xfs_inode.c 138 KB

12345678910111213141516171819202122232425262728293031323334353637383940414243444546474849505152535455565758596061626364656667686970717273747576777879808182838485868788899091929394959697989910010110210310410510610710810911011111211311411511611711811912012112212312412512612712812913013113213313413513613713813914014114214314414514614714814915015115215315415515615715815916016116216316416516616716816917017117217317417517617717817918018118218318418518618718818919019119219319419519619719819920020120220320420520620720820921021121221321421521621721821922022122222322422522622722822923023123223323423523623723823924024124224324424524624724824925025125225325425525625725825926026126226326426526626726826927027127227327427527627727827928028128228328428528628728828929029129229329429529629729829930030130230330430530630730830931031131231331431531631731831932032132232332432532632732832933033133233333433533633733833934034134234334434534634734834935035135235335435535635735835936036136236336436536636736836937037137237337437537637737837938038138238338438538638738838939039139239339439539639739839940040140240340440540640740840941041141241341441541641741841942042142242342442542642742842943043143243343443543643743843944044144244344444544644744844945045145245345445545645745845946046146246346446546646746846947047147247347447547647747847948048148248348448548648748848949049149249349449549649749849950050150250350450550650750850951051151251351451551651751851952052152252352452552652752852953053153253353453553653753853954054154254354454554654754854955055155255355455555655755855956056156256356456556656756856957057157257357457557657757857958058158258358458558658758858959059159259359459559659759859960060160260360460560660760860961061161261361461561661761861962062162262362462562662762862963063163263363463563663763863964064164264364464564664764864965065165265365465565665765865966066166266366466566666766866967067167267367467567667767867968068168268368468568668768868969069169269369469569669769869970070170270370470570670770870971071171271371471571671771871972072172272372472572672772872973073173273373473573673773873974074174274374474574674774874975075175275375475575675775875976076176276376476576676776876977077177277377477577677777877978078178278378478578678778878979079179279379479579679779879980080180280380480580680780880981081181281381481581681781881982082182282382482582682782882983083183283383483583683783883984084184284384484584684784884985085185285385485585685785885986086186286386486586686786886987087187287387487587687787887988088188288388488588688788888989089189289389489589689789889990090190290390490590690790890991091191291391491591691791891992092192292392492592692792892993093193293393493593693793893994094194294394494594694794894995095195295395495595695795895996096196296396496596696796896997097197297397497597697797897998098198298398498598698798898999099199299399499599699799899910001001100210031004100510061007100810091010101110121013101410151016101710181019102010211022102310241025102610271028102910301031103210331034103510361037103810391040104110421043104410451046104710481049105010511052105310541055105610571058105910601061106210631064106510661067106810691070107110721073107410751076107710781079108010811082108310841085108610871088108910901091109210931094109510961097109810991100110111021103110411051106110711081109111011111112111311141115111611171118111911201121112211231124112511261127112811291130113111321133113411351136113711381139114011411142114311441145114611471148114911501151115211531154115511561157115811591160116111621163116411651166116711681169117011711172117311741175117611771178117911801181118211831184118511861187118811891190119111921193119411951196119711981199120012011202120312041205120612071208120912101211121212131214121512161217121812191220122112221223122412251226122712281229123012311232123312341235123612371238123912401241124212431244124512461247124812491250125112521253125412551256125712581259126012611262126312641265126612671268126912701271127212731274127512761277127812791280128112821283128412851286128712881289129012911292129312941295129612971298129913001301130213031304130513061307130813091310131113121313131413151316131713181319132013211322132313241325132613271328132913301331133213331334133513361337133813391340134113421343134413451346134713481349135013511352135313541355135613571358135913601361136213631364136513661367136813691370137113721373137413751376137713781379138013811382138313841385138613871388138913901391139213931394139513961397139813991400140114021403140414051406140714081409141014111412141314141415141614171418141914201421142214231424142514261427142814291430143114321433143414351436143714381439144014411442144314441445144614471448144914501451145214531454145514561457145814591460146114621463146414651466146714681469147014711472147314741475147614771478147914801481148214831484148514861487148814891490149114921493149414951496149714981499150015011502150315041505150615071508150915101511151215131514151515161517151815191520152115221523152415251526152715281529153015311532153315341535153615371538153915401541154215431544154515461547154815491550155115521553155415551556155715581559156015611562156315641565156615671568156915701571157215731574157515761577157815791580158115821583158415851586158715881589159015911592159315941595159615971598159916001601160216031604160516061607160816091610161116121613161416151616161716181619162016211622162316241625162616271628162916301631163216331634163516361637163816391640164116421643164416451646164716481649165016511652165316541655165616571658165916601661166216631664166516661667166816691670167116721673167416751676167716781679168016811682168316841685168616871688168916901691169216931694169516961697169816991700170117021703170417051706170717081709171017111712171317141715171617171718171917201721172217231724172517261727172817291730173117321733173417351736173717381739174017411742174317441745174617471748174917501751175217531754175517561757175817591760176117621763176417651766176717681769177017711772177317741775177617771778177917801781178217831784178517861787178817891790179117921793179417951796179717981799180018011802180318041805180618071808180918101811181218131814181518161817181818191820182118221823182418251826182718281829183018311832183318341835183618371838183918401841184218431844184518461847184818491850185118521853185418551856185718581859186018611862186318641865186618671868186918701871187218731874187518761877187818791880188118821883188418851886188718881889189018911892189318941895189618971898189919001901190219031904190519061907190819091910191119121913191419151916191719181919192019211922192319241925192619271928192919301931193219331934193519361937193819391940194119421943194419451946194719481949195019511952195319541955195619571958195919601961196219631964196519661967196819691970197119721973197419751976197719781979198019811982198319841985198619871988198919901991199219931994199519961997199819992000200120022003200420052006200720082009201020112012201320142015201620172018201920202021202220232024202520262027202820292030203120322033203420352036203720382039204020412042204320442045204620472048204920502051205220532054205520562057205820592060206120622063206420652066206720682069207020712072207320742075207620772078207920802081208220832084208520862087208820892090209120922093209420952096209720982099210021012102210321042105210621072108210921102111211221132114211521162117211821192120212121222123212421252126212721282129213021312132213321342135213621372138213921402141214221432144214521462147214821492150215121522153215421552156215721582159216021612162216321642165216621672168216921702171217221732174217521762177217821792180218121822183218421852186218721882189219021912192219321942195219621972198219922002201220222032204220522062207220822092210221122122213221422152216221722182219222022212222222322242225222622272228222922302231223222332234223522362237223822392240224122422243224422452246224722482249225022512252225322542255225622572258225922602261226222632264226522662267226822692270227122722273227422752276227722782279228022812282228322842285228622872288228922902291229222932294229522962297229822992300230123022303230423052306230723082309231023112312231323142315231623172318231923202321232223232324232523262327232823292330233123322333233423352336233723382339234023412342234323442345234623472348234923502351235223532354235523562357235823592360236123622363236423652366236723682369237023712372237323742375237623772378237923802381238223832384238523862387238823892390239123922393239423952396239723982399240024012402240324042405240624072408240924102411241224132414241524162417241824192420242124222423242424252426242724282429243024312432243324342435243624372438243924402441244224432444244524462447244824492450245124522453245424552456245724582459246024612462246324642465246624672468246924702471247224732474247524762477247824792480248124822483248424852486248724882489249024912492249324942495249624972498249925002501250225032504250525062507250825092510251125122513251425152516251725182519252025212522252325242525252625272528252925302531253225332534253525362537253825392540254125422543254425452546254725482549255025512552255325542555255625572558255925602561256225632564256525662567256825692570257125722573257425752576257725782579258025812582258325842585258625872588258925902591259225932594259525962597259825992600260126022603260426052606260726082609261026112612261326142615261626172618261926202621262226232624262526262627262826292630263126322633263426352636263726382639264026412642264326442645264626472648264926502651265226532654265526562657265826592660266126622663266426652666266726682669267026712672267326742675267626772678267926802681268226832684268526862687268826892690269126922693269426952696269726982699270027012702270327042705270627072708270927102711271227132714271527162717271827192720272127222723272427252726272727282729273027312732273327342735273627372738273927402741274227432744274527462747274827492750275127522753275427552756275727582759276027612762276327642765276627672768276927702771277227732774277527762777277827792780278127822783278427852786278727882789279027912792279327942795279627972798279928002801280228032804280528062807280828092810281128122813281428152816281728182819282028212822282328242825282628272828282928302831283228332834283528362837283828392840284128422843284428452846284728482849285028512852285328542855285628572858285928602861286228632864286528662867286828692870287128722873287428752876287728782879288028812882288328842885288628872888288928902891289228932894289528962897289828992900290129022903290429052906290729082909291029112912291329142915291629172918291929202921292229232924292529262927292829292930293129322933293429352936293729382939294029412942294329442945294629472948294929502951295229532954295529562957295829592960296129622963296429652966296729682969297029712972297329742975297629772978297929802981298229832984298529862987298829892990299129922993299429952996299729982999300030013002300330043005300630073008300930103011301230133014301530163017301830193020302130223023302430253026302730283029303030313032303330343035303630373038303930403041304230433044304530463047304830493050305130523053305430553056305730583059306030613062306330643065306630673068306930703071307230733074307530763077307830793080308130823083308430853086308730883089309030913092309330943095309630973098309931003101310231033104310531063107310831093110311131123113311431153116311731183119312031213122312331243125312631273128312931303131313231333134313531363137313831393140314131423143314431453146314731483149315031513152315331543155315631573158315931603161316231633164316531663167316831693170317131723173317431753176317731783179318031813182318331843185318631873188318931903191319231933194319531963197319831993200320132023203320432053206320732083209321032113212321332143215321632173218321932203221322232233224322532263227322832293230323132323233323432353236323732383239324032413242324332443245324632473248324932503251325232533254325532563257325832593260326132623263326432653266326732683269327032713272327332743275327632773278327932803281328232833284328532863287328832893290329132923293329432953296329732983299330033013302330333043305330633073308330933103311331233133314331533163317331833193320332133223323332433253326332733283329333033313332333333343335333633373338333933403341334233433344334533463347334833493350335133523353335433553356335733583359336033613362336333643365336633673368336933703371337233733374337533763377337833793380338133823383338433853386338733883389339033913392339333943395339633973398339934003401340234033404340534063407340834093410341134123413341434153416341734183419342034213422342334243425342634273428342934303431343234333434343534363437343834393440344134423443344434453446344734483449345034513452345334543455345634573458345934603461346234633464346534663467346834693470347134723473347434753476347734783479348034813482348334843485348634873488348934903491349234933494349534963497349834993500350135023503350435053506350735083509351035113512351335143515351635173518351935203521352235233524352535263527352835293530353135323533353435353536353735383539354035413542354335443545354635473548354935503551355235533554355535563557355835593560356135623563356435653566356735683569357035713572357335743575357635773578357935803581358235833584358535863587358835893590359135923593359435953596359735983599360036013602360336043605360636073608360936103611361236133614361536163617361836193620362136223623362436253626362736283629363036313632363336343635363636373638363936403641364236433644364536463647364836493650365136523653365436553656365736583659366036613662366336643665366636673668366936703671367236733674367536763677367836793680368136823683368436853686368736883689369036913692369336943695369636973698369937003701370237033704370537063707370837093710371137123713371437153716371737183719372037213722372337243725372637273728372937303731373237333734373537363737373837393740374137423743374437453746374737483749375037513752375337543755375637573758375937603761376237633764376537663767376837693770377137723773377437753776377737783779378037813782378337843785378637873788378937903791379237933794379537963797379837993800380138023803380438053806380738083809381038113812381338143815381638173818381938203821382238233824382538263827382838293830383138323833383438353836383738383839384038413842384338443845384638473848384938503851385238533854385538563857385838593860386138623863386438653866386738683869387038713872387338743875387638773878387938803881388238833884388538863887388838893890389138923893389438953896389738983899390039013902390339043905390639073908390939103911391239133914391539163917391839193920392139223923392439253926392739283929393039313932393339343935393639373938393939403941394239433944394539463947394839493950395139523953395439553956395739583959396039613962396339643965396639673968396939703971397239733974397539763977397839793980398139823983398439853986398739883989399039913992399339943995399639973998399940004001400240034004400540064007400840094010401140124013401440154016401740184019402040214022402340244025402640274028402940304031403240334034403540364037403840394040404140424043404440454046404740484049405040514052405340544055405640574058405940604061406240634064406540664067406840694070407140724073407440754076407740784079408040814082408340844085408640874088408940904091409240934094409540964097409840994100410141024103410441054106410741084109411041114112411341144115411641174118411941204121412241234124412541264127412841294130413141324133413441354136413741384139414041414142414341444145414641474148414941504151415241534154415541564157415841594160416141624163416441654166416741684169417041714172417341744175417641774178417941804181418241834184418541864187418841894190419141924193419441954196419741984199420042014202420342044205420642074208420942104211421242134214421542164217421842194220422142224223422442254226422742284229423042314232423342344235423642374238423942404241424242434244424542464247424842494250425142524253425442554256425742584259426042614262426342644265426642674268426942704271427242734274427542764277427842794280428142824283428442854286428742884289429042914292429342944295429642974298429943004301430243034304430543064307430843094310431143124313431443154316431743184319432043214322432343244325432643274328432943304331433243334334433543364337433843394340434143424343434443454346434743484349435043514352435343544355435643574358435943604361436243634364436543664367436843694370437143724373437443754376437743784379438043814382438343844385438643874388438943904391439243934394439543964397439843994400440144024403440444054406440744084409441044114412441344144415441644174418441944204421442244234424442544264427442844294430443144324433443444354436443744384439444044414442444344444445444644474448444944504451445244534454445544564457445844594460446144624463446444654466446744684469447044714472447344744475447644774478447944804481448244834484448544864487448844894490449144924493449444954496449744984499450045014502450345044505450645074508450945104511451245134514451545164517451845194520452145224523452445254526452745284529453045314532453345344535453645374538453945404541454245434544454545464547454845494550455145524553455445554556455745584559456045614562456345644565456645674568456945704571457245734574457545764577457845794580458145824583458445854586458745884589459045914592459345944595459645974598459946004601460246034604460546064607460846094610461146124613461446154616461746184619462046214622462346244625462646274628462946304631463246334634463546364637463846394640464146424643464446454646464746484649465046514652465346544655465646574658465946604661466246634664466546664667466846694670467146724673467446754676467746784679468046814682468346844685468646874688468946904691469246934694469546964697469846994700470147024703470447054706470747084709471047114712471347144715471647174718471947204721472247234724472547264727472847294730473147324733473447354736473747384739474047414742474347444745474647474748474947504751475247534754475547564757475847594760476147624763476447654766476747684769477047714772477347744775477647774778477947804781478247834784478547864787478847894790479147924793479447954796479747984799480048014802480348044805
  1. /*
  2. * Copyright (c) 2000-2006 Silicon Graphics, Inc.
  3. * All Rights Reserved.
  4. *
  5. * This program is free software; you can redistribute it and/or
  6. * modify it under the terms of the GNU General Public License as
  7. * published by the Free Software Foundation.
  8. *
  9. * This program is distributed in the hope that it would be useful,
  10. * but WITHOUT ANY WARRANTY; without even the implied warranty of
  11. * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
  12. * GNU General Public License for more details.
  13. *
  14. * You should have received a copy of the GNU General Public License
  15. * along with this program; if not, write the Free Software Foundation,
  16. * Inc., 51 Franklin St, Fifth Floor, Boston, MA 02110-1301 USA
  17. */
  18. #include "xfs.h"
  19. #include "xfs_fs.h"
  20. #include "xfs_types.h"
  21. #include "xfs_bit.h"
  22. #include "xfs_log.h"
  23. #include "xfs_inum.h"
  24. #include "xfs_imap.h"
  25. #include "xfs_trans.h"
  26. #include "xfs_trans_priv.h"
  27. #include "xfs_sb.h"
  28. #include "xfs_ag.h"
  29. #include "xfs_dir2.h"
  30. #include "xfs_dmapi.h"
  31. #include "xfs_mount.h"
  32. #include "xfs_bmap_btree.h"
  33. #include "xfs_alloc_btree.h"
  34. #include "xfs_ialloc_btree.h"
  35. #include "xfs_dir2_sf.h"
  36. #include "xfs_attr_sf.h"
  37. #include "xfs_dinode.h"
  38. #include "xfs_inode.h"
  39. #include "xfs_buf_item.h"
  40. #include "xfs_inode_item.h"
  41. #include "xfs_btree.h"
  42. #include "xfs_alloc.h"
  43. #include "xfs_ialloc.h"
  44. #include "xfs_bmap.h"
  45. #include "xfs_rw.h"
  46. #include "xfs_error.h"
  47. #include "xfs_utils.h"
  48. #include "xfs_dir2_trace.h"
  49. #include "xfs_quota.h"
  50. #include "xfs_acl.h"
  51. #include "xfs_filestream.h"
  52. #include <linux/log2.h>
  53. kmem_zone_t *xfs_ifork_zone;
  54. kmem_zone_t *xfs_inode_zone;
  55. kmem_zone_t *xfs_chashlist_zone;
  56. /*
  57. * Used in xfs_itruncate(). This is the maximum number of extents
  58. * freed from a file in a single transaction.
  59. */
  60. #define XFS_ITRUNC_MAX_EXTENTS 2
  61. STATIC int xfs_iflush_int(xfs_inode_t *, xfs_buf_t *);
  62. STATIC int xfs_iformat_local(xfs_inode_t *, xfs_dinode_t *, int, int);
  63. STATIC int xfs_iformat_extents(xfs_inode_t *, xfs_dinode_t *, int);
  64. STATIC int xfs_iformat_btree(xfs_inode_t *, xfs_dinode_t *, int);
  65. #ifdef DEBUG
  66. /*
  67. * Make sure that the extents in the given memory buffer
  68. * are valid.
  69. */
  70. STATIC void
  71. xfs_validate_extents(
  72. xfs_ifork_t *ifp,
  73. int nrecs,
  74. int disk,
  75. xfs_exntfmt_t fmt)
  76. {
  77. xfs_bmbt_rec_t *ep;
  78. xfs_bmbt_irec_t irec;
  79. xfs_bmbt_rec_t rec;
  80. int i;
  81. for (i = 0; i < nrecs; i++) {
  82. ep = xfs_iext_get_ext(ifp, i);
  83. rec.l0 = get_unaligned((__uint64_t*)&ep->l0);
  84. rec.l1 = get_unaligned((__uint64_t*)&ep->l1);
  85. if (disk)
  86. xfs_bmbt_disk_get_all(&rec, &irec);
  87. else
  88. xfs_bmbt_get_all(&rec, &irec);
  89. if (fmt == XFS_EXTFMT_NOSTATE)
  90. ASSERT(irec.br_state == XFS_EXT_NORM);
  91. }
  92. }
  93. #else /* DEBUG */
  94. #define xfs_validate_extents(ifp, nrecs, disk, fmt)
  95. #endif /* DEBUG */
  96. /*
  97. * Check that none of the inode's in the buffer have a next
  98. * unlinked field of 0.
  99. */
  100. #if defined(DEBUG)
  101. void
  102. xfs_inobp_check(
  103. xfs_mount_t *mp,
  104. xfs_buf_t *bp)
  105. {
  106. int i;
  107. int j;
  108. xfs_dinode_t *dip;
  109. j = mp->m_inode_cluster_size >> mp->m_sb.sb_inodelog;
  110. for (i = 0; i < j; i++) {
  111. dip = (xfs_dinode_t *)xfs_buf_offset(bp,
  112. i * mp->m_sb.sb_inodesize);
  113. if (!dip->di_next_unlinked) {
  114. xfs_fs_cmn_err(CE_ALERT, mp,
  115. "Detected a bogus zero next_unlinked field in incore inode buffer 0x%p. About to pop an ASSERT.",
  116. bp);
  117. ASSERT(dip->di_next_unlinked);
  118. }
  119. }
  120. }
  121. #endif
  122. /*
  123. * This routine is called to map an inode number within a file
  124. * system to the buffer containing the on-disk version of the
  125. * inode. It returns a pointer to the buffer containing the
  126. * on-disk inode in the bpp parameter, and in the dip parameter
  127. * it returns a pointer to the on-disk inode within that buffer.
  128. *
  129. * If a non-zero error is returned, then the contents of bpp and
  130. * dipp are undefined.
  131. *
  132. * Use xfs_imap() to determine the size and location of the
  133. * buffer to read from disk.
  134. */
  135. STATIC int
  136. xfs_inotobp(
  137. xfs_mount_t *mp,
  138. xfs_trans_t *tp,
  139. xfs_ino_t ino,
  140. xfs_dinode_t **dipp,
  141. xfs_buf_t **bpp,
  142. int *offset)
  143. {
  144. int di_ok;
  145. xfs_imap_t imap;
  146. xfs_buf_t *bp;
  147. int error;
  148. xfs_dinode_t *dip;
  149. /*
  150. * Call the space management code to find the location of the
  151. * inode on disk.
  152. */
  153. imap.im_blkno = 0;
  154. error = xfs_imap(mp, tp, ino, &imap, XFS_IMAP_LOOKUP);
  155. if (error != 0) {
  156. cmn_err(CE_WARN,
  157. "xfs_inotobp: xfs_imap() returned an "
  158. "error %d on %s. Returning error.", error, mp->m_fsname);
  159. return error;
  160. }
  161. /*
  162. * If the inode number maps to a block outside the bounds of the
  163. * file system then return NULL rather than calling read_buf
  164. * and panicing when we get an error from the driver.
  165. */
  166. if ((imap.im_blkno + imap.im_len) >
  167. XFS_FSB_TO_BB(mp, mp->m_sb.sb_dblocks)) {
  168. cmn_err(CE_WARN,
  169. "xfs_inotobp: inode number (%llu + %d) maps to a block outside the bounds "
  170. "of the file system %s. Returning EINVAL.",
  171. (unsigned long long)imap.im_blkno,
  172. imap.im_len, mp->m_fsname);
  173. return XFS_ERROR(EINVAL);
  174. }
  175. /*
  176. * Read in the buffer. If tp is NULL, xfs_trans_read_buf() will
  177. * default to just a read_buf() call.
  178. */
  179. error = xfs_trans_read_buf(mp, tp, mp->m_ddev_targp, imap.im_blkno,
  180. (int)imap.im_len, XFS_BUF_LOCK, &bp);
  181. if (error) {
  182. cmn_err(CE_WARN,
  183. "xfs_inotobp: xfs_trans_read_buf() returned an "
  184. "error %d on %s. Returning error.", error, mp->m_fsname);
  185. return error;
  186. }
  187. dip = (xfs_dinode_t *)xfs_buf_offset(bp, 0);
  188. di_ok =
  189. INT_GET(dip->di_core.di_magic, ARCH_CONVERT) == XFS_DINODE_MAGIC &&
  190. XFS_DINODE_GOOD_VERSION(INT_GET(dip->di_core.di_version, ARCH_CONVERT));
  191. if (unlikely(XFS_TEST_ERROR(!di_ok, mp, XFS_ERRTAG_ITOBP_INOTOBP,
  192. XFS_RANDOM_ITOBP_INOTOBP))) {
  193. XFS_CORRUPTION_ERROR("xfs_inotobp", XFS_ERRLEVEL_LOW, mp, dip);
  194. xfs_trans_brelse(tp, bp);
  195. cmn_err(CE_WARN,
  196. "xfs_inotobp: XFS_TEST_ERROR() returned an "
  197. "error on %s. Returning EFSCORRUPTED.", mp->m_fsname);
  198. return XFS_ERROR(EFSCORRUPTED);
  199. }
  200. xfs_inobp_check(mp, bp);
  201. /*
  202. * Set *dipp to point to the on-disk inode in the buffer.
  203. */
  204. *dipp = (xfs_dinode_t *)xfs_buf_offset(bp, imap.im_boffset);
  205. *bpp = bp;
  206. *offset = imap.im_boffset;
  207. return 0;
  208. }
  209. /*
  210. * This routine is called to map an inode to the buffer containing
  211. * the on-disk version of the inode. It returns a pointer to the
  212. * buffer containing the on-disk inode in the bpp parameter, and in
  213. * the dip parameter it returns a pointer to the on-disk inode within
  214. * that buffer.
  215. *
  216. * If a non-zero error is returned, then the contents of bpp and
  217. * dipp are undefined.
  218. *
  219. * If the inode is new and has not yet been initialized, use xfs_imap()
  220. * to determine the size and location of the buffer to read from disk.
  221. * If the inode has already been mapped to its buffer and read in once,
  222. * then use the mapping information stored in the inode rather than
  223. * calling xfs_imap(). This allows us to avoid the overhead of looking
  224. * at the inode btree for small block file systems (see xfs_dilocate()).
  225. * We can tell whether the inode has been mapped in before by comparing
  226. * its disk block address to 0. Only uninitialized inodes will have
  227. * 0 for the disk block address.
  228. */
  229. int
  230. xfs_itobp(
  231. xfs_mount_t *mp,
  232. xfs_trans_t *tp,
  233. xfs_inode_t *ip,
  234. xfs_dinode_t **dipp,
  235. xfs_buf_t **bpp,
  236. xfs_daddr_t bno,
  237. uint imap_flags)
  238. {
  239. xfs_imap_t imap;
  240. xfs_buf_t *bp;
  241. int error;
  242. int i;
  243. int ni;
  244. if (ip->i_blkno == (xfs_daddr_t)0) {
  245. /*
  246. * Call the space management code to find the location of the
  247. * inode on disk.
  248. */
  249. imap.im_blkno = bno;
  250. if ((error = xfs_imap(mp, tp, ip->i_ino, &imap,
  251. XFS_IMAP_LOOKUP | imap_flags)))
  252. return error;
  253. /*
  254. * If the inode number maps to a block outside the bounds
  255. * of the file system then return NULL rather than calling
  256. * read_buf and panicing when we get an error from the
  257. * driver.
  258. */
  259. if ((imap.im_blkno + imap.im_len) >
  260. XFS_FSB_TO_BB(mp, mp->m_sb.sb_dblocks)) {
  261. #ifdef DEBUG
  262. xfs_fs_cmn_err(CE_ALERT, mp, "xfs_itobp: "
  263. "(imap.im_blkno (0x%llx) "
  264. "+ imap.im_len (0x%llx)) > "
  265. " XFS_FSB_TO_BB(mp, "
  266. "mp->m_sb.sb_dblocks) (0x%llx)",
  267. (unsigned long long) imap.im_blkno,
  268. (unsigned long long) imap.im_len,
  269. XFS_FSB_TO_BB(mp, mp->m_sb.sb_dblocks));
  270. #endif /* DEBUG */
  271. return XFS_ERROR(EINVAL);
  272. }
  273. /*
  274. * Fill in the fields in the inode that will be used to
  275. * map the inode to its buffer from now on.
  276. */
  277. ip->i_blkno = imap.im_blkno;
  278. ip->i_len = imap.im_len;
  279. ip->i_boffset = imap.im_boffset;
  280. } else {
  281. /*
  282. * We've already mapped the inode once, so just use the
  283. * mapping that we saved the first time.
  284. */
  285. imap.im_blkno = ip->i_blkno;
  286. imap.im_len = ip->i_len;
  287. imap.im_boffset = ip->i_boffset;
  288. }
  289. ASSERT(bno == 0 || bno == imap.im_blkno);
  290. /*
  291. * Read in the buffer. If tp is NULL, xfs_trans_read_buf() will
  292. * default to just a read_buf() call.
  293. */
  294. error = xfs_trans_read_buf(mp, tp, mp->m_ddev_targp, imap.im_blkno,
  295. (int)imap.im_len, XFS_BUF_LOCK, &bp);
  296. if (error) {
  297. #ifdef DEBUG
  298. xfs_fs_cmn_err(CE_ALERT, mp, "xfs_itobp: "
  299. "xfs_trans_read_buf() returned error %d, "
  300. "imap.im_blkno 0x%llx, imap.im_len 0x%llx",
  301. error, (unsigned long long) imap.im_blkno,
  302. (unsigned long long) imap.im_len);
  303. #endif /* DEBUG */
  304. return error;
  305. }
  306. /*
  307. * Validate the magic number and version of every inode in the buffer
  308. * (if DEBUG kernel) or the first inode in the buffer, otherwise.
  309. * No validation is done here in userspace (xfs_repair).
  310. */
  311. #if !defined(__KERNEL__)
  312. ni = 0;
  313. #elif defined(DEBUG)
  314. ni = BBTOB(imap.im_len) >> mp->m_sb.sb_inodelog;
  315. #else /* usual case */
  316. ni = 1;
  317. #endif
  318. for (i = 0; i < ni; i++) {
  319. int di_ok;
  320. xfs_dinode_t *dip;
  321. dip = (xfs_dinode_t *)xfs_buf_offset(bp,
  322. (i << mp->m_sb.sb_inodelog));
  323. di_ok = INT_GET(dip->di_core.di_magic, ARCH_CONVERT) == XFS_DINODE_MAGIC &&
  324. XFS_DINODE_GOOD_VERSION(INT_GET(dip->di_core.di_version, ARCH_CONVERT));
  325. if (unlikely(XFS_TEST_ERROR(!di_ok, mp,
  326. XFS_ERRTAG_ITOBP_INOTOBP,
  327. XFS_RANDOM_ITOBP_INOTOBP))) {
  328. if (imap_flags & XFS_IMAP_BULKSTAT) {
  329. xfs_trans_brelse(tp, bp);
  330. return XFS_ERROR(EINVAL);
  331. }
  332. #ifdef DEBUG
  333. cmn_err(CE_ALERT,
  334. "Device %s - bad inode magic/vsn "
  335. "daddr %lld #%d (magic=%x)",
  336. XFS_BUFTARG_NAME(mp->m_ddev_targp),
  337. (unsigned long long)imap.im_blkno, i,
  338. INT_GET(dip->di_core.di_magic, ARCH_CONVERT));
  339. #endif
  340. XFS_CORRUPTION_ERROR("xfs_itobp", XFS_ERRLEVEL_HIGH,
  341. mp, dip);
  342. xfs_trans_brelse(tp, bp);
  343. return XFS_ERROR(EFSCORRUPTED);
  344. }
  345. }
  346. xfs_inobp_check(mp, bp);
  347. /*
  348. * Mark the buffer as an inode buffer now that it looks good
  349. */
  350. XFS_BUF_SET_VTYPE(bp, B_FS_INO);
  351. /*
  352. * Set *dipp to point to the on-disk inode in the buffer.
  353. */
  354. *dipp = (xfs_dinode_t *)xfs_buf_offset(bp, imap.im_boffset);
  355. *bpp = bp;
  356. return 0;
  357. }
  358. /*
  359. * Move inode type and inode format specific information from the
  360. * on-disk inode to the in-core inode. For fifos, devs, and sockets
  361. * this means set if_rdev to the proper value. For files, directories,
  362. * and symlinks this means to bring in the in-line data or extent
  363. * pointers. For a file in B-tree format, only the root is immediately
  364. * brought in-core. The rest will be in-lined in if_extents when it
  365. * is first referenced (see xfs_iread_extents()).
  366. */
  367. STATIC int
  368. xfs_iformat(
  369. xfs_inode_t *ip,
  370. xfs_dinode_t *dip)
  371. {
  372. xfs_attr_shortform_t *atp;
  373. int size;
  374. int error;
  375. xfs_fsize_t di_size;
  376. ip->i_df.if_ext_max =
  377. XFS_IFORK_DSIZE(ip) / (uint)sizeof(xfs_bmbt_rec_t);
  378. error = 0;
  379. if (unlikely(
  380. INT_GET(dip->di_core.di_nextents, ARCH_CONVERT) +
  381. INT_GET(dip->di_core.di_anextents, ARCH_CONVERT) >
  382. INT_GET(dip->di_core.di_nblocks, ARCH_CONVERT))) {
  383. xfs_fs_repair_cmn_err(CE_WARN, ip->i_mount,
  384. "corrupt dinode %Lu, extent total = %d, nblocks = %Lu.",
  385. (unsigned long long)ip->i_ino,
  386. (int)(INT_GET(dip->di_core.di_nextents, ARCH_CONVERT)
  387. + INT_GET(dip->di_core.di_anextents, ARCH_CONVERT)),
  388. (unsigned long long)
  389. INT_GET(dip->di_core.di_nblocks, ARCH_CONVERT));
  390. XFS_CORRUPTION_ERROR("xfs_iformat(1)", XFS_ERRLEVEL_LOW,
  391. ip->i_mount, dip);
  392. return XFS_ERROR(EFSCORRUPTED);
  393. }
  394. if (unlikely(INT_GET(dip->di_core.di_forkoff, ARCH_CONVERT) > ip->i_mount->m_sb.sb_inodesize)) {
  395. xfs_fs_repair_cmn_err(CE_WARN, ip->i_mount,
  396. "corrupt dinode %Lu, forkoff = 0x%x.",
  397. (unsigned long long)ip->i_ino,
  398. (int)(INT_GET(dip->di_core.di_forkoff, ARCH_CONVERT)));
  399. XFS_CORRUPTION_ERROR("xfs_iformat(2)", XFS_ERRLEVEL_LOW,
  400. ip->i_mount, dip);
  401. return XFS_ERROR(EFSCORRUPTED);
  402. }
  403. switch (ip->i_d.di_mode & S_IFMT) {
  404. case S_IFIFO:
  405. case S_IFCHR:
  406. case S_IFBLK:
  407. case S_IFSOCK:
  408. if (unlikely(INT_GET(dip->di_core.di_format, ARCH_CONVERT) != XFS_DINODE_FMT_DEV)) {
  409. XFS_CORRUPTION_ERROR("xfs_iformat(3)", XFS_ERRLEVEL_LOW,
  410. ip->i_mount, dip);
  411. return XFS_ERROR(EFSCORRUPTED);
  412. }
  413. ip->i_d.di_size = 0;
  414. ip->i_size = 0;
  415. ip->i_df.if_u2.if_rdev = INT_GET(dip->di_u.di_dev, ARCH_CONVERT);
  416. break;
  417. case S_IFREG:
  418. case S_IFLNK:
  419. case S_IFDIR:
  420. switch (INT_GET(dip->di_core.di_format, ARCH_CONVERT)) {
  421. case XFS_DINODE_FMT_LOCAL:
  422. /*
  423. * no local regular files yet
  424. */
  425. if (unlikely((INT_GET(dip->di_core.di_mode, ARCH_CONVERT) & S_IFMT) == S_IFREG)) {
  426. xfs_fs_repair_cmn_err(CE_WARN, ip->i_mount,
  427. "corrupt inode %Lu "
  428. "(local format for regular file).",
  429. (unsigned long long) ip->i_ino);
  430. XFS_CORRUPTION_ERROR("xfs_iformat(4)",
  431. XFS_ERRLEVEL_LOW,
  432. ip->i_mount, dip);
  433. return XFS_ERROR(EFSCORRUPTED);
  434. }
  435. di_size = INT_GET(dip->di_core.di_size, ARCH_CONVERT);
  436. if (unlikely(di_size > XFS_DFORK_DSIZE(dip, ip->i_mount))) {
  437. xfs_fs_repair_cmn_err(CE_WARN, ip->i_mount,
  438. "corrupt inode %Lu "
  439. "(bad size %Ld for local inode).",
  440. (unsigned long long) ip->i_ino,
  441. (long long) di_size);
  442. XFS_CORRUPTION_ERROR("xfs_iformat(5)",
  443. XFS_ERRLEVEL_LOW,
  444. ip->i_mount, dip);
  445. return XFS_ERROR(EFSCORRUPTED);
  446. }
  447. size = (int)di_size;
  448. error = xfs_iformat_local(ip, dip, XFS_DATA_FORK, size);
  449. break;
  450. case XFS_DINODE_FMT_EXTENTS:
  451. error = xfs_iformat_extents(ip, dip, XFS_DATA_FORK);
  452. break;
  453. case XFS_DINODE_FMT_BTREE:
  454. error = xfs_iformat_btree(ip, dip, XFS_DATA_FORK);
  455. break;
  456. default:
  457. XFS_ERROR_REPORT("xfs_iformat(6)", XFS_ERRLEVEL_LOW,
  458. ip->i_mount);
  459. return XFS_ERROR(EFSCORRUPTED);
  460. }
  461. break;
  462. default:
  463. XFS_ERROR_REPORT("xfs_iformat(7)", XFS_ERRLEVEL_LOW, ip->i_mount);
  464. return XFS_ERROR(EFSCORRUPTED);
  465. }
  466. if (error) {
  467. return error;
  468. }
  469. if (!XFS_DFORK_Q(dip))
  470. return 0;
  471. ASSERT(ip->i_afp == NULL);
  472. ip->i_afp = kmem_zone_zalloc(xfs_ifork_zone, KM_SLEEP);
  473. ip->i_afp->if_ext_max =
  474. XFS_IFORK_ASIZE(ip) / (uint)sizeof(xfs_bmbt_rec_t);
  475. switch (INT_GET(dip->di_core.di_aformat, ARCH_CONVERT)) {
  476. case XFS_DINODE_FMT_LOCAL:
  477. atp = (xfs_attr_shortform_t *)XFS_DFORK_APTR(dip);
  478. size = be16_to_cpu(atp->hdr.totsize);
  479. error = xfs_iformat_local(ip, dip, XFS_ATTR_FORK, size);
  480. break;
  481. case XFS_DINODE_FMT_EXTENTS:
  482. error = xfs_iformat_extents(ip, dip, XFS_ATTR_FORK);
  483. break;
  484. case XFS_DINODE_FMT_BTREE:
  485. error = xfs_iformat_btree(ip, dip, XFS_ATTR_FORK);
  486. break;
  487. default:
  488. error = XFS_ERROR(EFSCORRUPTED);
  489. break;
  490. }
  491. if (error) {
  492. kmem_zone_free(xfs_ifork_zone, ip->i_afp);
  493. ip->i_afp = NULL;
  494. xfs_idestroy_fork(ip, XFS_DATA_FORK);
  495. }
  496. return error;
  497. }
  498. /*
  499. * The file is in-lined in the on-disk inode.
  500. * If it fits into if_inline_data, then copy
  501. * it there, otherwise allocate a buffer for it
  502. * and copy the data there. Either way, set
  503. * if_data to point at the data.
  504. * If we allocate a buffer for the data, make
  505. * sure that its size is a multiple of 4 and
  506. * record the real size in i_real_bytes.
  507. */
  508. STATIC int
  509. xfs_iformat_local(
  510. xfs_inode_t *ip,
  511. xfs_dinode_t *dip,
  512. int whichfork,
  513. int size)
  514. {
  515. xfs_ifork_t *ifp;
  516. int real_size;
  517. /*
  518. * If the size is unreasonable, then something
  519. * is wrong and we just bail out rather than crash in
  520. * kmem_alloc() or memcpy() below.
  521. */
  522. if (unlikely(size > XFS_DFORK_SIZE(dip, ip->i_mount, whichfork))) {
  523. xfs_fs_repair_cmn_err(CE_WARN, ip->i_mount,
  524. "corrupt inode %Lu "
  525. "(bad size %d for local fork, size = %d).",
  526. (unsigned long long) ip->i_ino, size,
  527. XFS_DFORK_SIZE(dip, ip->i_mount, whichfork));
  528. XFS_CORRUPTION_ERROR("xfs_iformat_local", XFS_ERRLEVEL_LOW,
  529. ip->i_mount, dip);
  530. return XFS_ERROR(EFSCORRUPTED);
  531. }
  532. ifp = XFS_IFORK_PTR(ip, whichfork);
  533. real_size = 0;
  534. if (size == 0)
  535. ifp->if_u1.if_data = NULL;
  536. else if (size <= sizeof(ifp->if_u2.if_inline_data))
  537. ifp->if_u1.if_data = ifp->if_u2.if_inline_data;
  538. else {
  539. real_size = roundup(size, 4);
  540. ifp->if_u1.if_data = kmem_alloc(real_size, KM_SLEEP);
  541. }
  542. ifp->if_bytes = size;
  543. ifp->if_real_bytes = real_size;
  544. if (size)
  545. memcpy(ifp->if_u1.if_data, XFS_DFORK_PTR(dip, whichfork), size);
  546. ifp->if_flags &= ~XFS_IFEXTENTS;
  547. ifp->if_flags |= XFS_IFINLINE;
  548. return 0;
  549. }
  550. /*
  551. * The file consists of a set of extents all
  552. * of which fit into the on-disk inode.
  553. * If there are few enough extents to fit into
  554. * the if_inline_ext, then copy them there.
  555. * Otherwise allocate a buffer for them and copy
  556. * them into it. Either way, set if_extents
  557. * to point at the extents.
  558. */
  559. STATIC int
  560. xfs_iformat_extents(
  561. xfs_inode_t *ip,
  562. xfs_dinode_t *dip,
  563. int whichfork)
  564. {
  565. xfs_bmbt_rec_t *ep, *dp;
  566. xfs_ifork_t *ifp;
  567. int nex;
  568. int size;
  569. int i;
  570. ifp = XFS_IFORK_PTR(ip, whichfork);
  571. nex = XFS_DFORK_NEXTENTS(dip, whichfork);
  572. size = nex * (uint)sizeof(xfs_bmbt_rec_t);
  573. /*
  574. * If the number of extents is unreasonable, then something
  575. * is wrong and we just bail out rather than crash in
  576. * kmem_alloc() or memcpy() below.
  577. */
  578. if (unlikely(size < 0 || size > XFS_DFORK_SIZE(dip, ip->i_mount, whichfork))) {
  579. xfs_fs_repair_cmn_err(CE_WARN, ip->i_mount,
  580. "corrupt inode %Lu ((a)extents = %d).",
  581. (unsigned long long) ip->i_ino, nex);
  582. XFS_CORRUPTION_ERROR("xfs_iformat_extents(1)", XFS_ERRLEVEL_LOW,
  583. ip->i_mount, dip);
  584. return XFS_ERROR(EFSCORRUPTED);
  585. }
  586. ifp->if_real_bytes = 0;
  587. if (nex == 0)
  588. ifp->if_u1.if_extents = NULL;
  589. else if (nex <= XFS_INLINE_EXTS)
  590. ifp->if_u1.if_extents = ifp->if_u2.if_inline_ext;
  591. else
  592. xfs_iext_add(ifp, 0, nex);
  593. ifp->if_bytes = size;
  594. if (size) {
  595. dp = (xfs_bmbt_rec_t *) XFS_DFORK_PTR(dip, whichfork);
  596. xfs_validate_extents(ifp, nex, 1, XFS_EXTFMT_INODE(ip));
  597. for (i = 0; i < nex; i++, dp++) {
  598. ep = xfs_iext_get_ext(ifp, i);
  599. ep->l0 = INT_GET(get_unaligned((__uint64_t*)&dp->l0),
  600. ARCH_CONVERT);
  601. ep->l1 = INT_GET(get_unaligned((__uint64_t*)&dp->l1),
  602. ARCH_CONVERT);
  603. }
  604. xfs_bmap_trace_exlist("xfs_iformat_extents", ip, nex,
  605. whichfork);
  606. if (whichfork != XFS_DATA_FORK ||
  607. XFS_EXTFMT_INODE(ip) == XFS_EXTFMT_NOSTATE)
  608. if (unlikely(xfs_check_nostate_extents(
  609. ifp, 0, nex))) {
  610. XFS_ERROR_REPORT("xfs_iformat_extents(2)",
  611. XFS_ERRLEVEL_LOW,
  612. ip->i_mount);
  613. return XFS_ERROR(EFSCORRUPTED);
  614. }
  615. }
  616. ifp->if_flags |= XFS_IFEXTENTS;
  617. return 0;
  618. }
  619. /*
  620. * The file has too many extents to fit into
  621. * the inode, so they are in B-tree format.
  622. * Allocate a buffer for the root of the B-tree
  623. * and copy the root into it. The i_extents
  624. * field will remain NULL until all of the
  625. * extents are read in (when they are needed).
  626. */
  627. STATIC int
  628. xfs_iformat_btree(
  629. xfs_inode_t *ip,
  630. xfs_dinode_t *dip,
  631. int whichfork)
  632. {
  633. xfs_bmdr_block_t *dfp;
  634. xfs_ifork_t *ifp;
  635. /* REFERENCED */
  636. int nrecs;
  637. int size;
  638. ifp = XFS_IFORK_PTR(ip, whichfork);
  639. dfp = (xfs_bmdr_block_t *)XFS_DFORK_PTR(dip, whichfork);
  640. size = XFS_BMAP_BROOT_SPACE(dfp);
  641. nrecs = XFS_BMAP_BROOT_NUMRECS(dfp);
  642. /*
  643. * blow out if -- fork has less extents than can fit in
  644. * fork (fork shouldn't be a btree format), root btree
  645. * block has more records than can fit into the fork,
  646. * or the number of extents is greater than the number of
  647. * blocks.
  648. */
  649. if (unlikely(XFS_IFORK_NEXTENTS(ip, whichfork) <= ifp->if_ext_max
  650. || XFS_BMDR_SPACE_CALC(nrecs) >
  651. XFS_DFORK_SIZE(dip, ip->i_mount, whichfork)
  652. || XFS_IFORK_NEXTENTS(ip, whichfork) > ip->i_d.di_nblocks)) {
  653. xfs_fs_repair_cmn_err(CE_WARN, ip->i_mount,
  654. "corrupt inode %Lu (btree).",
  655. (unsigned long long) ip->i_ino);
  656. XFS_ERROR_REPORT("xfs_iformat_btree", XFS_ERRLEVEL_LOW,
  657. ip->i_mount);
  658. return XFS_ERROR(EFSCORRUPTED);
  659. }
  660. ifp->if_broot_bytes = size;
  661. ifp->if_broot = kmem_alloc(size, KM_SLEEP);
  662. ASSERT(ifp->if_broot != NULL);
  663. /*
  664. * Copy and convert from the on-disk structure
  665. * to the in-memory structure.
  666. */
  667. xfs_bmdr_to_bmbt(dfp, XFS_DFORK_SIZE(dip, ip->i_mount, whichfork),
  668. ifp->if_broot, size);
  669. ifp->if_flags &= ~XFS_IFEXTENTS;
  670. ifp->if_flags |= XFS_IFBROOT;
  671. return 0;
  672. }
  673. /*
  674. * xfs_xlate_dinode_core - translate an xfs_inode_core_t between ondisk
  675. * and native format
  676. *
  677. * buf = on-disk representation
  678. * dip = native representation
  679. * dir = direction - +ve -> disk to native
  680. * -ve -> native to disk
  681. */
  682. void
  683. xfs_xlate_dinode_core(
  684. xfs_caddr_t buf,
  685. xfs_dinode_core_t *dip,
  686. int dir)
  687. {
  688. xfs_dinode_core_t *buf_core = (xfs_dinode_core_t *)buf;
  689. xfs_dinode_core_t *mem_core = (xfs_dinode_core_t *)dip;
  690. xfs_arch_t arch = ARCH_CONVERT;
  691. ASSERT(dir);
  692. INT_XLATE(buf_core->di_magic, mem_core->di_magic, dir, arch);
  693. INT_XLATE(buf_core->di_mode, mem_core->di_mode, dir, arch);
  694. INT_XLATE(buf_core->di_version, mem_core->di_version, dir, arch);
  695. INT_XLATE(buf_core->di_format, mem_core->di_format, dir, arch);
  696. INT_XLATE(buf_core->di_onlink, mem_core->di_onlink, dir, arch);
  697. INT_XLATE(buf_core->di_uid, mem_core->di_uid, dir, arch);
  698. INT_XLATE(buf_core->di_gid, mem_core->di_gid, dir, arch);
  699. INT_XLATE(buf_core->di_nlink, mem_core->di_nlink, dir, arch);
  700. INT_XLATE(buf_core->di_projid, mem_core->di_projid, dir, arch);
  701. if (dir > 0) {
  702. memcpy(mem_core->di_pad, buf_core->di_pad,
  703. sizeof(buf_core->di_pad));
  704. } else {
  705. memcpy(buf_core->di_pad, mem_core->di_pad,
  706. sizeof(buf_core->di_pad));
  707. }
  708. INT_XLATE(buf_core->di_flushiter, mem_core->di_flushiter, dir, arch);
  709. INT_XLATE(buf_core->di_atime.t_sec, mem_core->di_atime.t_sec,
  710. dir, arch);
  711. INT_XLATE(buf_core->di_atime.t_nsec, mem_core->di_atime.t_nsec,
  712. dir, arch);
  713. INT_XLATE(buf_core->di_mtime.t_sec, mem_core->di_mtime.t_sec,
  714. dir, arch);
  715. INT_XLATE(buf_core->di_mtime.t_nsec, mem_core->di_mtime.t_nsec,
  716. dir, arch);
  717. INT_XLATE(buf_core->di_ctime.t_sec, mem_core->di_ctime.t_sec,
  718. dir, arch);
  719. INT_XLATE(buf_core->di_ctime.t_nsec, mem_core->di_ctime.t_nsec,
  720. dir, arch);
  721. INT_XLATE(buf_core->di_size, mem_core->di_size, dir, arch);
  722. INT_XLATE(buf_core->di_nblocks, mem_core->di_nblocks, dir, arch);
  723. INT_XLATE(buf_core->di_extsize, mem_core->di_extsize, dir, arch);
  724. INT_XLATE(buf_core->di_nextents, mem_core->di_nextents, dir, arch);
  725. INT_XLATE(buf_core->di_anextents, mem_core->di_anextents, dir, arch);
  726. INT_XLATE(buf_core->di_forkoff, mem_core->di_forkoff, dir, arch);
  727. INT_XLATE(buf_core->di_aformat, mem_core->di_aformat, dir, arch);
  728. INT_XLATE(buf_core->di_dmevmask, mem_core->di_dmevmask, dir, arch);
  729. INT_XLATE(buf_core->di_dmstate, mem_core->di_dmstate, dir, arch);
  730. INT_XLATE(buf_core->di_flags, mem_core->di_flags, dir, arch);
  731. INT_XLATE(buf_core->di_gen, mem_core->di_gen, dir, arch);
  732. }
  733. STATIC uint
  734. _xfs_dic2xflags(
  735. __uint16_t di_flags)
  736. {
  737. uint flags = 0;
  738. if (di_flags & XFS_DIFLAG_ANY) {
  739. if (di_flags & XFS_DIFLAG_REALTIME)
  740. flags |= XFS_XFLAG_REALTIME;
  741. if (di_flags & XFS_DIFLAG_PREALLOC)
  742. flags |= XFS_XFLAG_PREALLOC;
  743. if (di_flags & XFS_DIFLAG_IMMUTABLE)
  744. flags |= XFS_XFLAG_IMMUTABLE;
  745. if (di_flags & XFS_DIFLAG_APPEND)
  746. flags |= XFS_XFLAG_APPEND;
  747. if (di_flags & XFS_DIFLAG_SYNC)
  748. flags |= XFS_XFLAG_SYNC;
  749. if (di_flags & XFS_DIFLAG_NOATIME)
  750. flags |= XFS_XFLAG_NOATIME;
  751. if (di_flags & XFS_DIFLAG_NODUMP)
  752. flags |= XFS_XFLAG_NODUMP;
  753. if (di_flags & XFS_DIFLAG_RTINHERIT)
  754. flags |= XFS_XFLAG_RTINHERIT;
  755. if (di_flags & XFS_DIFLAG_PROJINHERIT)
  756. flags |= XFS_XFLAG_PROJINHERIT;
  757. if (di_flags & XFS_DIFLAG_NOSYMLINKS)
  758. flags |= XFS_XFLAG_NOSYMLINKS;
  759. if (di_flags & XFS_DIFLAG_EXTSIZE)
  760. flags |= XFS_XFLAG_EXTSIZE;
  761. if (di_flags & XFS_DIFLAG_EXTSZINHERIT)
  762. flags |= XFS_XFLAG_EXTSZINHERIT;
  763. if (di_flags & XFS_DIFLAG_NODEFRAG)
  764. flags |= XFS_XFLAG_NODEFRAG;
  765. if (di_flags & XFS_DIFLAG_FILESTREAM)
  766. flags |= XFS_XFLAG_FILESTREAM;
  767. }
  768. return flags;
  769. }
  770. uint
  771. xfs_ip2xflags(
  772. xfs_inode_t *ip)
  773. {
  774. xfs_dinode_core_t *dic = &ip->i_d;
  775. return _xfs_dic2xflags(dic->di_flags) |
  776. (XFS_CFORK_Q(dic) ? XFS_XFLAG_HASATTR : 0);
  777. }
  778. uint
  779. xfs_dic2xflags(
  780. xfs_dinode_core_t *dic)
  781. {
  782. return _xfs_dic2xflags(INT_GET(dic->di_flags, ARCH_CONVERT)) |
  783. (XFS_CFORK_Q_DISK(dic) ? XFS_XFLAG_HASATTR : 0);
  784. }
  785. /*
  786. * Given a mount structure and an inode number, return a pointer
  787. * to a newly allocated in-core inode corresponding to the given
  788. * inode number.
  789. *
  790. * Initialize the inode's attributes and extent pointers if it
  791. * already has them (it will not if the inode has no links).
  792. */
  793. int
  794. xfs_iread(
  795. xfs_mount_t *mp,
  796. xfs_trans_t *tp,
  797. xfs_ino_t ino,
  798. xfs_inode_t **ipp,
  799. xfs_daddr_t bno,
  800. uint imap_flags)
  801. {
  802. xfs_buf_t *bp;
  803. xfs_dinode_t *dip;
  804. xfs_inode_t *ip;
  805. int error;
  806. ASSERT(xfs_inode_zone != NULL);
  807. ip = kmem_zone_zalloc(xfs_inode_zone, KM_SLEEP);
  808. ip->i_ino = ino;
  809. ip->i_mount = mp;
  810. spin_lock_init(&ip->i_flags_lock);
  811. /*
  812. * Get pointer's to the on-disk inode and the buffer containing it.
  813. * If the inode number refers to a block outside the file system
  814. * then xfs_itobp() will return NULL. In this case we should
  815. * return NULL as well. Set i_blkno to 0 so that xfs_itobp() will
  816. * know that this is a new incore inode.
  817. */
  818. error = xfs_itobp(mp, tp, ip, &dip, &bp, bno, imap_flags);
  819. if (error) {
  820. kmem_zone_free(xfs_inode_zone, ip);
  821. return error;
  822. }
  823. /*
  824. * Initialize inode's trace buffers.
  825. * Do this before xfs_iformat in case it adds entries.
  826. */
  827. #ifdef XFS_BMAP_TRACE
  828. ip->i_xtrace = ktrace_alloc(XFS_BMAP_KTRACE_SIZE, KM_SLEEP);
  829. #endif
  830. #ifdef XFS_BMBT_TRACE
  831. ip->i_btrace = ktrace_alloc(XFS_BMBT_KTRACE_SIZE, KM_SLEEP);
  832. #endif
  833. #ifdef XFS_RW_TRACE
  834. ip->i_rwtrace = ktrace_alloc(XFS_RW_KTRACE_SIZE, KM_SLEEP);
  835. #endif
  836. #ifdef XFS_ILOCK_TRACE
  837. ip->i_lock_trace = ktrace_alloc(XFS_ILOCK_KTRACE_SIZE, KM_SLEEP);
  838. #endif
  839. #ifdef XFS_DIR2_TRACE
  840. ip->i_dir_trace = ktrace_alloc(XFS_DIR2_KTRACE_SIZE, KM_SLEEP);
  841. #endif
  842. /*
  843. * If we got something that isn't an inode it means someone
  844. * (nfs or dmi) has a stale handle.
  845. */
  846. if (INT_GET(dip->di_core.di_magic, ARCH_CONVERT) != XFS_DINODE_MAGIC) {
  847. kmem_zone_free(xfs_inode_zone, ip);
  848. xfs_trans_brelse(tp, bp);
  849. #ifdef DEBUG
  850. xfs_fs_cmn_err(CE_ALERT, mp, "xfs_iread: "
  851. "dip->di_core.di_magic (0x%x) != "
  852. "XFS_DINODE_MAGIC (0x%x)",
  853. INT_GET(dip->di_core.di_magic, ARCH_CONVERT),
  854. XFS_DINODE_MAGIC);
  855. #endif /* DEBUG */
  856. return XFS_ERROR(EINVAL);
  857. }
  858. /*
  859. * If the on-disk inode is already linked to a directory
  860. * entry, copy all of the inode into the in-core inode.
  861. * xfs_iformat() handles copying in the inode format
  862. * specific information.
  863. * Otherwise, just get the truly permanent information.
  864. */
  865. if (dip->di_core.di_mode) {
  866. xfs_xlate_dinode_core((xfs_caddr_t)&dip->di_core,
  867. &(ip->i_d), 1);
  868. error = xfs_iformat(ip, dip);
  869. if (error) {
  870. kmem_zone_free(xfs_inode_zone, ip);
  871. xfs_trans_brelse(tp, bp);
  872. #ifdef DEBUG
  873. xfs_fs_cmn_err(CE_ALERT, mp, "xfs_iread: "
  874. "xfs_iformat() returned error %d",
  875. error);
  876. #endif /* DEBUG */
  877. return error;
  878. }
  879. } else {
  880. ip->i_d.di_magic = INT_GET(dip->di_core.di_magic, ARCH_CONVERT);
  881. ip->i_d.di_version = INT_GET(dip->di_core.di_version, ARCH_CONVERT);
  882. ip->i_d.di_gen = INT_GET(dip->di_core.di_gen, ARCH_CONVERT);
  883. ip->i_d.di_flushiter = INT_GET(dip->di_core.di_flushiter, ARCH_CONVERT);
  884. /*
  885. * Make sure to pull in the mode here as well in
  886. * case the inode is released without being used.
  887. * This ensures that xfs_inactive() will see that
  888. * the inode is already free and not try to mess
  889. * with the uninitialized part of it.
  890. */
  891. ip->i_d.di_mode = 0;
  892. /*
  893. * Initialize the per-fork minima and maxima for a new
  894. * inode here. xfs_iformat will do it for old inodes.
  895. */
  896. ip->i_df.if_ext_max =
  897. XFS_IFORK_DSIZE(ip) / (uint)sizeof(xfs_bmbt_rec_t);
  898. }
  899. INIT_LIST_HEAD(&ip->i_reclaim);
  900. /*
  901. * The inode format changed when we moved the link count and
  902. * made it 32 bits long. If this is an old format inode,
  903. * convert it in memory to look like a new one. If it gets
  904. * flushed to disk we will convert back before flushing or
  905. * logging it. We zero out the new projid field and the old link
  906. * count field. We'll handle clearing the pad field (the remains
  907. * of the old uuid field) when we actually convert the inode to
  908. * the new format. We don't change the version number so that we
  909. * can distinguish this from a real new format inode.
  910. */
  911. if (ip->i_d.di_version == XFS_DINODE_VERSION_1) {
  912. ip->i_d.di_nlink = ip->i_d.di_onlink;
  913. ip->i_d.di_onlink = 0;
  914. ip->i_d.di_projid = 0;
  915. }
  916. ip->i_delayed_blks = 0;
  917. ip->i_size = ip->i_d.di_size;
  918. /*
  919. * Mark the buffer containing the inode as something to keep
  920. * around for a while. This helps to keep recently accessed
  921. * meta-data in-core longer.
  922. */
  923. XFS_BUF_SET_REF(bp, XFS_INO_REF);
  924. /*
  925. * Use xfs_trans_brelse() to release the buffer containing the
  926. * on-disk inode, because it was acquired with xfs_trans_read_buf()
  927. * in xfs_itobp() above. If tp is NULL, this is just a normal
  928. * brelse(). If we're within a transaction, then xfs_trans_brelse()
  929. * will only release the buffer if it is not dirty within the
  930. * transaction. It will be OK to release the buffer in this case,
  931. * because inodes on disk are never destroyed and we will be
  932. * locking the new in-core inode before putting it in the hash
  933. * table where other processes can find it. Thus we don't have
  934. * to worry about the inode being changed just because we released
  935. * the buffer.
  936. */
  937. xfs_trans_brelse(tp, bp);
  938. *ipp = ip;
  939. return 0;
  940. }
  941. /*
  942. * Read in extents from a btree-format inode.
  943. * Allocate and fill in if_extents. Real work is done in xfs_bmap.c.
  944. */
  945. int
  946. xfs_iread_extents(
  947. xfs_trans_t *tp,
  948. xfs_inode_t *ip,
  949. int whichfork)
  950. {
  951. int error;
  952. xfs_ifork_t *ifp;
  953. xfs_extnum_t nextents;
  954. size_t size;
  955. if (unlikely(XFS_IFORK_FORMAT(ip, whichfork) != XFS_DINODE_FMT_BTREE)) {
  956. XFS_ERROR_REPORT("xfs_iread_extents", XFS_ERRLEVEL_LOW,
  957. ip->i_mount);
  958. return XFS_ERROR(EFSCORRUPTED);
  959. }
  960. nextents = XFS_IFORK_NEXTENTS(ip, whichfork);
  961. size = nextents * sizeof(xfs_bmbt_rec_t);
  962. ifp = XFS_IFORK_PTR(ip, whichfork);
  963. /*
  964. * We know that the size is valid (it's checked in iformat_btree)
  965. */
  966. ifp->if_lastex = NULLEXTNUM;
  967. ifp->if_bytes = ifp->if_real_bytes = 0;
  968. ifp->if_flags |= XFS_IFEXTENTS;
  969. xfs_iext_add(ifp, 0, nextents);
  970. error = xfs_bmap_read_extents(tp, ip, whichfork);
  971. if (error) {
  972. xfs_iext_destroy(ifp);
  973. ifp->if_flags &= ~XFS_IFEXTENTS;
  974. return error;
  975. }
  976. xfs_validate_extents(ifp, nextents, 0, XFS_EXTFMT_INODE(ip));
  977. return 0;
  978. }
  979. /*
  980. * Allocate an inode on disk and return a copy of its in-core version.
  981. * The in-core inode is locked exclusively. Set mode, nlink, and rdev
  982. * appropriately within the inode. The uid and gid for the inode are
  983. * set according to the contents of the given cred structure.
  984. *
  985. * Use xfs_dialloc() to allocate the on-disk inode. If xfs_dialloc()
  986. * has a free inode available, call xfs_iget()
  987. * to obtain the in-core version of the allocated inode. Finally,
  988. * fill in the inode and log its initial contents. In this case,
  989. * ialloc_context would be set to NULL and call_again set to false.
  990. *
  991. * If xfs_dialloc() does not have an available inode,
  992. * it will replenish its supply by doing an allocation. Since we can
  993. * only do one allocation within a transaction without deadlocks, we
  994. * must commit the current transaction before returning the inode itself.
  995. * In this case, therefore, we will set call_again to true and return.
  996. * The caller should then commit the current transaction, start a new
  997. * transaction, and call xfs_ialloc() again to actually get the inode.
  998. *
  999. * To ensure that some other process does not grab the inode that
  1000. * was allocated during the first call to xfs_ialloc(), this routine
  1001. * also returns the [locked] bp pointing to the head of the freelist
  1002. * as ialloc_context. The caller should hold this buffer across
  1003. * the commit and pass it back into this routine on the second call.
  1004. */
  1005. int
  1006. xfs_ialloc(
  1007. xfs_trans_t *tp,
  1008. xfs_inode_t *pip,
  1009. mode_t mode,
  1010. xfs_nlink_t nlink,
  1011. xfs_dev_t rdev,
  1012. cred_t *cr,
  1013. xfs_prid_t prid,
  1014. int okalloc,
  1015. xfs_buf_t **ialloc_context,
  1016. boolean_t *call_again,
  1017. xfs_inode_t **ipp)
  1018. {
  1019. xfs_ino_t ino;
  1020. xfs_inode_t *ip;
  1021. bhv_vnode_t *vp;
  1022. uint flags;
  1023. int error;
  1024. /*
  1025. * Call the space management code to pick
  1026. * the on-disk inode to be allocated.
  1027. */
  1028. error = xfs_dialloc(tp, pip->i_ino, mode, okalloc,
  1029. ialloc_context, call_again, &ino);
  1030. if (error != 0) {
  1031. return error;
  1032. }
  1033. if (*call_again || ino == NULLFSINO) {
  1034. *ipp = NULL;
  1035. return 0;
  1036. }
  1037. ASSERT(*ialloc_context == NULL);
  1038. /*
  1039. * Get the in-core inode with the lock held exclusively.
  1040. * This is because we're setting fields here we need
  1041. * to prevent others from looking at until we're done.
  1042. */
  1043. error = xfs_trans_iget(tp->t_mountp, tp, ino,
  1044. XFS_IGET_CREATE, XFS_ILOCK_EXCL, &ip);
  1045. if (error != 0) {
  1046. return error;
  1047. }
  1048. ASSERT(ip != NULL);
  1049. vp = XFS_ITOV(ip);
  1050. ip->i_d.di_mode = (__uint16_t)mode;
  1051. ip->i_d.di_onlink = 0;
  1052. ip->i_d.di_nlink = nlink;
  1053. ASSERT(ip->i_d.di_nlink == nlink);
  1054. ip->i_d.di_uid = current_fsuid(cr);
  1055. ip->i_d.di_gid = current_fsgid(cr);
  1056. ip->i_d.di_projid = prid;
  1057. memset(&(ip->i_d.di_pad[0]), 0, sizeof(ip->i_d.di_pad));
  1058. /*
  1059. * If the superblock version is up to where we support new format
  1060. * inodes and this is currently an old format inode, then change
  1061. * the inode version number now. This way we only do the conversion
  1062. * here rather than here and in the flush/logging code.
  1063. */
  1064. if (XFS_SB_VERSION_HASNLINK(&tp->t_mountp->m_sb) &&
  1065. ip->i_d.di_version == XFS_DINODE_VERSION_1) {
  1066. ip->i_d.di_version = XFS_DINODE_VERSION_2;
  1067. /*
  1068. * We've already zeroed the old link count, the projid field,
  1069. * and the pad field.
  1070. */
  1071. }
  1072. /*
  1073. * Project ids won't be stored on disk if we are using a version 1 inode.
  1074. */
  1075. if ((prid != 0) && (ip->i_d.di_version == XFS_DINODE_VERSION_1))
  1076. xfs_bump_ino_vers2(tp, ip);
  1077. if (XFS_INHERIT_GID(pip, vp->v_vfsp)) {
  1078. ip->i_d.di_gid = pip->i_d.di_gid;
  1079. if ((pip->i_d.di_mode & S_ISGID) && (mode & S_IFMT) == S_IFDIR) {
  1080. ip->i_d.di_mode |= S_ISGID;
  1081. }
  1082. }
  1083. /*
  1084. * If the group ID of the new file does not match the effective group
  1085. * ID or one of the supplementary group IDs, the S_ISGID bit is cleared
  1086. * (and only if the irix_sgid_inherit compatibility variable is set).
  1087. */
  1088. if ((irix_sgid_inherit) &&
  1089. (ip->i_d.di_mode & S_ISGID) &&
  1090. (!in_group_p((gid_t)ip->i_d.di_gid))) {
  1091. ip->i_d.di_mode &= ~S_ISGID;
  1092. }
  1093. ip->i_d.di_size = 0;
  1094. ip->i_size = 0;
  1095. ip->i_d.di_nextents = 0;
  1096. ASSERT(ip->i_d.di_nblocks == 0);
  1097. xfs_ichgtime(ip, XFS_ICHGTIME_CHG|XFS_ICHGTIME_ACC|XFS_ICHGTIME_MOD);
  1098. /*
  1099. * di_gen will have been taken care of in xfs_iread.
  1100. */
  1101. ip->i_d.di_extsize = 0;
  1102. ip->i_d.di_dmevmask = 0;
  1103. ip->i_d.di_dmstate = 0;
  1104. ip->i_d.di_flags = 0;
  1105. flags = XFS_ILOG_CORE;
  1106. switch (mode & S_IFMT) {
  1107. case S_IFIFO:
  1108. case S_IFCHR:
  1109. case S_IFBLK:
  1110. case S_IFSOCK:
  1111. ip->i_d.di_format = XFS_DINODE_FMT_DEV;
  1112. ip->i_df.if_u2.if_rdev = rdev;
  1113. ip->i_df.if_flags = 0;
  1114. flags |= XFS_ILOG_DEV;
  1115. break;
  1116. case S_IFREG:
  1117. if (xfs_inode_is_filestream(pip)) {
  1118. error = xfs_filestream_associate(pip, ip);
  1119. if (error < 0)
  1120. return -error;
  1121. if (!error)
  1122. xfs_iflags_set(ip, XFS_IFILESTREAM);
  1123. }
  1124. /* fall through */
  1125. case S_IFDIR:
  1126. if (pip->i_d.di_flags & XFS_DIFLAG_ANY) {
  1127. uint di_flags = 0;
  1128. if ((mode & S_IFMT) == S_IFDIR) {
  1129. if (pip->i_d.di_flags & XFS_DIFLAG_RTINHERIT)
  1130. di_flags |= XFS_DIFLAG_RTINHERIT;
  1131. if (pip->i_d.di_flags & XFS_DIFLAG_EXTSZINHERIT) {
  1132. di_flags |= XFS_DIFLAG_EXTSZINHERIT;
  1133. ip->i_d.di_extsize = pip->i_d.di_extsize;
  1134. }
  1135. } else if ((mode & S_IFMT) == S_IFREG) {
  1136. if (pip->i_d.di_flags & XFS_DIFLAG_RTINHERIT) {
  1137. di_flags |= XFS_DIFLAG_REALTIME;
  1138. ip->i_iocore.io_flags |= XFS_IOCORE_RT;
  1139. }
  1140. if (pip->i_d.di_flags & XFS_DIFLAG_EXTSZINHERIT) {
  1141. di_flags |= XFS_DIFLAG_EXTSIZE;
  1142. ip->i_d.di_extsize = pip->i_d.di_extsize;
  1143. }
  1144. }
  1145. if ((pip->i_d.di_flags & XFS_DIFLAG_NOATIME) &&
  1146. xfs_inherit_noatime)
  1147. di_flags |= XFS_DIFLAG_NOATIME;
  1148. if ((pip->i_d.di_flags & XFS_DIFLAG_NODUMP) &&
  1149. xfs_inherit_nodump)
  1150. di_flags |= XFS_DIFLAG_NODUMP;
  1151. if ((pip->i_d.di_flags & XFS_DIFLAG_SYNC) &&
  1152. xfs_inherit_sync)
  1153. di_flags |= XFS_DIFLAG_SYNC;
  1154. if ((pip->i_d.di_flags & XFS_DIFLAG_NOSYMLINKS) &&
  1155. xfs_inherit_nosymlinks)
  1156. di_flags |= XFS_DIFLAG_NOSYMLINKS;
  1157. if (pip->i_d.di_flags & XFS_DIFLAG_PROJINHERIT)
  1158. di_flags |= XFS_DIFLAG_PROJINHERIT;
  1159. if ((pip->i_d.di_flags & XFS_DIFLAG_NODEFRAG) &&
  1160. xfs_inherit_nodefrag)
  1161. di_flags |= XFS_DIFLAG_NODEFRAG;
  1162. if (pip->i_d.di_flags & XFS_DIFLAG_FILESTREAM)
  1163. di_flags |= XFS_DIFLAG_FILESTREAM;
  1164. ip->i_d.di_flags |= di_flags;
  1165. }
  1166. /* FALLTHROUGH */
  1167. case S_IFLNK:
  1168. ip->i_d.di_format = XFS_DINODE_FMT_EXTENTS;
  1169. ip->i_df.if_flags = XFS_IFEXTENTS;
  1170. ip->i_df.if_bytes = ip->i_df.if_real_bytes = 0;
  1171. ip->i_df.if_u1.if_extents = NULL;
  1172. break;
  1173. default:
  1174. ASSERT(0);
  1175. }
  1176. /*
  1177. * Attribute fork settings for new inode.
  1178. */
  1179. ip->i_d.di_aformat = XFS_DINODE_FMT_EXTENTS;
  1180. ip->i_d.di_anextents = 0;
  1181. /*
  1182. * Log the new values stuffed into the inode.
  1183. */
  1184. xfs_trans_log_inode(tp, ip, flags);
  1185. /* now that we have an i_mode we can setup inode ops and unlock */
  1186. bhv_vfs_init_vnode(XFS_MTOVFS(tp->t_mountp), vp, XFS_ITOBHV(ip), 1);
  1187. *ipp = ip;
  1188. return 0;
  1189. }
  1190. /*
  1191. * Check to make sure that there are no blocks allocated to the
  1192. * file beyond the size of the file. We don't check this for
  1193. * files with fixed size extents or real time extents, but we
  1194. * at least do it for regular files.
  1195. */
  1196. #ifdef DEBUG
  1197. void
  1198. xfs_isize_check(
  1199. xfs_mount_t *mp,
  1200. xfs_inode_t *ip,
  1201. xfs_fsize_t isize)
  1202. {
  1203. xfs_fileoff_t map_first;
  1204. int nimaps;
  1205. xfs_bmbt_irec_t imaps[2];
  1206. if ((ip->i_d.di_mode & S_IFMT) != S_IFREG)
  1207. return;
  1208. if (ip->i_d.di_flags & (XFS_DIFLAG_REALTIME | XFS_DIFLAG_EXTSIZE))
  1209. return;
  1210. nimaps = 2;
  1211. map_first = XFS_B_TO_FSB(mp, (xfs_ufsize_t)isize);
  1212. /*
  1213. * The filesystem could be shutting down, so bmapi may return
  1214. * an error.
  1215. */
  1216. if (xfs_bmapi(NULL, ip, map_first,
  1217. (XFS_B_TO_FSB(mp,
  1218. (xfs_ufsize_t)XFS_MAXIOFFSET(mp)) -
  1219. map_first),
  1220. XFS_BMAPI_ENTIRE, NULL, 0, imaps, &nimaps,
  1221. NULL, NULL))
  1222. return;
  1223. ASSERT(nimaps == 1);
  1224. ASSERT(imaps[0].br_startblock == HOLESTARTBLOCK);
  1225. }
  1226. #endif /* DEBUG */
  1227. /*
  1228. * Calculate the last possible buffered byte in a file. This must
  1229. * include data that was buffered beyond the EOF by the write code.
  1230. * This also needs to deal with overflowing the xfs_fsize_t type
  1231. * which can happen for sizes near the limit.
  1232. *
  1233. * We also need to take into account any blocks beyond the EOF. It
  1234. * may be the case that they were buffered by a write which failed.
  1235. * In that case the pages will still be in memory, but the inode size
  1236. * will never have been updated.
  1237. */
  1238. xfs_fsize_t
  1239. xfs_file_last_byte(
  1240. xfs_inode_t *ip)
  1241. {
  1242. xfs_mount_t *mp;
  1243. xfs_fsize_t last_byte;
  1244. xfs_fileoff_t last_block;
  1245. xfs_fileoff_t size_last_block;
  1246. int error;
  1247. ASSERT(ismrlocked(&(ip->i_iolock), MR_UPDATE | MR_ACCESS));
  1248. mp = ip->i_mount;
  1249. /*
  1250. * Only check for blocks beyond the EOF if the extents have
  1251. * been read in. This eliminates the need for the inode lock,
  1252. * and it also saves us from looking when it really isn't
  1253. * necessary.
  1254. */
  1255. if (ip->i_df.if_flags & XFS_IFEXTENTS) {
  1256. error = xfs_bmap_last_offset(NULL, ip, &last_block,
  1257. XFS_DATA_FORK);
  1258. if (error) {
  1259. last_block = 0;
  1260. }
  1261. } else {
  1262. last_block = 0;
  1263. }
  1264. size_last_block = XFS_B_TO_FSB(mp, (xfs_ufsize_t)ip->i_size);
  1265. last_block = XFS_FILEOFF_MAX(last_block, size_last_block);
  1266. last_byte = XFS_FSB_TO_B(mp, last_block);
  1267. if (last_byte < 0) {
  1268. return XFS_MAXIOFFSET(mp);
  1269. }
  1270. last_byte += (1 << mp->m_writeio_log);
  1271. if (last_byte < 0) {
  1272. return XFS_MAXIOFFSET(mp);
  1273. }
  1274. return last_byte;
  1275. }
  1276. #if defined(XFS_RW_TRACE)
  1277. STATIC void
  1278. xfs_itrunc_trace(
  1279. int tag,
  1280. xfs_inode_t *ip,
  1281. int flag,
  1282. xfs_fsize_t new_size,
  1283. xfs_off_t toss_start,
  1284. xfs_off_t toss_finish)
  1285. {
  1286. if (ip->i_rwtrace == NULL) {
  1287. return;
  1288. }
  1289. ktrace_enter(ip->i_rwtrace,
  1290. (void*)((long)tag),
  1291. (void*)ip,
  1292. (void*)(unsigned long)((ip->i_d.di_size >> 32) & 0xffffffff),
  1293. (void*)(unsigned long)(ip->i_d.di_size & 0xffffffff),
  1294. (void*)((long)flag),
  1295. (void*)(unsigned long)((new_size >> 32) & 0xffffffff),
  1296. (void*)(unsigned long)(new_size & 0xffffffff),
  1297. (void*)(unsigned long)((toss_start >> 32) & 0xffffffff),
  1298. (void*)(unsigned long)(toss_start & 0xffffffff),
  1299. (void*)(unsigned long)((toss_finish >> 32) & 0xffffffff),
  1300. (void*)(unsigned long)(toss_finish & 0xffffffff),
  1301. (void*)(unsigned long)current_cpu(),
  1302. (void*)(unsigned long)current_pid(),
  1303. (void*)NULL,
  1304. (void*)NULL,
  1305. (void*)NULL);
  1306. }
  1307. #else
  1308. #define xfs_itrunc_trace(tag, ip, flag, new_size, toss_start, toss_finish)
  1309. #endif
  1310. /*
  1311. * Start the truncation of the file to new_size. The new size
  1312. * must be smaller than the current size. This routine will
  1313. * clear the buffer and page caches of file data in the removed
  1314. * range, and xfs_itruncate_finish() will remove the underlying
  1315. * disk blocks.
  1316. *
  1317. * The inode must have its I/O lock locked EXCLUSIVELY, and it
  1318. * must NOT have the inode lock held at all. This is because we're
  1319. * calling into the buffer/page cache code and we can't hold the
  1320. * inode lock when we do so.
  1321. *
  1322. * We need to wait for any direct I/Os in flight to complete before we
  1323. * proceed with the truncate. This is needed to prevent the extents
  1324. * being read or written by the direct I/Os from being removed while the
  1325. * I/O is in flight as there is no other method of synchronising
  1326. * direct I/O with the truncate operation. Also, because we hold
  1327. * the IOLOCK in exclusive mode, we prevent new direct I/Os from being
  1328. * started until the truncate completes and drops the lock. Essentially,
  1329. * the vn_iowait() call forms an I/O barrier that provides strict ordering
  1330. * between direct I/Os and the truncate operation.
  1331. *
  1332. * The flags parameter can have either the value XFS_ITRUNC_DEFINITE
  1333. * or XFS_ITRUNC_MAYBE. The XFS_ITRUNC_MAYBE value should be used
  1334. * in the case that the caller is locking things out of order and
  1335. * may not be able to call xfs_itruncate_finish() with the inode lock
  1336. * held without dropping the I/O lock. If the caller must drop the
  1337. * I/O lock before calling xfs_itruncate_finish(), then xfs_itruncate_start()
  1338. * must be called again with all the same restrictions as the initial
  1339. * call.
  1340. */
  1341. int
  1342. xfs_itruncate_start(
  1343. xfs_inode_t *ip,
  1344. uint flags,
  1345. xfs_fsize_t new_size)
  1346. {
  1347. xfs_fsize_t last_byte;
  1348. xfs_off_t toss_start;
  1349. xfs_mount_t *mp;
  1350. bhv_vnode_t *vp;
  1351. int error = 0;
  1352. ASSERT(ismrlocked(&ip->i_iolock, MR_UPDATE) != 0);
  1353. ASSERT((new_size == 0) || (new_size <= ip->i_size));
  1354. ASSERT((flags == XFS_ITRUNC_DEFINITE) ||
  1355. (flags == XFS_ITRUNC_MAYBE));
  1356. mp = ip->i_mount;
  1357. vp = XFS_ITOV(ip);
  1358. vn_iowait(vp); /* wait for the completion of any pending DIOs */
  1359. /*
  1360. * Call toss_pages or flushinval_pages to get rid of pages
  1361. * overlapping the region being removed. We have to use
  1362. * the less efficient flushinval_pages in the case that the
  1363. * caller may not be able to finish the truncate without
  1364. * dropping the inode's I/O lock. Make sure
  1365. * to catch any pages brought in by buffers overlapping
  1366. * the EOF by searching out beyond the isize by our
  1367. * block size. We round new_size up to a block boundary
  1368. * so that we don't toss things on the same block as
  1369. * new_size but before it.
  1370. *
  1371. * Before calling toss_page or flushinval_pages, make sure to
  1372. * call remapf() over the same region if the file is mapped.
  1373. * This frees up mapped file references to the pages in the
  1374. * given range and for the flushinval_pages case it ensures
  1375. * that we get the latest mapped changes flushed out.
  1376. */
  1377. toss_start = XFS_B_TO_FSB(mp, (xfs_ufsize_t)new_size);
  1378. toss_start = XFS_FSB_TO_B(mp, toss_start);
  1379. if (toss_start < 0) {
  1380. /*
  1381. * The place to start tossing is beyond our maximum
  1382. * file size, so there is no way that the data extended
  1383. * out there.
  1384. */
  1385. return 0;
  1386. }
  1387. last_byte = xfs_file_last_byte(ip);
  1388. xfs_itrunc_trace(XFS_ITRUNC_START, ip, flags, new_size, toss_start,
  1389. last_byte);
  1390. if (last_byte > toss_start) {
  1391. if (flags & XFS_ITRUNC_DEFINITE) {
  1392. bhv_vop_toss_pages(vp, toss_start, -1, FI_REMAPF_LOCKED);
  1393. } else {
  1394. error = bhv_vop_flushinval_pages(vp, toss_start, -1, FI_REMAPF_LOCKED);
  1395. }
  1396. }
  1397. #ifdef DEBUG
  1398. if (new_size == 0) {
  1399. ASSERT(VN_CACHED(vp) == 0);
  1400. }
  1401. #endif
  1402. return error;
  1403. }
  1404. /*
  1405. * Shrink the file to the given new_size. The new
  1406. * size must be smaller than the current size.
  1407. * This will free up the underlying blocks
  1408. * in the removed range after a call to xfs_itruncate_start()
  1409. * or xfs_atruncate_start().
  1410. *
  1411. * The transaction passed to this routine must have made
  1412. * a permanent log reservation of at least XFS_ITRUNCATE_LOG_RES.
  1413. * This routine may commit the given transaction and
  1414. * start new ones, so make sure everything involved in
  1415. * the transaction is tidy before calling here.
  1416. * Some transaction will be returned to the caller to be
  1417. * committed. The incoming transaction must already include
  1418. * the inode, and both inode locks must be held exclusively.
  1419. * The inode must also be "held" within the transaction. On
  1420. * return the inode will be "held" within the returned transaction.
  1421. * This routine does NOT require any disk space to be reserved
  1422. * for it within the transaction.
  1423. *
  1424. * The fork parameter must be either xfs_attr_fork or xfs_data_fork,
  1425. * and it indicates the fork which is to be truncated. For the
  1426. * attribute fork we only support truncation to size 0.
  1427. *
  1428. * We use the sync parameter to indicate whether or not the first
  1429. * transaction we perform might have to be synchronous. For the attr fork,
  1430. * it needs to be so if the unlink of the inode is not yet known to be
  1431. * permanent in the log. This keeps us from freeing and reusing the
  1432. * blocks of the attribute fork before the unlink of the inode becomes
  1433. * permanent.
  1434. *
  1435. * For the data fork, we normally have to run synchronously if we're
  1436. * being called out of the inactive path or we're being called
  1437. * out of the create path where we're truncating an existing file.
  1438. * Either way, the truncate needs to be sync so blocks don't reappear
  1439. * in the file with altered data in case of a crash. wsync filesystems
  1440. * can run the first case async because anything that shrinks the inode
  1441. * has to run sync so by the time we're called here from inactive, the
  1442. * inode size is permanently set to 0.
  1443. *
  1444. * Calls from the truncate path always need to be sync unless we're
  1445. * in a wsync filesystem and the file has already been unlinked.
  1446. *
  1447. * The caller is responsible for correctly setting the sync parameter.
  1448. * It gets too hard for us to guess here which path we're being called
  1449. * out of just based on inode state.
  1450. */
  1451. int
  1452. xfs_itruncate_finish(
  1453. xfs_trans_t **tp,
  1454. xfs_inode_t *ip,
  1455. xfs_fsize_t new_size,
  1456. int fork,
  1457. int sync)
  1458. {
  1459. xfs_fsblock_t first_block;
  1460. xfs_fileoff_t first_unmap_block;
  1461. xfs_fileoff_t last_block;
  1462. xfs_filblks_t unmap_len=0;
  1463. xfs_mount_t *mp;
  1464. xfs_trans_t *ntp;
  1465. int done;
  1466. int committed;
  1467. xfs_bmap_free_t free_list;
  1468. int error;
  1469. ASSERT(ismrlocked(&ip->i_iolock, MR_UPDATE) != 0);
  1470. ASSERT(ismrlocked(&ip->i_lock, MR_UPDATE) != 0);
  1471. ASSERT((new_size == 0) || (new_size <= ip->i_size));
  1472. ASSERT(*tp != NULL);
  1473. ASSERT((*tp)->t_flags & XFS_TRANS_PERM_LOG_RES);
  1474. ASSERT(ip->i_transp == *tp);
  1475. ASSERT(ip->i_itemp != NULL);
  1476. ASSERT(ip->i_itemp->ili_flags & XFS_ILI_HOLD);
  1477. ntp = *tp;
  1478. mp = (ntp)->t_mountp;
  1479. ASSERT(! XFS_NOT_DQATTACHED(mp, ip));
  1480. /*
  1481. * We only support truncating the entire attribute fork.
  1482. */
  1483. if (fork == XFS_ATTR_FORK) {
  1484. new_size = 0LL;
  1485. }
  1486. first_unmap_block = XFS_B_TO_FSB(mp, (xfs_ufsize_t)new_size);
  1487. xfs_itrunc_trace(XFS_ITRUNC_FINISH1, ip, 0, new_size, 0, 0);
  1488. /*
  1489. * The first thing we do is set the size to new_size permanently
  1490. * on disk. This way we don't have to worry about anyone ever
  1491. * being able to look at the data being freed even in the face
  1492. * of a crash. What we're getting around here is the case where
  1493. * we free a block, it is allocated to another file, it is written
  1494. * to, and then we crash. If the new data gets written to the
  1495. * file but the log buffers containing the free and reallocation
  1496. * don't, then we'd end up with garbage in the blocks being freed.
  1497. * As long as we make the new_size permanent before actually
  1498. * freeing any blocks it doesn't matter if they get writtten to.
  1499. *
  1500. * The callers must signal into us whether or not the size
  1501. * setting here must be synchronous. There are a few cases
  1502. * where it doesn't have to be synchronous. Those cases
  1503. * occur if the file is unlinked and we know the unlink is
  1504. * permanent or if the blocks being truncated are guaranteed
  1505. * to be beyond the inode eof (regardless of the link count)
  1506. * and the eof value is permanent. Both of these cases occur
  1507. * only on wsync-mounted filesystems. In those cases, we're
  1508. * guaranteed that no user will ever see the data in the blocks
  1509. * that are being truncated so the truncate can run async.
  1510. * In the free beyond eof case, the file may wind up with
  1511. * more blocks allocated to it than it needs if we crash
  1512. * and that won't get fixed until the next time the file
  1513. * is re-opened and closed but that's ok as that shouldn't
  1514. * be too many blocks.
  1515. *
  1516. * However, we can't just make all wsync xactions run async
  1517. * because there's one call out of the create path that needs
  1518. * to run sync where it's truncating an existing file to size
  1519. * 0 whose size is > 0.
  1520. *
  1521. * It's probably possible to come up with a test in this
  1522. * routine that would correctly distinguish all the above
  1523. * cases from the values of the function parameters and the
  1524. * inode state but for sanity's sake, I've decided to let the
  1525. * layers above just tell us. It's simpler to correctly figure
  1526. * out in the layer above exactly under what conditions we
  1527. * can run async and I think it's easier for others read and
  1528. * follow the logic in case something has to be changed.
  1529. * cscope is your friend -- rcc.
  1530. *
  1531. * The attribute fork is much simpler.
  1532. *
  1533. * For the attribute fork we allow the caller to tell us whether
  1534. * the unlink of the inode that led to this call is yet permanent
  1535. * in the on disk log. If it is not and we will be freeing extents
  1536. * in this inode then we make the first transaction synchronous
  1537. * to make sure that the unlink is permanent by the time we free
  1538. * the blocks.
  1539. */
  1540. if (fork == XFS_DATA_FORK) {
  1541. if (ip->i_d.di_nextents > 0) {
  1542. /*
  1543. * If we are not changing the file size then do
  1544. * not update the on-disk file size - we may be
  1545. * called from xfs_inactive_free_eofblocks(). If we
  1546. * update the on-disk file size and then the system
  1547. * crashes before the contents of the file are
  1548. * flushed to disk then the files may be full of
  1549. * holes (ie NULL files bug).
  1550. */
  1551. if (ip->i_size != new_size) {
  1552. ip->i_d.di_size = new_size;
  1553. ip->i_size = new_size;
  1554. xfs_trans_log_inode(ntp, ip, XFS_ILOG_CORE);
  1555. }
  1556. }
  1557. } else if (sync) {
  1558. ASSERT(!(mp->m_flags & XFS_MOUNT_WSYNC));
  1559. if (ip->i_d.di_anextents > 0)
  1560. xfs_trans_set_sync(ntp);
  1561. }
  1562. ASSERT(fork == XFS_DATA_FORK ||
  1563. (fork == XFS_ATTR_FORK &&
  1564. ((sync && !(mp->m_flags & XFS_MOUNT_WSYNC)) ||
  1565. (sync == 0 && (mp->m_flags & XFS_MOUNT_WSYNC)))));
  1566. /*
  1567. * Since it is possible for space to become allocated beyond
  1568. * the end of the file (in a crash where the space is allocated
  1569. * but the inode size is not yet updated), simply remove any
  1570. * blocks which show up between the new EOF and the maximum
  1571. * possible file size. If the first block to be removed is
  1572. * beyond the maximum file size (ie it is the same as last_block),
  1573. * then there is nothing to do.
  1574. */
  1575. last_block = XFS_B_TO_FSB(mp, (xfs_ufsize_t)XFS_MAXIOFFSET(mp));
  1576. ASSERT(first_unmap_block <= last_block);
  1577. done = 0;
  1578. if (last_block == first_unmap_block) {
  1579. done = 1;
  1580. } else {
  1581. unmap_len = last_block - first_unmap_block + 1;
  1582. }
  1583. while (!done) {
  1584. /*
  1585. * Free up up to XFS_ITRUNC_MAX_EXTENTS. xfs_bunmapi()
  1586. * will tell us whether it freed the entire range or
  1587. * not. If this is a synchronous mount (wsync),
  1588. * then we can tell bunmapi to keep all the
  1589. * transactions asynchronous since the unlink
  1590. * transaction that made this inode inactive has
  1591. * already hit the disk. There's no danger of
  1592. * the freed blocks being reused, there being a
  1593. * crash, and the reused blocks suddenly reappearing
  1594. * in this file with garbage in them once recovery
  1595. * runs.
  1596. */
  1597. XFS_BMAP_INIT(&free_list, &first_block);
  1598. error = XFS_BUNMAPI(mp, ntp, &ip->i_iocore,
  1599. first_unmap_block, unmap_len,
  1600. XFS_BMAPI_AFLAG(fork) |
  1601. (sync ? 0 : XFS_BMAPI_ASYNC),
  1602. XFS_ITRUNC_MAX_EXTENTS,
  1603. &first_block, &free_list,
  1604. NULL, &done);
  1605. if (error) {
  1606. /*
  1607. * If the bunmapi call encounters an error,
  1608. * return to the caller where the transaction
  1609. * can be properly aborted. We just need to
  1610. * make sure we're not holding any resources
  1611. * that we were not when we came in.
  1612. */
  1613. xfs_bmap_cancel(&free_list);
  1614. return error;
  1615. }
  1616. /*
  1617. * Duplicate the transaction that has the permanent
  1618. * reservation and commit the old transaction.
  1619. */
  1620. error = xfs_bmap_finish(tp, &free_list, &committed);
  1621. ntp = *tp;
  1622. if (error) {
  1623. /*
  1624. * If the bmap finish call encounters an error,
  1625. * return to the caller where the transaction
  1626. * can be properly aborted. We just need to
  1627. * make sure we're not holding any resources
  1628. * that we were not when we came in.
  1629. *
  1630. * Aborting from this point might lose some
  1631. * blocks in the file system, but oh well.
  1632. */
  1633. xfs_bmap_cancel(&free_list);
  1634. if (committed) {
  1635. /*
  1636. * If the passed in transaction committed
  1637. * in xfs_bmap_finish(), then we want to
  1638. * add the inode to this one before returning.
  1639. * This keeps things simple for the higher
  1640. * level code, because it always knows that
  1641. * the inode is locked and held in the
  1642. * transaction that returns to it whether
  1643. * errors occur or not. We don't mark the
  1644. * inode dirty so that this transaction can
  1645. * be easily aborted if possible.
  1646. */
  1647. xfs_trans_ijoin(ntp, ip,
  1648. XFS_ILOCK_EXCL | XFS_IOLOCK_EXCL);
  1649. xfs_trans_ihold(ntp, ip);
  1650. }
  1651. return error;
  1652. }
  1653. if (committed) {
  1654. /*
  1655. * The first xact was committed,
  1656. * so add the inode to the new one.
  1657. * Mark it dirty so it will be logged
  1658. * and moved forward in the log as
  1659. * part of every commit.
  1660. */
  1661. xfs_trans_ijoin(ntp, ip,
  1662. XFS_ILOCK_EXCL | XFS_IOLOCK_EXCL);
  1663. xfs_trans_ihold(ntp, ip);
  1664. xfs_trans_log_inode(ntp, ip, XFS_ILOG_CORE);
  1665. }
  1666. ntp = xfs_trans_dup(ntp);
  1667. (void) xfs_trans_commit(*tp, 0);
  1668. *tp = ntp;
  1669. error = xfs_trans_reserve(ntp, 0, XFS_ITRUNCATE_LOG_RES(mp), 0,
  1670. XFS_TRANS_PERM_LOG_RES,
  1671. XFS_ITRUNCATE_LOG_COUNT);
  1672. /*
  1673. * Add the inode being truncated to the next chained
  1674. * transaction.
  1675. */
  1676. xfs_trans_ijoin(ntp, ip, XFS_ILOCK_EXCL | XFS_IOLOCK_EXCL);
  1677. xfs_trans_ihold(ntp, ip);
  1678. if (error)
  1679. return (error);
  1680. }
  1681. /*
  1682. * Only update the size in the case of the data fork, but
  1683. * always re-log the inode so that our permanent transaction
  1684. * can keep on rolling it forward in the log.
  1685. */
  1686. if (fork == XFS_DATA_FORK) {
  1687. xfs_isize_check(mp, ip, new_size);
  1688. /*
  1689. * If we are not changing the file size then do
  1690. * not update the on-disk file size - we may be
  1691. * called from xfs_inactive_free_eofblocks(). If we
  1692. * update the on-disk file size and then the system
  1693. * crashes before the contents of the file are
  1694. * flushed to disk then the files may be full of
  1695. * holes (ie NULL files bug).
  1696. */
  1697. if (ip->i_size != new_size) {
  1698. ip->i_d.di_size = new_size;
  1699. ip->i_size = new_size;
  1700. }
  1701. }
  1702. xfs_trans_log_inode(ntp, ip, XFS_ILOG_CORE);
  1703. ASSERT((new_size != 0) ||
  1704. (fork == XFS_ATTR_FORK) ||
  1705. (ip->i_delayed_blks == 0));
  1706. ASSERT((new_size != 0) ||
  1707. (fork == XFS_ATTR_FORK) ||
  1708. (ip->i_d.di_nextents == 0));
  1709. xfs_itrunc_trace(XFS_ITRUNC_FINISH2, ip, 0, new_size, 0, 0);
  1710. return 0;
  1711. }
  1712. /*
  1713. * xfs_igrow_start
  1714. *
  1715. * Do the first part of growing a file: zero any data in the last
  1716. * block that is beyond the old EOF. We need to do this before
  1717. * the inode is joined to the transaction to modify the i_size.
  1718. * That way we can drop the inode lock and call into the buffer
  1719. * cache to get the buffer mapping the EOF.
  1720. */
  1721. int
  1722. xfs_igrow_start(
  1723. xfs_inode_t *ip,
  1724. xfs_fsize_t new_size,
  1725. cred_t *credp)
  1726. {
  1727. int error;
  1728. ASSERT(ismrlocked(&(ip->i_lock), MR_UPDATE) != 0);
  1729. ASSERT(ismrlocked(&(ip->i_iolock), MR_UPDATE) != 0);
  1730. ASSERT(new_size > ip->i_size);
  1731. /*
  1732. * Zero any pages that may have been created by
  1733. * xfs_write_file() beyond the end of the file
  1734. * and any blocks between the old and new file sizes.
  1735. */
  1736. error = xfs_zero_eof(XFS_ITOV(ip), &ip->i_iocore, new_size,
  1737. ip->i_size);
  1738. return error;
  1739. }
  1740. /*
  1741. * xfs_igrow_finish
  1742. *
  1743. * This routine is called to extend the size of a file.
  1744. * The inode must have both the iolock and the ilock locked
  1745. * for update and it must be a part of the current transaction.
  1746. * The xfs_igrow_start() function must have been called previously.
  1747. * If the change_flag is not zero, the inode change timestamp will
  1748. * be updated.
  1749. */
  1750. void
  1751. xfs_igrow_finish(
  1752. xfs_trans_t *tp,
  1753. xfs_inode_t *ip,
  1754. xfs_fsize_t new_size,
  1755. int change_flag)
  1756. {
  1757. ASSERT(ismrlocked(&(ip->i_lock), MR_UPDATE) != 0);
  1758. ASSERT(ismrlocked(&(ip->i_iolock), MR_UPDATE) != 0);
  1759. ASSERT(ip->i_transp == tp);
  1760. ASSERT(new_size > ip->i_size);
  1761. /*
  1762. * Update the file size. Update the inode change timestamp
  1763. * if change_flag set.
  1764. */
  1765. ip->i_d.di_size = new_size;
  1766. ip->i_size = new_size;
  1767. if (change_flag)
  1768. xfs_ichgtime(ip, XFS_ICHGTIME_CHG);
  1769. xfs_trans_log_inode(tp, ip, XFS_ILOG_CORE);
  1770. }
  1771. /*
  1772. * This is called when the inode's link count goes to 0.
  1773. * We place the on-disk inode on a list in the AGI. It
  1774. * will be pulled from this list when the inode is freed.
  1775. */
  1776. int
  1777. xfs_iunlink(
  1778. xfs_trans_t *tp,
  1779. xfs_inode_t *ip)
  1780. {
  1781. xfs_mount_t *mp;
  1782. xfs_agi_t *agi;
  1783. xfs_dinode_t *dip;
  1784. xfs_buf_t *agibp;
  1785. xfs_buf_t *ibp;
  1786. xfs_agnumber_t agno;
  1787. xfs_daddr_t agdaddr;
  1788. xfs_agino_t agino;
  1789. short bucket_index;
  1790. int offset;
  1791. int error;
  1792. int agi_ok;
  1793. ASSERT(ip->i_d.di_nlink == 0);
  1794. ASSERT(ip->i_d.di_mode != 0);
  1795. ASSERT(ip->i_transp == tp);
  1796. mp = tp->t_mountp;
  1797. agno = XFS_INO_TO_AGNO(mp, ip->i_ino);
  1798. agdaddr = XFS_AG_DADDR(mp, agno, XFS_AGI_DADDR(mp));
  1799. /*
  1800. * Get the agi buffer first. It ensures lock ordering
  1801. * on the list.
  1802. */
  1803. error = xfs_trans_read_buf(mp, tp, mp->m_ddev_targp, agdaddr,
  1804. XFS_FSS_TO_BB(mp, 1), 0, &agibp);
  1805. if (error) {
  1806. return error;
  1807. }
  1808. /*
  1809. * Validate the magic number of the agi block.
  1810. */
  1811. agi = XFS_BUF_TO_AGI(agibp);
  1812. agi_ok =
  1813. be32_to_cpu(agi->agi_magicnum) == XFS_AGI_MAGIC &&
  1814. XFS_AGI_GOOD_VERSION(be32_to_cpu(agi->agi_versionnum));
  1815. if (unlikely(XFS_TEST_ERROR(!agi_ok, mp, XFS_ERRTAG_IUNLINK,
  1816. XFS_RANDOM_IUNLINK))) {
  1817. XFS_CORRUPTION_ERROR("xfs_iunlink", XFS_ERRLEVEL_LOW, mp, agi);
  1818. xfs_trans_brelse(tp, agibp);
  1819. return XFS_ERROR(EFSCORRUPTED);
  1820. }
  1821. /*
  1822. * Get the index into the agi hash table for the
  1823. * list this inode will go on.
  1824. */
  1825. agino = XFS_INO_TO_AGINO(mp, ip->i_ino);
  1826. ASSERT(agino != 0);
  1827. bucket_index = agino % XFS_AGI_UNLINKED_BUCKETS;
  1828. ASSERT(agi->agi_unlinked[bucket_index]);
  1829. ASSERT(be32_to_cpu(agi->agi_unlinked[bucket_index]) != agino);
  1830. if (be32_to_cpu(agi->agi_unlinked[bucket_index]) != NULLAGINO) {
  1831. /*
  1832. * There is already another inode in the bucket we need
  1833. * to add ourselves to. Add us at the front of the list.
  1834. * Here we put the head pointer into our next pointer,
  1835. * and then we fall through to point the head at us.
  1836. */
  1837. error = xfs_itobp(mp, tp, ip, &dip, &ibp, 0, 0);
  1838. if (error) {
  1839. return error;
  1840. }
  1841. ASSERT(INT_GET(dip->di_next_unlinked, ARCH_CONVERT) == NULLAGINO);
  1842. ASSERT(dip->di_next_unlinked);
  1843. /* both on-disk, don't endian flip twice */
  1844. dip->di_next_unlinked = agi->agi_unlinked[bucket_index];
  1845. offset = ip->i_boffset +
  1846. offsetof(xfs_dinode_t, di_next_unlinked);
  1847. xfs_trans_inode_buf(tp, ibp);
  1848. xfs_trans_log_buf(tp, ibp, offset,
  1849. (offset + sizeof(xfs_agino_t) - 1));
  1850. xfs_inobp_check(mp, ibp);
  1851. }
  1852. /*
  1853. * Point the bucket head pointer at the inode being inserted.
  1854. */
  1855. ASSERT(agino != 0);
  1856. agi->agi_unlinked[bucket_index] = cpu_to_be32(agino);
  1857. offset = offsetof(xfs_agi_t, agi_unlinked) +
  1858. (sizeof(xfs_agino_t) * bucket_index);
  1859. xfs_trans_log_buf(tp, agibp, offset,
  1860. (offset + sizeof(xfs_agino_t) - 1));
  1861. return 0;
  1862. }
  1863. /*
  1864. * Pull the on-disk inode from the AGI unlinked list.
  1865. */
  1866. STATIC int
  1867. xfs_iunlink_remove(
  1868. xfs_trans_t *tp,
  1869. xfs_inode_t *ip)
  1870. {
  1871. xfs_ino_t next_ino;
  1872. xfs_mount_t *mp;
  1873. xfs_agi_t *agi;
  1874. xfs_dinode_t *dip;
  1875. xfs_buf_t *agibp;
  1876. xfs_buf_t *ibp;
  1877. xfs_agnumber_t agno;
  1878. xfs_daddr_t agdaddr;
  1879. xfs_agino_t agino;
  1880. xfs_agino_t next_agino;
  1881. xfs_buf_t *last_ibp;
  1882. xfs_dinode_t *last_dip = NULL;
  1883. short bucket_index;
  1884. int offset, last_offset = 0;
  1885. int error;
  1886. int agi_ok;
  1887. /*
  1888. * First pull the on-disk inode from the AGI unlinked list.
  1889. */
  1890. mp = tp->t_mountp;
  1891. agno = XFS_INO_TO_AGNO(mp, ip->i_ino);
  1892. agdaddr = XFS_AG_DADDR(mp, agno, XFS_AGI_DADDR(mp));
  1893. /*
  1894. * Get the agi buffer first. It ensures lock ordering
  1895. * on the list.
  1896. */
  1897. error = xfs_trans_read_buf(mp, tp, mp->m_ddev_targp, agdaddr,
  1898. XFS_FSS_TO_BB(mp, 1), 0, &agibp);
  1899. if (error) {
  1900. cmn_err(CE_WARN,
  1901. "xfs_iunlink_remove: xfs_trans_read_buf() returned an error %d on %s. Returning error.",
  1902. error, mp->m_fsname);
  1903. return error;
  1904. }
  1905. /*
  1906. * Validate the magic number of the agi block.
  1907. */
  1908. agi = XFS_BUF_TO_AGI(agibp);
  1909. agi_ok =
  1910. be32_to_cpu(agi->agi_magicnum) == XFS_AGI_MAGIC &&
  1911. XFS_AGI_GOOD_VERSION(be32_to_cpu(agi->agi_versionnum));
  1912. if (unlikely(XFS_TEST_ERROR(!agi_ok, mp, XFS_ERRTAG_IUNLINK_REMOVE,
  1913. XFS_RANDOM_IUNLINK_REMOVE))) {
  1914. XFS_CORRUPTION_ERROR("xfs_iunlink_remove", XFS_ERRLEVEL_LOW,
  1915. mp, agi);
  1916. xfs_trans_brelse(tp, agibp);
  1917. cmn_err(CE_WARN,
  1918. "xfs_iunlink_remove: XFS_TEST_ERROR() returned an error on %s. Returning EFSCORRUPTED.",
  1919. mp->m_fsname);
  1920. return XFS_ERROR(EFSCORRUPTED);
  1921. }
  1922. /*
  1923. * Get the index into the agi hash table for the
  1924. * list this inode will go on.
  1925. */
  1926. agino = XFS_INO_TO_AGINO(mp, ip->i_ino);
  1927. ASSERT(agino != 0);
  1928. bucket_index = agino % XFS_AGI_UNLINKED_BUCKETS;
  1929. ASSERT(be32_to_cpu(agi->agi_unlinked[bucket_index]) != NULLAGINO);
  1930. ASSERT(agi->agi_unlinked[bucket_index]);
  1931. if (be32_to_cpu(agi->agi_unlinked[bucket_index]) == agino) {
  1932. /*
  1933. * We're at the head of the list. Get the inode's
  1934. * on-disk buffer to see if there is anyone after us
  1935. * on the list. Only modify our next pointer if it
  1936. * is not already NULLAGINO. This saves us the overhead
  1937. * of dealing with the buffer when there is no need to
  1938. * change it.
  1939. */
  1940. error = xfs_itobp(mp, tp, ip, &dip, &ibp, 0, 0);
  1941. if (error) {
  1942. cmn_err(CE_WARN,
  1943. "xfs_iunlink_remove: xfs_itobp() returned an error %d on %s. Returning error.",
  1944. error, mp->m_fsname);
  1945. return error;
  1946. }
  1947. next_agino = INT_GET(dip->di_next_unlinked, ARCH_CONVERT);
  1948. ASSERT(next_agino != 0);
  1949. if (next_agino != NULLAGINO) {
  1950. INT_SET(dip->di_next_unlinked, ARCH_CONVERT, NULLAGINO);
  1951. offset = ip->i_boffset +
  1952. offsetof(xfs_dinode_t, di_next_unlinked);
  1953. xfs_trans_inode_buf(tp, ibp);
  1954. xfs_trans_log_buf(tp, ibp, offset,
  1955. (offset + sizeof(xfs_agino_t) - 1));
  1956. xfs_inobp_check(mp, ibp);
  1957. } else {
  1958. xfs_trans_brelse(tp, ibp);
  1959. }
  1960. /*
  1961. * Point the bucket head pointer at the next inode.
  1962. */
  1963. ASSERT(next_agino != 0);
  1964. ASSERT(next_agino != agino);
  1965. agi->agi_unlinked[bucket_index] = cpu_to_be32(next_agino);
  1966. offset = offsetof(xfs_agi_t, agi_unlinked) +
  1967. (sizeof(xfs_agino_t) * bucket_index);
  1968. xfs_trans_log_buf(tp, agibp, offset,
  1969. (offset + sizeof(xfs_agino_t) - 1));
  1970. } else {
  1971. /*
  1972. * We need to search the list for the inode being freed.
  1973. */
  1974. next_agino = be32_to_cpu(agi->agi_unlinked[bucket_index]);
  1975. last_ibp = NULL;
  1976. while (next_agino != agino) {
  1977. /*
  1978. * If the last inode wasn't the one pointing to
  1979. * us, then release its buffer since we're not
  1980. * going to do anything with it.
  1981. */
  1982. if (last_ibp != NULL) {
  1983. xfs_trans_brelse(tp, last_ibp);
  1984. }
  1985. next_ino = XFS_AGINO_TO_INO(mp, agno, next_agino);
  1986. error = xfs_inotobp(mp, tp, next_ino, &last_dip,
  1987. &last_ibp, &last_offset);
  1988. if (error) {
  1989. cmn_err(CE_WARN,
  1990. "xfs_iunlink_remove: xfs_inotobp() returned an error %d on %s. Returning error.",
  1991. error, mp->m_fsname);
  1992. return error;
  1993. }
  1994. next_agino = INT_GET(last_dip->di_next_unlinked, ARCH_CONVERT);
  1995. ASSERT(next_agino != NULLAGINO);
  1996. ASSERT(next_agino != 0);
  1997. }
  1998. /*
  1999. * Now last_ibp points to the buffer previous to us on
  2000. * the unlinked list. Pull us from the list.
  2001. */
  2002. error = xfs_itobp(mp, tp, ip, &dip, &ibp, 0, 0);
  2003. if (error) {
  2004. cmn_err(CE_WARN,
  2005. "xfs_iunlink_remove: xfs_itobp() returned an error %d on %s. Returning error.",
  2006. error, mp->m_fsname);
  2007. return error;
  2008. }
  2009. next_agino = INT_GET(dip->di_next_unlinked, ARCH_CONVERT);
  2010. ASSERT(next_agino != 0);
  2011. ASSERT(next_agino != agino);
  2012. if (next_agino != NULLAGINO) {
  2013. INT_SET(dip->di_next_unlinked, ARCH_CONVERT, NULLAGINO);
  2014. offset = ip->i_boffset +
  2015. offsetof(xfs_dinode_t, di_next_unlinked);
  2016. xfs_trans_inode_buf(tp, ibp);
  2017. xfs_trans_log_buf(tp, ibp, offset,
  2018. (offset + sizeof(xfs_agino_t) - 1));
  2019. xfs_inobp_check(mp, ibp);
  2020. } else {
  2021. xfs_trans_brelse(tp, ibp);
  2022. }
  2023. /*
  2024. * Point the previous inode on the list to the next inode.
  2025. */
  2026. INT_SET(last_dip->di_next_unlinked, ARCH_CONVERT, next_agino);
  2027. ASSERT(next_agino != 0);
  2028. offset = last_offset + offsetof(xfs_dinode_t, di_next_unlinked);
  2029. xfs_trans_inode_buf(tp, last_ibp);
  2030. xfs_trans_log_buf(tp, last_ibp, offset,
  2031. (offset + sizeof(xfs_agino_t) - 1));
  2032. xfs_inobp_check(mp, last_ibp);
  2033. }
  2034. return 0;
  2035. }
  2036. STATIC_INLINE int xfs_inode_clean(xfs_inode_t *ip)
  2037. {
  2038. return (((ip->i_itemp == NULL) ||
  2039. !(ip->i_itemp->ili_format.ilf_fields & XFS_ILOG_ALL)) &&
  2040. (ip->i_update_core == 0));
  2041. }
  2042. STATIC void
  2043. xfs_ifree_cluster(
  2044. xfs_inode_t *free_ip,
  2045. xfs_trans_t *tp,
  2046. xfs_ino_t inum)
  2047. {
  2048. xfs_mount_t *mp = free_ip->i_mount;
  2049. int blks_per_cluster;
  2050. int nbufs;
  2051. int ninodes;
  2052. int i, j, found, pre_flushed;
  2053. xfs_daddr_t blkno;
  2054. xfs_buf_t *bp;
  2055. xfs_ihash_t *ih;
  2056. xfs_inode_t *ip, **ip_found;
  2057. xfs_inode_log_item_t *iip;
  2058. xfs_log_item_t *lip;
  2059. SPLDECL(s);
  2060. if (mp->m_sb.sb_blocksize >= XFS_INODE_CLUSTER_SIZE(mp)) {
  2061. blks_per_cluster = 1;
  2062. ninodes = mp->m_sb.sb_inopblock;
  2063. nbufs = XFS_IALLOC_BLOCKS(mp);
  2064. } else {
  2065. blks_per_cluster = XFS_INODE_CLUSTER_SIZE(mp) /
  2066. mp->m_sb.sb_blocksize;
  2067. ninodes = blks_per_cluster * mp->m_sb.sb_inopblock;
  2068. nbufs = XFS_IALLOC_BLOCKS(mp) / blks_per_cluster;
  2069. }
  2070. ip_found = kmem_alloc(ninodes * sizeof(xfs_inode_t *), KM_NOFS);
  2071. for (j = 0; j < nbufs; j++, inum += ninodes) {
  2072. blkno = XFS_AGB_TO_DADDR(mp, XFS_INO_TO_AGNO(mp, inum),
  2073. XFS_INO_TO_AGBNO(mp, inum));
  2074. /*
  2075. * Look for each inode in memory and attempt to lock it,
  2076. * we can be racing with flush and tail pushing here.
  2077. * any inode we get the locks on, add to an array of
  2078. * inode items to process later.
  2079. *
  2080. * The get the buffer lock, we could beat a flush
  2081. * or tail pushing thread to the lock here, in which
  2082. * case they will go looking for the inode buffer
  2083. * and fail, we need some other form of interlock
  2084. * here.
  2085. */
  2086. found = 0;
  2087. for (i = 0; i < ninodes; i++) {
  2088. ih = XFS_IHASH(mp, inum + i);
  2089. read_lock(&ih->ih_lock);
  2090. for (ip = ih->ih_next; ip != NULL; ip = ip->i_next) {
  2091. if (ip->i_ino == inum + i)
  2092. break;
  2093. }
  2094. /* Inode not in memory or we found it already,
  2095. * nothing to do
  2096. */
  2097. if (!ip || xfs_iflags_test(ip, XFS_ISTALE)) {
  2098. read_unlock(&ih->ih_lock);
  2099. continue;
  2100. }
  2101. if (xfs_inode_clean(ip)) {
  2102. read_unlock(&ih->ih_lock);
  2103. continue;
  2104. }
  2105. /* If we can get the locks then add it to the
  2106. * list, otherwise by the time we get the bp lock
  2107. * below it will already be attached to the
  2108. * inode buffer.
  2109. */
  2110. /* This inode will already be locked - by us, lets
  2111. * keep it that way.
  2112. */
  2113. if (ip == free_ip) {
  2114. if (xfs_iflock_nowait(ip)) {
  2115. xfs_iflags_set(ip, XFS_ISTALE);
  2116. if (xfs_inode_clean(ip)) {
  2117. xfs_ifunlock(ip);
  2118. } else {
  2119. ip_found[found++] = ip;
  2120. }
  2121. }
  2122. read_unlock(&ih->ih_lock);
  2123. continue;
  2124. }
  2125. if (xfs_ilock_nowait(ip, XFS_ILOCK_EXCL)) {
  2126. if (xfs_iflock_nowait(ip)) {
  2127. xfs_iflags_set(ip, XFS_ISTALE);
  2128. if (xfs_inode_clean(ip)) {
  2129. xfs_ifunlock(ip);
  2130. xfs_iunlock(ip, XFS_ILOCK_EXCL);
  2131. } else {
  2132. ip_found[found++] = ip;
  2133. }
  2134. } else {
  2135. xfs_iunlock(ip, XFS_ILOCK_EXCL);
  2136. }
  2137. }
  2138. read_unlock(&ih->ih_lock);
  2139. }
  2140. bp = xfs_trans_get_buf(tp, mp->m_ddev_targp, blkno,
  2141. mp->m_bsize * blks_per_cluster,
  2142. XFS_BUF_LOCK);
  2143. pre_flushed = 0;
  2144. lip = XFS_BUF_FSPRIVATE(bp, xfs_log_item_t *);
  2145. while (lip) {
  2146. if (lip->li_type == XFS_LI_INODE) {
  2147. iip = (xfs_inode_log_item_t *)lip;
  2148. ASSERT(iip->ili_logged == 1);
  2149. lip->li_cb = (void(*)(xfs_buf_t*,xfs_log_item_t*)) xfs_istale_done;
  2150. AIL_LOCK(mp,s);
  2151. iip->ili_flush_lsn = iip->ili_item.li_lsn;
  2152. AIL_UNLOCK(mp, s);
  2153. xfs_iflags_set(iip->ili_inode, XFS_ISTALE);
  2154. pre_flushed++;
  2155. }
  2156. lip = lip->li_bio_list;
  2157. }
  2158. for (i = 0; i < found; i++) {
  2159. ip = ip_found[i];
  2160. iip = ip->i_itemp;
  2161. if (!iip) {
  2162. ip->i_update_core = 0;
  2163. xfs_ifunlock(ip);
  2164. xfs_iunlock(ip, XFS_ILOCK_EXCL);
  2165. continue;
  2166. }
  2167. iip->ili_last_fields = iip->ili_format.ilf_fields;
  2168. iip->ili_format.ilf_fields = 0;
  2169. iip->ili_logged = 1;
  2170. AIL_LOCK(mp,s);
  2171. iip->ili_flush_lsn = iip->ili_item.li_lsn;
  2172. AIL_UNLOCK(mp, s);
  2173. xfs_buf_attach_iodone(bp,
  2174. (void(*)(xfs_buf_t*,xfs_log_item_t*))
  2175. xfs_istale_done, (xfs_log_item_t *)iip);
  2176. if (ip != free_ip) {
  2177. xfs_iunlock(ip, XFS_ILOCK_EXCL);
  2178. }
  2179. }
  2180. if (found || pre_flushed)
  2181. xfs_trans_stale_inode_buf(tp, bp);
  2182. xfs_trans_binval(tp, bp);
  2183. }
  2184. kmem_free(ip_found, ninodes * sizeof(xfs_inode_t *));
  2185. }
  2186. /*
  2187. * This is called to return an inode to the inode free list.
  2188. * The inode should already be truncated to 0 length and have
  2189. * no pages associated with it. This routine also assumes that
  2190. * the inode is already a part of the transaction.
  2191. *
  2192. * The on-disk copy of the inode will have been added to the list
  2193. * of unlinked inodes in the AGI. We need to remove the inode from
  2194. * that list atomically with respect to freeing it here.
  2195. */
  2196. int
  2197. xfs_ifree(
  2198. xfs_trans_t *tp,
  2199. xfs_inode_t *ip,
  2200. xfs_bmap_free_t *flist)
  2201. {
  2202. int error;
  2203. int delete;
  2204. xfs_ino_t first_ino;
  2205. ASSERT(ismrlocked(&ip->i_lock, MR_UPDATE));
  2206. ASSERT(ip->i_transp == tp);
  2207. ASSERT(ip->i_d.di_nlink == 0);
  2208. ASSERT(ip->i_d.di_nextents == 0);
  2209. ASSERT(ip->i_d.di_anextents == 0);
  2210. ASSERT((ip->i_d.di_size == 0 && ip->i_size == 0) ||
  2211. ((ip->i_d.di_mode & S_IFMT) != S_IFREG));
  2212. ASSERT(ip->i_d.di_nblocks == 0);
  2213. /*
  2214. * Pull the on-disk inode from the AGI unlinked list.
  2215. */
  2216. error = xfs_iunlink_remove(tp, ip);
  2217. if (error != 0) {
  2218. return error;
  2219. }
  2220. error = xfs_difree(tp, ip->i_ino, flist, &delete, &first_ino);
  2221. if (error != 0) {
  2222. return error;
  2223. }
  2224. ip->i_d.di_mode = 0; /* mark incore inode as free */
  2225. ip->i_d.di_flags = 0;
  2226. ip->i_d.di_dmevmask = 0;
  2227. ip->i_d.di_forkoff = 0; /* mark the attr fork not in use */
  2228. ip->i_df.if_ext_max =
  2229. XFS_IFORK_DSIZE(ip) / (uint)sizeof(xfs_bmbt_rec_t);
  2230. ip->i_d.di_format = XFS_DINODE_FMT_EXTENTS;
  2231. ip->i_d.di_aformat = XFS_DINODE_FMT_EXTENTS;
  2232. /*
  2233. * Bump the generation count so no one will be confused
  2234. * by reincarnations of this inode.
  2235. */
  2236. ip->i_d.di_gen++;
  2237. xfs_trans_log_inode(tp, ip, XFS_ILOG_CORE);
  2238. if (delete) {
  2239. xfs_ifree_cluster(ip, tp, first_ino);
  2240. }
  2241. return 0;
  2242. }
  2243. /*
  2244. * Reallocate the space for if_broot based on the number of records
  2245. * being added or deleted as indicated in rec_diff. Move the records
  2246. * and pointers in if_broot to fit the new size. When shrinking this
  2247. * will eliminate holes between the records and pointers created by
  2248. * the caller. When growing this will create holes to be filled in
  2249. * by the caller.
  2250. *
  2251. * The caller must not request to add more records than would fit in
  2252. * the on-disk inode root. If the if_broot is currently NULL, then
  2253. * if we adding records one will be allocated. The caller must also
  2254. * not request that the number of records go below zero, although
  2255. * it can go to zero.
  2256. *
  2257. * ip -- the inode whose if_broot area is changing
  2258. * ext_diff -- the change in the number of records, positive or negative,
  2259. * requested for the if_broot array.
  2260. */
  2261. void
  2262. xfs_iroot_realloc(
  2263. xfs_inode_t *ip,
  2264. int rec_diff,
  2265. int whichfork)
  2266. {
  2267. int cur_max;
  2268. xfs_ifork_t *ifp;
  2269. xfs_bmbt_block_t *new_broot;
  2270. int new_max;
  2271. size_t new_size;
  2272. char *np;
  2273. char *op;
  2274. /*
  2275. * Handle the degenerate case quietly.
  2276. */
  2277. if (rec_diff == 0) {
  2278. return;
  2279. }
  2280. ifp = XFS_IFORK_PTR(ip, whichfork);
  2281. if (rec_diff > 0) {
  2282. /*
  2283. * If there wasn't any memory allocated before, just
  2284. * allocate it now and get out.
  2285. */
  2286. if (ifp->if_broot_bytes == 0) {
  2287. new_size = (size_t)XFS_BMAP_BROOT_SPACE_CALC(rec_diff);
  2288. ifp->if_broot = (xfs_bmbt_block_t*)kmem_alloc(new_size,
  2289. KM_SLEEP);
  2290. ifp->if_broot_bytes = (int)new_size;
  2291. return;
  2292. }
  2293. /*
  2294. * If there is already an existing if_broot, then we need
  2295. * to realloc() it and shift the pointers to their new
  2296. * location. The records don't change location because
  2297. * they are kept butted up against the btree block header.
  2298. */
  2299. cur_max = XFS_BMAP_BROOT_MAXRECS(ifp->if_broot_bytes);
  2300. new_max = cur_max + rec_diff;
  2301. new_size = (size_t)XFS_BMAP_BROOT_SPACE_CALC(new_max);
  2302. ifp->if_broot = (xfs_bmbt_block_t *)
  2303. kmem_realloc(ifp->if_broot,
  2304. new_size,
  2305. (size_t)XFS_BMAP_BROOT_SPACE_CALC(cur_max), /* old size */
  2306. KM_SLEEP);
  2307. op = (char *)XFS_BMAP_BROOT_PTR_ADDR(ifp->if_broot, 1,
  2308. ifp->if_broot_bytes);
  2309. np = (char *)XFS_BMAP_BROOT_PTR_ADDR(ifp->if_broot, 1,
  2310. (int)new_size);
  2311. ifp->if_broot_bytes = (int)new_size;
  2312. ASSERT(ifp->if_broot_bytes <=
  2313. XFS_IFORK_SIZE(ip, whichfork) + XFS_BROOT_SIZE_ADJ);
  2314. memmove(np, op, cur_max * (uint)sizeof(xfs_dfsbno_t));
  2315. return;
  2316. }
  2317. /*
  2318. * rec_diff is less than 0. In this case, we are shrinking the
  2319. * if_broot buffer. It must already exist. If we go to zero
  2320. * records, just get rid of the root and clear the status bit.
  2321. */
  2322. ASSERT((ifp->if_broot != NULL) && (ifp->if_broot_bytes > 0));
  2323. cur_max = XFS_BMAP_BROOT_MAXRECS(ifp->if_broot_bytes);
  2324. new_max = cur_max + rec_diff;
  2325. ASSERT(new_max >= 0);
  2326. if (new_max > 0)
  2327. new_size = (size_t)XFS_BMAP_BROOT_SPACE_CALC(new_max);
  2328. else
  2329. new_size = 0;
  2330. if (new_size > 0) {
  2331. new_broot = (xfs_bmbt_block_t *)kmem_alloc(new_size, KM_SLEEP);
  2332. /*
  2333. * First copy over the btree block header.
  2334. */
  2335. memcpy(new_broot, ifp->if_broot, sizeof(xfs_bmbt_block_t));
  2336. } else {
  2337. new_broot = NULL;
  2338. ifp->if_flags &= ~XFS_IFBROOT;
  2339. }
  2340. /*
  2341. * Only copy the records and pointers if there are any.
  2342. */
  2343. if (new_max > 0) {
  2344. /*
  2345. * First copy the records.
  2346. */
  2347. op = (char *)XFS_BMAP_BROOT_REC_ADDR(ifp->if_broot, 1,
  2348. ifp->if_broot_bytes);
  2349. np = (char *)XFS_BMAP_BROOT_REC_ADDR(new_broot, 1,
  2350. (int)new_size);
  2351. memcpy(np, op, new_max * (uint)sizeof(xfs_bmbt_rec_t));
  2352. /*
  2353. * Then copy the pointers.
  2354. */
  2355. op = (char *)XFS_BMAP_BROOT_PTR_ADDR(ifp->if_broot, 1,
  2356. ifp->if_broot_bytes);
  2357. np = (char *)XFS_BMAP_BROOT_PTR_ADDR(new_broot, 1,
  2358. (int)new_size);
  2359. memcpy(np, op, new_max * (uint)sizeof(xfs_dfsbno_t));
  2360. }
  2361. kmem_free(ifp->if_broot, ifp->if_broot_bytes);
  2362. ifp->if_broot = new_broot;
  2363. ifp->if_broot_bytes = (int)new_size;
  2364. ASSERT(ifp->if_broot_bytes <=
  2365. XFS_IFORK_SIZE(ip, whichfork) + XFS_BROOT_SIZE_ADJ);
  2366. return;
  2367. }
  2368. /*
  2369. * This is called when the amount of space needed for if_data
  2370. * is increased or decreased. The change in size is indicated by
  2371. * the number of bytes that need to be added or deleted in the
  2372. * byte_diff parameter.
  2373. *
  2374. * If the amount of space needed has decreased below the size of the
  2375. * inline buffer, then switch to using the inline buffer. Otherwise,
  2376. * use kmem_realloc() or kmem_alloc() to adjust the size of the buffer
  2377. * to what is needed.
  2378. *
  2379. * ip -- the inode whose if_data area is changing
  2380. * byte_diff -- the change in the number of bytes, positive or negative,
  2381. * requested for the if_data array.
  2382. */
  2383. void
  2384. xfs_idata_realloc(
  2385. xfs_inode_t *ip,
  2386. int byte_diff,
  2387. int whichfork)
  2388. {
  2389. xfs_ifork_t *ifp;
  2390. int new_size;
  2391. int real_size;
  2392. if (byte_diff == 0) {
  2393. return;
  2394. }
  2395. ifp = XFS_IFORK_PTR(ip, whichfork);
  2396. new_size = (int)ifp->if_bytes + byte_diff;
  2397. ASSERT(new_size >= 0);
  2398. if (new_size == 0) {
  2399. if (ifp->if_u1.if_data != ifp->if_u2.if_inline_data) {
  2400. kmem_free(ifp->if_u1.if_data, ifp->if_real_bytes);
  2401. }
  2402. ifp->if_u1.if_data = NULL;
  2403. real_size = 0;
  2404. } else if (new_size <= sizeof(ifp->if_u2.if_inline_data)) {
  2405. /*
  2406. * If the valid extents/data can fit in if_inline_ext/data,
  2407. * copy them from the malloc'd vector and free it.
  2408. */
  2409. if (ifp->if_u1.if_data == NULL) {
  2410. ifp->if_u1.if_data = ifp->if_u2.if_inline_data;
  2411. } else if (ifp->if_u1.if_data != ifp->if_u2.if_inline_data) {
  2412. ASSERT(ifp->if_real_bytes != 0);
  2413. memcpy(ifp->if_u2.if_inline_data, ifp->if_u1.if_data,
  2414. new_size);
  2415. kmem_free(ifp->if_u1.if_data, ifp->if_real_bytes);
  2416. ifp->if_u1.if_data = ifp->if_u2.if_inline_data;
  2417. }
  2418. real_size = 0;
  2419. } else {
  2420. /*
  2421. * Stuck with malloc/realloc.
  2422. * For inline data, the underlying buffer must be
  2423. * a multiple of 4 bytes in size so that it can be
  2424. * logged and stay on word boundaries. We enforce
  2425. * that here.
  2426. */
  2427. real_size = roundup(new_size, 4);
  2428. if (ifp->if_u1.if_data == NULL) {
  2429. ASSERT(ifp->if_real_bytes == 0);
  2430. ifp->if_u1.if_data = kmem_alloc(real_size, KM_SLEEP);
  2431. } else if (ifp->if_u1.if_data != ifp->if_u2.if_inline_data) {
  2432. /*
  2433. * Only do the realloc if the underlying size
  2434. * is really changing.
  2435. */
  2436. if (ifp->if_real_bytes != real_size) {
  2437. ifp->if_u1.if_data =
  2438. kmem_realloc(ifp->if_u1.if_data,
  2439. real_size,
  2440. ifp->if_real_bytes,
  2441. KM_SLEEP);
  2442. }
  2443. } else {
  2444. ASSERT(ifp->if_real_bytes == 0);
  2445. ifp->if_u1.if_data = kmem_alloc(real_size, KM_SLEEP);
  2446. memcpy(ifp->if_u1.if_data, ifp->if_u2.if_inline_data,
  2447. ifp->if_bytes);
  2448. }
  2449. }
  2450. ifp->if_real_bytes = real_size;
  2451. ifp->if_bytes = new_size;
  2452. ASSERT(ifp->if_bytes <= XFS_IFORK_SIZE(ip, whichfork));
  2453. }
  2454. /*
  2455. * Map inode to disk block and offset.
  2456. *
  2457. * mp -- the mount point structure for the current file system
  2458. * tp -- the current transaction
  2459. * ino -- the inode number of the inode to be located
  2460. * imap -- this structure is filled in with the information necessary
  2461. * to retrieve the given inode from disk
  2462. * flags -- flags to pass to xfs_dilocate indicating whether or not
  2463. * lookups in the inode btree were OK or not
  2464. */
  2465. int
  2466. xfs_imap(
  2467. xfs_mount_t *mp,
  2468. xfs_trans_t *tp,
  2469. xfs_ino_t ino,
  2470. xfs_imap_t *imap,
  2471. uint flags)
  2472. {
  2473. xfs_fsblock_t fsbno;
  2474. int len;
  2475. int off;
  2476. int error;
  2477. fsbno = imap->im_blkno ?
  2478. XFS_DADDR_TO_FSB(mp, imap->im_blkno) : NULLFSBLOCK;
  2479. error = xfs_dilocate(mp, tp, ino, &fsbno, &len, &off, flags);
  2480. if (error != 0) {
  2481. return error;
  2482. }
  2483. imap->im_blkno = XFS_FSB_TO_DADDR(mp, fsbno);
  2484. imap->im_len = XFS_FSB_TO_BB(mp, len);
  2485. imap->im_agblkno = XFS_FSB_TO_AGBNO(mp, fsbno);
  2486. imap->im_ioffset = (ushort)off;
  2487. imap->im_boffset = (ushort)(off << mp->m_sb.sb_inodelog);
  2488. return 0;
  2489. }
  2490. void
  2491. xfs_idestroy_fork(
  2492. xfs_inode_t *ip,
  2493. int whichfork)
  2494. {
  2495. xfs_ifork_t *ifp;
  2496. ifp = XFS_IFORK_PTR(ip, whichfork);
  2497. if (ifp->if_broot != NULL) {
  2498. kmem_free(ifp->if_broot, ifp->if_broot_bytes);
  2499. ifp->if_broot = NULL;
  2500. }
  2501. /*
  2502. * If the format is local, then we can't have an extents
  2503. * array so just look for an inline data array. If we're
  2504. * not local then we may or may not have an extents list,
  2505. * so check and free it up if we do.
  2506. */
  2507. if (XFS_IFORK_FORMAT(ip, whichfork) == XFS_DINODE_FMT_LOCAL) {
  2508. if ((ifp->if_u1.if_data != ifp->if_u2.if_inline_data) &&
  2509. (ifp->if_u1.if_data != NULL)) {
  2510. ASSERT(ifp->if_real_bytes != 0);
  2511. kmem_free(ifp->if_u1.if_data, ifp->if_real_bytes);
  2512. ifp->if_u1.if_data = NULL;
  2513. ifp->if_real_bytes = 0;
  2514. }
  2515. } else if ((ifp->if_flags & XFS_IFEXTENTS) &&
  2516. ((ifp->if_flags & XFS_IFEXTIREC) ||
  2517. ((ifp->if_u1.if_extents != NULL) &&
  2518. (ifp->if_u1.if_extents != ifp->if_u2.if_inline_ext)))) {
  2519. ASSERT(ifp->if_real_bytes != 0);
  2520. xfs_iext_destroy(ifp);
  2521. }
  2522. ASSERT(ifp->if_u1.if_extents == NULL ||
  2523. ifp->if_u1.if_extents == ifp->if_u2.if_inline_ext);
  2524. ASSERT(ifp->if_real_bytes == 0);
  2525. if (whichfork == XFS_ATTR_FORK) {
  2526. kmem_zone_free(xfs_ifork_zone, ip->i_afp);
  2527. ip->i_afp = NULL;
  2528. }
  2529. }
  2530. /*
  2531. * This is called free all the memory associated with an inode.
  2532. * It must free the inode itself and any buffers allocated for
  2533. * if_extents/if_data and if_broot. It must also free the lock
  2534. * associated with the inode.
  2535. */
  2536. void
  2537. xfs_idestroy(
  2538. xfs_inode_t *ip)
  2539. {
  2540. switch (ip->i_d.di_mode & S_IFMT) {
  2541. case S_IFREG:
  2542. case S_IFDIR:
  2543. case S_IFLNK:
  2544. xfs_idestroy_fork(ip, XFS_DATA_FORK);
  2545. break;
  2546. }
  2547. if (ip->i_afp)
  2548. xfs_idestroy_fork(ip, XFS_ATTR_FORK);
  2549. mrfree(&ip->i_lock);
  2550. mrfree(&ip->i_iolock);
  2551. freesema(&ip->i_flock);
  2552. #ifdef XFS_BMAP_TRACE
  2553. ktrace_free(ip->i_xtrace);
  2554. #endif
  2555. #ifdef XFS_BMBT_TRACE
  2556. ktrace_free(ip->i_btrace);
  2557. #endif
  2558. #ifdef XFS_RW_TRACE
  2559. ktrace_free(ip->i_rwtrace);
  2560. #endif
  2561. #ifdef XFS_ILOCK_TRACE
  2562. ktrace_free(ip->i_lock_trace);
  2563. #endif
  2564. #ifdef XFS_DIR2_TRACE
  2565. ktrace_free(ip->i_dir_trace);
  2566. #endif
  2567. if (ip->i_itemp) {
  2568. /*
  2569. * Only if we are shutting down the fs will we see an
  2570. * inode still in the AIL. If it is there, we should remove
  2571. * it to prevent a use-after-free from occurring.
  2572. */
  2573. xfs_mount_t *mp = ip->i_mount;
  2574. xfs_log_item_t *lip = &ip->i_itemp->ili_item;
  2575. int s;
  2576. ASSERT(((lip->li_flags & XFS_LI_IN_AIL) == 0) ||
  2577. XFS_FORCED_SHUTDOWN(ip->i_mount));
  2578. if (lip->li_flags & XFS_LI_IN_AIL) {
  2579. AIL_LOCK(mp, s);
  2580. if (lip->li_flags & XFS_LI_IN_AIL)
  2581. xfs_trans_delete_ail(mp, lip, s);
  2582. else
  2583. AIL_UNLOCK(mp, s);
  2584. }
  2585. xfs_inode_item_destroy(ip);
  2586. }
  2587. kmem_zone_free(xfs_inode_zone, ip);
  2588. }
  2589. /*
  2590. * Increment the pin count of the given buffer.
  2591. * This value is protected by ipinlock spinlock in the mount structure.
  2592. */
  2593. void
  2594. xfs_ipin(
  2595. xfs_inode_t *ip)
  2596. {
  2597. ASSERT(ismrlocked(&ip->i_lock, MR_UPDATE));
  2598. atomic_inc(&ip->i_pincount);
  2599. }
  2600. /*
  2601. * Decrement the pin count of the given inode, and wake up
  2602. * anyone in xfs_iwait_unpin() if the count goes to 0. The
  2603. * inode must have been previously pinned with a call to xfs_ipin().
  2604. */
  2605. void
  2606. xfs_iunpin(
  2607. xfs_inode_t *ip)
  2608. {
  2609. ASSERT(atomic_read(&ip->i_pincount) > 0);
  2610. if (atomic_dec_and_lock(&ip->i_pincount, &ip->i_flags_lock)) {
  2611. /*
  2612. * If the inode is currently being reclaimed, the link between
  2613. * the bhv_vnode and the xfs_inode will be broken after the
  2614. * XFS_IRECLAIM* flag is set. Hence, if these flags are not
  2615. * set, then we can move forward and mark the linux inode dirty
  2616. * knowing that it is still valid as it won't freed until after
  2617. * the bhv_vnode<->xfs_inode link is broken in xfs_reclaim. The
  2618. * i_flags_lock is used to synchronise the setting of the
  2619. * XFS_IRECLAIM* flags and the breaking of the link, and so we
  2620. * can execute atomically w.r.t to reclaim by holding this lock
  2621. * here.
  2622. *
  2623. * However, we still need to issue the unpin wakeup call as the
  2624. * inode reclaim may be blocked waiting for the inode to become
  2625. * unpinned.
  2626. */
  2627. if (!__xfs_iflags_test(ip, XFS_IRECLAIM|XFS_IRECLAIMABLE)) {
  2628. bhv_vnode_t *vp = XFS_ITOV_NULL(ip);
  2629. struct inode *inode = NULL;
  2630. BUG_ON(vp == NULL);
  2631. inode = vn_to_inode(vp);
  2632. BUG_ON(inode->i_state & I_CLEAR);
  2633. /* make sync come back and flush this inode */
  2634. if (!(inode->i_state & (I_NEW|I_FREEING)))
  2635. mark_inode_dirty_sync(inode);
  2636. }
  2637. spin_unlock(&ip->i_flags_lock);
  2638. wake_up(&ip->i_ipin_wait);
  2639. }
  2640. }
  2641. /*
  2642. * This is called to wait for the given inode to be unpinned.
  2643. * It will sleep until this happens. The caller must have the
  2644. * inode locked in at least shared mode so that the buffer cannot
  2645. * be subsequently pinned once someone is waiting for it to be
  2646. * unpinned.
  2647. */
  2648. STATIC void
  2649. xfs_iunpin_wait(
  2650. xfs_inode_t *ip)
  2651. {
  2652. xfs_inode_log_item_t *iip;
  2653. xfs_lsn_t lsn;
  2654. ASSERT(ismrlocked(&ip->i_lock, MR_UPDATE | MR_ACCESS));
  2655. if (atomic_read(&ip->i_pincount) == 0) {
  2656. return;
  2657. }
  2658. iip = ip->i_itemp;
  2659. if (iip && iip->ili_last_lsn) {
  2660. lsn = iip->ili_last_lsn;
  2661. } else {
  2662. lsn = (xfs_lsn_t)0;
  2663. }
  2664. /*
  2665. * Give the log a push so we don't wait here too long.
  2666. */
  2667. xfs_log_force(ip->i_mount, lsn, XFS_LOG_FORCE);
  2668. wait_event(ip->i_ipin_wait, (atomic_read(&ip->i_pincount) == 0));
  2669. }
  2670. /*
  2671. * xfs_iextents_copy()
  2672. *
  2673. * This is called to copy the REAL extents (as opposed to the delayed
  2674. * allocation extents) from the inode into the given buffer. It
  2675. * returns the number of bytes copied into the buffer.
  2676. *
  2677. * If there are no delayed allocation extents, then we can just
  2678. * memcpy() the extents into the buffer. Otherwise, we need to
  2679. * examine each extent in turn and skip those which are delayed.
  2680. */
  2681. int
  2682. xfs_iextents_copy(
  2683. xfs_inode_t *ip,
  2684. xfs_bmbt_rec_t *buffer,
  2685. int whichfork)
  2686. {
  2687. int copied;
  2688. xfs_bmbt_rec_t *dest_ep;
  2689. xfs_bmbt_rec_t *ep;
  2690. #ifdef XFS_BMAP_TRACE
  2691. static char fname[] = "xfs_iextents_copy";
  2692. #endif
  2693. int i;
  2694. xfs_ifork_t *ifp;
  2695. int nrecs;
  2696. xfs_fsblock_t start_block;
  2697. ifp = XFS_IFORK_PTR(ip, whichfork);
  2698. ASSERT(ismrlocked(&ip->i_lock, MR_UPDATE|MR_ACCESS));
  2699. ASSERT(ifp->if_bytes > 0);
  2700. nrecs = ifp->if_bytes / (uint)sizeof(xfs_bmbt_rec_t);
  2701. xfs_bmap_trace_exlist(fname, ip, nrecs, whichfork);
  2702. ASSERT(nrecs > 0);
  2703. /*
  2704. * There are some delayed allocation extents in the
  2705. * inode, so copy the extents one at a time and skip
  2706. * the delayed ones. There must be at least one
  2707. * non-delayed extent.
  2708. */
  2709. dest_ep = buffer;
  2710. copied = 0;
  2711. for (i = 0; i < nrecs; i++) {
  2712. ep = xfs_iext_get_ext(ifp, i);
  2713. start_block = xfs_bmbt_get_startblock(ep);
  2714. if (ISNULLSTARTBLOCK(start_block)) {
  2715. /*
  2716. * It's a delayed allocation extent, so skip it.
  2717. */
  2718. continue;
  2719. }
  2720. /* Translate to on disk format */
  2721. put_unaligned(INT_GET(ep->l0, ARCH_CONVERT),
  2722. (__uint64_t*)&dest_ep->l0);
  2723. put_unaligned(INT_GET(ep->l1, ARCH_CONVERT),
  2724. (__uint64_t*)&dest_ep->l1);
  2725. dest_ep++;
  2726. copied++;
  2727. }
  2728. ASSERT(copied != 0);
  2729. xfs_validate_extents(ifp, copied, 1, XFS_EXTFMT_INODE(ip));
  2730. return (copied * (uint)sizeof(xfs_bmbt_rec_t));
  2731. }
  2732. /*
  2733. * Each of the following cases stores data into the same region
  2734. * of the on-disk inode, so only one of them can be valid at
  2735. * any given time. While it is possible to have conflicting formats
  2736. * and log flags, e.g. having XFS_ILOG_?DATA set when the fork is
  2737. * in EXTENTS format, this can only happen when the fork has
  2738. * changed formats after being modified but before being flushed.
  2739. * In these cases, the format always takes precedence, because the
  2740. * format indicates the current state of the fork.
  2741. */
  2742. /*ARGSUSED*/
  2743. STATIC int
  2744. xfs_iflush_fork(
  2745. xfs_inode_t *ip,
  2746. xfs_dinode_t *dip,
  2747. xfs_inode_log_item_t *iip,
  2748. int whichfork,
  2749. xfs_buf_t *bp)
  2750. {
  2751. char *cp;
  2752. xfs_ifork_t *ifp;
  2753. xfs_mount_t *mp;
  2754. #ifdef XFS_TRANS_DEBUG
  2755. int first;
  2756. #endif
  2757. static const short brootflag[2] =
  2758. { XFS_ILOG_DBROOT, XFS_ILOG_ABROOT };
  2759. static const short dataflag[2] =
  2760. { XFS_ILOG_DDATA, XFS_ILOG_ADATA };
  2761. static const short extflag[2] =
  2762. { XFS_ILOG_DEXT, XFS_ILOG_AEXT };
  2763. if (iip == NULL)
  2764. return 0;
  2765. ifp = XFS_IFORK_PTR(ip, whichfork);
  2766. /*
  2767. * This can happen if we gave up in iformat in an error path,
  2768. * for the attribute fork.
  2769. */
  2770. if (ifp == NULL) {
  2771. ASSERT(whichfork == XFS_ATTR_FORK);
  2772. return 0;
  2773. }
  2774. cp = XFS_DFORK_PTR(dip, whichfork);
  2775. mp = ip->i_mount;
  2776. switch (XFS_IFORK_FORMAT(ip, whichfork)) {
  2777. case XFS_DINODE_FMT_LOCAL:
  2778. if ((iip->ili_format.ilf_fields & dataflag[whichfork]) &&
  2779. (ifp->if_bytes > 0)) {
  2780. ASSERT(ifp->if_u1.if_data != NULL);
  2781. ASSERT(ifp->if_bytes <= XFS_IFORK_SIZE(ip, whichfork));
  2782. memcpy(cp, ifp->if_u1.if_data, ifp->if_bytes);
  2783. }
  2784. break;
  2785. case XFS_DINODE_FMT_EXTENTS:
  2786. ASSERT((ifp->if_flags & XFS_IFEXTENTS) ||
  2787. !(iip->ili_format.ilf_fields & extflag[whichfork]));
  2788. ASSERT((xfs_iext_get_ext(ifp, 0) != NULL) ||
  2789. (ifp->if_bytes == 0));
  2790. ASSERT((xfs_iext_get_ext(ifp, 0) == NULL) ||
  2791. (ifp->if_bytes > 0));
  2792. if ((iip->ili_format.ilf_fields & extflag[whichfork]) &&
  2793. (ifp->if_bytes > 0)) {
  2794. ASSERT(XFS_IFORK_NEXTENTS(ip, whichfork) > 0);
  2795. (void)xfs_iextents_copy(ip, (xfs_bmbt_rec_t *)cp,
  2796. whichfork);
  2797. }
  2798. break;
  2799. case XFS_DINODE_FMT_BTREE:
  2800. if ((iip->ili_format.ilf_fields & brootflag[whichfork]) &&
  2801. (ifp->if_broot_bytes > 0)) {
  2802. ASSERT(ifp->if_broot != NULL);
  2803. ASSERT(ifp->if_broot_bytes <=
  2804. (XFS_IFORK_SIZE(ip, whichfork) +
  2805. XFS_BROOT_SIZE_ADJ));
  2806. xfs_bmbt_to_bmdr(ifp->if_broot, ifp->if_broot_bytes,
  2807. (xfs_bmdr_block_t *)cp,
  2808. XFS_DFORK_SIZE(dip, mp, whichfork));
  2809. }
  2810. break;
  2811. case XFS_DINODE_FMT_DEV:
  2812. if (iip->ili_format.ilf_fields & XFS_ILOG_DEV) {
  2813. ASSERT(whichfork == XFS_DATA_FORK);
  2814. INT_SET(dip->di_u.di_dev, ARCH_CONVERT, ip->i_df.if_u2.if_rdev);
  2815. }
  2816. break;
  2817. case XFS_DINODE_FMT_UUID:
  2818. if (iip->ili_format.ilf_fields & XFS_ILOG_UUID) {
  2819. ASSERT(whichfork == XFS_DATA_FORK);
  2820. memcpy(&dip->di_u.di_muuid, &ip->i_df.if_u2.if_uuid,
  2821. sizeof(uuid_t));
  2822. }
  2823. break;
  2824. default:
  2825. ASSERT(0);
  2826. break;
  2827. }
  2828. return 0;
  2829. }
  2830. /*
  2831. * xfs_iflush() will write a modified inode's changes out to the
  2832. * inode's on disk home. The caller must have the inode lock held
  2833. * in at least shared mode and the inode flush semaphore must be
  2834. * held as well. The inode lock will still be held upon return from
  2835. * the call and the caller is free to unlock it.
  2836. * The inode flush lock will be unlocked when the inode reaches the disk.
  2837. * The flags indicate how the inode's buffer should be written out.
  2838. */
  2839. int
  2840. xfs_iflush(
  2841. xfs_inode_t *ip,
  2842. uint flags)
  2843. {
  2844. xfs_inode_log_item_t *iip;
  2845. xfs_buf_t *bp;
  2846. xfs_dinode_t *dip;
  2847. xfs_mount_t *mp;
  2848. int error;
  2849. /* REFERENCED */
  2850. xfs_chash_t *ch;
  2851. xfs_inode_t *iq;
  2852. int clcount; /* count of inodes clustered */
  2853. int bufwasdelwri;
  2854. enum { INT_DELWRI = (1 << 0), INT_ASYNC = (1 << 1) };
  2855. SPLDECL(s);
  2856. XFS_STATS_INC(xs_iflush_count);
  2857. ASSERT(ismrlocked(&ip->i_lock, MR_UPDATE|MR_ACCESS));
  2858. ASSERT(issemalocked(&(ip->i_flock)));
  2859. ASSERT(ip->i_d.di_format != XFS_DINODE_FMT_BTREE ||
  2860. ip->i_d.di_nextents > ip->i_df.if_ext_max);
  2861. iip = ip->i_itemp;
  2862. mp = ip->i_mount;
  2863. /*
  2864. * If the inode isn't dirty, then just release the inode
  2865. * flush lock and do nothing.
  2866. */
  2867. if ((ip->i_update_core == 0) &&
  2868. ((iip == NULL) || !(iip->ili_format.ilf_fields & XFS_ILOG_ALL))) {
  2869. ASSERT((iip != NULL) ?
  2870. !(iip->ili_item.li_flags & XFS_LI_IN_AIL) : 1);
  2871. xfs_ifunlock(ip);
  2872. return 0;
  2873. }
  2874. /*
  2875. * We can't flush the inode until it is unpinned, so
  2876. * wait for it. We know noone new can pin it, because
  2877. * we are holding the inode lock shared and you need
  2878. * to hold it exclusively to pin the inode.
  2879. */
  2880. xfs_iunpin_wait(ip);
  2881. /*
  2882. * This may have been unpinned because the filesystem is shutting
  2883. * down forcibly. If that's the case we must not write this inode
  2884. * to disk, because the log record didn't make it to disk!
  2885. */
  2886. if (XFS_FORCED_SHUTDOWN(mp)) {
  2887. ip->i_update_core = 0;
  2888. if (iip)
  2889. iip->ili_format.ilf_fields = 0;
  2890. xfs_ifunlock(ip);
  2891. return XFS_ERROR(EIO);
  2892. }
  2893. /*
  2894. * Get the buffer containing the on-disk inode.
  2895. */
  2896. error = xfs_itobp(mp, NULL, ip, &dip, &bp, 0, 0);
  2897. if (error) {
  2898. xfs_ifunlock(ip);
  2899. return error;
  2900. }
  2901. /*
  2902. * Decide how buffer will be flushed out. This is done before
  2903. * the call to xfs_iflush_int because this field is zeroed by it.
  2904. */
  2905. if (iip != NULL && iip->ili_format.ilf_fields != 0) {
  2906. /*
  2907. * Flush out the inode buffer according to the directions
  2908. * of the caller. In the cases where the caller has given
  2909. * us a choice choose the non-delwri case. This is because
  2910. * the inode is in the AIL and we need to get it out soon.
  2911. */
  2912. switch (flags) {
  2913. case XFS_IFLUSH_SYNC:
  2914. case XFS_IFLUSH_DELWRI_ELSE_SYNC:
  2915. flags = 0;
  2916. break;
  2917. case XFS_IFLUSH_ASYNC:
  2918. case XFS_IFLUSH_DELWRI_ELSE_ASYNC:
  2919. flags = INT_ASYNC;
  2920. break;
  2921. case XFS_IFLUSH_DELWRI:
  2922. flags = INT_DELWRI;
  2923. break;
  2924. default:
  2925. ASSERT(0);
  2926. flags = 0;
  2927. break;
  2928. }
  2929. } else {
  2930. switch (flags) {
  2931. case XFS_IFLUSH_DELWRI_ELSE_SYNC:
  2932. case XFS_IFLUSH_DELWRI_ELSE_ASYNC:
  2933. case XFS_IFLUSH_DELWRI:
  2934. flags = INT_DELWRI;
  2935. break;
  2936. case XFS_IFLUSH_ASYNC:
  2937. flags = INT_ASYNC;
  2938. break;
  2939. case XFS_IFLUSH_SYNC:
  2940. flags = 0;
  2941. break;
  2942. default:
  2943. ASSERT(0);
  2944. flags = 0;
  2945. break;
  2946. }
  2947. }
  2948. /*
  2949. * First flush out the inode that xfs_iflush was called with.
  2950. */
  2951. error = xfs_iflush_int(ip, bp);
  2952. if (error) {
  2953. goto corrupt_out;
  2954. }
  2955. /*
  2956. * inode clustering:
  2957. * see if other inodes can be gathered into this write
  2958. */
  2959. ip->i_chash->chl_buf = bp;
  2960. ch = XFS_CHASH(mp, ip->i_blkno);
  2961. s = mutex_spinlock(&ch->ch_lock);
  2962. clcount = 0;
  2963. for (iq = ip->i_cnext; iq != ip; iq = iq->i_cnext) {
  2964. /*
  2965. * Do an un-protected check to see if the inode is dirty and
  2966. * is a candidate for flushing. These checks will be repeated
  2967. * later after the appropriate locks are acquired.
  2968. */
  2969. iip = iq->i_itemp;
  2970. if ((iq->i_update_core == 0) &&
  2971. ((iip == NULL) ||
  2972. !(iip->ili_format.ilf_fields & XFS_ILOG_ALL)) &&
  2973. xfs_ipincount(iq) == 0) {
  2974. continue;
  2975. }
  2976. /*
  2977. * Try to get locks. If any are unavailable,
  2978. * then this inode cannot be flushed and is skipped.
  2979. */
  2980. /* get inode locks (just i_lock) */
  2981. if (xfs_ilock_nowait(iq, XFS_ILOCK_SHARED)) {
  2982. /* get inode flush lock */
  2983. if (xfs_iflock_nowait(iq)) {
  2984. /* check if pinned */
  2985. if (xfs_ipincount(iq) == 0) {
  2986. /* arriving here means that
  2987. * this inode can be flushed.
  2988. * first re-check that it's
  2989. * dirty
  2990. */
  2991. iip = iq->i_itemp;
  2992. if ((iq->i_update_core != 0)||
  2993. ((iip != NULL) &&
  2994. (iip->ili_format.ilf_fields & XFS_ILOG_ALL))) {
  2995. clcount++;
  2996. error = xfs_iflush_int(iq, bp);
  2997. if (error) {
  2998. xfs_iunlock(iq,
  2999. XFS_ILOCK_SHARED);
  3000. goto cluster_corrupt_out;
  3001. }
  3002. } else {
  3003. xfs_ifunlock(iq);
  3004. }
  3005. } else {
  3006. xfs_ifunlock(iq);
  3007. }
  3008. }
  3009. xfs_iunlock(iq, XFS_ILOCK_SHARED);
  3010. }
  3011. }
  3012. mutex_spinunlock(&ch->ch_lock, s);
  3013. if (clcount) {
  3014. XFS_STATS_INC(xs_icluster_flushcnt);
  3015. XFS_STATS_ADD(xs_icluster_flushinode, clcount);
  3016. }
  3017. /*
  3018. * If the buffer is pinned then push on the log so we won't
  3019. * get stuck waiting in the write for too long.
  3020. */
  3021. if (XFS_BUF_ISPINNED(bp)){
  3022. xfs_log_force(mp, (xfs_lsn_t)0, XFS_LOG_FORCE);
  3023. }
  3024. if (flags & INT_DELWRI) {
  3025. xfs_bdwrite(mp, bp);
  3026. } else if (flags & INT_ASYNC) {
  3027. xfs_bawrite(mp, bp);
  3028. } else {
  3029. error = xfs_bwrite(mp, bp);
  3030. }
  3031. return error;
  3032. corrupt_out:
  3033. xfs_buf_relse(bp);
  3034. xfs_force_shutdown(mp, SHUTDOWN_CORRUPT_INCORE);
  3035. xfs_iflush_abort(ip);
  3036. /*
  3037. * Unlocks the flush lock
  3038. */
  3039. return XFS_ERROR(EFSCORRUPTED);
  3040. cluster_corrupt_out:
  3041. /* Corruption detected in the clustering loop. Invalidate the
  3042. * inode buffer and shut down the filesystem.
  3043. */
  3044. mutex_spinunlock(&ch->ch_lock, s);
  3045. /*
  3046. * Clean up the buffer. If it was B_DELWRI, just release it --
  3047. * brelse can handle it with no problems. If not, shut down the
  3048. * filesystem before releasing the buffer.
  3049. */
  3050. if ((bufwasdelwri= XFS_BUF_ISDELAYWRITE(bp))) {
  3051. xfs_buf_relse(bp);
  3052. }
  3053. xfs_force_shutdown(mp, SHUTDOWN_CORRUPT_INCORE);
  3054. if(!bufwasdelwri) {
  3055. /*
  3056. * Just like incore_relse: if we have b_iodone functions,
  3057. * mark the buffer as an error and call them. Otherwise
  3058. * mark it as stale and brelse.
  3059. */
  3060. if (XFS_BUF_IODONE_FUNC(bp)) {
  3061. XFS_BUF_CLR_BDSTRAT_FUNC(bp);
  3062. XFS_BUF_UNDONE(bp);
  3063. XFS_BUF_STALE(bp);
  3064. XFS_BUF_SHUT(bp);
  3065. XFS_BUF_ERROR(bp,EIO);
  3066. xfs_biodone(bp);
  3067. } else {
  3068. XFS_BUF_STALE(bp);
  3069. xfs_buf_relse(bp);
  3070. }
  3071. }
  3072. xfs_iflush_abort(iq);
  3073. /*
  3074. * Unlocks the flush lock
  3075. */
  3076. return XFS_ERROR(EFSCORRUPTED);
  3077. }
  3078. STATIC int
  3079. xfs_iflush_int(
  3080. xfs_inode_t *ip,
  3081. xfs_buf_t *bp)
  3082. {
  3083. xfs_inode_log_item_t *iip;
  3084. xfs_dinode_t *dip;
  3085. xfs_mount_t *mp;
  3086. #ifdef XFS_TRANS_DEBUG
  3087. int first;
  3088. #endif
  3089. SPLDECL(s);
  3090. ASSERT(ismrlocked(&ip->i_lock, MR_UPDATE|MR_ACCESS));
  3091. ASSERT(issemalocked(&(ip->i_flock)));
  3092. ASSERT(ip->i_d.di_format != XFS_DINODE_FMT_BTREE ||
  3093. ip->i_d.di_nextents > ip->i_df.if_ext_max);
  3094. iip = ip->i_itemp;
  3095. mp = ip->i_mount;
  3096. /*
  3097. * If the inode isn't dirty, then just release the inode
  3098. * flush lock and do nothing.
  3099. */
  3100. if ((ip->i_update_core == 0) &&
  3101. ((iip == NULL) || !(iip->ili_format.ilf_fields & XFS_ILOG_ALL))) {
  3102. xfs_ifunlock(ip);
  3103. return 0;
  3104. }
  3105. /* set *dip = inode's place in the buffer */
  3106. dip = (xfs_dinode_t *)xfs_buf_offset(bp, ip->i_boffset);
  3107. /*
  3108. * Clear i_update_core before copying out the data.
  3109. * This is for coordination with our timestamp updates
  3110. * that don't hold the inode lock. They will always
  3111. * update the timestamps BEFORE setting i_update_core,
  3112. * so if we clear i_update_core after they set it we
  3113. * are guaranteed to see their updates to the timestamps.
  3114. * I believe that this depends on strongly ordered memory
  3115. * semantics, but we have that. We use the SYNCHRONIZE
  3116. * macro to make sure that the compiler does not reorder
  3117. * the i_update_core access below the data copy below.
  3118. */
  3119. ip->i_update_core = 0;
  3120. SYNCHRONIZE();
  3121. /*
  3122. * Make sure to get the latest atime from the Linux inode.
  3123. */
  3124. xfs_synchronize_atime(ip);
  3125. if (XFS_TEST_ERROR(INT_GET(dip->di_core.di_magic,ARCH_CONVERT) != XFS_DINODE_MAGIC,
  3126. mp, XFS_ERRTAG_IFLUSH_1, XFS_RANDOM_IFLUSH_1)) {
  3127. xfs_cmn_err(XFS_PTAG_IFLUSH, CE_ALERT, mp,
  3128. "xfs_iflush: Bad inode %Lu magic number 0x%x, ptr 0x%p",
  3129. ip->i_ino, (int) INT_GET(dip->di_core.di_magic, ARCH_CONVERT), dip);
  3130. goto corrupt_out;
  3131. }
  3132. if (XFS_TEST_ERROR(ip->i_d.di_magic != XFS_DINODE_MAGIC,
  3133. mp, XFS_ERRTAG_IFLUSH_2, XFS_RANDOM_IFLUSH_2)) {
  3134. xfs_cmn_err(XFS_PTAG_IFLUSH, CE_ALERT, mp,
  3135. "xfs_iflush: Bad inode %Lu, ptr 0x%p, magic number 0x%x",
  3136. ip->i_ino, ip, ip->i_d.di_magic);
  3137. goto corrupt_out;
  3138. }
  3139. if ((ip->i_d.di_mode & S_IFMT) == S_IFREG) {
  3140. if (XFS_TEST_ERROR(
  3141. (ip->i_d.di_format != XFS_DINODE_FMT_EXTENTS) &&
  3142. (ip->i_d.di_format != XFS_DINODE_FMT_BTREE),
  3143. mp, XFS_ERRTAG_IFLUSH_3, XFS_RANDOM_IFLUSH_3)) {
  3144. xfs_cmn_err(XFS_PTAG_IFLUSH, CE_ALERT, mp,
  3145. "xfs_iflush: Bad regular inode %Lu, ptr 0x%p",
  3146. ip->i_ino, ip);
  3147. goto corrupt_out;
  3148. }
  3149. } else if ((ip->i_d.di_mode & S_IFMT) == S_IFDIR) {
  3150. if (XFS_TEST_ERROR(
  3151. (ip->i_d.di_format != XFS_DINODE_FMT_EXTENTS) &&
  3152. (ip->i_d.di_format != XFS_DINODE_FMT_BTREE) &&
  3153. (ip->i_d.di_format != XFS_DINODE_FMT_LOCAL),
  3154. mp, XFS_ERRTAG_IFLUSH_4, XFS_RANDOM_IFLUSH_4)) {
  3155. xfs_cmn_err(XFS_PTAG_IFLUSH, CE_ALERT, mp,
  3156. "xfs_iflush: Bad directory inode %Lu, ptr 0x%p",
  3157. ip->i_ino, ip);
  3158. goto corrupt_out;
  3159. }
  3160. }
  3161. if (XFS_TEST_ERROR(ip->i_d.di_nextents + ip->i_d.di_anextents >
  3162. ip->i_d.di_nblocks, mp, XFS_ERRTAG_IFLUSH_5,
  3163. XFS_RANDOM_IFLUSH_5)) {
  3164. xfs_cmn_err(XFS_PTAG_IFLUSH, CE_ALERT, mp,
  3165. "xfs_iflush: detected corrupt incore inode %Lu, total extents = %d, nblocks = %Ld, ptr 0x%p",
  3166. ip->i_ino,
  3167. ip->i_d.di_nextents + ip->i_d.di_anextents,
  3168. ip->i_d.di_nblocks,
  3169. ip);
  3170. goto corrupt_out;
  3171. }
  3172. if (XFS_TEST_ERROR(ip->i_d.di_forkoff > mp->m_sb.sb_inodesize,
  3173. mp, XFS_ERRTAG_IFLUSH_6, XFS_RANDOM_IFLUSH_6)) {
  3174. xfs_cmn_err(XFS_PTAG_IFLUSH, CE_ALERT, mp,
  3175. "xfs_iflush: bad inode %Lu, forkoff 0x%x, ptr 0x%p",
  3176. ip->i_ino, ip->i_d.di_forkoff, ip);
  3177. goto corrupt_out;
  3178. }
  3179. /*
  3180. * bump the flush iteration count, used to detect flushes which
  3181. * postdate a log record during recovery.
  3182. */
  3183. ip->i_d.di_flushiter++;
  3184. /*
  3185. * Copy the dirty parts of the inode into the on-disk
  3186. * inode. We always copy out the core of the inode,
  3187. * because if the inode is dirty at all the core must
  3188. * be.
  3189. */
  3190. xfs_xlate_dinode_core((xfs_caddr_t)&(dip->di_core), &(ip->i_d), -1);
  3191. /* Wrap, we never let the log put out DI_MAX_FLUSH */
  3192. if (ip->i_d.di_flushiter == DI_MAX_FLUSH)
  3193. ip->i_d.di_flushiter = 0;
  3194. /*
  3195. * If this is really an old format inode and the superblock version
  3196. * has not been updated to support only new format inodes, then
  3197. * convert back to the old inode format. If the superblock version
  3198. * has been updated, then make the conversion permanent.
  3199. */
  3200. ASSERT(ip->i_d.di_version == XFS_DINODE_VERSION_1 ||
  3201. XFS_SB_VERSION_HASNLINK(&mp->m_sb));
  3202. if (ip->i_d.di_version == XFS_DINODE_VERSION_1) {
  3203. if (!XFS_SB_VERSION_HASNLINK(&mp->m_sb)) {
  3204. /*
  3205. * Convert it back.
  3206. */
  3207. ASSERT(ip->i_d.di_nlink <= XFS_MAXLINK_1);
  3208. INT_SET(dip->di_core.di_onlink, ARCH_CONVERT, ip->i_d.di_nlink);
  3209. } else {
  3210. /*
  3211. * The superblock version has already been bumped,
  3212. * so just make the conversion to the new inode
  3213. * format permanent.
  3214. */
  3215. ip->i_d.di_version = XFS_DINODE_VERSION_2;
  3216. INT_SET(dip->di_core.di_version, ARCH_CONVERT, XFS_DINODE_VERSION_2);
  3217. ip->i_d.di_onlink = 0;
  3218. dip->di_core.di_onlink = 0;
  3219. memset(&(ip->i_d.di_pad[0]), 0, sizeof(ip->i_d.di_pad));
  3220. memset(&(dip->di_core.di_pad[0]), 0,
  3221. sizeof(dip->di_core.di_pad));
  3222. ASSERT(ip->i_d.di_projid == 0);
  3223. }
  3224. }
  3225. if (xfs_iflush_fork(ip, dip, iip, XFS_DATA_FORK, bp) == EFSCORRUPTED) {
  3226. goto corrupt_out;
  3227. }
  3228. if (XFS_IFORK_Q(ip)) {
  3229. /*
  3230. * The only error from xfs_iflush_fork is on the data fork.
  3231. */
  3232. (void) xfs_iflush_fork(ip, dip, iip, XFS_ATTR_FORK, bp);
  3233. }
  3234. xfs_inobp_check(mp, bp);
  3235. /*
  3236. * We've recorded everything logged in the inode, so we'd
  3237. * like to clear the ilf_fields bits so we don't log and
  3238. * flush things unnecessarily. However, we can't stop
  3239. * logging all this information until the data we've copied
  3240. * into the disk buffer is written to disk. If we did we might
  3241. * overwrite the copy of the inode in the log with all the
  3242. * data after re-logging only part of it, and in the face of
  3243. * a crash we wouldn't have all the data we need to recover.
  3244. *
  3245. * What we do is move the bits to the ili_last_fields field.
  3246. * When logging the inode, these bits are moved back to the
  3247. * ilf_fields field. In the xfs_iflush_done() routine we
  3248. * clear ili_last_fields, since we know that the information
  3249. * those bits represent is permanently on disk. As long as
  3250. * the flush completes before the inode is logged again, then
  3251. * both ilf_fields and ili_last_fields will be cleared.
  3252. *
  3253. * We can play with the ilf_fields bits here, because the inode
  3254. * lock must be held exclusively in order to set bits there
  3255. * and the flush lock protects the ili_last_fields bits.
  3256. * Set ili_logged so the flush done
  3257. * routine can tell whether or not to look in the AIL.
  3258. * Also, store the current LSN of the inode so that we can tell
  3259. * whether the item has moved in the AIL from xfs_iflush_done().
  3260. * In order to read the lsn we need the AIL lock, because
  3261. * it is a 64 bit value that cannot be read atomically.
  3262. */
  3263. if (iip != NULL && iip->ili_format.ilf_fields != 0) {
  3264. iip->ili_last_fields = iip->ili_format.ilf_fields;
  3265. iip->ili_format.ilf_fields = 0;
  3266. iip->ili_logged = 1;
  3267. ASSERT(sizeof(xfs_lsn_t) == 8); /* don't lock if it shrinks */
  3268. AIL_LOCK(mp,s);
  3269. iip->ili_flush_lsn = iip->ili_item.li_lsn;
  3270. AIL_UNLOCK(mp, s);
  3271. /*
  3272. * Attach the function xfs_iflush_done to the inode's
  3273. * buffer. This will remove the inode from the AIL
  3274. * and unlock the inode's flush lock when the inode is
  3275. * completely written to disk.
  3276. */
  3277. xfs_buf_attach_iodone(bp, (void(*)(xfs_buf_t*,xfs_log_item_t*))
  3278. xfs_iflush_done, (xfs_log_item_t *)iip);
  3279. ASSERT(XFS_BUF_FSPRIVATE(bp, void *) != NULL);
  3280. ASSERT(XFS_BUF_IODONE_FUNC(bp) != NULL);
  3281. } else {
  3282. /*
  3283. * We're flushing an inode which is not in the AIL and has
  3284. * not been logged but has i_update_core set. For this
  3285. * case we can use a B_DELWRI flush and immediately drop
  3286. * the inode flush lock because we can avoid the whole
  3287. * AIL state thing. It's OK to drop the flush lock now,
  3288. * because we've already locked the buffer and to do anything
  3289. * you really need both.
  3290. */
  3291. if (iip != NULL) {
  3292. ASSERT(iip->ili_logged == 0);
  3293. ASSERT(iip->ili_last_fields == 0);
  3294. ASSERT((iip->ili_item.li_flags & XFS_LI_IN_AIL) == 0);
  3295. }
  3296. xfs_ifunlock(ip);
  3297. }
  3298. return 0;
  3299. corrupt_out:
  3300. return XFS_ERROR(EFSCORRUPTED);
  3301. }
  3302. /*
  3303. * Flush all inactive inodes in mp.
  3304. */
  3305. void
  3306. xfs_iflush_all(
  3307. xfs_mount_t *mp)
  3308. {
  3309. xfs_inode_t *ip;
  3310. bhv_vnode_t *vp;
  3311. again:
  3312. XFS_MOUNT_ILOCK(mp);
  3313. ip = mp->m_inodes;
  3314. if (ip == NULL)
  3315. goto out;
  3316. do {
  3317. /* Make sure we skip markers inserted by sync */
  3318. if (ip->i_mount == NULL) {
  3319. ip = ip->i_mnext;
  3320. continue;
  3321. }
  3322. vp = XFS_ITOV_NULL(ip);
  3323. if (!vp) {
  3324. XFS_MOUNT_IUNLOCK(mp);
  3325. xfs_finish_reclaim(ip, 0, XFS_IFLUSH_ASYNC);
  3326. goto again;
  3327. }
  3328. ASSERT(vn_count(vp) == 0);
  3329. ip = ip->i_mnext;
  3330. } while (ip != mp->m_inodes);
  3331. out:
  3332. XFS_MOUNT_IUNLOCK(mp);
  3333. }
  3334. /*
  3335. * xfs_iaccess: check accessibility of inode for mode.
  3336. */
  3337. int
  3338. xfs_iaccess(
  3339. xfs_inode_t *ip,
  3340. mode_t mode,
  3341. cred_t *cr)
  3342. {
  3343. int error;
  3344. mode_t orgmode = mode;
  3345. struct inode *inode = vn_to_inode(XFS_ITOV(ip));
  3346. if (mode & S_IWUSR) {
  3347. umode_t imode = inode->i_mode;
  3348. if (IS_RDONLY(inode) &&
  3349. (S_ISREG(imode) || S_ISDIR(imode) || S_ISLNK(imode)))
  3350. return XFS_ERROR(EROFS);
  3351. if (IS_IMMUTABLE(inode))
  3352. return XFS_ERROR(EACCES);
  3353. }
  3354. /*
  3355. * If there's an Access Control List it's used instead of
  3356. * the mode bits.
  3357. */
  3358. if ((error = _ACL_XFS_IACCESS(ip, mode, cr)) != -1)
  3359. return error ? XFS_ERROR(error) : 0;
  3360. if (current_fsuid(cr) != ip->i_d.di_uid) {
  3361. mode >>= 3;
  3362. if (!in_group_p((gid_t)ip->i_d.di_gid))
  3363. mode >>= 3;
  3364. }
  3365. /*
  3366. * If the DACs are ok we don't need any capability check.
  3367. */
  3368. if ((ip->i_d.di_mode & mode) == mode)
  3369. return 0;
  3370. /*
  3371. * Read/write DACs are always overridable.
  3372. * Executable DACs are overridable if at least one exec bit is set.
  3373. */
  3374. if (!(orgmode & S_IXUSR) ||
  3375. (inode->i_mode & S_IXUGO) || S_ISDIR(inode->i_mode))
  3376. if (capable_cred(cr, CAP_DAC_OVERRIDE))
  3377. return 0;
  3378. if ((orgmode == S_IRUSR) ||
  3379. (S_ISDIR(inode->i_mode) && (!(orgmode & S_IWUSR)))) {
  3380. if (capable_cred(cr, CAP_DAC_READ_SEARCH))
  3381. return 0;
  3382. #ifdef NOISE
  3383. cmn_err(CE_NOTE, "Ick: mode=%o, orgmode=%o", mode, orgmode);
  3384. #endif /* NOISE */
  3385. return XFS_ERROR(EACCES);
  3386. }
  3387. return XFS_ERROR(EACCES);
  3388. }
  3389. /*
  3390. * xfs_iroundup: round up argument to next power of two
  3391. */
  3392. uint
  3393. xfs_iroundup(
  3394. uint v)
  3395. {
  3396. int i;
  3397. uint m;
  3398. if ((v & (v - 1)) == 0)
  3399. return v;
  3400. ASSERT((v & 0x80000000) == 0);
  3401. if ((v & (v + 1)) == 0)
  3402. return v + 1;
  3403. for (i = 0, m = 1; i < 31; i++, m <<= 1) {
  3404. if (v & m)
  3405. continue;
  3406. v |= m;
  3407. if ((v & (v + 1)) == 0)
  3408. return v + 1;
  3409. }
  3410. ASSERT(0);
  3411. return( 0 );
  3412. }
  3413. #ifdef XFS_ILOCK_TRACE
  3414. ktrace_t *xfs_ilock_trace_buf;
  3415. void
  3416. xfs_ilock_trace(xfs_inode_t *ip, int lock, unsigned int lockflags, inst_t *ra)
  3417. {
  3418. ktrace_enter(ip->i_lock_trace,
  3419. (void *)ip,
  3420. (void *)(unsigned long)lock, /* 1 = LOCK, 3=UNLOCK, etc */
  3421. (void *)(unsigned long)lockflags, /* XFS_ILOCK_EXCL etc */
  3422. (void *)ra, /* caller of ilock */
  3423. (void *)(unsigned long)current_cpu(),
  3424. (void *)(unsigned long)current_pid(),
  3425. NULL,NULL,NULL,NULL,NULL,NULL,NULL,NULL,NULL,NULL);
  3426. }
  3427. #endif
  3428. /*
  3429. * Return a pointer to the extent record at file index idx.
  3430. */
  3431. xfs_bmbt_rec_t *
  3432. xfs_iext_get_ext(
  3433. xfs_ifork_t *ifp, /* inode fork pointer */
  3434. xfs_extnum_t idx) /* index of target extent */
  3435. {
  3436. ASSERT(idx >= 0);
  3437. if ((ifp->if_flags & XFS_IFEXTIREC) && (idx == 0)) {
  3438. return ifp->if_u1.if_ext_irec->er_extbuf;
  3439. } else if (ifp->if_flags & XFS_IFEXTIREC) {
  3440. xfs_ext_irec_t *erp; /* irec pointer */
  3441. int erp_idx = 0; /* irec index */
  3442. xfs_extnum_t page_idx = idx; /* ext index in target list */
  3443. erp = xfs_iext_idx_to_irec(ifp, &page_idx, &erp_idx, 0);
  3444. return &erp->er_extbuf[page_idx];
  3445. } else if (ifp->if_bytes) {
  3446. return &ifp->if_u1.if_extents[idx];
  3447. } else {
  3448. return NULL;
  3449. }
  3450. }
  3451. /*
  3452. * Insert new item(s) into the extent records for incore inode
  3453. * fork 'ifp'. 'count' new items are inserted at index 'idx'.
  3454. */
  3455. void
  3456. xfs_iext_insert(
  3457. xfs_ifork_t *ifp, /* inode fork pointer */
  3458. xfs_extnum_t idx, /* starting index of new items */
  3459. xfs_extnum_t count, /* number of inserted items */
  3460. xfs_bmbt_irec_t *new) /* items to insert */
  3461. {
  3462. xfs_bmbt_rec_t *ep; /* extent record pointer */
  3463. xfs_extnum_t i; /* extent record index */
  3464. ASSERT(ifp->if_flags & XFS_IFEXTENTS);
  3465. xfs_iext_add(ifp, idx, count);
  3466. for (i = idx; i < idx + count; i++, new++) {
  3467. ep = xfs_iext_get_ext(ifp, i);
  3468. xfs_bmbt_set_all(ep, new);
  3469. }
  3470. }
  3471. /*
  3472. * This is called when the amount of space required for incore file
  3473. * extents needs to be increased. The ext_diff parameter stores the
  3474. * number of new extents being added and the idx parameter contains
  3475. * the extent index where the new extents will be added. If the new
  3476. * extents are being appended, then we just need to (re)allocate and
  3477. * initialize the space. Otherwise, if the new extents are being
  3478. * inserted into the middle of the existing entries, a bit more work
  3479. * is required to make room for the new extents to be inserted. The
  3480. * caller is responsible for filling in the new extent entries upon
  3481. * return.
  3482. */
  3483. void
  3484. xfs_iext_add(
  3485. xfs_ifork_t *ifp, /* inode fork pointer */
  3486. xfs_extnum_t idx, /* index to begin adding exts */
  3487. int ext_diff) /* number of extents to add */
  3488. {
  3489. int byte_diff; /* new bytes being added */
  3490. int new_size; /* size of extents after adding */
  3491. xfs_extnum_t nextents; /* number of extents in file */
  3492. nextents = ifp->if_bytes / (uint)sizeof(xfs_bmbt_rec_t);
  3493. ASSERT((idx >= 0) && (idx <= nextents));
  3494. byte_diff = ext_diff * sizeof(xfs_bmbt_rec_t);
  3495. new_size = ifp->if_bytes + byte_diff;
  3496. /*
  3497. * If the new number of extents (nextents + ext_diff)
  3498. * fits inside the inode, then continue to use the inline
  3499. * extent buffer.
  3500. */
  3501. if (nextents + ext_diff <= XFS_INLINE_EXTS) {
  3502. if (idx < nextents) {
  3503. memmove(&ifp->if_u2.if_inline_ext[idx + ext_diff],
  3504. &ifp->if_u2.if_inline_ext[idx],
  3505. (nextents - idx) * sizeof(xfs_bmbt_rec_t));
  3506. memset(&ifp->if_u2.if_inline_ext[idx], 0, byte_diff);
  3507. }
  3508. ifp->if_u1.if_extents = ifp->if_u2.if_inline_ext;
  3509. ifp->if_real_bytes = 0;
  3510. ifp->if_lastex = nextents + ext_diff;
  3511. }
  3512. /*
  3513. * Otherwise use a linear (direct) extent list.
  3514. * If the extents are currently inside the inode,
  3515. * xfs_iext_realloc_direct will switch us from
  3516. * inline to direct extent allocation mode.
  3517. */
  3518. else if (nextents + ext_diff <= XFS_LINEAR_EXTS) {
  3519. xfs_iext_realloc_direct(ifp, new_size);
  3520. if (idx < nextents) {
  3521. memmove(&ifp->if_u1.if_extents[idx + ext_diff],
  3522. &ifp->if_u1.if_extents[idx],
  3523. (nextents - idx) * sizeof(xfs_bmbt_rec_t));
  3524. memset(&ifp->if_u1.if_extents[idx], 0, byte_diff);
  3525. }
  3526. }
  3527. /* Indirection array */
  3528. else {
  3529. xfs_ext_irec_t *erp;
  3530. int erp_idx = 0;
  3531. int page_idx = idx;
  3532. ASSERT(nextents + ext_diff > XFS_LINEAR_EXTS);
  3533. if (ifp->if_flags & XFS_IFEXTIREC) {
  3534. erp = xfs_iext_idx_to_irec(ifp, &page_idx, &erp_idx, 1);
  3535. } else {
  3536. xfs_iext_irec_init(ifp);
  3537. ASSERT(ifp->if_flags & XFS_IFEXTIREC);
  3538. erp = ifp->if_u1.if_ext_irec;
  3539. }
  3540. /* Extents fit in target extent page */
  3541. if (erp && erp->er_extcount + ext_diff <= XFS_LINEAR_EXTS) {
  3542. if (page_idx < erp->er_extcount) {
  3543. memmove(&erp->er_extbuf[page_idx + ext_diff],
  3544. &erp->er_extbuf[page_idx],
  3545. (erp->er_extcount - page_idx) *
  3546. sizeof(xfs_bmbt_rec_t));
  3547. memset(&erp->er_extbuf[page_idx], 0, byte_diff);
  3548. }
  3549. erp->er_extcount += ext_diff;
  3550. xfs_iext_irec_update_extoffs(ifp, erp_idx + 1, ext_diff);
  3551. }
  3552. /* Insert a new extent page */
  3553. else if (erp) {
  3554. xfs_iext_add_indirect_multi(ifp,
  3555. erp_idx, page_idx, ext_diff);
  3556. }
  3557. /*
  3558. * If extent(s) are being appended to the last page in
  3559. * the indirection array and the new extent(s) don't fit
  3560. * in the page, then erp is NULL and erp_idx is set to
  3561. * the next index needed in the indirection array.
  3562. */
  3563. else {
  3564. int count = ext_diff;
  3565. while (count) {
  3566. erp = xfs_iext_irec_new(ifp, erp_idx);
  3567. erp->er_extcount = count;
  3568. count -= MIN(count, (int)XFS_LINEAR_EXTS);
  3569. if (count) {
  3570. erp_idx++;
  3571. }
  3572. }
  3573. }
  3574. }
  3575. ifp->if_bytes = new_size;
  3576. }
  3577. /*
  3578. * This is called when incore extents are being added to the indirection
  3579. * array and the new extents do not fit in the target extent list. The
  3580. * erp_idx parameter contains the irec index for the target extent list
  3581. * in the indirection array, and the idx parameter contains the extent
  3582. * index within the list. The number of extents being added is stored
  3583. * in the count parameter.
  3584. *
  3585. * |-------| |-------|
  3586. * | | | | idx - number of extents before idx
  3587. * | idx | | count |
  3588. * | | | | count - number of extents being inserted at idx
  3589. * |-------| |-------|
  3590. * | count | | nex2 | nex2 - number of extents after idx + count
  3591. * |-------| |-------|
  3592. */
  3593. void
  3594. xfs_iext_add_indirect_multi(
  3595. xfs_ifork_t *ifp, /* inode fork pointer */
  3596. int erp_idx, /* target extent irec index */
  3597. xfs_extnum_t idx, /* index within target list */
  3598. int count) /* new extents being added */
  3599. {
  3600. int byte_diff; /* new bytes being added */
  3601. xfs_ext_irec_t *erp; /* pointer to irec entry */
  3602. xfs_extnum_t ext_diff; /* number of extents to add */
  3603. xfs_extnum_t ext_cnt; /* new extents still needed */
  3604. xfs_extnum_t nex2; /* extents after idx + count */
  3605. xfs_bmbt_rec_t *nex2_ep = NULL; /* temp list for nex2 extents */
  3606. int nlists; /* number of irec's (lists) */
  3607. ASSERT(ifp->if_flags & XFS_IFEXTIREC);
  3608. erp = &ifp->if_u1.if_ext_irec[erp_idx];
  3609. nex2 = erp->er_extcount - idx;
  3610. nlists = ifp->if_real_bytes / XFS_IEXT_BUFSZ;
  3611. /*
  3612. * Save second part of target extent list
  3613. * (all extents past */
  3614. if (nex2) {
  3615. byte_diff = nex2 * sizeof(xfs_bmbt_rec_t);
  3616. nex2_ep = (xfs_bmbt_rec_t *) kmem_alloc(byte_diff, KM_SLEEP);
  3617. memmove(nex2_ep, &erp->er_extbuf[idx], byte_diff);
  3618. erp->er_extcount -= nex2;
  3619. xfs_iext_irec_update_extoffs(ifp, erp_idx + 1, -nex2);
  3620. memset(&erp->er_extbuf[idx], 0, byte_diff);
  3621. }
  3622. /*
  3623. * Add the new extents to the end of the target
  3624. * list, then allocate new irec record(s) and
  3625. * extent buffer(s) as needed to store the rest
  3626. * of the new extents.
  3627. */
  3628. ext_cnt = count;
  3629. ext_diff = MIN(ext_cnt, (int)XFS_LINEAR_EXTS - erp->er_extcount);
  3630. if (ext_diff) {
  3631. erp->er_extcount += ext_diff;
  3632. xfs_iext_irec_update_extoffs(ifp, erp_idx + 1, ext_diff);
  3633. ext_cnt -= ext_diff;
  3634. }
  3635. while (ext_cnt) {
  3636. erp_idx++;
  3637. erp = xfs_iext_irec_new(ifp, erp_idx);
  3638. ext_diff = MIN(ext_cnt, (int)XFS_LINEAR_EXTS);
  3639. erp->er_extcount = ext_diff;
  3640. xfs_iext_irec_update_extoffs(ifp, erp_idx + 1, ext_diff);
  3641. ext_cnt -= ext_diff;
  3642. }
  3643. /* Add nex2 extents back to indirection array */
  3644. if (nex2) {
  3645. xfs_extnum_t ext_avail;
  3646. int i;
  3647. byte_diff = nex2 * sizeof(xfs_bmbt_rec_t);
  3648. ext_avail = XFS_LINEAR_EXTS - erp->er_extcount;
  3649. i = 0;
  3650. /*
  3651. * If nex2 extents fit in the current page, append
  3652. * nex2_ep after the new extents.
  3653. */
  3654. if (nex2 <= ext_avail) {
  3655. i = erp->er_extcount;
  3656. }
  3657. /*
  3658. * Otherwise, check if space is available in the
  3659. * next page.
  3660. */
  3661. else if ((erp_idx < nlists - 1) &&
  3662. (nex2 <= (ext_avail = XFS_LINEAR_EXTS -
  3663. ifp->if_u1.if_ext_irec[erp_idx+1].er_extcount))) {
  3664. erp_idx++;
  3665. erp++;
  3666. /* Create a hole for nex2 extents */
  3667. memmove(&erp->er_extbuf[nex2], erp->er_extbuf,
  3668. erp->er_extcount * sizeof(xfs_bmbt_rec_t));
  3669. }
  3670. /*
  3671. * Final choice, create a new extent page for
  3672. * nex2 extents.
  3673. */
  3674. else {
  3675. erp_idx++;
  3676. erp = xfs_iext_irec_new(ifp, erp_idx);
  3677. }
  3678. memmove(&erp->er_extbuf[i], nex2_ep, byte_diff);
  3679. kmem_free(nex2_ep, byte_diff);
  3680. erp->er_extcount += nex2;
  3681. xfs_iext_irec_update_extoffs(ifp, erp_idx + 1, nex2);
  3682. }
  3683. }
  3684. /*
  3685. * This is called when the amount of space required for incore file
  3686. * extents needs to be decreased. The ext_diff parameter stores the
  3687. * number of extents to be removed and the idx parameter contains
  3688. * the extent index where the extents will be removed from.
  3689. *
  3690. * If the amount of space needed has decreased below the linear
  3691. * limit, XFS_IEXT_BUFSZ, then switch to using the contiguous
  3692. * extent array. Otherwise, use kmem_realloc() to adjust the
  3693. * size to what is needed.
  3694. */
  3695. void
  3696. xfs_iext_remove(
  3697. xfs_ifork_t *ifp, /* inode fork pointer */
  3698. xfs_extnum_t idx, /* index to begin removing exts */
  3699. int ext_diff) /* number of extents to remove */
  3700. {
  3701. xfs_extnum_t nextents; /* number of extents in file */
  3702. int new_size; /* size of extents after removal */
  3703. ASSERT(ext_diff > 0);
  3704. nextents = ifp->if_bytes / (uint)sizeof(xfs_bmbt_rec_t);
  3705. new_size = (nextents - ext_diff) * sizeof(xfs_bmbt_rec_t);
  3706. if (new_size == 0) {
  3707. xfs_iext_destroy(ifp);
  3708. } else if (ifp->if_flags & XFS_IFEXTIREC) {
  3709. xfs_iext_remove_indirect(ifp, idx, ext_diff);
  3710. } else if (ifp->if_real_bytes) {
  3711. xfs_iext_remove_direct(ifp, idx, ext_diff);
  3712. } else {
  3713. xfs_iext_remove_inline(ifp, idx, ext_diff);
  3714. }
  3715. ifp->if_bytes = new_size;
  3716. }
  3717. /*
  3718. * This removes ext_diff extents from the inline buffer, beginning
  3719. * at extent index idx.
  3720. */
  3721. void
  3722. xfs_iext_remove_inline(
  3723. xfs_ifork_t *ifp, /* inode fork pointer */
  3724. xfs_extnum_t idx, /* index to begin removing exts */
  3725. int ext_diff) /* number of extents to remove */
  3726. {
  3727. int nextents; /* number of extents in file */
  3728. ASSERT(!(ifp->if_flags & XFS_IFEXTIREC));
  3729. ASSERT(idx < XFS_INLINE_EXTS);
  3730. nextents = ifp->if_bytes / (uint)sizeof(xfs_bmbt_rec_t);
  3731. ASSERT(((nextents - ext_diff) > 0) &&
  3732. (nextents - ext_diff) < XFS_INLINE_EXTS);
  3733. if (idx + ext_diff < nextents) {
  3734. memmove(&ifp->if_u2.if_inline_ext[idx],
  3735. &ifp->if_u2.if_inline_ext[idx + ext_diff],
  3736. (nextents - (idx + ext_diff)) *
  3737. sizeof(xfs_bmbt_rec_t));
  3738. memset(&ifp->if_u2.if_inline_ext[nextents - ext_diff],
  3739. 0, ext_diff * sizeof(xfs_bmbt_rec_t));
  3740. } else {
  3741. memset(&ifp->if_u2.if_inline_ext[idx], 0,
  3742. ext_diff * sizeof(xfs_bmbt_rec_t));
  3743. }
  3744. }
  3745. /*
  3746. * This removes ext_diff extents from a linear (direct) extent list,
  3747. * beginning at extent index idx. If the extents are being removed
  3748. * from the end of the list (ie. truncate) then we just need to re-
  3749. * allocate the list to remove the extra space. Otherwise, if the
  3750. * extents are being removed from the middle of the existing extent
  3751. * entries, then we first need to move the extent records beginning
  3752. * at idx + ext_diff up in the list to overwrite the records being
  3753. * removed, then remove the extra space via kmem_realloc.
  3754. */
  3755. void
  3756. xfs_iext_remove_direct(
  3757. xfs_ifork_t *ifp, /* inode fork pointer */
  3758. xfs_extnum_t idx, /* index to begin removing exts */
  3759. int ext_diff) /* number of extents to remove */
  3760. {
  3761. xfs_extnum_t nextents; /* number of extents in file */
  3762. int new_size; /* size of extents after removal */
  3763. ASSERT(!(ifp->if_flags & XFS_IFEXTIREC));
  3764. new_size = ifp->if_bytes -
  3765. (ext_diff * sizeof(xfs_bmbt_rec_t));
  3766. nextents = ifp->if_bytes / (uint)sizeof(xfs_bmbt_rec_t);
  3767. if (new_size == 0) {
  3768. xfs_iext_destroy(ifp);
  3769. return;
  3770. }
  3771. /* Move extents up in the list (if needed) */
  3772. if (idx + ext_diff < nextents) {
  3773. memmove(&ifp->if_u1.if_extents[idx],
  3774. &ifp->if_u1.if_extents[idx + ext_diff],
  3775. (nextents - (idx + ext_diff)) *
  3776. sizeof(xfs_bmbt_rec_t));
  3777. }
  3778. memset(&ifp->if_u1.if_extents[nextents - ext_diff],
  3779. 0, ext_diff * sizeof(xfs_bmbt_rec_t));
  3780. /*
  3781. * Reallocate the direct extent list. If the extents
  3782. * will fit inside the inode then xfs_iext_realloc_direct
  3783. * will switch from direct to inline extent allocation
  3784. * mode for us.
  3785. */
  3786. xfs_iext_realloc_direct(ifp, new_size);
  3787. ifp->if_bytes = new_size;
  3788. }
  3789. /*
  3790. * This is called when incore extents are being removed from the
  3791. * indirection array and the extents being removed span multiple extent
  3792. * buffers. The idx parameter contains the file extent index where we
  3793. * want to begin removing extents, and the count parameter contains
  3794. * how many extents need to be removed.
  3795. *
  3796. * |-------| |-------|
  3797. * | nex1 | | | nex1 - number of extents before idx
  3798. * |-------| | count |
  3799. * | | | | count - number of extents being removed at idx
  3800. * | count | |-------|
  3801. * | | | nex2 | nex2 - number of extents after idx + count
  3802. * |-------| |-------|
  3803. */
  3804. void
  3805. xfs_iext_remove_indirect(
  3806. xfs_ifork_t *ifp, /* inode fork pointer */
  3807. xfs_extnum_t idx, /* index to begin removing extents */
  3808. int count) /* number of extents to remove */
  3809. {
  3810. xfs_ext_irec_t *erp; /* indirection array pointer */
  3811. int erp_idx = 0; /* indirection array index */
  3812. xfs_extnum_t ext_cnt; /* extents left to remove */
  3813. xfs_extnum_t ext_diff; /* extents to remove in current list */
  3814. xfs_extnum_t nex1; /* number of extents before idx */
  3815. xfs_extnum_t nex2; /* extents after idx + count */
  3816. int nlists; /* entries in indirection array */
  3817. int page_idx = idx; /* index in target extent list */
  3818. ASSERT(ifp->if_flags & XFS_IFEXTIREC);
  3819. erp = xfs_iext_idx_to_irec(ifp, &page_idx, &erp_idx, 0);
  3820. ASSERT(erp != NULL);
  3821. nlists = ifp->if_real_bytes / XFS_IEXT_BUFSZ;
  3822. nex1 = page_idx;
  3823. ext_cnt = count;
  3824. while (ext_cnt) {
  3825. nex2 = MAX((erp->er_extcount - (nex1 + ext_cnt)), 0);
  3826. ext_diff = MIN(ext_cnt, (erp->er_extcount - nex1));
  3827. /*
  3828. * Check for deletion of entire list;
  3829. * xfs_iext_irec_remove() updates extent offsets.
  3830. */
  3831. if (ext_diff == erp->er_extcount) {
  3832. xfs_iext_irec_remove(ifp, erp_idx);
  3833. ext_cnt -= ext_diff;
  3834. nex1 = 0;
  3835. if (ext_cnt) {
  3836. ASSERT(erp_idx < ifp->if_real_bytes /
  3837. XFS_IEXT_BUFSZ);
  3838. erp = &ifp->if_u1.if_ext_irec[erp_idx];
  3839. nex1 = 0;
  3840. continue;
  3841. } else {
  3842. break;
  3843. }
  3844. }
  3845. /* Move extents up (if needed) */
  3846. if (nex2) {
  3847. memmove(&erp->er_extbuf[nex1],
  3848. &erp->er_extbuf[nex1 + ext_diff],
  3849. nex2 * sizeof(xfs_bmbt_rec_t));
  3850. }
  3851. /* Zero out rest of page */
  3852. memset(&erp->er_extbuf[nex1 + nex2], 0, (XFS_IEXT_BUFSZ -
  3853. ((nex1 + nex2) * sizeof(xfs_bmbt_rec_t))));
  3854. /* Update remaining counters */
  3855. erp->er_extcount -= ext_diff;
  3856. xfs_iext_irec_update_extoffs(ifp, erp_idx + 1, -ext_diff);
  3857. ext_cnt -= ext_diff;
  3858. nex1 = 0;
  3859. erp_idx++;
  3860. erp++;
  3861. }
  3862. ifp->if_bytes -= count * sizeof(xfs_bmbt_rec_t);
  3863. xfs_iext_irec_compact(ifp);
  3864. }
  3865. /*
  3866. * Create, destroy, or resize a linear (direct) block of extents.
  3867. */
  3868. void
  3869. xfs_iext_realloc_direct(
  3870. xfs_ifork_t *ifp, /* inode fork pointer */
  3871. int new_size) /* new size of extents */
  3872. {
  3873. int rnew_size; /* real new size of extents */
  3874. rnew_size = new_size;
  3875. ASSERT(!(ifp->if_flags & XFS_IFEXTIREC) ||
  3876. ((new_size >= 0) && (new_size <= XFS_IEXT_BUFSZ) &&
  3877. (new_size != ifp->if_real_bytes)));
  3878. /* Free extent records */
  3879. if (new_size == 0) {
  3880. xfs_iext_destroy(ifp);
  3881. }
  3882. /* Resize direct extent list and zero any new bytes */
  3883. else if (ifp->if_real_bytes) {
  3884. /* Check if extents will fit inside the inode */
  3885. if (new_size <= XFS_INLINE_EXTS * sizeof(xfs_bmbt_rec_t)) {
  3886. xfs_iext_direct_to_inline(ifp, new_size /
  3887. (uint)sizeof(xfs_bmbt_rec_t));
  3888. ifp->if_bytes = new_size;
  3889. return;
  3890. }
  3891. if (!is_power_of_2(new_size)){
  3892. rnew_size = xfs_iroundup(new_size);
  3893. }
  3894. if (rnew_size != ifp->if_real_bytes) {
  3895. ifp->if_u1.if_extents = (xfs_bmbt_rec_t *)
  3896. kmem_realloc(ifp->if_u1.if_extents,
  3897. rnew_size,
  3898. ifp->if_real_bytes,
  3899. KM_SLEEP);
  3900. }
  3901. if (rnew_size > ifp->if_real_bytes) {
  3902. memset(&ifp->if_u1.if_extents[ifp->if_bytes /
  3903. (uint)sizeof(xfs_bmbt_rec_t)], 0,
  3904. rnew_size - ifp->if_real_bytes);
  3905. }
  3906. }
  3907. /*
  3908. * Switch from the inline extent buffer to a direct
  3909. * extent list. Be sure to include the inline extent
  3910. * bytes in new_size.
  3911. */
  3912. else {
  3913. new_size += ifp->if_bytes;
  3914. if (!is_power_of_2(new_size)) {
  3915. rnew_size = xfs_iroundup(new_size);
  3916. }
  3917. xfs_iext_inline_to_direct(ifp, rnew_size);
  3918. }
  3919. ifp->if_real_bytes = rnew_size;
  3920. ifp->if_bytes = new_size;
  3921. }
  3922. /*
  3923. * Switch from linear (direct) extent records to inline buffer.
  3924. */
  3925. void
  3926. xfs_iext_direct_to_inline(
  3927. xfs_ifork_t *ifp, /* inode fork pointer */
  3928. xfs_extnum_t nextents) /* number of extents in file */
  3929. {
  3930. ASSERT(ifp->if_flags & XFS_IFEXTENTS);
  3931. ASSERT(nextents <= XFS_INLINE_EXTS);
  3932. /*
  3933. * The inline buffer was zeroed when we switched
  3934. * from inline to direct extent allocation mode,
  3935. * so we don't need to clear it here.
  3936. */
  3937. memcpy(ifp->if_u2.if_inline_ext, ifp->if_u1.if_extents,
  3938. nextents * sizeof(xfs_bmbt_rec_t));
  3939. kmem_free(ifp->if_u1.if_extents, ifp->if_real_bytes);
  3940. ifp->if_u1.if_extents = ifp->if_u2.if_inline_ext;
  3941. ifp->if_real_bytes = 0;
  3942. }
  3943. /*
  3944. * Switch from inline buffer to linear (direct) extent records.
  3945. * new_size should already be rounded up to the next power of 2
  3946. * by the caller (when appropriate), so use new_size as it is.
  3947. * However, since new_size may be rounded up, we can't update
  3948. * if_bytes here. It is the caller's responsibility to update
  3949. * if_bytes upon return.
  3950. */
  3951. void
  3952. xfs_iext_inline_to_direct(
  3953. xfs_ifork_t *ifp, /* inode fork pointer */
  3954. int new_size) /* number of extents in file */
  3955. {
  3956. ifp->if_u1.if_extents = (xfs_bmbt_rec_t *)
  3957. kmem_alloc(new_size, KM_SLEEP);
  3958. memset(ifp->if_u1.if_extents, 0, new_size);
  3959. if (ifp->if_bytes) {
  3960. memcpy(ifp->if_u1.if_extents, ifp->if_u2.if_inline_ext,
  3961. ifp->if_bytes);
  3962. memset(ifp->if_u2.if_inline_ext, 0, XFS_INLINE_EXTS *
  3963. sizeof(xfs_bmbt_rec_t));
  3964. }
  3965. ifp->if_real_bytes = new_size;
  3966. }
  3967. /*
  3968. * Resize an extent indirection array to new_size bytes.
  3969. */
  3970. void
  3971. xfs_iext_realloc_indirect(
  3972. xfs_ifork_t *ifp, /* inode fork pointer */
  3973. int new_size) /* new indirection array size */
  3974. {
  3975. int nlists; /* number of irec's (ex lists) */
  3976. int size; /* current indirection array size */
  3977. ASSERT(ifp->if_flags & XFS_IFEXTIREC);
  3978. nlists = ifp->if_real_bytes / XFS_IEXT_BUFSZ;
  3979. size = nlists * sizeof(xfs_ext_irec_t);
  3980. ASSERT(ifp->if_real_bytes);
  3981. ASSERT((new_size >= 0) && (new_size != size));
  3982. if (new_size == 0) {
  3983. xfs_iext_destroy(ifp);
  3984. } else {
  3985. ifp->if_u1.if_ext_irec = (xfs_ext_irec_t *)
  3986. kmem_realloc(ifp->if_u1.if_ext_irec,
  3987. new_size, size, KM_SLEEP);
  3988. }
  3989. }
  3990. /*
  3991. * Switch from indirection array to linear (direct) extent allocations.
  3992. */
  3993. void
  3994. xfs_iext_indirect_to_direct(
  3995. xfs_ifork_t *ifp) /* inode fork pointer */
  3996. {
  3997. xfs_bmbt_rec_t *ep; /* extent record pointer */
  3998. xfs_extnum_t nextents; /* number of extents in file */
  3999. int size; /* size of file extents */
  4000. ASSERT(ifp->if_flags & XFS_IFEXTIREC);
  4001. nextents = ifp->if_bytes / (uint)sizeof(xfs_bmbt_rec_t);
  4002. ASSERT(nextents <= XFS_LINEAR_EXTS);
  4003. size = nextents * sizeof(xfs_bmbt_rec_t);
  4004. xfs_iext_irec_compact_full(ifp);
  4005. ASSERT(ifp->if_real_bytes == XFS_IEXT_BUFSZ);
  4006. ep = ifp->if_u1.if_ext_irec->er_extbuf;
  4007. kmem_free(ifp->if_u1.if_ext_irec, sizeof(xfs_ext_irec_t));
  4008. ifp->if_flags &= ~XFS_IFEXTIREC;
  4009. ifp->if_u1.if_extents = ep;
  4010. ifp->if_bytes = size;
  4011. if (nextents < XFS_LINEAR_EXTS) {
  4012. xfs_iext_realloc_direct(ifp, size);
  4013. }
  4014. }
  4015. /*
  4016. * Free incore file extents.
  4017. */
  4018. void
  4019. xfs_iext_destroy(
  4020. xfs_ifork_t *ifp) /* inode fork pointer */
  4021. {
  4022. if (ifp->if_flags & XFS_IFEXTIREC) {
  4023. int erp_idx;
  4024. int nlists;
  4025. nlists = ifp->if_real_bytes / XFS_IEXT_BUFSZ;
  4026. for (erp_idx = nlists - 1; erp_idx >= 0 ; erp_idx--) {
  4027. xfs_iext_irec_remove(ifp, erp_idx);
  4028. }
  4029. ifp->if_flags &= ~XFS_IFEXTIREC;
  4030. } else if (ifp->if_real_bytes) {
  4031. kmem_free(ifp->if_u1.if_extents, ifp->if_real_bytes);
  4032. } else if (ifp->if_bytes) {
  4033. memset(ifp->if_u2.if_inline_ext, 0, XFS_INLINE_EXTS *
  4034. sizeof(xfs_bmbt_rec_t));
  4035. }
  4036. ifp->if_u1.if_extents = NULL;
  4037. ifp->if_real_bytes = 0;
  4038. ifp->if_bytes = 0;
  4039. }
  4040. /*
  4041. * Return a pointer to the extent record for file system block bno.
  4042. */
  4043. xfs_bmbt_rec_t * /* pointer to found extent record */
  4044. xfs_iext_bno_to_ext(
  4045. xfs_ifork_t *ifp, /* inode fork pointer */
  4046. xfs_fileoff_t bno, /* block number to search for */
  4047. xfs_extnum_t *idxp) /* index of target extent */
  4048. {
  4049. xfs_bmbt_rec_t *base; /* pointer to first extent */
  4050. xfs_filblks_t blockcount = 0; /* number of blocks in extent */
  4051. xfs_bmbt_rec_t *ep = NULL; /* pointer to target extent */
  4052. xfs_ext_irec_t *erp = NULL; /* indirection array pointer */
  4053. int high; /* upper boundary in search */
  4054. xfs_extnum_t idx = 0; /* index of target extent */
  4055. int low; /* lower boundary in search */
  4056. xfs_extnum_t nextents; /* number of file extents */
  4057. xfs_fileoff_t startoff = 0; /* start offset of extent */
  4058. nextents = ifp->if_bytes / (uint)sizeof(xfs_bmbt_rec_t);
  4059. if (nextents == 0) {
  4060. *idxp = 0;
  4061. return NULL;
  4062. }
  4063. low = 0;
  4064. if (ifp->if_flags & XFS_IFEXTIREC) {
  4065. /* Find target extent list */
  4066. int erp_idx = 0;
  4067. erp = xfs_iext_bno_to_irec(ifp, bno, &erp_idx);
  4068. base = erp->er_extbuf;
  4069. high = erp->er_extcount - 1;
  4070. } else {
  4071. base = ifp->if_u1.if_extents;
  4072. high = nextents - 1;
  4073. }
  4074. /* Binary search extent records */
  4075. while (low <= high) {
  4076. idx = (low + high) >> 1;
  4077. ep = base + idx;
  4078. startoff = xfs_bmbt_get_startoff(ep);
  4079. blockcount = xfs_bmbt_get_blockcount(ep);
  4080. if (bno < startoff) {
  4081. high = idx - 1;
  4082. } else if (bno >= startoff + blockcount) {
  4083. low = idx + 1;
  4084. } else {
  4085. /* Convert back to file-based extent index */
  4086. if (ifp->if_flags & XFS_IFEXTIREC) {
  4087. idx += erp->er_extoff;
  4088. }
  4089. *idxp = idx;
  4090. return ep;
  4091. }
  4092. }
  4093. /* Convert back to file-based extent index */
  4094. if (ifp->if_flags & XFS_IFEXTIREC) {
  4095. idx += erp->er_extoff;
  4096. }
  4097. if (bno >= startoff + blockcount) {
  4098. if (++idx == nextents) {
  4099. ep = NULL;
  4100. } else {
  4101. ep = xfs_iext_get_ext(ifp, idx);
  4102. }
  4103. }
  4104. *idxp = idx;
  4105. return ep;
  4106. }
  4107. /*
  4108. * Return a pointer to the indirection array entry containing the
  4109. * extent record for filesystem block bno. Store the index of the
  4110. * target irec in *erp_idxp.
  4111. */
  4112. xfs_ext_irec_t * /* pointer to found extent record */
  4113. xfs_iext_bno_to_irec(
  4114. xfs_ifork_t *ifp, /* inode fork pointer */
  4115. xfs_fileoff_t bno, /* block number to search for */
  4116. int *erp_idxp) /* irec index of target ext list */
  4117. {
  4118. xfs_ext_irec_t *erp = NULL; /* indirection array pointer */
  4119. xfs_ext_irec_t *erp_next; /* next indirection array entry */
  4120. int erp_idx; /* indirection array index */
  4121. int nlists; /* number of extent irec's (lists) */
  4122. int high; /* binary search upper limit */
  4123. int low; /* binary search lower limit */
  4124. ASSERT(ifp->if_flags & XFS_IFEXTIREC);
  4125. nlists = ifp->if_real_bytes / XFS_IEXT_BUFSZ;
  4126. erp_idx = 0;
  4127. low = 0;
  4128. high = nlists - 1;
  4129. while (low <= high) {
  4130. erp_idx = (low + high) >> 1;
  4131. erp = &ifp->if_u1.if_ext_irec[erp_idx];
  4132. erp_next = erp_idx < nlists - 1 ? erp + 1 : NULL;
  4133. if (bno < xfs_bmbt_get_startoff(erp->er_extbuf)) {
  4134. high = erp_idx - 1;
  4135. } else if (erp_next && bno >=
  4136. xfs_bmbt_get_startoff(erp_next->er_extbuf)) {
  4137. low = erp_idx + 1;
  4138. } else {
  4139. break;
  4140. }
  4141. }
  4142. *erp_idxp = erp_idx;
  4143. return erp;
  4144. }
  4145. /*
  4146. * Return a pointer to the indirection array entry containing the
  4147. * extent record at file extent index *idxp. Store the index of the
  4148. * target irec in *erp_idxp and store the page index of the target
  4149. * extent record in *idxp.
  4150. */
  4151. xfs_ext_irec_t *
  4152. xfs_iext_idx_to_irec(
  4153. xfs_ifork_t *ifp, /* inode fork pointer */
  4154. xfs_extnum_t *idxp, /* extent index (file -> page) */
  4155. int *erp_idxp, /* pointer to target irec */
  4156. int realloc) /* new bytes were just added */
  4157. {
  4158. xfs_ext_irec_t *prev; /* pointer to previous irec */
  4159. xfs_ext_irec_t *erp = NULL; /* pointer to current irec */
  4160. int erp_idx; /* indirection array index */
  4161. int nlists; /* number of irec's (ex lists) */
  4162. int high; /* binary search upper limit */
  4163. int low; /* binary search lower limit */
  4164. xfs_extnum_t page_idx = *idxp; /* extent index in target list */
  4165. ASSERT(ifp->if_flags & XFS_IFEXTIREC);
  4166. ASSERT(page_idx >= 0 && page_idx <=
  4167. ifp->if_bytes / (uint)sizeof(xfs_bmbt_rec_t));
  4168. nlists = ifp->if_real_bytes / XFS_IEXT_BUFSZ;
  4169. erp_idx = 0;
  4170. low = 0;
  4171. high = nlists - 1;
  4172. /* Binary search extent irec's */
  4173. while (low <= high) {
  4174. erp_idx = (low + high) >> 1;
  4175. erp = &ifp->if_u1.if_ext_irec[erp_idx];
  4176. prev = erp_idx > 0 ? erp - 1 : NULL;
  4177. if (page_idx < erp->er_extoff || (page_idx == erp->er_extoff &&
  4178. realloc && prev && prev->er_extcount < XFS_LINEAR_EXTS)) {
  4179. high = erp_idx - 1;
  4180. } else if (page_idx > erp->er_extoff + erp->er_extcount ||
  4181. (page_idx == erp->er_extoff + erp->er_extcount &&
  4182. !realloc)) {
  4183. low = erp_idx + 1;
  4184. } else if (page_idx == erp->er_extoff + erp->er_extcount &&
  4185. erp->er_extcount == XFS_LINEAR_EXTS) {
  4186. ASSERT(realloc);
  4187. page_idx = 0;
  4188. erp_idx++;
  4189. erp = erp_idx < nlists ? erp + 1 : NULL;
  4190. break;
  4191. } else {
  4192. page_idx -= erp->er_extoff;
  4193. break;
  4194. }
  4195. }
  4196. *idxp = page_idx;
  4197. *erp_idxp = erp_idx;
  4198. return(erp);
  4199. }
  4200. /*
  4201. * Allocate and initialize an indirection array once the space needed
  4202. * for incore extents increases above XFS_IEXT_BUFSZ.
  4203. */
  4204. void
  4205. xfs_iext_irec_init(
  4206. xfs_ifork_t *ifp) /* inode fork pointer */
  4207. {
  4208. xfs_ext_irec_t *erp; /* indirection array pointer */
  4209. xfs_extnum_t nextents; /* number of extents in file */
  4210. ASSERT(!(ifp->if_flags & XFS_IFEXTIREC));
  4211. nextents = ifp->if_bytes / (uint)sizeof(xfs_bmbt_rec_t);
  4212. ASSERT(nextents <= XFS_LINEAR_EXTS);
  4213. erp = (xfs_ext_irec_t *)
  4214. kmem_alloc(sizeof(xfs_ext_irec_t), KM_SLEEP);
  4215. if (nextents == 0) {
  4216. ifp->if_u1.if_extents = (xfs_bmbt_rec_t *)
  4217. kmem_alloc(XFS_IEXT_BUFSZ, KM_SLEEP);
  4218. } else if (!ifp->if_real_bytes) {
  4219. xfs_iext_inline_to_direct(ifp, XFS_IEXT_BUFSZ);
  4220. } else if (ifp->if_real_bytes < XFS_IEXT_BUFSZ) {
  4221. xfs_iext_realloc_direct(ifp, XFS_IEXT_BUFSZ);
  4222. }
  4223. erp->er_extbuf = ifp->if_u1.if_extents;
  4224. erp->er_extcount = nextents;
  4225. erp->er_extoff = 0;
  4226. ifp->if_flags |= XFS_IFEXTIREC;
  4227. ifp->if_real_bytes = XFS_IEXT_BUFSZ;
  4228. ifp->if_bytes = nextents * sizeof(xfs_bmbt_rec_t);
  4229. ifp->if_u1.if_ext_irec = erp;
  4230. return;
  4231. }
  4232. /*
  4233. * Allocate and initialize a new entry in the indirection array.
  4234. */
  4235. xfs_ext_irec_t *
  4236. xfs_iext_irec_new(
  4237. xfs_ifork_t *ifp, /* inode fork pointer */
  4238. int erp_idx) /* index for new irec */
  4239. {
  4240. xfs_ext_irec_t *erp; /* indirection array pointer */
  4241. int i; /* loop counter */
  4242. int nlists; /* number of irec's (ex lists) */
  4243. ASSERT(ifp->if_flags & XFS_IFEXTIREC);
  4244. nlists = ifp->if_real_bytes / XFS_IEXT_BUFSZ;
  4245. /* Resize indirection array */
  4246. xfs_iext_realloc_indirect(ifp, ++nlists *
  4247. sizeof(xfs_ext_irec_t));
  4248. /*
  4249. * Move records down in the array so the
  4250. * new page can use erp_idx.
  4251. */
  4252. erp = ifp->if_u1.if_ext_irec;
  4253. for (i = nlists - 1; i > erp_idx; i--) {
  4254. memmove(&erp[i], &erp[i-1], sizeof(xfs_ext_irec_t));
  4255. }
  4256. ASSERT(i == erp_idx);
  4257. /* Initialize new extent record */
  4258. erp = ifp->if_u1.if_ext_irec;
  4259. erp[erp_idx].er_extbuf = (xfs_bmbt_rec_t *)
  4260. kmem_alloc(XFS_IEXT_BUFSZ, KM_SLEEP);
  4261. ifp->if_real_bytes = nlists * XFS_IEXT_BUFSZ;
  4262. memset(erp[erp_idx].er_extbuf, 0, XFS_IEXT_BUFSZ);
  4263. erp[erp_idx].er_extcount = 0;
  4264. erp[erp_idx].er_extoff = erp_idx > 0 ?
  4265. erp[erp_idx-1].er_extoff + erp[erp_idx-1].er_extcount : 0;
  4266. return (&erp[erp_idx]);
  4267. }
  4268. /*
  4269. * Remove a record from the indirection array.
  4270. */
  4271. void
  4272. xfs_iext_irec_remove(
  4273. xfs_ifork_t *ifp, /* inode fork pointer */
  4274. int erp_idx) /* irec index to remove */
  4275. {
  4276. xfs_ext_irec_t *erp; /* indirection array pointer */
  4277. int i; /* loop counter */
  4278. int nlists; /* number of irec's (ex lists) */
  4279. ASSERT(ifp->if_flags & XFS_IFEXTIREC);
  4280. nlists = ifp->if_real_bytes / XFS_IEXT_BUFSZ;
  4281. erp = &ifp->if_u1.if_ext_irec[erp_idx];
  4282. if (erp->er_extbuf) {
  4283. xfs_iext_irec_update_extoffs(ifp, erp_idx + 1,
  4284. -erp->er_extcount);
  4285. kmem_free(erp->er_extbuf, XFS_IEXT_BUFSZ);
  4286. }
  4287. /* Compact extent records */
  4288. erp = ifp->if_u1.if_ext_irec;
  4289. for (i = erp_idx; i < nlists - 1; i++) {
  4290. memmove(&erp[i], &erp[i+1], sizeof(xfs_ext_irec_t));
  4291. }
  4292. /*
  4293. * Manually free the last extent record from the indirection
  4294. * array. A call to xfs_iext_realloc_indirect() with a size
  4295. * of zero would result in a call to xfs_iext_destroy() which
  4296. * would in turn call this function again, creating a nasty
  4297. * infinite loop.
  4298. */
  4299. if (--nlists) {
  4300. xfs_iext_realloc_indirect(ifp,
  4301. nlists * sizeof(xfs_ext_irec_t));
  4302. } else {
  4303. kmem_free(ifp->if_u1.if_ext_irec,
  4304. sizeof(xfs_ext_irec_t));
  4305. }
  4306. ifp->if_real_bytes = nlists * XFS_IEXT_BUFSZ;
  4307. }
  4308. /*
  4309. * This is called to clean up large amounts of unused memory allocated
  4310. * by the indirection array. Before compacting anything though, verify
  4311. * that the indirection array is still needed and switch back to the
  4312. * linear extent list (or even the inline buffer) if possible. The
  4313. * compaction policy is as follows:
  4314. *
  4315. * Full Compaction: Extents fit into a single page (or inline buffer)
  4316. * Full Compaction: Extents occupy less than 10% of allocated space
  4317. * Partial Compaction: Extents occupy > 10% and < 50% of allocated space
  4318. * No Compaction: Extents occupy at least 50% of allocated space
  4319. */
  4320. void
  4321. xfs_iext_irec_compact(
  4322. xfs_ifork_t *ifp) /* inode fork pointer */
  4323. {
  4324. xfs_extnum_t nextents; /* number of extents in file */
  4325. int nlists; /* number of irec's (ex lists) */
  4326. ASSERT(ifp->if_flags & XFS_IFEXTIREC);
  4327. nlists = ifp->if_real_bytes / XFS_IEXT_BUFSZ;
  4328. nextents = ifp->if_bytes / (uint)sizeof(xfs_bmbt_rec_t);
  4329. if (nextents == 0) {
  4330. xfs_iext_destroy(ifp);
  4331. } else if (nextents <= XFS_INLINE_EXTS) {
  4332. xfs_iext_indirect_to_direct(ifp);
  4333. xfs_iext_direct_to_inline(ifp, nextents);
  4334. } else if (nextents <= XFS_LINEAR_EXTS) {
  4335. xfs_iext_indirect_to_direct(ifp);
  4336. } else if (nextents < (nlists * XFS_LINEAR_EXTS) >> 3) {
  4337. xfs_iext_irec_compact_full(ifp);
  4338. } else if (nextents < (nlists * XFS_LINEAR_EXTS) >> 1) {
  4339. xfs_iext_irec_compact_pages(ifp);
  4340. }
  4341. }
  4342. /*
  4343. * Combine extents from neighboring extent pages.
  4344. */
  4345. void
  4346. xfs_iext_irec_compact_pages(
  4347. xfs_ifork_t *ifp) /* inode fork pointer */
  4348. {
  4349. xfs_ext_irec_t *erp, *erp_next;/* pointers to irec entries */
  4350. int erp_idx = 0; /* indirection array index */
  4351. int nlists; /* number of irec's (ex lists) */
  4352. ASSERT(ifp->if_flags & XFS_IFEXTIREC);
  4353. nlists = ifp->if_real_bytes / XFS_IEXT_BUFSZ;
  4354. while (erp_idx < nlists - 1) {
  4355. erp = &ifp->if_u1.if_ext_irec[erp_idx];
  4356. erp_next = erp + 1;
  4357. if (erp_next->er_extcount <=
  4358. (XFS_LINEAR_EXTS - erp->er_extcount)) {
  4359. memmove(&erp->er_extbuf[erp->er_extcount],
  4360. erp_next->er_extbuf, erp_next->er_extcount *
  4361. sizeof(xfs_bmbt_rec_t));
  4362. erp->er_extcount += erp_next->er_extcount;
  4363. /*
  4364. * Free page before removing extent record
  4365. * so er_extoffs don't get modified in
  4366. * xfs_iext_irec_remove.
  4367. */
  4368. kmem_free(erp_next->er_extbuf, XFS_IEXT_BUFSZ);
  4369. erp_next->er_extbuf = NULL;
  4370. xfs_iext_irec_remove(ifp, erp_idx + 1);
  4371. nlists = ifp->if_real_bytes / XFS_IEXT_BUFSZ;
  4372. } else {
  4373. erp_idx++;
  4374. }
  4375. }
  4376. }
  4377. /*
  4378. * Fully compact the extent records managed by the indirection array.
  4379. */
  4380. void
  4381. xfs_iext_irec_compact_full(
  4382. xfs_ifork_t *ifp) /* inode fork pointer */
  4383. {
  4384. xfs_bmbt_rec_t *ep, *ep_next; /* extent record pointers */
  4385. xfs_ext_irec_t *erp, *erp_next; /* extent irec pointers */
  4386. int erp_idx = 0; /* extent irec index */
  4387. int ext_avail; /* empty entries in ex list */
  4388. int ext_diff; /* number of exts to add */
  4389. int nlists; /* number of irec's (ex lists) */
  4390. ASSERT(ifp->if_flags & XFS_IFEXTIREC);
  4391. nlists = ifp->if_real_bytes / XFS_IEXT_BUFSZ;
  4392. erp = ifp->if_u1.if_ext_irec;
  4393. ep = &erp->er_extbuf[erp->er_extcount];
  4394. erp_next = erp + 1;
  4395. ep_next = erp_next->er_extbuf;
  4396. while (erp_idx < nlists - 1) {
  4397. ext_avail = XFS_LINEAR_EXTS - erp->er_extcount;
  4398. ext_diff = MIN(ext_avail, erp_next->er_extcount);
  4399. memcpy(ep, ep_next, ext_diff * sizeof(xfs_bmbt_rec_t));
  4400. erp->er_extcount += ext_diff;
  4401. erp_next->er_extcount -= ext_diff;
  4402. /* Remove next page */
  4403. if (erp_next->er_extcount == 0) {
  4404. /*
  4405. * Free page before removing extent record
  4406. * so er_extoffs don't get modified in
  4407. * xfs_iext_irec_remove.
  4408. */
  4409. kmem_free(erp_next->er_extbuf,
  4410. erp_next->er_extcount * sizeof(xfs_bmbt_rec_t));
  4411. erp_next->er_extbuf = NULL;
  4412. xfs_iext_irec_remove(ifp, erp_idx + 1);
  4413. erp = &ifp->if_u1.if_ext_irec[erp_idx];
  4414. nlists = ifp->if_real_bytes / XFS_IEXT_BUFSZ;
  4415. /* Update next page */
  4416. } else {
  4417. /* Move rest of page up to become next new page */
  4418. memmove(erp_next->er_extbuf, ep_next,
  4419. erp_next->er_extcount * sizeof(xfs_bmbt_rec_t));
  4420. ep_next = erp_next->er_extbuf;
  4421. memset(&ep_next[erp_next->er_extcount], 0,
  4422. (XFS_LINEAR_EXTS - erp_next->er_extcount) *
  4423. sizeof(xfs_bmbt_rec_t));
  4424. }
  4425. if (erp->er_extcount == XFS_LINEAR_EXTS) {
  4426. erp_idx++;
  4427. if (erp_idx < nlists)
  4428. erp = &ifp->if_u1.if_ext_irec[erp_idx];
  4429. else
  4430. break;
  4431. }
  4432. ep = &erp->er_extbuf[erp->er_extcount];
  4433. erp_next = erp + 1;
  4434. ep_next = erp_next->er_extbuf;
  4435. }
  4436. }
  4437. /*
  4438. * This is called to update the er_extoff field in the indirection
  4439. * array when extents have been added or removed from one of the
  4440. * extent lists. erp_idx contains the irec index to begin updating
  4441. * at and ext_diff contains the number of extents that were added
  4442. * or removed.
  4443. */
  4444. void
  4445. xfs_iext_irec_update_extoffs(
  4446. xfs_ifork_t *ifp, /* inode fork pointer */
  4447. int erp_idx, /* irec index to update */
  4448. int ext_diff) /* number of new extents */
  4449. {
  4450. int i; /* loop counter */
  4451. int nlists; /* number of irec's (ex lists */
  4452. ASSERT(ifp->if_flags & XFS_IFEXTIREC);
  4453. nlists = ifp->if_real_bytes / XFS_IEXT_BUFSZ;
  4454. for (i = erp_idx; i < nlists; i++) {
  4455. ifp->if_u1.if_ext_irec[i].er_extoff += ext_diff;
  4456. }
  4457. }