OReilly Building Embedded Linux Systems Apr 2003 ISBN 059600222X

C.3LegalClarificationsAbouttheKernelby
LinusTorvalds
ThisisafairlylongexplanationbyLinusTorvaldsregardingthekernel's
licensingandhowthislicensingappliestoforeigncode:

Feelfreetopost/addthis.Iwroteitsometimeagoforacorp
lawyerwhowonderedwhatthe"GPLexception"was.Namesandcom
removednotbecauseIthinktheyareashamed,butbecauseIdon
peopletoreadtoomuchintothem.
Linus
----Date:Fri,19Oct200113:16:45-0700(PDT)
From:LinusTorvalds<>
To:XxxxXxxxxx<>
Subject:Re:GPL,RichardStallman,andtheLinuxkernel

[Thisisnot,ofcourse,alegaldocument,butifyouwantto
toanybodyelse,feelfreetodoso.Andifyouwanttoargue
pointswithmeorpointsomehtingout,I'malwaysinterested.
point;-]

OnFri,19Oct2001,XxxxXxxxxxwrote:
>
>I'vebeenexchanginge-mailwithRichardStallmanforacoupl
>weeksaboutthefinerpointsoftheGPL.
Ifeelyourpain.

>I'vehavespenttimepouringthroughmailinglistarchives,u
>andwebsearchenginestofindoutwhat'salreadybeencovere
>yourstatementofallowingdynamicallyloadedkernelmodules

>proprietarycodetoco-existwiththeLinuxkernel.SofarI
>beenunabletofindanythingbeyondvaguestatementsattribut
>you.Iftheseissuesareaddressedsomewherealready,please
>me.

Well,itreallyboilsdowntotheequivalentof"_all_derived
havetobeGPL'd".Anexternalmoduledoesn'treallychangethe
thatrespect.

Thereare(mainlyhistorical)examplesofUNIXdevicedriversa
UNIXfilesystemsthatwerepre-existingpiecesofwork,andwhi
fairlywell-definedandclearinterfacesandthatIpersonally
reallyconsideranykindof"derivedwork"atall,andthatwer
acceptable.TheclearestexampleofthisisprobablytheAFS(t
Filesystem),buttherehavebeenvariousdevicedriversported
too.

>Issue#1
>========
>CurrentlytheGPLversion2licenseistheonlylicensecover
>Linuxkernel.Icannotfindanyalternativelicenseexplaini
>loadablekernelmoduleexceptionwhichmakesyourpositiondi
>tolegallyanalyze.
>
>Thereisanoteatthetopofwww.kernel.org/pub/linux/kernel
>butthatstates"userprograms"whichwouldclearlynotapply
>kernelmodules.
>
>Couldyouclarifyinwritingwhattheexceptionpreciselysta

Well,therereallyisnoexception.However,copyrightlawobvi
hingesonthedefinitionof"derivedwork",andassuchanythin
alwaysbearguedonthatpoint.

Ipersonallyconsideranythinga"derivedwork"thatneedsspec
inthekerneltofunctionwithLinux(ieitis_not_acceptable
smallpieceofGPL-codeasahookforthelargerpiece),astha


impliesthatthebiggermoduleneeds"help"fromthemainkerne

Similarly,Iconsideranythingthathasintimateknowledgeabou
internalstobeaderivedwork.

Whatisleftinthegrayareatendstobeclearlyseparatemodu
thathadalifeoutsideLinuxfromthebeginning,andthatdos
self-containtedthatdoesn'treallyhaveanyimpactontherest
kernel.Adevicedriverthatwasoriginallywrittenforsomethi
andthatdoesn'tneedanybutthestandardUNIXread/writekind
interfaces,forexample.

>Issue#2
>========
>I'vefoundstatementsattributedtoyouthatyouthinkonly1
>thecodeinthecurrentkernelwaswrittenbyyou.Bynotbe
>solecopyrightholderoftheLinuxkernel,astatedexception
>theGPLseemsinvalidunlessallkernelcopyrightholdersagr
>thisexception.HowdoestheexceptioncoverGPL'dkernelco
>writtenbyyou?Haseveryonecontributingtothekernelforf
>theircopyrighttoyouoragreedwiththeexception?

Well,seeaboveaboutthelackofexception,andaboutthefund
grayareain_any_copyrightissue.The"derivedwork"issueis
agrayarea,andIknowlawyersdon'tlikethem.Crazypeople(
judges)have,asweknow,claimedthatevenobviousspoofsofa
containnothingoftheoriginalworkitself,canberuledtobe

Idon'tholdviewsthatextreme,butatthesametimeIdocons
modulewrittenforLinuxandusingkernelinfrastructurestoge
done,evenifnotactuallycopyinganyexistingLinuxcode,to
derivedworkbydefault.You'dhavetohaveastrongcaseto_n
consideryourcodeaderivedwork..

>Issue#3
>========
>Thisissueisrelatedtoissue#1.Exactlywhatiscoveredb


>exception?Forexample,allcodeshippedwiththeLinuxkern
>archiveandtypicallyinstalledunder/usr/src/linux,allcod
>/usr/src/linuxexcept/usr/src/linux/drivers,orjustthecod
>the/usr/src/linux/kerneldirectory?
Seeabove,andIthinkyou'llseemypoint.

The"userprogram"exceptionisnotanexceptionatall,forex
justamoreclearlystatedlimitationonthe"derivedwork"iss
usestandardUNIXsystemcalls(withacceptedLinuxextensions)
programobviouslydoesn't"derive"fromthekernelitself.

Wheneveryoulinkintothekernel,eitherdirectlyorthrougha

thecaseisjusta_lot_moremuddy.Butasstated,bydefault
obviouslyderived-theveryfactthatyou_need_todosomethi
fundamentalaslinkingagainstthekernelverymucharguesthat
moduleisnotastand-alonething,regardlessofwherethemodu
codeitselfhascomefrom.

>Issue#4
>========
>ThislastissueisnotsomuchaissuefortheLinuxkernel
>exception,butarequestforcomment.
>
>RichardandIbothagreethata"plug-in"anda"dynamically
>loadedkernelmodule"areeffectivelythesameundertheGPL.
Agreed.

TheLinuxkernelmoduleshad(alongtimeago),amorelimited
andnotverymanyfunctionswereactuallyexported.Sofiveor
ago,wecouldbelievablyclaimthat"ifyouonlyusetheseNin
thatareexportedfromthestandardkernel,you'vekindofimpl
proventhatyoudonotneedthekernelinfrastructure".

Thatwasneverreallydocumentedeither(moreofaguidelinefo
otherswhenwelookedatthe"derivedwork"issue),andasmodu


more-and-moreusednotforexternalstuff,butjustfordynamic
standardlinuxmodulesthatweredistributedaspartoftheker
the"limitedinterfaces"argumentisnolongeraverygoodguid
"derivedwork".

Sothesedays,weexportmanyinternalinterfaces,notbecause
thinkthattheywould"taint"thelinker,butsimplybecauseit
tododynamicrun-timeloadingofmodulesevenwithstandardke
modulesthat_are_supposedtoknowalotaboutkernelinternal
obviously"derivedworks"..

>Howeverwedisagreethataplug-inforaGPL'dprogramfalls
>undertheGPLasassertedintheGPLFAQfoundintheanswer:
> />
Ithinkyoureallyjustdisagreeonwhatisderived,andwhati
Richardisveryextreme:_anything_thatlinksisderived,rega
whattheargumentsagainstitare.I'mlessextreme,andIbet
lessso(atleastyoumightliketoargueso).

>Myassertionisthatplug-insarewrittentoaninterface,no
>program.SinceinterfacesarenotGPL'd,aplug-incannotbe
>untiltheplug-inandprogramareplacedtogetherandrun.T
>donebytheenduser,nottheplug-increator.
Iagree,butalsodisrespectfullydisagree;)

It'sanissueofwhata"plug-in"is-isitawayfortheprog
internallyloadmoremodulesasitneedsthem,orisit_meant_
public,publishedinterface.

Forexample,the"systemcall"interfacecouldbeconsidereda
interface",andrunningausermodeprogramunderLinuxcoulde
construedasrunninga"plung-in"fortheLinuxkernel.No?

Andthere,Iobviouslyabsolutelyagreewithyou100%:theinte
published,andit's_meant_forexternalandindependentusers.

interfacethatwegotogreatlengthstopreserveaswellaswe
it'saninterfacethatisdesignedtobeindependentofkernel

Butmaybesomebodywrotehisprogramwiththeintentiontodyna
load"actors"astheywereneeded,asawaytomaintainagood
andtotrytokeeptheproblemspaceswell-defined.Inthatcas
"plug-in"maytechnicallyfollowallthesamerulesasthesyst
interface,eventhoughtheauthordoesn'tintenditthatway.

SoIthinkit'stoalargedegreeamatterofintent,butitco
arguablyalsobeconsideredamatterofstabilityanddocumenta
"requirerecompilationoftheplug-inbetweenversionchanges"
toimplythatit'saninternalinterface,while"documentedbin
compatibilityacrossmanyreleases"impliesamorestableexter
interface,andlessofaderivedwork)
Doesthatmakesensetoyou?

>IaskedRichardtocommentonseveralscenariosinvolvingplu
>explainwhetherornottheywereinviolationoftheGPL.So
>asonlyaddressedoneandhaseffectivelyadmittedahole.T
>theoneIaskedthathe'srespondedto:
>[A]non-GPL'dplug-inwriterwritesaplug-inforanon-G
>program.AnotherauthorwritesaGPL'dprogrammakingth
>firstauthor'splug-inscompatiblewithhisprogram.Are
>theplug-inauthor'splug-insnowretroactivelyrequired
>GPL'd?
>
>Hisresponse:

>No,becausetheplug-inwasnotwrittentoextendthispr
>
>IfinditsuspiciousthatwhetherornottheGPLwouldapply
>plug-independsonthemindsetoftheauthor.

Theabovemakesnosenseifyouthinkofitasa"plugin"issu
makessenseifyouthinkofitasa"derivedwork"issue,along
taking"intent"intoaccount.


Iknowlawyerstendtonotlikethenotionof"intent",because
inanotherwholerangeofgrayareas,butit'sobviouslyalega

Ok,enoughblatheringfromme.I'djustliketofinishoffwith
comments,justtoclarifymypersonalstand:

-I'mobviouslynottheonlycopyrightholderofLinux,andI
purposeforseveralreasons.OnereasonisjustbecauseIha
paperworkandothercr*pthatgoesalongwithcopyrightassi

AnotheristhatIdon'tmuchlikecopyrightassignmentsata
authoristheauthor,andhemaybeboundbymyrequirement
butthatdoesn'tmeanthatheshouldgivehiscopyrighttom

Athirdreason,andthemostrelevantreasonhere,isthatI
peopleto_know_thatIcannotcontrolthesources.Icanwr
notetosaythat"foruseXXX,IdonotconsidermoduleYYY
derivedworkofmykernel",butthatwouldnotreallymatter
AnyotherLinuxcopyrightholdermightstillsueyou.

Thisthirdreasoniswhatmakespeoplewhootherwisemightn
realizethatIcannotscrewpeopleover.Iamboundbythes
agreementthatIrequireofeverybodyelse,andtheonlyspe
Ireallyhaveisatotallynon-legalissue:peopletrustme.

(Yes,IrealizethatIprobablywouldenduphavingmoreleg
thanmost,evenapartfromthefactthatIstillamthelarg
copyrightholder,ifonlybecauseofappearances)

-Idon'treallycareaboutcopyrightlawitself.WhatIcare
ownmorals.WhetherI'deversuesomebodyornot(andquite
it'sthelastthingIeverwanttodo-ifIneverendupta
lawyersinaprofessionalcontext,I'llbeperfectlyhappy.
disrespectintended)willbeentirelyuptowhetherIconsid
peopledotome"moral"ornot.Whichiswhyintentmatters
lot-boththeintentoftheperson/corporationdoignthein


_and_theintentofmeandothersinissueslikethemodule
interface.

Anotherwayofputtingthis:Idon'tcareabout"legallooph
word-wrangling.

-Finally:Idon'ttrusttheFSF.IliketheGPLalot-altho
necessarilyasalegalpieceofpaper,butmoreasanintent
explainswhy,ifyou'velookedattheLinuxCOPYINGfile,yo
noticedtheexplicitcommentabout"only_this_particularv
theGPLcoversthekernelbydefault".

That'sbecauseIagreewiththeGPLas-is,butIdonotagre
FSFonmanyothermatters.Idon'tlikesoftwarepatentsmuc
example,butIdonotwantthecodeIwritetobeusedasa
againstcompaniesthathavethem.TheFSFhaslongbeendisc
isdraftingthe"nextgeneration"GPL,andtheygenerallysu
peopleusingtheGPLshouldsay"v2oratyourchoiceanyla
version".

Linuxdoesn'tdothat.TheLinuxkernelisv2ONLY,apartfr
fileswheretheauthorputintheFSFextension(andseeabo
copyrightassignmentswhyIwouldneverremovesuchanexten

The"v2only"issuemightchangesomeday,butonlyafteralld
copyrightholdersagreeonit,andonlyafterwe'veseenwhatt
suggests.FromwhatI'veseensofarfromtheFSFdrafts,we're
tochangeourv2-onlystance,buttheremightofcoursebelega
whywe'dhavetodosomethinglikeit(iesomebodychallenging
incourt,andpartofittobefoundunenforceableorsimilarw
obviouslymeanthatwe'dhavetoreconsiderthelicense).
Linus

PS.Historically,binary-onlymoduleshavenotworkedwellunde
quiteregardlessofanycopyrightissues.Thekerneljustdevel
quicklyforbinarymodulestoworkwell,andnobodyreallysupp


CompanieslikeRedHatetctendtorefusetohaveanythingtodo
binarymodules,becauseifsomethinggoeswrongthereisnothin
doaboutit.SoIjustwantedtoletyouknowthatthe_legal_
justthebeginning.Eventhoughyouprobablydon'tpersonallyc