md.c 193 KB

1234567891011121314151617181920212223242526272829303132333435363738394041424344454647484950515253545556575859606162636465666768697071727374757677787980818283848586878889909192939495969798991001011021031041051061071081091101111121131141151161171181191201211221231241251261271281291301311321331341351361371381391401411421431441451461471481491501511521531541551561571581591601611621631641651661671681691701711721731741751761771781791801811821831841851861871881891901911921931941951961971981992002012022032042052062072082092102112122132142152162172182192202212222232242252262272282292302312322332342352362372382392402412422432442452462472482492502512522532542552562572582592602612622632642652662672682692702712722732742752762772782792802812822832842852862872882892902912922932942952962972982993003013023033043053063073083093103113123133143153163173183193203213223233243253263273283293303313323333343353363373383393403413423433443453463473483493503513523533543553563573583593603613623633643653663673683693703713723733743753763773783793803813823833843853863873883893903913923933943953963973983994004014024034044054064074084094104114124134144154164174184194204214224234244254264274284294304314324334344354364374384394404414424434444454464474484494504514524534544554564574584594604614624634644654664674684694704714724734744754764774784794804814824834844854864874884894904914924934944954964974984995005015025035045055065075085095105115125135145155165175185195205215225235245255265275285295305315325335345355365375385395405415425435445455465475485495505515525535545555565575585595605615625635645655665675685695705715725735745755765775785795805815825835845855865875885895905915925935945955965975985996006016026036046056066076086096106116126136146156166176186196206216226236246256266276286296306316326336346356366376386396406416426436446456466476486496506516526536546556566576586596606616626636646656666676686696706716726736746756766776786796806816826836846856866876886896906916926936946956966976986997007017027037047057067077087097107117127137147157167177187197207217227237247257267277287297307317327337347357367377387397407417427437447457467477487497507517527537547557567577587597607617627637647657667677687697707717727737747757767777787797807817827837847857867877887897907917927937947957967977987998008018028038048058068078088098108118128138148158168178188198208218228238248258268278288298308318328338348358368378388398408418428438448458468478488498508518528538548558568578588598608618628638648658668678688698708718728738748758768778788798808818828838848858868878888898908918928938948958968978988999009019029039049059069079089099109119129139149159169179189199209219229239249259269279289299309319329339349359369379389399409419429439449459469479489499509519529539549559569579589599609619629639649659669679689699709719729739749759769779789799809819829839849859869879889899909919929939949959969979989991000100110021003100410051006100710081009101010111012101310141015101610171018101910201021102210231024102510261027102810291030103110321033103410351036103710381039104010411042104310441045104610471048104910501051105210531054105510561057105810591060106110621063106410651066106710681069107010711072107310741075107610771078107910801081108210831084108510861087108810891090109110921093109410951096109710981099110011011102110311041105110611071108110911101111111211131114111511161117111811191120112111221123112411251126112711281129113011311132113311341135113611371138113911401141114211431144114511461147114811491150115111521153115411551156115711581159116011611162116311641165116611671168116911701171117211731174117511761177117811791180118111821183118411851186118711881189119011911192119311941195119611971198119912001201120212031204120512061207120812091210121112121213121412151216121712181219122012211222122312241225122612271228122912301231123212331234123512361237123812391240124112421243124412451246124712481249125012511252125312541255125612571258125912601261126212631264126512661267126812691270127112721273127412751276127712781279128012811282128312841285128612871288128912901291129212931294129512961297129812991300130113021303130413051306130713081309131013111312131313141315131613171318131913201321132213231324132513261327132813291330133113321333133413351336133713381339134013411342134313441345134613471348134913501351135213531354135513561357135813591360136113621363136413651366136713681369137013711372137313741375137613771378137913801381138213831384138513861387138813891390139113921393139413951396139713981399140014011402140314041405140614071408140914101411141214131414141514161417141814191420142114221423142414251426142714281429143014311432143314341435143614371438143914401441144214431444144514461447144814491450145114521453145414551456145714581459146014611462146314641465146614671468146914701471147214731474147514761477147814791480148114821483148414851486148714881489149014911492149314941495149614971498149915001501150215031504150515061507150815091510151115121513151415151516151715181519152015211522152315241525152615271528152915301531153215331534153515361537153815391540154115421543154415451546154715481549155015511552155315541555155615571558155915601561156215631564156515661567156815691570157115721573157415751576157715781579158015811582158315841585158615871588158915901591159215931594159515961597159815991600160116021603160416051606160716081609161016111612161316141615161616171618161916201621162216231624162516261627162816291630163116321633163416351636163716381639164016411642164316441645164616471648164916501651165216531654165516561657165816591660166116621663166416651666166716681669167016711672167316741675167616771678167916801681168216831684168516861687168816891690169116921693169416951696169716981699170017011702170317041705170617071708170917101711171217131714171517161717171817191720172117221723172417251726172717281729173017311732173317341735173617371738173917401741174217431744174517461747174817491750175117521753175417551756175717581759176017611762176317641765176617671768176917701771177217731774177517761777177817791780178117821783178417851786178717881789179017911792179317941795179617971798179918001801180218031804180518061807180818091810181118121813181418151816181718181819182018211822182318241825182618271828182918301831183218331834183518361837183818391840184118421843184418451846184718481849185018511852185318541855185618571858185918601861186218631864186518661867186818691870187118721873187418751876187718781879188018811882188318841885188618871888188918901891189218931894189518961897189818991900190119021903190419051906190719081909191019111912191319141915191619171918191919201921192219231924192519261927192819291930193119321933193419351936193719381939194019411942194319441945194619471948194919501951195219531954195519561957195819591960196119621963196419651966196719681969197019711972197319741975197619771978197919801981198219831984198519861987198819891990199119921993199419951996199719981999200020012002200320042005200620072008200920102011201220132014201520162017201820192020202120222023202420252026202720282029203020312032203320342035203620372038203920402041204220432044204520462047204820492050205120522053205420552056205720582059206020612062206320642065206620672068206920702071207220732074207520762077207820792080208120822083208420852086208720882089209020912092209320942095209620972098209921002101210221032104210521062107210821092110211121122113211421152116211721182119212021212122212321242125212621272128212921302131213221332134213521362137213821392140214121422143214421452146214721482149215021512152215321542155215621572158215921602161216221632164216521662167216821692170217121722173217421752176217721782179218021812182218321842185218621872188218921902191219221932194219521962197219821992200220122022203220422052206220722082209221022112212221322142215221622172218221922202221222222232224222522262227222822292230223122322233223422352236223722382239224022412242224322442245224622472248224922502251225222532254225522562257225822592260226122622263226422652266226722682269227022712272227322742275227622772278227922802281228222832284228522862287228822892290229122922293229422952296229722982299230023012302230323042305230623072308230923102311231223132314231523162317231823192320232123222323232423252326232723282329233023312332233323342335233623372338233923402341234223432344234523462347234823492350235123522353235423552356235723582359236023612362236323642365236623672368236923702371237223732374237523762377237823792380238123822383238423852386238723882389239023912392239323942395239623972398239924002401240224032404240524062407240824092410241124122413241424152416241724182419242024212422242324242425242624272428242924302431243224332434243524362437243824392440244124422443244424452446244724482449245024512452245324542455245624572458245924602461246224632464246524662467246824692470247124722473247424752476247724782479248024812482248324842485248624872488248924902491249224932494249524962497249824992500250125022503250425052506250725082509251025112512251325142515251625172518251925202521252225232524252525262527252825292530253125322533253425352536253725382539254025412542254325442545254625472548254925502551255225532554255525562557255825592560256125622563256425652566256725682569257025712572257325742575257625772578257925802581258225832584258525862587258825892590259125922593259425952596259725982599260026012602260326042605260626072608260926102611261226132614261526162617261826192620262126222623262426252626262726282629263026312632263326342635263626372638263926402641264226432644264526462647264826492650265126522653265426552656265726582659266026612662266326642665266626672668266926702671267226732674267526762677267826792680268126822683268426852686268726882689269026912692269326942695269626972698269927002701270227032704270527062707270827092710271127122713271427152716271727182719272027212722272327242725272627272728272927302731273227332734273527362737273827392740274127422743274427452746274727482749275027512752275327542755275627572758275927602761276227632764276527662767276827692770277127722773277427752776277727782779278027812782278327842785278627872788278927902791279227932794279527962797279827992800280128022803280428052806280728082809281028112812281328142815281628172818281928202821282228232824282528262827282828292830283128322833283428352836283728382839284028412842284328442845284628472848284928502851285228532854285528562857285828592860286128622863286428652866286728682869287028712872287328742875287628772878287928802881288228832884288528862887288828892890289128922893289428952896289728982899290029012902290329042905290629072908290929102911291229132914291529162917291829192920292129222923292429252926292729282929293029312932293329342935293629372938293929402941294229432944294529462947294829492950295129522953295429552956295729582959296029612962296329642965296629672968296929702971297229732974297529762977297829792980298129822983298429852986298729882989299029912992299329942995299629972998299930003001300230033004300530063007300830093010301130123013301430153016301730183019302030213022302330243025302630273028302930303031303230333034303530363037303830393040304130423043304430453046304730483049305030513052305330543055305630573058305930603061306230633064306530663067306830693070307130723073307430753076307730783079308030813082308330843085308630873088308930903091309230933094309530963097309830993100310131023103310431053106310731083109311031113112311331143115311631173118311931203121312231233124312531263127312831293130313131323133313431353136313731383139314031413142314331443145314631473148314931503151315231533154315531563157315831593160316131623163316431653166316731683169317031713172317331743175317631773178317931803181318231833184318531863187318831893190319131923193319431953196319731983199320032013202320332043205320632073208320932103211321232133214321532163217321832193220322132223223322432253226322732283229323032313232323332343235323632373238323932403241324232433244324532463247324832493250325132523253325432553256325732583259326032613262326332643265326632673268326932703271327232733274327532763277327832793280328132823283328432853286328732883289329032913292329332943295329632973298329933003301330233033304330533063307330833093310331133123313331433153316331733183319332033213322332333243325332633273328332933303331333233333334333533363337333833393340334133423343334433453346334733483349335033513352335333543355335633573358335933603361336233633364336533663367336833693370337133723373337433753376337733783379338033813382338333843385338633873388338933903391339233933394339533963397339833993400340134023403340434053406340734083409341034113412341334143415341634173418341934203421342234233424342534263427342834293430343134323433343434353436343734383439344034413442344334443445344634473448344934503451345234533454345534563457345834593460346134623463346434653466346734683469347034713472347334743475347634773478347934803481348234833484348534863487348834893490349134923493349434953496349734983499350035013502350335043505350635073508350935103511351235133514351535163517351835193520352135223523352435253526352735283529353035313532353335343535353635373538353935403541354235433544354535463547354835493550355135523553355435553556355735583559356035613562356335643565356635673568356935703571357235733574357535763577357835793580358135823583358435853586358735883589359035913592359335943595359635973598359936003601360236033604360536063607360836093610361136123613361436153616361736183619362036213622362336243625362636273628362936303631363236333634363536363637363836393640364136423643364436453646364736483649365036513652365336543655365636573658365936603661366236633664366536663667366836693670367136723673367436753676367736783679368036813682368336843685368636873688368936903691369236933694369536963697369836993700370137023703370437053706370737083709371037113712371337143715371637173718371937203721372237233724372537263727372837293730373137323733373437353736373737383739374037413742374337443745374637473748374937503751375237533754375537563757375837593760376137623763376437653766376737683769377037713772377337743775377637773778377937803781378237833784378537863787378837893790379137923793379437953796379737983799380038013802380338043805380638073808380938103811381238133814381538163817381838193820382138223823382438253826382738283829383038313832383338343835383638373838383938403841384238433844384538463847384838493850385138523853385438553856385738583859386038613862386338643865386638673868386938703871387238733874387538763877387838793880388138823883388438853886388738883889389038913892389338943895389638973898389939003901390239033904390539063907390839093910391139123913391439153916391739183919392039213922392339243925392639273928392939303931393239333934393539363937393839393940394139423943394439453946394739483949395039513952395339543955395639573958395939603961396239633964396539663967396839693970397139723973397439753976397739783979398039813982398339843985398639873988398939903991399239933994399539963997399839994000400140024003400440054006400740084009401040114012401340144015401640174018401940204021402240234024402540264027402840294030403140324033403440354036403740384039404040414042404340444045404640474048404940504051405240534054405540564057405840594060406140624063406440654066406740684069407040714072407340744075407640774078407940804081408240834084408540864087408840894090409140924093409440954096409740984099410041014102410341044105410641074108410941104111411241134114411541164117411841194120412141224123412441254126412741284129413041314132413341344135413641374138413941404141414241434144414541464147414841494150415141524153415441554156415741584159416041614162416341644165416641674168416941704171417241734174417541764177417841794180418141824183418441854186418741884189419041914192419341944195419641974198419942004201420242034204420542064207420842094210421142124213421442154216421742184219422042214222422342244225422642274228422942304231423242334234423542364237423842394240424142424243424442454246424742484249425042514252425342544255425642574258425942604261426242634264426542664267426842694270427142724273427442754276427742784279428042814282428342844285428642874288428942904291429242934294429542964297429842994300430143024303430443054306430743084309431043114312431343144315431643174318431943204321432243234324432543264327432843294330433143324333433443354336433743384339434043414342434343444345434643474348434943504351435243534354435543564357435843594360436143624363436443654366436743684369437043714372437343744375437643774378437943804381438243834384438543864387438843894390439143924393439443954396439743984399440044014402440344044405440644074408440944104411441244134414441544164417441844194420442144224423442444254426442744284429443044314432443344344435443644374438443944404441444244434444444544464447444844494450445144524453445444554456445744584459446044614462446344644465446644674468446944704471447244734474447544764477447844794480448144824483448444854486448744884489449044914492449344944495449644974498449945004501450245034504450545064507450845094510451145124513451445154516451745184519452045214522452345244525452645274528452945304531453245334534453545364537453845394540454145424543454445454546454745484549455045514552455345544555455645574558455945604561456245634564456545664567456845694570457145724573457445754576457745784579458045814582458345844585458645874588458945904591459245934594459545964597459845994600460146024603460446054606460746084609461046114612461346144615461646174618461946204621462246234624462546264627462846294630463146324633463446354636463746384639464046414642464346444645464646474648464946504651465246534654465546564657465846594660466146624663466446654666466746684669467046714672467346744675467646774678467946804681468246834684468546864687468846894690469146924693469446954696469746984699470047014702470347044705470647074708470947104711471247134714471547164717471847194720472147224723472447254726472747284729473047314732473347344735473647374738473947404741474247434744474547464747474847494750475147524753475447554756475747584759476047614762476347644765476647674768476947704771477247734774477547764777477847794780478147824783478447854786478747884789479047914792479347944795479647974798479948004801480248034804480548064807480848094810481148124813481448154816481748184819482048214822482348244825482648274828482948304831483248334834483548364837483848394840484148424843484448454846484748484849485048514852485348544855485648574858485948604861486248634864486548664867486848694870487148724873487448754876487748784879488048814882488348844885488648874888488948904891489248934894489548964897489848994900490149024903490449054906490749084909491049114912491349144915491649174918491949204921492249234924492549264927492849294930493149324933493449354936493749384939494049414942494349444945494649474948494949504951495249534954495549564957495849594960496149624963496449654966496749684969497049714972497349744975497649774978497949804981498249834984498549864987498849894990499149924993499449954996499749984999500050015002500350045005500650075008500950105011501250135014501550165017501850195020502150225023502450255026502750285029503050315032503350345035503650375038503950405041504250435044504550465047504850495050505150525053505450555056505750585059506050615062506350645065506650675068506950705071507250735074507550765077507850795080508150825083508450855086508750885089509050915092509350945095509650975098509951005101510251035104510551065107510851095110511151125113511451155116511751185119512051215122512351245125512651275128512951305131513251335134513551365137513851395140514151425143514451455146514751485149515051515152515351545155515651575158515951605161516251635164516551665167516851695170517151725173517451755176517751785179518051815182518351845185518651875188518951905191519251935194519551965197519851995200520152025203520452055206520752085209521052115212521352145215521652175218521952205221522252235224522552265227522852295230523152325233523452355236523752385239524052415242524352445245524652475248524952505251525252535254525552565257525852595260526152625263526452655266526752685269527052715272527352745275527652775278527952805281528252835284528552865287528852895290529152925293529452955296529752985299530053015302530353045305530653075308530953105311531253135314531553165317531853195320532153225323532453255326532753285329533053315332533353345335533653375338533953405341534253435344534553465347534853495350535153525353535453555356535753585359536053615362536353645365536653675368536953705371537253735374537553765377537853795380538153825383538453855386538753885389539053915392539353945395539653975398539954005401540254035404540554065407540854095410541154125413541454155416541754185419542054215422542354245425542654275428542954305431543254335434543554365437543854395440544154425443544454455446544754485449545054515452545354545455545654575458545954605461546254635464546554665467546854695470547154725473547454755476547754785479548054815482548354845485548654875488548954905491549254935494549554965497549854995500550155025503550455055506550755085509551055115512551355145515551655175518551955205521552255235524552555265527552855295530553155325533553455355536553755385539554055415542554355445545554655475548554955505551555255535554555555565557555855595560556155625563556455655566556755685569557055715572557355745575557655775578557955805581558255835584558555865587558855895590559155925593559455955596559755985599560056015602560356045605560656075608560956105611561256135614561556165617561856195620562156225623562456255626562756285629563056315632563356345635563656375638563956405641564256435644564556465647564856495650565156525653565456555656565756585659566056615662566356645665566656675668566956705671567256735674567556765677567856795680568156825683568456855686568756885689569056915692569356945695569656975698569957005701570257035704570557065707570857095710571157125713571457155716571757185719572057215722572357245725572657275728572957305731573257335734573557365737573857395740574157425743574457455746574757485749575057515752575357545755575657575758575957605761576257635764576557665767576857695770577157725773577457755776577757785779578057815782578357845785578657875788578957905791579257935794579557965797579857995800580158025803580458055806580758085809581058115812581358145815581658175818581958205821582258235824582558265827582858295830583158325833583458355836583758385839584058415842584358445845584658475848584958505851585258535854585558565857585858595860586158625863586458655866586758685869587058715872587358745875587658775878587958805881588258835884588558865887588858895890589158925893589458955896589758985899590059015902590359045905590659075908590959105911591259135914591559165917591859195920592159225923592459255926592759285929593059315932593359345935593659375938593959405941594259435944594559465947594859495950595159525953595459555956595759585959596059615962596359645965596659675968596959705971597259735974597559765977597859795980598159825983598459855986598759885989599059915992599359945995599659975998599960006001600260036004600560066007600860096010601160126013601460156016601760186019602060216022602360246025602660276028602960306031603260336034603560366037603860396040604160426043604460456046604760486049605060516052605360546055605660576058605960606061606260636064606560666067606860696070607160726073607460756076607760786079608060816082608360846085608660876088608960906091609260936094609560966097609860996100610161026103610461056106610761086109611061116112611361146115611661176118611961206121612261236124612561266127612861296130613161326133613461356136613761386139614061416142614361446145614661476148614961506151615261536154615561566157615861596160616161626163616461656166616761686169617061716172617361746175617661776178617961806181618261836184618561866187618861896190619161926193619461956196619761986199620062016202620362046205620662076208620962106211621262136214621562166217621862196220622162226223622462256226622762286229623062316232623362346235623662376238623962406241624262436244624562466247624862496250625162526253625462556256625762586259626062616262626362646265626662676268626962706271627262736274627562766277627862796280628162826283628462856286628762886289629062916292629362946295629662976298629963006301630263036304630563066307630863096310631163126313631463156316631763186319632063216322632363246325632663276328632963306331633263336334633563366337633863396340634163426343634463456346634763486349635063516352635363546355635663576358635963606361636263636364636563666367636863696370637163726373637463756376637763786379638063816382638363846385638663876388638963906391639263936394639563966397639863996400640164026403640464056406640764086409641064116412641364146415641664176418641964206421642264236424642564266427642864296430643164326433643464356436643764386439644064416442644364446445644664476448644964506451645264536454645564566457645864596460646164626463646464656466646764686469647064716472647364746475647664776478647964806481648264836484648564866487648864896490649164926493649464956496649764986499650065016502650365046505650665076508650965106511651265136514651565166517651865196520652165226523652465256526652765286529653065316532653365346535653665376538653965406541654265436544654565466547654865496550655165526553655465556556655765586559656065616562656365646565656665676568656965706571657265736574657565766577657865796580658165826583658465856586658765886589659065916592659365946595659665976598659966006601660266036604660566066607660866096610661166126613661466156616661766186619662066216622662366246625662666276628662966306631663266336634663566366637663866396640664166426643664466456646664766486649665066516652665366546655665666576658665966606661666266636664666566666667666866696670667166726673667466756676667766786679668066816682668366846685668666876688668966906691669266936694669566966697669866996700670167026703670467056706670767086709671067116712671367146715671667176718671967206721672267236724672567266727672867296730673167326733673467356736673767386739674067416742674367446745674667476748674967506751675267536754675567566757675867596760676167626763676467656766676767686769677067716772677367746775677667776778677967806781678267836784678567866787678867896790679167926793679467956796679767986799680068016802680368046805680668076808680968106811681268136814681568166817681868196820682168226823682468256826682768286829683068316832683368346835683668376838683968406841684268436844684568466847684868496850685168526853685468556856685768586859686068616862686368646865686668676868686968706871687268736874687568766877687868796880688168826883688468856886688768886889689068916892689368946895689668976898689969006901690269036904690569066907690869096910691169126913691469156916691769186919692069216922692369246925692669276928692969306931693269336934693569366937693869396940694169426943694469456946694769486949695069516952695369546955695669576958695969606961696269636964696569666967696869696970697169726973697469756976697769786979698069816982698369846985698669876988698969906991699269936994699569966997699869997000700170027003700470057006700770087009701070117012701370147015701670177018701970207021702270237024702570267027702870297030703170327033703470357036703770387039704070417042704370447045704670477048704970507051705270537054705570567057705870597060706170627063706470657066706770687069707070717072707370747075707670777078707970807081708270837084708570867087708870897090709170927093709470957096709770987099710071017102710371047105710671077108710971107111711271137114711571167117711871197120712171227123712471257126712771287129713071317132713371347135713671377138713971407141714271437144714571467147714871497150715171527153715471557156715771587159716071617162716371647165716671677168716971707171717271737174717571767177717871797180718171827183718471857186718771887189719071917192719371947195719671977198719972007201720272037204720572067207720872097210721172127213721472157216721772187219722072217222722372247225722672277228722972307231723272337234723572367237723872397240724172427243724472457246724772487249725072517252725372547255725672577258725972607261726272637264726572667267726872697270727172727273727472757276727772787279728072817282728372847285728672877288728972907291729272937294729572967297729872997300730173027303730473057306730773087309731073117312731373147315731673177318731973207321732273237324732573267327732873297330733173327333733473357336733773387339734073417342734373447345734673477348734973507351735273537354735573567357735873597360736173627363736473657366736773687369737073717372737373747375737673777378737973807381738273837384738573867387738873897390739173927393739473957396739773987399740074017402740374047405740674077408740974107411741274137414741574167417741874197420742174227423742474257426742774287429743074317432743374347435743674377438743974407441744274437444744574467447744874497450745174527453745474557456745774587459746074617462746374647465746674677468746974707471747274737474747574767477747874797480748174827483748474857486748774887489749074917492749374947495749674977498749975007501750275037504750575067507750875097510751175127513751475157516751775187519752075217522752375247525752675277528752975307531753275337534753575367537753875397540754175427543754475457546754775487549755075517552755375547555755675577558
  1. /*
  2. md.c : Multiple Devices driver for Linux
  3. Copyright (C) 1998, 1999, 2000 Ingo Molnar
  4. completely rewritten, based on the MD driver code from Marc Zyngier
  5. Changes:
  6. - RAID-1/RAID-5 extensions by Miguel de Icaza, Gadi Oxman, Ingo Molnar
  7. - RAID-6 extensions by H. Peter Anvin <hpa@zytor.com>
  8. - boot support for linear and striped mode by Harald Hoyer <HarryH@Royal.Net>
  9. - kerneld support by Boris Tobotras <boris@xtalk.msk.su>
  10. - kmod support by: Cyrus Durgin
  11. - RAID0 bugfixes: Mark Anthony Lisher <markal@iname.com>
  12. - Devfs support by Richard Gooch <rgooch@atnf.csiro.au>
  13. - lots of fixes and improvements to the RAID1/RAID5 and generic
  14. RAID code (such as request based resynchronization):
  15. Neil Brown <neilb@cse.unsw.edu.au>.
  16. - persistent bitmap code
  17. Copyright (C) 2003-2004, Paul Clements, SteelEye Technology, Inc.
  18. This program is free software; you can redistribute it and/or modify
  19. it under the terms of the GNU General Public License as published by
  20. the Free Software Foundation; either version 2, or (at your option)
  21. any later version.
  22. You should have received a copy of the GNU General Public License
  23. (for example /usr/src/linux/COPYING); if not, write to the Free
  24. Software Foundation, Inc., 675 Mass Ave, Cambridge, MA 02139, USA.
  25. */
  26. #include <linux/kthread.h>
  27. #include <linux/blkdev.h>
  28. #include <linux/sysctl.h>
  29. #include <linux/seq_file.h>
  30. #include <linux/mutex.h>
  31. #include <linux/buffer_head.h> /* for invalidate_bdev */
  32. #include <linux/poll.h>
  33. #include <linux/ctype.h>
  34. #include <linux/string.h>
  35. #include <linux/hdreg.h>
  36. #include <linux/proc_fs.h>
  37. #include <linux/random.h>
  38. #include <linux/reboot.h>
  39. #include <linux/file.h>
  40. #include <linux/compat.h>
  41. #include <linux/delay.h>
  42. #include <linux/raid/md_p.h>
  43. #include <linux/raid/md_u.h>
  44. #include <linux/slab.h>
  45. #include "md.h"
  46. #include "bitmap.h"
  47. #define DEBUG 0
  48. #define dprintk(x...) ((void)(DEBUG && printk(x)))
  49. #ifndef MODULE
  50. static void autostart_arrays(int part);
  51. #endif
  52. static LIST_HEAD(pers_list);
  53. static DEFINE_SPINLOCK(pers_lock);
  54. static void md_print_devices(void);
  55. static DECLARE_WAIT_QUEUE_HEAD(resync_wait);
  56. static struct workqueue_struct *md_wq;
  57. static struct workqueue_struct *md_misc_wq;
  58. #define MD_BUG(x...) { printk("md: bug in file %s, line %d\n", __FILE__, __LINE__); md_print_devices(); }
  59. /*
  60. * Default number of read corrections we'll attempt on an rdev
  61. * before ejecting it from the array. We divide the read error
  62. * count by 2 for every hour elapsed between read errors.
  63. */
  64. #define MD_DEFAULT_MAX_CORRECTED_READ_ERRORS 20
  65. /*
  66. * Current RAID-1,4,5 parallel reconstruction 'guaranteed speed limit'
  67. * is 1000 KB/sec, so the extra system load does not show up that much.
  68. * Increase it if you want to have more _guaranteed_ speed. Note that
  69. * the RAID driver will use the maximum available bandwidth if the IO
  70. * subsystem is idle. There is also an 'absolute maximum' reconstruction
  71. * speed limit - in case reconstruction slows down your system despite
  72. * idle IO detection.
  73. *
  74. * you can change it via /proc/sys/dev/raid/speed_limit_min and _max.
  75. * or /sys/block/mdX/md/sync_speed_{min,max}
  76. */
  77. static int sysctl_speed_limit_min = 1000;
  78. static int sysctl_speed_limit_max = 200000;
  79. static inline int speed_min(mddev_t *mddev)
  80. {
  81. return mddev->sync_speed_min ?
  82. mddev->sync_speed_min : sysctl_speed_limit_min;
  83. }
  84. static inline int speed_max(mddev_t *mddev)
  85. {
  86. return mddev->sync_speed_max ?
  87. mddev->sync_speed_max : sysctl_speed_limit_max;
  88. }
  89. static struct ctl_table_header *raid_table_header;
  90. static ctl_table raid_table[] = {
  91. {
  92. .procname = "speed_limit_min",
  93. .data = &sysctl_speed_limit_min,
  94. .maxlen = sizeof(int),
  95. .mode = S_IRUGO|S_IWUSR,
  96. .proc_handler = proc_dointvec,
  97. },
  98. {
  99. .procname = "speed_limit_max",
  100. .data = &sysctl_speed_limit_max,
  101. .maxlen = sizeof(int),
  102. .mode = S_IRUGO|S_IWUSR,
  103. .proc_handler = proc_dointvec,
  104. },
  105. { }
  106. };
  107. static ctl_table raid_dir_table[] = {
  108. {
  109. .procname = "raid",
  110. .maxlen = 0,
  111. .mode = S_IRUGO|S_IXUGO,
  112. .child = raid_table,
  113. },
  114. { }
  115. };
  116. static ctl_table raid_root_table[] = {
  117. {
  118. .procname = "dev",
  119. .maxlen = 0,
  120. .mode = 0555,
  121. .child = raid_dir_table,
  122. },
  123. { }
  124. };
  125. static const struct block_device_operations md_fops;
  126. static int start_readonly;
  127. /* bio_clone_mddev
  128. * like bio_clone, but with a local bio set
  129. */
  130. static void mddev_bio_destructor(struct bio *bio)
  131. {
  132. mddev_t *mddev, **mddevp;
  133. mddevp = (void*)bio;
  134. mddev = mddevp[-1];
  135. bio_free(bio, mddev->bio_set);
  136. }
  137. struct bio *bio_alloc_mddev(gfp_t gfp_mask, int nr_iovecs,
  138. mddev_t *mddev)
  139. {
  140. struct bio *b;
  141. mddev_t **mddevp;
  142. if (!mddev || !mddev->bio_set)
  143. return bio_alloc(gfp_mask, nr_iovecs);
  144. b = bio_alloc_bioset(gfp_mask, nr_iovecs,
  145. mddev->bio_set);
  146. if (!b)
  147. return NULL;
  148. mddevp = (void*)b;
  149. mddevp[-1] = mddev;
  150. b->bi_destructor = mddev_bio_destructor;
  151. return b;
  152. }
  153. EXPORT_SYMBOL_GPL(bio_alloc_mddev);
  154. struct bio *bio_clone_mddev(struct bio *bio, gfp_t gfp_mask,
  155. mddev_t *mddev)
  156. {
  157. struct bio *b;
  158. mddev_t **mddevp;
  159. if (!mddev || !mddev->bio_set)
  160. return bio_clone(bio, gfp_mask);
  161. b = bio_alloc_bioset(gfp_mask, bio->bi_max_vecs,
  162. mddev->bio_set);
  163. if (!b)
  164. return NULL;
  165. mddevp = (void*)b;
  166. mddevp[-1] = mddev;
  167. b->bi_destructor = mddev_bio_destructor;
  168. __bio_clone(b, bio);
  169. if (bio_integrity(bio)) {
  170. int ret;
  171. ret = bio_integrity_clone(b, bio, gfp_mask, mddev->bio_set);
  172. if (ret < 0) {
  173. bio_put(b);
  174. return NULL;
  175. }
  176. }
  177. return b;
  178. }
  179. EXPORT_SYMBOL_GPL(bio_clone_mddev);
  180. /*
  181. * We have a system wide 'event count' that is incremented
  182. * on any 'interesting' event, and readers of /proc/mdstat
  183. * can use 'poll' or 'select' to find out when the event
  184. * count increases.
  185. *
  186. * Events are:
  187. * start array, stop array, error, add device, remove device,
  188. * start build, activate spare
  189. */
  190. static DECLARE_WAIT_QUEUE_HEAD(md_event_waiters);
  191. static atomic_t md_event_count;
  192. void md_new_event(mddev_t *mddev)
  193. {
  194. atomic_inc(&md_event_count);
  195. wake_up(&md_event_waiters);
  196. }
  197. EXPORT_SYMBOL_GPL(md_new_event);
  198. /* Alternate version that can be called from interrupts
  199. * when calling sysfs_notify isn't needed.
  200. */
  201. static void md_new_event_inintr(mddev_t *mddev)
  202. {
  203. atomic_inc(&md_event_count);
  204. wake_up(&md_event_waiters);
  205. }
  206. /*
  207. * Enables to iterate over all existing md arrays
  208. * all_mddevs_lock protects this list.
  209. */
  210. static LIST_HEAD(all_mddevs);
  211. static DEFINE_SPINLOCK(all_mddevs_lock);
  212. /*
  213. * iterates through all used mddevs in the system.
  214. * We take care to grab the all_mddevs_lock whenever navigating
  215. * the list, and to always hold a refcount when unlocked.
  216. * Any code which breaks out of this loop while own
  217. * a reference to the current mddev and must mddev_put it.
  218. */
  219. #define for_each_mddev(mddev,tmp) \
  220. \
  221. for (({ spin_lock(&all_mddevs_lock); \
  222. tmp = all_mddevs.next; \
  223. mddev = NULL;}); \
  224. ({ if (tmp != &all_mddevs) \
  225. mddev_get(list_entry(tmp, mddev_t, all_mddevs));\
  226. spin_unlock(&all_mddevs_lock); \
  227. if (mddev) mddev_put(mddev); \
  228. mddev = list_entry(tmp, mddev_t, all_mddevs); \
  229. tmp != &all_mddevs;}); \
  230. ({ spin_lock(&all_mddevs_lock); \
  231. tmp = tmp->next;}) \
  232. )
  233. /* Rather than calling directly into the personality make_request function,
  234. * IO requests come here first so that we can check if the device is
  235. * being suspended pending a reconfiguration.
  236. * We hold a refcount over the call to ->make_request. By the time that
  237. * call has finished, the bio has been linked into some internal structure
  238. * and so is visible to ->quiesce(), so we don't need the refcount any more.
  239. */
  240. static int md_make_request(struct request_queue *q, struct bio *bio)
  241. {
  242. const int rw = bio_data_dir(bio);
  243. mddev_t *mddev = q->queuedata;
  244. int rv;
  245. int cpu;
  246. unsigned int sectors;
  247. if (mddev == NULL || mddev->pers == NULL
  248. || !mddev->ready) {
  249. bio_io_error(bio);
  250. return 0;
  251. }
  252. smp_rmb(); /* Ensure implications of 'active' are visible */
  253. rcu_read_lock();
  254. if (mddev->suspended) {
  255. DEFINE_WAIT(__wait);
  256. for (;;) {
  257. prepare_to_wait(&mddev->sb_wait, &__wait,
  258. TASK_UNINTERRUPTIBLE);
  259. if (!mddev->suspended)
  260. break;
  261. rcu_read_unlock();
  262. schedule();
  263. rcu_read_lock();
  264. }
  265. finish_wait(&mddev->sb_wait, &__wait);
  266. }
  267. atomic_inc(&mddev->active_io);
  268. rcu_read_unlock();
  269. /*
  270. * save the sectors now since our bio can
  271. * go away inside make_request
  272. */
  273. sectors = bio_sectors(bio);
  274. rv = mddev->pers->make_request(mddev, bio);
  275. cpu = part_stat_lock();
  276. part_stat_inc(cpu, &mddev->gendisk->part0, ios[rw]);
  277. part_stat_add(cpu, &mddev->gendisk->part0, sectors[rw], sectors);
  278. part_stat_unlock();
  279. if (atomic_dec_and_test(&mddev->active_io) && mddev->suspended)
  280. wake_up(&mddev->sb_wait);
  281. return rv;
  282. }
  283. /* mddev_suspend makes sure no new requests are submitted
  284. * to the device, and that any requests that have been submitted
  285. * are completely handled.
  286. * Once ->stop is called and completes, the module will be completely
  287. * unused.
  288. */
  289. void mddev_suspend(mddev_t *mddev)
  290. {
  291. BUG_ON(mddev->suspended);
  292. mddev->suspended = 1;
  293. synchronize_rcu();
  294. wait_event(mddev->sb_wait, atomic_read(&mddev->active_io) == 0);
  295. mddev->pers->quiesce(mddev, 1);
  296. }
  297. EXPORT_SYMBOL_GPL(mddev_suspend);
  298. void mddev_resume(mddev_t *mddev)
  299. {
  300. mddev->suspended = 0;
  301. wake_up(&mddev->sb_wait);
  302. mddev->pers->quiesce(mddev, 0);
  303. md_wakeup_thread(mddev->thread);
  304. md_wakeup_thread(mddev->sync_thread); /* possibly kick off a reshape */
  305. }
  306. EXPORT_SYMBOL_GPL(mddev_resume);
  307. int mddev_congested(mddev_t *mddev, int bits)
  308. {
  309. return mddev->suspended;
  310. }
  311. EXPORT_SYMBOL(mddev_congested);
  312. /*
  313. * Generic flush handling for md
  314. */
  315. static void md_end_flush(struct bio *bio, int err)
  316. {
  317. mdk_rdev_t *rdev = bio->bi_private;
  318. mddev_t *mddev = rdev->mddev;
  319. rdev_dec_pending(rdev, mddev);
  320. if (atomic_dec_and_test(&mddev->flush_pending)) {
  321. /* The pre-request flush has finished */
  322. queue_work(md_wq, &mddev->flush_work);
  323. }
  324. bio_put(bio);
  325. }
  326. static void md_submit_flush_data(struct work_struct *ws);
  327. static void submit_flushes(struct work_struct *ws)
  328. {
  329. mddev_t *mddev = container_of(ws, mddev_t, flush_work);
  330. mdk_rdev_t *rdev;
  331. INIT_WORK(&mddev->flush_work, md_submit_flush_data);
  332. atomic_set(&mddev->flush_pending, 1);
  333. rcu_read_lock();
  334. list_for_each_entry_rcu(rdev, &mddev->disks, same_set)
  335. if (rdev->raid_disk >= 0 &&
  336. !test_bit(Faulty, &rdev->flags)) {
  337. /* Take two references, one is dropped
  338. * when request finishes, one after
  339. * we reclaim rcu_read_lock
  340. */
  341. struct bio *bi;
  342. atomic_inc(&rdev->nr_pending);
  343. atomic_inc(&rdev->nr_pending);
  344. rcu_read_unlock();
  345. bi = bio_alloc_mddev(GFP_KERNEL, 0, mddev);
  346. bi->bi_end_io = md_end_flush;
  347. bi->bi_private = rdev;
  348. bi->bi_bdev = rdev->bdev;
  349. atomic_inc(&mddev->flush_pending);
  350. submit_bio(WRITE_FLUSH, bi);
  351. rcu_read_lock();
  352. rdev_dec_pending(rdev, mddev);
  353. }
  354. rcu_read_unlock();
  355. if (atomic_dec_and_test(&mddev->flush_pending))
  356. queue_work(md_wq, &mddev->flush_work);
  357. }
  358. static void md_submit_flush_data(struct work_struct *ws)
  359. {
  360. mddev_t *mddev = container_of(ws, mddev_t, flush_work);
  361. struct bio *bio = mddev->flush_bio;
  362. if (bio->bi_size == 0)
  363. /* an empty barrier - all done */
  364. bio_endio(bio, 0);
  365. else {
  366. bio->bi_rw &= ~REQ_FLUSH;
  367. if (mddev->pers->make_request(mddev, bio))
  368. generic_make_request(bio);
  369. }
  370. mddev->flush_bio = NULL;
  371. wake_up(&mddev->sb_wait);
  372. }
  373. void md_flush_request(mddev_t *mddev, struct bio *bio)
  374. {
  375. spin_lock_irq(&mddev->write_lock);
  376. wait_event_lock_irq(mddev->sb_wait,
  377. !mddev->flush_bio,
  378. mddev->write_lock, /*nothing*/);
  379. mddev->flush_bio = bio;
  380. spin_unlock_irq(&mddev->write_lock);
  381. INIT_WORK(&mddev->flush_work, submit_flushes);
  382. queue_work(md_wq, &mddev->flush_work);
  383. }
  384. EXPORT_SYMBOL(md_flush_request);
  385. /* Support for plugging.
  386. * This mirrors the plugging support in request_queue, but does not
  387. * require having a whole queue or request structures.
  388. * We allocate an md_plug_cb for each md device and each thread it gets
  389. * plugged on. This links tot the private plug_handle structure in the
  390. * personality data where we keep a count of the number of outstanding
  391. * plugs so other code can see if a plug is active.
  392. */
  393. struct md_plug_cb {
  394. struct blk_plug_cb cb;
  395. mddev_t *mddev;
  396. };
  397. static void plugger_unplug(struct blk_plug_cb *cb)
  398. {
  399. struct md_plug_cb *mdcb = container_of(cb, struct md_plug_cb, cb);
  400. if (atomic_dec_and_test(&mdcb->mddev->plug_cnt))
  401. md_wakeup_thread(mdcb->mddev->thread);
  402. kfree(mdcb);
  403. }
  404. /* Check that an unplug wakeup will come shortly.
  405. * If not, wakeup the md thread immediately
  406. */
  407. int mddev_check_plugged(mddev_t *mddev)
  408. {
  409. struct blk_plug *plug = current->plug;
  410. struct md_plug_cb *mdcb;
  411. if (!plug)
  412. return 0;
  413. list_for_each_entry(mdcb, &plug->cb_list, cb.list) {
  414. if (mdcb->cb.callback == plugger_unplug &&
  415. mdcb->mddev == mddev) {
  416. /* Already on the list, move to top */
  417. if (mdcb != list_first_entry(&plug->cb_list,
  418. struct md_plug_cb,
  419. cb.list))
  420. list_move(&mdcb->cb.list, &plug->cb_list);
  421. return 1;
  422. }
  423. }
  424. /* Not currently on the callback list */
  425. mdcb = kmalloc(sizeof(*mdcb), GFP_ATOMIC);
  426. if (!mdcb)
  427. return 0;
  428. mdcb->mddev = mddev;
  429. mdcb->cb.callback = plugger_unplug;
  430. atomic_inc(&mddev->plug_cnt);
  431. list_add(&mdcb->cb.list, &plug->cb_list);
  432. return 1;
  433. }
  434. EXPORT_SYMBOL_GPL(mddev_check_plugged);
  435. static inline mddev_t *mddev_get(mddev_t *mddev)
  436. {
  437. atomic_inc(&mddev->active);
  438. return mddev;
  439. }
  440. static void mddev_delayed_delete(struct work_struct *ws);
  441. static void mddev_put(mddev_t *mddev)
  442. {
  443. struct bio_set *bs = NULL;
  444. if (!atomic_dec_and_lock(&mddev->active, &all_mddevs_lock))
  445. return;
  446. if (!mddev->raid_disks && list_empty(&mddev->disks) &&
  447. mddev->ctime == 0 && !mddev->hold_active) {
  448. /* Array is not configured at all, and not held active,
  449. * so destroy it */
  450. list_del(&mddev->all_mddevs);
  451. bs = mddev->bio_set;
  452. mddev->bio_set = NULL;
  453. if (mddev->gendisk) {
  454. /* We did a probe so need to clean up. Call
  455. * queue_work inside the spinlock so that
  456. * flush_workqueue() after mddev_find will
  457. * succeed in waiting for the work to be done.
  458. */
  459. INIT_WORK(&mddev->del_work, mddev_delayed_delete);
  460. queue_work(md_misc_wq, &mddev->del_work);
  461. } else
  462. kfree(mddev);
  463. }
  464. spin_unlock(&all_mddevs_lock);
  465. if (bs)
  466. bioset_free(bs);
  467. }
  468. void mddev_init(mddev_t *mddev)
  469. {
  470. mutex_init(&mddev->open_mutex);
  471. mutex_init(&mddev->reconfig_mutex);
  472. mutex_init(&mddev->bitmap_info.mutex);
  473. INIT_LIST_HEAD(&mddev->disks);
  474. INIT_LIST_HEAD(&mddev->all_mddevs);
  475. init_timer(&mddev->safemode_timer);
  476. atomic_set(&mddev->active, 1);
  477. atomic_set(&mddev->openers, 0);
  478. atomic_set(&mddev->active_io, 0);
  479. atomic_set(&mddev->plug_cnt, 0);
  480. spin_lock_init(&mddev->write_lock);
  481. atomic_set(&mddev->flush_pending, 0);
  482. init_waitqueue_head(&mddev->sb_wait);
  483. init_waitqueue_head(&mddev->recovery_wait);
  484. mddev->reshape_position = MaxSector;
  485. mddev->resync_min = 0;
  486. mddev->resync_max = MaxSector;
  487. mddev->level = LEVEL_NONE;
  488. }
  489. EXPORT_SYMBOL_GPL(mddev_init);
  490. static mddev_t * mddev_find(dev_t unit)
  491. {
  492. mddev_t *mddev, *new = NULL;
  493. if (unit && MAJOR(unit) != MD_MAJOR)
  494. unit &= ~((1<<MdpMinorShift)-1);
  495. retry:
  496. spin_lock(&all_mddevs_lock);
  497. if (unit) {
  498. list_for_each_entry(mddev, &all_mddevs, all_mddevs)
  499. if (mddev->unit == unit) {
  500. mddev_get(mddev);
  501. spin_unlock(&all_mddevs_lock);
  502. kfree(new);
  503. return mddev;
  504. }
  505. if (new) {
  506. list_add(&new->all_mddevs, &all_mddevs);
  507. spin_unlock(&all_mddevs_lock);
  508. new->hold_active = UNTIL_IOCTL;
  509. return new;
  510. }
  511. } else if (new) {
  512. /* find an unused unit number */
  513. static int next_minor = 512;
  514. int start = next_minor;
  515. int is_free = 0;
  516. int dev = 0;
  517. while (!is_free) {
  518. dev = MKDEV(MD_MAJOR, next_minor);
  519. next_minor++;
  520. if (next_minor > MINORMASK)
  521. next_minor = 0;
  522. if (next_minor == start) {
  523. /* Oh dear, all in use. */
  524. spin_unlock(&all_mddevs_lock);
  525. kfree(new);
  526. return NULL;
  527. }
  528. is_free = 1;
  529. list_for_each_entry(mddev, &all_mddevs, all_mddevs)
  530. if (mddev->unit == dev) {
  531. is_free = 0;
  532. break;
  533. }
  534. }
  535. new->unit = dev;
  536. new->md_minor = MINOR(dev);
  537. new->hold_active = UNTIL_STOP;
  538. list_add(&new->all_mddevs, &all_mddevs);
  539. spin_unlock(&all_mddevs_lock);
  540. return new;
  541. }
  542. spin_unlock(&all_mddevs_lock);
  543. new = kzalloc(sizeof(*new), GFP_KERNEL);
  544. if (!new)
  545. return NULL;
  546. new->unit = unit;
  547. if (MAJOR(unit) == MD_MAJOR)
  548. new->md_minor = MINOR(unit);
  549. else
  550. new->md_minor = MINOR(unit) >> MdpMinorShift;
  551. mddev_init(new);
  552. goto retry;
  553. }
  554. static inline int mddev_lock(mddev_t * mddev)
  555. {
  556. return mutex_lock_interruptible(&mddev->reconfig_mutex);
  557. }
  558. static inline int mddev_is_locked(mddev_t *mddev)
  559. {
  560. return mutex_is_locked(&mddev->reconfig_mutex);
  561. }
  562. static inline int mddev_trylock(mddev_t * mddev)
  563. {
  564. return mutex_trylock(&mddev->reconfig_mutex);
  565. }
  566. static struct attribute_group md_redundancy_group;
  567. static void mddev_unlock(mddev_t * mddev)
  568. {
  569. if (mddev->to_remove) {
  570. /* These cannot be removed under reconfig_mutex as
  571. * an access to the files will try to take reconfig_mutex
  572. * while holding the file unremovable, which leads to
  573. * a deadlock.
  574. * So hold set sysfs_active while the remove in happeing,
  575. * and anything else which might set ->to_remove or my
  576. * otherwise change the sysfs namespace will fail with
  577. * -EBUSY if sysfs_active is still set.
  578. * We set sysfs_active under reconfig_mutex and elsewhere
  579. * test it under the same mutex to ensure its correct value
  580. * is seen.
  581. */
  582. struct attribute_group *to_remove = mddev->to_remove;
  583. mddev->to_remove = NULL;
  584. mddev->sysfs_active = 1;
  585. mutex_unlock(&mddev->reconfig_mutex);
  586. if (mddev->kobj.sd) {
  587. if (to_remove != &md_redundancy_group)
  588. sysfs_remove_group(&mddev->kobj, to_remove);
  589. if (mddev->pers == NULL ||
  590. mddev->pers->sync_request == NULL) {
  591. sysfs_remove_group(&mddev->kobj, &md_redundancy_group);
  592. if (mddev->sysfs_action)
  593. sysfs_put(mddev->sysfs_action);
  594. mddev->sysfs_action = NULL;
  595. }
  596. }
  597. mddev->sysfs_active = 0;
  598. } else
  599. mutex_unlock(&mddev->reconfig_mutex);
  600. md_wakeup_thread(mddev->thread);
  601. }
  602. static mdk_rdev_t * find_rdev_nr(mddev_t *mddev, int nr)
  603. {
  604. mdk_rdev_t *rdev;
  605. list_for_each_entry(rdev, &mddev->disks, same_set)
  606. if (rdev->desc_nr == nr)
  607. return rdev;
  608. return NULL;
  609. }
  610. static mdk_rdev_t * find_rdev(mddev_t * mddev, dev_t dev)
  611. {
  612. mdk_rdev_t *rdev;
  613. list_for_each_entry(rdev, &mddev->disks, same_set)
  614. if (rdev->bdev->bd_dev == dev)
  615. return rdev;
  616. return NULL;
  617. }
  618. static struct mdk_personality *find_pers(int level, char *clevel)
  619. {
  620. struct mdk_personality *pers;
  621. list_for_each_entry(pers, &pers_list, list) {
  622. if (level != LEVEL_NONE && pers->level == level)
  623. return pers;
  624. if (strcmp(pers->name, clevel)==0)
  625. return pers;
  626. }
  627. return NULL;
  628. }
  629. /* return the offset of the super block in 512byte sectors */
  630. static inline sector_t calc_dev_sboffset(mdk_rdev_t *rdev)
  631. {
  632. sector_t num_sectors = i_size_read(rdev->bdev->bd_inode) / 512;
  633. return MD_NEW_SIZE_SECTORS(num_sectors);
  634. }
  635. static int alloc_disk_sb(mdk_rdev_t * rdev)
  636. {
  637. if (rdev->sb_page)
  638. MD_BUG();
  639. rdev->sb_page = alloc_page(GFP_KERNEL);
  640. if (!rdev->sb_page) {
  641. printk(KERN_ALERT "md: out of memory.\n");
  642. return -ENOMEM;
  643. }
  644. return 0;
  645. }
  646. static void free_disk_sb(mdk_rdev_t * rdev)
  647. {
  648. if (rdev->sb_page) {
  649. put_page(rdev->sb_page);
  650. rdev->sb_loaded = 0;
  651. rdev->sb_page = NULL;
  652. rdev->sb_start = 0;
  653. rdev->sectors = 0;
  654. }
  655. }
  656. static void super_written(struct bio *bio, int error)
  657. {
  658. mdk_rdev_t *rdev = bio->bi_private;
  659. mddev_t *mddev = rdev->mddev;
  660. if (error || !test_bit(BIO_UPTODATE, &bio->bi_flags)) {
  661. printk("md: super_written gets error=%d, uptodate=%d\n",
  662. error, test_bit(BIO_UPTODATE, &bio->bi_flags));
  663. WARN_ON(test_bit(BIO_UPTODATE, &bio->bi_flags));
  664. md_error(mddev, rdev);
  665. }
  666. if (atomic_dec_and_test(&mddev->pending_writes))
  667. wake_up(&mddev->sb_wait);
  668. bio_put(bio);
  669. }
  670. void md_super_write(mddev_t *mddev, mdk_rdev_t *rdev,
  671. sector_t sector, int size, struct page *page)
  672. {
  673. /* write first size bytes of page to sector of rdev
  674. * Increment mddev->pending_writes before returning
  675. * and decrement it on completion, waking up sb_wait
  676. * if zero is reached.
  677. * If an error occurred, call md_error
  678. */
  679. struct bio *bio = bio_alloc_mddev(GFP_NOIO, 1, mddev);
  680. bio->bi_bdev = rdev->meta_bdev ? rdev->meta_bdev : rdev->bdev;
  681. bio->bi_sector = sector;
  682. bio_add_page(bio, page, size, 0);
  683. bio->bi_private = rdev;
  684. bio->bi_end_io = super_written;
  685. atomic_inc(&mddev->pending_writes);
  686. submit_bio(REQ_WRITE | REQ_SYNC | REQ_FLUSH | REQ_FUA, bio);
  687. }
  688. void md_super_wait(mddev_t *mddev)
  689. {
  690. /* wait for all superblock writes that were scheduled to complete */
  691. DEFINE_WAIT(wq);
  692. for(;;) {
  693. prepare_to_wait(&mddev->sb_wait, &wq, TASK_UNINTERRUPTIBLE);
  694. if (atomic_read(&mddev->pending_writes)==0)
  695. break;
  696. schedule();
  697. }
  698. finish_wait(&mddev->sb_wait, &wq);
  699. }
  700. static void bi_complete(struct bio *bio, int error)
  701. {
  702. complete((struct completion*)bio->bi_private);
  703. }
  704. int sync_page_io(mdk_rdev_t *rdev, sector_t sector, int size,
  705. struct page *page, int rw, bool metadata_op)
  706. {
  707. struct bio *bio = bio_alloc_mddev(GFP_NOIO, 1, rdev->mddev);
  708. struct completion event;
  709. int ret;
  710. rw |= REQ_SYNC;
  711. bio->bi_bdev = (metadata_op && rdev->meta_bdev) ?
  712. rdev->meta_bdev : rdev->bdev;
  713. if (metadata_op)
  714. bio->bi_sector = sector + rdev->sb_start;
  715. else
  716. bio->bi_sector = sector + rdev->data_offset;
  717. bio_add_page(bio, page, size, 0);
  718. init_completion(&event);
  719. bio->bi_private = &event;
  720. bio->bi_end_io = bi_complete;
  721. submit_bio(rw, bio);
  722. wait_for_completion(&event);
  723. ret = test_bit(BIO_UPTODATE, &bio->bi_flags);
  724. bio_put(bio);
  725. return ret;
  726. }
  727. EXPORT_SYMBOL_GPL(sync_page_io);
  728. static int read_disk_sb(mdk_rdev_t * rdev, int size)
  729. {
  730. char b[BDEVNAME_SIZE];
  731. if (!rdev->sb_page) {
  732. MD_BUG();
  733. return -EINVAL;
  734. }
  735. if (rdev->sb_loaded)
  736. return 0;
  737. if (!sync_page_io(rdev, 0, size, rdev->sb_page, READ, true))
  738. goto fail;
  739. rdev->sb_loaded = 1;
  740. return 0;
  741. fail:
  742. printk(KERN_WARNING "md: disabled device %s, could not read superblock.\n",
  743. bdevname(rdev->bdev,b));
  744. return -EINVAL;
  745. }
  746. static int uuid_equal(mdp_super_t *sb1, mdp_super_t *sb2)
  747. {
  748. return sb1->set_uuid0 == sb2->set_uuid0 &&
  749. sb1->set_uuid1 == sb2->set_uuid1 &&
  750. sb1->set_uuid2 == sb2->set_uuid2 &&
  751. sb1->set_uuid3 == sb2->set_uuid3;
  752. }
  753. static int sb_equal(mdp_super_t *sb1, mdp_super_t *sb2)
  754. {
  755. int ret;
  756. mdp_super_t *tmp1, *tmp2;
  757. tmp1 = kmalloc(sizeof(*tmp1),GFP_KERNEL);
  758. tmp2 = kmalloc(sizeof(*tmp2),GFP_KERNEL);
  759. if (!tmp1 || !tmp2) {
  760. ret = 0;
  761. printk(KERN_INFO "md.c sb_equal(): failed to allocate memory!\n");
  762. goto abort;
  763. }
  764. *tmp1 = *sb1;
  765. *tmp2 = *sb2;
  766. /*
  767. * nr_disks is not constant
  768. */
  769. tmp1->nr_disks = 0;
  770. tmp2->nr_disks = 0;
  771. ret = (memcmp(tmp1, tmp2, MD_SB_GENERIC_CONSTANT_WORDS * 4) == 0);
  772. abort:
  773. kfree(tmp1);
  774. kfree(tmp2);
  775. return ret;
  776. }
  777. static u32 md_csum_fold(u32 csum)
  778. {
  779. csum = (csum & 0xffff) + (csum >> 16);
  780. return (csum & 0xffff) + (csum >> 16);
  781. }
  782. static unsigned int calc_sb_csum(mdp_super_t * sb)
  783. {
  784. u64 newcsum = 0;
  785. u32 *sb32 = (u32*)sb;
  786. int i;
  787. unsigned int disk_csum, csum;
  788. disk_csum = sb->sb_csum;
  789. sb->sb_csum = 0;
  790. for (i = 0; i < MD_SB_BYTES/4 ; i++)
  791. newcsum += sb32[i];
  792. csum = (newcsum & 0xffffffff) + (newcsum>>32);
  793. #ifdef CONFIG_ALPHA
  794. /* This used to use csum_partial, which was wrong for several
  795. * reasons including that different results are returned on
  796. * different architectures. It isn't critical that we get exactly
  797. * the same return value as before (we always csum_fold before
  798. * testing, and that removes any differences). However as we
  799. * know that csum_partial always returned a 16bit value on
  800. * alphas, do a fold to maximise conformity to previous behaviour.
  801. */
  802. sb->sb_csum = md_csum_fold(disk_csum);
  803. #else
  804. sb->sb_csum = disk_csum;
  805. #endif
  806. return csum;
  807. }
  808. /*
  809. * Handle superblock details.
  810. * We want to be able to handle multiple superblock formats
  811. * so we have a common interface to them all, and an array of
  812. * different handlers.
  813. * We rely on user-space to write the initial superblock, and support
  814. * reading and updating of superblocks.
  815. * Interface methods are:
  816. * int load_super(mdk_rdev_t *dev, mdk_rdev_t *refdev, int minor_version)
  817. * loads and validates a superblock on dev.
  818. * if refdev != NULL, compare superblocks on both devices
  819. * Return:
  820. * 0 - dev has a superblock that is compatible with refdev
  821. * 1 - dev has a superblock that is compatible and newer than refdev
  822. * so dev should be used as the refdev in future
  823. * -EINVAL superblock incompatible or invalid
  824. * -othererror e.g. -EIO
  825. *
  826. * int validate_super(mddev_t *mddev, mdk_rdev_t *dev)
  827. * Verify that dev is acceptable into mddev.
  828. * The first time, mddev->raid_disks will be 0, and data from
  829. * dev should be merged in. Subsequent calls check that dev
  830. * is new enough. Return 0 or -EINVAL
  831. *
  832. * void sync_super(mddev_t *mddev, mdk_rdev_t *dev)
  833. * Update the superblock for rdev with data in mddev
  834. * This does not write to disc.
  835. *
  836. */
  837. struct super_type {
  838. char *name;
  839. struct module *owner;
  840. int (*load_super)(mdk_rdev_t *rdev, mdk_rdev_t *refdev,
  841. int minor_version);
  842. int (*validate_super)(mddev_t *mddev, mdk_rdev_t *rdev);
  843. void (*sync_super)(mddev_t *mddev, mdk_rdev_t *rdev);
  844. unsigned long long (*rdev_size_change)(mdk_rdev_t *rdev,
  845. sector_t num_sectors);
  846. };
  847. /*
  848. * Check that the given mddev has no bitmap.
  849. *
  850. * This function is called from the run method of all personalities that do not
  851. * support bitmaps. It prints an error message and returns non-zero if mddev
  852. * has a bitmap. Otherwise, it returns 0.
  853. *
  854. */
  855. int md_check_no_bitmap(mddev_t *mddev)
  856. {
  857. if (!mddev->bitmap_info.file && !mddev->bitmap_info.offset)
  858. return 0;
  859. printk(KERN_ERR "%s: bitmaps are not supported for %s\n",
  860. mdname(mddev), mddev->pers->name);
  861. return 1;
  862. }
  863. EXPORT_SYMBOL(md_check_no_bitmap);
  864. /*
  865. * load_super for 0.90.0
  866. */
  867. static int super_90_load(mdk_rdev_t *rdev, mdk_rdev_t *refdev, int minor_version)
  868. {
  869. char b[BDEVNAME_SIZE], b2[BDEVNAME_SIZE];
  870. mdp_super_t *sb;
  871. int ret;
  872. /*
  873. * Calculate the position of the superblock (512byte sectors),
  874. * it's at the end of the disk.
  875. *
  876. * It also happens to be a multiple of 4Kb.
  877. */
  878. rdev->sb_start = calc_dev_sboffset(rdev);
  879. ret = read_disk_sb(rdev, MD_SB_BYTES);
  880. if (ret) return ret;
  881. ret = -EINVAL;
  882. bdevname(rdev->bdev, b);
  883. sb = (mdp_super_t*)page_address(rdev->sb_page);
  884. if (sb->md_magic != MD_SB_MAGIC) {
  885. printk(KERN_ERR "md: invalid raid superblock magic on %s\n",
  886. b);
  887. goto abort;
  888. }
  889. if (sb->major_version != 0 ||
  890. sb->minor_version < 90 ||
  891. sb->minor_version > 91) {
  892. printk(KERN_WARNING "Bad version number %d.%d on %s\n",
  893. sb->major_version, sb->minor_version,
  894. b);
  895. goto abort;
  896. }
  897. if (sb->raid_disks <= 0)
  898. goto abort;
  899. if (md_csum_fold(calc_sb_csum(sb)) != md_csum_fold(sb->sb_csum)) {
  900. printk(KERN_WARNING "md: invalid superblock checksum on %s\n",
  901. b);
  902. goto abort;
  903. }
  904. rdev->preferred_minor = sb->md_minor;
  905. rdev->data_offset = 0;
  906. rdev->sb_size = MD_SB_BYTES;
  907. if (sb->level == LEVEL_MULTIPATH)
  908. rdev->desc_nr = -1;
  909. else
  910. rdev->desc_nr = sb->this_disk.number;
  911. if (!refdev) {
  912. ret = 1;
  913. } else {
  914. __u64 ev1, ev2;
  915. mdp_super_t *refsb = (mdp_super_t*)page_address(refdev->sb_page);
  916. if (!uuid_equal(refsb, sb)) {
  917. printk(KERN_WARNING "md: %s has different UUID to %s\n",
  918. b, bdevname(refdev->bdev,b2));
  919. goto abort;
  920. }
  921. if (!sb_equal(refsb, sb)) {
  922. printk(KERN_WARNING "md: %s has same UUID"
  923. " but different superblock to %s\n",
  924. b, bdevname(refdev->bdev, b2));
  925. goto abort;
  926. }
  927. ev1 = md_event(sb);
  928. ev2 = md_event(refsb);
  929. if (ev1 > ev2)
  930. ret = 1;
  931. else
  932. ret = 0;
  933. }
  934. rdev->sectors = rdev->sb_start;
  935. if (rdev->sectors < sb->size * 2 && sb->level > 1)
  936. /* "this cannot possibly happen" ... */
  937. ret = -EINVAL;
  938. abort:
  939. return ret;
  940. }
  941. /*
  942. * validate_super for 0.90.0
  943. */
  944. static int super_90_validate(mddev_t *mddev, mdk_rdev_t *rdev)
  945. {
  946. mdp_disk_t *desc;
  947. mdp_super_t *sb = (mdp_super_t *)page_address(rdev->sb_page);
  948. __u64 ev1 = md_event(sb);
  949. rdev->raid_disk = -1;
  950. clear_bit(Faulty, &rdev->flags);
  951. clear_bit(In_sync, &rdev->flags);
  952. clear_bit(WriteMostly, &rdev->flags);
  953. if (mddev->raid_disks == 0) {
  954. mddev->major_version = 0;
  955. mddev->minor_version = sb->minor_version;
  956. mddev->patch_version = sb->patch_version;
  957. mddev->external = 0;
  958. mddev->chunk_sectors = sb->chunk_size >> 9;
  959. mddev->ctime = sb->ctime;
  960. mddev->utime = sb->utime;
  961. mddev->level = sb->level;
  962. mddev->clevel[0] = 0;
  963. mddev->layout = sb->layout;
  964. mddev->raid_disks = sb->raid_disks;
  965. mddev->dev_sectors = sb->size * 2;
  966. mddev->events = ev1;
  967. mddev->bitmap_info.offset = 0;
  968. mddev->bitmap_info.default_offset = MD_SB_BYTES >> 9;
  969. if (mddev->minor_version >= 91) {
  970. mddev->reshape_position = sb->reshape_position;
  971. mddev->delta_disks = sb->delta_disks;
  972. mddev->new_level = sb->new_level;
  973. mddev->new_layout = sb->new_layout;
  974. mddev->new_chunk_sectors = sb->new_chunk >> 9;
  975. } else {
  976. mddev->reshape_position = MaxSector;
  977. mddev->delta_disks = 0;
  978. mddev->new_level = mddev->level;
  979. mddev->new_layout = mddev->layout;
  980. mddev->new_chunk_sectors = mddev->chunk_sectors;
  981. }
  982. if (sb->state & (1<<MD_SB_CLEAN))
  983. mddev->recovery_cp = MaxSector;
  984. else {
  985. if (sb->events_hi == sb->cp_events_hi &&
  986. sb->events_lo == sb->cp_events_lo) {
  987. mddev->recovery_cp = sb->recovery_cp;
  988. } else
  989. mddev->recovery_cp = 0;
  990. }
  991. memcpy(mddev->uuid+0, &sb->set_uuid0, 4);
  992. memcpy(mddev->uuid+4, &sb->set_uuid1, 4);
  993. memcpy(mddev->uuid+8, &sb->set_uuid2, 4);
  994. memcpy(mddev->uuid+12,&sb->set_uuid3, 4);
  995. mddev->max_disks = MD_SB_DISKS;
  996. if (sb->state & (1<<MD_SB_BITMAP_PRESENT) &&
  997. mddev->bitmap_info.file == NULL)
  998. mddev->bitmap_info.offset =
  999. mddev->bitmap_info.default_offset;
  1000. } else if (mddev->pers == NULL) {
  1001. /* Insist on good event counter while assembling, except
  1002. * for spares (which don't need an event count) */
  1003. ++ev1;
  1004. if (sb->disks[rdev->desc_nr].state & (
  1005. (1<<MD_DISK_SYNC) | (1 << MD_DISK_ACTIVE)))
  1006. if (ev1 < mddev->events)
  1007. return -EINVAL;
  1008. } else if (mddev->bitmap) {
  1009. /* if adding to array with a bitmap, then we can accept an
  1010. * older device ... but not too old.
  1011. */
  1012. if (ev1 < mddev->bitmap->events_cleared)
  1013. return 0;
  1014. } else {
  1015. if (ev1 < mddev->events)
  1016. /* just a hot-add of a new device, leave raid_disk at -1 */
  1017. return 0;
  1018. }
  1019. if (mddev->level != LEVEL_MULTIPATH) {
  1020. desc = sb->disks + rdev->desc_nr;
  1021. if (desc->state & (1<<MD_DISK_FAULTY))
  1022. set_bit(Faulty, &rdev->flags);
  1023. else if (desc->state & (1<<MD_DISK_SYNC) /* &&
  1024. desc->raid_disk < mddev->raid_disks */) {
  1025. set_bit(In_sync, &rdev->flags);
  1026. rdev->raid_disk = desc->raid_disk;
  1027. } else if (desc->state & (1<<MD_DISK_ACTIVE)) {
  1028. /* active but not in sync implies recovery up to
  1029. * reshape position. We don't know exactly where
  1030. * that is, so set to zero for now */
  1031. if (mddev->minor_version >= 91) {
  1032. rdev->recovery_offset = 0;
  1033. rdev->raid_disk = desc->raid_disk;
  1034. }
  1035. }
  1036. if (desc->state & (1<<MD_DISK_WRITEMOSTLY))
  1037. set_bit(WriteMostly, &rdev->flags);
  1038. } else /* MULTIPATH are always insync */
  1039. set_bit(In_sync, &rdev->flags);
  1040. return 0;
  1041. }
  1042. /*
  1043. * sync_super for 0.90.0
  1044. */
  1045. static void super_90_sync(mddev_t *mddev, mdk_rdev_t *rdev)
  1046. {
  1047. mdp_super_t *sb;
  1048. mdk_rdev_t *rdev2;
  1049. int next_spare = mddev->raid_disks;
  1050. /* make rdev->sb match mddev data..
  1051. *
  1052. * 1/ zero out disks
  1053. * 2/ Add info for each disk, keeping track of highest desc_nr (next_spare);
  1054. * 3/ any empty disks < next_spare become removed
  1055. *
  1056. * disks[0] gets initialised to REMOVED because
  1057. * we cannot be sure from other fields if it has
  1058. * been initialised or not.
  1059. */
  1060. int i;
  1061. int active=0, working=0,failed=0,spare=0,nr_disks=0;
  1062. rdev->sb_size = MD_SB_BYTES;
  1063. sb = (mdp_super_t*)page_address(rdev->sb_page);
  1064. memset(sb, 0, sizeof(*sb));
  1065. sb->md_magic = MD_SB_MAGIC;
  1066. sb->major_version = mddev->major_version;
  1067. sb->patch_version = mddev->patch_version;
  1068. sb->gvalid_words = 0; /* ignored */
  1069. memcpy(&sb->set_uuid0, mddev->uuid+0, 4);
  1070. memcpy(&sb->set_uuid1, mddev->uuid+4, 4);
  1071. memcpy(&sb->set_uuid2, mddev->uuid+8, 4);
  1072. memcpy(&sb->set_uuid3, mddev->uuid+12,4);
  1073. sb->ctime = mddev->ctime;
  1074. sb->level = mddev->level;
  1075. sb->size = mddev->dev_sectors / 2;
  1076. sb->raid_disks = mddev->raid_disks;
  1077. sb->md_minor = mddev->md_minor;
  1078. sb->not_persistent = 0;
  1079. sb->utime = mddev->utime;
  1080. sb->state = 0;
  1081. sb->events_hi = (mddev->events>>32);
  1082. sb->events_lo = (u32)mddev->events;
  1083. if (mddev->reshape_position == MaxSector)
  1084. sb->minor_version = 90;
  1085. else {
  1086. sb->minor_version = 91;
  1087. sb->reshape_position = mddev->reshape_position;
  1088. sb->new_level = mddev->new_level;
  1089. sb->delta_disks = mddev->delta_disks;
  1090. sb->new_layout = mddev->new_layout;
  1091. sb->new_chunk = mddev->new_chunk_sectors << 9;
  1092. }
  1093. mddev->minor_version = sb->minor_version;
  1094. if (mddev->in_sync)
  1095. {
  1096. sb->recovery_cp = mddev->recovery_cp;
  1097. sb->cp_events_hi = (mddev->events>>32);
  1098. sb->cp_events_lo = (u32)mddev->events;
  1099. if (mddev->recovery_cp == MaxSector)
  1100. sb->state = (1<< MD_SB_CLEAN);
  1101. } else
  1102. sb->recovery_cp = 0;
  1103. sb->layout = mddev->layout;
  1104. sb->chunk_size = mddev->chunk_sectors << 9;
  1105. if (mddev->bitmap && mddev->bitmap_info.file == NULL)
  1106. sb->state |= (1<<MD_SB_BITMAP_PRESENT);
  1107. sb->disks[0].state = (1<<MD_DISK_REMOVED);
  1108. list_for_each_entry(rdev2, &mddev->disks, same_set) {
  1109. mdp_disk_t *d;
  1110. int desc_nr;
  1111. int is_active = test_bit(In_sync, &rdev2->flags);
  1112. if (rdev2->raid_disk >= 0 &&
  1113. sb->minor_version >= 91)
  1114. /* we have nowhere to store the recovery_offset,
  1115. * but if it is not below the reshape_position,
  1116. * we can piggy-back on that.
  1117. */
  1118. is_active = 1;
  1119. if (rdev2->raid_disk < 0 ||
  1120. test_bit(Faulty, &rdev2->flags))
  1121. is_active = 0;
  1122. if (is_active)
  1123. desc_nr = rdev2->raid_disk;
  1124. else
  1125. desc_nr = next_spare++;
  1126. rdev2->desc_nr = desc_nr;
  1127. d = &sb->disks[rdev2->desc_nr];
  1128. nr_disks++;
  1129. d->number = rdev2->desc_nr;
  1130. d->major = MAJOR(rdev2->bdev->bd_dev);
  1131. d->minor = MINOR(rdev2->bdev->bd_dev);
  1132. if (is_active)
  1133. d->raid_disk = rdev2->raid_disk;
  1134. else
  1135. d->raid_disk = rdev2->desc_nr; /* compatibility */
  1136. if (test_bit(Faulty, &rdev2->flags))
  1137. d->state = (1<<MD_DISK_FAULTY);
  1138. else if (is_active) {
  1139. d->state = (1<<MD_DISK_ACTIVE);
  1140. if (test_bit(In_sync, &rdev2->flags))
  1141. d->state |= (1<<MD_DISK_SYNC);
  1142. active++;
  1143. working++;
  1144. } else {
  1145. d->state = 0;
  1146. spare++;
  1147. working++;
  1148. }
  1149. if (test_bit(WriteMostly, &rdev2->flags))
  1150. d->state |= (1<<MD_DISK_WRITEMOSTLY);
  1151. }
  1152. /* now set the "removed" and "faulty" bits on any missing devices */
  1153. for (i=0 ; i < mddev->raid_disks ; i++) {
  1154. mdp_disk_t *d = &sb->disks[i];
  1155. if (d->state == 0 && d->number == 0) {
  1156. d->number = i;
  1157. d->raid_disk = i;
  1158. d->state = (1<<MD_DISK_REMOVED);
  1159. d->state |= (1<<MD_DISK_FAULTY);
  1160. failed++;
  1161. }
  1162. }
  1163. sb->nr_disks = nr_disks;
  1164. sb->active_disks = active;
  1165. sb->working_disks = working;
  1166. sb->failed_disks = failed;
  1167. sb->spare_disks = spare;
  1168. sb->this_disk = sb->disks[rdev->desc_nr];
  1169. sb->sb_csum = calc_sb_csum(sb);
  1170. }
  1171. /*
  1172. * rdev_size_change for 0.90.0
  1173. */
  1174. static unsigned long long
  1175. super_90_rdev_size_change(mdk_rdev_t *rdev, sector_t num_sectors)
  1176. {
  1177. if (num_sectors && num_sectors < rdev->mddev->dev_sectors)
  1178. return 0; /* component must fit device */
  1179. if (rdev->mddev->bitmap_info.offset)
  1180. return 0; /* can't move bitmap */
  1181. rdev->sb_start = calc_dev_sboffset(rdev);
  1182. if (!num_sectors || num_sectors > rdev->sb_start)
  1183. num_sectors = rdev->sb_start;
  1184. md_super_write(rdev->mddev, rdev, rdev->sb_start, rdev->sb_size,
  1185. rdev->sb_page);
  1186. md_super_wait(rdev->mddev);
  1187. return num_sectors;
  1188. }
  1189. /*
  1190. * version 1 superblock
  1191. */
  1192. static __le32 calc_sb_1_csum(struct mdp_superblock_1 * sb)
  1193. {
  1194. __le32 disk_csum;
  1195. u32 csum;
  1196. unsigned long long newcsum;
  1197. int size = 256 + le32_to_cpu(sb->max_dev)*2;
  1198. __le32 *isuper = (__le32*)sb;
  1199. int i;
  1200. disk_csum = sb->sb_csum;
  1201. sb->sb_csum = 0;
  1202. newcsum = 0;
  1203. for (i=0; size>=4; size -= 4 )
  1204. newcsum += le32_to_cpu(*isuper++);
  1205. if (size == 2)
  1206. newcsum += le16_to_cpu(*(__le16*) isuper);
  1207. csum = (newcsum & 0xffffffff) + (newcsum >> 32);
  1208. sb->sb_csum = disk_csum;
  1209. return cpu_to_le32(csum);
  1210. }
  1211. static int super_1_load(mdk_rdev_t *rdev, mdk_rdev_t *refdev, int minor_version)
  1212. {
  1213. struct mdp_superblock_1 *sb;
  1214. int ret;
  1215. sector_t sb_start;
  1216. char b[BDEVNAME_SIZE], b2[BDEVNAME_SIZE];
  1217. int bmask;
  1218. /*
  1219. * Calculate the position of the superblock in 512byte sectors.
  1220. * It is always aligned to a 4K boundary and
  1221. * depeding on minor_version, it can be:
  1222. * 0: At least 8K, but less than 12K, from end of device
  1223. * 1: At start of device
  1224. * 2: 4K from start of device.
  1225. */
  1226. switch(minor_version) {
  1227. case 0:
  1228. sb_start = i_size_read(rdev->bdev->bd_inode) >> 9;
  1229. sb_start -= 8*2;
  1230. sb_start &= ~(sector_t)(4*2-1);
  1231. break;
  1232. case 1:
  1233. sb_start = 0;
  1234. break;
  1235. case 2:
  1236. sb_start = 8;
  1237. break;
  1238. default:
  1239. return -EINVAL;
  1240. }
  1241. rdev->sb_start = sb_start;
  1242. /* superblock is rarely larger than 1K, but it can be larger,
  1243. * and it is safe to read 4k, so we do that
  1244. */
  1245. ret = read_disk_sb(rdev, 4096);
  1246. if (ret) return ret;
  1247. sb = (struct mdp_superblock_1*)page_address(rdev->sb_page);
  1248. if (sb->magic != cpu_to_le32(MD_SB_MAGIC) ||
  1249. sb->major_version != cpu_to_le32(1) ||
  1250. le32_to_cpu(sb->max_dev) > (4096-256)/2 ||
  1251. le64_to_cpu(sb->super_offset) != rdev->sb_start ||
  1252. (le32_to_cpu(sb->feature_map) & ~MD_FEATURE_ALL) != 0)
  1253. return -EINVAL;
  1254. if (calc_sb_1_csum(sb) != sb->sb_csum) {
  1255. printk("md: invalid superblock checksum on %s\n",
  1256. bdevname(rdev->bdev,b));
  1257. return -EINVAL;
  1258. }
  1259. if (le64_to_cpu(sb->data_size) < 10) {
  1260. printk("md: data_size too small on %s\n",
  1261. bdevname(rdev->bdev,b));
  1262. return -EINVAL;
  1263. }
  1264. rdev->preferred_minor = 0xffff;
  1265. rdev->data_offset = le64_to_cpu(sb->data_offset);
  1266. atomic_set(&rdev->corrected_errors, le32_to_cpu(sb->cnt_corrected_read));
  1267. rdev->sb_size = le32_to_cpu(sb->max_dev) * 2 + 256;
  1268. bmask = queue_logical_block_size(rdev->bdev->bd_disk->queue)-1;
  1269. if (rdev->sb_size & bmask)
  1270. rdev->sb_size = (rdev->sb_size | bmask) + 1;
  1271. if (minor_version
  1272. && rdev->data_offset < sb_start + (rdev->sb_size/512))
  1273. return -EINVAL;
  1274. if (sb->level == cpu_to_le32(LEVEL_MULTIPATH))
  1275. rdev->desc_nr = -1;
  1276. else
  1277. rdev->desc_nr = le32_to_cpu(sb->dev_number);
  1278. if (!refdev) {
  1279. ret = 1;
  1280. } else {
  1281. __u64 ev1, ev2;
  1282. struct mdp_superblock_1 *refsb =
  1283. (struct mdp_superblock_1*)page_address(refdev->sb_page);
  1284. if (memcmp(sb->set_uuid, refsb->set_uuid, 16) != 0 ||
  1285. sb->level != refsb->level ||
  1286. sb->layout != refsb->layout ||
  1287. sb->chunksize != refsb->chunksize) {
  1288. printk(KERN_WARNING "md: %s has strangely different"
  1289. " superblock to %s\n",
  1290. bdevname(rdev->bdev,b),
  1291. bdevname(refdev->bdev,b2));
  1292. return -EINVAL;
  1293. }
  1294. ev1 = le64_to_cpu(sb->events);
  1295. ev2 = le64_to_cpu(refsb->events);
  1296. if (ev1 > ev2)
  1297. ret = 1;
  1298. else
  1299. ret = 0;
  1300. }
  1301. if (minor_version)
  1302. rdev->sectors = (i_size_read(rdev->bdev->bd_inode) >> 9) -
  1303. le64_to_cpu(sb->data_offset);
  1304. else
  1305. rdev->sectors = rdev->sb_start;
  1306. if (rdev->sectors < le64_to_cpu(sb->data_size))
  1307. return -EINVAL;
  1308. rdev->sectors = le64_to_cpu(sb->data_size);
  1309. if (le64_to_cpu(sb->size) > rdev->sectors)
  1310. return -EINVAL;
  1311. return ret;
  1312. }
  1313. static int super_1_validate(mddev_t *mddev, mdk_rdev_t *rdev)
  1314. {
  1315. struct mdp_superblock_1 *sb = (struct mdp_superblock_1*)page_address(rdev->sb_page);
  1316. __u64 ev1 = le64_to_cpu(sb->events);
  1317. rdev->raid_disk = -1;
  1318. clear_bit(Faulty, &rdev->flags);
  1319. clear_bit(In_sync, &rdev->flags);
  1320. clear_bit(WriteMostly, &rdev->flags);
  1321. if (mddev->raid_disks == 0) {
  1322. mddev->major_version = 1;
  1323. mddev->patch_version = 0;
  1324. mddev->external = 0;
  1325. mddev->chunk_sectors = le32_to_cpu(sb->chunksize);
  1326. mddev->ctime = le64_to_cpu(sb->ctime) & ((1ULL << 32)-1);
  1327. mddev->utime = le64_to_cpu(sb->utime) & ((1ULL << 32)-1);
  1328. mddev->level = le32_to_cpu(sb->level);
  1329. mddev->clevel[0] = 0;
  1330. mddev->layout = le32_to_cpu(sb->layout);
  1331. mddev->raid_disks = le32_to_cpu(sb->raid_disks);
  1332. mddev->dev_sectors = le64_to_cpu(sb->size);
  1333. mddev->events = ev1;
  1334. mddev->bitmap_info.offset = 0;
  1335. mddev->bitmap_info.default_offset = 1024 >> 9;
  1336. mddev->recovery_cp = le64_to_cpu(sb->resync_offset);
  1337. memcpy(mddev->uuid, sb->set_uuid, 16);
  1338. mddev->max_disks = (4096-256)/2;
  1339. if ((le32_to_cpu(sb->feature_map) & MD_FEATURE_BITMAP_OFFSET) &&
  1340. mddev->bitmap_info.file == NULL )
  1341. mddev->bitmap_info.offset =
  1342. (__s32)le32_to_cpu(sb->bitmap_offset);
  1343. if ((le32_to_cpu(sb->feature_map) & MD_FEATURE_RESHAPE_ACTIVE)) {
  1344. mddev->reshape_position = le64_to_cpu(sb->reshape_position);
  1345. mddev->delta_disks = le32_to_cpu(sb->delta_disks);
  1346. mddev->new_level = le32_to_cpu(sb->new_level);
  1347. mddev->new_layout = le32_to_cpu(sb->new_layout);
  1348. mddev->new_chunk_sectors = le32_to_cpu(sb->new_chunk);
  1349. } else {
  1350. mddev->reshape_position = MaxSector;
  1351. mddev->delta_disks = 0;
  1352. mddev->new_level = mddev->level;
  1353. mddev->new_layout = mddev->layout;
  1354. mddev->new_chunk_sectors = mddev->chunk_sectors;
  1355. }
  1356. } else if (mddev->pers == NULL) {
  1357. /* Insist of good event counter while assembling, except for
  1358. * spares (which don't need an event count) */
  1359. ++ev1;
  1360. if (rdev->desc_nr >= 0 &&
  1361. rdev->desc_nr < le32_to_cpu(sb->max_dev) &&
  1362. le16_to_cpu(sb->dev_roles[rdev->desc_nr]) < 0xfffe)
  1363. if (ev1 < mddev->events)
  1364. return -EINVAL;
  1365. } else if (mddev->bitmap) {
  1366. /* If adding to array with a bitmap, then we can accept an
  1367. * older device, but not too old.
  1368. */
  1369. if (ev1 < mddev->bitmap->events_cleared)
  1370. return 0;
  1371. } else {
  1372. if (ev1 < mddev->events)
  1373. /* just a hot-add of a new device, leave raid_disk at -1 */
  1374. return 0;
  1375. }
  1376. if (mddev->level != LEVEL_MULTIPATH) {
  1377. int role;
  1378. if (rdev->desc_nr < 0 ||
  1379. rdev->desc_nr >= le32_to_cpu(sb->max_dev)) {
  1380. role = 0xffff;
  1381. rdev->desc_nr = -1;
  1382. } else
  1383. role = le16_to_cpu(sb->dev_roles[rdev->desc_nr]);
  1384. switch(role) {
  1385. case 0xffff: /* spare */
  1386. break;
  1387. case 0xfffe: /* faulty */
  1388. set_bit(Faulty, &rdev->flags);
  1389. break;
  1390. default:
  1391. if ((le32_to_cpu(sb->feature_map) &
  1392. MD_FEATURE_RECOVERY_OFFSET))
  1393. rdev->recovery_offset = le64_to_cpu(sb->recovery_offset);
  1394. else
  1395. set_bit(In_sync, &rdev->flags);
  1396. rdev->raid_disk = role;
  1397. break;
  1398. }
  1399. if (sb->devflags & WriteMostly1)
  1400. set_bit(WriteMostly, &rdev->flags);
  1401. } else /* MULTIPATH are always insync */
  1402. set_bit(In_sync, &rdev->flags);
  1403. return 0;
  1404. }
  1405. static void super_1_sync(mddev_t *mddev, mdk_rdev_t *rdev)
  1406. {
  1407. struct mdp_superblock_1 *sb;
  1408. mdk_rdev_t *rdev2;
  1409. int max_dev, i;
  1410. /* make rdev->sb match mddev and rdev data. */
  1411. sb = (struct mdp_superblock_1*)page_address(rdev->sb_page);
  1412. sb->feature_map = 0;
  1413. sb->pad0 = 0;
  1414. sb->recovery_offset = cpu_to_le64(0);
  1415. memset(sb->pad1, 0, sizeof(sb->pad1));
  1416. memset(sb->pad2, 0, sizeof(sb->pad2));
  1417. memset(sb->pad3, 0, sizeof(sb->pad3));
  1418. sb->utime = cpu_to_le64((__u64)mddev->utime);
  1419. sb->events = cpu_to_le64(mddev->events);
  1420. if (mddev->in_sync)
  1421. sb->resync_offset = cpu_to_le64(mddev->recovery_cp);
  1422. else
  1423. sb->resync_offset = cpu_to_le64(0);
  1424. sb->cnt_corrected_read = cpu_to_le32(atomic_read(&rdev->corrected_errors));
  1425. sb->raid_disks = cpu_to_le32(mddev->raid_disks);
  1426. sb->size = cpu_to_le64(mddev->dev_sectors);
  1427. sb->chunksize = cpu_to_le32(mddev->chunk_sectors);
  1428. sb->level = cpu_to_le32(mddev->level);
  1429. sb->layout = cpu_to_le32(mddev->layout);
  1430. if (mddev->bitmap && mddev->bitmap_info.file == NULL) {
  1431. sb->bitmap_offset = cpu_to_le32((__u32)mddev->bitmap_info.offset);
  1432. sb->feature_map = cpu_to_le32(MD_FEATURE_BITMAP_OFFSET);
  1433. }
  1434. if (rdev->raid_disk >= 0 &&
  1435. !test_bit(In_sync, &rdev->flags)) {
  1436. sb->feature_map |=
  1437. cpu_to_le32(MD_FEATURE_RECOVERY_OFFSET);
  1438. sb->recovery_offset =
  1439. cpu_to_le64(rdev->recovery_offset);
  1440. }
  1441. if (mddev->reshape_position != MaxSector) {
  1442. sb->feature_map |= cpu_to_le32(MD_FEATURE_RESHAPE_ACTIVE);
  1443. sb->reshape_position = cpu_to_le64(mddev->reshape_position);
  1444. sb->new_layout = cpu_to_le32(mddev->new_layout);
  1445. sb->delta_disks = cpu_to_le32(mddev->delta_disks);
  1446. sb->new_level = cpu_to_le32(mddev->new_level);
  1447. sb->new_chunk = cpu_to_le32(mddev->new_chunk_sectors);
  1448. }
  1449. max_dev = 0;
  1450. list_for_each_entry(rdev2, &mddev->disks, same_set)
  1451. if (rdev2->desc_nr+1 > max_dev)
  1452. max_dev = rdev2->desc_nr+1;
  1453. if (max_dev > le32_to_cpu(sb->max_dev)) {
  1454. int bmask;
  1455. sb->max_dev = cpu_to_le32(max_dev);
  1456. rdev->sb_size = max_dev * 2 + 256;
  1457. bmask = queue_logical_block_size(rdev->bdev->bd_disk->queue)-1;
  1458. if (rdev->sb_size & bmask)
  1459. rdev->sb_size = (rdev->sb_size | bmask) + 1;
  1460. } else
  1461. max_dev = le32_to_cpu(sb->max_dev);
  1462. for (i=0; i<max_dev;i++)
  1463. sb->dev_roles[i] = cpu_to_le16(0xfffe);
  1464. list_for_each_entry(rdev2, &mddev->disks, same_set) {
  1465. i = rdev2->desc_nr;
  1466. if (test_bit(Faulty, &rdev2->flags))
  1467. sb->dev_roles[i] = cpu_to_le16(0xfffe);
  1468. else if (test_bit(In_sync, &rdev2->flags))
  1469. sb->dev_roles[i] = cpu_to_le16(rdev2->raid_disk);
  1470. else if (rdev2->raid_disk >= 0)
  1471. sb->dev_roles[i] = cpu_to_le16(rdev2->raid_disk);
  1472. else
  1473. sb->dev_roles[i] = cpu_to_le16(0xffff);
  1474. }
  1475. sb->sb_csum = calc_sb_1_csum(sb);
  1476. }
  1477. static unsigned long long
  1478. super_1_rdev_size_change(mdk_rdev_t *rdev, sector_t num_sectors)
  1479. {
  1480. struct mdp_superblock_1 *sb;
  1481. sector_t max_sectors;
  1482. if (num_sectors && num_sectors < rdev->mddev->dev_sectors)
  1483. return 0; /* component must fit device */
  1484. if (rdev->sb_start < rdev->data_offset) {
  1485. /* minor versions 1 and 2; superblock before data */
  1486. max_sectors = i_size_read(rdev->bdev->bd_inode) >> 9;
  1487. max_sectors -= rdev->data_offset;
  1488. if (!num_sectors || num_sectors > max_sectors)
  1489. num_sectors = max_sectors;
  1490. } else if (rdev->mddev->bitmap_info.offset) {
  1491. /* minor version 0 with bitmap we can't move */
  1492. return 0;
  1493. } else {
  1494. /* minor version 0; superblock after data */
  1495. sector_t sb_start;
  1496. sb_start = (i_size_read(rdev->bdev->bd_inode) >> 9) - 8*2;
  1497. sb_start &= ~(sector_t)(4*2 - 1);
  1498. max_sectors = rdev->sectors + sb_start - rdev->sb_start;
  1499. if (!num_sectors || num_sectors > max_sectors)
  1500. num_sectors = max_sectors;
  1501. rdev->sb_start = sb_start;
  1502. }
  1503. sb = (struct mdp_superblock_1 *) page_address(rdev->sb_page);
  1504. sb->data_size = cpu_to_le64(num_sectors);
  1505. sb->super_offset = rdev->sb_start;
  1506. sb->sb_csum = calc_sb_1_csum(sb);
  1507. md_super_write(rdev->mddev, rdev, rdev->sb_start, rdev->sb_size,
  1508. rdev->sb_page);
  1509. md_super_wait(rdev->mddev);
  1510. return num_sectors;
  1511. }
  1512. static struct super_type super_types[] = {
  1513. [0] = {
  1514. .name = "0.90.0",
  1515. .owner = THIS_MODULE,
  1516. .load_super = super_90_load,
  1517. .validate_super = super_90_validate,
  1518. .sync_super = super_90_sync,
  1519. .rdev_size_change = super_90_rdev_size_change,
  1520. },
  1521. [1] = {
  1522. .name = "md-1",
  1523. .owner = THIS_MODULE,
  1524. .load_super = super_1_load,
  1525. .validate_super = super_1_validate,
  1526. .sync_super = super_1_sync,
  1527. .rdev_size_change = super_1_rdev_size_change,
  1528. },
  1529. };
  1530. static void sync_super(mddev_t *mddev, mdk_rdev_t *rdev)
  1531. {
  1532. if (mddev->sync_super) {
  1533. mddev->sync_super(mddev, rdev);
  1534. return;
  1535. }
  1536. BUG_ON(mddev->major_version >= ARRAY_SIZE(super_types));
  1537. super_types[mddev->major_version].sync_super(mddev, rdev);
  1538. }
  1539. static int match_mddev_units(mddev_t *mddev1, mddev_t *mddev2)
  1540. {
  1541. mdk_rdev_t *rdev, *rdev2;
  1542. rcu_read_lock();
  1543. rdev_for_each_rcu(rdev, mddev1)
  1544. rdev_for_each_rcu(rdev2, mddev2)
  1545. if (rdev->bdev->bd_contains ==
  1546. rdev2->bdev->bd_contains) {
  1547. rcu_read_unlock();
  1548. return 1;
  1549. }
  1550. rcu_read_unlock();
  1551. return 0;
  1552. }
  1553. static LIST_HEAD(pending_raid_disks);
  1554. /*
  1555. * Try to register data integrity profile for an mddev
  1556. *
  1557. * This is called when an array is started and after a disk has been kicked
  1558. * from the array. It only succeeds if all working and active component devices
  1559. * are integrity capable with matching profiles.
  1560. */
  1561. int md_integrity_register(mddev_t *mddev)
  1562. {
  1563. mdk_rdev_t *rdev, *reference = NULL;
  1564. if (list_empty(&mddev->disks))
  1565. return 0; /* nothing to do */
  1566. if (!mddev->gendisk || blk_get_integrity(mddev->gendisk))
  1567. return 0; /* shouldn't register, or already is */
  1568. list_for_each_entry(rdev, &mddev->disks, same_set) {
  1569. /* skip spares and non-functional disks */
  1570. if (test_bit(Faulty, &rdev->flags))
  1571. continue;
  1572. if (rdev->raid_disk < 0)
  1573. continue;
  1574. if (!reference) {
  1575. /* Use the first rdev as the reference */
  1576. reference = rdev;
  1577. continue;
  1578. }
  1579. /* does this rdev's profile match the reference profile? */
  1580. if (blk_integrity_compare(reference->bdev->bd_disk,
  1581. rdev->bdev->bd_disk) < 0)
  1582. return -EINVAL;
  1583. }
  1584. if (!reference || !bdev_get_integrity(reference->bdev))
  1585. return 0;
  1586. /*
  1587. * All component devices are integrity capable and have matching
  1588. * profiles, register the common profile for the md device.
  1589. */
  1590. if (blk_integrity_register(mddev->gendisk,
  1591. bdev_get_integrity(reference->bdev)) != 0) {
  1592. printk(KERN_ERR "md: failed to register integrity for %s\n",
  1593. mdname(mddev));
  1594. return -EINVAL;
  1595. }
  1596. printk(KERN_NOTICE "md: data integrity enabled on %s\n", mdname(mddev));
  1597. if (bioset_integrity_create(mddev->bio_set, BIO_POOL_SIZE)) {
  1598. printk(KERN_ERR "md: failed to create integrity pool for %s\n",
  1599. mdname(mddev));
  1600. return -EINVAL;
  1601. }
  1602. return 0;
  1603. }
  1604. EXPORT_SYMBOL(md_integrity_register);
  1605. /* Disable data integrity if non-capable/non-matching disk is being added */
  1606. void md_integrity_add_rdev(mdk_rdev_t *rdev, mddev_t *mddev)
  1607. {
  1608. struct blk_integrity *bi_rdev = bdev_get_integrity(rdev->bdev);
  1609. struct blk_integrity *bi_mddev = blk_get_integrity(mddev->gendisk);
  1610. if (!bi_mddev) /* nothing to do */
  1611. return;
  1612. if (rdev->raid_disk < 0) /* skip spares */
  1613. return;
  1614. if (bi_rdev && blk_integrity_compare(mddev->gendisk,
  1615. rdev->bdev->bd_disk) >= 0)
  1616. return;
  1617. printk(KERN_NOTICE "disabling data integrity on %s\n", mdname(mddev));
  1618. blk_integrity_unregister(mddev->gendisk);
  1619. }
  1620. EXPORT_SYMBOL(md_integrity_add_rdev);
  1621. static int bind_rdev_to_array(mdk_rdev_t * rdev, mddev_t * mddev)
  1622. {
  1623. char b[BDEVNAME_SIZE];
  1624. struct kobject *ko;
  1625. char *s;
  1626. int err;
  1627. if (rdev->mddev) {
  1628. MD_BUG();
  1629. return -EINVAL;
  1630. }
  1631. /* prevent duplicates */
  1632. if (find_rdev(mddev, rdev->bdev->bd_dev))
  1633. return -EEXIST;
  1634. /* make sure rdev->sectors exceeds mddev->dev_sectors */
  1635. if (rdev->sectors && (mddev->dev_sectors == 0 ||
  1636. rdev->sectors < mddev->dev_sectors)) {
  1637. if (mddev->pers) {
  1638. /* Cannot change size, so fail
  1639. * If mddev->level <= 0, then we don't care
  1640. * about aligning sizes (e.g. linear)
  1641. */
  1642. if (mddev->level > 0)
  1643. return -ENOSPC;
  1644. } else
  1645. mddev->dev_sectors = rdev->sectors;
  1646. }
  1647. /* Verify rdev->desc_nr is unique.
  1648. * If it is -1, assign a free number, else
  1649. * check number is not in use
  1650. */
  1651. if (rdev->desc_nr < 0) {
  1652. int choice = 0;
  1653. if (mddev->pers) choice = mddev->raid_disks;
  1654. while (find_rdev_nr(mddev, choice))
  1655. choice++;
  1656. rdev->desc_nr = choice;
  1657. } else {
  1658. if (find_rdev_nr(mddev, rdev->desc_nr))
  1659. return -EBUSY;
  1660. }
  1661. if (mddev->max_disks && rdev->desc_nr >= mddev->max_disks) {
  1662. printk(KERN_WARNING "md: %s: array is limited to %d devices\n",
  1663. mdname(mddev), mddev->max_disks);
  1664. return -EBUSY;
  1665. }
  1666. bdevname(rdev->bdev,b);
  1667. while ( (s=strchr(b, '/')) != NULL)
  1668. *s = '!';
  1669. rdev->mddev = mddev;
  1670. printk(KERN_INFO "md: bind<%s>\n", b);
  1671. if ((err = kobject_add(&rdev->kobj, &mddev->kobj, "dev-%s", b)))
  1672. goto fail;
  1673. ko = &part_to_dev(rdev->bdev->bd_part)->kobj;
  1674. if (sysfs_create_link(&rdev->kobj, ko, "block"))
  1675. /* failure here is OK */;
  1676. rdev->sysfs_state = sysfs_get_dirent_safe(rdev->kobj.sd, "state");
  1677. list_add_rcu(&rdev->same_set, &mddev->disks);
  1678. bd_link_disk_holder(rdev->bdev, mddev->gendisk);
  1679. /* May as well allow recovery to be retried once */
  1680. mddev->recovery_disabled = 0;
  1681. return 0;
  1682. fail:
  1683. printk(KERN_WARNING "md: failed to register dev-%s for %s\n",
  1684. b, mdname(mddev));
  1685. return err;
  1686. }
  1687. static void md_delayed_delete(struct work_struct *ws)
  1688. {
  1689. mdk_rdev_t *rdev = container_of(ws, mdk_rdev_t, del_work);
  1690. kobject_del(&rdev->kobj);
  1691. kobject_put(&rdev->kobj);
  1692. }
  1693. static void unbind_rdev_from_array(mdk_rdev_t * rdev)
  1694. {
  1695. char b[BDEVNAME_SIZE];
  1696. if (!rdev->mddev) {
  1697. MD_BUG();
  1698. return;
  1699. }
  1700. bd_unlink_disk_holder(rdev->bdev, rdev->mddev->gendisk);
  1701. list_del_rcu(&rdev->same_set);
  1702. printk(KERN_INFO "md: unbind<%s>\n", bdevname(rdev->bdev,b));
  1703. rdev->mddev = NULL;
  1704. sysfs_remove_link(&rdev->kobj, "block");
  1705. sysfs_put(rdev->sysfs_state);
  1706. rdev->sysfs_state = NULL;
  1707. /* We need to delay this, otherwise we can deadlock when
  1708. * writing to 'remove' to "dev/state". We also need
  1709. * to delay it due to rcu usage.
  1710. */
  1711. synchronize_rcu();
  1712. INIT_WORK(&rdev->del_work, md_delayed_delete);
  1713. kobject_get(&rdev->kobj);
  1714. queue_work(md_misc_wq, &rdev->del_work);
  1715. }
  1716. /*
  1717. * prevent the device from being mounted, repartitioned or
  1718. * otherwise reused by a RAID array (or any other kernel
  1719. * subsystem), by bd_claiming the device.
  1720. */
  1721. static int lock_rdev(mdk_rdev_t *rdev, dev_t dev, int shared)
  1722. {
  1723. int err = 0;
  1724. struct block_device *bdev;
  1725. char b[BDEVNAME_SIZE];
  1726. bdev = blkdev_get_by_dev(dev, FMODE_READ|FMODE_WRITE|FMODE_EXCL,
  1727. shared ? (mdk_rdev_t *)lock_rdev : rdev);
  1728. if (IS_ERR(bdev)) {
  1729. printk(KERN_ERR "md: could not open %s.\n",
  1730. __bdevname(dev, b));
  1731. return PTR_ERR(bdev);
  1732. }
  1733. rdev->bdev = bdev;
  1734. return err;
  1735. }
  1736. static void unlock_rdev(mdk_rdev_t *rdev)
  1737. {
  1738. struct block_device *bdev = rdev->bdev;
  1739. rdev->bdev = NULL;
  1740. if (!bdev)
  1741. MD_BUG();
  1742. blkdev_put(bdev, FMODE_READ|FMODE_WRITE|FMODE_EXCL);
  1743. }
  1744. void md_autodetect_dev(dev_t dev);
  1745. static void export_rdev(mdk_rdev_t * rdev)
  1746. {
  1747. char b[BDEVNAME_SIZE];
  1748. printk(KERN_INFO "md: export_rdev(%s)\n",
  1749. bdevname(rdev->bdev,b));
  1750. if (rdev->mddev)
  1751. MD_BUG();
  1752. free_disk_sb(rdev);
  1753. #ifndef MODULE
  1754. if (test_bit(AutoDetected, &rdev->flags))
  1755. md_autodetect_dev(rdev->bdev->bd_dev);
  1756. #endif
  1757. unlock_rdev(rdev);
  1758. kobject_put(&rdev->kobj);
  1759. }
  1760. static void kick_rdev_from_array(mdk_rdev_t * rdev)
  1761. {
  1762. unbind_rdev_from_array(rdev);
  1763. export_rdev(rdev);
  1764. }
  1765. static void export_array(mddev_t *mddev)
  1766. {
  1767. mdk_rdev_t *rdev, *tmp;
  1768. rdev_for_each(rdev, tmp, mddev) {
  1769. if (!rdev->mddev) {
  1770. MD_BUG();
  1771. continue;
  1772. }
  1773. kick_rdev_from_array(rdev);
  1774. }
  1775. if (!list_empty(&mddev->disks))
  1776. MD_BUG();
  1777. mddev->raid_disks = 0;
  1778. mddev->major_version = 0;
  1779. }
  1780. static void print_desc(mdp_disk_t *desc)
  1781. {
  1782. printk(" DISK<N:%d,(%d,%d),R:%d,S:%d>\n", desc->number,
  1783. desc->major,desc->minor,desc->raid_disk,desc->state);
  1784. }
  1785. static void print_sb_90(mdp_super_t *sb)
  1786. {
  1787. int i;
  1788. printk(KERN_INFO
  1789. "md: SB: (V:%d.%d.%d) ID:<%08x.%08x.%08x.%08x> CT:%08x\n",
  1790. sb->major_version, sb->minor_version, sb->patch_version,
  1791. sb->set_uuid0, sb->set_uuid1, sb->set_uuid2, sb->set_uuid3,
  1792. sb->ctime);
  1793. printk(KERN_INFO "md: L%d S%08d ND:%d RD:%d md%d LO:%d CS:%d\n",
  1794. sb->level, sb->size, sb->nr_disks, sb->raid_disks,
  1795. sb->md_minor, sb->layout, sb->chunk_size);
  1796. printk(KERN_INFO "md: UT:%08x ST:%d AD:%d WD:%d"
  1797. " FD:%d SD:%d CSUM:%08x E:%08lx\n",
  1798. sb->utime, sb->state, sb->active_disks, sb->working_disks,
  1799. sb->failed_disks, sb->spare_disks,
  1800. sb->sb_csum, (unsigned long)sb->events_lo);
  1801. printk(KERN_INFO);
  1802. for (i = 0; i < MD_SB_DISKS; i++) {
  1803. mdp_disk_t *desc;
  1804. desc = sb->disks + i;
  1805. if (desc->number || desc->major || desc->minor ||
  1806. desc->raid_disk || (desc->state && (desc->state != 4))) {
  1807. printk(" D %2d: ", i);
  1808. print_desc(desc);
  1809. }
  1810. }
  1811. printk(KERN_INFO "md: THIS: ");
  1812. print_desc(&sb->this_disk);
  1813. }
  1814. static void print_sb_1(struct mdp_superblock_1 *sb)
  1815. {
  1816. __u8 *uuid;
  1817. uuid = sb->set_uuid;
  1818. printk(KERN_INFO
  1819. "md: SB: (V:%u) (F:0x%08x) Array-ID:<%pU>\n"
  1820. "md: Name: \"%s\" CT:%llu\n",
  1821. le32_to_cpu(sb->major_version),
  1822. le32_to_cpu(sb->feature_map),
  1823. uuid,
  1824. sb->set_name,
  1825. (unsigned long long)le64_to_cpu(sb->ctime)
  1826. & MD_SUPERBLOCK_1_TIME_SEC_MASK);
  1827. uuid = sb->device_uuid;
  1828. printk(KERN_INFO
  1829. "md: L%u SZ%llu RD:%u LO:%u CS:%u DO:%llu DS:%llu SO:%llu"
  1830. " RO:%llu\n"
  1831. "md: Dev:%08x UUID: %pU\n"
  1832. "md: (F:0x%08x) UT:%llu Events:%llu ResyncOffset:%llu CSUM:0x%08x\n"
  1833. "md: (MaxDev:%u) \n",
  1834. le32_to_cpu(sb->level),
  1835. (unsigned long long)le64_to_cpu(sb->size),
  1836. le32_to_cpu(sb->raid_disks),
  1837. le32_to_cpu(sb->layout),
  1838. le32_to_cpu(sb->chunksize),
  1839. (unsigned long long)le64_to_cpu(sb->data_offset),
  1840. (unsigned long long)le64_to_cpu(sb->data_size),
  1841. (unsigned long long)le64_to_cpu(sb->super_offset),
  1842. (unsigned long long)le64_to_cpu(sb->recovery_offset),
  1843. le32_to_cpu(sb->dev_number),
  1844. uuid,
  1845. sb->devflags,
  1846. (unsigned long long)le64_to_cpu(sb->utime) & MD_SUPERBLOCK_1_TIME_SEC_MASK,
  1847. (unsigned long long)le64_to_cpu(sb->events),
  1848. (unsigned long long)le64_to_cpu(sb->resync_offset),
  1849. le32_to_cpu(sb->sb_csum),
  1850. le32_to_cpu(sb->max_dev)
  1851. );
  1852. }
  1853. static void print_rdev(mdk_rdev_t *rdev, int major_version)
  1854. {
  1855. char b[BDEVNAME_SIZE];
  1856. printk(KERN_INFO "md: rdev %s, Sect:%08llu F:%d S:%d DN:%u\n",
  1857. bdevname(rdev->bdev, b), (unsigned long long)rdev->sectors,
  1858. test_bit(Faulty, &rdev->flags), test_bit(In_sync, &rdev->flags),
  1859. rdev->desc_nr);
  1860. if (rdev->sb_loaded) {
  1861. printk(KERN_INFO "md: rdev superblock (MJ:%d):\n", major_version);
  1862. switch (major_version) {
  1863. case 0:
  1864. print_sb_90((mdp_super_t*)page_address(rdev->sb_page));
  1865. break;
  1866. case 1:
  1867. print_sb_1((struct mdp_superblock_1 *)page_address(rdev->sb_page));
  1868. break;
  1869. }
  1870. } else
  1871. printk(KERN_INFO "md: no rdev superblock!\n");
  1872. }
  1873. static void md_print_devices(void)
  1874. {
  1875. struct list_head *tmp;
  1876. mdk_rdev_t *rdev;
  1877. mddev_t *mddev;
  1878. char b[BDEVNAME_SIZE];
  1879. printk("\n");
  1880. printk("md: **********************************\n");
  1881. printk("md: * <COMPLETE RAID STATE PRINTOUT> *\n");
  1882. printk("md: **********************************\n");
  1883. for_each_mddev(mddev, tmp) {
  1884. if (mddev->bitmap)
  1885. bitmap_print_sb(mddev->bitmap);
  1886. else
  1887. printk("%s: ", mdname(mddev));
  1888. list_for_each_entry(rdev, &mddev->disks, same_set)
  1889. printk("<%s>", bdevname(rdev->bdev,b));
  1890. printk("\n");
  1891. list_for_each_entry(rdev, &mddev->disks, same_set)
  1892. print_rdev(rdev, mddev->major_version);
  1893. }
  1894. printk("md: **********************************\n");
  1895. printk("\n");
  1896. }
  1897. static void sync_sbs(mddev_t * mddev, int nospares)
  1898. {
  1899. /* Update each superblock (in-memory image), but
  1900. * if we are allowed to, skip spares which already
  1901. * have the right event counter, or have one earlier
  1902. * (which would mean they aren't being marked as dirty
  1903. * with the rest of the array)
  1904. */
  1905. mdk_rdev_t *rdev;
  1906. list_for_each_entry(rdev, &mddev->disks, same_set) {
  1907. if (rdev->sb_events == mddev->events ||
  1908. (nospares &&
  1909. rdev->raid_disk < 0 &&
  1910. rdev->sb_events+1 == mddev->events)) {
  1911. /* Don't update this superblock */
  1912. rdev->sb_loaded = 2;
  1913. } else {
  1914. sync_super(mddev, rdev);
  1915. rdev->sb_loaded = 1;
  1916. }
  1917. }
  1918. }
  1919. static void md_update_sb(mddev_t * mddev, int force_change)
  1920. {
  1921. mdk_rdev_t *rdev;
  1922. int sync_req;
  1923. int nospares = 0;
  1924. repeat:
  1925. /* First make sure individual recovery_offsets are correct */
  1926. list_for_each_entry(rdev, &mddev->disks, same_set) {
  1927. if (rdev->raid_disk >= 0 &&
  1928. mddev->delta_disks >= 0 &&
  1929. !test_bit(In_sync, &rdev->flags) &&
  1930. mddev->curr_resync_completed > rdev->recovery_offset)
  1931. rdev->recovery_offset = mddev->curr_resync_completed;
  1932. }
  1933. if (!mddev->persistent) {
  1934. clear_bit(MD_CHANGE_CLEAN, &mddev->flags);
  1935. clear_bit(MD_CHANGE_DEVS, &mddev->flags);
  1936. if (!mddev->external)
  1937. clear_bit(MD_CHANGE_PENDING, &mddev->flags);
  1938. wake_up(&mddev->sb_wait);
  1939. return;
  1940. }
  1941. spin_lock_irq(&mddev->write_lock);
  1942. mddev->utime = get_seconds();
  1943. if (test_and_clear_bit(MD_CHANGE_DEVS, &mddev->flags))
  1944. force_change = 1;
  1945. if (test_and_clear_bit(MD_CHANGE_CLEAN, &mddev->flags))
  1946. /* just a clean<-> dirty transition, possibly leave spares alone,
  1947. * though if events isn't the right even/odd, we will have to do
  1948. * spares after all
  1949. */
  1950. nospares = 1;
  1951. if (force_change)
  1952. nospares = 0;
  1953. if (mddev->degraded)
  1954. /* If the array is degraded, then skipping spares is both
  1955. * dangerous and fairly pointless.
  1956. * Dangerous because a device that was removed from the array
  1957. * might have a event_count that still looks up-to-date,
  1958. * so it can be re-added without a resync.
  1959. * Pointless because if there are any spares to skip,
  1960. * then a recovery will happen and soon that array won't
  1961. * be degraded any more and the spare can go back to sleep then.
  1962. */
  1963. nospares = 0;
  1964. sync_req = mddev->in_sync;
  1965. /* If this is just a dirty<->clean transition, and the array is clean
  1966. * and 'events' is odd, we can roll back to the previous clean state */
  1967. if (nospares
  1968. && (mddev->in_sync && mddev->recovery_cp == MaxSector)
  1969. && mddev->can_decrease_events
  1970. && mddev->events != 1) {
  1971. mddev->events--;
  1972. mddev->can_decrease_events = 0;
  1973. } else {
  1974. /* otherwise we have to go forward and ... */
  1975. mddev->events ++;
  1976. mddev->can_decrease_events = nospares;
  1977. }
  1978. if (!mddev->events) {
  1979. /*
  1980. * oops, this 64-bit counter should never wrap.
  1981. * Either we are in around ~1 trillion A.C., assuming
  1982. * 1 reboot per second, or we have a bug:
  1983. */
  1984. MD_BUG();
  1985. mddev->events --;
  1986. }
  1987. sync_sbs(mddev, nospares);
  1988. spin_unlock_irq(&mddev->write_lock);
  1989. dprintk(KERN_INFO
  1990. "md: updating %s RAID superblock on device (in sync %d)\n",
  1991. mdname(mddev),mddev->in_sync);
  1992. bitmap_update_sb(mddev->bitmap);
  1993. list_for_each_entry(rdev, &mddev->disks, same_set) {
  1994. char b[BDEVNAME_SIZE];
  1995. dprintk(KERN_INFO "md: ");
  1996. if (rdev->sb_loaded != 1)
  1997. continue; /* no noise on spare devices */
  1998. if (test_bit(Faulty, &rdev->flags))
  1999. dprintk("(skipping faulty ");
  2000. dprintk("%s ", bdevname(rdev->bdev,b));
  2001. if (!test_bit(Faulty, &rdev->flags)) {
  2002. md_super_write(mddev,rdev,
  2003. rdev->sb_start, rdev->sb_size,
  2004. rdev->sb_page);
  2005. dprintk(KERN_INFO "(write) %s's sb offset: %llu\n",
  2006. bdevname(rdev->bdev,b),
  2007. (unsigned long long)rdev->sb_start);
  2008. rdev->sb_events = mddev->events;
  2009. } else
  2010. dprintk(")\n");
  2011. if (mddev->level == LEVEL_MULTIPATH)
  2012. /* only need to write one superblock... */
  2013. break;
  2014. }
  2015. md_super_wait(mddev);
  2016. /* if there was a failure, MD_CHANGE_DEVS was set, and we re-write super */
  2017. spin_lock_irq(&mddev->write_lock);
  2018. if (mddev->in_sync != sync_req ||
  2019. test_bit(MD_CHANGE_DEVS, &mddev->flags)) {
  2020. /* have to write it out again */
  2021. spin_unlock_irq(&mddev->write_lock);
  2022. goto repeat;
  2023. }
  2024. clear_bit(MD_CHANGE_PENDING, &mddev->flags);
  2025. spin_unlock_irq(&mddev->write_lock);
  2026. wake_up(&mddev->sb_wait);
  2027. if (test_bit(MD_RECOVERY_RUNNING, &mddev->recovery))
  2028. sysfs_notify(&mddev->kobj, NULL, "sync_completed");
  2029. }
  2030. /* words written to sysfs files may, or may not, be \n terminated.
  2031. * We want to accept with case. For this we use cmd_match.
  2032. */
  2033. static int cmd_match(const char *cmd, const char *str)
  2034. {
  2035. /* See if cmd, written into a sysfs file, matches
  2036. * str. They must either be the same, or cmd can
  2037. * have a trailing newline
  2038. */
  2039. while (*cmd && *str && *cmd == *str) {
  2040. cmd++;
  2041. str++;
  2042. }
  2043. if (*cmd == '\n')
  2044. cmd++;
  2045. if (*str || *cmd)
  2046. return 0;
  2047. return 1;
  2048. }
  2049. struct rdev_sysfs_entry {
  2050. struct attribute attr;
  2051. ssize_t (*show)(mdk_rdev_t *, char *);
  2052. ssize_t (*store)(mdk_rdev_t *, const char *, size_t);
  2053. };
  2054. static ssize_t
  2055. state_show(mdk_rdev_t *rdev, char *page)
  2056. {
  2057. char *sep = "";
  2058. size_t len = 0;
  2059. if (test_bit(Faulty, &rdev->flags)) {
  2060. len+= sprintf(page+len, "%sfaulty",sep);
  2061. sep = ",";
  2062. }
  2063. if (test_bit(In_sync, &rdev->flags)) {
  2064. len += sprintf(page+len, "%sin_sync",sep);
  2065. sep = ",";
  2066. }
  2067. if (test_bit(WriteMostly, &rdev->flags)) {
  2068. len += sprintf(page+len, "%swrite_mostly",sep);
  2069. sep = ",";
  2070. }
  2071. if (test_bit(Blocked, &rdev->flags)) {
  2072. len += sprintf(page+len, "%sblocked", sep);
  2073. sep = ",";
  2074. }
  2075. if (!test_bit(Faulty, &rdev->flags) &&
  2076. !test_bit(In_sync, &rdev->flags)) {
  2077. len += sprintf(page+len, "%sspare", sep);
  2078. sep = ",";
  2079. }
  2080. return len+sprintf(page+len, "\n");
  2081. }
  2082. static ssize_t
  2083. state_store(mdk_rdev_t *rdev, const char *buf, size_t len)
  2084. {
  2085. /* can write
  2086. * faulty - simulates and error
  2087. * remove - disconnects the device
  2088. * writemostly - sets write_mostly
  2089. * -writemostly - clears write_mostly
  2090. * blocked - sets the Blocked flag
  2091. * -blocked - clears the Blocked flag
  2092. * insync - sets Insync providing device isn't active
  2093. */
  2094. int err = -EINVAL;
  2095. if (cmd_match(buf, "faulty") && rdev->mddev->pers) {
  2096. md_error(rdev->mddev, rdev);
  2097. err = 0;
  2098. } else if (cmd_match(buf, "remove")) {
  2099. if (rdev->raid_disk >= 0)
  2100. err = -EBUSY;
  2101. else {
  2102. mddev_t *mddev = rdev->mddev;
  2103. kick_rdev_from_array(rdev);
  2104. if (mddev->pers)
  2105. md_update_sb(mddev, 1);
  2106. md_new_event(mddev);
  2107. err = 0;
  2108. }
  2109. } else if (cmd_match(buf, "writemostly")) {
  2110. set_bit(WriteMostly, &rdev->flags);
  2111. err = 0;
  2112. } else if (cmd_match(buf, "-writemostly")) {
  2113. clear_bit(WriteMostly, &rdev->flags);
  2114. err = 0;
  2115. } else if (cmd_match(buf, "blocked")) {
  2116. set_bit(Blocked, &rdev->flags);
  2117. err = 0;
  2118. } else if (cmd_match(buf, "-blocked")) {
  2119. clear_bit(Blocked, &rdev->flags);
  2120. wake_up(&rdev->blocked_wait);
  2121. set_bit(MD_RECOVERY_NEEDED, &rdev->mddev->recovery);
  2122. md_wakeup_thread(rdev->mddev->thread);
  2123. err = 0;
  2124. } else if (cmd_match(buf, "insync") && rdev->raid_disk == -1) {
  2125. set_bit(In_sync, &rdev->flags);
  2126. err = 0;
  2127. }
  2128. if (!err)
  2129. sysfs_notify_dirent_safe(rdev->sysfs_state);
  2130. return err ? err : len;
  2131. }
  2132. static struct rdev_sysfs_entry rdev_state =
  2133. __ATTR(state, S_IRUGO|S_IWUSR, state_show, state_store);
  2134. static ssize_t
  2135. errors_show(mdk_rdev_t *rdev, char *page)
  2136. {
  2137. return sprintf(page, "%d\n", atomic_read(&rdev->corrected_errors));
  2138. }
  2139. static ssize_t
  2140. errors_store(mdk_rdev_t *rdev, const char *buf, size_t len)
  2141. {
  2142. char *e;
  2143. unsigned long n = simple_strtoul(buf, &e, 10);
  2144. if (*buf && (*e == 0 || *e == '\n')) {
  2145. atomic_set(&rdev->corrected_errors, n);
  2146. return len;
  2147. }
  2148. return -EINVAL;
  2149. }
  2150. static struct rdev_sysfs_entry rdev_errors =
  2151. __ATTR(errors, S_IRUGO|S_IWUSR, errors_show, errors_store);
  2152. static ssize_t
  2153. slot_show(mdk_rdev_t *rdev, char *page)
  2154. {
  2155. if (rdev->raid_disk < 0)
  2156. return sprintf(page, "none\n");
  2157. else
  2158. return sprintf(page, "%d\n", rdev->raid_disk);
  2159. }
  2160. static ssize_t
  2161. slot_store(mdk_rdev_t *rdev, const char *buf, size_t len)
  2162. {
  2163. char *e;
  2164. int err;
  2165. char nm[20];
  2166. int slot = simple_strtoul(buf, &e, 10);
  2167. if (strncmp(buf, "none", 4)==0)
  2168. slot = -1;
  2169. else if (e==buf || (*e && *e!= '\n'))
  2170. return -EINVAL;
  2171. if (rdev->mddev->pers && slot == -1) {
  2172. /* Setting 'slot' on an active array requires also
  2173. * updating the 'rd%d' link, and communicating
  2174. * with the personality with ->hot_*_disk.
  2175. * For now we only support removing
  2176. * failed/spare devices. This normally happens automatically,
  2177. * but not when the metadata is externally managed.
  2178. */
  2179. if (rdev->raid_disk == -1)
  2180. return -EEXIST;
  2181. /* personality does all needed checks */
  2182. if (rdev->mddev->pers->hot_remove_disk == NULL)
  2183. return -EINVAL;
  2184. err = rdev->mddev->pers->
  2185. hot_remove_disk(rdev->mddev, rdev->raid_disk);
  2186. if (err)
  2187. return err;
  2188. sprintf(nm, "rd%d", rdev->raid_disk);
  2189. sysfs_remove_link(&rdev->mddev->kobj, nm);
  2190. rdev->raid_disk = -1;
  2191. set_bit(MD_RECOVERY_NEEDED, &rdev->mddev->recovery);
  2192. md_wakeup_thread(rdev->mddev->thread);
  2193. } else if (rdev->mddev->pers) {
  2194. mdk_rdev_t *rdev2;
  2195. /* Activating a spare .. or possibly reactivating
  2196. * if we ever get bitmaps working here.
  2197. */
  2198. if (rdev->raid_disk != -1)
  2199. return -EBUSY;
  2200. if (test_bit(MD_RECOVERY_RUNNING, &rdev->mddev->recovery))
  2201. return -EBUSY;
  2202. if (rdev->mddev->pers->hot_add_disk == NULL)
  2203. return -EINVAL;
  2204. list_for_each_entry(rdev2, &rdev->mddev->disks, same_set)
  2205. if (rdev2->raid_disk == slot)
  2206. return -EEXIST;
  2207. if (slot >= rdev->mddev->raid_disks &&
  2208. slot >= rdev->mddev->raid_disks + rdev->mddev->delta_disks)
  2209. return -ENOSPC;
  2210. rdev->raid_disk = slot;
  2211. if (test_bit(In_sync, &rdev->flags))
  2212. rdev->saved_raid_disk = slot;
  2213. else
  2214. rdev->saved_raid_disk = -1;
  2215. err = rdev->mddev->pers->
  2216. hot_add_disk(rdev->mddev, rdev);
  2217. if (err) {
  2218. rdev->raid_disk = -1;
  2219. return err;
  2220. } else
  2221. sysfs_notify_dirent_safe(rdev->sysfs_state);
  2222. sprintf(nm, "rd%d", rdev->raid_disk);
  2223. if (sysfs_create_link(&rdev->mddev->kobj, &rdev->kobj, nm))
  2224. /* failure here is OK */;
  2225. /* don't wakeup anyone, leave that to userspace. */
  2226. } else {
  2227. if (slot >= rdev->mddev->raid_disks &&
  2228. slot >= rdev->mddev->raid_disks + rdev->mddev->delta_disks)
  2229. return -ENOSPC;
  2230. rdev->raid_disk = slot;
  2231. /* assume it is working */
  2232. clear_bit(Faulty, &rdev->flags);
  2233. clear_bit(WriteMostly, &rdev->flags);
  2234. set_bit(In_sync, &rdev->flags);
  2235. sysfs_notify_dirent_safe(rdev->sysfs_state);
  2236. }
  2237. return len;
  2238. }
  2239. static struct rdev_sysfs_entry rdev_slot =
  2240. __ATTR(slot, S_IRUGO|S_IWUSR, slot_show, slot_store);
  2241. static ssize_t
  2242. offset_show(mdk_rdev_t *rdev, char *page)
  2243. {
  2244. return sprintf(page, "%llu\n", (unsigned long long)rdev->data_offset);
  2245. }
  2246. static ssize_t
  2247. offset_store(mdk_rdev_t *rdev, const char *buf, size_t len)
  2248. {
  2249. char *e;
  2250. unsigned long long offset = simple_strtoull(buf, &e, 10);
  2251. if (e==buf || (*e && *e != '\n'))
  2252. return -EINVAL;
  2253. if (rdev->mddev->pers && rdev->raid_disk >= 0)
  2254. return -EBUSY;
  2255. if (rdev->sectors && rdev->mddev->external)
  2256. /* Must set offset before size, so overlap checks
  2257. * can be sane */
  2258. return -EBUSY;
  2259. rdev->data_offset = offset;
  2260. return len;
  2261. }
  2262. static struct rdev_sysfs_entry rdev_offset =
  2263. __ATTR(offset, S_IRUGO|S_IWUSR, offset_show, offset_store);
  2264. static ssize_t
  2265. rdev_size_show(mdk_rdev_t *rdev, char *page)
  2266. {
  2267. return sprintf(page, "%llu\n", (unsigned long long)rdev->sectors / 2);
  2268. }
  2269. static int overlaps(sector_t s1, sector_t l1, sector_t s2, sector_t l2)
  2270. {
  2271. /* check if two start/length pairs overlap */
  2272. if (s1+l1 <= s2)
  2273. return 0;
  2274. if (s2+l2 <= s1)
  2275. return 0;
  2276. return 1;
  2277. }
  2278. static int strict_blocks_to_sectors(const char *buf, sector_t *sectors)
  2279. {
  2280. unsigned long long blocks;
  2281. sector_t new;
  2282. if (strict_strtoull(buf, 10, &blocks) < 0)
  2283. return -EINVAL;
  2284. if (blocks & 1ULL << (8 * sizeof(blocks) - 1))
  2285. return -EINVAL; /* sector conversion overflow */
  2286. new = blocks * 2;
  2287. if (new != blocks * 2)
  2288. return -EINVAL; /* unsigned long long to sector_t overflow */
  2289. *sectors = new;
  2290. return 0;
  2291. }
  2292. static ssize_t
  2293. rdev_size_store(mdk_rdev_t *rdev, const char *buf, size_t len)
  2294. {
  2295. mddev_t *my_mddev = rdev->mddev;
  2296. sector_t oldsectors = rdev->sectors;
  2297. sector_t sectors;
  2298. if (strict_blocks_to_sectors(buf, &sectors) < 0)
  2299. return -EINVAL;
  2300. if (my_mddev->pers && rdev->raid_disk >= 0) {
  2301. if (my_mddev->persistent) {
  2302. sectors = super_types[my_mddev->major_version].
  2303. rdev_size_change(rdev, sectors);
  2304. if (!sectors)
  2305. return -EBUSY;
  2306. } else if (!sectors)
  2307. sectors = (i_size_read(rdev->bdev->bd_inode) >> 9) -
  2308. rdev->data_offset;
  2309. }
  2310. if (sectors < my_mddev->dev_sectors)
  2311. return -EINVAL; /* component must fit device */
  2312. rdev->sectors = sectors;
  2313. if (sectors > oldsectors && my_mddev->external) {
  2314. /* need to check that all other rdevs with the same ->bdev
  2315. * do not overlap. We need to unlock the mddev to avoid
  2316. * a deadlock. We have already changed rdev->sectors, and if
  2317. * we have to change it back, we will have the lock again.
  2318. */
  2319. mddev_t *mddev;
  2320. int overlap = 0;
  2321. struct list_head *tmp;
  2322. mddev_unlock(my_mddev);
  2323. for_each_mddev(mddev, tmp) {
  2324. mdk_rdev_t *rdev2;
  2325. mddev_lock(mddev);
  2326. list_for_each_entry(rdev2, &mddev->disks, same_set)
  2327. if (rdev->bdev == rdev2->bdev &&
  2328. rdev != rdev2 &&
  2329. overlaps(rdev->data_offset, rdev->sectors,
  2330. rdev2->data_offset,
  2331. rdev2->sectors)) {
  2332. overlap = 1;
  2333. break;
  2334. }
  2335. mddev_unlock(mddev);
  2336. if (overlap) {
  2337. mddev_put(mddev);
  2338. break;
  2339. }
  2340. }
  2341. mddev_lock(my_mddev);
  2342. if (overlap) {
  2343. /* Someone else could have slipped in a size
  2344. * change here, but doing so is just silly.
  2345. * We put oldsectors back because we *know* it is
  2346. * safe, and trust userspace not to race with
  2347. * itself
  2348. */
  2349. rdev->sectors = oldsectors;
  2350. return -EBUSY;
  2351. }
  2352. }
  2353. return len;
  2354. }
  2355. static struct rdev_sysfs_entry rdev_size =
  2356. __ATTR(size, S_IRUGO|S_IWUSR, rdev_size_show, rdev_size_store);
  2357. static ssize_t recovery_start_show(mdk_rdev_t *rdev, char *page)
  2358. {
  2359. unsigned long long recovery_start = rdev->recovery_offset;
  2360. if (test_bit(In_sync, &rdev->flags) ||
  2361. recovery_start == MaxSector)
  2362. return sprintf(page, "none\n");
  2363. return sprintf(page, "%llu\n", recovery_start);
  2364. }
  2365. static ssize_t recovery_start_store(mdk_rdev_t *rdev, const char *buf, size_t len)
  2366. {
  2367. unsigned long long recovery_start;
  2368. if (cmd_match(buf, "none"))
  2369. recovery_start = MaxSector;
  2370. else if (strict_strtoull(buf, 10, &recovery_start))
  2371. return -EINVAL;
  2372. if (rdev->mddev->pers &&
  2373. rdev->raid_disk >= 0)
  2374. return -EBUSY;
  2375. rdev->recovery_offset = recovery_start;
  2376. if (recovery_start == MaxSector)
  2377. set_bit(In_sync, &rdev->flags);
  2378. else
  2379. clear_bit(In_sync, &rdev->flags);
  2380. return len;
  2381. }
  2382. static struct rdev_sysfs_entry rdev_recovery_start =
  2383. __ATTR(recovery_start, S_IRUGO|S_IWUSR, recovery_start_show, recovery_start_store);
  2384. static struct attribute *rdev_default_attrs[] = {
  2385. &rdev_state.attr,
  2386. &rdev_errors.attr,
  2387. &rdev_slot.attr,
  2388. &rdev_offset.attr,
  2389. &rdev_size.attr,
  2390. &rdev_recovery_start.attr,
  2391. NULL,
  2392. };
  2393. static ssize_t
  2394. rdev_attr_show(struct kobject *kobj, struct attribute *attr, char *page)
  2395. {
  2396. struct rdev_sysfs_entry *entry = container_of(attr, struct rdev_sysfs_entry, attr);
  2397. mdk_rdev_t *rdev = container_of(kobj, mdk_rdev_t, kobj);
  2398. mddev_t *mddev = rdev->mddev;
  2399. ssize_t rv;
  2400. if (!entry->show)
  2401. return -EIO;
  2402. rv = mddev ? mddev_lock(mddev) : -EBUSY;
  2403. if (!rv) {
  2404. if (rdev->mddev == NULL)
  2405. rv = -EBUSY;
  2406. else
  2407. rv = entry->show(rdev, page);
  2408. mddev_unlock(mddev);
  2409. }
  2410. return rv;
  2411. }
  2412. static ssize_t
  2413. rdev_attr_store(struct kobject *kobj, struct attribute *attr,
  2414. const char *page, size_t length)
  2415. {
  2416. struct rdev_sysfs_entry *entry = container_of(attr, struct rdev_sysfs_entry, attr);
  2417. mdk_rdev_t *rdev = container_of(kobj, mdk_rdev_t, kobj);
  2418. ssize_t rv;
  2419. mddev_t *mddev = rdev->mddev;
  2420. if (!entry->store)
  2421. return -EIO;
  2422. if (!capable(CAP_SYS_ADMIN))
  2423. return -EACCES;
  2424. rv = mddev ? mddev_lock(mddev): -EBUSY;
  2425. if (!rv) {
  2426. if (rdev->mddev == NULL)
  2427. rv = -EBUSY;
  2428. else
  2429. rv = entry->store(rdev, page, length);
  2430. mddev_unlock(mddev);
  2431. }
  2432. return rv;
  2433. }
  2434. static void rdev_free(struct kobject *ko)
  2435. {
  2436. mdk_rdev_t *rdev = container_of(ko, mdk_rdev_t, kobj);
  2437. kfree(rdev);
  2438. }
  2439. static const struct sysfs_ops rdev_sysfs_ops = {
  2440. .show = rdev_attr_show,
  2441. .store = rdev_attr_store,
  2442. };
  2443. static struct kobj_type rdev_ktype = {
  2444. .release = rdev_free,
  2445. .sysfs_ops = &rdev_sysfs_ops,
  2446. .default_attrs = rdev_default_attrs,
  2447. };
  2448. void md_rdev_init(mdk_rdev_t *rdev)
  2449. {
  2450. rdev->desc_nr = -1;
  2451. rdev->saved_raid_disk = -1;
  2452. rdev->raid_disk = -1;
  2453. rdev->flags = 0;
  2454. rdev->data_offset = 0;
  2455. rdev->sb_events = 0;
  2456. rdev->last_read_error.tv_sec = 0;
  2457. rdev->last_read_error.tv_nsec = 0;
  2458. atomic_set(&rdev->nr_pending, 0);
  2459. atomic_set(&rdev->read_errors, 0);
  2460. atomic_set(&rdev->corrected_errors, 0);
  2461. INIT_LIST_HEAD(&rdev->same_set);
  2462. init_waitqueue_head(&rdev->blocked_wait);
  2463. }
  2464. EXPORT_SYMBOL_GPL(md_rdev_init);
  2465. /*
  2466. * Import a device. If 'super_format' >= 0, then sanity check the superblock
  2467. *
  2468. * mark the device faulty if:
  2469. *
  2470. * - the device is nonexistent (zero size)
  2471. * - the device has no valid superblock
  2472. *
  2473. * a faulty rdev _never_ has rdev->sb set.
  2474. */
  2475. static mdk_rdev_t *md_import_device(dev_t newdev, int super_format, int super_minor)
  2476. {
  2477. char b[BDEVNAME_SIZE];
  2478. int err;
  2479. mdk_rdev_t *rdev;
  2480. sector_t size;
  2481. rdev = kzalloc(sizeof(*rdev), GFP_KERNEL);
  2482. if (!rdev) {
  2483. printk(KERN_ERR "md: could not alloc mem for new device!\n");
  2484. return ERR_PTR(-ENOMEM);
  2485. }
  2486. md_rdev_init(rdev);
  2487. if ((err = alloc_disk_sb(rdev)))
  2488. goto abort_free;
  2489. err = lock_rdev(rdev, newdev, super_format == -2);
  2490. if (err)
  2491. goto abort_free;
  2492. kobject_init(&rdev->kobj, &rdev_ktype);
  2493. size = i_size_read(rdev->bdev->bd_inode) >> BLOCK_SIZE_BITS;
  2494. if (!size) {
  2495. printk(KERN_WARNING
  2496. "md: %s has zero or unknown size, marking faulty!\n",
  2497. bdevname(rdev->bdev,b));
  2498. err = -EINVAL;
  2499. goto abort_free;
  2500. }
  2501. if (super_format >= 0) {
  2502. err = super_types[super_format].
  2503. load_super(rdev, NULL, super_minor);
  2504. if (err == -EINVAL) {
  2505. printk(KERN_WARNING
  2506. "md: %s does not have a valid v%d.%d "
  2507. "superblock, not importing!\n",
  2508. bdevname(rdev->bdev,b),
  2509. super_format, super_minor);
  2510. goto abort_free;
  2511. }
  2512. if (err < 0) {
  2513. printk(KERN_WARNING
  2514. "md: could not read %s's sb, not importing!\n",
  2515. bdevname(rdev->bdev,b));
  2516. goto abort_free;
  2517. }
  2518. }
  2519. return rdev;
  2520. abort_free:
  2521. if (rdev->sb_page) {
  2522. if (rdev->bdev)
  2523. unlock_rdev(rdev);
  2524. free_disk_sb(rdev);
  2525. }
  2526. kfree(rdev);
  2527. return ERR_PTR(err);
  2528. }
  2529. /*
  2530. * Check a full RAID array for plausibility
  2531. */
  2532. static void analyze_sbs(mddev_t * mddev)
  2533. {
  2534. int i;
  2535. mdk_rdev_t *rdev, *freshest, *tmp;
  2536. char b[BDEVNAME_SIZE];
  2537. freshest = NULL;
  2538. rdev_for_each(rdev, tmp, mddev)
  2539. switch (super_types[mddev->major_version].
  2540. load_super(rdev, freshest, mddev->minor_version)) {
  2541. case 1:
  2542. freshest = rdev;
  2543. break;
  2544. case 0:
  2545. break;
  2546. default:
  2547. printk( KERN_ERR \
  2548. "md: fatal superblock inconsistency in %s"
  2549. " -- removing from array\n",
  2550. bdevname(rdev->bdev,b));
  2551. kick_rdev_from_array(rdev);
  2552. }
  2553. super_types[mddev->major_version].
  2554. validate_super(mddev, freshest);
  2555. i = 0;
  2556. rdev_for_each(rdev, tmp, mddev) {
  2557. if (mddev->max_disks &&
  2558. (rdev->desc_nr >= mddev->max_disks ||
  2559. i > mddev->max_disks)) {
  2560. printk(KERN_WARNING
  2561. "md: %s: %s: only %d devices permitted\n",
  2562. mdname(mddev), bdevname(rdev->bdev, b),
  2563. mddev->max_disks);
  2564. kick_rdev_from_array(rdev);
  2565. continue;
  2566. }
  2567. if (rdev != freshest)
  2568. if (super_types[mddev->major_version].
  2569. validate_super(mddev, rdev)) {
  2570. printk(KERN_WARNING "md: kicking non-fresh %s"
  2571. " from array!\n",
  2572. bdevname(rdev->bdev,b));
  2573. kick_rdev_from_array(rdev);
  2574. continue;
  2575. }
  2576. if (mddev->level == LEVEL_MULTIPATH) {
  2577. rdev->desc_nr = i++;
  2578. rdev->raid_disk = rdev->desc_nr;
  2579. set_bit(In_sync, &rdev->flags);
  2580. } else if (rdev->raid_disk >= (mddev->raid_disks - min(0, mddev->delta_disks))) {
  2581. rdev->raid_disk = -1;
  2582. clear_bit(In_sync, &rdev->flags);
  2583. }
  2584. }
  2585. }
  2586. /* Read a fixed-point number.
  2587. * Numbers in sysfs attributes should be in "standard" units where
  2588. * possible, so time should be in seconds.
  2589. * However we internally use a a much smaller unit such as
  2590. * milliseconds or jiffies.
  2591. * This function takes a decimal number with a possible fractional
  2592. * component, and produces an integer which is the result of
  2593. * multiplying that number by 10^'scale'.
  2594. * all without any floating-point arithmetic.
  2595. */
  2596. int strict_strtoul_scaled(const char *cp, unsigned long *res, int scale)
  2597. {
  2598. unsigned long result = 0;
  2599. long decimals = -1;
  2600. while (isdigit(*cp) || (*cp == '.' && decimals < 0)) {
  2601. if (*cp == '.')
  2602. decimals = 0;
  2603. else if (decimals < scale) {
  2604. unsigned int value;
  2605. value = *cp - '0';
  2606. result = result * 10 + value;
  2607. if (decimals >= 0)
  2608. decimals++;
  2609. }
  2610. cp++;
  2611. }
  2612. if (*cp == '\n')
  2613. cp++;
  2614. if (*cp)
  2615. return -EINVAL;
  2616. if (decimals < 0)
  2617. decimals = 0;
  2618. while (decimals < scale) {
  2619. result *= 10;
  2620. decimals ++;
  2621. }
  2622. *res = result;
  2623. return 0;
  2624. }
  2625. static void md_safemode_timeout(unsigned long data);
  2626. static ssize_t
  2627. safe_delay_show(mddev_t *mddev, char *page)
  2628. {
  2629. int msec = (mddev->safemode_delay*1000)/HZ;
  2630. return sprintf(page, "%d.%03d\n", msec/1000, msec%1000);
  2631. }
  2632. static ssize_t
  2633. safe_delay_store(mddev_t *mddev, const char *cbuf, size_t len)
  2634. {
  2635. unsigned long msec;
  2636. if (strict_strtoul_scaled(cbuf, &msec, 3) < 0)
  2637. return -EINVAL;
  2638. if (msec == 0)
  2639. mddev->safemode_delay = 0;
  2640. else {
  2641. unsigned long old_delay = mddev->safemode_delay;
  2642. mddev->safemode_delay = (msec*HZ)/1000;
  2643. if (mddev->safemode_delay == 0)
  2644. mddev->safemode_delay = 1;
  2645. if (mddev->safemode_delay < old_delay)
  2646. md_safemode_timeout((unsigned long)mddev);
  2647. }
  2648. return len;
  2649. }
  2650. static struct md_sysfs_entry md_safe_delay =
  2651. __ATTR(safe_mode_delay, S_IRUGO|S_IWUSR,safe_delay_show, safe_delay_store);
  2652. static ssize_t
  2653. level_show(mddev_t *mddev, char *page)
  2654. {
  2655. struct mdk_personality *p = mddev->pers;
  2656. if (p)
  2657. return sprintf(page, "%s\n", p->name);
  2658. else if (mddev->clevel[0])
  2659. return sprintf(page, "%s\n", mddev->clevel);
  2660. else if (mddev->level != LEVEL_NONE)
  2661. return sprintf(page, "%d\n", mddev->level);
  2662. else
  2663. return 0;
  2664. }
  2665. static ssize_t
  2666. level_store(mddev_t *mddev, const char *buf, size_t len)
  2667. {
  2668. char clevel[16];
  2669. ssize_t rv = len;
  2670. struct mdk_personality *pers;
  2671. long level;
  2672. void *priv;
  2673. mdk_rdev_t *rdev;
  2674. if (mddev->pers == NULL) {
  2675. if (len == 0)
  2676. return 0;
  2677. if (len >= sizeof(mddev->clevel))
  2678. return -ENOSPC;
  2679. strncpy(mddev->clevel, buf, len);
  2680. if (mddev->clevel[len-1] == '\n')
  2681. len--;
  2682. mddev->clevel[len] = 0;
  2683. mddev->level = LEVEL_NONE;
  2684. return rv;
  2685. }
  2686. /* request to change the personality. Need to ensure:
  2687. * - array is not engaged in resync/recovery/reshape
  2688. * - old personality can be suspended
  2689. * - new personality will access other array.
  2690. */
  2691. if (mddev->sync_thread ||
  2692. mddev->reshape_position != MaxSector ||
  2693. mddev->sysfs_active)
  2694. return -EBUSY;
  2695. if (!mddev->pers->quiesce) {
  2696. printk(KERN_WARNING "md: %s: %s does not support online personality change\n",
  2697. mdname(mddev), mddev->pers->name);
  2698. return -EINVAL;
  2699. }
  2700. /* Now find the new personality */
  2701. if (len == 0 || len >= sizeof(clevel))
  2702. return -EINVAL;
  2703. strncpy(clevel, buf, len);
  2704. if (clevel[len-1] == '\n')
  2705. len--;
  2706. clevel[len] = 0;
  2707. if (strict_strtol(clevel, 10, &level))
  2708. level = LEVEL_NONE;
  2709. if (request_module("md-%s", clevel) != 0)
  2710. request_module("md-level-%s", clevel);
  2711. spin_lock(&pers_lock);
  2712. pers = find_pers(level, clevel);
  2713. if (!pers || !try_module_get(pers->owner)) {
  2714. spin_unlock(&pers_lock);
  2715. printk(KERN_WARNING "md: personality %s not loaded\n", clevel);
  2716. return -EINVAL;
  2717. }
  2718. spin_unlock(&pers_lock);
  2719. if (pers == mddev->pers) {
  2720. /* Nothing to do! */
  2721. module_put(pers->owner);
  2722. return rv;
  2723. }
  2724. if (!pers->takeover) {
  2725. module_put(pers->owner);
  2726. printk(KERN_WARNING "md: %s: %s does not support personality takeover\n",
  2727. mdname(mddev), clevel);
  2728. return -EINVAL;
  2729. }
  2730. list_for_each_entry(rdev, &mddev->disks, same_set)
  2731. rdev->new_raid_disk = rdev->raid_disk;
  2732. /* ->takeover must set new_* and/or delta_disks
  2733. * if it succeeds, and may set them when it fails.
  2734. */
  2735. priv = pers->takeover(mddev);
  2736. if (IS_ERR(priv)) {
  2737. mddev->new_level = mddev->level;
  2738. mddev->new_layout = mddev->layout;
  2739. mddev->new_chunk_sectors = mddev->chunk_sectors;
  2740. mddev->raid_disks -= mddev->delta_disks;
  2741. mddev->delta_disks = 0;
  2742. module_put(pers->owner);
  2743. printk(KERN_WARNING "md: %s: %s would not accept array\n",
  2744. mdname(mddev), clevel);
  2745. return PTR_ERR(priv);
  2746. }
  2747. /* Looks like we have a winner */
  2748. mddev_suspend(mddev);
  2749. mddev->pers->stop(mddev);
  2750. if (mddev->pers->sync_request == NULL &&
  2751. pers->sync_request != NULL) {
  2752. /* need to add the md_redundancy_group */
  2753. if (sysfs_create_group(&mddev->kobj, &md_redundancy_group))
  2754. printk(KERN_WARNING
  2755. "md: cannot register extra attributes for %s\n",
  2756. mdname(mddev));
  2757. mddev->sysfs_action = sysfs_get_dirent(mddev->kobj.sd, NULL, "sync_action");
  2758. }
  2759. if (mddev->pers->sync_request != NULL &&
  2760. pers->sync_request == NULL) {
  2761. /* need to remove the md_redundancy_group */
  2762. if (mddev->to_remove == NULL)
  2763. mddev->to_remove = &md_redundancy_group;
  2764. }
  2765. if (mddev->pers->sync_request == NULL &&
  2766. mddev->external) {
  2767. /* We are converting from a no-redundancy array
  2768. * to a redundancy array and metadata is managed
  2769. * externally so we need to be sure that writes
  2770. * won't block due to a need to transition
  2771. * clean->dirty
  2772. * until external management is started.
  2773. */
  2774. mddev->in_sync = 0;
  2775. mddev->safemode_delay = 0;
  2776. mddev->safemode = 0;
  2777. }
  2778. list_for_each_entry(rdev, &mddev->disks, same_set) {
  2779. char nm[20];
  2780. if (rdev->raid_disk < 0)
  2781. continue;
  2782. if (rdev->new_raid_disk >= mddev->raid_disks)
  2783. rdev->new_raid_disk = -1;
  2784. if (rdev->new_raid_disk == rdev->raid_disk)
  2785. continue;
  2786. sprintf(nm, "rd%d", rdev->raid_disk);
  2787. sysfs_remove_link(&mddev->kobj, nm);
  2788. }
  2789. list_for_each_entry(rdev, &mddev->disks, same_set) {
  2790. if (rdev->raid_disk < 0)
  2791. continue;
  2792. if (rdev->new_raid_disk == rdev->raid_disk)
  2793. continue;
  2794. rdev->raid_disk = rdev->new_raid_disk;
  2795. if (rdev->raid_disk < 0)
  2796. clear_bit(In_sync, &rdev->flags);
  2797. else {
  2798. char nm[20];
  2799. sprintf(nm, "rd%d", rdev->raid_disk);
  2800. if(sysfs_create_link(&mddev->kobj, &rdev->kobj, nm))
  2801. printk("md: cannot register %s for %s after level change\n",
  2802. nm, mdname(mddev));
  2803. }
  2804. }
  2805. module_put(mddev->pers->owner);
  2806. mddev->pers = pers;
  2807. mddev->private = priv;
  2808. strlcpy(mddev->clevel, pers->name, sizeof(mddev->clevel));
  2809. mddev->level = mddev->new_level;
  2810. mddev->layout = mddev->new_layout;
  2811. mddev->chunk_sectors = mddev->new_chunk_sectors;
  2812. mddev->delta_disks = 0;
  2813. mddev->degraded = 0;
  2814. if (mddev->pers->sync_request == NULL) {
  2815. /* this is now an array without redundancy, so
  2816. * it must always be in_sync
  2817. */
  2818. mddev->in_sync = 1;
  2819. del_timer_sync(&mddev->safemode_timer);
  2820. }
  2821. pers->run(mddev);
  2822. mddev_resume(mddev);
  2823. set_bit(MD_CHANGE_DEVS, &mddev->flags);
  2824. set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
  2825. md_wakeup_thread(mddev->thread);
  2826. sysfs_notify(&mddev->kobj, NULL, "level");
  2827. md_new_event(mddev);
  2828. return rv;
  2829. }
  2830. static struct md_sysfs_entry md_level =
  2831. __ATTR(level, S_IRUGO|S_IWUSR, level_show, level_store);
  2832. static ssize_t
  2833. layout_show(mddev_t *mddev, char *page)
  2834. {
  2835. /* just a number, not meaningful for all levels */
  2836. if (mddev->reshape_position != MaxSector &&
  2837. mddev->layout != mddev->new_layout)
  2838. return sprintf(page, "%d (%d)\n",
  2839. mddev->new_layout, mddev->layout);
  2840. return sprintf(page, "%d\n", mddev->layout);
  2841. }
  2842. static ssize_t
  2843. layout_store(mddev_t *mddev, const char *buf, size_t len)
  2844. {
  2845. char *e;
  2846. unsigned long n = simple_strtoul(buf, &e, 10);
  2847. if (!*buf || (*e && *e != '\n'))
  2848. return -EINVAL;
  2849. if (mddev->pers) {
  2850. int err;
  2851. if (mddev->pers->check_reshape == NULL)
  2852. return -EBUSY;
  2853. mddev->new_layout = n;
  2854. err = mddev->pers->check_reshape(mddev);
  2855. if (err) {
  2856. mddev->new_layout = mddev->layout;
  2857. return err;
  2858. }
  2859. } else {
  2860. mddev->new_layout = n;
  2861. if (mddev->reshape_position == MaxSector)
  2862. mddev->layout = n;
  2863. }
  2864. return len;
  2865. }
  2866. static struct md_sysfs_entry md_layout =
  2867. __ATTR(layout, S_IRUGO|S_IWUSR, layout_show, layout_store);
  2868. static ssize_t
  2869. raid_disks_show(mddev_t *mddev, char *page)
  2870. {
  2871. if (mddev->raid_disks == 0)
  2872. return 0;
  2873. if (mddev->reshape_position != MaxSector &&
  2874. mddev->delta_disks != 0)
  2875. return sprintf(page, "%d (%d)\n", mddev->raid_disks,
  2876. mddev->raid_disks - mddev->delta_disks);
  2877. return sprintf(page, "%d\n", mddev->raid_disks);
  2878. }
  2879. static int update_raid_disks(mddev_t *mddev, int raid_disks);
  2880. static ssize_t
  2881. raid_disks_store(mddev_t *mddev, const char *buf, size_t len)
  2882. {
  2883. char *e;
  2884. int rv = 0;
  2885. unsigned long n = simple_strtoul(buf, &e, 10);
  2886. if (!*buf || (*e && *e != '\n'))
  2887. return -EINVAL;
  2888. if (mddev->pers)
  2889. rv = update_raid_disks(mddev, n);
  2890. else if (mddev->reshape_position != MaxSector) {
  2891. int olddisks = mddev->raid_disks - mddev->delta_disks;
  2892. mddev->delta_disks = n - olddisks;
  2893. mddev->raid_disks = n;
  2894. } else
  2895. mddev->raid_disks = n;
  2896. return rv ? rv : len;
  2897. }
  2898. static struct md_sysfs_entry md_raid_disks =
  2899. __ATTR(raid_disks, S_IRUGO|S_IWUSR, raid_disks_show, raid_disks_store);
  2900. static ssize_t
  2901. chunk_size_show(mddev_t *mddev, char *page)
  2902. {
  2903. if (mddev->reshape_position != MaxSector &&
  2904. mddev->chunk_sectors != mddev->new_chunk_sectors)
  2905. return sprintf(page, "%d (%d)\n",
  2906. mddev->new_chunk_sectors << 9,
  2907. mddev->chunk_sectors << 9);
  2908. return sprintf(page, "%d\n", mddev->chunk_sectors << 9);
  2909. }
  2910. static ssize_t
  2911. chunk_size_store(mddev_t *mddev, const char *buf, size_t len)
  2912. {
  2913. char *e;
  2914. unsigned long n = simple_strtoul(buf, &e, 10);
  2915. if (!*buf || (*e && *e != '\n'))
  2916. return -EINVAL;
  2917. if (mddev->pers) {
  2918. int err;
  2919. if (mddev->pers->check_reshape == NULL)
  2920. return -EBUSY;
  2921. mddev->new_chunk_sectors = n >> 9;
  2922. err = mddev->pers->check_reshape(mddev);
  2923. if (err) {
  2924. mddev->new_chunk_sectors = mddev->chunk_sectors;
  2925. return err;
  2926. }
  2927. } else {
  2928. mddev->new_chunk_sectors = n >> 9;
  2929. if (mddev->reshape_position == MaxSector)
  2930. mddev->chunk_sectors = n >> 9;
  2931. }
  2932. return len;
  2933. }
  2934. static struct md_sysfs_entry md_chunk_size =
  2935. __ATTR(chunk_size, S_IRUGO|S_IWUSR, chunk_size_show, chunk_size_store);
  2936. static ssize_t
  2937. resync_start_show(mddev_t *mddev, char *page)
  2938. {
  2939. if (mddev->recovery_cp == MaxSector)
  2940. return sprintf(page, "none\n");
  2941. return sprintf(page, "%llu\n", (unsigned long long)mddev->recovery_cp);
  2942. }
  2943. static ssize_t
  2944. resync_start_store(mddev_t *mddev, const char *buf, size_t len)
  2945. {
  2946. char *e;
  2947. unsigned long long n = simple_strtoull(buf, &e, 10);
  2948. if (mddev->pers && !test_bit(MD_RECOVERY_FROZEN, &mddev->recovery))
  2949. return -EBUSY;
  2950. if (cmd_match(buf, "none"))
  2951. n = MaxSector;
  2952. else if (!*buf || (*e && *e != '\n'))
  2953. return -EINVAL;
  2954. mddev->recovery_cp = n;
  2955. return len;
  2956. }
  2957. static struct md_sysfs_entry md_resync_start =
  2958. __ATTR(resync_start, S_IRUGO|S_IWUSR, resync_start_show, resync_start_store);
  2959. /*
  2960. * The array state can be:
  2961. *
  2962. * clear
  2963. * No devices, no size, no level
  2964. * Equivalent to STOP_ARRAY ioctl
  2965. * inactive
  2966. * May have some settings, but array is not active
  2967. * all IO results in error
  2968. * When written, doesn't tear down array, but just stops it
  2969. * suspended (not supported yet)
  2970. * All IO requests will block. The array can be reconfigured.
  2971. * Writing this, if accepted, will block until array is quiescent
  2972. * readonly
  2973. * no resync can happen. no superblocks get written.
  2974. * write requests fail
  2975. * read-auto
  2976. * like readonly, but behaves like 'clean' on a write request.
  2977. *
  2978. * clean - no pending writes, but otherwise active.
  2979. * When written to inactive array, starts without resync
  2980. * If a write request arrives then
  2981. * if metadata is known, mark 'dirty' and switch to 'active'.
  2982. * if not known, block and switch to write-pending
  2983. * If written to an active array that has pending writes, then fails.
  2984. * active
  2985. * fully active: IO and resync can be happening.
  2986. * When written to inactive array, starts with resync
  2987. *
  2988. * write-pending
  2989. * clean, but writes are blocked waiting for 'active' to be written.
  2990. *
  2991. * active-idle
  2992. * like active, but no writes have been seen for a while (100msec).
  2993. *
  2994. */
  2995. enum array_state { clear, inactive, suspended, readonly, read_auto, clean, active,
  2996. write_pending, active_idle, bad_word};
  2997. static char *array_states[] = {
  2998. "clear", "inactive", "suspended", "readonly", "read-auto", "clean", "active",
  2999. "write-pending", "active-idle", NULL };
  3000. static int match_word(const char *word, char **list)
  3001. {
  3002. int n;
  3003. for (n=0; list[n]; n++)
  3004. if (cmd_match(word, list[n]))
  3005. break;
  3006. return n;
  3007. }
  3008. static ssize_t
  3009. array_state_show(mddev_t *mddev, char *page)
  3010. {
  3011. enum array_state st = inactive;
  3012. if (mddev->pers)
  3013. switch(mddev->ro) {
  3014. case 1:
  3015. st = readonly;
  3016. break;
  3017. case 2:
  3018. st = read_auto;
  3019. break;
  3020. case 0:
  3021. if (mddev->in_sync)
  3022. st = clean;
  3023. else if (test_bit(MD_CHANGE_PENDING, &mddev->flags))
  3024. st = write_pending;
  3025. else if (mddev->safemode)
  3026. st = active_idle;
  3027. else
  3028. st = active;
  3029. }
  3030. else {
  3031. if (list_empty(&mddev->disks) &&
  3032. mddev->raid_disks == 0 &&
  3033. mddev->dev_sectors == 0)
  3034. st = clear;
  3035. else
  3036. st = inactive;
  3037. }
  3038. return sprintf(page, "%s\n", array_states[st]);
  3039. }
  3040. static int do_md_stop(mddev_t * mddev, int ro, int is_open);
  3041. static int md_set_readonly(mddev_t * mddev, int is_open);
  3042. static int do_md_run(mddev_t * mddev);
  3043. static int restart_array(mddev_t *mddev);
  3044. static ssize_t
  3045. array_state_store(mddev_t *mddev, const char *buf, size_t len)
  3046. {
  3047. int err = -EINVAL;
  3048. enum array_state st = match_word(buf, array_states);
  3049. switch(st) {
  3050. case bad_word:
  3051. break;
  3052. case clear:
  3053. /* stopping an active array */
  3054. if (atomic_read(&mddev->openers) > 0)
  3055. return -EBUSY;
  3056. err = do_md_stop(mddev, 0, 0);
  3057. break;
  3058. case inactive:
  3059. /* stopping an active array */
  3060. if (mddev->pers) {
  3061. if (atomic_read(&mddev->openers) > 0)
  3062. return -EBUSY;
  3063. err = do_md_stop(mddev, 2, 0);
  3064. } else
  3065. err = 0; /* already inactive */
  3066. break;
  3067. case suspended:
  3068. break; /* not supported yet */
  3069. case readonly:
  3070. if (mddev->pers)
  3071. err = md_set_readonly(mddev, 0);
  3072. else {
  3073. mddev->ro = 1;
  3074. set_disk_ro(mddev->gendisk, 1);
  3075. err = do_md_run(mddev);
  3076. }
  3077. break;
  3078. case read_auto:
  3079. if (mddev->pers) {
  3080. if (mddev->ro == 0)
  3081. err = md_set_readonly(mddev, 0);
  3082. else if (mddev->ro == 1)
  3083. err = restart_array(mddev);
  3084. if (err == 0) {
  3085. mddev->ro = 2;
  3086. set_disk_ro(mddev->gendisk, 0);
  3087. }
  3088. } else {
  3089. mddev->ro = 2;
  3090. err = do_md_run(mddev);
  3091. }
  3092. break;
  3093. case clean:
  3094. if (mddev->pers) {
  3095. restart_array(mddev);
  3096. spin_lock_irq(&mddev->write_lock);
  3097. if (atomic_read(&mddev->writes_pending) == 0) {
  3098. if (mddev->in_sync == 0) {
  3099. mddev->in_sync = 1;
  3100. if (mddev->safemode == 1)
  3101. mddev->safemode = 0;
  3102. set_bit(MD_CHANGE_CLEAN, &mddev->flags);
  3103. }
  3104. err = 0;
  3105. } else
  3106. err = -EBUSY;
  3107. spin_unlock_irq(&mddev->write_lock);
  3108. } else
  3109. err = -EINVAL;
  3110. break;
  3111. case active:
  3112. if (mddev->pers) {
  3113. restart_array(mddev);
  3114. clear_bit(MD_CHANGE_PENDING, &mddev->flags);
  3115. wake_up(&mddev->sb_wait);
  3116. err = 0;
  3117. } else {
  3118. mddev->ro = 0;
  3119. set_disk_ro(mddev->gendisk, 0);
  3120. err = do_md_run(mddev);
  3121. }
  3122. break;
  3123. case write_pending:
  3124. case active_idle:
  3125. /* these cannot be set */
  3126. break;
  3127. }
  3128. if (err)
  3129. return err;
  3130. else {
  3131. sysfs_notify_dirent_safe(mddev->sysfs_state);
  3132. return len;
  3133. }
  3134. }
  3135. static struct md_sysfs_entry md_array_state =
  3136. __ATTR(array_state, S_IRUGO|S_IWUSR, array_state_show, array_state_store);
  3137. static ssize_t
  3138. max_corrected_read_errors_show(mddev_t *mddev, char *page) {
  3139. return sprintf(page, "%d\n",
  3140. atomic_read(&mddev->max_corr_read_errors));
  3141. }
  3142. static ssize_t
  3143. max_corrected_read_errors_store(mddev_t *mddev, const char *buf, size_t len)
  3144. {
  3145. char *e;
  3146. unsigned long n = simple_strtoul(buf, &e, 10);
  3147. if (*buf && (*e == 0 || *e == '\n')) {
  3148. atomic_set(&mddev->max_corr_read_errors, n);
  3149. return len;
  3150. }
  3151. return -EINVAL;
  3152. }
  3153. static struct md_sysfs_entry max_corr_read_errors =
  3154. __ATTR(max_read_errors, S_IRUGO|S_IWUSR, max_corrected_read_errors_show,
  3155. max_corrected_read_errors_store);
  3156. static ssize_t
  3157. null_show(mddev_t *mddev, char *page)
  3158. {
  3159. return -EINVAL;
  3160. }
  3161. static ssize_t
  3162. new_dev_store(mddev_t *mddev, const char *buf, size_t len)
  3163. {
  3164. /* buf must be %d:%d\n? giving major and minor numbers */
  3165. /* The new device is added to the array.
  3166. * If the array has a persistent superblock, we read the
  3167. * superblock to initialise info and check validity.
  3168. * Otherwise, only checking done is that in bind_rdev_to_array,
  3169. * which mainly checks size.
  3170. */
  3171. char *e;
  3172. int major = simple_strtoul(buf, &e, 10);
  3173. int minor;
  3174. dev_t dev;
  3175. mdk_rdev_t *rdev;
  3176. int err;
  3177. if (!*buf || *e != ':' || !e[1] || e[1] == '\n')
  3178. return -EINVAL;
  3179. minor = simple_strtoul(e+1, &e, 10);
  3180. if (*e && *e != '\n')
  3181. return -EINVAL;
  3182. dev = MKDEV(major, minor);
  3183. if (major != MAJOR(dev) ||
  3184. minor != MINOR(dev))
  3185. return -EOVERFLOW;
  3186. if (mddev->persistent) {
  3187. rdev = md_import_device(dev, mddev->major_version,
  3188. mddev->minor_version);
  3189. if (!IS_ERR(rdev) && !list_empty(&mddev->disks)) {
  3190. mdk_rdev_t *rdev0 = list_entry(mddev->disks.next,
  3191. mdk_rdev_t, same_set);
  3192. err = super_types[mddev->major_version]
  3193. .load_super(rdev, rdev0, mddev->minor_version);
  3194. if (err < 0)
  3195. goto out;
  3196. }
  3197. } else if (mddev->external)
  3198. rdev = md_import_device(dev, -2, -1);
  3199. else
  3200. rdev = md_import_device(dev, -1, -1);
  3201. if (IS_ERR(rdev))
  3202. return PTR_ERR(rdev);
  3203. err = bind_rdev_to_array(rdev, mddev);
  3204. out:
  3205. if (err)
  3206. export_rdev(rdev);
  3207. return err ? err : len;
  3208. }
  3209. static struct md_sysfs_entry md_new_device =
  3210. __ATTR(new_dev, S_IWUSR, null_show, new_dev_store);
  3211. static ssize_t
  3212. bitmap_store(mddev_t *mddev, const char *buf, size_t len)
  3213. {
  3214. char *end;
  3215. unsigned long chunk, end_chunk;
  3216. if (!mddev->bitmap)
  3217. goto out;
  3218. /* buf should be <chunk> <chunk> ... or <chunk>-<chunk> ... (range) */
  3219. while (*buf) {
  3220. chunk = end_chunk = simple_strtoul(buf, &end, 0);
  3221. if (buf == end) break;
  3222. if (*end == '-') { /* range */
  3223. buf = end + 1;
  3224. end_chunk = simple_strtoul(buf, &end, 0);
  3225. if (buf == end) break;
  3226. }
  3227. if (*end && !isspace(*end)) break;
  3228. bitmap_dirty_bits(mddev->bitmap, chunk, end_chunk);
  3229. buf = skip_spaces(end);
  3230. }
  3231. bitmap_unplug(mddev->bitmap); /* flush the bits to disk */
  3232. out:
  3233. return len;
  3234. }
  3235. static struct md_sysfs_entry md_bitmap =
  3236. __ATTR(bitmap_set_bits, S_IWUSR, null_show, bitmap_store);
  3237. static ssize_t
  3238. size_show(mddev_t *mddev, char *page)
  3239. {
  3240. return sprintf(page, "%llu\n",
  3241. (unsigned long long)mddev->dev_sectors / 2);
  3242. }
  3243. static int update_size(mddev_t *mddev, sector_t num_sectors);
  3244. static ssize_t
  3245. size_store(mddev_t *mddev, const char *buf, size_t len)
  3246. {
  3247. /* If array is inactive, we can reduce the component size, but
  3248. * not increase it (except from 0).
  3249. * If array is active, we can try an on-line resize
  3250. */
  3251. sector_t sectors;
  3252. int err = strict_blocks_to_sectors(buf, &sectors);
  3253. if (err < 0)
  3254. return err;
  3255. if (mddev->pers) {
  3256. err = update_size(mddev, sectors);
  3257. md_update_sb(mddev, 1);
  3258. } else {
  3259. if (mddev->dev_sectors == 0 ||
  3260. mddev->dev_sectors > sectors)
  3261. mddev->dev_sectors = sectors;
  3262. else
  3263. err = -ENOSPC;
  3264. }
  3265. return err ? err : len;
  3266. }
  3267. static struct md_sysfs_entry md_size =
  3268. __ATTR(component_size, S_IRUGO|S_IWUSR, size_show, size_store);
  3269. /* Metdata version.
  3270. * This is one of
  3271. * 'none' for arrays with no metadata (good luck...)
  3272. * 'external' for arrays with externally managed metadata,
  3273. * or N.M for internally known formats
  3274. */
  3275. static ssize_t
  3276. metadata_show(mddev_t *mddev, char *page)
  3277. {
  3278. if (mddev->persistent)
  3279. return sprintf(page, "%d.%d\n",
  3280. mddev->major_version, mddev->minor_version);
  3281. else if (mddev->external)
  3282. return sprintf(page, "external:%s\n", mddev->metadata_type);
  3283. else
  3284. return sprintf(page, "none\n");
  3285. }
  3286. static ssize_t
  3287. metadata_store(mddev_t *mddev, const char *buf, size_t len)
  3288. {
  3289. int major, minor;
  3290. char *e;
  3291. /* Changing the details of 'external' metadata is
  3292. * always permitted. Otherwise there must be
  3293. * no devices attached to the array.
  3294. */
  3295. if (mddev->external && strncmp(buf, "external:", 9) == 0)
  3296. ;
  3297. else if (!list_empty(&mddev->disks))
  3298. return -EBUSY;
  3299. if (cmd_match(buf, "none")) {
  3300. mddev->persistent = 0;
  3301. mddev->external = 0;
  3302. mddev->major_version = 0;
  3303. mddev->minor_version = 90;
  3304. return len;
  3305. }
  3306. if (strncmp(buf, "external:", 9) == 0) {
  3307. size_t namelen = len-9;
  3308. if (namelen >= sizeof(mddev->metadata_type))
  3309. namelen = sizeof(mddev->metadata_type)-1;
  3310. strncpy(mddev->metadata_type, buf+9, namelen);
  3311. mddev->metadata_type[namelen] = 0;
  3312. if (namelen && mddev->metadata_type[namelen-1] == '\n')
  3313. mddev->metadata_type[--namelen] = 0;
  3314. mddev->persistent = 0;
  3315. mddev->external = 1;
  3316. mddev->major_version = 0;
  3317. mddev->minor_version = 90;
  3318. return len;
  3319. }
  3320. major = simple_strtoul(buf, &e, 10);
  3321. if (e==buf || *e != '.')
  3322. return -EINVAL;
  3323. buf = e+1;
  3324. minor = simple_strtoul(buf, &e, 10);
  3325. if (e==buf || (*e && *e != '\n') )
  3326. return -EINVAL;
  3327. if (major >= ARRAY_SIZE(super_types) || super_types[major].name == NULL)
  3328. return -ENOENT;
  3329. mddev->major_version = major;
  3330. mddev->minor_version = minor;
  3331. mddev->persistent = 1;
  3332. mddev->external = 0;
  3333. return len;
  3334. }
  3335. static struct md_sysfs_entry md_metadata =
  3336. __ATTR(metadata_version, S_IRUGO|S_IWUSR, metadata_show, metadata_store);
  3337. static ssize_t
  3338. action_show(mddev_t *mddev, char *page)
  3339. {
  3340. char *type = "idle";
  3341. if (test_bit(MD_RECOVERY_FROZEN, &mddev->recovery))
  3342. type = "frozen";
  3343. else if (test_bit(MD_RECOVERY_RUNNING, &mddev->recovery) ||
  3344. (!mddev->ro && test_bit(MD_RECOVERY_NEEDED, &mddev->recovery))) {
  3345. if (test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery))
  3346. type = "reshape";
  3347. else if (test_bit(MD_RECOVERY_SYNC, &mddev->recovery)) {
  3348. if (!test_bit(MD_RECOVERY_REQUESTED, &mddev->recovery))
  3349. type = "resync";
  3350. else if (test_bit(MD_RECOVERY_CHECK, &mddev->recovery))
  3351. type = "check";
  3352. else
  3353. type = "repair";
  3354. } else if (test_bit(MD_RECOVERY_RECOVER, &mddev->recovery))
  3355. type = "recover";
  3356. }
  3357. return sprintf(page, "%s\n", type);
  3358. }
  3359. static void reap_sync_thread(mddev_t *mddev);
  3360. static ssize_t
  3361. action_store(mddev_t *mddev, const char *page, size_t len)
  3362. {
  3363. if (!mddev->pers || !mddev->pers->sync_request)
  3364. return -EINVAL;
  3365. if (cmd_match(page, "frozen"))
  3366. set_bit(MD_RECOVERY_FROZEN, &mddev->recovery);
  3367. else
  3368. clear_bit(MD_RECOVERY_FROZEN, &mddev->recovery);
  3369. if (cmd_match(page, "idle") || cmd_match(page, "frozen")) {
  3370. if (mddev->sync_thread) {
  3371. set_bit(MD_RECOVERY_INTR, &mddev->recovery);
  3372. reap_sync_thread(mddev);
  3373. }
  3374. } else if (test_bit(MD_RECOVERY_RUNNING, &mddev->recovery) ||
  3375. test_bit(MD_RECOVERY_NEEDED, &mddev->recovery))
  3376. return -EBUSY;
  3377. else if (cmd_match(page, "resync"))
  3378. set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
  3379. else if (cmd_match(page, "recover")) {
  3380. set_bit(MD_RECOVERY_RECOVER, &mddev->recovery);
  3381. set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
  3382. } else if (cmd_match(page, "reshape")) {
  3383. int err;
  3384. if (mddev->pers->start_reshape == NULL)
  3385. return -EINVAL;
  3386. err = mddev->pers->start_reshape(mddev);
  3387. if (err)
  3388. return err;
  3389. sysfs_notify(&mddev->kobj, NULL, "degraded");
  3390. } else {
  3391. if (cmd_match(page, "check"))
  3392. set_bit(MD_RECOVERY_CHECK, &mddev->recovery);
  3393. else if (!cmd_match(page, "repair"))
  3394. return -EINVAL;
  3395. set_bit(MD_RECOVERY_REQUESTED, &mddev->recovery);
  3396. set_bit(MD_RECOVERY_SYNC, &mddev->recovery);
  3397. }
  3398. set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
  3399. md_wakeup_thread(mddev->thread);
  3400. sysfs_notify_dirent_safe(mddev->sysfs_action);
  3401. return len;
  3402. }
  3403. static ssize_t
  3404. mismatch_cnt_show(mddev_t *mddev, char *page)
  3405. {
  3406. return sprintf(page, "%llu\n",
  3407. (unsigned long long) mddev->resync_mismatches);
  3408. }
  3409. static struct md_sysfs_entry md_scan_mode =
  3410. __ATTR(sync_action, S_IRUGO|S_IWUSR, action_show, action_store);
  3411. static struct md_sysfs_entry md_mismatches = __ATTR_RO(mismatch_cnt);
  3412. static ssize_t
  3413. sync_min_show(mddev_t *mddev, char *page)
  3414. {
  3415. return sprintf(page, "%d (%s)\n", speed_min(mddev),
  3416. mddev->sync_speed_min ? "local": "system");
  3417. }
  3418. static ssize_t
  3419. sync_min_store(mddev_t *mddev, const char *buf, size_t len)
  3420. {
  3421. int min;
  3422. char *e;
  3423. if (strncmp(buf, "system", 6)==0) {
  3424. mddev->sync_speed_min = 0;
  3425. return len;
  3426. }
  3427. min = simple_strtoul(buf, &e, 10);
  3428. if (buf == e || (*e && *e != '\n') || min <= 0)
  3429. return -EINVAL;
  3430. mddev->sync_speed_min = min;
  3431. return len;
  3432. }
  3433. static struct md_sysfs_entry md_sync_min =
  3434. __ATTR(sync_speed_min, S_IRUGO|S_IWUSR, sync_min_show, sync_min_store);
  3435. static ssize_t
  3436. sync_max_show(mddev_t *mddev, char *page)
  3437. {
  3438. return sprintf(page, "%d (%s)\n", speed_max(mddev),
  3439. mddev->sync_speed_max ? "local": "system");
  3440. }
  3441. static ssize_t
  3442. sync_max_store(mddev_t *mddev, const char *buf, size_t len)
  3443. {
  3444. int max;
  3445. char *e;
  3446. if (strncmp(buf, "system", 6)==0) {
  3447. mddev->sync_speed_max = 0;
  3448. return len;
  3449. }
  3450. max = simple_strtoul(buf, &e, 10);
  3451. if (buf == e || (*e && *e != '\n') || max <= 0)
  3452. return -EINVAL;
  3453. mddev->sync_speed_max = max;
  3454. return len;
  3455. }
  3456. static struct md_sysfs_entry md_sync_max =
  3457. __ATTR(sync_speed_max, S_IRUGO|S_IWUSR, sync_max_show, sync_max_store);
  3458. static ssize_t
  3459. degraded_show(mddev_t *mddev, char *page)
  3460. {
  3461. return sprintf(page, "%d\n", mddev->degraded);
  3462. }
  3463. static struct md_sysfs_entry md_degraded = __ATTR_RO(degraded);
  3464. static ssize_t
  3465. sync_force_parallel_show(mddev_t *mddev, char *page)
  3466. {
  3467. return sprintf(page, "%d\n", mddev->parallel_resync);
  3468. }
  3469. static ssize_t
  3470. sync_force_parallel_store(mddev_t *mddev, const char *buf, size_t len)
  3471. {
  3472. long n;
  3473. if (strict_strtol(buf, 10, &n))
  3474. return -EINVAL;
  3475. if (n != 0 && n != 1)
  3476. return -EINVAL;
  3477. mddev->parallel_resync = n;
  3478. if (mddev->sync_thread)
  3479. wake_up(&resync_wait);
  3480. return len;
  3481. }
  3482. /* force parallel resync, even with shared block devices */
  3483. static struct md_sysfs_entry md_sync_force_parallel =
  3484. __ATTR(sync_force_parallel, S_IRUGO|S_IWUSR,
  3485. sync_force_parallel_show, sync_force_parallel_store);
  3486. static ssize_t
  3487. sync_speed_show(mddev_t *mddev, char *page)
  3488. {
  3489. unsigned long resync, dt, db;
  3490. if (mddev->curr_resync == 0)
  3491. return sprintf(page, "none\n");
  3492. resync = mddev->curr_mark_cnt - atomic_read(&mddev->recovery_active);
  3493. dt = (jiffies - mddev->resync_mark) / HZ;
  3494. if (!dt) dt++;
  3495. db = resync - mddev->resync_mark_cnt;
  3496. return sprintf(page, "%lu\n", db/dt/2); /* K/sec */
  3497. }
  3498. static struct md_sysfs_entry md_sync_speed = __ATTR_RO(sync_speed);
  3499. static ssize_t
  3500. sync_completed_show(mddev_t *mddev, char *page)
  3501. {
  3502. unsigned long long max_sectors, resync;
  3503. if (!test_bit(MD_RECOVERY_RUNNING, &mddev->recovery))
  3504. return sprintf(page, "none\n");
  3505. if (test_bit(MD_RECOVERY_SYNC, &mddev->recovery))
  3506. max_sectors = mddev->resync_max_sectors;
  3507. else
  3508. max_sectors = mddev->dev_sectors;
  3509. resync = mddev->curr_resync_completed;
  3510. return sprintf(page, "%llu / %llu\n", resync, max_sectors);
  3511. }
  3512. static struct md_sysfs_entry md_sync_completed = __ATTR_RO(sync_completed);
  3513. static ssize_t
  3514. min_sync_show(mddev_t *mddev, char *page)
  3515. {
  3516. return sprintf(page, "%llu\n",
  3517. (unsigned long long)mddev->resync_min);
  3518. }
  3519. static ssize_t
  3520. min_sync_store(mddev_t *mddev, const char *buf, size_t len)
  3521. {
  3522. unsigned long long min;
  3523. if (strict_strtoull(buf, 10, &min))
  3524. return -EINVAL;
  3525. if (min > mddev->resync_max)
  3526. return -EINVAL;
  3527. if (test_bit(MD_RECOVERY_RUNNING, &mddev->recovery))
  3528. return -EBUSY;
  3529. /* Must be a multiple of chunk_size */
  3530. if (mddev->chunk_sectors) {
  3531. sector_t temp = min;
  3532. if (sector_div(temp, mddev->chunk_sectors))
  3533. return -EINVAL;
  3534. }
  3535. mddev->resync_min = min;
  3536. return len;
  3537. }
  3538. static struct md_sysfs_entry md_min_sync =
  3539. __ATTR(sync_min, S_IRUGO|S_IWUSR, min_sync_show, min_sync_store);
  3540. static ssize_t
  3541. max_sync_show(mddev_t *mddev, char *page)
  3542. {
  3543. if (mddev->resync_max == MaxSector)
  3544. return sprintf(page, "max\n");
  3545. else
  3546. return sprintf(page, "%llu\n",
  3547. (unsigned long long)mddev->resync_max);
  3548. }
  3549. static ssize_t
  3550. max_sync_store(mddev_t *mddev, const char *buf, size_t len)
  3551. {
  3552. if (strncmp(buf, "max", 3) == 0)
  3553. mddev->resync_max = MaxSector;
  3554. else {
  3555. unsigned long long max;
  3556. if (strict_strtoull(buf, 10, &max))
  3557. return -EINVAL;
  3558. if (max < mddev->resync_min)
  3559. return -EINVAL;
  3560. if (max < mddev->resync_max &&
  3561. mddev->ro == 0 &&
  3562. test_bit(MD_RECOVERY_RUNNING, &mddev->recovery))
  3563. return -EBUSY;
  3564. /* Must be a multiple of chunk_size */
  3565. if (mddev->chunk_sectors) {
  3566. sector_t temp = max;
  3567. if (sector_div(temp, mddev->chunk_sectors))
  3568. return -EINVAL;
  3569. }
  3570. mddev->resync_max = max;
  3571. }
  3572. wake_up(&mddev->recovery_wait);
  3573. return len;
  3574. }
  3575. static struct md_sysfs_entry md_max_sync =
  3576. __ATTR(sync_max, S_IRUGO|S_IWUSR, max_sync_show, max_sync_store);
  3577. static ssize_t
  3578. suspend_lo_show(mddev_t *mddev, char *page)
  3579. {
  3580. return sprintf(page, "%llu\n", (unsigned long long)mddev->suspend_lo);
  3581. }
  3582. static ssize_t
  3583. suspend_lo_store(mddev_t *mddev, const char *buf, size_t len)
  3584. {
  3585. char *e;
  3586. unsigned long long new = simple_strtoull(buf, &e, 10);
  3587. unsigned long long old = mddev->suspend_lo;
  3588. if (mddev->pers == NULL ||
  3589. mddev->pers->quiesce == NULL)
  3590. return -EINVAL;
  3591. if (buf == e || (*e && *e != '\n'))
  3592. return -EINVAL;
  3593. mddev->suspend_lo = new;
  3594. if (new >= old)
  3595. /* Shrinking suspended region */
  3596. mddev->pers->quiesce(mddev, 2);
  3597. else {
  3598. /* Expanding suspended region - need to wait */
  3599. mddev->pers->quiesce(mddev, 1);
  3600. mddev->pers->quiesce(mddev, 0);
  3601. }
  3602. return len;
  3603. }
  3604. static struct md_sysfs_entry md_suspend_lo =
  3605. __ATTR(suspend_lo, S_IRUGO|S_IWUSR, suspend_lo_show, suspend_lo_store);
  3606. static ssize_t
  3607. suspend_hi_show(mddev_t *mddev, char *page)
  3608. {
  3609. return sprintf(page, "%llu\n", (unsigned long long)mddev->suspend_hi);
  3610. }
  3611. static ssize_t
  3612. suspend_hi_store(mddev_t *mddev, const char *buf, size_t len)
  3613. {
  3614. char *e;
  3615. unsigned long long new = simple_strtoull(buf, &e, 10);
  3616. unsigned long long old = mddev->suspend_hi;
  3617. if (mddev->pers == NULL ||
  3618. mddev->pers->quiesce == NULL)
  3619. return -EINVAL;
  3620. if (buf == e || (*e && *e != '\n'))
  3621. return -EINVAL;
  3622. mddev->suspend_hi = new;
  3623. if (new <= old)
  3624. /* Shrinking suspended region */
  3625. mddev->pers->quiesce(mddev, 2);
  3626. else {
  3627. /* Expanding suspended region - need to wait */
  3628. mddev->pers->quiesce(mddev, 1);
  3629. mddev->pers->quiesce(mddev, 0);
  3630. }
  3631. return len;
  3632. }
  3633. static struct md_sysfs_entry md_suspend_hi =
  3634. __ATTR(suspend_hi, S_IRUGO|S_IWUSR, suspend_hi_show, suspend_hi_store);
  3635. static ssize_t
  3636. reshape_position_show(mddev_t *mddev, char *page)
  3637. {
  3638. if (mddev->reshape_position != MaxSector)
  3639. return sprintf(page, "%llu\n",
  3640. (unsigned long long)mddev->reshape_position);
  3641. strcpy(page, "none\n");
  3642. return 5;
  3643. }
  3644. static ssize_t
  3645. reshape_position_store(mddev_t *mddev, const char *buf, size_t len)
  3646. {
  3647. char *e;
  3648. unsigned long long new = simple_strtoull(buf, &e, 10);
  3649. if (mddev->pers)
  3650. return -EBUSY;
  3651. if (buf == e || (*e && *e != '\n'))
  3652. return -EINVAL;
  3653. mddev->reshape_position = new;
  3654. mddev->delta_disks = 0;
  3655. mddev->new_level = mddev->level;
  3656. mddev->new_layout = mddev->layout;
  3657. mddev->new_chunk_sectors = mddev->chunk_sectors;
  3658. return len;
  3659. }
  3660. static struct md_sysfs_entry md_reshape_position =
  3661. __ATTR(reshape_position, S_IRUGO|S_IWUSR, reshape_position_show,
  3662. reshape_position_store);
  3663. static ssize_t
  3664. array_size_show(mddev_t *mddev, char *page)
  3665. {
  3666. if (mddev->external_size)
  3667. return sprintf(page, "%llu\n",
  3668. (unsigned long long)mddev->array_sectors/2);
  3669. else
  3670. return sprintf(page, "default\n");
  3671. }
  3672. static ssize_t
  3673. array_size_store(mddev_t *mddev, const char *buf, size_t len)
  3674. {
  3675. sector_t sectors;
  3676. if (strncmp(buf, "default", 7) == 0) {
  3677. if (mddev->pers)
  3678. sectors = mddev->pers->size(mddev, 0, 0);
  3679. else
  3680. sectors = mddev->array_sectors;
  3681. mddev->external_size = 0;
  3682. } else {
  3683. if (strict_blocks_to_sectors(buf, &sectors) < 0)
  3684. return -EINVAL;
  3685. if (mddev->pers && mddev->pers->size(mddev, 0, 0) < sectors)
  3686. return -E2BIG;
  3687. mddev->external_size = 1;
  3688. }
  3689. mddev->array_sectors = sectors;
  3690. if (mddev->pers) {
  3691. set_capacity(mddev->gendisk, mddev->array_sectors);
  3692. revalidate_disk(mddev->gendisk);
  3693. }
  3694. return len;
  3695. }
  3696. static struct md_sysfs_entry md_array_size =
  3697. __ATTR(array_size, S_IRUGO|S_IWUSR, array_size_show,
  3698. array_size_store);
  3699. static struct attribute *md_default_attrs[] = {
  3700. &md_level.attr,
  3701. &md_layout.attr,
  3702. &md_raid_disks.attr,
  3703. &md_chunk_size.attr,
  3704. &md_size.attr,
  3705. &md_resync_start.attr,
  3706. &md_metadata.attr,
  3707. &md_new_device.attr,
  3708. &md_safe_delay.attr,
  3709. &md_array_state.attr,
  3710. &md_reshape_position.attr,
  3711. &md_array_size.attr,
  3712. &max_corr_read_errors.attr,
  3713. NULL,
  3714. };
  3715. static struct attribute *md_redundancy_attrs[] = {
  3716. &md_scan_mode.attr,
  3717. &md_mismatches.attr,
  3718. &md_sync_min.attr,
  3719. &md_sync_max.attr,
  3720. &md_sync_speed.attr,
  3721. &md_sync_force_parallel.attr,
  3722. &md_sync_completed.attr,
  3723. &md_min_sync.attr,
  3724. &md_max_sync.attr,
  3725. &md_suspend_lo.attr,
  3726. &md_suspend_hi.attr,
  3727. &md_bitmap.attr,
  3728. &md_degraded.attr,
  3729. NULL,
  3730. };
  3731. static struct attribute_group md_redundancy_group = {
  3732. .name = NULL,
  3733. .attrs = md_redundancy_attrs,
  3734. };
  3735. static ssize_t
  3736. md_attr_show(struct kobject *kobj, struct attribute *attr, char *page)
  3737. {
  3738. struct md_sysfs_entry *entry = container_of(attr, struct md_sysfs_entry, attr);
  3739. mddev_t *mddev = container_of(kobj, struct mddev_s, kobj);
  3740. ssize_t rv;
  3741. if (!entry->show)
  3742. return -EIO;
  3743. rv = mddev_lock(mddev);
  3744. if (!rv) {
  3745. rv = entry->show(mddev, page);
  3746. mddev_unlock(mddev);
  3747. }
  3748. return rv;
  3749. }
  3750. static ssize_t
  3751. md_attr_store(struct kobject *kobj, struct attribute *attr,
  3752. const char *page, size_t length)
  3753. {
  3754. struct md_sysfs_entry *entry = container_of(attr, struct md_sysfs_entry, attr);
  3755. mddev_t *mddev = container_of(kobj, struct mddev_s, kobj);
  3756. ssize_t rv;
  3757. if (!entry->store)
  3758. return -EIO;
  3759. if (!capable(CAP_SYS_ADMIN))
  3760. return -EACCES;
  3761. rv = mddev_lock(mddev);
  3762. if (mddev->hold_active == UNTIL_IOCTL)
  3763. mddev->hold_active = 0;
  3764. if (!rv) {
  3765. rv = entry->store(mddev, page, length);
  3766. mddev_unlock(mddev);
  3767. }
  3768. return rv;
  3769. }
  3770. static void md_free(struct kobject *ko)
  3771. {
  3772. mddev_t *mddev = container_of(ko, mddev_t, kobj);
  3773. if (mddev->sysfs_state)
  3774. sysfs_put(mddev->sysfs_state);
  3775. if (mddev->gendisk) {
  3776. del_gendisk(mddev->gendisk);
  3777. put_disk(mddev->gendisk);
  3778. }
  3779. if (mddev->queue)
  3780. blk_cleanup_queue(mddev->queue);
  3781. kfree(mddev);
  3782. }
  3783. static const struct sysfs_ops md_sysfs_ops = {
  3784. .show = md_attr_show,
  3785. .store = md_attr_store,
  3786. };
  3787. static struct kobj_type md_ktype = {
  3788. .release = md_free,
  3789. .sysfs_ops = &md_sysfs_ops,
  3790. .default_attrs = md_default_attrs,
  3791. };
  3792. int mdp_major = 0;
  3793. static void mddev_delayed_delete(struct work_struct *ws)
  3794. {
  3795. mddev_t *mddev = container_of(ws, mddev_t, del_work);
  3796. sysfs_remove_group(&mddev->kobj, &md_bitmap_group);
  3797. kobject_del(&mddev->kobj);
  3798. kobject_put(&mddev->kobj);
  3799. }
  3800. static int md_alloc(dev_t dev, char *name)
  3801. {
  3802. static DEFINE_MUTEX(disks_mutex);
  3803. mddev_t *mddev = mddev_find(dev);
  3804. struct gendisk *disk;
  3805. int partitioned;
  3806. int shift;
  3807. int unit;
  3808. int error;
  3809. if (!mddev)
  3810. return -ENODEV;
  3811. partitioned = (MAJOR(mddev->unit) != MD_MAJOR);
  3812. shift = partitioned ? MdpMinorShift : 0;
  3813. unit = MINOR(mddev->unit) >> shift;
  3814. /* wait for any previous instance of this device to be
  3815. * completely removed (mddev_delayed_delete).
  3816. */
  3817. flush_workqueue(md_misc_wq);
  3818. mutex_lock(&disks_mutex);
  3819. error = -EEXIST;
  3820. if (mddev->gendisk)
  3821. goto abort;
  3822. if (name) {
  3823. /* Need to ensure that 'name' is not a duplicate.
  3824. */
  3825. mddev_t *mddev2;
  3826. spin_lock(&all_mddevs_lock);
  3827. list_for_each_entry(mddev2, &all_mddevs, all_mddevs)
  3828. if (mddev2->gendisk &&
  3829. strcmp(mddev2->gendisk->disk_name, name) == 0) {
  3830. spin_unlock(&all_mddevs_lock);
  3831. goto abort;
  3832. }
  3833. spin_unlock(&all_mddevs_lock);
  3834. }
  3835. error = -ENOMEM;
  3836. mddev->queue = blk_alloc_queue(GFP_KERNEL);
  3837. if (!mddev->queue)
  3838. goto abort;
  3839. mddev->queue->queuedata = mddev;
  3840. blk_queue_make_request(mddev->queue, md_make_request);
  3841. disk = alloc_disk(1 << shift);
  3842. if (!disk) {
  3843. blk_cleanup_queue(mddev->queue);
  3844. mddev->queue = NULL;
  3845. goto abort;
  3846. }
  3847. disk->major = MAJOR(mddev->unit);
  3848. disk->first_minor = unit << shift;
  3849. if (name)
  3850. strcpy(disk->disk_name, name);
  3851. else if (partitioned)
  3852. sprintf(disk->disk_name, "md_d%d", unit);
  3853. else
  3854. sprintf(disk->disk_name, "md%d", unit);
  3855. disk->fops = &md_fops;
  3856. disk->private_data = mddev;
  3857. disk->queue = mddev->queue;
  3858. blk_queue_flush(mddev->queue, REQ_FLUSH | REQ_FUA);
  3859. /* Allow extended partitions. This makes the
  3860. * 'mdp' device redundant, but we can't really
  3861. * remove it now.
  3862. */
  3863. disk->flags |= GENHD_FL_EXT_DEVT;
  3864. mddev->gendisk = disk;
  3865. /* As soon as we call add_disk(), another thread could get
  3866. * through to md_open, so make sure it doesn't get too far
  3867. */
  3868. mutex_lock(&mddev->open_mutex);
  3869. add_disk(disk);
  3870. error = kobject_init_and_add(&mddev->kobj, &md_ktype,
  3871. &disk_to_dev(disk)->kobj, "%s", "md");
  3872. if (error) {
  3873. /* This isn't possible, but as kobject_init_and_add is marked
  3874. * __must_check, we must do something with the result
  3875. */
  3876. printk(KERN_WARNING "md: cannot register %s/md - name in use\n",
  3877. disk->disk_name);
  3878. error = 0;
  3879. }
  3880. if (mddev->kobj.sd &&
  3881. sysfs_create_group(&mddev->kobj, &md_bitmap_group))
  3882. printk(KERN_DEBUG "pointless warning\n");
  3883. mutex_unlock(&mddev->open_mutex);
  3884. abort:
  3885. mutex_unlock(&disks_mutex);
  3886. if (!error && mddev->kobj.sd) {
  3887. kobject_uevent(&mddev->kobj, KOBJ_ADD);
  3888. mddev->sysfs_state = sysfs_get_dirent_safe(mddev->kobj.sd, "array_state");
  3889. }
  3890. mddev_put(mddev);
  3891. return error;
  3892. }
  3893. static struct kobject *md_probe(dev_t dev, int *part, void *data)
  3894. {
  3895. md_alloc(dev, NULL);
  3896. return NULL;
  3897. }
  3898. static int add_named_array(const char *val, struct kernel_param *kp)
  3899. {
  3900. /* val must be "md_*" where * is not all digits.
  3901. * We allocate an array with a large free minor number, and
  3902. * set the name to val. val must not already be an active name.
  3903. */
  3904. int len = strlen(val);
  3905. char buf[DISK_NAME_LEN];
  3906. while (len && val[len-1] == '\n')
  3907. len--;
  3908. if (len >= DISK_NAME_LEN)
  3909. return -E2BIG;
  3910. strlcpy(buf, val, len+1);
  3911. if (strncmp(buf, "md_", 3) != 0)
  3912. return -EINVAL;
  3913. return md_alloc(0, buf);
  3914. }
  3915. static void md_safemode_timeout(unsigned long data)
  3916. {
  3917. mddev_t *mddev = (mddev_t *) data;
  3918. if (!atomic_read(&mddev->writes_pending)) {
  3919. mddev->safemode = 1;
  3920. if (mddev->external)
  3921. sysfs_notify_dirent_safe(mddev->sysfs_state);
  3922. }
  3923. md_wakeup_thread(mddev->thread);
  3924. }
  3925. static int start_dirty_degraded;
  3926. int md_run(mddev_t *mddev)
  3927. {
  3928. int err;
  3929. mdk_rdev_t *rdev;
  3930. struct mdk_personality *pers;
  3931. if (list_empty(&mddev->disks))
  3932. /* cannot run an array with no devices.. */
  3933. return -EINVAL;
  3934. if (mddev->pers)
  3935. return -EBUSY;
  3936. /* Cannot run until previous stop completes properly */
  3937. if (mddev->sysfs_active)
  3938. return -EBUSY;
  3939. /*
  3940. * Analyze all RAID superblock(s)
  3941. */
  3942. if (!mddev->raid_disks) {
  3943. if (!mddev->persistent)
  3944. return -EINVAL;
  3945. analyze_sbs(mddev);
  3946. }
  3947. if (mddev->level != LEVEL_NONE)
  3948. request_module("md-level-%d", mddev->level);
  3949. else if (mddev->clevel[0])
  3950. request_module("md-%s", mddev->clevel);
  3951. /*
  3952. * Drop all container device buffers, from now on
  3953. * the only valid external interface is through the md
  3954. * device.
  3955. */
  3956. list_for_each_entry(rdev, &mddev->disks, same_set) {
  3957. if (test_bit(Faulty, &rdev->flags))
  3958. continue;
  3959. sync_blockdev(rdev->bdev);
  3960. invalidate_bdev(rdev->bdev);
  3961. /* perform some consistency tests on the device.
  3962. * We don't want the data to overlap the metadata,
  3963. * Internal Bitmap issues have been handled elsewhere.
  3964. */
  3965. if (rdev->meta_bdev) {
  3966. /* Nothing to check */;
  3967. } else if (rdev->data_offset < rdev->sb_start) {
  3968. if (mddev->dev_sectors &&
  3969. rdev->data_offset + mddev->dev_sectors
  3970. > rdev->sb_start) {
  3971. printk("md: %s: data overlaps metadata\n",
  3972. mdname(mddev));
  3973. return -EINVAL;
  3974. }
  3975. } else {
  3976. if (rdev->sb_start + rdev->sb_size/512
  3977. > rdev->data_offset) {
  3978. printk("md: %s: metadata overlaps data\n",
  3979. mdname(mddev));
  3980. return -EINVAL;
  3981. }
  3982. }
  3983. sysfs_notify_dirent_safe(rdev->sysfs_state);
  3984. }
  3985. if (mddev->bio_set == NULL)
  3986. mddev->bio_set = bioset_create(BIO_POOL_SIZE, sizeof(mddev));
  3987. spin_lock(&pers_lock);
  3988. pers = find_pers(mddev->level, mddev->clevel);
  3989. if (!pers || !try_module_get(pers->owner)) {
  3990. spin_unlock(&pers_lock);
  3991. if (mddev->level != LEVEL_NONE)
  3992. printk(KERN_WARNING "md: personality for level %d is not loaded!\n",
  3993. mddev->level);
  3994. else
  3995. printk(KERN_WARNING "md: personality for level %s is not loaded!\n",
  3996. mddev->clevel);
  3997. return -EINVAL;
  3998. }
  3999. mddev->pers = pers;
  4000. spin_unlock(&pers_lock);
  4001. if (mddev->level != pers->level) {
  4002. mddev->level = pers->level;
  4003. mddev->new_level = pers->level;
  4004. }
  4005. strlcpy(mddev->clevel, pers->name, sizeof(mddev->clevel));
  4006. if (mddev->reshape_position != MaxSector &&
  4007. pers->start_reshape == NULL) {
  4008. /* This personality cannot handle reshaping... */
  4009. mddev->pers = NULL;
  4010. module_put(pers->owner);
  4011. return -EINVAL;
  4012. }
  4013. if (pers->sync_request) {
  4014. /* Warn if this is a potentially silly
  4015. * configuration.
  4016. */
  4017. char b[BDEVNAME_SIZE], b2[BDEVNAME_SIZE];
  4018. mdk_rdev_t *rdev2;
  4019. int warned = 0;
  4020. list_for_each_entry(rdev, &mddev->disks, same_set)
  4021. list_for_each_entry(rdev2, &mddev->disks, same_set) {
  4022. if (rdev < rdev2 &&
  4023. rdev->bdev->bd_contains ==
  4024. rdev2->bdev->bd_contains) {
  4025. printk(KERN_WARNING
  4026. "%s: WARNING: %s appears to be"
  4027. " on the same physical disk as"
  4028. " %s.\n",
  4029. mdname(mddev),
  4030. bdevname(rdev->bdev,b),
  4031. bdevname(rdev2->bdev,b2));
  4032. warned = 1;
  4033. }
  4034. }
  4035. if (warned)
  4036. printk(KERN_WARNING
  4037. "True protection against single-disk"
  4038. " failure might be compromised.\n");
  4039. }
  4040. mddev->recovery = 0;
  4041. /* may be over-ridden by personality */
  4042. mddev->resync_max_sectors = mddev->dev_sectors;
  4043. mddev->ok_start_degraded = start_dirty_degraded;
  4044. if (start_readonly && mddev->ro == 0)
  4045. mddev->ro = 2; /* read-only, but switch on first write */
  4046. err = mddev->pers->run(mddev);
  4047. if (err)
  4048. printk(KERN_ERR "md: pers->run() failed ...\n");
  4049. else if (mddev->pers->size(mddev, 0, 0) < mddev->array_sectors) {
  4050. WARN_ONCE(!mddev->external_size, "%s: default size too small,"
  4051. " but 'external_size' not in effect?\n", __func__);
  4052. printk(KERN_ERR
  4053. "md: invalid array_size %llu > default size %llu\n",
  4054. (unsigned long long)mddev->array_sectors / 2,
  4055. (unsigned long long)mddev->pers->size(mddev, 0, 0) / 2);
  4056. err = -EINVAL;
  4057. mddev->pers->stop(mddev);
  4058. }
  4059. if (err == 0 && mddev->pers->sync_request) {
  4060. err = bitmap_create(mddev);
  4061. if (err) {
  4062. printk(KERN_ERR "%s: failed to create bitmap (%d)\n",
  4063. mdname(mddev), err);
  4064. mddev->pers->stop(mddev);
  4065. }
  4066. }
  4067. if (err) {
  4068. module_put(mddev->pers->owner);
  4069. mddev->pers = NULL;
  4070. bitmap_destroy(mddev);
  4071. return err;
  4072. }
  4073. if (mddev->pers->sync_request) {
  4074. if (mddev->kobj.sd &&
  4075. sysfs_create_group(&mddev->kobj, &md_redundancy_group))
  4076. printk(KERN_WARNING
  4077. "md: cannot register extra attributes for %s\n",
  4078. mdname(mddev));
  4079. mddev->sysfs_action = sysfs_get_dirent_safe(mddev->kobj.sd, "sync_action");
  4080. } else if (mddev->ro == 2) /* auto-readonly not meaningful */
  4081. mddev->ro = 0;
  4082. atomic_set(&mddev->writes_pending,0);
  4083. atomic_set(&mddev->max_corr_read_errors,
  4084. MD_DEFAULT_MAX_CORRECTED_READ_ERRORS);
  4085. mddev->safemode = 0;
  4086. mddev->safemode_timer.function = md_safemode_timeout;
  4087. mddev->safemode_timer.data = (unsigned long) mddev;
  4088. mddev->safemode_delay = (200 * HZ)/1000 +1; /* 200 msec delay */
  4089. mddev->in_sync = 1;
  4090. smp_wmb();
  4091. mddev->ready = 1;
  4092. list_for_each_entry(rdev, &mddev->disks, same_set)
  4093. if (rdev->raid_disk >= 0) {
  4094. char nm[20];
  4095. sprintf(nm, "rd%d", rdev->raid_disk);
  4096. if (sysfs_create_link(&mddev->kobj, &rdev->kobj, nm))
  4097. /* failure here is OK */;
  4098. }
  4099. set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
  4100. if (mddev->flags)
  4101. md_update_sb(mddev, 0);
  4102. md_new_event(mddev);
  4103. sysfs_notify_dirent_safe(mddev->sysfs_state);
  4104. sysfs_notify_dirent_safe(mddev->sysfs_action);
  4105. sysfs_notify(&mddev->kobj, NULL, "degraded");
  4106. return 0;
  4107. }
  4108. EXPORT_SYMBOL_GPL(md_run);
  4109. static int do_md_run(mddev_t *mddev)
  4110. {
  4111. int err;
  4112. err = md_run(mddev);
  4113. if (err)
  4114. goto out;
  4115. err = bitmap_load(mddev);
  4116. if (err) {
  4117. bitmap_destroy(mddev);
  4118. goto out;
  4119. }
  4120. md_wakeup_thread(mddev->thread);
  4121. md_wakeup_thread(mddev->sync_thread); /* possibly kick off a reshape */
  4122. set_capacity(mddev->gendisk, mddev->array_sectors);
  4123. revalidate_disk(mddev->gendisk);
  4124. mddev->changed = 1;
  4125. kobject_uevent(&disk_to_dev(mddev->gendisk)->kobj, KOBJ_CHANGE);
  4126. out:
  4127. return err;
  4128. }
  4129. static int restart_array(mddev_t *mddev)
  4130. {
  4131. struct gendisk *disk = mddev->gendisk;
  4132. /* Complain if it has no devices */
  4133. if (list_empty(&mddev->disks))
  4134. return -ENXIO;
  4135. if (!mddev->pers)
  4136. return -EINVAL;
  4137. if (!mddev->ro)
  4138. return -EBUSY;
  4139. mddev->safemode = 0;
  4140. mddev->ro = 0;
  4141. set_disk_ro(disk, 0);
  4142. printk(KERN_INFO "md: %s switched to read-write mode.\n",
  4143. mdname(mddev));
  4144. /* Kick recovery or resync if necessary */
  4145. set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
  4146. md_wakeup_thread(mddev->thread);
  4147. md_wakeup_thread(mddev->sync_thread);
  4148. sysfs_notify_dirent_safe(mddev->sysfs_state);
  4149. return 0;
  4150. }
  4151. /* similar to deny_write_access, but accounts for our holding a reference
  4152. * to the file ourselves */
  4153. static int deny_bitmap_write_access(struct file * file)
  4154. {
  4155. struct inode *inode = file->f_mapping->host;
  4156. spin_lock(&inode->i_lock);
  4157. if (atomic_read(&inode->i_writecount) > 1) {
  4158. spin_unlock(&inode->i_lock);
  4159. return -ETXTBSY;
  4160. }
  4161. atomic_set(&inode->i_writecount, -1);
  4162. spin_unlock(&inode->i_lock);
  4163. return 0;
  4164. }
  4165. void restore_bitmap_write_access(struct file *file)
  4166. {
  4167. struct inode *inode = file->f_mapping->host;
  4168. spin_lock(&inode->i_lock);
  4169. atomic_set(&inode->i_writecount, 1);
  4170. spin_unlock(&inode->i_lock);
  4171. }
  4172. static void md_clean(mddev_t *mddev)
  4173. {
  4174. mddev->array_sectors = 0;
  4175. mddev->external_size = 0;
  4176. mddev->dev_sectors = 0;
  4177. mddev->raid_disks = 0;
  4178. mddev->recovery_cp = 0;
  4179. mddev->resync_min = 0;
  4180. mddev->resync_max = MaxSector;
  4181. mddev->reshape_position = MaxSector;
  4182. mddev->external = 0;
  4183. mddev->persistent = 0;
  4184. mddev->level = LEVEL_NONE;
  4185. mddev->clevel[0] = 0;
  4186. mddev->flags = 0;
  4187. mddev->ro = 0;
  4188. mddev->metadata_type[0] = 0;
  4189. mddev->chunk_sectors = 0;
  4190. mddev->ctime = mddev->utime = 0;
  4191. mddev->layout = 0;
  4192. mddev->max_disks = 0;
  4193. mddev->events = 0;
  4194. mddev->can_decrease_events = 0;
  4195. mddev->delta_disks = 0;
  4196. mddev->new_level = LEVEL_NONE;
  4197. mddev->new_layout = 0;
  4198. mddev->new_chunk_sectors = 0;
  4199. mddev->curr_resync = 0;
  4200. mddev->resync_mismatches = 0;
  4201. mddev->suspend_lo = mddev->suspend_hi = 0;
  4202. mddev->sync_speed_min = mddev->sync_speed_max = 0;
  4203. mddev->recovery = 0;
  4204. mddev->in_sync = 0;
  4205. mddev->changed = 0;
  4206. mddev->degraded = 0;
  4207. mddev->safemode = 0;
  4208. mddev->bitmap_info.offset = 0;
  4209. mddev->bitmap_info.default_offset = 0;
  4210. mddev->bitmap_info.chunksize = 0;
  4211. mddev->bitmap_info.daemon_sleep = 0;
  4212. mddev->bitmap_info.max_write_behind = 0;
  4213. }
  4214. static void __md_stop_writes(mddev_t *mddev)
  4215. {
  4216. if (mddev->sync_thread) {
  4217. set_bit(MD_RECOVERY_FROZEN, &mddev->recovery);
  4218. set_bit(MD_RECOVERY_INTR, &mddev->recovery);
  4219. reap_sync_thread(mddev);
  4220. }
  4221. del_timer_sync(&mddev->safemode_timer);
  4222. bitmap_flush(mddev);
  4223. md_super_wait(mddev);
  4224. if (!mddev->in_sync || mddev->flags) {
  4225. /* mark array as shutdown cleanly */
  4226. mddev->in_sync = 1;
  4227. md_update_sb(mddev, 1);
  4228. }
  4229. }
  4230. void md_stop_writes(mddev_t *mddev)
  4231. {
  4232. mddev_lock(mddev);
  4233. __md_stop_writes(mddev);
  4234. mddev_unlock(mddev);
  4235. }
  4236. EXPORT_SYMBOL_GPL(md_stop_writes);
  4237. void md_stop(mddev_t *mddev)
  4238. {
  4239. mddev->ready = 0;
  4240. mddev->pers->stop(mddev);
  4241. if (mddev->pers->sync_request && mddev->to_remove == NULL)
  4242. mddev->to_remove = &md_redundancy_group;
  4243. module_put(mddev->pers->owner);
  4244. mddev->pers = NULL;
  4245. clear_bit(MD_RECOVERY_FROZEN, &mddev->recovery);
  4246. }
  4247. EXPORT_SYMBOL_GPL(md_stop);
  4248. static int md_set_readonly(mddev_t *mddev, int is_open)
  4249. {
  4250. int err = 0;
  4251. mutex_lock(&mddev->open_mutex);
  4252. if (atomic_read(&mddev->openers) > is_open) {
  4253. printk("md: %s still in use.\n",mdname(mddev));
  4254. err = -EBUSY;
  4255. goto out;
  4256. }
  4257. if (mddev->pers) {
  4258. __md_stop_writes(mddev);
  4259. err = -ENXIO;
  4260. if (mddev->ro==1)
  4261. goto out;
  4262. mddev->ro = 1;
  4263. set_disk_ro(mddev->gendisk, 1);
  4264. clear_bit(MD_RECOVERY_FROZEN, &mddev->recovery);
  4265. sysfs_notify_dirent_safe(mddev->sysfs_state);
  4266. err = 0;
  4267. }
  4268. out:
  4269. mutex_unlock(&mddev->open_mutex);
  4270. return err;
  4271. }
  4272. /* mode:
  4273. * 0 - completely stop and dis-assemble array
  4274. * 2 - stop but do not disassemble array
  4275. */
  4276. static int do_md_stop(mddev_t * mddev, int mode, int is_open)
  4277. {
  4278. struct gendisk *disk = mddev->gendisk;
  4279. mdk_rdev_t *rdev;
  4280. mutex_lock(&mddev->open_mutex);
  4281. if (atomic_read(&mddev->openers) > is_open ||
  4282. mddev->sysfs_active) {
  4283. printk("md: %s still in use.\n",mdname(mddev));
  4284. mutex_unlock(&mddev->open_mutex);
  4285. return -EBUSY;
  4286. }
  4287. if (mddev->pers) {
  4288. if (mddev->ro)
  4289. set_disk_ro(disk, 0);
  4290. __md_stop_writes(mddev);
  4291. md_stop(mddev);
  4292. mddev->queue->merge_bvec_fn = NULL;
  4293. mddev->queue->backing_dev_info.congested_fn = NULL;
  4294. /* tell userspace to handle 'inactive' */
  4295. sysfs_notify_dirent_safe(mddev->sysfs_state);
  4296. list_for_each_entry(rdev, &mddev->disks, same_set)
  4297. if (rdev->raid_disk >= 0) {
  4298. char nm[20];
  4299. sprintf(nm, "rd%d", rdev->raid_disk);
  4300. sysfs_remove_link(&mddev->kobj, nm);
  4301. }
  4302. set_capacity(disk, 0);
  4303. mutex_unlock(&mddev->open_mutex);
  4304. mddev->changed = 1;
  4305. revalidate_disk(disk);
  4306. if (mddev->ro)
  4307. mddev->ro = 0;
  4308. } else
  4309. mutex_unlock(&mddev->open_mutex);
  4310. /*
  4311. * Free resources if final stop
  4312. */
  4313. if (mode == 0) {
  4314. printk(KERN_INFO "md: %s stopped.\n", mdname(mddev));
  4315. bitmap_destroy(mddev);
  4316. if (mddev->bitmap_info.file) {
  4317. restore_bitmap_write_access(mddev->bitmap_info.file);
  4318. fput(mddev->bitmap_info.file);
  4319. mddev->bitmap_info.file = NULL;
  4320. }
  4321. mddev->bitmap_info.offset = 0;
  4322. export_array(mddev);
  4323. md_clean(mddev);
  4324. kobject_uevent(&disk_to_dev(mddev->gendisk)->kobj, KOBJ_CHANGE);
  4325. if (mddev->hold_active == UNTIL_STOP)
  4326. mddev->hold_active = 0;
  4327. }
  4328. blk_integrity_unregister(disk);
  4329. md_new_event(mddev);
  4330. sysfs_notify_dirent_safe(mddev->sysfs_state);
  4331. return 0;
  4332. }
  4333. #ifndef MODULE
  4334. static void autorun_array(mddev_t *mddev)
  4335. {
  4336. mdk_rdev_t *rdev;
  4337. int err;
  4338. if (list_empty(&mddev->disks))
  4339. return;
  4340. printk(KERN_INFO "md: running: ");
  4341. list_for_each_entry(rdev, &mddev->disks, same_set) {
  4342. char b[BDEVNAME_SIZE];
  4343. printk("<%s>", bdevname(rdev->bdev,b));
  4344. }
  4345. printk("\n");
  4346. err = do_md_run(mddev);
  4347. if (err) {
  4348. printk(KERN_WARNING "md: do_md_run() returned %d\n", err);
  4349. do_md_stop(mddev, 0, 0);
  4350. }
  4351. }
  4352. /*
  4353. * lets try to run arrays based on all disks that have arrived
  4354. * until now. (those are in pending_raid_disks)
  4355. *
  4356. * the method: pick the first pending disk, collect all disks with
  4357. * the same UUID, remove all from the pending list and put them into
  4358. * the 'same_array' list. Then order this list based on superblock
  4359. * update time (freshest comes first), kick out 'old' disks and
  4360. * compare superblocks. If everything's fine then run it.
  4361. *
  4362. * If "unit" is allocated, then bump its reference count
  4363. */
  4364. static void autorun_devices(int part)
  4365. {
  4366. mdk_rdev_t *rdev0, *rdev, *tmp;
  4367. mddev_t *mddev;
  4368. char b[BDEVNAME_SIZE];
  4369. printk(KERN_INFO "md: autorun ...\n");
  4370. while (!list_empty(&pending_raid_disks)) {
  4371. int unit;
  4372. dev_t dev;
  4373. LIST_HEAD(candidates);
  4374. rdev0 = list_entry(pending_raid_disks.next,
  4375. mdk_rdev_t, same_set);
  4376. printk(KERN_INFO "md: considering %s ...\n",
  4377. bdevname(rdev0->bdev,b));
  4378. INIT_LIST_HEAD(&candidates);
  4379. rdev_for_each_list(rdev, tmp, &pending_raid_disks)
  4380. if (super_90_load(rdev, rdev0, 0) >= 0) {
  4381. printk(KERN_INFO "md: adding %s ...\n",
  4382. bdevname(rdev->bdev,b));
  4383. list_move(&rdev->same_set, &candidates);
  4384. }
  4385. /*
  4386. * now we have a set of devices, with all of them having
  4387. * mostly sane superblocks. It's time to allocate the
  4388. * mddev.
  4389. */
  4390. if (part) {
  4391. dev = MKDEV(mdp_major,
  4392. rdev0->preferred_minor << MdpMinorShift);
  4393. unit = MINOR(dev) >> MdpMinorShift;
  4394. } else {
  4395. dev = MKDEV(MD_MAJOR, rdev0->preferred_minor);
  4396. unit = MINOR(dev);
  4397. }
  4398. if (rdev0->preferred_minor != unit) {
  4399. printk(KERN_INFO "md: unit number in %s is bad: %d\n",
  4400. bdevname(rdev0->bdev, b), rdev0->preferred_minor);
  4401. break;
  4402. }
  4403. md_probe(dev, NULL, NULL);
  4404. mddev = mddev_find(dev);
  4405. if (!mddev || !mddev->gendisk) {
  4406. if (mddev)
  4407. mddev_put(mddev);
  4408. printk(KERN_ERR
  4409. "md: cannot allocate memory for md drive.\n");
  4410. break;
  4411. }
  4412. if (mddev_lock(mddev))
  4413. printk(KERN_WARNING "md: %s locked, cannot run\n",
  4414. mdname(mddev));
  4415. else if (mddev->raid_disks || mddev->major_version
  4416. || !list_empty(&mddev->disks)) {
  4417. printk(KERN_WARNING
  4418. "md: %s already running, cannot run %s\n",
  4419. mdname(mddev), bdevname(rdev0->bdev,b));
  4420. mddev_unlock(mddev);
  4421. } else {
  4422. printk(KERN_INFO "md: created %s\n", mdname(mddev));
  4423. mddev->persistent = 1;
  4424. rdev_for_each_list(rdev, tmp, &candidates) {
  4425. list_del_init(&rdev->same_set);
  4426. if (bind_rdev_to_array(rdev, mddev))
  4427. export_rdev(rdev);
  4428. }
  4429. autorun_array(mddev);
  4430. mddev_unlock(mddev);
  4431. }
  4432. /* on success, candidates will be empty, on error
  4433. * it won't...
  4434. */
  4435. rdev_for_each_list(rdev, tmp, &candidates) {
  4436. list_del_init(&rdev->same_set);
  4437. export_rdev(rdev);
  4438. }
  4439. mddev_put(mddev);
  4440. }
  4441. printk(KERN_INFO "md: ... autorun DONE.\n");
  4442. }
  4443. #endif /* !MODULE */
  4444. static int get_version(void __user * arg)
  4445. {
  4446. mdu_version_t ver;
  4447. ver.major = MD_MAJOR_VERSION;
  4448. ver.minor = MD_MINOR_VERSION;
  4449. ver.patchlevel = MD_PATCHLEVEL_VERSION;
  4450. if (copy_to_user(arg, &ver, sizeof(ver)))
  4451. return -EFAULT;
  4452. return 0;
  4453. }
  4454. static int get_array_info(mddev_t * mddev, void __user * arg)
  4455. {
  4456. mdu_array_info_t info;
  4457. int nr,working,insync,failed,spare;
  4458. mdk_rdev_t *rdev;
  4459. nr=working=insync=failed=spare=0;
  4460. list_for_each_entry(rdev, &mddev->disks, same_set) {
  4461. nr++;
  4462. if (test_bit(Faulty, &rdev->flags))
  4463. failed++;
  4464. else {
  4465. working++;
  4466. if (test_bit(In_sync, &rdev->flags))
  4467. insync++;
  4468. else
  4469. spare++;
  4470. }
  4471. }
  4472. info.major_version = mddev->major_version;
  4473. info.minor_version = mddev->minor_version;
  4474. info.patch_version = MD_PATCHLEVEL_VERSION;
  4475. info.ctime = mddev->ctime;
  4476. info.level = mddev->level;
  4477. info.size = mddev->dev_sectors / 2;
  4478. if (info.size != mddev->dev_sectors / 2) /* overflow */
  4479. info.size = -1;
  4480. info.nr_disks = nr;
  4481. info.raid_disks = mddev->raid_disks;
  4482. info.md_minor = mddev->md_minor;
  4483. info.not_persistent= !mddev->persistent;
  4484. info.utime = mddev->utime;
  4485. info.state = 0;
  4486. if (mddev->in_sync)
  4487. info.state = (1<<MD_SB_CLEAN);
  4488. if (mddev->bitmap && mddev->bitmap_info.offset)
  4489. info.state = (1<<MD_SB_BITMAP_PRESENT);
  4490. info.active_disks = insync;
  4491. info.working_disks = working;
  4492. info.failed_disks = failed;
  4493. info.spare_disks = spare;
  4494. info.layout = mddev->layout;
  4495. info.chunk_size = mddev->chunk_sectors << 9;
  4496. if (copy_to_user(arg, &info, sizeof(info)))
  4497. return -EFAULT;
  4498. return 0;
  4499. }
  4500. static int get_bitmap_file(mddev_t * mddev, void __user * arg)
  4501. {
  4502. mdu_bitmap_file_t *file = NULL; /* too big for stack allocation */
  4503. char *ptr, *buf = NULL;
  4504. int err = -ENOMEM;
  4505. if (md_allow_write(mddev))
  4506. file = kmalloc(sizeof(*file), GFP_NOIO);
  4507. else
  4508. file = kmalloc(sizeof(*file), GFP_KERNEL);
  4509. if (!file)
  4510. goto out;
  4511. /* bitmap disabled, zero the first byte and copy out */
  4512. if (!mddev->bitmap || !mddev->bitmap->file) {
  4513. file->pathname[0] = '\0';
  4514. goto copy_out;
  4515. }
  4516. buf = kmalloc(sizeof(file->pathname), GFP_KERNEL);
  4517. if (!buf)
  4518. goto out;
  4519. ptr = d_path(&mddev->bitmap->file->f_path, buf, sizeof(file->pathname));
  4520. if (IS_ERR(ptr))
  4521. goto out;
  4522. strcpy(file->pathname, ptr);
  4523. copy_out:
  4524. err = 0;
  4525. if (copy_to_user(arg, file, sizeof(*file)))
  4526. err = -EFAULT;
  4527. out:
  4528. kfree(buf);
  4529. kfree(file);
  4530. return err;
  4531. }
  4532. static int get_disk_info(mddev_t * mddev, void __user * arg)
  4533. {
  4534. mdu_disk_info_t info;
  4535. mdk_rdev_t *rdev;
  4536. if (copy_from_user(&info, arg, sizeof(info)))
  4537. return -EFAULT;
  4538. rdev = find_rdev_nr(mddev, info.number);
  4539. if (rdev) {
  4540. info.major = MAJOR(rdev->bdev->bd_dev);
  4541. info.minor = MINOR(rdev->bdev->bd_dev);
  4542. info.raid_disk = rdev->raid_disk;
  4543. info.state = 0;
  4544. if (test_bit(Faulty, &rdev->flags))
  4545. info.state |= (1<<MD_DISK_FAULTY);
  4546. else if (test_bit(In_sync, &rdev->flags)) {
  4547. info.state |= (1<<MD_DISK_ACTIVE);
  4548. info.state |= (1<<MD_DISK_SYNC);
  4549. }
  4550. if (test_bit(WriteMostly, &rdev->flags))
  4551. info.state |= (1<<MD_DISK_WRITEMOSTLY);
  4552. } else {
  4553. info.major = info.minor = 0;
  4554. info.raid_disk = -1;
  4555. info.state = (1<<MD_DISK_REMOVED);
  4556. }
  4557. if (copy_to_user(arg, &info, sizeof(info)))
  4558. return -EFAULT;
  4559. return 0;
  4560. }
  4561. static int add_new_disk(mddev_t * mddev, mdu_disk_info_t *info)
  4562. {
  4563. char b[BDEVNAME_SIZE], b2[BDEVNAME_SIZE];
  4564. mdk_rdev_t *rdev;
  4565. dev_t dev = MKDEV(info->major,info->minor);
  4566. if (info->major != MAJOR(dev) || info->minor != MINOR(dev))
  4567. return -EOVERFLOW;
  4568. if (!mddev->raid_disks) {
  4569. int err;
  4570. /* expecting a device which has a superblock */
  4571. rdev = md_import_device(dev, mddev->major_version, mddev->minor_version);
  4572. if (IS_ERR(rdev)) {
  4573. printk(KERN_WARNING
  4574. "md: md_import_device returned %ld\n",
  4575. PTR_ERR(rdev));
  4576. return PTR_ERR(rdev);
  4577. }
  4578. if (!list_empty(&mddev->disks)) {
  4579. mdk_rdev_t *rdev0 = list_entry(mddev->disks.next,
  4580. mdk_rdev_t, same_set);
  4581. err = super_types[mddev->major_version]
  4582. .load_super(rdev, rdev0, mddev->minor_version);
  4583. if (err < 0) {
  4584. printk(KERN_WARNING
  4585. "md: %s has different UUID to %s\n",
  4586. bdevname(rdev->bdev,b),
  4587. bdevname(rdev0->bdev,b2));
  4588. export_rdev(rdev);
  4589. return -EINVAL;
  4590. }
  4591. }
  4592. err = bind_rdev_to_array(rdev, mddev);
  4593. if (err)
  4594. export_rdev(rdev);
  4595. return err;
  4596. }
  4597. /*
  4598. * add_new_disk can be used once the array is assembled
  4599. * to add "hot spares". They must already have a superblock
  4600. * written
  4601. */
  4602. if (mddev->pers) {
  4603. int err;
  4604. if (!mddev->pers->hot_add_disk) {
  4605. printk(KERN_WARNING
  4606. "%s: personality does not support diskops!\n",
  4607. mdname(mddev));
  4608. return -EINVAL;
  4609. }
  4610. if (mddev->persistent)
  4611. rdev = md_import_device(dev, mddev->major_version,
  4612. mddev->minor_version);
  4613. else
  4614. rdev = md_import_device(dev, -1, -1);
  4615. if (IS_ERR(rdev)) {
  4616. printk(KERN_WARNING
  4617. "md: md_import_device returned %ld\n",
  4618. PTR_ERR(rdev));
  4619. return PTR_ERR(rdev);
  4620. }
  4621. /* set saved_raid_disk if appropriate */
  4622. if (!mddev->persistent) {
  4623. if (info->state & (1<<MD_DISK_SYNC) &&
  4624. info->raid_disk < mddev->raid_disks) {
  4625. rdev->raid_disk = info->raid_disk;
  4626. set_bit(In_sync, &rdev->flags);
  4627. } else
  4628. rdev->raid_disk = -1;
  4629. } else
  4630. super_types[mddev->major_version].
  4631. validate_super(mddev, rdev);
  4632. if ((info->state & (1<<MD_DISK_SYNC)) &&
  4633. (!test_bit(In_sync, &rdev->flags) ||
  4634. rdev->raid_disk != info->raid_disk)) {
  4635. /* This was a hot-add request, but events doesn't
  4636. * match, so reject it.
  4637. */
  4638. export_rdev(rdev);
  4639. return -EINVAL;
  4640. }
  4641. if (test_bit(In_sync, &rdev->flags))
  4642. rdev->saved_raid_disk = rdev->raid_disk;
  4643. else
  4644. rdev->saved_raid_disk = -1;
  4645. clear_bit(In_sync, &rdev->flags); /* just to be sure */
  4646. if (info->state & (1<<MD_DISK_WRITEMOSTLY))
  4647. set_bit(WriteMostly, &rdev->flags);
  4648. else
  4649. clear_bit(WriteMostly, &rdev->flags);
  4650. rdev->raid_disk = -1;
  4651. err = bind_rdev_to_array(rdev, mddev);
  4652. if (!err && !mddev->pers->hot_remove_disk) {
  4653. /* If there is hot_add_disk but no hot_remove_disk
  4654. * then added disks for geometry changes,
  4655. * and should be added immediately.
  4656. */
  4657. super_types[mddev->major_version].
  4658. validate_super(mddev, rdev);
  4659. err = mddev->pers->hot_add_disk(mddev, rdev);
  4660. if (err)
  4661. unbind_rdev_from_array(rdev);
  4662. }
  4663. if (err)
  4664. export_rdev(rdev);
  4665. else
  4666. sysfs_notify_dirent_safe(rdev->sysfs_state);
  4667. md_update_sb(mddev, 1);
  4668. if (mddev->degraded)
  4669. set_bit(MD_RECOVERY_RECOVER, &mddev->recovery);
  4670. set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
  4671. if (!err)
  4672. md_new_event(mddev);
  4673. md_wakeup_thread(mddev->thread);
  4674. return err;
  4675. }
  4676. /* otherwise, add_new_disk is only allowed
  4677. * for major_version==0 superblocks
  4678. */
  4679. if (mddev->major_version != 0) {
  4680. printk(KERN_WARNING "%s: ADD_NEW_DISK not supported\n",
  4681. mdname(mddev));
  4682. return -EINVAL;
  4683. }
  4684. if (!(info->state & (1<<MD_DISK_FAULTY))) {
  4685. int err;
  4686. rdev = md_import_device(dev, -1, 0);
  4687. if (IS_ERR(rdev)) {
  4688. printk(KERN_WARNING
  4689. "md: error, md_import_device() returned %ld\n",
  4690. PTR_ERR(rdev));
  4691. return PTR_ERR(rdev);
  4692. }
  4693. rdev->desc_nr = info->number;
  4694. if (info->raid_disk < mddev->raid_disks)
  4695. rdev->raid_disk = info->raid_disk;
  4696. else
  4697. rdev->raid_disk = -1;
  4698. if (rdev->raid_disk < mddev->raid_disks)
  4699. if (info->state & (1<<MD_DISK_SYNC))
  4700. set_bit(In_sync, &rdev->flags);
  4701. if (info->state & (1<<MD_DISK_WRITEMOSTLY))
  4702. set_bit(WriteMostly, &rdev->flags);
  4703. if (!mddev->persistent) {
  4704. printk(KERN_INFO "md: nonpersistent superblock ...\n");
  4705. rdev->sb_start = i_size_read(rdev->bdev->bd_inode) / 512;
  4706. } else
  4707. rdev->sb_start = calc_dev_sboffset(rdev);
  4708. rdev->sectors = rdev->sb_start;
  4709. err = bind_rdev_to_array(rdev, mddev);
  4710. if (err) {
  4711. export_rdev(rdev);
  4712. return err;
  4713. }
  4714. }
  4715. return 0;
  4716. }
  4717. static int hot_remove_disk(mddev_t * mddev, dev_t dev)
  4718. {
  4719. char b[BDEVNAME_SIZE];
  4720. mdk_rdev_t *rdev;
  4721. rdev = find_rdev(mddev, dev);
  4722. if (!rdev)
  4723. return -ENXIO;
  4724. if (rdev->raid_disk >= 0)
  4725. goto busy;
  4726. kick_rdev_from_array(rdev);
  4727. md_update_sb(mddev, 1);
  4728. md_new_event(mddev);
  4729. return 0;
  4730. busy:
  4731. printk(KERN_WARNING "md: cannot remove active disk %s from %s ...\n",
  4732. bdevname(rdev->bdev,b), mdname(mddev));
  4733. return -EBUSY;
  4734. }
  4735. static int hot_add_disk(mddev_t * mddev, dev_t dev)
  4736. {
  4737. char b[BDEVNAME_SIZE];
  4738. int err;
  4739. mdk_rdev_t *rdev;
  4740. if (!mddev->pers)
  4741. return -ENODEV;
  4742. if (mddev->major_version != 0) {
  4743. printk(KERN_WARNING "%s: HOT_ADD may only be used with"
  4744. " version-0 superblocks.\n",
  4745. mdname(mddev));
  4746. return -EINVAL;
  4747. }
  4748. if (!mddev->pers->hot_add_disk) {
  4749. printk(KERN_WARNING
  4750. "%s: personality does not support diskops!\n",
  4751. mdname(mddev));
  4752. return -EINVAL;
  4753. }
  4754. rdev = md_import_device(dev, -1, 0);
  4755. if (IS_ERR(rdev)) {
  4756. printk(KERN_WARNING
  4757. "md: error, md_import_device() returned %ld\n",
  4758. PTR_ERR(rdev));
  4759. return -EINVAL;
  4760. }
  4761. if (mddev->persistent)
  4762. rdev->sb_start = calc_dev_sboffset(rdev);
  4763. else
  4764. rdev->sb_start = i_size_read(rdev->bdev->bd_inode) / 512;
  4765. rdev->sectors = rdev->sb_start;
  4766. if (test_bit(Faulty, &rdev->flags)) {
  4767. printk(KERN_WARNING
  4768. "md: can not hot-add faulty %s disk to %s!\n",
  4769. bdevname(rdev->bdev,b), mdname(mddev));
  4770. err = -EINVAL;
  4771. goto abort_export;
  4772. }
  4773. clear_bit(In_sync, &rdev->flags);
  4774. rdev->desc_nr = -1;
  4775. rdev->saved_raid_disk = -1;
  4776. err = bind_rdev_to_array(rdev, mddev);
  4777. if (err)
  4778. goto abort_export;
  4779. /*
  4780. * The rest should better be atomic, we can have disk failures
  4781. * noticed in interrupt contexts ...
  4782. */
  4783. rdev->raid_disk = -1;
  4784. md_update_sb(mddev, 1);
  4785. /*
  4786. * Kick recovery, maybe this spare has to be added to the
  4787. * array immediately.
  4788. */
  4789. set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
  4790. md_wakeup_thread(mddev->thread);
  4791. md_new_event(mddev);
  4792. return 0;
  4793. abort_export:
  4794. export_rdev(rdev);
  4795. return err;
  4796. }
  4797. static int set_bitmap_file(mddev_t *mddev, int fd)
  4798. {
  4799. int err;
  4800. if (mddev->pers) {
  4801. if (!mddev->pers->quiesce)
  4802. return -EBUSY;
  4803. if (mddev->recovery || mddev->sync_thread)
  4804. return -EBUSY;
  4805. /* we should be able to change the bitmap.. */
  4806. }
  4807. if (fd >= 0) {
  4808. if (mddev->bitmap)
  4809. return -EEXIST; /* cannot add when bitmap is present */
  4810. mddev->bitmap_info.file = fget(fd);
  4811. if (mddev->bitmap_info.file == NULL) {
  4812. printk(KERN_ERR "%s: error: failed to get bitmap file\n",
  4813. mdname(mddev));
  4814. return -EBADF;
  4815. }
  4816. err = deny_bitmap_write_access(mddev->bitmap_info.file);
  4817. if (err) {
  4818. printk(KERN_ERR "%s: error: bitmap file is already in use\n",
  4819. mdname(mddev));
  4820. fput(mddev->bitmap_info.file);
  4821. mddev->bitmap_info.file = NULL;
  4822. return err;
  4823. }
  4824. mddev->bitmap_info.offset = 0; /* file overrides offset */
  4825. } else if (mddev->bitmap == NULL)
  4826. return -ENOENT; /* cannot remove what isn't there */
  4827. err = 0;
  4828. if (mddev->pers) {
  4829. mddev->pers->quiesce(mddev, 1);
  4830. if (fd >= 0) {
  4831. err = bitmap_create(mddev);
  4832. if (!err)
  4833. err = bitmap_load(mddev);
  4834. }
  4835. if (fd < 0 || err) {
  4836. bitmap_destroy(mddev);
  4837. fd = -1; /* make sure to put the file */
  4838. }
  4839. mddev->pers->quiesce(mddev, 0);
  4840. }
  4841. if (fd < 0) {
  4842. if (mddev->bitmap_info.file) {
  4843. restore_bitmap_write_access(mddev->bitmap_info.file);
  4844. fput(mddev->bitmap_info.file);
  4845. }
  4846. mddev->bitmap_info.file = NULL;
  4847. }
  4848. return err;
  4849. }
  4850. /*
  4851. * set_array_info is used two different ways
  4852. * The original usage is when creating a new array.
  4853. * In this usage, raid_disks is > 0 and it together with
  4854. * level, size, not_persistent,layout,chunksize determine the
  4855. * shape of the array.
  4856. * This will always create an array with a type-0.90.0 superblock.
  4857. * The newer usage is when assembling an array.
  4858. * In this case raid_disks will be 0, and the major_version field is
  4859. * use to determine which style super-blocks are to be found on the devices.
  4860. * The minor and patch _version numbers are also kept incase the
  4861. * super_block handler wishes to interpret them.
  4862. */
  4863. static int set_array_info(mddev_t * mddev, mdu_array_info_t *info)
  4864. {
  4865. if (info->raid_disks == 0) {
  4866. /* just setting version number for superblock loading */
  4867. if (info->major_version < 0 ||
  4868. info->major_version >= ARRAY_SIZE(super_types) ||
  4869. super_types[info->major_version].name == NULL) {
  4870. /* maybe try to auto-load a module? */
  4871. printk(KERN_INFO
  4872. "md: superblock version %d not known\n",
  4873. info->major_version);
  4874. return -EINVAL;
  4875. }
  4876. mddev->major_version = info->major_version;
  4877. mddev->minor_version = info->minor_version;
  4878. mddev->patch_version = info->patch_version;
  4879. mddev->persistent = !info->not_persistent;
  4880. /* ensure mddev_put doesn't delete this now that there
  4881. * is some minimal configuration.
  4882. */
  4883. mddev->ctime = get_seconds();
  4884. return 0;
  4885. }
  4886. mddev->major_version = MD_MAJOR_VERSION;
  4887. mddev->minor_version = MD_MINOR_VERSION;
  4888. mddev->patch_version = MD_PATCHLEVEL_VERSION;
  4889. mddev->ctime = get_seconds();
  4890. mddev->level = info->level;
  4891. mddev->clevel[0] = 0;
  4892. mddev->dev_sectors = 2 * (sector_t)info->size;
  4893. mddev->raid_disks = info->raid_disks;
  4894. /* don't set md_minor, it is determined by which /dev/md* was
  4895. * openned
  4896. */
  4897. if (info->state & (1<<MD_SB_CLEAN))
  4898. mddev->recovery_cp = MaxSector;
  4899. else
  4900. mddev->recovery_cp = 0;
  4901. mddev->persistent = ! info->not_persistent;
  4902. mddev->external = 0;
  4903. mddev->layout = info->layout;
  4904. mddev->chunk_sectors = info->chunk_size >> 9;
  4905. mddev->max_disks = MD_SB_DISKS;
  4906. if (mddev->persistent)
  4907. mddev->flags = 0;
  4908. set_bit(MD_CHANGE_DEVS, &mddev->flags);
  4909. mddev->bitmap_info.default_offset = MD_SB_BYTES >> 9;
  4910. mddev->bitmap_info.offset = 0;
  4911. mddev->reshape_position = MaxSector;
  4912. /*
  4913. * Generate a 128 bit UUID
  4914. */
  4915. get_random_bytes(mddev->uuid, 16);
  4916. mddev->new_level = mddev->level;
  4917. mddev->new_chunk_sectors = mddev->chunk_sectors;
  4918. mddev->new_layout = mddev->layout;
  4919. mddev->delta_disks = 0;
  4920. return 0;
  4921. }
  4922. void md_set_array_sectors(mddev_t *mddev, sector_t array_sectors)
  4923. {
  4924. WARN(!mddev_is_locked(mddev), "%s: unlocked mddev!\n", __func__);
  4925. if (mddev->external_size)
  4926. return;
  4927. mddev->array_sectors = array_sectors;
  4928. }
  4929. EXPORT_SYMBOL(md_set_array_sectors);
  4930. static int update_size(mddev_t *mddev, sector_t num_sectors)
  4931. {
  4932. mdk_rdev_t *rdev;
  4933. int rv;
  4934. int fit = (num_sectors == 0);
  4935. if (mddev->pers->resize == NULL)
  4936. return -EINVAL;
  4937. /* The "num_sectors" is the number of sectors of each device that
  4938. * is used. This can only make sense for arrays with redundancy.
  4939. * linear and raid0 always use whatever space is available. We can only
  4940. * consider changing this number if no resync or reconstruction is
  4941. * happening, and if the new size is acceptable. It must fit before the
  4942. * sb_start or, if that is <data_offset, it must fit before the size
  4943. * of each device. If num_sectors is zero, we find the largest size
  4944. * that fits.
  4945. */
  4946. if (mddev->sync_thread)
  4947. return -EBUSY;
  4948. if (mddev->bitmap)
  4949. /* Sorry, cannot grow a bitmap yet, just remove it,
  4950. * grow, and re-add.
  4951. */
  4952. return -EBUSY;
  4953. list_for_each_entry(rdev, &mddev->disks, same_set) {
  4954. sector_t avail = rdev->sectors;
  4955. if (fit && (num_sectors == 0 || num_sectors > avail))
  4956. num_sectors = avail;
  4957. if (avail < num_sectors)
  4958. return -ENOSPC;
  4959. }
  4960. rv = mddev->pers->resize(mddev, num_sectors);
  4961. if (!rv)
  4962. revalidate_disk(mddev->gendisk);
  4963. return rv;
  4964. }
  4965. static int update_raid_disks(mddev_t *mddev, int raid_disks)
  4966. {
  4967. int rv;
  4968. /* change the number of raid disks */
  4969. if (mddev->pers->check_reshape == NULL)
  4970. return -EINVAL;
  4971. if (raid_disks <= 0 ||
  4972. (mddev->max_disks && raid_disks >= mddev->max_disks))
  4973. return -EINVAL;
  4974. if (mddev->sync_thread || mddev->reshape_position != MaxSector)
  4975. return -EBUSY;
  4976. mddev->delta_disks = raid_disks - mddev->raid_disks;
  4977. rv = mddev->pers->check_reshape(mddev);
  4978. if (rv < 0)
  4979. mddev->delta_disks = 0;
  4980. return rv;
  4981. }
  4982. /*
  4983. * update_array_info is used to change the configuration of an
  4984. * on-line array.
  4985. * The version, ctime,level,size,raid_disks,not_persistent, layout,chunk_size
  4986. * fields in the info are checked against the array.
  4987. * Any differences that cannot be handled will cause an error.
  4988. * Normally, only one change can be managed at a time.
  4989. */
  4990. static int update_array_info(mddev_t *mddev, mdu_array_info_t *info)
  4991. {
  4992. int rv = 0;
  4993. int cnt = 0;
  4994. int state = 0;
  4995. /* calculate expected state,ignoring low bits */
  4996. if (mddev->bitmap && mddev->bitmap_info.offset)
  4997. state |= (1 << MD_SB_BITMAP_PRESENT);
  4998. if (mddev->major_version != info->major_version ||
  4999. mddev->minor_version != info->minor_version ||
  5000. /* mddev->patch_version != info->patch_version || */
  5001. mddev->ctime != info->ctime ||
  5002. mddev->level != info->level ||
  5003. /* mddev->layout != info->layout || */
  5004. !mddev->persistent != info->not_persistent||
  5005. mddev->chunk_sectors != info->chunk_size >> 9 ||
  5006. /* ignore bottom 8 bits of state, and allow SB_BITMAP_PRESENT to change */
  5007. ((state^info->state) & 0xfffffe00)
  5008. )
  5009. return -EINVAL;
  5010. /* Check there is only one change */
  5011. if (info->size >= 0 && mddev->dev_sectors / 2 != info->size)
  5012. cnt++;
  5013. if (mddev->raid_disks != info->raid_disks)
  5014. cnt++;
  5015. if (mddev->layout != info->layout)
  5016. cnt++;
  5017. if ((state ^ info->state) & (1<<MD_SB_BITMAP_PRESENT))
  5018. cnt++;
  5019. if (cnt == 0)
  5020. return 0;
  5021. if (cnt > 1)
  5022. return -EINVAL;
  5023. if (mddev->layout != info->layout) {
  5024. /* Change layout
  5025. * we don't need to do anything at the md level, the
  5026. * personality will take care of it all.
  5027. */
  5028. if (mddev->pers->check_reshape == NULL)
  5029. return -EINVAL;
  5030. else {
  5031. mddev->new_layout = info->layout;
  5032. rv = mddev->pers->check_reshape(mddev);
  5033. if (rv)
  5034. mddev->new_layout = mddev->layout;
  5035. return rv;
  5036. }
  5037. }
  5038. if (info->size >= 0 && mddev->dev_sectors / 2 != info->size)
  5039. rv = update_size(mddev, (sector_t)info->size * 2);
  5040. if (mddev->raid_disks != info->raid_disks)
  5041. rv = update_raid_disks(mddev, info->raid_disks);
  5042. if ((state ^ info->state) & (1<<MD_SB_BITMAP_PRESENT)) {
  5043. if (mddev->pers->quiesce == NULL)
  5044. return -EINVAL;
  5045. if (mddev->recovery || mddev->sync_thread)
  5046. return -EBUSY;
  5047. if (info->state & (1<<MD_SB_BITMAP_PRESENT)) {
  5048. /* add the bitmap */
  5049. if (mddev->bitmap)
  5050. return -EEXIST;
  5051. if (mddev->bitmap_info.default_offset == 0)
  5052. return -EINVAL;
  5053. mddev->bitmap_info.offset =
  5054. mddev->bitmap_info.default_offset;
  5055. mddev->pers->quiesce(mddev, 1);
  5056. rv = bitmap_create(mddev);
  5057. if (!rv)
  5058. rv = bitmap_load(mddev);
  5059. if (rv)
  5060. bitmap_destroy(mddev);
  5061. mddev->pers->quiesce(mddev, 0);
  5062. } else {
  5063. /* remove the bitmap */
  5064. if (!mddev->bitmap)
  5065. return -ENOENT;
  5066. if (mddev->bitmap->file)
  5067. return -EINVAL;
  5068. mddev->pers->quiesce(mddev, 1);
  5069. bitmap_destroy(mddev);
  5070. mddev->pers->quiesce(mddev, 0);
  5071. mddev->bitmap_info.offset = 0;
  5072. }
  5073. }
  5074. md_update_sb(mddev, 1);
  5075. return rv;
  5076. }
  5077. static int set_disk_faulty(mddev_t *mddev, dev_t dev)
  5078. {
  5079. mdk_rdev_t *rdev;
  5080. if (mddev->pers == NULL)
  5081. return -ENODEV;
  5082. rdev = find_rdev(mddev, dev);
  5083. if (!rdev)
  5084. return -ENODEV;
  5085. md_error(mddev, rdev);
  5086. return 0;
  5087. }
  5088. /*
  5089. * We have a problem here : there is no easy way to give a CHS
  5090. * virtual geometry. We currently pretend that we have a 2 heads
  5091. * 4 sectors (with a BIG number of cylinders...). This drives
  5092. * dosfs just mad... ;-)
  5093. */
  5094. static int md_getgeo(struct block_device *bdev, struct hd_geometry *geo)
  5095. {
  5096. mddev_t *mddev = bdev->bd_disk->private_data;
  5097. geo->heads = 2;
  5098. geo->sectors = 4;
  5099. geo->cylinders = mddev->array_sectors / 8;
  5100. return 0;
  5101. }
  5102. static int md_ioctl(struct block_device *bdev, fmode_t mode,
  5103. unsigned int cmd, unsigned long arg)
  5104. {
  5105. int err = 0;
  5106. void __user *argp = (void __user *)arg;
  5107. mddev_t *mddev = NULL;
  5108. int ro;
  5109. if (!capable(CAP_SYS_ADMIN))
  5110. return -EACCES;
  5111. /*
  5112. * Commands dealing with the RAID driver but not any
  5113. * particular array:
  5114. */
  5115. switch (cmd)
  5116. {
  5117. case RAID_VERSION:
  5118. err = get_version(argp);
  5119. goto done;
  5120. case PRINT_RAID_DEBUG:
  5121. err = 0;
  5122. md_print_devices();
  5123. goto done;
  5124. #ifndef MODULE
  5125. case RAID_AUTORUN:
  5126. err = 0;
  5127. autostart_arrays(arg);
  5128. goto done;
  5129. #endif
  5130. default:;
  5131. }
  5132. /*
  5133. * Commands creating/starting a new array:
  5134. */
  5135. mddev = bdev->bd_disk->private_data;
  5136. if (!mddev) {
  5137. BUG();
  5138. goto abort;
  5139. }
  5140. err = mddev_lock(mddev);
  5141. if (err) {
  5142. printk(KERN_INFO
  5143. "md: ioctl lock interrupted, reason %d, cmd %d\n",
  5144. err, cmd);
  5145. goto abort;
  5146. }
  5147. switch (cmd)
  5148. {
  5149. case SET_ARRAY_INFO:
  5150. {
  5151. mdu_array_info_t info;
  5152. if (!arg)
  5153. memset(&info, 0, sizeof(info));
  5154. else if (copy_from_user(&info, argp, sizeof(info))) {
  5155. err = -EFAULT;
  5156. goto abort_unlock;
  5157. }
  5158. if (mddev->pers) {
  5159. err = update_array_info(mddev, &info);
  5160. if (err) {
  5161. printk(KERN_WARNING "md: couldn't update"
  5162. " array info. %d\n", err);
  5163. goto abort_unlock;
  5164. }
  5165. goto done_unlock;
  5166. }
  5167. if (!list_empty(&mddev->disks)) {
  5168. printk(KERN_WARNING
  5169. "md: array %s already has disks!\n",
  5170. mdname(mddev));
  5171. err = -EBUSY;
  5172. goto abort_unlock;
  5173. }
  5174. if (mddev->raid_disks) {
  5175. printk(KERN_WARNING
  5176. "md: array %s already initialised!\n",
  5177. mdname(mddev));
  5178. err = -EBUSY;
  5179. goto abort_unlock;
  5180. }
  5181. err = set_array_info(mddev, &info);
  5182. if (err) {
  5183. printk(KERN_WARNING "md: couldn't set"
  5184. " array info. %d\n", err);
  5185. goto abort_unlock;
  5186. }
  5187. }
  5188. goto done_unlock;
  5189. default:;
  5190. }
  5191. /*
  5192. * Commands querying/configuring an existing array:
  5193. */
  5194. /* if we are not initialised yet, only ADD_NEW_DISK, STOP_ARRAY,
  5195. * RUN_ARRAY, and GET_ and SET_BITMAP_FILE are allowed */
  5196. if ((!mddev->raid_disks && !mddev->external)
  5197. && cmd != ADD_NEW_DISK && cmd != STOP_ARRAY
  5198. && cmd != RUN_ARRAY && cmd != SET_BITMAP_FILE
  5199. && cmd != GET_BITMAP_FILE) {
  5200. err = -ENODEV;
  5201. goto abort_unlock;
  5202. }
  5203. /*
  5204. * Commands even a read-only array can execute:
  5205. */
  5206. switch (cmd)
  5207. {
  5208. case GET_ARRAY_INFO:
  5209. err = get_array_info(mddev, argp);
  5210. goto done_unlock;
  5211. case GET_BITMAP_FILE:
  5212. err = get_bitmap_file(mddev, argp);
  5213. goto done_unlock;
  5214. case GET_DISK_INFO:
  5215. err = get_disk_info(mddev, argp);
  5216. goto done_unlock;
  5217. case RESTART_ARRAY_RW:
  5218. err = restart_array(mddev);
  5219. goto done_unlock;
  5220. case STOP_ARRAY:
  5221. err = do_md_stop(mddev, 0, 1);
  5222. goto done_unlock;
  5223. case STOP_ARRAY_RO:
  5224. err = md_set_readonly(mddev, 1);
  5225. goto done_unlock;
  5226. case BLKROSET:
  5227. if (get_user(ro, (int __user *)(arg))) {
  5228. err = -EFAULT;
  5229. goto done_unlock;
  5230. }
  5231. err = -EINVAL;
  5232. /* if the bdev is going readonly the value of mddev->ro
  5233. * does not matter, no writes are coming
  5234. */
  5235. if (ro)
  5236. goto done_unlock;
  5237. /* are we are already prepared for writes? */
  5238. if (mddev->ro != 1)
  5239. goto done_unlock;
  5240. /* transitioning to readauto need only happen for
  5241. * arrays that call md_write_start
  5242. */
  5243. if (mddev->pers) {
  5244. err = restart_array(mddev);
  5245. if (err == 0) {
  5246. mddev->ro = 2;
  5247. set_disk_ro(mddev->gendisk, 0);
  5248. }
  5249. }
  5250. goto done_unlock;
  5251. }
  5252. /*
  5253. * The remaining ioctls are changing the state of the
  5254. * superblock, so we do not allow them on read-only arrays.
  5255. * However non-MD ioctls (e.g. get-size) will still come through
  5256. * here and hit the 'default' below, so only disallow
  5257. * 'md' ioctls, and switch to rw mode if started auto-readonly.
  5258. */
  5259. if (_IOC_TYPE(cmd) == MD_MAJOR && mddev->ro && mddev->pers) {
  5260. if (mddev->ro == 2) {
  5261. mddev->ro = 0;
  5262. sysfs_notify_dirent_safe(mddev->sysfs_state);
  5263. set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
  5264. md_wakeup_thread(mddev->thread);
  5265. } else {
  5266. err = -EROFS;
  5267. goto abort_unlock;
  5268. }
  5269. }
  5270. switch (cmd)
  5271. {
  5272. case ADD_NEW_DISK:
  5273. {
  5274. mdu_disk_info_t info;
  5275. if (copy_from_user(&info, argp, sizeof(info)))
  5276. err = -EFAULT;
  5277. else
  5278. err = add_new_disk(mddev, &info);
  5279. goto done_unlock;
  5280. }
  5281. case HOT_REMOVE_DISK:
  5282. err = hot_remove_disk(mddev, new_decode_dev(arg));
  5283. goto done_unlock;
  5284. case HOT_ADD_DISK:
  5285. err = hot_add_disk(mddev, new_decode_dev(arg));
  5286. goto done_unlock;
  5287. case SET_DISK_FAULTY:
  5288. err = set_disk_faulty(mddev, new_decode_dev(arg));
  5289. goto done_unlock;
  5290. case RUN_ARRAY:
  5291. err = do_md_run(mddev);
  5292. goto done_unlock;
  5293. case SET_BITMAP_FILE:
  5294. err = set_bitmap_file(mddev, (int)arg);
  5295. goto done_unlock;
  5296. default:
  5297. err = -EINVAL;
  5298. goto abort_unlock;
  5299. }
  5300. done_unlock:
  5301. abort_unlock:
  5302. if (mddev->hold_active == UNTIL_IOCTL &&
  5303. err != -EINVAL)
  5304. mddev->hold_active = 0;
  5305. mddev_unlock(mddev);
  5306. return err;
  5307. done:
  5308. if (err)
  5309. MD_BUG();
  5310. abort:
  5311. return err;
  5312. }
  5313. #ifdef CONFIG_COMPAT
  5314. static int md_compat_ioctl(struct block_device *bdev, fmode_t mode,
  5315. unsigned int cmd, unsigned long arg)
  5316. {
  5317. switch (cmd) {
  5318. case HOT_REMOVE_DISK:
  5319. case HOT_ADD_DISK:
  5320. case SET_DISK_FAULTY:
  5321. case SET_BITMAP_FILE:
  5322. /* These take in integer arg, do not convert */
  5323. break;
  5324. default:
  5325. arg = (unsigned long)compat_ptr(arg);
  5326. break;
  5327. }
  5328. return md_ioctl(bdev, mode, cmd, arg);
  5329. }
  5330. #endif /* CONFIG_COMPAT */
  5331. static int md_open(struct block_device *bdev, fmode_t mode)
  5332. {
  5333. /*
  5334. * Succeed if we can lock the mddev, which confirms that
  5335. * it isn't being stopped right now.
  5336. */
  5337. mddev_t *mddev = mddev_find(bdev->bd_dev);
  5338. int err;
  5339. if (mddev->gendisk != bdev->bd_disk) {
  5340. /* we are racing with mddev_put which is discarding this
  5341. * bd_disk.
  5342. */
  5343. mddev_put(mddev);
  5344. /* Wait until bdev->bd_disk is definitely gone */
  5345. flush_workqueue(md_misc_wq);
  5346. /* Then retry the open from the top */
  5347. return -ERESTARTSYS;
  5348. }
  5349. BUG_ON(mddev != bdev->bd_disk->private_data);
  5350. if ((err = mutex_lock_interruptible(&mddev->open_mutex)))
  5351. goto out;
  5352. err = 0;
  5353. atomic_inc(&mddev->openers);
  5354. mutex_unlock(&mddev->open_mutex);
  5355. check_disk_change(bdev);
  5356. out:
  5357. return err;
  5358. }
  5359. static int md_release(struct gendisk *disk, fmode_t mode)
  5360. {
  5361. mddev_t *mddev = disk->private_data;
  5362. BUG_ON(!mddev);
  5363. atomic_dec(&mddev->openers);
  5364. mddev_put(mddev);
  5365. return 0;
  5366. }
  5367. static int md_media_changed(struct gendisk *disk)
  5368. {
  5369. mddev_t *mddev = disk->private_data;
  5370. return mddev->changed;
  5371. }
  5372. static int md_revalidate(struct gendisk *disk)
  5373. {
  5374. mddev_t *mddev = disk->private_data;
  5375. mddev->changed = 0;
  5376. return 0;
  5377. }
  5378. static const struct block_device_operations md_fops =
  5379. {
  5380. .owner = THIS_MODULE,
  5381. .open = md_open,
  5382. .release = md_release,
  5383. .ioctl = md_ioctl,
  5384. #ifdef CONFIG_COMPAT
  5385. .compat_ioctl = md_compat_ioctl,
  5386. #endif
  5387. .getgeo = md_getgeo,
  5388. .media_changed = md_media_changed,
  5389. .revalidate_disk= md_revalidate,
  5390. };
  5391. static int md_thread(void * arg)
  5392. {
  5393. mdk_thread_t *thread = arg;
  5394. /*
  5395. * md_thread is a 'system-thread', it's priority should be very
  5396. * high. We avoid resource deadlocks individually in each
  5397. * raid personality. (RAID5 does preallocation) We also use RR and
  5398. * the very same RT priority as kswapd, thus we will never get
  5399. * into a priority inversion deadlock.
  5400. *
  5401. * we definitely have to have equal or higher priority than
  5402. * bdflush, otherwise bdflush will deadlock if there are too
  5403. * many dirty RAID5 blocks.
  5404. */
  5405. allow_signal(SIGKILL);
  5406. while (!kthread_should_stop()) {
  5407. /* We need to wait INTERRUPTIBLE so that
  5408. * we don't add to the load-average.
  5409. * That means we need to be sure no signals are
  5410. * pending
  5411. */
  5412. if (signal_pending(current))
  5413. flush_signals(current);
  5414. wait_event_interruptible_timeout
  5415. (thread->wqueue,
  5416. test_bit(THREAD_WAKEUP, &thread->flags)
  5417. || kthread_should_stop(),
  5418. thread->timeout);
  5419. clear_bit(THREAD_WAKEUP, &thread->flags);
  5420. if (!kthread_should_stop())
  5421. thread->run(thread->mddev);
  5422. }
  5423. return 0;
  5424. }
  5425. void md_wakeup_thread(mdk_thread_t *thread)
  5426. {
  5427. if (thread) {
  5428. dprintk("md: waking up MD thread %s.\n", thread->tsk->comm);
  5429. set_bit(THREAD_WAKEUP, &thread->flags);
  5430. wake_up(&thread->wqueue);
  5431. }
  5432. }
  5433. mdk_thread_t *md_register_thread(void (*run) (mddev_t *), mddev_t *mddev,
  5434. const char *name)
  5435. {
  5436. mdk_thread_t *thread;
  5437. thread = kzalloc(sizeof(mdk_thread_t), GFP_KERNEL);
  5438. if (!thread)
  5439. return NULL;
  5440. init_waitqueue_head(&thread->wqueue);
  5441. thread->run = run;
  5442. thread->mddev = mddev;
  5443. thread->timeout = MAX_SCHEDULE_TIMEOUT;
  5444. thread->tsk = kthread_run(md_thread, thread,
  5445. "%s_%s",
  5446. mdname(thread->mddev),
  5447. name ?: mddev->pers->name);
  5448. if (IS_ERR(thread->tsk)) {
  5449. kfree(thread);
  5450. return NULL;
  5451. }
  5452. return thread;
  5453. }
  5454. void md_unregister_thread(mdk_thread_t *thread)
  5455. {
  5456. if (!thread)
  5457. return;
  5458. dprintk("interrupting MD-thread pid %d\n", task_pid_nr(thread->tsk));
  5459. kthread_stop(thread->tsk);
  5460. kfree(thread);
  5461. }
  5462. void md_error(mddev_t *mddev, mdk_rdev_t *rdev)
  5463. {
  5464. if (!mddev) {
  5465. MD_BUG();
  5466. return;
  5467. }
  5468. if (!rdev || test_bit(Faulty, &rdev->flags))
  5469. return;
  5470. if (mddev->external)
  5471. set_bit(Blocked, &rdev->flags);
  5472. /*
  5473. dprintk("md_error dev:%s, rdev:(%d:%d), (caller: %p,%p,%p,%p).\n",
  5474. mdname(mddev),
  5475. MAJOR(rdev->bdev->bd_dev), MINOR(rdev->bdev->bd_dev),
  5476. __builtin_return_address(0),__builtin_return_address(1),
  5477. __builtin_return_address(2),__builtin_return_address(3));
  5478. */
  5479. if (!mddev->pers)
  5480. return;
  5481. if (!mddev->pers->error_handler)
  5482. return;
  5483. mddev->pers->error_handler(mddev,rdev);
  5484. if (mddev->degraded)
  5485. set_bit(MD_RECOVERY_RECOVER, &mddev->recovery);
  5486. sysfs_notify_dirent_safe(rdev->sysfs_state);
  5487. set_bit(MD_RECOVERY_INTR, &mddev->recovery);
  5488. set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
  5489. md_wakeup_thread(mddev->thread);
  5490. if (mddev->event_work.func)
  5491. queue_work(md_misc_wq, &mddev->event_work);
  5492. md_new_event_inintr(mddev);
  5493. }
  5494. /* seq_file implementation /proc/mdstat */
  5495. static void status_unused(struct seq_file *seq)
  5496. {
  5497. int i = 0;
  5498. mdk_rdev_t *rdev;
  5499. seq_printf(seq, "unused devices: ");
  5500. list_for_each_entry(rdev, &pending_raid_disks, same_set) {
  5501. char b[BDEVNAME_SIZE];
  5502. i++;
  5503. seq_printf(seq, "%s ",
  5504. bdevname(rdev->bdev,b));
  5505. }
  5506. if (!i)
  5507. seq_printf(seq, "<none>");
  5508. seq_printf(seq, "\n");
  5509. }
  5510. static void status_resync(struct seq_file *seq, mddev_t * mddev)
  5511. {
  5512. sector_t max_sectors, resync, res;
  5513. unsigned long dt, db;
  5514. sector_t rt;
  5515. int scale;
  5516. unsigned int per_milli;
  5517. resync = mddev->curr_resync - atomic_read(&mddev->recovery_active);
  5518. if (test_bit(MD_RECOVERY_SYNC, &mddev->recovery))
  5519. max_sectors = mddev->resync_max_sectors;
  5520. else
  5521. max_sectors = mddev->dev_sectors;
  5522. /*
  5523. * Should not happen.
  5524. */
  5525. if (!max_sectors) {
  5526. MD_BUG();
  5527. return;
  5528. }
  5529. /* Pick 'scale' such that (resync>>scale)*1000 will fit
  5530. * in a sector_t, and (max_sectors>>scale) will fit in a
  5531. * u32, as those are the requirements for sector_div.
  5532. * Thus 'scale' must be at least 10
  5533. */
  5534. scale = 10;
  5535. if (sizeof(sector_t) > sizeof(unsigned long)) {
  5536. while ( max_sectors/2 > (1ULL<<(scale+32)))
  5537. scale++;
  5538. }
  5539. res = (resync>>scale)*1000;
  5540. sector_div(res, (u32)((max_sectors>>scale)+1));
  5541. per_milli = res;
  5542. {
  5543. int i, x = per_milli/50, y = 20-x;
  5544. seq_printf(seq, "[");
  5545. for (i = 0; i < x; i++)
  5546. seq_printf(seq, "=");
  5547. seq_printf(seq, ">");
  5548. for (i = 0; i < y; i++)
  5549. seq_printf(seq, ".");
  5550. seq_printf(seq, "] ");
  5551. }
  5552. seq_printf(seq, " %s =%3u.%u%% (%llu/%llu)",
  5553. (test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery)?
  5554. "reshape" :
  5555. (test_bit(MD_RECOVERY_CHECK, &mddev->recovery)?
  5556. "check" :
  5557. (test_bit(MD_RECOVERY_SYNC, &mddev->recovery) ?
  5558. "resync" : "recovery"))),
  5559. per_milli/10, per_milli % 10,
  5560. (unsigned long long) resync/2,
  5561. (unsigned long long) max_sectors/2);
  5562. /*
  5563. * dt: time from mark until now
  5564. * db: blocks written from mark until now
  5565. * rt: remaining time
  5566. *
  5567. * rt is a sector_t, so could be 32bit or 64bit.
  5568. * So we divide before multiply in case it is 32bit and close
  5569. * to the limit.
  5570. * We scale the divisor (db) by 32 to avoid losing precision
  5571. * near the end of resync when the number of remaining sectors
  5572. * is close to 'db'.
  5573. * We then divide rt by 32 after multiplying by db to compensate.
  5574. * The '+1' avoids division by zero if db is very small.
  5575. */
  5576. dt = ((jiffies - mddev->resync_mark) / HZ);
  5577. if (!dt) dt++;
  5578. db = (mddev->curr_mark_cnt - atomic_read(&mddev->recovery_active))
  5579. - mddev->resync_mark_cnt;
  5580. rt = max_sectors - resync; /* number of remaining sectors */
  5581. sector_div(rt, db/32+1);
  5582. rt *= dt;
  5583. rt >>= 5;
  5584. seq_printf(seq, " finish=%lu.%lumin", (unsigned long)rt / 60,
  5585. ((unsigned long)rt % 60)/6);
  5586. seq_printf(seq, " speed=%ldK/sec", db/2/dt);
  5587. }
  5588. static void *md_seq_start(struct seq_file *seq, loff_t *pos)
  5589. {
  5590. struct list_head *tmp;
  5591. loff_t l = *pos;
  5592. mddev_t *mddev;
  5593. if (l >= 0x10000)
  5594. return NULL;
  5595. if (!l--)
  5596. /* header */
  5597. return (void*)1;
  5598. spin_lock(&all_mddevs_lock);
  5599. list_for_each(tmp,&all_mddevs)
  5600. if (!l--) {
  5601. mddev = list_entry(tmp, mddev_t, all_mddevs);
  5602. mddev_get(mddev);
  5603. spin_unlock(&all_mddevs_lock);
  5604. return mddev;
  5605. }
  5606. spin_unlock(&all_mddevs_lock);
  5607. if (!l--)
  5608. return (void*)2;/* tail */
  5609. return NULL;
  5610. }
  5611. static void *md_seq_next(struct seq_file *seq, void *v, loff_t *pos)
  5612. {
  5613. struct list_head *tmp;
  5614. mddev_t *next_mddev, *mddev = v;
  5615. ++*pos;
  5616. if (v == (void*)2)
  5617. return NULL;
  5618. spin_lock(&all_mddevs_lock);
  5619. if (v == (void*)1)
  5620. tmp = all_mddevs.next;
  5621. else
  5622. tmp = mddev->all_mddevs.next;
  5623. if (tmp != &all_mddevs)
  5624. next_mddev = mddev_get(list_entry(tmp,mddev_t,all_mddevs));
  5625. else {
  5626. next_mddev = (void*)2;
  5627. *pos = 0x10000;
  5628. }
  5629. spin_unlock(&all_mddevs_lock);
  5630. if (v != (void*)1)
  5631. mddev_put(mddev);
  5632. return next_mddev;
  5633. }
  5634. static void md_seq_stop(struct seq_file *seq, void *v)
  5635. {
  5636. mddev_t *mddev = v;
  5637. if (mddev && v != (void*)1 && v != (void*)2)
  5638. mddev_put(mddev);
  5639. }
  5640. struct mdstat_info {
  5641. int event;
  5642. };
  5643. static int md_seq_show(struct seq_file *seq, void *v)
  5644. {
  5645. mddev_t *mddev = v;
  5646. sector_t sectors;
  5647. mdk_rdev_t *rdev;
  5648. struct mdstat_info *mi = seq->private;
  5649. struct bitmap *bitmap;
  5650. if (v == (void*)1) {
  5651. struct mdk_personality *pers;
  5652. seq_printf(seq, "Personalities : ");
  5653. spin_lock(&pers_lock);
  5654. list_for_each_entry(pers, &pers_list, list)
  5655. seq_printf(seq, "[%s] ", pers->name);
  5656. spin_unlock(&pers_lock);
  5657. seq_printf(seq, "\n");
  5658. mi->event = atomic_read(&md_event_count);
  5659. return 0;
  5660. }
  5661. if (v == (void*)2) {
  5662. status_unused(seq);
  5663. return 0;
  5664. }
  5665. if (mddev_lock(mddev) < 0)
  5666. return -EINTR;
  5667. if (mddev->pers || mddev->raid_disks || !list_empty(&mddev->disks)) {
  5668. seq_printf(seq, "%s : %sactive", mdname(mddev),
  5669. mddev->pers ? "" : "in");
  5670. if (mddev->pers) {
  5671. if (mddev->ro==1)
  5672. seq_printf(seq, " (read-only)");
  5673. if (mddev->ro==2)
  5674. seq_printf(seq, " (auto-read-only)");
  5675. seq_printf(seq, " %s", mddev->pers->name);
  5676. }
  5677. sectors = 0;
  5678. list_for_each_entry(rdev, &mddev->disks, same_set) {
  5679. char b[BDEVNAME_SIZE];
  5680. seq_printf(seq, " %s[%d]",
  5681. bdevname(rdev->bdev,b), rdev->desc_nr);
  5682. if (test_bit(WriteMostly, &rdev->flags))
  5683. seq_printf(seq, "(W)");
  5684. if (test_bit(Faulty, &rdev->flags)) {
  5685. seq_printf(seq, "(F)");
  5686. continue;
  5687. } else if (rdev->raid_disk < 0)
  5688. seq_printf(seq, "(S)"); /* spare */
  5689. sectors += rdev->sectors;
  5690. }
  5691. if (!list_empty(&mddev->disks)) {
  5692. if (mddev->pers)
  5693. seq_printf(seq, "\n %llu blocks",
  5694. (unsigned long long)
  5695. mddev->array_sectors / 2);
  5696. else
  5697. seq_printf(seq, "\n %llu blocks",
  5698. (unsigned long long)sectors / 2);
  5699. }
  5700. if (mddev->persistent) {
  5701. if (mddev->major_version != 0 ||
  5702. mddev->minor_version != 90) {
  5703. seq_printf(seq," super %d.%d",
  5704. mddev->major_version,
  5705. mddev->minor_version);
  5706. }
  5707. } else if (mddev->external)
  5708. seq_printf(seq, " super external:%s",
  5709. mddev->metadata_type);
  5710. else
  5711. seq_printf(seq, " super non-persistent");
  5712. if (mddev->pers) {
  5713. mddev->pers->status(seq, mddev);
  5714. seq_printf(seq, "\n ");
  5715. if (mddev->pers->sync_request) {
  5716. if (mddev->curr_resync > 2) {
  5717. status_resync(seq, mddev);
  5718. seq_printf(seq, "\n ");
  5719. } else if (mddev->curr_resync == 1 || mddev->curr_resync == 2)
  5720. seq_printf(seq, "\tresync=DELAYED\n ");
  5721. else if (mddev->recovery_cp < MaxSector)
  5722. seq_printf(seq, "\tresync=PENDING\n ");
  5723. }
  5724. } else
  5725. seq_printf(seq, "\n ");
  5726. if ((bitmap = mddev->bitmap)) {
  5727. unsigned long chunk_kb;
  5728. unsigned long flags;
  5729. spin_lock_irqsave(&bitmap->lock, flags);
  5730. chunk_kb = mddev->bitmap_info.chunksize >> 10;
  5731. seq_printf(seq, "bitmap: %lu/%lu pages [%luKB], "
  5732. "%lu%s chunk",
  5733. bitmap->pages - bitmap->missing_pages,
  5734. bitmap->pages,
  5735. (bitmap->pages - bitmap->missing_pages)
  5736. << (PAGE_SHIFT - 10),
  5737. chunk_kb ? chunk_kb : mddev->bitmap_info.chunksize,
  5738. chunk_kb ? "KB" : "B");
  5739. if (bitmap->file) {
  5740. seq_printf(seq, ", file: ");
  5741. seq_path(seq, &bitmap->file->f_path, " \t\n");
  5742. }
  5743. seq_printf(seq, "\n");
  5744. spin_unlock_irqrestore(&bitmap->lock, flags);
  5745. }
  5746. seq_printf(seq, "\n");
  5747. }
  5748. mddev_unlock(mddev);
  5749. return 0;
  5750. }
  5751. static const struct seq_operations md_seq_ops = {
  5752. .start = md_seq_start,
  5753. .next = md_seq_next,
  5754. .stop = md_seq_stop,
  5755. .show = md_seq_show,
  5756. };
  5757. static int md_seq_open(struct inode *inode, struct file *file)
  5758. {
  5759. int error;
  5760. struct mdstat_info *mi = kmalloc(sizeof(*mi), GFP_KERNEL);
  5761. if (mi == NULL)
  5762. return -ENOMEM;
  5763. error = seq_open(file, &md_seq_ops);
  5764. if (error)
  5765. kfree(mi);
  5766. else {
  5767. struct seq_file *p = file->private_data;
  5768. p->private = mi;
  5769. mi->event = atomic_read(&md_event_count);
  5770. }
  5771. return error;
  5772. }
  5773. static unsigned int mdstat_poll(struct file *filp, poll_table *wait)
  5774. {
  5775. struct seq_file *m = filp->private_data;
  5776. struct mdstat_info *mi = m->private;
  5777. int mask;
  5778. poll_wait(filp, &md_event_waiters, wait);
  5779. /* always allow read */
  5780. mask = POLLIN | POLLRDNORM;
  5781. if (mi->event != atomic_read(&md_event_count))
  5782. mask |= POLLERR | POLLPRI;
  5783. return mask;
  5784. }
  5785. static const struct file_operations md_seq_fops = {
  5786. .owner = THIS_MODULE,
  5787. .open = md_seq_open,
  5788. .read = seq_read,
  5789. .llseek = seq_lseek,
  5790. .release = seq_release_private,
  5791. .poll = mdstat_poll,
  5792. };
  5793. int register_md_personality(struct mdk_personality *p)
  5794. {
  5795. spin_lock(&pers_lock);
  5796. list_add_tail(&p->list, &pers_list);
  5797. printk(KERN_INFO "md: %s personality registered for level %d\n", p->name, p->level);
  5798. spin_unlock(&pers_lock);
  5799. return 0;
  5800. }
  5801. int unregister_md_personality(struct mdk_personality *p)
  5802. {
  5803. printk(KERN_INFO "md: %s personality unregistered\n", p->name);
  5804. spin_lock(&pers_lock);
  5805. list_del_init(&p->list);
  5806. spin_unlock(&pers_lock);
  5807. return 0;
  5808. }
  5809. static int is_mddev_idle(mddev_t *mddev, int init)
  5810. {
  5811. mdk_rdev_t * rdev;
  5812. int idle;
  5813. int curr_events;
  5814. idle = 1;
  5815. rcu_read_lock();
  5816. rdev_for_each_rcu(rdev, mddev) {
  5817. struct gendisk *disk = rdev->bdev->bd_contains->bd_disk;
  5818. curr_events = (int)part_stat_read(&disk->part0, sectors[0]) +
  5819. (int)part_stat_read(&disk->part0, sectors[1]) -
  5820. atomic_read(&disk->sync_io);
  5821. /* sync IO will cause sync_io to increase before the disk_stats
  5822. * as sync_io is counted when a request starts, and
  5823. * disk_stats is counted when it completes.
  5824. * So resync activity will cause curr_events to be smaller than
  5825. * when there was no such activity.
  5826. * non-sync IO will cause disk_stat to increase without
  5827. * increasing sync_io so curr_events will (eventually)
  5828. * be larger than it was before. Once it becomes
  5829. * substantially larger, the test below will cause
  5830. * the array to appear non-idle, and resync will slow
  5831. * down.
  5832. * If there is a lot of outstanding resync activity when
  5833. * we set last_event to curr_events, then all that activity
  5834. * completing might cause the array to appear non-idle
  5835. * and resync will be slowed down even though there might
  5836. * not have been non-resync activity. This will only
  5837. * happen once though. 'last_events' will soon reflect
  5838. * the state where there is little or no outstanding
  5839. * resync requests, and further resync activity will
  5840. * always make curr_events less than last_events.
  5841. *
  5842. */
  5843. if (init || curr_events - rdev->last_events > 64) {
  5844. rdev->last_events = curr_events;
  5845. idle = 0;
  5846. }
  5847. }
  5848. rcu_read_unlock();
  5849. return idle;
  5850. }
  5851. void md_done_sync(mddev_t *mddev, int blocks, int ok)
  5852. {
  5853. /* another "blocks" (512byte) blocks have been synced */
  5854. atomic_sub(blocks, &mddev->recovery_active);
  5855. wake_up(&mddev->recovery_wait);
  5856. if (!ok) {
  5857. set_bit(MD_RECOVERY_INTR, &mddev->recovery);
  5858. md_wakeup_thread(mddev->thread);
  5859. // stop recovery, signal do_sync ....
  5860. }
  5861. }
  5862. /* md_write_start(mddev, bi)
  5863. * If we need to update some array metadata (e.g. 'active' flag
  5864. * in superblock) before writing, schedule a superblock update
  5865. * and wait for it to complete.
  5866. */
  5867. void md_write_start(mddev_t *mddev, struct bio *bi)
  5868. {
  5869. int did_change = 0;
  5870. if (bio_data_dir(bi) != WRITE)
  5871. return;
  5872. BUG_ON(mddev->ro == 1);
  5873. if (mddev->ro == 2) {
  5874. /* need to switch to read/write */
  5875. mddev->ro = 0;
  5876. set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
  5877. md_wakeup_thread(mddev->thread);
  5878. md_wakeup_thread(mddev->sync_thread);
  5879. did_change = 1;
  5880. }
  5881. atomic_inc(&mddev->writes_pending);
  5882. if (mddev->safemode == 1)
  5883. mddev->safemode = 0;
  5884. if (mddev->in_sync) {
  5885. spin_lock_irq(&mddev->write_lock);
  5886. if (mddev->in_sync) {
  5887. mddev->in_sync = 0;
  5888. set_bit(MD_CHANGE_CLEAN, &mddev->flags);
  5889. set_bit(MD_CHANGE_PENDING, &mddev->flags);
  5890. md_wakeup_thread(mddev->thread);
  5891. did_change = 1;
  5892. }
  5893. spin_unlock_irq(&mddev->write_lock);
  5894. }
  5895. if (did_change)
  5896. sysfs_notify_dirent_safe(mddev->sysfs_state);
  5897. wait_event(mddev->sb_wait,
  5898. !test_bit(MD_CHANGE_PENDING, &mddev->flags));
  5899. }
  5900. void md_write_end(mddev_t *mddev)
  5901. {
  5902. if (atomic_dec_and_test(&mddev->writes_pending)) {
  5903. if (mddev->safemode == 2)
  5904. md_wakeup_thread(mddev->thread);
  5905. else if (mddev->safemode_delay)
  5906. mod_timer(&mddev->safemode_timer, jiffies + mddev->safemode_delay);
  5907. }
  5908. }
  5909. /* md_allow_write(mddev)
  5910. * Calling this ensures that the array is marked 'active' so that writes
  5911. * may proceed without blocking. It is important to call this before
  5912. * attempting a GFP_KERNEL allocation while holding the mddev lock.
  5913. * Must be called with mddev_lock held.
  5914. *
  5915. * In the ->external case MD_CHANGE_CLEAN can not be cleared until mddev->lock
  5916. * is dropped, so return -EAGAIN after notifying userspace.
  5917. */
  5918. int md_allow_write(mddev_t *mddev)
  5919. {
  5920. if (!mddev->pers)
  5921. return 0;
  5922. if (mddev->ro)
  5923. return 0;
  5924. if (!mddev->pers->sync_request)
  5925. return 0;
  5926. spin_lock_irq(&mddev->write_lock);
  5927. if (mddev->in_sync) {
  5928. mddev->in_sync = 0;
  5929. set_bit(MD_CHANGE_CLEAN, &mddev->flags);
  5930. set_bit(MD_CHANGE_PENDING, &mddev->flags);
  5931. if (mddev->safemode_delay &&
  5932. mddev->safemode == 0)
  5933. mddev->safemode = 1;
  5934. spin_unlock_irq(&mddev->write_lock);
  5935. md_update_sb(mddev, 0);
  5936. sysfs_notify_dirent_safe(mddev->sysfs_state);
  5937. } else
  5938. spin_unlock_irq(&mddev->write_lock);
  5939. if (test_bit(MD_CHANGE_PENDING, &mddev->flags))
  5940. return -EAGAIN;
  5941. else
  5942. return 0;
  5943. }
  5944. EXPORT_SYMBOL_GPL(md_allow_write);
  5945. #define SYNC_MARKS 10
  5946. #define SYNC_MARK_STEP (3*HZ)
  5947. void md_do_sync(mddev_t *mddev)
  5948. {
  5949. mddev_t *mddev2;
  5950. unsigned int currspeed = 0,
  5951. window;
  5952. sector_t max_sectors,j, io_sectors;
  5953. unsigned long mark[SYNC_MARKS];
  5954. sector_t mark_cnt[SYNC_MARKS];
  5955. int last_mark,m;
  5956. struct list_head *tmp;
  5957. sector_t last_check;
  5958. int skipped = 0;
  5959. mdk_rdev_t *rdev;
  5960. char *desc;
  5961. /* just incase thread restarts... */
  5962. if (test_bit(MD_RECOVERY_DONE, &mddev->recovery))
  5963. return;
  5964. if (mddev->ro) /* never try to sync a read-only array */
  5965. return;
  5966. if (test_bit(MD_RECOVERY_SYNC, &mddev->recovery)) {
  5967. if (test_bit(MD_RECOVERY_CHECK, &mddev->recovery))
  5968. desc = "data-check";
  5969. else if (test_bit(MD_RECOVERY_REQUESTED, &mddev->recovery))
  5970. desc = "requested-resync";
  5971. else
  5972. desc = "resync";
  5973. } else if (test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery))
  5974. desc = "reshape";
  5975. else
  5976. desc = "recovery";
  5977. /* we overload curr_resync somewhat here.
  5978. * 0 == not engaged in resync at all
  5979. * 2 == checking that there is no conflict with another sync
  5980. * 1 == like 2, but have yielded to allow conflicting resync to
  5981. * commense
  5982. * other == active in resync - this many blocks
  5983. *
  5984. * Before starting a resync we must have set curr_resync to
  5985. * 2, and then checked that every "conflicting" array has curr_resync
  5986. * less than ours. When we find one that is the same or higher
  5987. * we wait on resync_wait. To avoid deadlock, we reduce curr_resync
  5988. * to 1 if we choose to yield (based arbitrarily on address of mddev structure).
  5989. * This will mean we have to start checking from the beginning again.
  5990. *
  5991. */
  5992. do {
  5993. mddev->curr_resync = 2;
  5994. try_again:
  5995. if (kthread_should_stop())
  5996. set_bit(MD_RECOVERY_INTR, &mddev->recovery);
  5997. if (test_bit(MD_RECOVERY_INTR, &mddev->recovery))
  5998. goto skip;
  5999. for_each_mddev(mddev2, tmp) {
  6000. if (mddev2 == mddev)
  6001. continue;
  6002. if (!mddev->parallel_resync
  6003. && mddev2->curr_resync
  6004. && match_mddev_units(mddev, mddev2)) {
  6005. DEFINE_WAIT(wq);
  6006. if (mddev < mddev2 && mddev->curr_resync == 2) {
  6007. /* arbitrarily yield */
  6008. mddev->curr_resync = 1;
  6009. wake_up(&resync_wait);
  6010. }
  6011. if (mddev > mddev2 && mddev->curr_resync == 1)
  6012. /* no need to wait here, we can wait the next
  6013. * time 'round when curr_resync == 2
  6014. */
  6015. continue;
  6016. /* We need to wait 'interruptible' so as not to
  6017. * contribute to the load average, and not to
  6018. * be caught by 'softlockup'
  6019. */
  6020. prepare_to_wait(&resync_wait, &wq, TASK_INTERRUPTIBLE);
  6021. if (!kthread_should_stop() &&
  6022. mddev2->curr_resync >= mddev->curr_resync) {
  6023. printk(KERN_INFO "md: delaying %s of %s"
  6024. " until %s has finished (they"
  6025. " share one or more physical units)\n",
  6026. desc, mdname(mddev), mdname(mddev2));
  6027. mddev_put(mddev2);
  6028. if (signal_pending(current))
  6029. flush_signals(current);
  6030. schedule();
  6031. finish_wait(&resync_wait, &wq);
  6032. goto try_again;
  6033. }
  6034. finish_wait(&resync_wait, &wq);
  6035. }
  6036. }
  6037. } while (mddev->curr_resync < 2);
  6038. j = 0;
  6039. if (test_bit(MD_RECOVERY_SYNC, &mddev->recovery)) {
  6040. /* resync follows the size requested by the personality,
  6041. * which defaults to physical size, but can be virtual size
  6042. */
  6043. max_sectors = mddev->resync_max_sectors;
  6044. mddev->resync_mismatches = 0;
  6045. /* we don't use the checkpoint if there's a bitmap */
  6046. if (test_bit(MD_RECOVERY_REQUESTED, &mddev->recovery))
  6047. j = mddev->resync_min;
  6048. else if (!mddev->bitmap)
  6049. j = mddev->recovery_cp;
  6050. } else if (test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery))
  6051. max_sectors = mddev->dev_sectors;
  6052. else {
  6053. /* recovery follows the physical size of devices */
  6054. max_sectors = mddev->dev_sectors;
  6055. j = MaxSector;
  6056. rcu_read_lock();
  6057. list_for_each_entry_rcu(rdev, &mddev->disks, same_set)
  6058. if (rdev->raid_disk >= 0 &&
  6059. !test_bit(Faulty, &rdev->flags) &&
  6060. !test_bit(In_sync, &rdev->flags) &&
  6061. rdev->recovery_offset < j)
  6062. j = rdev->recovery_offset;
  6063. rcu_read_unlock();
  6064. }
  6065. printk(KERN_INFO "md: %s of RAID array %s\n", desc, mdname(mddev));
  6066. printk(KERN_INFO "md: minimum _guaranteed_ speed:"
  6067. " %d KB/sec/disk.\n", speed_min(mddev));
  6068. printk(KERN_INFO "md: using maximum available idle IO bandwidth "
  6069. "(but not more than %d KB/sec) for %s.\n",
  6070. speed_max(mddev), desc);
  6071. is_mddev_idle(mddev, 1); /* this initializes IO event counters */
  6072. io_sectors = 0;
  6073. for (m = 0; m < SYNC_MARKS; m++) {
  6074. mark[m] = jiffies;
  6075. mark_cnt[m] = io_sectors;
  6076. }
  6077. last_mark = 0;
  6078. mddev->resync_mark = mark[last_mark];
  6079. mddev->resync_mark_cnt = mark_cnt[last_mark];
  6080. /*
  6081. * Tune reconstruction:
  6082. */
  6083. window = 32*(PAGE_SIZE/512);
  6084. printk(KERN_INFO "md: using %dk window, over a total of %lluk.\n",
  6085. window/2, (unsigned long long)max_sectors/2);
  6086. atomic_set(&mddev->recovery_active, 0);
  6087. last_check = 0;
  6088. if (j>2) {
  6089. printk(KERN_INFO
  6090. "md: resuming %s of %s from checkpoint.\n",
  6091. desc, mdname(mddev));
  6092. mddev->curr_resync = j;
  6093. }
  6094. mddev->curr_resync_completed = j;
  6095. while (j < max_sectors) {
  6096. sector_t sectors;
  6097. skipped = 0;
  6098. if (!test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery) &&
  6099. ((mddev->curr_resync > mddev->curr_resync_completed &&
  6100. (mddev->curr_resync - mddev->curr_resync_completed)
  6101. > (max_sectors >> 4)) ||
  6102. (j - mddev->curr_resync_completed)*2
  6103. >= mddev->resync_max - mddev->curr_resync_completed
  6104. )) {
  6105. /* time to update curr_resync_completed */
  6106. wait_event(mddev->recovery_wait,
  6107. atomic_read(&mddev->recovery_active) == 0);
  6108. mddev->curr_resync_completed = j;
  6109. set_bit(MD_CHANGE_CLEAN, &mddev->flags);
  6110. sysfs_notify(&mddev->kobj, NULL, "sync_completed");
  6111. }
  6112. while (j >= mddev->resync_max && !kthread_should_stop()) {
  6113. /* As this condition is controlled by user-space,
  6114. * we can block indefinitely, so use '_interruptible'
  6115. * to avoid triggering warnings.
  6116. */
  6117. flush_signals(current); /* just in case */
  6118. wait_event_interruptible(mddev->recovery_wait,
  6119. mddev->resync_max > j
  6120. || kthread_should_stop());
  6121. }
  6122. if (kthread_should_stop())
  6123. goto interrupted;
  6124. sectors = mddev->pers->sync_request(mddev, j, &skipped,
  6125. currspeed < speed_min(mddev));
  6126. if (sectors == 0) {
  6127. set_bit(MD_RECOVERY_INTR, &mddev->recovery);
  6128. goto out;
  6129. }
  6130. if (!skipped) { /* actual IO requested */
  6131. io_sectors += sectors;
  6132. atomic_add(sectors, &mddev->recovery_active);
  6133. }
  6134. j += sectors;
  6135. if (j>1) mddev->curr_resync = j;
  6136. mddev->curr_mark_cnt = io_sectors;
  6137. if (last_check == 0)
  6138. /* this is the earliers that rebuilt will be
  6139. * visible in /proc/mdstat
  6140. */
  6141. md_new_event(mddev);
  6142. if (last_check + window > io_sectors || j == max_sectors)
  6143. continue;
  6144. last_check = io_sectors;
  6145. if (test_bit(MD_RECOVERY_INTR, &mddev->recovery))
  6146. break;
  6147. repeat:
  6148. if (time_after_eq(jiffies, mark[last_mark] + SYNC_MARK_STEP )) {
  6149. /* step marks */
  6150. int next = (last_mark+1) % SYNC_MARKS;
  6151. mddev->resync_mark = mark[next];
  6152. mddev->resync_mark_cnt = mark_cnt[next];
  6153. mark[next] = jiffies;
  6154. mark_cnt[next] = io_sectors - atomic_read(&mddev->recovery_active);
  6155. last_mark = next;
  6156. }
  6157. if (kthread_should_stop())
  6158. goto interrupted;
  6159. /*
  6160. * this loop exits only if either when we are slower than
  6161. * the 'hard' speed limit, or the system was IO-idle for
  6162. * a jiffy.
  6163. * the system might be non-idle CPU-wise, but we only care
  6164. * about not overloading the IO subsystem. (things like an
  6165. * e2fsck being done on the RAID array should execute fast)
  6166. */
  6167. cond_resched();
  6168. currspeed = ((unsigned long)(io_sectors-mddev->resync_mark_cnt))/2
  6169. /((jiffies-mddev->resync_mark)/HZ +1) +1;
  6170. if (currspeed > speed_min(mddev)) {
  6171. if ((currspeed > speed_max(mddev)) ||
  6172. !is_mddev_idle(mddev, 0)) {
  6173. msleep(500);
  6174. goto repeat;
  6175. }
  6176. }
  6177. }
  6178. printk(KERN_INFO "md: %s: %s done.\n",mdname(mddev), desc);
  6179. /*
  6180. * this also signals 'finished resyncing' to md_stop
  6181. */
  6182. out:
  6183. wait_event(mddev->recovery_wait, !atomic_read(&mddev->recovery_active));
  6184. /* tell personality that we are finished */
  6185. mddev->pers->sync_request(mddev, max_sectors, &skipped, 1);
  6186. if (!test_bit(MD_RECOVERY_CHECK, &mddev->recovery) &&
  6187. mddev->curr_resync > 2) {
  6188. if (test_bit(MD_RECOVERY_SYNC, &mddev->recovery)) {
  6189. if (test_bit(MD_RECOVERY_INTR, &mddev->recovery)) {
  6190. if (mddev->curr_resync >= mddev->recovery_cp) {
  6191. printk(KERN_INFO
  6192. "md: checkpointing %s of %s.\n",
  6193. desc, mdname(mddev));
  6194. mddev->recovery_cp = mddev->curr_resync;
  6195. }
  6196. } else
  6197. mddev->recovery_cp = MaxSector;
  6198. } else {
  6199. if (!test_bit(MD_RECOVERY_INTR, &mddev->recovery))
  6200. mddev->curr_resync = MaxSector;
  6201. rcu_read_lock();
  6202. list_for_each_entry_rcu(rdev, &mddev->disks, same_set)
  6203. if (rdev->raid_disk >= 0 &&
  6204. mddev->delta_disks >= 0 &&
  6205. !test_bit(Faulty, &rdev->flags) &&
  6206. !test_bit(In_sync, &rdev->flags) &&
  6207. rdev->recovery_offset < mddev->curr_resync)
  6208. rdev->recovery_offset = mddev->curr_resync;
  6209. rcu_read_unlock();
  6210. }
  6211. }
  6212. set_bit(MD_CHANGE_DEVS, &mddev->flags);
  6213. skip:
  6214. if (!test_bit(MD_RECOVERY_INTR, &mddev->recovery)) {
  6215. /* We completed so min/max setting can be forgotten if used. */
  6216. if (test_bit(MD_RECOVERY_REQUESTED, &mddev->recovery))
  6217. mddev->resync_min = 0;
  6218. mddev->resync_max = MaxSector;
  6219. } else if (test_bit(MD_RECOVERY_REQUESTED, &mddev->recovery))
  6220. mddev->resync_min = mddev->curr_resync_completed;
  6221. mddev->curr_resync = 0;
  6222. wake_up(&resync_wait);
  6223. set_bit(MD_RECOVERY_DONE, &mddev->recovery);
  6224. md_wakeup_thread(mddev->thread);
  6225. return;
  6226. interrupted:
  6227. /*
  6228. * got a signal, exit.
  6229. */
  6230. printk(KERN_INFO
  6231. "md: md_do_sync() got signal ... exiting\n");
  6232. set_bit(MD_RECOVERY_INTR, &mddev->recovery);
  6233. goto out;
  6234. }
  6235. EXPORT_SYMBOL_GPL(md_do_sync);
  6236. static int remove_and_add_spares(mddev_t *mddev)
  6237. {
  6238. mdk_rdev_t *rdev;
  6239. int spares = 0;
  6240. mddev->curr_resync_completed = 0;
  6241. list_for_each_entry(rdev, &mddev->disks, same_set)
  6242. if (rdev->raid_disk >= 0 &&
  6243. !test_bit(Blocked, &rdev->flags) &&
  6244. (test_bit(Faulty, &rdev->flags) ||
  6245. ! test_bit(In_sync, &rdev->flags)) &&
  6246. atomic_read(&rdev->nr_pending)==0) {
  6247. if (mddev->pers->hot_remove_disk(
  6248. mddev, rdev->raid_disk)==0) {
  6249. char nm[20];
  6250. sprintf(nm,"rd%d", rdev->raid_disk);
  6251. sysfs_remove_link(&mddev->kobj, nm);
  6252. rdev->raid_disk = -1;
  6253. }
  6254. }
  6255. if (mddev->degraded && !mddev->recovery_disabled) {
  6256. list_for_each_entry(rdev, &mddev->disks, same_set) {
  6257. if (rdev->raid_disk >= 0 &&
  6258. !test_bit(In_sync, &rdev->flags) &&
  6259. !test_bit(Faulty, &rdev->flags) &&
  6260. !test_bit(Blocked, &rdev->flags))
  6261. spares++;
  6262. if (rdev->raid_disk < 0
  6263. && !test_bit(Faulty, &rdev->flags)) {
  6264. rdev->recovery_offset = 0;
  6265. if (mddev->pers->
  6266. hot_add_disk(mddev, rdev) == 0) {
  6267. char nm[20];
  6268. sprintf(nm, "rd%d", rdev->raid_disk);
  6269. if (sysfs_create_link(&mddev->kobj,
  6270. &rdev->kobj, nm))
  6271. /* failure here is OK */;
  6272. spares++;
  6273. md_new_event(mddev);
  6274. set_bit(MD_CHANGE_DEVS, &mddev->flags);
  6275. } else
  6276. break;
  6277. }
  6278. }
  6279. }
  6280. return spares;
  6281. }
  6282. static void reap_sync_thread(mddev_t *mddev)
  6283. {
  6284. mdk_rdev_t *rdev;
  6285. /* resync has finished, collect result */
  6286. md_unregister_thread(mddev->sync_thread);
  6287. mddev->sync_thread = NULL;
  6288. if (!test_bit(MD_RECOVERY_INTR, &mddev->recovery) &&
  6289. !test_bit(MD_RECOVERY_REQUESTED, &mddev->recovery)) {
  6290. /* success...*/
  6291. /* activate any spares */
  6292. if (mddev->pers->spare_active(mddev))
  6293. sysfs_notify(&mddev->kobj, NULL,
  6294. "degraded");
  6295. }
  6296. if (test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery) &&
  6297. mddev->pers->finish_reshape)
  6298. mddev->pers->finish_reshape(mddev);
  6299. md_update_sb(mddev, 1);
  6300. /* if array is no-longer degraded, then any saved_raid_disk
  6301. * information must be scrapped
  6302. */
  6303. if (!mddev->degraded)
  6304. list_for_each_entry(rdev, &mddev->disks, same_set)
  6305. rdev->saved_raid_disk = -1;
  6306. clear_bit(MD_RECOVERY_RUNNING, &mddev->recovery);
  6307. clear_bit(MD_RECOVERY_SYNC, &mddev->recovery);
  6308. clear_bit(MD_RECOVERY_RESHAPE, &mddev->recovery);
  6309. clear_bit(MD_RECOVERY_REQUESTED, &mddev->recovery);
  6310. clear_bit(MD_RECOVERY_CHECK, &mddev->recovery);
  6311. /* flag recovery needed just to double check */
  6312. set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
  6313. sysfs_notify_dirent_safe(mddev->sysfs_action);
  6314. md_new_event(mddev);
  6315. }
  6316. /*
  6317. * This routine is regularly called by all per-raid-array threads to
  6318. * deal with generic issues like resync and super-block update.
  6319. * Raid personalities that don't have a thread (linear/raid0) do not
  6320. * need this as they never do any recovery or update the superblock.
  6321. *
  6322. * It does not do any resync itself, but rather "forks" off other threads
  6323. * to do that as needed.
  6324. * When it is determined that resync is needed, we set MD_RECOVERY_RUNNING in
  6325. * "->recovery" and create a thread at ->sync_thread.
  6326. * When the thread finishes it sets MD_RECOVERY_DONE
  6327. * and wakeups up this thread which will reap the thread and finish up.
  6328. * This thread also removes any faulty devices (with nr_pending == 0).
  6329. *
  6330. * The overall approach is:
  6331. * 1/ if the superblock needs updating, update it.
  6332. * 2/ If a recovery thread is running, don't do anything else.
  6333. * 3/ If recovery has finished, clean up, possibly marking spares active.
  6334. * 4/ If there are any faulty devices, remove them.
  6335. * 5/ If array is degraded, try to add spares devices
  6336. * 6/ If array has spares or is not in-sync, start a resync thread.
  6337. */
  6338. void md_check_recovery(mddev_t *mddev)
  6339. {
  6340. if (mddev->suspended)
  6341. return;
  6342. if (mddev->bitmap)
  6343. bitmap_daemon_work(mddev);
  6344. if (mddev->ro)
  6345. return;
  6346. if (signal_pending(current)) {
  6347. if (mddev->pers->sync_request && !mddev->external) {
  6348. printk(KERN_INFO "md: %s in immediate safe mode\n",
  6349. mdname(mddev));
  6350. mddev->safemode = 2;
  6351. }
  6352. flush_signals(current);
  6353. }
  6354. if (mddev->ro && !test_bit(MD_RECOVERY_NEEDED, &mddev->recovery))
  6355. return;
  6356. if ( ! (
  6357. (mddev->flags & ~ (1<<MD_CHANGE_PENDING)) ||
  6358. test_bit(MD_RECOVERY_NEEDED, &mddev->recovery) ||
  6359. test_bit(MD_RECOVERY_DONE, &mddev->recovery) ||
  6360. (mddev->external == 0 && mddev->safemode == 1) ||
  6361. (mddev->safemode == 2 && ! atomic_read(&mddev->writes_pending)
  6362. && !mddev->in_sync && mddev->recovery_cp == MaxSector)
  6363. ))
  6364. return;
  6365. if (mddev_trylock(mddev)) {
  6366. int spares = 0;
  6367. if (mddev->ro) {
  6368. /* Only thing we do on a ro array is remove
  6369. * failed devices.
  6370. */
  6371. mdk_rdev_t *rdev;
  6372. list_for_each_entry(rdev, &mddev->disks, same_set)
  6373. if (rdev->raid_disk >= 0 &&
  6374. !test_bit(Blocked, &rdev->flags) &&
  6375. test_bit(Faulty, &rdev->flags) &&
  6376. atomic_read(&rdev->nr_pending)==0) {
  6377. if (mddev->pers->hot_remove_disk(
  6378. mddev, rdev->raid_disk)==0) {
  6379. char nm[20];
  6380. sprintf(nm,"rd%d", rdev->raid_disk);
  6381. sysfs_remove_link(&mddev->kobj, nm);
  6382. rdev->raid_disk = -1;
  6383. }
  6384. }
  6385. clear_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
  6386. goto unlock;
  6387. }
  6388. if (!mddev->external) {
  6389. int did_change = 0;
  6390. spin_lock_irq(&mddev->write_lock);
  6391. if (mddev->safemode &&
  6392. !atomic_read(&mddev->writes_pending) &&
  6393. !mddev->in_sync &&
  6394. mddev->recovery_cp == MaxSector) {
  6395. mddev->in_sync = 1;
  6396. did_change = 1;
  6397. set_bit(MD_CHANGE_CLEAN, &mddev->flags);
  6398. }
  6399. if (mddev->safemode == 1)
  6400. mddev->safemode = 0;
  6401. spin_unlock_irq(&mddev->write_lock);
  6402. if (did_change)
  6403. sysfs_notify_dirent_safe(mddev->sysfs_state);
  6404. }
  6405. if (mddev->flags)
  6406. md_update_sb(mddev, 0);
  6407. if (test_bit(MD_RECOVERY_RUNNING, &mddev->recovery) &&
  6408. !test_bit(MD_RECOVERY_DONE, &mddev->recovery)) {
  6409. /* resync/recovery still happening */
  6410. clear_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
  6411. goto unlock;
  6412. }
  6413. if (mddev->sync_thread) {
  6414. reap_sync_thread(mddev);
  6415. goto unlock;
  6416. }
  6417. /* Set RUNNING before clearing NEEDED to avoid
  6418. * any transients in the value of "sync_action".
  6419. */
  6420. set_bit(MD_RECOVERY_RUNNING, &mddev->recovery);
  6421. clear_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
  6422. /* Clear some bits that don't mean anything, but
  6423. * might be left set
  6424. */
  6425. clear_bit(MD_RECOVERY_INTR, &mddev->recovery);
  6426. clear_bit(MD_RECOVERY_DONE, &mddev->recovery);
  6427. if (test_bit(MD_RECOVERY_FROZEN, &mddev->recovery))
  6428. goto unlock;
  6429. /* no recovery is running.
  6430. * remove any failed drives, then
  6431. * add spares if possible.
  6432. * Spare are also removed and re-added, to allow
  6433. * the personality to fail the re-add.
  6434. */
  6435. if (mddev->reshape_position != MaxSector) {
  6436. if (mddev->pers->check_reshape == NULL ||
  6437. mddev->pers->check_reshape(mddev) != 0)
  6438. /* Cannot proceed */
  6439. goto unlock;
  6440. set_bit(MD_RECOVERY_RESHAPE, &mddev->recovery);
  6441. clear_bit(MD_RECOVERY_RECOVER, &mddev->recovery);
  6442. } else if ((spares = remove_and_add_spares(mddev))) {
  6443. clear_bit(MD_RECOVERY_SYNC, &mddev->recovery);
  6444. clear_bit(MD_RECOVERY_CHECK, &mddev->recovery);
  6445. clear_bit(MD_RECOVERY_REQUESTED, &mddev->recovery);
  6446. set_bit(MD_RECOVERY_RECOVER, &mddev->recovery);
  6447. } else if (mddev->recovery_cp < MaxSector) {
  6448. set_bit(MD_RECOVERY_SYNC, &mddev->recovery);
  6449. clear_bit(MD_RECOVERY_RECOVER, &mddev->recovery);
  6450. } else if (!test_bit(MD_RECOVERY_SYNC, &mddev->recovery))
  6451. /* nothing to be done ... */
  6452. goto unlock;
  6453. if (mddev->pers->sync_request) {
  6454. if (spares && mddev->bitmap && ! mddev->bitmap->file) {
  6455. /* We are adding a device or devices to an array
  6456. * which has the bitmap stored on all devices.
  6457. * So make sure all bitmap pages get written
  6458. */
  6459. bitmap_write_all(mddev->bitmap);
  6460. }
  6461. mddev->sync_thread = md_register_thread(md_do_sync,
  6462. mddev,
  6463. "resync");
  6464. if (!mddev->sync_thread) {
  6465. printk(KERN_ERR "%s: could not start resync"
  6466. " thread...\n",
  6467. mdname(mddev));
  6468. /* leave the spares where they are, it shouldn't hurt */
  6469. clear_bit(MD_RECOVERY_RUNNING, &mddev->recovery);
  6470. clear_bit(MD_RECOVERY_SYNC, &mddev->recovery);
  6471. clear_bit(MD_RECOVERY_RESHAPE, &mddev->recovery);
  6472. clear_bit(MD_RECOVERY_REQUESTED, &mddev->recovery);
  6473. clear_bit(MD_RECOVERY_CHECK, &mddev->recovery);
  6474. } else
  6475. md_wakeup_thread(mddev->sync_thread);
  6476. sysfs_notify_dirent_safe(mddev->sysfs_action);
  6477. md_new_event(mddev);
  6478. }
  6479. unlock:
  6480. if (!mddev->sync_thread) {
  6481. clear_bit(MD_RECOVERY_RUNNING, &mddev->recovery);
  6482. if (test_and_clear_bit(MD_RECOVERY_RECOVER,
  6483. &mddev->recovery))
  6484. if (mddev->sysfs_action)
  6485. sysfs_notify_dirent_safe(mddev->sysfs_action);
  6486. }
  6487. mddev_unlock(mddev);
  6488. }
  6489. }
  6490. void md_wait_for_blocked_rdev(mdk_rdev_t *rdev, mddev_t *mddev)
  6491. {
  6492. sysfs_notify_dirent_safe(rdev->sysfs_state);
  6493. wait_event_timeout(rdev->blocked_wait,
  6494. !test_bit(Blocked, &rdev->flags),
  6495. msecs_to_jiffies(5000));
  6496. rdev_dec_pending(rdev, mddev);
  6497. }
  6498. EXPORT_SYMBOL(md_wait_for_blocked_rdev);
  6499. static int md_notify_reboot(struct notifier_block *this,
  6500. unsigned long code, void *x)
  6501. {
  6502. struct list_head *tmp;
  6503. mddev_t *mddev;
  6504. if ((code == SYS_DOWN) || (code == SYS_HALT) || (code == SYS_POWER_OFF)) {
  6505. printk(KERN_INFO "md: stopping all md devices.\n");
  6506. for_each_mddev(mddev, tmp)
  6507. if (mddev_trylock(mddev)) {
  6508. /* Force a switch to readonly even array
  6509. * appears to still be in use. Hence
  6510. * the '100'.
  6511. */
  6512. md_set_readonly(mddev, 100);
  6513. mddev_unlock(mddev);
  6514. }
  6515. /*
  6516. * certain more exotic SCSI devices are known to be
  6517. * volatile wrt too early system reboots. While the
  6518. * right place to handle this issue is the given
  6519. * driver, we do want to have a safe RAID driver ...
  6520. */
  6521. mdelay(1000*1);
  6522. }
  6523. return NOTIFY_DONE;
  6524. }
  6525. static struct notifier_block md_notifier = {
  6526. .notifier_call = md_notify_reboot,
  6527. .next = NULL,
  6528. .priority = INT_MAX, /* before any real devices */
  6529. };
  6530. static void md_geninit(void)
  6531. {
  6532. dprintk("md: sizeof(mdp_super_t) = %d\n", (int)sizeof(mdp_super_t));
  6533. proc_create("mdstat", S_IRUGO, NULL, &md_seq_fops);
  6534. }
  6535. static int __init md_init(void)
  6536. {
  6537. int ret = -ENOMEM;
  6538. md_wq = alloc_workqueue("md", WQ_MEM_RECLAIM, 0);
  6539. if (!md_wq)
  6540. goto err_wq;
  6541. md_misc_wq = alloc_workqueue("md_misc", 0, 0);
  6542. if (!md_misc_wq)
  6543. goto err_misc_wq;
  6544. if ((ret = register_blkdev(MD_MAJOR, "md")) < 0)
  6545. goto err_md;
  6546. if ((ret = register_blkdev(0, "mdp")) < 0)
  6547. goto err_mdp;
  6548. mdp_major = ret;
  6549. blk_register_region(MKDEV(MD_MAJOR, 0), 1UL<<MINORBITS, THIS_MODULE,
  6550. md_probe, NULL, NULL);
  6551. blk_register_region(MKDEV(mdp_major, 0), 1UL<<MINORBITS, THIS_MODULE,
  6552. md_probe, NULL, NULL);
  6553. register_reboot_notifier(&md_notifier);
  6554. raid_table_header = register_sysctl_table(raid_root_table);
  6555. md_geninit();
  6556. return 0;
  6557. err_mdp:
  6558. unregister_blkdev(MD_MAJOR, "md");
  6559. err_md:
  6560. destroy_workqueue(md_misc_wq);
  6561. err_misc_wq:
  6562. destroy_workqueue(md_wq);
  6563. err_wq:
  6564. return ret;
  6565. }
  6566. #ifndef MODULE
  6567. /*
  6568. * Searches all registered partitions for autorun RAID arrays
  6569. * at boot time.
  6570. */
  6571. static LIST_HEAD(all_detected_devices);
  6572. struct detected_devices_node {
  6573. struct list_head list;
  6574. dev_t dev;
  6575. };
  6576. void md_autodetect_dev(dev_t dev)
  6577. {
  6578. struct detected_devices_node *node_detected_dev;
  6579. node_detected_dev = kzalloc(sizeof(*node_detected_dev), GFP_KERNEL);
  6580. if (node_detected_dev) {
  6581. node_detected_dev->dev = dev;
  6582. list_add_tail(&node_detected_dev->list, &all_detected_devices);
  6583. } else {
  6584. printk(KERN_CRIT "md: md_autodetect_dev: kzalloc failed"
  6585. ", skipping dev(%d,%d)\n", MAJOR(dev), MINOR(dev));
  6586. }
  6587. }
  6588. static void autostart_arrays(int part)
  6589. {
  6590. mdk_rdev_t *rdev;
  6591. struct detected_devices_node *node_detected_dev;
  6592. dev_t dev;
  6593. int i_scanned, i_passed;
  6594. i_scanned = 0;
  6595. i_passed = 0;
  6596. printk(KERN_INFO "md: Autodetecting RAID arrays.\n");
  6597. while (!list_empty(&all_detected_devices) && i_scanned < INT_MAX) {
  6598. i_scanned++;
  6599. node_detected_dev = list_entry(all_detected_devices.next,
  6600. struct detected_devices_node, list);
  6601. list_del(&node_detected_dev->list);
  6602. dev = node_detected_dev->dev;
  6603. kfree(node_detected_dev);
  6604. rdev = md_import_device(dev,0, 90);
  6605. if (IS_ERR(rdev))
  6606. continue;
  6607. if (test_bit(Faulty, &rdev->flags)) {
  6608. MD_BUG();
  6609. continue;
  6610. }
  6611. set_bit(AutoDetected, &rdev->flags);
  6612. list_add(&rdev->same_set, &pending_raid_disks);
  6613. i_passed++;
  6614. }
  6615. printk(KERN_INFO "md: Scanned %d and added %d devices.\n",
  6616. i_scanned, i_passed);
  6617. autorun_devices(part);
  6618. }
  6619. #endif /* !MODULE */
  6620. static __exit void md_exit(void)
  6621. {
  6622. mddev_t *mddev;
  6623. struct list_head *tmp;
  6624. blk_unregister_region(MKDEV(MD_MAJOR,0), 1U << MINORBITS);
  6625. blk_unregister_region(MKDEV(mdp_major,0), 1U << MINORBITS);
  6626. unregister_blkdev(MD_MAJOR,"md");
  6627. unregister_blkdev(mdp_major, "mdp");
  6628. unregister_reboot_notifier(&md_notifier);
  6629. unregister_sysctl_table(raid_table_header);
  6630. remove_proc_entry("mdstat", NULL);
  6631. for_each_mddev(mddev, tmp) {
  6632. export_array(mddev);
  6633. mddev->hold_active = 0;
  6634. }
  6635. destroy_workqueue(md_misc_wq);
  6636. destroy_workqueue(md_wq);
  6637. }
  6638. subsys_initcall(md_init);
  6639. module_exit(md_exit)
  6640. static int get_ro(char *buffer, struct kernel_param *kp)
  6641. {
  6642. return sprintf(buffer, "%d", start_readonly);
  6643. }
  6644. static int set_ro(const char *val, struct kernel_param *kp)
  6645. {
  6646. char *e;
  6647. int num = simple_strtoul(val, &e, 10);
  6648. if (*val && (*e == '\0' || *e == '\n')) {
  6649. start_readonly = num;
  6650. return 0;
  6651. }
  6652. return -EINVAL;
  6653. }
  6654. module_param_call(start_ro, set_ro, get_ro, NULL, S_IRUSR|S_IWUSR);
  6655. module_param(start_dirty_degraded, int, S_IRUGO|S_IWUSR);
  6656. module_param_call(new_array, add_named_array, NULL, NULL, S_IWUSR);
  6657. EXPORT_SYMBOL(register_md_personality);
  6658. EXPORT_SYMBOL(unregister_md_personality);
  6659. EXPORT_SYMBOL(md_error);
  6660. EXPORT_SYMBOL(md_done_sync);
  6661. EXPORT_SYMBOL(md_write_start);
  6662. EXPORT_SYMBOL(md_write_end);
  6663. EXPORT_SYMBOL(md_register_thread);
  6664. EXPORT_SYMBOL(md_unregister_thread);
  6665. EXPORT_SYMBOL(md_wakeup_thread);
  6666. EXPORT_SYMBOL(md_check_recovery);
  6667. MODULE_LICENSE("GPL");
  6668. MODULE_DESCRIPTION("MD RAID framework");
  6669. MODULE_ALIAS("md");
  6670. MODULE_ALIAS_BLOCKDEV_MAJOR(MD_MAJOR);