xfs_log_recover.c 110 KB

12345678910111213141516171819202122232425262728293031323334353637383940414243444546474849505152535455565758596061626364656667686970717273747576777879808182838485868788899091929394959697989910010110210310410510610710810911011111211311411511611711811912012112212312412512612712812913013113213313413513613713813914014114214314414514614714814915015115215315415515615715815916016116216316416516616716816917017117217317417517617717817918018118218318418518618718818919019119219319419519619719819920020120220320420520620720820921021121221321421521621721821922022122222322422522622722822923023123223323423523623723823924024124224324424524624724824925025125225325425525625725825926026126226326426526626726826927027127227327427527627727827928028128228328428528628728828929029129229329429529629729829930030130230330430530630730830931031131231331431531631731831932032132232332432532632732832933033133233333433533633733833934034134234334434534634734834935035135235335435535635735835936036136236336436536636736836937037137237337437537637737837938038138238338438538638738838939039139239339439539639739839940040140240340440540640740840941041141241341441541641741841942042142242342442542642742842943043143243343443543643743843944044144244344444544644744844945045145245345445545645745845946046146246346446546646746846947047147247347447547647747847948048148248348448548648748848949049149249349449549649749849950050150250350450550650750850951051151251351451551651751851952052152252352452552652752852953053153253353453553653753853954054154254354454554654754854955055155255355455555655755855956056156256356456556656756856957057157257357457557657757857958058158258358458558658758858959059159259359459559659759859960060160260360460560660760860961061161261361461561661761861962062162262362462562662762862963063163263363463563663763863964064164264364464564664764864965065165265365465565665765865966066166266366466566666766866967067167267367467567667767867968068168268368468568668768868969069169269369469569669769869970070170270370470570670770870971071171271371471571671771871972072172272372472572672772872973073173273373473573673773873974074174274374474574674774874975075175275375475575675775875976076176276376476576676776876977077177277377477577677777877978078178278378478578678778878979079179279379479579679779879980080180280380480580680780880981081181281381481581681781881982082182282382482582682782882983083183283383483583683783883984084184284384484584684784884985085185285385485585685785885986086186286386486586686786886987087187287387487587687787887988088188288388488588688788888989089189289389489589689789889990090190290390490590690790890991091191291391491591691791891992092192292392492592692792892993093193293393493593693793893994094194294394494594694794894995095195295395495595695795895996096196296396496596696796896997097197297397497597697797897998098198298398498598698798898999099199299399499599699799899910001001100210031004100510061007100810091010101110121013101410151016101710181019102010211022102310241025102610271028102910301031103210331034103510361037103810391040104110421043104410451046104710481049105010511052105310541055105610571058105910601061106210631064106510661067106810691070107110721073107410751076107710781079108010811082108310841085108610871088108910901091109210931094109510961097109810991100110111021103110411051106110711081109111011111112111311141115111611171118111911201121112211231124112511261127112811291130113111321133113411351136113711381139114011411142114311441145114611471148114911501151115211531154115511561157115811591160116111621163116411651166116711681169117011711172117311741175117611771178117911801181118211831184118511861187118811891190119111921193119411951196119711981199120012011202120312041205120612071208120912101211121212131214121512161217121812191220122112221223122412251226122712281229123012311232123312341235123612371238123912401241124212431244124512461247124812491250125112521253125412551256125712581259126012611262126312641265126612671268126912701271127212731274127512761277127812791280128112821283128412851286128712881289129012911292129312941295129612971298129913001301130213031304130513061307130813091310131113121313131413151316131713181319132013211322132313241325132613271328132913301331133213331334133513361337133813391340134113421343134413451346134713481349135013511352135313541355135613571358135913601361136213631364136513661367136813691370137113721373137413751376137713781379138013811382138313841385138613871388138913901391139213931394139513961397139813991400140114021403140414051406140714081409141014111412141314141415141614171418141914201421142214231424142514261427142814291430143114321433143414351436143714381439144014411442144314441445144614471448144914501451145214531454145514561457145814591460146114621463146414651466146714681469147014711472147314741475147614771478147914801481148214831484148514861487148814891490149114921493149414951496149714981499150015011502150315041505150615071508150915101511151215131514151515161517151815191520152115221523152415251526152715281529153015311532153315341535153615371538153915401541154215431544154515461547154815491550155115521553155415551556155715581559156015611562156315641565156615671568156915701571157215731574157515761577157815791580158115821583158415851586158715881589159015911592159315941595159615971598159916001601160216031604160516061607160816091610161116121613161416151616161716181619162016211622162316241625162616271628162916301631163216331634163516361637163816391640164116421643164416451646164716481649165016511652165316541655165616571658165916601661166216631664166516661667166816691670167116721673167416751676167716781679168016811682168316841685168616871688168916901691169216931694169516961697169816991700170117021703170417051706170717081709171017111712171317141715171617171718171917201721172217231724172517261727172817291730173117321733173417351736173717381739174017411742174317441745174617471748174917501751175217531754175517561757175817591760176117621763176417651766176717681769177017711772177317741775177617771778177917801781178217831784178517861787178817891790179117921793179417951796179717981799180018011802180318041805180618071808180918101811181218131814181518161817181818191820182118221823182418251826182718281829183018311832183318341835183618371838183918401841184218431844184518461847184818491850185118521853185418551856185718581859186018611862186318641865186618671868186918701871187218731874187518761877187818791880188118821883188418851886188718881889189018911892189318941895189618971898189919001901190219031904190519061907190819091910191119121913191419151916191719181919192019211922192319241925192619271928192919301931193219331934193519361937193819391940194119421943194419451946194719481949195019511952195319541955195619571958195919601961196219631964196519661967196819691970197119721973197419751976197719781979198019811982198319841985198619871988198919901991199219931994199519961997199819992000200120022003200420052006200720082009201020112012201320142015201620172018201920202021202220232024202520262027202820292030203120322033203420352036203720382039204020412042204320442045204620472048204920502051205220532054205520562057205820592060206120622063206420652066206720682069207020712072207320742075207620772078207920802081208220832084208520862087208820892090209120922093209420952096209720982099210021012102210321042105210621072108210921102111211221132114211521162117211821192120212121222123212421252126212721282129213021312132213321342135213621372138213921402141214221432144214521462147214821492150215121522153215421552156215721582159216021612162216321642165216621672168216921702171217221732174217521762177217821792180218121822183218421852186218721882189219021912192219321942195219621972198219922002201220222032204220522062207220822092210221122122213221422152216221722182219222022212222222322242225222622272228222922302231223222332234223522362237223822392240224122422243224422452246224722482249225022512252225322542255225622572258225922602261226222632264226522662267226822692270227122722273227422752276227722782279228022812282228322842285228622872288228922902291229222932294229522962297229822992300230123022303230423052306230723082309231023112312231323142315231623172318231923202321232223232324232523262327232823292330233123322333233423352336233723382339234023412342234323442345234623472348234923502351235223532354235523562357235823592360236123622363236423652366236723682369237023712372237323742375237623772378237923802381238223832384238523862387238823892390239123922393239423952396239723982399240024012402240324042405240624072408240924102411241224132414241524162417241824192420242124222423242424252426242724282429243024312432243324342435243624372438243924402441244224432444244524462447244824492450245124522453245424552456245724582459246024612462246324642465246624672468246924702471247224732474247524762477247824792480248124822483248424852486248724882489249024912492249324942495249624972498249925002501250225032504250525062507250825092510251125122513251425152516251725182519252025212522252325242525252625272528252925302531253225332534253525362537253825392540254125422543254425452546254725482549255025512552255325542555255625572558255925602561256225632564256525662567256825692570257125722573257425752576257725782579258025812582258325842585258625872588258925902591259225932594259525962597259825992600260126022603260426052606260726082609261026112612261326142615261626172618261926202621262226232624262526262627262826292630263126322633263426352636263726382639264026412642264326442645264626472648264926502651265226532654265526562657265826592660266126622663266426652666266726682669267026712672267326742675267626772678267926802681268226832684268526862687268826892690269126922693269426952696269726982699270027012702270327042705270627072708270927102711271227132714271527162717271827192720272127222723272427252726272727282729273027312732273327342735273627372738273927402741274227432744274527462747274827492750275127522753275427552756275727582759276027612762276327642765276627672768276927702771277227732774277527762777277827792780278127822783278427852786278727882789279027912792279327942795279627972798279928002801280228032804280528062807280828092810281128122813281428152816281728182819282028212822282328242825282628272828282928302831283228332834283528362837283828392840284128422843284428452846284728482849285028512852285328542855285628572858285928602861286228632864286528662867286828692870287128722873287428752876287728782879288028812882288328842885288628872888288928902891289228932894289528962897289828992900290129022903290429052906290729082909291029112912291329142915291629172918291929202921292229232924292529262927292829292930293129322933293429352936293729382939294029412942294329442945294629472948294929502951295229532954295529562957295829592960296129622963296429652966296729682969297029712972297329742975297629772978297929802981298229832984298529862987298829892990299129922993299429952996299729982999300030013002300330043005300630073008300930103011301230133014301530163017301830193020302130223023302430253026302730283029303030313032303330343035303630373038303930403041304230433044304530463047304830493050305130523053305430553056305730583059306030613062306330643065306630673068306930703071307230733074307530763077307830793080308130823083308430853086308730883089309030913092309330943095309630973098309931003101310231033104310531063107310831093110311131123113311431153116311731183119312031213122312331243125312631273128312931303131313231333134313531363137313831393140314131423143314431453146314731483149315031513152315331543155315631573158315931603161316231633164316531663167316831693170317131723173317431753176317731783179318031813182318331843185318631873188318931903191319231933194319531963197319831993200320132023203320432053206320732083209321032113212321332143215321632173218321932203221322232233224322532263227322832293230323132323233323432353236323732383239324032413242324332443245324632473248324932503251325232533254325532563257325832593260326132623263326432653266326732683269327032713272327332743275327632773278327932803281328232833284328532863287328832893290329132923293329432953296329732983299330033013302330333043305330633073308330933103311331233133314331533163317331833193320332133223323332433253326332733283329333033313332333333343335333633373338333933403341334233433344334533463347334833493350335133523353335433553356335733583359336033613362336333643365336633673368336933703371337233733374337533763377337833793380338133823383338433853386338733883389339033913392339333943395339633973398339934003401340234033404340534063407340834093410341134123413341434153416341734183419342034213422342334243425342634273428342934303431343234333434343534363437343834393440344134423443344434453446344734483449345034513452345334543455345634573458345934603461346234633464346534663467346834693470347134723473347434753476347734783479348034813482348334843485348634873488348934903491349234933494349534963497349834993500350135023503350435053506350735083509351035113512351335143515351635173518351935203521352235233524352535263527352835293530353135323533353435353536353735383539354035413542354335443545354635473548354935503551355235533554355535563557355835593560356135623563356435653566356735683569357035713572357335743575357635773578357935803581358235833584358535863587358835893590359135923593359435953596359735983599360036013602360336043605360636073608360936103611361236133614361536163617361836193620362136223623362436253626362736283629363036313632363336343635363636373638363936403641364236433644364536463647364836493650365136523653365436553656365736583659366036613662366336643665366636673668366936703671367236733674367536763677367836793680368136823683368436853686368736883689369036913692369336943695369636973698369937003701370237033704370537063707370837093710371137123713371437153716371737183719372037213722372337243725372637273728372937303731373237333734373537363737373837393740374137423743374437453746374737483749375037513752375337543755375637573758375937603761376237633764376537663767376837693770377137723773377437753776377737783779378037813782378337843785378637873788378937903791379237933794379537963797379837993800380138023803380438053806380738083809381038113812381338143815381638173818381938203821382238233824382538263827382838293830383138323833383438353836383738383839384038413842384338443845384638473848384938503851385238533854385538563857385838593860386138623863386438653866386738683869387038713872387338743875387638773878387938803881388238833884388538863887388838893890389138923893389438953896389738983899390039013902390339043905390639073908390939103911391239133914391539163917391839193920392139223923392439253926392739283929393039313932393339343935393639373938393939403941394239433944394539463947394839493950395139523953395439553956395739583959396039613962396339643965396639673968396939703971397239733974397539763977397839793980398139823983398439853986398739883989399039913992399339943995399639973998399940004001400240034004400540064007400840094010401140124013401440154016401740184019402040214022402340244025402640274028402940304031403240334034403540364037403840394040404140424043404440454046404740484049405040514052405340544055405640574058
  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. #include "xfs_utils.h"
  50. STATIC int xlog_find_zeroed(xlog_t *, xfs_daddr_t *);
  51. STATIC int xlog_clear_stale_blocks(xlog_t *, xfs_lsn_t);
  52. STATIC void xlog_recover_insert_item_backq(xlog_recover_item_t **q,
  53. xlog_recover_item_t *item);
  54. #if defined(DEBUG)
  55. STATIC void xlog_recover_check_summary(xlog_t *);
  56. #else
  57. #define xlog_recover_check_summary(log)
  58. #endif
  59. /*
  60. * Sector aligned buffer routines for buffer create/read/write/access
  61. */
  62. #define XLOG_SECTOR_ROUNDUP_BBCOUNT(log, bbs) \
  63. ( ((log)->l_sectbb_mask && (bbs & (log)->l_sectbb_mask)) ? \
  64. ((bbs + (log)->l_sectbb_mask + 1) & ~(log)->l_sectbb_mask) : (bbs) )
  65. #define XLOG_SECTOR_ROUNDDOWN_BLKNO(log, bno) ((bno) & ~(log)->l_sectbb_mask)
  66. xfs_buf_t *
  67. xlog_get_bp(
  68. xlog_t *log,
  69. int num_bblks)
  70. {
  71. ASSERT(num_bblks > 0);
  72. if (log->l_sectbb_log) {
  73. if (num_bblks > 1)
  74. num_bblks += XLOG_SECTOR_ROUNDUP_BBCOUNT(log, 1);
  75. num_bblks = XLOG_SECTOR_ROUNDUP_BBCOUNT(log, num_bblks);
  76. }
  77. return xfs_buf_get_noaddr(BBTOB(num_bblks), log->l_mp->m_logdev_targp);
  78. }
  79. void
  80. xlog_put_bp(
  81. xfs_buf_t *bp)
  82. {
  83. xfs_buf_free(bp);
  84. }
  85. /*
  86. * nbblks should be uint, but oh well. Just want to catch that 32-bit length.
  87. */
  88. int
  89. xlog_bread(
  90. xlog_t *log,
  91. xfs_daddr_t blk_no,
  92. int nbblks,
  93. xfs_buf_t *bp)
  94. {
  95. int error;
  96. if (log->l_sectbb_log) {
  97. blk_no = XLOG_SECTOR_ROUNDDOWN_BLKNO(log, blk_no);
  98. nbblks = XLOG_SECTOR_ROUNDUP_BBCOUNT(log, nbblks);
  99. }
  100. ASSERT(nbblks > 0);
  101. ASSERT(BBTOB(nbblks) <= XFS_BUF_SIZE(bp));
  102. ASSERT(bp);
  103. XFS_BUF_SET_ADDR(bp, log->l_logBBstart + blk_no);
  104. XFS_BUF_READ(bp);
  105. XFS_BUF_BUSY(bp);
  106. XFS_BUF_SET_COUNT(bp, BBTOB(nbblks));
  107. XFS_BUF_SET_TARGET(bp, log->l_mp->m_logdev_targp);
  108. xfsbdstrat(log->l_mp, bp);
  109. error = xfs_iowait(bp);
  110. if (error)
  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 = ", __func__);
  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", be32_to_cpu(head->h_fmt));
  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(be32_to_cpu(head->h_magicno) == 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(be32_to_cpu(head->h_fmt) != 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(be32_to_cpu(head->h_magicno) == 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. STATIC 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 = xlog_get_cycle(offset);
  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 = xlog_get_cycle(buf);
  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 == be32_to_cpu(head->h_magicno))
  401. break;
  402. if (!smallmem)
  403. offset -= BBSIZE;
  404. }
  405. /*
  406. * We hit the beginning of the physical log & still no header. Return
  407. * to caller. If caller can handle a return of -1, then this routine
  408. * will be called again for the end of the physical log.
  409. */
  410. if (i == -1) {
  411. error = -1;
  412. goto out;
  413. }
  414. /*
  415. * We have the final block of the good log (the first block
  416. * of the log record _before_ the head. So we check the uuid.
  417. */
  418. if ((error = xlog_header_check_mount(log->l_mp, head)))
  419. goto out;
  420. /*
  421. * We may have found a log record header before we expected one.
  422. * last_blk will be the 1st block # with a given cycle #. We may end
  423. * up reading an entire log record. In this case, we don't want to
  424. * reset last_blk. Only when last_blk points in the middle of a log
  425. * record do we update last_blk.
  426. */
  427. if (xfs_sb_version_haslogv2(&log->l_mp->m_sb)) {
  428. uint h_size = be32_to_cpu(head->h_size);
  429. xhdrs = h_size / XLOG_HEADER_CYCLE_SIZE;
  430. if (h_size % XLOG_HEADER_CYCLE_SIZE)
  431. xhdrs++;
  432. } else {
  433. xhdrs = 1;
  434. }
  435. if (*last_blk - i + extra_bblks !=
  436. BTOBB(be32_to_cpu(head->h_len)) + xhdrs)
  437. *last_blk = i;
  438. out:
  439. xlog_put_bp(bp);
  440. return error;
  441. }
  442. /*
  443. * Head is defined to be the point of the log where the next log write
  444. * write could go. This means that incomplete LR writes at the end are
  445. * eliminated when calculating the head. We aren't guaranteed that previous
  446. * LR have complete transactions. We only know that a cycle number of
  447. * current cycle number -1 won't be present in the log if we start writing
  448. * from our current block number.
  449. *
  450. * last_blk contains the block number of the first block with a given
  451. * cycle number.
  452. *
  453. * Return: zero if normal, non-zero if error.
  454. */
  455. STATIC int
  456. xlog_find_head(
  457. xlog_t *log,
  458. xfs_daddr_t *return_head_blk)
  459. {
  460. xfs_buf_t *bp;
  461. xfs_caddr_t offset;
  462. xfs_daddr_t new_blk, first_blk, start_blk, last_blk, head_blk;
  463. int num_scan_bblks;
  464. uint first_half_cycle, last_half_cycle;
  465. uint stop_on_cycle;
  466. int error, log_bbnum = log->l_logBBsize;
  467. /* Is the end of the log device zeroed? */
  468. if ((error = xlog_find_zeroed(log, &first_blk)) == -1) {
  469. *return_head_blk = first_blk;
  470. /* Is the whole lot zeroed? */
  471. if (!first_blk) {
  472. /* Linux XFS shouldn't generate totally zeroed logs -
  473. * mkfs etc write a dummy unmount record to a fresh
  474. * log so we can store the uuid in there
  475. */
  476. xlog_warn("XFS: totally zeroed log");
  477. }
  478. return 0;
  479. } else if (error) {
  480. xlog_warn("XFS: empty log check failed");
  481. return error;
  482. }
  483. first_blk = 0; /* get cycle # of 1st block */
  484. bp = xlog_get_bp(log, 1);
  485. if (!bp)
  486. return ENOMEM;
  487. if ((error = xlog_bread(log, 0, 1, bp)))
  488. goto bp_err;
  489. offset = xlog_align(log, 0, 1, bp);
  490. first_half_cycle = xlog_get_cycle(offset);
  491. last_blk = head_blk = log_bbnum - 1; /* get cycle # of last block */
  492. if ((error = xlog_bread(log, last_blk, 1, bp)))
  493. goto bp_err;
  494. offset = xlog_align(log, last_blk, 1, bp);
  495. last_half_cycle = xlog_get_cycle(offset);
  496. ASSERT(last_half_cycle != 0);
  497. /*
  498. * If the 1st half cycle number is equal to the last half cycle number,
  499. * then the entire log is stamped with the same cycle number. In this
  500. * case, head_blk can't be set to zero (which makes sense). The below
  501. * math doesn't work out properly with head_blk equal to zero. Instead,
  502. * we set it to log_bbnum which is an invalid block number, but this
  503. * value makes the math correct. If head_blk doesn't changed through
  504. * all the tests below, *head_blk is set to zero at the very end rather
  505. * than log_bbnum. In a sense, log_bbnum and zero are the same block
  506. * in a circular file.
  507. */
  508. if (first_half_cycle == last_half_cycle) {
  509. /*
  510. * In this case we believe that the entire log should have
  511. * cycle number last_half_cycle. We need to scan backwards
  512. * from the end verifying that there are no holes still
  513. * containing last_half_cycle - 1. If we find such a hole,
  514. * then the start of that hole will be the new head. The
  515. * simple case looks like
  516. * x | x ... | x - 1 | x
  517. * Another case that fits this picture would be
  518. * x | x + 1 | x ... | x
  519. * In this case the head really is somewhere at the end of the
  520. * log, as one of the latest writes at the beginning was
  521. * incomplete.
  522. * One more case is
  523. * x | x + 1 | x ... | x - 1 | x
  524. * This is really the combination of the above two cases, and
  525. * the head has to end up at the start of the x-1 hole at the
  526. * end of the log.
  527. *
  528. * In the 256k log case, we will read from the beginning to the
  529. * end of the log and search for cycle numbers equal to x-1.
  530. * We don't worry about the x+1 blocks that we encounter,
  531. * because we know that they cannot be the head since the log
  532. * started with x.
  533. */
  534. head_blk = log_bbnum;
  535. stop_on_cycle = last_half_cycle - 1;
  536. } else {
  537. /*
  538. * In this case we want to find the first block with cycle
  539. * number matching last_half_cycle. We expect the log to be
  540. * some variation on
  541. * x + 1 ... | x ...
  542. * The first block with cycle number x (last_half_cycle) will
  543. * be where the new head belongs. First we do a binary search
  544. * for the first occurrence of last_half_cycle. The binary
  545. * search may not be totally accurate, so then we scan back
  546. * from there looking for occurrences of last_half_cycle before
  547. * us. If that backwards scan wraps around the beginning of
  548. * the log, then we look for occurrences of last_half_cycle - 1
  549. * at the end of the log. The cases we're looking for look
  550. * like
  551. * x + 1 ... | x | x + 1 | x ...
  552. * ^ binary search stopped here
  553. * or
  554. * x + 1 ... | x ... | x - 1 | x
  555. * <---------> less than scan distance
  556. */
  557. stop_on_cycle = last_half_cycle;
  558. if ((error = xlog_find_cycle_start(log, bp, first_blk,
  559. &head_blk, last_half_cycle)))
  560. goto bp_err;
  561. }
  562. /*
  563. * Now validate the answer. Scan back some number of maximum possible
  564. * blocks and make sure each one has the expected cycle number. The
  565. * maximum is determined by the total possible amount of buffering
  566. * in the in-core log. The following number can be made tighter if
  567. * we actually look at the block size of the filesystem.
  568. */
  569. num_scan_bblks = XLOG_TOTAL_REC_SHIFT(log);
  570. if (head_blk >= num_scan_bblks) {
  571. /*
  572. * We are guaranteed that the entire check can be performed
  573. * in one buffer.
  574. */
  575. start_blk = head_blk - num_scan_bblks;
  576. if ((error = xlog_find_verify_cycle(log,
  577. start_blk, num_scan_bblks,
  578. stop_on_cycle, &new_blk)))
  579. goto bp_err;
  580. if (new_blk != -1)
  581. head_blk = new_blk;
  582. } else { /* need to read 2 parts of log */
  583. /*
  584. * We are going to scan backwards in the log in two parts.
  585. * First we scan the physical end of the log. In this part
  586. * of the log, we are looking for blocks with cycle number
  587. * last_half_cycle - 1.
  588. * If we find one, then we know that the log starts there, as
  589. * we've found a hole that didn't get written in going around
  590. * the end of the physical log. The simple case for this is
  591. * x + 1 ... | x ... | x - 1 | x
  592. * <---------> less than scan distance
  593. * If all of the blocks at the end of the log have cycle number
  594. * last_half_cycle, then we check the blocks at the start of
  595. * the log looking for occurrences of last_half_cycle. If we
  596. * find one, then our current estimate for the location of the
  597. * first occurrence of last_half_cycle is wrong and we move
  598. * back to the hole we've found. This case looks like
  599. * x + 1 ... | x | x + 1 | x ...
  600. * ^ binary search stopped here
  601. * Another case we need to handle that only occurs in 256k
  602. * logs is
  603. * x + 1 ... | x ... | x+1 | x ...
  604. * ^ binary search stops here
  605. * In a 256k log, the scan at the end of the log will see the
  606. * x + 1 blocks. We need to skip past those since that is
  607. * certainly not the head of the log. By searching for
  608. * last_half_cycle-1 we accomplish that.
  609. */
  610. start_blk = log_bbnum - num_scan_bblks + head_blk;
  611. ASSERT(head_blk <= INT_MAX &&
  612. (xfs_daddr_t) num_scan_bblks - head_blk >= 0);
  613. if ((error = xlog_find_verify_cycle(log, start_blk,
  614. num_scan_bblks - (int)head_blk,
  615. (stop_on_cycle - 1), &new_blk)))
  616. goto bp_err;
  617. if (new_blk != -1) {
  618. head_blk = new_blk;
  619. goto bad_blk;
  620. }
  621. /*
  622. * Scan beginning of log now. The last part of the physical
  623. * log is good. This scan needs to verify that it doesn't find
  624. * the last_half_cycle.
  625. */
  626. start_blk = 0;
  627. ASSERT(head_blk <= INT_MAX);
  628. if ((error = xlog_find_verify_cycle(log,
  629. start_blk, (int)head_blk,
  630. stop_on_cycle, &new_blk)))
  631. goto bp_err;
  632. if (new_blk != -1)
  633. head_blk = new_blk;
  634. }
  635. bad_blk:
  636. /*
  637. * Now we need to make sure head_blk is not pointing to a block in
  638. * the middle of a log record.
  639. */
  640. num_scan_bblks = XLOG_REC_SHIFT(log);
  641. if (head_blk >= num_scan_bblks) {
  642. start_blk = head_blk - num_scan_bblks; /* don't read head_blk */
  643. /* start ptr at last block ptr before head_blk */
  644. if ((error = xlog_find_verify_log_record(log, start_blk,
  645. &head_blk, 0)) == -1) {
  646. error = XFS_ERROR(EIO);
  647. goto bp_err;
  648. } else if (error)
  649. goto bp_err;
  650. } else {
  651. start_blk = 0;
  652. ASSERT(head_blk <= INT_MAX);
  653. if ((error = xlog_find_verify_log_record(log, start_blk,
  654. &head_blk, 0)) == -1) {
  655. /* We hit the beginning of the log during our search */
  656. start_blk = log_bbnum - num_scan_bblks + head_blk;
  657. new_blk = log_bbnum;
  658. ASSERT(start_blk <= INT_MAX &&
  659. (xfs_daddr_t) log_bbnum-start_blk >= 0);
  660. ASSERT(head_blk <= INT_MAX);
  661. if ((error = xlog_find_verify_log_record(log,
  662. start_blk, &new_blk,
  663. (int)head_blk)) == -1) {
  664. error = XFS_ERROR(EIO);
  665. goto bp_err;
  666. } else if (error)
  667. goto bp_err;
  668. if (new_blk != log_bbnum)
  669. head_blk = new_blk;
  670. } else if (error)
  671. goto bp_err;
  672. }
  673. xlog_put_bp(bp);
  674. if (head_blk == log_bbnum)
  675. *return_head_blk = 0;
  676. else
  677. *return_head_blk = head_blk;
  678. /*
  679. * When returning here, we have a good block number. Bad block
  680. * means that during a previous crash, we didn't have a clean break
  681. * from cycle number N to cycle number N-1. In this case, we need
  682. * to find the first block with cycle number N-1.
  683. */
  684. return 0;
  685. bp_err:
  686. xlog_put_bp(bp);
  687. if (error)
  688. xlog_warn("XFS: failed to find log head");
  689. return error;
  690. }
  691. /*
  692. * Find the sync block number or the tail of the log.
  693. *
  694. * This will be the block number of the last record to have its
  695. * associated buffers synced to disk. Every log record header has
  696. * a sync lsn embedded in it. LSNs hold block numbers, so it is easy
  697. * to get a sync block number. The only concern is to figure out which
  698. * log record header to believe.
  699. *
  700. * The following algorithm uses the log record header with the largest
  701. * lsn. The entire log record does not need to be valid. We only care
  702. * that the header is valid.
  703. *
  704. * We could speed up search by using current head_blk buffer, but it is not
  705. * available.
  706. */
  707. int
  708. xlog_find_tail(
  709. xlog_t *log,
  710. xfs_daddr_t *head_blk,
  711. xfs_daddr_t *tail_blk)
  712. {
  713. xlog_rec_header_t *rhead;
  714. xlog_op_header_t *op_head;
  715. xfs_caddr_t offset = NULL;
  716. xfs_buf_t *bp;
  717. int error, i, found;
  718. xfs_daddr_t umount_data_blk;
  719. xfs_daddr_t after_umount_blk;
  720. xfs_lsn_t tail_lsn;
  721. int hblks;
  722. found = 0;
  723. /*
  724. * Find previous log record
  725. */
  726. if ((error = xlog_find_head(log, head_blk)))
  727. return error;
  728. bp = xlog_get_bp(log, 1);
  729. if (!bp)
  730. return ENOMEM;
  731. if (*head_blk == 0) { /* special case */
  732. if ((error = xlog_bread(log, 0, 1, bp)))
  733. goto bread_err;
  734. offset = xlog_align(log, 0, 1, bp);
  735. if (xlog_get_cycle(offset) == 0) {
  736. *tail_blk = 0;
  737. /* leave all other log inited values alone */
  738. goto exit;
  739. }
  740. }
  741. /*
  742. * Search backwards looking for log record header block
  743. */
  744. ASSERT(*head_blk < INT_MAX);
  745. for (i = (int)(*head_blk) - 1; i >= 0; i--) {
  746. if ((error = xlog_bread(log, i, 1, bp)))
  747. goto bread_err;
  748. offset = xlog_align(log, i, 1, bp);
  749. if (XLOG_HEADER_MAGIC_NUM == be32_to_cpu(*(__be32 *)offset)) {
  750. found = 1;
  751. break;
  752. }
  753. }
  754. /*
  755. * If we haven't found the log record header block, start looking
  756. * again from the end of the physical log. XXXmiken: There should be
  757. * a check here to make sure we didn't search more than N blocks in
  758. * the previous code.
  759. */
  760. if (!found) {
  761. for (i = log->l_logBBsize - 1; i >= (int)(*head_blk); i--) {
  762. if ((error = xlog_bread(log, i, 1, bp)))
  763. goto bread_err;
  764. offset = xlog_align(log, i, 1, bp);
  765. if (XLOG_HEADER_MAGIC_NUM ==
  766. be32_to_cpu(*(__be32 *)offset)) {
  767. found = 2;
  768. break;
  769. }
  770. }
  771. }
  772. if (!found) {
  773. xlog_warn("XFS: xlog_find_tail: couldn't find sync record");
  774. ASSERT(0);
  775. return XFS_ERROR(EIO);
  776. }
  777. /* find blk_no of tail of log */
  778. rhead = (xlog_rec_header_t *)offset;
  779. *tail_blk = BLOCK_LSN(be64_to_cpu(rhead->h_tail_lsn));
  780. /*
  781. * Reset log values according to the state of the log when we
  782. * crashed. In the case where head_blk == 0, we bump curr_cycle
  783. * one because the next write starts a new cycle rather than
  784. * continuing the cycle of the last good log record. At this
  785. * point we have guaranteed that all partial log records have been
  786. * accounted for. Therefore, we know that the last good log record
  787. * written was complete and ended exactly on the end boundary
  788. * of the physical log.
  789. */
  790. log->l_prev_block = i;
  791. log->l_curr_block = (int)*head_blk;
  792. log->l_curr_cycle = be32_to_cpu(rhead->h_cycle);
  793. if (found == 2)
  794. log->l_curr_cycle++;
  795. log->l_tail_lsn = be64_to_cpu(rhead->h_tail_lsn);
  796. log->l_last_sync_lsn = be64_to_cpu(rhead->h_lsn);
  797. log->l_grant_reserve_cycle = log->l_curr_cycle;
  798. log->l_grant_reserve_bytes = BBTOB(log->l_curr_block);
  799. log->l_grant_write_cycle = log->l_curr_cycle;
  800. log->l_grant_write_bytes = BBTOB(log->l_curr_block);
  801. /*
  802. * Look for unmount record. If we find it, then we know there
  803. * was a clean unmount. Since 'i' could be the last block in
  804. * the physical log, we convert to a log block before comparing
  805. * to the head_blk.
  806. *
  807. * Save the current tail lsn to use to pass to
  808. * xlog_clear_stale_blocks() below. We won't want to clear the
  809. * unmount record if there is one, so we pass the lsn of the
  810. * unmount record rather than the block after it.
  811. */
  812. if (xfs_sb_version_haslogv2(&log->l_mp->m_sb)) {
  813. int h_size = be32_to_cpu(rhead->h_size);
  814. int h_version = be32_to_cpu(rhead->h_version);
  815. if ((h_version & XLOG_VERSION_2) &&
  816. (h_size > XLOG_HEADER_CYCLE_SIZE)) {
  817. hblks = h_size / XLOG_HEADER_CYCLE_SIZE;
  818. if (h_size % XLOG_HEADER_CYCLE_SIZE)
  819. hblks++;
  820. } else {
  821. hblks = 1;
  822. }
  823. } else {
  824. hblks = 1;
  825. }
  826. after_umount_blk = (i + hblks + (int)
  827. BTOBB(be32_to_cpu(rhead->h_len))) % log->l_logBBsize;
  828. tail_lsn = log->l_tail_lsn;
  829. if (*head_blk == after_umount_blk &&
  830. be32_to_cpu(rhead->h_num_logops) == 1) {
  831. umount_data_blk = (i + hblks) % log->l_logBBsize;
  832. if ((error = xlog_bread(log, umount_data_blk, 1, bp))) {
  833. goto bread_err;
  834. }
  835. offset = xlog_align(log, umount_data_blk, 1, bp);
  836. op_head = (xlog_op_header_t *)offset;
  837. if (op_head->oh_flags & XLOG_UNMOUNT_TRANS) {
  838. /*
  839. * Set tail and last sync so that newly written
  840. * log records will point recovery to after the
  841. * current unmount record.
  842. */
  843. log->l_tail_lsn =
  844. xlog_assign_lsn(log->l_curr_cycle,
  845. after_umount_blk);
  846. log->l_last_sync_lsn =
  847. xlog_assign_lsn(log->l_curr_cycle,
  848. after_umount_blk);
  849. *tail_blk = after_umount_blk;
  850. /*
  851. * Note that the unmount was clean. If the unmount
  852. * was not clean, we need to know this to rebuild the
  853. * superblock counters from the perag headers if we
  854. * have a filesystem using non-persistent counters.
  855. */
  856. log->l_mp->m_flags |= XFS_MOUNT_WAS_CLEAN;
  857. }
  858. }
  859. /*
  860. * Make sure that there are no blocks in front of the head
  861. * with the same cycle number as the head. This can happen
  862. * because we allow multiple outstanding log writes concurrently,
  863. * and the later writes might make it out before earlier ones.
  864. *
  865. * We use the lsn from before modifying it so that we'll never
  866. * overwrite the unmount record after a clean unmount.
  867. *
  868. * Do this only if we are going to recover the filesystem
  869. *
  870. * NOTE: This used to say "if (!readonly)"
  871. * However on Linux, we can & do recover a read-only filesystem.
  872. * We only skip recovery if NORECOVERY is specified on mount,
  873. * in which case we would not be here.
  874. *
  875. * But... if the -device- itself is readonly, just skip this.
  876. * We can't recover this device anyway, so it won't matter.
  877. */
  878. if (!xfs_readonly_buftarg(log->l_mp->m_logdev_targp)) {
  879. error = xlog_clear_stale_blocks(log, tail_lsn);
  880. }
  881. bread_err:
  882. exit:
  883. xlog_put_bp(bp);
  884. if (error)
  885. xlog_warn("XFS: failed to locate log tail");
  886. return error;
  887. }
  888. /*
  889. * Is the log zeroed at all?
  890. *
  891. * The last binary search should be changed to perform an X block read
  892. * once X becomes small enough. You can then search linearly through
  893. * the X blocks. This will cut down on the number of reads we need to do.
  894. *
  895. * If the log is partially zeroed, this routine will pass back the blkno
  896. * of the first block with cycle number 0. It won't have a complete LR
  897. * preceding it.
  898. *
  899. * Return:
  900. * 0 => the log is completely written to
  901. * -1 => use *blk_no as the first block of the log
  902. * >0 => error has occurred
  903. */
  904. STATIC int
  905. xlog_find_zeroed(
  906. xlog_t *log,
  907. xfs_daddr_t *blk_no)
  908. {
  909. xfs_buf_t *bp;
  910. xfs_caddr_t offset;
  911. uint first_cycle, last_cycle;
  912. xfs_daddr_t new_blk, last_blk, start_blk;
  913. xfs_daddr_t num_scan_bblks;
  914. int error, log_bbnum = log->l_logBBsize;
  915. *blk_no = 0;
  916. /* check totally zeroed log */
  917. bp = xlog_get_bp(log, 1);
  918. if (!bp)
  919. return ENOMEM;
  920. if ((error = xlog_bread(log, 0, 1, bp)))
  921. goto bp_err;
  922. offset = xlog_align(log, 0, 1, bp);
  923. first_cycle = xlog_get_cycle(offset);
  924. if (first_cycle == 0) { /* completely zeroed log */
  925. *blk_no = 0;
  926. xlog_put_bp(bp);
  927. return -1;
  928. }
  929. /* check partially zeroed log */
  930. if ((error = xlog_bread(log, log_bbnum-1, 1, bp)))
  931. goto bp_err;
  932. offset = xlog_align(log, log_bbnum-1, 1, bp);
  933. last_cycle = xlog_get_cycle(offset);
  934. if (last_cycle != 0) { /* log completely written to */
  935. xlog_put_bp(bp);
  936. return 0;
  937. } else if (first_cycle != 1) {
  938. /*
  939. * If the cycle of the last block is zero, the cycle of
  940. * the first block must be 1. If it's not, maybe we're
  941. * not looking at a log... Bail out.
  942. */
  943. xlog_warn("XFS: Log inconsistent or not a log (last==0, first!=1)");
  944. return XFS_ERROR(EINVAL);
  945. }
  946. /* we have a partially zeroed log */
  947. last_blk = log_bbnum-1;
  948. if ((error = xlog_find_cycle_start(log, bp, 0, &last_blk, 0)))
  949. goto bp_err;
  950. /*
  951. * Validate the answer. Because there is no way to guarantee that
  952. * the entire log is made up of log records which are the same size,
  953. * we scan over the defined maximum blocks. At this point, the maximum
  954. * is not chosen to mean anything special. XXXmiken
  955. */
  956. num_scan_bblks = XLOG_TOTAL_REC_SHIFT(log);
  957. ASSERT(num_scan_bblks <= INT_MAX);
  958. if (last_blk < num_scan_bblks)
  959. num_scan_bblks = last_blk;
  960. start_blk = last_blk - num_scan_bblks;
  961. /*
  962. * We search for any instances of cycle number 0 that occur before
  963. * our current estimate of the head. What we're trying to detect is
  964. * 1 ... | 0 | 1 | 0...
  965. * ^ binary search ends here
  966. */
  967. if ((error = xlog_find_verify_cycle(log, start_blk,
  968. (int)num_scan_bblks, 0, &new_blk)))
  969. goto bp_err;
  970. if (new_blk != -1)
  971. last_blk = new_blk;
  972. /*
  973. * Potentially backup over partial log record write. We don't need
  974. * to search the end of the log because we know it is zero.
  975. */
  976. if ((error = xlog_find_verify_log_record(log, start_blk,
  977. &last_blk, 0)) == -1) {
  978. error = XFS_ERROR(EIO);
  979. goto bp_err;
  980. } else if (error)
  981. goto bp_err;
  982. *blk_no = last_blk;
  983. bp_err:
  984. xlog_put_bp(bp);
  985. if (error)
  986. return error;
  987. return -1;
  988. }
  989. /*
  990. * These are simple subroutines used by xlog_clear_stale_blocks() below
  991. * to initialize a buffer full of empty log record headers and write
  992. * them into the log.
  993. */
  994. STATIC void
  995. xlog_add_record(
  996. xlog_t *log,
  997. xfs_caddr_t buf,
  998. int cycle,
  999. int block,
  1000. int tail_cycle,
  1001. int tail_block)
  1002. {
  1003. xlog_rec_header_t *recp = (xlog_rec_header_t *)buf;
  1004. memset(buf, 0, BBSIZE);
  1005. recp->h_magicno = cpu_to_be32(XLOG_HEADER_MAGIC_NUM);
  1006. recp->h_cycle = cpu_to_be32(cycle);
  1007. recp->h_version = cpu_to_be32(
  1008. xfs_sb_version_haslogv2(&log->l_mp->m_sb) ? 2 : 1);
  1009. recp->h_lsn = cpu_to_be64(xlog_assign_lsn(cycle, block));
  1010. recp->h_tail_lsn = cpu_to_be64(xlog_assign_lsn(tail_cycle, tail_block));
  1011. recp->h_fmt = cpu_to_be32(XLOG_FMT);
  1012. memcpy(&recp->h_fs_uuid, &log->l_mp->m_sb.sb_uuid, sizeof(uuid_t));
  1013. }
  1014. STATIC int
  1015. xlog_write_log_records(
  1016. xlog_t *log,
  1017. int cycle,
  1018. int start_block,
  1019. int blocks,
  1020. int tail_cycle,
  1021. int tail_block)
  1022. {
  1023. xfs_caddr_t offset;
  1024. xfs_buf_t *bp;
  1025. int balign, ealign;
  1026. int sectbb = XLOG_SECTOR_ROUNDUP_BBCOUNT(log, 1);
  1027. int end_block = start_block + blocks;
  1028. int bufblks;
  1029. int error = 0;
  1030. int i, j = 0;
  1031. bufblks = 1 << ffs(blocks);
  1032. while (!(bp = xlog_get_bp(log, bufblks))) {
  1033. bufblks >>= 1;
  1034. if (bufblks <= log->l_sectbb_log)
  1035. return ENOMEM;
  1036. }
  1037. /* We may need to do a read at the start to fill in part of
  1038. * the buffer in the starting sector not covered by the first
  1039. * write below.
  1040. */
  1041. balign = XLOG_SECTOR_ROUNDDOWN_BLKNO(log, start_block);
  1042. if (balign != start_block) {
  1043. if ((error = xlog_bread(log, start_block, 1, bp))) {
  1044. xlog_put_bp(bp);
  1045. return error;
  1046. }
  1047. j = start_block - balign;
  1048. }
  1049. for (i = start_block; i < end_block; i += bufblks) {
  1050. int bcount, endcount;
  1051. bcount = min(bufblks, end_block - start_block);
  1052. endcount = bcount - j;
  1053. /* We may need to do a read at the end to fill in part of
  1054. * the buffer in the final sector not covered by the write.
  1055. * If this is the same sector as the above read, skip it.
  1056. */
  1057. ealign = XLOG_SECTOR_ROUNDDOWN_BLKNO(log, end_block);
  1058. if (j == 0 && (start_block + endcount > ealign)) {
  1059. offset = XFS_BUF_PTR(bp);
  1060. balign = BBTOB(ealign - start_block);
  1061. error = XFS_BUF_SET_PTR(bp, offset + balign,
  1062. BBTOB(sectbb));
  1063. if (!error)
  1064. error = xlog_bread(log, ealign, sectbb, bp);
  1065. if (!error)
  1066. error = XFS_BUF_SET_PTR(bp, offset, bufblks);
  1067. if (error)
  1068. break;
  1069. }
  1070. offset = xlog_align(log, start_block, endcount, bp);
  1071. for (; j < endcount; j++) {
  1072. xlog_add_record(log, offset, cycle, i+j,
  1073. tail_cycle, tail_block);
  1074. offset += BBSIZE;
  1075. }
  1076. error = xlog_bwrite(log, start_block, endcount, bp);
  1077. if (error)
  1078. break;
  1079. start_block += endcount;
  1080. j = 0;
  1081. }
  1082. xlog_put_bp(bp);
  1083. return error;
  1084. }
  1085. /*
  1086. * This routine is called to blow away any incomplete log writes out
  1087. * in front of the log head. We do this so that we won't become confused
  1088. * if we come up, write only a little bit more, and then crash again.
  1089. * If we leave the partial log records out there, this situation could
  1090. * cause us to think those partial writes are valid blocks since they
  1091. * have the current cycle number. We get rid of them by overwriting them
  1092. * with empty log records with the old cycle number rather than the
  1093. * current one.
  1094. *
  1095. * The tail lsn is passed in rather than taken from
  1096. * the log so that we will not write over the unmount record after a
  1097. * clean unmount in a 512 block log. Doing so would leave the log without
  1098. * any valid log records in it until a new one was written. If we crashed
  1099. * during that time we would not be able to recover.
  1100. */
  1101. STATIC int
  1102. xlog_clear_stale_blocks(
  1103. xlog_t *log,
  1104. xfs_lsn_t tail_lsn)
  1105. {
  1106. int tail_cycle, head_cycle;
  1107. int tail_block, head_block;
  1108. int tail_distance, max_distance;
  1109. int distance;
  1110. int error;
  1111. tail_cycle = CYCLE_LSN(tail_lsn);
  1112. tail_block = BLOCK_LSN(tail_lsn);
  1113. head_cycle = log->l_curr_cycle;
  1114. head_block = log->l_curr_block;
  1115. /*
  1116. * Figure out the distance between the new head of the log
  1117. * and the tail. We want to write over any blocks beyond the
  1118. * head that we may have written just before the crash, but
  1119. * we don't want to overwrite the tail of the log.
  1120. */
  1121. if (head_cycle == tail_cycle) {
  1122. /*
  1123. * The tail is behind the head in the physical log,
  1124. * so the distance from the head to the tail is the
  1125. * distance from the head to the end of the log plus
  1126. * the distance from the beginning of the log to the
  1127. * tail.
  1128. */
  1129. if (unlikely(head_block < tail_block || head_block >= log->l_logBBsize)) {
  1130. XFS_ERROR_REPORT("xlog_clear_stale_blocks(1)",
  1131. XFS_ERRLEVEL_LOW, log->l_mp);
  1132. return XFS_ERROR(EFSCORRUPTED);
  1133. }
  1134. tail_distance = tail_block + (log->l_logBBsize - head_block);
  1135. } else {
  1136. /*
  1137. * The head is behind the tail in the physical log,
  1138. * so the distance from the head to the tail is just
  1139. * the tail block minus the head block.
  1140. */
  1141. if (unlikely(head_block >= tail_block || head_cycle != (tail_cycle + 1))){
  1142. XFS_ERROR_REPORT("xlog_clear_stale_blocks(2)",
  1143. XFS_ERRLEVEL_LOW, log->l_mp);
  1144. return XFS_ERROR(EFSCORRUPTED);
  1145. }
  1146. tail_distance = tail_block - head_block;
  1147. }
  1148. /*
  1149. * If the head is right up against the tail, we can't clear
  1150. * anything.
  1151. */
  1152. if (tail_distance <= 0) {
  1153. ASSERT(tail_distance == 0);
  1154. return 0;
  1155. }
  1156. max_distance = XLOG_TOTAL_REC_SHIFT(log);
  1157. /*
  1158. * Take the smaller of the maximum amount of outstanding I/O
  1159. * we could have and the distance to the tail to clear out.
  1160. * We take the smaller so that we don't overwrite the tail and
  1161. * we don't waste all day writing from the head to the tail
  1162. * for no reason.
  1163. */
  1164. max_distance = MIN(max_distance, tail_distance);
  1165. if ((head_block + max_distance) <= log->l_logBBsize) {
  1166. /*
  1167. * We can stomp all the blocks we need to without
  1168. * wrapping around the end of the log. Just do it
  1169. * in a single write. Use the cycle number of the
  1170. * current cycle minus one so that the log will look like:
  1171. * n ... | n - 1 ...
  1172. */
  1173. error = xlog_write_log_records(log, (head_cycle - 1),
  1174. head_block, max_distance, tail_cycle,
  1175. tail_block);
  1176. if (error)
  1177. return error;
  1178. } else {
  1179. /*
  1180. * We need to wrap around the end of the physical log in
  1181. * order to clear all the blocks. Do it in two separate
  1182. * I/Os. The first write should be from the head to the
  1183. * end of the physical log, and it should use the current
  1184. * cycle number minus one just like above.
  1185. */
  1186. distance = log->l_logBBsize - head_block;
  1187. error = xlog_write_log_records(log, (head_cycle - 1),
  1188. head_block, distance, tail_cycle,
  1189. tail_block);
  1190. if (error)
  1191. return error;
  1192. /*
  1193. * Now write the blocks at the start of the physical log.
  1194. * This writes the remainder of the blocks we want to clear.
  1195. * It uses the current cycle number since we're now on the
  1196. * same cycle as the head so that we get:
  1197. * n ... n ... | n - 1 ...
  1198. * ^^^^^ blocks we're writing
  1199. */
  1200. distance = max_distance - (log->l_logBBsize - head_block);
  1201. error = xlog_write_log_records(log, head_cycle, 0, distance,
  1202. tail_cycle, tail_block);
  1203. if (error)
  1204. return error;
  1205. }
  1206. return 0;
  1207. }
  1208. /******************************************************************************
  1209. *
  1210. * Log recover routines
  1211. *
  1212. ******************************************************************************
  1213. */
  1214. STATIC xlog_recover_t *
  1215. xlog_recover_find_tid(
  1216. xlog_recover_t *q,
  1217. xlog_tid_t tid)
  1218. {
  1219. xlog_recover_t *p = q;
  1220. while (p != NULL) {
  1221. if (p->r_log_tid == tid)
  1222. break;
  1223. p = p->r_next;
  1224. }
  1225. return p;
  1226. }
  1227. STATIC void
  1228. xlog_recover_put_hashq(
  1229. xlog_recover_t **q,
  1230. xlog_recover_t *trans)
  1231. {
  1232. trans->r_next = *q;
  1233. *q = trans;
  1234. }
  1235. STATIC void
  1236. xlog_recover_add_item(
  1237. xlog_recover_item_t **itemq)
  1238. {
  1239. xlog_recover_item_t *item;
  1240. item = kmem_zalloc(sizeof(xlog_recover_item_t), KM_SLEEP);
  1241. xlog_recover_insert_item_backq(itemq, item);
  1242. }
  1243. STATIC int
  1244. xlog_recover_add_to_cont_trans(
  1245. xlog_recover_t *trans,
  1246. xfs_caddr_t dp,
  1247. int len)
  1248. {
  1249. xlog_recover_item_t *item;
  1250. xfs_caddr_t ptr, old_ptr;
  1251. int old_len;
  1252. item = trans->r_itemq;
  1253. if (item == NULL) {
  1254. /* finish copying rest of trans header */
  1255. xlog_recover_add_item(&trans->r_itemq);
  1256. ptr = (xfs_caddr_t) &trans->r_theader +
  1257. sizeof(xfs_trans_header_t) - len;
  1258. memcpy(ptr, dp, len); /* d, s, l */
  1259. return 0;
  1260. }
  1261. item = item->ri_prev;
  1262. old_ptr = item->ri_buf[item->ri_cnt-1].i_addr;
  1263. old_len = item->ri_buf[item->ri_cnt-1].i_len;
  1264. ptr = kmem_realloc(old_ptr, len+old_len, old_len, 0u);
  1265. memcpy(&ptr[old_len], dp, len); /* d, s, l */
  1266. item->ri_buf[item->ri_cnt-1].i_len += len;
  1267. item->ri_buf[item->ri_cnt-1].i_addr = ptr;
  1268. return 0;
  1269. }
  1270. /*
  1271. * The next region to add is the start of a new region. It could be
  1272. * a whole region or it could be the first part of a new region. Because
  1273. * of this, the assumption here is that the type and size fields of all
  1274. * format structures fit into the first 32 bits of the structure.
  1275. *
  1276. * This works because all regions must be 32 bit aligned. Therefore, we
  1277. * either have both fields or we have neither field. In the case we have
  1278. * neither field, the data part of the region is zero length. We only have
  1279. * a log_op_header and can throw away the header since a new one will appear
  1280. * later. If we have at least 4 bytes, then we can determine how many regions
  1281. * will appear in the current log item.
  1282. */
  1283. STATIC int
  1284. xlog_recover_add_to_trans(
  1285. xlog_recover_t *trans,
  1286. xfs_caddr_t dp,
  1287. int len)
  1288. {
  1289. xfs_inode_log_format_t *in_f; /* any will do */
  1290. xlog_recover_item_t *item;
  1291. xfs_caddr_t ptr;
  1292. if (!len)
  1293. return 0;
  1294. item = trans->r_itemq;
  1295. if (item == NULL) {
  1296. ASSERT(*(uint *)dp == XFS_TRANS_HEADER_MAGIC);
  1297. if (len == sizeof(xfs_trans_header_t))
  1298. xlog_recover_add_item(&trans->r_itemq);
  1299. memcpy(&trans->r_theader, dp, len); /* d, s, l */
  1300. return 0;
  1301. }
  1302. ptr = kmem_alloc(len, KM_SLEEP);
  1303. memcpy(ptr, dp, len);
  1304. in_f = (xfs_inode_log_format_t *)ptr;
  1305. if (item->ri_prev->ri_total != 0 &&
  1306. item->ri_prev->ri_total == item->ri_prev->ri_cnt) {
  1307. xlog_recover_add_item(&trans->r_itemq);
  1308. }
  1309. item = trans->r_itemq;
  1310. item = item->ri_prev;
  1311. if (item->ri_total == 0) { /* first region to be added */
  1312. item->ri_total = in_f->ilf_size;
  1313. ASSERT(item->ri_total <= XLOG_MAX_REGIONS_IN_ITEM);
  1314. item->ri_buf = kmem_zalloc((item->ri_total *
  1315. sizeof(xfs_log_iovec_t)), KM_SLEEP);
  1316. }
  1317. ASSERT(item->ri_total > item->ri_cnt);
  1318. /* Description region is ri_buf[0] */
  1319. item->ri_buf[item->ri_cnt].i_addr = ptr;
  1320. item->ri_buf[item->ri_cnt].i_len = len;
  1321. item->ri_cnt++;
  1322. return 0;
  1323. }
  1324. STATIC void
  1325. xlog_recover_new_tid(
  1326. xlog_recover_t **q,
  1327. xlog_tid_t tid,
  1328. xfs_lsn_t lsn)
  1329. {
  1330. xlog_recover_t *trans;
  1331. trans = kmem_zalloc(sizeof(xlog_recover_t), KM_SLEEP);
  1332. trans->r_log_tid = tid;
  1333. trans->r_lsn = lsn;
  1334. xlog_recover_put_hashq(q, trans);
  1335. }
  1336. STATIC int
  1337. xlog_recover_unlink_tid(
  1338. xlog_recover_t **q,
  1339. xlog_recover_t *trans)
  1340. {
  1341. xlog_recover_t *tp;
  1342. int found = 0;
  1343. ASSERT(trans != NULL);
  1344. if (trans == *q) {
  1345. *q = (*q)->r_next;
  1346. } else {
  1347. tp = *q;
  1348. while (tp) {
  1349. if (tp->r_next == trans) {
  1350. found = 1;
  1351. break;
  1352. }
  1353. tp = tp->r_next;
  1354. }
  1355. if (!found) {
  1356. xlog_warn(
  1357. "XFS: xlog_recover_unlink_tid: trans not found");
  1358. ASSERT(0);
  1359. return XFS_ERROR(EIO);
  1360. }
  1361. tp->r_next = tp->r_next->r_next;
  1362. }
  1363. return 0;
  1364. }
  1365. STATIC void
  1366. xlog_recover_insert_item_backq(
  1367. xlog_recover_item_t **q,
  1368. xlog_recover_item_t *item)
  1369. {
  1370. if (*q == NULL) {
  1371. item->ri_prev = item->ri_next = item;
  1372. *q = item;
  1373. } else {
  1374. item->ri_next = *q;
  1375. item->ri_prev = (*q)->ri_prev;
  1376. (*q)->ri_prev = item;
  1377. item->ri_prev->ri_next = item;
  1378. }
  1379. }
  1380. STATIC void
  1381. xlog_recover_insert_item_frontq(
  1382. xlog_recover_item_t **q,
  1383. xlog_recover_item_t *item)
  1384. {
  1385. xlog_recover_insert_item_backq(q, item);
  1386. *q = item;
  1387. }
  1388. STATIC int
  1389. xlog_recover_reorder_trans(
  1390. xlog_recover_t *trans)
  1391. {
  1392. xlog_recover_item_t *first_item, *itemq, *itemq_next;
  1393. xfs_buf_log_format_t *buf_f;
  1394. ushort flags = 0;
  1395. first_item = itemq = trans->r_itemq;
  1396. trans->r_itemq = NULL;
  1397. do {
  1398. itemq_next = itemq->ri_next;
  1399. buf_f = (xfs_buf_log_format_t *)itemq->ri_buf[0].i_addr;
  1400. switch (ITEM_TYPE(itemq)) {
  1401. case XFS_LI_BUF:
  1402. flags = buf_f->blf_flags;
  1403. if (!(flags & XFS_BLI_CANCEL)) {
  1404. xlog_recover_insert_item_frontq(&trans->r_itemq,
  1405. itemq);
  1406. break;
  1407. }
  1408. case XFS_LI_INODE:
  1409. case XFS_LI_DQUOT:
  1410. case XFS_LI_QUOTAOFF:
  1411. case XFS_LI_EFD:
  1412. case XFS_LI_EFI:
  1413. xlog_recover_insert_item_backq(&trans->r_itemq, itemq);
  1414. break;
  1415. default:
  1416. xlog_warn(
  1417. "XFS: xlog_recover_reorder_trans: unrecognized type of log operation");
  1418. ASSERT(0);
  1419. return XFS_ERROR(EIO);
  1420. }
  1421. itemq = itemq_next;
  1422. } while (first_item != itemq);
  1423. return 0;
  1424. }
  1425. /*
  1426. * Build up the table of buf cancel records so that we don't replay
  1427. * cancelled data in the second pass. For buffer records that are
  1428. * not cancel records, there is nothing to do here so we just return.
  1429. *
  1430. * If we get a cancel record which is already in the table, this indicates
  1431. * that the buffer was cancelled multiple times. In order to ensure
  1432. * that during pass 2 we keep the record in the table until we reach its
  1433. * last occurrence in the log, we keep a reference count in the cancel
  1434. * record in the table to tell us how many times we expect to see this
  1435. * record during the second pass.
  1436. */
  1437. STATIC void
  1438. xlog_recover_do_buffer_pass1(
  1439. xlog_t *log,
  1440. xfs_buf_log_format_t *buf_f)
  1441. {
  1442. xfs_buf_cancel_t *bcp;
  1443. xfs_buf_cancel_t *nextp;
  1444. xfs_buf_cancel_t *prevp;
  1445. xfs_buf_cancel_t **bucket;
  1446. xfs_daddr_t blkno = 0;
  1447. uint len = 0;
  1448. ushort flags = 0;
  1449. switch (buf_f->blf_type) {
  1450. case XFS_LI_BUF:
  1451. blkno = buf_f->blf_blkno;
  1452. len = buf_f->blf_len;
  1453. flags = buf_f->blf_flags;
  1454. break;
  1455. }
  1456. /*
  1457. * If this isn't a cancel buffer item, then just return.
  1458. */
  1459. if (!(flags & XFS_BLI_CANCEL))
  1460. return;
  1461. /*
  1462. * Insert an xfs_buf_cancel record into the hash table of
  1463. * them. If there is already an identical record, bump
  1464. * its reference count.
  1465. */
  1466. bucket = &log->l_buf_cancel_table[(__uint64_t)blkno %
  1467. XLOG_BC_TABLE_SIZE];
  1468. /*
  1469. * If the hash bucket is empty then just insert a new record into
  1470. * the bucket.
  1471. */
  1472. if (*bucket == NULL) {
  1473. bcp = (xfs_buf_cancel_t *)kmem_alloc(sizeof(xfs_buf_cancel_t),
  1474. KM_SLEEP);
  1475. bcp->bc_blkno = blkno;
  1476. bcp->bc_len = len;
  1477. bcp->bc_refcount = 1;
  1478. bcp->bc_next = NULL;
  1479. *bucket = bcp;
  1480. return;
  1481. }
  1482. /*
  1483. * The hash bucket is not empty, so search for duplicates of our
  1484. * record. If we find one them just bump its refcount. If not
  1485. * then add us at the end of the list.
  1486. */
  1487. prevp = NULL;
  1488. nextp = *bucket;
  1489. while (nextp != NULL) {
  1490. if (nextp->bc_blkno == blkno && nextp->bc_len == len) {
  1491. nextp->bc_refcount++;
  1492. return;
  1493. }
  1494. prevp = nextp;
  1495. nextp = nextp->bc_next;
  1496. }
  1497. ASSERT(prevp != NULL);
  1498. bcp = (xfs_buf_cancel_t *)kmem_alloc(sizeof(xfs_buf_cancel_t),
  1499. KM_SLEEP);
  1500. bcp->bc_blkno = blkno;
  1501. bcp->bc_len = len;
  1502. bcp->bc_refcount = 1;
  1503. bcp->bc_next = NULL;
  1504. prevp->bc_next = bcp;
  1505. }
  1506. /*
  1507. * Check to see whether the buffer being recovered has a corresponding
  1508. * entry in the buffer cancel record table. If it does then return 1
  1509. * so that it will be cancelled, otherwise return 0. If the buffer is
  1510. * actually a buffer cancel item (XFS_BLI_CANCEL is set), then decrement
  1511. * the refcount on the entry in the table and remove it from the table
  1512. * if this is the last reference.
  1513. *
  1514. * We remove the cancel record from the table when we encounter its
  1515. * last occurrence in the log so that if the same buffer is re-used
  1516. * again after its last cancellation we actually replay the changes
  1517. * made at that point.
  1518. */
  1519. STATIC int
  1520. xlog_check_buffer_cancelled(
  1521. xlog_t *log,
  1522. xfs_daddr_t blkno,
  1523. uint len,
  1524. ushort flags)
  1525. {
  1526. xfs_buf_cancel_t *bcp;
  1527. xfs_buf_cancel_t *prevp;
  1528. xfs_buf_cancel_t **bucket;
  1529. if (log->l_buf_cancel_table == NULL) {
  1530. /*
  1531. * There is nothing in the table built in pass one,
  1532. * so this buffer must not be cancelled.
  1533. */
  1534. ASSERT(!(flags & XFS_BLI_CANCEL));
  1535. return 0;
  1536. }
  1537. bucket = &log->l_buf_cancel_table[(__uint64_t)blkno %
  1538. XLOG_BC_TABLE_SIZE];
  1539. bcp = *bucket;
  1540. if (bcp == NULL) {
  1541. /*
  1542. * There is no corresponding entry in the table built
  1543. * in pass one, so this buffer has not been cancelled.
  1544. */
  1545. ASSERT(!(flags & XFS_BLI_CANCEL));
  1546. return 0;
  1547. }
  1548. /*
  1549. * Search for an entry in the buffer cancel table that
  1550. * matches our buffer.
  1551. */
  1552. prevp = NULL;
  1553. while (bcp != NULL) {
  1554. if (bcp->bc_blkno == blkno && bcp->bc_len == len) {
  1555. /*
  1556. * We've go a match, so return 1 so that the
  1557. * recovery of this buffer is cancelled.
  1558. * If this buffer is actually a buffer cancel
  1559. * log item, then decrement the refcount on the
  1560. * one in the table and remove it if this is the
  1561. * last reference.
  1562. */
  1563. if (flags & XFS_BLI_CANCEL) {
  1564. bcp->bc_refcount--;
  1565. if (bcp->bc_refcount == 0) {
  1566. if (prevp == NULL) {
  1567. *bucket = bcp->bc_next;
  1568. } else {
  1569. prevp->bc_next = bcp->bc_next;
  1570. }
  1571. kmem_free(bcp);
  1572. }
  1573. }
  1574. return 1;
  1575. }
  1576. prevp = bcp;
  1577. bcp = bcp->bc_next;
  1578. }
  1579. /*
  1580. * We didn't find a corresponding entry in the table, so
  1581. * return 0 so that the buffer is NOT cancelled.
  1582. */
  1583. ASSERT(!(flags & XFS_BLI_CANCEL));
  1584. return 0;
  1585. }
  1586. STATIC int
  1587. xlog_recover_do_buffer_pass2(
  1588. xlog_t *log,
  1589. xfs_buf_log_format_t *buf_f)
  1590. {
  1591. xfs_daddr_t blkno = 0;
  1592. ushort flags = 0;
  1593. uint len = 0;
  1594. switch (buf_f->blf_type) {
  1595. case XFS_LI_BUF:
  1596. blkno = buf_f->blf_blkno;
  1597. flags = buf_f->blf_flags;
  1598. len = buf_f->blf_len;
  1599. break;
  1600. }
  1601. return xlog_check_buffer_cancelled(log, blkno, len, flags);
  1602. }
  1603. /*
  1604. * Perform recovery for a buffer full of inodes. In these buffers,
  1605. * the only data which should be recovered is that which corresponds
  1606. * to the di_next_unlinked pointers in the on disk inode structures.
  1607. * The rest of the data for the inodes is always logged through the
  1608. * inodes themselves rather than the inode buffer and is recovered
  1609. * in xlog_recover_do_inode_trans().
  1610. *
  1611. * The only time when buffers full of inodes are fully recovered is
  1612. * when the buffer is full of newly allocated inodes. In this case
  1613. * the buffer will not be marked as an inode buffer and so will be
  1614. * sent to xlog_recover_do_reg_buffer() below during recovery.
  1615. */
  1616. STATIC int
  1617. xlog_recover_do_inode_buffer(
  1618. xfs_mount_t *mp,
  1619. xlog_recover_item_t *item,
  1620. xfs_buf_t *bp,
  1621. xfs_buf_log_format_t *buf_f)
  1622. {
  1623. int i;
  1624. int item_index;
  1625. int bit;
  1626. int nbits;
  1627. int reg_buf_offset;
  1628. int reg_buf_bytes;
  1629. int next_unlinked_offset;
  1630. int inodes_per_buf;
  1631. xfs_agino_t *logged_nextp;
  1632. xfs_agino_t *buffer_nextp;
  1633. unsigned int *data_map = NULL;
  1634. unsigned int map_size = 0;
  1635. switch (buf_f->blf_type) {
  1636. case XFS_LI_BUF:
  1637. data_map = buf_f->blf_data_map;
  1638. map_size = buf_f->blf_map_size;
  1639. break;
  1640. }
  1641. /*
  1642. * Set the variables corresponding to the current region to
  1643. * 0 so that we'll initialize them on the first pass through
  1644. * the loop.
  1645. */
  1646. reg_buf_offset = 0;
  1647. reg_buf_bytes = 0;
  1648. bit = 0;
  1649. nbits = 0;
  1650. item_index = 0;
  1651. inodes_per_buf = XFS_BUF_COUNT(bp) >> mp->m_sb.sb_inodelog;
  1652. for (i = 0; i < inodes_per_buf; i++) {
  1653. next_unlinked_offset = (i * mp->m_sb.sb_inodesize) +
  1654. offsetof(xfs_dinode_t, di_next_unlinked);
  1655. while (next_unlinked_offset >=
  1656. (reg_buf_offset + reg_buf_bytes)) {
  1657. /*
  1658. * The next di_next_unlinked field is beyond
  1659. * the current logged region. Find the next
  1660. * logged region that contains or is beyond
  1661. * the current di_next_unlinked field.
  1662. */
  1663. bit += nbits;
  1664. bit = xfs_next_bit(data_map, map_size, bit);
  1665. /*
  1666. * If there are no more logged regions in the
  1667. * buffer, then we're done.
  1668. */
  1669. if (bit == -1) {
  1670. return 0;
  1671. }
  1672. nbits = xfs_contig_bits(data_map, map_size,
  1673. bit);
  1674. ASSERT(nbits > 0);
  1675. reg_buf_offset = bit << XFS_BLI_SHIFT;
  1676. reg_buf_bytes = nbits << XFS_BLI_SHIFT;
  1677. item_index++;
  1678. }
  1679. /*
  1680. * If the current logged region starts after the current
  1681. * di_next_unlinked field, then move on to the next
  1682. * di_next_unlinked field.
  1683. */
  1684. if (next_unlinked_offset < reg_buf_offset) {
  1685. continue;
  1686. }
  1687. ASSERT(item->ri_buf[item_index].i_addr != NULL);
  1688. ASSERT((item->ri_buf[item_index].i_len % XFS_BLI_CHUNK) == 0);
  1689. ASSERT((reg_buf_offset + reg_buf_bytes) <= XFS_BUF_COUNT(bp));
  1690. /*
  1691. * The current logged region contains a copy of the
  1692. * current di_next_unlinked field. Extract its value
  1693. * and copy it to the buffer copy.
  1694. */
  1695. logged_nextp = (xfs_agino_t *)
  1696. ((char *)(item->ri_buf[item_index].i_addr) +
  1697. (next_unlinked_offset - reg_buf_offset));
  1698. if (unlikely(*logged_nextp == 0)) {
  1699. xfs_fs_cmn_err(CE_ALERT, mp,
  1700. "bad inode buffer log record (ptr = 0x%p, bp = 0x%p). XFS trying to replay bad (0) inode di_next_unlinked field",
  1701. item, bp);
  1702. XFS_ERROR_REPORT("xlog_recover_do_inode_buf",
  1703. XFS_ERRLEVEL_LOW, mp);
  1704. return XFS_ERROR(EFSCORRUPTED);
  1705. }
  1706. buffer_nextp = (xfs_agino_t *)xfs_buf_offset(bp,
  1707. next_unlinked_offset);
  1708. *buffer_nextp = *logged_nextp;
  1709. }
  1710. return 0;
  1711. }
  1712. /*
  1713. * Perform a 'normal' buffer recovery. Each logged region of the
  1714. * buffer should be copied over the corresponding region in the
  1715. * given buffer. The bitmap in the buf log format structure indicates
  1716. * where to place the logged data.
  1717. */
  1718. /*ARGSUSED*/
  1719. STATIC void
  1720. xlog_recover_do_reg_buffer(
  1721. xlog_recover_item_t *item,
  1722. xfs_buf_t *bp,
  1723. xfs_buf_log_format_t *buf_f)
  1724. {
  1725. int i;
  1726. int bit;
  1727. int nbits;
  1728. unsigned int *data_map = NULL;
  1729. unsigned int map_size = 0;
  1730. int error;
  1731. switch (buf_f->blf_type) {
  1732. case XFS_LI_BUF:
  1733. data_map = buf_f->blf_data_map;
  1734. map_size = buf_f->blf_map_size;
  1735. break;
  1736. }
  1737. bit = 0;
  1738. i = 1; /* 0 is the buf format structure */
  1739. while (1) {
  1740. bit = xfs_next_bit(data_map, map_size, bit);
  1741. if (bit == -1)
  1742. break;
  1743. nbits = xfs_contig_bits(data_map, map_size, bit);
  1744. ASSERT(nbits > 0);
  1745. ASSERT(item->ri_buf[i].i_addr != NULL);
  1746. ASSERT(item->ri_buf[i].i_len % XFS_BLI_CHUNK == 0);
  1747. ASSERT(XFS_BUF_COUNT(bp) >=
  1748. ((uint)bit << XFS_BLI_SHIFT)+(nbits<<XFS_BLI_SHIFT));
  1749. /*
  1750. * Do a sanity check if this is a dquot buffer. Just checking
  1751. * the first dquot in the buffer should do. XXXThis is
  1752. * probably a good thing to do for other buf types also.
  1753. */
  1754. error = 0;
  1755. if (buf_f->blf_flags &
  1756. (XFS_BLI_UDQUOT_BUF|XFS_BLI_PDQUOT_BUF|XFS_BLI_GDQUOT_BUF)) {
  1757. error = xfs_qm_dqcheck((xfs_disk_dquot_t *)
  1758. item->ri_buf[i].i_addr,
  1759. -1, 0, XFS_QMOPT_DOWARN,
  1760. "dquot_buf_recover");
  1761. }
  1762. if (!error)
  1763. memcpy(xfs_buf_offset(bp,
  1764. (uint)bit << XFS_BLI_SHIFT), /* dest */
  1765. item->ri_buf[i].i_addr, /* source */
  1766. nbits<<XFS_BLI_SHIFT); /* length */
  1767. i++;
  1768. bit += nbits;
  1769. }
  1770. /* Shouldn't be any more regions */
  1771. ASSERT(i == item->ri_total);
  1772. }
  1773. /*
  1774. * Do some primitive error checking on ondisk dquot data structures.
  1775. */
  1776. int
  1777. xfs_qm_dqcheck(
  1778. xfs_disk_dquot_t *ddq,
  1779. xfs_dqid_t id,
  1780. uint type, /* used only when IO_dorepair is true */
  1781. uint flags,
  1782. char *str)
  1783. {
  1784. xfs_dqblk_t *d = (xfs_dqblk_t *)ddq;
  1785. int errs = 0;
  1786. /*
  1787. * We can encounter an uninitialized dquot buffer for 2 reasons:
  1788. * 1. If we crash while deleting the quotainode(s), and those blks got
  1789. * used for user data. This is because we take the path of regular
  1790. * file deletion; however, the size field of quotainodes is never
  1791. * updated, so all the tricks that we play in itruncate_finish
  1792. * don't quite matter.
  1793. *
  1794. * 2. We don't play the quota buffers when there's a quotaoff logitem.
  1795. * But the allocation will be replayed so we'll end up with an
  1796. * uninitialized quota block.
  1797. *
  1798. * This is all fine; things are still consistent, and we haven't lost
  1799. * any quota information. Just don't complain about bad dquot blks.
  1800. */
  1801. if (be16_to_cpu(ddq->d_magic) != XFS_DQUOT_MAGIC) {
  1802. if (flags & XFS_QMOPT_DOWARN)
  1803. cmn_err(CE_ALERT,
  1804. "%s : XFS dquot ID 0x%x, magic 0x%x != 0x%x",
  1805. str, id, be16_to_cpu(ddq->d_magic), XFS_DQUOT_MAGIC);
  1806. errs++;
  1807. }
  1808. if (ddq->d_version != XFS_DQUOT_VERSION) {
  1809. if (flags & XFS_QMOPT_DOWARN)
  1810. cmn_err(CE_ALERT,
  1811. "%s : XFS dquot ID 0x%x, version 0x%x != 0x%x",
  1812. str, id, ddq->d_version, XFS_DQUOT_VERSION);
  1813. errs++;
  1814. }
  1815. if (ddq->d_flags != XFS_DQ_USER &&
  1816. ddq->d_flags != XFS_DQ_PROJ &&
  1817. ddq->d_flags != XFS_DQ_GROUP) {
  1818. if (flags & XFS_QMOPT_DOWARN)
  1819. cmn_err(CE_ALERT,
  1820. "%s : XFS dquot ID 0x%x, unknown flags 0x%x",
  1821. str, id, ddq->d_flags);
  1822. errs++;
  1823. }
  1824. if (id != -1 && id != be32_to_cpu(ddq->d_id)) {
  1825. if (flags & XFS_QMOPT_DOWARN)
  1826. cmn_err(CE_ALERT,
  1827. "%s : ondisk-dquot 0x%p, ID mismatch: "
  1828. "0x%x expected, found id 0x%x",
  1829. str, ddq, id, be32_to_cpu(ddq->d_id));
  1830. errs++;
  1831. }
  1832. if (!errs && ddq->d_id) {
  1833. if (ddq->d_blk_softlimit &&
  1834. be64_to_cpu(ddq->d_bcount) >=
  1835. be64_to_cpu(ddq->d_blk_softlimit)) {
  1836. if (!ddq->d_btimer) {
  1837. if (flags & XFS_QMOPT_DOWARN)
  1838. cmn_err(CE_ALERT,
  1839. "%s : Dquot ID 0x%x (0x%p) "
  1840. "BLK TIMER NOT STARTED",
  1841. str, (int)be32_to_cpu(ddq->d_id), ddq);
  1842. errs++;
  1843. }
  1844. }
  1845. if (ddq->d_ino_softlimit &&
  1846. be64_to_cpu(ddq->d_icount) >=
  1847. be64_to_cpu(ddq->d_ino_softlimit)) {
  1848. if (!ddq->d_itimer) {
  1849. if (flags & XFS_QMOPT_DOWARN)
  1850. cmn_err(CE_ALERT,
  1851. "%s : Dquot ID 0x%x (0x%p) "
  1852. "INODE TIMER NOT STARTED",
  1853. str, (int)be32_to_cpu(ddq->d_id), ddq);
  1854. errs++;
  1855. }
  1856. }
  1857. if (ddq->d_rtb_softlimit &&
  1858. be64_to_cpu(ddq->d_rtbcount) >=
  1859. be64_to_cpu(ddq->d_rtb_softlimit)) {
  1860. if (!ddq->d_rtbtimer) {
  1861. if (flags & XFS_QMOPT_DOWARN)
  1862. cmn_err(CE_ALERT,
  1863. "%s : Dquot ID 0x%x (0x%p) "
  1864. "RTBLK TIMER NOT STARTED",
  1865. str, (int)be32_to_cpu(ddq->d_id), ddq);
  1866. errs++;
  1867. }
  1868. }
  1869. }
  1870. if (!errs || !(flags & XFS_QMOPT_DQREPAIR))
  1871. return errs;
  1872. if (flags & XFS_QMOPT_DOWARN)
  1873. cmn_err(CE_NOTE, "Re-initializing dquot ID 0x%x", id);
  1874. /*
  1875. * Typically, a repair is only requested by quotacheck.
  1876. */
  1877. ASSERT(id != -1);
  1878. ASSERT(flags & XFS_QMOPT_DQREPAIR);
  1879. memset(d, 0, sizeof(xfs_dqblk_t));
  1880. d->dd_diskdq.d_magic = cpu_to_be16(XFS_DQUOT_MAGIC);
  1881. d->dd_diskdq.d_version = XFS_DQUOT_VERSION;
  1882. d->dd_diskdq.d_flags = type;
  1883. d->dd_diskdq.d_id = cpu_to_be32(id);
  1884. return errs;
  1885. }
  1886. /*
  1887. * Perform a dquot buffer recovery.
  1888. * Simple algorithm: if we have found a QUOTAOFF logitem of the same type
  1889. * (ie. USR or GRP), then just toss this buffer away; don't recover it.
  1890. * Else, treat it as a regular buffer and do recovery.
  1891. */
  1892. STATIC void
  1893. xlog_recover_do_dquot_buffer(
  1894. xfs_mount_t *mp,
  1895. xlog_t *log,
  1896. xlog_recover_item_t *item,
  1897. xfs_buf_t *bp,
  1898. xfs_buf_log_format_t *buf_f)
  1899. {
  1900. uint type;
  1901. /*
  1902. * Filesystems are required to send in quota flags at mount time.
  1903. */
  1904. if (mp->m_qflags == 0) {
  1905. return;
  1906. }
  1907. type = 0;
  1908. if (buf_f->blf_flags & XFS_BLI_UDQUOT_BUF)
  1909. type |= XFS_DQ_USER;
  1910. if (buf_f->blf_flags & XFS_BLI_PDQUOT_BUF)
  1911. type |= XFS_DQ_PROJ;
  1912. if (buf_f->blf_flags & XFS_BLI_GDQUOT_BUF)
  1913. type |= XFS_DQ_GROUP;
  1914. /*
  1915. * This type of quotas was turned off, so ignore this buffer
  1916. */
  1917. if (log->l_quotaoffs_flag & type)
  1918. return;
  1919. xlog_recover_do_reg_buffer(item, bp, buf_f);
  1920. }
  1921. /*
  1922. * This routine replays a modification made to a buffer at runtime.
  1923. * There are actually two types of buffer, regular and inode, which
  1924. * are handled differently. Inode buffers are handled differently
  1925. * in that we only recover a specific set of data from them, namely
  1926. * the inode di_next_unlinked fields. This is because all other inode
  1927. * data is actually logged via inode records and any data we replay
  1928. * here which overlaps that may be stale.
  1929. *
  1930. * When meta-data buffers are freed at run time we log a buffer item
  1931. * with the XFS_BLI_CANCEL bit set to indicate that previous copies
  1932. * of the buffer in the log should not be replayed at recovery time.
  1933. * This is so that if the blocks covered by the buffer are reused for
  1934. * file data before we crash we don't end up replaying old, freed
  1935. * meta-data into a user's file.
  1936. *
  1937. * To handle the cancellation of buffer log items, we make two passes
  1938. * over the log during recovery. During the first we build a table of
  1939. * those buffers which have been cancelled, and during the second we
  1940. * only replay those buffers which do not have corresponding cancel
  1941. * records in the table. See xlog_recover_do_buffer_pass[1,2] above
  1942. * for more details on the implementation of the table of cancel records.
  1943. */
  1944. STATIC int
  1945. xlog_recover_do_buffer_trans(
  1946. xlog_t *log,
  1947. xlog_recover_item_t *item,
  1948. int pass)
  1949. {
  1950. xfs_buf_log_format_t *buf_f;
  1951. xfs_mount_t *mp;
  1952. xfs_buf_t *bp;
  1953. int error;
  1954. int cancel;
  1955. xfs_daddr_t blkno;
  1956. int len;
  1957. ushort flags;
  1958. buf_f = (xfs_buf_log_format_t *)item->ri_buf[0].i_addr;
  1959. if (pass == XLOG_RECOVER_PASS1) {
  1960. /*
  1961. * In this pass we're only looking for buf items
  1962. * with the XFS_BLI_CANCEL bit set.
  1963. */
  1964. xlog_recover_do_buffer_pass1(log, buf_f);
  1965. return 0;
  1966. } else {
  1967. /*
  1968. * In this pass we want to recover all the buffers
  1969. * which have not been cancelled and are not
  1970. * cancellation buffers themselves. The routine
  1971. * we call here will tell us whether or not to
  1972. * continue with the replay of this buffer.
  1973. */
  1974. cancel = xlog_recover_do_buffer_pass2(log, buf_f);
  1975. if (cancel) {
  1976. return 0;
  1977. }
  1978. }
  1979. switch (buf_f->blf_type) {
  1980. case XFS_LI_BUF:
  1981. blkno = buf_f->blf_blkno;
  1982. len = buf_f->blf_len;
  1983. flags = buf_f->blf_flags;
  1984. break;
  1985. default:
  1986. xfs_fs_cmn_err(CE_ALERT, log->l_mp,
  1987. "xfs_log_recover: unknown buffer type 0x%x, logdev %s",
  1988. buf_f->blf_type, log->l_mp->m_logname ?
  1989. log->l_mp->m_logname : "internal");
  1990. XFS_ERROR_REPORT("xlog_recover_do_buffer_trans",
  1991. XFS_ERRLEVEL_LOW, log->l_mp);
  1992. return XFS_ERROR(EFSCORRUPTED);
  1993. }
  1994. mp = log->l_mp;
  1995. if (flags & XFS_BLI_INODE_BUF) {
  1996. bp = xfs_buf_read_flags(mp->m_ddev_targp, blkno, len,
  1997. XFS_BUF_LOCK);
  1998. } else {
  1999. bp = xfs_buf_read(mp->m_ddev_targp, blkno, len, 0);
  2000. }
  2001. if (XFS_BUF_ISERROR(bp)) {
  2002. xfs_ioerror_alert("xlog_recover_do..(read#1)", log->l_mp,
  2003. bp, blkno);
  2004. error = XFS_BUF_GETERROR(bp);
  2005. xfs_buf_relse(bp);
  2006. return error;
  2007. }
  2008. error = 0;
  2009. if (flags & XFS_BLI_INODE_BUF) {
  2010. error = xlog_recover_do_inode_buffer(mp, item, bp, buf_f);
  2011. } else if (flags &
  2012. (XFS_BLI_UDQUOT_BUF|XFS_BLI_PDQUOT_BUF|XFS_BLI_GDQUOT_BUF)) {
  2013. xlog_recover_do_dquot_buffer(mp, log, item, bp, buf_f);
  2014. } else {
  2015. xlog_recover_do_reg_buffer(item, bp, buf_f);
  2016. }
  2017. if (error)
  2018. return XFS_ERROR(error);
  2019. /*
  2020. * Perform delayed write on the buffer. Asynchronous writes will be
  2021. * slower when taking into account all the buffers to be flushed.
  2022. *
  2023. * Also make sure that only inode buffers with good sizes stay in
  2024. * the buffer cache. The kernel moves inodes in buffers of 1 block
  2025. * or XFS_INODE_CLUSTER_SIZE bytes, whichever is bigger. The inode
  2026. * buffers in the log can be a different size if the log was generated
  2027. * by an older kernel using unclustered inode buffers or a newer kernel
  2028. * running with a different inode cluster size. Regardless, if the
  2029. * the inode buffer size isn't MAX(blocksize, XFS_INODE_CLUSTER_SIZE)
  2030. * for *our* value of XFS_INODE_CLUSTER_SIZE, then we need to keep
  2031. * the buffer out of the buffer cache so that the buffer won't
  2032. * overlap with future reads of those inodes.
  2033. */
  2034. if (XFS_DINODE_MAGIC ==
  2035. be16_to_cpu(*((__be16 *)xfs_buf_offset(bp, 0))) &&
  2036. (XFS_BUF_COUNT(bp) != MAX(log->l_mp->m_sb.sb_blocksize,
  2037. (__uint32_t)XFS_INODE_CLUSTER_SIZE(log->l_mp)))) {
  2038. XFS_BUF_STALE(bp);
  2039. error = xfs_bwrite(mp, bp);
  2040. } else {
  2041. ASSERT(XFS_BUF_FSPRIVATE(bp, void *) == NULL ||
  2042. XFS_BUF_FSPRIVATE(bp, xfs_mount_t *) == mp);
  2043. XFS_BUF_SET_FSPRIVATE(bp, mp);
  2044. XFS_BUF_SET_IODONE_FUNC(bp, xlog_recover_iodone);
  2045. xfs_bdwrite(mp, bp);
  2046. }
  2047. return (error);
  2048. }
  2049. STATIC int
  2050. xlog_recover_do_inode_trans(
  2051. xlog_t *log,
  2052. xlog_recover_item_t *item,
  2053. int pass)
  2054. {
  2055. xfs_inode_log_format_t *in_f;
  2056. xfs_mount_t *mp;
  2057. xfs_buf_t *bp;
  2058. xfs_imap_t imap;
  2059. xfs_dinode_t *dip;
  2060. xfs_ino_t ino;
  2061. int len;
  2062. xfs_caddr_t src;
  2063. xfs_caddr_t dest;
  2064. int error;
  2065. int attr_index;
  2066. uint fields;
  2067. xfs_icdinode_t *dicp;
  2068. int need_free = 0;
  2069. if (pass == XLOG_RECOVER_PASS1) {
  2070. return 0;
  2071. }
  2072. if (item->ri_buf[0].i_len == sizeof(xfs_inode_log_format_t)) {
  2073. in_f = (xfs_inode_log_format_t *)item->ri_buf[0].i_addr;
  2074. } else {
  2075. in_f = (xfs_inode_log_format_t *)kmem_alloc(
  2076. sizeof(xfs_inode_log_format_t), KM_SLEEP);
  2077. need_free = 1;
  2078. error = xfs_inode_item_format_convert(&item->ri_buf[0], in_f);
  2079. if (error)
  2080. goto error;
  2081. }
  2082. ino = in_f->ilf_ino;
  2083. mp = log->l_mp;
  2084. if (ITEM_TYPE(item) == XFS_LI_INODE) {
  2085. imap.im_blkno = (xfs_daddr_t)in_f->ilf_blkno;
  2086. imap.im_len = in_f->ilf_len;
  2087. imap.im_boffset = in_f->ilf_boffset;
  2088. } else {
  2089. /*
  2090. * It's an old inode format record. We don't know where
  2091. * its cluster is located on disk, and we can't allow
  2092. * xfs_imap() to figure it out because the inode btrees
  2093. * are not ready to be used. Therefore do not pass the
  2094. * XFS_IMAP_LOOKUP flag to xfs_imap(). This will give
  2095. * us only the single block in which the inode lives
  2096. * rather than its cluster, so we must make sure to
  2097. * invalidate the buffer when we write it out below.
  2098. */
  2099. imap.im_blkno = 0;
  2100. error = xfs_imap(log->l_mp, NULL, ino, &imap, 0);
  2101. if (error)
  2102. goto error;
  2103. }
  2104. /*
  2105. * Inode buffers can be freed, look out for it,
  2106. * and do not replay the inode.
  2107. */
  2108. if (xlog_check_buffer_cancelled(log, imap.im_blkno, imap.im_len, 0)) {
  2109. error = 0;
  2110. goto error;
  2111. }
  2112. bp = xfs_buf_read_flags(mp->m_ddev_targp, imap.im_blkno, imap.im_len,
  2113. XFS_BUF_LOCK);
  2114. if (XFS_BUF_ISERROR(bp)) {
  2115. xfs_ioerror_alert("xlog_recover_do..(read#2)", mp,
  2116. bp, imap.im_blkno);
  2117. error = XFS_BUF_GETERROR(bp);
  2118. xfs_buf_relse(bp);
  2119. goto error;
  2120. }
  2121. error = 0;
  2122. ASSERT(in_f->ilf_fields & XFS_ILOG_CORE);
  2123. dip = (xfs_dinode_t *)xfs_buf_offset(bp, imap.im_boffset);
  2124. /*
  2125. * Make sure the place we're flushing out to really looks
  2126. * like an inode!
  2127. */
  2128. if (unlikely(be16_to_cpu(dip->di_core.di_magic) != XFS_DINODE_MAGIC)) {
  2129. xfs_buf_relse(bp);
  2130. xfs_fs_cmn_err(CE_ALERT, mp,
  2131. "xfs_inode_recover: Bad inode magic number, dino ptr = 0x%p, dino bp = 0x%p, ino = %Ld",
  2132. dip, bp, ino);
  2133. XFS_ERROR_REPORT("xlog_recover_do_inode_trans(1)",
  2134. XFS_ERRLEVEL_LOW, mp);
  2135. error = EFSCORRUPTED;
  2136. goto error;
  2137. }
  2138. dicp = (xfs_icdinode_t *)(item->ri_buf[1].i_addr);
  2139. if (unlikely(dicp->di_magic != XFS_DINODE_MAGIC)) {
  2140. xfs_buf_relse(bp);
  2141. xfs_fs_cmn_err(CE_ALERT, mp,
  2142. "xfs_inode_recover: Bad inode log record, rec ptr 0x%p, ino %Ld",
  2143. item, ino);
  2144. XFS_ERROR_REPORT("xlog_recover_do_inode_trans(2)",
  2145. XFS_ERRLEVEL_LOW, mp);
  2146. error = EFSCORRUPTED;
  2147. goto error;
  2148. }
  2149. /* Skip replay when the on disk inode is newer than the log one */
  2150. if (dicp->di_flushiter < be16_to_cpu(dip->di_core.di_flushiter)) {
  2151. /*
  2152. * Deal with the wrap case, DI_MAX_FLUSH is less
  2153. * than smaller numbers
  2154. */
  2155. if (be16_to_cpu(dip->di_core.di_flushiter) == DI_MAX_FLUSH &&
  2156. dicp->di_flushiter < (DI_MAX_FLUSH >> 1)) {
  2157. /* do nothing */
  2158. } else {
  2159. xfs_buf_relse(bp);
  2160. error = 0;
  2161. goto error;
  2162. }
  2163. }
  2164. /* Take the opportunity to reset the flush iteration count */
  2165. dicp->di_flushiter = 0;
  2166. if (unlikely((dicp->di_mode & S_IFMT) == S_IFREG)) {
  2167. if ((dicp->di_format != XFS_DINODE_FMT_EXTENTS) &&
  2168. (dicp->di_format != XFS_DINODE_FMT_BTREE)) {
  2169. XFS_CORRUPTION_ERROR("xlog_recover_do_inode_trans(3)",
  2170. XFS_ERRLEVEL_LOW, mp, dicp);
  2171. xfs_buf_relse(bp);
  2172. xfs_fs_cmn_err(CE_ALERT, mp,
  2173. "xfs_inode_recover: Bad regular inode log record, rec ptr 0x%p, ino ptr = 0x%p, ino bp = 0x%p, ino %Ld",
  2174. item, dip, bp, ino);
  2175. error = EFSCORRUPTED;
  2176. goto error;
  2177. }
  2178. } else if (unlikely((dicp->di_mode & S_IFMT) == S_IFDIR)) {
  2179. if ((dicp->di_format != XFS_DINODE_FMT_EXTENTS) &&
  2180. (dicp->di_format != XFS_DINODE_FMT_BTREE) &&
  2181. (dicp->di_format != XFS_DINODE_FMT_LOCAL)) {
  2182. XFS_CORRUPTION_ERROR("xlog_recover_do_inode_trans(4)",
  2183. XFS_ERRLEVEL_LOW, mp, dicp);
  2184. xfs_buf_relse(bp);
  2185. xfs_fs_cmn_err(CE_ALERT, mp,
  2186. "xfs_inode_recover: Bad dir inode log record, rec ptr 0x%p, ino ptr = 0x%p, ino bp = 0x%p, ino %Ld",
  2187. item, dip, bp, ino);
  2188. error = EFSCORRUPTED;
  2189. goto error;
  2190. }
  2191. }
  2192. if (unlikely(dicp->di_nextents + dicp->di_anextents > dicp->di_nblocks)){
  2193. XFS_CORRUPTION_ERROR("xlog_recover_do_inode_trans(5)",
  2194. XFS_ERRLEVEL_LOW, mp, dicp);
  2195. xfs_buf_relse(bp);
  2196. xfs_fs_cmn_err(CE_ALERT, mp,
  2197. "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",
  2198. item, dip, bp, ino,
  2199. dicp->di_nextents + dicp->di_anextents,
  2200. dicp->di_nblocks);
  2201. error = EFSCORRUPTED;
  2202. goto error;
  2203. }
  2204. if (unlikely(dicp->di_forkoff > mp->m_sb.sb_inodesize)) {
  2205. XFS_CORRUPTION_ERROR("xlog_recover_do_inode_trans(6)",
  2206. XFS_ERRLEVEL_LOW, mp, dicp);
  2207. xfs_buf_relse(bp);
  2208. xfs_fs_cmn_err(CE_ALERT, mp,
  2209. "xfs_inode_recover: Bad inode log rec ptr 0x%p, dino ptr 0x%p, dino bp 0x%p, ino %Ld, forkoff 0x%x",
  2210. item, dip, bp, ino, dicp->di_forkoff);
  2211. error = EFSCORRUPTED;
  2212. goto error;
  2213. }
  2214. if (unlikely(item->ri_buf[1].i_len > sizeof(xfs_dinode_core_t))) {
  2215. XFS_CORRUPTION_ERROR("xlog_recover_do_inode_trans(7)",
  2216. XFS_ERRLEVEL_LOW, mp, dicp);
  2217. xfs_buf_relse(bp);
  2218. xfs_fs_cmn_err(CE_ALERT, mp,
  2219. "xfs_inode_recover: Bad inode log record length %d, rec ptr 0x%p",
  2220. item->ri_buf[1].i_len, item);
  2221. error = EFSCORRUPTED;
  2222. goto error;
  2223. }
  2224. /* The core is in in-core format */
  2225. xfs_dinode_to_disk(&dip->di_core,
  2226. (xfs_icdinode_t *)item->ri_buf[1].i_addr);
  2227. /* the rest is in on-disk format */
  2228. if (item->ri_buf[1].i_len > sizeof(xfs_dinode_core_t)) {
  2229. memcpy((xfs_caddr_t) dip + sizeof(xfs_dinode_core_t),
  2230. item->ri_buf[1].i_addr + sizeof(xfs_dinode_core_t),
  2231. item->ri_buf[1].i_len - sizeof(xfs_dinode_core_t));
  2232. }
  2233. fields = in_f->ilf_fields;
  2234. switch (fields & (XFS_ILOG_DEV | XFS_ILOG_UUID)) {
  2235. case XFS_ILOG_DEV:
  2236. dip->di_u.di_dev = cpu_to_be32(in_f->ilf_u.ilfu_rdev);
  2237. break;
  2238. case XFS_ILOG_UUID:
  2239. dip->di_u.di_muuid = in_f->ilf_u.ilfu_uuid;
  2240. break;
  2241. }
  2242. if (in_f->ilf_size == 2)
  2243. goto write_inode_buffer;
  2244. len = item->ri_buf[2].i_len;
  2245. src = item->ri_buf[2].i_addr;
  2246. ASSERT(in_f->ilf_size <= 4);
  2247. ASSERT((in_f->ilf_size == 3) || (fields & XFS_ILOG_AFORK));
  2248. ASSERT(!(fields & XFS_ILOG_DFORK) ||
  2249. (len == in_f->ilf_dsize));
  2250. switch (fields & XFS_ILOG_DFORK) {
  2251. case XFS_ILOG_DDATA:
  2252. case XFS_ILOG_DEXT:
  2253. memcpy(&dip->di_u, src, len);
  2254. break;
  2255. case XFS_ILOG_DBROOT:
  2256. xfs_bmbt_to_bmdr(mp, (struct xfs_btree_block *)src, len,
  2257. &dip->di_u.di_bmbt,
  2258. XFS_DFORK_DSIZE(dip, mp));
  2259. break;
  2260. default:
  2261. /*
  2262. * There are no data fork flags set.
  2263. */
  2264. ASSERT((fields & XFS_ILOG_DFORK) == 0);
  2265. break;
  2266. }
  2267. /*
  2268. * If we logged any attribute data, recover it. There may or
  2269. * may not have been any other non-core data logged in this
  2270. * transaction.
  2271. */
  2272. if (in_f->ilf_fields & XFS_ILOG_AFORK) {
  2273. if (in_f->ilf_fields & XFS_ILOG_DFORK) {
  2274. attr_index = 3;
  2275. } else {
  2276. attr_index = 2;
  2277. }
  2278. len = item->ri_buf[attr_index].i_len;
  2279. src = item->ri_buf[attr_index].i_addr;
  2280. ASSERT(len == in_f->ilf_asize);
  2281. switch (in_f->ilf_fields & XFS_ILOG_AFORK) {
  2282. case XFS_ILOG_ADATA:
  2283. case XFS_ILOG_AEXT:
  2284. dest = XFS_DFORK_APTR(dip);
  2285. ASSERT(len <= XFS_DFORK_ASIZE(dip, mp));
  2286. memcpy(dest, src, len);
  2287. break;
  2288. case XFS_ILOG_ABROOT:
  2289. dest = XFS_DFORK_APTR(dip);
  2290. xfs_bmbt_to_bmdr(mp, (struct xfs_btree_block *)src,
  2291. len, (xfs_bmdr_block_t*)dest,
  2292. XFS_DFORK_ASIZE(dip, mp));
  2293. break;
  2294. default:
  2295. xlog_warn("XFS: xlog_recover_do_inode_trans: Invalid flag");
  2296. ASSERT(0);
  2297. xfs_buf_relse(bp);
  2298. error = EIO;
  2299. goto error;
  2300. }
  2301. }
  2302. write_inode_buffer:
  2303. if (ITEM_TYPE(item) == XFS_LI_INODE) {
  2304. ASSERT(XFS_BUF_FSPRIVATE(bp, void *) == NULL ||
  2305. XFS_BUF_FSPRIVATE(bp, xfs_mount_t *) == mp);
  2306. XFS_BUF_SET_FSPRIVATE(bp, mp);
  2307. XFS_BUF_SET_IODONE_FUNC(bp, xlog_recover_iodone);
  2308. xfs_bdwrite(mp, bp);
  2309. } else {
  2310. XFS_BUF_STALE(bp);
  2311. error = xfs_bwrite(mp, bp);
  2312. }
  2313. error:
  2314. if (need_free)
  2315. kmem_free(in_f);
  2316. return XFS_ERROR(error);
  2317. }
  2318. /*
  2319. * Recover QUOTAOFF records. We simply make a note of it in the xlog_t
  2320. * structure, so that we know not to do any dquot item or dquot buffer recovery,
  2321. * of that type.
  2322. */
  2323. STATIC int
  2324. xlog_recover_do_quotaoff_trans(
  2325. xlog_t *log,
  2326. xlog_recover_item_t *item,
  2327. int pass)
  2328. {
  2329. xfs_qoff_logformat_t *qoff_f;
  2330. if (pass == XLOG_RECOVER_PASS2) {
  2331. return (0);
  2332. }
  2333. qoff_f = (xfs_qoff_logformat_t *)item->ri_buf[0].i_addr;
  2334. ASSERT(qoff_f);
  2335. /*
  2336. * The logitem format's flag tells us if this was user quotaoff,
  2337. * group/project quotaoff or both.
  2338. */
  2339. if (qoff_f->qf_flags & XFS_UQUOTA_ACCT)
  2340. log->l_quotaoffs_flag |= XFS_DQ_USER;
  2341. if (qoff_f->qf_flags & XFS_PQUOTA_ACCT)
  2342. log->l_quotaoffs_flag |= XFS_DQ_PROJ;
  2343. if (qoff_f->qf_flags & XFS_GQUOTA_ACCT)
  2344. log->l_quotaoffs_flag |= XFS_DQ_GROUP;
  2345. return (0);
  2346. }
  2347. /*
  2348. * Recover a dquot record
  2349. */
  2350. STATIC int
  2351. xlog_recover_do_dquot_trans(
  2352. xlog_t *log,
  2353. xlog_recover_item_t *item,
  2354. int pass)
  2355. {
  2356. xfs_mount_t *mp;
  2357. xfs_buf_t *bp;
  2358. struct xfs_disk_dquot *ddq, *recddq;
  2359. int error;
  2360. xfs_dq_logformat_t *dq_f;
  2361. uint type;
  2362. if (pass == XLOG_RECOVER_PASS1) {
  2363. return 0;
  2364. }
  2365. mp = log->l_mp;
  2366. /*
  2367. * Filesystems are required to send in quota flags at mount time.
  2368. */
  2369. if (mp->m_qflags == 0)
  2370. return (0);
  2371. recddq = (xfs_disk_dquot_t *)item->ri_buf[1].i_addr;
  2372. ASSERT(recddq);
  2373. /*
  2374. * This type of quotas was turned off, so ignore this record.
  2375. */
  2376. type = recddq->d_flags & (XFS_DQ_USER | XFS_DQ_PROJ | XFS_DQ_GROUP);
  2377. ASSERT(type);
  2378. if (log->l_quotaoffs_flag & type)
  2379. return (0);
  2380. /*
  2381. * At this point we know that quota was _not_ turned off.
  2382. * Since the mount flags are not indicating to us otherwise, this
  2383. * must mean that quota is on, and the dquot needs to be replayed.
  2384. * Remember that we may not have fully recovered the superblock yet,
  2385. * so we can't do the usual trick of looking at the SB quota bits.
  2386. *
  2387. * The other possibility, of course, is that the quota subsystem was
  2388. * removed since the last mount - ENOSYS.
  2389. */
  2390. dq_f = (xfs_dq_logformat_t *)item->ri_buf[0].i_addr;
  2391. ASSERT(dq_f);
  2392. if ((error = xfs_qm_dqcheck(recddq,
  2393. dq_f->qlf_id,
  2394. 0, XFS_QMOPT_DOWARN,
  2395. "xlog_recover_do_dquot_trans (log copy)"))) {
  2396. return XFS_ERROR(EIO);
  2397. }
  2398. ASSERT(dq_f->qlf_len == 1);
  2399. error = xfs_read_buf(mp, mp->m_ddev_targp,
  2400. dq_f->qlf_blkno,
  2401. XFS_FSB_TO_BB(mp, dq_f->qlf_len),
  2402. 0, &bp);
  2403. if (error) {
  2404. xfs_ioerror_alert("xlog_recover_do..(read#3)", mp,
  2405. bp, dq_f->qlf_blkno);
  2406. return error;
  2407. }
  2408. ASSERT(bp);
  2409. ddq = (xfs_disk_dquot_t *)xfs_buf_offset(bp, dq_f->qlf_boffset);
  2410. /*
  2411. * At least the magic num portion should be on disk because this
  2412. * was among a chunk of dquots created earlier, and we did some
  2413. * minimal initialization then.
  2414. */
  2415. if (xfs_qm_dqcheck(ddq, dq_f->qlf_id, 0, XFS_QMOPT_DOWARN,
  2416. "xlog_recover_do_dquot_trans")) {
  2417. xfs_buf_relse(bp);
  2418. return XFS_ERROR(EIO);
  2419. }
  2420. memcpy(ddq, recddq, item->ri_buf[1].i_len);
  2421. ASSERT(dq_f->qlf_size == 2);
  2422. ASSERT(XFS_BUF_FSPRIVATE(bp, void *) == NULL ||
  2423. XFS_BUF_FSPRIVATE(bp, xfs_mount_t *) == mp);
  2424. XFS_BUF_SET_FSPRIVATE(bp, mp);
  2425. XFS_BUF_SET_IODONE_FUNC(bp, xlog_recover_iodone);
  2426. xfs_bdwrite(mp, bp);
  2427. return (0);
  2428. }
  2429. /*
  2430. * This routine is called to create an in-core extent free intent
  2431. * item from the efi format structure which was logged on disk.
  2432. * It allocates an in-core efi, copies the extents from the format
  2433. * structure into it, and adds the efi to the AIL with the given
  2434. * LSN.
  2435. */
  2436. STATIC int
  2437. xlog_recover_do_efi_trans(
  2438. xlog_t *log,
  2439. xlog_recover_item_t *item,
  2440. xfs_lsn_t lsn,
  2441. int pass)
  2442. {
  2443. int error;
  2444. xfs_mount_t *mp;
  2445. xfs_efi_log_item_t *efip;
  2446. xfs_efi_log_format_t *efi_formatp;
  2447. if (pass == XLOG_RECOVER_PASS1) {
  2448. return 0;
  2449. }
  2450. efi_formatp = (xfs_efi_log_format_t *)item->ri_buf[0].i_addr;
  2451. mp = log->l_mp;
  2452. efip = xfs_efi_init(mp, efi_formatp->efi_nextents);
  2453. if ((error = xfs_efi_copy_format(&(item->ri_buf[0]),
  2454. &(efip->efi_format)))) {
  2455. xfs_efi_item_free(efip);
  2456. return error;
  2457. }
  2458. efip->efi_next_extent = efi_formatp->efi_nextents;
  2459. efip->efi_flags |= XFS_EFI_COMMITTED;
  2460. spin_lock(&log->l_ailp->xa_lock);
  2461. /*
  2462. * xfs_trans_update_ail() drops the AIL lock.
  2463. */
  2464. xfs_trans_update_ail(mp, (xfs_log_item_t *)efip, lsn);
  2465. return 0;
  2466. }
  2467. /*
  2468. * This routine is called when an efd format structure is found in
  2469. * a committed transaction in the log. It's purpose is to cancel
  2470. * the corresponding efi if it was still in the log. To do this
  2471. * it searches the AIL for the efi with an id equal to that in the
  2472. * efd format structure. If we find it, we remove the efi from the
  2473. * AIL and free it.
  2474. */
  2475. STATIC void
  2476. xlog_recover_do_efd_trans(
  2477. xlog_t *log,
  2478. xlog_recover_item_t *item,
  2479. int pass)
  2480. {
  2481. xfs_mount_t *mp;
  2482. xfs_efd_log_format_t *efd_formatp;
  2483. xfs_efi_log_item_t *efip = NULL;
  2484. xfs_log_item_t *lip;
  2485. __uint64_t efi_id;
  2486. struct xfs_ail_cursor cur;
  2487. struct xfs_ail *ailp;
  2488. if (pass == XLOG_RECOVER_PASS1) {
  2489. return;
  2490. }
  2491. efd_formatp = (xfs_efd_log_format_t *)item->ri_buf[0].i_addr;
  2492. ASSERT((item->ri_buf[0].i_len == (sizeof(xfs_efd_log_format_32_t) +
  2493. ((efd_formatp->efd_nextents - 1) * sizeof(xfs_extent_32_t)))) ||
  2494. (item->ri_buf[0].i_len == (sizeof(xfs_efd_log_format_64_t) +
  2495. ((efd_formatp->efd_nextents - 1) * sizeof(xfs_extent_64_t)))));
  2496. efi_id = efd_formatp->efd_efi_id;
  2497. /*
  2498. * Search for the efi with the id in the efd format structure
  2499. * in the AIL.
  2500. */
  2501. mp = log->l_mp;
  2502. ailp = log->l_ailp;
  2503. spin_lock(&ailp->xa_lock);
  2504. lip = xfs_trans_ail_cursor_first(ailp, &cur, 0);
  2505. while (lip != NULL) {
  2506. if (lip->li_type == XFS_LI_EFI) {
  2507. efip = (xfs_efi_log_item_t *)lip;
  2508. if (efip->efi_format.efi_id == efi_id) {
  2509. /*
  2510. * xfs_trans_delete_ail() drops the
  2511. * AIL lock.
  2512. */
  2513. xfs_trans_delete_ail(mp, lip);
  2514. xfs_efi_item_free(efip);
  2515. spin_lock(&ailp->xa_lock);
  2516. break;
  2517. }
  2518. }
  2519. lip = xfs_trans_ail_cursor_next(ailp, &cur);
  2520. }
  2521. xfs_trans_ail_cursor_done(ailp, &cur);
  2522. spin_unlock(&ailp->xa_lock);
  2523. }
  2524. /*
  2525. * Perform the transaction
  2526. *
  2527. * If the transaction modifies a buffer or inode, do it now. Otherwise,
  2528. * EFIs and EFDs get queued up by adding entries into the AIL for them.
  2529. */
  2530. STATIC int
  2531. xlog_recover_do_trans(
  2532. xlog_t *log,
  2533. xlog_recover_t *trans,
  2534. int pass)
  2535. {
  2536. int error = 0;
  2537. xlog_recover_item_t *item, *first_item;
  2538. if ((error = xlog_recover_reorder_trans(trans)))
  2539. return error;
  2540. first_item = item = trans->r_itemq;
  2541. do {
  2542. /*
  2543. * we don't need to worry about the block number being
  2544. * truncated in > 1 TB buffers because in user-land,
  2545. * we're now n32 or 64-bit so xfs_daddr_t is 64-bits so
  2546. * the blknos will get through the user-mode buffer
  2547. * cache properly. The only bad case is o32 kernels
  2548. * where xfs_daddr_t is 32-bits but mount will warn us
  2549. * off a > 1 TB filesystem before we get here.
  2550. */
  2551. if ((ITEM_TYPE(item) == XFS_LI_BUF)) {
  2552. if ((error = xlog_recover_do_buffer_trans(log, item,
  2553. pass)))
  2554. break;
  2555. } else if ((ITEM_TYPE(item) == XFS_LI_INODE)) {
  2556. if ((error = xlog_recover_do_inode_trans(log, item,
  2557. pass)))
  2558. break;
  2559. } else if (ITEM_TYPE(item) == XFS_LI_EFI) {
  2560. if ((error = xlog_recover_do_efi_trans(log, item, trans->r_lsn,
  2561. pass)))
  2562. break;
  2563. } else if (ITEM_TYPE(item) == XFS_LI_EFD) {
  2564. xlog_recover_do_efd_trans(log, item, pass);
  2565. } else if (ITEM_TYPE(item) == XFS_LI_DQUOT) {
  2566. if ((error = xlog_recover_do_dquot_trans(log, item,
  2567. pass)))
  2568. break;
  2569. } else if ((ITEM_TYPE(item) == XFS_LI_QUOTAOFF)) {
  2570. if ((error = xlog_recover_do_quotaoff_trans(log, item,
  2571. pass)))
  2572. break;
  2573. } else {
  2574. xlog_warn("XFS: xlog_recover_do_trans");
  2575. ASSERT(0);
  2576. error = XFS_ERROR(EIO);
  2577. break;
  2578. }
  2579. item = item->ri_next;
  2580. } while (first_item != item);
  2581. return error;
  2582. }
  2583. /*
  2584. * Free up any resources allocated by the transaction
  2585. *
  2586. * Remember that EFIs, EFDs, and IUNLINKs are handled later.
  2587. */
  2588. STATIC void
  2589. xlog_recover_free_trans(
  2590. xlog_recover_t *trans)
  2591. {
  2592. xlog_recover_item_t *first_item, *item, *free_item;
  2593. int i;
  2594. item = first_item = trans->r_itemq;
  2595. do {
  2596. free_item = item;
  2597. item = item->ri_next;
  2598. /* Free the regions in the item. */
  2599. for (i = 0; i < free_item->ri_cnt; i++) {
  2600. kmem_free(free_item->ri_buf[i].i_addr);
  2601. }
  2602. /* Free the item itself */
  2603. kmem_free(free_item->ri_buf);
  2604. kmem_free(free_item);
  2605. } while (first_item != item);
  2606. /* Free the transaction recover structure */
  2607. kmem_free(trans);
  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 + be32_to_cpu(rhead->h_len);
  2658. num_logops = be32_to_cpu(rhead->h_num_logops);
  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 = be32_to_cpu(ohead->oh_tid);
  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. be64_to_cpu(rhead->h_lsn));
  2680. } else {
  2681. if (dp + be32_to_cpu(ohead->oh_len) > lp) {
  2682. xlog_warn(
  2683. "XFS: xlog_recover_process_data: bad length");
  2684. WARN_ON(1);
  2685. return (XFS_ERROR(EIO));
  2686. }
  2687. flags = ohead->oh_flags & ~XLOG_END_TRANS;
  2688. if (flags & XLOG_WAS_CONT_TRANS)
  2689. flags &= ~XLOG_CONTINUE_TRANS;
  2690. switch (flags) {
  2691. case XLOG_COMMIT_TRANS:
  2692. error = xlog_recover_commit_trans(log,
  2693. &rhash[hash], trans, pass);
  2694. break;
  2695. case XLOG_UNMOUNT_TRANS:
  2696. error = xlog_recover_unmount_trans(trans);
  2697. break;
  2698. case XLOG_WAS_CONT_TRANS:
  2699. error = xlog_recover_add_to_cont_trans(trans,
  2700. dp, be32_to_cpu(ohead->oh_len));
  2701. break;
  2702. case XLOG_START_TRANS:
  2703. xlog_warn(
  2704. "XFS: xlog_recover_process_data: bad transaction");
  2705. ASSERT(0);
  2706. error = XFS_ERROR(EIO);
  2707. break;
  2708. case 0:
  2709. case XLOG_CONTINUE_TRANS:
  2710. error = xlog_recover_add_to_trans(trans,
  2711. dp, be32_to_cpu(ohead->oh_len));
  2712. break;
  2713. default:
  2714. xlog_warn(
  2715. "XFS: xlog_recover_process_data: bad flag");
  2716. ASSERT(0);
  2717. error = XFS_ERROR(EIO);
  2718. break;
  2719. }
  2720. if (error)
  2721. return error;
  2722. }
  2723. dp += be32_to_cpu(ohead->oh_len);
  2724. num_logops--;
  2725. }
  2726. return 0;
  2727. }
  2728. /*
  2729. * Process an extent free intent item that was recovered from
  2730. * the log. We need to free the extents that it describes.
  2731. */
  2732. STATIC int
  2733. xlog_recover_process_efi(
  2734. xfs_mount_t *mp,
  2735. xfs_efi_log_item_t *efip)
  2736. {
  2737. xfs_efd_log_item_t *efdp;
  2738. xfs_trans_t *tp;
  2739. int i;
  2740. int error = 0;
  2741. xfs_extent_t *extp;
  2742. xfs_fsblock_t startblock_fsb;
  2743. ASSERT(!(efip->efi_flags & XFS_EFI_RECOVERED));
  2744. /*
  2745. * First check the validity of the extents described by the
  2746. * EFI. If any are bad, then assume that all are bad and
  2747. * just toss the EFI.
  2748. */
  2749. for (i = 0; i < efip->efi_format.efi_nextents; i++) {
  2750. extp = &(efip->efi_format.efi_extents[i]);
  2751. startblock_fsb = XFS_BB_TO_FSB(mp,
  2752. XFS_FSB_TO_DADDR(mp, extp->ext_start));
  2753. if ((startblock_fsb == 0) ||
  2754. (extp->ext_len == 0) ||
  2755. (startblock_fsb >= mp->m_sb.sb_dblocks) ||
  2756. (extp->ext_len >= mp->m_sb.sb_agblocks)) {
  2757. /*
  2758. * This will pull the EFI from the AIL and
  2759. * free the memory associated with it.
  2760. */
  2761. xfs_efi_release(efip, efip->efi_format.efi_nextents);
  2762. return XFS_ERROR(EIO);
  2763. }
  2764. }
  2765. tp = xfs_trans_alloc(mp, 0);
  2766. error = xfs_trans_reserve(tp, 0, XFS_ITRUNCATE_LOG_RES(mp), 0, 0, 0);
  2767. if (error)
  2768. goto abort_error;
  2769. efdp = xfs_trans_get_efd(tp, efip, efip->efi_format.efi_nextents);
  2770. for (i = 0; i < efip->efi_format.efi_nextents; i++) {
  2771. extp = &(efip->efi_format.efi_extents[i]);
  2772. error = xfs_free_extent(tp, extp->ext_start, extp->ext_len);
  2773. if (error)
  2774. goto abort_error;
  2775. xfs_trans_log_efd_extent(tp, efdp, extp->ext_start,
  2776. extp->ext_len);
  2777. }
  2778. efip->efi_flags |= XFS_EFI_RECOVERED;
  2779. error = xfs_trans_commit(tp, 0);
  2780. return error;
  2781. abort_error:
  2782. xfs_trans_cancel(tp, XFS_TRANS_ABORT);
  2783. return error;
  2784. }
  2785. /*
  2786. * When this is called, all of the EFIs which did not have
  2787. * corresponding EFDs should be in the AIL. What we do now
  2788. * is free the extents associated with each one.
  2789. *
  2790. * Since we process the EFIs in normal transactions, they
  2791. * will be removed at some point after the commit. This prevents
  2792. * us from just walking down the list processing each one.
  2793. * We'll use a flag in the EFI to skip those that we've already
  2794. * processed and use the AIL iteration mechanism's generation
  2795. * count to try to speed this up at least a bit.
  2796. *
  2797. * When we start, we know that the EFIs are the only things in
  2798. * the AIL. As we process them, however, other items are added
  2799. * to the AIL. Since everything added to the AIL must come after
  2800. * everything already in the AIL, we stop processing as soon as
  2801. * we see something other than an EFI in the AIL.
  2802. */
  2803. STATIC int
  2804. xlog_recover_process_efis(
  2805. xlog_t *log)
  2806. {
  2807. xfs_log_item_t *lip;
  2808. xfs_efi_log_item_t *efip;
  2809. int error = 0;
  2810. struct xfs_ail_cursor cur;
  2811. struct xfs_ail *ailp;
  2812. ailp = log->l_ailp;
  2813. spin_lock(&ailp->xa_lock);
  2814. lip = xfs_trans_ail_cursor_first(ailp, &cur, 0);
  2815. while (lip != NULL) {
  2816. /*
  2817. * We're done when we see something other than an EFI.
  2818. * There should be no EFIs left in the AIL now.
  2819. */
  2820. if (lip->li_type != XFS_LI_EFI) {
  2821. #ifdef DEBUG
  2822. for (; lip; lip = xfs_trans_ail_cursor_next(ailp, &cur))
  2823. ASSERT(lip->li_type != XFS_LI_EFI);
  2824. #endif
  2825. break;
  2826. }
  2827. /*
  2828. * Skip EFIs that we've already processed.
  2829. */
  2830. efip = (xfs_efi_log_item_t *)lip;
  2831. if (efip->efi_flags & XFS_EFI_RECOVERED) {
  2832. lip = xfs_trans_ail_cursor_next(ailp, &cur);
  2833. continue;
  2834. }
  2835. spin_unlock(&ailp->xa_lock);
  2836. error = xlog_recover_process_efi(log->l_mp, efip);
  2837. spin_lock(&ailp->xa_lock);
  2838. if (error)
  2839. goto out;
  2840. lip = xfs_trans_ail_cursor_next(ailp, &cur);
  2841. }
  2842. out:
  2843. xfs_trans_ail_cursor_done(ailp, &cur);
  2844. spin_unlock(&ailp->xa_lock);
  2845. return error;
  2846. }
  2847. /*
  2848. * This routine performs a transaction to null out a bad inode pointer
  2849. * in an agi unlinked inode hash bucket.
  2850. */
  2851. STATIC void
  2852. xlog_recover_clear_agi_bucket(
  2853. xfs_mount_t *mp,
  2854. xfs_agnumber_t agno,
  2855. int bucket)
  2856. {
  2857. xfs_trans_t *tp;
  2858. xfs_agi_t *agi;
  2859. xfs_buf_t *agibp;
  2860. int offset;
  2861. int error;
  2862. tp = xfs_trans_alloc(mp, XFS_TRANS_CLEAR_AGI_BUCKET);
  2863. error = xfs_trans_reserve(tp, 0, XFS_CLEAR_AGI_BUCKET_LOG_RES(mp), 0, 0, 0);
  2864. if (!error)
  2865. error = xfs_trans_read_buf(mp, tp, mp->m_ddev_targp,
  2866. XFS_AG_DADDR(mp, agno, XFS_AGI_DADDR(mp)),
  2867. XFS_FSS_TO_BB(mp, 1), 0, &agibp);
  2868. if (error)
  2869. goto out_abort;
  2870. error = EINVAL;
  2871. agi = XFS_BUF_TO_AGI(agibp);
  2872. if (be32_to_cpu(agi->agi_magicnum) != XFS_AGI_MAGIC)
  2873. goto out_abort;
  2874. agi->agi_unlinked[bucket] = cpu_to_be32(NULLAGINO);
  2875. offset = offsetof(xfs_agi_t, agi_unlinked) +
  2876. (sizeof(xfs_agino_t) * bucket);
  2877. xfs_trans_log_buf(tp, agibp, offset,
  2878. (offset + sizeof(xfs_agino_t) - 1));
  2879. error = xfs_trans_commit(tp, 0);
  2880. if (error)
  2881. goto out_error;
  2882. return;
  2883. out_abort:
  2884. xfs_trans_cancel(tp, XFS_TRANS_ABORT);
  2885. out_error:
  2886. xfs_fs_cmn_err(CE_WARN, mp, "xlog_recover_clear_agi_bucket: "
  2887. "failed to clear agi %d. Continuing.", agno);
  2888. return;
  2889. }
  2890. /*
  2891. * xlog_iunlink_recover
  2892. *
  2893. * This is called during recovery to process any inodes which
  2894. * we unlinked but not freed when the system crashed. These
  2895. * inodes will be on the lists in the AGI blocks. What we do
  2896. * here is scan all the AGIs and fully truncate and free any
  2897. * inodes found on the lists. Each inode is removed from the
  2898. * lists when it has been fully truncated and is freed. The
  2899. * freeing of the inode and its removal from the list must be
  2900. * atomic.
  2901. */
  2902. void
  2903. xlog_recover_process_iunlinks(
  2904. xlog_t *log)
  2905. {
  2906. xfs_mount_t *mp;
  2907. xfs_agnumber_t agno;
  2908. xfs_agi_t *agi;
  2909. xfs_buf_t *agibp;
  2910. xfs_buf_t *ibp;
  2911. xfs_dinode_t *dip;
  2912. xfs_inode_t *ip;
  2913. xfs_agino_t agino;
  2914. xfs_ino_t ino;
  2915. int bucket;
  2916. int error;
  2917. uint mp_dmevmask;
  2918. mp = log->l_mp;
  2919. /*
  2920. * Prevent any DMAPI event from being sent while in this function.
  2921. */
  2922. mp_dmevmask = mp->m_dmevmask;
  2923. mp->m_dmevmask = 0;
  2924. for (agno = 0; agno < mp->m_sb.sb_agcount; agno++) {
  2925. /*
  2926. * Find the agi for this ag.
  2927. */
  2928. agibp = xfs_buf_read(mp->m_ddev_targp,
  2929. XFS_AG_DADDR(mp, agno, XFS_AGI_DADDR(mp)),
  2930. XFS_FSS_TO_BB(mp, 1), 0);
  2931. if (XFS_BUF_ISERROR(agibp)) {
  2932. xfs_ioerror_alert("xlog_recover_process_iunlinks(#1)",
  2933. log->l_mp, agibp,
  2934. XFS_AG_DADDR(mp, agno, XFS_AGI_DADDR(mp)));
  2935. }
  2936. agi = XFS_BUF_TO_AGI(agibp);
  2937. ASSERT(XFS_AGI_MAGIC == be32_to_cpu(agi->agi_magicnum));
  2938. for (bucket = 0; bucket < XFS_AGI_UNLINKED_BUCKETS; bucket++) {
  2939. agino = be32_to_cpu(agi->agi_unlinked[bucket]);
  2940. while (agino != NULLAGINO) {
  2941. /*
  2942. * Release the agi buffer so that it can
  2943. * be acquired in the normal course of the
  2944. * transaction to truncate and free the inode.
  2945. */
  2946. xfs_buf_relse(agibp);
  2947. ino = XFS_AGINO_TO_INO(mp, agno, agino);
  2948. error = xfs_iget(mp, NULL, ino, 0, 0, &ip, 0);
  2949. ASSERT(error || (ip != NULL));
  2950. if (!error) {
  2951. /*
  2952. * Get the on disk inode to find the
  2953. * next inode in the bucket.
  2954. */
  2955. error = xfs_itobp(mp, NULL, ip, &dip,
  2956. &ibp, 0, 0,
  2957. XFS_BUF_LOCK);
  2958. ASSERT(error || (dip != NULL));
  2959. }
  2960. if (!error) {
  2961. ASSERT(ip->i_d.di_nlink == 0);
  2962. /* setup for the next pass */
  2963. agino = be32_to_cpu(
  2964. dip->di_next_unlinked);
  2965. xfs_buf_relse(ibp);
  2966. /*
  2967. * Prevent any DMAPI event from
  2968. * being sent when the
  2969. * reference on the inode is
  2970. * dropped.
  2971. */
  2972. ip->i_d.di_dmevmask = 0;
  2973. /*
  2974. * If this is a new inode, handle
  2975. * it specially. Otherwise,
  2976. * just drop our reference to the
  2977. * inode. If there are no
  2978. * other references, this will
  2979. * send the inode to
  2980. * xfs_inactive() which will
  2981. * truncate the file and free
  2982. * the inode.
  2983. */
  2984. if (ip->i_d.di_mode == 0)
  2985. xfs_iput_new(ip, 0);
  2986. else
  2987. IRELE(ip);
  2988. } else {
  2989. /*
  2990. * We can't read in the inode
  2991. * this bucket points to, or
  2992. * this inode is messed up. Just
  2993. * ditch this bucket of inodes. We
  2994. * will lose some inodes and space,
  2995. * but at least we won't hang. Call
  2996. * xlog_recover_clear_agi_bucket()
  2997. * to perform a transaction to clear
  2998. * the inode pointer in the bucket.
  2999. */
  3000. xlog_recover_clear_agi_bucket(mp, agno,
  3001. bucket);
  3002. agino = NULLAGINO;
  3003. }
  3004. /*
  3005. * Reacquire the agibuffer and continue around
  3006. * the loop.
  3007. */
  3008. agibp = xfs_buf_read(mp->m_ddev_targp,
  3009. XFS_AG_DADDR(mp, agno,
  3010. XFS_AGI_DADDR(mp)),
  3011. XFS_FSS_TO_BB(mp, 1), 0);
  3012. if (XFS_BUF_ISERROR(agibp)) {
  3013. xfs_ioerror_alert(
  3014. "xlog_recover_process_iunlinks(#2)",
  3015. log->l_mp, agibp,
  3016. XFS_AG_DADDR(mp, agno,
  3017. XFS_AGI_DADDR(mp)));
  3018. }
  3019. agi = XFS_BUF_TO_AGI(agibp);
  3020. ASSERT(XFS_AGI_MAGIC == be32_to_cpu(
  3021. agi->agi_magicnum));
  3022. }
  3023. }
  3024. /*
  3025. * Release the buffer for the current agi so we can
  3026. * go on to the next one.
  3027. */
  3028. xfs_buf_relse(agibp);
  3029. }
  3030. mp->m_dmevmask = mp_dmevmask;
  3031. }
  3032. #ifdef DEBUG
  3033. STATIC void
  3034. xlog_pack_data_checksum(
  3035. xlog_t *log,
  3036. xlog_in_core_t *iclog,
  3037. int size)
  3038. {
  3039. int i;
  3040. __be32 *up;
  3041. uint chksum = 0;
  3042. up = (__be32 *)iclog->ic_datap;
  3043. /* divide length by 4 to get # words */
  3044. for (i = 0; i < (size >> 2); i++) {
  3045. chksum ^= be32_to_cpu(*up);
  3046. up++;
  3047. }
  3048. iclog->ic_header.h_chksum = cpu_to_be32(chksum);
  3049. }
  3050. #else
  3051. #define xlog_pack_data_checksum(log, iclog, size)
  3052. #endif
  3053. /*
  3054. * Stamp cycle number in every block
  3055. */
  3056. void
  3057. xlog_pack_data(
  3058. xlog_t *log,
  3059. xlog_in_core_t *iclog,
  3060. int roundoff)
  3061. {
  3062. int i, j, k;
  3063. int size = iclog->ic_offset + roundoff;
  3064. __be32 cycle_lsn;
  3065. xfs_caddr_t dp;
  3066. xlog_in_core_2_t *xhdr;
  3067. xlog_pack_data_checksum(log, iclog, size);
  3068. cycle_lsn = CYCLE_LSN_DISK(iclog->ic_header.h_lsn);
  3069. dp = iclog->ic_datap;
  3070. for (i = 0; i < BTOBB(size) &&
  3071. i < (XLOG_HEADER_CYCLE_SIZE / BBSIZE); i++) {
  3072. iclog->ic_header.h_cycle_data[i] = *(__be32 *)dp;
  3073. *(__be32 *)dp = cycle_lsn;
  3074. dp += BBSIZE;
  3075. }
  3076. if (xfs_sb_version_haslogv2(&log->l_mp->m_sb)) {
  3077. xhdr = (xlog_in_core_2_t *)&iclog->ic_header;
  3078. for ( ; i < BTOBB(size); i++) {
  3079. j = i / (XLOG_HEADER_CYCLE_SIZE / BBSIZE);
  3080. k = i % (XLOG_HEADER_CYCLE_SIZE / BBSIZE);
  3081. xhdr[j].hic_xheader.xh_cycle_data[k] = *(__be32 *)dp;
  3082. *(__be32 *)dp = cycle_lsn;
  3083. dp += BBSIZE;
  3084. }
  3085. for (i = 1; i < log->l_iclog_heads; i++) {
  3086. xhdr[i].hic_xheader.xh_cycle = cycle_lsn;
  3087. }
  3088. }
  3089. }
  3090. #if defined(DEBUG) && defined(XFS_LOUD_RECOVERY)
  3091. STATIC void
  3092. xlog_unpack_data_checksum(
  3093. xlog_rec_header_t *rhead,
  3094. xfs_caddr_t dp,
  3095. xlog_t *log)
  3096. {
  3097. __be32 *up = (__be32 *)dp;
  3098. uint chksum = 0;
  3099. int i;
  3100. /* divide length by 4 to get # words */
  3101. for (i=0; i < be32_to_cpu(rhead->h_len) >> 2; i++) {
  3102. chksum ^= be32_to_cpu(*up);
  3103. up++;
  3104. }
  3105. if (chksum != be32_to_cpu(rhead->h_chksum)) {
  3106. if (rhead->h_chksum ||
  3107. ((log->l_flags & XLOG_CHKSUM_MISMATCH) == 0)) {
  3108. cmn_err(CE_DEBUG,
  3109. "XFS: LogR chksum mismatch: was (0x%x) is (0x%x)\n",
  3110. be32_to_cpu(rhead->h_chksum), chksum);
  3111. cmn_err(CE_DEBUG,
  3112. "XFS: Disregard message if filesystem was created with non-DEBUG kernel");
  3113. if (xfs_sb_version_haslogv2(&log->l_mp->m_sb)) {
  3114. cmn_err(CE_DEBUG,
  3115. "XFS: LogR this is a LogV2 filesystem\n");
  3116. }
  3117. log->l_flags |= XLOG_CHKSUM_MISMATCH;
  3118. }
  3119. }
  3120. }
  3121. #else
  3122. #define xlog_unpack_data_checksum(rhead, dp, log)
  3123. #endif
  3124. STATIC void
  3125. xlog_unpack_data(
  3126. xlog_rec_header_t *rhead,
  3127. xfs_caddr_t dp,
  3128. xlog_t *log)
  3129. {
  3130. int i, j, k;
  3131. xlog_in_core_2_t *xhdr;
  3132. for (i = 0; i < BTOBB(be32_to_cpu(rhead->h_len)) &&
  3133. i < (XLOG_HEADER_CYCLE_SIZE / BBSIZE); i++) {
  3134. *(__be32 *)dp = *(__be32 *)&rhead->h_cycle_data[i];
  3135. dp += BBSIZE;
  3136. }
  3137. if (xfs_sb_version_haslogv2(&log->l_mp->m_sb)) {
  3138. xhdr = (xlog_in_core_2_t *)rhead;
  3139. for ( ; i < BTOBB(be32_to_cpu(rhead->h_len)); i++) {
  3140. j = i / (XLOG_HEADER_CYCLE_SIZE / BBSIZE);
  3141. k = i % (XLOG_HEADER_CYCLE_SIZE / BBSIZE);
  3142. *(__be32 *)dp = xhdr[j].hic_xheader.xh_cycle_data[k];
  3143. dp += BBSIZE;
  3144. }
  3145. }
  3146. xlog_unpack_data_checksum(rhead, dp, log);
  3147. }
  3148. STATIC int
  3149. xlog_valid_rec_header(
  3150. xlog_t *log,
  3151. xlog_rec_header_t *rhead,
  3152. xfs_daddr_t blkno)
  3153. {
  3154. int hlen;
  3155. if (unlikely(be32_to_cpu(rhead->h_magicno) != XLOG_HEADER_MAGIC_NUM)) {
  3156. XFS_ERROR_REPORT("xlog_valid_rec_header(1)",
  3157. XFS_ERRLEVEL_LOW, log->l_mp);
  3158. return XFS_ERROR(EFSCORRUPTED);
  3159. }
  3160. if (unlikely(
  3161. (!rhead->h_version ||
  3162. (be32_to_cpu(rhead->h_version) & (~XLOG_VERSION_OKBITS))))) {
  3163. xlog_warn("XFS: %s: unrecognised log version (%d).",
  3164. __func__, be32_to_cpu(rhead->h_version));
  3165. return XFS_ERROR(EIO);
  3166. }
  3167. /* LR body must have data or it wouldn't have been written */
  3168. hlen = be32_to_cpu(rhead->h_len);
  3169. if (unlikely( hlen <= 0 || hlen > INT_MAX )) {
  3170. XFS_ERROR_REPORT("xlog_valid_rec_header(2)",
  3171. XFS_ERRLEVEL_LOW, log->l_mp);
  3172. return XFS_ERROR(EFSCORRUPTED);
  3173. }
  3174. if (unlikely( blkno > log->l_logBBsize || blkno > INT_MAX )) {
  3175. XFS_ERROR_REPORT("xlog_valid_rec_header(3)",
  3176. XFS_ERRLEVEL_LOW, log->l_mp);
  3177. return XFS_ERROR(EFSCORRUPTED);
  3178. }
  3179. return 0;
  3180. }
  3181. /*
  3182. * Read the log from tail to head and process the log records found.
  3183. * Handle the two cases where the tail and head are in the same cycle
  3184. * and where the active portion of the log wraps around the end of
  3185. * the physical log separately. The pass parameter is passed through
  3186. * to the routines called to process the data and is not looked at
  3187. * here.
  3188. */
  3189. STATIC int
  3190. xlog_do_recovery_pass(
  3191. xlog_t *log,
  3192. xfs_daddr_t head_blk,
  3193. xfs_daddr_t tail_blk,
  3194. int pass)
  3195. {
  3196. xlog_rec_header_t *rhead;
  3197. xfs_daddr_t blk_no;
  3198. xfs_caddr_t bufaddr, offset;
  3199. xfs_buf_t *hbp, *dbp;
  3200. int error = 0, h_size;
  3201. int bblks, split_bblks;
  3202. int hblks, split_hblks, wrapped_hblks;
  3203. xlog_recover_t *rhash[XLOG_RHASH_SIZE];
  3204. ASSERT(head_blk != tail_blk);
  3205. /*
  3206. * Read the header of the tail block and get the iclog buffer size from
  3207. * h_size. Use this to tell how many sectors make up the log header.
  3208. */
  3209. if (xfs_sb_version_haslogv2(&log->l_mp->m_sb)) {
  3210. /*
  3211. * When using variable length iclogs, read first sector of
  3212. * iclog header and extract the header size from it. Get a
  3213. * new hbp that is the correct size.
  3214. */
  3215. hbp = xlog_get_bp(log, 1);
  3216. if (!hbp)
  3217. return ENOMEM;
  3218. if ((error = xlog_bread(log, tail_blk, 1, hbp)))
  3219. goto bread_err1;
  3220. offset = xlog_align(log, tail_blk, 1, hbp);
  3221. rhead = (xlog_rec_header_t *)offset;
  3222. error = xlog_valid_rec_header(log, rhead, tail_blk);
  3223. if (error)
  3224. goto bread_err1;
  3225. h_size = be32_to_cpu(rhead->h_size);
  3226. if ((be32_to_cpu(rhead->h_version) & 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(be32_to_cpu(rhead->h_len));
  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. wrapped_hblks = hblks - split_hblks;
  3317. bufaddr = XFS_BUF_PTR(hbp);
  3318. error = XFS_BUF_SET_PTR(hbp,
  3319. bufaddr + BBTOB(split_hblks),
  3320. BBTOB(hblks - split_hblks));
  3321. if (!error)
  3322. error = xlog_bread(log, 0,
  3323. wrapped_hblks, hbp);
  3324. if (!error)
  3325. error = XFS_BUF_SET_PTR(hbp, bufaddr,
  3326. BBTOB(hblks));
  3327. if (error)
  3328. goto bread_err2;
  3329. if (!offset)
  3330. offset = xlog_align(log, 0,
  3331. wrapped_hblks, hbp);
  3332. }
  3333. rhead = (xlog_rec_header_t *)offset;
  3334. error = xlog_valid_rec_header(log, rhead,
  3335. split_hblks ? blk_no : 0);
  3336. if (error)
  3337. goto bread_err2;
  3338. bblks = (int)BTOBB(be32_to_cpu(rhead->h_len));
  3339. blk_no += hblks;
  3340. /* Read in data for log record */
  3341. if (blk_no + bblks <= log->l_logBBsize) {
  3342. error = xlog_bread(log, blk_no, bblks, dbp);
  3343. if (error)
  3344. goto bread_err2;
  3345. offset = xlog_align(log, blk_no, bblks, dbp);
  3346. } else {
  3347. /* This log record is split across the
  3348. * physical end of log */
  3349. offset = NULL;
  3350. split_bblks = 0;
  3351. if (blk_no != log->l_logBBsize) {
  3352. /* some data is before the physical
  3353. * end of log */
  3354. ASSERT(!wrapped_hblks);
  3355. ASSERT(blk_no <= INT_MAX);
  3356. split_bblks =
  3357. log->l_logBBsize - (int)blk_no;
  3358. ASSERT(split_bblks > 0);
  3359. if ((error = xlog_bread(log, blk_no,
  3360. split_bblks, dbp)))
  3361. goto bread_err2;
  3362. offset = xlog_align(log, blk_no,
  3363. split_bblks, dbp);
  3364. }
  3365. /*
  3366. * Note: this black magic still works with
  3367. * large sector sizes (non-512) only because:
  3368. * - we increased the buffer size originally
  3369. * by 1 sector giving us enough extra space
  3370. * for the second read;
  3371. * - the log start is guaranteed to be sector
  3372. * aligned;
  3373. * - we read the log end (LR header start)
  3374. * _first_, then the log start (LR header end)
  3375. * - order is important.
  3376. */
  3377. bufaddr = XFS_BUF_PTR(dbp);
  3378. error = XFS_BUF_SET_PTR(dbp,
  3379. bufaddr + BBTOB(split_bblks),
  3380. BBTOB(bblks - split_bblks));
  3381. if (!error)
  3382. error = xlog_bread(log, wrapped_hblks,
  3383. bblks - split_bblks,
  3384. dbp);
  3385. if (!error)
  3386. error = XFS_BUF_SET_PTR(dbp, bufaddr,
  3387. h_size);
  3388. if (error)
  3389. goto bread_err2;
  3390. if (!offset)
  3391. offset = xlog_align(log, wrapped_hblks,
  3392. bblks - split_bblks, dbp);
  3393. }
  3394. xlog_unpack_data(rhead, offset, log);
  3395. if ((error = xlog_recover_process_data(log, rhash,
  3396. rhead, offset, pass)))
  3397. goto bread_err2;
  3398. blk_no += bblks;
  3399. }
  3400. ASSERT(blk_no >= log->l_logBBsize);
  3401. blk_no -= log->l_logBBsize;
  3402. /* read first part of physical log */
  3403. while (blk_no < head_blk) {
  3404. if ((error = xlog_bread(log, blk_no, hblks, hbp)))
  3405. goto bread_err2;
  3406. offset = xlog_align(log, blk_no, hblks, hbp);
  3407. rhead = (xlog_rec_header_t *)offset;
  3408. error = xlog_valid_rec_header(log, rhead, blk_no);
  3409. if (error)
  3410. goto bread_err2;
  3411. bblks = (int)BTOBB(be32_to_cpu(rhead->h_len));
  3412. if ((error = xlog_bread(log, blk_no+hblks, bblks, dbp)))
  3413. goto bread_err2;
  3414. offset = xlog_align(log, blk_no+hblks, bblks, dbp);
  3415. xlog_unpack_data(rhead, offset, log);
  3416. if ((error = xlog_recover_process_data(log, rhash,
  3417. rhead, offset, pass)))
  3418. goto bread_err2;
  3419. blk_no += bblks + hblks;
  3420. }
  3421. }
  3422. bread_err2:
  3423. xlog_put_bp(dbp);
  3424. bread_err1:
  3425. xlog_put_bp(hbp);
  3426. return error;
  3427. }
  3428. /*
  3429. * Do the recovery of the log. We actually do this in two phases.
  3430. * The two passes are necessary in order to implement the function
  3431. * of cancelling a record written into the log. The first pass
  3432. * determines those things which have been cancelled, and the
  3433. * second pass replays log items normally except for those which
  3434. * have been cancelled. The handling of the replay and cancellations
  3435. * takes place in the log item type specific routines.
  3436. *
  3437. * The table of items which have cancel records in the log is allocated
  3438. * and freed at this level, since only here do we know when all of
  3439. * the log recovery has been completed.
  3440. */
  3441. STATIC int
  3442. xlog_do_log_recovery(
  3443. xlog_t *log,
  3444. xfs_daddr_t head_blk,
  3445. xfs_daddr_t tail_blk)
  3446. {
  3447. int error;
  3448. ASSERT(head_blk != tail_blk);
  3449. /*
  3450. * First do a pass to find all of the cancelled buf log items.
  3451. * Store them in the buf_cancel_table for use in the second pass.
  3452. */
  3453. log->l_buf_cancel_table =
  3454. (xfs_buf_cancel_t **)kmem_zalloc(XLOG_BC_TABLE_SIZE *
  3455. sizeof(xfs_buf_cancel_t*),
  3456. KM_SLEEP);
  3457. error = xlog_do_recovery_pass(log, head_blk, tail_blk,
  3458. XLOG_RECOVER_PASS1);
  3459. if (error != 0) {
  3460. kmem_free(log->l_buf_cancel_table);
  3461. log->l_buf_cancel_table = NULL;
  3462. return error;
  3463. }
  3464. /*
  3465. * Then do a second pass to actually recover the items in the log.
  3466. * When it is complete free the table of buf cancel items.
  3467. */
  3468. error = xlog_do_recovery_pass(log, head_blk, tail_blk,
  3469. XLOG_RECOVER_PASS2);
  3470. #ifdef DEBUG
  3471. if (!error) {
  3472. int i;
  3473. for (i = 0; i < XLOG_BC_TABLE_SIZE; i++)
  3474. ASSERT(log->l_buf_cancel_table[i] == NULL);
  3475. }
  3476. #endif /* DEBUG */
  3477. kmem_free(log->l_buf_cancel_table);
  3478. log->l_buf_cancel_table = NULL;
  3479. return error;
  3480. }
  3481. /*
  3482. * Do the actual recovery
  3483. */
  3484. STATIC int
  3485. xlog_do_recover(
  3486. xlog_t *log,
  3487. xfs_daddr_t head_blk,
  3488. xfs_daddr_t tail_blk)
  3489. {
  3490. int error;
  3491. xfs_buf_t *bp;
  3492. xfs_sb_t *sbp;
  3493. /*
  3494. * First replay the images in the log.
  3495. */
  3496. error = xlog_do_log_recovery(log, head_blk, tail_blk);
  3497. if (error) {
  3498. return error;
  3499. }
  3500. XFS_bflush(log->l_mp->m_ddev_targp);
  3501. /*
  3502. * If IO errors happened during recovery, bail out.
  3503. */
  3504. if (XFS_FORCED_SHUTDOWN(log->l_mp)) {
  3505. return (EIO);
  3506. }
  3507. /*
  3508. * We now update the tail_lsn since much of the recovery has completed
  3509. * and there may be space available to use. If there were no extent
  3510. * or iunlinks, we can free up the entire log and set the tail_lsn to
  3511. * be the last_sync_lsn. This was set in xlog_find_tail to be the
  3512. * lsn of the last known good LR on disk. If there are extent frees
  3513. * or iunlinks they will have some entries in the AIL; so we look at
  3514. * the AIL to determine how to set the tail_lsn.
  3515. */
  3516. xlog_assign_tail_lsn(log->l_mp);
  3517. /*
  3518. * Now that we've finished replaying all buffer and inode
  3519. * updates, re-read in the superblock.
  3520. */
  3521. bp = xfs_getsb(log->l_mp, 0);
  3522. XFS_BUF_UNDONE(bp);
  3523. ASSERT(!(XFS_BUF_ISWRITE(bp)));
  3524. ASSERT(!(XFS_BUF_ISDELAYWRITE(bp)));
  3525. XFS_BUF_READ(bp);
  3526. XFS_BUF_UNASYNC(bp);
  3527. xfsbdstrat(log->l_mp, bp);
  3528. error = xfs_iowait(bp);
  3529. if (error) {
  3530. xfs_ioerror_alert("xlog_do_recover",
  3531. log->l_mp, bp, XFS_BUF_ADDR(bp));
  3532. ASSERT(0);
  3533. xfs_buf_relse(bp);
  3534. return error;
  3535. }
  3536. /* Convert superblock from on-disk format */
  3537. sbp = &log->l_mp->m_sb;
  3538. xfs_sb_from_disk(sbp, XFS_BUF_TO_SBP(bp));
  3539. ASSERT(sbp->sb_magicnum == XFS_SB_MAGIC);
  3540. ASSERT(xfs_sb_good_version(sbp));
  3541. xfs_buf_relse(bp);
  3542. /* We've re-read the superblock so re-initialize per-cpu counters */
  3543. xfs_icsb_reinit_counters(log->l_mp);
  3544. xlog_recover_check_summary(log);
  3545. /* Normal transactions can now occur */
  3546. log->l_flags &= ~XLOG_ACTIVE_RECOVERY;
  3547. return 0;
  3548. }
  3549. /*
  3550. * Perform recovery and re-initialize some log variables in xlog_find_tail.
  3551. *
  3552. * Return error or zero.
  3553. */
  3554. int
  3555. xlog_recover(
  3556. xlog_t *log)
  3557. {
  3558. xfs_daddr_t head_blk, tail_blk;
  3559. int error;
  3560. /* find the tail of the log */
  3561. if ((error = xlog_find_tail(log, &head_blk, &tail_blk)))
  3562. return error;
  3563. if (tail_blk != head_blk) {
  3564. /* There used to be a comment here:
  3565. *
  3566. * disallow recovery on read-only mounts. note -- mount
  3567. * checks for ENOSPC and turns it into an intelligent
  3568. * error message.
  3569. * ...but this is no longer true. Now, unless you specify
  3570. * NORECOVERY (in which case this function would never be
  3571. * called), we just go ahead and recover. We do this all
  3572. * under the vfs layer, so we can get away with it unless
  3573. * the device itself is read-only, in which case we fail.
  3574. */
  3575. if ((error = xfs_dev_is_read_only(log->l_mp, "recovery"))) {
  3576. return error;
  3577. }
  3578. cmn_err(CE_NOTE,
  3579. "Starting XFS recovery on filesystem: %s (logdev: %s)",
  3580. log->l_mp->m_fsname, log->l_mp->m_logname ?
  3581. log->l_mp->m_logname : "internal");
  3582. error = xlog_do_recover(log, head_blk, tail_blk);
  3583. log->l_flags |= XLOG_RECOVERY_NEEDED;
  3584. }
  3585. return error;
  3586. }
  3587. /*
  3588. * In the first part of recovery we replay inodes and buffers and build
  3589. * up the list of extent free items which need to be processed. Here
  3590. * we process the extent free items and clean up the on disk unlinked
  3591. * inode lists. This is separated from the first part of recovery so
  3592. * that the root and real-time bitmap inodes can be read in from disk in
  3593. * between the two stages. This is necessary so that we can free space
  3594. * in the real-time portion of the file system.
  3595. */
  3596. int
  3597. xlog_recover_finish(
  3598. xlog_t *log)
  3599. {
  3600. /*
  3601. * Now we're ready to do the transactions needed for the
  3602. * rest of recovery. Start with completing all the extent
  3603. * free intent records and then process the unlinked inode
  3604. * lists. At this point, we essentially run in normal mode
  3605. * except that we're still performing recovery actions
  3606. * rather than accepting new requests.
  3607. */
  3608. if (log->l_flags & XLOG_RECOVERY_NEEDED) {
  3609. int error;
  3610. error = xlog_recover_process_efis(log);
  3611. if (error) {
  3612. cmn_err(CE_ALERT,
  3613. "Failed to recover EFIs on filesystem: %s",
  3614. log->l_mp->m_fsname);
  3615. return error;
  3616. }
  3617. /*
  3618. * Sync the log to get all the EFIs out of the AIL.
  3619. * This isn't absolutely necessary, but it helps in
  3620. * case the unlink transactions would have problems
  3621. * pushing the EFIs out of the way.
  3622. */
  3623. xfs_log_force(log->l_mp, (xfs_lsn_t)0,
  3624. (XFS_LOG_FORCE | XFS_LOG_SYNC));
  3625. xlog_recover_process_iunlinks(log);
  3626. xlog_recover_check_summary(log);
  3627. cmn_err(CE_NOTE,
  3628. "Ending XFS recovery on filesystem: %s (logdev: %s)",
  3629. log->l_mp->m_fsname, log->l_mp->m_logname ?
  3630. log->l_mp->m_logname : "internal");
  3631. log->l_flags &= ~XLOG_RECOVERY_NEEDED;
  3632. } else {
  3633. cmn_err(CE_DEBUG,
  3634. "!Ending clean XFS mount for filesystem: %s\n",
  3635. log->l_mp->m_fsname);
  3636. }
  3637. return 0;
  3638. }
  3639. #if defined(DEBUG)
  3640. /*
  3641. * Read all of the agf and agi counters and check that they
  3642. * are consistent with the superblock counters.
  3643. */
  3644. void
  3645. xlog_recover_check_summary(
  3646. xlog_t *log)
  3647. {
  3648. xfs_mount_t *mp;
  3649. xfs_agf_t *agfp;
  3650. xfs_agi_t *agip;
  3651. xfs_buf_t *agfbp;
  3652. xfs_buf_t *agibp;
  3653. xfs_daddr_t agfdaddr;
  3654. xfs_daddr_t agidaddr;
  3655. xfs_buf_t *sbbp;
  3656. #ifdef XFS_LOUD_RECOVERY
  3657. xfs_sb_t *sbp;
  3658. #endif
  3659. xfs_agnumber_t agno;
  3660. __uint64_t freeblks;
  3661. __uint64_t itotal;
  3662. __uint64_t ifree;
  3663. mp = log->l_mp;
  3664. freeblks = 0LL;
  3665. itotal = 0LL;
  3666. ifree = 0LL;
  3667. for (agno = 0; agno < mp->m_sb.sb_agcount; agno++) {
  3668. agfdaddr = XFS_AG_DADDR(mp, agno, XFS_AGF_DADDR(mp));
  3669. agfbp = xfs_buf_read(mp->m_ddev_targp, agfdaddr,
  3670. XFS_FSS_TO_BB(mp, 1), 0);
  3671. if (XFS_BUF_ISERROR(agfbp)) {
  3672. xfs_ioerror_alert("xlog_recover_check_summary(agf)",
  3673. mp, agfbp, agfdaddr);
  3674. }
  3675. agfp = XFS_BUF_TO_AGF(agfbp);
  3676. ASSERT(XFS_AGF_MAGIC == be32_to_cpu(agfp->agf_magicnum));
  3677. ASSERT(XFS_AGF_GOOD_VERSION(be32_to_cpu(agfp->agf_versionnum)));
  3678. ASSERT(be32_to_cpu(agfp->agf_seqno) == agno);
  3679. freeblks += be32_to_cpu(agfp->agf_freeblks) +
  3680. be32_to_cpu(agfp->agf_flcount);
  3681. xfs_buf_relse(agfbp);
  3682. agidaddr = XFS_AG_DADDR(mp, agno, XFS_AGI_DADDR(mp));
  3683. agibp = xfs_buf_read(mp->m_ddev_targp, agidaddr,
  3684. XFS_FSS_TO_BB(mp, 1), 0);
  3685. if (XFS_BUF_ISERROR(agibp)) {
  3686. xfs_ioerror_alert("xlog_recover_check_summary(agi)",
  3687. mp, agibp, agidaddr);
  3688. }
  3689. agip = XFS_BUF_TO_AGI(agibp);
  3690. ASSERT(XFS_AGI_MAGIC == be32_to_cpu(agip->agi_magicnum));
  3691. ASSERT(XFS_AGI_GOOD_VERSION(be32_to_cpu(agip->agi_versionnum)));
  3692. ASSERT(be32_to_cpu(agip->agi_seqno) == agno);
  3693. itotal += be32_to_cpu(agip->agi_count);
  3694. ifree += be32_to_cpu(agip->agi_freecount);
  3695. xfs_buf_relse(agibp);
  3696. }
  3697. sbbp = xfs_getsb(mp, 0);
  3698. #ifdef XFS_LOUD_RECOVERY
  3699. sbp = &mp->m_sb;
  3700. xfs_sb_from_disk(sbp, XFS_BUF_TO_SBP(sbbp));
  3701. cmn_err(CE_NOTE,
  3702. "xlog_recover_check_summary: sb_icount %Lu itotal %Lu",
  3703. sbp->sb_icount, itotal);
  3704. cmn_err(CE_NOTE,
  3705. "xlog_recover_check_summary: sb_ifree %Lu itotal %Lu",
  3706. sbp->sb_ifree, ifree);
  3707. cmn_err(CE_NOTE,
  3708. "xlog_recover_check_summary: sb_fdblocks %Lu freeblks %Lu",
  3709. sbp->sb_fdblocks, freeblks);
  3710. #if 0
  3711. /*
  3712. * This is turned off until I account for the allocation
  3713. * btree blocks which live in free space.
  3714. */
  3715. ASSERT(sbp->sb_icount == itotal);
  3716. ASSERT(sbp->sb_ifree == ifree);
  3717. ASSERT(sbp->sb_fdblocks == freeblks);
  3718. #endif
  3719. #endif
  3720. xfs_buf_relse(sbbp);
  3721. }
  3722. #endif /* DEBUG */