inode.c 159 KB

1234567891011121314151617181920212223242526272829303132333435363738394041424344454647484950515253545556575859606162636465666768697071727374757677787980818283848586878889909192939495969798991001011021031041051061071081091101111121131141151161171181191201211221231241251261271281291301311321331341351361371381391401411421431441451461471481491501511521531541551561571581591601611621631641651661671681691701711721731741751761771781791801811821831841851861871881891901911921931941951961971981992002012022032042052062072082092102112122132142152162172182192202212222232242252262272282292302312322332342352362372382392402412422432442452462472482492502512522532542552562572582592602612622632642652662672682692702712722732742752762772782792802812822832842852862872882892902912922932942952962972982993003013023033043053063073083093103113123133143153163173183193203213223233243253263273283293303313323333343353363373383393403413423433443453463473483493503513523533543553563573583593603613623633643653663673683693703713723733743753763773783793803813823833843853863873883893903913923933943953963973983994004014024034044054064074084094104114124134144154164174184194204214224234244254264274284294304314324334344354364374384394404414424434444454464474484494504514524534544554564574584594604614624634644654664674684694704714724734744754764774784794804814824834844854864874884894904914924934944954964974984995005015025035045055065075085095105115125135145155165175185195205215225235245255265275285295305315325335345355365375385395405415425435445455465475485495505515525535545555565575585595605615625635645655665675685695705715725735745755765775785795805815825835845855865875885895905915925935945955965975985996006016026036046056066076086096106116126136146156166176186196206216226236246256266276286296306316326336346356366376386396406416426436446456466476486496506516526536546556566576586596606616626636646656666676686696706716726736746756766776786796806816826836846856866876886896906916926936946956966976986997007017027037047057067077087097107117127137147157167177187197207217227237247257267277287297307317327337347357367377387397407417427437447457467477487497507517527537547557567577587597607617627637647657667677687697707717727737747757767777787797807817827837847857867877887897907917927937947957967977987998008018028038048058068078088098108118128138148158168178188198208218228238248258268278288298308318328338348358368378388398408418428438448458468478488498508518528538548558568578588598608618628638648658668678688698708718728738748758768778788798808818828838848858868878888898908918928938948958968978988999009019029039049059069079089099109119129139149159169179189199209219229239249259269279289299309319329339349359369379389399409419429439449459469479489499509519529539549559569579589599609619629639649659669679689699709719729739749759769779789799809819829839849859869879889899909919929939949959969979989991000100110021003100410051006100710081009101010111012101310141015101610171018101910201021102210231024102510261027102810291030103110321033103410351036103710381039104010411042104310441045104610471048104910501051105210531054105510561057105810591060106110621063106410651066106710681069107010711072107310741075107610771078107910801081108210831084108510861087108810891090109110921093109410951096109710981099110011011102110311041105110611071108110911101111111211131114111511161117111811191120112111221123112411251126112711281129113011311132113311341135113611371138113911401141114211431144114511461147114811491150115111521153115411551156115711581159116011611162116311641165116611671168116911701171117211731174117511761177117811791180118111821183118411851186118711881189119011911192119311941195119611971198119912001201120212031204120512061207120812091210121112121213121412151216121712181219122012211222122312241225122612271228122912301231123212331234123512361237123812391240124112421243124412451246124712481249125012511252125312541255125612571258125912601261126212631264126512661267126812691270127112721273127412751276127712781279128012811282128312841285128612871288128912901291129212931294129512961297129812991300130113021303130413051306130713081309131013111312131313141315131613171318131913201321132213231324132513261327132813291330133113321333133413351336133713381339134013411342134313441345134613471348134913501351135213531354135513561357135813591360136113621363136413651366136713681369137013711372137313741375137613771378137913801381138213831384138513861387138813891390139113921393139413951396139713981399140014011402140314041405140614071408140914101411141214131414141514161417141814191420142114221423142414251426142714281429143014311432143314341435143614371438143914401441144214431444144514461447144814491450145114521453145414551456145714581459146014611462146314641465146614671468146914701471147214731474147514761477147814791480148114821483148414851486148714881489149014911492149314941495149614971498149915001501150215031504150515061507150815091510151115121513151415151516151715181519152015211522152315241525152615271528152915301531153215331534153515361537153815391540154115421543154415451546154715481549155015511552155315541555155615571558155915601561156215631564156515661567156815691570157115721573157415751576157715781579158015811582158315841585158615871588158915901591159215931594159515961597159815991600160116021603160416051606160716081609161016111612161316141615161616171618161916201621162216231624162516261627162816291630163116321633163416351636163716381639164016411642164316441645164616471648164916501651165216531654165516561657165816591660166116621663166416651666166716681669167016711672167316741675167616771678167916801681168216831684168516861687168816891690169116921693169416951696169716981699170017011702170317041705170617071708170917101711171217131714171517161717171817191720172117221723172417251726172717281729173017311732173317341735173617371738173917401741174217431744174517461747174817491750175117521753175417551756175717581759176017611762176317641765176617671768176917701771177217731774177517761777177817791780178117821783178417851786178717881789179017911792179317941795179617971798179918001801180218031804180518061807180818091810181118121813181418151816181718181819182018211822182318241825182618271828182918301831183218331834183518361837183818391840184118421843184418451846184718481849185018511852185318541855185618571858185918601861186218631864186518661867186818691870187118721873187418751876187718781879188018811882188318841885188618871888188918901891189218931894189518961897189818991900190119021903190419051906190719081909191019111912191319141915191619171918191919201921192219231924192519261927192819291930193119321933193419351936193719381939194019411942194319441945194619471948194919501951195219531954195519561957195819591960196119621963196419651966196719681969197019711972197319741975197619771978197919801981198219831984198519861987198819891990199119921993199419951996199719981999200020012002200320042005200620072008200920102011201220132014201520162017201820192020202120222023202420252026202720282029203020312032203320342035203620372038203920402041204220432044204520462047204820492050205120522053205420552056205720582059206020612062206320642065206620672068206920702071207220732074207520762077207820792080208120822083208420852086208720882089209020912092209320942095209620972098209921002101210221032104210521062107210821092110211121122113211421152116211721182119212021212122212321242125212621272128212921302131213221332134213521362137213821392140214121422143214421452146214721482149215021512152215321542155215621572158215921602161216221632164216521662167216821692170217121722173217421752176217721782179218021812182218321842185218621872188218921902191219221932194219521962197219821992200220122022203220422052206220722082209221022112212221322142215221622172218221922202221222222232224222522262227222822292230223122322233223422352236223722382239224022412242224322442245224622472248224922502251225222532254225522562257225822592260226122622263226422652266226722682269227022712272227322742275227622772278227922802281228222832284228522862287228822892290229122922293229422952296229722982299230023012302230323042305230623072308230923102311231223132314231523162317231823192320232123222323232423252326232723282329233023312332233323342335233623372338233923402341234223432344234523462347234823492350235123522353235423552356235723582359236023612362236323642365236623672368236923702371237223732374237523762377237823792380238123822383238423852386238723882389239023912392239323942395239623972398239924002401240224032404240524062407240824092410241124122413241424152416241724182419242024212422242324242425242624272428242924302431243224332434243524362437243824392440244124422443244424452446244724482449245024512452245324542455245624572458245924602461246224632464246524662467246824692470247124722473247424752476247724782479248024812482248324842485248624872488248924902491249224932494249524962497249824992500250125022503250425052506250725082509251025112512251325142515251625172518251925202521252225232524252525262527252825292530253125322533253425352536253725382539254025412542254325442545254625472548254925502551255225532554255525562557255825592560256125622563256425652566256725682569257025712572257325742575257625772578257925802581258225832584258525862587258825892590259125922593259425952596259725982599260026012602260326042605260626072608260926102611261226132614261526162617261826192620262126222623262426252626262726282629263026312632263326342635263626372638263926402641264226432644264526462647264826492650265126522653265426552656265726582659266026612662266326642665266626672668266926702671267226732674267526762677267826792680268126822683268426852686268726882689269026912692269326942695269626972698269927002701270227032704270527062707270827092710271127122713271427152716271727182719272027212722272327242725272627272728272927302731273227332734273527362737273827392740274127422743274427452746274727482749275027512752275327542755275627572758275927602761276227632764276527662767276827692770277127722773277427752776277727782779278027812782278327842785278627872788278927902791279227932794279527962797279827992800280128022803280428052806280728082809281028112812281328142815281628172818281928202821282228232824282528262827282828292830283128322833283428352836283728382839284028412842284328442845284628472848284928502851285228532854285528562857285828592860286128622863286428652866286728682869287028712872287328742875287628772878287928802881288228832884288528862887288828892890289128922893289428952896289728982899290029012902290329042905290629072908290929102911291229132914291529162917291829192920292129222923292429252926292729282929293029312932293329342935293629372938293929402941294229432944294529462947294829492950295129522953295429552956295729582959296029612962296329642965296629672968296929702971297229732974297529762977297829792980298129822983298429852986298729882989299029912992299329942995299629972998299930003001300230033004300530063007300830093010301130123013301430153016301730183019302030213022302330243025302630273028302930303031303230333034303530363037303830393040304130423043304430453046304730483049305030513052305330543055305630573058305930603061306230633064306530663067306830693070307130723073307430753076307730783079308030813082308330843085308630873088308930903091309230933094309530963097309830993100310131023103310431053106310731083109311031113112311331143115311631173118311931203121312231233124312531263127312831293130313131323133313431353136313731383139314031413142314331443145314631473148314931503151315231533154315531563157315831593160316131623163316431653166316731683169317031713172317331743175317631773178317931803181318231833184318531863187318831893190319131923193319431953196319731983199320032013202320332043205320632073208320932103211321232133214321532163217321832193220322132223223322432253226322732283229323032313232323332343235323632373238323932403241324232433244324532463247324832493250325132523253325432553256325732583259326032613262326332643265326632673268326932703271327232733274327532763277327832793280328132823283328432853286328732883289329032913292329332943295329632973298329933003301330233033304330533063307330833093310331133123313331433153316331733183319332033213322332333243325332633273328332933303331333233333334333533363337333833393340334133423343334433453346334733483349335033513352335333543355335633573358335933603361336233633364336533663367336833693370337133723373337433753376337733783379338033813382338333843385338633873388338933903391339233933394339533963397339833993400340134023403340434053406340734083409341034113412341334143415341634173418341934203421342234233424342534263427342834293430343134323433343434353436343734383439344034413442344334443445344634473448344934503451345234533454345534563457345834593460346134623463346434653466346734683469347034713472347334743475347634773478347934803481348234833484348534863487348834893490349134923493349434953496349734983499350035013502350335043505350635073508350935103511351235133514351535163517351835193520352135223523352435253526352735283529353035313532353335343535353635373538353935403541354235433544354535463547354835493550355135523553355435553556355735583559356035613562356335643565356635673568356935703571357235733574357535763577357835793580358135823583358435853586358735883589359035913592359335943595359635973598359936003601360236033604360536063607360836093610361136123613361436153616361736183619362036213622362336243625362636273628362936303631363236333634363536363637363836393640364136423643364436453646364736483649365036513652365336543655365636573658365936603661366236633664366536663667366836693670367136723673367436753676367736783679368036813682368336843685368636873688368936903691369236933694369536963697369836993700370137023703370437053706370737083709371037113712371337143715371637173718371937203721372237233724372537263727372837293730373137323733373437353736373737383739374037413742374337443745374637473748374937503751375237533754375537563757375837593760376137623763376437653766376737683769377037713772377337743775377637773778377937803781378237833784378537863787378837893790379137923793379437953796379737983799380038013802380338043805380638073808380938103811381238133814381538163817381838193820382138223823382438253826382738283829383038313832383338343835383638373838383938403841384238433844384538463847384838493850385138523853385438553856385738583859386038613862386338643865386638673868386938703871387238733874387538763877387838793880388138823883388438853886388738883889389038913892389338943895389638973898389939003901390239033904390539063907390839093910391139123913391439153916391739183919392039213922392339243925392639273928392939303931393239333934393539363937393839393940394139423943394439453946394739483949395039513952395339543955395639573958395939603961396239633964396539663967396839693970397139723973397439753976397739783979398039813982398339843985398639873988398939903991399239933994399539963997399839994000400140024003400440054006400740084009401040114012401340144015401640174018401940204021402240234024402540264027402840294030403140324033403440354036403740384039404040414042404340444045404640474048404940504051405240534054405540564057405840594060406140624063406440654066406740684069407040714072407340744075407640774078407940804081408240834084408540864087408840894090409140924093409440954096409740984099410041014102410341044105410641074108410941104111411241134114411541164117411841194120412141224123412441254126412741284129413041314132413341344135413641374138413941404141414241434144414541464147414841494150415141524153415441554156415741584159416041614162416341644165416641674168416941704171417241734174417541764177417841794180418141824183418441854186418741884189419041914192419341944195419641974198419942004201420242034204420542064207420842094210421142124213421442154216421742184219422042214222422342244225422642274228422942304231423242334234423542364237423842394240424142424243424442454246424742484249425042514252425342544255425642574258425942604261426242634264426542664267426842694270427142724273427442754276427742784279428042814282428342844285428642874288428942904291429242934294429542964297429842994300430143024303430443054306430743084309431043114312431343144315431643174318431943204321432243234324432543264327432843294330433143324333433443354336433743384339434043414342434343444345434643474348434943504351435243534354435543564357435843594360436143624363436443654366436743684369437043714372437343744375437643774378437943804381438243834384438543864387438843894390439143924393439443954396439743984399440044014402440344044405440644074408440944104411441244134414441544164417441844194420442144224423442444254426442744284429443044314432443344344435443644374438443944404441444244434444444544464447444844494450445144524453445444554456445744584459446044614462446344644465446644674468446944704471447244734474447544764477447844794480448144824483448444854486448744884489449044914492449344944495449644974498449945004501450245034504450545064507450845094510451145124513451445154516451745184519452045214522452345244525452645274528452945304531453245334534453545364537453845394540454145424543454445454546454745484549455045514552455345544555455645574558455945604561456245634564456545664567456845694570457145724573457445754576457745784579458045814582458345844585458645874588458945904591459245934594459545964597459845994600460146024603460446054606460746084609461046114612461346144615461646174618461946204621462246234624462546264627462846294630463146324633463446354636463746384639464046414642464346444645464646474648464946504651465246534654465546564657465846594660466146624663466446654666466746684669467046714672467346744675467646774678467946804681468246834684468546864687468846894690469146924693469446954696469746984699470047014702470347044705470647074708470947104711471247134714471547164717471847194720472147224723472447254726472747284729473047314732473347344735473647374738473947404741474247434744474547464747474847494750475147524753475447554756475747584759476047614762476347644765476647674768476947704771477247734774477547764777477847794780478147824783478447854786478747884789479047914792479347944795479647974798479948004801480248034804480548064807480848094810481148124813481448154816481748184819482048214822482348244825482648274828482948304831483248334834483548364837483848394840484148424843484448454846484748484849485048514852485348544855485648574858485948604861486248634864486548664867486848694870487148724873487448754876487748784879488048814882488348844885488648874888488948904891489248934894489548964897489848994900490149024903490449054906490749084909491049114912491349144915491649174918491949204921492249234924492549264927492849294930493149324933493449354936493749384939494049414942494349444945494649474948494949504951495249534954495549564957495849594960496149624963496449654966496749684969497049714972497349744975497649774978497949804981498249834984498549864987498849894990499149924993499449954996499749984999500050015002500350045005500650075008500950105011501250135014501550165017501850195020502150225023502450255026502750285029503050315032503350345035503650375038503950405041504250435044504550465047504850495050505150525053505450555056505750585059506050615062506350645065506650675068506950705071507250735074507550765077507850795080508150825083508450855086508750885089509050915092509350945095509650975098509951005101510251035104510551065107510851095110511151125113511451155116511751185119512051215122512351245125512651275128512951305131513251335134513551365137513851395140514151425143514451455146514751485149515051515152515351545155515651575158515951605161516251635164516551665167516851695170517151725173517451755176517751785179518051815182518351845185518651875188518951905191519251935194519551965197519851995200520152025203520452055206520752085209521052115212521352145215521652175218521952205221522252235224522552265227522852295230523152325233523452355236523752385239524052415242524352445245524652475248524952505251525252535254525552565257525852595260526152625263526452655266526752685269527052715272527352745275527652775278527952805281528252835284528552865287528852895290529152925293529452955296529752985299530053015302530353045305530653075308530953105311531253135314531553165317531853195320532153225323532453255326532753285329533053315332533353345335533653375338533953405341534253435344534553465347534853495350535153525353535453555356535753585359536053615362536353645365536653675368536953705371537253735374537553765377537853795380538153825383538453855386538753885389539053915392539353945395539653975398539954005401540254035404540554065407540854095410541154125413541454155416541754185419542054215422542354245425542654275428542954305431543254335434543554365437543854395440544154425443544454455446544754485449545054515452545354545455545654575458545954605461546254635464546554665467546854695470547154725473547454755476547754785479548054815482
  1. /*
  2. * linux/fs/ext4/inode.c
  3. *
  4. * Copyright (C) 1992, 1993, 1994, 1995
  5. * Remy Card (card@masi.ibp.fr)
  6. * Laboratoire MASI - Institut Blaise Pascal
  7. * Universite Pierre et Marie Curie (Paris VI)
  8. *
  9. * from
  10. *
  11. * linux/fs/minix/inode.c
  12. *
  13. * Copyright (C) 1991, 1992 Linus Torvalds
  14. *
  15. * Goal-directed block allocation by Stephen Tweedie
  16. * (sct@redhat.com), 1993, 1998
  17. * Big-endian to little-endian byte-swapping/bitmaps by
  18. * David S. Miller (davem@caip.rutgers.edu), 1995
  19. * 64-bit file support on 64-bit platforms by Jakub Jelinek
  20. * (jj@sunsite.ms.mff.cuni.cz)
  21. *
  22. * Assorted race fixes, rewrite of ext4_get_block() by Al Viro, 2000
  23. */
  24. #include <linux/module.h>
  25. #include <linux/fs.h>
  26. #include <linux/time.h>
  27. #include <linux/jbd2.h>
  28. #include <linux/highuid.h>
  29. #include <linux/pagemap.h>
  30. #include <linux/quotaops.h>
  31. #include <linux/string.h>
  32. #include <linux/buffer_head.h>
  33. #include <linux/writeback.h>
  34. #include <linux/pagevec.h>
  35. #include <linux/mpage.h>
  36. #include <linux/namei.h>
  37. #include <linux/uio.h>
  38. #include <linux/bio.h>
  39. #include "ext4_jbd2.h"
  40. #include "xattr.h"
  41. #include "acl.h"
  42. #include "ext4_extents.h"
  43. #define MPAGE_DA_EXTENT_TAIL 0x01
  44. static inline int ext4_begin_ordered_truncate(struct inode *inode,
  45. loff_t new_size)
  46. {
  47. return jbd2_journal_begin_ordered_truncate(
  48. EXT4_SB(inode->i_sb)->s_journal,
  49. &EXT4_I(inode)->jinode,
  50. new_size);
  51. }
  52. static void ext4_invalidatepage(struct page *page, unsigned long offset);
  53. /*
  54. * Test whether an inode is a fast symlink.
  55. */
  56. static int ext4_inode_is_fast_symlink(struct inode *inode)
  57. {
  58. int ea_blocks = EXT4_I(inode)->i_file_acl ?
  59. (inode->i_sb->s_blocksize >> 9) : 0;
  60. return (S_ISLNK(inode->i_mode) && inode->i_blocks - ea_blocks == 0);
  61. }
  62. /*
  63. * The ext4 forget function must perform a revoke if we are freeing data
  64. * which has been journaled. Metadata (eg. indirect blocks) must be
  65. * revoked in all cases.
  66. *
  67. * "bh" may be NULL: a metadata block may have been freed from memory
  68. * but there may still be a record of it in the journal, and that record
  69. * still needs to be revoked.
  70. *
  71. * If the handle isn't valid we're not journaling so there's nothing to do.
  72. */
  73. int ext4_forget(handle_t *handle, int is_metadata, struct inode *inode,
  74. struct buffer_head *bh, ext4_fsblk_t blocknr)
  75. {
  76. int err;
  77. if (!ext4_handle_valid(handle))
  78. return 0;
  79. might_sleep();
  80. BUFFER_TRACE(bh, "enter");
  81. jbd_debug(4, "forgetting bh %p: is_metadata = %d, mode %o, "
  82. "data mode %lx\n",
  83. bh, is_metadata, inode->i_mode,
  84. test_opt(inode->i_sb, DATA_FLAGS));
  85. /* Never use the revoke function if we are doing full data
  86. * journaling: there is no need to, and a V1 superblock won't
  87. * support it. Otherwise, only skip the revoke on un-journaled
  88. * data blocks. */
  89. if (test_opt(inode->i_sb, DATA_FLAGS) == EXT4_MOUNT_JOURNAL_DATA ||
  90. (!is_metadata && !ext4_should_journal_data(inode))) {
  91. if (bh) {
  92. BUFFER_TRACE(bh, "call jbd2_journal_forget");
  93. return ext4_journal_forget(handle, bh);
  94. }
  95. return 0;
  96. }
  97. /*
  98. * data!=journal && (is_metadata || should_journal_data(inode))
  99. */
  100. BUFFER_TRACE(bh, "call ext4_journal_revoke");
  101. err = ext4_journal_revoke(handle, blocknr, bh);
  102. if (err)
  103. ext4_abort(inode->i_sb, __func__,
  104. "error %d when attempting revoke", err);
  105. BUFFER_TRACE(bh, "exit");
  106. return err;
  107. }
  108. /*
  109. * Work out how many blocks we need to proceed with the next chunk of a
  110. * truncate transaction.
  111. */
  112. static unsigned long blocks_for_truncate(struct inode *inode)
  113. {
  114. ext4_lblk_t needed;
  115. needed = inode->i_blocks >> (inode->i_sb->s_blocksize_bits - 9);
  116. /* Give ourselves just enough room to cope with inodes in which
  117. * i_blocks is corrupt: we've seen disk corruptions in the past
  118. * which resulted in random data in an inode which looked enough
  119. * like a regular file for ext4 to try to delete it. Things
  120. * will go a bit crazy if that happens, but at least we should
  121. * try not to panic the whole kernel. */
  122. if (needed < 2)
  123. needed = 2;
  124. /* But we need to bound the transaction so we don't overflow the
  125. * journal. */
  126. if (needed > EXT4_MAX_TRANS_DATA)
  127. needed = EXT4_MAX_TRANS_DATA;
  128. return EXT4_DATA_TRANS_BLOCKS(inode->i_sb) + needed;
  129. }
  130. /*
  131. * Truncate transactions can be complex and absolutely huge. So we need to
  132. * be able to restart the transaction at a conventient checkpoint to make
  133. * sure we don't overflow the journal.
  134. *
  135. * start_transaction gets us a new handle for a truncate transaction,
  136. * and extend_transaction tries to extend the existing one a bit. If
  137. * extend fails, we need to propagate the failure up and restart the
  138. * transaction in the top-level truncate loop. --sct
  139. */
  140. static handle_t *start_transaction(struct inode *inode)
  141. {
  142. handle_t *result;
  143. result = ext4_journal_start(inode, blocks_for_truncate(inode));
  144. if (!IS_ERR(result))
  145. return result;
  146. ext4_std_error(inode->i_sb, PTR_ERR(result));
  147. return result;
  148. }
  149. /*
  150. * Try to extend this transaction for the purposes of truncation.
  151. *
  152. * Returns 0 if we managed to create more room. If we can't create more
  153. * room, and the transaction must be restarted we return 1.
  154. */
  155. static int try_to_extend_transaction(handle_t *handle, struct inode *inode)
  156. {
  157. if (!ext4_handle_valid(handle))
  158. return 0;
  159. if (ext4_handle_has_enough_credits(handle, EXT4_RESERVE_TRANS_BLOCKS+1))
  160. return 0;
  161. if (!ext4_journal_extend(handle, blocks_for_truncate(inode)))
  162. return 0;
  163. return 1;
  164. }
  165. /*
  166. * Restart the transaction associated with *handle. This does a commit,
  167. * so before we call here everything must be consistently dirtied against
  168. * this transaction.
  169. */
  170. static int ext4_journal_test_restart(handle_t *handle, struct inode *inode)
  171. {
  172. BUG_ON(EXT4_JOURNAL(inode) == NULL);
  173. jbd_debug(2, "restarting handle %p\n", handle);
  174. return ext4_journal_restart(handle, blocks_for_truncate(inode));
  175. }
  176. /*
  177. * Called at the last iput() if i_nlink is zero.
  178. */
  179. void ext4_delete_inode(struct inode *inode)
  180. {
  181. handle_t *handle;
  182. int err;
  183. if (ext4_should_order_data(inode))
  184. ext4_begin_ordered_truncate(inode, 0);
  185. truncate_inode_pages(&inode->i_data, 0);
  186. if (is_bad_inode(inode))
  187. goto no_delete;
  188. handle = ext4_journal_start(inode, blocks_for_truncate(inode)+3);
  189. if (IS_ERR(handle)) {
  190. ext4_std_error(inode->i_sb, PTR_ERR(handle));
  191. /*
  192. * If we're going to skip the normal cleanup, we still need to
  193. * make sure that the in-core orphan linked list is properly
  194. * cleaned up.
  195. */
  196. ext4_orphan_del(NULL, inode);
  197. goto no_delete;
  198. }
  199. if (IS_SYNC(inode))
  200. ext4_handle_sync(handle);
  201. inode->i_size = 0;
  202. err = ext4_mark_inode_dirty(handle, inode);
  203. if (err) {
  204. ext4_warning(inode->i_sb, __func__,
  205. "couldn't mark inode dirty (err %d)", err);
  206. goto stop_handle;
  207. }
  208. if (inode->i_blocks)
  209. ext4_truncate(inode);
  210. /*
  211. * ext4_ext_truncate() doesn't reserve any slop when it
  212. * restarts journal transactions; therefore there may not be
  213. * enough credits left in the handle to remove the inode from
  214. * the orphan list and set the dtime field.
  215. */
  216. if (!ext4_handle_has_enough_credits(handle, 3)) {
  217. err = ext4_journal_extend(handle, 3);
  218. if (err > 0)
  219. err = ext4_journal_restart(handle, 3);
  220. if (err != 0) {
  221. ext4_warning(inode->i_sb, __func__,
  222. "couldn't extend journal (err %d)", err);
  223. stop_handle:
  224. ext4_journal_stop(handle);
  225. goto no_delete;
  226. }
  227. }
  228. /*
  229. * Kill off the orphan record which ext4_truncate created.
  230. * AKPM: I think this can be inside the above `if'.
  231. * Note that ext4_orphan_del() has to be able to cope with the
  232. * deletion of a non-existent orphan - this is because we don't
  233. * know if ext4_truncate() actually created an orphan record.
  234. * (Well, we could do this if we need to, but heck - it works)
  235. */
  236. ext4_orphan_del(handle, inode);
  237. EXT4_I(inode)->i_dtime = get_seconds();
  238. /*
  239. * One subtle ordering requirement: if anything has gone wrong
  240. * (transaction abort, IO errors, whatever), then we can still
  241. * do these next steps (the fs will already have been marked as
  242. * having errors), but we can't free the inode if the mark_dirty
  243. * fails.
  244. */
  245. if (ext4_mark_inode_dirty(handle, inode))
  246. /* If that failed, just do the required in-core inode clear. */
  247. clear_inode(inode);
  248. else
  249. ext4_free_inode(handle, inode);
  250. ext4_journal_stop(handle);
  251. return;
  252. no_delete:
  253. clear_inode(inode); /* We must guarantee clearing of inode... */
  254. }
  255. typedef struct {
  256. __le32 *p;
  257. __le32 key;
  258. struct buffer_head *bh;
  259. } Indirect;
  260. static inline void add_chain(Indirect *p, struct buffer_head *bh, __le32 *v)
  261. {
  262. p->key = *(p->p = v);
  263. p->bh = bh;
  264. }
  265. /**
  266. * ext4_block_to_path - parse the block number into array of offsets
  267. * @inode: inode in question (we are only interested in its superblock)
  268. * @i_block: block number to be parsed
  269. * @offsets: array to store the offsets in
  270. * @boundary: set this non-zero if the referred-to block is likely to be
  271. * followed (on disk) by an indirect block.
  272. *
  273. * To store the locations of file's data ext4 uses a data structure common
  274. * for UNIX filesystems - tree of pointers anchored in the inode, with
  275. * data blocks at leaves and indirect blocks in intermediate nodes.
  276. * This function translates the block number into path in that tree -
  277. * return value is the path length and @offsets[n] is the offset of
  278. * pointer to (n+1)th node in the nth one. If @block is out of range
  279. * (negative or too large) warning is printed and zero returned.
  280. *
  281. * Note: function doesn't find node addresses, so no IO is needed. All
  282. * we need to know is the capacity of indirect blocks (taken from the
  283. * inode->i_sb).
  284. */
  285. /*
  286. * Portability note: the last comparison (check that we fit into triple
  287. * indirect block) is spelled differently, because otherwise on an
  288. * architecture with 32-bit longs and 8Kb pages we might get into trouble
  289. * if our filesystem had 8Kb blocks. We might use long long, but that would
  290. * kill us on x86. Oh, well, at least the sign propagation does not matter -
  291. * i_block would have to be negative in the very beginning, so we would not
  292. * get there at all.
  293. */
  294. static int ext4_block_to_path(struct inode *inode,
  295. ext4_lblk_t i_block,
  296. ext4_lblk_t offsets[4], int *boundary)
  297. {
  298. int ptrs = EXT4_ADDR_PER_BLOCK(inode->i_sb);
  299. int ptrs_bits = EXT4_ADDR_PER_BLOCK_BITS(inode->i_sb);
  300. const long direct_blocks = EXT4_NDIR_BLOCKS,
  301. indirect_blocks = ptrs,
  302. double_blocks = (1 << (ptrs_bits * 2));
  303. int n = 0;
  304. int final = 0;
  305. if (i_block < 0) {
  306. ext4_warning(inode->i_sb, "ext4_block_to_path", "block < 0");
  307. } else if (i_block < direct_blocks) {
  308. offsets[n++] = i_block;
  309. final = direct_blocks;
  310. } else if ((i_block -= direct_blocks) < indirect_blocks) {
  311. offsets[n++] = EXT4_IND_BLOCK;
  312. offsets[n++] = i_block;
  313. final = ptrs;
  314. } else if ((i_block -= indirect_blocks) < double_blocks) {
  315. offsets[n++] = EXT4_DIND_BLOCK;
  316. offsets[n++] = i_block >> ptrs_bits;
  317. offsets[n++] = i_block & (ptrs - 1);
  318. final = ptrs;
  319. } else if (((i_block -= double_blocks) >> (ptrs_bits * 2)) < ptrs) {
  320. offsets[n++] = EXT4_TIND_BLOCK;
  321. offsets[n++] = i_block >> (ptrs_bits * 2);
  322. offsets[n++] = (i_block >> ptrs_bits) & (ptrs - 1);
  323. offsets[n++] = i_block & (ptrs - 1);
  324. final = ptrs;
  325. } else {
  326. ext4_warning(inode->i_sb, "ext4_block_to_path",
  327. "block %lu > max in inode %lu",
  328. i_block + direct_blocks +
  329. indirect_blocks + double_blocks, inode->i_ino);
  330. }
  331. if (boundary)
  332. *boundary = final - 1 - (i_block & (ptrs - 1));
  333. return n;
  334. }
  335. static int __ext4_check_blockref(const char *function, struct inode *inode,
  336. __le32 *p, unsigned int max)
  337. {
  338. __le32 *bref = p;
  339. unsigned int blk;
  340. while (bref < p+max) {
  341. blk = le32_to_cpu(*bref++);
  342. if (blk &&
  343. unlikely(!ext4_data_block_valid(EXT4_SB(inode->i_sb),
  344. blk, 1))) {
  345. ext4_error(inode->i_sb, function,
  346. "invalid block reference %u "
  347. "in inode #%lu", blk, inode->i_ino);
  348. return -EIO;
  349. }
  350. }
  351. return 0;
  352. }
  353. #define ext4_check_indirect_blockref(inode, bh) \
  354. __ext4_check_blockref(__func__, inode, (__le32 *)(bh)->b_data, \
  355. EXT4_ADDR_PER_BLOCK((inode)->i_sb))
  356. #define ext4_check_inode_blockref(inode) \
  357. __ext4_check_blockref(__func__, inode, EXT4_I(inode)->i_data, \
  358. EXT4_NDIR_BLOCKS)
  359. /**
  360. * ext4_get_branch - read the chain of indirect blocks leading to data
  361. * @inode: inode in question
  362. * @depth: depth of the chain (1 - direct pointer, etc.)
  363. * @offsets: offsets of pointers in inode/indirect blocks
  364. * @chain: place to store the result
  365. * @err: here we store the error value
  366. *
  367. * Function fills the array of triples <key, p, bh> and returns %NULL
  368. * if everything went OK or the pointer to the last filled triple
  369. * (incomplete one) otherwise. Upon the return chain[i].key contains
  370. * the number of (i+1)-th block in the chain (as it is stored in memory,
  371. * i.e. little-endian 32-bit), chain[i].p contains the address of that
  372. * number (it points into struct inode for i==0 and into the bh->b_data
  373. * for i>0) and chain[i].bh points to the buffer_head of i-th indirect
  374. * block for i>0 and NULL for i==0. In other words, it holds the block
  375. * numbers of the chain, addresses they were taken from (and where we can
  376. * verify that chain did not change) and buffer_heads hosting these
  377. * numbers.
  378. *
  379. * Function stops when it stumbles upon zero pointer (absent block)
  380. * (pointer to last triple returned, *@err == 0)
  381. * or when it gets an IO error reading an indirect block
  382. * (ditto, *@err == -EIO)
  383. * or when it reads all @depth-1 indirect blocks successfully and finds
  384. * the whole chain, all way to the data (returns %NULL, *err == 0).
  385. *
  386. * Need to be called with
  387. * down_read(&EXT4_I(inode)->i_data_sem)
  388. */
  389. static Indirect *ext4_get_branch(struct inode *inode, int depth,
  390. ext4_lblk_t *offsets,
  391. Indirect chain[4], int *err)
  392. {
  393. struct super_block *sb = inode->i_sb;
  394. Indirect *p = chain;
  395. struct buffer_head *bh;
  396. *err = 0;
  397. /* i_data is not going away, no lock needed */
  398. add_chain(chain, NULL, EXT4_I(inode)->i_data + *offsets);
  399. if (!p->key)
  400. goto no_block;
  401. while (--depth) {
  402. bh = sb_getblk(sb, le32_to_cpu(p->key));
  403. if (unlikely(!bh))
  404. goto failure;
  405. if (!bh_uptodate_or_lock(bh)) {
  406. if (bh_submit_read(bh) < 0) {
  407. put_bh(bh);
  408. goto failure;
  409. }
  410. /* validate block references */
  411. if (ext4_check_indirect_blockref(inode, bh)) {
  412. put_bh(bh);
  413. goto failure;
  414. }
  415. }
  416. add_chain(++p, bh, (__le32 *)bh->b_data + *++offsets);
  417. /* Reader: end */
  418. if (!p->key)
  419. goto no_block;
  420. }
  421. return NULL;
  422. failure:
  423. *err = -EIO;
  424. no_block:
  425. return p;
  426. }
  427. /**
  428. * ext4_find_near - find a place for allocation with sufficient locality
  429. * @inode: owner
  430. * @ind: descriptor of indirect block.
  431. *
  432. * This function returns the preferred place for block allocation.
  433. * It is used when heuristic for sequential allocation fails.
  434. * Rules are:
  435. * + if there is a block to the left of our position - allocate near it.
  436. * + if pointer will live in indirect block - allocate near that block.
  437. * + if pointer will live in inode - allocate in the same
  438. * cylinder group.
  439. *
  440. * In the latter case we colour the starting block by the callers PID to
  441. * prevent it from clashing with concurrent allocations for a different inode
  442. * in the same block group. The PID is used here so that functionally related
  443. * files will be close-by on-disk.
  444. *
  445. * Caller must make sure that @ind is valid and will stay that way.
  446. */
  447. static ext4_fsblk_t ext4_find_near(struct inode *inode, Indirect *ind)
  448. {
  449. struct ext4_inode_info *ei = EXT4_I(inode);
  450. __le32 *start = ind->bh ? (__le32 *) ind->bh->b_data : ei->i_data;
  451. __le32 *p;
  452. ext4_fsblk_t bg_start;
  453. ext4_fsblk_t last_block;
  454. ext4_grpblk_t colour;
  455. ext4_group_t block_group;
  456. int flex_size = ext4_flex_bg_size(EXT4_SB(inode->i_sb));
  457. /* Try to find previous block */
  458. for (p = ind->p - 1; p >= start; p--) {
  459. if (*p)
  460. return le32_to_cpu(*p);
  461. }
  462. /* No such thing, so let's try location of indirect block */
  463. if (ind->bh)
  464. return ind->bh->b_blocknr;
  465. /*
  466. * It is going to be referred to from the inode itself? OK, just put it
  467. * into the same cylinder group then.
  468. */
  469. block_group = ei->i_block_group;
  470. if (flex_size >= EXT4_FLEX_SIZE_DIR_ALLOC_SCHEME) {
  471. block_group &= ~(flex_size-1);
  472. if (S_ISREG(inode->i_mode))
  473. block_group++;
  474. }
  475. bg_start = ext4_group_first_block_no(inode->i_sb, block_group);
  476. last_block = ext4_blocks_count(EXT4_SB(inode->i_sb)->s_es) - 1;
  477. /*
  478. * If we are doing delayed allocation, we don't need take
  479. * colour into account.
  480. */
  481. if (test_opt(inode->i_sb, DELALLOC))
  482. return bg_start;
  483. if (bg_start + EXT4_BLOCKS_PER_GROUP(inode->i_sb) <= last_block)
  484. colour = (current->pid % 16) *
  485. (EXT4_BLOCKS_PER_GROUP(inode->i_sb) / 16);
  486. else
  487. colour = (current->pid % 16) * ((last_block - bg_start) / 16);
  488. return bg_start + colour;
  489. }
  490. /**
  491. * ext4_find_goal - find a preferred place for allocation.
  492. * @inode: owner
  493. * @block: block we want
  494. * @partial: pointer to the last triple within a chain
  495. *
  496. * Normally this function find the preferred place for block allocation,
  497. * returns it.
  498. */
  499. static ext4_fsblk_t ext4_find_goal(struct inode *inode, ext4_lblk_t block,
  500. Indirect *partial)
  501. {
  502. /*
  503. * XXX need to get goal block from mballoc's data structures
  504. */
  505. return ext4_find_near(inode, partial);
  506. }
  507. /**
  508. * ext4_blks_to_allocate: Look up the block map and count the number
  509. * of direct blocks need to be allocated for the given branch.
  510. *
  511. * @branch: chain of indirect blocks
  512. * @k: number of blocks need for indirect blocks
  513. * @blks: number of data blocks to be mapped.
  514. * @blocks_to_boundary: the offset in the indirect block
  515. *
  516. * return the total number of blocks to be allocate, including the
  517. * direct and indirect blocks.
  518. */
  519. static int ext4_blks_to_allocate(Indirect *branch, int k, unsigned int blks,
  520. int blocks_to_boundary)
  521. {
  522. unsigned int count = 0;
  523. /*
  524. * Simple case, [t,d]Indirect block(s) has not allocated yet
  525. * then it's clear blocks on that path have not allocated
  526. */
  527. if (k > 0) {
  528. /* right now we don't handle cross boundary allocation */
  529. if (blks < blocks_to_boundary + 1)
  530. count += blks;
  531. else
  532. count += blocks_to_boundary + 1;
  533. return count;
  534. }
  535. count++;
  536. while (count < blks && count <= blocks_to_boundary &&
  537. le32_to_cpu(*(branch[0].p + count)) == 0) {
  538. count++;
  539. }
  540. return count;
  541. }
  542. /**
  543. * ext4_alloc_blocks: multiple allocate blocks needed for a branch
  544. * @indirect_blks: the number of blocks need to allocate for indirect
  545. * blocks
  546. *
  547. * @new_blocks: on return it will store the new block numbers for
  548. * the indirect blocks(if needed) and the first direct block,
  549. * @blks: on return it will store the total number of allocated
  550. * direct blocks
  551. */
  552. static int ext4_alloc_blocks(handle_t *handle, struct inode *inode,
  553. ext4_lblk_t iblock, ext4_fsblk_t goal,
  554. int indirect_blks, int blks,
  555. ext4_fsblk_t new_blocks[4], int *err)
  556. {
  557. struct ext4_allocation_request ar;
  558. int target, i;
  559. unsigned long count = 0, blk_allocated = 0;
  560. int index = 0;
  561. ext4_fsblk_t current_block = 0;
  562. int ret = 0;
  563. /*
  564. * Here we try to allocate the requested multiple blocks at once,
  565. * on a best-effort basis.
  566. * To build a branch, we should allocate blocks for
  567. * the indirect blocks(if not allocated yet), and at least
  568. * the first direct block of this branch. That's the
  569. * minimum number of blocks need to allocate(required)
  570. */
  571. /* first we try to allocate the indirect blocks */
  572. target = indirect_blks;
  573. while (target > 0) {
  574. count = target;
  575. /* allocating blocks for indirect blocks and direct blocks */
  576. current_block = ext4_new_meta_blocks(handle, inode,
  577. goal, &count, err);
  578. if (*err)
  579. goto failed_out;
  580. target -= count;
  581. /* allocate blocks for indirect blocks */
  582. while (index < indirect_blks && count) {
  583. new_blocks[index++] = current_block++;
  584. count--;
  585. }
  586. if (count > 0) {
  587. /*
  588. * save the new block number
  589. * for the first direct block
  590. */
  591. new_blocks[index] = current_block;
  592. printk(KERN_INFO "%s returned more blocks than "
  593. "requested\n", __func__);
  594. WARN_ON(1);
  595. break;
  596. }
  597. }
  598. target = blks - count ;
  599. blk_allocated = count;
  600. if (!target)
  601. goto allocated;
  602. /* Now allocate data blocks */
  603. memset(&ar, 0, sizeof(ar));
  604. ar.inode = inode;
  605. ar.goal = goal;
  606. ar.len = target;
  607. ar.logical = iblock;
  608. if (S_ISREG(inode->i_mode))
  609. /* enable in-core preallocation only for regular files */
  610. ar.flags = EXT4_MB_HINT_DATA;
  611. current_block = ext4_mb_new_blocks(handle, &ar, err);
  612. if (*err && (target == blks)) {
  613. /*
  614. * if the allocation failed and we didn't allocate
  615. * any blocks before
  616. */
  617. goto failed_out;
  618. }
  619. if (!*err) {
  620. if (target == blks) {
  621. /*
  622. * save the new block number
  623. * for the first direct block
  624. */
  625. new_blocks[index] = current_block;
  626. }
  627. blk_allocated += ar.len;
  628. }
  629. allocated:
  630. /* total number of blocks allocated for direct blocks */
  631. ret = blk_allocated;
  632. *err = 0;
  633. return ret;
  634. failed_out:
  635. for (i = 0; i < index; i++)
  636. ext4_free_blocks(handle, inode, new_blocks[i], 1, 0);
  637. return ret;
  638. }
  639. /**
  640. * ext4_alloc_branch - allocate and set up a chain of blocks.
  641. * @inode: owner
  642. * @indirect_blks: number of allocated indirect blocks
  643. * @blks: number of allocated direct blocks
  644. * @offsets: offsets (in the blocks) to store the pointers to next.
  645. * @branch: place to store the chain in.
  646. *
  647. * This function allocates blocks, zeroes out all but the last one,
  648. * links them into chain and (if we are synchronous) writes them to disk.
  649. * In other words, it prepares a branch that can be spliced onto the
  650. * inode. It stores the information about that chain in the branch[], in
  651. * the same format as ext4_get_branch() would do. We are calling it after
  652. * we had read the existing part of chain and partial points to the last
  653. * triple of that (one with zero ->key). Upon the exit we have the same
  654. * picture as after the successful ext4_get_block(), except that in one
  655. * place chain is disconnected - *branch->p is still zero (we did not
  656. * set the last link), but branch->key contains the number that should
  657. * be placed into *branch->p to fill that gap.
  658. *
  659. * If allocation fails we free all blocks we've allocated (and forget
  660. * their buffer_heads) and return the error value the from failed
  661. * ext4_alloc_block() (normally -ENOSPC). Otherwise we set the chain
  662. * as described above and return 0.
  663. */
  664. static int ext4_alloc_branch(handle_t *handle, struct inode *inode,
  665. ext4_lblk_t iblock, int indirect_blks,
  666. int *blks, ext4_fsblk_t goal,
  667. ext4_lblk_t *offsets, Indirect *branch)
  668. {
  669. int blocksize = inode->i_sb->s_blocksize;
  670. int i, n = 0;
  671. int err = 0;
  672. struct buffer_head *bh;
  673. int num;
  674. ext4_fsblk_t new_blocks[4];
  675. ext4_fsblk_t current_block;
  676. num = ext4_alloc_blocks(handle, inode, iblock, goal, indirect_blks,
  677. *blks, new_blocks, &err);
  678. if (err)
  679. return err;
  680. branch[0].key = cpu_to_le32(new_blocks[0]);
  681. /*
  682. * metadata blocks and data blocks are allocated.
  683. */
  684. for (n = 1; n <= indirect_blks; n++) {
  685. /*
  686. * Get buffer_head for parent block, zero it out
  687. * and set the pointer to new one, then send
  688. * parent to disk.
  689. */
  690. bh = sb_getblk(inode->i_sb, new_blocks[n-1]);
  691. branch[n].bh = bh;
  692. lock_buffer(bh);
  693. BUFFER_TRACE(bh, "call get_create_access");
  694. err = ext4_journal_get_create_access(handle, bh);
  695. if (err) {
  696. unlock_buffer(bh);
  697. brelse(bh);
  698. goto failed;
  699. }
  700. memset(bh->b_data, 0, blocksize);
  701. branch[n].p = (__le32 *) bh->b_data + offsets[n];
  702. branch[n].key = cpu_to_le32(new_blocks[n]);
  703. *branch[n].p = branch[n].key;
  704. if (n == indirect_blks) {
  705. current_block = new_blocks[n];
  706. /*
  707. * End of chain, update the last new metablock of
  708. * the chain to point to the new allocated
  709. * data blocks numbers
  710. */
  711. for (i=1; i < num; i++)
  712. *(branch[n].p + i) = cpu_to_le32(++current_block);
  713. }
  714. BUFFER_TRACE(bh, "marking uptodate");
  715. set_buffer_uptodate(bh);
  716. unlock_buffer(bh);
  717. BUFFER_TRACE(bh, "call ext4_handle_dirty_metadata");
  718. err = ext4_handle_dirty_metadata(handle, inode, bh);
  719. if (err)
  720. goto failed;
  721. }
  722. *blks = num;
  723. return err;
  724. failed:
  725. /* Allocation failed, free what we already allocated */
  726. for (i = 1; i <= n ; i++) {
  727. BUFFER_TRACE(branch[i].bh, "call jbd2_journal_forget");
  728. ext4_journal_forget(handle, branch[i].bh);
  729. }
  730. for (i = 0; i < indirect_blks; i++)
  731. ext4_free_blocks(handle, inode, new_blocks[i], 1, 0);
  732. ext4_free_blocks(handle, inode, new_blocks[i], num, 0);
  733. return err;
  734. }
  735. /**
  736. * ext4_splice_branch - splice the allocated branch onto inode.
  737. * @inode: owner
  738. * @block: (logical) number of block we are adding
  739. * @chain: chain of indirect blocks (with a missing link - see
  740. * ext4_alloc_branch)
  741. * @where: location of missing link
  742. * @num: number of indirect blocks we are adding
  743. * @blks: number of direct blocks we are adding
  744. *
  745. * This function fills the missing link and does all housekeeping needed in
  746. * inode (->i_blocks, etc.). In case of success we end up with the full
  747. * chain to new block and return 0.
  748. */
  749. static int ext4_splice_branch(handle_t *handle, struct inode *inode,
  750. ext4_lblk_t block, Indirect *where, int num, int blks)
  751. {
  752. int i;
  753. int err = 0;
  754. ext4_fsblk_t current_block;
  755. /*
  756. * If we're splicing into a [td]indirect block (as opposed to the
  757. * inode) then we need to get write access to the [td]indirect block
  758. * before the splice.
  759. */
  760. if (where->bh) {
  761. BUFFER_TRACE(where->bh, "get_write_access");
  762. err = ext4_journal_get_write_access(handle, where->bh);
  763. if (err)
  764. goto err_out;
  765. }
  766. /* That's it */
  767. *where->p = where->key;
  768. /*
  769. * Update the host buffer_head or inode to point to more just allocated
  770. * direct blocks blocks
  771. */
  772. if (num == 0 && blks > 1) {
  773. current_block = le32_to_cpu(where->key) + 1;
  774. for (i = 1; i < blks; i++)
  775. *(where->p + i) = cpu_to_le32(current_block++);
  776. }
  777. /* We are done with atomic stuff, now do the rest of housekeeping */
  778. inode->i_ctime = ext4_current_time(inode);
  779. ext4_mark_inode_dirty(handle, inode);
  780. /* had we spliced it onto indirect block? */
  781. if (where->bh) {
  782. /*
  783. * If we spliced it onto an indirect block, we haven't
  784. * altered the inode. Note however that if it is being spliced
  785. * onto an indirect block at the very end of the file (the
  786. * file is growing) then we *will* alter the inode to reflect
  787. * the new i_size. But that is not done here - it is done in
  788. * generic_commit_write->__mark_inode_dirty->ext4_dirty_inode.
  789. */
  790. jbd_debug(5, "splicing indirect only\n");
  791. BUFFER_TRACE(where->bh, "call ext4_handle_dirty_metadata");
  792. err = ext4_handle_dirty_metadata(handle, inode, where->bh);
  793. if (err)
  794. goto err_out;
  795. } else {
  796. /*
  797. * OK, we spliced it into the inode itself on a direct block.
  798. * Inode was dirtied above.
  799. */
  800. jbd_debug(5, "splicing direct\n");
  801. }
  802. return err;
  803. err_out:
  804. for (i = 1; i <= num; i++) {
  805. BUFFER_TRACE(where[i].bh, "call jbd2_journal_forget");
  806. ext4_journal_forget(handle, where[i].bh);
  807. ext4_free_blocks(handle, inode,
  808. le32_to_cpu(where[i-1].key), 1, 0);
  809. }
  810. ext4_free_blocks(handle, inode, le32_to_cpu(where[num].key), blks, 0);
  811. return err;
  812. }
  813. /*
  814. * The ext4_ind_get_blocks() function handles non-extents inodes
  815. * (i.e., using the traditional indirect/double-indirect i_blocks
  816. * scheme) for ext4_get_blocks().
  817. *
  818. * Allocation strategy is simple: if we have to allocate something, we will
  819. * have to go the whole way to leaf. So let's do it before attaching anything
  820. * to tree, set linkage between the newborn blocks, write them if sync is
  821. * required, recheck the path, free and repeat if check fails, otherwise
  822. * set the last missing link (that will protect us from any truncate-generated
  823. * removals - all blocks on the path are immune now) and possibly force the
  824. * write on the parent block.
  825. * That has a nice additional property: no special recovery from the failed
  826. * allocations is needed - we simply release blocks and do not touch anything
  827. * reachable from inode.
  828. *
  829. * `handle' can be NULL if create == 0.
  830. *
  831. * return > 0, # of blocks mapped or allocated.
  832. * return = 0, if plain lookup failed.
  833. * return < 0, error case.
  834. *
  835. * The ext4_ind_get_blocks() function should be called with
  836. * down_write(&EXT4_I(inode)->i_data_sem) if allocating filesystem
  837. * blocks (i.e., flags has EXT4_GET_BLOCKS_CREATE set) or
  838. * down_read(&EXT4_I(inode)->i_data_sem) if not allocating file system
  839. * blocks.
  840. */
  841. static int ext4_ind_get_blocks(handle_t *handle, struct inode *inode,
  842. ext4_lblk_t iblock, unsigned int maxblocks,
  843. struct buffer_head *bh_result,
  844. int flags)
  845. {
  846. int err = -EIO;
  847. ext4_lblk_t offsets[4];
  848. Indirect chain[4];
  849. Indirect *partial;
  850. ext4_fsblk_t goal;
  851. int indirect_blks;
  852. int blocks_to_boundary = 0;
  853. int depth;
  854. int count = 0;
  855. ext4_fsblk_t first_block = 0;
  856. J_ASSERT(!(EXT4_I(inode)->i_flags & EXT4_EXTENTS_FL));
  857. J_ASSERT(handle != NULL || (flags & EXT4_GET_BLOCKS_CREATE) == 0);
  858. depth = ext4_block_to_path(inode, iblock, offsets,
  859. &blocks_to_boundary);
  860. if (depth == 0)
  861. goto out;
  862. partial = ext4_get_branch(inode, depth, offsets, chain, &err);
  863. /* Simplest case - block found, no allocation needed */
  864. if (!partial) {
  865. first_block = le32_to_cpu(chain[depth - 1].key);
  866. clear_buffer_new(bh_result);
  867. count++;
  868. /*map more blocks*/
  869. while (count < maxblocks && count <= blocks_to_boundary) {
  870. ext4_fsblk_t blk;
  871. blk = le32_to_cpu(*(chain[depth-1].p + count));
  872. if (blk == first_block + count)
  873. count++;
  874. else
  875. break;
  876. }
  877. goto got_it;
  878. }
  879. /* Next simple case - plain lookup or failed read of indirect block */
  880. if ((flags & EXT4_GET_BLOCKS_CREATE) == 0 || err == -EIO)
  881. goto cleanup;
  882. /*
  883. * Okay, we need to do block allocation.
  884. */
  885. goal = ext4_find_goal(inode, iblock, partial);
  886. /* the number of blocks need to allocate for [d,t]indirect blocks */
  887. indirect_blks = (chain + depth) - partial - 1;
  888. /*
  889. * Next look up the indirect map to count the totoal number of
  890. * direct blocks to allocate for this branch.
  891. */
  892. count = ext4_blks_to_allocate(partial, indirect_blks,
  893. maxblocks, blocks_to_boundary);
  894. /*
  895. * Block out ext4_truncate while we alter the tree
  896. */
  897. err = ext4_alloc_branch(handle, inode, iblock, indirect_blks,
  898. &count, goal,
  899. offsets + (partial - chain), partial);
  900. /*
  901. * The ext4_splice_branch call will free and forget any buffers
  902. * on the new chain if there is a failure, but that risks using
  903. * up transaction credits, especially for bitmaps where the
  904. * credits cannot be returned. Can we handle this somehow? We
  905. * may need to return -EAGAIN upwards in the worst case. --sct
  906. */
  907. if (!err)
  908. err = ext4_splice_branch(handle, inode, iblock,
  909. partial, indirect_blks, count);
  910. else
  911. goto cleanup;
  912. set_buffer_new(bh_result);
  913. got_it:
  914. map_bh(bh_result, inode->i_sb, le32_to_cpu(chain[depth-1].key));
  915. if (count > blocks_to_boundary)
  916. set_buffer_boundary(bh_result);
  917. err = count;
  918. /* Clean up and exit */
  919. partial = chain + depth - 1; /* the whole chain */
  920. cleanup:
  921. while (partial > chain) {
  922. BUFFER_TRACE(partial->bh, "call brelse");
  923. brelse(partial->bh);
  924. partial--;
  925. }
  926. BUFFER_TRACE(bh_result, "returned");
  927. out:
  928. return err;
  929. }
  930. qsize_t ext4_get_reserved_space(struct inode *inode)
  931. {
  932. unsigned long long total;
  933. spin_lock(&EXT4_I(inode)->i_block_reservation_lock);
  934. total = EXT4_I(inode)->i_reserved_data_blocks +
  935. EXT4_I(inode)->i_reserved_meta_blocks;
  936. spin_unlock(&EXT4_I(inode)->i_block_reservation_lock);
  937. return total;
  938. }
  939. /*
  940. * Calculate the number of metadata blocks need to reserve
  941. * to allocate @blocks for non extent file based file
  942. */
  943. static int ext4_indirect_calc_metadata_amount(struct inode *inode, int blocks)
  944. {
  945. int icap = EXT4_ADDR_PER_BLOCK(inode->i_sb);
  946. int ind_blks, dind_blks, tind_blks;
  947. /* number of new indirect blocks needed */
  948. ind_blks = (blocks + icap - 1) / icap;
  949. dind_blks = (ind_blks + icap - 1) / icap;
  950. tind_blks = 1;
  951. return ind_blks + dind_blks + tind_blks;
  952. }
  953. /*
  954. * Calculate the number of metadata blocks need to reserve
  955. * to allocate given number of blocks
  956. */
  957. static int ext4_calc_metadata_amount(struct inode *inode, int blocks)
  958. {
  959. if (!blocks)
  960. return 0;
  961. if (EXT4_I(inode)->i_flags & EXT4_EXTENTS_FL)
  962. return ext4_ext_calc_metadata_amount(inode, blocks);
  963. return ext4_indirect_calc_metadata_amount(inode, blocks);
  964. }
  965. static void ext4_da_update_reserve_space(struct inode *inode, int used)
  966. {
  967. struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
  968. int total, mdb, mdb_free;
  969. spin_lock(&EXT4_I(inode)->i_block_reservation_lock);
  970. /* recalculate the number of metablocks still need to be reserved */
  971. total = EXT4_I(inode)->i_reserved_data_blocks - used;
  972. mdb = ext4_calc_metadata_amount(inode, total);
  973. /* figure out how many metablocks to release */
  974. BUG_ON(mdb > EXT4_I(inode)->i_reserved_meta_blocks);
  975. mdb_free = EXT4_I(inode)->i_reserved_meta_blocks - mdb;
  976. if (mdb_free) {
  977. /* Account for allocated meta_blocks */
  978. mdb_free -= EXT4_I(inode)->i_allocated_meta_blocks;
  979. /* update fs dirty blocks counter */
  980. percpu_counter_sub(&sbi->s_dirtyblocks_counter, mdb_free);
  981. EXT4_I(inode)->i_allocated_meta_blocks = 0;
  982. EXT4_I(inode)->i_reserved_meta_blocks = mdb;
  983. }
  984. /* update per-inode reservations */
  985. BUG_ON(used > EXT4_I(inode)->i_reserved_data_blocks);
  986. EXT4_I(inode)->i_reserved_data_blocks -= used;
  987. spin_unlock(&EXT4_I(inode)->i_block_reservation_lock);
  988. /*
  989. * free those over-booking quota for metadata blocks
  990. */
  991. if (mdb_free)
  992. vfs_dq_release_reservation_block(inode, mdb_free);
  993. /*
  994. * If we have done all the pending block allocations and if
  995. * there aren't any writers on the inode, we can discard the
  996. * inode's preallocations.
  997. */
  998. if (!total && (atomic_read(&inode->i_writecount) == 0))
  999. ext4_discard_preallocations(inode);
  1000. }
  1001. static int check_block_validity(struct inode *inode, sector_t logical,
  1002. sector_t phys, int len)
  1003. {
  1004. if (!ext4_data_block_valid(EXT4_SB(inode->i_sb), phys, len)) {
  1005. ext4_error(inode->i_sb, "check_block_validity",
  1006. "inode #%lu logical block %llu mapped to %llu "
  1007. "(size %d)", inode->i_ino,
  1008. (unsigned long long) logical,
  1009. (unsigned long long) phys, len);
  1010. WARN_ON(1);
  1011. return -EIO;
  1012. }
  1013. return 0;
  1014. }
  1015. /*
  1016. * The ext4_get_blocks() function tries to look up the requested blocks,
  1017. * and returns if the blocks are already mapped.
  1018. *
  1019. * Otherwise it takes the write lock of the i_data_sem and allocate blocks
  1020. * and store the allocated blocks in the result buffer head and mark it
  1021. * mapped.
  1022. *
  1023. * If file type is extents based, it will call ext4_ext_get_blocks(),
  1024. * Otherwise, call with ext4_ind_get_blocks() to handle indirect mapping
  1025. * based files
  1026. *
  1027. * On success, it returns the number of blocks being mapped or allocate.
  1028. * if create==0 and the blocks are pre-allocated and uninitialized block,
  1029. * the result buffer head is unmapped. If the create ==1, it will make sure
  1030. * the buffer head is mapped.
  1031. *
  1032. * It returns 0 if plain look up failed (blocks have not been allocated), in
  1033. * that casem, buffer head is unmapped
  1034. *
  1035. * It returns the error in case of allocation failure.
  1036. */
  1037. int ext4_get_blocks(handle_t *handle, struct inode *inode, sector_t block,
  1038. unsigned int max_blocks, struct buffer_head *bh,
  1039. int flags)
  1040. {
  1041. int retval;
  1042. clear_buffer_mapped(bh);
  1043. clear_buffer_unwritten(bh);
  1044. /*
  1045. * Try to see if we can get the block without requesting a new
  1046. * file system block.
  1047. */
  1048. down_read((&EXT4_I(inode)->i_data_sem));
  1049. if (EXT4_I(inode)->i_flags & EXT4_EXTENTS_FL) {
  1050. retval = ext4_ext_get_blocks(handle, inode, block, max_blocks,
  1051. bh, 0);
  1052. } else {
  1053. retval = ext4_ind_get_blocks(handle, inode, block, max_blocks,
  1054. bh, 0);
  1055. }
  1056. up_read((&EXT4_I(inode)->i_data_sem));
  1057. if (retval > 0 && buffer_mapped(bh)) {
  1058. int ret = check_block_validity(inode, block,
  1059. bh->b_blocknr, retval);
  1060. if (ret != 0)
  1061. return ret;
  1062. }
  1063. /* If it is only a block(s) look up */
  1064. if ((flags & EXT4_GET_BLOCKS_CREATE) == 0)
  1065. return retval;
  1066. /*
  1067. * Returns if the blocks have already allocated
  1068. *
  1069. * Note that if blocks have been preallocated
  1070. * ext4_ext_get_block() returns th create = 0
  1071. * with buffer head unmapped.
  1072. */
  1073. if (retval > 0 && buffer_mapped(bh))
  1074. return retval;
  1075. /*
  1076. * When we call get_blocks without the create flag, the
  1077. * BH_Unwritten flag could have gotten set if the blocks
  1078. * requested were part of a uninitialized extent. We need to
  1079. * clear this flag now that we are committed to convert all or
  1080. * part of the uninitialized extent to be an initialized
  1081. * extent. This is because we need to avoid the combination
  1082. * of BH_Unwritten and BH_Mapped flags being simultaneously
  1083. * set on the buffer_head.
  1084. */
  1085. clear_buffer_unwritten(bh);
  1086. /*
  1087. * New blocks allocate and/or writing to uninitialized extent
  1088. * will possibly result in updating i_data, so we take
  1089. * the write lock of i_data_sem, and call get_blocks()
  1090. * with create == 1 flag.
  1091. */
  1092. down_write((&EXT4_I(inode)->i_data_sem));
  1093. /*
  1094. * if the caller is from delayed allocation writeout path
  1095. * we have already reserved fs blocks for allocation
  1096. * let the underlying get_block() function know to
  1097. * avoid double accounting
  1098. */
  1099. if (flags & EXT4_GET_BLOCKS_DELALLOC_RESERVE)
  1100. EXT4_I(inode)->i_delalloc_reserved_flag = 1;
  1101. /*
  1102. * We need to check for EXT4 here because migrate
  1103. * could have changed the inode type in between
  1104. */
  1105. if (EXT4_I(inode)->i_flags & EXT4_EXTENTS_FL) {
  1106. retval = ext4_ext_get_blocks(handle, inode, block, max_blocks,
  1107. bh, flags);
  1108. } else {
  1109. retval = ext4_ind_get_blocks(handle, inode, block,
  1110. max_blocks, bh, flags);
  1111. if (retval > 0 && buffer_new(bh)) {
  1112. /*
  1113. * We allocated new blocks which will result in
  1114. * i_data's format changing. Force the migrate
  1115. * to fail by clearing migrate flags
  1116. */
  1117. EXT4_I(inode)->i_flags = EXT4_I(inode)->i_flags &
  1118. ~EXT4_EXT_MIGRATE;
  1119. }
  1120. }
  1121. if (flags & EXT4_GET_BLOCKS_DELALLOC_RESERVE)
  1122. EXT4_I(inode)->i_delalloc_reserved_flag = 0;
  1123. /*
  1124. * Update reserved blocks/metadata blocks after successful
  1125. * block allocation which had been deferred till now.
  1126. */
  1127. if ((retval > 0) && (flags & EXT4_GET_BLOCKS_UPDATE_RESERVE_SPACE))
  1128. ext4_da_update_reserve_space(inode, retval);
  1129. up_write((&EXT4_I(inode)->i_data_sem));
  1130. if (retval > 0 && buffer_mapped(bh)) {
  1131. int ret = check_block_validity(inode, block,
  1132. bh->b_blocknr, retval);
  1133. if (ret != 0)
  1134. return ret;
  1135. }
  1136. return retval;
  1137. }
  1138. /* Maximum number of blocks we map for direct IO at once. */
  1139. #define DIO_MAX_BLOCKS 4096
  1140. int ext4_get_block(struct inode *inode, sector_t iblock,
  1141. struct buffer_head *bh_result, int create)
  1142. {
  1143. handle_t *handle = ext4_journal_current_handle();
  1144. int ret = 0, started = 0;
  1145. unsigned max_blocks = bh_result->b_size >> inode->i_blkbits;
  1146. int dio_credits;
  1147. if (create && !handle) {
  1148. /* Direct IO write... */
  1149. if (max_blocks > DIO_MAX_BLOCKS)
  1150. max_blocks = DIO_MAX_BLOCKS;
  1151. dio_credits = ext4_chunk_trans_blocks(inode, max_blocks);
  1152. handle = ext4_journal_start(inode, dio_credits);
  1153. if (IS_ERR(handle)) {
  1154. ret = PTR_ERR(handle);
  1155. goto out;
  1156. }
  1157. started = 1;
  1158. }
  1159. ret = ext4_get_blocks(handle, inode, iblock, max_blocks, bh_result,
  1160. create ? EXT4_GET_BLOCKS_CREATE : 0);
  1161. if (ret > 0) {
  1162. bh_result->b_size = (ret << inode->i_blkbits);
  1163. ret = 0;
  1164. }
  1165. if (started)
  1166. ext4_journal_stop(handle);
  1167. out:
  1168. return ret;
  1169. }
  1170. /*
  1171. * `handle' can be NULL if create is zero
  1172. */
  1173. struct buffer_head *ext4_getblk(handle_t *handle, struct inode *inode,
  1174. ext4_lblk_t block, int create, int *errp)
  1175. {
  1176. struct buffer_head dummy;
  1177. int fatal = 0, err;
  1178. int flags = 0;
  1179. J_ASSERT(handle != NULL || create == 0);
  1180. dummy.b_state = 0;
  1181. dummy.b_blocknr = -1000;
  1182. buffer_trace_init(&dummy.b_history);
  1183. if (create)
  1184. flags |= EXT4_GET_BLOCKS_CREATE;
  1185. err = ext4_get_blocks(handle, inode, block, 1, &dummy, flags);
  1186. /*
  1187. * ext4_get_blocks() returns number of blocks mapped. 0 in
  1188. * case of a HOLE.
  1189. */
  1190. if (err > 0) {
  1191. if (err > 1)
  1192. WARN_ON(1);
  1193. err = 0;
  1194. }
  1195. *errp = err;
  1196. if (!err && buffer_mapped(&dummy)) {
  1197. struct buffer_head *bh;
  1198. bh = sb_getblk(inode->i_sb, dummy.b_blocknr);
  1199. if (!bh) {
  1200. *errp = -EIO;
  1201. goto err;
  1202. }
  1203. if (buffer_new(&dummy)) {
  1204. J_ASSERT(create != 0);
  1205. J_ASSERT(handle != NULL);
  1206. /*
  1207. * Now that we do not always journal data, we should
  1208. * keep in mind whether this should always journal the
  1209. * new buffer as metadata. For now, regular file
  1210. * writes use ext4_get_block instead, so it's not a
  1211. * problem.
  1212. */
  1213. lock_buffer(bh);
  1214. BUFFER_TRACE(bh, "call get_create_access");
  1215. fatal = ext4_journal_get_create_access(handle, bh);
  1216. if (!fatal && !buffer_uptodate(bh)) {
  1217. memset(bh->b_data, 0, inode->i_sb->s_blocksize);
  1218. set_buffer_uptodate(bh);
  1219. }
  1220. unlock_buffer(bh);
  1221. BUFFER_TRACE(bh, "call ext4_handle_dirty_metadata");
  1222. err = ext4_handle_dirty_metadata(handle, inode, bh);
  1223. if (!fatal)
  1224. fatal = err;
  1225. } else {
  1226. BUFFER_TRACE(bh, "not a new buffer");
  1227. }
  1228. if (fatal) {
  1229. *errp = fatal;
  1230. brelse(bh);
  1231. bh = NULL;
  1232. }
  1233. return bh;
  1234. }
  1235. err:
  1236. return NULL;
  1237. }
  1238. struct buffer_head *ext4_bread(handle_t *handle, struct inode *inode,
  1239. ext4_lblk_t block, int create, int *err)
  1240. {
  1241. struct buffer_head *bh;
  1242. bh = ext4_getblk(handle, inode, block, create, err);
  1243. if (!bh)
  1244. return bh;
  1245. if (buffer_uptodate(bh))
  1246. return bh;
  1247. ll_rw_block(READ_META, 1, &bh);
  1248. wait_on_buffer(bh);
  1249. if (buffer_uptodate(bh))
  1250. return bh;
  1251. put_bh(bh);
  1252. *err = -EIO;
  1253. return NULL;
  1254. }
  1255. static int walk_page_buffers(handle_t *handle,
  1256. struct buffer_head *head,
  1257. unsigned from,
  1258. unsigned to,
  1259. int *partial,
  1260. int (*fn)(handle_t *handle,
  1261. struct buffer_head *bh))
  1262. {
  1263. struct buffer_head *bh;
  1264. unsigned block_start, block_end;
  1265. unsigned blocksize = head->b_size;
  1266. int err, ret = 0;
  1267. struct buffer_head *next;
  1268. for (bh = head, block_start = 0;
  1269. ret == 0 && (bh != head || !block_start);
  1270. block_start = block_end, bh = next)
  1271. {
  1272. next = bh->b_this_page;
  1273. block_end = block_start + blocksize;
  1274. if (block_end <= from || block_start >= to) {
  1275. if (partial && !buffer_uptodate(bh))
  1276. *partial = 1;
  1277. continue;
  1278. }
  1279. err = (*fn)(handle, bh);
  1280. if (!ret)
  1281. ret = err;
  1282. }
  1283. return ret;
  1284. }
  1285. /*
  1286. * To preserve ordering, it is essential that the hole instantiation and
  1287. * the data write be encapsulated in a single transaction. We cannot
  1288. * close off a transaction and start a new one between the ext4_get_block()
  1289. * and the commit_write(). So doing the jbd2_journal_start at the start of
  1290. * prepare_write() is the right place.
  1291. *
  1292. * Also, this function can nest inside ext4_writepage() ->
  1293. * block_write_full_page(). In that case, we *know* that ext4_writepage()
  1294. * has generated enough buffer credits to do the whole page. So we won't
  1295. * block on the journal in that case, which is good, because the caller may
  1296. * be PF_MEMALLOC.
  1297. *
  1298. * By accident, ext4 can be reentered when a transaction is open via
  1299. * quota file writes. If we were to commit the transaction while thus
  1300. * reentered, there can be a deadlock - we would be holding a quota
  1301. * lock, and the commit would never complete if another thread had a
  1302. * transaction open and was blocking on the quota lock - a ranking
  1303. * violation.
  1304. *
  1305. * So what we do is to rely on the fact that jbd2_journal_stop/journal_start
  1306. * will _not_ run commit under these circumstances because handle->h_ref
  1307. * is elevated. We'll still have enough credits for the tiny quotafile
  1308. * write.
  1309. */
  1310. static int do_journal_get_write_access(handle_t *handle,
  1311. struct buffer_head *bh)
  1312. {
  1313. if (!buffer_mapped(bh) || buffer_freed(bh))
  1314. return 0;
  1315. return ext4_journal_get_write_access(handle, bh);
  1316. }
  1317. static int ext4_write_begin(struct file *file, struct address_space *mapping,
  1318. loff_t pos, unsigned len, unsigned flags,
  1319. struct page **pagep, void **fsdata)
  1320. {
  1321. struct inode *inode = mapping->host;
  1322. int ret, needed_blocks;
  1323. handle_t *handle;
  1324. int retries = 0;
  1325. struct page *page;
  1326. pgoff_t index;
  1327. unsigned from, to;
  1328. trace_mark(ext4_write_begin,
  1329. "dev %s ino %lu pos %llu len %u flags %u",
  1330. inode->i_sb->s_id, inode->i_ino,
  1331. (unsigned long long) pos, len, flags);
  1332. /*
  1333. * Reserve one block more for addition to orphan list in case
  1334. * we allocate blocks but write fails for some reason
  1335. */
  1336. needed_blocks = ext4_writepage_trans_blocks(inode) + 1;
  1337. index = pos >> PAGE_CACHE_SHIFT;
  1338. from = pos & (PAGE_CACHE_SIZE - 1);
  1339. to = from + len;
  1340. retry:
  1341. handle = ext4_journal_start(inode, needed_blocks);
  1342. if (IS_ERR(handle)) {
  1343. ret = PTR_ERR(handle);
  1344. goto out;
  1345. }
  1346. /* We cannot recurse into the filesystem as the transaction is already
  1347. * started */
  1348. flags |= AOP_FLAG_NOFS;
  1349. page = grab_cache_page_write_begin(mapping, index, flags);
  1350. if (!page) {
  1351. ext4_journal_stop(handle);
  1352. ret = -ENOMEM;
  1353. goto out;
  1354. }
  1355. *pagep = page;
  1356. ret = block_write_begin(file, mapping, pos, len, flags, pagep, fsdata,
  1357. ext4_get_block);
  1358. if (!ret && ext4_should_journal_data(inode)) {
  1359. ret = walk_page_buffers(handle, page_buffers(page),
  1360. from, to, NULL, do_journal_get_write_access);
  1361. }
  1362. if (ret) {
  1363. unlock_page(page);
  1364. page_cache_release(page);
  1365. /*
  1366. * block_write_begin may have instantiated a few blocks
  1367. * outside i_size. Trim these off again. Don't need
  1368. * i_size_read because we hold i_mutex.
  1369. *
  1370. * Add inode to orphan list in case we crash before
  1371. * truncate finishes
  1372. */
  1373. if (pos + len > inode->i_size)
  1374. ext4_orphan_add(handle, inode);
  1375. ext4_journal_stop(handle);
  1376. if (pos + len > inode->i_size) {
  1377. vmtruncate(inode, inode->i_size);
  1378. /*
  1379. * If vmtruncate failed early the inode might
  1380. * still be on the orphan list; we need to
  1381. * make sure the inode is removed from the
  1382. * orphan list in that case.
  1383. */
  1384. if (inode->i_nlink)
  1385. ext4_orphan_del(NULL, inode);
  1386. }
  1387. }
  1388. if (ret == -ENOSPC && ext4_should_retry_alloc(inode->i_sb, &retries))
  1389. goto retry;
  1390. out:
  1391. return ret;
  1392. }
  1393. /* For write_end() in data=journal mode */
  1394. static int write_end_fn(handle_t *handle, struct buffer_head *bh)
  1395. {
  1396. if (!buffer_mapped(bh) || buffer_freed(bh))
  1397. return 0;
  1398. set_buffer_uptodate(bh);
  1399. return ext4_handle_dirty_metadata(handle, NULL, bh);
  1400. }
  1401. static int ext4_generic_write_end(struct file *file,
  1402. struct address_space *mapping,
  1403. loff_t pos, unsigned len, unsigned copied,
  1404. struct page *page, void *fsdata)
  1405. {
  1406. int i_size_changed = 0;
  1407. struct inode *inode = mapping->host;
  1408. handle_t *handle = ext4_journal_current_handle();
  1409. copied = block_write_end(file, mapping, pos, len, copied, page, fsdata);
  1410. /*
  1411. * No need to use i_size_read() here, the i_size
  1412. * cannot change under us because we hold i_mutex.
  1413. *
  1414. * But it's important to update i_size while still holding page lock:
  1415. * page writeout could otherwise come in and zero beyond i_size.
  1416. */
  1417. if (pos + copied > inode->i_size) {
  1418. i_size_write(inode, pos + copied);
  1419. i_size_changed = 1;
  1420. }
  1421. if (pos + copied > EXT4_I(inode)->i_disksize) {
  1422. /* We need to mark inode dirty even if
  1423. * new_i_size is less that inode->i_size
  1424. * bu greater than i_disksize.(hint delalloc)
  1425. */
  1426. ext4_update_i_disksize(inode, (pos + copied));
  1427. i_size_changed = 1;
  1428. }
  1429. unlock_page(page);
  1430. page_cache_release(page);
  1431. /*
  1432. * Don't mark the inode dirty under page lock. First, it unnecessarily
  1433. * makes the holding time of page lock longer. Second, it forces lock
  1434. * ordering of page lock and transaction start for journaling
  1435. * filesystems.
  1436. */
  1437. if (i_size_changed)
  1438. ext4_mark_inode_dirty(handle, inode);
  1439. return copied;
  1440. }
  1441. /*
  1442. * We need to pick up the new inode size which generic_commit_write gave us
  1443. * `file' can be NULL - eg, when called from page_symlink().
  1444. *
  1445. * ext4 never places buffers on inode->i_mapping->private_list. metadata
  1446. * buffers are managed internally.
  1447. */
  1448. static int ext4_ordered_write_end(struct file *file,
  1449. struct address_space *mapping,
  1450. loff_t pos, unsigned len, unsigned copied,
  1451. struct page *page, void *fsdata)
  1452. {
  1453. handle_t *handle = ext4_journal_current_handle();
  1454. struct inode *inode = mapping->host;
  1455. int ret = 0, ret2;
  1456. trace_mark(ext4_ordered_write_end,
  1457. "dev %s ino %lu pos %llu len %u copied %u",
  1458. inode->i_sb->s_id, inode->i_ino,
  1459. (unsigned long long) pos, len, copied);
  1460. ret = ext4_jbd2_file_inode(handle, inode);
  1461. if (ret == 0) {
  1462. ret2 = ext4_generic_write_end(file, mapping, pos, len, copied,
  1463. page, fsdata);
  1464. copied = ret2;
  1465. if (pos + len > inode->i_size)
  1466. /* if we have allocated more blocks and copied
  1467. * less. We will have blocks allocated outside
  1468. * inode->i_size. So truncate them
  1469. */
  1470. ext4_orphan_add(handle, inode);
  1471. if (ret2 < 0)
  1472. ret = ret2;
  1473. }
  1474. ret2 = ext4_journal_stop(handle);
  1475. if (!ret)
  1476. ret = ret2;
  1477. if (pos + len > inode->i_size) {
  1478. vmtruncate(inode, inode->i_size);
  1479. /*
  1480. * If vmtruncate failed early the inode might still be
  1481. * on the orphan list; we need to make sure the inode
  1482. * is removed from the orphan list in that case.
  1483. */
  1484. if (inode->i_nlink)
  1485. ext4_orphan_del(NULL, inode);
  1486. }
  1487. return ret ? ret : copied;
  1488. }
  1489. static int ext4_writeback_write_end(struct file *file,
  1490. struct address_space *mapping,
  1491. loff_t pos, unsigned len, unsigned copied,
  1492. struct page *page, void *fsdata)
  1493. {
  1494. handle_t *handle = ext4_journal_current_handle();
  1495. struct inode *inode = mapping->host;
  1496. int ret = 0, ret2;
  1497. trace_mark(ext4_writeback_write_end,
  1498. "dev %s ino %lu pos %llu len %u copied %u",
  1499. inode->i_sb->s_id, inode->i_ino,
  1500. (unsigned long long) pos, len, copied);
  1501. ret2 = ext4_generic_write_end(file, mapping, pos, len, copied,
  1502. page, fsdata);
  1503. copied = ret2;
  1504. if (pos + len > inode->i_size)
  1505. /* if we have allocated more blocks and copied
  1506. * less. We will have blocks allocated outside
  1507. * inode->i_size. So truncate them
  1508. */
  1509. ext4_orphan_add(handle, inode);
  1510. if (ret2 < 0)
  1511. ret = ret2;
  1512. ret2 = ext4_journal_stop(handle);
  1513. if (!ret)
  1514. ret = ret2;
  1515. if (pos + len > inode->i_size) {
  1516. vmtruncate(inode, inode->i_size);
  1517. /*
  1518. * If vmtruncate failed early the inode might still be
  1519. * on the orphan list; we need to make sure the inode
  1520. * is removed from the orphan list in that case.
  1521. */
  1522. if (inode->i_nlink)
  1523. ext4_orphan_del(NULL, inode);
  1524. }
  1525. return ret ? ret : copied;
  1526. }
  1527. static int ext4_journalled_write_end(struct file *file,
  1528. struct address_space *mapping,
  1529. loff_t pos, unsigned len, unsigned copied,
  1530. struct page *page, void *fsdata)
  1531. {
  1532. handle_t *handle = ext4_journal_current_handle();
  1533. struct inode *inode = mapping->host;
  1534. int ret = 0, ret2;
  1535. int partial = 0;
  1536. unsigned from, to;
  1537. loff_t new_i_size;
  1538. trace_mark(ext4_journalled_write_end,
  1539. "dev %s ino %lu pos %llu len %u copied %u",
  1540. inode->i_sb->s_id, inode->i_ino,
  1541. (unsigned long long) pos, len, copied);
  1542. from = pos & (PAGE_CACHE_SIZE - 1);
  1543. to = from + len;
  1544. if (copied < len) {
  1545. if (!PageUptodate(page))
  1546. copied = 0;
  1547. page_zero_new_buffers(page, from+copied, to);
  1548. }
  1549. ret = walk_page_buffers(handle, page_buffers(page), from,
  1550. to, &partial, write_end_fn);
  1551. if (!partial)
  1552. SetPageUptodate(page);
  1553. new_i_size = pos + copied;
  1554. if (new_i_size > inode->i_size)
  1555. i_size_write(inode, pos+copied);
  1556. EXT4_I(inode)->i_state |= EXT4_STATE_JDATA;
  1557. if (new_i_size > EXT4_I(inode)->i_disksize) {
  1558. ext4_update_i_disksize(inode, new_i_size);
  1559. ret2 = ext4_mark_inode_dirty(handle, inode);
  1560. if (!ret)
  1561. ret = ret2;
  1562. }
  1563. unlock_page(page);
  1564. page_cache_release(page);
  1565. if (pos + len > inode->i_size)
  1566. /* if we have allocated more blocks and copied
  1567. * less. We will have blocks allocated outside
  1568. * inode->i_size. So truncate them
  1569. */
  1570. ext4_orphan_add(handle, inode);
  1571. ret2 = ext4_journal_stop(handle);
  1572. if (!ret)
  1573. ret = ret2;
  1574. if (pos + len > inode->i_size) {
  1575. vmtruncate(inode, inode->i_size);
  1576. /*
  1577. * If vmtruncate failed early the inode might still be
  1578. * on the orphan list; we need to make sure the inode
  1579. * is removed from the orphan list in that case.
  1580. */
  1581. if (inode->i_nlink)
  1582. ext4_orphan_del(NULL, inode);
  1583. }
  1584. return ret ? ret : copied;
  1585. }
  1586. static int ext4_da_reserve_space(struct inode *inode, int nrblocks)
  1587. {
  1588. int retries = 0;
  1589. struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
  1590. unsigned long md_needed, mdblocks, total = 0;
  1591. /*
  1592. * recalculate the amount of metadata blocks to reserve
  1593. * in order to allocate nrblocks
  1594. * worse case is one extent per block
  1595. */
  1596. repeat:
  1597. spin_lock(&EXT4_I(inode)->i_block_reservation_lock);
  1598. total = EXT4_I(inode)->i_reserved_data_blocks + nrblocks;
  1599. mdblocks = ext4_calc_metadata_amount(inode, total);
  1600. BUG_ON(mdblocks < EXT4_I(inode)->i_reserved_meta_blocks);
  1601. md_needed = mdblocks - EXT4_I(inode)->i_reserved_meta_blocks;
  1602. total = md_needed + nrblocks;
  1603. /*
  1604. * Make quota reservation here to prevent quota overflow
  1605. * later. Real quota accounting is done at pages writeout
  1606. * time.
  1607. */
  1608. if (vfs_dq_reserve_block(inode, total)) {
  1609. spin_unlock(&EXT4_I(inode)->i_block_reservation_lock);
  1610. return -EDQUOT;
  1611. }
  1612. if (ext4_claim_free_blocks(sbi, total)) {
  1613. spin_unlock(&EXT4_I(inode)->i_block_reservation_lock);
  1614. if (ext4_should_retry_alloc(inode->i_sb, &retries)) {
  1615. yield();
  1616. goto repeat;
  1617. }
  1618. vfs_dq_release_reservation_block(inode, total);
  1619. return -ENOSPC;
  1620. }
  1621. EXT4_I(inode)->i_reserved_data_blocks += nrblocks;
  1622. EXT4_I(inode)->i_reserved_meta_blocks = mdblocks;
  1623. spin_unlock(&EXT4_I(inode)->i_block_reservation_lock);
  1624. return 0; /* success */
  1625. }
  1626. static void ext4_da_release_space(struct inode *inode, int to_free)
  1627. {
  1628. struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
  1629. int total, mdb, mdb_free, release;
  1630. if (!to_free)
  1631. return; /* Nothing to release, exit */
  1632. spin_lock(&EXT4_I(inode)->i_block_reservation_lock);
  1633. if (!EXT4_I(inode)->i_reserved_data_blocks) {
  1634. /*
  1635. * if there is no reserved blocks, but we try to free some
  1636. * then the counter is messed up somewhere.
  1637. * but since this function is called from invalidate
  1638. * page, it's harmless to return without any action
  1639. */
  1640. printk(KERN_INFO "ext4 delalloc try to release %d reserved "
  1641. "blocks for inode %lu, but there is no reserved "
  1642. "data blocks\n", to_free, inode->i_ino);
  1643. spin_unlock(&EXT4_I(inode)->i_block_reservation_lock);
  1644. return;
  1645. }
  1646. /* recalculate the number of metablocks still need to be reserved */
  1647. total = EXT4_I(inode)->i_reserved_data_blocks - to_free;
  1648. mdb = ext4_calc_metadata_amount(inode, total);
  1649. /* figure out how many metablocks to release */
  1650. BUG_ON(mdb > EXT4_I(inode)->i_reserved_meta_blocks);
  1651. mdb_free = EXT4_I(inode)->i_reserved_meta_blocks - mdb;
  1652. release = to_free + mdb_free;
  1653. /* update fs dirty blocks counter for truncate case */
  1654. percpu_counter_sub(&sbi->s_dirtyblocks_counter, release);
  1655. /* update per-inode reservations */
  1656. BUG_ON(to_free > EXT4_I(inode)->i_reserved_data_blocks);
  1657. EXT4_I(inode)->i_reserved_data_blocks -= to_free;
  1658. BUG_ON(mdb > EXT4_I(inode)->i_reserved_meta_blocks);
  1659. EXT4_I(inode)->i_reserved_meta_blocks = mdb;
  1660. spin_unlock(&EXT4_I(inode)->i_block_reservation_lock);
  1661. vfs_dq_release_reservation_block(inode, release);
  1662. }
  1663. static void ext4_da_page_release_reservation(struct page *page,
  1664. unsigned long offset)
  1665. {
  1666. int to_release = 0;
  1667. struct buffer_head *head, *bh;
  1668. unsigned int curr_off = 0;
  1669. head = page_buffers(page);
  1670. bh = head;
  1671. do {
  1672. unsigned int next_off = curr_off + bh->b_size;
  1673. if ((offset <= curr_off) && (buffer_delay(bh))) {
  1674. to_release++;
  1675. clear_buffer_delay(bh);
  1676. }
  1677. curr_off = next_off;
  1678. } while ((bh = bh->b_this_page) != head);
  1679. ext4_da_release_space(page->mapping->host, to_release);
  1680. }
  1681. /*
  1682. * Delayed allocation stuff
  1683. */
  1684. struct mpage_da_data {
  1685. struct inode *inode;
  1686. sector_t b_blocknr; /* start block number of extent */
  1687. size_t b_size; /* size of extent */
  1688. unsigned long b_state; /* state of the extent */
  1689. unsigned long first_page, next_page; /* extent of pages */
  1690. struct writeback_control *wbc;
  1691. int io_done;
  1692. int pages_written;
  1693. int retval;
  1694. };
  1695. /*
  1696. * mpage_da_submit_io - walks through extent of pages and try to write
  1697. * them with writepage() call back
  1698. *
  1699. * @mpd->inode: inode
  1700. * @mpd->first_page: first page of the extent
  1701. * @mpd->next_page: page after the last page of the extent
  1702. *
  1703. * By the time mpage_da_submit_io() is called we expect all blocks
  1704. * to be allocated. this may be wrong if allocation failed.
  1705. *
  1706. * As pages are already locked by write_cache_pages(), we can't use it
  1707. */
  1708. static int mpage_da_submit_io(struct mpage_da_data *mpd)
  1709. {
  1710. long pages_skipped;
  1711. struct pagevec pvec;
  1712. unsigned long index, end;
  1713. int ret = 0, err, nr_pages, i;
  1714. struct inode *inode = mpd->inode;
  1715. struct address_space *mapping = inode->i_mapping;
  1716. BUG_ON(mpd->next_page <= mpd->first_page);
  1717. /*
  1718. * We need to start from the first_page to the next_page - 1
  1719. * to make sure we also write the mapped dirty buffer_heads.
  1720. * If we look at mpd->b_blocknr we would only be looking
  1721. * at the currently mapped buffer_heads.
  1722. */
  1723. index = mpd->first_page;
  1724. end = mpd->next_page - 1;
  1725. pagevec_init(&pvec, 0);
  1726. while (index <= end) {
  1727. nr_pages = pagevec_lookup(&pvec, mapping, index, PAGEVEC_SIZE);
  1728. if (nr_pages == 0)
  1729. break;
  1730. for (i = 0; i < nr_pages; i++) {
  1731. struct page *page = pvec.pages[i];
  1732. index = page->index;
  1733. if (index > end)
  1734. break;
  1735. index++;
  1736. BUG_ON(!PageLocked(page));
  1737. BUG_ON(PageWriteback(page));
  1738. pages_skipped = mpd->wbc->pages_skipped;
  1739. err = mapping->a_ops->writepage(page, mpd->wbc);
  1740. if (!err && (pages_skipped == mpd->wbc->pages_skipped))
  1741. /*
  1742. * have successfully written the page
  1743. * without skipping the same
  1744. */
  1745. mpd->pages_written++;
  1746. /*
  1747. * In error case, we have to continue because
  1748. * remaining pages are still locked
  1749. * XXX: unlock and re-dirty them?
  1750. */
  1751. if (ret == 0)
  1752. ret = err;
  1753. }
  1754. pagevec_release(&pvec);
  1755. }
  1756. return ret;
  1757. }
  1758. /*
  1759. * mpage_put_bnr_to_bhs - walk blocks and assign them actual numbers
  1760. *
  1761. * @mpd->inode - inode to walk through
  1762. * @exbh->b_blocknr - first block on a disk
  1763. * @exbh->b_size - amount of space in bytes
  1764. * @logical - first logical block to start assignment with
  1765. *
  1766. * the function goes through all passed space and put actual disk
  1767. * block numbers into buffer heads, dropping BH_Delay and BH_Unwritten
  1768. */
  1769. static void mpage_put_bnr_to_bhs(struct mpage_da_data *mpd, sector_t logical,
  1770. struct buffer_head *exbh)
  1771. {
  1772. struct inode *inode = mpd->inode;
  1773. struct address_space *mapping = inode->i_mapping;
  1774. int blocks = exbh->b_size >> inode->i_blkbits;
  1775. sector_t pblock = exbh->b_blocknr, cur_logical;
  1776. struct buffer_head *head, *bh;
  1777. pgoff_t index, end;
  1778. struct pagevec pvec;
  1779. int nr_pages, i;
  1780. index = logical >> (PAGE_CACHE_SHIFT - inode->i_blkbits);
  1781. end = (logical + blocks - 1) >> (PAGE_CACHE_SHIFT - inode->i_blkbits);
  1782. cur_logical = index << (PAGE_CACHE_SHIFT - inode->i_blkbits);
  1783. pagevec_init(&pvec, 0);
  1784. while (index <= end) {
  1785. /* XXX: optimize tail */
  1786. nr_pages = pagevec_lookup(&pvec, mapping, index, PAGEVEC_SIZE);
  1787. if (nr_pages == 0)
  1788. break;
  1789. for (i = 0; i < nr_pages; i++) {
  1790. struct page *page = pvec.pages[i];
  1791. index = page->index;
  1792. if (index > end)
  1793. break;
  1794. index++;
  1795. BUG_ON(!PageLocked(page));
  1796. BUG_ON(PageWriteback(page));
  1797. BUG_ON(!page_has_buffers(page));
  1798. bh = page_buffers(page);
  1799. head = bh;
  1800. /* skip blocks out of the range */
  1801. do {
  1802. if (cur_logical >= logical)
  1803. break;
  1804. cur_logical++;
  1805. } while ((bh = bh->b_this_page) != head);
  1806. do {
  1807. if (cur_logical >= logical + blocks)
  1808. break;
  1809. if (buffer_delay(bh) ||
  1810. buffer_unwritten(bh)) {
  1811. BUG_ON(bh->b_bdev != inode->i_sb->s_bdev);
  1812. if (buffer_delay(bh)) {
  1813. clear_buffer_delay(bh);
  1814. bh->b_blocknr = pblock;
  1815. } else {
  1816. /*
  1817. * unwritten already should have
  1818. * blocknr assigned. Verify that
  1819. */
  1820. clear_buffer_unwritten(bh);
  1821. BUG_ON(bh->b_blocknr != pblock);
  1822. }
  1823. } else if (buffer_mapped(bh))
  1824. BUG_ON(bh->b_blocknr != pblock);
  1825. cur_logical++;
  1826. pblock++;
  1827. } while ((bh = bh->b_this_page) != head);
  1828. }
  1829. pagevec_release(&pvec);
  1830. }
  1831. }
  1832. /*
  1833. * __unmap_underlying_blocks - just a helper function to unmap
  1834. * set of blocks described by @bh
  1835. */
  1836. static inline void __unmap_underlying_blocks(struct inode *inode,
  1837. struct buffer_head *bh)
  1838. {
  1839. struct block_device *bdev = inode->i_sb->s_bdev;
  1840. int blocks, i;
  1841. blocks = bh->b_size >> inode->i_blkbits;
  1842. for (i = 0; i < blocks; i++)
  1843. unmap_underlying_metadata(bdev, bh->b_blocknr + i);
  1844. }
  1845. static void ext4_da_block_invalidatepages(struct mpage_da_data *mpd,
  1846. sector_t logical, long blk_cnt)
  1847. {
  1848. int nr_pages, i;
  1849. pgoff_t index, end;
  1850. struct pagevec pvec;
  1851. struct inode *inode = mpd->inode;
  1852. struct address_space *mapping = inode->i_mapping;
  1853. index = logical >> (PAGE_CACHE_SHIFT - inode->i_blkbits);
  1854. end = (logical + blk_cnt - 1) >>
  1855. (PAGE_CACHE_SHIFT - inode->i_blkbits);
  1856. while (index <= end) {
  1857. nr_pages = pagevec_lookup(&pvec, mapping, index, PAGEVEC_SIZE);
  1858. if (nr_pages == 0)
  1859. break;
  1860. for (i = 0; i < nr_pages; i++) {
  1861. struct page *page = pvec.pages[i];
  1862. index = page->index;
  1863. if (index > end)
  1864. break;
  1865. index++;
  1866. BUG_ON(!PageLocked(page));
  1867. BUG_ON(PageWriteback(page));
  1868. block_invalidatepage(page, 0);
  1869. ClearPageUptodate(page);
  1870. unlock_page(page);
  1871. }
  1872. }
  1873. return;
  1874. }
  1875. static void ext4_print_free_blocks(struct inode *inode)
  1876. {
  1877. struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
  1878. printk(KERN_EMERG "Total free blocks count %lld\n",
  1879. ext4_count_free_blocks(inode->i_sb));
  1880. printk(KERN_EMERG "Free/Dirty block details\n");
  1881. printk(KERN_EMERG "free_blocks=%lld\n",
  1882. (long long)percpu_counter_sum(&sbi->s_freeblocks_counter));
  1883. printk(KERN_EMERG "dirty_blocks=%lld\n",
  1884. (long long)percpu_counter_sum(&sbi->s_dirtyblocks_counter));
  1885. printk(KERN_EMERG "Block reservation details\n");
  1886. printk(KERN_EMERG "i_reserved_data_blocks=%u\n",
  1887. EXT4_I(inode)->i_reserved_data_blocks);
  1888. printk(KERN_EMERG "i_reserved_meta_blocks=%u\n",
  1889. EXT4_I(inode)->i_reserved_meta_blocks);
  1890. return;
  1891. }
  1892. /*
  1893. * mpage_da_map_blocks - go through given space
  1894. *
  1895. * @mpd - bh describing space
  1896. *
  1897. * The function skips space we know is already mapped to disk blocks.
  1898. *
  1899. */
  1900. static int mpage_da_map_blocks(struct mpage_da_data *mpd)
  1901. {
  1902. int err, blks, get_blocks_flags;
  1903. struct buffer_head new;
  1904. sector_t next = mpd->b_blocknr;
  1905. unsigned max_blocks = mpd->b_size >> mpd->inode->i_blkbits;
  1906. loff_t disksize = EXT4_I(mpd->inode)->i_disksize;
  1907. handle_t *handle = NULL;
  1908. /*
  1909. * We consider only non-mapped and non-allocated blocks
  1910. */
  1911. if ((mpd->b_state & (1 << BH_Mapped)) &&
  1912. !(mpd->b_state & (1 << BH_Delay)) &&
  1913. !(mpd->b_state & (1 << BH_Unwritten)))
  1914. return 0;
  1915. /*
  1916. * If we didn't accumulate anything to write simply return
  1917. */
  1918. if (!mpd->b_size)
  1919. return 0;
  1920. handle = ext4_journal_current_handle();
  1921. BUG_ON(!handle);
  1922. /*
  1923. * Call ext4_get_blocks() to allocate any delayed allocation
  1924. * blocks, or to convert an uninitialized extent to be
  1925. * initialized (in the case where we have written into
  1926. * one or more preallocated blocks).
  1927. *
  1928. * We pass in the magic EXT4_GET_BLOCKS_DELALLOC_RESERVE to
  1929. * indicate that we are on the delayed allocation path. This
  1930. * affects functions in many different parts of the allocation
  1931. * call path. This flag exists primarily because we don't
  1932. * want to change *many* call functions, so ext4_get_blocks()
  1933. * will set the magic i_delalloc_reserved_flag once the
  1934. * inode's allocation semaphore is taken.
  1935. *
  1936. * If the blocks in questions were delalloc blocks, set
  1937. * EXT4_GET_BLOCKS_DELALLOC_RESERVE so the delalloc accounting
  1938. * variables are updated after the blocks have been allocated.
  1939. */
  1940. new.b_state = 0;
  1941. get_blocks_flags = (EXT4_GET_BLOCKS_CREATE |
  1942. EXT4_GET_BLOCKS_DELALLOC_RESERVE);
  1943. if (mpd->b_state & (1 << BH_Delay))
  1944. get_blocks_flags |= EXT4_GET_BLOCKS_UPDATE_RESERVE_SPACE;
  1945. blks = ext4_get_blocks(handle, mpd->inode, next, max_blocks,
  1946. &new, get_blocks_flags);
  1947. if (blks < 0) {
  1948. err = blks;
  1949. /*
  1950. * If get block returns with error we simply
  1951. * return. Later writepage will redirty the page and
  1952. * writepages will find the dirty page again
  1953. */
  1954. if (err == -EAGAIN)
  1955. return 0;
  1956. if (err == -ENOSPC &&
  1957. ext4_count_free_blocks(mpd->inode->i_sb)) {
  1958. mpd->retval = err;
  1959. return 0;
  1960. }
  1961. /*
  1962. * get block failure will cause us to loop in
  1963. * writepages, because a_ops->writepage won't be able
  1964. * to make progress. The page will be redirtied by
  1965. * writepage and writepages will again try to write
  1966. * the same.
  1967. */
  1968. printk(KERN_EMERG "%s block allocation failed for inode %lu "
  1969. "at logical offset %llu with max blocks "
  1970. "%zd with error %d\n",
  1971. __func__, mpd->inode->i_ino,
  1972. (unsigned long long)next,
  1973. mpd->b_size >> mpd->inode->i_blkbits, err);
  1974. printk(KERN_EMERG "This should not happen.!! "
  1975. "Data will be lost\n");
  1976. if (err == -ENOSPC) {
  1977. ext4_print_free_blocks(mpd->inode);
  1978. }
  1979. /* invalidate all the pages */
  1980. ext4_da_block_invalidatepages(mpd, next,
  1981. mpd->b_size >> mpd->inode->i_blkbits);
  1982. return err;
  1983. }
  1984. BUG_ON(blks == 0);
  1985. new.b_size = (blks << mpd->inode->i_blkbits);
  1986. if (buffer_new(&new))
  1987. __unmap_underlying_blocks(mpd->inode, &new);
  1988. /*
  1989. * If blocks are delayed marked, we need to
  1990. * put actual blocknr and drop delayed bit
  1991. */
  1992. if ((mpd->b_state & (1 << BH_Delay)) ||
  1993. (mpd->b_state & (1 << BH_Unwritten)))
  1994. mpage_put_bnr_to_bhs(mpd, next, &new);
  1995. if (ext4_should_order_data(mpd->inode)) {
  1996. err = ext4_jbd2_file_inode(handle, mpd->inode);
  1997. if (err)
  1998. return err;
  1999. }
  2000. /*
  2001. * Update on-disk size along with block allocation.
  2002. */
  2003. disksize = ((loff_t) next + blks) << mpd->inode->i_blkbits;
  2004. if (disksize > i_size_read(mpd->inode))
  2005. disksize = i_size_read(mpd->inode);
  2006. if (disksize > EXT4_I(mpd->inode)->i_disksize) {
  2007. ext4_update_i_disksize(mpd->inode, disksize);
  2008. return ext4_mark_inode_dirty(handle, mpd->inode);
  2009. }
  2010. return 0;
  2011. }
  2012. #define BH_FLAGS ((1 << BH_Uptodate) | (1 << BH_Mapped) | \
  2013. (1 << BH_Delay) | (1 << BH_Unwritten))
  2014. /*
  2015. * mpage_add_bh_to_extent - try to add one more block to extent of blocks
  2016. *
  2017. * @mpd->lbh - extent of blocks
  2018. * @logical - logical number of the block in the file
  2019. * @bh - bh of the block (used to access block's state)
  2020. *
  2021. * the function is used to collect contig. blocks in same state
  2022. */
  2023. static void mpage_add_bh_to_extent(struct mpage_da_data *mpd,
  2024. sector_t logical, size_t b_size,
  2025. unsigned long b_state)
  2026. {
  2027. sector_t next;
  2028. int nrblocks = mpd->b_size >> mpd->inode->i_blkbits;
  2029. /* check if thereserved journal credits might overflow */
  2030. if (!(EXT4_I(mpd->inode)->i_flags & EXT4_EXTENTS_FL)) {
  2031. if (nrblocks >= EXT4_MAX_TRANS_DATA) {
  2032. /*
  2033. * With non-extent format we are limited by the journal
  2034. * credit available. Total credit needed to insert
  2035. * nrblocks contiguous blocks is dependent on the
  2036. * nrblocks. So limit nrblocks.
  2037. */
  2038. goto flush_it;
  2039. } else if ((nrblocks + (b_size >> mpd->inode->i_blkbits)) >
  2040. EXT4_MAX_TRANS_DATA) {
  2041. /*
  2042. * Adding the new buffer_head would make it cross the
  2043. * allowed limit for which we have journal credit
  2044. * reserved. So limit the new bh->b_size
  2045. */
  2046. b_size = (EXT4_MAX_TRANS_DATA - nrblocks) <<
  2047. mpd->inode->i_blkbits;
  2048. /* we will do mpage_da_submit_io in the next loop */
  2049. }
  2050. }
  2051. /*
  2052. * First block in the extent
  2053. */
  2054. if (mpd->b_size == 0) {
  2055. mpd->b_blocknr = logical;
  2056. mpd->b_size = b_size;
  2057. mpd->b_state = b_state & BH_FLAGS;
  2058. return;
  2059. }
  2060. next = mpd->b_blocknr + nrblocks;
  2061. /*
  2062. * Can we merge the block to our big extent?
  2063. */
  2064. if (logical == next && (b_state & BH_FLAGS) == mpd->b_state) {
  2065. mpd->b_size += b_size;
  2066. return;
  2067. }
  2068. flush_it:
  2069. /*
  2070. * We couldn't merge the block to our extent, so we
  2071. * need to flush current extent and start new one
  2072. */
  2073. if (mpage_da_map_blocks(mpd) == 0)
  2074. mpage_da_submit_io(mpd);
  2075. mpd->io_done = 1;
  2076. return;
  2077. }
  2078. static int ext4_bh_unmapped_or_delay(handle_t *handle, struct buffer_head *bh)
  2079. {
  2080. /*
  2081. * unmapped buffer is possible for holes.
  2082. * delay buffer is possible with delayed allocation.
  2083. * We also need to consider unwritten buffer as unmapped.
  2084. */
  2085. return (!buffer_mapped(bh) || buffer_delay(bh) ||
  2086. buffer_unwritten(bh)) && buffer_dirty(bh);
  2087. }
  2088. /*
  2089. * __mpage_da_writepage - finds extent of pages and blocks
  2090. *
  2091. * @page: page to consider
  2092. * @wbc: not used, we just follow rules
  2093. * @data: context
  2094. *
  2095. * The function finds extents of pages and scan them for all blocks.
  2096. */
  2097. static int __mpage_da_writepage(struct page *page,
  2098. struct writeback_control *wbc, void *data)
  2099. {
  2100. struct mpage_da_data *mpd = data;
  2101. struct inode *inode = mpd->inode;
  2102. struct buffer_head *bh, *head;
  2103. sector_t logical;
  2104. if (mpd->io_done) {
  2105. /*
  2106. * Rest of the page in the page_vec
  2107. * redirty then and skip then. We will
  2108. * try to to write them again after
  2109. * starting a new transaction
  2110. */
  2111. redirty_page_for_writepage(wbc, page);
  2112. unlock_page(page);
  2113. return MPAGE_DA_EXTENT_TAIL;
  2114. }
  2115. /*
  2116. * Can we merge this page to current extent?
  2117. */
  2118. if (mpd->next_page != page->index) {
  2119. /*
  2120. * Nope, we can't. So, we map non-allocated blocks
  2121. * and start IO on them using writepage()
  2122. */
  2123. if (mpd->next_page != mpd->first_page) {
  2124. if (mpage_da_map_blocks(mpd) == 0)
  2125. mpage_da_submit_io(mpd);
  2126. /*
  2127. * skip rest of the page in the page_vec
  2128. */
  2129. mpd->io_done = 1;
  2130. redirty_page_for_writepage(wbc, page);
  2131. unlock_page(page);
  2132. return MPAGE_DA_EXTENT_TAIL;
  2133. }
  2134. /*
  2135. * Start next extent of pages ...
  2136. */
  2137. mpd->first_page = page->index;
  2138. /*
  2139. * ... and blocks
  2140. */
  2141. mpd->b_size = 0;
  2142. mpd->b_state = 0;
  2143. mpd->b_blocknr = 0;
  2144. }
  2145. mpd->next_page = page->index + 1;
  2146. logical = (sector_t) page->index <<
  2147. (PAGE_CACHE_SHIFT - inode->i_blkbits);
  2148. if (!page_has_buffers(page)) {
  2149. mpage_add_bh_to_extent(mpd, logical, PAGE_CACHE_SIZE,
  2150. (1 << BH_Dirty) | (1 << BH_Uptodate));
  2151. if (mpd->io_done)
  2152. return MPAGE_DA_EXTENT_TAIL;
  2153. } else {
  2154. /*
  2155. * Page with regular buffer heads, just add all dirty ones
  2156. */
  2157. head = page_buffers(page);
  2158. bh = head;
  2159. do {
  2160. BUG_ON(buffer_locked(bh));
  2161. /*
  2162. * We need to try to allocate
  2163. * unmapped blocks in the same page.
  2164. * Otherwise we won't make progress
  2165. * with the page in ext4_da_writepage
  2166. */
  2167. if (ext4_bh_unmapped_or_delay(NULL, bh)) {
  2168. mpage_add_bh_to_extent(mpd, logical,
  2169. bh->b_size,
  2170. bh->b_state);
  2171. if (mpd->io_done)
  2172. return MPAGE_DA_EXTENT_TAIL;
  2173. } else if (buffer_dirty(bh) && (buffer_mapped(bh))) {
  2174. /*
  2175. * mapped dirty buffer. We need to update
  2176. * the b_state because we look at
  2177. * b_state in mpage_da_map_blocks. We don't
  2178. * update b_size because if we find an
  2179. * unmapped buffer_head later we need to
  2180. * use the b_state flag of that buffer_head.
  2181. */
  2182. if (mpd->b_size == 0)
  2183. mpd->b_state = bh->b_state & BH_FLAGS;
  2184. }
  2185. logical++;
  2186. } while ((bh = bh->b_this_page) != head);
  2187. }
  2188. return 0;
  2189. }
  2190. /*
  2191. * This is a special get_blocks_t callback which is used by
  2192. * ext4_da_write_begin(). It will either return mapped block or
  2193. * reserve space for a single block.
  2194. *
  2195. * For delayed buffer_head we have BH_Mapped, BH_New, BH_Delay set.
  2196. * We also have b_blocknr = -1 and b_bdev initialized properly
  2197. *
  2198. * For unwritten buffer_head we have BH_Mapped, BH_New, BH_Unwritten set.
  2199. * We also have b_blocknr = physicalblock mapping unwritten extent and b_bdev
  2200. * initialized properly.
  2201. */
  2202. static int ext4_da_get_block_prep(struct inode *inode, sector_t iblock,
  2203. struct buffer_head *bh_result, int create)
  2204. {
  2205. int ret = 0;
  2206. sector_t invalid_block = ~((sector_t) 0xffff);
  2207. if (invalid_block < ext4_blocks_count(EXT4_SB(inode->i_sb)->s_es))
  2208. invalid_block = ~0;
  2209. BUG_ON(create == 0);
  2210. BUG_ON(bh_result->b_size != inode->i_sb->s_blocksize);
  2211. /*
  2212. * first, we need to know whether the block is allocated already
  2213. * preallocated blocks are unmapped but should treated
  2214. * the same as allocated blocks.
  2215. */
  2216. ret = ext4_get_blocks(NULL, inode, iblock, 1, bh_result, 0);
  2217. if ((ret == 0) && !buffer_delay(bh_result)) {
  2218. /* the block isn't (pre)allocated yet, let's reserve space */
  2219. /*
  2220. * XXX: __block_prepare_write() unmaps passed block,
  2221. * is it OK?
  2222. */
  2223. ret = ext4_da_reserve_space(inode, 1);
  2224. if (ret)
  2225. /* not enough space to reserve */
  2226. return ret;
  2227. map_bh(bh_result, inode->i_sb, invalid_block);
  2228. set_buffer_new(bh_result);
  2229. set_buffer_delay(bh_result);
  2230. } else if (ret > 0) {
  2231. bh_result->b_size = (ret << inode->i_blkbits);
  2232. if (buffer_unwritten(bh_result)) {
  2233. /* A delayed write to unwritten bh should
  2234. * be marked new and mapped. Mapped ensures
  2235. * that we don't do get_block multiple times
  2236. * when we write to the same offset and new
  2237. * ensures that we do proper zero out for
  2238. * partial write.
  2239. */
  2240. set_buffer_new(bh_result);
  2241. set_buffer_mapped(bh_result);
  2242. }
  2243. ret = 0;
  2244. }
  2245. return ret;
  2246. }
  2247. /*
  2248. * This function is used as a standard get_block_t calback function
  2249. * when there is no desire to allocate any blocks. It is used as a
  2250. * callback function for block_prepare_write(), nobh_writepage(), and
  2251. * block_write_full_page(). These functions should only try to map a
  2252. * single block at a time.
  2253. *
  2254. * Since this function doesn't do block allocations even if the caller
  2255. * requests it by passing in create=1, it is critically important that
  2256. * any caller checks to make sure that any buffer heads are returned
  2257. * by this function are either all already mapped or marked for
  2258. * delayed allocation before calling nobh_writepage() or
  2259. * block_write_full_page(). Otherwise, b_blocknr could be left
  2260. * unitialized, and the page write functions will be taken by
  2261. * surprise.
  2262. */
  2263. static int noalloc_get_block_write(struct inode *inode, sector_t iblock,
  2264. struct buffer_head *bh_result, int create)
  2265. {
  2266. int ret = 0;
  2267. unsigned max_blocks = bh_result->b_size >> inode->i_blkbits;
  2268. BUG_ON(bh_result->b_size != inode->i_sb->s_blocksize);
  2269. /*
  2270. * we don't want to do block allocation in writepage
  2271. * so call get_block_wrap with create = 0
  2272. */
  2273. ret = ext4_get_blocks(NULL, inode, iblock, max_blocks, bh_result, 0);
  2274. BUG_ON(create && ret == 0);
  2275. if (ret > 0) {
  2276. bh_result->b_size = (ret << inode->i_blkbits);
  2277. ret = 0;
  2278. }
  2279. return ret;
  2280. }
  2281. /*
  2282. * This function can get called via...
  2283. * - ext4_da_writepages after taking page lock (have journal handle)
  2284. * - journal_submit_inode_data_buffers (no journal handle)
  2285. * - shrink_page_list via pdflush (no journal handle)
  2286. * - grab_page_cache when doing write_begin (have journal handle)
  2287. */
  2288. static int ext4_da_writepage(struct page *page,
  2289. struct writeback_control *wbc)
  2290. {
  2291. int ret = 0;
  2292. loff_t size;
  2293. unsigned int len;
  2294. struct buffer_head *page_bufs;
  2295. struct inode *inode = page->mapping->host;
  2296. trace_mark(ext4_da_writepage,
  2297. "dev %s ino %lu page_index %lu",
  2298. inode->i_sb->s_id, inode->i_ino, page->index);
  2299. size = i_size_read(inode);
  2300. if (page->index == size >> PAGE_CACHE_SHIFT)
  2301. len = size & ~PAGE_CACHE_MASK;
  2302. else
  2303. len = PAGE_CACHE_SIZE;
  2304. if (page_has_buffers(page)) {
  2305. page_bufs = page_buffers(page);
  2306. if (walk_page_buffers(NULL, page_bufs, 0, len, NULL,
  2307. ext4_bh_unmapped_or_delay)) {
  2308. /*
  2309. * We don't want to do block allocation
  2310. * So redirty the page and return
  2311. * We may reach here when we do a journal commit
  2312. * via journal_submit_inode_data_buffers.
  2313. * If we don't have mapping block we just ignore
  2314. * them. We can also reach here via shrink_page_list
  2315. */
  2316. redirty_page_for_writepage(wbc, page);
  2317. unlock_page(page);
  2318. return 0;
  2319. }
  2320. } else {
  2321. /*
  2322. * The test for page_has_buffers() is subtle:
  2323. * We know the page is dirty but it lost buffers. That means
  2324. * that at some moment in time after write_begin()/write_end()
  2325. * has been called all buffers have been clean and thus they
  2326. * must have been written at least once. So they are all
  2327. * mapped and we can happily proceed with mapping them
  2328. * and writing the page.
  2329. *
  2330. * Try to initialize the buffer_heads and check whether
  2331. * all are mapped and non delay. We don't want to
  2332. * do block allocation here.
  2333. */
  2334. ret = block_prepare_write(page, 0, PAGE_CACHE_SIZE,
  2335. noalloc_get_block_write);
  2336. if (!ret) {
  2337. page_bufs = page_buffers(page);
  2338. /* check whether all are mapped and non delay */
  2339. if (walk_page_buffers(NULL, page_bufs, 0, len, NULL,
  2340. ext4_bh_unmapped_or_delay)) {
  2341. redirty_page_for_writepage(wbc, page);
  2342. unlock_page(page);
  2343. return 0;
  2344. }
  2345. } else {
  2346. /*
  2347. * We can't do block allocation here
  2348. * so just redity the page and unlock
  2349. * and return
  2350. */
  2351. redirty_page_for_writepage(wbc, page);
  2352. unlock_page(page);
  2353. return 0;
  2354. }
  2355. /* now mark the buffer_heads as dirty and uptodate */
  2356. block_commit_write(page, 0, PAGE_CACHE_SIZE);
  2357. }
  2358. if (test_opt(inode->i_sb, NOBH) && ext4_should_writeback_data(inode))
  2359. ret = nobh_writepage(page, noalloc_get_block_write, wbc);
  2360. else
  2361. ret = block_write_full_page(page, noalloc_get_block_write,
  2362. wbc);
  2363. return ret;
  2364. }
  2365. /*
  2366. * This is called via ext4_da_writepages() to
  2367. * calulate the total number of credits to reserve to fit
  2368. * a single extent allocation into a single transaction,
  2369. * ext4_da_writpeages() will loop calling this before
  2370. * the block allocation.
  2371. */
  2372. static int ext4_da_writepages_trans_blocks(struct inode *inode)
  2373. {
  2374. int max_blocks = EXT4_I(inode)->i_reserved_data_blocks;
  2375. /*
  2376. * With non-extent format the journal credit needed to
  2377. * insert nrblocks contiguous block is dependent on
  2378. * number of contiguous block. So we will limit
  2379. * number of contiguous block to a sane value
  2380. */
  2381. if (!(inode->i_flags & EXT4_EXTENTS_FL) &&
  2382. (max_blocks > EXT4_MAX_TRANS_DATA))
  2383. max_blocks = EXT4_MAX_TRANS_DATA;
  2384. return ext4_chunk_trans_blocks(inode, max_blocks);
  2385. }
  2386. static int ext4_da_writepages(struct address_space *mapping,
  2387. struct writeback_control *wbc)
  2388. {
  2389. pgoff_t index;
  2390. int range_whole = 0;
  2391. handle_t *handle = NULL;
  2392. struct mpage_da_data mpd;
  2393. struct inode *inode = mapping->host;
  2394. int no_nrwrite_index_update;
  2395. int pages_written = 0;
  2396. long pages_skipped;
  2397. int range_cyclic, cycled = 1, io_done = 0;
  2398. int needed_blocks, ret = 0, nr_to_writebump = 0;
  2399. struct ext4_sb_info *sbi = EXT4_SB(mapping->host->i_sb);
  2400. trace_mark(ext4_da_writepages,
  2401. "dev %s ino %lu nr_t_write %ld "
  2402. "pages_skipped %ld range_start %llu "
  2403. "range_end %llu nonblocking %d "
  2404. "for_kupdate %d for_reclaim %d "
  2405. "for_writepages %d range_cyclic %d",
  2406. inode->i_sb->s_id, inode->i_ino,
  2407. wbc->nr_to_write, wbc->pages_skipped,
  2408. (unsigned long long) wbc->range_start,
  2409. (unsigned long long) wbc->range_end,
  2410. wbc->nonblocking, wbc->for_kupdate,
  2411. wbc->for_reclaim, wbc->for_writepages,
  2412. wbc->range_cyclic);
  2413. /*
  2414. * No pages to write? This is mainly a kludge to avoid starting
  2415. * a transaction for special inodes like journal inode on last iput()
  2416. * because that could violate lock ordering on umount
  2417. */
  2418. if (!mapping->nrpages || !mapping_tagged(mapping, PAGECACHE_TAG_DIRTY))
  2419. return 0;
  2420. /*
  2421. * If the filesystem has aborted, it is read-only, so return
  2422. * right away instead of dumping stack traces later on that
  2423. * will obscure the real source of the problem. We test
  2424. * EXT4_MOUNT_ABORT instead of sb->s_flag's MS_RDONLY because
  2425. * the latter could be true if the filesystem is mounted
  2426. * read-only, and in that case, ext4_da_writepages should
  2427. * *never* be called, so if that ever happens, we would want
  2428. * the stack trace.
  2429. */
  2430. if (unlikely(sbi->s_mount_opt & EXT4_MOUNT_ABORT))
  2431. return -EROFS;
  2432. /*
  2433. * Make sure nr_to_write is >= sbi->s_mb_stream_request
  2434. * This make sure small files blocks are allocated in
  2435. * single attempt. This ensure that small files
  2436. * get less fragmented.
  2437. */
  2438. if (wbc->nr_to_write < sbi->s_mb_stream_request) {
  2439. nr_to_writebump = sbi->s_mb_stream_request - wbc->nr_to_write;
  2440. wbc->nr_to_write = sbi->s_mb_stream_request;
  2441. }
  2442. if (wbc->range_start == 0 && wbc->range_end == LLONG_MAX)
  2443. range_whole = 1;
  2444. range_cyclic = wbc->range_cyclic;
  2445. if (wbc->range_cyclic) {
  2446. index = mapping->writeback_index;
  2447. if (index)
  2448. cycled = 0;
  2449. wbc->range_start = index << PAGE_CACHE_SHIFT;
  2450. wbc->range_end = LLONG_MAX;
  2451. wbc->range_cyclic = 0;
  2452. } else
  2453. index = wbc->range_start >> PAGE_CACHE_SHIFT;
  2454. mpd.wbc = wbc;
  2455. mpd.inode = mapping->host;
  2456. /*
  2457. * we don't want write_cache_pages to update
  2458. * nr_to_write and writeback_index
  2459. */
  2460. no_nrwrite_index_update = wbc->no_nrwrite_index_update;
  2461. wbc->no_nrwrite_index_update = 1;
  2462. pages_skipped = wbc->pages_skipped;
  2463. retry:
  2464. while (!ret && wbc->nr_to_write > 0) {
  2465. /*
  2466. * we insert one extent at a time. So we need
  2467. * credit needed for single extent allocation.
  2468. * journalled mode is currently not supported
  2469. * by delalloc
  2470. */
  2471. BUG_ON(ext4_should_journal_data(inode));
  2472. needed_blocks = ext4_da_writepages_trans_blocks(inode);
  2473. /* start a new transaction*/
  2474. handle = ext4_journal_start(inode, needed_blocks);
  2475. if (IS_ERR(handle)) {
  2476. ret = PTR_ERR(handle);
  2477. printk(KERN_CRIT "%s: jbd2_start: "
  2478. "%ld pages, ino %lu; err %d\n", __func__,
  2479. wbc->nr_to_write, inode->i_ino, ret);
  2480. dump_stack();
  2481. goto out_writepages;
  2482. }
  2483. /*
  2484. * Now call __mpage_da_writepage to find the next
  2485. * contiguous region of logical blocks that need
  2486. * blocks to be allocated by ext4. We don't actually
  2487. * submit the blocks for I/O here, even though
  2488. * write_cache_pages thinks it will, and will set the
  2489. * pages as clean for write before calling
  2490. * __mpage_da_writepage().
  2491. */
  2492. mpd.b_size = 0;
  2493. mpd.b_state = 0;
  2494. mpd.b_blocknr = 0;
  2495. mpd.first_page = 0;
  2496. mpd.next_page = 0;
  2497. mpd.io_done = 0;
  2498. mpd.pages_written = 0;
  2499. mpd.retval = 0;
  2500. ret = write_cache_pages(mapping, wbc, __mpage_da_writepage,
  2501. &mpd);
  2502. /*
  2503. * If we have a contigous extent of pages and we
  2504. * haven't done the I/O yet, map the blocks and submit
  2505. * them for I/O.
  2506. */
  2507. if (!mpd.io_done && mpd.next_page != mpd.first_page) {
  2508. if (mpage_da_map_blocks(&mpd) == 0)
  2509. mpage_da_submit_io(&mpd);
  2510. mpd.io_done = 1;
  2511. ret = MPAGE_DA_EXTENT_TAIL;
  2512. }
  2513. wbc->nr_to_write -= mpd.pages_written;
  2514. ext4_journal_stop(handle);
  2515. if ((mpd.retval == -ENOSPC) && sbi->s_journal) {
  2516. /* commit the transaction which would
  2517. * free blocks released in the transaction
  2518. * and try again
  2519. */
  2520. jbd2_journal_force_commit_nested(sbi->s_journal);
  2521. wbc->pages_skipped = pages_skipped;
  2522. ret = 0;
  2523. } else if (ret == MPAGE_DA_EXTENT_TAIL) {
  2524. /*
  2525. * got one extent now try with
  2526. * rest of the pages
  2527. */
  2528. pages_written += mpd.pages_written;
  2529. wbc->pages_skipped = pages_skipped;
  2530. ret = 0;
  2531. io_done = 1;
  2532. } else if (wbc->nr_to_write)
  2533. /*
  2534. * There is no more writeout needed
  2535. * or we requested for a noblocking writeout
  2536. * and we found the device congested
  2537. */
  2538. break;
  2539. }
  2540. if (!io_done && !cycled) {
  2541. cycled = 1;
  2542. index = 0;
  2543. wbc->range_start = index << PAGE_CACHE_SHIFT;
  2544. wbc->range_end = mapping->writeback_index - 1;
  2545. goto retry;
  2546. }
  2547. if (pages_skipped != wbc->pages_skipped)
  2548. printk(KERN_EMERG "This should not happen leaving %s "
  2549. "with nr_to_write = %ld ret = %d\n",
  2550. __func__, wbc->nr_to_write, ret);
  2551. /* Update index */
  2552. index += pages_written;
  2553. wbc->range_cyclic = range_cyclic;
  2554. if (wbc->range_cyclic || (range_whole && wbc->nr_to_write > 0))
  2555. /*
  2556. * set the writeback_index so that range_cyclic
  2557. * mode will write it back later
  2558. */
  2559. mapping->writeback_index = index;
  2560. out_writepages:
  2561. if (!no_nrwrite_index_update)
  2562. wbc->no_nrwrite_index_update = 0;
  2563. wbc->nr_to_write -= nr_to_writebump;
  2564. trace_mark(ext4_da_writepage_result,
  2565. "dev %s ino %lu ret %d pages_written %d "
  2566. "pages_skipped %ld congestion %d "
  2567. "more_io %d no_nrwrite_index_update %d",
  2568. inode->i_sb->s_id, inode->i_ino, ret,
  2569. pages_written, wbc->pages_skipped,
  2570. wbc->encountered_congestion, wbc->more_io,
  2571. wbc->no_nrwrite_index_update);
  2572. return ret;
  2573. }
  2574. #define FALL_BACK_TO_NONDELALLOC 1
  2575. static int ext4_nonda_switch(struct super_block *sb)
  2576. {
  2577. s64 free_blocks, dirty_blocks;
  2578. struct ext4_sb_info *sbi = EXT4_SB(sb);
  2579. /*
  2580. * switch to non delalloc mode if we are running low
  2581. * on free block. The free block accounting via percpu
  2582. * counters can get slightly wrong with percpu_counter_batch getting
  2583. * accumulated on each CPU without updating global counters
  2584. * Delalloc need an accurate free block accounting. So switch
  2585. * to non delalloc when we are near to error range.
  2586. */
  2587. free_blocks = percpu_counter_read_positive(&sbi->s_freeblocks_counter);
  2588. dirty_blocks = percpu_counter_read_positive(&sbi->s_dirtyblocks_counter);
  2589. if (2 * free_blocks < 3 * dirty_blocks ||
  2590. free_blocks < (dirty_blocks + EXT4_FREEBLOCKS_WATERMARK)) {
  2591. /*
  2592. * free block count is less that 150% of dirty blocks
  2593. * or free blocks is less that watermark
  2594. */
  2595. return 1;
  2596. }
  2597. return 0;
  2598. }
  2599. static int ext4_da_write_begin(struct file *file, struct address_space *mapping,
  2600. loff_t pos, unsigned len, unsigned flags,
  2601. struct page **pagep, void **fsdata)
  2602. {
  2603. int ret, retries = 0;
  2604. struct page *page;
  2605. pgoff_t index;
  2606. unsigned from, to;
  2607. struct inode *inode = mapping->host;
  2608. handle_t *handle;
  2609. index = pos >> PAGE_CACHE_SHIFT;
  2610. from = pos & (PAGE_CACHE_SIZE - 1);
  2611. to = from + len;
  2612. if (ext4_nonda_switch(inode->i_sb)) {
  2613. *fsdata = (void *)FALL_BACK_TO_NONDELALLOC;
  2614. return ext4_write_begin(file, mapping, pos,
  2615. len, flags, pagep, fsdata);
  2616. }
  2617. *fsdata = (void *)0;
  2618. trace_mark(ext4_da_write_begin,
  2619. "dev %s ino %lu pos %llu len %u flags %u",
  2620. inode->i_sb->s_id, inode->i_ino,
  2621. (unsigned long long) pos, len, flags);
  2622. retry:
  2623. /*
  2624. * With delayed allocation, we don't log the i_disksize update
  2625. * if there is delayed block allocation. But we still need
  2626. * to journalling the i_disksize update if writes to the end
  2627. * of file which has an already mapped buffer.
  2628. */
  2629. handle = ext4_journal_start(inode, 1);
  2630. if (IS_ERR(handle)) {
  2631. ret = PTR_ERR(handle);
  2632. goto out;
  2633. }
  2634. /* We cannot recurse into the filesystem as the transaction is already
  2635. * started */
  2636. flags |= AOP_FLAG_NOFS;
  2637. page = grab_cache_page_write_begin(mapping, index, flags);
  2638. if (!page) {
  2639. ext4_journal_stop(handle);
  2640. ret = -ENOMEM;
  2641. goto out;
  2642. }
  2643. *pagep = page;
  2644. ret = block_write_begin(file, mapping, pos, len, flags, pagep, fsdata,
  2645. ext4_da_get_block_prep);
  2646. if (ret < 0) {
  2647. unlock_page(page);
  2648. ext4_journal_stop(handle);
  2649. page_cache_release(page);
  2650. /*
  2651. * block_write_begin may have instantiated a few blocks
  2652. * outside i_size. Trim these off again. Don't need
  2653. * i_size_read because we hold i_mutex.
  2654. */
  2655. if (pos + len > inode->i_size)
  2656. vmtruncate(inode, inode->i_size);
  2657. }
  2658. if (ret == -ENOSPC && ext4_should_retry_alloc(inode->i_sb, &retries))
  2659. goto retry;
  2660. out:
  2661. return ret;
  2662. }
  2663. /*
  2664. * Check if we should update i_disksize
  2665. * when write to the end of file but not require block allocation
  2666. */
  2667. static int ext4_da_should_update_i_disksize(struct page *page,
  2668. unsigned long offset)
  2669. {
  2670. struct buffer_head *bh;
  2671. struct inode *inode = page->mapping->host;
  2672. unsigned int idx;
  2673. int i;
  2674. bh = page_buffers(page);
  2675. idx = offset >> inode->i_blkbits;
  2676. for (i = 0; i < idx; i++)
  2677. bh = bh->b_this_page;
  2678. if (!buffer_mapped(bh) || (buffer_delay(bh)) || buffer_unwritten(bh))
  2679. return 0;
  2680. return 1;
  2681. }
  2682. static int ext4_da_write_end(struct file *file,
  2683. struct address_space *mapping,
  2684. loff_t pos, unsigned len, unsigned copied,
  2685. struct page *page, void *fsdata)
  2686. {
  2687. struct inode *inode = mapping->host;
  2688. int ret = 0, ret2;
  2689. handle_t *handle = ext4_journal_current_handle();
  2690. loff_t new_i_size;
  2691. unsigned long start, end;
  2692. int write_mode = (int)(unsigned long)fsdata;
  2693. if (write_mode == FALL_BACK_TO_NONDELALLOC) {
  2694. if (ext4_should_order_data(inode)) {
  2695. return ext4_ordered_write_end(file, mapping, pos,
  2696. len, copied, page, fsdata);
  2697. } else if (ext4_should_writeback_data(inode)) {
  2698. return ext4_writeback_write_end(file, mapping, pos,
  2699. len, copied, page, fsdata);
  2700. } else {
  2701. BUG();
  2702. }
  2703. }
  2704. trace_mark(ext4_da_write_end,
  2705. "dev %s ino %lu pos %llu len %u copied %u",
  2706. inode->i_sb->s_id, inode->i_ino,
  2707. (unsigned long long) pos, len, copied);
  2708. start = pos & (PAGE_CACHE_SIZE - 1);
  2709. end = start + copied - 1;
  2710. /*
  2711. * generic_write_end() will run mark_inode_dirty() if i_size
  2712. * changes. So let's piggyback the i_disksize mark_inode_dirty
  2713. * into that.
  2714. */
  2715. new_i_size = pos + copied;
  2716. if (new_i_size > EXT4_I(inode)->i_disksize) {
  2717. if (ext4_da_should_update_i_disksize(page, end)) {
  2718. down_write(&EXT4_I(inode)->i_data_sem);
  2719. if (new_i_size > EXT4_I(inode)->i_disksize) {
  2720. /*
  2721. * Updating i_disksize when extending file
  2722. * without needing block allocation
  2723. */
  2724. if (ext4_should_order_data(inode))
  2725. ret = ext4_jbd2_file_inode(handle,
  2726. inode);
  2727. EXT4_I(inode)->i_disksize = new_i_size;
  2728. }
  2729. up_write(&EXT4_I(inode)->i_data_sem);
  2730. /* We need to mark inode dirty even if
  2731. * new_i_size is less that inode->i_size
  2732. * bu greater than i_disksize.(hint delalloc)
  2733. */
  2734. ext4_mark_inode_dirty(handle, inode);
  2735. }
  2736. }
  2737. ret2 = generic_write_end(file, mapping, pos, len, copied,
  2738. page, fsdata);
  2739. copied = ret2;
  2740. if (ret2 < 0)
  2741. ret = ret2;
  2742. ret2 = ext4_journal_stop(handle);
  2743. if (!ret)
  2744. ret = ret2;
  2745. return ret ? ret : copied;
  2746. }
  2747. static void ext4_da_invalidatepage(struct page *page, unsigned long offset)
  2748. {
  2749. /*
  2750. * Drop reserved blocks
  2751. */
  2752. BUG_ON(!PageLocked(page));
  2753. if (!page_has_buffers(page))
  2754. goto out;
  2755. ext4_da_page_release_reservation(page, offset);
  2756. out:
  2757. ext4_invalidatepage(page, offset);
  2758. return;
  2759. }
  2760. /*
  2761. * Force all delayed allocation blocks to be allocated for a given inode.
  2762. */
  2763. int ext4_alloc_da_blocks(struct inode *inode)
  2764. {
  2765. if (!EXT4_I(inode)->i_reserved_data_blocks &&
  2766. !EXT4_I(inode)->i_reserved_meta_blocks)
  2767. return 0;
  2768. /*
  2769. * We do something simple for now. The filemap_flush() will
  2770. * also start triggering a write of the data blocks, which is
  2771. * not strictly speaking necessary (and for users of
  2772. * laptop_mode, not even desirable). However, to do otherwise
  2773. * would require replicating code paths in:
  2774. *
  2775. * ext4_da_writepages() ->
  2776. * write_cache_pages() ---> (via passed in callback function)
  2777. * __mpage_da_writepage() -->
  2778. * mpage_add_bh_to_extent()
  2779. * mpage_da_map_blocks()
  2780. *
  2781. * The problem is that write_cache_pages(), located in
  2782. * mm/page-writeback.c, marks pages clean in preparation for
  2783. * doing I/O, which is not desirable if we're not planning on
  2784. * doing I/O at all.
  2785. *
  2786. * We could call write_cache_pages(), and then redirty all of
  2787. * the pages by calling redirty_page_for_writeback() but that
  2788. * would be ugly in the extreme. So instead we would need to
  2789. * replicate parts of the code in the above functions,
  2790. * simplifying them becuase we wouldn't actually intend to
  2791. * write out the pages, but rather only collect contiguous
  2792. * logical block extents, call the multi-block allocator, and
  2793. * then update the buffer heads with the block allocations.
  2794. *
  2795. * For now, though, we'll cheat by calling filemap_flush(),
  2796. * which will map the blocks, and start the I/O, but not
  2797. * actually wait for the I/O to complete.
  2798. */
  2799. return filemap_flush(inode->i_mapping);
  2800. }
  2801. /*
  2802. * bmap() is special. It gets used by applications such as lilo and by
  2803. * the swapper to find the on-disk block of a specific piece of data.
  2804. *
  2805. * Naturally, this is dangerous if the block concerned is still in the
  2806. * journal. If somebody makes a swapfile on an ext4 data-journaling
  2807. * filesystem and enables swap, then they may get a nasty shock when the
  2808. * data getting swapped to that swapfile suddenly gets overwritten by
  2809. * the original zero's written out previously to the journal and
  2810. * awaiting writeback in the kernel's buffer cache.
  2811. *
  2812. * So, if we see any bmap calls here on a modified, data-journaled file,
  2813. * take extra steps to flush any blocks which might be in the cache.
  2814. */
  2815. static sector_t ext4_bmap(struct address_space *mapping, sector_t block)
  2816. {
  2817. struct inode *inode = mapping->host;
  2818. journal_t *journal;
  2819. int err;
  2820. if (mapping_tagged(mapping, PAGECACHE_TAG_DIRTY) &&
  2821. test_opt(inode->i_sb, DELALLOC)) {
  2822. /*
  2823. * With delalloc we want to sync the file
  2824. * so that we can make sure we allocate
  2825. * blocks for file
  2826. */
  2827. filemap_write_and_wait(mapping);
  2828. }
  2829. if (EXT4_JOURNAL(inode) && EXT4_I(inode)->i_state & EXT4_STATE_JDATA) {
  2830. /*
  2831. * This is a REALLY heavyweight approach, but the use of
  2832. * bmap on dirty files is expected to be extremely rare:
  2833. * only if we run lilo or swapon on a freshly made file
  2834. * do we expect this to happen.
  2835. *
  2836. * (bmap requires CAP_SYS_RAWIO so this does not
  2837. * represent an unprivileged user DOS attack --- we'd be
  2838. * in trouble if mortal users could trigger this path at
  2839. * will.)
  2840. *
  2841. * NB. EXT4_STATE_JDATA is not set on files other than
  2842. * regular files. If somebody wants to bmap a directory
  2843. * or symlink and gets confused because the buffer
  2844. * hasn't yet been flushed to disk, they deserve
  2845. * everything they get.
  2846. */
  2847. EXT4_I(inode)->i_state &= ~EXT4_STATE_JDATA;
  2848. journal = EXT4_JOURNAL(inode);
  2849. jbd2_journal_lock_updates(journal);
  2850. err = jbd2_journal_flush(journal);
  2851. jbd2_journal_unlock_updates(journal);
  2852. if (err)
  2853. return 0;
  2854. }
  2855. return generic_block_bmap(mapping, block, ext4_get_block);
  2856. }
  2857. static int bget_one(handle_t *handle, struct buffer_head *bh)
  2858. {
  2859. get_bh(bh);
  2860. return 0;
  2861. }
  2862. static int bput_one(handle_t *handle, struct buffer_head *bh)
  2863. {
  2864. put_bh(bh);
  2865. return 0;
  2866. }
  2867. /*
  2868. * Note that we don't need to start a transaction unless we're journaling data
  2869. * because we should have holes filled from ext4_page_mkwrite(). We even don't
  2870. * need to file the inode to the transaction's list in ordered mode because if
  2871. * we are writing back data added by write(), the inode is already there and if
  2872. * we are writing back data modified via mmap(), noone guarantees in which
  2873. * transaction the data will hit the disk. In case we are journaling data, we
  2874. * cannot start transaction directly because transaction start ranks above page
  2875. * lock so we have to do some magic.
  2876. *
  2877. * In all journaling modes block_write_full_page() will start the I/O.
  2878. *
  2879. * Problem:
  2880. *
  2881. * ext4_writepage() -> kmalloc() -> __alloc_pages() -> page_launder() ->
  2882. * ext4_writepage()
  2883. *
  2884. * Similar for:
  2885. *
  2886. * ext4_file_write() -> generic_file_write() -> __alloc_pages() -> ...
  2887. *
  2888. * Same applies to ext4_get_block(). We will deadlock on various things like
  2889. * lock_journal and i_data_sem
  2890. *
  2891. * Setting PF_MEMALLOC here doesn't work - too many internal memory
  2892. * allocations fail.
  2893. *
  2894. * 16May01: If we're reentered then journal_current_handle() will be
  2895. * non-zero. We simply *return*.
  2896. *
  2897. * 1 July 2001: @@@ FIXME:
  2898. * In journalled data mode, a data buffer may be metadata against the
  2899. * current transaction. But the same file is part of a shared mapping
  2900. * and someone does a writepage() on it.
  2901. *
  2902. * We will move the buffer onto the async_data list, but *after* it has
  2903. * been dirtied. So there's a small window where we have dirty data on
  2904. * BJ_Metadata.
  2905. *
  2906. * Note that this only applies to the last partial page in the file. The
  2907. * bit which block_write_full_page() uses prepare/commit for. (That's
  2908. * broken code anyway: it's wrong for msync()).
  2909. *
  2910. * It's a rare case: affects the final partial page, for journalled data
  2911. * where the file is subject to bith write() and writepage() in the same
  2912. * transction. To fix it we'll need a custom block_write_full_page().
  2913. * We'll probably need that anyway for journalling writepage() output.
  2914. *
  2915. * We don't honour synchronous mounts for writepage(). That would be
  2916. * disastrous. Any write() or metadata operation will sync the fs for
  2917. * us.
  2918. *
  2919. */
  2920. static int __ext4_normal_writepage(struct page *page,
  2921. struct writeback_control *wbc)
  2922. {
  2923. struct inode *inode = page->mapping->host;
  2924. if (test_opt(inode->i_sb, NOBH))
  2925. return nobh_writepage(page, noalloc_get_block_write, wbc);
  2926. else
  2927. return block_write_full_page(page, noalloc_get_block_write,
  2928. wbc);
  2929. }
  2930. static int ext4_normal_writepage(struct page *page,
  2931. struct writeback_control *wbc)
  2932. {
  2933. struct inode *inode = page->mapping->host;
  2934. loff_t size = i_size_read(inode);
  2935. loff_t len;
  2936. trace_mark(ext4_normal_writepage,
  2937. "dev %s ino %lu page_index %lu",
  2938. inode->i_sb->s_id, inode->i_ino, page->index);
  2939. J_ASSERT(PageLocked(page));
  2940. if (page->index == size >> PAGE_CACHE_SHIFT)
  2941. len = size & ~PAGE_CACHE_MASK;
  2942. else
  2943. len = PAGE_CACHE_SIZE;
  2944. if (page_has_buffers(page)) {
  2945. /* if page has buffers it should all be mapped
  2946. * and allocated. If there are not buffers attached
  2947. * to the page we know the page is dirty but it lost
  2948. * buffers. That means that at some moment in time
  2949. * after write_begin() / write_end() has been called
  2950. * all buffers have been clean and thus they must have been
  2951. * written at least once. So they are all mapped and we can
  2952. * happily proceed with mapping them and writing the page.
  2953. */
  2954. BUG_ON(walk_page_buffers(NULL, page_buffers(page), 0, len, NULL,
  2955. ext4_bh_unmapped_or_delay));
  2956. }
  2957. if (!ext4_journal_current_handle())
  2958. return __ext4_normal_writepage(page, wbc);
  2959. redirty_page_for_writepage(wbc, page);
  2960. unlock_page(page);
  2961. return 0;
  2962. }
  2963. static int __ext4_journalled_writepage(struct page *page,
  2964. struct writeback_control *wbc)
  2965. {
  2966. struct address_space *mapping = page->mapping;
  2967. struct inode *inode = mapping->host;
  2968. struct buffer_head *page_bufs;
  2969. handle_t *handle = NULL;
  2970. int ret = 0;
  2971. int err;
  2972. ret = block_prepare_write(page, 0, PAGE_CACHE_SIZE,
  2973. noalloc_get_block_write);
  2974. if (ret != 0)
  2975. goto out_unlock;
  2976. page_bufs = page_buffers(page);
  2977. walk_page_buffers(handle, page_bufs, 0, PAGE_CACHE_SIZE, NULL,
  2978. bget_one);
  2979. /* As soon as we unlock the page, it can go away, but we have
  2980. * references to buffers so we are safe */
  2981. unlock_page(page);
  2982. handle = ext4_journal_start(inode, ext4_writepage_trans_blocks(inode));
  2983. if (IS_ERR(handle)) {
  2984. ret = PTR_ERR(handle);
  2985. goto out;
  2986. }
  2987. ret = walk_page_buffers(handle, page_bufs, 0,
  2988. PAGE_CACHE_SIZE, NULL, do_journal_get_write_access);
  2989. err = walk_page_buffers(handle, page_bufs, 0,
  2990. PAGE_CACHE_SIZE, NULL, write_end_fn);
  2991. if (ret == 0)
  2992. ret = err;
  2993. err = ext4_journal_stop(handle);
  2994. if (!ret)
  2995. ret = err;
  2996. walk_page_buffers(handle, page_bufs, 0,
  2997. PAGE_CACHE_SIZE, NULL, bput_one);
  2998. EXT4_I(inode)->i_state |= EXT4_STATE_JDATA;
  2999. goto out;
  3000. out_unlock:
  3001. unlock_page(page);
  3002. out:
  3003. return ret;
  3004. }
  3005. static int ext4_journalled_writepage(struct page *page,
  3006. struct writeback_control *wbc)
  3007. {
  3008. struct inode *inode = page->mapping->host;
  3009. loff_t size = i_size_read(inode);
  3010. loff_t len;
  3011. trace_mark(ext4_journalled_writepage,
  3012. "dev %s ino %lu page_index %lu",
  3013. inode->i_sb->s_id, inode->i_ino, page->index);
  3014. J_ASSERT(PageLocked(page));
  3015. if (page->index == size >> PAGE_CACHE_SHIFT)
  3016. len = size & ~PAGE_CACHE_MASK;
  3017. else
  3018. len = PAGE_CACHE_SIZE;
  3019. if (page_has_buffers(page)) {
  3020. /* if page has buffers it should all be mapped
  3021. * and allocated. If there are not buffers attached
  3022. * to the page we know the page is dirty but it lost
  3023. * buffers. That means that at some moment in time
  3024. * after write_begin() / write_end() has been called
  3025. * all buffers have been clean and thus they must have been
  3026. * written at least once. So they are all mapped and we can
  3027. * happily proceed with mapping them and writing the page.
  3028. */
  3029. BUG_ON(walk_page_buffers(NULL, page_buffers(page), 0, len, NULL,
  3030. ext4_bh_unmapped_or_delay));
  3031. }
  3032. if (ext4_journal_current_handle())
  3033. goto no_write;
  3034. if (PageChecked(page)) {
  3035. /*
  3036. * It's mmapped pagecache. Add buffers and journal it. There
  3037. * doesn't seem much point in redirtying the page here.
  3038. */
  3039. ClearPageChecked(page);
  3040. return __ext4_journalled_writepage(page, wbc);
  3041. } else {
  3042. /*
  3043. * It may be a page full of checkpoint-mode buffers. We don't
  3044. * really know unless we go poke around in the buffer_heads.
  3045. * But block_write_full_page will do the right thing.
  3046. */
  3047. return block_write_full_page(page, noalloc_get_block_write,
  3048. wbc);
  3049. }
  3050. no_write:
  3051. redirty_page_for_writepage(wbc, page);
  3052. unlock_page(page);
  3053. return 0;
  3054. }
  3055. static int ext4_readpage(struct file *file, struct page *page)
  3056. {
  3057. return mpage_readpage(page, ext4_get_block);
  3058. }
  3059. static int
  3060. ext4_readpages(struct file *file, struct address_space *mapping,
  3061. struct list_head *pages, unsigned nr_pages)
  3062. {
  3063. return mpage_readpages(mapping, pages, nr_pages, ext4_get_block);
  3064. }
  3065. static void ext4_invalidatepage(struct page *page, unsigned long offset)
  3066. {
  3067. journal_t *journal = EXT4_JOURNAL(page->mapping->host);
  3068. /*
  3069. * If it's a full truncate we just forget about the pending dirtying
  3070. */
  3071. if (offset == 0)
  3072. ClearPageChecked(page);
  3073. if (journal)
  3074. jbd2_journal_invalidatepage(journal, page, offset);
  3075. else
  3076. block_invalidatepage(page, offset);
  3077. }
  3078. static int ext4_releasepage(struct page *page, gfp_t wait)
  3079. {
  3080. journal_t *journal = EXT4_JOURNAL(page->mapping->host);
  3081. WARN_ON(PageChecked(page));
  3082. if (!page_has_buffers(page))
  3083. return 0;
  3084. if (journal)
  3085. return jbd2_journal_try_to_free_buffers(journal, page, wait);
  3086. else
  3087. return try_to_free_buffers(page);
  3088. }
  3089. /*
  3090. * If the O_DIRECT write will extend the file then add this inode to the
  3091. * orphan list. So recovery will truncate it back to the original size
  3092. * if the machine crashes during the write.
  3093. *
  3094. * If the O_DIRECT write is intantiating holes inside i_size and the machine
  3095. * crashes then stale disk data _may_ be exposed inside the file. But current
  3096. * VFS code falls back into buffered path in that case so we are safe.
  3097. */
  3098. static ssize_t ext4_direct_IO(int rw, struct kiocb *iocb,
  3099. const struct iovec *iov, loff_t offset,
  3100. unsigned long nr_segs)
  3101. {
  3102. struct file *file = iocb->ki_filp;
  3103. struct inode *inode = file->f_mapping->host;
  3104. struct ext4_inode_info *ei = EXT4_I(inode);
  3105. handle_t *handle;
  3106. ssize_t ret;
  3107. int orphan = 0;
  3108. size_t count = iov_length(iov, nr_segs);
  3109. if (rw == WRITE) {
  3110. loff_t final_size = offset + count;
  3111. if (final_size > inode->i_size) {
  3112. /* Credits for sb + inode write */
  3113. handle = ext4_journal_start(inode, 2);
  3114. if (IS_ERR(handle)) {
  3115. ret = PTR_ERR(handle);
  3116. goto out;
  3117. }
  3118. ret = ext4_orphan_add(handle, inode);
  3119. if (ret) {
  3120. ext4_journal_stop(handle);
  3121. goto out;
  3122. }
  3123. orphan = 1;
  3124. ei->i_disksize = inode->i_size;
  3125. ext4_journal_stop(handle);
  3126. }
  3127. }
  3128. ret = blockdev_direct_IO(rw, iocb, inode, inode->i_sb->s_bdev, iov,
  3129. offset, nr_segs,
  3130. ext4_get_block, NULL);
  3131. if (orphan) {
  3132. int err;
  3133. /* Credits for sb + inode write */
  3134. handle = ext4_journal_start(inode, 2);
  3135. if (IS_ERR(handle)) {
  3136. /* This is really bad luck. We've written the data
  3137. * but cannot extend i_size. Bail out and pretend
  3138. * the write failed... */
  3139. ret = PTR_ERR(handle);
  3140. goto out;
  3141. }
  3142. if (inode->i_nlink)
  3143. ext4_orphan_del(handle, inode);
  3144. if (ret > 0) {
  3145. loff_t end = offset + ret;
  3146. if (end > inode->i_size) {
  3147. ei->i_disksize = end;
  3148. i_size_write(inode, end);
  3149. /*
  3150. * We're going to return a positive `ret'
  3151. * here due to non-zero-length I/O, so there's
  3152. * no way of reporting error returns from
  3153. * ext4_mark_inode_dirty() to userspace. So
  3154. * ignore it.
  3155. */
  3156. ext4_mark_inode_dirty(handle, inode);
  3157. }
  3158. }
  3159. err = ext4_journal_stop(handle);
  3160. if (ret == 0)
  3161. ret = err;
  3162. }
  3163. out:
  3164. return ret;
  3165. }
  3166. /*
  3167. * Pages can be marked dirty completely asynchronously from ext4's journalling
  3168. * activity. By filemap_sync_pte(), try_to_unmap_one(), etc. We cannot do
  3169. * much here because ->set_page_dirty is called under VFS locks. The page is
  3170. * not necessarily locked.
  3171. *
  3172. * We cannot just dirty the page and leave attached buffers clean, because the
  3173. * buffers' dirty state is "definitive". We cannot just set the buffers dirty
  3174. * or jbddirty because all the journalling code will explode.
  3175. *
  3176. * So what we do is to mark the page "pending dirty" and next time writepage
  3177. * is called, propagate that into the buffers appropriately.
  3178. */
  3179. static int ext4_journalled_set_page_dirty(struct page *page)
  3180. {
  3181. SetPageChecked(page);
  3182. return __set_page_dirty_nobuffers(page);
  3183. }
  3184. static const struct address_space_operations ext4_ordered_aops = {
  3185. .readpage = ext4_readpage,
  3186. .readpages = ext4_readpages,
  3187. .writepage = ext4_normal_writepage,
  3188. .sync_page = block_sync_page,
  3189. .write_begin = ext4_write_begin,
  3190. .write_end = ext4_ordered_write_end,
  3191. .bmap = ext4_bmap,
  3192. .invalidatepage = ext4_invalidatepage,
  3193. .releasepage = ext4_releasepage,
  3194. .direct_IO = ext4_direct_IO,
  3195. .migratepage = buffer_migrate_page,
  3196. .is_partially_uptodate = block_is_partially_uptodate,
  3197. };
  3198. static const struct address_space_operations ext4_writeback_aops = {
  3199. .readpage = ext4_readpage,
  3200. .readpages = ext4_readpages,
  3201. .writepage = ext4_normal_writepage,
  3202. .sync_page = block_sync_page,
  3203. .write_begin = ext4_write_begin,
  3204. .write_end = ext4_writeback_write_end,
  3205. .bmap = ext4_bmap,
  3206. .invalidatepage = ext4_invalidatepage,
  3207. .releasepage = ext4_releasepage,
  3208. .direct_IO = ext4_direct_IO,
  3209. .migratepage = buffer_migrate_page,
  3210. .is_partially_uptodate = block_is_partially_uptodate,
  3211. };
  3212. static const struct address_space_operations ext4_journalled_aops = {
  3213. .readpage = ext4_readpage,
  3214. .readpages = ext4_readpages,
  3215. .writepage = ext4_journalled_writepage,
  3216. .sync_page = block_sync_page,
  3217. .write_begin = ext4_write_begin,
  3218. .write_end = ext4_journalled_write_end,
  3219. .set_page_dirty = ext4_journalled_set_page_dirty,
  3220. .bmap = ext4_bmap,
  3221. .invalidatepage = ext4_invalidatepage,
  3222. .releasepage = ext4_releasepage,
  3223. .is_partially_uptodate = block_is_partially_uptodate,
  3224. };
  3225. static const struct address_space_operations ext4_da_aops = {
  3226. .readpage = ext4_readpage,
  3227. .readpages = ext4_readpages,
  3228. .writepage = ext4_da_writepage,
  3229. .writepages = ext4_da_writepages,
  3230. .sync_page = block_sync_page,
  3231. .write_begin = ext4_da_write_begin,
  3232. .write_end = ext4_da_write_end,
  3233. .bmap = ext4_bmap,
  3234. .invalidatepage = ext4_da_invalidatepage,
  3235. .releasepage = ext4_releasepage,
  3236. .direct_IO = ext4_direct_IO,
  3237. .migratepage = buffer_migrate_page,
  3238. .is_partially_uptodate = block_is_partially_uptodate,
  3239. };
  3240. void ext4_set_aops(struct inode *inode)
  3241. {
  3242. if (ext4_should_order_data(inode) &&
  3243. test_opt(inode->i_sb, DELALLOC))
  3244. inode->i_mapping->a_ops = &ext4_da_aops;
  3245. else if (ext4_should_order_data(inode))
  3246. inode->i_mapping->a_ops = &ext4_ordered_aops;
  3247. else if (ext4_should_writeback_data(inode) &&
  3248. test_opt(inode->i_sb, DELALLOC))
  3249. inode->i_mapping->a_ops = &ext4_da_aops;
  3250. else if (ext4_should_writeback_data(inode))
  3251. inode->i_mapping->a_ops = &ext4_writeback_aops;
  3252. else
  3253. inode->i_mapping->a_ops = &ext4_journalled_aops;
  3254. }
  3255. /*
  3256. * ext4_block_truncate_page() zeroes out a mapping from file offset `from'
  3257. * up to the end of the block which corresponds to `from'.
  3258. * This required during truncate. We need to physically zero the tail end
  3259. * of that block so it doesn't yield old data if the file is later grown.
  3260. */
  3261. int ext4_block_truncate_page(handle_t *handle,
  3262. struct address_space *mapping, loff_t from)
  3263. {
  3264. ext4_fsblk_t index = from >> PAGE_CACHE_SHIFT;
  3265. unsigned offset = from & (PAGE_CACHE_SIZE-1);
  3266. unsigned blocksize, length, pos;
  3267. ext4_lblk_t iblock;
  3268. struct inode *inode = mapping->host;
  3269. struct buffer_head *bh;
  3270. struct page *page;
  3271. int err = 0;
  3272. page = grab_cache_page(mapping, from >> PAGE_CACHE_SHIFT);
  3273. if (!page)
  3274. return -EINVAL;
  3275. blocksize = inode->i_sb->s_blocksize;
  3276. length = blocksize - (offset & (blocksize - 1));
  3277. iblock = index << (PAGE_CACHE_SHIFT - inode->i_sb->s_blocksize_bits);
  3278. /*
  3279. * For "nobh" option, we can only work if we don't need to
  3280. * read-in the page - otherwise we create buffers to do the IO.
  3281. */
  3282. if (!page_has_buffers(page) && test_opt(inode->i_sb, NOBH) &&
  3283. ext4_should_writeback_data(inode) && PageUptodate(page)) {
  3284. zero_user(page, offset, length);
  3285. set_page_dirty(page);
  3286. goto unlock;
  3287. }
  3288. if (!page_has_buffers(page))
  3289. create_empty_buffers(page, blocksize, 0);
  3290. /* Find the buffer that contains "offset" */
  3291. bh = page_buffers(page);
  3292. pos = blocksize;
  3293. while (offset >= pos) {
  3294. bh = bh->b_this_page;
  3295. iblock++;
  3296. pos += blocksize;
  3297. }
  3298. err = 0;
  3299. if (buffer_freed(bh)) {
  3300. BUFFER_TRACE(bh, "freed: skip");
  3301. goto unlock;
  3302. }
  3303. if (!buffer_mapped(bh)) {
  3304. BUFFER_TRACE(bh, "unmapped");
  3305. ext4_get_block(inode, iblock, bh, 0);
  3306. /* unmapped? It's a hole - nothing to do */
  3307. if (!buffer_mapped(bh)) {
  3308. BUFFER_TRACE(bh, "still unmapped");
  3309. goto unlock;
  3310. }
  3311. }
  3312. /* Ok, it's mapped. Make sure it's up-to-date */
  3313. if (PageUptodate(page))
  3314. set_buffer_uptodate(bh);
  3315. if (!buffer_uptodate(bh)) {
  3316. err = -EIO;
  3317. ll_rw_block(READ, 1, &bh);
  3318. wait_on_buffer(bh);
  3319. /* Uhhuh. Read error. Complain and punt. */
  3320. if (!buffer_uptodate(bh))
  3321. goto unlock;
  3322. }
  3323. if (ext4_should_journal_data(inode)) {
  3324. BUFFER_TRACE(bh, "get write access");
  3325. err = ext4_journal_get_write_access(handle, bh);
  3326. if (err)
  3327. goto unlock;
  3328. }
  3329. zero_user(page, offset, length);
  3330. BUFFER_TRACE(bh, "zeroed end of block");
  3331. err = 0;
  3332. if (ext4_should_journal_data(inode)) {
  3333. err = ext4_handle_dirty_metadata(handle, inode, bh);
  3334. } else {
  3335. if (ext4_should_order_data(inode))
  3336. err = ext4_jbd2_file_inode(handle, inode);
  3337. mark_buffer_dirty(bh);
  3338. }
  3339. unlock:
  3340. unlock_page(page);
  3341. page_cache_release(page);
  3342. return err;
  3343. }
  3344. /*
  3345. * Probably it should be a library function... search for first non-zero word
  3346. * or memcmp with zero_page, whatever is better for particular architecture.
  3347. * Linus?
  3348. */
  3349. static inline int all_zeroes(__le32 *p, __le32 *q)
  3350. {
  3351. while (p < q)
  3352. if (*p++)
  3353. return 0;
  3354. return 1;
  3355. }
  3356. /**
  3357. * ext4_find_shared - find the indirect blocks for partial truncation.
  3358. * @inode: inode in question
  3359. * @depth: depth of the affected branch
  3360. * @offsets: offsets of pointers in that branch (see ext4_block_to_path)
  3361. * @chain: place to store the pointers to partial indirect blocks
  3362. * @top: place to the (detached) top of branch
  3363. *
  3364. * This is a helper function used by ext4_truncate().
  3365. *
  3366. * When we do truncate() we may have to clean the ends of several
  3367. * indirect blocks but leave the blocks themselves alive. Block is
  3368. * partially truncated if some data below the new i_size is refered
  3369. * from it (and it is on the path to the first completely truncated
  3370. * data block, indeed). We have to free the top of that path along
  3371. * with everything to the right of the path. Since no allocation
  3372. * past the truncation point is possible until ext4_truncate()
  3373. * finishes, we may safely do the latter, but top of branch may
  3374. * require special attention - pageout below the truncation point
  3375. * might try to populate it.
  3376. *
  3377. * We atomically detach the top of branch from the tree, store the
  3378. * block number of its root in *@top, pointers to buffer_heads of
  3379. * partially truncated blocks - in @chain[].bh and pointers to
  3380. * their last elements that should not be removed - in
  3381. * @chain[].p. Return value is the pointer to last filled element
  3382. * of @chain.
  3383. *
  3384. * The work left to caller to do the actual freeing of subtrees:
  3385. * a) free the subtree starting from *@top
  3386. * b) free the subtrees whose roots are stored in
  3387. * (@chain[i].p+1 .. end of @chain[i].bh->b_data)
  3388. * c) free the subtrees growing from the inode past the @chain[0].
  3389. * (no partially truncated stuff there). */
  3390. static Indirect *ext4_find_shared(struct inode *inode, int depth,
  3391. ext4_lblk_t offsets[4], Indirect chain[4], __le32 *top)
  3392. {
  3393. Indirect *partial, *p;
  3394. int k, err;
  3395. *top = 0;
  3396. /* Make k index the deepest non-null offest + 1 */
  3397. for (k = depth; k > 1 && !offsets[k-1]; k--)
  3398. ;
  3399. partial = ext4_get_branch(inode, k, offsets, chain, &err);
  3400. /* Writer: pointers */
  3401. if (!partial)
  3402. partial = chain + k-1;
  3403. /*
  3404. * If the branch acquired continuation since we've looked at it -
  3405. * fine, it should all survive and (new) top doesn't belong to us.
  3406. */
  3407. if (!partial->key && *partial->p)
  3408. /* Writer: end */
  3409. goto no_top;
  3410. for (p = partial; (p > chain) && all_zeroes((__le32 *) p->bh->b_data, p->p); p--)
  3411. ;
  3412. /*
  3413. * OK, we've found the last block that must survive. The rest of our
  3414. * branch should be detached before unlocking. However, if that rest
  3415. * of branch is all ours and does not grow immediately from the inode
  3416. * it's easier to cheat and just decrement partial->p.
  3417. */
  3418. if (p == chain + k - 1 && p > chain) {
  3419. p->p--;
  3420. } else {
  3421. *top = *p->p;
  3422. /* Nope, don't do this in ext4. Must leave the tree intact */
  3423. #if 0
  3424. *p->p = 0;
  3425. #endif
  3426. }
  3427. /* Writer: end */
  3428. while (partial > p) {
  3429. brelse(partial->bh);
  3430. partial--;
  3431. }
  3432. no_top:
  3433. return partial;
  3434. }
  3435. /*
  3436. * Zero a number of block pointers in either an inode or an indirect block.
  3437. * If we restart the transaction we must again get write access to the
  3438. * indirect block for further modification.
  3439. *
  3440. * We release `count' blocks on disk, but (last - first) may be greater
  3441. * than `count' because there can be holes in there.
  3442. */
  3443. static void ext4_clear_blocks(handle_t *handle, struct inode *inode,
  3444. struct buffer_head *bh, ext4_fsblk_t block_to_free,
  3445. unsigned long count, __le32 *first, __le32 *last)
  3446. {
  3447. __le32 *p;
  3448. if (try_to_extend_transaction(handle, inode)) {
  3449. if (bh) {
  3450. BUFFER_TRACE(bh, "call ext4_handle_dirty_metadata");
  3451. ext4_handle_dirty_metadata(handle, inode, bh);
  3452. }
  3453. ext4_mark_inode_dirty(handle, inode);
  3454. ext4_journal_test_restart(handle, inode);
  3455. if (bh) {
  3456. BUFFER_TRACE(bh, "retaking write access");
  3457. ext4_journal_get_write_access(handle, bh);
  3458. }
  3459. }
  3460. /*
  3461. * Any buffers which are on the journal will be in memory. We find
  3462. * them on the hash table so jbd2_journal_revoke() will run jbd2_journal_forget()
  3463. * on them. We've already detached each block from the file, so
  3464. * bforget() in jbd2_journal_forget() should be safe.
  3465. *
  3466. * AKPM: turn on bforget in jbd2_journal_forget()!!!
  3467. */
  3468. for (p = first; p < last; p++) {
  3469. u32 nr = le32_to_cpu(*p);
  3470. if (nr) {
  3471. struct buffer_head *tbh;
  3472. *p = 0;
  3473. tbh = sb_find_get_block(inode->i_sb, nr);
  3474. ext4_forget(handle, 0, inode, tbh, nr);
  3475. }
  3476. }
  3477. ext4_free_blocks(handle, inode, block_to_free, count, 0);
  3478. }
  3479. /**
  3480. * ext4_free_data - free a list of data blocks
  3481. * @handle: handle for this transaction
  3482. * @inode: inode we are dealing with
  3483. * @this_bh: indirect buffer_head which contains *@first and *@last
  3484. * @first: array of block numbers
  3485. * @last: points immediately past the end of array
  3486. *
  3487. * We are freeing all blocks refered from that array (numbers are stored as
  3488. * little-endian 32-bit) and updating @inode->i_blocks appropriately.
  3489. *
  3490. * We accumulate contiguous runs of blocks to free. Conveniently, if these
  3491. * blocks are contiguous then releasing them at one time will only affect one
  3492. * or two bitmap blocks (+ group descriptor(s) and superblock) and we won't
  3493. * actually use a lot of journal space.
  3494. *
  3495. * @this_bh will be %NULL if @first and @last point into the inode's direct
  3496. * block pointers.
  3497. */
  3498. static void ext4_free_data(handle_t *handle, struct inode *inode,
  3499. struct buffer_head *this_bh,
  3500. __le32 *first, __le32 *last)
  3501. {
  3502. ext4_fsblk_t block_to_free = 0; /* Starting block # of a run */
  3503. unsigned long count = 0; /* Number of blocks in the run */
  3504. __le32 *block_to_free_p = NULL; /* Pointer into inode/ind
  3505. corresponding to
  3506. block_to_free */
  3507. ext4_fsblk_t nr; /* Current block # */
  3508. __le32 *p; /* Pointer into inode/ind
  3509. for current block */
  3510. int err;
  3511. if (this_bh) { /* For indirect block */
  3512. BUFFER_TRACE(this_bh, "get_write_access");
  3513. err = ext4_journal_get_write_access(handle, this_bh);
  3514. /* Important: if we can't update the indirect pointers
  3515. * to the blocks, we can't free them. */
  3516. if (err)
  3517. return;
  3518. }
  3519. for (p = first; p < last; p++) {
  3520. nr = le32_to_cpu(*p);
  3521. if (nr) {
  3522. /* accumulate blocks to free if they're contiguous */
  3523. if (count == 0) {
  3524. block_to_free = nr;
  3525. block_to_free_p = p;
  3526. count = 1;
  3527. } else if (nr == block_to_free + count) {
  3528. count++;
  3529. } else {
  3530. ext4_clear_blocks(handle, inode, this_bh,
  3531. block_to_free,
  3532. count, block_to_free_p, p);
  3533. block_to_free = nr;
  3534. block_to_free_p = p;
  3535. count = 1;
  3536. }
  3537. }
  3538. }
  3539. if (count > 0)
  3540. ext4_clear_blocks(handle, inode, this_bh, block_to_free,
  3541. count, block_to_free_p, p);
  3542. if (this_bh) {
  3543. BUFFER_TRACE(this_bh, "call ext4_handle_dirty_metadata");
  3544. /*
  3545. * The buffer head should have an attached journal head at this
  3546. * point. However, if the data is corrupted and an indirect
  3547. * block pointed to itself, it would have been detached when
  3548. * the block was cleared. Check for this instead of OOPSing.
  3549. */
  3550. if ((EXT4_JOURNAL(inode) == NULL) || bh2jh(this_bh))
  3551. ext4_handle_dirty_metadata(handle, inode, this_bh);
  3552. else
  3553. ext4_error(inode->i_sb, __func__,
  3554. "circular indirect block detected, "
  3555. "inode=%lu, block=%llu",
  3556. inode->i_ino,
  3557. (unsigned long long) this_bh->b_blocknr);
  3558. }
  3559. }
  3560. /**
  3561. * ext4_free_branches - free an array of branches
  3562. * @handle: JBD handle for this transaction
  3563. * @inode: inode we are dealing with
  3564. * @parent_bh: the buffer_head which contains *@first and *@last
  3565. * @first: array of block numbers
  3566. * @last: pointer immediately past the end of array
  3567. * @depth: depth of the branches to free
  3568. *
  3569. * We are freeing all blocks refered from these branches (numbers are
  3570. * stored as little-endian 32-bit) and updating @inode->i_blocks
  3571. * appropriately.
  3572. */
  3573. static void ext4_free_branches(handle_t *handle, struct inode *inode,
  3574. struct buffer_head *parent_bh,
  3575. __le32 *first, __le32 *last, int depth)
  3576. {
  3577. ext4_fsblk_t nr;
  3578. __le32 *p;
  3579. if (ext4_handle_is_aborted(handle))
  3580. return;
  3581. if (depth--) {
  3582. struct buffer_head *bh;
  3583. int addr_per_block = EXT4_ADDR_PER_BLOCK(inode->i_sb);
  3584. p = last;
  3585. while (--p >= first) {
  3586. nr = le32_to_cpu(*p);
  3587. if (!nr)
  3588. continue; /* A hole */
  3589. /* Go read the buffer for the next level down */
  3590. bh = sb_bread(inode->i_sb, nr);
  3591. /*
  3592. * A read failure? Report error and clear slot
  3593. * (should be rare).
  3594. */
  3595. if (!bh) {
  3596. ext4_error(inode->i_sb, "ext4_free_branches",
  3597. "Read failure, inode=%lu, block=%llu",
  3598. inode->i_ino, nr);
  3599. continue;
  3600. }
  3601. /* This zaps the entire block. Bottom up. */
  3602. BUFFER_TRACE(bh, "free child branches");
  3603. ext4_free_branches(handle, inode, bh,
  3604. (__le32 *) bh->b_data,
  3605. (__le32 *) bh->b_data + addr_per_block,
  3606. depth);
  3607. /*
  3608. * We've probably journalled the indirect block several
  3609. * times during the truncate. But it's no longer
  3610. * needed and we now drop it from the transaction via
  3611. * jbd2_journal_revoke().
  3612. *
  3613. * That's easy if it's exclusively part of this
  3614. * transaction. But if it's part of the committing
  3615. * transaction then jbd2_journal_forget() will simply
  3616. * brelse() it. That means that if the underlying
  3617. * block is reallocated in ext4_get_block(),
  3618. * unmap_underlying_metadata() will find this block
  3619. * and will try to get rid of it. damn, damn.
  3620. *
  3621. * If this block has already been committed to the
  3622. * journal, a revoke record will be written. And
  3623. * revoke records must be emitted *before* clearing
  3624. * this block's bit in the bitmaps.
  3625. */
  3626. ext4_forget(handle, 1, inode, bh, bh->b_blocknr);
  3627. /*
  3628. * Everything below this this pointer has been
  3629. * released. Now let this top-of-subtree go.
  3630. *
  3631. * We want the freeing of this indirect block to be
  3632. * atomic in the journal with the updating of the
  3633. * bitmap block which owns it. So make some room in
  3634. * the journal.
  3635. *
  3636. * We zero the parent pointer *after* freeing its
  3637. * pointee in the bitmaps, so if extend_transaction()
  3638. * for some reason fails to put the bitmap changes and
  3639. * the release into the same transaction, recovery
  3640. * will merely complain about releasing a free block,
  3641. * rather than leaking blocks.
  3642. */
  3643. if (ext4_handle_is_aborted(handle))
  3644. return;
  3645. if (try_to_extend_transaction(handle, inode)) {
  3646. ext4_mark_inode_dirty(handle, inode);
  3647. ext4_journal_test_restart(handle, inode);
  3648. }
  3649. ext4_free_blocks(handle, inode, nr, 1, 1);
  3650. if (parent_bh) {
  3651. /*
  3652. * The block which we have just freed is
  3653. * pointed to by an indirect block: journal it
  3654. */
  3655. BUFFER_TRACE(parent_bh, "get_write_access");
  3656. if (!ext4_journal_get_write_access(handle,
  3657. parent_bh)){
  3658. *p = 0;
  3659. BUFFER_TRACE(parent_bh,
  3660. "call ext4_handle_dirty_metadata");
  3661. ext4_handle_dirty_metadata(handle,
  3662. inode,
  3663. parent_bh);
  3664. }
  3665. }
  3666. }
  3667. } else {
  3668. /* We have reached the bottom of the tree. */
  3669. BUFFER_TRACE(parent_bh, "free data blocks");
  3670. ext4_free_data(handle, inode, parent_bh, first, last);
  3671. }
  3672. }
  3673. int ext4_can_truncate(struct inode *inode)
  3674. {
  3675. if (IS_APPEND(inode) || IS_IMMUTABLE(inode))
  3676. return 0;
  3677. if (S_ISREG(inode->i_mode))
  3678. return 1;
  3679. if (S_ISDIR(inode->i_mode))
  3680. return 1;
  3681. if (S_ISLNK(inode->i_mode))
  3682. return !ext4_inode_is_fast_symlink(inode);
  3683. return 0;
  3684. }
  3685. /*
  3686. * ext4_truncate()
  3687. *
  3688. * We block out ext4_get_block() block instantiations across the entire
  3689. * transaction, and VFS/VM ensures that ext4_truncate() cannot run
  3690. * simultaneously on behalf of the same inode.
  3691. *
  3692. * As we work through the truncate and commmit bits of it to the journal there
  3693. * is one core, guiding principle: the file's tree must always be consistent on
  3694. * disk. We must be able to restart the truncate after a crash.
  3695. *
  3696. * The file's tree may be transiently inconsistent in memory (although it
  3697. * probably isn't), but whenever we close off and commit a journal transaction,
  3698. * the contents of (the filesystem + the journal) must be consistent and
  3699. * restartable. It's pretty simple, really: bottom up, right to left (although
  3700. * left-to-right works OK too).
  3701. *
  3702. * Note that at recovery time, journal replay occurs *before* the restart of
  3703. * truncate against the orphan inode list.
  3704. *
  3705. * The committed inode has the new, desired i_size (which is the same as
  3706. * i_disksize in this case). After a crash, ext4_orphan_cleanup() will see
  3707. * that this inode's truncate did not complete and it will again call
  3708. * ext4_truncate() to have another go. So there will be instantiated blocks
  3709. * to the right of the truncation point in a crashed ext4 filesystem. But
  3710. * that's fine - as long as they are linked from the inode, the post-crash
  3711. * ext4_truncate() run will find them and release them.
  3712. */
  3713. void ext4_truncate(struct inode *inode)
  3714. {
  3715. handle_t *handle;
  3716. struct ext4_inode_info *ei = EXT4_I(inode);
  3717. __le32 *i_data = ei->i_data;
  3718. int addr_per_block = EXT4_ADDR_PER_BLOCK(inode->i_sb);
  3719. struct address_space *mapping = inode->i_mapping;
  3720. ext4_lblk_t offsets[4];
  3721. Indirect chain[4];
  3722. Indirect *partial;
  3723. __le32 nr = 0;
  3724. int n;
  3725. ext4_lblk_t last_block;
  3726. unsigned blocksize = inode->i_sb->s_blocksize;
  3727. if (!ext4_can_truncate(inode))
  3728. return;
  3729. if (ei->i_disksize && inode->i_size == 0 &&
  3730. !test_opt(inode->i_sb, NO_AUTO_DA_ALLOC))
  3731. ei->i_state |= EXT4_STATE_DA_ALLOC_CLOSE;
  3732. if (EXT4_I(inode)->i_flags & EXT4_EXTENTS_FL) {
  3733. ext4_ext_truncate(inode);
  3734. return;
  3735. }
  3736. handle = start_transaction(inode);
  3737. if (IS_ERR(handle))
  3738. return; /* AKPM: return what? */
  3739. last_block = (inode->i_size + blocksize-1)
  3740. >> EXT4_BLOCK_SIZE_BITS(inode->i_sb);
  3741. if (inode->i_size & (blocksize - 1))
  3742. if (ext4_block_truncate_page(handle, mapping, inode->i_size))
  3743. goto out_stop;
  3744. n = ext4_block_to_path(inode, last_block, offsets, NULL);
  3745. if (n == 0)
  3746. goto out_stop; /* error */
  3747. /*
  3748. * OK. This truncate is going to happen. We add the inode to the
  3749. * orphan list, so that if this truncate spans multiple transactions,
  3750. * and we crash, we will resume the truncate when the filesystem
  3751. * recovers. It also marks the inode dirty, to catch the new size.
  3752. *
  3753. * Implication: the file must always be in a sane, consistent
  3754. * truncatable state while each transaction commits.
  3755. */
  3756. if (ext4_orphan_add(handle, inode))
  3757. goto out_stop;
  3758. /*
  3759. * From here we block out all ext4_get_block() callers who want to
  3760. * modify the block allocation tree.
  3761. */
  3762. down_write(&ei->i_data_sem);
  3763. ext4_discard_preallocations(inode);
  3764. /*
  3765. * The orphan list entry will now protect us from any crash which
  3766. * occurs before the truncate completes, so it is now safe to propagate
  3767. * the new, shorter inode size (held for now in i_size) into the
  3768. * on-disk inode. We do this via i_disksize, which is the value which
  3769. * ext4 *really* writes onto the disk inode.
  3770. */
  3771. ei->i_disksize = inode->i_size;
  3772. if (n == 1) { /* direct blocks */
  3773. ext4_free_data(handle, inode, NULL, i_data+offsets[0],
  3774. i_data + EXT4_NDIR_BLOCKS);
  3775. goto do_indirects;
  3776. }
  3777. partial = ext4_find_shared(inode, n, offsets, chain, &nr);
  3778. /* Kill the top of shared branch (not detached) */
  3779. if (nr) {
  3780. if (partial == chain) {
  3781. /* Shared branch grows from the inode */
  3782. ext4_free_branches(handle, inode, NULL,
  3783. &nr, &nr+1, (chain+n-1) - partial);
  3784. *partial->p = 0;
  3785. /*
  3786. * We mark the inode dirty prior to restart,
  3787. * and prior to stop. No need for it here.
  3788. */
  3789. } else {
  3790. /* Shared branch grows from an indirect block */
  3791. BUFFER_TRACE(partial->bh, "get_write_access");
  3792. ext4_free_branches(handle, inode, partial->bh,
  3793. partial->p,
  3794. partial->p+1, (chain+n-1) - partial);
  3795. }
  3796. }
  3797. /* Clear the ends of indirect blocks on the shared branch */
  3798. while (partial > chain) {
  3799. ext4_free_branches(handle, inode, partial->bh, partial->p + 1,
  3800. (__le32*)partial->bh->b_data+addr_per_block,
  3801. (chain+n-1) - partial);
  3802. BUFFER_TRACE(partial->bh, "call brelse");
  3803. brelse (partial->bh);
  3804. partial--;
  3805. }
  3806. do_indirects:
  3807. /* Kill the remaining (whole) subtrees */
  3808. switch (offsets[0]) {
  3809. default:
  3810. nr = i_data[EXT4_IND_BLOCK];
  3811. if (nr) {
  3812. ext4_free_branches(handle, inode, NULL, &nr, &nr+1, 1);
  3813. i_data[EXT4_IND_BLOCK] = 0;
  3814. }
  3815. case EXT4_IND_BLOCK:
  3816. nr = i_data[EXT4_DIND_BLOCK];
  3817. if (nr) {
  3818. ext4_free_branches(handle, inode, NULL, &nr, &nr+1, 2);
  3819. i_data[EXT4_DIND_BLOCK] = 0;
  3820. }
  3821. case EXT4_DIND_BLOCK:
  3822. nr = i_data[EXT4_TIND_BLOCK];
  3823. if (nr) {
  3824. ext4_free_branches(handle, inode, NULL, &nr, &nr+1, 3);
  3825. i_data[EXT4_TIND_BLOCK] = 0;
  3826. }
  3827. case EXT4_TIND_BLOCK:
  3828. ;
  3829. }
  3830. up_write(&ei->i_data_sem);
  3831. inode->i_mtime = inode->i_ctime = ext4_current_time(inode);
  3832. ext4_mark_inode_dirty(handle, inode);
  3833. /*
  3834. * In a multi-transaction truncate, we only make the final transaction
  3835. * synchronous
  3836. */
  3837. if (IS_SYNC(inode))
  3838. ext4_handle_sync(handle);
  3839. out_stop:
  3840. /*
  3841. * If this was a simple ftruncate(), and the file will remain alive
  3842. * then we need to clear up the orphan record which we created above.
  3843. * However, if this was a real unlink then we were called by
  3844. * ext4_delete_inode(), and we allow that function to clean up the
  3845. * orphan info for us.
  3846. */
  3847. if (inode->i_nlink)
  3848. ext4_orphan_del(handle, inode);
  3849. ext4_journal_stop(handle);
  3850. }
  3851. /*
  3852. * ext4_get_inode_loc returns with an extra refcount against the inode's
  3853. * underlying buffer_head on success. If 'in_mem' is true, we have all
  3854. * data in memory that is needed to recreate the on-disk version of this
  3855. * inode.
  3856. */
  3857. static int __ext4_get_inode_loc(struct inode *inode,
  3858. struct ext4_iloc *iloc, int in_mem)
  3859. {
  3860. struct ext4_group_desc *gdp;
  3861. struct buffer_head *bh;
  3862. struct super_block *sb = inode->i_sb;
  3863. ext4_fsblk_t block;
  3864. int inodes_per_block, inode_offset;
  3865. iloc->bh = NULL;
  3866. if (!ext4_valid_inum(sb, inode->i_ino))
  3867. return -EIO;
  3868. iloc->block_group = (inode->i_ino - 1) / EXT4_INODES_PER_GROUP(sb);
  3869. gdp = ext4_get_group_desc(sb, iloc->block_group, NULL);
  3870. if (!gdp)
  3871. return -EIO;
  3872. /*
  3873. * Figure out the offset within the block group inode table
  3874. */
  3875. inodes_per_block = (EXT4_BLOCK_SIZE(sb) / EXT4_INODE_SIZE(sb));
  3876. inode_offset = ((inode->i_ino - 1) %
  3877. EXT4_INODES_PER_GROUP(sb));
  3878. block = ext4_inode_table(sb, gdp) + (inode_offset / inodes_per_block);
  3879. iloc->offset = (inode_offset % inodes_per_block) * EXT4_INODE_SIZE(sb);
  3880. bh = sb_getblk(sb, block);
  3881. if (!bh) {
  3882. ext4_error(sb, "ext4_get_inode_loc", "unable to read "
  3883. "inode block - inode=%lu, block=%llu",
  3884. inode->i_ino, block);
  3885. return -EIO;
  3886. }
  3887. if (!buffer_uptodate(bh)) {
  3888. lock_buffer(bh);
  3889. /*
  3890. * If the buffer has the write error flag, we have failed
  3891. * to write out another inode in the same block. In this
  3892. * case, we don't have to read the block because we may
  3893. * read the old inode data successfully.
  3894. */
  3895. if (buffer_write_io_error(bh) && !buffer_uptodate(bh))
  3896. set_buffer_uptodate(bh);
  3897. if (buffer_uptodate(bh)) {
  3898. /* someone brought it uptodate while we waited */
  3899. unlock_buffer(bh);
  3900. goto has_buffer;
  3901. }
  3902. /*
  3903. * If we have all information of the inode in memory and this
  3904. * is the only valid inode in the block, we need not read the
  3905. * block.
  3906. */
  3907. if (in_mem) {
  3908. struct buffer_head *bitmap_bh;
  3909. int i, start;
  3910. start = inode_offset & ~(inodes_per_block - 1);
  3911. /* Is the inode bitmap in cache? */
  3912. bitmap_bh = sb_getblk(sb, ext4_inode_bitmap(sb, gdp));
  3913. if (!bitmap_bh)
  3914. goto make_io;
  3915. /*
  3916. * If the inode bitmap isn't in cache then the
  3917. * optimisation may end up performing two reads instead
  3918. * of one, so skip it.
  3919. */
  3920. if (!buffer_uptodate(bitmap_bh)) {
  3921. brelse(bitmap_bh);
  3922. goto make_io;
  3923. }
  3924. for (i = start; i < start + inodes_per_block; i++) {
  3925. if (i == inode_offset)
  3926. continue;
  3927. if (ext4_test_bit(i, bitmap_bh->b_data))
  3928. break;
  3929. }
  3930. brelse(bitmap_bh);
  3931. if (i == start + inodes_per_block) {
  3932. /* all other inodes are free, so skip I/O */
  3933. memset(bh->b_data, 0, bh->b_size);
  3934. set_buffer_uptodate(bh);
  3935. unlock_buffer(bh);
  3936. goto has_buffer;
  3937. }
  3938. }
  3939. make_io:
  3940. /*
  3941. * If we need to do any I/O, try to pre-readahead extra
  3942. * blocks from the inode table.
  3943. */
  3944. if (EXT4_SB(sb)->s_inode_readahead_blks) {
  3945. ext4_fsblk_t b, end, table;
  3946. unsigned num;
  3947. table = ext4_inode_table(sb, gdp);
  3948. /* s_inode_readahead_blks is always a power of 2 */
  3949. b = block & ~(EXT4_SB(sb)->s_inode_readahead_blks-1);
  3950. if (table > b)
  3951. b = table;
  3952. end = b + EXT4_SB(sb)->s_inode_readahead_blks;
  3953. num = EXT4_INODES_PER_GROUP(sb);
  3954. if (EXT4_HAS_RO_COMPAT_FEATURE(sb,
  3955. EXT4_FEATURE_RO_COMPAT_GDT_CSUM))
  3956. num -= ext4_itable_unused_count(sb, gdp);
  3957. table += num / inodes_per_block;
  3958. if (end > table)
  3959. end = table;
  3960. while (b <= end)
  3961. sb_breadahead(sb, b++);
  3962. }
  3963. /*
  3964. * There are other valid inodes in the buffer, this inode
  3965. * has in-inode xattrs, or we don't have this inode in memory.
  3966. * Read the block from disk.
  3967. */
  3968. get_bh(bh);
  3969. bh->b_end_io = end_buffer_read_sync;
  3970. submit_bh(READ_META, bh);
  3971. wait_on_buffer(bh);
  3972. if (!buffer_uptodate(bh)) {
  3973. ext4_error(sb, __func__,
  3974. "unable to read inode block - inode=%lu, "
  3975. "block=%llu", inode->i_ino, block);
  3976. brelse(bh);
  3977. return -EIO;
  3978. }
  3979. }
  3980. has_buffer:
  3981. iloc->bh = bh;
  3982. return 0;
  3983. }
  3984. int ext4_get_inode_loc(struct inode *inode, struct ext4_iloc *iloc)
  3985. {
  3986. /* We have all inode data except xattrs in memory here. */
  3987. return __ext4_get_inode_loc(inode, iloc,
  3988. !(EXT4_I(inode)->i_state & EXT4_STATE_XATTR));
  3989. }
  3990. void ext4_set_inode_flags(struct inode *inode)
  3991. {
  3992. unsigned int flags = EXT4_I(inode)->i_flags;
  3993. inode->i_flags &= ~(S_SYNC|S_APPEND|S_IMMUTABLE|S_NOATIME|S_DIRSYNC);
  3994. if (flags & EXT4_SYNC_FL)
  3995. inode->i_flags |= S_SYNC;
  3996. if (flags & EXT4_APPEND_FL)
  3997. inode->i_flags |= S_APPEND;
  3998. if (flags & EXT4_IMMUTABLE_FL)
  3999. inode->i_flags |= S_IMMUTABLE;
  4000. if (flags & EXT4_NOATIME_FL)
  4001. inode->i_flags |= S_NOATIME;
  4002. if (flags & EXT4_DIRSYNC_FL)
  4003. inode->i_flags |= S_DIRSYNC;
  4004. }
  4005. /* Propagate flags from i_flags to EXT4_I(inode)->i_flags */
  4006. void ext4_get_inode_flags(struct ext4_inode_info *ei)
  4007. {
  4008. unsigned int flags = ei->vfs_inode.i_flags;
  4009. ei->i_flags &= ~(EXT4_SYNC_FL|EXT4_APPEND_FL|
  4010. EXT4_IMMUTABLE_FL|EXT4_NOATIME_FL|EXT4_DIRSYNC_FL);
  4011. if (flags & S_SYNC)
  4012. ei->i_flags |= EXT4_SYNC_FL;
  4013. if (flags & S_APPEND)
  4014. ei->i_flags |= EXT4_APPEND_FL;
  4015. if (flags & S_IMMUTABLE)
  4016. ei->i_flags |= EXT4_IMMUTABLE_FL;
  4017. if (flags & S_NOATIME)
  4018. ei->i_flags |= EXT4_NOATIME_FL;
  4019. if (flags & S_DIRSYNC)
  4020. ei->i_flags |= EXT4_DIRSYNC_FL;
  4021. }
  4022. static blkcnt_t ext4_inode_blocks(struct ext4_inode *raw_inode,
  4023. struct ext4_inode_info *ei)
  4024. {
  4025. blkcnt_t i_blocks ;
  4026. struct inode *inode = &(ei->vfs_inode);
  4027. struct super_block *sb = inode->i_sb;
  4028. if (EXT4_HAS_RO_COMPAT_FEATURE(sb,
  4029. EXT4_FEATURE_RO_COMPAT_HUGE_FILE)) {
  4030. /* we are using combined 48 bit field */
  4031. i_blocks = ((u64)le16_to_cpu(raw_inode->i_blocks_high)) << 32 |
  4032. le32_to_cpu(raw_inode->i_blocks_lo);
  4033. if (ei->i_flags & EXT4_HUGE_FILE_FL) {
  4034. /* i_blocks represent file system block size */
  4035. return i_blocks << (inode->i_blkbits - 9);
  4036. } else {
  4037. return i_blocks;
  4038. }
  4039. } else {
  4040. return le32_to_cpu(raw_inode->i_blocks_lo);
  4041. }
  4042. }
  4043. struct inode *ext4_iget(struct super_block *sb, unsigned long ino)
  4044. {
  4045. struct ext4_iloc iloc;
  4046. struct ext4_inode *raw_inode;
  4047. struct ext4_inode_info *ei;
  4048. struct buffer_head *bh;
  4049. struct inode *inode;
  4050. long ret;
  4051. int block;
  4052. inode = iget_locked(sb, ino);
  4053. if (!inode)
  4054. return ERR_PTR(-ENOMEM);
  4055. if (!(inode->i_state & I_NEW))
  4056. return inode;
  4057. ei = EXT4_I(inode);
  4058. #ifdef CONFIG_EXT4_FS_POSIX_ACL
  4059. ei->i_acl = EXT4_ACL_NOT_CACHED;
  4060. ei->i_default_acl = EXT4_ACL_NOT_CACHED;
  4061. #endif
  4062. ret = __ext4_get_inode_loc(inode, &iloc, 0);
  4063. if (ret < 0)
  4064. goto bad_inode;
  4065. bh = iloc.bh;
  4066. raw_inode = ext4_raw_inode(&iloc);
  4067. inode->i_mode = le16_to_cpu(raw_inode->i_mode);
  4068. inode->i_uid = (uid_t)le16_to_cpu(raw_inode->i_uid_low);
  4069. inode->i_gid = (gid_t)le16_to_cpu(raw_inode->i_gid_low);
  4070. if (!(test_opt(inode->i_sb, NO_UID32))) {
  4071. inode->i_uid |= le16_to_cpu(raw_inode->i_uid_high) << 16;
  4072. inode->i_gid |= le16_to_cpu(raw_inode->i_gid_high) << 16;
  4073. }
  4074. inode->i_nlink = le16_to_cpu(raw_inode->i_links_count);
  4075. ei->i_state = 0;
  4076. ei->i_dir_start_lookup = 0;
  4077. ei->i_dtime = le32_to_cpu(raw_inode->i_dtime);
  4078. /* We now have enough fields to check if the inode was active or not.
  4079. * This is needed because nfsd might try to access dead inodes
  4080. * the test is that same one that e2fsck uses
  4081. * NeilBrown 1999oct15
  4082. */
  4083. if (inode->i_nlink == 0) {
  4084. if (inode->i_mode == 0 ||
  4085. !(EXT4_SB(inode->i_sb)->s_mount_state & EXT4_ORPHAN_FS)) {
  4086. /* this inode is deleted */
  4087. brelse(bh);
  4088. ret = -ESTALE;
  4089. goto bad_inode;
  4090. }
  4091. /* The only unlinked inodes we let through here have
  4092. * valid i_mode and are being read by the orphan
  4093. * recovery code: that's fine, we're about to complete
  4094. * the process of deleting those. */
  4095. }
  4096. ei->i_flags = le32_to_cpu(raw_inode->i_flags);
  4097. inode->i_blocks = ext4_inode_blocks(raw_inode, ei);
  4098. ei->i_file_acl = le32_to_cpu(raw_inode->i_file_acl_lo);
  4099. if (EXT4_HAS_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_64BIT))
  4100. ei->i_file_acl |=
  4101. ((__u64)le16_to_cpu(raw_inode->i_file_acl_high)) << 32;
  4102. inode->i_size = ext4_isize(raw_inode);
  4103. ei->i_disksize = inode->i_size;
  4104. inode->i_generation = le32_to_cpu(raw_inode->i_generation);
  4105. ei->i_block_group = iloc.block_group;
  4106. ei->i_last_alloc_group = ~0;
  4107. /*
  4108. * NOTE! The in-memory inode i_data array is in little-endian order
  4109. * even on big-endian machines: we do NOT byteswap the block numbers!
  4110. */
  4111. for (block = 0; block < EXT4_N_BLOCKS; block++)
  4112. ei->i_data[block] = raw_inode->i_block[block];
  4113. INIT_LIST_HEAD(&ei->i_orphan);
  4114. if (EXT4_INODE_SIZE(inode->i_sb) > EXT4_GOOD_OLD_INODE_SIZE) {
  4115. ei->i_extra_isize = le16_to_cpu(raw_inode->i_extra_isize);
  4116. if (EXT4_GOOD_OLD_INODE_SIZE + ei->i_extra_isize >
  4117. EXT4_INODE_SIZE(inode->i_sb)) {
  4118. brelse(bh);
  4119. ret = -EIO;
  4120. goto bad_inode;
  4121. }
  4122. if (ei->i_extra_isize == 0) {
  4123. /* The extra space is currently unused. Use it. */
  4124. ei->i_extra_isize = sizeof(struct ext4_inode) -
  4125. EXT4_GOOD_OLD_INODE_SIZE;
  4126. } else {
  4127. __le32 *magic = (void *)raw_inode +
  4128. EXT4_GOOD_OLD_INODE_SIZE +
  4129. ei->i_extra_isize;
  4130. if (*magic == cpu_to_le32(EXT4_XATTR_MAGIC))
  4131. ei->i_state |= EXT4_STATE_XATTR;
  4132. }
  4133. } else
  4134. ei->i_extra_isize = 0;
  4135. EXT4_INODE_GET_XTIME(i_ctime, inode, raw_inode);
  4136. EXT4_INODE_GET_XTIME(i_mtime, inode, raw_inode);
  4137. EXT4_INODE_GET_XTIME(i_atime, inode, raw_inode);
  4138. EXT4_EINODE_GET_XTIME(i_crtime, ei, raw_inode);
  4139. inode->i_version = le32_to_cpu(raw_inode->i_disk_version);
  4140. if (EXT4_INODE_SIZE(inode->i_sb) > EXT4_GOOD_OLD_INODE_SIZE) {
  4141. if (EXT4_FITS_IN_INODE(raw_inode, ei, i_version_hi))
  4142. inode->i_version |=
  4143. (__u64)(le32_to_cpu(raw_inode->i_version_hi)) << 32;
  4144. }
  4145. ret = 0;
  4146. if (ei->i_file_acl &&
  4147. ((ei->i_file_acl <
  4148. (le32_to_cpu(EXT4_SB(sb)->s_es->s_first_data_block) +
  4149. EXT4_SB(sb)->s_gdb_count)) ||
  4150. (ei->i_file_acl >= ext4_blocks_count(EXT4_SB(sb)->s_es)))) {
  4151. ext4_error(sb, __func__,
  4152. "bad extended attribute block %llu in inode #%lu",
  4153. ei->i_file_acl, inode->i_ino);
  4154. ret = -EIO;
  4155. goto bad_inode;
  4156. } else if (ei->i_flags & EXT4_EXTENTS_FL) {
  4157. if (S_ISREG(inode->i_mode) || S_ISDIR(inode->i_mode) ||
  4158. (S_ISLNK(inode->i_mode) &&
  4159. !ext4_inode_is_fast_symlink(inode)))
  4160. /* Validate extent which is part of inode */
  4161. ret = ext4_ext_check_inode(inode);
  4162. } else if (S_ISREG(inode->i_mode) || S_ISDIR(inode->i_mode) ||
  4163. (S_ISLNK(inode->i_mode) &&
  4164. !ext4_inode_is_fast_symlink(inode))) {
  4165. /* Validate block references which are part of inode */
  4166. ret = ext4_check_inode_blockref(inode);
  4167. }
  4168. if (ret) {
  4169. brelse(bh);
  4170. goto bad_inode;
  4171. }
  4172. if (S_ISREG(inode->i_mode)) {
  4173. inode->i_op = &ext4_file_inode_operations;
  4174. inode->i_fop = &ext4_file_operations;
  4175. ext4_set_aops(inode);
  4176. } else if (S_ISDIR(inode->i_mode)) {
  4177. inode->i_op = &ext4_dir_inode_operations;
  4178. inode->i_fop = &ext4_dir_operations;
  4179. } else if (S_ISLNK(inode->i_mode)) {
  4180. if (ext4_inode_is_fast_symlink(inode)) {
  4181. inode->i_op = &ext4_fast_symlink_inode_operations;
  4182. nd_terminate_link(ei->i_data, inode->i_size,
  4183. sizeof(ei->i_data) - 1);
  4184. } else {
  4185. inode->i_op = &ext4_symlink_inode_operations;
  4186. ext4_set_aops(inode);
  4187. }
  4188. } else if (S_ISCHR(inode->i_mode) || S_ISBLK(inode->i_mode) ||
  4189. S_ISFIFO(inode->i_mode) || S_ISSOCK(inode->i_mode)) {
  4190. inode->i_op = &ext4_special_inode_operations;
  4191. if (raw_inode->i_block[0])
  4192. init_special_inode(inode, inode->i_mode,
  4193. old_decode_dev(le32_to_cpu(raw_inode->i_block[0])));
  4194. else
  4195. init_special_inode(inode, inode->i_mode,
  4196. new_decode_dev(le32_to_cpu(raw_inode->i_block[1])));
  4197. } else {
  4198. brelse(bh);
  4199. ret = -EIO;
  4200. ext4_error(inode->i_sb, __func__,
  4201. "bogus i_mode (%o) for inode=%lu",
  4202. inode->i_mode, inode->i_ino);
  4203. goto bad_inode;
  4204. }
  4205. brelse(iloc.bh);
  4206. ext4_set_inode_flags(inode);
  4207. unlock_new_inode(inode);
  4208. return inode;
  4209. bad_inode:
  4210. iget_failed(inode);
  4211. return ERR_PTR(ret);
  4212. }
  4213. static int ext4_inode_blocks_set(handle_t *handle,
  4214. struct ext4_inode *raw_inode,
  4215. struct ext4_inode_info *ei)
  4216. {
  4217. struct inode *inode = &(ei->vfs_inode);
  4218. u64 i_blocks = inode->i_blocks;
  4219. struct super_block *sb = inode->i_sb;
  4220. if (i_blocks <= ~0U) {
  4221. /*
  4222. * i_blocks can be represnted in a 32 bit variable
  4223. * as multiple of 512 bytes
  4224. */
  4225. raw_inode->i_blocks_lo = cpu_to_le32(i_blocks);
  4226. raw_inode->i_blocks_high = 0;
  4227. ei->i_flags &= ~EXT4_HUGE_FILE_FL;
  4228. return 0;
  4229. }
  4230. if (!EXT4_HAS_RO_COMPAT_FEATURE(sb, EXT4_FEATURE_RO_COMPAT_HUGE_FILE))
  4231. return -EFBIG;
  4232. if (i_blocks <= 0xffffffffffffULL) {
  4233. /*
  4234. * i_blocks can be represented in a 48 bit variable
  4235. * as multiple of 512 bytes
  4236. */
  4237. raw_inode->i_blocks_lo = cpu_to_le32(i_blocks);
  4238. raw_inode->i_blocks_high = cpu_to_le16(i_blocks >> 32);
  4239. ei->i_flags &= ~EXT4_HUGE_FILE_FL;
  4240. } else {
  4241. ei->i_flags |= EXT4_HUGE_FILE_FL;
  4242. /* i_block is stored in file system block size */
  4243. i_blocks = i_blocks >> (inode->i_blkbits - 9);
  4244. raw_inode->i_blocks_lo = cpu_to_le32(i_blocks);
  4245. raw_inode->i_blocks_high = cpu_to_le16(i_blocks >> 32);
  4246. }
  4247. return 0;
  4248. }
  4249. /*
  4250. * Post the struct inode info into an on-disk inode location in the
  4251. * buffer-cache. This gobbles the caller's reference to the
  4252. * buffer_head in the inode location struct.
  4253. *
  4254. * The caller must have write access to iloc->bh.
  4255. */
  4256. static int ext4_do_update_inode(handle_t *handle,
  4257. struct inode *inode,
  4258. struct ext4_iloc *iloc)
  4259. {
  4260. struct ext4_inode *raw_inode = ext4_raw_inode(iloc);
  4261. struct ext4_inode_info *ei = EXT4_I(inode);
  4262. struct buffer_head *bh = iloc->bh;
  4263. int err = 0, rc, block;
  4264. /* For fields not not tracking in the in-memory inode,
  4265. * initialise them to zero for new inodes. */
  4266. if (ei->i_state & EXT4_STATE_NEW)
  4267. memset(raw_inode, 0, EXT4_SB(inode->i_sb)->s_inode_size);
  4268. ext4_get_inode_flags(ei);
  4269. raw_inode->i_mode = cpu_to_le16(inode->i_mode);
  4270. if (!(test_opt(inode->i_sb, NO_UID32))) {
  4271. raw_inode->i_uid_low = cpu_to_le16(low_16_bits(inode->i_uid));
  4272. raw_inode->i_gid_low = cpu_to_le16(low_16_bits(inode->i_gid));
  4273. /*
  4274. * Fix up interoperability with old kernels. Otherwise, old inodes get
  4275. * re-used with the upper 16 bits of the uid/gid intact
  4276. */
  4277. if (!ei->i_dtime) {
  4278. raw_inode->i_uid_high =
  4279. cpu_to_le16(high_16_bits(inode->i_uid));
  4280. raw_inode->i_gid_high =
  4281. cpu_to_le16(high_16_bits(inode->i_gid));
  4282. } else {
  4283. raw_inode->i_uid_high = 0;
  4284. raw_inode->i_gid_high = 0;
  4285. }
  4286. } else {
  4287. raw_inode->i_uid_low =
  4288. cpu_to_le16(fs_high2lowuid(inode->i_uid));
  4289. raw_inode->i_gid_low =
  4290. cpu_to_le16(fs_high2lowgid(inode->i_gid));
  4291. raw_inode->i_uid_high = 0;
  4292. raw_inode->i_gid_high = 0;
  4293. }
  4294. raw_inode->i_links_count = cpu_to_le16(inode->i_nlink);
  4295. EXT4_INODE_SET_XTIME(i_ctime, inode, raw_inode);
  4296. EXT4_INODE_SET_XTIME(i_mtime, inode, raw_inode);
  4297. EXT4_INODE_SET_XTIME(i_atime, inode, raw_inode);
  4298. EXT4_EINODE_SET_XTIME(i_crtime, ei, raw_inode);
  4299. if (ext4_inode_blocks_set(handle, raw_inode, ei))
  4300. goto out_brelse;
  4301. raw_inode->i_dtime = cpu_to_le32(ei->i_dtime);
  4302. /* clear the migrate flag in the raw_inode */
  4303. raw_inode->i_flags = cpu_to_le32(ei->i_flags & ~EXT4_EXT_MIGRATE);
  4304. if (EXT4_SB(inode->i_sb)->s_es->s_creator_os !=
  4305. cpu_to_le32(EXT4_OS_HURD))
  4306. raw_inode->i_file_acl_high =
  4307. cpu_to_le16(ei->i_file_acl >> 32);
  4308. raw_inode->i_file_acl_lo = cpu_to_le32(ei->i_file_acl);
  4309. ext4_isize_set(raw_inode, ei->i_disksize);
  4310. if (ei->i_disksize > 0x7fffffffULL) {
  4311. struct super_block *sb = inode->i_sb;
  4312. if (!EXT4_HAS_RO_COMPAT_FEATURE(sb,
  4313. EXT4_FEATURE_RO_COMPAT_LARGE_FILE) ||
  4314. EXT4_SB(sb)->s_es->s_rev_level ==
  4315. cpu_to_le32(EXT4_GOOD_OLD_REV)) {
  4316. /* If this is the first large file
  4317. * created, add a flag to the superblock.
  4318. */
  4319. err = ext4_journal_get_write_access(handle,
  4320. EXT4_SB(sb)->s_sbh);
  4321. if (err)
  4322. goto out_brelse;
  4323. ext4_update_dynamic_rev(sb);
  4324. EXT4_SET_RO_COMPAT_FEATURE(sb,
  4325. EXT4_FEATURE_RO_COMPAT_LARGE_FILE);
  4326. sb->s_dirt = 1;
  4327. ext4_handle_sync(handle);
  4328. err = ext4_handle_dirty_metadata(handle, inode,
  4329. EXT4_SB(sb)->s_sbh);
  4330. }
  4331. }
  4332. raw_inode->i_generation = cpu_to_le32(inode->i_generation);
  4333. if (S_ISCHR(inode->i_mode) || S_ISBLK(inode->i_mode)) {
  4334. if (old_valid_dev(inode->i_rdev)) {
  4335. raw_inode->i_block[0] =
  4336. cpu_to_le32(old_encode_dev(inode->i_rdev));
  4337. raw_inode->i_block[1] = 0;
  4338. } else {
  4339. raw_inode->i_block[0] = 0;
  4340. raw_inode->i_block[1] =
  4341. cpu_to_le32(new_encode_dev(inode->i_rdev));
  4342. raw_inode->i_block[2] = 0;
  4343. }
  4344. } else for (block = 0; block < EXT4_N_BLOCKS; block++)
  4345. raw_inode->i_block[block] = ei->i_data[block];
  4346. raw_inode->i_disk_version = cpu_to_le32(inode->i_version);
  4347. if (ei->i_extra_isize) {
  4348. if (EXT4_FITS_IN_INODE(raw_inode, ei, i_version_hi))
  4349. raw_inode->i_version_hi =
  4350. cpu_to_le32(inode->i_version >> 32);
  4351. raw_inode->i_extra_isize = cpu_to_le16(ei->i_extra_isize);
  4352. }
  4353. BUFFER_TRACE(bh, "call ext4_handle_dirty_metadata");
  4354. rc = ext4_handle_dirty_metadata(handle, inode, bh);
  4355. if (!err)
  4356. err = rc;
  4357. ei->i_state &= ~EXT4_STATE_NEW;
  4358. out_brelse:
  4359. brelse(bh);
  4360. ext4_std_error(inode->i_sb, err);
  4361. return err;
  4362. }
  4363. /*
  4364. * ext4_write_inode()
  4365. *
  4366. * We are called from a few places:
  4367. *
  4368. * - Within generic_file_write() for O_SYNC files.
  4369. * Here, there will be no transaction running. We wait for any running
  4370. * trasnaction to commit.
  4371. *
  4372. * - Within sys_sync(), kupdate and such.
  4373. * We wait on commit, if tol to.
  4374. *
  4375. * - Within prune_icache() (PF_MEMALLOC == true)
  4376. * Here we simply return. We can't afford to block kswapd on the
  4377. * journal commit.
  4378. *
  4379. * In all cases it is actually safe for us to return without doing anything,
  4380. * because the inode has been copied into a raw inode buffer in
  4381. * ext4_mark_inode_dirty(). This is a correctness thing for O_SYNC and for
  4382. * knfsd.
  4383. *
  4384. * Note that we are absolutely dependent upon all inode dirtiers doing the
  4385. * right thing: they *must* call mark_inode_dirty() after dirtying info in
  4386. * which we are interested.
  4387. *
  4388. * It would be a bug for them to not do this. The code:
  4389. *
  4390. * mark_inode_dirty(inode)
  4391. * stuff();
  4392. * inode->i_size = expr;
  4393. *
  4394. * is in error because a kswapd-driven write_inode() could occur while
  4395. * `stuff()' is running, and the new i_size will be lost. Plus the inode
  4396. * will no longer be on the superblock's dirty inode list.
  4397. */
  4398. int ext4_write_inode(struct inode *inode, int wait)
  4399. {
  4400. if (current->flags & PF_MEMALLOC)
  4401. return 0;
  4402. if (ext4_journal_current_handle()) {
  4403. jbd_debug(1, "called recursively, non-PF_MEMALLOC!\n");
  4404. dump_stack();
  4405. return -EIO;
  4406. }
  4407. if (!wait)
  4408. return 0;
  4409. return ext4_force_commit(inode->i_sb);
  4410. }
  4411. /*
  4412. * ext4_setattr()
  4413. *
  4414. * Called from notify_change.
  4415. *
  4416. * We want to trap VFS attempts to truncate the file as soon as
  4417. * possible. In particular, we want to make sure that when the VFS
  4418. * shrinks i_size, we put the inode on the orphan list and modify
  4419. * i_disksize immediately, so that during the subsequent flushing of
  4420. * dirty pages and freeing of disk blocks, we can guarantee that any
  4421. * commit will leave the blocks being flushed in an unused state on
  4422. * disk. (On recovery, the inode will get truncated and the blocks will
  4423. * be freed, so we have a strong guarantee that no future commit will
  4424. * leave these blocks visible to the user.)
  4425. *
  4426. * Another thing we have to assure is that if we are in ordered mode
  4427. * and inode is still attached to the committing transaction, we must
  4428. * we start writeout of all the dirty pages which are being truncated.
  4429. * This way we are sure that all the data written in the previous
  4430. * transaction are already on disk (truncate waits for pages under
  4431. * writeback).
  4432. *
  4433. * Called with inode->i_mutex down.
  4434. */
  4435. int ext4_setattr(struct dentry *dentry, struct iattr *attr)
  4436. {
  4437. struct inode *inode = dentry->d_inode;
  4438. int error, rc = 0;
  4439. const unsigned int ia_valid = attr->ia_valid;
  4440. error = inode_change_ok(inode, attr);
  4441. if (error)
  4442. return error;
  4443. if ((ia_valid & ATTR_UID && attr->ia_uid != inode->i_uid) ||
  4444. (ia_valid & ATTR_GID && attr->ia_gid != inode->i_gid)) {
  4445. handle_t *handle;
  4446. /* (user+group)*(old+new) structure, inode write (sb,
  4447. * inode block, ? - but truncate inode update has it) */
  4448. handle = ext4_journal_start(inode, 2*(EXT4_QUOTA_INIT_BLOCKS(inode->i_sb)+
  4449. EXT4_QUOTA_DEL_BLOCKS(inode->i_sb))+3);
  4450. if (IS_ERR(handle)) {
  4451. error = PTR_ERR(handle);
  4452. goto err_out;
  4453. }
  4454. error = vfs_dq_transfer(inode, attr) ? -EDQUOT : 0;
  4455. if (error) {
  4456. ext4_journal_stop(handle);
  4457. return error;
  4458. }
  4459. /* Update corresponding info in inode so that everything is in
  4460. * one transaction */
  4461. if (attr->ia_valid & ATTR_UID)
  4462. inode->i_uid = attr->ia_uid;
  4463. if (attr->ia_valid & ATTR_GID)
  4464. inode->i_gid = attr->ia_gid;
  4465. error = ext4_mark_inode_dirty(handle, inode);
  4466. ext4_journal_stop(handle);
  4467. }
  4468. if (attr->ia_valid & ATTR_SIZE) {
  4469. if (!(EXT4_I(inode)->i_flags & EXT4_EXTENTS_FL)) {
  4470. struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
  4471. if (attr->ia_size > sbi->s_bitmap_maxbytes) {
  4472. error = -EFBIG;
  4473. goto err_out;
  4474. }
  4475. }
  4476. }
  4477. if (S_ISREG(inode->i_mode) &&
  4478. attr->ia_valid & ATTR_SIZE && attr->ia_size < inode->i_size) {
  4479. handle_t *handle;
  4480. handle = ext4_journal_start(inode, 3);
  4481. if (IS_ERR(handle)) {
  4482. error = PTR_ERR(handle);
  4483. goto err_out;
  4484. }
  4485. error = ext4_orphan_add(handle, inode);
  4486. EXT4_I(inode)->i_disksize = attr->ia_size;
  4487. rc = ext4_mark_inode_dirty(handle, inode);
  4488. if (!error)
  4489. error = rc;
  4490. ext4_journal_stop(handle);
  4491. if (ext4_should_order_data(inode)) {
  4492. error = ext4_begin_ordered_truncate(inode,
  4493. attr->ia_size);
  4494. if (error) {
  4495. /* Do as much error cleanup as possible */
  4496. handle = ext4_journal_start(inode, 3);
  4497. if (IS_ERR(handle)) {
  4498. ext4_orphan_del(NULL, inode);
  4499. goto err_out;
  4500. }
  4501. ext4_orphan_del(handle, inode);
  4502. ext4_journal_stop(handle);
  4503. goto err_out;
  4504. }
  4505. }
  4506. }
  4507. rc = inode_setattr(inode, attr);
  4508. /* If inode_setattr's call to ext4_truncate failed to get a
  4509. * transaction handle at all, we need to clean up the in-core
  4510. * orphan list manually. */
  4511. if (inode->i_nlink)
  4512. ext4_orphan_del(NULL, inode);
  4513. if (!rc && (ia_valid & ATTR_MODE))
  4514. rc = ext4_acl_chmod(inode);
  4515. err_out:
  4516. ext4_std_error(inode->i_sb, error);
  4517. if (!error)
  4518. error = rc;
  4519. return error;
  4520. }
  4521. int ext4_getattr(struct vfsmount *mnt, struct dentry *dentry,
  4522. struct kstat *stat)
  4523. {
  4524. struct inode *inode;
  4525. unsigned long delalloc_blocks;
  4526. inode = dentry->d_inode;
  4527. generic_fillattr(inode, stat);
  4528. /*
  4529. * We can't update i_blocks if the block allocation is delayed
  4530. * otherwise in the case of system crash before the real block
  4531. * allocation is done, we will have i_blocks inconsistent with
  4532. * on-disk file blocks.
  4533. * We always keep i_blocks updated together with real
  4534. * allocation. But to not confuse with user, stat
  4535. * will return the blocks that include the delayed allocation
  4536. * blocks for this file.
  4537. */
  4538. spin_lock(&EXT4_I(inode)->i_block_reservation_lock);
  4539. delalloc_blocks = EXT4_I(inode)->i_reserved_data_blocks;
  4540. spin_unlock(&EXT4_I(inode)->i_block_reservation_lock);
  4541. stat->blocks += (delalloc_blocks << inode->i_sb->s_blocksize_bits)>>9;
  4542. return 0;
  4543. }
  4544. static int ext4_indirect_trans_blocks(struct inode *inode, int nrblocks,
  4545. int chunk)
  4546. {
  4547. int indirects;
  4548. /* if nrblocks are contiguous */
  4549. if (chunk) {
  4550. /*
  4551. * With N contiguous data blocks, it need at most
  4552. * N/EXT4_ADDR_PER_BLOCK(inode->i_sb) indirect blocks
  4553. * 2 dindirect blocks
  4554. * 1 tindirect block
  4555. */
  4556. indirects = nrblocks / EXT4_ADDR_PER_BLOCK(inode->i_sb);
  4557. return indirects + 3;
  4558. }
  4559. /*
  4560. * if nrblocks are not contiguous, worse case, each block touch
  4561. * a indirect block, and each indirect block touch a double indirect
  4562. * block, plus a triple indirect block
  4563. */
  4564. indirects = nrblocks * 2 + 1;
  4565. return indirects;
  4566. }
  4567. static int ext4_index_trans_blocks(struct inode *inode, int nrblocks, int chunk)
  4568. {
  4569. if (!(EXT4_I(inode)->i_flags & EXT4_EXTENTS_FL))
  4570. return ext4_indirect_trans_blocks(inode, nrblocks, chunk);
  4571. return ext4_ext_index_trans_blocks(inode, nrblocks, chunk);
  4572. }
  4573. /*
  4574. * Account for index blocks, block groups bitmaps and block group
  4575. * descriptor blocks if modify datablocks and index blocks
  4576. * worse case, the indexs blocks spread over different block groups
  4577. *
  4578. * If datablocks are discontiguous, they are possible to spread over
  4579. * different block groups too. If they are contiugous, with flexbg,
  4580. * they could still across block group boundary.
  4581. *
  4582. * Also account for superblock, inode, quota and xattr blocks
  4583. */
  4584. int ext4_meta_trans_blocks(struct inode *inode, int nrblocks, int chunk)
  4585. {
  4586. ext4_group_t groups, ngroups = ext4_get_groups_count(inode->i_sb);
  4587. int gdpblocks;
  4588. int idxblocks;
  4589. int ret = 0;
  4590. /*
  4591. * How many index blocks need to touch to modify nrblocks?
  4592. * The "Chunk" flag indicating whether the nrblocks is
  4593. * physically contiguous on disk
  4594. *
  4595. * For Direct IO and fallocate, they calls get_block to allocate
  4596. * one single extent at a time, so they could set the "Chunk" flag
  4597. */
  4598. idxblocks = ext4_index_trans_blocks(inode, nrblocks, chunk);
  4599. ret = idxblocks;
  4600. /*
  4601. * Now let's see how many group bitmaps and group descriptors need
  4602. * to account
  4603. */
  4604. groups = idxblocks;
  4605. if (chunk)
  4606. groups += 1;
  4607. else
  4608. groups += nrblocks;
  4609. gdpblocks = groups;
  4610. if (groups > ngroups)
  4611. groups = ngroups;
  4612. if (groups > EXT4_SB(inode->i_sb)->s_gdb_count)
  4613. gdpblocks = EXT4_SB(inode->i_sb)->s_gdb_count;
  4614. /* bitmaps and block group descriptor blocks */
  4615. ret += groups + gdpblocks;
  4616. /* Blocks for super block, inode, quota and xattr blocks */
  4617. ret += EXT4_META_TRANS_BLOCKS(inode->i_sb);
  4618. return ret;
  4619. }
  4620. /*
  4621. * Calulate the total number of credits to reserve to fit
  4622. * the modification of a single pages into a single transaction,
  4623. * which may include multiple chunks of block allocations.
  4624. *
  4625. * This could be called via ext4_write_begin()
  4626. *
  4627. * We need to consider the worse case, when
  4628. * one new block per extent.
  4629. */
  4630. int ext4_writepage_trans_blocks(struct inode *inode)
  4631. {
  4632. int bpp = ext4_journal_blocks_per_page(inode);
  4633. int ret;
  4634. ret = ext4_meta_trans_blocks(inode, bpp, 0);
  4635. /* Account for data blocks for journalled mode */
  4636. if (ext4_should_journal_data(inode))
  4637. ret += bpp;
  4638. return ret;
  4639. }
  4640. /*
  4641. * Calculate the journal credits for a chunk of data modification.
  4642. *
  4643. * This is called from DIO, fallocate or whoever calling
  4644. * ext4_get_blocks() to map/allocate a chunk of contigous disk blocks.
  4645. *
  4646. * journal buffers for data blocks are not included here, as DIO
  4647. * and fallocate do no need to journal data buffers.
  4648. */
  4649. int ext4_chunk_trans_blocks(struct inode *inode, int nrblocks)
  4650. {
  4651. return ext4_meta_trans_blocks(inode, nrblocks, 1);
  4652. }
  4653. /*
  4654. * The caller must have previously called ext4_reserve_inode_write().
  4655. * Give this, we know that the caller already has write access to iloc->bh.
  4656. */
  4657. int ext4_mark_iloc_dirty(handle_t *handle,
  4658. struct inode *inode, struct ext4_iloc *iloc)
  4659. {
  4660. int err = 0;
  4661. if (test_opt(inode->i_sb, I_VERSION))
  4662. inode_inc_iversion(inode);
  4663. /* the do_update_inode consumes one bh->b_count */
  4664. get_bh(iloc->bh);
  4665. /* ext4_do_update_inode() does jbd2_journal_dirty_metadata */
  4666. err = ext4_do_update_inode(handle, inode, iloc);
  4667. put_bh(iloc->bh);
  4668. return err;
  4669. }
  4670. /*
  4671. * On success, We end up with an outstanding reference count against
  4672. * iloc->bh. This _must_ be cleaned up later.
  4673. */
  4674. int
  4675. ext4_reserve_inode_write(handle_t *handle, struct inode *inode,
  4676. struct ext4_iloc *iloc)
  4677. {
  4678. int err;
  4679. err = ext4_get_inode_loc(inode, iloc);
  4680. if (!err) {
  4681. BUFFER_TRACE(iloc->bh, "get_write_access");
  4682. err = ext4_journal_get_write_access(handle, iloc->bh);
  4683. if (err) {
  4684. brelse(iloc->bh);
  4685. iloc->bh = NULL;
  4686. }
  4687. }
  4688. ext4_std_error(inode->i_sb, err);
  4689. return err;
  4690. }
  4691. /*
  4692. * Expand an inode by new_extra_isize bytes.
  4693. * Returns 0 on success or negative error number on failure.
  4694. */
  4695. static int ext4_expand_extra_isize(struct inode *inode,
  4696. unsigned int new_extra_isize,
  4697. struct ext4_iloc iloc,
  4698. handle_t *handle)
  4699. {
  4700. struct ext4_inode *raw_inode;
  4701. struct ext4_xattr_ibody_header *header;
  4702. struct ext4_xattr_entry *entry;
  4703. if (EXT4_I(inode)->i_extra_isize >= new_extra_isize)
  4704. return 0;
  4705. raw_inode = ext4_raw_inode(&iloc);
  4706. header = IHDR(inode, raw_inode);
  4707. entry = IFIRST(header);
  4708. /* No extended attributes present */
  4709. if (!(EXT4_I(inode)->i_state & EXT4_STATE_XATTR) ||
  4710. header->h_magic != cpu_to_le32(EXT4_XATTR_MAGIC)) {
  4711. memset((void *)raw_inode + EXT4_GOOD_OLD_INODE_SIZE, 0,
  4712. new_extra_isize);
  4713. EXT4_I(inode)->i_extra_isize = new_extra_isize;
  4714. return 0;
  4715. }
  4716. /* try to expand with EAs present */
  4717. return ext4_expand_extra_isize_ea(inode, new_extra_isize,
  4718. raw_inode, handle);
  4719. }
  4720. /*
  4721. * What we do here is to mark the in-core inode as clean with respect to inode
  4722. * dirtiness (it may still be data-dirty).
  4723. * This means that the in-core inode may be reaped by prune_icache
  4724. * without having to perform any I/O. This is a very good thing,
  4725. * because *any* task may call prune_icache - even ones which
  4726. * have a transaction open against a different journal.
  4727. *
  4728. * Is this cheating? Not really. Sure, we haven't written the
  4729. * inode out, but prune_icache isn't a user-visible syncing function.
  4730. * Whenever the user wants stuff synced (sys_sync, sys_msync, sys_fsync)
  4731. * we start and wait on commits.
  4732. *
  4733. * Is this efficient/effective? Well, we're being nice to the system
  4734. * by cleaning up our inodes proactively so they can be reaped
  4735. * without I/O. But we are potentially leaving up to five seconds'
  4736. * worth of inodes floating about which prune_icache wants us to
  4737. * write out. One way to fix that would be to get prune_icache()
  4738. * to do a write_super() to free up some memory. It has the desired
  4739. * effect.
  4740. */
  4741. int ext4_mark_inode_dirty(handle_t *handle, struct inode *inode)
  4742. {
  4743. struct ext4_iloc iloc;
  4744. struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
  4745. static unsigned int mnt_count;
  4746. int err, ret;
  4747. might_sleep();
  4748. err = ext4_reserve_inode_write(handle, inode, &iloc);
  4749. if (ext4_handle_valid(handle) &&
  4750. EXT4_I(inode)->i_extra_isize < sbi->s_want_extra_isize &&
  4751. !(EXT4_I(inode)->i_state & EXT4_STATE_NO_EXPAND)) {
  4752. /*
  4753. * We need extra buffer credits since we may write into EA block
  4754. * with this same handle. If journal_extend fails, then it will
  4755. * only result in a minor loss of functionality for that inode.
  4756. * If this is felt to be critical, then e2fsck should be run to
  4757. * force a large enough s_min_extra_isize.
  4758. */
  4759. if ((jbd2_journal_extend(handle,
  4760. EXT4_DATA_TRANS_BLOCKS(inode->i_sb))) == 0) {
  4761. ret = ext4_expand_extra_isize(inode,
  4762. sbi->s_want_extra_isize,
  4763. iloc, handle);
  4764. if (ret) {
  4765. EXT4_I(inode)->i_state |= EXT4_STATE_NO_EXPAND;
  4766. if (mnt_count !=
  4767. le16_to_cpu(sbi->s_es->s_mnt_count)) {
  4768. ext4_warning(inode->i_sb, __func__,
  4769. "Unable to expand inode %lu. Delete"
  4770. " some EAs or run e2fsck.",
  4771. inode->i_ino);
  4772. mnt_count =
  4773. le16_to_cpu(sbi->s_es->s_mnt_count);
  4774. }
  4775. }
  4776. }
  4777. }
  4778. if (!err)
  4779. err = ext4_mark_iloc_dirty(handle, inode, &iloc);
  4780. return err;
  4781. }
  4782. /*
  4783. * ext4_dirty_inode() is called from __mark_inode_dirty()
  4784. *
  4785. * We're really interested in the case where a file is being extended.
  4786. * i_size has been changed by generic_commit_write() and we thus need
  4787. * to include the updated inode in the current transaction.
  4788. *
  4789. * Also, vfs_dq_alloc_block() will always dirty the inode when blocks
  4790. * are allocated to the file.
  4791. *
  4792. * If the inode is marked synchronous, we don't honour that here - doing
  4793. * so would cause a commit on atime updates, which we don't bother doing.
  4794. * We handle synchronous inodes at the highest possible level.
  4795. */
  4796. void ext4_dirty_inode(struct inode *inode)
  4797. {
  4798. handle_t *current_handle = ext4_journal_current_handle();
  4799. handle_t *handle;
  4800. if (!ext4_handle_valid(current_handle)) {
  4801. ext4_mark_inode_dirty(current_handle, inode);
  4802. return;
  4803. }
  4804. handle = ext4_journal_start(inode, 2);
  4805. if (IS_ERR(handle))
  4806. goto out;
  4807. if (current_handle &&
  4808. current_handle->h_transaction != handle->h_transaction) {
  4809. /* This task has a transaction open against a different fs */
  4810. printk(KERN_EMERG "%s: transactions do not match!\n",
  4811. __func__);
  4812. } else {
  4813. jbd_debug(5, "marking dirty. outer handle=%p\n",
  4814. current_handle);
  4815. ext4_mark_inode_dirty(handle, inode);
  4816. }
  4817. ext4_journal_stop(handle);
  4818. out:
  4819. return;
  4820. }
  4821. #if 0
  4822. /*
  4823. * Bind an inode's backing buffer_head into this transaction, to prevent
  4824. * it from being flushed to disk early. Unlike
  4825. * ext4_reserve_inode_write, this leaves behind no bh reference and
  4826. * returns no iloc structure, so the caller needs to repeat the iloc
  4827. * lookup to mark the inode dirty later.
  4828. */
  4829. static int ext4_pin_inode(handle_t *handle, struct inode *inode)
  4830. {
  4831. struct ext4_iloc iloc;
  4832. int err = 0;
  4833. if (handle) {
  4834. err = ext4_get_inode_loc(inode, &iloc);
  4835. if (!err) {
  4836. BUFFER_TRACE(iloc.bh, "get_write_access");
  4837. err = jbd2_journal_get_write_access(handle, iloc.bh);
  4838. if (!err)
  4839. err = ext4_handle_dirty_metadata(handle,
  4840. inode,
  4841. iloc.bh);
  4842. brelse(iloc.bh);
  4843. }
  4844. }
  4845. ext4_std_error(inode->i_sb, err);
  4846. return err;
  4847. }
  4848. #endif
  4849. int ext4_change_inode_journal_flag(struct inode *inode, int val)
  4850. {
  4851. journal_t *journal;
  4852. handle_t *handle;
  4853. int err;
  4854. /*
  4855. * We have to be very careful here: changing a data block's
  4856. * journaling status dynamically is dangerous. If we write a
  4857. * data block to the journal, change the status and then delete
  4858. * that block, we risk forgetting to revoke the old log record
  4859. * from the journal and so a subsequent replay can corrupt data.
  4860. * So, first we make sure that the journal is empty and that
  4861. * nobody is changing anything.
  4862. */
  4863. journal = EXT4_JOURNAL(inode);
  4864. if (!journal)
  4865. return 0;
  4866. if (is_journal_aborted(journal))
  4867. return -EROFS;
  4868. jbd2_journal_lock_updates(journal);
  4869. jbd2_journal_flush(journal);
  4870. /*
  4871. * OK, there are no updates running now, and all cached data is
  4872. * synced to disk. We are now in a completely consistent state
  4873. * which doesn't have anything in the journal, and we know that
  4874. * no filesystem updates are running, so it is safe to modify
  4875. * the inode's in-core data-journaling state flag now.
  4876. */
  4877. if (val)
  4878. EXT4_I(inode)->i_flags |= EXT4_JOURNAL_DATA_FL;
  4879. else
  4880. EXT4_I(inode)->i_flags &= ~EXT4_JOURNAL_DATA_FL;
  4881. ext4_set_aops(inode);
  4882. jbd2_journal_unlock_updates(journal);
  4883. /* Finally we can mark the inode as dirty. */
  4884. handle = ext4_journal_start(inode, 1);
  4885. if (IS_ERR(handle))
  4886. return PTR_ERR(handle);
  4887. err = ext4_mark_inode_dirty(handle, inode);
  4888. ext4_handle_sync(handle);
  4889. ext4_journal_stop(handle);
  4890. ext4_std_error(inode->i_sb, err);
  4891. return err;
  4892. }
  4893. static int ext4_bh_unmapped(handle_t *handle, struct buffer_head *bh)
  4894. {
  4895. return !buffer_mapped(bh);
  4896. }
  4897. int ext4_page_mkwrite(struct vm_area_struct *vma, struct vm_fault *vmf)
  4898. {
  4899. struct page *page = vmf->page;
  4900. loff_t size;
  4901. unsigned long len;
  4902. int ret = -EINVAL;
  4903. void *fsdata;
  4904. struct file *file = vma->vm_file;
  4905. struct inode *inode = file->f_path.dentry->d_inode;
  4906. struct address_space *mapping = inode->i_mapping;
  4907. /*
  4908. * Get i_alloc_sem to stop truncates messing with the inode. We cannot
  4909. * get i_mutex because we are already holding mmap_sem.
  4910. */
  4911. down_read(&inode->i_alloc_sem);
  4912. size = i_size_read(inode);
  4913. if (page->mapping != mapping || size <= page_offset(page)
  4914. || !PageUptodate(page)) {
  4915. /* page got truncated from under us? */
  4916. goto out_unlock;
  4917. }
  4918. ret = 0;
  4919. if (PageMappedToDisk(page))
  4920. goto out_unlock;
  4921. if (page->index == size >> PAGE_CACHE_SHIFT)
  4922. len = size & ~PAGE_CACHE_MASK;
  4923. else
  4924. len = PAGE_CACHE_SIZE;
  4925. if (page_has_buffers(page)) {
  4926. /* return if we have all the buffers mapped */
  4927. if (!walk_page_buffers(NULL, page_buffers(page), 0, len, NULL,
  4928. ext4_bh_unmapped))
  4929. goto out_unlock;
  4930. }
  4931. /*
  4932. * OK, we need to fill the hole... Do write_begin write_end
  4933. * to do block allocation/reservation.We are not holding
  4934. * inode.i__mutex here. That allow * parallel write_begin,
  4935. * write_end call. lock_page prevent this from happening
  4936. * on the same page though
  4937. */
  4938. ret = mapping->a_ops->write_begin(file, mapping, page_offset(page),
  4939. len, AOP_FLAG_UNINTERRUPTIBLE, &page, &fsdata);
  4940. if (ret < 0)
  4941. goto out_unlock;
  4942. ret = mapping->a_ops->write_end(file, mapping, page_offset(page),
  4943. len, len, page, fsdata);
  4944. if (ret < 0)
  4945. goto out_unlock;
  4946. ret = 0;
  4947. out_unlock:
  4948. if (ret)
  4949. ret = VM_FAULT_SIGBUS;
  4950. up_read(&inode->i_alloc_sem);
  4951. return ret;
  4952. }