bond_main.c 122 KB

1234567891011121314151617181920212223242526272829303132333435363738394041424344454647484950515253545556575859606162636465666768697071727374757677787980818283848586878889909192939495969798991001011021031041051061071081091101111121131141151161171181191201211221231241251261271281291301311321331341351361371381391401411421431441451461471481491501511521531541551561571581591601611621631641651661671681691701711721731741751761771781791801811821831841851861871881891901911921931941951961971981992002012022032042052062072082092102112122132142152162172182192202212222232242252262272282292302312322332342352362372382392402412422432442452462472482492502512522532542552562572582592602612622632642652662672682692702712722732742752762772782792802812822832842852862872882892902912922932942952962972982993003013023033043053063073083093103113123133143153163173183193203213223233243253263273283293303313323333343353363373383393403413423433443453463473483493503513523533543553563573583593603613623633643653663673683693703713723733743753763773783793803813823833843853863873883893903913923933943953963973983994004014024034044054064074084094104114124134144154164174184194204214224234244254264274284294304314324334344354364374384394404414424434444454464474484494504514524534544554564574584594604614624634644654664674684694704714724734744754764774784794804814824834844854864874884894904914924934944954964974984995005015025035045055065075085095105115125135145155165175185195205215225235245255265275285295305315325335345355365375385395405415425435445455465475485495505515525535545555565575585595605615625635645655665675685695705715725735745755765775785795805815825835845855865875885895905915925935945955965975985996006016026036046056066076086096106116126136146156166176186196206216226236246256266276286296306316326336346356366376386396406416426436446456466476486496506516526536546556566576586596606616626636646656666676686696706716726736746756766776786796806816826836846856866876886896906916926936946956966976986997007017027037047057067077087097107117127137147157167177187197207217227237247257267277287297307317327337347357367377387397407417427437447457467477487497507517527537547557567577587597607617627637647657667677687697707717727737747757767777787797807817827837847857867877887897907917927937947957967977987998008018028038048058068078088098108118128138148158168178188198208218228238248258268278288298308318328338348358368378388398408418428438448458468478488498508518528538548558568578588598608618628638648658668678688698708718728738748758768778788798808818828838848858868878888898908918928938948958968978988999009019029039049059069079089099109119129139149159169179189199209219229239249259269279289299309319329339349359369379389399409419429439449459469479489499509519529539549559569579589599609619629639649659669679689699709719729739749759769779789799809819829839849859869879889899909919929939949959969979989991000100110021003100410051006100710081009101010111012101310141015101610171018101910201021102210231024102510261027102810291030103110321033103410351036103710381039104010411042104310441045104610471048104910501051105210531054105510561057105810591060106110621063106410651066106710681069107010711072107310741075107610771078107910801081108210831084108510861087108810891090109110921093109410951096109710981099110011011102110311041105110611071108110911101111111211131114111511161117111811191120112111221123112411251126112711281129113011311132113311341135113611371138113911401141114211431144114511461147114811491150115111521153115411551156115711581159116011611162116311641165116611671168116911701171117211731174117511761177117811791180118111821183118411851186118711881189119011911192119311941195119611971198119912001201120212031204120512061207120812091210121112121213121412151216121712181219122012211222122312241225122612271228122912301231123212331234123512361237123812391240124112421243124412451246124712481249125012511252125312541255125612571258125912601261126212631264126512661267126812691270127112721273127412751276127712781279128012811282128312841285128612871288128912901291129212931294129512961297129812991300130113021303130413051306130713081309131013111312131313141315131613171318131913201321132213231324132513261327132813291330133113321333133413351336133713381339134013411342134313441345134613471348134913501351135213531354135513561357135813591360136113621363136413651366136713681369137013711372137313741375137613771378137913801381138213831384138513861387138813891390139113921393139413951396139713981399140014011402140314041405140614071408140914101411141214131414141514161417141814191420142114221423142414251426142714281429143014311432143314341435143614371438143914401441144214431444144514461447144814491450145114521453145414551456145714581459146014611462146314641465146614671468146914701471147214731474147514761477147814791480148114821483148414851486148714881489149014911492149314941495149614971498149915001501150215031504150515061507150815091510151115121513151415151516151715181519152015211522152315241525152615271528152915301531153215331534153515361537153815391540154115421543154415451546154715481549155015511552155315541555155615571558155915601561156215631564156515661567156815691570157115721573157415751576157715781579158015811582158315841585158615871588158915901591159215931594159515961597159815991600160116021603160416051606160716081609161016111612161316141615161616171618161916201621162216231624162516261627162816291630163116321633163416351636163716381639164016411642164316441645164616471648164916501651165216531654165516561657165816591660166116621663166416651666166716681669167016711672167316741675167616771678167916801681168216831684168516861687168816891690169116921693169416951696169716981699170017011702170317041705170617071708170917101711171217131714171517161717171817191720172117221723172417251726172717281729173017311732173317341735173617371738173917401741174217431744174517461747174817491750175117521753175417551756175717581759176017611762176317641765176617671768176917701771177217731774177517761777177817791780178117821783178417851786178717881789179017911792179317941795179617971798179918001801180218031804180518061807180818091810181118121813181418151816181718181819182018211822182318241825182618271828182918301831183218331834183518361837183818391840184118421843184418451846184718481849185018511852185318541855185618571858185918601861186218631864186518661867186818691870187118721873187418751876187718781879188018811882188318841885188618871888188918901891189218931894189518961897189818991900190119021903190419051906190719081909191019111912191319141915191619171918191919201921192219231924192519261927192819291930193119321933193419351936193719381939194019411942194319441945194619471948194919501951195219531954195519561957195819591960196119621963196419651966196719681969197019711972197319741975197619771978197919801981198219831984198519861987198819891990199119921993199419951996199719981999200020012002200320042005200620072008200920102011201220132014201520162017201820192020202120222023202420252026202720282029203020312032203320342035203620372038203920402041204220432044204520462047204820492050205120522053205420552056205720582059206020612062206320642065206620672068206920702071207220732074207520762077207820792080208120822083208420852086208720882089209020912092209320942095209620972098209921002101210221032104210521062107210821092110211121122113211421152116211721182119212021212122212321242125212621272128212921302131213221332134213521362137213821392140214121422143214421452146214721482149215021512152215321542155215621572158215921602161216221632164216521662167216821692170217121722173217421752176217721782179218021812182218321842185218621872188218921902191219221932194219521962197219821992200220122022203220422052206220722082209221022112212221322142215221622172218221922202221222222232224222522262227222822292230223122322233223422352236223722382239224022412242224322442245224622472248224922502251225222532254225522562257225822592260226122622263226422652266226722682269227022712272227322742275227622772278227922802281228222832284228522862287228822892290229122922293229422952296229722982299230023012302230323042305230623072308230923102311231223132314231523162317231823192320232123222323232423252326232723282329233023312332233323342335233623372338233923402341234223432344234523462347234823492350235123522353235423552356235723582359236023612362236323642365236623672368236923702371237223732374237523762377237823792380238123822383238423852386238723882389239023912392239323942395239623972398239924002401240224032404240524062407240824092410241124122413241424152416241724182419242024212422242324242425242624272428242924302431243224332434243524362437243824392440244124422443244424452446244724482449245024512452245324542455245624572458245924602461246224632464246524662467246824692470247124722473247424752476247724782479248024812482248324842485248624872488248924902491249224932494249524962497249824992500250125022503250425052506250725082509251025112512251325142515251625172518251925202521252225232524252525262527252825292530253125322533253425352536253725382539254025412542254325442545254625472548254925502551255225532554255525562557255825592560256125622563256425652566256725682569257025712572257325742575257625772578257925802581258225832584258525862587258825892590259125922593259425952596259725982599260026012602260326042605260626072608260926102611261226132614261526162617261826192620262126222623262426252626262726282629263026312632263326342635263626372638263926402641264226432644264526462647264826492650265126522653265426552656265726582659266026612662266326642665266626672668266926702671267226732674267526762677267826792680268126822683268426852686268726882689269026912692269326942695269626972698269927002701270227032704270527062707270827092710271127122713271427152716271727182719272027212722272327242725272627272728272927302731273227332734273527362737273827392740274127422743274427452746274727482749275027512752275327542755275627572758275927602761276227632764276527662767276827692770277127722773277427752776277727782779278027812782278327842785278627872788278927902791279227932794279527962797279827992800280128022803280428052806280728082809281028112812281328142815281628172818281928202821282228232824282528262827282828292830283128322833283428352836283728382839284028412842284328442845284628472848284928502851285228532854285528562857285828592860286128622863286428652866286728682869287028712872287328742875287628772878287928802881288228832884288528862887288828892890289128922893289428952896289728982899290029012902290329042905290629072908290929102911291229132914291529162917291829192920292129222923292429252926292729282929293029312932293329342935293629372938293929402941294229432944294529462947294829492950295129522953295429552956295729582959296029612962296329642965296629672968296929702971297229732974297529762977297829792980298129822983298429852986298729882989299029912992299329942995299629972998299930003001300230033004300530063007300830093010301130123013301430153016301730183019302030213022302330243025302630273028302930303031303230333034303530363037303830393040304130423043304430453046304730483049305030513052305330543055305630573058305930603061306230633064306530663067306830693070307130723073307430753076307730783079308030813082308330843085308630873088308930903091309230933094309530963097309830993100310131023103310431053106310731083109311031113112311331143115311631173118311931203121312231233124312531263127312831293130313131323133313431353136313731383139314031413142314331443145314631473148314931503151315231533154315531563157315831593160316131623163316431653166316731683169317031713172317331743175317631773178317931803181318231833184318531863187318831893190319131923193319431953196319731983199320032013202320332043205320632073208320932103211321232133214321532163217321832193220322132223223322432253226322732283229323032313232323332343235323632373238323932403241324232433244324532463247324832493250325132523253325432553256325732583259326032613262326332643265326632673268326932703271327232733274327532763277327832793280328132823283328432853286328732883289329032913292329332943295329632973298329933003301330233033304330533063307330833093310331133123313331433153316331733183319332033213322332333243325332633273328332933303331333233333334333533363337333833393340334133423343334433453346334733483349335033513352335333543355335633573358335933603361336233633364336533663367336833693370337133723373337433753376337733783379338033813382338333843385338633873388338933903391339233933394339533963397339833993400340134023403340434053406340734083409341034113412341334143415341634173418341934203421342234233424342534263427342834293430343134323433343434353436343734383439344034413442344334443445344634473448344934503451345234533454345534563457345834593460346134623463346434653466346734683469347034713472347334743475347634773478347934803481348234833484348534863487348834893490349134923493349434953496349734983499350035013502350335043505350635073508350935103511351235133514351535163517351835193520352135223523352435253526352735283529353035313532353335343535353635373538353935403541354235433544354535463547354835493550355135523553355435553556355735583559356035613562356335643565356635673568356935703571357235733574357535763577357835793580358135823583358435853586358735883589359035913592359335943595359635973598359936003601360236033604360536063607360836093610361136123613361436153616361736183619362036213622362336243625362636273628362936303631363236333634363536363637363836393640364136423643364436453646364736483649365036513652365336543655365636573658365936603661366236633664366536663667366836693670367136723673367436753676367736783679368036813682368336843685368636873688368936903691369236933694369536963697369836993700370137023703370437053706370737083709371037113712371337143715371637173718371937203721372237233724372537263727372837293730373137323733373437353736373737383739374037413742374337443745374637473748374937503751375237533754375537563757375837593760376137623763376437653766376737683769377037713772377337743775377637773778377937803781378237833784378537863787378837893790379137923793379437953796379737983799380038013802380338043805380638073808380938103811381238133814381538163817381838193820382138223823382438253826382738283829383038313832383338343835383638373838383938403841384238433844384538463847384838493850385138523853385438553856385738583859386038613862386338643865386638673868386938703871387238733874387538763877387838793880388138823883388438853886388738883889389038913892389338943895389638973898389939003901390239033904390539063907390839093910391139123913391439153916391739183919392039213922392339243925392639273928392939303931393239333934393539363937393839393940394139423943394439453946394739483949395039513952395339543955395639573958395939603961396239633964396539663967396839693970397139723973397439753976397739783979398039813982398339843985398639873988398939903991399239933994399539963997399839994000400140024003400440054006400740084009401040114012401340144015401640174018401940204021402240234024402540264027402840294030403140324033403440354036403740384039404040414042404340444045404640474048404940504051405240534054405540564057405840594060406140624063406440654066406740684069407040714072407340744075407640774078407940804081408240834084408540864087408840894090409140924093409440954096409740984099410041014102410341044105410641074108410941104111411241134114411541164117411841194120412141224123412441254126412741284129413041314132413341344135413641374138413941404141414241434144414541464147414841494150415141524153415441554156415741584159416041614162416341644165416641674168416941704171417241734174417541764177417841794180418141824183418441854186418741884189419041914192419341944195419641974198419942004201420242034204420542064207420842094210421142124213421442154216421742184219422042214222422342244225422642274228422942304231423242334234423542364237423842394240424142424243424442454246424742484249425042514252425342544255425642574258425942604261426242634264426542664267426842694270427142724273427442754276427742784279428042814282428342844285428642874288428942904291429242934294429542964297429842994300430143024303430443054306430743084309431043114312431343144315431643174318431943204321432243234324432543264327432843294330433143324333433443354336433743384339434043414342434343444345434643474348434943504351435243534354435543564357435843594360436143624363436443654366436743684369437043714372437343744375437643774378437943804381438243834384438543864387438843894390439143924393439443954396439743984399440044014402440344044405440644074408440944104411441244134414441544164417441844194420442144224423442444254426442744284429443044314432443344344435443644374438443944404441444244434444444544464447444844494450445144524453445444554456445744584459446044614462446344644465446644674468446944704471447244734474447544764477447844794480448144824483448444854486448744884489449044914492449344944495449644974498449945004501450245034504
  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. */
  33. #define pr_fmt(fmt) KBUILD_MODNAME ": " fmt
  34. #include <linux/kernel.h>
  35. #include <linux/module.h>
  36. #include <linux/types.h>
  37. #include <linux/fcntl.h>
  38. #include <linux/interrupt.h>
  39. #include <linux/ptrace.h>
  40. #include <linux/ioport.h>
  41. #include <linux/in.h>
  42. #include <net/ip.h>
  43. #include <linux/ip.h>
  44. #include <linux/tcp.h>
  45. #include <linux/udp.h>
  46. #include <linux/slab.h>
  47. #include <linux/string.h>
  48. #include <linux/init.h>
  49. #include <linux/timer.h>
  50. #include <linux/socket.h>
  51. #include <linux/ctype.h>
  52. #include <linux/inet.h>
  53. #include <linux/bitops.h>
  54. #include <linux/io.h>
  55. #include <asm/dma.h>
  56. #include <linux/uaccess.h>
  57. #include <linux/errno.h>
  58. #include <linux/netdevice.h>
  59. #include <linux/inetdevice.h>
  60. #include <linux/igmp.h>
  61. #include <linux/etherdevice.h>
  62. #include <linux/skbuff.h>
  63. #include <net/sock.h>
  64. #include <linux/rtnetlink.h>
  65. #include <linux/smp.h>
  66. #include <linux/if_ether.h>
  67. #include <net/arp.h>
  68. #include <linux/mii.h>
  69. #include <linux/ethtool.h>
  70. #include <linux/if_vlan.h>
  71. #include <linux/if_bonding.h>
  72. #include <linux/jiffies.h>
  73. #include <linux/preempt.h>
  74. #include <net/route.h>
  75. #include <net/net_namespace.h>
  76. #include <net/netns/generic.h>
  77. #include <net/pkt_sched.h>
  78. #include <linux/rculist.h>
  79. #include <net/flow_keys.h>
  80. #include <linux/reciprocal_div.h>
  81. #include "bonding.h"
  82. #include "bond_3ad.h"
  83. #include "bond_alb.h"
  84. /*---------------------------- Module parameters ----------------------------*/
  85. /* monitor all links that often (in milliseconds). <=0 disables monitoring */
  86. #define BOND_LINK_MON_INTERV 0
  87. #define BOND_LINK_ARP_INTERV 0
  88. static int max_bonds = BOND_DEFAULT_MAX_BONDS;
  89. static int tx_queues = BOND_DEFAULT_TX_QUEUES;
  90. static int num_peer_notif = 1;
  91. static int miimon = BOND_LINK_MON_INTERV;
  92. static int updelay;
  93. static int downdelay;
  94. static int use_carrier = 1;
  95. static char *mode;
  96. static char *primary;
  97. static char *primary_reselect;
  98. static char *lacp_rate;
  99. static int min_links;
  100. static char *ad_select;
  101. static char *xmit_hash_policy;
  102. static int arp_interval = BOND_LINK_ARP_INTERV;
  103. static char *arp_ip_target[BOND_MAX_ARP_TARGETS];
  104. static char *arp_validate;
  105. static char *arp_all_targets;
  106. static char *fail_over_mac;
  107. static int all_slaves_active;
  108. static struct bond_params bonding_defaults;
  109. static int resend_igmp = BOND_DEFAULT_RESEND_IGMP;
  110. static int packets_per_slave = 1;
  111. module_param(max_bonds, int, 0);
  112. MODULE_PARM_DESC(max_bonds, "Max number of bonded devices");
  113. module_param(tx_queues, int, 0);
  114. MODULE_PARM_DESC(tx_queues, "Max number of transmit queues (default = 16)");
  115. module_param_named(num_grat_arp, num_peer_notif, int, 0644);
  116. MODULE_PARM_DESC(num_grat_arp, "Number of peer notifications to send on "
  117. "failover event (alias of num_unsol_na)");
  118. module_param_named(num_unsol_na, num_peer_notif, int, 0644);
  119. MODULE_PARM_DESC(num_unsol_na, "Number of peer notifications to send on "
  120. "failover event (alias of num_grat_arp)");
  121. module_param(miimon, int, 0);
  122. MODULE_PARM_DESC(miimon, "Link check interval in milliseconds");
  123. module_param(updelay, int, 0);
  124. MODULE_PARM_DESC(updelay, "Delay before considering link up, in milliseconds");
  125. module_param(downdelay, int, 0);
  126. MODULE_PARM_DESC(downdelay, "Delay before considering link down, "
  127. "in milliseconds");
  128. module_param(use_carrier, int, 0);
  129. MODULE_PARM_DESC(use_carrier, "Use netif_carrier_ok (vs MII ioctls) in miimon; "
  130. "0 for off, 1 for on (default)");
  131. module_param(mode, charp, 0);
  132. MODULE_PARM_DESC(mode, "Mode of operation; 0 for balance-rr, "
  133. "1 for active-backup, 2 for balance-xor, "
  134. "3 for broadcast, 4 for 802.3ad, 5 for balance-tlb, "
  135. "6 for balance-alb");
  136. module_param(primary, charp, 0);
  137. MODULE_PARM_DESC(primary, "Primary network device to use");
  138. module_param(primary_reselect, charp, 0);
  139. MODULE_PARM_DESC(primary_reselect, "Reselect primary slave "
  140. "once it comes up; "
  141. "0 for always (default), "
  142. "1 for only if speed of primary is "
  143. "better, "
  144. "2 for only on active slave "
  145. "failure");
  146. module_param(lacp_rate, charp, 0);
  147. MODULE_PARM_DESC(lacp_rate, "LACPDU tx rate to request from 802.3ad partner; "
  148. "0 for slow, 1 for fast");
  149. module_param(ad_select, charp, 0);
  150. MODULE_PARM_DESC(ad_select, "803.ad aggregation selection logic; "
  151. "0 for stable (default), 1 for bandwidth, "
  152. "2 for count");
  153. module_param(min_links, int, 0);
  154. MODULE_PARM_DESC(min_links, "Minimum number of available links before turning on carrier");
  155. module_param(xmit_hash_policy, charp, 0);
  156. MODULE_PARM_DESC(xmit_hash_policy, "balance-xor and 802.3ad hashing method; "
  157. "0 for layer 2 (default), 1 for layer 3+4, "
  158. "2 for layer 2+3, 3 for encap layer 2+3, "
  159. "4 for encap layer 3+4");
  160. module_param(arp_interval, int, 0);
  161. MODULE_PARM_DESC(arp_interval, "arp interval in milliseconds");
  162. module_param_array(arp_ip_target, charp, NULL, 0);
  163. MODULE_PARM_DESC(arp_ip_target, "arp targets in n.n.n.n form");
  164. module_param(arp_validate, charp, 0);
  165. MODULE_PARM_DESC(arp_validate, "validate src/dst of ARP probes; "
  166. "0 for none (default), 1 for active, "
  167. "2 for backup, 3 for all");
  168. module_param(arp_all_targets, charp, 0);
  169. MODULE_PARM_DESC(arp_all_targets, "fail on any/all arp targets timeout; 0 for any (default), 1 for all");
  170. module_param(fail_over_mac, charp, 0);
  171. MODULE_PARM_DESC(fail_over_mac, "For active-backup, do not set all slaves to "
  172. "the same MAC; 0 for none (default), "
  173. "1 for active, 2 for follow");
  174. module_param(all_slaves_active, int, 0);
  175. MODULE_PARM_DESC(all_slaves_active, "Keep all frames received on an interface"
  176. "by setting active flag for all slaves; "
  177. "0 for never (default), 1 for always.");
  178. module_param(resend_igmp, int, 0);
  179. MODULE_PARM_DESC(resend_igmp, "Number of IGMP membership reports to send on "
  180. "link failure");
  181. module_param(packets_per_slave, int, 0);
  182. MODULE_PARM_DESC(packets_per_slave, "Packets to send per slave in balance-rr "
  183. "mode; 0 for a random slave, 1 packet per "
  184. "slave (default), >1 packets per slave.");
  185. /*----------------------------- Global variables ----------------------------*/
  186. #ifdef CONFIG_NET_POLL_CONTROLLER
  187. atomic_t netpoll_block_tx = ATOMIC_INIT(0);
  188. #endif
  189. int bond_net_id __read_mostly;
  190. static __be32 arp_target[BOND_MAX_ARP_TARGETS];
  191. static int arp_ip_count;
  192. static int bond_mode = BOND_MODE_ROUNDROBIN;
  193. static int xmit_hashtype = BOND_XMIT_POLICY_LAYER2;
  194. static int lacp_fast;
  195. const struct bond_parm_tbl bond_lacp_tbl[] = {
  196. { "slow", AD_LACP_SLOW},
  197. { "fast", AD_LACP_FAST},
  198. { NULL, -1},
  199. };
  200. const struct bond_parm_tbl bond_mode_tbl[] = {
  201. { "balance-rr", BOND_MODE_ROUNDROBIN},
  202. { "active-backup", BOND_MODE_ACTIVEBACKUP},
  203. { "balance-xor", BOND_MODE_XOR},
  204. { "broadcast", BOND_MODE_BROADCAST},
  205. { "802.3ad", BOND_MODE_8023AD},
  206. { "balance-tlb", BOND_MODE_TLB},
  207. { "balance-alb", BOND_MODE_ALB},
  208. { NULL, -1},
  209. };
  210. const struct bond_parm_tbl xmit_hashtype_tbl[] = {
  211. { "layer2", BOND_XMIT_POLICY_LAYER2},
  212. { "layer3+4", BOND_XMIT_POLICY_LAYER34},
  213. { "layer2+3", BOND_XMIT_POLICY_LAYER23},
  214. { "encap2+3", BOND_XMIT_POLICY_ENCAP23},
  215. { "encap3+4", BOND_XMIT_POLICY_ENCAP34},
  216. { NULL, -1},
  217. };
  218. const struct bond_parm_tbl arp_all_targets_tbl[] = {
  219. { "any", BOND_ARP_TARGETS_ANY},
  220. { "all", BOND_ARP_TARGETS_ALL},
  221. { NULL, -1},
  222. };
  223. const struct bond_parm_tbl arp_validate_tbl[] = {
  224. { "none", BOND_ARP_VALIDATE_NONE},
  225. { "active", BOND_ARP_VALIDATE_ACTIVE},
  226. { "backup", BOND_ARP_VALIDATE_BACKUP},
  227. { "all", BOND_ARP_VALIDATE_ALL},
  228. { NULL, -1},
  229. };
  230. const struct bond_parm_tbl fail_over_mac_tbl[] = {
  231. { "none", BOND_FOM_NONE},
  232. { "active", BOND_FOM_ACTIVE},
  233. { "follow", BOND_FOM_FOLLOW},
  234. { NULL, -1},
  235. };
  236. const struct bond_parm_tbl pri_reselect_tbl[] = {
  237. { "always", BOND_PRI_RESELECT_ALWAYS},
  238. { "better", BOND_PRI_RESELECT_BETTER},
  239. { "failure", BOND_PRI_RESELECT_FAILURE},
  240. { NULL, -1},
  241. };
  242. struct bond_parm_tbl ad_select_tbl[] = {
  243. { "stable", BOND_AD_STABLE},
  244. { "bandwidth", BOND_AD_BANDWIDTH},
  245. { "count", BOND_AD_COUNT},
  246. { NULL, -1},
  247. };
  248. /*-------------------------- Forward declarations ---------------------------*/
  249. static int bond_init(struct net_device *bond_dev);
  250. static void bond_uninit(struct net_device *bond_dev);
  251. /*---------------------------- General routines -----------------------------*/
  252. const char *bond_mode_name(int mode)
  253. {
  254. static const char *names[] = {
  255. [BOND_MODE_ROUNDROBIN] = "load balancing (round-robin)",
  256. [BOND_MODE_ACTIVEBACKUP] = "fault-tolerance (active-backup)",
  257. [BOND_MODE_XOR] = "load balancing (xor)",
  258. [BOND_MODE_BROADCAST] = "fault-tolerance (broadcast)",
  259. [BOND_MODE_8023AD] = "IEEE 802.3ad Dynamic link aggregation",
  260. [BOND_MODE_TLB] = "transmit load balancing",
  261. [BOND_MODE_ALB] = "adaptive load balancing",
  262. };
  263. if (mode < BOND_MODE_ROUNDROBIN || mode > BOND_MODE_ALB)
  264. return "unknown";
  265. return names[mode];
  266. }
  267. /*---------------------------------- VLAN -----------------------------------*/
  268. /**
  269. * bond_dev_queue_xmit - Prepare skb for xmit.
  270. *
  271. * @bond: bond device that got this skb for tx.
  272. * @skb: hw accel VLAN tagged skb to transmit
  273. * @slave_dev: slave that is supposed to xmit this skbuff
  274. */
  275. int bond_dev_queue_xmit(struct bonding *bond, struct sk_buff *skb,
  276. struct net_device *slave_dev)
  277. {
  278. skb->dev = slave_dev;
  279. BUILD_BUG_ON(sizeof(skb->queue_mapping) !=
  280. sizeof(qdisc_skb_cb(skb)->slave_dev_queue_mapping));
  281. skb->queue_mapping = qdisc_skb_cb(skb)->slave_dev_queue_mapping;
  282. if (unlikely(netpoll_tx_running(bond->dev)))
  283. bond_netpoll_send_skb(bond_get_slave_by_dev(bond, slave_dev), skb);
  284. else
  285. dev_queue_xmit(skb);
  286. return 0;
  287. }
  288. /*
  289. * In the following 2 functions, bond_vlan_rx_add_vid and bond_vlan_rx_kill_vid,
  290. * We don't protect the slave list iteration with a lock because:
  291. * a. This operation is performed in IOCTL context,
  292. * b. The operation is protected by the RTNL semaphore in the 8021q code,
  293. * c. Holding a lock with BH disabled while directly calling a base driver
  294. * entry point is generally a BAD idea.
  295. *
  296. * The design of synchronization/protection for this operation in the 8021q
  297. * module is good for one or more VLAN devices over a single physical device
  298. * and cannot be extended for a teaming solution like bonding, so there is a
  299. * potential race condition here where a net device from the vlan group might
  300. * be referenced (either by a base driver or the 8021q code) while it is being
  301. * removed from the system. However, it turns out we're not making matters
  302. * worse, and if it works for regular VLAN usage it will work here too.
  303. */
  304. /**
  305. * bond_vlan_rx_add_vid - Propagates adding an id to slaves
  306. * @bond_dev: bonding net device that got called
  307. * @vid: vlan id being added
  308. */
  309. static int bond_vlan_rx_add_vid(struct net_device *bond_dev,
  310. __be16 proto, u16 vid)
  311. {
  312. struct bonding *bond = netdev_priv(bond_dev);
  313. struct slave *slave, *rollback_slave;
  314. struct list_head *iter;
  315. int res;
  316. bond_for_each_slave(bond, slave, iter) {
  317. res = vlan_vid_add(slave->dev, proto, vid);
  318. if (res)
  319. goto unwind;
  320. }
  321. return 0;
  322. unwind:
  323. /* unwind to the slave that failed */
  324. bond_for_each_slave(bond, rollback_slave, iter) {
  325. if (rollback_slave == slave)
  326. break;
  327. vlan_vid_del(rollback_slave->dev, proto, vid);
  328. }
  329. return res;
  330. }
  331. /**
  332. * bond_vlan_rx_kill_vid - Propagates deleting an id to slaves
  333. * @bond_dev: bonding net device that got called
  334. * @vid: vlan id being removed
  335. */
  336. static int bond_vlan_rx_kill_vid(struct net_device *bond_dev,
  337. __be16 proto, u16 vid)
  338. {
  339. struct bonding *bond = netdev_priv(bond_dev);
  340. struct list_head *iter;
  341. struct slave *slave;
  342. bond_for_each_slave(bond, slave, iter)
  343. vlan_vid_del(slave->dev, proto, vid);
  344. if (bond_is_lb(bond))
  345. bond_alb_clear_vlan(bond, vid);
  346. return 0;
  347. }
  348. /*------------------------------- Link status -------------------------------*/
  349. /*
  350. * Set the carrier state for the master according to the state of its
  351. * slaves. If any slaves are up, the master is up. In 802.3ad mode,
  352. * do special 802.3ad magic.
  353. *
  354. * Returns zero if carrier state does not change, nonzero if it does.
  355. */
  356. static int bond_set_carrier(struct bonding *bond)
  357. {
  358. struct list_head *iter;
  359. struct slave *slave;
  360. if (!bond_has_slaves(bond))
  361. goto down;
  362. if (bond->params.mode == BOND_MODE_8023AD)
  363. return bond_3ad_set_carrier(bond);
  364. bond_for_each_slave(bond, slave, iter) {
  365. if (slave->link == BOND_LINK_UP) {
  366. if (!netif_carrier_ok(bond->dev)) {
  367. netif_carrier_on(bond->dev);
  368. return 1;
  369. }
  370. return 0;
  371. }
  372. }
  373. down:
  374. if (netif_carrier_ok(bond->dev)) {
  375. netif_carrier_off(bond->dev);
  376. return 1;
  377. }
  378. return 0;
  379. }
  380. /*
  381. * Get link speed and duplex from the slave's base driver
  382. * using ethtool. If for some reason the call fails or the
  383. * values are invalid, set speed and duplex to -1,
  384. * and return.
  385. */
  386. static void bond_update_speed_duplex(struct slave *slave)
  387. {
  388. struct net_device *slave_dev = slave->dev;
  389. struct ethtool_cmd ecmd;
  390. u32 slave_speed;
  391. int res;
  392. slave->speed = SPEED_UNKNOWN;
  393. slave->duplex = DUPLEX_UNKNOWN;
  394. res = __ethtool_get_settings(slave_dev, &ecmd);
  395. if (res < 0)
  396. return;
  397. slave_speed = ethtool_cmd_speed(&ecmd);
  398. if (slave_speed == 0 || slave_speed == ((__u32) -1))
  399. return;
  400. switch (ecmd.duplex) {
  401. case DUPLEX_FULL:
  402. case DUPLEX_HALF:
  403. break;
  404. default:
  405. return;
  406. }
  407. slave->speed = slave_speed;
  408. slave->duplex = ecmd.duplex;
  409. return;
  410. }
  411. /*
  412. * if <dev> supports MII link status reporting, check its link status.
  413. *
  414. * We either do MII/ETHTOOL ioctls, or check netif_carrier_ok(),
  415. * depending upon the setting of the use_carrier parameter.
  416. *
  417. * Return either BMSR_LSTATUS, meaning that the link is up (or we
  418. * can't tell and just pretend it is), or 0, meaning that the link is
  419. * down.
  420. *
  421. * If reporting is non-zero, instead of faking link up, return -1 if
  422. * both ETHTOOL and MII ioctls fail (meaning the device does not
  423. * support them). If use_carrier is set, return whatever it says.
  424. * It'd be nice if there was a good way to tell if a driver supports
  425. * netif_carrier, but there really isn't.
  426. */
  427. static int bond_check_dev_link(struct bonding *bond,
  428. struct net_device *slave_dev, int reporting)
  429. {
  430. const struct net_device_ops *slave_ops = slave_dev->netdev_ops;
  431. int (*ioctl)(struct net_device *, struct ifreq *, int);
  432. struct ifreq ifr;
  433. struct mii_ioctl_data *mii;
  434. if (!reporting && !netif_running(slave_dev))
  435. return 0;
  436. if (bond->params.use_carrier)
  437. return netif_carrier_ok(slave_dev) ? BMSR_LSTATUS : 0;
  438. /* Try to get link status using Ethtool first. */
  439. if (slave_dev->ethtool_ops->get_link)
  440. return slave_dev->ethtool_ops->get_link(slave_dev) ?
  441. BMSR_LSTATUS : 0;
  442. /* Ethtool can't be used, fallback to MII ioctls. */
  443. ioctl = slave_ops->ndo_do_ioctl;
  444. if (ioctl) {
  445. /* TODO: set pointer to correct ioctl on a per team member */
  446. /* bases to make this more efficient. that is, once */
  447. /* we determine the correct ioctl, we will always */
  448. /* call it and not the others for that team */
  449. /* member. */
  450. /*
  451. * We cannot assume that SIOCGMIIPHY will also read a
  452. * register; not all network drivers (e.g., e100)
  453. * support that.
  454. */
  455. /* Yes, the mii is overlaid on the ifreq.ifr_ifru */
  456. strncpy(ifr.ifr_name, slave_dev->name, IFNAMSIZ);
  457. mii = if_mii(&ifr);
  458. if (IOCTL(slave_dev, &ifr, SIOCGMIIPHY) == 0) {
  459. mii->reg_num = MII_BMSR;
  460. if (IOCTL(slave_dev, &ifr, SIOCGMIIREG) == 0)
  461. return mii->val_out & BMSR_LSTATUS;
  462. }
  463. }
  464. /*
  465. * If reporting, report that either there's no dev->do_ioctl,
  466. * or both SIOCGMIIREG and get_link failed (meaning that we
  467. * cannot report link status). If not reporting, pretend
  468. * we're ok.
  469. */
  470. return reporting ? -1 : BMSR_LSTATUS;
  471. }
  472. /*----------------------------- Multicast list ------------------------------*/
  473. /*
  474. * Push the promiscuity flag down to appropriate slaves
  475. */
  476. static int bond_set_promiscuity(struct bonding *bond, int inc)
  477. {
  478. struct list_head *iter;
  479. int err = 0;
  480. if (USES_PRIMARY(bond->params.mode)) {
  481. /* write lock already acquired */
  482. if (bond->curr_active_slave) {
  483. err = dev_set_promiscuity(bond->curr_active_slave->dev,
  484. inc);
  485. }
  486. } else {
  487. struct slave *slave;
  488. bond_for_each_slave(bond, slave, iter) {
  489. err = dev_set_promiscuity(slave->dev, inc);
  490. if (err)
  491. return err;
  492. }
  493. }
  494. return err;
  495. }
  496. /*
  497. * Push the allmulti flag down to all slaves
  498. */
  499. static int bond_set_allmulti(struct bonding *bond, int inc)
  500. {
  501. struct list_head *iter;
  502. int err = 0;
  503. if (USES_PRIMARY(bond->params.mode)) {
  504. /* write lock already acquired */
  505. if (bond->curr_active_slave) {
  506. err = dev_set_allmulti(bond->curr_active_slave->dev,
  507. inc);
  508. }
  509. } else {
  510. struct slave *slave;
  511. bond_for_each_slave(bond, slave, iter) {
  512. err = dev_set_allmulti(slave->dev, inc);
  513. if (err)
  514. return err;
  515. }
  516. }
  517. return err;
  518. }
  519. /*
  520. * Retrieve the list of registered multicast addresses for the bonding
  521. * device and retransmit an IGMP JOIN request to the current active
  522. * slave.
  523. */
  524. static void bond_resend_igmp_join_requests_delayed(struct work_struct *work)
  525. {
  526. struct bonding *bond = container_of(work, struct bonding,
  527. mcast_work.work);
  528. if (!rtnl_trylock()) {
  529. queue_delayed_work(bond->wq, &bond->mcast_work, 1);
  530. return;
  531. }
  532. call_netdevice_notifiers(NETDEV_RESEND_IGMP, bond->dev);
  533. if (bond->igmp_retrans > 1) {
  534. bond->igmp_retrans--;
  535. queue_delayed_work(bond->wq, &bond->mcast_work, HZ/5);
  536. }
  537. rtnl_unlock();
  538. }
  539. /* Flush bond's hardware addresses from slave
  540. */
  541. static void bond_hw_addr_flush(struct net_device *bond_dev,
  542. struct net_device *slave_dev)
  543. {
  544. struct bonding *bond = netdev_priv(bond_dev);
  545. dev_uc_unsync(slave_dev, bond_dev);
  546. dev_mc_unsync(slave_dev, bond_dev);
  547. if (bond->params.mode == BOND_MODE_8023AD) {
  548. /* del lacpdu mc addr from mc list */
  549. u8 lacpdu_multicast[ETH_ALEN] = MULTICAST_LACPDU_ADDR;
  550. dev_mc_del(slave_dev, lacpdu_multicast);
  551. }
  552. }
  553. /*--------------------------- Active slave change ---------------------------*/
  554. /* Update the hardware address list and promisc/allmulti for the new and
  555. * old active slaves (if any). Modes that are !USES_PRIMARY keep all
  556. * slaves up date at all times; only the USES_PRIMARY modes need to call
  557. * this function to swap these settings during a failover.
  558. */
  559. static void bond_hw_addr_swap(struct bonding *bond, struct slave *new_active,
  560. struct slave *old_active)
  561. {
  562. ASSERT_RTNL();
  563. if (old_active) {
  564. if (bond->dev->flags & IFF_PROMISC)
  565. dev_set_promiscuity(old_active->dev, -1);
  566. if (bond->dev->flags & IFF_ALLMULTI)
  567. dev_set_allmulti(old_active->dev, -1);
  568. bond_hw_addr_flush(bond->dev, old_active->dev);
  569. }
  570. if (new_active) {
  571. /* FIXME: Signal errors upstream. */
  572. if (bond->dev->flags & IFF_PROMISC)
  573. dev_set_promiscuity(new_active->dev, 1);
  574. if (bond->dev->flags & IFF_ALLMULTI)
  575. dev_set_allmulti(new_active->dev, 1);
  576. netif_addr_lock_bh(bond->dev);
  577. dev_uc_sync(new_active->dev, bond->dev);
  578. dev_mc_sync(new_active->dev, bond->dev);
  579. netif_addr_unlock_bh(bond->dev);
  580. }
  581. }
  582. /**
  583. * bond_set_dev_addr - clone slave's address to bond
  584. * @bond_dev: bond net device
  585. * @slave_dev: slave net device
  586. *
  587. * Should be called with RTNL held.
  588. */
  589. static void bond_set_dev_addr(struct net_device *bond_dev,
  590. struct net_device *slave_dev)
  591. {
  592. pr_debug("bond_dev=%p slave_dev=%p slave_dev->addr_len=%d\n",
  593. bond_dev, slave_dev, slave_dev->addr_len);
  594. memcpy(bond_dev->dev_addr, slave_dev->dev_addr, slave_dev->addr_len);
  595. bond_dev->addr_assign_type = NET_ADDR_STOLEN;
  596. call_netdevice_notifiers(NETDEV_CHANGEADDR, bond_dev);
  597. }
  598. /*
  599. * bond_do_fail_over_mac
  600. *
  601. * Perform special MAC address swapping for fail_over_mac settings
  602. *
  603. * Called with RTNL, curr_slave_lock for write_bh.
  604. */
  605. static void bond_do_fail_over_mac(struct bonding *bond,
  606. struct slave *new_active,
  607. struct slave *old_active)
  608. __releases(&bond->curr_slave_lock)
  609. __acquires(&bond->curr_slave_lock)
  610. {
  611. u8 tmp_mac[ETH_ALEN];
  612. struct sockaddr saddr;
  613. int rv;
  614. switch (bond->params.fail_over_mac) {
  615. case BOND_FOM_ACTIVE:
  616. if (new_active) {
  617. write_unlock_bh(&bond->curr_slave_lock);
  618. bond_set_dev_addr(bond->dev, new_active->dev);
  619. write_lock_bh(&bond->curr_slave_lock);
  620. }
  621. break;
  622. case BOND_FOM_FOLLOW:
  623. /*
  624. * if new_active && old_active, swap them
  625. * if just old_active, do nothing (going to no active slave)
  626. * if just new_active, set new_active to bond's MAC
  627. */
  628. if (!new_active)
  629. return;
  630. write_unlock_bh(&bond->curr_slave_lock);
  631. if (old_active) {
  632. memcpy(tmp_mac, new_active->dev->dev_addr, ETH_ALEN);
  633. memcpy(saddr.sa_data, old_active->dev->dev_addr,
  634. ETH_ALEN);
  635. saddr.sa_family = new_active->dev->type;
  636. } else {
  637. memcpy(saddr.sa_data, bond->dev->dev_addr, ETH_ALEN);
  638. saddr.sa_family = bond->dev->type;
  639. }
  640. rv = dev_set_mac_address(new_active->dev, &saddr);
  641. if (rv) {
  642. pr_err("%s: Error %d setting MAC of slave %s\n",
  643. bond->dev->name, -rv, new_active->dev->name);
  644. goto out;
  645. }
  646. if (!old_active)
  647. goto out;
  648. memcpy(saddr.sa_data, tmp_mac, ETH_ALEN);
  649. saddr.sa_family = old_active->dev->type;
  650. rv = dev_set_mac_address(old_active->dev, &saddr);
  651. if (rv)
  652. pr_err("%s: Error %d setting MAC of slave %s\n",
  653. bond->dev->name, -rv, new_active->dev->name);
  654. out:
  655. write_lock_bh(&bond->curr_slave_lock);
  656. break;
  657. default:
  658. pr_err("%s: bond_do_fail_over_mac impossible: bad policy %d\n",
  659. bond->dev->name, bond->params.fail_over_mac);
  660. break;
  661. }
  662. }
  663. static bool bond_should_change_active(struct bonding *bond)
  664. {
  665. struct slave *prim = bond->primary_slave;
  666. struct slave *curr = bond->curr_active_slave;
  667. if (!prim || !curr || curr->link != BOND_LINK_UP)
  668. return true;
  669. if (bond->force_primary) {
  670. bond->force_primary = false;
  671. return true;
  672. }
  673. if (bond->params.primary_reselect == BOND_PRI_RESELECT_BETTER &&
  674. (prim->speed < curr->speed ||
  675. (prim->speed == curr->speed && prim->duplex <= curr->duplex)))
  676. return false;
  677. if (bond->params.primary_reselect == BOND_PRI_RESELECT_FAILURE)
  678. return false;
  679. return true;
  680. }
  681. /**
  682. * find_best_interface - select the best available slave to be the active one
  683. * @bond: our bonding struct
  684. */
  685. static struct slave *bond_find_best_slave(struct bonding *bond)
  686. {
  687. struct slave *slave, *bestslave = NULL;
  688. struct list_head *iter;
  689. int mintime = bond->params.updelay;
  690. if (bond->primary_slave && bond->primary_slave->link == BOND_LINK_UP &&
  691. bond_should_change_active(bond))
  692. return bond->primary_slave;
  693. bond_for_each_slave(bond, slave, iter) {
  694. if (slave->link == BOND_LINK_UP)
  695. return slave;
  696. if (slave->link == BOND_LINK_BACK && IS_UP(slave->dev) &&
  697. slave->delay < mintime) {
  698. mintime = slave->delay;
  699. bestslave = slave;
  700. }
  701. }
  702. return bestslave;
  703. }
  704. static bool bond_should_notify_peers(struct bonding *bond)
  705. {
  706. struct slave *slave;
  707. rcu_read_lock();
  708. slave = rcu_dereference(bond->curr_active_slave);
  709. rcu_read_unlock();
  710. pr_debug("bond_should_notify_peers: bond %s slave %s\n",
  711. bond->dev->name, slave ? slave->dev->name : "NULL");
  712. if (!slave || !bond->send_peer_notif ||
  713. test_bit(__LINK_STATE_LINKWATCH_PENDING, &slave->dev->state))
  714. return false;
  715. return true;
  716. }
  717. /**
  718. * change_active_interface - change the active slave into the specified one
  719. * @bond: our bonding struct
  720. * @new: the new slave to make the active one
  721. *
  722. * Set the new slave to the bond's settings and unset them on the old
  723. * curr_active_slave.
  724. * Setting include flags, mc-list, promiscuity, allmulti, etc.
  725. *
  726. * If @new's link state is %BOND_LINK_BACK we'll set it to %BOND_LINK_UP,
  727. * because it is apparently the best available slave we have, even though its
  728. * updelay hasn't timed out yet.
  729. *
  730. * If new_active is not NULL, caller must hold curr_slave_lock for write_bh.
  731. */
  732. void bond_change_active_slave(struct bonding *bond, struct slave *new_active)
  733. {
  734. struct slave *old_active = bond->curr_active_slave;
  735. if (old_active == new_active)
  736. return;
  737. if (new_active) {
  738. new_active->jiffies = jiffies;
  739. if (new_active->link == BOND_LINK_BACK) {
  740. if (USES_PRIMARY(bond->params.mode)) {
  741. pr_info("%s: making interface %s the new active one %d ms earlier.\n",
  742. bond->dev->name, new_active->dev->name,
  743. (bond->params.updelay - new_active->delay) * bond->params.miimon);
  744. }
  745. new_active->delay = 0;
  746. new_active->link = BOND_LINK_UP;
  747. if (bond->params.mode == BOND_MODE_8023AD)
  748. bond_3ad_handle_link_change(new_active, BOND_LINK_UP);
  749. if (bond_is_lb(bond))
  750. bond_alb_handle_link_change(bond, new_active, BOND_LINK_UP);
  751. } else {
  752. if (USES_PRIMARY(bond->params.mode)) {
  753. pr_info("%s: making interface %s the new active one.\n",
  754. bond->dev->name, new_active->dev->name);
  755. }
  756. }
  757. }
  758. if (USES_PRIMARY(bond->params.mode))
  759. bond_hw_addr_swap(bond, new_active, old_active);
  760. if (bond_is_lb(bond)) {
  761. bond_alb_handle_active_change(bond, new_active);
  762. if (old_active)
  763. bond_set_slave_inactive_flags(old_active);
  764. if (new_active)
  765. bond_set_slave_active_flags(new_active);
  766. } else {
  767. rcu_assign_pointer(bond->curr_active_slave, new_active);
  768. }
  769. if (bond->params.mode == BOND_MODE_ACTIVEBACKUP) {
  770. if (old_active)
  771. bond_set_slave_inactive_flags(old_active);
  772. if (new_active) {
  773. bool should_notify_peers = false;
  774. bond_set_slave_active_flags(new_active);
  775. if (bond->params.fail_over_mac)
  776. bond_do_fail_over_mac(bond, new_active,
  777. old_active);
  778. if (netif_running(bond->dev)) {
  779. bond->send_peer_notif =
  780. bond->params.num_peer_notif;
  781. should_notify_peers =
  782. bond_should_notify_peers(bond);
  783. }
  784. write_unlock_bh(&bond->curr_slave_lock);
  785. call_netdevice_notifiers(NETDEV_BONDING_FAILOVER, bond->dev);
  786. if (should_notify_peers)
  787. call_netdevice_notifiers(NETDEV_NOTIFY_PEERS,
  788. bond->dev);
  789. write_lock_bh(&bond->curr_slave_lock);
  790. }
  791. }
  792. /* resend IGMP joins since active slave has changed or
  793. * all were sent on curr_active_slave.
  794. * resend only if bond is brought up with the affected
  795. * bonding modes and the retransmission is enabled */
  796. if (netif_running(bond->dev) && (bond->params.resend_igmp > 0) &&
  797. ((USES_PRIMARY(bond->params.mode) && new_active) ||
  798. bond->params.mode == BOND_MODE_ROUNDROBIN)) {
  799. bond->igmp_retrans = bond->params.resend_igmp;
  800. queue_delayed_work(bond->wq, &bond->mcast_work, 1);
  801. }
  802. }
  803. /**
  804. * bond_select_active_slave - select a new active slave, if needed
  805. * @bond: our bonding struct
  806. *
  807. * This functions should be called when one of the following occurs:
  808. * - The old curr_active_slave has been released or lost its link.
  809. * - The primary_slave has got its link back.
  810. * - A slave has got its link back and there's no old curr_active_slave.
  811. *
  812. * Caller must hold curr_slave_lock for write_bh.
  813. */
  814. void bond_select_active_slave(struct bonding *bond)
  815. {
  816. struct slave *best_slave;
  817. int rv;
  818. best_slave = bond_find_best_slave(bond);
  819. if (best_slave != bond->curr_active_slave) {
  820. bond_change_active_slave(bond, best_slave);
  821. rv = bond_set_carrier(bond);
  822. if (!rv)
  823. return;
  824. if (netif_carrier_ok(bond->dev)) {
  825. pr_info("%s: first active interface up!\n",
  826. bond->dev->name);
  827. } else {
  828. pr_info("%s: now running without any active interface !\n",
  829. bond->dev->name);
  830. }
  831. }
  832. }
  833. #ifdef CONFIG_NET_POLL_CONTROLLER
  834. static inline int slave_enable_netpoll(struct slave *slave)
  835. {
  836. struct netpoll *np;
  837. int err = 0;
  838. np = kzalloc(sizeof(*np), GFP_ATOMIC);
  839. err = -ENOMEM;
  840. if (!np)
  841. goto out;
  842. err = __netpoll_setup(np, slave->dev, GFP_ATOMIC);
  843. if (err) {
  844. kfree(np);
  845. goto out;
  846. }
  847. slave->np = np;
  848. out:
  849. return err;
  850. }
  851. static inline void slave_disable_netpoll(struct slave *slave)
  852. {
  853. struct netpoll *np = slave->np;
  854. if (!np)
  855. return;
  856. slave->np = NULL;
  857. __netpoll_free_async(np);
  858. }
  859. static inline bool slave_dev_support_netpoll(struct net_device *slave_dev)
  860. {
  861. if (slave_dev->priv_flags & IFF_DISABLE_NETPOLL)
  862. return false;
  863. if (!slave_dev->netdev_ops->ndo_poll_controller)
  864. return false;
  865. return true;
  866. }
  867. static void bond_poll_controller(struct net_device *bond_dev)
  868. {
  869. }
  870. static void bond_netpoll_cleanup(struct net_device *bond_dev)
  871. {
  872. struct bonding *bond = netdev_priv(bond_dev);
  873. struct list_head *iter;
  874. struct slave *slave;
  875. bond_for_each_slave(bond, slave, iter)
  876. if (IS_UP(slave->dev))
  877. slave_disable_netpoll(slave);
  878. }
  879. static int bond_netpoll_setup(struct net_device *dev, struct netpoll_info *ni, gfp_t gfp)
  880. {
  881. struct bonding *bond = netdev_priv(dev);
  882. struct list_head *iter;
  883. struct slave *slave;
  884. int err = 0;
  885. bond_for_each_slave(bond, slave, iter) {
  886. err = slave_enable_netpoll(slave);
  887. if (err) {
  888. bond_netpoll_cleanup(dev);
  889. break;
  890. }
  891. }
  892. return err;
  893. }
  894. #else
  895. static inline int slave_enable_netpoll(struct slave *slave)
  896. {
  897. return 0;
  898. }
  899. static inline void slave_disable_netpoll(struct slave *slave)
  900. {
  901. }
  902. static void bond_netpoll_cleanup(struct net_device *bond_dev)
  903. {
  904. }
  905. #endif
  906. /*---------------------------------- IOCTL ----------------------------------*/
  907. static netdev_features_t bond_fix_features(struct net_device *dev,
  908. netdev_features_t features)
  909. {
  910. struct bonding *bond = netdev_priv(dev);
  911. struct list_head *iter;
  912. netdev_features_t mask;
  913. struct slave *slave;
  914. if (!bond_has_slaves(bond)) {
  915. /* Disable adding VLANs to empty bond. But why? --mq */
  916. features |= NETIF_F_VLAN_CHALLENGED;
  917. return features;
  918. }
  919. mask = features;
  920. features &= ~NETIF_F_ONE_FOR_ALL;
  921. features |= NETIF_F_ALL_FOR_ALL;
  922. bond_for_each_slave(bond, slave, iter) {
  923. features = netdev_increment_features(features,
  924. slave->dev->features,
  925. mask);
  926. }
  927. features = netdev_add_tso_features(features, mask);
  928. return features;
  929. }
  930. #define BOND_VLAN_FEATURES (NETIF_F_ALL_CSUM | NETIF_F_SG | \
  931. NETIF_F_FRAGLIST | NETIF_F_ALL_TSO | \
  932. NETIF_F_HIGHDMA | NETIF_F_LRO)
  933. static void bond_compute_features(struct bonding *bond)
  934. {
  935. unsigned int flags, dst_release_flag = IFF_XMIT_DST_RELEASE;
  936. netdev_features_t vlan_features = BOND_VLAN_FEATURES;
  937. struct net_device *bond_dev = bond->dev;
  938. struct list_head *iter;
  939. struct slave *slave;
  940. unsigned short max_hard_header_len = ETH_HLEN;
  941. unsigned int gso_max_size = GSO_MAX_SIZE;
  942. u16 gso_max_segs = GSO_MAX_SEGS;
  943. if (!bond_has_slaves(bond))
  944. goto done;
  945. bond_for_each_slave(bond, slave, iter) {
  946. vlan_features = netdev_increment_features(vlan_features,
  947. slave->dev->vlan_features, BOND_VLAN_FEATURES);
  948. dst_release_flag &= slave->dev->priv_flags;
  949. if (slave->dev->hard_header_len > max_hard_header_len)
  950. max_hard_header_len = slave->dev->hard_header_len;
  951. gso_max_size = min(gso_max_size, slave->dev->gso_max_size);
  952. gso_max_segs = min(gso_max_segs, slave->dev->gso_max_segs);
  953. }
  954. done:
  955. bond_dev->vlan_features = vlan_features;
  956. bond_dev->hard_header_len = max_hard_header_len;
  957. bond_dev->gso_max_segs = gso_max_segs;
  958. netif_set_gso_max_size(bond_dev, gso_max_size);
  959. flags = bond_dev->priv_flags & ~IFF_XMIT_DST_RELEASE;
  960. bond_dev->priv_flags = flags | dst_release_flag;
  961. netdev_change_features(bond_dev);
  962. }
  963. static void bond_setup_by_slave(struct net_device *bond_dev,
  964. struct net_device *slave_dev)
  965. {
  966. bond_dev->header_ops = slave_dev->header_ops;
  967. bond_dev->type = slave_dev->type;
  968. bond_dev->hard_header_len = slave_dev->hard_header_len;
  969. bond_dev->addr_len = slave_dev->addr_len;
  970. memcpy(bond_dev->broadcast, slave_dev->broadcast,
  971. slave_dev->addr_len);
  972. }
  973. /* On bonding slaves other than the currently active slave, suppress
  974. * duplicates except for alb non-mcast/bcast.
  975. */
  976. static bool bond_should_deliver_exact_match(struct sk_buff *skb,
  977. struct slave *slave,
  978. struct bonding *bond)
  979. {
  980. if (bond_is_slave_inactive(slave)) {
  981. if (bond->params.mode == BOND_MODE_ALB &&
  982. skb->pkt_type != PACKET_BROADCAST &&
  983. skb->pkt_type != PACKET_MULTICAST)
  984. return false;
  985. return true;
  986. }
  987. return false;
  988. }
  989. static rx_handler_result_t bond_handle_frame(struct sk_buff **pskb)
  990. {
  991. struct sk_buff *skb = *pskb;
  992. struct slave *slave;
  993. struct bonding *bond;
  994. int (*recv_probe)(const struct sk_buff *, struct bonding *,
  995. struct slave *);
  996. int ret = RX_HANDLER_ANOTHER;
  997. skb = skb_share_check(skb, GFP_ATOMIC);
  998. if (unlikely(!skb))
  999. return RX_HANDLER_CONSUMED;
  1000. *pskb = skb;
  1001. slave = bond_slave_get_rcu(skb->dev);
  1002. bond = slave->bond;
  1003. if (bond->params.arp_interval)
  1004. slave->dev->last_rx = jiffies;
  1005. recv_probe = ACCESS_ONCE(bond->recv_probe);
  1006. if (recv_probe) {
  1007. ret = recv_probe(skb, bond, slave);
  1008. if (ret == RX_HANDLER_CONSUMED) {
  1009. consume_skb(skb);
  1010. return ret;
  1011. }
  1012. }
  1013. if (bond_should_deliver_exact_match(skb, slave, bond)) {
  1014. return RX_HANDLER_EXACT;
  1015. }
  1016. skb->dev = bond->dev;
  1017. if (bond->params.mode == BOND_MODE_ALB &&
  1018. bond->dev->priv_flags & IFF_BRIDGE_PORT &&
  1019. skb->pkt_type == PACKET_HOST) {
  1020. if (unlikely(skb_cow_head(skb,
  1021. skb->data - skb_mac_header(skb)))) {
  1022. kfree_skb(skb);
  1023. return RX_HANDLER_CONSUMED;
  1024. }
  1025. memcpy(eth_hdr(skb)->h_dest, bond->dev->dev_addr, ETH_ALEN);
  1026. }
  1027. return ret;
  1028. }
  1029. static int bond_master_upper_dev_link(struct net_device *bond_dev,
  1030. struct net_device *slave_dev,
  1031. struct slave *slave)
  1032. {
  1033. int err;
  1034. err = netdev_master_upper_dev_link_private(slave_dev, bond_dev, slave);
  1035. if (err)
  1036. return err;
  1037. slave_dev->flags |= IFF_SLAVE;
  1038. rtmsg_ifinfo(RTM_NEWLINK, slave_dev, IFF_SLAVE, GFP_KERNEL);
  1039. return 0;
  1040. }
  1041. static void bond_upper_dev_unlink(struct net_device *bond_dev,
  1042. struct net_device *slave_dev)
  1043. {
  1044. netdev_upper_dev_unlink(slave_dev, bond_dev);
  1045. slave_dev->flags &= ~IFF_SLAVE;
  1046. rtmsg_ifinfo(RTM_NEWLINK, slave_dev, IFF_SLAVE, GFP_KERNEL);
  1047. }
  1048. /* enslave device <slave> to bond device <master> */
  1049. int bond_enslave(struct net_device *bond_dev, struct net_device *slave_dev)
  1050. {
  1051. struct bonding *bond = netdev_priv(bond_dev);
  1052. const struct net_device_ops *slave_ops = slave_dev->netdev_ops;
  1053. struct slave *new_slave = NULL, *prev_slave;
  1054. struct sockaddr addr;
  1055. int link_reporting;
  1056. int res = 0, i;
  1057. if (!bond->params.use_carrier &&
  1058. slave_dev->ethtool_ops->get_link == NULL &&
  1059. slave_ops->ndo_do_ioctl == NULL) {
  1060. pr_warning("%s: Warning: no link monitoring support for %s\n",
  1061. bond_dev->name, slave_dev->name);
  1062. }
  1063. /* already enslaved */
  1064. if (slave_dev->flags & IFF_SLAVE) {
  1065. pr_debug("Error, Device was already enslaved\n");
  1066. return -EBUSY;
  1067. }
  1068. /* vlan challenged mutual exclusion */
  1069. /* no need to lock since we're protected by rtnl_lock */
  1070. if (slave_dev->features & NETIF_F_VLAN_CHALLENGED) {
  1071. pr_debug("%s: NETIF_F_VLAN_CHALLENGED\n", slave_dev->name);
  1072. if (vlan_uses_dev(bond_dev)) {
  1073. pr_err("%s: Error: cannot enslave VLAN challenged slave %s on VLAN enabled bond %s\n",
  1074. bond_dev->name, slave_dev->name, bond_dev->name);
  1075. return -EPERM;
  1076. } else {
  1077. pr_warning("%s: Warning: enslaved VLAN challenged slave %s. Adding VLANs will be blocked as long as %s is part of bond %s\n",
  1078. bond_dev->name, slave_dev->name,
  1079. slave_dev->name, bond_dev->name);
  1080. }
  1081. } else {
  1082. pr_debug("%s: ! NETIF_F_VLAN_CHALLENGED\n", slave_dev->name);
  1083. }
  1084. /*
  1085. * Old ifenslave binaries are no longer supported. These can
  1086. * be identified with moderate accuracy by the state of the slave:
  1087. * the current ifenslave will set the interface down prior to
  1088. * enslaving it; the old ifenslave will not.
  1089. */
  1090. if ((slave_dev->flags & IFF_UP)) {
  1091. pr_err("%s is up. This may be due to an out of date ifenslave.\n",
  1092. slave_dev->name);
  1093. res = -EPERM;
  1094. goto err_undo_flags;
  1095. }
  1096. /* set bonding device ether type by slave - bonding netdevices are
  1097. * created with ether_setup, so when the slave type is not ARPHRD_ETHER
  1098. * there is a need to override some of the type dependent attribs/funcs.
  1099. *
  1100. * bond ether type mutual exclusion - don't allow slaves of dissimilar
  1101. * ether type (eg ARPHRD_ETHER and ARPHRD_INFINIBAND) share the same bond
  1102. */
  1103. if (!bond_has_slaves(bond)) {
  1104. if (bond_dev->type != slave_dev->type) {
  1105. pr_debug("%s: change device type from %d to %d\n",
  1106. bond_dev->name,
  1107. bond_dev->type, slave_dev->type);
  1108. res = call_netdevice_notifiers(NETDEV_PRE_TYPE_CHANGE,
  1109. bond_dev);
  1110. res = notifier_to_errno(res);
  1111. if (res) {
  1112. pr_err("%s: refused to change device type\n",
  1113. bond_dev->name);
  1114. res = -EBUSY;
  1115. goto err_undo_flags;
  1116. }
  1117. /* Flush unicast and multicast addresses */
  1118. dev_uc_flush(bond_dev);
  1119. dev_mc_flush(bond_dev);
  1120. if (slave_dev->type != ARPHRD_ETHER)
  1121. bond_setup_by_slave(bond_dev, slave_dev);
  1122. else {
  1123. ether_setup(bond_dev);
  1124. bond_dev->priv_flags &= ~IFF_TX_SKB_SHARING;
  1125. }
  1126. call_netdevice_notifiers(NETDEV_POST_TYPE_CHANGE,
  1127. bond_dev);
  1128. }
  1129. } else if (bond_dev->type != slave_dev->type) {
  1130. pr_err("%s ether type (%d) is different from other slaves (%d), can not enslave it.\n",
  1131. slave_dev->name,
  1132. slave_dev->type, bond_dev->type);
  1133. res = -EINVAL;
  1134. goto err_undo_flags;
  1135. }
  1136. if (slave_ops->ndo_set_mac_address == NULL) {
  1137. if (!bond_has_slaves(bond)) {
  1138. pr_warning("%s: Warning: The first slave device specified does not support setting the MAC address. Setting fail_over_mac to active.",
  1139. bond_dev->name);
  1140. bond->params.fail_over_mac = BOND_FOM_ACTIVE;
  1141. } else if (bond->params.fail_over_mac != BOND_FOM_ACTIVE) {
  1142. pr_err("%s: Error: The slave device specified does not support setting the MAC address, but fail_over_mac is not set to active.\n",
  1143. bond_dev->name);
  1144. res = -EOPNOTSUPP;
  1145. goto err_undo_flags;
  1146. }
  1147. }
  1148. call_netdevice_notifiers(NETDEV_JOIN, slave_dev);
  1149. /* If this is the first slave, then we need to set the master's hardware
  1150. * address to be the same as the slave's. */
  1151. if (!bond_has_slaves(bond) &&
  1152. bond->dev->addr_assign_type == NET_ADDR_RANDOM)
  1153. bond_set_dev_addr(bond->dev, slave_dev);
  1154. new_slave = kzalloc(sizeof(struct slave), GFP_KERNEL);
  1155. if (!new_slave) {
  1156. res = -ENOMEM;
  1157. goto err_undo_flags;
  1158. }
  1159. /*
  1160. * Set the new_slave's queue_id to be zero. Queue ID mapping
  1161. * is set via sysfs or module option if desired.
  1162. */
  1163. new_slave->queue_id = 0;
  1164. /* Save slave's original mtu and then set it to match the bond */
  1165. new_slave->original_mtu = slave_dev->mtu;
  1166. res = dev_set_mtu(slave_dev, bond->dev->mtu);
  1167. if (res) {
  1168. pr_debug("Error %d calling dev_set_mtu\n", res);
  1169. goto err_free;
  1170. }
  1171. /*
  1172. * Save slave's original ("permanent") mac address for modes
  1173. * that need it, and for restoring it upon release, and then
  1174. * set it to the master's address
  1175. */
  1176. memcpy(new_slave->perm_hwaddr, slave_dev->dev_addr, ETH_ALEN);
  1177. if (!bond->params.fail_over_mac) {
  1178. /*
  1179. * Set slave to master's mac address. The application already
  1180. * set the master's mac address to that of the first slave
  1181. */
  1182. memcpy(addr.sa_data, bond_dev->dev_addr, bond_dev->addr_len);
  1183. addr.sa_family = slave_dev->type;
  1184. res = dev_set_mac_address(slave_dev, &addr);
  1185. if (res) {
  1186. pr_debug("Error %d calling set_mac_address\n", res);
  1187. goto err_restore_mtu;
  1188. }
  1189. }
  1190. /* open the slave since the application closed it */
  1191. res = dev_open(slave_dev);
  1192. if (res) {
  1193. pr_debug("Opening slave %s failed\n", slave_dev->name);
  1194. goto err_restore_mac;
  1195. }
  1196. new_slave->bond = bond;
  1197. new_slave->dev = slave_dev;
  1198. slave_dev->priv_flags |= IFF_BONDING;
  1199. if (bond_is_lb(bond)) {
  1200. /* bond_alb_init_slave() must be called before all other stages since
  1201. * it might fail and we do not want to have to undo everything
  1202. */
  1203. res = bond_alb_init_slave(bond, new_slave);
  1204. if (res)
  1205. goto err_close;
  1206. }
  1207. /* If the mode USES_PRIMARY, then the following is handled by
  1208. * bond_change_active_slave().
  1209. */
  1210. if (!USES_PRIMARY(bond->params.mode)) {
  1211. /* set promiscuity level to new slave */
  1212. if (bond_dev->flags & IFF_PROMISC) {
  1213. res = dev_set_promiscuity(slave_dev, 1);
  1214. if (res)
  1215. goto err_close;
  1216. }
  1217. /* set allmulti level to new slave */
  1218. if (bond_dev->flags & IFF_ALLMULTI) {
  1219. res = dev_set_allmulti(slave_dev, 1);
  1220. if (res)
  1221. goto err_close;
  1222. }
  1223. netif_addr_lock_bh(bond_dev);
  1224. dev_mc_sync_multiple(slave_dev, bond_dev);
  1225. dev_uc_sync_multiple(slave_dev, bond_dev);
  1226. netif_addr_unlock_bh(bond_dev);
  1227. }
  1228. if (bond->params.mode == BOND_MODE_8023AD) {
  1229. /* add lacpdu mc addr to mc list */
  1230. u8 lacpdu_multicast[ETH_ALEN] = MULTICAST_LACPDU_ADDR;
  1231. dev_mc_add(slave_dev, lacpdu_multicast);
  1232. }
  1233. res = vlan_vids_add_by_dev(slave_dev, bond_dev);
  1234. if (res) {
  1235. pr_err("%s: Error: Couldn't add bond vlan ids to %s\n",
  1236. bond_dev->name, slave_dev->name);
  1237. goto err_close;
  1238. }
  1239. prev_slave = bond_last_slave(bond);
  1240. new_slave->delay = 0;
  1241. new_slave->link_failure_count = 0;
  1242. bond_update_speed_duplex(new_slave);
  1243. new_slave->last_arp_rx = jiffies -
  1244. (msecs_to_jiffies(bond->params.arp_interval) + 1);
  1245. for (i = 0; i < BOND_MAX_ARP_TARGETS; i++)
  1246. new_slave->target_last_arp_rx[i] = new_slave->last_arp_rx;
  1247. if (bond->params.miimon && !bond->params.use_carrier) {
  1248. link_reporting = bond_check_dev_link(bond, slave_dev, 1);
  1249. if ((link_reporting == -1) && !bond->params.arp_interval) {
  1250. /*
  1251. * miimon is set but a bonded network driver
  1252. * does not support ETHTOOL/MII and
  1253. * arp_interval is not set. Note: if
  1254. * use_carrier is enabled, we will never go
  1255. * here (because netif_carrier is always
  1256. * supported); thus, we don't need to change
  1257. * the messages for netif_carrier.
  1258. */
  1259. pr_warning("%s: Warning: MII and ETHTOOL support not available for interface %s, and arp_interval/arp_ip_target module parameters not specified, thus bonding will not detect link failures! see bonding.txt for details.\n",
  1260. bond_dev->name, slave_dev->name);
  1261. } else if (link_reporting == -1) {
  1262. /* unable get link status using mii/ethtool */
  1263. pr_warning("%s: Warning: can't get link status from interface %s; the network driver associated with this interface does not support MII or ETHTOOL link status reporting, thus miimon has no effect on this interface.\n",
  1264. bond_dev->name, slave_dev->name);
  1265. }
  1266. }
  1267. /* check for initial state */
  1268. if (bond->params.miimon) {
  1269. if (bond_check_dev_link(bond, slave_dev, 0) == BMSR_LSTATUS) {
  1270. if (bond->params.updelay) {
  1271. new_slave->link = BOND_LINK_BACK;
  1272. new_slave->delay = bond->params.updelay;
  1273. } else {
  1274. new_slave->link = BOND_LINK_UP;
  1275. }
  1276. } else {
  1277. new_slave->link = BOND_LINK_DOWN;
  1278. }
  1279. } else if (bond->params.arp_interval) {
  1280. new_slave->link = (netif_carrier_ok(slave_dev) ?
  1281. BOND_LINK_UP : BOND_LINK_DOWN);
  1282. } else {
  1283. new_slave->link = BOND_LINK_UP;
  1284. }
  1285. if (new_slave->link != BOND_LINK_DOWN)
  1286. new_slave->jiffies = jiffies;
  1287. pr_debug("Initial state of slave_dev is BOND_LINK_%s\n",
  1288. new_slave->link == BOND_LINK_DOWN ? "DOWN" :
  1289. (new_slave->link == BOND_LINK_UP ? "UP" : "BACK"));
  1290. if (USES_PRIMARY(bond->params.mode) && bond->params.primary[0]) {
  1291. /* if there is a primary slave, remember it */
  1292. if (strcmp(bond->params.primary, new_slave->dev->name) == 0) {
  1293. bond->primary_slave = new_slave;
  1294. bond->force_primary = true;
  1295. }
  1296. }
  1297. switch (bond->params.mode) {
  1298. case BOND_MODE_ACTIVEBACKUP:
  1299. bond_set_slave_inactive_flags(new_slave);
  1300. break;
  1301. case BOND_MODE_8023AD:
  1302. /* in 802.3ad mode, the internal mechanism
  1303. * will activate the slaves in the selected
  1304. * aggregator
  1305. */
  1306. bond_set_slave_inactive_flags(new_slave);
  1307. /* if this is the first slave */
  1308. if (!prev_slave) {
  1309. SLAVE_AD_INFO(new_slave).id = 1;
  1310. /* Initialize AD with the number of times that the AD timer is called in 1 second
  1311. * can be called only after the mac address of the bond is set
  1312. */
  1313. bond_3ad_initialize(bond, 1000/AD_TIMER_INTERVAL);
  1314. } else {
  1315. SLAVE_AD_INFO(new_slave).id =
  1316. SLAVE_AD_INFO(prev_slave).id + 1;
  1317. }
  1318. bond_3ad_bind_slave(new_slave);
  1319. break;
  1320. case BOND_MODE_TLB:
  1321. case BOND_MODE_ALB:
  1322. bond_set_active_slave(new_slave);
  1323. bond_set_slave_inactive_flags(new_slave);
  1324. break;
  1325. default:
  1326. pr_debug("This slave is always active in trunk mode\n");
  1327. /* always active in trunk mode */
  1328. bond_set_active_slave(new_slave);
  1329. /* In trunking mode there is little meaning to curr_active_slave
  1330. * anyway (it holds no special properties of the bond device),
  1331. * so we can change it without calling change_active_interface()
  1332. */
  1333. if (!bond->curr_active_slave && new_slave->link == BOND_LINK_UP)
  1334. rcu_assign_pointer(bond->curr_active_slave, new_slave);
  1335. break;
  1336. } /* switch(bond_mode) */
  1337. #ifdef CONFIG_NET_POLL_CONTROLLER
  1338. slave_dev->npinfo = bond->dev->npinfo;
  1339. if (slave_dev->npinfo) {
  1340. if (slave_enable_netpoll(new_slave)) {
  1341. read_unlock(&bond->lock);
  1342. pr_info("Error, %s: master_dev is using netpoll, "
  1343. "but new slave device does not support netpoll.\n",
  1344. bond_dev->name);
  1345. res = -EBUSY;
  1346. goto err_detach;
  1347. }
  1348. }
  1349. #endif
  1350. res = netdev_rx_handler_register(slave_dev, bond_handle_frame,
  1351. new_slave);
  1352. if (res) {
  1353. pr_debug("Error %d calling netdev_rx_handler_register\n", res);
  1354. goto err_detach;
  1355. }
  1356. res = bond_master_upper_dev_link(bond_dev, slave_dev, new_slave);
  1357. if (res) {
  1358. pr_debug("Error %d calling bond_master_upper_dev_link\n", res);
  1359. goto err_unregister;
  1360. }
  1361. bond->slave_cnt++;
  1362. bond_compute_features(bond);
  1363. bond_set_carrier(bond);
  1364. if (USES_PRIMARY(bond->params.mode)) {
  1365. write_lock_bh(&bond->curr_slave_lock);
  1366. bond_select_active_slave(bond);
  1367. write_unlock_bh(&bond->curr_slave_lock);
  1368. }
  1369. pr_info("%s: enslaving %s as a%s interface with a%s link.\n",
  1370. bond_dev->name, slave_dev->name,
  1371. bond_is_active_slave(new_slave) ? "n active" : " backup",
  1372. new_slave->link != BOND_LINK_DOWN ? "n up" : " down");
  1373. /* enslave is successful */
  1374. return 0;
  1375. /* Undo stages on error */
  1376. err_unregister:
  1377. netdev_rx_handler_unregister(slave_dev);
  1378. err_detach:
  1379. if (!USES_PRIMARY(bond->params.mode))
  1380. bond_hw_addr_flush(bond_dev, slave_dev);
  1381. vlan_vids_del_by_dev(slave_dev, bond_dev);
  1382. if (bond->primary_slave == new_slave)
  1383. bond->primary_slave = NULL;
  1384. if (bond->curr_active_slave == new_slave) {
  1385. write_lock_bh(&bond->curr_slave_lock);
  1386. bond_change_active_slave(bond, NULL);
  1387. bond_select_active_slave(bond);
  1388. write_unlock_bh(&bond->curr_slave_lock);
  1389. }
  1390. slave_disable_netpoll(new_slave);
  1391. err_close:
  1392. slave_dev->priv_flags &= ~IFF_BONDING;
  1393. dev_close(slave_dev);
  1394. err_restore_mac:
  1395. if (!bond->params.fail_over_mac) {
  1396. /* XXX TODO - fom follow mode needs to change master's
  1397. * MAC if this slave's MAC is in use by the bond, or at
  1398. * least print a warning.
  1399. */
  1400. memcpy(addr.sa_data, new_slave->perm_hwaddr, ETH_ALEN);
  1401. addr.sa_family = slave_dev->type;
  1402. dev_set_mac_address(slave_dev, &addr);
  1403. }
  1404. err_restore_mtu:
  1405. dev_set_mtu(slave_dev, new_slave->original_mtu);
  1406. err_free:
  1407. kfree(new_slave);
  1408. err_undo_flags:
  1409. /* Enslave of first slave has failed and we need to fix master's mac */
  1410. if (!bond_has_slaves(bond) &&
  1411. ether_addr_equal(bond_dev->dev_addr, slave_dev->dev_addr))
  1412. eth_hw_addr_random(bond_dev);
  1413. return res;
  1414. }
  1415. /*
  1416. * Try to release the slave device <slave> from the bond device <master>
  1417. * It is legal to access curr_active_slave without a lock because all the function
  1418. * is write-locked. If "all" is true it means that the function is being called
  1419. * while destroying a bond interface and all slaves are being released.
  1420. *
  1421. * The rules for slave state should be:
  1422. * for Active/Backup:
  1423. * Active stays on all backups go down
  1424. * for Bonded connections:
  1425. * The first up interface should be left on and all others downed.
  1426. */
  1427. static int __bond_release_one(struct net_device *bond_dev,
  1428. struct net_device *slave_dev,
  1429. bool all)
  1430. {
  1431. struct bonding *bond = netdev_priv(bond_dev);
  1432. struct slave *slave, *oldcurrent;
  1433. struct sockaddr addr;
  1434. int old_flags = bond_dev->flags;
  1435. netdev_features_t old_features = bond_dev->features;
  1436. /* slave is not a slave or master is not master of this slave */
  1437. if (!(slave_dev->flags & IFF_SLAVE) ||
  1438. !netdev_has_upper_dev(slave_dev, bond_dev)) {
  1439. pr_err("%s: Error: cannot release %s.\n",
  1440. bond_dev->name, slave_dev->name);
  1441. return -EINVAL;
  1442. }
  1443. block_netpoll_tx();
  1444. slave = bond_get_slave_by_dev(bond, slave_dev);
  1445. if (!slave) {
  1446. /* not a slave of this bond */
  1447. pr_info("%s: %s not enslaved\n",
  1448. bond_dev->name, slave_dev->name);
  1449. unblock_netpoll_tx();
  1450. return -EINVAL;
  1451. }
  1452. /* release the slave from its bond */
  1453. bond->slave_cnt--;
  1454. bond_upper_dev_unlink(bond_dev, slave_dev);
  1455. /* unregister rx_handler early so bond_handle_frame wouldn't be called
  1456. * for this slave anymore.
  1457. */
  1458. netdev_rx_handler_unregister(slave_dev);
  1459. write_lock_bh(&bond->lock);
  1460. /* Inform AD package of unbinding of slave. */
  1461. if (bond->params.mode == BOND_MODE_8023AD)
  1462. bond_3ad_unbind_slave(slave);
  1463. write_unlock_bh(&bond->lock);
  1464. pr_info("%s: releasing %s interface %s\n",
  1465. bond_dev->name,
  1466. bond_is_active_slave(slave) ? "active" : "backup",
  1467. slave_dev->name);
  1468. oldcurrent = bond->curr_active_slave;
  1469. bond->current_arp_slave = NULL;
  1470. if (!all && !bond->params.fail_over_mac) {
  1471. if (ether_addr_equal(bond_dev->dev_addr, slave->perm_hwaddr) &&
  1472. bond_has_slaves(bond))
  1473. pr_warn("%s: Warning: the permanent HWaddr of %s - %pM - is still in use by %s. Set the HWaddr of %s to a different address to avoid conflicts.\n",
  1474. bond_dev->name, slave_dev->name,
  1475. slave->perm_hwaddr,
  1476. bond_dev->name, slave_dev->name);
  1477. }
  1478. if (bond->primary_slave == slave)
  1479. bond->primary_slave = NULL;
  1480. if (oldcurrent == slave) {
  1481. write_lock_bh(&bond->curr_slave_lock);
  1482. bond_change_active_slave(bond, NULL);
  1483. write_unlock_bh(&bond->curr_slave_lock);
  1484. }
  1485. if (bond_is_lb(bond)) {
  1486. /* Must be called only after the slave has been
  1487. * detached from the list and the curr_active_slave
  1488. * has been cleared (if our_slave == old_current),
  1489. * but before a new active slave is selected.
  1490. */
  1491. bond_alb_deinit_slave(bond, slave);
  1492. }
  1493. if (all) {
  1494. rcu_assign_pointer(bond->curr_active_slave, NULL);
  1495. } else if (oldcurrent == slave) {
  1496. /*
  1497. * Note that we hold RTNL over this sequence, so there
  1498. * is no concern that another slave add/remove event
  1499. * will interfere.
  1500. */
  1501. write_lock_bh(&bond->curr_slave_lock);
  1502. bond_select_active_slave(bond);
  1503. write_unlock_bh(&bond->curr_slave_lock);
  1504. }
  1505. if (!bond_has_slaves(bond)) {
  1506. bond_set_carrier(bond);
  1507. eth_hw_addr_random(bond_dev);
  1508. if (vlan_uses_dev(bond_dev)) {
  1509. pr_warning("%s: Warning: clearing HW address of %s while it still has VLANs.\n",
  1510. bond_dev->name, bond_dev->name);
  1511. pr_warning("%s: When re-adding slaves, make sure the bond's HW address matches its VLANs'.\n",
  1512. bond_dev->name);
  1513. }
  1514. }
  1515. unblock_netpoll_tx();
  1516. synchronize_rcu();
  1517. if (!bond_has_slaves(bond)) {
  1518. call_netdevice_notifiers(NETDEV_CHANGEADDR, bond->dev);
  1519. call_netdevice_notifiers(NETDEV_RELEASE, bond->dev);
  1520. }
  1521. bond_compute_features(bond);
  1522. if (!(bond_dev->features & NETIF_F_VLAN_CHALLENGED) &&
  1523. (old_features & NETIF_F_VLAN_CHALLENGED))
  1524. pr_info("%s: last VLAN challenged slave %s left bond %s. VLAN blocking is removed\n",
  1525. bond_dev->name, slave_dev->name, bond_dev->name);
  1526. /* must do this from outside any spinlocks */
  1527. vlan_vids_del_by_dev(slave_dev, bond_dev);
  1528. /* If the mode USES_PRIMARY, then this cases was handled above by
  1529. * bond_change_active_slave(..., NULL)
  1530. */
  1531. if (!USES_PRIMARY(bond->params.mode)) {
  1532. /* unset promiscuity level from slave
  1533. * NOTE: The NETDEV_CHANGEADDR call above may change the value
  1534. * of the IFF_PROMISC flag in the bond_dev, but we need the
  1535. * value of that flag before that change, as that was the value
  1536. * when this slave was attached, so we cache at the start of the
  1537. * function and use it here. Same goes for ALLMULTI below
  1538. */
  1539. if (old_flags & IFF_PROMISC)
  1540. dev_set_promiscuity(slave_dev, -1);
  1541. /* unset allmulti level from slave */
  1542. if (old_flags & IFF_ALLMULTI)
  1543. dev_set_allmulti(slave_dev, -1);
  1544. bond_hw_addr_flush(bond_dev, slave_dev);
  1545. }
  1546. slave_disable_netpoll(slave);
  1547. /* close slave before restoring its mac address */
  1548. dev_close(slave_dev);
  1549. if (bond->params.fail_over_mac != BOND_FOM_ACTIVE) {
  1550. /* restore original ("permanent") mac address */
  1551. memcpy(addr.sa_data, slave->perm_hwaddr, ETH_ALEN);
  1552. addr.sa_family = slave_dev->type;
  1553. dev_set_mac_address(slave_dev, &addr);
  1554. }
  1555. dev_set_mtu(slave_dev, slave->original_mtu);
  1556. slave_dev->priv_flags &= ~IFF_BONDING;
  1557. kfree(slave);
  1558. return 0; /* deletion OK */
  1559. }
  1560. /* A wrapper used because of ndo_del_link */
  1561. int bond_release(struct net_device *bond_dev, struct net_device *slave_dev)
  1562. {
  1563. return __bond_release_one(bond_dev, slave_dev, false);
  1564. }
  1565. /*
  1566. * First release a slave and then destroy the bond if no more slaves are left.
  1567. * Must be under rtnl_lock when this function is called.
  1568. */
  1569. static int bond_release_and_destroy(struct net_device *bond_dev,
  1570. struct net_device *slave_dev)
  1571. {
  1572. struct bonding *bond = netdev_priv(bond_dev);
  1573. int ret;
  1574. ret = bond_release(bond_dev, slave_dev);
  1575. if (ret == 0 && !bond_has_slaves(bond)) {
  1576. bond_dev->priv_flags |= IFF_DISABLE_NETPOLL;
  1577. pr_info("%s: destroying bond %s.\n",
  1578. bond_dev->name, bond_dev->name);
  1579. unregister_netdevice(bond_dev);
  1580. }
  1581. return ret;
  1582. }
  1583. static int bond_info_query(struct net_device *bond_dev, struct ifbond *info)
  1584. {
  1585. struct bonding *bond = netdev_priv(bond_dev);
  1586. info->bond_mode = bond->params.mode;
  1587. info->miimon = bond->params.miimon;
  1588. read_lock(&bond->lock);
  1589. info->num_slaves = bond->slave_cnt;
  1590. read_unlock(&bond->lock);
  1591. return 0;
  1592. }
  1593. static int bond_slave_info_query(struct net_device *bond_dev, struct ifslave *info)
  1594. {
  1595. struct bonding *bond = netdev_priv(bond_dev);
  1596. struct list_head *iter;
  1597. int i = 0, res = -ENODEV;
  1598. struct slave *slave;
  1599. read_lock(&bond->lock);
  1600. bond_for_each_slave(bond, slave, iter) {
  1601. if (i++ == (int)info->slave_id) {
  1602. res = 0;
  1603. strcpy(info->slave_name, slave->dev->name);
  1604. info->link = slave->link;
  1605. info->state = bond_slave_state(slave);
  1606. info->link_failure_count = slave->link_failure_count;
  1607. break;
  1608. }
  1609. }
  1610. read_unlock(&bond->lock);
  1611. return res;
  1612. }
  1613. /*-------------------------------- Monitoring -------------------------------*/
  1614. static int bond_miimon_inspect(struct bonding *bond)
  1615. {
  1616. int link_state, commit = 0;
  1617. struct list_head *iter;
  1618. struct slave *slave;
  1619. bool ignore_updelay;
  1620. ignore_updelay = !bond->curr_active_slave ? true : false;
  1621. bond_for_each_slave_rcu(bond, slave, iter) {
  1622. slave->new_link = BOND_LINK_NOCHANGE;
  1623. link_state = bond_check_dev_link(bond, slave->dev, 0);
  1624. switch (slave->link) {
  1625. case BOND_LINK_UP:
  1626. if (link_state)
  1627. continue;
  1628. slave->link = BOND_LINK_FAIL;
  1629. slave->delay = bond->params.downdelay;
  1630. if (slave->delay) {
  1631. pr_info("%s: link status down for %sinterface %s, disabling it in %d ms.\n",
  1632. bond->dev->name,
  1633. (bond->params.mode ==
  1634. BOND_MODE_ACTIVEBACKUP) ?
  1635. (bond_is_active_slave(slave) ?
  1636. "active " : "backup ") : "",
  1637. slave->dev->name,
  1638. bond->params.downdelay * bond->params.miimon);
  1639. }
  1640. /*FALLTHRU*/
  1641. case BOND_LINK_FAIL:
  1642. if (link_state) {
  1643. /*
  1644. * recovered before downdelay expired
  1645. */
  1646. slave->link = BOND_LINK_UP;
  1647. slave->jiffies = jiffies;
  1648. pr_info("%s: link status up again after %d ms for interface %s.\n",
  1649. bond->dev->name,
  1650. (bond->params.downdelay - slave->delay) *
  1651. bond->params.miimon,
  1652. slave->dev->name);
  1653. continue;
  1654. }
  1655. if (slave->delay <= 0) {
  1656. slave->new_link = BOND_LINK_DOWN;
  1657. commit++;
  1658. continue;
  1659. }
  1660. slave->delay--;
  1661. break;
  1662. case BOND_LINK_DOWN:
  1663. if (!link_state)
  1664. continue;
  1665. slave->link = BOND_LINK_BACK;
  1666. slave->delay = bond->params.updelay;
  1667. if (slave->delay) {
  1668. pr_info("%s: link status up for interface %s, enabling it in %d ms.\n",
  1669. bond->dev->name, slave->dev->name,
  1670. ignore_updelay ? 0 :
  1671. bond->params.updelay *
  1672. bond->params.miimon);
  1673. }
  1674. /*FALLTHRU*/
  1675. case BOND_LINK_BACK:
  1676. if (!link_state) {
  1677. slave->link = BOND_LINK_DOWN;
  1678. pr_info("%s: link status down again after %d ms for interface %s.\n",
  1679. bond->dev->name,
  1680. (bond->params.updelay - slave->delay) *
  1681. bond->params.miimon,
  1682. slave->dev->name);
  1683. continue;
  1684. }
  1685. if (ignore_updelay)
  1686. slave->delay = 0;
  1687. if (slave->delay <= 0) {
  1688. slave->new_link = BOND_LINK_UP;
  1689. commit++;
  1690. ignore_updelay = false;
  1691. continue;
  1692. }
  1693. slave->delay--;
  1694. break;
  1695. }
  1696. }
  1697. return commit;
  1698. }
  1699. static void bond_miimon_commit(struct bonding *bond)
  1700. {
  1701. struct list_head *iter;
  1702. struct slave *slave;
  1703. bond_for_each_slave(bond, slave, iter) {
  1704. switch (slave->new_link) {
  1705. case BOND_LINK_NOCHANGE:
  1706. continue;
  1707. case BOND_LINK_UP:
  1708. slave->link = BOND_LINK_UP;
  1709. slave->jiffies = jiffies;
  1710. if (bond->params.mode == BOND_MODE_8023AD) {
  1711. /* prevent it from being the active one */
  1712. bond_set_backup_slave(slave);
  1713. } else if (bond->params.mode != BOND_MODE_ACTIVEBACKUP) {
  1714. /* make it immediately active */
  1715. bond_set_active_slave(slave);
  1716. } else if (slave != bond->primary_slave) {
  1717. /* prevent it from being the active one */
  1718. bond_set_backup_slave(slave);
  1719. }
  1720. pr_info("%s: link status definitely up for interface %s, %u Mbps %s duplex.\n",
  1721. bond->dev->name, slave->dev->name,
  1722. slave->speed == SPEED_UNKNOWN ? 0 : slave->speed,
  1723. slave->duplex ? "full" : "half");
  1724. /* notify ad that the link status has changed */
  1725. if (bond->params.mode == BOND_MODE_8023AD)
  1726. bond_3ad_handle_link_change(slave, BOND_LINK_UP);
  1727. if (bond_is_lb(bond))
  1728. bond_alb_handle_link_change(bond, slave,
  1729. BOND_LINK_UP);
  1730. if (!bond->curr_active_slave ||
  1731. (slave == bond->primary_slave))
  1732. goto do_failover;
  1733. continue;
  1734. case BOND_LINK_DOWN:
  1735. if (slave->link_failure_count < UINT_MAX)
  1736. slave->link_failure_count++;
  1737. slave->link = BOND_LINK_DOWN;
  1738. if (bond->params.mode == BOND_MODE_ACTIVEBACKUP ||
  1739. bond->params.mode == BOND_MODE_8023AD)
  1740. bond_set_slave_inactive_flags(slave);
  1741. pr_info("%s: link status definitely down for interface %s, disabling it\n",
  1742. bond->dev->name, slave->dev->name);
  1743. if (bond->params.mode == BOND_MODE_8023AD)
  1744. bond_3ad_handle_link_change(slave,
  1745. BOND_LINK_DOWN);
  1746. if (bond_is_lb(bond))
  1747. bond_alb_handle_link_change(bond, slave,
  1748. BOND_LINK_DOWN);
  1749. if (slave == bond->curr_active_slave)
  1750. goto do_failover;
  1751. continue;
  1752. default:
  1753. pr_err("%s: invalid new link %d on slave %s\n",
  1754. bond->dev->name, slave->new_link,
  1755. slave->dev->name);
  1756. slave->new_link = BOND_LINK_NOCHANGE;
  1757. continue;
  1758. }
  1759. do_failover:
  1760. ASSERT_RTNL();
  1761. block_netpoll_tx();
  1762. write_lock_bh(&bond->curr_slave_lock);
  1763. bond_select_active_slave(bond);
  1764. write_unlock_bh(&bond->curr_slave_lock);
  1765. unblock_netpoll_tx();
  1766. }
  1767. bond_set_carrier(bond);
  1768. }
  1769. /*
  1770. * bond_mii_monitor
  1771. *
  1772. * Really a wrapper that splits the mii monitor into two phases: an
  1773. * inspection, then (if inspection indicates something needs to be done)
  1774. * an acquisition of appropriate locks followed by a commit phase to
  1775. * implement whatever link state changes are indicated.
  1776. */
  1777. void bond_mii_monitor(struct work_struct *work)
  1778. {
  1779. struct bonding *bond = container_of(work, struct bonding,
  1780. mii_work.work);
  1781. bool should_notify_peers = false;
  1782. unsigned long delay;
  1783. delay = msecs_to_jiffies(bond->params.miimon);
  1784. if (!bond_has_slaves(bond))
  1785. goto re_arm;
  1786. rcu_read_lock();
  1787. should_notify_peers = bond_should_notify_peers(bond);
  1788. if (bond_miimon_inspect(bond)) {
  1789. rcu_read_unlock();
  1790. /* Race avoidance with bond_close cancel of workqueue */
  1791. if (!rtnl_trylock()) {
  1792. delay = 1;
  1793. should_notify_peers = false;
  1794. goto re_arm;
  1795. }
  1796. bond_miimon_commit(bond);
  1797. rtnl_unlock(); /* might sleep, hold no other locks */
  1798. } else
  1799. rcu_read_unlock();
  1800. re_arm:
  1801. if (bond->params.miimon)
  1802. queue_delayed_work(bond->wq, &bond->mii_work, delay);
  1803. if (should_notify_peers) {
  1804. if (!rtnl_trylock())
  1805. return;
  1806. call_netdevice_notifiers(NETDEV_NOTIFY_PEERS, bond->dev);
  1807. rtnl_unlock();
  1808. }
  1809. }
  1810. static bool bond_has_this_ip(struct bonding *bond, __be32 ip)
  1811. {
  1812. struct net_device *upper;
  1813. struct list_head *iter;
  1814. bool ret = false;
  1815. if (ip == bond_confirm_addr(bond->dev, 0, ip))
  1816. return true;
  1817. rcu_read_lock();
  1818. netdev_for_each_all_upper_dev_rcu(bond->dev, upper, iter) {
  1819. if (ip == bond_confirm_addr(upper, 0, ip)) {
  1820. ret = true;
  1821. break;
  1822. }
  1823. }
  1824. rcu_read_unlock();
  1825. return ret;
  1826. }
  1827. /*
  1828. * We go to the (large) trouble of VLAN tagging ARP frames because
  1829. * switches in VLAN mode (especially if ports are configured as
  1830. * "native" to a VLAN) might not pass non-tagged frames.
  1831. */
  1832. static void bond_arp_send(struct net_device *slave_dev, int arp_op, __be32 dest_ip, __be32 src_ip, unsigned short vlan_id)
  1833. {
  1834. struct sk_buff *skb;
  1835. pr_debug("arp %d on slave %s: dst %pI4 src %pI4 vid %d\n", arp_op,
  1836. slave_dev->name, &dest_ip, &src_ip, vlan_id);
  1837. skb = arp_create(arp_op, ETH_P_ARP, dest_ip, slave_dev, src_ip,
  1838. NULL, slave_dev->dev_addr, NULL);
  1839. if (!skb) {
  1840. pr_err("ARP packet allocation failed\n");
  1841. return;
  1842. }
  1843. if (vlan_id) {
  1844. skb = vlan_put_tag(skb, htons(ETH_P_8021Q), vlan_id);
  1845. if (!skb) {
  1846. pr_err("failed to insert VLAN tag\n");
  1847. return;
  1848. }
  1849. }
  1850. arp_xmit(skb);
  1851. }
  1852. static void bond_arp_send_all(struct bonding *bond, struct slave *slave)
  1853. {
  1854. struct net_device *upper, *vlan_upper;
  1855. struct list_head *iter, *vlan_iter;
  1856. struct rtable *rt;
  1857. __be32 *targets = bond->params.arp_targets, addr;
  1858. int i, vlan_id;
  1859. for (i = 0; i < BOND_MAX_ARP_TARGETS && targets[i]; i++) {
  1860. pr_debug("basa: target %pI4\n", &targets[i]);
  1861. /* Find out through which dev should the packet go */
  1862. rt = ip_route_output(dev_net(bond->dev), targets[i], 0,
  1863. RTO_ONLINK, 0);
  1864. if (IS_ERR(rt)) {
  1865. pr_debug("%s: no route to arp_ip_target %pI4\n",
  1866. bond->dev->name, &targets[i]);
  1867. continue;
  1868. }
  1869. vlan_id = 0;
  1870. /* bond device itself */
  1871. if (rt->dst.dev == bond->dev)
  1872. goto found;
  1873. rcu_read_lock();
  1874. /* first we search only for vlan devices. for every vlan
  1875. * found we verify its upper dev list, searching for the
  1876. * rt->dst.dev. If found we save the tag of the vlan and
  1877. * proceed to send the packet.
  1878. *
  1879. * TODO: QinQ?
  1880. */
  1881. netdev_for_each_all_upper_dev_rcu(bond->dev, vlan_upper,
  1882. vlan_iter) {
  1883. if (!is_vlan_dev(vlan_upper))
  1884. continue;
  1885. netdev_for_each_all_upper_dev_rcu(vlan_upper, upper,
  1886. iter) {
  1887. if (upper == rt->dst.dev) {
  1888. vlan_id = vlan_dev_vlan_id(vlan_upper);
  1889. rcu_read_unlock();
  1890. goto found;
  1891. }
  1892. }
  1893. }
  1894. /* if the device we're looking for is not on top of any of
  1895. * our upper vlans, then just search for any dev that
  1896. * matches, and in case it's a vlan - save the id
  1897. */
  1898. netdev_for_each_all_upper_dev_rcu(bond->dev, upper, iter) {
  1899. if (upper == rt->dst.dev) {
  1900. /* if it's a vlan - get its VID */
  1901. if (is_vlan_dev(upper))
  1902. vlan_id = vlan_dev_vlan_id(upper);
  1903. rcu_read_unlock();
  1904. goto found;
  1905. }
  1906. }
  1907. rcu_read_unlock();
  1908. /* Not our device - skip */
  1909. pr_debug("%s: no path to arp_ip_target %pI4 via rt.dev %s\n",
  1910. bond->dev->name, &targets[i],
  1911. rt->dst.dev ? rt->dst.dev->name : "NULL");
  1912. ip_rt_put(rt);
  1913. continue;
  1914. found:
  1915. addr = bond_confirm_addr(rt->dst.dev, targets[i], 0);
  1916. ip_rt_put(rt);
  1917. bond_arp_send(slave->dev, ARPOP_REQUEST, targets[i],
  1918. addr, vlan_id);
  1919. }
  1920. }
  1921. static void bond_validate_arp(struct bonding *bond, struct slave *slave, __be32 sip, __be32 tip)
  1922. {
  1923. int i;
  1924. if (!sip || !bond_has_this_ip(bond, tip)) {
  1925. pr_debug("bva: sip %pI4 tip %pI4 not found\n", &sip, &tip);
  1926. return;
  1927. }
  1928. i = bond_get_targets_ip(bond->params.arp_targets, sip);
  1929. if (i == -1) {
  1930. pr_debug("bva: sip %pI4 not found in targets\n", &sip);
  1931. return;
  1932. }
  1933. slave->last_arp_rx = jiffies;
  1934. slave->target_last_arp_rx[i] = jiffies;
  1935. }
  1936. int bond_arp_rcv(const struct sk_buff *skb, struct bonding *bond,
  1937. struct slave *slave)
  1938. {
  1939. struct arphdr *arp = (struct arphdr *)skb->data;
  1940. unsigned char *arp_ptr;
  1941. __be32 sip, tip;
  1942. int alen;
  1943. if (skb->protocol != __cpu_to_be16(ETH_P_ARP))
  1944. return RX_HANDLER_ANOTHER;
  1945. read_lock(&bond->lock);
  1946. if (!slave_do_arp_validate(bond, slave))
  1947. goto out_unlock;
  1948. alen = arp_hdr_len(bond->dev);
  1949. pr_debug("bond_arp_rcv: bond %s skb->dev %s\n",
  1950. bond->dev->name, skb->dev->name);
  1951. if (alen > skb_headlen(skb)) {
  1952. arp = kmalloc(alen, GFP_ATOMIC);
  1953. if (!arp)
  1954. goto out_unlock;
  1955. if (skb_copy_bits(skb, 0, arp, alen) < 0)
  1956. goto out_unlock;
  1957. }
  1958. if (arp->ar_hln != bond->dev->addr_len ||
  1959. skb->pkt_type == PACKET_OTHERHOST ||
  1960. skb->pkt_type == PACKET_LOOPBACK ||
  1961. arp->ar_hrd != htons(ARPHRD_ETHER) ||
  1962. arp->ar_pro != htons(ETH_P_IP) ||
  1963. arp->ar_pln != 4)
  1964. goto out_unlock;
  1965. arp_ptr = (unsigned char *)(arp + 1);
  1966. arp_ptr += bond->dev->addr_len;
  1967. memcpy(&sip, arp_ptr, 4);
  1968. arp_ptr += 4 + bond->dev->addr_len;
  1969. memcpy(&tip, arp_ptr, 4);
  1970. pr_debug("bond_arp_rcv: %s %s/%d av %d sv %d sip %pI4 tip %pI4\n",
  1971. bond->dev->name, slave->dev->name, bond_slave_state(slave),
  1972. bond->params.arp_validate, slave_do_arp_validate(bond, slave),
  1973. &sip, &tip);
  1974. /*
  1975. * Backup slaves won't see the ARP reply, but do come through
  1976. * here for each ARP probe (so we swap the sip/tip to validate
  1977. * the probe). In a "redundant switch, common router" type of
  1978. * configuration, the ARP probe will (hopefully) travel from
  1979. * the active, through one switch, the router, then the other
  1980. * switch before reaching the backup.
  1981. *
  1982. * We 'trust' the arp requests if there is an active slave and
  1983. * it received valid arp reply(s) after it became active. This
  1984. * is done to avoid endless looping when we can't reach the
  1985. * arp_ip_target and fool ourselves with our own arp requests.
  1986. */
  1987. if (bond_is_active_slave(slave))
  1988. bond_validate_arp(bond, slave, sip, tip);
  1989. else if (bond->curr_active_slave &&
  1990. time_after(slave_last_rx(bond, bond->curr_active_slave),
  1991. bond->curr_active_slave->jiffies))
  1992. bond_validate_arp(bond, slave, tip, sip);
  1993. out_unlock:
  1994. read_unlock(&bond->lock);
  1995. if (arp != (struct arphdr *)skb->data)
  1996. kfree(arp);
  1997. return RX_HANDLER_ANOTHER;
  1998. }
  1999. /* function to verify if we're in the arp_interval timeslice, returns true if
  2000. * (last_act - arp_interval) <= jiffies <= (last_act + mod * arp_interval +
  2001. * arp_interval/2) . the arp_interval/2 is needed for really fast networks.
  2002. */
  2003. static bool bond_time_in_interval(struct bonding *bond, unsigned long last_act,
  2004. int mod)
  2005. {
  2006. int delta_in_ticks = msecs_to_jiffies(bond->params.arp_interval);
  2007. return time_in_range(jiffies,
  2008. last_act - delta_in_ticks,
  2009. last_act + mod * delta_in_ticks + delta_in_ticks/2);
  2010. }
  2011. /*
  2012. * this function is called regularly to monitor each slave's link
  2013. * ensuring that traffic is being sent and received when arp monitoring
  2014. * is used in load-balancing mode. if the adapter has been dormant, then an
  2015. * arp is transmitted to generate traffic. see activebackup_arp_monitor for
  2016. * arp monitoring in active backup mode.
  2017. */
  2018. void bond_loadbalance_arp_mon(struct work_struct *work)
  2019. {
  2020. struct bonding *bond = container_of(work, struct bonding,
  2021. arp_work.work);
  2022. struct slave *slave, *oldcurrent;
  2023. struct list_head *iter;
  2024. int do_failover = 0;
  2025. if (!bond_has_slaves(bond))
  2026. goto re_arm;
  2027. rcu_read_lock();
  2028. oldcurrent = ACCESS_ONCE(bond->curr_active_slave);
  2029. /* see if any of the previous devices are up now (i.e. they have
  2030. * xmt and rcv traffic). the curr_active_slave does not come into
  2031. * the picture unless it is null. also, slave->jiffies is not needed
  2032. * here because we send an arp on each slave and give a slave as
  2033. * long as it needs to get the tx/rx within the delta.
  2034. * TODO: what about up/down delay in arp mode? it wasn't here before
  2035. * so it can wait
  2036. */
  2037. bond_for_each_slave_rcu(bond, slave, iter) {
  2038. unsigned long trans_start = dev_trans_start(slave->dev);
  2039. if (slave->link != BOND_LINK_UP) {
  2040. if (bond_time_in_interval(bond, trans_start, 1) &&
  2041. bond_time_in_interval(bond, slave->dev->last_rx, 1)) {
  2042. slave->link = BOND_LINK_UP;
  2043. bond_set_active_slave(slave);
  2044. /* primary_slave has no meaning in round-robin
  2045. * mode. the window of a slave being up and
  2046. * curr_active_slave being null after enslaving
  2047. * is closed.
  2048. */
  2049. if (!oldcurrent) {
  2050. pr_info("%s: link status definitely up for interface %s, ",
  2051. bond->dev->name,
  2052. slave->dev->name);
  2053. do_failover = 1;
  2054. } else {
  2055. pr_info("%s: interface %s is now up\n",
  2056. bond->dev->name,
  2057. slave->dev->name);
  2058. }
  2059. }
  2060. } else {
  2061. /* slave->link == BOND_LINK_UP */
  2062. /* not all switches will respond to an arp request
  2063. * when the source ip is 0, so don't take the link down
  2064. * if we don't know our ip yet
  2065. */
  2066. if (!bond_time_in_interval(bond, trans_start, 2) ||
  2067. !bond_time_in_interval(bond, slave->dev->last_rx, 2)) {
  2068. slave->link = BOND_LINK_DOWN;
  2069. bond_set_backup_slave(slave);
  2070. if (slave->link_failure_count < UINT_MAX)
  2071. slave->link_failure_count++;
  2072. pr_info("%s: interface %s is now down.\n",
  2073. bond->dev->name,
  2074. slave->dev->name);
  2075. if (slave == oldcurrent)
  2076. do_failover = 1;
  2077. }
  2078. }
  2079. /* note: if switch is in round-robin mode, all links
  2080. * must tx arp to ensure all links rx an arp - otherwise
  2081. * links may oscillate or not come up at all; if switch is
  2082. * in something like xor mode, there is nothing we can
  2083. * do - all replies will be rx'ed on same link causing slaves
  2084. * to be unstable during low/no traffic periods
  2085. */
  2086. if (IS_UP(slave->dev))
  2087. bond_arp_send_all(bond, slave);
  2088. }
  2089. rcu_read_unlock();
  2090. if (do_failover) {
  2091. /* the bond_select_active_slave must hold RTNL
  2092. * and curr_slave_lock for write.
  2093. */
  2094. if (!rtnl_trylock())
  2095. goto re_arm;
  2096. block_netpoll_tx();
  2097. write_lock_bh(&bond->curr_slave_lock);
  2098. bond_select_active_slave(bond);
  2099. write_unlock_bh(&bond->curr_slave_lock);
  2100. unblock_netpoll_tx();
  2101. rtnl_unlock();
  2102. }
  2103. re_arm:
  2104. if (bond->params.arp_interval)
  2105. queue_delayed_work(bond->wq, &bond->arp_work,
  2106. msecs_to_jiffies(bond->params.arp_interval));
  2107. }
  2108. /*
  2109. * Called to inspect slaves for active-backup mode ARP monitor link state
  2110. * changes. Sets new_link in slaves to specify what action should take
  2111. * place for the slave. Returns 0 if no changes are found, >0 if changes
  2112. * to link states must be committed.
  2113. *
  2114. * Called with rcu_read_lock hold.
  2115. */
  2116. static int bond_ab_arp_inspect(struct bonding *bond)
  2117. {
  2118. unsigned long trans_start, last_rx;
  2119. struct list_head *iter;
  2120. struct slave *slave;
  2121. int commit = 0;
  2122. bond_for_each_slave_rcu(bond, slave, iter) {
  2123. slave->new_link = BOND_LINK_NOCHANGE;
  2124. last_rx = slave_last_rx(bond, slave);
  2125. if (slave->link != BOND_LINK_UP) {
  2126. if (bond_time_in_interval(bond, last_rx, 1)) {
  2127. slave->new_link = BOND_LINK_UP;
  2128. commit++;
  2129. }
  2130. continue;
  2131. }
  2132. /*
  2133. * Give slaves 2*delta after being enslaved or made
  2134. * active. This avoids bouncing, as the last receive
  2135. * times need a full ARP monitor cycle to be updated.
  2136. */
  2137. if (bond_time_in_interval(bond, slave->jiffies, 2))
  2138. continue;
  2139. /*
  2140. * Backup slave is down if:
  2141. * - No current_arp_slave AND
  2142. * - more than 3*delta since last receive AND
  2143. * - the bond has an IP address
  2144. *
  2145. * Note: a non-null current_arp_slave indicates
  2146. * the curr_active_slave went down and we are
  2147. * searching for a new one; under this condition
  2148. * we only take the curr_active_slave down - this
  2149. * gives each slave a chance to tx/rx traffic
  2150. * before being taken out
  2151. */
  2152. if (!bond_is_active_slave(slave) &&
  2153. !bond->current_arp_slave &&
  2154. !bond_time_in_interval(bond, last_rx, 3)) {
  2155. slave->new_link = BOND_LINK_DOWN;
  2156. commit++;
  2157. }
  2158. /*
  2159. * Active slave is down if:
  2160. * - more than 2*delta since transmitting OR
  2161. * - (more than 2*delta since receive AND
  2162. * the bond has an IP address)
  2163. */
  2164. trans_start = dev_trans_start(slave->dev);
  2165. if (bond_is_active_slave(slave) &&
  2166. (!bond_time_in_interval(bond, trans_start, 2) ||
  2167. !bond_time_in_interval(bond, last_rx, 2))) {
  2168. slave->new_link = BOND_LINK_DOWN;
  2169. commit++;
  2170. }
  2171. }
  2172. return commit;
  2173. }
  2174. /*
  2175. * Called to commit link state changes noted by inspection step of
  2176. * active-backup mode ARP monitor.
  2177. *
  2178. * Called with RTNL hold.
  2179. */
  2180. static void bond_ab_arp_commit(struct bonding *bond)
  2181. {
  2182. unsigned long trans_start;
  2183. struct list_head *iter;
  2184. struct slave *slave;
  2185. bond_for_each_slave(bond, slave, iter) {
  2186. switch (slave->new_link) {
  2187. case BOND_LINK_NOCHANGE:
  2188. continue;
  2189. case BOND_LINK_UP:
  2190. trans_start = dev_trans_start(slave->dev);
  2191. if (bond->curr_active_slave != slave ||
  2192. (!bond->curr_active_slave &&
  2193. bond_time_in_interval(bond, trans_start, 1))) {
  2194. slave->link = BOND_LINK_UP;
  2195. if (bond->current_arp_slave) {
  2196. bond_set_slave_inactive_flags(
  2197. bond->current_arp_slave);
  2198. bond->current_arp_slave = NULL;
  2199. }
  2200. pr_info("%s: link status definitely up for interface %s.\n",
  2201. bond->dev->name, slave->dev->name);
  2202. if (!bond->curr_active_slave ||
  2203. (slave == bond->primary_slave))
  2204. goto do_failover;
  2205. }
  2206. continue;
  2207. case BOND_LINK_DOWN:
  2208. if (slave->link_failure_count < UINT_MAX)
  2209. slave->link_failure_count++;
  2210. slave->link = BOND_LINK_DOWN;
  2211. bond_set_slave_inactive_flags(slave);
  2212. pr_info("%s: link status definitely down for interface %s, disabling it\n",
  2213. bond->dev->name, slave->dev->name);
  2214. if (slave == bond->curr_active_slave) {
  2215. bond->current_arp_slave = NULL;
  2216. goto do_failover;
  2217. }
  2218. continue;
  2219. default:
  2220. pr_err("%s: impossible: new_link %d on slave %s\n",
  2221. bond->dev->name, slave->new_link,
  2222. slave->dev->name);
  2223. continue;
  2224. }
  2225. do_failover:
  2226. ASSERT_RTNL();
  2227. block_netpoll_tx();
  2228. write_lock_bh(&bond->curr_slave_lock);
  2229. bond_select_active_slave(bond);
  2230. write_unlock_bh(&bond->curr_slave_lock);
  2231. unblock_netpoll_tx();
  2232. }
  2233. bond_set_carrier(bond);
  2234. }
  2235. /*
  2236. * Send ARP probes for active-backup mode ARP monitor.
  2237. *
  2238. * Called with rcu_read_lock hold.
  2239. */
  2240. static void bond_ab_arp_probe(struct bonding *bond)
  2241. {
  2242. struct slave *slave, *before = NULL, *new_slave = NULL,
  2243. *curr_arp_slave = rcu_dereference(bond->current_arp_slave);
  2244. struct list_head *iter;
  2245. bool found = false;
  2246. read_lock(&bond->curr_slave_lock);
  2247. if (curr_arp_slave && bond->curr_active_slave)
  2248. pr_info("PROBE: c_arp %s && cas %s BAD\n",
  2249. curr_arp_slave->dev->name,
  2250. bond->curr_active_slave->dev->name);
  2251. if (bond->curr_active_slave) {
  2252. bond_arp_send_all(bond, bond->curr_active_slave);
  2253. read_unlock(&bond->curr_slave_lock);
  2254. return;
  2255. }
  2256. read_unlock(&bond->curr_slave_lock);
  2257. /* if we don't have a curr_active_slave, search for the next available
  2258. * backup slave from the current_arp_slave and make it the candidate
  2259. * for becoming the curr_active_slave
  2260. */
  2261. if (!curr_arp_slave) {
  2262. curr_arp_slave = bond_first_slave_rcu(bond);
  2263. if (!curr_arp_slave)
  2264. return;
  2265. }
  2266. bond_set_slave_inactive_flags(curr_arp_slave);
  2267. bond_for_each_slave_rcu(bond, slave, iter) {
  2268. if (!found && !before && IS_UP(slave->dev))
  2269. before = slave;
  2270. if (found && !new_slave && IS_UP(slave->dev))
  2271. new_slave = slave;
  2272. /* if the link state is up at this point, we
  2273. * mark it down - this can happen if we have
  2274. * simultaneous link failures and
  2275. * reselect_active_interface doesn't make this
  2276. * one the current slave so it is still marked
  2277. * up when it is actually down
  2278. */
  2279. if (!IS_UP(slave->dev) && slave->link == BOND_LINK_UP) {
  2280. slave->link = BOND_LINK_DOWN;
  2281. if (slave->link_failure_count < UINT_MAX)
  2282. slave->link_failure_count++;
  2283. bond_set_slave_inactive_flags(slave);
  2284. pr_info("%s: backup interface %s is now down.\n",
  2285. bond->dev->name, slave->dev->name);
  2286. }
  2287. if (slave == curr_arp_slave)
  2288. found = true;
  2289. }
  2290. if (!new_slave && before)
  2291. new_slave = before;
  2292. if (!new_slave)
  2293. return;
  2294. new_slave->link = BOND_LINK_BACK;
  2295. bond_set_slave_active_flags(new_slave);
  2296. bond_arp_send_all(bond, new_slave);
  2297. new_slave->jiffies = jiffies;
  2298. rcu_assign_pointer(bond->current_arp_slave, new_slave);
  2299. }
  2300. void bond_activebackup_arp_mon(struct work_struct *work)
  2301. {
  2302. struct bonding *bond = container_of(work, struct bonding,
  2303. arp_work.work);
  2304. bool should_notify_peers = false;
  2305. int delta_in_ticks;
  2306. delta_in_ticks = msecs_to_jiffies(bond->params.arp_interval);
  2307. if (!bond_has_slaves(bond))
  2308. goto re_arm;
  2309. rcu_read_lock();
  2310. should_notify_peers = bond_should_notify_peers(bond);
  2311. if (bond_ab_arp_inspect(bond)) {
  2312. rcu_read_unlock();
  2313. /* Race avoidance with bond_close flush of workqueue */
  2314. if (!rtnl_trylock()) {
  2315. delta_in_ticks = 1;
  2316. should_notify_peers = false;
  2317. goto re_arm;
  2318. }
  2319. bond_ab_arp_commit(bond);
  2320. rtnl_unlock();
  2321. rcu_read_lock();
  2322. }
  2323. bond_ab_arp_probe(bond);
  2324. rcu_read_unlock();
  2325. re_arm:
  2326. if (bond->params.arp_interval)
  2327. queue_delayed_work(bond->wq, &bond->arp_work, delta_in_ticks);
  2328. if (should_notify_peers) {
  2329. if (!rtnl_trylock())
  2330. return;
  2331. call_netdevice_notifiers(NETDEV_NOTIFY_PEERS, bond->dev);
  2332. rtnl_unlock();
  2333. }
  2334. }
  2335. /*-------------------------- netdev event handling --------------------------*/
  2336. /*
  2337. * Change device name
  2338. */
  2339. static int bond_event_changename(struct bonding *bond)
  2340. {
  2341. bond_remove_proc_entry(bond);
  2342. bond_create_proc_entry(bond);
  2343. bond_debug_reregister(bond);
  2344. return NOTIFY_DONE;
  2345. }
  2346. static int bond_master_netdev_event(unsigned long event,
  2347. struct net_device *bond_dev)
  2348. {
  2349. struct bonding *event_bond = netdev_priv(bond_dev);
  2350. switch (event) {
  2351. case NETDEV_CHANGENAME:
  2352. return bond_event_changename(event_bond);
  2353. case NETDEV_UNREGISTER:
  2354. bond_remove_proc_entry(event_bond);
  2355. break;
  2356. case NETDEV_REGISTER:
  2357. bond_create_proc_entry(event_bond);
  2358. break;
  2359. case NETDEV_NOTIFY_PEERS:
  2360. if (event_bond->send_peer_notif)
  2361. event_bond->send_peer_notif--;
  2362. break;
  2363. default:
  2364. break;
  2365. }
  2366. return NOTIFY_DONE;
  2367. }
  2368. static int bond_slave_netdev_event(unsigned long event,
  2369. struct net_device *slave_dev)
  2370. {
  2371. struct slave *slave = bond_slave_get_rtnl(slave_dev);
  2372. struct bonding *bond;
  2373. struct net_device *bond_dev;
  2374. u32 old_speed;
  2375. u8 old_duplex;
  2376. /* A netdev event can be generated while enslaving a device
  2377. * before netdev_rx_handler_register is called in which case
  2378. * slave will be NULL
  2379. */
  2380. if (!slave)
  2381. return NOTIFY_DONE;
  2382. bond_dev = slave->bond->dev;
  2383. bond = slave->bond;
  2384. switch (event) {
  2385. case NETDEV_UNREGISTER:
  2386. if (bond_dev->type != ARPHRD_ETHER)
  2387. bond_release_and_destroy(bond_dev, slave_dev);
  2388. else
  2389. bond_release(bond_dev, slave_dev);
  2390. break;
  2391. case NETDEV_UP:
  2392. case NETDEV_CHANGE:
  2393. old_speed = slave->speed;
  2394. old_duplex = slave->duplex;
  2395. bond_update_speed_duplex(slave);
  2396. if (bond->params.mode == BOND_MODE_8023AD) {
  2397. if (old_speed != slave->speed)
  2398. bond_3ad_adapter_speed_changed(slave);
  2399. if (old_duplex != slave->duplex)
  2400. bond_3ad_adapter_duplex_changed(slave);
  2401. }
  2402. break;
  2403. case NETDEV_DOWN:
  2404. /*
  2405. * ... Or is it this?
  2406. */
  2407. break;
  2408. case NETDEV_CHANGEMTU:
  2409. /*
  2410. * TODO: Should slaves be allowed to
  2411. * independently alter their MTU? For
  2412. * an active-backup bond, slaves need
  2413. * not be the same type of device, so
  2414. * MTUs may vary. For other modes,
  2415. * slaves arguably should have the
  2416. * same MTUs. To do this, we'd need to
  2417. * take over the slave's change_mtu
  2418. * function for the duration of their
  2419. * servitude.
  2420. */
  2421. break;
  2422. case NETDEV_CHANGENAME:
  2423. /*
  2424. * TODO: handle changing the primary's name
  2425. */
  2426. break;
  2427. case NETDEV_FEAT_CHANGE:
  2428. bond_compute_features(bond);
  2429. break;
  2430. case NETDEV_RESEND_IGMP:
  2431. /* Propagate to master device */
  2432. call_netdevice_notifiers(event, slave->bond->dev);
  2433. break;
  2434. default:
  2435. break;
  2436. }
  2437. return NOTIFY_DONE;
  2438. }
  2439. /*
  2440. * bond_netdev_event: handle netdev notifier chain events.
  2441. *
  2442. * This function receives events for the netdev chain. The caller (an
  2443. * ioctl handler calling blocking_notifier_call_chain) holds the necessary
  2444. * locks for us to safely manipulate the slave devices (RTNL lock,
  2445. * dev_probe_lock).
  2446. */
  2447. static int bond_netdev_event(struct notifier_block *this,
  2448. unsigned long event, void *ptr)
  2449. {
  2450. struct net_device *event_dev = netdev_notifier_info_to_dev(ptr);
  2451. pr_debug("event_dev: %s, event: %lx\n",
  2452. event_dev ? event_dev->name : "None",
  2453. event);
  2454. if (!(event_dev->priv_flags & IFF_BONDING))
  2455. return NOTIFY_DONE;
  2456. if (event_dev->flags & IFF_MASTER) {
  2457. pr_debug("IFF_MASTER\n");
  2458. return bond_master_netdev_event(event, event_dev);
  2459. }
  2460. if (event_dev->flags & IFF_SLAVE) {
  2461. pr_debug("IFF_SLAVE\n");
  2462. return bond_slave_netdev_event(event, event_dev);
  2463. }
  2464. return NOTIFY_DONE;
  2465. }
  2466. static struct notifier_block bond_netdev_notifier = {
  2467. .notifier_call = bond_netdev_event,
  2468. };
  2469. /*---------------------------- Hashing Policies -----------------------------*/
  2470. /* L2 hash helper */
  2471. static inline u32 bond_eth_hash(struct sk_buff *skb)
  2472. {
  2473. struct ethhdr *data = (struct ethhdr *)skb->data;
  2474. if (skb_headlen(skb) >= offsetof(struct ethhdr, h_proto))
  2475. return data->h_dest[5] ^ data->h_source[5];
  2476. return 0;
  2477. }
  2478. /* Extract the appropriate headers based on bond's xmit policy */
  2479. static bool bond_flow_dissect(struct bonding *bond, struct sk_buff *skb,
  2480. struct flow_keys *fk)
  2481. {
  2482. const struct ipv6hdr *iph6;
  2483. const struct iphdr *iph;
  2484. int noff, proto = -1;
  2485. if (bond->params.xmit_policy > BOND_XMIT_POLICY_LAYER23)
  2486. return skb_flow_dissect(skb, fk);
  2487. fk->ports = 0;
  2488. noff = skb_network_offset(skb);
  2489. if (skb->protocol == htons(ETH_P_IP)) {
  2490. if (!pskb_may_pull(skb, noff + sizeof(*iph)))
  2491. return false;
  2492. iph = ip_hdr(skb);
  2493. fk->src = iph->saddr;
  2494. fk->dst = iph->daddr;
  2495. noff += iph->ihl << 2;
  2496. if (!ip_is_fragment(iph))
  2497. proto = iph->protocol;
  2498. } else if (skb->protocol == htons(ETH_P_IPV6)) {
  2499. if (!pskb_may_pull(skb, noff + sizeof(*iph6)))
  2500. return false;
  2501. iph6 = ipv6_hdr(skb);
  2502. fk->src = (__force __be32)ipv6_addr_hash(&iph6->saddr);
  2503. fk->dst = (__force __be32)ipv6_addr_hash(&iph6->daddr);
  2504. noff += sizeof(*iph6);
  2505. proto = iph6->nexthdr;
  2506. } else {
  2507. return false;
  2508. }
  2509. if (bond->params.xmit_policy == BOND_XMIT_POLICY_LAYER34 && proto >= 0)
  2510. fk->ports = skb_flow_get_ports(skb, noff, proto);
  2511. return true;
  2512. }
  2513. /**
  2514. * bond_xmit_hash - generate a hash value based on the xmit policy
  2515. * @bond: bonding device
  2516. * @skb: buffer to use for headers
  2517. * @count: modulo value
  2518. *
  2519. * This function will extract the necessary headers from the skb buffer and use
  2520. * them to generate a hash based on the xmit_policy set in the bonding device
  2521. * which will be reduced modulo count before returning.
  2522. */
  2523. int bond_xmit_hash(struct bonding *bond, struct sk_buff *skb, int count)
  2524. {
  2525. struct flow_keys flow;
  2526. u32 hash;
  2527. if (bond->params.xmit_policy == BOND_XMIT_POLICY_LAYER2 ||
  2528. !bond_flow_dissect(bond, skb, &flow))
  2529. return bond_eth_hash(skb) % count;
  2530. if (bond->params.xmit_policy == BOND_XMIT_POLICY_LAYER23 ||
  2531. bond->params.xmit_policy == BOND_XMIT_POLICY_ENCAP23)
  2532. hash = bond_eth_hash(skb);
  2533. else
  2534. hash = (__force u32)flow.ports;
  2535. hash ^= (__force u32)flow.dst ^ (__force u32)flow.src;
  2536. hash ^= (hash >> 16);
  2537. hash ^= (hash >> 8);
  2538. return hash % count;
  2539. }
  2540. /*-------------------------- Device entry points ----------------------------*/
  2541. static void bond_work_init_all(struct bonding *bond)
  2542. {
  2543. INIT_DELAYED_WORK(&bond->mcast_work,
  2544. bond_resend_igmp_join_requests_delayed);
  2545. INIT_DELAYED_WORK(&bond->alb_work, bond_alb_monitor);
  2546. INIT_DELAYED_WORK(&bond->mii_work, bond_mii_monitor);
  2547. if (bond->params.mode == BOND_MODE_ACTIVEBACKUP)
  2548. INIT_DELAYED_WORK(&bond->arp_work, bond_activebackup_arp_mon);
  2549. else
  2550. INIT_DELAYED_WORK(&bond->arp_work, bond_loadbalance_arp_mon);
  2551. INIT_DELAYED_WORK(&bond->ad_work, bond_3ad_state_machine_handler);
  2552. }
  2553. static void bond_work_cancel_all(struct bonding *bond)
  2554. {
  2555. cancel_delayed_work_sync(&bond->mii_work);
  2556. cancel_delayed_work_sync(&bond->arp_work);
  2557. cancel_delayed_work_sync(&bond->alb_work);
  2558. cancel_delayed_work_sync(&bond->ad_work);
  2559. cancel_delayed_work_sync(&bond->mcast_work);
  2560. }
  2561. static int bond_open(struct net_device *bond_dev)
  2562. {
  2563. struct bonding *bond = netdev_priv(bond_dev);
  2564. struct list_head *iter;
  2565. struct slave *slave;
  2566. /* reset slave->backup and slave->inactive */
  2567. read_lock(&bond->lock);
  2568. if (bond_has_slaves(bond)) {
  2569. read_lock(&bond->curr_slave_lock);
  2570. bond_for_each_slave(bond, slave, iter) {
  2571. if ((bond->params.mode == BOND_MODE_ACTIVEBACKUP)
  2572. && (slave != bond->curr_active_slave)) {
  2573. bond_set_slave_inactive_flags(slave);
  2574. } else {
  2575. bond_set_slave_active_flags(slave);
  2576. }
  2577. }
  2578. read_unlock(&bond->curr_slave_lock);
  2579. }
  2580. read_unlock(&bond->lock);
  2581. bond_work_init_all(bond);
  2582. if (bond_is_lb(bond)) {
  2583. /* bond_alb_initialize must be called before the timer
  2584. * is started.
  2585. */
  2586. if (bond_alb_initialize(bond, (bond->params.mode == BOND_MODE_ALB)))
  2587. return -ENOMEM;
  2588. queue_delayed_work(bond->wq, &bond->alb_work, 0);
  2589. }
  2590. if (bond->params.miimon) /* link check interval, in milliseconds. */
  2591. queue_delayed_work(bond->wq, &bond->mii_work, 0);
  2592. if (bond->params.arp_interval) { /* arp interval, in milliseconds. */
  2593. queue_delayed_work(bond->wq, &bond->arp_work, 0);
  2594. if (bond->params.arp_validate)
  2595. bond->recv_probe = bond_arp_rcv;
  2596. }
  2597. if (bond->params.mode == BOND_MODE_8023AD) {
  2598. queue_delayed_work(bond->wq, &bond->ad_work, 0);
  2599. /* register to receive LACPDUs */
  2600. bond->recv_probe = bond_3ad_lacpdu_recv;
  2601. bond_3ad_initiate_agg_selection(bond, 1);
  2602. }
  2603. return 0;
  2604. }
  2605. static int bond_close(struct net_device *bond_dev)
  2606. {
  2607. struct bonding *bond = netdev_priv(bond_dev);
  2608. bond_work_cancel_all(bond);
  2609. bond->send_peer_notif = 0;
  2610. if (bond_is_lb(bond))
  2611. bond_alb_deinitialize(bond);
  2612. bond->recv_probe = NULL;
  2613. return 0;
  2614. }
  2615. static struct rtnl_link_stats64 *bond_get_stats(struct net_device *bond_dev,
  2616. struct rtnl_link_stats64 *stats)
  2617. {
  2618. struct bonding *bond = netdev_priv(bond_dev);
  2619. struct rtnl_link_stats64 temp;
  2620. struct list_head *iter;
  2621. struct slave *slave;
  2622. memset(stats, 0, sizeof(*stats));
  2623. read_lock_bh(&bond->lock);
  2624. bond_for_each_slave(bond, slave, iter) {
  2625. const struct rtnl_link_stats64 *sstats =
  2626. dev_get_stats(slave->dev, &temp);
  2627. stats->rx_packets += sstats->rx_packets;
  2628. stats->rx_bytes += sstats->rx_bytes;
  2629. stats->rx_errors += sstats->rx_errors;
  2630. stats->rx_dropped += sstats->rx_dropped;
  2631. stats->tx_packets += sstats->tx_packets;
  2632. stats->tx_bytes += sstats->tx_bytes;
  2633. stats->tx_errors += sstats->tx_errors;
  2634. stats->tx_dropped += sstats->tx_dropped;
  2635. stats->multicast += sstats->multicast;
  2636. stats->collisions += sstats->collisions;
  2637. stats->rx_length_errors += sstats->rx_length_errors;
  2638. stats->rx_over_errors += sstats->rx_over_errors;
  2639. stats->rx_crc_errors += sstats->rx_crc_errors;
  2640. stats->rx_frame_errors += sstats->rx_frame_errors;
  2641. stats->rx_fifo_errors += sstats->rx_fifo_errors;
  2642. stats->rx_missed_errors += sstats->rx_missed_errors;
  2643. stats->tx_aborted_errors += sstats->tx_aborted_errors;
  2644. stats->tx_carrier_errors += sstats->tx_carrier_errors;
  2645. stats->tx_fifo_errors += sstats->tx_fifo_errors;
  2646. stats->tx_heartbeat_errors += sstats->tx_heartbeat_errors;
  2647. stats->tx_window_errors += sstats->tx_window_errors;
  2648. }
  2649. read_unlock_bh(&bond->lock);
  2650. return stats;
  2651. }
  2652. static int bond_do_ioctl(struct net_device *bond_dev, struct ifreq *ifr, int cmd)
  2653. {
  2654. struct bonding *bond = netdev_priv(bond_dev);
  2655. struct net_device *slave_dev = NULL;
  2656. struct ifbond k_binfo;
  2657. struct ifbond __user *u_binfo = NULL;
  2658. struct ifslave k_sinfo;
  2659. struct ifslave __user *u_sinfo = NULL;
  2660. struct mii_ioctl_data *mii = NULL;
  2661. struct net *net;
  2662. int res = 0;
  2663. pr_debug("bond_ioctl: master=%s, cmd=%d\n", bond_dev->name, cmd);
  2664. switch (cmd) {
  2665. case SIOCGMIIPHY:
  2666. mii = if_mii(ifr);
  2667. if (!mii)
  2668. return -EINVAL;
  2669. mii->phy_id = 0;
  2670. /* Fall Through */
  2671. case SIOCGMIIREG:
  2672. /*
  2673. * We do this again just in case we were called by SIOCGMIIREG
  2674. * instead of SIOCGMIIPHY.
  2675. */
  2676. mii = if_mii(ifr);
  2677. if (!mii)
  2678. return -EINVAL;
  2679. if (mii->reg_num == 1) {
  2680. mii->val_out = 0;
  2681. read_lock(&bond->lock);
  2682. read_lock(&bond->curr_slave_lock);
  2683. if (netif_carrier_ok(bond->dev))
  2684. mii->val_out = BMSR_LSTATUS;
  2685. read_unlock(&bond->curr_slave_lock);
  2686. read_unlock(&bond->lock);
  2687. }
  2688. return 0;
  2689. case BOND_INFO_QUERY_OLD:
  2690. case SIOCBONDINFOQUERY:
  2691. u_binfo = (struct ifbond __user *)ifr->ifr_data;
  2692. if (copy_from_user(&k_binfo, u_binfo, sizeof(ifbond)))
  2693. return -EFAULT;
  2694. res = bond_info_query(bond_dev, &k_binfo);
  2695. if (res == 0 &&
  2696. copy_to_user(u_binfo, &k_binfo, sizeof(ifbond)))
  2697. return -EFAULT;
  2698. return res;
  2699. case BOND_SLAVE_INFO_QUERY_OLD:
  2700. case SIOCBONDSLAVEINFOQUERY:
  2701. u_sinfo = (struct ifslave __user *)ifr->ifr_data;
  2702. if (copy_from_user(&k_sinfo, u_sinfo, sizeof(ifslave)))
  2703. return -EFAULT;
  2704. res = bond_slave_info_query(bond_dev, &k_sinfo);
  2705. if (res == 0 &&
  2706. copy_to_user(u_sinfo, &k_sinfo, sizeof(ifslave)))
  2707. return -EFAULT;
  2708. return res;
  2709. default:
  2710. /* Go on */
  2711. break;
  2712. }
  2713. net = dev_net(bond_dev);
  2714. if (!ns_capable(net->user_ns, CAP_NET_ADMIN))
  2715. return -EPERM;
  2716. slave_dev = dev_get_by_name(net, ifr->ifr_slave);
  2717. pr_debug("slave_dev=%p:\n", slave_dev);
  2718. if (!slave_dev)
  2719. res = -ENODEV;
  2720. else {
  2721. pr_debug("slave_dev->name=%s:\n", slave_dev->name);
  2722. switch (cmd) {
  2723. case BOND_ENSLAVE_OLD:
  2724. case SIOCBONDENSLAVE:
  2725. res = bond_enslave(bond_dev, slave_dev);
  2726. break;
  2727. case BOND_RELEASE_OLD:
  2728. case SIOCBONDRELEASE:
  2729. res = bond_release(bond_dev, slave_dev);
  2730. break;
  2731. case BOND_SETHWADDR_OLD:
  2732. case SIOCBONDSETHWADDR:
  2733. bond_set_dev_addr(bond_dev, slave_dev);
  2734. res = 0;
  2735. break;
  2736. case BOND_CHANGE_ACTIVE_OLD:
  2737. case SIOCBONDCHANGEACTIVE:
  2738. res = bond_option_active_slave_set(bond, slave_dev);
  2739. break;
  2740. default:
  2741. res = -EOPNOTSUPP;
  2742. }
  2743. dev_put(slave_dev);
  2744. }
  2745. return res;
  2746. }
  2747. static void bond_change_rx_flags(struct net_device *bond_dev, int change)
  2748. {
  2749. struct bonding *bond = netdev_priv(bond_dev);
  2750. if (change & IFF_PROMISC)
  2751. bond_set_promiscuity(bond,
  2752. bond_dev->flags & IFF_PROMISC ? 1 : -1);
  2753. if (change & IFF_ALLMULTI)
  2754. bond_set_allmulti(bond,
  2755. bond_dev->flags & IFF_ALLMULTI ? 1 : -1);
  2756. }
  2757. static void bond_set_rx_mode(struct net_device *bond_dev)
  2758. {
  2759. struct bonding *bond = netdev_priv(bond_dev);
  2760. struct list_head *iter;
  2761. struct slave *slave;
  2762. rcu_read_lock();
  2763. if (USES_PRIMARY(bond->params.mode)) {
  2764. slave = rcu_dereference(bond->curr_active_slave);
  2765. if (slave) {
  2766. dev_uc_sync(slave->dev, bond_dev);
  2767. dev_mc_sync(slave->dev, bond_dev);
  2768. }
  2769. } else {
  2770. bond_for_each_slave_rcu(bond, slave, iter) {
  2771. dev_uc_sync_multiple(slave->dev, bond_dev);
  2772. dev_mc_sync_multiple(slave->dev, bond_dev);
  2773. }
  2774. }
  2775. rcu_read_unlock();
  2776. }
  2777. static int bond_neigh_init(struct neighbour *n)
  2778. {
  2779. struct bonding *bond = netdev_priv(n->dev);
  2780. const struct net_device_ops *slave_ops;
  2781. struct neigh_parms parms;
  2782. struct slave *slave;
  2783. int ret;
  2784. slave = bond_first_slave(bond);
  2785. if (!slave)
  2786. return 0;
  2787. slave_ops = slave->dev->netdev_ops;
  2788. if (!slave_ops->ndo_neigh_setup)
  2789. return 0;
  2790. parms.neigh_setup = NULL;
  2791. parms.neigh_cleanup = NULL;
  2792. ret = slave_ops->ndo_neigh_setup(slave->dev, &parms);
  2793. if (ret)
  2794. return ret;
  2795. /*
  2796. * Assign slave's neigh_cleanup to neighbour in case cleanup is called
  2797. * after the last slave has been detached. Assumes that all slaves
  2798. * utilize the same neigh_cleanup (true at this writing as only user
  2799. * is ipoib).
  2800. */
  2801. n->parms->neigh_cleanup = parms.neigh_cleanup;
  2802. if (!parms.neigh_setup)
  2803. return 0;
  2804. return parms.neigh_setup(n);
  2805. }
  2806. /*
  2807. * The bonding ndo_neigh_setup is called at init time beofre any
  2808. * slave exists. So we must declare proxy setup function which will
  2809. * be used at run time to resolve the actual slave neigh param setup.
  2810. *
  2811. * It's also called by master devices (such as vlans) to setup their
  2812. * underlying devices. In that case - do nothing, we're already set up from
  2813. * our init.
  2814. */
  2815. static int bond_neigh_setup(struct net_device *dev,
  2816. struct neigh_parms *parms)
  2817. {
  2818. /* modify only our neigh_parms */
  2819. if (parms->dev == dev)
  2820. parms->neigh_setup = bond_neigh_init;
  2821. return 0;
  2822. }
  2823. /*
  2824. * Change the MTU of all of a master's slaves to match the master
  2825. */
  2826. static int bond_change_mtu(struct net_device *bond_dev, int new_mtu)
  2827. {
  2828. struct bonding *bond = netdev_priv(bond_dev);
  2829. struct slave *slave, *rollback_slave;
  2830. struct list_head *iter;
  2831. int res = 0;
  2832. pr_debug("bond=%p, name=%s, new_mtu=%d\n", bond,
  2833. (bond_dev ? bond_dev->name : "None"), new_mtu);
  2834. /* Can't hold bond->lock with bh disabled here since
  2835. * some base drivers panic. On the other hand we can't
  2836. * hold bond->lock without bh disabled because we'll
  2837. * deadlock. The only solution is to rely on the fact
  2838. * that we're under rtnl_lock here, and the slaves
  2839. * list won't change. This doesn't solve the problem
  2840. * of setting the slave's MTU while it is
  2841. * transmitting, but the assumption is that the base
  2842. * driver can handle that.
  2843. *
  2844. * TODO: figure out a way to safely iterate the slaves
  2845. * list, but without holding a lock around the actual
  2846. * call to the base driver.
  2847. */
  2848. bond_for_each_slave(bond, slave, iter) {
  2849. pr_debug("s %p c_m %p\n",
  2850. slave,
  2851. slave->dev->netdev_ops->ndo_change_mtu);
  2852. res = dev_set_mtu(slave->dev, new_mtu);
  2853. if (res) {
  2854. /* If we failed to set the slave's mtu to the new value
  2855. * we must abort the operation even in ACTIVE_BACKUP
  2856. * mode, because if we allow the backup slaves to have
  2857. * different mtu values than the active slave we'll
  2858. * need to change their mtu when doing a failover. That
  2859. * means changing their mtu from timer context, which
  2860. * is probably not a good idea.
  2861. */
  2862. pr_debug("err %d %s\n", res, slave->dev->name);
  2863. goto unwind;
  2864. }
  2865. }
  2866. bond_dev->mtu = new_mtu;
  2867. return 0;
  2868. unwind:
  2869. /* unwind from head to the slave that failed */
  2870. bond_for_each_slave(bond, rollback_slave, iter) {
  2871. int tmp_res;
  2872. if (rollback_slave == slave)
  2873. break;
  2874. tmp_res = dev_set_mtu(rollback_slave->dev, bond_dev->mtu);
  2875. if (tmp_res) {
  2876. pr_debug("unwind err %d dev %s\n",
  2877. tmp_res, rollback_slave->dev->name);
  2878. }
  2879. }
  2880. return res;
  2881. }
  2882. /*
  2883. * Change HW address
  2884. *
  2885. * Note that many devices must be down to change the HW address, and
  2886. * downing the master releases all slaves. We can make bonds full of
  2887. * bonding devices to test this, however.
  2888. */
  2889. static int bond_set_mac_address(struct net_device *bond_dev, void *addr)
  2890. {
  2891. struct bonding *bond = netdev_priv(bond_dev);
  2892. struct slave *slave, *rollback_slave;
  2893. struct sockaddr *sa = addr, tmp_sa;
  2894. struct list_head *iter;
  2895. int res = 0;
  2896. if (bond->params.mode == BOND_MODE_ALB)
  2897. return bond_alb_set_mac_address(bond_dev, addr);
  2898. pr_debug("bond=%p, name=%s\n",
  2899. bond, bond_dev ? bond_dev->name : "None");
  2900. /* If fail_over_mac is enabled, do nothing and return success.
  2901. * Returning an error causes ifenslave to fail.
  2902. */
  2903. if (bond->params.fail_over_mac)
  2904. return 0;
  2905. if (!is_valid_ether_addr(sa->sa_data))
  2906. return -EADDRNOTAVAIL;
  2907. /* Can't hold bond->lock with bh disabled here since
  2908. * some base drivers panic. On the other hand we can't
  2909. * hold bond->lock without bh disabled because we'll
  2910. * deadlock. The only solution is to rely on the fact
  2911. * that we're under rtnl_lock here, and the slaves
  2912. * list won't change. This doesn't solve the problem
  2913. * of setting the slave's hw address while it is
  2914. * transmitting, but the assumption is that the base
  2915. * driver can handle that.
  2916. *
  2917. * TODO: figure out a way to safely iterate the slaves
  2918. * list, but without holding a lock around the actual
  2919. * call to the base driver.
  2920. */
  2921. bond_for_each_slave(bond, slave, iter) {
  2922. const struct net_device_ops *slave_ops = slave->dev->netdev_ops;
  2923. pr_debug("slave %p %s\n", slave, slave->dev->name);
  2924. if (slave_ops->ndo_set_mac_address == NULL) {
  2925. res = -EOPNOTSUPP;
  2926. pr_debug("EOPNOTSUPP %s\n", slave->dev->name);
  2927. goto unwind;
  2928. }
  2929. res = dev_set_mac_address(slave->dev, addr);
  2930. if (res) {
  2931. /* TODO: consider downing the slave
  2932. * and retry ?
  2933. * User should expect communications
  2934. * breakage anyway until ARP finish
  2935. * updating, so...
  2936. */
  2937. pr_debug("err %d %s\n", res, slave->dev->name);
  2938. goto unwind;
  2939. }
  2940. }
  2941. /* success */
  2942. memcpy(bond_dev->dev_addr, sa->sa_data, bond_dev->addr_len);
  2943. return 0;
  2944. unwind:
  2945. memcpy(tmp_sa.sa_data, bond_dev->dev_addr, bond_dev->addr_len);
  2946. tmp_sa.sa_family = bond_dev->type;
  2947. /* unwind from head to the slave that failed */
  2948. bond_for_each_slave(bond, rollback_slave, iter) {
  2949. int tmp_res;
  2950. if (rollback_slave == slave)
  2951. break;
  2952. tmp_res = dev_set_mac_address(rollback_slave->dev, &tmp_sa);
  2953. if (tmp_res) {
  2954. pr_debug("unwind err %d dev %s\n",
  2955. tmp_res, rollback_slave->dev->name);
  2956. }
  2957. }
  2958. return res;
  2959. }
  2960. /**
  2961. * bond_xmit_slave_id - transmit skb through slave with slave_id
  2962. * @bond: bonding device that is transmitting
  2963. * @skb: buffer to transmit
  2964. * @slave_id: slave id up to slave_cnt-1 through which to transmit
  2965. *
  2966. * This function tries to transmit through slave with slave_id but in case
  2967. * it fails, it tries to find the first available slave for transmission.
  2968. * The skb is consumed in all cases, thus the function is void.
  2969. */
  2970. void bond_xmit_slave_id(struct bonding *bond, struct sk_buff *skb, int slave_id)
  2971. {
  2972. struct list_head *iter;
  2973. struct slave *slave;
  2974. int i = slave_id;
  2975. /* Here we start from the slave with slave_id */
  2976. bond_for_each_slave_rcu(bond, slave, iter) {
  2977. if (--i < 0) {
  2978. if (slave_can_tx(slave)) {
  2979. bond_dev_queue_xmit(bond, skb, slave->dev);
  2980. return;
  2981. }
  2982. }
  2983. }
  2984. /* Here we start from the first slave up to slave_id */
  2985. i = slave_id;
  2986. bond_for_each_slave_rcu(bond, slave, iter) {
  2987. if (--i < 0)
  2988. break;
  2989. if (slave_can_tx(slave)) {
  2990. bond_dev_queue_xmit(bond, skb, slave->dev);
  2991. return;
  2992. }
  2993. }
  2994. /* no slave that can tx has been found */
  2995. kfree_skb(skb);
  2996. }
  2997. /**
  2998. * bond_rr_gen_slave_id - generate slave id based on packets_per_slave
  2999. * @bond: bonding device to use
  3000. *
  3001. * Based on the value of the bonding device's packets_per_slave parameter
  3002. * this function generates a slave id, which is usually used as the next
  3003. * slave to transmit through.
  3004. */
  3005. static u32 bond_rr_gen_slave_id(struct bonding *bond)
  3006. {
  3007. int packets_per_slave = bond->params.packets_per_slave;
  3008. u32 slave_id;
  3009. switch (packets_per_slave) {
  3010. case 0:
  3011. slave_id = prandom_u32();
  3012. break;
  3013. case 1:
  3014. slave_id = bond->rr_tx_counter;
  3015. break;
  3016. default:
  3017. slave_id = reciprocal_divide(bond->rr_tx_counter,
  3018. packets_per_slave);
  3019. break;
  3020. }
  3021. bond->rr_tx_counter++;
  3022. return slave_id;
  3023. }
  3024. static int bond_xmit_roundrobin(struct sk_buff *skb, struct net_device *bond_dev)
  3025. {
  3026. struct bonding *bond = netdev_priv(bond_dev);
  3027. struct iphdr *iph = ip_hdr(skb);
  3028. struct slave *slave;
  3029. u32 slave_id;
  3030. /* Start with the curr_active_slave that joined the bond as the
  3031. * default for sending IGMP traffic. For failover purposes one
  3032. * needs to maintain some consistency for the interface that will
  3033. * send the join/membership reports. The curr_active_slave found
  3034. * will send all of this type of traffic.
  3035. */
  3036. if (iph->protocol == IPPROTO_IGMP && skb->protocol == htons(ETH_P_IP)) {
  3037. slave = rcu_dereference(bond->curr_active_slave);
  3038. if (slave && slave_can_tx(slave))
  3039. bond_dev_queue_xmit(bond, skb, slave->dev);
  3040. else
  3041. bond_xmit_slave_id(bond, skb, 0);
  3042. } else {
  3043. slave_id = bond_rr_gen_slave_id(bond);
  3044. bond_xmit_slave_id(bond, skb, slave_id % bond->slave_cnt);
  3045. }
  3046. return NETDEV_TX_OK;
  3047. }
  3048. /*
  3049. * in active-backup mode, we know that bond->curr_active_slave is always valid if
  3050. * the bond has a usable interface.
  3051. */
  3052. static int bond_xmit_activebackup(struct sk_buff *skb, struct net_device *bond_dev)
  3053. {
  3054. struct bonding *bond = netdev_priv(bond_dev);
  3055. struct slave *slave;
  3056. slave = rcu_dereference(bond->curr_active_slave);
  3057. if (slave)
  3058. bond_dev_queue_xmit(bond, skb, slave->dev);
  3059. else
  3060. kfree_skb(skb);
  3061. return NETDEV_TX_OK;
  3062. }
  3063. /* In bond_xmit_xor() , we determine the output device by using a pre-
  3064. * determined xmit_hash_policy(), If the selected device is not enabled,
  3065. * find the next active slave.
  3066. */
  3067. static int bond_xmit_xor(struct sk_buff *skb, struct net_device *bond_dev)
  3068. {
  3069. struct bonding *bond = netdev_priv(bond_dev);
  3070. bond_xmit_slave_id(bond, skb, bond_xmit_hash(bond, skb, bond->slave_cnt));
  3071. return NETDEV_TX_OK;
  3072. }
  3073. /* in broadcast mode, we send everything to all usable interfaces. */
  3074. static int bond_xmit_broadcast(struct sk_buff *skb, struct net_device *bond_dev)
  3075. {
  3076. struct bonding *bond = netdev_priv(bond_dev);
  3077. struct slave *slave = NULL;
  3078. struct list_head *iter;
  3079. bond_for_each_slave_rcu(bond, slave, iter) {
  3080. if (bond_is_last_slave(bond, slave))
  3081. break;
  3082. if (IS_UP(slave->dev) && slave->link == BOND_LINK_UP) {
  3083. struct sk_buff *skb2 = skb_clone(skb, GFP_ATOMIC);
  3084. if (!skb2) {
  3085. pr_err("%s: Error: bond_xmit_broadcast(): skb_clone() failed\n",
  3086. bond_dev->name);
  3087. continue;
  3088. }
  3089. /* bond_dev_queue_xmit always returns 0 */
  3090. bond_dev_queue_xmit(bond, skb2, slave->dev);
  3091. }
  3092. }
  3093. if (slave && IS_UP(slave->dev) && slave->link == BOND_LINK_UP)
  3094. bond_dev_queue_xmit(bond, skb, slave->dev);
  3095. else
  3096. kfree_skb(skb);
  3097. return NETDEV_TX_OK;
  3098. }
  3099. /*------------------------- Device initialization ---------------------------*/
  3100. /*
  3101. * Lookup the slave that corresponds to a qid
  3102. */
  3103. static inline int bond_slave_override(struct bonding *bond,
  3104. struct sk_buff *skb)
  3105. {
  3106. struct slave *slave = NULL;
  3107. struct slave *check_slave;
  3108. struct list_head *iter;
  3109. int res = 1;
  3110. if (!skb->queue_mapping)
  3111. return 1;
  3112. /* Find out if any slaves have the same mapping as this skb. */
  3113. bond_for_each_slave_rcu(bond, check_slave, iter) {
  3114. if (check_slave->queue_id == skb->queue_mapping) {
  3115. slave = check_slave;
  3116. break;
  3117. }
  3118. }
  3119. /* If the slave isn't UP, use default transmit policy. */
  3120. if (slave && slave->queue_id && IS_UP(slave->dev) &&
  3121. (slave->link == BOND_LINK_UP)) {
  3122. res = bond_dev_queue_xmit(bond, skb, slave->dev);
  3123. }
  3124. return res;
  3125. }
  3126. static u16 bond_select_queue(struct net_device *dev, struct sk_buff *skb)
  3127. {
  3128. /*
  3129. * This helper function exists to help dev_pick_tx get the correct
  3130. * destination queue. Using a helper function skips a call to
  3131. * skb_tx_hash and will put the skbs in the queue we expect on their
  3132. * way down to the bonding driver.
  3133. */
  3134. u16 txq = skb_rx_queue_recorded(skb) ? skb_get_rx_queue(skb) : 0;
  3135. /*
  3136. * Save the original txq to restore before passing to the driver
  3137. */
  3138. qdisc_skb_cb(skb)->slave_dev_queue_mapping = skb->queue_mapping;
  3139. if (unlikely(txq >= dev->real_num_tx_queues)) {
  3140. do {
  3141. txq -= dev->real_num_tx_queues;
  3142. } while (txq >= dev->real_num_tx_queues);
  3143. }
  3144. return txq;
  3145. }
  3146. static netdev_tx_t __bond_start_xmit(struct sk_buff *skb, struct net_device *dev)
  3147. {
  3148. struct bonding *bond = netdev_priv(dev);
  3149. if (TX_QUEUE_OVERRIDE(bond->params.mode)) {
  3150. if (!bond_slave_override(bond, skb))
  3151. return NETDEV_TX_OK;
  3152. }
  3153. switch (bond->params.mode) {
  3154. case BOND_MODE_ROUNDROBIN:
  3155. return bond_xmit_roundrobin(skb, dev);
  3156. case BOND_MODE_ACTIVEBACKUP:
  3157. return bond_xmit_activebackup(skb, dev);
  3158. case BOND_MODE_XOR:
  3159. return bond_xmit_xor(skb, dev);
  3160. case BOND_MODE_BROADCAST:
  3161. return bond_xmit_broadcast(skb, dev);
  3162. case BOND_MODE_8023AD:
  3163. return bond_3ad_xmit_xor(skb, dev);
  3164. case BOND_MODE_ALB:
  3165. case BOND_MODE_TLB:
  3166. return bond_alb_xmit(skb, dev);
  3167. default:
  3168. /* Should never happen, mode already checked */
  3169. pr_err("%s: Error: Unknown bonding mode %d\n",
  3170. dev->name, bond->params.mode);
  3171. WARN_ON_ONCE(1);
  3172. kfree_skb(skb);
  3173. return NETDEV_TX_OK;
  3174. }
  3175. }
  3176. static netdev_tx_t bond_start_xmit(struct sk_buff *skb, struct net_device *dev)
  3177. {
  3178. struct bonding *bond = netdev_priv(dev);
  3179. netdev_tx_t ret = NETDEV_TX_OK;
  3180. /*
  3181. * If we risk deadlock from transmitting this in the
  3182. * netpoll path, tell netpoll to queue the frame for later tx
  3183. */
  3184. if (is_netpoll_tx_blocked(dev))
  3185. return NETDEV_TX_BUSY;
  3186. rcu_read_lock();
  3187. if (bond_has_slaves(bond))
  3188. ret = __bond_start_xmit(skb, dev);
  3189. else
  3190. kfree_skb(skb);
  3191. rcu_read_unlock();
  3192. return ret;
  3193. }
  3194. static int bond_ethtool_get_settings(struct net_device *bond_dev,
  3195. struct ethtool_cmd *ecmd)
  3196. {
  3197. struct bonding *bond = netdev_priv(bond_dev);
  3198. unsigned long speed = 0;
  3199. struct list_head *iter;
  3200. struct slave *slave;
  3201. ecmd->duplex = DUPLEX_UNKNOWN;
  3202. ecmd->port = PORT_OTHER;
  3203. /* Since SLAVE_IS_OK returns false for all inactive or down slaves, we
  3204. * do not need to check mode. Though link speed might not represent
  3205. * the true receive or transmit bandwidth (not all modes are symmetric)
  3206. * this is an accurate maximum.
  3207. */
  3208. read_lock(&bond->lock);
  3209. bond_for_each_slave(bond, slave, iter) {
  3210. if (SLAVE_IS_OK(slave)) {
  3211. if (slave->speed != SPEED_UNKNOWN)
  3212. speed += slave->speed;
  3213. if (ecmd->duplex == DUPLEX_UNKNOWN &&
  3214. slave->duplex != DUPLEX_UNKNOWN)
  3215. ecmd->duplex = slave->duplex;
  3216. }
  3217. }
  3218. ethtool_cmd_speed_set(ecmd, speed ? : SPEED_UNKNOWN);
  3219. read_unlock(&bond->lock);
  3220. return 0;
  3221. }
  3222. static void bond_ethtool_get_drvinfo(struct net_device *bond_dev,
  3223. struct ethtool_drvinfo *drvinfo)
  3224. {
  3225. strlcpy(drvinfo->driver, DRV_NAME, sizeof(drvinfo->driver));
  3226. strlcpy(drvinfo->version, DRV_VERSION, sizeof(drvinfo->version));
  3227. snprintf(drvinfo->fw_version, sizeof(drvinfo->fw_version), "%d",
  3228. BOND_ABI_VERSION);
  3229. }
  3230. static const struct ethtool_ops bond_ethtool_ops = {
  3231. .get_drvinfo = bond_ethtool_get_drvinfo,
  3232. .get_settings = bond_ethtool_get_settings,
  3233. .get_link = ethtool_op_get_link,
  3234. };
  3235. static const struct net_device_ops bond_netdev_ops = {
  3236. .ndo_init = bond_init,
  3237. .ndo_uninit = bond_uninit,
  3238. .ndo_open = bond_open,
  3239. .ndo_stop = bond_close,
  3240. .ndo_start_xmit = bond_start_xmit,
  3241. .ndo_select_queue = bond_select_queue,
  3242. .ndo_get_stats64 = bond_get_stats,
  3243. .ndo_do_ioctl = bond_do_ioctl,
  3244. .ndo_change_rx_flags = bond_change_rx_flags,
  3245. .ndo_set_rx_mode = bond_set_rx_mode,
  3246. .ndo_change_mtu = bond_change_mtu,
  3247. .ndo_set_mac_address = bond_set_mac_address,
  3248. .ndo_neigh_setup = bond_neigh_setup,
  3249. .ndo_vlan_rx_add_vid = bond_vlan_rx_add_vid,
  3250. .ndo_vlan_rx_kill_vid = bond_vlan_rx_kill_vid,
  3251. #ifdef CONFIG_NET_POLL_CONTROLLER
  3252. .ndo_netpoll_setup = bond_netpoll_setup,
  3253. .ndo_netpoll_cleanup = bond_netpoll_cleanup,
  3254. .ndo_poll_controller = bond_poll_controller,
  3255. #endif
  3256. .ndo_add_slave = bond_enslave,
  3257. .ndo_del_slave = bond_release,
  3258. .ndo_fix_features = bond_fix_features,
  3259. };
  3260. static const struct device_type bond_type = {
  3261. .name = "bond",
  3262. };
  3263. static void bond_destructor(struct net_device *bond_dev)
  3264. {
  3265. struct bonding *bond = netdev_priv(bond_dev);
  3266. if (bond->wq)
  3267. destroy_workqueue(bond->wq);
  3268. free_netdev(bond_dev);
  3269. }
  3270. void bond_setup(struct net_device *bond_dev)
  3271. {
  3272. struct bonding *bond = netdev_priv(bond_dev);
  3273. /* initialize rwlocks */
  3274. rwlock_init(&bond->lock);
  3275. rwlock_init(&bond->curr_slave_lock);
  3276. bond->params = bonding_defaults;
  3277. /* Initialize pointers */
  3278. bond->dev = bond_dev;
  3279. /* Initialize the device entry points */
  3280. ether_setup(bond_dev);
  3281. bond_dev->netdev_ops = &bond_netdev_ops;
  3282. bond_dev->ethtool_ops = &bond_ethtool_ops;
  3283. bond_dev->destructor = bond_destructor;
  3284. SET_NETDEV_DEVTYPE(bond_dev, &bond_type);
  3285. /* Initialize the device options */
  3286. bond_dev->tx_queue_len = 0;
  3287. bond_dev->flags |= IFF_MASTER|IFF_MULTICAST;
  3288. bond_dev->priv_flags |= IFF_BONDING;
  3289. bond_dev->priv_flags &= ~(IFF_XMIT_DST_RELEASE | IFF_TX_SKB_SHARING);
  3290. /* At first, we block adding VLANs. That's the only way to
  3291. * prevent problems that occur when adding VLANs over an
  3292. * empty bond. The block will be removed once non-challenged
  3293. * slaves are enslaved.
  3294. */
  3295. bond_dev->features |= NETIF_F_VLAN_CHALLENGED;
  3296. /* don't acquire bond device's netif_tx_lock when
  3297. * transmitting */
  3298. bond_dev->features |= NETIF_F_LLTX;
  3299. /* By default, we declare the bond to be fully
  3300. * VLAN hardware accelerated capable. Special
  3301. * care is taken in the various xmit functions
  3302. * when there are slaves that are not hw accel
  3303. * capable
  3304. */
  3305. bond_dev->hw_features = BOND_VLAN_FEATURES |
  3306. NETIF_F_HW_VLAN_CTAG_TX |
  3307. NETIF_F_HW_VLAN_CTAG_RX |
  3308. NETIF_F_HW_VLAN_CTAG_FILTER;
  3309. bond_dev->hw_features &= ~(NETIF_F_ALL_CSUM & ~NETIF_F_HW_CSUM);
  3310. bond_dev->features |= bond_dev->hw_features;
  3311. }
  3312. /*
  3313. * Destroy a bonding device.
  3314. * Must be under rtnl_lock when this function is called.
  3315. */
  3316. static void bond_uninit(struct net_device *bond_dev)
  3317. {
  3318. struct bonding *bond = netdev_priv(bond_dev);
  3319. struct list_head *iter;
  3320. struct slave *slave;
  3321. bond_netpoll_cleanup(bond_dev);
  3322. /* Release the bonded slaves */
  3323. bond_for_each_slave(bond, slave, iter)
  3324. __bond_release_one(bond_dev, slave->dev, true);
  3325. pr_info("%s: released all slaves\n", bond_dev->name);
  3326. list_del(&bond->bond_list);
  3327. bond_debug_unregister(bond);
  3328. }
  3329. /*------------------------- Module initialization ---------------------------*/
  3330. /*
  3331. * Convert string input module parms. Accept either the
  3332. * number of the mode or its string name. A bit complicated because
  3333. * some mode names are substrings of other names, and calls from sysfs
  3334. * may have whitespace in the name (trailing newlines, for example).
  3335. */
  3336. int bond_parse_parm(const char *buf, const struct bond_parm_tbl *tbl)
  3337. {
  3338. int modeint = -1, i, rv;
  3339. char *p, modestr[BOND_MAX_MODENAME_LEN + 1] = { 0, };
  3340. for (p = (char *)buf; *p; p++)
  3341. if (!(isdigit(*p) || isspace(*p)))
  3342. break;
  3343. if (*p)
  3344. rv = sscanf(buf, "%20s", modestr);
  3345. else
  3346. rv = sscanf(buf, "%d", &modeint);
  3347. if (!rv)
  3348. return -1;
  3349. for (i = 0; tbl[i].modename; i++) {
  3350. if (modeint == tbl[i].mode)
  3351. return tbl[i].mode;
  3352. if (strcmp(modestr, tbl[i].modename) == 0)
  3353. return tbl[i].mode;
  3354. }
  3355. return -1;
  3356. }
  3357. static int bond_check_params(struct bond_params *params)
  3358. {
  3359. int arp_validate_value, fail_over_mac_value, primary_reselect_value, i;
  3360. int arp_all_targets_value;
  3361. /*
  3362. * Convert string parameters.
  3363. */
  3364. if (mode) {
  3365. bond_mode = bond_parse_parm(mode, bond_mode_tbl);
  3366. if (bond_mode == -1) {
  3367. pr_err("Error: Invalid bonding mode \"%s\"\n",
  3368. mode == NULL ? "NULL" : mode);
  3369. return -EINVAL;
  3370. }
  3371. }
  3372. if (xmit_hash_policy) {
  3373. if ((bond_mode != BOND_MODE_XOR) &&
  3374. (bond_mode != BOND_MODE_8023AD)) {
  3375. pr_info("xmit_hash_policy param is irrelevant in mode %s\n",
  3376. bond_mode_name(bond_mode));
  3377. } else {
  3378. xmit_hashtype = bond_parse_parm(xmit_hash_policy,
  3379. xmit_hashtype_tbl);
  3380. if (xmit_hashtype == -1) {
  3381. pr_err("Error: Invalid xmit_hash_policy \"%s\"\n",
  3382. xmit_hash_policy == NULL ? "NULL" :
  3383. xmit_hash_policy);
  3384. return -EINVAL;
  3385. }
  3386. }
  3387. }
  3388. if (lacp_rate) {
  3389. if (bond_mode != BOND_MODE_8023AD) {
  3390. pr_info("lacp_rate param is irrelevant in mode %s\n",
  3391. bond_mode_name(bond_mode));
  3392. } else {
  3393. lacp_fast = bond_parse_parm(lacp_rate, bond_lacp_tbl);
  3394. if (lacp_fast == -1) {
  3395. pr_err("Error: Invalid lacp rate \"%s\"\n",
  3396. lacp_rate == NULL ? "NULL" : lacp_rate);
  3397. return -EINVAL;
  3398. }
  3399. }
  3400. }
  3401. if (ad_select) {
  3402. params->ad_select = bond_parse_parm(ad_select, ad_select_tbl);
  3403. if (params->ad_select == -1) {
  3404. pr_err("Error: Invalid ad_select \"%s\"\n",
  3405. ad_select == NULL ? "NULL" : ad_select);
  3406. return -EINVAL;
  3407. }
  3408. if (bond_mode != BOND_MODE_8023AD) {
  3409. pr_warning("ad_select param only affects 802.3ad mode\n");
  3410. }
  3411. } else {
  3412. params->ad_select = BOND_AD_STABLE;
  3413. }
  3414. if (max_bonds < 0) {
  3415. pr_warning("Warning: max_bonds (%d) not in range %d-%d, so it was reset to BOND_DEFAULT_MAX_BONDS (%d)\n",
  3416. max_bonds, 0, INT_MAX, BOND_DEFAULT_MAX_BONDS);
  3417. max_bonds = BOND_DEFAULT_MAX_BONDS;
  3418. }
  3419. if (miimon < 0) {
  3420. pr_warning("Warning: miimon module parameter (%d), not in range 0-%d, so it was reset to %d\n",
  3421. miimon, INT_MAX, BOND_LINK_MON_INTERV);
  3422. miimon = BOND_LINK_MON_INTERV;
  3423. }
  3424. if (updelay < 0) {
  3425. pr_warning("Warning: updelay module parameter (%d), not in range 0-%d, so it was reset to 0\n",
  3426. updelay, INT_MAX);
  3427. updelay = 0;
  3428. }
  3429. if (downdelay < 0) {
  3430. pr_warning("Warning: downdelay module parameter (%d), not in range 0-%d, so it was reset to 0\n",
  3431. downdelay, INT_MAX);
  3432. downdelay = 0;
  3433. }
  3434. if ((use_carrier != 0) && (use_carrier != 1)) {
  3435. pr_warning("Warning: use_carrier module parameter (%d), not of valid value (0/1), so it was set to 1\n",
  3436. use_carrier);
  3437. use_carrier = 1;
  3438. }
  3439. if (num_peer_notif < 0 || num_peer_notif > 255) {
  3440. pr_warning("Warning: num_grat_arp/num_unsol_na (%d) not in range 0-255 so it was reset to 1\n",
  3441. num_peer_notif);
  3442. num_peer_notif = 1;
  3443. }
  3444. /* reset values for 802.3ad */
  3445. if (bond_mode == BOND_MODE_8023AD) {
  3446. if (!miimon) {
  3447. pr_warning("Warning: miimon must be specified, otherwise bonding will not detect link failure, speed and duplex which are essential for 802.3ad operation\n");
  3448. pr_warning("Forcing miimon to 100msec\n");
  3449. miimon = BOND_DEFAULT_MIIMON;
  3450. }
  3451. }
  3452. if (tx_queues < 1 || tx_queues > 255) {
  3453. pr_warning("Warning: tx_queues (%d) should be between "
  3454. "1 and 255, resetting to %d\n",
  3455. tx_queues, BOND_DEFAULT_TX_QUEUES);
  3456. tx_queues = BOND_DEFAULT_TX_QUEUES;
  3457. }
  3458. if ((all_slaves_active != 0) && (all_slaves_active != 1)) {
  3459. pr_warning("Warning: all_slaves_active module parameter (%d), "
  3460. "not of valid value (0/1), so it was set to "
  3461. "0\n", all_slaves_active);
  3462. all_slaves_active = 0;
  3463. }
  3464. if (resend_igmp < 0 || resend_igmp > 255) {
  3465. pr_warning("Warning: resend_igmp (%d) should be between "
  3466. "0 and 255, resetting to %d\n",
  3467. resend_igmp, BOND_DEFAULT_RESEND_IGMP);
  3468. resend_igmp = BOND_DEFAULT_RESEND_IGMP;
  3469. }
  3470. if (packets_per_slave < 0 || packets_per_slave > USHRT_MAX) {
  3471. pr_warn("Warning: packets_per_slave (%d) should be between 0 and %u resetting to 1\n",
  3472. packets_per_slave, USHRT_MAX);
  3473. packets_per_slave = 1;
  3474. }
  3475. /* reset values for TLB/ALB */
  3476. if ((bond_mode == BOND_MODE_TLB) ||
  3477. (bond_mode == BOND_MODE_ALB)) {
  3478. if (!miimon) {
  3479. pr_warning("Warning: miimon must be specified, otherwise bonding will not detect link failure and link speed which are essential for TLB/ALB load balancing\n");
  3480. pr_warning("Forcing miimon to 100msec\n");
  3481. miimon = BOND_DEFAULT_MIIMON;
  3482. }
  3483. }
  3484. if (bond_mode == BOND_MODE_ALB) {
  3485. pr_notice("In ALB mode you might experience client disconnections upon reconnection of a link if the bonding module updelay parameter (%d msec) is incompatible with the forwarding delay time of the switch\n",
  3486. updelay);
  3487. }
  3488. if (!miimon) {
  3489. if (updelay || downdelay) {
  3490. /* just warn the user the up/down delay will have
  3491. * no effect since miimon is zero...
  3492. */
  3493. pr_warning("Warning: miimon module parameter not set and updelay (%d) or downdelay (%d) module parameter is set; updelay and downdelay have no effect unless miimon is set\n",
  3494. updelay, downdelay);
  3495. }
  3496. } else {
  3497. /* don't allow arp monitoring */
  3498. if (arp_interval) {
  3499. pr_warning("Warning: miimon (%d) and arp_interval (%d) can't be used simultaneously, disabling ARP monitoring\n",
  3500. miimon, arp_interval);
  3501. arp_interval = 0;
  3502. }
  3503. if ((updelay % miimon) != 0) {
  3504. pr_warning("Warning: updelay (%d) is not a multiple of miimon (%d), updelay rounded to %d ms\n",
  3505. updelay, miimon,
  3506. (updelay / miimon) * miimon);
  3507. }
  3508. updelay /= miimon;
  3509. if ((downdelay % miimon) != 0) {
  3510. pr_warning("Warning: downdelay (%d) is not a multiple of miimon (%d), downdelay rounded to %d ms\n",
  3511. downdelay, miimon,
  3512. (downdelay / miimon) * miimon);
  3513. }
  3514. downdelay /= miimon;
  3515. }
  3516. if (arp_interval < 0) {
  3517. pr_warning("Warning: arp_interval module parameter (%d) , not in range 0-%d, so it was reset to %d\n",
  3518. arp_interval, INT_MAX, BOND_LINK_ARP_INTERV);
  3519. arp_interval = BOND_LINK_ARP_INTERV;
  3520. }
  3521. for (arp_ip_count = 0, i = 0;
  3522. (arp_ip_count < BOND_MAX_ARP_TARGETS) && arp_ip_target[i]; i++) {
  3523. /* not complete check, but should be good enough to
  3524. catch mistakes */
  3525. __be32 ip;
  3526. if (!in4_pton(arp_ip_target[i], -1, (u8 *)&ip, -1, NULL) ||
  3527. IS_IP_TARGET_UNUSABLE_ADDRESS(ip)) {
  3528. pr_warning("Warning: bad arp_ip_target module parameter (%s), ARP monitoring will not be performed\n",
  3529. arp_ip_target[i]);
  3530. arp_interval = 0;
  3531. } else {
  3532. if (bond_get_targets_ip(arp_target, ip) == -1)
  3533. arp_target[arp_ip_count++] = ip;
  3534. else
  3535. pr_warning("Warning: duplicate address %pI4 in arp_ip_target, skipping\n",
  3536. &ip);
  3537. }
  3538. }
  3539. if (arp_interval && !arp_ip_count) {
  3540. /* don't allow arping if no arp_ip_target given... */
  3541. pr_warning("Warning: arp_interval module parameter (%d) specified without providing an arp_ip_target parameter, arp_interval was reset to 0\n",
  3542. arp_interval);
  3543. arp_interval = 0;
  3544. }
  3545. if (arp_validate) {
  3546. if (bond_mode != BOND_MODE_ACTIVEBACKUP) {
  3547. pr_err("arp_validate only supported in active-backup mode\n");
  3548. return -EINVAL;
  3549. }
  3550. if (!arp_interval) {
  3551. pr_err("arp_validate requires arp_interval\n");
  3552. return -EINVAL;
  3553. }
  3554. arp_validate_value = bond_parse_parm(arp_validate,
  3555. arp_validate_tbl);
  3556. if (arp_validate_value == -1) {
  3557. pr_err("Error: invalid arp_validate \"%s\"\n",
  3558. arp_validate == NULL ? "NULL" : arp_validate);
  3559. return -EINVAL;
  3560. }
  3561. } else
  3562. arp_validate_value = 0;
  3563. arp_all_targets_value = 0;
  3564. if (arp_all_targets) {
  3565. arp_all_targets_value = bond_parse_parm(arp_all_targets,
  3566. arp_all_targets_tbl);
  3567. if (arp_all_targets_value == -1) {
  3568. pr_err("Error: invalid arp_all_targets_value \"%s\"\n",
  3569. arp_all_targets);
  3570. arp_all_targets_value = 0;
  3571. }
  3572. }
  3573. if (miimon) {
  3574. pr_info("MII link monitoring set to %d ms\n", miimon);
  3575. } else if (arp_interval) {
  3576. pr_info("ARP monitoring set to %d ms, validate %s, with %d target(s):",
  3577. arp_interval,
  3578. arp_validate_tbl[arp_validate_value].modename,
  3579. arp_ip_count);
  3580. for (i = 0; i < arp_ip_count; i++)
  3581. pr_info(" %s", arp_ip_target[i]);
  3582. pr_info("\n");
  3583. } else if (max_bonds) {
  3584. /* miimon and arp_interval not set, we need one so things
  3585. * work as expected, see bonding.txt for details
  3586. */
  3587. pr_debug("Warning: either miimon or arp_interval and arp_ip_target module parameters must be specified, otherwise bonding will not detect link failures! see bonding.txt for details.\n");
  3588. }
  3589. if (primary && !USES_PRIMARY(bond_mode)) {
  3590. /* currently, using a primary only makes sense
  3591. * in active backup, TLB or ALB modes
  3592. */
  3593. pr_warning("Warning: %s primary device specified but has no effect in %s mode\n",
  3594. primary, bond_mode_name(bond_mode));
  3595. primary = NULL;
  3596. }
  3597. if (primary && primary_reselect) {
  3598. primary_reselect_value = bond_parse_parm(primary_reselect,
  3599. pri_reselect_tbl);
  3600. if (primary_reselect_value == -1) {
  3601. pr_err("Error: Invalid primary_reselect \"%s\"\n",
  3602. primary_reselect ==
  3603. NULL ? "NULL" : primary_reselect);
  3604. return -EINVAL;
  3605. }
  3606. } else {
  3607. primary_reselect_value = BOND_PRI_RESELECT_ALWAYS;
  3608. }
  3609. if (fail_over_mac) {
  3610. fail_over_mac_value = bond_parse_parm(fail_over_mac,
  3611. fail_over_mac_tbl);
  3612. if (fail_over_mac_value == -1) {
  3613. pr_err("Error: invalid fail_over_mac \"%s\"\n",
  3614. arp_validate == NULL ? "NULL" : arp_validate);
  3615. return -EINVAL;
  3616. }
  3617. if (bond_mode != BOND_MODE_ACTIVEBACKUP)
  3618. pr_warning("Warning: fail_over_mac only affects active-backup mode.\n");
  3619. } else {
  3620. fail_over_mac_value = BOND_FOM_NONE;
  3621. }
  3622. /* fill params struct with the proper values */
  3623. params->mode = bond_mode;
  3624. params->xmit_policy = xmit_hashtype;
  3625. params->miimon = miimon;
  3626. params->num_peer_notif = num_peer_notif;
  3627. params->arp_interval = arp_interval;
  3628. params->arp_validate = arp_validate_value;
  3629. params->arp_all_targets = arp_all_targets_value;
  3630. params->updelay = updelay;
  3631. params->downdelay = downdelay;
  3632. params->use_carrier = use_carrier;
  3633. params->lacp_fast = lacp_fast;
  3634. params->primary[0] = 0;
  3635. params->primary_reselect = primary_reselect_value;
  3636. params->fail_over_mac = fail_over_mac_value;
  3637. params->tx_queues = tx_queues;
  3638. params->all_slaves_active = all_slaves_active;
  3639. params->resend_igmp = resend_igmp;
  3640. params->min_links = min_links;
  3641. params->lp_interval = BOND_ALB_DEFAULT_LP_INTERVAL;
  3642. if (packets_per_slave > 1)
  3643. params->packets_per_slave = reciprocal_value(packets_per_slave);
  3644. else
  3645. params->packets_per_slave = packets_per_slave;
  3646. if (primary) {
  3647. strncpy(params->primary, primary, IFNAMSIZ);
  3648. params->primary[IFNAMSIZ - 1] = 0;
  3649. }
  3650. memcpy(params->arp_targets, arp_target, sizeof(arp_target));
  3651. return 0;
  3652. }
  3653. static struct lock_class_key bonding_netdev_xmit_lock_key;
  3654. static struct lock_class_key bonding_netdev_addr_lock_key;
  3655. static struct lock_class_key bonding_tx_busylock_key;
  3656. static void bond_set_lockdep_class_one(struct net_device *dev,
  3657. struct netdev_queue *txq,
  3658. void *_unused)
  3659. {
  3660. lockdep_set_class(&txq->_xmit_lock,
  3661. &bonding_netdev_xmit_lock_key);
  3662. }
  3663. static void bond_set_lockdep_class(struct net_device *dev)
  3664. {
  3665. lockdep_set_class(&dev->addr_list_lock,
  3666. &bonding_netdev_addr_lock_key);
  3667. netdev_for_each_tx_queue(dev, bond_set_lockdep_class_one, NULL);
  3668. dev->qdisc_tx_busylock = &bonding_tx_busylock_key;
  3669. }
  3670. /*
  3671. * Called from registration process
  3672. */
  3673. static int bond_init(struct net_device *bond_dev)
  3674. {
  3675. struct bonding *bond = netdev_priv(bond_dev);
  3676. struct bond_net *bn = net_generic(dev_net(bond_dev), bond_net_id);
  3677. struct alb_bond_info *bond_info = &(BOND_ALB_INFO(bond));
  3678. pr_debug("Begin bond_init for %s\n", bond_dev->name);
  3679. /*
  3680. * Initialize locks that may be required during
  3681. * en/deslave operations. All of the bond_open work
  3682. * (of which this is part) should really be moved to
  3683. * a phase prior to dev_open
  3684. */
  3685. spin_lock_init(&(bond_info->tx_hashtbl_lock));
  3686. spin_lock_init(&(bond_info->rx_hashtbl_lock));
  3687. bond->wq = create_singlethread_workqueue(bond_dev->name);
  3688. if (!bond->wq)
  3689. return -ENOMEM;
  3690. bond_set_lockdep_class(bond_dev);
  3691. list_add_tail(&bond->bond_list, &bn->dev_list);
  3692. bond_prepare_sysfs_group(bond);
  3693. bond_debug_register(bond);
  3694. /* Ensure valid dev_addr */
  3695. if (is_zero_ether_addr(bond_dev->dev_addr) &&
  3696. bond_dev->addr_assign_type == NET_ADDR_PERM)
  3697. eth_hw_addr_random(bond_dev);
  3698. return 0;
  3699. }
  3700. unsigned int bond_get_num_tx_queues(void)
  3701. {
  3702. return tx_queues;
  3703. }
  3704. /* Create a new bond based on the specified name and bonding parameters.
  3705. * If name is NULL, obtain a suitable "bond%d" name for us.
  3706. * Caller must NOT hold rtnl_lock; we need to release it here before we
  3707. * set up our sysfs entries.
  3708. */
  3709. int bond_create(struct net *net, const char *name)
  3710. {
  3711. struct net_device *bond_dev;
  3712. int res;
  3713. rtnl_lock();
  3714. bond_dev = alloc_netdev_mq(sizeof(struct bonding),
  3715. name ? name : "bond%d",
  3716. bond_setup, tx_queues);
  3717. if (!bond_dev) {
  3718. pr_err("%s: eek! can't alloc netdev!\n", name);
  3719. rtnl_unlock();
  3720. return -ENOMEM;
  3721. }
  3722. dev_net_set(bond_dev, net);
  3723. bond_dev->rtnl_link_ops = &bond_link_ops;
  3724. res = register_netdevice(bond_dev);
  3725. netif_carrier_off(bond_dev);
  3726. rtnl_unlock();
  3727. if (res < 0)
  3728. bond_destructor(bond_dev);
  3729. return res;
  3730. }
  3731. static int __net_init bond_net_init(struct net *net)
  3732. {
  3733. struct bond_net *bn = net_generic(net, bond_net_id);
  3734. bn->net = net;
  3735. INIT_LIST_HEAD(&bn->dev_list);
  3736. bond_create_proc_dir(bn);
  3737. bond_create_sysfs(bn);
  3738. return 0;
  3739. }
  3740. static void __net_exit bond_net_exit(struct net *net)
  3741. {
  3742. struct bond_net *bn = net_generic(net, bond_net_id);
  3743. struct bonding *bond, *tmp_bond;
  3744. LIST_HEAD(list);
  3745. bond_destroy_sysfs(bn);
  3746. bond_destroy_proc_dir(bn);
  3747. /* Kill off any bonds created after unregistering bond rtnl ops */
  3748. rtnl_lock();
  3749. list_for_each_entry_safe(bond, tmp_bond, &bn->dev_list, bond_list)
  3750. unregister_netdevice_queue(bond->dev, &list);
  3751. unregister_netdevice_many(&list);
  3752. rtnl_unlock();
  3753. }
  3754. static struct pernet_operations bond_net_ops = {
  3755. .init = bond_net_init,
  3756. .exit = bond_net_exit,
  3757. .id = &bond_net_id,
  3758. .size = sizeof(struct bond_net),
  3759. };
  3760. static int __init bonding_init(void)
  3761. {
  3762. int i;
  3763. int res;
  3764. pr_info("%s", bond_version);
  3765. res = bond_check_params(&bonding_defaults);
  3766. if (res)
  3767. goto out;
  3768. res = register_pernet_subsys(&bond_net_ops);
  3769. if (res)
  3770. goto out;
  3771. res = bond_netlink_init();
  3772. if (res)
  3773. goto err_link;
  3774. bond_create_debugfs();
  3775. for (i = 0; i < max_bonds; i++) {
  3776. res = bond_create(&init_net, NULL);
  3777. if (res)
  3778. goto err;
  3779. }
  3780. register_netdevice_notifier(&bond_netdev_notifier);
  3781. out:
  3782. return res;
  3783. err:
  3784. bond_netlink_fini();
  3785. err_link:
  3786. unregister_pernet_subsys(&bond_net_ops);
  3787. goto out;
  3788. }
  3789. static void __exit bonding_exit(void)
  3790. {
  3791. unregister_netdevice_notifier(&bond_netdev_notifier);
  3792. bond_destroy_debugfs();
  3793. bond_netlink_fini();
  3794. unregister_pernet_subsys(&bond_net_ops);
  3795. #ifdef CONFIG_NET_POLL_CONTROLLER
  3796. /*
  3797. * Make sure we don't have an imbalance on our netpoll blocking
  3798. */
  3799. WARN_ON(atomic_read(&netpoll_block_tx));
  3800. #endif
  3801. }
  3802. module_init(bonding_init);
  3803. module_exit(bonding_exit);
  3804. MODULE_LICENSE("GPL");
  3805. MODULE_VERSION(DRV_VERSION);
  3806. MODULE_DESCRIPTION(DRV_DESCRIPTION ", v" DRV_VERSION);
  3807. MODULE_AUTHOR("Thomas Davis, tadavis@lbl.gov and many others");