xfs_inode.c 137 KB

1234567891011121314151617181920212223242526272829303132333435363738394041424344454647484950515253545556575859606162636465666768697071727374757677787980818283848586878889909192939495969798991001011021031041051061071081091101111121131141151161171181191201211221231241251261271281291301311321331341351361371381391401411421431441451461471481491501511521531541551561571581591601611621631641651661671681691701711721731741751761771781791801811821831841851861871881891901911921931941951961971981992002012022032042052062072082092102112122132142152162172182192202212222232242252262272282292302312322332342352362372382392402412422432442452462472482492502512522532542552562572582592602612622632642652662672682692702712722732742752762772782792802812822832842852862872882892902912922932942952962972982993003013023033043053063073083093103113123133143153163173183193203213223233243253263273283293303313323333343353363373383393403413423433443453463473483493503513523533543553563573583593603613623633643653663673683693703713723733743753763773783793803813823833843853863873883893903913923933943953963973983994004014024034044054064074084094104114124134144154164174184194204214224234244254264274284294304314324334344354364374384394404414424434444454464474484494504514524534544554564574584594604614624634644654664674684694704714724734744754764774784794804814824834844854864874884894904914924934944954964974984995005015025035045055065075085095105115125135145155165175185195205215225235245255265275285295305315325335345355365375385395405415425435445455465475485495505515525535545555565575585595605615625635645655665675685695705715725735745755765775785795805815825835845855865875885895905915925935945955965975985996006016026036046056066076086096106116126136146156166176186196206216226236246256266276286296306316326336346356366376386396406416426436446456466476486496506516526536546556566576586596606616626636646656666676686696706716726736746756766776786796806816826836846856866876886896906916926936946956966976986997007017027037047057067077087097107117127137147157167177187197207217227237247257267277287297307317327337347357367377387397407417427437447457467477487497507517527537547557567577587597607617627637647657667677687697707717727737747757767777787797807817827837847857867877887897907917927937947957967977987998008018028038048058068078088098108118128138148158168178188198208218228238248258268278288298308318328338348358368378388398408418428438448458468478488498508518528538548558568578588598608618628638648658668678688698708718728738748758768778788798808818828838848858868878888898908918928938948958968978988999009019029039049059069079089099109119129139149159169179189199209219229239249259269279289299309319329339349359369379389399409419429439449459469479489499509519529539549559569579589599609619629639649659669679689699709719729739749759769779789799809819829839849859869879889899909919929939949959969979989991000100110021003100410051006100710081009101010111012101310141015101610171018101910201021102210231024102510261027102810291030103110321033103410351036103710381039104010411042104310441045104610471048104910501051105210531054105510561057105810591060106110621063106410651066106710681069107010711072107310741075107610771078107910801081108210831084108510861087108810891090109110921093109410951096109710981099110011011102110311041105110611071108110911101111111211131114111511161117111811191120112111221123112411251126112711281129113011311132113311341135113611371138113911401141114211431144114511461147114811491150115111521153115411551156115711581159116011611162116311641165116611671168116911701171117211731174117511761177117811791180118111821183118411851186118711881189119011911192119311941195119611971198119912001201120212031204120512061207120812091210121112121213121412151216121712181219122012211222122312241225122612271228122912301231123212331234123512361237123812391240124112421243124412451246124712481249125012511252125312541255125612571258125912601261126212631264126512661267126812691270127112721273127412751276127712781279128012811282128312841285128612871288128912901291129212931294129512961297129812991300130113021303130413051306130713081309131013111312131313141315131613171318131913201321132213231324132513261327132813291330133113321333133413351336133713381339134013411342134313441345134613471348134913501351135213531354135513561357135813591360136113621363136413651366136713681369137013711372137313741375137613771378137913801381138213831384138513861387138813891390139113921393139413951396139713981399140014011402140314041405140614071408140914101411141214131414141514161417141814191420142114221423142414251426142714281429143014311432143314341435143614371438143914401441144214431444144514461447144814491450145114521453145414551456145714581459146014611462146314641465146614671468146914701471147214731474147514761477147814791480148114821483148414851486148714881489149014911492149314941495149614971498149915001501150215031504150515061507150815091510151115121513151415151516151715181519152015211522152315241525152615271528152915301531153215331534153515361537153815391540154115421543154415451546154715481549155015511552155315541555155615571558155915601561156215631564156515661567156815691570157115721573157415751576157715781579158015811582158315841585158615871588158915901591159215931594159515961597159815991600160116021603160416051606160716081609161016111612161316141615161616171618161916201621162216231624162516261627162816291630163116321633163416351636163716381639164016411642164316441645164616471648164916501651165216531654165516561657165816591660166116621663166416651666166716681669167016711672167316741675167616771678167916801681168216831684168516861687168816891690169116921693169416951696169716981699170017011702170317041705170617071708170917101711171217131714171517161717171817191720172117221723172417251726172717281729173017311732173317341735173617371738173917401741174217431744174517461747174817491750175117521753175417551756175717581759176017611762176317641765176617671768176917701771177217731774177517761777177817791780178117821783178417851786178717881789179017911792179317941795179617971798179918001801180218031804180518061807180818091810181118121813181418151816181718181819182018211822182318241825182618271828182918301831183218331834183518361837183818391840184118421843184418451846184718481849185018511852185318541855185618571858185918601861186218631864186518661867186818691870187118721873187418751876187718781879188018811882188318841885188618871888188918901891189218931894189518961897189818991900190119021903190419051906190719081909191019111912191319141915191619171918191919201921192219231924192519261927192819291930193119321933193419351936193719381939194019411942194319441945194619471948194919501951195219531954195519561957195819591960196119621963196419651966196719681969197019711972197319741975197619771978197919801981198219831984198519861987198819891990199119921993199419951996199719981999200020012002200320042005200620072008200920102011201220132014201520162017201820192020202120222023202420252026202720282029203020312032203320342035203620372038203920402041204220432044204520462047204820492050205120522053205420552056205720582059206020612062206320642065206620672068206920702071207220732074207520762077207820792080208120822083208420852086208720882089209020912092209320942095209620972098209921002101210221032104210521062107210821092110211121122113211421152116211721182119212021212122212321242125212621272128212921302131213221332134213521362137213821392140214121422143214421452146214721482149215021512152215321542155215621572158215921602161216221632164216521662167216821692170217121722173217421752176217721782179218021812182218321842185218621872188218921902191219221932194219521962197219821992200220122022203220422052206220722082209221022112212221322142215221622172218221922202221222222232224222522262227222822292230223122322233223422352236223722382239224022412242224322442245224622472248224922502251225222532254225522562257225822592260226122622263226422652266226722682269227022712272227322742275227622772278227922802281228222832284228522862287228822892290229122922293229422952296229722982299230023012302230323042305230623072308230923102311231223132314231523162317231823192320232123222323232423252326232723282329233023312332233323342335233623372338233923402341234223432344234523462347234823492350235123522353235423552356235723582359236023612362236323642365236623672368236923702371237223732374237523762377237823792380238123822383238423852386238723882389239023912392239323942395239623972398239924002401240224032404240524062407240824092410241124122413241424152416241724182419242024212422242324242425242624272428242924302431243224332434243524362437243824392440244124422443244424452446244724482449245024512452245324542455245624572458245924602461246224632464246524662467246824692470247124722473247424752476247724782479248024812482248324842485248624872488248924902491249224932494249524962497249824992500250125022503250425052506250725082509251025112512251325142515251625172518251925202521252225232524252525262527252825292530253125322533253425352536253725382539254025412542254325442545254625472548254925502551255225532554255525562557255825592560256125622563256425652566256725682569257025712572257325742575257625772578257925802581258225832584258525862587258825892590259125922593259425952596259725982599260026012602260326042605260626072608260926102611261226132614261526162617261826192620262126222623262426252626262726282629263026312632263326342635263626372638263926402641264226432644264526462647264826492650265126522653265426552656265726582659266026612662266326642665266626672668266926702671267226732674267526762677267826792680268126822683268426852686268726882689269026912692269326942695269626972698269927002701270227032704270527062707270827092710271127122713271427152716271727182719272027212722272327242725272627272728272927302731273227332734273527362737273827392740274127422743274427452746274727482749275027512752275327542755275627572758275927602761276227632764276527662767276827692770277127722773277427752776277727782779278027812782278327842785278627872788278927902791279227932794279527962797279827992800280128022803280428052806280728082809281028112812281328142815281628172818281928202821282228232824282528262827282828292830283128322833283428352836283728382839284028412842284328442845284628472848284928502851285228532854285528562857285828592860286128622863286428652866286728682869287028712872287328742875287628772878287928802881288228832884288528862887288828892890289128922893289428952896289728982899290029012902290329042905290629072908290929102911291229132914291529162917291829192920292129222923292429252926292729282929293029312932293329342935293629372938293929402941294229432944294529462947294829492950295129522953295429552956295729582959296029612962296329642965296629672968296929702971297229732974297529762977297829792980298129822983298429852986298729882989299029912992299329942995299629972998299930003001300230033004300530063007300830093010301130123013301430153016301730183019302030213022302330243025302630273028302930303031303230333034303530363037303830393040304130423043304430453046304730483049305030513052305330543055305630573058305930603061306230633064306530663067306830693070307130723073307430753076307730783079308030813082308330843085308630873088308930903091309230933094309530963097309830993100310131023103310431053106310731083109311031113112311331143115311631173118311931203121312231233124312531263127312831293130313131323133313431353136313731383139314031413142314331443145314631473148314931503151315231533154315531563157315831593160316131623163316431653166316731683169317031713172317331743175317631773178317931803181318231833184318531863187318831893190319131923193319431953196319731983199320032013202320332043205320632073208320932103211321232133214321532163217321832193220322132223223322432253226322732283229323032313232323332343235323632373238323932403241324232433244324532463247324832493250325132523253325432553256325732583259326032613262326332643265326632673268326932703271327232733274327532763277327832793280328132823283328432853286328732883289329032913292329332943295329632973298329933003301330233033304330533063307330833093310331133123313331433153316331733183319332033213322332333243325332633273328332933303331333233333334333533363337333833393340334133423343334433453346334733483349335033513352335333543355335633573358335933603361336233633364336533663367336833693370337133723373337433753376337733783379338033813382338333843385338633873388338933903391339233933394339533963397339833993400340134023403340434053406340734083409341034113412341334143415341634173418341934203421342234233424342534263427342834293430343134323433343434353436343734383439344034413442344334443445344634473448344934503451345234533454345534563457345834593460346134623463346434653466346734683469347034713472347334743475347634773478347934803481348234833484348534863487348834893490349134923493349434953496349734983499350035013502350335043505350635073508350935103511351235133514351535163517351835193520352135223523352435253526352735283529353035313532353335343535353635373538353935403541354235433544354535463547354835493550355135523553355435553556355735583559356035613562356335643565356635673568356935703571357235733574357535763577357835793580358135823583358435853586358735883589359035913592359335943595359635973598359936003601360236033604360536063607360836093610361136123613361436153616361736183619362036213622362336243625362636273628362936303631363236333634363536363637363836393640364136423643364436453646364736483649365036513652365336543655365636573658365936603661366236633664366536663667366836693670367136723673367436753676367736783679368036813682368336843685368636873688368936903691369236933694369536963697369836993700370137023703370437053706370737083709371037113712371337143715371637173718371937203721372237233724372537263727372837293730373137323733373437353736373737383739374037413742374337443745374637473748374937503751375237533754375537563757375837593760376137623763376437653766376737683769377037713772377337743775377637773778377937803781378237833784378537863787378837893790379137923793379437953796379737983799380038013802380338043805380638073808380938103811381238133814381538163817381838193820382138223823382438253826382738283829383038313832383338343835383638373838383938403841384238433844384538463847384838493850385138523853385438553856385738583859386038613862386338643865386638673868386938703871387238733874387538763877387838793880388138823883388438853886388738883889389038913892389338943895389638973898389939003901390239033904390539063907390839093910391139123913391439153916391739183919392039213922392339243925392639273928392939303931393239333934393539363937393839393940394139423943394439453946394739483949395039513952395339543955395639573958395939603961396239633964396539663967396839693970397139723973397439753976397739783979398039813982398339843985398639873988398939903991399239933994399539963997399839994000400140024003400440054006400740084009401040114012401340144015401640174018401940204021402240234024402540264027402840294030403140324033403440354036403740384039404040414042404340444045404640474048404940504051405240534054405540564057405840594060406140624063406440654066406740684069407040714072407340744075407640774078407940804081408240834084408540864087408840894090409140924093409440954096409740984099410041014102410341044105410641074108410941104111411241134114411541164117411841194120412141224123412441254126412741284129413041314132413341344135413641374138413941404141414241434144414541464147414841494150415141524153415441554156415741584159416041614162416341644165416641674168416941704171417241734174417541764177417841794180418141824183418441854186418741884189419041914192419341944195419641974198419942004201420242034204420542064207420842094210421142124213421442154216421742184219422042214222422342244225422642274228422942304231423242334234423542364237423842394240424142424243424442454246424742484249425042514252425342544255425642574258425942604261426242634264426542664267426842694270427142724273427442754276427742784279428042814282428342844285428642874288428942904291429242934294429542964297429842994300430143024303430443054306430743084309431043114312431343144315431643174318431943204321432243234324432543264327432843294330433143324333433443354336433743384339434043414342434343444345434643474348434943504351435243534354435543564357435843594360436143624363436443654366436743684369437043714372437343744375437643774378437943804381438243834384438543864387438843894390439143924393439443954396439743984399440044014402440344044405440644074408440944104411441244134414441544164417441844194420442144224423442444254426442744284429443044314432443344344435443644374438443944404441444244434444444544464447444844494450445144524453445444554456445744584459446044614462446344644465446644674468446944704471447244734474447544764477447844794480448144824483448444854486448744884489449044914492449344944495449644974498449945004501450245034504450545064507450845094510451145124513451445154516451745184519452045214522452345244525452645274528452945304531453245334534453545364537453845394540454145424543454445454546454745484549455045514552455345544555455645574558455945604561456245634564456545664567456845694570457145724573457445754576457745784579458045814582458345844585458645874588458945904591459245934594459545964597459845994600460146024603460446054606460746084609461046114612461346144615461646174618461946204621462246234624462546264627462846294630463146324633463446354636463746384639464046414642464346444645464646474648464946504651465246534654465546564657465846594660466146624663466446654666466746684669467046714672467346744675467646774678467946804681468246834684468546864687468846894690469146924693469446954696469746984699470047014702470347044705470647074708470947104711471247134714471547164717471847194720472147224723472447254726472747284729473047314732473347344735473647374738473947404741474247434744474547464747
  1. /*
  2. * Copyright (c) 2000-2006 Silicon Graphics, Inc.
  3. * All Rights Reserved.
  4. *
  5. * This program is free software; you can redistribute it and/or
  6. * modify it under the terms of the GNU General Public License as
  7. * published by the Free Software Foundation.
  8. *
  9. * This program is distributed in the hope that it would be useful,
  10. * but WITHOUT ANY WARRANTY; without even the implied warranty of
  11. * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
  12. * GNU General Public License for more details.
  13. *
  14. * You should have received a copy of the GNU General Public License
  15. * along with this program; if not, write the Free Software Foundation,
  16. * Inc., 51 Franklin St, Fifth Floor, Boston, MA 02110-1301 USA
  17. */
  18. #include "xfs.h"
  19. #include "xfs_fs.h"
  20. #include "xfs_types.h"
  21. #include "xfs_bit.h"
  22. #include "xfs_log.h"
  23. #include "xfs_inum.h"
  24. #include "xfs_imap.h"
  25. #include "xfs_trans.h"
  26. #include "xfs_trans_priv.h"
  27. #include "xfs_sb.h"
  28. #include "xfs_ag.h"
  29. #include "xfs_dir.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_dir_sf.h"
  37. #include "xfs_dir2_sf.h"
  38. #include "xfs_attr_sf.h"
  39. #include "xfs_dinode.h"
  40. #include "xfs_inode.h"
  41. #include "xfs_buf_item.h"
  42. #include "xfs_inode_item.h"
  43. #include "xfs_btree.h"
  44. #include "xfs_alloc.h"
  45. #include "xfs_ialloc.h"
  46. #include "xfs_bmap.h"
  47. #include "xfs_rw.h"
  48. #include "xfs_error.h"
  49. #include "xfs_utils.h"
  50. #include "xfs_dir2_trace.h"
  51. #include "xfs_quota.h"
  52. #include "xfs_mac.h"
  53. #include "xfs_acl.h"
  54. kmem_zone_t *xfs_ifork_zone;
  55. kmem_zone_t *xfs_inode_zone;
  56. kmem_zone_t *xfs_chashlist_zone;
  57. /*
  58. * Used in xfs_itruncate(). This is the maximum number of extents
  59. * freed from a file in a single transaction.
  60. */
  61. #define XFS_ITRUNC_MAX_EXTENTS 2
  62. STATIC int xfs_iflush_int(xfs_inode_t *, xfs_buf_t *);
  63. STATIC int xfs_iformat_local(xfs_inode_t *, xfs_dinode_t *, int, int);
  64. STATIC int xfs_iformat_extents(xfs_inode_t *, xfs_dinode_t *, int);
  65. STATIC int xfs_iformat_btree(xfs_inode_t *, xfs_dinode_t *, int);
  66. #ifdef DEBUG
  67. /*
  68. * Make sure that the extents in the given memory buffer
  69. * are valid.
  70. */
  71. STATIC void
  72. xfs_validate_extents(
  73. xfs_ifork_t *ifp,
  74. int nrecs,
  75. int disk,
  76. xfs_exntfmt_t fmt)
  77. {
  78. xfs_bmbt_rec_t *ep;
  79. xfs_bmbt_irec_t irec;
  80. xfs_bmbt_rec_t rec;
  81. int i;
  82. for (i = 0; i < nrecs; i++) {
  83. ep = xfs_iext_get_ext(ifp, i);
  84. rec.l0 = get_unaligned((__uint64_t*)&ep->l0);
  85. rec.l1 = get_unaligned((__uint64_t*)&ep->l1);
  86. if (disk)
  87. xfs_bmbt_disk_get_all(&rec, &irec);
  88. else
  89. xfs_bmbt_get_all(&rec, &irec);
  90. if (fmt == XFS_EXTFMT_NOSTATE)
  91. ASSERT(irec.br_state == XFS_EXT_NORM);
  92. }
  93. }
  94. #else /* DEBUG */
  95. #define xfs_validate_extents(ifp, nrecs, disk, fmt)
  96. #endif /* DEBUG */
  97. /*
  98. * Check that none of the inode's in the buffer have a next
  99. * unlinked field of 0.
  100. */
  101. #if defined(DEBUG)
  102. void
  103. xfs_inobp_check(
  104. xfs_mount_t *mp,
  105. xfs_buf_t *bp)
  106. {
  107. int i;
  108. int j;
  109. xfs_dinode_t *dip;
  110. j = mp->m_inode_cluster_size >> mp->m_sb.sb_inodelog;
  111. for (i = 0; i < j; i++) {
  112. dip = (xfs_dinode_t *)xfs_buf_offset(bp,
  113. i * mp->m_sb.sb_inodesize);
  114. if (!dip->di_next_unlinked) {
  115. xfs_fs_cmn_err(CE_ALERT, mp,
  116. "Detected a bogus zero next_unlinked field in incore inode buffer 0x%p. About to pop an ASSERT.",
  117. bp);
  118. ASSERT(dip->di_next_unlinked);
  119. }
  120. }
  121. }
  122. #endif
  123. /*
  124. * This routine is called to map an inode number within a file
  125. * system to the buffer containing the on-disk version of the
  126. * inode. It returns a pointer to the buffer containing the
  127. * on-disk inode in the bpp parameter, and in the dip parameter
  128. * it returns a pointer to the on-disk inode within that buffer.
  129. *
  130. * If a non-zero error is returned, then the contents of bpp and
  131. * dipp are undefined.
  132. *
  133. * Use xfs_imap() to determine the size and location of the
  134. * buffer to read from disk.
  135. */
  136. STATIC int
  137. xfs_inotobp(
  138. xfs_mount_t *mp,
  139. xfs_trans_t *tp,
  140. xfs_ino_t ino,
  141. xfs_dinode_t **dipp,
  142. xfs_buf_t **bpp,
  143. int *offset)
  144. {
  145. int di_ok;
  146. xfs_imap_t imap;
  147. xfs_buf_t *bp;
  148. int error;
  149. xfs_dinode_t *dip;
  150. /*
  151. * Call the space management code to find the location of the
  152. * inode on disk.
  153. */
  154. imap.im_blkno = 0;
  155. error = xfs_imap(mp, tp, ino, &imap, XFS_IMAP_LOOKUP);
  156. if (error != 0) {
  157. cmn_err(CE_WARN,
  158. "xfs_inotobp: xfs_imap() returned an "
  159. "error %d on %s. Returning error.", error, mp->m_fsname);
  160. return error;
  161. }
  162. /*
  163. * If the inode number maps to a block outside the bounds of the
  164. * file system then return NULL rather than calling read_buf
  165. * and panicing when we get an error from the driver.
  166. */
  167. if ((imap.im_blkno + imap.im_len) >
  168. XFS_FSB_TO_BB(mp, mp->m_sb.sb_dblocks)) {
  169. cmn_err(CE_WARN,
  170. "xfs_inotobp: inode number (%llu + %d) maps to a block outside the bounds "
  171. "of the file system %s. Returning EINVAL.",
  172. (unsigned long long)imap.im_blkno,
  173. imap.im_len, mp->m_fsname);
  174. return XFS_ERROR(EINVAL);
  175. }
  176. /*
  177. * Read in the buffer. If tp is NULL, xfs_trans_read_buf() will
  178. * default to just a read_buf() call.
  179. */
  180. error = xfs_trans_read_buf(mp, tp, mp->m_ddev_targp, imap.im_blkno,
  181. (int)imap.im_len, XFS_BUF_LOCK, &bp);
  182. if (error) {
  183. cmn_err(CE_WARN,
  184. "xfs_inotobp: xfs_trans_read_buf() returned an "
  185. "error %d on %s. Returning error.", error, mp->m_fsname);
  186. return error;
  187. }
  188. dip = (xfs_dinode_t *)xfs_buf_offset(bp, 0);
  189. di_ok =
  190. INT_GET(dip->di_core.di_magic, ARCH_CONVERT) == XFS_DINODE_MAGIC &&
  191. XFS_DINODE_GOOD_VERSION(INT_GET(dip->di_core.di_version, ARCH_CONVERT));
  192. if (unlikely(XFS_TEST_ERROR(!di_ok, mp, XFS_ERRTAG_ITOBP_INOTOBP,
  193. XFS_RANDOM_ITOBP_INOTOBP))) {
  194. XFS_CORRUPTION_ERROR("xfs_inotobp", XFS_ERRLEVEL_LOW, mp, dip);
  195. xfs_trans_brelse(tp, bp);
  196. cmn_err(CE_WARN,
  197. "xfs_inotobp: XFS_TEST_ERROR() returned an "
  198. "error on %s. Returning EFSCORRUPTED.", mp->m_fsname);
  199. return XFS_ERROR(EFSCORRUPTED);
  200. }
  201. xfs_inobp_check(mp, bp);
  202. /*
  203. * Set *dipp to point to the on-disk inode in the buffer.
  204. */
  205. *dipp = (xfs_dinode_t *)xfs_buf_offset(bp, imap.im_boffset);
  206. *bpp = bp;
  207. *offset = imap.im_boffset;
  208. return 0;
  209. }
  210. /*
  211. * This routine is called to map an inode to the buffer containing
  212. * the on-disk version of the inode. It returns a pointer to the
  213. * buffer containing the on-disk inode in the bpp parameter, and in
  214. * the dip parameter it returns a pointer to the on-disk inode within
  215. * that buffer.
  216. *
  217. * If a non-zero error is returned, then the contents of bpp and
  218. * dipp are undefined.
  219. *
  220. * If the inode is new and has not yet been initialized, use xfs_imap()
  221. * to determine the size and location of the buffer to read from disk.
  222. * If the inode has already been mapped to its buffer and read in once,
  223. * then use the mapping information stored in the inode rather than
  224. * calling xfs_imap(). This allows us to avoid the overhead of looking
  225. * at the inode btree for small block file systems (see xfs_dilocate()).
  226. * We can tell whether the inode has been mapped in before by comparing
  227. * its disk block address to 0. Only uninitialized inodes will have
  228. * 0 for the disk block address.
  229. */
  230. int
  231. xfs_itobp(
  232. xfs_mount_t *mp,
  233. xfs_trans_t *tp,
  234. xfs_inode_t *ip,
  235. xfs_dinode_t **dipp,
  236. xfs_buf_t **bpp,
  237. xfs_daddr_t bno,
  238. uint imap_flags)
  239. {
  240. xfs_imap_t imap;
  241. xfs_buf_t *bp;
  242. int error;
  243. int i;
  244. int ni;
  245. if (ip->i_blkno == (xfs_daddr_t)0) {
  246. /*
  247. * Call the space management code to find the location of the
  248. * inode on disk.
  249. */
  250. imap.im_blkno = bno;
  251. if ((error = xfs_imap(mp, tp, ip->i_ino, &imap,
  252. XFS_IMAP_LOOKUP | imap_flags)))
  253. return error;
  254. /*
  255. * If the inode number maps to a block outside the bounds
  256. * of the file system then return NULL rather than calling
  257. * read_buf and panicing when we get an error from the
  258. * driver.
  259. */
  260. if ((imap.im_blkno + imap.im_len) >
  261. XFS_FSB_TO_BB(mp, mp->m_sb.sb_dblocks)) {
  262. #ifdef DEBUG
  263. xfs_fs_cmn_err(CE_ALERT, mp, "xfs_itobp: "
  264. "(imap.im_blkno (0x%llx) "
  265. "+ imap.im_len (0x%llx)) > "
  266. " XFS_FSB_TO_BB(mp, "
  267. "mp->m_sb.sb_dblocks) (0x%llx)",
  268. (unsigned long long) imap.im_blkno,
  269. (unsigned long long) imap.im_len,
  270. XFS_FSB_TO_BB(mp, mp->m_sb.sb_dblocks));
  271. #endif /* DEBUG */
  272. return XFS_ERROR(EINVAL);
  273. }
  274. /*
  275. * Fill in the fields in the inode that will be used to
  276. * map the inode to its buffer from now on.
  277. */
  278. ip->i_blkno = imap.im_blkno;
  279. ip->i_len = imap.im_len;
  280. ip->i_boffset = imap.im_boffset;
  281. } else {
  282. /*
  283. * We've already mapped the inode once, so just use the
  284. * mapping that we saved the first time.
  285. */
  286. imap.im_blkno = ip->i_blkno;
  287. imap.im_len = ip->i_len;
  288. imap.im_boffset = ip->i_boffset;
  289. }
  290. ASSERT(bno == 0 || bno == imap.im_blkno);
  291. /*
  292. * Read in the buffer. If tp is NULL, xfs_trans_read_buf() will
  293. * default to just a read_buf() call.
  294. */
  295. error = xfs_trans_read_buf(mp, tp, mp->m_ddev_targp, imap.im_blkno,
  296. (int)imap.im_len, XFS_BUF_LOCK, &bp);
  297. if (error) {
  298. #ifdef DEBUG
  299. xfs_fs_cmn_err(CE_ALERT, mp, "xfs_itobp: "
  300. "xfs_trans_read_buf() returned error %d, "
  301. "imap.im_blkno 0x%llx, imap.im_len 0x%llx",
  302. error, (unsigned long long) imap.im_blkno,
  303. (unsigned long long) imap.im_len);
  304. #endif /* DEBUG */
  305. return error;
  306. }
  307. /*
  308. * Validate the magic number and version of every inode in the buffer
  309. * (if DEBUG kernel) or the first inode in the buffer, otherwise.
  310. * No validation is done here in userspace (xfs_repair).
  311. */
  312. #if !defined(__KERNEL__)
  313. ni = 0;
  314. #elif defined(DEBUG)
  315. ni = (imap_flags & XFS_IMAP_BULKSTAT) ? 0 :
  316. (BBTOB(imap.im_len) >> mp->m_sb.sb_inodelog);
  317. #else /* usual case */
  318. ni = (imap_flags & XFS_IMAP_BULKSTAT) ? 0 : 1;
  319. #endif
  320. for (i = 0; i < ni; i++) {
  321. int di_ok;
  322. xfs_dinode_t *dip;
  323. dip = (xfs_dinode_t *)xfs_buf_offset(bp,
  324. (i << mp->m_sb.sb_inodelog));
  325. di_ok = INT_GET(dip->di_core.di_magic, ARCH_CONVERT) == XFS_DINODE_MAGIC &&
  326. XFS_DINODE_GOOD_VERSION(INT_GET(dip->di_core.di_version, ARCH_CONVERT));
  327. if (unlikely(XFS_TEST_ERROR(!di_ok, mp, XFS_ERRTAG_ITOBP_INOTOBP,
  328. XFS_RANDOM_ITOBP_INOTOBP))) {
  329. #ifdef DEBUG
  330. if (!(imap_flags & XFS_IMAP_BULKSTAT))
  331. cmn_err(CE_ALERT,
  332. "Device %s - bad inode magic/vsn "
  333. "daddr %lld #%d (magic=%x)",
  334. XFS_BUFTARG_NAME(mp->m_ddev_targp),
  335. (unsigned long long)imap.im_blkno, i,
  336. INT_GET(dip->di_core.di_magic, ARCH_CONVERT));
  337. #endif
  338. XFS_CORRUPTION_ERROR("xfs_itobp", XFS_ERRLEVEL_HIGH,
  339. mp, dip);
  340. xfs_trans_brelse(tp, bp);
  341. return XFS_ERROR(EFSCORRUPTED);
  342. }
  343. }
  344. xfs_inobp_check(mp, bp);
  345. /*
  346. * Mark the buffer as an inode buffer now that it looks good
  347. */
  348. XFS_BUF_SET_VTYPE(bp, B_FS_INO);
  349. /*
  350. * Set *dipp to point to the on-disk inode in the buffer.
  351. */
  352. *dipp = (xfs_dinode_t *)xfs_buf_offset(bp, imap.im_boffset);
  353. *bpp = bp;
  354. return 0;
  355. }
  356. /*
  357. * Move inode type and inode format specific information from the
  358. * on-disk inode to the in-core inode. For fifos, devs, and sockets
  359. * this means set if_rdev to the proper value. For files, directories,
  360. * and symlinks this means to bring in the in-line data or extent
  361. * pointers. For a file in B-tree format, only the root is immediately
  362. * brought in-core. The rest will be in-lined in if_extents when it
  363. * is first referenced (see xfs_iread_extents()).
  364. */
  365. STATIC int
  366. xfs_iformat(
  367. xfs_inode_t *ip,
  368. xfs_dinode_t *dip)
  369. {
  370. xfs_attr_shortform_t *atp;
  371. int size;
  372. int error;
  373. xfs_fsize_t di_size;
  374. ip->i_df.if_ext_max =
  375. XFS_IFORK_DSIZE(ip) / (uint)sizeof(xfs_bmbt_rec_t);
  376. error = 0;
  377. if (unlikely(
  378. INT_GET(dip->di_core.di_nextents, ARCH_CONVERT) +
  379. INT_GET(dip->di_core.di_anextents, ARCH_CONVERT) >
  380. INT_GET(dip->di_core.di_nblocks, ARCH_CONVERT))) {
  381. xfs_fs_repair_cmn_err(CE_WARN, ip->i_mount,
  382. "corrupt dinode %Lu, extent total = %d, nblocks = %Lu.",
  383. (unsigned long long)ip->i_ino,
  384. (int)(INT_GET(dip->di_core.di_nextents, ARCH_CONVERT)
  385. + INT_GET(dip->di_core.di_anextents, ARCH_CONVERT)),
  386. (unsigned long long)
  387. INT_GET(dip->di_core.di_nblocks, ARCH_CONVERT));
  388. XFS_CORRUPTION_ERROR("xfs_iformat(1)", XFS_ERRLEVEL_LOW,
  389. ip->i_mount, dip);
  390. return XFS_ERROR(EFSCORRUPTED);
  391. }
  392. if (unlikely(INT_GET(dip->di_core.di_forkoff, ARCH_CONVERT) > ip->i_mount->m_sb.sb_inodesize)) {
  393. xfs_fs_repair_cmn_err(CE_WARN, ip->i_mount,
  394. "corrupt dinode %Lu, forkoff = 0x%x.",
  395. (unsigned long long)ip->i_ino,
  396. (int)(INT_GET(dip->di_core.di_forkoff, ARCH_CONVERT)));
  397. XFS_CORRUPTION_ERROR("xfs_iformat(2)", XFS_ERRLEVEL_LOW,
  398. ip->i_mount, dip);
  399. return XFS_ERROR(EFSCORRUPTED);
  400. }
  401. switch (ip->i_d.di_mode & S_IFMT) {
  402. case S_IFIFO:
  403. case S_IFCHR:
  404. case S_IFBLK:
  405. case S_IFSOCK:
  406. if (unlikely(INT_GET(dip->di_core.di_format, ARCH_CONVERT) != XFS_DINODE_FMT_DEV)) {
  407. XFS_CORRUPTION_ERROR("xfs_iformat(3)", XFS_ERRLEVEL_LOW,
  408. ip->i_mount, dip);
  409. return XFS_ERROR(EFSCORRUPTED);
  410. }
  411. ip->i_d.di_size = 0;
  412. ip->i_df.if_u2.if_rdev = INT_GET(dip->di_u.di_dev, ARCH_CONVERT);
  413. break;
  414. case S_IFREG:
  415. case S_IFLNK:
  416. case S_IFDIR:
  417. switch (INT_GET(dip->di_core.di_format, ARCH_CONVERT)) {
  418. case XFS_DINODE_FMT_LOCAL:
  419. /*
  420. * no local regular files yet
  421. */
  422. if (unlikely((INT_GET(dip->di_core.di_mode, ARCH_CONVERT) & S_IFMT) == S_IFREG)) {
  423. xfs_fs_repair_cmn_err(CE_WARN, ip->i_mount,
  424. "corrupt inode %Lu "
  425. "(local format for regular file).",
  426. (unsigned long long) ip->i_ino);
  427. XFS_CORRUPTION_ERROR("xfs_iformat(4)",
  428. XFS_ERRLEVEL_LOW,
  429. ip->i_mount, dip);
  430. return XFS_ERROR(EFSCORRUPTED);
  431. }
  432. di_size = INT_GET(dip->di_core.di_size, ARCH_CONVERT);
  433. if (unlikely(di_size > XFS_DFORK_DSIZE(dip, ip->i_mount))) {
  434. xfs_fs_repair_cmn_err(CE_WARN, ip->i_mount,
  435. "corrupt inode %Lu "
  436. "(bad size %Ld for local inode).",
  437. (unsigned long long) ip->i_ino,
  438. (long long) di_size);
  439. XFS_CORRUPTION_ERROR("xfs_iformat(5)",
  440. XFS_ERRLEVEL_LOW,
  441. ip->i_mount, dip);
  442. return XFS_ERROR(EFSCORRUPTED);
  443. }
  444. size = (int)di_size;
  445. error = xfs_iformat_local(ip, dip, XFS_DATA_FORK, size);
  446. break;
  447. case XFS_DINODE_FMT_EXTENTS:
  448. error = xfs_iformat_extents(ip, dip, XFS_DATA_FORK);
  449. break;
  450. case XFS_DINODE_FMT_BTREE:
  451. error = xfs_iformat_btree(ip, dip, XFS_DATA_FORK);
  452. break;
  453. default:
  454. XFS_ERROR_REPORT("xfs_iformat(6)", XFS_ERRLEVEL_LOW,
  455. ip->i_mount);
  456. return XFS_ERROR(EFSCORRUPTED);
  457. }
  458. break;
  459. default:
  460. XFS_ERROR_REPORT("xfs_iformat(7)", XFS_ERRLEVEL_LOW, ip->i_mount);
  461. return XFS_ERROR(EFSCORRUPTED);
  462. }
  463. if (error) {
  464. return error;
  465. }
  466. if (!XFS_DFORK_Q(dip))
  467. return 0;
  468. ASSERT(ip->i_afp == NULL);
  469. ip->i_afp = kmem_zone_zalloc(xfs_ifork_zone, KM_SLEEP);
  470. ip->i_afp->if_ext_max =
  471. XFS_IFORK_ASIZE(ip) / (uint)sizeof(xfs_bmbt_rec_t);
  472. switch (INT_GET(dip->di_core.di_aformat, ARCH_CONVERT)) {
  473. case XFS_DINODE_FMT_LOCAL:
  474. atp = (xfs_attr_shortform_t *)XFS_DFORK_APTR(dip);
  475. size = be16_to_cpu(atp->hdr.totsize);
  476. error = xfs_iformat_local(ip, dip, XFS_ATTR_FORK, size);
  477. break;
  478. case XFS_DINODE_FMT_EXTENTS:
  479. error = xfs_iformat_extents(ip, dip, XFS_ATTR_FORK);
  480. break;
  481. case XFS_DINODE_FMT_BTREE:
  482. error = xfs_iformat_btree(ip, dip, XFS_ATTR_FORK);
  483. break;
  484. default:
  485. error = XFS_ERROR(EFSCORRUPTED);
  486. break;
  487. }
  488. if (error) {
  489. kmem_zone_free(xfs_ifork_zone, ip->i_afp);
  490. ip->i_afp = NULL;
  491. xfs_idestroy_fork(ip, XFS_DATA_FORK);
  492. }
  493. return error;
  494. }
  495. /*
  496. * The file is in-lined in the on-disk inode.
  497. * If it fits into if_inline_data, then copy
  498. * it there, otherwise allocate a buffer for it
  499. * and copy the data there. Either way, set
  500. * if_data to point at the data.
  501. * If we allocate a buffer for the data, make
  502. * sure that its size is a multiple of 4 and
  503. * record the real size in i_real_bytes.
  504. */
  505. STATIC int
  506. xfs_iformat_local(
  507. xfs_inode_t *ip,
  508. xfs_dinode_t *dip,
  509. int whichfork,
  510. int size)
  511. {
  512. xfs_ifork_t *ifp;
  513. int real_size;
  514. /*
  515. * If the size 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 > XFS_DFORK_SIZE(dip, ip->i_mount, whichfork))) {
  520. xfs_fs_repair_cmn_err(CE_WARN, ip->i_mount,
  521. "corrupt inode %Lu "
  522. "(bad size %d for local fork, size = %d).",
  523. (unsigned long long) ip->i_ino, size,
  524. XFS_DFORK_SIZE(dip, ip->i_mount, whichfork));
  525. XFS_CORRUPTION_ERROR("xfs_iformat_local", XFS_ERRLEVEL_LOW,
  526. ip->i_mount, dip);
  527. return XFS_ERROR(EFSCORRUPTED);
  528. }
  529. ifp = XFS_IFORK_PTR(ip, whichfork);
  530. real_size = 0;
  531. if (size == 0)
  532. ifp->if_u1.if_data = NULL;
  533. else if (size <= sizeof(ifp->if_u2.if_inline_data))
  534. ifp->if_u1.if_data = ifp->if_u2.if_inline_data;
  535. else {
  536. real_size = roundup(size, 4);
  537. ifp->if_u1.if_data = kmem_alloc(real_size, KM_SLEEP);
  538. }
  539. ifp->if_bytes = size;
  540. ifp->if_real_bytes = real_size;
  541. if (size)
  542. memcpy(ifp->if_u1.if_data, XFS_DFORK_PTR(dip, whichfork), size);
  543. ifp->if_flags &= ~XFS_IFEXTENTS;
  544. ifp->if_flags |= XFS_IFINLINE;
  545. return 0;
  546. }
  547. /*
  548. * The file consists of a set of extents all
  549. * of which fit into the on-disk inode.
  550. * If there are few enough extents to fit into
  551. * the if_inline_ext, then copy them there.
  552. * Otherwise allocate a buffer for them and copy
  553. * them into it. Either way, set if_extents
  554. * to point at the extents.
  555. */
  556. STATIC int
  557. xfs_iformat_extents(
  558. xfs_inode_t *ip,
  559. xfs_dinode_t *dip,
  560. int whichfork)
  561. {
  562. xfs_bmbt_rec_t *ep, *dp;
  563. xfs_ifork_t *ifp;
  564. int nex;
  565. int size;
  566. int i;
  567. ifp = XFS_IFORK_PTR(ip, whichfork);
  568. nex = XFS_DFORK_NEXTENTS(dip, whichfork);
  569. size = nex * (uint)sizeof(xfs_bmbt_rec_t);
  570. /*
  571. * If the number of extents is unreasonable, then something
  572. * is wrong and we just bail out rather than crash in
  573. * kmem_alloc() or memcpy() below.
  574. */
  575. if (unlikely(size < 0 || size > XFS_DFORK_SIZE(dip, ip->i_mount, whichfork))) {
  576. xfs_fs_repair_cmn_err(CE_WARN, ip->i_mount,
  577. "corrupt inode %Lu ((a)extents = %d).",
  578. (unsigned long long) ip->i_ino, nex);
  579. XFS_CORRUPTION_ERROR("xfs_iformat_extents(1)", XFS_ERRLEVEL_LOW,
  580. ip->i_mount, dip);
  581. return XFS_ERROR(EFSCORRUPTED);
  582. }
  583. ifp->if_real_bytes = 0;
  584. if (nex == 0)
  585. ifp->if_u1.if_extents = NULL;
  586. else if (nex <= XFS_INLINE_EXTS)
  587. ifp->if_u1.if_extents = ifp->if_u2.if_inline_ext;
  588. else
  589. xfs_iext_add(ifp, 0, nex);
  590. ifp->if_bytes = size;
  591. if (size) {
  592. dp = (xfs_bmbt_rec_t *) XFS_DFORK_PTR(dip, whichfork);
  593. xfs_validate_extents(ifp, nex, 1, XFS_EXTFMT_INODE(ip));
  594. for (i = 0; i < nex; i++, dp++) {
  595. ep = xfs_iext_get_ext(ifp, i);
  596. ep->l0 = INT_GET(get_unaligned((__uint64_t*)&dp->l0),
  597. ARCH_CONVERT);
  598. ep->l1 = INT_GET(get_unaligned((__uint64_t*)&dp->l1),
  599. ARCH_CONVERT);
  600. }
  601. xfs_bmap_trace_exlist("xfs_iformat_extents", ip, nex,
  602. whichfork);
  603. if (whichfork != XFS_DATA_FORK ||
  604. XFS_EXTFMT_INODE(ip) == XFS_EXTFMT_NOSTATE)
  605. if (unlikely(xfs_check_nostate_extents(
  606. ifp, 0, nex))) {
  607. XFS_ERROR_REPORT("xfs_iformat_extents(2)",
  608. XFS_ERRLEVEL_LOW,
  609. ip->i_mount);
  610. return XFS_ERROR(EFSCORRUPTED);
  611. }
  612. }
  613. ifp->if_flags |= XFS_IFEXTENTS;
  614. return 0;
  615. }
  616. /*
  617. * The file has too many extents to fit into
  618. * the inode, so they are in B-tree format.
  619. * Allocate a buffer for the root of the B-tree
  620. * and copy the root into it. The i_extents
  621. * field will remain NULL until all of the
  622. * extents are read in (when they are needed).
  623. */
  624. STATIC int
  625. xfs_iformat_btree(
  626. xfs_inode_t *ip,
  627. xfs_dinode_t *dip,
  628. int whichfork)
  629. {
  630. xfs_bmdr_block_t *dfp;
  631. xfs_ifork_t *ifp;
  632. /* REFERENCED */
  633. int nrecs;
  634. int size;
  635. ifp = XFS_IFORK_PTR(ip, whichfork);
  636. dfp = (xfs_bmdr_block_t *)XFS_DFORK_PTR(dip, whichfork);
  637. size = XFS_BMAP_BROOT_SPACE(dfp);
  638. nrecs = XFS_BMAP_BROOT_NUMRECS(dfp);
  639. /*
  640. * blow out if -- fork has less extents than can fit in
  641. * fork (fork shouldn't be a btree format), root btree
  642. * block has more records than can fit into the fork,
  643. * or the number of extents is greater than the number of
  644. * blocks.
  645. */
  646. if (unlikely(XFS_IFORK_NEXTENTS(ip, whichfork) <= ifp->if_ext_max
  647. || XFS_BMDR_SPACE_CALC(nrecs) >
  648. XFS_DFORK_SIZE(dip, ip->i_mount, whichfork)
  649. || XFS_IFORK_NEXTENTS(ip, whichfork) > ip->i_d.di_nblocks)) {
  650. xfs_fs_repair_cmn_err(CE_WARN, ip->i_mount,
  651. "corrupt inode %Lu (btree).",
  652. (unsigned long long) ip->i_ino);
  653. XFS_ERROR_REPORT("xfs_iformat_btree", XFS_ERRLEVEL_LOW,
  654. ip->i_mount);
  655. return XFS_ERROR(EFSCORRUPTED);
  656. }
  657. ifp->if_broot_bytes = size;
  658. ifp->if_broot = kmem_alloc(size, KM_SLEEP);
  659. ASSERT(ifp->if_broot != NULL);
  660. /*
  661. * Copy and convert from the on-disk structure
  662. * to the in-memory structure.
  663. */
  664. xfs_bmdr_to_bmbt(dfp, XFS_DFORK_SIZE(dip, ip->i_mount, whichfork),
  665. ifp->if_broot, size);
  666. ifp->if_flags &= ~XFS_IFEXTENTS;
  667. ifp->if_flags |= XFS_IFBROOT;
  668. return 0;
  669. }
  670. /*
  671. * xfs_xlate_dinode_core - translate an xfs_inode_core_t between ondisk
  672. * and native format
  673. *
  674. * buf = on-disk representation
  675. * dip = native representation
  676. * dir = direction - +ve -> disk to native
  677. * -ve -> native to disk
  678. */
  679. void
  680. xfs_xlate_dinode_core(
  681. xfs_caddr_t buf,
  682. xfs_dinode_core_t *dip,
  683. int dir)
  684. {
  685. xfs_dinode_core_t *buf_core = (xfs_dinode_core_t *)buf;
  686. xfs_dinode_core_t *mem_core = (xfs_dinode_core_t *)dip;
  687. xfs_arch_t arch = ARCH_CONVERT;
  688. ASSERT(dir);
  689. INT_XLATE(buf_core->di_magic, mem_core->di_magic, dir, arch);
  690. INT_XLATE(buf_core->di_mode, mem_core->di_mode, dir, arch);
  691. INT_XLATE(buf_core->di_version, mem_core->di_version, dir, arch);
  692. INT_XLATE(buf_core->di_format, mem_core->di_format, dir, arch);
  693. INT_XLATE(buf_core->di_onlink, mem_core->di_onlink, dir, arch);
  694. INT_XLATE(buf_core->di_uid, mem_core->di_uid, dir, arch);
  695. INT_XLATE(buf_core->di_gid, mem_core->di_gid, dir, arch);
  696. INT_XLATE(buf_core->di_nlink, mem_core->di_nlink, dir, arch);
  697. INT_XLATE(buf_core->di_projid, mem_core->di_projid, dir, arch);
  698. if (dir > 0) {
  699. memcpy(mem_core->di_pad, buf_core->di_pad,
  700. sizeof(buf_core->di_pad));
  701. } else {
  702. memcpy(buf_core->di_pad, mem_core->di_pad,
  703. sizeof(buf_core->di_pad));
  704. }
  705. INT_XLATE(buf_core->di_flushiter, mem_core->di_flushiter, dir, arch);
  706. INT_XLATE(buf_core->di_atime.t_sec, mem_core->di_atime.t_sec,
  707. dir, arch);
  708. INT_XLATE(buf_core->di_atime.t_nsec, mem_core->di_atime.t_nsec,
  709. dir, arch);
  710. INT_XLATE(buf_core->di_mtime.t_sec, mem_core->di_mtime.t_sec,
  711. dir, arch);
  712. INT_XLATE(buf_core->di_mtime.t_nsec, mem_core->di_mtime.t_nsec,
  713. dir, arch);
  714. INT_XLATE(buf_core->di_ctime.t_sec, mem_core->di_ctime.t_sec,
  715. dir, arch);
  716. INT_XLATE(buf_core->di_ctime.t_nsec, mem_core->di_ctime.t_nsec,
  717. dir, arch);
  718. INT_XLATE(buf_core->di_size, mem_core->di_size, dir, arch);
  719. INT_XLATE(buf_core->di_nblocks, mem_core->di_nblocks, dir, arch);
  720. INT_XLATE(buf_core->di_extsize, mem_core->di_extsize, dir, arch);
  721. INT_XLATE(buf_core->di_nextents, mem_core->di_nextents, dir, arch);
  722. INT_XLATE(buf_core->di_anextents, mem_core->di_anextents, dir, arch);
  723. INT_XLATE(buf_core->di_forkoff, mem_core->di_forkoff, dir, arch);
  724. INT_XLATE(buf_core->di_aformat, mem_core->di_aformat, dir, arch);
  725. INT_XLATE(buf_core->di_dmevmask, mem_core->di_dmevmask, dir, arch);
  726. INT_XLATE(buf_core->di_dmstate, mem_core->di_dmstate, dir, arch);
  727. INT_XLATE(buf_core->di_flags, mem_core->di_flags, dir, arch);
  728. INT_XLATE(buf_core->di_gen, mem_core->di_gen, dir, arch);
  729. }
  730. STATIC uint
  731. _xfs_dic2xflags(
  732. __uint16_t di_flags)
  733. {
  734. uint flags = 0;
  735. if (di_flags & XFS_DIFLAG_ANY) {
  736. if (di_flags & XFS_DIFLAG_REALTIME)
  737. flags |= XFS_XFLAG_REALTIME;
  738. if (di_flags & XFS_DIFLAG_PREALLOC)
  739. flags |= XFS_XFLAG_PREALLOC;
  740. if (di_flags & XFS_DIFLAG_IMMUTABLE)
  741. flags |= XFS_XFLAG_IMMUTABLE;
  742. if (di_flags & XFS_DIFLAG_APPEND)
  743. flags |= XFS_XFLAG_APPEND;
  744. if (di_flags & XFS_DIFLAG_SYNC)
  745. flags |= XFS_XFLAG_SYNC;
  746. if (di_flags & XFS_DIFLAG_NOATIME)
  747. flags |= XFS_XFLAG_NOATIME;
  748. if (di_flags & XFS_DIFLAG_NODUMP)
  749. flags |= XFS_XFLAG_NODUMP;
  750. if (di_flags & XFS_DIFLAG_RTINHERIT)
  751. flags |= XFS_XFLAG_RTINHERIT;
  752. if (di_flags & XFS_DIFLAG_PROJINHERIT)
  753. flags |= XFS_XFLAG_PROJINHERIT;
  754. if (di_flags & XFS_DIFLAG_NOSYMLINKS)
  755. flags |= XFS_XFLAG_NOSYMLINKS;
  756. if (di_flags & XFS_DIFLAG_EXTSIZE)
  757. flags |= XFS_XFLAG_EXTSIZE;
  758. if (di_flags & XFS_DIFLAG_EXTSZINHERIT)
  759. flags |= XFS_XFLAG_EXTSZINHERIT;
  760. if (di_flags & XFS_DIFLAG_NODEFRAG)
  761. flags |= XFS_XFLAG_NODEFRAG;
  762. }
  763. return flags;
  764. }
  765. uint
  766. xfs_ip2xflags(
  767. xfs_inode_t *ip)
  768. {
  769. xfs_dinode_core_t *dic = &ip->i_d;
  770. return _xfs_dic2xflags(dic->di_flags) |
  771. (XFS_CFORK_Q(dic) ? XFS_XFLAG_HASATTR : 0);
  772. }
  773. uint
  774. xfs_dic2xflags(
  775. xfs_dinode_core_t *dic)
  776. {
  777. return _xfs_dic2xflags(INT_GET(dic->di_flags, ARCH_CONVERT)) |
  778. (XFS_CFORK_Q_DISK(dic) ? XFS_XFLAG_HASATTR : 0);
  779. }
  780. /*
  781. * Given a mount structure and an inode number, return a pointer
  782. * to a newly allocated in-core inode corresponding to the given
  783. * inode number.
  784. *
  785. * Initialize the inode's attributes and extent pointers if it
  786. * already has them (it will not if the inode has no links).
  787. */
  788. int
  789. xfs_iread(
  790. xfs_mount_t *mp,
  791. xfs_trans_t *tp,
  792. xfs_ino_t ino,
  793. xfs_inode_t **ipp,
  794. xfs_daddr_t bno)
  795. {
  796. xfs_buf_t *bp;
  797. xfs_dinode_t *dip;
  798. xfs_inode_t *ip;
  799. int error;
  800. ASSERT(xfs_inode_zone != NULL);
  801. ip = kmem_zone_zalloc(xfs_inode_zone, KM_SLEEP);
  802. ip->i_ino = ino;
  803. ip->i_mount = mp;
  804. /*
  805. * Get pointer's to the on-disk inode and the buffer containing it.
  806. * If the inode number refers to a block outside the file system
  807. * then xfs_itobp() will return NULL. In this case we should
  808. * return NULL as well. Set i_blkno to 0 so that xfs_itobp() will
  809. * know that this is a new incore inode.
  810. */
  811. error = xfs_itobp(mp, tp, ip, &dip, &bp, bno, 0);
  812. if (error) {
  813. kmem_zone_free(xfs_inode_zone, ip);
  814. return error;
  815. }
  816. /*
  817. * Initialize inode's trace buffers.
  818. * Do this before xfs_iformat in case it adds entries.
  819. */
  820. #ifdef XFS_BMAP_TRACE
  821. ip->i_xtrace = ktrace_alloc(XFS_BMAP_KTRACE_SIZE, KM_SLEEP);
  822. #endif
  823. #ifdef XFS_BMBT_TRACE
  824. ip->i_btrace = ktrace_alloc(XFS_BMBT_KTRACE_SIZE, KM_SLEEP);
  825. #endif
  826. #ifdef XFS_RW_TRACE
  827. ip->i_rwtrace = ktrace_alloc(XFS_RW_KTRACE_SIZE, KM_SLEEP);
  828. #endif
  829. #ifdef XFS_ILOCK_TRACE
  830. ip->i_lock_trace = ktrace_alloc(XFS_ILOCK_KTRACE_SIZE, KM_SLEEP);
  831. #endif
  832. #ifdef XFS_DIR2_TRACE
  833. ip->i_dir_trace = ktrace_alloc(XFS_DIR2_KTRACE_SIZE, KM_SLEEP);
  834. #endif
  835. /*
  836. * If we got something that isn't an inode it means someone
  837. * (nfs or dmi) has a stale handle.
  838. */
  839. if (INT_GET(dip->di_core.di_magic, ARCH_CONVERT) != XFS_DINODE_MAGIC) {
  840. kmem_zone_free(xfs_inode_zone, ip);
  841. xfs_trans_brelse(tp, bp);
  842. #ifdef DEBUG
  843. xfs_fs_cmn_err(CE_ALERT, mp, "xfs_iread: "
  844. "dip->di_core.di_magic (0x%x) != "
  845. "XFS_DINODE_MAGIC (0x%x)",
  846. INT_GET(dip->di_core.di_magic, ARCH_CONVERT),
  847. XFS_DINODE_MAGIC);
  848. #endif /* DEBUG */
  849. return XFS_ERROR(EINVAL);
  850. }
  851. /*
  852. * If the on-disk inode is already linked to a directory
  853. * entry, copy all of the inode into the in-core inode.
  854. * xfs_iformat() handles copying in the inode format
  855. * specific information.
  856. * Otherwise, just get the truly permanent information.
  857. */
  858. if (dip->di_core.di_mode) {
  859. xfs_xlate_dinode_core((xfs_caddr_t)&dip->di_core,
  860. &(ip->i_d), 1);
  861. error = xfs_iformat(ip, dip);
  862. if (error) {
  863. kmem_zone_free(xfs_inode_zone, ip);
  864. xfs_trans_brelse(tp, bp);
  865. #ifdef DEBUG
  866. xfs_fs_cmn_err(CE_ALERT, mp, "xfs_iread: "
  867. "xfs_iformat() returned error %d",
  868. error);
  869. #endif /* DEBUG */
  870. return error;
  871. }
  872. } else {
  873. ip->i_d.di_magic = INT_GET(dip->di_core.di_magic, ARCH_CONVERT);
  874. ip->i_d.di_version = INT_GET(dip->di_core.di_version, ARCH_CONVERT);
  875. ip->i_d.di_gen = INT_GET(dip->di_core.di_gen, ARCH_CONVERT);
  876. ip->i_d.di_flushiter = INT_GET(dip->di_core.di_flushiter, ARCH_CONVERT);
  877. /*
  878. * Make sure to pull in the mode here as well in
  879. * case the inode is released without being used.
  880. * This ensures that xfs_inactive() will see that
  881. * the inode is already free and not try to mess
  882. * with the uninitialized part of it.
  883. */
  884. ip->i_d.di_mode = 0;
  885. /*
  886. * Initialize the per-fork minima and maxima for a new
  887. * inode here. xfs_iformat will do it for old inodes.
  888. */
  889. ip->i_df.if_ext_max =
  890. XFS_IFORK_DSIZE(ip) / (uint)sizeof(xfs_bmbt_rec_t);
  891. }
  892. INIT_LIST_HEAD(&ip->i_reclaim);
  893. /*
  894. * The inode format changed when we moved the link count and
  895. * made it 32 bits long. If this is an old format inode,
  896. * convert it in memory to look like a new one. If it gets
  897. * flushed to disk we will convert back before flushing or
  898. * logging it. We zero out the new projid field and the old link
  899. * count field. We'll handle clearing the pad field (the remains
  900. * of the old uuid field) when we actually convert the inode to
  901. * the new format. We don't change the version number so that we
  902. * can distinguish this from a real new format inode.
  903. */
  904. if (ip->i_d.di_version == XFS_DINODE_VERSION_1) {
  905. ip->i_d.di_nlink = ip->i_d.di_onlink;
  906. ip->i_d.di_onlink = 0;
  907. ip->i_d.di_projid = 0;
  908. }
  909. ip->i_delayed_blks = 0;
  910. /*
  911. * Mark the buffer containing the inode as something to keep
  912. * around for a while. This helps to keep recently accessed
  913. * meta-data in-core longer.
  914. */
  915. XFS_BUF_SET_REF(bp, XFS_INO_REF);
  916. /*
  917. * Use xfs_trans_brelse() to release the buffer containing the
  918. * on-disk inode, because it was acquired with xfs_trans_read_buf()
  919. * in xfs_itobp() above. If tp is NULL, this is just a normal
  920. * brelse(). If we're within a transaction, then xfs_trans_brelse()
  921. * will only release the buffer if it is not dirty within the
  922. * transaction. It will be OK to release the buffer in this case,
  923. * because inodes on disk are never destroyed and we will be
  924. * locking the new in-core inode before putting it in the hash
  925. * table where other processes can find it. Thus we don't have
  926. * to worry about the inode being changed just because we released
  927. * the buffer.
  928. */
  929. xfs_trans_brelse(tp, bp);
  930. *ipp = ip;
  931. return 0;
  932. }
  933. /*
  934. * Read in extents from a btree-format inode.
  935. * Allocate and fill in if_extents. Real work is done in xfs_bmap.c.
  936. */
  937. int
  938. xfs_iread_extents(
  939. xfs_trans_t *tp,
  940. xfs_inode_t *ip,
  941. int whichfork)
  942. {
  943. int error;
  944. xfs_ifork_t *ifp;
  945. xfs_extnum_t nextents;
  946. size_t size;
  947. if (unlikely(XFS_IFORK_FORMAT(ip, whichfork) != XFS_DINODE_FMT_BTREE)) {
  948. XFS_ERROR_REPORT("xfs_iread_extents", XFS_ERRLEVEL_LOW,
  949. ip->i_mount);
  950. return XFS_ERROR(EFSCORRUPTED);
  951. }
  952. nextents = XFS_IFORK_NEXTENTS(ip, whichfork);
  953. size = nextents * sizeof(xfs_bmbt_rec_t);
  954. ifp = XFS_IFORK_PTR(ip, whichfork);
  955. /*
  956. * We know that the size is valid (it's checked in iformat_btree)
  957. */
  958. ifp->if_lastex = NULLEXTNUM;
  959. ifp->if_bytes = ifp->if_real_bytes = 0;
  960. ifp->if_flags |= XFS_IFEXTENTS;
  961. xfs_iext_add(ifp, 0, nextents);
  962. error = xfs_bmap_read_extents(tp, ip, whichfork);
  963. if (error) {
  964. xfs_iext_destroy(ifp);
  965. ifp->if_flags &= ~XFS_IFEXTENTS;
  966. return error;
  967. }
  968. xfs_validate_extents(ifp, nextents, 0, XFS_EXTFMT_INODE(ip));
  969. return 0;
  970. }
  971. /*
  972. * Allocate an inode on disk and return a copy of its in-core version.
  973. * The in-core inode is locked exclusively. Set mode, nlink, and rdev
  974. * appropriately within the inode. The uid and gid for the inode are
  975. * set according to the contents of the given cred structure.
  976. *
  977. * Use xfs_dialloc() to allocate the on-disk inode. If xfs_dialloc()
  978. * has a free inode available, call xfs_iget()
  979. * to obtain the in-core version of the allocated inode. Finally,
  980. * fill in the inode and log its initial contents. In this case,
  981. * ialloc_context would be set to NULL and call_again set to false.
  982. *
  983. * If xfs_dialloc() does not have an available inode,
  984. * it will replenish its supply by doing an allocation. Since we can
  985. * only do one allocation within a transaction without deadlocks, we
  986. * must commit the current transaction before returning the inode itself.
  987. * In this case, therefore, we will set call_again to true and return.
  988. * The caller should then commit the current transaction, start a new
  989. * transaction, and call xfs_ialloc() again to actually get the inode.
  990. *
  991. * To ensure that some other process does not grab the inode that
  992. * was allocated during the first call to xfs_ialloc(), this routine
  993. * also returns the [locked] bp pointing to the head of the freelist
  994. * as ialloc_context. The caller should hold this buffer across
  995. * the commit and pass it back into this routine on the second call.
  996. */
  997. int
  998. xfs_ialloc(
  999. xfs_trans_t *tp,
  1000. xfs_inode_t *pip,
  1001. mode_t mode,
  1002. xfs_nlink_t nlink,
  1003. xfs_dev_t rdev,
  1004. cred_t *cr,
  1005. xfs_prid_t prid,
  1006. int okalloc,
  1007. xfs_buf_t **ialloc_context,
  1008. boolean_t *call_again,
  1009. xfs_inode_t **ipp)
  1010. {
  1011. xfs_ino_t ino;
  1012. xfs_inode_t *ip;
  1013. bhv_vnode_t *vp;
  1014. uint flags;
  1015. int error;
  1016. /*
  1017. * Call the space management code to pick
  1018. * the on-disk inode to be allocated.
  1019. */
  1020. error = xfs_dialloc(tp, pip->i_ino, mode, okalloc,
  1021. ialloc_context, call_again, &ino);
  1022. if (error != 0) {
  1023. return error;
  1024. }
  1025. if (*call_again || ino == NULLFSINO) {
  1026. *ipp = NULL;
  1027. return 0;
  1028. }
  1029. ASSERT(*ialloc_context == NULL);
  1030. /*
  1031. * Get the in-core inode with the lock held exclusively.
  1032. * This is because we're setting fields here we need
  1033. * to prevent others from looking at until we're done.
  1034. */
  1035. error = xfs_trans_iget(tp->t_mountp, tp, ino,
  1036. IGET_CREATE, XFS_ILOCK_EXCL, &ip);
  1037. if (error != 0) {
  1038. return error;
  1039. }
  1040. ASSERT(ip != NULL);
  1041. vp = XFS_ITOV(ip);
  1042. ip->i_d.di_mode = (__uint16_t)mode;
  1043. ip->i_d.di_onlink = 0;
  1044. ip->i_d.di_nlink = nlink;
  1045. ASSERT(ip->i_d.di_nlink == nlink);
  1046. ip->i_d.di_uid = current_fsuid(cr);
  1047. ip->i_d.di_gid = current_fsgid(cr);
  1048. ip->i_d.di_projid = prid;
  1049. memset(&(ip->i_d.di_pad[0]), 0, sizeof(ip->i_d.di_pad));
  1050. /*
  1051. * If the superblock version is up to where we support new format
  1052. * inodes and this is currently an old format inode, then change
  1053. * the inode version number now. This way we only do the conversion
  1054. * here rather than here and in the flush/logging code.
  1055. */
  1056. if (XFS_SB_VERSION_HASNLINK(&tp->t_mountp->m_sb) &&
  1057. ip->i_d.di_version == XFS_DINODE_VERSION_1) {
  1058. ip->i_d.di_version = XFS_DINODE_VERSION_2;
  1059. /*
  1060. * We've already zeroed the old link count, the projid field,
  1061. * and the pad field.
  1062. */
  1063. }
  1064. /*
  1065. * Project ids won't be stored on disk if we are using a version 1 inode.
  1066. */
  1067. if ( (prid != 0) && (ip->i_d.di_version == XFS_DINODE_VERSION_1))
  1068. xfs_bump_ino_vers2(tp, ip);
  1069. if (XFS_INHERIT_GID(pip, vp->v_vfsp)) {
  1070. ip->i_d.di_gid = pip->i_d.di_gid;
  1071. if ((pip->i_d.di_mode & S_ISGID) && (mode & S_IFMT) == S_IFDIR) {
  1072. ip->i_d.di_mode |= S_ISGID;
  1073. }
  1074. }
  1075. /*
  1076. * If the group ID of the new file does not match the effective group
  1077. * ID or one of the supplementary group IDs, the S_ISGID bit is cleared
  1078. * (and only if the irix_sgid_inherit compatibility variable is set).
  1079. */
  1080. if ((irix_sgid_inherit) &&
  1081. (ip->i_d.di_mode & S_ISGID) &&
  1082. (!in_group_p((gid_t)ip->i_d.di_gid))) {
  1083. ip->i_d.di_mode &= ~S_ISGID;
  1084. }
  1085. ip->i_d.di_size = 0;
  1086. ip->i_d.di_nextents = 0;
  1087. ASSERT(ip->i_d.di_nblocks == 0);
  1088. xfs_ichgtime(ip, XFS_ICHGTIME_CHG|XFS_ICHGTIME_ACC|XFS_ICHGTIME_MOD);
  1089. /*
  1090. * di_gen will have been taken care of in xfs_iread.
  1091. */
  1092. ip->i_d.di_extsize = 0;
  1093. ip->i_d.di_dmevmask = 0;
  1094. ip->i_d.di_dmstate = 0;
  1095. ip->i_d.di_flags = 0;
  1096. flags = XFS_ILOG_CORE;
  1097. switch (mode & S_IFMT) {
  1098. case S_IFIFO:
  1099. case S_IFCHR:
  1100. case S_IFBLK:
  1101. case S_IFSOCK:
  1102. ip->i_d.di_format = XFS_DINODE_FMT_DEV;
  1103. ip->i_df.if_u2.if_rdev = rdev;
  1104. ip->i_df.if_flags = 0;
  1105. flags |= XFS_ILOG_DEV;
  1106. break;
  1107. case S_IFREG:
  1108. case S_IFDIR:
  1109. if (unlikely(pip->i_d.di_flags & XFS_DIFLAG_ANY)) {
  1110. uint di_flags = 0;
  1111. if ((mode & S_IFMT) == S_IFDIR) {
  1112. if (pip->i_d.di_flags & XFS_DIFLAG_RTINHERIT)
  1113. di_flags |= XFS_DIFLAG_RTINHERIT;
  1114. if (pip->i_d.di_flags & XFS_DIFLAG_EXTSZINHERIT) {
  1115. di_flags |= XFS_DIFLAG_EXTSZINHERIT;
  1116. ip->i_d.di_extsize = pip->i_d.di_extsize;
  1117. }
  1118. } else if ((mode & S_IFMT) == S_IFREG) {
  1119. if (pip->i_d.di_flags & XFS_DIFLAG_RTINHERIT) {
  1120. di_flags |= XFS_DIFLAG_REALTIME;
  1121. ip->i_iocore.io_flags |= XFS_IOCORE_RT;
  1122. }
  1123. if (pip->i_d.di_flags & XFS_DIFLAG_EXTSZINHERIT) {
  1124. di_flags |= XFS_DIFLAG_EXTSIZE;
  1125. ip->i_d.di_extsize = pip->i_d.di_extsize;
  1126. }
  1127. }
  1128. if ((pip->i_d.di_flags & XFS_DIFLAG_NOATIME) &&
  1129. xfs_inherit_noatime)
  1130. di_flags |= XFS_DIFLAG_NOATIME;
  1131. if ((pip->i_d.di_flags & XFS_DIFLAG_NODUMP) &&
  1132. xfs_inherit_nodump)
  1133. di_flags |= XFS_DIFLAG_NODUMP;
  1134. if ((pip->i_d.di_flags & XFS_DIFLAG_SYNC) &&
  1135. xfs_inherit_sync)
  1136. di_flags |= XFS_DIFLAG_SYNC;
  1137. if ((pip->i_d.di_flags & XFS_DIFLAG_NOSYMLINKS) &&
  1138. xfs_inherit_nosymlinks)
  1139. di_flags |= XFS_DIFLAG_NOSYMLINKS;
  1140. if (pip->i_d.di_flags & XFS_DIFLAG_PROJINHERIT)
  1141. di_flags |= XFS_DIFLAG_PROJINHERIT;
  1142. if ((pip->i_d.di_flags & XFS_DIFLAG_NODEFRAG) &&
  1143. xfs_inherit_nodefrag)
  1144. di_flags |= XFS_DIFLAG_NODEFRAG;
  1145. ip->i_d.di_flags |= di_flags;
  1146. }
  1147. /* FALLTHROUGH */
  1148. case S_IFLNK:
  1149. ip->i_d.di_format = XFS_DINODE_FMT_EXTENTS;
  1150. ip->i_df.if_flags = XFS_IFEXTENTS;
  1151. ip->i_df.if_bytes = ip->i_df.if_real_bytes = 0;
  1152. ip->i_df.if_u1.if_extents = NULL;
  1153. break;
  1154. default:
  1155. ASSERT(0);
  1156. }
  1157. /*
  1158. * Attribute fork settings for new inode.
  1159. */
  1160. ip->i_d.di_aformat = XFS_DINODE_FMT_EXTENTS;
  1161. ip->i_d.di_anextents = 0;
  1162. /*
  1163. * Log the new values stuffed into the inode.
  1164. */
  1165. xfs_trans_log_inode(tp, ip, flags);
  1166. /* now that we have an i_mode we can setup inode ops and unlock */
  1167. bhv_vfs_init_vnode(XFS_MTOVFS(tp->t_mountp), vp, XFS_ITOBHV(ip), 1);
  1168. *ipp = ip;
  1169. return 0;
  1170. }
  1171. /*
  1172. * Check to make sure that there are no blocks allocated to the
  1173. * file beyond the size of the file. We don't check this for
  1174. * files with fixed size extents or real time extents, but we
  1175. * at least do it for regular files.
  1176. */
  1177. #ifdef DEBUG
  1178. void
  1179. xfs_isize_check(
  1180. xfs_mount_t *mp,
  1181. xfs_inode_t *ip,
  1182. xfs_fsize_t isize)
  1183. {
  1184. xfs_fileoff_t map_first;
  1185. int nimaps;
  1186. xfs_bmbt_irec_t imaps[2];
  1187. if ((ip->i_d.di_mode & S_IFMT) != S_IFREG)
  1188. return;
  1189. if (ip->i_d.di_flags & (XFS_DIFLAG_REALTIME | XFS_DIFLAG_EXTSIZE))
  1190. return;
  1191. nimaps = 2;
  1192. map_first = XFS_B_TO_FSB(mp, (xfs_ufsize_t)isize);
  1193. /*
  1194. * The filesystem could be shutting down, so bmapi may return
  1195. * an error.
  1196. */
  1197. if (xfs_bmapi(NULL, ip, map_first,
  1198. (XFS_B_TO_FSB(mp,
  1199. (xfs_ufsize_t)XFS_MAXIOFFSET(mp)) -
  1200. map_first),
  1201. XFS_BMAPI_ENTIRE, NULL, 0, imaps, &nimaps,
  1202. NULL, NULL))
  1203. return;
  1204. ASSERT(nimaps == 1);
  1205. ASSERT(imaps[0].br_startblock == HOLESTARTBLOCK);
  1206. }
  1207. #endif /* DEBUG */
  1208. /*
  1209. * Calculate the last possible buffered byte in a file. This must
  1210. * include data that was buffered beyond the EOF by the write code.
  1211. * This also needs to deal with overflowing the xfs_fsize_t type
  1212. * which can happen for sizes near the limit.
  1213. *
  1214. * We also need to take into account any blocks beyond the EOF. It
  1215. * may be the case that they were buffered by a write which failed.
  1216. * In that case the pages will still be in memory, but the inode size
  1217. * will never have been updated.
  1218. */
  1219. xfs_fsize_t
  1220. xfs_file_last_byte(
  1221. xfs_inode_t *ip)
  1222. {
  1223. xfs_mount_t *mp;
  1224. xfs_fsize_t last_byte;
  1225. xfs_fileoff_t last_block;
  1226. xfs_fileoff_t size_last_block;
  1227. int error;
  1228. ASSERT(ismrlocked(&(ip->i_iolock), MR_UPDATE | MR_ACCESS));
  1229. mp = ip->i_mount;
  1230. /*
  1231. * Only check for blocks beyond the EOF if the extents have
  1232. * been read in. This eliminates the need for the inode lock,
  1233. * and it also saves us from looking when it really isn't
  1234. * necessary.
  1235. */
  1236. if (ip->i_df.if_flags & XFS_IFEXTENTS) {
  1237. error = xfs_bmap_last_offset(NULL, ip, &last_block,
  1238. XFS_DATA_FORK);
  1239. if (error) {
  1240. last_block = 0;
  1241. }
  1242. } else {
  1243. last_block = 0;
  1244. }
  1245. size_last_block = XFS_B_TO_FSB(mp, (xfs_ufsize_t)ip->i_d.di_size);
  1246. last_block = XFS_FILEOFF_MAX(last_block, size_last_block);
  1247. last_byte = XFS_FSB_TO_B(mp, last_block);
  1248. if (last_byte < 0) {
  1249. return XFS_MAXIOFFSET(mp);
  1250. }
  1251. last_byte += (1 << mp->m_writeio_log);
  1252. if (last_byte < 0) {
  1253. return XFS_MAXIOFFSET(mp);
  1254. }
  1255. return last_byte;
  1256. }
  1257. #if defined(XFS_RW_TRACE)
  1258. STATIC void
  1259. xfs_itrunc_trace(
  1260. int tag,
  1261. xfs_inode_t *ip,
  1262. int flag,
  1263. xfs_fsize_t new_size,
  1264. xfs_off_t toss_start,
  1265. xfs_off_t toss_finish)
  1266. {
  1267. if (ip->i_rwtrace == NULL) {
  1268. return;
  1269. }
  1270. ktrace_enter(ip->i_rwtrace,
  1271. (void*)((long)tag),
  1272. (void*)ip,
  1273. (void*)(unsigned long)((ip->i_d.di_size >> 32) & 0xffffffff),
  1274. (void*)(unsigned long)(ip->i_d.di_size & 0xffffffff),
  1275. (void*)((long)flag),
  1276. (void*)(unsigned long)((new_size >> 32) & 0xffffffff),
  1277. (void*)(unsigned long)(new_size & 0xffffffff),
  1278. (void*)(unsigned long)((toss_start >> 32) & 0xffffffff),
  1279. (void*)(unsigned long)(toss_start & 0xffffffff),
  1280. (void*)(unsigned long)((toss_finish >> 32) & 0xffffffff),
  1281. (void*)(unsigned long)(toss_finish & 0xffffffff),
  1282. (void*)(unsigned long)current_cpu(),
  1283. (void*)(unsigned long)current_pid(),
  1284. (void*)NULL,
  1285. (void*)NULL,
  1286. (void*)NULL);
  1287. }
  1288. #else
  1289. #define xfs_itrunc_trace(tag, ip, flag, new_size, toss_start, toss_finish)
  1290. #endif
  1291. /*
  1292. * Start the truncation of the file to new_size. The new size
  1293. * must be smaller than the current size. This routine will
  1294. * clear the buffer and page caches of file data in the removed
  1295. * range, and xfs_itruncate_finish() will remove the underlying
  1296. * disk blocks.
  1297. *
  1298. * The inode must have its I/O lock locked EXCLUSIVELY, and it
  1299. * must NOT have the inode lock held at all. This is because we're
  1300. * calling into the buffer/page cache code and we can't hold the
  1301. * inode lock when we do so.
  1302. *
  1303. * We need to wait for any direct I/Os in flight to complete before we
  1304. * proceed with the truncate. This is needed to prevent the extents
  1305. * being read or written by the direct I/Os from being removed while the
  1306. * I/O is in flight as there is no other method of synchronising
  1307. * direct I/O with the truncate operation. Also, because we hold
  1308. * the IOLOCK in exclusive mode, we prevent new direct I/Os from being
  1309. * started until the truncate completes and drops the lock. Essentially,
  1310. * the vn_iowait() call forms an I/O barrier that provides strict ordering
  1311. * between direct I/Os and the truncate operation.
  1312. *
  1313. * The flags parameter can have either the value XFS_ITRUNC_DEFINITE
  1314. * or XFS_ITRUNC_MAYBE. The XFS_ITRUNC_MAYBE value should be used
  1315. * in the case that the caller is locking things out of order and
  1316. * may not be able to call xfs_itruncate_finish() with the inode lock
  1317. * held without dropping the I/O lock. If the caller must drop the
  1318. * I/O lock before calling xfs_itruncate_finish(), then xfs_itruncate_start()
  1319. * must be called again with all the same restrictions as the initial
  1320. * call.
  1321. */
  1322. void
  1323. xfs_itruncate_start(
  1324. xfs_inode_t *ip,
  1325. uint flags,
  1326. xfs_fsize_t new_size)
  1327. {
  1328. xfs_fsize_t last_byte;
  1329. xfs_off_t toss_start;
  1330. xfs_mount_t *mp;
  1331. bhv_vnode_t *vp;
  1332. ASSERT(ismrlocked(&ip->i_iolock, MR_UPDATE) != 0);
  1333. ASSERT((new_size == 0) || (new_size <= ip->i_d.di_size));
  1334. ASSERT((flags == XFS_ITRUNC_DEFINITE) ||
  1335. (flags == XFS_ITRUNC_MAYBE));
  1336. mp = ip->i_mount;
  1337. vp = XFS_ITOV(ip);
  1338. vn_iowait(vp); /* wait for the completion of any pending DIOs */
  1339. /*
  1340. * Call toss_pages or flushinval_pages to get rid of pages
  1341. * overlapping the region being removed. We have to use
  1342. * the less efficient flushinval_pages in the case that the
  1343. * caller may not be able to finish the truncate without
  1344. * dropping the inode's I/O lock. Make sure
  1345. * to catch any pages brought in by buffers overlapping
  1346. * the EOF by searching out beyond the isize by our
  1347. * block size. We round new_size up to a block boundary
  1348. * so that we don't toss things on the same block as
  1349. * new_size but before it.
  1350. *
  1351. * Before calling toss_page or flushinval_pages, make sure to
  1352. * call remapf() over the same region if the file is mapped.
  1353. * This frees up mapped file references to the pages in the
  1354. * given range and for the flushinval_pages case it ensures
  1355. * that we get the latest mapped changes flushed out.
  1356. */
  1357. toss_start = XFS_B_TO_FSB(mp, (xfs_ufsize_t)new_size);
  1358. toss_start = XFS_FSB_TO_B(mp, toss_start);
  1359. if (toss_start < 0) {
  1360. /*
  1361. * The place to start tossing is beyond our maximum
  1362. * file size, so there is no way that the data extended
  1363. * out there.
  1364. */
  1365. return;
  1366. }
  1367. last_byte = xfs_file_last_byte(ip);
  1368. xfs_itrunc_trace(XFS_ITRUNC_START, ip, flags, new_size, toss_start,
  1369. last_byte);
  1370. if (last_byte > toss_start) {
  1371. if (flags & XFS_ITRUNC_DEFINITE) {
  1372. bhv_vop_toss_pages(vp, toss_start, -1, FI_REMAPF_LOCKED);
  1373. } else {
  1374. bhv_vop_flushinval_pages(vp, toss_start, -1, FI_REMAPF_LOCKED);
  1375. }
  1376. }
  1377. #ifdef DEBUG
  1378. if (new_size == 0) {
  1379. ASSERT(VN_CACHED(vp) == 0);
  1380. }
  1381. #endif
  1382. }
  1383. /*
  1384. * Shrink the file to the given new_size. The new
  1385. * size must be smaller than the current size.
  1386. * This will free up the underlying blocks
  1387. * in the removed range after a call to xfs_itruncate_start()
  1388. * or xfs_atruncate_start().
  1389. *
  1390. * The transaction passed to this routine must have made
  1391. * a permanent log reservation of at least XFS_ITRUNCATE_LOG_RES.
  1392. * This routine may commit the given transaction and
  1393. * start new ones, so make sure everything involved in
  1394. * the transaction is tidy before calling here.
  1395. * Some transaction will be returned to the caller to be
  1396. * committed. The incoming transaction must already include
  1397. * the inode, and both inode locks must be held exclusively.
  1398. * The inode must also be "held" within the transaction. On
  1399. * return the inode will be "held" within the returned transaction.
  1400. * This routine does NOT require any disk space to be reserved
  1401. * for it within the transaction.
  1402. *
  1403. * The fork parameter must be either xfs_attr_fork or xfs_data_fork,
  1404. * and it indicates the fork which is to be truncated. For the
  1405. * attribute fork we only support truncation to size 0.
  1406. *
  1407. * We use the sync parameter to indicate whether or not the first
  1408. * transaction we perform might have to be synchronous. For the attr fork,
  1409. * it needs to be so if the unlink of the inode is not yet known to be
  1410. * permanent in the log. This keeps us from freeing and reusing the
  1411. * blocks of the attribute fork before the unlink of the inode becomes
  1412. * permanent.
  1413. *
  1414. * For the data fork, we normally have to run synchronously if we're
  1415. * being called out of the inactive path or we're being called
  1416. * out of the create path where we're truncating an existing file.
  1417. * Either way, the truncate needs to be sync so blocks don't reappear
  1418. * in the file with altered data in case of a crash. wsync filesystems
  1419. * can run the first case async because anything that shrinks the inode
  1420. * has to run sync so by the time we're called here from inactive, the
  1421. * inode size is permanently set to 0.
  1422. *
  1423. * Calls from the truncate path always need to be sync unless we're
  1424. * in a wsync filesystem and the file has already been unlinked.
  1425. *
  1426. * The caller is responsible for correctly setting the sync parameter.
  1427. * It gets too hard for us to guess here which path we're being called
  1428. * out of just based on inode state.
  1429. */
  1430. int
  1431. xfs_itruncate_finish(
  1432. xfs_trans_t **tp,
  1433. xfs_inode_t *ip,
  1434. xfs_fsize_t new_size,
  1435. int fork,
  1436. int sync)
  1437. {
  1438. xfs_fsblock_t first_block;
  1439. xfs_fileoff_t first_unmap_block;
  1440. xfs_fileoff_t last_block;
  1441. xfs_filblks_t unmap_len=0;
  1442. xfs_mount_t *mp;
  1443. xfs_trans_t *ntp;
  1444. int done;
  1445. int committed;
  1446. xfs_bmap_free_t free_list;
  1447. int error;
  1448. ASSERT(ismrlocked(&ip->i_iolock, MR_UPDATE) != 0);
  1449. ASSERT(ismrlocked(&ip->i_lock, MR_UPDATE) != 0);
  1450. ASSERT((new_size == 0) || (new_size <= ip->i_d.di_size));
  1451. ASSERT(*tp != NULL);
  1452. ASSERT((*tp)->t_flags & XFS_TRANS_PERM_LOG_RES);
  1453. ASSERT(ip->i_transp == *tp);
  1454. ASSERT(ip->i_itemp != NULL);
  1455. ASSERT(ip->i_itemp->ili_flags & XFS_ILI_HOLD);
  1456. ntp = *tp;
  1457. mp = (ntp)->t_mountp;
  1458. ASSERT(! XFS_NOT_DQATTACHED(mp, ip));
  1459. /*
  1460. * We only support truncating the entire attribute fork.
  1461. */
  1462. if (fork == XFS_ATTR_FORK) {
  1463. new_size = 0LL;
  1464. }
  1465. first_unmap_block = XFS_B_TO_FSB(mp, (xfs_ufsize_t)new_size);
  1466. xfs_itrunc_trace(XFS_ITRUNC_FINISH1, ip, 0, new_size, 0, 0);
  1467. /*
  1468. * The first thing we do is set the size to new_size permanently
  1469. * on disk. This way we don't have to worry about anyone ever
  1470. * being able to look at the data being freed even in the face
  1471. * of a crash. What we're getting around here is the case where
  1472. * we free a block, it is allocated to another file, it is written
  1473. * to, and then we crash. If the new data gets written to the
  1474. * file but the log buffers containing the free and reallocation
  1475. * don't, then we'd end up with garbage in the blocks being freed.
  1476. * As long as we make the new_size permanent before actually
  1477. * freeing any blocks it doesn't matter if they get writtten to.
  1478. *
  1479. * The callers must signal into us whether or not the size
  1480. * setting here must be synchronous. There are a few cases
  1481. * where it doesn't have to be synchronous. Those cases
  1482. * occur if the file is unlinked and we know the unlink is
  1483. * permanent or if the blocks being truncated are guaranteed
  1484. * to be beyond the inode eof (regardless of the link count)
  1485. * and the eof value is permanent. Both of these cases occur
  1486. * only on wsync-mounted filesystems. In those cases, we're
  1487. * guaranteed that no user will ever see the data in the blocks
  1488. * that are being truncated so the truncate can run async.
  1489. * In the free beyond eof case, the file may wind up with
  1490. * more blocks allocated to it than it needs if we crash
  1491. * and that won't get fixed until the next time the file
  1492. * is re-opened and closed but that's ok as that shouldn't
  1493. * be too many blocks.
  1494. *
  1495. * However, we can't just make all wsync xactions run async
  1496. * because there's one call out of the create path that needs
  1497. * to run sync where it's truncating an existing file to size
  1498. * 0 whose size is > 0.
  1499. *
  1500. * It's probably possible to come up with a test in this
  1501. * routine that would correctly distinguish all the above
  1502. * cases from the values of the function parameters and the
  1503. * inode state but for sanity's sake, I've decided to let the
  1504. * layers above just tell us. It's simpler to correctly figure
  1505. * out in the layer above exactly under what conditions we
  1506. * can run async and I think it's easier for others read and
  1507. * follow the logic in case something has to be changed.
  1508. * cscope is your friend -- rcc.
  1509. *
  1510. * The attribute fork is much simpler.
  1511. *
  1512. * For the attribute fork we allow the caller to tell us whether
  1513. * the unlink of the inode that led to this call is yet permanent
  1514. * in the on disk log. If it is not and we will be freeing extents
  1515. * in this inode then we make the first transaction synchronous
  1516. * to make sure that the unlink is permanent by the time we free
  1517. * the blocks.
  1518. */
  1519. if (fork == XFS_DATA_FORK) {
  1520. if (ip->i_d.di_nextents > 0) {
  1521. ip->i_d.di_size = new_size;
  1522. xfs_trans_log_inode(ntp, ip, XFS_ILOG_CORE);
  1523. }
  1524. } else if (sync) {
  1525. ASSERT(!(mp->m_flags & XFS_MOUNT_WSYNC));
  1526. if (ip->i_d.di_anextents > 0)
  1527. xfs_trans_set_sync(ntp);
  1528. }
  1529. ASSERT(fork == XFS_DATA_FORK ||
  1530. (fork == XFS_ATTR_FORK &&
  1531. ((sync && !(mp->m_flags & XFS_MOUNT_WSYNC)) ||
  1532. (sync == 0 && (mp->m_flags & XFS_MOUNT_WSYNC)))));
  1533. /*
  1534. * Since it is possible for space to become allocated beyond
  1535. * the end of the file (in a crash where the space is allocated
  1536. * but the inode size is not yet updated), simply remove any
  1537. * blocks which show up between the new EOF and the maximum
  1538. * possible file size. If the first block to be removed is
  1539. * beyond the maximum file size (ie it is the same as last_block),
  1540. * then there is nothing to do.
  1541. */
  1542. last_block = XFS_B_TO_FSB(mp, (xfs_ufsize_t)XFS_MAXIOFFSET(mp));
  1543. ASSERT(first_unmap_block <= last_block);
  1544. done = 0;
  1545. if (last_block == first_unmap_block) {
  1546. done = 1;
  1547. } else {
  1548. unmap_len = last_block - first_unmap_block + 1;
  1549. }
  1550. while (!done) {
  1551. /*
  1552. * Free up up to XFS_ITRUNC_MAX_EXTENTS. xfs_bunmapi()
  1553. * will tell us whether it freed the entire range or
  1554. * not. If this is a synchronous mount (wsync),
  1555. * then we can tell bunmapi to keep all the
  1556. * transactions asynchronous since the unlink
  1557. * transaction that made this inode inactive has
  1558. * already hit the disk. There's no danger of
  1559. * the freed blocks being reused, there being a
  1560. * crash, and the reused blocks suddenly reappearing
  1561. * in this file with garbage in them once recovery
  1562. * runs.
  1563. */
  1564. XFS_BMAP_INIT(&free_list, &first_block);
  1565. error = XFS_BUNMAPI(mp, ntp, &ip->i_iocore,
  1566. first_unmap_block, unmap_len,
  1567. XFS_BMAPI_AFLAG(fork) |
  1568. (sync ? 0 : XFS_BMAPI_ASYNC),
  1569. XFS_ITRUNC_MAX_EXTENTS,
  1570. &first_block, &free_list,
  1571. NULL, &done);
  1572. if (error) {
  1573. /*
  1574. * If the bunmapi call encounters an error,
  1575. * return to the caller where the transaction
  1576. * can be properly aborted. We just need to
  1577. * make sure we're not holding any resources
  1578. * that we were not when we came in.
  1579. */
  1580. xfs_bmap_cancel(&free_list);
  1581. return error;
  1582. }
  1583. /*
  1584. * Duplicate the transaction that has the permanent
  1585. * reservation and commit the old transaction.
  1586. */
  1587. error = xfs_bmap_finish(tp, &free_list, first_block,
  1588. &committed);
  1589. ntp = *tp;
  1590. if (error) {
  1591. /*
  1592. * If the bmap finish call encounters an error,
  1593. * return to the caller where the transaction
  1594. * can be properly aborted. We just need to
  1595. * make sure we're not holding any resources
  1596. * that we were not when we came in.
  1597. *
  1598. * Aborting from this point might lose some
  1599. * blocks in the file system, but oh well.
  1600. */
  1601. xfs_bmap_cancel(&free_list);
  1602. if (committed) {
  1603. /*
  1604. * If the passed in transaction committed
  1605. * in xfs_bmap_finish(), then we want to
  1606. * add the inode to this one before returning.
  1607. * This keeps things simple for the higher
  1608. * level code, because it always knows that
  1609. * the inode is locked and held in the
  1610. * transaction that returns to it whether
  1611. * errors occur or not. We don't mark the
  1612. * inode dirty so that this transaction can
  1613. * be easily aborted if possible.
  1614. */
  1615. xfs_trans_ijoin(ntp, ip,
  1616. XFS_ILOCK_EXCL | XFS_IOLOCK_EXCL);
  1617. xfs_trans_ihold(ntp, ip);
  1618. }
  1619. return error;
  1620. }
  1621. if (committed) {
  1622. /*
  1623. * The first xact was committed,
  1624. * so add the inode to the new one.
  1625. * Mark it dirty so it will be logged
  1626. * and moved forward in the log as
  1627. * part of every commit.
  1628. */
  1629. xfs_trans_ijoin(ntp, ip,
  1630. XFS_ILOCK_EXCL | XFS_IOLOCK_EXCL);
  1631. xfs_trans_ihold(ntp, ip);
  1632. xfs_trans_log_inode(ntp, ip, XFS_ILOG_CORE);
  1633. }
  1634. ntp = xfs_trans_dup(ntp);
  1635. (void) xfs_trans_commit(*tp, 0, NULL);
  1636. *tp = ntp;
  1637. error = xfs_trans_reserve(ntp, 0, XFS_ITRUNCATE_LOG_RES(mp), 0,
  1638. XFS_TRANS_PERM_LOG_RES,
  1639. XFS_ITRUNCATE_LOG_COUNT);
  1640. /*
  1641. * Add the inode being truncated to the next chained
  1642. * transaction.
  1643. */
  1644. xfs_trans_ijoin(ntp, ip, XFS_ILOCK_EXCL | XFS_IOLOCK_EXCL);
  1645. xfs_trans_ihold(ntp, ip);
  1646. if (error)
  1647. return (error);
  1648. }
  1649. /*
  1650. * Only update the size in the case of the data fork, but
  1651. * always re-log the inode so that our permanent transaction
  1652. * can keep on rolling it forward in the log.
  1653. */
  1654. if (fork == XFS_DATA_FORK) {
  1655. xfs_isize_check(mp, ip, new_size);
  1656. ip->i_d.di_size = new_size;
  1657. }
  1658. xfs_trans_log_inode(ntp, ip, XFS_ILOG_CORE);
  1659. ASSERT((new_size != 0) ||
  1660. (fork == XFS_ATTR_FORK) ||
  1661. (ip->i_delayed_blks == 0));
  1662. ASSERT((new_size != 0) ||
  1663. (fork == XFS_ATTR_FORK) ||
  1664. (ip->i_d.di_nextents == 0));
  1665. xfs_itrunc_trace(XFS_ITRUNC_FINISH2, ip, 0, new_size, 0, 0);
  1666. return 0;
  1667. }
  1668. /*
  1669. * xfs_igrow_start
  1670. *
  1671. * Do the first part of growing a file: zero any data in the last
  1672. * block that is beyond the old EOF. We need to do this before
  1673. * the inode is joined to the transaction to modify the i_size.
  1674. * That way we can drop the inode lock and call into the buffer
  1675. * cache to get the buffer mapping the EOF.
  1676. */
  1677. int
  1678. xfs_igrow_start(
  1679. xfs_inode_t *ip,
  1680. xfs_fsize_t new_size,
  1681. cred_t *credp)
  1682. {
  1683. int error;
  1684. ASSERT(ismrlocked(&(ip->i_lock), MR_UPDATE) != 0);
  1685. ASSERT(ismrlocked(&(ip->i_iolock), MR_UPDATE) != 0);
  1686. ASSERT(new_size > ip->i_d.di_size);
  1687. /*
  1688. * Zero any pages that may have been created by
  1689. * xfs_write_file() beyond the end of the file
  1690. * and any blocks between the old and new file sizes.
  1691. */
  1692. error = xfs_zero_eof(XFS_ITOV(ip), &ip->i_iocore, new_size,
  1693. ip->i_d.di_size, new_size);
  1694. return error;
  1695. }
  1696. /*
  1697. * xfs_igrow_finish
  1698. *
  1699. * This routine is called to extend the size of a file.
  1700. * The inode must have both the iolock and the ilock locked
  1701. * for update and it must be a part of the current transaction.
  1702. * The xfs_igrow_start() function must have been called previously.
  1703. * If the change_flag is not zero, the inode change timestamp will
  1704. * be updated.
  1705. */
  1706. void
  1707. xfs_igrow_finish(
  1708. xfs_trans_t *tp,
  1709. xfs_inode_t *ip,
  1710. xfs_fsize_t new_size,
  1711. int change_flag)
  1712. {
  1713. ASSERT(ismrlocked(&(ip->i_lock), MR_UPDATE) != 0);
  1714. ASSERT(ismrlocked(&(ip->i_iolock), MR_UPDATE) != 0);
  1715. ASSERT(ip->i_transp == tp);
  1716. ASSERT(new_size > ip->i_d.di_size);
  1717. /*
  1718. * Update the file size. Update the inode change timestamp
  1719. * if change_flag set.
  1720. */
  1721. ip->i_d.di_size = new_size;
  1722. if (change_flag)
  1723. xfs_ichgtime(ip, XFS_ICHGTIME_CHG);
  1724. xfs_trans_log_inode(tp, ip, XFS_ILOG_CORE);
  1725. }
  1726. /*
  1727. * This is called when the inode's link count goes to 0.
  1728. * We place the on-disk inode on a list in the AGI. It
  1729. * will be pulled from this list when the inode is freed.
  1730. */
  1731. int
  1732. xfs_iunlink(
  1733. xfs_trans_t *tp,
  1734. xfs_inode_t *ip)
  1735. {
  1736. xfs_mount_t *mp;
  1737. xfs_agi_t *agi;
  1738. xfs_dinode_t *dip;
  1739. xfs_buf_t *agibp;
  1740. xfs_buf_t *ibp;
  1741. xfs_agnumber_t agno;
  1742. xfs_daddr_t agdaddr;
  1743. xfs_agino_t agino;
  1744. short bucket_index;
  1745. int offset;
  1746. int error;
  1747. int agi_ok;
  1748. ASSERT(ip->i_d.di_nlink == 0);
  1749. ASSERT(ip->i_d.di_mode != 0);
  1750. ASSERT(ip->i_transp == tp);
  1751. mp = tp->t_mountp;
  1752. agno = XFS_INO_TO_AGNO(mp, ip->i_ino);
  1753. agdaddr = XFS_AG_DADDR(mp, agno, XFS_AGI_DADDR(mp));
  1754. /*
  1755. * Get the agi buffer first. It ensures lock ordering
  1756. * on the list.
  1757. */
  1758. error = xfs_trans_read_buf(mp, tp, mp->m_ddev_targp, agdaddr,
  1759. XFS_FSS_TO_BB(mp, 1), 0, &agibp);
  1760. if (error) {
  1761. return error;
  1762. }
  1763. /*
  1764. * Validate the magic number of the agi block.
  1765. */
  1766. agi = XFS_BUF_TO_AGI(agibp);
  1767. agi_ok =
  1768. be32_to_cpu(agi->agi_magicnum) == XFS_AGI_MAGIC &&
  1769. XFS_AGI_GOOD_VERSION(be32_to_cpu(agi->agi_versionnum));
  1770. if (unlikely(XFS_TEST_ERROR(!agi_ok, mp, XFS_ERRTAG_IUNLINK,
  1771. XFS_RANDOM_IUNLINK))) {
  1772. XFS_CORRUPTION_ERROR("xfs_iunlink", XFS_ERRLEVEL_LOW, mp, agi);
  1773. xfs_trans_brelse(tp, agibp);
  1774. return XFS_ERROR(EFSCORRUPTED);
  1775. }
  1776. /*
  1777. * Get the index into the agi hash table for the
  1778. * list this inode will go on.
  1779. */
  1780. agino = XFS_INO_TO_AGINO(mp, ip->i_ino);
  1781. ASSERT(agino != 0);
  1782. bucket_index = agino % XFS_AGI_UNLINKED_BUCKETS;
  1783. ASSERT(agi->agi_unlinked[bucket_index]);
  1784. ASSERT(be32_to_cpu(agi->agi_unlinked[bucket_index]) != agino);
  1785. if (be32_to_cpu(agi->agi_unlinked[bucket_index]) != NULLAGINO) {
  1786. /*
  1787. * There is already another inode in the bucket we need
  1788. * to add ourselves to. Add us at the front of the list.
  1789. * Here we put the head pointer into our next pointer,
  1790. * and then we fall through to point the head at us.
  1791. */
  1792. error = xfs_itobp(mp, tp, ip, &dip, &ibp, 0, 0);
  1793. if (error) {
  1794. return error;
  1795. }
  1796. ASSERT(INT_GET(dip->di_next_unlinked, ARCH_CONVERT) == NULLAGINO);
  1797. ASSERT(dip->di_next_unlinked);
  1798. /* both on-disk, don't endian flip twice */
  1799. dip->di_next_unlinked = agi->agi_unlinked[bucket_index];
  1800. offset = ip->i_boffset +
  1801. offsetof(xfs_dinode_t, di_next_unlinked);
  1802. xfs_trans_inode_buf(tp, ibp);
  1803. xfs_trans_log_buf(tp, ibp, offset,
  1804. (offset + sizeof(xfs_agino_t) - 1));
  1805. xfs_inobp_check(mp, ibp);
  1806. }
  1807. /*
  1808. * Point the bucket head pointer at the inode being inserted.
  1809. */
  1810. ASSERT(agino != 0);
  1811. agi->agi_unlinked[bucket_index] = cpu_to_be32(agino);
  1812. offset = offsetof(xfs_agi_t, agi_unlinked) +
  1813. (sizeof(xfs_agino_t) * bucket_index);
  1814. xfs_trans_log_buf(tp, agibp, offset,
  1815. (offset + sizeof(xfs_agino_t) - 1));
  1816. return 0;
  1817. }
  1818. /*
  1819. * Pull the on-disk inode from the AGI unlinked list.
  1820. */
  1821. STATIC int
  1822. xfs_iunlink_remove(
  1823. xfs_trans_t *tp,
  1824. xfs_inode_t *ip)
  1825. {
  1826. xfs_ino_t next_ino;
  1827. xfs_mount_t *mp;
  1828. xfs_agi_t *agi;
  1829. xfs_dinode_t *dip;
  1830. xfs_buf_t *agibp;
  1831. xfs_buf_t *ibp;
  1832. xfs_agnumber_t agno;
  1833. xfs_daddr_t agdaddr;
  1834. xfs_agino_t agino;
  1835. xfs_agino_t next_agino;
  1836. xfs_buf_t *last_ibp;
  1837. xfs_dinode_t *last_dip;
  1838. short bucket_index;
  1839. int offset, last_offset;
  1840. int error;
  1841. int agi_ok;
  1842. /*
  1843. * First pull the on-disk inode from the AGI unlinked list.
  1844. */
  1845. mp = tp->t_mountp;
  1846. agno = XFS_INO_TO_AGNO(mp, ip->i_ino);
  1847. agdaddr = XFS_AG_DADDR(mp, agno, XFS_AGI_DADDR(mp));
  1848. /*
  1849. * Get the agi buffer first. It ensures lock ordering
  1850. * on the list.
  1851. */
  1852. error = xfs_trans_read_buf(mp, tp, mp->m_ddev_targp, agdaddr,
  1853. XFS_FSS_TO_BB(mp, 1), 0, &agibp);
  1854. if (error) {
  1855. cmn_err(CE_WARN,
  1856. "xfs_iunlink_remove: xfs_trans_read_buf() returned an error %d on %s. Returning error.",
  1857. error, mp->m_fsname);
  1858. return error;
  1859. }
  1860. /*
  1861. * Validate the magic number of the agi block.
  1862. */
  1863. agi = XFS_BUF_TO_AGI(agibp);
  1864. agi_ok =
  1865. be32_to_cpu(agi->agi_magicnum) == XFS_AGI_MAGIC &&
  1866. XFS_AGI_GOOD_VERSION(be32_to_cpu(agi->agi_versionnum));
  1867. if (unlikely(XFS_TEST_ERROR(!agi_ok, mp, XFS_ERRTAG_IUNLINK_REMOVE,
  1868. XFS_RANDOM_IUNLINK_REMOVE))) {
  1869. XFS_CORRUPTION_ERROR("xfs_iunlink_remove", XFS_ERRLEVEL_LOW,
  1870. mp, agi);
  1871. xfs_trans_brelse(tp, agibp);
  1872. cmn_err(CE_WARN,
  1873. "xfs_iunlink_remove: XFS_TEST_ERROR() returned an error on %s. Returning EFSCORRUPTED.",
  1874. mp->m_fsname);
  1875. return XFS_ERROR(EFSCORRUPTED);
  1876. }
  1877. /*
  1878. * Get the index into the agi hash table for the
  1879. * list this inode will go on.
  1880. */
  1881. agino = XFS_INO_TO_AGINO(mp, ip->i_ino);
  1882. ASSERT(agino != 0);
  1883. bucket_index = agino % XFS_AGI_UNLINKED_BUCKETS;
  1884. ASSERT(be32_to_cpu(agi->agi_unlinked[bucket_index]) != NULLAGINO);
  1885. ASSERT(agi->agi_unlinked[bucket_index]);
  1886. if (be32_to_cpu(agi->agi_unlinked[bucket_index]) == agino) {
  1887. /*
  1888. * We're at the head of the list. Get the inode's
  1889. * on-disk buffer to see if there is anyone after us
  1890. * on the list. Only modify our next pointer if it
  1891. * is not already NULLAGINO. This saves us the overhead
  1892. * of dealing with the buffer when there is no need to
  1893. * change it.
  1894. */
  1895. error = xfs_itobp(mp, tp, ip, &dip, &ibp, 0, 0);
  1896. if (error) {
  1897. cmn_err(CE_WARN,
  1898. "xfs_iunlink_remove: xfs_itobp() returned an error %d on %s. Returning error.",
  1899. error, mp->m_fsname);
  1900. return error;
  1901. }
  1902. next_agino = INT_GET(dip->di_next_unlinked, ARCH_CONVERT);
  1903. ASSERT(next_agino != 0);
  1904. if (next_agino != NULLAGINO) {
  1905. INT_SET(dip->di_next_unlinked, ARCH_CONVERT, NULLAGINO);
  1906. offset = ip->i_boffset +
  1907. offsetof(xfs_dinode_t, di_next_unlinked);
  1908. xfs_trans_inode_buf(tp, ibp);
  1909. xfs_trans_log_buf(tp, ibp, offset,
  1910. (offset + sizeof(xfs_agino_t) - 1));
  1911. xfs_inobp_check(mp, ibp);
  1912. } else {
  1913. xfs_trans_brelse(tp, ibp);
  1914. }
  1915. /*
  1916. * Point the bucket head pointer at the next inode.
  1917. */
  1918. ASSERT(next_agino != 0);
  1919. ASSERT(next_agino != agino);
  1920. agi->agi_unlinked[bucket_index] = cpu_to_be32(next_agino);
  1921. offset = offsetof(xfs_agi_t, agi_unlinked) +
  1922. (sizeof(xfs_agino_t) * bucket_index);
  1923. xfs_trans_log_buf(tp, agibp, offset,
  1924. (offset + sizeof(xfs_agino_t) - 1));
  1925. } else {
  1926. /*
  1927. * We need to search the list for the inode being freed.
  1928. */
  1929. next_agino = be32_to_cpu(agi->agi_unlinked[bucket_index]);
  1930. last_ibp = NULL;
  1931. while (next_agino != agino) {
  1932. /*
  1933. * If the last inode wasn't the one pointing to
  1934. * us, then release its buffer since we're not
  1935. * going to do anything with it.
  1936. */
  1937. if (last_ibp != NULL) {
  1938. xfs_trans_brelse(tp, last_ibp);
  1939. }
  1940. next_ino = XFS_AGINO_TO_INO(mp, agno, next_agino);
  1941. error = xfs_inotobp(mp, tp, next_ino, &last_dip,
  1942. &last_ibp, &last_offset);
  1943. if (error) {
  1944. cmn_err(CE_WARN,
  1945. "xfs_iunlink_remove: xfs_inotobp() returned an error %d on %s. Returning error.",
  1946. error, mp->m_fsname);
  1947. return error;
  1948. }
  1949. next_agino = INT_GET(last_dip->di_next_unlinked, ARCH_CONVERT);
  1950. ASSERT(next_agino != NULLAGINO);
  1951. ASSERT(next_agino != 0);
  1952. }
  1953. /*
  1954. * Now last_ibp points to the buffer previous to us on
  1955. * the unlinked list. Pull us from the list.
  1956. */
  1957. error = xfs_itobp(mp, tp, ip, &dip, &ibp, 0, 0);
  1958. if (error) {
  1959. cmn_err(CE_WARN,
  1960. "xfs_iunlink_remove: xfs_itobp() returned an error %d on %s. Returning error.",
  1961. error, mp->m_fsname);
  1962. return error;
  1963. }
  1964. next_agino = INT_GET(dip->di_next_unlinked, ARCH_CONVERT);
  1965. ASSERT(next_agino != 0);
  1966. ASSERT(next_agino != agino);
  1967. if (next_agino != NULLAGINO) {
  1968. INT_SET(dip->di_next_unlinked, ARCH_CONVERT, NULLAGINO);
  1969. offset = ip->i_boffset +
  1970. offsetof(xfs_dinode_t, di_next_unlinked);
  1971. xfs_trans_inode_buf(tp, ibp);
  1972. xfs_trans_log_buf(tp, ibp, offset,
  1973. (offset + sizeof(xfs_agino_t) - 1));
  1974. xfs_inobp_check(mp, ibp);
  1975. } else {
  1976. xfs_trans_brelse(tp, ibp);
  1977. }
  1978. /*
  1979. * Point the previous inode on the list to the next inode.
  1980. */
  1981. INT_SET(last_dip->di_next_unlinked, ARCH_CONVERT, next_agino);
  1982. ASSERT(next_agino != 0);
  1983. offset = last_offset + offsetof(xfs_dinode_t, di_next_unlinked);
  1984. xfs_trans_inode_buf(tp, last_ibp);
  1985. xfs_trans_log_buf(tp, last_ibp, offset,
  1986. (offset + sizeof(xfs_agino_t) - 1));
  1987. xfs_inobp_check(mp, last_ibp);
  1988. }
  1989. return 0;
  1990. }
  1991. static __inline__ int xfs_inode_clean(xfs_inode_t *ip)
  1992. {
  1993. return (((ip->i_itemp == NULL) ||
  1994. !(ip->i_itemp->ili_format.ilf_fields & XFS_ILOG_ALL)) &&
  1995. (ip->i_update_core == 0));
  1996. }
  1997. STATIC void
  1998. xfs_ifree_cluster(
  1999. xfs_inode_t *free_ip,
  2000. xfs_trans_t *tp,
  2001. xfs_ino_t inum)
  2002. {
  2003. xfs_mount_t *mp = free_ip->i_mount;
  2004. int blks_per_cluster;
  2005. int nbufs;
  2006. int ninodes;
  2007. int i, j, found, pre_flushed;
  2008. xfs_daddr_t blkno;
  2009. xfs_buf_t *bp;
  2010. xfs_ihash_t *ih;
  2011. xfs_inode_t *ip, **ip_found;
  2012. xfs_inode_log_item_t *iip;
  2013. xfs_log_item_t *lip;
  2014. SPLDECL(s);
  2015. if (mp->m_sb.sb_blocksize >= XFS_INODE_CLUSTER_SIZE(mp)) {
  2016. blks_per_cluster = 1;
  2017. ninodes = mp->m_sb.sb_inopblock;
  2018. nbufs = XFS_IALLOC_BLOCKS(mp);
  2019. } else {
  2020. blks_per_cluster = XFS_INODE_CLUSTER_SIZE(mp) /
  2021. mp->m_sb.sb_blocksize;
  2022. ninodes = blks_per_cluster * mp->m_sb.sb_inopblock;
  2023. nbufs = XFS_IALLOC_BLOCKS(mp) / blks_per_cluster;
  2024. }
  2025. ip_found = kmem_alloc(ninodes * sizeof(xfs_inode_t *), KM_NOFS);
  2026. for (j = 0; j < nbufs; j++, inum += ninodes) {
  2027. blkno = XFS_AGB_TO_DADDR(mp, XFS_INO_TO_AGNO(mp, inum),
  2028. XFS_INO_TO_AGBNO(mp, inum));
  2029. /*
  2030. * Look for each inode in memory and attempt to lock it,
  2031. * we can be racing with flush and tail pushing here.
  2032. * any inode we get the locks on, add to an array of
  2033. * inode items to process later.
  2034. *
  2035. * The get the buffer lock, we could beat a flush
  2036. * or tail pushing thread to the lock here, in which
  2037. * case they will go looking for the inode buffer
  2038. * and fail, we need some other form of interlock
  2039. * here.
  2040. */
  2041. found = 0;
  2042. for (i = 0; i < ninodes; i++) {
  2043. ih = XFS_IHASH(mp, inum + i);
  2044. read_lock(&ih->ih_lock);
  2045. for (ip = ih->ih_next; ip != NULL; ip = ip->i_next) {
  2046. if (ip->i_ino == inum + i)
  2047. break;
  2048. }
  2049. /* Inode not in memory or we found it already,
  2050. * nothing to do
  2051. */
  2052. if (!ip || (ip->i_flags & XFS_ISTALE)) {
  2053. read_unlock(&ih->ih_lock);
  2054. continue;
  2055. }
  2056. if (xfs_inode_clean(ip)) {
  2057. read_unlock(&ih->ih_lock);
  2058. continue;
  2059. }
  2060. /* If we can get the locks then add it to the
  2061. * list, otherwise by the time we get the bp lock
  2062. * below it will already be attached to the
  2063. * inode buffer.
  2064. */
  2065. /* This inode will already be locked - by us, lets
  2066. * keep it that way.
  2067. */
  2068. if (ip == free_ip) {
  2069. if (xfs_iflock_nowait(ip)) {
  2070. ip->i_flags |= XFS_ISTALE;
  2071. if (xfs_inode_clean(ip)) {
  2072. xfs_ifunlock(ip);
  2073. } else {
  2074. ip_found[found++] = ip;
  2075. }
  2076. }
  2077. read_unlock(&ih->ih_lock);
  2078. continue;
  2079. }
  2080. if (xfs_ilock_nowait(ip, XFS_ILOCK_EXCL)) {
  2081. if (xfs_iflock_nowait(ip)) {
  2082. ip->i_flags |= XFS_ISTALE;
  2083. if (xfs_inode_clean(ip)) {
  2084. xfs_ifunlock(ip);
  2085. xfs_iunlock(ip, XFS_ILOCK_EXCL);
  2086. } else {
  2087. ip_found[found++] = ip;
  2088. }
  2089. } else {
  2090. xfs_iunlock(ip, XFS_ILOCK_EXCL);
  2091. }
  2092. }
  2093. read_unlock(&ih->ih_lock);
  2094. }
  2095. bp = xfs_trans_get_buf(tp, mp->m_ddev_targp, blkno,
  2096. mp->m_bsize * blks_per_cluster,
  2097. XFS_BUF_LOCK);
  2098. pre_flushed = 0;
  2099. lip = XFS_BUF_FSPRIVATE(bp, xfs_log_item_t *);
  2100. while (lip) {
  2101. if (lip->li_type == XFS_LI_INODE) {
  2102. iip = (xfs_inode_log_item_t *)lip;
  2103. ASSERT(iip->ili_logged == 1);
  2104. lip->li_cb = (void(*)(xfs_buf_t*,xfs_log_item_t*)) xfs_istale_done;
  2105. AIL_LOCK(mp,s);
  2106. iip->ili_flush_lsn = iip->ili_item.li_lsn;
  2107. AIL_UNLOCK(mp, s);
  2108. iip->ili_inode->i_flags |= XFS_ISTALE;
  2109. pre_flushed++;
  2110. }
  2111. lip = lip->li_bio_list;
  2112. }
  2113. for (i = 0; i < found; i++) {
  2114. ip = ip_found[i];
  2115. iip = ip->i_itemp;
  2116. if (!iip) {
  2117. ip->i_update_core = 0;
  2118. xfs_ifunlock(ip);
  2119. xfs_iunlock(ip, XFS_ILOCK_EXCL);
  2120. continue;
  2121. }
  2122. iip->ili_last_fields = iip->ili_format.ilf_fields;
  2123. iip->ili_format.ilf_fields = 0;
  2124. iip->ili_logged = 1;
  2125. AIL_LOCK(mp,s);
  2126. iip->ili_flush_lsn = iip->ili_item.li_lsn;
  2127. AIL_UNLOCK(mp, s);
  2128. xfs_buf_attach_iodone(bp,
  2129. (void(*)(xfs_buf_t*,xfs_log_item_t*))
  2130. xfs_istale_done, (xfs_log_item_t *)iip);
  2131. if (ip != free_ip) {
  2132. xfs_iunlock(ip, XFS_ILOCK_EXCL);
  2133. }
  2134. }
  2135. if (found || pre_flushed)
  2136. xfs_trans_stale_inode_buf(tp, bp);
  2137. xfs_trans_binval(tp, bp);
  2138. }
  2139. kmem_free(ip_found, ninodes * sizeof(xfs_inode_t *));
  2140. }
  2141. /*
  2142. * This is called to return an inode to the inode free list.
  2143. * The inode should already be truncated to 0 length and have
  2144. * no pages associated with it. This routine also assumes that
  2145. * the inode is already a part of the transaction.
  2146. *
  2147. * The on-disk copy of the inode will have been added to the list
  2148. * of unlinked inodes in the AGI. We need to remove the inode from
  2149. * that list atomically with respect to freeing it here.
  2150. */
  2151. int
  2152. xfs_ifree(
  2153. xfs_trans_t *tp,
  2154. xfs_inode_t *ip,
  2155. xfs_bmap_free_t *flist)
  2156. {
  2157. int error;
  2158. int delete;
  2159. xfs_ino_t first_ino;
  2160. ASSERT(ismrlocked(&ip->i_lock, MR_UPDATE));
  2161. ASSERT(ip->i_transp == tp);
  2162. ASSERT(ip->i_d.di_nlink == 0);
  2163. ASSERT(ip->i_d.di_nextents == 0);
  2164. ASSERT(ip->i_d.di_anextents == 0);
  2165. ASSERT((ip->i_d.di_size == 0) ||
  2166. ((ip->i_d.di_mode & S_IFMT) != S_IFREG));
  2167. ASSERT(ip->i_d.di_nblocks == 0);
  2168. /*
  2169. * Pull the on-disk inode from the AGI unlinked list.
  2170. */
  2171. error = xfs_iunlink_remove(tp, ip);
  2172. if (error != 0) {
  2173. return error;
  2174. }
  2175. error = xfs_difree(tp, ip->i_ino, flist, &delete, &first_ino);
  2176. if (error != 0) {
  2177. return error;
  2178. }
  2179. ip->i_d.di_mode = 0; /* mark incore inode as free */
  2180. ip->i_d.di_flags = 0;
  2181. ip->i_d.di_dmevmask = 0;
  2182. ip->i_d.di_forkoff = 0; /* mark the attr fork not in use */
  2183. ip->i_df.if_ext_max =
  2184. XFS_IFORK_DSIZE(ip) / (uint)sizeof(xfs_bmbt_rec_t);
  2185. ip->i_d.di_format = XFS_DINODE_FMT_EXTENTS;
  2186. ip->i_d.di_aformat = XFS_DINODE_FMT_EXTENTS;
  2187. /*
  2188. * Bump the generation count so no one will be confused
  2189. * by reincarnations of this inode.
  2190. */
  2191. ip->i_d.di_gen++;
  2192. xfs_trans_log_inode(tp, ip, XFS_ILOG_CORE);
  2193. if (delete) {
  2194. xfs_ifree_cluster(ip, tp, first_ino);
  2195. }
  2196. return 0;
  2197. }
  2198. /*
  2199. * Reallocate the space for if_broot based on the number of records
  2200. * being added or deleted as indicated in rec_diff. Move the records
  2201. * and pointers in if_broot to fit the new size. When shrinking this
  2202. * will eliminate holes between the records and pointers created by
  2203. * the caller. When growing this will create holes to be filled in
  2204. * by the caller.
  2205. *
  2206. * The caller must not request to add more records than would fit in
  2207. * the on-disk inode root. If the if_broot is currently NULL, then
  2208. * if we adding records one will be allocated. The caller must also
  2209. * not request that the number of records go below zero, although
  2210. * it can go to zero.
  2211. *
  2212. * ip -- the inode whose if_broot area is changing
  2213. * ext_diff -- the change in the number of records, positive or negative,
  2214. * requested for the if_broot array.
  2215. */
  2216. void
  2217. xfs_iroot_realloc(
  2218. xfs_inode_t *ip,
  2219. int rec_diff,
  2220. int whichfork)
  2221. {
  2222. int cur_max;
  2223. xfs_ifork_t *ifp;
  2224. xfs_bmbt_block_t *new_broot;
  2225. int new_max;
  2226. size_t new_size;
  2227. char *np;
  2228. char *op;
  2229. /*
  2230. * Handle the degenerate case quietly.
  2231. */
  2232. if (rec_diff == 0) {
  2233. return;
  2234. }
  2235. ifp = XFS_IFORK_PTR(ip, whichfork);
  2236. if (rec_diff > 0) {
  2237. /*
  2238. * If there wasn't any memory allocated before, just
  2239. * allocate it now and get out.
  2240. */
  2241. if (ifp->if_broot_bytes == 0) {
  2242. new_size = (size_t)XFS_BMAP_BROOT_SPACE_CALC(rec_diff);
  2243. ifp->if_broot = (xfs_bmbt_block_t*)kmem_alloc(new_size,
  2244. KM_SLEEP);
  2245. ifp->if_broot_bytes = (int)new_size;
  2246. return;
  2247. }
  2248. /*
  2249. * If there is already an existing if_broot, then we need
  2250. * to realloc() it and shift the pointers to their new
  2251. * location. The records don't change location because
  2252. * they are kept butted up against the btree block header.
  2253. */
  2254. cur_max = XFS_BMAP_BROOT_MAXRECS(ifp->if_broot_bytes);
  2255. new_max = cur_max + rec_diff;
  2256. new_size = (size_t)XFS_BMAP_BROOT_SPACE_CALC(new_max);
  2257. ifp->if_broot = (xfs_bmbt_block_t *)
  2258. kmem_realloc(ifp->if_broot,
  2259. new_size,
  2260. (size_t)XFS_BMAP_BROOT_SPACE_CALC(cur_max), /* old size */
  2261. KM_SLEEP);
  2262. op = (char *)XFS_BMAP_BROOT_PTR_ADDR(ifp->if_broot, 1,
  2263. ifp->if_broot_bytes);
  2264. np = (char *)XFS_BMAP_BROOT_PTR_ADDR(ifp->if_broot, 1,
  2265. (int)new_size);
  2266. ifp->if_broot_bytes = (int)new_size;
  2267. ASSERT(ifp->if_broot_bytes <=
  2268. XFS_IFORK_SIZE(ip, whichfork) + XFS_BROOT_SIZE_ADJ);
  2269. memmove(np, op, cur_max * (uint)sizeof(xfs_dfsbno_t));
  2270. return;
  2271. }
  2272. /*
  2273. * rec_diff is less than 0. In this case, we are shrinking the
  2274. * if_broot buffer. It must already exist. If we go to zero
  2275. * records, just get rid of the root and clear the status bit.
  2276. */
  2277. ASSERT((ifp->if_broot != NULL) && (ifp->if_broot_bytes > 0));
  2278. cur_max = XFS_BMAP_BROOT_MAXRECS(ifp->if_broot_bytes);
  2279. new_max = cur_max + rec_diff;
  2280. ASSERT(new_max >= 0);
  2281. if (new_max > 0)
  2282. new_size = (size_t)XFS_BMAP_BROOT_SPACE_CALC(new_max);
  2283. else
  2284. new_size = 0;
  2285. if (new_size > 0) {
  2286. new_broot = (xfs_bmbt_block_t *)kmem_alloc(new_size, KM_SLEEP);
  2287. /*
  2288. * First copy over the btree block header.
  2289. */
  2290. memcpy(new_broot, ifp->if_broot, sizeof(xfs_bmbt_block_t));
  2291. } else {
  2292. new_broot = NULL;
  2293. ifp->if_flags &= ~XFS_IFBROOT;
  2294. }
  2295. /*
  2296. * Only copy the records and pointers if there are any.
  2297. */
  2298. if (new_max > 0) {
  2299. /*
  2300. * First copy the records.
  2301. */
  2302. op = (char *)XFS_BMAP_BROOT_REC_ADDR(ifp->if_broot, 1,
  2303. ifp->if_broot_bytes);
  2304. np = (char *)XFS_BMAP_BROOT_REC_ADDR(new_broot, 1,
  2305. (int)new_size);
  2306. memcpy(np, op, new_max * (uint)sizeof(xfs_bmbt_rec_t));
  2307. /*
  2308. * Then copy the pointers.
  2309. */
  2310. op = (char *)XFS_BMAP_BROOT_PTR_ADDR(ifp->if_broot, 1,
  2311. ifp->if_broot_bytes);
  2312. np = (char *)XFS_BMAP_BROOT_PTR_ADDR(new_broot, 1,
  2313. (int)new_size);
  2314. memcpy(np, op, new_max * (uint)sizeof(xfs_dfsbno_t));
  2315. }
  2316. kmem_free(ifp->if_broot, ifp->if_broot_bytes);
  2317. ifp->if_broot = new_broot;
  2318. ifp->if_broot_bytes = (int)new_size;
  2319. ASSERT(ifp->if_broot_bytes <=
  2320. XFS_IFORK_SIZE(ip, whichfork) + XFS_BROOT_SIZE_ADJ);
  2321. return;
  2322. }
  2323. /*
  2324. * This is called when the amount of space needed for if_data
  2325. * is increased or decreased. The change in size is indicated by
  2326. * the number of bytes that need to be added or deleted in the
  2327. * byte_diff parameter.
  2328. *
  2329. * If the amount of space needed has decreased below the size of the
  2330. * inline buffer, then switch to using the inline buffer. Otherwise,
  2331. * use kmem_realloc() or kmem_alloc() to adjust the size of the buffer
  2332. * to what is needed.
  2333. *
  2334. * ip -- the inode whose if_data area is changing
  2335. * byte_diff -- the change in the number of bytes, positive or negative,
  2336. * requested for the if_data array.
  2337. */
  2338. void
  2339. xfs_idata_realloc(
  2340. xfs_inode_t *ip,
  2341. int byte_diff,
  2342. int whichfork)
  2343. {
  2344. xfs_ifork_t *ifp;
  2345. int new_size;
  2346. int real_size;
  2347. if (byte_diff == 0) {
  2348. return;
  2349. }
  2350. ifp = XFS_IFORK_PTR(ip, whichfork);
  2351. new_size = (int)ifp->if_bytes + byte_diff;
  2352. ASSERT(new_size >= 0);
  2353. if (new_size == 0) {
  2354. if (ifp->if_u1.if_data != ifp->if_u2.if_inline_data) {
  2355. kmem_free(ifp->if_u1.if_data, ifp->if_real_bytes);
  2356. }
  2357. ifp->if_u1.if_data = NULL;
  2358. real_size = 0;
  2359. } else if (new_size <= sizeof(ifp->if_u2.if_inline_data)) {
  2360. /*
  2361. * If the valid extents/data can fit in if_inline_ext/data,
  2362. * copy them from the malloc'd vector and free it.
  2363. */
  2364. if (ifp->if_u1.if_data == NULL) {
  2365. ifp->if_u1.if_data = ifp->if_u2.if_inline_data;
  2366. } else if (ifp->if_u1.if_data != ifp->if_u2.if_inline_data) {
  2367. ASSERT(ifp->if_real_bytes != 0);
  2368. memcpy(ifp->if_u2.if_inline_data, ifp->if_u1.if_data,
  2369. new_size);
  2370. kmem_free(ifp->if_u1.if_data, ifp->if_real_bytes);
  2371. ifp->if_u1.if_data = ifp->if_u2.if_inline_data;
  2372. }
  2373. real_size = 0;
  2374. } else {
  2375. /*
  2376. * Stuck with malloc/realloc.
  2377. * For inline data, the underlying buffer must be
  2378. * a multiple of 4 bytes in size so that it can be
  2379. * logged and stay on word boundaries. We enforce
  2380. * that here.
  2381. */
  2382. real_size = roundup(new_size, 4);
  2383. if (ifp->if_u1.if_data == NULL) {
  2384. ASSERT(ifp->if_real_bytes == 0);
  2385. ifp->if_u1.if_data = kmem_alloc(real_size, KM_SLEEP);
  2386. } else if (ifp->if_u1.if_data != ifp->if_u2.if_inline_data) {
  2387. /*
  2388. * Only do the realloc if the underlying size
  2389. * is really changing.
  2390. */
  2391. if (ifp->if_real_bytes != real_size) {
  2392. ifp->if_u1.if_data =
  2393. kmem_realloc(ifp->if_u1.if_data,
  2394. real_size,
  2395. ifp->if_real_bytes,
  2396. KM_SLEEP);
  2397. }
  2398. } else {
  2399. ASSERT(ifp->if_real_bytes == 0);
  2400. ifp->if_u1.if_data = kmem_alloc(real_size, KM_SLEEP);
  2401. memcpy(ifp->if_u1.if_data, ifp->if_u2.if_inline_data,
  2402. ifp->if_bytes);
  2403. }
  2404. }
  2405. ifp->if_real_bytes = real_size;
  2406. ifp->if_bytes = new_size;
  2407. ASSERT(ifp->if_bytes <= XFS_IFORK_SIZE(ip, whichfork));
  2408. }
  2409. /*
  2410. * Map inode to disk block and offset.
  2411. *
  2412. * mp -- the mount point structure for the current file system
  2413. * tp -- the current transaction
  2414. * ino -- the inode number of the inode to be located
  2415. * imap -- this structure is filled in with the information necessary
  2416. * to retrieve the given inode from disk
  2417. * flags -- flags to pass to xfs_dilocate indicating whether or not
  2418. * lookups in the inode btree were OK or not
  2419. */
  2420. int
  2421. xfs_imap(
  2422. xfs_mount_t *mp,
  2423. xfs_trans_t *tp,
  2424. xfs_ino_t ino,
  2425. xfs_imap_t *imap,
  2426. uint flags)
  2427. {
  2428. xfs_fsblock_t fsbno;
  2429. int len;
  2430. int off;
  2431. int error;
  2432. fsbno = imap->im_blkno ?
  2433. XFS_DADDR_TO_FSB(mp, imap->im_blkno) : NULLFSBLOCK;
  2434. error = xfs_dilocate(mp, tp, ino, &fsbno, &len, &off, flags);
  2435. if (error != 0) {
  2436. return error;
  2437. }
  2438. imap->im_blkno = XFS_FSB_TO_DADDR(mp, fsbno);
  2439. imap->im_len = XFS_FSB_TO_BB(mp, len);
  2440. imap->im_agblkno = XFS_FSB_TO_AGBNO(mp, fsbno);
  2441. imap->im_ioffset = (ushort)off;
  2442. imap->im_boffset = (ushort)(off << mp->m_sb.sb_inodelog);
  2443. return 0;
  2444. }
  2445. void
  2446. xfs_idestroy_fork(
  2447. xfs_inode_t *ip,
  2448. int whichfork)
  2449. {
  2450. xfs_ifork_t *ifp;
  2451. ifp = XFS_IFORK_PTR(ip, whichfork);
  2452. if (ifp->if_broot != NULL) {
  2453. kmem_free(ifp->if_broot, ifp->if_broot_bytes);
  2454. ifp->if_broot = NULL;
  2455. }
  2456. /*
  2457. * If the format is local, then we can't have an extents
  2458. * array so just look for an inline data array. If we're
  2459. * not local then we may or may not have an extents list,
  2460. * so check and free it up if we do.
  2461. */
  2462. if (XFS_IFORK_FORMAT(ip, whichfork) == XFS_DINODE_FMT_LOCAL) {
  2463. if ((ifp->if_u1.if_data != ifp->if_u2.if_inline_data) &&
  2464. (ifp->if_u1.if_data != NULL)) {
  2465. ASSERT(ifp->if_real_bytes != 0);
  2466. kmem_free(ifp->if_u1.if_data, ifp->if_real_bytes);
  2467. ifp->if_u1.if_data = NULL;
  2468. ifp->if_real_bytes = 0;
  2469. }
  2470. } else if ((ifp->if_flags & XFS_IFEXTENTS) &&
  2471. ((ifp->if_flags & XFS_IFEXTIREC) ||
  2472. ((ifp->if_u1.if_extents != NULL) &&
  2473. (ifp->if_u1.if_extents != ifp->if_u2.if_inline_ext)))) {
  2474. ASSERT(ifp->if_real_bytes != 0);
  2475. xfs_iext_destroy(ifp);
  2476. }
  2477. ASSERT(ifp->if_u1.if_extents == NULL ||
  2478. ifp->if_u1.if_extents == ifp->if_u2.if_inline_ext);
  2479. ASSERT(ifp->if_real_bytes == 0);
  2480. if (whichfork == XFS_ATTR_FORK) {
  2481. kmem_zone_free(xfs_ifork_zone, ip->i_afp);
  2482. ip->i_afp = NULL;
  2483. }
  2484. }
  2485. /*
  2486. * This is called free all the memory associated with an inode.
  2487. * It must free the inode itself and any buffers allocated for
  2488. * if_extents/if_data and if_broot. It must also free the lock
  2489. * associated with the inode.
  2490. */
  2491. void
  2492. xfs_idestroy(
  2493. xfs_inode_t *ip)
  2494. {
  2495. switch (ip->i_d.di_mode & S_IFMT) {
  2496. case S_IFREG:
  2497. case S_IFDIR:
  2498. case S_IFLNK:
  2499. xfs_idestroy_fork(ip, XFS_DATA_FORK);
  2500. break;
  2501. }
  2502. if (ip->i_afp)
  2503. xfs_idestroy_fork(ip, XFS_ATTR_FORK);
  2504. mrfree(&ip->i_lock);
  2505. mrfree(&ip->i_iolock);
  2506. freesema(&ip->i_flock);
  2507. #ifdef XFS_BMAP_TRACE
  2508. ktrace_free(ip->i_xtrace);
  2509. #endif
  2510. #ifdef XFS_BMBT_TRACE
  2511. ktrace_free(ip->i_btrace);
  2512. #endif
  2513. #ifdef XFS_RW_TRACE
  2514. ktrace_free(ip->i_rwtrace);
  2515. #endif
  2516. #ifdef XFS_ILOCK_TRACE
  2517. ktrace_free(ip->i_lock_trace);
  2518. #endif
  2519. #ifdef XFS_DIR2_TRACE
  2520. ktrace_free(ip->i_dir_trace);
  2521. #endif
  2522. if (ip->i_itemp) {
  2523. /* XXXdpd should be able to assert this but shutdown
  2524. * is leaving the AIL behind. */
  2525. ASSERT(((ip->i_itemp->ili_item.li_flags & XFS_LI_IN_AIL) == 0) ||
  2526. XFS_FORCED_SHUTDOWN(ip->i_mount));
  2527. xfs_inode_item_destroy(ip);
  2528. }
  2529. kmem_zone_free(xfs_inode_zone, ip);
  2530. }
  2531. /*
  2532. * Increment the pin count of the given buffer.
  2533. * This value is protected by ipinlock spinlock in the mount structure.
  2534. */
  2535. void
  2536. xfs_ipin(
  2537. xfs_inode_t *ip)
  2538. {
  2539. ASSERT(ismrlocked(&ip->i_lock, MR_UPDATE));
  2540. atomic_inc(&ip->i_pincount);
  2541. }
  2542. /*
  2543. * Decrement the pin count of the given inode, and wake up
  2544. * anyone in xfs_iwait_unpin() if the count goes to 0. The
  2545. * inode must have been previously pinned with a call to xfs_ipin().
  2546. */
  2547. void
  2548. xfs_iunpin(
  2549. xfs_inode_t *ip)
  2550. {
  2551. ASSERT(atomic_read(&ip->i_pincount) > 0);
  2552. if (atomic_dec_and_test(&ip->i_pincount)) {
  2553. /*
  2554. * If the inode is currently being reclaimed, the
  2555. * linux inode _and_ the xfs vnode may have been
  2556. * freed so we cannot reference either of them safely.
  2557. * Hence we should not try to do anything to them
  2558. * if the xfs inode is currently in the reclaim
  2559. * path.
  2560. *
  2561. * However, we still need to issue the unpin wakeup
  2562. * call as the inode reclaim may be blocked waiting for
  2563. * the inode to become unpinned.
  2564. */
  2565. if (!(ip->i_flags & (XFS_IRECLAIM|XFS_IRECLAIMABLE))) {
  2566. bhv_vnode_t *vp = XFS_ITOV_NULL(ip);
  2567. /* make sync come back and flush this inode */
  2568. if (vp) {
  2569. struct inode *inode = vn_to_inode(vp);
  2570. if (!(inode->i_state &
  2571. (I_NEW|I_FREEING|I_CLEAR)))
  2572. mark_inode_dirty_sync(inode);
  2573. }
  2574. }
  2575. wake_up(&ip->i_ipin_wait);
  2576. }
  2577. }
  2578. /*
  2579. * This is called to wait for the given inode to be unpinned.
  2580. * It will sleep until this happens. The caller must have the
  2581. * inode locked in at least shared mode so that the buffer cannot
  2582. * be subsequently pinned once someone is waiting for it to be
  2583. * unpinned.
  2584. */
  2585. STATIC void
  2586. xfs_iunpin_wait(
  2587. xfs_inode_t *ip)
  2588. {
  2589. xfs_inode_log_item_t *iip;
  2590. xfs_lsn_t lsn;
  2591. ASSERT(ismrlocked(&ip->i_lock, MR_UPDATE | MR_ACCESS));
  2592. if (atomic_read(&ip->i_pincount) == 0) {
  2593. return;
  2594. }
  2595. iip = ip->i_itemp;
  2596. if (iip && iip->ili_last_lsn) {
  2597. lsn = iip->ili_last_lsn;
  2598. } else {
  2599. lsn = (xfs_lsn_t)0;
  2600. }
  2601. /*
  2602. * Give the log a push so we don't wait here too long.
  2603. */
  2604. xfs_log_force(ip->i_mount, lsn, XFS_LOG_FORCE);
  2605. wait_event(ip->i_ipin_wait, (atomic_read(&ip->i_pincount) == 0));
  2606. }
  2607. /*
  2608. * xfs_iextents_copy()
  2609. *
  2610. * This is called to copy the REAL extents (as opposed to the delayed
  2611. * allocation extents) from the inode into the given buffer. It
  2612. * returns the number of bytes copied into the buffer.
  2613. *
  2614. * If there are no delayed allocation extents, then we can just
  2615. * memcpy() the extents into the buffer. Otherwise, we need to
  2616. * examine each extent in turn and skip those which are delayed.
  2617. */
  2618. int
  2619. xfs_iextents_copy(
  2620. xfs_inode_t *ip,
  2621. xfs_bmbt_rec_t *buffer,
  2622. int whichfork)
  2623. {
  2624. int copied;
  2625. xfs_bmbt_rec_t *dest_ep;
  2626. xfs_bmbt_rec_t *ep;
  2627. #ifdef XFS_BMAP_TRACE
  2628. static char fname[] = "xfs_iextents_copy";
  2629. #endif
  2630. int i;
  2631. xfs_ifork_t *ifp;
  2632. int nrecs;
  2633. xfs_fsblock_t start_block;
  2634. ifp = XFS_IFORK_PTR(ip, whichfork);
  2635. ASSERT(ismrlocked(&ip->i_lock, MR_UPDATE|MR_ACCESS));
  2636. ASSERT(ifp->if_bytes > 0);
  2637. nrecs = ifp->if_bytes / (uint)sizeof(xfs_bmbt_rec_t);
  2638. xfs_bmap_trace_exlist(fname, ip, nrecs, whichfork);
  2639. ASSERT(nrecs > 0);
  2640. /*
  2641. * There are some delayed allocation extents in the
  2642. * inode, so copy the extents one at a time and skip
  2643. * the delayed ones. There must be at least one
  2644. * non-delayed extent.
  2645. */
  2646. dest_ep = buffer;
  2647. copied = 0;
  2648. for (i = 0; i < nrecs; i++) {
  2649. ep = xfs_iext_get_ext(ifp, i);
  2650. start_block = xfs_bmbt_get_startblock(ep);
  2651. if (ISNULLSTARTBLOCK(start_block)) {
  2652. /*
  2653. * It's a delayed allocation extent, so skip it.
  2654. */
  2655. continue;
  2656. }
  2657. /* Translate to on disk format */
  2658. put_unaligned(INT_GET(ep->l0, ARCH_CONVERT),
  2659. (__uint64_t*)&dest_ep->l0);
  2660. put_unaligned(INT_GET(ep->l1, ARCH_CONVERT),
  2661. (__uint64_t*)&dest_ep->l1);
  2662. dest_ep++;
  2663. copied++;
  2664. }
  2665. ASSERT(copied != 0);
  2666. xfs_validate_extents(ifp, copied, 1, 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. if (whichfork == XFS_DATA_FORK) {
  2722. if (unlikely(XFS_DIR_SHORTFORM_VALIDATE_ONDISK(mp, dip))) {
  2723. XFS_ERROR_REPORT("xfs_iflush_fork",
  2724. XFS_ERRLEVEL_LOW, mp);
  2725. return XFS_ERROR(EFSCORRUPTED);
  2726. }
  2727. }
  2728. break;
  2729. case XFS_DINODE_FMT_EXTENTS:
  2730. ASSERT((ifp->if_flags & XFS_IFEXTENTS) ||
  2731. !(iip->ili_format.ilf_fields & extflag[whichfork]));
  2732. ASSERT((xfs_iext_get_ext(ifp, 0) != NULL) ||
  2733. (ifp->if_bytes == 0));
  2734. ASSERT((xfs_iext_get_ext(ifp, 0) == NULL) ||
  2735. (ifp->if_bytes > 0));
  2736. if ((iip->ili_format.ilf_fields & extflag[whichfork]) &&
  2737. (ifp->if_bytes > 0)) {
  2738. ASSERT(XFS_IFORK_NEXTENTS(ip, whichfork) > 0);
  2739. (void)xfs_iextents_copy(ip, (xfs_bmbt_rec_t *)cp,
  2740. whichfork);
  2741. }
  2742. break;
  2743. case XFS_DINODE_FMT_BTREE:
  2744. if ((iip->ili_format.ilf_fields & brootflag[whichfork]) &&
  2745. (ifp->if_broot_bytes > 0)) {
  2746. ASSERT(ifp->if_broot != NULL);
  2747. ASSERT(ifp->if_broot_bytes <=
  2748. (XFS_IFORK_SIZE(ip, whichfork) +
  2749. XFS_BROOT_SIZE_ADJ));
  2750. xfs_bmbt_to_bmdr(ifp->if_broot, ifp->if_broot_bytes,
  2751. (xfs_bmdr_block_t *)cp,
  2752. XFS_DFORK_SIZE(dip, mp, whichfork));
  2753. }
  2754. break;
  2755. case XFS_DINODE_FMT_DEV:
  2756. if (iip->ili_format.ilf_fields & XFS_ILOG_DEV) {
  2757. ASSERT(whichfork == XFS_DATA_FORK);
  2758. INT_SET(dip->di_u.di_dev, ARCH_CONVERT, ip->i_df.if_u2.if_rdev);
  2759. }
  2760. break;
  2761. case XFS_DINODE_FMT_UUID:
  2762. if (iip->ili_format.ilf_fields & XFS_ILOG_UUID) {
  2763. ASSERT(whichfork == XFS_DATA_FORK);
  2764. memcpy(&dip->di_u.di_muuid, &ip->i_df.if_u2.if_uuid,
  2765. sizeof(uuid_t));
  2766. }
  2767. break;
  2768. default:
  2769. ASSERT(0);
  2770. break;
  2771. }
  2772. return 0;
  2773. }
  2774. /*
  2775. * xfs_iflush() will write a modified inode's changes out to the
  2776. * inode's on disk home. The caller must have the inode lock held
  2777. * in at least shared mode and the inode flush semaphore must be
  2778. * held as well. The inode lock will still be held upon return from
  2779. * the call and the caller is free to unlock it.
  2780. * The inode flush lock will be unlocked when the inode reaches the disk.
  2781. * The flags indicate how the inode's buffer should be written out.
  2782. */
  2783. int
  2784. xfs_iflush(
  2785. xfs_inode_t *ip,
  2786. uint flags)
  2787. {
  2788. xfs_inode_log_item_t *iip;
  2789. xfs_buf_t *bp;
  2790. xfs_dinode_t *dip;
  2791. xfs_mount_t *mp;
  2792. int error;
  2793. /* REFERENCED */
  2794. xfs_chash_t *ch;
  2795. xfs_inode_t *iq;
  2796. int clcount; /* count of inodes clustered */
  2797. int bufwasdelwri;
  2798. enum { INT_DELWRI = (1 << 0), INT_ASYNC = (1 << 1) };
  2799. SPLDECL(s);
  2800. XFS_STATS_INC(xs_iflush_count);
  2801. ASSERT(ismrlocked(&ip->i_lock, MR_UPDATE|MR_ACCESS));
  2802. ASSERT(issemalocked(&(ip->i_flock)));
  2803. ASSERT(ip->i_d.di_format != XFS_DINODE_FMT_BTREE ||
  2804. ip->i_d.di_nextents > ip->i_df.if_ext_max);
  2805. iip = ip->i_itemp;
  2806. mp = ip->i_mount;
  2807. /*
  2808. * If the inode isn't dirty, then just release the inode
  2809. * flush lock and do nothing.
  2810. */
  2811. if ((ip->i_update_core == 0) &&
  2812. ((iip == NULL) || !(iip->ili_format.ilf_fields & XFS_ILOG_ALL))) {
  2813. ASSERT((iip != NULL) ?
  2814. !(iip->ili_item.li_flags & XFS_LI_IN_AIL) : 1);
  2815. xfs_ifunlock(ip);
  2816. return 0;
  2817. }
  2818. /*
  2819. * We can't flush the inode until it is unpinned, so
  2820. * wait for it. We know noone new can pin it, because
  2821. * we are holding the inode lock shared and you need
  2822. * to hold it exclusively to pin the inode.
  2823. */
  2824. xfs_iunpin_wait(ip);
  2825. /*
  2826. * This may have been unpinned because the filesystem is shutting
  2827. * down forcibly. If that's the case we must not write this inode
  2828. * to disk, because the log record didn't make it to disk!
  2829. */
  2830. if (XFS_FORCED_SHUTDOWN(mp)) {
  2831. ip->i_update_core = 0;
  2832. if (iip)
  2833. iip->ili_format.ilf_fields = 0;
  2834. xfs_ifunlock(ip);
  2835. return XFS_ERROR(EIO);
  2836. }
  2837. /*
  2838. * Get the buffer containing the on-disk inode.
  2839. */
  2840. error = xfs_itobp(mp, NULL, ip, &dip, &bp, 0, 0);
  2841. if (error) {
  2842. xfs_ifunlock(ip);
  2843. return error;
  2844. }
  2845. /*
  2846. * Decide how buffer will be flushed out. This is done before
  2847. * the call to xfs_iflush_int because this field is zeroed by it.
  2848. */
  2849. if (iip != NULL && iip->ili_format.ilf_fields != 0) {
  2850. /*
  2851. * Flush out the inode buffer according to the directions
  2852. * of the caller. In the cases where the caller has given
  2853. * us a choice choose the non-delwri case. This is because
  2854. * the inode is in the AIL and we need to get it out soon.
  2855. */
  2856. switch (flags) {
  2857. case XFS_IFLUSH_SYNC:
  2858. case XFS_IFLUSH_DELWRI_ELSE_SYNC:
  2859. flags = 0;
  2860. break;
  2861. case XFS_IFLUSH_ASYNC:
  2862. case XFS_IFLUSH_DELWRI_ELSE_ASYNC:
  2863. flags = INT_ASYNC;
  2864. break;
  2865. case XFS_IFLUSH_DELWRI:
  2866. flags = INT_DELWRI;
  2867. break;
  2868. default:
  2869. ASSERT(0);
  2870. flags = 0;
  2871. break;
  2872. }
  2873. } else {
  2874. switch (flags) {
  2875. case XFS_IFLUSH_DELWRI_ELSE_SYNC:
  2876. case XFS_IFLUSH_DELWRI_ELSE_ASYNC:
  2877. case XFS_IFLUSH_DELWRI:
  2878. flags = INT_DELWRI;
  2879. break;
  2880. case XFS_IFLUSH_ASYNC:
  2881. flags = INT_ASYNC;
  2882. break;
  2883. case XFS_IFLUSH_SYNC:
  2884. flags = 0;
  2885. break;
  2886. default:
  2887. ASSERT(0);
  2888. flags = 0;
  2889. break;
  2890. }
  2891. }
  2892. /*
  2893. * First flush out the inode that xfs_iflush was called with.
  2894. */
  2895. error = xfs_iflush_int(ip, bp);
  2896. if (error) {
  2897. goto corrupt_out;
  2898. }
  2899. /*
  2900. * inode clustering:
  2901. * see if other inodes can be gathered into this write
  2902. */
  2903. ip->i_chash->chl_buf = bp;
  2904. ch = XFS_CHASH(mp, ip->i_blkno);
  2905. s = mutex_spinlock(&ch->ch_lock);
  2906. clcount = 0;
  2907. for (iq = ip->i_cnext; iq != ip; iq = iq->i_cnext) {
  2908. /*
  2909. * Do an un-protected check to see if the inode is dirty and
  2910. * is a candidate for flushing. These checks will be repeated
  2911. * later after the appropriate locks are acquired.
  2912. */
  2913. iip = iq->i_itemp;
  2914. if ((iq->i_update_core == 0) &&
  2915. ((iip == NULL) ||
  2916. !(iip->ili_format.ilf_fields & XFS_ILOG_ALL)) &&
  2917. xfs_ipincount(iq) == 0) {
  2918. continue;
  2919. }
  2920. /*
  2921. * Try to get locks. If any are unavailable,
  2922. * then this inode cannot be flushed and is skipped.
  2923. */
  2924. /* get inode locks (just i_lock) */
  2925. if (xfs_ilock_nowait(iq, XFS_ILOCK_SHARED)) {
  2926. /* get inode flush lock */
  2927. if (xfs_iflock_nowait(iq)) {
  2928. /* check if pinned */
  2929. if (xfs_ipincount(iq) == 0) {
  2930. /* arriving here means that
  2931. * this inode can be flushed.
  2932. * first re-check that it's
  2933. * dirty
  2934. */
  2935. iip = iq->i_itemp;
  2936. if ((iq->i_update_core != 0)||
  2937. ((iip != NULL) &&
  2938. (iip->ili_format.ilf_fields & XFS_ILOG_ALL))) {
  2939. clcount++;
  2940. error = xfs_iflush_int(iq, bp);
  2941. if (error) {
  2942. xfs_iunlock(iq,
  2943. XFS_ILOCK_SHARED);
  2944. goto cluster_corrupt_out;
  2945. }
  2946. } else {
  2947. xfs_ifunlock(iq);
  2948. }
  2949. } else {
  2950. xfs_ifunlock(iq);
  2951. }
  2952. }
  2953. xfs_iunlock(iq, XFS_ILOCK_SHARED);
  2954. }
  2955. }
  2956. mutex_spinunlock(&ch->ch_lock, s);
  2957. if (clcount) {
  2958. XFS_STATS_INC(xs_icluster_flushcnt);
  2959. XFS_STATS_ADD(xs_icluster_flushinode, clcount);
  2960. }
  2961. /*
  2962. * If the buffer is pinned then push on the log so we won't
  2963. * get stuck waiting in the write for too long.
  2964. */
  2965. if (XFS_BUF_ISPINNED(bp)){
  2966. xfs_log_force(mp, (xfs_lsn_t)0, XFS_LOG_FORCE);
  2967. }
  2968. if (flags & INT_DELWRI) {
  2969. xfs_bdwrite(mp, bp);
  2970. } else if (flags & INT_ASYNC) {
  2971. xfs_bawrite(mp, bp);
  2972. } else {
  2973. error = xfs_bwrite(mp, bp);
  2974. }
  2975. return error;
  2976. corrupt_out:
  2977. xfs_buf_relse(bp);
  2978. xfs_force_shutdown(mp, SHUTDOWN_CORRUPT_INCORE);
  2979. xfs_iflush_abort(ip);
  2980. /*
  2981. * Unlocks the flush lock
  2982. */
  2983. return XFS_ERROR(EFSCORRUPTED);
  2984. cluster_corrupt_out:
  2985. /* Corruption detected in the clustering loop. Invalidate the
  2986. * inode buffer and shut down the filesystem.
  2987. */
  2988. mutex_spinunlock(&ch->ch_lock, s);
  2989. /*
  2990. * Clean up the buffer. If it was B_DELWRI, just release it --
  2991. * brelse can handle it with no problems. If not, shut down the
  2992. * filesystem before releasing the buffer.
  2993. */
  2994. if ((bufwasdelwri= XFS_BUF_ISDELAYWRITE(bp))) {
  2995. xfs_buf_relse(bp);
  2996. }
  2997. xfs_force_shutdown(mp, SHUTDOWN_CORRUPT_INCORE);
  2998. if(!bufwasdelwri) {
  2999. /*
  3000. * Just like incore_relse: if we have b_iodone functions,
  3001. * mark the buffer as an error and call them. Otherwise
  3002. * mark it as stale and brelse.
  3003. */
  3004. if (XFS_BUF_IODONE_FUNC(bp)) {
  3005. XFS_BUF_CLR_BDSTRAT_FUNC(bp);
  3006. XFS_BUF_UNDONE(bp);
  3007. XFS_BUF_STALE(bp);
  3008. XFS_BUF_SHUT(bp);
  3009. XFS_BUF_ERROR(bp,EIO);
  3010. xfs_biodone(bp);
  3011. } else {
  3012. XFS_BUF_STALE(bp);
  3013. xfs_buf_relse(bp);
  3014. }
  3015. }
  3016. xfs_iflush_abort(iq);
  3017. /*
  3018. * Unlocks the flush lock
  3019. */
  3020. return XFS_ERROR(EFSCORRUPTED);
  3021. }
  3022. STATIC int
  3023. xfs_iflush_int(
  3024. xfs_inode_t *ip,
  3025. xfs_buf_t *bp)
  3026. {
  3027. xfs_inode_log_item_t *iip;
  3028. xfs_dinode_t *dip;
  3029. xfs_mount_t *mp;
  3030. #ifdef XFS_TRANS_DEBUG
  3031. int first;
  3032. #endif
  3033. SPLDECL(s);
  3034. ASSERT(ismrlocked(&ip->i_lock, MR_UPDATE|MR_ACCESS));
  3035. ASSERT(issemalocked(&(ip->i_flock)));
  3036. ASSERT(ip->i_d.di_format != XFS_DINODE_FMT_BTREE ||
  3037. ip->i_d.di_nextents > ip->i_df.if_ext_max);
  3038. iip = ip->i_itemp;
  3039. mp = ip->i_mount;
  3040. /*
  3041. * If the inode isn't dirty, then just release the inode
  3042. * flush lock and do nothing.
  3043. */
  3044. if ((ip->i_update_core == 0) &&
  3045. ((iip == NULL) || !(iip->ili_format.ilf_fields & XFS_ILOG_ALL))) {
  3046. xfs_ifunlock(ip);
  3047. return 0;
  3048. }
  3049. /* set *dip = inode's place in the buffer */
  3050. dip = (xfs_dinode_t *)xfs_buf_offset(bp, ip->i_boffset);
  3051. /*
  3052. * Clear i_update_core before copying out the data.
  3053. * This is for coordination with our timestamp updates
  3054. * that don't hold the inode lock. They will always
  3055. * update the timestamps BEFORE setting i_update_core,
  3056. * so if we clear i_update_core after they set it we
  3057. * are guaranteed to see their updates to the timestamps.
  3058. * I believe that this depends on strongly ordered memory
  3059. * semantics, but we have that. We use the SYNCHRONIZE
  3060. * macro to make sure that the compiler does not reorder
  3061. * the i_update_core access below the data copy below.
  3062. */
  3063. ip->i_update_core = 0;
  3064. SYNCHRONIZE();
  3065. /*
  3066. * Make sure to get the latest atime from the Linux inode.
  3067. */
  3068. xfs_synchronize_atime(ip);
  3069. if (XFS_TEST_ERROR(INT_GET(dip->di_core.di_magic,ARCH_CONVERT) != XFS_DINODE_MAGIC,
  3070. mp, XFS_ERRTAG_IFLUSH_1, XFS_RANDOM_IFLUSH_1)) {
  3071. xfs_cmn_err(XFS_PTAG_IFLUSH, CE_ALERT, mp,
  3072. "xfs_iflush: Bad inode %Lu magic number 0x%x, ptr 0x%p",
  3073. ip->i_ino, (int) INT_GET(dip->di_core.di_magic, ARCH_CONVERT), dip);
  3074. goto corrupt_out;
  3075. }
  3076. if (XFS_TEST_ERROR(ip->i_d.di_magic != XFS_DINODE_MAGIC,
  3077. mp, XFS_ERRTAG_IFLUSH_2, XFS_RANDOM_IFLUSH_2)) {
  3078. xfs_cmn_err(XFS_PTAG_IFLUSH, CE_ALERT, mp,
  3079. "xfs_iflush: Bad inode %Lu, ptr 0x%p, magic number 0x%x",
  3080. ip->i_ino, ip, ip->i_d.di_magic);
  3081. goto corrupt_out;
  3082. }
  3083. if ((ip->i_d.di_mode & S_IFMT) == S_IFREG) {
  3084. if (XFS_TEST_ERROR(
  3085. (ip->i_d.di_format != XFS_DINODE_FMT_EXTENTS) &&
  3086. (ip->i_d.di_format != XFS_DINODE_FMT_BTREE),
  3087. mp, XFS_ERRTAG_IFLUSH_3, XFS_RANDOM_IFLUSH_3)) {
  3088. xfs_cmn_err(XFS_PTAG_IFLUSH, CE_ALERT, mp,
  3089. "xfs_iflush: Bad regular inode %Lu, ptr 0x%p",
  3090. ip->i_ino, ip);
  3091. goto corrupt_out;
  3092. }
  3093. } else if ((ip->i_d.di_mode & S_IFMT) == S_IFDIR) {
  3094. if (XFS_TEST_ERROR(
  3095. (ip->i_d.di_format != XFS_DINODE_FMT_EXTENTS) &&
  3096. (ip->i_d.di_format != XFS_DINODE_FMT_BTREE) &&
  3097. (ip->i_d.di_format != XFS_DINODE_FMT_LOCAL),
  3098. mp, XFS_ERRTAG_IFLUSH_4, XFS_RANDOM_IFLUSH_4)) {
  3099. xfs_cmn_err(XFS_PTAG_IFLUSH, CE_ALERT, mp,
  3100. "xfs_iflush: Bad directory inode %Lu, ptr 0x%p",
  3101. ip->i_ino, ip);
  3102. goto corrupt_out;
  3103. }
  3104. }
  3105. if (XFS_TEST_ERROR(ip->i_d.di_nextents + ip->i_d.di_anextents >
  3106. ip->i_d.di_nblocks, mp, XFS_ERRTAG_IFLUSH_5,
  3107. XFS_RANDOM_IFLUSH_5)) {
  3108. xfs_cmn_err(XFS_PTAG_IFLUSH, CE_ALERT, mp,
  3109. "xfs_iflush: detected corrupt incore inode %Lu, total extents = %d, nblocks = %Ld, ptr 0x%p",
  3110. ip->i_ino,
  3111. ip->i_d.di_nextents + ip->i_d.di_anextents,
  3112. ip->i_d.di_nblocks,
  3113. ip);
  3114. goto corrupt_out;
  3115. }
  3116. if (XFS_TEST_ERROR(ip->i_d.di_forkoff > mp->m_sb.sb_inodesize,
  3117. mp, XFS_ERRTAG_IFLUSH_6, XFS_RANDOM_IFLUSH_6)) {
  3118. xfs_cmn_err(XFS_PTAG_IFLUSH, CE_ALERT, mp,
  3119. "xfs_iflush: bad inode %Lu, forkoff 0x%x, ptr 0x%p",
  3120. ip->i_ino, ip->i_d.di_forkoff, ip);
  3121. goto corrupt_out;
  3122. }
  3123. /*
  3124. * bump the flush iteration count, used to detect flushes which
  3125. * postdate a log record during recovery.
  3126. */
  3127. ip->i_d.di_flushiter++;
  3128. /*
  3129. * Copy the dirty parts of the inode into the on-disk
  3130. * inode. We always copy out the core of the inode,
  3131. * because if the inode is dirty at all the core must
  3132. * be.
  3133. */
  3134. xfs_xlate_dinode_core((xfs_caddr_t)&(dip->di_core), &(ip->i_d), -1);
  3135. /* Wrap, we never let the log put out DI_MAX_FLUSH */
  3136. if (ip->i_d.di_flushiter == DI_MAX_FLUSH)
  3137. ip->i_d.di_flushiter = 0;
  3138. /*
  3139. * If this is really an old format inode and the superblock version
  3140. * has not been updated to support only new format inodes, then
  3141. * convert back to the old inode format. If the superblock version
  3142. * has been updated, then make the conversion permanent.
  3143. */
  3144. ASSERT(ip->i_d.di_version == XFS_DINODE_VERSION_1 ||
  3145. XFS_SB_VERSION_HASNLINK(&mp->m_sb));
  3146. if (ip->i_d.di_version == XFS_DINODE_VERSION_1) {
  3147. if (!XFS_SB_VERSION_HASNLINK(&mp->m_sb)) {
  3148. /*
  3149. * Convert it back.
  3150. */
  3151. ASSERT(ip->i_d.di_nlink <= XFS_MAXLINK_1);
  3152. INT_SET(dip->di_core.di_onlink, ARCH_CONVERT, ip->i_d.di_nlink);
  3153. } else {
  3154. /*
  3155. * The superblock version has already been bumped,
  3156. * so just make the conversion to the new inode
  3157. * format permanent.
  3158. */
  3159. ip->i_d.di_version = XFS_DINODE_VERSION_2;
  3160. INT_SET(dip->di_core.di_version, ARCH_CONVERT, XFS_DINODE_VERSION_2);
  3161. ip->i_d.di_onlink = 0;
  3162. dip->di_core.di_onlink = 0;
  3163. memset(&(ip->i_d.di_pad[0]), 0, sizeof(ip->i_d.di_pad));
  3164. memset(&(dip->di_core.di_pad[0]), 0,
  3165. sizeof(dip->di_core.di_pad));
  3166. ASSERT(ip->i_d.di_projid == 0);
  3167. }
  3168. }
  3169. if (xfs_iflush_fork(ip, dip, iip, XFS_DATA_FORK, bp) == EFSCORRUPTED) {
  3170. goto corrupt_out;
  3171. }
  3172. if (XFS_IFORK_Q(ip)) {
  3173. /*
  3174. * The only error from xfs_iflush_fork is on the data fork.
  3175. */
  3176. (void) xfs_iflush_fork(ip, dip, iip, XFS_ATTR_FORK, bp);
  3177. }
  3178. xfs_inobp_check(mp, bp);
  3179. /*
  3180. * We've recorded everything logged in the inode, so we'd
  3181. * like to clear the ilf_fields bits so we don't log and
  3182. * flush things unnecessarily. However, we can't stop
  3183. * logging all this information until the data we've copied
  3184. * into the disk buffer is written to disk. If we did we might
  3185. * overwrite the copy of the inode in the log with all the
  3186. * data after re-logging only part of it, and in the face of
  3187. * a crash we wouldn't have all the data we need to recover.
  3188. *
  3189. * What we do is move the bits to the ili_last_fields field.
  3190. * When logging the inode, these bits are moved back to the
  3191. * ilf_fields field. In the xfs_iflush_done() routine we
  3192. * clear ili_last_fields, since we know that the information
  3193. * those bits represent is permanently on disk. As long as
  3194. * the flush completes before the inode is logged again, then
  3195. * both ilf_fields and ili_last_fields will be cleared.
  3196. *
  3197. * We can play with the ilf_fields bits here, because the inode
  3198. * lock must be held exclusively in order to set bits there
  3199. * and the flush lock protects the ili_last_fields bits.
  3200. * Set ili_logged so the flush done
  3201. * routine can tell whether or not to look in the AIL.
  3202. * Also, store the current LSN of the inode so that we can tell
  3203. * whether the item has moved in the AIL from xfs_iflush_done().
  3204. * In order to read the lsn we need the AIL lock, because
  3205. * it is a 64 bit value that cannot be read atomically.
  3206. */
  3207. if (iip != NULL && iip->ili_format.ilf_fields != 0) {
  3208. iip->ili_last_fields = iip->ili_format.ilf_fields;
  3209. iip->ili_format.ilf_fields = 0;
  3210. iip->ili_logged = 1;
  3211. ASSERT(sizeof(xfs_lsn_t) == 8); /* don't lock if it shrinks */
  3212. AIL_LOCK(mp,s);
  3213. iip->ili_flush_lsn = iip->ili_item.li_lsn;
  3214. AIL_UNLOCK(mp, s);
  3215. /*
  3216. * Attach the function xfs_iflush_done to the inode's
  3217. * buffer. This will remove the inode from the AIL
  3218. * and unlock the inode's flush lock when the inode is
  3219. * completely written to disk.
  3220. */
  3221. xfs_buf_attach_iodone(bp, (void(*)(xfs_buf_t*,xfs_log_item_t*))
  3222. xfs_iflush_done, (xfs_log_item_t *)iip);
  3223. ASSERT(XFS_BUF_FSPRIVATE(bp, void *) != NULL);
  3224. ASSERT(XFS_BUF_IODONE_FUNC(bp) != NULL);
  3225. } else {
  3226. /*
  3227. * We're flushing an inode which is not in the AIL and has
  3228. * not been logged but has i_update_core set. For this
  3229. * case we can use a B_DELWRI flush and immediately drop
  3230. * the inode flush lock because we can avoid the whole
  3231. * AIL state thing. It's OK to drop the flush lock now,
  3232. * because we've already locked the buffer and to do anything
  3233. * you really need both.
  3234. */
  3235. if (iip != NULL) {
  3236. ASSERT(iip->ili_logged == 0);
  3237. ASSERT(iip->ili_last_fields == 0);
  3238. ASSERT((iip->ili_item.li_flags & XFS_LI_IN_AIL) == 0);
  3239. }
  3240. xfs_ifunlock(ip);
  3241. }
  3242. return 0;
  3243. corrupt_out:
  3244. return XFS_ERROR(EFSCORRUPTED);
  3245. }
  3246. /*
  3247. * Flush all inactive inodes in mp.
  3248. */
  3249. void
  3250. xfs_iflush_all(
  3251. xfs_mount_t *mp)
  3252. {
  3253. xfs_inode_t *ip;
  3254. bhv_vnode_t *vp;
  3255. again:
  3256. XFS_MOUNT_ILOCK(mp);
  3257. ip = mp->m_inodes;
  3258. if (ip == NULL)
  3259. goto out;
  3260. do {
  3261. /* Make sure we skip markers inserted by sync */
  3262. if (ip->i_mount == NULL) {
  3263. ip = ip->i_mnext;
  3264. continue;
  3265. }
  3266. vp = XFS_ITOV_NULL(ip);
  3267. if (!vp) {
  3268. XFS_MOUNT_IUNLOCK(mp);
  3269. xfs_finish_reclaim(ip, 0, XFS_IFLUSH_ASYNC);
  3270. goto again;
  3271. }
  3272. ASSERT(vn_count(vp) == 0);
  3273. ip = ip->i_mnext;
  3274. } while (ip != mp->m_inodes);
  3275. out:
  3276. XFS_MOUNT_IUNLOCK(mp);
  3277. }
  3278. /*
  3279. * xfs_iaccess: check accessibility of inode for mode.
  3280. */
  3281. int
  3282. xfs_iaccess(
  3283. xfs_inode_t *ip,
  3284. mode_t mode,
  3285. cred_t *cr)
  3286. {
  3287. int error;
  3288. mode_t orgmode = mode;
  3289. struct inode *inode = vn_to_inode(XFS_ITOV(ip));
  3290. if (mode & S_IWUSR) {
  3291. umode_t imode = inode->i_mode;
  3292. if (IS_RDONLY(inode) &&
  3293. (S_ISREG(imode) || S_ISDIR(imode) || S_ISLNK(imode)))
  3294. return XFS_ERROR(EROFS);
  3295. if (IS_IMMUTABLE(inode))
  3296. return XFS_ERROR(EACCES);
  3297. }
  3298. /*
  3299. * If there's an Access Control List it's used instead of
  3300. * the mode bits.
  3301. */
  3302. if ((error = _ACL_XFS_IACCESS(ip, mode, cr)) != -1)
  3303. return error ? XFS_ERROR(error) : 0;
  3304. if (current_fsuid(cr) != ip->i_d.di_uid) {
  3305. mode >>= 3;
  3306. if (!in_group_p((gid_t)ip->i_d.di_gid))
  3307. mode >>= 3;
  3308. }
  3309. /*
  3310. * If the DACs are ok we don't need any capability check.
  3311. */
  3312. if ((ip->i_d.di_mode & mode) == mode)
  3313. return 0;
  3314. /*
  3315. * Read/write DACs are always overridable.
  3316. * Executable DACs are overridable if at least one exec bit is set.
  3317. */
  3318. if (!(orgmode & S_IXUSR) ||
  3319. (inode->i_mode & S_IXUGO) || S_ISDIR(inode->i_mode))
  3320. if (capable_cred(cr, CAP_DAC_OVERRIDE))
  3321. return 0;
  3322. if ((orgmode == S_IRUSR) ||
  3323. (S_ISDIR(inode->i_mode) && (!(orgmode & S_IWUSR)))) {
  3324. if (capable_cred(cr, CAP_DAC_READ_SEARCH))
  3325. return 0;
  3326. #ifdef NOISE
  3327. cmn_err(CE_NOTE, "Ick: mode=%o, orgmode=%o", mode, orgmode);
  3328. #endif /* NOISE */
  3329. return XFS_ERROR(EACCES);
  3330. }
  3331. return XFS_ERROR(EACCES);
  3332. }
  3333. /*
  3334. * xfs_iroundup: round up argument to next power of two
  3335. */
  3336. uint
  3337. xfs_iroundup(
  3338. uint v)
  3339. {
  3340. int i;
  3341. uint m;
  3342. if ((v & (v - 1)) == 0)
  3343. return v;
  3344. ASSERT((v & 0x80000000) == 0);
  3345. if ((v & (v + 1)) == 0)
  3346. return v + 1;
  3347. for (i = 0, m = 1; i < 31; i++, m <<= 1) {
  3348. if (v & m)
  3349. continue;
  3350. v |= m;
  3351. if ((v & (v + 1)) == 0)
  3352. return v + 1;
  3353. }
  3354. ASSERT(0);
  3355. return( 0 );
  3356. }
  3357. #ifdef XFS_ILOCK_TRACE
  3358. ktrace_t *xfs_ilock_trace_buf;
  3359. void
  3360. xfs_ilock_trace(xfs_inode_t *ip, int lock, unsigned int lockflags, inst_t *ra)
  3361. {
  3362. ktrace_enter(ip->i_lock_trace,
  3363. (void *)ip,
  3364. (void *)(unsigned long)lock, /* 1 = LOCK, 3=UNLOCK, etc */
  3365. (void *)(unsigned long)lockflags, /* XFS_ILOCK_EXCL etc */
  3366. (void *)ra, /* caller of ilock */
  3367. (void *)(unsigned long)current_cpu(),
  3368. (void *)(unsigned long)current_pid(),
  3369. NULL,NULL,NULL,NULL,NULL,NULL,NULL,NULL,NULL,NULL);
  3370. }
  3371. #endif
  3372. /*
  3373. * Return a pointer to the extent record at file index idx.
  3374. */
  3375. xfs_bmbt_rec_t *
  3376. xfs_iext_get_ext(
  3377. xfs_ifork_t *ifp, /* inode fork pointer */
  3378. xfs_extnum_t idx) /* index of target extent */
  3379. {
  3380. ASSERT(idx >= 0);
  3381. if ((ifp->if_flags & XFS_IFEXTIREC) && (idx == 0)) {
  3382. return ifp->if_u1.if_ext_irec->er_extbuf;
  3383. } else if (ifp->if_flags & XFS_IFEXTIREC) {
  3384. xfs_ext_irec_t *erp; /* irec pointer */
  3385. int erp_idx = 0; /* irec index */
  3386. xfs_extnum_t page_idx = idx; /* ext index in target list */
  3387. erp = xfs_iext_idx_to_irec(ifp, &page_idx, &erp_idx, 0);
  3388. return &erp->er_extbuf[page_idx];
  3389. } else if (ifp->if_bytes) {
  3390. return &ifp->if_u1.if_extents[idx];
  3391. } else {
  3392. return NULL;
  3393. }
  3394. }
  3395. /*
  3396. * Insert new item(s) into the extent records for incore inode
  3397. * fork 'ifp'. 'count' new items are inserted at index 'idx'.
  3398. */
  3399. void
  3400. xfs_iext_insert(
  3401. xfs_ifork_t *ifp, /* inode fork pointer */
  3402. xfs_extnum_t idx, /* starting index of new items */
  3403. xfs_extnum_t count, /* number of inserted items */
  3404. xfs_bmbt_irec_t *new) /* items to insert */
  3405. {
  3406. xfs_bmbt_rec_t *ep; /* extent record pointer */
  3407. xfs_extnum_t i; /* extent record index */
  3408. ASSERT(ifp->if_flags & XFS_IFEXTENTS);
  3409. xfs_iext_add(ifp, idx, count);
  3410. for (i = idx; i < idx + count; i++, new++) {
  3411. ep = xfs_iext_get_ext(ifp, i);
  3412. xfs_bmbt_set_all(ep, new);
  3413. }
  3414. }
  3415. /*
  3416. * This is called when the amount of space required for incore file
  3417. * extents needs to be increased. The ext_diff parameter stores the
  3418. * number of new extents being added and the idx parameter contains
  3419. * the extent index where the new extents will be added. If the new
  3420. * extents are being appended, then we just need to (re)allocate and
  3421. * initialize the space. Otherwise, if the new extents are being
  3422. * inserted into the middle of the existing entries, a bit more work
  3423. * is required to make room for the new extents to be inserted. The
  3424. * caller is responsible for filling in the new extent entries upon
  3425. * return.
  3426. */
  3427. void
  3428. xfs_iext_add(
  3429. xfs_ifork_t *ifp, /* inode fork pointer */
  3430. xfs_extnum_t idx, /* index to begin adding exts */
  3431. int ext_diff) /* number of extents to add */
  3432. {
  3433. int byte_diff; /* new bytes being added */
  3434. int new_size; /* size of extents after adding */
  3435. xfs_extnum_t nextents; /* number of extents in file */
  3436. nextents = ifp->if_bytes / (uint)sizeof(xfs_bmbt_rec_t);
  3437. ASSERT((idx >= 0) && (idx <= nextents));
  3438. byte_diff = ext_diff * sizeof(xfs_bmbt_rec_t);
  3439. new_size = ifp->if_bytes + byte_diff;
  3440. /*
  3441. * If the new number of extents (nextents + ext_diff)
  3442. * fits inside the inode, then continue to use the inline
  3443. * extent buffer.
  3444. */
  3445. if (nextents + ext_diff <= XFS_INLINE_EXTS) {
  3446. if (idx < nextents) {
  3447. memmove(&ifp->if_u2.if_inline_ext[idx + ext_diff],
  3448. &ifp->if_u2.if_inline_ext[idx],
  3449. (nextents - idx) * sizeof(xfs_bmbt_rec_t));
  3450. memset(&ifp->if_u2.if_inline_ext[idx], 0, byte_diff);
  3451. }
  3452. ifp->if_u1.if_extents = ifp->if_u2.if_inline_ext;
  3453. ifp->if_real_bytes = 0;
  3454. ifp->if_lastex = nextents + ext_diff;
  3455. }
  3456. /*
  3457. * Otherwise use a linear (direct) extent list.
  3458. * If the extents are currently inside the inode,
  3459. * xfs_iext_realloc_direct will switch us from
  3460. * inline to direct extent allocation mode.
  3461. */
  3462. else if (nextents + ext_diff <= XFS_LINEAR_EXTS) {
  3463. xfs_iext_realloc_direct(ifp, new_size);
  3464. if (idx < nextents) {
  3465. memmove(&ifp->if_u1.if_extents[idx + ext_diff],
  3466. &ifp->if_u1.if_extents[idx],
  3467. (nextents - idx) * sizeof(xfs_bmbt_rec_t));
  3468. memset(&ifp->if_u1.if_extents[idx], 0, byte_diff);
  3469. }
  3470. }
  3471. /* Indirection array */
  3472. else {
  3473. xfs_ext_irec_t *erp;
  3474. int erp_idx = 0;
  3475. int page_idx = idx;
  3476. ASSERT(nextents + ext_diff > XFS_LINEAR_EXTS);
  3477. if (ifp->if_flags & XFS_IFEXTIREC) {
  3478. erp = xfs_iext_idx_to_irec(ifp, &page_idx, &erp_idx, 1);
  3479. } else {
  3480. xfs_iext_irec_init(ifp);
  3481. ASSERT(ifp->if_flags & XFS_IFEXTIREC);
  3482. erp = ifp->if_u1.if_ext_irec;
  3483. }
  3484. /* Extents fit in target extent page */
  3485. if (erp && erp->er_extcount + ext_diff <= XFS_LINEAR_EXTS) {
  3486. if (page_idx < erp->er_extcount) {
  3487. memmove(&erp->er_extbuf[page_idx + ext_diff],
  3488. &erp->er_extbuf[page_idx],
  3489. (erp->er_extcount - page_idx) *
  3490. sizeof(xfs_bmbt_rec_t));
  3491. memset(&erp->er_extbuf[page_idx], 0, byte_diff);
  3492. }
  3493. erp->er_extcount += ext_diff;
  3494. xfs_iext_irec_update_extoffs(ifp, erp_idx + 1, ext_diff);
  3495. }
  3496. /* Insert a new extent page */
  3497. else if (erp) {
  3498. xfs_iext_add_indirect_multi(ifp,
  3499. erp_idx, page_idx, ext_diff);
  3500. }
  3501. /*
  3502. * If extent(s) are being appended to the last page in
  3503. * the indirection array and the new extent(s) don't fit
  3504. * in the page, then erp is NULL and erp_idx is set to
  3505. * the next index needed in the indirection array.
  3506. */
  3507. else {
  3508. int count = ext_diff;
  3509. while (count) {
  3510. erp = xfs_iext_irec_new(ifp, erp_idx);
  3511. erp->er_extcount = count;
  3512. count -= MIN(count, (int)XFS_LINEAR_EXTS);
  3513. if (count) {
  3514. erp_idx++;
  3515. }
  3516. }
  3517. }
  3518. }
  3519. ifp->if_bytes = new_size;
  3520. }
  3521. /*
  3522. * This is called when incore extents are being added to the indirection
  3523. * array and the new extents do not fit in the target extent list. The
  3524. * erp_idx parameter contains the irec index for the target extent list
  3525. * in the indirection array, and the idx parameter contains the extent
  3526. * index within the list. The number of extents being added is stored
  3527. * in the count parameter.
  3528. *
  3529. * |-------| |-------|
  3530. * | | | | idx - number of extents before idx
  3531. * | idx | | count |
  3532. * | | | | count - number of extents being inserted at idx
  3533. * |-------| |-------|
  3534. * | count | | nex2 | nex2 - number of extents after idx + count
  3535. * |-------| |-------|
  3536. */
  3537. void
  3538. xfs_iext_add_indirect_multi(
  3539. xfs_ifork_t *ifp, /* inode fork pointer */
  3540. int erp_idx, /* target extent irec index */
  3541. xfs_extnum_t idx, /* index within target list */
  3542. int count) /* new extents being added */
  3543. {
  3544. int byte_diff; /* new bytes being added */
  3545. xfs_ext_irec_t *erp; /* pointer to irec entry */
  3546. xfs_extnum_t ext_diff; /* number of extents to add */
  3547. xfs_extnum_t ext_cnt; /* new extents still needed */
  3548. xfs_extnum_t nex2; /* extents after idx + count */
  3549. xfs_bmbt_rec_t *nex2_ep = NULL; /* temp list for nex2 extents */
  3550. int nlists; /* number of irec's (lists) */
  3551. ASSERT(ifp->if_flags & XFS_IFEXTIREC);
  3552. erp = &ifp->if_u1.if_ext_irec[erp_idx];
  3553. nex2 = erp->er_extcount - idx;
  3554. nlists = ifp->if_real_bytes / XFS_IEXT_BUFSZ;
  3555. /*
  3556. * Save second part of target extent list
  3557. * (all extents past */
  3558. if (nex2) {
  3559. byte_diff = nex2 * sizeof(xfs_bmbt_rec_t);
  3560. nex2_ep = (xfs_bmbt_rec_t *) kmem_alloc(byte_diff, KM_SLEEP);
  3561. memmove(nex2_ep, &erp->er_extbuf[idx], byte_diff);
  3562. erp->er_extcount -= nex2;
  3563. xfs_iext_irec_update_extoffs(ifp, erp_idx + 1, -nex2);
  3564. memset(&erp->er_extbuf[idx], 0, byte_diff);
  3565. }
  3566. /*
  3567. * Add the new extents to the end of the target
  3568. * list, then allocate new irec record(s) and
  3569. * extent buffer(s) as needed to store the rest
  3570. * of the new extents.
  3571. */
  3572. ext_cnt = count;
  3573. ext_diff = MIN(ext_cnt, (int)XFS_LINEAR_EXTS - erp->er_extcount);
  3574. if (ext_diff) {
  3575. erp->er_extcount += ext_diff;
  3576. xfs_iext_irec_update_extoffs(ifp, erp_idx + 1, ext_diff);
  3577. ext_cnt -= ext_diff;
  3578. }
  3579. while (ext_cnt) {
  3580. erp_idx++;
  3581. erp = xfs_iext_irec_new(ifp, erp_idx);
  3582. ext_diff = MIN(ext_cnt, (int)XFS_LINEAR_EXTS);
  3583. erp->er_extcount = ext_diff;
  3584. xfs_iext_irec_update_extoffs(ifp, erp_idx + 1, ext_diff);
  3585. ext_cnt -= ext_diff;
  3586. }
  3587. /* Add nex2 extents back to indirection array */
  3588. if (nex2) {
  3589. xfs_extnum_t ext_avail;
  3590. int i;
  3591. byte_diff = nex2 * sizeof(xfs_bmbt_rec_t);
  3592. ext_avail = XFS_LINEAR_EXTS - erp->er_extcount;
  3593. i = 0;
  3594. /*
  3595. * If nex2 extents fit in the current page, append
  3596. * nex2_ep after the new extents.
  3597. */
  3598. if (nex2 <= ext_avail) {
  3599. i = erp->er_extcount;
  3600. }
  3601. /*
  3602. * Otherwise, check if space is available in the
  3603. * next page.
  3604. */
  3605. else if ((erp_idx < nlists - 1) &&
  3606. (nex2 <= (ext_avail = XFS_LINEAR_EXTS -
  3607. ifp->if_u1.if_ext_irec[erp_idx+1].er_extcount))) {
  3608. erp_idx++;
  3609. erp++;
  3610. /* Create a hole for nex2 extents */
  3611. memmove(&erp->er_extbuf[nex2], erp->er_extbuf,
  3612. erp->er_extcount * sizeof(xfs_bmbt_rec_t));
  3613. }
  3614. /*
  3615. * Final choice, create a new extent page for
  3616. * nex2 extents.
  3617. */
  3618. else {
  3619. erp_idx++;
  3620. erp = xfs_iext_irec_new(ifp, erp_idx);
  3621. }
  3622. memmove(&erp->er_extbuf[i], nex2_ep, byte_diff);
  3623. kmem_free(nex2_ep, byte_diff);
  3624. erp->er_extcount += nex2;
  3625. xfs_iext_irec_update_extoffs(ifp, erp_idx + 1, nex2);
  3626. }
  3627. }
  3628. /*
  3629. * This is called when the amount of space required for incore file
  3630. * extents needs to be decreased. The ext_diff parameter stores the
  3631. * number of extents to be removed and the idx parameter contains
  3632. * the extent index where the extents will be removed from.
  3633. *
  3634. * If the amount of space needed has decreased below the linear
  3635. * limit, XFS_IEXT_BUFSZ, then switch to using the contiguous
  3636. * extent array. Otherwise, use kmem_realloc() to adjust the
  3637. * size to what is needed.
  3638. */
  3639. void
  3640. xfs_iext_remove(
  3641. xfs_ifork_t *ifp, /* inode fork pointer */
  3642. xfs_extnum_t idx, /* index to begin removing exts */
  3643. int ext_diff) /* number of extents to remove */
  3644. {
  3645. xfs_extnum_t nextents; /* number of extents in file */
  3646. int new_size; /* size of extents after removal */
  3647. ASSERT(ext_diff > 0);
  3648. nextents = ifp->if_bytes / (uint)sizeof(xfs_bmbt_rec_t);
  3649. new_size = (nextents - ext_diff) * sizeof(xfs_bmbt_rec_t);
  3650. if (new_size == 0) {
  3651. xfs_iext_destroy(ifp);
  3652. } else if (ifp->if_flags & XFS_IFEXTIREC) {
  3653. xfs_iext_remove_indirect(ifp, idx, ext_diff);
  3654. } else if (ifp->if_real_bytes) {
  3655. xfs_iext_remove_direct(ifp, idx, ext_diff);
  3656. } else {
  3657. xfs_iext_remove_inline(ifp, idx, ext_diff);
  3658. }
  3659. ifp->if_bytes = new_size;
  3660. }
  3661. /*
  3662. * This removes ext_diff extents from the inline buffer, beginning
  3663. * at extent index idx.
  3664. */
  3665. void
  3666. xfs_iext_remove_inline(
  3667. xfs_ifork_t *ifp, /* inode fork pointer */
  3668. xfs_extnum_t idx, /* index to begin removing exts */
  3669. int ext_diff) /* number of extents to remove */
  3670. {
  3671. int nextents; /* number of extents in file */
  3672. ASSERT(!(ifp->if_flags & XFS_IFEXTIREC));
  3673. ASSERT(idx < XFS_INLINE_EXTS);
  3674. nextents = ifp->if_bytes / (uint)sizeof(xfs_bmbt_rec_t);
  3675. ASSERT(((nextents - ext_diff) > 0) &&
  3676. (nextents - ext_diff) < XFS_INLINE_EXTS);
  3677. if (idx + ext_diff < nextents) {
  3678. memmove(&ifp->if_u2.if_inline_ext[idx],
  3679. &ifp->if_u2.if_inline_ext[idx + ext_diff],
  3680. (nextents - (idx + ext_diff)) *
  3681. sizeof(xfs_bmbt_rec_t));
  3682. memset(&ifp->if_u2.if_inline_ext[nextents - ext_diff],
  3683. 0, ext_diff * sizeof(xfs_bmbt_rec_t));
  3684. } else {
  3685. memset(&ifp->if_u2.if_inline_ext[idx], 0,
  3686. ext_diff * sizeof(xfs_bmbt_rec_t));
  3687. }
  3688. }
  3689. /*
  3690. * This removes ext_diff extents from a linear (direct) extent list,
  3691. * beginning at extent index idx. If the extents are being removed
  3692. * from the end of the list (ie. truncate) then we just need to re-
  3693. * allocate the list to remove the extra space. Otherwise, if the
  3694. * extents are being removed from the middle of the existing extent
  3695. * entries, then we first need to move the extent records beginning
  3696. * at idx + ext_diff up in the list to overwrite the records being
  3697. * removed, then remove the extra space via kmem_realloc.
  3698. */
  3699. void
  3700. xfs_iext_remove_direct(
  3701. xfs_ifork_t *ifp, /* inode fork pointer */
  3702. xfs_extnum_t idx, /* index to begin removing exts */
  3703. int ext_diff) /* number of extents to remove */
  3704. {
  3705. xfs_extnum_t nextents; /* number of extents in file */
  3706. int new_size; /* size of extents after removal */
  3707. ASSERT(!(ifp->if_flags & XFS_IFEXTIREC));
  3708. new_size = ifp->if_bytes -
  3709. (ext_diff * sizeof(xfs_bmbt_rec_t));
  3710. nextents = ifp->if_bytes / (uint)sizeof(xfs_bmbt_rec_t);
  3711. if (new_size == 0) {
  3712. xfs_iext_destroy(ifp);
  3713. return;
  3714. }
  3715. /* Move extents up in the list (if needed) */
  3716. if (idx + ext_diff < nextents) {
  3717. memmove(&ifp->if_u1.if_extents[idx],
  3718. &ifp->if_u1.if_extents[idx + ext_diff],
  3719. (nextents - (idx + ext_diff)) *
  3720. sizeof(xfs_bmbt_rec_t));
  3721. }
  3722. memset(&ifp->if_u1.if_extents[nextents - ext_diff],
  3723. 0, ext_diff * sizeof(xfs_bmbt_rec_t));
  3724. /*
  3725. * Reallocate the direct extent list. If the extents
  3726. * will fit inside the inode then xfs_iext_realloc_direct
  3727. * will switch from direct to inline extent allocation
  3728. * mode for us.
  3729. */
  3730. xfs_iext_realloc_direct(ifp, new_size);
  3731. ifp->if_bytes = new_size;
  3732. }
  3733. /*
  3734. * This is called when incore extents are being removed from the
  3735. * indirection array and the extents being removed span multiple extent
  3736. * buffers. The idx parameter contains the file extent index where we
  3737. * want to begin removing extents, and the count parameter contains
  3738. * how many extents need to be removed.
  3739. *
  3740. * |-------| |-------|
  3741. * | nex1 | | | nex1 - number of extents before idx
  3742. * |-------| | count |
  3743. * | | | | count - number of extents being removed at idx
  3744. * | count | |-------|
  3745. * | | | nex2 | nex2 - number of extents after idx + count
  3746. * |-------| |-------|
  3747. */
  3748. void
  3749. xfs_iext_remove_indirect(
  3750. xfs_ifork_t *ifp, /* inode fork pointer */
  3751. xfs_extnum_t idx, /* index to begin removing extents */
  3752. int count) /* number of extents to remove */
  3753. {
  3754. xfs_ext_irec_t *erp; /* indirection array pointer */
  3755. int erp_idx = 0; /* indirection array index */
  3756. xfs_extnum_t ext_cnt; /* extents left to remove */
  3757. xfs_extnum_t ext_diff; /* extents to remove in current list */
  3758. xfs_extnum_t nex1; /* number of extents before idx */
  3759. xfs_extnum_t nex2; /* extents after idx + count */
  3760. int nlists; /* entries in indirection array */
  3761. int page_idx = idx; /* index in target extent list */
  3762. ASSERT(ifp->if_flags & XFS_IFEXTIREC);
  3763. erp = xfs_iext_idx_to_irec(ifp, &page_idx, &erp_idx, 0);
  3764. ASSERT(erp != NULL);
  3765. nlists = ifp->if_real_bytes / XFS_IEXT_BUFSZ;
  3766. nex1 = page_idx;
  3767. ext_cnt = count;
  3768. while (ext_cnt) {
  3769. nex2 = MAX((erp->er_extcount - (nex1 + ext_cnt)), 0);
  3770. ext_diff = MIN(ext_cnt, (erp->er_extcount - nex1));
  3771. /*
  3772. * Check for deletion of entire list;
  3773. * xfs_iext_irec_remove() updates extent offsets.
  3774. */
  3775. if (ext_diff == erp->er_extcount) {
  3776. xfs_iext_irec_remove(ifp, erp_idx);
  3777. ext_cnt -= ext_diff;
  3778. nex1 = 0;
  3779. if (ext_cnt) {
  3780. ASSERT(erp_idx < ifp->if_real_bytes /
  3781. XFS_IEXT_BUFSZ);
  3782. erp = &ifp->if_u1.if_ext_irec[erp_idx];
  3783. nex1 = 0;
  3784. continue;
  3785. } else {
  3786. break;
  3787. }
  3788. }
  3789. /* Move extents up (if needed) */
  3790. if (nex2) {
  3791. memmove(&erp->er_extbuf[nex1],
  3792. &erp->er_extbuf[nex1 + ext_diff],
  3793. nex2 * sizeof(xfs_bmbt_rec_t));
  3794. }
  3795. /* Zero out rest of page */
  3796. memset(&erp->er_extbuf[nex1 + nex2], 0, (XFS_IEXT_BUFSZ -
  3797. ((nex1 + nex2) * sizeof(xfs_bmbt_rec_t))));
  3798. /* Update remaining counters */
  3799. erp->er_extcount -= ext_diff;
  3800. xfs_iext_irec_update_extoffs(ifp, erp_idx + 1, -ext_diff);
  3801. ext_cnt -= ext_diff;
  3802. nex1 = 0;
  3803. erp_idx++;
  3804. erp++;
  3805. }
  3806. ifp->if_bytes -= count * sizeof(xfs_bmbt_rec_t);
  3807. xfs_iext_irec_compact(ifp);
  3808. }
  3809. /*
  3810. * Create, destroy, or resize a linear (direct) block of extents.
  3811. */
  3812. void
  3813. xfs_iext_realloc_direct(
  3814. xfs_ifork_t *ifp, /* inode fork pointer */
  3815. int new_size) /* new size of extents */
  3816. {
  3817. int rnew_size; /* real new size of extents */
  3818. rnew_size = new_size;
  3819. ASSERT(!(ifp->if_flags & XFS_IFEXTIREC) ||
  3820. ((new_size >= 0) && (new_size <= XFS_IEXT_BUFSZ) &&
  3821. (new_size != ifp->if_real_bytes)));
  3822. /* Free extent records */
  3823. if (new_size == 0) {
  3824. xfs_iext_destroy(ifp);
  3825. }
  3826. /* Resize direct extent list and zero any new bytes */
  3827. else if (ifp->if_real_bytes) {
  3828. /* Check if extents will fit inside the inode */
  3829. if (new_size <= XFS_INLINE_EXTS * sizeof(xfs_bmbt_rec_t)) {
  3830. xfs_iext_direct_to_inline(ifp, new_size /
  3831. (uint)sizeof(xfs_bmbt_rec_t));
  3832. ifp->if_bytes = new_size;
  3833. return;
  3834. }
  3835. if ((new_size & (new_size - 1)) != 0) {
  3836. rnew_size = xfs_iroundup(new_size);
  3837. }
  3838. if (rnew_size != ifp->if_real_bytes) {
  3839. ifp->if_u1.if_extents = (xfs_bmbt_rec_t *)
  3840. kmem_realloc(ifp->if_u1.if_extents,
  3841. rnew_size,
  3842. ifp->if_real_bytes,
  3843. KM_SLEEP);
  3844. }
  3845. if (rnew_size > ifp->if_real_bytes) {
  3846. memset(&ifp->if_u1.if_extents[ifp->if_bytes /
  3847. (uint)sizeof(xfs_bmbt_rec_t)], 0,
  3848. rnew_size - ifp->if_real_bytes);
  3849. }
  3850. }
  3851. /*
  3852. * Switch from the inline extent buffer to a direct
  3853. * extent list. Be sure to include the inline extent
  3854. * bytes in new_size.
  3855. */
  3856. else {
  3857. new_size += ifp->if_bytes;
  3858. if ((new_size & (new_size - 1)) != 0) {
  3859. rnew_size = xfs_iroundup(new_size);
  3860. }
  3861. xfs_iext_inline_to_direct(ifp, rnew_size);
  3862. }
  3863. ifp->if_real_bytes = rnew_size;
  3864. ifp->if_bytes = new_size;
  3865. }
  3866. /*
  3867. * Switch from linear (direct) extent records to inline buffer.
  3868. */
  3869. void
  3870. xfs_iext_direct_to_inline(
  3871. xfs_ifork_t *ifp, /* inode fork pointer */
  3872. xfs_extnum_t nextents) /* number of extents in file */
  3873. {
  3874. ASSERT(ifp->if_flags & XFS_IFEXTENTS);
  3875. ASSERT(nextents <= XFS_INLINE_EXTS);
  3876. /*
  3877. * The inline buffer was zeroed when we switched
  3878. * from inline to direct extent allocation mode,
  3879. * so we don't need to clear it here.
  3880. */
  3881. memcpy(ifp->if_u2.if_inline_ext, ifp->if_u1.if_extents,
  3882. nextents * sizeof(xfs_bmbt_rec_t));
  3883. kmem_free(ifp->if_u1.if_extents, ifp->if_real_bytes);
  3884. ifp->if_u1.if_extents = ifp->if_u2.if_inline_ext;
  3885. ifp->if_real_bytes = 0;
  3886. }
  3887. /*
  3888. * Switch from inline buffer to linear (direct) extent records.
  3889. * new_size should already be rounded up to the next power of 2
  3890. * by the caller (when appropriate), so use new_size as it is.
  3891. * However, since new_size may be rounded up, we can't update
  3892. * if_bytes here. It is the caller's responsibility to update
  3893. * if_bytes upon return.
  3894. */
  3895. void
  3896. xfs_iext_inline_to_direct(
  3897. xfs_ifork_t *ifp, /* inode fork pointer */
  3898. int new_size) /* number of extents in file */
  3899. {
  3900. ifp->if_u1.if_extents = (xfs_bmbt_rec_t *)
  3901. kmem_alloc(new_size, KM_SLEEP);
  3902. memset(ifp->if_u1.if_extents, 0, new_size);
  3903. if (ifp->if_bytes) {
  3904. memcpy(ifp->if_u1.if_extents, ifp->if_u2.if_inline_ext,
  3905. ifp->if_bytes);
  3906. memset(ifp->if_u2.if_inline_ext, 0, XFS_INLINE_EXTS *
  3907. sizeof(xfs_bmbt_rec_t));
  3908. }
  3909. ifp->if_real_bytes = new_size;
  3910. }
  3911. /*
  3912. * Resize an extent indirection array to new_size bytes.
  3913. */
  3914. void
  3915. xfs_iext_realloc_indirect(
  3916. xfs_ifork_t *ifp, /* inode fork pointer */
  3917. int new_size) /* new indirection array size */
  3918. {
  3919. int nlists; /* number of irec's (ex lists) */
  3920. int size; /* current indirection array size */
  3921. ASSERT(ifp->if_flags & XFS_IFEXTIREC);
  3922. nlists = ifp->if_real_bytes / XFS_IEXT_BUFSZ;
  3923. size = nlists * sizeof(xfs_ext_irec_t);
  3924. ASSERT(ifp->if_real_bytes);
  3925. ASSERT((new_size >= 0) && (new_size != size));
  3926. if (new_size == 0) {
  3927. xfs_iext_destroy(ifp);
  3928. } else {
  3929. ifp->if_u1.if_ext_irec = (xfs_ext_irec_t *)
  3930. kmem_realloc(ifp->if_u1.if_ext_irec,
  3931. new_size, size, KM_SLEEP);
  3932. }
  3933. }
  3934. /*
  3935. * Switch from indirection array to linear (direct) extent allocations.
  3936. */
  3937. void
  3938. xfs_iext_indirect_to_direct(
  3939. xfs_ifork_t *ifp) /* inode fork pointer */
  3940. {
  3941. xfs_bmbt_rec_t *ep; /* extent record pointer */
  3942. xfs_extnum_t nextents; /* number of extents in file */
  3943. int size; /* size of file extents */
  3944. ASSERT(ifp->if_flags & XFS_IFEXTIREC);
  3945. nextents = ifp->if_bytes / (uint)sizeof(xfs_bmbt_rec_t);
  3946. ASSERT(nextents <= XFS_LINEAR_EXTS);
  3947. size = nextents * sizeof(xfs_bmbt_rec_t);
  3948. xfs_iext_irec_compact_full(ifp);
  3949. ASSERT(ifp->if_real_bytes == XFS_IEXT_BUFSZ);
  3950. ep = ifp->if_u1.if_ext_irec->er_extbuf;
  3951. kmem_free(ifp->if_u1.if_ext_irec, sizeof(xfs_ext_irec_t));
  3952. ifp->if_flags &= ~XFS_IFEXTIREC;
  3953. ifp->if_u1.if_extents = ep;
  3954. ifp->if_bytes = size;
  3955. if (nextents < XFS_LINEAR_EXTS) {
  3956. xfs_iext_realloc_direct(ifp, size);
  3957. }
  3958. }
  3959. /*
  3960. * Free incore file extents.
  3961. */
  3962. void
  3963. xfs_iext_destroy(
  3964. xfs_ifork_t *ifp) /* inode fork pointer */
  3965. {
  3966. if (ifp->if_flags & XFS_IFEXTIREC) {
  3967. int erp_idx;
  3968. int nlists;
  3969. nlists = ifp->if_real_bytes / XFS_IEXT_BUFSZ;
  3970. for (erp_idx = nlists - 1; erp_idx >= 0 ; erp_idx--) {
  3971. xfs_iext_irec_remove(ifp, erp_idx);
  3972. }
  3973. ifp->if_flags &= ~XFS_IFEXTIREC;
  3974. } else if (ifp->if_real_bytes) {
  3975. kmem_free(ifp->if_u1.if_extents, ifp->if_real_bytes);
  3976. } else if (ifp->if_bytes) {
  3977. memset(ifp->if_u2.if_inline_ext, 0, XFS_INLINE_EXTS *
  3978. sizeof(xfs_bmbt_rec_t));
  3979. }
  3980. ifp->if_u1.if_extents = NULL;
  3981. ifp->if_real_bytes = 0;
  3982. ifp->if_bytes = 0;
  3983. }
  3984. /*
  3985. * Return a pointer to the extent record for file system block bno.
  3986. */
  3987. xfs_bmbt_rec_t * /* pointer to found extent record */
  3988. xfs_iext_bno_to_ext(
  3989. xfs_ifork_t *ifp, /* inode fork pointer */
  3990. xfs_fileoff_t bno, /* block number to search for */
  3991. xfs_extnum_t *idxp) /* index of target extent */
  3992. {
  3993. xfs_bmbt_rec_t *base; /* pointer to first extent */
  3994. xfs_filblks_t blockcount = 0; /* number of blocks in extent */
  3995. xfs_bmbt_rec_t *ep = NULL; /* pointer to target extent */
  3996. xfs_ext_irec_t *erp = NULL; /* indirection array pointer */
  3997. int high; /* upper boundary in search */
  3998. xfs_extnum_t idx = 0; /* index of target extent */
  3999. int low; /* lower boundary in search */
  4000. xfs_extnum_t nextents; /* number of file extents */
  4001. xfs_fileoff_t startoff = 0; /* start offset of extent */
  4002. nextents = ifp->if_bytes / (uint)sizeof(xfs_bmbt_rec_t);
  4003. if (nextents == 0) {
  4004. *idxp = 0;
  4005. return NULL;
  4006. }
  4007. low = 0;
  4008. if (ifp->if_flags & XFS_IFEXTIREC) {
  4009. /* Find target extent list */
  4010. int erp_idx = 0;
  4011. erp = xfs_iext_bno_to_irec(ifp, bno, &erp_idx);
  4012. base = erp->er_extbuf;
  4013. high = erp->er_extcount - 1;
  4014. } else {
  4015. base = ifp->if_u1.if_extents;
  4016. high = nextents - 1;
  4017. }
  4018. /* Binary search extent records */
  4019. while (low <= high) {
  4020. idx = (low + high) >> 1;
  4021. ep = base + idx;
  4022. startoff = xfs_bmbt_get_startoff(ep);
  4023. blockcount = xfs_bmbt_get_blockcount(ep);
  4024. if (bno < startoff) {
  4025. high = idx - 1;
  4026. } else if (bno >= startoff + blockcount) {
  4027. low = idx + 1;
  4028. } else {
  4029. /* Convert back to file-based extent index */
  4030. if (ifp->if_flags & XFS_IFEXTIREC) {
  4031. idx += erp->er_extoff;
  4032. }
  4033. *idxp = idx;
  4034. return ep;
  4035. }
  4036. }
  4037. /* Convert back to file-based extent index */
  4038. if (ifp->if_flags & XFS_IFEXTIREC) {
  4039. idx += erp->er_extoff;
  4040. }
  4041. if (bno >= startoff + blockcount) {
  4042. if (++idx == nextents) {
  4043. ep = NULL;
  4044. } else {
  4045. ep = xfs_iext_get_ext(ifp, idx);
  4046. }
  4047. }
  4048. *idxp = idx;
  4049. return ep;
  4050. }
  4051. /*
  4052. * Return a pointer to the indirection array entry containing the
  4053. * extent record for filesystem block bno. Store the index of the
  4054. * target irec in *erp_idxp.
  4055. */
  4056. xfs_ext_irec_t * /* pointer to found extent record */
  4057. xfs_iext_bno_to_irec(
  4058. xfs_ifork_t *ifp, /* inode fork pointer */
  4059. xfs_fileoff_t bno, /* block number to search for */
  4060. int *erp_idxp) /* irec index of target ext list */
  4061. {
  4062. xfs_ext_irec_t *erp = NULL; /* indirection array pointer */
  4063. xfs_ext_irec_t *erp_next; /* next indirection array entry */
  4064. int erp_idx; /* indirection array index */
  4065. int nlists; /* number of extent irec's (lists) */
  4066. int high; /* binary search upper limit */
  4067. int low; /* binary search lower limit */
  4068. ASSERT(ifp->if_flags & XFS_IFEXTIREC);
  4069. nlists = ifp->if_real_bytes / XFS_IEXT_BUFSZ;
  4070. erp_idx = 0;
  4071. low = 0;
  4072. high = nlists - 1;
  4073. while (low <= high) {
  4074. erp_idx = (low + high) >> 1;
  4075. erp = &ifp->if_u1.if_ext_irec[erp_idx];
  4076. erp_next = erp_idx < nlists - 1 ? erp + 1 : NULL;
  4077. if (bno < xfs_bmbt_get_startoff(erp->er_extbuf)) {
  4078. high = erp_idx - 1;
  4079. } else if (erp_next && bno >=
  4080. xfs_bmbt_get_startoff(erp_next->er_extbuf)) {
  4081. low = erp_idx + 1;
  4082. } else {
  4083. break;
  4084. }
  4085. }
  4086. *erp_idxp = erp_idx;
  4087. return erp;
  4088. }
  4089. /*
  4090. * Return a pointer to the indirection array entry containing the
  4091. * extent record at file extent index *idxp. Store the index of the
  4092. * target irec in *erp_idxp and store the page index of the target
  4093. * extent record in *idxp.
  4094. */
  4095. xfs_ext_irec_t *
  4096. xfs_iext_idx_to_irec(
  4097. xfs_ifork_t *ifp, /* inode fork pointer */
  4098. xfs_extnum_t *idxp, /* extent index (file -> page) */
  4099. int *erp_idxp, /* pointer to target irec */
  4100. int realloc) /* new bytes were just added */
  4101. {
  4102. xfs_ext_irec_t *prev; /* pointer to previous irec */
  4103. xfs_ext_irec_t *erp = NULL; /* pointer to current irec */
  4104. int erp_idx; /* indirection array index */
  4105. int nlists; /* number of irec's (ex lists) */
  4106. int high; /* binary search upper limit */
  4107. int low; /* binary search lower limit */
  4108. xfs_extnum_t page_idx = *idxp; /* extent index in target list */
  4109. ASSERT(ifp->if_flags & XFS_IFEXTIREC);
  4110. ASSERT(page_idx >= 0 && page_idx <=
  4111. ifp->if_bytes / (uint)sizeof(xfs_bmbt_rec_t));
  4112. nlists = ifp->if_real_bytes / XFS_IEXT_BUFSZ;
  4113. erp_idx = 0;
  4114. low = 0;
  4115. high = nlists - 1;
  4116. /* Binary search extent irec's */
  4117. while (low <= high) {
  4118. erp_idx = (low + high) >> 1;
  4119. erp = &ifp->if_u1.if_ext_irec[erp_idx];
  4120. prev = erp_idx > 0 ? erp - 1 : NULL;
  4121. if (page_idx < erp->er_extoff || (page_idx == erp->er_extoff &&
  4122. realloc && prev && prev->er_extcount < XFS_LINEAR_EXTS)) {
  4123. high = erp_idx - 1;
  4124. } else if (page_idx > erp->er_extoff + erp->er_extcount ||
  4125. (page_idx == erp->er_extoff + erp->er_extcount &&
  4126. !realloc)) {
  4127. low = erp_idx + 1;
  4128. } else if (page_idx == erp->er_extoff + erp->er_extcount &&
  4129. erp->er_extcount == XFS_LINEAR_EXTS) {
  4130. ASSERT(realloc);
  4131. page_idx = 0;
  4132. erp_idx++;
  4133. erp = erp_idx < nlists ? erp + 1 : NULL;
  4134. break;
  4135. } else {
  4136. page_idx -= erp->er_extoff;
  4137. break;
  4138. }
  4139. }
  4140. *idxp = page_idx;
  4141. *erp_idxp = erp_idx;
  4142. return(erp);
  4143. }
  4144. /*
  4145. * Allocate and initialize an indirection array once the space needed
  4146. * for incore extents increases above XFS_IEXT_BUFSZ.
  4147. */
  4148. void
  4149. xfs_iext_irec_init(
  4150. xfs_ifork_t *ifp) /* inode fork pointer */
  4151. {
  4152. xfs_ext_irec_t *erp; /* indirection array pointer */
  4153. xfs_extnum_t nextents; /* number of extents in file */
  4154. ASSERT(!(ifp->if_flags & XFS_IFEXTIREC));
  4155. nextents = ifp->if_bytes / (uint)sizeof(xfs_bmbt_rec_t);
  4156. ASSERT(nextents <= XFS_LINEAR_EXTS);
  4157. erp = (xfs_ext_irec_t *)
  4158. kmem_alloc(sizeof(xfs_ext_irec_t), KM_SLEEP);
  4159. if (nextents == 0) {
  4160. ifp->if_u1.if_extents = (xfs_bmbt_rec_t *)
  4161. kmem_alloc(XFS_IEXT_BUFSZ, KM_SLEEP);
  4162. } else if (!ifp->if_real_bytes) {
  4163. xfs_iext_inline_to_direct(ifp, XFS_IEXT_BUFSZ);
  4164. } else if (ifp->if_real_bytes < XFS_IEXT_BUFSZ) {
  4165. xfs_iext_realloc_direct(ifp, XFS_IEXT_BUFSZ);
  4166. }
  4167. erp->er_extbuf = ifp->if_u1.if_extents;
  4168. erp->er_extcount = nextents;
  4169. erp->er_extoff = 0;
  4170. ifp->if_flags |= XFS_IFEXTIREC;
  4171. ifp->if_real_bytes = XFS_IEXT_BUFSZ;
  4172. ifp->if_bytes = nextents * sizeof(xfs_bmbt_rec_t);
  4173. ifp->if_u1.if_ext_irec = erp;
  4174. return;
  4175. }
  4176. /*
  4177. * Allocate and initialize a new entry in the indirection array.
  4178. */
  4179. xfs_ext_irec_t *
  4180. xfs_iext_irec_new(
  4181. xfs_ifork_t *ifp, /* inode fork pointer */
  4182. int erp_idx) /* index for new irec */
  4183. {
  4184. xfs_ext_irec_t *erp; /* indirection array pointer */
  4185. int i; /* loop counter */
  4186. int nlists; /* number of irec's (ex lists) */
  4187. ASSERT(ifp->if_flags & XFS_IFEXTIREC);
  4188. nlists = ifp->if_real_bytes / XFS_IEXT_BUFSZ;
  4189. /* Resize indirection array */
  4190. xfs_iext_realloc_indirect(ifp, ++nlists *
  4191. sizeof(xfs_ext_irec_t));
  4192. /*
  4193. * Move records down in the array so the
  4194. * new page can use erp_idx.
  4195. */
  4196. erp = ifp->if_u1.if_ext_irec;
  4197. for (i = nlists - 1; i > erp_idx; i--) {
  4198. memmove(&erp[i], &erp[i-1], sizeof(xfs_ext_irec_t));
  4199. }
  4200. ASSERT(i == erp_idx);
  4201. /* Initialize new extent record */
  4202. erp = ifp->if_u1.if_ext_irec;
  4203. erp[erp_idx].er_extbuf = (xfs_bmbt_rec_t *)
  4204. kmem_alloc(XFS_IEXT_BUFSZ, KM_SLEEP);
  4205. ifp->if_real_bytes = nlists * XFS_IEXT_BUFSZ;
  4206. memset(erp[erp_idx].er_extbuf, 0, XFS_IEXT_BUFSZ);
  4207. erp[erp_idx].er_extcount = 0;
  4208. erp[erp_idx].er_extoff = erp_idx > 0 ?
  4209. erp[erp_idx-1].er_extoff + erp[erp_idx-1].er_extcount : 0;
  4210. return (&erp[erp_idx]);
  4211. }
  4212. /*
  4213. * Remove a record from the indirection array.
  4214. */
  4215. void
  4216. xfs_iext_irec_remove(
  4217. xfs_ifork_t *ifp, /* inode fork pointer */
  4218. int erp_idx) /* irec index to remove */
  4219. {
  4220. xfs_ext_irec_t *erp; /* indirection array pointer */
  4221. int i; /* loop counter */
  4222. int nlists; /* number of irec's (ex lists) */
  4223. ASSERT(ifp->if_flags & XFS_IFEXTIREC);
  4224. nlists = ifp->if_real_bytes / XFS_IEXT_BUFSZ;
  4225. erp = &ifp->if_u1.if_ext_irec[erp_idx];
  4226. if (erp->er_extbuf) {
  4227. xfs_iext_irec_update_extoffs(ifp, erp_idx + 1,
  4228. -erp->er_extcount);
  4229. kmem_free(erp->er_extbuf, XFS_IEXT_BUFSZ);
  4230. }
  4231. /* Compact extent records */
  4232. erp = ifp->if_u1.if_ext_irec;
  4233. for (i = erp_idx; i < nlists - 1; i++) {
  4234. memmove(&erp[i], &erp[i+1], sizeof(xfs_ext_irec_t));
  4235. }
  4236. /*
  4237. * Manually free the last extent record from the indirection
  4238. * array. A call to xfs_iext_realloc_indirect() with a size
  4239. * of zero would result in a call to xfs_iext_destroy() which
  4240. * would in turn call this function again, creating a nasty
  4241. * infinite loop.
  4242. */
  4243. if (--nlists) {
  4244. xfs_iext_realloc_indirect(ifp,
  4245. nlists * sizeof(xfs_ext_irec_t));
  4246. } else {
  4247. kmem_free(ifp->if_u1.if_ext_irec,
  4248. sizeof(xfs_ext_irec_t));
  4249. }
  4250. ifp->if_real_bytes = nlists * XFS_IEXT_BUFSZ;
  4251. }
  4252. /*
  4253. * This is called to clean up large amounts of unused memory allocated
  4254. * by the indirection array. Before compacting anything though, verify
  4255. * that the indirection array is still needed and switch back to the
  4256. * linear extent list (or even the inline buffer) if possible. The
  4257. * compaction policy is as follows:
  4258. *
  4259. * Full Compaction: Extents fit into a single page (or inline buffer)
  4260. * Full Compaction: Extents occupy less than 10% of allocated space
  4261. * Partial Compaction: Extents occupy > 10% and < 50% of allocated space
  4262. * No Compaction: Extents occupy at least 50% of allocated space
  4263. */
  4264. void
  4265. xfs_iext_irec_compact(
  4266. xfs_ifork_t *ifp) /* inode fork pointer */
  4267. {
  4268. xfs_extnum_t nextents; /* number of extents in file */
  4269. int nlists; /* number of irec's (ex lists) */
  4270. ASSERT(ifp->if_flags & XFS_IFEXTIREC);
  4271. nlists = ifp->if_real_bytes / XFS_IEXT_BUFSZ;
  4272. nextents = ifp->if_bytes / (uint)sizeof(xfs_bmbt_rec_t);
  4273. if (nextents == 0) {
  4274. xfs_iext_destroy(ifp);
  4275. } else if (nextents <= XFS_INLINE_EXTS) {
  4276. xfs_iext_indirect_to_direct(ifp);
  4277. xfs_iext_direct_to_inline(ifp, nextents);
  4278. } else if (nextents <= XFS_LINEAR_EXTS) {
  4279. xfs_iext_indirect_to_direct(ifp);
  4280. } else if (nextents < (nlists * XFS_LINEAR_EXTS) >> 3) {
  4281. xfs_iext_irec_compact_full(ifp);
  4282. } else if (nextents < (nlists * XFS_LINEAR_EXTS) >> 1) {
  4283. xfs_iext_irec_compact_pages(ifp);
  4284. }
  4285. }
  4286. /*
  4287. * Combine extents from neighboring extent pages.
  4288. */
  4289. void
  4290. xfs_iext_irec_compact_pages(
  4291. xfs_ifork_t *ifp) /* inode fork pointer */
  4292. {
  4293. xfs_ext_irec_t *erp, *erp_next;/* pointers to irec entries */
  4294. int erp_idx = 0; /* indirection array index */
  4295. int nlists; /* number of irec's (ex lists) */
  4296. ASSERT(ifp->if_flags & XFS_IFEXTIREC);
  4297. nlists = ifp->if_real_bytes / XFS_IEXT_BUFSZ;
  4298. while (erp_idx < nlists - 1) {
  4299. erp = &ifp->if_u1.if_ext_irec[erp_idx];
  4300. erp_next = erp + 1;
  4301. if (erp_next->er_extcount <=
  4302. (XFS_LINEAR_EXTS - erp->er_extcount)) {
  4303. memmove(&erp->er_extbuf[erp->er_extcount],
  4304. erp_next->er_extbuf, erp_next->er_extcount *
  4305. sizeof(xfs_bmbt_rec_t));
  4306. erp->er_extcount += erp_next->er_extcount;
  4307. /*
  4308. * Free page before removing extent record
  4309. * so er_extoffs don't get modified in
  4310. * xfs_iext_irec_remove.
  4311. */
  4312. kmem_free(erp_next->er_extbuf, XFS_IEXT_BUFSZ);
  4313. erp_next->er_extbuf = NULL;
  4314. xfs_iext_irec_remove(ifp, erp_idx + 1);
  4315. nlists = ifp->if_real_bytes / XFS_IEXT_BUFSZ;
  4316. } else {
  4317. erp_idx++;
  4318. }
  4319. }
  4320. }
  4321. /*
  4322. * Fully compact the extent records managed by the indirection array.
  4323. */
  4324. void
  4325. xfs_iext_irec_compact_full(
  4326. xfs_ifork_t *ifp) /* inode fork pointer */
  4327. {
  4328. xfs_bmbt_rec_t *ep, *ep_next; /* extent record pointers */
  4329. xfs_ext_irec_t *erp, *erp_next; /* extent irec pointers */
  4330. int erp_idx = 0; /* extent irec index */
  4331. int ext_avail; /* empty entries in ex list */
  4332. int ext_diff; /* number of exts to add */
  4333. int nlists; /* number of irec's (ex lists) */
  4334. ASSERT(ifp->if_flags & XFS_IFEXTIREC);
  4335. nlists = ifp->if_real_bytes / XFS_IEXT_BUFSZ;
  4336. erp = ifp->if_u1.if_ext_irec;
  4337. ep = &erp->er_extbuf[erp->er_extcount];
  4338. erp_next = erp + 1;
  4339. ep_next = erp_next->er_extbuf;
  4340. while (erp_idx < nlists - 1) {
  4341. ext_avail = XFS_LINEAR_EXTS - erp->er_extcount;
  4342. ext_diff = MIN(ext_avail, erp_next->er_extcount);
  4343. memcpy(ep, ep_next, ext_diff * sizeof(xfs_bmbt_rec_t));
  4344. erp->er_extcount += ext_diff;
  4345. erp_next->er_extcount -= ext_diff;
  4346. /* Remove next page */
  4347. if (erp_next->er_extcount == 0) {
  4348. /*
  4349. * Free page before removing extent record
  4350. * so er_extoffs don't get modified in
  4351. * xfs_iext_irec_remove.
  4352. */
  4353. kmem_free(erp_next->er_extbuf,
  4354. erp_next->er_extcount * sizeof(xfs_bmbt_rec_t));
  4355. erp_next->er_extbuf = NULL;
  4356. xfs_iext_irec_remove(ifp, erp_idx + 1);
  4357. erp = &ifp->if_u1.if_ext_irec[erp_idx];
  4358. nlists = ifp->if_real_bytes / XFS_IEXT_BUFSZ;
  4359. /* Update next page */
  4360. } else {
  4361. /* Move rest of page up to become next new page */
  4362. memmove(erp_next->er_extbuf, ep_next,
  4363. erp_next->er_extcount * sizeof(xfs_bmbt_rec_t));
  4364. ep_next = erp_next->er_extbuf;
  4365. memset(&ep_next[erp_next->er_extcount], 0,
  4366. (XFS_LINEAR_EXTS - erp_next->er_extcount) *
  4367. sizeof(xfs_bmbt_rec_t));
  4368. }
  4369. if (erp->er_extcount == XFS_LINEAR_EXTS) {
  4370. erp_idx++;
  4371. if (erp_idx < nlists)
  4372. erp = &ifp->if_u1.if_ext_irec[erp_idx];
  4373. else
  4374. break;
  4375. }
  4376. ep = &erp->er_extbuf[erp->er_extcount];
  4377. erp_next = erp + 1;
  4378. ep_next = erp_next->er_extbuf;
  4379. }
  4380. }
  4381. /*
  4382. * This is called to update the er_extoff field in the indirection
  4383. * array when extents have been added or removed from one of the
  4384. * extent lists. erp_idx contains the irec index to begin updating
  4385. * at and ext_diff contains the number of extents that were added
  4386. * or removed.
  4387. */
  4388. void
  4389. xfs_iext_irec_update_extoffs(
  4390. xfs_ifork_t *ifp, /* inode fork pointer */
  4391. int erp_idx, /* irec index to update */
  4392. int ext_diff) /* number of new extents */
  4393. {
  4394. int i; /* loop counter */
  4395. int nlists; /* number of irec's (ex lists */
  4396. ASSERT(ifp->if_flags & XFS_IFEXTIREC);
  4397. nlists = ifp->if_real_bytes / XFS_IEXT_BUFSZ;
  4398. for (i = erp_idx; i < nlists; i++) {
  4399. ifp->if_u1.if_ext_irec[i].er_extoff += ext_diff;
  4400. }
  4401. }