xfs_inode.c 135 KB

12345678910111213141516171819202122232425262728293031323334353637383940414243444546474849505152535455565758596061626364656667686970717273747576777879808182838485868788899091929394959697989910010110210310410510610710810911011111211311411511611711811912012112212312412512612712812913013113213313413513613713813914014114214314414514614714814915015115215315415515615715815916016116216316416516616716816917017117217317417517617717817918018118218318418518618718818919019119219319419519619719819920020120220320420520620720820921021121221321421521621721821922022122222322422522622722822923023123223323423523623723823924024124224324424524624724824925025125225325425525625725825926026126226326426526626726826927027127227327427527627727827928028128228328428528628728828929029129229329429529629729829930030130230330430530630730830931031131231331431531631731831932032132232332432532632732832933033133233333433533633733833934034134234334434534634734834935035135235335435535635735835936036136236336436536636736836937037137237337437537637737837938038138238338438538638738838939039139239339439539639739839940040140240340440540640740840941041141241341441541641741841942042142242342442542642742842943043143243343443543643743843944044144244344444544644744844945045145245345445545645745845946046146246346446546646746846947047147247347447547647747847948048148248348448548648748848949049149249349449549649749849950050150250350450550650750850951051151251351451551651751851952052152252352452552652752852953053153253353453553653753853954054154254354454554654754854955055155255355455555655755855956056156256356456556656756856957057157257357457557657757857958058158258358458558658758858959059159259359459559659759859960060160260360460560660760860961061161261361461561661761861962062162262362462562662762862963063163263363463563663763863964064164264364464564664764864965065165265365465565665765865966066166266366466566666766866967067167267367467567667767867968068168268368468568668768868969069169269369469569669769869970070170270370470570670770870971071171271371471571671771871972072172272372472572672772872973073173273373473573673773873974074174274374474574674774874975075175275375475575675775875976076176276376476576676776876977077177277377477577677777877978078178278378478578678778878979079179279379479579679779879980080180280380480580680780880981081181281381481581681781881982082182282382482582682782882983083183283383483583683783883984084184284384484584684784884985085185285385485585685785885986086186286386486586686786886987087187287387487587687787887988088188288388488588688788888989089189289389489589689789889990090190290390490590690790890991091191291391491591691791891992092192292392492592692792892993093193293393493593693793893994094194294394494594694794894995095195295395495595695795895996096196296396496596696796896997097197297397497597697797897998098198298398498598698798898999099199299399499599699799899910001001100210031004100510061007100810091010101110121013101410151016101710181019102010211022102310241025102610271028102910301031103210331034103510361037103810391040104110421043104410451046104710481049105010511052105310541055105610571058105910601061106210631064106510661067106810691070107110721073107410751076107710781079108010811082108310841085108610871088108910901091109210931094109510961097109810991100110111021103110411051106110711081109111011111112111311141115111611171118111911201121112211231124112511261127112811291130113111321133113411351136113711381139114011411142114311441145114611471148114911501151115211531154115511561157115811591160116111621163116411651166116711681169117011711172117311741175117611771178117911801181118211831184118511861187118811891190119111921193119411951196119711981199120012011202120312041205120612071208120912101211121212131214121512161217121812191220122112221223122412251226122712281229123012311232123312341235123612371238123912401241124212431244124512461247124812491250125112521253125412551256125712581259126012611262126312641265126612671268126912701271127212731274127512761277127812791280128112821283128412851286128712881289129012911292129312941295129612971298129913001301130213031304130513061307130813091310131113121313131413151316131713181319132013211322132313241325132613271328132913301331133213331334133513361337133813391340134113421343134413451346134713481349135013511352135313541355135613571358135913601361136213631364136513661367136813691370137113721373137413751376137713781379138013811382138313841385138613871388138913901391139213931394139513961397139813991400140114021403140414051406140714081409141014111412141314141415141614171418141914201421142214231424142514261427142814291430143114321433143414351436143714381439144014411442144314441445144614471448144914501451145214531454145514561457145814591460146114621463146414651466146714681469147014711472147314741475147614771478147914801481148214831484148514861487148814891490149114921493149414951496149714981499150015011502150315041505150615071508150915101511151215131514151515161517151815191520152115221523152415251526152715281529153015311532153315341535153615371538153915401541154215431544154515461547154815491550155115521553155415551556155715581559156015611562156315641565156615671568156915701571157215731574157515761577157815791580158115821583158415851586158715881589159015911592159315941595159615971598159916001601160216031604160516061607160816091610161116121613161416151616161716181619162016211622162316241625162616271628162916301631163216331634163516361637163816391640164116421643164416451646164716481649165016511652165316541655165616571658165916601661166216631664166516661667166816691670167116721673167416751676167716781679168016811682168316841685168616871688168916901691169216931694169516961697169816991700170117021703170417051706170717081709171017111712171317141715171617171718171917201721172217231724172517261727172817291730173117321733173417351736173717381739174017411742174317441745174617471748174917501751175217531754175517561757175817591760176117621763176417651766176717681769177017711772177317741775177617771778177917801781178217831784178517861787178817891790179117921793179417951796179717981799180018011802180318041805180618071808180918101811181218131814181518161817181818191820182118221823182418251826182718281829183018311832183318341835183618371838183918401841184218431844184518461847184818491850185118521853185418551856185718581859186018611862186318641865186618671868186918701871187218731874187518761877187818791880188118821883188418851886188718881889189018911892189318941895189618971898189919001901190219031904190519061907190819091910191119121913191419151916191719181919192019211922192319241925192619271928192919301931193219331934193519361937193819391940194119421943194419451946194719481949195019511952195319541955195619571958195919601961196219631964196519661967196819691970197119721973197419751976197719781979198019811982198319841985198619871988198919901991199219931994199519961997199819992000200120022003200420052006200720082009201020112012201320142015201620172018201920202021202220232024202520262027202820292030203120322033203420352036203720382039204020412042204320442045204620472048204920502051205220532054205520562057205820592060206120622063206420652066206720682069207020712072207320742075207620772078207920802081208220832084208520862087208820892090209120922093209420952096209720982099210021012102210321042105210621072108210921102111211221132114211521162117211821192120212121222123212421252126212721282129213021312132213321342135213621372138213921402141214221432144214521462147214821492150215121522153215421552156215721582159216021612162216321642165216621672168216921702171217221732174217521762177217821792180218121822183218421852186218721882189219021912192219321942195219621972198219922002201220222032204220522062207220822092210221122122213221422152216221722182219222022212222222322242225222622272228222922302231223222332234223522362237223822392240224122422243224422452246224722482249225022512252225322542255225622572258225922602261226222632264226522662267226822692270227122722273227422752276227722782279228022812282228322842285228622872288228922902291229222932294229522962297229822992300230123022303230423052306230723082309231023112312231323142315231623172318231923202321232223232324232523262327232823292330233123322333233423352336233723382339234023412342234323442345234623472348234923502351235223532354235523562357235823592360236123622363236423652366236723682369237023712372237323742375237623772378237923802381238223832384238523862387238823892390239123922393239423952396239723982399240024012402240324042405240624072408240924102411241224132414241524162417241824192420242124222423242424252426242724282429243024312432243324342435243624372438243924402441244224432444244524462447244824492450245124522453245424552456245724582459246024612462246324642465246624672468246924702471247224732474247524762477247824792480248124822483248424852486248724882489249024912492249324942495249624972498249925002501250225032504250525062507250825092510251125122513251425152516251725182519252025212522252325242525252625272528252925302531253225332534253525362537253825392540254125422543254425452546254725482549255025512552255325542555255625572558255925602561256225632564256525662567256825692570257125722573257425752576257725782579258025812582258325842585258625872588258925902591259225932594259525962597259825992600260126022603260426052606260726082609261026112612261326142615261626172618261926202621262226232624262526262627262826292630263126322633263426352636263726382639264026412642264326442645264626472648264926502651265226532654265526562657265826592660266126622663266426652666266726682669267026712672267326742675267626772678267926802681268226832684268526862687268826892690269126922693269426952696269726982699270027012702270327042705270627072708270927102711271227132714271527162717271827192720272127222723272427252726272727282729273027312732273327342735273627372738273927402741274227432744274527462747274827492750275127522753275427552756275727582759276027612762276327642765276627672768276927702771277227732774277527762777277827792780278127822783278427852786278727882789279027912792279327942795279627972798279928002801280228032804280528062807280828092810281128122813281428152816281728182819282028212822282328242825282628272828282928302831283228332834283528362837283828392840284128422843284428452846284728482849285028512852285328542855285628572858285928602861286228632864286528662867286828692870287128722873287428752876287728782879288028812882288328842885288628872888288928902891289228932894289528962897289828992900290129022903290429052906290729082909291029112912291329142915291629172918291929202921292229232924292529262927292829292930293129322933293429352936293729382939294029412942294329442945294629472948294929502951295229532954295529562957295829592960296129622963296429652966296729682969297029712972297329742975297629772978297929802981298229832984298529862987298829892990299129922993299429952996299729982999300030013002300330043005300630073008300930103011301230133014301530163017301830193020302130223023302430253026302730283029303030313032303330343035303630373038303930403041304230433044304530463047304830493050305130523053305430553056305730583059306030613062306330643065306630673068306930703071307230733074307530763077307830793080308130823083308430853086308730883089309030913092309330943095309630973098309931003101310231033104310531063107310831093110311131123113311431153116311731183119312031213122312331243125312631273128312931303131313231333134313531363137313831393140314131423143314431453146314731483149315031513152315331543155315631573158315931603161316231633164316531663167316831693170317131723173317431753176317731783179318031813182318331843185318631873188318931903191319231933194319531963197319831993200320132023203320432053206320732083209321032113212321332143215321632173218321932203221322232233224322532263227322832293230323132323233323432353236323732383239324032413242324332443245324632473248324932503251325232533254325532563257325832593260326132623263326432653266326732683269327032713272327332743275327632773278327932803281328232833284328532863287328832893290329132923293329432953296329732983299330033013302330333043305330633073308330933103311331233133314331533163317331833193320332133223323332433253326332733283329333033313332333333343335333633373338333933403341334233433344334533463347334833493350335133523353335433553356335733583359336033613362336333643365336633673368336933703371337233733374337533763377337833793380338133823383338433853386338733883389339033913392339333943395339633973398339934003401340234033404340534063407340834093410341134123413341434153416341734183419342034213422342334243425342634273428342934303431343234333434343534363437343834393440344134423443344434453446344734483449345034513452345334543455345634573458345934603461346234633464346534663467346834693470347134723473347434753476347734783479348034813482348334843485348634873488348934903491349234933494349534963497349834993500350135023503350435053506350735083509351035113512351335143515351635173518351935203521352235233524352535263527352835293530353135323533353435353536353735383539354035413542354335443545354635473548354935503551355235533554355535563557355835593560356135623563356435653566356735683569357035713572357335743575357635773578357935803581358235833584358535863587358835893590359135923593359435953596359735983599360036013602360336043605360636073608360936103611361236133614361536163617361836193620362136223623362436253626362736283629363036313632363336343635363636373638363936403641364236433644364536463647364836493650365136523653365436553656365736583659366036613662366336643665366636673668366936703671367236733674367536763677367836793680368136823683368436853686368736883689369036913692369336943695369636973698369937003701370237033704370537063707370837093710371137123713371437153716371737183719372037213722372337243725372637273728372937303731373237333734373537363737373837393740374137423743374437453746374737483749375037513752375337543755375637573758375937603761376237633764376537663767376837693770377137723773377437753776377737783779378037813782378337843785378637873788378937903791379237933794379537963797379837993800380138023803380438053806380738083809381038113812381338143815381638173818381938203821382238233824382538263827382838293830383138323833383438353836383738383839384038413842384338443845384638473848384938503851385238533854385538563857385838593860386138623863386438653866386738683869387038713872387338743875387638773878387938803881388238833884388538863887388838893890389138923893389438953896389738983899390039013902390339043905390639073908390939103911391239133914391539163917391839193920392139223923392439253926392739283929393039313932393339343935393639373938393939403941394239433944394539463947394839493950395139523953395439553956395739583959396039613962396339643965396639673968396939703971397239733974397539763977397839793980398139823983398439853986398739883989399039913992399339943995399639973998399940004001400240034004400540064007400840094010401140124013401440154016401740184019402040214022402340244025402640274028402940304031403240334034403540364037403840394040404140424043404440454046404740484049405040514052405340544055405640574058405940604061406240634064406540664067406840694070407140724073407440754076407740784079408040814082408340844085408640874088408940904091409240934094409540964097409840994100410141024103410441054106410741084109411041114112411341144115411641174118411941204121412241234124412541264127412841294130413141324133413441354136413741384139414041414142414341444145414641474148414941504151415241534154415541564157415841594160416141624163416441654166416741684169417041714172417341744175417641774178417941804181418241834184418541864187418841894190419141924193419441954196419741984199420042014202420342044205420642074208420942104211421242134214421542164217421842194220422142224223422442254226422742284229423042314232423342344235423642374238423942404241424242434244424542464247424842494250425142524253425442554256425742584259426042614262426342644265426642674268426942704271427242734274427542764277427842794280428142824283428442854286428742884289429042914292429342944295429642974298429943004301430243034304430543064307430843094310431143124313431443154316431743184319432043214322432343244325432643274328432943304331433243334334433543364337433843394340434143424343434443454346434743484349435043514352435343544355435643574358435943604361436243634364436543664367436843694370437143724373437443754376437743784379438043814382438343844385438643874388438943904391439243934394439543964397439843994400440144024403440444054406440744084409441044114412441344144415441644174418441944204421442244234424442544264427442844294430443144324433443444354436443744384439444044414442444344444445444644474448444944504451445244534454445544564457445844594460446144624463446444654466446744684469447044714472447344744475447644774478447944804481448244834484448544864487448844894490449144924493449444954496449744984499450045014502450345044505450645074508450945104511451245134514451545164517451845194520452145224523452445254526452745284529453045314532453345344535453645374538453945404541454245434544454545464547454845494550455145524553455445554556455745584559456045614562456345644565456645674568456945704571457245734574457545764577457845794580458145824583458445854586458745884589459045914592459345944595459645974598459946004601460246034604460546064607460846094610461146124613461446154616461746184619462046214622462346244625462646274628462946304631463246334634463546364637463846394640464146424643464446454646464746484649465046514652465346544655465646574658465946604661466246634664466546664667466846694670467146724673467446754676467746784679468046814682468346844685
  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 <linux/log2.h>
  19. #include "xfs.h"
  20. #include "xfs_fs.h"
  21. #include "xfs_types.h"
  22. #include "xfs_bit.h"
  23. #include "xfs_log.h"
  24. #include "xfs_inum.h"
  25. #include "xfs_imap.h"
  26. #include "xfs_trans.h"
  27. #include "xfs_trans_priv.h"
  28. #include "xfs_sb.h"
  29. #include "xfs_ag.h"
  30. #include "xfs_dir2.h"
  31. #include "xfs_dmapi.h"
  32. #include "xfs_mount.h"
  33. #include "xfs_bmap_btree.h"
  34. #include "xfs_alloc_btree.h"
  35. #include "xfs_ialloc_btree.h"
  36. #include "xfs_dir2_sf.h"
  37. #include "xfs_attr_sf.h"
  38. #include "xfs_dinode.h"
  39. #include "xfs_inode.h"
  40. #include "xfs_buf_item.h"
  41. #include "xfs_inode_item.h"
  42. #include "xfs_btree.h"
  43. #include "xfs_alloc.h"
  44. #include "xfs_ialloc.h"
  45. #include "xfs_bmap.h"
  46. #include "xfs_rw.h"
  47. #include "xfs_error.h"
  48. #include "xfs_utils.h"
  49. #include "xfs_dir2_trace.h"
  50. #include "xfs_quota.h"
  51. #include "xfs_acl.h"
  52. #include "xfs_filestream.h"
  53. #include "xfs_vnodeops.h"
  54. kmem_zone_t *xfs_ifork_zone;
  55. kmem_zone_t *xfs_inode_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. xfs_exntfmt_t fmt)
  75. {
  76. xfs_bmbt_irec_t irec;
  77. xfs_bmbt_rec_host_t rec;
  78. int i;
  79. for (i = 0; i < nrecs; i++) {
  80. xfs_bmbt_rec_host_t *ep = xfs_iext_get_ext(ifp, i);
  81. rec.l0 = get_unaligned(&ep->l0);
  82. rec.l1 = get_unaligned(&ep->l1);
  83. xfs_bmbt_get_all(&rec, &irec);
  84. if (fmt == XFS_EXTFMT_NOSTATE)
  85. ASSERT(irec.br_state == XFS_EXT_NORM);
  86. }
  87. }
  88. #else /* DEBUG */
  89. #define xfs_validate_extents(ifp, nrecs, fmt)
  90. #endif /* DEBUG */
  91. /*
  92. * Check that none of the inode's in the buffer have a next
  93. * unlinked field of 0.
  94. */
  95. #if defined(DEBUG)
  96. void
  97. xfs_inobp_check(
  98. xfs_mount_t *mp,
  99. xfs_buf_t *bp)
  100. {
  101. int i;
  102. int j;
  103. xfs_dinode_t *dip;
  104. j = mp->m_inode_cluster_size >> mp->m_sb.sb_inodelog;
  105. for (i = 0; i < j; i++) {
  106. dip = (xfs_dinode_t *)xfs_buf_offset(bp,
  107. i * mp->m_sb.sb_inodesize);
  108. if (!dip->di_next_unlinked) {
  109. xfs_fs_cmn_err(CE_ALERT, mp,
  110. "Detected a bogus zero next_unlinked field in incore inode buffer 0x%p. About to pop an ASSERT.",
  111. bp);
  112. ASSERT(dip->di_next_unlinked);
  113. }
  114. }
  115. }
  116. #endif
  117. /*
  118. * Find the buffer associated with the given inode map
  119. * We do basic validation checks on the buffer once it has been
  120. * retrieved from disk.
  121. */
  122. STATIC int
  123. xfs_imap_to_bp(
  124. xfs_mount_t *mp,
  125. xfs_trans_t *tp,
  126. xfs_imap_t *imap,
  127. xfs_buf_t **bpp,
  128. uint buf_flags,
  129. uint imap_flags)
  130. {
  131. int error;
  132. int i;
  133. int ni;
  134. xfs_buf_t *bp;
  135. error = xfs_trans_read_buf(mp, tp, mp->m_ddev_targp, imap->im_blkno,
  136. (int)imap->im_len, buf_flags, &bp);
  137. if (error) {
  138. if (error != EAGAIN) {
  139. cmn_err(CE_WARN,
  140. "xfs_imap_to_bp: xfs_trans_read_buf()returned "
  141. "an error %d on %s. Returning error.",
  142. error, mp->m_fsname);
  143. } else {
  144. ASSERT(buf_flags & XFS_BUF_TRYLOCK);
  145. }
  146. return error;
  147. }
  148. /*
  149. * Validate the magic number and version of every inode in the buffer
  150. * (if DEBUG kernel) or the first inode in the buffer, otherwise.
  151. */
  152. #ifdef DEBUG
  153. ni = BBTOB(imap->im_len) >> mp->m_sb.sb_inodelog;
  154. #else /* usual case */
  155. ni = 1;
  156. #endif
  157. for (i = 0; i < ni; i++) {
  158. int di_ok;
  159. xfs_dinode_t *dip;
  160. dip = (xfs_dinode_t *)xfs_buf_offset(bp,
  161. (i << mp->m_sb.sb_inodelog));
  162. di_ok = be16_to_cpu(dip->di_core.di_magic) == XFS_DINODE_MAGIC &&
  163. XFS_DINODE_GOOD_VERSION(dip->di_core.di_version);
  164. if (unlikely(XFS_TEST_ERROR(!di_ok, mp,
  165. XFS_ERRTAG_ITOBP_INOTOBP,
  166. XFS_RANDOM_ITOBP_INOTOBP))) {
  167. if (imap_flags & XFS_IMAP_BULKSTAT) {
  168. xfs_trans_brelse(tp, bp);
  169. return XFS_ERROR(EINVAL);
  170. }
  171. XFS_CORRUPTION_ERROR("xfs_imap_to_bp",
  172. XFS_ERRLEVEL_HIGH, mp, dip);
  173. #ifdef DEBUG
  174. cmn_err(CE_PANIC,
  175. "Device %s - bad inode magic/vsn "
  176. "daddr %lld #%d (magic=%x)",
  177. XFS_BUFTARG_NAME(mp->m_ddev_targp),
  178. (unsigned long long)imap->im_blkno, i,
  179. be16_to_cpu(dip->di_core.di_magic));
  180. #endif
  181. xfs_trans_brelse(tp, bp);
  182. return XFS_ERROR(EFSCORRUPTED);
  183. }
  184. }
  185. xfs_inobp_check(mp, bp);
  186. /*
  187. * Mark the buffer as an inode buffer now that it looks good
  188. */
  189. XFS_BUF_SET_VTYPE(bp, B_FS_INO);
  190. *bpp = bp;
  191. return 0;
  192. }
  193. /*
  194. * This routine is called to map an inode number within a file
  195. * system to the buffer containing the on-disk version of the
  196. * inode. It returns a pointer to the buffer containing the
  197. * on-disk inode in the bpp parameter, and in the dip parameter
  198. * it returns a pointer to the on-disk inode within that buffer.
  199. *
  200. * If a non-zero error is returned, then the contents of bpp and
  201. * dipp are undefined.
  202. *
  203. * Use xfs_imap() to determine the size and location of the
  204. * buffer to read from disk.
  205. */
  206. STATIC int
  207. xfs_inotobp(
  208. xfs_mount_t *mp,
  209. xfs_trans_t *tp,
  210. xfs_ino_t ino,
  211. xfs_dinode_t **dipp,
  212. xfs_buf_t **bpp,
  213. int *offset)
  214. {
  215. xfs_imap_t imap;
  216. xfs_buf_t *bp;
  217. int error;
  218. imap.im_blkno = 0;
  219. error = xfs_imap(mp, tp, ino, &imap, XFS_IMAP_LOOKUP);
  220. if (error)
  221. return error;
  222. error = xfs_imap_to_bp(mp, tp, &imap, &bp, XFS_BUF_LOCK, 0);
  223. if (error)
  224. return error;
  225. *dipp = (xfs_dinode_t *)xfs_buf_offset(bp, imap.im_boffset);
  226. *bpp = bp;
  227. *offset = imap.im_boffset;
  228. return 0;
  229. }
  230. /*
  231. * This routine is called to map an inode to the buffer containing
  232. * the on-disk version of the inode. It returns a pointer to the
  233. * buffer containing the on-disk inode in the bpp parameter, and in
  234. * the dip parameter it returns a pointer to the on-disk inode within
  235. * that buffer.
  236. *
  237. * If a non-zero error is returned, then the contents of bpp and
  238. * dipp are undefined.
  239. *
  240. * If the inode is new and has not yet been initialized, use xfs_imap()
  241. * to determine the size and location of the buffer to read from disk.
  242. * If the inode has already been mapped to its buffer and read in once,
  243. * then use the mapping information stored in the inode rather than
  244. * calling xfs_imap(). This allows us to avoid the overhead of looking
  245. * at the inode btree for small block file systems (see xfs_dilocate()).
  246. * We can tell whether the inode has been mapped in before by comparing
  247. * its disk block address to 0. Only uninitialized inodes will have
  248. * 0 for the disk block address.
  249. */
  250. int
  251. xfs_itobp(
  252. xfs_mount_t *mp,
  253. xfs_trans_t *tp,
  254. xfs_inode_t *ip,
  255. xfs_dinode_t **dipp,
  256. xfs_buf_t **bpp,
  257. xfs_daddr_t bno,
  258. uint imap_flags,
  259. uint buf_flags)
  260. {
  261. xfs_imap_t imap;
  262. xfs_buf_t *bp;
  263. int error;
  264. if (ip->i_blkno == (xfs_daddr_t)0) {
  265. imap.im_blkno = bno;
  266. error = xfs_imap(mp, tp, ip->i_ino, &imap,
  267. XFS_IMAP_LOOKUP | imap_flags);
  268. if (error)
  269. return error;
  270. /*
  271. * Fill in the fields in the inode that will be used to
  272. * map the inode to its buffer from now on.
  273. */
  274. ip->i_blkno = imap.im_blkno;
  275. ip->i_len = imap.im_len;
  276. ip->i_boffset = imap.im_boffset;
  277. } else {
  278. /*
  279. * We've already mapped the inode once, so just use the
  280. * mapping that we saved the first time.
  281. */
  282. imap.im_blkno = ip->i_blkno;
  283. imap.im_len = ip->i_len;
  284. imap.im_boffset = ip->i_boffset;
  285. }
  286. ASSERT(bno == 0 || bno == imap.im_blkno);
  287. error = xfs_imap_to_bp(mp, tp, &imap, &bp, buf_flags, imap_flags);
  288. if (error)
  289. return error;
  290. if (!bp) {
  291. ASSERT(buf_flags & XFS_BUF_TRYLOCK);
  292. ASSERT(tp == NULL);
  293. *bpp = NULL;
  294. return EAGAIN;
  295. }
  296. *dipp = (xfs_dinode_t *)xfs_buf_offset(bp, imap.im_boffset);
  297. *bpp = bp;
  298. return 0;
  299. }
  300. /*
  301. * Move inode type and inode format specific information from the
  302. * on-disk inode to the in-core inode. For fifos, devs, and sockets
  303. * this means set if_rdev to the proper value. For files, directories,
  304. * and symlinks this means to bring in the in-line data or extent
  305. * pointers. For a file in B-tree format, only the root is immediately
  306. * brought in-core. The rest will be in-lined in if_extents when it
  307. * is first referenced (see xfs_iread_extents()).
  308. */
  309. STATIC int
  310. xfs_iformat(
  311. xfs_inode_t *ip,
  312. xfs_dinode_t *dip)
  313. {
  314. xfs_attr_shortform_t *atp;
  315. int size;
  316. int error;
  317. xfs_fsize_t di_size;
  318. ip->i_df.if_ext_max =
  319. XFS_IFORK_DSIZE(ip) / (uint)sizeof(xfs_bmbt_rec_t);
  320. error = 0;
  321. if (unlikely(be32_to_cpu(dip->di_core.di_nextents) +
  322. be16_to_cpu(dip->di_core.di_anextents) >
  323. be64_to_cpu(dip->di_core.di_nblocks))) {
  324. xfs_fs_repair_cmn_err(CE_WARN, ip->i_mount,
  325. "corrupt dinode %Lu, extent total = %d, nblocks = %Lu.",
  326. (unsigned long long)ip->i_ino,
  327. (int)(be32_to_cpu(dip->di_core.di_nextents) +
  328. be16_to_cpu(dip->di_core.di_anextents)),
  329. (unsigned long long)
  330. be64_to_cpu(dip->di_core.di_nblocks));
  331. XFS_CORRUPTION_ERROR("xfs_iformat(1)", XFS_ERRLEVEL_LOW,
  332. ip->i_mount, dip);
  333. return XFS_ERROR(EFSCORRUPTED);
  334. }
  335. if (unlikely(dip->di_core.di_forkoff > ip->i_mount->m_sb.sb_inodesize)) {
  336. xfs_fs_repair_cmn_err(CE_WARN, ip->i_mount,
  337. "corrupt dinode %Lu, forkoff = 0x%x.",
  338. (unsigned long long)ip->i_ino,
  339. dip->di_core.di_forkoff);
  340. XFS_CORRUPTION_ERROR("xfs_iformat(2)", XFS_ERRLEVEL_LOW,
  341. ip->i_mount, dip);
  342. return XFS_ERROR(EFSCORRUPTED);
  343. }
  344. switch (ip->i_d.di_mode & S_IFMT) {
  345. case S_IFIFO:
  346. case S_IFCHR:
  347. case S_IFBLK:
  348. case S_IFSOCK:
  349. if (unlikely(dip->di_core.di_format != XFS_DINODE_FMT_DEV)) {
  350. XFS_CORRUPTION_ERROR("xfs_iformat(3)", XFS_ERRLEVEL_LOW,
  351. ip->i_mount, dip);
  352. return XFS_ERROR(EFSCORRUPTED);
  353. }
  354. ip->i_d.di_size = 0;
  355. ip->i_size = 0;
  356. ip->i_df.if_u2.if_rdev = be32_to_cpu(dip->di_u.di_dev);
  357. break;
  358. case S_IFREG:
  359. case S_IFLNK:
  360. case S_IFDIR:
  361. switch (dip->di_core.di_format) {
  362. case XFS_DINODE_FMT_LOCAL:
  363. /*
  364. * no local regular files yet
  365. */
  366. if (unlikely((be16_to_cpu(dip->di_core.di_mode) & S_IFMT) == S_IFREG)) {
  367. xfs_fs_repair_cmn_err(CE_WARN, ip->i_mount,
  368. "corrupt inode %Lu "
  369. "(local format for regular file).",
  370. (unsigned long long) ip->i_ino);
  371. XFS_CORRUPTION_ERROR("xfs_iformat(4)",
  372. XFS_ERRLEVEL_LOW,
  373. ip->i_mount, dip);
  374. return XFS_ERROR(EFSCORRUPTED);
  375. }
  376. di_size = be64_to_cpu(dip->di_core.di_size);
  377. if (unlikely(di_size > XFS_DFORK_DSIZE(dip, ip->i_mount))) {
  378. xfs_fs_repair_cmn_err(CE_WARN, ip->i_mount,
  379. "corrupt inode %Lu "
  380. "(bad size %Ld for local inode).",
  381. (unsigned long long) ip->i_ino,
  382. (long long) di_size);
  383. XFS_CORRUPTION_ERROR("xfs_iformat(5)",
  384. XFS_ERRLEVEL_LOW,
  385. ip->i_mount, dip);
  386. return XFS_ERROR(EFSCORRUPTED);
  387. }
  388. size = (int)di_size;
  389. error = xfs_iformat_local(ip, dip, XFS_DATA_FORK, size);
  390. break;
  391. case XFS_DINODE_FMT_EXTENTS:
  392. error = xfs_iformat_extents(ip, dip, XFS_DATA_FORK);
  393. break;
  394. case XFS_DINODE_FMT_BTREE:
  395. error = xfs_iformat_btree(ip, dip, XFS_DATA_FORK);
  396. break;
  397. default:
  398. XFS_ERROR_REPORT("xfs_iformat(6)", XFS_ERRLEVEL_LOW,
  399. ip->i_mount);
  400. return XFS_ERROR(EFSCORRUPTED);
  401. }
  402. break;
  403. default:
  404. XFS_ERROR_REPORT("xfs_iformat(7)", XFS_ERRLEVEL_LOW, ip->i_mount);
  405. return XFS_ERROR(EFSCORRUPTED);
  406. }
  407. if (error) {
  408. return error;
  409. }
  410. if (!XFS_DFORK_Q(dip))
  411. return 0;
  412. ASSERT(ip->i_afp == NULL);
  413. ip->i_afp = kmem_zone_zalloc(xfs_ifork_zone, KM_SLEEP);
  414. ip->i_afp->if_ext_max =
  415. XFS_IFORK_ASIZE(ip) / (uint)sizeof(xfs_bmbt_rec_t);
  416. switch (dip->di_core.di_aformat) {
  417. case XFS_DINODE_FMT_LOCAL:
  418. atp = (xfs_attr_shortform_t *)XFS_DFORK_APTR(dip);
  419. size = be16_to_cpu(atp->hdr.totsize);
  420. error = xfs_iformat_local(ip, dip, XFS_ATTR_FORK, size);
  421. break;
  422. case XFS_DINODE_FMT_EXTENTS:
  423. error = xfs_iformat_extents(ip, dip, XFS_ATTR_FORK);
  424. break;
  425. case XFS_DINODE_FMT_BTREE:
  426. error = xfs_iformat_btree(ip, dip, XFS_ATTR_FORK);
  427. break;
  428. default:
  429. error = XFS_ERROR(EFSCORRUPTED);
  430. break;
  431. }
  432. if (error) {
  433. kmem_zone_free(xfs_ifork_zone, ip->i_afp);
  434. ip->i_afp = NULL;
  435. xfs_idestroy_fork(ip, XFS_DATA_FORK);
  436. }
  437. return error;
  438. }
  439. /*
  440. * The file is in-lined in the on-disk inode.
  441. * If it fits into if_inline_data, then copy
  442. * it there, otherwise allocate a buffer for it
  443. * and copy the data there. Either way, set
  444. * if_data to point at the data.
  445. * If we allocate a buffer for the data, make
  446. * sure that its size is a multiple of 4 and
  447. * record the real size in i_real_bytes.
  448. */
  449. STATIC int
  450. xfs_iformat_local(
  451. xfs_inode_t *ip,
  452. xfs_dinode_t *dip,
  453. int whichfork,
  454. int size)
  455. {
  456. xfs_ifork_t *ifp;
  457. int real_size;
  458. /*
  459. * If the size is unreasonable, then something
  460. * is wrong and we just bail out rather than crash in
  461. * kmem_alloc() or memcpy() below.
  462. */
  463. if (unlikely(size > XFS_DFORK_SIZE(dip, ip->i_mount, whichfork))) {
  464. xfs_fs_repair_cmn_err(CE_WARN, ip->i_mount,
  465. "corrupt inode %Lu "
  466. "(bad size %d for local fork, size = %d).",
  467. (unsigned long long) ip->i_ino, size,
  468. XFS_DFORK_SIZE(dip, ip->i_mount, whichfork));
  469. XFS_CORRUPTION_ERROR("xfs_iformat_local", XFS_ERRLEVEL_LOW,
  470. ip->i_mount, dip);
  471. return XFS_ERROR(EFSCORRUPTED);
  472. }
  473. ifp = XFS_IFORK_PTR(ip, whichfork);
  474. real_size = 0;
  475. if (size == 0)
  476. ifp->if_u1.if_data = NULL;
  477. else if (size <= sizeof(ifp->if_u2.if_inline_data))
  478. ifp->if_u1.if_data = ifp->if_u2.if_inline_data;
  479. else {
  480. real_size = roundup(size, 4);
  481. ifp->if_u1.if_data = kmem_alloc(real_size, KM_SLEEP);
  482. }
  483. ifp->if_bytes = size;
  484. ifp->if_real_bytes = real_size;
  485. if (size)
  486. memcpy(ifp->if_u1.if_data, XFS_DFORK_PTR(dip, whichfork), size);
  487. ifp->if_flags &= ~XFS_IFEXTENTS;
  488. ifp->if_flags |= XFS_IFINLINE;
  489. return 0;
  490. }
  491. /*
  492. * The file consists of a set of extents all
  493. * of which fit into the on-disk inode.
  494. * If there are few enough extents to fit into
  495. * the if_inline_ext, then copy them there.
  496. * Otherwise allocate a buffer for them and copy
  497. * them into it. Either way, set if_extents
  498. * to point at the extents.
  499. */
  500. STATIC int
  501. xfs_iformat_extents(
  502. xfs_inode_t *ip,
  503. xfs_dinode_t *dip,
  504. int whichfork)
  505. {
  506. xfs_bmbt_rec_t *dp;
  507. xfs_ifork_t *ifp;
  508. int nex;
  509. int size;
  510. int i;
  511. ifp = XFS_IFORK_PTR(ip, whichfork);
  512. nex = XFS_DFORK_NEXTENTS(dip, whichfork);
  513. size = nex * (uint)sizeof(xfs_bmbt_rec_t);
  514. /*
  515. * If the number of extents is unreasonable, then something
  516. * is wrong and we just bail out rather than crash in
  517. * kmem_alloc() or memcpy() below.
  518. */
  519. if (unlikely(size < 0 || size > XFS_DFORK_SIZE(dip, ip->i_mount, whichfork))) {
  520. xfs_fs_repair_cmn_err(CE_WARN, ip->i_mount,
  521. "corrupt inode %Lu ((a)extents = %d).",
  522. (unsigned long long) ip->i_ino, nex);
  523. XFS_CORRUPTION_ERROR("xfs_iformat_extents(1)", XFS_ERRLEVEL_LOW,
  524. ip->i_mount, dip);
  525. return XFS_ERROR(EFSCORRUPTED);
  526. }
  527. ifp->if_real_bytes = 0;
  528. if (nex == 0)
  529. ifp->if_u1.if_extents = NULL;
  530. else if (nex <= XFS_INLINE_EXTS)
  531. ifp->if_u1.if_extents = ifp->if_u2.if_inline_ext;
  532. else
  533. xfs_iext_add(ifp, 0, nex);
  534. ifp->if_bytes = size;
  535. if (size) {
  536. dp = (xfs_bmbt_rec_t *) XFS_DFORK_PTR(dip, whichfork);
  537. xfs_validate_extents(ifp, nex, XFS_EXTFMT_INODE(ip));
  538. for (i = 0; i < nex; i++, dp++) {
  539. xfs_bmbt_rec_host_t *ep = xfs_iext_get_ext(ifp, i);
  540. ep->l0 = be64_to_cpu(get_unaligned(&dp->l0));
  541. ep->l1 = be64_to_cpu(get_unaligned(&dp->l1));
  542. }
  543. XFS_BMAP_TRACE_EXLIST(ip, nex, whichfork);
  544. if (whichfork != XFS_DATA_FORK ||
  545. XFS_EXTFMT_INODE(ip) == XFS_EXTFMT_NOSTATE)
  546. if (unlikely(xfs_check_nostate_extents(
  547. ifp, 0, nex))) {
  548. XFS_ERROR_REPORT("xfs_iformat_extents(2)",
  549. XFS_ERRLEVEL_LOW,
  550. ip->i_mount);
  551. return XFS_ERROR(EFSCORRUPTED);
  552. }
  553. }
  554. ifp->if_flags |= XFS_IFEXTENTS;
  555. return 0;
  556. }
  557. /*
  558. * The file has too many extents to fit into
  559. * the inode, so they are in B-tree format.
  560. * Allocate a buffer for the root of the B-tree
  561. * and copy the root into it. The i_extents
  562. * field will remain NULL until all of the
  563. * extents are read in (when they are needed).
  564. */
  565. STATIC int
  566. xfs_iformat_btree(
  567. xfs_inode_t *ip,
  568. xfs_dinode_t *dip,
  569. int whichfork)
  570. {
  571. xfs_bmdr_block_t *dfp;
  572. xfs_ifork_t *ifp;
  573. /* REFERENCED */
  574. int nrecs;
  575. int size;
  576. ifp = XFS_IFORK_PTR(ip, whichfork);
  577. dfp = (xfs_bmdr_block_t *)XFS_DFORK_PTR(dip, whichfork);
  578. size = XFS_BMAP_BROOT_SPACE(dfp);
  579. nrecs = XFS_BMAP_BROOT_NUMRECS(dfp);
  580. /*
  581. * blow out if -- fork has less extents than can fit in
  582. * fork (fork shouldn't be a btree format), root btree
  583. * block has more records than can fit into the fork,
  584. * or the number of extents is greater than the number of
  585. * blocks.
  586. */
  587. if (unlikely(XFS_IFORK_NEXTENTS(ip, whichfork) <= ifp->if_ext_max
  588. || XFS_BMDR_SPACE_CALC(nrecs) >
  589. XFS_DFORK_SIZE(dip, ip->i_mount, whichfork)
  590. || XFS_IFORK_NEXTENTS(ip, whichfork) > ip->i_d.di_nblocks)) {
  591. xfs_fs_repair_cmn_err(CE_WARN, ip->i_mount,
  592. "corrupt inode %Lu (btree).",
  593. (unsigned long long) ip->i_ino);
  594. XFS_ERROR_REPORT("xfs_iformat_btree", XFS_ERRLEVEL_LOW,
  595. ip->i_mount);
  596. return XFS_ERROR(EFSCORRUPTED);
  597. }
  598. ifp->if_broot_bytes = size;
  599. ifp->if_broot = kmem_alloc(size, KM_SLEEP);
  600. ASSERT(ifp->if_broot != NULL);
  601. /*
  602. * Copy and convert from the on-disk structure
  603. * to the in-memory structure.
  604. */
  605. xfs_bmdr_to_bmbt(dfp, XFS_DFORK_SIZE(dip, ip->i_mount, whichfork),
  606. ifp->if_broot, size);
  607. ifp->if_flags &= ~XFS_IFEXTENTS;
  608. ifp->if_flags |= XFS_IFBROOT;
  609. return 0;
  610. }
  611. void
  612. xfs_dinode_from_disk(
  613. xfs_icdinode_t *to,
  614. xfs_dinode_core_t *from)
  615. {
  616. to->di_magic = be16_to_cpu(from->di_magic);
  617. to->di_mode = be16_to_cpu(from->di_mode);
  618. to->di_version = from ->di_version;
  619. to->di_format = from->di_format;
  620. to->di_onlink = be16_to_cpu(from->di_onlink);
  621. to->di_uid = be32_to_cpu(from->di_uid);
  622. to->di_gid = be32_to_cpu(from->di_gid);
  623. to->di_nlink = be32_to_cpu(from->di_nlink);
  624. to->di_projid = be16_to_cpu(from->di_projid);
  625. memcpy(to->di_pad, from->di_pad, sizeof(to->di_pad));
  626. to->di_flushiter = be16_to_cpu(from->di_flushiter);
  627. to->di_atime.t_sec = be32_to_cpu(from->di_atime.t_sec);
  628. to->di_atime.t_nsec = be32_to_cpu(from->di_atime.t_nsec);
  629. to->di_mtime.t_sec = be32_to_cpu(from->di_mtime.t_sec);
  630. to->di_mtime.t_nsec = be32_to_cpu(from->di_mtime.t_nsec);
  631. to->di_ctime.t_sec = be32_to_cpu(from->di_ctime.t_sec);
  632. to->di_ctime.t_nsec = be32_to_cpu(from->di_ctime.t_nsec);
  633. to->di_size = be64_to_cpu(from->di_size);
  634. to->di_nblocks = be64_to_cpu(from->di_nblocks);
  635. to->di_extsize = be32_to_cpu(from->di_extsize);
  636. to->di_nextents = be32_to_cpu(from->di_nextents);
  637. to->di_anextents = be16_to_cpu(from->di_anextents);
  638. to->di_forkoff = from->di_forkoff;
  639. to->di_aformat = from->di_aformat;
  640. to->di_dmevmask = be32_to_cpu(from->di_dmevmask);
  641. to->di_dmstate = be16_to_cpu(from->di_dmstate);
  642. to->di_flags = be16_to_cpu(from->di_flags);
  643. to->di_gen = be32_to_cpu(from->di_gen);
  644. }
  645. void
  646. xfs_dinode_to_disk(
  647. xfs_dinode_core_t *to,
  648. xfs_icdinode_t *from)
  649. {
  650. to->di_magic = cpu_to_be16(from->di_magic);
  651. to->di_mode = cpu_to_be16(from->di_mode);
  652. to->di_version = from ->di_version;
  653. to->di_format = from->di_format;
  654. to->di_onlink = cpu_to_be16(from->di_onlink);
  655. to->di_uid = cpu_to_be32(from->di_uid);
  656. to->di_gid = cpu_to_be32(from->di_gid);
  657. to->di_nlink = cpu_to_be32(from->di_nlink);
  658. to->di_projid = cpu_to_be16(from->di_projid);
  659. memcpy(to->di_pad, from->di_pad, sizeof(to->di_pad));
  660. to->di_flushiter = cpu_to_be16(from->di_flushiter);
  661. to->di_atime.t_sec = cpu_to_be32(from->di_atime.t_sec);
  662. to->di_atime.t_nsec = cpu_to_be32(from->di_atime.t_nsec);
  663. to->di_mtime.t_sec = cpu_to_be32(from->di_mtime.t_sec);
  664. to->di_mtime.t_nsec = cpu_to_be32(from->di_mtime.t_nsec);
  665. to->di_ctime.t_sec = cpu_to_be32(from->di_ctime.t_sec);
  666. to->di_ctime.t_nsec = cpu_to_be32(from->di_ctime.t_nsec);
  667. to->di_size = cpu_to_be64(from->di_size);
  668. to->di_nblocks = cpu_to_be64(from->di_nblocks);
  669. to->di_extsize = cpu_to_be32(from->di_extsize);
  670. to->di_nextents = cpu_to_be32(from->di_nextents);
  671. to->di_anextents = cpu_to_be16(from->di_anextents);
  672. to->di_forkoff = from->di_forkoff;
  673. to->di_aformat = from->di_aformat;
  674. to->di_dmevmask = cpu_to_be32(from->di_dmevmask);
  675. to->di_dmstate = cpu_to_be16(from->di_dmstate);
  676. to->di_flags = cpu_to_be16(from->di_flags);
  677. to->di_gen = cpu_to_be32(from->di_gen);
  678. }
  679. STATIC uint
  680. _xfs_dic2xflags(
  681. __uint16_t di_flags)
  682. {
  683. uint flags = 0;
  684. if (di_flags & XFS_DIFLAG_ANY) {
  685. if (di_flags & XFS_DIFLAG_REALTIME)
  686. flags |= XFS_XFLAG_REALTIME;
  687. if (di_flags & XFS_DIFLAG_PREALLOC)
  688. flags |= XFS_XFLAG_PREALLOC;
  689. if (di_flags & XFS_DIFLAG_IMMUTABLE)
  690. flags |= XFS_XFLAG_IMMUTABLE;
  691. if (di_flags & XFS_DIFLAG_APPEND)
  692. flags |= XFS_XFLAG_APPEND;
  693. if (di_flags & XFS_DIFLAG_SYNC)
  694. flags |= XFS_XFLAG_SYNC;
  695. if (di_flags & XFS_DIFLAG_NOATIME)
  696. flags |= XFS_XFLAG_NOATIME;
  697. if (di_flags & XFS_DIFLAG_NODUMP)
  698. flags |= XFS_XFLAG_NODUMP;
  699. if (di_flags & XFS_DIFLAG_RTINHERIT)
  700. flags |= XFS_XFLAG_RTINHERIT;
  701. if (di_flags & XFS_DIFLAG_PROJINHERIT)
  702. flags |= XFS_XFLAG_PROJINHERIT;
  703. if (di_flags & XFS_DIFLAG_NOSYMLINKS)
  704. flags |= XFS_XFLAG_NOSYMLINKS;
  705. if (di_flags & XFS_DIFLAG_EXTSIZE)
  706. flags |= XFS_XFLAG_EXTSIZE;
  707. if (di_flags & XFS_DIFLAG_EXTSZINHERIT)
  708. flags |= XFS_XFLAG_EXTSZINHERIT;
  709. if (di_flags & XFS_DIFLAG_NODEFRAG)
  710. flags |= XFS_XFLAG_NODEFRAG;
  711. if (di_flags & XFS_DIFLAG_FILESTREAM)
  712. flags |= XFS_XFLAG_FILESTREAM;
  713. }
  714. return flags;
  715. }
  716. uint
  717. xfs_ip2xflags(
  718. xfs_inode_t *ip)
  719. {
  720. xfs_icdinode_t *dic = &ip->i_d;
  721. return _xfs_dic2xflags(dic->di_flags) |
  722. (XFS_IFORK_Q(ip) ? XFS_XFLAG_HASATTR : 0);
  723. }
  724. uint
  725. xfs_dic2xflags(
  726. xfs_dinode_t *dip)
  727. {
  728. xfs_dinode_core_t *dic = &dip->di_core;
  729. return _xfs_dic2xflags(be16_to_cpu(dic->di_flags)) |
  730. (XFS_DFORK_Q(dip) ? XFS_XFLAG_HASATTR : 0);
  731. }
  732. /*
  733. * Given a mount structure and an inode number, return a pointer
  734. * to a newly allocated in-core inode corresponding to the given
  735. * inode number.
  736. *
  737. * Initialize the inode's attributes and extent pointers if it
  738. * already has them (it will not if the inode has no links).
  739. */
  740. int
  741. xfs_iread(
  742. xfs_mount_t *mp,
  743. xfs_trans_t *tp,
  744. xfs_ino_t ino,
  745. xfs_inode_t **ipp,
  746. xfs_daddr_t bno,
  747. uint imap_flags)
  748. {
  749. xfs_buf_t *bp;
  750. xfs_dinode_t *dip;
  751. xfs_inode_t *ip;
  752. int error;
  753. ASSERT(xfs_inode_zone != NULL);
  754. ip = kmem_zone_zalloc(xfs_inode_zone, KM_SLEEP);
  755. ip->i_ino = ino;
  756. ip->i_mount = mp;
  757. atomic_set(&ip->i_iocount, 0);
  758. spin_lock_init(&ip->i_flags_lock);
  759. /*
  760. * Get pointer's to the on-disk inode and the buffer containing it.
  761. * If the inode number refers to a block outside the file system
  762. * then xfs_itobp() will return NULL. In this case we should
  763. * return NULL as well. Set i_blkno to 0 so that xfs_itobp() will
  764. * know that this is a new incore inode.
  765. */
  766. error = xfs_itobp(mp, tp, ip, &dip, &bp, bno, imap_flags, XFS_BUF_LOCK);
  767. if (error) {
  768. kmem_zone_free(xfs_inode_zone, ip);
  769. return error;
  770. }
  771. /*
  772. * Initialize inode's trace buffers.
  773. * Do this before xfs_iformat in case it adds entries.
  774. */
  775. #ifdef XFS_INODE_TRACE
  776. ip->i_trace = ktrace_alloc(INODE_TRACE_SIZE, KM_SLEEP);
  777. #endif
  778. #ifdef XFS_BMAP_TRACE
  779. ip->i_xtrace = ktrace_alloc(XFS_BMAP_KTRACE_SIZE, KM_SLEEP);
  780. #endif
  781. #ifdef XFS_BMBT_TRACE
  782. ip->i_btrace = ktrace_alloc(XFS_BMBT_KTRACE_SIZE, KM_SLEEP);
  783. #endif
  784. #ifdef XFS_RW_TRACE
  785. ip->i_rwtrace = ktrace_alloc(XFS_RW_KTRACE_SIZE, KM_SLEEP);
  786. #endif
  787. #ifdef XFS_ILOCK_TRACE
  788. ip->i_lock_trace = ktrace_alloc(XFS_ILOCK_KTRACE_SIZE, KM_SLEEP);
  789. #endif
  790. #ifdef XFS_DIR2_TRACE
  791. ip->i_dir_trace = ktrace_alloc(XFS_DIR2_KTRACE_SIZE, KM_SLEEP);
  792. #endif
  793. /*
  794. * If we got something that isn't an inode it means someone
  795. * (nfs or dmi) has a stale handle.
  796. */
  797. if (be16_to_cpu(dip->di_core.di_magic) != XFS_DINODE_MAGIC) {
  798. kmem_zone_free(xfs_inode_zone, ip);
  799. xfs_trans_brelse(tp, bp);
  800. #ifdef DEBUG
  801. xfs_fs_cmn_err(CE_ALERT, mp, "xfs_iread: "
  802. "dip->di_core.di_magic (0x%x) != "
  803. "XFS_DINODE_MAGIC (0x%x)",
  804. be16_to_cpu(dip->di_core.di_magic),
  805. XFS_DINODE_MAGIC);
  806. #endif /* DEBUG */
  807. return XFS_ERROR(EINVAL);
  808. }
  809. /*
  810. * If the on-disk inode is already linked to a directory
  811. * entry, copy all of the inode into the in-core inode.
  812. * xfs_iformat() handles copying in the inode format
  813. * specific information.
  814. * Otherwise, just get the truly permanent information.
  815. */
  816. if (dip->di_core.di_mode) {
  817. xfs_dinode_from_disk(&ip->i_d, &dip->di_core);
  818. error = xfs_iformat(ip, dip);
  819. if (error) {
  820. kmem_zone_free(xfs_inode_zone, ip);
  821. xfs_trans_brelse(tp, bp);
  822. #ifdef DEBUG
  823. xfs_fs_cmn_err(CE_ALERT, mp, "xfs_iread: "
  824. "xfs_iformat() returned error %d",
  825. error);
  826. #endif /* DEBUG */
  827. return error;
  828. }
  829. } else {
  830. ip->i_d.di_magic = be16_to_cpu(dip->di_core.di_magic);
  831. ip->i_d.di_version = dip->di_core.di_version;
  832. ip->i_d.di_gen = be32_to_cpu(dip->di_core.di_gen);
  833. ip->i_d.di_flushiter = be16_to_cpu(dip->di_core.di_flushiter);
  834. /*
  835. * Make sure to pull in the mode here as well in
  836. * case the inode is released without being used.
  837. * This ensures that xfs_inactive() will see that
  838. * the inode is already free and not try to mess
  839. * with the uninitialized part of it.
  840. */
  841. ip->i_d.di_mode = 0;
  842. /*
  843. * Initialize the per-fork minima and maxima for a new
  844. * inode here. xfs_iformat will do it for old inodes.
  845. */
  846. ip->i_df.if_ext_max =
  847. XFS_IFORK_DSIZE(ip) / (uint)sizeof(xfs_bmbt_rec_t);
  848. }
  849. INIT_LIST_HEAD(&ip->i_reclaim);
  850. /*
  851. * The inode format changed when we moved the link count and
  852. * made it 32 bits long. If this is an old format inode,
  853. * convert it in memory to look like a new one. If it gets
  854. * flushed to disk we will convert back before flushing or
  855. * logging it. We zero out the new projid field and the old link
  856. * count field. We'll handle clearing the pad field (the remains
  857. * of the old uuid field) when we actually convert the inode to
  858. * the new format. We don't change the version number so that we
  859. * can distinguish this from a real new format inode.
  860. */
  861. if (ip->i_d.di_version == XFS_DINODE_VERSION_1) {
  862. ip->i_d.di_nlink = ip->i_d.di_onlink;
  863. ip->i_d.di_onlink = 0;
  864. ip->i_d.di_projid = 0;
  865. }
  866. ip->i_delayed_blks = 0;
  867. ip->i_size = ip->i_d.di_size;
  868. /*
  869. * Mark the buffer containing the inode as something to keep
  870. * around for a while. This helps to keep recently accessed
  871. * meta-data in-core longer.
  872. */
  873. XFS_BUF_SET_REF(bp, XFS_INO_REF);
  874. /*
  875. * Use xfs_trans_brelse() to release the buffer containing the
  876. * on-disk inode, because it was acquired with xfs_trans_read_buf()
  877. * in xfs_itobp() above. If tp is NULL, this is just a normal
  878. * brelse(). If we're within a transaction, then xfs_trans_brelse()
  879. * will only release the buffer if it is not dirty within the
  880. * transaction. It will be OK to release the buffer in this case,
  881. * because inodes on disk are never destroyed and we will be
  882. * locking the new in-core inode before putting it in the hash
  883. * table where other processes can find it. Thus we don't have
  884. * to worry about the inode being changed just because we released
  885. * the buffer.
  886. */
  887. xfs_trans_brelse(tp, bp);
  888. *ipp = ip;
  889. return 0;
  890. }
  891. /*
  892. * Read in extents from a btree-format inode.
  893. * Allocate and fill in if_extents. Real work is done in xfs_bmap.c.
  894. */
  895. int
  896. xfs_iread_extents(
  897. xfs_trans_t *tp,
  898. xfs_inode_t *ip,
  899. int whichfork)
  900. {
  901. int error;
  902. xfs_ifork_t *ifp;
  903. xfs_extnum_t nextents;
  904. size_t size;
  905. if (unlikely(XFS_IFORK_FORMAT(ip, whichfork) != XFS_DINODE_FMT_BTREE)) {
  906. XFS_ERROR_REPORT("xfs_iread_extents", XFS_ERRLEVEL_LOW,
  907. ip->i_mount);
  908. return XFS_ERROR(EFSCORRUPTED);
  909. }
  910. nextents = XFS_IFORK_NEXTENTS(ip, whichfork);
  911. size = nextents * sizeof(xfs_bmbt_rec_t);
  912. ifp = XFS_IFORK_PTR(ip, whichfork);
  913. /*
  914. * We know that the size is valid (it's checked in iformat_btree)
  915. */
  916. ifp->if_lastex = NULLEXTNUM;
  917. ifp->if_bytes = ifp->if_real_bytes = 0;
  918. ifp->if_flags |= XFS_IFEXTENTS;
  919. xfs_iext_add(ifp, 0, nextents);
  920. error = xfs_bmap_read_extents(tp, ip, whichfork);
  921. if (error) {
  922. xfs_iext_destroy(ifp);
  923. ifp->if_flags &= ~XFS_IFEXTENTS;
  924. return error;
  925. }
  926. xfs_validate_extents(ifp, nextents, XFS_EXTFMT_INODE(ip));
  927. return 0;
  928. }
  929. /*
  930. * Allocate an inode on disk and return a copy of its in-core version.
  931. * The in-core inode is locked exclusively. Set mode, nlink, and rdev
  932. * appropriately within the inode. The uid and gid for the inode are
  933. * set according to the contents of the given cred structure.
  934. *
  935. * Use xfs_dialloc() to allocate the on-disk inode. If xfs_dialloc()
  936. * has a free inode available, call xfs_iget()
  937. * to obtain the in-core version of the allocated inode. Finally,
  938. * fill in the inode and log its initial contents. In this case,
  939. * ialloc_context would be set to NULL and call_again set to false.
  940. *
  941. * If xfs_dialloc() does not have an available inode,
  942. * it will replenish its supply by doing an allocation. Since we can
  943. * only do one allocation within a transaction without deadlocks, we
  944. * must commit the current transaction before returning the inode itself.
  945. * In this case, therefore, we will set call_again to true and return.
  946. * The caller should then commit the current transaction, start a new
  947. * transaction, and call xfs_ialloc() again to actually get the inode.
  948. *
  949. * To ensure that some other process does not grab the inode that
  950. * was allocated during the first call to xfs_ialloc(), this routine
  951. * also returns the [locked] bp pointing to the head of the freelist
  952. * as ialloc_context. The caller should hold this buffer across
  953. * the commit and pass it back into this routine on the second call.
  954. *
  955. * If we are allocating quota inodes, we do not have a parent inode
  956. * to attach to or associate with (i.e. pip == NULL) because they
  957. * are not linked into the directory structure - they are attached
  958. * directly to the superblock - and so have no parent.
  959. */
  960. int
  961. xfs_ialloc(
  962. xfs_trans_t *tp,
  963. xfs_inode_t *pip,
  964. mode_t mode,
  965. xfs_nlink_t nlink,
  966. xfs_dev_t rdev,
  967. cred_t *cr,
  968. xfs_prid_t prid,
  969. int okalloc,
  970. xfs_buf_t **ialloc_context,
  971. boolean_t *call_again,
  972. xfs_inode_t **ipp)
  973. {
  974. xfs_ino_t ino;
  975. xfs_inode_t *ip;
  976. bhv_vnode_t *vp;
  977. uint flags;
  978. int error;
  979. /*
  980. * Call the space management code to pick
  981. * the on-disk inode to be allocated.
  982. */
  983. error = xfs_dialloc(tp, pip ? pip->i_ino : 0, mode, okalloc,
  984. ialloc_context, call_again, &ino);
  985. if (error != 0) {
  986. return error;
  987. }
  988. if (*call_again || ino == NULLFSINO) {
  989. *ipp = NULL;
  990. return 0;
  991. }
  992. ASSERT(*ialloc_context == NULL);
  993. /*
  994. * Get the in-core inode with the lock held exclusively.
  995. * This is because we're setting fields here we need
  996. * to prevent others from looking at until we're done.
  997. */
  998. error = xfs_trans_iget(tp->t_mountp, tp, ino,
  999. XFS_IGET_CREATE, XFS_ILOCK_EXCL, &ip);
  1000. if (error != 0) {
  1001. return error;
  1002. }
  1003. ASSERT(ip != NULL);
  1004. vp = XFS_ITOV(ip);
  1005. ip->i_d.di_mode = (__uint16_t)mode;
  1006. ip->i_d.di_onlink = 0;
  1007. ip->i_d.di_nlink = nlink;
  1008. ASSERT(ip->i_d.di_nlink == nlink);
  1009. ip->i_d.di_uid = current_fsuid(cr);
  1010. ip->i_d.di_gid = current_fsgid(cr);
  1011. ip->i_d.di_projid = prid;
  1012. memset(&(ip->i_d.di_pad[0]), 0, sizeof(ip->i_d.di_pad));
  1013. /*
  1014. * If the superblock version is up to where we support new format
  1015. * inodes and this is currently an old format inode, then change
  1016. * the inode version number now. This way we only do the conversion
  1017. * here rather than here and in the flush/logging code.
  1018. */
  1019. if (xfs_sb_version_hasnlink(&tp->t_mountp->m_sb) &&
  1020. ip->i_d.di_version == XFS_DINODE_VERSION_1) {
  1021. ip->i_d.di_version = XFS_DINODE_VERSION_2;
  1022. /*
  1023. * We've already zeroed the old link count, the projid field,
  1024. * and the pad field.
  1025. */
  1026. }
  1027. /*
  1028. * Project ids won't be stored on disk if we are using a version 1 inode.
  1029. */
  1030. if ((prid != 0) && (ip->i_d.di_version == XFS_DINODE_VERSION_1))
  1031. xfs_bump_ino_vers2(tp, ip);
  1032. if (pip && XFS_INHERIT_GID(pip)) {
  1033. ip->i_d.di_gid = pip->i_d.di_gid;
  1034. if ((pip->i_d.di_mode & S_ISGID) && (mode & S_IFMT) == S_IFDIR) {
  1035. ip->i_d.di_mode |= S_ISGID;
  1036. }
  1037. }
  1038. /*
  1039. * If the group ID of the new file does not match the effective group
  1040. * ID or one of the supplementary group IDs, the S_ISGID bit is cleared
  1041. * (and only if the irix_sgid_inherit compatibility variable is set).
  1042. */
  1043. if ((irix_sgid_inherit) &&
  1044. (ip->i_d.di_mode & S_ISGID) &&
  1045. (!in_group_p((gid_t)ip->i_d.di_gid))) {
  1046. ip->i_d.di_mode &= ~S_ISGID;
  1047. }
  1048. ip->i_d.di_size = 0;
  1049. ip->i_size = 0;
  1050. ip->i_d.di_nextents = 0;
  1051. ASSERT(ip->i_d.di_nblocks == 0);
  1052. xfs_ichgtime(ip, XFS_ICHGTIME_CHG|XFS_ICHGTIME_ACC|XFS_ICHGTIME_MOD);
  1053. /*
  1054. * di_gen will have been taken care of in xfs_iread.
  1055. */
  1056. ip->i_d.di_extsize = 0;
  1057. ip->i_d.di_dmevmask = 0;
  1058. ip->i_d.di_dmstate = 0;
  1059. ip->i_d.di_flags = 0;
  1060. flags = XFS_ILOG_CORE;
  1061. switch (mode & S_IFMT) {
  1062. case S_IFIFO:
  1063. case S_IFCHR:
  1064. case S_IFBLK:
  1065. case S_IFSOCK:
  1066. ip->i_d.di_format = XFS_DINODE_FMT_DEV;
  1067. ip->i_df.if_u2.if_rdev = rdev;
  1068. ip->i_df.if_flags = 0;
  1069. flags |= XFS_ILOG_DEV;
  1070. break;
  1071. case S_IFREG:
  1072. if (pip && xfs_inode_is_filestream(pip)) {
  1073. error = xfs_filestream_associate(pip, ip);
  1074. if (error < 0)
  1075. return -error;
  1076. if (!error)
  1077. xfs_iflags_set(ip, XFS_IFILESTREAM);
  1078. }
  1079. /* fall through */
  1080. case S_IFDIR:
  1081. if (pip && (pip->i_d.di_flags & XFS_DIFLAG_ANY)) {
  1082. uint di_flags = 0;
  1083. if ((mode & S_IFMT) == S_IFDIR) {
  1084. if (pip->i_d.di_flags & XFS_DIFLAG_RTINHERIT)
  1085. di_flags |= XFS_DIFLAG_RTINHERIT;
  1086. if (pip->i_d.di_flags & XFS_DIFLAG_EXTSZINHERIT) {
  1087. di_flags |= XFS_DIFLAG_EXTSZINHERIT;
  1088. ip->i_d.di_extsize = pip->i_d.di_extsize;
  1089. }
  1090. } else if ((mode & S_IFMT) == S_IFREG) {
  1091. if (pip->i_d.di_flags & XFS_DIFLAG_RTINHERIT)
  1092. di_flags |= XFS_DIFLAG_REALTIME;
  1093. if (pip->i_d.di_flags & XFS_DIFLAG_EXTSZINHERIT) {
  1094. di_flags |= XFS_DIFLAG_EXTSIZE;
  1095. ip->i_d.di_extsize = pip->i_d.di_extsize;
  1096. }
  1097. }
  1098. if ((pip->i_d.di_flags & XFS_DIFLAG_NOATIME) &&
  1099. xfs_inherit_noatime)
  1100. di_flags |= XFS_DIFLAG_NOATIME;
  1101. if ((pip->i_d.di_flags & XFS_DIFLAG_NODUMP) &&
  1102. xfs_inherit_nodump)
  1103. di_flags |= XFS_DIFLAG_NODUMP;
  1104. if ((pip->i_d.di_flags & XFS_DIFLAG_SYNC) &&
  1105. xfs_inherit_sync)
  1106. di_flags |= XFS_DIFLAG_SYNC;
  1107. if ((pip->i_d.di_flags & XFS_DIFLAG_NOSYMLINKS) &&
  1108. xfs_inherit_nosymlinks)
  1109. di_flags |= XFS_DIFLAG_NOSYMLINKS;
  1110. if (pip->i_d.di_flags & XFS_DIFLAG_PROJINHERIT)
  1111. di_flags |= XFS_DIFLAG_PROJINHERIT;
  1112. if ((pip->i_d.di_flags & XFS_DIFLAG_NODEFRAG) &&
  1113. xfs_inherit_nodefrag)
  1114. di_flags |= XFS_DIFLAG_NODEFRAG;
  1115. if (pip->i_d.di_flags & XFS_DIFLAG_FILESTREAM)
  1116. di_flags |= XFS_DIFLAG_FILESTREAM;
  1117. ip->i_d.di_flags |= di_flags;
  1118. }
  1119. /* FALLTHROUGH */
  1120. case S_IFLNK:
  1121. ip->i_d.di_format = XFS_DINODE_FMT_EXTENTS;
  1122. ip->i_df.if_flags = XFS_IFEXTENTS;
  1123. ip->i_df.if_bytes = ip->i_df.if_real_bytes = 0;
  1124. ip->i_df.if_u1.if_extents = NULL;
  1125. break;
  1126. default:
  1127. ASSERT(0);
  1128. }
  1129. /*
  1130. * Attribute fork settings for new inode.
  1131. */
  1132. ip->i_d.di_aformat = XFS_DINODE_FMT_EXTENTS;
  1133. ip->i_d.di_anextents = 0;
  1134. /*
  1135. * Log the new values stuffed into the inode.
  1136. */
  1137. xfs_trans_log_inode(tp, ip, flags);
  1138. /* now that we have an i_mode we can setup inode ops and unlock */
  1139. xfs_initialize_vnode(tp->t_mountp, vp, ip);
  1140. *ipp = ip;
  1141. return 0;
  1142. }
  1143. /*
  1144. * Check to make sure that there are no blocks allocated to the
  1145. * file beyond the size of the file. We don't check this for
  1146. * files with fixed size extents or real time extents, but we
  1147. * at least do it for regular files.
  1148. */
  1149. #ifdef DEBUG
  1150. void
  1151. xfs_isize_check(
  1152. xfs_mount_t *mp,
  1153. xfs_inode_t *ip,
  1154. xfs_fsize_t isize)
  1155. {
  1156. xfs_fileoff_t map_first;
  1157. int nimaps;
  1158. xfs_bmbt_irec_t imaps[2];
  1159. if ((ip->i_d.di_mode & S_IFMT) != S_IFREG)
  1160. return;
  1161. if (XFS_IS_REALTIME_INODE(ip))
  1162. return;
  1163. if (ip->i_d.di_flags & XFS_DIFLAG_EXTSIZE)
  1164. return;
  1165. nimaps = 2;
  1166. map_first = XFS_B_TO_FSB(mp, (xfs_ufsize_t)isize);
  1167. /*
  1168. * The filesystem could be shutting down, so bmapi may return
  1169. * an error.
  1170. */
  1171. if (xfs_bmapi(NULL, ip, map_first,
  1172. (XFS_B_TO_FSB(mp,
  1173. (xfs_ufsize_t)XFS_MAXIOFFSET(mp)) -
  1174. map_first),
  1175. XFS_BMAPI_ENTIRE, NULL, 0, imaps, &nimaps,
  1176. NULL, NULL))
  1177. return;
  1178. ASSERT(nimaps == 1);
  1179. ASSERT(imaps[0].br_startblock == HOLESTARTBLOCK);
  1180. }
  1181. #endif /* DEBUG */
  1182. /*
  1183. * Calculate the last possible buffered byte in a file. This must
  1184. * include data that was buffered beyond the EOF by the write code.
  1185. * This also needs to deal with overflowing the xfs_fsize_t type
  1186. * which can happen for sizes near the limit.
  1187. *
  1188. * We also need to take into account any blocks beyond the EOF. It
  1189. * may be the case that they were buffered by a write which failed.
  1190. * In that case the pages will still be in memory, but the inode size
  1191. * will never have been updated.
  1192. */
  1193. xfs_fsize_t
  1194. xfs_file_last_byte(
  1195. xfs_inode_t *ip)
  1196. {
  1197. xfs_mount_t *mp;
  1198. xfs_fsize_t last_byte;
  1199. xfs_fileoff_t last_block;
  1200. xfs_fileoff_t size_last_block;
  1201. int error;
  1202. ASSERT(ismrlocked(&(ip->i_iolock), MR_UPDATE | MR_ACCESS));
  1203. mp = ip->i_mount;
  1204. /*
  1205. * Only check for blocks beyond the EOF if the extents have
  1206. * been read in. This eliminates the need for the inode lock,
  1207. * and it also saves us from looking when it really isn't
  1208. * necessary.
  1209. */
  1210. if (ip->i_df.if_flags & XFS_IFEXTENTS) {
  1211. error = xfs_bmap_last_offset(NULL, ip, &last_block,
  1212. XFS_DATA_FORK);
  1213. if (error) {
  1214. last_block = 0;
  1215. }
  1216. } else {
  1217. last_block = 0;
  1218. }
  1219. size_last_block = XFS_B_TO_FSB(mp, (xfs_ufsize_t)ip->i_size);
  1220. last_block = XFS_FILEOFF_MAX(last_block, size_last_block);
  1221. last_byte = XFS_FSB_TO_B(mp, last_block);
  1222. if (last_byte < 0) {
  1223. return XFS_MAXIOFFSET(mp);
  1224. }
  1225. last_byte += (1 << mp->m_writeio_log);
  1226. if (last_byte < 0) {
  1227. return XFS_MAXIOFFSET(mp);
  1228. }
  1229. return last_byte;
  1230. }
  1231. #if defined(XFS_RW_TRACE)
  1232. STATIC void
  1233. xfs_itrunc_trace(
  1234. int tag,
  1235. xfs_inode_t *ip,
  1236. int flag,
  1237. xfs_fsize_t new_size,
  1238. xfs_off_t toss_start,
  1239. xfs_off_t toss_finish)
  1240. {
  1241. if (ip->i_rwtrace == NULL) {
  1242. return;
  1243. }
  1244. ktrace_enter(ip->i_rwtrace,
  1245. (void*)((long)tag),
  1246. (void*)ip,
  1247. (void*)(unsigned long)((ip->i_d.di_size >> 32) & 0xffffffff),
  1248. (void*)(unsigned long)(ip->i_d.di_size & 0xffffffff),
  1249. (void*)((long)flag),
  1250. (void*)(unsigned long)((new_size >> 32) & 0xffffffff),
  1251. (void*)(unsigned long)(new_size & 0xffffffff),
  1252. (void*)(unsigned long)((toss_start >> 32) & 0xffffffff),
  1253. (void*)(unsigned long)(toss_start & 0xffffffff),
  1254. (void*)(unsigned long)((toss_finish >> 32) & 0xffffffff),
  1255. (void*)(unsigned long)(toss_finish & 0xffffffff),
  1256. (void*)(unsigned long)current_cpu(),
  1257. (void*)(unsigned long)current_pid(),
  1258. (void*)NULL,
  1259. (void*)NULL,
  1260. (void*)NULL);
  1261. }
  1262. #else
  1263. #define xfs_itrunc_trace(tag, ip, flag, new_size, toss_start, toss_finish)
  1264. #endif
  1265. /*
  1266. * Start the truncation of the file to new_size. The new size
  1267. * must be smaller than the current size. This routine will
  1268. * clear the buffer and page caches of file data in the removed
  1269. * range, and xfs_itruncate_finish() will remove the underlying
  1270. * disk blocks.
  1271. *
  1272. * The inode must have its I/O lock locked EXCLUSIVELY, and it
  1273. * must NOT have the inode lock held at all. This is because we're
  1274. * calling into the buffer/page cache code and we can't hold the
  1275. * inode lock when we do so.
  1276. *
  1277. * We need to wait for any direct I/Os in flight to complete before we
  1278. * proceed with the truncate. This is needed to prevent the extents
  1279. * being read or written by the direct I/Os from being removed while the
  1280. * I/O is in flight as there is no other method of synchronising
  1281. * direct I/O with the truncate operation. Also, because we hold
  1282. * the IOLOCK in exclusive mode, we prevent new direct I/Os from being
  1283. * started until the truncate completes and drops the lock. Essentially,
  1284. * the vn_iowait() call forms an I/O barrier that provides strict ordering
  1285. * between direct I/Os and the truncate operation.
  1286. *
  1287. * The flags parameter can have either the value XFS_ITRUNC_DEFINITE
  1288. * or XFS_ITRUNC_MAYBE. The XFS_ITRUNC_MAYBE value should be used
  1289. * in the case that the caller is locking things out of order and
  1290. * may not be able to call xfs_itruncate_finish() with the inode lock
  1291. * held without dropping the I/O lock. If the caller must drop the
  1292. * I/O lock before calling xfs_itruncate_finish(), then xfs_itruncate_start()
  1293. * must be called again with all the same restrictions as the initial
  1294. * call.
  1295. */
  1296. int
  1297. xfs_itruncate_start(
  1298. xfs_inode_t *ip,
  1299. uint flags,
  1300. xfs_fsize_t new_size)
  1301. {
  1302. xfs_fsize_t last_byte;
  1303. xfs_off_t toss_start;
  1304. xfs_mount_t *mp;
  1305. bhv_vnode_t *vp;
  1306. int error = 0;
  1307. ASSERT(ismrlocked(&ip->i_iolock, MR_UPDATE) != 0);
  1308. ASSERT((new_size == 0) || (new_size <= ip->i_size));
  1309. ASSERT((flags == XFS_ITRUNC_DEFINITE) ||
  1310. (flags == XFS_ITRUNC_MAYBE));
  1311. mp = ip->i_mount;
  1312. vp = XFS_ITOV(ip);
  1313. /* wait for the completion of any pending DIOs */
  1314. if (new_size < ip->i_size)
  1315. vn_iowait(ip);
  1316. /*
  1317. * Call toss_pages or flushinval_pages to get rid of pages
  1318. * overlapping the region being removed. We have to use
  1319. * the less efficient flushinval_pages in the case that the
  1320. * caller may not be able to finish the truncate without
  1321. * dropping the inode's I/O lock. Make sure
  1322. * to catch any pages brought in by buffers overlapping
  1323. * the EOF by searching out beyond the isize by our
  1324. * block size. We round new_size up to a block boundary
  1325. * so that we don't toss things on the same block as
  1326. * new_size but before it.
  1327. *
  1328. * Before calling toss_page or flushinval_pages, make sure to
  1329. * call remapf() over the same region if the file is mapped.
  1330. * This frees up mapped file references to the pages in the
  1331. * given range and for the flushinval_pages case it ensures
  1332. * that we get the latest mapped changes flushed out.
  1333. */
  1334. toss_start = XFS_B_TO_FSB(mp, (xfs_ufsize_t)new_size);
  1335. toss_start = XFS_FSB_TO_B(mp, toss_start);
  1336. if (toss_start < 0) {
  1337. /*
  1338. * The place to start tossing is beyond our maximum
  1339. * file size, so there is no way that the data extended
  1340. * out there.
  1341. */
  1342. return 0;
  1343. }
  1344. last_byte = xfs_file_last_byte(ip);
  1345. xfs_itrunc_trace(XFS_ITRUNC_START, ip, flags, new_size, toss_start,
  1346. last_byte);
  1347. if (last_byte > toss_start) {
  1348. if (flags & XFS_ITRUNC_DEFINITE) {
  1349. xfs_tosspages(ip, toss_start,
  1350. -1, FI_REMAPF_LOCKED);
  1351. } else {
  1352. error = xfs_flushinval_pages(ip, toss_start,
  1353. -1, FI_REMAPF_LOCKED);
  1354. }
  1355. }
  1356. #ifdef DEBUG
  1357. if (new_size == 0) {
  1358. ASSERT(VN_CACHED(vp) == 0);
  1359. }
  1360. #endif
  1361. return error;
  1362. }
  1363. /*
  1364. * Shrink the file to the given new_size. The new
  1365. * size must be smaller than the current size.
  1366. * This will free up the underlying blocks
  1367. * in the removed range after a call to xfs_itruncate_start()
  1368. * or xfs_atruncate_start().
  1369. *
  1370. * The transaction passed to this routine must have made
  1371. * a permanent log reservation of at least XFS_ITRUNCATE_LOG_RES.
  1372. * This routine may commit the given transaction and
  1373. * start new ones, so make sure everything involved in
  1374. * the transaction is tidy before calling here.
  1375. * Some transaction will be returned to the caller to be
  1376. * committed. The incoming transaction must already include
  1377. * the inode, and both inode locks must be held exclusively.
  1378. * The inode must also be "held" within the transaction. On
  1379. * return the inode will be "held" within the returned transaction.
  1380. * This routine does NOT require any disk space to be reserved
  1381. * for it within the transaction.
  1382. *
  1383. * The fork parameter must be either xfs_attr_fork or xfs_data_fork,
  1384. * and it indicates the fork which is to be truncated. For the
  1385. * attribute fork we only support truncation to size 0.
  1386. *
  1387. * We use the sync parameter to indicate whether or not the first
  1388. * transaction we perform might have to be synchronous. For the attr fork,
  1389. * it needs to be so if the unlink of the inode is not yet known to be
  1390. * permanent in the log. This keeps us from freeing and reusing the
  1391. * blocks of the attribute fork before the unlink of the inode becomes
  1392. * permanent.
  1393. *
  1394. * For the data fork, we normally have to run synchronously if we're
  1395. * being called out of the inactive path or we're being called
  1396. * out of the create path where we're truncating an existing file.
  1397. * Either way, the truncate needs to be sync so blocks don't reappear
  1398. * in the file with altered data in case of a crash. wsync filesystems
  1399. * can run the first case async because anything that shrinks the inode
  1400. * has to run sync so by the time we're called here from inactive, the
  1401. * inode size is permanently set to 0.
  1402. *
  1403. * Calls from the truncate path always need to be sync unless we're
  1404. * in a wsync filesystem and the file has already been unlinked.
  1405. *
  1406. * The caller is responsible for correctly setting the sync parameter.
  1407. * It gets too hard for us to guess here which path we're being called
  1408. * out of just based on inode state.
  1409. */
  1410. int
  1411. xfs_itruncate_finish(
  1412. xfs_trans_t **tp,
  1413. xfs_inode_t *ip,
  1414. xfs_fsize_t new_size,
  1415. int fork,
  1416. int sync)
  1417. {
  1418. xfs_fsblock_t first_block;
  1419. xfs_fileoff_t first_unmap_block;
  1420. xfs_fileoff_t last_block;
  1421. xfs_filblks_t unmap_len=0;
  1422. xfs_mount_t *mp;
  1423. xfs_trans_t *ntp;
  1424. int done;
  1425. int committed;
  1426. xfs_bmap_free_t free_list;
  1427. int error;
  1428. ASSERT(ismrlocked(&ip->i_iolock, MR_UPDATE) != 0);
  1429. ASSERT(ismrlocked(&ip->i_lock, MR_UPDATE) != 0);
  1430. ASSERT((new_size == 0) || (new_size <= ip->i_size));
  1431. ASSERT(*tp != NULL);
  1432. ASSERT((*tp)->t_flags & XFS_TRANS_PERM_LOG_RES);
  1433. ASSERT(ip->i_transp == *tp);
  1434. ASSERT(ip->i_itemp != NULL);
  1435. ASSERT(ip->i_itemp->ili_flags & XFS_ILI_HOLD);
  1436. ntp = *tp;
  1437. mp = (ntp)->t_mountp;
  1438. ASSERT(! XFS_NOT_DQATTACHED(mp, ip));
  1439. /*
  1440. * We only support truncating the entire attribute fork.
  1441. */
  1442. if (fork == XFS_ATTR_FORK) {
  1443. new_size = 0LL;
  1444. }
  1445. first_unmap_block = XFS_B_TO_FSB(mp, (xfs_ufsize_t)new_size);
  1446. xfs_itrunc_trace(XFS_ITRUNC_FINISH1, ip, 0, new_size, 0, 0);
  1447. /*
  1448. * The first thing we do is set the size to new_size permanently
  1449. * on disk. This way we don't have to worry about anyone ever
  1450. * being able to look at the data being freed even in the face
  1451. * of a crash. What we're getting around here is the case where
  1452. * we free a block, it is allocated to another file, it is written
  1453. * to, and then we crash. If the new data gets written to the
  1454. * file but the log buffers containing the free and reallocation
  1455. * don't, then we'd end up with garbage in the blocks being freed.
  1456. * As long as we make the new_size permanent before actually
  1457. * freeing any blocks it doesn't matter if they get writtten to.
  1458. *
  1459. * The callers must signal into us whether or not the size
  1460. * setting here must be synchronous. There are a few cases
  1461. * where it doesn't have to be synchronous. Those cases
  1462. * occur if the file is unlinked and we know the unlink is
  1463. * permanent or if the blocks being truncated are guaranteed
  1464. * to be beyond the inode eof (regardless of the link count)
  1465. * and the eof value is permanent. Both of these cases occur
  1466. * only on wsync-mounted filesystems. In those cases, we're
  1467. * guaranteed that no user will ever see the data in the blocks
  1468. * that are being truncated so the truncate can run async.
  1469. * In the free beyond eof case, the file may wind up with
  1470. * more blocks allocated to it than it needs if we crash
  1471. * and that won't get fixed until the next time the file
  1472. * is re-opened and closed but that's ok as that shouldn't
  1473. * be too many blocks.
  1474. *
  1475. * However, we can't just make all wsync xactions run async
  1476. * because there's one call out of the create path that needs
  1477. * to run sync where it's truncating an existing file to size
  1478. * 0 whose size is > 0.
  1479. *
  1480. * It's probably possible to come up with a test in this
  1481. * routine that would correctly distinguish all the above
  1482. * cases from the values of the function parameters and the
  1483. * inode state but for sanity's sake, I've decided to let the
  1484. * layers above just tell us. It's simpler to correctly figure
  1485. * out in the layer above exactly under what conditions we
  1486. * can run async and I think it's easier for others read and
  1487. * follow the logic in case something has to be changed.
  1488. * cscope is your friend -- rcc.
  1489. *
  1490. * The attribute fork is much simpler.
  1491. *
  1492. * For the attribute fork we allow the caller to tell us whether
  1493. * the unlink of the inode that led to this call is yet permanent
  1494. * in the on disk log. If it is not and we will be freeing extents
  1495. * in this inode then we make the first transaction synchronous
  1496. * to make sure that the unlink is permanent by the time we free
  1497. * the blocks.
  1498. */
  1499. if (fork == XFS_DATA_FORK) {
  1500. if (ip->i_d.di_nextents > 0) {
  1501. /*
  1502. * If we are not changing the file size then do
  1503. * not update the on-disk file size - we may be
  1504. * called from xfs_inactive_free_eofblocks(). If we
  1505. * update the on-disk file size and then the system
  1506. * crashes before the contents of the file are
  1507. * flushed to disk then the files may be full of
  1508. * holes (ie NULL files bug).
  1509. */
  1510. if (ip->i_size != new_size) {
  1511. ip->i_d.di_size = new_size;
  1512. ip->i_size = new_size;
  1513. xfs_trans_log_inode(ntp, ip, XFS_ILOG_CORE);
  1514. }
  1515. }
  1516. } else if (sync) {
  1517. ASSERT(!(mp->m_flags & XFS_MOUNT_WSYNC));
  1518. if (ip->i_d.di_anextents > 0)
  1519. xfs_trans_set_sync(ntp);
  1520. }
  1521. ASSERT(fork == XFS_DATA_FORK ||
  1522. (fork == XFS_ATTR_FORK &&
  1523. ((sync && !(mp->m_flags & XFS_MOUNT_WSYNC)) ||
  1524. (sync == 0 && (mp->m_flags & XFS_MOUNT_WSYNC)))));
  1525. /*
  1526. * Since it is possible for space to become allocated beyond
  1527. * the end of the file (in a crash where the space is allocated
  1528. * but the inode size is not yet updated), simply remove any
  1529. * blocks which show up between the new EOF and the maximum
  1530. * possible file size. If the first block to be removed is
  1531. * beyond the maximum file size (ie it is the same as last_block),
  1532. * then there is nothing to do.
  1533. */
  1534. last_block = XFS_B_TO_FSB(mp, (xfs_ufsize_t)XFS_MAXIOFFSET(mp));
  1535. ASSERT(first_unmap_block <= last_block);
  1536. done = 0;
  1537. if (last_block == first_unmap_block) {
  1538. done = 1;
  1539. } else {
  1540. unmap_len = last_block - first_unmap_block + 1;
  1541. }
  1542. while (!done) {
  1543. /*
  1544. * Free up up to XFS_ITRUNC_MAX_EXTENTS. xfs_bunmapi()
  1545. * will tell us whether it freed the entire range or
  1546. * not. If this is a synchronous mount (wsync),
  1547. * then we can tell bunmapi to keep all the
  1548. * transactions asynchronous since the unlink
  1549. * transaction that made this inode inactive has
  1550. * already hit the disk. There's no danger of
  1551. * the freed blocks being reused, there being a
  1552. * crash, and the reused blocks suddenly reappearing
  1553. * in this file with garbage in them once recovery
  1554. * runs.
  1555. */
  1556. XFS_BMAP_INIT(&free_list, &first_block);
  1557. error = xfs_bunmapi(ntp, ip,
  1558. first_unmap_block, unmap_len,
  1559. XFS_BMAPI_AFLAG(fork) |
  1560. (sync ? 0 : XFS_BMAPI_ASYNC),
  1561. XFS_ITRUNC_MAX_EXTENTS,
  1562. &first_block, &free_list,
  1563. NULL, &done);
  1564. if (error) {
  1565. /*
  1566. * If the bunmapi call encounters an error,
  1567. * return to the caller where the transaction
  1568. * can be properly aborted. We just need to
  1569. * make sure we're not holding any resources
  1570. * that we were not when we came in.
  1571. */
  1572. xfs_bmap_cancel(&free_list);
  1573. return error;
  1574. }
  1575. /*
  1576. * Duplicate the transaction that has the permanent
  1577. * reservation and commit the old transaction.
  1578. */
  1579. error = xfs_bmap_finish(tp, &free_list, &committed);
  1580. ntp = *tp;
  1581. if (error) {
  1582. /*
  1583. * If the bmap finish call encounters an error,
  1584. * return to the caller where the transaction
  1585. * can be properly aborted. We just need to
  1586. * make sure we're not holding any resources
  1587. * that we were not when we came in.
  1588. *
  1589. * Aborting from this point might lose some
  1590. * blocks in the file system, but oh well.
  1591. */
  1592. xfs_bmap_cancel(&free_list);
  1593. if (committed)
  1594. goto error_join;
  1595. return error;
  1596. }
  1597. if (committed) {
  1598. /*
  1599. * The first xact was committed, so add the inode to
  1600. * the new one. Mark it dirty so it will be logged and
  1601. * moved forward in the log as part of every commit.
  1602. */
  1603. xfs_trans_ijoin(ntp, ip,
  1604. XFS_ILOCK_EXCL | XFS_IOLOCK_EXCL);
  1605. xfs_trans_ihold(ntp, ip);
  1606. xfs_trans_log_inode(ntp, ip, XFS_ILOG_CORE);
  1607. }
  1608. ntp = xfs_trans_dup(ntp);
  1609. error = xfs_trans_commit(*tp, 0);
  1610. *tp = ntp;
  1611. if (error)
  1612. goto error_join;
  1613. error = xfs_trans_reserve(ntp, 0, XFS_ITRUNCATE_LOG_RES(mp), 0,
  1614. XFS_TRANS_PERM_LOG_RES,
  1615. XFS_ITRUNCATE_LOG_COUNT);
  1616. if (error)
  1617. goto error_join;
  1618. }
  1619. /*
  1620. * Only update the size in the case of the data fork, but
  1621. * always re-log the inode so that our permanent transaction
  1622. * can keep on rolling it forward in the log.
  1623. */
  1624. if (fork == XFS_DATA_FORK) {
  1625. xfs_isize_check(mp, ip, new_size);
  1626. /*
  1627. * If we are not changing the file size then do
  1628. * not update the on-disk file size - we may be
  1629. * called from xfs_inactive_free_eofblocks(). If we
  1630. * update the on-disk file size and then the system
  1631. * crashes before the contents of the file are
  1632. * flushed to disk then the files may be full of
  1633. * holes (ie NULL files bug).
  1634. */
  1635. if (ip->i_size != new_size) {
  1636. ip->i_d.di_size = new_size;
  1637. ip->i_size = new_size;
  1638. }
  1639. }
  1640. xfs_trans_log_inode(ntp, ip, XFS_ILOG_CORE);
  1641. ASSERT((new_size != 0) ||
  1642. (fork == XFS_ATTR_FORK) ||
  1643. (ip->i_delayed_blks == 0));
  1644. ASSERT((new_size != 0) ||
  1645. (fork == XFS_ATTR_FORK) ||
  1646. (ip->i_d.di_nextents == 0));
  1647. xfs_itrunc_trace(XFS_ITRUNC_FINISH2, ip, 0, new_size, 0, 0);
  1648. return 0;
  1649. error_join:
  1650. /*
  1651. * Add the inode being truncated to the next chained transaction. This
  1652. * keeps things simple for the higher level code, because it always
  1653. * knows that the inode is locked and held in the transaction that
  1654. * returns to it whether errors occur or not. We don't mark the inode
  1655. * dirty so that this transaction can be easily aborted if possible.
  1656. */
  1657. xfs_trans_ijoin(ntp, ip, XFS_ILOCK_EXCL | XFS_IOLOCK_EXCL);
  1658. xfs_trans_ihold(ntp, ip);
  1659. return error;
  1660. }
  1661. /*
  1662. * xfs_igrow_start
  1663. *
  1664. * Do the first part of growing a file: zero any data in the last
  1665. * block that is beyond the old EOF. We need to do this before
  1666. * the inode is joined to the transaction to modify the i_size.
  1667. * That way we can drop the inode lock and call into the buffer
  1668. * cache to get the buffer mapping the EOF.
  1669. */
  1670. int
  1671. xfs_igrow_start(
  1672. xfs_inode_t *ip,
  1673. xfs_fsize_t new_size,
  1674. cred_t *credp)
  1675. {
  1676. ASSERT(ismrlocked(&(ip->i_lock), MR_UPDATE) != 0);
  1677. ASSERT(ismrlocked(&(ip->i_iolock), MR_UPDATE) != 0);
  1678. ASSERT(new_size > ip->i_size);
  1679. /*
  1680. * Zero any pages that may have been created by
  1681. * xfs_write_file() beyond the end of the file
  1682. * and any blocks between the old and new file sizes.
  1683. */
  1684. return xfs_zero_eof(ip, new_size, ip->i_size);
  1685. }
  1686. /*
  1687. * xfs_igrow_finish
  1688. *
  1689. * This routine is called to extend the size of a file.
  1690. * The inode must have both the iolock and the ilock locked
  1691. * for update and it must be a part of the current transaction.
  1692. * The xfs_igrow_start() function must have been called previously.
  1693. * If the change_flag is not zero, the inode change timestamp will
  1694. * be updated.
  1695. */
  1696. void
  1697. xfs_igrow_finish(
  1698. xfs_trans_t *tp,
  1699. xfs_inode_t *ip,
  1700. xfs_fsize_t new_size,
  1701. int change_flag)
  1702. {
  1703. ASSERT(ismrlocked(&(ip->i_lock), MR_UPDATE) != 0);
  1704. ASSERT(ismrlocked(&(ip->i_iolock), MR_UPDATE) != 0);
  1705. ASSERT(ip->i_transp == tp);
  1706. ASSERT(new_size > ip->i_size);
  1707. /*
  1708. * Update the file size. Update the inode change timestamp
  1709. * if change_flag set.
  1710. */
  1711. ip->i_d.di_size = new_size;
  1712. ip->i_size = new_size;
  1713. if (change_flag)
  1714. xfs_ichgtime(ip, XFS_ICHGTIME_CHG);
  1715. xfs_trans_log_inode(tp, ip, XFS_ILOG_CORE);
  1716. }
  1717. /*
  1718. * This is called when the inode's link count goes to 0.
  1719. * We place the on-disk inode on a list in the AGI. It
  1720. * will be pulled from this list when the inode is freed.
  1721. */
  1722. int
  1723. xfs_iunlink(
  1724. xfs_trans_t *tp,
  1725. xfs_inode_t *ip)
  1726. {
  1727. xfs_mount_t *mp;
  1728. xfs_agi_t *agi;
  1729. xfs_dinode_t *dip;
  1730. xfs_buf_t *agibp;
  1731. xfs_buf_t *ibp;
  1732. xfs_agnumber_t agno;
  1733. xfs_daddr_t agdaddr;
  1734. xfs_agino_t agino;
  1735. short bucket_index;
  1736. int offset;
  1737. int error;
  1738. int agi_ok;
  1739. ASSERT(ip->i_d.di_nlink == 0);
  1740. ASSERT(ip->i_d.di_mode != 0);
  1741. ASSERT(ip->i_transp == tp);
  1742. mp = tp->t_mountp;
  1743. agno = XFS_INO_TO_AGNO(mp, ip->i_ino);
  1744. agdaddr = XFS_AG_DADDR(mp, agno, XFS_AGI_DADDR(mp));
  1745. /*
  1746. * Get the agi buffer first. It ensures lock ordering
  1747. * on the list.
  1748. */
  1749. error = xfs_trans_read_buf(mp, tp, mp->m_ddev_targp, agdaddr,
  1750. XFS_FSS_TO_BB(mp, 1), 0, &agibp);
  1751. if (error)
  1752. return error;
  1753. /*
  1754. * Validate the magic number of the agi block.
  1755. */
  1756. agi = XFS_BUF_TO_AGI(agibp);
  1757. agi_ok =
  1758. be32_to_cpu(agi->agi_magicnum) == XFS_AGI_MAGIC &&
  1759. XFS_AGI_GOOD_VERSION(be32_to_cpu(agi->agi_versionnum));
  1760. if (unlikely(XFS_TEST_ERROR(!agi_ok, mp, XFS_ERRTAG_IUNLINK,
  1761. XFS_RANDOM_IUNLINK))) {
  1762. XFS_CORRUPTION_ERROR("xfs_iunlink", XFS_ERRLEVEL_LOW, mp, agi);
  1763. xfs_trans_brelse(tp, agibp);
  1764. return XFS_ERROR(EFSCORRUPTED);
  1765. }
  1766. /*
  1767. * Get the index into the agi hash table for the
  1768. * list this inode will go on.
  1769. */
  1770. agino = XFS_INO_TO_AGINO(mp, ip->i_ino);
  1771. ASSERT(agino != 0);
  1772. bucket_index = agino % XFS_AGI_UNLINKED_BUCKETS;
  1773. ASSERT(agi->agi_unlinked[bucket_index]);
  1774. ASSERT(be32_to_cpu(agi->agi_unlinked[bucket_index]) != agino);
  1775. if (be32_to_cpu(agi->agi_unlinked[bucket_index]) != NULLAGINO) {
  1776. /*
  1777. * There is already another inode in the bucket we need
  1778. * to add ourselves to. Add us at the front of the list.
  1779. * Here we put the head pointer into our next pointer,
  1780. * and then we fall through to point the head at us.
  1781. */
  1782. error = xfs_itobp(mp, tp, ip, &dip, &ibp, 0, 0, XFS_BUF_LOCK);
  1783. if (error)
  1784. return error;
  1785. ASSERT(be32_to_cpu(dip->di_next_unlinked) == NULLAGINO);
  1786. /* both on-disk, don't endian flip twice */
  1787. dip->di_next_unlinked = agi->agi_unlinked[bucket_index];
  1788. offset = ip->i_boffset +
  1789. offsetof(xfs_dinode_t, di_next_unlinked);
  1790. xfs_trans_inode_buf(tp, ibp);
  1791. xfs_trans_log_buf(tp, ibp, offset,
  1792. (offset + sizeof(xfs_agino_t) - 1));
  1793. xfs_inobp_check(mp, ibp);
  1794. }
  1795. /*
  1796. * Point the bucket head pointer at the inode being inserted.
  1797. */
  1798. ASSERT(agino != 0);
  1799. agi->agi_unlinked[bucket_index] = cpu_to_be32(agino);
  1800. offset = offsetof(xfs_agi_t, agi_unlinked) +
  1801. (sizeof(xfs_agino_t) * bucket_index);
  1802. xfs_trans_log_buf(tp, agibp, offset,
  1803. (offset + sizeof(xfs_agino_t) - 1));
  1804. return 0;
  1805. }
  1806. /*
  1807. * Pull the on-disk inode from the AGI unlinked list.
  1808. */
  1809. STATIC int
  1810. xfs_iunlink_remove(
  1811. xfs_trans_t *tp,
  1812. xfs_inode_t *ip)
  1813. {
  1814. xfs_ino_t next_ino;
  1815. xfs_mount_t *mp;
  1816. xfs_agi_t *agi;
  1817. xfs_dinode_t *dip;
  1818. xfs_buf_t *agibp;
  1819. xfs_buf_t *ibp;
  1820. xfs_agnumber_t agno;
  1821. xfs_daddr_t agdaddr;
  1822. xfs_agino_t agino;
  1823. xfs_agino_t next_agino;
  1824. xfs_buf_t *last_ibp;
  1825. xfs_dinode_t *last_dip = NULL;
  1826. short bucket_index;
  1827. int offset, last_offset = 0;
  1828. int error;
  1829. int agi_ok;
  1830. /*
  1831. * First pull the on-disk inode from the AGI unlinked list.
  1832. */
  1833. mp = tp->t_mountp;
  1834. agno = XFS_INO_TO_AGNO(mp, ip->i_ino);
  1835. agdaddr = XFS_AG_DADDR(mp, agno, XFS_AGI_DADDR(mp));
  1836. /*
  1837. * Get the agi buffer first. It ensures lock ordering
  1838. * on the list.
  1839. */
  1840. error = xfs_trans_read_buf(mp, tp, mp->m_ddev_targp, agdaddr,
  1841. XFS_FSS_TO_BB(mp, 1), 0, &agibp);
  1842. if (error) {
  1843. cmn_err(CE_WARN,
  1844. "xfs_iunlink_remove: xfs_trans_read_buf() returned an error %d on %s. Returning error.",
  1845. error, mp->m_fsname);
  1846. return error;
  1847. }
  1848. /*
  1849. * Validate the magic number of the agi block.
  1850. */
  1851. agi = XFS_BUF_TO_AGI(agibp);
  1852. agi_ok =
  1853. be32_to_cpu(agi->agi_magicnum) == XFS_AGI_MAGIC &&
  1854. XFS_AGI_GOOD_VERSION(be32_to_cpu(agi->agi_versionnum));
  1855. if (unlikely(XFS_TEST_ERROR(!agi_ok, mp, XFS_ERRTAG_IUNLINK_REMOVE,
  1856. XFS_RANDOM_IUNLINK_REMOVE))) {
  1857. XFS_CORRUPTION_ERROR("xfs_iunlink_remove", XFS_ERRLEVEL_LOW,
  1858. mp, agi);
  1859. xfs_trans_brelse(tp, agibp);
  1860. cmn_err(CE_WARN,
  1861. "xfs_iunlink_remove: XFS_TEST_ERROR() returned an error on %s. Returning EFSCORRUPTED.",
  1862. mp->m_fsname);
  1863. return XFS_ERROR(EFSCORRUPTED);
  1864. }
  1865. /*
  1866. * Get the index into the agi hash table for the
  1867. * list this inode will go on.
  1868. */
  1869. agino = XFS_INO_TO_AGINO(mp, ip->i_ino);
  1870. ASSERT(agino != 0);
  1871. bucket_index = agino % XFS_AGI_UNLINKED_BUCKETS;
  1872. ASSERT(be32_to_cpu(agi->agi_unlinked[bucket_index]) != NULLAGINO);
  1873. ASSERT(agi->agi_unlinked[bucket_index]);
  1874. if (be32_to_cpu(agi->agi_unlinked[bucket_index]) == agino) {
  1875. /*
  1876. * We're at the head of the list. Get the inode's
  1877. * on-disk buffer to see if there is anyone after us
  1878. * on the list. Only modify our next pointer if it
  1879. * is not already NULLAGINO. This saves us the overhead
  1880. * of dealing with the buffer when there is no need to
  1881. * change it.
  1882. */
  1883. error = xfs_itobp(mp, tp, ip, &dip, &ibp, 0, 0, XFS_BUF_LOCK);
  1884. if (error) {
  1885. cmn_err(CE_WARN,
  1886. "xfs_iunlink_remove: xfs_itobp() returned an error %d on %s. Returning error.",
  1887. error, mp->m_fsname);
  1888. return error;
  1889. }
  1890. next_agino = be32_to_cpu(dip->di_next_unlinked);
  1891. ASSERT(next_agino != 0);
  1892. if (next_agino != NULLAGINO) {
  1893. dip->di_next_unlinked = cpu_to_be32(NULLAGINO);
  1894. offset = ip->i_boffset +
  1895. offsetof(xfs_dinode_t, di_next_unlinked);
  1896. xfs_trans_inode_buf(tp, ibp);
  1897. xfs_trans_log_buf(tp, ibp, offset,
  1898. (offset + sizeof(xfs_agino_t) - 1));
  1899. xfs_inobp_check(mp, ibp);
  1900. } else {
  1901. xfs_trans_brelse(tp, ibp);
  1902. }
  1903. /*
  1904. * Point the bucket head pointer at the next inode.
  1905. */
  1906. ASSERT(next_agino != 0);
  1907. ASSERT(next_agino != agino);
  1908. agi->agi_unlinked[bucket_index] = cpu_to_be32(next_agino);
  1909. offset = offsetof(xfs_agi_t, agi_unlinked) +
  1910. (sizeof(xfs_agino_t) * bucket_index);
  1911. xfs_trans_log_buf(tp, agibp, offset,
  1912. (offset + sizeof(xfs_agino_t) - 1));
  1913. } else {
  1914. /*
  1915. * We need to search the list for the inode being freed.
  1916. */
  1917. next_agino = be32_to_cpu(agi->agi_unlinked[bucket_index]);
  1918. last_ibp = NULL;
  1919. while (next_agino != agino) {
  1920. /*
  1921. * If the last inode wasn't the one pointing to
  1922. * us, then release its buffer since we're not
  1923. * going to do anything with it.
  1924. */
  1925. if (last_ibp != NULL) {
  1926. xfs_trans_brelse(tp, last_ibp);
  1927. }
  1928. next_ino = XFS_AGINO_TO_INO(mp, agno, next_agino);
  1929. error = xfs_inotobp(mp, tp, next_ino, &last_dip,
  1930. &last_ibp, &last_offset);
  1931. if (error) {
  1932. cmn_err(CE_WARN,
  1933. "xfs_iunlink_remove: xfs_inotobp() returned an error %d on %s. Returning error.",
  1934. error, mp->m_fsname);
  1935. return error;
  1936. }
  1937. next_agino = be32_to_cpu(last_dip->di_next_unlinked);
  1938. ASSERT(next_agino != NULLAGINO);
  1939. ASSERT(next_agino != 0);
  1940. }
  1941. /*
  1942. * Now last_ibp points to the buffer previous to us on
  1943. * the unlinked list. Pull us from the list.
  1944. */
  1945. error = xfs_itobp(mp, tp, ip, &dip, &ibp, 0, 0, XFS_BUF_LOCK);
  1946. if (error) {
  1947. cmn_err(CE_WARN,
  1948. "xfs_iunlink_remove: xfs_itobp() returned an error %d on %s. Returning error.",
  1949. error, mp->m_fsname);
  1950. return error;
  1951. }
  1952. next_agino = be32_to_cpu(dip->di_next_unlinked);
  1953. ASSERT(next_agino != 0);
  1954. ASSERT(next_agino != agino);
  1955. if (next_agino != NULLAGINO) {
  1956. dip->di_next_unlinked = cpu_to_be32(NULLAGINO);
  1957. offset = ip->i_boffset +
  1958. offsetof(xfs_dinode_t, di_next_unlinked);
  1959. xfs_trans_inode_buf(tp, ibp);
  1960. xfs_trans_log_buf(tp, ibp, offset,
  1961. (offset + sizeof(xfs_agino_t) - 1));
  1962. xfs_inobp_check(mp, ibp);
  1963. } else {
  1964. xfs_trans_brelse(tp, ibp);
  1965. }
  1966. /*
  1967. * Point the previous inode on the list to the next inode.
  1968. */
  1969. last_dip->di_next_unlinked = cpu_to_be32(next_agino);
  1970. ASSERT(next_agino != 0);
  1971. offset = last_offset + offsetof(xfs_dinode_t, di_next_unlinked);
  1972. xfs_trans_inode_buf(tp, last_ibp);
  1973. xfs_trans_log_buf(tp, last_ibp, offset,
  1974. (offset + sizeof(xfs_agino_t) - 1));
  1975. xfs_inobp_check(mp, last_ibp);
  1976. }
  1977. return 0;
  1978. }
  1979. STATIC void
  1980. xfs_ifree_cluster(
  1981. xfs_inode_t *free_ip,
  1982. xfs_trans_t *tp,
  1983. xfs_ino_t inum)
  1984. {
  1985. xfs_mount_t *mp = free_ip->i_mount;
  1986. int blks_per_cluster;
  1987. int nbufs;
  1988. int ninodes;
  1989. int i, j, found, pre_flushed;
  1990. xfs_daddr_t blkno;
  1991. xfs_buf_t *bp;
  1992. xfs_inode_t *ip, **ip_found;
  1993. xfs_inode_log_item_t *iip;
  1994. xfs_log_item_t *lip;
  1995. xfs_perag_t *pag = xfs_get_perag(mp, inum);
  1996. if (mp->m_sb.sb_blocksize >= XFS_INODE_CLUSTER_SIZE(mp)) {
  1997. blks_per_cluster = 1;
  1998. ninodes = mp->m_sb.sb_inopblock;
  1999. nbufs = XFS_IALLOC_BLOCKS(mp);
  2000. } else {
  2001. blks_per_cluster = XFS_INODE_CLUSTER_SIZE(mp) /
  2002. mp->m_sb.sb_blocksize;
  2003. ninodes = blks_per_cluster * mp->m_sb.sb_inopblock;
  2004. nbufs = XFS_IALLOC_BLOCKS(mp) / blks_per_cluster;
  2005. }
  2006. ip_found = kmem_alloc(ninodes * sizeof(xfs_inode_t *), KM_NOFS);
  2007. for (j = 0; j < nbufs; j++, inum += ninodes) {
  2008. blkno = XFS_AGB_TO_DADDR(mp, XFS_INO_TO_AGNO(mp, inum),
  2009. XFS_INO_TO_AGBNO(mp, inum));
  2010. /*
  2011. * Look for each inode in memory and attempt to lock it,
  2012. * we can be racing with flush and tail pushing here.
  2013. * any inode we get the locks on, add to an array of
  2014. * inode items to process later.
  2015. *
  2016. * The get the buffer lock, we could beat a flush
  2017. * or tail pushing thread to the lock here, in which
  2018. * case they will go looking for the inode buffer
  2019. * and fail, we need some other form of interlock
  2020. * here.
  2021. */
  2022. found = 0;
  2023. for (i = 0; i < ninodes; i++) {
  2024. read_lock(&pag->pag_ici_lock);
  2025. ip = radix_tree_lookup(&pag->pag_ici_root,
  2026. XFS_INO_TO_AGINO(mp, (inum + i)));
  2027. /* Inode not in memory or we found it already,
  2028. * nothing to do
  2029. */
  2030. if (!ip || xfs_iflags_test(ip, XFS_ISTALE)) {
  2031. read_unlock(&pag->pag_ici_lock);
  2032. continue;
  2033. }
  2034. if (xfs_inode_clean(ip)) {
  2035. read_unlock(&pag->pag_ici_lock);
  2036. continue;
  2037. }
  2038. /* If we can get the locks then add it to the
  2039. * list, otherwise by the time we get the bp lock
  2040. * below it will already be attached to the
  2041. * inode buffer.
  2042. */
  2043. /* This inode will already be locked - by us, lets
  2044. * keep it that way.
  2045. */
  2046. if (ip == free_ip) {
  2047. if (xfs_iflock_nowait(ip)) {
  2048. xfs_iflags_set(ip, XFS_ISTALE);
  2049. if (xfs_inode_clean(ip)) {
  2050. xfs_ifunlock(ip);
  2051. } else {
  2052. ip_found[found++] = ip;
  2053. }
  2054. }
  2055. read_unlock(&pag->pag_ici_lock);
  2056. continue;
  2057. }
  2058. if (xfs_ilock_nowait(ip, XFS_ILOCK_EXCL)) {
  2059. if (xfs_iflock_nowait(ip)) {
  2060. xfs_iflags_set(ip, XFS_ISTALE);
  2061. if (xfs_inode_clean(ip)) {
  2062. xfs_ifunlock(ip);
  2063. xfs_iunlock(ip, XFS_ILOCK_EXCL);
  2064. } else {
  2065. ip_found[found++] = ip;
  2066. }
  2067. } else {
  2068. xfs_iunlock(ip, XFS_ILOCK_EXCL);
  2069. }
  2070. }
  2071. read_unlock(&pag->pag_ici_lock);
  2072. }
  2073. bp = xfs_trans_get_buf(tp, mp->m_ddev_targp, blkno,
  2074. mp->m_bsize * blks_per_cluster,
  2075. XFS_BUF_LOCK);
  2076. pre_flushed = 0;
  2077. lip = XFS_BUF_FSPRIVATE(bp, xfs_log_item_t *);
  2078. while (lip) {
  2079. if (lip->li_type == XFS_LI_INODE) {
  2080. iip = (xfs_inode_log_item_t *)lip;
  2081. ASSERT(iip->ili_logged == 1);
  2082. lip->li_cb = (void(*)(xfs_buf_t*,xfs_log_item_t*)) xfs_istale_done;
  2083. spin_lock(&mp->m_ail_lock);
  2084. iip->ili_flush_lsn = iip->ili_item.li_lsn;
  2085. spin_unlock(&mp->m_ail_lock);
  2086. xfs_iflags_set(iip->ili_inode, XFS_ISTALE);
  2087. pre_flushed++;
  2088. }
  2089. lip = lip->li_bio_list;
  2090. }
  2091. for (i = 0; i < found; i++) {
  2092. ip = ip_found[i];
  2093. iip = ip->i_itemp;
  2094. if (!iip) {
  2095. ip->i_update_core = 0;
  2096. xfs_ifunlock(ip);
  2097. xfs_iunlock(ip, XFS_ILOCK_EXCL);
  2098. continue;
  2099. }
  2100. iip->ili_last_fields = iip->ili_format.ilf_fields;
  2101. iip->ili_format.ilf_fields = 0;
  2102. iip->ili_logged = 1;
  2103. spin_lock(&mp->m_ail_lock);
  2104. iip->ili_flush_lsn = iip->ili_item.li_lsn;
  2105. spin_unlock(&mp->m_ail_lock);
  2106. xfs_buf_attach_iodone(bp,
  2107. (void(*)(xfs_buf_t*,xfs_log_item_t*))
  2108. xfs_istale_done, (xfs_log_item_t *)iip);
  2109. if (ip != free_ip) {
  2110. xfs_iunlock(ip, XFS_ILOCK_EXCL);
  2111. }
  2112. }
  2113. if (found || pre_flushed)
  2114. xfs_trans_stale_inode_buf(tp, bp);
  2115. xfs_trans_binval(tp, bp);
  2116. }
  2117. kmem_free(ip_found, ninodes * sizeof(xfs_inode_t *));
  2118. xfs_put_perag(mp, pag);
  2119. }
  2120. /*
  2121. * This is called to return an inode to the inode free list.
  2122. * The inode should already be truncated to 0 length and have
  2123. * no pages associated with it. This routine also assumes that
  2124. * the inode is already a part of the transaction.
  2125. *
  2126. * The on-disk copy of the inode will have been added to the list
  2127. * of unlinked inodes in the AGI. We need to remove the inode from
  2128. * that list atomically with respect to freeing it here.
  2129. */
  2130. int
  2131. xfs_ifree(
  2132. xfs_trans_t *tp,
  2133. xfs_inode_t *ip,
  2134. xfs_bmap_free_t *flist)
  2135. {
  2136. int error;
  2137. int delete;
  2138. xfs_ino_t first_ino;
  2139. xfs_dinode_t *dip;
  2140. xfs_buf_t *ibp;
  2141. ASSERT(ismrlocked(&ip->i_lock, MR_UPDATE));
  2142. ASSERT(ip->i_transp == tp);
  2143. ASSERT(ip->i_d.di_nlink == 0);
  2144. ASSERT(ip->i_d.di_nextents == 0);
  2145. ASSERT(ip->i_d.di_anextents == 0);
  2146. ASSERT((ip->i_d.di_size == 0 && ip->i_size == 0) ||
  2147. ((ip->i_d.di_mode & S_IFMT) != S_IFREG));
  2148. ASSERT(ip->i_d.di_nblocks == 0);
  2149. /*
  2150. * Pull the on-disk inode from the AGI unlinked list.
  2151. */
  2152. error = xfs_iunlink_remove(tp, ip);
  2153. if (error != 0) {
  2154. return error;
  2155. }
  2156. error = xfs_difree(tp, ip->i_ino, flist, &delete, &first_ino);
  2157. if (error != 0) {
  2158. return error;
  2159. }
  2160. ip->i_d.di_mode = 0; /* mark incore inode as free */
  2161. ip->i_d.di_flags = 0;
  2162. ip->i_d.di_dmevmask = 0;
  2163. ip->i_d.di_forkoff = 0; /* mark the attr fork not in use */
  2164. ip->i_df.if_ext_max =
  2165. XFS_IFORK_DSIZE(ip) / (uint)sizeof(xfs_bmbt_rec_t);
  2166. ip->i_d.di_format = XFS_DINODE_FMT_EXTENTS;
  2167. ip->i_d.di_aformat = XFS_DINODE_FMT_EXTENTS;
  2168. /*
  2169. * Bump the generation count so no one will be confused
  2170. * by reincarnations of this inode.
  2171. */
  2172. ip->i_d.di_gen++;
  2173. xfs_trans_log_inode(tp, ip, XFS_ILOG_CORE);
  2174. error = xfs_itobp(ip->i_mount, tp, ip, &dip, &ibp, 0, 0, XFS_BUF_LOCK);
  2175. if (error)
  2176. return error;
  2177. /*
  2178. * Clear the on-disk di_mode. This is to prevent xfs_bulkstat
  2179. * from picking up this inode when it is reclaimed (its incore state
  2180. * initialzed but not flushed to disk yet). The in-core di_mode is
  2181. * already cleared and a corresponding transaction logged.
  2182. * The hack here just synchronizes the in-core to on-disk
  2183. * di_mode value in advance before the actual inode sync to disk.
  2184. * This is OK because the inode is already unlinked and would never
  2185. * change its di_mode again for this inode generation.
  2186. * This is a temporary hack that would require a proper fix
  2187. * in the future.
  2188. */
  2189. dip->di_core.di_mode = 0;
  2190. if (delete) {
  2191. xfs_ifree_cluster(ip, tp, first_ino);
  2192. }
  2193. return 0;
  2194. }
  2195. /*
  2196. * Reallocate the space for if_broot based on the number of records
  2197. * being added or deleted as indicated in rec_diff. Move the records
  2198. * and pointers in if_broot to fit the new size. When shrinking this
  2199. * will eliminate holes between the records and pointers created by
  2200. * the caller. When growing this will create holes to be filled in
  2201. * by the caller.
  2202. *
  2203. * The caller must not request to add more records than would fit in
  2204. * the on-disk inode root. If the if_broot is currently NULL, then
  2205. * if we adding records one will be allocated. The caller must also
  2206. * not request that the number of records go below zero, although
  2207. * it can go to zero.
  2208. *
  2209. * ip -- the inode whose if_broot area is changing
  2210. * ext_diff -- the change in the number of records, positive or negative,
  2211. * requested for the if_broot array.
  2212. */
  2213. void
  2214. xfs_iroot_realloc(
  2215. xfs_inode_t *ip,
  2216. int rec_diff,
  2217. int whichfork)
  2218. {
  2219. int cur_max;
  2220. xfs_ifork_t *ifp;
  2221. xfs_bmbt_block_t *new_broot;
  2222. int new_max;
  2223. size_t new_size;
  2224. char *np;
  2225. char *op;
  2226. /*
  2227. * Handle the degenerate case quietly.
  2228. */
  2229. if (rec_diff == 0) {
  2230. return;
  2231. }
  2232. ifp = XFS_IFORK_PTR(ip, whichfork);
  2233. if (rec_diff > 0) {
  2234. /*
  2235. * If there wasn't any memory allocated before, just
  2236. * allocate it now and get out.
  2237. */
  2238. if (ifp->if_broot_bytes == 0) {
  2239. new_size = (size_t)XFS_BMAP_BROOT_SPACE_CALC(rec_diff);
  2240. ifp->if_broot = (xfs_bmbt_block_t*)kmem_alloc(new_size,
  2241. KM_SLEEP);
  2242. ifp->if_broot_bytes = (int)new_size;
  2243. return;
  2244. }
  2245. /*
  2246. * If there is already an existing if_broot, then we need
  2247. * to realloc() it and shift the pointers to their new
  2248. * location. The records don't change location because
  2249. * they are kept butted up against the btree block header.
  2250. */
  2251. cur_max = XFS_BMAP_BROOT_MAXRECS(ifp->if_broot_bytes);
  2252. new_max = cur_max + rec_diff;
  2253. new_size = (size_t)XFS_BMAP_BROOT_SPACE_CALC(new_max);
  2254. ifp->if_broot = (xfs_bmbt_block_t *)
  2255. kmem_realloc(ifp->if_broot,
  2256. new_size,
  2257. (size_t)XFS_BMAP_BROOT_SPACE_CALC(cur_max), /* old size */
  2258. KM_SLEEP);
  2259. op = (char *)XFS_BMAP_BROOT_PTR_ADDR(ifp->if_broot, 1,
  2260. ifp->if_broot_bytes);
  2261. np = (char *)XFS_BMAP_BROOT_PTR_ADDR(ifp->if_broot, 1,
  2262. (int)new_size);
  2263. ifp->if_broot_bytes = (int)new_size;
  2264. ASSERT(ifp->if_broot_bytes <=
  2265. XFS_IFORK_SIZE(ip, whichfork) + XFS_BROOT_SIZE_ADJ);
  2266. memmove(np, op, cur_max * (uint)sizeof(xfs_dfsbno_t));
  2267. return;
  2268. }
  2269. /*
  2270. * rec_diff is less than 0. In this case, we are shrinking the
  2271. * if_broot buffer. It must already exist. If we go to zero
  2272. * records, just get rid of the root and clear the status bit.
  2273. */
  2274. ASSERT((ifp->if_broot != NULL) && (ifp->if_broot_bytes > 0));
  2275. cur_max = XFS_BMAP_BROOT_MAXRECS(ifp->if_broot_bytes);
  2276. new_max = cur_max + rec_diff;
  2277. ASSERT(new_max >= 0);
  2278. if (new_max > 0)
  2279. new_size = (size_t)XFS_BMAP_BROOT_SPACE_CALC(new_max);
  2280. else
  2281. new_size = 0;
  2282. if (new_size > 0) {
  2283. new_broot = (xfs_bmbt_block_t *)kmem_alloc(new_size, KM_SLEEP);
  2284. /*
  2285. * First copy over the btree block header.
  2286. */
  2287. memcpy(new_broot, ifp->if_broot, sizeof(xfs_bmbt_block_t));
  2288. } else {
  2289. new_broot = NULL;
  2290. ifp->if_flags &= ~XFS_IFBROOT;
  2291. }
  2292. /*
  2293. * Only copy the records and pointers if there are any.
  2294. */
  2295. if (new_max > 0) {
  2296. /*
  2297. * First copy the records.
  2298. */
  2299. op = (char *)XFS_BMAP_BROOT_REC_ADDR(ifp->if_broot, 1,
  2300. ifp->if_broot_bytes);
  2301. np = (char *)XFS_BMAP_BROOT_REC_ADDR(new_broot, 1,
  2302. (int)new_size);
  2303. memcpy(np, op, new_max * (uint)sizeof(xfs_bmbt_rec_t));
  2304. /*
  2305. * Then copy the pointers.
  2306. */
  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(new_broot, 1,
  2310. (int)new_size);
  2311. memcpy(np, op, new_max * (uint)sizeof(xfs_dfsbno_t));
  2312. }
  2313. kmem_free(ifp->if_broot, ifp->if_broot_bytes);
  2314. ifp->if_broot = new_broot;
  2315. ifp->if_broot_bytes = (int)new_size;
  2316. ASSERT(ifp->if_broot_bytes <=
  2317. XFS_IFORK_SIZE(ip, whichfork) + XFS_BROOT_SIZE_ADJ);
  2318. return;
  2319. }
  2320. /*
  2321. * This is called when the amount of space needed for if_data
  2322. * is increased or decreased. The change in size is indicated by
  2323. * the number of bytes that need to be added or deleted in the
  2324. * byte_diff parameter.
  2325. *
  2326. * If the amount of space needed has decreased below the size of the
  2327. * inline buffer, then switch to using the inline buffer. Otherwise,
  2328. * use kmem_realloc() or kmem_alloc() to adjust the size of the buffer
  2329. * to what is needed.
  2330. *
  2331. * ip -- the inode whose if_data area is changing
  2332. * byte_diff -- the change in the number of bytes, positive or negative,
  2333. * requested for the if_data array.
  2334. */
  2335. void
  2336. xfs_idata_realloc(
  2337. xfs_inode_t *ip,
  2338. int byte_diff,
  2339. int whichfork)
  2340. {
  2341. xfs_ifork_t *ifp;
  2342. int new_size;
  2343. int real_size;
  2344. if (byte_diff == 0) {
  2345. return;
  2346. }
  2347. ifp = XFS_IFORK_PTR(ip, whichfork);
  2348. new_size = (int)ifp->if_bytes + byte_diff;
  2349. ASSERT(new_size >= 0);
  2350. if (new_size == 0) {
  2351. if (ifp->if_u1.if_data != ifp->if_u2.if_inline_data) {
  2352. kmem_free(ifp->if_u1.if_data, ifp->if_real_bytes);
  2353. }
  2354. ifp->if_u1.if_data = NULL;
  2355. real_size = 0;
  2356. } else if (new_size <= sizeof(ifp->if_u2.if_inline_data)) {
  2357. /*
  2358. * If the valid extents/data can fit in if_inline_ext/data,
  2359. * copy them from the malloc'd vector and free it.
  2360. */
  2361. if (ifp->if_u1.if_data == NULL) {
  2362. ifp->if_u1.if_data = ifp->if_u2.if_inline_data;
  2363. } else if (ifp->if_u1.if_data != ifp->if_u2.if_inline_data) {
  2364. ASSERT(ifp->if_real_bytes != 0);
  2365. memcpy(ifp->if_u2.if_inline_data, ifp->if_u1.if_data,
  2366. new_size);
  2367. kmem_free(ifp->if_u1.if_data, ifp->if_real_bytes);
  2368. ifp->if_u1.if_data = ifp->if_u2.if_inline_data;
  2369. }
  2370. real_size = 0;
  2371. } else {
  2372. /*
  2373. * Stuck with malloc/realloc.
  2374. * For inline data, the underlying buffer must be
  2375. * a multiple of 4 bytes in size so that it can be
  2376. * logged and stay on word boundaries. We enforce
  2377. * that here.
  2378. */
  2379. real_size = roundup(new_size, 4);
  2380. if (ifp->if_u1.if_data == NULL) {
  2381. ASSERT(ifp->if_real_bytes == 0);
  2382. ifp->if_u1.if_data = kmem_alloc(real_size, KM_SLEEP);
  2383. } else if (ifp->if_u1.if_data != ifp->if_u2.if_inline_data) {
  2384. /*
  2385. * Only do the realloc if the underlying size
  2386. * is really changing.
  2387. */
  2388. if (ifp->if_real_bytes != real_size) {
  2389. ifp->if_u1.if_data =
  2390. kmem_realloc(ifp->if_u1.if_data,
  2391. real_size,
  2392. ifp->if_real_bytes,
  2393. KM_SLEEP);
  2394. }
  2395. } else {
  2396. ASSERT(ifp->if_real_bytes == 0);
  2397. ifp->if_u1.if_data = kmem_alloc(real_size, KM_SLEEP);
  2398. memcpy(ifp->if_u1.if_data, ifp->if_u2.if_inline_data,
  2399. ifp->if_bytes);
  2400. }
  2401. }
  2402. ifp->if_real_bytes = real_size;
  2403. ifp->if_bytes = new_size;
  2404. ASSERT(ifp->if_bytes <= XFS_IFORK_SIZE(ip, whichfork));
  2405. }
  2406. /*
  2407. * Map inode to disk block and offset.
  2408. *
  2409. * mp -- the mount point structure for the current file system
  2410. * tp -- the current transaction
  2411. * ino -- the inode number of the inode to be located
  2412. * imap -- this structure is filled in with the information necessary
  2413. * to retrieve the given inode from disk
  2414. * flags -- flags to pass to xfs_dilocate indicating whether or not
  2415. * lookups in the inode btree were OK or not
  2416. */
  2417. int
  2418. xfs_imap(
  2419. xfs_mount_t *mp,
  2420. xfs_trans_t *tp,
  2421. xfs_ino_t ino,
  2422. xfs_imap_t *imap,
  2423. uint flags)
  2424. {
  2425. xfs_fsblock_t fsbno;
  2426. int len;
  2427. int off;
  2428. int error;
  2429. fsbno = imap->im_blkno ?
  2430. XFS_DADDR_TO_FSB(mp, imap->im_blkno) : NULLFSBLOCK;
  2431. error = xfs_dilocate(mp, tp, ino, &fsbno, &len, &off, flags);
  2432. if (error)
  2433. return error;
  2434. imap->im_blkno = XFS_FSB_TO_DADDR(mp, fsbno);
  2435. imap->im_len = XFS_FSB_TO_BB(mp, len);
  2436. imap->im_agblkno = XFS_FSB_TO_AGBNO(mp, fsbno);
  2437. imap->im_ioffset = (ushort)off;
  2438. imap->im_boffset = (ushort)(off << mp->m_sb.sb_inodelog);
  2439. /*
  2440. * If the inode number maps to a block outside the bounds
  2441. * of the file system then return NULL rather than calling
  2442. * read_buf and panicing when we get an error from the
  2443. * driver.
  2444. */
  2445. if ((imap->im_blkno + imap->im_len) >
  2446. XFS_FSB_TO_BB(mp, mp->m_sb.sb_dblocks)) {
  2447. xfs_fs_cmn_err(CE_ALERT, mp, "xfs_imap: "
  2448. "(imap->im_blkno (0x%llx) + imap->im_len (0x%llx)) > "
  2449. " XFS_FSB_TO_BB(mp, mp->m_sb.sb_dblocks) (0x%llx)",
  2450. (unsigned long long) imap->im_blkno,
  2451. (unsigned long long) imap->im_len,
  2452. XFS_FSB_TO_BB(mp, mp->m_sb.sb_dblocks));
  2453. return EINVAL;
  2454. }
  2455. return 0;
  2456. }
  2457. void
  2458. xfs_idestroy_fork(
  2459. xfs_inode_t *ip,
  2460. int whichfork)
  2461. {
  2462. xfs_ifork_t *ifp;
  2463. ifp = XFS_IFORK_PTR(ip, whichfork);
  2464. if (ifp->if_broot != NULL) {
  2465. kmem_free(ifp->if_broot, ifp->if_broot_bytes);
  2466. ifp->if_broot = NULL;
  2467. }
  2468. /*
  2469. * If the format is local, then we can't have an extents
  2470. * array so just look for an inline data array. If we're
  2471. * not local then we may or may not have an extents list,
  2472. * so check and free it up if we do.
  2473. */
  2474. if (XFS_IFORK_FORMAT(ip, whichfork) == XFS_DINODE_FMT_LOCAL) {
  2475. if ((ifp->if_u1.if_data != ifp->if_u2.if_inline_data) &&
  2476. (ifp->if_u1.if_data != NULL)) {
  2477. ASSERT(ifp->if_real_bytes != 0);
  2478. kmem_free(ifp->if_u1.if_data, ifp->if_real_bytes);
  2479. ifp->if_u1.if_data = NULL;
  2480. ifp->if_real_bytes = 0;
  2481. }
  2482. } else if ((ifp->if_flags & XFS_IFEXTENTS) &&
  2483. ((ifp->if_flags & XFS_IFEXTIREC) ||
  2484. ((ifp->if_u1.if_extents != NULL) &&
  2485. (ifp->if_u1.if_extents != ifp->if_u2.if_inline_ext)))) {
  2486. ASSERT(ifp->if_real_bytes != 0);
  2487. xfs_iext_destroy(ifp);
  2488. }
  2489. ASSERT(ifp->if_u1.if_extents == NULL ||
  2490. ifp->if_u1.if_extents == ifp->if_u2.if_inline_ext);
  2491. ASSERT(ifp->if_real_bytes == 0);
  2492. if (whichfork == XFS_ATTR_FORK) {
  2493. kmem_zone_free(xfs_ifork_zone, ip->i_afp);
  2494. ip->i_afp = NULL;
  2495. }
  2496. }
  2497. /*
  2498. * This is called free all the memory associated with an inode.
  2499. * It must free the inode itself and any buffers allocated for
  2500. * if_extents/if_data and if_broot. It must also free the lock
  2501. * associated with the inode.
  2502. */
  2503. void
  2504. xfs_idestroy(
  2505. xfs_inode_t *ip)
  2506. {
  2507. switch (ip->i_d.di_mode & S_IFMT) {
  2508. case S_IFREG:
  2509. case S_IFDIR:
  2510. case S_IFLNK:
  2511. xfs_idestroy_fork(ip, XFS_DATA_FORK);
  2512. break;
  2513. }
  2514. if (ip->i_afp)
  2515. xfs_idestroy_fork(ip, XFS_ATTR_FORK);
  2516. mrfree(&ip->i_lock);
  2517. mrfree(&ip->i_iolock);
  2518. freesema(&ip->i_flock);
  2519. #ifdef XFS_INODE_TRACE
  2520. ktrace_free(ip->i_trace);
  2521. #endif
  2522. #ifdef XFS_BMAP_TRACE
  2523. ktrace_free(ip->i_xtrace);
  2524. #endif
  2525. #ifdef XFS_BMBT_TRACE
  2526. ktrace_free(ip->i_btrace);
  2527. #endif
  2528. #ifdef XFS_RW_TRACE
  2529. ktrace_free(ip->i_rwtrace);
  2530. #endif
  2531. #ifdef XFS_ILOCK_TRACE
  2532. ktrace_free(ip->i_lock_trace);
  2533. #endif
  2534. #ifdef XFS_DIR2_TRACE
  2535. ktrace_free(ip->i_dir_trace);
  2536. #endif
  2537. if (ip->i_itemp) {
  2538. /*
  2539. * Only if we are shutting down the fs will we see an
  2540. * inode still in the AIL. If it is there, we should remove
  2541. * it to prevent a use-after-free from occurring.
  2542. */
  2543. xfs_mount_t *mp = ip->i_mount;
  2544. xfs_log_item_t *lip = &ip->i_itemp->ili_item;
  2545. ASSERT(((lip->li_flags & XFS_LI_IN_AIL) == 0) ||
  2546. XFS_FORCED_SHUTDOWN(ip->i_mount));
  2547. if (lip->li_flags & XFS_LI_IN_AIL) {
  2548. spin_lock(&mp->m_ail_lock);
  2549. if (lip->li_flags & XFS_LI_IN_AIL)
  2550. xfs_trans_delete_ail(mp, lip);
  2551. else
  2552. spin_unlock(&mp->m_ail_lock);
  2553. }
  2554. xfs_inode_item_destroy(ip);
  2555. }
  2556. kmem_zone_free(xfs_inode_zone, ip);
  2557. }
  2558. /*
  2559. * Increment the pin count of the given buffer.
  2560. * This value is protected by ipinlock spinlock in the mount structure.
  2561. */
  2562. void
  2563. xfs_ipin(
  2564. xfs_inode_t *ip)
  2565. {
  2566. ASSERT(ismrlocked(&ip->i_lock, MR_UPDATE));
  2567. atomic_inc(&ip->i_pincount);
  2568. }
  2569. /*
  2570. * Decrement the pin count of the given inode, and wake up
  2571. * anyone in xfs_iwait_unpin() if the count goes to 0. The
  2572. * inode must have been previously pinned with a call to xfs_ipin().
  2573. */
  2574. void
  2575. xfs_iunpin(
  2576. xfs_inode_t *ip)
  2577. {
  2578. ASSERT(atomic_read(&ip->i_pincount) > 0);
  2579. if (atomic_dec_and_test(&ip->i_pincount))
  2580. wake_up(&ip->i_ipin_wait);
  2581. }
  2582. /*
  2583. * This is called to unpin an inode. It can be directed to wait or to return
  2584. * immediately without waiting for the inode to be unpinned. The caller must
  2585. * have the inode locked in at least shared mode so that the buffer cannot be
  2586. * subsequently pinned once someone is waiting for it to be unpinned.
  2587. */
  2588. STATIC void
  2589. __xfs_iunpin_wait(
  2590. xfs_inode_t *ip,
  2591. int wait)
  2592. {
  2593. xfs_inode_log_item_t *iip = ip->i_itemp;
  2594. ASSERT(ismrlocked(&ip->i_lock, MR_UPDATE | MR_ACCESS));
  2595. if (atomic_read(&ip->i_pincount) == 0)
  2596. return;
  2597. /* Give the log a push to start the unpinning I/O */
  2598. xfs_log_force(ip->i_mount, (iip && iip->ili_last_lsn) ?
  2599. iip->ili_last_lsn : 0, XFS_LOG_FORCE);
  2600. if (wait)
  2601. wait_event(ip->i_ipin_wait, (atomic_read(&ip->i_pincount) == 0));
  2602. }
  2603. static inline void
  2604. xfs_iunpin_wait(
  2605. xfs_inode_t *ip)
  2606. {
  2607. __xfs_iunpin_wait(ip, 1);
  2608. }
  2609. static inline void
  2610. xfs_iunpin_nowait(
  2611. xfs_inode_t *ip)
  2612. {
  2613. __xfs_iunpin_wait(ip, 0);
  2614. }
  2615. /*
  2616. * xfs_iextents_copy()
  2617. *
  2618. * This is called to copy the REAL extents (as opposed to the delayed
  2619. * allocation extents) from the inode into the given buffer. It
  2620. * returns the number of bytes copied into the buffer.
  2621. *
  2622. * If there are no delayed allocation extents, then we can just
  2623. * memcpy() the extents into the buffer. Otherwise, we need to
  2624. * examine each extent in turn and skip those which are delayed.
  2625. */
  2626. int
  2627. xfs_iextents_copy(
  2628. xfs_inode_t *ip,
  2629. xfs_bmbt_rec_t *dp,
  2630. int whichfork)
  2631. {
  2632. int copied;
  2633. int i;
  2634. xfs_ifork_t *ifp;
  2635. int nrecs;
  2636. xfs_fsblock_t start_block;
  2637. ifp = XFS_IFORK_PTR(ip, whichfork);
  2638. ASSERT(ismrlocked(&ip->i_lock, MR_UPDATE|MR_ACCESS));
  2639. ASSERT(ifp->if_bytes > 0);
  2640. nrecs = ifp->if_bytes / (uint)sizeof(xfs_bmbt_rec_t);
  2641. XFS_BMAP_TRACE_EXLIST(ip, nrecs, whichfork);
  2642. ASSERT(nrecs > 0);
  2643. /*
  2644. * There are some delayed allocation extents in the
  2645. * inode, so copy the extents one at a time and skip
  2646. * the delayed ones. There must be at least one
  2647. * non-delayed extent.
  2648. */
  2649. copied = 0;
  2650. for (i = 0; i < nrecs; i++) {
  2651. xfs_bmbt_rec_host_t *ep = xfs_iext_get_ext(ifp, i);
  2652. start_block = xfs_bmbt_get_startblock(ep);
  2653. if (ISNULLSTARTBLOCK(start_block)) {
  2654. /*
  2655. * It's a delayed allocation extent, so skip it.
  2656. */
  2657. continue;
  2658. }
  2659. /* Translate to on disk format */
  2660. put_unaligned(cpu_to_be64(ep->l0), &dp->l0);
  2661. put_unaligned(cpu_to_be64(ep->l1), &dp->l1);
  2662. dp++;
  2663. copied++;
  2664. }
  2665. ASSERT(copied != 0);
  2666. xfs_validate_extents(ifp, copied, XFS_EXTFMT_INODE(ip));
  2667. return (copied * (uint)sizeof(xfs_bmbt_rec_t));
  2668. }
  2669. /*
  2670. * Each of the following cases stores data into the same region
  2671. * of the on-disk inode, so only one of them can be valid at
  2672. * any given time. While it is possible to have conflicting formats
  2673. * and log flags, e.g. having XFS_ILOG_?DATA set when the fork is
  2674. * in EXTENTS format, this can only happen when the fork has
  2675. * changed formats after being modified but before being flushed.
  2676. * In these cases, the format always takes precedence, because the
  2677. * format indicates the current state of the fork.
  2678. */
  2679. /*ARGSUSED*/
  2680. STATIC int
  2681. xfs_iflush_fork(
  2682. xfs_inode_t *ip,
  2683. xfs_dinode_t *dip,
  2684. xfs_inode_log_item_t *iip,
  2685. int whichfork,
  2686. xfs_buf_t *bp)
  2687. {
  2688. char *cp;
  2689. xfs_ifork_t *ifp;
  2690. xfs_mount_t *mp;
  2691. #ifdef XFS_TRANS_DEBUG
  2692. int first;
  2693. #endif
  2694. static const short brootflag[2] =
  2695. { XFS_ILOG_DBROOT, XFS_ILOG_ABROOT };
  2696. static const short dataflag[2] =
  2697. { XFS_ILOG_DDATA, XFS_ILOG_ADATA };
  2698. static const short extflag[2] =
  2699. { XFS_ILOG_DEXT, XFS_ILOG_AEXT };
  2700. if (iip == NULL)
  2701. return 0;
  2702. ifp = XFS_IFORK_PTR(ip, whichfork);
  2703. /*
  2704. * This can happen if we gave up in iformat in an error path,
  2705. * for the attribute fork.
  2706. */
  2707. if (ifp == NULL) {
  2708. ASSERT(whichfork == XFS_ATTR_FORK);
  2709. return 0;
  2710. }
  2711. cp = XFS_DFORK_PTR(dip, whichfork);
  2712. mp = ip->i_mount;
  2713. switch (XFS_IFORK_FORMAT(ip, whichfork)) {
  2714. case XFS_DINODE_FMT_LOCAL:
  2715. if ((iip->ili_format.ilf_fields & dataflag[whichfork]) &&
  2716. (ifp->if_bytes > 0)) {
  2717. ASSERT(ifp->if_u1.if_data != NULL);
  2718. ASSERT(ifp->if_bytes <= XFS_IFORK_SIZE(ip, whichfork));
  2719. memcpy(cp, ifp->if_u1.if_data, ifp->if_bytes);
  2720. }
  2721. break;
  2722. case XFS_DINODE_FMT_EXTENTS:
  2723. ASSERT((ifp->if_flags & XFS_IFEXTENTS) ||
  2724. !(iip->ili_format.ilf_fields & extflag[whichfork]));
  2725. ASSERT((xfs_iext_get_ext(ifp, 0) != NULL) ||
  2726. (ifp->if_bytes == 0));
  2727. ASSERT((xfs_iext_get_ext(ifp, 0) == NULL) ||
  2728. (ifp->if_bytes > 0));
  2729. if ((iip->ili_format.ilf_fields & extflag[whichfork]) &&
  2730. (ifp->if_bytes > 0)) {
  2731. ASSERT(XFS_IFORK_NEXTENTS(ip, whichfork) > 0);
  2732. (void)xfs_iextents_copy(ip, (xfs_bmbt_rec_t *)cp,
  2733. whichfork);
  2734. }
  2735. break;
  2736. case XFS_DINODE_FMT_BTREE:
  2737. if ((iip->ili_format.ilf_fields & brootflag[whichfork]) &&
  2738. (ifp->if_broot_bytes > 0)) {
  2739. ASSERT(ifp->if_broot != NULL);
  2740. ASSERT(ifp->if_broot_bytes <=
  2741. (XFS_IFORK_SIZE(ip, whichfork) +
  2742. XFS_BROOT_SIZE_ADJ));
  2743. xfs_bmbt_to_bmdr(ifp->if_broot, ifp->if_broot_bytes,
  2744. (xfs_bmdr_block_t *)cp,
  2745. XFS_DFORK_SIZE(dip, mp, whichfork));
  2746. }
  2747. break;
  2748. case XFS_DINODE_FMT_DEV:
  2749. if (iip->ili_format.ilf_fields & XFS_ILOG_DEV) {
  2750. ASSERT(whichfork == XFS_DATA_FORK);
  2751. dip->di_u.di_dev = cpu_to_be32(ip->i_df.if_u2.if_rdev);
  2752. }
  2753. break;
  2754. case XFS_DINODE_FMT_UUID:
  2755. if (iip->ili_format.ilf_fields & XFS_ILOG_UUID) {
  2756. ASSERT(whichfork == XFS_DATA_FORK);
  2757. memcpy(&dip->di_u.di_muuid, &ip->i_df.if_u2.if_uuid,
  2758. sizeof(uuid_t));
  2759. }
  2760. break;
  2761. default:
  2762. ASSERT(0);
  2763. break;
  2764. }
  2765. return 0;
  2766. }
  2767. STATIC int
  2768. xfs_iflush_cluster(
  2769. xfs_inode_t *ip,
  2770. xfs_buf_t *bp)
  2771. {
  2772. xfs_mount_t *mp = ip->i_mount;
  2773. xfs_perag_t *pag = xfs_get_perag(mp, ip->i_ino);
  2774. unsigned long first_index, mask;
  2775. int ilist_size;
  2776. xfs_inode_t **ilist;
  2777. xfs_inode_t *iq;
  2778. int nr_found;
  2779. int clcount = 0;
  2780. int bufwasdelwri;
  2781. int i;
  2782. ASSERT(pag->pagi_inodeok);
  2783. ASSERT(pag->pag_ici_init);
  2784. ilist_size = XFS_INODE_CLUSTER_SIZE(mp) * sizeof(xfs_inode_t *);
  2785. ilist = kmem_alloc(ilist_size, KM_MAYFAIL);
  2786. if (!ilist)
  2787. return 0;
  2788. mask = ~(((XFS_INODE_CLUSTER_SIZE(mp) >> mp->m_sb.sb_inodelog)) - 1);
  2789. first_index = XFS_INO_TO_AGINO(mp, ip->i_ino) & mask;
  2790. read_lock(&pag->pag_ici_lock);
  2791. /* really need a gang lookup range call here */
  2792. nr_found = radix_tree_gang_lookup(&pag->pag_ici_root, (void**)ilist,
  2793. first_index,
  2794. XFS_INODE_CLUSTER_SIZE(mp));
  2795. if (nr_found == 0)
  2796. goto out_free;
  2797. for (i = 0; i < nr_found; i++) {
  2798. iq = ilist[i];
  2799. if (iq == ip)
  2800. continue;
  2801. /* if the inode lies outside this cluster, we're done. */
  2802. if ((XFS_INO_TO_AGINO(mp, iq->i_ino) & mask) != first_index)
  2803. break;
  2804. /*
  2805. * Do an un-protected check to see if the inode is dirty and
  2806. * is a candidate for flushing. These checks will be repeated
  2807. * later after the appropriate locks are acquired.
  2808. */
  2809. if (xfs_inode_clean(iq) && xfs_ipincount(iq) == 0)
  2810. continue;
  2811. /*
  2812. * Try to get locks. If any are unavailable or it is pinned,
  2813. * then this inode cannot be flushed and is skipped.
  2814. */
  2815. if (!xfs_ilock_nowait(iq, XFS_ILOCK_SHARED))
  2816. continue;
  2817. if (!xfs_iflock_nowait(iq)) {
  2818. xfs_iunlock(iq, XFS_ILOCK_SHARED);
  2819. continue;
  2820. }
  2821. if (xfs_ipincount(iq)) {
  2822. xfs_ifunlock(iq);
  2823. xfs_iunlock(iq, XFS_ILOCK_SHARED);
  2824. continue;
  2825. }
  2826. /*
  2827. * arriving here means that this inode can be flushed. First
  2828. * re-check that it's dirty before flushing.
  2829. */
  2830. if (!xfs_inode_clean(iq)) {
  2831. int error;
  2832. error = xfs_iflush_int(iq, bp);
  2833. if (error) {
  2834. xfs_iunlock(iq, XFS_ILOCK_SHARED);
  2835. goto cluster_corrupt_out;
  2836. }
  2837. clcount++;
  2838. } else {
  2839. xfs_ifunlock(iq);
  2840. }
  2841. xfs_iunlock(iq, XFS_ILOCK_SHARED);
  2842. }
  2843. if (clcount) {
  2844. XFS_STATS_INC(xs_icluster_flushcnt);
  2845. XFS_STATS_ADD(xs_icluster_flushinode, clcount);
  2846. }
  2847. out_free:
  2848. read_unlock(&pag->pag_ici_lock);
  2849. kmem_free(ilist, ilist_size);
  2850. return 0;
  2851. cluster_corrupt_out:
  2852. /*
  2853. * Corruption detected in the clustering loop. Invalidate the
  2854. * inode buffer and shut down the filesystem.
  2855. */
  2856. read_unlock(&pag->pag_ici_lock);
  2857. /*
  2858. * Clean up the buffer. If it was B_DELWRI, just release it --
  2859. * brelse can handle it with no problems. If not, shut down the
  2860. * filesystem before releasing the buffer.
  2861. */
  2862. bufwasdelwri = XFS_BUF_ISDELAYWRITE(bp);
  2863. if (bufwasdelwri)
  2864. xfs_buf_relse(bp);
  2865. xfs_force_shutdown(mp, SHUTDOWN_CORRUPT_INCORE);
  2866. if (!bufwasdelwri) {
  2867. /*
  2868. * Just like incore_relse: if we have b_iodone functions,
  2869. * mark the buffer as an error and call them. Otherwise
  2870. * mark it as stale and brelse.
  2871. */
  2872. if (XFS_BUF_IODONE_FUNC(bp)) {
  2873. XFS_BUF_CLR_BDSTRAT_FUNC(bp);
  2874. XFS_BUF_UNDONE(bp);
  2875. XFS_BUF_STALE(bp);
  2876. XFS_BUF_SHUT(bp);
  2877. XFS_BUF_ERROR(bp,EIO);
  2878. xfs_biodone(bp);
  2879. } else {
  2880. XFS_BUF_STALE(bp);
  2881. xfs_buf_relse(bp);
  2882. }
  2883. }
  2884. /*
  2885. * Unlocks the flush lock
  2886. */
  2887. xfs_iflush_abort(iq);
  2888. kmem_free(ilist, ilist_size);
  2889. return XFS_ERROR(EFSCORRUPTED);
  2890. }
  2891. /*
  2892. * xfs_iflush() will write a modified inode's changes out to the
  2893. * inode's on disk home. The caller must have the inode lock held
  2894. * in at least shared mode and the inode flush semaphore must be
  2895. * held as well. The inode lock will still be held upon return from
  2896. * the call and the caller is free to unlock it.
  2897. * The inode flush lock will be unlocked when the inode reaches the disk.
  2898. * The flags indicate how the inode's buffer should be written out.
  2899. */
  2900. int
  2901. xfs_iflush(
  2902. xfs_inode_t *ip,
  2903. uint flags)
  2904. {
  2905. xfs_inode_log_item_t *iip;
  2906. xfs_buf_t *bp;
  2907. xfs_dinode_t *dip;
  2908. xfs_mount_t *mp;
  2909. int error;
  2910. int noblock = (flags == XFS_IFLUSH_ASYNC_NOBLOCK);
  2911. enum { INT_DELWRI = (1 << 0), INT_ASYNC = (1 << 1) };
  2912. XFS_STATS_INC(xs_iflush_count);
  2913. ASSERT(ismrlocked(&ip->i_lock, MR_UPDATE|MR_ACCESS));
  2914. ASSERT(issemalocked(&(ip->i_flock)));
  2915. ASSERT(ip->i_d.di_format != XFS_DINODE_FMT_BTREE ||
  2916. ip->i_d.di_nextents > ip->i_df.if_ext_max);
  2917. iip = ip->i_itemp;
  2918. mp = ip->i_mount;
  2919. /*
  2920. * If the inode isn't dirty, then just release the inode
  2921. * flush lock and do nothing.
  2922. */
  2923. if (xfs_inode_clean(ip)) {
  2924. ASSERT((iip != NULL) ?
  2925. !(iip->ili_item.li_flags & XFS_LI_IN_AIL) : 1);
  2926. xfs_ifunlock(ip);
  2927. return 0;
  2928. }
  2929. /*
  2930. * We can't flush the inode until it is unpinned, so wait for it if we
  2931. * are allowed to block. We know noone new can pin it, because we are
  2932. * holding the inode lock shared and you need to hold it exclusively to
  2933. * pin the inode.
  2934. *
  2935. * If we are not allowed to block, force the log out asynchronously so
  2936. * that when we come back the inode will be unpinned. If other inodes
  2937. * in the same cluster are dirty, they will probably write the inode
  2938. * out for us if they occur after the log force completes.
  2939. */
  2940. if (noblock && xfs_ipincount(ip)) {
  2941. xfs_iunpin_nowait(ip);
  2942. xfs_ifunlock(ip);
  2943. return EAGAIN;
  2944. }
  2945. xfs_iunpin_wait(ip);
  2946. /*
  2947. * This may have been unpinned because the filesystem is shutting
  2948. * down forcibly. If that's the case we must not write this inode
  2949. * to disk, because the log record didn't make it to disk!
  2950. */
  2951. if (XFS_FORCED_SHUTDOWN(mp)) {
  2952. ip->i_update_core = 0;
  2953. if (iip)
  2954. iip->ili_format.ilf_fields = 0;
  2955. xfs_ifunlock(ip);
  2956. return XFS_ERROR(EIO);
  2957. }
  2958. /*
  2959. * Decide how buffer will be flushed out. This is done before
  2960. * the call to xfs_iflush_int because this field is zeroed by it.
  2961. */
  2962. if (iip != NULL && iip->ili_format.ilf_fields != 0) {
  2963. /*
  2964. * Flush out the inode buffer according to the directions
  2965. * of the caller. In the cases where the caller has given
  2966. * us a choice choose the non-delwri case. This is because
  2967. * the inode is in the AIL and we need to get it out soon.
  2968. */
  2969. switch (flags) {
  2970. case XFS_IFLUSH_SYNC:
  2971. case XFS_IFLUSH_DELWRI_ELSE_SYNC:
  2972. flags = 0;
  2973. break;
  2974. case XFS_IFLUSH_ASYNC_NOBLOCK:
  2975. case XFS_IFLUSH_ASYNC:
  2976. case XFS_IFLUSH_DELWRI_ELSE_ASYNC:
  2977. flags = INT_ASYNC;
  2978. break;
  2979. case XFS_IFLUSH_DELWRI:
  2980. flags = INT_DELWRI;
  2981. break;
  2982. default:
  2983. ASSERT(0);
  2984. flags = 0;
  2985. break;
  2986. }
  2987. } else {
  2988. switch (flags) {
  2989. case XFS_IFLUSH_DELWRI_ELSE_SYNC:
  2990. case XFS_IFLUSH_DELWRI_ELSE_ASYNC:
  2991. case XFS_IFLUSH_DELWRI:
  2992. flags = INT_DELWRI;
  2993. break;
  2994. case XFS_IFLUSH_ASYNC_NOBLOCK:
  2995. case XFS_IFLUSH_ASYNC:
  2996. flags = INT_ASYNC;
  2997. break;
  2998. case XFS_IFLUSH_SYNC:
  2999. flags = 0;
  3000. break;
  3001. default:
  3002. ASSERT(0);
  3003. flags = 0;
  3004. break;
  3005. }
  3006. }
  3007. /*
  3008. * Get the buffer containing the on-disk inode.
  3009. */
  3010. error = xfs_itobp(mp, NULL, ip, &dip, &bp, 0, 0,
  3011. noblock ? XFS_BUF_TRYLOCK : XFS_BUF_LOCK);
  3012. if (error || !bp) {
  3013. xfs_ifunlock(ip);
  3014. return error;
  3015. }
  3016. /*
  3017. * First flush out the inode that xfs_iflush was called with.
  3018. */
  3019. error = xfs_iflush_int(ip, bp);
  3020. if (error)
  3021. goto corrupt_out;
  3022. /*
  3023. * If the buffer is pinned then push on the log now so we won't
  3024. * get stuck waiting in the write for too long.
  3025. */
  3026. if (XFS_BUF_ISPINNED(bp))
  3027. xfs_log_force(mp, (xfs_lsn_t)0, XFS_LOG_FORCE);
  3028. /*
  3029. * inode clustering:
  3030. * see if other inodes can be gathered into this write
  3031. */
  3032. error = xfs_iflush_cluster(ip, bp);
  3033. if (error)
  3034. goto cluster_corrupt_out;
  3035. if (flags & INT_DELWRI) {
  3036. xfs_bdwrite(mp, bp);
  3037. } else if (flags & INT_ASYNC) {
  3038. error = xfs_bawrite(mp, bp);
  3039. } else {
  3040. error = xfs_bwrite(mp, bp);
  3041. }
  3042. return error;
  3043. corrupt_out:
  3044. xfs_buf_relse(bp);
  3045. xfs_force_shutdown(mp, SHUTDOWN_CORRUPT_INCORE);
  3046. cluster_corrupt_out:
  3047. /*
  3048. * Unlocks the flush lock
  3049. */
  3050. xfs_iflush_abort(ip);
  3051. return XFS_ERROR(EFSCORRUPTED);
  3052. }
  3053. STATIC int
  3054. xfs_iflush_int(
  3055. xfs_inode_t *ip,
  3056. xfs_buf_t *bp)
  3057. {
  3058. xfs_inode_log_item_t *iip;
  3059. xfs_dinode_t *dip;
  3060. xfs_mount_t *mp;
  3061. #ifdef XFS_TRANS_DEBUG
  3062. int first;
  3063. #endif
  3064. ASSERT(ismrlocked(&ip->i_lock, MR_UPDATE|MR_ACCESS));
  3065. ASSERT(issemalocked(&(ip->i_flock)));
  3066. ASSERT(ip->i_d.di_format != XFS_DINODE_FMT_BTREE ||
  3067. ip->i_d.di_nextents > ip->i_df.if_ext_max);
  3068. iip = ip->i_itemp;
  3069. mp = ip->i_mount;
  3070. /*
  3071. * If the inode isn't dirty, then just release the inode
  3072. * flush lock and do nothing.
  3073. */
  3074. if (xfs_inode_clean(ip)) {
  3075. xfs_ifunlock(ip);
  3076. return 0;
  3077. }
  3078. /* set *dip = inode's place in the buffer */
  3079. dip = (xfs_dinode_t *)xfs_buf_offset(bp, ip->i_boffset);
  3080. /*
  3081. * Clear i_update_core before copying out the data.
  3082. * This is for coordination with our timestamp updates
  3083. * that don't hold the inode lock. They will always
  3084. * update the timestamps BEFORE setting i_update_core,
  3085. * so if we clear i_update_core after they set it we
  3086. * are guaranteed to see their updates to the timestamps.
  3087. * I believe that this depends on strongly ordered memory
  3088. * semantics, but we have that. We use the SYNCHRONIZE
  3089. * macro to make sure that the compiler does not reorder
  3090. * the i_update_core access below the data copy below.
  3091. */
  3092. ip->i_update_core = 0;
  3093. SYNCHRONIZE();
  3094. /*
  3095. * Make sure to get the latest atime from the Linux inode.
  3096. */
  3097. xfs_synchronize_atime(ip);
  3098. if (XFS_TEST_ERROR(be16_to_cpu(dip->di_core.di_magic) != XFS_DINODE_MAGIC,
  3099. mp, XFS_ERRTAG_IFLUSH_1, XFS_RANDOM_IFLUSH_1)) {
  3100. xfs_cmn_err(XFS_PTAG_IFLUSH, CE_ALERT, mp,
  3101. "xfs_iflush: Bad inode %Lu magic number 0x%x, ptr 0x%p",
  3102. ip->i_ino, be16_to_cpu(dip->di_core.di_magic), dip);
  3103. goto corrupt_out;
  3104. }
  3105. if (XFS_TEST_ERROR(ip->i_d.di_magic != XFS_DINODE_MAGIC,
  3106. mp, XFS_ERRTAG_IFLUSH_2, XFS_RANDOM_IFLUSH_2)) {
  3107. xfs_cmn_err(XFS_PTAG_IFLUSH, CE_ALERT, mp,
  3108. "xfs_iflush: Bad inode %Lu, ptr 0x%p, magic number 0x%x",
  3109. ip->i_ino, ip, ip->i_d.di_magic);
  3110. goto corrupt_out;
  3111. }
  3112. if ((ip->i_d.di_mode & S_IFMT) == S_IFREG) {
  3113. if (XFS_TEST_ERROR(
  3114. (ip->i_d.di_format != XFS_DINODE_FMT_EXTENTS) &&
  3115. (ip->i_d.di_format != XFS_DINODE_FMT_BTREE),
  3116. mp, XFS_ERRTAG_IFLUSH_3, XFS_RANDOM_IFLUSH_3)) {
  3117. xfs_cmn_err(XFS_PTAG_IFLUSH, CE_ALERT, mp,
  3118. "xfs_iflush: Bad regular inode %Lu, ptr 0x%p",
  3119. ip->i_ino, ip);
  3120. goto corrupt_out;
  3121. }
  3122. } else if ((ip->i_d.di_mode & S_IFMT) == S_IFDIR) {
  3123. if (XFS_TEST_ERROR(
  3124. (ip->i_d.di_format != XFS_DINODE_FMT_EXTENTS) &&
  3125. (ip->i_d.di_format != XFS_DINODE_FMT_BTREE) &&
  3126. (ip->i_d.di_format != XFS_DINODE_FMT_LOCAL),
  3127. mp, XFS_ERRTAG_IFLUSH_4, XFS_RANDOM_IFLUSH_4)) {
  3128. xfs_cmn_err(XFS_PTAG_IFLUSH, CE_ALERT, mp,
  3129. "xfs_iflush: Bad directory inode %Lu, ptr 0x%p",
  3130. ip->i_ino, ip);
  3131. goto corrupt_out;
  3132. }
  3133. }
  3134. if (XFS_TEST_ERROR(ip->i_d.di_nextents + ip->i_d.di_anextents >
  3135. ip->i_d.di_nblocks, mp, XFS_ERRTAG_IFLUSH_5,
  3136. XFS_RANDOM_IFLUSH_5)) {
  3137. xfs_cmn_err(XFS_PTAG_IFLUSH, CE_ALERT, mp,
  3138. "xfs_iflush: detected corrupt incore inode %Lu, total extents = %d, nblocks = %Ld, ptr 0x%p",
  3139. ip->i_ino,
  3140. ip->i_d.di_nextents + ip->i_d.di_anextents,
  3141. ip->i_d.di_nblocks,
  3142. ip);
  3143. goto corrupt_out;
  3144. }
  3145. if (XFS_TEST_ERROR(ip->i_d.di_forkoff > mp->m_sb.sb_inodesize,
  3146. mp, XFS_ERRTAG_IFLUSH_6, XFS_RANDOM_IFLUSH_6)) {
  3147. xfs_cmn_err(XFS_PTAG_IFLUSH, CE_ALERT, mp,
  3148. "xfs_iflush: bad inode %Lu, forkoff 0x%x, ptr 0x%p",
  3149. ip->i_ino, ip->i_d.di_forkoff, ip);
  3150. goto corrupt_out;
  3151. }
  3152. /*
  3153. * bump the flush iteration count, used to detect flushes which
  3154. * postdate a log record during recovery.
  3155. */
  3156. ip->i_d.di_flushiter++;
  3157. /*
  3158. * Copy the dirty parts of the inode into the on-disk
  3159. * inode. We always copy out the core of the inode,
  3160. * because if the inode is dirty at all the core must
  3161. * be.
  3162. */
  3163. xfs_dinode_to_disk(&dip->di_core, &ip->i_d);
  3164. /* Wrap, we never let the log put out DI_MAX_FLUSH */
  3165. if (ip->i_d.di_flushiter == DI_MAX_FLUSH)
  3166. ip->i_d.di_flushiter = 0;
  3167. /*
  3168. * If this is really an old format inode and the superblock version
  3169. * has not been updated to support only new format inodes, then
  3170. * convert back to the old inode format. If the superblock version
  3171. * has been updated, then make the conversion permanent.
  3172. */
  3173. ASSERT(ip->i_d.di_version == XFS_DINODE_VERSION_1 ||
  3174. xfs_sb_version_hasnlink(&mp->m_sb));
  3175. if (ip->i_d.di_version == XFS_DINODE_VERSION_1) {
  3176. if (!xfs_sb_version_hasnlink(&mp->m_sb)) {
  3177. /*
  3178. * Convert it back.
  3179. */
  3180. ASSERT(ip->i_d.di_nlink <= XFS_MAXLINK_1);
  3181. dip->di_core.di_onlink = cpu_to_be16(ip->i_d.di_nlink);
  3182. } else {
  3183. /*
  3184. * The superblock version has already been bumped,
  3185. * so just make the conversion to the new inode
  3186. * format permanent.
  3187. */
  3188. ip->i_d.di_version = XFS_DINODE_VERSION_2;
  3189. dip->di_core.di_version = XFS_DINODE_VERSION_2;
  3190. ip->i_d.di_onlink = 0;
  3191. dip->di_core.di_onlink = 0;
  3192. memset(&(ip->i_d.di_pad[0]), 0, sizeof(ip->i_d.di_pad));
  3193. memset(&(dip->di_core.di_pad[0]), 0,
  3194. sizeof(dip->di_core.di_pad));
  3195. ASSERT(ip->i_d.di_projid == 0);
  3196. }
  3197. }
  3198. if (xfs_iflush_fork(ip, dip, iip, XFS_DATA_FORK, bp) == EFSCORRUPTED) {
  3199. goto corrupt_out;
  3200. }
  3201. if (XFS_IFORK_Q(ip)) {
  3202. /*
  3203. * The only error from xfs_iflush_fork is on the data fork.
  3204. */
  3205. (void) xfs_iflush_fork(ip, dip, iip, XFS_ATTR_FORK, bp);
  3206. }
  3207. xfs_inobp_check(mp, bp);
  3208. /*
  3209. * We've recorded everything logged in the inode, so we'd
  3210. * like to clear the ilf_fields bits so we don't log and
  3211. * flush things unnecessarily. However, we can't stop
  3212. * logging all this information until the data we've copied
  3213. * into the disk buffer is written to disk. If we did we might
  3214. * overwrite the copy of the inode in the log with all the
  3215. * data after re-logging only part of it, and in the face of
  3216. * a crash we wouldn't have all the data we need to recover.
  3217. *
  3218. * What we do is move the bits to the ili_last_fields field.
  3219. * When logging the inode, these bits are moved back to the
  3220. * ilf_fields field. In the xfs_iflush_done() routine we
  3221. * clear ili_last_fields, since we know that the information
  3222. * those bits represent is permanently on disk. As long as
  3223. * the flush completes before the inode is logged again, then
  3224. * both ilf_fields and ili_last_fields will be cleared.
  3225. *
  3226. * We can play with the ilf_fields bits here, because the inode
  3227. * lock must be held exclusively in order to set bits there
  3228. * and the flush lock protects the ili_last_fields bits.
  3229. * Set ili_logged so the flush done
  3230. * routine can tell whether or not to look in the AIL.
  3231. * Also, store the current LSN of the inode so that we can tell
  3232. * whether the item has moved in the AIL from xfs_iflush_done().
  3233. * In order to read the lsn we need the AIL lock, because
  3234. * it is a 64 bit value that cannot be read atomically.
  3235. */
  3236. if (iip != NULL && iip->ili_format.ilf_fields != 0) {
  3237. iip->ili_last_fields = iip->ili_format.ilf_fields;
  3238. iip->ili_format.ilf_fields = 0;
  3239. iip->ili_logged = 1;
  3240. ASSERT(sizeof(xfs_lsn_t) == 8); /* don't lock if it shrinks */
  3241. spin_lock(&mp->m_ail_lock);
  3242. iip->ili_flush_lsn = iip->ili_item.li_lsn;
  3243. spin_unlock(&mp->m_ail_lock);
  3244. /*
  3245. * Attach the function xfs_iflush_done to the inode's
  3246. * buffer. This will remove the inode from the AIL
  3247. * and unlock the inode's flush lock when the inode is
  3248. * completely written to disk.
  3249. */
  3250. xfs_buf_attach_iodone(bp, (void(*)(xfs_buf_t*,xfs_log_item_t*))
  3251. xfs_iflush_done, (xfs_log_item_t *)iip);
  3252. ASSERT(XFS_BUF_FSPRIVATE(bp, void *) != NULL);
  3253. ASSERT(XFS_BUF_IODONE_FUNC(bp) != NULL);
  3254. } else {
  3255. /*
  3256. * We're flushing an inode which is not in the AIL and has
  3257. * not been logged but has i_update_core set. For this
  3258. * case we can use a B_DELWRI flush and immediately drop
  3259. * the inode flush lock because we can avoid the whole
  3260. * AIL state thing. It's OK to drop the flush lock now,
  3261. * because we've already locked the buffer and to do anything
  3262. * you really need both.
  3263. */
  3264. if (iip != NULL) {
  3265. ASSERT(iip->ili_logged == 0);
  3266. ASSERT(iip->ili_last_fields == 0);
  3267. ASSERT((iip->ili_item.li_flags & XFS_LI_IN_AIL) == 0);
  3268. }
  3269. xfs_ifunlock(ip);
  3270. }
  3271. return 0;
  3272. corrupt_out:
  3273. return XFS_ERROR(EFSCORRUPTED);
  3274. }
  3275. /*
  3276. * Flush all inactive inodes in mp.
  3277. */
  3278. void
  3279. xfs_iflush_all(
  3280. xfs_mount_t *mp)
  3281. {
  3282. xfs_inode_t *ip;
  3283. bhv_vnode_t *vp;
  3284. again:
  3285. XFS_MOUNT_ILOCK(mp);
  3286. ip = mp->m_inodes;
  3287. if (ip == NULL)
  3288. goto out;
  3289. do {
  3290. /* Make sure we skip markers inserted by sync */
  3291. if (ip->i_mount == NULL) {
  3292. ip = ip->i_mnext;
  3293. continue;
  3294. }
  3295. vp = XFS_ITOV_NULL(ip);
  3296. if (!vp) {
  3297. XFS_MOUNT_IUNLOCK(mp);
  3298. xfs_finish_reclaim(ip, 0, XFS_IFLUSH_ASYNC);
  3299. goto again;
  3300. }
  3301. ASSERT(vn_count(vp) == 0);
  3302. ip = ip->i_mnext;
  3303. } while (ip != mp->m_inodes);
  3304. out:
  3305. XFS_MOUNT_IUNLOCK(mp);
  3306. }
  3307. #ifdef XFS_ILOCK_TRACE
  3308. ktrace_t *xfs_ilock_trace_buf;
  3309. void
  3310. xfs_ilock_trace(xfs_inode_t *ip, int lock, unsigned int lockflags, inst_t *ra)
  3311. {
  3312. ktrace_enter(ip->i_lock_trace,
  3313. (void *)ip,
  3314. (void *)(unsigned long)lock, /* 1 = LOCK, 3=UNLOCK, etc */
  3315. (void *)(unsigned long)lockflags, /* XFS_ILOCK_EXCL etc */
  3316. (void *)ra, /* caller of ilock */
  3317. (void *)(unsigned long)current_cpu(),
  3318. (void *)(unsigned long)current_pid(),
  3319. NULL,NULL,NULL,NULL,NULL,NULL,NULL,NULL,NULL,NULL);
  3320. }
  3321. #endif
  3322. /*
  3323. * Return a pointer to the extent record at file index idx.
  3324. */
  3325. xfs_bmbt_rec_host_t *
  3326. xfs_iext_get_ext(
  3327. xfs_ifork_t *ifp, /* inode fork pointer */
  3328. xfs_extnum_t idx) /* index of target extent */
  3329. {
  3330. ASSERT(idx >= 0);
  3331. if ((ifp->if_flags & XFS_IFEXTIREC) && (idx == 0)) {
  3332. return ifp->if_u1.if_ext_irec->er_extbuf;
  3333. } else if (ifp->if_flags & XFS_IFEXTIREC) {
  3334. xfs_ext_irec_t *erp; /* irec pointer */
  3335. int erp_idx = 0; /* irec index */
  3336. xfs_extnum_t page_idx = idx; /* ext index in target list */
  3337. erp = xfs_iext_idx_to_irec(ifp, &page_idx, &erp_idx, 0);
  3338. return &erp->er_extbuf[page_idx];
  3339. } else if (ifp->if_bytes) {
  3340. return &ifp->if_u1.if_extents[idx];
  3341. } else {
  3342. return NULL;
  3343. }
  3344. }
  3345. /*
  3346. * Insert new item(s) into the extent records for incore inode
  3347. * fork 'ifp'. 'count' new items are inserted at index 'idx'.
  3348. */
  3349. void
  3350. xfs_iext_insert(
  3351. xfs_ifork_t *ifp, /* inode fork pointer */
  3352. xfs_extnum_t idx, /* starting index of new items */
  3353. xfs_extnum_t count, /* number of inserted items */
  3354. xfs_bmbt_irec_t *new) /* items to insert */
  3355. {
  3356. xfs_extnum_t i; /* extent record index */
  3357. ASSERT(ifp->if_flags & XFS_IFEXTENTS);
  3358. xfs_iext_add(ifp, idx, count);
  3359. for (i = idx; i < idx + count; i++, new++)
  3360. xfs_bmbt_set_all(xfs_iext_get_ext(ifp, i), new);
  3361. }
  3362. /*
  3363. * This is called when the amount of space required for incore file
  3364. * extents needs to be increased. The ext_diff parameter stores the
  3365. * number of new extents being added and the idx parameter contains
  3366. * the extent index where the new extents will be added. If the new
  3367. * extents are being appended, then we just need to (re)allocate and
  3368. * initialize the space. Otherwise, if the new extents are being
  3369. * inserted into the middle of the existing entries, a bit more work
  3370. * is required to make room for the new extents to be inserted. The
  3371. * caller is responsible for filling in the new extent entries upon
  3372. * return.
  3373. */
  3374. void
  3375. xfs_iext_add(
  3376. xfs_ifork_t *ifp, /* inode fork pointer */
  3377. xfs_extnum_t idx, /* index to begin adding exts */
  3378. int ext_diff) /* number of extents to add */
  3379. {
  3380. int byte_diff; /* new bytes being added */
  3381. int new_size; /* size of extents after adding */
  3382. xfs_extnum_t nextents; /* number of extents in file */
  3383. nextents = ifp->if_bytes / (uint)sizeof(xfs_bmbt_rec_t);
  3384. ASSERT((idx >= 0) && (idx <= nextents));
  3385. byte_diff = ext_diff * sizeof(xfs_bmbt_rec_t);
  3386. new_size = ifp->if_bytes + byte_diff;
  3387. /*
  3388. * If the new number of extents (nextents + ext_diff)
  3389. * fits inside the inode, then continue to use the inline
  3390. * extent buffer.
  3391. */
  3392. if (nextents + ext_diff <= XFS_INLINE_EXTS) {
  3393. if (idx < nextents) {
  3394. memmove(&ifp->if_u2.if_inline_ext[idx + ext_diff],
  3395. &ifp->if_u2.if_inline_ext[idx],
  3396. (nextents - idx) * sizeof(xfs_bmbt_rec_t));
  3397. memset(&ifp->if_u2.if_inline_ext[idx], 0, byte_diff);
  3398. }
  3399. ifp->if_u1.if_extents = ifp->if_u2.if_inline_ext;
  3400. ifp->if_real_bytes = 0;
  3401. ifp->if_lastex = nextents + ext_diff;
  3402. }
  3403. /*
  3404. * Otherwise use a linear (direct) extent list.
  3405. * If the extents are currently inside the inode,
  3406. * xfs_iext_realloc_direct will switch us from
  3407. * inline to direct extent allocation mode.
  3408. */
  3409. else if (nextents + ext_diff <= XFS_LINEAR_EXTS) {
  3410. xfs_iext_realloc_direct(ifp, new_size);
  3411. if (idx < nextents) {
  3412. memmove(&ifp->if_u1.if_extents[idx + ext_diff],
  3413. &ifp->if_u1.if_extents[idx],
  3414. (nextents - idx) * sizeof(xfs_bmbt_rec_t));
  3415. memset(&ifp->if_u1.if_extents[idx], 0, byte_diff);
  3416. }
  3417. }
  3418. /* Indirection array */
  3419. else {
  3420. xfs_ext_irec_t *erp;
  3421. int erp_idx = 0;
  3422. int page_idx = idx;
  3423. ASSERT(nextents + ext_diff > XFS_LINEAR_EXTS);
  3424. if (ifp->if_flags & XFS_IFEXTIREC) {
  3425. erp = xfs_iext_idx_to_irec(ifp, &page_idx, &erp_idx, 1);
  3426. } else {
  3427. xfs_iext_irec_init(ifp);
  3428. ASSERT(ifp->if_flags & XFS_IFEXTIREC);
  3429. erp = ifp->if_u1.if_ext_irec;
  3430. }
  3431. /* Extents fit in target extent page */
  3432. if (erp && erp->er_extcount + ext_diff <= XFS_LINEAR_EXTS) {
  3433. if (page_idx < erp->er_extcount) {
  3434. memmove(&erp->er_extbuf[page_idx + ext_diff],
  3435. &erp->er_extbuf[page_idx],
  3436. (erp->er_extcount - page_idx) *
  3437. sizeof(xfs_bmbt_rec_t));
  3438. memset(&erp->er_extbuf[page_idx], 0, byte_diff);
  3439. }
  3440. erp->er_extcount += ext_diff;
  3441. xfs_iext_irec_update_extoffs(ifp, erp_idx + 1, ext_diff);
  3442. }
  3443. /* Insert a new extent page */
  3444. else if (erp) {
  3445. xfs_iext_add_indirect_multi(ifp,
  3446. erp_idx, page_idx, ext_diff);
  3447. }
  3448. /*
  3449. * If extent(s) are being appended to the last page in
  3450. * the indirection array and the new extent(s) don't fit
  3451. * in the page, then erp is NULL and erp_idx is set to
  3452. * the next index needed in the indirection array.
  3453. */
  3454. else {
  3455. int count = ext_diff;
  3456. while (count) {
  3457. erp = xfs_iext_irec_new(ifp, erp_idx);
  3458. erp->er_extcount = count;
  3459. count -= MIN(count, (int)XFS_LINEAR_EXTS);
  3460. if (count) {
  3461. erp_idx++;
  3462. }
  3463. }
  3464. }
  3465. }
  3466. ifp->if_bytes = new_size;
  3467. }
  3468. /*
  3469. * This is called when incore extents are being added to the indirection
  3470. * array and the new extents do not fit in the target extent list. The
  3471. * erp_idx parameter contains the irec index for the target extent list
  3472. * in the indirection array, and the idx parameter contains the extent
  3473. * index within the list. The number of extents being added is stored
  3474. * in the count parameter.
  3475. *
  3476. * |-------| |-------|
  3477. * | | | | idx - number of extents before idx
  3478. * | idx | | count |
  3479. * | | | | count - number of extents being inserted at idx
  3480. * |-------| |-------|
  3481. * | count | | nex2 | nex2 - number of extents after idx + count
  3482. * |-------| |-------|
  3483. */
  3484. void
  3485. xfs_iext_add_indirect_multi(
  3486. xfs_ifork_t *ifp, /* inode fork pointer */
  3487. int erp_idx, /* target extent irec index */
  3488. xfs_extnum_t idx, /* index within target list */
  3489. int count) /* new extents being added */
  3490. {
  3491. int byte_diff; /* new bytes being added */
  3492. xfs_ext_irec_t *erp; /* pointer to irec entry */
  3493. xfs_extnum_t ext_diff; /* number of extents to add */
  3494. xfs_extnum_t ext_cnt; /* new extents still needed */
  3495. xfs_extnum_t nex2; /* extents after idx + count */
  3496. xfs_bmbt_rec_t *nex2_ep = NULL; /* temp list for nex2 extents */
  3497. int nlists; /* number of irec's (lists) */
  3498. ASSERT(ifp->if_flags & XFS_IFEXTIREC);
  3499. erp = &ifp->if_u1.if_ext_irec[erp_idx];
  3500. nex2 = erp->er_extcount - idx;
  3501. nlists = ifp->if_real_bytes / XFS_IEXT_BUFSZ;
  3502. /*
  3503. * Save second part of target extent list
  3504. * (all extents past */
  3505. if (nex2) {
  3506. byte_diff = nex2 * sizeof(xfs_bmbt_rec_t);
  3507. nex2_ep = (xfs_bmbt_rec_t *) kmem_alloc(byte_diff, KM_SLEEP);
  3508. memmove(nex2_ep, &erp->er_extbuf[idx], byte_diff);
  3509. erp->er_extcount -= nex2;
  3510. xfs_iext_irec_update_extoffs(ifp, erp_idx + 1, -nex2);
  3511. memset(&erp->er_extbuf[idx], 0, byte_diff);
  3512. }
  3513. /*
  3514. * Add the new extents to the end of the target
  3515. * list, then allocate new irec record(s) and
  3516. * extent buffer(s) as needed to store the rest
  3517. * of the new extents.
  3518. */
  3519. ext_cnt = count;
  3520. ext_diff = MIN(ext_cnt, (int)XFS_LINEAR_EXTS - erp->er_extcount);
  3521. if (ext_diff) {
  3522. erp->er_extcount += ext_diff;
  3523. xfs_iext_irec_update_extoffs(ifp, erp_idx + 1, ext_diff);
  3524. ext_cnt -= ext_diff;
  3525. }
  3526. while (ext_cnt) {
  3527. erp_idx++;
  3528. erp = xfs_iext_irec_new(ifp, erp_idx);
  3529. ext_diff = MIN(ext_cnt, (int)XFS_LINEAR_EXTS);
  3530. erp->er_extcount = ext_diff;
  3531. xfs_iext_irec_update_extoffs(ifp, erp_idx + 1, ext_diff);
  3532. ext_cnt -= ext_diff;
  3533. }
  3534. /* Add nex2 extents back to indirection array */
  3535. if (nex2) {
  3536. xfs_extnum_t ext_avail;
  3537. int i;
  3538. byte_diff = nex2 * sizeof(xfs_bmbt_rec_t);
  3539. ext_avail = XFS_LINEAR_EXTS - erp->er_extcount;
  3540. i = 0;
  3541. /*
  3542. * If nex2 extents fit in the current page, append
  3543. * nex2_ep after the new extents.
  3544. */
  3545. if (nex2 <= ext_avail) {
  3546. i = erp->er_extcount;
  3547. }
  3548. /*
  3549. * Otherwise, check if space is available in the
  3550. * next page.
  3551. */
  3552. else if ((erp_idx < nlists - 1) &&
  3553. (nex2 <= (ext_avail = XFS_LINEAR_EXTS -
  3554. ifp->if_u1.if_ext_irec[erp_idx+1].er_extcount))) {
  3555. erp_idx++;
  3556. erp++;
  3557. /* Create a hole for nex2 extents */
  3558. memmove(&erp->er_extbuf[nex2], erp->er_extbuf,
  3559. erp->er_extcount * sizeof(xfs_bmbt_rec_t));
  3560. }
  3561. /*
  3562. * Final choice, create a new extent page for
  3563. * nex2 extents.
  3564. */
  3565. else {
  3566. erp_idx++;
  3567. erp = xfs_iext_irec_new(ifp, erp_idx);
  3568. }
  3569. memmove(&erp->er_extbuf[i], nex2_ep, byte_diff);
  3570. kmem_free(nex2_ep, byte_diff);
  3571. erp->er_extcount += nex2;
  3572. xfs_iext_irec_update_extoffs(ifp, erp_idx + 1, nex2);
  3573. }
  3574. }
  3575. /*
  3576. * This is called when the amount of space required for incore file
  3577. * extents needs to be decreased. The ext_diff parameter stores the
  3578. * number of extents to be removed and the idx parameter contains
  3579. * the extent index where the extents will be removed from.
  3580. *
  3581. * If the amount of space needed has decreased below the linear
  3582. * limit, XFS_IEXT_BUFSZ, then switch to using the contiguous
  3583. * extent array. Otherwise, use kmem_realloc() to adjust the
  3584. * size to what is needed.
  3585. */
  3586. void
  3587. xfs_iext_remove(
  3588. xfs_ifork_t *ifp, /* inode fork pointer */
  3589. xfs_extnum_t idx, /* index to begin removing exts */
  3590. int ext_diff) /* number of extents to remove */
  3591. {
  3592. xfs_extnum_t nextents; /* number of extents in file */
  3593. int new_size; /* size of extents after removal */
  3594. ASSERT(ext_diff > 0);
  3595. nextents = ifp->if_bytes / (uint)sizeof(xfs_bmbt_rec_t);
  3596. new_size = (nextents - ext_diff) * sizeof(xfs_bmbt_rec_t);
  3597. if (new_size == 0) {
  3598. xfs_iext_destroy(ifp);
  3599. } else if (ifp->if_flags & XFS_IFEXTIREC) {
  3600. xfs_iext_remove_indirect(ifp, idx, ext_diff);
  3601. } else if (ifp->if_real_bytes) {
  3602. xfs_iext_remove_direct(ifp, idx, ext_diff);
  3603. } else {
  3604. xfs_iext_remove_inline(ifp, idx, ext_diff);
  3605. }
  3606. ifp->if_bytes = new_size;
  3607. }
  3608. /*
  3609. * This removes ext_diff extents from the inline buffer, beginning
  3610. * at extent index idx.
  3611. */
  3612. void
  3613. xfs_iext_remove_inline(
  3614. xfs_ifork_t *ifp, /* inode fork pointer */
  3615. xfs_extnum_t idx, /* index to begin removing exts */
  3616. int ext_diff) /* number of extents to remove */
  3617. {
  3618. int nextents; /* number of extents in file */
  3619. ASSERT(!(ifp->if_flags & XFS_IFEXTIREC));
  3620. ASSERT(idx < XFS_INLINE_EXTS);
  3621. nextents = ifp->if_bytes / (uint)sizeof(xfs_bmbt_rec_t);
  3622. ASSERT(((nextents - ext_diff) > 0) &&
  3623. (nextents - ext_diff) < XFS_INLINE_EXTS);
  3624. if (idx + ext_diff < nextents) {
  3625. memmove(&ifp->if_u2.if_inline_ext[idx],
  3626. &ifp->if_u2.if_inline_ext[idx + ext_diff],
  3627. (nextents - (idx + ext_diff)) *
  3628. sizeof(xfs_bmbt_rec_t));
  3629. memset(&ifp->if_u2.if_inline_ext[nextents - ext_diff],
  3630. 0, ext_diff * sizeof(xfs_bmbt_rec_t));
  3631. } else {
  3632. memset(&ifp->if_u2.if_inline_ext[idx], 0,
  3633. ext_diff * sizeof(xfs_bmbt_rec_t));
  3634. }
  3635. }
  3636. /*
  3637. * This removes ext_diff extents from a linear (direct) extent list,
  3638. * beginning at extent index idx. If the extents are being removed
  3639. * from the end of the list (ie. truncate) then we just need to re-
  3640. * allocate the list to remove the extra space. Otherwise, if the
  3641. * extents are being removed from the middle of the existing extent
  3642. * entries, then we first need to move the extent records beginning
  3643. * at idx + ext_diff up in the list to overwrite the records being
  3644. * removed, then remove the extra space via kmem_realloc.
  3645. */
  3646. void
  3647. xfs_iext_remove_direct(
  3648. xfs_ifork_t *ifp, /* inode fork pointer */
  3649. xfs_extnum_t idx, /* index to begin removing exts */
  3650. int ext_diff) /* number of extents to remove */
  3651. {
  3652. xfs_extnum_t nextents; /* number of extents in file */
  3653. int new_size; /* size of extents after removal */
  3654. ASSERT(!(ifp->if_flags & XFS_IFEXTIREC));
  3655. new_size = ifp->if_bytes -
  3656. (ext_diff * sizeof(xfs_bmbt_rec_t));
  3657. nextents = ifp->if_bytes / (uint)sizeof(xfs_bmbt_rec_t);
  3658. if (new_size == 0) {
  3659. xfs_iext_destroy(ifp);
  3660. return;
  3661. }
  3662. /* Move extents up in the list (if needed) */
  3663. if (idx + ext_diff < nextents) {
  3664. memmove(&ifp->if_u1.if_extents[idx],
  3665. &ifp->if_u1.if_extents[idx + ext_diff],
  3666. (nextents - (idx + ext_diff)) *
  3667. sizeof(xfs_bmbt_rec_t));
  3668. }
  3669. memset(&ifp->if_u1.if_extents[nextents - ext_diff],
  3670. 0, ext_diff * sizeof(xfs_bmbt_rec_t));
  3671. /*
  3672. * Reallocate the direct extent list. If the extents
  3673. * will fit inside the inode then xfs_iext_realloc_direct
  3674. * will switch from direct to inline extent allocation
  3675. * mode for us.
  3676. */
  3677. xfs_iext_realloc_direct(ifp, new_size);
  3678. ifp->if_bytes = new_size;
  3679. }
  3680. /*
  3681. * This is called when incore extents are being removed from the
  3682. * indirection array and the extents being removed span multiple extent
  3683. * buffers. The idx parameter contains the file extent index where we
  3684. * want to begin removing extents, and the count parameter contains
  3685. * how many extents need to be removed.
  3686. *
  3687. * |-------| |-------|
  3688. * | nex1 | | | nex1 - number of extents before idx
  3689. * |-------| | count |
  3690. * | | | | count - number of extents being removed at idx
  3691. * | count | |-------|
  3692. * | | | nex2 | nex2 - number of extents after idx + count
  3693. * |-------| |-------|
  3694. */
  3695. void
  3696. xfs_iext_remove_indirect(
  3697. xfs_ifork_t *ifp, /* inode fork pointer */
  3698. xfs_extnum_t idx, /* index to begin removing extents */
  3699. int count) /* number of extents to remove */
  3700. {
  3701. xfs_ext_irec_t *erp; /* indirection array pointer */
  3702. int erp_idx = 0; /* indirection array index */
  3703. xfs_extnum_t ext_cnt; /* extents left to remove */
  3704. xfs_extnum_t ext_diff; /* extents to remove in current list */
  3705. xfs_extnum_t nex1; /* number of extents before idx */
  3706. xfs_extnum_t nex2; /* extents after idx + count */
  3707. int nlists; /* entries in indirection array */
  3708. int page_idx = idx; /* index in target extent list */
  3709. ASSERT(ifp->if_flags & XFS_IFEXTIREC);
  3710. erp = xfs_iext_idx_to_irec(ifp, &page_idx, &erp_idx, 0);
  3711. ASSERT(erp != NULL);
  3712. nlists = ifp->if_real_bytes / XFS_IEXT_BUFSZ;
  3713. nex1 = page_idx;
  3714. ext_cnt = count;
  3715. while (ext_cnt) {
  3716. nex2 = MAX((erp->er_extcount - (nex1 + ext_cnt)), 0);
  3717. ext_diff = MIN(ext_cnt, (erp->er_extcount - nex1));
  3718. /*
  3719. * Check for deletion of entire list;
  3720. * xfs_iext_irec_remove() updates extent offsets.
  3721. */
  3722. if (ext_diff == erp->er_extcount) {
  3723. xfs_iext_irec_remove(ifp, erp_idx);
  3724. ext_cnt -= ext_diff;
  3725. nex1 = 0;
  3726. if (ext_cnt) {
  3727. ASSERT(erp_idx < ifp->if_real_bytes /
  3728. XFS_IEXT_BUFSZ);
  3729. erp = &ifp->if_u1.if_ext_irec[erp_idx];
  3730. nex1 = 0;
  3731. continue;
  3732. } else {
  3733. break;
  3734. }
  3735. }
  3736. /* Move extents up (if needed) */
  3737. if (nex2) {
  3738. memmove(&erp->er_extbuf[nex1],
  3739. &erp->er_extbuf[nex1 + ext_diff],
  3740. nex2 * sizeof(xfs_bmbt_rec_t));
  3741. }
  3742. /* Zero out rest of page */
  3743. memset(&erp->er_extbuf[nex1 + nex2], 0, (XFS_IEXT_BUFSZ -
  3744. ((nex1 + nex2) * sizeof(xfs_bmbt_rec_t))));
  3745. /* Update remaining counters */
  3746. erp->er_extcount -= ext_diff;
  3747. xfs_iext_irec_update_extoffs(ifp, erp_idx + 1, -ext_diff);
  3748. ext_cnt -= ext_diff;
  3749. nex1 = 0;
  3750. erp_idx++;
  3751. erp++;
  3752. }
  3753. ifp->if_bytes -= count * sizeof(xfs_bmbt_rec_t);
  3754. xfs_iext_irec_compact(ifp);
  3755. }
  3756. /*
  3757. * Create, destroy, or resize a linear (direct) block of extents.
  3758. */
  3759. void
  3760. xfs_iext_realloc_direct(
  3761. xfs_ifork_t *ifp, /* inode fork pointer */
  3762. int new_size) /* new size of extents */
  3763. {
  3764. int rnew_size; /* real new size of extents */
  3765. rnew_size = new_size;
  3766. ASSERT(!(ifp->if_flags & XFS_IFEXTIREC) ||
  3767. ((new_size >= 0) && (new_size <= XFS_IEXT_BUFSZ) &&
  3768. (new_size != ifp->if_real_bytes)));
  3769. /* Free extent records */
  3770. if (new_size == 0) {
  3771. xfs_iext_destroy(ifp);
  3772. }
  3773. /* Resize direct extent list and zero any new bytes */
  3774. else if (ifp->if_real_bytes) {
  3775. /* Check if extents will fit inside the inode */
  3776. if (new_size <= XFS_INLINE_EXTS * sizeof(xfs_bmbt_rec_t)) {
  3777. xfs_iext_direct_to_inline(ifp, new_size /
  3778. (uint)sizeof(xfs_bmbt_rec_t));
  3779. ifp->if_bytes = new_size;
  3780. return;
  3781. }
  3782. if (!is_power_of_2(new_size)){
  3783. rnew_size = roundup_pow_of_two(new_size);
  3784. }
  3785. if (rnew_size != ifp->if_real_bytes) {
  3786. ifp->if_u1.if_extents =
  3787. kmem_realloc(ifp->if_u1.if_extents,
  3788. rnew_size,
  3789. ifp->if_real_bytes,
  3790. KM_SLEEP);
  3791. }
  3792. if (rnew_size > ifp->if_real_bytes) {
  3793. memset(&ifp->if_u1.if_extents[ifp->if_bytes /
  3794. (uint)sizeof(xfs_bmbt_rec_t)], 0,
  3795. rnew_size - ifp->if_real_bytes);
  3796. }
  3797. }
  3798. /*
  3799. * Switch from the inline extent buffer to a direct
  3800. * extent list. Be sure to include the inline extent
  3801. * bytes in new_size.
  3802. */
  3803. else {
  3804. new_size += ifp->if_bytes;
  3805. if (!is_power_of_2(new_size)) {
  3806. rnew_size = roundup_pow_of_two(new_size);
  3807. }
  3808. xfs_iext_inline_to_direct(ifp, rnew_size);
  3809. }
  3810. ifp->if_real_bytes = rnew_size;
  3811. ifp->if_bytes = new_size;
  3812. }
  3813. /*
  3814. * Switch from linear (direct) extent records to inline buffer.
  3815. */
  3816. void
  3817. xfs_iext_direct_to_inline(
  3818. xfs_ifork_t *ifp, /* inode fork pointer */
  3819. xfs_extnum_t nextents) /* number of extents in file */
  3820. {
  3821. ASSERT(ifp->if_flags & XFS_IFEXTENTS);
  3822. ASSERT(nextents <= XFS_INLINE_EXTS);
  3823. /*
  3824. * The inline buffer was zeroed when we switched
  3825. * from inline to direct extent allocation mode,
  3826. * so we don't need to clear it here.
  3827. */
  3828. memcpy(ifp->if_u2.if_inline_ext, ifp->if_u1.if_extents,
  3829. nextents * sizeof(xfs_bmbt_rec_t));
  3830. kmem_free(ifp->if_u1.if_extents, ifp->if_real_bytes);
  3831. ifp->if_u1.if_extents = ifp->if_u2.if_inline_ext;
  3832. ifp->if_real_bytes = 0;
  3833. }
  3834. /*
  3835. * Switch from inline buffer to linear (direct) extent records.
  3836. * new_size should already be rounded up to the next power of 2
  3837. * by the caller (when appropriate), so use new_size as it is.
  3838. * However, since new_size may be rounded up, we can't update
  3839. * if_bytes here. It is the caller's responsibility to update
  3840. * if_bytes upon return.
  3841. */
  3842. void
  3843. xfs_iext_inline_to_direct(
  3844. xfs_ifork_t *ifp, /* inode fork pointer */
  3845. int new_size) /* number of extents in file */
  3846. {
  3847. ifp->if_u1.if_extents = kmem_alloc(new_size, KM_SLEEP);
  3848. memset(ifp->if_u1.if_extents, 0, new_size);
  3849. if (ifp->if_bytes) {
  3850. memcpy(ifp->if_u1.if_extents, ifp->if_u2.if_inline_ext,
  3851. ifp->if_bytes);
  3852. memset(ifp->if_u2.if_inline_ext, 0, XFS_INLINE_EXTS *
  3853. sizeof(xfs_bmbt_rec_t));
  3854. }
  3855. ifp->if_real_bytes = new_size;
  3856. }
  3857. /*
  3858. * Resize an extent indirection array to new_size bytes.
  3859. */
  3860. void
  3861. xfs_iext_realloc_indirect(
  3862. xfs_ifork_t *ifp, /* inode fork pointer */
  3863. int new_size) /* new indirection array size */
  3864. {
  3865. int nlists; /* number of irec's (ex lists) */
  3866. int size; /* current indirection array size */
  3867. ASSERT(ifp->if_flags & XFS_IFEXTIREC);
  3868. nlists = ifp->if_real_bytes / XFS_IEXT_BUFSZ;
  3869. size = nlists * sizeof(xfs_ext_irec_t);
  3870. ASSERT(ifp->if_real_bytes);
  3871. ASSERT((new_size >= 0) && (new_size != size));
  3872. if (new_size == 0) {
  3873. xfs_iext_destroy(ifp);
  3874. } else {
  3875. ifp->if_u1.if_ext_irec = (xfs_ext_irec_t *)
  3876. kmem_realloc(ifp->if_u1.if_ext_irec,
  3877. new_size, size, KM_SLEEP);
  3878. }
  3879. }
  3880. /*
  3881. * Switch from indirection array to linear (direct) extent allocations.
  3882. */
  3883. void
  3884. xfs_iext_indirect_to_direct(
  3885. xfs_ifork_t *ifp) /* inode fork pointer */
  3886. {
  3887. xfs_bmbt_rec_host_t *ep; /* extent record pointer */
  3888. xfs_extnum_t nextents; /* number of extents in file */
  3889. int size; /* size of file extents */
  3890. ASSERT(ifp->if_flags & XFS_IFEXTIREC);
  3891. nextents = ifp->if_bytes / (uint)sizeof(xfs_bmbt_rec_t);
  3892. ASSERT(nextents <= XFS_LINEAR_EXTS);
  3893. size = nextents * sizeof(xfs_bmbt_rec_t);
  3894. xfs_iext_irec_compact_full(ifp);
  3895. ASSERT(ifp->if_real_bytes == XFS_IEXT_BUFSZ);
  3896. ep = ifp->if_u1.if_ext_irec->er_extbuf;
  3897. kmem_free(ifp->if_u1.if_ext_irec, sizeof(xfs_ext_irec_t));
  3898. ifp->if_flags &= ~XFS_IFEXTIREC;
  3899. ifp->if_u1.if_extents = ep;
  3900. ifp->if_bytes = size;
  3901. if (nextents < XFS_LINEAR_EXTS) {
  3902. xfs_iext_realloc_direct(ifp, size);
  3903. }
  3904. }
  3905. /*
  3906. * Free incore file extents.
  3907. */
  3908. void
  3909. xfs_iext_destroy(
  3910. xfs_ifork_t *ifp) /* inode fork pointer */
  3911. {
  3912. if (ifp->if_flags & XFS_IFEXTIREC) {
  3913. int erp_idx;
  3914. int nlists;
  3915. nlists = ifp->if_real_bytes / XFS_IEXT_BUFSZ;
  3916. for (erp_idx = nlists - 1; erp_idx >= 0 ; erp_idx--) {
  3917. xfs_iext_irec_remove(ifp, erp_idx);
  3918. }
  3919. ifp->if_flags &= ~XFS_IFEXTIREC;
  3920. } else if (ifp->if_real_bytes) {
  3921. kmem_free(ifp->if_u1.if_extents, ifp->if_real_bytes);
  3922. } else if (ifp->if_bytes) {
  3923. memset(ifp->if_u2.if_inline_ext, 0, XFS_INLINE_EXTS *
  3924. sizeof(xfs_bmbt_rec_t));
  3925. }
  3926. ifp->if_u1.if_extents = NULL;
  3927. ifp->if_real_bytes = 0;
  3928. ifp->if_bytes = 0;
  3929. }
  3930. /*
  3931. * Return a pointer to the extent record for file system block bno.
  3932. */
  3933. xfs_bmbt_rec_host_t * /* pointer to found extent record */
  3934. xfs_iext_bno_to_ext(
  3935. xfs_ifork_t *ifp, /* inode fork pointer */
  3936. xfs_fileoff_t bno, /* block number to search for */
  3937. xfs_extnum_t *idxp) /* index of target extent */
  3938. {
  3939. xfs_bmbt_rec_host_t *base; /* pointer to first extent */
  3940. xfs_filblks_t blockcount = 0; /* number of blocks in extent */
  3941. xfs_bmbt_rec_host_t *ep = NULL; /* pointer to target extent */
  3942. xfs_ext_irec_t *erp = NULL; /* indirection array pointer */
  3943. int high; /* upper boundary in search */
  3944. xfs_extnum_t idx = 0; /* index of target extent */
  3945. int low; /* lower boundary in search */
  3946. xfs_extnum_t nextents; /* number of file extents */
  3947. xfs_fileoff_t startoff = 0; /* start offset of extent */
  3948. nextents = ifp->if_bytes / (uint)sizeof(xfs_bmbt_rec_t);
  3949. if (nextents == 0) {
  3950. *idxp = 0;
  3951. return NULL;
  3952. }
  3953. low = 0;
  3954. if (ifp->if_flags & XFS_IFEXTIREC) {
  3955. /* Find target extent list */
  3956. int erp_idx = 0;
  3957. erp = xfs_iext_bno_to_irec(ifp, bno, &erp_idx);
  3958. base = erp->er_extbuf;
  3959. high = erp->er_extcount - 1;
  3960. } else {
  3961. base = ifp->if_u1.if_extents;
  3962. high = nextents - 1;
  3963. }
  3964. /* Binary search extent records */
  3965. while (low <= high) {
  3966. idx = (low + high) >> 1;
  3967. ep = base + idx;
  3968. startoff = xfs_bmbt_get_startoff(ep);
  3969. blockcount = xfs_bmbt_get_blockcount(ep);
  3970. if (bno < startoff) {
  3971. high = idx - 1;
  3972. } else if (bno >= startoff + blockcount) {
  3973. low = idx + 1;
  3974. } else {
  3975. /* Convert back to file-based extent index */
  3976. if (ifp->if_flags & XFS_IFEXTIREC) {
  3977. idx += erp->er_extoff;
  3978. }
  3979. *idxp = idx;
  3980. return ep;
  3981. }
  3982. }
  3983. /* Convert back to file-based extent index */
  3984. if (ifp->if_flags & XFS_IFEXTIREC) {
  3985. idx += erp->er_extoff;
  3986. }
  3987. if (bno >= startoff + blockcount) {
  3988. if (++idx == nextents) {
  3989. ep = NULL;
  3990. } else {
  3991. ep = xfs_iext_get_ext(ifp, idx);
  3992. }
  3993. }
  3994. *idxp = idx;
  3995. return ep;
  3996. }
  3997. /*
  3998. * Return a pointer to the indirection array entry containing the
  3999. * extent record for filesystem block bno. Store the index of the
  4000. * target irec in *erp_idxp.
  4001. */
  4002. xfs_ext_irec_t * /* pointer to found extent record */
  4003. xfs_iext_bno_to_irec(
  4004. xfs_ifork_t *ifp, /* inode fork pointer */
  4005. xfs_fileoff_t bno, /* block number to search for */
  4006. int *erp_idxp) /* irec index of target ext list */
  4007. {
  4008. xfs_ext_irec_t *erp = NULL; /* indirection array pointer */
  4009. xfs_ext_irec_t *erp_next; /* next indirection array entry */
  4010. int erp_idx; /* indirection array index */
  4011. int nlists; /* number of extent irec's (lists) */
  4012. int high; /* binary search upper limit */
  4013. int low; /* binary search lower limit */
  4014. ASSERT(ifp->if_flags & XFS_IFEXTIREC);
  4015. nlists = ifp->if_real_bytes / XFS_IEXT_BUFSZ;
  4016. erp_idx = 0;
  4017. low = 0;
  4018. high = nlists - 1;
  4019. while (low <= high) {
  4020. erp_idx = (low + high) >> 1;
  4021. erp = &ifp->if_u1.if_ext_irec[erp_idx];
  4022. erp_next = erp_idx < nlists - 1 ? erp + 1 : NULL;
  4023. if (bno < xfs_bmbt_get_startoff(erp->er_extbuf)) {
  4024. high = erp_idx - 1;
  4025. } else if (erp_next && bno >=
  4026. xfs_bmbt_get_startoff(erp_next->er_extbuf)) {
  4027. low = erp_idx + 1;
  4028. } else {
  4029. break;
  4030. }
  4031. }
  4032. *erp_idxp = erp_idx;
  4033. return erp;
  4034. }
  4035. /*
  4036. * Return a pointer to the indirection array entry containing the
  4037. * extent record at file extent index *idxp. Store the index of the
  4038. * target irec in *erp_idxp and store the page index of the target
  4039. * extent record in *idxp.
  4040. */
  4041. xfs_ext_irec_t *
  4042. xfs_iext_idx_to_irec(
  4043. xfs_ifork_t *ifp, /* inode fork pointer */
  4044. xfs_extnum_t *idxp, /* extent index (file -> page) */
  4045. int *erp_idxp, /* pointer to target irec */
  4046. int realloc) /* new bytes were just added */
  4047. {
  4048. xfs_ext_irec_t *prev; /* pointer to previous irec */
  4049. xfs_ext_irec_t *erp = NULL; /* pointer to current irec */
  4050. int erp_idx; /* indirection array index */
  4051. int nlists; /* number of irec's (ex lists) */
  4052. int high; /* binary search upper limit */
  4053. int low; /* binary search lower limit */
  4054. xfs_extnum_t page_idx = *idxp; /* extent index in target list */
  4055. ASSERT(ifp->if_flags & XFS_IFEXTIREC);
  4056. ASSERT(page_idx >= 0 && page_idx <=
  4057. ifp->if_bytes / (uint)sizeof(xfs_bmbt_rec_t));
  4058. nlists = ifp->if_real_bytes / XFS_IEXT_BUFSZ;
  4059. erp_idx = 0;
  4060. low = 0;
  4061. high = nlists - 1;
  4062. /* Binary search extent irec's */
  4063. while (low <= high) {
  4064. erp_idx = (low + high) >> 1;
  4065. erp = &ifp->if_u1.if_ext_irec[erp_idx];
  4066. prev = erp_idx > 0 ? erp - 1 : NULL;
  4067. if (page_idx < erp->er_extoff || (page_idx == erp->er_extoff &&
  4068. realloc && prev && prev->er_extcount < XFS_LINEAR_EXTS)) {
  4069. high = erp_idx - 1;
  4070. } else if (page_idx > erp->er_extoff + erp->er_extcount ||
  4071. (page_idx == erp->er_extoff + erp->er_extcount &&
  4072. !realloc)) {
  4073. low = erp_idx + 1;
  4074. } else if (page_idx == erp->er_extoff + erp->er_extcount &&
  4075. erp->er_extcount == XFS_LINEAR_EXTS) {
  4076. ASSERT(realloc);
  4077. page_idx = 0;
  4078. erp_idx++;
  4079. erp = erp_idx < nlists ? erp + 1 : NULL;
  4080. break;
  4081. } else {
  4082. page_idx -= erp->er_extoff;
  4083. break;
  4084. }
  4085. }
  4086. *idxp = page_idx;
  4087. *erp_idxp = erp_idx;
  4088. return(erp);
  4089. }
  4090. /*
  4091. * Allocate and initialize an indirection array once the space needed
  4092. * for incore extents increases above XFS_IEXT_BUFSZ.
  4093. */
  4094. void
  4095. xfs_iext_irec_init(
  4096. xfs_ifork_t *ifp) /* inode fork pointer */
  4097. {
  4098. xfs_ext_irec_t *erp; /* indirection array pointer */
  4099. xfs_extnum_t nextents; /* number of extents in file */
  4100. ASSERT(!(ifp->if_flags & XFS_IFEXTIREC));
  4101. nextents = ifp->if_bytes / (uint)sizeof(xfs_bmbt_rec_t);
  4102. ASSERT(nextents <= XFS_LINEAR_EXTS);
  4103. erp = (xfs_ext_irec_t *)
  4104. kmem_alloc(sizeof(xfs_ext_irec_t), KM_SLEEP);
  4105. if (nextents == 0) {
  4106. ifp->if_u1.if_extents = kmem_alloc(XFS_IEXT_BUFSZ, KM_SLEEP);
  4107. } else if (!ifp->if_real_bytes) {
  4108. xfs_iext_inline_to_direct(ifp, XFS_IEXT_BUFSZ);
  4109. } else if (ifp->if_real_bytes < XFS_IEXT_BUFSZ) {
  4110. xfs_iext_realloc_direct(ifp, XFS_IEXT_BUFSZ);
  4111. }
  4112. erp->er_extbuf = ifp->if_u1.if_extents;
  4113. erp->er_extcount = nextents;
  4114. erp->er_extoff = 0;
  4115. ifp->if_flags |= XFS_IFEXTIREC;
  4116. ifp->if_real_bytes = XFS_IEXT_BUFSZ;
  4117. ifp->if_bytes = nextents * sizeof(xfs_bmbt_rec_t);
  4118. ifp->if_u1.if_ext_irec = erp;
  4119. return;
  4120. }
  4121. /*
  4122. * Allocate and initialize a new entry in the indirection array.
  4123. */
  4124. xfs_ext_irec_t *
  4125. xfs_iext_irec_new(
  4126. xfs_ifork_t *ifp, /* inode fork pointer */
  4127. int erp_idx) /* index for new irec */
  4128. {
  4129. xfs_ext_irec_t *erp; /* indirection array pointer */
  4130. int i; /* loop counter */
  4131. int nlists; /* number of irec's (ex lists) */
  4132. ASSERT(ifp->if_flags & XFS_IFEXTIREC);
  4133. nlists = ifp->if_real_bytes / XFS_IEXT_BUFSZ;
  4134. /* Resize indirection array */
  4135. xfs_iext_realloc_indirect(ifp, ++nlists *
  4136. sizeof(xfs_ext_irec_t));
  4137. /*
  4138. * Move records down in the array so the
  4139. * new page can use erp_idx.
  4140. */
  4141. erp = ifp->if_u1.if_ext_irec;
  4142. for (i = nlists - 1; i > erp_idx; i--) {
  4143. memmove(&erp[i], &erp[i-1], sizeof(xfs_ext_irec_t));
  4144. }
  4145. ASSERT(i == erp_idx);
  4146. /* Initialize new extent record */
  4147. erp = ifp->if_u1.if_ext_irec;
  4148. erp[erp_idx].er_extbuf = kmem_alloc(XFS_IEXT_BUFSZ, KM_SLEEP);
  4149. ifp->if_real_bytes = nlists * XFS_IEXT_BUFSZ;
  4150. memset(erp[erp_idx].er_extbuf, 0, XFS_IEXT_BUFSZ);
  4151. erp[erp_idx].er_extcount = 0;
  4152. erp[erp_idx].er_extoff = erp_idx > 0 ?
  4153. erp[erp_idx-1].er_extoff + erp[erp_idx-1].er_extcount : 0;
  4154. return (&erp[erp_idx]);
  4155. }
  4156. /*
  4157. * Remove a record from the indirection array.
  4158. */
  4159. void
  4160. xfs_iext_irec_remove(
  4161. xfs_ifork_t *ifp, /* inode fork pointer */
  4162. int erp_idx) /* irec index to remove */
  4163. {
  4164. xfs_ext_irec_t *erp; /* indirection array pointer */
  4165. int i; /* loop counter */
  4166. int nlists; /* number of irec's (ex lists) */
  4167. ASSERT(ifp->if_flags & XFS_IFEXTIREC);
  4168. nlists = ifp->if_real_bytes / XFS_IEXT_BUFSZ;
  4169. erp = &ifp->if_u1.if_ext_irec[erp_idx];
  4170. if (erp->er_extbuf) {
  4171. xfs_iext_irec_update_extoffs(ifp, erp_idx + 1,
  4172. -erp->er_extcount);
  4173. kmem_free(erp->er_extbuf, XFS_IEXT_BUFSZ);
  4174. }
  4175. /* Compact extent records */
  4176. erp = ifp->if_u1.if_ext_irec;
  4177. for (i = erp_idx; i < nlists - 1; i++) {
  4178. memmove(&erp[i], &erp[i+1], sizeof(xfs_ext_irec_t));
  4179. }
  4180. /*
  4181. * Manually free the last extent record from the indirection
  4182. * array. A call to xfs_iext_realloc_indirect() with a size
  4183. * of zero would result in a call to xfs_iext_destroy() which
  4184. * would in turn call this function again, creating a nasty
  4185. * infinite loop.
  4186. */
  4187. if (--nlists) {
  4188. xfs_iext_realloc_indirect(ifp,
  4189. nlists * sizeof(xfs_ext_irec_t));
  4190. } else {
  4191. kmem_free(ifp->if_u1.if_ext_irec,
  4192. sizeof(xfs_ext_irec_t));
  4193. }
  4194. ifp->if_real_bytes = nlists * XFS_IEXT_BUFSZ;
  4195. }
  4196. /*
  4197. * This is called to clean up large amounts of unused memory allocated
  4198. * by the indirection array. Before compacting anything though, verify
  4199. * that the indirection array is still needed and switch back to the
  4200. * linear extent list (or even the inline buffer) if possible. The
  4201. * compaction policy is as follows:
  4202. *
  4203. * Full Compaction: Extents fit into a single page (or inline buffer)
  4204. * Full Compaction: Extents occupy less than 10% of allocated space
  4205. * Partial Compaction: Extents occupy > 10% and < 50% of allocated space
  4206. * No Compaction: Extents occupy at least 50% of allocated space
  4207. */
  4208. void
  4209. xfs_iext_irec_compact(
  4210. xfs_ifork_t *ifp) /* inode fork pointer */
  4211. {
  4212. xfs_extnum_t nextents; /* number of extents in file */
  4213. int nlists; /* number of irec's (ex lists) */
  4214. ASSERT(ifp->if_flags & XFS_IFEXTIREC);
  4215. nlists = ifp->if_real_bytes / XFS_IEXT_BUFSZ;
  4216. nextents = ifp->if_bytes / (uint)sizeof(xfs_bmbt_rec_t);
  4217. if (nextents == 0) {
  4218. xfs_iext_destroy(ifp);
  4219. } else if (nextents <= XFS_INLINE_EXTS) {
  4220. xfs_iext_indirect_to_direct(ifp);
  4221. xfs_iext_direct_to_inline(ifp, nextents);
  4222. } else if (nextents <= XFS_LINEAR_EXTS) {
  4223. xfs_iext_indirect_to_direct(ifp);
  4224. } else if (nextents < (nlists * XFS_LINEAR_EXTS) >> 3) {
  4225. xfs_iext_irec_compact_full(ifp);
  4226. } else if (nextents < (nlists * XFS_LINEAR_EXTS) >> 1) {
  4227. xfs_iext_irec_compact_pages(ifp);
  4228. }
  4229. }
  4230. /*
  4231. * Combine extents from neighboring extent pages.
  4232. */
  4233. void
  4234. xfs_iext_irec_compact_pages(
  4235. xfs_ifork_t *ifp) /* inode fork pointer */
  4236. {
  4237. xfs_ext_irec_t *erp, *erp_next;/* pointers to irec entries */
  4238. int erp_idx = 0; /* indirection array index */
  4239. int nlists; /* number of irec's (ex lists) */
  4240. ASSERT(ifp->if_flags & XFS_IFEXTIREC);
  4241. nlists = ifp->if_real_bytes / XFS_IEXT_BUFSZ;
  4242. while (erp_idx < nlists - 1) {
  4243. erp = &ifp->if_u1.if_ext_irec[erp_idx];
  4244. erp_next = erp + 1;
  4245. if (erp_next->er_extcount <=
  4246. (XFS_LINEAR_EXTS - erp->er_extcount)) {
  4247. memmove(&erp->er_extbuf[erp->er_extcount],
  4248. erp_next->er_extbuf, erp_next->er_extcount *
  4249. sizeof(xfs_bmbt_rec_t));
  4250. erp->er_extcount += erp_next->er_extcount;
  4251. /*
  4252. * Free page before removing extent record
  4253. * so er_extoffs don't get modified in
  4254. * xfs_iext_irec_remove.
  4255. */
  4256. kmem_free(erp_next->er_extbuf, XFS_IEXT_BUFSZ);
  4257. erp_next->er_extbuf = NULL;
  4258. xfs_iext_irec_remove(ifp, erp_idx + 1);
  4259. nlists = ifp->if_real_bytes / XFS_IEXT_BUFSZ;
  4260. } else {
  4261. erp_idx++;
  4262. }
  4263. }
  4264. }
  4265. /*
  4266. * Fully compact the extent records managed by the indirection array.
  4267. */
  4268. void
  4269. xfs_iext_irec_compact_full(
  4270. xfs_ifork_t *ifp) /* inode fork pointer */
  4271. {
  4272. xfs_bmbt_rec_host_t *ep, *ep_next; /* extent record pointers */
  4273. xfs_ext_irec_t *erp, *erp_next; /* extent irec pointers */
  4274. int erp_idx = 0; /* extent irec index */
  4275. int ext_avail; /* empty entries in ex list */
  4276. int ext_diff; /* number of exts to add */
  4277. int nlists; /* number of irec's (ex lists) */
  4278. ASSERT(ifp->if_flags & XFS_IFEXTIREC);
  4279. nlists = ifp->if_real_bytes / XFS_IEXT_BUFSZ;
  4280. erp = ifp->if_u1.if_ext_irec;
  4281. ep = &erp->er_extbuf[erp->er_extcount];
  4282. erp_next = erp + 1;
  4283. ep_next = erp_next->er_extbuf;
  4284. while (erp_idx < nlists - 1) {
  4285. ext_avail = XFS_LINEAR_EXTS - erp->er_extcount;
  4286. ext_diff = MIN(ext_avail, erp_next->er_extcount);
  4287. memcpy(ep, ep_next, ext_diff * sizeof(xfs_bmbt_rec_t));
  4288. erp->er_extcount += ext_diff;
  4289. erp_next->er_extcount -= ext_diff;
  4290. /* Remove next page */
  4291. if (erp_next->er_extcount == 0) {
  4292. /*
  4293. * Free page before removing extent record
  4294. * so er_extoffs don't get modified in
  4295. * xfs_iext_irec_remove.
  4296. */
  4297. kmem_free(erp_next->er_extbuf,
  4298. erp_next->er_extcount * sizeof(xfs_bmbt_rec_t));
  4299. erp_next->er_extbuf = NULL;
  4300. xfs_iext_irec_remove(ifp, erp_idx + 1);
  4301. erp = &ifp->if_u1.if_ext_irec[erp_idx];
  4302. nlists = ifp->if_real_bytes / XFS_IEXT_BUFSZ;
  4303. /* Update next page */
  4304. } else {
  4305. /* Move rest of page up to become next new page */
  4306. memmove(erp_next->er_extbuf, ep_next,
  4307. erp_next->er_extcount * sizeof(xfs_bmbt_rec_t));
  4308. ep_next = erp_next->er_extbuf;
  4309. memset(&ep_next[erp_next->er_extcount], 0,
  4310. (XFS_LINEAR_EXTS - erp_next->er_extcount) *
  4311. sizeof(xfs_bmbt_rec_t));
  4312. }
  4313. if (erp->er_extcount == XFS_LINEAR_EXTS) {
  4314. erp_idx++;
  4315. if (erp_idx < nlists)
  4316. erp = &ifp->if_u1.if_ext_irec[erp_idx];
  4317. else
  4318. break;
  4319. }
  4320. ep = &erp->er_extbuf[erp->er_extcount];
  4321. erp_next = erp + 1;
  4322. ep_next = erp_next->er_extbuf;
  4323. }
  4324. }
  4325. /*
  4326. * This is called to update the er_extoff field in the indirection
  4327. * array when extents have been added or removed from one of the
  4328. * extent lists. erp_idx contains the irec index to begin updating
  4329. * at and ext_diff contains the number of extents that were added
  4330. * or removed.
  4331. */
  4332. void
  4333. xfs_iext_irec_update_extoffs(
  4334. xfs_ifork_t *ifp, /* inode fork pointer */
  4335. int erp_idx, /* irec index to update */
  4336. int ext_diff) /* number of new extents */
  4337. {
  4338. int i; /* loop counter */
  4339. int nlists; /* number of irec's (ex lists */
  4340. ASSERT(ifp->if_flags & XFS_IFEXTIREC);
  4341. nlists = ifp->if_real_bytes / XFS_IEXT_BUFSZ;
  4342. for (i = erp_idx; i < nlists; i++) {
  4343. ifp->if_u1.if_ext_irec[i].er_extoff += ext_diff;
  4344. }
  4345. }