hijackattackshavetakenplaceintheInternetformonthsorevenyears.Weachievethisbyleveragingalarge-scaledatacollectionsystemdubbedSPAMTRACERthatweintroducedin[47].Thislargedatasetisthenminedbymeansofamulti-stagescoringand lteringprocess,whoseresultsareenrichedusingexternaldatasourcesandclusteredinordertorevealthemodusoperandioftheattackers.
Beforemovingintothecoreofthepaper,wewouldliketomakeitclearwhatthispaperisnotabout:
ThispaperdoesnotofferyetanotherBGPhijackdetectiontechnique.Wedonotwanttoincrementallyimprovethestateoftheartinthatspace.Thenovelty,atthecontrary,residesintheidenti edhijacksandintheirdetailedanalysis.
Alongthepreviouspoint,wedonotclaimthatwehavefoundallBGPhijacksthatcouldhavebeenfoundinourdataset.Whatmattersisthatourresultsmustbeseenasaproofoftheexistenceoftheserecurringattacks.Someofthemwerecon rmedbyanISPwhowasunwittinglyinvolvedinseveralhijackcasesapparentlyperformedbyaRussianspammer.Webelievethatothersarestillhiddeninourdataset.However, guringouthowto ndthemallisleftasanexerciseforfuturework.
Thispaperdoesnotdiscusstheoptimalchoiceofthefewparametersusedinouralgorithms.Runningarigoroussensitivityanalysistofurtherimproveourresultsissomethingweareworkingon.However,thekeycontributionofthispaperisnotthemethoditselfbut,instead,theidenti cationofrealmaliciousattacksroutinelyhappeninginthewild.
Alternatively,someexistingproposals[29],[26],[28],[34]aimatdetectingIPpre xhijackingbypassivelymonitoringtheroutinginfrastructure.Howeverduetothestrongsimilar-itybetweenIPpre xhijackingandsomelegitimateroutingchangesthosemethodssufferfrommanyfalsepositives.Otherproposals[23],[40],[49],[50]leverageactiveprob-ingofnetworkstogetherwithpassivemonitoringtoimprovethedetectionbyassessingtheimpactofBGProutingchangesonthedataplane.InordertostudyBGPhijackingspammersweusetheSPAMTRACERmethodologyweintroducedin[47]forcollectingacomprehensivesetofrouting-levelfeaturesaboutspamnetworks.In[38]Roughanetal.advocatesthatneitherBGPnortraceroutemeasurementsweredesignedtoinfertheAS-levelconnectivityoftheInternetandcapturethecomplexinter-ASrelationships,henceallresultsinferredfromsuchdatacanonlybeasaccurateasthedata.However,wetrytobalancethislimitationbysettingupourowndatacollectionprocessallowingustocollectthemostappropriatedataforstudyingtherouting-levelbehaviorofspammers.
In2006,Ramachandranetal.[35]introducedanewphe-nomenoncalled“BGPspectrumagility”,wheretheyclaimedtohaveobserved,overaperiodofafewmonths,spamfromasetoflarge(i.e.,/8)previouslyunannouncedIPaddressblockshijackedforaveryshortperiodoftime(i.e.,lessthanoneday).Later,Huetal.[23]andDuanetal.[20]con rmedtheseobservations.However,wehaverecentlyshownin[46]throughapracticalcasestudythatcorrelatingBGPabnormaleventswithmaliciousnetworktraf cisinsuf cienttoconclu-sivelyidentifymaliciousBGPhijacks.Meanwhile,Schlampetal.[39]describedauniquecasewhereacoupleofIPaddressblockswerehijackedformonthstoperformmaliciousactivities,suchasspamming.
ThesefewpublicationsshowtheexistenceandtherealityofBGPhijacksinthewildbutthescarcityoftheattacksob-servedsince2006givetheimpressionthatthisthreatremainshighlyanecdotalandthatnoinfrastructureseemstohavebeenputinplacebyhackerstoautomatizeef ciently,systemati-cally,thelaunchingofBGPhijacks.Theresultswepresenthereafterwillportrayaverydifferentsituationinwhich,everyday,severalBGPhijacksaretakingplace.Furthermore,weshowthatthisisbynomeansanewphenomenon.Ourdatahighlightthatthishastakenplacefor18months,withoutanyonenoticingitapparently.
III.
EXPERIMENTALSETUP
Therestofthispaperisorganisedasfollows.Section2presentstherelatedwork.Section3describestheenviron-mentalsetupwehavebuilt,namelythedatacollectionandanalysisprocesses.Section4isthecoreofthepaperandgoesintothedetailsofalltheresultsobtainedwhenusingthisenvironmentforseveralmonths.Section5offerssomeinsightsontheeffectivenessofcurrentcounter-measurestodefeattheattackswehavefound.Section6summarizesthelessonslearnedandconcludesthepaper.
II.
RELATEDWORK
BGPhijackingdefensesolutionsaretwofold:(i)detectiontechniquesaimatmonitoringtheInternetroutinginfrastructureandtriggeralarmsuponabnormalroutingchanges,and(ii)othertechniquesaimatprovidingBGPwithnewmechanismstomitigateorpreventhijacking.Unlikehijackmitigationorpreventiontechniques,detectionmethodsrequirenochangestoroutersoftware,whichusuallymakesthemreadilyandeasilydeployable.
SometechniqueshavebeenproposedtobringsecurityintoBGP[24],[25],[30],usuallyusingcryptographytosignsomeelementsofBGPupdatestoensureroutinginformationauthenticityandintegrity.InthelastfewyearsaBGPsecurityframeworkrelyingonaRPKI[31]tosecureIPpre xorigina-tion[24]hasgainedalotofattentionandisnowprogressivelybeingdeployed.
Wehavesetupacomprehensiveexperimentalenviron-menttostudytheBGPhijackingspammersphenomenon.ThecompletesetupisdepictedinFigure1.Ourgoalhereistocollectroutingdatarelatedtospamnetworks,extractthisdataIPaddressblocksexhibitinganroutingbehaviorandretaintheonesmostlikelyindicatingtheymightresultfromaBGPhijack,manually(in)validateeachcandidatehijackbytakingofexternaldatasources,and nallyinvestigatetherootcausebehindsomevalidatedmalicioushijackstoobtainnewinsightsintohijackingspammersbehavior.
OurexperimentalsetupbuildsuponSPAMTRACER[47],asystemdesignedforthecollectionandanalysisofroutingdatarelatedtospamnetworks.Theassumptionbehindthis
热门文档