extent-tree.c 154 KB

123456789101112131415161718192021222324252627282930313233343536373839404142434445464748495051525354555657585960616263646566676869707172737475767778798081828384858687888990919293949596979899100101102103104105106107108109110111112113114115116117118119120121122123124125126127128129130131132133134135136137138139140141142143144145146147148149150151152153154155156157158159160161162163164165166167168169170171172173174175176177178179180181182183184185186187188189190191192193194195196197198199200201202203204205206207208209210211212213214215216217218219220221222223224225226227228229230231232233234235236237238239240241242243244245246247248249250251252253254255256257258259260261262263264265266267268269270271272273274275276277278279280281282283284285286287288289290291292293294295296297298299300301302303304305306307308309310311312313314315316317318319320321322323324325326327328329330331332333334335336337338339340341342343344345346347348349350351352353354355356357358359360361362363364365366367368369370371372373374375376377378379380381382383384385386387388389390391392393394395396397398399400401402403404405406407408409410411412413414415416417418419420421422423424425426427428429430431432433434435436437438439440441442443444445446447448449450451452453454455456457458459460461462463464465466467468469470471472473474475476477478479480481482483484485486487488489490491492493494495496497498499500501502503504505506507508509510511512513514515516517518519520521522523524525526527528529530531532533534535536537538539540541542543544545546547548549550551552553554555556557558559560561562563564565566567568569570571572573574575576577578579580581582583584585586587588589590591592593594595596597598599600601602603604605606607608609610611612613614615616617618619620621622623624625626627628629630631632633634635636637638639640641642643644645646647648649650651652653654655656657658659660661662663664665666667668669670671672673674675676677678679680681682683684685686687688689690691692693694695696697698699700701702703704705706707708709710711712713714715716717718719720721722723724725726727728729730731732733734735736737738739740741742743744745746747748749750751752753754755756757758759760761762763764765766767768769770771772773774775776777778779780781782783784785786787788789790791792793794795796797798799800801802803804805806807808809810811812813814815816817818819820821822823824825826827828829830831832833834835836837838839840841842843844845846847848849850851852853854855856857858859860861862863864865866867868869870871872873874875876877878879880881882883884885886887888889890891892893894895896897898899900901902903904905906907908909910911912913914915916917918919920921922923924925926927928929930931932933934935936937938939940941942943944945946947948949950951952953954955956957958959960961962963964965966967968969970971972973974975976977978979980981982983984985986987988989990991992993994995996997998999100010011002100310041005100610071008100910101011101210131014101510161017101810191020102110221023102410251026102710281029103010311032103310341035103610371038103910401041104210431044104510461047104810491050105110521053105410551056105710581059106010611062106310641065106610671068106910701071107210731074107510761077107810791080108110821083108410851086108710881089109010911092109310941095109610971098109911001101110211031104110511061107110811091110111111121113111411151116111711181119112011211122112311241125112611271128112911301131113211331134113511361137113811391140114111421143114411451146114711481149115011511152115311541155115611571158115911601161116211631164116511661167116811691170117111721173117411751176117711781179118011811182118311841185118611871188118911901191119211931194119511961197119811991200120112021203120412051206120712081209121012111212121312141215121612171218121912201221122212231224122512261227122812291230123112321233123412351236123712381239124012411242124312441245124612471248124912501251125212531254125512561257125812591260126112621263126412651266126712681269127012711272127312741275127612771278127912801281128212831284128512861287128812891290129112921293129412951296129712981299130013011302130313041305130613071308130913101311131213131314131513161317131813191320132113221323132413251326132713281329133013311332133313341335133613371338133913401341134213431344134513461347134813491350135113521353135413551356135713581359136013611362136313641365136613671368136913701371137213731374137513761377137813791380138113821383138413851386138713881389139013911392139313941395139613971398139914001401140214031404140514061407140814091410141114121413141414151416141714181419142014211422142314241425142614271428142914301431143214331434143514361437143814391440144114421443144414451446144714481449145014511452145314541455145614571458145914601461146214631464146514661467146814691470147114721473147414751476147714781479148014811482148314841485148614871488148914901491149214931494149514961497149814991500150115021503150415051506150715081509151015111512151315141515151615171518151915201521152215231524152515261527152815291530153115321533153415351536153715381539154015411542154315441545154615471548154915501551155215531554155515561557155815591560156115621563156415651566156715681569157015711572157315741575157615771578157915801581158215831584158515861587158815891590159115921593159415951596159715981599160016011602160316041605160616071608160916101611161216131614161516161617161816191620162116221623162416251626162716281629163016311632163316341635163616371638163916401641164216431644164516461647164816491650165116521653165416551656165716581659166016611662166316641665166616671668166916701671167216731674167516761677167816791680168116821683168416851686168716881689169016911692169316941695169616971698169917001701170217031704170517061707170817091710171117121713171417151716171717181719172017211722172317241725172617271728172917301731173217331734173517361737173817391740174117421743174417451746174717481749175017511752175317541755175617571758175917601761176217631764176517661767176817691770177117721773177417751776177717781779178017811782178317841785178617871788178917901791179217931794179517961797179817991800180118021803180418051806180718081809181018111812181318141815181618171818181918201821182218231824182518261827182818291830183118321833183418351836183718381839184018411842184318441845184618471848184918501851185218531854185518561857185818591860186118621863186418651866186718681869187018711872187318741875187618771878187918801881188218831884188518861887188818891890189118921893189418951896189718981899190019011902190319041905190619071908190919101911191219131914191519161917191819191920192119221923192419251926192719281929193019311932193319341935193619371938193919401941194219431944194519461947194819491950195119521953195419551956195719581959196019611962196319641965196619671968196919701971197219731974197519761977197819791980198119821983198419851986198719881989199019911992199319941995199619971998199920002001200220032004200520062007200820092010201120122013201420152016201720182019202020212022202320242025202620272028202920302031203220332034203520362037203820392040204120422043204420452046204720482049205020512052205320542055205620572058205920602061206220632064206520662067206820692070207120722073207420752076207720782079208020812082208320842085208620872088208920902091209220932094209520962097209820992100210121022103210421052106210721082109211021112112211321142115211621172118211921202121212221232124212521262127212821292130213121322133213421352136213721382139214021412142214321442145214621472148214921502151215221532154215521562157215821592160216121622163216421652166216721682169217021712172217321742175217621772178217921802181218221832184218521862187218821892190219121922193219421952196219721982199220022012202220322042205220622072208220922102211221222132214221522162217221822192220222122222223222422252226222722282229223022312232223322342235223622372238223922402241224222432244224522462247224822492250225122522253225422552256225722582259226022612262226322642265226622672268226922702271227222732274227522762277227822792280228122822283228422852286228722882289229022912292229322942295229622972298229923002301230223032304230523062307230823092310231123122313231423152316231723182319232023212322232323242325232623272328232923302331233223332334233523362337233823392340234123422343234423452346234723482349235023512352235323542355235623572358235923602361236223632364236523662367236823692370237123722373237423752376237723782379238023812382238323842385238623872388238923902391239223932394239523962397239823992400240124022403240424052406240724082409241024112412241324142415241624172418241924202421242224232424242524262427242824292430243124322433243424352436243724382439244024412442244324442445244624472448244924502451245224532454245524562457245824592460246124622463246424652466246724682469247024712472247324742475247624772478247924802481248224832484248524862487248824892490249124922493249424952496249724982499250025012502250325042505250625072508250925102511251225132514251525162517251825192520252125222523252425252526252725282529253025312532253325342535253625372538253925402541254225432544254525462547254825492550255125522553255425552556255725582559256025612562256325642565256625672568256925702571257225732574257525762577257825792580258125822583258425852586258725882589259025912592259325942595259625972598259926002601260226032604260526062607260826092610261126122613261426152616261726182619262026212622262326242625262626272628262926302631263226332634263526362637263826392640264126422643264426452646264726482649265026512652265326542655265626572658265926602661266226632664266526662667266826692670267126722673267426752676267726782679268026812682268326842685268626872688268926902691269226932694269526962697269826992700270127022703270427052706270727082709271027112712271327142715271627172718271927202721272227232724272527262727272827292730273127322733273427352736273727382739274027412742274327442745274627472748274927502751275227532754275527562757275827592760276127622763276427652766276727682769277027712772277327742775277627772778277927802781278227832784278527862787278827892790279127922793279427952796279727982799280028012802280328042805280628072808280928102811281228132814281528162817281828192820282128222823282428252826282728282829283028312832283328342835283628372838283928402841284228432844284528462847284828492850285128522853285428552856285728582859286028612862286328642865286628672868286928702871287228732874287528762877287828792880288128822883288428852886288728882889289028912892289328942895289628972898289929002901290229032904290529062907290829092910291129122913291429152916291729182919292029212922292329242925292629272928292929302931293229332934293529362937293829392940294129422943294429452946294729482949295029512952295329542955295629572958295929602961296229632964296529662967296829692970297129722973297429752976297729782979298029812982298329842985298629872988298929902991299229932994299529962997299829993000300130023003300430053006300730083009301030113012301330143015301630173018301930203021302230233024302530263027302830293030303130323033303430353036303730383039304030413042304330443045304630473048304930503051305230533054305530563057305830593060306130623063306430653066306730683069307030713072307330743075307630773078307930803081308230833084308530863087308830893090309130923093309430953096309730983099310031013102310331043105310631073108310931103111311231133114311531163117311831193120312131223123312431253126312731283129313031313132313331343135313631373138313931403141314231433144314531463147314831493150315131523153315431553156315731583159316031613162316331643165316631673168316931703171317231733174317531763177317831793180318131823183318431853186318731883189319031913192319331943195319631973198319932003201320232033204320532063207320832093210321132123213321432153216321732183219322032213222322332243225322632273228322932303231323232333234323532363237323832393240324132423243324432453246324732483249325032513252325332543255325632573258325932603261326232633264326532663267326832693270327132723273327432753276327732783279328032813282328332843285328632873288328932903291329232933294329532963297329832993300330133023303330433053306330733083309331033113312331333143315331633173318331933203321332233233324332533263327332833293330333133323333333433353336333733383339334033413342334333443345334633473348334933503351335233533354335533563357335833593360336133623363336433653366336733683369337033713372337333743375337633773378337933803381338233833384338533863387338833893390339133923393339433953396339733983399340034013402340334043405340634073408340934103411341234133414341534163417341834193420342134223423342434253426342734283429343034313432343334343435343634373438343934403441344234433444344534463447344834493450345134523453345434553456345734583459346034613462346334643465346634673468346934703471347234733474347534763477347834793480348134823483348434853486348734883489349034913492349334943495349634973498349935003501350235033504350535063507350835093510351135123513351435153516351735183519352035213522352335243525352635273528352935303531353235333534353535363537353835393540354135423543354435453546354735483549355035513552355335543555355635573558355935603561356235633564356535663567356835693570357135723573357435753576357735783579358035813582358335843585358635873588358935903591359235933594359535963597359835993600360136023603360436053606360736083609361036113612361336143615361636173618361936203621362236233624362536263627362836293630363136323633363436353636363736383639364036413642364336443645364636473648364936503651365236533654365536563657365836593660366136623663366436653666366736683669367036713672367336743675367636773678367936803681368236833684368536863687368836893690369136923693369436953696369736983699370037013702370337043705370637073708370937103711371237133714371537163717371837193720372137223723372437253726372737283729373037313732373337343735373637373738373937403741374237433744374537463747374837493750375137523753375437553756375737583759376037613762376337643765376637673768376937703771377237733774377537763777377837793780378137823783378437853786378737883789379037913792379337943795379637973798379938003801380238033804380538063807380838093810381138123813381438153816381738183819382038213822382338243825382638273828382938303831383238333834383538363837383838393840384138423843384438453846384738483849385038513852385338543855385638573858385938603861386238633864386538663867386838693870387138723873387438753876387738783879388038813882388338843885388638873888388938903891389238933894389538963897389838993900390139023903390439053906390739083909391039113912391339143915391639173918391939203921392239233924392539263927392839293930393139323933393439353936393739383939394039413942394339443945394639473948394939503951395239533954395539563957395839593960396139623963396439653966396739683969397039713972397339743975397639773978397939803981398239833984398539863987398839893990399139923993399439953996399739983999400040014002400340044005400640074008400940104011401240134014401540164017401840194020402140224023402440254026402740284029403040314032403340344035403640374038403940404041404240434044404540464047404840494050405140524053405440554056405740584059406040614062406340644065406640674068406940704071407240734074407540764077407840794080408140824083408440854086408740884089409040914092409340944095409640974098409941004101410241034104410541064107410841094110411141124113411441154116411741184119412041214122412341244125412641274128412941304131413241334134413541364137413841394140414141424143414441454146414741484149415041514152415341544155415641574158415941604161416241634164416541664167416841694170417141724173417441754176417741784179418041814182418341844185418641874188418941904191419241934194419541964197419841994200420142024203420442054206420742084209421042114212421342144215421642174218421942204221422242234224422542264227422842294230423142324233423442354236423742384239424042414242424342444245424642474248424942504251425242534254425542564257425842594260426142624263426442654266426742684269427042714272427342744275427642774278427942804281428242834284428542864287428842894290429142924293429442954296429742984299430043014302430343044305430643074308430943104311431243134314431543164317431843194320432143224323432443254326432743284329433043314332433343344335433643374338433943404341434243434344434543464347434843494350435143524353435443554356435743584359436043614362436343644365436643674368436943704371437243734374437543764377437843794380438143824383438443854386438743884389439043914392439343944395439643974398439944004401440244034404440544064407440844094410441144124413441444154416441744184419442044214422442344244425442644274428442944304431443244334434443544364437443844394440444144424443444444454446444744484449445044514452445344544455445644574458445944604461446244634464446544664467446844694470447144724473447444754476447744784479448044814482448344844485448644874488448944904491449244934494449544964497449844994500450145024503450445054506450745084509451045114512451345144515451645174518451945204521452245234524452545264527452845294530453145324533453445354536453745384539454045414542454345444545454645474548454945504551455245534554455545564557455845594560456145624563456445654566456745684569457045714572457345744575457645774578457945804581458245834584458545864587458845894590459145924593459445954596459745984599460046014602460346044605460646074608460946104611461246134614461546164617461846194620462146224623462446254626462746284629463046314632463346344635463646374638463946404641464246434644464546464647464846494650465146524653465446554656465746584659466046614662466346644665466646674668466946704671467246734674467546764677467846794680468146824683468446854686468746884689469046914692469346944695469646974698469947004701470247034704470547064707470847094710471147124713471447154716471747184719472047214722472347244725472647274728472947304731473247334734473547364737473847394740474147424743474447454746474747484749475047514752475347544755475647574758475947604761476247634764476547664767476847694770477147724773477447754776477747784779478047814782478347844785478647874788478947904791479247934794479547964797479847994800480148024803480448054806480748084809481048114812481348144815481648174818481948204821482248234824482548264827482848294830483148324833483448354836483748384839484048414842484348444845484648474848484948504851485248534854485548564857485848594860486148624863486448654866486748684869487048714872487348744875487648774878487948804881488248834884488548864887488848894890489148924893489448954896489748984899490049014902490349044905490649074908490949104911491249134914491549164917491849194920492149224923492449254926492749284929493049314932493349344935493649374938493949404941494249434944494549464947494849494950495149524953495449554956495749584959496049614962496349644965496649674968496949704971497249734974497549764977497849794980498149824983498449854986498749884989499049914992499349944995499649974998499950005001500250035004500550065007500850095010501150125013501450155016501750185019502050215022502350245025502650275028502950305031503250335034503550365037503850395040504150425043504450455046504750485049505050515052505350545055505650575058505950605061506250635064506550665067506850695070507150725073507450755076507750785079508050815082508350845085508650875088508950905091509250935094509550965097509850995100510151025103510451055106510751085109511051115112511351145115511651175118511951205121512251235124512551265127512851295130513151325133513451355136513751385139514051415142514351445145514651475148514951505151515251535154515551565157515851595160516151625163516451655166516751685169517051715172517351745175517651775178517951805181518251835184518551865187518851895190519151925193519451955196519751985199520052015202520352045205520652075208520952105211521252135214521552165217521852195220522152225223522452255226522752285229523052315232523352345235523652375238523952405241524252435244524552465247524852495250525152525253525452555256525752585259526052615262526352645265526652675268526952705271527252735274527552765277527852795280528152825283528452855286528752885289529052915292529352945295529652975298529953005301530253035304530553065307530853095310531153125313531453155316531753185319532053215322532353245325532653275328532953305331533253335334533553365337533853395340534153425343534453455346534753485349535053515352535353545355535653575358535953605361536253635364536553665367536853695370537153725373537453755376537753785379538053815382538353845385538653875388538953905391539253935394539553965397539853995400540154025403540454055406540754085409541054115412541354145415541654175418541954205421542254235424542554265427542854295430543154325433543454355436543754385439544054415442544354445445544654475448544954505451545254535454545554565457545854595460546154625463546454655466546754685469547054715472547354745475547654775478547954805481548254835484548554865487548854895490549154925493549454955496549754985499550055015502550355045505550655075508550955105511551255135514551555165517551855195520552155225523552455255526552755285529553055315532553355345535553655375538553955405541554255435544554555465547554855495550555155525553555455555556555755585559556055615562556355645565556655675568556955705571557255735574557555765577557855795580558155825583558455855586558755885589559055915592559355945595559655975598559956005601560256035604560556065607560856095610561156125613561456155616561756185619562056215622562356245625562656275628562956305631563256335634563556365637563856395640564156425643564456455646564756485649565056515652565356545655565656575658565956605661566256635664566556665667566856695670567156725673567456755676567756785679568056815682568356845685568656875688568956905691569256935694569556965697569856995700570157025703570457055706570757085709571057115712571357145715571657175718571957205721572257235724572557265727572857295730573157325733573457355736573757385739574057415742574357445745574657475748574957505751575257535754575557565757575857595760576157625763576457655766576757685769577057715772577357745775577657775778577957805781578257835784578557865787578857895790579157925793579457955796579757985799580058015802580358045805580658075808580958105811581258135814581558165817581858195820582158225823582458255826582758285829583058315832583358345835583658375838583958405841584258435844584558465847584858495850585158525853585458555856585758585859586058615862586358645865586658675868586958705871587258735874587558765877587858795880588158825883588458855886588758885889589058915892589358945895589658975898589959005901590259035904590559065907590859095910591159125913591459155916591759185919592059215922592359245925592659275928592959305931593259335934593559365937593859395940594159425943594459455946594759485949595059515952595359545955595659575958595959605961596259635964596559665967596859695970597159725973
  1. /*
  2. * Copyright (C) 2007 Oracle. All rights reserved.
  3. *
  4. * This program is free software; you can redistribute it and/or
  5. * modify it under the terms of the GNU General Public
  6. * License v2 as published by the Free Software Foundation.
  7. *
  8. * This program is distributed in the hope that it will be useful,
  9. * but WITHOUT ANY WARRANTY; without even the implied warranty of
  10. * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
  11. * General Public License for more details.
  12. *
  13. * You should have received a copy of the GNU General Public
  14. * License along with this program; if not, write to the
  15. * Free Software Foundation, Inc., 59 Temple Place - Suite 330,
  16. * Boston, MA 021110-1307, USA.
  17. */
  18. #include <linux/sched.h>
  19. #include <linux/pagemap.h>
  20. #include <linux/writeback.h>
  21. #include <linux/blkdev.h>
  22. #include <linux/version.h>
  23. #include "compat.h"
  24. #include "hash.h"
  25. #include "crc32c.h"
  26. #include "ctree.h"
  27. #include "disk-io.h"
  28. #include "print-tree.h"
  29. #include "transaction.h"
  30. #include "volumes.h"
  31. #include "locking.h"
  32. #include "ref-cache.h"
  33. #include "compat.h"
  34. #define PENDING_EXTENT_INSERT 0
  35. #define PENDING_EXTENT_DELETE 1
  36. #define PENDING_BACKREF_UPDATE 2
  37. struct pending_extent_op {
  38. int type;
  39. u64 bytenr;
  40. u64 num_bytes;
  41. u64 parent;
  42. u64 orig_parent;
  43. u64 generation;
  44. u64 orig_generation;
  45. int level;
  46. struct list_head list;
  47. int del;
  48. };
  49. static int finish_current_insert(struct btrfs_trans_handle *trans, struct
  50. btrfs_root *extent_root, int all);
  51. static int del_pending_extents(struct btrfs_trans_handle *trans, struct
  52. btrfs_root *extent_root, int all);
  53. static int pin_down_bytes(struct btrfs_trans_handle *trans,
  54. struct btrfs_root *root,
  55. u64 bytenr, u64 num_bytes, int is_data);
  56. static int update_block_group(struct btrfs_trans_handle *trans,
  57. struct btrfs_root *root,
  58. u64 bytenr, u64 num_bytes, int alloc,
  59. int mark_free);
  60. static int block_group_bits(struct btrfs_block_group_cache *cache, u64 bits)
  61. {
  62. return (cache->flags & bits) == bits;
  63. }
  64. /*
  65. * this adds the block group to the fs_info rb tree for the block group
  66. * cache
  67. */
  68. static int btrfs_add_block_group_cache(struct btrfs_fs_info *info,
  69. struct btrfs_block_group_cache *block_group)
  70. {
  71. struct rb_node **p;
  72. struct rb_node *parent = NULL;
  73. struct btrfs_block_group_cache *cache;
  74. spin_lock(&info->block_group_cache_lock);
  75. p = &info->block_group_cache_tree.rb_node;
  76. while (*p) {
  77. parent = *p;
  78. cache = rb_entry(parent, struct btrfs_block_group_cache,
  79. cache_node);
  80. if (block_group->key.objectid < cache->key.objectid) {
  81. p = &(*p)->rb_left;
  82. } else if (block_group->key.objectid > cache->key.objectid) {
  83. p = &(*p)->rb_right;
  84. } else {
  85. spin_unlock(&info->block_group_cache_lock);
  86. return -EEXIST;
  87. }
  88. }
  89. rb_link_node(&block_group->cache_node, parent, p);
  90. rb_insert_color(&block_group->cache_node,
  91. &info->block_group_cache_tree);
  92. spin_unlock(&info->block_group_cache_lock);
  93. return 0;
  94. }
  95. /*
  96. * This will return the block group at or after bytenr if contains is 0, else
  97. * it will return the block group that contains the bytenr
  98. */
  99. static struct btrfs_block_group_cache *
  100. block_group_cache_tree_search(struct btrfs_fs_info *info, u64 bytenr,
  101. int contains)
  102. {
  103. struct btrfs_block_group_cache *cache, *ret = NULL;
  104. struct rb_node *n;
  105. u64 end, start;
  106. spin_lock(&info->block_group_cache_lock);
  107. n = info->block_group_cache_tree.rb_node;
  108. while (n) {
  109. cache = rb_entry(n, struct btrfs_block_group_cache,
  110. cache_node);
  111. end = cache->key.objectid + cache->key.offset - 1;
  112. start = cache->key.objectid;
  113. if (bytenr < start) {
  114. if (!contains && (!ret || start < ret->key.objectid))
  115. ret = cache;
  116. n = n->rb_left;
  117. } else if (bytenr > start) {
  118. if (contains && bytenr <= end) {
  119. ret = cache;
  120. break;
  121. }
  122. n = n->rb_right;
  123. } else {
  124. ret = cache;
  125. break;
  126. }
  127. }
  128. if (ret)
  129. atomic_inc(&ret->count);
  130. spin_unlock(&info->block_group_cache_lock);
  131. return ret;
  132. }
  133. /*
  134. * this is only called by cache_block_group, since we could have freed extents
  135. * we need to check the pinned_extents for any extents that can't be used yet
  136. * since their free space will be released as soon as the transaction commits.
  137. */
  138. static int add_new_free_space(struct btrfs_block_group_cache *block_group,
  139. struct btrfs_fs_info *info, u64 start, u64 end)
  140. {
  141. u64 extent_start, extent_end, size;
  142. int ret;
  143. mutex_lock(&info->pinned_mutex);
  144. while (start < end) {
  145. ret = find_first_extent_bit(&info->pinned_extents, start,
  146. &extent_start, &extent_end,
  147. EXTENT_DIRTY);
  148. if (ret)
  149. break;
  150. if (extent_start == start) {
  151. start = extent_end + 1;
  152. } else if (extent_start > start && extent_start < end) {
  153. size = extent_start - start;
  154. ret = btrfs_add_free_space(block_group, start,
  155. size);
  156. BUG_ON(ret);
  157. start = extent_end + 1;
  158. } else {
  159. break;
  160. }
  161. }
  162. if (start < end) {
  163. size = end - start;
  164. ret = btrfs_add_free_space(block_group, start, size);
  165. BUG_ON(ret);
  166. }
  167. mutex_unlock(&info->pinned_mutex);
  168. return 0;
  169. }
  170. static int remove_sb_from_cache(struct btrfs_root *root,
  171. struct btrfs_block_group_cache *cache)
  172. {
  173. u64 bytenr;
  174. u64 *logical;
  175. int stripe_len;
  176. int i, nr, ret;
  177. for (i = 0; i < BTRFS_SUPER_MIRROR_MAX; i++) {
  178. bytenr = btrfs_sb_offset(i);
  179. ret = btrfs_rmap_block(&root->fs_info->mapping_tree,
  180. cache->key.objectid, bytenr, 0,
  181. &logical, &nr, &stripe_len);
  182. BUG_ON(ret);
  183. while (nr--) {
  184. btrfs_remove_free_space(cache, logical[nr],
  185. stripe_len);
  186. }
  187. kfree(logical);
  188. }
  189. return 0;
  190. }
  191. static int cache_block_group(struct btrfs_root *root,
  192. struct btrfs_block_group_cache *block_group)
  193. {
  194. struct btrfs_path *path;
  195. int ret = 0;
  196. struct btrfs_key key;
  197. struct extent_buffer *leaf;
  198. int slot;
  199. u64 last;
  200. if (!block_group)
  201. return 0;
  202. root = root->fs_info->extent_root;
  203. if (block_group->cached)
  204. return 0;
  205. path = btrfs_alloc_path();
  206. if (!path)
  207. return -ENOMEM;
  208. path->reada = 2;
  209. /*
  210. * we get into deadlocks with paths held by callers of this function.
  211. * since the alloc_mutex is protecting things right now, just
  212. * skip the locking here
  213. */
  214. path->skip_locking = 1;
  215. last = max_t(u64, block_group->key.objectid, BTRFS_SUPER_INFO_OFFSET);
  216. key.objectid = last;
  217. key.offset = 0;
  218. btrfs_set_key_type(&key, BTRFS_EXTENT_ITEM_KEY);
  219. ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
  220. if (ret < 0)
  221. goto err;
  222. while(1) {
  223. leaf = path->nodes[0];
  224. slot = path->slots[0];
  225. if (slot >= btrfs_header_nritems(leaf)) {
  226. ret = btrfs_next_leaf(root, path);
  227. if (ret < 0)
  228. goto err;
  229. if (ret == 0)
  230. continue;
  231. else
  232. break;
  233. }
  234. btrfs_item_key_to_cpu(leaf, &key, slot);
  235. if (key.objectid < block_group->key.objectid)
  236. goto next;
  237. if (key.objectid >= block_group->key.objectid +
  238. block_group->key.offset)
  239. break;
  240. if (btrfs_key_type(&key) == BTRFS_EXTENT_ITEM_KEY) {
  241. add_new_free_space(block_group, root->fs_info, last,
  242. key.objectid);
  243. last = key.objectid + key.offset;
  244. }
  245. next:
  246. path->slots[0]++;
  247. }
  248. add_new_free_space(block_group, root->fs_info, last,
  249. block_group->key.objectid +
  250. block_group->key.offset);
  251. remove_sb_from_cache(root, block_group);
  252. block_group->cached = 1;
  253. ret = 0;
  254. err:
  255. btrfs_free_path(path);
  256. return ret;
  257. }
  258. /*
  259. * return the block group that starts at or after bytenr
  260. */
  261. static struct btrfs_block_group_cache *btrfs_lookup_first_block_group(struct
  262. btrfs_fs_info *info,
  263. u64 bytenr)
  264. {
  265. struct btrfs_block_group_cache *cache;
  266. cache = block_group_cache_tree_search(info, bytenr, 0);
  267. return cache;
  268. }
  269. /*
  270. * return the block group that contains teh given bytenr
  271. */
  272. struct btrfs_block_group_cache *btrfs_lookup_block_group(struct
  273. btrfs_fs_info *info,
  274. u64 bytenr)
  275. {
  276. struct btrfs_block_group_cache *cache;
  277. cache = block_group_cache_tree_search(info, bytenr, 1);
  278. return cache;
  279. }
  280. static inline void put_block_group(struct btrfs_block_group_cache *cache)
  281. {
  282. if (atomic_dec_and_test(&cache->count))
  283. kfree(cache);
  284. }
  285. static struct btrfs_space_info *__find_space_info(struct btrfs_fs_info *info,
  286. u64 flags)
  287. {
  288. struct list_head *head = &info->space_info;
  289. struct list_head *cur;
  290. struct btrfs_space_info *found;
  291. list_for_each(cur, head) {
  292. found = list_entry(cur, struct btrfs_space_info, list);
  293. if (found->flags == flags)
  294. return found;
  295. }
  296. return NULL;
  297. }
  298. static u64 div_factor(u64 num, int factor)
  299. {
  300. if (factor == 10)
  301. return num;
  302. num *= factor;
  303. do_div(num, 10);
  304. return num;
  305. }
  306. u64 btrfs_find_block_group(struct btrfs_root *root,
  307. u64 search_start, u64 search_hint, int owner)
  308. {
  309. struct btrfs_block_group_cache *cache;
  310. u64 used;
  311. u64 last = max(search_hint, search_start);
  312. u64 group_start = 0;
  313. int full_search = 0;
  314. int factor = 9;
  315. int wrapped = 0;
  316. again:
  317. while (1) {
  318. cache = btrfs_lookup_first_block_group(root->fs_info, last);
  319. if (!cache)
  320. break;
  321. spin_lock(&cache->lock);
  322. last = cache->key.objectid + cache->key.offset;
  323. used = btrfs_block_group_used(&cache->item);
  324. if ((full_search || !cache->ro) &&
  325. block_group_bits(cache, BTRFS_BLOCK_GROUP_METADATA)) {
  326. if (used + cache->pinned + cache->reserved <
  327. div_factor(cache->key.offset, factor)) {
  328. group_start = cache->key.objectid;
  329. spin_unlock(&cache->lock);
  330. put_block_group(cache);
  331. goto found;
  332. }
  333. }
  334. spin_unlock(&cache->lock);
  335. put_block_group(cache);
  336. cond_resched();
  337. }
  338. if (!wrapped) {
  339. last = search_start;
  340. wrapped = 1;
  341. goto again;
  342. }
  343. if (!full_search && factor < 10) {
  344. last = search_start;
  345. full_search = 1;
  346. factor = 10;
  347. goto again;
  348. }
  349. found:
  350. return group_start;
  351. }
  352. /* simple helper to search for an existing extent at a given offset */
  353. int btrfs_lookup_extent(struct btrfs_root *root, u64 start, u64 len)
  354. {
  355. int ret;
  356. struct btrfs_key key;
  357. struct btrfs_path *path;
  358. path = btrfs_alloc_path();
  359. BUG_ON(!path);
  360. key.objectid = start;
  361. key.offset = len;
  362. btrfs_set_key_type(&key, BTRFS_EXTENT_ITEM_KEY);
  363. ret = btrfs_search_slot(NULL, root->fs_info->extent_root, &key, path,
  364. 0, 0);
  365. btrfs_free_path(path);
  366. return ret;
  367. }
  368. /*
  369. * Back reference rules. Back refs have three main goals:
  370. *
  371. * 1) differentiate between all holders of references to an extent so that
  372. * when a reference is dropped we can make sure it was a valid reference
  373. * before freeing the extent.
  374. *
  375. * 2) Provide enough information to quickly find the holders of an extent
  376. * if we notice a given block is corrupted or bad.
  377. *
  378. * 3) Make it easy to migrate blocks for FS shrinking or storage pool
  379. * maintenance. This is actually the same as #2, but with a slightly
  380. * different use case.
  381. *
  382. * File extents can be referenced by:
  383. *
  384. * - multiple snapshots, subvolumes, or different generations in one subvol
  385. * - different files inside a single subvolume
  386. * - different offsets inside a file (bookend extents in file.c)
  387. *
  388. * The extent ref structure has fields for:
  389. *
  390. * - Objectid of the subvolume root
  391. * - Generation number of the tree holding the reference
  392. * - objectid of the file holding the reference
  393. * - number of references holding by parent node (alway 1 for tree blocks)
  394. *
  395. * Btree leaf may hold multiple references to a file extent. In most cases,
  396. * these references are from same file and the corresponding offsets inside
  397. * the file are close together.
  398. *
  399. * When a file extent is allocated the fields are filled in:
  400. * (root_key.objectid, trans->transid, inode objectid, 1)
  401. *
  402. * When a leaf is cow'd new references are added for every file extent found
  403. * in the leaf. It looks similar to the create case, but trans->transid will
  404. * be different when the block is cow'd.
  405. *
  406. * (root_key.objectid, trans->transid, inode objectid,
  407. * number of references in the leaf)
  408. *
  409. * When a file extent is removed either during snapshot deletion or
  410. * file truncation, we find the corresponding back reference and check
  411. * the following fields:
  412. *
  413. * (btrfs_header_owner(leaf), btrfs_header_generation(leaf),
  414. * inode objectid)
  415. *
  416. * Btree extents can be referenced by:
  417. *
  418. * - Different subvolumes
  419. * - Different generations of the same subvolume
  420. *
  421. * When a tree block is created, back references are inserted:
  422. *
  423. * (root->root_key.objectid, trans->transid, level, 1)
  424. *
  425. * When a tree block is cow'd, new back references are added for all the
  426. * blocks it points to. If the tree block isn't in reference counted root,
  427. * the old back references are removed. These new back references are of
  428. * the form (trans->transid will have increased since creation):
  429. *
  430. * (root->root_key.objectid, trans->transid, level, 1)
  431. *
  432. * When a backref is in deleting, the following fields are checked:
  433. *
  434. * if backref was for a tree root:
  435. * (btrfs_header_owner(itself), btrfs_header_generation(itself), level)
  436. * else
  437. * (btrfs_header_owner(parent), btrfs_header_generation(parent), level)
  438. *
  439. * Back Reference Key composing:
  440. *
  441. * The key objectid corresponds to the first byte in the extent, the key
  442. * type is set to BTRFS_EXTENT_REF_KEY, and the key offset is the first
  443. * byte of parent extent. If a extent is tree root, the key offset is set
  444. * to the key objectid.
  445. */
  446. static int noinline lookup_extent_backref(struct btrfs_trans_handle *trans,
  447. struct btrfs_root *root,
  448. struct btrfs_path *path,
  449. u64 bytenr, u64 parent,
  450. u64 ref_root, u64 ref_generation,
  451. u64 owner_objectid, int del)
  452. {
  453. struct btrfs_key key;
  454. struct btrfs_extent_ref *ref;
  455. struct extent_buffer *leaf;
  456. u64 ref_objectid;
  457. int ret;
  458. key.objectid = bytenr;
  459. key.type = BTRFS_EXTENT_REF_KEY;
  460. key.offset = parent;
  461. ret = btrfs_search_slot(trans, root, &key, path, del ? -1 : 0, 1);
  462. if (ret < 0)
  463. goto out;
  464. if (ret > 0) {
  465. ret = -ENOENT;
  466. goto out;
  467. }
  468. leaf = path->nodes[0];
  469. ref = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_ref);
  470. ref_objectid = btrfs_ref_objectid(leaf, ref);
  471. if (btrfs_ref_root(leaf, ref) != ref_root ||
  472. btrfs_ref_generation(leaf, ref) != ref_generation ||
  473. (ref_objectid != owner_objectid &&
  474. ref_objectid != BTRFS_MULTIPLE_OBJECTIDS)) {
  475. ret = -EIO;
  476. WARN_ON(1);
  477. goto out;
  478. }
  479. ret = 0;
  480. out:
  481. return ret;
  482. }
  483. /*
  484. * updates all the backrefs that are pending on update_list for the
  485. * extent_root
  486. */
  487. static int noinline update_backrefs(struct btrfs_trans_handle *trans,
  488. struct btrfs_root *extent_root,
  489. struct btrfs_path *path,
  490. struct list_head *update_list)
  491. {
  492. struct btrfs_key key;
  493. struct btrfs_extent_ref *ref;
  494. struct btrfs_fs_info *info = extent_root->fs_info;
  495. struct pending_extent_op *op;
  496. struct extent_buffer *leaf;
  497. int ret = 0;
  498. struct list_head *cur = update_list->next;
  499. u64 ref_objectid;
  500. u64 ref_root = extent_root->root_key.objectid;
  501. op = list_entry(cur, struct pending_extent_op, list);
  502. search:
  503. key.objectid = op->bytenr;
  504. key.type = BTRFS_EXTENT_REF_KEY;
  505. key.offset = op->orig_parent;
  506. ret = btrfs_search_slot(trans, extent_root, &key, path, 0, 1);
  507. BUG_ON(ret);
  508. leaf = path->nodes[0];
  509. loop:
  510. ref = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_ref);
  511. ref_objectid = btrfs_ref_objectid(leaf, ref);
  512. if (btrfs_ref_root(leaf, ref) != ref_root ||
  513. btrfs_ref_generation(leaf, ref) != op->orig_generation ||
  514. (ref_objectid != op->level &&
  515. ref_objectid != BTRFS_MULTIPLE_OBJECTIDS)) {
  516. printk(KERN_ERR "couldn't find %Lu, parent %Lu, root %Lu, "
  517. "owner %u\n", op->bytenr, op->orig_parent,
  518. ref_root, op->level);
  519. btrfs_print_leaf(extent_root, leaf);
  520. BUG();
  521. }
  522. key.objectid = op->bytenr;
  523. key.offset = op->parent;
  524. key.type = BTRFS_EXTENT_REF_KEY;
  525. ret = btrfs_set_item_key_safe(trans, extent_root, path, &key);
  526. BUG_ON(ret);
  527. ref = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_ref);
  528. btrfs_set_ref_generation(leaf, ref, op->generation);
  529. cur = cur->next;
  530. list_del_init(&op->list);
  531. unlock_extent(&info->extent_ins, op->bytenr,
  532. op->bytenr + op->num_bytes - 1, GFP_NOFS);
  533. kfree(op);
  534. if (cur == update_list) {
  535. btrfs_mark_buffer_dirty(path->nodes[0]);
  536. btrfs_release_path(extent_root, path);
  537. goto out;
  538. }
  539. op = list_entry(cur, struct pending_extent_op, list);
  540. path->slots[0]++;
  541. while (path->slots[0] < btrfs_header_nritems(leaf)) {
  542. btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
  543. if (key.objectid == op->bytenr &&
  544. key.type == BTRFS_EXTENT_REF_KEY)
  545. goto loop;
  546. path->slots[0]++;
  547. }
  548. btrfs_mark_buffer_dirty(path->nodes[0]);
  549. btrfs_release_path(extent_root, path);
  550. goto search;
  551. out:
  552. return 0;
  553. }
  554. static int noinline insert_extents(struct btrfs_trans_handle *trans,
  555. struct btrfs_root *extent_root,
  556. struct btrfs_path *path,
  557. struct list_head *insert_list, int nr)
  558. {
  559. struct btrfs_key *keys;
  560. u32 *data_size;
  561. struct pending_extent_op *op;
  562. struct extent_buffer *leaf;
  563. struct list_head *cur = insert_list->next;
  564. struct btrfs_fs_info *info = extent_root->fs_info;
  565. u64 ref_root = extent_root->root_key.objectid;
  566. int i = 0, last = 0, ret;
  567. int total = nr * 2;
  568. if (!nr)
  569. return 0;
  570. keys = kzalloc(total * sizeof(struct btrfs_key), GFP_NOFS);
  571. if (!keys)
  572. return -ENOMEM;
  573. data_size = kzalloc(total * sizeof(u32), GFP_NOFS);
  574. if (!data_size) {
  575. kfree(keys);
  576. return -ENOMEM;
  577. }
  578. list_for_each_entry(op, insert_list, list) {
  579. keys[i].objectid = op->bytenr;
  580. keys[i].offset = op->num_bytes;
  581. keys[i].type = BTRFS_EXTENT_ITEM_KEY;
  582. data_size[i] = sizeof(struct btrfs_extent_item);
  583. i++;
  584. keys[i].objectid = op->bytenr;
  585. keys[i].offset = op->parent;
  586. keys[i].type = BTRFS_EXTENT_REF_KEY;
  587. data_size[i] = sizeof(struct btrfs_extent_ref);
  588. i++;
  589. }
  590. op = list_entry(cur, struct pending_extent_op, list);
  591. i = 0;
  592. while (i < total) {
  593. int c;
  594. ret = btrfs_insert_some_items(trans, extent_root, path,
  595. keys+i, data_size+i, total-i);
  596. BUG_ON(ret < 0);
  597. if (last && ret > 1)
  598. BUG();
  599. leaf = path->nodes[0];
  600. for (c = 0; c < ret; c++) {
  601. int ref_first = keys[i].type == BTRFS_EXTENT_REF_KEY;
  602. /*
  603. * if the first item we inserted was a backref, then
  604. * the EXTENT_ITEM will be the odd c's, else it will
  605. * be the even c's
  606. */
  607. if ((ref_first && (c % 2)) ||
  608. (!ref_first && !(c % 2))) {
  609. struct btrfs_extent_item *itm;
  610. itm = btrfs_item_ptr(leaf, path->slots[0] + c,
  611. struct btrfs_extent_item);
  612. btrfs_set_extent_refs(path->nodes[0], itm, 1);
  613. op->del++;
  614. } else {
  615. struct btrfs_extent_ref *ref;
  616. ref = btrfs_item_ptr(leaf, path->slots[0] + c,
  617. struct btrfs_extent_ref);
  618. btrfs_set_ref_root(leaf, ref, ref_root);
  619. btrfs_set_ref_generation(leaf, ref,
  620. op->generation);
  621. btrfs_set_ref_objectid(leaf, ref, op->level);
  622. btrfs_set_ref_num_refs(leaf, ref, 1);
  623. op->del++;
  624. }
  625. /*
  626. * using del to see when its ok to free up the
  627. * pending_extent_op. In the case where we insert the
  628. * last item on the list in order to help do batching
  629. * we need to not free the extent op until we actually
  630. * insert the extent_item
  631. */
  632. if (op->del == 2) {
  633. unlock_extent(&info->extent_ins, op->bytenr,
  634. op->bytenr + op->num_bytes - 1,
  635. GFP_NOFS);
  636. cur = cur->next;
  637. list_del_init(&op->list);
  638. kfree(op);
  639. if (cur != insert_list)
  640. op = list_entry(cur,
  641. struct pending_extent_op,
  642. list);
  643. }
  644. }
  645. btrfs_mark_buffer_dirty(leaf);
  646. btrfs_release_path(extent_root, path);
  647. /*
  648. * Ok backref's and items usually go right next to eachother,
  649. * but if we could only insert 1 item that means that we
  650. * inserted on the end of a leaf, and we have no idea what may
  651. * be on the next leaf so we just play it safe. In order to
  652. * try and help this case we insert the last thing on our
  653. * insert list so hopefully it will end up being the last
  654. * thing on the leaf and everything else will be before it,
  655. * which will let us insert a whole bunch of items at the same
  656. * time.
  657. */
  658. if (ret == 1 && !last && (i + ret < total)) {
  659. /*
  660. * last: where we will pick up the next time around
  661. * i: our current key to insert, will be total - 1
  662. * cur: the current op we are screwing with
  663. * op: duh
  664. */
  665. last = i + ret;
  666. i = total - 1;
  667. cur = insert_list->prev;
  668. op = list_entry(cur, struct pending_extent_op, list);
  669. } else if (last) {
  670. /*
  671. * ok we successfully inserted the last item on the
  672. * list, lets reset everything
  673. *
  674. * i: our current key to insert, so where we left off
  675. * last time
  676. * last: done with this
  677. * cur: the op we are messing with
  678. * op: duh
  679. * total: since we inserted the last key, we need to
  680. * decrement total so we dont overflow
  681. */
  682. i = last;
  683. last = 0;
  684. total--;
  685. if (i < total) {
  686. cur = insert_list->next;
  687. op = list_entry(cur, struct pending_extent_op,
  688. list);
  689. }
  690. } else {
  691. i += ret;
  692. }
  693. cond_resched();
  694. }
  695. ret = 0;
  696. kfree(keys);
  697. kfree(data_size);
  698. return ret;
  699. }
  700. static int noinline insert_extent_backref(struct btrfs_trans_handle *trans,
  701. struct btrfs_root *root,
  702. struct btrfs_path *path,
  703. u64 bytenr, u64 parent,
  704. u64 ref_root, u64 ref_generation,
  705. u64 owner_objectid)
  706. {
  707. struct btrfs_key key;
  708. struct extent_buffer *leaf;
  709. struct btrfs_extent_ref *ref;
  710. u32 num_refs;
  711. int ret;
  712. key.objectid = bytenr;
  713. key.type = BTRFS_EXTENT_REF_KEY;
  714. key.offset = parent;
  715. ret = btrfs_insert_empty_item(trans, root, path, &key, sizeof(*ref));
  716. if (ret == 0) {
  717. leaf = path->nodes[0];
  718. ref = btrfs_item_ptr(leaf, path->slots[0],
  719. struct btrfs_extent_ref);
  720. btrfs_set_ref_root(leaf, ref, ref_root);
  721. btrfs_set_ref_generation(leaf, ref, ref_generation);
  722. btrfs_set_ref_objectid(leaf, ref, owner_objectid);
  723. btrfs_set_ref_num_refs(leaf, ref, 1);
  724. } else if (ret == -EEXIST) {
  725. u64 existing_owner;
  726. BUG_ON(owner_objectid < BTRFS_FIRST_FREE_OBJECTID);
  727. leaf = path->nodes[0];
  728. ref = btrfs_item_ptr(leaf, path->slots[0],
  729. struct btrfs_extent_ref);
  730. if (btrfs_ref_root(leaf, ref) != ref_root ||
  731. btrfs_ref_generation(leaf, ref) != ref_generation) {
  732. ret = -EIO;
  733. WARN_ON(1);
  734. goto out;
  735. }
  736. num_refs = btrfs_ref_num_refs(leaf, ref);
  737. BUG_ON(num_refs == 0);
  738. btrfs_set_ref_num_refs(leaf, ref, num_refs + 1);
  739. existing_owner = btrfs_ref_objectid(leaf, ref);
  740. if (existing_owner != owner_objectid &&
  741. existing_owner != BTRFS_MULTIPLE_OBJECTIDS) {
  742. btrfs_set_ref_objectid(leaf, ref,
  743. BTRFS_MULTIPLE_OBJECTIDS);
  744. }
  745. ret = 0;
  746. } else {
  747. goto out;
  748. }
  749. btrfs_mark_buffer_dirty(path->nodes[0]);
  750. out:
  751. btrfs_release_path(root, path);
  752. return ret;
  753. }
  754. static int noinline remove_extent_backref(struct btrfs_trans_handle *trans,
  755. struct btrfs_root *root,
  756. struct btrfs_path *path)
  757. {
  758. struct extent_buffer *leaf;
  759. struct btrfs_extent_ref *ref;
  760. u32 num_refs;
  761. int ret = 0;
  762. leaf = path->nodes[0];
  763. ref = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_ref);
  764. num_refs = btrfs_ref_num_refs(leaf, ref);
  765. BUG_ON(num_refs == 0);
  766. num_refs -= 1;
  767. if (num_refs == 0) {
  768. ret = btrfs_del_item(trans, root, path);
  769. } else {
  770. btrfs_set_ref_num_refs(leaf, ref, num_refs);
  771. btrfs_mark_buffer_dirty(leaf);
  772. }
  773. btrfs_release_path(root, path);
  774. return ret;
  775. }
  776. #ifdef BIO_RW_DISCARD
  777. static void btrfs_issue_discard(struct block_device *bdev,
  778. u64 start, u64 len)
  779. {
  780. #if LINUX_VERSION_CODE >= KERNEL_VERSION(2,6,28)
  781. blkdev_issue_discard(bdev, start >> 9, len >> 9, GFP_KERNEL);
  782. #else
  783. blkdev_issue_discard(bdev, start >> 9, len >> 9);
  784. #endif
  785. }
  786. #endif
  787. static int noinline free_extents(struct btrfs_trans_handle *trans,
  788. struct btrfs_root *extent_root,
  789. struct list_head *del_list)
  790. {
  791. struct btrfs_fs_info *info = extent_root->fs_info;
  792. struct btrfs_path *path;
  793. struct btrfs_key key, found_key;
  794. struct extent_buffer *leaf;
  795. struct list_head *cur;
  796. struct pending_extent_op *op;
  797. struct btrfs_extent_item *ei;
  798. int ret, num_to_del, extent_slot = 0, found_extent = 0;
  799. u32 refs;
  800. u64 bytes_freed = 0;
  801. path = btrfs_alloc_path();
  802. if (!path)
  803. return -ENOMEM;
  804. path->reada = 1;
  805. search:
  806. /* search for the backref for the current ref we want to delete */
  807. cur = del_list->next;
  808. op = list_entry(cur, struct pending_extent_op, list);
  809. ret = lookup_extent_backref(trans, extent_root, path, op->bytenr,
  810. op->orig_parent,
  811. extent_root->root_key.objectid,
  812. op->orig_generation, op->level, 1);
  813. if (ret) {
  814. printk("Unable to find backref byte nr %Lu root %Lu gen %Lu "
  815. "owner %u\n", op->bytenr,
  816. extent_root->root_key.objectid, op->orig_generation,
  817. op->level);
  818. btrfs_print_leaf(extent_root, path->nodes[0]);
  819. WARN_ON(1);
  820. goto out;
  821. }
  822. extent_slot = path->slots[0];
  823. num_to_del = 1;
  824. found_extent = 0;
  825. /*
  826. * if we aren't the first item on the leaf we can move back one and see
  827. * if our ref is right next to our extent item
  828. */
  829. if (likely(extent_slot)) {
  830. extent_slot--;
  831. btrfs_item_key_to_cpu(path->nodes[0], &found_key,
  832. extent_slot);
  833. if (found_key.objectid == op->bytenr &&
  834. found_key.type == BTRFS_EXTENT_ITEM_KEY &&
  835. found_key.offset == op->num_bytes) {
  836. num_to_del++;
  837. found_extent = 1;
  838. }
  839. }
  840. /*
  841. * if we didn't find the extent we need to delete the backref and then
  842. * search for the extent item key so we can update its ref count
  843. */
  844. if (!found_extent) {
  845. key.objectid = op->bytenr;
  846. key.type = BTRFS_EXTENT_ITEM_KEY;
  847. key.offset = op->num_bytes;
  848. ret = remove_extent_backref(trans, extent_root, path);
  849. BUG_ON(ret);
  850. btrfs_release_path(extent_root, path);
  851. ret = btrfs_search_slot(trans, extent_root, &key, path, -1, 1);
  852. BUG_ON(ret);
  853. extent_slot = path->slots[0];
  854. }
  855. /* this is where we update the ref count for the extent */
  856. leaf = path->nodes[0];
  857. ei = btrfs_item_ptr(leaf, extent_slot, struct btrfs_extent_item);
  858. refs = btrfs_extent_refs(leaf, ei);
  859. BUG_ON(refs == 0);
  860. refs--;
  861. btrfs_set_extent_refs(leaf, ei, refs);
  862. btrfs_mark_buffer_dirty(leaf);
  863. /*
  864. * This extent needs deleting. The reason cur_slot is extent_slot +
  865. * num_to_del is because extent_slot points to the slot where the extent
  866. * is, and if the backref was not right next to the extent we will be
  867. * deleting at least 1 item, and will want to start searching at the
  868. * slot directly next to extent_slot. However if we did find the
  869. * backref next to the extent item them we will be deleting at least 2
  870. * items and will want to start searching directly after the ref slot
  871. */
  872. if (!refs) {
  873. struct list_head *pos, *n, *end;
  874. int cur_slot = extent_slot+num_to_del;
  875. u64 super_used;
  876. u64 root_used;
  877. path->slots[0] = extent_slot;
  878. bytes_freed = op->num_bytes;
  879. mutex_lock(&info->pinned_mutex);
  880. ret = pin_down_bytes(trans, extent_root, op->bytenr,
  881. op->num_bytes, op->level >=
  882. BTRFS_FIRST_FREE_OBJECTID);
  883. mutex_unlock(&info->pinned_mutex);
  884. BUG_ON(ret < 0);
  885. op->del = ret;
  886. /*
  887. * we need to see if we can delete multiple things at once, so
  888. * start looping through the list of extents we are wanting to
  889. * delete and see if their extent/backref's are right next to
  890. * eachother and the extents only have 1 ref
  891. */
  892. for (pos = cur->next; pos != del_list; pos = pos->next) {
  893. struct pending_extent_op *tmp;
  894. tmp = list_entry(pos, struct pending_extent_op, list);
  895. /* we only want to delete extent+ref at this stage */
  896. if (cur_slot >= btrfs_header_nritems(leaf) - 1)
  897. break;
  898. btrfs_item_key_to_cpu(leaf, &found_key, cur_slot);
  899. if (found_key.objectid != tmp->bytenr ||
  900. found_key.type != BTRFS_EXTENT_ITEM_KEY ||
  901. found_key.offset != tmp->num_bytes)
  902. break;
  903. /* check to make sure this extent only has one ref */
  904. ei = btrfs_item_ptr(leaf, cur_slot,
  905. struct btrfs_extent_item);
  906. if (btrfs_extent_refs(leaf, ei) != 1)
  907. break;
  908. btrfs_item_key_to_cpu(leaf, &found_key, cur_slot+1);
  909. if (found_key.objectid != tmp->bytenr ||
  910. found_key.type != BTRFS_EXTENT_REF_KEY ||
  911. found_key.offset != tmp->orig_parent)
  912. break;
  913. /*
  914. * the ref is right next to the extent, we can set the
  915. * ref count to 0 since we will delete them both now
  916. */
  917. btrfs_set_extent_refs(leaf, ei, 0);
  918. /* pin down the bytes for this extent */
  919. mutex_lock(&info->pinned_mutex);
  920. ret = pin_down_bytes(trans, extent_root, tmp->bytenr,
  921. tmp->num_bytes, tmp->level >=
  922. BTRFS_FIRST_FREE_OBJECTID);
  923. mutex_unlock(&info->pinned_mutex);
  924. BUG_ON(ret < 0);
  925. /*
  926. * use the del field to tell if we need to go ahead and
  927. * free up the extent when we delete the item or not.
  928. */
  929. tmp->del = ret;
  930. bytes_freed += tmp->num_bytes;
  931. num_to_del += 2;
  932. cur_slot += 2;
  933. }
  934. end = pos;
  935. /* update the free space counters */
  936. spin_lock_irq(&info->delalloc_lock);
  937. super_used = btrfs_super_bytes_used(&info->super_copy);
  938. btrfs_set_super_bytes_used(&info->super_copy,
  939. super_used - bytes_freed);
  940. spin_unlock_irq(&info->delalloc_lock);
  941. root_used = btrfs_root_used(&extent_root->root_item);
  942. btrfs_set_root_used(&extent_root->root_item,
  943. root_used - bytes_freed);
  944. /* delete the items */
  945. ret = btrfs_del_items(trans, extent_root, path,
  946. path->slots[0], num_to_del);
  947. BUG_ON(ret);
  948. /*
  949. * loop through the extents we deleted and do the cleanup work
  950. * on them
  951. */
  952. for (pos = cur, n = pos->next; pos != end;
  953. pos = n, n = pos->next) {
  954. struct pending_extent_op *tmp;
  955. #ifdef BIO_RW_DISCARD
  956. u64 map_length;
  957. struct btrfs_multi_bio *multi = NULL;
  958. #endif
  959. tmp = list_entry(pos, struct pending_extent_op, list);
  960. /*
  961. * remember tmp->del tells us wether or not we pinned
  962. * down the extent
  963. */
  964. ret = update_block_group(trans, extent_root,
  965. tmp->bytenr, tmp->num_bytes, 0,
  966. tmp->del);
  967. BUG_ON(ret);
  968. #ifdef BIO_RW_DISCARD
  969. map_length = tmp->num_bytes;
  970. ret = btrfs_map_block(&info->mapping_tree, READ,
  971. tmp->bytenr, &map_length, &multi,
  972. 0);
  973. if (!ret) {
  974. struct btrfs_bio_stripe *stripe;
  975. int i;
  976. stripe = multi->stripes;
  977. if (map_length > tmp->num_bytes)
  978. map_length = tmp->num_bytes;
  979. for (i = 0; i < multi->num_stripes;
  980. i++, stripe++)
  981. btrfs_issue_discard(stripe->dev->bdev,
  982. stripe->physical,
  983. map_length);
  984. kfree(multi);
  985. }
  986. #endif
  987. list_del_init(&tmp->list);
  988. unlock_extent(&info->extent_ins, tmp->bytenr,
  989. tmp->bytenr + tmp->num_bytes - 1,
  990. GFP_NOFS);
  991. kfree(tmp);
  992. }
  993. } else if (refs && found_extent) {
  994. /*
  995. * the ref and extent were right next to eachother, but the
  996. * extent still has a ref, so just free the backref and keep
  997. * going
  998. */
  999. ret = remove_extent_backref(trans, extent_root, path);
  1000. BUG_ON(ret);
  1001. list_del_init(&op->list);
  1002. unlock_extent(&info->extent_ins, op->bytenr,
  1003. op->bytenr + op->num_bytes - 1, GFP_NOFS);
  1004. kfree(op);
  1005. } else {
  1006. /*
  1007. * the extent has multiple refs and the backref we were looking
  1008. * for was not right next to it, so just unlock and go next,
  1009. * we're good to go
  1010. */
  1011. list_del_init(&op->list);
  1012. unlock_extent(&info->extent_ins, op->bytenr,
  1013. op->bytenr + op->num_bytes - 1, GFP_NOFS);
  1014. kfree(op);
  1015. }
  1016. btrfs_release_path(extent_root, path);
  1017. if (!list_empty(del_list))
  1018. goto search;
  1019. out:
  1020. btrfs_free_path(path);
  1021. return ret;
  1022. }
  1023. static int __btrfs_update_extent_ref(struct btrfs_trans_handle *trans,
  1024. struct btrfs_root *root, u64 bytenr,
  1025. u64 orig_parent, u64 parent,
  1026. u64 orig_root, u64 ref_root,
  1027. u64 orig_generation, u64 ref_generation,
  1028. u64 owner_objectid)
  1029. {
  1030. int ret;
  1031. struct btrfs_root *extent_root = root->fs_info->extent_root;
  1032. struct btrfs_path *path;
  1033. if (root == root->fs_info->extent_root) {
  1034. struct pending_extent_op *extent_op;
  1035. u64 num_bytes;
  1036. BUG_ON(owner_objectid >= BTRFS_MAX_LEVEL);
  1037. num_bytes = btrfs_level_size(root, (int)owner_objectid);
  1038. mutex_lock(&root->fs_info->extent_ins_mutex);
  1039. if (test_range_bit(&root->fs_info->extent_ins, bytenr,
  1040. bytenr + num_bytes - 1, EXTENT_WRITEBACK, 0)) {
  1041. u64 priv;
  1042. ret = get_state_private(&root->fs_info->extent_ins,
  1043. bytenr, &priv);
  1044. BUG_ON(ret);
  1045. extent_op = (struct pending_extent_op *)
  1046. (unsigned long)priv;
  1047. BUG_ON(extent_op->parent != orig_parent);
  1048. BUG_ON(extent_op->generation != orig_generation);
  1049. extent_op->parent = parent;
  1050. extent_op->generation = ref_generation;
  1051. } else {
  1052. extent_op = kmalloc(sizeof(*extent_op), GFP_NOFS);
  1053. BUG_ON(!extent_op);
  1054. extent_op->type = PENDING_BACKREF_UPDATE;
  1055. extent_op->bytenr = bytenr;
  1056. extent_op->num_bytes = num_bytes;
  1057. extent_op->parent = parent;
  1058. extent_op->orig_parent = orig_parent;
  1059. extent_op->generation = ref_generation;
  1060. extent_op->orig_generation = orig_generation;
  1061. extent_op->level = (int)owner_objectid;
  1062. INIT_LIST_HEAD(&extent_op->list);
  1063. extent_op->del = 0;
  1064. set_extent_bits(&root->fs_info->extent_ins,
  1065. bytenr, bytenr + num_bytes - 1,
  1066. EXTENT_WRITEBACK, GFP_NOFS);
  1067. set_state_private(&root->fs_info->extent_ins,
  1068. bytenr, (unsigned long)extent_op);
  1069. }
  1070. mutex_unlock(&root->fs_info->extent_ins_mutex);
  1071. return 0;
  1072. }
  1073. path = btrfs_alloc_path();
  1074. if (!path)
  1075. return -ENOMEM;
  1076. ret = lookup_extent_backref(trans, extent_root, path,
  1077. bytenr, orig_parent, orig_root,
  1078. orig_generation, owner_objectid, 1);
  1079. if (ret)
  1080. goto out;
  1081. ret = remove_extent_backref(trans, extent_root, path);
  1082. if (ret)
  1083. goto out;
  1084. ret = insert_extent_backref(trans, extent_root, path, bytenr,
  1085. parent, ref_root, ref_generation,
  1086. owner_objectid);
  1087. BUG_ON(ret);
  1088. finish_current_insert(trans, extent_root, 0);
  1089. del_pending_extents(trans, extent_root, 0);
  1090. out:
  1091. btrfs_free_path(path);
  1092. return ret;
  1093. }
  1094. int btrfs_update_extent_ref(struct btrfs_trans_handle *trans,
  1095. struct btrfs_root *root, u64 bytenr,
  1096. u64 orig_parent, u64 parent,
  1097. u64 ref_root, u64 ref_generation,
  1098. u64 owner_objectid)
  1099. {
  1100. int ret;
  1101. if (ref_root == BTRFS_TREE_LOG_OBJECTID &&
  1102. owner_objectid < BTRFS_FIRST_FREE_OBJECTID)
  1103. return 0;
  1104. ret = __btrfs_update_extent_ref(trans, root, bytenr, orig_parent,
  1105. parent, ref_root, ref_root,
  1106. ref_generation, ref_generation,
  1107. owner_objectid);
  1108. return ret;
  1109. }
  1110. static int __btrfs_inc_extent_ref(struct btrfs_trans_handle *trans,
  1111. struct btrfs_root *root, u64 bytenr,
  1112. u64 orig_parent, u64 parent,
  1113. u64 orig_root, u64 ref_root,
  1114. u64 orig_generation, u64 ref_generation,
  1115. u64 owner_objectid)
  1116. {
  1117. struct btrfs_path *path;
  1118. int ret;
  1119. struct btrfs_key key;
  1120. struct extent_buffer *l;
  1121. struct btrfs_extent_item *item;
  1122. u32 refs;
  1123. path = btrfs_alloc_path();
  1124. if (!path)
  1125. return -ENOMEM;
  1126. path->reada = 1;
  1127. key.objectid = bytenr;
  1128. key.type = BTRFS_EXTENT_ITEM_KEY;
  1129. key.offset = (u64)-1;
  1130. ret = btrfs_search_slot(trans, root->fs_info->extent_root, &key, path,
  1131. 0, 1);
  1132. if (ret < 0)
  1133. return ret;
  1134. BUG_ON(ret == 0 || path->slots[0] == 0);
  1135. path->slots[0]--;
  1136. l = path->nodes[0];
  1137. btrfs_item_key_to_cpu(l, &key, path->slots[0]);
  1138. if (key.objectid != bytenr) {
  1139. btrfs_print_leaf(root->fs_info->extent_root, path->nodes[0]);
  1140. printk("wanted %Lu found %Lu\n", bytenr, key.objectid);
  1141. BUG();
  1142. }
  1143. BUG_ON(key.type != BTRFS_EXTENT_ITEM_KEY);
  1144. item = btrfs_item_ptr(l, path->slots[0], struct btrfs_extent_item);
  1145. refs = btrfs_extent_refs(l, item);
  1146. btrfs_set_extent_refs(l, item, refs + 1);
  1147. btrfs_mark_buffer_dirty(path->nodes[0]);
  1148. btrfs_release_path(root->fs_info->extent_root, path);
  1149. path->reada = 1;
  1150. ret = insert_extent_backref(trans, root->fs_info->extent_root,
  1151. path, bytenr, parent,
  1152. ref_root, ref_generation,
  1153. owner_objectid);
  1154. BUG_ON(ret);
  1155. finish_current_insert(trans, root->fs_info->extent_root, 0);
  1156. del_pending_extents(trans, root->fs_info->extent_root, 0);
  1157. btrfs_free_path(path);
  1158. return 0;
  1159. }
  1160. int btrfs_inc_extent_ref(struct btrfs_trans_handle *trans,
  1161. struct btrfs_root *root,
  1162. u64 bytenr, u64 num_bytes, u64 parent,
  1163. u64 ref_root, u64 ref_generation,
  1164. u64 owner_objectid)
  1165. {
  1166. int ret;
  1167. if (ref_root == BTRFS_TREE_LOG_OBJECTID &&
  1168. owner_objectid < BTRFS_FIRST_FREE_OBJECTID)
  1169. return 0;
  1170. ret = __btrfs_inc_extent_ref(trans, root, bytenr, 0, parent,
  1171. 0, ref_root, 0, ref_generation,
  1172. owner_objectid);
  1173. return ret;
  1174. }
  1175. int btrfs_extent_post_op(struct btrfs_trans_handle *trans,
  1176. struct btrfs_root *root)
  1177. {
  1178. finish_current_insert(trans, root->fs_info->extent_root, 1);
  1179. del_pending_extents(trans, root->fs_info->extent_root, 1);
  1180. return 0;
  1181. }
  1182. int btrfs_lookup_extent_ref(struct btrfs_trans_handle *trans,
  1183. struct btrfs_root *root, u64 bytenr,
  1184. u64 num_bytes, u32 *refs)
  1185. {
  1186. struct btrfs_path *path;
  1187. int ret;
  1188. struct btrfs_key key;
  1189. struct extent_buffer *l;
  1190. struct btrfs_extent_item *item;
  1191. WARN_ON(num_bytes < root->sectorsize);
  1192. path = btrfs_alloc_path();
  1193. path->reada = 1;
  1194. key.objectid = bytenr;
  1195. key.offset = num_bytes;
  1196. btrfs_set_key_type(&key, BTRFS_EXTENT_ITEM_KEY);
  1197. ret = btrfs_search_slot(trans, root->fs_info->extent_root, &key, path,
  1198. 0, 0);
  1199. if (ret < 0)
  1200. goto out;
  1201. if (ret != 0) {
  1202. btrfs_print_leaf(root, path->nodes[0]);
  1203. printk("failed to find block number %Lu\n", bytenr);
  1204. BUG();
  1205. }
  1206. l = path->nodes[0];
  1207. item = btrfs_item_ptr(l, path->slots[0], struct btrfs_extent_item);
  1208. *refs = btrfs_extent_refs(l, item);
  1209. out:
  1210. btrfs_free_path(path);
  1211. return 0;
  1212. }
  1213. int btrfs_cross_ref_exist(struct btrfs_trans_handle *trans,
  1214. struct btrfs_root *root, u64 objectid, u64 bytenr)
  1215. {
  1216. struct btrfs_root *extent_root = root->fs_info->extent_root;
  1217. struct btrfs_path *path;
  1218. struct extent_buffer *leaf;
  1219. struct btrfs_extent_ref *ref_item;
  1220. struct btrfs_key key;
  1221. struct btrfs_key found_key;
  1222. u64 ref_root;
  1223. u64 last_snapshot;
  1224. u32 nritems;
  1225. int ret;
  1226. key.objectid = bytenr;
  1227. key.offset = (u64)-1;
  1228. key.type = BTRFS_EXTENT_ITEM_KEY;
  1229. path = btrfs_alloc_path();
  1230. ret = btrfs_search_slot(NULL, extent_root, &key, path, 0, 0);
  1231. if (ret < 0)
  1232. goto out;
  1233. BUG_ON(ret == 0);
  1234. ret = -ENOENT;
  1235. if (path->slots[0] == 0)
  1236. goto out;
  1237. path->slots[0]--;
  1238. leaf = path->nodes[0];
  1239. btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
  1240. if (found_key.objectid != bytenr ||
  1241. found_key.type != BTRFS_EXTENT_ITEM_KEY)
  1242. goto out;
  1243. last_snapshot = btrfs_root_last_snapshot(&root->root_item);
  1244. while (1) {
  1245. leaf = path->nodes[0];
  1246. nritems = btrfs_header_nritems(leaf);
  1247. if (path->slots[0] >= nritems) {
  1248. ret = btrfs_next_leaf(extent_root, path);
  1249. if (ret < 0)
  1250. goto out;
  1251. if (ret == 0)
  1252. continue;
  1253. break;
  1254. }
  1255. btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
  1256. if (found_key.objectid != bytenr)
  1257. break;
  1258. if (found_key.type != BTRFS_EXTENT_REF_KEY) {
  1259. path->slots[0]++;
  1260. continue;
  1261. }
  1262. ref_item = btrfs_item_ptr(leaf, path->slots[0],
  1263. struct btrfs_extent_ref);
  1264. ref_root = btrfs_ref_root(leaf, ref_item);
  1265. if ((ref_root != root->root_key.objectid &&
  1266. ref_root != BTRFS_TREE_LOG_OBJECTID) ||
  1267. objectid != btrfs_ref_objectid(leaf, ref_item)) {
  1268. ret = 1;
  1269. goto out;
  1270. }
  1271. if (btrfs_ref_generation(leaf, ref_item) <= last_snapshot) {
  1272. ret = 1;
  1273. goto out;
  1274. }
  1275. path->slots[0]++;
  1276. }
  1277. ret = 0;
  1278. out:
  1279. btrfs_free_path(path);
  1280. return ret;
  1281. }
  1282. int btrfs_cache_ref(struct btrfs_trans_handle *trans, struct btrfs_root *root,
  1283. struct extent_buffer *buf, u32 nr_extents)
  1284. {
  1285. struct btrfs_key key;
  1286. struct btrfs_file_extent_item *fi;
  1287. u64 root_gen;
  1288. u32 nritems;
  1289. int i;
  1290. int level;
  1291. int ret = 0;
  1292. int shared = 0;
  1293. if (!root->ref_cows)
  1294. return 0;
  1295. if (root->root_key.objectid != BTRFS_TREE_RELOC_OBJECTID) {
  1296. shared = 0;
  1297. root_gen = root->root_key.offset;
  1298. } else {
  1299. shared = 1;
  1300. root_gen = trans->transid - 1;
  1301. }
  1302. level = btrfs_header_level(buf);
  1303. nritems = btrfs_header_nritems(buf);
  1304. if (level == 0) {
  1305. struct btrfs_leaf_ref *ref;
  1306. struct btrfs_extent_info *info;
  1307. ref = btrfs_alloc_leaf_ref(root, nr_extents);
  1308. if (!ref) {
  1309. ret = -ENOMEM;
  1310. goto out;
  1311. }
  1312. ref->root_gen = root_gen;
  1313. ref->bytenr = buf->start;
  1314. ref->owner = btrfs_header_owner(buf);
  1315. ref->generation = btrfs_header_generation(buf);
  1316. ref->nritems = nr_extents;
  1317. info = ref->extents;
  1318. for (i = 0; nr_extents > 0 && i < nritems; i++) {
  1319. u64 disk_bytenr;
  1320. btrfs_item_key_to_cpu(buf, &key, i);
  1321. if (btrfs_key_type(&key) != BTRFS_EXTENT_DATA_KEY)
  1322. continue;
  1323. fi = btrfs_item_ptr(buf, i,
  1324. struct btrfs_file_extent_item);
  1325. if (btrfs_file_extent_type(buf, fi) ==
  1326. BTRFS_FILE_EXTENT_INLINE)
  1327. continue;
  1328. disk_bytenr = btrfs_file_extent_disk_bytenr(buf, fi);
  1329. if (disk_bytenr == 0)
  1330. continue;
  1331. info->bytenr = disk_bytenr;
  1332. info->num_bytes =
  1333. btrfs_file_extent_disk_num_bytes(buf, fi);
  1334. info->objectid = key.objectid;
  1335. info->offset = key.offset;
  1336. info++;
  1337. }
  1338. ret = btrfs_add_leaf_ref(root, ref, shared);
  1339. if (ret == -EEXIST && shared) {
  1340. struct btrfs_leaf_ref *old;
  1341. old = btrfs_lookup_leaf_ref(root, ref->bytenr);
  1342. BUG_ON(!old);
  1343. btrfs_remove_leaf_ref(root, old);
  1344. btrfs_free_leaf_ref(root, old);
  1345. ret = btrfs_add_leaf_ref(root, ref, shared);
  1346. }
  1347. WARN_ON(ret);
  1348. btrfs_free_leaf_ref(root, ref);
  1349. }
  1350. out:
  1351. return ret;
  1352. }
  1353. int btrfs_inc_ref(struct btrfs_trans_handle *trans, struct btrfs_root *root,
  1354. struct extent_buffer *orig_buf, struct extent_buffer *buf,
  1355. u32 *nr_extents)
  1356. {
  1357. u64 bytenr;
  1358. u64 ref_root;
  1359. u64 orig_root;
  1360. u64 ref_generation;
  1361. u64 orig_generation;
  1362. u32 nritems;
  1363. u32 nr_file_extents = 0;
  1364. struct btrfs_key key;
  1365. struct btrfs_file_extent_item *fi;
  1366. int i;
  1367. int level;
  1368. int ret = 0;
  1369. int faili = 0;
  1370. int (*process_func)(struct btrfs_trans_handle *, struct btrfs_root *,
  1371. u64, u64, u64, u64, u64, u64, u64, u64);
  1372. ref_root = btrfs_header_owner(buf);
  1373. ref_generation = btrfs_header_generation(buf);
  1374. orig_root = btrfs_header_owner(orig_buf);
  1375. orig_generation = btrfs_header_generation(orig_buf);
  1376. nritems = btrfs_header_nritems(buf);
  1377. level = btrfs_header_level(buf);
  1378. if (root->ref_cows) {
  1379. process_func = __btrfs_inc_extent_ref;
  1380. } else {
  1381. if (level == 0 &&
  1382. root->root_key.objectid != BTRFS_TREE_LOG_OBJECTID)
  1383. goto out;
  1384. if (level != 0 &&
  1385. root->root_key.objectid == BTRFS_TREE_LOG_OBJECTID)
  1386. goto out;
  1387. process_func = __btrfs_update_extent_ref;
  1388. }
  1389. for (i = 0; i < nritems; i++) {
  1390. cond_resched();
  1391. if (level == 0) {
  1392. btrfs_item_key_to_cpu(buf, &key, i);
  1393. if (btrfs_key_type(&key) != BTRFS_EXTENT_DATA_KEY)
  1394. continue;
  1395. fi = btrfs_item_ptr(buf, i,
  1396. struct btrfs_file_extent_item);
  1397. if (btrfs_file_extent_type(buf, fi) ==
  1398. BTRFS_FILE_EXTENT_INLINE)
  1399. continue;
  1400. bytenr = btrfs_file_extent_disk_bytenr(buf, fi);
  1401. if (bytenr == 0)
  1402. continue;
  1403. nr_file_extents++;
  1404. ret = process_func(trans, root, bytenr,
  1405. orig_buf->start, buf->start,
  1406. orig_root, ref_root,
  1407. orig_generation, ref_generation,
  1408. key.objectid);
  1409. if (ret) {
  1410. faili = i;
  1411. WARN_ON(1);
  1412. goto fail;
  1413. }
  1414. } else {
  1415. bytenr = btrfs_node_blockptr(buf, i);
  1416. ret = process_func(trans, root, bytenr,
  1417. orig_buf->start, buf->start,
  1418. orig_root, ref_root,
  1419. orig_generation, ref_generation,
  1420. level - 1);
  1421. if (ret) {
  1422. faili = i;
  1423. WARN_ON(1);
  1424. goto fail;
  1425. }
  1426. }
  1427. }
  1428. out:
  1429. if (nr_extents) {
  1430. if (level == 0)
  1431. *nr_extents = nr_file_extents;
  1432. else
  1433. *nr_extents = nritems;
  1434. }
  1435. return 0;
  1436. fail:
  1437. WARN_ON(1);
  1438. return ret;
  1439. }
  1440. int btrfs_update_ref(struct btrfs_trans_handle *trans,
  1441. struct btrfs_root *root, struct extent_buffer *orig_buf,
  1442. struct extent_buffer *buf, int start_slot, int nr)
  1443. {
  1444. u64 bytenr;
  1445. u64 ref_root;
  1446. u64 orig_root;
  1447. u64 ref_generation;
  1448. u64 orig_generation;
  1449. struct btrfs_key key;
  1450. struct btrfs_file_extent_item *fi;
  1451. int i;
  1452. int ret;
  1453. int slot;
  1454. int level;
  1455. BUG_ON(start_slot < 0);
  1456. BUG_ON(start_slot + nr > btrfs_header_nritems(buf));
  1457. ref_root = btrfs_header_owner(buf);
  1458. ref_generation = btrfs_header_generation(buf);
  1459. orig_root = btrfs_header_owner(orig_buf);
  1460. orig_generation = btrfs_header_generation(orig_buf);
  1461. level = btrfs_header_level(buf);
  1462. if (!root->ref_cows) {
  1463. if (level == 0 &&
  1464. root->root_key.objectid != BTRFS_TREE_LOG_OBJECTID)
  1465. return 0;
  1466. if (level != 0 &&
  1467. root->root_key.objectid == BTRFS_TREE_LOG_OBJECTID)
  1468. return 0;
  1469. }
  1470. for (i = 0, slot = start_slot; i < nr; i++, slot++) {
  1471. cond_resched();
  1472. if (level == 0) {
  1473. btrfs_item_key_to_cpu(buf, &key, slot);
  1474. if (btrfs_key_type(&key) != BTRFS_EXTENT_DATA_KEY)
  1475. continue;
  1476. fi = btrfs_item_ptr(buf, slot,
  1477. struct btrfs_file_extent_item);
  1478. if (btrfs_file_extent_type(buf, fi) ==
  1479. BTRFS_FILE_EXTENT_INLINE)
  1480. continue;
  1481. bytenr = btrfs_file_extent_disk_bytenr(buf, fi);
  1482. if (bytenr == 0)
  1483. continue;
  1484. ret = __btrfs_update_extent_ref(trans, root, bytenr,
  1485. orig_buf->start, buf->start,
  1486. orig_root, ref_root,
  1487. orig_generation, ref_generation,
  1488. key.objectid);
  1489. if (ret)
  1490. goto fail;
  1491. } else {
  1492. bytenr = btrfs_node_blockptr(buf, slot);
  1493. ret = __btrfs_update_extent_ref(trans, root, bytenr,
  1494. orig_buf->start, buf->start,
  1495. orig_root, ref_root,
  1496. orig_generation, ref_generation,
  1497. level - 1);
  1498. if (ret)
  1499. goto fail;
  1500. }
  1501. }
  1502. return 0;
  1503. fail:
  1504. WARN_ON(1);
  1505. return -1;
  1506. }
  1507. static int write_one_cache_group(struct btrfs_trans_handle *trans,
  1508. struct btrfs_root *root,
  1509. struct btrfs_path *path,
  1510. struct btrfs_block_group_cache *cache)
  1511. {
  1512. int ret;
  1513. int pending_ret;
  1514. struct btrfs_root *extent_root = root->fs_info->extent_root;
  1515. unsigned long bi;
  1516. struct extent_buffer *leaf;
  1517. ret = btrfs_search_slot(trans, extent_root, &cache->key, path, 0, 1);
  1518. if (ret < 0)
  1519. goto fail;
  1520. BUG_ON(ret);
  1521. leaf = path->nodes[0];
  1522. bi = btrfs_item_ptr_offset(leaf, path->slots[0]);
  1523. write_extent_buffer(leaf, &cache->item, bi, sizeof(cache->item));
  1524. btrfs_mark_buffer_dirty(leaf);
  1525. btrfs_release_path(extent_root, path);
  1526. fail:
  1527. finish_current_insert(trans, extent_root, 0);
  1528. pending_ret = del_pending_extents(trans, extent_root, 0);
  1529. if (ret)
  1530. return ret;
  1531. if (pending_ret)
  1532. return pending_ret;
  1533. return 0;
  1534. }
  1535. int btrfs_write_dirty_block_groups(struct btrfs_trans_handle *trans,
  1536. struct btrfs_root *root)
  1537. {
  1538. struct btrfs_block_group_cache *cache, *entry;
  1539. struct rb_node *n;
  1540. int err = 0;
  1541. int werr = 0;
  1542. struct btrfs_path *path;
  1543. u64 last = 0;
  1544. path = btrfs_alloc_path();
  1545. if (!path)
  1546. return -ENOMEM;
  1547. while(1) {
  1548. cache = NULL;
  1549. spin_lock(&root->fs_info->block_group_cache_lock);
  1550. for (n = rb_first(&root->fs_info->block_group_cache_tree);
  1551. n; n = rb_next(n)) {
  1552. entry = rb_entry(n, struct btrfs_block_group_cache,
  1553. cache_node);
  1554. if (entry->dirty) {
  1555. cache = entry;
  1556. break;
  1557. }
  1558. }
  1559. spin_unlock(&root->fs_info->block_group_cache_lock);
  1560. if (!cache)
  1561. break;
  1562. cache->dirty = 0;
  1563. last += cache->key.offset;
  1564. err = write_one_cache_group(trans, root,
  1565. path, cache);
  1566. /*
  1567. * if we fail to write the cache group, we want
  1568. * to keep it marked dirty in hopes that a later
  1569. * write will work
  1570. */
  1571. if (err) {
  1572. werr = err;
  1573. continue;
  1574. }
  1575. }
  1576. btrfs_free_path(path);
  1577. return werr;
  1578. }
  1579. int btrfs_extent_readonly(struct btrfs_root *root, u64 bytenr)
  1580. {
  1581. struct btrfs_block_group_cache *block_group;
  1582. int readonly = 0;
  1583. block_group = btrfs_lookup_block_group(root->fs_info, bytenr);
  1584. if (!block_group || block_group->ro)
  1585. readonly = 1;
  1586. if (block_group)
  1587. put_block_group(block_group);
  1588. return readonly;
  1589. }
  1590. static int update_space_info(struct btrfs_fs_info *info, u64 flags,
  1591. u64 total_bytes, u64 bytes_used,
  1592. struct btrfs_space_info **space_info)
  1593. {
  1594. struct btrfs_space_info *found;
  1595. found = __find_space_info(info, flags);
  1596. if (found) {
  1597. spin_lock(&found->lock);
  1598. found->total_bytes += total_bytes;
  1599. found->bytes_used += bytes_used;
  1600. found->full = 0;
  1601. spin_unlock(&found->lock);
  1602. *space_info = found;
  1603. return 0;
  1604. }
  1605. found = kzalloc(sizeof(*found), GFP_NOFS);
  1606. if (!found)
  1607. return -ENOMEM;
  1608. list_add(&found->list, &info->space_info);
  1609. INIT_LIST_HEAD(&found->block_groups);
  1610. init_rwsem(&found->groups_sem);
  1611. spin_lock_init(&found->lock);
  1612. found->flags = flags;
  1613. found->total_bytes = total_bytes;
  1614. found->bytes_used = bytes_used;
  1615. found->bytes_pinned = 0;
  1616. found->bytes_reserved = 0;
  1617. found->bytes_readonly = 0;
  1618. found->full = 0;
  1619. found->force_alloc = 0;
  1620. *space_info = found;
  1621. return 0;
  1622. }
  1623. static void set_avail_alloc_bits(struct btrfs_fs_info *fs_info, u64 flags)
  1624. {
  1625. u64 extra_flags = flags & (BTRFS_BLOCK_GROUP_RAID0 |
  1626. BTRFS_BLOCK_GROUP_RAID1 |
  1627. BTRFS_BLOCK_GROUP_RAID10 |
  1628. BTRFS_BLOCK_GROUP_DUP);
  1629. if (extra_flags) {
  1630. if (flags & BTRFS_BLOCK_GROUP_DATA)
  1631. fs_info->avail_data_alloc_bits |= extra_flags;
  1632. if (flags & BTRFS_BLOCK_GROUP_METADATA)
  1633. fs_info->avail_metadata_alloc_bits |= extra_flags;
  1634. if (flags & BTRFS_BLOCK_GROUP_SYSTEM)
  1635. fs_info->avail_system_alloc_bits |= extra_flags;
  1636. }
  1637. }
  1638. static void set_block_group_readonly(struct btrfs_block_group_cache *cache)
  1639. {
  1640. spin_lock(&cache->space_info->lock);
  1641. spin_lock(&cache->lock);
  1642. if (!cache->ro) {
  1643. cache->space_info->bytes_readonly += cache->key.offset -
  1644. btrfs_block_group_used(&cache->item);
  1645. cache->ro = 1;
  1646. }
  1647. spin_unlock(&cache->lock);
  1648. spin_unlock(&cache->space_info->lock);
  1649. }
  1650. u64 btrfs_reduce_alloc_profile(struct btrfs_root *root, u64 flags)
  1651. {
  1652. u64 num_devices = root->fs_info->fs_devices->rw_devices;
  1653. if (num_devices == 1)
  1654. flags &= ~(BTRFS_BLOCK_GROUP_RAID1 | BTRFS_BLOCK_GROUP_RAID0);
  1655. if (num_devices < 4)
  1656. flags &= ~BTRFS_BLOCK_GROUP_RAID10;
  1657. if ((flags & BTRFS_BLOCK_GROUP_DUP) &&
  1658. (flags & (BTRFS_BLOCK_GROUP_RAID1 |
  1659. BTRFS_BLOCK_GROUP_RAID10))) {
  1660. flags &= ~BTRFS_BLOCK_GROUP_DUP;
  1661. }
  1662. if ((flags & BTRFS_BLOCK_GROUP_RAID1) &&
  1663. (flags & BTRFS_BLOCK_GROUP_RAID10)) {
  1664. flags &= ~BTRFS_BLOCK_GROUP_RAID1;
  1665. }
  1666. if ((flags & BTRFS_BLOCK_GROUP_RAID0) &&
  1667. ((flags & BTRFS_BLOCK_GROUP_RAID1) |
  1668. (flags & BTRFS_BLOCK_GROUP_RAID10) |
  1669. (flags & BTRFS_BLOCK_GROUP_DUP)))
  1670. flags &= ~BTRFS_BLOCK_GROUP_RAID0;
  1671. return flags;
  1672. }
  1673. static int do_chunk_alloc(struct btrfs_trans_handle *trans,
  1674. struct btrfs_root *extent_root, u64 alloc_bytes,
  1675. u64 flags, int force)
  1676. {
  1677. struct btrfs_space_info *space_info;
  1678. u64 thresh;
  1679. int ret = 0;
  1680. mutex_lock(&extent_root->fs_info->chunk_mutex);
  1681. flags = btrfs_reduce_alloc_profile(extent_root, flags);
  1682. space_info = __find_space_info(extent_root->fs_info, flags);
  1683. if (!space_info) {
  1684. ret = update_space_info(extent_root->fs_info, flags,
  1685. 0, 0, &space_info);
  1686. BUG_ON(ret);
  1687. }
  1688. BUG_ON(!space_info);
  1689. spin_lock(&space_info->lock);
  1690. if (space_info->force_alloc) {
  1691. force = 1;
  1692. space_info->force_alloc = 0;
  1693. }
  1694. if (space_info->full) {
  1695. spin_unlock(&space_info->lock);
  1696. goto out;
  1697. }
  1698. thresh = space_info->total_bytes - space_info->bytes_readonly;
  1699. thresh = div_factor(thresh, 6);
  1700. if (!force &&
  1701. (space_info->bytes_used + space_info->bytes_pinned +
  1702. space_info->bytes_reserved + alloc_bytes) < thresh) {
  1703. spin_unlock(&space_info->lock);
  1704. goto out;
  1705. }
  1706. spin_unlock(&space_info->lock);
  1707. ret = btrfs_alloc_chunk(trans, extent_root, flags);
  1708. if (ret) {
  1709. printk("space info full %Lu\n", flags);
  1710. space_info->full = 1;
  1711. }
  1712. out:
  1713. mutex_unlock(&extent_root->fs_info->chunk_mutex);
  1714. return ret;
  1715. }
  1716. static int update_block_group(struct btrfs_trans_handle *trans,
  1717. struct btrfs_root *root,
  1718. u64 bytenr, u64 num_bytes, int alloc,
  1719. int mark_free)
  1720. {
  1721. struct btrfs_block_group_cache *cache;
  1722. struct btrfs_fs_info *info = root->fs_info;
  1723. u64 total = num_bytes;
  1724. u64 old_val;
  1725. u64 byte_in_group;
  1726. while(total) {
  1727. cache = btrfs_lookup_block_group(info, bytenr);
  1728. if (!cache)
  1729. return -1;
  1730. byte_in_group = bytenr - cache->key.objectid;
  1731. WARN_ON(byte_in_group > cache->key.offset);
  1732. spin_lock(&cache->space_info->lock);
  1733. spin_lock(&cache->lock);
  1734. cache->dirty = 1;
  1735. old_val = btrfs_block_group_used(&cache->item);
  1736. num_bytes = min(total, cache->key.offset - byte_in_group);
  1737. if (alloc) {
  1738. old_val += num_bytes;
  1739. cache->space_info->bytes_used += num_bytes;
  1740. if (cache->ro)
  1741. cache->space_info->bytes_readonly -= num_bytes;
  1742. btrfs_set_block_group_used(&cache->item, old_val);
  1743. spin_unlock(&cache->lock);
  1744. spin_unlock(&cache->space_info->lock);
  1745. } else {
  1746. old_val -= num_bytes;
  1747. cache->space_info->bytes_used -= num_bytes;
  1748. if (cache->ro)
  1749. cache->space_info->bytes_readonly += num_bytes;
  1750. btrfs_set_block_group_used(&cache->item, old_val);
  1751. spin_unlock(&cache->lock);
  1752. spin_unlock(&cache->space_info->lock);
  1753. if (mark_free) {
  1754. int ret;
  1755. ret = btrfs_add_free_space(cache, bytenr,
  1756. num_bytes);
  1757. WARN_ON(ret);
  1758. }
  1759. }
  1760. put_block_group(cache);
  1761. total -= num_bytes;
  1762. bytenr += num_bytes;
  1763. }
  1764. return 0;
  1765. }
  1766. static u64 first_logical_byte(struct btrfs_root *root, u64 search_start)
  1767. {
  1768. struct btrfs_block_group_cache *cache;
  1769. u64 bytenr;
  1770. cache = btrfs_lookup_first_block_group(root->fs_info, search_start);
  1771. if (!cache)
  1772. return 0;
  1773. bytenr = cache->key.objectid;
  1774. put_block_group(cache);
  1775. return bytenr;
  1776. }
  1777. int btrfs_update_pinned_extents(struct btrfs_root *root,
  1778. u64 bytenr, u64 num, int pin)
  1779. {
  1780. u64 len;
  1781. struct btrfs_block_group_cache *cache;
  1782. struct btrfs_fs_info *fs_info = root->fs_info;
  1783. WARN_ON(!mutex_is_locked(&root->fs_info->pinned_mutex));
  1784. if (pin) {
  1785. set_extent_dirty(&fs_info->pinned_extents,
  1786. bytenr, bytenr + num - 1, GFP_NOFS);
  1787. } else {
  1788. clear_extent_dirty(&fs_info->pinned_extents,
  1789. bytenr, bytenr + num - 1, GFP_NOFS);
  1790. }
  1791. while (num > 0) {
  1792. cache = btrfs_lookup_block_group(fs_info, bytenr);
  1793. BUG_ON(!cache);
  1794. len = min(num, cache->key.offset -
  1795. (bytenr - cache->key.objectid));
  1796. if (pin) {
  1797. spin_lock(&cache->space_info->lock);
  1798. spin_lock(&cache->lock);
  1799. cache->pinned += len;
  1800. cache->space_info->bytes_pinned += len;
  1801. spin_unlock(&cache->lock);
  1802. spin_unlock(&cache->space_info->lock);
  1803. fs_info->total_pinned += len;
  1804. } else {
  1805. spin_lock(&cache->space_info->lock);
  1806. spin_lock(&cache->lock);
  1807. cache->pinned -= len;
  1808. cache->space_info->bytes_pinned -= len;
  1809. spin_unlock(&cache->lock);
  1810. spin_unlock(&cache->space_info->lock);
  1811. fs_info->total_pinned -= len;
  1812. if (cache->cached)
  1813. btrfs_add_free_space(cache, bytenr, len);
  1814. }
  1815. put_block_group(cache);
  1816. bytenr += len;
  1817. num -= len;
  1818. }
  1819. return 0;
  1820. }
  1821. static int update_reserved_extents(struct btrfs_root *root,
  1822. u64 bytenr, u64 num, int reserve)
  1823. {
  1824. u64 len;
  1825. struct btrfs_block_group_cache *cache;
  1826. struct btrfs_fs_info *fs_info = root->fs_info;
  1827. while (num > 0) {
  1828. cache = btrfs_lookup_block_group(fs_info, bytenr);
  1829. BUG_ON(!cache);
  1830. len = min(num, cache->key.offset -
  1831. (bytenr - cache->key.objectid));
  1832. spin_lock(&cache->space_info->lock);
  1833. spin_lock(&cache->lock);
  1834. if (reserve) {
  1835. cache->reserved += len;
  1836. cache->space_info->bytes_reserved += len;
  1837. } else {
  1838. cache->reserved -= len;
  1839. cache->space_info->bytes_reserved -= len;
  1840. }
  1841. spin_unlock(&cache->lock);
  1842. spin_unlock(&cache->space_info->lock);
  1843. put_block_group(cache);
  1844. bytenr += len;
  1845. num -= len;
  1846. }
  1847. return 0;
  1848. }
  1849. int btrfs_copy_pinned(struct btrfs_root *root, struct extent_io_tree *copy)
  1850. {
  1851. u64 last = 0;
  1852. u64 start;
  1853. u64 end;
  1854. struct extent_io_tree *pinned_extents = &root->fs_info->pinned_extents;
  1855. int ret;
  1856. mutex_lock(&root->fs_info->pinned_mutex);
  1857. while(1) {
  1858. ret = find_first_extent_bit(pinned_extents, last,
  1859. &start, &end, EXTENT_DIRTY);
  1860. if (ret)
  1861. break;
  1862. set_extent_dirty(copy, start, end, GFP_NOFS);
  1863. last = end + 1;
  1864. }
  1865. mutex_unlock(&root->fs_info->pinned_mutex);
  1866. return 0;
  1867. }
  1868. int btrfs_finish_extent_commit(struct btrfs_trans_handle *trans,
  1869. struct btrfs_root *root,
  1870. struct extent_io_tree *unpin)
  1871. {
  1872. u64 start;
  1873. u64 end;
  1874. int ret;
  1875. mutex_lock(&root->fs_info->pinned_mutex);
  1876. while(1) {
  1877. ret = find_first_extent_bit(unpin, 0, &start, &end,
  1878. EXTENT_DIRTY);
  1879. if (ret)
  1880. break;
  1881. btrfs_update_pinned_extents(root, start, end + 1 - start, 0);
  1882. clear_extent_dirty(unpin, start, end, GFP_NOFS);
  1883. if (need_resched()) {
  1884. mutex_unlock(&root->fs_info->pinned_mutex);
  1885. cond_resched();
  1886. mutex_lock(&root->fs_info->pinned_mutex);
  1887. }
  1888. }
  1889. mutex_unlock(&root->fs_info->pinned_mutex);
  1890. return 0;
  1891. }
  1892. static int finish_current_insert(struct btrfs_trans_handle *trans,
  1893. struct btrfs_root *extent_root, int all)
  1894. {
  1895. u64 start;
  1896. u64 end;
  1897. u64 priv;
  1898. u64 search = 0;
  1899. u64 skipped = 0;
  1900. struct btrfs_fs_info *info = extent_root->fs_info;
  1901. struct btrfs_path *path;
  1902. struct pending_extent_op *extent_op, *tmp;
  1903. struct list_head insert_list, update_list;
  1904. int ret;
  1905. int num_inserts = 0, max_inserts;
  1906. path = btrfs_alloc_path();
  1907. INIT_LIST_HEAD(&insert_list);
  1908. INIT_LIST_HEAD(&update_list);
  1909. max_inserts = extent_root->leafsize /
  1910. (2 * sizeof(struct btrfs_key) + 2 * sizeof(struct btrfs_item) +
  1911. sizeof(struct btrfs_extent_ref) +
  1912. sizeof(struct btrfs_extent_item));
  1913. again:
  1914. mutex_lock(&info->extent_ins_mutex);
  1915. while (1) {
  1916. ret = find_first_extent_bit(&info->extent_ins, search, &start,
  1917. &end, EXTENT_WRITEBACK);
  1918. if (ret) {
  1919. if (skipped && all && !num_inserts) {
  1920. skipped = 0;
  1921. search = 0;
  1922. continue;
  1923. }
  1924. mutex_unlock(&info->extent_ins_mutex);
  1925. break;
  1926. }
  1927. ret = try_lock_extent(&info->extent_ins, start, end, GFP_NOFS);
  1928. if (!ret) {
  1929. skipped = 1;
  1930. search = end + 1;
  1931. if (need_resched()) {
  1932. mutex_unlock(&info->extent_ins_mutex);
  1933. cond_resched();
  1934. mutex_lock(&info->extent_ins_mutex);
  1935. }
  1936. continue;
  1937. }
  1938. ret = get_state_private(&info->extent_ins, start, &priv);
  1939. BUG_ON(ret);
  1940. extent_op = (struct pending_extent_op *)(unsigned long) priv;
  1941. if (extent_op->type == PENDING_EXTENT_INSERT) {
  1942. num_inserts++;
  1943. list_add_tail(&extent_op->list, &insert_list);
  1944. search = end + 1;
  1945. if (num_inserts == max_inserts) {
  1946. mutex_unlock(&info->extent_ins_mutex);
  1947. break;
  1948. }
  1949. } else if (extent_op->type == PENDING_BACKREF_UPDATE) {
  1950. list_add_tail(&extent_op->list, &update_list);
  1951. search = end + 1;
  1952. } else {
  1953. BUG();
  1954. }
  1955. }
  1956. /*
  1957. * process the update list, clear the writeback bit for it, and if
  1958. * somebody marked this thing for deletion then just unlock it and be
  1959. * done, the free_extents will handle it
  1960. */
  1961. mutex_lock(&info->extent_ins_mutex);
  1962. list_for_each_entry_safe(extent_op, tmp, &update_list, list) {
  1963. clear_extent_bits(&info->extent_ins, extent_op->bytenr,
  1964. extent_op->bytenr + extent_op->num_bytes - 1,
  1965. EXTENT_WRITEBACK, GFP_NOFS);
  1966. if (extent_op->del) {
  1967. list_del_init(&extent_op->list);
  1968. unlock_extent(&info->extent_ins, extent_op->bytenr,
  1969. extent_op->bytenr + extent_op->num_bytes
  1970. - 1, GFP_NOFS);
  1971. kfree(extent_op);
  1972. }
  1973. }
  1974. mutex_unlock(&info->extent_ins_mutex);
  1975. /*
  1976. * still have things left on the update list, go ahead an update
  1977. * everything
  1978. */
  1979. if (!list_empty(&update_list)) {
  1980. ret = update_backrefs(trans, extent_root, path, &update_list);
  1981. BUG_ON(ret);
  1982. }
  1983. /*
  1984. * if no inserts need to be done, but we skipped some extents and we
  1985. * need to make sure everything is cleaned then reset everything and
  1986. * go back to the beginning
  1987. */
  1988. if (!num_inserts && all && skipped) {
  1989. search = 0;
  1990. skipped = 0;
  1991. INIT_LIST_HEAD(&update_list);
  1992. INIT_LIST_HEAD(&insert_list);
  1993. goto again;
  1994. } else if (!num_inserts) {
  1995. goto out;
  1996. }
  1997. /*
  1998. * process the insert extents list. Again if we are deleting this
  1999. * extent, then just unlock it, pin down the bytes if need be, and be
  2000. * done with it. Saves us from having to actually insert the extent
  2001. * into the tree and then subsequently come along and delete it
  2002. */
  2003. mutex_lock(&info->extent_ins_mutex);
  2004. list_for_each_entry_safe(extent_op, tmp, &insert_list, list) {
  2005. clear_extent_bits(&info->extent_ins, extent_op->bytenr,
  2006. extent_op->bytenr + extent_op->num_bytes - 1,
  2007. EXTENT_WRITEBACK, GFP_NOFS);
  2008. if (extent_op->del) {
  2009. list_del_init(&extent_op->list);
  2010. unlock_extent(&info->extent_ins, extent_op->bytenr,
  2011. extent_op->bytenr + extent_op->num_bytes
  2012. - 1, GFP_NOFS);
  2013. mutex_lock(&extent_root->fs_info->pinned_mutex);
  2014. ret = pin_down_bytes(trans, extent_root,
  2015. extent_op->bytenr,
  2016. extent_op->num_bytes, 0);
  2017. mutex_unlock(&extent_root->fs_info->pinned_mutex);
  2018. ret = update_block_group(trans, extent_root,
  2019. extent_op->bytenr,
  2020. extent_op->num_bytes,
  2021. 0, ret > 0);
  2022. BUG_ON(ret);
  2023. kfree(extent_op);
  2024. num_inserts--;
  2025. }
  2026. }
  2027. mutex_unlock(&info->extent_ins_mutex);
  2028. ret = insert_extents(trans, extent_root, path, &insert_list,
  2029. num_inserts);
  2030. BUG_ON(ret);
  2031. /*
  2032. * if we broke out of the loop in order to insert stuff because we hit
  2033. * the maximum number of inserts at a time we can handle, then loop
  2034. * back and pick up where we left off
  2035. */
  2036. if (num_inserts == max_inserts) {
  2037. INIT_LIST_HEAD(&insert_list);
  2038. INIT_LIST_HEAD(&update_list);
  2039. num_inserts = 0;
  2040. goto again;
  2041. }
  2042. /*
  2043. * again, if we need to make absolutely sure there are no more pending
  2044. * extent operations left and we know that we skipped some, go back to
  2045. * the beginning and do it all again
  2046. */
  2047. if (all && skipped) {
  2048. INIT_LIST_HEAD(&insert_list);
  2049. INIT_LIST_HEAD(&update_list);
  2050. search = 0;
  2051. skipped = 0;
  2052. num_inserts = 0;
  2053. goto again;
  2054. }
  2055. out:
  2056. btrfs_free_path(path);
  2057. return 0;
  2058. }
  2059. static int pin_down_bytes(struct btrfs_trans_handle *trans,
  2060. struct btrfs_root *root,
  2061. u64 bytenr, u64 num_bytes, int is_data)
  2062. {
  2063. int err = 0;
  2064. struct extent_buffer *buf;
  2065. if (is_data)
  2066. goto pinit;
  2067. buf = btrfs_find_tree_block(root, bytenr, num_bytes);
  2068. if (!buf)
  2069. goto pinit;
  2070. /* we can reuse a block if it hasn't been written
  2071. * and it is from this transaction. We can't
  2072. * reuse anything from the tree log root because
  2073. * it has tiny sub-transactions.
  2074. */
  2075. if (btrfs_buffer_uptodate(buf, 0) &&
  2076. btrfs_try_tree_lock(buf)) {
  2077. u64 header_owner = btrfs_header_owner(buf);
  2078. u64 header_transid = btrfs_header_generation(buf);
  2079. if (header_owner != BTRFS_TREE_LOG_OBJECTID &&
  2080. header_owner != BTRFS_TREE_RELOC_OBJECTID &&
  2081. header_transid == trans->transid &&
  2082. !btrfs_header_flag(buf, BTRFS_HEADER_FLAG_WRITTEN)) {
  2083. clean_tree_block(NULL, root, buf);
  2084. btrfs_tree_unlock(buf);
  2085. free_extent_buffer(buf);
  2086. return 1;
  2087. }
  2088. btrfs_tree_unlock(buf);
  2089. }
  2090. free_extent_buffer(buf);
  2091. pinit:
  2092. btrfs_update_pinned_extents(root, bytenr, num_bytes, 1);
  2093. BUG_ON(err < 0);
  2094. return 0;
  2095. }
  2096. /*
  2097. * remove an extent from the root, returns 0 on success
  2098. */
  2099. static int __free_extent(struct btrfs_trans_handle *trans,
  2100. struct btrfs_root *root,
  2101. u64 bytenr, u64 num_bytes, u64 parent,
  2102. u64 root_objectid, u64 ref_generation,
  2103. u64 owner_objectid, int pin, int mark_free)
  2104. {
  2105. struct btrfs_path *path;
  2106. struct btrfs_key key;
  2107. struct btrfs_fs_info *info = root->fs_info;
  2108. struct btrfs_root *extent_root = info->extent_root;
  2109. struct extent_buffer *leaf;
  2110. int ret;
  2111. int extent_slot = 0;
  2112. int found_extent = 0;
  2113. int num_to_del = 1;
  2114. struct btrfs_extent_item *ei;
  2115. u32 refs;
  2116. key.objectid = bytenr;
  2117. btrfs_set_key_type(&key, BTRFS_EXTENT_ITEM_KEY);
  2118. key.offset = num_bytes;
  2119. path = btrfs_alloc_path();
  2120. if (!path)
  2121. return -ENOMEM;
  2122. path->reada = 1;
  2123. ret = lookup_extent_backref(trans, extent_root, path,
  2124. bytenr, parent, root_objectid,
  2125. ref_generation, owner_objectid, 1);
  2126. if (ret == 0) {
  2127. struct btrfs_key found_key;
  2128. extent_slot = path->slots[0];
  2129. while(extent_slot > 0) {
  2130. extent_slot--;
  2131. btrfs_item_key_to_cpu(path->nodes[0], &found_key,
  2132. extent_slot);
  2133. if (found_key.objectid != bytenr)
  2134. break;
  2135. if (found_key.type == BTRFS_EXTENT_ITEM_KEY &&
  2136. found_key.offset == num_bytes) {
  2137. found_extent = 1;
  2138. break;
  2139. }
  2140. if (path->slots[0] - extent_slot > 5)
  2141. break;
  2142. }
  2143. if (!found_extent) {
  2144. ret = remove_extent_backref(trans, extent_root, path);
  2145. BUG_ON(ret);
  2146. btrfs_release_path(extent_root, path);
  2147. ret = btrfs_search_slot(trans, extent_root,
  2148. &key, path, -1, 1);
  2149. if (ret) {
  2150. printk(KERN_ERR "umm, got %d back from search"
  2151. ", was looking for %Lu\n", ret,
  2152. bytenr);
  2153. btrfs_print_leaf(extent_root, path->nodes[0]);
  2154. }
  2155. BUG_ON(ret);
  2156. extent_slot = path->slots[0];
  2157. }
  2158. } else {
  2159. btrfs_print_leaf(extent_root, path->nodes[0]);
  2160. WARN_ON(1);
  2161. printk("Unable to find ref byte nr %Lu root %Lu "
  2162. "gen %Lu owner %Lu\n", bytenr,
  2163. root_objectid, ref_generation, owner_objectid);
  2164. }
  2165. leaf = path->nodes[0];
  2166. ei = btrfs_item_ptr(leaf, extent_slot,
  2167. struct btrfs_extent_item);
  2168. refs = btrfs_extent_refs(leaf, ei);
  2169. BUG_ON(refs == 0);
  2170. refs -= 1;
  2171. btrfs_set_extent_refs(leaf, ei, refs);
  2172. btrfs_mark_buffer_dirty(leaf);
  2173. if (refs == 0 && found_extent && path->slots[0] == extent_slot + 1) {
  2174. struct btrfs_extent_ref *ref;
  2175. ref = btrfs_item_ptr(leaf, path->slots[0],
  2176. struct btrfs_extent_ref);
  2177. BUG_ON(btrfs_ref_num_refs(leaf, ref) != 1);
  2178. /* if the back ref and the extent are next to each other
  2179. * they get deleted below in one shot
  2180. */
  2181. path->slots[0] = extent_slot;
  2182. num_to_del = 2;
  2183. } else if (found_extent) {
  2184. /* otherwise delete the extent back ref */
  2185. ret = remove_extent_backref(trans, extent_root, path);
  2186. BUG_ON(ret);
  2187. /* if refs are 0, we need to setup the path for deletion */
  2188. if (refs == 0) {
  2189. btrfs_release_path(extent_root, path);
  2190. ret = btrfs_search_slot(trans, extent_root, &key, path,
  2191. -1, 1);
  2192. BUG_ON(ret);
  2193. }
  2194. }
  2195. if (refs == 0) {
  2196. u64 super_used;
  2197. u64 root_used;
  2198. #ifdef BIO_RW_DISCARD
  2199. u64 map_length = num_bytes;
  2200. struct btrfs_multi_bio *multi = NULL;
  2201. #endif
  2202. if (pin) {
  2203. mutex_lock(&root->fs_info->pinned_mutex);
  2204. ret = pin_down_bytes(trans, root, bytenr, num_bytes,
  2205. owner_objectid >= BTRFS_FIRST_FREE_OBJECTID);
  2206. mutex_unlock(&root->fs_info->pinned_mutex);
  2207. if (ret > 0)
  2208. mark_free = 1;
  2209. BUG_ON(ret < 0);
  2210. }
  2211. /* block accounting for super block */
  2212. spin_lock_irq(&info->delalloc_lock);
  2213. super_used = btrfs_super_bytes_used(&info->super_copy);
  2214. btrfs_set_super_bytes_used(&info->super_copy,
  2215. super_used - num_bytes);
  2216. spin_unlock_irq(&info->delalloc_lock);
  2217. /* block accounting for root item */
  2218. root_used = btrfs_root_used(&root->root_item);
  2219. btrfs_set_root_used(&root->root_item,
  2220. root_used - num_bytes);
  2221. ret = btrfs_del_items(trans, extent_root, path, path->slots[0],
  2222. num_to_del);
  2223. BUG_ON(ret);
  2224. btrfs_release_path(extent_root, path);
  2225. ret = update_block_group(trans, root, bytenr, num_bytes, 0,
  2226. mark_free);
  2227. BUG_ON(ret);
  2228. if (owner_objectid >= BTRFS_FIRST_FREE_OBJECTID) {
  2229. ret = btrfs_del_csums(trans, root, bytenr, num_bytes);
  2230. BUG_ON(ret);
  2231. }
  2232. #ifdef BIO_RW_DISCARD
  2233. /* Tell the block device(s) that the sectors can be discarded */
  2234. ret = btrfs_map_block(&root->fs_info->mapping_tree, READ,
  2235. bytenr, &map_length, &multi, 0);
  2236. if (!ret) {
  2237. struct btrfs_bio_stripe *stripe = multi->stripes;
  2238. int i;
  2239. if (map_length > num_bytes)
  2240. map_length = num_bytes;
  2241. for (i = 0; i < multi->num_stripes; i++, stripe++) {
  2242. btrfs_issue_discard(stripe->dev->bdev,
  2243. stripe->physical,
  2244. map_length);
  2245. }
  2246. kfree(multi);
  2247. }
  2248. #endif
  2249. }
  2250. btrfs_free_path(path);
  2251. finish_current_insert(trans, extent_root, 0);
  2252. return ret;
  2253. }
  2254. /*
  2255. * find all the blocks marked as pending in the radix tree and remove
  2256. * them from the extent map
  2257. */
  2258. static int del_pending_extents(struct btrfs_trans_handle *trans, struct
  2259. btrfs_root *extent_root, int all)
  2260. {
  2261. int ret;
  2262. int err = 0;
  2263. u64 start;
  2264. u64 end;
  2265. u64 priv;
  2266. u64 search = 0;
  2267. int nr = 0, skipped = 0;
  2268. struct extent_io_tree *pending_del;
  2269. struct extent_io_tree *extent_ins;
  2270. struct pending_extent_op *extent_op;
  2271. struct btrfs_fs_info *info = extent_root->fs_info;
  2272. struct list_head delete_list;
  2273. INIT_LIST_HEAD(&delete_list);
  2274. extent_ins = &extent_root->fs_info->extent_ins;
  2275. pending_del = &extent_root->fs_info->pending_del;
  2276. again:
  2277. mutex_lock(&info->extent_ins_mutex);
  2278. while(1) {
  2279. ret = find_first_extent_bit(pending_del, search, &start, &end,
  2280. EXTENT_WRITEBACK);
  2281. if (ret) {
  2282. if (all && skipped && !nr) {
  2283. search = 0;
  2284. continue;
  2285. }
  2286. mutex_unlock(&info->extent_ins_mutex);
  2287. break;
  2288. }
  2289. ret = try_lock_extent(extent_ins, start, end, GFP_NOFS);
  2290. if (!ret) {
  2291. search = end+1;
  2292. skipped = 1;
  2293. if (need_resched()) {
  2294. mutex_unlock(&info->extent_ins_mutex);
  2295. cond_resched();
  2296. mutex_lock(&info->extent_ins_mutex);
  2297. }
  2298. continue;
  2299. }
  2300. BUG_ON(ret < 0);
  2301. ret = get_state_private(pending_del, start, &priv);
  2302. BUG_ON(ret);
  2303. extent_op = (struct pending_extent_op *)(unsigned long)priv;
  2304. clear_extent_bits(pending_del, start, end, EXTENT_WRITEBACK,
  2305. GFP_NOFS);
  2306. if (!test_range_bit(extent_ins, start, end,
  2307. EXTENT_WRITEBACK, 0)) {
  2308. list_add_tail(&extent_op->list, &delete_list);
  2309. nr++;
  2310. } else {
  2311. kfree(extent_op);
  2312. ret = get_state_private(&info->extent_ins, start,
  2313. &priv);
  2314. BUG_ON(ret);
  2315. extent_op = (struct pending_extent_op *)
  2316. (unsigned long)priv;
  2317. clear_extent_bits(&info->extent_ins, start, end,
  2318. EXTENT_WRITEBACK, GFP_NOFS);
  2319. if (extent_op->type == PENDING_BACKREF_UPDATE) {
  2320. list_add_tail(&extent_op->list, &delete_list);
  2321. search = end + 1;
  2322. nr++;
  2323. continue;
  2324. }
  2325. mutex_lock(&extent_root->fs_info->pinned_mutex);
  2326. ret = pin_down_bytes(trans, extent_root, start,
  2327. end + 1 - start, 0);
  2328. mutex_unlock(&extent_root->fs_info->pinned_mutex);
  2329. ret = update_block_group(trans, extent_root, start,
  2330. end + 1 - start, 0, ret > 0);
  2331. unlock_extent(extent_ins, start, end, GFP_NOFS);
  2332. BUG_ON(ret);
  2333. kfree(extent_op);
  2334. }
  2335. if (ret)
  2336. err = ret;
  2337. search = end + 1;
  2338. if (need_resched()) {
  2339. mutex_unlock(&info->extent_ins_mutex);
  2340. cond_resched();
  2341. mutex_lock(&info->extent_ins_mutex);
  2342. }
  2343. }
  2344. if (nr) {
  2345. ret = free_extents(trans, extent_root, &delete_list);
  2346. BUG_ON(ret);
  2347. }
  2348. if (all && skipped) {
  2349. INIT_LIST_HEAD(&delete_list);
  2350. search = 0;
  2351. nr = 0;
  2352. goto again;
  2353. }
  2354. return err;
  2355. }
  2356. /*
  2357. * remove an extent from the root, returns 0 on success
  2358. */
  2359. static int __btrfs_free_extent(struct btrfs_trans_handle *trans,
  2360. struct btrfs_root *root,
  2361. u64 bytenr, u64 num_bytes, u64 parent,
  2362. u64 root_objectid, u64 ref_generation,
  2363. u64 owner_objectid, int pin)
  2364. {
  2365. struct btrfs_root *extent_root = root->fs_info->extent_root;
  2366. int pending_ret;
  2367. int ret;
  2368. WARN_ON(num_bytes < root->sectorsize);
  2369. if (root == extent_root) {
  2370. struct pending_extent_op *extent_op = NULL;
  2371. mutex_lock(&root->fs_info->extent_ins_mutex);
  2372. if (test_range_bit(&root->fs_info->extent_ins, bytenr,
  2373. bytenr + num_bytes - 1, EXTENT_WRITEBACK, 0)) {
  2374. u64 priv;
  2375. ret = get_state_private(&root->fs_info->extent_ins,
  2376. bytenr, &priv);
  2377. BUG_ON(ret);
  2378. extent_op = (struct pending_extent_op *)
  2379. (unsigned long)priv;
  2380. extent_op->del = 1;
  2381. if (extent_op->type == PENDING_EXTENT_INSERT) {
  2382. mutex_unlock(&root->fs_info->extent_ins_mutex);
  2383. return 0;
  2384. }
  2385. }
  2386. if (extent_op) {
  2387. ref_generation = extent_op->orig_generation;
  2388. parent = extent_op->orig_parent;
  2389. }
  2390. extent_op = kmalloc(sizeof(*extent_op), GFP_NOFS);
  2391. BUG_ON(!extent_op);
  2392. extent_op->type = PENDING_EXTENT_DELETE;
  2393. extent_op->bytenr = bytenr;
  2394. extent_op->num_bytes = num_bytes;
  2395. extent_op->parent = parent;
  2396. extent_op->orig_parent = parent;
  2397. extent_op->generation = ref_generation;
  2398. extent_op->orig_generation = ref_generation;
  2399. extent_op->level = (int)owner_objectid;
  2400. INIT_LIST_HEAD(&extent_op->list);
  2401. extent_op->del = 0;
  2402. set_extent_bits(&root->fs_info->pending_del,
  2403. bytenr, bytenr + num_bytes - 1,
  2404. EXTENT_WRITEBACK, GFP_NOFS);
  2405. set_state_private(&root->fs_info->pending_del,
  2406. bytenr, (unsigned long)extent_op);
  2407. mutex_unlock(&root->fs_info->extent_ins_mutex);
  2408. return 0;
  2409. }
  2410. /* if metadata always pin */
  2411. if (owner_objectid < BTRFS_FIRST_FREE_OBJECTID) {
  2412. if (root->root_key.objectid == BTRFS_TREE_LOG_OBJECTID) {
  2413. struct btrfs_block_group_cache *cache;
  2414. /* btrfs_free_reserved_extent */
  2415. cache = btrfs_lookup_block_group(root->fs_info, bytenr);
  2416. BUG_ON(!cache);
  2417. btrfs_add_free_space(cache, bytenr, num_bytes);
  2418. put_block_group(cache);
  2419. update_reserved_extents(root, bytenr, num_bytes, 0);
  2420. return 0;
  2421. }
  2422. pin = 1;
  2423. }
  2424. /* if data pin when any transaction has committed this */
  2425. if (ref_generation != trans->transid)
  2426. pin = 1;
  2427. ret = __free_extent(trans, root, bytenr, num_bytes, parent,
  2428. root_objectid, ref_generation,
  2429. owner_objectid, pin, pin == 0);
  2430. finish_current_insert(trans, root->fs_info->extent_root, 0);
  2431. pending_ret = del_pending_extents(trans, root->fs_info->extent_root, 0);
  2432. return ret ? ret : pending_ret;
  2433. }
  2434. int btrfs_free_extent(struct btrfs_trans_handle *trans,
  2435. struct btrfs_root *root,
  2436. u64 bytenr, u64 num_bytes, u64 parent,
  2437. u64 root_objectid, u64 ref_generation,
  2438. u64 owner_objectid, int pin)
  2439. {
  2440. int ret;
  2441. ret = __btrfs_free_extent(trans, root, bytenr, num_bytes, parent,
  2442. root_objectid, ref_generation,
  2443. owner_objectid, pin);
  2444. return ret;
  2445. }
  2446. static u64 stripe_align(struct btrfs_root *root, u64 val)
  2447. {
  2448. u64 mask = ((u64)root->stripesize - 1);
  2449. u64 ret = (val + mask) & ~mask;
  2450. return ret;
  2451. }
  2452. /*
  2453. * walks the btree of allocated extents and find a hole of a given size.
  2454. * The key ins is changed to record the hole:
  2455. * ins->objectid == block start
  2456. * ins->flags = BTRFS_EXTENT_ITEM_KEY
  2457. * ins->offset == number of blocks
  2458. * Any available blocks before search_start are skipped.
  2459. */
  2460. static int noinline find_free_extent(struct btrfs_trans_handle *trans,
  2461. struct btrfs_root *orig_root,
  2462. u64 num_bytes, u64 empty_size,
  2463. u64 search_start, u64 search_end,
  2464. u64 hint_byte, struct btrfs_key *ins,
  2465. u64 exclude_start, u64 exclude_nr,
  2466. int data)
  2467. {
  2468. int ret = 0;
  2469. struct btrfs_root * root = orig_root->fs_info->extent_root;
  2470. u64 total_needed = num_bytes;
  2471. u64 *last_ptr = NULL;
  2472. u64 last_wanted = 0;
  2473. struct btrfs_block_group_cache *block_group = NULL;
  2474. int chunk_alloc_done = 0;
  2475. int empty_cluster = 2 * 1024 * 1024;
  2476. int allowed_chunk_alloc = 0;
  2477. struct list_head *head = NULL, *cur = NULL;
  2478. int loop = 0;
  2479. int extra_loop = 0;
  2480. struct btrfs_space_info *space_info;
  2481. WARN_ON(num_bytes < root->sectorsize);
  2482. btrfs_set_key_type(ins, BTRFS_EXTENT_ITEM_KEY);
  2483. ins->objectid = 0;
  2484. ins->offset = 0;
  2485. if (orig_root->ref_cows || empty_size)
  2486. allowed_chunk_alloc = 1;
  2487. if (data & BTRFS_BLOCK_GROUP_METADATA) {
  2488. last_ptr = &root->fs_info->last_alloc;
  2489. empty_cluster = 64 * 1024;
  2490. }
  2491. if ((data & BTRFS_BLOCK_GROUP_DATA) && btrfs_test_opt(root, SSD))
  2492. last_ptr = &root->fs_info->last_data_alloc;
  2493. if (last_ptr) {
  2494. if (*last_ptr) {
  2495. hint_byte = *last_ptr;
  2496. last_wanted = *last_ptr;
  2497. } else
  2498. empty_size += empty_cluster;
  2499. } else {
  2500. empty_cluster = 0;
  2501. }
  2502. search_start = max(search_start, first_logical_byte(root, 0));
  2503. search_start = max(search_start, hint_byte);
  2504. if (last_wanted && search_start != last_wanted) {
  2505. last_wanted = 0;
  2506. empty_size += empty_cluster;
  2507. }
  2508. total_needed += empty_size;
  2509. block_group = btrfs_lookup_block_group(root->fs_info, search_start);
  2510. if (!block_group)
  2511. block_group = btrfs_lookup_first_block_group(root->fs_info,
  2512. search_start);
  2513. space_info = __find_space_info(root->fs_info, data);
  2514. down_read(&space_info->groups_sem);
  2515. while (1) {
  2516. struct btrfs_free_space *free_space;
  2517. /*
  2518. * the only way this happens if our hint points to a block
  2519. * group thats not of the proper type, while looping this
  2520. * should never happen
  2521. */
  2522. if (empty_size)
  2523. extra_loop = 1;
  2524. if (!block_group)
  2525. goto new_group_no_lock;
  2526. if (unlikely(!block_group->cached)) {
  2527. mutex_lock(&block_group->cache_mutex);
  2528. ret = cache_block_group(root, block_group);
  2529. mutex_unlock(&block_group->cache_mutex);
  2530. if (ret)
  2531. break;
  2532. }
  2533. mutex_lock(&block_group->alloc_mutex);
  2534. if (unlikely(!block_group_bits(block_group, data)))
  2535. goto new_group;
  2536. if (unlikely(block_group->ro))
  2537. goto new_group;
  2538. free_space = btrfs_find_free_space(block_group, search_start,
  2539. total_needed);
  2540. if (free_space) {
  2541. u64 start = block_group->key.objectid;
  2542. u64 end = block_group->key.objectid +
  2543. block_group->key.offset;
  2544. search_start = stripe_align(root, free_space->offset);
  2545. /* move on to the next group */
  2546. if (search_start + num_bytes >= search_end)
  2547. goto new_group;
  2548. /* move on to the next group */
  2549. if (search_start + num_bytes > end)
  2550. goto new_group;
  2551. if (last_wanted && search_start != last_wanted) {
  2552. total_needed += empty_cluster;
  2553. empty_size += empty_cluster;
  2554. last_wanted = 0;
  2555. /*
  2556. * if search_start is still in this block group
  2557. * then we just re-search this block group
  2558. */
  2559. if (search_start >= start &&
  2560. search_start < end) {
  2561. mutex_unlock(&block_group->alloc_mutex);
  2562. continue;
  2563. }
  2564. /* else we go to the next block group */
  2565. goto new_group;
  2566. }
  2567. if (exclude_nr > 0 &&
  2568. (search_start + num_bytes > exclude_start &&
  2569. search_start < exclude_start + exclude_nr)) {
  2570. search_start = exclude_start + exclude_nr;
  2571. /*
  2572. * if search_start is still in this block group
  2573. * then we just re-search this block group
  2574. */
  2575. if (search_start >= start &&
  2576. search_start < end) {
  2577. mutex_unlock(&block_group->alloc_mutex);
  2578. last_wanted = 0;
  2579. continue;
  2580. }
  2581. /* else we go to the next block group */
  2582. goto new_group;
  2583. }
  2584. ins->objectid = search_start;
  2585. ins->offset = num_bytes;
  2586. btrfs_remove_free_space_lock(block_group, search_start,
  2587. num_bytes);
  2588. /* we are all good, lets return */
  2589. mutex_unlock(&block_group->alloc_mutex);
  2590. break;
  2591. }
  2592. new_group:
  2593. mutex_unlock(&block_group->alloc_mutex);
  2594. put_block_group(block_group);
  2595. block_group = NULL;
  2596. new_group_no_lock:
  2597. /* don't try to compare new allocations against the
  2598. * last allocation any more
  2599. */
  2600. last_wanted = 0;
  2601. /*
  2602. * Here's how this works.
  2603. * loop == 0: we were searching a block group via a hint
  2604. * and didn't find anything, so we start at
  2605. * the head of the block groups and keep searching
  2606. * loop == 1: we're searching through all of the block groups
  2607. * if we hit the head again we have searched
  2608. * all of the block groups for this space and we
  2609. * need to try and allocate, if we cant error out.
  2610. * loop == 2: we allocated more space and are looping through
  2611. * all of the block groups again.
  2612. */
  2613. if (loop == 0) {
  2614. head = &space_info->block_groups;
  2615. cur = head->next;
  2616. loop++;
  2617. } else if (loop == 1 && cur == head) {
  2618. int keep_going;
  2619. /* at this point we give up on the empty_size
  2620. * allocations and just try to allocate the min
  2621. * space.
  2622. *
  2623. * The extra_loop field was set if an empty_size
  2624. * allocation was attempted above, and if this
  2625. * is try we need to try the loop again without
  2626. * the additional empty_size.
  2627. */
  2628. total_needed -= empty_size;
  2629. empty_size = 0;
  2630. keep_going = extra_loop;
  2631. loop++;
  2632. if (allowed_chunk_alloc && !chunk_alloc_done) {
  2633. up_read(&space_info->groups_sem);
  2634. ret = do_chunk_alloc(trans, root, num_bytes +
  2635. 2 * 1024 * 1024, data, 1);
  2636. down_read(&space_info->groups_sem);
  2637. if (ret < 0)
  2638. goto loop_check;
  2639. head = &space_info->block_groups;
  2640. /*
  2641. * we've allocated a new chunk, keep
  2642. * trying
  2643. */
  2644. keep_going = 1;
  2645. chunk_alloc_done = 1;
  2646. } else if (!allowed_chunk_alloc) {
  2647. space_info->force_alloc = 1;
  2648. }
  2649. loop_check:
  2650. if (keep_going) {
  2651. cur = head->next;
  2652. extra_loop = 0;
  2653. } else {
  2654. break;
  2655. }
  2656. } else if (cur == head) {
  2657. break;
  2658. }
  2659. block_group = list_entry(cur, struct btrfs_block_group_cache,
  2660. list);
  2661. atomic_inc(&block_group->count);
  2662. search_start = block_group->key.objectid;
  2663. cur = cur->next;
  2664. }
  2665. /* we found what we needed */
  2666. if (ins->objectid) {
  2667. if (!(data & BTRFS_BLOCK_GROUP_DATA))
  2668. trans->block_group = block_group->key.objectid;
  2669. if (last_ptr)
  2670. *last_ptr = ins->objectid + ins->offset;
  2671. ret = 0;
  2672. } else if (!ret) {
  2673. printk(KERN_ERR "we were searching for %Lu bytes, num_bytes %Lu,"
  2674. " loop %d, allowed_alloc %d\n", total_needed, num_bytes,
  2675. loop, allowed_chunk_alloc);
  2676. ret = -ENOSPC;
  2677. }
  2678. if (block_group)
  2679. put_block_group(block_group);
  2680. up_read(&space_info->groups_sem);
  2681. return ret;
  2682. }
  2683. static void dump_space_info(struct btrfs_space_info *info, u64 bytes)
  2684. {
  2685. struct btrfs_block_group_cache *cache;
  2686. struct list_head *l;
  2687. printk(KERN_INFO "space_info has %Lu free, is %sfull\n",
  2688. info->total_bytes - info->bytes_used - info->bytes_pinned -
  2689. info->bytes_reserved, (info->full) ? "" : "not ");
  2690. down_read(&info->groups_sem);
  2691. list_for_each(l, &info->block_groups) {
  2692. cache = list_entry(l, struct btrfs_block_group_cache, list);
  2693. spin_lock(&cache->lock);
  2694. printk(KERN_INFO "block group %Lu has %Lu bytes, %Lu used "
  2695. "%Lu pinned %Lu reserved\n",
  2696. cache->key.objectid, cache->key.offset,
  2697. btrfs_block_group_used(&cache->item),
  2698. cache->pinned, cache->reserved);
  2699. btrfs_dump_free_space(cache, bytes);
  2700. spin_unlock(&cache->lock);
  2701. }
  2702. up_read(&info->groups_sem);
  2703. }
  2704. static int __btrfs_reserve_extent(struct btrfs_trans_handle *trans,
  2705. struct btrfs_root *root,
  2706. u64 num_bytes, u64 min_alloc_size,
  2707. u64 empty_size, u64 hint_byte,
  2708. u64 search_end, struct btrfs_key *ins,
  2709. u64 data)
  2710. {
  2711. int ret;
  2712. u64 search_start = 0;
  2713. u64 alloc_profile;
  2714. struct btrfs_fs_info *info = root->fs_info;
  2715. if (data) {
  2716. alloc_profile = info->avail_data_alloc_bits &
  2717. info->data_alloc_profile;
  2718. data = BTRFS_BLOCK_GROUP_DATA | alloc_profile;
  2719. } else if (root == root->fs_info->chunk_root) {
  2720. alloc_profile = info->avail_system_alloc_bits &
  2721. info->system_alloc_profile;
  2722. data = BTRFS_BLOCK_GROUP_SYSTEM | alloc_profile;
  2723. } else {
  2724. alloc_profile = info->avail_metadata_alloc_bits &
  2725. info->metadata_alloc_profile;
  2726. data = BTRFS_BLOCK_GROUP_METADATA | alloc_profile;
  2727. }
  2728. again:
  2729. data = btrfs_reduce_alloc_profile(root, data);
  2730. /*
  2731. * the only place that sets empty_size is btrfs_realloc_node, which
  2732. * is not called recursively on allocations
  2733. */
  2734. if (empty_size || root->ref_cows) {
  2735. if (!(data & BTRFS_BLOCK_GROUP_METADATA)) {
  2736. ret = do_chunk_alloc(trans, root->fs_info->extent_root,
  2737. 2 * 1024 * 1024,
  2738. BTRFS_BLOCK_GROUP_METADATA |
  2739. (info->metadata_alloc_profile &
  2740. info->avail_metadata_alloc_bits), 0);
  2741. }
  2742. ret = do_chunk_alloc(trans, root->fs_info->extent_root,
  2743. num_bytes + 2 * 1024 * 1024, data, 0);
  2744. }
  2745. WARN_ON(num_bytes < root->sectorsize);
  2746. ret = find_free_extent(trans, root, num_bytes, empty_size,
  2747. search_start, search_end, hint_byte, ins,
  2748. trans->alloc_exclude_start,
  2749. trans->alloc_exclude_nr, data);
  2750. if (ret == -ENOSPC && num_bytes > min_alloc_size) {
  2751. num_bytes = num_bytes >> 1;
  2752. num_bytes = num_bytes & ~(root->sectorsize - 1);
  2753. num_bytes = max(num_bytes, min_alloc_size);
  2754. do_chunk_alloc(trans, root->fs_info->extent_root,
  2755. num_bytes, data, 1);
  2756. goto again;
  2757. }
  2758. if (ret) {
  2759. struct btrfs_space_info *sinfo;
  2760. sinfo = __find_space_info(root->fs_info, data);
  2761. printk("allocation failed flags %Lu, wanted %Lu\n",
  2762. data, num_bytes);
  2763. dump_space_info(sinfo, num_bytes);
  2764. BUG();
  2765. }
  2766. return ret;
  2767. }
  2768. int btrfs_free_reserved_extent(struct btrfs_root *root, u64 start, u64 len)
  2769. {
  2770. struct btrfs_block_group_cache *cache;
  2771. cache = btrfs_lookup_block_group(root->fs_info, start);
  2772. if (!cache) {
  2773. printk(KERN_ERR "Unable to find block group for %Lu\n", start);
  2774. return -ENOSPC;
  2775. }
  2776. btrfs_add_free_space(cache, start, len);
  2777. put_block_group(cache);
  2778. update_reserved_extents(root, start, len, 0);
  2779. return 0;
  2780. }
  2781. int btrfs_reserve_extent(struct btrfs_trans_handle *trans,
  2782. struct btrfs_root *root,
  2783. u64 num_bytes, u64 min_alloc_size,
  2784. u64 empty_size, u64 hint_byte,
  2785. u64 search_end, struct btrfs_key *ins,
  2786. u64 data)
  2787. {
  2788. int ret;
  2789. ret = __btrfs_reserve_extent(trans, root, num_bytes, min_alloc_size,
  2790. empty_size, hint_byte, search_end, ins,
  2791. data);
  2792. update_reserved_extents(root, ins->objectid, ins->offset, 1);
  2793. return ret;
  2794. }
  2795. static int __btrfs_alloc_reserved_extent(struct btrfs_trans_handle *trans,
  2796. struct btrfs_root *root, u64 parent,
  2797. u64 root_objectid, u64 ref_generation,
  2798. u64 owner, struct btrfs_key *ins)
  2799. {
  2800. int ret;
  2801. int pending_ret;
  2802. u64 super_used;
  2803. u64 root_used;
  2804. u64 num_bytes = ins->offset;
  2805. u32 sizes[2];
  2806. struct btrfs_fs_info *info = root->fs_info;
  2807. struct btrfs_root *extent_root = info->extent_root;
  2808. struct btrfs_extent_item *extent_item;
  2809. struct btrfs_extent_ref *ref;
  2810. struct btrfs_path *path;
  2811. struct btrfs_key keys[2];
  2812. if (parent == 0)
  2813. parent = ins->objectid;
  2814. /* block accounting for super block */
  2815. spin_lock_irq(&info->delalloc_lock);
  2816. super_used = btrfs_super_bytes_used(&info->super_copy);
  2817. btrfs_set_super_bytes_used(&info->super_copy, super_used + num_bytes);
  2818. spin_unlock_irq(&info->delalloc_lock);
  2819. /* block accounting for root item */
  2820. root_used = btrfs_root_used(&root->root_item);
  2821. btrfs_set_root_used(&root->root_item, root_used + num_bytes);
  2822. if (root == extent_root) {
  2823. struct pending_extent_op *extent_op;
  2824. extent_op = kmalloc(sizeof(*extent_op), GFP_NOFS);
  2825. BUG_ON(!extent_op);
  2826. extent_op->type = PENDING_EXTENT_INSERT;
  2827. extent_op->bytenr = ins->objectid;
  2828. extent_op->num_bytes = ins->offset;
  2829. extent_op->parent = parent;
  2830. extent_op->orig_parent = 0;
  2831. extent_op->generation = ref_generation;
  2832. extent_op->orig_generation = 0;
  2833. extent_op->level = (int)owner;
  2834. INIT_LIST_HEAD(&extent_op->list);
  2835. extent_op->del = 0;
  2836. mutex_lock(&root->fs_info->extent_ins_mutex);
  2837. set_extent_bits(&root->fs_info->extent_ins, ins->objectid,
  2838. ins->objectid + ins->offset - 1,
  2839. EXTENT_WRITEBACK, GFP_NOFS);
  2840. set_state_private(&root->fs_info->extent_ins,
  2841. ins->objectid, (unsigned long)extent_op);
  2842. mutex_unlock(&root->fs_info->extent_ins_mutex);
  2843. goto update_block;
  2844. }
  2845. memcpy(&keys[0], ins, sizeof(*ins));
  2846. keys[1].objectid = ins->objectid;
  2847. keys[1].type = BTRFS_EXTENT_REF_KEY;
  2848. keys[1].offset = parent;
  2849. sizes[0] = sizeof(*extent_item);
  2850. sizes[1] = sizeof(*ref);
  2851. path = btrfs_alloc_path();
  2852. BUG_ON(!path);
  2853. ret = btrfs_insert_empty_items(trans, extent_root, path, keys,
  2854. sizes, 2);
  2855. BUG_ON(ret);
  2856. extent_item = btrfs_item_ptr(path->nodes[0], path->slots[0],
  2857. struct btrfs_extent_item);
  2858. btrfs_set_extent_refs(path->nodes[0], extent_item, 1);
  2859. ref = btrfs_item_ptr(path->nodes[0], path->slots[0] + 1,
  2860. struct btrfs_extent_ref);
  2861. btrfs_set_ref_root(path->nodes[0], ref, root_objectid);
  2862. btrfs_set_ref_generation(path->nodes[0], ref, ref_generation);
  2863. btrfs_set_ref_objectid(path->nodes[0], ref, owner);
  2864. btrfs_set_ref_num_refs(path->nodes[0], ref, 1);
  2865. btrfs_mark_buffer_dirty(path->nodes[0]);
  2866. trans->alloc_exclude_start = 0;
  2867. trans->alloc_exclude_nr = 0;
  2868. btrfs_free_path(path);
  2869. finish_current_insert(trans, extent_root, 0);
  2870. pending_ret = del_pending_extents(trans, extent_root, 0);
  2871. if (ret)
  2872. goto out;
  2873. if (pending_ret) {
  2874. ret = pending_ret;
  2875. goto out;
  2876. }
  2877. update_block:
  2878. ret = update_block_group(trans, root, ins->objectid, ins->offset, 1, 0);
  2879. if (ret) {
  2880. printk("update block group failed for %Lu %Lu\n",
  2881. ins->objectid, ins->offset);
  2882. BUG();
  2883. }
  2884. out:
  2885. return ret;
  2886. }
  2887. int btrfs_alloc_reserved_extent(struct btrfs_trans_handle *trans,
  2888. struct btrfs_root *root, u64 parent,
  2889. u64 root_objectid, u64 ref_generation,
  2890. u64 owner, struct btrfs_key *ins)
  2891. {
  2892. int ret;
  2893. if (root_objectid == BTRFS_TREE_LOG_OBJECTID)
  2894. return 0;
  2895. ret = __btrfs_alloc_reserved_extent(trans, root, parent, root_objectid,
  2896. ref_generation, owner, ins);
  2897. update_reserved_extents(root, ins->objectid, ins->offset, 0);
  2898. return ret;
  2899. }
  2900. /*
  2901. * this is used by the tree logging recovery code. It records that
  2902. * an extent has been allocated and makes sure to clear the free
  2903. * space cache bits as well
  2904. */
  2905. int btrfs_alloc_logged_extent(struct btrfs_trans_handle *trans,
  2906. struct btrfs_root *root, u64 parent,
  2907. u64 root_objectid, u64 ref_generation,
  2908. u64 owner, struct btrfs_key *ins)
  2909. {
  2910. int ret;
  2911. struct btrfs_block_group_cache *block_group;
  2912. block_group = btrfs_lookup_block_group(root->fs_info, ins->objectid);
  2913. mutex_lock(&block_group->cache_mutex);
  2914. cache_block_group(root, block_group);
  2915. mutex_unlock(&block_group->cache_mutex);
  2916. ret = btrfs_remove_free_space(block_group, ins->objectid,
  2917. ins->offset);
  2918. BUG_ON(ret);
  2919. put_block_group(block_group);
  2920. ret = __btrfs_alloc_reserved_extent(trans, root, parent, root_objectid,
  2921. ref_generation, owner, ins);
  2922. return ret;
  2923. }
  2924. /*
  2925. * finds a free extent and does all the dirty work required for allocation
  2926. * returns the key for the extent through ins, and a tree buffer for
  2927. * the first block of the extent through buf.
  2928. *
  2929. * returns 0 if everything worked, non-zero otherwise.
  2930. */
  2931. int btrfs_alloc_extent(struct btrfs_trans_handle *trans,
  2932. struct btrfs_root *root,
  2933. u64 num_bytes, u64 parent, u64 min_alloc_size,
  2934. u64 root_objectid, u64 ref_generation,
  2935. u64 owner_objectid, u64 empty_size, u64 hint_byte,
  2936. u64 search_end, struct btrfs_key *ins, u64 data)
  2937. {
  2938. int ret;
  2939. ret = __btrfs_reserve_extent(trans, root, num_bytes,
  2940. min_alloc_size, empty_size, hint_byte,
  2941. search_end, ins, data);
  2942. BUG_ON(ret);
  2943. if (root_objectid != BTRFS_TREE_LOG_OBJECTID) {
  2944. ret = __btrfs_alloc_reserved_extent(trans, root, parent,
  2945. root_objectid, ref_generation,
  2946. owner_objectid, ins);
  2947. BUG_ON(ret);
  2948. } else {
  2949. update_reserved_extents(root, ins->objectid, ins->offset, 1);
  2950. }
  2951. return ret;
  2952. }
  2953. struct extent_buffer *btrfs_init_new_buffer(struct btrfs_trans_handle *trans,
  2954. struct btrfs_root *root,
  2955. u64 bytenr, u32 blocksize)
  2956. {
  2957. struct extent_buffer *buf;
  2958. buf = btrfs_find_create_tree_block(root, bytenr, blocksize);
  2959. if (!buf)
  2960. return ERR_PTR(-ENOMEM);
  2961. btrfs_set_header_generation(buf, trans->transid);
  2962. btrfs_tree_lock(buf);
  2963. clean_tree_block(trans, root, buf);
  2964. btrfs_set_buffer_uptodate(buf);
  2965. if (root->root_key.objectid == BTRFS_TREE_LOG_OBJECTID) {
  2966. set_extent_dirty(&root->dirty_log_pages, buf->start,
  2967. buf->start + buf->len - 1, GFP_NOFS);
  2968. } else {
  2969. set_extent_dirty(&trans->transaction->dirty_pages, buf->start,
  2970. buf->start + buf->len - 1, GFP_NOFS);
  2971. }
  2972. trans->blocks_used++;
  2973. return buf;
  2974. }
  2975. /*
  2976. * helper function to allocate a block for a given tree
  2977. * returns the tree buffer or NULL.
  2978. */
  2979. struct extent_buffer *btrfs_alloc_free_block(struct btrfs_trans_handle *trans,
  2980. struct btrfs_root *root,
  2981. u32 blocksize, u64 parent,
  2982. u64 root_objectid,
  2983. u64 ref_generation,
  2984. int level,
  2985. u64 hint,
  2986. u64 empty_size)
  2987. {
  2988. struct btrfs_key ins;
  2989. int ret;
  2990. struct extent_buffer *buf;
  2991. ret = btrfs_alloc_extent(trans, root, blocksize, parent, blocksize,
  2992. root_objectid, ref_generation, level,
  2993. empty_size, hint, (u64)-1, &ins, 0);
  2994. if (ret) {
  2995. BUG_ON(ret > 0);
  2996. return ERR_PTR(ret);
  2997. }
  2998. buf = btrfs_init_new_buffer(trans, root, ins.objectid, blocksize);
  2999. return buf;
  3000. }
  3001. int btrfs_drop_leaf_ref(struct btrfs_trans_handle *trans,
  3002. struct btrfs_root *root, struct extent_buffer *leaf)
  3003. {
  3004. u64 leaf_owner;
  3005. u64 leaf_generation;
  3006. struct btrfs_key key;
  3007. struct btrfs_file_extent_item *fi;
  3008. int i;
  3009. int nritems;
  3010. int ret;
  3011. BUG_ON(!btrfs_is_leaf(leaf));
  3012. nritems = btrfs_header_nritems(leaf);
  3013. leaf_owner = btrfs_header_owner(leaf);
  3014. leaf_generation = btrfs_header_generation(leaf);
  3015. for (i = 0; i < nritems; i++) {
  3016. u64 disk_bytenr;
  3017. cond_resched();
  3018. btrfs_item_key_to_cpu(leaf, &key, i);
  3019. if (btrfs_key_type(&key) != BTRFS_EXTENT_DATA_KEY)
  3020. continue;
  3021. fi = btrfs_item_ptr(leaf, i, struct btrfs_file_extent_item);
  3022. if (btrfs_file_extent_type(leaf, fi) ==
  3023. BTRFS_FILE_EXTENT_INLINE)
  3024. continue;
  3025. /*
  3026. * FIXME make sure to insert a trans record that
  3027. * repeats the snapshot del on crash
  3028. */
  3029. disk_bytenr = btrfs_file_extent_disk_bytenr(leaf, fi);
  3030. if (disk_bytenr == 0)
  3031. continue;
  3032. ret = __btrfs_free_extent(trans, root, disk_bytenr,
  3033. btrfs_file_extent_disk_num_bytes(leaf, fi),
  3034. leaf->start, leaf_owner, leaf_generation,
  3035. key.objectid, 0);
  3036. BUG_ON(ret);
  3037. atomic_inc(&root->fs_info->throttle_gen);
  3038. wake_up(&root->fs_info->transaction_throttle);
  3039. cond_resched();
  3040. }
  3041. return 0;
  3042. }
  3043. static int noinline cache_drop_leaf_ref(struct btrfs_trans_handle *trans,
  3044. struct btrfs_root *root,
  3045. struct btrfs_leaf_ref *ref)
  3046. {
  3047. int i;
  3048. int ret;
  3049. struct btrfs_extent_info *info = ref->extents;
  3050. for (i = 0; i < ref->nritems; i++) {
  3051. ret = __btrfs_free_extent(trans, root, info->bytenr,
  3052. info->num_bytes, ref->bytenr,
  3053. ref->owner, ref->generation,
  3054. info->objectid, 0);
  3055. atomic_inc(&root->fs_info->throttle_gen);
  3056. wake_up(&root->fs_info->transaction_throttle);
  3057. cond_resched();
  3058. BUG_ON(ret);
  3059. info++;
  3060. }
  3061. return 0;
  3062. }
  3063. static int drop_snap_lookup_refcount(struct btrfs_root *root, u64 start, u64 len,
  3064. u32 *refs)
  3065. {
  3066. int ret;
  3067. ret = btrfs_lookup_extent_ref(NULL, root, start, len, refs);
  3068. BUG_ON(ret);
  3069. #if 0 // some debugging code in case we see problems here
  3070. /* if the refs count is one, it won't get increased again. But
  3071. * if the ref count is > 1, someone may be decreasing it at
  3072. * the same time we are.
  3073. */
  3074. if (*refs != 1) {
  3075. struct extent_buffer *eb = NULL;
  3076. eb = btrfs_find_create_tree_block(root, start, len);
  3077. if (eb)
  3078. btrfs_tree_lock(eb);
  3079. mutex_lock(&root->fs_info->alloc_mutex);
  3080. ret = lookup_extent_ref(NULL, root, start, len, refs);
  3081. BUG_ON(ret);
  3082. mutex_unlock(&root->fs_info->alloc_mutex);
  3083. if (eb) {
  3084. btrfs_tree_unlock(eb);
  3085. free_extent_buffer(eb);
  3086. }
  3087. if (*refs == 1) {
  3088. printk("block %llu went down to one during drop_snap\n",
  3089. (unsigned long long)start);
  3090. }
  3091. }
  3092. #endif
  3093. cond_resched();
  3094. return ret;
  3095. }
  3096. /*
  3097. * helper function for drop_snapshot, this walks down the tree dropping ref
  3098. * counts as it goes.
  3099. */
  3100. static int noinline walk_down_tree(struct btrfs_trans_handle *trans,
  3101. struct btrfs_root *root,
  3102. struct btrfs_path *path, int *level)
  3103. {
  3104. u64 root_owner;
  3105. u64 root_gen;
  3106. u64 bytenr;
  3107. u64 ptr_gen;
  3108. struct extent_buffer *next;
  3109. struct extent_buffer *cur;
  3110. struct extent_buffer *parent;
  3111. struct btrfs_leaf_ref *ref;
  3112. u32 blocksize;
  3113. int ret;
  3114. u32 refs;
  3115. WARN_ON(*level < 0);
  3116. WARN_ON(*level >= BTRFS_MAX_LEVEL);
  3117. ret = drop_snap_lookup_refcount(root, path->nodes[*level]->start,
  3118. path->nodes[*level]->len, &refs);
  3119. BUG_ON(ret);
  3120. if (refs > 1)
  3121. goto out;
  3122. /*
  3123. * walk down to the last node level and free all the leaves
  3124. */
  3125. while(*level >= 0) {
  3126. WARN_ON(*level < 0);
  3127. WARN_ON(*level >= BTRFS_MAX_LEVEL);
  3128. cur = path->nodes[*level];
  3129. if (btrfs_header_level(cur) != *level)
  3130. WARN_ON(1);
  3131. if (path->slots[*level] >=
  3132. btrfs_header_nritems(cur))
  3133. break;
  3134. if (*level == 0) {
  3135. ret = btrfs_drop_leaf_ref(trans, root, cur);
  3136. BUG_ON(ret);
  3137. break;
  3138. }
  3139. bytenr = btrfs_node_blockptr(cur, path->slots[*level]);
  3140. ptr_gen = btrfs_node_ptr_generation(cur, path->slots[*level]);
  3141. blocksize = btrfs_level_size(root, *level - 1);
  3142. ret = drop_snap_lookup_refcount(root, bytenr, blocksize, &refs);
  3143. BUG_ON(ret);
  3144. if (refs != 1) {
  3145. parent = path->nodes[*level];
  3146. root_owner = btrfs_header_owner(parent);
  3147. root_gen = btrfs_header_generation(parent);
  3148. path->slots[*level]++;
  3149. ret = __btrfs_free_extent(trans, root, bytenr,
  3150. blocksize, parent->start,
  3151. root_owner, root_gen,
  3152. *level - 1, 1);
  3153. BUG_ON(ret);
  3154. atomic_inc(&root->fs_info->throttle_gen);
  3155. wake_up(&root->fs_info->transaction_throttle);
  3156. cond_resched();
  3157. continue;
  3158. }
  3159. /*
  3160. * at this point, we have a single ref, and since the
  3161. * only place referencing this extent is a dead root
  3162. * the reference count should never go higher.
  3163. * So, we don't need to check it again
  3164. */
  3165. if (*level == 1) {
  3166. ref = btrfs_lookup_leaf_ref(root, bytenr);
  3167. if (ref && ref->generation != ptr_gen) {
  3168. btrfs_free_leaf_ref(root, ref);
  3169. ref = NULL;
  3170. }
  3171. if (ref) {
  3172. ret = cache_drop_leaf_ref(trans, root, ref);
  3173. BUG_ON(ret);
  3174. btrfs_remove_leaf_ref(root, ref);
  3175. btrfs_free_leaf_ref(root, ref);
  3176. *level = 0;
  3177. break;
  3178. }
  3179. if (printk_ratelimit()) {
  3180. printk("leaf ref miss for bytenr %llu\n",
  3181. (unsigned long long)bytenr);
  3182. }
  3183. }
  3184. next = btrfs_find_tree_block(root, bytenr, blocksize);
  3185. if (!next || !btrfs_buffer_uptodate(next, ptr_gen)) {
  3186. free_extent_buffer(next);
  3187. next = read_tree_block(root, bytenr, blocksize,
  3188. ptr_gen);
  3189. cond_resched();
  3190. #if 0
  3191. /*
  3192. * this is a debugging check and can go away
  3193. * the ref should never go all the way down to 1
  3194. * at this point
  3195. */
  3196. ret = lookup_extent_ref(NULL, root, bytenr, blocksize,
  3197. &refs);
  3198. BUG_ON(ret);
  3199. WARN_ON(refs != 1);
  3200. #endif
  3201. }
  3202. WARN_ON(*level <= 0);
  3203. if (path->nodes[*level-1])
  3204. free_extent_buffer(path->nodes[*level-1]);
  3205. path->nodes[*level-1] = next;
  3206. *level = btrfs_header_level(next);
  3207. path->slots[*level] = 0;
  3208. cond_resched();
  3209. }
  3210. out:
  3211. WARN_ON(*level < 0);
  3212. WARN_ON(*level >= BTRFS_MAX_LEVEL);
  3213. if (path->nodes[*level] == root->node) {
  3214. parent = path->nodes[*level];
  3215. bytenr = path->nodes[*level]->start;
  3216. } else {
  3217. parent = path->nodes[*level + 1];
  3218. bytenr = btrfs_node_blockptr(parent, path->slots[*level + 1]);
  3219. }
  3220. blocksize = btrfs_level_size(root, *level);
  3221. root_owner = btrfs_header_owner(parent);
  3222. root_gen = btrfs_header_generation(parent);
  3223. ret = __btrfs_free_extent(trans, root, bytenr, blocksize,
  3224. parent->start, root_owner, root_gen,
  3225. *level, 1);
  3226. free_extent_buffer(path->nodes[*level]);
  3227. path->nodes[*level] = NULL;
  3228. *level += 1;
  3229. BUG_ON(ret);
  3230. cond_resched();
  3231. return 0;
  3232. }
  3233. /*
  3234. * helper function for drop_subtree, this function is similar to
  3235. * walk_down_tree. The main difference is that it checks reference
  3236. * counts while tree blocks are locked.
  3237. */
  3238. static int noinline walk_down_subtree(struct btrfs_trans_handle *trans,
  3239. struct btrfs_root *root,
  3240. struct btrfs_path *path, int *level)
  3241. {
  3242. struct extent_buffer *next;
  3243. struct extent_buffer *cur;
  3244. struct extent_buffer *parent;
  3245. u64 bytenr;
  3246. u64 ptr_gen;
  3247. u32 blocksize;
  3248. u32 refs;
  3249. int ret;
  3250. cur = path->nodes[*level];
  3251. ret = btrfs_lookup_extent_ref(trans, root, cur->start, cur->len,
  3252. &refs);
  3253. BUG_ON(ret);
  3254. if (refs > 1)
  3255. goto out;
  3256. while (*level >= 0) {
  3257. cur = path->nodes[*level];
  3258. if (*level == 0) {
  3259. ret = btrfs_drop_leaf_ref(trans, root, cur);
  3260. BUG_ON(ret);
  3261. clean_tree_block(trans, root, cur);
  3262. break;
  3263. }
  3264. if (path->slots[*level] >= btrfs_header_nritems(cur)) {
  3265. clean_tree_block(trans, root, cur);
  3266. break;
  3267. }
  3268. bytenr = btrfs_node_blockptr(cur, path->slots[*level]);
  3269. blocksize = btrfs_level_size(root, *level - 1);
  3270. ptr_gen = btrfs_node_ptr_generation(cur, path->slots[*level]);
  3271. next = read_tree_block(root, bytenr, blocksize, ptr_gen);
  3272. btrfs_tree_lock(next);
  3273. ret = btrfs_lookup_extent_ref(trans, root, bytenr, blocksize,
  3274. &refs);
  3275. BUG_ON(ret);
  3276. if (refs > 1) {
  3277. parent = path->nodes[*level];
  3278. ret = btrfs_free_extent(trans, root, bytenr,
  3279. blocksize, parent->start,
  3280. btrfs_header_owner(parent),
  3281. btrfs_header_generation(parent),
  3282. *level - 1, 1);
  3283. BUG_ON(ret);
  3284. path->slots[*level]++;
  3285. btrfs_tree_unlock(next);
  3286. free_extent_buffer(next);
  3287. continue;
  3288. }
  3289. *level = btrfs_header_level(next);
  3290. path->nodes[*level] = next;
  3291. path->slots[*level] = 0;
  3292. path->locks[*level] = 1;
  3293. cond_resched();
  3294. }
  3295. out:
  3296. parent = path->nodes[*level + 1];
  3297. bytenr = path->nodes[*level]->start;
  3298. blocksize = path->nodes[*level]->len;
  3299. ret = btrfs_free_extent(trans, root, bytenr, blocksize,
  3300. parent->start, btrfs_header_owner(parent),
  3301. btrfs_header_generation(parent), *level, 1);
  3302. BUG_ON(ret);
  3303. if (path->locks[*level]) {
  3304. btrfs_tree_unlock(path->nodes[*level]);
  3305. path->locks[*level] = 0;
  3306. }
  3307. free_extent_buffer(path->nodes[*level]);
  3308. path->nodes[*level] = NULL;
  3309. *level += 1;
  3310. cond_resched();
  3311. return 0;
  3312. }
  3313. /*
  3314. * helper for dropping snapshots. This walks back up the tree in the path
  3315. * to find the first node higher up where we haven't yet gone through
  3316. * all the slots
  3317. */
  3318. static int noinline walk_up_tree(struct btrfs_trans_handle *trans,
  3319. struct btrfs_root *root,
  3320. struct btrfs_path *path,
  3321. int *level, int max_level)
  3322. {
  3323. u64 root_owner;
  3324. u64 root_gen;
  3325. struct btrfs_root_item *root_item = &root->root_item;
  3326. int i;
  3327. int slot;
  3328. int ret;
  3329. for (i = *level; i < max_level && path->nodes[i]; i++) {
  3330. slot = path->slots[i];
  3331. if (slot < btrfs_header_nritems(path->nodes[i]) - 1) {
  3332. struct extent_buffer *node;
  3333. struct btrfs_disk_key disk_key;
  3334. node = path->nodes[i];
  3335. path->slots[i]++;
  3336. *level = i;
  3337. WARN_ON(*level == 0);
  3338. btrfs_node_key(node, &disk_key, path->slots[i]);
  3339. memcpy(&root_item->drop_progress,
  3340. &disk_key, sizeof(disk_key));
  3341. root_item->drop_level = i;
  3342. return 0;
  3343. } else {
  3344. struct extent_buffer *parent;
  3345. if (path->nodes[*level] == root->node)
  3346. parent = path->nodes[*level];
  3347. else
  3348. parent = path->nodes[*level + 1];
  3349. root_owner = btrfs_header_owner(parent);
  3350. root_gen = btrfs_header_generation(parent);
  3351. clean_tree_block(trans, root, path->nodes[*level]);
  3352. ret = btrfs_free_extent(trans, root,
  3353. path->nodes[*level]->start,
  3354. path->nodes[*level]->len,
  3355. parent->start, root_owner,
  3356. root_gen, *level, 1);
  3357. BUG_ON(ret);
  3358. if (path->locks[*level]) {
  3359. btrfs_tree_unlock(path->nodes[*level]);
  3360. path->locks[*level] = 0;
  3361. }
  3362. free_extent_buffer(path->nodes[*level]);
  3363. path->nodes[*level] = NULL;
  3364. *level = i + 1;
  3365. }
  3366. }
  3367. return 1;
  3368. }
  3369. /*
  3370. * drop the reference count on the tree rooted at 'snap'. This traverses
  3371. * the tree freeing any blocks that have a ref count of zero after being
  3372. * decremented.
  3373. */
  3374. int btrfs_drop_snapshot(struct btrfs_trans_handle *trans, struct btrfs_root
  3375. *root)
  3376. {
  3377. int ret = 0;
  3378. int wret;
  3379. int level;
  3380. struct btrfs_path *path;
  3381. int i;
  3382. int orig_level;
  3383. struct btrfs_root_item *root_item = &root->root_item;
  3384. WARN_ON(!mutex_is_locked(&root->fs_info->drop_mutex));
  3385. path = btrfs_alloc_path();
  3386. BUG_ON(!path);
  3387. level = btrfs_header_level(root->node);
  3388. orig_level = level;
  3389. if (btrfs_disk_key_objectid(&root_item->drop_progress) == 0) {
  3390. path->nodes[level] = root->node;
  3391. extent_buffer_get(root->node);
  3392. path->slots[level] = 0;
  3393. } else {
  3394. struct btrfs_key key;
  3395. struct btrfs_disk_key found_key;
  3396. struct extent_buffer *node;
  3397. btrfs_disk_key_to_cpu(&key, &root_item->drop_progress);
  3398. level = root_item->drop_level;
  3399. path->lowest_level = level;
  3400. wret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
  3401. if (wret < 0) {
  3402. ret = wret;
  3403. goto out;
  3404. }
  3405. node = path->nodes[level];
  3406. btrfs_node_key(node, &found_key, path->slots[level]);
  3407. WARN_ON(memcmp(&found_key, &root_item->drop_progress,
  3408. sizeof(found_key)));
  3409. /*
  3410. * unlock our path, this is safe because only this
  3411. * function is allowed to delete this snapshot
  3412. */
  3413. for (i = 0; i < BTRFS_MAX_LEVEL; i++) {
  3414. if (path->nodes[i] && path->locks[i]) {
  3415. path->locks[i] = 0;
  3416. btrfs_tree_unlock(path->nodes[i]);
  3417. }
  3418. }
  3419. }
  3420. while(1) {
  3421. wret = walk_down_tree(trans, root, path, &level);
  3422. if (wret > 0)
  3423. break;
  3424. if (wret < 0)
  3425. ret = wret;
  3426. wret = walk_up_tree(trans, root, path, &level,
  3427. BTRFS_MAX_LEVEL);
  3428. if (wret > 0)
  3429. break;
  3430. if (wret < 0)
  3431. ret = wret;
  3432. if (trans->transaction->in_commit) {
  3433. ret = -EAGAIN;
  3434. break;
  3435. }
  3436. atomic_inc(&root->fs_info->throttle_gen);
  3437. wake_up(&root->fs_info->transaction_throttle);
  3438. }
  3439. for (i = 0; i <= orig_level; i++) {
  3440. if (path->nodes[i]) {
  3441. free_extent_buffer(path->nodes[i]);
  3442. path->nodes[i] = NULL;
  3443. }
  3444. }
  3445. out:
  3446. btrfs_free_path(path);
  3447. return ret;
  3448. }
  3449. int btrfs_drop_subtree(struct btrfs_trans_handle *trans,
  3450. struct btrfs_root *root,
  3451. struct extent_buffer *node,
  3452. struct extent_buffer *parent)
  3453. {
  3454. struct btrfs_path *path;
  3455. int level;
  3456. int parent_level;
  3457. int ret = 0;
  3458. int wret;
  3459. path = btrfs_alloc_path();
  3460. BUG_ON(!path);
  3461. BUG_ON(!btrfs_tree_locked(parent));
  3462. parent_level = btrfs_header_level(parent);
  3463. extent_buffer_get(parent);
  3464. path->nodes[parent_level] = parent;
  3465. path->slots[parent_level] = btrfs_header_nritems(parent);
  3466. BUG_ON(!btrfs_tree_locked(node));
  3467. level = btrfs_header_level(node);
  3468. extent_buffer_get(node);
  3469. path->nodes[level] = node;
  3470. path->slots[level] = 0;
  3471. while (1) {
  3472. wret = walk_down_subtree(trans, root, path, &level);
  3473. if (wret < 0)
  3474. ret = wret;
  3475. if (wret != 0)
  3476. break;
  3477. wret = walk_up_tree(trans, root, path, &level, parent_level);
  3478. if (wret < 0)
  3479. ret = wret;
  3480. if (wret != 0)
  3481. break;
  3482. }
  3483. btrfs_free_path(path);
  3484. return ret;
  3485. }
  3486. static unsigned long calc_ra(unsigned long start, unsigned long last,
  3487. unsigned long nr)
  3488. {
  3489. return min(last, start + nr - 1);
  3490. }
  3491. static int noinline relocate_inode_pages(struct inode *inode, u64 start,
  3492. u64 len)
  3493. {
  3494. u64 page_start;
  3495. u64 page_end;
  3496. unsigned long first_index;
  3497. unsigned long last_index;
  3498. unsigned long i;
  3499. struct page *page;
  3500. struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
  3501. struct file_ra_state *ra;
  3502. struct btrfs_ordered_extent *ordered;
  3503. unsigned int total_read = 0;
  3504. unsigned int total_dirty = 0;
  3505. int ret = 0;
  3506. ra = kzalloc(sizeof(*ra), GFP_NOFS);
  3507. mutex_lock(&inode->i_mutex);
  3508. first_index = start >> PAGE_CACHE_SHIFT;
  3509. last_index = (start + len - 1) >> PAGE_CACHE_SHIFT;
  3510. /* make sure the dirty trick played by the caller work */
  3511. ret = invalidate_inode_pages2_range(inode->i_mapping,
  3512. first_index, last_index);
  3513. if (ret)
  3514. goto out_unlock;
  3515. file_ra_state_init(ra, inode->i_mapping);
  3516. for (i = first_index ; i <= last_index; i++) {
  3517. if (total_read % ra->ra_pages == 0) {
  3518. btrfs_force_ra(inode->i_mapping, ra, NULL, i,
  3519. calc_ra(i, last_index, ra->ra_pages));
  3520. }
  3521. total_read++;
  3522. again:
  3523. if (((u64)i << PAGE_CACHE_SHIFT) > i_size_read(inode))
  3524. BUG_ON(1);
  3525. page = grab_cache_page(inode->i_mapping, i);
  3526. if (!page) {
  3527. ret = -ENOMEM;
  3528. goto out_unlock;
  3529. }
  3530. if (!PageUptodate(page)) {
  3531. btrfs_readpage(NULL, page);
  3532. lock_page(page);
  3533. if (!PageUptodate(page)) {
  3534. unlock_page(page);
  3535. page_cache_release(page);
  3536. ret = -EIO;
  3537. goto out_unlock;
  3538. }
  3539. }
  3540. wait_on_page_writeback(page);
  3541. page_start = (u64)page->index << PAGE_CACHE_SHIFT;
  3542. page_end = page_start + PAGE_CACHE_SIZE - 1;
  3543. lock_extent(io_tree, page_start, page_end, GFP_NOFS);
  3544. ordered = btrfs_lookup_ordered_extent(inode, page_start);
  3545. if (ordered) {
  3546. unlock_extent(io_tree, page_start, page_end, GFP_NOFS);
  3547. unlock_page(page);
  3548. page_cache_release(page);
  3549. btrfs_start_ordered_extent(inode, ordered, 1);
  3550. btrfs_put_ordered_extent(ordered);
  3551. goto again;
  3552. }
  3553. set_page_extent_mapped(page);
  3554. btrfs_set_extent_delalloc(inode, page_start, page_end);
  3555. if (i == first_index)
  3556. set_extent_bits(io_tree, page_start, page_end,
  3557. EXTENT_BOUNDARY, GFP_NOFS);
  3558. set_page_dirty(page);
  3559. total_dirty++;
  3560. unlock_extent(io_tree, page_start, page_end, GFP_NOFS);
  3561. unlock_page(page);
  3562. page_cache_release(page);
  3563. }
  3564. out_unlock:
  3565. kfree(ra);
  3566. mutex_unlock(&inode->i_mutex);
  3567. balance_dirty_pages_ratelimited_nr(inode->i_mapping, total_dirty);
  3568. return ret;
  3569. }
  3570. static int noinline relocate_data_extent(struct inode *reloc_inode,
  3571. struct btrfs_key *extent_key,
  3572. u64 offset)
  3573. {
  3574. struct btrfs_root *root = BTRFS_I(reloc_inode)->root;
  3575. struct extent_map_tree *em_tree = &BTRFS_I(reloc_inode)->extent_tree;
  3576. struct extent_map *em;
  3577. u64 start = extent_key->objectid - offset;
  3578. u64 end = start + extent_key->offset - 1;
  3579. em = alloc_extent_map(GFP_NOFS);
  3580. BUG_ON(!em || IS_ERR(em));
  3581. em->start = start;
  3582. em->len = extent_key->offset;
  3583. em->block_len = extent_key->offset;
  3584. em->block_start = extent_key->objectid;
  3585. em->bdev = root->fs_info->fs_devices->latest_bdev;
  3586. set_bit(EXTENT_FLAG_PINNED, &em->flags);
  3587. /* setup extent map to cheat btrfs_readpage */
  3588. lock_extent(&BTRFS_I(reloc_inode)->io_tree, start, end, GFP_NOFS);
  3589. while (1) {
  3590. int ret;
  3591. spin_lock(&em_tree->lock);
  3592. ret = add_extent_mapping(em_tree, em);
  3593. spin_unlock(&em_tree->lock);
  3594. if (ret != -EEXIST) {
  3595. free_extent_map(em);
  3596. break;
  3597. }
  3598. btrfs_drop_extent_cache(reloc_inode, start, end, 0);
  3599. }
  3600. unlock_extent(&BTRFS_I(reloc_inode)->io_tree, start, end, GFP_NOFS);
  3601. return relocate_inode_pages(reloc_inode, start, extent_key->offset);
  3602. }
  3603. struct btrfs_ref_path {
  3604. u64 extent_start;
  3605. u64 nodes[BTRFS_MAX_LEVEL];
  3606. u64 root_objectid;
  3607. u64 root_generation;
  3608. u64 owner_objectid;
  3609. u32 num_refs;
  3610. int lowest_level;
  3611. int current_level;
  3612. int shared_level;
  3613. struct btrfs_key node_keys[BTRFS_MAX_LEVEL];
  3614. u64 new_nodes[BTRFS_MAX_LEVEL];
  3615. };
  3616. struct disk_extent {
  3617. u64 ram_bytes;
  3618. u64 disk_bytenr;
  3619. u64 disk_num_bytes;
  3620. u64 offset;
  3621. u64 num_bytes;
  3622. u8 compression;
  3623. u8 encryption;
  3624. u16 other_encoding;
  3625. };
  3626. static int is_cowonly_root(u64 root_objectid)
  3627. {
  3628. if (root_objectid == BTRFS_ROOT_TREE_OBJECTID ||
  3629. root_objectid == BTRFS_EXTENT_TREE_OBJECTID ||
  3630. root_objectid == BTRFS_CHUNK_TREE_OBJECTID ||
  3631. root_objectid == BTRFS_DEV_TREE_OBJECTID ||
  3632. root_objectid == BTRFS_TREE_LOG_OBJECTID ||
  3633. root_objectid == BTRFS_CSUM_TREE_OBJECTID)
  3634. return 1;
  3635. return 0;
  3636. }
  3637. static int noinline __next_ref_path(struct btrfs_trans_handle *trans,
  3638. struct btrfs_root *extent_root,
  3639. struct btrfs_ref_path *ref_path,
  3640. int first_time)
  3641. {
  3642. struct extent_buffer *leaf;
  3643. struct btrfs_path *path;
  3644. struct btrfs_extent_ref *ref;
  3645. struct btrfs_key key;
  3646. struct btrfs_key found_key;
  3647. u64 bytenr;
  3648. u32 nritems;
  3649. int level;
  3650. int ret = 1;
  3651. path = btrfs_alloc_path();
  3652. if (!path)
  3653. return -ENOMEM;
  3654. if (first_time) {
  3655. ref_path->lowest_level = -1;
  3656. ref_path->current_level = -1;
  3657. ref_path->shared_level = -1;
  3658. goto walk_up;
  3659. }
  3660. walk_down:
  3661. level = ref_path->current_level - 1;
  3662. while (level >= -1) {
  3663. u64 parent;
  3664. if (level < ref_path->lowest_level)
  3665. break;
  3666. if (level >= 0) {
  3667. bytenr = ref_path->nodes[level];
  3668. } else {
  3669. bytenr = ref_path->extent_start;
  3670. }
  3671. BUG_ON(bytenr == 0);
  3672. parent = ref_path->nodes[level + 1];
  3673. ref_path->nodes[level + 1] = 0;
  3674. ref_path->current_level = level;
  3675. BUG_ON(parent == 0);
  3676. key.objectid = bytenr;
  3677. key.offset = parent + 1;
  3678. key.type = BTRFS_EXTENT_REF_KEY;
  3679. ret = btrfs_search_slot(trans, extent_root, &key, path, 0, 0);
  3680. if (ret < 0)
  3681. goto out;
  3682. BUG_ON(ret == 0);
  3683. leaf = path->nodes[0];
  3684. nritems = btrfs_header_nritems(leaf);
  3685. if (path->slots[0] >= nritems) {
  3686. ret = btrfs_next_leaf(extent_root, path);
  3687. if (ret < 0)
  3688. goto out;
  3689. if (ret > 0)
  3690. goto next;
  3691. leaf = path->nodes[0];
  3692. }
  3693. btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
  3694. if (found_key.objectid == bytenr &&
  3695. found_key.type == BTRFS_EXTENT_REF_KEY) {
  3696. if (level < ref_path->shared_level)
  3697. ref_path->shared_level = level;
  3698. goto found;
  3699. }
  3700. next:
  3701. level--;
  3702. btrfs_release_path(extent_root, path);
  3703. cond_resched();
  3704. }
  3705. /* reached lowest level */
  3706. ret = 1;
  3707. goto out;
  3708. walk_up:
  3709. level = ref_path->current_level;
  3710. while (level < BTRFS_MAX_LEVEL - 1) {
  3711. u64 ref_objectid;
  3712. if (level >= 0) {
  3713. bytenr = ref_path->nodes[level];
  3714. } else {
  3715. bytenr = ref_path->extent_start;
  3716. }
  3717. BUG_ON(bytenr == 0);
  3718. key.objectid = bytenr;
  3719. key.offset = 0;
  3720. key.type = BTRFS_EXTENT_REF_KEY;
  3721. ret = btrfs_search_slot(trans, extent_root, &key, path, 0, 0);
  3722. if (ret < 0)
  3723. goto out;
  3724. leaf = path->nodes[0];
  3725. nritems = btrfs_header_nritems(leaf);
  3726. if (path->slots[0] >= nritems) {
  3727. ret = btrfs_next_leaf(extent_root, path);
  3728. if (ret < 0)
  3729. goto out;
  3730. if (ret > 0) {
  3731. /* the extent was freed by someone */
  3732. if (ref_path->lowest_level == level)
  3733. goto out;
  3734. btrfs_release_path(extent_root, path);
  3735. goto walk_down;
  3736. }
  3737. leaf = path->nodes[0];
  3738. }
  3739. btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
  3740. if (found_key.objectid != bytenr ||
  3741. found_key.type != BTRFS_EXTENT_REF_KEY) {
  3742. /* the extent was freed by someone */
  3743. if (ref_path->lowest_level == level) {
  3744. ret = 1;
  3745. goto out;
  3746. }
  3747. btrfs_release_path(extent_root, path);
  3748. goto walk_down;
  3749. }
  3750. found:
  3751. ref = btrfs_item_ptr(leaf, path->slots[0],
  3752. struct btrfs_extent_ref);
  3753. ref_objectid = btrfs_ref_objectid(leaf, ref);
  3754. if (ref_objectid < BTRFS_FIRST_FREE_OBJECTID) {
  3755. if (first_time) {
  3756. level = (int)ref_objectid;
  3757. BUG_ON(level >= BTRFS_MAX_LEVEL);
  3758. ref_path->lowest_level = level;
  3759. ref_path->current_level = level;
  3760. ref_path->nodes[level] = bytenr;
  3761. } else {
  3762. WARN_ON(ref_objectid != level);
  3763. }
  3764. } else {
  3765. WARN_ON(level != -1);
  3766. }
  3767. first_time = 0;
  3768. if (ref_path->lowest_level == level) {
  3769. ref_path->owner_objectid = ref_objectid;
  3770. ref_path->num_refs = btrfs_ref_num_refs(leaf, ref);
  3771. }
  3772. /*
  3773. * the block is tree root or the block isn't in reference
  3774. * counted tree.
  3775. */
  3776. if (found_key.objectid == found_key.offset ||
  3777. is_cowonly_root(btrfs_ref_root(leaf, ref))) {
  3778. ref_path->root_objectid = btrfs_ref_root(leaf, ref);
  3779. ref_path->root_generation =
  3780. btrfs_ref_generation(leaf, ref);
  3781. if (level < 0) {
  3782. /* special reference from the tree log */
  3783. ref_path->nodes[0] = found_key.offset;
  3784. ref_path->current_level = 0;
  3785. }
  3786. ret = 0;
  3787. goto out;
  3788. }
  3789. level++;
  3790. BUG_ON(ref_path->nodes[level] != 0);
  3791. ref_path->nodes[level] = found_key.offset;
  3792. ref_path->current_level = level;
  3793. /*
  3794. * the reference was created in the running transaction,
  3795. * no need to continue walking up.
  3796. */
  3797. if (btrfs_ref_generation(leaf, ref) == trans->transid) {
  3798. ref_path->root_objectid = btrfs_ref_root(leaf, ref);
  3799. ref_path->root_generation =
  3800. btrfs_ref_generation(leaf, ref);
  3801. ret = 0;
  3802. goto out;
  3803. }
  3804. btrfs_release_path(extent_root, path);
  3805. cond_resched();
  3806. }
  3807. /* reached max tree level, but no tree root found. */
  3808. BUG();
  3809. out:
  3810. btrfs_free_path(path);
  3811. return ret;
  3812. }
  3813. static int btrfs_first_ref_path(struct btrfs_trans_handle *trans,
  3814. struct btrfs_root *extent_root,
  3815. struct btrfs_ref_path *ref_path,
  3816. u64 extent_start)
  3817. {
  3818. memset(ref_path, 0, sizeof(*ref_path));
  3819. ref_path->extent_start = extent_start;
  3820. return __next_ref_path(trans, extent_root, ref_path, 1);
  3821. }
  3822. static int btrfs_next_ref_path(struct btrfs_trans_handle *trans,
  3823. struct btrfs_root *extent_root,
  3824. struct btrfs_ref_path *ref_path)
  3825. {
  3826. return __next_ref_path(trans, extent_root, ref_path, 0);
  3827. }
  3828. static int noinline get_new_locations(struct inode *reloc_inode,
  3829. struct btrfs_key *extent_key,
  3830. u64 offset, int no_fragment,
  3831. struct disk_extent **extents,
  3832. int *nr_extents)
  3833. {
  3834. struct btrfs_root *root = BTRFS_I(reloc_inode)->root;
  3835. struct btrfs_path *path;
  3836. struct btrfs_file_extent_item *fi;
  3837. struct extent_buffer *leaf;
  3838. struct disk_extent *exts = *extents;
  3839. struct btrfs_key found_key;
  3840. u64 cur_pos;
  3841. u64 last_byte;
  3842. u32 nritems;
  3843. int nr = 0;
  3844. int max = *nr_extents;
  3845. int ret;
  3846. WARN_ON(!no_fragment && *extents);
  3847. if (!exts) {
  3848. max = 1;
  3849. exts = kmalloc(sizeof(*exts) * max, GFP_NOFS);
  3850. if (!exts)
  3851. return -ENOMEM;
  3852. }
  3853. path = btrfs_alloc_path();
  3854. BUG_ON(!path);
  3855. cur_pos = extent_key->objectid - offset;
  3856. last_byte = extent_key->objectid + extent_key->offset;
  3857. ret = btrfs_lookup_file_extent(NULL, root, path, reloc_inode->i_ino,
  3858. cur_pos, 0);
  3859. if (ret < 0)
  3860. goto out;
  3861. if (ret > 0) {
  3862. ret = -ENOENT;
  3863. goto out;
  3864. }
  3865. while (1) {
  3866. leaf = path->nodes[0];
  3867. nritems = btrfs_header_nritems(leaf);
  3868. if (path->slots[0] >= nritems) {
  3869. ret = btrfs_next_leaf(root, path);
  3870. if (ret < 0)
  3871. goto out;
  3872. if (ret > 0)
  3873. break;
  3874. leaf = path->nodes[0];
  3875. }
  3876. btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
  3877. if (found_key.offset != cur_pos ||
  3878. found_key.type != BTRFS_EXTENT_DATA_KEY ||
  3879. found_key.objectid != reloc_inode->i_ino)
  3880. break;
  3881. fi = btrfs_item_ptr(leaf, path->slots[0],
  3882. struct btrfs_file_extent_item);
  3883. if (btrfs_file_extent_type(leaf, fi) !=
  3884. BTRFS_FILE_EXTENT_REG ||
  3885. btrfs_file_extent_disk_bytenr(leaf, fi) == 0)
  3886. break;
  3887. if (nr == max) {
  3888. struct disk_extent *old = exts;
  3889. max *= 2;
  3890. exts = kzalloc(sizeof(*exts) * max, GFP_NOFS);
  3891. memcpy(exts, old, sizeof(*exts) * nr);
  3892. if (old != *extents)
  3893. kfree(old);
  3894. }
  3895. exts[nr].disk_bytenr =
  3896. btrfs_file_extent_disk_bytenr(leaf, fi);
  3897. exts[nr].disk_num_bytes =
  3898. btrfs_file_extent_disk_num_bytes(leaf, fi);
  3899. exts[nr].offset = btrfs_file_extent_offset(leaf, fi);
  3900. exts[nr].num_bytes = btrfs_file_extent_num_bytes(leaf, fi);
  3901. exts[nr].ram_bytes = btrfs_file_extent_ram_bytes(leaf, fi);
  3902. exts[nr].compression = btrfs_file_extent_compression(leaf, fi);
  3903. exts[nr].encryption = btrfs_file_extent_encryption(leaf, fi);
  3904. exts[nr].other_encoding = btrfs_file_extent_other_encoding(leaf,
  3905. fi);
  3906. BUG_ON(exts[nr].offset > 0);
  3907. BUG_ON(exts[nr].compression || exts[nr].encryption);
  3908. BUG_ON(exts[nr].num_bytes != exts[nr].disk_num_bytes);
  3909. cur_pos += exts[nr].num_bytes;
  3910. nr++;
  3911. if (cur_pos + offset >= last_byte)
  3912. break;
  3913. if (no_fragment) {
  3914. ret = 1;
  3915. goto out;
  3916. }
  3917. path->slots[0]++;
  3918. }
  3919. WARN_ON(cur_pos + offset > last_byte);
  3920. if (cur_pos + offset < last_byte) {
  3921. ret = -ENOENT;
  3922. goto out;
  3923. }
  3924. ret = 0;
  3925. out:
  3926. btrfs_free_path(path);
  3927. if (ret) {
  3928. if (exts != *extents)
  3929. kfree(exts);
  3930. } else {
  3931. *extents = exts;
  3932. *nr_extents = nr;
  3933. }
  3934. return ret;
  3935. }
  3936. static int noinline replace_one_extent(struct btrfs_trans_handle *trans,
  3937. struct btrfs_root *root,
  3938. struct btrfs_path *path,
  3939. struct btrfs_key *extent_key,
  3940. struct btrfs_key *leaf_key,
  3941. struct btrfs_ref_path *ref_path,
  3942. struct disk_extent *new_extents,
  3943. int nr_extents)
  3944. {
  3945. struct extent_buffer *leaf;
  3946. struct btrfs_file_extent_item *fi;
  3947. struct inode *inode = NULL;
  3948. struct btrfs_key key;
  3949. u64 lock_start = 0;
  3950. u64 lock_end = 0;
  3951. u64 num_bytes;
  3952. u64 ext_offset;
  3953. u64 first_pos;
  3954. u32 nritems;
  3955. int nr_scaned = 0;
  3956. int extent_locked = 0;
  3957. int extent_type;
  3958. int ret;
  3959. memcpy(&key, leaf_key, sizeof(key));
  3960. first_pos = INT_LIMIT(loff_t) - extent_key->offset;
  3961. if (ref_path->owner_objectid != BTRFS_MULTIPLE_OBJECTIDS) {
  3962. if (key.objectid < ref_path->owner_objectid ||
  3963. (key.objectid == ref_path->owner_objectid &&
  3964. key.type < BTRFS_EXTENT_DATA_KEY)) {
  3965. key.objectid = ref_path->owner_objectid;
  3966. key.type = BTRFS_EXTENT_DATA_KEY;
  3967. key.offset = 0;
  3968. }
  3969. }
  3970. while (1) {
  3971. ret = btrfs_search_slot(trans, root, &key, path, 0, 1);
  3972. if (ret < 0)
  3973. goto out;
  3974. leaf = path->nodes[0];
  3975. nritems = btrfs_header_nritems(leaf);
  3976. next:
  3977. if (extent_locked && ret > 0) {
  3978. /*
  3979. * the file extent item was modified by someone
  3980. * before the extent got locked.
  3981. */
  3982. unlock_extent(&BTRFS_I(inode)->io_tree, lock_start,
  3983. lock_end, GFP_NOFS);
  3984. extent_locked = 0;
  3985. }
  3986. if (path->slots[0] >= nritems) {
  3987. if (++nr_scaned > 2)
  3988. break;
  3989. BUG_ON(extent_locked);
  3990. ret = btrfs_next_leaf(root, path);
  3991. if (ret < 0)
  3992. goto out;
  3993. if (ret > 0)
  3994. break;
  3995. leaf = path->nodes[0];
  3996. nritems = btrfs_header_nritems(leaf);
  3997. }
  3998. btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
  3999. if (ref_path->owner_objectid != BTRFS_MULTIPLE_OBJECTIDS) {
  4000. if ((key.objectid > ref_path->owner_objectid) ||
  4001. (key.objectid == ref_path->owner_objectid &&
  4002. key.type > BTRFS_EXTENT_DATA_KEY) ||
  4003. (key.offset >= first_pos + extent_key->offset))
  4004. break;
  4005. }
  4006. if (inode && key.objectid != inode->i_ino) {
  4007. BUG_ON(extent_locked);
  4008. btrfs_release_path(root, path);
  4009. mutex_unlock(&inode->i_mutex);
  4010. iput(inode);
  4011. inode = NULL;
  4012. continue;
  4013. }
  4014. if (key.type != BTRFS_EXTENT_DATA_KEY) {
  4015. path->slots[0]++;
  4016. ret = 1;
  4017. goto next;
  4018. }
  4019. fi = btrfs_item_ptr(leaf, path->slots[0],
  4020. struct btrfs_file_extent_item);
  4021. extent_type = btrfs_file_extent_type(leaf, fi);
  4022. if ((extent_type != BTRFS_FILE_EXTENT_REG &&
  4023. extent_type != BTRFS_FILE_EXTENT_PREALLOC) ||
  4024. (btrfs_file_extent_disk_bytenr(leaf, fi) !=
  4025. extent_key->objectid)) {
  4026. path->slots[0]++;
  4027. ret = 1;
  4028. goto next;
  4029. }
  4030. num_bytes = btrfs_file_extent_num_bytes(leaf, fi);
  4031. ext_offset = btrfs_file_extent_offset(leaf, fi);
  4032. if (first_pos > key.offset - ext_offset)
  4033. first_pos = key.offset - ext_offset;
  4034. if (!extent_locked) {
  4035. lock_start = key.offset;
  4036. lock_end = lock_start + num_bytes - 1;
  4037. } else {
  4038. if (lock_start > key.offset ||
  4039. lock_end + 1 < key.offset + num_bytes) {
  4040. unlock_extent(&BTRFS_I(inode)->io_tree,
  4041. lock_start, lock_end, GFP_NOFS);
  4042. extent_locked = 0;
  4043. }
  4044. }
  4045. if (!inode) {
  4046. btrfs_release_path(root, path);
  4047. inode = btrfs_iget_locked(root->fs_info->sb,
  4048. key.objectid, root);
  4049. if (inode->i_state & I_NEW) {
  4050. BTRFS_I(inode)->root = root;
  4051. BTRFS_I(inode)->location.objectid =
  4052. key.objectid;
  4053. BTRFS_I(inode)->location.type =
  4054. BTRFS_INODE_ITEM_KEY;
  4055. BTRFS_I(inode)->location.offset = 0;
  4056. btrfs_read_locked_inode(inode);
  4057. unlock_new_inode(inode);
  4058. }
  4059. /*
  4060. * some code call btrfs_commit_transaction while
  4061. * holding the i_mutex, so we can't use mutex_lock
  4062. * here.
  4063. */
  4064. if (is_bad_inode(inode) ||
  4065. !mutex_trylock(&inode->i_mutex)) {
  4066. iput(inode);
  4067. inode = NULL;
  4068. key.offset = (u64)-1;
  4069. goto skip;
  4070. }
  4071. }
  4072. if (!extent_locked) {
  4073. struct btrfs_ordered_extent *ordered;
  4074. btrfs_release_path(root, path);
  4075. lock_extent(&BTRFS_I(inode)->io_tree, lock_start,
  4076. lock_end, GFP_NOFS);
  4077. ordered = btrfs_lookup_first_ordered_extent(inode,
  4078. lock_end);
  4079. if (ordered &&
  4080. ordered->file_offset <= lock_end &&
  4081. ordered->file_offset + ordered->len > lock_start) {
  4082. unlock_extent(&BTRFS_I(inode)->io_tree,
  4083. lock_start, lock_end, GFP_NOFS);
  4084. btrfs_start_ordered_extent(inode, ordered, 1);
  4085. btrfs_put_ordered_extent(ordered);
  4086. key.offset += num_bytes;
  4087. goto skip;
  4088. }
  4089. if (ordered)
  4090. btrfs_put_ordered_extent(ordered);
  4091. extent_locked = 1;
  4092. continue;
  4093. }
  4094. if (nr_extents == 1) {
  4095. /* update extent pointer in place */
  4096. btrfs_set_file_extent_disk_bytenr(leaf, fi,
  4097. new_extents[0].disk_bytenr);
  4098. btrfs_set_file_extent_disk_num_bytes(leaf, fi,
  4099. new_extents[0].disk_num_bytes);
  4100. btrfs_mark_buffer_dirty(leaf);
  4101. btrfs_drop_extent_cache(inode, key.offset,
  4102. key.offset + num_bytes - 1, 0);
  4103. ret = btrfs_inc_extent_ref(trans, root,
  4104. new_extents[0].disk_bytenr,
  4105. new_extents[0].disk_num_bytes,
  4106. leaf->start,
  4107. root->root_key.objectid,
  4108. trans->transid,
  4109. key.objectid);
  4110. BUG_ON(ret);
  4111. ret = btrfs_free_extent(trans, root,
  4112. extent_key->objectid,
  4113. extent_key->offset,
  4114. leaf->start,
  4115. btrfs_header_owner(leaf),
  4116. btrfs_header_generation(leaf),
  4117. key.objectid, 0);
  4118. BUG_ON(ret);
  4119. btrfs_release_path(root, path);
  4120. key.offset += num_bytes;
  4121. } else {
  4122. BUG_ON(1);
  4123. #if 0
  4124. u64 alloc_hint;
  4125. u64 extent_len;
  4126. int i;
  4127. /*
  4128. * drop old extent pointer at first, then insert the
  4129. * new pointers one bye one
  4130. */
  4131. btrfs_release_path(root, path);
  4132. ret = btrfs_drop_extents(trans, root, inode, key.offset,
  4133. key.offset + num_bytes,
  4134. key.offset, &alloc_hint);
  4135. BUG_ON(ret);
  4136. for (i = 0; i < nr_extents; i++) {
  4137. if (ext_offset >= new_extents[i].num_bytes) {
  4138. ext_offset -= new_extents[i].num_bytes;
  4139. continue;
  4140. }
  4141. extent_len = min(new_extents[i].num_bytes -
  4142. ext_offset, num_bytes);
  4143. ret = btrfs_insert_empty_item(trans, root,
  4144. path, &key,
  4145. sizeof(*fi));
  4146. BUG_ON(ret);
  4147. leaf = path->nodes[0];
  4148. fi = btrfs_item_ptr(leaf, path->slots[0],
  4149. struct btrfs_file_extent_item);
  4150. btrfs_set_file_extent_generation(leaf, fi,
  4151. trans->transid);
  4152. btrfs_set_file_extent_type(leaf, fi,
  4153. BTRFS_FILE_EXTENT_REG);
  4154. btrfs_set_file_extent_disk_bytenr(leaf, fi,
  4155. new_extents[i].disk_bytenr);
  4156. btrfs_set_file_extent_disk_num_bytes(leaf, fi,
  4157. new_extents[i].disk_num_bytes);
  4158. btrfs_set_file_extent_ram_bytes(leaf, fi,
  4159. new_extents[i].ram_bytes);
  4160. btrfs_set_file_extent_compression(leaf, fi,
  4161. new_extents[i].compression);
  4162. btrfs_set_file_extent_encryption(leaf, fi,
  4163. new_extents[i].encryption);
  4164. btrfs_set_file_extent_other_encoding(leaf, fi,
  4165. new_extents[i].other_encoding);
  4166. btrfs_set_file_extent_num_bytes(leaf, fi,
  4167. extent_len);
  4168. ext_offset += new_extents[i].offset;
  4169. btrfs_set_file_extent_offset(leaf, fi,
  4170. ext_offset);
  4171. btrfs_mark_buffer_dirty(leaf);
  4172. btrfs_drop_extent_cache(inode, key.offset,
  4173. key.offset + extent_len - 1, 0);
  4174. ret = btrfs_inc_extent_ref(trans, root,
  4175. new_extents[i].disk_bytenr,
  4176. new_extents[i].disk_num_bytes,
  4177. leaf->start,
  4178. root->root_key.objectid,
  4179. trans->transid, key.objectid);
  4180. BUG_ON(ret);
  4181. btrfs_release_path(root, path);
  4182. inode_add_bytes(inode, extent_len);
  4183. ext_offset = 0;
  4184. num_bytes -= extent_len;
  4185. key.offset += extent_len;
  4186. if (num_bytes == 0)
  4187. break;
  4188. }
  4189. BUG_ON(i >= nr_extents);
  4190. #endif
  4191. }
  4192. if (extent_locked) {
  4193. unlock_extent(&BTRFS_I(inode)->io_tree, lock_start,
  4194. lock_end, GFP_NOFS);
  4195. extent_locked = 0;
  4196. }
  4197. skip:
  4198. if (ref_path->owner_objectid != BTRFS_MULTIPLE_OBJECTIDS &&
  4199. key.offset >= first_pos + extent_key->offset)
  4200. break;
  4201. cond_resched();
  4202. }
  4203. ret = 0;
  4204. out:
  4205. btrfs_release_path(root, path);
  4206. if (inode) {
  4207. mutex_unlock(&inode->i_mutex);
  4208. if (extent_locked) {
  4209. unlock_extent(&BTRFS_I(inode)->io_tree, lock_start,
  4210. lock_end, GFP_NOFS);
  4211. }
  4212. iput(inode);
  4213. }
  4214. return ret;
  4215. }
  4216. int btrfs_reloc_tree_cache_ref(struct btrfs_trans_handle *trans,
  4217. struct btrfs_root *root,
  4218. struct extent_buffer *buf, u64 orig_start)
  4219. {
  4220. int level;
  4221. int ret;
  4222. BUG_ON(btrfs_header_generation(buf) != trans->transid);
  4223. BUG_ON(root->root_key.objectid != BTRFS_TREE_RELOC_OBJECTID);
  4224. level = btrfs_header_level(buf);
  4225. if (level == 0) {
  4226. struct btrfs_leaf_ref *ref;
  4227. struct btrfs_leaf_ref *orig_ref;
  4228. orig_ref = btrfs_lookup_leaf_ref(root, orig_start);
  4229. if (!orig_ref)
  4230. return -ENOENT;
  4231. ref = btrfs_alloc_leaf_ref(root, orig_ref->nritems);
  4232. if (!ref) {
  4233. btrfs_free_leaf_ref(root, orig_ref);
  4234. return -ENOMEM;
  4235. }
  4236. ref->nritems = orig_ref->nritems;
  4237. memcpy(ref->extents, orig_ref->extents,
  4238. sizeof(ref->extents[0]) * ref->nritems);
  4239. btrfs_free_leaf_ref(root, orig_ref);
  4240. ref->root_gen = trans->transid;
  4241. ref->bytenr = buf->start;
  4242. ref->owner = btrfs_header_owner(buf);
  4243. ref->generation = btrfs_header_generation(buf);
  4244. ret = btrfs_add_leaf_ref(root, ref, 0);
  4245. WARN_ON(ret);
  4246. btrfs_free_leaf_ref(root, ref);
  4247. }
  4248. return 0;
  4249. }
  4250. static int noinline invalidate_extent_cache(struct btrfs_root *root,
  4251. struct extent_buffer *leaf,
  4252. struct btrfs_block_group_cache *group,
  4253. struct btrfs_root *target_root)
  4254. {
  4255. struct btrfs_key key;
  4256. struct inode *inode = NULL;
  4257. struct btrfs_file_extent_item *fi;
  4258. u64 num_bytes;
  4259. u64 skip_objectid = 0;
  4260. u32 nritems;
  4261. u32 i;
  4262. nritems = btrfs_header_nritems(leaf);
  4263. for (i = 0; i < nritems; i++) {
  4264. btrfs_item_key_to_cpu(leaf, &key, i);
  4265. if (key.objectid == skip_objectid ||
  4266. key.type != BTRFS_EXTENT_DATA_KEY)
  4267. continue;
  4268. fi = btrfs_item_ptr(leaf, i, struct btrfs_file_extent_item);
  4269. if (btrfs_file_extent_type(leaf, fi) ==
  4270. BTRFS_FILE_EXTENT_INLINE)
  4271. continue;
  4272. if (btrfs_file_extent_disk_bytenr(leaf, fi) == 0)
  4273. continue;
  4274. if (!inode || inode->i_ino != key.objectid) {
  4275. iput(inode);
  4276. inode = btrfs_ilookup(target_root->fs_info->sb,
  4277. key.objectid, target_root, 1);
  4278. }
  4279. if (!inode) {
  4280. skip_objectid = key.objectid;
  4281. continue;
  4282. }
  4283. num_bytes = btrfs_file_extent_num_bytes(leaf, fi);
  4284. lock_extent(&BTRFS_I(inode)->io_tree, key.offset,
  4285. key.offset + num_bytes - 1, GFP_NOFS);
  4286. btrfs_drop_extent_cache(inode, key.offset,
  4287. key.offset + num_bytes - 1, 1);
  4288. unlock_extent(&BTRFS_I(inode)->io_tree, key.offset,
  4289. key.offset + num_bytes - 1, GFP_NOFS);
  4290. cond_resched();
  4291. }
  4292. iput(inode);
  4293. return 0;
  4294. }
  4295. static int noinline replace_extents_in_leaf(struct btrfs_trans_handle *trans,
  4296. struct btrfs_root *root,
  4297. struct extent_buffer *leaf,
  4298. struct btrfs_block_group_cache *group,
  4299. struct inode *reloc_inode)
  4300. {
  4301. struct btrfs_key key;
  4302. struct btrfs_key extent_key;
  4303. struct btrfs_file_extent_item *fi;
  4304. struct btrfs_leaf_ref *ref;
  4305. struct disk_extent *new_extent;
  4306. u64 bytenr;
  4307. u64 num_bytes;
  4308. u32 nritems;
  4309. u32 i;
  4310. int ext_index;
  4311. int nr_extent;
  4312. int ret;
  4313. new_extent = kmalloc(sizeof(*new_extent), GFP_NOFS);
  4314. BUG_ON(!new_extent);
  4315. ref = btrfs_lookup_leaf_ref(root, leaf->start);
  4316. BUG_ON(!ref);
  4317. ext_index = -1;
  4318. nritems = btrfs_header_nritems(leaf);
  4319. for (i = 0; i < nritems; i++) {
  4320. btrfs_item_key_to_cpu(leaf, &key, i);
  4321. if (btrfs_key_type(&key) != BTRFS_EXTENT_DATA_KEY)
  4322. continue;
  4323. fi = btrfs_item_ptr(leaf, i, struct btrfs_file_extent_item);
  4324. if (btrfs_file_extent_type(leaf, fi) ==
  4325. BTRFS_FILE_EXTENT_INLINE)
  4326. continue;
  4327. bytenr = btrfs_file_extent_disk_bytenr(leaf, fi);
  4328. num_bytes = btrfs_file_extent_disk_num_bytes(leaf, fi);
  4329. if (bytenr == 0)
  4330. continue;
  4331. ext_index++;
  4332. if (bytenr >= group->key.objectid + group->key.offset ||
  4333. bytenr + num_bytes <= group->key.objectid)
  4334. continue;
  4335. extent_key.objectid = bytenr;
  4336. extent_key.offset = num_bytes;
  4337. extent_key.type = BTRFS_EXTENT_ITEM_KEY;
  4338. nr_extent = 1;
  4339. ret = get_new_locations(reloc_inode, &extent_key,
  4340. group->key.objectid, 1,
  4341. &new_extent, &nr_extent);
  4342. if (ret > 0)
  4343. continue;
  4344. BUG_ON(ret < 0);
  4345. BUG_ON(ref->extents[ext_index].bytenr != bytenr);
  4346. BUG_ON(ref->extents[ext_index].num_bytes != num_bytes);
  4347. ref->extents[ext_index].bytenr = new_extent->disk_bytenr;
  4348. ref->extents[ext_index].num_bytes = new_extent->disk_num_bytes;
  4349. btrfs_set_file_extent_disk_bytenr(leaf, fi,
  4350. new_extent->disk_bytenr);
  4351. btrfs_set_file_extent_disk_num_bytes(leaf, fi,
  4352. new_extent->disk_num_bytes);
  4353. btrfs_mark_buffer_dirty(leaf);
  4354. ret = btrfs_inc_extent_ref(trans, root,
  4355. new_extent->disk_bytenr,
  4356. new_extent->disk_num_bytes,
  4357. leaf->start,
  4358. root->root_key.objectid,
  4359. trans->transid, key.objectid);
  4360. BUG_ON(ret);
  4361. ret = btrfs_free_extent(trans, root,
  4362. bytenr, num_bytes, leaf->start,
  4363. btrfs_header_owner(leaf),
  4364. btrfs_header_generation(leaf),
  4365. key.objectid, 0);
  4366. BUG_ON(ret);
  4367. cond_resched();
  4368. }
  4369. kfree(new_extent);
  4370. BUG_ON(ext_index + 1 != ref->nritems);
  4371. btrfs_free_leaf_ref(root, ref);
  4372. return 0;
  4373. }
  4374. int btrfs_free_reloc_root(struct btrfs_trans_handle *trans,
  4375. struct btrfs_root *root)
  4376. {
  4377. struct btrfs_root *reloc_root;
  4378. int ret;
  4379. if (root->reloc_root) {
  4380. reloc_root = root->reloc_root;
  4381. root->reloc_root = NULL;
  4382. list_add(&reloc_root->dead_list,
  4383. &root->fs_info->dead_reloc_roots);
  4384. btrfs_set_root_bytenr(&reloc_root->root_item,
  4385. reloc_root->node->start);
  4386. btrfs_set_root_level(&root->root_item,
  4387. btrfs_header_level(reloc_root->node));
  4388. memset(&reloc_root->root_item.drop_progress, 0,
  4389. sizeof(struct btrfs_disk_key));
  4390. reloc_root->root_item.drop_level = 0;
  4391. ret = btrfs_update_root(trans, root->fs_info->tree_root,
  4392. &reloc_root->root_key,
  4393. &reloc_root->root_item);
  4394. BUG_ON(ret);
  4395. }
  4396. return 0;
  4397. }
  4398. int btrfs_drop_dead_reloc_roots(struct btrfs_root *root)
  4399. {
  4400. struct btrfs_trans_handle *trans;
  4401. struct btrfs_root *reloc_root;
  4402. struct btrfs_root *prev_root = NULL;
  4403. struct list_head dead_roots;
  4404. int ret;
  4405. unsigned long nr;
  4406. INIT_LIST_HEAD(&dead_roots);
  4407. list_splice_init(&root->fs_info->dead_reloc_roots, &dead_roots);
  4408. while (!list_empty(&dead_roots)) {
  4409. reloc_root = list_entry(dead_roots.prev,
  4410. struct btrfs_root, dead_list);
  4411. list_del_init(&reloc_root->dead_list);
  4412. BUG_ON(reloc_root->commit_root != NULL);
  4413. while (1) {
  4414. trans = btrfs_join_transaction(root, 1);
  4415. BUG_ON(!trans);
  4416. mutex_lock(&root->fs_info->drop_mutex);
  4417. ret = btrfs_drop_snapshot(trans, reloc_root);
  4418. if (ret != -EAGAIN)
  4419. break;
  4420. mutex_unlock(&root->fs_info->drop_mutex);
  4421. nr = trans->blocks_used;
  4422. ret = btrfs_end_transaction(trans, root);
  4423. BUG_ON(ret);
  4424. btrfs_btree_balance_dirty(root, nr);
  4425. }
  4426. free_extent_buffer(reloc_root->node);
  4427. ret = btrfs_del_root(trans, root->fs_info->tree_root,
  4428. &reloc_root->root_key);
  4429. BUG_ON(ret);
  4430. mutex_unlock(&root->fs_info->drop_mutex);
  4431. nr = trans->blocks_used;
  4432. ret = btrfs_end_transaction(trans, root);
  4433. BUG_ON(ret);
  4434. btrfs_btree_balance_dirty(root, nr);
  4435. kfree(prev_root);
  4436. prev_root = reloc_root;
  4437. }
  4438. if (prev_root) {
  4439. btrfs_remove_leaf_refs(prev_root, (u64)-1, 0);
  4440. kfree(prev_root);
  4441. }
  4442. return 0;
  4443. }
  4444. int btrfs_add_dead_reloc_root(struct btrfs_root *root)
  4445. {
  4446. list_add(&root->dead_list, &root->fs_info->dead_reloc_roots);
  4447. return 0;
  4448. }
  4449. int btrfs_cleanup_reloc_trees(struct btrfs_root *root)
  4450. {
  4451. struct btrfs_root *reloc_root;
  4452. struct btrfs_trans_handle *trans;
  4453. struct btrfs_key location;
  4454. int found;
  4455. int ret;
  4456. mutex_lock(&root->fs_info->tree_reloc_mutex);
  4457. ret = btrfs_find_dead_roots(root, BTRFS_TREE_RELOC_OBJECTID, NULL);
  4458. BUG_ON(ret);
  4459. found = !list_empty(&root->fs_info->dead_reloc_roots);
  4460. mutex_unlock(&root->fs_info->tree_reloc_mutex);
  4461. if (found) {
  4462. trans = btrfs_start_transaction(root, 1);
  4463. BUG_ON(!trans);
  4464. ret = btrfs_commit_transaction(trans, root);
  4465. BUG_ON(ret);
  4466. }
  4467. location.objectid = BTRFS_DATA_RELOC_TREE_OBJECTID;
  4468. location.offset = (u64)-1;
  4469. location.type = BTRFS_ROOT_ITEM_KEY;
  4470. reloc_root = btrfs_read_fs_root_no_name(root->fs_info, &location);
  4471. BUG_ON(!reloc_root);
  4472. btrfs_orphan_cleanup(reloc_root);
  4473. return 0;
  4474. }
  4475. static int noinline init_reloc_tree(struct btrfs_trans_handle *trans,
  4476. struct btrfs_root *root)
  4477. {
  4478. struct btrfs_root *reloc_root;
  4479. struct extent_buffer *eb;
  4480. struct btrfs_root_item *root_item;
  4481. struct btrfs_key root_key;
  4482. int ret;
  4483. BUG_ON(!root->ref_cows);
  4484. if (root->reloc_root)
  4485. return 0;
  4486. root_item = kmalloc(sizeof(*root_item), GFP_NOFS);
  4487. BUG_ON(!root_item);
  4488. ret = btrfs_copy_root(trans, root, root->commit_root,
  4489. &eb, BTRFS_TREE_RELOC_OBJECTID);
  4490. BUG_ON(ret);
  4491. root_key.objectid = BTRFS_TREE_RELOC_OBJECTID;
  4492. root_key.offset = root->root_key.objectid;
  4493. root_key.type = BTRFS_ROOT_ITEM_KEY;
  4494. memcpy(root_item, &root->root_item, sizeof(root_item));
  4495. btrfs_set_root_refs(root_item, 0);
  4496. btrfs_set_root_bytenr(root_item, eb->start);
  4497. btrfs_set_root_level(root_item, btrfs_header_level(eb));
  4498. btrfs_set_root_generation(root_item, trans->transid);
  4499. btrfs_tree_unlock(eb);
  4500. free_extent_buffer(eb);
  4501. ret = btrfs_insert_root(trans, root->fs_info->tree_root,
  4502. &root_key, root_item);
  4503. BUG_ON(ret);
  4504. kfree(root_item);
  4505. reloc_root = btrfs_read_fs_root_no_radix(root->fs_info->tree_root,
  4506. &root_key);
  4507. BUG_ON(!reloc_root);
  4508. reloc_root->last_trans = trans->transid;
  4509. reloc_root->commit_root = NULL;
  4510. reloc_root->ref_tree = &root->fs_info->reloc_ref_tree;
  4511. root->reloc_root = reloc_root;
  4512. return 0;
  4513. }
  4514. /*
  4515. * Core function of space balance.
  4516. *
  4517. * The idea is using reloc trees to relocate tree blocks in reference
  4518. * counted roots. There is one reloc tree for each subvol, and all
  4519. * reloc trees share same root key objectid. Reloc trees are snapshots
  4520. * of the latest committed roots of subvols (root->commit_root).
  4521. *
  4522. * To relocate a tree block referenced by a subvol, there are two steps.
  4523. * COW the block through subvol's reloc tree, then update block pointer
  4524. * in the subvol to point to the new block. Since all reloc trees share
  4525. * same root key objectid, doing special handing for tree blocks owned
  4526. * by them is easy. Once a tree block has been COWed in one reloc tree,
  4527. * we can use the resulting new block directly when the same block is
  4528. * required to COW again through other reloc trees. By this way, relocated
  4529. * tree blocks are shared between reloc trees, so they are also shared
  4530. * between subvols.
  4531. */
  4532. static int noinline relocate_one_path(struct btrfs_trans_handle *trans,
  4533. struct btrfs_root *root,
  4534. struct btrfs_path *path,
  4535. struct btrfs_key *first_key,
  4536. struct btrfs_ref_path *ref_path,
  4537. struct btrfs_block_group_cache *group,
  4538. struct inode *reloc_inode)
  4539. {
  4540. struct btrfs_root *reloc_root;
  4541. struct extent_buffer *eb = NULL;
  4542. struct btrfs_key *keys;
  4543. u64 *nodes;
  4544. int level;
  4545. int shared_level;
  4546. int lowest_level = 0;
  4547. int ret;
  4548. if (ref_path->owner_objectid < BTRFS_FIRST_FREE_OBJECTID)
  4549. lowest_level = ref_path->owner_objectid;
  4550. if (!root->ref_cows) {
  4551. path->lowest_level = lowest_level;
  4552. ret = btrfs_search_slot(trans, root, first_key, path, 0, 1);
  4553. BUG_ON(ret < 0);
  4554. path->lowest_level = 0;
  4555. btrfs_release_path(root, path);
  4556. return 0;
  4557. }
  4558. mutex_lock(&root->fs_info->tree_reloc_mutex);
  4559. ret = init_reloc_tree(trans, root);
  4560. BUG_ON(ret);
  4561. reloc_root = root->reloc_root;
  4562. shared_level = ref_path->shared_level;
  4563. ref_path->shared_level = BTRFS_MAX_LEVEL - 1;
  4564. keys = ref_path->node_keys;
  4565. nodes = ref_path->new_nodes;
  4566. memset(&keys[shared_level + 1], 0,
  4567. sizeof(*keys) * (BTRFS_MAX_LEVEL - shared_level - 1));
  4568. memset(&nodes[shared_level + 1], 0,
  4569. sizeof(*nodes) * (BTRFS_MAX_LEVEL - shared_level - 1));
  4570. if (nodes[lowest_level] == 0) {
  4571. path->lowest_level = lowest_level;
  4572. ret = btrfs_search_slot(trans, reloc_root, first_key, path,
  4573. 0, 1);
  4574. BUG_ON(ret);
  4575. for (level = lowest_level; level < BTRFS_MAX_LEVEL; level++) {
  4576. eb = path->nodes[level];
  4577. if (!eb || eb == reloc_root->node)
  4578. break;
  4579. nodes[level] = eb->start;
  4580. if (level == 0)
  4581. btrfs_item_key_to_cpu(eb, &keys[level], 0);
  4582. else
  4583. btrfs_node_key_to_cpu(eb, &keys[level], 0);
  4584. }
  4585. if (nodes[0] &&
  4586. ref_path->owner_objectid >= BTRFS_FIRST_FREE_OBJECTID) {
  4587. eb = path->nodes[0];
  4588. ret = replace_extents_in_leaf(trans, reloc_root, eb,
  4589. group, reloc_inode);
  4590. BUG_ON(ret);
  4591. }
  4592. btrfs_release_path(reloc_root, path);
  4593. } else {
  4594. ret = btrfs_merge_path(trans, reloc_root, keys, nodes,
  4595. lowest_level);
  4596. BUG_ON(ret);
  4597. }
  4598. /*
  4599. * replace tree blocks in the fs tree with tree blocks in
  4600. * the reloc tree.
  4601. */
  4602. ret = btrfs_merge_path(trans, root, keys, nodes, lowest_level);
  4603. BUG_ON(ret < 0);
  4604. if (ref_path->owner_objectid >= BTRFS_FIRST_FREE_OBJECTID) {
  4605. ret = btrfs_search_slot(trans, reloc_root, first_key, path,
  4606. 0, 0);
  4607. BUG_ON(ret);
  4608. extent_buffer_get(path->nodes[0]);
  4609. eb = path->nodes[0];
  4610. btrfs_release_path(reloc_root, path);
  4611. ret = invalidate_extent_cache(reloc_root, eb, group, root);
  4612. BUG_ON(ret);
  4613. free_extent_buffer(eb);
  4614. }
  4615. mutex_unlock(&root->fs_info->tree_reloc_mutex);
  4616. path->lowest_level = 0;
  4617. return 0;
  4618. }
  4619. static int noinline relocate_tree_block(struct btrfs_trans_handle *trans,
  4620. struct btrfs_root *root,
  4621. struct btrfs_path *path,
  4622. struct btrfs_key *first_key,
  4623. struct btrfs_ref_path *ref_path)
  4624. {
  4625. int ret;
  4626. ret = relocate_one_path(trans, root, path, first_key,
  4627. ref_path, NULL, NULL);
  4628. BUG_ON(ret);
  4629. if (root == root->fs_info->extent_root)
  4630. btrfs_extent_post_op(trans, root);
  4631. return 0;
  4632. }
  4633. static int noinline del_extent_zero(struct btrfs_trans_handle *trans,
  4634. struct btrfs_root *extent_root,
  4635. struct btrfs_path *path,
  4636. struct btrfs_key *extent_key)
  4637. {
  4638. int ret;
  4639. ret = btrfs_search_slot(trans, extent_root, extent_key, path, -1, 1);
  4640. if (ret)
  4641. goto out;
  4642. ret = btrfs_del_item(trans, extent_root, path);
  4643. out:
  4644. btrfs_release_path(extent_root, path);
  4645. return ret;
  4646. }
  4647. static struct btrfs_root noinline *read_ref_root(struct btrfs_fs_info *fs_info,
  4648. struct btrfs_ref_path *ref_path)
  4649. {
  4650. struct btrfs_key root_key;
  4651. root_key.objectid = ref_path->root_objectid;
  4652. root_key.type = BTRFS_ROOT_ITEM_KEY;
  4653. if (is_cowonly_root(ref_path->root_objectid))
  4654. root_key.offset = 0;
  4655. else
  4656. root_key.offset = (u64)-1;
  4657. return btrfs_read_fs_root_no_name(fs_info, &root_key);
  4658. }
  4659. static int noinline relocate_one_extent(struct btrfs_root *extent_root,
  4660. struct btrfs_path *path,
  4661. struct btrfs_key *extent_key,
  4662. struct btrfs_block_group_cache *group,
  4663. struct inode *reloc_inode, int pass)
  4664. {
  4665. struct btrfs_trans_handle *trans;
  4666. struct btrfs_root *found_root;
  4667. struct btrfs_ref_path *ref_path = NULL;
  4668. struct disk_extent *new_extents = NULL;
  4669. int nr_extents = 0;
  4670. int loops;
  4671. int ret;
  4672. int level;
  4673. struct btrfs_key first_key;
  4674. u64 prev_block = 0;
  4675. trans = btrfs_start_transaction(extent_root, 1);
  4676. BUG_ON(!trans);
  4677. if (extent_key->objectid == 0) {
  4678. ret = del_extent_zero(trans, extent_root, path, extent_key);
  4679. goto out;
  4680. }
  4681. ref_path = kmalloc(sizeof(*ref_path), GFP_NOFS);
  4682. if (!ref_path) {
  4683. ret = -ENOMEM;
  4684. goto out;
  4685. }
  4686. for (loops = 0; ; loops++) {
  4687. if (loops == 0) {
  4688. ret = btrfs_first_ref_path(trans, extent_root, ref_path,
  4689. extent_key->objectid);
  4690. } else {
  4691. ret = btrfs_next_ref_path(trans, extent_root, ref_path);
  4692. }
  4693. if (ret < 0)
  4694. goto out;
  4695. if (ret > 0)
  4696. break;
  4697. if (ref_path->root_objectid == BTRFS_TREE_LOG_OBJECTID ||
  4698. ref_path->root_objectid == BTRFS_TREE_RELOC_OBJECTID)
  4699. continue;
  4700. found_root = read_ref_root(extent_root->fs_info, ref_path);
  4701. BUG_ON(!found_root);
  4702. /*
  4703. * for reference counted tree, only process reference paths
  4704. * rooted at the latest committed root.
  4705. */
  4706. if (found_root->ref_cows &&
  4707. ref_path->root_generation != found_root->root_key.offset)
  4708. continue;
  4709. if (ref_path->owner_objectid >= BTRFS_FIRST_FREE_OBJECTID) {
  4710. if (pass == 0) {
  4711. /*
  4712. * copy data extents to new locations
  4713. */
  4714. u64 group_start = group->key.objectid;
  4715. ret = relocate_data_extent(reloc_inode,
  4716. extent_key,
  4717. group_start);
  4718. if (ret < 0)
  4719. goto out;
  4720. break;
  4721. }
  4722. level = 0;
  4723. } else {
  4724. level = ref_path->owner_objectid;
  4725. }
  4726. if (prev_block != ref_path->nodes[level]) {
  4727. struct extent_buffer *eb;
  4728. u64 block_start = ref_path->nodes[level];
  4729. u64 block_size = btrfs_level_size(found_root, level);
  4730. eb = read_tree_block(found_root, block_start,
  4731. block_size, 0);
  4732. btrfs_tree_lock(eb);
  4733. BUG_ON(level != btrfs_header_level(eb));
  4734. if (level == 0)
  4735. btrfs_item_key_to_cpu(eb, &first_key, 0);
  4736. else
  4737. btrfs_node_key_to_cpu(eb, &first_key, 0);
  4738. btrfs_tree_unlock(eb);
  4739. free_extent_buffer(eb);
  4740. prev_block = block_start;
  4741. }
  4742. btrfs_record_root_in_trans(found_root);
  4743. if (ref_path->owner_objectid >= BTRFS_FIRST_FREE_OBJECTID) {
  4744. /*
  4745. * try to update data extent references while
  4746. * keeping metadata shared between snapshots.
  4747. */
  4748. if (pass == 1) {
  4749. ret = relocate_one_path(trans, found_root,
  4750. path, &first_key, ref_path,
  4751. group, reloc_inode);
  4752. if (ret < 0)
  4753. goto out;
  4754. continue;
  4755. }
  4756. /*
  4757. * use fallback method to process the remaining
  4758. * references.
  4759. */
  4760. if (!new_extents) {
  4761. u64 group_start = group->key.objectid;
  4762. new_extents = kmalloc(sizeof(*new_extents),
  4763. GFP_NOFS);
  4764. nr_extents = 1;
  4765. ret = get_new_locations(reloc_inode,
  4766. extent_key,
  4767. group_start, 1,
  4768. &new_extents,
  4769. &nr_extents);
  4770. if (ret)
  4771. goto out;
  4772. }
  4773. ret = replace_one_extent(trans, found_root,
  4774. path, extent_key,
  4775. &first_key, ref_path,
  4776. new_extents, nr_extents);
  4777. } else {
  4778. ret = relocate_tree_block(trans, found_root, path,
  4779. &first_key, ref_path);
  4780. }
  4781. if (ret < 0)
  4782. goto out;
  4783. }
  4784. ret = 0;
  4785. out:
  4786. btrfs_end_transaction(trans, extent_root);
  4787. kfree(new_extents);
  4788. kfree(ref_path);
  4789. return ret;
  4790. }
  4791. static u64 update_block_group_flags(struct btrfs_root *root, u64 flags)
  4792. {
  4793. u64 num_devices;
  4794. u64 stripped = BTRFS_BLOCK_GROUP_RAID0 |
  4795. BTRFS_BLOCK_GROUP_RAID1 | BTRFS_BLOCK_GROUP_RAID10;
  4796. num_devices = root->fs_info->fs_devices->rw_devices;
  4797. if (num_devices == 1) {
  4798. stripped |= BTRFS_BLOCK_GROUP_DUP;
  4799. stripped = flags & ~stripped;
  4800. /* turn raid0 into single device chunks */
  4801. if (flags & BTRFS_BLOCK_GROUP_RAID0)
  4802. return stripped;
  4803. /* turn mirroring into duplication */
  4804. if (flags & (BTRFS_BLOCK_GROUP_RAID1 |
  4805. BTRFS_BLOCK_GROUP_RAID10))
  4806. return stripped | BTRFS_BLOCK_GROUP_DUP;
  4807. return flags;
  4808. } else {
  4809. /* they already had raid on here, just return */
  4810. if (flags & stripped)
  4811. return flags;
  4812. stripped |= BTRFS_BLOCK_GROUP_DUP;
  4813. stripped = flags & ~stripped;
  4814. /* switch duplicated blocks with raid1 */
  4815. if (flags & BTRFS_BLOCK_GROUP_DUP)
  4816. return stripped | BTRFS_BLOCK_GROUP_RAID1;
  4817. /* turn single device chunks into raid0 */
  4818. return stripped | BTRFS_BLOCK_GROUP_RAID0;
  4819. }
  4820. return flags;
  4821. }
  4822. static int __alloc_chunk_for_shrink(struct btrfs_root *root,
  4823. struct btrfs_block_group_cache *shrink_block_group,
  4824. int force)
  4825. {
  4826. struct btrfs_trans_handle *trans;
  4827. u64 new_alloc_flags;
  4828. u64 calc;
  4829. spin_lock(&shrink_block_group->lock);
  4830. if (btrfs_block_group_used(&shrink_block_group->item) > 0) {
  4831. spin_unlock(&shrink_block_group->lock);
  4832. trans = btrfs_start_transaction(root, 1);
  4833. spin_lock(&shrink_block_group->lock);
  4834. new_alloc_flags = update_block_group_flags(root,
  4835. shrink_block_group->flags);
  4836. if (new_alloc_flags != shrink_block_group->flags) {
  4837. calc =
  4838. btrfs_block_group_used(&shrink_block_group->item);
  4839. } else {
  4840. calc = shrink_block_group->key.offset;
  4841. }
  4842. spin_unlock(&shrink_block_group->lock);
  4843. do_chunk_alloc(trans, root->fs_info->extent_root,
  4844. calc + 2 * 1024 * 1024, new_alloc_flags, force);
  4845. btrfs_end_transaction(trans, root);
  4846. } else
  4847. spin_unlock(&shrink_block_group->lock);
  4848. return 0;
  4849. }
  4850. static int __insert_orphan_inode(struct btrfs_trans_handle *trans,
  4851. struct btrfs_root *root,
  4852. u64 objectid, u64 size)
  4853. {
  4854. struct btrfs_path *path;
  4855. struct btrfs_inode_item *item;
  4856. struct extent_buffer *leaf;
  4857. int ret;
  4858. path = btrfs_alloc_path();
  4859. if (!path)
  4860. return -ENOMEM;
  4861. ret = btrfs_insert_empty_inode(trans, root, path, objectid);
  4862. if (ret)
  4863. goto out;
  4864. leaf = path->nodes[0];
  4865. item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_inode_item);
  4866. memset_extent_buffer(leaf, 0, (unsigned long)item, sizeof(*item));
  4867. btrfs_set_inode_generation(leaf, item, 1);
  4868. btrfs_set_inode_size(leaf, item, size);
  4869. btrfs_set_inode_mode(leaf, item, S_IFREG | 0600);
  4870. btrfs_set_inode_flags(leaf, item, BTRFS_INODE_NOCOMPRESS);
  4871. btrfs_mark_buffer_dirty(leaf);
  4872. btrfs_release_path(root, path);
  4873. out:
  4874. btrfs_free_path(path);
  4875. return ret;
  4876. }
  4877. static struct inode noinline *create_reloc_inode(struct btrfs_fs_info *fs_info,
  4878. struct btrfs_block_group_cache *group)
  4879. {
  4880. struct inode *inode = NULL;
  4881. struct btrfs_trans_handle *trans;
  4882. struct btrfs_root *root;
  4883. struct btrfs_key root_key;
  4884. u64 objectid = BTRFS_FIRST_FREE_OBJECTID;
  4885. int err = 0;
  4886. root_key.objectid = BTRFS_DATA_RELOC_TREE_OBJECTID;
  4887. root_key.type = BTRFS_ROOT_ITEM_KEY;
  4888. root_key.offset = (u64)-1;
  4889. root = btrfs_read_fs_root_no_name(fs_info, &root_key);
  4890. if (IS_ERR(root))
  4891. return ERR_CAST(root);
  4892. trans = btrfs_start_transaction(root, 1);
  4893. BUG_ON(!trans);
  4894. err = btrfs_find_free_objectid(trans, root, objectid, &objectid);
  4895. if (err)
  4896. goto out;
  4897. err = __insert_orphan_inode(trans, root, objectid, group->key.offset);
  4898. BUG_ON(err);
  4899. err = btrfs_insert_file_extent(trans, root, objectid, 0, 0, 0,
  4900. group->key.offset, 0, group->key.offset,
  4901. 0, 0, 0);
  4902. BUG_ON(err);
  4903. inode = btrfs_iget_locked(root->fs_info->sb, objectid, root);
  4904. if (inode->i_state & I_NEW) {
  4905. BTRFS_I(inode)->root = root;
  4906. BTRFS_I(inode)->location.objectid = objectid;
  4907. BTRFS_I(inode)->location.type = BTRFS_INODE_ITEM_KEY;
  4908. BTRFS_I(inode)->location.offset = 0;
  4909. btrfs_read_locked_inode(inode);
  4910. unlock_new_inode(inode);
  4911. BUG_ON(is_bad_inode(inode));
  4912. } else {
  4913. BUG_ON(1);
  4914. }
  4915. BTRFS_I(inode)->index_cnt = group->key.objectid;
  4916. err = btrfs_orphan_add(trans, inode);
  4917. out:
  4918. btrfs_end_transaction(trans, root);
  4919. if (err) {
  4920. if (inode)
  4921. iput(inode);
  4922. inode = ERR_PTR(err);
  4923. }
  4924. return inode;
  4925. }
  4926. int btrfs_reloc_clone_csums(struct inode *inode, u64 file_pos, u64 len)
  4927. {
  4928. struct btrfs_ordered_sum *sums;
  4929. struct btrfs_sector_sum *sector_sum;
  4930. struct btrfs_ordered_extent *ordered;
  4931. struct btrfs_root *root = BTRFS_I(inode)->root;
  4932. struct list_head list;
  4933. size_t offset;
  4934. int ret;
  4935. u64 disk_bytenr;
  4936. INIT_LIST_HEAD(&list);
  4937. ordered = btrfs_lookup_ordered_extent(inode, file_pos);
  4938. BUG_ON(ordered->file_offset != file_pos || ordered->len != len);
  4939. disk_bytenr = file_pos + BTRFS_I(inode)->index_cnt;
  4940. ret = btrfs_lookup_csums_range(root, disk_bytenr,
  4941. disk_bytenr + len - 1, &list);
  4942. while (!list_empty(&list)) {
  4943. sums = list_entry(list.next, struct btrfs_ordered_sum, list);
  4944. list_del_init(&sums->list);
  4945. sector_sum = sums->sums;
  4946. sums->bytenr = ordered->start;
  4947. offset = 0;
  4948. while (offset < sums->len) {
  4949. sector_sum->bytenr += ordered->start - disk_bytenr;
  4950. sector_sum++;
  4951. offset += root->sectorsize;
  4952. }
  4953. btrfs_add_ordered_sum(inode, ordered, sums);
  4954. }
  4955. btrfs_put_ordered_extent(ordered);
  4956. return 0;
  4957. }
  4958. int btrfs_relocate_block_group(struct btrfs_root *root, u64 group_start)
  4959. {
  4960. struct btrfs_trans_handle *trans;
  4961. struct btrfs_path *path;
  4962. struct btrfs_fs_info *info = root->fs_info;
  4963. struct extent_buffer *leaf;
  4964. struct inode *reloc_inode;
  4965. struct btrfs_block_group_cache *block_group;
  4966. struct btrfs_key key;
  4967. u64 skipped;
  4968. u64 cur_byte;
  4969. u64 total_found;
  4970. u32 nritems;
  4971. int ret;
  4972. int progress;
  4973. int pass = 0;
  4974. root = root->fs_info->extent_root;
  4975. block_group = btrfs_lookup_block_group(info, group_start);
  4976. BUG_ON(!block_group);
  4977. printk("btrfs relocating block group %llu flags %llu\n",
  4978. (unsigned long long)block_group->key.objectid,
  4979. (unsigned long long)block_group->flags);
  4980. path = btrfs_alloc_path();
  4981. BUG_ON(!path);
  4982. reloc_inode = create_reloc_inode(info, block_group);
  4983. BUG_ON(IS_ERR(reloc_inode));
  4984. __alloc_chunk_for_shrink(root, block_group, 1);
  4985. set_block_group_readonly(block_group);
  4986. btrfs_start_delalloc_inodes(info->tree_root);
  4987. btrfs_wait_ordered_extents(info->tree_root, 0);
  4988. again:
  4989. skipped = 0;
  4990. total_found = 0;
  4991. progress = 0;
  4992. key.objectid = block_group->key.objectid;
  4993. key.offset = 0;
  4994. key.type = 0;
  4995. cur_byte = key.objectid;
  4996. trans = btrfs_start_transaction(info->tree_root, 1);
  4997. btrfs_commit_transaction(trans, info->tree_root);
  4998. mutex_lock(&root->fs_info->cleaner_mutex);
  4999. btrfs_clean_old_snapshots(info->tree_root);
  5000. btrfs_remove_leaf_refs(info->tree_root, (u64)-1, 1);
  5001. mutex_unlock(&root->fs_info->cleaner_mutex);
  5002. while(1) {
  5003. ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
  5004. if (ret < 0)
  5005. goto out;
  5006. next:
  5007. leaf = path->nodes[0];
  5008. nritems = btrfs_header_nritems(leaf);
  5009. if (path->slots[0] >= nritems) {
  5010. ret = btrfs_next_leaf(root, path);
  5011. if (ret < 0)
  5012. goto out;
  5013. if (ret == 1) {
  5014. ret = 0;
  5015. break;
  5016. }
  5017. leaf = path->nodes[0];
  5018. nritems = btrfs_header_nritems(leaf);
  5019. }
  5020. btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
  5021. if (key.objectid >= block_group->key.objectid +
  5022. block_group->key.offset)
  5023. break;
  5024. if (progress && need_resched()) {
  5025. btrfs_release_path(root, path);
  5026. cond_resched();
  5027. progress = 0;
  5028. continue;
  5029. }
  5030. progress = 1;
  5031. if (btrfs_key_type(&key) != BTRFS_EXTENT_ITEM_KEY ||
  5032. key.objectid + key.offset <= cur_byte) {
  5033. path->slots[0]++;
  5034. goto next;
  5035. }
  5036. total_found++;
  5037. cur_byte = key.objectid + key.offset;
  5038. btrfs_release_path(root, path);
  5039. __alloc_chunk_for_shrink(root, block_group, 0);
  5040. ret = relocate_one_extent(root, path, &key, block_group,
  5041. reloc_inode, pass);
  5042. BUG_ON(ret < 0);
  5043. if (ret > 0)
  5044. skipped++;
  5045. key.objectid = cur_byte;
  5046. key.type = 0;
  5047. key.offset = 0;
  5048. }
  5049. btrfs_release_path(root, path);
  5050. if (pass == 0) {
  5051. btrfs_wait_ordered_range(reloc_inode, 0, (u64)-1);
  5052. invalidate_mapping_pages(reloc_inode->i_mapping, 0, -1);
  5053. WARN_ON(reloc_inode->i_mapping->nrpages);
  5054. }
  5055. if (total_found > 0) {
  5056. printk("btrfs found %llu extents in pass %d\n",
  5057. (unsigned long long)total_found, pass);
  5058. pass++;
  5059. if (total_found == skipped && pass > 2) {
  5060. iput(reloc_inode);
  5061. reloc_inode = create_reloc_inode(info, block_group);
  5062. pass = 0;
  5063. }
  5064. goto again;
  5065. }
  5066. /* delete reloc_inode */
  5067. iput(reloc_inode);
  5068. /* unpin extents in this range */
  5069. trans = btrfs_start_transaction(info->tree_root, 1);
  5070. btrfs_commit_transaction(trans, info->tree_root);
  5071. spin_lock(&block_group->lock);
  5072. WARN_ON(block_group->pinned > 0);
  5073. WARN_ON(block_group->reserved > 0);
  5074. WARN_ON(btrfs_block_group_used(&block_group->item) > 0);
  5075. spin_unlock(&block_group->lock);
  5076. put_block_group(block_group);
  5077. ret = 0;
  5078. out:
  5079. btrfs_free_path(path);
  5080. return ret;
  5081. }
  5082. static int find_first_block_group(struct btrfs_root *root,
  5083. struct btrfs_path *path, struct btrfs_key *key)
  5084. {
  5085. int ret = 0;
  5086. struct btrfs_key found_key;
  5087. struct extent_buffer *leaf;
  5088. int slot;
  5089. ret = btrfs_search_slot(NULL, root, key, path, 0, 0);
  5090. if (ret < 0)
  5091. goto out;
  5092. while(1) {
  5093. slot = path->slots[0];
  5094. leaf = path->nodes[0];
  5095. if (slot >= btrfs_header_nritems(leaf)) {
  5096. ret = btrfs_next_leaf(root, path);
  5097. if (ret == 0)
  5098. continue;
  5099. if (ret < 0)
  5100. goto out;
  5101. break;
  5102. }
  5103. btrfs_item_key_to_cpu(leaf, &found_key, slot);
  5104. if (found_key.objectid >= key->objectid &&
  5105. found_key.type == BTRFS_BLOCK_GROUP_ITEM_KEY) {
  5106. ret = 0;
  5107. goto out;
  5108. }
  5109. path->slots[0]++;
  5110. }
  5111. ret = -ENOENT;
  5112. out:
  5113. return ret;
  5114. }
  5115. int btrfs_free_block_groups(struct btrfs_fs_info *info)
  5116. {
  5117. struct btrfs_block_group_cache *block_group;
  5118. struct rb_node *n;
  5119. spin_lock(&info->block_group_cache_lock);
  5120. while ((n = rb_last(&info->block_group_cache_tree)) != NULL) {
  5121. block_group = rb_entry(n, struct btrfs_block_group_cache,
  5122. cache_node);
  5123. rb_erase(&block_group->cache_node,
  5124. &info->block_group_cache_tree);
  5125. spin_unlock(&info->block_group_cache_lock);
  5126. btrfs_remove_free_space_cache(block_group);
  5127. down_write(&block_group->space_info->groups_sem);
  5128. list_del(&block_group->list);
  5129. up_write(&block_group->space_info->groups_sem);
  5130. WARN_ON(atomic_read(&block_group->count) != 1);
  5131. kfree(block_group);
  5132. spin_lock(&info->block_group_cache_lock);
  5133. }
  5134. spin_unlock(&info->block_group_cache_lock);
  5135. return 0;
  5136. }
  5137. int btrfs_read_block_groups(struct btrfs_root *root)
  5138. {
  5139. struct btrfs_path *path;
  5140. int ret;
  5141. struct btrfs_block_group_cache *cache;
  5142. struct btrfs_fs_info *info = root->fs_info;
  5143. struct btrfs_space_info *space_info;
  5144. struct btrfs_key key;
  5145. struct btrfs_key found_key;
  5146. struct extent_buffer *leaf;
  5147. root = info->extent_root;
  5148. key.objectid = 0;
  5149. key.offset = 0;
  5150. btrfs_set_key_type(&key, BTRFS_BLOCK_GROUP_ITEM_KEY);
  5151. path = btrfs_alloc_path();
  5152. if (!path)
  5153. return -ENOMEM;
  5154. while(1) {
  5155. ret = find_first_block_group(root, path, &key);
  5156. if (ret > 0) {
  5157. ret = 0;
  5158. goto error;
  5159. }
  5160. if (ret != 0)
  5161. goto error;
  5162. leaf = path->nodes[0];
  5163. btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
  5164. cache = kzalloc(sizeof(*cache), GFP_NOFS);
  5165. if (!cache) {
  5166. ret = -ENOMEM;
  5167. break;
  5168. }
  5169. atomic_set(&cache->count, 1);
  5170. spin_lock_init(&cache->lock);
  5171. mutex_init(&cache->alloc_mutex);
  5172. mutex_init(&cache->cache_mutex);
  5173. INIT_LIST_HEAD(&cache->list);
  5174. read_extent_buffer(leaf, &cache->item,
  5175. btrfs_item_ptr_offset(leaf, path->slots[0]),
  5176. sizeof(cache->item));
  5177. memcpy(&cache->key, &found_key, sizeof(found_key));
  5178. key.objectid = found_key.objectid + found_key.offset;
  5179. btrfs_release_path(root, path);
  5180. cache->flags = btrfs_block_group_flags(&cache->item);
  5181. ret = update_space_info(info, cache->flags, found_key.offset,
  5182. btrfs_block_group_used(&cache->item),
  5183. &space_info);
  5184. BUG_ON(ret);
  5185. cache->space_info = space_info;
  5186. down_write(&space_info->groups_sem);
  5187. list_add_tail(&cache->list, &space_info->block_groups);
  5188. up_write(&space_info->groups_sem);
  5189. ret = btrfs_add_block_group_cache(root->fs_info, cache);
  5190. BUG_ON(ret);
  5191. set_avail_alloc_bits(root->fs_info, cache->flags);
  5192. if (btrfs_chunk_readonly(root, cache->key.objectid))
  5193. set_block_group_readonly(cache);
  5194. }
  5195. ret = 0;
  5196. error:
  5197. btrfs_free_path(path);
  5198. return ret;
  5199. }
  5200. int btrfs_make_block_group(struct btrfs_trans_handle *trans,
  5201. struct btrfs_root *root, u64 bytes_used,
  5202. u64 type, u64 chunk_objectid, u64 chunk_offset,
  5203. u64 size)
  5204. {
  5205. int ret;
  5206. struct btrfs_root *extent_root;
  5207. struct btrfs_block_group_cache *cache;
  5208. extent_root = root->fs_info->extent_root;
  5209. root->fs_info->last_trans_new_blockgroup = trans->transid;
  5210. cache = kzalloc(sizeof(*cache), GFP_NOFS);
  5211. if (!cache)
  5212. return -ENOMEM;
  5213. cache->key.objectid = chunk_offset;
  5214. cache->key.offset = size;
  5215. cache->key.type = BTRFS_BLOCK_GROUP_ITEM_KEY;
  5216. atomic_set(&cache->count, 1);
  5217. spin_lock_init(&cache->lock);
  5218. mutex_init(&cache->alloc_mutex);
  5219. mutex_init(&cache->cache_mutex);
  5220. INIT_LIST_HEAD(&cache->list);
  5221. btrfs_set_block_group_used(&cache->item, bytes_used);
  5222. btrfs_set_block_group_chunk_objectid(&cache->item, chunk_objectid);
  5223. cache->flags = type;
  5224. btrfs_set_block_group_flags(&cache->item, type);
  5225. ret = update_space_info(root->fs_info, cache->flags, size, bytes_used,
  5226. &cache->space_info);
  5227. BUG_ON(ret);
  5228. down_write(&cache->space_info->groups_sem);
  5229. list_add_tail(&cache->list, &cache->space_info->block_groups);
  5230. up_write(&cache->space_info->groups_sem);
  5231. ret = btrfs_add_block_group_cache(root->fs_info, cache);
  5232. BUG_ON(ret);
  5233. ret = btrfs_insert_item(trans, extent_root, &cache->key, &cache->item,
  5234. sizeof(cache->item));
  5235. BUG_ON(ret);
  5236. finish_current_insert(trans, extent_root, 0);
  5237. ret = del_pending_extents(trans, extent_root, 0);
  5238. BUG_ON(ret);
  5239. set_avail_alloc_bits(extent_root->fs_info, type);
  5240. return 0;
  5241. }
  5242. int btrfs_remove_block_group(struct btrfs_trans_handle *trans,
  5243. struct btrfs_root *root, u64 group_start)
  5244. {
  5245. struct btrfs_path *path;
  5246. struct btrfs_block_group_cache *block_group;
  5247. struct btrfs_key key;
  5248. int ret;
  5249. root = root->fs_info->extent_root;
  5250. block_group = btrfs_lookup_block_group(root->fs_info, group_start);
  5251. BUG_ON(!block_group);
  5252. BUG_ON(!block_group->ro);
  5253. memcpy(&key, &block_group->key, sizeof(key));
  5254. path = btrfs_alloc_path();
  5255. BUG_ON(!path);
  5256. btrfs_remove_free_space_cache(block_group);
  5257. rb_erase(&block_group->cache_node,
  5258. &root->fs_info->block_group_cache_tree);
  5259. down_write(&block_group->space_info->groups_sem);
  5260. list_del(&block_group->list);
  5261. up_write(&block_group->space_info->groups_sem);
  5262. spin_lock(&block_group->space_info->lock);
  5263. block_group->space_info->total_bytes -= block_group->key.offset;
  5264. block_group->space_info->bytes_readonly -= block_group->key.offset;
  5265. spin_unlock(&block_group->space_info->lock);
  5266. block_group->space_info->full = 0;
  5267. put_block_group(block_group);
  5268. put_block_group(block_group);
  5269. ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
  5270. if (ret > 0)
  5271. ret = -EIO;
  5272. if (ret < 0)
  5273. goto out;
  5274. ret = btrfs_del_item(trans, root, path);
  5275. out:
  5276. btrfs_free_path(path);
  5277. return ret;
  5278. }