xfs_inode.c 138 KB

1234567891011121314151617181920212223242526272829303132333435363738394041424344454647484950515253545556575859606162636465666768697071727374757677787980818283848586878889909192939495969798991001011021031041051061071081091101111121131141151161171181191201211221231241251261271281291301311321331341351361371381391401411421431441451461471481491501511521531541551561571581591601611621631641651661671681691701711721731741751761771781791801811821831841851861871881891901911921931941951961971981992002012022032042052062072082092102112122132142152162172182192202212222232242252262272282292302312322332342352362372382392402412422432442452462472482492502512522532542552562572582592602612622632642652662672682692702712722732742752762772782792802812822832842852862872882892902912922932942952962972982993003013023033043053063073083093103113123133143153163173183193203213223233243253263273283293303313323333343353363373383393403413423433443453463473483493503513523533543553563573583593603613623633643653663673683693703713723733743753763773783793803813823833843853863873883893903913923933943953963973983994004014024034044054064074084094104114124134144154164174184194204214224234244254264274284294304314324334344354364374384394404414424434444454464474484494504514524534544554564574584594604614624634644654664674684694704714724734744754764774784794804814824834844854864874884894904914924934944954964974984995005015025035045055065075085095105115125135145155165175185195205215225235245255265275285295305315325335345355365375385395405415425435445455465475485495505515525535545555565575585595605615625635645655665675685695705715725735745755765775785795805815825835845855865875885895905915925935945955965975985996006016026036046056066076086096106116126136146156166176186196206216226236246256266276286296306316326336346356366376386396406416426436446456466476486496506516526536546556566576586596606616626636646656666676686696706716726736746756766776786796806816826836846856866876886896906916926936946956966976986997007017027037047057067077087097107117127137147157167177187197207217227237247257267277287297307317327337347357367377387397407417427437447457467477487497507517527537547557567577587597607617627637647657667677687697707717727737747757767777787797807817827837847857867877887897907917927937947957967977987998008018028038048058068078088098108118128138148158168178188198208218228238248258268278288298308318328338348358368378388398408418428438448458468478488498508518528538548558568578588598608618628638648658668678688698708718728738748758768778788798808818828838848858868878888898908918928938948958968978988999009019029039049059069079089099109119129139149159169179189199209219229239249259269279289299309319329339349359369379389399409419429439449459469479489499509519529539549559569579589599609619629639649659669679689699709719729739749759769779789799809819829839849859869879889899909919929939949959969979989991000100110021003100410051006100710081009101010111012101310141015101610171018101910201021102210231024102510261027102810291030103110321033103410351036103710381039104010411042104310441045104610471048104910501051105210531054105510561057105810591060106110621063106410651066106710681069107010711072107310741075107610771078107910801081108210831084108510861087108810891090109110921093109410951096109710981099110011011102110311041105110611071108110911101111111211131114111511161117111811191120112111221123112411251126112711281129113011311132113311341135113611371138113911401141114211431144114511461147114811491150115111521153115411551156115711581159116011611162116311641165116611671168116911701171117211731174117511761177117811791180118111821183118411851186118711881189119011911192119311941195119611971198119912001201120212031204120512061207120812091210121112121213121412151216121712181219122012211222122312241225122612271228122912301231123212331234123512361237123812391240124112421243124412451246124712481249125012511252125312541255125612571258125912601261126212631264126512661267126812691270127112721273127412751276127712781279128012811282128312841285128612871288128912901291129212931294129512961297129812991300130113021303130413051306130713081309131013111312131313141315131613171318131913201321132213231324132513261327132813291330133113321333133413351336133713381339134013411342134313441345134613471348134913501351135213531354135513561357135813591360136113621363136413651366136713681369137013711372137313741375137613771378137913801381138213831384138513861387138813891390139113921393139413951396139713981399140014011402140314041405140614071408140914101411141214131414141514161417141814191420142114221423142414251426142714281429143014311432143314341435143614371438143914401441144214431444144514461447144814491450145114521453145414551456145714581459146014611462146314641465146614671468146914701471147214731474147514761477147814791480148114821483148414851486148714881489149014911492149314941495149614971498149915001501150215031504150515061507150815091510151115121513151415151516151715181519152015211522152315241525152615271528152915301531153215331534153515361537153815391540154115421543154415451546154715481549155015511552155315541555155615571558155915601561156215631564156515661567156815691570157115721573157415751576157715781579158015811582158315841585158615871588158915901591159215931594159515961597159815991600160116021603160416051606160716081609161016111612161316141615161616171618161916201621162216231624162516261627162816291630163116321633163416351636163716381639164016411642164316441645164616471648164916501651165216531654165516561657165816591660166116621663166416651666166716681669167016711672167316741675167616771678167916801681168216831684168516861687168816891690169116921693169416951696169716981699170017011702170317041705170617071708170917101711171217131714171517161717171817191720172117221723172417251726172717281729173017311732173317341735173617371738173917401741174217431744174517461747174817491750175117521753175417551756175717581759176017611762176317641765176617671768176917701771177217731774177517761777177817791780178117821783178417851786178717881789179017911792179317941795179617971798179918001801180218031804180518061807180818091810181118121813181418151816181718181819182018211822182318241825182618271828182918301831183218331834183518361837183818391840184118421843184418451846184718481849185018511852185318541855185618571858185918601861186218631864186518661867186818691870187118721873187418751876187718781879188018811882188318841885188618871888188918901891189218931894189518961897189818991900190119021903190419051906190719081909191019111912191319141915191619171918191919201921192219231924192519261927192819291930193119321933193419351936193719381939194019411942194319441945194619471948194919501951195219531954195519561957195819591960196119621963196419651966196719681969197019711972197319741975197619771978197919801981198219831984198519861987198819891990199119921993199419951996199719981999200020012002200320042005200620072008200920102011201220132014201520162017201820192020202120222023202420252026202720282029203020312032203320342035203620372038203920402041204220432044204520462047204820492050205120522053205420552056205720582059206020612062206320642065206620672068206920702071207220732074207520762077207820792080208120822083208420852086208720882089209020912092209320942095209620972098209921002101210221032104210521062107210821092110211121122113211421152116211721182119212021212122212321242125212621272128212921302131213221332134213521362137213821392140214121422143214421452146214721482149215021512152215321542155215621572158215921602161216221632164216521662167216821692170217121722173217421752176217721782179218021812182218321842185218621872188218921902191219221932194219521962197219821992200220122022203220422052206220722082209221022112212221322142215221622172218221922202221222222232224222522262227222822292230223122322233223422352236223722382239224022412242224322442245224622472248224922502251225222532254225522562257225822592260226122622263226422652266226722682269227022712272227322742275227622772278227922802281228222832284228522862287228822892290229122922293229422952296229722982299230023012302230323042305230623072308230923102311231223132314231523162317231823192320232123222323232423252326232723282329233023312332233323342335233623372338233923402341234223432344234523462347234823492350235123522353235423552356235723582359236023612362236323642365236623672368236923702371237223732374237523762377237823792380238123822383238423852386238723882389239023912392239323942395239623972398239924002401240224032404240524062407240824092410241124122413241424152416241724182419242024212422242324242425242624272428242924302431243224332434243524362437243824392440244124422443244424452446244724482449245024512452245324542455245624572458245924602461246224632464246524662467246824692470247124722473247424752476247724782479248024812482248324842485248624872488248924902491249224932494249524962497249824992500250125022503250425052506250725082509251025112512251325142515251625172518251925202521252225232524252525262527252825292530253125322533253425352536253725382539254025412542254325442545254625472548254925502551255225532554255525562557255825592560256125622563256425652566256725682569257025712572257325742575257625772578257925802581258225832584258525862587258825892590259125922593259425952596259725982599260026012602260326042605260626072608260926102611261226132614261526162617261826192620262126222623262426252626262726282629263026312632263326342635263626372638263926402641264226432644264526462647264826492650265126522653265426552656265726582659266026612662266326642665266626672668266926702671267226732674267526762677267826792680268126822683268426852686268726882689269026912692269326942695269626972698269927002701270227032704270527062707270827092710271127122713271427152716271727182719272027212722272327242725272627272728272927302731273227332734273527362737273827392740274127422743274427452746274727482749275027512752275327542755275627572758275927602761276227632764276527662767276827692770277127722773277427752776277727782779278027812782278327842785278627872788278927902791279227932794279527962797279827992800280128022803280428052806280728082809281028112812281328142815281628172818281928202821282228232824282528262827282828292830283128322833283428352836283728382839284028412842284328442845284628472848284928502851285228532854285528562857285828592860286128622863286428652866286728682869287028712872287328742875287628772878287928802881288228832884288528862887288828892890289128922893289428952896289728982899290029012902290329042905290629072908290929102911291229132914291529162917291829192920292129222923292429252926292729282929293029312932293329342935293629372938293929402941294229432944294529462947294829492950295129522953295429552956295729582959296029612962296329642965296629672968296929702971297229732974297529762977297829792980298129822983298429852986298729882989299029912992299329942995299629972998299930003001300230033004300530063007300830093010301130123013301430153016301730183019302030213022302330243025302630273028302930303031303230333034303530363037303830393040304130423043304430453046304730483049305030513052305330543055305630573058305930603061306230633064306530663067306830693070307130723073307430753076307730783079308030813082308330843085308630873088308930903091309230933094309530963097309830993100310131023103310431053106310731083109311031113112311331143115311631173118311931203121312231233124312531263127312831293130313131323133313431353136313731383139314031413142314331443145314631473148314931503151315231533154315531563157315831593160316131623163316431653166316731683169317031713172317331743175317631773178317931803181318231833184318531863187318831893190319131923193319431953196319731983199320032013202320332043205320632073208320932103211321232133214321532163217321832193220322132223223322432253226322732283229323032313232323332343235323632373238323932403241324232433244324532463247324832493250325132523253325432553256325732583259326032613262326332643265326632673268326932703271327232733274327532763277327832793280328132823283328432853286328732883289329032913292329332943295329632973298329933003301330233033304330533063307330833093310331133123313331433153316331733183319332033213322332333243325332633273328332933303331333233333334333533363337333833393340334133423343334433453346334733483349335033513352335333543355335633573358335933603361336233633364336533663367336833693370337133723373337433753376337733783379338033813382338333843385338633873388338933903391339233933394339533963397339833993400340134023403340434053406340734083409341034113412341334143415341634173418341934203421342234233424342534263427342834293430343134323433343434353436343734383439344034413442344334443445344634473448344934503451345234533454345534563457345834593460346134623463346434653466346734683469347034713472347334743475347634773478347934803481348234833484348534863487348834893490349134923493349434953496349734983499350035013502350335043505350635073508350935103511351235133514351535163517351835193520352135223523352435253526352735283529353035313532353335343535353635373538353935403541354235433544354535463547354835493550355135523553355435553556355735583559356035613562356335643565356635673568356935703571357235733574357535763577357835793580358135823583358435853586358735883589359035913592359335943595359635973598359936003601360236033604360536063607360836093610361136123613361436153616361736183619362036213622362336243625362636273628362936303631363236333634363536363637363836393640364136423643364436453646364736483649365036513652365336543655365636573658365936603661366236633664366536663667366836693670367136723673367436753676367736783679368036813682368336843685368636873688368936903691369236933694369536963697369836993700370137023703370437053706370737083709371037113712371337143715371637173718371937203721372237233724372537263727372837293730373137323733373437353736373737383739374037413742374337443745374637473748374937503751375237533754375537563757375837593760376137623763376437653766376737683769377037713772377337743775377637773778377937803781378237833784378537863787378837893790379137923793379437953796379737983799380038013802380338043805380638073808380938103811381238133814381538163817381838193820382138223823382438253826382738283829383038313832383338343835383638373838383938403841384238433844384538463847384838493850385138523853385438553856385738583859386038613862386338643865386638673868386938703871387238733874387538763877387838793880388138823883388438853886388738883889389038913892389338943895389638973898389939003901390239033904390539063907390839093910391139123913391439153916391739183919392039213922392339243925392639273928392939303931393239333934393539363937393839393940394139423943394439453946394739483949395039513952395339543955395639573958395939603961396239633964396539663967396839693970397139723973397439753976397739783979398039813982398339843985398639873988398939903991399239933994399539963997399839994000400140024003400440054006400740084009401040114012401340144015401640174018401940204021402240234024402540264027402840294030403140324033403440354036403740384039404040414042404340444045404640474048404940504051405240534054405540564057405840594060406140624063406440654066406740684069407040714072407340744075407640774078407940804081408240834084408540864087408840894090409140924093409440954096409740984099410041014102410341044105410641074108410941104111411241134114411541164117411841194120412141224123412441254126412741284129413041314132413341344135413641374138413941404141414241434144414541464147414841494150415141524153415441554156415741584159416041614162416341644165416641674168416941704171417241734174417541764177417841794180418141824183418441854186418741884189419041914192419341944195419641974198419942004201420242034204420542064207420842094210421142124213421442154216421742184219422042214222422342244225422642274228422942304231423242334234423542364237423842394240424142424243424442454246424742484249425042514252425342544255425642574258425942604261426242634264426542664267426842694270427142724273427442754276427742784279428042814282428342844285428642874288428942904291429242934294429542964297429842994300430143024303430443054306430743084309431043114312431343144315431643174318431943204321432243234324432543264327432843294330433143324333433443354336433743384339434043414342434343444345434643474348434943504351435243534354435543564357435843594360436143624363436443654366436743684369437043714372437343744375437643774378437943804381438243834384438543864387438843894390439143924393439443954396439743984399440044014402440344044405440644074408440944104411441244134414441544164417441844194420442144224423442444254426442744284429443044314432443344344435443644374438443944404441444244434444444544464447444844494450445144524453445444554456445744584459446044614462446344644465446644674468446944704471447244734474447544764477447844794480448144824483448444854486448744884489449044914492449344944495449644974498449945004501450245034504450545064507450845094510451145124513451445154516451745184519452045214522452345244525452645274528452945304531453245334534453545364537453845394540454145424543454445454546454745484549455045514552455345544555455645574558455945604561456245634564456545664567456845694570457145724573457445754576457745784579458045814582458345844585458645874588458945904591459245934594459545964597459845994600460146024603460446054606460746084609461046114612461346144615461646174618461946204621462246234624462546264627462846294630463146324633463446354636463746384639464046414642464346444645464646474648464946504651465246534654465546564657465846594660466146624663466446654666466746684669467046714672467346744675467646774678467946804681468246834684468546864687468846894690469146924693469446954696469746984699470047014702470347044705470647074708470947104711471247134714471547164717471847194720472147224723472447254726472747284729473047314732473347344735473647374738473947404741474247434744474547464747474847494750475147524753475447554756475747584759476047614762476347644765476647674768476947704771477247734774477547764777477847794780478147824783478447854786478747884789479047914792
  1. /*
  2. * Copyright (c) 2000-2006 Silicon Graphics, Inc.
  3. * All Rights Reserved.
  4. *
  5. * This program is free software; you can redistribute it and/or
  6. * modify it under the terms of the GNU General Public License as
  7. * published by the Free Software Foundation.
  8. *
  9. * This program is distributed in the hope that it would be useful,
  10. * but WITHOUT ANY WARRANTY; without even the implied warranty of
  11. * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
  12. * GNU General Public License for more details.
  13. *
  14. * You should have received a copy of the GNU General Public License
  15. * along with this program; if not, write the Free Software Foundation,
  16. * Inc., 51 Franklin St, Fifth Floor, Boston, MA 02110-1301 USA
  17. */
  18. #include "xfs.h"
  19. #include "xfs_fs.h"
  20. #include "xfs_types.h"
  21. #include "xfs_bit.h"
  22. #include "xfs_log.h"
  23. #include "xfs_inum.h"
  24. #include "xfs_imap.h"
  25. #include "xfs_trans.h"
  26. #include "xfs_trans_priv.h"
  27. #include "xfs_sb.h"
  28. #include "xfs_ag.h"
  29. #include "xfs_dir2.h"
  30. #include "xfs_dmapi.h"
  31. #include "xfs_mount.h"
  32. #include "xfs_bmap_btree.h"
  33. #include "xfs_alloc_btree.h"
  34. #include "xfs_ialloc_btree.h"
  35. #include "xfs_dir2_sf.h"
  36. #include "xfs_attr_sf.h"
  37. #include "xfs_dinode.h"
  38. #include "xfs_inode.h"
  39. #include "xfs_buf_item.h"
  40. #include "xfs_inode_item.h"
  41. #include "xfs_btree.h"
  42. #include "xfs_alloc.h"
  43. #include "xfs_ialloc.h"
  44. #include "xfs_bmap.h"
  45. #include "xfs_rw.h"
  46. #include "xfs_error.h"
  47. #include "xfs_utils.h"
  48. #include "xfs_dir2_trace.h"
  49. #include "xfs_quota.h"
  50. #include "xfs_acl.h"
  51. #include <linux/log2.h>
  52. kmem_zone_t *xfs_ifork_zone;
  53. kmem_zone_t *xfs_inode_zone;
  54. kmem_zone_t *xfs_chashlist_zone;
  55. /*
  56. * Used in xfs_itruncate(). This is the maximum number of extents
  57. * freed from a file in a single transaction.
  58. */
  59. #define XFS_ITRUNC_MAX_EXTENTS 2
  60. STATIC int xfs_iflush_int(xfs_inode_t *, xfs_buf_t *);
  61. STATIC int xfs_iformat_local(xfs_inode_t *, xfs_dinode_t *, int, int);
  62. STATIC int xfs_iformat_extents(xfs_inode_t *, xfs_dinode_t *, int);
  63. STATIC int xfs_iformat_btree(xfs_inode_t *, xfs_dinode_t *, int);
  64. #ifdef DEBUG
  65. /*
  66. * Make sure that the extents in the given memory buffer
  67. * are valid.
  68. */
  69. STATIC void
  70. xfs_validate_extents(
  71. xfs_ifork_t *ifp,
  72. int nrecs,
  73. int disk,
  74. xfs_exntfmt_t fmt)
  75. {
  76. xfs_bmbt_rec_t *ep;
  77. xfs_bmbt_irec_t irec;
  78. xfs_bmbt_rec_t rec;
  79. int i;
  80. for (i = 0; i < nrecs; i++) {
  81. ep = xfs_iext_get_ext(ifp, i);
  82. rec.l0 = get_unaligned((__uint64_t*)&ep->l0);
  83. rec.l1 = get_unaligned((__uint64_t*)&ep->l1);
  84. if (disk)
  85. xfs_bmbt_disk_get_all(&rec, &irec);
  86. else
  87. xfs_bmbt_get_all(&rec, &irec);
  88. if (fmt == XFS_EXTFMT_NOSTATE)
  89. ASSERT(irec.br_state == XFS_EXT_NORM);
  90. }
  91. }
  92. #else /* DEBUG */
  93. #define xfs_validate_extents(ifp, nrecs, disk, fmt)
  94. #endif /* DEBUG */
  95. /*
  96. * Check that none of the inode's in the buffer have a next
  97. * unlinked field of 0.
  98. */
  99. #if defined(DEBUG)
  100. void
  101. xfs_inobp_check(
  102. xfs_mount_t *mp,
  103. xfs_buf_t *bp)
  104. {
  105. int i;
  106. int j;
  107. xfs_dinode_t *dip;
  108. j = mp->m_inode_cluster_size >> mp->m_sb.sb_inodelog;
  109. for (i = 0; i < j; i++) {
  110. dip = (xfs_dinode_t *)xfs_buf_offset(bp,
  111. i * mp->m_sb.sb_inodesize);
  112. if (!dip->di_next_unlinked) {
  113. xfs_fs_cmn_err(CE_ALERT, mp,
  114. "Detected a bogus zero next_unlinked field in incore inode buffer 0x%p. About to pop an ASSERT.",
  115. bp);
  116. ASSERT(dip->di_next_unlinked);
  117. }
  118. }
  119. }
  120. #endif
  121. /*
  122. * This routine is called to map an inode number within a file
  123. * system to the buffer containing the on-disk version of the
  124. * inode. It returns a pointer to the buffer containing the
  125. * on-disk inode in the bpp parameter, and in the dip parameter
  126. * it returns a pointer to the on-disk inode within that buffer.
  127. *
  128. * If a non-zero error is returned, then the contents of bpp and
  129. * dipp are undefined.
  130. *
  131. * Use xfs_imap() to determine the size and location of the
  132. * buffer to read from disk.
  133. */
  134. STATIC int
  135. xfs_inotobp(
  136. xfs_mount_t *mp,
  137. xfs_trans_t *tp,
  138. xfs_ino_t ino,
  139. xfs_dinode_t **dipp,
  140. xfs_buf_t **bpp,
  141. int *offset)
  142. {
  143. int di_ok;
  144. xfs_imap_t imap;
  145. xfs_buf_t *bp;
  146. int error;
  147. xfs_dinode_t *dip;
  148. /*
  149. * Call the space management code to find the location of the
  150. * inode on disk.
  151. */
  152. imap.im_blkno = 0;
  153. error = xfs_imap(mp, tp, ino, &imap, XFS_IMAP_LOOKUP);
  154. if (error != 0) {
  155. cmn_err(CE_WARN,
  156. "xfs_inotobp: xfs_imap() returned an "
  157. "error %d on %s. Returning error.", error, mp->m_fsname);
  158. return error;
  159. }
  160. /*
  161. * If the inode number maps to a block outside the bounds of the
  162. * file system then return NULL rather than calling read_buf
  163. * and panicing when we get an error from the driver.
  164. */
  165. if ((imap.im_blkno + imap.im_len) >
  166. XFS_FSB_TO_BB(mp, mp->m_sb.sb_dblocks)) {
  167. cmn_err(CE_WARN,
  168. "xfs_inotobp: inode number (%llu + %d) maps to a block outside the bounds "
  169. "of the file system %s. Returning EINVAL.",
  170. (unsigned long long)imap.im_blkno,
  171. imap.im_len, mp->m_fsname);
  172. return XFS_ERROR(EINVAL);
  173. }
  174. /*
  175. * Read in the buffer. If tp is NULL, xfs_trans_read_buf() will
  176. * default to just a read_buf() call.
  177. */
  178. error = xfs_trans_read_buf(mp, tp, mp->m_ddev_targp, imap.im_blkno,
  179. (int)imap.im_len, XFS_BUF_LOCK, &bp);
  180. if (error) {
  181. cmn_err(CE_WARN,
  182. "xfs_inotobp: xfs_trans_read_buf() returned an "
  183. "error %d on %s. Returning error.", error, mp->m_fsname);
  184. return error;
  185. }
  186. dip = (xfs_dinode_t *)xfs_buf_offset(bp, 0);
  187. di_ok =
  188. INT_GET(dip->di_core.di_magic, ARCH_CONVERT) == XFS_DINODE_MAGIC &&
  189. XFS_DINODE_GOOD_VERSION(INT_GET(dip->di_core.di_version, ARCH_CONVERT));
  190. if (unlikely(XFS_TEST_ERROR(!di_ok, mp, XFS_ERRTAG_ITOBP_INOTOBP,
  191. XFS_RANDOM_ITOBP_INOTOBP))) {
  192. XFS_CORRUPTION_ERROR("xfs_inotobp", XFS_ERRLEVEL_LOW, mp, dip);
  193. xfs_trans_brelse(tp, bp);
  194. cmn_err(CE_WARN,
  195. "xfs_inotobp: XFS_TEST_ERROR() returned an "
  196. "error on %s. Returning EFSCORRUPTED.", mp->m_fsname);
  197. return XFS_ERROR(EFSCORRUPTED);
  198. }
  199. xfs_inobp_check(mp, bp);
  200. /*
  201. * Set *dipp to point to the on-disk inode in the buffer.
  202. */
  203. *dipp = (xfs_dinode_t *)xfs_buf_offset(bp, imap.im_boffset);
  204. *bpp = bp;
  205. *offset = imap.im_boffset;
  206. return 0;
  207. }
  208. /*
  209. * This routine is called to map an inode to the buffer containing
  210. * the on-disk version of the inode. It returns a pointer to the
  211. * buffer containing the on-disk inode in the bpp parameter, and in
  212. * the dip parameter it returns a pointer to the on-disk inode within
  213. * that buffer.
  214. *
  215. * If a non-zero error is returned, then the contents of bpp and
  216. * dipp are undefined.
  217. *
  218. * If the inode is new and has not yet been initialized, use xfs_imap()
  219. * to determine the size and location of the buffer to read from disk.
  220. * If the inode has already been mapped to its buffer and read in once,
  221. * then use the mapping information stored in the inode rather than
  222. * calling xfs_imap(). This allows us to avoid the overhead of looking
  223. * at the inode btree for small block file systems (see xfs_dilocate()).
  224. * We can tell whether the inode has been mapped in before by comparing
  225. * its disk block address to 0. Only uninitialized inodes will have
  226. * 0 for the disk block address.
  227. */
  228. int
  229. xfs_itobp(
  230. xfs_mount_t *mp,
  231. xfs_trans_t *tp,
  232. xfs_inode_t *ip,
  233. xfs_dinode_t **dipp,
  234. xfs_buf_t **bpp,
  235. xfs_daddr_t bno,
  236. uint imap_flags)
  237. {
  238. xfs_imap_t imap;
  239. xfs_buf_t *bp;
  240. int error;
  241. int i;
  242. int ni;
  243. if (ip->i_blkno == (xfs_daddr_t)0) {
  244. /*
  245. * Call the space management code to find the location of the
  246. * inode on disk.
  247. */
  248. imap.im_blkno = bno;
  249. if ((error = xfs_imap(mp, tp, ip->i_ino, &imap,
  250. XFS_IMAP_LOOKUP | imap_flags)))
  251. return error;
  252. /*
  253. * If the inode number maps to a block outside the bounds
  254. * of the file system then return NULL rather than calling
  255. * read_buf and panicing when we get an error from the
  256. * driver.
  257. */
  258. if ((imap.im_blkno + imap.im_len) >
  259. XFS_FSB_TO_BB(mp, mp->m_sb.sb_dblocks)) {
  260. #ifdef DEBUG
  261. xfs_fs_cmn_err(CE_ALERT, mp, "xfs_itobp: "
  262. "(imap.im_blkno (0x%llx) "
  263. "+ imap.im_len (0x%llx)) > "
  264. " XFS_FSB_TO_BB(mp, "
  265. "mp->m_sb.sb_dblocks) (0x%llx)",
  266. (unsigned long long) imap.im_blkno,
  267. (unsigned long long) imap.im_len,
  268. XFS_FSB_TO_BB(mp, mp->m_sb.sb_dblocks));
  269. #endif /* DEBUG */
  270. return XFS_ERROR(EINVAL);
  271. }
  272. /*
  273. * Fill in the fields in the inode that will be used to
  274. * map the inode to its buffer from now on.
  275. */
  276. ip->i_blkno = imap.im_blkno;
  277. ip->i_len = imap.im_len;
  278. ip->i_boffset = imap.im_boffset;
  279. } else {
  280. /*
  281. * We've already mapped the inode once, so just use the
  282. * mapping that we saved the first time.
  283. */
  284. imap.im_blkno = ip->i_blkno;
  285. imap.im_len = ip->i_len;
  286. imap.im_boffset = ip->i_boffset;
  287. }
  288. ASSERT(bno == 0 || bno == imap.im_blkno);
  289. /*
  290. * Read in the buffer. If tp is NULL, xfs_trans_read_buf() will
  291. * default to just a read_buf() call.
  292. */
  293. error = xfs_trans_read_buf(mp, tp, mp->m_ddev_targp, imap.im_blkno,
  294. (int)imap.im_len, XFS_BUF_LOCK, &bp);
  295. if (error) {
  296. #ifdef DEBUG
  297. xfs_fs_cmn_err(CE_ALERT, mp, "xfs_itobp: "
  298. "xfs_trans_read_buf() returned error %d, "
  299. "imap.im_blkno 0x%llx, imap.im_len 0x%llx",
  300. error, (unsigned long long) imap.im_blkno,
  301. (unsigned long long) imap.im_len);
  302. #endif /* DEBUG */
  303. return error;
  304. }
  305. /*
  306. * Validate the magic number and version of every inode in the buffer
  307. * (if DEBUG kernel) or the first inode in the buffer, otherwise.
  308. * No validation is done here in userspace (xfs_repair).
  309. */
  310. #if !defined(__KERNEL__)
  311. ni = 0;
  312. #elif defined(DEBUG)
  313. ni = BBTOB(imap.im_len) >> mp->m_sb.sb_inodelog;
  314. #else /* usual case */
  315. ni = 1;
  316. #endif
  317. for (i = 0; i < ni; i++) {
  318. int di_ok;
  319. xfs_dinode_t *dip;
  320. dip = (xfs_dinode_t *)xfs_buf_offset(bp,
  321. (i << mp->m_sb.sb_inodelog));
  322. di_ok = INT_GET(dip->di_core.di_magic, ARCH_CONVERT) == XFS_DINODE_MAGIC &&
  323. XFS_DINODE_GOOD_VERSION(INT_GET(dip->di_core.di_version, ARCH_CONVERT));
  324. if (unlikely(XFS_TEST_ERROR(!di_ok, mp,
  325. XFS_ERRTAG_ITOBP_INOTOBP,
  326. XFS_RANDOM_ITOBP_INOTOBP))) {
  327. if (imap_flags & XFS_IMAP_BULKSTAT) {
  328. xfs_trans_brelse(tp, bp);
  329. return XFS_ERROR(EINVAL);
  330. }
  331. #ifdef DEBUG
  332. cmn_err(CE_ALERT,
  333. "Device %s - bad inode magic/vsn "
  334. "daddr %lld #%d (magic=%x)",
  335. XFS_BUFTARG_NAME(mp->m_ddev_targp),
  336. (unsigned long long)imap.im_blkno, i,
  337. INT_GET(dip->di_core.di_magic, ARCH_CONVERT));
  338. #endif
  339. XFS_CORRUPTION_ERROR("xfs_itobp", XFS_ERRLEVEL_HIGH,
  340. mp, dip);
  341. xfs_trans_brelse(tp, bp);
  342. return XFS_ERROR(EFSCORRUPTED);
  343. }
  344. }
  345. xfs_inobp_check(mp, bp);
  346. /*
  347. * Mark the buffer as an inode buffer now that it looks good
  348. */
  349. XFS_BUF_SET_VTYPE(bp, B_FS_INO);
  350. /*
  351. * Set *dipp to point to the on-disk inode in the buffer.
  352. */
  353. *dipp = (xfs_dinode_t *)xfs_buf_offset(bp, imap.im_boffset);
  354. *bpp = bp;
  355. return 0;
  356. }
  357. /*
  358. * Move inode type and inode format specific information from the
  359. * on-disk inode to the in-core inode. For fifos, devs, and sockets
  360. * this means set if_rdev to the proper value. For files, directories,
  361. * and symlinks this means to bring in the in-line data or extent
  362. * pointers. For a file in B-tree format, only the root is immediately
  363. * brought in-core. The rest will be in-lined in if_extents when it
  364. * is first referenced (see xfs_iread_extents()).
  365. */
  366. STATIC int
  367. xfs_iformat(
  368. xfs_inode_t *ip,
  369. xfs_dinode_t *dip)
  370. {
  371. xfs_attr_shortform_t *atp;
  372. int size;
  373. int error;
  374. xfs_fsize_t di_size;
  375. ip->i_df.if_ext_max =
  376. XFS_IFORK_DSIZE(ip) / (uint)sizeof(xfs_bmbt_rec_t);
  377. error = 0;
  378. if (unlikely(
  379. INT_GET(dip->di_core.di_nextents, ARCH_CONVERT) +
  380. INT_GET(dip->di_core.di_anextents, ARCH_CONVERT) >
  381. INT_GET(dip->di_core.di_nblocks, ARCH_CONVERT))) {
  382. xfs_fs_repair_cmn_err(CE_WARN, ip->i_mount,
  383. "corrupt dinode %Lu, extent total = %d, nblocks = %Lu.",
  384. (unsigned long long)ip->i_ino,
  385. (int)(INT_GET(dip->di_core.di_nextents, ARCH_CONVERT)
  386. + INT_GET(dip->di_core.di_anextents, ARCH_CONVERT)),
  387. (unsigned long long)
  388. INT_GET(dip->di_core.di_nblocks, ARCH_CONVERT));
  389. XFS_CORRUPTION_ERROR("xfs_iformat(1)", XFS_ERRLEVEL_LOW,
  390. ip->i_mount, dip);
  391. return XFS_ERROR(EFSCORRUPTED);
  392. }
  393. if (unlikely(INT_GET(dip->di_core.di_forkoff, ARCH_CONVERT) > ip->i_mount->m_sb.sb_inodesize)) {
  394. xfs_fs_repair_cmn_err(CE_WARN, ip->i_mount,
  395. "corrupt dinode %Lu, forkoff = 0x%x.",
  396. (unsigned long long)ip->i_ino,
  397. (int)(INT_GET(dip->di_core.di_forkoff, ARCH_CONVERT)));
  398. XFS_CORRUPTION_ERROR("xfs_iformat(2)", XFS_ERRLEVEL_LOW,
  399. ip->i_mount, dip);
  400. return XFS_ERROR(EFSCORRUPTED);
  401. }
  402. switch (ip->i_d.di_mode & S_IFMT) {
  403. case S_IFIFO:
  404. case S_IFCHR:
  405. case S_IFBLK:
  406. case S_IFSOCK:
  407. if (unlikely(INT_GET(dip->di_core.di_format, ARCH_CONVERT) != XFS_DINODE_FMT_DEV)) {
  408. XFS_CORRUPTION_ERROR("xfs_iformat(3)", XFS_ERRLEVEL_LOW,
  409. ip->i_mount, dip);
  410. return XFS_ERROR(EFSCORRUPTED);
  411. }
  412. ip->i_d.di_size = 0;
  413. ip->i_size = 0;
  414. ip->i_df.if_u2.if_rdev = INT_GET(dip->di_u.di_dev, ARCH_CONVERT);
  415. break;
  416. case S_IFREG:
  417. case S_IFLNK:
  418. case S_IFDIR:
  419. switch (INT_GET(dip->di_core.di_format, ARCH_CONVERT)) {
  420. case XFS_DINODE_FMT_LOCAL:
  421. /*
  422. * no local regular files yet
  423. */
  424. if (unlikely((INT_GET(dip->di_core.di_mode, ARCH_CONVERT) & S_IFMT) == S_IFREG)) {
  425. xfs_fs_repair_cmn_err(CE_WARN, ip->i_mount,
  426. "corrupt inode %Lu "
  427. "(local format for regular file).",
  428. (unsigned long long) ip->i_ino);
  429. XFS_CORRUPTION_ERROR("xfs_iformat(4)",
  430. XFS_ERRLEVEL_LOW,
  431. ip->i_mount, dip);
  432. return XFS_ERROR(EFSCORRUPTED);
  433. }
  434. di_size = INT_GET(dip->di_core.di_size, ARCH_CONVERT);
  435. if (unlikely(di_size > XFS_DFORK_DSIZE(dip, ip->i_mount))) {
  436. xfs_fs_repair_cmn_err(CE_WARN, ip->i_mount,
  437. "corrupt inode %Lu "
  438. "(bad size %Ld for local inode).",
  439. (unsigned long long) ip->i_ino,
  440. (long long) di_size);
  441. XFS_CORRUPTION_ERROR("xfs_iformat(5)",
  442. XFS_ERRLEVEL_LOW,
  443. ip->i_mount, dip);
  444. return XFS_ERROR(EFSCORRUPTED);
  445. }
  446. size = (int)di_size;
  447. error = xfs_iformat_local(ip, dip, XFS_DATA_FORK, size);
  448. break;
  449. case XFS_DINODE_FMT_EXTENTS:
  450. error = xfs_iformat_extents(ip, dip, XFS_DATA_FORK);
  451. break;
  452. case XFS_DINODE_FMT_BTREE:
  453. error = xfs_iformat_btree(ip, dip, XFS_DATA_FORK);
  454. break;
  455. default:
  456. XFS_ERROR_REPORT("xfs_iformat(6)", XFS_ERRLEVEL_LOW,
  457. ip->i_mount);
  458. return XFS_ERROR(EFSCORRUPTED);
  459. }
  460. break;
  461. default:
  462. XFS_ERROR_REPORT("xfs_iformat(7)", XFS_ERRLEVEL_LOW, ip->i_mount);
  463. return XFS_ERROR(EFSCORRUPTED);
  464. }
  465. if (error) {
  466. return error;
  467. }
  468. if (!XFS_DFORK_Q(dip))
  469. return 0;
  470. ASSERT(ip->i_afp == NULL);
  471. ip->i_afp = kmem_zone_zalloc(xfs_ifork_zone, KM_SLEEP);
  472. ip->i_afp->if_ext_max =
  473. XFS_IFORK_ASIZE(ip) / (uint)sizeof(xfs_bmbt_rec_t);
  474. switch (INT_GET(dip->di_core.di_aformat, ARCH_CONVERT)) {
  475. case XFS_DINODE_FMT_LOCAL:
  476. atp = (xfs_attr_shortform_t *)XFS_DFORK_APTR(dip);
  477. size = be16_to_cpu(atp->hdr.totsize);
  478. error = xfs_iformat_local(ip, dip, XFS_ATTR_FORK, size);
  479. break;
  480. case XFS_DINODE_FMT_EXTENTS:
  481. error = xfs_iformat_extents(ip, dip, XFS_ATTR_FORK);
  482. break;
  483. case XFS_DINODE_FMT_BTREE:
  484. error = xfs_iformat_btree(ip, dip, XFS_ATTR_FORK);
  485. break;
  486. default:
  487. error = XFS_ERROR(EFSCORRUPTED);
  488. break;
  489. }
  490. if (error) {
  491. kmem_zone_free(xfs_ifork_zone, ip->i_afp);
  492. ip->i_afp = NULL;
  493. xfs_idestroy_fork(ip, XFS_DATA_FORK);
  494. }
  495. return error;
  496. }
  497. /*
  498. * The file is in-lined in the on-disk inode.
  499. * If it fits into if_inline_data, then copy
  500. * it there, otherwise allocate a buffer for it
  501. * and copy the data there. Either way, set
  502. * if_data to point at the data.
  503. * If we allocate a buffer for the data, make
  504. * sure that its size is a multiple of 4 and
  505. * record the real size in i_real_bytes.
  506. */
  507. STATIC int
  508. xfs_iformat_local(
  509. xfs_inode_t *ip,
  510. xfs_dinode_t *dip,
  511. int whichfork,
  512. int size)
  513. {
  514. xfs_ifork_t *ifp;
  515. int real_size;
  516. /*
  517. * If the size is unreasonable, then something
  518. * is wrong and we just bail out rather than crash in
  519. * kmem_alloc() or memcpy() below.
  520. */
  521. if (unlikely(size > XFS_DFORK_SIZE(dip, ip->i_mount, whichfork))) {
  522. xfs_fs_repair_cmn_err(CE_WARN, ip->i_mount,
  523. "corrupt inode %Lu "
  524. "(bad size %d for local fork, size = %d).",
  525. (unsigned long long) ip->i_ino, size,
  526. XFS_DFORK_SIZE(dip, ip->i_mount, whichfork));
  527. XFS_CORRUPTION_ERROR("xfs_iformat_local", XFS_ERRLEVEL_LOW,
  528. ip->i_mount, dip);
  529. return XFS_ERROR(EFSCORRUPTED);
  530. }
  531. ifp = XFS_IFORK_PTR(ip, whichfork);
  532. real_size = 0;
  533. if (size == 0)
  534. ifp->if_u1.if_data = NULL;
  535. else if (size <= sizeof(ifp->if_u2.if_inline_data))
  536. ifp->if_u1.if_data = ifp->if_u2.if_inline_data;
  537. else {
  538. real_size = roundup(size, 4);
  539. ifp->if_u1.if_data = kmem_alloc(real_size, KM_SLEEP);
  540. }
  541. ifp->if_bytes = size;
  542. ifp->if_real_bytes = real_size;
  543. if (size)
  544. memcpy(ifp->if_u1.if_data, XFS_DFORK_PTR(dip, whichfork), size);
  545. ifp->if_flags &= ~XFS_IFEXTENTS;
  546. ifp->if_flags |= XFS_IFINLINE;
  547. return 0;
  548. }
  549. /*
  550. * The file consists of a set of extents all
  551. * of which fit into the on-disk inode.
  552. * If there are few enough extents to fit into
  553. * the if_inline_ext, then copy them there.
  554. * Otherwise allocate a buffer for them and copy
  555. * them into it. Either way, set if_extents
  556. * to point at the extents.
  557. */
  558. STATIC int
  559. xfs_iformat_extents(
  560. xfs_inode_t *ip,
  561. xfs_dinode_t *dip,
  562. int whichfork)
  563. {
  564. xfs_bmbt_rec_t *ep, *dp;
  565. xfs_ifork_t *ifp;
  566. int nex;
  567. int size;
  568. int i;
  569. ifp = XFS_IFORK_PTR(ip, whichfork);
  570. nex = XFS_DFORK_NEXTENTS(dip, whichfork);
  571. size = nex * (uint)sizeof(xfs_bmbt_rec_t);
  572. /*
  573. * If the number of extents is unreasonable, then something
  574. * is wrong and we just bail out rather than crash in
  575. * kmem_alloc() or memcpy() below.
  576. */
  577. if (unlikely(size < 0 || size > XFS_DFORK_SIZE(dip, ip->i_mount, whichfork))) {
  578. xfs_fs_repair_cmn_err(CE_WARN, ip->i_mount,
  579. "corrupt inode %Lu ((a)extents = %d).",
  580. (unsigned long long) ip->i_ino, nex);
  581. XFS_CORRUPTION_ERROR("xfs_iformat_extents(1)", XFS_ERRLEVEL_LOW,
  582. ip->i_mount, dip);
  583. return XFS_ERROR(EFSCORRUPTED);
  584. }
  585. ifp->if_real_bytes = 0;
  586. if (nex == 0)
  587. ifp->if_u1.if_extents = NULL;
  588. else if (nex <= XFS_INLINE_EXTS)
  589. ifp->if_u1.if_extents = ifp->if_u2.if_inline_ext;
  590. else
  591. xfs_iext_add(ifp, 0, nex);
  592. ifp->if_bytes = size;
  593. if (size) {
  594. dp = (xfs_bmbt_rec_t *) XFS_DFORK_PTR(dip, whichfork);
  595. xfs_validate_extents(ifp, nex, 1, XFS_EXTFMT_INODE(ip));
  596. for (i = 0; i < nex; i++, dp++) {
  597. ep = xfs_iext_get_ext(ifp, i);
  598. ep->l0 = INT_GET(get_unaligned((__uint64_t*)&dp->l0),
  599. ARCH_CONVERT);
  600. ep->l1 = INT_GET(get_unaligned((__uint64_t*)&dp->l1),
  601. ARCH_CONVERT);
  602. }
  603. xfs_bmap_trace_exlist("xfs_iformat_extents", ip, nex,
  604. whichfork);
  605. if (whichfork != XFS_DATA_FORK ||
  606. XFS_EXTFMT_INODE(ip) == XFS_EXTFMT_NOSTATE)
  607. if (unlikely(xfs_check_nostate_extents(
  608. ifp, 0, nex))) {
  609. XFS_ERROR_REPORT("xfs_iformat_extents(2)",
  610. XFS_ERRLEVEL_LOW,
  611. ip->i_mount);
  612. return XFS_ERROR(EFSCORRUPTED);
  613. }
  614. }
  615. ifp->if_flags |= XFS_IFEXTENTS;
  616. return 0;
  617. }
  618. /*
  619. * The file has too many extents to fit into
  620. * the inode, so they are in B-tree format.
  621. * Allocate a buffer for the root of the B-tree
  622. * and copy the root into it. The i_extents
  623. * field will remain NULL until all of the
  624. * extents are read in (when they are needed).
  625. */
  626. STATIC int
  627. xfs_iformat_btree(
  628. xfs_inode_t *ip,
  629. xfs_dinode_t *dip,
  630. int whichfork)
  631. {
  632. xfs_bmdr_block_t *dfp;
  633. xfs_ifork_t *ifp;
  634. /* REFERENCED */
  635. int nrecs;
  636. int size;
  637. ifp = XFS_IFORK_PTR(ip, whichfork);
  638. dfp = (xfs_bmdr_block_t *)XFS_DFORK_PTR(dip, whichfork);
  639. size = XFS_BMAP_BROOT_SPACE(dfp);
  640. nrecs = XFS_BMAP_BROOT_NUMRECS(dfp);
  641. /*
  642. * blow out if -- fork has less extents than can fit in
  643. * fork (fork shouldn't be a btree format), root btree
  644. * block has more records than can fit into the fork,
  645. * or the number of extents is greater than the number of
  646. * blocks.
  647. */
  648. if (unlikely(XFS_IFORK_NEXTENTS(ip, whichfork) <= ifp->if_ext_max
  649. || XFS_BMDR_SPACE_CALC(nrecs) >
  650. XFS_DFORK_SIZE(dip, ip->i_mount, whichfork)
  651. || XFS_IFORK_NEXTENTS(ip, whichfork) > ip->i_d.di_nblocks)) {
  652. xfs_fs_repair_cmn_err(CE_WARN, ip->i_mount,
  653. "corrupt inode %Lu (btree).",
  654. (unsigned long long) ip->i_ino);
  655. XFS_ERROR_REPORT("xfs_iformat_btree", XFS_ERRLEVEL_LOW,
  656. ip->i_mount);
  657. return XFS_ERROR(EFSCORRUPTED);
  658. }
  659. ifp->if_broot_bytes = size;
  660. ifp->if_broot = kmem_alloc(size, KM_SLEEP);
  661. ASSERT(ifp->if_broot != NULL);
  662. /*
  663. * Copy and convert from the on-disk structure
  664. * to the in-memory structure.
  665. */
  666. xfs_bmdr_to_bmbt(dfp, XFS_DFORK_SIZE(dip, ip->i_mount, whichfork),
  667. ifp->if_broot, size);
  668. ifp->if_flags &= ~XFS_IFEXTENTS;
  669. ifp->if_flags |= XFS_IFBROOT;
  670. return 0;
  671. }
  672. /*
  673. * xfs_xlate_dinode_core - translate an xfs_inode_core_t between ondisk
  674. * and native format
  675. *
  676. * buf = on-disk representation
  677. * dip = native representation
  678. * dir = direction - +ve -> disk to native
  679. * -ve -> native to disk
  680. */
  681. void
  682. xfs_xlate_dinode_core(
  683. xfs_caddr_t buf,
  684. xfs_dinode_core_t *dip,
  685. int dir)
  686. {
  687. xfs_dinode_core_t *buf_core = (xfs_dinode_core_t *)buf;
  688. xfs_dinode_core_t *mem_core = (xfs_dinode_core_t *)dip;
  689. xfs_arch_t arch = ARCH_CONVERT;
  690. ASSERT(dir);
  691. INT_XLATE(buf_core->di_magic, mem_core->di_magic, dir, arch);
  692. INT_XLATE(buf_core->di_mode, mem_core->di_mode, dir, arch);
  693. INT_XLATE(buf_core->di_version, mem_core->di_version, dir, arch);
  694. INT_XLATE(buf_core->di_format, mem_core->di_format, dir, arch);
  695. INT_XLATE(buf_core->di_onlink, mem_core->di_onlink, dir, arch);
  696. INT_XLATE(buf_core->di_uid, mem_core->di_uid, dir, arch);
  697. INT_XLATE(buf_core->di_gid, mem_core->di_gid, dir, arch);
  698. INT_XLATE(buf_core->di_nlink, mem_core->di_nlink, dir, arch);
  699. INT_XLATE(buf_core->di_projid, mem_core->di_projid, dir, arch);
  700. if (dir > 0) {
  701. memcpy(mem_core->di_pad, buf_core->di_pad,
  702. sizeof(buf_core->di_pad));
  703. } else {
  704. memcpy(buf_core->di_pad, mem_core->di_pad,
  705. sizeof(buf_core->di_pad));
  706. }
  707. INT_XLATE(buf_core->di_flushiter, mem_core->di_flushiter, dir, arch);
  708. INT_XLATE(buf_core->di_atime.t_sec, mem_core->di_atime.t_sec,
  709. dir, arch);
  710. INT_XLATE(buf_core->di_atime.t_nsec, mem_core->di_atime.t_nsec,
  711. dir, arch);
  712. INT_XLATE(buf_core->di_mtime.t_sec, mem_core->di_mtime.t_sec,
  713. dir, arch);
  714. INT_XLATE(buf_core->di_mtime.t_nsec, mem_core->di_mtime.t_nsec,
  715. dir, arch);
  716. INT_XLATE(buf_core->di_ctime.t_sec, mem_core->di_ctime.t_sec,
  717. dir, arch);
  718. INT_XLATE(buf_core->di_ctime.t_nsec, mem_core->di_ctime.t_nsec,
  719. dir, arch);
  720. INT_XLATE(buf_core->di_size, mem_core->di_size, dir, arch);
  721. INT_XLATE(buf_core->di_nblocks, mem_core->di_nblocks, dir, arch);
  722. INT_XLATE(buf_core->di_extsize, mem_core->di_extsize, dir, arch);
  723. INT_XLATE(buf_core->di_nextents, mem_core->di_nextents, dir, arch);
  724. INT_XLATE(buf_core->di_anextents, mem_core->di_anextents, dir, arch);
  725. INT_XLATE(buf_core->di_forkoff, mem_core->di_forkoff, dir, arch);
  726. INT_XLATE(buf_core->di_aformat, mem_core->di_aformat, dir, arch);
  727. INT_XLATE(buf_core->di_dmevmask, mem_core->di_dmevmask, dir, arch);
  728. INT_XLATE(buf_core->di_dmstate, mem_core->di_dmstate, dir, arch);
  729. INT_XLATE(buf_core->di_flags, mem_core->di_flags, dir, arch);
  730. INT_XLATE(buf_core->di_gen, mem_core->di_gen, dir, arch);
  731. }
  732. STATIC uint
  733. _xfs_dic2xflags(
  734. __uint16_t di_flags)
  735. {
  736. uint flags = 0;
  737. if (di_flags & XFS_DIFLAG_ANY) {
  738. if (di_flags & XFS_DIFLAG_REALTIME)
  739. flags |= XFS_XFLAG_REALTIME;
  740. if (di_flags & XFS_DIFLAG_PREALLOC)
  741. flags |= XFS_XFLAG_PREALLOC;
  742. if (di_flags & XFS_DIFLAG_IMMUTABLE)
  743. flags |= XFS_XFLAG_IMMUTABLE;
  744. if (di_flags & XFS_DIFLAG_APPEND)
  745. flags |= XFS_XFLAG_APPEND;
  746. if (di_flags & XFS_DIFLAG_SYNC)
  747. flags |= XFS_XFLAG_SYNC;
  748. if (di_flags & XFS_DIFLAG_NOATIME)
  749. flags |= XFS_XFLAG_NOATIME;
  750. if (di_flags & XFS_DIFLAG_NODUMP)
  751. flags |= XFS_XFLAG_NODUMP;
  752. if (di_flags & XFS_DIFLAG_RTINHERIT)
  753. flags |= XFS_XFLAG_RTINHERIT;
  754. if (di_flags & XFS_DIFLAG_PROJINHERIT)
  755. flags |= XFS_XFLAG_PROJINHERIT;
  756. if (di_flags & XFS_DIFLAG_NOSYMLINKS)
  757. flags |= XFS_XFLAG_NOSYMLINKS;
  758. if (di_flags & XFS_DIFLAG_EXTSIZE)
  759. flags |= XFS_XFLAG_EXTSIZE;
  760. if (di_flags & XFS_DIFLAG_EXTSZINHERIT)
  761. flags |= XFS_XFLAG_EXTSZINHERIT;
  762. if (di_flags & XFS_DIFLAG_NODEFRAG)
  763. flags |= XFS_XFLAG_NODEFRAG;
  764. }
  765. return flags;
  766. }
  767. uint
  768. xfs_ip2xflags(
  769. xfs_inode_t *ip)
  770. {
  771. xfs_dinode_core_t *dic = &ip->i_d;
  772. return _xfs_dic2xflags(dic->di_flags) |
  773. (XFS_CFORK_Q(dic) ? XFS_XFLAG_HASATTR : 0);
  774. }
  775. uint
  776. xfs_dic2xflags(
  777. xfs_dinode_core_t *dic)
  778. {
  779. return _xfs_dic2xflags(INT_GET(dic->di_flags, ARCH_CONVERT)) |
  780. (XFS_CFORK_Q_DISK(dic) ? XFS_XFLAG_HASATTR : 0);
  781. }
  782. /*
  783. * Given a mount structure and an inode number, return a pointer
  784. * to a newly allocated in-core inode corresponding to the given
  785. * inode number.
  786. *
  787. * Initialize the inode's attributes and extent pointers if it
  788. * already has them (it will not if the inode has no links).
  789. */
  790. int
  791. xfs_iread(
  792. xfs_mount_t *mp,
  793. xfs_trans_t *tp,
  794. xfs_ino_t ino,
  795. xfs_inode_t **ipp,
  796. xfs_daddr_t bno,
  797. uint imap_flags)
  798. {
  799. xfs_buf_t *bp;
  800. xfs_dinode_t *dip;
  801. xfs_inode_t *ip;
  802. int error;
  803. ASSERT(xfs_inode_zone != NULL);
  804. ip = kmem_zone_zalloc(xfs_inode_zone, KM_SLEEP);
  805. ip->i_ino = ino;
  806. ip->i_mount = mp;
  807. spin_lock_init(&ip->i_flags_lock);
  808. /*
  809. * Get pointer's to the on-disk inode and the buffer containing it.
  810. * If the inode number refers to a block outside the file system
  811. * then xfs_itobp() will return NULL. In this case we should
  812. * return NULL as well. Set i_blkno to 0 so that xfs_itobp() will
  813. * know that this is a new incore inode.
  814. */
  815. error = xfs_itobp(mp, tp, ip, &dip, &bp, bno, imap_flags);
  816. if (error) {
  817. kmem_zone_free(xfs_inode_zone, ip);
  818. return error;
  819. }
  820. /*
  821. * Initialize inode's trace buffers.
  822. * Do this before xfs_iformat in case it adds entries.
  823. */
  824. #ifdef XFS_BMAP_TRACE
  825. ip->i_xtrace = ktrace_alloc(XFS_BMAP_KTRACE_SIZE, KM_SLEEP);
  826. #endif
  827. #ifdef XFS_BMBT_TRACE
  828. ip->i_btrace = ktrace_alloc(XFS_BMBT_KTRACE_SIZE, KM_SLEEP);
  829. #endif
  830. #ifdef XFS_RW_TRACE
  831. ip->i_rwtrace = ktrace_alloc(XFS_RW_KTRACE_SIZE, KM_SLEEP);
  832. #endif
  833. #ifdef XFS_ILOCK_TRACE
  834. ip->i_lock_trace = ktrace_alloc(XFS_ILOCK_KTRACE_SIZE, KM_SLEEP);
  835. #endif
  836. #ifdef XFS_DIR2_TRACE
  837. ip->i_dir_trace = ktrace_alloc(XFS_DIR2_KTRACE_SIZE, KM_SLEEP);
  838. #endif
  839. /*
  840. * If we got something that isn't an inode it means someone
  841. * (nfs or dmi) has a stale handle.
  842. */
  843. if (INT_GET(dip->di_core.di_magic, ARCH_CONVERT) != XFS_DINODE_MAGIC) {
  844. kmem_zone_free(xfs_inode_zone, ip);
  845. xfs_trans_brelse(tp, bp);
  846. #ifdef DEBUG
  847. xfs_fs_cmn_err(CE_ALERT, mp, "xfs_iread: "
  848. "dip->di_core.di_magic (0x%x) != "
  849. "XFS_DINODE_MAGIC (0x%x)",
  850. INT_GET(dip->di_core.di_magic, ARCH_CONVERT),
  851. XFS_DINODE_MAGIC);
  852. #endif /* DEBUG */
  853. return XFS_ERROR(EINVAL);
  854. }
  855. /*
  856. * If the on-disk inode is already linked to a directory
  857. * entry, copy all of the inode into the in-core inode.
  858. * xfs_iformat() handles copying in the inode format
  859. * specific information.
  860. * Otherwise, just get the truly permanent information.
  861. */
  862. if (dip->di_core.di_mode) {
  863. xfs_xlate_dinode_core((xfs_caddr_t)&dip->di_core,
  864. &(ip->i_d), 1);
  865. error = xfs_iformat(ip, dip);
  866. if (error) {
  867. kmem_zone_free(xfs_inode_zone, ip);
  868. xfs_trans_brelse(tp, bp);
  869. #ifdef DEBUG
  870. xfs_fs_cmn_err(CE_ALERT, mp, "xfs_iread: "
  871. "xfs_iformat() returned error %d",
  872. error);
  873. #endif /* DEBUG */
  874. return error;
  875. }
  876. } else {
  877. ip->i_d.di_magic = INT_GET(dip->di_core.di_magic, ARCH_CONVERT);
  878. ip->i_d.di_version = INT_GET(dip->di_core.di_version, ARCH_CONVERT);
  879. ip->i_d.di_gen = INT_GET(dip->di_core.di_gen, ARCH_CONVERT);
  880. ip->i_d.di_flushiter = INT_GET(dip->di_core.di_flushiter, ARCH_CONVERT);
  881. /*
  882. * Make sure to pull in the mode here as well in
  883. * case the inode is released without being used.
  884. * This ensures that xfs_inactive() will see that
  885. * the inode is already free and not try to mess
  886. * with the uninitialized part of it.
  887. */
  888. ip->i_d.di_mode = 0;
  889. /*
  890. * Initialize the per-fork minima and maxima for a new
  891. * inode here. xfs_iformat will do it for old inodes.
  892. */
  893. ip->i_df.if_ext_max =
  894. XFS_IFORK_DSIZE(ip) / (uint)sizeof(xfs_bmbt_rec_t);
  895. }
  896. INIT_LIST_HEAD(&ip->i_reclaim);
  897. /*
  898. * The inode format changed when we moved the link count and
  899. * made it 32 bits long. If this is an old format inode,
  900. * convert it in memory to look like a new one. If it gets
  901. * flushed to disk we will convert back before flushing or
  902. * logging it. We zero out the new projid field and the old link
  903. * count field. We'll handle clearing the pad field (the remains
  904. * of the old uuid field) when we actually convert the inode to
  905. * the new format. We don't change the version number so that we
  906. * can distinguish this from a real new format inode.
  907. */
  908. if (ip->i_d.di_version == XFS_DINODE_VERSION_1) {
  909. ip->i_d.di_nlink = ip->i_d.di_onlink;
  910. ip->i_d.di_onlink = 0;
  911. ip->i_d.di_projid = 0;
  912. }
  913. ip->i_delayed_blks = 0;
  914. ip->i_size = ip->i_d.di_size;
  915. /*
  916. * Mark the buffer containing the inode as something to keep
  917. * around for a while. This helps to keep recently accessed
  918. * meta-data in-core longer.
  919. */
  920. XFS_BUF_SET_REF(bp, XFS_INO_REF);
  921. /*
  922. * Use xfs_trans_brelse() to release the buffer containing the
  923. * on-disk inode, because it was acquired with xfs_trans_read_buf()
  924. * in xfs_itobp() above. If tp is NULL, this is just a normal
  925. * brelse(). If we're within a transaction, then xfs_trans_brelse()
  926. * will only release the buffer if it is not dirty within the
  927. * transaction. It will be OK to release the buffer in this case,
  928. * because inodes on disk are never destroyed and we will be
  929. * locking the new in-core inode before putting it in the hash
  930. * table where other processes can find it. Thus we don't have
  931. * to worry about the inode being changed just because we released
  932. * the buffer.
  933. */
  934. xfs_trans_brelse(tp, bp);
  935. *ipp = ip;
  936. return 0;
  937. }
  938. /*
  939. * Read in extents from a btree-format inode.
  940. * Allocate and fill in if_extents. Real work is done in xfs_bmap.c.
  941. */
  942. int
  943. xfs_iread_extents(
  944. xfs_trans_t *tp,
  945. xfs_inode_t *ip,
  946. int whichfork)
  947. {
  948. int error;
  949. xfs_ifork_t *ifp;
  950. xfs_extnum_t nextents;
  951. size_t size;
  952. if (unlikely(XFS_IFORK_FORMAT(ip, whichfork) != XFS_DINODE_FMT_BTREE)) {
  953. XFS_ERROR_REPORT("xfs_iread_extents", XFS_ERRLEVEL_LOW,
  954. ip->i_mount);
  955. return XFS_ERROR(EFSCORRUPTED);
  956. }
  957. nextents = XFS_IFORK_NEXTENTS(ip, whichfork);
  958. size = nextents * sizeof(xfs_bmbt_rec_t);
  959. ifp = XFS_IFORK_PTR(ip, whichfork);
  960. /*
  961. * We know that the size is valid (it's checked in iformat_btree)
  962. */
  963. ifp->if_lastex = NULLEXTNUM;
  964. ifp->if_bytes = ifp->if_real_bytes = 0;
  965. ifp->if_flags |= XFS_IFEXTENTS;
  966. xfs_iext_add(ifp, 0, nextents);
  967. error = xfs_bmap_read_extents(tp, ip, whichfork);
  968. if (error) {
  969. xfs_iext_destroy(ifp);
  970. ifp->if_flags &= ~XFS_IFEXTENTS;
  971. return error;
  972. }
  973. xfs_validate_extents(ifp, nextents, 0, XFS_EXTFMT_INODE(ip));
  974. return 0;
  975. }
  976. /*
  977. * Allocate an inode on disk and return a copy of its in-core version.
  978. * The in-core inode is locked exclusively. Set mode, nlink, and rdev
  979. * appropriately within the inode. The uid and gid for the inode are
  980. * set according to the contents of the given cred structure.
  981. *
  982. * Use xfs_dialloc() to allocate the on-disk inode. If xfs_dialloc()
  983. * has a free inode available, call xfs_iget()
  984. * to obtain the in-core version of the allocated inode. Finally,
  985. * fill in the inode and log its initial contents. In this case,
  986. * ialloc_context would be set to NULL and call_again set to false.
  987. *
  988. * If xfs_dialloc() does not have an available inode,
  989. * it will replenish its supply by doing an allocation. Since we can
  990. * only do one allocation within a transaction without deadlocks, we
  991. * must commit the current transaction before returning the inode itself.
  992. * In this case, therefore, we will set call_again to true and return.
  993. * The caller should then commit the current transaction, start a new
  994. * transaction, and call xfs_ialloc() again to actually get the inode.
  995. *
  996. * To ensure that some other process does not grab the inode that
  997. * was allocated during the first call to xfs_ialloc(), this routine
  998. * also returns the [locked] bp pointing to the head of the freelist
  999. * as ialloc_context. The caller should hold this buffer across
  1000. * the commit and pass it back into this routine on the second call.
  1001. */
  1002. int
  1003. xfs_ialloc(
  1004. xfs_trans_t *tp,
  1005. xfs_inode_t *pip,
  1006. mode_t mode,
  1007. xfs_nlink_t nlink,
  1008. xfs_dev_t rdev,
  1009. cred_t *cr,
  1010. xfs_prid_t prid,
  1011. int okalloc,
  1012. xfs_buf_t **ialloc_context,
  1013. boolean_t *call_again,
  1014. xfs_inode_t **ipp)
  1015. {
  1016. xfs_ino_t ino;
  1017. xfs_inode_t *ip;
  1018. bhv_vnode_t *vp;
  1019. uint flags;
  1020. int error;
  1021. /*
  1022. * Call the space management code to pick
  1023. * the on-disk inode to be allocated.
  1024. */
  1025. error = xfs_dialloc(tp, pip->i_ino, mode, okalloc,
  1026. ialloc_context, call_again, &ino);
  1027. if (error != 0) {
  1028. return error;
  1029. }
  1030. if (*call_again || ino == NULLFSINO) {
  1031. *ipp = NULL;
  1032. return 0;
  1033. }
  1034. ASSERT(*ialloc_context == NULL);
  1035. /*
  1036. * Get the in-core inode with the lock held exclusively.
  1037. * This is because we're setting fields here we need
  1038. * to prevent others from looking at until we're done.
  1039. */
  1040. error = xfs_trans_iget(tp->t_mountp, tp, ino,
  1041. XFS_IGET_CREATE, XFS_ILOCK_EXCL, &ip);
  1042. if (error != 0) {
  1043. return error;
  1044. }
  1045. ASSERT(ip != NULL);
  1046. vp = XFS_ITOV(ip);
  1047. ip->i_d.di_mode = (__uint16_t)mode;
  1048. ip->i_d.di_onlink = 0;
  1049. ip->i_d.di_nlink = nlink;
  1050. ASSERT(ip->i_d.di_nlink == nlink);
  1051. ip->i_d.di_uid = current_fsuid(cr);
  1052. ip->i_d.di_gid = current_fsgid(cr);
  1053. ip->i_d.di_projid = prid;
  1054. memset(&(ip->i_d.di_pad[0]), 0, sizeof(ip->i_d.di_pad));
  1055. /*
  1056. * If the superblock version is up to where we support new format
  1057. * inodes and this is currently an old format inode, then change
  1058. * the inode version number now. This way we only do the conversion
  1059. * here rather than here and in the flush/logging code.
  1060. */
  1061. if (XFS_SB_VERSION_HASNLINK(&tp->t_mountp->m_sb) &&
  1062. ip->i_d.di_version == XFS_DINODE_VERSION_1) {
  1063. ip->i_d.di_version = XFS_DINODE_VERSION_2;
  1064. /*
  1065. * We've already zeroed the old link count, the projid field,
  1066. * and the pad field.
  1067. */
  1068. }
  1069. /*
  1070. * Project ids won't be stored on disk if we are using a version 1 inode.
  1071. */
  1072. if ( (prid != 0) && (ip->i_d.di_version == XFS_DINODE_VERSION_1))
  1073. xfs_bump_ino_vers2(tp, ip);
  1074. if (XFS_INHERIT_GID(pip, vp->v_vfsp)) {
  1075. ip->i_d.di_gid = pip->i_d.di_gid;
  1076. if ((pip->i_d.di_mode & S_ISGID) && (mode & S_IFMT) == S_IFDIR) {
  1077. ip->i_d.di_mode |= S_ISGID;
  1078. }
  1079. }
  1080. /*
  1081. * If the group ID of the new file does not match the effective group
  1082. * ID or one of the supplementary group IDs, the S_ISGID bit is cleared
  1083. * (and only if the irix_sgid_inherit compatibility variable is set).
  1084. */
  1085. if ((irix_sgid_inherit) &&
  1086. (ip->i_d.di_mode & S_ISGID) &&
  1087. (!in_group_p((gid_t)ip->i_d.di_gid))) {
  1088. ip->i_d.di_mode &= ~S_ISGID;
  1089. }
  1090. ip->i_d.di_size = 0;
  1091. ip->i_size = 0;
  1092. ip->i_d.di_nextents = 0;
  1093. ASSERT(ip->i_d.di_nblocks == 0);
  1094. xfs_ichgtime(ip, XFS_ICHGTIME_CHG|XFS_ICHGTIME_ACC|XFS_ICHGTIME_MOD);
  1095. /*
  1096. * di_gen will have been taken care of in xfs_iread.
  1097. */
  1098. ip->i_d.di_extsize = 0;
  1099. ip->i_d.di_dmevmask = 0;
  1100. ip->i_d.di_dmstate = 0;
  1101. ip->i_d.di_flags = 0;
  1102. flags = XFS_ILOG_CORE;
  1103. switch (mode & S_IFMT) {
  1104. case S_IFIFO:
  1105. case S_IFCHR:
  1106. case S_IFBLK:
  1107. case S_IFSOCK:
  1108. ip->i_d.di_format = XFS_DINODE_FMT_DEV;
  1109. ip->i_df.if_u2.if_rdev = rdev;
  1110. ip->i_df.if_flags = 0;
  1111. flags |= XFS_ILOG_DEV;
  1112. break;
  1113. case S_IFREG:
  1114. case S_IFDIR:
  1115. if (unlikely(pip->i_d.di_flags & XFS_DIFLAG_ANY)) {
  1116. uint di_flags = 0;
  1117. if ((mode & S_IFMT) == S_IFDIR) {
  1118. if (pip->i_d.di_flags & XFS_DIFLAG_RTINHERIT)
  1119. di_flags |= XFS_DIFLAG_RTINHERIT;
  1120. if (pip->i_d.di_flags & XFS_DIFLAG_EXTSZINHERIT) {
  1121. di_flags |= XFS_DIFLAG_EXTSZINHERIT;
  1122. ip->i_d.di_extsize = pip->i_d.di_extsize;
  1123. }
  1124. } else if ((mode & S_IFMT) == S_IFREG) {
  1125. if (pip->i_d.di_flags & XFS_DIFLAG_RTINHERIT) {
  1126. di_flags |= XFS_DIFLAG_REALTIME;
  1127. ip->i_iocore.io_flags |= XFS_IOCORE_RT;
  1128. }
  1129. if (pip->i_d.di_flags & XFS_DIFLAG_EXTSZINHERIT) {
  1130. di_flags |= XFS_DIFLAG_EXTSIZE;
  1131. ip->i_d.di_extsize = pip->i_d.di_extsize;
  1132. }
  1133. }
  1134. if ((pip->i_d.di_flags & XFS_DIFLAG_NOATIME) &&
  1135. xfs_inherit_noatime)
  1136. di_flags |= XFS_DIFLAG_NOATIME;
  1137. if ((pip->i_d.di_flags & XFS_DIFLAG_NODUMP) &&
  1138. xfs_inherit_nodump)
  1139. di_flags |= XFS_DIFLAG_NODUMP;
  1140. if ((pip->i_d.di_flags & XFS_DIFLAG_SYNC) &&
  1141. xfs_inherit_sync)
  1142. di_flags |= XFS_DIFLAG_SYNC;
  1143. if ((pip->i_d.di_flags & XFS_DIFLAG_NOSYMLINKS) &&
  1144. xfs_inherit_nosymlinks)
  1145. di_flags |= XFS_DIFLAG_NOSYMLINKS;
  1146. if (pip->i_d.di_flags & XFS_DIFLAG_PROJINHERIT)
  1147. di_flags |= XFS_DIFLAG_PROJINHERIT;
  1148. if ((pip->i_d.di_flags & XFS_DIFLAG_NODEFRAG) &&
  1149. xfs_inherit_nodefrag)
  1150. di_flags |= XFS_DIFLAG_NODEFRAG;
  1151. ip->i_d.di_flags |= di_flags;
  1152. }
  1153. /* FALLTHROUGH */
  1154. case S_IFLNK:
  1155. ip->i_d.di_format = XFS_DINODE_FMT_EXTENTS;
  1156. ip->i_df.if_flags = XFS_IFEXTENTS;
  1157. ip->i_df.if_bytes = ip->i_df.if_real_bytes = 0;
  1158. ip->i_df.if_u1.if_extents = NULL;
  1159. break;
  1160. default:
  1161. ASSERT(0);
  1162. }
  1163. /*
  1164. * Attribute fork settings for new inode.
  1165. */
  1166. ip->i_d.di_aformat = XFS_DINODE_FMT_EXTENTS;
  1167. ip->i_d.di_anextents = 0;
  1168. /*
  1169. * Log the new values stuffed into the inode.
  1170. */
  1171. xfs_trans_log_inode(tp, ip, flags);
  1172. /* now that we have an i_mode we can setup inode ops and unlock */
  1173. bhv_vfs_init_vnode(XFS_MTOVFS(tp->t_mountp), vp, XFS_ITOBHV(ip), 1);
  1174. *ipp = ip;
  1175. return 0;
  1176. }
  1177. /*
  1178. * Check to make sure that there are no blocks allocated to the
  1179. * file beyond the size of the file. We don't check this for
  1180. * files with fixed size extents or real time extents, but we
  1181. * at least do it for regular files.
  1182. */
  1183. #ifdef DEBUG
  1184. void
  1185. xfs_isize_check(
  1186. xfs_mount_t *mp,
  1187. xfs_inode_t *ip,
  1188. xfs_fsize_t isize)
  1189. {
  1190. xfs_fileoff_t map_first;
  1191. int nimaps;
  1192. xfs_bmbt_irec_t imaps[2];
  1193. if ((ip->i_d.di_mode & S_IFMT) != S_IFREG)
  1194. return;
  1195. if (ip->i_d.di_flags & (XFS_DIFLAG_REALTIME | XFS_DIFLAG_EXTSIZE))
  1196. return;
  1197. nimaps = 2;
  1198. map_first = XFS_B_TO_FSB(mp, (xfs_ufsize_t)isize);
  1199. /*
  1200. * The filesystem could be shutting down, so bmapi may return
  1201. * an error.
  1202. */
  1203. if (xfs_bmapi(NULL, ip, map_first,
  1204. (XFS_B_TO_FSB(mp,
  1205. (xfs_ufsize_t)XFS_MAXIOFFSET(mp)) -
  1206. map_first),
  1207. XFS_BMAPI_ENTIRE, NULL, 0, imaps, &nimaps,
  1208. NULL, NULL))
  1209. return;
  1210. ASSERT(nimaps == 1);
  1211. ASSERT(imaps[0].br_startblock == HOLESTARTBLOCK);
  1212. }
  1213. #endif /* DEBUG */
  1214. /*
  1215. * Calculate the last possible buffered byte in a file. This must
  1216. * include data that was buffered beyond the EOF by the write code.
  1217. * This also needs to deal with overflowing the xfs_fsize_t type
  1218. * which can happen for sizes near the limit.
  1219. *
  1220. * We also need to take into account any blocks beyond the EOF. It
  1221. * may be the case that they were buffered by a write which failed.
  1222. * In that case the pages will still be in memory, but the inode size
  1223. * will never have been updated.
  1224. */
  1225. xfs_fsize_t
  1226. xfs_file_last_byte(
  1227. xfs_inode_t *ip)
  1228. {
  1229. xfs_mount_t *mp;
  1230. xfs_fsize_t last_byte;
  1231. xfs_fileoff_t last_block;
  1232. xfs_fileoff_t size_last_block;
  1233. int error;
  1234. ASSERT(ismrlocked(&(ip->i_iolock), MR_UPDATE | MR_ACCESS));
  1235. mp = ip->i_mount;
  1236. /*
  1237. * Only check for blocks beyond the EOF if the extents have
  1238. * been read in. This eliminates the need for the inode lock,
  1239. * and it also saves us from looking when it really isn't
  1240. * necessary.
  1241. */
  1242. if (ip->i_df.if_flags & XFS_IFEXTENTS) {
  1243. error = xfs_bmap_last_offset(NULL, ip, &last_block,
  1244. XFS_DATA_FORK);
  1245. if (error) {
  1246. last_block = 0;
  1247. }
  1248. } else {
  1249. last_block = 0;
  1250. }
  1251. size_last_block = XFS_B_TO_FSB(mp, (xfs_ufsize_t)ip->i_size);
  1252. last_block = XFS_FILEOFF_MAX(last_block, size_last_block);
  1253. last_byte = XFS_FSB_TO_B(mp, last_block);
  1254. if (last_byte < 0) {
  1255. return XFS_MAXIOFFSET(mp);
  1256. }
  1257. last_byte += (1 << mp->m_writeio_log);
  1258. if (last_byte < 0) {
  1259. return XFS_MAXIOFFSET(mp);
  1260. }
  1261. return last_byte;
  1262. }
  1263. #if defined(XFS_RW_TRACE)
  1264. STATIC void
  1265. xfs_itrunc_trace(
  1266. int tag,
  1267. xfs_inode_t *ip,
  1268. int flag,
  1269. xfs_fsize_t new_size,
  1270. xfs_off_t toss_start,
  1271. xfs_off_t toss_finish)
  1272. {
  1273. if (ip->i_rwtrace == NULL) {
  1274. return;
  1275. }
  1276. ktrace_enter(ip->i_rwtrace,
  1277. (void*)((long)tag),
  1278. (void*)ip,
  1279. (void*)(unsigned long)((ip->i_d.di_size >> 32) & 0xffffffff),
  1280. (void*)(unsigned long)(ip->i_d.di_size & 0xffffffff),
  1281. (void*)((long)flag),
  1282. (void*)(unsigned long)((new_size >> 32) & 0xffffffff),
  1283. (void*)(unsigned long)(new_size & 0xffffffff),
  1284. (void*)(unsigned long)((toss_start >> 32) & 0xffffffff),
  1285. (void*)(unsigned long)(toss_start & 0xffffffff),
  1286. (void*)(unsigned long)((toss_finish >> 32) & 0xffffffff),
  1287. (void*)(unsigned long)(toss_finish & 0xffffffff),
  1288. (void*)(unsigned long)current_cpu(),
  1289. (void*)(unsigned long)current_pid(),
  1290. (void*)NULL,
  1291. (void*)NULL,
  1292. (void*)NULL);
  1293. }
  1294. #else
  1295. #define xfs_itrunc_trace(tag, ip, flag, new_size, toss_start, toss_finish)
  1296. #endif
  1297. /*
  1298. * Start the truncation of the file to new_size. The new size
  1299. * must be smaller than the current size. This routine will
  1300. * clear the buffer and page caches of file data in the removed
  1301. * range, and xfs_itruncate_finish() will remove the underlying
  1302. * disk blocks.
  1303. *
  1304. * The inode must have its I/O lock locked EXCLUSIVELY, and it
  1305. * must NOT have the inode lock held at all. This is because we're
  1306. * calling into the buffer/page cache code and we can't hold the
  1307. * inode lock when we do so.
  1308. *
  1309. * We need to wait for any direct I/Os in flight to complete before we
  1310. * proceed with the truncate. This is needed to prevent the extents
  1311. * being read or written by the direct I/Os from being removed while the
  1312. * I/O is in flight as there is no other method of synchronising
  1313. * direct I/O with the truncate operation. Also, because we hold
  1314. * the IOLOCK in exclusive mode, we prevent new direct I/Os from being
  1315. * started until the truncate completes and drops the lock. Essentially,
  1316. * the vn_iowait() call forms an I/O barrier that provides strict ordering
  1317. * between direct I/Os and the truncate operation.
  1318. *
  1319. * The flags parameter can have either the value XFS_ITRUNC_DEFINITE
  1320. * or XFS_ITRUNC_MAYBE. The XFS_ITRUNC_MAYBE value should be used
  1321. * in the case that the caller is locking things out of order and
  1322. * may not be able to call xfs_itruncate_finish() with the inode lock
  1323. * held without dropping the I/O lock. If the caller must drop the
  1324. * I/O lock before calling xfs_itruncate_finish(), then xfs_itruncate_start()
  1325. * must be called again with all the same restrictions as the initial
  1326. * call.
  1327. */
  1328. int
  1329. xfs_itruncate_start(
  1330. xfs_inode_t *ip,
  1331. uint flags,
  1332. xfs_fsize_t new_size)
  1333. {
  1334. xfs_fsize_t last_byte;
  1335. xfs_off_t toss_start;
  1336. xfs_mount_t *mp;
  1337. bhv_vnode_t *vp;
  1338. int error = 0;
  1339. ASSERT(ismrlocked(&ip->i_iolock, MR_UPDATE) != 0);
  1340. ASSERT((new_size == 0) || (new_size <= ip->i_size));
  1341. ASSERT((flags == XFS_ITRUNC_DEFINITE) ||
  1342. (flags == XFS_ITRUNC_MAYBE));
  1343. mp = ip->i_mount;
  1344. vp = XFS_ITOV(ip);
  1345. vn_iowait(vp); /* wait for the completion of any pending DIOs */
  1346. /*
  1347. * Call toss_pages or flushinval_pages to get rid of pages
  1348. * overlapping the region being removed. We have to use
  1349. * the less efficient flushinval_pages in the case that the
  1350. * caller may not be able to finish the truncate without
  1351. * dropping the inode's I/O lock. Make sure
  1352. * to catch any pages brought in by buffers overlapping
  1353. * the EOF by searching out beyond the isize by our
  1354. * block size. We round new_size up to a block boundary
  1355. * so that we don't toss things on the same block as
  1356. * new_size but before it.
  1357. *
  1358. * Before calling toss_page or flushinval_pages, make sure to
  1359. * call remapf() over the same region if the file is mapped.
  1360. * This frees up mapped file references to the pages in the
  1361. * given range and for the flushinval_pages case it ensures
  1362. * that we get the latest mapped changes flushed out.
  1363. */
  1364. toss_start = XFS_B_TO_FSB(mp, (xfs_ufsize_t)new_size);
  1365. toss_start = XFS_FSB_TO_B(mp, toss_start);
  1366. if (toss_start < 0) {
  1367. /*
  1368. * The place to start tossing is beyond our maximum
  1369. * file size, so there is no way that the data extended
  1370. * out there.
  1371. */
  1372. return 0;
  1373. }
  1374. last_byte = xfs_file_last_byte(ip);
  1375. xfs_itrunc_trace(XFS_ITRUNC_START, ip, flags, new_size, toss_start,
  1376. last_byte);
  1377. if (last_byte > toss_start) {
  1378. if (flags & XFS_ITRUNC_DEFINITE) {
  1379. bhv_vop_toss_pages(vp, toss_start, -1, FI_REMAPF_LOCKED);
  1380. } else {
  1381. error = bhv_vop_flushinval_pages(vp, toss_start, -1, FI_REMAPF_LOCKED);
  1382. }
  1383. }
  1384. #ifdef DEBUG
  1385. if (new_size == 0) {
  1386. ASSERT(VN_CACHED(vp) == 0);
  1387. }
  1388. #endif
  1389. return error;
  1390. }
  1391. /*
  1392. * Shrink the file to the given new_size. The new
  1393. * size must be smaller than the current size.
  1394. * This will free up the underlying blocks
  1395. * in the removed range after a call to xfs_itruncate_start()
  1396. * or xfs_atruncate_start().
  1397. *
  1398. * The transaction passed to this routine must have made
  1399. * a permanent log reservation of at least XFS_ITRUNCATE_LOG_RES.
  1400. * This routine may commit the given transaction and
  1401. * start new ones, so make sure everything involved in
  1402. * the transaction is tidy before calling here.
  1403. * Some transaction will be returned to the caller to be
  1404. * committed. The incoming transaction must already include
  1405. * the inode, and both inode locks must be held exclusively.
  1406. * The inode must also be "held" within the transaction. On
  1407. * return the inode will be "held" within the returned transaction.
  1408. * This routine does NOT require any disk space to be reserved
  1409. * for it within the transaction.
  1410. *
  1411. * The fork parameter must be either xfs_attr_fork or xfs_data_fork,
  1412. * and it indicates the fork which is to be truncated. For the
  1413. * attribute fork we only support truncation to size 0.
  1414. *
  1415. * We use the sync parameter to indicate whether or not the first
  1416. * transaction we perform might have to be synchronous. For the attr fork,
  1417. * it needs to be so if the unlink of the inode is not yet known to be
  1418. * permanent in the log. This keeps us from freeing and reusing the
  1419. * blocks of the attribute fork before the unlink of the inode becomes
  1420. * permanent.
  1421. *
  1422. * For the data fork, we normally have to run synchronously if we're
  1423. * being called out of the inactive path or we're being called
  1424. * out of the create path where we're truncating an existing file.
  1425. * Either way, the truncate needs to be sync so blocks don't reappear
  1426. * in the file with altered data in case of a crash. wsync filesystems
  1427. * can run the first case async because anything that shrinks the inode
  1428. * has to run sync so by the time we're called here from inactive, the
  1429. * inode size is permanently set to 0.
  1430. *
  1431. * Calls from the truncate path always need to be sync unless we're
  1432. * in a wsync filesystem and the file has already been unlinked.
  1433. *
  1434. * The caller is responsible for correctly setting the sync parameter.
  1435. * It gets too hard for us to guess here which path we're being called
  1436. * out of just based on inode state.
  1437. */
  1438. int
  1439. xfs_itruncate_finish(
  1440. xfs_trans_t **tp,
  1441. xfs_inode_t *ip,
  1442. xfs_fsize_t new_size,
  1443. int fork,
  1444. int sync)
  1445. {
  1446. xfs_fsblock_t first_block;
  1447. xfs_fileoff_t first_unmap_block;
  1448. xfs_fileoff_t last_block;
  1449. xfs_filblks_t unmap_len=0;
  1450. xfs_mount_t *mp;
  1451. xfs_trans_t *ntp;
  1452. int done;
  1453. int committed;
  1454. xfs_bmap_free_t free_list;
  1455. int error;
  1456. ASSERT(ismrlocked(&ip->i_iolock, MR_UPDATE) != 0);
  1457. ASSERT(ismrlocked(&ip->i_lock, MR_UPDATE) != 0);
  1458. ASSERT((new_size == 0) || (new_size <= ip->i_size));
  1459. ASSERT(*tp != NULL);
  1460. ASSERT((*tp)->t_flags & XFS_TRANS_PERM_LOG_RES);
  1461. ASSERT(ip->i_transp == *tp);
  1462. ASSERT(ip->i_itemp != NULL);
  1463. ASSERT(ip->i_itemp->ili_flags & XFS_ILI_HOLD);
  1464. ntp = *tp;
  1465. mp = (ntp)->t_mountp;
  1466. ASSERT(! XFS_NOT_DQATTACHED(mp, ip));
  1467. /*
  1468. * We only support truncating the entire attribute fork.
  1469. */
  1470. if (fork == XFS_ATTR_FORK) {
  1471. new_size = 0LL;
  1472. }
  1473. first_unmap_block = XFS_B_TO_FSB(mp, (xfs_ufsize_t)new_size);
  1474. xfs_itrunc_trace(XFS_ITRUNC_FINISH1, ip, 0, new_size, 0, 0);
  1475. /*
  1476. * The first thing we do is set the size to new_size permanently
  1477. * on disk. This way we don't have to worry about anyone ever
  1478. * being able to look at the data being freed even in the face
  1479. * of a crash. What we're getting around here is the case where
  1480. * we free a block, it is allocated to another file, it is written
  1481. * to, and then we crash. If the new data gets written to the
  1482. * file but the log buffers containing the free and reallocation
  1483. * don't, then we'd end up with garbage in the blocks being freed.
  1484. * As long as we make the new_size permanent before actually
  1485. * freeing any blocks it doesn't matter if they get writtten to.
  1486. *
  1487. * The callers must signal into us whether or not the size
  1488. * setting here must be synchronous. There are a few cases
  1489. * where it doesn't have to be synchronous. Those cases
  1490. * occur if the file is unlinked and we know the unlink is
  1491. * permanent or if the blocks being truncated are guaranteed
  1492. * to be beyond the inode eof (regardless of the link count)
  1493. * and the eof value is permanent. Both of these cases occur
  1494. * only on wsync-mounted filesystems. In those cases, we're
  1495. * guaranteed that no user will ever see the data in the blocks
  1496. * that are being truncated so the truncate can run async.
  1497. * In the free beyond eof case, the file may wind up with
  1498. * more blocks allocated to it than it needs if we crash
  1499. * and that won't get fixed until the next time the file
  1500. * is re-opened and closed but that's ok as that shouldn't
  1501. * be too many blocks.
  1502. *
  1503. * However, we can't just make all wsync xactions run async
  1504. * because there's one call out of the create path that needs
  1505. * to run sync where it's truncating an existing file to size
  1506. * 0 whose size is > 0.
  1507. *
  1508. * It's probably possible to come up with a test in this
  1509. * routine that would correctly distinguish all the above
  1510. * cases from the values of the function parameters and the
  1511. * inode state but for sanity's sake, I've decided to let the
  1512. * layers above just tell us. It's simpler to correctly figure
  1513. * out in the layer above exactly under what conditions we
  1514. * can run async and I think it's easier for others read and
  1515. * follow the logic in case something has to be changed.
  1516. * cscope is your friend -- rcc.
  1517. *
  1518. * The attribute fork is much simpler.
  1519. *
  1520. * For the attribute fork we allow the caller to tell us whether
  1521. * the unlink of the inode that led to this call is yet permanent
  1522. * in the on disk log. If it is not and we will be freeing extents
  1523. * in this inode then we make the first transaction synchronous
  1524. * to make sure that the unlink is permanent by the time we free
  1525. * the blocks.
  1526. */
  1527. if (fork == XFS_DATA_FORK) {
  1528. if (ip->i_d.di_nextents > 0) {
  1529. /*
  1530. * If we are not changing the file size then do
  1531. * not update the on-disk file size - we may be
  1532. * called from xfs_inactive_free_eofblocks(). If we
  1533. * update the on-disk file size and then the system
  1534. * crashes before the contents of the file are
  1535. * flushed to disk then the files may be full of
  1536. * holes (ie NULL files bug).
  1537. */
  1538. if (ip->i_size != new_size) {
  1539. ip->i_d.di_size = new_size;
  1540. ip->i_size = new_size;
  1541. xfs_trans_log_inode(ntp, ip, XFS_ILOG_CORE);
  1542. }
  1543. }
  1544. } else if (sync) {
  1545. ASSERT(!(mp->m_flags & XFS_MOUNT_WSYNC));
  1546. if (ip->i_d.di_anextents > 0)
  1547. xfs_trans_set_sync(ntp);
  1548. }
  1549. ASSERT(fork == XFS_DATA_FORK ||
  1550. (fork == XFS_ATTR_FORK &&
  1551. ((sync && !(mp->m_flags & XFS_MOUNT_WSYNC)) ||
  1552. (sync == 0 && (mp->m_flags & XFS_MOUNT_WSYNC)))));
  1553. /*
  1554. * Since it is possible for space to become allocated beyond
  1555. * the end of the file (in a crash where the space is allocated
  1556. * but the inode size is not yet updated), simply remove any
  1557. * blocks which show up between the new EOF and the maximum
  1558. * possible file size. If the first block to be removed is
  1559. * beyond the maximum file size (ie it is the same as last_block),
  1560. * then there is nothing to do.
  1561. */
  1562. last_block = XFS_B_TO_FSB(mp, (xfs_ufsize_t)XFS_MAXIOFFSET(mp));
  1563. ASSERT(first_unmap_block <= last_block);
  1564. done = 0;
  1565. if (last_block == first_unmap_block) {
  1566. done = 1;
  1567. } else {
  1568. unmap_len = last_block - first_unmap_block + 1;
  1569. }
  1570. while (!done) {
  1571. /*
  1572. * Free up up to XFS_ITRUNC_MAX_EXTENTS. xfs_bunmapi()
  1573. * will tell us whether it freed the entire range or
  1574. * not. If this is a synchronous mount (wsync),
  1575. * then we can tell bunmapi to keep all the
  1576. * transactions asynchronous since the unlink
  1577. * transaction that made this inode inactive has
  1578. * already hit the disk. There's no danger of
  1579. * the freed blocks being reused, there being a
  1580. * crash, and the reused blocks suddenly reappearing
  1581. * in this file with garbage in them once recovery
  1582. * runs.
  1583. */
  1584. XFS_BMAP_INIT(&free_list, &first_block);
  1585. error = XFS_BUNMAPI(mp, ntp, &ip->i_iocore,
  1586. first_unmap_block, unmap_len,
  1587. XFS_BMAPI_AFLAG(fork) |
  1588. (sync ? 0 : XFS_BMAPI_ASYNC),
  1589. XFS_ITRUNC_MAX_EXTENTS,
  1590. &first_block, &free_list,
  1591. NULL, &done);
  1592. if (error) {
  1593. /*
  1594. * If the bunmapi call encounters an error,
  1595. * return to the caller where the transaction
  1596. * can be properly aborted. We just need to
  1597. * make sure we're not holding any resources
  1598. * that we were not when we came in.
  1599. */
  1600. xfs_bmap_cancel(&free_list);
  1601. return error;
  1602. }
  1603. /*
  1604. * Duplicate the transaction that has the permanent
  1605. * reservation and commit the old transaction.
  1606. */
  1607. error = xfs_bmap_finish(tp, &free_list, &committed);
  1608. ntp = *tp;
  1609. if (error) {
  1610. /*
  1611. * If the bmap finish call encounters an error,
  1612. * return to the caller where the transaction
  1613. * can be properly aborted. We just need to
  1614. * make sure we're not holding any resources
  1615. * that we were not when we came in.
  1616. *
  1617. * Aborting from this point might lose some
  1618. * blocks in the file system, but oh well.
  1619. */
  1620. xfs_bmap_cancel(&free_list);
  1621. if (committed) {
  1622. /*
  1623. * If the passed in transaction committed
  1624. * in xfs_bmap_finish(), then we want to
  1625. * add the inode to this one before returning.
  1626. * This keeps things simple for the higher
  1627. * level code, because it always knows that
  1628. * the inode is locked and held in the
  1629. * transaction that returns to it whether
  1630. * errors occur or not. We don't mark the
  1631. * inode dirty so that this transaction can
  1632. * be easily aborted if possible.
  1633. */
  1634. xfs_trans_ijoin(ntp, ip,
  1635. XFS_ILOCK_EXCL | XFS_IOLOCK_EXCL);
  1636. xfs_trans_ihold(ntp, ip);
  1637. }
  1638. return error;
  1639. }
  1640. if (committed) {
  1641. /*
  1642. * The first xact was committed,
  1643. * so add the inode to the new one.
  1644. * Mark it dirty so it will be logged
  1645. * and moved forward in the log as
  1646. * part of every commit.
  1647. */
  1648. xfs_trans_ijoin(ntp, ip,
  1649. XFS_ILOCK_EXCL | XFS_IOLOCK_EXCL);
  1650. xfs_trans_ihold(ntp, ip);
  1651. xfs_trans_log_inode(ntp, ip, XFS_ILOG_CORE);
  1652. }
  1653. ntp = xfs_trans_dup(ntp);
  1654. (void) xfs_trans_commit(*tp, 0);
  1655. *tp = ntp;
  1656. error = xfs_trans_reserve(ntp, 0, XFS_ITRUNCATE_LOG_RES(mp), 0,
  1657. XFS_TRANS_PERM_LOG_RES,
  1658. XFS_ITRUNCATE_LOG_COUNT);
  1659. /*
  1660. * Add the inode being truncated to the next chained
  1661. * transaction.
  1662. */
  1663. xfs_trans_ijoin(ntp, ip, XFS_ILOCK_EXCL | XFS_IOLOCK_EXCL);
  1664. xfs_trans_ihold(ntp, ip);
  1665. if (error)
  1666. return (error);
  1667. }
  1668. /*
  1669. * Only update the size in the case of the data fork, but
  1670. * always re-log the inode so that our permanent transaction
  1671. * can keep on rolling it forward in the log.
  1672. */
  1673. if (fork == XFS_DATA_FORK) {
  1674. xfs_isize_check(mp, ip, new_size);
  1675. /*
  1676. * If we are not changing the file size then do
  1677. * not update the on-disk file size - we may be
  1678. * called from xfs_inactive_free_eofblocks(). If we
  1679. * update the on-disk file size and then the system
  1680. * crashes before the contents of the file are
  1681. * flushed to disk then the files may be full of
  1682. * holes (ie NULL files bug).
  1683. */
  1684. if (ip->i_size != new_size) {
  1685. ip->i_d.di_size = new_size;
  1686. ip->i_size = new_size;
  1687. }
  1688. }
  1689. xfs_trans_log_inode(ntp, ip, XFS_ILOG_CORE);
  1690. ASSERT((new_size != 0) ||
  1691. (fork == XFS_ATTR_FORK) ||
  1692. (ip->i_delayed_blks == 0));
  1693. ASSERT((new_size != 0) ||
  1694. (fork == XFS_ATTR_FORK) ||
  1695. (ip->i_d.di_nextents == 0));
  1696. xfs_itrunc_trace(XFS_ITRUNC_FINISH2, ip, 0, new_size, 0, 0);
  1697. return 0;
  1698. }
  1699. /*
  1700. * xfs_igrow_start
  1701. *
  1702. * Do the first part of growing a file: zero any data in the last
  1703. * block that is beyond the old EOF. We need to do this before
  1704. * the inode is joined to the transaction to modify the i_size.
  1705. * That way we can drop the inode lock and call into the buffer
  1706. * cache to get the buffer mapping the EOF.
  1707. */
  1708. int
  1709. xfs_igrow_start(
  1710. xfs_inode_t *ip,
  1711. xfs_fsize_t new_size,
  1712. cred_t *credp)
  1713. {
  1714. int error;
  1715. ASSERT(ismrlocked(&(ip->i_lock), MR_UPDATE) != 0);
  1716. ASSERT(ismrlocked(&(ip->i_iolock), MR_UPDATE) != 0);
  1717. ASSERT(new_size > ip->i_size);
  1718. /*
  1719. * Zero any pages that may have been created by
  1720. * xfs_write_file() beyond the end of the file
  1721. * and any blocks between the old and new file sizes.
  1722. */
  1723. error = xfs_zero_eof(XFS_ITOV(ip), &ip->i_iocore, new_size,
  1724. ip->i_size);
  1725. return error;
  1726. }
  1727. /*
  1728. * xfs_igrow_finish
  1729. *
  1730. * This routine is called to extend the size of a file.
  1731. * The inode must have both the iolock and the ilock locked
  1732. * for update and it must be a part of the current transaction.
  1733. * The xfs_igrow_start() function must have been called previously.
  1734. * If the change_flag is not zero, the inode change timestamp will
  1735. * be updated.
  1736. */
  1737. void
  1738. xfs_igrow_finish(
  1739. xfs_trans_t *tp,
  1740. xfs_inode_t *ip,
  1741. xfs_fsize_t new_size,
  1742. int change_flag)
  1743. {
  1744. ASSERT(ismrlocked(&(ip->i_lock), MR_UPDATE) != 0);
  1745. ASSERT(ismrlocked(&(ip->i_iolock), MR_UPDATE) != 0);
  1746. ASSERT(ip->i_transp == tp);
  1747. ASSERT(new_size > ip->i_size);
  1748. /*
  1749. * Update the file size. Update the inode change timestamp
  1750. * if change_flag set.
  1751. */
  1752. ip->i_d.di_size = new_size;
  1753. ip->i_size = new_size;
  1754. if (change_flag)
  1755. xfs_ichgtime(ip, XFS_ICHGTIME_CHG);
  1756. xfs_trans_log_inode(tp, ip, XFS_ILOG_CORE);
  1757. }
  1758. /*
  1759. * This is called when the inode's link count goes to 0.
  1760. * We place the on-disk inode on a list in the AGI. It
  1761. * will be pulled from this list when the inode is freed.
  1762. */
  1763. int
  1764. xfs_iunlink(
  1765. xfs_trans_t *tp,
  1766. xfs_inode_t *ip)
  1767. {
  1768. xfs_mount_t *mp;
  1769. xfs_agi_t *agi;
  1770. xfs_dinode_t *dip;
  1771. xfs_buf_t *agibp;
  1772. xfs_buf_t *ibp;
  1773. xfs_agnumber_t agno;
  1774. xfs_daddr_t agdaddr;
  1775. xfs_agino_t agino;
  1776. short bucket_index;
  1777. int offset;
  1778. int error;
  1779. int agi_ok;
  1780. ASSERT(ip->i_d.di_nlink == 0);
  1781. ASSERT(ip->i_d.di_mode != 0);
  1782. ASSERT(ip->i_transp == tp);
  1783. mp = tp->t_mountp;
  1784. agno = XFS_INO_TO_AGNO(mp, ip->i_ino);
  1785. agdaddr = XFS_AG_DADDR(mp, agno, XFS_AGI_DADDR(mp));
  1786. /*
  1787. * Get the agi buffer first. It ensures lock ordering
  1788. * on the list.
  1789. */
  1790. error = xfs_trans_read_buf(mp, tp, mp->m_ddev_targp, agdaddr,
  1791. XFS_FSS_TO_BB(mp, 1), 0, &agibp);
  1792. if (error) {
  1793. return error;
  1794. }
  1795. /*
  1796. * Validate the magic number of the agi block.
  1797. */
  1798. agi = XFS_BUF_TO_AGI(agibp);
  1799. agi_ok =
  1800. be32_to_cpu(agi->agi_magicnum) == XFS_AGI_MAGIC &&
  1801. XFS_AGI_GOOD_VERSION(be32_to_cpu(agi->agi_versionnum));
  1802. if (unlikely(XFS_TEST_ERROR(!agi_ok, mp, XFS_ERRTAG_IUNLINK,
  1803. XFS_RANDOM_IUNLINK))) {
  1804. XFS_CORRUPTION_ERROR("xfs_iunlink", XFS_ERRLEVEL_LOW, mp, agi);
  1805. xfs_trans_brelse(tp, agibp);
  1806. return XFS_ERROR(EFSCORRUPTED);
  1807. }
  1808. /*
  1809. * Get the index into the agi hash table for the
  1810. * list this inode will go on.
  1811. */
  1812. agino = XFS_INO_TO_AGINO(mp, ip->i_ino);
  1813. ASSERT(agino != 0);
  1814. bucket_index = agino % XFS_AGI_UNLINKED_BUCKETS;
  1815. ASSERT(agi->agi_unlinked[bucket_index]);
  1816. ASSERT(be32_to_cpu(agi->agi_unlinked[bucket_index]) != agino);
  1817. if (be32_to_cpu(agi->agi_unlinked[bucket_index]) != NULLAGINO) {
  1818. /*
  1819. * There is already another inode in the bucket we need
  1820. * to add ourselves to. Add us at the front of the list.
  1821. * Here we put the head pointer into our next pointer,
  1822. * and then we fall through to point the head at us.
  1823. */
  1824. error = xfs_itobp(mp, tp, ip, &dip, &ibp, 0, 0);
  1825. if (error) {
  1826. return error;
  1827. }
  1828. ASSERT(INT_GET(dip->di_next_unlinked, ARCH_CONVERT) == NULLAGINO);
  1829. ASSERT(dip->di_next_unlinked);
  1830. /* both on-disk, don't endian flip twice */
  1831. dip->di_next_unlinked = agi->agi_unlinked[bucket_index];
  1832. offset = ip->i_boffset +
  1833. offsetof(xfs_dinode_t, di_next_unlinked);
  1834. xfs_trans_inode_buf(tp, ibp);
  1835. xfs_trans_log_buf(tp, ibp, offset,
  1836. (offset + sizeof(xfs_agino_t) - 1));
  1837. xfs_inobp_check(mp, ibp);
  1838. }
  1839. /*
  1840. * Point the bucket head pointer at the inode being inserted.
  1841. */
  1842. ASSERT(agino != 0);
  1843. agi->agi_unlinked[bucket_index] = cpu_to_be32(agino);
  1844. offset = offsetof(xfs_agi_t, agi_unlinked) +
  1845. (sizeof(xfs_agino_t) * bucket_index);
  1846. xfs_trans_log_buf(tp, agibp, offset,
  1847. (offset + sizeof(xfs_agino_t) - 1));
  1848. return 0;
  1849. }
  1850. /*
  1851. * Pull the on-disk inode from the AGI unlinked list.
  1852. */
  1853. STATIC int
  1854. xfs_iunlink_remove(
  1855. xfs_trans_t *tp,
  1856. xfs_inode_t *ip)
  1857. {
  1858. xfs_ino_t next_ino;
  1859. xfs_mount_t *mp;
  1860. xfs_agi_t *agi;
  1861. xfs_dinode_t *dip;
  1862. xfs_buf_t *agibp;
  1863. xfs_buf_t *ibp;
  1864. xfs_agnumber_t agno;
  1865. xfs_daddr_t agdaddr;
  1866. xfs_agino_t agino;
  1867. xfs_agino_t next_agino;
  1868. xfs_buf_t *last_ibp;
  1869. xfs_dinode_t *last_dip = NULL;
  1870. short bucket_index;
  1871. int offset, last_offset = 0;
  1872. int error;
  1873. int agi_ok;
  1874. /*
  1875. * First pull the on-disk inode from the AGI unlinked list.
  1876. */
  1877. mp = tp->t_mountp;
  1878. agno = XFS_INO_TO_AGNO(mp, ip->i_ino);
  1879. agdaddr = XFS_AG_DADDR(mp, agno, XFS_AGI_DADDR(mp));
  1880. /*
  1881. * Get the agi buffer first. It ensures lock ordering
  1882. * on the list.
  1883. */
  1884. error = xfs_trans_read_buf(mp, tp, mp->m_ddev_targp, agdaddr,
  1885. XFS_FSS_TO_BB(mp, 1), 0, &agibp);
  1886. if (error) {
  1887. cmn_err(CE_WARN,
  1888. "xfs_iunlink_remove: xfs_trans_read_buf() returned an error %d on %s. Returning error.",
  1889. error, mp->m_fsname);
  1890. return error;
  1891. }
  1892. /*
  1893. * Validate the magic number of the agi block.
  1894. */
  1895. agi = XFS_BUF_TO_AGI(agibp);
  1896. agi_ok =
  1897. be32_to_cpu(agi->agi_magicnum) == XFS_AGI_MAGIC &&
  1898. XFS_AGI_GOOD_VERSION(be32_to_cpu(agi->agi_versionnum));
  1899. if (unlikely(XFS_TEST_ERROR(!agi_ok, mp, XFS_ERRTAG_IUNLINK_REMOVE,
  1900. XFS_RANDOM_IUNLINK_REMOVE))) {
  1901. XFS_CORRUPTION_ERROR("xfs_iunlink_remove", XFS_ERRLEVEL_LOW,
  1902. mp, agi);
  1903. xfs_trans_brelse(tp, agibp);
  1904. cmn_err(CE_WARN,
  1905. "xfs_iunlink_remove: XFS_TEST_ERROR() returned an error on %s. Returning EFSCORRUPTED.",
  1906. mp->m_fsname);
  1907. return XFS_ERROR(EFSCORRUPTED);
  1908. }
  1909. /*
  1910. * Get the index into the agi hash table for the
  1911. * list this inode will go on.
  1912. */
  1913. agino = XFS_INO_TO_AGINO(mp, ip->i_ino);
  1914. ASSERT(agino != 0);
  1915. bucket_index = agino % XFS_AGI_UNLINKED_BUCKETS;
  1916. ASSERT(be32_to_cpu(agi->agi_unlinked[bucket_index]) != NULLAGINO);
  1917. ASSERT(agi->agi_unlinked[bucket_index]);
  1918. if (be32_to_cpu(agi->agi_unlinked[bucket_index]) == agino) {
  1919. /*
  1920. * We're at the head of the list. Get the inode's
  1921. * on-disk buffer to see if there is anyone after us
  1922. * on the list. Only modify our next pointer if it
  1923. * is not already NULLAGINO. This saves us the overhead
  1924. * of dealing with the buffer when there is no need to
  1925. * change it.
  1926. */
  1927. error = xfs_itobp(mp, tp, ip, &dip, &ibp, 0, 0);
  1928. if (error) {
  1929. cmn_err(CE_WARN,
  1930. "xfs_iunlink_remove: xfs_itobp() returned an error %d on %s. Returning error.",
  1931. error, mp->m_fsname);
  1932. return error;
  1933. }
  1934. next_agino = INT_GET(dip->di_next_unlinked, ARCH_CONVERT);
  1935. ASSERT(next_agino != 0);
  1936. if (next_agino != NULLAGINO) {
  1937. INT_SET(dip->di_next_unlinked, ARCH_CONVERT, NULLAGINO);
  1938. offset = ip->i_boffset +
  1939. offsetof(xfs_dinode_t, di_next_unlinked);
  1940. xfs_trans_inode_buf(tp, ibp);
  1941. xfs_trans_log_buf(tp, ibp, offset,
  1942. (offset + sizeof(xfs_agino_t) - 1));
  1943. xfs_inobp_check(mp, ibp);
  1944. } else {
  1945. xfs_trans_brelse(tp, ibp);
  1946. }
  1947. /*
  1948. * Point the bucket head pointer at the next inode.
  1949. */
  1950. ASSERT(next_agino != 0);
  1951. ASSERT(next_agino != agino);
  1952. agi->agi_unlinked[bucket_index] = cpu_to_be32(next_agino);
  1953. offset = offsetof(xfs_agi_t, agi_unlinked) +
  1954. (sizeof(xfs_agino_t) * bucket_index);
  1955. xfs_trans_log_buf(tp, agibp, offset,
  1956. (offset + sizeof(xfs_agino_t) - 1));
  1957. } else {
  1958. /*
  1959. * We need to search the list for the inode being freed.
  1960. */
  1961. next_agino = be32_to_cpu(agi->agi_unlinked[bucket_index]);
  1962. last_ibp = NULL;
  1963. while (next_agino != agino) {
  1964. /*
  1965. * If the last inode wasn't the one pointing to
  1966. * us, then release its buffer since we're not
  1967. * going to do anything with it.
  1968. */
  1969. if (last_ibp != NULL) {
  1970. xfs_trans_brelse(tp, last_ibp);
  1971. }
  1972. next_ino = XFS_AGINO_TO_INO(mp, agno, next_agino);
  1973. error = xfs_inotobp(mp, tp, next_ino, &last_dip,
  1974. &last_ibp, &last_offset);
  1975. if (error) {
  1976. cmn_err(CE_WARN,
  1977. "xfs_iunlink_remove: xfs_inotobp() returned an error %d on %s. Returning error.",
  1978. error, mp->m_fsname);
  1979. return error;
  1980. }
  1981. next_agino = INT_GET(last_dip->di_next_unlinked, ARCH_CONVERT);
  1982. ASSERT(next_agino != NULLAGINO);
  1983. ASSERT(next_agino != 0);
  1984. }
  1985. /*
  1986. * Now last_ibp points to the buffer previous to us on
  1987. * the unlinked list. Pull us from the list.
  1988. */
  1989. error = xfs_itobp(mp, tp, ip, &dip, &ibp, 0, 0);
  1990. if (error) {
  1991. cmn_err(CE_WARN,
  1992. "xfs_iunlink_remove: xfs_itobp() returned an error %d on %s. Returning error.",
  1993. error, mp->m_fsname);
  1994. return error;
  1995. }
  1996. next_agino = INT_GET(dip->di_next_unlinked, ARCH_CONVERT);
  1997. ASSERT(next_agino != 0);
  1998. ASSERT(next_agino != agino);
  1999. if (next_agino != NULLAGINO) {
  2000. INT_SET(dip->di_next_unlinked, ARCH_CONVERT, NULLAGINO);
  2001. offset = ip->i_boffset +
  2002. offsetof(xfs_dinode_t, di_next_unlinked);
  2003. xfs_trans_inode_buf(tp, ibp);
  2004. xfs_trans_log_buf(tp, ibp, offset,
  2005. (offset + sizeof(xfs_agino_t) - 1));
  2006. xfs_inobp_check(mp, ibp);
  2007. } else {
  2008. xfs_trans_brelse(tp, ibp);
  2009. }
  2010. /*
  2011. * Point the previous inode on the list to the next inode.
  2012. */
  2013. INT_SET(last_dip->di_next_unlinked, ARCH_CONVERT, next_agino);
  2014. ASSERT(next_agino != 0);
  2015. offset = last_offset + offsetof(xfs_dinode_t, di_next_unlinked);
  2016. xfs_trans_inode_buf(tp, last_ibp);
  2017. xfs_trans_log_buf(tp, last_ibp, offset,
  2018. (offset + sizeof(xfs_agino_t) - 1));
  2019. xfs_inobp_check(mp, last_ibp);
  2020. }
  2021. return 0;
  2022. }
  2023. STATIC_INLINE int xfs_inode_clean(xfs_inode_t *ip)
  2024. {
  2025. return (((ip->i_itemp == NULL) ||
  2026. !(ip->i_itemp->ili_format.ilf_fields & XFS_ILOG_ALL)) &&
  2027. (ip->i_update_core == 0));
  2028. }
  2029. STATIC void
  2030. xfs_ifree_cluster(
  2031. xfs_inode_t *free_ip,
  2032. xfs_trans_t *tp,
  2033. xfs_ino_t inum)
  2034. {
  2035. xfs_mount_t *mp = free_ip->i_mount;
  2036. int blks_per_cluster;
  2037. int nbufs;
  2038. int ninodes;
  2039. int i, j, found, pre_flushed;
  2040. xfs_daddr_t blkno;
  2041. xfs_buf_t *bp;
  2042. xfs_ihash_t *ih;
  2043. xfs_inode_t *ip, **ip_found;
  2044. xfs_inode_log_item_t *iip;
  2045. xfs_log_item_t *lip;
  2046. SPLDECL(s);
  2047. if (mp->m_sb.sb_blocksize >= XFS_INODE_CLUSTER_SIZE(mp)) {
  2048. blks_per_cluster = 1;
  2049. ninodes = mp->m_sb.sb_inopblock;
  2050. nbufs = XFS_IALLOC_BLOCKS(mp);
  2051. } else {
  2052. blks_per_cluster = XFS_INODE_CLUSTER_SIZE(mp) /
  2053. mp->m_sb.sb_blocksize;
  2054. ninodes = blks_per_cluster * mp->m_sb.sb_inopblock;
  2055. nbufs = XFS_IALLOC_BLOCKS(mp) / blks_per_cluster;
  2056. }
  2057. ip_found = kmem_alloc(ninodes * sizeof(xfs_inode_t *), KM_NOFS);
  2058. for (j = 0; j < nbufs; j++, inum += ninodes) {
  2059. blkno = XFS_AGB_TO_DADDR(mp, XFS_INO_TO_AGNO(mp, inum),
  2060. XFS_INO_TO_AGBNO(mp, inum));
  2061. /*
  2062. * Look for each inode in memory and attempt to lock it,
  2063. * we can be racing with flush and tail pushing here.
  2064. * any inode we get the locks on, add to an array of
  2065. * inode items to process later.
  2066. *
  2067. * The get the buffer lock, we could beat a flush
  2068. * or tail pushing thread to the lock here, in which
  2069. * case they will go looking for the inode buffer
  2070. * and fail, we need some other form of interlock
  2071. * here.
  2072. */
  2073. found = 0;
  2074. for (i = 0; i < ninodes; i++) {
  2075. ih = XFS_IHASH(mp, inum + i);
  2076. read_lock(&ih->ih_lock);
  2077. for (ip = ih->ih_next; ip != NULL; ip = ip->i_next) {
  2078. if (ip->i_ino == inum + i)
  2079. break;
  2080. }
  2081. /* Inode not in memory or we found it already,
  2082. * nothing to do
  2083. */
  2084. if (!ip || xfs_iflags_test(ip, XFS_ISTALE)) {
  2085. read_unlock(&ih->ih_lock);
  2086. continue;
  2087. }
  2088. if (xfs_inode_clean(ip)) {
  2089. read_unlock(&ih->ih_lock);
  2090. continue;
  2091. }
  2092. /* If we can get the locks then add it to the
  2093. * list, otherwise by the time we get the bp lock
  2094. * below it will already be attached to the
  2095. * inode buffer.
  2096. */
  2097. /* This inode will already be locked - by us, lets
  2098. * keep it that way.
  2099. */
  2100. if (ip == free_ip) {
  2101. if (xfs_iflock_nowait(ip)) {
  2102. xfs_iflags_set(ip, XFS_ISTALE);
  2103. if (xfs_inode_clean(ip)) {
  2104. xfs_ifunlock(ip);
  2105. } else {
  2106. ip_found[found++] = ip;
  2107. }
  2108. }
  2109. read_unlock(&ih->ih_lock);
  2110. continue;
  2111. }
  2112. if (xfs_ilock_nowait(ip, XFS_ILOCK_EXCL)) {
  2113. if (xfs_iflock_nowait(ip)) {
  2114. xfs_iflags_set(ip, XFS_ISTALE);
  2115. if (xfs_inode_clean(ip)) {
  2116. xfs_ifunlock(ip);
  2117. xfs_iunlock(ip, XFS_ILOCK_EXCL);
  2118. } else {
  2119. ip_found[found++] = ip;
  2120. }
  2121. } else {
  2122. xfs_iunlock(ip, XFS_ILOCK_EXCL);
  2123. }
  2124. }
  2125. read_unlock(&ih->ih_lock);
  2126. }
  2127. bp = xfs_trans_get_buf(tp, mp->m_ddev_targp, blkno,
  2128. mp->m_bsize * blks_per_cluster,
  2129. XFS_BUF_LOCK);
  2130. pre_flushed = 0;
  2131. lip = XFS_BUF_FSPRIVATE(bp, xfs_log_item_t *);
  2132. while (lip) {
  2133. if (lip->li_type == XFS_LI_INODE) {
  2134. iip = (xfs_inode_log_item_t *)lip;
  2135. ASSERT(iip->ili_logged == 1);
  2136. lip->li_cb = (void(*)(xfs_buf_t*,xfs_log_item_t*)) xfs_istale_done;
  2137. AIL_LOCK(mp,s);
  2138. iip->ili_flush_lsn = iip->ili_item.li_lsn;
  2139. AIL_UNLOCK(mp, s);
  2140. xfs_iflags_set(iip->ili_inode, XFS_ISTALE);
  2141. pre_flushed++;
  2142. }
  2143. lip = lip->li_bio_list;
  2144. }
  2145. for (i = 0; i < found; i++) {
  2146. ip = ip_found[i];
  2147. iip = ip->i_itemp;
  2148. if (!iip) {
  2149. ip->i_update_core = 0;
  2150. xfs_ifunlock(ip);
  2151. xfs_iunlock(ip, XFS_ILOCK_EXCL);
  2152. continue;
  2153. }
  2154. iip->ili_last_fields = iip->ili_format.ilf_fields;
  2155. iip->ili_format.ilf_fields = 0;
  2156. iip->ili_logged = 1;
  2157. AIL_LOCK(mp,s);
  2158. iip->ili_flush_lsn = iip->ili_item.li_lsn;
  2159. AIL_UNLOCK(mp, s);
  2160. xfs_buf_attach_iodone(bp,
  2161. (void(*)(xfs_buf_t*,xfs_log_item_t*))
  2162. xfs_istale_done, (xfs_log_item_t *)iip);
  2163. if (ip != free_ip) {
  2164. xfs_iunlock(ip, XFS_ILOCK_EXCL);
  2165. }
  2166. }
  2167. if (found || pre_flushed)
  2168. xfs_trans_stale_inode_buf(tp, bp);
  2169. xfs_trans_binval(tp, bp);
  2170. }
  2171. kmem_free(ip_found, ninodes * sizeof(xfs_inode_t *));
  2172. }
  2173. /*
  2174. * This is called to return an inode to the inode free list.
  2175. * The inode should already be truncated to 0 length and have
  2176. * no pages associated with it. This routine also assumes that
  2177. * the inode is already a part of the transaction.
  2178. *
  2179. * The on-disk copy of the inode will have been added to the list
  2180. * of unlinked inodes in the AGI. We need to remove the inode from
  2181. * that list atomically with respect to freeing it here.
  2182. */
  2183. int
  2184. xfs_ifree(
  2185. xfs_trans_t *tp,
  2186. xfs_inode_t *ip,
  2187. xfs_bmap_free_t *flist)
  2188. {
  2189. int error;
  2190. int delete;
  2191. xfs_ino_t first_ino;
  2192. ASSERT(ismrlocked(&ip->i_lock, MR_UPDATE));
  2193. ASSERT(ip->i_transp == tp);
  2194. ASSERT(ip->i_d.di_nlink == 0);
  2195. ASSERT(ip->i_d.di_nextents == 0);
  2196. ASSERT(ip->i_d.di_anextents == 0);
  2197. ASSERT((ip->i_d.di_size == 0 && ip->i_size == 0) ||
  2198. ((ip->i_d.di_mode & S_IFMT) != S_IFREG));
  2199. ASSERT(ip->i_d.di_nblocks == 0);
  2200. /*
  2201. * Pull the on-disk inode from the AGI unlinked list.
  2202. */
  2203. error = xfs_iunlink_remove(tp, ip);
  2204. if (error != 0) {
  2205. return error;
  2206. }
  2207. error = xfs_difree(tp, ip->i_ino, flist, &delete, &first_ino);
  2208. if (error != 0) {
  2209. return error;
  2210. }
  2211. ip->i_d.di_mode = 0; /* mark incore inode as free */
  2212. ip->i_d.di_flags = 0;
  2213. ip->i_d.di_dmevmask = 0;
  2214. ip->i_d.di_forkoff = 0; /* mark the attr fork not in use */
  2215. ip->i_df.if_ext_max =
  2216. XFS_IFORK_DSIZE(ip) / (uint)sizeof(xfs_bmbt_rec_t);
  2217. ip->i_d.di_format = XFS_DINODE_FMT_EXTENTS;
  2218. ip->i_d.di_aformat = XFS_DINODE_FMT_EXTENTS;
  2219. /*
  2220. * Bump the generation count so no one will be confused
  2221. * by reincarnations of this inode.
  2222. */
  2223. ip->i_d.di_gen++;
  2224. xfs_trans_log_inode(tp, ip, XFS_ILOG_CORE);
  2225. if (delete) {
  2226. xfs_ifree_cluster(ip, tp, first_ino);
  2227. }
  2228. return 0;
  2229. }
  2230. /*
  2231. * Reallocate the space for if_broot based on the number of records
  2232. * being added or deleted as indicated in rec_diff. Move the records
  2233. * and pointers in if_broot to fit the new size. When shrinking this
  2234. * will eliminate holes between the records and pointers created by
  2235. * the caller. When growing this will create holes to be filled in
  2236. * by the caller.
  2237. *
  2238. * The caller must not request to add more records than would fit in
  2239. * the on-disk inode root. If the if_broot is currently NULL, then
  2240. * if we adding records one will be allocated. The caller must also
  2241. * not request that the number of records go below zero, although
  2242. * it can go to zero.
  2243. *
  2244. * ip -- the inode whose if_broot area is changing
  2245. * ext_diff -- the change in the number of records, positive or negative,
  2246. * requested for the if_broot array.
  2247. */
  2248. void
  2249. xfs_iroot_realloc(
  2250. xfs_inode_t *ip,
  2251. int rec_diff,
  2252. int whichfork)
  2253. {
  2254. int cur_max;
  2255. xfs_ifork_t *ifp;
  2256. xfs_bmbt_block_t *new_broot;
  2257. int new_max;
  2258. size_t new_size;
  2259. char *np;
  2260. char *op;
  2261. /*
  2262. * Handle the degenerate case quietly.
  2263. */
  2264. if (rec_diff == 0) {
  2265. return;
  2266. }
  2267. ifp = XFS_IFORK_PTR(ip, whichfork);
  2268. if (rec_diff > 0) {
  2269. /*
  2270. * If there wasn't any memory allocated before, just
  2271. * allocate it now and get out.
  2272. */
  2273. if (ifp->if_broot_bytes == 0) {
  2274. new_size = (size_t)XFS_BMAP_BROOT_SPACE_CALC(rec_diff);
  2275. ifp->if_broot = (xfs_bmbt_block_t*)kmem_alloc(new_size,
  2276. KM_SLEEP);
  2277. ifp->if_broot_bytes = (int)new_size;
  2278. return;
  2279. }
  2280. /*
  2281. * If there is already an existing if_broot, then we need
  2282. * to realloc() it and shift the pointers to their new
  2283. * location. The records don't change location because
  2284. * they are kept butted up against the btree block header.
  2285. */
  2286. cur_max = XFS_BMAP_BROOT_MAXRECS(ifp->if_broot_bytes);
  2287. new_max = cur_max + rec_diff;
  2288. new_size = (size_t)XFS_BMAP_BROOT_SPACE_CALC(new_max);
  2289. ifp->if_broot = (xfs_bmbt_block_t *)
  2290. kmem_realloc(ifp->if_broot,
  2291. new_size,
  2292. (size_t)XFS_BMAP_BROOT_SPACE_CALC(cur_max), /* old size */
  2293. KM_SLEEP);
  2294. op = (char *)XFS_BMAP_BROOT_PTR_ADDR(ifp->if_broot, 1,
  2295. ifp->if_broot_bytes);
  2296. np = (char *)XFS_BMAP_BROOT_PTR_ADDR(ifp->if_broot, 1,
  2297. (int)new_size);
  2298. ifp->if_broot_bytes = (int)new_size;
  2299. ASSERT(ifp->if_broot_bytes <=
  2300. XFS_IFORK_SIZE(ip, whichfork) + XFS_BROOT_SIZE_ADJ);
  2301. memmove(np, op, cur_max * (uint)sizeof(xfs_dfsbno_t));
  2302. return;
  2303. }
  2304. /*
  2305. * rec_diff is less than 0. In this case, we are shrinking the
  2306. * if_broot buffer. It must already exist. If we go to zero
  2307. * records, just get rid of the root and clear the status bit.
  2308. */
  2309. ASSERT((ifp->if_broot != NULL) && (ifp->if_broot_bytes > 0));
  2310. cur_max = XFS_BMAP_BROOT_MAXRECS(ifp->if_broot_bytes);
  2311. new_max = cur_max + rec_diff;
  2312. ASSERT(new_max >= 0);
  2313. if (new_max > 0)
  2314. new_size = (size_t)XFS_BMAP_BROOT_SPACE_CALC(new_max);
  2315. else
  2316. new_size = 0;
  2317. if (new_size > 0) {
  2318. new_broot = (xfs_bmbt_block_t *)kmem_alloc(new_size, KM_SLEEP);
  2319. /*
  2320. * First copy over the btree block header.
  2321. */
  2322. memcpy(new_broot, ifp->if_broot, sizeof(xfs_bmbt_block_t));
  2323. } else {
  2324. new_broot = NULL;
  2325. ifp->if_flags &= ~XFS_IFBROOT;
  2326. }
  2327. /*
  2328. * Only copy the records and pointers if there are any.
  2329. */
  2330. if (new_max > 0) {
  2331. /*
  2332. * First copy the records.
  2333. */
  2334. op = (char *)XFS_BMAP_BROOT_REC_ADDR(ifp->if_broot, 1,
  2335. ifp->if_broot_bytes);
  2336. np = (char *)XFS_BMAP_BROOT_REC_ADDR(new_broot, 1,
  2337. (int)new_size);
  2338. memcpy(np, op, new_max * (uint)sizeof(xfs_bmbt_rec_t));
  2339. /*
  2340. * Then copy the pointers.
  2341. */
  2342. op = (char *)XFS_BMAP_BROOT_PTR_ADDR(ifp->if_broot, 1,
  2343. ifp->if_broot_bytes);
  2344. np = (char *)XFS_BMAP_BROOT_PTR_ADDR(new_broot, 1,
  2345. (int)new_size);
  2346. memcpy(np, op, new_max * (uint)sizeof(xfs_dfsbno_t));
  2347. }
  2348. kmem_free(ifp->if_broot, ifp->if_broot_bytes);
  2349. ifp->if_broot = new_broot;
  2350. ifp->if_broot_bytes = (int)new_size;
  2351. ASSERT(ifp->if_broot_bytes <=
  2352. XFS_IFORK_SIZE(ip, whichfork) + XFS_BROOT_SIZE_ADJ);
  2353. return;
  2354. }
  2355. /*
  2356. * This is called when the amount of space needed for if_data
  2357. * is increased or decreased. The change in size is indicated by
  2358. * the number of bytes that need to be added or deleted in the
  2359. * byte_diff parameter.
  2360. *
  2361. * If the amount of space needed has decreased below the size of the
  2362. * inline buffer, then switch to using the inline buffer. Otherwise,
  2363. * use kmem_realloc() or kmem_alloc() to adjust the size of the buffer
  2364. * to what is needed.
  2365. *
  2366. * ip -- the inode whose if_data area is changing
  2367. * byte_diff -- the change in the number of bytes, positive or negative,
  2368. * requested for the if_data array.
  2369. */
  2370. void
  2371. xfs_idata_realloc(
  2372. xfs_inode_t *ip,
  2373. int byte_diff,
  2374. int whichfork)
  2375. {
  2376. xfs_ifork_t *ifp;
  2377. int new_size;
  2378. int real_size;
  2379. if (byte_diff == 0) {
  2380. return;
  2381. }
  2382. ifp = XFS_IFORK_PTR(ip, whichfork);
  2383. new_size = (int)ifp->if_bytes + byte_diff;
  2384. ASSERT(new_size >= 0);
  2385. if (new_size == 0) {
  2386. if (ifp->if_u1.if_data != ifp->if_u2.if_inline_data) {
  2387. kmem_free(ifp->if_u1.if_data, ifp->if_real_bytes);
  2388. }
  2389. ifp->if_u1.if_data = NULL;
  2390. real_size = 0;
  2391. } else if (new_size <= sizeof(ifp->if_u2.if_inline_data)) {
  2392. /*
  2393. * If the valid extents/data can fit in if_inline_ext/data,
  2394. * copy them from the malloc'd vector and free it.
  2395. */
  2396. if (ifp->if_u1.if_data == NULL) {
  2397. ifp->if_u1.if_data = ifp->if_u2.if_inline_data;
  2398. } else if (ifp->if_u1.if_data != ifp->if_u2.if_inline_data) {
  2399. ASSERT(ifp->if_real_bytes != 0);
  2400. memcpy(ifp->if_u2.if_inline_data, ifp->if_u1.if_data,
  2401. new_size);
  2402. kmem_free(ifp->if_u1.if_data, ifp->if_real_bytes);
  2403. ifp->if_u1.if_data = ifp->if_u2.if_inline_data;
  2404. }
  2405. real_size = 0;
  2406. } else {
  2407. /*
  2408. * Stuck with malloc/realloc.
  2409. * For inline data, the underlying buffer must be
  2410. * a multiple of 4 bytes in size so that it can be
  2411. * logged and stay on word boundaries. We enforce
  2412. * that here.
  2413. */
  2414. real_size = roundup(new_size, 4);
  2415. if (ifp->if_u1.if_data == NULL) {
  2416. ASSERT(ifp->if_real_bytes == 0);
  2417. ifp->if_u1.if_data = kmem_alloc(real_size, KM_SLEEP);
  2418. } else if (ifp->if_u1.if_data != ifp->if_u2.if_inline_data) {
  2419. /*
  2420. * Only do the realloc if the underlying size
  2421. * is really changing.
  2422. */
  2423. if (ifp->if_real_bytes != real_size) {
  2424. ifp->if_u1.if_data =
  2425. kmem_realloc(ifp->if_u1.if_data,
  2426. real_size,
  2427. ifp->if_real_bytes,
  2428. KM_SLEEP);
  2429. }
  2430. } else {
  2431. ASSERT(ifp->if_real_bytes == 0);
  2432. ifp->if_u1.if_data = kmem_alloc(real_size, KM_SLEEP);
  2433. memcpy(ifp->if_u1.if_data, ifp->if_u2.if_inline_data,
  2434. ifp->if_bytes);
  2435. }
  2436. }
  2437. ifp->if_real_bytes = real_size;
  2438. ifp->if_bytes = new_size;
  2439. ASSERT(ifp->if_bytes <= XFS_IFORK_SIZE(ip, whichfork));
  2440. }
  2441. /*
  2442. * Map inode to disk block and offset.
  2443. *
  2444. * mp -- the mount point structure for the current file system
  2445. * tp -- the current transaction
  2446. * ino -- the inode number of the inode to be located
  2447. * imap -- this structure is filled in with the information necessary
  2448. * to retrieve the given inode from disk
  2449. * flags -- flags to pass to xfs_dilocate indicating whether or not
  2450. * lookups in the inode btree were OK or not
  2451. */
  2452. int
  2453. xfs_imap(
  2454. xfs_mount_t *mp,
  2455. xfs_trans_t *tp,
  2456. xfs_ino_t ino,
  2457. xfs_imap_t *imap,
  2458. uint flags)
  2459. {
  2460. xfs_fsblock_t fsbno;
  2461. int len;
  2462. int off;
  2463. int error;
  2464. fsbno = imap->im_blkno ?
  2465. XFS_DADDR_TO_FSB(mp, imap->im_blkno) : NULLFSBLOCK;
  2466. error = xfs_dilocate(mp, tp, ino, &fsbno, &len, &off, flags);
  2467. if (error != 0) {
  2468. return error;
  2469. }
  2470. imap->im_blkno = XFS_FSB_TO_DADDR(mp, fsbno);
  2471. imap->im_len = XFS_FSB_TO_BB(mp, len);
  2472. imap->im_agblkno = XFS_FSB_TO_AGBNO(mp, fsbno);
  2473. imap->im_ioffset = (ushort)off;
  2474. imap->im_boffset = (ushort)(off << mp->m_sb.sb_inodelog);
  2475. return 0;
  2476. }
  2477. void
  2478. xfs_idestroy_fork(
  2479. xfs_inode_t *ip,
  2480. int whichfork)
  2481. {
  2482. xfs_ifork_t *ifp;
  2483. ifp = XFS_IFORK_PTR(ip, whichfork);
  2484. if (ifp->if_broot != NULL) {
  2485. kmem_free(ifp->if_broot, ifp->if_broot_bytes);
  2486. ifp->if_broot = NULL;
  2487. }
  2488. /*
  2489. * If the format is local, then we can't have an extents
  2490. * array so just look for an inline data array. If we're
  2491. * not local then we may or may not have an extents list,
  2492. * so check and free it up if we do.
  2493. */
  2494. if (XFS_IFORK_FORMAT(ip, whichfork) == XFS_DINODE_FMT_LOCAL) {
  2495. if ((ifp->if_u1.if_data != ifp->if_u2.if_inline_data) &&
  2496. (ifp->if_u1.if_data != NULL)) {
  2497. ASSERT(ifp->if_real_bytes != 0);
  2498. kmem_free(ifp->if_u1.if_data, ifp->if_real_bytes);
  2499. ifp->if_u1.if_data = NULL;
  2500. ifp->if_real_bytes = 0;
  2501. }
  2502. } else if ((ifp->if_flags & XFS_IFEXTENTS) &&
  2503. ((ifp->if_flags & XFS_IFEXTIREC) ||
  2504. ((ifp->if_u1.if_extents != NULL) &&
  2505. (ifp->if_u1.if_extents != ifp->if_u2.if_inline_ext)))) {
  2506. ASSERT(ifp->if_real_bytes != 0);
  2507. xfs_iext_destroy(ifp);
  2508. }
  2509. ASSERT(ifp->if_u1.if_extents == NULL ||
  2510. ifp->if_u1.if_extents == ifp->if_u2.if_inline_ext);
  2511. ASSERT(ifp->if_real_bytes == 0);
  2512. if (whichfork == XFS_ATTR_FORK) {
  2513. kmem_zone_free(xfs_ifork_zone, ip->i_afp);
  2514. ip->i_afp = NULL;
  2515. }
  2516. }
  2517. /*
  2518. * This is called free all the memory associated with an inode.
  2519. * It must free the inode itself and any buffers allocated for
  2520. * if_extents/if_data and if_broot. It must also free the lock
  2521. * associated with the inode.
  2522. */
  2523. void
  2524. xfs_idestroy(
  2525. xfs_inode_t *ip)
  2526. {
  2527. switch (ip->i_d.di_mode & S_IFMT) {
  2528. case S_IFREG:
  2529. case S_IFDIR:
  2530. case S_IFLNK:
  2531. xfs_idestroy_fork(ip, XFS_DATA_FORK);
  2532. break;
  2533. }
  2534. if (ip->i_afp)
  2535. xfs_idestroy_fork(ip, XFS_ATTR_FORK);
  2536. mrfree(&ip->i_lock);
  2537. mrfree(&ip->i_iolock);
  2538. freesema(&ip->i_flock);
  2539. #ifdef XFS_BMAP_TRACE
  2540. ktrace_free(ip->i_xtrace);
  2541. #endif
  2542. #ifdef XFS_BMBT_TRACE
  2543. ktrace_free(ip->i_btrace);
  2544. #endif
  2545. #ifdef XFS_RW_TRACE
  2546. ktrace_free(ip->i_rwtrace);
  2547. #endif
  2548. #ifdef XFS_ILOCK_TRACE
  2549. ktrace_free(ip->i_lock_trace);
  2550. #endif
  2551. #ifdef XFS_DIR2_TRACE
  2552. ktrace_free(ip->i_dir_trace);
  2553. #endif
  2554. if (ip->i_itemp) {
  2555. /*
  2556. * Only if we are shutting down the fs will we see an
  2557. * inode still in the AIL. If it is there, we should remove
  2558. * it to prevent a use-after-free from occurring.
  2559. */
  2560. xfs_mount_t *mp = ip->i_mount;
  2561. xfs_log_item_t *lip = &ip->i_itemp->ili_item;
  2562. int s;
  2563. ASSERT(((lip->li_flags & XFS_LI_IN_AIL) == 0) ||
  2564. XFS_FORCED_SHUTDOWN(ip->i_mount));
  2565. if (lip->li_flags & XFS_LI_IN_AIL) {
  2566. AIL_LOCK(mp, s);
  2567. if (lip->li_flags & XFS_LI_IN_AIL)
  2568. xfs_trans_delete_ail(mp, lip, s);
  2569. else
  2570. AIL_UNLOCK(mp, s);
  2571. }
  2572. xfs_inode_item_destroy(ip);
  2573. }
  2574. kmem_zone_free(xfs_inode_zone, ip);
  2575. }
  2576. /*
  2577. * Increment the pin count of the given buffer.
  2578. * This value is protected by ipinlock spinlock in the mount structure.
  2579. */
  2580. void
  2581. xfs_ipin(
  2582. xfs_inode_t *ip)
  2583. {
  2584. ASSERT(ismrlocked(&ip->i_lock, MR_UPDATE));
  2585. atomic_inc(&ip->i_pincount);
  2586. }
  2587. /*
  2588. * Decrement the pin count of the given inode, and wake up
  2589. * anyone in xfs_iwait_unpin() if the count goes to 0. The
  2590. * inode must have been previously pinned with a call to xfs_ipin().
  2591. */
  2592. void
  2593. xfs_iunpin(
  2594. xfs_inode_t *ip)
  2595. {
  2596. ASSERT(atomic_read(&ip->i_pincount) > 0);
  2597. if (atomic_dec_and_lock(&ip->i_pincount, &ip->i_flags_lock)) {
  2598. /*
  2599. * If the inode is currently being reclaimed, the link between
  2600. * the bhv_vnode and the xfs_inode will be broken after the
  2601. * XFS_IRECLAIM* flag is set. Hence, if these flags are not
  2602. * set, then we can move forward and mark the linux inode dirty
  2603. * knowing that it is still valid as it won't freed until after
  2604. * the bhv_vnode<->xfs_inode link is broken in xfs_reclaim. The
  2605. * i_flags_lock is used to synchronise the setting of the
  2606. * XFS_IRECLAIM* flags and the breaking of the link, and so we
  2607. * can execute atomically w.r.t to reclaim by holding this lock
  2608. * here.
  2609. *
  2610. * However, we still need to issue the unpin wakeup call as the
  2611. * inode reclaim may be blocked waiting for the inode to become
  2612. * unpinned.
  2613. */
  2614. if (!__xfs_iflags_test(ip, XFS_IRECLAIM|XFS_IRECLAIMABLE)) {
  2615. bhv_vnode_t *vp = XFS_ITOV_NULL(ip);
  2616. struct inode *inode = NULL;
  2617. BUG_ON(vp == NULL);
  2618. inode = vn_to_inode(vp);
  2619. BUG_ON(inode->i_state & I_CLEAR);
  2620. /* make sync come back and flush this inode */
  2621. if (!(inode->i_state & (I_NEW|I_FREEING)))
  2622. mark_inode_dirty_sync(inode);
  2623. }
  2624. spin_unlock(&ip->i_flags_lock);
  2625. wake_up(&ip->i_ipin_wait);
  2626. }
  2627. }
  2628. /*
  2629. * This is called to wait for the given inode to be unpinned.
  2630. * It will sleep until this happens. The caller must have the
  2631. * inode locked in at least shared mode so that the buffer cannot
  2632. * be subsequently pinned once someone is waiting for it to be
  2633. * unpinned.
  2634. */
  2635. STATIC void
  2636. xfs_iunpin_wait(
  2637. xfs_inode_t *ip)
  2638. {
  2639. xfs_inode_log_item_t *iip;
  2640. xfs_lsn_t lsn;
  2641. ASSERT(ismrlocked(&ip->i_lock, MR_UPDATE | MR_ACCESS));
  2642. if (atomic_read(&ip->i_pincount) == 0) {
  2643. return;
  2644. }
  2645. iip = ip->i_itemp;
  2646. if (iip && iip->ili_last_lsn) {
  2647. lsn = iip->ili_last_lsn;
  2648. } else {
  2649. lsn = (xfs_lsn_t)0;
  2650. }
  2651. /*
  2652. * Give the log a push so we don't wait here too long.
  2653. */
  2654. xfs_log_force(ip->i_mount, lsn, XFS_LOG_FORCE);
  2655. wait_event(ip->i_ipin_wait, (atomic_read(&ip->i_pincount) == 0));
  2656. }
  2657. /*
  2658. * xfs_iextents_copy()
  2659. *
  2660. * This is called to copy the REAL extents (as opposed to the delayed
  2661. * allocation extents) from the inode into the given buffer. It
  2662. * returns the number of bytes copied into the buffer.
  2663. *
  2664. * If there are no delayed allocation extents, then we can just
  2665. * memcpy() the extents into the buffer. Otherwise, we need to
  2666. * examine each extent in turn and skip those which are delayed.
  2667. */
  2668. int
  2669. xfs_iextents_copy(
  2670. xfs_inode_t *ip,
  2671. xfs_bmbt_rec_t *buffer,
  2672. int whichfork)
  2673. {
  2674. int copied;
  2675. xfs_bmbt_rec_t *dest_ep;
  2676. xfs_bmbt_rec_t *ep;
  2677. #ifdef XFS_BMAP_TRACE
  2678. static char fname[] = "xfs_iextents_copy";
  2679. #endif
  2680. int i;
  2681. xfs_ifork_t *ifp;
  2682. int nrecs;
  2683. xfs_fsblock_t start_block;
  2684. ifp = XFS_IFORK_PTR(ip, whichfork);
  2685. ASSERT(ismrlocked(&ip->i_lock, MR_UPDATE|MR_ACCESS));
  2686. ASSERT(ifp->if_bytes > 0);
  2687. nrecs = ifp->if_bytes / (uint)sizeof(xfs_bmbt_rec_t);
  2688. xfs_bmap_trace_exlist(fname, ip, nrecs, whichfork);
  2689. ASSERT(nrecs > 0);
  2690. /*
  2691. * There are some delayed allocation extents in the
  2692. * inode, so copy the extents one at a time and skip
  2693. * the delayed ones. There must be at least one
  2694. * non-delayed extent.
  2695. */
  2696. dest_ep = buffer;
  2697. copied = 0;
  2698. for (i = 0; i < nrecs; i++) {
  2699. ep = xfs_iext_get_ext(ifp, i);
  2700. start_block = xfs_bmbt_get_startblock(ep);
  2701. if (ISNULLSTARTBLOCK(start_block)) {
  2702. /*
  2703. * It's a delayed allocation extent, so skip it.
  2704. */
  2705. continue;
  2706. }
  2707. /* Translate to on disk format */
  2708. put_unaligned(INT_GET(ep->l0, ARCH_CONVERT),
  2709. (__uint64_t*)&dest_ep->l0);
  2710. put_unaligned(INT_GET(ep->l1, ARCH_CONVERT),
  2711. (__uint64_t*)&dest_ep->l1);
  2712. dest_ep++;
  2713. copied++;
  2714. }
  2715. ASSERT(copied != 0);
  2716. xfs_validate_extents(ifp, copied, 1, XFS_EXTFMT_INODE(ip));
  2717. return (copied * (uint)sizeof(xfs_bmbt_rec_t));
  2718. }
  2719. /*
  2720. * Each of the following cases stores data into the same region
  2721. * of the on-disk inode, so only one of them can be valid at
  2722. * any given time. While it is possible to have conflicting formats
  2723. * and log flags, e.g. having XFS_ILOG_?DATA set when the fork is
  2724. * in EXTENTS format, this can only happen when the fork has
  2725. * changed formats after being modified but before being flushed.
  2726. * In these cases, the format always takes precedence, because the
  2727. * format indicates the current state of the fork.
  2728. */
  2729. /*ARGSUSED*/
  2730. STATIC int
  2731. xfs_iflush_fork(
  2732. xfs_inode_t *ip,
  2733. xfs_dinode_t *dip,
  2734. xfs_inode_log_item_t *iip,
  2735. int whichfork,
  2736. xfs_buf_t *bp)
  2737. {
  2738. char *cp;
  2739. xfs_ifork_t *ifp;
  2740. xfs_mount_t *mp;
  2741. #ifdef XFS_TRANS_DEBUG
  2742. int first;
  2743. #endif
  2744. static const short brootflag[2] =
  2745. { XFS_ILOG_DBROOT, XFS_ILOG_ABROOT };
  2746. static const short dataflag[2] =
  2747. { XFS_ILOG_DDATA, XFS_ILOG_ADATA };
  2748. static const short extflag[2] =
  2749. { XFS_ILOG_DEXT, XFS_ILOG_AEXT };
  2750. if (iip == NULL)
  2751. return 0;
  2752. ifp = XFS_IFORK_PTR(ip, whichfork);
  2753. /*
  2754. * This can happen if we gave up in iformat in an error path,
  2755. * for the attribute fork.
  2756. */
  2757. if (ifp == NULL) {
  2758. ASSERT(whichfork == XFS_ATTR_FORK);
  2759. return 0;
  2760. }
  2761. cp = XFS_DFORK_PTR(dip, whichfork);
  2762. mp = ip->i_mount;
  2763. switch (XFS_IFORK_FORMAT(ip, whichfork)) {
  2764. case XFS_DINODE_FMT_LOCAL:
  2765. if ((iip->ili_format.ilf_fields & dataflag[whichfork]) &&
  2766. (ifp->if_bytes > 0)) {
  2767. ASSERT(ifp->if_u1.if_data != NULL);
  2768. ASSERT(ifp->if_bytes <= XFS_IFORK_SIZE(ip, whichfork));
  2769. memcpy(cp, ifp->if_u1.if_data, ifp->if_bytes);
  2770. }
  2771. break;
  2772. case XFS_DINODE_FMT_EXTENTS:
  2773. ASSERT((ifp->if_flags & XFS_IFEXTENTS) ||
  2774. !(iip->ili_format.ilf_fields & extflag[whichfork]));
  2775. ASSERT((xfs_iext_get_ext(ifp, 0) != NULL) ||
  2776. (ifp->if_bytes == 0));
  2777. ASSERT((xfs_iext_get_ext(ifp, 0) == NULL) ||
  2778. (ifp->if_bytes > 0));
  2779. if ((iip->ili_format.ilf_fields & extflag[whichfork]) &&
  2780. (ifp->if_bytes > 0)) {
  2781. ASSERT(XFS_IFORK_NEXTENTS(ip, whichfork) > 0);
  2782. (void)xfs_iextents_copy(ip, (xfs_bmbt_rec_t *)cp,
  2783. whichfork);
  2784. }
  2785. break;
  2786. case XFS_DINODE_FMT_BTREE:
  2787. if ((iip->ili_format.ilf_fields & brootflag[whichfork]) &&
  2788. (ifp->if_broot_bytes > 0)) {
  2789. ASSERT(ifp->if_broot != NULL);
  2790. ASSERT(ifp->if_broot_bytes <=
  2791. (XFS_IFORK_SIZE(ip, whichfork) +
  2792. XFS_BROOT_SIZE_ADJ));
  2793. xfs_bmbt_to_bmdr(ifp->if_broot, ifp->if_broot_bytes,
  2794. (xfs_bmdr_block_t *)cp,
  2795. XFS_DFORK_SIZE(dip, mp, whichfork));
  2796. }
  2797. break;
  2798. case XFS_DINODE_FMT_DEV:
  2799. if (iip->ili_format.ilf_fields & XFS_ILOG_DEV) {
  2800. ASSERT(whichfork == XFS_DATA_FORK);
  2801. INT_SET(dip->di_u.di_dev, ARCH_CONVERT, ip->i_df.if_u2.if_rdev);
  2802. }
  2803. break;
  2804. case XFS_DINODE_FMT_UUID:
  2805. if (iip->ili_format.ilf_fields & XFS_ILOG_UUID) {
  2806. ASSERT(whichfork == XFS_DATA_FORK);
  2807. memcpy(&dip->di_u.di_muuid, &ip->i_df.if_u2.if_uuid,
  2808. sizeof(uuid_t));
  2809. }
  2810. break;
  2811. default:
  2812. ASSERT(0);
  2813. break;
  2814. }
  2815. return 0;
  2816. }
  2817. /*
  2818. * xfs_iflush() will write a modified inode's changes out to the
  2819. * inode's on disk home. The caller must have the inode lock held
  2820. * in at least shared mode and the inode flush semaphore must be
  2821. * held as well. The inode lock will still be held upon return from
  2822. * the call and the caller is free to unlock it.
  2823. * The inode flush lock will be unlocked when the inode reaches the disk.
  2824. * The flags indicate how the inode's buffer should be written out.
  2825. */
  2826. int
  2827. xfs_iflush(
  2828. xfs_inode_t *ip,
  2829. uint flags)
  2830. {
  2831. xfs_inode_log_item_t *iip;
  2832. xfs_buf_t *bp;
  2833. xfs_dinode_t *dip;
  2834. xfs_mount_t *mp;
  2835. int error;
  2836. /* REFERENCED */
  2837. xfs_chash_t *ch;
  2838. xfs_inode_t *iq;
  2839. int clcount; /* count of inodes clustered */
  2840. int bufwasdelwri;
  2841. enum { INT_DELWRI = (1 << 0), INT_ASYNC = (1 << 1) };
  2842. SPLDECL(s);
  2843. XFS_STATS_INC(xs_iflush_count);
  2844. ASSERT(ismrlocked(&ip->i_lock, MR_UPDATE|MR_ACCESS));
  2845. ASSERT(issemalocked(&(ip->i_flock)));
  2846. ASSERT(ip->i_d.di_format != XFS_DINODE_FMT_BTREE ||
  2847. ip->i_d.di_nextents > ip->i_df.if_ext_max);
  2848. iip = ip->i_itemp;
  2849. mp = ip->i_mount;
  2850. /*
  2851. * If the inode isn't dirty, then just release the inode
  2852. * flush lock and do nothing.
  2853. */
  2854. if ((ip->i_update_core == 0) &&
  2855. ((iip == NULL) || !(iip->ili_format.ilf_fields & XFS_ILOG_ALL))) {
  2856. ASSERT((iip != NULL) ?
  2857. !(iip->ili_item.li_flags & XFS_LI_IN_AIL) : 1);
  2858. xfs_ifunlock(ip);
  2859. return 0;
  2860. }
  2861. /*
  2862. * We can't flush the inode until it is unpinned, so
  2863. * wait for it. We know noone new can pin it, because
  2864. * we are holding the inode lock shared and you need
  2865. * to hold it exclusively to pin the inode.
  2866. */
  2867. xfs_iunpin_wait(ip);
  2868. /*
  2869. * This may have been unpinned because the filesystem is shutting
  2870. * down forcibly. If that's the case we must not write this inode
  2871. * to disk, because the log record didn't make it to disk!
  2872. */
  2873. if (XFS_FORCED_SHUTDOWN(mp)) {
  2874. ip->i_update_core = 0;
  2875. if (iip)
  2876. iip->ili_format.ilf_fields = 0;
  2877. xfs_ifunlock(ip);
  2878. return XFS_ERROR(EIO);
  2879. }
  2880. /*
  2881. * Get the buffer containing the on-disk inode.
  2882. */
  2883. error = xfs_itobp(mp, NULL, ip, &dip, &bp, 0, 0);
  2884. if (error) {
  2885. xfs_ifunlock(ip);
  2886. return error;
  2887. }
  2888. /*
  2889. * Decide how buffer will be flushed out. This is done before
  2890. * the call to xfs_iflush_int because this field is zeroed by it.
  2891. */
  2892. if (iip != NULL && iip->ili_format.ilf_fields != 0) {
  2893. /*
  2894. * Flush out the inode buffer according to the directions
  2895. * of the caller. In the cases where the caller has given
  2896. * us a choice choose the non-delwri case. This is because
  2897. * the inode is in the AIL and we need to get it out soon.
  2898. */
  2899. switch (flags) {
  2900. case XFS_IFLUSH_SYNC:
  2901. case XFS_IFLUSH_DELWRI_ELSE_SYNC:
  2902. flags = 0;
  2903. break;
  2904. case XFS_IFLUSH_ASYNC:
  2905. case XFS_IFLUSH_DELWRI_ELSE_ASYNC:
  2906. flags = INT_ASYNC;
  2907. break;
  2908. case XFS_IFLUSH_DELWRI:
  2909. flags = INT_DELWRI;
  2910. break;
  2911. default:
  2912. ASSERT(0);
  2913. flags = 0;
  2914. break;
  2915. }
  2916. } else {
  2917. switch (flags) {
  2918. case XFS_IFLUSH_DELWRI_ELSE_SYNC:
  2919. case XFS_IFLUSH_DELWRI_ELSE_ASYNC:
  2920. case XFS_IFLUSH_DELWRI:
  2921. flags = INT_DELWRI;
  2922. break;
  2923. case XFS_IFLUSH_ASYNC:
  2924. flags = INT_ASYNC;
  2925. break;
  2926. case XFS_IFLUSH_SYNC:
  2927. flags = 0;
  2928. break;
  2929. default:
  2930. ASSERT(0);
  2931. flags = 0;
  2932. break;
  2933. }
  2934. }
  2935. /*
  2936. * First flush out the inode that xfs_iflush was called with.
  2937. */
  2938. error = xfs_iflush_int(ip, bp);
  2939. if (error) {
  2940. goto corrupt_out;
  2941. }
  2942. /*
  2943. * inode clustering:
  2944. * see if other inodes can be gathered into this write
  2945. */
  2946. ip->i_chash->chl_buf = bp;
  2947. ch = XFS_CHASH(mp, ip->i_blkno);
  2948. s = mutex_spinlock(&ch->ch_lock);
  2949. clcount = 0;
  2950. for (iq = ip->i_cnext; iq != ip; iq = iq->i_cnext) {
  2951. /*
  2952. * Do an un-protected check to see if the inode is dirty and
  2953. * is a candidate for flushing. These checks will be repeated
  2954. * later after the appropriate locks are acquired.
  2955. */
  2956. iip = iq->i_itemp;
  2957. if ((iq->i_update_core == 0) &&
  2958. ((iip == NULL) ||
  2959. !(iip->ili_format.ilf_fields & XFS_ILOG_ALL)) &&
  2960. xfs_ipincount(iq) == 0) {
  2961. continue;
  2962. }
  2963. /*
  2964. * Try to get locks. If any are unavailable,
  2965. * then this inode cannot be flushed and is skipped.
  2966. */
  2967. /* get inode locks (just i_lock) */
  2968. if (xfs_ilock_nowait(iq, XFS_ILOCK_SHARED)) {
  2969. /* get inode flush lock */
  2970. if (xfs_iflock_nowait(iq)) {
  2971. /* check if pinned */
  2972. if (xfs_ipincount(iq) == 0) {
  2973. /* arriving here means that
  2974. * this inode can be flushed.
  2975. * first re-check that it's
  2976. * dirty
  2977. */
  2978. iip = iq->i_itemp;
  2979. if ((iq->i_update_core != 0)||
  2980. ((iip != NULL) &&
  2981. (iip->ili_format.ilf_fields & XFS_ILOG_ALL))) {
  2982. clcount++;
  2983. error = xfs_iflush_int(iq, bp);
  2984. if (error) {
  2985. xfs_iunlock(iq,
  2986. XFS_ILOCK_SHARED);
  2987. goto cluster_corrupt_out;
  2988. }
  2989. } else {
  2990. xfs_ifunlock(iq);
  2991. }
  2992. } else {
  2993. xfs_ifunlock(iq);
  2994. }
  2995. }
  2996. xfs_iunlock(iq, XFS_ILOCK_SHARED);
  2997. }
  2998. }
  2999. mutex_spinunlock(&ch->ch_lock, s);
  3000. if (clcount) {
  3001. XFS_STATS_INC(xs_icluster_flushcnt);
  3002. XFS_STATS_ADD(xs_icluster_flushinode, clcount);
  3003. }
  3004. /*
  3005. * If the buffer is pinned then push on the log so we won't
  3006. * get stuck waiting in the write for too long.
  3007. */
  3008. if (XFS_BUF_ISPINNED(bp)){
  3009. xfs_log_force(mp, (xfs_lsn_t)0, XFS_LOG_FORCE);
  3010. }
  3011. if (flags & INT_DELWRI) {
  3012. xfs_bdwrite(mp, bp);
  3013. } else if (flags & INT_ASYNC) {
  3014. xfs_bawrite(mp, bp);
  3015. } else {
  3016. error = xfs_bwrite(mp, bp);
  3017. }
  3018. return error;
  3019. corrupt_out:
  3020. xfs_buf_relse(bp);
  3021. xfs_force_shutdown(mp, SHUTDOWN_CORRUPT_INCORE);
  3022. xfs_iflush_abort(ip);
  3023. /*
  3024. * Unlocks the flush lock
  3025. */
  3026. return XFS_ERROR(EFSCORRUPTED);
  3027. cluster_corrupt_out:
  3028. /* Corruption detected in the clustering loop. Invalidate the
  3029. * inode buffer and shut down the filesystem.
  3030. */
  3031. mutex_spinunlock(&ch->ch_lock, s);
  3032. /*
  3033. * Clean up the buffer. If it was B_DELWRI, just release it --
  3034. * brelse can handle it with no problems. If not, shut down the
  3035. * filesystem before releasing the buffer.
  3036. */
  3037. if ((bufwasdelwri= XFS_BUF_ISDELAYWRITE(bp))) {
  3038. xfs_buf_relse(bp);
  3039. }
  3040. xfs_force_shutdown(mp, SHUTDOWN_CORRUPT_INCORE);
  3041. if(!bufwasdelwri) {
  3042. /*
  3043. * Just like incore_relse: if we have b_iodone functions,
  3044. * mark the buffer as an error and call them. Otherwise
  3045. * mark it as stale and brelse.
  3046. */
  3047. if (XFS_BUF_IODONE_FUNC(bp)) {
  3048. XFS_BUF_CLR_BDSTRAT_FUNC(bp);
  3049. XFS_BUF_UNDONE(bp);
  3050. XFS_BUF_STALE(bp);
  3051. XFS_BUF_SHUT(bp);
  3052. XFS_BUF_ERROR(bp,EIO);
  3053. xfs_biodone(bp);
  3054. } else {
  3055. XFS_BUF_STALE(bp);
  3056. xfs_buf_relse(bp);
  3057. }
  3058. }
  3059. xfs_iflush_abort(iq);
  3060. /*
  3061. * Unlocks the flush lock
  3062. */
  3063. return XFS_ERROR(EFSCORRUPTED);
  3064. }
  3065. STATIC int
  3066. xfs_iflush_int(
  3067. xfs_inode_t *ip,
  3068. xfs_buf_t *bp)
  3069. {
  3070. xfs_inode_log_item_t *iip;
  3071. xfs_dinode_t *dip;
  3072. xfs_mount_t *mp;
  3073. #ifdef XFS_TRANS_DEBUG
  3074. int first;
  3075. #endif
  3076. SPLDECL(s);
  3077. ASSERT(ismrlocked(&ip->i_lock, MR_UPDATE|MR_ACCESS));
  3078. ASSERT(issemalocked(&(ip->i_flock)));
  3079. ASSERT(ip->i_d.di_format != XFS_DINODE_FMT_BTREE ||
  3080. ip->i_d.di_nextents > ip->i_df.if_ext_max);
  3081. iip = ip->i_itemp;
  3082. mp = ip->i_mount;
  3083. /*
  3084. * If the inode isn't dirty, then just release the inode
  3085. * flush lock and do nothing.
  3086. */
  3087. if ((ip->i_update_core == 0) &&
  3088. ((iip == NULL) || !(iip->ili_format.ilf_fields & XFS_ILOG_ALL))) {
  3089. xfs_ifunlock(ip);
  3090. return 0;
  3091. }
  3092. /* set *dip = inode's place in the buffer */
  3093. dip = (xfs_dinode_t *)xfs_buf_offset(bp, ip->i_boffset);
  3094. /*
  3095. * Clear i_update_core before copying out the data.
  3096. * This is for coordination with our timestamp updates
  3097. * that don't hold the inode lock. They will always
  3098. * update the timestamps BEFORE setting i_update_core,
  3099. * so if we clear i_update_core after they set it we
  3100. * are guaranteed to see their updates to the timestamps.
  3101. * I believe that this depends on strongly ordered memory
  3102. * semantics, but we have that. We use the SYNCHRONIZE
  3103. * macro to make sure that the compiler does not reorder
  3104. * the i_update_core access below the data copy below.
  3105. */
  3106. ip->i_update_core = 0;
  3107. SYNCHRONIZE();
  3108. /*
  3109. * Make sure to get the latest atime from the Linux inode.
  3110. */
  3111. xfs_synchronize_atime(ip);
  3112. if (XFS_TEST_ERROR(INT_GET(dip->di_core.di_magic,ARCH_CONVERT) != XFS_DINODE_MAGIC,
  3113. mp, XFS_ERRTAG_IFLUSH_1, XFS_RANDOM_IFLUSH_1)) {
  3114. xfs_cmn_err(XFS_PTAG_IFLUSH, CE_ALERT, mp,
  3115. "xfs_iflush: Bad inode %Lu magic number 0x%x, ptr 0x%p",
  3116. ip->i_ino, (int) INT_GET(dip->di_core.di_magic, ARCH_CONVERT), dip);
  3117. goto corrupt_out;
  3118. }
  3119. if (XFS_TEST_ERROR(ip->i_d.di_magic != XFS_DINODE_MAGIC,
  3120. mp, XFS_ERRTAG_IFLUSH_2, XFS_RANDOM_IFLUSH_2)) {
  3121. xfs_cmn_err(XFS_PTAG_IFLUSH, CE_ALERT, mp,
  3122. "xfs_iflush: Bad inode %Lu, ptr 0x%p, magic number 0x%x",
  3123. ip->i_ino, ip, ip->i_d.di_magic);
  3124. goto corrupt_out;
  3125. }
  3126. if ((ip->i_d.di_mode & S_IFMT) == S_IFREG) {
  3127. if (XFS_TEST_ERROR(
  3128. (ip->i_d.di_format != XFS_DINODE_FMT_EXTENTS) &&
  3129. (ip->i_d.di_format != XFS_DINODE_FMT_BTREE),
  3130. mp, XFS_ERRTAG_IFLUSH_3, XFS_RANDOM_IFLUSH_3)) {
  3131. xfs_cmn_err(XFS_PTAG_IFLUSH, CE_ALERT, mp,
  3132. "xfs_iflush: Bad regular inode %Lu, ptr 0x%p",
  3133. ip->i_ino, ip);
  3134. goto corrupt_out;
  3135. }
  3136. } else if ((ip->i_d.di_mode & S_IFMT) == S_IFDIR) {
  3137. if (XFS_TEST_ERROR(
  3138. (ip->i_d.di_format != XFS_DINODE_FMT_EXTENTS) &&
  3139. (ip->i_d.di_format != XFS_DINODE_FMT_BTREE) &&
  3140. (ip->i_d.di_format != XFS_DINODE_FMT_LOCAL),
  3141. mp, XFS_ERRTAG_IFLUSH_4, XFS_RANDOM_IFLUSH_4)) {
  3142. xfs_cmn_err(XFS_PTAG_IFLUSH, CE_ALERT, mp,
  3143. "xfs_iflush: Bad directory inode %Lu, ptr 0x%p",
  3144. ip->i_ino, ip);
  3145. goto corrupt_out;
  3146. }
  3147. }
  3148. if (XFS_TEST_ERROR(ip->i_d.di_nextents + ip->i_d.di_anextents >
  3149. ip->i_d.di_nblocks, mp, XFS_ERRTAG_IFLUSH_5,
  3150. XFS_RANDOM_IFLUSH_5)) {
  3151. xfs_cmn_err(XFS_PTAG_IFLUSH, CE_ALERT, mp,
  3152. "xfs_iflush: detected corrupt incore inode %Lu, total extents = %d, nblocks = %Ld, ptr 0x%p",
  3153. ip->i_ino,
  3154. ip->i_d.di_nextents + ip->i_d.di_anextents,
  3155. ip->i_d.di_nblocks,
  3156. ip);
  3157. goto corrupt_out;
  3158. }
  3159. if (XFS_TEST_ERROR(ip->i_d.di_forkoff > mp->m_sb.sb_inodesize,
  3160. mp, XFS_ERRTAG_IFLUSH_6, XFS_RANDOM_IFLUSH_6)) {
  3161. xfs_cmn_err(XFS_PTAG_IFLUSH, CE_ALERT, mp,
  3162. "xfs_iflush: bad inode %Lu, forkoff 0x%x, ptr 0x%p",
  3163. ip->i_ino, ip->i_d.di_forkoff, ip);
  3164. goto corrupt_out;
  3165. }
  3166. /*
  3167. * bump the flush iteration count, used to detect flushes which
  3168. * postdate a log record during recovery.
  3169. */
  3170. ip->i_d.di_flushiter++;
  3171. /*
  3172. * Copy the dirty parts of the inode into the on-disk
  3173. * inode. We always copy out the core of the inode,
  3174. * because if the inode is dirty at all the core must
  3175. * be.
  3176. */
  3177. xfs_xlate_dinode_core((xfs_caddr_t)&(dip->di_core), &(ip->i_d), -1);
  3178. /* Wrap, we never let the log put out DI_MAX_FLUSH */
  3179. if (ip->i_d.di_flushiter == DI_MAX_FLUSH)
  3180. ip->i_d.di_flushiter = 0;
  3181. /*
  3182. * If this is really an old format inode and the superblock version
  3183. * has not been updated to support only new format inodes, then
  3184. * convert back to the old inode format. If the superblock version
  3185. * has been updated, then make the conversion permanent.
  3186. */
  3187. ASSERT(ip->i_d.di_version == XFS_DINODE_VERSION_1 ||
  3188. XFS_SB_VERSION_HASNLINK(&mp->m_sb));
  3189. if (ip->i_d.di_version == XFS_DINODE_VERSION_1) {
  3190. if (!XFS_SB_VERSION_HASNLINK(&mp->m_sb)) {
  3191. /*
  3192. * Convert it back.
  3193. */
  3194. ASSERT(ip->i_d.di_nlink <= XFS_MAXLINK_1);
  3195. INT_SET(dip->di_core.di_onlink, ARCH_CONVERT, ip->i_d.di_nlink);
  3196. } else {
  3197. /*
  3198. * The superblock version has already been bumped,
  3199. * so just make the conversion to the new inode
  3200. * format permanent.
  3201. */
  3202. ip->i_d.di_version = XFS_DINODE_VERSION_2;
  3203. INT_SET(dip->di_core.di_version, ARCH_CONVERT, XFS_DINODE_VERSION_2);
  3204. ip->i_d.di_onlink = 0;
  3205. dip->di_core.di_onlink = 0;
  3206. memset(&(ip->i_d.di_pad[0]), 0, sizeof(ip->i_d.di_pad));
  3207. memset(&(dip->di_core.di_pad[0]), 0,
  3208. sizeof(dip->di_core.di_pad));
  3209. ASSERT(ip->i_d.di_projid == 0);
  3210. }
  3211. }
  3212. if (xfs_iflush_fork(ip, dip, iip, XFS_DATA_FORK, bp) == EFSCORRUPTED) {
  3213. goto corrupt_out;
  3214. }
  3215. if (XFS_IFORK_Q(ip)) {
  3216. /*
  3217. * The only error from xfs_iflush_fork is on the data fork.
  3218. */
  3219. (void) xfs_iflush_fork(ip, dip, iip, XFS_ATTR_FORK, bp);
  3220. }
  3221. xfs_inobp_check(mp, bp);
  3222. /*
  3223. * We've recorded everything logged in the inode, so we'd
  3224. * like to clear the ilf_fields bits so we don't log and
  3225. * flush things unnecessarily. However, we can't stop
  3226. * logging all this information until the data we've copied
  3227. * into the disk buffer is written to disk. If we did we might
  3228. * overwrite the copy of the inode in the log with all the
  3229. * data after re-logging only part of it, and in the face of
  3230. * a crash we wouldn't have all the data we need to recover.
  3231. *
  3232. * What we do is move the bits to the ili_last_fields field.
  3233. * When logging the inode, these bits are moved back to the
  3234. * ilf_fields field. In the xfs_iflush_done() routine we
  3235. * clear ili_last_fields, since we know that the information
  3236. * those bits represent is permanently on disk. As long as
  3237. * the flush completes before the inode is logged again, then
  3238. * both ilf_fields and ili_last_fields will be cleared.
  3239. *
  3240. * We can play with the ilf_fields bits here, because the inode
  3241. * lock must be held exclusively in order to set bits there
  3242. * and the flush lock protects the ili_last_fields bits.
  3243. * Set ili_logged so the flush done
  3244. * routine can tell whether or not to look in the AIL.
  3245. * Also, store the current LSN of the inode so that we can tell
  3246. * whether the item has moved in the AIL from xfs_iflush_done().
  3247. * In order to read the lsn we need the AIL lock, because
  3248. * it is a 64 bit value that cannot be read atomically.
  3249. */
  3250. if (iip != NULL && iip->ili_format.ilf_fields != 0) {
  3251. iip->ili_last_fields = iip->ili_format.ilf_fields;
  3252. iip->ili_format.ilf_fields = 0;
  3253. iip->ili_logged = 1;
  3254. ASSERT(sizeof(xfs_lsn_t) == 8); /* don't lock if it shrinks */
  3255. AIL_LOCK(mp,s);
  3256. iip->ili_flush_lsn = iip->ili_item.li_lsn;
  3257. AIL_UNLOCK(mp, s);
  3258. /*
  3259. * Attach the function xfs_iflush_done to the inode's
  3260. * buffer. This will remove the inode from the AIL
  3261. * and unlock the inode's flush lock when the inode is
  3262. * completely written to disk.
  3263. */
  3264. xfs_buf_attach_iodone(bp, (void(*)(xfs_buf_t*,xfs_log_item_t*))
  3265. xfs_iflush_done, (xfs_log_item_t *)iip);
  3266. ASSERT(XFS_BUF_FSPRIVATE(bp, void *) != NULL);
  3267. ASSERT(XFS_BUF_IODONE_FUNC(bp) != NULL);
  3268. } else {
  3269. /*
  3270. * We're flushing an inode which is not in the AIL and has
  3271. * not been logged but has i_update_core set. For this
  3272. * case we can use a B_DELWRI flush and immediately drop
  3273. * the inode flush lock because we can avoid the whole
  3274. * AIL state thing. It's OK to drop the flush lock now,
  3275. * because we've already locked the buffer and to do anything
  3276. * you really need both.
  3277. */
  3278. if (iip != NULL) {
  3279. ASSERT(iip->ili_logged == 0);
  3280. ASSERT(iip->ili_last_fields == 0);
  3281. ASSERT((iip->ili_item.li_flags & XFS_LI_IN_AIL) == 0);
  3282. }
  3283. xfs_ifunlock(ip);
  3284. }
  3285. return 0;
  3286. corrupt_out:
  3287. return XFS_ERROR(EFSCORRUPTED);
  3288. }
  3289. /*
  3290. * Flush all inactive inodes in mp.
  3291. */
  3292. void
  3293. xfs_iflush_all(
  3294. xfs_mount_t *mp)
  3295. {
  3296. xfs_inode_t *ip;
  3297. bhv_vnode_t *vp;
  3298. again:
  3299. XFS_MOUNT_ILOCK(mp);
  3300. ip = mp->m_inodes;
  3301. if (ip == NULL)
  3302. goto out;
  3303. do {
  3304. /* Make sure we skip markers inserted by sync */
  3305. if (ip->i_mount == NULL) {
  3306. ip = ip->i_mnext;
  3307. continue;
  3308. }
  3309. vp = XFS_ITOV_NULL(ip);
  3310. if (!vp) {
  3311. XFS_MOUNT_IUNLOCK(mp);
  3312. xfs_finish_reclaim(ip, 0, XFS_IFLUSH_ASYNC);
  3313. goto again;
  3314. }
  3315. ASSERT(vn_count(vp) == 0);
  3316. ip = ip->i_mnext;
  3317. } while (ip != mp->m_inodes);
  3318. out:
  3319. XFS_MOUNT_IUNLOCK(mp);
  3320. }
  3321. /*
  3322. * xfs_iaccess: check accessibility of inode for mode.
  3323. */
  3324. int
  3325. xfs_iaccess(
  3326. xfs_inode_t *ip,
  3327. mode_t mode,
  3328. cred_t *cr)
  3329. {
  3330. int error;
  3331. mode_t orgmode = mode;
  3332. struct inode *inode = vn_to_inode(XFS_ITOV(ip));
  3333. if (mode & S_IWUSR) {
  3334. umode_t imode = inode->i_mode;
  3335. if (IS_RDONLY(inode) &&
  3336. (S_ISREG(imode) || S_ISDIR(imode) || S_ISLNK(imode)))
  3337. return XFS_ERROR(EROFS);
  3338. if (IS_IMMUTABLE(inode))
  3339. return XFS_ERROR(EACCES);
  3340. }
  3341. /*
  3342. * If there's an Access Control List it's used instead of
  3343. * the mode bits.
  3344. */
  3345. if ((error = _ACL_XFS_IACCESS(ip, mode, cr)) != -1)
  3346. return error ? XFS_ERROR(error) : 0;
  3347. if (current_fsuid(cr) != ip->i_d.di_uid) {
  3348. mode >>= 3;
  3349. if (!in_group_p((gid_t)ip->i_d.di_gid))
  3350. mode >>= 3;
  3351. }
  3352. /*
  3353. * If the DACs are ok we don't need any capability check.
  3354. */
  3355. if ((ip->i_d.di_mode & mode) == mode)
  3356. return 0;
  3357. /*
  3358. * Read/write DACs are always overridable.
  3359. * Executable DACs are overridable if at least one exec bit is set.
  3360. */
  3361. if (!(orgmode & S_IXUSR) ||
  3362. (inode->i_mode & S_IXUGO) || S_ISDIR(inode->i_mode))
  3363. if (capable_cred(cr, CAP_DAC_OVERRIDE))
  3364. return 0;
  3365. if ((orgmode == S_IRUSR) ||
  3366. (S_ISDIR(inode->i_mode) && (!(orgmode & S_IWUSR)))) {
  3367. if (capable_cred(cr, CAP_DAC_READ_SEARCH))
  3368. return 0;
  3369. #ifdef NOISE
  3370. cmn_err(CE_NOTE, "Ick: mode=%o, orgmode=%o", mode, orgmode);
  3371. #endif /* NOISE */
  3372. return XFS_ERROR(EACCES);
  3373. }
  3374. return XFS_ERROR(EACCES);
  3375. }
  3376. /*
  3377. * xfs_iroundup: round up argument to next power of two
  3378. */
  3379. uint
  3380. xfs_iroundup(
  3381. uint v)
  3382. {
  3383. int i;
  3384. uint m;
  3385. if ((v & (v - 1)) == 0)
  3386. return v;
  3387. ASSERT((v & 0x80000000) == 0);
  3388. if ((v & (v + 1)) == 0)
  3389. return v + 1;
  3390. for (i = 0, m = 1; i < 31; i++, m <<= 1) {
  3391. if (v & m)
  3392. continue;
  3393. v |= m;
  3394. if ((v & (v + 1)) == 0)
  3395. return v + 1;
  3396. }
  3397. ASSERT(0);
  3398. return( 0 );
  3399. }
  3400. #ifdef XFS_ILOCK_TRACE
  3401. ktrace_t *xfs_ilock_trace_buf;
  3402. void
  3403. xfs_ilock_trace(xfs_inode_t *ip, int lock, unsigned int lockflags, inst_t *ra)
  3404. {
  3405. ktrace_enter(ip->i_lock_trace,
  3406. (void *)ip,
  3407. (void *)(unsigned long)lock, /* 1 = LOCK, 3=UNLOCK, etc */
  3408. (void *)(unsigned long)lockflags, /* XFS_ILOCK_EXCL etc */
  3409. (void *)ra, /* caller of ilock */
  3410. (void *)(unsigned long)current_cpu(),
  3411. (void *)(unsigned long)current_pid(),
  3412. NULL,NULL,NULL,NULL,NULL,NULL,NULL,NULL,NULL,NULL);
  3413. }
  3414. #endif
  3415. /*
  3416. * Return a pointer to the extent record at file index idx.
  3417. */
  3418. xfs_bmbt_rec_t *
  3419. xfs_iext_get_ext(
  3420. xfs_ifork_t *ifp, /* inode fork pointer */
  3421. xfs_extnum_t idx) /* index of target extent */
  3422. {
  3423. ASSERT(idx >= 0);
  3424. if ((ifp->if_flags & XFS_IFEXTIREC) && (idx == 0)) {
  3425. return ifp->if_u1.if_ext_irec->er_extbuf;
  3426. } else if (ifp->if_flags & XFS_IFEXTIREC) {
  3427. xfs_ext_irec_t *erp; /* irec pointer */
  3428. int erp_idx = 0; /* irec index */
  3429. xfs_extnum_t page_idx = idx; /* ext index in target list */
  3430. erp = xfs_iext_idx_to_irec(ifp, &page_idx, &erp_idx, 0);
  3431. return &erp->er_extbuf[page_idx];
  3432. } else if (ifp->if_bytes) {
  3433. return &ifp->if_u1.if_extents[idx];
  3434. } else {
  3435. return NULL;
  3436. }
  3437. }
  3438. /*
  3439. * Insert new item(s) into the extent records for incore inode
  3440. * fork 'ifp'. 'count' new items are inserted at index 'idx'.
  3441. */
  3442. void
  3443. xfs_iext_insert(
  3444. xfs_ifork_t *ifp, /* inode fork pointer */
  3445. xfs_extnum_t idx, /* starting index of new items */
  3446. xfs_extnum_t count, /* number of inserted items */
  3447. xfs_bmbt_irec_t *new) /* items to insert */
  3448. {
  3449. xfs_bmbt_rec_t *ep; /* extent record pointer */
  3450. xfs_extnum_t i; /* extent record index */
  3451. ASSERT(ifp->if_flags & XFS_IFEXTENTS);
  3452. xfs_iext_add(ifp, idx, count);
  3453. for (i = idx; i < idx + count; i++, new++) {
  3454. ep = xfs_iext_get_ext(ifp, i);
  3455. xfs_bmbt_set_all(ep, new);
  3456. }
  3457. }
  3458. /*
  3459. * This is called when the amount of space required for incore file
  3460. * extents needs to be increased. The ext_diff parameter stores the
  3461. * number of new extents being added and the idx parameter contains
  3462. * the extent index where the new extents will be added. If the new
  3463. * extents are being appended, then we just need to (re)allocate and
  3464. * initialize the space. Otherwise, if the new extents are being
  3465. * inserted into the middle of the existing entries, a bit more work
  3466. * is required to make room for the new extents to be inserted. The
  3467. * caller is responsible for filling in the new extent entries upon
  3468. * return.
  3469. */
  3470. void
  3471. xfs_iext_add(
  3472. xfs_ifork_t *ifp, /* inode fork pointer */
  3473. xfs_extnum_t idx, /* index to begin adding exts */
  3474. int ext_diff) /* number of extents to add */
  3475. {
  3476. int byte_diff; /* new bytes being added */
  3477. int new_size; /* size of extents after adding */
  3478. xfs_extnum_t nextents; /* number of extents in file */
  3479. nextents = ifp->if_bytes / (uint)sizeof(xfs_bmbt_rec_t);
  3480. ASSERT((idx >= 0) && (idx <= nextents));
  3481. byte_diff = ext_diff * sizeof(xfs_bmbt_rec_t);
  3482. new_size = ifp->if_bytes + byte_diff;
  3483. /*
  3484. * If the new number of extents (nextents + ext_diff)
  3485. * fits inside the inode, then continue to use the inline
  3486. * extent buffer.
  3487. */
  3488. if (nextents + ext_diff <= XFS_INLINE_EXTS) {
  3489. if (idx < nextents) {
  3490. memmove(&ifp->if_u2.if_inline_ext[idx + ext_diff],
  3491. &ifp->if_u2.if_inline_ext[idx],
  3492. (nextents - idx) * sizeof(xfs_bmbt_rec_t));
  3493. memset(&ifp->if_u2.if_inline_ext[idx], 0, byte_diff);
  3494. }
  3495. ifp->if_u1.if_extents = ifp->if_u2.if_inline_ext;
  3496. ifp->if_real_bytes = 0;
  3497. ifp->if_lastex = nextents + ext_diff;
  3498. }
  3499. /*
  3500. * Otherwise use a linear (direct) extent list.
  3501. * If the extents are currently inside the inode,
  3502. * xfs_iext_realloc_direct will switch us from
  3503. * inline to direct extent allocation mode.
  3504. */
  3505. else if (nextents + ext_diff <= XFS_LINEAR_EXTS) {
  3506. xfs_iext_realloc_direct(ifp, new_size);
  3507. if (idx < nextents) {
  3508. memmove(&ifp->if_u1.if_extents[idx + ext_diff],
  3509. &ifp->if_u1.if_extents[idx],
  3510. (nextents - idx) * sizeof(xfs_bmbt_rec_t));
  3511. memset(&ifp->if_u1.if_extents[idx], 0, byte_diff);
  3512. }
  3513. }
  3514. /* Indirection array */
  3515. else {
  3516. xfs_ext_irec_t *erp;
  3517. int erp_idx = 0;
  3518. int page_idx = idx;
  3519. ASSERT(nextents + ext_diff > XFS_LINEAR_EXTS);
  3520. if (ifp->if_flags & XFS_IFEXTIREC) {
  3521. erp = xfs_iext_idx_to_irec(ifp, &page_idx, &erp_idx, 1);
  3522. } else {
  3523. xfs_iext_irec_init(ifp);
  3524. ASSERT(ifp->if_flags & XFS_IFEXTIREC);
  3525. erp = ifp->if_u1.if_ext_irec;
  3526. }
  3527. /* Extents fit in target extent page */
  3528. if (erp && erp->er_extcount + ext_diff <= XFS_LINEAR_EXTS) {
  3529. if (page_idx < erp->er_extcount) {
  3530. memmove(&erp->er_extbuf[page_idx + ext_diff],
  3531. &erp->er_extbuf[page_idx],
  3532. (erp->er_extcount - page_idx) *
  3533. sizeof(xfs_bmbt_rec_t));
  3534. memset(&erp->er_extbuf[page_idx], 0, byte_diff);
  3535. }
  3536. erp->er_extcount += ext_diff;
  3537. xfs_iext_irec_update_extoffs(ifp, erp_idx + 1, ext_diff);
  3538. }
  3539. /* Insert a new extent page */
  3540. else if (erp) {
  3541. xfs_iext_add_indirect_multi(ifp,
  3542. erp_idx, page_idx, ext_diff);
  3543. }
  3544. /*
  3545. * If extent(s) are being appended to the last page in
  3546. * the indirection array and the new extent(s) don't fit
  3547. * in the page, then erp is NULL and erp_idx is set to
  3548. * the next index needed in the indirection array.
  3549. */
  3550. else {
  3551. int count = ext_diff;
  3552. while (count) {
  3553. erp = xfs_iext_irec_new(ifp, erp_idx);
  3554. erp->er_extcount = count;
  3555. count -= MIN(count, (int)XFS_LINEAR_EXTS);
  3556. if (count) {
  3557. erp_idx++;
  3558. }
  3559. }
  3560. }
  3561. }
  3562. ifp->if_bytes = new_size;
  3563. }
  3564. /*
  3565. * This is called when incore extents are being added to the indirection
  3566. * array and the new extents do not fit in the target extent list. The
  3567. * erp_idx parameter contains the irec index for the target extent list
  3568. * in the indirection array, and the idx parameter contains the extent
  3569. * index within the list. The number of extents being added is stored
  3570. * in the count parameter.
  3571. *
  3572. * |-------| |-------|
  3573. * | | | | idx - number of extents before idx
  3574. * | idx | | count |
  3575. * | | | | count - number of extents being inserted at idx
  3576. * |-------| |-------|
  3577. * | count | | nex2 | nex2 - number of extents after idx + count
  3578. * |-------| |-------|
  3579. */
  3580. void
  3581. xfs_iext_add_indirect_multi(
  3582. xfs_ifork_t *ifp, /* inode fork pointer */
  3583. int erp_idx, /* target extent irec index */
  3584. xfs_extnum_t idx, /* index within target list */
  3585. int count) /* new extents being added */
  3586. {
  3587. int byte_diff; /* new bytes being added */
  3588. xfs_ext_irec_t *erp; /* pointer to irec entry */
  3589. xfs_extnum_t ext_diff; /* number of extents to add */
  3590. xfs_extnum_t ext_cnt; /* new extents still needed */
  3591. xfs_extnum_t nex2; /* extents after idx + count */
  3592. xfs_bmbt_rec_t *nex2_ep = NULL; /* temp list for nex2 extents */
  3593. int nlists; /* number of irec's (lists) */
  3594. ASSERT(ifp->if_flags & XFS_IFEXTIREC);
  3595. erp = &ifp->if_u1.if_ext_irec[erp_idx];
  3596. nex2 = erp->er_extcount - idx;
  3597. nlists = ifp->if_real_bytes / XFS_IEXT_BUFSZ;
  3598. /*
  3599. * Save second part of target extent list
  3600. * (all extents past */
  3601. if (nex2) {
  3602. byte_diff = nex2 * sizeof(xfs_bmbt_rec_t);
  3603. nex2_ep = (xfs_bmbt_rec_t *) kmem_alloc(byte_diff, KM_SLEEP);
  3604. memmove(nex2_ep, &erp->er_extbuf[idx], byte_diff);
  3605. erp->er_extcount -= nex2;
  3606. xfs_iext_irec_update_extoffs(ifp, erp_idx + 1, -nex2);
  3607. memset(&erp->er_extbuf[idx], 0, byte_diff);
  3608. }
  3609. /*
  3610. * Add the new extents to the end of the target
  3611. * list, then allocate new irec record(s) and
  3612. * extent buffer(s) as needed to store the rest
  3613. * of the new extents.
  3614. */
  3615. ext_cnt = count;
  3616. ext_diff = MIN(ext_cnt, (int)XFS_LINEAR_EXTS - erp->er_extcount);
  3617. if (ext_diff) {
  3618. erp->er_extcount += ext_diff;
  3619. xfs_iext_irec_update_extoffs(ifp, erp_idx + 1, ext_diff);
  3620. ext_cnt -= ext_diff;
  3621. }
  3622. while (ext_cnt) {
  3623. erp_idx++;
  3624. erp = xfs_iext_irec_new(ifp, erp_idx);
  3625. ext_diff = MIN(ext_cnt, (int)XFS_LINEAR_EXTS);
  3626. erp->er_extcount = ext_diff;
  3627. xfs_iext_irec_update_extoffs(ifp, erp_idx + 1, ext_diff);
  3628. ext_cnt -= ext_diff;
  3629. }
  3630. /* Add nex2 extents back to indirection array */
  3631. if (nex2) {
  3632. xfs_extnum_t ext_avail;
  3633. int i;
  3634. byte_diff = nex2 * sizeof(xfs_bmbt_rec_t);
  3635. ext_avail = XFS_LINEAR_EXTS - erp->er_extcount;
  3636. i = 0;
  3637. /*
  3638. * If nex2 extents fit in the current page, append
  3639. * nex2_ep after the new extents.
  3640. */
  3641. if (nex2 <= ext_avail) {
  3642. i = erp->er_extcount;
  3643. }
  3644. /*
  3645. * Otherwise, check if space is available in the
  3646. * next page.
  3647. */
  3648. else if ((erp_idx < nlists - 1) &&
  3649. (nex2 <= (ext_avail = XFS_LINEAR_EXTS -
  3650. ifp->if_u1.if_ext_irec[erp_idx+1].er_extcount))) {
  3651. erp_idx++;
  3652. erp++;
  3653. /* Create a hole for nex2 extents */
  3654. memmove(&erp->er_extbuf[nex2], erp->er_extbuf,
  3655. erp->er_extcount * sizeof(xfs_bmbt_rec_t));
  3656. }
  3657. /*
  3658. * Final choice, create a new extent page for
  3659. * nex2 extents.
  3660. */
  3661. else {
  3662. erp_idx++;
  3663. erp = xfs_iext_irec_new(ifp, erp_idx);
  3664. }
  3665. memmove(&erp->er_extbuf[i], nex2_ep, byte_diff);
  3666. kmem_free(nex2_ep, byte_diff);
  3667. erp->er_extcount += nex2;
  3668. xfs_iext_irec_update_extoffs(ifp, erp_idx + 1, nex2);
  3669. }
  3670. }
  3671. /*
  3672. * This is called when the amount of space required for incore file
  3673. * extents needs to be decreased. The ext_diff parameter stores the
  3674. * number of extents to be removed and the idx parameter contains
  3675. * the extent index where the extents will be removed from.
  3676. *
  3677. * If the amount of space needed has decreased below the linear
  3678. * limit, XFS_IEXT_BUFSZ, then switch to using the contiguous
  3679. * extent array. Otherwise, use kmem_realloc() to adjust the
  3680. * size to what is needed.
  3681. */
  3682. void
  3683. xfs_iext_remove(
  3684. xfs_ifork_t *ifp, /* inode fork pointer */
  3685. xfs_extnum_t idx, /* index to begin removing exts */
  3686. int ext_diff) /* number of extents to remove */
  3687. {
  3688. xfs_extnum_t nextents; /* number of extents in file */
  3689. int new_size; /* size of extents after removal */
  3690. ASSERT(ext_diff > 0);
  3691. nextents = ifp->if_bytes / (uint)sizeof(xfs_bmbt_rec_t);
  3692. new_size = (nextents - ext_diff) * sizeof(xfs_bmbt_rec_t);
  3693. if (new_size == 0) {
  3694. xfs_iext_destroy(ifp);
  3695. } else if (ifp->if_flags & XFS_IFEXTIREC) {
  3696. xfs_iext_remove_indirect(ifp, idx, ext_diff);
  3697. } else if (ifp->if_real_bytes) {
  3698. xfs_iext_remove_direct(ifp, idx, ext_diff);
  3699. } else {
  3700. xfs_iext_remove_inline(ifp, idx, ext_diff);
  3701. }
  3702. ifp->if_bytes = new_size;
  3703. }
  3704. /*
  3705. * This removes ext_diff extents from the inline buffer, beginning
  3706. * at extent index idx.
  3707. */
  3708. void
  3709. xfs_iext_remove_inline(
  3710. xfs_ifork_t *ifp, /* inode fork pointer */
  3711. xfs_extnum_t idx, /* index to begin removing exts */
  3712. int ext_diff) /* number of extents to remove */
  3713. {
  3714. int nextents; /* number of extents in file */
  3715. ASSERT(!(ifp->if_flags & XFS_IFEXTIREC));
  3716. ASSERT(idx < XFS_INLINE_EXTS);
  3717. nextents = ifp->if_bytes / (uint)sizeof(xfs_bmbt_rec_t);
  3718. ASSERT(((nextents - ext_diff) > 0) &&
  3719. (nextents - ext_diff) < XFS_INLINE_EXTS);
  3720. if (idx + ext_diff < nextents) {
  3721. memmove(&ifp->if_u2.if_inline_ext[idx],
  3722. &ifp->if_u2.if_inline_ext[idx + ext_diff],
  3723. (nextents - (idx + ext_diff)) *
  3724. sizeof(xfs_bmbt_rec_t));
  3725. memset(&ifp->if_u2.if_inline_ext[nextents - ext_diff],
  3726. 0, ext_diff * sizeof(xfs_bmbt_rec_t));
  3727. } else {
  3728. memset(&ifp->if_u2.if_inline_ext[idx], 0,
  3729. ext_diff * sizeof(xfs_bmbt_rec_t));
  3730. }
  3731. }
  3732. /*
  3733. * This removes ext_diff extents from a linear (direct) extent list,
  3734. * beginning at extent index idx. If the extents are being removed
  3735. * from the end of the list (ie. truncate) then we just need to re-
  3736. * allocate the list to remove the extra space. Otherwise, if the
  3737. * extents are being removed from the middle of the existing extent
  3738. * entries, then we first need to move the extent records beginning
  3739. * at idx + ext_diff up in the list to overwrite the records being
  3740. * removed, then remove the extra space via kmem_realloc.
  3741. */
  3742. void
  3743. xfs_iext_remove_direct(
  3744. xfs_ifork_t *ifp, /* inode fork pointer */
  3745. xfs_extnum_t idx, /* index to begin removing exts */
  3746. int ext_diff) /* number of extents to remove */
  3747. {
  3748. xfs_extnum_t nextents; /* number of extents in file */
  3749. int new_size; /* size of extents after removal */
  3750. ASSERT(!(ifp->if_flags & XFS_IFEXTIREC));
  3751. new_size = ifp->if_bytes -
  3752. (ext_diff * sizeof(xfs_bmbt_rec_t));
  3753. nextents = ifp->if_bytes / (uint)sizeof(xfs_bmbt_rec_t);
  3754. if (new_size == 0) {
  3755. xfs_iext_destroy(ifp);
  3756. return;
  3757. }
  3758. /* Move extents up in the list (if needed) */
  3759. if (idx + ext_diff < nextents) {
  3760. memmove(&ifp->if_u1.if_extents[idx],
  3761. &ifp->if_u1.if_extents[idx + ext_diff],
  3762. (nextents - (idx + ext_diff)) *
  3763. sizeof(xfs_bmbt_rec_t));
  3764. }
  3765. memset(&ifp->if_u1.if_extents[nextents - ext_diff],
  3766. 0, ext_diff * sizeof(xfs_bmbt_rec_t));
  3767. /*
  3768. * Reallocate the direct extent list. If the extents
  3769. * will fit inside the inode then xfs_iext_realloc_direct
  3770. * will switch from direct to inline extent allocation
  3771. * mode for us.
  3772. */
  3773. xfs_iext_realloc_direct(ifp, new_size);
  3774. ifp->if_bytes = new_size;
  3775. }
  3776. /*
  3777. * This is called when incore extents are being removed from the
  3778. * indirection array and the extents being removed span multiple extent
  3779. * buffers. The idx parameter contains the file extent index where we
  3780. * want to begin removing extents, and the count parameter contains
  3781. * how many extents need to be removed.
  3782. *
  3783. * |-------| |-------|
  3784. * | nex1 | | | nex1 - number of extents before idx
  3785. * |-------| | count |
  3786. * | | | | count - number of extents being removed at idx
  3787. * | count | |-------|
  3788. * | | | nex2 | nex2 - number of extents after idx + count
  3789. * |-------| |-------|
  3790. */
  3791. void
  3792. xfs_iext_remove_indirect(
  3793. xfs_ifork_t *ifp, /* inode fork pointer */
  3794. xfs_extnum_t idx, /* index to begin removing extents */
  3795. int count) /* number of extents to remove */
  3796. {
  3797. xfs_ext_irec_t *erp; /* indirection array pointer */
  3798. int erp_idx = 0; /* indirection array index */
  3799. xfs_extnum_t ext_cnt; /* extents left to remove */
  3800. xfs_extnum_t ext_diff; /* extents to remove in current list */
  3801. xfs_extnum_t nex1; /* number of extents before idx */
  3802. xfs_extnum_t nex2; /* extents after idx + count */
  3803. int nlists; /* entries in indirection array */
  3804. int page_idx = idx; /* index in target extent list */
  3805. ASSERT(ifp->if_flags & XFS_IFEXTIREC);
  3806. erp = xfs_iext_idx_to_irec(ifp, &page_idx, &erp_idx, 0);
  3807. ASSERT(erp != NULL);
  3808. nlists = ifp->if_real_bytes / XFS_IEXT_BUFSZ;
  3809. nex1 = page_idx;
  3810. ext_cnt = count;
  3811. while (ext_cnt) {
  3812. nex2 = MAX((erp->er_extcount - (nex1 + ext_cnt)), 0);
  3813. ext_diff = MIN(ext_cnt, (erp->er_extcount - nex1));
  3814. /*
  3815. * Check for deletion of entire list;
  3816. * xfs_iext_irec_remove() updates extent offsets.
  3817. */
  3818. if (ext_diff == erp->er_extcount) {
  3819. xfs_iext_irec_remove(ifp, erp_idx);
  3820. ext_cnt -= ext_diff;
  3821. nex1 = 0;
  3822. if (ext_cnt) {
  3823. ASSERT(erp_idx < ifp->if_real_bytes /
  3824. XFS_IEXT_BUFSZ);
  3825. erp = &ifp->if_u1.if_ext_irec[erp_idx];
  3826. nex1 = 0;
  3827. continue;
  3828. } else {
  3829. break;
  3830. }
  3831. }
  3832. /* Move extents up (if needed) */
  3833. if (nex2) {
  3834. memmove(&erp->er_extbuf[nex1],
  3835. &erp->er_extbuf[nex1 + ext_diff],
  3836. nex2 * sizeof(xfs_bmbt_rec_t));
  3837. }
  3838. /* Zero out rest of page */
  3839. memset(&erp->er_extbuf[nex1 + nex2], 0, (XFS_IEXT_BUFSZ -
  3840. ((nex1 + nex2) * sizeof(xfs_bmbt_rec_t))));
  3841. /* Update remaining counters */
  3842. erp->er_extcount -= ext_diff;
  3843. xfs_iext_irec_update_extoffs(ifp, erp_idx + 1, -ext_diff);
  3844. ext_cnt -= ext_diff;
  3845. nex1 = 0;
  3846. erp_idx++;
  3847. erp++;
  3848. }
  3849. ifp->if_bytes -= count * sizeof(xfs_bmbt_rec_t);
  3850. xfs_iext_irec_compact(ifp);
  3851. }
  3852. /*
  3853. * Create, destroy, or resize a linear (direct) block of extents.
  3854. */
  3855. void
  3856. xfs_iext_realloc_direct(
  3857. xfs_ifork_t *ifp, /* inode fork pointer */
  3858. int new_size) /* new size of extents */
  3859. {
  3860. int rnew_size; /* real new size of extents */
  3861. rnew_size = new_size;
  3862. ASSERT(!(ifp->if_flags & XFS_IFEXTIREC) ||
  3863. ((new_size >= 0) && (new_size <= XFS_IEXT_BUFSZ) &&
  3864. (new_size != ifp->if_real_bytes)));
  3865. /* Free extent records */
  3866. if (new_size == 0) {
  3867. xfs_iext_destroy(ifp);
  3868. }
  3869. /* Resize direct extent list and zero any new bytes */
  3870. else if (ifp->if_real_bytes) {
  3871. /* Check if extents will fit inside the inode */
  3872. if (new_size <= XFS_INLINE_EXTS * sizeof(xfs_bmbt_rec_t)) {
  3873. xfs_iext_direct_to_inline(ifp, new_size /
  3874. (uint)sizeof(xfs_bmbt_rec_t));
  3875. ifp->if_bytes = new_size;
  3876. return;
  3877. }
  3878. if (!is_power_of_2(new_size)){
  3879. rnew_size = xfs_iroundup(new_size);
  3880. }
  3881. if (rnew_size != ifp->if_real_bytes) {
  3882. ifp->if_u1.if_extents = (xfs_bmbt_rec_t *)
  3883. kmem_realloc(ifp->if_u1.if_extents,
  3884. rnew_size,
  3885. ifp->if_real_bytes,
  3886. KM_SLEEP);
  3887. }
  3888. if (rnew_size > ifp->if_real_bytes) {
  3889. memset(&ifp->if_u1.if_extents[ifp->if_bytes /
  3890. (uint)sizeof(xfs_bmbt_rec_t)], 0,
  3891. rnew_size - ifp->if_real_bytes);
  3892. }
  3893. }
  3894. /*
  3895. * Switch from the inline extent buffer to a direct
  3896. * extent list. Be sure to include the inline extent
  3897. * bytes in new_size.
  3898. */
  3899. else {
  3900. new_size += ifp->if_bytes;
  3901. if (!is_power_of_2(new_size)) {
  3902. rnew_size = xfs_iroundup(new_size);
  3903. }
  3904. xfs_iext_inline_to_direct(ifp, rnew_size);
  3905. }
  3906. ifp->if_real_bytes = rnew_size;
  3907. ifp->if_bytes = new_size;
  3908. }
  3909. /*
  3910. * Switch from linear (direct) extent records to inline buffer.
  3911. */
  3912. void
  3913. xfs_iext_direct_to_inline(
  3914. xfs_ifork_t *ifp, /* inode fork pointer */
  3915. xfs_extnum_t nextents) /* number of extents in file */
  3916. {
  3917. ASSERT(ifp->if_flags & XFS_IFEXTENTS);
  3918. ASSERT(nextents <= XFS_INLINE_EXTS);
  3919. /*
  3920. * The inline buffer was zeroed when we switched
  3921. * from inline to direct extent allocation mode,
  3922. * so we don't need to clear it here.
  3923. */
  3924. memcpy(ifp->if_u2.if_inline_ext, ifp->if_u1.if_extents,
  3925. nextents * sizeof(xfs_bmbt_rec_t));
  3926. kmem_free(ifp->if_u1.if_extents, ifp->if_real_bytes);
  3927. ifp->if_u1.if_extents = ifp->if_u2.if_inline_ext;
  3928. ifp->if_real_bytes = 0;
  3929. }
  3930. /*
  3931. * Switch from inline buffer to linear (direct) extent records.
  3932. * new_size should already be rounded up to the next power of 2
  3933. * by the caller (when appropriate), so use new_size as it is.
  3934. * However, since new_size may be rounded up, we can't update
  3935. * if_bytes here. It is the caller's responsibility to update
  3936. * if_bytes upon return.
  3937. */
  3938. void
  3939. xfs_iext_inline_to_direct(
  3940. xfs_ifork_t *ifp, /* inode fork pointer */
  3941. int new_size) /* number of extents in file */
  3942. {
  3943. ifp->if_u1.if_extents = (xfs_bmbt_rec_t *)
  3944. kmem_alloc(new_size, KM_SLEEP);
  3945. memset(ifp->if_u1.if_extents, 0, new_size);
  3946. if (ifp->if_bytes) {
  3947. memcpy(ifp->if_u1.if_extents, ifp->if_u2.if_inline_ext,
  3948. ifp->if_bytes);
  3949. memset(ifp->if_u2.if_inline_ext, 0, XFS_INLINE_EXTS *
  3950. sizeof(xfs_bmbt_rec_t));
  3951. }
  3952. ifp->if_real_bytes = new_size;
  3953. }
  3954. /*
  3955. * Resize an extent indirection array to new_size bytes.
  3956. */
  3957. void
  3958. xfs_iext_realloc_indirect(
  3959. xfs_ifork_t *ifp, /* inode fork pointer */
  3960. int new_size) /* new indirection array size */
  3961. {
  3962. int nlists; /* number of irec's (ex lists) */
  3963. int size; /* current indirection array size */
  3964. ASSERT(ifp->if_flags & XFS_IFEXTIREC);
  3965. nlists = ifp->if_real_bytes / XFS_IEXT_BUFSZ;
  3966. size = nlists * sizeof(xfs_ext_irec_t);
  3967. ASSERT(ifp->if_real_bytes);
  3968. ASSERT((new_size >= 0) && (new_size != size));
  3969. if (new_size == 0) {
  3970. xfs_iext_destroy(ifp);
  3971. } else {
  3972. ifp->if_u1.if_ext_irec = (xfs_ext_irec_t *)
  3973. kmem_realloc(ifp->if_u1.if_ext_irec,
  3974. new_size, size, KM_SLEEP);
  3975. }
  3976. }
  3977. /*
  3978. * Switch from indirection array to linear (direct) extent allocations.
  3979. */
  3980. void
  3981. xfs_iext_indirect_to_direct(
  3982. xfs_ifork_t *ifp) /* inode fork pointer */
  3983. {
  3984. xfs_bmbt_rec_t *ep; /* extent record pointer */
  3985. xfs_extnum_t nextents; /* number of extents in file */
  3986. int size; /* size of file extents */
  3987. ASSERT(ifp->if_flags & XFS_IFEXTIREC);
  3988. nextents = ifp->if_bytes / (uint)sizeof(xfs_bmbt_rec_t);
  3989. ASSERT(nextents <= XFS_LINEAR_EXTS);
  3990. size = nextents * sizeof(xfs_bmbt_rec_t);
  3991. xfs_iext_irec_compact_full(ifp);
  3992. ASSERT(ifp->if_real_bytes == XFS_IEXT_BUFSZ);
  3993. ep = ifp->if_u1.if_ext_irec->er_extbuf;
  3994. kmem_free(ifp->if_u1.if_ext_irec, sizeof(xfs_ext_irec_t));
  3995. ifp->if_flags &= ~XFS_IFEXTIREC;
  3996. ifp->if_u1.if_extents = ep;
  3997. ifp->if_bytes = size;
  3998. if (nextents < XFS_LINEAR_EXTS) {
  3999. xfs_iext_realloc_direct(ifp, size);
  4000. }
  4001. }
  4002. /*
  4003. * Free incore file extents.
  4004. */
  4005. void
  4006. xfs_iext_destroy(
  4007. xfs_ifork_t *ifp) /* inode fork pointer */
  4008. {
  4009. if (ifp->if_flags & XFS_IFEXTIREC) {
  4010. int erp_idx;
  4011. int nlists;
  4012. nlists = ifp->if_real_bytes / XFS_IEXT_BUFSZ;
  4013. for (erp_idx = nlists - 1; erp_idx >= 0 ; erp_idx--) {
  4014. xfs_iext_irec_remove(ifp, erp_idx);
  4015. }
  4016. ifp->if_flags &= ~XFS_IFEXTIREC;
  4017. } else if (ifp->if_real_bytes) {
  4018. kmem_free(ifp->if_u1.if_extents, ifp->if_real_bytes);
  4019. } else if (ifp->if_bytes) {
  4020. memset(ifp->if_u2.if_inline_ext, 0, XFS_INLINE_EXTS *
  4021. sizeof(xfs_bmbt_rec_t));
  4022. }
  4023. ifp->if_u1.if_extents = NULL;
  4024. ifp->if_real_bytes = 0;
  4025. ifp->if_bytes = 0;
  4026. }
  4027. /*
  4028. * Return a pointer to the extent record for file system block bno.
  4029. */
  4030. xfs_bmbt_rec_t * /* pointer to found extent record */
  4031. xfs_iext_bno_to_ext(
  4032. xfs_ifork_t *ifp, /* inode fork pointer */
  4033. xfs_fileoff_t bno, /* block number to search for */
  4034. xfs_extnum_t *idxp) /* index of target extent */
  4035. {
  4036. xfs_bmbt_rec_t *base; /* pointer to first extent */
  4037. xfs_filblks_t blockcount = 0; /* number of blocks in extent */
  4038. xfs_bmbt_rec_t *ep = NULL; /* pointer to target extent */
  4039. xfs_ext_irec_t *erp = NULL; /* indirection array pointer */
  4040. int high; /* upper boundary in search */
  4041. xfs_extnum_t idx = 0; /* index of target extent */
  4042. int low; /* lower boundary in search */
  4043. xfs_extnum_t nextents; /* number of file extents */
  4044. xfs_fileoff_t startoff = 0; /* start offset of extent */
  4045. nextents = ifp->if_bytes / (uint)sizeof(xfs_bmbt_rec_t);
  4046. if (nextents == 0) {
  4047. *idxp = 0;
  4048. return NULL;
  4049. }
  4050. low = 0;
  4051. if (ifp->if_flags & XFS_IFEXTIREC) {
  4052. /* Find target extent list */
  4053. int erp_idx = 0;
  4054. erp = xfs_iext_bno_to_irec(ifp, bno, &erp_idx);
  4055. base = erp->er_extbuf;
  4056. high = erp->er_extcount - 1;
  4057. } else {
  4058. base = ifp->if_u1.if_extents;
  4059. high = nextents - 1;
  4060. }
  4061. /* Binary search extent records */
  4062. while (low <= high) {
  4063. idx = (low + high) >> 1;
  4064. ep = base + idx;
  4065. startoff = xfs_bmbt_get_startoff(ep);
  4066. blockcount = xfs_bmbt_get_blockcount(ep);
  4067. if (bno < startoff) {
  4068. high = idx - 1;
  4069. } else if (bno >= startoff + blockcount) {
  4070. low = idx + 1;
  4071. } else {
  4072. /* Convert back to file-based extent index */
  4073. if (ifp->if_flags & XFS_IFEXTIREC) {
  4074. idx += erp->er_extoff;
  4075. }
  4076. *idxp = idx;
  4077. return ep;
  4078. }
  4079. }
  4080. /* Convert back to file-based extent index */
  4081. if (ifp->if_flags & XFS_IFEXTIREC) {
  4082. idx += erp->er_extoff;
  4083. }
  4084. if (bno >= startoff + blockcount) {
  4085. if (++idx == nextents) {
  4086. ep = NULL;
  4087. } else {
  4088. ep = xfs_iext_get_ext(ifp, idx);
  4089. }
  4090. }
  4091. *idxp = idx;
  4092. return ep;
  4093. }
  4094. /*
  4095. * Return a pointer to the indirection array entry containing the
  4096. * extent record for filesystem block bno. Store the index of the
  4097. * target irec in *erp_idxp.
  4098. */
  4099. xfs_ext_irec_t * /* pointer to found extent record */
  4100. xfs_iext_bno_to_irec(
  4101. xfs_ifork_t *ifp, /* inode fork pointer */
  4102. xfs_fileoff_t bno, /* block number to search for */
  4103. int *erp_idxp) /* irec index of target ext list */
  4104. {
  4105. xfs_ext_irec_t *erp = NULL; /* indirection array pointer */
  4106. xfs_ext_irec_t *erp_next; /* next indirection array entry */
  4107. int erp_idx; /* indirection array index */
  4108. int nlists; /* number of extent irec's (lists) */
  4109. int high; /* binary search upper limit */
  4110. int low; /* binary search lower limit */
  4111. ASSERT(ifp->if_flags & XFS_IFEXTIREC);
  4112. nlists = ifp->if_real_bytes / XFS_IEXT_BUFSZ;
  4113. erp_idx = 0;
  4114. low = 0;
  4115. high = nlists - 1;
  4116. while (low <= high) {
  4117. erp_idx = (low + high) >> 1;
  4118. erp = &ifp->if_u1.if_ext_irec[erp_idx];
  4119. erp_next = erp_idx < nlists - 1 ? erp + 1 : NULL;
  4120. if (bno < xfs_bmbt_get_startoff(erp->er_extbuf)) {
  4121. high = erp_idx - 1;
  4122. } else if (erp_next && bno >=
  4123. xfs_bmbt_get_startoff(erp_next->er_extbuf)) {
  4124. low = erp_idx + 1;
  4125. } else {
  4126. break;
  4127. }
  4128. }
  4129. *erp_idxp = erp_idx;
  4130. return erp;
  4131. }
  4132. /*
  4133. * Return a pointer to the indirection array entry containing the
  4134. * extent record at file extent index *idxp. Store the index of the
  4135. * target irec in *erp_idxp and store the page index of the target
  4136. * extent record in *idxp.
  4137. */
  4138. xfs_ext_irec_t *
  4139. xfs_iext_idx_to_irec(
  4140. xfs_ifork_t *ifp, /* inode fork pointer */
  4141. xfs_extnum_t *idxp, /* extent index (file -> page) */
  4142. int *erp_idxp, /* pointer to target irec */
  4143. int realloc) /* new bytes were just added */
  4144. {
  4145. xfs_ext_irec_t *prev; /* pointer to previous irec */
  4146. xfs_ext_irec_t *erp = NULL; /* pointer to current irec */
  4147. int erp_idx; /* indirection array index */
  4148. int nlists; /* number of irec's (ex lists) */
  4149. int high; /* binary search upper limit */
  4150. int low; /* binary search lower limit */
  4151. xfs_extnum_t page_idx = *idxp; /* extent index in target list */
  4152. ASSERT(ifp->if_flags & XFS_IFEXTIREC);
  4153. ASSERT(page_idx >= 0 && page_idx <=
  4154. ifp->if_bytes / (uint)sizeof(xfs_bmbt_rec_t));
  4155. nlists = ifp->if_real_bytes / XFS_IEXT_BUFSZ;
  4156. erp_idx = 0;
  4157. low = 0;
  4158. high = nlists - 1;
  4159. /* Binary search extent irec's */
  4160. while (low <= high) {
  4161. erp_idx = (low + high) >> 1;
  4162. erp = &ifp->if_u1.if_ext_irec[erp_idx];
  4163. prev = erp_idx > 0 ? erp - 1 : NULL;
  4164. if (page_idx < erp->er_extoff || (page_idx == erp->er_extoff &&
  4165. realloc && prev && prev->er_extcount < XFS_LINEAR_EXTS)) {
  4166. high = erp_idx - 1;
  4167. } else if (page_idx > erp->er_extoff + erp->er_extcount ||
  4168. (page_idx == erp->er_extoff + erp->er_extcount &&
  4169. !realloc)) {
  4170. low = erp_idx + 1;
  4171. } else if (page_idx == erp->er_extoff + erp->er_extcount &&
  4172. erp->er_extcount == XFS_LINEAR_EXTS) {
  4173. ASSERT(realloc);
  4174. page_idx = 0;
  4175. erp_idx++;
  4176. erp = erp_idx < nlists ? erp + 1 : NULL;
  4177. break;
  4178. } else {
  4179. page_idx -= erp->er_extoff;
  4180. break;
  4181. }
  4182. }
  4183. *idxp = page_idx;
  4184. *erp_idxp = erp_idx;
  4185. return(erp);
  4186. }
  4187. /*
  4188. * Allocate and initialize an indirection array once the space needed
  4189. * for incore extents increases above XFS_IEXT_BUFSZ.
  4190. */
  4191. void
  4192. xfs_iext_irec_init(
  4193. xfs_ifork_t *ifp) /* inode fork pointer */
  4194. {
  4195. xfs_ext_irec_t *erp; /* indirection array pointer */
  4196. xfs_extnum_t nextents; /* number of extents in file */
  4197. ASSERT(!(ifp->if_flags & XFS_IFEXTIREC));
  4198. nextents = ifp->if_bytes / (uint)sizeof(xfs_bmbt_rec_t);
  4199. ASSERT(nextents <= XFS_LINEAR_EXTS);
  4200. erp = (xfs_ext_irec_t *)
  4201. kmem_alloc(sizeof(xfs_ext_irec_t), KM_SLEEP);
  4202. if (nextents == 0) {
  4203. ifp->if_u1.if_extents = (xfs_bmbt_rec_t *)
  4204. kmem_alloc(XFS_IEXT_BUFSZ, KM_SLEEP);
  4205. } else if (!ifp->if_real_bytes) {
  4206. xfs_iext_inline_to_direct(ifp, XFS_IEXT_BUFSZ);
  4207. } else if (ifp->if_real_bytes < XFS_IEXT_BUFSZ) {
  4208. xfs_iext_realloc_direct(ifp, XFS_IEXT_BUFSZ);
  4209. }
  4210. erp->er_extbuf = ifp->if_u1.if_extents;
  4211. erp->er_extcount = nextents;
  4212. erp->er_extoff = 0;
  4213. ifp->if_flags |= XFS_IFEXTIREC;
  4214. ifp->if_real_bytes = XFS_IEXT_BUFSZ;
  4215. ifp->if_bytes = nextents * sizeof(xfs_bmbt_rec_t);
  4216. ifp->if_u1.if_ext_irec = erp;
  4217. return;
  4218. }
  4219. /*
  4220. * Allocate and initialize a new entry in the indirection array.
  4221. */
  4222. xfs_ext_irec_t *
  4223. xfs_iext_irec_new(
  4224. xfs_ifork_t *ifp, /* inode fork pointer */
  4225. int erp_idx) /* index for new irec */
  4226. {
  4227. xfs_ext_irec_t *erp; /* indirection array pointer */
  4228. int i; /* loop counter */
  4229. int nlists; /* number of irec's (ex lists) */
  4230. ASSERT(ifp->if_flags & XFS_IFEXTIREC);
  4231. nlists = ifp->if_real_bytes / XFS_IEXT_BUFSZ;
  4232. /* Resize indirection array */
  4233. xfs_iext_realloc_indirect(ifp, ++nlists *
  4234. sizeof(xfs_ext_irec_t));
  4235. /*
  4236. * Move records down in the array so the
  4237. * new page can use erp_idx.
  4238. */
  4239. erp = ifp->if_u1.if_ext_irec;
  4240. for (i = nlists - 1; i > erp_idx; i--) {
  4241. memmove(&erp[i], &erp[i-1], sizeof(xfs_ext_irec_t));
  4242. }
  4243. ASSERT(i == erp_idx);
  4244. /* Initialize new extent record */
  4245. erp = ifp->if_u1.if_ext_irec;
  4246. erp[erp_idx].er_extbuf = (xfs_bmbt_rec_t *)
  4247. kmem_alloc(XFS_IEXT_BUFSZ, KM_SLEEP);
  4248. ifp->if_real_bytes = nlists * XFS_IEXT_BUFSZ;
  4249. memset(erp[erp_idx].er_extbuf, 0, XFS_IEXT_BUFSZ);
  4250. erp[erp_idx].er_extcount = 0;
  4251. erp[erp_idx].er_extoff = erp_idx > 0 ?
  4252. erp[erp_idx-1].er_extoff + erp[erp_idx-1].er_extcount : 0;
  4253. return (&erp[erp_idx]);
  4254. }
  4255. /*
  4256. * Remove a record from the indirection array.
  4257. */
  4258. void
  4259. xfs_iext_irec_remove(
  4260. xfs_ifork_t *ifp, /* inode fork pointer */
  4261. int erp_idx) /* irec index to remove */
  4262. {
  4263. xfs_ext_irec_t *erp; /* indirection array pointer */
  4264. int i; /* loop counter */
  4265. int nlists; /* number of irec's (ex lists) */
  4266. ASSERT(ifp->if_flags & XFS_IFEXTIREC);
  4267. nlists = ifp->if_real_bytes / XFS_IEXT_BUFSZ;
  4268. erp = &ifp->if_u1.if_ext_irec[erp_idx];
  4269. if (erp->er_extbuf) {
  4270. xfs_iext_irec_update_extoffs(ifp, erp_idx + 1,
  4271. -erp->er_extcount);
  4272. kmem_free(erp->er_extbuf, XFS_IEXT_BUFSZ);
  4273. }
  4274. /* Compact extent records */
  4275. erp = ifp->if_u1.if_ext_irec;
  4276. for (i = erp_idx; i < nlists - 1; i++) {
  4277. memmove(&erp[i], &erp[i+1], sizeof(xfs_ext_irec_t));
  4278. }
  4279. /*
  4280. * Manually free the last extent record from the indirection
  4281. * array. A call to xfs_iext_realloc_indirect() with a size
  4282. * of zero would result in a call to xfs_iext_destroy() which
  4283. * would in turn call this function again, creating a nasty
  4284. * infinite loop.
  4285. */
  4286. if (--nlists) {
  4287. xfs_iext_realloc_indirect(ifp,
  4288. nlists * sizeof(xfs_ext_irec_t));
  4289. } else {
  4290. kmem_free(ifp->if_u1.if_ext_irec,
  4291. sizeof(xfs_ext_irec_t));
  4292. }
  4293. ifp->if_real_bytes = nlists * XFS_IEXT_BUFSZ;
  4294. }
  4295. /*
  4296. * This is called to clean up large amounts of unused memory allocated
  4297. * by the indirection array. Before compacting anything though, verify
  4298. * that the indirection array is still needed and switch back to the
  4299. * linear extent list (or even the inline buffer) if possible. The
  4300. * compaction policy is as follows:
  4301. *
  4302. * Full Compaction: Extents fit into a single page (or inline buffer)
  4303. * Full Compaction: Extents occupy less than 10% of allocated space
  4304. * Partial Compaction: Extents occupy > 10% and < 50% of allocated space
  4305. * No Compaction: Extents occupy at least 50% of allocated space
  4306. */
  4307. void
  4308. xfs_iext_irec_compact(
  4309. xfs_ifork_t *ifp) /* inode fork pointer */
  4310. {
  4311. xfs_extnum_t nextents; /* number of extents in file */
  4312. int nlists; /* number of irec's (ex lists) */
  4313. ASSERT(ifp->if_flags & XFS_IFEXTIREC);
  4314. nlists = ifp->if_real_bytes / XFS_IEXT_BUFSZ;
  4315. nextents = ifp->if_bytes / (uint)sizeof(xfs_bmbt_rec_t);
  4316. if (nextents == 0) {
  4317. xfs_iext_destroy(ifp);
  4318. } else if (nextents <= XFS_INLINE_EXTS) {
  4319. xfs_iext_indirect_to_direct(ifp);
  4320. xfs_iext_direct_to_inline(ifp, nextents);
  4321. } else if (nextents <= XFS_LINEAR_EXTS) {
  4322. xfs_iext_indirect_to_direct(ifp);
  4323. } else if (nextents < (nlists * XFS_LINEAR_EXTS) >> 3) {
  4324. xfs_iext_irec_compact_full(ifp);
  4325. } else if (nextents < (nlists * XFS_LINEAR_EXTS) >> 1) {
  4326. xfs_iext_irec_compact_pages(ifp);
  4327. }
  4328. }
  4329. /*
  4330. * Combine extents from neighboring extent pages.
  4331. */
  4332. void
  4333. xfs_iext_irec_compact_pages(
  4334. xfs_ifork_t *ifp) /* inode fork pointer */
  4335. {
  4336. xfs_ext_irec_t *erp, *erp_next;/* pointers to irec entries */
  4337. int erp_idx = 0; /* indirection array index */
  4338. int nlists; /* number of irec's (ex lists) */
  4339. ASSERT(ifp->if_flags & XFS_IFEXTIREC);
  4340. nlists = ifp->if_real_bytes / XFS_IEXT_BUFSZ;
  4341. while (erp_idx < nlists - 1) {
  4342. erp = &ifp->if_u1.if_ext_irec[erp_idx];
  4343. erp_next = erp + 1;
  4344. if (erp_next->er_extcount <=
  4345. (XFS_LINEAR_EXTS - erp->er_extcount)) {
  4346. memmove(&erp->er_extbuf[erp->er_extcount],
  4347. erp_next->er_extbuf, erp_next->er_extcount *
  4348. sizeof(xfs_bmbt_rec_t));
  4349. erp->er_extcount += erp_next->er_extcount;
  4350. /*
  4351. * Free page before removing extent record
  4352. * so er_extoffs don't get modified in
  4353. * xfs_iext_irec_remove.
  4354. */
  4355. kmem_free(erp_next->er_extbuf, XFS_IEXT_BUFSZ);
  4356. erp_next->er_extbuf = NULL;
  4357. xfs_iext_irec_remove(ifp, erp_idx + 1);
  4358. nlists = ifp->if_real_bytes / XFS_IEXT_BUFSZ;
  4359. } else {
  4360. erp_idx++;
  4361. }
  4362. }
  4363. }
  4364. /*
  4365. * Fully compact the extent records managed by the indirection array.
  4366. */
  4367. void
  4368. xfs_iext_irec_compact_full(
  4369. xfs_ifork_t *ifp) /* inode fork pointer */
  4370. {
  4371. xfs_bmbt_rec_t *ep, *ep_next; /* extent record pointers */
  4372. xfs_ext_irec_t *erp, *erp_next; /* extent irec pointers */
  4373. int erp_idx = 0; /* extent irec index */
  4374. int ext_avail; /* empty entries in ex list */
  4375. int ext_diff; /* number of exts to add */
  4376. int nlists; /* number of irec's (ex lists) */
  4377. ASSERT(ifp->if_flags & XFS_IFEXTIREC);
  4378. nlists = ifp->if_real_bytes / XFS_IEXT_BUFSZ;
  4379. erp = ifp->if_u1.if_ext_irec;
  4380. ep = &erp->er_extbuf[erp->er_extcount];
  4381. erp_next = erp + 1;
  4382. ep_next = erp_next->er_extbuf;
  4383. while (erp_idx < nlists - 1) {
  4384. ext_avail = XFS_LINEAR_EXTS - erp->er_extcount;
  4385. ext_diff = MIN(ext_avail, erp_next->er_extcount);
  4386. memcpy(ep, ep_next, ext_diff * sizeof(xfs_bmbt_rec_t));
  4387. erp->er_extcount += ext_diff;
  4388. erp_next->er_extcount -= ext_diff;
  4389. /* Remove next page */
  4390. if (erp_next->er_extcount == 0) {
  4391. /*
  4392. * Free page before removing extent record
  4393. * so er_extoffs don't get modified in
  4394. * xfs_iext_irec_remove.
  4395. */
  4396. kmem_free(erp_next->er_extbuf,
  4397. erp_next->er_extcount * sizeof(xfs_bmbt_rec_t));
  4398. erp_next->er_extbuf = NULL;
  4399. xfs_iext_irec_remove(ifp, erp_idx + 1);
  4400. erp = &ifp->if_u1.if_ext_irec[erp_idx];
  4401. nlists = ifp->if_real_bytes / XFS_IEXT_BUFSZ;
  4402. /* Update next page */
  4403. } else {
  4404. /* Move rest of page up to become next new page */
  4405. memmove(erp_next->er_extbuf, ep_next,
  4406. erp_next->er_extcount * sizeof(xfs_bmbt_rec_t));
  4407. ep_next = erp_next->er_extbuf;
  4408. memset(&ep_next[erp_next->er_extcount], 0,
  4409. (XFS_LINEAR_EXTS - erp_next->er_extcount) *
  4410. sizeof(xfs_bmbt_rec_t));
  4411. }
  4412. if (erp->er_extcount == XFS_LINEAR_EXTS) {
  4413. erp_idx++;
  4414. if (erp_idx < nlists)
  4415. erp = &ifp->if_u1.if_ext_irec[erp_idx];
  4416. else
  4417. break;
  4418. }
  4419. ep = &erp->er_extbuf[erp->er_extcount];
  4420. erp_next = erp + 1;
  4421. ep_next = erp_next->er_extbuf;
  4422. }
  4423. }
  4424. /*
  4425. * This is called to update the er_extoff field in the indirection
  4426. * array when extents have been added or removed from one of the
  4427. * extent lists. erp_idx contains the irec index to begin updating
  4428. * at and ext_diff contains the number of extents that were added
  4429. * or removed.
  4430. */
  4431. void
  4432. xfs_iext_irec_update_extoffs(
  4433. xfs_ifork_t *ifp, /* inode fork pointer */
  4434. int erp_idx, /* irec index to update */
  4435. int ext_diff) /* number of new extents */
  4436. {
  4437. int i; /* loop counter */
  4438. int nlists; /* number of irec's (ex lists */
  4439. ASSERT(ifp->if_flags & XFS_IFEXTIREC);
  4440. nlists = ifp->if_real_bytes / XFS_IEXT_BUFSZ;
  4441. for (i = erp_idx; i < nlists; i++) {
  4442. ifp->if_u1.if_ext_irec[i].er_extoff += ext_diff;
  4443. }
  4444. }