)) {
  2065. in_f = (xfs_inode_log_format_t *)item->ri_buf[0].i_addr;
  2066. } else {
  2067. in_f = (xfs_inode_log_format_t *)kmem_alloc(
  2068. sizeof(xfs_inode_log_format_t), KM_SLEEP);
  2069. need_free = 1;
  2070. error = xfs_inode_item_format_convert(&item->ri_buf[0], in_f);
  2071. if (error)
  2072. goto error;
  2073. }
  2074. ino = in_f->ilf_ino;
  2075. mp = log->l_mp;
  2076. if (ITEM_TYPE(item) == XFS_LI_INODE) {
  2077. imap.im_blkno = (xfs_daddr_t)in_f->ilf_blkno;
  2078. imap.im_len = in_f->ilf_len;
  2079. imap.im_boffset = in_f->ilf_boffset;
  2080. } else {
  2081. /*
  2082. * It's an old inode format record. We don't know where
  2083. * its cluster is located on disk, and we can't allow
  2084. * xfs_imap() to figure it out because the inode btrees
  2085. * are not ready to be used. Therefore do not pass the
  2086. * XFS_IMAP_LOOKUP flag to xfs_imap(). This will give
  2087. * us only the single block in which the inode lives
  2088. * rather than its cluster, so we must make sure to
  2089. * invalidate the buffer when we write it out below.
  2090. */
  2091. imap.im_blkno = 0;
  2092. xfs_imap(log->l_mp, NULL, ino, &imap, 0);
  2093. }
  2094. /*
  2095. * Inode buffers can be freed, look out for it,
  2096. * and do not replay the inode.
  2097. */
  2098. if (xlog_check_buffer_cancelled(log, imap.im_blkno, imap.im_len, 0)) {
  2099. error = 0;
  2100. goto error;
  2101. }
  2102. bp = xfs_buf_read_flags(mp->m_ddev_targp, imap.im_blkno, imap.im_len,
  2103. XFS_BUF_LOCK);
  2104. if (XFS_BUF_ISERROR(bp)) {
  2105. xfs_ioerror_alert("xlog_recover_do..(read#2)", mp,
  2106. bp, imap.im_blkno);
  2107. error = XFS_BUF_GETERROR(bp);
  2108. xfs_buf_relse(bp);
  2109. goto error;
  2110. }
  2111. error = 0;
  2112. ASSERT(in_f->ilf_fields & XFS_ILOG_CORE);
  2113. dip = (xfs_dinode_t *)xfs_buf_offset(bp, imap.im_boffset);
  2114. /*
  2115. * Make sure the place we're flushing out to really looks
  2116. * like an inode!
  2117. */
  2118. if (unlikely(INT_GET(dip->di_core.di_magic, ARCH_CONVERT) != XFS_DINODE_MAGIC)) {
  2119. xfs_buf_relse(bp);
  2120. xfs_fs_cmn_err(CE_ALERT, mp,
  2121. "xfs_inode_recover: Bad inode magic number, dino ptr = 0x%p, dino bp = 0x%p, ino = %Ld",
  2122. dip, bp, ino);
  2123. XFS_ERROR_REPORT("xlog_recover_do_inode_trans(1)",
  2124. XFS_ERRLEVEL_LOW, mp);
  2125. error = EFSCORRUPTED;
  2126. goto error;
  2127. }
  2128. dicp = (xfs_dinode_core_t*)(item->ri_buf[1].i_addr);
  2129. if (unlikely(dicp->di_magic != XFS_DINODE_MAGIC)) {
  2130. xfs_buf_relse(bp);
  2131. xfs_fs_cmn_err(CE_ALERT, mp,
  2132. "xfs_inode_recover: Bad inode log record, rec ptr 0x%p, ino %Ld",
  2133. item, ino);
  2134. XFS_ERROR_REPORT("xlog_recover_do_inode_trans(2)",
  2135. XFS_ERRLEVEL_LOW, mp);
  2136. error = EFSCORRUPTED;
  2137. goto error;
  2138. }
  2139. /* Skip replay when the on disk inode is newer than the log one */
  2140. if (dicp->di_flushiter <
  2141. INT_GET(dip->di_core.di_flushiter, ARCH_CONVERT)) {
  2142. /*
  2143. * Deal with the wrap case, DI_MAX_FLUSH is less
  2144. * than smaller numbers
  2145. */
  2146. if ((INT_GET(dip->di_core.di_flushiter, ARCH_CONVERT)
  2147. == DI_MAX_FLUSH) &&
  2148. (dicp->di_flushiter < (DI_MAX_FLUSH>>1))) {
  2149. /* do nothing */
  2150. } else {
  2151. xfs_buf_relse(bp);
  2152. error = 0;
  2153. goto error;
  2154. }
  2155. }
  2156. /* Take the opportunity to reset the flush iteration count */
  2157. dicp->di_flushiter = 0;
  2158. if (unlikely((dicp->di_mode & S_IFMT) == S_IFREG)) {
  2159. if ((dicp->di_format != XFS_DINODE_FMT_EXTENTS) &&
  2160. (dicp->di_format != XFS_DINODE_FMT_BTREE)) {
  2161. XFS_CORRUPTION_ERROR("xlog_recover_do_inode_trans(3)",
  2162. XFS_ERRLEVEL_LOW, mp, dicp);
  2163. xfs_buf_relse(bp);
  2164. xfs_fs_cmn_err(CE_ALERT, mp,
  2165. "xfs_inode_recover: Bad regular inode log record, rec ptr 0x%p, ino ptr = 0x%p, ino bp = 0x%p, ino %Ld",
  2166. item, dip, bp, ino);
  2167. error = EFSCORRUPTED;
  2168. goto error;
  2169. }
  2170. } else if (unlikely((dicp->di_mode & S_IFMT) == S_IFDIR)) {
  2171. if ((dicp->di_format != XFS_DINODE_FMT_EXTENTS) &&
  2172. (dicp->di_format != XFS_DINODE_FMT_BTREE) &&
  2173. (dicp->di_format != XFS_DINODE_FMT_LOCAL)) {
  2174. XFS_CORRUPTION_ERROR("xlog_recover_do_inode_trans(4)",
  2175. XFS_ERRLEVEL_LOW, mp, dicp);
  2176. xfs_buf_relse(bp);
  2177. xfs_fs_cmn_err(CE_ALERT, mp,
  2178. "xfs_inode_recover: Bad dir inode log record, rec ptr 0x%p, ino ptr = 0x%p, ino bp = 0x%p, ino %Ld",
  2179. item, dip, bp, ino);
  2180. error = EFSCORRUPTED;
  2181. goto error;
  2182. }
  2183. }
  2184. if (unlikely(dicp->di_nextents + dicp->di_anextents > dicp->di_nblocks)){
  2185. XFS_CORRUPTION_ERROR("xlog_recover_do_inode_trans(5)",
  2186. XFS_ERRLEVEL_LOW, mp, dicp);
  2187. xfs_buf_relse(bp);
  2188. xfs_fs_cmn_err(CE_ALERT, mp,
  2189. "xfs_inode_recover: Bad inode log record, rec ptr 0x%p, dino ptr 0x%p, dino bp 0x%p, ino %Ld, total extents = %d, nblocks = %Ld",
  2190. item, dip, bp, ino,
  2191. dicp->di_nextents + dicp->di_anextents,
  2192. dicp->di_nblocks);
  2193. error = EFSCORRUPTED;
  2194. goto error;
  2195. }
  2196. if (unlikely(dicp->di_forkoff > mp->m_sb.sb_inodesize)) {
  2197. XFS_CORRUPTION_ERROR("xlog_recover_do_inode_trans(6)",
  2198. XFS_ERRLEVEL_LOW, mp, dicp);
  2199. xfs_buf_relse(bp);
  2200. xfs_fs_cmn_err(CE_ALERT, mp,
  2201. "xfs_inode_recover: Bad inode log rec ptr 0x%p, dino ptr 0x%p, dino bp 0x%p, ino %Ld, forkoff 0x%x",
  2202. item, dip, bp, ino, dicp->di_forkoff);
  2203. error = EFSCORRUPTED;
  2204. goto error;
  2205. }
  2206. if (unlikely(item->ri_buf[1].i_len > sizeof(xfs_dinode_core_t))) {
  2207. XFS_CORRUPTION_ERROR("xlog_recover_do_inode_trans(7)",
  2208. XFS_ERRLEVEL_LOW, mp, dicp);
  2209. xfs_buf_relse(bp);
  2210. xfs_fs_cmn_err(CE_ALERT, mp,
  2211. "xfs_inode_recover: Bad inode log record length %d, rec ptr 0x%p",
  2212. item->ri_buf[1].i_len, item);
  2213. error = EFSCORRUPTED;
  2214. goto error;
  2215. }
  2216. /* The core is in in-core format */
  2217. xfs_xlate_dinode_core((xfs_caddr_t)&dip->di_core,
  2218. (xfs_dinode_core_t*)item->ri_buf[1].i_addr, -1);
  2219. /* the rest is in on-disk format */
  2220. if (item->ri_buf[1].i_len > sizeof(xfs_dinode_core_t)) {
  2221. memcpy((xfs_caddr_t) dip + sizeof(xfs_dinode_core_t),
  2222. item->ri_buf[1].i_addr + sizeof(xfs_dinode_core_t),
  2223. item->ri_buf[1].i_len - sizeof(xfs_dinode_core_t));
  2224. }
  2225. fields = in_f->ilf_fields;
  2226. switch (fields & (XFS_ILOG_DEV | XFS_ILOG_UUID)) {
  2227. case XFS_ILOG_DEV:
  2228. INT_SET(dip->di_u.di_dev, ARCH_CONVERT, in_f->ilf_u.ilfu_rdev);
  2229. break;
  2230. case XFS_ILOG_UUID:
  2231. dip->di_u.di_muuid = in_f->ilf_u.ilfu_uuid;
  2232. break;
  2233. }
  2234. if (in_f->ilf_size == 2)
  2235. goto write_inode_buffer;
  2236. len = item->ri_buf[2].i_len;
  2237. src = item->ri_buf[2].i_addr;
  2238. ASSERT(in_f->ilf_size <= 4);
  2239. ASSERT((in_f->ilf_size == 3) || (fields & XFS_ILOG_AFORK));
  2240. ASSERT(!(fields & XFS_ILOG_DFORK) ||
  2241. (len == in_f->ilf_dsize));
  2242. switch (fields & XFS_ILOG_DFORK) {
  2243. case XFS_ILOG_DDATA:
  2244. case XFS_ILOG_DEXT:
  2245. memcpy(&dip->di_u, src, len);
  2246. break;
  2247. case XFS_ILOG_DBROOT:
  2248. xfs_bmbt_to_bmdr((xfs_bmbt_block_t *)src, len,
  2249. &(dip->di_u.di_bmbt),
  2250. XFS_DFORK_DSIZE(dip, mp));
  2251. break;
  2252. default:
  2253. /*
  2254. * There are no data fork flags set.
  2255. */
  2256. ASSERT((fields & XFS_ILOG_DFORK) == 0);
  2257. break;
  2258. }
  2259. /*
  2260. * If we logged any attribute data, recover it. There may or
  2261. * may not have been any other non-core data logged in this
  2262. * transaction.
  2263. */
  2264. if (in_f->ilf_fields & XFS_ILOG_AFORK) {
  2265. if (in_f->ilf_fields & XFS_ILOG_DFORK) {
  2266. attr_index = 3;
  2267. } else {
  2268. attr_index = 2;
  2269. }
  2270. len = item->ri_buf[attr_index].i_len;
  2271. src = item->ri_buf[attr_index].i_addr;
  2272. ASSERT(len == in_f->ilf_asize);
  2273. switch (in_f->ilf_fields & XFS_ILOG_AFORK) {
  2274. case XFS_ILOG_ADATA:
  2275. case XFS_ILOG_AEXT:
  2276. dest = XFS_DFORK_APTR(dip);
  2277. ASSERT(len <= XFS_DFORK_ASIZE(dip, mp));
  2278. memcpy(dest, src, len);
  2279. break;
  2280. case XFS_ILOG_ABROOT:
  2281. dest = XFS_DFORK_APTR(dip);
  2282. xfs_bmbt_to_bmdr((xfs_bmbt_block_t *)src, len,
  2283. (xfs_bmdr_block_t*)dest,
  2284. XFS_DFORK_ASIZE(dip, mp));
  2285. break;
  2286. default:
  2287. xlog_warn("XFS: xlog_recover_do_inode_trans: Invalid flag");
  2288. ASSERT(0);
  2289. xfs_buf_relse(bp);
  2290. error = EIO;
  2291. goto error;
  2292. }
  2293. }
  2294. write_inode_buffer:
  2295. if (ITEM_TYPE(item) == XFS_LI_INODE) {
  2296. ASSERT(XFS_BUF_FSPRIVATE(bp, void *) == NULL ||
  2297. XFS_BUF_FSPRIVATE(bp, xfs_mount_t *) == mp);
  2298. XFS_BUF_SET_FSPRIVATE(bp, mp);
  2299. XFS_BUF_SET_IODONE_FUNC(bp, xlog_recover_iodone);
  2300. xfs_bdwrite(mp, bp);
  2301. } else {
  2302. XFS_BUF_STALE(bp);
  2303. error = xfs_bwrite(mp, bp);
  2304. }
  2305. error:
  2306. if (need_free)
  2307. kmem_free(in_f, sizeof(*in_f));
  2308. return XFS_ERROR(error);
  2309. }
  2310. /*
  2311. * Recover QUOTAOFF records. We simply make a note of it in the xlog_t
  2312. * structure, so that we know not to do any dquot item or dquot buffer recovery,
  2313. * of that type.
  2314. */
  2315. STATIC int
  2316. xlog_recover_do_quotaoff_trans(
  2317. xlog_t *log,
  2318. xlog_recover_item_t *item,
  2319. int pass)
  2320. {
  2321. xfs_qoff_logformat_t *qoff_f;
  2322. if (pass == XLOG_RECOVER_PASS2) {
  2323. return (0);
  2324. }
  2325. qoff_f = (xfs_qoff_logformat_t *)item->ri_buf[0].i_addr;
  2326. ASSERT(qoff_f);
  2327. /*
  2328. * The logitem format's flag tells us if this was user quotaoff,
  2329. * group/project quotaoff or both.
  2330. */
  2331. if (qoff_f->qf_flags & XFS_UQUOTA_ACCT)
  2332. log->l_quotaoffs_flag |= XFS_DQ_USER;
  2333. if (qoff_f->qf_flags & XFS_PQUOTA_ACCT)
  2334. log->l_quotaoffs_flag |= XFS_DQ_PROJ;
  2335. if (qoff_f->qf_flags & XFS_GQUOTA_ACCT)
  2336. log->l_quotaoffs_flag |= XFS_DQ_GROUP;
  2337. return (0);
  2338. }
  2339. /*
  2340. * Recover a dquot record
  2341. */
  2342. STATIC int
  2343. xlog_recover_do_dquot_trans(
  2344. xlog_t *log,
  2345. xlog_recover_item_t *item,
  2346. int pass)
  2347. {
  2348. xfs_mount_t *mp;
  2349. xfs_buf_t *bp;
  2350. struct xfs_disk_dquot *ddq, *recddq;
  2351. int error;
  2352. xfs_dq_logformat_t *dq_f;
  2353. uint type;
  2354. if (pass == XLOG_RECOVER_PASS1) {
  2355. return 0;
  2356. }
  2357. mp = log->l_mp;
  2358. /*
  2359. * Filesystems are required to send in quota flags at mount time.
  2360. */
  2361. if (mp->m_qflags == 0)
  2362. return (0);
  2363. recddq = (xfs_disk_dquot_t *)item->ri_buf[1].i_addr;
  2364. ASSERT(recddq);
  2365. /*
  2366. * This type of quotas was turned off, so ignore this record.
  2367. */
  2368. type = INT_GET(recddq->d_flags, ARCH_CONVERT) &
  2369. (XFS_DQ_USER | XFS_DQ_PROJ | XFS_DQ_GROUP);
  2370. ASSERT(type);
  2371. if (log->l_quotaoffs_flag & type)
  2372. return (0);
  2373. /*
  2374. * At this point we know that quota was _not_ turned off.
  2375. * Since the mount flags are not indicating to us otherwise, this
  2376. * must mean that quota is on, and the dquot needs to be replayed.
  2377. * Remember that we may not have fully recovered the superblock yet,
  2378. * so we can't do the usual trick of looking at the SB quota bits.
  2379. *
  2380. * The other possibility, of course, is that the quota subsystem was
  2381. * removed since the last mount - ENOSYS.
  2382. */
  2383. dq_f = (xfs_dq_logformat_t *)item->ri_buf[0].i_addr;
  2384. ASSERT(dq_f);
  2385. if ((error = xfs_qm_dqcheck(recddq,
  2386. dq_f->qlf_id,
  2387. 0, XFS_QMOPT_DOWARN,
  2388. "xlog_recover_do_dquot_trans (log copy)"))) {
  2389. return XFS_ERROR(EIO);
  2390. }
  2391. ASSERT(dq_f->qlf_len == 1);
  2392. error = xfs_read_buf(mp, mp->m_ddev_targp,
  2393. dq_f->qlf_blkno,
  2394. XFS_FSB_TO_BB(mp, dq_f->qlf_len),
  2395. 0, &bp);
  2396. if (error) {
  2397. xfs_ioerror_alert("xlog_recover_do..(read#3)", mp,
  2398. bp, dq_f->qlf_blkno);
  2399. return error;
  2400. }
  2401. ASSERT(bp);
  2402. ddq = (xfs_disk_dquot_t *)xfs_buf_offset(bp, dq_f->qlf_boffset);
  2403. /*
  2404. * At least the magic num portion should be on disk because this
  2405. * was among a chunk of dquots created earlier, and we did some
  2406. * minimal initialization then.
  2407. */
  2408. if (xfs_qm_dqcheck(ddq, dq_f->qlf_id, 0, XFS_QMOPT_DOWARN,
  2409. "xlog_recover_do_dquot_trans")) {
  2410. xfs_buf_relse(bp);
  2411. return XFS_ERROR(EIO);
  2412. }
  2413. memcpy(ddq, recddq, item->ri_buf[1].i_len);
  2414. ASSERT(dq_f->qlf_size == 2);
  2415. ASSERT(XFS_BUF_FSPRIVATE(bp, void *) == NULL ||
  2416. XFS_BUF_FSPRIVATE(bp, xfs_mount_t *) == mp);
  2417. XFS_BUF_SET_FSPRIVATE(bp, mp);
  2418. XFS_BUF_SET_IODONE_FUNC(bp, xlog_recover_iodone);
  2419. xfs_bdwrite(mp, bp);
  2420. return (0);
  2421. }
  2422. /*
  2423. * This routine is called to create an in-core extent free intent
  2424. * item from the efi format structure which was logged on disk.
  2425. * It allocates an in-core efi, copies the extents from the format
  2426. * structure into it, and adds the efi to the AIL with the given
  2427. * LSN.
  2428. */
  2429. STATIC int
  2430. xlog_recover_do_efi_trans(
  2431. xlog_t *log,
  2432. xlog_recover_item_t *item,
  2433. xfs_lsn_t lsn,
  2434. int pass)
  2435. {
  2436. int error;
  2437. xfs_mount_t *mp;
  2438. xfs_efi_log_item_t *efip;
  2439. xfs_efi_log_format_t *efi_formatp;
  2440. SPLDECL(s);
  2441. if (pass == XLOG_RECOVER_PASS1) {
  2442. return 0;
  2443. }
  2444. efi_formatp = (xfs_efi_log_format_t *)item->ri_buf[0].i_addr;
  2445. mp = log->l_mp;
  2446. efip = xfs_efi_init(mp, efi_formatp->efi_nextents);
  2447. if ((error = xfs_efi_copy_format(&(item->ri_buf[0]),
  2448. &(efip->efi_format)))) {
  2449. xfs_efi_item_free(efip);
  2450. return error;
  2451. }
  2452. efip->efi_next_extent = efi_formatp->efi_nextents;
  2453. efip->efi_flags |= XFS_EFI_COMMITTED;
  2454. AIL_LOCK(mp,s);
  2455. /*
  2456. * xfs_trans_update_ail() drops the AIL lock.
  2457. */
  2458. xfs_trans_update_ail(mp, (xfs_log_item_t *)efip, lsn, s);
  2459. return 0;
  2460. }
  2461. /*
  2462. * This routine is called when an efd format structure is found in
  2463. * a committed transaction in the log. It's purpose is to cancel
  2464. * the corresponding efi if it was still in the log. To do this
  2465. * it searches the AIL for the efi with an id equal to that in the
  2466. * efd format structure. If we find it, we remove the efi from the
  2467. * AIL and free it.
  2468. */
  2469. STATIC void
  2470. xlog_recover_do_efd_trans(
  2471. xlog_t *log,
  2472. xlog_recover_item_t *item,
  2473. int pass)
  2474. {
  2475. xfs_mount_t *mp;
  2476. xfs_efd_log_format_t *efd_formatp;
  2477. xfs_efi_log_item_t *efip = NULL;
  2478. xfs_log_item_t *lip;
  2479. int gen;
  2480. __uint64_t efi_id;
  2481. SPLDECL(s);
  2482. if (pass == XLOG_RECOVER_PASS1) {
  2483. return;
  2484. }
  2485. efd_formatp = (xfs_efd_log_format_t *)item->ri_buf[0].i_addr;
  2486. ASSERT((item->ri_buf[0].i_len == (sizeof(xfs_efd_log_format_32_t) +
  2487. ((efd_formatp->efd_nextents - 1) * sizeof(xfs_extent_32_t)))) ||
  2488. (item->ri_buf[0].i_len == (sizeof(xfs_efd_log_format_64_t) +
  2489. ((efd_formatp->efd_nextents - 1) * sizeof(xfs_extent_64_t)))));
  2490. efi_id = efd_formatp->efd_efi_id;
  2491. /*
  2492. * Search for the efi with the id in the efd format structure
  2493. * in the AIL.
  2494. */
  2495. mp = log->l_mp;
  2496. AIL_LOCK(mp,s);
  2497. lip = xfs_trans_first_ail(mp, &gen);
  2498. while (lip != NULL) {
  2499. if (lip->li_type == XFS_LI_EFI) {
  2500. efip = (xfs_efi_log_item_t *)lip;
  2501. if (efip->efi_format.efi_id == efi_id) {
  2502. /*
  2503. * xfs_trans_delete_ail() drops the
  2504. * AIL lock.
  2505. */
  2506. xfs_trans_delete_ail(mp, lip, s);
  2507. break;
  2508. }
  2509. }
  2510. lip = xfs_trans_next_ail(mp, lip, &gen, NULL);
  2511. }
  2512. /*
  2513. * If we found it, then free it up. If it wasn't there, it
  2514. * must have been overwritten in the log. Oh well.
  2515. */
  2516. if (lip != NULL) {
  2517. xfs_efi_item_free(efip);
  2518. } else {
  2519. AIL_UNLOCK(mp, s);
  2520. }
  2521. }
  2522. /*
  2523. * Perform the transaction
  2524. *
  2525. * If the transaction modifies a buffer or inode, do it now. Otherwise,
  2526. * EFIs and EFDs get queued up by adding entries into the AIL for them.
  2527. */
  2528. STATIC int
  2529. xlog_recover_do_trans(
  2530. xlog_t *log,
  2531. xlog_recover_t *trans,
  2532. int pass)
  2533. {
  2534. int error = 0;
  2535. xlog_recover_item_t *item, *first_item;
  2536. if ((error = xlog_recover_reorder_trans(log, trans)))
  2537. return error;
  2538. first_item = item = trans->r_itemq;
  2539. do {
  2540. /*
  2541. * we don't need to worry about the block number being
  2542. * truncated in > 1 TB buffers because in user-land,
  2543. * we're now n32 or 64-bit so xfs_daddr_t is 64-bits so
  2544. * the blknos will get through the user-mode buffer
  2545. * cache properly. The only bad case is o32 kernels
  2546. * where xfs_daddr_t is 32-bits but mount will warn us
  2547. * off a > 1 TB filesystem before we get here.
  2548. */
  2549. if ((ITEM_TYPE(item) == XFS_LI_BUF)) {
  2550. if ((error = xlog_recover_do_buffer_trans(log, item,
  2551. pass)))
  2552. break;
  2553. } else if ((ITEM_TYPE(item) == XFS_LI_INODE)) {
  2554. if ((error = xlog_recover_do_inode_trans(log, item,
  2555. pass)))
  2556. break;
  2557. } else if (ITEM_TYPE(item) == XFS_LI_EFI) {
  2558. if ((error = xlog_recover_do_efi_trans(log, item, trans->r_lsn,
  2559. pass)))
  2560. break;
  2561. } else if (ITEM_TYPE(item) == XFS_LI_EFD) {
  2562. xlog_recover_do_efd_trans(log, item, pass);
  2563. } else if (ITEM_TYPE(item) == XFS_LI_DQUOT) {
  2564. if ((error = xlog_recover_do_dquot_trans(log, item,
  2565. pass)))
  2566. break;
  2567. } else if ((ITEM_TYPE(item) == XFS_LI_QUOTAOFF)) {
  2568. if ((error = xlog_recover_do_quotaoff_trans(log, item,
  2569. pass)))
  2570. break;
  2571. } else {
  2572. xlog_warn("XFS: xlog_recover_do_trans");
  2573. ASSERT(0);
  2574. error = XFS_ERROR(EIO);
  2575. break;
  2576. }
  2577. item = item->ri_next;
  2578. } while (first_item != item);
  2579. return error;
  2580. }
  2581. /*
  2582. * Free up any resources allocated by the transaction
  2583. *
  2584. * Remember that EFIs, EFDs, and IUNLINKs are handled later.
  2585. */
  2586. STATIC void
  2587. xlog_recover_free_trans(
  2588. xlog_recover_t *trans)
  2589. {
  2590. xlog_recover_item_t *first_item, *item, *free_item;
  2591. int i;
  2592. item = first_item = trans->r_itemq;
  2593. do {
  2594. free_item = item;
  2595. item = item->ri_next;
  2596. /* Free the regions in the item. */
  2597. for (i = 0; i < free_item->ri_cnt; i++) {
  2598. kmem_free(free_item->ri_buf[i].i_addr,
  2599. free_item->ri_buf[i].i_len);
  2600. }
  2601. /* Free the item itself */
  2602. kmem_free(free_item->ri_buf,
  2603. (free_item->ri_total * sizeof(xfs_log_iovec_t)));
  2604. kmem_free(free_item, sizeof(xlog_recover_item_t));
  2605. } while (first_item != item);
  2606. /* Free the transaction recover structure */
  2607. kmem_free(trans, sizeof(xlog_recover_t));
  2608. }
  2609. STATIC int
  2610. xlog_recover_commit_trans(
  2611. xlog_t *log,
  2612. xlog_recover_t **q,
  2613. xlog_recover_t *trans,
  2614. int pass)
  2615. {
  2616. int error;
  2617. if ((error = xlog_recover_unlink_tid(q, trans)))
  2618. return error;
  2619. if ((error = xlog_recover_do_trans(log, trans, pass)))
  2620. return error;
  2621. xlog_recover_free_trans(trans); /* no error */
  2622. return 0;
  2623. }
  2624. STATIC int
  2625. xlog_recover_unmount_trans(
  2626. xlog_recover_t *trans)
  2627. {
  2628. /* Do nothing now */
  2629. xlog_warn("XFS: xlog_recover_unmount_trans: Unmount LR");
  2630. return 0;
  2631. }
  2632. /*
  2633. * There are two valid states of the r_state field. 0 indicates that the
  2634. * transaction structure is in a normal state. We have either seen the
  2635. * start of the transaction or the last operation we added was not a partial
  2636. * operation. If the last operation we added to the transaction was a
  2637. * partial operation, we need to mark r_state with XLOG_WAS_CONT_TRANS.
  2638. *
  2639. * NOTE: skip LRs with 0 data length.
  2640. */
  2641. STATIC int
  2642. xlog_recover_process_data(
  2643. xlog_t *log,
  2644. xlog_recover_t *rhash[],
  2645. xlog_rec_header_t *rhead,
  2646. xfs_caddr_t dp,
  2647. int pass)
  2648. {
  2649. xfs_caddr_t lp;
  2650. int num_logops;
  2651. xlog_op_header_t *ohead;
  2652. xlog_recover_t *trans;
  2653. xlog_tid_t tid;
  2654. int error;
  2655. unsigned long hash;
  2656. uint flags;
  2657. lp = dp + INT_GET(rhead->h_len, ARCH_CONVERT);
  2658. num_logops = INT_GET(rhead->h_num_logops, ARCH_CONVERT);
  2659. /* check the log format matches our own - else we can't recover */
  2660. if (xlog_header_check_recover(log->l_mp, rhead))
  2661. return (XFS_ERROR(EIO));
  2662. while ((dp < lp) && num_logops) {
  2663. ASSERT(dp + sizeof(xlog_op_header_t) <= lp);
  2664. ohead = (xlog_op_header_t *)dp;
  2665. dp += sizeof(xlog_op_header_t);
  2666. if (ohead->oh_clientid != XFS_TRANSACTION &&
  2667. ohead->oh_clientid != XFS_LOG) {
  2668. xlog_warn(
  2669. "XFS: xlog_recover_process_data: bad clientid");
  2670. ASSERT(0);
  2671. return (XFS_ERROR(EIO));
  2672. }
  2673. tid = INT_GET(ohead->oh_tid, ARCH_CONVERT);
  2674. hash = XLOG_RHASH(tid);
  2675. trans = xlog_recover_find_tid(rhash[hash], tid);
  2676. if (trans == NULL) { /* not found; add new tid */
  2677. if (ohead->oh_flags & XLOG_START_TRANS)
  2678. xlog_recover_new_tid(&rhash[hash], tid,
  2679. INT_GET(rhead->h_lsn, ARCH_CONVERT));
  2680. } else {
  2681. ASSERT(dp+INT_GET(ohead->oh_len, ARCH_CONVERT) <= lp);
  2682. flags = ohead->oh_flags & ~XLOG_END_TRANS;
  2683. if (flags & XLOG_WAS_CONT_TRANS)
  2684. flags &= ~XLOG_CONTINUE_TRANS;
  2685. switch (flags) {
  2686. case XLOG_COMMIT_TRANS:
  2687. error = xlog_recover_commit_trans(log,
  2688. &rhash[hash], trans, pass);
  2689. break;
  2690. case XLOG_UNMOUNT_TRANS:
  2691. error = xlog_recover_unmount_trans(trans);
  2692. break;
  2693. case XLOG_WAS_CONT_TRANS:
  2694. error = xlog_recover_add_to_cont_trans(trans,
  2695. dp, INT_GET(ohead->oh_len,
  2696. ARCH_CONVERT));
  2697. break;
  2698. case XLOG_START_TRANS:
  2699. xlog_warn(
  2700. "XFS: xlog_recover_process_data: bad transaction");
  2701. ASSERT(0);
  2702. error = XFS_ERROR(EIO);
  2703. break;
  2704. case 0:
  2705. case XLOG_CONTINUE_TRANS:
  2706. error = xlog_recover_add_to_trans(trans,
  2707. dp, INT_GET(ohead->oh_len,
  2708. ARCH_CONVERT));
  2709. break;
  2710. default:
  2711. xlog_warn(
  2712. "XFS: xlog_recover_process_data: bad flag");
  2713. ASSERT(0);
  2714. error = XFS_ERROR(EIO);
  2715. break;
  2716. }
  2717. if (error)
  2718. return error;
  2719. }
  2720. dp += INT_GET(ohead->oh_len, ARCH_CONVERT);
  2721. num_logops--;
  2722. }
  2723. return 0;
  2724. }
  2725. /*
  2726. * Process an extent free intent item that was recovered from
  2727. * the log. We need to free the extents that it describes.
  2728. */
  2729. STATIC void
  2730. xlog_recover_process_efi(
  2731. xfs_mount_t *mp,
  2732. xfs_efi_log_item_t *efip)
  2733. {
  2734. xfs_efd_log_item_t *efdp;
  2735. xfs_trans_t *tp;
  2736. int i;
  2737. xfs_extent_t *extp;
  2738. xfs_fsblock_t startblock_fsb;
  2739. ASSERT(!(efip->efi_flags & XFS_EFI_RECOVERED));
  2740. /*
  2741. * First check the validity of the extents described by the
  2742. * EFI. If any are bad, then assume that all are bad and
  2743. * just toss the EFI.
  2744. */
  2745. for (i = 0; i < efip->efi_format.efi_nextents; i++) {
  2746. extp = &(efip->efi_format.efi_extents[i]);
  2747. startblock_fsb = XFS_BB_TO_FSB(mp,
  2748. XFS_FSB_TO_DADDR(mp, extp->ext_start));
  2749. if ((startblock_fsb == 0) ||
  2750. (extp->ext_len == 0) ||
  2751. (startblock_fsb >= mp->m_sb.sb_dblocks) ||
  2752. (extp->ext_len >= mp->m_sb.sb_agblocks)) {
  2753. /*
  2754. * This will pull the EFI from the AIL and
  2755. * free the memory associated with it.
  2756. */
  2757. xfs_efi_release(efip, efip->efi_format.efi_nextents);
  2758. return;
  2759. }
  2760. }
  2761. tp = xfs_trans_alloc(mp, 0);
  2762. xfs_trans_reserve(tp, 0, XFS_ITRUNCATE_LOG_RES(mp), 0, 0, 0);
  2763. efdp = xfs_trans_get_efd(tp, efip, efip->efi_format.efi_nextents);
  2764. for (i = 0; i < efip->efi_format.efi_nextents; i++) {
  2765. extp = &(efip->efi_format.efi_extents[i]);
  2766. xfs_free_extent(tp, extp->ext_start, extp->ext_len);
  2767. xfs_trans_log_efd_extent(tp, efdp, extp->ext_start,
  2768. extp->ext_len);
  2769. }
  2770. efip->efi_flags |= XFS_EFI_RECOVERED;
  2771. xfs_trans_commit(tp, 0, NULL);
  2772. }
  2773. /*
  2774. * Verify that once we've encountered something other than an EFI
  2775. * in the AIL that there are no more EFIs in the AIL.
  2776. */
  2777. #if defined(DEBUG)
  2778. STATIC void
  2779. xlog_recover_check_ail(
  2780. xfs_mount_t *mp,
  2781. xfs_log_item_t *lip,
  2782. int gen)
  2783. {
  2784. int orig_gen = gen;
  2785. do {
  2786. ASSERT(lip->li_type != XFS_LI_EFI);
  2787. lip = xfs_trans_next_ail(mp, lip, &gen, NULL);
  2788. /*
  2789. * The check will be bogus if we restart from the
  2790. * beginning of the AIL, so ASSERT that we don't.
  2791. * We never should since we're holding the AIL lock
  2792. * the entire time.
  2793. */
  2794. ASSERT(gen == orig_gen);
  2795. } while (lip != NULL);
  2796. }
  2797. #endif /* DEBUG */
  2798. /*
  2799. * When this is called, all of the EFIs which did not have
  2800. * corresponding EFDs should be in the AIL. What we do now
  2801. * is free the extents associated with each one.
  2802. *
  2803. * Since we process the EFIs in normal transactions, they
  2804. * will be removed at some point after the commit. This prevents
  2805. * us from just walking down the list processing each one.
  2806. * We'll use a flag in the EFI to skip those that we've already
  2807. * processed and use the AIL iteration mechanism's generation
  2808. * count to try to speed this up at least a bit.
  2809. *
  2810. * When we start, we know that the EFIs are the only things in
  2811. * the AIL. As we process them, however, other items are added
  2812. * to the AIL. Since everything added to the AIL must come after
  2813. * everything already in the AIL, we stop processing as soon as
  2814. * we see something other than an EFI in the AIL.
  2815. */
  2816. STATIC void
  2817. xlog_recover_process_efis(
  2818. xlog_t *log)
  2819. {
  2820. xfs_log_item_t *lip;
  2821. xfs_efi_log_item_t *efip;
  2822. int gen;
  2823. xfs_mount_t *mp;
  2824. SPLDECL(s);
  2825. mp = log->l_mp;
  2826. AIL_LOCK(mp,s);
  2827. lip = xfs_trans_first_ail(mp, &gen);
  2828. while (lip != NULL) {
  2829. /*
  2830. * We're done when we see something other than an EFI.
  2831. */
  2832. if (lip->li_type != XFS_LI_EFI) {
  2833. xlog_recover_check_ail(mp, lip, gen);
  2834. break;
  2835. }
  2836. /*
  2837. * Skip EFIs that we've already processed.
  2838. */
  2839. efip = (xfs_efi_log_item_t *)lip;
  2840. if (efip->efi_flags & XFS_EFI_RECOVERED) {
  2841. lip = xfs_trans_next_ail(mp, lip, &gen, NULL);
  2842. continue;
  2843. }
  2844. AIL_UNLOCK(mp, s);
  2845. xlog_recover_process_efi(mp, efip);
  2846. AIL_LOCK(mp,s);
  2847. lip = xfs_trans_next_ail(mp, lip, &gen, NULL);
  2848. }
  2849. AIL_UNLOCK(mp, s);
  2850. }
  2851. /*
  2852. * This routine performs a transaction to null out a bad inode pointer
  2853. * in an agi unlinked inode hash bucket.
  2854. */
  2855. STATIC void
  2856. xlog_recover_clear_agi_bucket(
  2857. xfs_mount_t *mp,
  2858. xfs_agnumber_t agno,
  2859. int bucket)
  2860. {
  2861. xfs_trans_t *tp;
  2862. xfs_agi_t *agi;
  2863. xfs_buf_t *agibp;
  2864. int offset;
  2865. int error;
  2866. tp = xfs_trans_alloc(mp, XFS_TRANS_CLEAR_AGI_BUCKET);
  2867. xfs_trans_reserve(tp, 0, XFS_CLEAR_AGI_BUCKET_LOG_RES(mp), 0, 0, 0);
  2868. error = xfs_trans_read_buf(mp, tp, mp->m_ddev_targp,
  2869. XFS_AG_DADDR(mp, agno, XFS_AGI_DADDR(mp)),
  2870. XFS_FSS_TO_BB(mp, 1), 0, &agibp);
  2871. if (error) {
  2872. xfs_trans_cancel(tp, XFS_TRANS_ABORT);
  2873. return;
  2874. }
  2875. agi = XFS_BUF_TO_AGI(agibp);
  2876. if (be32_to_cpu(agi->agi_magicnum) != XFS_AGI_MAGIC) {
  2877. xfs_trans_cancel(tp, XFS_TRANS_ABORT);
  2878. return;
  2879. }
  2880. agi->agi_unlinked[bucket] = cpu_to_be32(NULLAGINO);
  2881. offset = offsetof(xfs_agi_t, agi_unlinked) +
  2882. (sizeof(xfs_agino_t) * bucket);
  2883. xfs_trans_log_buf(tp, agibp, offset,
  2884. (offset + sizeof(xfs_agino_t) - 1));
  2885. (void) xfs_trans_commit(tp, 0, NULL);
  2886. }
  2887. /*
  2888. * xlog_iunlink_recover
  2889. *
  2890. * This is called during recovery to process any inodes which
  2891. * we unlinked but not freed when the system crashed. These
  2892. * inodes will be on the lists in the AGI blocks. What we do
  2893. * here is scan all the AGIs and fully truncate and free any
  2894. * inodes found on the lists. Each inode is removed from the
  2895. * lists when it has been fully truncated and is freed. The
  2896. * freeing of the inode and its removal from the list must be
  2897. * atomic.
  2898. */
  2899. void
  2900. xlog_recover_process_iunlinks(
  2901. xlog_t *log)
  2902. {
  2903. xfs_mount_t *mp;
  2904. xfs_agnumber_t agno;
  2905. xfs_agi_t *agi;
  2906. xfs_buf_t *agibp;
  2907. xfs_buf_t *ibp;
  2908. xfs_dinode_t *dip;
  2909. xfs_inode_t *ip;
  2910. xfs_agino_t agino;
  2911. xfs_ino_t ino;
  2912. int bucket;
  2913. int error;
  2914. uint mp_dmevmask;
  2915. mp = log->l_mp;
  2916. /*
  2917. * Prevent any DMAPI event from being sent while in this function.
  2918. */
  2919. mp_dmevmask = mp->m_dmevmask;
  2920. mp->m_dmevmask = 0;
  2921. for (agno = 0; agno < mp->m_sb.sb_agcount; agno++) {
  2922. /*
  2923. * Find the agi for this ag.
  2924. */
  2925. agibp = xfs_buf_read(mp->m_ddev_targp,
  2926. XFS_AG_DADDR(mp, agno, XFS_AGI_DADDR(mp)),
  2927. XFS_FSS_TO_BB(mp, 1), 0);
  2928. if (XFS_BUF_ISERROR(agibp)) {
  2929. xfs_ioerror_alert("xlog_recover_process_iunlinks(#1)",
  2930. log->l_mp, agibp,
  2931. XFS_AG_DADDR(mp, agno, XFS_AGI_DADDR(mp)));
  2932. }
  2933. agi = XFS_BUF_TO_AGI(agibp);
  2934. ASSERT(XFS_AGI_MAGIC == be32_to_cpu(agi->agi_magicnum));
  2935. for (bucket = 0; bucket < XFS_AGI_UNLINKED_BUCKETS; bucket++) {
  2936. agino = be32_to_cpu(agi->agi_unlinked[bucket]);
  2937. while (agino != NULLAGINO) {
  2938. /*
  2939. * Release the agi buffer so that it can
  2940. * be acquired in the normal course of the
  2941. * transaction to truncate and free the inode.
  2942. */
  2943. xfs_buf_relse(agibp);
  2944. ino = XFS_AGINO_TO_INO(mp, agno, agino);
  2945. error = xfs_iget(mp, NULL, ino, 0, 0, &ip, 0);
  2946. ASSERT(error || (ip != NULL));
  2947. if (!error) {
  2948. /*
  2949. * Get the on disk inode to find the
  2950. * next inode in the bucket.
  2951. */
  2952. error = xfs_itobp(mp, NULL, ip, &dip,
  2953. &ibp, 0, 0);
  2954. ASSERT(error || (dip != NULL));
  2955. }
  2956. if (!error) {
  2957. ASSERT(ip->i_d.di_nlink == 0);
  2958. /* setup for the next pass */
  2959. agino = INT_GET(dip->di_next_unlinked,
  2960. ARCH_CONVERT);
  2961. xfs_buf_relse(ibp);
  2962. /*
  2963. * Prevent any DMAPI event from
  2964. * being sent when the
  2965. * reference on the inode is
  2966. * dropped.
  2967. */
  2968. ip->i_d.di_dmevmask = 0;
  2969. /*
  2970. * If this is a new inode, handle
  2971. * it specially. Otherwise,
  2972. * just drop our reference to the
  2973. * inode. If there are no
  2974. * other references, this will
  2975. * send the inode to
  2976. * xfs_inactive() which will
  2977. * truncate the file and free
  2978. * the inode.
  2979. */
  2980. if (ip->i_d.di_mode == 0)
  2981. xfs_iput_new(ip, 0);
  2982. else
  2983. VN_RELE(XFS_ITOV(ip));
  2984. } else {
  2985. /*
  2986. * We can't read in the inode
  2987. * this bucket points to, or
  2988. * this inode is messed up. Just
  2989. * ditch this bucket of inodes. We
  2990. * will lose some inodes and space,
  2991. * but at least we won't hang. Call
  2992. * xlog_recover_clear_agi_bucket()
  2993. * to perform a transaction to clear
  2994. * the inode pointer in the bucket.
  2995. */
  2996. xlog_recover_clear_agi_bucket(mp, agno,
  2997. bucket);
  2998. agino = NULLAGINO;
  2999. }
  3000. /*
  3001. * Reacquire the agibuffer and continue around
  3002. * the loop.
  3003. */
  3004. agibp = xfs_buf_read(mp->m_ddev_targp,
  3005. XFS_AG_DADDR(mp, agno,
  3006. XFS_AGI_DADDR(mp)),
  3007. XFS_FSS_TO_BB(mp, 1), 0);
  3008. if (XFS_BUF_ISERROR(agibp)) {
  3009. xfs_ioerror_alert(
  3010. "xlog_recover_process_iunlinks(#2)",
  3011. log->l_mp, agibp,
  3012. XFS_AG_DADDR(mp, agno,
  3013. XFS_AGI_DADDR(mp)));
  3014. }
  3015. agi = XFS_BUF_TO_AGI(agibp);
  3016. ASSERT(XFS_AGI_MAGIC == be32_to_cpu(
  3017. agi->agi_magicnum));
  3018. }
  3019. }
  3020. /*
  3021. * Release the buffer for the current agi so we can
  3022. * go on to the next one.
  3023. */
  3024. xfs_buf_relse(agibp);
  3025. }
  3026. mp->m_dmevmask = mp_dmevmask;
  3027. }
  3028. #ifdef DEBUG
  3029. STATIC void
  3030. xlog_pack_data_checksum(
  3031. xlog_t *log,
  3032. xlog_in_core_t *iclog,
  3033. int size)
  3034. {
  3035. int i;
  3036. uint *up;
  3037. uint chksum = 0;
  3038. up = (uint *)iclog->ic_datap;
  3039. /* divide length by 4 to get # words */
  3040. for (i = 0; i < (size >> 2); i++) {
  3041. chksum ^= INT_GET(*up, ARCH_CONVERT);
  3042. up++;
  3043. }
  3044. INT_SET(iclog->ic_header.h_chksum, ARCH_CONVERT, chksum);
  3045. }
  3046. #else
  3047. #define xlog_pack_data_checksum(log, iclog, size)
  3048. #endif
  3049. /*
  3050. * Stamp cycle number in every block
  3051. */
  3052. void
  3053. xlog_pack_data(
  3054. xlog_t *log,
  3055. xlog_in_core_t *iclog,
  3056. int roundoff)
  3057. {
  3058. int i, j, k;
  3059. int size = iclog->ic_offset + roundoff;
  3060. uint cycle_lsn;
  3061. xfs_caddr_t dp;
  3062. xlog_in_core_2_t *xhdr;
  3063. xlog_pack_data_checksum(log, iclog, size);
  3064. cycle_lsn = CYCLE_LSN_DISK(iclog->ic_header.h_lsn);
  3065. dp = iclog->ic_datap;
  3066. for (i = 0; i < BTOBB(size) &&
  3067. i < (XLOG_HEADER_CYCLE_SIZE / BBSIZE); i++) {
  3068. iclog->ic_header.h_cycle_data[i] = *(uint *)dp;
  3069. *(uint *)dp = cycle_lsn;
  3070. dp += BBSIZE;
  3071. }
  3072. if (XFS_SB_VERSION_HASLOGV2(&log->l_mp->m_sb)) {
  3073. xhdr = (xlog_in_core_2_t *)&iclog->ic_header;
  3074. for ( ; i < BTOBB(size); i++) {
  3075. j = i / (XLOG_HEADER_CYCLE_SIZE / BBSIZE);
  3076. k = i % (XLOG_HEADER_CYCLE_SIZE / BBSIZE);
  3077. xhdr[j].hic_xheader.xh_cycle_data[k] = *(uint *)dp;
  3078. *(uint *)dp = cycle_lsn;
  3079. dp += BBSIZE;
  3080. }
  3081. for (i = 1; i < log->l_iclog_heads; i++) {
  3082. xhdr[i].hic_xheader.xh_cycle = cycle_lsn;
  3083. }
  3084. }
  3085. }
  3086. #if defined(DEBUG) && defined(XFS_LOUD_RECOVERY)
  3087. STATIC void
  3088. xlog_unpack_data_checksum(
  3089. xlog_rec_header_t *rhead,
  3090. xfs_caddr_t dp,
  3091. xlog_t *log)
  3092. {
  3093. uint *up = (uint *)dp;
  3094. uint chksum = 0;
  3095. int i;
  3096. /* divide length by 4 to get # words */
  3097. for (i=0; i < INT_GET(rhead->h_len, ARCH_CONVERT) >> 2; i++) {
  3098. chksum ^= INT_GET(*up, ARCH_CONVERT);
  3099. up++;
  3100. }
  3101. if (chksum != INT_GET(rhead->h_chksum, ARCH_CONVERT)) {
  3102. if (rhead->h_chksum ||
  3103. ((log->l_flags & XLOG_CHKSUM_MISMATCH) == 0)) {
  3104. cmn_err(CE_DEBUG,
  3105. "XFS: LogR chksum mismatch: was (0x%x) is (0x%x)\n",
  3106. INT_GET(rhead->h_chksum, ARCH_CONVERT), chksum);
  3107. cmn_err(CE_DEBUG,
  3108. "XFS: Disregard message if filesystem was created with non-DEBUG kernel");
  3109. if (XFS_SB_VERSION_HASLOGV2(&log->l_mp->m_sb)) {
  3110. cmn_err(CE_DEBUG,
  3111. "XFS: LogR this is a LogV2 filesystem\n");
  3112. }
  3113. log->l_flags |= XLOG_CHKSUM_MISMATCH;
  3114. }
  3115. }
  3116. }
  3117. #else
  3118. #define xlog_unpack_data_checksum(rhead, dp, log)
  3119. #endif
  3120. STATIC void
  3121. xlog_unpack_data(
  3122. xlog_rec_header_t *rhead,
  3123. xfs_caddr_t dp,
  3124. xlog_t *log)
  3125. {
  3126. int i, j, k;
  3127. xlog_in_core_2_t *xhdr;
  3128. for (i = 0; i < BTOBB(INT_GET(rhead->h_len, ARCH_CONVERT)) &&
  3129. i < (XLOG_HEADER_CYCLE_SIZE / BBSIZE); i++) {
  3130. *(uint *)dp = *(uint *)&rhead->h_cycle_data[i];
  3131. dp += BBSIZE;
  3132. }
  3133. if (XFS_SB_VERSION_HASLOGV2(&log->l_mp->m_sb)) {
  3134. xhdr = (xlog_in_core_2_t *)rhead;
  3135. for ( ; i < BTOBB(INT_GET(rhead->h_len, ARCH_CONVERT)); i++) {
  3136. j = i / (XLOG_HEADER_CYCLE_SIZE / BBSIZE);
  3137. k = i % (XLOG_HEADER_CYCLE_SIZE / BBSIZE);
  3138. *(uint *)dp = xhdr[j].hic_xheader.xh_cycle_data[k];
  3139. dp += BBSIZE;
  3140. }
  3141. }
  3142. xlog_unpack_data_checksum(rhead, dp, log);
  3143. }
  3144. STATIC int
  3145. xlog_valid_rec_header(
  3146. xlog_t *log,
  3147. xlog_rec_header_t *rhead,
  3148. xfs_daddr_t blkno)
  3149. {
  3150. int hlen;
  3151. if (unlikely(
  3152. (INT_GET(rhead->h_magicno, ARCH_CONVERT) !=
  3153. XLOG_HEADER_MAGIC_NUM))) {
  3154. XFS_ERROR_REPORT("xlog_valid_rec_header(1)",
  3155. XFS_ERRLEVEL_LOW, log->l_mp);
  3156. return XFS_ERROR(EFSCORRUPTED);
  3157. }
  3158. if (unlikely(
  3159. (!rhead->h_version ||
  3160. (INT_GET(rhead->h_version, ARCH_CONVERT) &
  3161. (~XLOG_VERSION_OKBITS)) != 0))) {
  3162. xlog_warn("XFS: %s: unrecognised log version (%d).",
  3163. __FUNCTION__, INT_GET(rhead->h_version, ARCH_CONVERT));
  3164. return XFS_ERROR(EIO);
  3165. }
  3166. /* LR body must have data or it wouldn't have been written */
  3167. hlen = INT_GET(rhead->h_len, ARCH_CONVERT);
  3168. if (unlikely( hlen <= 0 || hlen > INT_MAX )) {
  3169. XFS_ERROR_REPORT("xlog_valid_rec_header(2)",
  3170. XFS_ERRLEVEL_LOW, log->l_mp);
  3171. return XFS_ERROR(EFSCORRUPTED);
  3172. }
  3173. if (unlikely( blkno > log->l_logBBsize || blkno > INT_MAX )) {
  3174. XFS_ERROR_REPORT("xlog_valid_rec_header(3)",
  3175. XFS_ERRLEVEL_LOW, log->l_mp);
  3176. return XFS_ERROR(EFSCORRUPTED);
  3177. }
  3178. return 0;
  3179. }
  3180. /*
  3181. * Read the log from tail to head and process the log records found.
  3182. * Handle the two cases where the tail and head are in the same cycle
  3183. * and where the active portion of the log wraps around the end of
  3184. * the physical log separately. The pass parameter is passed through
  3185. * to the routines called to process the data and is not looked at
  3186. * here.
  3187. */
  3188. STATIC int
  3189. xlog_do_recovery_pass(
  3190. xlog_t *log,
  3191. xfs_daddr_t head_blk,
  3192. xfs_daddr_t tail_blk,
  3193. int pass)
  3194. {
  3195. xlog_rec_header_t *rhead;
  3196. xfs_daddr_t blk_no;
  3197. xfs_caddr_t bufaddr, offset;
  3198. xfs_buf_t *hbp, *dbp;
  3199. int error = 0, h_size;
  3200. int bblks, split_bblks;
  3201. int hblks, split_hblks, wrapped_hblks;
  3202. xlog_recover_t *rhash[XLOG_RHASH_SIZE];
  3203. ASSERT(head_blk != tail_blk);
  3204. /*
  3205. * Read the header of the tail block and get the iclog buffer size from
  3206. * h_size. Use this to tell how many sectors make up the log header.
  3207. */
  3208. if (XFS_SB_VERSION_HASLOGV2(&log->l_mp->m_sb)) {
  3209. /*
  3210. * When using variable length iclogs, read first sector of
  3211. * iclog header and extract the header size from it. Get a
  3212. * new hbp that is the correct size.
  3213. */
  3214. hbp = xlog_get_bp(log, 1);
  3215. if (!hbp)
  3216. return ENOMEM;
  3217. if ((error = xlog_bread(log, tail_blk, 1, hbp)))
  3218. goto bread_err1;
  3219. offset = xlog_align(log, tail_blk, 1, hbp);
  3220. rhead = (xlog_rec_header_t *)offset;
  3221. error = xlog_valid_rec_header(log, rhead, tail_blk);
  3222. if (error)
  3223. goto bread_err1;
  3224. h_size = INT_GET(rhead->h_size, ARCH_CONVERT);
  3225. if ((INT_GET(rhead->h_version, ARCH_CONVERT)
  3226. & XLOG_VERSION_2) &&
  3227. (h_size > XLOG_HEADER_CYCLE_SIZE)) {
  3228. hblks = h_size / XLOG_HEADER_CYCLE_SIZE;
  3229. if (h_size % XLOG_HEADER_CYCLE_SIZE)
  3230. hblks++;
  3231. xlog_put_bp(hbp);
  3232. hbp = xlog_get_bp(log, hblks);
  3233. } else {
  3234. hblks = 1;
  3235. }
  3236. } else {
  3237. ASSERT(log->l_sectbb_log == 0);
  3238. hblks = 1;
  3239. hbp = xlog_get_bp(log, 1);
  3240. h_size = XLOG_BIG_RECORD_BSIZE;
  3241. }
  3242. if (!hbp)
  3243. return ENOMEM;
  3244. dbp = xlog_get_bp(log, BTOBB(h_size));
  3245. if (!dbp) {
  3246. xlog_put_bp(hbp);
  3247. return ENOMEM;
  3248. }
  3249. memset(rhash, 0, sizeof(rhash));
  3250. if (tail_blk <= head_blk) {
  3251. for (blk_no = tail_blk; blk_no < head_blk; ) {
  3252. if ((error = xlog_bread(log, blk_no, hblks, hbp)))
  3253. goto bread_err2;
  3254. offset = xlog_align(log, blk_no, hblks, hbp);
  3255. rhead = (xlog_rec_header_t *)offset;
  3256. error = xlog_valid_rec_header(log, rhead, blk_no);
  3257. if (error)
  3258. goto bread_err2;
  3259. /* blocks in data section */
  3260. bblks = (int)BTOBB(INT_GET(rhead->h_len, ARCH_CONVERT));
  3261. error = xlog_bread(log, blk_no + hblks, bblks, dbp);
  3262. if (error)
  3263. goto bread_err2;
  3264. offset = xlog_align(log, blk_no + hblks, bblks, dbp);
  3265. xlog_unpack_data(rhead, offset, log);
  3266. if ((error = xlog_recover_process_data(log,
  3267. rhash, rhead, offset, pass)))
  3268. goto bread_err2;
  3269. blk_no += bblks + hblks;
  3270. }
  3271. } else {
  3272. /*
  3273. * Perform recovery around the end of the physical log.
  3274. * When the head is not on the same cycle number as the tail,
  3275. * we can't do a sequential recovery as above.
  3276. */
  3277. blk_no = tail_blk;
  3278. while (blk_no < log->l_logBBsize) {
  3279. /*
  3280. * Check for header wrapping around physical end-of-log
  3281. */
  3282. offset = NULL;
  3283. split_hblks = 0;
  3284. wrapped_hblks = 0;
  3285. if (blk_no + hblks <= log->l_logBBsize) {
  3286. /* Read header in one read */
  3287. error = xlog_bread(log, blk_no, hblks, hbp);
  3288. if (error)
  3289. goto bread_err2;
  3290. offset = xlog_align(log, blk_no, hblks, hbp);
  3291. } else {
  3292. /* This LR is split across physical log end */
  3293. if (blk_no != log->l_logBBsize) {
  3294. /* some data before physical log end */
  3295. ASSERT(blk_no <= INT_MAX);
  3296. split_hblks = log->l_logBBsize - (int)blk_no;
  3297. ASSERT(split_hblks > 0);
  3298. if ((error = xlog_bread(log, blk_no,
  3299. split_hblks, hbp)))
  3300. goto bread_err2;
  3301. offset = xlog_align(log, blk_no,
  3302. split_hblks, hbp);
  3303. }
  3304. /*
  3305. * Note: this black magic still works with
  3306. * large sector sizes (non-512) only because:
  3307. * - we increased the buffer size originally
  3308. * by 1 sector giving us enough extra space
  3309. * for the second read;
  3310. * - the log start is guaranteed to be sector
  3311. * aligned;
  3312. * - we read the log end (LR header start)
  3313. * _first_, then the log start (LR header end)
  3314. * - order is important.
  3315. */
  3316. bufaddr = XFS_BUF_PTR(hbp);
  3317. XFS_BUF_SET_PTR(hbp,
  3318. bufaddr + BBTOB(split_hblks),
  3319. BBTOB(hblks - split_hblks));
  3320. wrapped_hblks = hblks - split_hblks;
  3321. error = xlog_bread(log, 0, wrapped_hblks, hbp);
  3322. if (error)
  3323. goto bread_err2;
  3324. XFS_BUF_SET_PTR(hbp, bufaddr, BBTOB(hblks));
  3325. if (!offset)
  3326. offset = xlog_align(log, 0,
  3327. wrapped_hblks, hbp);
  3328. }
  3329. rhead = (xlog_rec_header_t *)offset;
  3330. error = xlog_valid_rec_header(log, rhead,
  3331. split_hblks ? blk_no : 0);
  3332. if (error)
  3333. goto bread_err2;
  3334. bblks = (int)BTOBB(INT_GET(rhead->h_len, ARCH_CONVERT));
  3335. blk_no += hblks;
  3336. /* Read in data for log record */
  3337. if (blk_no + bblks <= log->l_logBBsize) {
  3338. error = xlog_bread(log, blk_no, bblks, dbp);
  3339. if (error)
  3340. goto bread_err2;
  3341. offset = xlog_align(log, blk_no, bblks, dbp);
  3342. } else {
  3343. /* This log record is split across the
  3344. * physical end of log */
  3345. offset = NULL;
  3346. split_bblks = 0;
  3347. if (blk_no != log->l_logBBsize) {
  3348. /* some data is before the physical
  3349. * end of log */
  3350. ASSERT(!wrapped_hblks);
  3351. ASSERT(blk_no <= INT_MAX);
  3352. split_bblks =
  3353. log->l_logBBsize - (int)blk_no;
  3354. ASSERT(split_bblks > 0);
  3355. if ((error = xlog_bread(log, blk_no,
  3356. split_bblks, dbp)))
  3357. goto bread_err2;
  3358. offset = xlog_align(log, blk_no,
  3359. split_bblks, dbp);
  3360. }
  3361. /*
  3362. * Note: this black magic still works with
  3363. * large sector sizes (non-512) only because:
  3364. * - we increased the buffer size originally
  3365. * by 1 sector giving us enough extra space
  3366. * for the second read;
  3367. * - the log start is guaranteed to be sector
  3368. * aligned;
  3369. * - we read the log end (LR header start)
  3370. * _first_, then the log start (LR header end)
  3371. * - order is important.
  3372. */
  3373. bufaddr = XFS_BUF_PTR(dbp);
  3374. XFS_BUF_SET_PTR(dbp,
  3375. bufaddr + BBTOB(split_bblks),
  3376. BBTOB(bblks - split_bblks));
  3377. if ((error = xlog_bread(log, wrapped_hblks,
  3378. bblks - split_bblks, dbp)))
  3379. goto bread_err2;
  3380. XFS_BUF_SET_PTR(dbp, bufaddr, h_size);
  3381. if (!offset)
  3382. offset = xlog_align(log, wrapped_hblks,
  3383. bblks - split_bblks, dbp);
  3384. }
  3385. xlog_unpack_data(rhead, offset, log);
  3386. if ((error = xlog_recover_process_data(log, rhash,
  3387. rhead, offset, pass)))
  3388. goto bread_err2;
  3389. blk_no += bblks;
  3390. }
  3391. ASSERT(blk_no >= log->l_logBBsize);
  3392. blk_no -= log->l_logBBsize;
  3393. /* read first part of physical log */
  3394. while (blk_no < head_blk) {
  3395. if ((error = xlog_bread(log, blk_no, hblks, hbp)))
  3396. goto bread_err2;
  3397. offset = xlog_align(log, blk_no, hblks, hbp);
  3398. rhead = (xlog_rec_header_t *)offset;
  3399. error = xlog_valid_rec_header(log, rhead, blk_no);
  3400. if (error)
  3401. goto bread_err2;
  3402. bblks = (int)BTOBB(INT_GET(rhead->h_len, ARCH_CONVERT));
  3403. if ((error = xlog_bread(log, blk_no+hblks, bblks, dbp)))
  3404. goto bread_err2;
  3405. offset = xlog_align(log, blk_no+hblks, bblks, dbp);
  3406. xlog_unpack_data(rhead, offset, log);
  3407. if ((error = xlog_recover_process_data(log, rhash,
  3408. rhead, offset, pass)))
  3409. goto bread_err2;
  3410. blk_no += bblks + hblks;
  3411. }
  3412. }
  3413. bread_err2:
  3414. xlog_put_bp(dbp);
  3415. bread_err1:
  3416. xlog_put_bp(hbp);
  3417. return error;
  3418. }
  3419. /*
  3420. * Do the recovery of the log. We actually do this in two phases.
  3421. * The two passes are necessary in order to implement the function
  3422. * of cancelling a record written into the log. The first pass
  3423. * determines those things which have been cancelled, and the
  3424. * second pass replays log items normally except for those which
  3425. * have been cancelled. The handling of the replay and cancellations
  3426. * takes place in the log item type specific routines.
  3427. *
  3428. * The table of items which have cancel records in the log is allocated
  3429. * and freed at this level, since only here do we know when all of
  3430. * the log recovery has been completed.
  3431. */
  3432. STATIC int
  3433. xlog_do_log_recovery(
  3434. xlog_t *log,
  3435. xfs_daddr_t head_blk,
  3436. xfs_daddr_t tail_blk)
  3437. {
  3438. int error;
  3439. ASSERT(head_blk != tail_blk);
  3440. /*
  3441. * First do a pass to find all of the cancelled buf log items.
  3442. * Store them in the buf_cancel_table for use in the second pass.
  3443. */
  3444. log->l_buf_cancel_table =
  3445. (xfs_buf_cancel_t **)kmem_zalloc(XLOG_BC_TABLE_SIZE *
  3446. sizeof(xfs_buf_cancel_t*),
  3447. KM_SLEEP);
  3448. error = xlog_do_recovery_pass(log, head_blk, tail_blk,
  3449. XLOG_RECOVER_PASS1);
  3450. if (error != 0) {
  3451. kmem_free(log->l_buf_cancel_table,
  3452. XLOG_BC_TABLE_SIZE * sizeof(xfs_buf_cancel_t*));
  3453. log->l_buf_cancel_table = NULL;
  3454. return error;
  3455. }
  3456. /*
  3457. * Then do a second pass to actually recover the items in the log.
  3458. * When it is complete free the table of buf cancel items.
  3459. */
  3460. error = xlog_do_recovery_pass(log, head_blk, tail_blk,
  3461. XLOG_RECOVER_PASS2);
  3462. #ifdef DEBUG
  3463. if (!error) {
  3464. int i;
  3465. for (i = 0; i < XLOG_BC_TABLE_SIZE; i++)
  3466. ASSERT(log->l_buf_cancel_table[i] == NULL);
  3467. }
  3468. #endif /* DEBUG */
  3469. kmem_free(log->l_buf_cancel_table,
  3470. XLOG_BC_TABLE_SIZE * sizeof(xfs_buf_cancel_t*));
  3471. log->l_buf_cancel_table = NULL;
  3472. return error;
  3473. }
  3474. /*
  3475. * Do the actual recovery
  3476. */
  3477. STATIC int
  3478. xlog_do_recover(
  3479. xlog_t *log,
  3480. xfs_daddr_t head_blk,
  3481. xfs_daddr_t tail_blk)
  3482. {
  3483. int error;
  3484. xfs_buf_t *bp;
  3485. xfs_sb_t *sbp;
  3486. /*
  3487. * First replay the images in the log.
  3488. */
  3489. error = xlog_do_log_recovery(log, head_blk, tail_blk);
  3490. if (error) {
  3491. return error;
  3492. }
  3493. XFS_bflush(log->l_mp->m_ddev_targp);
  3494. /*
  3495. * If IO errors happened during recovery, bail out.
  3496. */
  3497. if (XFS_FORCED_SHUTDOWN(log->l_mp)) {
  3498. return (EIO);
  3499. }
  3500. /*
  3501. * We now update the tail_lsn since much of the recovery has completed
  3502. * and there may be space available to use. If there were no extent
  3503. * or iunlinks, we can free up the entire log and set the tail_lsn to
  3504. * be the last_sync_lsn. This was set in xlog_find_tail to be the
  3505. * lsn of the last known good LR on disk. If there are extent frees
  3506. * or iunlinks they will have some entries in the AIL; so we look at
  3507. * the AIL to determine how to set the tail_lsn.
  3508. */
  3509. xlog_assign_tail_lsn(log->l_mp);
  3510. /*
  3511. * Now that we've finished replaying all buffer and inode
  3512. * updates, re-read in the superblock.
  3513. */
  3514. bp = xfs_getsb(log->l_mp, 0);
  3515. XFS_BUF_UNDONE(bp);
  3516. XFS_BUF_READ(bp);
  3517. xfsbdstrat(log->l_mp, bp);
  3518. if ((error = xfs_iowait(bp))) {
  3519. xfs_ioerror_alert("xlog_do_recover",
  3520. log->l_mp, bp, XFS_BUF_ADDR(bp));
  3521. ASSERT(0);
  3522. xfs_buf_relse(bp);
  3523. return error;
  3524. }
  3525. /* Convert superblock from on-disk format */
  3526. sbp = &log->l_mp->m_sb;
  3527. xfs_xlatesb(XFS_BUF_TO_SBP(bp), sbp, 1, XFS_SB_ALL_BITS);
  3528. ASSERT(sbp->sb_magicnum == XFS_SB_MAGIC);
  3529. ASSERT(XFS_SB_GOOD_VERSION(sbp));
  3530. xfs_buf_relse(bp);
  3531. /* We've re-read the superblock so re-initialize per-cpu counters */
  3532. xfs_icsb_reinit_counters(log->l_mp);
  3533. xlog_recover_check_summary(log);
  3534. /* Normal transactions can now occur */
  3535. log->l_flags &= ~XLOG_ACTIVE_RECOVERY;
  3536. return 0;
  3537. }
  3538. /*
  3539. * Perform recovery and re-initialize some log variables in xlog_find_tail.
  3540. *
  3541. * Return error or zero.
  3542. */
  3543. int
  3544. xlog_recover(
  3545. xlog_t *log)
  3546. {
  3547. xfs_daddr_t head_blk, tail_blk;
  3548. int error;
  3549. /* find the tail of the log */
  3550. if ((error = xlog_find_tail(log, &head_blk, &tail_blk)))
  3551. return error;
  3552. if (tail_blk != head_blk) {
  3553. /* There used to be a comment here:
  3554. *
  3555. * disallow recovery on read-only mounts. note -- mount
  3556. * checks for ENOSPC and turns it into an intelligent
  3557. * error message.
  3558. * ...but this is no longer true. Now, unless you specify
  3559. * NORECOVERY (in which case this function would never be
  3560. * called), we just go ahead and recover. We do this all
  3561. * under the vfs layer, so we can get away with it unless
  3562. * the device itself is read-only, in which case we fail.
  3563. */
  3564. if ((error = xfs_dev_is_read_only(log->l_mp,
  3565. "recovery required"))) {
  3566. return error;
  3567. }
  3568. cmn_err(CE_NOTE,
  3569. "Starting XFS recovery on filesystem: %s (logdev: %s)",
  3570. log->l_mp->m_fsname, log->l_mp->m_logname ?
  3571. log->l_mp->m_logname : "internal");
  3572. error = xlog_do_recover(log, head_blk, tail_blk);
  3573. log->l_flags |= XLOG_RECOVERY_NEEDED;
  3574. }
  3575. return error;
  3576. }
  3577. /*
  3578. * In the first part of recovery we replay inodes and buffers and build
  3579. * up the list of extent free items which need to be processed. Here
  3580. * we process the extent free items and clean up the on disk unlinked
  3581. * inode lists. This is separated from the first part of recovery so
  3582. * that the root and real-time bitmap inodes can be read in from disk in
  3583. * between the two stages. This is necessary so that we can free space
  3584. * in the real-time portion of the file system.
  3585. */
  3586. int
  3587. xlog_recover_finish(
  3588. xlog_t *log,
  3589. int mfsi_flags)
  3590. {
  3591. /*
  3592. * Now we're ready to do the transactions needed for the
  3593. * rest of recovery. Start with completing all the extent
  3594. * free intent records and then process the unlinked inode
  3595. * lists. At this point, we essentially run in normal mode
  3596. * except that we're still performing recovery actions
  3597. * rather than accepting new requests.
  3598. */
  3599. if (log->l_flags & XLOG_RECOVERY_NEEDED) {
  3600. xlog_recover_process_efis(log);
  3601. /*
  3602. * Sync the log to get all the EFIs out of the AIL.
  3603. * This isn't absolutely necessary, but it helps in
  3604. * case the unlink transactions would have problems
  3605. * pushing the EFIs out of the way.
  3606. */
  3607. xfs_log_force(log->l_mp, (xfs_lsn_t)0,
  3608. (XFS_LOG_FORCE | XFS_LOG_SYNC));
  3609. if ( (mfsi_flags & XFS_MFSI_NOUNLINK) == 0 ) {
  3610. xlog_recover_process_iunlinks(log);
  3611. }
  3612. xlog_recover_check_summary(log);
  3613. cmn_err(CE_NOTE,
  3614. "Ending XFS recovery on filesystem: %s (logdev: %s)",
  3615. log->l_mp->m_fsname, log->l_mp->m_logname ?
  3616. log->l_mp->m_logname : "internal");
  3617. log->l_flags &= ~XLOG_RECOVERY_NEEDED;
  3618. } else {
  3619. cmn_err(CE_DEBUG,
  3620. "!Ending clean XFS mount for filesystem: %s\n",
  3621. log->l_mp->m_fsname);
  3622. }
  3623. return 0;
  3624. }
  3625. #if defined(DEBUG)
  3626. /*
  3627. * Read all of the agf and agi counters and check that they
  3628. * are consistent with the superblock counters.
  3629. */
  3630. void
  3631. xlog_recover_check_summary(
  3632. xlog_t *log)
  3633. {
  3634. xfs_mount_t *mp;
  3635. xfs_agf_t *agfp;
  3636. xfs_agi_t *agip;
  3637. xfs_buf_t *agfbp;
  3638. xfs_buf_t *agibp;
  3639. xfs_daddr_t agfdaddr;
  3640. xfs_daddr_t agidaddr;
  3641. xfs_buf_t *sbbp;
  3642. #ifdef XFS_LOUD_RECOVERY
  3643. xfs_sb_t *sbp;
  3644. #endif
  3645. xfs_agnumber_t agno;
  3646. __uint64_t freeblks;
  3647. __uint64_t itotal;
  3648. __uint64_t ifree;
  3649. mp = log->l_mp;
  3650. freeblks = 0LL;
  3651. itotal = 0LL;
  3652. ifree = 0LL;
  3653. for (agno = 0; agno < mp->m_sb.sb_agcount; agno++) {
  3654. agfdaddr = XFS_AG_DADDR(mp, agno, XFS_AGF_DADDR(mp));
  3655. agfbp = xfs_buf_read(mp->m_ddev_targp, agfdaddr,
  3656. XFS_FSS_TO_BB(mp, 1), 0);
  3657. if (XFS_BUF_ISERROR(agfbp)) {
  3658. xfs_ioerror_alert("xlog_recover_check_summary(agf)",
  3659. mp, agfbp, agfdaddr);
  3660. }
  3661. agfp = XFS_BUF_TO_AGF(agfbp);
  3662. ASSERT(XFS_AGF_MAGIC == be32_to_cpu(agfp->agf_magicnum));
  3663. ASSERT(XFS_AGF_GOOD_VERSION(be32_to_cpu(agfp->agf_versionnum)));
  3664. ASSERT(be32_to_cpu(agfp->agf_seqno) == agno);
  3665. freeblks += be32_to_cpu(agfp->agf_freeblks) +
  3666. be32_to_cpu(agfp->agf_flcount);
  3667. xfs_buf_relse(agfbp);
  3668. agidaddr = XFS_AG_DADDR(mp, agno, XFS_AGI_DADDR(mp));
  3669. agibp = xfs_buf_read(mp->m_ddev_targp, agidaddr,
  3670. XFS_FSS_TO_BB(mp, 1), 0);
  3671. if (XFS_BUF_ISERROR(agibp)) {
  3672. xfs_ioerror_alert("xlog_recover_check_summary(agi)",
  3673. mp, agibp, agidaddr);
  3674. }
  3675. agip = XFS_BUF_TO_AGI(agibp);
  3676. ASSERT(XFS_AGI_MAGIC == be32_to_cpu(agip->agi_magicnum));
  3677. ASSERT(XFS_AGI_GOOD_VERSION(be32_to_cpu(agip->agi_versionnum)));
  3678. ASSERT(be32_to_cpu(agip->agi_seqno) == agno);
  3679. itotal += be32_to_cpu(agip->agi_count);
  3680. ifree += be32_to_cpu(agip->agi_freecount);
  3681. xfs_buf_relse(agibp);
  3682. }
  3683. sbbp = xfs_getsb(mp, 0);
  3684. #ifdef XFS_LOUD_RECOVERY
  3685. sbp = &mp->m_sb;
  3686. xfs_xlatesb(XFS_BUF_TO_SBP(sbbp), sbp, 1, XFS_SB_ALL_BITS);
  3687. cmn_err(CE_NOTE,
  3688. "xlog_recover_check_summary: sb_icount %Lu itotal %Lu",
  3689. sbp->sb_icount, itotal);
  3690. cmn_err(CE_NOTE,
  3691. "xlog_recover_check_summary: sb_ifree %Lu itotal %Lu",
  3692. sbp->sb_ifree, ifree);
  3693. cmn_err(CE_NOTE,
  3694. "xlog_recover_check_summary: sb_fdblocks %Lu freeblks %Lu",
  3695. sbp->sb_fdblocks, freeblks);
  3696. #if 0
  3697. /*
  3698. * This is turned off until I account for the allocation
  3699. * btree blocks which live in free space.
  3700. */
  3701. ASSERT(sbp->sb_icount == itotal);
  3702. ASSERT(sbp->sb_ifree == ifree);
  3703. ASSERT(sbp->sb_fdblocks == freeblks);
  3704. #endif
  3705. #endif
  3706. xfs_buf_relse(sbbp);
  3707. }
  3708. #endif /* DEBUG */