usersdnsfail

HydrasandIPFS:ADecentralisedPlaygroundforMalwareConstantinosPatsakisandFranCasinoUniversityofPiraeus,GreeceAbstractModernmalwarecantakevariousforms,andhasreachedaveryhighlevelofsophisticationintermsofitspenetration,persistence,communi-cationandhidingcapabilities.
Theuseofcryptography,andofcovertcommunicationchannelsoverpublicandwidelyusedprotocolsandser-vices,isbecominganorm.
Inthiswork,westartbyintroducingResourceIdentierGenerationAlgorithms.
Theseareanextensionofawell-knownmechanismcalledDomainGenerationAlgorithms(DGA),whicharefre-quentlyemployedbycybercriminalsforbotmanagementandcommuni-cation.
Ourextensionallows,beyondDNS,theuseofotherprotocols.
Moreconcretely,weshowcasetheexploitationoftheInterPlanetarylesystem(IPFS).
Thisisasolutionforthe"permanentweb",whichenjoysasteadilygrowingcommunityinterestandadoption.
TheIPFSis,inad-dition,oneofthemostprominentsolutionsforblockchainstorage.
WegobeyondthestraightforwardcaseofusingtheIPFSforhostingmali-ciouscontent,andexplorewaysinwhichabotmastercouldemployit,tomanageherbots,validatingourndingsexperimentally.
Finally,wediscusstheadvantagesofourapproachformalwareauthors,itsecacyandhighlightitsextensibilityforotherdistributedstorageservices.
Indexterms:Malware,Botnets,DomainGenerationAlgorithm,DGA,IPFS1IntroductionThecontinuousarmsrace[23]betweenmalwareauthorsandsecuritypracti-tionershasledtotheexistenceofawiderangeoftoolsandtechniqueswhichhaveturnedcybercrimeintoaprotable"business".
Thisbusinesshasvariousmonetisationsources,rangingfromspamming[33]toadinjection[9]butalsoincluding,notably,denialofservice,ransomware-basedextorsionandphishing.
Monetisationschemeswereinthepastchallengingandfrequentlyproblematicfortheattacker,astraditionallytheadversaryneedsto(atleastpartly)dis-closeheridentitytoreceivethemoney.
Yet,therecentavailabilityofprivate1arXiv:1905.
11880v2[cs.
CR]23Jan2020cryptocurrencies(suchasMonero,Zcash,etc.
)hassignicantlydecreasedthisexposure.
However,asecondriskofdeanonymizationstillexistsforthecybercriminal,assheneedstomaintaincontrolovertheinfecteddevicesorbots.
Itisclearthataclassicalschemewheretheadversary(botmaster)issuescommandsanddistributestheminacentralisedfashionhasinherentsingle-point-offailureis-suesthatimpedeitssuccess.
Forinstance,ifthereisauniquedomainorIPthatisusedtocontroltheinfecteddevices,onceitisidentieditwillbe(quickly)blocked.
Therefore,botmasterstrytoeitherhidetheircommandswithinle-gitimatenetworktracthatcannotbeblocked,e.
g.
socialnetworksorwidelyusedservices,ortodecentralisebotnetmanagement.
Itshouldbehighlightedthatatypicalbotnetconsistsofseveralthousandsofdeviceswhich,ontopoftraditionalworkstationsmayconsistofrouters,printers,smartphonesandothersmartdevices.
Atypicalstrategythatcybercriminalsusetocounterbothdeanonymiza-tionandblacklistingissuesistheuseofmultipledomainstomanageinfectedhosts.
Inthiscontext,botsrunDomainGenerationAlgorithms(DGAs)whichgeneratemultipledomainsatapredenedtime(generallyeachday)whichwillallbequeriedbythebotstoretrievecommands.
Inthisregard,theattackermaygeneratethousandsofbogusdomainsbutlimithercontroltoonlyahand-fulofthem.
However,thismeansthatthecontrolisperiodically(e.
g.
daily)transferredfromonedomaintoanotherwithoutanyspecicpatternforthedefendant'sperspectiveandwitheachgenerateddomainhavinganequalprob-abilityofbeingusedbytheattacker.
Theproblemisampliedbythelackofproperreportingfromdomainregistrars,whooftenfailtotimelyrespondtorequestsfromlawenforcementauthoritiestryingtotakedownbotnetsoratleasttracetheidentityofmalwareauthors.
Inspeciccases,registrarshavebeenevenfoundtomisbehave,e.
g.
byacceptingbribes[20].
Tofurtherhidetheiridentityandavoidblacklisting,botmastersmayalsoloopthroughasetofinfectedhoststohostthecontentthattheDGAspointto.
Therefore,theprob-lemresemblesthemythicalLernaeanHydrawhichspawnedtwoheadswhenoneofherheadswaschopped.
ThetypicalmanagementmodelusedbybotmasterswhenusingDGAsisillustratedinFigure1.
1.
1MaincontributionsPonderingonthenewdirectionsthatmalwarecanhave,thegoalofthisworkistoinvestigatetheuseoftheInterPlanetaryFileSystem(IPFS)1forthecoor-dinationofabotnet.
WhilewearguethattheuseofIPFSforthedistributionofmaliciouscontentisrelativelyclear,wediscussotherissuesthatemergefromthisperspective.
Moreprecisely,wearguethatthefactthatIPFScanprovideanonymity,persistenceofthecontent,fastdelivery,andarobustnetworkwherecontentcannotbeeasilyblocked,providesanideallandscapeformalwareau-thors.
ItshouldbenotedthatIPFShasalreadystartedbeingusedmaliciously.
1https://ipfs.
io2Figure1:ThetypicalDGAmanagementmodel.
Quiterecently,IPFSwasusedduringaphishingcampaignforhostingphishingforms2.
Inthisregard,weinvestigateIPFS-basedbotnets,theirapplicabilityandissuesthatemergefromtheirpossibleuse.
Tothebestofourknowledge,thisistherstworkthatprovidesaholis-ticarchitecturetodiscusstheexploitationofIPFSfrombotnets,beyondthetrivialscenarioofjuststoringmaliciousorillegalcontent.
Moreconcretely,weinvestigateanIPFSaddressgenerationalgorithmwhichusesaskewedPRNGtogeneratehashesthatwilleventuallypointtolesthatcontainanewsetofinstructionstobeexecutedbytheinfecteddevices.
Apparently,intheproposedsettingwedonothavenewdomainnamesbutnewIPFSaddresses,however,duetothewayIPFSworksitisthenameresolution.
TheuseofIPFSimpliesaseriesofinherentpropertiesandbenetsfortheadversarycomparedtosimilarworkswhichrelyontheuseofsocialnetworksorblockchain[26,24,31,1,29].
Ourproposedmethodologyisfurtherdistributed,morerobustandprovidesmorebandwidth.
ContrarytotraditionalDGAs,theIPFSavourthatwedis-cussinthisworkismorestealthsinceithidesthetracesofthebotmasterandprovidesabetterdistributionnetworkwithnoapparenttake-downmechanism.
Basedontheabove,weconsiderthatDGAsandthemethodologyusedforIPFScanbeconsidereddierentinstantiationsofabroaderconcept.
Therefore,weintroducetheconceptofaResourceIdentierGenerationAlgorithm(RIGA).
Wearguethatthegrowingtrendtowardsdecentralisedsystems,theintroductionofnewprotocolswithdierentcontentresolutionmethodswilleventuallybeexploitedfrommalwaretodisseminatethecontent.
Sinceanalgorithmcould2https://www.
bleepingcomputer.
com/news/security/phishing-attacks-distributed-through-cloudflares-ipfs-gateway/3generatethenamesthatcanbeusedfromdierentprotocolsthenotionofRIGAencapsulatesthecoreconceptofDGAs:generatemoreoutputthantheactualresourcecannotbedeterminedfromanoutsider.
Therefore,aRIGAisanaturalextensionofDGAsthatconsidersotherprotocols.
1.
2EthicalconsiderationsThegoalofthisworkisnottoinspiremalwareauthorstowritemoreecientmalwarebuttomotivateresearcherstondsolutionsforanemergingthreat.
Unarguably,duetoitsnatureIPFScanbeconsideredidealforspreadingmal-wareandillegal/maliciouscontentsinceamongothersthereislittleregulation,alotofanonymity,andthecontentcanbeconsideredpermanent.
ThefactthatmanybigorganisationslikeCloudarearebackingtheIPFSinitiativemeansthatIPFSisheretostay.
Tothisend,thereisaneedtostudyhowanadver-sarycouldexploitthisnewtechnologyandpreparepossiblecountermeasuresandinitiateadialoguewiththeIPFScommunitytoseehowcansimilarissuesberesolvedinthenearfuture.
Theresearchinthiseldmust,therefore,inves-tigateaspectsbeyondthetrivialones,,e.
g.
hostingmaliciouscontentorforms,whichasalreadydiscussedhavebeenalreadyobserved.
Tothisend,theproofofconceptcodeonlyprovidesthesimpleimplementationofaDGAthatisnothardenedtoprovidefurthersecuritytotheadversary.
1.
3OrganisationofthisworkTherestofthisworkisstructuredasfollows.
InthenextsectionweprovideanoverviewoftherelatedworkfocusingmainlyonDGAsandIPFS.
InSection3,weextendthenotionofDGAsandintroducetheconceptofRIGAs.
Then,inSection4wedetailtheproposedschemefordecentralisedmanagementofbotsthroughIPFS.
Section5illustratestheecacyofourproposalthroughvariousexperiments.
InSection6,wecompareourworktothecurrentstateoftheart.
Finally,weconcludethisworkdiscussingfuturedirectionsandopenissues.
2RelatedworkNowadays,mostmalware-basedcampaignsrelyontheuseofbotnets,com-mandedbyremoteservers(i.
e.
C&Cservers),whichsendinstructions/orderstoinfecteddevices[3].
Inthepast,themechanismusedtondsuchC&CserverswastohardcodeIPaddressesinthemalware.
However,suchtechniqueentailedasetofdrawbacksfortheattackers[25](e.
g.
easytotake-down).
Therefore,botnetsevolvedintopeer-to-peer(P2P)botnets[13],whichadoptedamyriadoftechniquessuchastheFastFluxapproach,whichimitatescontentdistributionnetworksbyresolvingadomainnametomultipleIPaddresses[14,18].
Ontopofthepreviouslydiscussedapproaches,malwareuseDGAs,whichimplementadeterministicpseudo-randomgenerator(PRNG)tocreateasetofdomainnames[35,28].
Hence,infecteddevicescheckthelistofgenerated4domainsandperformqueriesuntiltheyndtheC&Cserver,whoselocationmayalsochangedynamically.
Inthisscenario,blacklistingdomainsisrendereduselessasitimpliesmanypracticalissues.
2.
1DomainGenerationAlgorithmsAccordingtoPlohmannetal.
[30],DGAsarecategorisedas:Arithmetic-basedDGAsHash-basedDGAsWordlist-basedDGAsPermutation-basedDGAsIngeneral,arithmetic-basedDGAsuseaPRNGtocombineasetofchar-acters(typicallyASCII)andcreateadomainname.
Inthecaseofhash-basedDGAs,attackersbasicallycreatedomainsusingthehexrepresentationofahash.
TodetectbothDGA-basedfamilies,methodsreportedintheliteratureusefea-turessuchasentropy,lengthorlexicalcharacteristicstodeterminewhetheraDGAhasgeneratedadomainnameornot[5,40],aswellascharacteristicssuchastracinformation(e.
g.
NXDomainqueriesorWHOISinformation)[44,17,4,41,22,12,42].
Wordlist-basedDGAsappearedtoovercomethedrawbacksoftheprevioustwoDGAfamilies.
Inthiscase,attackersadopttheuseofEnglishwordliststogenerate"human-readable"domains,hinderingpreviousDGAdetectionap-proaches.
Inthiscontext,severalneuralnetwork-basedtechniquesexhibitgooddetectionaccuracy[38,2],aswellasothernovelmethodsbasedonmetricssuchasn-gramsandwordfeatureextraction[10,34].
Inthecaseofpermutation-basedDGAs,attackerscanusecombinationsbasedonaninitialdomaintobypassdetectors[6].
Thereexistsimilartechniquessuchasdomainshadowing,whichreliesontheuseofvaliddomainsthatwerepreviouslyhacked[21].
RegardlessoftheDGAused,thetypicalbotnetsuseInternetprotocolsforcommunication.
Nevertheless,moresophisticatedanoriginalapproachescanbefoundinthecaseofsocialnetworks[26,31]andblockchain[1,29,24],whichentailfurtherdetectionchallenges,sincealltracseemslegitimateandiscoveredunderstandardchannelssuchasHTTPS.
Inthecaseofblockchainnetworks,theproblemisexacerbatedduetoitsinherentpropertiessuchasimmutabilityandanonymity.
Foradetailedoverviewandclassicationofmethodsofhowmaliciousdo-mainscanbedetected,theinterestedreadermayreferto[43].
2.
2IPFSTheInterPlanetaryFileSystem(IPFS)[7]isadistributedP2Psystemforre-trievingandsharingIPFSobjects.
IPFSusesaMerkleDirectedAcyclicGraph5(DAG),whichisacryptographicallyauthenticateddatastructure,toaddresssuchobjects.
Therefore,insteadofidentifyingobjectsbytheirlocation(e.
g.
HTTPS),thesystemaddressesthembytheirrepresentationofthecontentit-self,whichisusuallytheirBase58SHA-256encodedhash3.
ThecontentsofanIPFSobjectaremainlystoredintwoelds:(i)thedataeld,whichisanunstructuredbinarydatablockofsizeof256KB(i.
e.
largerlesarestoredbyalistoflinkstolechunksthatareipfsobjectgetQmYWAifyw2V5.
.
.
|jq2{3"Links":[.
.
.
somelinks.
.
.
],4"Data":"binarydatablocks"5}Listing1:OverviewofanIPFSobjectstructure.
SimilardecentralisedsystemssuchasBitTorrentcansuccessfullycoordinatethetransferofdatabetweenmillionsofnodes,butitappliesonlytothetorrentecosystem.
Incontrast,IPFSimplementsageneralisedversionofthisproto-colcalledBitSwap,whichenablesfurtherpossibilitiessuchasbuilt-instoragemarketplaceslikeFilecoin5.
Filecoinisadistributedelectroniccurrencythatusesproof-of-retrievability,whichisavericationmechanismusedtoprovethatanodestoresaparticularle.
Inthiscontext,currencyisawardedforstoringles,whichisapracticethatisgainingmoreadepts(e.
g.
Ethereum'sSwarmandMistorMaidSafeandtheSAFEnetwork)6.
TheuseofaMerkleDAGstructureallowsthecreationofaversioncon-trolsystem(VCS).
Moreconcretely,IPFSstorestheobjecthistorysothatallversionsareaccessiblethroughouttime.
Thispermitscongurablesynchronisa-tionoflessincealluserscaneditthemlocallyandlaterpushthenewlestoIPFS.
VCSisenhancedbytheInterPlanetaryNamingSystem(IPNS),whichenablescontentlinking,sothatlescanbeaccessedusingthenodeIDaddress,allowinguserstoretrieveupdatedcontentswithoutknowingthenewhashesofsuchles.
AnotherrelevantcharacteristicofIPFSisthatitisaself-certifyinglesystem(SFS),whichmeansthatdataservedtoclientsisauthenticatedbytheirownlenameandthenodeprovidingit.
Therefore,nodesusetheirprivatekeysto"sign"dataobjectstheypublish,andtheauthenticityofthisdatacan3https://en.
bitcoin.
it/wiki/Base58Check_encoding#Base58_symbol_chart4https://multiformats.
io/multihash/5https://filecoin.
io/filecoin.
pdf6https://www.
ibtimes.
co.
uk/juan-benet-ipfs-talks-about-filecoin-15861226beveriedusingtheirpublic-key.
Inadditiontothesefeatures,IPFSenablesaseriesofpropertiesdescribedbrieyinTable1.
Table1:IPFSmainproperties.
PropertyDescriptionImmutability/TamperproofContent-basedaddressingguaranteesthateachleresolvestoaspecichashEqualityAllpeershavesimilar,ifnotequal,permissionsandpossibilities.
DecentralisationThenetworkistotallydistributedwithnocentralentitiesFaulttoleranceAhighnumberofindividualpeersguaranteestherobustnessofthenetworkAvailabilityTheavailabilityofthenetworkdependsonmultiplepeersandnotonasingleentityResilienttoattacksThelargenumberofpeersguaranteethepersistenceofthenet-workUnlimitedRe-sourcesAhighnumberofsimultaneoususerssharingtheirassets.
ScalabilityRequestsaremadetotheclosestpeerinsteadtoasinglecentrallocation,avoidingbottlenecksMarketplacemon-etisationWithsystemslikeFilecoin,whichincentiviseIPFSNowadays,theapplicationsandpossibilitiesofIPFSarebeingexploitedinamyriadofcontexts[15,19,7].
OfparticularrelevanceisthesymbioticrelationshipbetweenIPFSandblockchain,adistributedandlinkedimmutableledger[27].
Therefore,thecombinationofbothtechnologiesenhancestheirapplicationscenarios[36,8]byenablingo-chainstorageandanonymouslesharing,whichareusuallymanagedthroughsmartcontracts[37].
3ResourceIdentierGenerationAlgorithmsThecoreconceptofaDGAistocreateapseudo-randomsequenceofdomainsthatthebotmastermayusetohosthismaliciouscontent.
Duetothevastamountofdomainsthatcanbegenerated,thedefendantcannotblacklistallthedomains,norcanshedeterminewhichonewillbeusedbythebotmaster.
WhileinDGAsthisconceptisimplementedthroughtheDNSprotocol,onecouldextendthistoaccommodateforotherprotocolandapproaches.
Tothisend,weextendthenotionofaDGAtothatofaResourceIdentierGenerationAlgorithm(RIGA).
Inprinciple,aRIGAgeneratesasequenceofpossibleaddressesthatcanbeusedbyanadversarytohostmaliciouscontent.
Weassumethatwehaveanaddressresolutionspacewheretheadversarycanhostherdesiredmaliciouscontent,andeachaddresshasthesameprobabilityofhostingitforanexternalobserver.
Thesequenceoftheseaddressescanbepurelyrandomorpseudo-random,dependingonthesizeoftheaddressresolutionspace.
Forinstance,iftheaddressresolutionspaceisminimal,thentheadversaryknowsthatwiththeproperqueryratetheRIGAwouldeventuallygeneratethedesiredaddresses.
Clearly,inthisapproach,theRIGAdoesnotneedanyinput.
However,inthecaseofalargeaddressresolutionspace,theadversaryneedstoknowthatthe7Figure2:AnoverviewofthepotentialscopeofResourceIdentierGenerationAlgorithms.
RIGAwillproduceasetofpredenedaddresseswhichwillbegeneratedbyallinfecteddevices.
Therefore,theRIGAtakesasinputaparameterwhichwillbeusedasaseedbyaPRNG.
RegardlessofwhethertheRIGAwilluseaPRNGortruerandomnum-bers,theRIGAhasamappingmechanismtoconvertanumberintoaUniformResourceIdentier(URI)ofthecorrespondingprotocol,whichwillbeitsout-put.
InthecaseoftheDNSprotocol,theRIGAgeneratesaseriesofURLssomeofwhichtheadversarywillallocatetohosthercontent.
InthecaseofIPFS,theRIGAcouldgenerateaseriesofIPFS,IPNSorDNSLinkaddresses,asitwillbediscussedinthenextsection.
Similarly,RIGAscouldbeusedtogenerateURIsforSWARM[39],STORJ[32],Maidsafe[16]andotherdis-tributedstorageservices.
Finally,aRIGAcouldgeneratehierarchicaloratnamesforInformation-CentricNetworks,orhierarchicalnamesforthecaseofNetworkDenedNetworks.
AnoverviewofthepotentialscopeofRIGAsanditsrelationshipwithDGAsisillustratedinFigure2.
Whiletheprotocolmayimplysomeparameters,aRIGAwouldneedsomeadditionalmodicationstogeneratethedesiredoutput.
TypicalparametersneededforaRIGAincludethelengthofthedesiredoutputandthealphabetthatwillbeused.
Forinstance,theURLencodingimpliesaspecicalphabetwhichdiersfromtheoneusedinIPFS.
Moreover,oneRIGAmayusesomedictionariestoproducethenamesorpartofthem.
Forinstance,theTLDsinallDGAsarehard-codedintheirsourcecode,whilesomeRIGAsmayusedictionariestoproducethenames,,e.
g.
theMatsnuDGA.
SomeRIGAsmayrequirespecicsyntaxrulesonthealphabettocreateameaningname,,e.
g.
Base58Checkencodingorusethelettersofthealphabetwithspecicfrequencies8ofoccurrence,,e.
g.
theSymmiDGA.
Finally,onemayrequiretheRIGAtogenerateasetofpredenedURIswithouthavingtohard-codetheminthesourcecode.
Moredetailsaboutthelatterwillbediscussedinthenextsection.
4ProposedschemeforIPFSInthenextparagraphs,wedescribethewaythatIPFScanbeexploitedbymalwarefocusingonhowDGAscanuseit.
Therefore,theinfectionmechanismtospreadthemalwarethatpenetratedthesystemsisnotdiscussedandiscon-sideredbeyondthescopeofthiswork.
Theinfectionmechanismdependsonthevulnerabilitiesthatthesedevicesmighthave,thefactthattheyarewebdiscoverableandaccessiblewithdefaultcredentialsorthattheuserhasbeentrickedtovisitmaliciousURL.
NotethatthesamemethodcouldbeextendedforotherdistributedstorageserviceslikeSTORJ,SWARMetc.
asoutlinedintheprevioussection.
Forthesakeofclarity,inwhatfollowswewillonlyfocusonIPFS.
First,weprovideabriefoverviewoftheproposedbotnetmanagementmodelwithIPFS-basedDGAs,thenwedetailthenewDGAalgorithmandprovideaskewedPRNGforitsconstruction.
Afterwards,wediscussseedingandspread-ingissuesandtheeectivenessoftakedownmechanisms.
Finally,weclosethissectionprovidingadiscussionforupstreamcommunication.
4.
1OverviewGiventhecontinuousgrowthofmalware,inwhatfollowswediscussthepos-sibilityofexploitingIPFS.
Thereasonbehindthischoiceisthatwehaveseenseveraltransformationsofthemondusoperandiofmodernmalware,seeSection2.
Therefore,thegrowingcommunityofIPFSandtheseamlessprocedureofuploadingandsharingcontentisforsureexpectedtoattractpeoplewillingtoexploittheprovidedfeaturesfornefariousmeans.
Infact,theseconsiderationshavealreadysparkedseveraldiscussionsintheIPFScommunityon,e.
g.
howtotreatillegalcontentandwhetherdeletionofcontentshouldbemadepossible7.
Whilemanyoptfortheblacklistingapproachtocounterthisissue,wearguethatthismayonlypartiallyaddresstheseissues.
Thereasonismoreorlesssimilartotheonediscussedabouttheregularweb,iftheaddressesthathavetobeblacklistedarefartoomany,thenthisapproachisrendereduselesssinceitimpliesalotofcontinuouseortfromtheparticipatingnodes.
Tothisend,wediscussthepossibilityofhavingIPFS-basedRIGAs.
TheuseofIPFStoarmourabotnetenablesasetofopportunitiesformal-wareauthorsincludingimmutability,removalofcosts,easinesstospread,andanonymity.
Ingeneral,usingIPFS,thebotscandisseminatethecommandsoftheC&Cserveronlybypinningale.
Moreover,onceinfected,theycancreatedierentversionsoftheinstructionleandspreadanew"version"of7Seehttps://github.
com/ipfs/faq/issues/9https://github.
com/ipfs/faq/issues/36https://github.
com/ipfs/faq/issues/1569themalware,whichtranslatesintoaresilientandmutablebotnet.
Bydefault,thetracinIPFSisencrypted,andHTTPScanbeusedtopreventprotocolcensoringin,e.
g.
corporateenvironments.
Evenmore,sincenobodymanagesthecontentsnorthetrac,implementingltersordetectionschemesismoredicultthaninthecaseofcentralisedservices,anditisimpossibletoknowalltheavailablelesunlesstheirhashisknown.
Clearly,noonecanknowbeforehandwhetherahashlinkwillbeusedornot.
Evenmore,havingaccesstotheoriginalsourcecodeoftheRIGAisuselessasthecontentmaybecomealmostimmediatelyavailable.
Therefore,onecannotdeterminewhetheranyofthegeneratedhasheswillbeuploaded.
Moreover,sincethehashisnotreversible,onecannotdeterminethecontentofthesharedles.
ContrarytoDNS,thebotmasterdoesnothaveanycost,andthereisnodirectlinkwithheridentity.
Theadversary,fromnowonMalory,canuploadherleswhenevershewantsandtheywillbeimmediatelydistributedwithoutanobvioustake-downmechanism.
4.
2ModusoperandiBasedontheabove,Malorypreparessomeleswiththemaliciouscontext,orlinkstothatcontextintheformofles,,e.
g.
rendezvouspoints.
Thelatterapproachallowshertosetthecontentatalaterstage.
Havingtheseles,shewilluploadthemtoIPFSwhenshedeemsappropriate.
ToallowherbotstoaccesstheselessheusesaskewedPRNG;whichwillbedetailedafterwards,thatsheembedsinthemalwareofhercampaign.
Usingapredenedseed,thePRNGwilliteratethroughrandomIPFSlinks,whichmostlikelywillnotpointtoanycontent.
However,thePRNGwillalsogeneratethepredenedlinksatthedesiredinput.
ToallowthebotsretrievethecontentthatMaloryhaspreparedwhenaccess-ingcontentbeyondIPFS,,e.
g.
therendezvouspoints,Maloryembedsfurtherinformationinthemalware.
Thisinformationincludestheinstructionsetthatbotscanexecuteandhowitcanberetrieved.
Forinstance,themaliciouscontentcanbeembeddedusingsteganographicmethodsinimagesinsocialnetworkstoavoidraisingsuspicions.
Therefore,thebotsaregiventheinstructionson(i)howtocollecttheinformationfromIPFS,(ii)howtoparseretrievedcontent,(iii)howtonavigatetoasetofrendezvouspoints,(iv)howtoextractthecon-tentfromthesepoints,(v)howtoextractthecommandsfromthecollectedcontent,and(vi)howtoprovideherfeedbackintherendezvouspoints.
Clearly,thecontentthatMalorydistributescouldbedigitallysigned,soMalorywouldhavetoembedasetofpublickeys.
ThiswouldallowherbotstoverifythatMalorysubmittedthecommandandavoidtakeover/downeorts.
Fromthebots'side,MaloryhasinstructedthemtoqueryIPFSlesperiodi-callyusingtheRIGA.
Tothisend,Maloryhasembeddedthelistofwell-knownIPFSgateways8whichthebotsuseaccordingtoaround-robinschedule.
Thelatterisusedtominimisetheriskofalertsandblacklistingintheeventofhav-8https://ipfs.
github.
io/public-gateway-checker/10ingshortintervalsbetweenthequeries.
BotsusetheRIGA,generateadomainandtrytoaccessitviaoneoftheavailabledomainsusingapredenedtimeouttoguaranteethatthesearchforcontentwillnotresultintoanendlessloopifthecontentdoesnotexist(seeFigure2).
Ifthecontentexists,thebotretrievesitandexecutesthelistedcommandsorredirectsitselftoanewlocationtoretrievethemaccordingtotheretrievalalgorithmthatMaloryhasembedded.
TheconceptbehindredirectionisthatMalorymaynothavealreadydecidedheractionssoshewilluploadthematalaterstage,butshecannotchangetheIPFShash.
Therefore,shepreparesaplaceholdertodosoatalaterstage.
TheproposedmanagementmodelisillustratedinFigure3.
1importrequests2h="Qmc8N5wtMkvMySqxu4Agy2SGv"3h=h+"L2zxYGf4rWmHvMASoUQv6"4r=requests.
get('https://ipfs.
io/ipfs/%s'%h,timeout=5)Listing2:ExamplecodetorequestcontentfromIPFSwithatimeout.
Figure3:Proposedmanagementmodel.
4.
3APRNGforanIPFSRIGAInatypicalDGA,theauthorcreatesapseudo-randomnumbergenerator(PRNG)thatgeneratesthedomainsandselectsasmallsubsetofthepossiblevaluesandregistersthecorrespondingdomains.
However,intheIPFSscenario,thisisnotpossibleasIPFSusescryptographichashfunctionsandtheresultisencodedtoBase58encoding.
Therefore,itiscomputationallyimpossibletoselectsomehashesandcreatelesthathashintothesevalues.
11Toachievethesameeect,wecreateaskewedPRNGwhichgeneratesasetofpredenedhashesthatwillbeusedaspointersforlinkstocontentwhichbeusedbybotstoresolvenewMalory'scommands.
InwhatfollowswerepresentthisPRNGasafunctionfwhichiteratesthroughvariousstates.
Forthesakeofsimplicity,weassumethatwehaveacounterx∈{0,1,2,.
.
.
,U}=Dthatthebotsprogressivelyincreasestartingfrom0.
Whileweconsider0astheinitialseed,thiscandierastheseedmaybecollecteddynamically.
Moreover,weassumethatMaloryhaspreparedasetofklesand/orlinkstocontent(alsostoredasles).
Maloryselectsahashfunctionh(x):D→[0,2L1],,e.
g.
SHA-256,andhashestheleswithitcreatinghi,i∈{1,2,.
.
.
,k}.
Then,MaloryselectsthevaluesofthecounterswhenthePRNGwillevaluatetoherdesiredvaluesthatwedenoteasvi,i∈{1,2,.
.
.
,k}.
Therefore,MaloryrequiresfromthePRNGtoevaluateasfollows:f(vi)=hi,i∈{1,2,.
.
.
,k}SincethescopeofthisworkistodiscussthepossibleuseofIPFS-basedRIGAs,wedeliberatelyomitthehardeningofthePRNG.
Inthiscontext,oneofthemostobviouswaystocomputefisusingpolynomialinterpolation.
Hence,wecomputethepolynomialintermsofLagrangepolynomialsasfollows:f(x)=ni=00≤j≤n,j=ixxjxixjhimodpwherepisaprimenumberwithp>hi,i∈{1,2,.
.
.
,k}.
Clearly,evaluatingfovervi,i∈{1,2,.
.
.
,k}willproducethedesiredvalueshi,i∈{1,2,.
.
.
,k},andtherestofthevalueswillberandomvaluesintherangeof[0,2L1].
SinceIPFSrepresentsthehashoflesandobjectsusingMultihashformatandBase58encoding,theDGAwilliteratethroughthevaluesofDandevaluatethepolynomialatapredenedrateandconvertthevaluesfromintegersintheinterval[0,2L1]tothecorrespondingBase58encodingwhichcanbedirectlyassociatedwithanIPFSlink.
Itisevidentthatthemethodologyabovecanbeusedforanyhash-basednamingschemelikeIPFS.
AtrivialimplementationofthisPRNGisavailableonGitHub9.
WhiletheRIGAaboveecientlygeneratesthedomains,itdoesnotexploitallthefeaturesofIPFS.
Moreprecisely,sinceallobjectsinIPFSarecontent-addressedeverymutationofanobjectleadstoanewaddress,IPFSprovidestwoprotocolstocreatemutableaddressesinthisecosystem,IPNSandDNSLinkwiththelatterbeingmoreecient.
Thus,byhavingtheIPNSorDNSLinkaddressofanobjectonemayconstantlyaccessthenewestversionofanobject.
ThefactthattheprotocolsaboveallowonetoaccessmutablecontentcanbeexploitedbyMaloryinanotherway.
Inthisscenario,eachbotmayparseDrandomlyandevaluateafunctionfsimilartofwhichnowgeneratesIPNS9https://github.
com/kpatsakis/RIGA12links.
Malorycanupdatehercontentwhenevershewants,andthebotswillaccessitinarandomsequence.
TheuseofIPNSenablesfurtherpossibilities,suchaslinkingmaliciousles(e.
g.
duringabriefwindowoftime)andthensubstitutingthembynewnon-maliciouscontent,sothatnotracesareleft.
Moreover,oldcontentcanbeunpinnedandhenceremovedbytheparticipatingnodescreatingfurtherissuesintheanalysisandgapsinthetimelineofthecampaign.
4.
4ContentseedingandspreadingInIPFS,nodesandlescanbeuniquelyidentied.
Inthecaseofnodes,theirIDcorrespondstothehashoftheirpublickey.
Inthecaseofles,identicationisperformedbycomputingthehashoftheircontent,asstatedinSection2.
2.
Bydefault,eachnodeisconnectedtoaswarmofnodeswhichcanberetrievedwiththecommandipfsswarm.
However,itisnotpossibletoretrievethelistofallexistingnodesinIPFSnorthelesstoredbyeachnode.
Therefore,lesareonlyaccessiblebyusersthatknowtheircorrespondinghashes.
Despitetheselimitations,thelistofnodesthatstorealecanberetrievedusingthecommandipfsdhtfindprovs.
Thus,nodescanbecomeseedersofalebypinningitusingipfspinadd.
Theanonymityofnodesandlescanbeseenasachallengetomalwarespreading.
NotethatwhenuploadingcontenttoIPFS,thenodethatuploadsitbecomesitsprimaryanduniqueseeder.
IfamaliciousnodeactsastheuniqueC&Cserver,itcanbeeasilydetectedandisolated.
Tocounterthisdeciency,asubsetofinfecteddevicesmayinstallIPFSandautomaticallypinthesharedlefromMalory,immediatelybecomingseedersofthiscontent.
Practically,thismeansthatthebotnethasarobustinfrastructurewhichcannotbetakendownduetoitsdecentralisednature.
Asalreadydiscussed,currentlythereisnomechanismtodeletealeinIPFSandnoplantosupportthisfunctionalityinthenearfuture.
ItshouldbenotedthatMalorycanimprovetheseedingofhercontentandobscuretheidentityoftheinfectednodesthatparticipateinhercampaignbymotivatingothernodestosharehercontentusingFilecoins.
SincethecontentthatMalorywouldrequesttosharewouldbeminimalintermsofsize,atthepropercost,manynodescouldbeeasilyconvincedtosharetheproposedcontent.
Asaresult,Maloryhastheguaranteesnotonlyfromthesubsetofinfectedhostsbutfromseeminglybenignhoststhathercontentwillbedisseminated.
4.
5TakedownmechanismsInthetraditionalDGAscenario,organisationshaveoftenjoinedtheireortstotakedownbotnets.
Oneofthemostwell-knownexamplesisConckerforwhichinternationalcooperationwasmadethatmanagedtoregisteredallthedomainsthattheDGAwouldgenerate,preventingthebotmasterfromcontactingthebotsandeventuallylosetheircontrol.
13Evidently,thisisnotpossibleintheIPFSscenariosinceallthelinksareactuallyhashesoflesfromcryptographichashfunctions.
Therefore,evenifthemalwarehasa"killswitch"(asinthecaseofWannacry)thatcanbeactivatedbyaremotecommandthatitreceivesthroughanIPFSlink,noonecandothisforanIPFSlink.
ThelatterwouldrequireonetogeneratealewithaspecichashthatwouldbegeneratedfromtheDGA.
Sincecryptographichashfunctionsareimmunetopre-imageattacks,thisapproachisimpossible.
Similarly,tractoIPFSmightnotbeeasilyblocked.
ThefactthatseveralgatewayscanbeusedmeansthatonemustrestrictaccesstoallIPFSfacingcontent.
Giventhemomentum,whilethisoptionsoundspractical,itisnotrationalduetotheincreasingcontentthatisshared.
ItshouldbehighlightedthatwithcontinuoususeofIPFSinconjunctionwithblockchains,blockingIPFScontentinseveralcorporateenvironmentsbecomesinapplicable.
4.
6Receivingbots'feedbackWhiletheinfecteddeviceshavetoretrievecommandsfromthebotmaster,botsmustalsoreturnsomeinformationtoher.
Thisinformationmaybetrivial,,e.
g.
botIDXhasenteredthenetwork,ormaycontainvaluableinformation,,e.
g.
creditcards,credentialsetc.
WhilethisinformationcaneasilybestoredinIPFS,itisimpossibleforthebotmastertoretrieveitasshecannotguessthehashoftheinformationtoretrieveit.
Therefore,thebotmastermayuseIPFStopublishrendezvouspointsthatthebotswillusetopushinformationbacktothebotmasterbyeitherpostingitdirectlyintherendezvouspointortheIPFSlinkwheretheyhaveuploadedit.
Inthelattercase,infectednodesmaylaterunpinsuchles,leavingnotraceofwhatinformationwasexltrated.
Notethatsuchleswillbeaccessedonlybythebotmasterwhowillreceivethehashfromtherendezvouspoint.
Obviously,asinthecaseofMalory'scommands,thisinformationcanbeencryptedorembeddedusingsteganographyinanotherobjectbeforesharing.
5ExperimentalresultsTovalidatethefeasibilityofourapproach,weimplementedtwotests.
Therstonetriestodeterminewhetherthegatewaysimplementanymeasuretoblockfrequenthostrequests.
ThescopeofthisexperimentistoseewhetherahostthatwillhaveanIPFS-basedRIGAwouldbeblockedafteroftenrequeststoagateway.
ThesecondexperimenttriestoquantifyhowmuchtimeitwouldtakeMalorytomakehercontentavailabletoherbots.
Oncewecheckedtheeciencyofthenetworkbylocallypinningasetofles(i.
e.
contentisavailableintheorderofseconds,soitenablesreal-time)ourrsttestfocusedonstudyingwhetherthegatewaysapplyanythresholdonclientrequests.
Therefore,weselected20WikipediaarticleswhicharehostedonIPFSandcollectedtheiraddresses.
Then,wecollectedthelistofavailableIPFSgateways.
Foreachgateway,wesequentiallytriedtofetcheacharticle1450timesfromthesamehost.
NotethatweaddedamaximumtimeoutforeachquerytoberesolvedsinceIPFSdoesnotimplementanecienttimeoutmechanisminthecaseofnon-existingles.
TheresultsforeachgatewayareillustratedinTable2.
Inthereportedresults,droppedreferstorequeststhattimedout.
Fromthetable,itisevidentthatthegatewaysdonotresponduni-formly.
Therefore,therearesignicantdierencesintheamountoftimeneededforeachgatewaytorespondandtherearediscrepanciesevenwiththedierenttimeoutsoftherequests.
Apartfromtheswedneck.
xyzgatewaywhichhadbyfarthelongesttimetorespondtotherequestsandwithrelativelyhightimeouts(3.
1%and5.
4%)alltheothergatewaysalmostnevertimedout(4timeoutsintotal).
Practically,theaboveresultsindicatethatIPFS-basedRIGAscanper-formrequeststoIPFSgatewaysatarateofonerequestper2secondswithatimeoutof3secondsfacingnothrottlingissuesfromthegateways.
Clearly,inthecaseofnativeIPFSprotocol,therearenosuchconsiderations;however,weexaminetheHTTPSapproachoftheinfecteddevices.
Gateway5sectimeout3sectimeoutTimeDroppedTimeDroppedhttps://ipfs.
io/ipfs/373.
4920422.
8890https://gateway.
ipfs.
io/ipfs/391.
0090374.
1110https://ipfs.
infura.
io/ipfs/854.
95301123.
7620https://xmine128.
tk/ipfs/381.
5950383.
2060https://ipfs.
jes.
xxx/ipfs/438.
049001903.
1501https://siderus.
io/ipfs/280.
2610296.
1380https://www.
eternum.
io/ipfs/594.
1571609.
2380https://hardbin.
com/ipfs/457.
3640601.
4640https://ipfs.
wa.
hle.
rs/ipfs/1234.
26301043.
5000https://ipfs.
renehsz.
com/ipfs/1482.
9310482.
2820https://cloudflare-ipfs.
com/ipfs/285.
8931289.
6221https://ipns.
co/1848.
69501143.
3980https://gateway.
swedneck.
xyz/ipfs/5952.
236315887.
39554Table2:Statisticsfor1000requestsofWikipediaarticlesusingdierentIPFSgateways.
Time(inmilliseconds)referstothetotaltimeforallrequests.
Asalreadydiscussed,thesecondexperimentaimstodeterminethetimere-quiredforcontenttobecomeavailableinIPFS.
Tothisend,weuploadatotalof1000lestoIPFSwitha4KBsizeand,later,weretrievethemusingthefourmostecientgateways,accordingtoTable2(i.
e.
https://cloudflare-ipfs.
com/ipfs/,https://xmine128.
tk/ipfs/,https://siderus.
io/ipfs/andhttps://gateway.
ipfs.
io/ipfs/).
Ineachcase,weusearelaxedtime-outof5secondstogettheles,randomlyshiftingtoanotherifonefailstotimelydeliverit.
Therefore,wecapturethetimestampwhenaddingthelestoIPFSandafterretrievingthemandcomputetheirdierence.
NotethattheaforementionedprocedurebypassesanypossiblelocalitybiassincetheleisactuallyrequestedfromaremoteIPFSgateway.
Theaveragetimeandstan-darddeviationofthefullprocedurewere3647msand3715ms,respectively.
AsillustratedinFigure4,thereareseveraloutlierswhichsignicantlyincreasethe15Figure4:Boxplotofthetimeresults(inmilliseconds)forestimatingthetimerequiredforaletobecomeavailableonIPFS.
averagevalueandstandarddeviation.
Nevertheless,theresultsindicatethattheavailabilityiswellboundedinthescaleofjustseconds,enablingreal-timeadaptablemalwarecampaigns,wheremaliciouscontentcanberapidlyspread.
6DiscussionComparedwiththecurrentstateoftheart,theproposedmethodologyofRIGAsoverIPFShasseveraladvantages.
Asinmanyapproaches,theproposedmethodologyistotallydecentralised,providingrobustness,availability,andhighscalability.
Moreover,contrarytomanyapproaches,ourIPFSmethodologyes-tablishesconnectionswithwhitelisteddomainsoverencryptedchannels,usingastandardandwhitelistedprotocol(HTTPS),andwithlegitimateandtrustedcerticates.
AtypicalexampleisCloudare,oneofthebiggestcontentdeliverynetworkswhichactsasanexitnodeforIPFS.
NotethatduetothecontinuousadoptionofblockchaineventhedirectuseofIPFSprotocolisexpectedtobewhitelistedinawiderangeofdevicesquitesoon,asitisoneofthemostviablesolutionsforstoringdataontheblockchain.
ItisworthnotingthatAquinoetal.
exploredthepotentialexploitationofIPFSfordistributingthecommandsofaC&Cserverin[11].
However,theyonlyconsiderthestaticscenariowherethebotndsthecontentuploadedonIPFSasitwouldinthecaseofotherapproachese.
g.
usingsocialnetworks.
Asalreadydiscussed,abotnetwiththeproposedmethodologyisnoteasytotakedown.
ThiscanbeattributedtotheactualnatureofIPFS.
Despiteitsdecentralisedarchitecture,IPFSnotonlydoesnothaveadeletionmechanism,butitscommunityisnoteagertosupportit.
Theblacklistingapproachwhichhasbeenproposedinthecommunitytoaddressmaliciouscontentisnotcertainthateverynodeinthenetworkwillusethat.
Moreover,whenthesizeofthe16blacklistedcontentgrowsbeyondapoint,theadditionaloverheadisexpectedtodetermanynodesfromusingit.
Therefore,giventhepropermotivation(e.
g.
Filecoins)orbymakinginfecteddevicestopincontent,thesharedcontentisexpectedtostayonthenetworkandnotbetakendown.
SinceIPFSnodeIDsarecreatedusingthehashoftheirpublickey(gener-atedusinga2048-bitRSA)anodeIDcanberesetmultipletimes.
Moreover,IPFSobjectsdonotstorewhichnodesseedthem,andthus,itisnotpossibletorelatealewithitsinitialseeder,whichhinderstrackingmechanisms.
Inthisregard,IPNSremotepinningservicessuchasTextile10implementusefulmechanismswhichenabledecentralisedcontentmanagementwithhighavail-ability,enhancingtheprivacyandanonymityofpeers.
Nevertheless,althoughdevelopersareworkingtoenhancenetwork'sanonymity,thereisstillabiggaptocover.
Forexample,sincethearchitectureofIPFSconnectsnodesbytheirproximitytoenhancenetworkeciency,usingTorconnectionswoulddisclosetheuser'sprivate,publicand.
onionaddressesinformation.
Notwithstanding,contentretrievingfromagatewayusingTorconnectionsisanonymous11.
Otherfeaturessuchasend-to-endintegritycanalsobesatisedutilisingofipfs-secdo-mainsandDNSSECsothatusersdonotneedtotrustintermediategateways12,apropertyanalogoustothatprovidedbyHTTPS(i.
e.
withInternetServiceProviders(ISP)actingasintermediategateways).
Theuseofblockchains,asproposedinbyAlietal.
[1]mayalsoprovideadecentralisedapproachonpermanentstorage;however,theirproposedsolutionallowsthecommunicationissmallchunks,andtherefore,manymessageshavetobebroadcastfromthebotmaster.
TakingintoconsiderationthelowrateofveriedcommitsoftheBitcoinnetwork,thebotswillhavetowaitfarlongerthanthereportedvalues.
Itshouldalsobenotedthatbotswouldhavetocontinuallymonitortheblockchainwhichimpliesasignicantprocessingandunnecessaryoverhead.
Onthecontrary,theIPFSapproachcandelivercontentofarbitrarysizealmostinstantlyandwithouthavingtomonitorthenetworkforupdatescontinually.
IntheIPFSimplementationofRIGAs,contrarytoDGAs,theadversarydoesnothaveanadditionalcostthatmaydiscloseheridentity.
Moreprecisely,inDGAstheadversarymustpurchasesomedomainnamesfromregistrars.
Inthisregard,theadversarymayperformthetransactionsusingstolencardsandfakeidentities,however,thecostcanbesubstantialiftheamountofdomainsislarge.
Nevertheless,theIPFSapproachdoesnotimplyanycostfortheadversaryandnosubstantialwaitperiod.
Forinstance,whenpurchasingadomainname,dependingontheregistrar,theremightbeanidleperioduntilthedomainnamebecomesavailablewhichmayrangefromhourstodays.
IntheIPFSapproachthoughthereisnoidleperiodandcontentbecomealmostimmediatelyavailable.
Finally,itshouldbenotedthattheIPFSapproachismorestealth.
IntraditionalDGAs,therequesteddomainnameduetoitsentropyorWHOIS10https://www.
textile.
io/11https://dweb-primer.
ipfs.
io/avenues-for-access/lessons/tor-gateways.
html12https://blog.
cloudflare.
com/e2e-integrity/17datamaydisclosethatitisanoutputofaDGA.
However,intheIPFScase,thisisnotrelevant.
Therequestsaremadetolegitimateandencrypteddomains.
Therefore,thiswillnotleaveanytracesintheDNSqueries.
7ConclusionsCybercrimehasbecomeaveryprotable"business"havingvariousmonetisa-tionsources.
Modernmalwarecreatorsbypassmostofcurrentsecurityarchitec-turesbyusing,amongothers,cryptographicmethodsandcovertcommunicationchannels.
Inthiswork,weexplorethepossibilitiesofextendingDGAsandthenfocusonthecontextofarecentdecentraliseddatastoragesystem,namelyIPFS.
Tothisend,werstextendthenotionofDGAsintoamoregenericframe-work,namelyRIGA,whichisafamilyofmechanismstogenerateasequenceofpossibleaddressesthatcanbeusedtohostmaliciouscontent,regardlessofthecontext.
Next,weshowcasetheexploitationoftheIPFSnetworktoenablebotmanagementandmalwarespreadingthroughthecorrespondinginstantiationofaRIGA.
OurexperimentalresultsshowthatIPFScouldbeeectivelyexploitedtoconveysuchanattack,duetothelackofdefencemechanismsintermsofqueryresolution(gateways),thelackofaneectivedeletionmechanism,anditsspeed,enablingreal-timecampaigns.
Moreover,weprovideafunctionalPRNGimplementationofanIPFS-enabledRIGA.
Finally,wediscusstheadvantagesofourapproachformalwareauthorstoraiseawarenessandmotivateresearcherstondsolutionsforsuchanemergingthreat.
Futureworkwillfocusonstudy-ingtheimpactofsystemsliketheonepresentedinthispaperintechnologieswhichuseIPFS(e.
g.
blockchain)andtheexplorationofcountermeasures(e.
g.
implementinganeectivedeletionsysteminIPFS).
AcknowledgementsThisworkwassupportedbytheEuropeanCommissionundertheHorizon2020Programme(H2020),aspartoftheprojectYAKSHA(GrantAgreementno.
780498)andisbaseduponworkfromCOSTActionCA17124:DigForASPDig-italforensics:evidenceanalysisviaintelligentsystemsandpractices(EuropeanCooperationinScienceandTechnology).
References[1]SyedTahaAli,PatrickMcCorry,PeterHyun-JeenLee,andFengHao.
Zombiecoin2.
0:managingnext-generationbotnetsusingbitcoin.
Interna-tionalJournalofInformationSecurity,17(4):411–422,Aug2018.
[2]HyrumS.
Anderson,JonathanWoodbridge,andBobbyFilar.
DeepDGA:Adversarially-tuneddomaingenerationanddetection.
InProceedingsof18the2016ACMWorkshoponArticialIntelligenceandSecurity,AISec'16,pages13–21,NewYork,NY,USA,2016.
ACM.
[3]ManosAntonakakis,TimApril,MichaelBailey,MattBernhard,ElieBursztein,JaimeCochran,ZakirDurumeric,J.
AlexHalderman,LucaIn-vernizzi,MichalisKallitsis,DeepakKumar,ChazLever,ZaneMa,JoshuaMason,DamianMenscher,ChadSeaman,NickSullivan,KurtThomas,andYiZhou.
Understandingthemiraibotnet.
In26thUSENIXSecu-ritySymposium(USENIXSecurity17),pages1093–1110,Vancouver,BC,2017.
USENIXAssociation.
[4]ManosAntonakakis,RobertoPerdisci,YacinNadji,NikolaosVasiloglou,SaeedAbu-Nimeh,WenkeLee,andDavidDagon.
Fromthrow-awaytractobots:detectingtheriseofDGA-basedmalware.
InProceedingsofthe21stUSENIXconferenceonSecuritysymposium,pages24–24.
USENIXAssociation,2012.
[5]AdamJ.
AvivandAndreasHaeberlen.
Challengesinexperimentingwithbotnetdetectionsystems.
InProceedingsofthe4thConferenceonCyberSecurityExperimentationandTest,CSET'11,pages6–6,Berkeley,CA,USA,2011.
USENIXAssociation.
[6]JohannesBader.
TheDGAofpykspa"youskypeversionisold".
https://www.
johannesbader.
ch/2015/03/the-dga-of-pykspa/,2015.
[7]JuanBenet.
IPFS-contentaddressed,versioned,p2plesystem.
arXivpreprintarXiv:1407.
3561,2014.
[8]FranCasino,ThomasK.
Dasaklis,andConstantinosPatsakis.
Asystematicliteraturereviewofblockchain-basedapplications:Currentstatus,classi-cationandopenissues.
TelematicsandInformatics,36:55–81,2019.
[9]YizhengChen,PanagiotisKintis,ManosAntonakakis,YacinNadji,DavidDagon,andMichaelFarrell.
Measuringlowerboundsofthenancialabusetoonlineadvertisers:AfouryearcasestudyoftheTDSS/TDL4botnet.
Computers&Security,67:164–180,2017.
[10]RyanRCurtin,AndrewBGardner,SlawomirGrzonkowski,AlexeyKley-menov,andAlejandroMosquera.
DetectingDGAdomainswithrecurrentneuralnetworksandsideinformation.
arXivpreprintarXiv:1810.
02023,2018.
[11]BrunoMacabeusM.
deAquino,MarcusViniciusL.
deLima,JooPaoloCavalcanteM.
deOliveira,andCidcleyTeixeiradeSouza.
Protocolosipfseipnscomomeioparaocontroledebotnet:provadeconceito.
InAnaisdoWorkshopdeSeguranaCibernticaemDispositivosConectados(WSCDC-SBRC2018),volume1,PortoAlegre,RS,Brasil,2018.
SBC.
19[12]YimingGong,SuQitian,andZaifengZhang.
ADGAodysseyPDNSdrivenDGAanalysis.
https://pc.
nanog.
org/static/published/meetings/NANOG71/1444/20171004_Gong_A_Dga_Odyssey__v1.
pdf,2017.
[13]JulianBGrizzard,VikramSharma,ChrisNunnery,BrentByungHoonKang,andDavidDagon.
Peer-to-peerbotnets:Overviewandcasestudy.
HotBots,7:1–1,2007.
[14]ThorstenHolz,ChristianGorecki,KonradRieck,andFelixCFreiling.
Measuringanddetectingfast-uxservicenetworks.
InProceedingsoftheNetworkandDistributedSystemSecuritySymposium,2018.
[15]SteveHuckle,RituparnaBhattacharya,MartinWhite,andNataliaBelo.
Internetofthings,blockchainandsharedeconomyapplications.
ProcediaComputerScience,98:461–466,2016.
The7thInternationalConferenceonEmergingUbiquitousSystemsandPervasiveNetworks(EUSPN2016)/The6thInternationalConferenceonCurrentandFutureTrendsofInformationandCommunicationTechnologiesinHealthcare(ICTH-2016)/AliatedWorkshops.
[16]DavidIrvine.
Maidsafe.
net,March112010.
USPatentApp.
12/476,229.
[17]N.
Jiang,J.
Cao,Y.
Jin,L.
E.
Li,andZ.
Zhang.
Identifyingsuspiciousac-tivitiesthroughdnsfailuregraphanalysis.
InThe18thIEEEInternationalConferenceonNetworkProtocols,pages144–153,Oct2010.
[18]OrKatz,RavivPerets,andGuyMatzliach.
Diggingdeeper-anin-depthanalysisofafastuxnetwork.
https://www.
akamai.
com/us/en/multimedia/documents/white-paper/digging-deeper-in-depth-analysis-of-fast-flux-network.
pdf,2016.
[19]MatKelly,SawoodAlam,MichaelL.
Nelson,andMicheleC.
Weigle.
Inter-planetarywayback:Peer-to-peerpermanenceofwebarchives.
InNorbertFuhr,LaszloKovacs,ThomasRisse,andWolfgangNejdl,editors,ResearchandAdvancedTechnologyforDigitalLibraries,pages411–416,Cham,2016.
SpringerInternationalPublishing.
[20]BKrebs.
Mariposabotnetauthorsmayavoidjailtime.
https://krebsonsecurity.
com/2010/03/mariposa-botnet-authors-may-avoid-jail-time/,2010.
[21]DaipingLiu,ZhouLi,KunDu,HainingWang,BaojunLiu,andHaixinDuan.
Don'tletonerottenapplespoilthewholebarrel:Towardsauto-mateddetectionofshadoweddomains.
InProceedingsofthe2017ACMSIGSACConferenceonComputerandCommunicationsSecurity,CCS'17,pages537–552,NewYork,NY,USA,2017.
ACM.
[22]PratyusaK.
Manadhata,SandeepYadav,PrasadRao,andWilliamHorne.
Detectingmaliciousdomainsviagraphinference.
InMiroslawKutylowski20andJaideepVaidya,editors,ComputerSecurity-ESORICS2014,pages1–18,Cham,2014.
SpringerInternationalPublishing.
[23]SteveManseld-Devine.
Themalwarearmsrace.
ComputerFraud&Se-curity,2018(2):15–20,2018.
[24]JoannaMoubarak,EricFiliol,andMarounChamoun.
Developingak-arymalwareusingblockchain.
arXivpreprintarXiv:1804.
01488,2018.
[25]YacinNadji,RobertoPerdisci,andManosAntonakakis.
Stillbeheadinghy-dras:Botnettakedownsthenandnow.
IEEETransactionsonDependableandSecureComputing,14(5):535–549,2017.
[26]ShishirNagaraja,AmirHoumansadr,PratchPiyawongwisal,VijitSingh,PragyaAgarwal,andNikitaBorisov.
Stegobot:acovertsocialnetworkbotnet.
InInternationalWorkshoponInformationHiding,pages299–313.
Springer,2011.
[27]SatoshiNakamoto.
Bitcoin:Apeer-to-peerelectroniccashsystem,2008.
[28]R.
Perdisci,I.
Corona,andG.
Giacinto.
Earlydetectionofmaliciousuxnetworksvialarge-scalepassivednstracanalysis.
IEEETransactionsonDependableandSecureComputing,9(5):714–726,Sept2012.
[29]S.
Pletinckx,C.
Trap,andC.
Doerr.
Malwarecoordinationusingtheblockchain:Ananalysisofthecerberransomware.
In2018IEEECon-ferenceonCommunicationsandNetworkSecurity(CNS),pages1–9,May2018.
[30]DanielPlohmann,KhaledYakdan,MichaelKlatt,JohannesBader,andElmarGerhards-Padilla.
Acomprehensivemeasurementstudyofdomaingeneratingmalware.
In25thUSENIXSecuritySymposium(USENIXSe-curity16),pages263–278,Austin,TX,2016.
USENIXAssociation.
[31]BPrince.
Flashbackbotnetupdatedtoincludetwitterasc&c.
securityweek,2012.
[32]BrunoProduit.
UsingBlockchainTechnologyinDistributedStor-ageSystems.
https://courses.
cs.
ut.
ee/MTAT.
07.
022/2018_spring/uploads/Main/bruno-report-s17-18.
pdf,2018.
[33]JustinMRaoandDavidHReiley.
Theeconomicsofspam.
JournalofEconomicPerspectives,26(3):87–110,2012.
[34]StefanoSchiavoni,FedericoMaggi,LorenzoCavallaro,andStefanoZanero.
Phoenix:Dga-basedbotnettrackingandintelligence.
InSvenDietrich,editor,DetectionofIntrusionsandMalware,andVulnerabilityAssessment,pages192–211,Cham,2014.
SpringerInternationalPublishing.
[35]A.
K.
SoodandS.
Zeadally.
Ataxonomyofdomain-generationalgorithms.
IEEESecurityPrivacy,14(4):46–53,July2016.
21[36]MelanieSwan.
Blockchainthinking:Thebrainasadac(decentralizedautonomousorganization).
InTexasBitcoinConference,pages27–29.
Chicago,2015.
[37]NickSzabo.
Theideaofsmartcontracts,1997.
[38]DucTran,HieuMac,VanTong,HaiAnhTran,andLinhGiangNguyen.
Alstmbasedframeworkforhandlingmulticlassimbalanceindgabotnetdetection.
Neurocomputing,275:2401–2413,2018.
[39]ViktorTronetal.
Swarm.
https://swarm-gateways.
net/bzz:/theswarm.
eth/#the-thsph-orange-paper-series,2016.
[40]S.
Yadav,A.
K.
K.
Reddy,A.
L.
N.
Reddy,andS.
Ranjan.
Detectingalgorithmicallygenerateddomain-uxattackswithDNStracanalysis.
IEEE/ACMTransactionsonNetworking,20(5):1663–1677,Oct2012.
[41]SandeepYadavandA.
L.
NarasimhaReddy.
WinningwithDNSfail-ures:Strategiesforfasterbotnetdetection.
InMuttukrishnanRajarajan,FredPiper,HainingWang,andGeorgeKesidis,editors,SecurityandPri-vacyinCommunicationNetworks,pages446–459,Berlin,Heidelberg,2012.
SpringerBerlinHeidelberg.
[42]G.
Zhao,K.
Xu,L.
Xu,andB.
Wu.
DetectingAPTmalwareinfectionsbasedonmaliciousDNSandtracanalysis.
IEEEAccess,3:1132–1142,2015.
[43]YuryZhauniarovich,IssaKhalil,TingYu,andMarcDacier.
AsurveyonmaliciousdomainsdetectionthroughDNSdataanalysis.
ACMComputingSurveys,51(4):67:1–67:36,July2018.
[44]YonglinZhou,Qing-ShanLi,QidiMiao,andKangbinYim.
DGA-basedbotnetdetectionusingDNStrac.
J.
InternetServ.
Inf.
Secur.
,3:116–123,2013.
22