mdb.c 285 KB

123456789101112131415161718192021222324252627282930313233343536373839404142434445464748495051525354555657585960616263646566676869707172737475767778798081828384858687888990919293949596979899100101102103104105106107108109110111112113114115116117118119120121122123124125126127128129130131132133134135136137138139140141142143144145146147148149150151152153154155156157158159160161162163164165166167168169170171172173174175176177178179180181182183184185186187188189190191192193194195196197198199200201202203204205206207208209210211212213214215216217218219220221222223224225226227228229230231232233234235236237238239240241242243244245246247248249250251252253254255256257258259260261262263264265266267268269270271272273274275276277278279280281282283284285286287288289290291292293294295296297298299300301302303304305306307308309310311312313314315316317318319320321322323324325326327328329330331332333334335336337338339340341342343344345346347348349350351352353354355356357358359360361362363364365366367368369370371372373374375376377378379380381382383384385386387388389390391392393394395396397398399400401402403404405406407408409410411412413414415416417418419420421422423424425426427428429430431432433434435436437438439440441442443444445446447448449450451452453454455456457458459460461462463464465466467468469470471472473474475476477478479480481482483484485486487488489490491492493494495496497498499500501502503504505506507508509510511512513514515516517518519520521522523524525526527528529530531532533534535536537538539540541542543544545546547548549550551552553554555556557558559560561562563564565566567568569570571572573574575576577578579580581582583584585586587588589590591592593594595596597598599600601602603604605606607608609610611612613614615616617618619620621622623624625626627628629630631632633634635636637638639640641642643644645646647648649650651652653654655656657658659660661662663664665666667668669670671672673674675676677678679680681682683684685686687688689690691692693694695696697698699700701702703704705706707708709710711712713714715716717718719720721722723724725726727728729730731732733734735736737738739740741742743744745746747748749750751752753754755756757758759760761762763764765766767768769770771772773774775776777778779780781782783784785786787788789790791792793794795796797798799800801802803804805806807808809810811812813814815816817818819820821822823824825826827828829830831832833834835836837838839840841842843844845846847848849850851852853854855856857858859860861862863864865866867868869870871872873874875876877878879880881882883884885886887888889890891892893894895896897898899900901902903904905906907908909910911912913914915916917918919920921922923924925926927928929930931932933934935936937938939940941942943944945946947948949950951952953954955956957958959960961962963964965966967968969970971972973974975976977978979980981982983984985986987988989990991992993994995996997998999100010011002100310041005100610071008100910101011101210131014101510161017101810191020102110221023102410251026102710281029103010311032103310341035103610371038103910401041104210431044104510461047104810491050105110521053105410551056105710581059106010611062106310641065106610671068106910701071107210731074107510761077107810791080108110821083108410851086108710881089109010911092109310941095109610971098109911001101110211031104110511061107110811091110111111121113111411151116111711181119112011211122112311241125112611271128112911301131113211331134113511361137113811391140114111421143114411451146114711481149115011511152115311541155115611571158115911601161116211631164116511661167116811691170117111721173117411751176117711781179118011811182118311841185118611871188118911901191119211931194119511961197119811991200120112021203120412051206120712081209121012111212121312141215121612171218121912201221122212231224122512261227122812291230123112321233123412351236123712381239124012411242124312441245124612471248124912501251125212531254125512561257125812591260126112621263126412651266126712681269127012711272127312741275127612771278127912801281128212831284128512861287128812891290129112921293129412951296129712981299130013011302130313041305130613071308130913101311131213131314131513161317131813191320132113221323132413251326132713281329133013311332133313341335133613371338133913401341134213431344134513461347134813491350135113521353135413551356135713581359136013611362136313641365136613671368136913701371137213731374137513761377137813791380138113821383138413851386138713881389139013911392139313941395139613971398139914001401140214031404140514061407140814091410141114121413141414151416141714181419142014211422142314241425142614271428142914301431143214331434143514361437143814391440144114421443144414451446144714481449145014511452145314541455145614571458145914601461146214631464146514661467146814691470147114721473147414751476147714781479148014811482148314841485148614871488148914901491149214931494149514961497149814991500150115021503150415051506150715081509151015111512151315141515151615171518151915201521152215231524152515261527152815291530153115321533153415351536153715381539154015411542154315441545154615471548154915501551155215531554155515561557155815591560156115621563156415651566156715681569157015711572157315741575157615771578157915801581158215831584158515861587158815891590159115921593159415951596159715981599160016011602160316041605160616071608160916101611161216131614161516161617161816191620162116221623162416251626162716281629163016311632163316341635163616371638163916401641164216431644164516461647164816491650165116521653165416551656165716581659166016611662166316641665166616671668166916701671167216731674167516761677167816791680168116821683168416851686168716881689169016911692169316941695169616971698169917001701170217031704170517061707170817091710171117121713171417151716171717181719172017211722172317241725172617271728172917301731173217331734173517361737173817391740174117421743174417451746174717481749175017511752175317541755175617571758175917601761176217631764176517661767176817691770177117721773177417751776177717781779178017811782178317841785178617871788178917901791179217931794179517961797179817991800180118021803180418051806180718081809181018111812181318141815181618171818181918201821182218231824182518261827182818291830183118321833183418351836183718381839184018411842184318441845184618471848184918501851185218531854185518561857185818591860186118621863186418651866186718681869187018711872187318741875187618771878187918801881188218831884188518861887188818891890189118921893189418951896189718981899190019011902190319041905190619071908190919101911191219131914191519161917191819191920192119221923192419251926192719281929193019311932193319341935193619371938193919401941194219431944194519461947194819491950195119521953195419551956195719581959196019611962196319641965196619671968196919701971197219731974197519761977197819791980198119821983198419851986198719881989199019911992199319941995199619971998199920002001200220032004200520062007200820092010201120122013201420152016201720182019202020212022202320242025202620272028202920302031203220332034203520362037203820392040204120422043204420452046204720482049205020512052205320542055205620572058205920602061206220632064206520662067206820692070207120722073207420752076207720782079208020812082208320842085208620872088208920902091209220932094209520962097209820992100210121022103210421052106210721082109211021112112211321142115211621172118211921202121212221232124212521262127212821292130213121322133213421352136213721382139214021412142214321442145214621472148214921502151215221532154215521562157215821592160216121622163216421652166216721682169217021712172217321742175217621772178217921802181218221832184218521862187218821892190219121922193219421952196219721982199220022012202220322042205220622072208220922102211221222132214221522162217221822192220222122222223222422252226222722282229223022312232223322342235223622372238223922402241224222432244224522462247224822492250225122522253225422552256225722582259226022612262226322642265226622672268226922702271227222732274227522762277227822792280228122822283228422852286228722882289229022912292229322942295229622972298229923002301230223032304230523062307230823092310231123122313231423152316231723182319232023212322232323242325232623272328232923302331233223332334233523362337233823392340234123422343234423452346234723482349235023512352235323542355235623572358235923602361236223632364236523662367236823692370237123722373237423752376237723782379238023812382238323842385238623872388238923902391239223932394239523962397239823992400240124022403240424052406240724082409241024112412241324142415241624172418241924202421242224232424242524262427242824292430243124322433243424352436243724382439244024412442244324442445244624472448244924502451245224532454245524562457245824592460246124622463246424652466246724682469247024712472247324742475247624772478247924802481248224832484248524862487248824892490249124922493249424952496249724982499250025012502250325042505250625072508250925102511251225132514251525162517251825192520252125222523252425252526252725282529253025312532253325342535253625372538253925402541254225432544254525462547254825492550255125522553255425552556255725582559256025612562256325642565256625672568256925702571257225732574257525762577257825792580258125822583258425852586258725882589259025912592259325942595259625972598259926002601260226032604260526062607260826092610261126122613261426152616261726182619262026212622262326242625262626272628262926302631263226332634263526362637263826392640264126422643264426452646264726482649265026512652265326542655265626572658265926602661266226632664266526662667266826692670267126722673267426752676267726782679268026812682268326842685268626872688268926902691269226932694269526962697269826992700270127022703270427052706270727082709271027112712271327142715271627172718271927202721272227232724272527262727272827292730273127322733273427352736273727382739274027412742274327442745274627472748274927502751275227532754275527562757275827592760276127622763276427652766276727682769277027712772277327742775277627772778277927802781278227832784278527862787278827892790279127922793279427952796279727982799280028012802280328042805280628072808280928102811281228132814281528162817281828192820282128222823282428252826282728282829283028312832283328342835283628372838283928402841284228432844284528462847284828492850285128522853285428552856285728582859286028612862286328642865286628672868286928702871287228732874287528762877287828792880288128822883288428852886288728882889289028912892289328942895289628972898289929002901290229032904290529062907290829092910291129122913291429152916291729182919292029212922292329242925292629272928292929302931293229332934293529362937293829392940294129422943294429452946294729482949295029512952295329542955295629572958295929602961296229632964296529662967296829692970297129722973297429752976297729782979298029812982298329842985298629872988298929902991299229932994299529962997299829993000300130023003300430053006300730083009301030113012301330143015301630173018301930203021302230233024302530263027302830293030303130323033303430353036303730383039304030413042304330443045304630473048304930503051305230533054305530563057305830593060306130623063306430653066306730683069307030713072307330743075307630773078307930803081308230833084308530863087308830893090309130923093309430953096309730983099310031013102310331043105310631073108310931103111311231133114311531163117311831193120312131223123312431253126312731283129313031313132313331343135313631373138313931403141314231433144314531463147314831493150315131523153315431553156315731583159316031613162316331643165316631673168316931703171317231733174317531763177317831793180318131823183318431853186318731883189319031913192319331943195319631973198319932003201320232033204320532063207320832093210321132123213321432153216321732183219322032213222322332243225322632273228322932303231323232333234323532363237323832393240324132423243324432453246324732483249325032513252325332543255325632573258325932603261326232633264326532663267326832693270327132723273327432753276327732783279328032813282328332843285328632873288328932903291329232933294329532963297329832993300330133023303330433053306330733083309331033113312331333143315331633173318331933203321332233233324332533263327332833293330333133323333333433353336333733383339334033413342334333443345334633473348334933503351335233533354335533563357335833593360336133623363336433653366336733683369337033713372337333743375337633773378337933803381338233833384338533863387338833893390339133923393339433953396339733983399340034013402340334043405340634073408340934103411341234133414341534163417341834193420342134223423342434253426342734283429343034313432343334343435343634373438343934403441344234433444344534463447344834493450345134523453345434553456345734583459346034613462346334643465346634673468346934703471347234733474347534763477347834793480348134823483348434853486348734883489349034913492349334943495349634973498349935003501350235033504350535063507350835093510351135123513351435153516351735183519352035213522352335243525352635273528352935303531353235333534353535363537353835393540354135423543354435453546354735483549355035513552355335543555355635573558355935603561356235633564356535663567356835693570357135723573357435753576357735783579358035813582358335843585358635873588358935903591359235933594359535963597359835993600360136023603360436053606360736083609361036113612361336143615361636173618361936203621362236233624362536263627362836293630363136323633363436353636363736383639364036413642364336443645364636473648364936503651365236533654365536563657365836593660366136623663366436653666366736683669367036713672367336743675367636773678367936803681368236833684368536863687368836893690369136923693369436953696369736983699370037013702370337043705370637073708370937103711371237133714371537163717371837193720372137223723372437253726372737283729373037313732373337343735373637373738373937403741374237433744374537463747374837493750375137523753375437553756375737583759376037613762376337643765376637673768376937703771377237733774377537763777377837793780378137823783378437853786378737883789379037913792379337943795379637973798379938003801380238033804380538063807380838093810381138123813381438153816381738183819382038213822382338243825382638273828382938303831383238333834383538363837383838393840384138423843384438453846384738483849385038513852385338543855385638573858385938603861386238633864386538663867386838693870387138723873387438753876387738783879388038813882388338843885388638873888388938903891389238933894389538963897389838993900390139023903390439053906390739083909391039113912391339143915391639173918391939203921392239233924392539263927392839293930393139323933393439353936393739383939394039413942394339443945394639473948394939503951395239533954395539563957395839593960396139623963396439653966396739683969397039713972397339743975397639773978397939803981398239833984398539863987398839893990399139923993399439953996399739983999400040014002400340044005400640074008400940104011401240134014401540164017401840194020402140224023402440254026402740284029403040314032403340344035403640374038403940404041404240434044404540464047404840494050405140524053405440554056405740584059406040614062406340644065406640674068406940704071407240734074407540764077407840794080408140824083408440854086408740884089409040914092409340944095409640974098409941004101410241034104410541064107410841094110411141124113411441154116411741184119412041214122412341244125412641274128412941304131413241334134413541364137413841394140414141424143414441454146414741484149415041514152415341544155415641574158415941604161416241634164416541664167416841694170417141724173417441754176417741784179418041814182418341844185418641874188418941904191419241934194419541964197419841994200420142024203420442054206420742084209421042114212421342144215421642174218421942204221422242234224422542264227422842294230423142324233423442354236423742384239424042414242424342444245424642474248424942504251425242534254425542564257425842594260426142624263426442654266426742684269427042714272427342744275427642774278427942804281428242834284428542864287428842894290429142924293429442954296429742984299430043014302430343044305430643074308430943104311431243134314431543164317431843194320432143224323432443254326432743284329433043314332433343344335433643374338433943404341434243434344434543464347434843494350435143524353435443554356435743584359436043614362436343644365436643674368436943704371437243734374437543764377437843794380438143824383438443854386438743884389439043914392439343944395439643974398439944004401440244034404440544064407440844094410441144124413441444154416441744184419442044214422442344244425442644274428442944304431443244334434443544364437443844394440444144424443444444454446444744484449445044514452445344544455445644574458445944604461446244634464446544664467446844694470447144724473447444754476447744784479448044814482448344844485448644874488448944904491449244934494449544964497449844994500450145024503450445054506450745084509451045114512451345144515451645174518451945204521452245234524452545264527452845294530453145324533453445354536453745384539454045414542454345444545454645474548454945504551455245534554455545564557455845594560456145624563456445654566456745684569457045714572457345744575457645774578457945804581458245834584458545864587458845894590459145924593459445954596459745984599460046014602460346044605460646074608460946104611461246134614461546164617461846194620462146224623462446254626462746284629463046314632463346344635463646374638463946404641464246434644464546464647464846494650465146524653465446554656465746584659466046614662466346644665466646674668466946704671467246734674467546764677467846794680468146824683468446854686468746884689469046914692469346944695469646974698469947004701470247034704470547064707470847094710471147124713471447154716471747184719472047214722472347244725472647274728472947304731473247334734473547364737473847394740474147424743474447454746474747484749475047514752475347544755475647574758475947604761476247634764476547664767476847694770477147724773477447754776477747784779478047814782478347844785478647874788478947904791479247934794479547964797479847994800480148024803480448054806480748084809481048114812481348144815481648174818481948204821482248234824482548264827482848294830483148324833483448354836483748384839484048414842484348444845484648474848484948504851485248534854485548564857485848594860486148624863486448654866486748684869487048714872487348744875487648774878487948804881488248834884488548864887488848894890489148924893489448954896489748984899490049014902490349044905490649074908490949104911491249134914491549164917491849194920492149224923492449254926492749284929493049314932493349344935493649374938493949404941494249434944494549464947494849494950495149524953495449554956495749584959496049614962496349644965496649674968496949704971497249734974497549764977497849794980498149824983498449854986498749884989499049914992499349944995499649974998499950005001500250035004500550065007500850095010501150125013501450155016501750185019502050215022502350245025502650275028502950305031503250335034503550365037503850395040504150425043504450455046504750485049505050515052505350545055505650575058505950605061506250635064506550665067506850695070507150725073507450755076507750785079508050815082508350845085508650875088508950905091509250935094509550965097509850995100510151025103510451055106510751085109511051115112511351145115511651175118511951205121512251235124512551265127512851295130513151325133513451355136513751385139514051415142514351445145514651475148514951505151515251535154515551565157515851595160516151625163516451655166516751685169517051715172517351745175517651775178517951805181518251835184518551865187518851895190519151925193519451955196519751985199520052015202520352045205520652075208520952105211521252135214521552165217521852195220522152225223522452255226522752285229523052315232523352345235523652375238523952405241524252435244524552465247524852495250525152525253525452555256525752585259526052615262526352645265526652675268526952705271527252735274527552765277527852795280528152825283528452855286528752885289529052915292529352945295529652975298529953005301530253035304530553065307530853095310531153125313531453155316531753185319532053215322532353245325532653275328532953305331533253335334533553365337533853395340534153425343534453455346534753485349535053515352535353545355535653575358535953605361536253635364536553665367536853695370537153725373537453755376537753785379538053815382538353845385538653875388538953905391539253935394539553965397539853995400540154025403540454055406540754085409541054115412541354145415541654175418541954205421542254235424542554265427542854295430543154325433543454355436543754385439544054415442544354445445544654475448544954505451545254535454545554565457545854595460546154625463546454655466546754685469547054715472547354745475547654775478547954805481548254835484548554865487548854895490549154925493549454955496549754985499550055015502550355045505550655075508550955105511551255135514551555165517551855195520552155225523552455255526552755285529553055315532553355345535553655375538553955405541554255435544554555465547554855495550555155525553555455555556555755585559556055615562556355645565556655675568556955705571557255735574557555765577557855795580558155825583558455855586558755885589559055915592559355945595559655975598559956005601560256035604560556065607560856095610561156125613561456155616561756185619562056215622562356245625562656275628562956305631563256335634563556365637563856395640564156425643564456455646564756485649565056515652565356545655565656575658565956605661566256635664566556665667566856695670567156725673567456755676567756785679568056815682568356845685568656875688568956905691569256935694569556965697569856995700570157025703570457055706570757085709571057115712571357145715571657175718571957205721572257235724572557265727572857295730573157325733573457355736573757385739574057415742574357445745574657475748574957505751575257535754575557565757575857595760576157625763576457655766576757685769577057715772577357745775577657775778577957805781578257835784578557865787578857895790579157925793579457955796579757985799580058015802580358045805580658075808580958105811581258135814581558165817581858195820582158225823582458255826582758285829583058315832583358345835583658375838583958405841584258435844584558465847584858495850585158525853585458555856585758585859586058615862586358645865586658675868586958705871587258735874587558765877587858795880588158825883588458855886588758885889589058915892589358945895589658975898589959005901590259035904590559065907590859095910591159125913591459155916591759185919592059215922592359245925592659275928592959305931593259335934593559365937593859395940594159425943594459455946594759485949595059515952595359545955595659575958595959605961596259635964596559665967596859695970597159725973597459755976597759785979598059815982598359845985598659875988598959905991599259935994599559965997599859996000600160026003600460056006600760086009601060116012601360146015601660176018601960206021602260236024602560266027602860296030603160326033603460356036603760386039604060416042604360446045604660476048604960506051605260536054605560566057605860596060606160626063606460656066606760686069607060716072607360746075607660776078607960806081608260836084608560866087608860896090609160926093609460956096609760986099610061016102610361046105610661076108610961106111611261136114611561166117611861196120612161226123612461256126612761286129613061316132613361346135613661376138613961406141614261436144614561466147614861496150615161526153615461556156615761586159616061616162616361646165616661676168616961706171617261736174617561766177617861796180618161826183618461856186618761886189619061916192619361946195619661976198619962006201620262036204620562066207620862096210621162126213621462156216621762186219622062216222622362246225622662276228622962306231623262336234623562366237623862396240624162426243624462456246624762486249625062516252625362546255625662576258625962606261626262636264626562666267626862696270627162726273627462756276627762786279628062816282628362846285628662876288628962906291629262936294629562966297629862996300630163026303630463056306630763086309631063116312631363146315631663176318631963206321632263236324632563266327632863296330633163326333633463356336633763386339634063416342634363446345634663476348634963506351635263536354635563566357635863596360636163626363636463656366636763686369637063716372637363746375637663776378637963806381638263836384638563866387638863896390639163926393639463956396639763986399640064016402640364046405640664076408640964106411641264136414641564166417641864196420642164226423642464256426642764286429643064316432643364346435643664376438643964406441644264436444644564466447644864496450645164526453645464556456645764586459646064616462646364646465646664676468646964706471647264736474647564766477647864796480648164826483648464856486648764886489649064916492649364946495649664976498649965006501650265036504650565066507650865096510651165126513651465156516651765186519652065216522652365246525652665276528652965306531653265336534653565366537653865396540654165426543654465456546654765486549655065516552655365546555655665576558655965606561656265636564656565666567656865696570657165726573657465756576657765786579658065816582658365846585658665876588658965906591659265936594659565966597659865996600660166026603660466056606660766086609661066116612661366146615661666176618661966206621662266236624662566266627662866296630663166326633663466356636663766386639664066416642664366446645664666476648664966506651665266536654665566566657665866596660666166626663666466656666666766686669667066716672667366746675667666776678667966806681668266836684668566866687668866896690669166926693669466956696669766986699670067016702670367046705670667076708670967106711671267136714671567166717671867196720672167226723672467256726672767286729673067316732673367346735673667376738673967406741674267436744674567466747674867496750675167526753675467556756675767586759676067616762676367646765676667676768676967706771677267736774677567766777677867796780678167826783678467856786678767886789679067916792679367946795679667976798679968006801680268036804680568066807680868096810681168126813681468156816681768186819682068216822682368246825682668276828682968306831683268336834683568366837683868396840684168426843684468456846684768486849685068516852685368546855685668576858685968606861686268636864686568666867686868696870687168726873687468756876687768786879688068816882688368846885688668876888688968906891689268936894689568966897689868996900690169026903690469056906690769086909691069116912691369146915691669176918691969206921692269236924692569266927692869296930693169326933693469356936693769386939694069416942694369446945694669476948694969506951695269536954695569566957695869596960696169626963696469656966696769686969697069716972697369746975697669776978697969806981698269836984698569866987698869896990699169926993699469956996699769986999700070017002700370047005700670077008700970107011701270137014701570167017701870197020702170227023702470257026702770287029703070317032703370347035703670377038703970407041704270437044704570467047704870497050705170527053705470557056705770587059706070617062706370647065706670677068706970707071707270737074707570767077707870797080708170827083708470857086708770887089709070917092709370947095709670977098709971007101710271037104710571067107710871097110711171127113711471157116711771187119712071217122712371247125712671277128712971307131713271337134713571367137713871397140714171427143714471457146714771487149715071517152715371547155715671577158715971607161716271637164716571667167716871697170717171727173717471757176717771787179718071817182718371847185718671877188718971907191719271937194719571967197719871997200720172027203720472057206720772087209721072117212721372147215721672177218721972207221722272237224722572267227722872297230723172327233723472357236723772387239724072417242724372447245724672477248724972507251725272537254725572567257725872597260726172627263726472657266726772687269727072717272727372747275727672777278727972807281728272837284728572867287728872897290729172927293729472957296729772987299730073017302730373047305730673077308730973107311731273137314731573167317731873197320732173227323732473257326732773287329733073317332733373347335733673377338733973407341734273437344734573467347734873497350735173527353735473557356735773587359736073617362736373647365736673677368736973707371737273737374737573767377737873797380738173827383738473857386738773887389739073917392739373947395739673977398739974007401740274037404740574067407740874097410741174127413741474157416741774187419742074217422742374247425742674277428742974307431743274337434743574367437743874397440744174427443744474457446744774487449745074517452745374547455745674577458745974607461746274637464746574667467746874697470747174727473747474757476747774787479748074817482748374847485748674877488748974907491749274937494749574967497749874997500750175027503750475057506750775087509751075117512751375147515751675177518751975207521752275237524752575267527752875297530753175327533753475357536753775387539754075417542754375447545754675477548754975507551755275537554755575567557755875597560756175627563756475657566756775687569757075717572757375747575757675777578757975807581758275837584758575867587758875897590759175927593759475957596759775987599760076017602760376047605760676077608760976107611761276137614761576167617761876197620762176227623762476257626762776287629763076317632763376347635763676377638763976407641764276437644764576467647764876497650765176527653765476557656765776587659766076617662766376647665766676677668766976707671767276737674767576767677767876797680768176827683768476857686768776887689769076917692769376947695769676977698769977007701770277037704770577067707770877097710771177127713771477157716771777187719772077217722772377247725772677277728772977307731773277337734773577367737773877397740774177427743774477457746774777487749775077517752775377547755775677577758775977607761776277637764776577667767776877697770777177727773777477757776777777787779778077817782778377847785778677877788778977907791779277937794779577967797779877997800780178027803780478057806780778087809781078117812781378147815781678177818781978207821782278237824782578267827782878297830783178327833783478357836783778387839784078417842784378447845784678477848784978507851785278537854785578567857785878597860786178627863786478657866786778687869787078717872787378747875787678777878787978807881788278837884788578867887788878897890789178927893789478957896789778987899790079017902790379047905790679077908790979107911791279137914791579167917791879197920792179227923792479257926792779287929793079317932793379347935793679377938793979407941794279437944794579467947794879497950795179527953795479557956795779587959796079617962796379647965796679677968796979707971797279737974797579767977797879797980798179827983798479857986798779887989799079917992799379947995799679977998799980008001800280038004800580068007800880098010801180128013801480158016801780188019802080218022802380248025802680278028802980308031803280338034803580368037803880398040804180428043804480458046804780488049805080518052805380548055805680578058805980608061806280638064806580668067806880698070807180728073807480758076807780788079808080818082808380848085808680878088808980908091809280938094809580968097809880998100810181028103810481058106810781088109811081118112811381148115811681178118811981208121812281238124812581268127812881298130813181328133813481358136813781388139814081418142814381448145814681478148814981508151815281538154815581568157815881598160816181628163816481658166816781688169817081718172817381748175817681778178817981808181818281838184818581868187818881898190819181928193819481958196819781988199820082018202820382048205820682078208820982108211821282138214821582168217821882198220822182228223822482258226822782288229823082318232823382348235823682378238823982408241824282438244824582468247824882498250825182528253825482558256825782588259826082618262826382648265826682678268826982708271827282738274827582768277827882798280828182828283828482858286828782888289829082918292829382948295829682978298829983008301830283038304830583068307830883098310831183128313831483158316831783188319832083218322832383248325832683278328832983308331833283338334833583368337833883398340834183428343834483458346834783488349835083518352835383548355835683578358835983608361836283638364836583668367836883698370837183728373837483758376837783788379838083818382838383848385838683878388838983908391839283938394839583968397839883998400840184028403840484058406840784088409841084118412841384148415841684178418841984208421842284238424842584268427842884298430843184328433843484358436843784388439844084418442844384448445844684478448844984508451845284538454845584568457845884598460846184628463846484658466846784688469847084718472847384748475847684778478847984808481848284838484848584868487848884898490849184928493849484958496849784988499850085018502850385048505850685078508850985108511851285138514851585168517851885198520852185228523852485258526852785288529853085318532853385348535853685378538853985408541854285438544854585468547854885498550855185528553855485558556855785588559856085618562856385648565856685678568856985708571857285738574857585768577857885798580858185828583858485858586858785888589859085918592859385948595859685978598859986008601860286038604860586068607860886098610861186128613861486158616861786188619862086218622862386248625862686278628862986308631863286338634863586368637863886398640864186428643864486458646864786488649865086518652865386548655865686578658865986608661866286638664866586668667866886698670867186728673867486758676867786788679868086818682868386848685868686878688868986908691869286938694869586968697869886998700870187028703870487058706870787088709871087118712871387148715871687178718871987208721872287238724872587268727872887298730873187328733873487358736873787388739874087418742874387448745874687478748874987508751875287538754875587568757875887598760876187628763876487658766876787688769877087718772877387748775877687778778877987808781878287838784878587868787878887898790879187928793879487958796879787988799880088018802880388048805880688078808880988108811881288138814881588168817881888198820882188228823882488258826882788288829883088318832883388348835883688378838883988408841884288438844884588468847884888498850885188528853885488558856885788588859886088618862886388648865886688678868886988708871887288738874887588768877887888798880888188828883888488858886888788888889889088918892889388948895889688978898889989008901890289038904890589068907890889098910891189128913891489158916891789188919892089218922892389248925892689278928892989308931893289338934893589368937893889398940894189428943894489458946894789488949895089518952895389548955895689578958895989608961896289638964896589668967896889698970897189728973897489758976897789788979898089818982898389848985898689878988898989908991899289938994899589968997899889999000900190029003900490059006900790089009901090119012901390149015901690179018901990209021902290239024902590269027902890299030903190329033903490359036903790389039904090419042904390449045904690479048904990509051905290539054905590569057905890599060906190629063906490659066906790689069907090719072907390749075907690779078907990809081908290839084908590869087908890899090909190929093909490959096909790989099910091019102910391049105910691079108910991109111911291139114911591169117911891199120912191229123912491259126912791289129913091319132913391349135913691379138913991409141914291439144914591469147914891499150915191529153915491559156915791589159916091619162916391649165916691679168916991709171917291739174917591769177917891799180918191829183918491859186918791889189919091919192919391949195919691979198919992009201920292039204920592069207920892099210921192129213921492159216921792189219922092219222922392249225922692279228922992309231923292339234923592369237923892399240924192429243924492459246924792489249925092519252925392549255925692579258925992609261926292639264926592669267926892699270927192729273927492759276927792789279928092819282928392849285928692879288928992909291929292939294929592969297929892999300930193029303930493059306930793089309931093119312931393149315931693179318931993209321932293239324932593269327932893299330933193329333933493359336933793389339934093419342934393449345934693479348934993509351935293539354935593569357935893599360936193629363936493659366936793689369937093719372937393749375937693779378937993809381938293839384938593869387938893899390939193929393939493959396939793989399940094019402940394049405940694079408940994109411941294139414941594169417941894199420942194229423942494259426942794289429943094319432943394349435943694379438943994409441944294439444944594469447944894499450945194529453945494559456945794589459946094619462946394649465946694679468946994709471947294739474947594769477947894799480948194829483948494859486948794889489949094919492949394949495949694979498949995009501950295039504950595069507950895099510951195129513951495159516951795189519952095219522952395249525952695279528952995309531953295339534953595369537953895399540954195429543954495459546954795489549955095519552955395549555955695579558955995609561956295639564956595669567956895699570957195729573957495759576957795789579958095819582958395849585958695879588958995909591959295939594959595969597959895999600960196029603960496059606960796089609961096119612961396149615961696179618961996209621962296239624962596269627962896299630963196329633963496359636963796389639964096419642964396449645964696479648964996509651965296539654965596569657965896599660966196629663966496659666966796689669967096719672967396749675967696779678967996809681968296839684968596869687968896899690969196929693969496959696969796989699970097019702970397049705970697079708970997109711971297139714971597169717971897199720972197229723972497259726972797289729973097319732973397349735973697379738973997409741974297439744974597469747974897499750975197529753975497559756975797589759976097619762976397649765976697679768976997709771977297739774977597769777977897799780978197829783978497859786978797889789979097919792979397949795979697979798979998009801980298039804980598069807980898099810981198129813981498159816981798189819982098219822982398249825982698279828982998309831983298339834983598369837983898399840984198429843984498459846984798489849985098519852985398549855985698579858985998609861986298639864986598669867986898699870987198729873987498759876987798789879988098819882988398849885988698879888988998909891989298939894989598969897989898999900990199029903990499059906990799089909991099119912991399149915991699179918991999209921992299239924992599269927992899299930993199329933993499359936993799389939994099419942994399449945994699479948994999509951995299539954995599569957995899599960996199629963996499659966996799689969997099719972997399749975997699779978997999809981998299839984998599869987998899899990999199929993999499959996999799989999100001000110002100031000410005100061000710008100091001010011100121001310014100151001610017100181001910020100211002210023100241002510026100271002810029100301003110032100331003410035100361003710038100391004010041100421004310044100451004610047100481004910050100511005210053100541005510056100571005810059100601006110062100631006410065100661006710068100691007010071100721007310074100751007610077100781007910080100811008210083100841008510086100871008810089100901009110092100931009410095100961009710098100991010010101101021010310104101051010610107101081010910110101111011210113101141011510116101171011810119101201012110122101231012410125101261012710128101291013010131101321013310134101351013610137101381013910140101411014210143101441014510146101471014810149101501015110152101531015410155101561015710158101591016010161101621016310164101651016610167101681016910170101711017210173101741017510176101771017810179101801018110182101831018410185101861018710188101891019010191101921019310194101951019610197101981019910200102011020210203102041020510206102071020810209102101021110212102131021410215102161021710218102191022010221102221022310224102251022610227102281022910230102311023210233102341023510236102371023810239102401024110242102431024410245102461024710248102491025010251102521025310254102551025610257102581025910260102611026210263102641026510266102671026810269102701027110272102731027410275102761027710278102791028010281102821028310284102851028610287102881028910290102911029210293102941029510296102971029810299103001030110302103031030410305103061030710308103091031010311103121031310314103151031610317103181031910320
  1. /** @file mdb.c
  2. * @brief Lightning memory-mapped database library
  3. *
  4. * A Btree-based database management library modeled loosely on the
  5. * BerkeleyDB API, but much simplified.
  6. */
  7. /*
  8. * Copyright 2011-2021 Howard Chu, Symas Corp.
  9. * All rights reserved.
  10. *
  11. * Redistribution and use in source and binary forms, with or without
  12. * modification, are permitted only as authorized by the OpenLDAP
  13. * Public License.
  14. *
  15. * A copy of this license is available in the file LICENSE in the
  16. * top-level directory of the distribution or, alternatively, at
  17. * <http://www.OpenLDAP.org/license.html>.
  18. *
  19. * This code is derived from btree.c written by Martin Hedenfalk.
  20. *
  21. * Copyright (c) 2009, 2010 Martin Hedenfalk <martin@bzero.se>
  22. *
  23. * Permission to use, copy, modify, and distribute this software for any
  24. * purpose with or without fee is hereby granted, provided that the above
  25. * copyright notice and this permission notice appear in all copies.
  26. *
  27. * THE SOFTWARE IS PROVIDED "AS IS" AND THE AUTHOR DISCLAIMS ALL WARRANTIES
  28. * WITH REGARD TO THIS SOFTWARE INCLUDING ALL IMPLIED WARRANTIES OF
  29. * MERCHANTABILITY AND FITNESS. IN NO EVENT SHALL THE AUTHOR BE LIABLE FOR
  30. * ANY SPECIAL, DIRECT, INDIRECT, OR CONSEQUENTIAL DAMAGES OR ANY DAMAGES
  31. * WHATSOEVER RESULTING FROM LOSS OF USE, DATA OR PROFITS, WHETHER IN AN
  32. * ACTION OF CONTRACT, NEGLIGENCE OR OTHER TORTIOUS ACTION, ARISING OUT OF
  33. * OR IN CONNECTION WITH THE USE OR PERFORMANCE OF THIS SOFTWARE.
  34. */
  35. #ifndef _GNU_SOURCE
  36. #define _GNU_SOURCE 1
  37. #endif
  38. #if defined(__WIN64__)
  39. #define _FILE_OFFSET_BITS 64
  40. #endif
  41. #ifdef _WIN32
  42. #include <malloc.h>
  43. #include <windows.h>
  44. #include <wchar.h> /* get wcscpy() */
  45. /** getpid() returns int; MinGW defines pid_t but MinGW64 typedefs it
  46. * as int64 which is wrong. MSVC doesn't define it at all, so just
  47. * don't use it.
  48. */
  49. #define MDB_PID_T int
  50. #define MDB_THR_T DWORD
  51. #include <sys/types.h>
  52. #include <sys/stat.h>
  53. #ifdef __GNUC__
  54. # include <sys/param.h>
  55. #else
  56. # define LITTLE_ENDIAN 1234
  57. # define BIG_ENDIAN 4321
  58. # define BYTE_ORDER LITTLE_ENDIAN
  59. # ifndef SSIZE_MAX
  60. # define SSIZE_MAX INT_MAX
  61. # endif
  62. #endif
  63. #else
  64. #include <sys/types.h>
  65. #include <sys/stat.h>
  66. #define MDB_PID_T pid_t
  67. #define MDB_THR_T pthread_t
  68. #include <sys/param.h>
  69. #include <sys/uio.h>
  70. #include <sys/mman.h>
  71. #ifdef HAVE_SYS_FILE_H
  72. #include <sys/file.h>
  73. #endif
  74. #include <fcntl.h>
  75. #endif
  76. #if defined(__mips) && defined(__linux)
  77. /* MIPS has cache coherency issues, requires explicit cache control */
  78. #include <asm/cachectl.h>
  79. extern int cacheflush(char *addr, int nbytes, int cache);
  80. #define CACHEFLUSH(addr, bytes, cache) cacheflush(addr, bytes, cache)
  81. #else
  82. #define CACHEFLUSH(addr, bytes, cache)
  83. #endif
  84. #if defined(__linux) && !defined(MDB_FDATASYNC_WORKS)
  85. /** fdatasync is broken on ext3/ext4fs on older kernels, see
  86. * description in #mdb_env_open2 comments. You can safely
  87. * define MDB_FDATASYNC_WORKS if this code will only be run
  88. * on kernels 3.6 and newer.
  89. */
  90. #define BROKEN_FDATASYNC
  91. #endif
  92. #include <errno.h>
  93. #include <limits.h>
  94. #include <stddef.h>
  95. #include <inttypes.h>
  96. #include <stdio.h>
  97. #include <stdlib.h>
  98. #include <string.h>
  99. #include <time.h>
  100. #ifdef _MSC_VER
  101. #include <io.h>
  102. typedef SSIZE_T ssize_t;
  103. #else
  104. #include <unistd.h>
  105. #endif
  106. #if defined(__sun) || defined(ANDROID)
  107. /* Most platforms have posix_memalign, older may only have memalign */
  108. #define HAVE_MEMALIGN 1
  109. #include <malloc.h>
  110. /* On Solaris, we need the POSIX sigwait function */
  111. #if defined (__sun)
  112. # define _POSIX_PTHREAD_SEMANTICS 1
  113. #endif
  114. #endif
  115. #if !(defined(BYTE_ORDER) || defined(__BYTE_ORDER))
  116. #include <netinet/in.h>
  117. #include <resolv.h> /* defines BYTE_ORDER on HPUX and Solaris */
  118. #endif
  119. #if defined(__FreeBSD__) && defined(__FreeBSD_version) && __FreeBSD_version >= 1100110
  120. # define MDB_USE_POSIX_MUTEX 1
  121. # define MDB_USE_ROBUST 1
  122. #elif defined(__APPLE__) || defined (BSD) || defined(__FreeBSD_kernel__)
  123. # define MDB_USE_POSIX_SEM 1
  124. # define MDB_FDATASYNC fsync
  125. #elif defined(ANDROID)
  126. # define MDB_FDATASYNC fsync
  127. #endif
  128. #ifndef _WIN32
  129. #include <pthread.h>
  130. #include <signal.h>
  131. #ifdef MDB_USE_POSIX_SEM
  132. # define MDB_USE_HASH 1
  133. #include <semaphore.h>
  134. #else
  135. #define MDB_USE_POSIX_MUTEX 1
  136. #endif
  137. #endif
  138. #if defined(_WIN32) + defined(MDB_USE_POSIX_SEM) \
  139. + defined(MDB_USE_POSIX_MUTEX) != 1
  140. # error "Ambiguous shared-lock implementation"
  141. #endif
  142. #ifdef USE_VALGRIND
  143. #include <valgrind/memcheck.h>
  144. #define VGMEMP_CREATE(h,r,z) VALGRIND_CREATE_MEMPOOL(h,r,z)
  145. #define VGMEMP_ALLOC(h,a,s) VALGRIND_MEMPOOL_ALLOC(h,a,s)
  146. #define VGMEMP_FREE(h,a) VALGRIND_MEMPOOL_FREE(h,a)
  147. #define VGMEMP_DESTROY(h) VALGRIND_DESTROY_MEMPOOL(h)
  148. #define VGMEMP_DEFINED(a,s) VALGRIND_MAKE_MEM_DEFINED(a,s)
  149. #else
  150. #define VGMEMP_CREATE(h,r,z)
  151. #define VGMEMP_ALLOC(h,a,s)
  152. #define VGMEMP_FREE(h,a)
  153. #define VGMEMP_DESTROY(h)
  154. #define VGMEMP_DEFINED(a,s)
  155. #endif
  156. #ifndef BYTE_ORDER
  157. # if (defined(_LITTLE_ENDIAN) || defined(_BIG_ENDIAN)) && !(defined(_LITTLE_ENDIAN) && defined(_BIG_ENDIAN))
  158. /* Solaris just defines one or the other */
  159. # define LITTLE_ENDIAN 1234
  160. # define BIG_ENDIAN 4321
  161. # ifdef _LITTLE_ENDIAN
  162. # define BYTE_ORDER LITTLE_ENDIAN
  163. # else
  164. # define BYTE_ORDER BIG_ENDIAN
  165. # endif
  166. # else
  167. # define BYTE_ORDER __BYTE_ORDER
  168. # endif
  169. #endif
  170. #ifndef LITTLE_ENDIAN
  171. #define LITTLE_ENDIAN __LITTLE_ENDIAN
  172. #endif
  173. #ifndef BIG_ENDIAN
  174. #define BIG_ENDIAN __BIG_ENDIAN
  175. #endif
  176. #if defined(__i386) || defined(__x86_64) || defined(_M_IX86)
  177. #define MISALIGNED_OK 1
  178. #endif
  179. #include "lmdb.h"
  180. #include "midl.h"
  181. #if (BYTE_ORDER == LITTLE_ENDIAN) == (BYTE_ORDER == BIG_ENDIAN)
  182. # error "Unknown or unsupported endianness (BYTE_ORDER)"
  183. #elif (-6 & 5) || CHAR_BIT != 8 || UINT_MAX < 0xffffffff || ULONG_MAX % 0xFFFF
  184. # error "Two's complement, reasonably sized integer types, please"
  185. #endif
  186. #ifdef __GNUC__
  187. /** Put infrequently used env functions in separate section */
  188. # ifdef __APPLE__
  189. # define ESECT __attribute__ ((section("__TEXT,text_env")))
  190. # else
  191. # define ESECT __attribute__ ((section("text_env")))
  192. # endif
  193. #else
  194. #define ESECT
  195. #endif
  196. #ifdef _WIN32
  197. #define CALL_CONV WINAPI
  198. #else
  199. #define CALL_CONV
  200. #endif
  201. /** @defgroup internal LMDB Internals
  202. * @{
  203. */
  204. /** @defgroup compat Compatibility Macros
  205. * A bunch of macros to minimize the amount of platform-specific ifdefs
  206. * needed throughout the rest of the code. When the features this library
  207. * needs are similar enough to POSIX to be hidden in a one-or-two line
  208. * replacement, this macro approach is used.
  209. * @{
  210. */
  211. /** Features under development */
  212. #ifndef MDB_DEVEL
  213. #define MDB_DEVEL 0
  214. #endif
  215. /** Wrapper around __func__, which is a C99 feature */
  216. #if __STDC_VERSION__ >= 199901L
  217. # define mdb_func_ __func__
  218. #elif __GNUC__ >= 2 || _MSC_VER >= 1300
  219. # define mdb_func_ __FUNCTION__
  220. #else
  221. /* If a debug message says <mdb_unknown>(), update the #if statements above */
  222. # define mdb_func_ "<mdb_unknown>"
  223. #endif
  224. /* Internal error codes, not exposed outside liblmdb */
  225. #define MDB_NO_ROOT (MDB_LAST_ERRCODE + 10)
  226. #ifdef _WIN32
  227. #define MDB_OWNERDEAD ((int) WAIT_ABANDONED)
  228. #elif defined(MDB_USE_POSIX_MUTEX) && defined(EOWNERDEAD)
  229. #define MDB_OWNERDEAD EOWNERDEAD /**< #LOCK_MUTEX0() result if dead owner */
  230. #endif
  231. #ifdef __GLIBC__
  232. #define GLIBC_VER ((__GLIBC__ << 16 )| __GLIBC_MINOR__)
  233. #endif
  234. /** Some platforms define the EOWNERDEAD error code
  235. * even though they don't support Robust Mutexes.
  236. * Compile with -DMDB_USE_ROBUST=0, or use some other
  237. * mechanism like -DMDB_USE_POSIX_SEM instead of
  238. * -DMDB_USE_POSIX_MUTEX.
  239. * (Posix semaphores are not robust.)
  240. */
  241. #ifndef MDB_USE_ROBUST
  242. /* Android currently lacks Robust Mutex support. So does glibc < 2.4. */
  243. # if defined(MDB_USE_POSIX_MUTEX) && (defined(ANDROID) || \
  244. (defined(__GLIBC__) && GLIBC_VER < 0x020004))
  245. # define MDB_USE_ROBUST 0
  246. # else
  247. # define MDB_USE_ROBUST 1
  248. # endif
  249. #endif /* !MDB_USE_ROBUST */
  250. #if defined(MDB_USE_POSIX_MUTEX) && (MDB_USE_ROBUST)
  251. /* glibc < 2.12 only provided _np API */
  252. # if (defined(__GLIBC__) && GLIBC_VER < 0x02000c) || \
  253. (defined(PTHREAD_MUTEX_ROBUST_NP) && !defined(PTHREAD_MUTEX_ROBUST))
  254. # define PTHREAD_MUTEX_ROBUST PTHREAD_MUTEX_ROBUST_NP
  255. # define pthread_mutexattr_setrobust(attr, flag) pthread_mutexattr_setrobust_np(attr, flag)
  256. # define pthread_mutex_consistent(mutex) pthread_mutex_consistent_np(mutex)
  257. # endif
  258. #endif /* MDB_USE_POSIX_MUTEX && MDB_USE_ROBUST */
  259. #if defined(MDB_OWNERDEAD) && (MDB_USE_ROBUST)
  260. #define MDB_ROBUST_SUPPORTED 1
  261. #endif
  262. #ifdef _WIN32
  263. #define MDB_USE_HASH 1
  264. #define MDB_PIDLOCK 0
  265. #define THREAD_RET DWORD
  266. #define pthread_t HANDLE
  267. #define pthread_mutex_t HANDLE
  268. #define pthread_cond_t HANDLE
  269. typedef HANDLE mdb_mutex_t, mdb_mutexref_t;
  270. #define pthread_key_t DWORD
  271. #define pthread_self() GetCurrentThreadId()
  272. #define pthread_key_create(x,y) \
  273. ((*(x) = TlsAlloc()) == TLS_OUT_OF_INDEXES ? ErrCode() : 0)
  274. #define pthread_key_delete(x) TlsFree(x)
  275. #define pthread_getspecific(x) TlsGetValue(x)
  276. #define pthread_setspecific(x,y) (TlsSetValue(x,y) ? 0 : ErrCode())
  277. #define pthread_mutex_unlock(x) ReleaseMutex(*x)
  278. #define pthread_mutex_lock(x) WaitForSingleObject(*x, INFINITE)
  279. #define pthread_cond_signal(x) SetEvent(*x)
  280. #define pthread_cond_wait(cond,mutex) do{SignalObjectAndWait(*mutex, *cond, INFINITE, FALSE); WaitForSingleObject(*mutex, INFINITE);}while(0)
  281. #define THREAD_CREATE(thr,start,arg) \
  282. (((thr) = CreateThread(NULL, 0, start, arg, 0, NULL)) ? 0 : ErrCode())
  283. #define THREAD_FINISH(thr) \
  284. (WaitForSingleObject(thr, INFINITE) ? ErrCode() : 0)
  285. #define LOCK_MUTEX0(mutex) WaitForSingleObject(mutex, INFINITE)
  286. #define UNLOCK_MUTEX(mutex) ReleaseMutex(mutex)
  287. #define mdb_mutex_consistent(mutex) 0
  288. #define getpid() GetCurrentProcessId()
  289. #define MDB_FDATASYNC(fd) (!FlushFileBuffers(fd))
  290. #define MDB_MSYNC(addr,len,flags) (!FlushViewOfFile(addr,len))
  291. #define ErrCode() GetLastError()
  292. #define GET_PAGESIZE(x) {SYSTEM_INFO si; GetSystemInfo(&si); (x) = si.dwPageSize;}
  293. #define close(fd) (CloseHandle(fd) ? 0 : -1)
  294. #define munmap(ptr,len) UnmapViewOfFile(ptr)
  295. #ifdef PROCESS_QUERY_LIMITED_INFORMATION
  296. #define MDB_PROCESS_QUERY_LIMITED_INFORMATION PROCESS_QUERY_LIMITED_INFORMATION
  297. #else
  298. #define MDB_PROCESS_QUERY_LIMITED_INFORMATION 0x1000
  299. #endif
  300. #define Z "I"
  301. #else
  302. #define THREAD_RET void *
  303. #define THREAD_CREATE(thr,start,arg) pthread_create(&thr,NULL,start,arg)
  304. #define THREAD_FINISH(thr) pthread_join(thr,NULL)
  305. #define Z "z" /**< printf format modifier for size_t */
  306. /** For MDB_LOCK_FORMAT: True if readers take a pid lock in the lockfile */
  307. #define MDB_PIDLOCK 1
  308. #ifdef MDB_USE_POSIX_SEM
  309. typedef sem_t *mdb_mutex_t, *mdb_mutexref_t;
  310. #define LOCK_MUTEX0(mutex) mdb_sem_wait(mutex)
  311. #define UNLOCK_MUTEX(mutex) sem_post(mutex)
  312. static int
  313. mdb_sem_wait(sem_t *sem)
  314. {
  315. int rc;
  316. while ((rc = sem_wait(sem)) && (rc = errno) == EINTR) ;
  317. return rc;
  318. }
  319. #else /* MDB_USE_POSIX_MUTEX: */
  320. /** Shared mutex/semaphore as the original is stored.
  321. *
  322. * Not for copies. Instead it can be assigned to an #mdb_mutexref_t.
  323. * When mdb_mutexref_t is a pointer and mdb_mutex_t is not, then it
  324. * is array[size 1] so it can be assigned to the pointer.
  325. */
  326. typedef pthread_mutex_t mdb_mutex_t[1];
  327. /** Reference to an #mdb_mutex_t */
  328. typedef pthread_mutex_t *mdb_mutexref_t;
  329. /** Lock the reader or writer mutex.
  330. * Returns 0 or a code to give #mdb_mutex_failed(), as in #LOCK_MUTEX().
  331. */
  332. #define LOCK_MUTEX0(mutex) pthread_mutex_lock(mutex)
  333. /** Unlock the reader or writer mutex.
  334. */
  335. #define UNLOCK_MUTEX(mutex) pthread_mutex_unlock(mutex)
  336. /** Mark mutex-protected data as repaired, after death of previous owner.
  337. */
  338. #define mdb_mutex_consistent(mutex) pthread_mutex_consistent(mutex)
  339. #endif /* MDB_USE_POSIX_SEM */
  340. /** Get the error code for the last failed system function.
  341. */
  342. #define ErrCode() errno
  343. /** An abstraction for a file handle.
  344. * On POSIX systems file handles are small integers. On Windows
  345. * they're opaque pointers.
  346. */
  347. #define HANDLE int
  348. /** A value for an invalid file handle.
  349. * Mainly used to initialize file variables and signify that they are
  350. * unused.
  351. */
  352. #define INVALID_HANDLE_VALUE (-1)
  353. /** Get the size of a memory page for the system.
  354. * This is the basic size that the platform's memory manager uses, and is
  355. * fundamental to the use of memory-mapped files.
  356. */
  357. #define GET_PAGESIZE(x) ((x) = sysconf(_SC_PAGE_SIZE))
  358. #endif
  359. #if defined(_WIN32) || defined(MDB_USE_POSIX_SEM)
  360. #define MNAME_LEN 32
  361. #else
  362. #define MNAME_LEN (sizeof(pthread_mutex_t))
  363. #endif
  364. /** @} */
  365. #ifdef MDB_ROBUST_SUPPORTED
  366. /** Lock mutex, handle any error, set rc = result.
  367. * Return 0 on success, nonzero (not rc) on error.
  368. */
  369. #define LOCK_MUTEX(rc, env, mutex) \
  370. (((rc) = LOCK_MUTEX0(mutex)) && \
  371. ((rc) = mdb_mutex_failed(env, mutex, rc)))
  372. static int mdb_mutex_failed(MDB_env *env, mdb_mutexref_t mutex, int rc);
  373. #else
  374. #define LOCK_MUTEX(rc, env, mutex) ((rc) = LOCK_MUTEX0(mutex))
  375. #define mdb_mutex_failed(env, mutex, rc) (rc)
  376. #endif
  377. #ifndef _WIN32
  378. /** A flag for opening a file and requesting synchronous data writes.
  379. * This is only used when writing a meta page. It's not strictly needed;
  380. * we could just do a normal write and then immediately perform a flush.
  381. * But if this flag is available it saves us an extra system call.
  382. *
  383. * @note If O_DSYNC is undefined but exists in /usr/include,
  384. * preferably set some compiler flag to get the definition.
  385. */
  386. #ifndef MDB_DSYNC
  387. # ifdef O_DSYNC
  388. # define MDB_DSYNC O_DSYNC
  389. # else
  390. # define MDB_DSYNC O_SYNC
  391. # endif
  392. #endif
  393. #endif
  394. /** Function for flushing the data of a file. Define this to fsync
  395. * if fdatasync() is not supported.
  396. */
  397. #ifndef MDB_FDATASYNC
  398. # define MDB_FDATASYNC fdatasync
  399. #endif
  400. #ifndef MDB_MSYNC
  401. # define MDB_MSYNC(addr,len,flags) msync(addr,len,flags)
  402. #endif
  403. #ifndef MS_SYNC
  404. #define MS_SYNC 1
  405. #endif
  406. #ifndef MS_ASYNC
  407. #define MS_ASYNC 0
  408. #endif
  409. /** A page number in the database.
  410. * Note that 64 bit page numbers are overkill, since pages themselves
  411. * already represent 12-13 bits of addressable memory, and the OS will
  412. * always limit applications to a maximum of 63 bits of address space.
  413. *
  414. * @note In the #MDB_node structure, we only store 48 bits of this value,
  415. * which thus limits us to only 60 bits of addressable data.
  416. */
  417. typedef MDB_ID pgno_t;
  418. /** A transaction ID.
  419. * See struct MDB_txn.mt_txnid for details.
  420. */
  421. typedef MDB_ID txnid_t;
  422. /** @defgroup debug Debug Macros
  423. * @{
  424. */
  425. #ifndef MDB_DEBUG
  426. /** Enable debug output. Needs variable argument macros (a C99 feature).
  427. * Set this to 1 for copious tracing. Set to 2 to add dumps of all IDLs
  428. * read from and written to the database (used for free space management).
  429. */
  430. #define MDB_DEBUG 0
  431. #endif
  432. #if MDB_DEBUG
  433. static int mdb_debug;
  434. static txnid_t mdb_debug_start;
  435. /** Print a debug message with printf formatting.
  436. * Requires double parenthesis around 2 or more args.
  437. */
  438. # define DPRINTF(args) ((void) ((mdb_debug) && DPRINTF0 args))
  439. # define DPRINTF0(fmt, ...) \
  440. fprintf(stderr, "%s:%d " fmt "\n", mdb_func_, __LINE__, __VA_ARGS__)
  441. #else
  442. # define DPRINTF(args) ((void) 0)
  443. #endif
  444. /** Print a debug string.
  445. * The string is printed literally, with no format processing.
  446. */
  447. #define DPUTS(arg) DPRINTF(("%s", arg))
  448. /** Debugging output value of a cursor DBI: Negative in a sub-cursor. */
  449. #define DDBI(mc) \
  450. (((mc)->mc_flags & C_SUB) ? -(int)(mc)->mc_dbi : (int)(mc)->mc_dbi)
  451. /** @} */
  452. /** @brief The maximum size of a database page.
  453. *
  454. * It is 32k or 64k, since value-PAGEBASE must fit in
  455. * #MDB_page.%mp_upper.
  456. *
  457. * LMDB will use database pages < OS pages if needed.
  458. * That causes more I/O in write transactions: The OS must
  459. * know (read) the whole page before writing a partial page.
  460. *
  461. * Note that we don't currently support Huge pages. On Linux,
  462. * regular data files cannot use Huge pages, and in general
  463. * Huge pages aren't actually pageable. We rely on the OS
  464. * demand-pager to read our data and page it out when memory
  465. * pressure from other processes is high. So until OSs have
  466. * actual paging support for Huge pages, they're not viable.
  467. */
  468. #define MAX_PAGESIZE (PAGEBASE ? 0x10000 : 0x8000)
  469. /** The minimum number of keys required in a database page.
  470. * Setting this to a larger value will place a smaller bound on the
  471. * maximum size of a data item. Data items larger than this size will
  472. * be pushed into overflow pages instead of being stored directly in
  473. * the B-tree node. This value used to default to 4. With a page size
  474. * of 4096 bytes that meant that any item larger than 1024 bytes would
  475. * go into an overflow page. That also meant that on average 2-3KB of
  476. * each overflow page was wasted space. The value cannot be lower than
  477. * 2 because then there would no longer be a tree structure. With this
  478. * value, items larger than 2KB will go into overflow pages, and on
  479. * average only 1KB will be wasted.
  480. */
  481. #define MDB_MINKEYS 2
  482. /** A stamp that identifies a file as an LMDB file.
  483. * There's nothing special about this value other than that it is easily
  484. * recognizable, and it will reflect any byte order mismatches.
  485. */
  486. #define MDB_MAGIC 0xBEEFC0DE
  487. /** The version number for a database's datafile format. */
  488. #define MDB_DATA_VERSION ((MDB_DEVEL) ? 999 : 1)
  489. /** The version number for a database's lockfile format. */
  490. #define MDB_LOCK_VERSION 1
  491. /** @brief The max size of a key we can write, or 0 for computed max.
  492. *
  493. * This macro should normally be left alone or set to 0.
  494. * Note that a database with big keys or dupsort data cannot be
  495. * reliably modified by a liblmdb which uses a smaller max.
  496. * The default is 511 for backwards compat, or 0 when #MDB_DEVEL.
  497. *
  498. * Other values are allowed, for backwards compat. However:
  499. * A value bigger than the computed max can break if you do not
  500. * know what you are doing, and liblmdb <= 0.9.10 can break when
  501. * modifying a DB with keys/dupsort data bigger than its max.
  502. *
  503. * Data items in an #MDB_DUPSORT database are also limited to
  504. * this size, since they're actually keys of a sub-DB. Keys and
  505. * #MDB_DUPSORT data items must fit on a node in a regular page.
  506. */
  507. #ifndef MDB_MAXKEYSIZE
  508. #define MDB_MAXKEYSIZE ((MDB_DEVEL) ? 0 : 511)
  509. #endif
  510. /** The maximum size of a key we can write to the environment. */
  511. #if MDB_MAXKEYSIZE
  512. #define ENV_MAXKEY(env) (MDB_MAXKEYSIZE)
  513. #else
  514. #define ENV_MAXKEY(env) ((env)->me_maxkey)
  515. #endif
  516. /** @brief The maximum size of a data item.
  517. *
  518. * We only store a 32 bit value for node sizes.
  519. */
  520. #define MAXDATASIZE 0xffffffffUL
  521. #if MDB_DEBUG
  522. /** Key size which fits in a #DKBUF.
  523. * @ingroup debug
  524. */
  525. #define DKBUF_MAXKEYSIZE ((MDB_MAXKEYSIZE) > 0 ? (MDB_MAXKEYSIZE) : 511)
  526. /** A key buffer.
  527. * @ingroup debug
  528. * This is used for printing a hex dump of a key's contents.
  529. */
  530. #define DKBUF char kbuf[DKBUF_MAXKEYSIZE*2+1]
  531. /** Display a key in hex.
  532. * @ingroup debug
  533. * Invoke a function to display a key in hex.
  534. */
  535. #define DKEY(x) mdb_dkey(x, kbuf)
  536. #else
  537. #define DKBUF
  538. #define DKEY(x) 0
  539. #endif
  540. /** An invalid page number.
  541. * Mainly used to denote an empty tree.
  542. */
  543. #define P_INVALID (~(pgno_t)0)
  544. /** Test if the flags \b f are set in a flag word \b w. */
  545. #define F_ISSET(w, f) (((w) & (f)) == (f))
  546. /** Round \b n up to an even number. */
  547. #define EVEN(n) (((n) + 1U) & -2) /* sign-extending -2 to match n+1U */
  548. /** Used for offsets within a single page.
  549. * Since memory pages are typically 4 or 8KB in size, 12-13 bits,
  550. * this is plenty.
  551. */
  552. typedef uint16_t indx_t;
  553. /** Default size of memory map.
  554. * This is certainly too small for any actual applications. Apps should always set
  555. * the size explicitly using #mdb_env_set_mapsize().
  556. */
  557. #define DEFAULT_MAPSIZE 1048576
  558. /** @defgroup readers Reader Lock Table
  559. * Readers don't acquire any locks for their data access. Instead, they
  560. * simply record their transaction ID in the reader table. The reader
  561. * mutex is needed just to find an empty slot in the reader table. The
  562. * slot's address is saved in thread-specific data so that subsequent read
  563. * transactions started by the same thread need no further locking to proceed.
  564. *
  565. * If #MDB_NOTLS is set, the slot address is not saved in thread-specific data.
  566. *
  567. * No reader table is used if the database is on a read-only filesystem, or
  568. * if #MDB_NOLOCK is set.
  569. *
  570. * Since the database uses multi-version concurrency control, readers don't
  571. * actually need any locking. This table is used to keep track of which
  572. * readers are using data from which old transactions, so that we'll know
  573. * when a particular old transaction is no longer in use. Old transactions
  574. * that have discarded any data pages can then have those pages reclaimed
  575. * for use by a later write transaction.
  576. *
  577. * The lock table is constructed such that reader slots are aligned with the
  578. * processor's cache line size. Any slot is only ever used by one thread.
  579. * This alignment guarantees that there will be no contention or cache
  580. * thrashing as threads update their own slot info, and also eliminates
  581. * any need for locking when accessing a slot.
  582. *
  583. * A writer thread will scan every slot in the table to determine the oldest
  584. * outstanding reader transaction. Any freed pages older than this will be
  585. * reclaimed by the writer. The writer doesn't use any locks when scanning
  586. * this table. This means that there's no guarantee that the writer will
  587. * see the most up-to-date reader info, but that's not required for correct
  588. * operation - all we need is to know the upper bound on the oldest reader,
  589. * we don't care at all about the newest reader. So the only consequence of
  590. * reading stale information here is that old pages might hang around a
  591. * while longer before being reclaimed. That's actually good anyway, because
  592. * the longer we delay reclaiming old pages, the more likely it is that a
  593. * string of contiguous pages can be found after coalescing old pages from
  594. * many old transactions together.
  595. * @{
  596. */
  597. /** Number of slots in the reader table.
  598. * This value was chosen somewhat arbitrarily. 126 readers plus a
  599. * couple mutexes fit exactly into 8KB on my development machine.
  600. * Applications should set the table size using #mdb_env_set_maxreaders().
  601. */
  602. #define DEFAULT_READERS 126
  603. /** The size of a CPU cache line in bytes. We want our lock structures
  604. * aligned to this size to avoid false cache line sharing in the
  605. * lock table.
  606. * This value works for most CPUs. For Itanium this should be 128.
  607. */
  608. #ifndef CACHELINE
  609. #define CACHELINE 64
  610. #endif
  611. /** The information we store in a single slot of the reader table.
  612. * In addition to a transaction ID, we also record the process and
  613. * thread ID that owns a slot, so that we can detect stale information,
  614. * e.g. threads or processes that went away without cleaning up.
  615. * @note We currently don't check for stale records. We simply re-init
  616. * the table when we know that we're the only process opening the
  617. * lock file.
  618. */
  619. typedef struct MDB_rxbody {
  620. /** Current Transaction ID when this transaction began, or (txnid_t)-1.
  621. * Multiple readers that start at the same time will probably have the
  622. * same ID here. Again, it's not important to exclude them from
  623. * anything; all we need to know is which version of the DB they
  624. * started from so we can avoid overwriting any data used in that
  625. * particular version.
  626. */
  627. volatile txnid_t mrb_txnid;
  628. /** The process ID of the process owning this reader txn. */
  629. volatile MDB_PID_T mrb_pid;
  630. /** The thread ID of the thread owning this txn. */
  631. volatile MDB_THR_T mrb_tid;
  632. } MDB_rxbody;
  633. /** The actual reader record, with cacheline padding. */
  634. typedef struct MDB_reader {
  635. union {
  636. MDB_rxbody mrx;
  637. /** shorthand for mrb_txnid */
  638. #define mr_txnid mru.mrx.mrb_txnid
  639. #define mr_pid mru.mrx.mrb_pid
  640. #define mr_tid mru.mrx.mrb_tid
  641. /** cache line alignment */
  642. char pad[(sizeof(MDB_rxbody)+CACHELINE-1) & ~(CACHELINE-1)];
  643. } mru;
  644. } MDB_reader;
  645. /** The header for the reader table.
  646. * The table resides in a memory-mapped file. (This is a different file
  647. * than is used for the main database.)
  648. *
  649. * For POSIX the actual mutexes reside in the shared memory of this
  650. * mapped file. On Windows, mutexes are named objects allocated by the
  651. * kernel; we store the mutex names in this mapped file so that other
  652. * processes can grab them. This same approach is also used on
  653. * MacOSX/Darwin (using named semaphores) since MacOSX doesn't support
  654. * process-shared POSIX mutexes. For these cases where a named object
  655. * is used, the object name is derived from a 64 bit FNV hash of the
  656. * environment pathname. As such, naming collisions are extremely
  657. * unlikely. If a collision occurs, the results are unpredictable.
  658. */
  659. typedef struct MDB_txbody {
  660. /** Stamp identifying this as an LMDB file. It must be set
  661. * to #MDB_MAGIC. */
  662. uint32_t mtb_magic;
  663. /** Format of this lock file. Must be set to #MDB_LOCK_FORMAT. */
  664. uint32_t mtb_format;
  665. #if defined(_WIN32) || defined(MDB_USE_POSIX_SEM)
  666. char mtb_rmname[MNAME_LEN];
  667. #else
  668. /** Mutex protecting access to this table.
  669. * This is the reader table lock used with LOCK_MUTEX().
  670. */
  671. mdb_mutex_t mtb_rmutex;
  672. #endif
  673. /** The ID of the last transaction committed to the database.
  674. * This is recorded here only for convenience; the value can always
  675. * be determined by reading the main database meta pages.
  676. */
  677. volatile txnid_t mtb_txnid;
  678. /** The number of slots that have been used in the reader table.
  679. * This always records the maximum count, it is not decremented
  680. * when readers release their slots.
  681. */
  682. volatile unsigned mtb_numreaders;
  683. } MDB_txbody;
  684. /** The actual reader table definition. */
  685. typedef struct MDB_txninfo {
  686. union {
  687. MDB_txbody mtb;
  688. #define mti_magic mt1.mtb.mtb_magic
  689. #define mti_format mt1.mtb.mtb_format
  690. #define mti_rmutex mt1.mtb.mtb_rmutex
  691. #define mti_rmname mt1.mtb.mtb_rmname
  692. #define mti_txnid mt1.mtb.mtb_txnid
  693. #define mti_numreaders mt1.mtb.mtb_numreaders
  694. char pad[(sizeof(MDB_txbody)+CACHELINE-1) & ~(CACHELINE-1)];
  695. } mt1;
  696. union {
  697. #if defined(_WIN32) || defined(MDB_USE_POSIX_SEM)
  698. char mt2_wmname[MNAME_LEN];
  699. #define mti_wmname mt2.mt2_wmname
  700. #else
  701. mdb_mutex_t mt2_wmutex;
  702. #define mti_wmutex mt2.mt2_wmutex
  703. #endif
  704. char pad[(MNAME_LEN+CACHELINE-1) & ~(CACHELINE-1)];
  705. } mt2;
  706. MDB_reader mti_readers[1];
  707. } MDB_txninfo;
  708. /** Lockfile format signature: version, features and field layout */
  709. #define MDB_LOCK_FORMAT \
  710. ((uint32_t) \
  711. ((MDB_LOCK_VERSION) \
  712. /* Flags which describe functionality */ \
  713. + (((MDB_PIDLOCK) != 0) << 16)))
  714. /** @} */
  715. /** Common header for all page types. The page type depends on #mp_flags.
  716. *
  717. * #P_BRANCH and #P_LEAF pages have unsorted '#MDB_node's at the end, with
  718. * sorted #mp_ptrs[] entries referring to them. Exception: #P_LEAF2 pages
  719. * omit mp_ptrs and pack sorted #MDB_DUPFIXED values after the page header.
  720. *
  721. * #P_OVERFLOW records occupy one or more contiguous pages where only the
  722. * first has a page header. They hold the real data of #F_BIGDATA nodes.
  723. *
  724. * #P_SUBP sub-pages are small leaf "pages" with duplicate data.
  725. * A node with flag #F_DUPDATA but not #F_SUBDATA contains a sub-page.
  726. * (Duplicate data can also go in sub-databases, which use normal pages.)
  727. *
  728. * #P_META pages contain #MDB_meta, the start point of an LMDB snapshot.
  729. *
  730. * Each non-metapage up to #MDB_meta.%mm_last_pg is reachable exactly once
  731. * in the snapshot: Either used by a database or listed in a freeDB record.
  732. */
  733. typedef struct MDB_page {
  734. #define mp_pgno mp_p.p_pgno
  735. #define mp_next mp_p.p_next
  736. union {
  737. pgno_t p_pgno; /**< page number */
  738. struct MDB_page *p_next; /**< for in-memory list of freed pages */
  739. } mp_p;
  740. uint16_t mp_pad; /**< key size if this is a LEAF2 page */
  741. /** @defgroup mdb_page Page Flags
  742. * @ingroup internal
  743. * Flags for the page headers.
  744. * @{
  745. */
  746. #define P_BRANCH 0x01 /**< branch page */
  747. #define P_LEAF 0x02 /**< leaf page */
  748. #define P_OVERFLOW 0x04 /**< overflow page */
  749. #define P_META 0x08 /**< meta page */
  750. #define P_DIRTY 0x10 /**< dirty page, also set for #P_SUBP pages */
  751. #define P_LEAF2 0x20 /**< for #MDB_DUPFIXED records */
  752. #define P_SUBP 0x40 /**< for #MDB_DUPSORT sub-pages */
  753. #define P_LOOSE 0x4000 /**< page was dirtied then freed, can be reused */
  754. #define P_KEEP 0x8000 /**< leave this page alone during spill */
  755. /** @} */
  756. uint16_t mp_flags; /**< @ref mdb_page */
  757. #define mp_lower mp_pb.pb.pb_lower
  758. #define mp_upper mp_pb.pb.pb_upper
  759. #define mp_pages mp_pb.pb_pages
  760. union {
  761. struct {
  762. indx_t pb_lower; /**< lower bound of free space */
  763. indx_t pb_upper; /**< upper bound of free space */
  764. } pb;
  765. uint32_t pb_pages; /**< number of overflow pages */
  766. } mp_pb;
  767. indx_t mp_ptrs[1]; /**< dynamic size */
  768. } MDB_page;
  769. /** Size of the page header, excluding dynamic data at the end */
  770. #define PAGEHDRSZ ((unsigned) offsetof(MDB_page, mp_ptrs))
  771. /** Address of first usable data byte in a page, after the header */
  772. #define METADATA(p) ((void *)((char *)(p) + PAGEHDRSZ))
  773. /** ITS#7713, change PAGEBASE to handle 65536 byte pages */
  774. #define PAGEBASE ((MDB_DEVEL) ? PAGEHDRSZ : 0)
  775. /** Number of nodes on a page */
  776. #define NUMKEYS(p) (((p)->mp_lower - (PAGEHDRSZ-PAGEBASE)) >> 1)
  777. /** The amount of space remaining in the page */
  778. #define SIZELEFT(p) (indx_t)((p)->mp_upper - (p)->mp_lower)
  779. /** The percentage of space used in the page, in tenths of a percent. */
  780. #define PAGEFILL(env, p) (1000L * ((env)->me_psize - PAGEHDRSZ - SIZELEFT(p)) / \
  781. ((env)->me_psize - PAGEHDRSZ))
  782. /** The minimum page fill factor, in tenths of a percent.
  783. * Pages emptier than this are candidates for merging.
  784. */
  785. #define FILL_THRESHOLD 250
  786. /** Test if a page is a leaf page */
  787. #define IS_LEAF(p) F_ISSET((p)->mp_flags, P_LEAF)
  788. /** Test if a page is a LEAF2 page */
  789. #define IS_LEAF2(p) F_ISSET((p)->mp_flags, P_LEAF2)
  790. /** Test if a page is a branch page */
  791. #define IS_BRANCH(p) F_ISSET((p)->mp_flags, P_BRANCH)
  792. /** Test if a page is an overflow page */
  793. #define IS_OVERFLOW(p) F_ISSET((p)->mp_flags, P_OVERFLOW)
  794. /** Test if a page is a sub page */
  795. #define IS_SUBP(p) F_ISSET((p)->mp_flags, P_SUBP)
  796. /** The number of overflow pages needed to store the given size. */
  797. #define OVPAGES(size, psize) ((PAGEHDRSZ-1 + (size)) / (psize) + 1)
  798. /** Link in #MDB_txn.%mt_loose_pgs list.
  799. * Kept outside the page header, which is needed when reusing the page.
  800. */
  801. #define NEXT_LOOSE_PAGE(p) (*(MDB_page **)((p) + 2))
  802. /** Header for a single key/data pair within a page.
  803. * Used in pages of type #P_BRANCH and #P_LEAF without #P_LEAF2.
  804. * We guarantee 2-byte alignment for 'MDB_node's.
  805. *
  806. * #mn_lo and #mn_hi are used for data size on leaf nodes, and for child
  807. * pgno on branch nodes. On 64 bit platforms, #mn_flags is also used
  808. * for pgno. (Branch nodes have no flags). Lo and hi are in host byte
  809. * order in case some accesses can be optimized to 32-bit word access.
  810. *
  811. * Leaf node flags describe node contents. #F_BIGDATA says the node's
  812. * data part is the page number of an overflow page with actual data.
  813. * #F_DUPDATA and #F_SUBDATA can be combined giving duplicate data in
  814. * a sub-page/sub-database, and named databases (just #F_SUBDATA).
  815. */
  816. typedef struct MDB_node {
  817. /** part of data size or pgno
  818. * @{ */
  819. #if BYTE_ORDER == LITTLE_ENDIAN
  820. unsigned short mn_lo, mn_hi;
  821. #else
  822. unsigned short mn_hi, mn_lo;
  823. #endif
  824. /** @} */
  825. /** @defgroup mdb_node Node Flags
  826. * @ingroup internal
  827. * Flags for node headers.
  828. * @{
  829. */
  830. #define F_BIGDATA 0x01 /**< data put on overflow page */
  831. #define F_SUBDATA 0x02 /**< data is a sub-database */
  832. #define F_DUPDATA 0x04 /**< data has duplicates */
  833. /** valid flags for #mdb_node_add() */
  834. #define NODE_ADD_FLAGS (F_DUPDATA|F_SUBDATA|MDB_RESERVE|MDB_APPEND)
  835. /** @} */
  836. unsigned short mn_flags; /**< @ref mdb_node */
  837. unsigned short mn_ksize; /**< key size */
  838. char mn_data[1]; /**< key and data are appended here */
  839. } MDB_node;
  840. /** Size of the node header, excluding dynamic data at the end */
  841. #define NODESIZE offsetof(MDB_node, mn_data)
  842. /** Bit position of top word in page number, for shifting mn_flags */
  843. #define PGNO_TOPWORD ((pgno_t)-1 > 0xffffffffu ? 32 : 0)
  844. /** Size of a node in a branch page with a given key.
  845. * This is just the node header plus the key, there is no data.
  846. */
  847. #define INDXSIZE(k) (NODESIZE + ((k) == NULL ? 0 : (k)->mv_size))
  848. /** Size of a node in a leaf page with a given key and data.
  849. * This is node header plus key plus data size.
  850. */
  851. #define LEAFSIZE(k, d) (NODESIZE + (k)->mv_size + (d)->mv_size)
  852. /** Address of node \b i in page \b p */
  853. #define NODEPTR(p, i) ((MDB_node *)((char *)(p) + (p)->mp_ptrs[i] + PAGEBASE))
  854. /** Address of the key for the node */
  855. #define NODEKEY(node) (void *)((node)->mn_data)
  856. /** Address of the data for a node */
  857. #define NODEDATA(node) (void *)((char *)(node)->mn_data + (node)->mn_ksize)
  858. /** Get the page number pointed to by a branch node */
  859. #define NODEPGNO(node) \
  860. ((node)->mn_lo | ((pgno_t) (node)->mn_hi << 16) | \
  861. (PGNO_TOPWORD ? ((pgno_t) (node)->mn_flags << PGNO_TOPWORD) : 0))
  862. /** Set the page number in a branch node */
  863. #define SETPGNO(node,pgno) do { \
  864. (node)->mn_lo = (pgno) & 0xffff; (node)->mn_hi = (pgno) >> 16; \
  865. if (PGNO_TOPWORD) (node)->mn_flags = (pgno) >> PGNO_TOPWORD; } while(0)
  866. /** Get the size of the data in a leaf node */
  867. #define NODEDSZ(node) ((node)->mn_lo | ((unsigned)(node)->mn_hi << 16))
  868. /** Set the size of the data for a leaf node */
  869. #define SETDSZ(node,size) do { \
  870. (node)->mn_lo = (size) & 0xffff; (node)->mn_hi = (size) >> 16;} while(0)
  871. /** The size of a key in a node */
  872. #define NODEKSZ(node) ((node)->mn_ksize)
  873. /** Copy a page number from src to dst */
  874. #ifdef MISALIGNED_OK
  875. #define COPY_PGNO(dst,src) dst = src
  876. #else
  877. #if SIZE_MAX > 4294967295UL
  878. #define COPY_PGNO(dst,src) do { \
  879. unsigned short *s, *d; \
  880. s = (unsigned short *)&(src); \
  881. d = (unsigned short *)&(dst); \
  882. *d++ = *s++; \
  883. *d++ = *s++; \
  884. *d++ = *s++; \
  885. *d = *s; \
  886. } while (0)
  887. #else
  888. #define COPY_PGNO(dst,src) do { \
  889. unsigned short *s, *d; \
  890. s = (unsigned short *)&(src); \
  891. d = (unsigned short *)&(dst); \
  892. *d++ = *s++; \
  893. *d = *s; \
  894. } while (0)
  895. #endif
  896. #endif
  897. /** The address of a key in a LEAF2 page.
  898. * LEAF2 pages are used for #MDB_DUPFIXED sorted-duplicate sub-DBs.
  899. * There are no node headers, keys are stored contiguously.
  900. */
  901. #define LEAF2KEY(p, i, ks) ((char *)(p) + PAGEHDRSZ + ((i)*(ks)))
  902. /** Set the \b node's key into \b keyptr, if requested. */
  903. #define MDB_GET_KEY(node, keyptr) { if ((keyptr) != NULL) { \
  904. (keyptr)->mv_size = NODEKSZ(node); (keyptr)->mv_data = NODEKEY(node); } }
  905. /** Set the \b node's key into \b key. */
  906. #define MDB_GET_KEY2(node, key) { key.mv_size = NODEKSZ(node); key.mv_data = NODEKEY(node); }
  907. /** Information about a single database in the environment. */
  908. typedef struct MDB_db {
  909. uint32_t md_pad; /**< also ksize for LEAF2 pages */
  910. uint16_t md_flags; /**< @ref mdb_dbi_open */
  911. uint16_t md_depth; /**< depth of this tree */
  912. pgno_t md_branch_pages; /**< number of internal pages */
  913. pgno_t md_leaf_pages; /**< number of leaf pages */
  914. pgno_t md_overflow_pages; /**< number of overflow pages */
  915. size_t md_entries; /**< number of data items */
  916. pgno_t md_root; /**< the root page of this tree */
  917. } MDB_db;
  918. #define MDB_VALID 0x8000 /**< DB handle is valid, for me_dbflags */
  919. #define PERSISTENT_FLAGS (0xffff & ~(MDB_VALID))
  920. /** #mdb_dbi_open() flags */
  921. #define VALID_FLAGS (MDB_REVERSEKEY|MDB_DUPSORT|MDB_INTEGERKEY|MDB_DUPFIXED|\
  922. MDB_INTEGERDUP|MDB_REVERSEDUP|MDB_CREATE)
  923. /** Handle for the DB used to track free pages. */
  924. #define FREE_DBI 0
  925. /** Handle for the default DB. */
  926. #define MAIN_DBI 1
  927. /** Number of DBs in metapage (free and main) - also hardcoded elsewhere */
  928. #define CORE_DBS 2
  929. /** Number of meta pages - also hardcoded elsewhere */
  930. #define NUM_METAS 2
  931. /** Meta page content.
  932. * A meta page is the start point for accessing a database snapshot.
  933. * Pages 0-1 are meta pages. Transaction N writes meta page #(N % 2).
  934. */
  935. typedef struct MDB_meta {
  936. /** Stamp identifying this as an LMDB file. It must be set
  937. * to #MDB_MAGIC. */
  938. uint32_t mm_magic;
  939. /** Version number of this file. Must be set to #MDB_DATA_VERSION. */
  940. uint32_t mm_version;
  941. void *mm_address; /**< address for fixed mapping */
  942. size_t mm_mapsize; /**< size of mmap region */
  943. MDB_db mm_dbs[CORE_DBS]; /**< first is free space, 2nd is main db */
  944. /** The size of pages used in this DB */
  945. #define mm_psize mm_dbs[FREE_DBI].md_pad
  946. /** Any persistent environment flags. @ref mdb_env */
  947. #define mm_flags mm_dbs[FREE_DBI].md_flags
  948. /** Last used page in the datafile.
  949. * Actually the file may be shorter if the freeDB lists the final pages.
  950. */
  951. pgno_t mm_last_pg;
  952. volatile txnid_t mm_txnid; /**< txnid that committed this page */
  953. } MDB_meta;
  954. /** Buffer for a stack-allocated meta page.
  955. * The members define size and alignment, and silence type
  956. * aliasing warnings. They are not used directly; that could
  957. * mean incorrectly using several union members in parallel.
  958. */
  959. typedef union MDB_metabuf {
  960. MDB_page mb_page;
  961. struct {
  962. char mm_pad[PAGEHDRSZ];
  963. MDB_meta mm_meta;
  964. } mb_metabuf;
  965. } MDB_metabuf;
  966. /** Auxiliary DB info.
  967. * The information here is mostly static/read-only. There is
  968. * only a single copy of this record in the environment.
  969. */
  970. typedef struct MDB_dbx {
  971. MDB_val md_name; /**< name of the database */
  972. MDB_cmp_func *md_cmp; /**< function for comparing keys */
  973. MDB_cmp_func *md_dcmp; /**< function for comparing data items */
  974. MDB_rel_func *md_rel; /**< user relocate function */
  975. void *md_relctx; /**< user-provided context for md_rel */
  976. } MDB_dbx;
  977. /** A database transaction.
  978. * Every operation requires a transaction handle.
  979. */
  980. struct MDB_txn {
  981. MDB_txn *mt_parent; /**< parent of a nested txn */
  982. /** Nested txn under this txn, set together with flag #MDB_TXN_HAS_CHILD */
  983. MDB_txn *mt_child;
  984. pgno_t mt_next_pgno; /**< next unallocated page */
  985. /** The ID of this transaction. IDs are integers incrementing from 1.
  986. * Only committed write transactions increment the ID. If a transaction
  987. * aborts, the ID may be re-used by the next writer.
  988. */
  989. txnid_t mt_txnid;
  990. MDB_env *mt_env; /**< the DB environment */
  991. /** The list of pages that became unused during this transaction.
  992. */
  993. MDB_IDL mt_free_pgs;
  994. /** The list of loose pages that became unused and may be reused
  995. * in this transaction, linked through #NEXT_LOOSE_PAGE(page).
  996. */
  997. MDB_page *mt_loose_pgs;
  998. /** Number of loose pages (#mt_loose_pgs) */
  999. int mt_loose_count;
  1000. /** The sorted list of dirty pages we temporarily wrote to disk
  1001. * because the dirty list was full. page numbers in here are
  1002. * shifted left by 1, deleted slots have the LSB set.
  1003. */
  1004. MDB_IDL mt_spill_pgs;
  1005. union {
  1006. /** For write txns: Modified pages. Sorted when not MDB_WRITEMAP. */
  1007. MDB_ID2L dirty_list;
  1008. /** For read txns: This thread/txn's reader table slot, or NULL. */
  1009. MDB_reader *reader;
  1010. } mt_u;
  1011. /** Array of records for each DB known in the environment. */
  1012. MDB_dbx *mt_dbxs;
  1013. /** Array of MDB_db records for each known DB */
  1014. MDB_db *mt_dbs;
  1015. /** Array of sequence numbers for each DB handle */
  1016. unsigned int *mt_dbiseqs;
  1017. /** @defgroup mt_dbflag Transaction DB Flags
  1018. * @ingroup internal
  1019. * @{
  1020. */
  1021. #define DB_DIRTY 0x01 /**< DB was written in this txn */
  1022. #define DB_STALE 0x02 /**< Named-DB record is older than txnID */
  1023. #define DB_NEW 0x04 /**< Named-DB handle opened in this txn */
  1024. #define DB_VALID 0x08 /**< DB handle is valid, see also #MDB_VALID */
  1025. #define DB_USRVALID 0x10 /**< As #DB_VALID, but not set for #FREE_DBI */
  1026. #define DB_DUPDATA 0x20 /**< DB is #MDB_DUPSORT data */
  1027. /** @} */
  1028. /** In write txns, array of cursors for each DB */
  1029. MDB_cursor **mt_cursors;
  1030. /** Array of flags for each DB */
  1031. unsigned char *mt_dbflags;
  1032. /** Number of DB records in use, or 0 when the txn is finished.
  1033. * This number only ever increments until the txn finishes; we
  1034. * don't decrement it when individual DB handles are closed.
  1035. */
  1036. MDB_dbi mt_numdbs;
  1037. /** @defgroup mdb_txn Transaction Flags
  1038. * @ingroup internal
  1039. * @{
  1040. */
  1041. /** #mdb_txn_begin() flags */
  1042. #define MDB_TXN_BEGIN_FLAGS MDB_RDONLY
  1043. #define MDB_TXN_RDONLY MDB_RDONLY /**< read-only transaction */
  1044. /* internal txn flags */
  1045. #define MDB_TXN_WRITEMAP MDB_WRITEMAP /**< copy of #MDB_env flag in writers */
  1046. #define MDB_TXN_FINISHED 0x01 /**< txn is finished or never began */
  1047. #define MDB_TXN_ERROR 0x02 /**< txn is unusable after an error */
  1048. #define MDB_TXN_DIRTY 0x04 /**< must write, even if dirty list is empty */
  1049. #define MDB_TXN_SPILLS 0x08 /**< txn or a parent has spilled pages */
  1050. #define MDB_TXN_HAS_CHILD 0x10 /**< txn has an #MDB_txn.%mt_child */
  1051. /** most operations on the txn are currently illegal */
  1052. #define MDB_TXN_BLOCKED (MDB_TXN_FINISHED|MDB_TXN_ERROR|MDB_TXN_HAS_CHILD)
  1053. /** @} */
  1054. unsigned int mt_flags; /**< @ref mdb_txn */
  1055. /** #dirty_list room: Array size - \#dirty pages visible to this txn.
  1056. * Includes ancestor txns' dirty pages not hidden by other txns'
  1057. * dirty/spilled pages. Thus commit(nested txn) has room to merge
  1058. * dirty_list into mt_parent after freeing hidden mt_parent pages.
  1059. */
  1060. unsigned int mt_dirty_room;
  1061. };
  1062. /** Enough space for 2^32 nodes with minimum of 2 keys per node. I.e., plenty.
  1063. * At 4 keys per node, enough for 2^64 nodes, so there's probably no need to
  1064. * raise this on a 64 bit machine.
  1065. */
  1066. #define CURSOR_STACK 32
  1067. struct MDB_xcursor;
  1068. /** Cursors are used for all DB operations.
  1069. * A cursor holds a path of (page pointer, key index) from the DB
  1070. * root to a position in the DB, plus other state. #MDB_DUPSORT
  1071. * cursors include an xcursor to the current data item. Write txns
  1072. * track their cursors and keep them up to date when data moves.
  1073. * Exception: An xcursor's pointer to a #P_SUBP page can be stale.
  1074. * (A node with #F_DUPDATA but no #F_SUBDATA contains a subpage).
  1075. */
  1076. struct MDB_cursor {
  1077. /** Next cursor on this DB in this txn */
  1078. MDB_cursor *mc_next;
  1079. /** Backup of the original cursor if this cursor is a shadow */
  1080. MDB_cursor *mc_backup;
  1081. /** Context used for databases with #MDB_DUPSORT, otherwise NULL */
  1082. struct MDB_xcursor *mc_xcursor;
  1083. /** The transaction that owns this cursor */
  1084. MDB_txn *mc_txn;
  1085. /** The database handle this cursor operates on */
  1086. MDB_dbi mc_dbi;
  1087. /** The database record for this cursor */
  1088. MDB_db *mc_db;
  1089. /** The database auxiliary record for this cursor */
  1090. MDB_dbx *mc_dbx;
  1091. /** The @ref mt_dbflag for this database */
  1092. unsigned char *mc_dbflag;
  1093. unsigned short mc_snum; /**< number of pushed pages */
  1094. unsigned short mc_top; /**< index of top page, normally mc_snum-1 */
  1095. /** @defgroup mdb_cursor Cursor Flags
  1096. * @ingroup internal
  1097. * Cursor state flags.
  1098. * @{
  1099. */
  1100. #define C_INITIALIZED 0x01 /**< cursor has been initialized and is valid */
  1101. #define C_EOF 0x02 /**< No more data */
  1102. #define C_SUB 0x04 /**< Cursor is a sub-cursor */
  1103. #define C_DEL 0x08 /**< last op was a cursor_del */
  1104. #define C_UNTRACK 0x40 /**< Un-track cursor when closing */
  1105. /** @} */
  1106. unsigned int mc_flags; /**< @ref mdb_cursor */
  1107. MDB_page *mc_pg[CURSOR_STACK]; /**< stack of pushed pages */
  1108. indx_t mc_ki[CURSOR_STACK]; /**< stack of page indices */
  1109. };
  1110. /** Context for sorted-dup records.
  1111. * We could have gone to a fully recursive design, with arbitrarily
  1112. * deep nesting of sub-databases. But for now we only handle these
  1113. * levels - main DB, optional sub-DB, sorted-duplicate DB.
  1114. */
  1115. typedef struct MDB_xcursor {
  1116. /** A sub-cursor for traversing the Dup DB */
  1117. MDB_cursor mx_cursor;
  1118. /** The database record for this Dup DB */
  1119. MDB_db mx_db;
  1120. /** The auxiliary DB record for this Dup DB */
  1121. MDB_dbx mx_dbx;
  1122. /** The @ref mt_dbflag for this Dup DB */
  1123. unsigned char mx_dbflag;
  1124. } MDB_xcursor;
  1125. /** Check if there is an inited xcursor */
  1126. #define XCURSOR_INITED(mc) \
  1127. ((mc)->mc_xcursor && ((mc)->mc_xcursor->mx_cursor.mc_flags & C_INITIALIZED))
  1128. /** Update the xcursor's sub-page pointer, if any, in \b mc. Needed
  1129. * when the node which contains the sub-page may have moved. Called
  1130. * with leaf page \b mp = mc->mc_pg[\b top].
  1131. */
  1132. #define XCURSOR_REFRESH(mc, top, mp) do { \
  1133. MDB_page *xr_pg = (mp); \
  1134. MDB_node *xr_node; \
  1135. if (!XCURSOR_INITED(mc) || (mc)->mc_ki[top] >= NUMKEYS(xr_pg)) break; \
  1136. xr_node = NODEPTR(xr_pg, (mc)->mc_ki[top]); \
  1137. if ((xr_node->mn_flags & (F_DUPDATA|F_SUBDATA)) == F_DUPDATA) \
  1138. (mc)->mc_xcursor->mx_cursor.mc_pg[0] = NODEDATA(xr_node); \
  1139. } while (0)
  1140. /** State of FreeDB old pages, stored in the MDB_env */
  1141. typedef struct MDB_pgstate {
  1142. pgno_t *mf_pghead; /**< Reclaimed freeDB pages, or NULL before use */
  1143. txnid_t mf_pglast; /**< ID of last used record, or 0 if !mf_pghead */
  1144. } MDB_pgstate;
  1145. /** The database environment. */
  1146. struct MDB_env {
  1147. HANDLE me_fd; /**< The main data file */
  1148. HANDLE me_lfd; /**< The lock file */
  1149. HANDLE me_mfd; /**< For writing and syncing the meta pages */
  1150. /** Failed to update the meta page. Probably an I/O error. */
  1151. #define MDB_FATAL_ERROR 0x80000000U
  1152. /** Some fields are initialized. */
  1153. #define MDB_ENV_ACTIVE 0x20000000U
  1154. /** me_txkey is set */
  1155. #define MDB_ENV_TXKEY 0x10000000U
  1156. /** fdatasync is unreliable */
  1157. #define MDB_FSYNCONLY 0x08000000U
  1158. uint32_t me_flags; /**< @ref mdb_env */
  1159. unsigned int me_psize; /**< DB page size, inited from me_os_psize */
  1160. unsigned int me_os_psize; /**< OS page size, from #GET_PAGESIZE */
  1161. unsigned int me_maxreaders; /**< size of the reader table */
  1162. /** Max #MDB_txninfo.%mti_numreaders of interest to #mdb_env_close() */
  1163. volatile int me_close_readers;
  1164. MDB_dbi me_numdbs; /**< number of DBs opened */
  1165. MDB_dbi me_maxdbs; /**< size of the DB table */
  1166. MDB_PID_T me_pid; /**< process ID of this env */
  1167. char *me_path; /**< path to the DB files */
  1168. char *me_map; /**< the memory map of the data file */
  1169. MDB_txninfo *me_txns; /**< the memory map of the lock file or NULL */
  1170. MDB_meta *me_metas[NUM_METAS]; /**< pointers to the two meta pages */
  1171. void *me_pbuf; /**< scratch area for DUPSORT put() */
  1172. MDB_txn *me_txn; /**< current write transaction */
  1173. MDB_txn *me_txn0; /**< prealloc'd write transaction */
  1174. size_t me_mapsize; /**< size of the data memory map */
  1175. off_t me_size; /**< current file size */
  1176. pgno_t me_maxpg; /**< me_mapsize / me_psize */
  1177. MDB_dbx *me_dbxs; /**< array of static DB info */
  1178. uint16_t *me_dbflags; /**< array of flags from MDB_db.md_flags */
  1179. unsigned int *me_dbiseqs; /**< array of dbi sequence numbers */
  1180. pthread_key_t me_txkey; /**< thread-key for readers */
  1181. txnid_t me_pgoldest; /**< ID of oldest reader last time we looked */
  1182. MDB_pgstate me_pgstate; /**< state of old pages from freeDB */
  1183. # define me_pglast me_pgstate.mf_pglast
  1184. # define me_pghead me_pgstate.mf_pghead
  1185. MDB_page *me_dpages; /**< list of malloc'd blocks for re-use */
  1186. /** IDL of pages that became unused in a write txn */
  1187. MDB_IDL me_free_pgs;
  1188. /** ID2L of pages written during a write txn. Length MDB_IDL_UM_SIZE. */
  1189. MDB_ID2L me_dirty_list;
  1190. /** Max number of freelist items that can fit in a single overflow page */
  1191. int me_maxfree_1pg;
  1192. /** Max size of a node on a page */
  1193. unsigned int me_nodemax;
  1194. #if !(MDB_MAXKEYSIZE)
  1195. unsigned int me_maxkey; /**< max size of a key */
  1196. #endif
  1197. int me_live_reader; /**< have liveness lock in reader table */
  1198. #ifdef _WIN32
  1199. int me_pidquery; /**< Used in OpenProcess */
  1200. #endif
  1201. #ifdef MDB_USE_POSIX_MUTEX /* Posix mutexes reside in shared mem */
  1202. # define me_rmutex me_txns->mti_rmutex /**< Shared reader lock */
  1203. # define me_wmutex me_txns->mti_wmutex /**< Shared writer lock */
  1204. #else
  1205. mdb_mutex_t me_rmutex;
  1206. mdb_mutex_t me_wmutex;
  1207. #endif
  1208. void *me_userctx; /**< User-settable context */
  1209. MDB_assert_func *me_assert_func; /**< Callback for assertion failures */
  1210. };
  1211. /** Nested transaction */
  1212. typedef struct MDB_ntxn {
  1213. MDB_txn mnt_txn; /**< the transaction */
  1214. MDB_pgstate mnt_pgstate; /**< parent transaction's saved freestate */
  1215. } MDB_ntxn;
  1216. /** max number of pages to commit in one writev() call */
  1217. #define MDB_COMMIT_PAGES 64
  1218. #if defined(IOV_MAX) && IOV_MAX < MDB_COMMIT_PAGES
  1219. #undef MDB_COMMIT_PAGES
  1220. #define MDB_COMMIT_PAGES IOV_MAX
  1221. #endif
  1222. /** max bytes to write in one call */
  1223. #define MAX_WRITE (0x40000000U >> (sizeof(ssize_t) == 4))
  1224. /** Check \b txn and \b dbi arguments to a function */
  1225. #define TXN_DBI_EXIST(txn, dbi, validity) \
  1226. ((txn) && (dbi)<(txn)->mt_numdbs && ((txn)->mt_dbflags[dbi] & (validity)))
  1227. /** Check for misused \b dbi handles */
  1228. #define TXN_DBI_CHANGED(txn, dbi) \
  1229. ((txn)->mt_dbiseqs[dbi] != (txn)->mt_env->me_dbiseqs[dbi])
  1230. static int mdb_page_alloc(MDB_cursor *mc, int num, MDB_page **mp);
  1231. static int mdb_page_new(MDB_cursor *mc, uint32_t flags, int num, MDB_page **mp);
  1232. static int mdb_page_touch(MDB_cursor *mc);
  1233. #define MDB_END_NAMES {"committed", "empty-commit", "abort", "reset", \
  1234. "reset-tmp", "fail-begin", "fail-beginchild"}
  1235. enum {
  1236. /* mdb_txn_end operation number, for logging */
  1237. MDB_END_COMMITTED, MDB_END_EMPTY_COMMIT, MDB_END_ABORT, MDB_END_RESET,
  1238. MDB_END_RESET_TMP, MDB_END_FAIL_BEGIN, MDB_END_FAIL_BEGINCHILD
  1239. };
  1240. #define MDB_END_OPMASK 0x0F /**< mask for #mdb_txn_end() operation number */
  1241. #define MDB_END_UPDATE 0x10 /**< update env state (DBIs) */
  1242. #define MDB_END_FREE 0x20 /**< free txn unless it is #MDB_env.%me_txn0 */
  1243. #define MDB_END_SLOT MDB_NOTLS /**< release any reader slot if #MDB_NOTLS */
  1244. static void mdb_txn_end(MDB_txn *txn, unsigned mode);
  1245. static int mdb_page_get(MDB_cursor *mc, pgno_t pgno, MDB_page **mp, int *lvl);
  1246. static int mdb_page_search_root(MDB_cursor *mc,
  1247. MDB_val *key, int modify);
  1248. #define MDB_PS_MODIFY 1
  1249. #define MDB_PS_ROOTONLY 2
  1250. #define MDB_PS_FIRST 4
  1251. #define MDB_PS_LAST 8
  1252. static int mdb_page_search(MDB_cursor *mc,
  1253. MDB_val *key, int flags);
  1254. static int mdb_page_merge(MDB_cursor *csrc, MDB_cursor *cdst);
  1255. #define MDB_SPLIT_REPLACE MDB_APPENDDUP /**< newkey is not new */
  1256. static int mdb_page_split(MDB_cursor *mc, MDB_val *newkey, MDB_val *newdata,
  1257. pgno_t newpgno, unsigned int nflags);
  1258. static int mdb_env_read_header(MDB_env *env, MDB_meta *meta);
  1259. static MDB_meta *mdb_env_pick_meta(const MDB_env *env);
  1260. static int mdb_env_write_meta(MDB_txn *txn);
  1261. #if defined(MDB_USE_POSIX_MUTEX) && !defined(MDB_ROBUST_SUPPORTED) /* Drop unused excl arg */
  1262. # define mdb_env_close0(env, excl) mdb_env_close1(env)
  1263. #endif
  1264. static void mdb_env_close0(MDB_env *env, int excl);
  1265. static MDB_node *mdb_node_search(MDB_cursor *mc, MDB_val *key, int *exactp);
  1266. static int mdb_node_add(MDB_cursor *mc, indx_t indx,
  1267. MDB_val *key, MDB_val *data, pgno_t pgno, unsigned int flags);
  1268. static void mdb_node_del(MDB_cursor *mc, int ksize);
  1269. static void mdb_node_shrink(MDB_page *mp, indx_t indx);
  1270. static int mdb_node_move(MDB_cursor *csrc, MDB_cursor *cdst, int fromleft);
  1271. static int mdb_node_read(MDB_cursor *mc, MDB_node *leaf, MDB_val *data);
  1272. static size_t mdb_leaf_size(MDB_env *env, MDB_val *key, MDB_val *data);
  1273. static size_t mdb_branch_size(MDB_env *env, MDB_val *key);
  1274. static int mdb_rebalance(MDB_cursor *mc);
  1275. static int mdb_update_key(MDB_cursor *mc, MDB_val *key);
  1276. static void mdb_cursor_pop(MDB_cursor *mc);
  1277. static int mdb_cursor_push(MDB_cursor *mc, MDB_page *mp);
  1278. static int mdb_cursor_del0(MDB_cursor *mc);
  1279. static int mdb_del0(MDB_txn *txn, MDB_dbi dbi, MDB_val *key, MDB_val *data, unsigned flags);
  1280. static int mdb_cursor_sibling(MDB_cursor *mc, int move_right);
  1281. static int mdb_cursor_next(MDB_cursor *mc, MDB_val *key, MDB_val *data, MDB_cursor_op op);
  1282. static int mdb_cursor_prev(MDB_cursor *mc, MDB_val *key, MDB_val *data, MDB_cursor_op op);
  1283. static int mdb_cursor_set(MDB_cursor *mc, MDB_val *key, MDB_val *data, MDB_cursor_op op,
  1284. int *exactp);
  1285. static int mdb_cursor_first(MDB_cursor *mc, MDB_val *key, MDB_val *data);
  1286. static int mdb_cursor_last(MDB_cursor *mc, MDB_val *key, MDB_val *data);
  1287. static void mdb_cursor_init(MDB_cursor *mc, MDB_txn *txn, MDB_dbi dbi, MDB_xcursor *mx);
  1288. static void mdb_xcursor_init0(MDB_cursor *mc);
  1289. static void mdb_xcursor_init1(MDB_cursor *mc, MDB_node *node);
  1290. static void mdb_xcursor_init2(MDB_cursor *mc, MDB_xcursor *src_mx, int force);
  1291. static int mdb_drop0(MDB_cursor *mc, int subs);
  1292. static void mdb_default_cmp(MDB_txn *txn, MDB_dbi dbi);
  1293. static int mdb_reader_check0(MDB_env *env, int rlocked, int *dead);
  1294. /** @cond */
  1295. static MDB_cmp_func mdb_cmp_memn, mdb_cmp_memnr, mdb_cmp_int, mdb_cmp_cint, mdb_cmp_long;
  1296. /** @endcond */
  1297. /** Compare two items pointing at size_t's of unknown alignment. */
  1298. #ifdef MISALIGNED_OK
  1299. # define mdb_cmp_clong mdb_cmp_long
  1300. #else
  1301. # define mdb_cmp_clong mdb_cmp_cint
  1302. #endif
  1303. #ifdef _WIN32
  1304. static SECURITY_DESCRIPTOR mdb_null_sd;
  1305. static SECURITY_ATTRIBUTES mdb_all_sa;
  1306. static int mdb_sec_inited;
  1307. struct MDB_name;
  1308. static int utf8_to_utf16(const char *src, struct MDB_name *dst, int xtra);
  1309. #endif
  1310. /** Return the library version info. */
  1311. char * ESECT
  1312. mdb_version(int *major, int *minor, int *patch)
  1313. {
  1314. if (major) *major = MDB_VERSION_MAJOR;
  1315. if (minor) *minor = MDB_VERSION_MINOR;
  1316. if (patch) *patch = MDB_VERSION_PATCH;
  1317. return MDB_VERSION_STRING;
  1318. }
  1319. /** Table of descriptions for LMDB @ref errors */
  1320. static char *const mdb_errstr[] = {
  1321. "MDB_KEYEXIST: Key/data pair already exists",
  1322. "MDB_NOTFOUND: No matching key/data pair found",
  1323. "MDB_PAGE_NOTFOUND: Requested page not found",
  1324. "MDB_CORRUPTED: Located page was wrong type",
  1325. "MDB_PANIC: Update of meta page failed or environment had fatal error",
  1326. "MDB_VERSION_MISMATCH: Database environment version mismatch",
  1327. "MDB_INVALID: File is not an LMDB file",
  1328. "MDB_MAP_FULL: Environment mapsize limit reached",
  1329. "MDB_DBS_FULL: Environment maxdbs limit reached",
  1330. "MDB_READERS_FULL: Environment maxreaders limit reached",
  1331. "MDB_TLS_FULL: Thread-local storage keys full - too many environments open",
  1332. "MDB_TXN_FULL: Transaction has too many dirty pages - transaction too big",
  1333. "MDB_CURSOR_FULL: Internal error - cursor stack limit reached",
  1334. "MDB_PAGE_FULL: Internal error - page has no more space",
  1335. "MDB_MAP_RESIZED: Database contents grew beyond environment mapsize",
  1336. "MDB_INCOMPATIBLE: Operation and DB incompatible, or DB flags changed",
  1337. "MDB_BAD_RSLOT: Invalid reuse of reader locktable slot",
  1338. "MDB_BAD_TXN: Transaction must abort, has a child, or is invalid",
  1339. "MDB_BAD_VALSIZE: Unsupported size of key/DB name/data, or wrong DUPFIXED size",
  1340. "MDB_BAD_DBI: The specified DBI handle was closed/changed unexpectedly",
  1341. };
  1342. char *
  1343. mdb_strerror(int err)
  1344. {
  1345. #ifdef _WIN32
  1346. /** HACK: pad 4KB on stack over the buf. Return system msgs in buf.
  1347. * This works as long as no function between the call to mdb_strerror
  1348. * and the actual use of the message uses more than 4K of stack.
  1349. */
  1350. #define MSGSIZE 1024
  1351. #define PADSIZE 4096
  1352. char buf[MSGSIZE+PADSIZE], *ptr = buf;
  1353. #endif
  1354. int i;
  1355. if (!err)
  1356. return ("Successful return: 0");
  1357. if (err >= MDB_KEYEXIST && err <= MDB_LAST_ERRCODE) {
  1358. i = err - MDB_KEYEXIST;
  1359. return mdb_errstr[i];
  1360. }
  1361. #ifdef _WIN32
  1362. /* These are the C-runtime error codes we use. The comment indicates
  1363. * their numeric value, and the Win32 error they would correspond to
  1364. * if the error actually came from a Win32 API. A major mess, we should
  1365. * have used LMDB-specific error codes for everything.
  1366. */
  1367. switch(err) {
  1368. case ENOENT: /* 2, FILE_NOT_FOUND */
  1369. case EIO: /* 5, ACCESS_DENIED */
  1370. case ENOMEM: /* 12, INVALID_ACCESS */
  1371. case EACCES: /* 13, INVALID_DATA */
  1372. case EBUSY: /* 16, CURRENT_DIRECTORY */
  1373. case EINVAL: /* 22, BAD_COMMAND */
  1374. case ENOSPC: /* 28, OUT_OF_PAPER */
  1375. return strerror(err);
  1376. default:
  1377. ;
  1378. }
  1379. buf[0] = 0;
  1380. FormatMessageA(FORMAT_MESSAGE_FROM_SYSTEM |
  1381. FORMAT_MESSAGE_IGNORE_INSERTS,
  1382. NULL, err, 0, ptr, MSGSIZE, (va_list *)buf+MSGSIZE);
  1383. return ptr;
  1384. #else
  1385. return strerror(err);
  1386. #endif
  1387. }
  1388. /** assert(3) variant in cursor context */
  1389. #define mdb_cassert(mc, expr) mdb_assert0((mc)->mc_txn->mt_env, expr, #expr)
  1390. /** assert(3) variant in transaction context */
  1391. #define mdb_tassert(txn, expr) mdb_assert0((txn)->mt_env, expr, #expr)
  1392. /** assert(3) variant in environment context */
  1393. #define mdb_eassert(env, expr) mdb_assert0(env, expr, #expr)
  1394. #ifndef NDEBUG
  1395. # define mdb_assert0(env, expr, expr_txt) ((expr) ? (void)0 : \
  1396. mdb_assert_fail(env, expr_txt, mdb_func_, __FILE__, __LINE__))
  1397. static void ESECT
  1398. mdb_assert_fail(MDB_env *env, const char *expr_txt,
  1399. const char *func, const char *file, int line)
  1400. {
  1401. char buf[400];
  1402. sprintf(buf, "%.100s:%d: Assertion '%.200s' failed in %.40s()",
  1403. file, line, expr_txt, func);
  1404. if (env->me_assert_func)
  1405. env->me_assert_func(env, buf);
  1406. fprintf(stderr, "%s\n", buf);
  1407. abort();
  1408. }
  1409. #else
  1410. # define mdb_assert0(env, expr, expr_txt) ((void) 0)
  1411. #endif /* NDEBUG */
  1412. #if MDB_DEBUG
  1413. /** Return the page number of \b mp which may be sub-page, for debug output */
  1414. static pgno_t
  1415. mdb_dbg_pgno(MDB_page *mp)
  1416. {
  1417. pgno_t ret;
  1418. COPY_PGNO(ret, mp->mp_pgno);
  1419. return ret;
  1420. }
  1421. /** Display a key in hexadecimal and return the address of the result.
  1422. * @param[in] key the key to display
  1423. * @param[in] buf the buffer to write into. Should always be #DKBUF.
  1424. * @return The key in hexadecimal form.
  1425. */
  1426. char *
  1427. mdb_dkey(MDB_val *key, char *buf)
  1428. {
  1429. char *ptr = buf;
  1430. unsigned char *c = key->mv_data;
  1431. unsigned int i;
  1432. if (!key)
  1433. return "";
  1434. if (key->mv_size > DKBUF_MAXKEYSIZE)
  1435. return "MDB_MAXKEYSIZE";
  1436. /* may want to make this a dynamic check: if the key is mostly
  1437. * printable characters, print it as-is instead of converting to hex.
  1438. */
  1439. #if 1
  1440. buf[0] = '\0';
  1441. for (i=0; i<key->mv_size; i++)
  1442. ptr += sprintf(ptr, "%02x", *c++);
  1443. #else
  1444. sprintf(buf, "%.*s", key->mv_size, key->mv_data);
  1445. #endif
  1446. return buf;
  1447. }
  1448. static const char *
  1449. mdb_leafnode_type(MDB_node *n)
  1450. {
  1451. static char *const tp[2][2] = {{"", ": DB"}, {": sub-page", ": sub-DB"}};
  1452. return F_ISSET(n->mn_flags, F_BIGDATA) ? ": overflow page" :
  1453. tp[F_ISSET(n->mn_flags, F_DUPDATA)][F_ISSET(n->mn_flags, F_SUBDATA)];
  1454. }
  1455. /** Display all the keys in the page. */
  1456. void
  1457. mdb_page_list(MDB_page *mp)
  1458. {
  1459. pgno_t pgno = mdb_dbg_pgno(mp);
  1460. const char *type, *state = (mp->mp_flags & P_DIRTY) ? ", dirty" : "";
  1461. MDB_node *node;
  1462. unsigned int i, nkeys, nsize, total = 0;
  1463. MDB_val key;
  1464. DKBUF;
  1465. switch (mp->mp_flags & (P_BRANCH|P_LEAF|P_LEAF2|P_META|P_OVERFLOW|P_SUBP)) {
  1466. case P_BRANCH: type = "Branch page"; break;
  1467. case P_LEAF: type = "Leaf page"; break;
  1468. case P_LEAF|P_SUBP: type = "Sub-page"; break;
  1469. case P_LEAF|P_LEAF2: type = "LEAF2 page"; break;
  1470. case P_LEAF|P_LEAF2|P_SUBP: type = "LEAF2 sub-page"; break;
  1471. case P_OVERFLOW:
  1472. fprintf(stderr, "Overflow page %"Z"u pages %u%s\n",
  1473. pgno, mp->mp_pages, state);
  1474. return;
  1475. case P_META:
  1476. fprintf(stderr, "Meta-page %"Z"u txnid %"Z"u\n",
  1477. pgno, ((MDB_meta *)METADATA(mp))->mm_txnid);
  1478. return;
  1479. default:
  1480. fprintf(stderr, "Bad page %"Z"u flags 0x%X\n", pgno, mp->mp_flags);
  1481. return;
  1482. }
  1483. nkeys = NUMKEYS(mp);
  1484. fprintf(stderr, "%s %"Z"u numkeys %d%s\n", type, pgno, nkeys, state);
  1485. for (i=0; i<nkeys; i++) {
  1486. if (IS_LEAF2(mp)) { /* LEAF2 pages have no mp_ptrs[] or node headers */
  1487. key.mv_size = nsize = mp->mp_pad;
  1488. key.mv_data = LEAF2KEY(mp, i, nsize);
  1489. total += nsize;
  1490. fprintf(stderr, "key %d: nsize %d, %s\n", i, nsize, DKEY(&key));
  1491. continue;
  1492. }
  1493. node = NODEPTR(mp, i);
  1494. key.mv_size = node->mn_ksize;
  1495. key.mv_data = node->mn_data;
  1496. nsize = NODESIZE + key.mv_size;
  1497. if (IS_BRANCH(mp)) {
  1498. fprintf(stderr, "key %d: page %"Z"u, %s\n", i, NODEPGNO(node),
  1499. DKEY(&key));
  1500. total += nsize;
  1501. } else {
  1502. if (F_ISSET(node->mn_flags, F_BIGDATA))
  1503. nsize += sizeof(pgno_t);
  1504. else
  1505. nsize += NODEDSZ(node);
  1506. total += nsize;
  1507. nsize += sizeof(indx_t);
  1508. fprintf(stderr, "key %d: nsize %d, %s%s\n",
  1509. i, nsize, DKEY(&key), mdb_leafnode_type(node));
  1510. }
  1511. total = EVEN(total);
  1512. }
  1513. fprintf(stderr, "Total: header %d + contents %d + unused %d\n",
  1514. IS_LEAF2(mp) ? PAGEHDRSZ : PAGEBASE + mp->mp_lower, total, SIZELEFT(mp));
  1515. }
  1516. void
  1517. mdb_cursor_chk(MDB_cursor *mc)
  1518. {
  1519. unsigned int i;
  1520. MDB_node *node;
  1521. MDB_page *mp;
  1522. if (!mc->mc_snum || !(mc->mc_flags & C_INITIALIZED)) return;
  1523. for (i=0; i<mc->mc_top; i++) {
  1524. mp = mc->mc_pg[i];
  1525. node = NODEPTR(mp, mc->mc_ki[i]);
  1526. if (NODEPGNO(node) != mc->mc_pg[i+1]->mp_pgno)
  1527. printf("oops!\n");
  1528. }
  1529. if (mc->mc_ki[i] >= NUMKEYS(mc->mc_pg[i]))
  1530. printf("ack!\n");
  1531. if (XCURSOR_INITED(mc)) {
  1532. node = NODEPTR(mc->mc_pg[mc->mc_top], mc->mc_ki[mc->mc_top]);
  1533. if (((node->mn_flags & (F_DUPDATA|F_SUBDATA)) == F_DUPDATA) &&
  1534. mc->mc_xcursor->mx_cursor.mc_pg[0] != NODEDATA(node)) {
  1535. printf("blah!\n");
  1536. }
  1537. }
  1538. }
  1539. #endif
  1540. #if (MDB_DEBUG) > 2
  1541. /** Count all the pages in each DB and in the freelist
  1542. * and make sure it matches the actual number of pages
  1543. * being used.
  1544. * All named DBs must be open for a correct count.
  1545. */
  1546. static void mdb_audit(MDB_txn *txn)
  1547. {
  1548. MDB_cursor mc;
  1549. MDB_val key, data;
  1550. MDB_ID freecount, count;
  1551. MDB_dbi i;
  1552. int rc;
  1553. freecount = 0;
  1554. mdb_cursor_init(&mc, txn, FREE_DBI, NULL);
  1555. while ((rc = mdb_cursor_get(&mc, &key, &data, MDB_NEXT)) == 0)
  1556. freecount += *(MDB_ID *)data.mv_data;
  1557. mdb_tassert(txn, rc == MDB_NOTFOUND);
  1558. count = 0;
  1559. for (i = 0; i<txn->mt_numdbs; i++) {
  1560. MDB_xcursor mx;
  1561. if (!(txn->mt_dbflags[i] & DB_VALID))
  1562. continue;
  1563. mdb_cursor_init(&mc, txn, i, &mx);
  1564. if (txn->mt_dbs[i].md_root == P_INVALID)
  1565. continue;
  1566. count += txn->mt_dbs[i].md_branch_pages +
  1567. txn->mt_dbs[i].md_leaf_pages +
  1568. txn->mt_dbs[i].md_overflow_pages;
  1569. if (txn->mt_dbs[i].md_flags & MDB_DUPSORT) {
  1570. rc = mdb_page_search(&mc, NULL, MDB_PS_FIRST);
  1571. for (; rc == MDB_SUCCESS; rc = mdb_cursor_sibling(&mc, 1)) {
  1572. unsigned j;
  1573. MDB_page *mp;
  1574. mp = mc.mc_pg[mc.mc_top];
  1575. for (j=0; j<NUMKEYS(mp); j++) {
  1576. MDB_node *leaf = NODEPTR(mp, j);
  1577. if (leaf->mn_flags & F_SUBDATA) {
  1578. MDB_db db;
  1579. memcpy(&db, NODEDATA(leaf), sizeof(db));
  1580. count += db.md_branch_pages + db.md_leaf_pages +
  1581. db.md_overflow_pages;
  1582. }
  1583. }
  1584. }
  1585. mdb_tassert(txn, rc == MDB_NOTFOUND);
  1586. }
  1587. }
  1588. if (freecount + count + NUM_METAS != txn->mt_next_pgno) {
  1589. fprintf(stderr, "audit: %"Z"u freecount: %"Z"u count: %"Z"u total: %"Z"u next_pgno: %"Z"u\n",
  1590. txn->mt_txnid, freecount, count+NUM_METAS,
  1591. freecount+count+NUM_METAS, txn->mt_next_pgno);
  1592. }
  1593. }
  1594. #endif
  1595. int
  1596. mdb_cmp(MDB_txn *txn, MDB_dbi dbi, const MDB_val *a, const MDB_val *b)
  1597. {
  1598. return txn->mt_dbxs[dbi].md_cmp(a, b);
  1599. }
  1600. int
  1601. mdb_dcmp(MDB_txn *txn, MDB_dbi dbi, const MDB_val *a, const MDB_val *b)
  1602. {
  1603. MDB_cmp_func *dcmp = txn->mt_dbxs[dbi].md_dcmp;
  1604. #if UINT_MAX < SIZE_MAX
  1605. if (dcmp == mdb_cmp_int && a->mv_size == sizeof(size_t))
  1606. dcmp = mdb_cmp_clong;
  1607. #endif
  1608. return dcmp(a, b);
  1609. }
  1610. /** Allocate memory for a page.
  1611. * Re-use old malloc'd pages first for singletons, otherwise just malloc.
  1612. * Set #MDB_TXN_ERROR on failure.
  1613. */
  1614. static MDB_page *
  1615. mdb_page_malloc(MDB_txn *txn, unsigned num)
  1616. {
  1617. MDB_env *env = txn->mt_env;
  1618. MDB_page *ret = env->me_dpages;
  1619. size_t psize = env->me_psize, sz = psize, off;
  1620. /* For ! #MDB_NOMEMINIT, psize counts how much to init.
  1621. * For a single page alloc, we init everything after the page header.
  1622. * For multi-page, we init the final page; if the caller needed that
  1623. * many pages they will be filling in at least up to the last page.
  1624. */
  1625. if (num == 1) {
  1626. if (ret) {
  1627. VGMEMP_ALLOC(env, ret, sz);
  1628. VGMEMP_DEFINED(ret, sizeof(ret->mp_next));
  1629. env->me_dpages = ret->mp_next;
  1630. return ret;
  1631. }
  1632. psize -= off = PAGEHDRSZ;
  1633. } else {
  1634. sz *= num;
  1635. off = sz - psize;
  1636. }
  1637. if ((ret = malloc(sz)) != NULL) {
  1638. VGMEMP_ALLOC(env, ret, sz);
  1639. if (!(env->me_flags & MDB_NOMEMINIT)) {
  1640. memset((char *)ret + off, 0, psize);
  1641. ret->mp_pad = 0;
  1642. }
  1643. } else {
  1644. txn->mt_flags |= MDB_TXN_ERROR;
  1645. }
  1646. return ret;
  1647. }
  1648. /** Free a single page.
  1649. * Saves single pages to a list, for future reuse.
  1650. * (This is not used for multi-page overflow pages.)
  1651. */
  1652. static void
  1653. mdb_page_free(MDB_env *env, MDB_page *mp)
  1654. {
  1655. mp->mp_next = env->me_dpages;
  1656. VGMEMP_FREE(env, mp);
  1657. env->me_dpages = mp;
  1658. }
  1659. /** Free a dirty page */
  1660. static void
  1661. mdb_dpage_free(MDB_env *env, MDB_page *dp)
  1662. {
  1663. if (!IS_OVERFLOW(dp) || dp->mp_pages == 1) {
  1664. mdb_page_free(env, dp);
  1665. } else {
  1666. /* large pages just get freed directly */
  1667. VGMEMP_FREE(env, dp);
  1668. free(dp);
  1669. }
  1670. }
  1671. /** Return all dirty pages to dpage list */
  1672. static void
  1673. mdb_dlist_free(MDB_txn *txn)
  1674. {
  1675. MDB_env *env = txn->mt_env;
  1676. MDB_ID2L dl = txn->mt_u.dirty_list;
  1677. unsigned i, n = dl[0].mid;
  1678. for (i = 1; i <= n; i++) {
  1679. mdb_dpage_free(env, dl[i].mptr);
  1680. }
  1681. dl[0].mid = 0;
  1682. }
  1683. /** Loosen or free a single page.
  1684. * Saves single pages to a list for future reuse
  1685. * in this same txn. It has been pulled from the freeDB
  1686. * and already resides on the dirty list, but has been
  1687. * deleted. Use these pages first before pulling again
  1688. * from the freeDB.
  1689. *
  1690. * If the page wasn't dirtied in this txn, just add it
  1691. * to this txn's free list.
  1692. */
  1693. static int
  1694. mdb_page_loose(MDB_cursor *mc, MDB_page *mp)
  1695. {
  1696. int loose = 0;
  1697. pgno_t pgno = mp->mp_pgno;
  1698. MDB_txn *txn = mc->mc_txn;
  1699. if ((mp->mp_flags & P_DIRTY) && mc->mc_dbi != FREE_DBI) {
  1700. if (txn->mt_parent) {
  1701. MDB_ID2 *dl = txn->mt_u.dirty_list;
  1702. /* If txn has a parent, make sure the page is in our
  1703. * dirty list.
  1704. */
  1705. if (dl[0].mid) {
  1706. unsigned x = mdb_mid2l_search(dl, pgno);
  1707. if (x <= dl[0].mid && dl[x].mid == pgno) {
  1708. if (mp != dl[x].mptr) { /* bad cursor? */
  1709. mc->mc_flags &= ~(C_INITIALIZED|C_EOF);
  1710. txn->mt_flags |= MDB_TXN_ERROR;
  1711. return MDB_CORRUPTED;
  1712. }
  1713. /* ok, it's ours */
  1714. loose = 1;
  1715. }
  1716. }
  1717. } else {
  1718. /* no parent txn, so it's just ours */
  1719. loose = 1;
  1720. }
  1721. }
  1722. if (loose) {
  1723. DPRINTF(("loosen db %d page %"Z"u", DDBI(mc),
  1724. mp->mp_pgno));
  1725. NEXT_LOOSE_PAGE(mp) = txn->mt_loose_pgs;
  1726. txn->mt_loose_pgs = mp;
  1727. txn->mt_loose_count++;
  1728. mp->mp_flags |= P_LOOSE;
  1729. } else {
  1730. int rc = mdb_midl_append(&txn->mt_free_pgs, pgno);
  1731. if (rc)
  1732. return rc;
  1733. }
  1734. return MDB_SUCCESS;
  1735. }
  1736. /** Set or clear P_KEEP in dirty, non-overflow, non-sub pages watched by txn.
  1737. * @param[in] mc A cursor handle for the current operation.
  1738. * @param[in] pflags Flags of the pages to update:
  1739. * P_DIRTY to set P_KEEP, P_DIRTY|P_KEEP to clear it.
  1740. * @param[in] all No shortcuts. Needed except after a full #mdb_page_flush().
  1741. * @return 0 on success, non-zero on failure.
  1742. */
  1743. static int
  1744. mdb_pages_xkeep(MDB_cursor *mc, unsigned pflags, int all)
  1745. {
  1746. enum { Mask = P_SUBP|P_DIRTY|P_LOOSE|P_KEEP };
  1747. MDB_txn *txn = mc->mc_txn;
  1748. MDB_cursor *m3, *m0 = mc;
  1749. MDB_xcursor *mx;
  1750. MDB_page *dp, *mp;
  1751. MDB_node *leaf;
  1752. unsigned i, j;
  1753. int rc = MDB_SUCCESS, level;
  1754. /* Mark pages seen by cursors */
  1755. if (mc->mc_flags & C_UNTRACK)
  1756. mc = NULL; /* will find mc in mt_cursors */
  1757. for (i = txn->mt_numdbs;; mc = txn->mt_cursors[--i]) {
  1758. for (; mc; mc=mc->mc_next) {
  1759. if (!(mc->mc_flags & C_INITIALIZED))
  1760. continue;
  1761. for (m3 = mc;; m3 = &mx->mx_cursor) {
  1762. mp = NULL;
  1763. for (j=0; j<m3->mc_snum; j++) {
  1764. mp = m3->mc_pg[j];
  1765. if ((mp->mp_flags & Mask) == pflags)
  1766. mp->mp_flags ^= P_KEEP;
  1767. }
  1768. mx = m3->mc_xcursor;
  1769. /* Proceed to mx if it is at a sub-database */
  1770. if (! (mx && (mx->mx_cursor.mc_flags & C_INITIALIZED)))
  1771. break;
  1772. if (! (mp && (mp->mp_flags & P_LEAF)))
  1773. break;
  1774. leaf = NODEPTR(mp, m3->mc_ki[j-1]);
  1775. if (!(leaf->mn_flags & F_SUBDATA))
  1776. break;
  1777. }
  1778. }
  1779. if (i == 0)
  1780. break;
  1781. }
  1782. if (all) {
  1783. /* Mark dirty root pages */
  1784. for (i=0; i<txn->mt_numdbs; i++) {
  1785. if (txn->mt_dbflags[i] & DB_DIRTY) {
  1786. pgno_t pgno = txn->mt_dbs[i].md_root;
  1787. if (pgno == P_INVALID)
  1788. continue;
  1789. if ((rc = mdb_page_get(m0, pgno, &dp, &level)) != MDB_SUCCESS)
  1790. break;
  1791. if ((dp->mp_flags & Mask) == pflags && level <= 1)
  1792. dp->mp_flags ^= P_KEEP;
  1793. }
  1794. }
  1795. }
  1796. return rc;
  1797. }
  1798. static int mdb_page_flush(MDB_txn *txn, int keep);
  1799. /** Spill pages from the dirty list back to disk.
  1800. * This is intended to prevent running into #MDB_TXN_FULL situations,
  1801. * but note that they may still occur in a few cases:
  1802. * 1) our estimate of the txn size could be too small. Currently this
  1803. * seems unlikely, except with a large number of #MDB_MULTIPLE items.
  1804. * 2) child txns may run out of space if their parents dirtied a
  1805. * lot of pages and never spilled them. TODO: we probably should do
  1806. * a preemptive spill during #mdb_txn_begin() of a child txn, if
  1807. * the parent's dirty_room is below a given threshold.
  1808. *
  1809. * Otherwise, if not using nested txns, it is expected that apps will
  1810. * not run into #MDB_TXN_FULL any more. The pages are flushed to disk
  1811. * the same way as for a txn commit, e.g. their P_DIRTY flag is cleared.
  1812. * If the txn never references them again, they can be left alone.
  1813. * If the txn only reads them, they can be used without any fuss.
  1814. * If the txn writes them again, they can be dirtied immediately without
  1815. * going thru all of the work of #mdb_page_touch(). Such references are
  1816. * handled by #mdb_page_unspill().
  1817. *
  1818. * Also note, we never spill DB root pages, nor pages of active cursors,
  1819. * because we'll need these back again soon anyway. And in nested txns,
  1820. * we can't spill a page in a child txn if it was already spilled in a
  1821. * parent txn. That would alter the parent txns' data even though
  1822. * the child hasn't committed yet, and we'd have no way to undo it if
  1823. * the child aborted.
  1824. *
  1825. * @param[in] m0 cursor A cursor handle identifying the transaction and
  1826. * database for which we are checking space.
  1827. * @param[in] key For a put operation, the key being stored.
  1828. * @param[in] data For a put operation, the data being stored.
  1829. * @return 0 on success, non-zero on failure.
  1830. */
  1831. static int
  1832. mdb_page_spill(MDB_cursor *m0, MDB_val *key, MDB_val *data)
  1833. {
  1834. MDB_txn *txn = m0->mc_txn;
  1835. MDB_page *dp;
  1836. MDB_ID2L dl = txn->mt_u.dirty_list;
  1837. unsigned int i, j, need;
  1838. int rc;
  1839. if (m0->mc_flags & C_SUB)
  1840. return MDB_SUCCESS;
  1841. /* Estimate how much space this op will take */
  1842. i = m0->mc_db->md_depth;
  1843. /* Named DBs also dirty the main DB */
  1844. if (m0->mc_dbi >= CORE_DBS)
  1845. i += txn->mt_dbs[MAIN_DBI].md_depth;
  1846. /* For puts, roughly factor in the key+data size */
  1847. if (key)
  1848. i += (LEAFSIZE(key, data) + txn->mt_env->me_psize) / txn->mt_env->me_psize;
  1849. i += i; /* double it for good measure */
  1850. need = i;
  1851. if (txn->mt_dirty_room > i)
  1852. return MDB_SUCCESS;
  1853. if (!txn->mt_spill_pgs) {
  1854. txn->mt_spill_pgs = mdb_midl_alloc(MDB_IDL_UM_MAX);
  1855. if (!txn->mt_spill_pgs)
  1856. return ENOMEM;
  1857. } else {
  1858. /* purge deleted slots */
  1859. MDB_IDL sl = txn->mt_spill_pgs;
  1860. unsigned int num = sl[0];
  1861. j=0;
  1862. for (i=1; i<=num; i++) {
  1863. if (!(sl[i] & 1))
  1864. sl[++j] = sl[i];
  1865. }
  1866. sl[0] = j;
  1867. }
  1868. /* Preserve pages which may soon be dirtied again */
  1869. if ((rc = mdb_pages_xkeep(m0, P_DIRTY, 1)) != MDB_SUCCESS)
  1870. goto done;
  1871. /* Less aggressive spill - we originally spilled the entire dirty list,
  1872. * with a few exceptions for cursor pages and DB root pages. But this
  1873. * turns out to be a lot of wasted effort because in a large txn many
  1874. * of those pages will need to be used again. So now we spill only 1/8th
  1875. * of the dirty pages. Testing revealed this to be a good tradeoff,
  1876. * better than 1/2, 1/4, or 1/10.
  1877. */
  1878. if (need < MDB_IDL_UM_MAX / 8)
  1879. need = MDB_IDL_UM_MAX / 8;
  1880. /* Save the page IDs of all the pages we're flushing */
  1881. /* flush from the tail forward, this saves a lot of shifting later on. */
  1882. for (i=dl[0].mid; i && need; i--) {
  1883. MDB_ID pn = dl[i].mid << 1;
  1884. dp = dl[i].mptr;
  1885. if (dp->mp_flags & (P_LOOSE|P_KEEP))
  1886. continue;
  1887. /* Can't spill twice, make sure it's not already in a parent's
  1888. * spill list.
  1889. */
  1890. if (txn->mt_parent) {
  1891. MDB_txn *tx2;
  1892. for (tx2 = txn->mt_parent; tx2; tx2 = tx2->mt_parent) {
  1893. if (tx2->mt_spill_pgs) {
  1894. j = mdb_midl_search(tx2->mt_spill_pgs, pn);
  1895. if (j <= tx2->mt_spill_pgs[0] && tx2->mt_spill_pgs[j] == pn) {
  1896. dp->mp_flags |= P_KEEP;
  1897. break;
  1898. }
  1899. }
  1900. }
  1901. if (tx2)
  1902. continue;
  1903. }
  1904. if ((rc = mdb_midl_append(&txn->mt_spill_pgs, pn)))
  1905. goto done;
  1906. need--;
  1907. }
  1908. mdb_midl_sort(txn->mt_spill_pgs);
  1909. /* Flush the spilled part of dirty list */
  1910. if ((rc = mdb_page_flush(txn, i)) != MDB_SUCCESS)
  1911. goto done;
  1912. /* Reset any dirty pages we kept that page_flush didn't see */
  1913. rc = mdb_pages_xkeep(m0, P_DIRTY|P_KEEP, i);
  1914. done:
  1915. txn->mt_flags |= rc ? MDB_TXN_ERROR : MDB_TXN_SPILLS;
  1916. return rc;
  1917. }
  1918. /** Find oldest txnid still referenced. Expects txn->mt_txnid > 0. */
  1919. static txnid_t
  1920. mdb_find_oldest(MDB_txn *txn)
  1921. {
  1922. int i;
  1923. txnid_t mr, oldest = txn->mt_txnid - 1;
  1924. if (txn->mt_env->me_txns) {
  1925. MDB_reader *r = txn->mt_env->me_txns->mti_readers;
  1926. for (i = txn->mt_env->me_txns->mti_numreaders; --i >= 0; ) {
  1927. if (r[i].mr_pid) {
  1928. mr = r[i].mr_txnid;
  1929. if (oldest > mr)
  1930. oldest = mr;
  1931. }
  1932. }
  1933. }
  1934. return oldest;
  1935. }
  1936. /** Add a page to the txn's dirty list */
  1937. static void
  1938. mdb_page_dirty(MDB_txn *txn, MDB_page *mp)
  1939. {
  1940. MDB_ID2 mid;
  1941. int rc, (*insert)(MDB_ID2L, MDB_ID2 *);
  1942. if (txn->mt_flags & MDB_TXN_WRITEMAP) {
  1943. insert = mdb_mid2l_append;
  1944. } else {
  1945. insert = mdb_mid2l_insert;
  1946. }
  1947. mid.mid = mp->mp_pgno;
  1948. mid.mptr = mp;
  1949. rc = insert(txn->mt_u.dirty_list, &mid);
  1950. mdb_tassert(txn, rc == 0);
  1951. txn->mt_dirty_room--;
  1952. }
  1953. /** Allocate page numbers and memory for writing. Maintain me_pglast,
  1954. * me_pghead and mt_next_pgno. Set #MDB_TXN_ERROR on failure.
  1955. *
  1956. * If there are free pages available from older transactions, they
  1957. * are re-used first. Otherwise allocate a new page at mt_next_pgno.
  1958. * Do not modify the freedB, just merge freeDB records into me_pghead[]
  1959. * and move me_pglast to say which records were consumed. Only this
  1960. * function can create me_pghead and move me_pglast/mt_next_pgno.
  1961. * @param[in] mc cursor A cursor handle identifying the transaction and
  1962. * database for which we are allocating.
  1963. * @param[in] num the number of pages to allocate.
  1964. * @param[out] mp Address of the allocated page(s). Requests for multiple pages
  1965. * will always be satisfied by a single contiguous chunk of memory.
  1966. * @return 0 on success, non-zero on failure.
  1967. */
  1968. static int
  1969. mdb_page_alloc(MDB_cursor *mc, int num, MDB_page **mp)
  1970. {
  1971. #ifdef MDB_PARANOID /* Seems like we can ignore this now */
  1972. /* Get at most <Max_retries> more freeDB records once me_pghead
  1973. * has enough pages. If not enough, use new pages from the map.
  1974. * If <Paranoid> and mc is updating the freeDB, only get new
  1975. * records if me_pghead is empty. Then the freelist cannot play
  1976. * catch-up with itself by growing while trying to save it.
  1977. */
  1978. enum { Paranoid = 1, Max_retries = 500 };
  1979. #else
  1980. enum { Paranoid = 0, Max_retries = INT_MAX /*infinite*/ };
  1981. #endif
  1982. int rc, retry = num * 60;
  1983. MDB_txn *txn = mc->mc_txn;
  1984. MDB_env *env = txn->mt_env;
  1985. pgno_t pgno, *mop = env->me_pghead;
  1986. unsigned i, j, mop_len = mop ? mop[0] : 0, n2 = num-1;
  1987. MDB_page *np;
  1988. txnid_t oldest = 0, last;
  1989. MDB_cursor_op op;
  1990. MDB_cursor m2;
  1991. int found_old = 0;
  1992. /* If there are any loose pages, just use them */
  1993. if (num == 1 && txn->mt_loose_pgs) {
  1994. np = txn->mt_loose_pgs;
  1995. txn->mt_loose_pgs = NEXT_LOOSE_PAGE(np);
  1996. txn->mt_loose_count--;
  1997. DPRINTF(("db %d use loose page %"Z"u", DDBI(mc),
  1998. np->mp_pgno));
  1999. *mp = np;
  2000. return MDB_SUCCESS;
  2001. }
  2002. *mp = NULL;
  2003. /* If our dirty list is already full, we can't do anything */
  2004. if (txn->mt_dirty_room == 0) {
  2005. rc = MDB_TXN_FULL;
  2006. goto fail;
  2007. }
  2008. for (op = MDB_FIRST;; op = MDB_NEXT) {
  2009. MDB_val key, data;
  2010. MDB_node *leaf;
  2011. pgno_t *idl;
  2012. /* Seek a big enough contiguous page range. Prefer
  2013. * pages at the tail, just truncating the list.
  2014. */
  2015. if (mop_len > n2) {
  2016. i = mop_len;
  2017. do {
  2018. pgno = mop[i];
  2019. if (mop[i-n2] == pgno+n2)
  2020. goto search_done;
  2021. } while (--i > n2);
  2022. if (--retry < 0)
  2023. break;
  2024. }
  2025. if (op == MDB_FIRST) { /* 1st iteration */
  2026. /* Prepare to fetch more and coalesce */
  2027. last = env->me_pglast;
  2028. oldest = env->me_pgoldest;
  2029. mdb_cursor_init(&m2, txn, FREE_DBI, NULL);
  2030. if (last) {
  2031. op = MDB_SET_RANGE;
  2032. key.mv_data = &last; /* will look up last+1 */
  2033. key.mv_size = sizeof(last);
  2034. }
  2035. if (Paranoid && mc->mc_dbi == FREE_DBI)
  2036. retry = -1;
  2037. }
  2038. if (Paranoid && retry < 0 && mop_len)
  2039. break;
  2040. last++;
  2041. /* Do not fetch more if the record will be too recent */
  2042. if (oldest <= last) {
  2043. if (!found_old) {
  2044. oldest = mdb_find_oldest(txn);
  2045. env->me_pgoldest = oldest;
  2046. found_old = 1;
  2047. }
  2048. if (oldest <= last)
  2049. break;
  2050. }
  2051. rc = mdb_cursor_get(&m2, &key, NULL, op);
  2052. if (rc) {
  2053. if (rc == MDB_NOTFOUND)
  2054. break;
  2055. goto fail;
  2056. }
  2057. last = *(txnid_t*)key.mv_data;
  2058. if (oldest <= last) {
  2059. if (!found_old) {
  2060. oldest = mdb_find_oldest(txn);
  2061. env->me_pgoldest = oldest;
  2062. found_old = 1;
  2063. }
  2064. if (oldest <= last)
  2065. break;
  2066. }
  2067. np = m2.mc_pg[m2.mc_top];
  2068. leaf = NODEPTR(np, m2.mc_ki[m2.mc_top]);
  2069. if ((rc = mdb_node_read(&m2, leaf, &data)) != MDB_SUCCESS)
  2070. goto fail;
  2071. idl = (MDB_ID *) data.mv_data;
  2072. i = idl[0];
  2073. if (!mop) {
  2074. if (!(env->me_pghead = mop = mdb_midl_alloc(i))) {
  2075. rc = ENOMEM;
  2076. goto fail;
  2077. }
  2078. } else {
  2079. if ((rc = mdb_midl_need(&env->me_pghead, i)) != 0)
  2080. goto fail;
  2081. mop = env->me_pghead;
  2082. }
  2083. env->me_pglast = last;
  2084. #if (MDB_DEBUG) > 1
  2085. DPRINTF(("IDL read txn %"Z"u root %"Z"u num %u",
  2086. last, txn->mt_dbs[FREE_DBI].md_root, i));
  2087. for (j = i; j; j--)
  2088. DPRINTF(("IDL %"Z"u", idl[j]));
  2089. #endif
  2090. /* Merge in descending sorted order */
  2091. mdb_midl_xmerge(mop, idl);
  2092. mop_len = mop[0];
  2093. }
  2094. /* Use new pages from the map when nothing suitable in the freeDB */
  2095. i = 0;
  2096. pgno = txn->mt_next_pgno;
  2097. if (pgno + num >= env->me_maxpg) {
  2098. DPUTS("DB size maxed out");
  2099. rc = MDB_MAP_FULL;
  2100. goto fail;
  2101. }
  2102. search_done:
  2103. if (env->me_flags & MDB_WRITEMAP) {
  2104. np = (MDB_page *)(env->me_map + env->me_psize * pgno);
  2105. } else {
  2106. if (!(np = mdb_page_malloc(txn, num))) {
  2107. rc = ENOMEM;
  2108. goto fail;
  2109. }
  2110. }
  2111. if (i) {
  2112. mop[0] = mop_len -= num;
  2113. /* Move any stragglers down */
  2114. for (j = i-num; j < mop_len; )
  2115. mop[++j] = mop[++i];
  2116. } else {
  2117. txn->mt_next_pgno = pgno + num;
  2118. }
  2119. np->mp_pgno = pgno;
  2120. mdb_page_dirty(txn, np);
  2121. *mp = np;
  2122. return MDB_SUCCESS;
  2123. fail:
  2124. txn->mt_flags |= MDB_TXN_ERROR;
  2125. return rc;
  2126. }
  2127. /** Copy the used portions of a non-overflow page.
  2128. * @param[in] dst page to copy into
  2129. * @param[in] src page to copy from
  2130. * @param[in] psize size of a page
  2131. */
  2132. static void
  2133. mdb_page_copy(MDB_page *dst, MDB_page *src, unsigned int psize)
  2134. {
  2135. enum { Align = sizeof(pgno_t) };
  2136. indx_t upper = src->mp_upper, lower = src->mp_lower, unused = upper-lower;
  2137. /* If page isn't full, just copy the used portion. Adjust
  2138. * alignment so memcpy may copy words instead of bytes.
  2139. */
  2140. if ((unused &= -Align) && !IS_LEAF2(src)) {
  2141. upper = (upper + PAGEBASE) & -Align;
  2142. memcpy(dst, src, (lower + PAGEBASE + (Align-1)) & -Align);
  2143. memcpy((pgno_t *)((char *)dst+upper), (pgno_t *)((char *)src+upper),
  2144. psize - upper);
  2145. } else {
  2146. memcpy(dst, src, psize - unused);
  2147. }
  2148. }
  2149. /** Pull a page off the txn's spill list, if present.
  2150. * If a page being referenced was spilled to disk in this txn, bring
  2151. * it back and make it dirty/writable again.
  2152. * @param[in] txn the transaction handle.
  2153. * @param[in] mp the page being referenced. It must not be dirty.
  2154. * @param[out] ret the writable page, if any. ret is unchanged if
  2155. * mp wasn't spilled.
  2156. */
  2157. static int
  2158. mdb_page_unspill(MDB_txn *txn, MDB_page *mp, MDB_page **ret)
  2159. {
  2160. MDB_env *env = txn->mt_env;
  2161. const MDB_txn *tx2;
  2162. unsigned x;
  2163. pgno_t pgno = mp->mp_pgno, pn = pgno << 1;
  2164. for (tx2 = txn; tx2; tx2=tx2->mt_parent) {
  2165. if (!tx2->mt_spill_pgs)
  2166. continue;
  2167. x = mdb_midl_search(tx2->mt_spill_pgs, pn);
  2168. if (x <= tx2->mt_spill_pgs[0] && tx2->mt_spill_pgs[x] == pn) {
  2169. MDB_page *np;
  2170. int num;
  2171. if (txn->mt_dirty_room == 0)
  2172. return MDB_TXN_FULL;
  2173. if (IS_OVERFLOW(mp))
  2174. num = mp->mp_pages;
  2175. else
  2176. num = 1;
  2177. if (env->me_flags & MDB_WRITEMAP) {
  2178. np = mp;
  2179. } else {
  2180. np = mdb_page_malloc(txn, num);
  2181. if (!np)
  2182. return ENOMEM;
  2183. if (num > 1)
  2184. memcpy(np, mp, num * env->me_psize);
  2185. else
  2186. mdb_page_copy(np, mp, env->me_psize);
  2187. }
  2188. if (tx2 == txn) {
  2189. /* If in current txn, this page is no longer spilled.
  2190. * If it happens to be the last page, truncate the spill list.
  2191. * Otherwise mark it as deleted by setting the LSB.
  2192. */
  2193. if (x == txn->mt_spill_pgs[0])
  2194. txn->mt_spill_pgs[0]--;
  2195. else
  2196. txn->mt_spill_pgs[x] |= 1;
  2197. } /* otherwise, if belonging to a parent txn, the
  2198. * page remains spilled until child commits
  2199. */
  2200. mdb_page_dirty(txn, np);
  2201. np->mp_flags |= P_DIRTY;
  2202. *ret = np;
  2203. break;
  2204. }
  2205. }
  2206. return MDB_SUCCESS;
  2207. }
  2208. /** Touch a page: make it dirty and re-insert into tree with updated pgno.
  2209. * Set #MDB_TXN_ERROR on failure.
  2210. * @param[in] mc cursor pointing to the page to be touched
  2211. * @return 0 on success, non-zero on failure.
  2212. */
  2213. static int
  2214. mdb_page_touch(MDB_cursor *mc)
  2215. {
  2216. MDB_page *mp = mc->mc_pg[mc->mc_top], *np;
  2217. MDB_txn *txn = mc->mc_txn;
  2218. MDB_cursor *m2, *m3;
  2219. pgno_t pgno;
  2220. int rc;
  2221. if (!F_ISSET(mp->mp_flags, P_DIRTY)) {
  2222. if (txn->mt_flags & MDB_TXN_SPILLS) {
  2223. np = NULL;
  2224. rc = mdb_page_unspill(txn, mp, &np);
  2225. if (rc)
  2226. goto fail;
  2227. if (np)
  2228. goto done;
  2229. }
  2230. if ((rc = mdb_midl_need(&txn->mt_free_pgs, 1)) ||
  2231. (rc = mdb_page_alloc(mc, 1, &np)))
  2232. goto fail;
  2233. pgno = np->mp_pgno;
  2234. DPRINTF(("touched db %d page %"Z"u -> %"Z"u", DDBI(mc),
  2235. mp->mp_pgno, pgno));
  2236. mdb_cassert(mc, mp->mp_pgno != pgno);
  2237. mdb_midl_xappend(txn->mt_free_pgs, mp->mp_pgno);
  2238. /* Update the parent page, if any, to point to the new page */
  2239. if (mc->mc_top) {
  2240. MDB_page *parent = mc->mc_pg[mc->mc_top-1];
  2241. MDB_node *node = NODEPTR(parent, mc->mc_ki[mc->mc_top-1]);
  2242. SETPGNO(node, pgno);
  2243. } else {
  2244. mc->mc_db->md_root = pgno;
  2245. }
  2246. } else if (txn->mt_parent && !IS_SUBP(mp)) {
  2247. MDB_ID2 mid, *dl = txn->mt_u.dirty_list;
  2248. pgno = mp->mp_pgno;
  2249. /* If txn has a parent, make sure the page is in our
  2250. * dirty list.
  2251. */
  2252. if (dl[0].mid) {
  2253. unsigned x = mdb_mid2l_search(dl, pgno);
  2254. if (x <= dl[0].mid && dl[x].mid == pgno) {
  2255. if (mp != dl[x].mptr) { /* bad cursor? */
  2256. mc->mc_flags &= ~(C_INITIALIZED|C_EOF);
  2257. txn->mt_flags |= MDB_TXN_ERROR;
  2258. return MDB_CORRUPTED;
  2259. }
  2260. return 0;
  2261. }
  2262. }
  2263. mdb_cassert(mc, dl[0].mid < MDB_IDL_UM_MAX);
  2264. /* No - copy it */
  2265. np = mdb_page_malloc(txn, 1);
  2266. if (!np)
  2267. return ENOMEM;
  2268. mid.mid = pgno;
  2269. mid.mptr = np;
  2270. rc = mdb_mid2l_insert(dl, &mid);
  2271. mdb_cassert(mc, rc == 0);
  2272. } else {
  2273. return 0;
  2274. }
  2275. mdb_page_copy(np, mp, txn->mt_env->me_psize);
  2276. np->mp_pgno = pgno;
  2277. np->mp_flags |= P_DIRTY;
  2278. done:
  2279. /* Adjust cursors pointing to mp */
  2280. mc->mc_pg[mc->mc_top] = np;
  2281. m2 = txn->mt_cursors[mc->mc_dbi];
  2282. if (mc->mc_flags & C_SUB) {
  2283. for (; m2; m2=m2->mc_next) {
  2284. m3 = &m2->mc_xcursor->mx_cursor;
  2285. if (m3->mc_snum < mc->mc_snum) continue;
  2286. if (m3->mc_pg[mc->mc_top] == mp)
  2287. m3->mc_pg[mc->mc_top] = np;
  2288. }
  2289. } else {
  2290. for (; m2; m2=m2->mc_next) {
  2291. if (m2->mc_snum < mc->mc_snum) continue;
  2292. if (m2 == mc) continue;
  2293. if (m2->mc_pg[mc->mc_top] == mp) {
  2294. m2->mc_pg[mc->mc_top] = np;
  2295. if (IS_LEAF(np))
  2296. XCURSOR_REFRESH(m2, mc->mc_top, np);
  2297. }
  2298. }
  2299. }
  2300. return 0;
  2301. fail:
  2302. txn->mt_flags |= MDB_TXN_ERROR;
  2303. return rc;
  2304. }
  2305. int
  2306. mdb_env_sync(MDB_env *env, int force)
  2307. {
  2308. int rc = 0;
  2309. if (env->me_flags & MDB_RDONLY)
  2310. return EACCES;
  2311. if (force || !F_ISSET(env->me_flags, MDB_NOSYNC)) {
  2312. if (env->me_flags & MDB_WRITEMAP) {
  2313. int flags = ((env->me_flags & MDB_MAPASYNC) && !force)
  2314. ? MS_ASYNC : MS_SYNC;
  2315. if (MDB_MSYNC(env->me_map, env->me_mapsize, flags))
  2316. rc = ErrCode();
  2317. #ifdef _WIN32
  2318. else if (flags == MS_SYNC && MDB_FDATASYNC(env->me_fd))
  2319. rc = ErrCode();
  2320. #endif
  2321. } else {
  2322. #ifdef BROKEN_FDATASYNC
  2323. if (env->me_flags & MDB_FSYNCONLY) {
  2324. if (fsync(env->me_fd))
  2325. rc = ErrCode();
  2326. } else
  2327. #endif
  2328. if (MDB_FDATASYNC(env->me_fd))
  2329. rc = ErrCode();
  2330. }
  2331. }
  2332. return rc;
  2333. }
  2334. /** Back up parent txn's cursors, then grab the originals for tracking */
  2335. static int
  2336. mdb_cursor_shadow(MDB_txn *src, MDB_txn *dst)
  2337. {
  2338. MDB_cursor *mc, *bk;
  2339. MDB_xcursor *mx;
  2340. size_t size;
  2341. int i;
  2342. for (i = src->mt_numdbs; --i >= 0; ) {
  2343. if ((mc = src->mt_cursors[i]) != NULL) {
  2344. size = sizeof(MDB_cursor);
  2345. if (mc->mc_xcursor)
  2346. size += sizeof(MDB_xcursor);
  2347. for (; mc; mc = bk->mc_next) {
  2348. bk = malloc(size);
  2349. if (!bk)
  2350. return ENOMEM;
  2351. *bk = *mc;
  2352. mc->mc_backup = bk;
  2353. mc->mc_db = &dst->mt_dbs[i];
  2354. /* Kill pointers into src to reduce abuse: The
  2355. * user may not use mc until dst ends. But we need a valid
  2356. * txn pointer here for cursor fixups to keep working.
  2357. */
  2358. mc->mc_txn = dst;
  2359. mc->mc_dbflag = &dst->mt_dbflags[i];
  2360. if ((mx = mc->mc_xcursor) != NULL) {
  2361. *(MDB_xcursor *)(bk+1) = *mx;
  2362. mx->mx_cursor.mc_txn = dst;
  2363. }
  2364. mc->mc_next = dst->mt_cursors[i];
  2365. dst->mt_cursors[i] = mc;
  2366. }
  2367. }
  2368. }
  2369. return MDB_SUCCESS;
  2370. }
  2371. /** Close this write txn's cursors, give parent txn's cursors back to parent.
  2372. * @param[in] txn the transaction handle.
  2373. * @param[in] merge true to keep changes to parent cursors, false to revert.
  2374. * @return 0 on success, non-zero on failure.
  2375. */
  2376. static void
  2377. mdb_cursors_close(MDB_txn *txn, unsigned merge)
  2378. {
  2379. MDB_cursor **cursors = txn->mt_cursors, *mc, *next, *bk;
  2380. MDB_xcursor *mx;
  2381. int i;
  2382. for (i = txn->mt_numdbs; --i >= 0; ) {
  2383. for (mc = cursors[i]; mc; mc = next) {
  2384. next = mc->mc_next;
  2385. if ((bk = mc->mc_backup) != NULL) {
  2386. if (merge) {
  2387. /* Commit changes to parent txn */
  2388. mc->mc_next = bk->mc_next;
  2389. mc->mc_backup = bk->mc_backup;
  2390. mc->mc_txn = bk->mc_txn;
  2391. mc->mc_db = bk->mc_db;
  2392. mc->mc_dbflag = bk->mc_dbflag;
  2393. if ((mx = mc->mc_xcursor) != NULL)
  2394. mx->mx_cursor.mc_txn = bk->mc_txn;
  2395. } else {
  2396. /* Abort nested txn */
  2397. *mc = *bk;
  2398. if ((mx = mc->mc_xcursor) != NULL)
  2399. *mx = *(MDB_xcursor *)(bk+1);
  2400. }
  2401. mc = bk;
  2402. }
  2403. /* Only malloced cursors are permanently tracked. */
  2404. free(mc);
  2405. }
  2406. cursors[i] = NULL;
  2407. }
  2408. }
  2409. #if !(MDB_PIDLOCK) /* Currently the same as defined(_WIN32) */
  2410. enum Pidlock_op {
  2411. Pidset, Pidcheck
  2412. };
  2413. #else
  2414. enum Pidlock_op {
  2415. Pidset = F_SETLK, Pidcheck = F_GETLK
  2416. };
  2417. #endif
  2418. /** Set or check a pid lock. Set returns 0 on success.
  2419. * Check returns 0 if the process is certainly dead, nonzero if it may
  2420. * be alive (the lock exists or an error happened so we do not know).
  2421. *
  2422. * On Windows Pidset is a no-op, we merely check for the existence
  2423. * of the process with the given pid. On POSIX we use a single byte
  2424. * lock on the lockfile, set at an offset equal to the pid.
  2425. */
  2426. static int
  2427. mdb_reader_pid(MDB_env *env, enum Pidlock_op op, MDB_PID_T pid)
  2428. {
  2429. #if !(MDB_PIDLOCK) /* Currently the same as defined(_WIN32) */
  2430. int ret = 0;
  2431. HANDLE h;
  2432. if (op == Pidcheck) {
  2433. h = OpenProcess(env->me_pidquery, FALSE, pid);
  2434. /* No documented "no such process" code, but other program use this: */
  2435. if (!h)
  2436. return ErrCode() != ERROR_INVALID_PARAMETER;
  2437. /* A process exists until all handles to it close. Has it exited? */
  2438. ret = WaitForSingleObject(h, 0) != 0;
  2439. CloseHandle(h);
  2440. }
  2441. return ret;
  2442. #else
  2443. for (;;) {
  2444. int rc;
  2445. struct flock lock_info;
  2446. memset(&lock_info, 0, sizeof(lock_info));
  2447. lock_info.l_type = F_WRLCK;
  2448. lock_info.l_whence = SEEK_SET;
  2449. lock_info.l_start = pid;
  2450. lock_info.l_len = 1;
  2451. if ((rc = fcntl(env->me_lfd, op, &lock_info)) == 0) {
  2452. if (op == F_GETLK && lock_info.l_type != F_UNLCK)
  2453. rc = -1;
  2454. } else if ((rc = ErrCode()) == EINTR) {
  2455. continue;
  2456. }
  2457. return rc;
  2458. }
  2459. #endif
  2460. }
  2461. /** Common code for #mdb_txn_begin() and #mdb_txn_renew().
  2462. * @param[in] txn the transaction handle to initialize
  2463. * @return 0 on success, non-zero on failure.
  2464. */
  2465. static int
  2466. mdb_txn_renew0(MDB_txn *txn)
  2467. {
  2468. MDB_env *env = txn->mt_env;
  2469. MDB_txninfo *ti = env->me_txns;
  2470. MDB_meta *meta;
  2471. unsigned int i, nr, flags = txn->mt_flags;
  2472. uint16_t x;
  2473. int rc, new_notls = 0;
  2474. if ((flags &= MDB_TXN_RDONLY) != 0) {
  2475. if (!ti) {
  2476. meta = mdb_env_pick_meta(env);
  2477. txn->mt_txnid = meta->mm_txnid;
  2478. txn->mt_u.reader = NULL;
  2479. } else {
  2480. MDB_reader *r = (env->me_flags & MDB_NOTLS) ? txn->mt_u.reader :
  2481. pthread_getspecific(env->me_txkey);
  2482. if (r) {
  2483. if (r->mr_pid != env->me_pid || r->mr_txnid != (txnid_t)-1)
  2484. return MDB_BAD_RSLOT;
  2485. } else {
  2486. MDB_PID_T pid = env->me_pid;
  2487. MDB_THR_T tid = pthread_self();
  2488. mdb_mutexref_t rmutex = env->me_rmutex;
  2489. if (!env->me_live_reader) {
  2490. rc = mdb_reader_pid(env, Pidset, pid);
  2491. if (rc)
  2492. return rc;
  2493. env->me_live_reader = 1;
  2494. }
  2495. if (LOCK_MUTEX(rc, env, rmutex))
  2496. return rc;
  2497. nr = ti->mti_numreaders;
  2498. for (i=0; i<nr; i++)
  2499. if (ti->mti_readers[i].mr_pid == 0)
  2500. break;
  2501. if (i == env->me_maxreaders) {
  2502. UNLOCK_MUTEX(rmutex);
  2503. return MDB_READERS_FULL;
  2504. }
  2505. r = &ti->mti_readers[i];
  2506. /* Claim the reader slot, carefully since other code
  2507. * uses the reader table un-mutexed: First reset the
  2508. * slot, next publish it in mti_numreaders. After
  2509. * that, it is safe for mdb_env_close() to touch it.
  2510. * When it will be closed, we can finally claim it.
  2511. */
  2512. r->mr_pid = 0;
  2513. r->mr_txnid = (txnid_t)-1;
  2514. r->mr_tid = tid;
  2515. if (i == nr)
  2516. ti->mti_numreaders = ++nr;
  2517. env->me_close_readers = nr;
  2518. r->mr_pid = pid;
  2519. UNLOCK_MUTEX(rmutex);
  2520. new_notls = (env->me_flags & MDB_NOTLS);
  2521. if (!new_notls && (rc=pthread_setspecific(env->me_txkey, r))) {
  2522. r->mr_pid = 0;
  2523. return rc;
  2524. }
  2525. }
  2526. do /* LY: Retry on a race, ITS#7970. */
  2527. r->mr_txnid = ti->mti_txnid;
  2528. while(r->mr_txnid != ti->mti_txnid);
  2529. txn->mt_txnid = r->mr_txnid;
  2530. txn->mt_u.reader = r;
  2531. meta = env->me_metas[txn->mt_txnid & 1];
  2532. }
  2533. } else {
  2534. /* Not yet touching txn == env->me_txn0, it may be active */
  2535. if (ti) {
  2536. if (LOCK_MUTEX(rc, env, env->me_wmutex))
  2537. return rc;
  2538. txn->mt_txnid = ti->mti_txnid;
  2539. meta = env->me_metas[txn->mt_txnid & 1];
  2540. } else {
  2541. meta = mdb_env_pick_meta(env);
  2542. txn->mt_txnid = meta->mm_txnid;
  2543. }
  2544. txn->mt_txnid++;
  2545. #if MDB_DEBUG
  2546. if (txn->mt_txnid == mdb_debug_start)
  2547. mdb_debug = 1;
  2548. #endif
  2549. txn->mt_child = NULL;
  2550. txn->mt_loose_pgs = NULL;
  2551. txn->mt_loose_count = 0;
  2552. txn->mt_dirty_room = MDB_IDL_UM_MAX;
  2553. txn->mt_u.dirty_list = env->me_dirty_list;
  2554. txn->mt_u.dirty_list[0].mid = 0;
  2555. txn->mt_free_pgs = env->me_free_pgs;
  2556. txn->mt_free_pgs[0] = 0;
  2557. txn->mt_spill_pgs = NULL;
  2558. env->me_txn = txn;
  2559. memcpy(txn->mt_dbiseqs, env->me_dbiseqs, env->me_maxdbs * sizeof(unsigned int));
  2560. }
  2561. /* Copy the DB info and flags */
  2562. memcpy(txn->mt_dbs, meta->mm_dbs, CORE_DBS * sizeof(MDB_db));
  2563. /* Moved to here to avoid a data race in read TXNs */
  2564. txn->mt_next_pgno = meta->mm_last_pg+1;
  2565. txn->mt_flags = flags;
  2566. /* Setup db info */
  2567. txn->mt_numdbs = env->me_numdbs;
  2568. for (i=CORE_DBS; i<txn->mt_numdbs; i++) {
  2569. x = env->me_dbflags[i];
  2570. txn->mt_dbs[i].md_flags = x & PERSISTENT_FLAGS;
  2571. txn->mt_dbflags[i] = (x & MDB_VALID) ? DB_VALID|DB_USRVALID|DB_STALE : 0;
  2572. }
  2573. txn->mt_dbflags[MAIN_DBI] = DB_VALID|DB_USRVALID;
  2574. txn->mt_dbflags[FREE_DBI] = DB_VALID;
  2575. if (env->me_flags & MDB_FATAL_ERROR) {
  2576. DPUTS("environment had fatal error, must shutdown!");
  2577. rc = MDB_PANIC;
  2578. } else if (env->me_maxpg < txn->mt_next_pgno) {
  2579. rc = MDB_MAP_RESIZED;
  2580. } else {
  2581. return MDB_SUCCESS;
  2582. }
  2583. mdb_txn_end(txn, new_notls /*0 or MDB_END_SLOT*/ | MDB_END_FAIL_BEGIN);
  2584. return rc;
  2585. }
  2586. int
  2587. mdb_txn_renew(MDB_txn *txn)
  2588. {
  2589. int rc;
  2590. if (!txn || !F_ISSET(txn->mt_flags, MDB_TXN_RDONLY|MDB_TXN_FINISHED))
  2591. return EINVAL;
  2592. rc = mdb_txn_renew0(txn);
  2593. if (rc == MDB_SUCCESS) {
  2594. DPRINTF(("renew txn %"Z"u%c %p on mdbenv %p, root page %"Z"u",
  2595. txn->mt_txnid, (txn->mt_flags & MDB_TXN_RDONLY) ? 'r' : 'w',
  2596. (void *)txn, (void *)txn->mt_env, txn->mt_dbs[MAIN_DBI].md_root));
  2597. }
  2598. return rc;
  2599. }
  2600. int
  2601. mdb_txn_begin(MDB_env *env, MDB_txn *parent, unsigned int flags, MDB_txn **ret)
  2602. {
  2603. MDB_txn *txn;
  2604. MDB_ntxn *ntxn;
  2605. int rc, size, tsize;
  2606. flags &= MDB_TXN_BEGIN_FLAGS;
  2607. flags |= env->me_flags & MDB_WRITEMAP;
  2608. if (env->me_flags & MDB_RDONLY & ~flags) /* write txn in RDONLY env */
  2609. return EACCES;
  2610. if (parent) {
  2611. /* Nested transactions: Max 1 child, write txns only, no writemap */
  2612. flags |= parent->mt_flags;
  2613. if (flags & (MDB_RDONLY|MDB_WRITEMAP|MDB_TXN_BLOCKED)) {
  2614. return (parent->mt_flags & MDB_TXN_RDONLY) ? EINVAL : MDB_BAD_TXN;
  2615. }
  2616. /* Child txns save MDB_pgstate and use own copy of cursors */
  2617. size = env->me_maxdbs * (sizeof(MDB_db)+sizeof(MDB_cursor *)+1);
  2618. size += tsize = sizeof(MDB_ntxn);
  2619. } else if (flags & MDB_RDONLY) {
  2620. size = env->me_maxdbs * (sizeof(MDB_db)+1);
  2621. size += tsize = sizeof(MDB_txn);
  2622. } else {
  2623. /* Reuse preallocated write txn. However, do not touch it until
  2624. * mdb_txn_renew0() succeeds, since it currently may be active.
  2625. */
  2626. txn = env->me_txn0;
  2627. goto renew;
  2628. }
  2629. if ((txn = calloc(1, size)) == NULL) {
  2630. DPRINTF(("calloc: %s", strerror(errno)));
  2631. return ENOMEM;
  2632. }
  2633. txn->mt_dbxs = env->me_dbxs; /* static */
  2634. txn->mt_dbs = (MDB_db *) ((char *)txn + tsize);
  2635. txn->mt_dbflags = (unsigned char *)txn + size - env->me_maxdbs;
  2636. txn->mt_flags = flags;
  2637. txn->mt_env = env;
  2638. if (parent) {
  2639. unsigned int i;
  2640. txn->mt_cursors = (MDB_cursor **)(txn->mt_dbs + env->me_maxdbs);
  2641. txn->mt_dbiseqs = parent->mt_dbiseqs;
  2642. txn->mt_u.dirty_list = malloc(sizeof(MDB_ID2)*MDB_IDL_UM_SIZE);
  2643. if (!txn->mt_u.dirty_list ||
  2644. !(txn->mt_free_pgs = mdb_midl_alloc(MDB_IDL_UM_MAX)))
  2645. {
  2646. free(txn->mt_u.dirty_list);
  2647. free(txn);
  2648. return ENOMEM;
  2649. }
  2650. txn->mt_txnid = parent->mt_txnid;
  2651. txn->mt_dirty_room = parent->mt_dirty_room;
  2652. txn->mt_u.dirty_list[0].mid = 0;
  2653. txn->mt_spill_pgs = NULL;
  2654. txn->mt_next_pgno = parent->mt_next_pgno;
  2655. parent->mt_flags |= MDB_TXN_HAS_CHILD;
  2656. parent->mt_child = txn;
  2657. txn->mt_parent = parent;
  2658. txn->mt_numdbs = parent->mt_numdbs;
  2659. memcpy(txn->mt_dbs, parent->mt_dbs, txn->mt_numdbs * sizeof(MDB_db));
  2660. /* Copy parent's mt_dbflags, but clear DB_NEW */
  2661. for (i=0; i<txn->mt_numdbs; i++)
  2662. txn->mt_dbflags[i] = parent->mt_dbflags[i] & ~DB_NEW;
  2663. rc = 0;
  2664. ntxn = (MDB_ntxn *)txn;
  2665. ntxn->mnt_pgstate = env->me_pgstate; /* save parent me_pghead & co */
  2666. if (env->me_pghead) {
  2667. size = MDB_IDL_SIZEOF(env->me_pghead);
  2668. env->me_pghead = mdb_midl_alloc(env->me_pghead[0]);
  2669. if (env->me_pghead)
  2670. memcpy(env->me_pghead, ntxn->mnt_pgstate.mf_pghead, size);
  2671. else
  2672. rc = ENOMEM;
  2673. }
  2674. if (!rc)
  2675. rc = mdb_cursor_shadow(parent, txn);
  2676. if (rc)
  2677. mdb_txn_end(txn, MDB_END_FAIL_BEGINCHILD);
  2678. } else { /* MDB_RDONLY */
  2679. txn->mt_dbiseqs = env->me_dbiseqs;
  2680. renew:
  2681. rc = mdb_txn_renew0(txn);
  2682. }
  2683. if (rc) {
  2684. if (txn != env->me_txn0)
  2685. free(txn);
  2686. } else {
  2687. txn->mt_flags |= flags; /* could not change txn=me_txn0 earlier */
  2688. *ret = txn;
  2689. DPRINTF(("begin txn %"Z"u%c %p on mdbenv %p, root page %"Z"u",
  2690. txn->mt_txnid, (flags & MDB_RDONLY) ? 'r' : 'w',
  2691. (void *) txn, (void *) env, txn->mt_dbs[MAIN_DBI].md_root));
  2692. }
  2693. return rc;
  2694. }
  2695. MDB_env *
  2696. mdb_txn_env(MDB_txn *txn)
  2697. {
  2698. if(!txn) return NULL;
  2699. return txn->mt_env;
  2700. }
  2701. size_t
  2702. mdb_txn_id(MDB_txn *txn)
  2703. {
  2704. if(!txn) return 0;
  2705. return txn->mt_txnid;
  2706. }
  2707. /** Export or close DBI handles opened in this txn. */
  2708. static void
  2709. mdb_dbis_update(MDB_txn *txn, int keep)
  2710. {
  2711. int i;
  2712. MDB_dbi n = txn->mt_numdbs;
  2713. MDB_env *env = txn->mt_env;
  2714. unsigned char *tdbflags = txn->mt_dbflags;
  2715. for (i = n; --i >= CORE_DBS;) {
  2716. if (tdbflags[i] & DB_NEW) {
  2717. if (keep) {
  2718. env->me_dbflags[i] = txn->mt_dbs[i].md_flags | MDB_VALID;
  2719. } else {
  2720. char *ptr = env->me_dbxs[i].md_name.mv_data;
  2721. if (ptr) {
  2722. env->me_dbxs[i].md_name.mv_data = NULL;
  2723. env->me_dbxs[i].md_name.mv_size = 0;
  2724. env->me_dbflags[i] = 0;
  2725. env->me_dbiseqs[i]++;
  2726. free(ptr);
  2727. }
  2728. }
  2729. }
  2730. }
  2731. if (keep && env->me_numdbs < n)
  2732. env->me_numdbs = n;
  2733. }
  2734. /** End a transaction, except successful commit of a nested transaction.
  2735. * May be called twice for readonly txns: First reset it, then abort.
  2736. * @param[in] txn the transaction handle to end
  2737. * @param[in] mode why and how to end the transaction
  2738. */
  2739. static void
  2740. mdb_txn_end(MDB_txn *txn, unsigned mode)
  2741. {
  2742. MDB_env *env = txn->mt_env;
  2743. #if MDB_DEBUG
  2744. static const char *const names[] = MDB_END_NAMES;
  2745. #endif
  2746. /* Export or close DBI handles opened in this txn */
  2747. mdb_dbis_update(txn, mode & MDB_END_UPDATE);
  2748. DPRINTF(("%s txn %"Z"u%c %p on mdbenv %p, root page %"Z"u",
  2749. names[mode & MDB_END_OPMASK],
  2750. txn->mt_txnid, (txn->mt_flags & MDB_TXN_RDONLY) ? 'r' : 'w',
  2751. (void *) txn, (void *)env, txn->mt_dbs[MAIN_DBI].md_root));
  2752. if (F_ISSET(txn->mt_flags, MDB_TXN_RDONLY)) {
  2753. if (txn->mt_u.reader) {
  2754. txn->mt_u.reader->mr_txnid = (txnid_t)-1;
  2755. if (!(env->me_flags & MDB_NOTLS)) {
  2756. txn->mt_u.reader = NULL; /* txn does not own reader */
  2757. } else if (mode & MDB_END_SLOT) {
  2758. txn->mt_u.reader->mr_pid = 0;
  2759. txn->mt_u.reader = NULL;
  2760. } /* else txn owns the slot until it does MDB_END_SLOT */
  2761. }
  2762. txn->mt_numdbs = 0; /* prevent further DBI activity */
  2763. txn->mt_flags |= MDB_TXN_FINISHED;
  2764. } else if (!F_ISSET(txn->mt_flags, MDB_TXN_FINISHED)) {
  2765. pgno_t *pghead = env->me_pghead;
  2766. if (!(mode & MDB_END_UPDATE)) /* !(already closed cursors) */
  2767. mdb_cursors_close(txn, 0);
  2768. if (!(env->me_flags & MDB_WRITEMAP)) {
  2769. mdb_dlist_free(txn);
  2770. }
  2771. txn->mt_numdbs = 0;
  2772. txn->mt_flags = MDB_TXN_FINISHED;
  2773. if (!txn->mt_parent) {
  2774. mdb_midl_shrink(&txn->mt_free_pgs);
  2775. env->me_free_pgs = txn->mt_free_pgs;
  2776. /* me_pgstate: */
  2777. env->me_pghead = NULL;
  2778. env->me_pglast = 0;
  2779. env->me_txn = NULL;
  2780. mode = 0; /* txn == env->me_txn0, do not free() it */
  2781. /* The writer mutex was locked in mdb_txn_begin. */
  2782. if (env->me_txns)
  2783. UNLOCK_MUTEX(env->me_wmutex);
  2784. } else {
  2785. txn->mt_parent->mt_child = NULL;
  2786. txn->mt_parent->mt_flags &= ~MDB_TXN_HAS_CHILD;
  2787. env->me_pgstate = ((MDB_ntxn *)txn)->mnt_pgstate;
  2788. mdb_midl_free(txn->mt_free_pgs);
  2789. free(txn->mt_u.dirty_list);
  2790. }
  2791. mdb_midl_free(txn->mt_spill_pgs);
  2792. mdb_midl_free(pghead);
  2793. }
  2794. if (mode & MDB_END_FREE)
  2795. free(txn);
  2796. }
  2797. void
  2798. mdb_txn_reset(MDB_txn *txn)
  2799. {
  2800. if (txn == NULL)
  2801. return;
  2802. /* This call is only valid for read-only txns */
  2803. if (!(txn->mt_flags & MDB_TXN_RDONLY))
  2804. return;
  2805. mdb_txn_end(txn, MDB_END_RESET);
  2806. }
  2807. void
  2808. mdb_txn_abort(MDB_txn *txn)
  2809. {
  2810. if (txn == NULL)
  2811. return;
  2812. if (txn->mt_child)
  2813. mdb_txn_abort(txn->mt_child);
  2814. mdb_txn_end(txn, MDB_END_ABORT|MDB_END_SLOT|MDB_END_FREE);
  2815. }
  2816. /** Save the freelist as of this transaction to the freeDB.
  2817. * This changes the freelist. Keep trying until it stabilizes.
  2818. */
  2819. static int
  2820. mdb_freelist_save(MDB_txn *txn)
  2821. {
  2822. /* env->me_pghead[] can grow and shrink during this call.
  2823. * env->me_pglast and txn->mt_free_pgs[] can only grow.
  2824. * Page numbers cannot disappear from txn->mt_free_pgs[].
  2825. */
  2826. MDB_cursor mc;
  2827. MDB_env *env = txn->mt_env;
  2828. int rc, maxfree_1pg = env->me_maxfree_1pg, more = 1;
  2829. txnid_t pglast = 0, head_id = 0;
  2830. pgno_t freecnt = 0, *free_pgs, *mop;
  2831. ssize_t head_room = 0, total_room = 0, mop_len, clean_limit;
  2832. mdb_cursor_init(&mc, txn, FREE_DBI, NULL);
  2833. if (env->me_pghead) {
  2834. /* Make sure first page of freeDB is touched and on freelist */
  2835. rc = mdb_page_search(&mc, NULL, MDB_PS_FIRST|MDB_PS_MODIFY);
  2836. if (rc && rc != MDB_NOTFOUND)
  2837. return rc;
  2838. }
  2839. if (!env->me_pghead && txn->mt_loose_pgs) {
  2840. /* Put loose page numbers in mt_free_pgs, since
  2841. * we may be unable to return them to me_pghead.
  2842. */
  2843. MDB_page *mp = txn->mt_loose_pgs;
  2844. MDB_ID2 *dl = txn->mt_u.dirty_list;
  2845. unsigned x;
  2846. if ((rc = mdb_midl_need(&txn->mt_free_pgs, txn->mt_loose_count)) != 0)
  2847. return rc;
  2848. for (; mp; mp = NEXT_LOOSE_PAGE(mp)) {
  2849. mdb_midl_xappend(txn->mt_free_pgs, mp->mp_pgno);
  2850. /* must also remove from dirty list */
  2851. if (txn->mt_flags & MDB_TXN_WRITEMAP) {
  2852. for (x=1; x<=dl[0].mid; x++)
  2853. if (dl[x].mid == mp->mp_pgno)
  2854. break;
  2855. mdb_tassert(txn, x <= dl[0].mid);
  2856. } else {
  2857. x = mdb_mid2l_search(dl, mp->mp_pgno);
  2858. mdb_tassert(txn, dl[x].mid == mp->mp_pgno);
  2859. mdb_dpage_free(env, mp);
  2860. }
  2861. dl[x].mptr = NULL;
  2862. }
  2863. {
  2864. /* squash freed slots out of the dirty list */
  2865. unsigned y;
  2866. for (y=1; dl[y].mptr && y <= dl[0].mid; y++);
  2867. if (y <= dl[0].mid) {
  2868. for(x=y, y++;;) {
  2869. while (!dl[y].mptr && y <= dl[0].mid) y++;
  2870. if (y > dl[0].mid) break;
  2871. dl[x++] = dl[y++];
  2872. }
  2873. dl[0].mid = x-1;
  2874. } else {
  2875. /* all slots freed */
  2876. dl[0].mid = 0;
  2877. }
  2878. }
  2879. txn->mt_loose_pgs = NULL;
  2880. txn->mt_loose_count = 0;
  2881. }
  2882. /* MDB_RESERVE cancels meminit in ovpage malloc (when no WRITEMAP) */
  2883. clean_limit = (env->me_flags & (MDB_NOMEMINIT|MDB_WRITEMAP))
  2884. ? SSIZE_MAX : maxfree_1pg;
  2885. for (;;) {
  2886. /* Come back here after each Put() in case freelist changed */
  2887. MDB_val key, data;
  2888. pgno_t *pgs;
  2889. ssize_t j;
  2890. /* If using records from freeDB which we have not yet
  2891. * deleted, delete them and any we reserved for me_pghead.
  2892. */
  2893. while (pglast < env->me_pglast) {
  2894. rc = mdb_cursor_first(&mc, &key, NULL);
  2895. if (rc)
  2896. return rc;
  2897. pglast = head_id = *(txnid_t *)key.mv_data;
  2898. total_room = head_room = 0;
  2899. mdb_tassert(txn, pglast <= env->me_pglast);
  2900. rc = mdb_cursor_del(&mc, 0);
  2901. if (rc)
  2902. return rc;
  2903. }
  2904. /* Save the IDL of pages freed by this txn, to a single record */
  2905. if (freecnt < txn->mt_free_pgs[0]) {
  2906. if (!freecnt) {
  2907. /* Make sure last page of freeDB is touched and on freelist */
  2908. rc = mdb_page_search(&mc, NULL, MDB_PS_LAST|MDB_PS_MODIFY);
  2909. if (rc && rc != MDB_NOTFOUND)
  2910. return rc;
  2911. }
  2912. free_pgs = txn->mt_free_pgs;
  2913. /* Write to last page of freeDB */
  2914. key.mv_size = sizeof(txn->mt_txnid);
  2915. key.mv_data = &txn->mt_txnid;
  2916. do {
  2917. freecnt = free_pgs[0];
  2918. data.mv_size = MDB_IDL_SIZEOF(free_pgs);
  2919. rc = mdb_cursor_put(&mc, &key, &data, MDB_RESERVE);
  2920. if (rc)
  2921. return rc;
  2922. /* Retry if mt_free_pgs[] grew during the Put() */
  2923. free_pgs = txn->mt_free_pgs;
  2924. } while (freecnt < free_pgs[0]);
  2925. mdb_midl_sort(free_pgs);
  2926. memcpy(data.mv_data, free_pgs, data.mv_size);
  2927. #if (MDB_DEBUG) > 1
  2928. {
  2929. unsigned int i = free_pgs[0];
  2930. DPRINTF(("IDL write txn %"Z"u root %"Z"u num %u",
  2931. txn->mt_txnid, txn->mt_dbs[FREE_DBI].md_root, i));
  2932. for (; i; i--)
  2933. DPRINTF(("IDL %"Z"u", free_pgs[i]));
  2934. }
  2935. #endif
  2936. continue;
  2937. }
  2938. mop = env->me_pghead;
  2939. mop_len = (mop ? mop[0] : 0) + txn->mt_loose_count;
  2940. /* Reserve records for me_pghead[]. Split it if multi-page,
  2941. * to avoid searching freeDB for a page range. Use keys in
  2942. * range [1,me_pglast]: Smaller than txnid of oldest reader.
  2943. */
  2944. if (total_room >= mop_len) {
  2945. if (total_room == mop_len || --more < 0)
  2946. break;
  2947. } else if (head_room >= maxfree_1pg && head_id > 1) {
  2948. /* Keep current record (overflow page), add a new one */
  2949. head_id--;
  2950. head_room = 0;
  2951. }
  2952. /* (Re)write {key = head_id, IDL length = head_room} */
  2953. total_room -= head_room;
  2954. head_room = mop_len - total_room;
  2955. if (head_room > maxfree_1pg && head_id > 1) {
  2956. /* Overflow multi-page for part of me_pghead */
  2957. head_room /= head_id; /* amortize page sizes */
  2958. head_room += maxfree_1pg - head_room % (maxfree_1pg + 1);
  2959. } else if (head_room < 0) {
  2960. /* Rare case, not bothering to delete this record */
  2961. head_room = 0;
  2962. }
  2963. key.mv_size = sizeof(head_id);
  2964. key.mv_data = &head_id;
  2965. data.mv_size = (head_room + 1) * sizeof(pgno_t);
  2966. rc = mdb_cursor_put(&mc, &key, &data, MDB_RESERVE);
  2967. if (rc)
  2968. return rc;
  2969. /* IDL is initially empty, zero out at least the length */
  2970. pgs = (pgno_t *)data.mv_data;
  2971. j = head_room > clean_limit ? head_room : 0;
  2972. do {
  2973. pgs[j] = 0;
  2974. } while (--j >= 0);
  2975. total_room += head_room;
  2976. }
  2977. /* Return loose page numbers to me_pghead, though usually none are
  2978. * left at this point. The pages themselves remain in dirty_list.
  2979. */
  2980. if (txn->mt_loose_pgs) {
  2981. MDB_page *mp = txn->mt_loose_pgs;
  2982. unsigned count = txn->mt_loose_count;
  2983. MDB_IDL loose;
  2984. /* Room for loose pages + temp IDL with same */
  2985. if ((rc = mdb_midl_need(&env->me_pghead, 2*count+1)) != 0)
  2986. return rc;
  2987. mop = env->me_pghead;
  2988. loose = mop + MDB_IDL_ALLOCLEN(mop) - count;
  2989. for (count = 0; mp; mp = NEXT_LOOSE_PAGE(mp))
  2990. loose[ ++count ] = mp->mp_pgno;
  2991. loose[0] = count;
  2992. mdb_midl_sort(loose);
  2993. mdb_midl_xmerge(mop, loose);
  2994. txn->mt_loose_pgs = NULL;
  2995. txn->mt_loose_count = 0;
  2996. mop_len = mop[0];
  2997. }
  2998. /* Fill in the reserved me_pghead records */
  2999. rc = MDB_SUCCESS;
  3000. if (mop_len) {
  3001. MDB_val key, data;
  3002. mop += mop_len;
  3003. rc = mdb_cursor_first(&mc, &key, &data);
  3004. for (; !rc; rc = mdb_cursor_next(&mc, &key, &data, MDB_NEXT)) {
  3005. txnid_t id = *(txnid_t *)key.mv_data;
  3006. ssize_t len = (ssize_t)(data.mv_size / sizeof(MDB_ID)) - 1;
  3007. MDB_ID save;
  3008. mdb_tassert(txn, len >= 0 && id <= env->me_pglast);
  3009. key.mv_data = &id;
  3010. if (len > mop_len) {
  3011. len = mop_len;
  3012. data.mv_size = (len + 1) * sizeof(MDB_ID);
  3013. }
  3014. data.mv_data = mop -= len;
  3015. save = mop[0];
  3016. mop[0] = len;
  3017. rc = mdb_cursor_put(&mc, &key, &data, MDB_CURRENT);
  3018. mop[0] = save;
  3019. if (rc || !(mop_len -= len))
  3020. break;
  3021. }
  3022. }
  3023. return rc;
  3024. }
  3025. /** Flush (some) dirty pages to the map, after clearing their dirty flag.
  3026. * @param[in] txn the transaction that's being committed
  3027. * @param[in] keep number of initial pages in dirty_list to keep dirty.
  3028. * @return 0 on success, non-zero on failure.
  3029. */
  3030. static int
  3031. mdb_page_flush(MDB_txn *txn, int keep)
  3032. {
  3033. MDB_env *env = txn->mt_env;
  3034. MDB_ID2L dl = txn->mt_u.dirty_list;
  3035. unsigned psize = env->me_psize, j;
  3036. int i, pagecount = dl[0].mid, rc;
  3037. size_t size = 0, pos = 0;
  3038. pgno_t pgno = 0;
  3039. MDB_page *dp = NULL;
  3040. #ifdef _WIN32
  3041. OVERLAPPED ov;
  3042. #else
  3043. struct iovec iov[MDB_COMMIT_PAGES];
  3044. ssize_t wpos = 0, wsize = 0, wres;
  3045. size_t next_pos = 1; /* impossible pos, so pos != next_pos */
  3046. int n = 0;
  3047. #endif
  3048. j = i = keep;
  3049. if (env->me_flags & MDB_WRITEMAP) {
  3050. /* Clear dirty flags */
  3051. while (++i <= pagecount) {
  3052. dp = dl[i].mptr;
  3053. /* Don't flush this page yet */
  3054. if (dp->mp_flags & (P_LOOSE|P_KEEP)) {
  3055. dp->mp_flags &= ~P_KEEP;
  3056. dl[++j] = dl[i];
  3057. continue;
  3058. }
  3059. dp->mp_flags &= ~P_DIRTY;
  3060. }
  3061. goto done;
  3062. }
  3063. /* Write the pages */
  3064. for (;;) {
  3065. if (++i <= pagecount) {
  3066. dp = dl[i].mptr;
  3067. /* Don't flush this page yet */
  3068. if (dp->mp_flags & (P_LOOSE|P_KEEP)) {
  3069. dp->mp_flags &= ~P_KEEP;
  3070. dl[i].mid = 0;
  3071. continue;
  3072. }
  3073. pgno = dl[i].mid;
  3074. /* clear dirty flag */
  3075. dp->mp_flags &= ~P_DIRTY;
  3076. pos = pgno * psize;
  3077. size = psize;
  3078. if (IS_OVERFLOW(dp)) size *= dp->mp_pages;
  3079. }
  3080. #ifdef _WIN32
  3081. else break;
  3082. /* Windows actually supports scatter/gather I/O, but only on
  3083. * unbuffered file handles. Since we're relying on the OS page
  3084. * cache for all our data, that's self-defeating. So we just
  3085. * write pages one at a time. We use the ov structure to set
  3086. * the write offset, to at least save the overhead of a Seek
  3087. * system call.
  3088. */
  3089. DPRINTF(("committing page %"Z"u", pgno));
  3090. memset(&ov, 0, sizeof(ov));
  3091. ov.Offset = pos & 0xffffffff;
  3092. ov.OffsetHigh = pos >> 16 >> 16;
  3093. if (!WriteFile(env->me_fd, dp, size, NULL, &ov)) {
  3094. rc = ErrCode();
  3095. DPRINTF(("WriteFile: %d", rc));
  3096. return rc;
  3097. }
  3098. #else
  3099. /* Write up to MDB_COMMIT_PAGES dirty pages at a time. */
  3100. if (pos!=next_pos || n==MDB_COMMIT_PAGES || wsize+size>MAX_WRITE) {
  3101. if (n) {
  3102. retry_write:
  3103. /* Write previous page(s) */
  3104. #ifdef MDB_USE_PWRITEV
  3105. wres = pwritev(env->me_fd, iov, n, wpos);
  3106. #else
  3107. if (n == 1) {
  3108. wres = pwrite(env->me_fd, iov[0].iov_base, wsize, wpos);
  3109. } else {
  3110. retry_seek:
  3111. if (lseek(env->me_fd, wpos, SEEK_SET) == -1) {
  3112. rc = ErrCode();
  3113. if (rc == EINTR)
  3114. goto retry_seek;
  3115. DPRINTF(("lseek: %s", strerror(rc)));
  3116. return rc;
  3117. }
  3118. wres = writev(env->me_fd, iov, n);
  3119. }
  3120. #endif
  3121. if (wres != wsize) {
  3122. if (wres < 0) {
  3123. rc = ErrCode();
  3124. if (rc == EINTR)
  3125. goto retry_write;
  3126. DPRINTF(("Write error: %s", strerror(rc)));
  3127. } else {
  3128. rc = EIO; /* TODO: Use which error code? */
  3129. DPUTS("short write, filesystem full?");
  3130. }
  3131. return rc;
  3132. }
  3133. n = 0;
  3134. }
  3135. if (i > pagecount)
  3136. break;
  3137. wpos = pos;
  3138. wsize = 0;
  3139. }
  3140. DPRINTF(("committing page %"Z"u", pgno));
  3141. next_pos = pos + size;
  3142. iov[n].iov_len = size;
  3143. iov[n].iov_base = (char *)dp;
  3144. wsize += size;
  3145. n++;
  3146. #endif /* _WIN32 */
  3147. }
  3148. /* MIPS has cache coherency issues, this is a no-op everywhere else
  3149. * Note: for any size >= on-chip cache size, entire on-chip cache is
  3150. * flushed.
  3151. */
  3152. CACHEFLUSH(env->me_map, txn->mt_next_pgno * env->me_psize, DCACHE);
  3153. for (i = keep; ++i <= pagecount; ) {
  3154. dp = dl[i].mptr;
  3155. /* This is a page we skipped above */
  3156. if (!dl[i].mid) {
  3157. dl[++j] = dl[i];
  3158. dl[j].mid = dp->mp_pgno;
  3159. continue;
  3160. }
  3161. mdb_dpage_free(env, dp);
  3162. }
  3163. done:
  3164. i--;
  3165. txn->mt_dirty_room += i - j;
  3166. dl[0].mid = j;
  3167. return MDB_SUCCESS;
  3168. }
  3169. int
  3170. mdb_txn_commit(MDB_txn *txn)
  3171. {
  3172. int rc;
  3173. unsigned int i, end_mode;
  3174. MDB_env *env;
  3175. if (txn == NULL)
  3176. return EINVAL;
  3177. /* mdb_txn_end() mode for a commit which writes nothing */
  3178. end_mode = MDB_END_EMPTY_COMMIT|MDB_END_UPDATE|MDB_END_SLOT|MDB_END_FREE;
  3179. if (txn->mt_child) {
  3180. rc = mdb_txn_commit(txn->mt_child);
  3181. if (rc)
  3182. goto fail;
  3183. }
  3184. env = txn->mt_env;
  3185. if (F_ISSET(txn->mt_flags, MDB_TXN_RDONLY)) {
  3186. goto done;
  3187. }
  3188. if (txn->mt_flags & (MDB_TXN_FINISHED|MDB_TXN_ERROR)) {
  3189. DPUTS("txn has failed/finished, can't commit");
  3190. if (txn->mt_parent)
  3191. txn->mt_parent->mt_flags |= MDB_TXN_ERROR;
  3192. rc = MDB_BAD_TXN;
  3193. goto fail;
  3194. }
  3195. if (txn->mt_parent) {
  3196. MDB_txn *parent = txn->mt_parent;
  3197. MDB_page **lp;
  3198. MDB_ID2L dst, src;
  3199. MDB_IDL pspill;
  3200. unsigned x, y, len, ps_len;
  3201. /* Append our free list to parent's */
  3202. rc = mdb_midl_append_list(&parent->mt_free_pgs, txn->mt_free_pgs);
  3203. if (rc)
  3204. goto fail;
  3205. mdb_midl_free(txn->mt_free_pgs);
  3206. /* Failures after this must either undo the changes
  3207. * to the parent or set MDB_TXN_ERROR in the parent.
  3208. */
  3209. parent->mt_next_pgno = txn->mt_next_pgno;
  3210. parent->mt_flags = txn->mt_flags;
  3211. /* Merge our cursors into parent's and close them */
  3212. mdb_cursors_close(txn, 1);
  3213. /* Update parent's DB table. */
  3214. memcpy(parent->mt_dbs, txn->mt_dbs, txn->mt_numdbs * sizeof(MDB_db));
  3215. parent->mt_numdbs = txn->mt_numdbs;
  3216. parent->mt_dbflags[FREE_DBI] = txn->mt_dbflags[FREE_DBI];
  3217. parent->mt_dbflags[MAIN_DBI] = txn->mt_dbflags[MAIN_DBI];
  3218. for (i=CORE_DBS; i<txn->mt_numdbs; i++) {
  3219. /* preserve parent's DB_NEW status */
  3220. x = parent->mt_dbflags[i] & DB_NEW;
  3221. parent->mt_dbflags[i] = txn->mt_dbflags[i] | x;
  3222. }
  3223. dst = parent->mt_u.dirty_list;
  3224. src = txn->mt_u.dirty_list;
  3225. /* Remove anything in our dirty list from parent's spill list */
  3226. if ((pspill = parent->mt_spill_pgs) && (ps_len = pspill[0])) {
  3227. x = y = ps_len;
  3228. pspill[0] = (pgno_t)-1;
  3229. /* Mark our dirty pages as deleted in parent spill list */
  3230. for (i=0, len=src[0].mid; ++i <= len; ) {
  3231. MDB_ID pn = src[i].mid << 1;
  3232. while (pn > pspill[x])
  3233. x--;
  3234. if (pn == pspill[x]) {
  3235. pspill[x] = 1;
  3236. y = --x;
  3237. }
  3238. }
  3239. /* Squash deleted pagenums if we deleted any */
  3240. for (x=y; ++x <= ps_len; )
  3241. if (!(pspill[x] & 1))
  3242. pspill[++y] = pspill[x];
  3243. pspill[0] = y;
  3244. }
  3245. /* Remove anything in our spill list from parent's dirty list */
  3246. if (txn->mt_spill_pgs && txn->mt_spill_pgs[0]) {
  3247. for (i=1; i<=txn->mt_spill_pgs[0]; i++) {
  3248. MDB_ID pn = txn->mt_spill_pgs[i];
  3249. if (pn & 1)
  3250. continue; /* deleted spillpg */
  3251. pn >>= 1;
  3252. y = mdb_mid2l_search(dst, pn);
  3253. if (y <= dst[0].mid && dst[y].mid == pn) {
  3254. free(dst[y].mptr);
  3255. while (y < dst[0].mid) {
  3256. dst[y] = dst[y+1];
  3257. y++;
  3258. }
  3259. dst[0].mid--;
  3260. }
  3261. }
  3262. }
  3263. /* Find len = length of merging our dirty list with parent's */
  3264. x = dst[0].mid;
  3265. dst[0].mid = 0; /* simplify loops */
  3266. if (parent->mt_parent) {
  3267. len = x + src[0].mid;
  3268. y = mdb_mid2l_search(src, dst[x].mid + 1) - 1;
  3269. for (i = x; y && i; y--) {
  3270. pgno_t yp = src[y].mid;
  3271. while (yp < dst[i].mid)
  3272. i--;
  3273. if (yp == dst[i].mid) {
  3274. i--;
  3275. len--;
  3276. }
  3277. }
  3278. } else { /* Simplify the above for single-ancestor case */
  3279. len = MDB_IDL_UM_MAX - txn->mt_dirty_room;
  3280. }
  3281. /* Merge our dirty list with parent's */
  3282. y = src[0].mid;
  3283. for (i = len; y; dst[i--] = src[y--]) {
  3284. pgno_t yp = src[y].mid;
  3285. while (yp < dst[x].mid)
  3286. dst[i--] = dst[x--];
  3287. if (yp == dst[x].mid)
  3288. free(dst[x--].mptr);
  3289. }
  3290. mdb_tassert(txn, i == x);
  3291. dst[0].mid = len;
  3292. free(txn->mt_u.dirty_list);
  3293. parent->mt_dirty_room = txn->mt_dirty_room;
  3294. if (txn->mt_spill_pgs) {
  3295. if (parent->mt_spill_pgs) {
  3296. /* TODO: Prevent failure here, so parent does not fail */
  3297. rc = mdb_midl_append_list(&parent->mt_spill_pgs, txn->mt_spill_pgs);
  3298. if (rc)
  3299. parent->mt_flags |= MDB_TXN_ERROR;
  3300. mdb_midl_free(txn->mt_spill_pgs);
  3301. mdb_midl_sort(parent->mt_spill_pgs);
  3302. } else {
  3303. parent->mt_spill_pgs = txn->mt_spill_pgs;
  3304. }
  3305. }
  3306. /* Append our loose page list to parent's */
  3307. for (lp = &parent->mt_loose_pgs; *lp; lp = &NEXT_LOOSE_PAGE(*lp))
  3308. ;
  3309. *lp = txn->mt_loose_pgs;
  3310. parent->mt_loose_count += txn->mt_loose_count;
  3311. parent->mt_child = NULL;
  3312. mdb_midl_free(((MDB_ntxn *)txn)->mnt_pgstate.mf_pghead);
  3313. free(txn);
  3314. return rc;
  3315. }
  3316. if (txn != env->me_txn) {
  3317. DPUTS("attempt to commit unknown transaction");
  3318. rc = EINVAL;
  3319. goto fail;
  3320. }
  3321. mdb_cursors_close(txn, 0);
  3322. if (!txn->mt_u.dirty_list[0].mid &&
  3323. !(txn->mt_flags & (MDB_TXN_DIRTY|MDB_TXN_SPILLS)))
  3324. goto done;
  3325. DPRINTF(("committing txn %"Z"u %p on mdbenv %p, root page %"Z"u",
  3326. txn->mt_txnid, (void*)txn, (void*)env, txn->mt_dbs[MAIN_DBI].md_root));
  3327. /* Update DB root pointers */
  3328. if (txn->mt_numdbs > CORE_DBS) {
  3329. MDB_cursor mc;
  3330. MDB_dbi i;
  3331. MDB_val data;
  3332. data.mv_size = sizeof(MDB_db);
  3333. mdb_cursor_init(&mc, txn, MAIN_DBI, NULL);
  3334. for (i = CORE_DBS; i < txn->mt_numdbs; i++) {
  3335. if (txn->mt_dbflags[i] & DB_DIRTY) {
  3336. if (TXN_DBI_CHANGED(txn, i)) {
  3337. rc = MDB_BAD_DBI;
  3338. goto fail;
  3339. }
  3340. data.mv_data = &txn->mt_dbs[i];
  3341. rc = mdb_cursor_put(&mc, &txn->mt_dbxs[i].md_name, &data,
  3342. F_SUBDATA);
  3343. if (rc)
  3344. goto fail;
  3345. }
  3346. }
  3347. }
  3348. rc = mdb_freelist_save(txn);
  3349. if (rc)
  3350. goto fail;
  3351. mdb_midl_free(env->me_pghead);
  3352. env->me_pghead = NULL;
  3353. mdb_midl_shrink(&txn->mt_free_pgs);
  3354. #if (MDB_DEBUG) > 2
  3355. mdb_audit(txn);
  3356. #endif
  3357. if ((rc = mdb_page_flush(txn, 0)) ||
  3358. (rc = mdb_env_sync(env, 0)) ||
  3359. (rc = mdb_env_write_meta(txn)))
  3360. goto fail;
  3361. end_mode = MDB_END_COMMITTED|MDB_END_UPDATE;
  3362. done:
  3363. mdb_txn_end(txn, end_mode);
  3364. return MDB_SUCCESS;
  3365. fail:
  3366. mdb_txn_abort(txn);
  3367. return rc;
  3368. }
  3369. /** Read the environment parameters of a DB environment before
  3370. * mapping it into memory.
  3371. * @param[in] env the environment handle
  3372. * @param[out] meta address of where to store the meta information
  3373. * @return 0 on success, non-zero on failure.
  3374. */
  3375. static int ESECT
  3376. mdb_env_read_header(MDB_env *env, MDB_meta *meta)
  3377. {
  3378. MDB_metabuf pbuf;
  3379. MDB_page *p;
  3380. MDB_meta *m;
  3381. int i, rc, off;
  3382. enum { Size = sizeof(pbuf) };
  3383. /* We don't know the page size yet, so use a minimum value.
  3384. * Read both meta pages so we can use the latest one.
  3385. */
  3386. for (i=off=0; i<NUM_METAS; i++, off += meta->mm_psize) {
  3387. #ifdef _WIN32
  3388. DWORD len;
  3389. OVERLAPPED ov;
  3390. memset(&ov, 0, sizeof(ov));
  3391. ov.Offset = off;
  3392. rc = ReadFile(env->me_fd, &pbuf, Size, &len, &ov) ? (int)len : -1;
  3393. if (rc == -1 && ErrCode() == ERROR_HANDLE_EOF)
  3394. rc = 0;
  3395. #else
  3396. rc = pread(env->me_fd, &pbuf, Size, off);
  3397. #endif
  3398. if (rc != Size) {
  3399. if (rc == 0 && off == 0)
  3400. return ENOENT;
  3401. rc = rc < 0 ? (int) ErrCode() : MDB_INVALID;
  3402. DPRINTF(("read: %s", mdb_strerror(rc)));
  3403. return rc;
  3404. }
  3405. p = (MDB_page *)&pbuf;
  3406. if (!F_ISSET(p->mp_flags, P_META)) {
  3407. DPRINTF(("page %"Z"u not a meta page", p->mp_pgno));
  3408. return MDB_INVALID;
  3409. }
  3410. m = METADATA(p);
  3411. if (m->mm_magic != MDB_MAGIC) {
  3412. DPUTS("meta has invalid magic");
  3413. return MDB_INVALID;
  3414. }
  3415. if (m->mm_version != MDB_DATA_VERSION) {
  3416. DPRINTF(("database is version %u, expected version %u",
  3417. m->mm_version, MDB_DATA_VERSION));
  3418. return MDB_VERSION_MISMATCH;
  3419. }
  3420. if (off == 0 || m->mm_txnid > meta->mm_txnid)
  3421. *meta = *m;
  3422. }
  3423. return 0;
  3424. }
  3425. /** Fill in most of the zeroed #MDB_meta for an empty database environment */
  3426. static void ESECT
  3427. mdb_env_init_meta0(MDB_env *env, MDB_meta *meta)
  3428. {
  3429. meta->mm_magic = MDB_MAGIC;
  3430. meta->mm_version = MDB_DATA_VERSION;
  3431. meta->mm_mapsize = env->me_mapsize;
  3432. meta->mm_psize = env->me_psize;
  3433. meta->mm_last_pg = NUM_METAS-1;
  3434. meta->mm_flags = env->me_flags & 0xffff;
  3435. meta->mm_flags |= MDB_INTEGERKEY; /* this is mm_dbs[FREE_DBI].md_flags */
  3436. meta->mm_dbs[FREE_DBI].md_root = P_INVALID;
  3437. meta->mm_dbs[MAIN_DBI].md_root = P_INVALID;
  3438. }
  3439. /** Write the environment parameters of a freshly created DB environment.
  3440. * @param[in] env the environment handle
  3441. * @param[in] meta the #MDB_meta to write
  3442. * @return 0 on success, non-zero on failure.
  3443. */
  3444. static int ESECT
  3445. mdb_env_init_meta(MDB_env *env, MDB_meta *meta)
  3446. {
  3447. MDB_page *p, *q;
  3448. int rc;
  3449. unsigned int psize;
  3450. #ifdef _WIN32
  3451. DWORD len;
  3452. OVERLAPPED ov;
  3453. memset(&ov, 0, sizeof(ov));
  3454. #define DO_PWRITE(rc, fd, ptr, size, len, pos) do { \
  3455. ov.Offset = pos; \
  3456. rc = WriteFile(fd, ptr, size, &len, &ov); } while(0)
  3457. #else
  3458. int len;
  3459. #define DO_PWRITE(rc, fd, ptr, size, len, pos) do { \
  3460. len = pwrite(fd, ptr, size, pos); \
  3461. if (len == -1 && ErrCode() == EINTR) continue; \
  3462. rc = (len >= 0); break; } while(1)
  3463. #endif
  3464. DPUTS("writing new meta page");
  3465. psize = env->me_psize;
  3466. p = calloc(NUM_METAS, psize);
  3467. if (!p)
  3468. return ENOMEM;
  3469. p->mp_pgno = 0;
  3470. p->mp_flags = P_META;
  3471. *(MDB_meta *)METADATA(p) = *meta;
  3472. q = (MDB_page *)((char *)p + psize);
  3473. q->mp_pgno = 1;
  3474. q->mp_flags = P_META;
  3475. *(MDB_meta *)METADATA(q) = *meta;
  3476. DO_PWRITE(rc, env->me_fd, p, psize * NUM_METAS, len, 0);
  3477. if (!rc)
  3478. rc = ErrCode();
  3479. else if ((unsigned) len == psize * NUM_METAS)
  3480. rc = MDB_SUCCESS;
  3481. else
  3482. rc = ENOSPC;
  3483. free(p);
  3484. return rc;
  3485. }
  3486. /** Update the environment info to commit a transaction.
  3487. * @param[in] txn the transaction that's being committed
  3488. * @return 0 on success, non-zero on failure.
  3489. */
  3490. static int
  3491. mdb_env_write_meta(MDB_txn *txn)
  3492. {
  3493. MDB_env *env;
  3494. MDB_meta meta, metab, *mp;
  3495. unsigned flags;
  3496. size_t mapsize;
  3497. off_t off;
  3498. int rc, len, toggle;
  3499. char *ptr;
  3500. HANDLE mfd;
  3501. #ifdef _WIN32
  3502. OVERLAPPED ov;
  3503. #else
  3504. int r2;
  3505. #endif
  3506. toggle = txn->mt_txnid & 1;
  3507. DPRINTF(("writing meta page %d for root page %"Z"u",
  3508. toggle, txn->mt_dbs[MAIN_DBI].md_root));
  3509. env = txn->mt_env;
  3510. flags = env->me_flags;
  3511. mp = env->me_metas[toggle];
  3512. mapsize = env->me_metas[toggle ^ 1]->mm_mapsize;
  3513. /* Persist any increases of mapsize config */
  3514. if (mapsize < env->me_mapsize)
  3515. mapsize = env->me_mapsize;
  3516. if (flags & MDB_WRITEMAP) {
  3517. mp->mm_mapsize = mapsize;
  3518. mp->mm_dbs[FREE_DBI] = txn->mt_dbs[FREE_DBI];
  3519. mp->mm_dbs[MAIN_DBI] = txn->mt_dbs[MAIN_DBI];
  3520. mp->mm_last_pg = txn->mt_next_pgno - 1;
  3521. #if (__GNUC__ * 100 + __GNUC_MINOR__ >= 404) && /* TODO: portability */ \
  3522. !(defined(__i386__) || defined(__x86_64__))
  3523. /* LY: issue a memory barrier, if not x86. ITS#7969 */
  3524. __sync_synchronize();
  3525. #endif
  3526. mp->mm_txnid = txn->mt_txnid;
  3527. if (!(flags & (MDB_NOMETASYNC|MDB_NOSYNC))) {
  3528. unsigned meta_size = env->me_psize;
  3529. rc = (env->me_flags & MDB_MAPASYNC) ? MS_ASYNC : MS_SYNC;
  3530. ptr = (char *)mp - PAGEHDRSZ;
  3531. #ifndef _WIN32 /* POSIX msync() requires ptr = start of OS page */
  3532. r2 = (ptr - env->me_map) & (env->me_os_psize - 1);
  3533. ptr -= r2;
  3534. meta_size += r2;
  3535. #endif
  3536. if (MDB_MSYNC(ptr, meta_size, rc)) {
  3537. rc = ErrCode();
  3538. goto fail;
  3539. }
  3540. }
  3541. goto done;
  3542. }
  3543. metab.mm_txnid = mp->mm_txnid;
  3544. metab.mm_last_pg = mp->mm_last_pg;
  3545. meta.mm_mapsize = mapsize;
  3546. meta.mm_dbs[FREE_DBI] = txn->mt_dbs[FREE_DBI];
  3547. meta.mm_dbs[MAIN_DBI] = txn->mt_dbs[MAIN_DBI];
  3548. meta.mm_last_pg = txn->mt_next_pgno - 1;
  3549. meta.mm_txnid = txn->mt_txnid;
  3550. off = offsetof(MDB_meta, mm_mapsize);
  3551. ptr = (char *)&meta + off;
  3552. len = sizeof(MDB_meta) - off;
  3553. off += (char *)mp - env->me_map;
  3554. /* Write to the SYNC fd unless MDB_NOSYNC/MDB_NOMETASYNC.
  3555. * (me_mfd goes to the same file as me_fd, but writing to it
  3556. * also syncs to disk. Avoids a separate fdatasync() call.)
  3557. */
  3558. mfd = (flags & (MDB_NOSYNC|MDB_NOMETASYNC)) ? env->me_fd : env->me_mfd;
  3559. #ifdef _WIN32
  3560. {
  3561. memset(&ov, 0, sizeof(ov));
  3562. ov.Offset = off;
  3563. if (!WriteFile(mfd, ptr, len, (DWORD *)&rc, &ov))
  3564. rc = -1;
  3565. }
  3566. #else
  3567. retry_write:
  3568. rc = pwrite(mfd, ptr, len, off);
  3569. #endif
  3570. if (rc != len) {
  3571. rc = rc < 0 ? ErrCode() : EIO;
  3572. #ifndef _WIN32
  3573. if (rc == EINTR)
  3574. goto retry_write;
  3575. #endif
  3576. DPUTS("write failed, disk error?");
  3577. /* On a failure, the pagecache still contains the new data.
  3578. * Write some old data back, to prevent it from being used.
  3579. * Use the non-SYNC fd; we know it will fail anyway.
  3580. */
  3581. meta.mm_last_pg = metab.mm_last_pg;
  3582. meta.mm_txnid = metab.mm_txnid;
  3583. #ifdef _WIN32
  3584. memset(&ov, 0, sizeof(ov));
  3585. ov.Offset = off;
  3586. WriteFile(env->me_fd, ptr, len, NULL, &ov);
  3587. #else
  3588. r2 = pwrite(env->me_fd, ptr, len, off);
  3589. (void)r2; /* Silence warnings. We don't care about pwrite's return value */
  3590. #endif
  3591. fail:
  3592. env->me_flags |= MDB_FATAL_ERROR;
  3593. return rc;
  3594. }
  3595. /* MIPS has cache coherency issues, this is a no-op everywhere else */
  3596. CACHEFLUSH(env->me_map + off, len, DCACHE);
  3597. done:
  3598. /* Memory ordering issues are irrelevant; since the entire writer
  3599. * is wrapped by wmutex, all of these changes will become visible
  3600. * after the wmutex is unlocked. Since the DB is multi-version,
  3601. * readers will get consistent data regardless of how fresh or
  3602. * how stale their view of these values is.
  3603. */
  3604. if (env->me_txns)
  3605. env->me_txns->mti_txnid = txn->mt_txnid;
  3606. return MDB_SUCCESS;
  3607. }
  3608. /** Check both meta pages to see which one is newer.
  3609. * @param[in] env the environment handle
  3610. * @return newest #MDB_meta.
  3611. */
  3612. static MDB_meta *
  3613. mdb_env_pick_meta(const MDB_env *env)
  3614. {
  3615. MDB_meta *const *metas = env->me_metas;
  3616. return metas[ metas[0]->mm_txnid < metas[1]->mm_txnid ];
  3617. }
  3618. int ESECT
  3619. mdb_env_create(MDB_env **env)
  3620. {
  3621. MDB_env *e;
  3622. e = calloc(1, sizeof(MDB_env));
  3623. if (!e)
  3624. return ENOMEM;
  3625. e->me_maxreaders = DEFAULT_READERS;
  3626. e->me_maxdbs = e->me_numdbs = CORE_DBS;
  3627. e->me_fd = INVALID_HANDLE_VALUE;
  3628. e->me_lfd = INVALID_HANDLE_VALUE;
  3629. e->me_mfd = INVALID_HANDLE_VALUE;
  3630. #ifdef MDB_USE_POSIX_SEM
  3631. e->me_rmutex = SEM_FAILED;
  3632. e->me_wmutex = SEM_FAILED;
  3633. #endif
  3634. e->me_pid = getpid();
  3635. GET_PAGESIZE(e->me_os_psize);
  3636. VGMEMP_CREATE(e,0,0);
  3637. *env = e;
  3638. return MDB_SUCCESS;
  3639. }
  3640. static int ESECT
  3641. mdb_env_map(MDB_env *env, void *addr)
  3642. {
  3643. MDB_page *p;
  3644. unsigned int flags = env->me_flags;
  3645. #ifdef _WIN32
  3646. int rc;
  3647. HANDLE mh;
  3648. LONG sizelo, sizehi;
  3649. size_t msize;
  3650. if (flags & MDB_RDONLY) {
  3651. /* Don't set explicit map size, use whatever exists */
  3652. msize = 0;
  3653. sizelo = 0;
  3654. sizehi = 0;
  3655. } else {
  3656. msize = env->me_mapsize;
  3657. sizelo = msize & 0xffffffff;
  3658. sizehi = msize >> 16 >> 16; /* only needed on Win64 */
  3659. /* Windows won't create mappings for zero length files.
  3660. * and won't map more than the file size.
  3661. * Just set the maxsize right now.
  3662. */
  3663. if (!(flags & MDB_WRITEMAP) && (SetFilePointer(env->me_fd, sizelo, &sizehi, 0) != (DWORD)sizelo
  3664. || !SetEndOfFile(env->me_fd)
  3665. || SetFilePointer(env->me_fd, 0, NULL, 0) != 0))
  3666. return ErrCode();
  3667. }
  3668. mh = CreateFileMapping(env->me_fd, NULL, flags & MDB_WRITEMAP ?
  3669. PAGE_READWRITE : PAGE_READONLY,
  3670. sizehi, sizelo, NULL);
  3671. if (!mh)
  3672. return ErrCode();
  3673. env->me_map = MapViewOfFileEx(mh, flags & MDB_WRITEMAP ?
  3674. FILE_MAP_WRITE : FILE_MAP_READ,
  3675. 0, 0, msize, addr);
  3676. rc = env->me_map ? 0 : ErrCode();
  3677. CloseHandle(mh);
  3678. if (rc)
  3679. return rc;
  3680. #else
  3681. int mmap_flags = MAP_SHARED;
  3682. int prot = PROT_READ;
  3683. #ifdef MAP_NOSYNC /* Used on FreeBSD */
  3684. if (flags & MDB_NOSYNC)
  3685. mmap_flags |= MAP_NOSYNC;
  3686. #endif
  3687. if (flags & MDB_WRITEMAP) {
  3688. prot |= PROT_WRITE;
  3689. if (ftruncate(env->me_fd, env->me_mapsize) < 0)
  3690. return ErrCode();
  3691. }
  3692. env->me_map = mmap(addr, env->me_mapsize, prot, mmap_flags,
  3693. env->me_fd, 0);
  3694. if (env->me_map == MAP_FAILED) {
  3695. env->me_map = NULL;
  3696. return ErrCode();
  3697. }
  3698. if (flags & MDB_NORDAHEAD) {
  3699. /* Turn off readahead. It's harmful when the DB is larger than RAM. */
  3700. #ifdef MADV_RANDOM
  3701. madvise(env->me_map, env->me_mapsize, MADV_RANDOM);
  3702. #else
  3703. #ifdef POSIX_MADV_RANDOM
  3704. posix_madvise(env->me_map, env->me_mapsize, POSIX_MADV_RANDOM);
  3705. #endif /* POSIX_MADV_RANDOM */
  3706. #endif /* MADV_RANDOM */
  3707. }
  3708. #endif /* _WIN32 */
  3709. /* Can happen because the address argument to mmap() is just a
  3710. * hint. mmap() can pick another, e.g. if the range is in use.
  3711. * The MAP_FIXED flag would prevent that, but then mmap could
  3712. * instead unmap existing pages to make room for the new map.
  3713. */
  3714. if (addr && env->me_map != addr)
  3715. return EBUSY; /* TODO: Make a new MDB_* error code? */
  3716. p = (MDB_page *)env->me_map;
  3717. env->me_metas[0] = METADATA(p);
  3718. env->me_metas[1] = (MDB_meta *)((char *)env->me_metas[0] + env->me_psize);
  3719. return MDB_SUCCESS;
  3720. }
  3721. int ESECT
  3722. mdb_env_set_mapsize(MDB_env *env, size_t size)
  3723. {
  3724. /* If env is already open, caller is responsible for making
  3725. * sure there are no active txns.
  3726. */
  3727. if (env->me_map) {
  3728. int rc;
  3729. MDB_meta *meta;
  3730. void *old;
  3731. if (env->me_txn)
  3732. return EINVAL;
  3733. meta = mdb_env_pick_meta(env);
  3734. if (!size)
  3735. size = meta->mm_mapsize;
  3736. {
  3737. /* Silently round up to minimum if the size is too small */
  3738. size_t minsize = (meta->mm_last_pg + 1) * env->me_psize;
  3739. if (size < minsize)
  3740. size = minsize;
  3741. }
  3742. munmap(env->me_map, env->me_mapsize);
  3743. env->me_mapsize = size;
  3744. old = (env->me_flags & MDB_FIXEDMAP) ? env->me_map : NULL;
  3745. rc = mdb_env_map(env, old);
  3746. if (rc)
  3747. return rc;
  3748. }
  3749. env->me_mapsize = size;
  3750. if (env->me_psize)
  3751. env->me_maxpg = env->me_mapsize / env->me_psize;
  3752. return MDB_SUCCESS;
  3753. }
  3754. int ESECT
  3755. mdb_env_set_maxdbs(MDB_env *env, MDB_dbi dbs)
  3756. {
  3757. if (env->me_map)
  3758. return EINVAL;
  3759. env->me_maxdbs = dbs + CORE_DBS;
  3760. return MDB_SUCCESS;
  3761. }
  3762. int ESECT
  3763. mdb_env_set_maxreaders(MDB_env *env, unsigned int readers)
  3764. {
  3765. if (env->me_map || readers < 1)
  3766. return EINVAL;
  3767. env->me_maxreaders = readers;
  3768. return MDB_SUCCESS;
  3769. }
  3770. int ESECT
  3771. mdb_env_get_maxreaders(MDB_env *env, unsigned int *readers)
  3772. {
  3773. if (!env || !readers)
  3774. return EINVAL;
  3775. *readers = env->me_maxreaders;
  3776. return MDB_SUCCESS;
  3777. }
  3778. static int ESECT
  3779. mdb_fsize(HANDLE fd, size_t *size)
  3780. {
  3781. #ifdef _WIN32
  3782. LARGE_INTEGER fsize;
  3783. if (!GetFileSizeEx(fd, &fsize))
  3784. return ErrCode();
  3785. *size = fsize.QuadPart;
  3786. #else
  3787. struct stat st;
  3788. if (fstat(fd, &st))
  3789. return ErrCode();
  3790. *size = st.st_size;
  3791. #endif
  3792. return MDB_SUCCESS;
  3793. }
  3794. #ifdef _WIN32
  3795. typedef wchar_t mdb_nchar_t;
  3796. # define MDB_NAME(str) L##str
  3797. # define mdb_name_cpy wcscpy
  3798. #else
  3799. /** Character type for file names: char on Unix, wchar_t on Windows */
  3800. typedef char mdb_nchar_t;
  3801. # define MDB_NAME(str) str /**< #mdb_nchar_t[] string literal */
  3802. # define mdb_name_cpy strcpy /**< Copy name (#mdb_nchar_t string) */
  3803. #endif
  3804. /** Filename - string of #mdb_nchar_t[] */
  3805. typedef struct MDB_name {
  3806. int mn_len; /**< Length */
  3807. int mn_alloced; /**< True if #mn_val was malloced */
  3808. mdb_nchar_t *mn_val; /**< Contents */
  3809. } MDB_name;
  3810. /** Filename suffixes [datafile,lockfile][without,with MDB_NOSUBDIR] */
  3811. static const mdb_nchar_t *const mdb_suffixes[2][2] = {
  3812. { MDB_NAME("/data.mdb"), MDB_NAME("") },
  3813. { MDB_NAME("/lock.mdb"), MDB_NAME("-lock") }
  3814. };
  3815. #define MDB_SUFFLEN 9 /**< Max string length in #mdb_suffixes[] */
  3816. /** Set up filename + scratch area for filename suffix, for opening files.
  3817. * It should be freed with #mdb_fname_destroy().
  3818. * On Windows, paths are converted from char *UTF-8 to wchar_t *UTF-16.
  3819. *
  3820. * @param[in] path Pathname for #mdb_env_open().
  3821. * @param[in] envflags Whether a subdir and/or lockfile will be used.
  3822. * @param[out] fname Resulting filename, with room for a suffix if necessary.
  3823. */
  3824. static int ESECT
  3825. mdb_fname_init(const char *path, unsigned envflags, MDB_name *fname)
  3826. {
  3827. int no_suffix = F_ISSET(envflags, MDB_NOSUBDIR|MDB_NOLOCK);
  3828. fname->mn_alloced = 0;
  3829. #ifdef _WIN32
  3830. return utf8_to_utf16(path, fname, no_suffix ? 0 : MDB_SUFFLEN);
  3831. #else
  3832. fname->mn_len = strlen(path);
  3833. if (no_suffix)
  3834. fname->mn_val = (char *) path;
  3835. else if ((fname->mn_val = malloc(fname->mn_len + MDB_SUFFLEN+1)) != NULL) {
  3836. fname->mn_alloced = 1;
  3837. strcpy(fname->mn_val, path);
  3838. }
  3839. else
  3840. return ENOMEM;
  3841. return MDB_SUCCESS;
  3842. #endif
  3843. }
  3844. /** Destroy \b fname from #mdb_fname_init() */
  3845. #define mdb_fname_destroy(fname) \
  3846. do { if ((fname).mn_alloced) free((fname).mn_val); } while (0)
  3847. #ifdef O_CLOEXEC /* POSIX.1-2008: Set FD_CLOEXEC atomically at open() */
  3848. # define MDB_CLOEXEC O_CLOEXEC
  3849. #else
  3850. # define MDB_CLOEXEC 0
  3851. #endif
  3852. /** File type, access mode etc. for #mdb_fopen() */
  3853. enum mdb_fopen_type {
  3854. #ifdef _WIN32
  3855. MDB_O_RDONLY, MDB_O_RDWR, MDB_O_META, MDB_O_COPY, MDB_O_LOCKS
  3856. #else
  3857. /* A comment in mdb_fopen() explains some O_* flag choices. */
  3858. MDB_O_RDONLY= O_RDONLY, /**< for RDONLY me_fd */
  3859. MDB_O_RDWR = O_RDWR |O_CREAT, /**< for me_fd */
  3860. MDB_O_META = O_WRONLY|MDB_DSYNC |MDB_CLOEXEC, /**< for me_mfd */
  3861. MDB_O_COPY = O_WRONLY|O_CREAT|O_EXCL|MDB_CLOEXEC, /**< for #mdb_env_copy() */
  3862. /** Bitmask for open() flags in enum #mdb_fopen_type. The other bits
  3863. * distinguish otherwise-equal MDB_O_* constants from each other.
  3864. */
  3865. MDB_O_MASK = MDB_O_RDWR|MDB_CLOEXEC | MDB_O_RDONLY|MDB_O_META|MDB_O_COPY,
  3866. MDB_O_LOCKS = MDB_O_RDWR|MDB_CLOEXEC | ((MDB_O_MASK+1) & ~MDB_O_MASK) /**< for me_lfd */
  3867. #endif
  3868. };
  3869. /** Open an LMDB file.
  3870. * @param[in] env The LMDB environment.
  3871. * @param[in,out] fname Path from from #mdb_fname_init(). A suffix is
  3872. * appended if necessary to create the filename, without changing mn_len.
  3873. * @param[in] which Determines file type, access mode, etc.
  3874. * @param[in] mode The Unix permissions for the file, if we create it.
  3875. * @param[out] res Resulting file handle.
  3876. * @return 0 on success, non-zero on failure.
  3877. */
  3878. static int ESECT
  3879. mdb_fopen(const MDB_env *env, MDB_name *fname,
  3880. enum mdb_fopen_type which, mdb_mode_t mode,
  3881. HANDLE *res)
  3882. {
  3883. int rc = MDB_SUCCESS;
  3884. HANDLE fd;
  3885. #ifdef _WIN32
  3886. DWORD acc, share, disp, attrs;
  3887. #else
  3888. int flags;
  3889. #endif
  3890. if (fname->mn_alloced) /* modifiable copy */
  3891. mdb_name_cpy(fname->mn_val + fname->mn_len,
  3892. mdb_suffixes[which==MDB_O_LOCKS][F_ISSET(env->me_flags, MDB_NOSUBDIR)]);
  3893. /* The directory must already exist. Usually the file need not.
  3894. * MDB_O_META requires the file because we already created it using
  3895. * MDB_O_RDWR. MDB_O_COPY must not overwrite an existing file.
  3896. *
  3897. * With MDB_O_COPY we do not want the OS to cache the writes, since
  3898. * the source data is already in the OS cache.
  3899. *
  3900. * The lockfile needs FD_CLOEXEC (close file descriptor on exec*())
  3901. * to avoid the flock() issues noted under Caveats in lmdb.h.
  3902. * Also set it for other filehandles which the user cannot get at
  3903. * and close himself, which he may need after fork(). I.e. all but
  3904. * me_fd, which programs do use via mdb_env_get_fd().
  3905. */
  3906. #ifdef _WIN32
  3907. acc = GENERIC_READ|GENERIC_WRITE;
  3908. share = FILE_SHARE_READ|FILE_SHARE_WRITE;
  3909. disp = OPEN_ALWAYS;
  3910. attrs = FILE_ATTRIBUTE_NORMAL;
  3911. switch (which) {
  3912. case MDB_O_RDONLY: /* read-only datafile */
  3913. acc = GENERIC_READ;
  3914. disp = OPEN_EXISTING;
  3915. break;
  3916. case MDB_O_META: /* for writing metapages */
  3917. acc = GENERIC_WRITE;
  3918. disp = OPEN_EXISTING;
  3919. attrs = FILE_ATTRIBUTE_NORMAL|FILE_FLAG_WRITE_THROUGH;
  3920. break;
  3921. case MDB_O_COPY: /* mdb_env_copy() & co */
  3922. acc = GENERIC_WRITE;
  3923. share = 0;
  3924. disp = CREATE_NEW;
  3925. attrs = FILE_FLAG_NO_BUFFERING|FILE_FLAG_WRITE_THROUGH;
  3926. break;
  3927. default: break; /* silence gcc -Wswitch (not all enum values handled) */
  3928. }
  3929. fd = CreateFileW(fname->mn_val, acc, share, NULL, disp, attrs, NULL);
  3930. #else
  3931. fd = open(fname->mn_val, which & MDB_O_MASK, mode);
  3932. #endif
  3933. if (fd == INVALID_HANDLE_VALUE)
  3934. rc = ErrCode();
  3935. #ifndef _WIN32
  3936. else {
  3937. if (which != MDB_O_RDONLY && which != MDB_O_RDWR) {
  3938. /* Set CLOEXEC if we could not pass it to open() */
  3939. if (!MDB_CLOEXEC && (flags = fcntl(fd, F_GETFD)) != -1)
  3940. (void) fcntl(fd, F_SETFD, flags | FD_CLOEXEC);
  3941. }
  3942. if (which == MDB_O_COPY && env->me_psize >= env->me_os_psize) {
  3943. /* This may require buffer alignment. There is no portable
  3944. * way to ask how much, so we require OS pagesize alignment.
  3945. */
  3946. # ifdef F_NOCACHE /* __APPLE__ */
  3947. (void) fcntl(fd, F_NOCACHE, 1);
  3948. # elif defined O_DIRECT
  3949. /* open(...O_DIRECT...) would break on filesystems without
  3950. * O_DIRECT support (ITS#7682). Try to set it here instead.
  3951. */
  3952. if ((flags = fcntl(fd, F_GETFL)) != -1)
  3953. (void) fcntl(fd, F_SETFL, flags | O_DIRECT);
  3954. # endif
  3955. }
  3956. }
  3957. #endif /* !_WIN32 */
  3958. *res = fd;
  3959. return rc;
  3960. }
  3961. #ifdef BROKEN_FDATASYNC
  3962. #include <sys/utsname.h>
  3963. #include <sys/vfs.h>
  3964. #endif
  3965. /** Further setup required for opening an LMDB environment
  3966. */
  3967. static int ESECT
  3968. mdb_env_open2(MDB_env *env)
  3969. {
  3970. unsigned int flags = env->me_flags;
  3971. int i, newenv = 0, rc;
  3972. MDB_meta meta;
  3973. #ifdef _WIN32
  3974. /* See if we should use QueryLimited */
  3975. rc = GetVersion();
  3976. if ((rc & 0xff) > 5)
  3977. env->me_pidquery = MDB_PROCESS_QUERY_LIMITED_INFORMATION;
  3978. else
  3979. env->me_pidquery = PROCESS_QUERY_INFORMATION;
  3980. #endif /* _WIN32 */
  3981. #ifdef BROKEN_FDATASYNC
  3982. /* ext3/ext4 fdatasync is broken on some older Linux kernels.
  3983. * https://lkml.org/lkml/2012/9/3/83
  3984. * Kernels after 3.6-rc6 are known good.
  3985. * https://lkml.org/lkml/2012/9/10/556
  3986. * See if the DB is on ext3/ext4, then check for new enough kernel
  3987. * Kernels 2.6.32.60, 2.6.34.15, 3.2.30, and 3.5.4 are also known
  3988. * to be patched.
  3989. */
  3990. {
  3991. struct statfs st;
  3992. fstatfs(env->me_fd, &st);
  3993. while (st.f_type == 0xEF53) {
  3994. struct utsname uts;
  3995. int i;
  3996. uname(&uts);
  3997. if (uts.release[0] < '3') {
  3998. if (!strncmp(uts.release, "2.6.32.", 7)) {
  3999. i = atoi(uts.release+7);
  4000. if (i >= 60)
  4001. break; /* 2.6.32.60 and newer is OK */
  4002. } else if (!strncmp(uts.release, "2.6.34.", 7)) {
  4003. i = atoi(uts.release+7);
  4004. if (i >= 15)
  4005. break; /* 2.6.34.15 and newer is OK */
  4006. }
  4007. } else if (uts.release[0] == '3') {
  4008. i = atoi(uts.release+2);
  4009. if (i > 5)
  4010. break; /* 3.6 and newer is OK */
  4011. if (i == 5) {
  4012. i = atoi(uts.release+4);
  4013. if (i >= 4)
  4014. break; /* 3.5.4 and newer is OK */
  4015. } else if (i == 2) {
  4016. i = atoi(uts.release+4);
  4017. if (i >= 30)
  4018. break; /* 3.2.30 and newer is OK */
  4019. }
  4020. } else { /* 4.x and newer is OK */
  4021. break;
  4022. }
  4023. env->me_flags |= MDB_FSYNCONLY;
  4024. break;
  4025. }
  4026. }
  4027. #endif
  4028. if ((i = mdb_env_read_header(env, &meta)) != 0) {
  4029. if (i != ENOENT)
  4030. return i;
  4031. DPUTS("new mdbenv");
  4032. newenv = 1;
  4033. env->me_psize = env->me_os_psize;
  4034. if (env->me_psize > MAX_PAGESIZE)
  4035. env->me_psize = MAX_PAGESIZE;
  4036. memset(&meta, 0, sizeof(meta));
  4037. mdb_env_init_meta0(env, &meta);
  4038. meta.mm_mapsize = DEFAULT_MAPSIZE;
  4039. } else {
  4040. env->me_psize = meta.mm_psize;
  4041. }
  4042. /* Was a mapsize configured? */
  4043. if (!env->me_mapsize) {
  4044. env->me_mapsize = meta.mm_mapsize;
  4045. }
  4046. {
  4047. /* Make sure mapsize >= committed data size. Even when using
  4048. * mm_mapsize, which could be broken in old files (ITS#7789).
  4049. */
  4050. size_t minsize = (meta.mm_last_pg + 1) * meta.mm_psize;
  4051. if (env->me_mapsize < minsize)
  4052. env->me_mapsize = minsize;
  4053. }
  4054. meta.mm_mapsize = env->me_mapsize;
  4055. if (newenv && !(flags & MDB_FIXEDMAP)) {
  4056. /* mdb_env_map() may grow the datafile. Write the metapages
  4057. * first, so the file will be valid if initialization fails.
  4058. * Except with FIXEDMAP, since we do not yet know mm_address.
  4059. * We could fill in mm_address later, but then a different
  4060. * program might end up doing that - one with a memory layout
  4061. * and map address which does not suit the main program.
  4062. */
  4063. rc = mdb_env_init_meta(env, &meta);
  4064. if (rc)
  4065. return rc;
  4066. newenv = 0;
  4067. }
  4068. rc = mdb_env_map(env, (flags & MDB_FIXEDMAP) ? meta.mm_address : NULL);
  4069. if (rc)
  4070. return rc;
  4071. if (newenv) {
  4072. if (flags & MDB_FIXEDMAP)
  4073. meta.mm_address = env->me_map;
  4074. i = mdb_env_init_meta(env, &meta);
  4075. if (i != MDB_SUCCESS) {
  4076. return i;
  4077. }
  4078. }
  4079. env->me_maxfree_1pg = (env->me_psize - PAGEHDRSZ) / sizeof(pgno_t) - 1;
  4080. env->me_nodemax = (((env->me_psize - PAGEHDRSZ) / MDB_MINKEYS) & -2)
  4081. - sizeof(indx_t);
  4082. #if !(MDB_MAXKEYSIZE)
  4083. env->me_maxkey = env->me_nodemax - (NODESIZE + sizeof(MDB_db));
  4084. #endif
  4085. env->me_maxpg = env->me_mapsize / env->me_psize;
  4086. #if MDB_DEBUG
  4087. {
  4088. MDB_meta *meta = mdb_env_pick_meta(env);
  4089. MDB_db *db = &meta->mm_dbs[MAIN_DBI];
  4090. DPRINTF(("opened database version %u, pagesize %u",
  4091. meta->mm_version, env->me_psize));
  4092. DPRINTF(("using meta page %d", (int) (meta->mm_txnid & 1)));
  4093. DPRINTF(("depth: %u", db->md_depth));
  4094. DPRINTF(("entries: %"Z"u", db->md_entries));
  4095. DPRINTF(("branch pages: %"Z"u", db->md_branch_pages));
  4096. DPRINTF(("leaf pages: %"Z"u", db->md_leaf_pages));
  4097. DPRINTF(("overflow pages: %"Z"u", db->md_overflow_pages));
  4098. DPRINTF(("root: %"Z"u", db->md_root));
  4099. }
  4100. #endif
  4101. return MDB_SUCCESS;
  4102. }
  4103. /** Release a reader thread's slot in the reader lock table.
  4104. * This function is called automatically when a thread exits.
  4105. * @param[in] ptr This points to the slot in the reader lock table.
  4106. */
  4107. static void
  4108. mdb_env_reader_dest(void *ptr)
  4109. {
  4110. MDB_reader *reader = ptr;
  4111. #ifndef _WIN32
  4112. if (reader->mr_pid == getpid()) /* catch pthread_exit() in child process */
  4113. #endif
  4114. /* We omit the mutex, so do this atomically (i.e. skip mr_txnid) */
  4115. reader->mr_pid = 0;
  4116. }
  4117. #ifdef _WIN32
  4118. /** Junk for arranging thread-specific callbacks on Windows. This is
  4119. * necessarily platform and compiler-specific. Windows supports up
  4120. * to 1088 keys. Let's assume nobody opens more than 64 environments
  4121. * in a single process, for now. They can override this if needed.
  4122. */
  4123. #ifndef MAX_TLS_KEYS
  4124. #define MAX_TLS_KEYS 64
  4125. #endif
  4126. static pthread_key_t mdb_tls_keys[MAX_TLS_KEYS];
  4127. static int mdb_tls_nkeys;
  4128. static void NTAPI mdb_tls_callback(PVOID module, DWORD reason, PVOID ptr)
  4129. {
  4130. int i;
  4131. switch(reason) {
  4132. case DLL_PROCESS_ATTACH: break;
  4133. case DLL_THREAD_ATTACH: break;
  4134. case DLL_THREAD_DETACH:
  4135. for (i=0; i<mdb_tls_nkeys; i++) {
  4136. MDB_reader *r = pthread_getspecific(mdb_tls_keys[i]);
  4137. if (r) {
  4138. mdb_env_reader_dest(r);
  4139. }
  4140. }
  4141. break;
  4142. case DLL_PROCESS_DETACH: break;
  4143. }
  4144. }
  4145. #ifdef __GNUC__
  4146. #ifdef _WIN64
  4147. const PIMAGE_TLS_CALLBACK mdb_tls_cbp __attribute__((section (".CRT$XLB"))) = mdb_tls_callback;
  4148. #else
  4149. PIMAGE_TLS_CALLBACK mdb_tls_cbp __attribute__((section (".CRT$XLB"))) = mdb_tls_callback;
  4150. #endif
  4151. #else
  4152. #ifdef _WIN64
  4153. /* Force some symbol references.
  4154. * _tls_used forces the linker to create the TLS directory if not already done
  4155. * mdb_tls_cbp prevents whole-program-optimizer from dropping the symbol.
  4156. */
  4157. #pragma comment(linker, "/INCLUDE:_tls_used")
  4158. #pragma comment(linker, "/INCLUDE:mdb_tls_cbp")
  4159. #pragma const_seg(".CRT$XLB")
  4160. extern const PIMAGE_TLS_CALLBACK mdb_tls_cbp;
  4161. const PIMAGE_TLS_CALLBACK mdb_tls_cbp = mdb_tls_callback;
  4162. #pragma const_seg()
  4163. #else /* _WIN32 */
  4164. #pragma comment(linker, "/INCLUDE:__tls_used")
  4165. #pragma comment(linker, "/INCLUDE:_mdb_tls_cbp")
  4166. #pragma data_seg(".CRT$XLB")
  4167. PIMAGE_TLS_CALLBACK mdb_tls_cbp = mdb_tls_callback;
  4168. #pragma data_seg()
  4169. #endif /* WIN 32/64 */
  4170. #endif /* !__GNUC__ */
  4171. #endif
  4172. /** Downgrade the exclusive lock on the region back to shared */
  4173. static int ESECT
  4174. mdb_env_share_locks(MDB_env *env, int *excl)
  4175. {
  4176. int rc = 0;
  4177. MDB_meta *meta = mdb_env_pick_meta(env);
  4178. env->me_txns->mti_txnid = meta->mm_txnid;
  4179. #ifdef _WIN32
  4180. {
  4181. OVERLAPPED ov;
  4182. /* First acquire a shared lock. The Unlock will
  4183. * then release the existing exclusive lock.
  4184. */
  4185. memset(&ov, 0, sizeof(ov));
  4186. if (!LockFileEx(env->me_lfd, 0, 0, 1, 0, &ov)) {
  4187. rc = ErrCode();
  4188. } else {
  4189. UnlockFile(env->me_lfd, 0, 0, 1, 0);
  4190. *excl = 0;
  4191. }
  4192. }
  4193. #else
  4194. {
  4195. struct flock lock_info;
  4196. /* The shared lock replaces the existing lock */
  4197. memset((void *)&lock_info, 0, sizeof(lock_info));
  4198. lock_info.l_type = F_RDLCK;
  4199. lock_info.l_whence = SEEK_SET;
  4200. lock_info.l_start = 0;
  4201. lock_info.l_len = 1;
  4202. while ((rc = fcntl(env->me_lfd, F_SETLK, &lock_info)) &&
  4203. (rc = ErrCode()) == EINTR) ;
  4204. *excl = rc ? -1 : 0; /* error may mean we lost the lock */
  4205. }
  4206. #endif
  4207. return rc;
  4208. }
  4209. /** Try to get exclusive lock, otherwise shared.
  4210. * Maintain *excl = -1: no/unknown lock, 0: shared, 1: exclusive.
  4211. */
  4212. static int ESECT
  4213. mdb_env_excl_lock(MDB_env *env, int *excl)
  4214. {
  4215. int rc = 0;
  4216. #ifdef _WIN32
  4217. if (LockFile(env->me_lfd, 0, 0, 1, 0)) {
  4218. *excl = 1;
  4219. } else {
  4220. OVERLAPPED ov;
  4221. memset(&ov, 0, sizeof(ov));
  4222. if (LockFileEx(env->me_lfd, 0, 0, 1, 0, &ov)) {
  4223. *excl = 0;
  4224. } else {
  4225. rc = ErrCode();
  4226. }
  4227. }
  4228. #else
  4229. struct flock lock_info;
  4230. memset((void *)&lock_info, 0, sizeof(lock_info));
  4231. lock_info.l_type = F_WRLCK;
  4232. lock_info.l_whence = SEEK_SET;
  4233. lock_info.l_start = 0;
  4234. lock_info.l_len = 1;
  4235. while ((rc = fcntl(env->me_lfd, F_SETLK, &lock_info)) &&
  4236. (rc = ErrCode()) == EINTR) ;
  4237. if (!rc) {
  4238. *excl = 1;
  4239. } else
  4240. # ifndef MDB_USE_POSIX_MUTEX
  4241. if (*excl < 0) /* always true when MDB_USE_POSIX_MUTEX */
  4242. # endif
  4243. {
  4244. lock_info.l_type = F_RDLCK;
  4245. while ((rc = fcntl(env->me_lfd, F_SETLKW, &lock_info)) &&
  4246. (rc = ErrCode()) == EINTR) ;
  4247. if (rc == 0)
  4248. *excl = 0;
  4249. }
  4250. #endif
  4251. return rc;
  4252. }
  4253. #ifdef MDB_USE_HASH
  4254. /*
  4255. * hash_64 - 64 bit Fowler/Noll/Vo-0 FNV-1a hash code
  4256. *
  4257. * @(#) $Revision: 5.1 $
  4258. * @(#) $Id: hash_64a.c,v 5.1 2009/06/30 09:01:38 chongo Exp $
  4259. * @(#) $Source: /usr/local/src/cmd/fnv/RCS/hash_64a.c,v $
  4260. *
  4261. * http://www.isthe.com/chongo/tech/comp/fnv/index.html
  4262. *
  4263. ***
  4264. *
  4265. * Please do not copyright this code. This code is in the public domain.
  4266. *
  4267. * LANDON CURT NOLL DISCLAIMS ALL WARRANTIES WITH REGARD TO THIS SOFTWARE,
  4268. * INCLUDING ALL IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS. IN NO
  4269. * EVENT SHALL LANDON CURT NOLL BE LIABLE FOR ANY SPECIAL, INDIRECT OR
  4270. * CONSEQUENTIAL DAMAGES OR ANY DAMAGES WHATSOEVER RESULTING FROM LOSS OF
  4271. * USE, DATA OR PROFITS, WHETHER IN AN ACTION OF CONTRACT, NEGLIGENCE OR
  4272. * OTHER TORTIOUS ACTION, ARISING OUT OF OR IN CONNECTION WITH THE USE OR
  4273. * PERFORMANCE OF THIS SOFTWARE.
  4274. *
  4275. * By:
  4276. * chongo <Landon Curt Noll> /\oo/\
  4277. * http://www.isthe.com/chongo/
  4278. *
  4279. * Share and Enjoy! :-)
  4280. */
  4281. typedef unsigned long long mdb_hash_t;
  4282. #define MDB_HASH_INIT ((mdb_hash_t)0xcbf29ce484222325ULL)
  4283. /** perform a 64 bit Fowler/Noll/Vo FNV-1a hash on a buffer
  4284. * @param[in] val value to hash
  4285. * @param[in] hval initial value for hash
  4286. * @return 64 bit hash
  4287. *
  4288. * NOTE: To use the recommended 64 bit FNV-1a hash, use MDB_HASH_INIT as the
  4289. * hval arg on the first call.
  4290. */
  4291. static mdb_hash_t
  4292. mdb_hash_val(MDB_val *val, mdb_hash_t hval)
  4293. {
  4294. unsigned char *s = (unsigned char *)val->mv_data; /* unsigned string */
  4295. unsigned char *end = s + val->mv_size;
  4296. /*
  4297. * FNV-1a hash each octet of the string
  4298. */
  4299. while (s < end) {
  4300. /* xor the bottom with the current octet */
  4301. hval ^= (mdb_hash_t)*s++;
  4302. /* multiply by the 64 bit FNV magic prime mod 2^64 */
  4303. hval += (hval << 1) + (hval << 4) + (hval << 5) +
  4304. (hval << 7) + (hval << 8) + (hval << 40);
  4305. }
  4306. /* return our new hash value */
  4307. return hval;
  4308. }
  4309. /** Hash the string and output the encoded hash.
  4310. * This uses modified RFC1924 Ascii85 encoding to accommodate systems with
  4311. * very short name limits. We don't care about the encoding being reversible,
  4312. * we just want to preserve as many bits of the input as possible in a
  4313. * small printable string.
  4314. * @param[in] str string to hash
  4315. * @param[out] encbuf an array of 11 chars to hold the hash
  4316. */
  4317. static const char mdb_a85[]= "0123456789ABCDEFGHIJKLMNOPQRSTUVWXYZabcdefghijklmnopqrstuvwxyz!#$%&()*+-;<=>?@^_`{|}~";
  4318. static void ESECT
  4319. mdb_pack85(unsigned long l, char *out)
  4320. {
  4321. int i;
  4322. for (i=0; i<5; i++) {
  4323. *out++ = mdb_a85[l % 85];
  4324. l /= 85;
  4325. }
  4326. }
  4327. static void ESECT
  4328. mdb_hash_enc(MDB_val *val, char *encbuf)
  4329. {
  4330. mdb_hash_t h = mdb_hash_val(val, MDB_HASH_INIT);
  4331. mdb_pack85(h, encbuf);
  4332. mdb_pack85(h>>32, encbuf+5);
  4333. encbuf[10] = '\0';
  4334. }
  4335. #endif
  4336. /** Open and/or initialize the lock region for the environment.
  4337. * @param[in] env The LMDB environment.
  4338. * @param[in] fname Filename + scratch area, from #mdb_fname_init().
  4339. * @param[in] mode The Unix permissions for the file, if we create it.
  4340. * @param[in,out] excl In -1, out lock type: -1 none, 0 shared, 1 exclusive
  4341. * @return 0 on success, non-zero on failure.
  4342. */
  4343. static int ESECT
  4344. mdb_env_setup_locks(MDB_env *env, MDB_name *fname, int mode, int *excl)
  4345. {
  4346. #ifdef _WIN32
  4347. # define MDB_ERRCODE_ROFS ERROR_WRITE_PROTECT
  4348. #else
  4349. # define MDB_ERRCODE_ROFS EROFS
  4350. #endif
  4351. int rc;
  4352. off_t size, rsize;
  4353. rc = mdb_fopen(env, fname, MDB_O_LOCKS, mode, &env->me_lfd);
  4354. if (rc) {
  4355. /* Omit lockfile if read-only env on read-only filesystem */
  4356. if (rc == MDB_ERRCODE_ROFS && (env->me_flags & MDB_RDONLY)) {
  4357. return MDB_SUCCESS;
  4358. }
  4359. goto fail;
  4360. }
  4361. if (!(env->me_flags & MDB_NOTLS)) {
  4362. rc = pthread_key_create(&env->me_txkey, mdb_env_reader_dest);
  4363. if (rc)
  4364. goto fail;
  4365. env->me_flags |= MDB_ENV_TXKEY;
  4366. #ifdef _WIN32
  4367. /* Windows TLS callbacks need help finding their TLS info. */
  4368. if (mdb_tls_nkeys >= MAX_TLS_KEYS) {
  4369. rc = MDB_TLS_FULL;
  4370. goto fail;
  4371. }
  4372. mdb_tls_keys[mdb_tls_nkeys++] = env->me_txkey;
  4373. #endif
  4374. }
  4375. /* Try to get exclusive lock. If we succeed, then
  4376. * nobody is using the lock region and we should initialize it.
  4377. */
  4378. if ((rc = mdb_env_excl_lock(env, excl))) goto fail;
  4379. #ifdef _WIN32
  4380. size = GetFileSize(env->me_lfd, NULL);
  4381. #else
  4382. size = lseek(env->me_lfd, 0, SEEK_END);
  4383. if (size == -1) goto fail_errno;
  4384. #endif
  4385. rsize = (env->me_maxreaders-1) * sizeof(MDB_reader) + sizeof(MDB_txninfo);
  4386. if (size < rsize && *excl > 0) {
  4387. #ifdef _WIN32
  4388. if (SetFilePointer(env->me_lfd, rsize, NULL, FILE_BEGIN) != (DWORD)rsize
  4389. || !SetEndOfFile(env->me_lfd))
  4390. goto fail_errno;
  4391. #else
  4392. if (ftruncate(env->me_lfd, rsize) != 0) goto fail_errno;
  4393. #endif
  4394. } else {
  4395. rsize = size;
  4396. size = rsize - sizeof(MDB_txninfo);
  4397. env->me_maxreaders = size/sizeof(MDB_reader) + 1;
  4398. }
  4399. {
  4400. #ifdef _WIN32
  4401. HANDLE mh;
  4402. mh = CreateFileMapping(env->me_lfd, NULL, PAGE_READWRITE,
  4403. 0, 0, NULL);
  4404. if (!mh) goto fail_errno;
  4405. env->me_txns = MapViewOfFileEx(mh, FILE_MAP_WRITE, 0, 0, rsize, NULL);
  4406. CloseHandle(mh);
  4407. if (!env->me_txns) goto fail_errno;
  4408. #else
  4409. void *m = mmap(NULL, rsize, PROT_READ|PROT_WRITE, MAP_SHARED,
  4410. env->me_lfd, 0);
  4411. if (m == MAP_FAILED) goto fail_errno;
  4412. env->me_txns = m;
  4413. #endif
  4414. }
  4415. if (*excl > 0) {
  4416. #ifdef _WIN32
  4417. BY_HANDLE_FILE_INFORMATION stbuf;
  4418. struct {
  4419. DWORD volume;
  4420. DWORD nhigh;
  4421. DWORD nlow;
  4422. } idbuf;
  4423. MDB_val val;
  4424. char encbuf[11];
  4425. if (!mdb_sec_inited) {
  4426. InitializeSecurityDescriptor(&mdb_null_sd,
  4427. SECURITY_DESCRIPTOR_REVISION);
  4428. SetSecurityDescriptorDacl(&mdb_null_sd, TRUE, 0, FALSE);
  4429. mdb_all_sa.nLength = sizeof(SECURITY_ATTRIBUTES);
  4430. mdb_all_sa.bInheritHandle = FALSE;
  4431. mdb_all_sa.lpSecurityDescriptor = &mdb_null_sd;
  4432. mdb_sec_inited = 1;
  4433. }
  4434. if (!GetFileInformationByHandle(env->me_lfd, &stbuf)) goto fail_errno;
  4435. idbuf.volume = stbuf.dwVolumeSerialNumber;
  4436. idbuf.nhigh = stbuf.nFileIndexHigh;
  4437. idbuf.nlow = stbuf.nFileIndexLow;
  4438. val.mv_data = &idbuf;
  4439. val.mv_size = sizeof(idbuf);
  4440. mdb_hash_enc(&val, encbuf);
  4441. sprintf(env->me_txns->mti_rmname, "Global\\MDBr%s", encbuf);
  4442. sprintf(env->me_txns->mti_wmname, "Global\\MDBw%s", encbuf);
  4443. env->me_rmutex = CreateMutexA(&mdb_all_sa, FALSE, env->me_txns->mti_rmname);
  4444. if (!env->me_rmutex) goto fail_errno;
  4445. env->me_wmutex = CreateMutexA(&mdb_all_sa, FALSE, env->me_txns->mti_wmname);
  4446. if (!env->me_wmutex) goto fail_errno;
  4447. #elif defined(MDB_USE_POSIX_SEM)
  4448. struct stat stbuf;
  4449. struct {
  4450. dev_t dev;
  4451. ino_t ino;
  4452. } idbuf;
  4453. MDB_val val;
  4454. char encbuf[11];
  4455. #if defined(__NetBSD__)
  4456. #define MDB_SHORT_SEMNAMES 1 /* limited to 14 chars */
  4457. #endif
  4458. if (fstat(env->me_lfd, &stbuf)) goto fail_errno;
  4459. idbuf.dev = stbuf.st_dev;
  4460. idbuf.ino = stbuf.st_ino;
  4461. val.mv_data = &idbuf;
  4462. val.mv_size = sizeof(idbuf);
  4463. mdb_hash_enc(&val, encbuf);
  4464. #ifdef MDB_SHORT_SEMNAMES
  4465. encbuf[9] = '\0'; /* drop name from 15 chars to 14 chars */
  4466. #endif
  4467. sprintf(env->me_txns->mti_rmname, "/MDBr%s", encbuf);
  4468. sprintf(env->me_txns->mti_wmname, "/MDBw%s", encbuf);
  4469. /* Clean up after a previous run, if needed: Try to
  4470. * remove both semaphores before doing anything else.
  4471. */
  4472. sem_unlink(env->me_txns->mti_rmname);
  4473. sem_unlink(env->me_txns->mti_wmname);
  4474. env->me_rmutex = sem_open(env->me_txns->mti_rmname,
  4475. O_CREAT|O_EXCL, mode, 1);
  4476. if (env->me_rmutex == SEM_FAILED) goto fail_errno;
  4477. env->me_wmutex = sem_open(env->me_txns->mti_wmname,
  4478. O_CREAT|O_EXCL, mode, 1);
  4479. if (env->me_wmutex == SEM_FAILED) goto fail_errno;
  4480. #else /* MDB_USE_POSIX_MUTEX: */
  4481. pthread_mutexattr_t mattr;
  4482. /* Solaris needs this before initing a robust mutex. Otherwise
  4483. * it may skip the init and return EBUSY "seems someone already
  4484. * inited" or EINVAL "it was inited differently".
  4485. */
  4486. memset(env->me_txns->mti_rmutex, 0, sizeof(*env->me_txns->mti_rmutex));
  4487. memset(env->me_txns->mti_wmutex, 0, sizeof(*env->me_txns->mti_wmutex));
  4488. if ((rc = pthread_mutexattr_init(&mattr)))
  4489. goto fail;
  4490. rc = pthread_mutexattr_setpshared(&mattr, PTHREAD_PROCESS_SHARED);
  4491. #ifdef MDB_ROBUST_SUPPORTED
  4492. if (!rc) rc = pthread_mutexattr_setrobust(&mattr, PTHREAD_MUTEX_ROBUST);
  4493. #endif
  4494. if (!rc) rc = pthread_mutex_init(env->me_txns->mti_rmutex, &mattr);
  4495. if (!rc) rc = pthread_mutex_init(env->me_txns->mti_wmutex, &mattr);
  4496. pthread_mutexattr_destroy(&mattr);
  4497. if (rc)
  4498. goto fail;
  4499. #endif /* _WIN32 || MDB_USE_POSIX_SEM */
  4500. env->me_txns->mti_magic = MDB_MAGIC;
  4501. env->me_txns->mti_format = MDB_LOCK_FORMAT;
  4502. env->me_txns->mti_txnid = 0;
  4503. env->me_txns->mti_numreaders = 0;
  4504. } else {
  4505. if (env->me_txns->mti_magic != MDB_MAGIC) {
  4506. DPUTS("lock region has invalid magic");
  4507. rc = MDB_INVALID;
  4508. goto fail;
  4509. }
  4510. if (env->me_txns->mti_format != MDB_LOCK_FORMAT) {
  4511. DPRINTF(("lock region has format+version 0x%x, expected 0x%x",
  4512. env->me_txns->mti_format, MDB_LOCK_FORMAT));
  4513. rc = MDB_VERSION_MISMATCH;
  4514. goto fail;
  4515. }
  4516. rc = ErrCode();
  4517. if (rc && rc != EACCES && rc != EAGAIN) {
  4518. goto fail;
  4519. }
  4520. #ifdef _WIN32
  4521. env->me_rmutex = OpenMutexA(SYNCHRONIZE, FALSE, env->me_txns->mti_rmname);
  4522. if (!env->me_rmutex) goto fail_errno;
  4523. env->me_wmutex = OpenMutexA(SYNCHRONIZE, FALSE, env->me_txns->mti_wmname);
  4524. if (!env->me_wmutex) goto fail_errno;
  4525. #elif defined(MDB_USE_POSIX_SEM)
  4526. env->me_rmutex = sem_open(env->me_txns->mti_rmname, 0);
  4527. if (env->me_rmutex == SEM_FAILED) goto fail_errno;
  4528. env->me_wmutex = sem_open(env->me_txns->mti_wmname, 0);
  4529. if (env->me_wmutex == SEM_FAILED) goto fail_errno;
  4530. #endif
  4531. }
  4532. return MDB_SUCCESS;
  4533. fail_errno:
  4534. rc = ErrCode();
  4535. fail:
  4536. return rc;
  4537. }
  4538. /** Only a subset of the @ref mdb_env flags can be changed
  4539. * at runtime. Changing other flags requires closing the
  4540. * environment and re-opening it with the new flags.
  4541. */
  4542. #define CHANGEABLE (MDB_NOSYNC|MDB_NOMETASYNC|MDB_MAPASYNC|MDB_NOMEMINIT)
  4543. #define CHANGELESS (MDB_FIXEDMAP|MDB_NOSUBDIR|MDB_RDONLY| \
  4544. MDB_WRITEMAP|MDB_NOTLS|MDB_NOLOCK|MDB_NORDAHEAD)
  4545. #if VALID_FLAGS & PERSISTENT_FLAGS & (CHANGEABLE|CHANGELESS)
  4546. # error "Persistent DB flags & env flags overlap, but both go in mm_flags"
  4547. #endif
  4548. int ESECT
  4549. mdb_env_open(MDB_env *env, const char *path, unsigned int flags, mdb_mode_t mode)
  4550. {
  4551. int rc, excl = -1;
  4552. MDB_name fname;
  4553. if (env->me_fd!=INVALID_HANDLE_VALUE || (flags & ~(CHANGEABLE|CHANGELESS)))
  4554. return EINVAL;
  4555. flags |= env->me_flags;
  4556. rc = mdb_fname_init(path, flags, &fname);
  4557. if (rc)
  4558. return rc;
  4559. if (flags & MDB_RDONLY) {
  4560. /* silently ignore WRITEMAP when we're only getting read access */
  4561. flags &= ~MDB_WRITEMAP;
  4562. } else {
  4563. if (!((env->me_free_pgs = mdb_midl_alloc(MDB_IDL_UM_MAX)) &&
  4564. (env->me_dirty_list = calloc(MDB_IDL_UM_SIZE, sizeof(MDB_ID2)))))
  4565. rc = ENOMEM;
  4566. }
  4567. env->me_flags = flags |= MDB_ENV_ACTIVE;
  4568. if (rc)
  4569. goto leave;
  4570. env->me_path = strdup(path);
  4571. env->me_dbxs = calloc(env->me_maxdbs, sizeof(MDB_dbx));
  4572. env->me_dbflags = calloc(env->me_maxdbs, sizeof(uint16_t));
  4573. env->me_dbiseqs = calloc(env->me_maxdbs, sizeof(unsigned int));
  4574. if (!(env->me_dbxs && env->me_path && env->me_dbflags && env->me_dbiseqs)) {
  4575. rc = ENOMEM;
  4576. goto leave;
  4577. }
  4578. env->me_dbxs[FREE_DBI].md_cmp = mdb_cmp_long; /* aligned MDB_INTEGERKEY */
  4579. /* For RDONLY, get lockfile after we know datafile exists */
  4580. if (!(flags & (MDB_RDONLY|MDB_NOLOCK))) {
  4581. rc = mdb_env_setup_locks(env, &fname, mode, &excl);
  4582. if (rc)
  4583. goto leave;
  4584. }
  4585. rc = mdb_fopen(env, &fname,
  4586. (flags & MDB_RDONLY) ? MDB_O_RDONLY : MDB_O_RDWR,
  4587. mode, &env->me_fd);
  4588. if (rc)
  4589. goto leave;
  4590. if ((flags & (MDB_RDONLY|MDB_NOLOCK)) == MDB_RDONLY) {
  4591. rc = mdb_env_setup_locks(env, &fname, mode, &excl);
  4592. if (rc)
  4593. goto leave;
  4594. }
  4595. if ((rc = mdb_env_open2(env)) == MDB_SUCCESS) {
  4596. if (!(flags & (MDB_RDONLY|MDB_WRITEMAP))) {
  4597. /* Synchronous fd for meta writes. Needed even with
  4598. * MDB_NOSYNC/MDB_NOMETASYNC, in case these get reset.
  4599. */
  4600. rc = mdb_fopen(env, &fname, MDB_O_META, mode, &env->me_mfd);
  4601. if (rc)
  4602. goto leave;
  4603. }
  4604. DPRINTF(("opened dbenv %p", (void *) env));
  4605. if (excl > 0) {
  4606. rc = mdb_env_share_locks(env, &excl);
  4607. if (rc)
  4608. goto leave;
  4609. }
  4610. if (!(flags & MDB_RDONLY)) {
  4611. MDB_txn *txn;
  4612. int tsize = sizeof(MDB_txn), size = tsize + env->me_maxdbs *
  4613. (sizeof(MDB_db)+sizeof(MDB_cursor *)+sizeof(unsigned int)+1);
  4614. if ((env->me_pbuf = calloc(1, env->me_psize)) &&
  4615. (txn = calloc(1, size)))
  4616. {
  4617. txn->mt_dbs = (MDB_db *)((char *)txn + tsize);
  4618. txn->mt_cursors = (MDB_cursor **)(txn->mt_dbs + env->me_maxdbs);
  4619. txn->mt_dbiseqs = (unsigned int *)(txn->mt_cursors + env->me_maxdbs);
  4620. txn->mt_dbflags = (unsigned char *)(txn->mt_dbiseqs + env->me_maxdbs);
  4621. txn->mt_env = env;
  4622. txn->mt_dbxs = env->me_dbxs;
  4623. txn->mt_flags = MDB_TXN_FINISHED;
  4624. env->me_txn0 = txn;
  4625. } else {
  4626. rc = ENOMEM;
  4627. }
  4628. }
  4629. }
  4630. leave:
  4631. if (rc) {
  4632. mdb_env_close0(env, excl);
  4633. }
  4634. mdb_fname_destroy(fname);
  4635. return rc;
  4636. }
  4637. /** Destroy resources from mdb_env_open(), clear our readers & DBIs */
  4638. static void ESECT
  4639. mdb_env_close0(MDB_env *env, int excl)
  4640. {
  4641. int i;
  4642. if (!(env->me_flags & MDB_ENV_ACTIVE))
  4643. return;
  4644. /* Doing this here since me_dbxs may not exist during mdb_env_close */
  4645. if (env->me_dbxs) {
  4646. for (i = env->me_maxdbs; --i >= CORE_DBS; )
  4647. free(env->me_dbxs[i].md_name.mv_data);
  4648. free(env->me_dbxs);
  4649. }
  4650. free(env->me_pbuf);
  4651. free(env->me_dbiseqs);
  4652. free(env->me_dbflags);
  4653. free(env->me_path);
  4654. free(env->me_dirty_list);
  4655. free(env->me_txn0);
  4656. mdb_midl_free(env->me_free_pgs);
  4657. if (env->me_flags & MDB_ENV_TXKEY) {
  4658. pthread_key_delete(env->me_txkey);
  4659. #ifdef _WIN32
  4660. /* Delete our key from the global list */
  4661. for (i=0; i<mdb_tls_nkeys; i++)
  4662. if (mdb_tls_keys[i] == env->me_txkey) {
  4663. mdb_tls_keys[i] = mdb_tls_keys[mdb_tls_nkeys-1];
  4664. mdb_tls_nkeys--;
  4665. break;
  4666. }
  4667. #endif
  4668. }
  4669. if (env->me_map) {
  4670. munmap(env->me_map, env->me_mapsize);
  4671. }
  4672. if (env->me_mfd != INVALID_HANDLE_VALUE)
  4673. (void) close(env->me_mfd);
  4674. if (env->me_fd != INVALID_HANDLE_VALUE)
  4675. (void) close(env->me_fd);
  4676. if (env->me_txns) {
  4677. MDB_PID_T pid = getpid();
  4678. /* Clearing readers is done in this function because
  4679. * me_txkey with its destructor must be disabled first.
  4680. *
  4681. * We skip the the reader mutex, so we touch only
  4682. * data owned by this process (me_close_readers and
  4683. * our readers), and clear each reader atomically.
  4684. */
  4685. for (i = env->me_close_readers; --i >= 0; )
  4686. if (env->me_txns->mti_readers[i].mr_pid == pid)
  4687. env->me_txns->mti_readers[i].mr_pid = 0;
  4688. #ifdef _WIN32
  4689. if (env->me_rmutex) {
  4690. CloseHandle(env->me_rmutex);
  4691. if (env->me_wmutex) CloseHandle(env->me_wmutex);
  4692. }
  4693. /* Windows automatically destroys the mutexes when
  4694. * the last handle closes.
  4695. */
  4696. #elif defined(MDB_USE_POSIX_SEM)
  4697. if (env->me_rmutex != SEM_FAILED) {
  4698. sem_close(env->me_rmutex);
  4699. if (env->me_wmutex != SEM_FAILED)
  4700. sem_close(env->me_wmutex);
  4701. /* If we have the filelock: If we are the
  4702. * only remaining user, clean up semaphores.
  4703. */
  4704. if (excl == 0)
  4705. mdb_env_excl_lock(env, &excl);
  4706. if (excl > 0) {
  4707. sem_unlink(env->me_txns->mti_rmname);
  4708. sem_unlink(env->me_txns->mti_wmname);
  4709. }
  4710. }
  4711. #elif defined(MDB_ROBUST_SUPPORTED)
  4712. /* If we have the filelock: If we are the
  4713. * only remaining user, clean up robust
  4714. * mutexes.
  4715. */
  4716. if (excl == 0)
  4717. mdb_env_excl_lock(env, &excl);
  4718. if (excl > 0) {
  4719. pthread_mutex_destroy(env->me_txns->mti_rmutex);
  4720. pthread_mutex_destroy(env->me_txns->mti_wmutex);
  4721. }
  4722. #endif
  4723. munmap((void *)env->me_txns, (env->me_maxreaders-1)*sizeof(MDB_reader)+sizeof(MDB_txninfo));
  4724. }
  4725. if (env->me_lfd != INVALID_HANDLE_VALUE) {
  4726. #ifdef _WIN32
  4727. if (excl >= 0) {
  4728. /* Unlock the lockfile. Windows would have unlocked it
  4729. * after closing anyway, but not necessarily at once.
  4730. */
  4731. UnlockFile(env->me_lfd, 0, 0, 1, 0);
  4732. }
  4733. #endif
  4734. (void) close(env->me_lfd);
  4735. }
  4736. env->me_flags &= ~(MDB_ENV_ACTIVE|MDB_ENV_TXKEY);
  4737. }
  4738. void ESECT
  4739. mdb_env_close(MDB_env *env)
  4740. {
  4741. MDB_page *dp;
  4742. if (env == NULL)
  4743. return;
  4744. VGMEMP_DESTROY(env);
  4745. while ((dp = env->me_dpages) != NULL) {
  4746. VGMEMP_DEFINED(&dp->mp_next, sizeof(dp->mp_next));
  4747. env->me_dpages = dp->mp_next;
  4748. free(dp);
  4749. }
  4750. mdb_env_close0(env, 0);
  4751. free(env);
  4752. }
  4753. /** Compare two items pointing at aligned size_t's */
  4754. static int
  4755. mdb_cmp_long(const MDB_val *a, const MDB_val *b)
  4756. {
  4757. return (*(size_t *)a->mv_data < *(size_t *)b->mv_data) ? -1 :
  4758. *(size_t *)a->mv_data > *(size_t *)b->mv_data;
  4759. }
  4760. /** Compare two items pointing at aligned unsigned int's.
  4761. *
  4762. * This is also set as #MDB_INTEGERDUP|#MDB_DUPFIXED's #MDB_dbx.%md_dcmp,
  4763. * but #mdb_cmp_clong() is called instead if the data type is size_t.
  4764. */
  4765. static int
  4766. mdb_cmp_int(const MDB_val *a, const MDB_val *b)
  4767. {
  4768. return (*(unsigned int *)a->mv_data < *(unsigned int *)b->mv_data) ? -1 :
  4769. *(unsigned int *)a->mv_data > *(unsigned int *)b->mv_data;
  4770. }
  4771. /** Compare two items pointing at unsigned ints of unknown alignment.
  4772. * Nodes and keys are guaranteed to be 2-byte aligned.
  4773. */
  4774. static int
  4775. mdb_cmp_cint(const MDB_val *a, const MDB_val *b)
  4776. {
  4777. #if BYTE_ORDER == LITTLE_ENDIAN
  4778. unsigned short *u, *c;
  4779. int x;
  4780. u = (unsigned short *) ((char *) a->mv_data + a->mv_size);
  4781. c = (unsigned short *) ((char *) b->mv_data + a->mv_size);
  4782. do {
  4783. x = *--u - *--c;
  4784. } while(!x && u > (unsigned short *)a->mv_data);
  4785. return x;
  4786. #else
  4787. unsigned short *u, *c, *end;
  4788. int x;
  4789. end = (unsigned short *) ((char *) a->mv_data + a->mv_size);
  4790. u = (unsigned short *)a->mv_data;
  4791. c = (unsigned short *)b->mv_data;
  4792. do {
  4793. x = *u++ - *c++;
  4794. } while(!x && u < end);
  4795. return x;
  4796. #endif
  4797. }
  4798. /** Compare two items lexically */
  4799. static int
  4800. mdb_cmp_memn(const MDB_val *a, const MDB_val *b)
  4801. {
  4802. int diff;
  4803. ssize_t len_diff;
  4804. unsigned int len;
  4805. len = a->mv_size;
  4806. len_diff = (ssize_t) a->mv_size - (ssize_t) b->mv_size;
  4807. if (len_diff > 0) {
  4808. len = b->mv_size;
  4809. len_diff = 1;
  4810. }
  4811. diff = memcmp(a->mv_data, b->mv_data, len);
  4812. return diff ? diff : len_diff<0 ? -1 : len_diff;
  4813. }
  4814. /** Compare two items in reverse byte order */
  4815. static int
  4816. mdb_cmp_memnr(const MDB_val *a, const MDB_val *b)
  4817. {
  4818. const unsigned char *p1, *p2, *p1_lim;
  4819. ssize_t len_diff;
  4820. int diff;
  4821. p1_lim = (const unsigned char *)a->mv_data;
  4822. p1 = (const unsigned char *)a->mv_data + a->mv_size;
  4823. p2 = (const unsigned char *)b->mv_data + b->mv_size;
  4824. len_diff = (ssize_t) a->mv_size - (ssize_t) b->mv_size;
  4825. if (len_diff > 0) {
  4826. p1_lim += len_diff;
  4827. len_diff = 1;
  4828. }
  4829. while (p1 > p1_lim) {
  4830. diff = *--p1 - *--p2;
  4831. if (diff)
  4832. return diff;
  4833. }
  4834. return len_diff<0 ? -1 : len_diff;
  4835. }
  4836. /** Search for key within a page, using binary search.
  4837. * Returns the smallest entry larger or equal to the key.
  4838. * If exactp is non-null, stores whether the found entry was an exact match
  4839. * in *exactp (1 or 0).
  4840. * Updates the cursor index with the index of the found entry.
  4841. * If no entry larger or equal to the key is found, returns NULL.
  4842. */
  4843. static MDB_node *
  4844. mdb_node_search(MDB_cursor *mc, MDB_val *key, int *exactp)
  4845. {
  4846. unsigned int i = 0, nkeys;
  4847. int low, high;
  4848. int rc = 0;
  4849. MDB_page *mp = mc->mc_pg[mc->mc_top];
  4850. MDB_node *node = NULL;
  4851. MDB_val nodekey;
  4852. MDB_cmp_func *cmp;
  4853. DKBUF;
  4854. nkeys = NUMKEYS(mp);
  4855. DPRINTF(("searching %u keys in %s %spage %"Z"u",
  4856. nkeys, IS_LEAF(mp) ? "leaf" : "branch", IS_SUBP(mp) ? "sub-" : "",
  4857. mdb_dbg_pgno(mp)));
  4858. low = IS_LEAF(mp) ? 0 : 1;
  4859. high = nkeys - 1;
  4860. cmp = mc->mc_dbx->md_cmp;
  4861. /* Branch pages have no data, so if using integer keys,
  4862. * alignment is guaranteed. Use faster mdb_cmp_int.
  4863. */
  4864. if (cmp == mdb_cmp_cint && IS_BRANCH(mp)) {
  4865. if (NODEPTR(mp, 1)->mn_ksize == sizeof(size_t))
  4866. cmp = mdb_cmp_long;
  4867. else
  4868. cmp = mdb_cmp_int;
  4869. }
  4870. if (IS_LEAF2(mp)) {
  4871. nodekey.mv_size = mc->mc_db->md_pad;
  4872. node = NODEPTR(mp, 0); /* fake */
  4873. while (low <= high) {
  4874. i = (low + high) >> 1;
  4875. nodekey.mv_data = LEAF2KEY(mp, i, nodekey.mv_size);
  4876. rc = cmp(key, &nodekey);
  4877. DPRINTF(("found leaf index %u [%s], rc = %i",
  4878. i, DKEY(&nodekey), rc));
  4879. if (rc == 0)
  4880. break;
  4881. if (rc > 0)
  4882. low = i + 1;
  4883. else
  4884. high = i - 1;
  4885. }
  4886. } else {
  4887. while (low <= high) {
  4888. i = (low + high) >> 1;
  4889. node = NODEPTR(mp, i);
  4890. nodekey.mv_size = NODEKSZ(node);
  4891. nodekey.mv_data = NODEKEY(node);
  4892. rc = cmp(key, &nodekey);
  4893. #if MDB_DEBUG
  4894. if (IS_LEAF(mp))
  4895. DPRINTF(("found leaf index %u [%s], rc = %i",
  4896. i, DKEY(&nodekey), rc));
  4897. else
  4898. DPRINTF(("found branch index %u [%s -> %"Z"u], rc = %i",
  4899. i, DKEY(&nodekey), NODEPGNO(node), rc));
  4900. #endif
  4901. if (rc == 0)
  4902. break;
  4903. if (rc > 0)
  4904. low = i + 1;
  4905. else
  4906. high = i - 1;
  4907. }
  4908. }
  4909. if (rc > 0) { /* Found entry is less than the key. */
  4910. i++; /* Skip to get the smallest entry larger than key. */
  4911. if (!IS_LEAF2(mp))
  4912. node = NODEPTR(mp, i);
  4913. }
  4914. if (exactp)
  4915. *exactp = (rc == 0 && nkeys > 0);
  4916. /* store the key index */
  4917. mc->mc_ki[mc->mc_top] = i;
  4918. if (i >= nkeys)
  4919. /* There is no entry larger or equal to the key. */
  4920. return NULL;
  4921. /* nodeptr is fake for LEAF2 */
  4922. return node;
  4923. }
  4924. #if 0
  4925. static void
  4926. mdb_cursor_adjust(MDB_cursor *mc, func)
  4927. {
  4928. MDB_cursor *m2;
  4929. for (m2 = mc->mc_txn->mt_cursors[mc->mc_dbi]; m2; m2=m2->mc_next) {
  4930. if (m2->mc_pg[m2->mc_top] == mc->mc_pg[mc->mc_top]) {
  4931. func(mc, m2);
  4932. }
  4933. }
  4934. }
  4935. #endif
  4936. /** Pop a page off the top of the cursor's stack. */
  4937. static void
  4938. mdb_cursor_pop(MDB_cursor *mc)
  4939. {
  4940. if (mc->mc_snum) {
  4941. DPRINTF(("popping page %"Z"u off db %d cursor %p",
  4942. mc->mc_pg[mc->mc_top]->mp_pgno, DDBI(mc), (void *) mc));
  4943. mc->mc_snum--;
  4944. if (mc->mc_snum) {
  4945. mc->mc_top--;
  4946. } else {
  4947. mc->mc_flags &= ~C_INITIALIZED;
  4948. }
  4949. }
  4950. }
  4951. /** Push a page onto the top of the cursor's stack.
  4952. * Set #MDB_TXN_ERROR on failure.
  4953. */
  4954. static int
  4955. mdb_cursor_push(MDB_cursor *mc, MDB_page *mp)
  4956. {
  4957. DPRINTF(("pushing page %"Z"u on db %d cursor %p", mp->mp_pgno,
  4958. DDBI(mc), (void *) mc));
  4959. if (mc->mc_snum >= CURSOR_STACK) {
  4960. mc->mc_txn->mt_flags |= MDB_TXN_ERROR;
  4961. return MDB_CURSOR_FULL;
  4962. }
  4963. mc->mc_top = mc->mc_snum++;
  4964. mc->mc_pg[mc->mc_top] = mp;
  4965. mc->mc_ki[mc->mc_top] = 0;
  4966. return MDB_SUCCESS;
  4967. }
  4968. /** Find the address of the page corresponding to a given page number.
  4969. * Set #MDB_TXN_ERROR on failure.
  4970. * @param[in] mc the cursor accessing the page.
  4971. * @param[in] pgno the page number for the page to retrieve.
  4972. * @param[out] ret address of a pointer where the page's address will be stored.
  4973. * @param[out] lvl dirty_list inheritance level of found page. 1=current txn, 0=mapped page.
  4974. * @return 0 on success, non-zero on failure.
  4975. */
  4976. static int
  4977. mdb_page_get(MDB_cursor *mc, pgno_t pgno, MDB_page **ret, int *lvl)
  4978. {
  4979. MDB_txn *txn = mc->mc_txn;
  4980. MDB_env *env = txn->mt_env;
  4981. MDB_page *p = NULL;
  4982. int level;
  4983. if (! (txn->mt_flags & (MDB_TXN_RDONLY|MDB_TXN_WRITEMAP))) {
  4984. MDB_txn *tx2 = txn;
  4985. level = 1;
  4986. do {
  4987. MDB_ID2L dl = tx2->mt_u.dirty_list;
  4988. unsigned x;
  4989. /* Spilled pages were dirtied in this txn and flushed
  4990. * because the dirty list got full. Bring this page
  4991. * back in from the map (but don't unspill it here,
  4992. * leave that unless page_touch happens again).
  4993. */
  4994. if (tx2->mt_spill_pgs) {
  4995. MDB_ID pn = pgno << 1;
  4996. x = mdb_midl_search(tx2->mt_spill_pgs, pn);
  4997. if (x <= tx2->mt_spill_pgs[0] && tx2->mt_spill_pgs[x] == pn) {
  4998. p = (MDB_page *)(env->me_map + env->me_psize * pgno);
  4999. goto done;
  5000. }
  5001. }
  5002. if (dl[0].mid) {
  5003. unsigned x = mdb_mid2l_search(dl, pgno);
  5004. if (x <= dl[0].mid && dl[x].mid == pgno) {
  5005. p = dl[x].mptr;
  5006. goto done;
  5007. }
  5008. }
  5009. level++;
  5010. } while ((tx2 = tx2->mt_parent) != NULL);
  5011. }
  5012. if (pgno < txn->mt_next_pgno) {
  5013. level = 0;
  5014. p = (MDB_page *)(env->me_map + env->me_psize * pgno);
  5015. } else {
  5016. DPRINTF(("page %"Z"u not found", pgno));
  5017. txn->mt_flags |= MDB_TXN_ERROR;
  5018. return MDB_PAGE_NOTFOUND;
  5019. }
  5020. done:
  5021. *ret = p;
  5022. if (lvl)
  5023. *lvl = level;
  5024. return MDB_SUCCESS;
  5025. }
  5026. /** Finish #mdb_page_search() / #mdb_page_search_lowest().
  5027. * The cursor is at the root page, set up the rest of it.
  5028. */
  5029. static int
  5030. mdb_page_search_root(MDB_cursor *mc, MDB_val *key, int flags)
  5031. {
  5032. MDB_page *mp = mc->mc_pg[mc->mc_top];
  5033. int rc;
  5034. DKBUF;
  5035. while (IS_BRANCH(mp)) {
  5036. MDB_node *node;
  5037. indx_t i;
  5038. DPRINTF(("branch page %"Z"u has %u keys", mp->mp_pgno, NUMKEYS(mp)));
  5039. /* Don't assert on branch pages in the FreeDB. We can get here
  5040. * while in the process of rebalancing a FreeDB branch page; we must
  5041. * let that proceed. ITS#8336
  5042. */
  5043. mdb_cassert(mc, !mc->mc_dbi || NUMKEYS(mp) > 1);
  5044. DPRINTF(("found index 0 to page %"Z"u", NODEPGNO(NODEPTR(mp, 0))));
  5045. if (flags & (MDB_PS_FIRST|MDB_PS_LAST)) {
  5046. i = 0;
  5047. if (flags & MDB_PS_LAST) {
  5048. i = NUMKEYS(mp) - 1;
  5049. /* if already init'd, see if we're already in right place */
  5050. if (mc->mc_flags & C_INITIALIZED) {
  5051. if (mc->mc_ki[mc->mc_top] == i) {
  5052. mc->mc_top = mc->mc_snum++;
  5053. mp = mc->mc_pg[mc->mc_top];
  5054. goto ready;
  5055. }
  5056. }
  5057. }
  5058. } else {
  5059. int exact;
  5060. node = mdb_node_search(mc, key, &exact);
  5061. if (node == NULL)
  5062. i = NUMKEYS(mp) - 1;
  5063. else {
  5064. i = mc->mc_ki[mc->mc_top];
  5065. if (!exact) {
  5066. mdb_cassert(mc, i > 0);
  5067. i--;
  5068. }
  5069. }
  5070. DPRINTF(("following index %u for key [%s]", i, DKEY(key)));
  5071. }
  5072. mdb_cassert(mc, i < NUMKEYS(mp));
  5073. node = NODEPTR(mp, i);
  5074. if ((rc = mdb_page_get(mc, NODEPGNO(node), &mp, NULL)) != 0)
  5075. return rc;
  5076. mc->mc_ki[mc->mc_top] = i;
  5077. if ((rc = mdb_cursor_push(mc, mp)))
  5078. return rc;
  5079. ready:
  5080. if (flags & MDB_PS_MODIFY) {
  5081. if ((rc = mdb_page_touch(mc)) != 0)
  5082. return rc;
  5083. mp = mc->mc_pg[mc->mc_top];
  5084. }
  5085. }
  5086. if (!IS_LEAF(mp)) {
  5087. DPRINTF(("internal error, index points to a %02X page!?",
  5088. mp->mp_flags));
  5089. mc->mc_txn->mt_flags |= MDB_TXN_ERROR;
  5090. return MDB_CORRUPTED;
  5091. }
  5092. DPRINTF(("found leaf page %"Z"u for key [%s]", mp->mp_pgno,
  5093. key ? DKEY(key) : "null"));
  5094. mc->mc_flags |= C_INITIALIZED;
  5095. mc->mc_flags &= ~C_EOF;
  5096. return MDB_SUCCESS;
  5097. }
  5098. /** Search for the lowest key under the current branch page.
  5099. * This just bypasses a NUMKEYS check in the current page
  5100. * before calling mdb_page_search_root(), because the callers
  5101. * are all in situations where the current page is known to
  5102. * be underfilled.
  5103. */
  5104. static int
  5105. mdb_page_search_lowest(MDB_cursor *mc)
  5106. {
  5107. MDB_page *mp = mc->mc_pg[mc->mc_top];
  5108. MDB_node *node = NODEPTR(mp, 0);
  5109. int rc;
  5110. if ((rc = mdb_page_get(mc, NODEPGNO(node), &mp, NULL)) != 0)
  5111. return rc;
  5112. mc->mc_ki[mc->mc_top] = 0;
  5113. if ((rc = mdb_cursor_push(mc, mp)))
  5114. return rc;
  5115. return mdb_page_search_root(mc, NULL, MDB_PS_FIRST);
  5116. }
  5117. /** Search for the page a given key should be in.
  5118. * Push it and its parent pages on the cursor stack.
  5119. * @param[in,out] mc the cursor for this operation.
  5120. * @param[in] key the key to search for, or NULL for first/last page.
  5121. * @param[in] flags If MDB_PS_MODIFY is set, visited pages in the DB
  5122. * are touched (updated with new page numbers).
  5123. * If MDB_PS_FIRST or MDB_PS_LAST is set, find first or last leaf.
  5124. * This is used by #mdb_cursor_first() and #mdb_cursor_last().
  5125. * If MDB_PS_ROOTONLY set, just fetch root node, no further lookups.
  5126. * @return 0 on success, non-zero on failure.
  5127. */
  5128. static int
  5129. mdb_page_search(MDB_cursor *mc, MDB_val *key, int flags)
  5130. {
  5131. int rc;
  5132. pgno_t root;
  5133. /* Make sure the txn is still viable, then find the root from
  5134. * the txn's db table and set it as the root of the cursor's stack.
  5135. */
  5136. if (mc->mc_txn->mt_flags & MDB_TXN_BLOCKED) {
  5137. DPUTS("transaction may not be used now");
  5138. return MDB_BAD_TXN;
  5139. } else {
  5140. /* Make sure we're using an up-to-date root */
  5141. if (*mc->mc_dbflag & DB_STALE) {
  5142. MDB_cursor mc2;
  5143. if (TXN_DBI_CHANGED(mc->mc_txn, mc->mc_dbi))
  5144. return MDB_BAD_DBI;
  5145. mdb_cursor_init(&mc2, mc->mc_txn, MAIN_DBI, NULL);
  5146. rc = mdb_page_search(&mc2, &mc->mc_dbx->md_name, 0);
  5147. if (rc)
  5148. return rc;
  5149. {
  5150. MDB_val data;
  5151. int exact = 0;
  5152. uint16_t flags;
  5153. MDB_node *leaf = mdb_node_search(&mc2,
  5154. &mc->mc_dbx->md_name, &exact);
  5155. if (!exact)
  5156. return MDB_NOTFOUND;
  5157. if ((leaf->mn_flags & (F_DUPDATA|F_SUBDATA)) != F_SUBDATA)
  5158. return MDB_INCOMPATIBLE; /* not a named DB */
  5159. rc = mdb_node_read(&mc2, leaf, &data);
  5160. if (rc)
  5161. return rc;
  5162. memcpy(&flags, ((char *) data.mv_data + offsetof(MDB_db, md_flags)),
  5163. sizeof(uint16_t));
  5164. /* The txn may not know this DBI, or another process may
  5165. * have dropped and recreated the DB with other flags.
  5166. */
  5167. if ((mc->mc_db->md_flags & PERSISTENT_FLAGS) != flags)
  5168. return MDB_INCOMPATIBLE;
  5169. memcpy(mc->mc_db, data.mv_data, sizeof(MDB_db));
  5170. }
  5171. *mc->mc_dbflag &= ~DB_STALE;
  5172. }
  5173. root = mc->mc_db->md_root;
  5174. if (root == P_INVALID) { /* Tree is empty. */
  5175. DPUTS("tree is empty");
  5176. return MDB_NOTFOUND;
  5177. }
  5178. }
  5179. mdb_cassert(mc, root > 1);
  5180. if (!mc->mc_pg[0] || mc->mc_pg[0]->mp_pgno != root)
  5181. if ((rc = mdb_page_get(mc, root, &mc->mc_pg[0], NULL)) != 0)
  5182. return rc;
  5183. mc->mc_snum = 1;
  5184. mc->mc_top = 0;
  5185. DPRINTF(("db %d root page %"Z"u has flags 0x%X",
  5186. DDBI(mc), root, mc->mc_pg[0]->mp_flags));
  5187. if (flags & MDB_PS_MODIFY) {
  5188. if ((rc = mdb_page_touch(mc)))
  5189. return rc;
  5190. }
  5191. if (flags & MDB_PS_ROOTONLY)
  5192. return MDB_SUCCESS;
  5193. return mdb_page_search_root(mc, key, flags);
  5194. }
  5195. static int
  5196. mdb_ovpage_free(MDB_cursor *mc, MDB_page *mp)
  5197. {
  5198. MDB_txn *txn = mc->mc_txn;
  5199. pgno_t pg = mp->mp_pgno;
  5200. unsigned x = 0, ovpages = mp->mp_pages;
  5201. MDB_env *env = txn->mt_env;
  5202. MDB_IDL sl = txn->mt_spill_pgs;
  5203. MDB_ID pn = pg << 1;
  5204. int rc;
  5205. DPRINTF(("free ov page %"Z"u (%d)", pg, ovpages));
  5206. /* If the page is dirty or on the spill list we just acquired it,
  5207. * so we should give it back to our current free list, if any.
  5208. * Otherwise put it onto the list of pages we freed in this txn.
  5209. *
  5210. * Won't create me_pghead: me_pglast must be inited along with it.
  5211. * Unsupported in nested txns: They would need to hide the page
  5212. * range in ancestor txns' dirty and spilled lists.
  5213. */
  5214. if (env->me_pghead &&
  5215. !txn->mt_parent &&
  5216. ((mp->mp_flags & P_DIRTY) ||
  5217. (sl && (x = mdb_midl_search(sl, pn)) <= sl[0] && sl[x] == pn)))
  5218. {
  5219. unsigned i, j;
  5220. pgno_t *mop;
  5221. MDB_ID2 *dl, ix, iy;
  5222. rc = mdb_midl_need(&env->me_pghead, ovpages);
  5223. if (rc)
  5224. return rc;
  5225. if (!(mp->mp_flags & P_DIRTY)) {
  5226. /* This page is no longer spilled */
  5227. if (x == sl[0])
  5228. sl[0]--;
  5229. else
  5230. sl[x] |= 1;
  5231. goto release;
  5232. }
  5233. /* Remove from dirty list */
  5234. dl = txn->mt_u.dirty_list;
  5235. x = dl[0].mid--;
  5236. for (ix = dl[x]; ix.mptr != mp; ix = iy) {
  5237. if (x > 1) {
  5238. x--;
  5239. iy = dl[x];
  5240. dl[x] = ix;
  5241. } else {
  5242. mdb_cassert(mc, x > 1);
  5243. j = ++(dl[0].mid);
  5244. dl[j] = ix; /* Unsorted. OK when MDB_TXN_ERROR. */
  5245. txn->mt_flags |= MDB_TXN_ERROR;
  5246. return MDB_CORRUPTED;
  5247. }
  5248. }
  5249. txn->mt_dirty_room++;
  5250. if (!(env->me_flags & MDB_WRITEMAP))
  5251. mdb_dpage_free(env, mp);
  5252. release:
  5253. /* Insert in me_pghead */
  5254. mop = env->me_pghead;
  5255. j = mop[0] + ovpages;
  5256. for (i = mop[0]; i && mop[i] < pg; i--)
  5257. mop[j--] = mop[i];
  5258. while (j>i)
  5259. mop[j--] = pg++;
  5260. mop[0] += ovpages;
  5261. } else {
  5262. rc = mdb_midl_append_range(&txn->mt_free_pgs, pg, ovpages);
  5263. if (rc)
  5264. return rc;
  5265. }
  5266. mc->mc_db->md_overflow_pages -= ovpages;
  5267. return 0;
  5268. }
  5269. /** Return the data associated with a given node.
  5270. * @param[in] mc The cursor for this operation.
  5271. * @param[in] leaf The node being read.
  5272. * @param[out] data Updated to point to the node's data.
  5273. * @return 0 on success, non-zero on failure.
  5274. */
  5275. static int
  5276. mdb_node_read(MDB_cursor *mc, MDB_node *leaf, MDB_val *data)
  5277. {
  5278. MDB_page *omp; /* overflow page */
  5279. pgno_t pgno;
  5280. int rc;
  5281. if (!F_ISSET(leaf->mn_flags, F_BIGDATA)) {
  5282. data->mv_size = NODEDSZ(leaf);
  5283. data->mv_data = NODEDATA(leaf);
  5284. return MDB_SUCCESS;
  5285. }
  5286. /* Read overflow data.
  5287. */
  5288. data->mv_size = NODEDSZ(leaf);
  5289. memcpy(&pgno, NODEDATA(leaf), sizeof(pgno));
  5290. if ((rc = mdb_page_get(mc, pgno, &omp, NULL)) != 0) {
  5291. DPRINTF(("read overflow page %"Z"u failed", pgno));
  5292. return rc;
  5293. }
  5294. data->mv_data = METADATA(omp);
  5295. return MDB_SUCCESS;
  5296. }
  5297. int
  5298. mdb_get(MDB_txn *txn, MDB_dbi dbi,
  5299. MDB_val *key, MDB_val *data)
  5300. {
  5301. MDB_cursor mc;
  5302. MDB_xcursor mx;
  5303. int exact = 0;
  5304. DKBUF;
  5305. DPRINTF(("===> get db %u key [%s]", dbi, DKEY(key)));
  5306. if (!key || !data || !TXN_DBI_EXIST(txn, dbi, DB_USRVALID))
  5307. return EINVAL;
  5308. if (txn->mt_flags & MDB_TXN_BLOCKED)
  5309. return MDB_BAD_TXN;
  5310. mdb_cursor_init(&mc, txn, dbi, &mx);
  5311. return mdb_cursor_set(&mc, key, data, MDB_SET, &exact);
  5312. }
  5313. /** Find a sibling for a page.
  5314. * Replaces the page at the top of the cursor's stack with the
  5315. * specified sibling, if one exists.
  5316. * @param[in] mc The cursor for this operation.
  5317. * @param[in] move_right Non-zero if the right sibling is requested,
  5318. * otherwise the left sibling.
  5319. * @return 0 on success, non-zero on failure.
  5320. */
  5321. static int
  5322. mdb_cursor_sibling(MDB_cursor *mc, int move_right)
  5323. {
  5324. int rc;
  5325. MDB_node *indx;
  5326. MDB_page *mp;
  5327. if (mc->mc_snum < 2) {
  5328. return MDB_NOTFOUND; /* root has no siblings */
  5329. }
  5330. mdb_cursor_pop(mc);
  5331. DPRINTF(("parent page is page %"Z"u, index %u",
  5332. mc->mc_pg[mc->mc_top]->mp_pgno, mc->mc_ki[mc->mc_top]));
  5333. if (move_right ? (mc->mc_ki[mc->mc_top] + 1u >= NUMKEYS(mc->mc_pg[mc->mc_top]))
  5334. : (mc->mc_ki[mc->mc_top] == 0)) {
  5335. DPRINTF(("no more keys left, moving to %s sibling",
  5336. move_right ? "right" : "left"));
  5337. if ((rc = mdb_cursor_sibling(mc, move_right)) != MDB_SUCCESS) {
  5338. /* undo cursor_pop before returning */
  5339. mc->mc_top++;
  5340. mc->mc_snum++;
  5341. return rc;
  5342. }
  5343. } else {
  5344. if (move_right)
  5345. mc->mc_ki[mc->mc_top]++;
  5346. else
  5347. mc->mc_ki[mc->mc_top]--;
  5348. DPRINTF(("just moving to %s index key %u",
  5349. move_right ? "right" : "left", mc->mc_ki[mc->mc_top]));
  5350. }
  5351. mdb_cassert(mc, IS_BRANCH(mc->mc_pg[mc->mc_top]));
  5352. indx = NODEPTR(mc->mc_pg[mc->mc_top], mc->mc_ki[mc->mc_top]);
  5353. if ((rc = mdb_page_get(mc, NODEPGNO(indx), &mp, NULL)) != 0) {
  5354. /* mc will be inconsistent if caller does mc_snum++ as above */
  5355. mc->mc_flags &= ~(C_INITIALIZED|C_EOF);
  5356. return rc;
  5357. }
  5358. mdb_cursor_push(mc, mp);
  5359. if (!move_right)
  5360. mc->mc_ki[mc->mc_top] = NUMKEYS(mp)-1;
  5361. return MDB_SUCCESS;
  5362. }
  5363. /** Move the cursor to the next data item. */
  5364. static int
  5365. mdb_cursor_next(MDB_cursor *mc, MDB_val *key, MDB_val *data, MDB_cursor_op op)
  5366. {
  5367. MDB_page *mp;
  5368. MDB_node *leaf;
  5369. int rc;
  5370. if ((mc->mc_flags & C_DEL && op == MDB_NEXT_DUP))
  5371. return MDB_NOTFOUND;
  5372. if (!(mc->mc_flags & C_INITIALIZED))
  5373. return mdb_cursor_first(mc, key, data);
  5374. mp = mc->mc_pg[mc->mc_top];
  5375. if (mc->mc_flags & C_EOF) {
  5376. if (mc->mc_ki[mc->mc_top] >= NUMKEYS(mp)-1)
  5377. return MDB_NOTFOUND;
  5378. mc->mc_flags ^= C_EOF;
  5379. }
  5380. if (mc->mc_db->md_flags & MDB_DUPSORT) {
  5381. leaf = NODEPTR(mp, mc->mc_ki[mc->mc_top]);
  5382. if (F_ISSET(leaf->mn_flags, F_DUPDATA)) {
  5383. if (op == MDB_NEXT || op == MDB_NEXT_DUP) {
  5384. rc = mdb_cursor_next(&mc->mc_xcursor->mx_cursor, data, NULL, MDB_NEXT);
  5385. if (op != MDB_NEXT || rc != MDB_NOTFOUND) {
  5386. if (rc == MDB_SUCCESS)
  5387. MDB_GET_KEY(leaf, key);
  5388. return rc;
  5389. }
  5390. }
  5391. } else {
  5392. mc->mc_xcursor->mx_cursor.mc_flags &= ~(C_INITIALIZED|C_EOF);
  5393. if (op == MDB_NEXT_DUP)
  5394. return MDB_NOTFOUND;
  5395. }
  5396. }
  5397. DPRINTF(("cursor_next: top page is %"Z"u in cursor %p",
  5398. mdb_dbg_pgno(mp), (void *) mc));
  5399. if (mc->mc_flags & C_DEL) {
  5400. mc->mc_flags ^= C_DEL;
  5401. goto skip;
  5402. }
  5403. if (mc->mc_ki[mc->mc_top] + 1u >= NUMKEYS(mp)) {
  5404. DPUTS("=====> move to next sibling page");
  5405. if ((rc = mdb_cursor_sibling(mc, 1)) != MDB_SUCCESS) {
  5406. mc->mc_flags |= C_EOF;
  5407. return rc;
  5408. }
  5409. mp = mc->mc_pg[mc->mc_top];
  5410. DPRINTF(("next page is %"Z"u, key index %u", mp->mp_pgno, mc->mc_ki[mc->mc_top]));
  5411. } else
  5412. mc->mc_ki[mc->mc_top]++;
  5413. skip:
  5414. DPRINTF(("==> cursor points to page %"Z"u with %u keys, key index %u",
  5415. mdb_dbg_pgno(mp), NUMKEYS(mp), mc->mc_ki[mc->mc_top]));
  5416. if (IS_LEAF2(mp)) {
  5417. key->mv_size = mc->mc_db->md_pad;
  5418. key->mv_data = LEAF2KEY(mp, mc->mc_ki[mc->mc_top], key->mv_size);
  5419. return MDB_SUCCESS;
  5420. }
  5421. mdb_cassert(mc, IS_LEAF(mp));
  5422. leaf = NODEPTR(mp, mc->mc_ki[mc->mc_top]);
  5423. if (F_ISSET(leaf->mn_flags, F_DUPDATA)) {
  5424. mdb_xcursor_init1(mc, leaf);
  5425. rc = mdb_cursor_first(&mc->mc_xcursor->mx_cursor, data, NULL);
  5426. if (rc != MDB_SUCCESS)
  5427. return rc;
  5428. } else if (data) {
  5429. if ((rc = mdb_node_read(mc, leaf, data)) != MDB_SUCCESS)
  5430. return rc;
  5431. }
  5432. MDB_GET_KEY(leaf, key);
  5433. return MDB_SUCCESS;
  5434. }
  5435. /** Move the cursor to the previous data item. */
  5436. static int
  5437. mdb_cursor_prev(MDB_cursor *mc, MDB_val *key, MDB_val *data, MDB_cursor_op op)
  5438. {
  5439. MDB_page *mp;
  5440. MDB_node *leaf;
  5441. int rc;
  5442. if (!(mc->mc_flags & C_INITIALIZED)) {
  5443. rc = mdb_cursor_last(mc, key, data);
  5444. if (rc)
  5445. return rc;
  5446. mc->mc_ki[mc->mc_top]++;
  5447. }
  5448. mp = mc->mc_pg[mc->mc_top];
  5449. if ((mc->mc_db->md_flags & MDB_DUPSORT) &&
  5450. mc->mc_ki[mc->mc_top] < NUMKEYS(mp)) {
  5451. leaf = NODEPTR(mp, mc->mc_ki[mc->mc_top]);
  5452. if (F_ISSET(leaf->mn_flags, F_DUPDATA)) {
  5453. if (op == MDB_PREV || op == MDB_PREV_DUP) {
  5454. rc = mdb_cursor_prev(&mc->mc_xcursor->mx_cursor, data, NULL, MDB_PREV);
  5455. if (op != MDB_PREV || rc != MDB_NOTFOUND) {
  5456. if (rc == MDB_SUCCESS) {
  5457. MDB_GET_KEY(leaf, key);
  5458. mc->mc_flags &= ~C_EOF;
  5459. }
  5460. return rc;
  5461. }
  5462. }
  5463. } else {
  5464. mc->mc_xcursor->mx_cursor.mc_flags &= ~(C_INITIALIZED|C_EOF);
  5465. if (op == MDB_PREV_DUP)
  5466. return MDB_NOTFOUND;
  5467. }
  5468. }
  5469. DPRINTF(("cursor_prev: top page is %"Z"u in cursor %p",
  5470. mdb_dbg_pgno(mp), (void *) mc));
  5471. mc->mc_flags &= ~(C_EOF|C_DEL);
  5472. if (mc->mc_ki[mc->mc_top] == 0) {
  5473. DPUTS("=====> move to prev sibling page");
  5474. if ((rc = mdb_cursor_sibling(mc, 0)) != MDB_SUCCESS) {
  5475. return rc;
  5476. }
  5477. mp = mc->mc_pg[mc->mc_top];
  5478. mc->mc_ki[mc->mc_top] = NUMKEYS(mp) - 1;
  5479. DPRINTF(("prev page is %"Z"u, key index %u", mp->mp_pgno, mc->mc_ki[mc->mc_top]));
  5480. } else
  5481. mc->mc_ki[mc->mc_top]--;
  5482. DPRINTF(("==> cursor points to page %"Z"u with %u keys, key index %u",
  5483. mdb_dbg_pgno(mp), NUMKEYS(mp), mc->mc_ki[mc->mc_top]));
  5484. if (!IS_LEAF(mp))
  5485. return MDB_CORRUPTED;
  5486. if (IS_LEAF2(mp)) {
  5487. key->mv_size = mc->mc_db->md_pad;
  5488. key->mv_data = LEAF2KEY(mp, mc->mc_ki[mc->mc_top], key->mv_size);
  5489. return MDB_SUCCESS;
  5490. }
  5491. leaf = NODEPTR(mp, mc->mc_ki[mc->mc_top]);
  5492. if (F_ISSET(leaf->mn_flags, F_DUPDATA)) {
  5493. mdb_xcursor_init1(mc, leaf);
  5494. rc = mdb_cursor_last(&mc->mc_xcursor->mx_cursor, data, NULL);
  5495. if (rc != MDB_SUCCESS)
  5496. return rc;
  5497. } else if (data) {
  5498. if ((rc = mdb_node_read(mc, leaf, data)) != MDB_SUCCESS)
  5499. return rc;
  5500. }
  5501. MDB_GET_KEY(leaf, key);
  5502. return MDB_SUCCESS;
  5503. }
  5504. /** Set the cursor on a specific data item. */
  5505. static int
  5506. mdb_cursor_set(MDB_cursor *mc, MDB_val *key, MDB_val *data,
  5507. MDB_cursor_op op, int *exactp)
  5508. {
  5509. int rc;
  5510. MDB_page *mp;
  5511. MDB_node *leaf = NULL;
  5512. DKBUF;
  5513. if (key->mv_size == 0)
  5514. return MDB_BAD_VALSIZE;
  5515. if (mc->mc_xcursor)
  5516. mc->mc_xcursor->mx_cursor.mc_flags &= ~(C_INITIALIZED|C_EOF);
  5517. /* See if we're already on the right page */
  5518. if (mc->mc_flags & C_INITIALIZED) {
  5519. MDB_val nodekey;
  5520. mp = mc->mc_pg[mc->mc_top];
  5521. if (!NUMKEYS(mp)) {
  5522. mc->mc_ki[mc->mc_top] = 0;
  5523. return MDB_NOTFOUND;
  5524. }
  5525. if (mp->mp_flags & P_LEAF2) {
  5526. nodekey.mv_size = mc->mc_db->md_pad;
  5527. nodekey.mv_data = LEAF2KEY(mp, 0, nodekey.mv_size);
  5528. } else {
  5529. leaf = NODEPTR(mp, 0);
  5530. MDB_GET_KEY2(leaf, nodekey);
  5531. }
  5532. rc = mc->mc_dbx->md_cmp(key, &nodekey);
  5533. if (rc == 0) {
  5534. /* Probably happens rarely, but first node on the page
  5535. * was the one we wanted.
  5536. */
  5537. mc->mc_ki[mc->mc_top] = 0;
  5538. if (exactp)
  5539. *exactp = 1;
  5540. goto set1;
  5541. }
  5542. if (rc > 0) {
  5543. unsigned int i;
  5544. unsigned int nkeys = NUMKEYS(mp);
  5545. if (nkeys > 1) {
  5546. if (mp->mp_flags & P_LEAF2) {
  5547. nodekey.mv_data = LEAF2KEY(mp,
  5548. nkeys-1, nodekey.mv_size);
  5549. } else {
  5550. leaf = NODEPTR(mp, nkeys-1);
  5551. MDB_GET_KEY2(leaf, nodekey);
  5552. }
  5553. rc = mc->mc_dbx->md_cmp(key, &nodekey);
  5554. if (rc == 0) {
  5555. /* last node was the one we wanted */
  5556. mc->mc_ki[mc->mc_top] = nkeys-1;
  5557. if (exactp)
  5558. *exactp = 1;
  5559. goto set1;
  5560. }
  5561. if (rc < 0) {
  5562. if (mc->mc_ki[mc->mc_top] < NUMKEYS(mp)) {
  5563. /* This is definitely the right page, skip search_page */
  5564. if (mp->mp_flags & P_LEAF2) {
  5565. nodekey.mv_data = LEAF2KEY(mp,
  5566. mc->mc_ki[mc->mc_top], nodekey.mv_size);
  5567. } else {
  5568. leaf = NODEPTR(mp, mc->mc_ki[mc->mc_top]);
  5569. MDB_GET_KEY2(leaf, nodekey);
  5570. }
  5571. rc = mc->mc_dbx->md_cmp(key, &nodekey);
  5572. if (rc == 0) {
  5573. /* current node was the one we wanted */
  5574. if (exactp)
  5575. *exactp = 1;
  5576. goto set1;
  5577. }
  5578. }
  5579. rc = 0;
  5580. mc->mc_flags &= ~C_EOF;
  5581. goto set2;
  5582. }
  5583. }
  5584. /* If any parents have right-sibs, search.
  5585. * Otherwise, there's nothing further.
  5586. */
  5587. for (i=0; i<mc->mc_top; i++)
  5588. if (mc->mc_ki[i] <
  5589. NUMKEYS(mc->mc_pg[i])-1)
  5590. break;
  5591. if (i == mc->mc_top) {
  5592. /* There are no other pages */
  5593. mc->mc_ki[mc->mc_top] = nkeys;
  5594. return MDB_NOTFOUND;
  5595. }
  5596. }
  5597. if (!mc->mc_top) {
  5598. /* There are no other pages */
  5599. mc->mc_ki[mc->mc_top] = 0;
  5600. if (op == MDB_SET_RANGE && !exactp) {
  5601. rc = 0;
  5602. goto set1;
  5603. } else
  5604. return MDB_NOTFOUND;
  5605. }
  5606. } else {
  5607. mc->mc_pg[0] = 0;
  5608. }
  5609. rc = mdb_page_search(mc, key, 0);
  5610. if (rc != MDB_SUCCESS)
  5611. return rc;
  5612. mp = mc->mc_pg[mc->mc_top];
  5613. mdb_cassert(mc, IS_LEAF(mp));
  5614. set2:
  5615. leaf = mdb_node_search(mc, key, exactp);
  5616. if (exactp != NULL && !*exactp) {
  5617. /* MDB_SET specified and not an exact match. */
  5618. return MDB_NOTFOUND;
  5619. }
  5620. if (leaf == NULL) {
  5621. DPUTS("===> inexact leaf not found, goto sibling");
  5622. if ((rc = mdb_cursor_sibling(mc, 1)) != MDB_SUCCESS) {
  5623. mc->mc_flags |= C_EOF;
  5624. return rc; /* no entries matched */
  5625. }
  5626. mp = mc->mc_pg[mc->mc_top];
  5627. mdb_cassert(mc, IS_LEAF(mp));
  5628. leaf = NODEPTR(mp, 0);
  5629. }
  5630. set1:
  5631. mc->mc_flags |= C_INITIALIZED;
  5632. mc->mc_flags &= ~C_EOF;
  5633. if (IS_LEAF2(mp)) {
  5634. if (op == MDB_SET_RANGE || op == MDB_SET_KEY) {
  5635. key->mv_size = mc->mc_db->md_pad;
  5636. key->mv_data = LEAF2KEY(mp, mc->mc_ki[mc->mc_top], key->mv_size);
  5637. }
  5638. return MDB_SUCCESS;
  5639. }
  5640. if (F_ISSET(leaf->mn_flags, F_DUPDATA)) {
  5641. mdb_xcursor_init1(mc, leaf);
  5642. if (op == MDB_SET || op == MDB_SET_KEY || op == MDB_SET_RANGE) {
  5643. rc = mdb_cursor_first(&mc->mc_xcursor->mx_cursor, data, NULL);
  5644. } else {
  5645. int ex2, *ex2p;
  5646. if (op == MDB_GET_BOTH) {
  5647. ex2p = &ex2;
  5648. ex2 = 0;
  5649. } else {
  5650. ex2p = NULL;
  5651. }
  5652. rc = mdb_cursor_set(&mc->mc_xcursor->mx_cursor, data, NULL, MDB_SET_RANGE, ex2p);
  5653. if (rc != MDB_SUCCESS)
  5654. return rc;
  5655. }
  5656. } else if (data) {
  5657. if (op == MDB_GET_BOTH || op == MDB_GET_BOTH_RANGE) {
  5658. MDB_val olddata;
  5659. MDB_cmp_func *dcmp;
  5660. if ((rc = mdb_node_read(mc, leaf, &olddata)) != MDB_SUCCESS)
  5661. return rc;
  5662. dcmp = mc->mc_dbx->md_dcmp;
  5663. #if UINT_MAX < SIZE_MAX
  5664. if (dcmp == mdb_cmp_int && olddata.mv_size == sizeof(size_t))
  5665. dcmp = mdb_cmp_clong;
  5666. #endif
  5667. rc = dcmp(data, &olddata);
  5668. if (rc) {
  5669. if (op == MDB_GET_BOTH || rc > 0)
  5670. return MDB_NOTFOUND;
  5671. rc = 0;
  5672. }
  5673. *data = olddata;
  5674. } else {
  5675. if (mc->mc_xcursor)
  5676. mc->mc_xcursor->mx_cursor.mc_flags &= ~(C_INITIALIZED|C_EOF);
  5677. if ((rc = mdb_node_read(mc, leaf, data)) != MDB_SUCCESS)
  5678. return rc;
  5679. }
  5680. }
  5681. /* The key already matches in all other cases */
  5682. if (op == MDB_SET_RANGE || op == MDB_SET_KEY)
  5683. MDB_GET_KEY(leaf, key);
  5684. DPRINTF(("==> cursor placed on key [%s]", DKEY(key)));
  5685. return rc;
  5686. }
  5687. /** Move the cursor to the first item in the database. */
  5688. static int
  5689. mdb_cursor_first(MDB_cursor *mc, MDB_val *key, MDB_val *data)
  5690. {
  5691. int rc;
  5692. MDB_node *leaf;
  5693. if (mc->mc_xcursor)
  5694. mc->mc_xcursor->mx_cursor.mc_flags &= ~(C_INITIALIZED|C_EOF);
  5695. if (!(mc->mc_flags & C_INITIALIZED) || mc->mc_top) {
  5696. rc = mdb_page_search(mc, NULL, MDB_PS_FIRST);
  5697. if (rc != MDB_SUCCESS)
  5698. return rc;
  5699. }
  5700. mdb_cassert(mc, IS_LEAF(mc->mc_pg[mc->mc_top]));
  5701. leaf = NODEPTR(mc->mc_pg[mc->mc_top], 0);
  5702. mc->mc_flags |= C_INITIALIZED;
  5703. mc->mc_flags &= ~C_EOF;
  5704. mc->mc_ki[mc->mc_top] = 0;
  5705. if (IS_LEAF2(mc->mc_pg[mc->mc_top])) {
  5706. if ( key ) {
  5707. key->mv_size = mc->mc_db->md_pad;
  5708. key->mv_data = LEAF2KEY(mc->mc_pg[mc->mc_top], 0, key->mv_size);
  5709. }
  5710. return MDB_SUCCESS;
  5711. }
  5712. if (F_ISSET(leaf->mn_flags, F_DUPDATA)) {
  5713. mdb_xcursor_init1(mc, leaf);
  5714. rc = mdb_cursor_first(&mc->mc_xcursor->mx_cursor, data, NULL);
  5715. if (rc)
  5716. return rc;
  5717. } else if (data) {
  5718. if ((rc = mdb_node_read(mc, leaf, data)) != MDB_SUCCESS)
  5719. return rc;
  5720. }
  5721. MDB_GET_KEY(leaf, key);
  5722. return MDB_SUCCESS;
  5723. }
  5724. /** Move the cursor to the last item in the database. */
  5725. static int
  5726. mdb_cursor_last(MDB_cursor *mc, MDB_val *key, MDB_val *data)
  5727. {
  5728. int rc;
  5729. MDB_node *leaf;
  5730. if (mc->mc_xcursor)
  5731. mc->mc_xcursor->mx_cursor.mc_flags &= ~(C_INITIALIZED|C_EOF);
  5732. if (!(mc->mc_flags & C_INITIALIZED) || mc->mc_top) {
  5733. rc = mdb_page_search(mc, NULL, MDB_PS_LAST);
  5734. if (rc != MDB_SUCCESS)
  5735. return rc;
  5736. }
  5737. mdb_cassert(mc, IS_LEAF(mc->mc_pg[mc->mc_top]));
  5738. mc->mc_ki[mc->mc_top] = NUMKEYS(mc->mc_pg[mc->mc_top]) - 1;
  5739. mc->mc_flags |= C_INITIALIZED|C_EOF;
  5740. leaf = NODEPTR(mc->mc_pg[mc->mc_top], mc->mc_ki[mc->mc_top]);
  5741. if (IS_LEAF2(mc->mc_pg[mc->mc_top])) {
  5742. if (key) {
  5743. key->mv_size = mc->mc_db->md_pad;
  5744. key->mv_data = LEAF2KEY(mc->mc_pg[mc->mc_top], mc->mc_ki[mc->mc_top], key->mv_size);
  5745. }
  5746. return MDB_SUCCESS;
  5747. }
  5748. if (F_ISSET(leaf->mn_flags, F_DUPDATA)) {
  5749. mdb_xcursor_init1(mc, leaf);
  5750. rc = mdb_cursor_last(&mc->mc_xcursor->mx_cursor, data, NULL);
  5751. if (rc)
  5752. return rc;
  5753. } else if (data) {
  5754. if ((rc = mdb_node_read(mc, leaf, data)) != MDB_SUCCESS)
  5755. return rc;
  5756. }
  5757. MDB_GET_KEY(leaf, key);
  5758. return MDB_SUCCESS;
  5759. }
  5760. int
  5761. mdb_cursor_get(MDB_cursor *mc, MDB_val *key, MDB_val *data,
  5762. MDB_cursor_op op)
  5763. {
  5764. int rc;
  5765. int exact = 0;
  5766. int (*mfunc)(MDB_cursor *mc, MDB_val *key, MDB_val *data);
  5767. if (mc == NULL)
  5768. return EINVAL;
  5769. if (mc->mc_txn->mt_flags & MDB_TXN_BLOCKED)
  5770. return MDB_BAD_TXN;
  5771. switch (op) {
  5772. case MDB_GET_CURRENT:
  5773. if (!(mc->mc_flags & C_INITIALIZED)) {
  5774. rc = EINVAL;
  5775. } else {
  5776. MDB_page *mp = mc->mc_pg[mc->mc_top];
  5777. int nkeys = NUMKEYS(mp);
  5778. if (!nkeys || mc->mc_ki[mc->mc_top] >= nkeys) {
  5779. mc->mc_ki[mc->mc_top] = nkeys;
  5780. rc = MDB_NOTFOUND;
  5781. break;
  5782. }
  5783. rc = MDB_SUCCESS;
  5784. if (IS_LEAF2(mp)) {
  5785. key->mv_size = mc->mc_db->md_pad;
  5786. key->mv_data = LEAF2KEY(mp, mc->mc_ki[mc->mc_top], key->mv_size);
  5787. } else {
  5788. MDB_node *leaf = NODEPTR(mp, mc->mc_ki[mc->mc_top]);
  5789. MDB_GET_KEY(leaf, key);
  5790. if (data) {
  5791. if (F_ISSET(leaf->mn_flags, F_DUPDATA)) {
  5792. rc = mdb_cursor_get(&mc->mc_xcursor->mx_cursor, data, NULL, MDB_GET_CURRENT);
  5793. } else {
  5794. rc = mdb_node_read(mc, leaf, data);
  5795. }
  5796. }
  5797. }
  5798. }
  5799. break;
  5800. case MDB_GET_BOTH:
  5801. case MDB_GET_BOTH_RANGE:
  5802. if (data == NULL) {
  5803. rc = EINVAL;
  5804. break;
  5805. }
  5806. if (mc->mc_xcursor == NULL) {
  5807. rc = MDB_INCOMPATIBLE;
  5808. break;
  5809. }
  5810. /* FALLTHRU */
  5811. case MDB_SET:
  5812. case MDB_SET_KEY:
  5813. case MDB_SET_RANGE:
  5814. if (key == NULL) {
  5815. rc = EINVAL;
  5816. } else {
  5817. rc = mdb_cursor_set(mc, key, data, op,
  5818. op == MDB_SET_RANGE ? NULL : &exact);
  5819. }
  5820. break;
  5821. case MDB_GET_MULTIPLE:
  5822. if (data == NULL || !(mc->mc_flags & C_INITIALIZED)) {
  5823. rc = EINVAL;
  5824. break;
  5825. }
  5826. if (!(mc->mc_db->md_flags & MDB_DUPFIXED)) {
  5827. rc = MDB_INCOMPATIBLE;
  5828. break;
  5829. }
  5830. rc = MDB_SUCCESS;
  5831. if (!(mc->mc_xcursor->mx_cursor.mc_flags & C_INITIALIZED) ||
  5832. (mc->mc_xcursor->mx_cursor.mc_flags & C_EOF))
  5833. break;
  5834. goto fetchm;
  5835. case MDB_NEXT_MULTIPLE:
  5836. if (data == NULL) {
  5837. rc = EINVAL;
  5838. break;
  5839. }
  5840. if (!(mc->mc_db->md_flags & MDB_DUPFIXED)) {
  5841. rc = MDB_INCOMPATIBLE;
  5842. break;
  5843. }
  5844. rc = mdb_cursor_next(mc, key, data, MDB_NEXT_DUP);
  5845. if (rc == MDB_SUCCESS) {
  5846. if (mc->mc_xcursor->mx_cursor.mc_flags & C_INITIALIZED) {
  5847. MDB_cursor *mx;
  5848. fetchm:
  5849. mx = &mc->mc_xcursor->mx_cursor;
  5850. data->mv_size = NUMKEYS(mx->mc_pg[mx->mc_top]) *
  5851. mx->mc_db->md_pad;
  5852. data->mv_data = METADATA(mx->mc_pg[mx->mc_top]);
  5853. mx->mc_ki[mx->mc_top] = NUMKEYS(mx->mc_pg[mx->mc_top])-1;
  5854. } else {
  5855. rc = MDB_NOTFOUND;
  5856. }
  5857. }
  5858. break;
  5859. case MDB_PREV_MULTIPLE:
  5860. if (data == NULL) {
  5861. rc = EINVAL;
  5862. break;
  5863. }
  5864. if (!(mc->mc_db->md_flags & MDB_DUPFIXED)) {
  5865. rc = MDB_INCOMPATIBLE;
  5866. break;
  5867. }
  5868. if (!(mc->mc_flags & C_INITIALIZED))
  5869. rc = mdb_cursor_last(mc, key, data);
  5870. else
  5871. rc = MDB_SUCCESS;
  5872. if (rc == MDB_SUCCESS) {
  5873. MDB_cursor *mx = &mc->mc_xcursor->mx_cursor;
  5874. if (mx->mc_flags & C_INITIALIZED) {
  5875. rc = mdb_cursor_sibling(mx, 0);
  5876. if (rc == MDB_SUCCESS)
  5877. goto fetchm;
  5878. } else {
  5879. rc = MDB_NOTFOUND;
  5880. }
  5881. }
  5882. break;
  5883. case MDB_NEXT:
  5884. case MDB_NEXT_DUP:
  5885. case MDB_NEXT_NODUP:
  5886. rc = mdb_cursor_next(mc, key, data, op);
  5887. break;
  5888. case MDB_PREV:
  5889. case MDB_PREV_DUP:
  5890. case MDB_PREV_NODUP:
  5891. rc = mdb_cursor_prev(mc, key, data, op);
  5892. break;
  5893. case MDB_FIRST:
  5894. rc = mdb_cursor_first(mc, key, data);
  5895. break;
  5896. case MDB_FIRST_DUP:
  5897. mfunc = mdb_cursor_first;
  5898. mmove:
  5899. if (data == NULL || !(mc->mc_flags & C_INITIALIZED)) {
  5900. rc = EINVAL;
  5901. break;
  5902. }
  5903. if (mc->mc_xcursor == NULL) {
  5904. rc = MDB_INCOMPATIBLE;
  5905. break;
  5906. }
  5907. if (mc->mc_ki[mc->mc_top] >= NUMKEYS(mc->mc_pg[mc->mc_top])) {
  5908. mc->mc_ki[mc->mc_top] = NUMKEYS(mc->mc_pg[mc->mc_top]);
  5909. rc = MDB_NOTFOUND;
  5910. break;
  5911. }
  5912. {
  5913. MDB_node *leaf = NODEPTR(mc->mc_pg[mc->mc_top], mc->mc_ki[mc->mc_top]);
  5914. if (!F_ISSET(leaf->mn_flags, F_DUPDATA)) {
  5915. MDB_GET_KEY(leaf, key);
  5916. rc = mdb_node_read(mc, leaf, data);
  5917. break;
  5918. }
  5919. }
  5920. if (!(mc->mc_xcursor->mx_cursor.mc_flags & C_INITIALIZED)) {
  5921. rc = EINVAL;
  5922. break;
  5923. }
  5924. rc = mfunc(&mc->mc_xcursor->mx_cursor, data, NULL);
  5925. break;
  5926. case MDB_LAST:
  5927. rc = mdb_cursor_last(mc, key, data);
  5928. break;
  5929. case MDB_LAST_DUP:
  5930. mfunc = mdb_cursor_last;
  5931. goto mmove;
  5932. default:
  5933. DPRINTF(("unhandled/unimplemented cursor operation %u", op));
  5934. rc = EINVAL;
  5935. break;
  5936. }
  5937. if (mc->mc_flags & C_DEL)
  5938. mc->mc_flags ^= C_DEL;
  5939. return rc;
  5940. }
  5941. /** Touch all the pages in the cursor stack. Set mc_top.
  5942. * Makes sure all the pages are writable, before attempting a write operation.
  5943. * @param[in] mc The cursor to operate on.
  5944. */
  5945. static int
  5946. mdb_cursor_touch(MDB_cursor *mc)
  5947. {
  5948. int rc = MDB_SUCCESS;
  5949. if (mc->mc_dbi >= CORE_DBS && !(*mc->mc_dbflag & (DB_DIRTY|DB_DUPDATA))) {
  5950. /* Touch DB record of named DB */
  5951. MDB_cursor mc2;
  5952. MDB_xcursor mcx;
  5953. if (TXN_DBI_CHANGED(mc->mc_txn, mc->mc_dbi))
  5954. return MDB_BAD_DBI;
  5955. mdb_cursor_init(&mc2, mc->mc_txn, MAIN_DBI, &mcx);
  5956. rc = mdb_page_search(&mc2, &mc->mc_dbx->md_name, MDB_PS_MODIFY);
  5957. if (rc)
  5958. return rc;
  5959. *mc->mc_dbflag |= DB_DIRTY;
  5960. }
  5961. mc->mc_top = 0;
  5962. if (mc->mc_snum) {
  5963. do {
  5964. rc = mdb_page_touch(mc);
  5965. } while (!rc && ++(mc->mc_top) < mc->mc_snum);
  5966. mc->mc_top = mc->mc_snum-1;
  5967. }
  5968. return rc;
  5969. }
  5970. /** Do not spill pages to disk if txn is getting full, may fail instead */
  5971. #define MDB_NOSPILL 0x8000
  5972. int
  5973. mdb_cursor_put(MDB_cursor *mc, MDB_val *key, MDB_val *data,
  5974. unsigned int flags)
  5975. {
  5976. MDB_env *env;
  5977. MDB_node *leaf = NULL;
  5978. MDB_page *fp, *mp, *sub_root = NULL;
  5979. uint16_t fp_flags;
  5980. MDB_val xdata, *rdata, dkey, olddata;
  5981. MDB_db dummy;
  5982. int do_sub = 0, insert_key, insert_data;
  5983. unsigned int mcount = 0, dcount = 0, nospill;
  5984. size_t nsize;
  5985. int rc, rc2;
  5986. unsigned int nflags;
  5987. DKBUF;
  5988. if (mc == NULL || key == NULL)
  5989. return EINVAL;
  5990. env = mc->mc_txn->mt_env;
  5991. /* Check this first so counter will always be zero on any
  5992. * early failures.
  5993. */
  5994. if (flags & MDB_MULTIPLE) {
  5995. dcount = data[1].mv_size;
  5996. data[1].mv_size = 0;
  5997. if (!F_ISSET(mc->mc_db->md_flags, MDB_DUPFIXED))
  5998. return MDB_INCOMPATIBLE;
  5999. }
  6000. nospill = flags & MDB_NOSPILL;
  6001. flags &= ~MDB_NOSPILL;
  6002. if (mc->mc_txn->mt_flags & (MDB_TXN_RDONLY|MDB_TXN_BLOCKED))
  6003. return (mc->mc_txn->mt_flags & MDB_TXN_RDONLY) ? EACCES : MDB_BAD_TXN;
  6004. if (key->mv_size-1 >= ENV_MAXKEY(env))
  6005. return MDB_BAD_VALSIZE;
  6006. #if SIZE_MAX > MAXDATASIZE
  6007. if (data->mv_size > ((mc->mc_db->md_flags & MDB_DUPSORT) ? ENV_MAXKEY(env) : MAXDATASIZE))
  6008. return MDB_BAD_VALSIZE;
  6009. #else
  6010. if ((mc->mc_db->md_flags & MDB_DUPSORT) && data->mv_size > ENV_MAXKEY(env))
  6011. return MDB_BAD_VALSIZE;
  6012. #endif
  6013. DPRINTF(("==> put db %d key [%s], size %"Z"u, data size %"Z"u",
  6014. DDBI(mc), DKEY(key), key ? key->mv_size : 0, data->mv_size));
  6015. dkey.mv_size = 0;
  6016. if (flags & MDB_CURRENT) {
  6017. if (!(mc->mc_flags & C_INITIALIZED))
  6018. return EINVAL;
  6019. rc = MDB_SUCCESS;
  6020. } else if (mc->mc_db->md_root == P_INVALID) {
  6021. /* new database, cursor has nothing to point to */
  6022. mc->mc_snum = 0;
  6023. mc->mc_top = 0;
  6024. mc->mc_flags &= ~C_INITIALIZED;
  6025. rc = MDB_NO_ROOT;
  6026. } else {
  6027. int exact = 0;
  6028. MDB_val d2;
  6029. if (flags & MDB_APPEND) {
  6030. MDB_val k2;
  6031. rc = mdb_cursor_last(mc, &k2, &d2);
  6032. if (rc == 0) {
  6033. rc = mc->mc_dbx->md_cmp(key, &k2);
  6034. if (rc > 0) {
  6035. rc = MDB_NOTFOUND;
  6036. mc->mc_ki[mc->mc_top]++;
  6037. } else {
  6038. /* new key is <= last key */
  6039. rc = MDB_KEYEXIST;
  6040. }
  6041. }
  6042. } else {
  6043. rc = mdb_cursor_set(mc, key, &d2, MDB_SET, &exact);
  6044. }
  6045. if ((flags & MDB_NOOVERWRITE) && rc == 0) {
  6046. DPRINTF(("duplicate key [%s]", DKEY(key)));
  6047. *data = d2;
  6048. return MDB_KEYEXIST;
  6049. }
  6050. if (rc && rc != MDB_NOTFOUND)
  6051. return rc;
  6052. }
  6053. if (mc->mc_flags & C_DEL)
  6054. mc->mc_flags ^= C_DEL;
  6055. /* Cursor is positioned, check for room in the dirty list */
  6056. if (!nospill) {
  6057. if (flags & MDB_MULTIPLE) {
  6058. rdata = &xdata;
  6059. xdata.mv_size = data->mv_size * dcount;
  6060. } else {
  6061. rdata = data;
  6062. }
  6063. if ((rc2 = mdb_page_spill(mc, key, rdata)))
  6064. return rc2;
  6065. }
  6066. if (rc == MDB_NO_ROOT) {
  6067. MDB_page *np;
  6068. /* new database, write a root leaf page */
  6069. DPUTS("allocating new root leaf page");
  6070. if ((rc2 = mdb_page_new(mc, P_LEAF, 1, &np))) {
  6071. return rc2;
  6072. }
  6073. mdb_cursor_push(mc, np);
  6074. mc->mc_db->md_root = np->mp_pgno;
  6075. mc->mc_db->md_depth++;
  6076. *mc->mc_dbflag |= DB_DIRTY;
  6077. if ((mc->mc_db->md_flags & (MDB_DUPSORT|MDB_DUPFIXED))
  6078. == MDB_DUPFIXED)
  6079. np->mp_flags |= P_LEAF2;
  6080. mc->mc_flags |= C_INITIALIZED;
  6081. } else {
  6082. /* make sure all cursor pages are writable */
  6083. rc2 = mdb_cursor_touch(mc);
  6084. if (rc2)
  6085. return rc2;
  6086. }
  6087. insert_key = insert_data = rc;
  6088. if (insert_key) {
  6089. /* The key does not exist */
  6090. DPRINTF(("inserting key at index %i", mc->mc_ki[mc->mc_top]));
  6091. if ((mc->mc_db->md_flags & MDB_DUPSORT) &&
  6092. LEAFSIZE(key, data) > env->me_nodemax)
  6093. {
  6094. /* Too big for a node, insert in sub-DB. Set up an empty
  6095. * "old sub-page" for prep_subDB to expand to a full page.
  6096. */
  6097. fp_flags = P_LEAF|P_DIRTY;
  6098. fp = env->me_pbuf;
  6099. fp->mp_pad = data->mv_size; /* used if MDB_DUPFIXED */
  6100. fp->mp_lower = fp->mp_upper = (PAGEHDRSZ-PAGEBASE);
  6101. olddata.mv_size = PAGEHDRSZ;
  6102. goto prep_subDB;
  6103. }
  6104. } else {
  6105. /* there's only a key anyway, so this is a no-op */
  6106. if (IS_LEAF2(mc->mc_pg[mc->mc_top])) {
  6107. char *ptr;
  6108. unsigned int ksize = mc->mc_db->md_pad;
  6109. if (key->mv_size != ksize)
  6110. return MDB_BAD_VALSIZE;
  6111. ptr = LEAF2KEY(mc->mc_pg[mc->mc_top], mc->mc_ki[mc->mc_top], ksize);
  6112. memcpy(ptr, key->mv_data, ksize);
  6113. fix_parent:
  6114. /* if overwriting slot 0 of leaf, need to
  6115. * update branch key if there is a parent page
  6116. */
  6117. if (mc->mc_top && !mc->mc_ki[mc->mc_top]) {
  6118. unsigned short dtop = 1;
  6119. mc->mc_top--;
  6120. /* slot 0 is always an empty key, find real slot */
  6121. while (mc->mc_top && !mc->mc_ki[mc->mc_top]) {
  6122. mc->mc_top--;
  6123. dtop++;
  6124. }
  6125. if (mc->mc_ki[mc->mc_top])
  6126. rc2 = mdb_update_key(mc, key);
  6127. else
  6128. rc2 = MDB_SUCCESS;
  6129. mc->mc_top += dtop;
  6130. if (rc2)
  6131. return rc2;
  6132. }
  6133. return MDB_SUCCESS;
  6134. }
  6135. more:
  6136. leaf = NODEPTR(mc->mc_pg[mc->mc_top], mc->mc_ki[mc->mc_top]);
  6137. olddata.mv_size = NODEDSZ(leaf);
  6138. olddata.mv_data = NODEDATA(leaf);
  6139. /* DB has dups? */
  6140. if (F_ISSET(mc->mc_db->md_flags, MDB_DUPSORT)) {
  6141. /* Prepare (sub-)page/sub-DB to accept the new item,
  6142. * if needed. fp: old sub-page or a header faking
  6143. * it. mp: new (sub-)page. offset: growth in page
  6144. * size. xdata: node data with new page or DB.
  6145. */
  6146. unsigned i, offset = 0;
  6147. mp = fp = xdata.mv_data = env->me_pbuf;
  6148. mp->mp_pgno = mc->mc_pg[mc->mc_top]->mp_pgno;
  6149. /* Was a single item before, must convert now */
  6150. if (!F_ISSET(leaf->mn_flags, F_DUPDATA)) {
  6151. MDB_cmp_func *dcmp;
  6152. /* Just overwrite the current item */
  6153. if (flags == MDB_CURRENT)
  6154. goto current;
  6155. dcmp = mc->mc_dbx->md_dcmp;
  6156. #if UINT_MAX < SIZE_MAX
  6157. if (dcmp == mdb_cmp_int && olddata.mv_size == sizeof(size_t))
  6158. dcmp = mdb_cmp_clong;
  6159. #endif
  6160. /* does data match? */
  6161. if (!dcmp(data, &olddata)) {
  6162. if (flags & (MDB_NODUPDATA|MDB_APPENDDUP))
  6163. return MDB_KEYEXIST;
  6164. /* overwrite it */
  6165. goto current;
  6166. }
  6167. /* Back up original data item */
  6168. dkey.mv_size = olddata.mv_size;
  6169. dkey.mv_data = memcpy(fp+1, olddata.mv_data, olddata.mv_size);
  6170. /* Make sub-page header for the dup items, with dummy body */
  6171. fp->mp_flags = P_LEAF|P_DIRTY|P_SUBP;
  6172. fp->mp_lower = (PAGEHDRSZ-PAGEBASE);
  6173. xdata.mv_size = PAGEHDRSZ + dkey.mv_size + data->mv_size;
  6174. if (mc->mc_db->md_flags & MDB_DUPFIXED) {
  6175. fp->mp_flags |= P_LEAF2;
  6176. fp->mp_pad = data->mv_size;
  6177. xdata.mv_size += 2 * data->mv_size; /* leave space for 2 more */
  6178. } else {
  6179. xdata.mv_size += 2 * (sizeof(indx_t) + NODESIZE) +
  6180. (dkey.mv_size & 1) + (data->mv_size & 1);
  6181. }
  6182. fp->mp_upper = xdata.mv_size - PAGEBASE;
  6183. olddata.mv_size = xdata.mv_size; /* pretend olddata is fp */
  6184. } else if (leaf->mn_flags & F_SUBDATA) {
  6185. /* Data is on sub-DB, just store it */
  6186. flags |= F_DUPDATA|F_SUBDATA;
  6187. goto put_sub;
  6188. } else {
  6189. /* Data is on sub-page */
  6190. fp = olddata.mv_data;
  6191. switch (flags) {
  6192. default:
  6193. if (!(mc->mc_db->md_flags & MDB_DUPFIXED)) {
  6194. offset = EVEN(NODESIZE + sizeof(indx_t) +
  6195. data->mv_size);
  6196. break;
  6197. }
  6198. offset = fp->mp_pad;
  6199. if (SIZELEFT(fp) < offset) {
  6200. offset *= 4; /* space for 4 more */
  6201. break;
  6202. }
  6203. /* FALLTHRU */ /* Big enough MDB_DUPFIXED sub-page */
  6204. case MDB_CURRENT:
  6205. fp->mp_flags |= P_DIRTY;
  6206. COPY_PGNO(fp->mp_pgno, mp->mp_pgno);
  6207. mc->mc_xcursor->mx_cursor.mc_pg[0] = fp;
  6208. flags |= F_DUPDATA;
  6209. goto put_sub;
  6210. }
  6211. xdata.mv_size = olddata.mv_size + offset;
  6212. }
  6213. fp_flags = fp->mp_flags;
  6214. if (NODESIZE + NODEKSZ(leaf) + xdata.mv_size > env->me_nodemax) {
  6215. /* Too big for a sub-page, convert to sub-DB */
  6216. fp_flags &= ~P_SUBP;
  6217. prep_subDB:
  6218. if (mc->mc_db->md_flags & MDB_DUPFIXED) {
  6219. fp_flags |= P_LEAF2;
  6220. dummy.md_pad = fp->mp_pad;
  6221. dummy.md_flags = MDB_DUPFIXED;
  6222. if (mc->mc_db->md_flags & MDB_INTEGERDUP)
  6223. dummy.md_flags |= MDB_INTEGERKEY;
  6224. } else {
  6225. dummy.md_pad = 0;
  6226. dummy.md_flags = 0;
  6227. }
  6228. dummy.md_depth = 1;
  6229. dummy.md_branch_pages = 0;
  6230. dummy.md_leaf_pages = 1;
  6231. dummy.md_overflow_pages = 0;
  6232. dummy.md_entries = NUMKEYS(fp);
  6233. xdata.mv_size = sizeof(MDB_db);
  6234. xdata.mv_data = &dummy;
  6235. if ((rc = mdb_page_alloc(mc, 1, &mp)))
  6236. return rc;
  6237. offset = env->me_psize - olddata.mv_size;
  6238. flags |= F_DUPDATA|F_SUBDATA;
  6239. dummy.md_root = mp->mp_pgno;
  6240. sub_root = mp;
  6241. }
  6242. if (mp != fp) {
  6243. mp->mp_flags = fp_flags | P_DIRTY;
  6244. mp->mp_pad = fp->mp_pad;
  6245. mp->mp_lower = fp->mp_lower;
  6246. mp->mp_upper = fp->mp_upper + offset;
  6247. if (fp_flags & P_LEAF2) {
  6248. memcpy(METADATA(mp), METADATA(fp), NUMKEYS(fp) * fp->mp_pad);
  6249. } else {
  6250. memcpy((char *)mp + mp->mp_upper + PAGEBASE, (char *)fp + fp->mp_upper + PAGEBASE,
  6251. olddata.mv_size - fp->mp_upper - PAGEBASE);
  6252. memcpy((char *)(&mp->mp_ptrs), (char *)(&fp->mp_ptrs), NUMKEYS(fp) * sizeof(mp->mp_ptrs[0]));
  6253. for (i=0; i<NUMKEYS(fp); i++)
  6254. mp->mp_ptrs[i] += offset;
  6255. }
  6256. }
  6257. rdata = &xdata;
  6258. flags |= F_DUPDATA;
  6259. do_sub = 1;
  6260. if (!insert_key)
  6261. mdb_node_del(mc, 0);
  6262. goto new_sub;
  6263. }
  6264. current:
  6265. /* LMDB passes F_SUBDATA in 'flags' to write a DB record */
  6266. if ((leaf->mn_flags ^ flags) & F_SUBDATA)
  6267. return MDB_INCOMPATIBLE;
  6268. /* overflow page overwrites need special handling */
  6269. if (F_ISSET(leaf->mn_flags, F_BIGDATA)) {
  6270. MDB_page *omp;
  6271. pgno_t pg;
  6272. int level, ovpages, dpages = OVPAGES(data->mv_size, env->me_psize);
  6273. memcpy(&pg, olddata.mv_data, sizeof(pg));
  6274. if ((rc2 = mdb_page_get(mc, pg, &omp, &level)) != 0)
  6275. return rc2;
  6276. ovpages = omp->mp_pages;
  6277. /* Is the ov page large enough? */
  6278. if (ovpages >= dpages) {
  6279. if (!(omp->mp_flags & P_DIRTY) &&
  6280. (level || (env->me_flags & MDB_WRITEMAP)))
  6281. {
  6282. rc = mdb_page_unspill(mc->mc_txn, omp, &omp);
  6283. if (rc)
  6284. return rc;
  6285. level = 0; /* dirty in this txn or clean */
  6286. }
  6287. /* Is it dirty? */
  6288. if (omp->mp_flags & P_DIRTY) {
  6289. /* yes, overwrite it. Note in this case we don't
  6290. * bother to try shrinking the page if the new data
  6291. * is smaller than the overflow threshold.
  6292. */
  6293. if (level > 1) {
  6294. /* It is writable only in a parent txn */
  6295. size_t sz = (size_t) env->me_psize * ovpages, off;
  6296. MDB_page *np = mdb_page_malloc(mc->mc_txn, ovpages);
  6297. MDB_ID2 id2;
  6298. if (!np)
  6299. return ENOMEM;
  6300. id2.mid = pg;
  6301. id2.mptr = np;
  6302. /* Note - this page is already counted in parent's dirty_room */
  6303. rc2 = mdb_mid2l_insert(mc->mc_txn->mt_u.dirty_list, &id2);
  6304. mdb_cassert(mc, rc2 == 0);
  6305. /* Currently we make the page look as with put() in the
  6306. * parent txn, in case the user peeks at MDB_RESERVEd
  6307. * or unused parts. Some users treat ovpages specially.
  6308. */
  6309. if (!(flags & MDB_RESERVE)) {
  6310. /* Skip the part where LMDB will put *data.
  6311. * Copy end of page, adjusting alignment so
  6312. * compiler may copy words instead of bytes.
  6313. */
  6314. off = (PAGEHDRSZ + data->mv_size) & -sizeof(size_t);
  6315. memcpy((size_t *)((char *)np + off),
  6316. (size_t *)((char *)omp + off), sz - off);
  6317. sz = PAGEHDRSZ;
  6318. }
  6319. memcpy(np, omp, sz); /* Copy beginning of page */
  6320. omp = np;
  6321. }
  6322. SETDSZ(leaf, data->mv_size);
  6323. if (F_ISSET(flags, MDB_RESERVE))
  6324. data->mv_data = METADATA(omp);
  6325. else
  6326. memcpy(METADATA(omp), data->mv_data, data->mv_size);
  6327. return MDB_SUCCESS;
  6328. }
  6329. }
  6330. if ((rc2 = mdb_ovpage_free(mc, omp)) != MDB_SUCCESS)
  6331. return rc2;
  6332. } else if (data->mv_size == olddata.mv_size) {
  6333. /* same size, just replace it. Note that we could
  6334. * also reuse this node if the new data is smaller,
  6335. * but instead we opt to shrink the node in that case.
  6336. */
  6337. if (F_ISSET(flags, MDB_RESERVE))
  6338. data->mv_data = olddata.mv_data;
  6339. else if (!(mc->mc_flags & C_SUB))
  6340. memcpy(olddata.mv_data, data->mv_data, data->mv_size);
  6341. else {
  6342. memcpy(NODEKEY(leaf), key->mv_data, key->mv_size);
  6343. goto fix_parent;
  6344. }
  6345. return MDB_SUCCESS;
  6346. }
  6347. mdb_node_del(mc, 0);
  6348. }
  6349. rdata = data;
  6350. new_sub:
  6351. nflags = flags & NODE_ADD_FLAGS;
  6352. nsize = IS_LEAF2(mc->mc_pg[mc->mc_top]) ? key->mv_size : mdb_leaf_size(env, key, rdata);
  6353. if (SIZELEFT(mc->mc_pg[mc->mc_top]) < nsize) {
  6354. if (( flags & (F_DUPDATA|F_SUBDATA)) == F_DUPDATA )
  6355. nflags &= ~MDB_APPEND; /* sub-page may need room to grow */
  6356. if (!insert_key)
  6357. nflags |= MDB_SPLIT_REPLACE;
  6358. rc = mdb_page_split(mc, key, rdata, P_INVALID, nflags);
  6359. } else {
  6360. /* There is room already in this leaf page. */
  6361. rc = mdb_node_add(mc, mc->mc_ki[mc->mc_top], key, rdata, 0, nflags);
  6362. if (rc == 0) {
  6363. /* Adjust other cursors pointing to mp */
  6364. MDB_cursor *m2, *m3;
  6365. MDB_dbi dbi = mc->mc_dbi;
  6366. unsigned i = mc->mc_top;
  6367. MDB_page *mp = mc->mc_pg[i];
  6368. for (m2 = mc->mc_txn->mt_cursors[dbi]; m2; m2=m2->mc_next) {
  6369. if (mc->mc_flags & C_SUB)
  6370. m3 = &m2->mc_xcursor->mx_cursor;
  6371. else
  6372. m3 = m2;
  6373. if (m3 == mc || m3->mc_snum < mc->mc_snum || m3->mc_pg[i] != mp) continue;
  6374. if (m3->mc_ki[i] >= mc->mc_ki[i] && insert_key) {
  6375. m3->mc_ki[i]++;
  6376. }
  6377. XCURSOR_REFRESH(m3, i, mp);
  6378. }
  6379. }
  6380. }
  6381. if (rc == MDB_SUCCESS) {
  6382. /* Now store the actual data in the child DB. Note that we're
  6383. * storing the user data in the keys field, so there are strict
  6384. * size limits on dupdata. The actual data fields of the child
  6385. * DB are all zero size.
  6386. */
  6387. if (do_sub) {
  6388. int xflags, new_dupdata;
  6389. size_t ecount;
  6390. put_sub:
  6391. xdata.mv_size = 0;
  6392. xdata.mv_data = "";
  6393. leaf = NODEPTR(mc->mc_pg[mc->mc_top], mc->mc_ki[mc->mc_top]);
  6394. if ((flags & (MDB_CURRENT|MDB_APPENDDUP)) == MDB_CURRENT) {
  6395. xflags = MDB_CURRENT|MDB_NOSPILL;
  6396. } else {
  6397. mdb_xcursor_init1(mc, leaf);
  6398. xflags = (flags & MDB_NODUPDATA) ?
  6399. MDB_NOOVERWRITE|MDB_NOSPILL : MDB_NOSPILL;
  6400. }
  6401. if (sub_root)
  6402. mc->mc_xcursor->mx_cursor.mc_pg[0] = sub_root;
  6403. new_dupdata = (int)dkey.mv_size;
  6404. /* converted, write the original data first */
  6405. if (dkey.mv_size) {
  6406. rc = mdb_cursor_put(&mc->mc_xcursor->mx_cursor, &dkey, &xdata, xflags);
  6407. if (rc)
  6408. goto bad_sub;
  6409. /* we've done our job */
  6410. dkey.mv_size = 0;
  6411. }
  6412. if (!(leaf->mn_flags & F_SUBDATA) || sub_root) {
  6413. /* Adjust other cursors pointing to mp */
  6414. MDB_cursor *m2;
  6415. MDB_xcursor *mx = mc->mc_xcursor;
  6416. unsigned i = mc->mc_top;
  6417. MDB_page *mp = mc->mc_pg[i];
  6418. for (m2 = mc->mc_txn->mt_cursors[mc->mc_dbi]; m2; m2=m2->mc_next) {
  6419. if (m2 == mc || m2->mc_snum < mc->mc_snum) continue;
  6420. if (!(m2->mc_flags & C_INITIALIZED)) continue;
  6421. if (m2->mc_pg[i] == mp) {
  6422. if (m2->mc_ki[i] == mc->mc_ki[i]) {
  6423. mdb_xcursor_init2(m2, mx, new_dupdata);
  6424. } else if (!insert_key) {
  6425. XCURSOR_REFRESH(m2, i, mp);
  6426. }
  6427. }
  6428. }
  6429. }
  6430. ecount = mc->mc_xcursor->mx_db.md_entries;
  6431. if (flags & MDB_APPENDDUP)
  6432. xflags |= MDB_APPEND;
  6433. rc = mdb_cursor_put(&mc->mc_xcursor->mx_cursor, data, &xdata, xflags);
  6434. if (flags & F_SUBDATA) {
  6435. void *db = NODEDATA(leaf);
  6436. memcpy(db, &mc->mc_xcursor->mx_db, sizeof(MDB_db));
  6437. }
  6438. insert_data = mc->mc_xcursor->mx_db.md_entries - ecount;
  6439. }
  6440. /* Increment count unless we just replaced an existing item. */
  6441. if (insert_data)
  6442. mc->mc_db->md_entries++;
  6443. if (insert_key) {
  6444. /* Invalidate txn if we created an empty sub-DB */
  6445. if (rc)
  6446. goto bad_sub;
  6447. /* If we succeeded and the key didn't exist before,
  6448. * make sure the cursor is marked valid.
  6449. */
  6450. mc->mc_flags |= C_INITIALIZED;
  6451. }
  6452. if (flags & MDB_MULTIPLE) {
  6453. if (!rc) {
  6454. mcount++;
  6455. /* let caller know how many succeeded, if any */
  6456. data[1].mv_size = mcount;
  6457. if (mcount < dcount) {
  6458. data[0].mv_data = (char *)data[0].mv_data + data[0].mv_size;
  6459. insert_key = insert_data = 0;
  6460. goto more;
  6461. }
  6462. }
  6463. }
  6464. return rc;
  6465. bad_sub:
  6466. if (rc == MDB_KEYEXIST) /* should not happen, we deleted that item */
  6467. rc = MDB_CORRUPTED;
  6468. }
  6469. mc->mc_txn->mt_flags |= MDB_TXN_ERROR;
  6470. return rc;
  6471. }
  6472. int
  6473. mdb_cursor_del(MDB_cursor *mc, unsigned int flags)
  6474. {
  6475. MDB_node *leaf;
  6476. MDB_page *mp;
  6477. int rc;
  6478. if (mc->mc_txn->mt_flags & (MDB_TXN_RDONLY|MDB_TXN_BLOCKED))
  6479. return (mc->mc_txn->mt_flags & MDB_TXN_RDONLY) ? EACCES : MDB_BAD_TXN;
  6480. if (!(mc->mc_flags & C_INITIALIZED))
  6481. return EINVAL;
  6482. if (mc->mc_ki[mc->mc_top] >= NUMKEYS(mc->mc_pg[mc->mc_top]))
  6483. return MDB_NOTFOUND;
  6484. if (!(flags & MDB_NOSPILL) && (rc = mdb_page_spill(mc, NULL, NULL)))
  6485. return rc;
  6486. rc = mdb_cursor_touch(mc);
  6487. if (rc)
  6488. return rc;
  6489. mp = mc->mc_pg[mc->mc_top];
  6490. if (!IS_LEAF(mp))
  6491. return MDB_CORRUPTED;
  6492. if (IS_LEAF2(mp))
  6493. goto del_key;
  6494. leaf = NODEPTR(mp, mc->mc_ki[mc->mc_top]);
  6495. if (F_ISSET(leaf->mn_flags, F_DUPDATA)) {
  6496. if (flags & MDB_NODUPDATA) {
  6497. /* mdb_cursor_del0() will subtract the final entry */
  6498. mc->mc_db->md_entries -= mc->mc_xcursor->mx_db.md_entries - 1;
  6499. mc->mc_xcursor->mx_cursor.mc_flags &= ~C_INITIALIZED;
  6500. } else {
  6501. if (!F_ISSET(leaf->mn_flags, F_SUBDATA)) {
  6502. mc->mc_xcursor->mx_cursor.mc_pg[0] = NODEDATA(leaf);
  6503. }
  6504. rc = mdb_cursor_del(&mc->mc_xcursor->mx_cursor, MDB_NOSPILL);
  6505. if (rc)
  6506. return rc;
  6507. /* If sub-DB still has entries, we're done */
  6508. if (mc->mc_xcursor->mx_db.md_entries) {
  6509. if (leaf->mn_flags & F_SUBDATA) {
  6510. /* update subDB info */
  6511. void *db = NODEDATA(leaf);
  6512. memcpy(db, &mc->mc_xcursor->mx_db, sizeof(MDB_db));
  6513. } else {
  6514. MDB_cursor *m2;
  6515. /* shrink fake page */
  6516. mdb_node_shrink(mp, mc->mc_ki[mc->mc_top]);
  6517. leaf = NODEPTR(mp, mc->mc_ki[mc->mc_top]);
  6518. mc->mc_xcursor->mx_cursor.mc_pg[0] = NODEDATA(leaf);
  6519. /* fix other sub-DB cursors pointed at fake pages on this page */
  6520. for (m2 = mc->mc_txn->mt_cursors[mc->mc_dbi]; m2; m2=m2->mc_next) {
  6521. if (m2 == mc || m2->mc_snum < mc->mc_snum) continue;
  6522. if (!(m2->mc_flags & C_INITIALIZED)) continue;
  6523. if (m2->mc_pg[mc->mc_top] == mp) {
  6524. XCURSOR_REFRESH(m2, mc->mc_top, mp);
  6525. }
  6526. }
  6527. }
  6528. mc->mc_db->md_entries--;
  6529. return rc;
  6530. } else {
  6531. mc->mc_xcursor->mx_cursor.mc_flags &= ~C_INITIALIZED;
  6532. }
  6533. /* otherwise fall thru and delete the sub-DB */
  6534. }
  6535. if (leaf->mn_flags & F_SUBDATA) {
  6536. /* add all the child DB's pages to the free list */
  6537. rc = mdb_drop0(&mc->mc_xcursor->mx_cursor, 0);
  6538. if (rc)
  6539. goto fail;
  6540. }
  6541. }
  6542. /* LMDB passes F_SUBDATA in 'flags' to delete a DB record */
  6543. else if ((leaf->mn_flags ^ flags) & F_SUBDATA) {
  6544. rc = MDB_INCOMPATIBLE;
  6545. goto fail;
  6546. }
  6547. /* add overflow pages to free list */
  6548. if (F_ISSET(leaf->mn_flags, F_BIGDATA)) {
  6549. MDB_page *omp;
  6550. pgno_t pg;
  6551. memcpy(&pg, NODEDATA(leaf), sizeof(pg));
  6552. if ((rc = mdb_page_get(mc, pg, &omp, NULL)) ||
  6553. (rc = mdb_ovpage_free(mc, omp)))
  6554. goto fail;
  6555. }
  6556. del_key:
  6557. return mdb_cursor_del0(mc);
  6558. fail:
  6559. mc->mc_txn->mt_flags |= MDB_TXN_ERROR;
  6560. return rc;
  6561. }
  6562. /** Allocate and initialize new pages for a database.
  6563. * Set #MDB_TXN_ERROR on failure.
  6564. * @param[in] mc a cursor on the database being added to.
  6565. * @param[in] flags flags defining what type of page is being allocated.
  6566. * @param[in] num the number of pages to allocate. This is usually 1,
  6567. * unless allocating overflow pages for a large record.
  6568. * @param[out] mp Address of a page, or NULL on failure.
  6569. * @return 0 on success, non-zero on failure.
  6570. */
  6571. static int
  6572. mdb_page_new(MDB_cursor *mc, uint32_t flags, int num, MDB_page **mp)
  6573. {
  6574. MDB_page *np;
  6575. int rc;
  6576. if ((rc = mdb_page_alloc(mc, num, &np)))
  6577. return rc;
  6578. DPRINTF(("allocated new mpage %"Z"u, page size %u",
  6579. np->mp_pgno, mc->mc_txn->mt_env->me_psize));
  6580. np->mp_flags = flags | P_DIRTY;
  6581. np->mp_lower = (PAGEHDRSZ-PAGEBASE);
  6582. np->mp_upper = mc->mc_txn->mt_env->me_psize - PAGEBASE;
  6583. if (IS_BRANCH(np))
  6584. mc->mc_db->md_branch_pages++;
  6585. else if (IS_LEAF(np))
  6586. mc->mc_db->md_leaf_pages++;
  6587. else if (IS_OVERFLOW(np)) {
  6588. mc->mc_db->md_overflow_pages += num;
  6589. np->mp_pages = num;
  6590. }
  6591. *mp = np;
  6592. return 0;
  6593. }
  6594. /** Calculate the size of a leaf node.
  6595. * The size depends on the environment's page size; if a data item
  6596. * is too large it will be put onto an overflow page and the node
  6597. * size will only include the key and not the data. Sizes are always
  6598. * rounded up to an even number of bytes, to guarantee 2-byte alignment
  6599. * of the #MDB_node headers.
  6600. * @param[in] env The environment handle.
  6601. * @param[in] key The key for the node.
  6602. * @param[in] data The data for the node.
  6603. * @return The number of bytes needed to store the node.
  6604. */
  6605. static size_t
  6606. mdb_leaf_size(MDB_env *env, MDB_val *key, MDB_val *data)
  6607. {
  6608. size_t sz;
  6609. sz = LEAFSIZE(key, data);
  6610. if (sz > env->me_nodemax) {
  6611. /* put on overflow page */
  6612. sz -= data->mv_size - sizeof(pgno_t);
  6613. }
  6614. return EVEN(sz + sizeof(indx_t));
  6615. }
  6616. /** Calculate the size of a branch node.
  6617. * The size should depend on the environment's page size but since
  6618. * we currently don't support spilling large keys onto overflow
  6619. * pages, it's simply the size of the #MDB_node header plus the
  6620. * size of the key. Sizes are always rounded up to an even number
  6621. * of bytes, to guarantee 2-byte alignment of the #MDB_node headers.
  6622. * @param[in] env The environment handle.
  6623. * @param[in] key The key for the node.
  6624. * @return The number of bytes needed to store the node.
  6625. */
  6626. static size_t
  6627. mdb_branch_size(MDB_env *env, MDB_val *key)
  6628. {
  6629. size_t sz;
  6630. sz = INDXSIZE(key);
  6631. if (sz > env->me_nodemax) {
  6632. /* put on overflow page */
  6633. /* not implemented */
  6634. /* sz -= key->size - sizeof(pgno_t); */
  6635. }
  6636. return sz + sizeof(indx_t);
  6637. }
  6638. /** Add a node to the page pointed to by the cursor.
  6639. * Set #MDB_TXN_ERROR on failure.
  6640. * @param[in] mc The cursor for this operation.
  6641. * @param[in] indx The index on the page where the new node should be added.
  6642. * @param[in] key The key for the new node.
  6643. * @param[in] data The data for the new node, if any.
  6644. * @param[in] pgno The page number, if adding a branch node.
  6645. * @param[in] flags Flags for the node.
  6646. * @return 0 on success, non-zero on failure. Possible errors are:
  6647. * <ul>
  6648. * <li>ENOMEM - failed to allocate overflow pages for the node.
  6649. * <li>MDB_PAGE_FULL - there is insufficient room in the page. This error
  6650. * should never happen since all callers already calculate the
  6651. * page's free space before calling this function.
  6652. * </ul>
  6653. */
  6654. static int
  6655. mdb_node_add(MDB_cursor *mc, indx_t indx,
  6656. MDB_val *key, MDB_val *data, pgno_t pgno, unsigned int flags)
  6657. {
  6658. unsigned int i;
  6659. size_t node_size = NODESIZE;
  6660. ssize_t room;
  6661. indx_t ofs;
  6662. MDB_node *node;
  6663. MDB_page *mp = mc->mc_pg[mc->mc_top];
  6664. MDB_page *ofp = NULL; /* overflow page */
  6665. void *ndata;
  6666. DKBUF;
  6667. mdb_cassert(mc, mp->mp_upper >= mp->mp_lower);
  6668. DPRINTF(("add to %s %spage %"Z"u index %i, data size %"Z"u key size %"Z"u [%s]",
  6669. IS_LEAF(mp) ? "leaf" : "branch",
  6670. IS_SUBP(mp) ? "sub-" : "",
  6671. mdb_dbg_pgno(mp), indx, data ? data->mv_size : 0,
  6672. key ? key->mv_size : 0, key ? DKEY(key) : "null"));
  6673. if (IS_LEAF2(mp)) {
  6674. /* Move higher keys up one slot. */
  6675. int ksize = mc->mc_db->md_pad, dif;
  6676. char *ptr = LEAF2KEY(mp, indx, ksize);
  6677. dif = NUMKEYS(mp) - indx;
  6678. if (dif > 0)
  6679. memmove(ptr+ksize, ptr, dif*ksize);
  6680. /* insert new key */
  6681. memcpy(ptr, key->mv_data, ksize);
  6682. /* Just using these for counting */
  6683. mp->mp_lower += sizeof(indx_t);
  6684. mp->mp_upper -= ksize - sizeof(indx_t);
  6685. return MDB_SUCCESS;
  6686. }
  6687. room = (ssize_t)SIZELEFT(mp) - (ssize_t)sizeof(indx_t);
  6688. if (key != NULL)
  6689. node_size += key->mv_size;
  6690. if (IS_LEAF(mp)) {
  6691. mdb_cassert(mc, key && data);
  6692. if (F_ISSET(flags, F_BIGDATA)) {
  6693. /* Data already on overflow page. */
  6694. node_size += sizeof(pgno_t);
  6695. } else if (node_size + data->mv_size > mc->mc_txn->mt_env->me_nodemax) {
  6696. int ovpages = OVPAGES(data->mv_size, mc->mc_txn->mt_env->me_psize);
  6697. int rc;
  6698. /* Put data on overflow page. */
  6699. DPRINTF(("data size is %"Z"u, node would be %"Z"u, put data on overflow page",
  6700. data->mv_size, node_size+data->mv_size));
  6701. node_size = EVEN(node_size + sizeof(pgno_t));
  6702. if ((ssize_t)node_size > room)
  6703. goto full;
  6704. if ((rc = mdb_page_new(mc, P_OVERFLOW, ovpages, &ofp)))
  6705. return rc;
  6706. DPRINTF(("allocated overflow page %"Z"u", ofp->mp_pgno));
  6707. flags |= F_BIGDATA;
  6708. goto update;
  6709. } else {
  6710. node_size += data->mv_size;
  6711. }
  6712. }
  6713. node_size = EVEN(node_size);
  6714. if ((ssize_t)node_size > room)
  6715. goto full;
  6716. update:
  6717. /* Move higher pointers up one slot. */
  6718. for (i = NUMKEYS(mp); i > indx; i--)
  6719. mp->mp_ptrs[i] = mp->mp_ptrs[i - 1];
  6720. /* Adjust free space offsets. */
  6721. ofs = mp->mp_upper - node_size;
  6722. mdb_cassert(mc, ofs >= mp->mp_lower + sizeof(indx_t));
  6723. mp->mp_ptrs[indx] = ofs;
  6724. mp->mp_upper = ofs;
  6725. mp->mp_lower += sizeof(indx_t);
  6726. /* Write the node data. */
  6727. node = NODEPTR(mp, indx);
  6728. node->mn_ksize = (key == NULL) ? 0 : key->mv_size;
  6729. node->mn_flags = flags;
  6730. if (IS_LEAF(mp))
  6731. SETDSZ(node,data->mv_size);
  6732. else
  6733. SETPGNO(node,pgno);
  6734. if (key)
  6735. memcpy(NODEKEY(node), key->mv_data, key->mv_size);
  6736. if (IS_LEAF(mp)) {
  6737. ndata = NODEDATA(node);
  6738. if (ofp == NULL) {
  6739. if (F_ISSET(flags, F_BIGDATA))
  6740. memcpy(ndata, data->mv_data, sizeof(pgno_t));
  6741. else if (F_ISSET(flags, MDB_RESERVE))
  6742. data->mv_data = ndata;
  6743. else
  6744. memcpy(ndata, data->mv_data, data->mv_size);
  6745. } else {
  6746. memcpy(ndata, &ofp->mp_pgno, sizeof(pgno_t));
  6747. ndata = METADATA(ofp);
  6748. if (F_ISSET(flags, MDB_RESERVE))
  6749. data->mv_data = ndata;
  6750. else
  6751. memcpy(ndata, data->mv_data, data->mv_size);
  6752. }
  6753. }
  6754. return MDB_SUCCESS;
  6755. full:
  6756. DPRINTF(("not enough room in page %"Z"u, got %u ptrs",
  6757. mdb_dbg_pgno(mp), NUMKEYS(mp)));
  6758. DPRINTF(("upper-lower = %u - %u = %"Z"d", mp->mp_upper,mp->mp_lower,room));
  6759. DPRINTF(("node size = %"Z"u", node_size));
  6760. mc->mc_txn->mt_flags |= MDB_TXN_ERROR;
  6761. return MDB_PAGE_FULL;
  6762. }
  6763. /** Delete the specified node from a page.
  6764. * @param[in] mc Cursor pointing to the node to delete.
  6765. * @param[in] ksize The size of a node. Only used if the page is
  6766. * part of a #MDB_DUPFIXED database.
  6767. */
  6768. static void
  6769. mdb_node_del(MDB_cursor *mc, int ksize)
  6770. {
  6771. MDB_page *mp = mc->mc_pg[mc->mc_top];
  6772. indx_t indx = mc->mc_ki[mc->mc_top];
  6773. unsigned int sz;
  6774. indx_t i, j, numkeys, ptr;
  6775. MDB_node *node;
  6776. char *base;
  6777. DPRINTF(("delete node %u on %s page %"Z"u", indx,
  6778. IS_LEAF(mp) ? "leaf" : "branch", mdb_dbg_pgno(mp)));
  6779. numkeys = NUMKEYS(mp);
  6780. mdb_cassert(mc, indx < numkeys);
  6781. if (IS_LEAF2(mp)) {
  6782. int x = numkeys - 1 - indx;
  6783. base = LEAF2KEY(mp, indx, ksize);
  6784. if (x)
  6785. memmove(base, base + ksize, x * ksize);
  6786. mp->mp_lower -= sizeof(indx_t);
  6787. mp->mp_upper += ksize - sizeof(indx_t);
  6788. return;
  6789. }
  6790. node = NODEPTR(mp, indx);
  6791. sz = NODESIZE + node->mn_ksize;
  6792. if (IS_LEAF(mp)) {
  6793. if (F_ISSET(node->mn_flags, F_BIGDATA))
  6794. sz += sizeof(pgno_t);
  6795. else
  6796. sz += NODEDSZ(node);
  6797. }
  6798. sz = EVEN(sz);
  6799. ptr = mp->mp_ptrs[indx];
  6800. for (i = j = 0; i < numkeys; i++) {
  6801. if (i != indx) {
  6802. mp->mp_ptrs[j] = mp->mp_ptrs[i];
  6803. if (mp->mp_ptrs[i] < ptr)
  6804. mp->mp_ptrs[j] += sz;
  6805. j++;
  6806. }
  6807. }
  6808. base = (char *)mp + mp->mp_upper + PAGEBASE;
  6809. memmove(base + sz, base, ptr - mp->mp_upper);
  6810. mp->mp_lower -= sizeof(indx_t);
  6811. mp->mp_upper += sz;
  6812. }
  6813. /** Compact the main page after deleting a node on a subpage.
  6814. * @param[in] mp The main page to operate on.
  6815. * @param[in] indx The index of the subpage on the main page.
  6816. */
  6817. static void
  6818. mdb_node_shrink(MDB_page *mp, indx_t indx)
  6819. {
  6820. MDB_node *node;
  6821. MDB_page *sp, *xp;
  6822. char *base;
  6823. indx_t delta, nsize, len, ptr;
  6824. int i;
  6825. node = NODEPTR(mp, indx);
  6826. sp = (MDB_page *)NODEDATA(node);
  6827. delta = SIZELEFT(sp);
  6828. nsize = NODEDSZ(node) - delta;
  6829. /* Prepare to shift upward, set len = length(subpage part to shift) */
  6830. if (IS_LEAF2(sp)) {
  6831. len = nsize;
  6832. if (nsize & 1)
  6833. return; /* do not make the node uneven-sized */
  6834. } else {
  6835. xp = (MDB_page *)((char *)sp + delta); /* destination subpage */
  6836. for (i = NUMKEYS(sp); --i >= 0; )
  6837. xp->mp_ptrs[i] = sp->mp_ptrs[i] - delta;
  6838. len = PAGEHDRSZ;
  6839. }
  6840. sp->mp_upper = sp->mp_lower;
  6841. COPY_PGNO(sp->mp_pgno, mp->mp_pgno);
  6842. SETDSZ(node, nsize);
  6843. /* Shift <lower nodes...initial part of subpage> upward */
  6844. base = (char *)mp + mp->mp_upper + PAGEBASE;
  6845. memmove(base + delta, base, (char *)sp + len - base);
  6846. ptr = mp->mp_ptrs[indx];
  6847. for (i = NUMKEYS(mp); --i >= 0; ) {
  6848. if (mp->mp_ptrs[i] <= ptr)
  6849. mp->mp_ptrs[i] += delta;
  6850. }
  6851. mp->mp_upper += delta;
  6852. }
  6853. /** Initial setup of a sorted-dups cursor.
  6854. * Sorted duplicates are implemented as a sub-database for the given key.
  6855. * The duplicate data items are actually keys of the sub-database.
  6856. * Operations on the duplicate data items are performed using a sub-cursor
  6857. * initialized when the sub-database is first accessed. This function does
  6858. * the preliminary setup of the sub-cursor, filling in the fields that
  6859. * depend only on the parent DB.
  6860. * @param[in] mc The main cursor whose sorted-dups cursor is to be initialized.
  6861. */
  6862. static void
  6863. mdb_xcursor_init0(MDB_cursor *mc)
  6864. {
  6865. MDB_xcursor *mx = mc->mc_xcursor;
  6866. mx->mx_cursor.mc_xcursor = NULL;
  6867. mx->mx_cursor.mc_txn = mc->mc_txn;
  6868. mx->mx_cursor.mc_db = &mx->mx_db;
  6869. mx->mx_cursor.mc_dbx = &mx->mx_dbx;
  6870. mx->mx_cursor.mc_dbi = mc->mc_dbi;
  6871. mx->mx_cursor.mc_dbflag = &mx->mx_dbflag;
  6872. mx->mx_cursor.mc_snum = 0;
  6873. mx->mx_cursor.mc_top = 0;
  6874. mx->mx_cursor.mc_flags = C_SUB;
  6875. mx->mx_dbx.md_name.mv_size = 0;
  6876. mx->mx_dbx.md_name.mv_data = NULL;
  6877. mx->mx_dbx.md_cmp = mc->mc_dbx->md_dcmp;
  6878. mx->mx_dbx.md_dcmp = NULL;
  6879. mx->mx_dbx.md_rel = mc->mc_dbx->md_rel;
  6880. }
  6881. /** Final setup of a sorted-dups cursor.
  6882. * Sets up the fields that depend on the data from the main cursor.
  6883. * @param[in] mc The main cursor whose sorted-dups cursor is to be initialized.
  6884. * @param[in] node The data containing the #MDB_db record for the
  6885. * sorted-dup database.
  6886. */
  6887. static void
  6888. mdb_xcursor_init1(MDB_cursor *mc, MDB_node *node)
  6889. {
  6890. MDB_xcursor *mx = mc->mc_xcursor;
  6891. if (node->mn_flags & F_SUBDATA) {
  6892. memcpy(&mx->mx_db, NODEDATA(node), sizeof(MDB_db));
  6893. mx->mx_cursor.mc_pg[0] = 0;
  6894. mx->mx_cursor.mc_snum = 0;
  6895. mx->mx_cursor.mc_top = 0;
  6896. mx->mx_cursor.mc_flags = C_SUB;
  6897. } else {
  6898. MDB_page *fp = NODEDATA(node);
  6899. mx->mx_db.md_pad = 0;
  6900. mx->mx_db.md_flags = 0;
  6901. mx->mx_db.md_depth = 1;
  6902. mx->mx_db.md_branch_pages = 0;
  6903. mx->mx_db.md_leaf_pages = 1;
  6904. mx->mx_db.md_overflow_pages = 0;
  6905. mx->mx_db.md_entries = NUMKEYS(fp);
  6906. COPY_PGNO(mx->mx_db.md_root, fp->mp_pgno);
  6907. mx->mx_cursor.mc_snum = 1;
  6908. mx->mx_cursor.mc_top = 0;
  6909. mx->mx_cursor.mc_flags = C_INITIALIZED|C_SUB;
  6910. mx->mx_cursor.mc_pg[0] = fp;
  6911. mx->mx_cursor.mc_ki[0] = 0;
  6912. if (mc->mc_db->md_flags & MDB_DUPFIXED) {
  6913. mx->mx_db.md_flags = MDB_DUPFIXED;
  6914. mx->mx_db.md_pad = fp->mp_pad;
  6915. if (mc->mc_db->md_flags & MDB_INTEGERDUP)
  6916. mx->mx_db.md_flags |= MDB_INTEGERKEY;
  6917. }
  6918. }
  6919. DPRINTF(("Sub-db -%u root page %"Z"u", mx->mx_cursor.mc_dbi,
  6920. mx->mx_db.md_root));
  6921. mx->mx_dbflag = DB_VALID|DB_USRVALID|DB_DUPDATA;
  6922. #if UINT_MAX < SIZE_MAX
  6923. if (mx->mx_dbx.md_cmp == mdb_cmp_int && mx->mx_db.md_pad == sizeof(size_t))
  6924. mx->mx_dbx.md_cmp = mdb_cmp_clong;
  6925. #endif
  6926. }
  6927. /** Fixup a sorted-dups cursor due to underlying update.
  6928. * Sets up some fields that depend on the data from the main cursor.
  6929. * Almost the same as init1, but skips initialization steps if the
  6930. * xcursor had already been used.
  6931. * @param[in] mc The main cursor whose sorted-dups cursor is to be fixed up.
  6932. * @param[in] src_mx The xcursor of an up-to-date cursor.
  6933. * @param[in] new_dupdata True if converting from a non-#F_DUPDATA item.
  6934. */
  6935. static void
  6936. mdb_xcursor_init2(MDB_cursor *mc, MDB_xcursor *src_mx, int new_dupdata)
  6937. {
  6938. MDB_xcursor *mx = mc->mc_xcursor;
  6939. if (new_dupdata) {
  6940. mx->mx_cursor.mc_snum = 1;
  6941. mx->mx_cursor.mc_top = 0;
  6942. mx->mx_cursor.mc_flags |= C_INITIALIZED;
  6943. mx->mx_cursor.mc_ki[0] = 0;
  6944. mx->mx_dbflag = DB_VALID|DB_USRVALID|DB_DUPDATA;
  6945. #if UINT_MAX < SIZE_MAX
  6946. mx->mx_dbx.md_cmp = src_mx->mx_dbx.md_cmp;
  6947. #endif
  6948. } else if (!(mx->mx_cursor.mc_flags & C_INITIALIZED)) {
  6949. return;
  6950. }
  6951. mx->mx_db = src_mx->mx_db;
  6952. mx->mx_cursor.mc_pg[0] = src_mx->mx_cursor.mc_pg[0];
  6953. DPRINTF(("Sub-db -%u root page %"Z"u", mx->mx_cursor.mc_dbi,
  6954. mx->mx_db.md_root));
  6955. }
  6956. /** Initialize a cursor for a given transaction and database. */
  6957. static void
  6958. mdb_cursor_init(MDB_cursor *mc, MDB_txn *txn, MDB_dbi dbi, MDB_xcursor *mx)
  6959. {
  6960. mc->mc_next = NULL;
  6961. mc->mc_backup = NULL;
  6962. mc->mc_dbi = dbi;
  6963. mc->mc_txn = txn;
  6964. mc->mc_db = &txn->mt_dbs[dbi];
  6965. mc->mc_dbx = &txn->mt_dbxs[dbi];
  6966. mc->mc_dbflag = &txn->mt_dbflags[dbi];
  6967. mc->mc_snum = 0;
  6968. mc->mc_top = 0;
  6969. mc->mc_pg[0] = 0;
  6970. mc->mc_ki[0] = 0;
  6971. mc->mc_flags = 0;
  6972. if (txn->mt_dbs[dbi].md_flags & MDB_DUPSORT) {
  6973. mdb_tassert(txn, mx != NULL);
  6974. mc->mc_xcursor = mx;
  6975. mdb_xcursor_init0(mc);
  6976. } else {
  6977. mc->mc_xcursor = NULL;
  6978. }
  6979. if (*mc->mc_dbflag & DB_STALE) {
  6980. mdb_page_search(mc, NULL, MDB_PS_ROOTONLY);
  6981. }
  6982. }
  6983. int
  6984. mdb_cursor_open(MDB_txn *txn, MDB_dbi dbi, MDB_cursor **ret)
  6985. {
  6986. MDB_cursor *mc;
  6987. size_t size = sizeof(MDB_cursor);
  6988. if (!ret || !TXN_DBI_EXIST(txn, dbi, DB_VALID))
  6989. return EINVAL;
  6990. if (txn->mt_flags & MDB_TXN_BLOCKED)
  6991. return MDB_BAD_TXN;
  6992. if (dbi == FREE_DBI && !F_ISSET(txn->mt_flags, MDB_TXN_RDONLY))
  6993. return EINVAL;
  6994. if (txn->mt_dbs[dbi].md_flags & MDB_DUPSORT)
  6995. size += sizeof(MDB_xcursor);
  6996. if ((mc = malloc(size)) != NULL) {
  6997. mdb_cursor_init(mc, txn, dbi, (MDB_xcursor *)(mc + 1));
  6998. if (txn->mt_cursors) {
  6999. mc->mc_next = txn->mt_cursors[dbi];
  7000. txn->mt_cursors[dbi] = mc;
  7001. mc->mc_flags |= C_UNTRACK;
  7002. }
  7003. } else {
  7004. return ENOMEM;
  7005. }
  7006. *ret = mc;
  7007. return MDB_SUCCESS;
  7008. }
  7009. int
  7010. mdb_cursor_renew(MDB_txn *txn, MDB_cursor *mc)
  7011. {
  7012. if (!mc || !TXN_DBI_EXIST(txn, mc->mc_dbi, DB_VALID))
  7013. return EINVAL;
  7014. if ((mc->mc_flags & C_UNTRACK) || txn->mt_cursors)
  7015. return EINVAL;
  7016. if (txn->mt_flags & MDB_TXN_BLOCKED)
  7017. return MDB_BAD_TXN;
  7018. mdb_cursor_init(mc, txn, mc->mc_dbi, mc->mc_xcursor);
  7019. return MDB_SUCCESS;
  7020. }
  7021. /* Return the count of duplicate data items for the current key */
  7022. int
  7023. mdb_cursor_count(MDB_cursor *mc, size_t *countp)
  7024. {
  7025. MDB_node *leaf;
  7026. if (mc == NULL || countp == NULL)
  7027. return EINVAL;
  7028. if (mc->mc_xcursor == NULL)
  7029. return MDB_INCOMPATIBLE;
  7030. if (mc->mc_txn->mt_flags & MDB_TXN_BLOCKED)
  7031. return MDB_BAD_TXN;
  7032. if (!(mc->mc_flags & C_INITIALIZED))
  7033. return EINVAL;
  7034. if (!mc->mc_snum)
  7035. return MDB_NOTFOUND;
  7036. if (mc->mc_flags & C_EOF) {
  7037. if (mc->mc_ki[mc->mc_top] >= NUMKEYS(mc->mc_pg[mc->mc_top]))
  7038. return MDB_NOTFOUND;
  7039. mc->mc_flags ^= C_EOF;
  7040. }
  7041. leaf = NODEPTR(mc->mc_pg[mc->mc_top], mc->mc_ki[mc->mc_top]);
  7042. if (!F_ISSET(leaf->mn_flags, F_DUPDATA)) {
  7043. *countp = 1;
  7044. } else {
  7045. if (!(mc->mc_xcursor->mx_cursor.mc_flags & C_INITIALIZED))
  7046. return EINVAL;
  7047. *countp = mc->mc_xcursor->mx_db.md_entries;
  7048. }
  7049. return MDB_SUCCESS;
  7050. }
  7051. void
  7052. mdb_cursor_close(MDB_cursor *mc)
  7053. {
  7054. if (mc && !mc->mc_backup) {
  7055. /* remove from txn, if tracked */
  7056. if ((mc->mc_flags & C_UNTRACK) && mc->mc_txn->mt_cursors) {
  7057. MDB_cursor **prev = &mc->mc_txn->mt_cursors[mc->mc_dbi];
  7058. while (*prev && *prev != mc) prev = &(*prev)->mc_next;
  7059. if (*prev == mc)
  7060. *prev = mc->mc_next;
  7061. }
  7062. free(mc);
  7063. }
  7064. }
  7065. MDB_txn *
  7066. mdb_cursor_txn(MDB_cursor *mc)
  7067. {
  7068. if (!mc) return NULL;
  7069. return mc->mc_txn;
  7070. }
  7071. MDB_dbi
  7072. mdb_cursor_dbi(MDB_cursor *mc)
  7073. {
  7074. return mc->mc_dbi;
  7075. }
  7076. /** Replace the key for a branch node with a new key.
  7077. * Set #MDB_TXN_ERROR on failure.
  7078. * @param[in] mc Cursor pointing to the node to operate on.
  7079. * @param[in] key The new key to use.
  7080. * @return 0 on success, non-zero on failure.
  7081. */
  7082. static int
  7083. mdb_update_key(MDB_cursor *mc, MDB_val *key)
  7084. {
  7085. MDB_page *mp;
  7086. MDB_node *node;
  7087. char *base;
  7088. size_t len;
  7089. int delta, ksize, oksize;
  7090. indx_t ptr, i, numkeys, indx;
  7091. DKBUF;
  7092. indx = mc->mc_ki[mc->mc_top];
  7093. mp = mc->mc_pg[mc->mc_top];
  7094. node = NODEPTR(mp, indx);
  7095. ptr = mp->mp_ptrs[indx];
  7096. #if MDB_DEBUG
  7097. {
  7098. MDB_val k2;
  7099. char kbuf2[DKBUF_MAXKEYSIZE*2+1];
  7100. k2.mv_data = NODEKEY(node);
  7101. k2.mv_size = node->mn_ksize;
  7102. DPRINTF(("update key %u (ofs %u) [%s] to [%s] on page %"Z"u",
  7103. indx, ptr,
  7104. mdb_dkey(&k2, kbuf2),
  7105. DKEY(key),
  7106. mp->mp_pgno));
  7107. }
  7108. #endif
  7109. /* Sizes must be 2-byte aligned. */
  7110. ksize = EVEN(key->mv_size);
  7111. oksize = EVEN(node->mn_ksize);
  7112. delta = ksize - oksize;
  7113. /* Shift node contents if EVEN(key length) changed. */
  7114. if (delta) {
  7115. if (delta > 0 && SIZELEFT(mp) < delta) {
  7116. pgno_t pgno;
  7117. /* not enough space left, do a delete and split */
  7118. DPRINTF(("Not enough room, delta = %d, splitting...", delta));
  7119. pgno = NODEPGNO(node);
  7120. mdb_node_del(mc, 0);
  7121. return mdb_page_split(mc, key, NULL, pgno, MDB_SPLIT_REPLACE);
  7122. }
  7123. numkeys = NUMKEYS(mp);
  7124. for (i = 0; i < numkeys; i++) {
  7125. if (mp->mp_ptrs[i] <= ptr)
  7126. mp->mp_ptrs[i] -= delta;
  7127. }
  7128. base = (char *)mp + mp->mp_upper + PAGEBASE;
  7129. len = ptr - mp->mp_upper + NODESIZE;
  7130. memmove(base - delta, base, len);
  7131. mp->mp_upper -= delta;
  7132. node = NODEPTR(mp, indx);
  7133. }
  7134. /* But even if no shift was needed, update ksize */
  7135. if (node->mn_ksize != key->mv_size)
  7136. node->mn_ksize = key->mv_size;
  7137. if (key->mv_size)
  7138. memcpy(NODEKEY(node), key->mv_data, key->mv_size);
  7139. return MDB_SUCCESS;
  7140. }
  7141. static void
  7142. mdb_cursor_copy(const MDB_cursor *csrc, MDB_cursor *cdst);
  7143. /** Perform \b act while tracking temporary cursor \b mn */
  7144. #define WITH_CURSOR_TRACKING(mn, act) do { \
  7145. MDB_cursor dummy, *tracked, **tp = &(mn).mc_txn->mt_cursors[mn.mc_dbi]; \
  7146. if ((mn).mc_flags & C_SUB) { \
  7147. dummy.mc_flags = C_INITIALIZED; \
  7148. dummy.mc_xcursor = (MDB_xcursor *)&(mn); \
  7149. tracked = &dummy; \
  7150. } else { \
  7151. tracked = &(mn); \
  7152. } \
  7153. tracked->mc_next = *tp; \
  7154. *tp = tracked; \
  7155. { act; } \
  7156. *tp = tracked->mc_next; \
  7157. } while (0)
  7158. /** Move a node from csrc to cdst.
  7159. */
  7160. static int
  7161. mdb_node_move(MDB_cursor *csrc, MDB_cursor *cdst, int fromleft)
  7162. {
  7163. MDB_node *srcnode;
  7164. MDB_val key, data;
  7165. pgno_t srcpg;
  7166. MDB_cursor mn;
  7167. int rc;
  7168. unsigned short flags;
  7169. DKBUF;
  7170. /* Mark src and dst as dirty. */
  7171. if ((rc = mdb_page_touch(csrc)) ||
  7172. (rc = mdb_page_touch(cdst)))
  7173. return rc;
  7174. if (IS_LEAF2(csrc->mc_pg[csrc->mc_top])) {
  7175. key.mv_size = csrc->mc_db->md_pad;
  7176. key.mv_data = LEAF2KEY(csrc->mc_pg[csrc->mc_top], csrc->mc_ki[csrc->mc_top], key.mv_size);
  7177. data.mv_size = 0;
  7178. data.mv_data = NULL;
  7179. srcpg = 0;
  7180. flags = 0;
  7181. } else {
  7182. srcnode = NODEPTR(csrc->mc_pg[csrc->mc_top], csrc->mc_ki[csrc->mc_top]);
  7183. mdb_cassert(csrc, !((size_t)srcnode & 1));
  7184. srcpg = NODEPGNO(srcnode);
  7185. flags = srcnode->mn_flags;
  7186. if (csrc->mc_ki[csrc->mc_top] == 0 && IS_BRANCH(csrc->mc_pg[csrc->mc_top])) {
  7187. unsigned int snum = csrc->mc_snum;
  7188. MDB_node *s2;
  7189. /* must find the lowest key below src */
  7190. rc = mdb_page_search_lowest(csrc);
  7191. if (rc)
  7192. return rc;
  7193. if (IS_LEAF2(csrc->mc_pg[csrc->mc_top])) {
  7194. key.mv_size = csrc->mc_db->md_pad;
  7195. key.mv_data = LEAF2KEY(csrc->mc_pg[csrc->mc_top], 0, key.mv_size);
  7196. } else {
  7197. s2 = NODEPTR(csrc->mc_pg[csrc->mc_top], 0);
  7198. key.mv_size = NODEKSZ(s2);
  7199. key.mv_data = NODEKEY(s2);
  7200. }
  7201. csrc->mc_snum = snum--;
  7202. csrc->mc_top = snum;
  7203. } else {
  7204. key.mv_size = NODEKSZ(srcnode);
  7205. key.mv_data = NODEKEY(srcnode);
  7206. }
  7207. data.mv_size = NODEDSZ(srcnode);
  7208. data.mv_data = NODEDATA(srcnode);
  7209. }
  7210. mn.mc_xcursor = NULL;
  7211. if (IS_BRANCH(cdst->mc_pg[cdst->mc_top]) && cdst->mc_ki[cdst->mc_top] == 0) {
  7212. unsigned int snum = cdst->mc_snum;
  7213. MDB_node *s2;
  7214. MDB_val bkey;
  7215. /* must find the lowest key below dst */
  7216. mdb_cursor_copy(cdst, &mn);
  7217. rc = mdb_page_search_lowest(&mn);
  7218. if (rc)
  7219. return rc;
  7220. if (IS_LEAF2(mn.mc_pg[mn.mc_top])) {
  7221. bkey.mv_size = mn.mc_db->md_pad;
  7222. bkey.mv_data = LEAF2KEY(mn.mc_pg[mn.mc_top], 0, bkey.mv_size);
  7223. } else {
  7224. s2 = NODEPTR(mn.mc_pg[mn.mc_top], 0);
  7225. bkey.mv_size = NODEKSZ(s2);
  7226. bkey.mv_data = NODEKEY(s2);
  7227. }
  7228. mn.mc_snum = snum--;
  7229. mn.mc_top = snum;
  7230. mn.mc_ki[snum] = 0;
  7231. rc = mdb_update_key(&mn, &bkey);
  7232. if (rc)
  7233. return rc;
  7234. }
  7235. DPRINTF(("moving %s node %u [%s] on page %"Z"u to node %u on page %"Z"u",
  7236. IS_LEAF(csrc->mc_pg[csrc->mc_top]) ? "leaf" : "branch",
  7237. csrc->mc_ki[csrc->mc_top],
  7238. DKEY(&key),
  7239. csrc->mc_pg[csrc->mc_top]->mp_pgno,
  7240. cdst->mc_ki[cdst->mc_top], cdst->mc_pg[cdst->mc_top]->mp_pgno));
  7241. /* Add the node to the destination page.
  7242. */
  7243. rc = mdb_node_add(cdst, cdst->mc_ki[cdst->mc_top], &key, &data, srcpg, flags);
  7244. if (rc != MDB_SUCCESS)
  7245. return rc;
  7246. /* Delete the node from the source page.
  7247. */
  7248. mdb_node_del(csrc, key.mv_size);
  7249. {
  7250. /* Adjust other cursors pointing to mp */
  7251. MDB_cursor *m2, *m3;
  7252. MDB_dbi dbi = csrc->mc_dbi;
  7253. MDB_page *mpd, *mps;
  7254. mps = csrc->mc_pg[csrc->mc_top];
  7255. /* If we're adding on the left, bump others up */
  7256. if (fromleft) {
  7257. mpd = cdst->mc_pg[csrc->mc_top];
  7258. for (m2 = csrc->mc_txn->mt_cursors[dbi]; m2; m2=m2->mc_next) {
  7259. if (csrc->mc_flags & C_SUB)
  7260. m3 = &m2->mc_xcursor->mx_cursor;
  7261. else
  7262. m3 = m2;
  7263. if (!(m3->mc_flags & C_INITIALIZED) || m3->mc_top < csrc->mc_top)
  7264. continue;
  7265. if (m3 != cdst &&
  7266. m3->mc_pg[csrc->mc_top] == mpd &&
  7267. m3->mc_ki[csrc->mc_top] >= cdst->mc_ki[csrc->mc_top]) {
  7268. m3->mc_ki[csrc->mc_top]++;
  7269. }
  7270. if (m3 !=csrc &&
  7271. m3->mc_pg[csrc->mc_top] == mps &&
  7272. m3->mc_ki[csrc->mc_top] == csrc->mc_ki[csrc->mc_top]) {
  7273. m3->mc_pg[csrc->mc_top] = cdst->mc_pg[cdst->mc_top];
  7274. m3->mc_ki[csrc->mc_top] = cdst->mc_ki[cdst->mc_top];
  7275. m3->mc_ki[csrc->mc_top-1]++;
  7276. }
  7277. if (IS_LEAF(mps))
  7278. XCURSOR_REFRESH(m3, csrc->mc_top, m3->mc_pg[csrc->mc_top]);
  7279. }
  7280. } else
  7281. /* Adding on the right, bump others down */
  7282. {
  7283. for (m2 = csrc->mc_txn->mt_cursors[dbi]; m2; m2=m2->mc_next) {
  7284. if (csrc->mc_flags & C_SUB)
  7285. m3 = &m2->mc_xcursor->mx_cursor;
  7286. else
  7287. m3 = m2;
  7288. if (m3 == csrc) continue;
  7289. if (!(m3->mc_flags & C_INITIALIZED) || m3->mc_top < csrc->mc_top)
  7290. continue;
  7291. if (m3->mc_pg[csrc->mc_top] == mps) {
  7292. if (!m3->mc_ki[csrc->mc_top]) {
  7293. m3->mc_pg[csrc->mc_top] = cdst->mc_pg[cdst->mc_top];
  7294. m3->mc_ki[csrc->mc_top] = cdst->mc_ki[cdst->mc_top];
  7295. m3->mc_ki[csrc->mc_top-1]--;
  7296. } else {
  7297. m3->mc_ki[csrc->mc_top]--;
  7298. }
  7299. if (IS_LEAF(mps))
  7300. XCURSOR_REFRESH(m3, csrc->mc_top, m3->mc_pg[csrc->mc_top]);
  7301. }
  7302. }
  7303. }
  7304. }
  7305. /* Update the parent separators.
  7306. */
  7307. if (csrc->mc_ki[csrc->mc_top] == 0) {
  7308. if (csrc->mc_ki[csrc->mc_top-1] != 0) {
  7309. if (IS_LEAF2(csrc->mc_pg[csrc->mc_top])) {
  7310. key.mv_data = LEAF2KEY(csrc->mc_pg[csrc->mc_top], 0, key.mv_size);
  7311. } else {
  7312. srcnode = NODEPTR(csrc->mc_pg[csrc->mc_top], 0);
  7313. key.mv_size = NODEKSZ(srcnode);
  7314. key.mv_data = NODEKEY(srcnode);
  7315. }
  7316. DPRINTF(("update separator for source page %"Z"u to [%s]",
  7317. csrc->mc_pg[csrc->mc_top]->mp_pgno, DKEY(&key)));
  7318. mdb_cursor_copy(csrc, &mn);
  7319. mn.mc_snum--;
  7320. mn.mc_top--;
  7321. /* We want mdb_rebalance to find mn when doing fixups */
  7322. WITH_CURSOR_TRACKING(mn,
  7323. rc = mdb_update_key(&mn, &key));
  7324. if (rc)
  7325. return rc;
  7326. }
  7327. if (IS_BRANCH(csrc->mc_pg[csrc->mc_top])) {
  7328. MDB_val nullkey;
  7329. indx_t ix = csrc->mc_ki[csrc->mc_top];
  7330. nullkey.mv_size = 0;
  7331. csrc->mc_ki[csrc->mc_top] = 0;
  7332. rc = mdb_update_key(csrc, &nullkey);
  7333. csrc->mc_ki[csrc->mc_top] = ix;
  7334. mdb_cassert(csrc, rc == MDB_SUCCESS);
  7335. }
  7336. }
  7337. if (cdst->mc_ki[cdst->mc_top] == 0) {
  7338. if (cdst->mc_ki[cdst->mc_top-1] != 0) {
  7339. if (IS_LEAF2(csrc->mc_pg[csrc->mc_top])) {
  7340. key.mv_data = LEAF2KEY(cdst->mc_pg[cdst->mc_top], 0, key.mv_size);
  7341. } else {
  7342. srcnode = NODEPTR(cdst->mc_pg[cdst->mc_top], 0);
  7343. key.mv_size = NODEKSZ(srcnode);
  7344. key.mv_data = NODEKEY(srcnode);
  7345. }
  7346. DPRINTF(("update separator for destination page %"Z"u to [%s]",
  7347. cdst->mc_pg[cdst->mc_top]->mp_pgno, DKEY(&key)));
  7348. mdb_cursor_copy(cdst, &mn);
  7349. mn.mc_snum--;
  7350. mn.mc_top--;
  7351. /* We want mdb_rebalance to find mn when doing fixups */
  7352. WITH_CURSOR_TRACKING(mn,
  7353. rc = mdb_update_key(&mn, &key));
  7354. if (rc)
  7355. return rc;
  7356. }
  7357. if (IS_BRANCH(cdst->mc_pg[cdst->mc_top])) {
  7358. MDB_val nullkey;
  7359. indx_t ix = cdst->mc_ki[cdst->mc_top];
  7360. nullkey.mv_size = 0;
  7361. cdst->mc_ki[cdst->mc_top] = 0;
  7362. rc = mdb_update_key(cdst, &nullkey);
  7363. cdst->mc_ki[cdst->mc_top] = ix;
  7364. mdb_cassert(cdst, rc == MDB_SUCCESS);
  7365. }
  7366. }
  7367. return MDB_SUCCESS;
  7368. }
  7369. /** Merge one page into another.
  7370. * The nodes from the page pointed to by \b csrc will
  7371. * be copied to the page pointed to by \b cdst and then
  7372. * the \b csrc page will be freed.
  7373. * @param[in] csrc Cursor pointing to the source page.
  7374. * @param[in] cdst Cursor pointing to the destination page.
  7375. * @return 0 on success, non-zero on failure.
  7376. */
  7377. static int
  7378. mdb_page_merge(MDB_cursor *csrc, MDB_cursor *cdst)
  7379. {
  7380. MDB_page *psrc, *pdst;
  7381. MDB_node *srcnode;
  7382. MDB_val key, data;
  7383. unsigned nkeys;
  7384. int rc;
  7385. indx_t i, j;
  7386. psrc = csrc->mc_pg[csrc->mc_top];
  7387. pdst = cdst->mc_pg[cdst->mc_top];
  7388. DPRINTF(("merging page %"Z"u into %"Z"u", psrc->mp_pgno, pdst->mp_pgno));
  7389. mdb_cassert(csrc, csrc->mc_snum > 1); /* can't merge root page */
  7390. mdb_cassert(csrc, cdst->mc_snum > 1);
  7391. /* Mark dst as dirty. */
  7392. if ((rc = mdb_page_touch(cdst)))
  7393. return rc;
  7394. /* get dst page again now that we've touched it. */
  7395. pdst = cdst->mc_pg[cdst->mc_top];
  7396. /* Move all nodes from src to dst.
  7397. */
  7398. j = nkeys = NUMKEYS(pdst);
  7399. if (IS_LEAF2(psrc)) {
  7400. key.mv_size = csrc->mc_db->md_pad;
  7401. key.mv_data = METADATA(psrc);
  7402. for (i = 0; i < NUMKEYS(psrc); i++, j++) {
  7403. rc = mdb_node_add(cdst, j, &key, NULL, 0, 0);
  7404. if (rc != MDB_SUCCESS)
  7405. return rc;
  7406. key.mv_data = (char *)key.mv_data + key.mv_size;
  7407. }
  7408. } else {
  7409. for (i = 0; i < NUMKEYS(psrc); i++, j++) {
  7410. srcnode = NODEPTR(psrc, i);
  7411. if (i == 0 && IS_BRANCH(psrc)) {
  7412. MDB_cursor mn;
  7413. MDB_node *s2;
  7414. mdb_cursor_copy(csrc, &mn);
  7415. mn.mc_xcursor = NULL;
  7416. /* must find the lowest key below src */
  7417. rc = mdb_page_search_lowest(&mn);
  7418. if (rc)
  7419. return rc;
  7420. if (IS_LEAF2(mn.mc_pg[mn.mc_top])) {
  7421. key.mv_size = mn.mc_db->md_pad;
  7422. key.mv_data = LEAF2KEY(mn.mc_pg[mn.mc_top], 0, key.mv_size);
  7423. } else {
  7424. s2 = NODEPTR(mn.mc_pg[mn.mc_top], 0);
  7425. key.mv_size = NODEKSZ(s2);
  7426. key.mv_data = NODEKEY(s2);
  7427. }
  7428. } else {
  7429. key.mv_size = srcnode->mn_ksize;
  7430. key.mv_data = NODEKEY(srcnode);
  7431. }
  7432. data.mv_size = NODEDSZ(srcnode);
  7433. data.mv_data = NODEDATA(srcnode);
  7434. rc = mdb_node_add(cdst, j, &key, &data, NODEPGNO(srcnode), srcnode->mn_flags);
  7435. if (rc != MDB_SUCCESS)
  7436. return rc;
  7437. }
  7438. }
  7439. DPRINTF(("dst page %"Z"u now has %u keys (%.1f%% filled)",
  7440. pdst->mp_pgno, NUMKEYS(pdst),
  7441. (float)PAGEFILL(cdst->mc_txn->mt_env, pdst) / 10));
  7442. /* Unlink the src page from parent and add to free list.
  7443. */
  7444. csrc->mc_top--;
  7445. mdb_node_del(csrc, 0);
  7446. if (csrc->mc_ki[csrc->mc_top] == 0) {
  7447. key.mv_size = 0;
  7448. rc = mdb_update_key(csrc, &key);
  7449. if (rc) {
  7450. csrc->mc_top++;
  7451. return rc;
  7452. }
  7453. }
  7454. csrc->mc_top++;
  7455. psrc = csrc->mc_pg[csrc->mc_top];
  7456. /* If not operating on FreeDB, allow this page to be reused
  7457. * in this txn. Otherwise just add to free list.
  7458. */
  7459. rc = mdb_page_loose(csrc, psrc);
  7460. if (rc)
  7461. return rc;
  7462. if (IS_LEAF(psrc))
  7463. csrc->mc_db->md_leaf_pages--;
  7464. else
  7465. csrc->mc_db->md_branch_pages--;
  7466. {
  7467. /* Adjust other cursors pointing to mp */
  7468. MDB_cursor *m2, *m3;
  7469. MDB_dbi dbi = csrc->mc_dbi;
  7470. unsigned int top = csrc->mc_top;
  7471. for (m2 = csrc->mc_txn->mt_cursors[dbi]; m2; m2=m2->mc_next) {
  7472. if (csrc->mc_flags & C_SUB)
  7473. m3 = &m2->mc_xcursor->mx_cursor;
  7474. else
  7475. m3 = m2;
  7476. if (m3 == csrc) continue;
  7477. if (m3->mc_snum < csrc->mc_snum) continue;
  7478. if (m3->mc_pg[top] == psrc) {
  7479. m3->mc_pg[top] = pdst;
  7480. m3->mc_ki[top] += nkeys;
  7481. m3->mc_ki[top-1] = cdst->mc_ki[top-1];
  7482. } else if (m3->mc_pg[top-1] == csrc->mc_pg[top-1] &&
  7483. m3->mc_ki[top-1] > csrc->mc_ki[top-1]) {
  7484. m3->mc_ki[top-1]--;
  7485. }
  7486. if (IS_LEAF(psrc))
  7487. XCURSOR_REFRESH(m3, top, m3->mc_pg[top]);
  7488. }
  7489. }
  7490. {
  7491. unsigned int snum = cdst->mc_snum;
  7492. uint16_t depth = cdst->mc_db->md_depth;
  7493. mdb_cursor_pop(cdst);
  7494. rc = mdb_rebalance(cdst);
  7495. /* Did the tree height change? */
  7496. if (depth != cdst->mc_db->md_depth)
  7497. snum += cdst->mc_db->md_depth - depth;
  7498. cdst->mc_snum = snum;
  7499. cdst->mc_top = snum-1;
  7500. }
  7501. return rc;
  7502. }
  7503. /** Copy the contents of a cursor.
  7504. * @param[in] csrc The cursor to copy from.
  7505. * @param[out] cdst The cursor to copy to.
  7506. */
  7507. static void
  7508. mdb_cursor_copy(const MDB_cursor *csrc, MDB_cursor *cdst)
  7509. {
  7510. unsigned int i;
  7511. cdst->mc_txn = csrc->mc_txn;
  7512. cdst->mc_dbi = csrc->mc_dbi;
  7513. cdst->mc_db = csrc->mc_db;
  7514. cdst->mc_dbx = csrc->mc_dbx;
  7515. cdst->mc_snum = csrc->mc_snum;
  7516. cdst->mc_top = csrc->mc_top;
  7517. cdst->mc_flags = csrc->mc_flags;
  7518. for (i=0; i<csrc->mc_snum; i++) {
  7519. cdst->mc_pg[i] = csrc->mc_pg[i];
  7520. cdst->mc_ki[i] = csrc->mc_ki[i];
  7521. }
  7522. }
  7523. /** Rebalance the tree after a delete operation.
  7524. * @param[in] mc Cursor pointing to the page where rebalancing
  7525. * should begin.
  7526. * @return 0 on success, non-zero on failure.
  7527. */
  7528. static int
  7529. mdb_rebalance(MDB_cursor *mc)
  7530. {
  7531. MDB_node *node;
  7532. int rc, fromleft;
  7533. unsigned int ptop, minkeys, thresh;
  7534. MDB_cursor mn;
  7535. indx_t oldki;
  7536. if (IS_BRANCH(mc->mc_pg[mc->mc_top])) {
  7537. minkeys = 2;
  7538. thresh = 1;
  7539. } else {
  7540. minkeys = 1;
  7541. thresh = FILL_THRESHOLD;
  7542. }
  7543. DPRINTF(("rebalancing %s page %"Z"u (has %u keys, %.1f%% full)",
  7544. IS_LEAF(mc->mc_pg[mc->mc_top]) ? "leaf" : "branch",
  7545. mdb_dbg_pgno(mc->mc_pg[mc->mc_top]), NUMKEYS(mc->mc_pg[mc->mc_top]),
  7546. (float)PAGEFILL(mc->mc_txn->mt_env, mc->mc_pg[mc->mc_top]) / 10));
  7547. if (PAGEFILL(mc->mc_txn->mt_env, mc->mc_pg[mc->mc_top]) >= thresh &&
  7548. NUMKEYS(mc->mc_pg[mc->mc_top]) >= minkeys) {
  7549. DPRINTF(("no need to rebalance page %"Z"u, above fill threshold",
  7550. mdb_dbg_pgno(mc->mc_pg[mc->mc_top])));
  7551. return MDB_SUCCESS;
  7552. }
  7553. if (mc->mc_snum < 2) {
  7554. MDB_page *mp = mc->mc_pg[0];
  7555. if (IS_SUBP(mp)) {
  7556. DPUTS("Can't rebalance a subpage, ignoring");
  7557. return MDB_SUCCESS;
  7558. }
  7559. if (NUMKEYS(mp) == 0) {
  7560. DPUTS("tree is completely empty");
  7561. mc->mc_db->md_root = P_INVALID;
  7562. mc->mc_db->md_depth = 0;
  7563. mc->mc_db->md_leaf_pages = 0;
  7564. rc = mdb_midl_append(&mc->mc_txn->mt_free_pgs, mp->mp_pgno);
  7565. if (rc)
  7566. return rc;
  7567. /* Adjust cursors pointing to mp */
  7568. mc->mc_snum = 0;
  7569. mc->mc_top = 0;
  7570. mc->mc_flags &= ~C_INITIALIZED;
  7571. {
  7572. MDB_cursor *m2, *m3;
  7573. MDB_dbi dbi = mc->mc_dbi;
  7574. for (m2 = mc->mc_txn->mt_cursors[dbi]; m2; m2=m2->mc_next) {
  7575. if (mc->mc_flags & C_SUB)
  7576. m3 = &m2->mc_xcursor->mx_cursor;
  7577. else
  7578. m3 = m2;
  7579. if (!(m3->mc_flags & C_INITIALIZED) || (m3->mc_snum < mc->mc_snum))
  7580. continue;
  7581. if (m3->mc_pg[0] == mp) {
  7582. m3->mc_snum = 0;
  7583. m3->mc_top = 0;
  7584. m3->mc_flags &= ~C_INITIALIZED;
  7585. }
  7586. }
  7587. }
  7588. } else if (IS_BRANCH(mp) && NUMKEYS(mp) == 1) {
  7589. int i;
  7590. DPUTS("collapsing root page!");
  7591. rc = mdb_midl_append(&mc->mc_txn->mt_free_pgs, mp->mp_pgno);
  7592. if (rc)
  7593. return rc;
  7594. mc->mc_db->md_root = NODEPGNO(NODEPTR(mp, 0));
  7595. rc = mdb_page_get(mc, mc->mc_db->md_root, &mc->mc_pg[0], NULL);
  7596. if (rc)
  7597. return rc;
  7598. mc->mc_db->md_depth--;
  7599. mc->mc_db->md_branch_pages--;
  7600. mc->mc_ki[0] = mc->mc_ki[1];
  7601. for (i = 1; i<mc->mc_db->md_depth; i++) {
  7602. mc->mc_pg[i] = mc->mc_pg[i+1];
  7603. mc->mc_ki[i] = mc->mc_ki[i+1];
  7604. }
  7605. {
  7606. /* Adjust other cursors pointing to mp */
  7607. MDB_cursor *m2, *m3;
  7608. MDB_dbi dbi = mc->mc_dbi;
  7609. for (m2 = mc->mc_txn->mt_cursors[dbi]; m2; m2=m2->mc_next) {
  7610. if (mc->mc_flags & C_SUB)
  7611. m3 = &m2->mc_xcursor->mx_cursor;
  7612. else
  7613. m3 = m2;
  7614. if (m3 == mc) continue;
  7615. if (!(m3->mc_flags & C_INITIALIZED))
  7616. continue;
  7617. if (m3->mc_pg[0] == mp) {
  7618. for (i=0; i<mc->mc_db->md_depth; i++) {
  7619. m3->mc_pg[i] = m3->mc_pg[i+1];
  7620. m3->mc_ki[i] = m3->mc_ki[i+1];
  7621. }
  7622. m3->mc_snum--;
  7623. m3->mc_top--;
  7624. }
  7625. }
  7626. }
  7627. } else
  7628. DPUTS("root page doesn't need rebalancing");
  7629. return MDB_SUCCESS;
  7630. }
  7631. /* The parent (branch page) must have at least 2 pointers,
  7632. * otherwise the tree is invalid.
  7633. */
  7634. ptop = mc->mc_top-1;
  7635. mdb_cassert(mc, NUMKEYS(mc->mc_pg[ptop]) > 1);
  7636. /* Leaf page fill factor is below the threshold.
  7637. * Try to move keys from left or right neighbor, or
  7638. * merge with a neighbor page.
  7639. */
  7640. /* Find neighbors.
  7641. */
  7642. mdb_cursor_copy(mc, &mn);
  7643. mn.mc_xcursor = NULL;
  7644. oldki = mc->mc_ki[mc->mc_top];
  7645. if (mc->mc_ki[ptop] == 0) {
  7646. /* We're the leftmost leaf in our parent.
  7647. */
  7648. DPUTS("reading right neighbor");
  7649. mn.mc_ki[ptop]++;
  7650. node = NODEPTR(mc->mc_pg[ptop], mn.mc_ki[ptop]);
  7651. rc = mdb_page_get(mc, NODEPGNO(node), &mn.mc_pg[mn.mc_top], NULL);
  7652. if (rc)
  7653. return rc;
  7654. mn.mc_ki[mn.mc_top] = 0;
  7655. mc->mc_ki[mc->mc_top] = NUMKEYS(mc->mc_pg[mc->mc_top]);
  7656. fromleft = 0;
  7657. } else {
  7658. /* There is at least one neighbor to the left.
  7659. */
  7660. DPUTS("reading left neighbor");
  7661. mn.mc_ki[ptop]--;
  7662. node = NODEPTR(mc->mc_pg[ptop], mn.mc_ki[ptop]);
  7663. rc = mdb_page_get(mc, NODEPGNO(node), &mn.mc_pg[mn.mc_top], NULL);
  7664. if (rc)
  7665. return rc;
  7666. mn.mc_ki[mn.mc_top] = NUMKEYS(mn.mc_pg[mn.mc_top]) - 1;
  7667. mc->mc_ki[mc->mc_top] = 0;
  7668. fromleft = 1;
  7669. }
  7670. DPRINTF(("found neighbor page %"Z"u (%u keys, %.1f%% full)",
  7671. mn.mc_pg[mn.mc_top]->mp_pgno, NUMKEYS(mn.mc_pg[mn.mc_top]),
  7672. (float)PAGEFILL(mc->mc_txn->mt_env, mn.mc_pg[mn.mc_top]) / 10));
  7673. /* If the neighbor page is above threshold and has enough keys,
  7674. * move one key from it. Otherwise we should try to merge them.
  7675. * (A branch page must never have less than 2 keys.)
  7676. */
  7677. if (PAGEFILL(mc->mc_txn->mt_env, mn.mc_pg[mn.mc_top]) >= thresh && NUMKEYS(mn.mc_pg[mn.mc_top]) > minkeys) {
  7678. rc = mdb_node_move(&mn, mc, fromleft);
  7679. if (fromleft) {
  7680. /* if we inserted on left, bump position up */
  7681. oldki++;
  7682. }
  7683. } else {
  7684. if (!fromleft) {
  7685. rc = mdb_page_merge(&mn, mc);
  7686. } else {
  7687. oldki += NUMKEYS(mn.mc_pg[mn.mc_top]);
  7688. mn.mc_ki[mn.mc_top] += mc->mc_ki[mn.mc_top] + 1;
  7689. /* We want mdb_rebalance to find mn when doing fixups */
  7690. WITH_CURSOR_TRACKING(mn,
  7691. rc = mdb_page_merge(mc, &mn));
  7692. mdb_cursor_copy(&mn, mc);
  7693. }
  7694. mc->mc_flags &= ~C_EOF;
  7695. }
  7696. mc->mc_ki[mc->mc_top] = oldki;
  7697. return rc;
  7698. }
  7699. /** Complete a delete operation started by #mdb_cursor_del(). */
  7700. static int
  7701. mdb_cursor_del0(MDB_cursor *mc)
  7702. {
  7703. int rc;
  7704. MDB_page *mp;
  7705. indx_t ki;
  7706. unsigned int nkeys;
  7707. MDB_cursor *m2, *m3;
  7708. MDB_dbi dbi = mc->mc_dbi;
  7709. ki = mc->mc_ki[mc->mc_top];
  7710. mp = mc->mc_pg[mc->mc_top];
  7711. mdb_node_del(mc, mc->mc_db->md_pad);
  7712. mc->mc_db->md_entries--;
  7713. {
  7714. /* Adjust other cursors pointing to mp */
  7715. for (m2 = mc->mc_txn->mt_cursors[dbi]; m2; m2=m2->mc_next) {
  7716. m3 = (mc->mc_flags & C_SUB) ? &m2->mc_xcursor->mx_cursor : m2;
  7717. if (! (m2->mc_flags & m3->mc_flags & C_INITIALIZED))
  7718. continue;
  7719. if (m3 == mc || m3->mc_snum < mc->mc_snum)
  7720. continue;
  7721. if (m3->mc_pg[mc->mc_top] == mp) {
  7722. if (m3->mc_ki[mc->mc_top] == ki) {
  7723. m3->mc_flags |= C_DEL;
  7724. if (mc->mc_db->md_flags & MDB_DUPSORT) {
  7725. /* Sub-cursor referred into dataset which is gone */
  7726. m3->mc_xcursor->mx_cursor.mc_flags &= ~(C_INITIALIZED|C_EOF);
  7727. }
  7728. continue;
  7729. } else if (m3->mc_ki[mc->mc_top] > ki) {
  7730. m3->mc_ki[mc->mc_top]--;
  7731. }
  7732. XCURSOR_REFRESH(m3, mc->mc_top, mp);
  7733. }
  7734. }
  7735. }
  7736. rc = mdb_rebalance(mc);
  7737. if (rc)
  7738. goto fail;
  7739. /* DB is totally empty now, just bail out.
  7740. * Other cursors adjustments were already done
  7741. * by mdb_rebalance and aren't needed here.
  7742. */
  7743. if (!mc->mc_snum) {
  7744. mc->mc_flags |= C_EOF;
  7745. return rc;
  7746. }
  7747. mp = mc->mc_pg[mc->mc_top];
  7748. nkeys = NUMKEYS(mp);
  7749. /* Adjust other cursors pointing to mp */
  7750. for (m2 = mc->mc_txn->mt_cursors[dbi]; !rc && m2; m2=m2->mc_next) {
  7751. m3 = (mc->mc_flags & C_SUB) ? &m2->mc_xcursor->mx_cursor : m2;
  7752. if (!(m2->mc_flags & m3->mc_flags & C_INITIALIZED))
  7753. continue;
  7754. if (m3->mc_snum < mc->mc_snum)
  7755. continue;
  7756. if (m3->mc_pg[mc->mc_top] == mp) {
  7757. if (m3->mc_ki[mc->mc_top] >= mc->mc_ki[mc->mc_top]) {
  7758. /* if m3 points past last node in page, find next sibling */
  7759. if (m3->mc_ki[mc->mc_top] >= nkeys) {
  7760. rc = mdb_cursor_sibling(m3, 1);
  7761. if (rc == MDB_NOTFOUND) {
  7762. m3->mc_flags |= C_EOF;
  7763. rc = MDB_SUCCESS;
  7764. continue;
  7765. }
  7766. if (rc)
  7767. goto fail;
  7768. }
  7769. if (m3->mc_xcursor && !(m3->mc_flags & C_EOF)) {
  7770. MDB_node *node = NODEPTR(m3->mc_pg[m3->mc_top], m3->mc_ki[m3->mc_top]);
  7771. /* If this node has dupdata, it may need to be reinited
  7772. * because its data has moved.
  7773. * If the xcursor was not initd it must be reinited.
  7774. * Else if node points to a subDB, nothing is needed.
  7775. * Else (xcursor was initd, not a subDB) needs mc_pg[0] reset.
  7776. */
  7777. if (node->mn_flags & F_DUPDATA) {
  7778. if (m3->mc_xcursor->mx_cursor.mc_flags & C_INITIALIZED) {
  7779. if (!(node->mn_flags & F_SUBDATA))
  7780. m3->mc_xcursor->mx_cursor.mc_pg[0] = NODEDATA(node);
  7781. } else {
  7782. mdb_xcursor_init1(m3, node);
  7783. rc = mdb_cursor_first(&m3->mc_xcursor->mx_cursor, NULL, NULL);
  7784. if (rc)
  7785. goto fail;
  7786. }
  7787. }
  7788. m3->mc_xcursor->mx_cursor.mc_flags |= C_DEL;
  7789. }
  7790. }
  7791. }
  7792. }
  7793. mc->mc_flags |= C_DEL;
  7794. fail:
  7795. if (rc)
  7796. mc->mc_txn->mt_flags |= MDB_TXN_ERROR;
  7797. return rc;
  7798. }
  7799. int
  7800. mdb_del(MDB_txn *txn, MDB_dbi dbi,
  7801. MDB_val *key, MDB_val *data)
  7802. {
  7803. if (!key || !TXN_DBI_EXIST(txn, dbi, DB_USRVALID))
  7804. return EINVAL;
  7805. if (txn->mt_flags & (MDB_TXN_RDONLY|MDB_TXN_BLOCKED))
  7806. return (txn->mt_flags & MDB_TXN_RDONLY) ? EACCES : MDB_BAD_TXN;
  7807. if (!F_ISSET(txn->mt_dbs[dbi].md_flags, MDB_DUPSORT)) {
  7808. /* must ignore any data */
  7809. data = NULL;
  7810. }
  7811. return mdb_del0(txn, dbi, key, data, 0);
  7812. }
  7813. static int
  7814. mdb_del0(MDB_txn *txn, MDB_dbi dbi,
  7815. MDB_val *key, MDB_val *data, unsigned flags)
  7816. {
  7817. MDB_cursor mc;
  7818. MDB_xcursor mx;
  7819. MDB_cursor_op op;
  7820. MDB_val rdata, *xdata;
  7821. int rc, exact = 0;
  7822. DKBUF;
  7823. DPRINTF(("====> delete db %u key [%s]", dbi, DKEY(key)));
  7824. mdb_cursor_init(&mc, txn, dbi, &mx);
  7825. if (data) {
  7826. op = MDB_GET_BOTH;
  7827. rdata = *data;
  7828. xdata = &rdata;
  7829. } else {
  7830. op = MDB_SET;
  7831. xdata = NULL;
  7832. flags |= MDB_NODUPDATA;
  7833. }
  7834. rc = mdb_cursor_set(&mc, key, xdata, op, &exact);
  7835. if (rc == 0) {
  7836. /* let mdb_page_split know about this cursor if needed:
  7837. * delete will trigger a rebalance; if it needs to move
  7838. * a node from one page to another, it will have to
  7839. * update the parent's separator key(s). If the new sepkey
  7840. * is larger than the current one, the parent page may
  7841. * run out of space, triggering a split. We need this
  7842. * cursor to be consistent until the end of the rebalance.
  7843. */
  7844. mc.mc_flags |= C_UNTRACK;
  7845. mc.mc_next = txn->mt_cursors[dbi];
  7846. txn->mt_cursors[dbi] = &mc;
  7847. rc = mdb_cursor_del(&mc, flags);
  7848. txn->mt_cursors[dbi] = mc.mc_next;
  7849. }
  7850. return rc;
  7851. }
  7852. /** Split a page and insert a new node.
  7853. * Set #MDB_TXN_ERROR on failure.
  7854. * @param[in,out] mc Cursor pointing to the page and desired insertion index.
  7855. * The cursor will be updated to point to the actual page and index where
  7856. * the node got inserted after the split.
  7857. * @param[in] newkey The key for the newly inserted node.
  7858. * @param[in] newdata The data for the newly inserted node.
  7859. * @param[in] newpgno The page number, if the new node is a branch node.
  7860. * @param[in] nflags The #NODE_ADD_FLAGS for the new node.
  7861. * @return 0 on success, non-zero on failure.
  7862. */
  7863. static int
  7864. mdb_page_split(MDB_cursor *mc, MDB_val *newkey, MDB_val *newdata, pgno_t newpgno,
  7865. unsigned int nflags)
  7866. {
  7867. unsigned int flags;
  7868. int rc = MDB_SUCCESS, new_root = 0, did_split = 0;
  7869. indx_t newindx;
  7870. pgno_t pgno = 0;
  7871. int i, j, split_indx, nkeys, pmax;
  7872. MDB_env *env = mc->mc_txn->mt_env;
  7873. MDB_node *node;
  7874. MDB_val sepkey, rkey, xdata, *rdata = &xdata;
  7875. MDB_page *copy = NULL;
  7876. MDB_page *mp, *rp, *pp;
  7877. int ptop;
  7878. MDB_cursor mn;
  7879. DKBUF;
  7880. mp = mc->mc_pg[mc->mc_top];
  7881. newindx = mc->mc_ki[mc->mc_top];
  7882. nkeys = NUMKEYS(mp);
  7883. DPRINTF(("-----> splitting %s page %"Z"u and adding [%s] at index %i/%i",
  7884. IS_LEAF(mp) ? "leaf" : "branch", mp->mp_pgno,
  7885. DKEY(newkey), mc->mc_ki[mc->mc_top], nkeys));
  7886. /* Create a right sibling. */
  7887. if ((rc = mdb_page_new(mc, mp->mp_flags, 1, &rp)))
  7888. return rc;
  7889. rp->mp_pad = mp->mp_pad;
  7890. DPRINTF(("new right sibling: page %"Z"u", rp->mp_pgno));
  7891. /* Usually when splitting the root page, the cursor
  7892. * height is 1. But when called from mdb_update_key,
  7893. * the cursor height may be greater because it walks
  7894. * up the stack while finding the branch slot to update.
  7895. */
  7896. if (mc->mc_top < 1) {
  7897. if ((rc = mdb_page_new(mc, P_BRANCH, 1, &pp)))
  7898. goto done;
  7899. /* shift current top to make room for new parent */
  7900. for (i=mc->mc_snum; i>0; i--) {
  7901. mc->mc_pg[i] = mc->mc_pg[i-1];
  7902. mc->mc_ki[i] = mc->mc_ki[i-1];
  7903. }
  7904. mc->mc_pg[0] = pp;
  7905. mc->mc_ki[0] = 0;
  7906. mc->mc_db->md_root = pp->mp_pgno;
  7907. DPRINTF(("root split! new root = %"Z"u", pp->mp_pgno));
  7908. new_root = mc->mc_db->md_depth++;
  7909. /* Add left (implicit) pointer. */
  7910. if ((rc = mdb_node_add(mc, 0, NULL, NULL, mp->mp_pgno, 0)) != MDB_SUCCESS) {
  7911. /* undo the pre-push */
  7912. mc->mc_pg[0] = mc->mc_pg[1];
  7913. mc->mc_ki[0] = mc->mc_ki[1];
  7914. mc->mc_db->md_root = mp->mp_pgno;
  7915. mc->mc_db->md_depth--;
  7916. goto done;
  7917. }
  7918. mc->mc_snum++;
  7919. mc->mc_top++;
  7920. ptop = 0;
  7921. } else {
  7922. ptop = mc->mc_top-1;
  7923. DPRINTF(("parent branch page is %"Z"u", mc->mc_pg[ptop]->mp_pgno));
  7924. }
  7925. mdb_cursor_copy(mc, &mn);
  7926. mn.mc_xcursor = NULL;
  7927. mn.mc_pg[mn.mc_top] = rp;
  7928. mn.mc_ki[ptop] = mc->mc_ki[ptop]+1;
  7929. if (nflags & MDB_APPEND) {
  7930. mn.mc_ki[mn.mc_top] = 0;
  7931. sepkey = *newkey;
  7932. split_indx = newindx;
  7933. nkeys = 0;
  7934. } else {
  7935. split_indx = (nkeys+1) / 2;
  7936. if (IS_LEAF2(rp)) {
  7937. char *split, *ins;
  7938. int x;
  7939. unsigned int lsize, rsize, ksize;
  7940. /* Move half of the keys to the right sibling */
  7941. x = mc->mc_ki[mc->mc_top] - split_indx;
  7942. ksize = mc->mc_db->md_pad;
  7943. split = LEAF2KEY(mp, split_indx, ksize);
  7944. rsize = (nkeys - split_indx) * ksize;
  7945. lsize = (nkeys - split_indx) * sizeof(indx_t);
  7946. mp->mp_lower -= lsize;
  7947. rp->mp_lower += lsize;
  7948. mp->mp_upper += rsize - lsize;
  7949. rp->mp_upper -= rsize - lsize;
  7950. sepkey.mv_size = ksize;
  7951. if (newindx == split_indx) {
  7952. sepkey.mv_data = newkey->mv_data;
  7953. } else {
  7954. sepkey.mv_data = split;
  7955. }
  7956. if (x<0) {
  7957. ins = LEAF2KEY(mp, mc->mc_ki[mc->mc_top], ksize);
  7958. memcpy(rp->mp_ptrs, split, rsize);
  7959. sepkey.mv_data = rp->mp_ptrs;
  7960. memmove(ins+ksize, ins, (split_indx - mc->mc_ki[mc->mc_top]) * ksize);
  7961. memcpy(ins, newkey->mv_data, ksize);
  7962. mp->mp_lower += sizeof(indx_t);
  7963. mp->mp_upper -= ksize - sizeof(indx_t);
  7964. } else {
  7965. if (x)
  7966. memcpy(rp->mp_ptrs, split, x * ksize);
  7967. ins = LEAF2KEY(rp, x, ksize);
  7968. memcpy(ins, newkey->mv_data, ksize);
  7969. memcpy(ins+ksize, split + x * ksize, rsize - x * ksize);
  7970. rp->mp_lower += sizeof(indx_t);
  7971. rp->mp_upper -= ksize - sizeof(indx_t);
  7972. mc->mc_ki[mc->mc_top] = x;
  7973. }
  7974. } else {
  7975. int psize, nsize, k;
  7976. /* Maximum free space in an empty page */
  7977. pmax = env->me_psize - PAGEHDRSZ;
  7978. if (IS_LEAF(mp))
  7979. nsize = mdb_leaf_size(env, newkey, newdata);
  7980. else
  7981. nsize = mdb_branch_size(env, newkey);
  7982. nsize = EVEN(nsize);
  7983. /* grab a page to hold a temporary copy */
  7984. copy = mdb_page_malloc(mc->mc_txn, 1);
  7985. if (copy == NULL) {
  7986. rc = ENOMEM;
  7987. goto done;
  7988. }
  7989. copy->mp_pgno = mp->mp_pgno;
  7990. copy->mp_flags = mp->mp_flags;
  7991. copy->mp_lower = (PAGEHDRSZ-PAGEBASE);
  7992. copy->mp_upper = env->me_psize - PAGEBASE;
  7993. /* prepare to insert */
  7994. for (i=0, j=0; i<nkeys; i++) {
  7995. if (i == newindx) {
  7996. copy->mp_ptrs[j++] = 0;
  7997. }
  7998. copy->mp_ptrs[j++] = mp->mp_ptrs[i];
  7999. }
  8000. /* When items are relatively large the split point needs
  8001. * to be checked, because being off-by-one will make the
  8002. * difference between success or failure in mdb_node_add.
  8003. *
  8004. * It's also relevant if a page happens to be laid out
  8005. * such that one half of its nodes are all "small" and
  8006. * the other half of its nodes are "large." If the new
  8007. * item is also "large" and falls on the half with
  8008. * "large" nodes, it also may not fit.
  8009. *
  8010. * As a final tweak, if the new item goes on the last
  8011. * spot on the page (and thus, onto the new page), bias
  8012. * the split so the new page is emptier than the old page.
  8013. * This yields better packing during sequential inserts.
  8014. */
  8015. if (nkeys < 32 || nsize > pmax/16 || newindx >= nkeys) {
  8016. /* Find split point */
  8017. psize = 0;
  8018. if (newindx <= split_indx || newindx >= nkeys) {
  8019. i = 0; j = 1;
  8020. k = newindx >= nkeys ? nkeys : split_indx+1+IS_LEAF(mp);
  8021. } else {
  8022. i = nkeys; j = -1;
  8023. k = split_indx-1;
  8024. }
  8025. for (; i!=k; i+=j) {
  8026. if (i == newindx) {
  8027. psize += nsize;
  8028. node = NULL;
  8029. } else {
  8030. node = (MDB_node *)((char *)mp + copy->mp_ptrs[i] + PAGEBASE);
  8031. psize += NODESIZE + NODEKSZ(node) + sizeof(indx_t);
  8032. if (IS_LEAF(mp)) {
  8033. if (F_ISSET(node->mn_flags, F_BIGDATA))
  8034. psize += sizeof(pgno_t);
  8035. else
  8036. psize += NODEDSZ(node);
  8037. }
  8038. psize = EVEN(psize);
  8039. }
  8040. if (psize > pmax || i == k-j) {
  8041. split_indx = i + (j<0);
  8042. break;
  8043. }
  8044. }
  8045. }
  8046. if (split_indx == newindx) {
  8047. sepkey.mv_size = newkey->mv_size;
  8048. sepkey.mv_data = newkey->mv_data;
  8049. } else {
  8050. node = (MDB_node *)((char *)mp + copy->mp_ptrs[split_indx] + PAGEBASE);
  8051. sepkey.mv_size = node->mn_ksize;
  8052. sepkey.mv_data = NODEKEY(node);
  8053. }
  8054. }
  8055. }
  8056. DPRINTF(("separator is %d [%s]", split_indx, DKEY(&sepkey)));
  8057. /* Copy separator key to the parent.
  8058. */
  8059. if (SIZELEFT(mn.mc_pg[ptop]) < mdb_branch_size(env, &sepkey)) {
  8060. int snum = mc->mc_snum;
  8061. mn.mc_snum--;
  8062. mn.mc_top--;
  8063. did_split = 1;
  8064. /* We want other splits to find mn when doing fixups */
  8065. WITH_CURSOR_TRACKING(mn,
  8066. rc = mdb_page_split(&mn, &sepkey, NULL, rp->mp_pgno, 0));
  8067. if (rc)
  8068. goto done;
  8069. /* root split? */
  8070. if (mc->mc_snum > snum) {
  8071. ptop++;
  8072. }
  8073. /* Right page might now have changed parent.
  8074. * Check if left page also changed parent.
  8075. */
  8076. if (mn.mc_pg[ptop] != mc->mc_pg[ptop] &&
  8077. mc->mc_ki[ptop] >= NUMKEYS(mc->mc_pg[ptop])) {
  8078. for (i=0; i<ptop; i++) {
  8079. mc->mc_pg[i] = mn.mc_pg[i];
  8080. mc->mc_ki[i] = mn.mc_ki[i];
  8081. }
  8082. mc->mc_pg[ptop] = mn.mc_pg[ptop];
  8083. if (mn.mc_ki[ptop]) {
  8084. mc->mc_ki[ptop] = mn.mc_ki[ptop] - 1;
  8085. } else {
  8086. /* find right page's left sibling */
  8087. mc->mc_ki[ptop] = mn.mc_ki[ptop];
  8088. mdb_cursor_sibling(mc, 0);
  8089. }
  8090. }
  8091. } else {
  8092. mn.mc_top--;
  8093. rc = mdb_node_add(&mn, mn.mc_ki[ptop], &sepkey, NULL, rp->mp_pgno, 0);
  8094. mn.mc_top++;
  8095. }
  8096. if (rc != MDB_SUCCESS) {
  8097. goto done;
  8098. }
  8099. if (nflags & MDB_APPEND) {
  8100. mc->mc_pg[mc->mc_top] = rp;
  8101. mc->mc_ki[mc->mc_top] = 0;
  8102. rc = mdb_node_add(mc, 0, newkey, newdata, newpgno, nflags);
  8103. if (rc)
  8104. goto done;
  8105. for (i=0; i<mc->mc_top; i++)
  8106. mc->mc_ki[i] = mn.mc_ki[i];
  8107. } else if (!IS_LEAF2(mp)) {
  8108. /* Move nodes */
  8109. mc->mc_pg[mc->mc_top] = rp;
  8110. i = split_indx;
  8111. j = 0;
  8112. do {
  8113. if (i == newindx) {
  8114. rkey.mv_data = newkey->mv_data;
  8115. rkey.mv_size = newkey->mv_size;
  8116. if (IS_LEAF(mp)) {
  8117. rdata = newdata;
  8118. } else
  8119. pgno = newpgno;
  8120. flags = nflags;
  8121. /* Update index for the new key. */
  8122. mc->mc_ki[mc->mc_top] = j;
  8123. } else {
  8124. node = (MDB_node *)((char *)mp + copy->mp_ptrs[i] + PAGEBASE);
  8125. rkey.mv_data = NODEKEY(node);
  8126. rkey.mv_size = node->mn_ksize;
  8127. if (IS_LEAF(mp)) {
  8128. xdata.mv_data = NODEDATA(node);
  8129. xdata.mv_size = NODEDSZ(node);
  8130. rdata = &xdata;
  8131. } else
  8132. pgno = NODEPGNO(node);
  8133. flags = node->mn_flags;
  8134. }
  8135. if (!IS_LEAF(mp) && j == 0) {
  8136. /* First branch index doesn't need key data. */
  8137. rkey.mv_size = 0;
  8138. }
  8139. rc = mdb_node_add(mc, j, &rkey, rdata, pgno, flags);
  8140. if (rc)
  8141. goto done;
  8142. if (i == nkeys) {
  8143. i = 0;
  8144. j = 0;
  8145. mc->mc_pg[mc->mc_top] = copy;
  8146. } else {
  8147. i++;
  8148. j++;
  8149. }
  8150. } while (i != split_indx);
  8151. nkeys = NUMKEYS(copy);
  8152. for (i=0; i<nkeys; i++)
  8153. mp->mp_ptrs[i] = copy->mp_ptrs[i];
  8154. mp->mp_lower = copy->mp_lower;
  8155. mp->mp_upper = copy->mp_upper;
  8156. memcpy(NODEPTR(mp, nkeys-1), NODEPTR(copy, nkeys-1),
  8157. env->me_psize - copy->mp_upper - PAGEBASE);
  8158. /* reset back to original page */
  8159. if (newindx < split_indx) {
  8160. mc->mc_pg[mc->mc_top] = mp;
  8161. } else {
  8162. mc->mc_pg[mc->mc_top] = rp;
  8163. mc->mc_ki[ptop]++;
  8164. /* Make sure mc_ki is still valid.
  8165. */
  8166. if (mn.mc_pg[ptop] != mc->mc_pg[ptop] &&
  8167. mc->mc_ki[ptop] >= NUMKEYS(mc->mc_pg[ptop])) {
  8168. for (i=0; i<=ptop; i++) {
  8169. mc->mc_pg[i] = mn.mc_pg[i];
  8170. mc->mc_ki[i] = mn.mc_ki[i];
  8171. }
  8172. }
  8173. }
  8174. if (nflags & MDB_RESERVE) {
  8175. node = NODEPTR(mc->mc_pg[mc->mc_top], mc->mc_ki[mc->mc_top]);
  8176. if (!(node->mn_flags & F_BIGDATA))
  8177. newdata->mv_data = NODEDATA(node);
  8178. }
  8179. } else {
  8180. if (newindx >= split_indx) {
  8181. mc->mc_pg[mc->mc_top] = rp;
  8182. mc->mc_ki[ptop]++;
  8183. /* Make sure mc_ki is still valid.
  8184. */
  8185. if (mn.mc_pg[ptop] != mc->mc_pg[ptop] &&
  8186. mc->mc_ki[ptop] >= NUMKEYS(mc->mc_pg[ptop])) {
  8187. for (i=0; i<=ptop; i++) {
  8188. mc->mc_pg[i] = mn.mc_pg[i];
  8189. mc->mc_ki[i] = mn.mc_ki[i];
  8190. }
  8191. }
  8192. }
  8193. }
  8194. {
  8195. /* Adjust other cursors pointing to mp */
  8196. MDB_cursor *m2, *m3;
  8197. MDB_dbi dbi = mc->mc_dbi;
  8198. nkeys = NUMKEYS(mp);
  8199. for (m2 = mc->mc_txn->mt_cursors[dbi]; m2; m2=m2->mc_next) {
  8200. if (mc->mc_flags & C_SUB)
  8201. m3 = &m2->mc_xcursor->mx_cursor;
  8202. else
  8203. m3 = m2;
  8204. if (m3 == mc)
  8205. continue;
  8206. if (!(m2->mc_flags & m3->mc_flags & C_INITIALIZED))
  8207. continue;
  8208. if (new_root) {
  8209. int k;
  8210. /* sub cursors may be on different DB */
  8211. if (m3->mc_pg[0] != mp)
  8212. continue;
  8213. /* root split */
  8214. for (k=new_root; k>=0; k--) {
  8215. m3->mc_ki[k+1] = m3->mc_ki[k];
  8216. m3->mc_pg[k+1] = m3->mc_pg[k];
  8217. }
  8218. if (m3->mc_ki[0] >= nkeys) {
  8219. m3->mc_ki[0] = 1;
  8220. } else {
  8221. m3->mc_ki[0] = 0;
  8222. }
  8223. m3->mc_pg[0] = mc->mc_pg[0];
  8224. m3->mc_snum++;
  8225. m3->mc_top++;
  8226. }
  8227. if (m3->mc_top >= mc->mc_top && m3->mc_pg[mc->mc_top] == mp) {
  8228. if (m3->mc_ki[mc->mc_top] >= newindx && !(nflags & MDB_SPLIT_REPLACE))
  8229. m3->mc_ki[mc->mc_top]++;
  8230. if (m3->mc_ki[mc->mc_top] >= nkeys) {
  8231. m3->mc_pg[mc->mc_top] = rp;
  8232. m3->mc_ki[mc->mc_top] -= nkeys;
  8233. for (i=0; i<mc->mc_top; i++) {
  8234. m3->mc_ki[i] = mn.mc_ki[i];
  8235. m3->mc_pg[i] = mn.mc_pg[i];
  8236. }
  8237. }
  8238. } else if (!did_split && m3->mc_top >= ptop && m3->mc_pg[ptop] == mc->mc_pg[ptop] &&
  8239. m3->mc_ki[ptop] >= mc->mc_ki[ptop]) {
  8240. m3->mc_ki[ptop]++;
  8241. }
  8242. if (IS_LEAF(mp))
  8243. XCURSOR_REFRESH(m3, mc->mc_top, m3->mc_pg[mc->mc_top]);
  8244. }
  8245. }
  8246. DPRINTF(("mp left: %d, rp left: %d", SIZELEFT(mp), SIZELEFT(rp)));
  8247. done:
  8248. if (copy) /* tmp page */
  8249. mdb_page_free(env, copy);
  8250. if (rc)
  8251. mc->mc_txn->mt_flags |= MDB_TXN_ERROR;
  8252. return rc;
  8253. }
  8254. int
  8255. mdb_put(MDB_txn *txn, MDB_dbi dbi,
  8256. MDB_val *key, MDB_val *data, unsigned int flags)
  8257. {
  8258. MDB_cursor mc;
  8259. MDB_xcursor mx;
  8260. int rc;
  8261. if (!key || !data || !TXN_DBI_EXIST(txn, dbi, DB_USRVALID))
  8262. return EINVAL;
  8263. if (flags & ~(MDB_NOOVERWRITE|MDB_NODUPDATA|MDB_RESERVE|MDB_APPEND|MDB_APPENDDUP))
  8264. return EINVAL;
  8265. if (txn->mt_flags & (MDB_TXN_RDONLY|MDB_TXN_BLOCKED))
  8266. return (txn->mt_flags & MDB_TXN_RDONLY) ? EACCES : MDB_BAD_TXN;
  8267. mdb_cursor_init(&mc, txn, dbi, &mx);
  8268. mc.mc_next = txn->mt_cursors[dbi];
  8269. txn->mt_cursors[dbi] = &mc;
  8270. rc = mdb_cursor_put(&mc, key, data, flags);
  8271. txn->mt_cursors[dbi] = mc.mc_next;
  8272. return rc;
  8273. }
  8274. #ifndef MDB_WBUF
  8275. #define MDB_WBUF (1024*1024)
  8276. #endif
  8277. #define MDB_EOF 0x10 /**< #mdb_env_copyfd1() is done reading */
  8278. /** State needed for a double-buffering compacting copy. */
  8279. typedef struct mdb_copy {
  8280. MDB_env *mc_env;
  8281. MDB_txn *mc_txn;
  8282. pthread_mutex_t mc_mutex;
  8283. pthread_cond_t mc_cond; /**< Condition variable for #mc_new */
  8284. char *mc_wbuf[2];
  8285. char *mc_over[2];
  8286. int mc_wlen[2];
  8287. int mc_olen[2];
  8288. pgno_t mc_next_pgno;
  8289. HANDLE mc_fd;
  8290. int mc_toggle; /**< Buffer number in provider */
  8291. int mc_new; /**< (0-2 buffers to write) | (#MDB_EOF at end) */
  8292. /** Error code. Never cleared if set. Both threads can set nonzero
  8293. * to fail the copy. Not mutex-protected, LMDB expects atomic int.
  8294. */
  8295. volatile int mc_error;
  8296. } mdb_copy;
  8297. /** Dedicated writer thread for compacting copy. */
  8298. static THREAD_RET ESECT CALL_CONV
  8299. mdb_env_copythr(void *arg)
  8300. {
  8301. mdb_copy *my = arg;
  8302. char *ptr;
  8303. int toggle = 0, wsize, rc;
  8304. #ifdef _WIN32
  8305. DWORD len;
  8306. #define DO_WRITE(rc, fd, ptr, w2, len) rc = WriteFile(fd, ptr, w2, &len, NULL)
  8307. #else
  8308. int len;
  8309. #define DO_WRITE(rc, fd, ptr, w2, len) len = write(fd, ptr, w2); rc = (len >= 0)
  8310. #ifdef SIGPIPE
  8311. sigset_t set;
  8312. sigemptyset(&set);
  8313. sigaddset(&set, SIGPIPE);
  8314. if ((rc = pthread_sigmask(SIG_BLOCK, &set, NULL)) != 0)
  8315. my->mc_error = rc;
  8316. #endif
  8317. #endif
  8318. pthread_mutex_lock(&my->mc_mutex);
  8319. for(;;) {
  8320. while (!my->mc_new)
  8321. pthread_cond_wait(&my->mc_cond, &my->mc_mutex);
  8322. if (my->mc_new == 0 + MDB_EOF) /* 0 buffers, just EOF */
  8323. break;
  8324. wsize = my->mc_wlen[toggle];
  8325. ptr = my->mc_wbuf[toggle];
  8326. again:
  8327. rc = MDB_SUCCESS;
  8328. while (wsize > 0 && !my->mc_error) {
  8329. DO_WRITE(rc, my->mc_fd, ptr, wsize, len);
  8330. if (!rc) {
  8331. rc = ErrCode();
  8332. #if defined(SIGPIPE) && !defined(_WIN32)
  8333. if (rc == EPIPE) {
  8334. /* Collect the pending SIGPIPE, otherwise at least OS X
  8335. * gives it to the process on thread-exit (ITS#8504).
  8336. */
  8337. int tmp;
  8338. sigwait(&set, &tmp);
  8339. }
  8340. #endif
  8341. break;
  8342. } else if (len > 0) {
  8343. rc = MDB_SUCCESS;
  8344. ptr += len;
  8345. wsize -= len;
  8346. continue;
  8347. } else {
  8348. rc = EIO;
  8349. break;
  8350. }
  8351. }
  8352. if (rc) {
  8353. my->mc_error = rc;
  8354. }
  8355. /* If there's an overflow page tail, write it too */
  8356. if (my->mc_olen[toggle]) {
  8357. wsize = my->mc_olen[toggle];
  8358. ptr = my->mc_over[toggle];
  8359. my->mc_olen[toggle] = 0;
  8360. goto again;
  8361. }
  8362. my->mc_wlen[toggle] = 0;
  8363. toggle ^= 1;
  8364. /* Return the empty buffer to provider */
  8365. my->mc_new--;
  8366. pthread_cond_signal(&my->mc_cond);
  8367. }
  8368. pthread_mutex_unlock(&my->mc_mutex);
  8369. return (THREAD_RET)0;
  8370. #undef DO_WRITE
  8371. }
  8372. /** Give buffer and/or #MDB_EOF to writer thread, await unused buffer.
  8373. *
  8374. * @param[in] my control structure.
  8375. * @param[in] adjust (1 to hand off 1 buffer) | (MDB_EOF when ending).
  8376. */
  8377. static int ESECT
  8378. mdb_env_cthr_toggle(mdb_copy *my, int adjust)
  8379. {
  8380. pthread_mutex_lock(&my->mc_mutex);
  8381. my->mc_new += adjust;
  8382. pthread_cond_signal(&my->mc_cond);
  8383. while (my->mc_new & 2) /* both buffers in use */
  8384. pthread_cond_wait(&my->mc_cond, &my->mc_mutex);
  8385. pthread_mutex_unlock(&my->mc_mutex);
  8386. my->mc_toggle ^= (adjust & 1);
  8387. /* Both threads reset mc_wlen, to be safe from threading errors */
  8388. my->mc_wlen[my->mc_toggle] = 0;
  8389. return my->mc_error;
  8390. }
  8391. /** Depth-first tree traversal for compacting copy.
  8392. * @param[in] my control structure.
  8393. * @param[in,out] pg database root.
  8394. * @param[in] flags includes #F_DUPDATA if it is a sorted-duplicate sub-DB.
  8395. */
  8396. static int ESECT
  8397. mdb_env_cwalk(mdb_copy *my, pgno_t *pg, int flags)
  8398. {
  8399. MDB_cursor mc = {0};
  8400. MDB_node *ni;
  8401. MDB_page *mo, *mp, *leaf;
  8402. char *buf, *ptr;
  8403. int rc, toggle;
  8404. unsigned int i;
  8405. /* Empty DB, nothing to do */
  8406. if (*pg == P_INVALID)
  8407. return MDB_SUCCESS;
  8408. mc.mc_snum = 1;
  8409. mc.mc_txn = my->mc_txn;
  8410. rc = mdb_page_get(&mc, *pg, &mc.mc_pg[0], NULL);
  8411. if (rc)
  8412. return rc;
  8413. rc = mdb_page_search_root(&mc, NULL, MDB_PS_FIRST);
  8414. if (rc)
  8415. return rc;
  8416. /* Make cursor pages writable */
  8417. buf = ptr = malloc(my->mc_env->me_psize * mc.mc_snum);
  8418. if (buf == NULL)
  8419. return ENOMEM;
  8420. for (i=0; i<mc.mc_top; i++) {
  8421. mdb_page_copy((MDB_page *)ptr, mc.mc_pg[i], my->mc_env->me_psize);
  8422. mc.mc_pg[i] = (MDB_page *)ptr;
  8423. ptr += my->mc_env->me_psize;
  8424. }
  8425. /* This is writable space for a leaf page. Usually not needed. */
  8426. leaf = (MDB_page *)ptr;
  8427. toggle = my->mc_toggle;
  8428. while (mc.mc_snum > 0) {
  8429. unsigned n;
  8430. mp = mc.mc_pg[mc.mc_top];
  8431. n = NUMKEYS(mp);
  8432. if (IS_LEAF(mp)) {
  8433. if (!IS_LEAF2(mp) && !(flags & F_DUPDATA)) {
  8434. for (i=0; i<n; i++) {
  8435. ni = NODEPTR(mp, i);
  8436. if (ni->mn_flags & F_BIGDATA) {
  8437. MDB_page *omp;
  8438. pgno_t pg;
  8439. /* Need writable leaf */
  8440. if (mp != leaf) {
  8441. mc.mc_pg[mc.mc_top] = leaf;
  8442. mdb_page_copy(leaf, mp, my->mc_env->me_psize);
  8443. mp = leaf;
  8444. ni = NODEPTR(mp, i);
  8445. }
  8446. memcpy(&pg, NODEDATA(ni), sizeof(pg));
  8447. memcpy(NODEDATA(ni), &my->mc_next_pgno, sizeof(pgno_t));
  8448. rc = mdb_page_get(&mc, pg, &omp, NULL);
  8449. if (rc)
  8450. goto done;
  8451. if (my->mc_wlen[toggle] >= MDB_WBUF) {
  8452. rc = mdb_env_cthr_toggle(my, 1);
  8453. if (rc)
  8454. goto done;
  8455. toggle = my->mc_toggle;
  8456. }
  8457. mo = (MDB_page *)(my->mc_wbuf[toggle] + my->mc_wlen[toggle]);
  8458. memcpy(mo, omp, my->mc_env->me_psize);
  8459. mo->mp_pgno = my->mc_next_pgno;
  8460. my->mc_next_pgno += omp->mp_pages;
  8461. my->mc_wlen[toggle] += my->mc_env->me_psize;
  8462. if (omp->mp_pages > 1) {
  8463. my->mc_olen[toggle] = my->mc_env->me_psize * (omp->mp_pages - 1);
  8464. my->mc_over[toggle] = (char *)omp + my->mc_env->me_psize;
  8465. rc = mdb_env_cthr_toggle(my, 1);
  8466. if (rc)
  8467. goto done;
  8468. toggle = my->mc_toggle;
  8469. }
  8470. } else if (ni->mn_flags & F_SUBDATA) {
  8471. MDB_db db;
  8472. /* Need writable leaf */
  8473. if (mp != leaf) {
  8474. mc.mc_pg[mc.mc_top] = leaf;
  8475. mdb_page_copy(leaf, mp, my->mc_env->me_psize);
  8476. mp = leaf;
  8477. ni = NODEPTR(mp, i);
  8478. }
  8479. memcpy(&db, NODEDATA(ni), sizeof(db));
  8480. my->mc_toggle = toggle;
  8481. rc = mdb_env_cwalk(my, &db.md_root, ni->mn_flags & F_DUPDATA);
  8482. if (rc)
  8483. goto done;
  8484. toggle = my->mc_toggle;
  8485. memcpy(NODEDATA(ni), &db, sizeof(db));
  8486. }
  8487. }
  8488. }
  8489. } else {
  8490. mc.mc_ki[mc.mc_top]++;
  8491. if (mc.mc_ki[mc.mc_top] < n) {
  8492. pgno_t pg;
  8493. again:
  8494. ni = NODEPTR(mp, mc.mc_ki[mc.mc_top]);
  8495. pg = NODEPGNO(ni);
  8496. rc = mdb_page_get(&mc, pg, &mp, NULL);
  8497. if (rc)
  8498. goto done;
  8499. mc.mc_top++;
  8500. mc.mc_snum++;
  8501. mc.mc_ki[mc.mc_top] = 0;
  8502. if (IS_BRANCH(mp)) {
  8503. /* Whenever we advance to a sibling branch page,
  8504. * we must proceed all the way down to its first leaf.
  8505. */
  8506. mdb_page_copy(mc.mc_pg[mc.mc_top], mp, my->mc_env->me_psize);
  8507. goto again;
  8508. } else
  8509. mc.mc_pg[mc.mc_top] = mp;
  8510. continue;
  8511. }
  8512. }
  8513. if (my->mc_wlen[toggle] >= MDB_WBUF) {
  8514. rc = mdb_env_cthr_toggle(my, 1);
  8515. if (rc)
  8516. goto done;
  8517. toggle = my->mc_toggle;
  8518. }
  8519. mo = (MDB_page *)(my->mc_wbuf[toggle] + my->mc_wlen[toggle]);
  8520. mdb_page_copy(mo, mp, my->mc_env->me_psize);
  8521. mo->mp_pgno = my->mc_next_pgno++;
  8522. my->mc_wlen[toggle] += my->mc_env->me_psize;
  8523. if (mc.mc_top) {
  8524. /* Update parent if there is one */
  8525. ni = NODEPTR(mc.mc_pg[mc.mc_top-1], mc.mc_ki[mc.mc_top-1]);
  8526. SETPGNO(ni, mo->mp_pgno);
  8527. mdb_cursor_pop(&mc);
  8528. } else {
  8529. /* Otherwise we're done */
  8530. *pg = mo->mp_pgno;
  8531. break;
  8532. }
  8533. }
  8534. done:
  8535. free(buf);
  8536. return rc;
  8537. }
  8538. /** Copy environment with compaction. */
  8539. static int ESECT
  8540. mdb_env_copyfd1(MDB_env *env, HANDLE fd)
  8541. {
  8542. MDB_meta *mm;
  8543. MDB_page *mp;
  8544. mdb_copy my = {0};
  8545. MDB_txn *txn = NULL;
  8546. pthread_t thr;
  8547. pgno_t root, new_root;
  8548. int rc = MDB_SUCCESS;
  8549. #ifdef _WIN32
  8550. if (!(my.mc_mutex = CreateMutex(NULL, FALSE, NULL)) ||
  8551. !(my.mc_cond = CreateEvent(NULL, FALSE, FALSE, NULL))) {
  8552. rc = ErrCode();
  8553. goto done;
  8554. }
  8555. my.mc_wbuf[0] = _aligned_malloc(MDB_WBUF*2, env->me_os_psize);
  8556. if (my.mc_wbuf[0] == NULL) {
  8557. /* _aligned_malloc() sets errno, but we use Windows error codes */
  8558. rc = ERROR_NOT_ENOUGH_MEMORY;
  8559. goto done;
  8560. }
  8561. #else
  8562. if ((rc = pthread_mutex_init(&my.mc_mutex, NULL)) != 0)
  8563. return rc;
  8564. if ((rc = pthread_cond_init(&my.mc_cond, NULL)) != 0)
  8565. goto done2;
  8566. #ifdef HAVE_MEMALIGN
  8567. my.mc_wbuf[0] = memalign(env->me_os_psize, MDB_WBUF*2);
  8568. if (my.mc_wbuf[0] == NULL) {
  8569. rc = errno;
  8570. goto done;
  8571. }
  8572. #else
  8573. {
  8574. void *p;
  8575. if ((rc = posix_memalign(&p, env->me_os_psize, MDB_WBUF*2)) != 0)
  8576. goto done;
  8577. my.mc_wbuf[0] = p;
  8578. }
  8579. #endif
  8580. #endif
  8581. memset(my.mc_wbuf[0], 0, MDB_WBUF*2);
  8582. my.mc_wbuf[1] = my.mc_wbuf[0] + MDB_WBUF;
  8583. my.mc_next_pgno = NUM_METAS;
  8584. my.mc_env = env;
  8585. my.mc_fd = fd;
  8586. rc = THREAD_CREATE(thr, mdb_env_copythr, &my);
  8587. if (rc)
  8588. goto done;
  8589. rc = mdb_txn_begin(env, NULL, MDB_RDONLY, &txn);
  8590. if (rc)
  8591. goto finish;
  8592. mp = (MDB_page *)my.mc_wbuf[0];
  8593. memset(mp, 0, NUM_METAS * env->me_psize);
  8594. mp->mp_pgno = 0;
  8595. mp->mp_flags = P_META;
  8596. mm = (MDB_meta *)METADATA(mp);
  8597. mdb_env_init_meta0(env, mm);
  8598. mm->mm_address = env->me_metas[0]->mm_address;
  8599. mp = (MDB_page *)(my.mc_wbuf[0] + env->me_psize);
  8600. mp->mp_pgno = 1;
  8601. mp->mp_flags = P_META;
  8602. *(MDB_meta *)METADATA(mp) = *mm;
  8603. mm = (MDB_meta *)METADATA(mp);
  8604. /* Set metapage 1 with current main DB */
  8605. root = new_root = txn->mt_dbs[MAIN_DBI].md_root;
  8606. if (root != P_INVALID) {
  8607. /* Count free pages + freeDB pages. Subtract from last_pg
  8608. * to find the new last_pg, which also becomes the new root.
  8609. */
  8610. MDB_ID freecount = 0;
  8611. MDB_cursor mc;
  8612. MDB_val key, data;
  8613. mdb_cursor_init(&mc, txn, FREE_DBI, NULL);
  8614. while ((rc = mdb_cursor_get(&mc, &key, &data, MDB_NEXT)) == 0)
  8615. freecount += *(MDB_ID *)data.mv_data;
  8616. if (rc != MDB_NOTFOUND)
  8617. goto finish;
  8618. freecount += txn->mt_dbs[FREE_DBI].md_branch_pages +
  8619. txn->mt_dbs[FREE_DBI].md_leaf_pages +
  8620. txn->mt_dbs[FREE_DBI].md_overflow_pages;
  8621. new_root = txn->mt_next_pgno - 1 - freecount;
  8622. mm->mm_last_pg = new_root;
  8623. mm->mm_dbs[MAIN_DBI] = txn->mt_dbs[MAIN_DBI];
  8624. mm->mm_dbs[MAIN_DBI].md_root = new_root;
  8625. } else {
  8626. /* When the DB is empty, handle it specially to
  8627. * fix any breakage like page leaks from ITS#8174.
  8628. */
  8629. mm->mm_dbs[MAIN_DBI].md_flags = txn->mt_dbs[MAIN_DBI].md_flags;
  8630. }
  8631. if (root != P_INVALID || mm->mm_dbs[MAIN_DBI].md_flags) {
  8632. mm->mm_txnid = 1; /* use metapage 1 */
  8633. }
  8634. my.mc_wlen[0] = env->me_psize * NUM_METAS;
  8635. my.mc_txn = txn;
  8636. rc = mdb_env_cwalk(&my, &root, 0);
  8637. if (rc == MDB_SUCCESS && root != new_root) {
  8638. rc = MDB_INCOMPATIBLE; /* page leak or corrupt DB */
  8639. }
  8640. finish:
  8641. if (rc)
  8642. my.mc_error = rc;
  8643. mdb_env_cthr_toggle(&my, 1 | MDB_EOF);
  8644. rc = THREAD_FINISH(thr);
  8645. mdb_txn_abort(txn);
  8646. done:
  8647. #ifdef _WIN32
  8648. if (my.mc_wbuf[0]) _aligned_free(my.mc_wbuf[0]);
  8649. if (my.mc_cond) CloseHandle(my.mc_cond);
  8650. if (my.mc_mutex) CloseHandle(my.mc_mutex);
  8651. #else
  8652. free(my.mc_wbuf[0]);
  8653. pthread_cond_destroy(&my.mc_cond);
  8654. done2:
  8655. pthread_mutex_destroy(&my.mc_mutex);
  8656. #endif
  8657. return rc ? rc : my.mc_error;
  8658. }
  8659. /** Copy environment as-is. */
  8660. static int ESECT
  8661. mdb_env_copyfd0(MDB_env *env, HANDLE fd)
  8662. {
  8663. MDB_txn *txn = NULL;
  8664. mdb_mutexref_t wmutex = NULL;
  8665. int rc;
  8666. size_t wsize, w3;
  8667. char *ptr;
  8668. #ifdef _WIN32
  8669. DWORD len, w2;
  8670. #define DO_WRITE(rc, fd, ptr, w2, len) rc = WriteFile(fd, ptr, w2, &len, NULL)
  8671. #else
  8672. ssize_t len;
  8673. size_t w2;
  8674. #define DO_WRITE(rc, fd, ptr, w2, len) len = write(fd, ptr, w2); rc = (len >= 0)
  8675. #endif
  8676. /* Do the lock/unlock of the reader mutex before starting the
  8677. * write txn. Otherwise other read txns could block writers.
  8678. */
  8679. rc = mdb_txn_begin(env, NULL, MDB_RDONLY, &txn);
  8680. if (rc)
  8681. return rc;
  8682. if (env->me_txns) {
  8683. /* We must start the actual read txn after blocking writers */
  8684. mdb_txn_end(txn, MDB_END_RESET_TMP);
  8685. /* Temporarily block writers until we snapshot the meta pages */
  8686. wmutex = env->me_wmutex;
  8687. if (LOCK_MUTEX(rc, env, wmutex))
  8688. goto leave;
  8689. rc = mdb_txn_renew0(txn);
  8690. if (rc) {
  8691. UNLOCK_MUTEX(wmutex);
  8692. goto leave;
  8693. }
  8694. }
  8695. wsize = env->me_psize * NUM_METAS;
  8696. ptr = env->me_map;
  8697. w2 = wsize;
  8698. while (w2 > 0) {
  8699. DO_WRITE(rc, fd, ptr, w2, len);
  8700. if (!rc) {
  8701. rc = ErrCode();
  8702. break;
  8703. } else if (len > 0) {
  8704. rc = MDB_SUCCESS;
  8705. ptr += len;
  8706. w2 -= len;
  8707. continue;
  8708. } else {
  8709. /* Non-blocking or async handles are not supported */
  8710. rc = EIO;
  8711. break;
  8712. }
  8713. }
  8714. if (wmutex)
  8715. UNLOCK_MUTEX(wmutex);
  8716. if (rc)
  8717. goto leave;
  8718. w3 = txn->mt_next_pgno * env->me_psize;
  8719. {
  8720. size_t fsize = 0;
  8721. if ((rc = mdb_fsize(env->me_fd, &fsize)))
  8722. goto leave;
  8723. if (w3 > fsize)
  8724. w3 = fsize;
  8725. }
  8726. wsize = w3 - wsize;
  8727. while (wsize > 0) {
  8728. if (wsize > MAX_WRITE)
  8729. w2 = MAX_WRITE;
  8730. else
  8731. w2 = wsize;
  8732. DO_WRITE(rc, fd, ptr, w2, len);
  8733. if (!rc) {
  8734. rc = ErrCode();
  8735. break;
  8736. } else if (len > 0) {
  8737. rc = MDB_SUCCESS;
  8738. ptr += len;
  8739. wsize -= len;
  8740. continue;
  8741. } else {
  8742. rc = EIO;
  8743. break;
  8744. }
  8745. }
  8746. leave:
  8747. mdb_txn_abort(txn);
  8748. return rc;
  8749. }
  8750. int ESECT
  8751. mdb_env_copyfd2(MDB_env *env, HANDLE fd, unsigned int flags)
  8752. {
  8753. if (flags & MDB_CP_COMPACT)
  8754. return mdb_env_copyfd1(env, fd);
  8755. else
  8756. return mdb_env_copyfd0(env, fd);
  8757. }
  8758. int ESECT
  8759. mdb_env_copyfd(MDB_env *env, HANDLE fd)
  8760. {
  8761. return mdb_env_copyfd2(env, fd, 0);
  8762. }
  8763. int ESECT
  8764. mdb_env_copy2(MDB_env *env, const char *path, unsigned int flags)
  8765. {
  8766. int rc;
  8767. MDB_name fname;
  8768. HANDLE newfd = INVALID_HANDLE_VALUE;
  8769. rc = mdb_fname_init(path, env->me_flags | MDB_NOLOCK, &fname);
  8770. if (rc == MDB_SUCCESS) {
  8771. rc = mdb_fopen(env, &fname, MDB_O_COPY, 0666, &newfd);
  8772. mdb_fname_destroy(fname);
  8773. }
  8774. if (rc == MDB_SUCCESS) {
  8775. rc = mdb_env_copyfd2(env, newfd, flags);
  8776. if (close(newfd) < 0 && rc == MDB_SUCCESS)
  8777. rc = ErrCode();
  8778. }
  8779. return rc;
  8780. }
  8781. int ESECT
  8782. mdb_env_copy(MDB_env *env, const char *path)
  8783. {
  8784. return mdb_env_copy2(env, path, 0);
  8785. }
  8786. int ESECT
  8787. mdb_env_set_flags(MDB_env *env, unsigned int flag, int onoff)
  8788. {
  8789. if (flag & ~CHANGEABLE)
  8790. return EINVAL;
  8791. if (onoff)
  8792. env->me_flags |= flag;
  8793. else
  8794. env->me_flags &= ~flag;
  8795. return MDB_SUCCESS;
  8796. }
  8797. int ESECT
  8798. mdb_env_get_flags(MDB_env *env, unsigned int *arg)
  8799. {
  8800. if (!env || !arg)
  8801. return EINVAL;
  8802. *arg = env->me_flags & (CHANGEABLE|CHANGELESS);
  8803. return MDB_SUCCESS;
  8804. }
  8805. int ESECT
  8806. mdb_env_set_userctx(MDB_env *env, void *ctx)
  8807. {
  8808. if (!env)
  8809. return EINVAL;
  8810. env->me_userctx = ctx;
  8811. return MDB_SUCCESS;
  8812. }
  8813. void * ESECT
  8814. mdb_env_get_userctx(MDB_env *env)
  8815. {
  8816. return env ? env->me_userctx : NULL;
  8817. }
  8818. int ESECT
  8819. mdb_env_set_assert(MDB_env *env, MDB_assert_func *func)
  8820. {
  8821. if (!env)
  8822. return EINVAL;
  8823. #ifndef NDEBUG
  8824. env->me_assert_func = func;
  8825. #endif
  8826. return MDB_SUCCESS;
  8827. }
  8828. int ESECT
  8829. mdb_env_get_path(MDB_env *env, const char **arg)
  8830. {
  8831. if (!env || !arg)
  8832. return EINVAL;
  8833. *arg = env->me_path;
  8834. return MDB_SUCCESS;
  8835. }
  8836. int ESECT
  8837. mdb_env_get_fd(MDB_env *env, mdb_filehandle_t *arg)
  8838. {
  8839. if (!env || !arg)
  8840. return EINVAL;
  8841. *arg = env->me_fd;
  8842. return MDB_SUCCESS;
  8843. }
  8844. /** Common code for #mdb_stat() and #mdb_env_stat().
  8845. * @param[in] env the environment to operate in.
  8846. * @param[in] db the #MDB_db record containing the stats to return.
  8847. * @param[out] arg the address of an #MDB_stat structure to receive the stats.
  8848. * @return 0, this function always succeeds.
  8849. */
  8850. static int ESECT
  8851. mdb_stat0(MDB_env *env, MDB_db *db, MDB_stat *arg)
  8852. {
  8853. arg->ms_psize = env->me_psize;
  8854. arg->ms_depth = db->md_depth;
  8855. arg->ms_branch_pages = db->md_branch_pages;
  8856. arg->ms_leaf_pages = db->md_leaf_pages;
  8857. arg->ms_overflow_pages = db->md_overflow_pages;
  8858. arg->ms_entries = db->md_entries;
  8859. return MDB_SUCCESS;
  8860. }
  8861. int ESECT
  8862. mdb_env_stat(MDB_env *env, MDB_stat *arg)
  8863. {
  8864. MDB_meta *meta;
  8865. if (env == NULL || arg == NULL)
  8866. return EINVAL;
  8867. meta = mdb_env_pick_meta(env);
  8868. return mdb_stat0(env, &meta->mm_dbs[MAIN_DBI], arg);
  8869. }
  8870. int ESECT
  8871. mdb_env_info(MDB_env *env, MDB_envinfo *arg)
  8872. {
  8873. MDB_meta *meta;
  8874. if (env == NULL || arg == NULL)
  8875. return EINVAL;
  8876. meta = mdb_env_pick_meta(env);
  8877. arg->me_mapaddr = meta->mm_address;
  8878. arg->me_last_pgno = meta->mm_last_pg;
  8879. arg->me_last_txnid = meta->mm_txnid;
  8880. arg->me_mapsize = env->me_mapsize;
  8881. arg->me_maxreaders = env->me_maxreaders;
  8882. arg->me_numreaders = env->me_txns ? env->me_txns->mti_numreaders : 0;
  8883. return MDB_SUCCESS;
  8884. }
  8885. /** Set the default comparison functions for a database.
  8886. * Called immediately after a database is opened to set the defaults.
  8887. * The user can then override them with #mdb_set_compare() or
  8888. * #mdb_set_dupsort().
  8889. * @param[in] txn A transaction handle returned by #mdb_txn_begin()
  8890. * @param[in] dbi A database handle returned by #mdb_dbi_open()
  8891. */
  8892. static void
  8893. mdb_default_cmp(MDB_txn *txn, MDB_dbi dbi)
  8894. {
  8895. uint16_t f = txn->mt_dbs[dbi].md_flags;
  8896. txn->mt_dbxs[dbi].md_cmp =
  8897. (f & MDB_REVERSEKEY) ? mdb_cmp_memnr :
  8898. (f & MDB_INTEGERKEY) ? mdb_cmp_cint : mdb_cmp_memn;
  8899. txn->mt_dbxs[dbi].md_dcmp =
  8900. !(f & MDB_DUPSORT) ? 0 :
  8901. ((f & MDB_INTEGERDUP)
  8902. ? ((f & MDB_DUPFIXED) ? mdb_cmp_int : mdb_cmp_cint)
  8903. : ((f & MDB_REVERSEDUP) ? mdb_cmp_memnr : mdb_cmp_memn));
  8904. }
  8905. int mdb_dbi_open(MDB_txn *txn, const char *name, unsigned int flags, MDB_dbi *dbi)
  8906. {
  8907. MDB_val key, data;
  8908. MDB_dbi i;
  8909. MDB_cursor mc;
  8910. MDB_db dummy;
  8911. int rc, dbflag, exact;
  8912. unsigned int unused = 0, seq;
  8913. char *namedup;
  8914. size_t len;
  8915. if (flags & ~VALID_FLAGS)
  8916. return EINVAL;
  8917. if (txn->mt_flags & MDB_TXN_BLOCKED)
  8918. return MDB_BAD_TXN;
  8919. /* main DB? */
  8920. if (!name) {
  8921. *dbi = MAIN_DBI;
  8922. if (flags & PERSISTENT_FLAGS) {
  8923. uint16_t f2 = flags & PERSISTENT_FLAGS;
  8924. /* make sure flag changes get committed */
  8925. if ((txn->mt_dbs[MAIN_DBI].md_flags | f2) != txn->mt_dbs[MAIN_DBI].md_flags) {
  8926. txn->mt_dbs[MAIN_DBI].md_flags |= f2;
  8927. txn->mt_flags |= MDB_TXN_DIRTY;
  8928. }
  8929. }
  8930. mdb_default_cmp(txn, MAIN_DBI);
  8931. return MDB_SUCCESS;
  8932. }
  8933. if (txn->mt_dbxs[MAIN_DBI].md_cmp == NULL) {
  8934. mdb_default_cmp(txn, MAIN_DBI);
  8935. }
  8936. /* Is the DB already open? */
  8937. len = strlen(name);
  8938. for (i=CORE_DBS; i<txn->mt_numdbs; i++) {
  8939. if (!txn->mt_dbxs[i].md_name.mv_size) {
  8940. /* Remember this free slot */
  8941. if (!unused) unused = i;
  8942. continue;
  8943. }
  8944. if (len == txn->mt_dbxs[i].md_name.mv_size &&
  8945. !strncmp(name, txn->mt_dbxs[i].md_name.mv_data, len)) {
  8946. *dbi = i;
  8947. return MDB_SUCCESS;
  8948. }
  8949. }
  8950. /* If no free slot and max hit, fail */
  8951. if (!unused && txn->mt_numdbs >= txn->mt_env->me_maxdbs)
  8952. return MDB_DBS_FULL;
  8953. /* Cannot mix named databases with some mainDB flags */
  8954. if (txn->mt_dbs[MAIN_DBI].md_flags & (MDB_DUPSORT|MDB_INTEGERKEY))
  8955. return (flags & MDB_CREATE) ? MDB_INCOMPATIBLE : MDB_NOTFOUND;
  8956. /* Find the DB info */
  8957. dbflag = DB_NEW|DB_VALID|DB_USRVALID;
  8958. exact = 0;
  8959. key.mv_size = len;
  8960. key.mv_data = (void *)name;
  8961. mdb_cursor_init(&mc, txn, MAIN_DBI, NULL);
  8962. rc = mdb_cursor_set(&mc, &key, &data, MDB_SET, &exact);
  8963. if (rc == MDB_SUCCESS) {
  8964. /* make sure this is actually a DB */
  8965. MDB_node *node = NODEPTR(mc.mc_pg[mc.mc_top], mc.mc_ki[mc.mc_top]);
  8966. if ((node->mn_flags & (F_DUPDATA|F_SUBDATA)) != F_SUBDATA)
  8967. return MDB_INCOMPATIBLE;
  8968. } else {
  8969. if (rc != MDB_NOTFOUND || !(flags & MDB_CREATE))
  8970. return rc;
  8971. if (F_ISSET(txn->mt_flags, MDB_TXN_RDONLY))
  8972. return EACCES;
  8973. }
  8974. /* Done here so we cannot fail after creating a new DB */
  8975. if ((namedup = strdup(name)) == NULL)
  8976. return ENOMEM;
  8977. if (rc) {
  8978. /* MDB_NOTFOUND and MDB_CREATE: Create new DB */
  8979. data.mv_size = sizeof(MDB_db);
  8980. data.mv_data = &dummy;
  8981. memset(&dummy, 0, sizeof(dummy));
  8982. dummy.md_root = P_INVALID;
  8983. dummy.md_flags = flags & PERSISTENT_FLAGS;
  8984. WITH_CURSOR_TRACKING(mc,
  8985. rc = mdb_cursor_put(&mc, &key, &data, F_SUBDATA));
  8986. dbflag |= DB_DIRTY;
  8987. }
  8988. if (rc) {
  8989. free(namedup);
  8990. } else {
  8991. /* Got info, register DBI in this txn */
  8992. unsigned int slot = unused ? unused : txn->mt_numdbs;
  8993. txn->mt_dbxs[slot].md_name.mv_data = namedup;
  8994. txn->mt_dbxs[slot].md_name.mv_size = len;
  8995. txn->mt_dbxs[slot].md_rel = NULL;
  8996. txn->mt_dbflags[slot] = dbflag;
  8997. /* txn-> and env-> are the same in read txns, use
  8998. * tmp variable to avoid undefined assignment
  8999. */
  9000. seq = ++txn->mt_env->me_dbiseqs[slot];
  9001. txn->mt_dbiseqs[slot] = seq;
  9002. memcpy(&txn->mt_dbs[slot], data.mv_data, sizeof(MDB_db));
  9003. *dbi = slot;
  9004. mdb_default_cmp(txn, slot);
  9005. if (!unused) {
  9006. txn->mt_numdbs++;
  9007. }
  9008. }
  9009. return rc;
  9010. }
  9011. int ESECT
  9012. mdb_stat(MDB_txn *txn, MDB_dbi dbi, MDB_stat *arg)
  9013. {
  9014. if (!arg || !TXN_DBI_EXIST(txn, dbi, DB_VALID))
  9015. return EINVAL;
  9016. if (txn->mt_flags & MDB_TXN_BLOCKED)
  9017. return MDB_BAD_TXN;
  9018. if (txn->mt_dbflags[dbi] & DB_STALE) {
  9019. MDB_cursor mc;
  9020. MDB_xcursor mx;
  9021. /* Stale, must read the DB's root. cursor_init does it for us. */
  9022. mdb_cursor_init(&mc, txn, dbi, &mx);
  9023. }
  9024. return mdb_stat0(txn->mt_env, &txn->mt_dbs[dbi], arg);
  9025. }
  9026. void mdb_dbi_close(MDB_env *env, MDB_dbi dbi)
  9027. {
  9028. char *ptr;
  9029. if (dbi < CORE_DBS || dbi >= env->me_maxdbs)
  9030. return;
  9031. ptr = env->me_dbxs[dbi].md_name.mv_data;
  9032. /* If there was no name, this was already closed */
  9033. if (ptr) {
  9034. env->me_dbxs[dbi].md_name.mv_data = NULL;
  9035. env->me_dbxs[dbi].md_name.mv_size = 0;
  9036. env->me_dbflags[dbi] = 0;
  9037. env->me_dbiseqs[dbi]++;
  9038. free(ptr);
  9039. }
  9040. }
  9041. int mdb_dbi_flags(MDB_txn *txn, MDB_dbi dbi, unsigned int *flags)
  9042. {
  9043. /* We could return the flags for the FREE_DBI too but what's the point? */
  9044. if (!TXN_DBI_EXIST(txn, dbi, DB_USRVALID))
  9045. return EINVAL;
  9046. *flags = txn->mt_dbs[dbi].md_flags & PERSISTENT_FLAGS;
  9047. return MDB_SUCCESS;
  9048. }
  9049. /** Add all the DB's pages to the free list.
  9050. * @param[in] mc Cursor on the DB to free.
  9051. * @param[in] subs non-Zero to check for sub-DBs in this DB.
  9052. * @return 0 on success, non-zero on failure.
  9053. */
  9054. static int
  9055. mdb_drop0(MDB_cursor *mc, int subs)
  9056. {
  9057. int rc;
  9058. rc = mdb_page_search(mc, NULL, MDB_PS_FIRST);
  9059. if (rc == MDB_SUCCESS) {
  9060. MDB_txn *txn = mc->mc_txn;
  9061. MDB_node *ni;
  9062. MDB_cursor mx;
  9063. unsigned int i;
  9064. /* DUPSORT sub-DBs have no ovpages/DBs. Omit scanning leaves.
  9065. * This also avoids any P_LEAF2 pages, which have no nodes.
  9066. * Also if the DB doesn't have sub-DBs and has no overflow
  9067. * pages, omit scanning leaves.
  9068. */
  9069. if ((mc->mc_flags & C_SUB) ||
  9070. (!subs && !mc->mc_db->md_overflow_pages))
  9071. mdb_cursor_pop(mc);
  9072. mdb_cursor_copy(mc, &mx);
  9073. while (mc->mc_snum > 0) {
  9074. MDB_page *mp = mc->mc_pg[mc->mc_top];
  9075. unsigned n = NUMKEYS(mp);
  9076. if (IS_LEAF(mp)) {
  9077. for (i=0; i<n; i++) {
  9078. ni = NODEPTR(mp, i);
  9079. if (ni->mn_flags & F_BIGDATA) {
  9080. MDB_page *omp;
  9081. pgno_t pg;
  9082. memcpy(&pg, NODEDATA(ni), sizeof(pg));
  9083. rc = mdb_page_get(mc, pg, &omp, NULL);
  9084. if (rc != 0)
  9085. goto done;
  9086. mdb_cassert(mc, IS_OVERFLOW(omp));
  9087. rc = mdb_midl_append_range(&txn->mt_free_pgs,
  9088. pg, omp->mp_pages);
  9089. if (rc)
  9090. goto done;
  9091. mc->mc_db->md_overflow_pages -= omp->mp_pages;
  9092. if (!mc->mc_db->md_overflow_pages && !subs)
  9093. break;
  9094. } else if (subs && (ni->mn_flags & F_SUBDATA)) {
  9095. mdb_xcursor_init1(mc, ni);
  9096. rc = mdb_drop0(&mc->mc_xcursor->mx_cursor, 0);
  9097. if (rc)
  9098. goto done;
  9099. }
  9100. }
  9101. if (!subs && !mc->mc_db->md_overflow_pages)
  9102. goto pop;
  9103. } else {
  9104. if ((rc = mdb_midl_need(&txn->mt_free_pgs, n)) != 0)
  9105. goto done;
  9106. for (i=0; i<n; i++) {
  9107. pgno_t pg;
  9108. ni = NODEPTR(mp, i);
  9109. pg = NODEPGNO(ni);
  9110. /* free it */
  9111. mdb_midl_xappend(txn->mt_free_pgs, pg);
  9112. }
  9113. }
  9114. if (!mc->mc_top)
  9115. break;
  9116. mc->mc_ki[mc->mc_top] = i;
  9117. rc = mdb_cursor_sibling(mc, 1);
  9118. if (rc) {
  9119. if (rc != MDB_NOTFOUND)
  9120. goto done;
  9121. /* no more siblings, go back to beginning
  9122. * of previous level.
  9123. */
  9124. pop:
  9125. mdb_cursor_pop(mc);
  9126. mc->mc_ki[0] = 0;
  9127. for (i=1; i<mc->mc_snum; i++) {
  9128. mc->mc_ki[i] = 0;
  9129. mc->mc_pg[i] = mx.mc_pg[i];
  9130. }
  9131. }
  9132. }
  9133. /* free it */
  9134. rc = mdb_midl_append(&txn->mt_free_pgs, mc->mc_db->md_root);
  9135. done:
  9136. if (rc)
  9137. txn->mt_flags |= MDB_TXN_ERROR;
  9138. } else if (rc == MDB_NOTFOUND) {
  9139. rc = MDB_SUCCESS;
  9140. }
  9141. mc->mc_flags &= ~C_INITIALIZED;
  9142. return rc;
  9143. }
  9144. int mdb_drop(MDB_txn *txn, MDB_dbi dbi, int del)
  9145. {
  9146. MDB_cursor *mc, *m2;
  9147. int rc;
  9148. if ((unsigned)del > 1 || !TXN_DBI_EXIST(txn, dbi, DB_USRVALID))
  9149. return EINVAL;
  9150. if (F_ISSET(txn->mt_flags, MDB_TXN_RDONLY))
  9151. return EACCES;
  9152. if (TXN_DBI_CHANGED(txn, dbi))
  9153. return MDB_BAD_DBI;
  9154. rc = mdb_cursor_open(txn, dbi, &mc);
  9155. if (rc)
  9156. return rc;
  9157. rc = mdb_drop0(mc, mc->mc_db->md_flags & MDB_DUPSORT);
  9158. /* Invalidate the dropped DB's cursors */
  9159. for (m2 = txn->mt_cursors[dbi]; m2; m2 = m2->mc_next)
  9160. m2->mc_flags &= ~(C_INITIALIZED|C_EOF);
  9161. if (rc)
  9162. goto leave;
  9163. /* Can't delete the main DB */
  9164. if (del && dbi >= CORE_DBS) {
  9165. rc = mdb_del0(txn, MAIN_DBI, &mc->mc_dbx->md_name, NULL, F_SUBDATA);
  9166. if (!rc) {
  9167. txn->mt_dbflags[dbi] = DB_STALE;
  9168. mdb_dbi_close(txn->mt_env, dbi);
  9169. } else {
  9170. txn->mt_flags |= MDB_TXN_ERROR;
  9171. }
  9172. } else {
  9173. /* reset the DB record, mark it dirty */
  9174. txn->mt_dbflags[dbi] |= DB_DIRTY;
  9175. txn->mt_dbs[dbi].md_depth = 0;
  9176. txn->mt_dbs[dbi].md_branch_pages = 0;
  9177. txn->mt_dbs[dbi].md_leaf_pages = 0;
  9178. txn->mt_dbs[dbi].md_overflow_pages = 0;
  9179. txn->mt_dbs[dbi].md_entries = 0;
  9180. txn->mt_dbs[dbi].md_root = P_INVALID;
  9181. txn->mt_flags |= MDB_TXN_DIRTY;
  9182. }
  9183. leave:
  9184. mdb_cursor_close(mc);
  9185. return rc;
  9186. }
  9187. int mdb_set_compare(MDB_txn *txn, MDB_dbi dbi, MDB_cmp_func *cmp)
  9188. {
  9189. if (!TXN_DBI_EXIST(txn, dbi, DB_USRVALID))
  9190. return EINVAL;
  9191. txn->mt_dbxs[dbi].md_cmp = cmp;
  9192. return MDB_SUCCESS;
  9193. }
  9194. int mdb_set_dupsort(MDB_txn *txn, MDB_dbi dbi, MDB_cmp_func *cmp)
  9195. {
  9196. if (!TXN_DBI_EXIST(txn, dbi, DB_USRVALID))
  9197. return EINVAL;
  9198. txn->mt_dbxs[dbi].md_dcmp = cmp;
  9199. return MDB_SUCCESS;
  9200. }
  9201. int mdb_set_relfunc(MDB_txn *txn, MDB_dbi dbi, MDB_rel_func *rel)
  9202. {
  9203. if (!TXN_DBI_EXIST(txn, dbi, DB_USRVALID))
  9204. return EINVAL;
  9205. txn->mt_dbxs[dbi].md_rel = rel;
  9206. return MDB_SUCCESS;
  9207. }
  9208. int mdb_set_relctx(MDB_txn *txn, MDB_dbi dbi, void *ctx)
  9209. {
  9210. if (!TXN_DBI_EXIST(txn, dbi, DB_USRVALID))
  9211. return EINVAL;
  9212. txn->mt_dbxs[dbi].md_relctx = ctx;
  9213. return MDB_SUCCESS;
  9214. }
  9215. int ESECT
  9216. mdb_env_get_maxkeysize(MDB_env *env)
  9217. {
  9218. return ENV_MAXKEY(env);
  9219. }
  9220. int ESECT
  9221. mdb_reader_list(MDB_env *env, MDB_msg_func *func, void *ctx)
  9222. {
  9223. unsigned int i, rdrs;
  9224. MDB_reader *mr;
  9225. char buf[64];
  9226. int rc = 0, first = 1;
  9227. if (!env || !func)
  9228. return -1;
  9229. if (!env->me_txns) {
  9230. return func("(no reader locks)\n", ctx);
  9231. }
  9232. rdrs = env->me_txns->mti_numreaders;
  9233. mr = env->me_txns->mti_readers;
  9234. for (i=0; i<rdrs; i++) {
  9235. if (mr[i].mr_pid) {
  9236. txnid_t txnid = mr[i].mr_txnid;
  9237. sprintf(buf, txnid == (txnid_t)-1 ?
  9238. "%10d %"Z"x -\n" : "%10d %"Z"x %"Z"u\n",
  9239. (int)mr[i].mr_pid, (size_t)mr[i].mr_tid, txnid);
  9240. if (first) {
  9241. first = 0;
  9242. rc = func(" pid thread txnid\n", ctx);
  9243. if (rc < 0)
  9244. break;
  9245. }
  9246. rc = func(buf, ctx);
  9247. if (rc < 0)
  9248. break;
  9249. }
  9250. }
  9251. if (first) {
  9252. rc = func("(no active readers)\n", ctx);
  9253. }
  9254. return rc;
  9255. }
  9256. /** Insert pid into list if not already present.
  9257. * return -1 if already present.
  9258. */
  9259. static int ESECT
  9260. mdb_pid_insert(MDB_PID_T *ids, MDB_PID_T pid)
  9261. {
  9262. /* binary search of pid in list */
  9263. unsigned base = 0;
  9264. unsigned cursor = 1;
  9265. int val = 0;
  9266. unsigned n = ids[0];
  9267. while( 0 < n ) {
  9268. unsigned pivot = n >> 1;
  9269. cursor = base + pivot + 1;
  9270. val = pid - ids[cursor];
  9271. if( val < 0 ) {
  9272. n = pivot;
  9273. } else if ( val > 0 ) {
  9274. base = cursor;
  9275. n -= pivot + 1;
  9276. } else {
  9277. /* found, so it's a duplicate */
  9278. return -1;
  9279. }
  9280. }
  9281. if( val > 0 ) {
  9282. ++cursor;
  9283. }
  9284. ids[0]++;
  9285. for (n = ids[0]; n > cursor; n--)
  9286. ids[n] = ids[n-1];
  9287. ids[n] = pid;
  9288. return 0;
  9289. }
  9290. int ESECT
  9291. mdb_reader_check(MDB_env *env, int *dead)
  9292. {
  9293. if (!env)
  9294. return EINVAL;
  9295. if (dead)
  9296. *dead = 0;
  9297. return env->me_txns ? mdb_reader_check0(env, 0, dead) : MDB_SUCCESS;
  9298. }
  9299. /** As #mdb_reader_check(). \b rlocked is set if caller locked #me_rmutex. */
  9300. static int ESECT
  9301. mdb_reader_check0(MDB_env *env, int rlocked, int *dead)
  9302. {
  9303. mdb_mutexref_t rmutex = rlocked ? NULL : env->me_rmutex;
  9304. unsigned int i, j, rdrs;
  9305. MDB_reader *mr;
  9306. MDB_PID_T *pids, pid;
  9307. int rc = MDB_SUCCESS, count = 0;
  9308. rdrs = env->me_txns->mti_numreaders;
  9309. pids = malloc((rdrs+1) * sizeof(MDB_PID_T));
  9310. if (!pids)
  9311. return ENOMEM;
  9312. pids[0] = 0;
  9313. mr = env->me_txns->mti_readers;
  9314. for (i=0; i<rdrs; i++) {
  9315. pid = mr[i].mr_pid;
  9316. if (pid && pid != env->me_pid) {
  9317. if (mdb_pid_insert(pids, pid) == 0) {
  9318. if (!mdb_reader_pid(env, Pidcheck, pid)) {
  9319. /* Stale reader found */
  9320. j = i;
  9321. if (rmutex) {
  9322. if ((rc = LOCK_MUTEX0(rmutex)) != 0) {
  9323. if ((rc = mdb_mutex_failed(env, rmutex, rc)))
  9324. break;
  9325. rdrs = 0; /* the above checked all readers */
  9326. } else {
  9327. /* Recheck, a new process may have reused pid */
  9328. if (mdb_reader_pid(env, Pidcheck, pid))
  9329. j = rdrs;
  9330. }
  9331. }
  9332. for (; j<rdrs; j++)
  9333. if (mr[j].mr_pid == pid) {
  9334. DPRINTF(("clear stale reader pid %u txn %"Z"d",
  9335. (unsigned) pid, mr[j].mr_txnid));
  9336. mr[j].mr_pid = 0;
  9337. count++;
  9338. }
  9339. if (rmutex)
  9340. UNLOCK_MUTEX(rmutex);
  9341. }
  9342. }
  9343. }
  9344. }
  9345. free(pids);
  9346. if (dead)
  9347. *dead = count;
  9348. return rc;
  9349. }
  9350. #ifdef MDB_ROBUST_SUPPORTED
  9351. /** Handle #LOCK_MUTEX0() failure.
  9352. * Try to repair the lock file if the mutex owner died.
  9353. * @param[in] env the environment handle
  9354. * @param[in] mutex LOCK_MUTEX0() mutex
  9355. * @param[in] rc LOCK_MUTEX0() error (nonzero)
  9356. * @return 0 on success with the mutex locked, or an error code on failure.
  9357. */
  9358. static int ESECT
  9359. mdb_mutex_failed(MDB_env *env, mdb_mutexref_t mutex, int rc)
  9360. {
  9361. int rlocked, rc2;
  9362. MDB_meta *meta;
  9363. if (rc == MDB_OWNERDEAD) {
  9364. /* We own the mutex. Clean up after dead previous owner. */
  9365. rc = MDB_SUCCESS;
  9366. rlocked = (mutex == env->me_rmutex);
  9367. if (!rlocked) {
  9368. /* Keep mti_txnid updated, otherwise next writer can
  9369. * overwrite data which latest meta page refers to.
  9370. */
  9371. meta = mdb_env_pick_meta(env);
  9372. env->me_txns->mti_txnid = meta->mm_txnid;
  9373. /* env is hosed if the dead thread was ours */
  9374. if (env->me_txn) {
  9375. env->me_flags |= MDB_FATAL_ERROR;
  9376. env->me_txn = NULL;
  9377. rc = MDB_PANIC;
  9378. }
  9379. }
  9380. DPRINTF(("%cmutex owner died, %s", (rlocked ? 'r' : 'w'),
  9381. (rc ? "this process' env is hosed" : "recovering")));
  9382. rc2 = mdb_reader_check0(env, rlocked, NULL);
  9383. if (rc2 == 0)
  9384. rc2 = mdb_mutex_consistent(mutex);
  9385. if (rc || (rc = rc2)) {
  9386. DPRINTF(("LOCK_MUTEX recovery failed, %s", mdb_strerror(rc)));
  9387. UNLOCK_MUTEX(mutex);
  9388. }
  9389. } else {
  9390. #ifdef _WIN32
  9391. rc = ErrCode();
  9392. #endif
  9393. DPRINTF(("LOCK_MUTEX failed, %s", mdb_strerror(rc)));
  9394. }
  9395. return rc;
  9396. }
  9397. #endif /* MDB_ROBUST_SUPPORTED */
  9398. #if defined(_WIN32)
  9399. /** Convert \b src to new wchar_t[] string with room for \b xtra extra chars */
  9400. static int ESECT
  9401. utf8_to_utf16(const char *src, MDB_name *dst, int xtra)
  9402. {
  9403. int rc, need = 0;
  9404. wchar_t *result = NULL;
  9405. for (;;) { /* malloc result, then fill it in */
  9406. need = MultiByteToWideChar(CP_UTF8, 0, src, -1, result, need);
  9407. if (!need) {
  9408. rc = ErrCode();
  9409. free(result);
  9410. return rc;
  9411. }
  9412. if (!result) {
  9413. result = malloc(sizeof(wchar_t) * (need + xtra));
  9414. if (!result)
  9415. return ENOMEM;
  9416. continue;
  9417. }
  9418. dst->mn_alloced = 1;
  9419. dst->mn_len = need - 1;
  9420. dst->mn_val = result;
  9421. return MDB_SUCCESS;
  9422. }
  9423. }
  9424. #endif /* defined(_WIN32) */
  9425. /** @} */