bond_main.c 132 KB

1234567891011121314151617181920212223242526272829303132333435363738394041424344454647484950515253545556575859606162636465666768697071727374757677787980818283848586878889909192939495969798991001011021031041051061071081091101111121131141151161171181191201211221231241251261271281291301311321331341351361371381391401411421431441451461471481491501511521531541551561571581591601611621631641651661671681691701711721731741751761771781791801811821831841851861871881891901911921931941951961971981992002012022032042052062072082092102112122132142152162172182192202212222232242252262272282292302312322332342352362372382392402412422432442452462472482492502512522532542552562572582592602612622632642652662672682692702712722732742752762772782792802812822832842852862872882892902912922932942952962972982993003013023033043053063073083093103113123133143153163173183193203213223233243253263273283293303313323333343353363373383393403413423433443453463473483493503513523533543553563573583593603613623633643653663673683693703713723733743753763773783793803813823833843853863873883893903913923933943953963973983994004014024034044054064074084094104114124134144154164174184194204214224234244254264274284294304314324334344354364374384394404414424434444454464474484494504514524534544554564574584594604614624634644654664674684694704714724734744754764774784794804814824834844854864874884894904914924934944954964974984995005015025035045055065075085095105115125135145155165175185195205215225235245255265275285295305315325335345355365375385395405415425435445455465475485495505515525535545555565575585595605615625635645655665675685695705715725735745755765775785795805815825835845855865875885895905915925935945955965975985996006016026036046056066076086096106116126136146156166176186196206216226236246256266276286296306316326336346356366376386396406416426436446456466476486496506516526536546556566576586596606616626636646656666676686696706716726736746756766776786796806816826836846856866876886896906916926936946956966976986997007017027037047057067077087097107117127137147157167177187197207217227237247257267277287297307317327337347357367377387397407417427437447457467477487497507517527537547557567577587597607617627637647657667677687697707717727737747757767777787797807817827837847857867877887897907917927937947957967977987998008018028038048058068078088098108118128138148158168178188198208218228238248258268278288298308318328338348358368378388398408418428438448458468478488498508518528538548558568578588598608618628638648658668678688698708718728738748758768778788798808818828838848858868878888898908918928938948958968978988999009019029039049059069079089099109119129139149159169179189199209219229239249259269279289299309319329339349359369379389399409419429439449459469479489499509519529539549559569579589599609619629639649659669679689699709719729739749759769779789799809819829839849859869879889899909919929939949959969979989991000100110021003100410051006100710081009101010111012101310141015101610171018101910201021102210231024102510261027102810291030103110321033103410351036103710381039104010411042104310441045104610471048104910501051105210531054105510561057105810591060106110621063106410651066106710681069107010711072107310741075107610771078107910801081108210831084108510861087108810891090109110921093109410951096109710981099110011011102110311041105110611071108110911101111111211131114111511161117111811191120112111221123112411251126112711281129113011311132113311341135113611371138113911401141114211431144114511461147114811491150115111521153115411551156115711581159116011611162116311641165116611671168116911701171117211731174117511761177117811791180118111821183118411851186118711881189119011911192119311941195119611971198119912001201120212031204120512061207120812091210121112121213121412151216121712181219122012211222122312241225122612271228122912301231123212331234123512361237123812391240124112421243124412451246124712481249125012511252125312541255125612571258125912601261126212631264126512661267126812691270127112721273127412751276127712781279128012811282128312841285128612871288128912901291129212931294129512961297129812991300130113021303130413051306130713081309131013111312131313141315131613171318131913201321132213231324132513261327132813291330133113321333133413351336133713381339134013411342134313441345134613471348134913501351135213531354135513561357135813591360136113621363136413651366136713681369137013711372137313741375137613771378137913801381138213831384138513861387138813891390139113921393139413951396139713981399140014011402140314041405140614071408140914101411141214131414141514161417141814191420142114221423142414251426142714281429143014311432143314341435143614371438143914401441144214431444144514461447144814491450145114521453145414551456145714581459146014611462146314641465146614671468146914701471147214731474147514761477147814791480148114821483148414851486148714881489149014911492149314941495149614971498149915001501150215031504150515061507150815091510151115121513151415151516151715181519152015211522152315241525152615271528152915301531153215331534153515361537153815391540154115421543154415451546154715481549155015511552155315541555155615571558155915601561156215631564156515661567156815691570157115721573157415751576157715781579158015811582158315841585158615871588158915901591159215931594159515961597159815991600160116021603160416051606160716081609161016111612161316141615161616171618161916201621162216231624162516261627162816291630163116321633163416351636163716381639164016411642164316441645164616471648164916501651165216531654165516561657165816591660166116621663166416651666166716681669167016711672167316741675167616771678167916801681168216831684168516861687168816891690169116921693169416951696169716981699170017011702170317041705170617071708170917101711171217131714171517161717171817191720172117221723172417251726172717281729173017311732173317341735173617371738173917401741174217431744174517461747174817491750175117521753175417551756175717581759176017611762176317641765176617671768176917701771177217731774177517761777177817791780178117821783178417851786178717881789179017911792179317941795179617971798179918001801180218031804180518061807180818091810181118121813181418151816181718181819182018211822182318241825182618271828182918301831183218331834183518361837183818391840184118421843184418451846184718481849185018511852185318541855185618571858185918601861186218631864186518661867186818691870187118721873187418751876187718781879188018811882188318841885188618871888188918901891189218931894189518961897189818991900190119021903190419051906190719081909191019111912191319141915191619171918191919201921192219231924192519261927192819291930193119321933193419351936193719381939194019411942194319441945194619471948194919501951195219531954195519561957195819591960196119621963196419651966196719681969197019711972197319741975197619771978197919801981198219831984198519861987198819891990199119921993199419951996199719981999200020012002200320042005200620072008200920102011201220132014201520162017201820192020202120222023202420252026202720282029203020312032203320342035203620372038203920402041204220432044204520462047204820492050205120522053205420552056205720582059206020612062206320642065206620672068206920702071207220732074207520762077207820792080208120822083208420852086208720882089209020912092209320942095209620972098209921002101210221032104210521062107210821092110211121122113211421152116211721182119212021212122212321242125212621272128212921302131213221332134213521362137213821392140214121422143214421452146214721482149215021512152215321542155215621572158215921602161216221632164216521662167216821692170217121722173217421752176217721782179218021812182218321842185218621872188218921902191219221932194219521962197219821992200220122022203220422052206220722082209221022112212221322142215221622172218221922202221222222232224222522262227222822292230223122322233223422352236223722382239224022412242224322442245224622472248224922502251225222532254225522562257225822592260226122622263226422652266226722682269227022712272227322742275227622772278227922802281228222832284228522862287228822892290229122922293229422952296229722982299230023012302230323042305230623072308230923102311231223132314231523162317231823192320232123222323232423252326232723282329233023312332233323342335233623372338233923402341234223432344234523462347234823492350235123522353235423552356235723582359236023612362236323642365236623672368236923702371237223732374237523762377237823792380238123822383238423852386238723882389239023912392239323942395239623972398239924002401240224032404240524062407240824092410241124122413241424152416241724182419242024212422242324242425242624272428242924302431243224332434243524362437243824392440244124422443244424452446244724482449245024512452245324542455245624572458245924602461246224632464246524662467246824692470247124722473247424752476247724782479248024812482248324842485248624872488248924902491249224932494249524962497249824992500250125022503250425052506250725082509251025112512251325142515251625172518251925202521252225232524252525262527252825292530253125322533253425352536253725382539254025412542254325442545254625472548254925502551255225532554255525562557255825592560256125622563256425652566256725682569257025712572257325742575257625772578257925802581258225832584258525862587258825892590259125922593259425952596259725982599260026012602260326042605260626072608260926102611261226132614261526162617261826192620262126222623262426252626262726282629263026312632263326342635263626372638263926402641264226432644264526462647264826492650265126522653265426552656265726582659266026612662266326642665266626672668266926702671267226732674267526762677267826792680268126822683268426852686268726882689269026912692269326942695269626972698269927002701270227032704270527062707270827092710271127122713271427152716271727182719272027212722272327242725272627272728272927302731273227332734273527362737273827392740274127422743274427452746274727482749275027512752275327542755275627572758275927602761276227632764276527662767276827692770277127722773277427752776277727782779278027812782278327842785278627872788278927902791279227932794279527962797279827992800280128022803280428052806280728082809281028112812281328142815281628172818281928202821282228232824282528262827282828292830283128322833283428352836283728382839284028412842284328442845284628472848284928502851285228532854285528562857285828592860286128622863286428652866286728682869287028712872287328742875287628772878287928802881288228832884288528862887288828892890289128922893289428952896289728982899290029012902290329042905290629072908290929102911291229132914291529162917291829192920292129222923292429252926292729282929293029312932293329342935293629372938293929402941294229432944294529462947294829492950295129522953295429552956295729582959296029612962296329642965296629672968296929702971297229732974297529762977297829792980298129822983298429852986298729882989299029912992299329942995299629972998299930003001300230033004300530063007300830093010301130123013301430153016301730183019302030213022302330243025302630273028302930303031303230333034303530363037303830393040304130423043304430453046304730483049305030513052305330543055305630573058305930603061306230633064306530663067306830693070307130723073307430753076307730783079308030813082308330843085308630873088308930903091309230933094309530963097309830993100310131023103310431053106310731083109311031113112311331143115311631173118311931203121312231233124312531263127312831293130313131323133313431353136313731383139314031413142314331443145314631473148314931503151315231533154315531563157315831593160316131623163316431653166316731683169317031713172317331743175317631773178317931803181318231833184318531863187318831893190319131923193319431953196319731983199320032013202320332043205320632073208320932103211321232133214321532163217321832193220322132223223322432253226322732283229323032313232323332343235323632373238323932403241324232433244324532463247324832493250325132523253325432553256325732583259326032613262326332643265326632673268326932703271327232733274327532763277327832793280328132823283328432853286328732883289329032913292329332943295329632973298329933003301330233033304330533063307330833093310331133123313331433153316331733183319332033213322332333243325332633273328332933303331333233333334333533363337333833393340334133423343334433453346334733483349335033513352335333543355335633573358335933603361336233633364336533663367336833693370337133723373337433753376337733783379338033813382338333843385338633873388338933903391339233933394339533963397339833993400340134023403340434053406340734083409341034113412341334143415341634173418341934203421342234233424342534263427342834293430343134323433343434353436343734383439344034413442344334443445344634473448344934503451345234533454345534563457345834593460346134623463346434653466346734683469347034713472347334743475347634773478347934803481348234833484348534863487348834893490349134923493349434953496349734983499350035013502350335043505350635073508350935103511351235133514351535163517351835193520352135223523352435253526352735283529353035313532353335343535353635373538353935403541354235433544354535463547354835493550355135523553355435553556355735583559356035613562356335643565356635673568356935703571357235733574357535763577357835793580358135823583358435853586358735883589359035913592359335943595359635973598359936003601360236033604360536063607360836093610361136123613361436153616361736183619362036213622362336243625362636273628362936303631363236333634363536363637363836393640364136423643364436453646364736483649365036513652365336543655365636573658365936603661366236633664366536663667366836693670367136723673367436753676367736783679368036813682368336843685368636873688368936903691369236933694369536963697369836993700370137023703370437053706370737083709371037113712371337143715371637173718371937203721372237233724372537263727372837293730373137323733373437353736373737383739374037413742374337443745374637473748374937503751375237533754375537563757375837593760376137623763376437653766376737683769377037713772377337743775377637773778377937803781378237833784378537863787378837893790379137923793379437953796379737983799380038013802380338043805380638073808380938103811381238133814381538163817381838193820382138223823382438253826382738283829383038313832383338343835383638373838383938403841384238433844384538463847384838493850385138523853385438553856385738583859386038613862386338643865386638673868386938703871387238733874387538763877387838793880388138823883388438853886388738883889389038913892389338943895389638973898389939003901390239033904390539063907390839093910391139123913391439153916391739183919392039213922392339243925392639273928392939303931393239333934393539363937393839393940394139423943394439453946394739483949395039513952395339543955395639573958395939603961396239633964396539663967396839693970397139723973397439753976397739783979398039813982398339843985398639873988398939903991399239933994399539963997399839994000400140024003400440054006400740084009401040114012401340144015401640174018401940204021402240234024402540264027402840294030403140324033403440354036403740384039404040414042404340444045404640474048404940504051405240534054405540564057405840594060406140624063406440654066406740684069407040714072407340744075407640774078407940804081408240834084408540864087408840894090409140924093409440954096409740984099410041014102410341044105410641074108410941104111411241134114411541164117411841194120412141224123412441254126412741284129413041314132413341344135413641374138413941404141414241434144414541464147414841494150415141524153415441554156415741584159416041614162416341644165416641674168416941704171417241734174417541764177417841794180418141824183418441854186418741884189419041914192419341944195419641974198419942004201420242034204420542064207420842094210421142124213421442154216421742184219422042214222422342244225422642274228422942304231423242334234423542364237423842394240424142424243424442454246424742484249425042514252425342544255425642574258425942604261426242634264426542664267426842694270427142724273427442754276427742784279428042814282428342844285428642874288428942904291429242934294429542964297429842994300430143024303430443054306430743084309431043114312431343144315431643174318431943204321432243234324432543264327432843294330433143324333433443354336433743384339434043414342434343444345434643474348434943504351435243534354435543564357435843594360436143624363436443654366436743684369437043714372437343744375437643774378437943804381438243834384438543864387438843894390439143924393439443954396439743984399440044014402440344044405440644074408440944104411441244134414441544164417441844194420442144224423442444254426442744284429443044314432443344344435443644374438443944404441444244434444444544464447444844494450445144524453445444554456445744584459446044614462446344644465446644674468446944704471447244734474447544764477447844794480448144824483448444854486448744884489449044914492449344944495449644974498449945004501450245034504450545064507450845094510451145124513451445154516451745184519452045214522452345244525452645274528452945304531453245334534453545364537453845394540454145424543454445454546454745484549455045514552455345544555455645574558455945604561456245634564456545664567456845694570457145724573457445754576457745784579458045814582458345844585458645874588458945904591459245934594459545964597459845994600460146024603460446054606460746084609461046114612461346144615461646174618461946204621462246234624462546264627462846294630463146324633463446354636463746384639464046414642464346444645464646474648464946504651465246534654465546564657465846594660466146624663466446654666466746684669467046714672467346744675467646774678467946804681468246834684468546864687468846894690469146924693469446954696469746984699470047014702470347044705470647074708
  1. /*
  2. * originally based on the dummy device.
  3. *
  4. * Copyright 1999, Thomas Davis, tadavis@lbl.gov.
  5. * Licensed under the GPL. Based on dummy.c, and eql.c devices.
  6. *
  7. * bonding.c: an Ethernet Bonding driver
  8. *
  9. * This is useful to talk to a Cisco EtherChannel compatible equipment:
  10. * Cisco 5500
  11. * Sun Trunking (Solaris)
  12. * Alteon AceDirector Trunks
  13. * Linux Bonding
  14. * and probably many L2 switches ...
  15. *
  16. * How it works:
  17. * ifconfig bond0 ipaddress netmask up
  18. * will setup a network device, with an ip address. No mac address
  19. * will be assigned at this time. The hw mac address will come from
  20. * the first slave bonded to the channel. All slaves will then use
  21. * this hw mac address.
  22. *
  23. * ifconfig bond0 down
  24. * will release all slaves, marking them as down.
  25. *
  26. * ifenslave bond0 eth0
  27. * will attach eth0 to bond0 as a slave. eth0 hw mac address will either
  28. * a: be used as initial mac address
  29. * b: if a hw mac address already is there, eth0's hw mac address
  30. * will then be set from bond0.
  31. *
  32. * v0.1 - first working version.
  33. * v0.2 - changed stats to be calculated by summing slaves stats.
  34. *
  35. * Changes:
  36. * Arnaldo Carvalho de Melo <acme@conectiva.com.br>
  37. * - fix leaks on failure at bond_init
  38. *
  39. * 2000/09/30 - Willy Tarreau <willy at meta-x.org>
  40. * - added trivial code to release a slave device.
  41. * - fixed security bug (CAP_NET_ADMIN not checked)
  42. * - implemented MII link monitoring to disable dead links :
  43. * All MII capable slaves are checked every <miimon> milliseconds
  44. * (100 ms seems good). This value can be changed by passing it to
  45. * insmod. A value of zero disables the monitoring (default).
  46. * - fixed an infinite loop in bond_xmit_roundrobin() when there's no
  47. * good slave.
  48. * - made the code hopefully SMP safe
  49. *
  50. * 2000/10/03 - Willy Tarreau <willy at meta-x.org>
  51. * - optimized slave lists based on relevant suggestions from Thomas Davis
  52. * - implemented active-backup method to obtain HA with two switches:
  53. * stay as long as possible on the same active interface, while we
  54. * also monitor the backup one (MII link status) because we want to know
  55. * if we are able to switch at any time. ( pass "mode=1" to insmod )
  56. * - lots of stress testings because we need it to be more robust than the
  57. * wires ! :->
  58. *
  59. * 2000/10/09 - Willy Tarreau <willy at meta-x.org>
  60. * - added up and down delays after link state change.
  61. * - optimized the slaves chaining so that when we run forward, we never
  62. * repass through the bond itself, but we can find it by searching
  63. * backwards. Renders the deletion more difficult, but accelerates the
  64. * scan.
  65. * - smarter enslaving and releasing.
  66. * - finer and more robust SMP locking
  67. *
  68. * 2000/10/17 - Willy Tarreau <willy at meta-x.org>
  69. * - fixed two potential SMP race conditions
  70. *
  71. * 2000/10/18 - Willy Tarreau <willy at meta-x.org>
  72. * - small fixes to the monitoring FSM in case of zero delays
  73. * 2000/11/01 - Willy Tarreau <willy at meta-x.org>
  74. * - fixed first slave not automatically used in trunk mode.
  75. * 2000/11/10 : spelling of "EtherChannel" corrected.
  76. * 2000/11/13 : fixed a race condition in case of concurrent accesses to ioctl().
  77. * 2000/12/16 : fixed improper usage of rtnl_exlock_nowait().
  78. *
  79. * 2001/1/3 - Chad N. Tindel <ctindel at ieee dot org>
  80. * - The bonding driver now simulates MII status monitoring, just like
  81. * a normal network device. It will show that the link is down iff
  82. * every slave in the bond shows that their links are down. If at least
  83. * one slave is up, the bond's MII status will appear as up.
  84. *
  85. * 2001/2/7 - Chad N. Tindel <ctindel at ieee dot org>
  86. * - Applications can now query the bond from user space to get
  87. * information which may be useful. They do this by calling
  88. * the BOND_INFO_QUERY ioctl. Once the app knows how many slaves
  89. * are in the bond, it can call the BOND_SLAVE_INFO_QUERY ioctl to
  90. * get slave specific information (# link failures, etc). See
  91. * <linux/if_bonding.h> for more details. The structs of interest
  92. * are ifbond and ifslave.
  93. *
  94. * 2001/4/5 - Chad N. Tindel <ctindel at ieee dot org>
  95. * - Ported to 2.4 Kernel
  96. *
  97. * 2001/5/2 - Jeffrey E. Mast <jeff at mastfamily dot com>
  98. * - When a device is detached from a bond, the slave device is no longer
  99. * left thinking that is has a master.
  100. *
  101. * 2001/5/16 - Jeffrey E. Mast <jeff at mastfamily dot com>
  102. * - memset did not appropriately initialized the bond rw_locks. Used
  103. * rwlock_init to initialize to unlocked state to prevent deadlock when
  104. * first attempting a lock
  105. * - Called SET_MODULE_OWNER for bond device
  106. *
  107. * 2001/5/17 - Tim Anderson <tsa at mvista.com>
  108. * - 2 paths for releasing for slave release; 1 through ioctl
  109. * and 2) through close. Both paths need to release the same way.
  110. * - the free slave in bond release is changing slave status before
  111. * the free. The netdev_set_master() is intended to change slave state
  112. * so it should not be done as part of the release process.
  113. * - Simple rule for slave state at release: only the active in A/B and
  114. * only one in the trunked case.
  115. *
  116. * 2001/6/01 - Tim Anderson <tsa at mvista.com>
  117. * - Now call dev_close when releasing a slave so it doesn't screw up
  118. * out routing table.
  119. *
  120. * 2001/6/01 - Chad N. Tindel <ctindel at ieee dot org>
  121. * - Added /proc support for getting bond and slave information.
  122. * Information is in /proc/net/<bond device>/info.
  123. * - Changed the locking when calling bond_close to prevent deadlock.
  124. *
  125. * 2001/8/05 - Janice Girouard <girouard at us.ibm.com>
  126. * - correct problem where refcnt of slave is not incremented in bond_ioctl
  127. * so the system hangs when halting.
  128. * - correct locking problem when unable to malloc in bond_enslave.
  129. * - adding bond_xmit_xor logic.
  130. * - adding multiple bond device support.
  131. *
  132. * 2001/8/13 - Erik Habbinga <erik_habbinga at hp dot com>
  133. * - correct locking problem with rtnl_exlock_nowait
  134. *
  135. * 2001/8/23 - Janice Girouard <girouard at us.ibm.com>
  136. * - bzero initial dev_bonds, to correct oops
  137. * - convert SIOCDEVPRIVATE to new MII ioctl calls
  138. *
  139. * 2001/9/13 - Takao Indoh <indou dot takao at jp dot fujitsu dot com>
  140. * - Add the BOND_CHANGE_ACTIVE ioctl implementation
  141. *
  142. * 2001/9/14 - Mark Huth <mhuth at mvista dot com>
  143. * - Change MII_LINK_READY to not check for end of auto-negotiation,
  144. * but only for an up link.
  145. *
  146. * 2001/9/20 - Chad N. Tindel <ctindel at ieee dot org>
  147. * - Add the device field to bonding_t. Previously the net_device
  148. * corresponding to a bond wasn't available from the bonding_t
  149. * structure.
  150. *
  151. * 2001/9/25 - Janice Girouard <girouard at us.ibm.com>
  152. * - add arp_monitor for active backup mode
  153. *
  154. * 2001/10/23 - Takao Indoh <indou dot takao at jp dot fujitsu dot com>
  155. * - Various memory leak fixes
  156. *
  157. * 2001/11/5 - Mark Huth <mark dot huth at mvista dot com>
  158. * - Don't take rtnl lock in bond_mii_monitor as it deadlocks under
  159. * certain hotswap conditions.
  160. * Note: this same change may be required in bond_arp_monitor ???
  161. * - Remove possibility of calling bond_sethwaddr with NULL slave_dev ptr
  162. * - Handle hot swap ethernet interface deregistration events to remove
  163. * kernel oops following hot swap of enslaved interface
  164. *
  165. * 2002/1/2 - Chad N. Tindel <ctindel at ieee dot org>
  166. * - Restore original slave flags at release time.
  167. *
  168. * 2002/02/18 - Erik Habbinga <erik_habbinga at hp dot com>
  169. * - bond_release(): calling kfree on our_slave after call to
  170. * bond_restore_slave_flags, not before
  171. * - bond_enslave(): saving slave flags into original_flags before
  172. * call to netdev_set_master, so the IFF_SLAVE flag doesn't end
  173. * up in original_flags
  174. *
  175. * 2002/04/05 - Mark Smith <mark.smith at comdev dot cc> and
  176. * Steve Mead <steve.mead at comdev dot cc>
  177. * - Port Gleb Natapov's multicast support patchs from 2.4.12
  178. * to 2.4.18 adding support for multicast.
  179. *
  180. * 2002/06/10 - Tony Cureington <tony.cureington * hp_com>
  181. * - corrected uninitialized pointer (ifr.ifr_data) in bond_check_dev_link;
  182. * actually changed function to use MIIPHY, then MIIREG, and finally
  183. * ETHTOOL to determine the link status
  184. * - fixed bad ifr_data pointer assignments in bond_ioctl
  185. * - corrected mode 1 being reported as active-backup in bond_get_info;
  186. * also added text to distinguish type of load balancing (rr or xor)
  187. * - change arp_ip_target module param from "1-12s" (array of 12 ptrs)
  188. * to "s" (a single ptr)
  189. *
  190. * 2002/08/30 - Jay Vosburgh <fubar at us dot ibm dot com>
  191. * - Removed acquisition of xmit_lock in set_multicast_list; caused
  192. * deadlock on SMP (lock is held by caller).
  193. * - Revamped SIOCGMIIPHY, SIOCGMIIREG portion of bond_check_dev_link().
  194. *
  195. * 2002/09/18 - Jay Vosburgh <fubar at us dot ibm dot com>
  196. * - Fixed up bond_check_dev_link() (and callers): removed some magic
  197. * numbers, banished local MII_ defines, wrapped ioctl calls to
  198. * prevent EFAULT errors
  199. *
  200. * 2002/9/30 - Jay Vosburgh <fubar at us dot ibm dot com>
  201. * - make sure the ip target matches the arp_target before saving the
  202. * hw address.
  203. *
  204. * 2002/9/30 - Dan Eisner <eisner at 2robots dot com>
  205. * - make sure my_ip is set before taking down the link, since
  206. * not all switches respond if the source ip is not set.
  207. *
  208. * 2002/10/8 - Janice Girouard <girouard at us dot ibm dot com>
  209. * - read in the local ip address when enslaving a device
  210. * - add primary support
  211. * - make sure 2*arp_interval has passed when a new device
  212. * is brought on-line before taking it down.
  213. *
  214. * 2002/09/11 - Philippe De Muyter <phdm at macqel dot be>
  215. * - Added bond_xmit_broadcast logic.
  216. * - Added bond_mode() support function.
  217. *
  218. * 2002/10/26 - Laurent Deniel <laurent.deniel at free.fr>
  219. * - allow to register multicast addresses only on active slave
  220. * (useful in active-backup mode)
  221. * - add multicast module parameter
  222. * - fix deletion of multicast groups after unloading module
  223. *
  224. * 2002/11/06 - Kameshwara Rayaprolu <kameshwara.rao * wipro_com>
  225. * - Changes to prevent panic from closing the device twice; if we close
  226. * the device in bond_release, we must set the original_flags to down
  227. * so it won't be closed again by the network layer.
  228. *
  229. * 2002/11/07 - Tony Cureington <tony.cureington * hp_com>
  230. * - Fix arp_target_hw_addr memory leak
  231. * - Created activebackup_arp_monitor function to handle arp monitoring
  232. * in active backup mode - the bond_arp_monitor had several problems...
  233. * such as allowing slaves to tx arps sequentially without any delay
  234. * for a response
  235. * - Renamed bond_arp_monitor to loadbalance_arp_monitor and re-wrote
  236. * this function to just handle arp monitoring in load-balancing mode;
  237. * it is a lot more compact now
  238. * - Changes to ensure one and only one slave transmits in active-backup
  239. * mode
  240. * - Robustesize parameters; warn users about bad combinations of
  241. * parameters; also if miimon is specified and a network driver does
  242. * not support MII or ETHTOOL, inform the user of this
  243. * - Changes to support link_failure_count when in arp monitoring mode
  244. * - Fix up/down delay reported in /proc
  245. * - Added version; log version; make version available from "modinfo -d"
  246. * - Fixed problem in bond_check_dev_link - if the first IOCTL (SIOCGMIIPH)
  247. * failed, the ETHTOOL ioctl never got a chance
  248. *
  249. * 2002/11/16 - Laurent Deniel <laurent.deniel at free.fr>
  250. * - fix multicast handling in activebackup_arp_monitor
  251. * - remove one unnecessary and confusing curr_active_slave == slave test
  252. * in activebackup_arp_monitor
  253. *
  254. * 2002/11/17 - Laurent Deniel <laurent.deniel at free.fr>
  255. * - fix bond_slave_info_query when slave_id = num_slaves
  256. *
  257. * 2002/11/19 - Janice Girouard <girouard at us dot ibm dot com>
  258. * - correct ifr_data reference. Update ifr_data reference
  259. * to mii_ioctl_data struct values to avoid confusion.
  260. *
  261. * 2002/11/22 - Bert Barbe <bert.barbe at oracle dot com>
  262. * - Add support for multiple arp_ip_target
  263. *
  264. * 2002/12/13 - Jay Vosburgh <fubar at us dot ibm dot com>
  265. * - Changed to allow text strings for mode and multicast, e.g.,
  266. * insmod bonding mode=active-backup. The numbers still work.
  267. * One change: an invalid choice will cause module load failure,
  268. * rather than the previous behavior of just picking one.
  269. * - Minor cleanups; got rid of dup ctype stuff, atoi function
  270. *
  271. * 2003/02/07 - Jay Vosburgh <fubar at us dot ibm dot com>
  272. * - Added use_carrier module parameter that causes miimon to
  273. * use netif_carrier_ok() test instead of MII/ETHTOOL ioctls.
  274. * - Minor cleanups; consolidated ioctl calls to one function.
  275. *
  276. * 2003/02/07 - Tony Cureington <tony.cureington * hp_com>
  277. * - Fix bond_mii_monitor() logic error that could result in
  278. * bonding round-robin mode ignoring links after failover/recovery
  279. *
  280. * 2003/03/17 - Jay Vosburgh <fubar at us dot ibm dot com>
  281. * - kmalloc fix (GFP_KERNEL to GFP_ATOMIC) reported by
  282. * Shmulik dot Hen at intel.com.
  283. * - Based on discussion on mailing list, changed use of
  284. * update_slave_cnt(), created wrapper functions for adding/removing
  285. * slaves, changed bond_xmit_xor() to check slave_cnt instead of
  286. * checking slave and slave->dev (which only worked by accident).
  287. * - Misc code cleanup: get arp_send() prototype from header file,
  288. * add max_bonds to bonding.txt.
  289. *
  290. * 2003/03/18 - Tsippy Mendelson <tsippy.mendelson at intel dot com> and
  291. * Shmulik Hen <shmulik.hen at intel dot com>
  292. * - Make sure only bond_attach_slave() and bond_detach_slave() can
  293. * manipulate the slave list, including slave_cnt, even when in
  294. * bond_release_all().
  295. * - Fixed hang in bond_release() with traffic running:
  296. * netdev_set_master() must not be called from within the bond lock.
  297. *
  298. * 2003/03/18 - Tsippy Mendelson <tsippy.mendelson at intel dot com> and
  299. * Shmulik Hen <shmulik.hen at intel dot com>
  300. * - Fixed hang in bond_enslave() with traffic running:
  301. * netdev_set_master() must not be called from within the bond lock.
  302. *
  303. * 2003/03/18 - Amir Noam <amir.noam at intel dot com>
  304. * - Added support for getting slave's speed and duplex via ethtool.
  305. * Needed for 802.3ad and other future modes.
  306. *
  307. * 2003/03/18 - Tsippy Mendelson <tsippy.mendelson at intel dot com> and
  308. * Shmulik Hen <shmulik.hen at intel dot com>
  309. * - Enable support of modes that need to use the unique mac address of
  310. * each slave.
  311. * * bond_enslave(): Moved setting the slave's mac address, and
  312. * openning it, from the application to the driver. This breaks
  313. * backward comaptibility with old versions of ifenslave that open
  314. * the slave before enalsving it !!!.
  315. * * bond_release(): The driver also takes care of closing the slave
  316. * and restoring its original mac address.
  317. * - Removed the code that restores all base driver's flags.
  318. * Flags are automatically restored once all undo stages are done
  319. * properly.
  320. * - Block possibility of enslaving before the master is up. This
  321. * prevents putting the system in an unstable state.
  322. *
  323. * 2003/03/18 - Amir Noam <amir.noam at intel dot com>,
  324. * Tsippy Mendelson <tsippy.mendelson at intel dot com> and
  325. * Shmulik Hen <shmulik.hen at intel dot com>
  326. * - Added support for IEEE 802.3ad Dynamic link aggregation mode.
  327. *
  328. * 2003/05/01 - Amir Noam <amir.noam at intel dot com>
  329. * - Added ABI version control to restore compatibility between
  330. * new/old ifenslave and new/old bonding.
  331. *
  332. * 2003/05/01 - Shmulik Hen <shmulik.hen at intel dot com>
  333. * - Fixed bug in bond_release_all(): save old value of curr_active_slave
  334. * before setting it to NULL.
  335. * - Changed driver versioning scheme to include version number instead
  336. * of release date (that is already in another field). There are 3
  337. * fields X.Y.Z where:
  338. * X - Major version - big behavior changes
  339. * Y - Minor version - addition of features
  340. * Z - Extra version - minor changes and bug fixes
  341. * The current version is 1.0.0 as a base line.
  342. *
  343. * 2003/05/01 - Tsippy Mendelson <tsippy.mendelson at intel dot com> and
  344. * Amir Noam <amir.noam at intel dot com>
  345. * - Added support for lacp_rate module param.
  346. * - Code beautification and style changes (mainly in comments).
  347. * new version - 1.0.1
  348. *
  349. * 2003/05/01 - Shmulik Hen <shmulik.hen at intel dot com>
  350. * - Based on discussion on mailing list, changed locking scheme
  351. * to use lock/unlock or lock_bh/unlock_bh appropriately instead
  352. * of lock_irqsave/unlock_irqrestore. The new scheme helps exposing
  353. * hidden bugs and solves system hangs that occurred due to the fact
  354. * that holding lock_irqsave doesn't prevent softirqs from running.
  355. * This also increases total throughput since interrupts are not
  356. * blocked on each transmitted packets or monitor timeout.
  357. * new version - 2.0.0
  358. *
  359. * 2003/05/01 - Shmulik Hen <shmulik.hen at intel dot com>
  360. * - Added support for Transmit load balancing mode.
  361. * - Concentrate all assignments of curr_active_slave to a single point
  362. * so specific modes can take actions when the primary adapter is
  363. * changed.
  364. * - Take the updelay parameter into consideration during bond_enslave
  365. * since some adapters loose their link during setting the device.
  366. * - Renamed bond_3ad_link_status_changed() to
  367. * bond_3ad_handle_link_change() for compatibility with TLB.
  368. * new version - 2.1.0
  369. *
  370. * 2003/05/01 - Tsippy Mendelson <tsippy.mendelson at intel dot com>
  371. * - Added support for Adaptive load balancing mode which is
  372. * equivalent to Transmit load balancing + Receive load balancing.
  373. * new version - 2.2.0
  374. *
  375. * 2003/05/15 - Jay Vosburgh <fubar at us dot ibm dot com>
  376. * - Applied fix to activebackup_arp_monitor posted to bonding-devel
  377. * by Tony Cureington <tony.cureington * hp_com>. Fixes ARP
  378. * monitor endless failover bug. Version to 2.2.10
  379. *
  380. * 2003/05/20 - Amir Noam <amir.noam at intel dot com>
  381. * - Fixed bug in ABI version control - Don't commit to a specific
  382. * ABI version if receiving unsupported ioctl commands.
  383. *
  384. * 2003/05/22 - Jay Vosburgh <fubar at us dot ibm dot com>
  385. * - Fix ifenslave -c causing bond to loose existing routes;
  386. * added bond_set_mac_address() that doesn't require the
  387. * bond to be down.
  388. * - In conjunction with fix for ifenslave -c, in
  389. * bond_change_active(), changing to the already active slave
  390. * is no longer an error (it successfully does nothing).
  391. *
  392. * 2003/06/30 - Amir Noam <amir.noam at intel dot com>
  393. * - Fixed bond_change_active() for ALB/TLB modes.
  394. * Version to 2.2.14.
  395. *
  396. * 2003/07/29 - Amir Noam <amir.noam at intel dot com>
  397. * - Fixed ARP monitoring bug.
  398. * Version to 2.2.15.
  399. *
  400. * 2003/07/31 - Willy Tarreau <willy at ods dot org>
  401. * - Fixed kernel panic when using ARP monitoring without
  402. * setting bond's IP address.
  403. * Version to 2.2.16.
  404. *
  405. * 2003/08/06 - Amir Noam <amir.noam at intel dot com>
  406. * - Back port from 2.6: use alloc_netdev(); fix /proc handling;
  407. * made stats a part of bond struct so no need to allocate
  408. * and free it separately; use standard list operations instead
  409. * of pre-allocated array of bonds.
  410. * Version to 2.3.0.
  411. *
  412. * 2003/08/07 - Jay Vosburgh <fubar at us dot ibm dot com>,
  413. * Amir Noam <amir.noam at intel dot com> and
  414. * Shmulik Hen <shmulik.hen at intel dot com>
  415. * - Propagating master's settings: Distinguish between modes that
  416. * use a primary slave from those that don't, and propagate settings
  417. * accordingly; Consolidate change_active opeartions and add
  418. * reselect_active and find_best opeartions; Decouple promiscuous
  419. * handling from the multicast mode setting; Add support for changing
  420. * HW address and MTU with proper unwind; Consolidate procfs code,
  421. * add CHANGENAME handler; Enhance netdev notification handling.
  422. * Version to 2.4.0.
  423. *
  424. * 2003/09/15 - Stephen Hemminger <shemminger at osdl dot org>,
  425. * Amir Noam <amir.noam at intel dot com>
  426. * - Convert /proc to seq_file interface.
  427. * Change /proc/net/bondX/info to /proc/net/bonding/bondX.
  428. * Set version to 2.4.1.
  429. *
  430. * 2003/11/20 - Amir Noam <amir.noam at intel dot com>
  431. * - Fix /proc creation/destruction.
  432. *
  433. * 2003/12/01 - Shmulik Hen <shmulik.hen at intel dot com>
  434. * - Massive cleanup - Set version to 2.5.0
  435. * Code changes:
  436. * o Consolidate format of prints and debug prints.
  437. * o Remove bonding_t/slave_t typedefs and consolidate all casts.
  438. * o Remove dead code and unnecessary checks.
  439. * o Consolidate starting/stopping timers.
  440. * o Consolidate handling of primary module param throughout the code.
  441. * o Removed multicast module param support - all settings are done
  442. * according to mode.
  443. * o Slave list iteration - bond is no longer part of the list,
  444. * added cyclic list iteration macros.
  445. * o Consolidate error handling in all xmit functions.
  446. * Style changes:
  447. * o Consolidate function naming and declarations.
  448. * o Consolidate function params and local variables names.
  449. * o Consolidate return values.
  450. * o Consolidate curly braces.
  451. * o Consolidate conditionals format.
  452. * o Change struct member names and types.
  453. * o Chomp trailing spaces, remove empty lines, fix indentations.
  454. * o Re-organize code according to context.
  455. *
  456. * 2003/12/30 - Amir Noam <amir.noam at intel dot com>
  457. * - Fixed: Cannot remove and re-enslave the original active slave.
  458. * - Fixed: Releasing the original active slave causes mac address
  459. * duplication.
  460. * - Add support for slaves that use ethtool_ops.
  461. * Set version to 2.5.3.
  462. *
  463. * 2004/01/05 - Amir Noam <amir.noam at intel dot com>
  464. * - Save bonding parameters per bond instead of using the global values.
  465. * Set version to 2.5.4.
  466. *
  467. * 2004/01/14 - Shmulik Hen <shmulik.hen at intel dot com>
  468. * - Enhance VLAN support:
  469. * * Add support for VLAN hardware acceleration capable slaves.
  470. * * Add capability to tag self generated packets in ALB/TLB modes.
  471. * Set version to 2.6.0.
  472. * 2004/10/29 - Mitch Williams <mitch.a.williams at intel dot com>
  473. * - Fixed bug when unloading module while using 802.3ad. If
  474. * spinlock debugging is turned on, this causes a stack dump.
  475. * Solution is to move call to dev_remove_pack outside of the
  476. * spinlock.
  477. * Set version to 2.6.1.
  478. *
  479. */
  480. //#define BONDING_DEBUG 1
  481. #include <linux/config.h>
  482. #include <linux/kernel.h>
  483. #include <linux/module.h>
  484. #include <linux/sched.h>
  485. #include <linux/types.h>
  486. #include <linux/fcntl.h>
  487. #include <linux/interrupt.h>
  488. #include <linux/ptrace.h>
  489. #include <linux/ioport.h>
  490. #include <linux/in.h>
  491. #include <linux/ip.h>
  492. #include <linux/slab.h>
  493. #include <linux/string.h>
  494. #include <linux/init.h>
  495. #include <linux/timer.h>
  496. #include <linux/socket.h>
  497. #include <linux/ctype.h>
  498. #include <linux/inet.h>
  499. #include <linux/bitops.h>
  500. #include <asm/system.h>
  501. #include <asm/io.h>
  502. #include <asm/dma.h>
  503. #include <asm/uaccess.h>
  504. #include <linux/errno.h>
  505. #include <linux/netdevice.h>
  506. #include <linux/inetdevice.h>
  507. #include <linux/etherdevice.h>
  508. #include <linux/skbuff.h>
  509. #include <net/sock.h>
  510. #include <linux/rtnetlink.h>
  511. #include <linux/proc_fs.h>
  512. #include <linux/seq_file.h>
  513. #include <linux/smp.h>
  514. #include <linux/if_ether.h>
  515. #include <net/arp.h>
  516. #include <linux/mii.h>
  517. #include <linux/ethtool.h>
  518. #include <linux/if_vlan.h>
  519. #include <linux/if_bonding.h>
  520. #include "bonding.h"
  521. #include "bond_3ad.h"
  522. #include "bond_alb.h"
  523. /*---------------------------- Module parameters ----------------------------*/
  524. /* monitor all links that often (in milliseconds). <=0 disables monitoring */
  525. #define BOND_LINK_MON_INTERV 0
  526. #define BOND_LINK_ARP_INTERV 0
  527. static int max_bonds = BOND_DEFAULT_MAX_BONDS;
  528. static int miimon = BOND_LINK_MON_INTERV;
  529. static int updelay = 0;
  530. static int downdelay = 0;
  531. static int use_carrier = 1;
  532. static char *mode = NULL;
  533. static char *primary = NULL;
  534. static char *lacp_rate = NULL;
  535. static int arp_interval = BOND_LINK_ARP_INTERV;
  536. static char *arp_ip_target[BOND_MAX_ARP_TARGETS] = { NULL, };
  537. module_param(max_bonds, int, 0);
  538. MODULE_PARM_DESC(max_bonds, "Max number of bonded devices");
  539. module_param(miimon, int, 0);
  540. MODULE_PARM_DESC(miimon, "Link check interval in milliseconds");
  541. module_param(updelay, int, 0);
  542. MODULE_PARM_DESC(updelay, "Delay before considering link up, in milliseconds");
  543. module_param(downdelay, int, 0);
  544. MODULE_PARM_DESC(downdelay, "Delay before considering link down, in milliseconds");
  545. module_param(use_carrier, int, 0);
  546. MODULE_PARM_DESC(use_carrier, "Use netif_carrier_ok (vs MII ioctls) in miimon; 0 for off, 1 for on (default)");
  547. module_param(mode, charp, 0);
  548. MODULE_PARM_DESC(mode, "Mode of operation : 0 for round robin, 1 for active-backup, 2 for xor");
  549. module_param(primary, charp, 0);
  550. MODULE_PARM_DESC(primary, "Primary network device to use");
  551. module_param(lacp_rate, charp, 0);
  552. MODULE_PARM_DESC(lacp_rate, "LACPDU tx rate to request from 802.3ad partner (slow/fast)");
  553. module_param(arp_interval, int, 0);
  554. MODULE_PARM_DESC(arp_interval, "arp interval in milliseconds");
  555. module_param_array(arp_ip_target, charp, NULL, 0);
  556. MODULE_PARM_DESC(arp_ip_target, "arp targets in n.n.n.n form");
  557. /*----------------------------- Global variables ----------------------------*/
  558. static const char *version =
  559. DRV_DESCRIPTION ": v" DRV_VERSION " (" DRV_RELDATE ")\n";
  560. static LIST_HEAD(bond_dev_list);
  561. #ifdef CONFIG_PROC_FS
  562. static struct proc_dir_entry *bond_proc_dir = NULL;
  563. #endif
  564. static u32 arp_target[BOND_MAX_ARP_TARGETS] = { 0, } ;
  565. static int arp_ip_count = 0;
  566. static u32 my_ip = 0;
  567. static int bond_mode = BOND_MODE_ROUNDROBIN;
  568. static int lacp_fast = 0;
  569. static int app_abi_ver = 0;
  570. static int orig_app_abi_ver = -1; /* This is used to save the first ABI version
  571. * we receive from the application. Once set,
  572. * it won't be changed, and the module will
  573. * refuse to enslave/release interfaces if the
  574. * command comes from an application using
  575. * another ABI version.
  576. */
  577. struct bond_parm_tbl {
  578. char *modename;
  579. int mode;
  580. };
  581. static struct bond_parm_tbl bond_lacp_tbl[] = {
  582. { "slow", AD_LACP_SLOW},
  583. { "fast", AD_LACP_FAST},
  584. { NULL, -1},
  585. };
  586. static struct bond_parm_tbl bond_mode_tbl[] = {
  587. { "balance-rr", BOND_MODE_ROUNDROBIN},
  588. { "active-backup", BOND_MODE_ACTIVEBACKUP},
  589. { "balance-xor", BOND_MODE_XOR},
  590. { "broadcast", BOND_MODE_BROADCAST},
  591. { "802.3ad", BOND_MODE_8023AD},
  592. { "balance-tlb", BOND_MODE_TLB},
  593. { "balance-alb", BOND_MODE_ALB},
  594. { NULL, -1},
  595. };
  596. /*-------------------------- Forward declarations ---------------------------*/
  597. static inline void bond_set_mode_ops(struct net_device *bond_dev, int mode);
  598. /*---------------------------- General routines -----------------------------*/
  599. static const char *bond_mode_name(int mode)
  600. {
  601. switch (mode) {
  602. case BOND_MODE_ROUNDROBIN :
  603. return "load balancing (round-robin)";
  604. case BOND_MODE_ACTIVEBACKUP :
  605. return "fault-tolerance (active-backup)";
  606. case BOND_MODE_XOR :
  607. return "load balancing (xor)";
  608. case BOND_MODE_BROADCAST :
  609. return "fault-tolerance (broadcast)";
  610. case BOND_MODE_8023AD:
  611. return "IEEE 802.3ad Dynamic link aggregation";
  612. case BOND_MODE_TLB:
  613. return "transmit load balancing";
  614. case BOND_MODE_ALB:
  615. return "adaptive load balancing";
  616. default:
  617. return "unknown";
  618. }
  619. }
  620. /*---------------------------------- VLAN -----------------------------------*/
  621. /**
  622. * bond_add_vlan - add a new vlan id on bond
  623. * @bond: bond that got the notification
  624. * @vlan_id: the vlan id to add
  625. *
  626. * Returns -ENOMEM if allocation failed.
  627. */
  628. static int bond_add_vlan(struct bonding *bond, unsigned short vlan_id)
  629. {
  630. struct vlan_entry *vlan;
  631. dprintk("bond: %s, vlan id %d\n",
  632. (bond ? bond->dev->name: "None"), vlan_id);
  633. vlan = kmalloc(sizeof(struct vlan_entry), GFP_KERNEL);
  634. if (!vlan) {
  635. return -ENOMEM;
  636. }
  637. INIT_LIST_HEAD(&vlan->vlan_list);
  638. vlan->vlan_id = vlan_id;
  639. write_lock_bh(&bond->lock);
  640. list_add_tail(&vlan->vlan_list, &bond->vlan_list);
  641. write_unlock_bh(&bond->lock);
  642. dprintk("added VLAN ID %d on bond %s\n", vlan_id, bond->dev->name);
  643. return 0;
  644. }
  645. /**
  646. * bond_del_vlan - delete a vlan id from bond
  647. * @bond: bond that got the notification
  648. * @vlan_id: the vlan id to delete
  649. *
  650. * returns -ENODEV if @vlan_id was not found in @bond.
  651. */
  652. static int bond_del_vlan(struct bonding *bond, unsigned short vlan_id)
  653. {
  654. struct vlan_entry *vlan, *next;
  655. int res = -ENODEV;
  656. dprintk("bond: %s, vlan id %d\n", bond->dev->name, vlan_id);
  657. write_lock_bh(&bond->lock);
  658. list_for_each_entry_safe(vlan, next, &bond->vlan_list, vlan_list) {
  659. if (vlan->vlan_id == vlan_id) {
  660. list_del(&vlan->vlan_list);
  661. if ((bond->params.mode == BOND_MODE_TLB) ||
  662. (bond->params.mode == BOND_MODE_ALB)) {
  663. bond_alb_clear_vlan(bond, vlan_id);
  664. }
  665. dprintk("removed VLAN ID %d from bond %s\n", vlan_id,
  666. bond->dev->name);
  667. kfree(vlan);
  668. if (list_empty(&bond->vlan_list) &&
  669. (bond->slave_cnt == 0)) {
  670. /* Last VLAN removed and no slaves, so
  671. * restore block on adding VLANs. This will
  672. * be removed once new slaves that are not
  673. * VLAN challenged will be added.
  674. */
  675. bond->dev->features |= NETIF_F_VLAN_CHALLENGED;
  676. }
  677. res = 0;
  678. goto out;
  679. }
  680. }
  681. dprintk("couldn't find VLAN ID %d in bond %s\n", vlan_id,
  682. bond->dev->name);
  683. out:
  684. write_unlock_bh(&bond->lock);
  685. return res;
  686. }
  687. /**
  688. * bond_has_challenged_slaves
  689. * @bond: the bond we're working on
  690. *
  691. * Searches the slave list. Returns 1 if a vlan challenged slave
  692. * was found, 0 otherwise.
  693. *
  694. * Assumes bond->lock is held.
  695. */
  696. static int bond_has_challenged_slaves(struct bonding *bond)
  697. {
  698. struct slave *slave;
  699. int i;
  700. bond_for_each_slave(bond, slave, i) {
  701. if (slave->dev->features & NETIF_F_VLAN_CHALLENGED) {
  702. dprintk("found VLAN challenged slave - %s\n",
  703. slave->dev->name);
  704. return 1;
  705. }
  706. }
  707. dprintk("no VLAN challenged slaves found\n");
  708. return 0;
  709. }
  710. /**
  711. * bond_next_vlan - safely skip to the next item in the vlans list.
  712. * @bond: the bond we're working on
  713. * @curr: item we're advancing from
  714. *
  715. * Returns %NULL if list is empty, bond->next_vlan if @curr is %NULL,
  716. * or @curr->next otherwise (even if it is @curr itself again).
  717. *
  718. * Caller must hold bond->lock
  719. */
  720. struct vlan_entry *bond_next_vlan(struct bonding *bond, struct vlan_entry *curr)
  721. {
  722. struct vlan_entry *next, *last;
  723. if (list_empty(&bond->vlan_list)) {
  724. return NULL;
  725. }
  726. if (!curr) {
  727. next = list_entry(bond->vlan_list.next,
  728. struct vlan_entry, vlan_list);
  729. } else {
  730. last = list_entry(bond->vlan_list.prev,
  731. struct vlan_entry, vlan_list);
  732. if (last == curr) {
  733. next = list_entry(bond->vlan_list.next,
  734. struct vlan_entry, vlan_list);
  735. } else {
  736. next = list_entry(curr->vlan_list.next,
  737. struct vlan_entry, vlan_list);
  738. }
  739. }
  740. return next;
  741. }
  742. /**
  743. * bond_dev_queue_xmit - Prepare skb for xmit.
  744. *
  745. * @bond: bond device that got this skb for tx.
  746. * @skb: hw accel VLAN tagged skb to transmit
  747. * @slave_dev: slave that is supposed to xmit this skbuff
  748. *
  749. * When the bond gets an skb to transmit that is
  750. * already hardware accelerated VLAN tagged, and it
  751. * needs to relay this skb to a slave that is not
  752. * hw accel capable, the skb needs to be "unaccelerated",
  753. * i.e. strip the hwaccel tag and re-insert it as part
  754. * of the payload.
  755. */
  756. int bond_dev_queue_xmit(struct bonding *bond, struct sk_buff *skb, struct net_device *slave_dev)
  757. {
  758. unsigned short vlan_id;
  759. if (!list_empty(&bond->vlan_list) &&
  760. !(slave_dev->features & NETIF_F_HW_VLAN_TX) &&
  761. vlan_get_tag(skb, &vlan_id) == 0) {
  762. skb->dev = slave_dev;
  763. skb = vlan_put_tag(skb, vlan_id);
  764. if (!skb) {
  765. /* vlan_put_tag() frees the skb in case of error,
  766. * so return success here so the calling functions
  767. * won't attempt to free is again.
  768. */
  769. return 0;
  770. }
  771. } else {
  772. skb->dev = slave_dev;
  773. }
  774. skb->priority = 1;
  775. dev_queue_xmit(skb);
  776. return 0;
  777. }
  778. /*
  779. * In the following 3 functions, bond_vlan_rx_register(), bond_vlan_rx_add_vid
  780. * and bond_vlan_rx_kill_vid, We don't protect the slave list iteration with a
  781. * lock because:
  782. * a. This operation is performed in IOCTL context,
  783. * b. The operation is protected by the RTNL semaphore in the 8021q code,
  784. * c. Holding a lock with BH disabled while directly calling a base driver
  785. * entry point is generally a BAD idea.
  786. *
  787. * The design of synchronization/protection for this operation in the 8021q
  788. * module is good for one or more VLAN devices over a single physical device
  789. * and cannot be extended for a teaming solution like bonding, so there is a
  790. * potential race condition here where a net device from the vlan group might
  791. * be referenced (either by a base driver or the 8021q code) while it is being
  792. * removed from the system. However, it turns out we're not making matters
  793. * worse, and if it works for regular VLAN usage it will work here too.
  794. */
  795. /**
  796. * bond_vlan_rx_register - Propagates registration to slaves
  797. * @bond_dev: bonding net device that got called
  798. * @grp: vlan group being registered
  799. */
  800. static void bond_vlan_rx_register(struct net_device *bond_dev, struct vlan_group *grp)
  801. {
  802. struct bonding *bond = bond_dev->priv;
  803. struct slave *slave;
  804. int i;
  805. bond->vlgrp = grp;
  806. bond_for_each_slave(bond, slave, i) {
  807. struct net_device *slave_dev = slave->dev;
  808. if ((slave_dev->features & NETIF_F_HW_VLAN_RX) &&
  809. slave_dev->vlan_rx_register) {
  810. slave_dev->vlan_rx_register(slave_dev, grp);
  811. }
  812. }
  813. }
  814. /**
  815. * bond_vlan_rx_add_vid - Propagates adding an id to slaves
  816. * @bond_dev: bonding net device that got called
  817. * @vid: vlan id being added
  818. */
  819. static void bond_vlan_rx_add_vid(struct net_device *bond_dev, uint16_t vid)
  820. {
  821. struct bonding *bond = bond_dev->priv;
  822. struct slave *slave;
  823. int i, res;
  824. bond_for_each_slave(bond, slave, i) {
  825. struct net_device *slave_dev = slave->dev;
  826. if ((slave_dev->features & NETIF_F_HW_VLAN_FILTER) &&
  827. slave_dev->vlan_rx_add_vid) {
  828. slave_dev->vlan_rx_add_vid(slave_dev, vid);
  829. }
  830. }
  831. res = bond_add_vlan(bond, vid);
  832. if (res) {
  833. printk(KERN_ERR DRV_NAME
  834. ": %s: Failed to add vlan id %d\n",
  835. bond_dev->name, vid);
  836. }
  837. }
  838. /**
  839. * bond_vlan_rx_kill_vid - Propagates deleting an id to slaves
  840. * @bond_dev: bonding net device that got called
  841. * @vid: vlan id being removed
  842. */
  843. static void bond_vlan_rx_kill_vid(struct net_device *bond_dev, uint16_t vid)
  844. {
  845. struct bonding *bond = bond_dev->priv;
  846. struct slave *slave;
  847. struct net_device *vlan_dev;
  848. int i, res;
  849. bond_for_each_slave(bond, slave, i) {
  850. struct net_device *slave_dev = slave->dev;
  851. if ((slave_dev->features & NETIF_F_HW_VLAN_FILTER) &&
  852. slave_dev->vlan_rx_kill_vid) {
  853. /* Save and then restore vlan_dev in the grp array,
  854. * since the slave's driver might clear it.
  855. */
  856. vlan_dev = bond->vlgrp->vlan_devices[vid];
  857. slave_dev->vlan_rx_kill_vid(slave_dev, vid);
  858. bond->vlgrp->vlan_devices[vid] = vlan_dev;
  859. }
  860. }
  861. res = bond_del_vlan(bond, vid);
  862. if (res) {
  863. printk(KERN_ERR DRV_NAME
  864. ": %s: Failed to remove vlan id %d\n",
  865. bond_dev->name, vid);
  866. }
  867. }
  868. static void bond_add_vlans_on_slave(struct bonding *bond, struct net_device *slave_dev)
  869. {
  870. struct vlan_entry *vlan;
  871. write_lock_bh(&bond->lock);
  872. if (list_empty(&bond->vlan_list)) {
  873. goto out;
  874. }
  875. if ((slave_dev->features & NETIF_F_HW_VLAN_RX) &&
  876. slave_dev->vlan_rx_register) {
  877. slave_dev->vlan_rx_register(slave_dev, bond->vlgrp);
  878. }
  879. if (!(slave_dev->features & NETIF_F_HW_VLAN_FILTER) ||
  880. !(slave_dev->vlan_rx_add_vid)) {
  881. goto out;
  882. }
  883. list_for_each_entry(vlan, &bond->vlan_list, vlan_list) {
  884. slave_dev->vlan_rx_add_vid(slave_dev, vlan->vlan_id);
  885. }
  886. out:
  887. write_unlock_bh(&bond->lock);
  888. }
  889. static void bond_del_vlans_from_slave(struct bonding *bond, struct net_device *slave_dev)
  890. {
  891. struct vlan_entry *vlan;
  892. struct net_device *vlan_dev;
  893. write_lock_bh(&bond->lock);
  894. if (list_empty(&bond->vlan_list)) {
  895. goto out;
  896. }
  897. if (!(slave_dev->features & NETIF_F_HW_VLAN_FILTER) ||
  898. !(slave_dev->vlan_rx_kill_vid)) {
  899. goto unreg;
  900. }
  901. list_for_each_entry(vlan, &bond->vlan_list, vlan_list) {
  902. /* Save and then restore vlan_dev in the grp array,
  903. * since the slave's driver might clear it.
  904. */
  905. vlan_dev = bond->vlgrp->vlan_devices[vlan->vlan_id];
  906. slave_dev->vlan_rx_kill_vid(slave_dev, vlan->vlan_id);
  907. bond->vlgrp->vlan_devices[vlan->vlan_id] = vlan_dev;
  908. }
  909. unreg:
  910. if ((slave_dev->features & NETIF_F_HW_VLAN_RX) &&
  911. slave_dev->vlan_rx_register) {
  912. slave_dev->vlan_rx_register(slave_dev, NULL);
  913. }
  914. out:
  915. write_unlock_bh(&bond->lock);
  916. }
  917. /*------------------------------- Link status -------------------------------*/
  918. /*
  919. * Get link speed and duplex from the slave's base driver
  920. * using ethtool. If for some reason the call fails or the
  921. * values are invalid, fake speed and duplex to 100/Full
  922. * and return error.
  923. */
  924. static int bond_update_speed_duplex(struct slave *slave)
  925. {
  926. struct net_device *slave_dev = slave->dev;
  927. static int (* ioctl)(struct net_device *, struct ifreq *, int);
  928. struct ifreq ifr;
  929. struct ethtool_cmd etool;
  930. /* Fake speed and duplex */
  931. slave->speed = SPEED_100;
  932. slave->duplex = DUPLEX_FULL;
  933. if (slave_dev->ethtool_ops) {
  934. u32 res;
  935. if (!slave_dev->ethtool_ops->get_settings) {
  936. return -1;
  937. }
  938. res = slave_dev->ethtool_ops->get_settings(slave_dev, &etool);
  939. if (res < 0) {
  940. return -1;
  941. }
  942. goto verify;
  943. }
  944. ioctl = slave_dev->do_ioctl;
  945. strncpy(ifr.ifr_name, slave_dev->name, IFNAMSIZ);
  946. etool.cmd = ETHTOOL_GSET;
  947. ifr.ifr_data = (char*)&etool;
  948. if (!ioctl || (IOCTL(slave_dev, &ifr, SIOCETHTOOL) < 0)) {
  949. return -1;
  950. }
  951. verify:
  952. switch (etool.speed) {
  953. case SPEED_10:
  954. case SPEED_100:
  955. case SPEED_1000:
  956. break;
  957. default:
  958. return -1;
  959. }
  960. switch (etool.duplex) {
  961. case DUPLEX_FULL:
  962. case DUPLEX_HALF:
  963. break;
  964. default:
  965. return -1;
  966. }
  967. slave->speed = etool.speed;
  968. slave->duplex = etool.duplex;
  969. return 0;
  970. }
  971. /*
  972. * if <dev> supports MII link status reporting, check its link status.
  973. *
  974. * We either do MII/ETHTOOL ioctls, or check netif_carrier_ok(),
  975. * depening upon the setting of the use_carrier parameter.
  976. *
  977. * Return either BMSR_LSTATUS, meaning that the link is up (or we
  978. * can't tell and just pretend it is), or 0, meaning that the link is
  979. * down.
  980. *
  981. * If reporting is non-zero, instead of faking link up, return -1 if
  982. * both ETHTOOL and MII ioctls fail (meaning the device does not
  983. * support them). If use_carrier is set, return whatever it says.
  984. * It'd be nice if there was a good way to tell if a driver supports
  985. * netif_carrier, but there really isn't.
  986. */
  987. static int bond_check_dev_link(struct bonding *bond, struct net_device *slave_dev, int reporting)
  988. {
  989. static int (* ioctl)(struct net_device *, struct ifreq *, int);
  990. struct ifreq ifr;
  991. struct mii_ioctl_data *mii;
  992. struct ethtool_value etool;
  993. if (bond->params.use_carrier) {
  994. return netif_carrier_ok(slave_dev) ? BMSR_LSTATUS : 0;
  995. }
  996. ioctl = slave_dev->do_ioctl;
  997. if (ioctl) {
  998. /* TODO: set pointer to correct ioctl on a per team member */
  999. /* bases to make this more efficient. that is, once */
  1000. /* we determine the correct ioctl, we will always */
  1001. /* call it and not the others for that team */
  1002. /* member. */
  1003. /*
  1004. * We cannot assume that SIOCGMIIPHY will also read a
  1005. * register; not all network drivers (e.g., e100)
  1006. * support that.
  1007. */
  1008. /* Yes, the mii is overlaid on the ifreq.ifr_ifru */
  1009. strncpy(ifr.ifr_name, slave_dev->name, IFNAMSIZ);
  1010. mii = if_mii(&ifr);
  1011. if (IOCTL(slave_dev, &ifr, SIOCGMIIPHY) == 0) {
  1012. mii->reg_num = MII_BMSR;
  1013. if (IOCTL(slave_dev, &ifr, SIOCGMIIREG) == 0) {
  1014. return (mii->val_out & BMSR_LSTATUS);
  1015. }
  1016. }
  1017. }
  1018. /* try SIOCETHTOOL ioctl, some drivers cache ETHTOOL_GLINK */
  1019. /* for a period of time so we attempt to get link status */
  1020. /* from it last if the above MII ioctls fail... */
  1021. if (slave_dev->ethtool_ops) {
  1022. if (slave_dev->ethtool_ops->get_link) {
  1023. u32 link;
  1024. link = slave_dev->ethtool_ops->get_link(slave_dev);
  1025. return link ? BMSR_LSTATUS : 0;
  1026. }
  1027. }
  1028. if (ioctl) {
  1029. strncpy(ifr.ifr_name, slave_dev->name, IFNAMSIZ);
  1030. etool.cmd = ETHTOOL_GLINK;
  1031. ifr.ifr_data = (char*)&etool;
  1032. if (IOCTL(slave_dev, &ifr, SIOCETHTOOL) == 0) {
  1033. if (etool.data == 1) {
  1034. return BMSR_LSTATUS;
  1035. } else {
  1036. dprintk("SIOCETHTOOL shows link down\n");
  1037. return 0;
  1038. }
  1039. }
  1040. }
  1041. /*
  1042. * If reporting, report that either there's no dev->do_ioctl,
  1043. * or both SIOCGMIIREG and SIOCETHTOOL failed (meaning that we
  1044. * cannot report link status). If not reporting, pretend
  1045. * we're ok.
  1046. */
  1047. return (reporting ? -1 : BMSR_LSTATUS);
  1048. }
  1049. /*----------------------------- Multicast list ------------------------------*/
  1050. /*
  1051. * Returns 0 if dmi1 and dmi2 are the same, non-0 otherwise
  1052. */
  1053. static inline int bond_is_dmi_same(struct dev_mc_list *dmi1, struct dev_mc_list *dmi2)
  1054. {
  1055. return memcmp(dmi1->dmi_addr, dmi2->dmi_addr, dmi1->dmi_addrlen) == 0 &&
  1056. dmi1->dmi_addrlen == dmi2->dmi_addrlen;
  1057. }
  1058. /*
  1059. * returns dmi entry if found, NULL otherwise
  1060. */
  1061. static struct dev_mc_list *bond_mc_list_find_dmi(struct dev_mc_list *dmi, struct dev_mc_list *mc_list)
  1062. {
  1063. struct dev_mc_list *idmi;
  1064. for (idmi = mc_list; idmi; idmi = idmi->next) {
  1065. if (bond_is_dmi_same(dmi, idmi)) {
  1066. return idmi;
  1067. }
  1068. }
  1069. return NULL;
  1070. }
  1071. /*
  1072. * Push the promiscuity flag down to appropriate slaves
  1073. */
  1074. static void bond_set_promiscuity(struct bonding *bond, int inc)
  1075. {
  1076. if (USES_PRIMARY(bond->params.mode)) {
  1077. /* write lock already acquired */
  1078. if (bond->curr_active_slave) {
  1079. dev_set_promiscuity(bond->curr_active_slave->dev, inc);
  1080. }
  1081. } else {
  1082. struct slave *slave;
  1083. int i;
  1084. bond_for_each_slave(bond, slave, i) {
  1085. dev_set_promiscuity(slave->dev, inc);
  1086. }
  1087. }
  1088. }
  1089. /*
  1090. * Push the allmulti flag down to all slaves
  1091. */
  1092. static void bond_set_allmulti(struct bonding *bond, int inc)
  1093. {
  1094. if (USES_PRIMARY(bond->params.mode)) {
  1095. /* write lock already acquired */
  1096. if (bond->curr_active_slave) {
  1097. dev_set_allmulti(bond->curr_active_slave->dev, inc);
  1098. }
  1099. } else {
  1100. struct slave *slave;
  1101. int i;
  1102. bond_for_each_slave(bond, slave, i) {
  1103. dev_set_allmulti(slave->dev, inc);
  1104. }
  1105. }
  1106. }
  1107. /*
  1108. * Add a Multicast address to slaves
  1109. * according to mode
  1110. */
  1111. static void bond_mc_add(struct bonding *bond, void *addr, int alen)
  1112. {
  1113. if (USES_PRIMARY(bond->params.mode)) {
  1114. /* write lock already acquired */
  1115. if (bond->curr_active_slave) {
  1116. dev_mc_add(bond->curr_active_slave->dev, addr, alen, 0);
  1117. }
  1118. } else {
  1119. struct slave *slave;
  1120. int i;
  1121. bond_for_each_slave(bond, slave, i) {
  1122. dev_mc_add(slave->dev, addr, alen, 0);
  1123. }
  1124. }
  1125. }
  1126. /*
  1127. * Remove a multicast address from slave
  1128. * according to mode
  1129. */
  1130. static void bond_mc_delete(struct bonding *bond, void *addr, int alen)
  1131. {
  1132. if (USES_PRIMARY(bond->params.mode)) {
  1133. /* write lock already acquired */
  1134. if (bond->curr_active_slave) {
  1135. dev_mc_delete(bond->curr_active_slave->dev, addr, alen, 0);
  1136. }
  1137. } else {
  1138. struct slave *slave;
  1139. int i;
  1140. bond_for_each_slave(bond, slave, i) {
  1141. dev_mc_delete(slave->dev, addr, alen, 0);
  1142. }
  1143. }
  1144. }
  1145. /*
  1146. * Totally destroys the mc_list in bond
  1147. */
  1148. static void bond_mc_list_destroy(struct bonding *bond)
  1149. {
  1150. struct dev_mc_list *dmi;
  1151. dmi = bond->mc_list;
  1152. while (dmi) {
  1153. bond->mc_list = dmi->next;
  1154. kfree(dmi);
  1155. dmi = bond->mc_list;
  1156. }
  1157. }
  1158. /*
  1159. * Copy all the Multicast addresses from src to the bonding device dst
  1160. */
  1161. static int bond_mc_list_copy(struct dev_mc_list *mc_list, struct bonding *bond, int gpf_flag)
  1162. {
  1163. struct dev_mc_list *dmi, *new_dmi;
  1164. for (dmi = mc_list; dmi; dmi = dmi->next) {
  1165. new_dmi = kmalloc(sizeof(struct dev_mc_list), gpf_flag);
  1166. if (!new_dmi) {
  1167. /* FIXME: Potential memory leak !!! */
  1168. return -ENOMEM;
  1169. }
  1170. new_dmi->next = bond->mc_list;
  1171. bond->mc_list = new_dmi;
  1172. new_dmi->dmi_addrlen = dmi->dmi_addrlen;
  1173. memcpy(new_dmi->dmi_addr, dmi->dmi_addr, dmi->dmi_addrlen);
  1174. new_dmi->dmi_users = dmi->dmi_users;
  1175. new_dmi->dmi_gusers = dmi->dmi_gusers;
  1176. }
  1177. return 0;
  1178. }
  1179. /*
  1180. * flush all members of flush->mc_list from device dev->mc_list
  1181. */
  1182. static void bond_mc_list_flush(struct net_device *bond_dev, struct net_device *slave_dev)
  1183. {
  1184. struct bonding *bond = bond_dev->priv;
  1185. struct dev_mc_list *dmi;
  1186. for (dmi = bond_dev->mc_list; dmi; dmi = dmi->next) {
  1187. dev_mc_delete(slave_dev, dmi->dmi_addr, dmi->dmi_addrlen, 0);
  1188. }
  1189. if (bond->params.mode == BOND_MODE_8023AD) {
  1190. /* del lacpdu mc addr from mc list */
  1191. u8 lacpdu_multicast[ETH_ALEN] = MULTICAST_LACPDU_ADDR;
  1192. dev_mc_delete(slave_dev, lacpdu_multicast, ETH_ALEN, 0);
  1193. }
  1194. }
  1195. /*--------------------------- Active slave change ---------------------------*/
  1196. /*
  1197. * Update the mc list and multicast-related flags for the new and
  1198. * old active slaves (if any) according to the multicast mode, and
  1199. * promiscuous flags unconditionally.
  1200. */
  1201. static void bond_mc_swap(struct bonding *bond, struct slave *new_active, struct slave *old_active)
  1202. {
  1203. struct dev_mc_list *dmi;
  1204. if (!USES_PRIMARY(bond->params.mode)) {
  1205. /* nothing to do - mc list is already up-to-date on
  1206. * all slaves
  1207. */
  1208. return;
  1209. }
  1210. if (old_active) {
  1211. if (bond->dev->flags & IFF_PROMISC) {
  1212. dev_set_promiscuity(old_active->dev, -1);
  1213. }
  1214. if (bond->dev->flags & IFF_ALLMULTI) {
  1215. dev_set_allmulti(old_active->dev, -1);
  1216. }
  1217. for (dmi = bond->dev->mc_list; dmi; dmi = dmi->next) {
  1218. dev_mc_delete(old_active->dev, dmi->dmi_addr, dmi->dmi_addrlen, 0);
  1219. }
  1220. }
  1221. if (new_active) {
  1222. if (bond->dev->flags & IFF_PROMISC) {
  1223. dev_set_promiscuity(new_active->dev, 1);
  1224. }
  1225. if (bond->dev->flags & IFF_ALLMULTI) {
  1226. dev_set_allmulti(new_active->dev, 1);
  1227. }
  1228. for (dmi = bond->dev->mc_list; dmi; dmi = dmi->next) {
  1229. dev_mc_add(new_active->dev, dmi->dmi_addr, dmi->dmi_addrlen, 0);
  1230. }
  1231. }
  1232. }
  1233. /**
  1234. * find_best_interface - select the best available slave to be the active one
  1235. * @bond: our bonding struct
  1236. *
  1237. * Warning: Caller must hold curr_slave_lock for writing.
  1238. */
  1239. static struct slave *bond_find_best_slave(struct bonding *bond)
  1240. {
  1241. struct slave *new_active, *old_active;
  1242. struct slave *bestslave = NULL;
  1243. int mintime = bond->params.updelay;
  1244. int i;
  1245. new_active = old_active = bond->curr_active_slave;
  1246. if (!new_active) { /* there were no active slaves left */
  1247. if (bond->slave_cnt > 0) { /* found one slave */
  1248. new_active = bond->first_slave;
  1249. } else {
  1250. return NULL; /* still no slave, return NULL */
  1251. }
  1252. }
  1253. /* first try the primary link; if arping, a link must tx/rx traffic
  1254. * before it can be considered the curr_active_slave - also, we would skip
  1255. * slaves between the curr_active_slave and primary_slave that may be up
  1256. * and able to arp
  1257. */
  1258. if ((bond->primary_slave) &&
  1259. (!bond->params.arp_interval) &&
  1260. (IS_UP(bond->primary_slave->dev))) {
  1261. new_active = bond->primary_slave;
  1262. }
  1263. /* remember where to stop iterating over the slaves */
  1264. old_active = new_active;
  1265. bond_for_each_slave_from(bond, new_active, i, old_active) {
  1266. if (IS_UP(new_active->dev)) {
  1267. if (new_active->link == BOND_LINK_UP) {
  1268. return new_active;
  1269. } else if (new_active->link == BOND_LINK_BACK) {
  1270. /* link up, but waiting for stabilization */
  1271. if (new_active->delay < mintime) {
  1272. mintime = new_active->delay;
  1273. bestslave = new_active;
  1274. }
  1275. }
  1276. }
  1277. }
  1278. return bestslave;
  1279. }
  1280. /**
  1281. * change_active_interface - change the active slave into the specified one
  1282. * @bond: our bonding struct
  1283. * @new: the new slave to make the active one
  1284. *
  1285. * Set the new slave to the bond's settings and unset them on the old
  1286. * curr_active_slave.
  1287. * Setting include flags, mc-list, promiscuity, allmulti, etc.
  1288. *
  1289. * If @new's link state is %BOND_LINK_BACK we'll set it to %BOND_LINK_UP,
  1290. * because it is apparently the best available slave we have, even though its
  1291. * updelay hasn't timed out yet.
  1292. *
  1293. * Warning: Caller must hold curr_slave_lock for writing.
  1294. */
  1295. static void bond_change_active_slave(struct bonding *bond, struct slave *new_active)
  1296. {
  1297. struct slave *old_active = bond->curr_active_slave;
  1298. if (old_active == new_active) {
  1299. return;
  1300. }
  1301. if (new_active) {
  1302. if (new_active->link == BOND_LINK_BACK) {
  1303. if (USES_PRIMARY(bond->params.mode)) {
  1304. printk(KERN_INFO DRV_NAME
  1305. ": %s: making interface %s the new "
  1306. "active one %d ms earlier.\n",
  1307. bond->dev->name, new_active->dev->name,
  1308. (bond->params.updelay - new_active->delay) * bond->params.miimon);
  1309. }
  1310. new_active->delay = 0;
  1311. new_active->link = BOND_LINK_UP;
  1312. new_active->jiffies = jiffies;
  1313. if (bond->params.mode == BOND_MODE_8023AD) {
  1314. bond_3ad_handle_link_change(new_active, BOND_LINK_UP);
  1315. }
  1316. if ((bond->params.mode == BOND_MODE_TLB) ||
  1317. (bond->params.mode == BOND_MODE_ALB)) {
  1318. bond_alb_handle_link_change(bond, new_active, BOND_LINK_UP);
  1319. }
  1320. } else {
  1321. if (USES_PRIMARY(bond->params.mode)) {
  1322. printk(KERN_INFO DRV_NAME
  1323. ": %s: making interface %s the new "
  1324. "active one.\n",
  1325. bond->dev->name, new_active->dev->name);
  1326. }
  1327. }
  1328. }
  1329. if (bond->params.mode == BOND_MODE_ACTIVEBACKUP) {
  1330. if (old_active) {
  1331. bond_set_slave_inactive_flags(old_active);
  1332. }
  1333. if (new_active) {
  1334. bond_set_slave_active_flags(new_active);
  1335. }
  1336. }
  1337. if (USES_PRIMARY(bond->params.mode)) {
  1338. bond_mc_swap(bond, new_active, old_active);
  1339. }
  1340. if ((bond->params.mode == BOND_MODE_TLB) ||
  1341. (bond->params.mode == BOND_MODE_ALB)) {
  1342. bond_alb_handle_active_change(bond, new_active);
  1343. } else {
  1344. bond->curr_active_slave = new_active;
  1345. }
  1346. }
  1347. /**
  1348. * bond_select_active_slave - select a new active slave, if needed
  1349. * @bond: our bonding struct
  1350. *
  1351. * This functions shoud be called when one of the following occurs:
  1352. * - The old curr_active_slave has been released or lost its link.
  1353. * - The primary_slave has got its link back.
  1354. * - A slave has got its link back and there's no old curr_active_slave.
  1355. *
  1356. * Warning: Caller must hold curr_slave_lock for writing.
  1357. */
  1358. static void bond_select_active_slave(struct bonding *bond)
  1359. {
  1360. struct slave *best_slave;
  1361. best_slave = bond_find_best_slave(bond);
  1362. if (best_slave != bond->curr_active_slave) {
  1363. bond_change_active_slave(bond, best_slave);
  1364. }
  1365. }
  1366. /*--------------------------- slave list handling ---------------------------*/
  1367. /*
  1368. * This function attaches the slave to the end of list.
  1369. *
  1370. * bond->lock held for writing by caller.
  1371. */
  1372. static void bond_attach_slave(struct bonding *bond, struct slave *new_slave)
  1373. {
  1374. if (bond->first_slave == NULL) { /* attaching the first slave */
  1375. new_slave->next = new_slave;
  1376. new_slave->prev = new_slave;
  1377. bond->first_slave = new_slave;
  1378. } else {
  1379. new_slave->next = bond->first_slave;
  1380. new_slave->prev = bond->first_slave->prev;
  1381. new_slave->next->prev = new_slave;
  1382. new_slave->prev->next = new_slave;
  1383. }
  1384. bond->slave_cnt++;
  1385. }
  1386. /*
  1387. * This function detaches the slave from the list.
  1388. * WARNING: no check is made to verify if the slave effectively
  1389. * belongs to <bond>.
  1390. * Nothing is freed on return, structures are just unchained.
  1391. * If any slave pointer in bond was pointing to <slave>,
  1392. * it should be changed by the calling function.
  1393. *
  1394. * bond->lock held for writing by caller.
  1395. */
  1396. static void bond_detach_slave(struct bonding *bond, struct slave *slave)
  1397. {
  1398. if (slave->next) {
  1399. slave->next->prev = slave->prev;
  1400. }
  1401. if (slave->prev) {
  1402. slave->prev->next = slave->next;
  1403. }
  1404. if (bond->first_slave == slave) { /* slave is the first slave */
  1405. if (bond->slave_cnt > 1) { /* there are more slave */
  1406. bond->first_slave = slave->next;
  1407. } else {
  1408. bond->first_slave = NULL; /* slave was the last one */
  1409. }
  1410. }
  1411. slave->next = NULL;
  1412. slave->prev = NULL;
  1413. bond->slave_cnt--;
  1414. }
  1415. /*---------------------------------- IOCTL ----------------------------------*/
  1416. static int bond_sethwaddr(struct net_device *bond_dev, struct net_device *slave_dev)
  1417. {
  1418. dprintk("bond_dev=%p\n", bond_dev);
  1419. dprintk("slave_dev=%p\n", slave_dev);
  1420. dprintk("slave_dev->addr_len=%d\n", slave_dev->addr_len);
  1421. memcpy(bond_dev->dev_addr, slave_dev->dev_addr, slave_dev->addr_len);
  1422. return 0;
  1423. }
  1424. /* enslave device <slave> to bond device <master> */
  1425. static int bond_enslave(struct net_device *bond_dev, struct net_device *slave_dev)
  1426. {
  1427. struct bonding *bond = bond_dev->priv;
  1428. struct slave *new_slave = NULL;
  1429. struct dev_mc_list *dmi;
  1430. struct sockaddr addr;
  1431. int link_reporting;
  1432. int old_features = bond_dev->features;
  1433. int res = 0;
  1434. if (slave_dev->do_ioctl == NULL) {
  1435. printk(KERN_WARNING DRV_NAME
  1436. ": Warning : no link monitoring support for %s\n",
  1437. slave_dev->name);
  1438. }
  1439. /* bond must be initialized by bond_open() before enslaving */
  1440. if (!(bond_dev->flags & IFF_UP)) {
  1441. dprintk("Error, master_dev is not up\n");
  1442. return -EPERM;
  1443. }
  1444. /* already enslaved */
  1445. if (slave_dev->flags & IFF_SLAVE) {
  1446. dprintk("Error, Device was already enslaved\n");
  1447. return -EBUSY;
  1448. }
  1449. /* vlan challenged mutual exclusion */
  1450. /* no need to lock since we're protected by rtnl_lock */
  1451. if (slave_dev->features & NETIF_F_VLAN_CHALLENGED) {
  1452. dprintk("%s: NETIF_F_VLAN_CHALLENGED\n", slave_dev->name);
  1453. if (!list_empty(&bond->vlan_list)) {
  1454. printk(KERN_ERR DRV_NAME
  1455. ": Error: cannot enslave VLAN "
  1456. "challenged slave %s on VLAN enabled "
  1457. "bond %s\n", slave_dev->name,
  1458. bond_dev->name);
  1459. return -EPERM;
  1460. } else {
  1461. printk(KERN_WARNING DRV_NAME
  1462. ": Warning: enslaved VLAN challenged "
  1463. "slave %s. Adding VLANs will be blocked as "
  1464. "long as %s is part of bond %s\n",
  1465. slave_dev->name, slave_dev->name,
  1466. bond_dev->name);
  1467. bond_dev->features |= NETIF_F_VLAN_CHALLENGED;
  1468. }
  1469. } else {
  1470. dprintk("%s: ! NETIF_F_VLAN_CHALLENGED\n", slave_dev->name);
  1471. if (bond->slave_cnt == 0) {
  1472. /* First slave, and it is not VLAN challenged,
  1473. * so remove the block of adding VLANs over the bond.
  1474. */
  1475. bond_dev->features &= ~NETIF_F_VLAN_CHALLENGED;
  1476. }
  1477. }
  1478. if (app_abi_ver >= 1) {
  1479. /* The application is using an ABI, which requires the
  1480. * slave interface to be closed.
  1481. */
  1482. if ((slave_dev->flags & IFF_UP)) {
  1483. printk(KERN_ERR DRV_NAME
  1484. ": Error: %s is up\n",
  1485. slave_dev->name);
  1486. res = -EPERM;
  1487. goto err_undo_flags;
  1488. }
  1489. if (slave_dev->set_mac_address == NULL) {
  1490. printk(KERN_ERR DRV_NAME
  1491. ": Error: The slave device you specified does "
  1492. "not support setting the MAC address.\n");
  1493. printk(KERN_ERR
  1494. "Your kernel likely does not support slave "
  1495. "devices.\n");
  1496. res = -EOPNOTSUPP;
  1497. goto err_undo_flags;
  1498. }
  1499. } else {
  1500. /* The application is not using an ABI, which requires the
  1501. * slave interface to be open.
  1502. */
  1503. if (!(slave_dev->flags & IFF_UP)) {
  1504. printk(KERN_ERR DRV_NAME
  1505. ": Error: %s is not running\n",
  1506. slave_dev->name);
  1507. res = -EINVAL;
  1508. goto err_undo_flags;
  1509. }
  1510. if ((bond->params.mode == BOND_MODE_8023AD) ||
  1511. (bond->params.mode == BOND_MODE_TLB) ||
  1512. (bond->params.mode == BOND_MODE_ALB)) {
  1513. printk(KERN_ERR DRV_NAME
  1514. ": Error: to use %s mode, you must upgrade "
  1515. "ifenslave.\n",
  1516. bond_mode_name(bond->params.mode));
  1517. res = -EOPNOTSUPP;
  1518. goto err_undo_flags;
  1519. }
  1520. }
  1521. new_slave = kmalloc(sizeof(struct slave), GFP_KERNEL);
  1522. if (!new_slave) {
  1523. res = -ENOMEM;
  1524. goto err_undo_flags;
  1525. }
  1526. memset(new_slave, 0, sizeof(struct slave));
  1527. /* save slave's original flags before calling
  1528. * netdev_set_master and dev_open
  1529. */
  1530. new_slave->original_flags = slave_dev->flags;
  1531. if (app_abi_ver >= 1) {
  1532. /* save slave's original ("permanent") mac address for
  1533. * modes that needs it, and for restoring it upon release,
  1534. * and then set it to the master's address
  1535. */
  1536. memcpy(new_slave->perm_hwaddr, slave_dev->dev_addr, ETH_ALEN);
  1537. /* set slave to master's mac address
  1538. * The application already set the master's
  1539. * mac address to that of the first slave
  1540. */
  1541. memcpy(addr.sa_data, bond_dev->dev_addr, bond_dev->addr_len);
  1542. addr.sa_family = slave_dev->type;
  1543. res = dev_set_mac_address(slave_dev, &addr);
  1544. if (res) {
  1545. dprintk("Error %d calling set_mac_address\n", res);
  1546. goto err_free;
  1547. }
  1548. /* open the slave since the application closed it */
  1549. res = dev_open(slave_dev);
  1550. if (res) {
  1551. dprintk("Openning slave %s failed\n", slave_dev->name);
  1552. goto err_restore_mac;
  1553. }
  1554. }
  1555. res = netdev_set_master(slave_dev, bond_dev);
  1556. if (res) {
  1557. dprintk("Error %d calling netdev_set_master\n", res);
  1558. if (app_abi_ver < 1) {
  1559. goto err_free;
  1560. } else {
  1561. goto err_close;
  1562. }
  1563. }
  1564. new_slave->dev = slave_dev;
  1565. if ((bond->params.mode == BOND_MODE_TLB) ||
  1566. (bond->params.mode == BOND_MODE_ALB)) {
  1567. /* bond_alb_init_slave() must be called before all other stages since
  1568. * it might fail and we do not want to have to undo everything
  1569. */
  1570. res = bond_alb_init_slave(bond, new_slave);
  1571. if (res) {
  1572. goto err_unset_master;
  1573. }
  1574. }
  1575. /* If the mode USES_PRIMARY, then the new slave gets the
  1576. * master's promisc (and mc) settings only if it becomes the
  1577. * curr_active_slave, and that is taken care of later when calling
  1578. * bond_change_active()
  1579. */
  1580. if (!USES_PRIMARY(bond->params.mode)) {
  1581. /* set promiscuity level to new slave */
  1582. if (bond_dev->flags & IFF_PROMISC) {
  1583. dev_set_promiscuity(slave_dev, 1);
  1584. }
  1585. /* set allmulti level to new slave */
  1586. if (bond_dev->flags & IFF_ALLMULTI) {
  1587. dev_set_allmulti(slave_dev, 1);
  1588. }
  1589. /* upload master's mc_list to new slave */
  1590. for (dmi = bond_dev->mc_list; dmi; dmi = dmi->next) {
  1591. dev_mc_add (slave_dev, dmi->dmi_addr, dmi->dmi_addrlen, 0);
  1592. }
  1593. }
  1594. if (bond->params.mode == BOND_MODE_8023AD) {
  1595. /* add lacpdu mc addr to mc list */
  1596. u8 lacpdu_multicast[ETH_ALEN] = MULTICAST_LACPDU_ADDR;
  1597. dev_mc_add(slave_dev, lacpdu_multicast, ETH_ALEN, 0);
  1598. }
  1599. bond_add_vlans_on_slave(bond, slave_dev);
  1600. write_lock_bh(&bond->lock);
  1601. bond_attach_slave(bond, new_slave);
  1602. new_slave->delay = 0;
  1603. new_slave->link_failure_count = 0;
  1604. if (bond->params.miimon && !bond->params.use_carrier) {
  1605. link_reporting = bond_check_dev_link(bond, slave_dev, 1);
  1606. if ((link_reporting == -1) && !bond->params.arp_interval) {
  1607. /*
  1608. * miimon is set but a bonded network driver
  1609. * does not support ETHTOOL/MII and
  1610. * arp_interval is not set. Note: if
  1611. * use_carrier is enabled, we will never go
  1612. * here (because netif_carrier is always
  1613. * supported); thus, we don't need to change
  1614. * the messages for netif_carrier.
  1615. */
  1616. printk(KERN_WARNING DRV_NAME
  1617. ": Warning: MII and ETHTOOL support not "
  1618. "available for interface %s, and "
  1619. "arp_interval/arp_ip_target module parameters "
  1620. "not specified, thus bonding will not detect "
  1621. "link failures! see bonding.txt for details.\n",
  1622. slave_dev->name);
  1623. } else if (link_reporting == -1) {
  1624. /* unable get link status using mii/ethtool */
  1625. printk(KERN_WARNING DRV_NAME
  1626. ": Warning: can't get link status from "
  1627. "interface %s; the network driver associated "
  1628. "with this interface does not support MII or "
  1629. "ETHTOOL link status reporting, thus miimon "
  1630. "has no effect on this interface.\n",
  1631. slave_dev->name);
  1632. }
  1633. }
  1634. /* check for initial state */
  1635. if (!bond->params.miimon ||
  1636. (bond_check_dev_link(bond, slave_dev, 0) == BMSR_LSTATUS)) {
  1637. if (bond->params.updelay) {
  1638. dprintk("Initial state of slave_dev is "
  1639. "BOND_LINK_BACK\n");
  1640. new_slave->link = BOND_LINK_BACK;
  1641. new_slave->delay = bond->params.updelay;
  1642. } else {
  1643. dprintk("Initial state of slave_dev is "
  1644. "BOND_LINK_UP\n");
  1645. new_slave->link = BOND_LINK_UP;
  1646. }
  1647. new_slave->jiffies = jiffies;
  1648. } else {
  1649. dprintk("Initial state of slave_dev is "
  1650. "BOND_LINK_DOWN\n");
  1651. new_slave->link = BOND_LINK_DOWN;
  1652. }
  1653. if (bond_update_speed_duplex(new_slave) &&
  1654. (new_slave->link != BOND_LINK_DOWN)) {
  1655. printk(KERN_WARNING DRV_NAME
  1656. ": Warning: failed to get speed and duplex from %s, "
  1657. "assumed to be 100Mb/sec and Full.\n",
  1658. new_slave->dev->name);
  1659. if (bond->params.mode == BOND_MODE_8023AD) {
  1660. printk(KERN_WARNING
  1661. "Operation of 802.3ad mode requires ETHTOOL "
  1662. "support in base driver for proper aggregator "
  1663. "selection.\n");
  1664. }
  1665. }
  1666. if (USES_PRIMARY(bond->params.mode) && bond->params.primary[0]) {
  1667. /* if there is a primary slave, remember it */
  1668. if (strcmp(bond->params.primary, new_slave->dev->name) == 0) {
  1669. bond->primary_slave = new_slave;
  1670. }
  1671. }
  1672. switch (bond->params.mode) {
  1673. case BOND_MODE_ACTIVEBACKUP:
  1674. /* if we're in active-backup mode, we need one and only one active
  1675. * interface. The backup interfaces will have their NOARP flag set
  1676. * because we need them to be completely deaf and not to respond to
  1677. * any ARP request on the network to avoid fooling a switch. Thus,
  1678. * since we guarantee that curr_active_slave always point to the last
  1679. * usable interface, we just have to verify this interface's flag.
  1680. */
  1681. if (((!bond->curr_active_slave) ||
  1682. (bond->curr_active_slave->dev->flags & IFF_NOARP)) &&
  1683. (new_slave->link != BOND_LINK_DOWN)) {
  1684. dprintk("This is the first active slave\n");
  1685. /* first slave or no active slave yet, and this link
  1686. is OK, so make this interface the active one */
  1687. bond_change_active_slave(bond, new_slave);
  1688. } else {
  1689. dprintk("This is just a backup slave\n");
  1690. bond_set_slave_inactive_flags(new_slave);
  1691. }
  1692. break;
  1693. case BOND_MODE_8023AD:
  1694. /* in 802.3ad mode, the internal mechanism
  1695. * will activate the slaves in the selected
  1696. * aggregator
  1697. */
  1698. bond_set_slave_inactive_flags(new_slave);
  1699. /* if this is the first slave */
  1700. if (bond->slave_cnt == 1) {
  1701. SLAVE_AD_INFO(new_slave).id = 1;
  1702. /* Initialize AD with the number of times that the AD timer is called in 1 second
  1703. * can be called only after the mac address of the bond is set
  1704. */
  1705. bond_3ad_initialize(bond, 1000/AD_TIMER_INTERVAL,
  1706. bond->params.lacp_fast);
  1707. } else {
  1708. SLAVE_AD_INFO(new_slave).id =
  1709. SLAVE_AD_INFO(new_slave->prev).id + 1;
  1710. }
  1711. bond_3ad_bind_slave(new_slave);
  1712. break;
  1713. case BOND_MODE_TLB:
  1714. case BOND_MODE_ALB:
  1715. new_slave->state = BOND_STATE_ACTIVE;
  1716. if ((!bond->curr_active_slave) &&
  1717. (new_slave->link != BOND_LINK_DOWN)) {
  1718. /* first slave or no active slave yet, and this link
  1719. * is OK, so make this interface the active one
  1720. */
  1721. bond_change_active_slave(bond, new_slave);
  1722. }
  1723. break;
  1724. default:
  1725. dprintk("This slave is always active in trunk mode\n");
  1726. /* always active in trunk mode */
  1727. new_slave->state = BOND_STATE_ACTIVE;
  1728. /* In trunking mode there is little meaning to curr_active_slave
  1729. * anyway (it holds no special properties of the bond device),
  1730. * so we can change it without calling change_active_interface()
  1731. */
  1732. if (!bond->curr_active_slave) {
  1733. bond->curr_active_slave = new_slave;
  1734. }
  1735. break;
  1736. } /* switch(bond_mode) */
  1737. write_unlock_bh(&bond->lock);
  1738. if (app_abi_ver < 1) {
  1739. /*
  1740. * !!! This is to support old versions of ifenslave.
  1741. * We can remove this in 2.5 because our ifenslave takes
  1742. * care of this for us.
  1743. * We check to see if the master has a mac address yet.
  1744. * If not, we'll give it the mac address of our slave device.
  1745. */
  1746. int ndx = 0;
  1747. for (ndx = 0; ndx < bond_dev->addr_len; ndx++) {
  1748. dprintk("Checking ndx=%d of bond_dev->dev_addr\n",
  1749. ndx);
  1750. if (bond_dev->dev_addr[ndx] != 0) {
  1751. dprintk("Found non-zero byte at ndx=%d\n",
  1752. ndx);
  1753. break;
  1754. }
  1755. }
  1756. if (ndx == bond_dev->addr_len) {
  1757. /*
  1758. * We got all the way through the address and it was
  1759. * all 0's.
  1760. */
  1761. dprintk("%s doesn't have a MAC address yet. \n",
  1762. bond_dev->name);
  1763. dprintk("Going to give assign it from %s.\n",
  1764. slave_dev->name);
  1765. bond_sethwaddr(bond_dev, slave_dev);
  1766. }
  1767. }
  1768. printk(KERN_INFO DRV_NAME
  1769. ": %s: enslaving %s as a%s interface with a%s link.\n",
  1770. bond_dev->name, slave_dev->name,
  1771. new_slave->state == BOND_STATE_ACTIVE ? "n active" : " backup",
  1772. new_slave->link != BOND_LINK_DOWN ? "n up" : " down");
  1773. /* enslave is successful */
  1774. return 0;
  1775. /* Undo stages on error */
  1776. err_unset_master:
  1777. netdev_set_master(slave_dev, NULL);
  1778. err_close:
  1779. dev_close(slave_dev);
  1780. err_restore_mac:
  1781. memcpy(addr.sa_data, new_slave->perm_hwaddr, ETH_ALEN);
  1782. addr.sa_family = slave_dev->type;
  1783. dev_set_mac_address(slave_dev, &addr);
  1784. err_free:
  1785. kfree(new_slave);
  1786. err_undo_flags:
  1787. bond_dev->features = old_features;
  1788. return res;
  1789. }
  1790. /*
  1791. * Try to release the slave device <slave> from the bond device <master>
  1792. * It is legal to access curr_active_slave without a lock because all the function
  1793. * is write-locked.
  1794. *
  1795. * The rules for slave state should be:
  1796. * for Active/Backup:
  1797. * Active stays on all backups go down
  1798. * for Bonded connections:
  1799. * The first up interface should be left on and all others downed.
  1800. */
  1801. static int bond_release(struct net_device *bond_dev, struct net_device *slave_dev)
  1802. {
  1803. struct bonding *bond = bond_dev->priv;
  1804. struct slave *slave, *oldcurrent;
  1805. struct sockaddr addr;
  1806. int mac_addr_differ;
  1807. /* slave is not a slave or master is not master of this slave */
  1808. if (!(slave_dev->flags & IFF_SLAVE) ||
  1809. (slave_dev->master != bond_dev)) {
  1810. printk(KERN_ERR DRV_NAME
  1811. ": Error: %s: cannot release %s.\n",
  1812. bond_dev->name, slave_dev->name);
  1813. return -EINVAL;
  1814. }
  1815. write_lock_bh(&bond->lock);
  1816. slave = bond_get_slave_by_dev(bond, slave_dev);
  1817. if (!slave) {
  1818. /* not a slave of this bond */
  1819. printk(KERN_INFO DRV_NAME
  1820. ": %s: %s not enslaved\n",
  1821. bond_dev->name, slave_dev->name);
  1822. return -EINVAL;
  1823. }
  1824. mac_addr_differ = memcmp(bond_dev->dev_addr,
  1825. slave->perm_hwaddr,
  1826. ETH_ALEN);
  1827. if (!mac_addr_differ && (bond->slave_cnt > 1)) {
  1828. printk(KERN_WARNING DRV_NAME
  1829. ": Warning: the permanent HWaddr of %s "
  1830. "- %02X:%02X:%02X:%02X:%02X:%02X - is "
  1831. "still in use by %s. Set the HWaddr of "
  1832. "%s to a different address to avoid "
  1833. "conflicts.\n",
  1834. slave_dev->name,
  1835. slave->perm_hwaddr[0],
  1836. slave->perm_hwaddr[1],
  1837. slave->perm_hwaddr[2],
  1838. slave->perm_hwaddr[3],
  1839. slave->perm_hwaddr[4],
  1840. slave->perm_hwaddr[5],
  1841. bond_dev->name,
  1842. slave_dev->name);
  1843. }
  1844. /* Inform AD package of unbinding of slave. */
  1845. if (bond->params.mode == BOND_MODE_8023AD) {
  1846. /* must be called before the slave is
  1847. * detached from the list
  1848. */
  1849. bond_3ad_unbind_slave(slave);
  1850. }
  1851. printk(KERN_INFO DRV_NAME
  1852. ": %s: releasing %s interface %s\n",
  1853. bond_dev->name,
  1854. (slave->state == BOND_STATE_ACTIVE)
  1855. ? "active" : "backup",
  1856. slave_dev->name);
  1857. oldcurrent = bond->curr_active_slave;
  1858. bond->current_arp_slave = NULL;
  1859. /* release the slave from its bond */
  1860. bond_detach_slave(bond, slave);
  1861. if (bond->primary_slave == slave) {
  1862. bond->primary_slave = NULL;
  1863. }
  1864. if (oldcurrent == slave) {
  1865. bond_change_active_slave(bond, NULL);
  1866. }
  1867. if ((bond->params.mode == BOND_MODE_TLB) ||
  1868. (bond->params.mode == BOND_MODE_ALB)) {
  1869. /* Must be called only after the slave has been
  1870. * detached from the list and the curr_active_slave
  1871. * has been cleared (if our_slave == old_current),
  1872. * but before a new active slave is selected.
  1873. */
  1874. bond_alb_deinit_slave(bond, slave);
  1875. }
  1876. if (oldcurrent == slave) {
  1877. bond_select_active_slave(bond);
  1878. if (!bond->curr_active_slave) {
  1879. printk(KERN_INFO DRV_NAME
  1880. ": %s: now running without any active "
  1881. "interface !\n",
  1882. bond_dev->name);
  1883. }
  1884. }
  1885. if (bond->slave_cnt == 0) {
  1886. /* if the last slave was removed, zero the mac address
  1887. * of the master so it will be set by the application
  1888. * to the mac address of the first slave
  1889. */
  1890. memset(bond_dev->dev_addr, 0, bond_dev->addr_len);
  1891. if (list_empty(&bond->vlan_list)) {
  1892. bond_dev->features |= NETIF_F_VLAN_CHALLENGED;
  1893. } else {
  1894. printk(KERN_WARNING DRV_NAME
  1895. ": Warning: clearing HW address of %s while it "
  1896. "still has VLANs.\n",
  1897. bond_dev->name);
  1898. printk(KERN_WARNING DRV_NAME
  1899. ": When re-adding slaves, make sure the bond's "
  1900. "HW address matches its VLANs'.\n");
  1901. }
  1902. } else if ((bond_dev->features & NETIF_F_VLAN_CHALLENGED) &&
  1903. !bond_has_challenged_slaves(bond)) {
  1904. printk(KERN_INFO DRV_NAME
  1905. ": last VLAN challenged slave %s "
  1906. "left bond %s. VLAN blocking is removed\n",
  1907. slave_dev->name, bond_dev->name);
  1908. bond_dev->features &= ~NETIF_F_VLAN_CHALLENGED;
  1909. }
  1910. write_unlock_bh(&bond->lock);
  1911. bond_del_vlans_from_slave(bond, slave_dev);
  1912. /* If the mode USES_PRIMARY, then we should only remove its
  1913. * promisc and mc settings if it was the curr_active_slave, but that was
  1914. * already taken care of above when we detached the slave
  1915. */
  1916. if (!USES_PRIMARY(bond->params.mode)) {
  1917. /* unset promiscuity level from slave */
  1918. if (bond_dev->flags & IFF_PROMISC) {
  1919. dev_set_promiscuity(slave_dev, -1);
  1920. }
  1921. /* unset allmulti level from slave */
  1922. if (bond_dev->flags & IFF_ALLMULTI) {
  1923. dev_set_allmulti(slave_dev, -1);
  1924. }
  1925. /* flush master's mc_list from slave */
  1926. bond_mc_list_flush(bond_dev, slave_dev);
  1927. }
  1928. netdev_set_master(slave_dev, NULL);
  1929. /* close slave before restoring its mac address */
  1930. dev_close(slave_dev);
  1931. if (app_abi_ver >= 1) {
  1932. /* restore original ("permanent") mac address */
  1933. memcpy(addr.sa_data, slave->perm_hwaddr, ETH_ALEN);
  1934. addr.sa_family = slave_dev->type;
  1935. dev_set_mac_address(slave_dev, &addr);
  1936. }
  1937. /* restore the original state of the
  1938. * IFF_NOARP flag that might have been
  1939. * set by bond_set_slave_inactive_flags()
  1940. */
  1941. if ((slave->original_flags & IFF_NOARP) == 0) {
  1942. slave_dev->flags &= ~IFF_NOARP;
  1943. }
  1944. kfree(slave);
  1945. return 0; /* deletion OK */
  1946. }
  1947. /*
  1948. * This function releases all slaves.
  1949. */
  1950. static int bond_release_all(struct net_device *bond_dev)
  1951. {
  1952. struct bonding *bond = bond_dev->priv;
  1953. struct slave *slave;
  1954. struct net_device *slave_dev;
  1955. struct sockaddr addr;
  1956. write_lock_bh(&bond->lock);
  1957. if (bond->slave_cnt == 0) {
  1958. goto out;
  1959. }
  1960. bond->current_arp_slave = NULL;
  1961. bond->primary_slave = NULL;
  1962. bond_change_active_slave(bond, NULL);
  1963. while ((slave = bond->first_slave) != NULL) {
  1964. /* Inform AD package of unbinding of slave
  1965. * before slave is detached from the list.
  1966. */
  1967. if (bond->params.mode == BOND_MODE_8023AD) {
  1968. bond_3ad_unbind_slave(slave);
  1969. }
  1970. slave_dev = slave->dev;
  1971. bond_detach_slave(bond, slave);
  1972. if ((bond->params.mode == BOND_MODE_TLB) ||
  1973. (bond->params.mode == BOND_MODE_ALB)) {
  1974. /* must be called only after the slave
  1975. * has been detached from the list
  1976. */
  1977. bond_alb_deinit_slave(bond, slave);
  1978. }
  1979. /* now that the slave is detached, unlock and perform
  1980. * all the undo steps that should not be called from
  1981. * within a lock.
  1982. */
  1983. write_unlock_bh(&bond->lock);
  1984. bond_del_vlans_from_slave(bond, slave_dev);
  1985. /* If the mode USES_PRIMARY, then we should only remove its
  1986. * promisc and mc settings if it was the curr_active_slave, but that was
  1987. * already taken care of above when we detached the slave
  1988. */
  1989. if (!USES_PRIMARY(bond->params.mode)) {
  1990. /* unset promiscuity level from slave */
  1991. if (bond_dev->flags & IFF_PROMISC) {
  1992. dev_set_promiscuity(slave_dev, -1);
  1993. }
  1994. /* unset allmulti level from slave */
  1995. if (bond_dev->flags & IFF_ALLMULTI) {
  1996. dev_set_allmulti(slave_dev, -1);
  1997. }
  1998. /* flush master's mc_list from slave */
  1999. bond_mc_list_flush(bond_dev, slave_dev);
  2000. }
  2001. netdev_set_master(slave_dev, NULL);
  2002. /* close slave before restoring its mac address */
  2003. dev_close(slave_dev);
  2004. if (app_abi_ver >= 1) {
  2005. /* restore original ("permanent") mac address*/
  2006. memcpy(addr.sa_data, slave->perm_hwaddr, ETH_ALEN);
  2007. addr.sa_family = slave_dev->type;
  2008. dev_set_mac_address(slave_dev, &addr);
  2009. }
  2010. /* restore the original state of the IFF_NOARP flag that might have
  2011. * been set by bond_set_slave_inactive_flags()
  2012. */
  2013. if ((slave->original_flags & IFF_NOARP) == 0) {
  2014. slave_dev->flags &= ~IFF_NOARP;
  2015. }
  2016. kfree(slave);
  2017. /* re-acquire the lock before getting the next slave */
  2018. write_lock_bh(&bond->lock);
  2019. }
  2020. /* zero the mac address of the master so it will be
  2021. * set by the application to the mac address of the
  2022. * first slave
  2023. */
  2024. memset(bond_dev->dev_addr, 0, bond_dev->addr_len);
  2025. if (list_empty(&bond->vlan_list)) {
  2026. bond_dev->features |= NETIF_F_VLAN_CHALLENGED;
  2027. } else {
  2028. printk(KERN_WARNING DRV_NAME
  2029. ": Warning: clearing HW address of %s while it "
  2030. "still has VLANs.\n",
  2031. bond_dev->name);
  2032. printk(KERN_WARNING DRV_NAME
  2033. ": When re-adding slaves, make sure the bond's "
  2034. "HW address matches its VLANs'.\n");
  2035. }
  2036. printk(KERN_INFO DRV_NAME
  2037. ": %s: released all slaves\n",
  2038. bond_dev->name);
  2039. out:
  2040. write_unlock_bh(&bond->lock);
  2041. return 0;
  2042. }
  2043. /*
  2044. * This function changes the active slave to slave <slave_dev>.
  2045. * It returns -EINVAL in the following cases.
  2046. * - <slave_dev> is not found in the list.
  2047. * - There is not active slave now.
  2048. * - <slave_dev> is already active.
  2049. * - The link state of <slave_dev> is not BOND_LINK_UP.
  2050. * - <slave_dev> is not running.
  2051. * In these cases, this fuction does nothing.
  2052. * In the other cases, currnt_slave pointer is changed and 0 is returned.
  2053. */
  2054. static int bond_ioctl_change_active(struct net_device *bond_dev, struct net_device *slave_dev)
  2055. {
  2056. struct bonding *bond = bond_dev->priv;
  2057. struct slave *old_active = NULL;
  2058. struct slave *new_active = NULL;
  2059. int res = 0;
  2060. if (!USES_PRIMARY(bond->params.mode)) {
  2061. return -EINVAL;
  2062. }
  2063. /* Verify that master_dev is indeed the master of slave_dev */
  2064. if (!(slave_dev->flags & IFF_SLAVE) ||
  2065. (slave_dev->master != bond_dev)) {
  2066. return -EINVAL;
  2067. }
  2068. write_lock_bh(&bond->lock);
  2069. old_active = bond->curr_active_slave;
  2070. new_active = bond_get_slave_by_dev(bond, slave_dev);
  2071. /*
  2072. * Changing to the current active: do nothing; return success.
  2073. */
  2074. if (new_active && (new_active == old_active)) {
  2075. write_unlock_bh(&bond->lock);
  2076. return 0;
  2077. }
  2078. if ((new_active) &&
  2079. (old_active) &&
  2080. (new_active->link == BOND_LINK_UP) &&
  2081. IS_UP(new_active->dev)) {
  2082. bond_change_active_slave(bond, new_active);
  2083. } else {
  2084. res = -EINVAL;
  2085. }
  2086. write_unlock_bh(&bond->lock);
  2087. return res;
  2088. }
  2089. static int bond_ethtool_ioctl(struct net_device *bond_dev, struct ifreq *ifr)
  2090. {
  2091. struct ethtool_drvinfo info;
  2092. void __user *addr = ifr->ifr_data;
  2093. uint32_t cmd;
  2094. if (get_user(cmd, (uint32_t __user *)addr)) {
  2095. return -EFAULT;
  2096. }
  2097. switch (cmd) {
  2098. case ETHTOOL_GDRVINFO:
  2099. if (copy_from_user(&info, addr, sizeof(info))) {
  2100. return -EFAULT;
  2101. }
  2102. if (strcmp(info.driver, "ifenslave") == 0) {
  2103. int new_abi_ver;
  2104. char *endptr;
  2105. new_abi_ver = simple_strtoul(info.fw_version,
  2106. &endptr, 0);
  2107. if (*endptr) {
  2108. printk(KERN_ERR DRV_NAME
  2109. ": Error: got invalid ABI "
  2110. "version from application\n");
  2111. return -EINVAL;
  2112. }
  2113. if (orig_app_abi_ver == -1) {
  2114. orig_app_abi_ver = new_abi_ver;
  2115. }
  2116. app_abi_ver = new_abi_ver;
  2117. }
  2118. strncpy(info.driver, DRV_NAME, 32);
  2119. strncpy(info.version, DRV_VERSION, 32);
  2120. snprintf(info.fw_version, 32, "%d", BOND_ABI_VERSION);
  2121. if (copy_to_user(addr, &info, sizeof(info))) {
  2122. return -EFAULT;
  2123. }
  2124. return 0;
  2125. default:
  2126. return -EOPNOTSUPP;
  2127. }
  2128. }
  2129. static int bond_info_query(struct net_device *bond_dev, struct ifbond *info)
  2130. {
  2131. struct bonding *bond = bond_dev->priv;
  2132. info->bond_mode = bond->params.mode;
  2133. info->miimon = bond->params.miimon;
  2134. read_lock_bh(&bond->lock);
  2135. info->num_slaves = bond->slave_cnt;
  2136. read_unlock_bh(&bond->lock);
  2137. return 0;
  2138. }
  2139. static int bond_slave_info_query(struct net_device *bond_dev, struct ifslave *info)
  2140. {
  2141. struct bonding *bond = bond_dev->priv;
  2142. struct slave *slave;
  2143. int i, found = 0;
  2144. if (info->slave_id < 0) {
  2145. return -ENODEV;
  2146. }
  2147. read_lock_bh(&bond->lock);
  2148. bond_for_each_slave(bond, slave, i) {
  2149. if (i == (int)info->slave_id) {
  2150. found = 1;
  2151. break;
  2152. }
  2153. }
  2154. read_unlock_bh(&bond->lock);
  2155. if (found) {
  2156. strcpy(info->slave_name, slave->dev->name);
  2157. info->link = slave->link;
  2158. info->state = slave->state;
  2159. info->link_failure_count = slave->link_failure_count;
  2160. } else {
  2161. return -ENODEV;
  2162. }
  2163. return 0;
  2164. }
  2165. /*-------------------------------- Monitoring -------------------------------*/
  2166. /* this function is called regularly to monitor each slave's link. */
  2167. static void bond_mii_monitor(struct net_device *bond_dev)
  2168. {
  2169. struct bonding *bond = bond_dev->priv;
  2170. struct slave *slave, *oldcurrent;
  2171. int do_failover = 0;
  2172. int delta_in_ticks;
  2173. int i;
  2174. read_lock(&bond->lock);
  2175. delta_in_ticks = (bond->params.miimon * HZ) / 1000;
  2176. if (bond->kill_timers) {
  2177. goto out;
  2178. }
  2179. if (bond->slave_cnt == 0) {
  2180. goto re_arm;
  2181. }
  2182. /* we will try to read the link status of each of our slaves, and
  2183. * set their IFF_RUNNING flag appropriately. For each slave not
  2184. * supporting MII status, we won't do anything so that a user-space
  2185. * program could monitor the link itself if needed.
  2186. */
  2187. read_lock(&bond->curr_slave_lock);
  2188. oldcurrent = bond->curr_active_slave;
  2189. read_unlock(&bond->curr_slave_lock);
  2190. bond_for_each_slave(bond, slave, i) {
  2191. struct net_device *slave_dev = slave->dev;
  2192. int link_state;
  2193. u16 old_speed = slave->speed;
  2194. u8 old_duplex = slave->duplex;
  2195. link_state = bond_check_dev_link(bond, slave_dev, 0);
  2196. switch (slave->link) {
  2197. case BOND_LINK_UP: /* the link was up */
  2198. if (link_state == BMSR_LSTATUS) {
  2199. /* link stays up, nothing more to do */
  2200. break;
  2201. } else { /* link going down */
  2202. slave->link = BOND_LINK_FAIL;
  2203. slave->delay = bond->params.downdelay;
  2204. if (slave->link_failure_count < UINT_MAX) {
  2205. slave->link_failure_count++;
  2206. }
  2207. if (bond->params.downdelay) {
  2208. printk(KERN_INFO DRV_NAME
  2209. ": %s: link status down for %s "
  2210. "interface %s, disabling it in "
  2211. "%d ms.\n",
  2212. bond_dev->name,
  2213. IS_UP(slave_dev)
  2214. ? ((bond->params.mode == BOND_MODE_ACTIVEBACKUP)
  2215. ? ((slave == oldcurrent)
  2216. ? "active " : "backup ")
  2217. : "")
  2218. : "idle ",
  2219. slave_dev->name,
  2220. bond->params.downdelay * bond->params.miimon);
  2221. }
  2222. }
  2223. /* no break ! fall through the BOND_LINK_FAIL test to
  2224. ensure proper action to be taken
  2225. */
  2226. case BOND_LINK_FAIL: /* the link has just gone down */
  2227. if (link_state != BMSR_LSTATUS) {
  2228. /* link stays down */
  2229. if (slave->delay <= 0) {
  2230. /* link down for too long time */
  2231. slave->link = BOND_LINK_DOWN;
  2232. /* in active/backup mode, we must
  2233. * completely disable this interface
  2234. */
  2235. if ((bond->params.mode == BOND_MODE_ACTIVEBACKUP) ||
  2236. (bond->params.mode == BOND_MODE_8023AD)) {
  2237. bond_set_slave_inactive_flags(slave);
  2238. }
  2239. printk(KERN_INFO DRV_NAME
  2240. ": %s: link status definitely "
  2241. "down for interface %s, "
  2242. "disabling it\n",
  2243. bond_dev->name,
  2244. slave_dev->name);
  2245. /* notify ad that the link status has changed */
  2246. if (bond->params.mode == BOND_MODE_8023AD) {
  2247. bond_3ad_handle_link_change(slave, BOND_LINK_DOWN);
  2248. }
  2249. if ((bond->params.mode == BOND_MODE_TLB) ||
  2250. (bond->params.mode == BOND_MODE_ALB)) {
  2251. bond_alb_handle_link_change(bond, slave, BOND_LINK_DOWN);
  2252. }
  2253. if (slave == oldcurrent) {
  2254. do_failover = 1;
  2255. }
  2256. } else {
  2257. slave->delay--;
  2258. }
  2259. } else {
  2260. /* link up again */
  2261. slave->link = BOND_LINK_UP;
  2262. slave->jiffies = jiffies;
  2263. printk(KERN_INFO DRV_NAME
  2264. ": %s: link status up again after %d "
  2265. "ms for interface %s.\n",
  2266. bond_dev->name,
  2267. (bond->params.downdelay - slave->delay) * bond->params.miimon,
  2268. slave_dev->name);
  2269. }
  2270. break;
  2271. case BOND_LINK_DOWN: /* the link was down */
  2272. if (link_state != BMSR_LSTATUS) {
  2273. /* the link stays down, nothing more to do */
  2274. break;
  2275. } else { /* link going up */
  2276. slave->link = BOND_LINK_BACK;
  2277. slave->delay = bond->params.updelay;
  2278. if (bond->params.updelay) {
  2279. /* if updelay == 0, no need to
  2280. advertise about a 0 ms delay */
  2281. printk(KERN_INFO DRV_NAME
  2282. ": %s: link status up for "
  2283. "interface %s, enabling it "
  2284. "in %d ms.\n",
  2285. bond_dev->name,
  2286. slave_dev->name,
  2287. bond->params.updelay * bond->params.miimon);
  2288. }
  2289. }
  2290. /* no break ! fall through the BOND_LINK_BACK state in
  2291. case there's something to do.
  2292. */
  2293. case BOND_LINK_BACK: /* the link has just come back */
  2294. if (link_state != BMSR_LSTATUS) {
  2295. /* link down again */
  2296. slave->link = BOND_LINK_DOWN;
  2297. printk(KERN_INFO DRV_NAME
  2298. ": %s: link status down again after %d "
  2299. "ms for interface %s.\n",
  2300. bond_dev->name,
  2301. (bond->params.updelay - slave->delay) * bond->params.miimon,
  2302. slave_dev->name);
  2303. } else {
  2304. /* link stays up */
  2305. if (slave->delay == 0) {
  2306. /* now the link has been up for long time enough */
  2307. slave->link = BOND_LINK_UP;
  2308. slave->jiffies = jiffies;
  2309. if (bond->params.mode == BOND_MODE_8023AD) {
  2310. /* prevent it from being the active one */
  2311. slave->state = BOND_STATE_BACKUP;
  2312. } else if (bond->params.mode != BOND_MODE_ACTIVEBACKUP) {
  2313. /* make it immediately active */
  2314. slave->state = BOND_STATE_ACTIVE;
  2315. } else if (slave != bond->primary_slave) {
  2316. /* prevent it from being the active one */
  2317. slave->state = BOND_STATE_BACKUP;
  2318. }
  2319. printk(KERN_INFO DRV_NAME
  2320. ": %s: link status definitely "
  2321. "up for interface %s.\n",
  2322. bond_dev->name,
  2323. slave_dev->name);
  2324. /* notify ad that the link status has changed */
  2325. if (bond->params.mode == BOND_MODE_8023AD) {
  2326. bond_3ad_handle_link_change(slave, BOND_LINK_UP);
  2327. }
  2328. if ((bond->params.mode == BOND_MODE_TLB) ||
  2329. (bond->params.mode == BOND_MODE_ALB)) {
  2330. bond_alb_handle_link_change(bond, slave, BOND_LINK_UP);
  2331. }
  2332. if ((!oldcurrent) ||
  2333. (slave == bond->primary_slave)) {
  2334. do_failover = 1;
  2335. }
  2336. } else {
  2337. slave->delay--;
  2338. }
  2339. }
  2340. break;
  2341. default:
  2342. /* Should not happen */
  2343. printk(KERN_ERR "bonding: Error: %s Illegal value (link=%d)\n",
  2344. slave->dev->name, slave->link);
  2345. goto out;
  2346. } /* end of switch (slave->link) */
  2347. bond_update_speed_duplex(slave);
  2348. if (bond->params.mode == BOND_MODE_8023AD) {
  2349. if (old_speed != slave->speed) {
  2350. bond_3ad_adapter_speed_changed(slave);
  2351. }
  2352. if (old_duplex != slave->duplex) {
  2353. bond_3ad_adapter_duplex_changed(slave);
  2354. }
  2355. }
  2356. } /* end of for */
  2357. if (do_failover) {
  2358. write_lock(&bond->curr_slave_lock);
  2359. bond_select_active_slave(bond);
  2360. if (oldcurrent && !bond->curr_active_slave) {
  2361. printk(KERN_INFO DRV_NAME
  2362. ": %s: now running without any active "
  2363. "interface !\n",
  2364. bond_dev->name);
  2365. }
  2366. write_unlock(&bond->curr_slave_lock);
  2367. }
  2368. re_arm:
  2369. if (bond->params.miimon) {
  2370. mod_timer(&bond->mii_timer, jiffies + delta_in_ticks);
  2371. }
  2372. out:
  2373. read_unlock(&bond->lock);
  2374. }
  2375. static void bond_arp_send_all(struct bonding *bond, struct slave *slave)
  2376. {
  2377. int i;
  2378. u32 *targets = bond->params.arp_targets;
  2379. for (i = 0; (i < BOND_MAX_ARP_TARGETS) && targets[i]; i++) {
  2380. arp_send(ARPOP_REQUEST, ETH_P_ARP, targets[i], slave->dev,
  2381. my_ip, NULL, slave->dev->dev_addr,
  2382. NULL);
  2383. }
  2384. }
  2385. /*
  2386. * this function is called regularly to monitor each slave's link
  2387. * ensuring that traffic is being sent and received when arp monitoring
  2388. * is used in load-balancing mode. if the adapter has been dormant, then an
  2389. * arp is transmitted to generate traffic. see activebackup_arp_monitor for
  2390. * arp monitoring in active backup mode.
  2391. */
  2392. static void bond_loadbalance_arp_mon(struct net_device *bond_dev)
  2393. {
  2394. struct bonding *bond = bond_dev->priv;
  2395. struct slave *slave, *oldcurrent;
  2396. int do_failover = 0;
  2397. int delta_in_ticks;
  2398. int i;
  2399. read_lock(&bond->lock);
  2400. delta_in_ticks = (bond->params.arp_interval * HZ) / 1000;
  2401. if (bond->kill_timers) {
  2402. goto out;
  2403. }
  2404. if (bond->slave_cnt == 0) {
  2405. goto re_arm;
  2406. }
  2407. read_lock(&bond->curr_slave_lock);
  2408. oldcurrent = bond->curr_active_slave;
  2409. read_unlock(&bond->curr_slave_lock);
  2410. /* see if any of the previous devices are up now (i.e. they have
  2411. * xmt and rcv traffic). the curr_active_slave does not come into
  2412. * the picture unless it is null. also, slave->jiffies is not needed
  2413. * here because we send an arp on each slave and give a slave as
  2414. * long as it needs to get the tx/rx within the delta.
  2415. * TODO: what about up/down delay in arp mode? it wasn't here before
  2416. * so it can wait
  2417. */
  2418. bond_for_each_slave(bond, slave, i) {
  2419. if (slave->link != BOND_LINK_UP) {
  2420. if (((jiffies - slave->dev->trans_start) <= delta_in_ticks) &&
  2421. ((jiffies - slave->dev->last_rx) <= delta_in_ticks)) {
  2422. slave->link = BOND_LINK_UP;
  2423. slave->state = BOND_STATE_ACTIVE;
  2424. /* primary_slave has no meaning in round-robin
  2425. * mode. the window of a slave being up and
  2426. * curr_active_slave being null after enslaving
  2427. * is closed.
  2428. */
  2429. if (!oldcurrent) {
  2430. printk(KERN_INFO DRV_NAME
  2431. ": %s: link status definitely "
  2432. "up for interface %s, ",
  2433. bond_dev->name,
  2434. slave->dev->name);
  2435. do_failover = 1;
  2436. } else {
  2437. printk(KERN_INFO DRV_NAME
  2438. ": %s: interface %s is now up\n",
  2439. bond_dev->name,
  2440. slave->dev->name);
  2441. }
  2442. }
  2443. } else {
  2444. /* slave->link == BOND_LINK_UP */
  2445. /* not all switches will respond to an arp request
  2446. * when the source ip is 0, so don't take the link down
  2447. * if we don't know our ip yet
  2448. */
  2449. if (((jiffies - slave->dev->trans_start) >= (2*delta_in_ticks)) ||
  2450. (((jiffies - slave->dev->last_rx) >= (2*delta_in_ticks)) &&
  2451. my_ip)) {
  2452. slave->link = BOND_LINK_DOWN;
  2453. slave->state = BOND_STATE_BACKUP;
  2454. if (slave->link_failure_count < UINT_MAX) {
  2455. slave->link_failure_count++;
  2456. }
  2457. printk(KERN_INFO DRV_NAME
  2458. ": %s: interface %s is now down.\n",
  2459. bond_dev->name,
  2460. slave->dev->name);
  2461. if (slave == oldcurrent) {
  2462. do_failover = 1;
  2463. }
  2464. }
  2465. }
  2466. /* note: if switch is in round-robin mode, all links
  2467. * must tx arp to ensure all links rx an arp - otherwise
  2468. * links may oscillate or not come up at all; if switch is
  2469. * in something like xor mode, there is nothing we can
  2470. * do - all replies will be rx'ed on same link causing slaves
  2471. * to be unstable during low/no traffic periods
  2472. */
  2473. if (IS_UP(slave->dev)) {
  2474. bond_arp_send_all(bond, slave);
  2475. }
  2476. }
  2477. if (do_failover) {
  2478. write_lock(&bond->curr_slave_lock);
  2479. bond_select_active_slave(bond);
  2480. if (oldcurrent && !bond->curr_active_slave) {
  2481. printk(KERN_INFO DRV_NAME
  2482. ": %s: now running without any active "
  2483. "interface !\n",
  2484. bond_dev->name);
  2485. }
  2486. write_unlock(&bond->curr_slave_lock);
  2487. }
  2488. re_arm:
  2489. if (bond->params.arp_interval) {
  2490. mod_timer(&bond->arp_timer, jiffies + delta_in_ticks);
  2491. }
  2492. out:
  2493. read_unlock(&bond->lock);
  2494. }
  2495. /*
  2496. * When using arp monitoring in active-backup mode, this function is
  2497. * called to determine if any backup slaves have went down or a new
  2498. * current slave needs to be found.
  2499. * The backup slaves never generate traffic, they are considered up by merely
  2500. * receiving traffic. If the current slave goes down, each backup slave will
  2501. * be given the opportunity to tx/rx an arp before being taken down - this
  2502. * prevents all slaves from being taken down due to the current slave not
  2503. * sending any traffic for the backups to receive. The arps are not necessarily
  2504. * necessary, any tx and rx traffic will keep the current slave up. While any
  2505. * rx traffic will keep the backup slaves up, the current slave is responsible
  2506. * for generating traffic to keep them up regardless of any other traffic they
  2507. * may have received.
  2508. * see loadbalance_arp_monitor for arp monitoring in load balancing mode
  2509. */
  2510. static void bond_activebackup_arp_mon(struct net_device *bond_dev)
  2511. {
  2512. struct bonding *bond = bond_dev->priv;
  2513. struct slave *slave;
  2514. int delta_in_ticks;
  2515. int i;
  2516. read_lock(&bond->lock);
  2517. delta_in_ticks = (bond->params.arp_interval * HZ) / 1000;
  2518. if (bond->kill_timers) {
  2519. goto out;
  2520. }
  2521. if (bond->slave_cnt == 0) {
  2522. goto re_arm;
  2523. }
  2524. /* determine if any slave has come up or any backup slave has
  2525. * gone down
  2526. * TODO: what about up/down delay in arp mode? it wasn't here before
  2527. * so it can wait
  2528. */
  2529. bond_for_each_slave(bond, slave, i) {
  2530. if (slave->link != BOND_LINK_UP) {
  2531. if ((jiffies - slave->dev->last_rx) <= delta_in_ticks) {
  2532. slave->link = BOND_LINK_UP;
  2533. write_lock(&bond->curr_slave_lock);
  2534. if ((!bond->curr_active_slave) &&
  2535. ((jiffies - slave->dev->trans_start) <= delta_in_ticks)) {
  2536. bond_change_active_slave(bond, slave);
  2537. bond->current_arp_slave = NULL;
  2538. } else if (bond->curr_active_slave != slave) {
  2539. /* this slave has just come up but we
  2540. * already have a current slave; this
  2541. * can also happen if bond_enslave adds
  2542. * a new slave that is up while we are
  2543. * searching for a new slave
  2544. */
  2545. bond_set_slave_inactive_flags(slave);
  2546. bond->current_arp_slave = NULL;
  2547. }
  2548. if (slave == bond->curr_active_slave) {
  2549. printk(KERN_INFO DRV_NAME
  2550. ": %s: %s is up and now the "
  2551. "active interface\n",
  2552. bond_dev->name,
  2553. slave->dev->name);
  2554. } else {
  2555. printk(KERN_INFO DRV_NAME
  2556. ": %s: backup interface %s is "
  2557. "now up\n",
  2558. bond_dev->name,
  2559. slave->dev->name);
  2560. }
  2561. write_unlock(&bond->curr_slave_lock);
  2562. }
  2563. } else {
  2564. read_lock(&bond->curr_slave_lock);
  2565. if ((slave != bond->curr_active_slave) &&
  2566. (!bond->current_arp_slave) &&
  2567. (((jiffies - slave->dev->last_rx) >= 3*delta_in_ticks) &&
  2568. my_ip)) {
  2569. /* a backup slave has gone down; three times
  2570. * the delta allows the current slave to be
  2571. * taken out before the backup slave.
  2572. * note: a non-null current_arp_slave indicates
  2573. * the curr_active_slave went down and we are
  2574. * searching for a new one; under this
  2575. * condition we only take the curr_active_slave
  2576. * down - this gives each slave a chance to
  2577. * tx/rx traffic before being taken out
  2578. */
  2579. read_unlock(&bond->curr_slave_lock);
  2580. slave->link = BOND_LINK_DOWN;
  2581. if (slave->link_failure_count < UINT_MAX) {
  2582. slave->link_failure_count++;
  2583. }
  2584. bond_set_slave_inactive_flags(slave);
  2585. printk(KERN_INFO DRV_NAME
  2586. ": %s: backup interface %s is now down\n",
  2587. bond_dev->name,
  2588. slave->dev->name);
  2589. } else {
  2590. read_unlock(&bond->curr_slave_lock);
  2591. }
  2592. }
  2593. }
  2594. read_lock(&bond->curr_slave_lock);
  2595. slave = bond->curr_active_slave;
  2596. read_unlock(&bond->curr_slave_lock);
  2597. if (slave) {
  2598. /* if we have sent traffic in the past 2*arp_intervals but
  2599. * haven't xmit and rx traffic in that time interval, select
  2600. * a different slave. slave->jiffies is only updated when
  2601. * a slave first becomes the curr_active_slave - not necessarily
  2602. * after every arp; this ensures the slave has a full 2*delta
  2603. * before being taken out. if a primary is being used, check
  2604. * if it is up and needs to take over as the curr_active_slave
  2605. */
  2606. if ((((jiffies - slave->dev->trans_start) >= (2*delta_in_ticks)) ||
  2607. (((jiffies - slave->dev->last_rx) >= (2*delta_in_ticks)) &&
  2608. my_ip)) &&
  2609. ((jiffies - slave->jiffies) >= 2*delta_in_ticks)) {
  2610. slave->link = BOND_LINK_DOWN;
  2611. if (slave->link_failure_count < UINT_MAX) {
  2612. slave->link_failure_count++;
  2613. }
  2614. printk(KERN_INFO DRV_NAME
  2615. ": %s: link status down for active interface "
  2616. "%s, disabling it\n",
  2617. bond_dev->name,
  2618. slave->dev->name);
  2619. write_lock(&bond->curr_slave_lock);
  2620. bond_select_active_slave(bond);
  2621. slave = bond->curr_active_slave;
  2622. write_unlock(&bond->curr_slave_lock);
  2623. bond->current_arp_slave = slave;
  2624. if (slave) {
  2625. slave->jiffies = jiffies;
  2626. }
  2627. } else if ((bond->primary_slave) &&
  2628. (bond->primary_slave != slave) &&
  2629. (bond->primary_slave->link == BOND_LINK_UP)) {
  2630. /* at this point, slave is the curr_active_slave */
  2631. printk(KERN_INFO DRV_NAME
  2632. ": %s: changing from interface %s to primary "
  2633. "interface %s\n",
  2634. bond_dev->name,
  2635. slave->dev->name,
  2636. bond->primary_slave->dev->name);
  2637. /* primary is up so switch to it */
  2638. write_lock(&bond->curr_slave_lock);
  2639. bond_change_active_slave(bond, bond->primary_slave);
  2640. write_unlock(&bond->curr_slave_lock);
  2641. slave = bond->primary_slave;
  2642. slave->jiffies = jiffies;
  2643. } else {
  2644. bond->current_arp_slave = NULL;
  2645. }
  2646. /* the current slave must tx an arp to ensure backup slaves
  2647. * rx traffic
  2648. */
  2649. if (slave && my_ip) {
  2650. bond_arp_send_all(bond, slave);
  2651. }
  2652. }
  2653. /* if we don't have a curr_active_slave, search for the next available
  2654. * backup slave from the current_arp_slave and make it the candidate
  2655. * for becoming the curr_active_slave
  2656. */
  2657. if (!slave) {
  2658. if (!bond->current_arp_slave) {
  2659. bond->current_arp_slave = bond->first_slave;
  2660. }
  2661. if (bond->current_arp_slave) {
  2662. bond_set_slave_inactive_flags(bond->current_arp_slave);
  2663. /* search for next candidate */
  2664. bond_for_each_slave_from(bond, slave, i, bond->current_arp_slave) {
  2665. if (IS_UP(slave->dev)) {
  2666. slave->link = BOND_LINK_BACK;
  2667. bond_set_slave_active_flags(slave);
  2668. bond_arp_send_all(bond, slave);
  2669. slave->jiffies = jiffies;
  2670. bond->current_arp_slave = slave;
  2671. break;
  2672. }
  2673. /* if the link state is up at this point, we
  2674. * mark it down - this can happen if we have
  2675. * simultaneous link failures and
  2676. * reselect_active_interface doesn't make this
  2677. * one the current slave so it is still marked
  2678. * up when it is actually down
  2679. */
  2680. if (slave->link == BOND_LINK_UP) {
  2681. slave->link = BOND_LINK_DOWN;
  2682. if (slave->link_failure_count < UINT_MAX) {
  2683. slave->link_failure_count++;
  2684. }
  2685. bond_set_slave_inactive_flags(slave);
  2686. printk(KERN_INFO DRV_NAME
  2687. ": %s: backup interface %s is "
  2688. "now down.\n",
  2689. bond_dev->name,
  2690. slave->dev->name);
  2691. }
  2692. }
  2693. }
  2694. }
  2695. re_arm:
  2696. if (bond->params.arp_interval) {
  2697. mod_timer(&bond->arp_timer, jiffies + delta_in_ticks);
  2698. }
  2699. out:
  2700. read_unlock(&bond->lock);
  2701. }
  2702. /*------------------------------ proc/seq_file-------------------------------*/
  2703. #ifdef CONFIG_PROC_FS
  2704. #define SEQ_START_TOKEN ((void *)1)
  2705. static void *bond_info_seq_start(struct seq_file *seq, loff_t *pos)
  2706. {
  2707. struct bonding *bond = seq->private;
  2708. loff_t off = 0;
  2709. struct slave *slave;
  2710. int i;
  2711. /* make sure the bond won't be taken away */
  2712. read_lock(&dev_base_lock);
  2713. read_lock_bh(&bond->lock);
  2714. if (*pos == 0) {
  2715. return SEQ_START_TOKEN;
  2716. }
  2717. bond_for_each_slave(bond, slave, i) {
  2718. if (++off == *pos) {
  2719. return slave;
  2720. }
  2721. }
  2722. return NULL;
  2723. }
  2724. static void *bond_info_seq_next(struct seq_file *seq, void *v, loff_t *pos)
  2725. {
  2726. struct bonding *bond = seq->private;
  2727. struct slave *slave = v;
  2728. ++*pos;
  2729. if (v == SEQ_START_TOKEN) {
  2730. return bond->first_slave;
  2731. }
  2732. slave = slave->next;
  2733. return (slave == bond->first_slave) ? NULL : slave;
  2734. }
  2735. static void bond_info_seq_stop(struct seq_file *seq, void *v)
  2736. {
  2737. struct bonding *bond = seq->private;
  2738. read_unlock_bh(&bond->lock);
  2739. read_unlock(&dev_base_lock);
  2740. }
  2741. static void bond_info_show_master(struct seq_file *seq)
  2742. {
  2743. struct bonding *bond = seq->private;
  2744. struct slave *curr;
  2745. read_lock(&bond->curr_slave_lock);
  2746. curr = bond->curr_active_slave;
  2747. read_unlock(&bond->curr_slave_lock);
  2748. seq_printf(seq, "Bonding Mode: %s\n",
  2749. bond_mode_name(bond->params.mode));
  2750. if (USES_PRIMARY(bond->params.mode)) {
  2751. seq_printf(seq, "Primary Slave: %s\n",
  2752. (bond->params.primary[0]) ?
  2753. bond->params.primary : "None");
  2754. seq_printf(seq, "Currently Active Slave: %s\n",
  2755. (curr) ? curr->dev->name : "None");
  2756. }
  2757. seq_printf(seq, "MII Status: %s\n", (curr) ? "up" : "down");
  2758. seq_printf(seq, "MII Polling Interval (ms): %d\n", bond->params.miimon);
  2759. seq_printf(seq, "Up Delay (ms): %d\n",
  2760. bond->params.updelay * bond->params.miimon);
  2761. seq_printf(seq, "Down Delay (ms): %d\n",
  2762. bond->params.downdelay * bond->params.miimon);
  2763. if (bond->params.mode == BOND_MODE_8023AD) {
  2764. struct ad_info ad_info;
  2765. seq_puts(seq, "\n802.3ad info\n");
  2766. seq_printf(seq, "LACP rate: %s\n",
  2767. (bond->params.lacp_fast) ? "fast" : "slow");
  2768. if (bond_3ad_get_active_agg_info(bond, &ad_info)) {
  2769. seq_printf(seq, "bond %s has no active aggregator\n",
  2770. bond->dev->name);
  2771. } else {
  2772. seq_printf(seq, "Active Aggregator Info:\n");
  2773. seq_printf(seq, "\tAggregator ID: %d\n",
  2774. ad_info.aggregator_id);
  2775. seq_printf(seq, "\tNumber of ports: %d\n",
  2776. ad_info.ports);
  2777. seq_printf(seq, "\tActor Key: %d\n",
  2778. ad_info.actor_key);
  2779. seq_printf(seq, "\tPartner Key: %d\n",
  2780. ad_info.partner_key);
  2781. seq_printf(seq, "\tPartner Mac Address: %02x:%02x:%02x:%02x:%02x:%02x\n",
  2782. ad_info.partner_system[0],
  2783. ad_info.partner_system[1],
  2784. ad_info.partner_system[2],
  2785. ad_info.partner_system[3],
  2786. ad_info.partner_system[4],
  2787. ad_info.partner_system[5]);
  2788. }
  2789. }
  2790. }
  2791. static void bond_info_show_slave(struct seq_file *seq, const struct slave *slave)
  2792. {
  2793. struct bonding *bond = seq->private;
  2794. seq_printf(seq, "\nSlave Interface: %s\n", slave->dev->name);
  2795. seq_printf(seq, "MII Status: %s\n",
  2796. (slave->link == BOND_LINK_UP) ? "up" : "down");
  2797. seq_printf(seq, "Link Failure Count: %d\n",
  2798. slave->link_failure_count);
  2799. if (app_abi_ver >= 1) {
  2800. seq_printf(seq,
  2801. "Permanent HW addr: %02x:%02x:%02x:%02x:%02x:%02x\n",
  2802. slave->perm_hwaddr[0],
  2803. slave->perm_hwaddr[1],
  2804. slave->perm_hwaddr[2],
  2805. slave->perm_hwaddr[3],
  2806. slave->perm_hwaddr[4],
  2807. slave->perm_hwaddr[5]);
  2808. }
  2809. if (bond->params.mode == BOND_MODE_8023AD) {
  2810. const struct aggregator *agg
  2811. = SLAVE_AD_INFO(slave).port.aggregator;
  2812. if (agg) {
  2813. seq_printf(seq, "Aggregator ID: %d\n",
  2814. agg->aggregator_identifier);
  2815. } else {
  2816. seq_puts(seq, "Aggregator ID: N/A\n");
  2817. }
  2818. }
  2819. }
  2820. static int bond_info_seq_show(struct seq_file *seq, void *v)
  2821. {
  2822. if (v == SEQ_START_TOKEN) {
  2823. seq_printf(seq, "%s\n", version);
  2824. bond_info_show_master(seq);
  2825. } else {
  2826. bond_info_show_slave(seq, v);
  2827. }
  2828. return 0;
  2829. }
  2830. static struct seq_operations bond_info_seq_ops = {
  2831. .start = bond_info_seq_start,
  2832. .next = bond_info_seq_next,
  2833. .stop = bond_info_seq_stop,
  2834. .show = bond_info_seq_show,
  2835. };
  2836. static int bond_info_open(struct inode *inode, struct file *file)
  2837. {
  2838. struct seq_file *seq;
  2839. struct proc_dir_entry *proc;
  2840. int res;
  2841. res = seq_open(file, &bond_info_seq_ops);
  2842. if (!res) {
  2843. /* recover the pointer buried in proc_dir_entry data */
  2844. seq = file->private_data;
  2845. proc = PDE(inode);
  2846. seq->private = proc->data;
  2847. }
  2848. return res;
  2849. }
  2850. static struct file_operations bond_info_fops = {
  2851. .owner = THIS_MODULE,
  2852. .open = bond_info_open,
  2853. .read = seq_read,
  2854. .llseek = seq_lseek,
  2855. .release = seq_release,
  2856. };
  2857. static int bond_create_proc_entry(struct bonding *bond)
  2858. {
  2859. struct net_device *bond_dev = bond->dev;
  2860. if (bond_proc_dir) {
  2861. bond->proc_entry = create_proc_entry(bond_dev->name,
  2862. S_IRUGO,
  2863. bond_proc_dir);
  2864. if (bond->proc_entry == NULL) {
  2865. printk(KERN_WARNING DRV_NAME
  2866. ": Warning: Cannot create /proc/net/%s/%s\n",
  2867. DRV_NAME, bond_dev->name);
  2868. } else {
  2869. bond->proc_entry->data = bond;
  2870. bond->proc_entry->proc_fops = &bond_info_fops;
  2871. bond->proc_entry->owner = THIS_MODULE;
  2872. memcpy(bond->proc_file_name, bond_dev->name, IFNAMSIZ);
  2873. }
  2874. }
  2875. return 0;
  2876. }
  2877. static void bond_remove_proc_entry(struct bonding *bond)
  2878. {
  2879. if (bond_proc_dir && bond->proc_entry) {
  2880. remove_proc_entry(bond->proc_file_name, bond_proc_dir);
  2881. memset(bond->proc_file_name, 0, IFNAMSIZ);
  2882. bond->proc_entry = NULL;
  2883. }
  2884. }
  2885. /* Create the bonding directory under /proc/net, if doesn't exist yet.
  2886. * Caller must hold rtnl_lock.
  2887. */
  2888. static void bond_create_proc_dir(void)
  2889. {
  2890. int len = strlen(DRV_NAME);
  2891. for (bond_proc_dir = proc_net->subdir; bond_proc_dir;
  2892. bond_proc_dir = bond_proc_dir->next) {
  2893. if ((bond_proc_dir->namelen == len) &&
  2894. !memcmp(bond_proc_dir->name, DRV_NAME, len)) {
  2895. break;
  2896. }
  2897. }
  2898. if (!bond_proc_dir) {
  2899. bond_proc_dir = proc_mkdir(DRV_NAME, proc_net);
  2900. if (bond_proc_dir) {
  2901. bond_proc_dir->owner = THIS_MODULE;
  2902. } else {
  2903. printk(KERN_WARNING DRV_NAME
  2904. ": Warning: cannot create /proc/net/%s\n",
  2905. DRV_NAME);
  2906. }
  2907. }
  2908. }
  2909. /* Destroy the bonding directory under /proc/net, if empty.
  2910. * Caller must hold rtnl_lock.
  2911. */
  2912. static void bond_destroy_proc_dir(void)
  2913. {
  2914. struct proc_dir_entry *de;
  2915. if (!bond_proc_dir) {
  2916. return;
  2917. }
  2918. /* verify that the /proc dir is empty */
  2919. for (de = bond_proc_dir->subdir; de; de = de->next) {
  2920. /* ignore . and .. */
  2921. if (*(de->name) != '.') {
  2922. break;
  2923. }
  2924. }
  2925. if (de) {
  2926. if (bond_proc_dir->owner == THIS_MODULE) {
  2927. bond_proc_dir->owner = NULL;
  2928. }
  2929. } else {
  2930. remove_proc_entry(DRV_NAME, proc_net);
  2931. bond_proc_dir = NULL;
  2932. }
  2933. }
  2934. #endif /* CONFIG_PROC_FS */
  2935. /*-------------------------- netdev event handling --------------------------*/
  2936. /*
  2937. * Change device name
  2938. */
  2939. static int bond_event_changename(struct bonding *bond)
  2940. {
  2941. #ifdef CONFIG_PROC_FS
  2942. bond_remove_proc_entry(bond);
  2943. bond_create_proc_entry(bond);
  2944. #endif
  2945. return NOTIFY_DONE;
  2946. }
  2947. static int bond_master_netdev_event(unsigned long event, struct net_device *bond_dev)
  2948. {
  2949. struct bonding *event_bond = bond_dev->priv;
  2950. switch (event) {
  2951. case NETDEV_CHANGENAME:
  2952. return bond_event_changename(event_bond);
  2953. case NETDEV_UNREGISTER:
  2954. /*
  2955. * TODO: remove a bond from the list?
  2956. */
  2957. break;
  2958. default:
  2959. break;
  2960. }
  2961. return NOTIFY_DONE;
  2962. }
  2963. static int bond_slave_netdev_event(unsigned long event, struct net_device *slave_dev)
  2964. {
  2965. struct net_device *bond_dev = slave_dev->master;
  2966. switch (event) {
  2967. case NETDEV_UNREGISTER:
  2968. if (bond_dev) {
  2969. bond_release(bond_dev, slave_dev);
  2970. }
  2971. break;
  2972. case NETDEV_CHANGE:
  2973. /*
  2974. * TODO: is this what we get if somebody
  2975. * sets up a hierarchical bond, then rmmod's
  2976. * one of the slave bonding devices?
  2977. */
  2978. break;
  2979. case NETDEV_DOWN:
  2980. /*
  2981. * ... Or is it this?
  2982. */
  2983. break;
  2984. case NETDEV_CHANGEMTU:
  2985. /*
  2986. * TODO: Should slaves be allowed to
  2987. * independently alter their MTU? For
  2988. * an active-backup bond, slaves need
  2989. * not be the same type of device, so
  2990. * MTUs may vary. For other modes,
  2991. * slaves arguably should have the
  2992. * same MTUs. To do this, we'd need to
  2993. * take over the slave's change_mtu
  2994. * function for the duration of their
  2995. * servitude.
  2996. */
  2997. break;
  2998. case NETDEV_CHANGENAME:
  2999. /*
  3000. * TODO: handle changing the primary's name
  3001. */
  3002. break;
  3003. default:
  3004. break;
  3005. }
  3006. return NOTIFY_DONE;
  3007. }
  3008. /*
  3009. * bond_netdev_event: handle netdev notifier chain events.
  3010. *
  3011. * This function receives events for the netdev chain. The caller (an
  3012. * ioctl handler calling notifier_call_chain) holds the necessary
  3013. * locks for us to safely manipulate the slave devices (RTNL lock,
  3014. * dev_probe_lock).
  3015. */
  3016. static int bond_netdev_event(struct notifier_block *this, unsigned long event, void *ptr)
  3017. {
  3018. struct net_device *event_dev = (struct net_device *)ptr;
  3019. dprintk("event_dev: %s, event: %lx\n",
  3020. (event_dev ? event_dev->name : "None"),
  3021. event);
  3022. if (event_dev->flags & IFF_MASTER) {
  3023. dprintk("IFF_MASTER\n");
  3024. return bond_master_netdev_event(event, event_dev);
  3025. }
  3026. if (event_dev->flags & IFF_SLAVE) {
  3027. dprintk("IFF_SLAVE\n");
  3028. return bond_slave_netdev_event(event, event_dev);
  3029. }
  3030. return NOTIFY_DONE;
  3031. }
  3032. static struct notifier_block bond_netdev_notifier = {
  3033. .notifier_call = bond_netdev_event,
  3034. };
  3035. /*-------------------------- Packet type handling ---------------------------*/
  3036. /* register to receive lacpdus on a bond */
  3037. static void bond_register_lacpdu(struct bonding *bond)
  3038. {
  3039. struct packet_type *pk_type = &(BOND_AD_INFO(bond).ad_pkt_type);
  3040. /* initialize packet type */
  3041. pk_type->type = PKT_TYPE_LACPDU;
  3042. pk_type->dev = bond->dev;
  3043. pk_type->func = bond_3ad_lacpdu_recv;
  3044. dev_add_pack(pk_type);
  3045. }
  3046. /* unregister to receive lacpdus on a bond */
  3047. static void bond_unregister_lacpdu(struct bonding *bond)
  3048. {
  3049. dev_remove_pack(&(BOND_AD_INFO(bond).ad_pkt_type));
  3050. }
  3051. /*-------------------------- Device entry points ----------------------------*/
  3052. static int bond_open(struct net_device *bond_dev)
  3053. {
  3054. struct bonding *bond = bond_dev->priv;
  3055. struct timer_list *mii_timer = &bond->mii_timer;
  3056. struct timer_list *arp_timer = &bond->arp_timer;
  3057. bond->kill_timers = 0;
  3058. if ((bond->params.mode == BOND_MODE_TLB) ||
  3059. (bond->params.mode == BOND_MODE_ALB)) {
  3060. struct timer_list *alb_timer = &(BOND_ALB_INFO(bond).alb_timer);
  3061. /* bond_alb_initialize must be called before the timer
  3062. * is started.
  3063. */
  3064. if (bond_alb_initialize(bond, (bond->params.mode == BOND_MODE_ALB))) {
  3065. /* something went wrong - fail the open operation */
  3066. return -1;
  3067. }
  3068. init_timer(alb_timer);
  3069. alb_timer->expires = jiffies + 1;
  3070. alb_timer->data = (unsigned long)bond;
  3071. alb_timer->function = (void *)&bond_alb_monitor;
  3072. add_timer(alb_timer);
  3073. }
  3074. if (bond->params.miimon) { /* link check interval, in milliseconds. */
  3075. init_timer(mii_timer);
  3076. mii_timer->expires = jiffies + 1;
  3077. mii_timer->data = (unsigned long)bond_dev;
  3078. mii_timer->function = (void *)&bond_mii_monitor;
  3079. add_timer(mii_timer);
  3080. }
  3081. if (bond->params.arp_interval) { /* arp interval, in milliseconds. */
  3082. init_timer(arp_timer);
  3083. arp_timer->expires = jiffies + 1;
  3084. arp_timer->data = (unsigned long)bond_dev;
  3085. if (bond->params.mode == BOND_MODE_ACTIVEBACKUP) {
  3086. arp_timer->function = (void *)&bond_activebackup_arp_mon;
  3087. } else {
  3088. arp_timer->function = (void *)&bond_loadbalance_arp_mon;
  3089. }
  3090. add_timer(arp_timer);
  3091. }
  3092. if (bond->params.mode == BOND_MODE_8023AD) {
  3093. struct timer_list *ad_timer = &(BOND_AD_INFO(bond).ad_timer);
  3094. init_timer(ad_timer);
  3095. ad_timer->expires = jiffies + 1;
  3096. ad_timer->data = (unsigned long)bond;
  3097. ad_timer->function = (void *)&bond_3ad_state_machine_handler;
  3098. add_timer(ad_timer);
  3099. /* register to receive LACPDUs */
  3100. bond_register_lacpdu(bond);
  3101. }
  3102. return 0;
  3103. }
  3104. static int bond_close(struct net_device *bond_dev)
  3105. {
  3106. struct bonding *bond = bond_dev->priv;
  3107. if (bond->params.mode == BOND_MODE_8023AD) {
  3108. /* Unregister the receive of LACPDUs */
  3109. bond_unregister_lacpdu(bond);
  3110. }
  3111. write_lock_bh(&bond->lock);
  3112. bond_mc_list_destroy(bond);
  3113. /* signal timers not to re-arm */
  3114. bond->kill_timers = 1;
  3115. write_unlock_bh(&bond->lock);
  3116. /* del_timer_sync must run without holding the bond->lock
  3117. * because a running timer might be trying to hold it too
  3118. */
  3119. if (bond->params.miimon) { /* link check interval, in milliseconds. */
  3120. del_timer_sync(&bond->mii_timer);
  3121. }
  3122. if (bond->params.arp_interval) { /* arp interval, in milliseconds. */
  3123. del_timer_sync(&bond->arp_timer);
  3124. }
  3125. switch (bond->params.mode) {
  3126. case BOND_MODE_8023AD:
  3127. del_timer_sync(&(BOND_AD_INFO(bond).ad_timer));
  3128. break;
  3129. case BOND_MODE_TLB:
  3130. case BOND_MODE_ALB:
  3131. del_timer_sync(&(BOND_ALB_INFO(bond).alb_timer));
  3132. break;
  3133. default:
  3134. break;
  3135. }
  3136. /* Release the bonded slaves */
  3137. bond_release_all(bond_dev);
  3138. if ((bond->params.mode == BOND_MODE_TLB) ||
  3139. (bond->params.mode == BOND_MODE_ALB)) {
  3140. /* Must be called only after all
  3141. * slaves have been released
  3142. */
  3143. bond_alb_deinitialize(bond);
  3144. }
  3145. return 0;
  3146. }
  3147. static struct net_device_stats *bond_get_stats(struct net_device *bond_dev)
  3148. {
  3149. struct bonding *bond = bond_dev->priv;
  3150. struct net_device_stats *stats = &(bond->stats), *sstats;
  3151. struct slave *slave;
  3152. int i;
  3153. memset(stats, 0, sizeof(struct net_device_stats));
  3154. read_lock_bh(&bond->lock);
  3155. bond_for_each_slave(bond, slave, i) {
  3156. sstats = slave->dev->get_stats(slave->dev);
  3157. stats->rx_packets += sstats->rx_packets;
  3158. stats->rx_bytes += sstats->rx_bytes;
  3159. stats->rx_errors += sstats->rx_errors;
  3160. stats->rx_dropped += sstats->rx_dropped;
  3161. stats->tx_packets += sstats->tx_packets;
  3162. stats->tx_bytes += sstats->tx_bytes;
  3163. stats->tx_errors += sstats->tx_errors;
  3164. stats->tx_dropped += sstats->tx_dropped;
  3165. stats->multicast += sstats->multicast;
  3166. stats->collisions += sstats->collisions;
  3167. stats->rx_length_errors += sstats->rx_length_errors;
  3168. stats->rx_over_errors += sstats->rx_over_errors;
  3169. stats->rx_crc_errors += sstats->rx_crc_errors;
  3170. stats->rx_frame_errors += sstats->rx_frame_errors;
  3171. stats->rx_fifo_errors += sstats->rx_fifo_errors;
  3172. stats->rx_missed_errors += sstats->rx_missed_errors;
  3173. stats->tx_aborted_errors += sstats->tx_aborted_errors;
  3174. stats->tx_carrier_errors += sstats->tx_carrier_errors;
  3175. stats->tx_fifo_errors += sstats->tx_fifo_errors;
  3176. stats->tx_heartbeat_errors += sstats->tx_heartbeat_errors;
  3177. stats->tx_window_errors += sstats->tx_window_errors;
  3178. }
  3179. read_unlock_bh(&bond->lock);
  3180. return stats;
  3181. }
  3182. static int bond_do_ioctl(struct net_device *bond_dev, struct ifreq *ifr, int cmd)
  3183. {
  3184. struct net_device *slave_dev = NULL;
  3185. struct ifbond k_binfo;
  3186. struct ifbond __user *u_binfo = NULL;
  3187. struct ifslave k_sinfo;
  3188. struct ifslave __user *u_sinfo = NULL;
  3189. struct mii_ioctl_data *mii = NULL;
  3190. int prev_abi_ver = orig_app_abi_ver;
  3191. int res = 0;
  3192. dprintk("bond_ioctl: master=%s, cmd=%d\n",
  3193. bond_dev->name, cmd);
  3194. switch (cmd) {
  3195. case SIOCETHTOOL:
  3196. return bond_ethtool_ioctl(bond_dev, ifr);
  3197. case SIOCGMIIPHY:
  3198. mii = if_mii(ifr);
  3199. if (!mii) {
  3200. return -EINVAL;
  3201. }
  3202. mii->phy_id = 0;
  3203. /* Fall Through */
  3204. case SIOCGMIIREG:
  3205. /*
  3206. * We do this again just in case we were called by SIOCGMIIREG
  3207. * instead of SIOCGMIIPHY.
  3208. */
  3209. mii = if_mii(ifr);
  3210. if (!mii) {
  3211. return -EINVAL;
  3212. }
  3213. if (mii->reg_num == 1) {
  3214. struct bonding *bond = bond_dev->priv;
  3215. mii->val_out = 0;
  3216. read_lock_bh(&bond->lock);
  3217. read_lock(&bond->curr_slave_lock);
  3218. if (bond->curr_active_slave) {
  3219. mii->val_out = BMSR_LSTATUS;
  3220. }
  3221. read_unlock(&bond->curr_slave_lock);
  3222. read_unlock_bh(&bond->lock);
  3223. }
  3224. return 0;
  3225. case BOND_INFO_QUERY_OLD:
  3226. case SIOCBONDINFOQUERY:
  3227. u_binfo = (struct ifbond __user *)ifr->ifr_data;
  3228. if (copy_from_user(&k_binfo, u_binfo, sizeof(ifbond))) {
  3229. return -EFAULT;
  3230. }
  3231. res = bond_info_query(bond_dev, &k_binfo);
  3232. if (res == 0) {
  3233. if (copy_to_user(u_binfo, &k_binfo, sizeof(ifbond))) {
  3234. return -EFAULT;
  3235. }
  3236. }
  3237. return res;
  3238. case BOND_SLAVE_INFO_QUERY_OLD:
  3239. case SIOCBONDSLAVEINFOQUERY:
  3240. u_sinfo = (struct ifslave __user *)ifr->ifr_data;
  3241. if (copy_from_user(&k_sinfo, u_sinfo, sizeof(ifslave))) {
  3242. return -EFAULT;
  3243. }
  3244. res = bond_slave_info_query(bond_dev, &k_sinfo);
  3245. if (res == 0) {
  3246. if (copy_to_user(u_sinfo, &k_sinfo, sizeof(ifslave))) {
  3247. return -EFAULT;
  3248. }
  3249. }
  3250. return res;
  3251. default:
  3252. /* Go on */
  3253. break;
  3254. }
  3255. if (!capable(CAP_NET_ADMIN)) {
  3256. return -EPERM;
  3257. }
  3258. if (orig_app_abi_ver == -1) {
  3259. /* no orig_app_abi_ver was provided yet, so we'll use the
  3260. * current one from now on, even if it's 0
  3261. */
  3262. orig_app_abi_ver = app_abi_ver;
  3263. } else if (orig_app_abi_ver != app_abi_ver) {
  3264. printk(KERN_ERR DRV_NAME
  3265. ": Error: already using ifenslave ABI version %d; to "
  3266. "upgrade ifenslave to version %d, you must first "
  3267. "reload bonding.\n",
  3268. orig_app_abi_ver, app_abi_ver);
  3269. return -EINVAL;
  3270. }
  3271. slave_dev = dev_get_by_name(ifr->ifr_slave);
  3272. dprintk("slave_dev=%p: \n", slave_dev);
  3273. if (!slave_dev) {
  3274. res = -ENODEV;
  3275. } else {
  3276. dprintk("slave_dev->name=%s: \n", slave_dev->name);
  3277. switch (cmd) {
  3278. case BOND_ENSLAVE_OLD:
  3279. case SIOCBONDENSLAVE:
  3280. res = bond_enslave(bond_dev, slave_dev);
  3281. break;
  3282. case BOND_RELEASE_OLD:
  3283. case SIOCBONDRELEASE:
  3284. res = bond_release(bond_dev, slave_dev);
  3285. break;
  3286. case BOND_SETHWADDR_OLD:
  3287. case SIOCBONDSETHWADDR:
  3288. res = bond_sethwaddr(bond_dev, slave_dev);
  3289. break;
  3290. case BOND_CHANGE_ACTIVE_OLD:
  3291. case SIOCBONDCHANGEACTIVE:
  3292. res = bond_ioctl_change_active(bond_dev, slave_dev);
  3293. break;
  3294. default:
  3295. res = -EOPNOTSUPP;
  3296. }
  3297. dev_put(slave_dev);
  3298. }
  3299. if (res < 0) {
  3300. /* The ioctl failed, so there's no point in changing the
  3301. * orig_app_abi_ver. We'll restore it's value just in case
  3302. * we've changed it earlier in this function.
  3303. */
  3304. orig_app_abi_ver = prev_abi_ver;
  3305. }
  3306. return res;
  3307. }
  3308. static void bond_set_multicast_list(struct net_device *bond_dev)
  3309. {
  3310. struct bonding *bond = bond_dev->priv;
  3311. struct dev_mc_list *dmi;
  3312. write_lock_bh(&bond->lock);
  3313. /*
  3314. * Do promisc before checking multicast_mode
  3315. */
  3316. if ((bond_dev->flags & IFF_PROMISC) && !(bond->flags & IFF_PROMISC)) {
  3317. bond_set_promiscuity(bond, 1);
  3318. }
  3319. if (!(bond_dev->flags & IFF_PROMISC) && (bond->flags & IFF_PROMISC)) {
  3320. bond_set_promiscuity(bond, -1);
  3321. }
  3322. /* set allmulti flag to slaves */
  3323. if ((bond_dev->flags & IFF_ALLMULTI) && !(bond->flags & IFF_ALLMULTI)) {
  3324. bond_set_allmulti(bond, 1);
  3325. }
  3326. if (!(bond_dev->flags & IFF_ALLMULTI) && (bond->flags & IFF_ALLMULTI)) {
  3327. bond_set_allmulti(bond, -1);
  3328. }
  3329. bond->flags = bond_dev->flags;
  3330. /* looking for addresses to add to slaves' mc list */
  3331. for (dmi = bond_dev->mc_list; dmi; dmi = dmi->next) {
  3332. if (!bond_mc_list_find_dmi(dmi, bond->mc_list)) {
  3333. bond_mc_add(bond, dmi->dmi_addr, dmi->dmi_addrlen);
  3334. }
  3335. }
  3336. /* looking for addresses to delete from slaves' list */
  3337. for (dmi = bond->mc_list; dmi; dmi = dmi->next) {
  3338. if (!bond_mc_list_find_dmi(dmi, bond_dev->mc_list)) {
  3339. bond_mc_delete(bond, dmi->dmi_addr, dmi->dmi_addrlen);
  3340. }
  3341. }
  3342. /* save master's multicast list */
  3343. bond_mc_list_destroy(bond);
  3344. bond_mc_list_copy(bond_dev->mc_list, bond, GFP_ATOMIC);
  3345. write_unlock_bh(&bond->lock);
  3346. }
  3347. /*
  3348. * Change the MTU of all of a master's slaves to match the master
  3349. */
  3350. static int bond_change_mtu(struct net_device *bond_dev, int new_mtu)
  3351. {
  3352. struct bonding *bond = bond_dev->priv;
  3353. struct slave *slave, *stop_at;
  3354. int res = 0;
  3355. int i;
  3356. dprintk("bond=%p, name=%s, new_mtu=%d\n", bond,
  3357. (bond_dev ? bond_dev->name : "None"), new_mtu);
  3358. /* Can't hold bond->lock with bh disabled here since
  3359. * some base drivers panic. On the other hand we can't
  3360. * hold bond->lock without bh disabled because we'll
  3361. * deadlock. The only solution is to rely on the fact
  3362. * that we're under rtnl_lock here, and the slaves
  3363. * list won't change. This doesn't solve the problem
  3364. * of setting the slave's MTU while it is
  3365. * transmitting, but the assumption is that the base
  3366. * driver can handle that.
  3367. *
  3368. * TODO: figure out a way to safely iterate the slaves
  3369. * list, but without holding a lock around the actual
  3370. * call to the base driver.
  3371. */
  3372. bond_for_each_slave(bond, slave, i) {
  3373. dprintk("s %p s->p %p c_m %p\n", slave,
  3374. slave->prev, slave->dev->change_mtu);
  3375. res = dev_set_mtu(slave->dev, new_mtu);
  3376. if (res) {
  3377. /* If we failed to set the slave's mtu to the new value
  3378. * we must abort the operation even in ACTIVE_BACKUP
  3379. * mode, because if we allow the backup slaves to have
  3380. * different mtu values than the active slave we'll
  3381. * need to change their mtu when doing a failover. That
  3382. * means changing their mtu from timer context, which
  3383. * is probably not a good idea.
  3384. */
  3385. dprintk("err %d %s\n", res, slave->dev->name);
  3386. goto unwind;
  3387. }
  3388. }
  3389. bond_dev->mtu = new_mtu;
  3390. return 0;
  3391. unwind:
  3392. /* unwind from head to the slave that failed */
  3393. stop_at = slave;
  3394. bond_for_each_slave_from_to(bond, slave, i, bond->first_slave, stop_at) {
  3395. int tmp_res;
  3396. tmp_res = dev_set_mtu(slave->dev, bond_dev->mtu);
  3397. if (tmp_res) {
  3398. dprintk("unwind err %d dev %s\n", tmp_res,
  3399. slave->dev->name);
  3400. }
  3401. }
  3402. return res;
  3403. }
  3404. /*
  3405. * Change HW address
  3406. *
  3407. * Note that many devices must be down to change the HW address, and
  3408. * downing the master releases all slaves. We can make bonds full of
  3409. * bonding devices to test this, however.
  3410. */
  3411. static int bond_set_mac_address(struct net_device *bond_dev, void *addr)
  3412. {
  3413. struct bonding *bond = bond_dev->priv;
  3414. struct sockaddr *sa = addr, tmp_sa;
  3415. struct slave *slave, *stop_at;
  3416. int res = 0;
  3417. int i;
  3418. dprintk("bond=%p, name=%s\n", bond, (bond_dev ? bond_dev->name : "None"));
  3419. if (!is_valid_ether_addr(sa->sa_data)) {
  3420. return -EADDRNOTAVAIL;
  3421. }
  3422. /* Can't hold bond->lock with bh disabled here since
  3423. * some base drivers panic. On the other hand we can't
  3424. * hold bond->lock without bh disabled because we'll
  3425. * deadlock. The only solution is to rely on the fact
  3426. * that we're under rtnl_lock here, and the slaves
  3427. * list won't change. This doesn't solve the problem
  3428. * of setting the slave's hw address while it is
  3429. * transmitting, but the assumption is that the base
  3430. * driver can handle that.
  3431. *
  3432. * TODO: figure out a way to safely iterate the slaves
  3433. * list, but without holding a lock around the actual
  3434. * call to the base driver.
  3435. */
  3436. bond_for_each_slave(bond, slave, i) {
  3437. dprintk("slave %p %s\n", slave, slave->dev->name);
  3438. if (slave->dev->set_mac_address == NULL) {
  3439. res = -EOPNOTSUPP;
  3440. dprintk("EOPNOTSUPP %s\n", slave->dev->name);
  3441. goto unwind;
  3442. }
  3443. res = dev_set_mac_address(slave->dev, addr);
  3444. if (res) {
  3445. /* TODO: consider downing the slave
  3446. * and retry ?
  3447. * User should expect communications
  3448. * breakage anyway until ARP finish
  3449. * updating, so...
  3450. */
  3451. dprintk("err %d %s\n", res, slave->dev->name);
  3452. goto unwind;
  3453. }
  3454. }
  3455. /* success */
  3456. memcpy(bond_dev->dev_addr, sa->sa_data, bond_dev->addr_len);
  3457. return 0;
  3458. unwind:
  3459. memcpy(tmp_sa.sa_data, bond_dev->dev_addr, bond_dev->addr_len);
  3460. tmp_sa.sa_family = bond_dev->type;
  3461. /* unwind from head to the slave that failed */
  3462. stop_at = slave;
  3463. bond_for_each_slave_from_to(bond, slave, i, bond->first_slave, stop_at) {
  3464. int tmp_res;
  3465. tmp_res = dev_set_mac_address(slave->dev, &tmp_sa);
  3466. if (tmp_res) {
  3467. dprintk("unwind err %d dev %s\n", tmp_res,
  3468. slave->dev->name);
  3469. }
  3470. }
  3471. return res;
  3472. }
  3473. static int bond_xmit_roundrobin(struct sk_buff *skb, struct net_device *bond_dev)
  3474. {
  3475. struct bonding *bond = bond_dev->priv;
  3476. struct slave *slave, *start_at;
  3477. int i;
  3478. int res = 1;
  3479. read_lock(&bond->lock);
  3480. if (!BOND_IS_OK(bond)) {
  3481. goto out;
  3482. }
  3483. read_lock(&bond->curr_slave_lock);
  3484. slave = start_at = bond->curr_active_slave;
  3485. read_unlock(&bond->curr_slave_lock);
  3486. if (!slave) {
  3487. goto out;
  3488. }
  3489. bond_for_each_slave_from(bond, slave, i, start_at) {
  3490. if (IS_UP(slave->dev) &&
  3491. (slave->link == BOND_LINK_UP) &&
  3492. (slave->state == BOND_STATE_ACTIVE)) {
  3493. res = bond_dev_queue_xmit(bond, skb, slave->dev);
  3494. write_lock(&bond->curr_slave_lock);
  3495. bond->curr_active_slave = slave->next;
  3496. write_unlock(&bond->curr_slave_lock);
  3497. break;
  3498. }
  3499. }
  3500. out:
  3501. if (res) {
  3502. /* no suitable interface, frame not sent */
  3503. dev_kfree_skb(skb);
  3504. }
  3505. read_unlock(&bond->lock);
  3506. return 0;
  3507. }
  3508. /*
  3509. * in active-backup mode, we know that bond->curr_active_slave is always valid if
  3510. * the bond has a usable interface.
  3511. */
  3512. static int bond_xmit_activebackup(struct sk_buff *skb, struct net_device *bond_dev)
  3513. {
  3514. struct bonding *bond = bond_dev->priv;
  3515. int res = 1;
  3516. /* if we are sending arp packets, try to at least
  3517. identify our own ip address */
  3518. if (bond->params.arp_interval && !my_ip &&
  3519. (skb->protocol == __constant_htons(ETH_P_ARP))) {
  3520. char *the_ip = (char *)skb->data +
  3521. sizeof(struct ethhdr) +
  3522. sizeof(struct arphdr) +
  3523. ETH_ALEN;
  3524. memcpy(&my_ip, the_ip, 4);
  3525. }
  3526. read_lock(&bond->lock);
  3527. read_lock(&bond->curr_slave_lock);
  3528. if (!BOND_IS_OK(bond)) {
  3529. goto out;
  3530. }
  3531. if (bond->curr_active_slave) { /* one usable interface */
  3532. res = bond_dev_queue_xmit(bond, skb, bond->curr_active_slave->dev);
  3533. }
  3534. out:
  3535. if (res) {
  3536. /* no suitable interface, frame not sent */
  3537. dev_kfree_skb(skb);
  3538. }
  3539. read_unlock(&bond->curr_slave_lock);
  3540. read_unlock(&bond->lock);
  3541. return 0;
  3542. }
  3543. /*
  3544. * in XOR mode, we determine the output device by performing xor on
  3545. * the source and destination hw adresses. If this device is not
  3546. * enabled, find the next slave following this xor slave.
  3547. */
  3548. static int bond_xmit_xor(struct sk_buff *skb, struct net_device *bond_dev)
  3549. {
  3550. struct bonding *bond = bond_dev->priv;
  3551. struct ethhdr *data = (struct ethhdr *)skb->data;
  3552. struct slave *slave, *start_at;
  3553. int slave_no;
  3554. int i;
  3555. int res = 1;
  3556. read_lock(&bond->lock);
  3557. if (!BOND_IS_OK(bond)) {
  3558. goto out;
  3559. }
  3560. slave_no = (data->h_dest[5]^bond_dev->dev_addr[5]) % bond->slave_cnt;
  3561. bond_for_each_slave(bond, slave, i) {
  3562. slave_no--;
  3563. if (slave_no < 0) {
  3564. break;
  3565. }
  3566. }
  3567. start_at = slave;
  3568. bond_for_each_slave_from(bond, slave, i, start_at) {
  3569. if (IS_UP(slave->dev) &&
  3570. (slave->link == BOND_LINK_UP) &&
  3571. (slave->state == BOND_STATE_ACTIVE)) {
  3572. res = bond_dev_queue_xmit(bond, skb, slave->dev);
  3573. break;
  3574. }
  3575. }
  3576. out:
  3577. if (res) {
  3578. /* no suitable interface, frame not sent */
  3579. dev_kfree_skb(skb);
  3580. }
  3581. read_unlock(&bond->lock);
  3582. return 0;
  3583. }
  3584. /*
  3585. * in broadcast mode, we send everything to all usable interfaces.
  3586. */
  3587. static int bond_xmit_broadcast(struct sk_buff *skb, struct net_device *bond_dev)
  3588. {
  3589. struct bonding *bond = bond_dev->priv;
  3590. struct slave *slave, *start_at;
  3591. struct net_device *tx_dev = NULL;
  3592. int i;
  3593. int res = 1;
  3594. read_lock(&bond->lock);
  3595. if (!BOND_IS_OK(bond)) {
  3596. goto out;
  3597. }
  3598. read_lock(&bond->curr_slave_lock);
  3599. start_at = bond->curr_active_slave;
  3600. read_unlock(&bond->curr_slave_lock);
  3601. if (!start_at) {
  3602. goto out;
  3603. }
  3604. bond_for_each_slave_from(bond, slave, i, start_at) {
  3605. if (IS_UP(slave->dev) &&
  3606. (slave->link == BOND_LINK_UP) &&
  3607. (slave->state == BOND_STATE_ACTIVE)) {
  3608. if (tx_dev) {
  3609. struct sk_buff *skb2 = skb_clone(skb, GFP_ATOMIC);
  3610. if (!skb2) {
  3611. printk(KERN_ERR DRV_NAME
  3612. ": Error: bond_xmit_broadcast(): "
  3613. "skb_clone() failed\n");
  3614. continue;
  3615. }
  3616. res = bond_dev_queue_xmit(bond, skb2, tx_dev);
  3617. if (res) {
  3618. dev_kfree_skb(skb2);
  3619. continue;
  3620. }
  3621. }
  3622. tx_dev = slave->dev;
  3623. }
  3624. }
  3625. if (tx_dev) {
  3626. res = bond_dev_queue_xmit(bond, skb, tx_dev);
  3627. }
  3628. out:
  3629. if (res) {
  3630. /* no suitable interface, frame not sent */
  3631. dev_kfree_skb(skb);
  3632. }
  3633. /* frame sent to all suitable interfaces */
  3634. read_unlock(&bond->lock);
  3635. return 0;
  3636. }
  3637. /*------------------------- Device initialization ---------------------------*/
  3638. /*
  3639. * set bond mode specific net device operations
  3640. */
  3641. static inline void bond_set_mode_ops(struct net_device *bond_dev, int mode)
  3642. {
  3643. switch (mode) {
  3644. case BOND_MODE_ROUNDROBIN:
  3645. bond_dev->hard_start_xmit = bond_xmit_roundrobin;
  3646. break;
  3647. case BOND_MODE_ACTIVEBACKUP:
  3648. bond_dev->hard_start_xmit = bond_xmit_activebackup;
  3649. break;
  3650. case BOND_MODE_XOR:
  3651. bond_dev->hard_start_xmit = bond_xmit_xor;
  3652. break;
  3653. case BOND_MODE_BROADCAST:
  3654. bond_dev->hard_start_xmit = bond_xmit_broadcast;
  3655. break;
  3656. case BOND_MODE_8023AD:
  3657. bond_dev->hard_start_xmit = bond_3ad_xmit_xor;
  3658. break;
  3659. case BOND_MODE_TLB:
  3660. case BOND_MODE_ALB:
  3661. bond_dev->hard_start_xmit = bond_alb_xmit;
  3662. bond_dev->set_mac_address = bond_alb_set_mac_address;
  3663. break;
  3664. default:
  3665. /* Should never happen, mode already checked */
  3666. printk(KERN_ERR DRV_NAME
  3667. ": Error: Unknown bonding mode %d\n",
  3668. mode);
  3669. break;
  3670. }
  3671. }
  3672. /*
  3673. * Does not allocate but creates a /proc entry.
  3674. * Allowed to fail.
  3675. */
  3676. static int __init bond_init(struct net_device *bond_dev, struct bond_params *params)
  3677. {
  3678. struct bonding *bond = bond_dev->priv;
  3679. dprintk("Begin bond_init for %s\n", bond_dev->name);
  3680. /* initialize rwlocks */
  3681. rwlock_init(&bond->lock);
  3682. rwlock_init(&bond->curr_slave_lock);
  3683. bond->params = *params; /* copy params struct */
  3684. /* Initialize pointers */
  3685. bond->first_slave = NULL;
  3686. bond->curr_active_slave = NULL;
  3687. bond->current_arp_slave = NULL;
  3688. bond->primary_slave = NULL;
  3689. bond->dev = bond_dev;
  3690. INIT_LIST_HEAD(&bond->vlan_list);
  3691. /* Initialize the device entry points */
  3692. bond_dev->open = bond_open;
  3693. bond_dev->stop = bond_close;
  3694. bond_dev->get_stats = bond_get_stats;
  3695. bond_dev->do_ioctl = bond_do_ioctl;
  3696. bond_dev->set_multicast_list = bond_set_multicast_list;
  3697. bond_dev->change_mtu = bond_change_mtu;
  3698. bond_dev->set_mac_address = bond_set_mac_address;
  3699. bond_set_mode_ops(bond_dev, bond->params.mode);
  3700. bond_dev->destructor = free_netdev;
  3701. /* Initialize the device options */
  3702. bond_dev->tx_queue_len = 0;
  3703. bond_dev->flags |= IFF_MASTER|IFF_MULTICAST;
  3704. /* At first, we block adding VLANs. That's the only way to
  3705. * prevent problems that occur when adding VLANs over an
  3706. * empty bond. The block will be removed once non-challenged
  3707. * slaves are enslaved.
  3708. */
  3709. bond_dev->features |= NETIF_F_VLAN_CHALLENGED;
  3710. /* don't acquire bond device's xmit_lock when
  3711. * transmitting */
  3712. bond_dev->features |= NETIF_F_LLTX;
  3713. /* By default, we declare the bond to be fully
  3714. * VLAN hardware accelerated capable. Special
  3715. * care is taken in the various xmit functions
  3716. * when there are slaves that are not hw accel
  3717. * capable
  3718. */
  3719. bond_dev->vlan_rx_register = bond_vlan_rx_register;
  3720. bond_dev->vlan_rx_add_vid = bond_vlan_rx_add_vid;
  3721. bond_dev->vlan_rx_kill_vid = bond_vlan_rx_kill_vid;
  3722. bond_dev->features |= (NETIF_F_HW_VLAN_TX |
  3723. NETIF_F_HW_VLAN_RX |
  3724. NETIF_F_HW_VLAN_FILTER);
  3725. #ifdef CONFIG_PROC_FS
  3726. bond_create_proc_entry(bond);
  3727. #endif
  3728. list_add_tail(&bond->bond_list, &bond_dev_list);
  3729. return 0;
  3730. }
  3731. /* De-initialize device specific data.
  3732. * Caller must hold rtnl_lock.
  3733. */
  3734. static inline void bond_deinit(struct net_device *bond_dev)
  3735. {
  3736. struct bonding *bond = bond_dev->priv;
  3737. list_del(&bond->bond_list);
  3738. #ifdef CONFIG_PROC_FS
  3739. bond_remove_proc_entry(bond);
  3740. #endif
  3741. }
  3742. /* Unregister and free all bond devices.
  3743. * Caller must hold rtnl_lock.
  3744. */
  3745. static void bond_free_all(void)
  3746. {
  3747. struct bonding *bond, *nxt;
  3748. list_for_each_entry_safe(bond, nxt, &bond_dev_list, bond_list) {
  3749. struct net_device *bond_dev = bond->dev;
  3750. unregister_netdevice(bond_dev);
  3751. bond_deinit(bond_dev);
  3752. }
  3753. #ifdef CONFIG_PROC_FS
  3754. bond_destroy_proc_dir();
  3755. #endif
  3756. }
  3757. /*------------------------- Module initialization ---------------------------*/
  3758. /*
  3759. * Convert string input module parms. Accept either the
  3760. * number of the mode or its string name.
  3761. */
  3762. static inline int bond_parse_parm(char *mode_arg, struct bond_parm_tbl *tbl)
  3763. {
  3764. int i;
  3765. for (i = 0; tbl[i].modename; i++) {
  3766. if ((isdigit(*mode_arg) &&
  3767. tbl[i].mode == simple_strtol(mode_arg, NULL, 0)) ||
  3768. (strncmp(mode_arg, tbl[i].modename,
  3769. strlen(tbl[i].modename)) == 0)) {
  3770. return tbl[i].mode;
  3771. }
  3772. }
  3773. return -1;
  3774. }
  3775. static int bond_check_params(struct bond_params *params)
  3776. {
  3777. /*
  3778. * Convert string parameters.
  3779. */
  3780. if (mode) {
  3781. bond_mode = bond_parse_parm(mode, bond_mode_tbl);
  3782. if (bond_mode == -1) {
  3783. printk(KERN_ERR DRV_NAME
  3784. ": Error: Invalid bonding mode \"%s\"\n",
  3785. mode == NULL ? "NULL" : mode);
  3786. return -EINVAL;
  3787. }
  3788. }
  3789. if (lacp_rate) {
  3790. if (bond_mode != BOND_MODE_8023AD) {
  3791. printk(KERN_INFO DRV_NAME
  3792. ": lacp_rate param is irrelevant in mode %s\n",
  3793. bond_mode_name(bond_mode));
  3794. } else {
  3795. lacp_fast = bond_parse_parm(lacp_rate, bond_lacp_tbl);
  3796. if (lacp_fast == -1) {
  3797. printk(KERN_ERR DRV_NAME
  3798. ": Error: Invalid lacp rate \"%s\"\n",
  3799. lacp_rate == NULL ? "NULL" : lacp_rate);
  3800. return -EINVAL;
  3801. }
  3802. }
  3803. }
  3804. if (max_bonds < 1 || max_bonds > INT_MAX) {
  3805. printk(KERN_WARNING DRV_NAME
  3806. ": Warning: max_bonds (%d) not in range %d-%d, so it "
  3807. "was reset to BOND_DEFAULT_MAX_BONDS (%d)",
  3808. max_bonds, 1, INT_MAX, BOND_DEFAULT_MAX_BONDS);
  3809. max_bonds = BOND_DEFAULT_MAX_BONDS;
  3810. }
  3811. if (miimon < 0) {
  3812. printk(KERN_WARNING DRV_NAME
  3813. ": Warning: miimon module parameter (%d), "
  3814. "not in range 0-%d, so it was reset to %d\n",
  3815. miimon, INT_MAX, BOND_LINK_MON_INTERV);
  3816. miimon = BOND_LINK_MON_INTERV;
  3817. }
  3818. if (updelay < 0) {
  3819. printk(KERN_WARNING DRV_NAME
  3820. ": Warning: updelay module parameter (%d), "
  3821. "not in range 0-%d, so it was reset to 0\n",
  3822. updelay, INT_MAX);
  3823. updelay = 0;
  3824. }
  3825. if (downdelay < 0) {
  3826. printk(KERN_WARNING DRV_NAME
  3827. ": Warning: downdelay module parameter (%d), "
  3828. "not in range 0-%d, so it was reset to 0\n",
  3829. downdelay, INT_MAX);
  3830. downdelay = 0;
  3831. }
  3832. if ((use_carrier != 0) && (use_carrier != 1)) {
  3833. printk(KERN_WARNING DRV_NAME
  3834. ": Warning: use_carrier module parameter (%d), "
  3835. "not of valid value (0/1), so it was set to 1\n",
  3836. use_carrier);
  3837. use_carrier = 1;
  3838. }
  3839. /* reset values for 802.3ad */
  3840. if (bond_mode == BOND_MODE_8023AD) {
  3841. if (!miimon) {
  3842. printk(KERN_WARNING DRV_NAME
  3843. ": Warning: miimon must be specified, "
  3844. "otherwise bonding will not detect link "
  3845. "failure, speed and duplex which are "
  3846. "essential for 802.3ad operation\n");
  3847. printk(KERN_WARNING "Forcing miimon to 100msec\n");
  3848. miimon = 100;
  3849. }
  3850. }
  3851. /* reset values for TLB/ALB */
  3852. if ((bond_mode == BOND_MODE_TLB) ||
  3853. (bond_mode == BOND_MODE_ALB)) {
  3854. if (!miimon) {
  3855. printk(KERN_WARNING DRV_NAME
  3856. ": Warning: miimon must be specified, "
  3857. "otherwise bonding will not detect link "
  3858. "failure and link speed which are essential "
  3859. "for TLB/ALB load balancing\n");
  3860. printk(KERN_WARNING "Forcing miimon to 100msec\n");
  3861. miimon = 100;
  3862. }
  3863. }
  3864. if (bond_mode == BOND_MODE_ALB) {
  3865. printk(KERN_NOTICE DRV_NAME
  3866. ": In ALB mode you might experience client "
  3867. "disconnections upon reconnection of a link if the "
  3868. "bonding module updelay parameter (%d msec) is "
  3869. "incompatible with the forwarding delay time of the "
  3870. "switch\n",
  3871. updelay);
  3872. }
  3873. if (!miimon) {
  3874. if (updelay || downdelay) {
  3875. /* just warn the user the up/down delay will have
  3876. * no effect since miimon is zero...
  3877. */
  3878. printk(KERN_WARNING DRV_NAME
  3879. ": Warning: miimon module parameter not set "
  3880. "and updelay (%d) or downdelay (%d) module "
  3881. "parameter is set; updelay and downdelay have "
  3882. "no effect unless miimon is set\n",
  3883. updelay, downdelay);
  3884. }
  3885. } else {
  3886. /* don't allow arp monitoring */
  3887. if (arp_interval) {
  3888. printk(KERN_WARNING DRV_NAME
  3889. ": Warning: miimon (%d) and arp_interval (%d) "
  3890. "can't be used simultaneously, disabling ARP "
  3891. "monitoring\n",
  3892. miimon, arp_interval);
  3893. arp_interval = 0;
  3894. }
  3895. if ((updelay % miimon) != 0) {
  3896. printk(KERN_WARNING DRV_NAME
  3897. ": Warning: updelay (%d) is not a multiple "
  3898. "of miimon (%d), updelay rounded to %d ms\n",
  3899. updelay, miimon, (updelay / miimon) * miimon);
  3900. }
  3901. updelay /= miimon;
  3902. if ((downdelay % miimon) != 0) {
  3903. printk(KERN_WARNING DRV_NAME
  3904. ": Warning: downdelay (%d) is not a multiple "
  3905. "of miimon (%d), downdelay rounded to %d ms\n",
  3906. downdelay, miimon,
  3907. (downdelay / miimon) * miimon);
  3908. }
  3909. downdelay /= miimon;
  3910. }
  3911. if (arp_interval < 0) {
  3912. printk(KERN_WARNING DRV_NAME
  3913. ": Warning: arp_interval module parameter (%d) "
  3914. ", not in range 0-%d, so it was reset to %d\n",
  3915. arp_interval, INT_MAX, BOND_LINK_ARP_INTERV);
  3916. arp_interval = BOND_LINK_ARP_INTERV;
  3917. }
  3918. for (arp_ip_count = 0;
  3919. (arp_ip_count < BOND_MAX_ARP_TARGETS) && arp_ip_target[arp_ip_count];
  3920. arp_ip_count++) {
  3921. /* not complete check, but should be good enough to
  3922. catch mistakes */
  3923. if (!isdigit(arp_ip_target[arp_ip_count][0])) {
  3924. printk(KERN_WARNING DRV_NAME
  3925. ": Warning: bad arp_ip_target module parameter "
  3926. "(%s), ARP monitoring will not be performed\n",
  3927. arp_ip_target[arp_ip_count]);
  3928. arp_interval = 0;
  3929. } else {
  3930. u32 ip = in_aton(arp_ip_target[arp_ip_count]);
  3931. arp_target[arp_ip_count] = ip;
  3932. }
  3933. }
  3934. if (arp_interval && !arp_ip_count) {
  3935. /* don't allow arping if no arp_ip_target given... */
  3936. printk(KERN_WARNING DRV_NAME
  3937. ": Warning: arp_interval module parameter (%d) "
  3938. "specified without providing an arp_ip_target "
  3939. "parameter, arp_interval was reset to 0\n",
  3940. arp_interval);
  3941. arp_interval = 0;
  3942. }
  3943. if (miimon) {
  3944. printk(KERN_INFO DRV_NAME
  3945. ": MII link monitoring set to %d ms\n",
  3946. miimon);
  3947. } else if (arp_interval) {
  3948. int i;
  3949. printk(KERN_INFO DRV_NAME
  3950. ": ARP monitoring set to %d ms with %d target(s):",
  3951. arp_interval, arp_ip_count);
  3952. for (i = 0; i < arp_ip_count; i++)
  3953. printk (" %s", arp_ip_target[i]);
  3954. printk("\n");
  3955. } else {
  3956. /* miimon and arp_interval not set, we need one so things
  3957. * work as expected, see bonding.txt for details
  3958. */
  3959. printk(KERN_WARNING DRV_NAME
  3960. ": Warning: either miimon or arp_interval and "
  3961. "arp_ip_target module parameters must be specified, "
  3962. "otherwise bonding will not detect link failures! see "
  3963. "bonding.txt for details.\n");
  3964. }
  3965. if (primary && !USES_PRIMARY(bond_mode)) {
  3966. /* currently, using a primary only makes sense
  3967. * in active backup, TLB or ALB modes
  3968. */
  3969. printk(KERN_WARNING DRV_NAME
  3970. ": Warning: %s primary device specified but has no "
  3971. "effect in %s mode\n",
  3972. primary, bond_mode_name(bond_mode));
  3973. primary = NULL;
  3974. }
  3975. /* fill params struct with the proper values */
  3976. params->mode = bond_mode;
  3977. params->miimon = miimon;
  3978. params->arp_interval = arp_interval;
  3979. params->updelay = updelay;
  3980. params->downdelay = downdelay;
  3981. params->use_carrier = use_carrier;
  3982. params->lacp_fast = lacp_fast;
  3983. params->primary[0] = 0;
  3984. if (primary) {
  3985. strncpy(params->primary, primary, IFNAMSIZ);
  3986. params->primary[IFNAMSIZ - 1] = 0;
  3987. }
  3988. memcpy(params->arp_targets, arp_target, sizeof(arp_target));
  3989. return 0;
  3990. }
  3991. static int __init bonding_init(void)
  3992. {
  3993. struct bond_params params;
  3994. int i;
  3995. int res;
  3996. printk(KERN_INFO "%s", version);
  3997. res = bond_check_params(&params);
  3998. if (res) {
  3999. return res;
  4000. }
  4001. rtnl_lock();
  4002. #ifdef CONFIG_PROC_FS
  4003. bond_create_proc_dir();
  4004. #endif
  4005. for (i = 0; i < max_bonds; i++) {
  4006. struct net_device *bond_dev;
  4007. bond_dev = alloc_netdev(sizeof(struct bonding), "", ether_setup);
  4008. if (!bond_dev) {
  4009. res = -ENOMEM;
  4010. goto out_err;
  4011. }
  4012. res = dev_alloc_name(bond_dev, "bond%d");
  4013. if (res < 0) {
  4014. free_netdev(bond_dev);
  4015. goto out_err;
  4016. }
  4017. /* bond_init() must be called after dev_alloc_name() (for the
  4018. * /proc files), but before register_netdevice(), because we
  4019. * need to set function pointers.
  4020. */
  4021. res = bond_init(bond_dev, &params);
  4022. if (res < 0) {
  4023. free_netdev(bond_dev);
  4024. goto out_err;
  4025. }
  4026. SET_MODULE_OWNER(bond_dev);
  4027. res = register_netdevice(bond_dev);
  4028. if (res < 0) {
  4029. bond_deinit(bond_dev);
  4030. free_netdev(bond_dev);
  4031. goto out_err;
  4032. }
  4033. }
  4034. rtnl_unlock();
  4035. register_netdevice_notifier(&bond_netdev_notifier);
  4036. return 0;
  4037. out_err:
  4038. /* free and unregister all bonds that were successfully added */
  4039. bond_free_all();
  4040. rtnl_unlock();
  4041. return res;
  4042. }
  4043. static void __exit bonding_exit(void)
  4044. {
  4045. unregister_netdevice_notifier(&bond_netdev_notifier);
  4046. rtnl_lock();
  4047. bond_free_all();
  4048. rtnl_unlock();
  4049. }
  4050. module_init(bonding_init);
  4051. module_exit(bonding_exit);
  4052. MODULE_LICENSE("GPL");
  4053. MODULE_VERSION(DRV_VERSION);
  4054. MODULE_DESCRIPTION(DRV_DESCRIPTION ", v" DRV_VERSION);
  4055. MODULE_AUTHOR("Thomas Davis, tadavis@lbl.gov and many others");
  4056. MODULE_SUPPORTED_DEVICE("most ethernet devices");
  4057. /*
  4058. * Local variables:
  4059. * c-indent-level: 8
  4060. * c-basic-offset: 8
  4061. * tab-width: 8
  4062. * End:
  4063. */