68
Cyber Grand Challenge Frequently Asked Questions (FAQ) August 3, 2016 Defense Advanced Research Projects Agency Information Innovation Office 675 North Randolph Street Arlington, VA 222032114

Cyber Grand Challenge - MIT Lincoln Laboratory · Q157 Added Jan 27, 2016 Q158 Added Mar 16, 2016 Q159 – Q161 Added Mar 16, 2016 Q162 – Q163 Added Mar 25, 2016 Q164 Added May

  • Upload
    others

  • View
    0

  • Download
    0

Embed Size (px)

Citation preview

Page 1: Cyber Grand Challenge - MIT Lincoln Laboratory · Q157 Added Jan 27, 2016 Q158 Added Mar 16, 2016 Q159 – Q161 Added Mar 16, 2016 Q162 – Q163 Added Mar 25, 2016 Q164 Added May

Cyber Grand Challenge Frequently Asked Questions (FAQ) August 3, 2016  

                                                                Defense Advanced Research Projects Agency Information Innovation Office 675 North Randolph Street Arlington, VA  22203‐2114 

Page 2: Cyber Grand Challenge - MIT Lincoln Laboratory · Q157 Added Jan 27, 2016 Q158 Added Mar 16, 2016 Q159 – Q161 Added Mar 16, 2016 Q162 – Q163 Added Mar 25, 2016 Q164 Added May

               

 

Page 3: Cyber Grand Challenge - MIT Lincoln Laboratory · Q157 Added Jan 27, 2016 Q158 Added Mar 16, 2016 Q159 – Q161 Added Mar 16, 2016 Q162 – Q163 Added Mar 25, 2016 Q164 Added May

Document Change Summary  

Section  Description  Date 

InitialPublication Nov7,2013Q1 EastCoastCompetitorDaydatesadded Nov14,2013Q1 ModifiedtoincludeWestCoastCompetitorDay

informationandremovedCGCwebsiteURLNov27,2013

Q10 Added‐WhattypeofsecurityvulnerabilitieswillCGCaddress?

Nov27,2013

Q11–Q33 Added Dec17,2013Q34–Q56 Added Dec24,2013Q57–Q59 Added–UpdatetoscoringmethodsandinitialCGC

environmentAPI.Mar10,2014

Q26–Q27 Obsoletedentriesreplacedbyentries58and59. Mar10,2014Q60–Q64 Added Jul24,2014Q65–Q73 Added Aug29,2014Q74 Added Oct21,2014Q75–Q77 Scoredeventupdate1. Nov14,2014Q78–Q79 Tiebreakerpubliccommentupdate1. Feb13,2015Q80–Q92 Scoredeventupdate2. Feb13,2015Q93–Q102 Added Mar18,2015Q103–Q106 Added Apr10,2015Q107–Q111 Added May6,2015Q112 Added May12,2015Q112 Revised May22,2015Q113–Q119 Added May22,2015Q120–Q124 Added May28,2015Q125–Q126 Added May29,2015Q127–Q130 Added Jun2,2015Q131 Added Sep1,2015Q132–Q144 Added Oct20,2015Q145‐Q156 Added Dec10,2015Q157 Added Jan27,2016Q158 Added Mar16,2016Q159–Q161 Added Mar16,2016Q162–Q163 Added Mar25,2016Q164 Added May6,2016Q165–Q170 Added May19,2016Q171 Added May27,2016Q172‐Q174 Added Jun3,2016Q175 Added Jun8,2016

Page 4: Cyber Grand Challenge - MIT Lincoln Laboratory · Q157 Added Jan 27, 2016 Q158 Added Mar 16, 2016 Q159 – Q161 Added Mar 16, 2016 Q162 – Q163 Added Mar 25, 2016 Q164 Added May

Q176 Added Jun10,2016Q177–Q178 Added Jun23,2016Q179 Added Jun28,2016Q180 Added Jul12,2016Q181 Added Jul15,2016Q182 Added Jul21,2016Q183‐Q184 Added Jul22,2016Q185 PublishCFETeamPhrases Aug3,2016Q186 Added Aug3,2016

Page 5: Cyber Grand Challenge - MIT Lincoln Laboratory · Q157 Added Jan 27, 2016 Q158 Added Mar 16, 2016 Q159 – Q161 Added Mar 16, 2016 Q162 – Q163 Added Mar 25, 2016 Q164 Added May

Q186:EffectiveAugust3,2016at11:00pmPT,thisFAQisclosed.HowwillnewFAQentriesbepublished?A186:AllnewFAQentrieswillbepublishedintheCGCEventFAQ.[1]https://github.com/CyberGrandChallenge/Event‐FAQ

Page 6: Cyber Grand Challenge - MIT Lincoln Laboratory · Q157 Added Jan 27, 2016 Q158 Added Mar 16, 2016 Q159 – Q161 Added Mar 16, 2016 Q162 – Q163 Added Mar 25, 2016 Q164 Added May

Q185:WhatwerethecompetitorteamTeamPhrasesusedtocontributetothecalculationofthemasterseed?A185:TheTeamPhrasessolicitedfromfinalistsandusedaccordingtoA176oftheFAQarepublishedinthebelowJSON:[ { "phrase": "defa5b1925203b76ee19bb1102e620754fb655b11b52399da226354630e1f18b61f439b8cb2d520de99589c68fdc5312ab6b229879f7bda06d285cba98a961b7fe63ba3e9b96de11254196e2a73dab6099058af816a747a1182868b868e58eda8206bc33aba51964c4ef77aa4378d5665b66db8b18ae4eb6ed99e560b89ce2467b4bfff16dea49d6dcb88101392f91e0ca4c4ed672e30cc52b3f7c45a0a8d39c7a4b41b83a0f7e00b50c5ce123c38645a7c495b53d32df5b8b57dfb3a0933bce478930cd6b4692e8a57b0c335997c6e86a99114a0ca5c0751118ffe7d989b298d15e5f3df7cd9546290bedb7d79d87f91abafbb4a953078cac4aa53fd10caca1", "name": "CodeJitsu" }, { "phrase": "Remember: Innovation can occur anywhere -- If we knew what we were doing, it would not be called research", "name": "CSDS" }, { "phrase": "Deep Red [team photo: (-(-_(-_-)_-)-)] is excited to participate in the Cyber Grand Challenge. They do not battle with o==]:::::::::::>'s, but with 1's and 0's for treasure and eternal glory. Their specially chosen TeamPhrase shall surely secure their victory.", "name": "DeepRed" }, { "phrase": " ;.\n : . ..\n .NNN. cco XMM k. .\n .MMM. .OK0 kMM l:; l0Wo\n .,;'.,MMM. . . kMW xkl. ,MMo\n '0WMNKNMNMMM: dKK :dMMMMMXO: ,xXMMMWKk, 0MM. ;XKWW :0MMN0x..\n cMMM' lMMMMMX..KMX...lMMMMc.'cWM0 .KMMXl,llWMM: WMM ;0NM0:' ..'MMK'.\n NMMc d0XMMMx. ;MMd .XMMW0, XMMk :MMW0. NMMNMMNx. .,..MMK\n MMMc c:oXM0. .MMl 0:dxMMMWXx:..xOMMMKKXNWMMMWW0'.KMMXMMWO0' WMK\n MMM0c .lXMK .MM; .;dMMMWl.'MMNW0 .KMXl XW0MO. . MM0\n ,NMMd .XMMW. 'MM' oWM; .NMMMx OMNWN. :MMXl. XMMd 0WMMN 'MMX\n kOWMMWXWMXWMM. ;MN cMMXxdNMMMl. xMMMKkd0MMKc XMN' .KMWK. kMWKxo\n : l;Ko'. . .;c .cxxdOO .:odkxol' ,. . . cddl;..\n :kO. . .\n .", "name": "Disekt" }, { "phrase": "\n AAAAA\n AAAAAAAAAAAAAAA\n AAAAAAAAAAAAAAAAAAA\n AAAAAAAAAAAAAAAAAAAAAAA\n AAAAA AAAAAAAAAAA AAAAA\n AAAAAAA AAAAAAAAA AAAAAAA\n AAAAAAAAA AAAAAAA AAAAAAAAA\n AAAAAAAAAA AAAAAAAAAA\n AAAAAAAAAAA AAA AAAAAAAAAAA\n AAAAAAAAAAA A AAAAAAAAAAA\n AAAAAAAAAAA AAAAAAAAAAA\n AAAAAAAAAAA AAAAAAAAAAA\n AAAAAAAAAAAAAAAAAAAAA\n AAAAAAAAAAAAAAA\n AAAAA\n", "name": "ForAllSecure" }, { "phrase": "Shell we play a game?", "name": "Shellphish" }, { "phrase": "GrammaTech and UVA bring you Xandra, Defeneder of Humanity.", "name": "TECHx" } ]

Page 7: Cyber Grand Challenge - MIT Lincoln Laboratory · Q157 Added Jan 27, 2016 Q158 Added Mar 16, 2016 Q159 – Q161 Added Mar 16, 2016 Q162 – Q163 Added Mar 25, 2016 Q164 Added May

Q184:WenoticedtheTIAPI[1]publishesNNNvalues"between0and100,reflectingpercentage"forpollfeedbackbymultiplyingtheavailabilityscorespecifiedintheCFEscoringdocument[A58]by100.Doesthisconventionholdforscoresaswell(i.e.ascoreof4becomes400intheTIAPI)?A184:Yes[1]https://github.com/CyberGrandChallenge/cgc‐releasedocumentation/blob/master/ti‐api‐spec.txtQ183:IntheFinalistEventInformationdocument,youstate"thepossibilityexiststhatcompetitorHPCaccesswillnotberestoredpriortoCFEcomputation".CanwespecifyapowerupscheduleforourbladesintheeventthatourCRSmustbepoweredupcoldforCFE?A183:No.Q182:CanthesameCSIDappearintwodifferentcompetitions?A182:Noprotectionmechanismexiststhatwouldpreventthis.Q181:We'dliketooptimizeforidenticalbinaries;howmanyofourcompetitorswillfieldChallengeSetscontainingidenticalChallengeBinariesduringCFE?A181:Unknown.Q180:WhatisthetiebreakeralgorithmforCFE?A180:InthehighlyunlikelyeventofatiedscoreinCFE,DARPA/CGCwillusethefollowingtiebreakingalgorithmonthescoresofteamscompliantwiththeCGCRulesForCFE,aTiedBlockisanygroupofscoresinadjacentplacesforwhich1>thescoresareequaland2>theblockoccludesaprizeplace(1st,2nd,or3rd).Forinstance,threeteamstiedfor2nd,3rd,and4thplacewouldconstituteasingleTiedBlock,occluding2ndand3rdprizeplaces.ForeveryTiedBlockattheconclusionofCFEcomputation,thetiebreakingprocessisasfollows:a)Discardonescoredround'sworthofpointsfromeveryCRSinaTiedBlock

fromtheendoftheCFEcomputationrecordb)RemoveanyCRSfromtheTiedBlockwhosescoreisnolongertiedwiththe

Block

Page 8: Cyber Grand Challenge - MIT Lincoln Laboratory · Q157 Added Jan 27, 2016 Q158 Added Mar 16, 2016 Q159 – Q161 Added Mar 16, 2016 Q162 – Q163 Added Mar 25, 2016 Q164 Added May

c)IfanyTiedBlocksremain,repeatthetiebreakerata)Forexample,attheconclusionofCFEcomputation,1stplaceand2ndplacesystemshavethesamescore,constitutingaTiedBlock.3rdand4thplacesystemsarealsotied,constitutingasecondTiedBlock.Thetiebreakerisrunandthescoresfromthefinalroundareremoved.1stplaceand2ndplacearenolongertied,however3rdplaceand4thplaceremaintied.Forthepurposesofseparating3rdand4thonly,asecondroundisremoved.Thisprocessrepeatsuntil3rdand4thplaceareseparated.IntheeventthattheabovealgorithmcannotseparateaTiedBlockoverthefullCFEcomputationduration,DARPAwillissuearankingbyexpertjudgment.Q179:WilltheCFEEventPlanincludethenumberofroundseachCSwillbefielded?A179:Yes;DARPA/CGCwillneverPlantofieldaCSforlessthan10rounds.Q178:WewerereadingthroughtheCGCRulesandnotedthatAreaofExcellence#5says“mitigatesecurityflaws[using]anetworksecuritydevice”.Dowereallyhavetodealwithalltheproblemsofanetworksecuritydevice?WeuploadedsomeIDSsignaturesthatDoStheIDSdeviceandnoticedthatmax’ingouttheloadonthedevicecausedperformanceimpacttoeveryChallengeSetonourdefendedhost.CouldyouperhapsgiveusavirtualizednetworksecuritydeviceforeveryChallengeSet?A178:No.TheCGCRulesspecifythat“theCRSconfrontstheCFEnetworkfromthevantagepointofarealworldnetworkdefender”usinga“networksecuritydevice”anda“defendedhost”.Thesechallenges:load,resourcecontention,andnetworkconcurrencyarerealworldconstraintsthathavealwaysbeenpartoftheCapturetheFlagtraditionandwereguaranteedbytheCGCRulessincetheinceptionoftheChallenge.Competitorsareassuredthatnolate‐breakingchangeswillbemadetotheRulesthatwouldcreatealessrealisticChallenge. Q177:WhatistheofficialOSkernelofCFE? A177:Nosuchkernelhasbeenspecified,exists,orcanexist.DECREEhasbeenimplementedasalayeroveranopensourceoperatingsystemthatcontinuestoreceiveupdatesandsecuritypatchesfromaglobalcommunityofresearchersonatimelinethatcannotbepredictedbyDARPA/CGC.DARPA/CGCreservestheabilitytocontinuetosecureallpartsoftheCFEarchitectureincludingthekerneluptoandduringCFEagainstanythreattotheintegrityofCFE,toincludesecurityweaknesses,emergentthreats,executiondivergenceattacks,dataandmemoryattacks,hardware/softwareattacksutilizingRowhammer,memorydeduplication,etc.DARPA/CGChascommittedtotheDECREEABIandtheTIAPI;seealsoA137,A174.Anycompetitorapproachwhosesuccessisdictatedbyhardware,firmware,

Page 9: Cyber Grand Challenge - MIT Lincoln Laboratory · Q157 Added Jan 27, 2016 Q158 Added Mar 16, 2016 Q159 – Q161 Added Mar 16, 2016 Q162 – Q163 Added Mar 25, 2016 Q164 Added May

orkernelerratamayfailinarealworldexercise,toincludedeploymenttomillionsofcomputersortheCFEinfrastructure.Q176:HowwilltheCFEframeworkgeneraterandomnumbersduringtheevent?A176:AseedvaluewillinitializepseudorandomnumbergeneratorsusedinCFE,toincludethePRNGsusedtocreatetheflagpage,pollingschedule,etc.Thisseedvaluewillbearrivedatthroughacalculationconductedinpublicview,describedinthisentry.Thenameofthisseedvalueismasterseed,hereafterMS.EachteamisaskedtoprovidetoDARPA/CGC,bymidnightEDTonJune17,2016,aphrasethatwillcontributetothecalculationofMS.Thecalculationisexplainedbelow,where:

HisdefinedasthebinarydigestoutputofSHA384 HEXisanASCIItransforminwhicheverybyteofinputisconverted

intoa2‐characterhexadecimalrepresentation ASCIINUMisanASCIItransforminwhichthedecimalformofa

numberisprintedinASCII XORisdefinedasbit‐wiseexclusiveor thecommacharacter(“,”)isdefinedasconcatenation.

Thecalculationisasfollows:MS=H(TeamName1,TeamPhrase1)XORH(TeamName2,TeamPhrase2)XORH(TeamName3,TeamPhrase3)XORH(TeamName4,TeamPhrase4)XORH(TeamName5,TeamPhrase5)XORH(TeamName6,TeamPhrase6)XORH(TeamName7,TeamPhrase7)XORH(TeamNameDARPA,TeamPhraseDARPA);DARPAwillcommittoTeamPhraseDARPAonJune10,2016bychoosingarandomnumberrandpublishingHEX(H(H(“DARPA”,TeamPhraseDARPA),ASCIINUM(r)))inthisentry.DARPAwillpublishallcompetitorteamTeamPhrasesaheadoftheCFEintheCGCFAQ.DARPAwillpublishTeamPhraseDARPAandraftertheCFEintheCGCFAQ.

Page 10: Cyber Grand Challenge - MIT Lincoln Laboratory · Q157 Added Jan 27, 2016 Q158 Added Mar 16, 2016 Q159 – Q161 Added Mar 16, 2016 Q162 – Q163 Added Mar 25, 2016 Q164 Added May

TeamsmaycommunicatetheirTeamPhrasetoDARPA/CGCvia:Email:[email protected](S/MIMEauthsupportedperCQEinstructions).

Post:DARPA/I2O

Attn:CyberGrandChallenge675NorthRandolphStreetArlington,VA22203‐2114

ATeamPhrasemaybeofanylength.TeamsareencouragedtochooseaTeamPhrasethatcanbeexpressedinASCIIandwillsurvivegovernmentreviewforpublicposting.AnyTeamPhrasenotreceivedbymidnightEDTonJune17,2016willbesettotheNULLstring.HEX(H(H(“DARPA”,TeamPhraseDARPA),ASCIINUM(r)))is:

1773ec260768b2d17bd1cdd6fc54a19e619ee0cc58edcab0f19ef2a3212879670ede7121914ed1a55eac4565a0bf88ac Q175:IsDARPA/CGCacceptingsuggestionsforitscryptographiccommitmentstrategyforCFE? A175:TheCFEcommitmentmechanismshavebeendesignedovertwoyearsofChallengeexecutionandcryptographyreviewiscomplete.Nosuggestionswillbeaccepted.Q174:WillthescoringalgorithmdescribedinA58,accessedontheCFEinfrastructureperA13oftheTrialsFAQ[1],andprovidedasanoracletocompetitorsperA137bemodifiedpriortoCFE?A174:No.[1]https://github.com/CyberGrandChallenge/Event‐FAQ/blob/master/event_faq.mdQ173:WillourCRSknowinadvanceonwhichroundaChallengeSetwillberemoved?A173:No.

Page 11: Cyber Grand Challenge - MIT Lincoln Laboratory · Q157 Added Jan 27, 2016 Q158 Added Mar 16, 2016 Q159 – Q161 Added Mar 16, 2016 Q162 – Q163 Added Mar 25, 2016 Q164 Added May

Q172:WillDARPA/CGCprovideamappingthatidentifieseachopponentCRSduringCFE?A172:No.Q171:WhatistheautonomypolicyforCFE? 

A171:Followingconsiderationofresponsestotheperiodofpubliccomment,carefulobservationofTrials,andcontinuedobservationofSparringPartner,DARPA/CGChasdeterminedthatthefollowingautonomypolicywillgovernCFE: AllroundsofCFEwillbecomputedautonomously.SeealsoSection4oftheCGCRules. Q170:InCFE,ineveryround,willeveryteambescoredonthesamecorpusofChallengeSets?A170:Yes.Q169:WhatisthedurationofthecomputationofCFE?A169:CFEcomputationwillbeginonthemorningofAugust4th,2016,andoccuroverroughlytenhoursofelapsedtime,subjecttoconstraintsdescribedinA168.CFEcomputationisprojectedtofinishduringtheCFEliveevent[1].SuccessfulcomputationofCFErequiresaminimumof40scoredrounds.[1]DetailsontheCFEliveeventaredistributedonwww.cybergrandchallenge.comQ168:WhatisthefixedroundscheduleforCFE?A168:CFEwillnotbecomputedusingafixedroundschedule.CFEwillbecomputedusinganeventplan.Theeventplanwillbeconsultedifanyrisktoeventexecutionisencountered,suchas:‐Naturaldisaster‐Coolantleak‐Powerfailure[1]‐Majorsoftwarefaultgeneratedbyadversariallyformedmachineinputs

(DECREEkernelpanic,etal).‐Thermaloverload‐Threattostorageoftheeventrecord(infrastructurestoragefailure,primary

orsecondary)‐Otherinfrastructurehardwarefailure(suspectedorconfirmed)‐OtherHPChardwarefailure(suspectedorconfirmed)‐Previouslyundiagnosedinfrastructuresoftwarefault

Page 12: Cyber Grand Challenge - MIT Lincoln Laboratory · Q157 Added Jan 27, 2016 Q158 Added Mar 16, 2016 Q159 – Q161 Added Mar 16, 2016 Q162 – Q163 Added Mar 25, 2016 Q164 Added May

DARPA/CGCwillpublishacryptographiccommitmenttothiseventplananddistributetheplaintextofthiseventplanafterCFE.[1]https://lasvegassun.com/news/2016/may/11/power‐outage‐stops‐play‐for‐40‐minutes‐at‐vegas‐st/Q167:We’veprototypedupsomecodethatsendsallourcellphonesatextmessagewhenSparringPartnermatchesoccur.DoesDARPA/CGCobjecttosemi‐autonomousparticipationinSparringPartnermatches?A167:No.Q166:We’rerunninga“ChaosMonkey”[1]fulltimeonourCRSthatdoesthingslikerandomlycorruptmemory,shutdownnodes,corruptthedisk,andterminateprocesses‐gettingtheCRSreadyforafullyautonomousevent.Unfortunately,ourChaosMonkeyiscausingustoperformsub‐optimallyinourmatchesagainstSparringPartner.WillimperfectplayversusSparringPartnercountagainstourCFEscoreinanyway?A166:No.[1]http://techblog.netflix.com/2012/07/chaos‐monkey‐released‐into‐wild.htmlQ165:We’reconcernedthatourcompetitorsmayanalyzeourreplacementbinaries(RBs)throughtheconsensusevaluationprocessandco‐optportionsofourRB,makingallofourCBs/RBsaCRS‐formedinputemittedontotheCGCnetworkandpotentiallydestinedforacompetitor’sCB.Couldthishappen?A165:Yes.Q164:Onceremoved,willaChallengeSeteverbere‐introducedinCFE?A164:No.Q163:WhathappenswithanewlyintroducedChallengeSet(CS)?A163:InA157normalplaywasdescribed.ForanygivenChallengeSetM,itsinitialroundsarecoveredinthisentryasfollows:

Page 13: Cyber Grand Challenge - MIT Lincoln Laboratory · Q157 Added Jan 27, 2016 Q158 Added Mar 16, 2016 Q159 – Q161 Added Mar 16, 2016 Q162 – Q163 Added Mar 25, 2016 Q164 Added May

WhenMisfirstintroducedinroundi:

i:Misfielded.ThefirstroundMispolled.i:ThefirstroundinwhichCRSformedinputsforMmaybesubmitted.i+1:ThefirstroundPoVmodulesthatseektoprovevulnerabilityinMmaybethrown.ReplacementChallengeBinariesandIDSrulesfilesforMreceivedperA161priortoroundi+2:

i+2:availableforconsensusevaluation. i+3:fielded.ThefieldingofReplacementChallengeBinariesandIDSrulesfilesincursservicedowntimeinamannersimilartoA157.Q162:HowmanyChallengeSetswillbeactiveatonceduringCFE?A162:Amaximumof30ChallengeSetswillbeactiveduringanygivenround.Q161:IfweuploadabunchofIDSrulesfilesforthesameCS,whichonedoyouuse?A161:Theinfrastructurewillfieldthelastcompleteduploadperround.ThisappliestoReplacementChallengeBinariesaswell. Q160:WillaCSroundscorebezeroperA157ifallCBsandtheIDSrulesfileforthatCSareidenticalbetweenrounds? A160:No. Q159:IsaCRSrequiredtore‐submitidenticalReplacementChallengeBinaries,IDSrulesfilesorPoVmoduleseveryroundaCSisinplay? A159:No. Q158:WhatisDARPA’sprocesstoanswercompetitorquestions?A158:DARPA/CGCisacompetitionwithsubstantialprizesatstake;thisintroducesuniquerequirementsontraditionalsoftwareQ&Aprocesses:

Neverprovidecompetitiveadvantagetoaparticularcompetitor.Allanswerstoallquestionsmustbemadeavailabletoallcompetitorssimultaneously;thisensuresthatnocompetitorhasaccesstomoreinformationaboutthecompetitionthananyothercompetitor.

Neverprovidecompetitivedisadvantagetoaparticularcompetitor.IfacompetitorasksDARPA/CGC“oursecretsauceisketchup,pleasepredictthe

Page 14: Cyber Grand Challenge - MIT Lincoln Laboratory · Q157 Added Jan 27, 2016 Q158 Added Mar 16, 2016 Q159 – Q161 Added Mar 16, 2016 Q162 – Q163 Added Mar 25, 2016 Q164 Added May

performanceofketchupinCGC”,DARPA/CGCcannotusethequestionaswrittenasrevealingthetextofthequestionwouldrevealthesecretsauceandcreatecompetitivedisadvantage.

DARPA/CGCmustnotrestatepubliclyavailableinformation.Whenaquestion

hasalreadybeenansweredpublicly,restatingtheanswerprovidesthepossibilityofalternateinterpretationsthatharmtheintegrityofthecontest.Forthisreason,DARPA/CGCmustalwaysrepeatexistingpublicanswerswhentheyexist.

DARPA/CGCmustnotdistributeuncertainty.WhenasoftwarebugoccursDARPA/CGCmustindependentlyroot‐causeandauthoritativelyfixpriortoresponding,ascollaboratingwithanycompetitorduringthedebuggingprocesswouldprovidethatcompetitorwithadditionalinsightand,asaresult,competitiveadvantage.

DARPA/CGCmustworktominimizeFAQrewriting.Competitorsdependon

accurateFAQentries;revokinganentryandrewritingitcancausecompetitorstobackupandchangedevelopmentcourse.Forthisreason,manyFAQentriescannotbereleaseduntiltheyarebackedbyrepeatedtesting.

DARPA/CGCmustobserveanapprovalworkflowconsistentwithpublicrelease

ofinformationbyagovernmentagency.Allpubliclypostedanswersflowthroughanapprovalprocessbeforerelease.

CGChasbeenconstructedasatechnologycompetitionthatisgloballyaccessible,fairtoallcompetitorparties,high‐integrity,andastransparentaspossiblewithintheseconstraints.Q157:Howlongareroundsandhowdotheywork?A157:EachroundofCFEplaywillbefollowedbyashort,unscoredbreak.Thedurationofthesebreakswillbeaminimumof30secondswithnomaximumduration.Eachroundofplaywilllastatleast240seconds.Duringabreak,nogameactionswillbeinitiatedbyCFEinfrastructure.ACRSformedinputwhosesubmissioncompletesduringanunscoredbreakwillbesubmittedtotherounddirectlyfollowingthebreak.

Page 15: Cyber Grand Challenge - MIT Lincoln Laboratory · Q157 Added Jan 27, 2016 Q158 Added Mar 16, 2016 Q159 – Q161 Added Mar 16, 2016 Q162 – Q163 Added Mar 25, 2016 Q164 Added May

Inresponsetofeedbackreceivedduringtheperiodofpubliccomment,DARPA/CGChasformulatedaroundstructuretoenableconsensusevaluation.ForanyChallengeSet,IDSrulesfilesandReplacementChallengeBinariessubmittedduringaroundnarescheduledasfollowsduringnormalplay*: n:CRSformedinputsubmitted.

n+1:DARPA/CGCinspectsCRS‐formedinputs.n+1:ServiceisdowntopollersandPoVs(roundscoreiszero).n+1:IDSrulefileandReplacementChallengeBinariesavailableto

opponentsviaconsensusevaluation.n+2:IDSrulefileandReplacementChallengeBinariesarefielded.

PoVmodulessubmittedduringaroundmarescheduledasfollows: m:CRSformedinputsubmitted. m+1:PoVisthrownasspecified.*normalplaydoesnotincludetheinitialandfinalroundsthataChallengeSetisfieldedduringCFE;CSintroductionandremovalwillbecoveredseparately.Q156:OurCRShadtroublewithnonces,challenge‐responseandconcurrencyinCQEChallengeSets;weareworriedthatthiswillbeanobstacleforourCRSinCFE.WillCFEChallengeSetsincludethesefeatures?A156:Yes.ChallengeSetauthorshavebeeninstructedtoprovideincreasedchallengesetdifficultyfortheCFEphaseofthecompetition.TheguidelinesprovidedtotheChallengeSetauthorscanbefoundathttps://github.com/CyberGrandChallenge/cgc‐release‐documentation/blob/master/walk‐throughs/submitting‐a‐cb.mdQ155:WillaccesstotheDARPAcloudnodesbemaintained24x7inthemonthspriortoCFE?A155:No.DARPA/CGCwillschedulemaintenancewindowswheneverpossible.DARPA/CGCobservesnofixedmappingbetweenteamsandhardwareandreservestheabilitytore‐provisionwithintheDARPAcloudatanytime.

Page 16: Cyber Grand Challenge - MIT Lincoln Laboratory · Q157 Added Jan 27, 2016 Q158 Added Mar 16, 2016 Q159 – Q161 Added Mar 16, 2016 Q162 – Q163 Added Mar 25, 2016 Q164 Added May

Q154:WhatshouldIdoifmyDARPAcloudnodesfailduringdevelopmentpriortoCFE?A154:DARPAcloudissuescanbereportedtocybergrandchallenge@darpa.mil.InthecaseofasoftwarefaultDARPA/CGCmayautomaticallyre‐provisionthenode.InthecaseofahardwarefaultDARPAmaypursueremedy,replacement,and/orre‐provisioning.Q153:AretheDARPAcloudnodedisksbackedup?A153:No.CompetitorsareencouragedtofollowcloudcomputingbestpracticesandremainpreparedtodeploytheirCRSonafreshlyre‐provisionedcloud.AutomatedinstallationanddevopsbestpracticesmayassistintheeventofCFEdisasterrecovery.Q152:WhatguaranteesdoesDARPA/CGCmakeagainstHPCnodefailure?A152:DARPA/CGCcannoteliminatethepossibilityofnodefailure.DuringdevelopmentpriortoCFE,replacementnodeswillbeinstalledpercommercialwarrantyaspossible.DuringCFE,failednodeswillnotbereplaced.CRSdesignshouldtakeplaceinformedbythesefacts.Q151:WhatruntimelimitsareplacedonNetworkAppliancerules?A151:EachNetworkApplianceinstancewillutilizeonerulesfile.Foreachinstance,theNetworkAppliancemaintainstwoanalysisringbuffers,oneforincomingbytesreadfromeachsideoftheconnection.Theseringbuffersarecappedat100x1024bytes.Q150:WhatruntimelimitsareplacedonCFEPoVmodules?A150:ThecompetitionframeworkenforcesahardlimitonPoVmodulesof15secondsofwalltimeandlimitsphysicalmemoryusageto64x10242bytesperPoVinstantiation(e.g.aPoV“throw”).Q149:WhatruntimelimitsareplacedonCFEChallengeBinaries?A149:ThecompetitionframeworkenforcesahardlimitonChallengeBinariesof15secondsofwalltimeand3x10243bytesofvirtualaddressspace.Seealso,A58andA119.Q148:WhataretheCFEuploadlimits?A148:TheCFEcompetitionframeworkTeamInterfaceAPI[1]willacceptinputnolargerthan10x10242bytesforPoVmodules,10x10242bytesforNetworkAppliancerulesfilesand50x10242bytesforReplacementChallengeBinaries.

Page 17: Cyber Grand Challenge - MIT Lincoln Laboratory · Q157 Added Jan 27, 2016 Q158 Added Mar 16, 2016 Q159 – Q161 Added Mar 16, 2016 Q162 – Q163 Added Mar 25, 2016 Q164 Added May

Seealso:Q58,Q77,Q119,andQ132.[1]https://github.com/CyberGrandChallenge/cgc‐release‐documentation/blob/master/ti‐api‐spec.txtQ147:WillthevalueofMAX_THROWSintheTeamInterfaceAPIspecificationbehardcodedforCFE?A147:Yes:10.Seealso[1].[1]https://github.com/CyberGrandChallenge/cgc‐release‐documentation/blob/master/ti‐api‐spec.txtQ146:DoestheCFEframeworkfacilitatesavingstatebetweenPoVthrows?A146:No.Q145:HowwilltheNetworkApplianceimpactthePerformancecomponentoftheAvailabilityscore?A145:Networkappliancerulesetexecutiontimewillbemeasuredrelativetotheexecutiontimeofareferenceemptyruleset;thesemeasuredincreasesinexecutiontimewillbecombinedwithexistingCBexecutiontimemeasurementstodeterminetheexecutiontimeportionofthePerformancescore.SeealsoA58.Q144:InA58,RetainedFunctionalityisspecifiedintermsofnetworktestcases;arethesenetworktestcases“servicepolls”?A144:Yes.Q143:Canaservicepollfailbytimingout?A143:Yes.Q142:Whyisafacilityprovidedtoquerythelocationofthetype2PoVmemoryrangeandPoVlengthincfe‐pov‐markup‐spec.txt?A142:TheDECREEAPIallowscompetitionshostedusingDECREEtospecifythestartaddressofthetype2PoVmemorylocationandtherequiredsizeofaPoVthroughrequeststothecompetitionframework.ThisflexibilitywillassistinportingtheDECREEframeworktofuturearchitectures.ThisfacilityallowsresearchersutilizingDECREEflexibilityindesigningcurrentandfuturecompetitionsandresearch.DuringtheCyberGrandChallengeFinalEvent,theType2PoVmemoryrangeandPoVlengthwillbeconsistentwiththe“submitting‐a‐cb.md”walkthrough.SeealsoA29.

Page 18: Cyber Grand Challenge - MIT Lincoln Laboratory · Q157 Added Jan 27, 2016 Q158 Added Mar 16, 2016 Q159 – Q161 Added Mar 16, 2016 Q162 – Q163 Added Mar 25, 2016 Q164 Added May

Q141:WillteamshaveaccesstoascoringoracleduringCFEthatcanbeutilizedpriortofieldingRBs?A141:PerA4,A17,andA137:No.Q140:Isitpossibleforafieldedreplacementbinarytoreceiveanetworkinputthathasnottransitedthenetworkappliance?A140:NoQ139:Canfieldingapoorlyperformingnetworkappliancerulenegativelyaffectavailabilityscore?A139:Yes.Q138:DuringCQE,asecondaryscoringdocumentwasreleasedwhichprovideddegradationcurvesandformulae.WillasimilardocumentbereleasedforCFE?A138:AdditionalinformationaboutCFEscoringwillbereleasedthroughtheFAQ.Q137:Whatadditionalinformationabouttheimpactofnetworkappliancerules,executiontime,etc.ontheAvailabilityscorewillbereleased?A137:DARPA/CGCintendstoallowcompetitorstoanswerthesequestionsindependentlybyprovidingaccesstoacompetitionframeworkoracleviatheDARPAcloud.Atthetimeofthiswriting,constructionofthecompetitionframeworkcontinues.ThisaccesswillbegrantedpriortoCFETrialsandrevokedpriortoCFE.Q136:WhichChallengeSetswereexcludedfromcomputingCQErankinginA130?A136:Theexclusionproofdecryptsto:

Common-name,CSID YAN01_00007,fd0e1101 YAN01_00009,c9967603 YAN01_00010,31502e01 YAN01_00011,77f39101 YAN01_00012,9d97ef01

Page 19: Cyber Grand Challenge - MIT Lincoln Laboratory · Q157 Added Jan 27, 2016 Q158 Added Mar 16, 2016 Q159 – Q161 Added Mar 16, 2016 Q162 – Q163 Added Mar 25, 2016 Q164 Added May

Thedecryptioncanbeverifiedbytakingtheencryptedbase64blobfromA130anddecryptingitwiththekeyinthebelowscript:

PASS="The best way to predict the future is to invent it.qR59RBQiIbMnuFwJFJpNUHMusu6RjyrGxvDI17v9egiGRloXME9bGjwBjF32bc2qdNqLVwTz3CNuogh1XcWapd5IGQ1VEE4B9/Mt9TAKN0S1P97XZj4WFCH+KAhMzTR3AkYM/mTbtzXfhDzSOEgl7G+7T1IE0uYaYYYuEwm4uL4=" A130_PROOF="U2FsdGVkX1/21aa4u0ZhcTCAyGr8oX7aCimSvKfGG3HFvjy5sCWi7Dh0dY9mb2nizlldmCCeB7g2RbUv3R0PkvFVp+EeTeWj8MpTiBCP0cVS3/uXy2rWIZtkFpOYBXIqYdybwpawzlRshJ8BeA2/WCWBeupEhCWFjJT7XigmLyEZqIujQ/rsWJTLOTfWFWC0" echo "$A130_PROOF" | \ base64 --decode | \ openssl aes-256-cbc -d -pass "pass:$PASS"

Q135:DuringareviewoftheCQEcorpus,ourteamdiscoveredthatsomereplacementbinariesreleasedbyourcompetitorsdisruptthefunctionofourprogramanalysissoftware.We’reconcernedaboutthiseffectduringamoreautonomouscompetition(CFE).TheCGCEventParticipationAgreementexplicitlyallowsthiseffectfromCRS‐formedinputs;willDARPA/CGCprovideadditionalprotectionsagainstthiseffect?A135:No.Q134:Therequirementforfullautonomybringswithitaslewofengineeringdecisions,includingwatchdogtimers,redundantcomponents,parallelexecution,andhighassuranceprogramming.DoesDARPA/CGChavearecommendedpathtowardsfullautonomy?A134:No.SeealsoA4andA17.Q133:WhichCRS‐formedinputsnamedintheCGCEventParticipationAgreementcanbeidentifiedinthecurrentDECREEarchitecture?A133:Atthepresenttime:

‐ Replacementbinaries‐ Dataproducedbyreplacementbinaries‐ Networkappliancerules‐ Trafficmodifiedbynetworkappliancerules‐ Proofsofvulnerabilities‐ Dataproducedbyproofsofvulnerabilities

Page 20: Cyber Grand Challenge - MIT Lincoln Laboratory · Q157 Added Jan 27, 2016 Q158 Added Mar 16, 2016 Q159 – Q161 Added Mar 16, 2016 Q162 – Q163 Added Mar 25, 2016 Q164 Added May

Q132:IsaCRSpermittedtosubmitpovmlfilesasaProofofVulnerabilitytothecompetitionframeworkduringCFE?A132:No;seealsosection2.2oftheCRSTeamInterfaceAPIspec[1]thatdescribeshowtosubmitaPoVduringCFE.[1]https://github.com/CyberGrandChallenge/cgc‐release‐documentation/blob/master/ti‐api‐spec.txtQ131:WhatisthedurationandautonomypolicyoftheCFE?A131:InA58,DARPA/CGCindicatedthatCFEpointswouldaccrue"perChallengeBinaryperround".DARPA/CGCisopeningaperiodofpubliccommentthatwillcloseonSeptember25th,2015.ThisperiodofpubliccommentencouragesfeedbackontheappropriatetimingoftheCFEconsistentwiththegoalsofCGC:theabilitytodemonstrate,iffeasible,real‐timeproactiveandreactivemitigationofnovelsecurityfaultsonmachine‐scaletimelinesmeasuredinseconds.Feedbacktopicsinclude: 1>DurationofCFERounds,inseconds 2>TotalnumberofRoundsinCFE

Note:Feedbackfor1>and2>shouldbepairedtoindicateatotaleventduration.

3>Releaseand/orexpirationscheduleforChallengeSets(CS)inCFE.

4>PolicyonHumanInterference(handsondebugging).ApartiallistoftheorizedpoliciesonCFEdebuggingfollows:

4a>TeamscouldbeallowedtodebugtheirCRSatanytimeduringCFE.Duringdebugging,aCRSwouldbedisconnectedfromtheframeworkandunabletoaccumulatepointsduringdebugging.Afterdebuggingcompleted,anyChallengeSet(CS)previouslyrevealedtothedebuggingteamwouldbefruitofthepoisonedtreeandunavailableforscoringtothatteamfortheremainingdurationoftheevent.Forexample,ifteamalphachosetodebuginthefirstsecondwhenasingleCSwasfielded,thenallpointsforthatCSandtheabilitytofieldthatCSwouldbeunavailabletoalphaforthedurationoftheevent;ateambetathatchosetodebuginthefinalminutewitheveryCSfieldedwouldbeunabletoscoreanyfurtherpointsafterdebugging.

4b>Anumberofscheduleddebuggingbreakscouldbeannouncedpriortotheevent,dividingCFEintoStages.ForeachStage,allpreviousCS'swouldbewipedandfreshCS'swouldbeprovided.

Page 21: Cyber Grand Challenge - MIT Lincoln Laboratory · Q157 Added Jan 27, 2016 Q158 Added Mar 16, 2016 Q159 – Q161 Added Mar 16, 2016 Q162 – Q163 Added Mar 25, 2016 Q164 Added May

4c>CFEcouldbetotallyautonomous;catastrophicerrorscouldnotbecorrected.

4d>CompetitorswouldbepermittedtofullyinteractwithaCRSforaninitialperiodoftimeattheonsetofCFEtoensurethecorrectoperationoftheCRS.AfterthistimetheCRSwouldcontinueautonomouslyasin4c.

Competitorsareencouragedtoprovidecommentstocybergrandchallenge@darpa.milbySeptember25th,2015.

Q130:WhichChallengeSetswillbeexcludedfromcomputingCQEranking?

A130:SeeFAQA117.TheproofofexcludedCSselectionis:

U2FsdGVkX1/21aa4u0ZhcTCAyGr8oX7aCimSvKfGG3HFvjy5sCWi7Dh0dY9mb2nizlldmCCeB7g2RbUv3R0PkvFVp+EeTeWj8MpTiBCP0cVS3/uXy2rWIZtkFpOYBXIqYdybwpawzlRshJ8BeA2/WCWBeupEhCWFjJT7XigmLyEZqIujQ/rsWJTLOTfWFWC0Q129:AnemptywriteinaPoVXMLpassespoll‐validatebutcausesanexceptionincb‐replay;willthisbefixed?A129:AfixwillbereleasedinafutureversionofDECREE.DuringCQE,competitorsareencouragedtorefertoFAQentryA111withrespecttodebuggingtheirCRSduringtheevent.PriortoCQE,competitorsareencouragedtoensuretheirCRSdoesnotgenerateaPoVwithanemptywritestatementsimilartothebelow: <write><data></data></write>Q128:WhatistheofficialTwitterfeedfortheCGCQualificationEvent?A128:https://twitter.com/DARPA_CGC_CQEQ127:HowdoIconfirmthatIamrunningthelatestversionofDECREE?A127:InthelastDECREEreleaseemail,atest[1]wasissuedtocheckthevagrantversion.Ifyourcurrentsystemrespondswiththeoutputcitedbelow[1][2],thereisnoneedtoupdate.Ifyoursystemrespondswithdifferentoutput,DARPA/CGCadvisesthatyouupdate.[1]TheexpectedoutputofvagrantsshofanupgradedVMwillbe:Linuxcgc‐linux‐packer3.13.2‐cgc#1SMPMonApr1318:33:57UTC2015i686[2]vagrant@cgc‐linux‐packer$cat/etc/decree_versioncqe_development‐vm‐61

Page 22: Cyber Grand Challenge - MIT Lincoln Laboratory · Q157 Added Jan 27, 2016 Q158 Added Mar 16, 2016 Q159 – Q161 Added Mar 16, 2016 Q162 – Q163 Added Mar 25, 2016 Q164 Added May

Q126:WhereistheCQEchallengebundle?A126:TheCyberGrandChallengeCQEbundleisavailable!PleasedownloadandverifyatyourearliestpossibleconveniencepriortoCQE.Downloadinstructionsareavailable[1].PleasecontinuetomonitortheEventFAQ[2]andtheCQETwitterFeed[3]throughoutCQE.PackageName:cgc_qualifier_event.ar.gz.encPackageMD5:0ec6a5708aea0b0d7c5a9f1c37924423PackageSHA256:8274d4086ee1039e7901b3cc7221fa89e2f923dc5ad39924965f2ff5b110449d[1]https://github.com/CyberGrandChallenge/cgc‐release‐documentation/blob/master/walk‐throughs/cgc‐qualifier‐event‐api.md[2]https://github.com/CyberGrandChallenge/Event‐FAQ/blob/master/event_faq.md[3]https://twitter.com/darpa_cgc_cqeQ125:DoallhardwarecomponentsoftheCRSneedtobephysicallypresentataCQEsitevisit?A125:No.Q124:WereallsubmissionstoScoredEvent2scored?A124:No;somesubmissionstoScoredEvent2werenotscored.Allfinalsubmissionswerescoredandranked.PerA67abesteffortwasmadetocollectandscoremultiplesubmissions.DuringSE‐2,newsubmissionswerecollectedevery10‐20seconds;inthecaseofmultiplesubmissionsfromthesameteamtothesamechallengesetwithinthattimewindow,thelatestsubmissionwasscoredandtherestwerenotscored.Fiveteamswereaffected.Q123:Ismem_use(CB)themaximumRSSsizefortheCBprocess?A123:No;mem_use(CB)iscalculatedcongruentwiththeformulapresentedintheCBscoringwalkthrough[1].[1]https://github.com/CyberGrandChallenge/cgc‐release‐documentation/blob/master/walk‐throughs/scoring‐cbs.mdQ122:Isexec_time(CB)theuserCPUtime+systemCPUtimefortheCBprocess?Justusertime?Orsomethingelse,e.g.wallclock?Ifwall‐clock,whatinterval?A122:exec_time(CB)doesnotincludesystemtimeperA113;exec_timeisusertimemeasuredwithasensorasdescribedinA113.

Page 23: Cyber Grand Challenge - MIT Lincoln Laboratory · Q157 Added Jan 27, 2016 Q158 Added Mar 16, 2016 Q159 – Q161 Added Mar 16, 2016 Q162 – Q163 Added Mar 25, 2016 Q164 Added May

Q121:Aremem_use(CB)andexec_time(CB)computedfromthevaluesweseeinru_maxrss,ru_minfltand(ru_utime+ru_stime),respectively,intherusagestructurereturnedbyawait3/wait4systemcallfromaparent?A121:Therusagedatastructurereturnedbywait3/wait4containsru_maxrssandru_minfltthatareusedtocomputemem_use(CB),asdescribedinCBscoringwalkthrough[1].DARPA/CGCdoesnotderivetimingmeasurementsfromtherusagedatastructureperA113.[1]https://github.com/CyberGrandChallenge/cgc‐release‐documentation/blob/master/walk‐throughs/scoring‐cbs.mdQ120:HowarethemeasurementsforalltheservicepollsagainstaCBaggregatedintoasinglemem_use(CB)orexec_time(CB)component?Ismem_use(CB)thehighestmaximummemoryusageacrossanyservicepoll/input,orisitanaverage?Likewise,istheexec_time(CB)theaverageacrossservicepolls?A120:ThesecalculationsareperformedinamannercongruentwiththescoringCBswalkthrough[1].[1]https://github.com/CyberGrandChallenge/cgc‐release‐documentation/blob/master/walk‐throughs/scoring‐cbs.mdQ119:FilesizeandmemoryusagerestrictionsimposedonReplacementBinariesappearverysevereandpotentiallynotrepresentativeofreal‐worldconstraints;inthecaseofmemoryusagerestrictions,per‐pagegranularityimposesaheavyoverheadburdenduetorelativelysmallsizeofChallengeBinaries.Cantheselimitsbeadjustedandmemorymeasurementgranularityincreased?A119:CyberGrandChallengedoesnotimposeperformancepenaltiesbymeasuringtheincreaseinperformancebetweenacompetitorsubmittedRBandtheoriginalCB.Performancepenaltiesareassessedbasedonthedifferencebetweenasuccessfullydefended(patched)CBcreatedbyahumanexpertandacompetitorsubmittedRB[1].Thusadefenseisproventoexistandcompetitordefensesaremeasuredagainstknownsuccessfuldefenses.Fromitsinception,theCGChassettheambitiousgoalofmeasuringtheskillofautomatedsystemsagainsttheabilitiesofhumanexperts.Whilesomedefensivecompetitionshaveimposedabsoluteperformancelimitationsof5%orless[2],CGCimposedagracefuldegradationcurveandopenedthiscurvetopubliccommentinearly2014.ToprovidecontinuitytoallcompetitorsthismechanismwillnotbemodifiedaftertheperiodofpubliccommentperA100.Onthetopicofgranularityofmemorymeasurement,theCGCadoptedtheuseofpages,aMMUconstruct;traditionallyhumanexpertshaveskillfullyoptimizedtheuseofpages.

Page 24: Cyber Grand Challenge - MIT Lincoln Laboratory · Q157 Added Jan 27, 2016 Q158 Added Mar 16, 2016 Q159 – Q161 Added Mar 16, 2016 Q162 – Q163 Added Mar 25, 2016 Q164 Added May

[1]https://github.com/CyberGrandChallenge/cgc‐release‐documentation/blob/master/CQE%20Scoring.pdf[2]http://www.microsoft.com/security/bluehatprize/rules.aspxQ118:WhenhandlingmultipleCBsinaCS,isacrashinanyoralloftheCBsconsideredproofofavulnerability?A118:Yes,acrashinanyorallCBsinaCSthatmeetsthecriteriadescribedinA28isconsideredtohaveprovenavulnerability.Q117:WilleveryChallengeSetintheCQECorpusbescoredaspartoftheCQEevent?A117:No.Therewillbeover100ChallengeSetsintheCQECorpus;asmallnumberofthesearepurelydiagnosticandwillbeexcludedfromcomputingCQEranking.ForeachoftheseexcludedCSs:‐ PoVsagainsttheseCSswillnotbepartofCQEConsensusEvaluation‐ CSScoreswillnotbeusedforanypartofCQEcompetitorranking‐ CSScoreswillnotbeconsideredforanypartofsection3.1.3oftheCGCRulesToprovethatexcludedCSswerechoseninadvance,DARPA/CGCwillprovideacryptographicallyverifiableproofofthisselectionpriortoCQE.AfterCQE,aselectiondocumentmatchingthisproofwillbereleased.CompetitorsareadvisedtofieldaCRSthatattemptstosolveeveryCSinCQE.Q116:WhatisthemaximumsizeoftheuncompressedCQEChallengeBundle?A116:Sizewillnotexceed5000x10242bytes.Q115:WhyweretherenovulnerabilitiesthatlackedareferencePoVintherankingofScoredEvent2?A115:DARPA/CGChasrefineditsproceduresfollowingScoredEvent1.NotethatduetothepotentialconflictofinterestinvolvedinmakingCBchangeswhilesimultaneouslyhandlingde‐anonymizedscoringmaterialandPoVs,theDARPA/CGCscoringteamcannotdetermineflawuniquenessorformulateCBrepairsperAttack1inA110.SeealsoA9andA97.Q114:Thepublicspecificationforcb‐replay[1]indicatesthatthereadtagacceptsatimeouttag.Thisdoesnotappeartobeimplemented.Willthecb‐replayimplementationbeadjustedtomatchthespecbeforeCQE?A114:No‐‐themismatchwillbemaintainedthroughCQEforcontinuitypurposes.

Page 25: Cyber Grand Challenge - MIT Lincoln Laboratory · Q157 Added Jan 27, 2016 Q158 Added Mar 16, 2016 Q159 – Q161 Added Mar 16, 2016 Q162 – Q163 Added Mar 25, 2016 Q164 Added May

[1]https://github.com/CyberGrandChallenge/cgc‐release‐documentation/blob/master/replay.dtdQ113:FollowingupA99intheFAQ,isadditionalinformationavailableaboutthecorrelationbetweenoperatingsystemtimingestimatesandDARPA/CGCsensortimingfollowingSE2?A113:TheDARPA/CGCteam'stimingsensorwasdesignedtomeasureprocessexecutiontimewithunprecedentedaccuracyandprecision.Thissensorstopsmeasuringtimeimmediatelyduringthesystemcalltransitionfromusertokernel,andresumesmeasuringtimeimmediatelyduringthetransitionfromkerneltouser.Unfortunately,thetaskclockprovidedbyLinuxisnotasrigidlyaccurateduringthistransitionandcanmeasuresomekernelexecutionasuserexecution,causingdivergencebetweentheDARPA/CGCsensorandestimatesbasedonthesystemclock(usingtheimprovedestimationfacilityinDECREEdescribedinA99).Thisdivergenceincreasesbasedonthenumberandtypeofsystemcallsmadeduringexecution.Therandom()call,inparticular,isamajorsourceofdivergence.ThegraphsprovidedshowthecorrelationresultsoftheDARPA/CGCsensorandthesystemclock,withdivergenceincreasingwhenlargenumbersofsystemcallsaremade.DARPA/CGCexpectsthatformostChallengeSetexecution(unburdenedbyapreponderanceofsystemcalls)thecorrelationbetweenthetaskclockestimateandthehighaccuracyDARPA/CGCsensorshouldbehigh.GraphsofthiscorrelationoversamplesfromtheservicepollsforSE‐2ChallengeSetsareprovidedbelow.PleasenotethedivergentgraphforYAN01_00006,auniqueCBwhoseexecutionispredominantlysystemcalls.Thefollowingplotsshowlinearcorrelation;anrvalue[1]of1.00denotesmaximalpositivelinearcorrelation(optimal).Thex‐axisisgeneratedbytheoutputoftheDARPA/CGCsensor;they‐axisisgeneratedbytheOS‐basedestimatesissuedbyDECREEinA99.Eachdotinthescattergramisasingleservicepoll.[1]http://en.wikipedia.org/wiki/Pearson_product‐moment_correlation_coefficient

Page 26: Cyber Grand Challenge - MIT Lincoln Laboratory · Q157 Added Jan 27, 2016 Q158 Added Mar 16, 2016 Q159 – Q161 Added Mar 16, 2016 Q162 – Q163 Added Mar 25, 2016 Q164 Added May
Page 27: Cyber Grand Challenge - MIT Lincoln Laboratory · Q157 Added Jan 27, 2016 Q158 Added Mar 16, 2016 Q159 – Q161 Added Mar 16, 2016 Q162 – Q163 Added Mar 25, 2016 Q164 Added May
Page 28: Cyber Grand Challenge - MIT Lincoln Laboratory · Q157 Added Jan 27, 2016 Q158 Added Mar 16, 2016 Q159 – Q161 Added Mar 16, 2016 Q162 – Q163 Added Mar 25, 2016 Q164 Added May
Page 29: Cyber Grand Challenge - MIT Lincoln Laboratory · Q157 Added Jan 27, 2016 Q158 Added Mar 16, 2016 Q159 – Q161 Added Mar 16, 2016 Q162 – Q163 Added Mar 25, 2016 Q164 Added May
Page 30: Cyber Grand Challenge - MIT Lincoln Laboratory · Q157 Added Jan 27, 2016 Q158 Added Mar 16, 2016 Q159 – Q161 Added Mar 16, 2016 Q162 – Q163 Added Mar 25, 2016 Q164 Added May
Page 31: Cyber Grand Challenge - MIT Lincoln Laboratory · Q157 Added Jan 27, 2016 Q158 Added Mar 16, 2016 Q159 – Q161 Added Mar 16, 2016 Q162 – Q163 Added Mar 25, 2016 Q164 Added May
Page 32: Cyber Grand Challenge - MIT Lincoln Laboratory · Q157 Added Jan 27, 2016 Q158 Added Mar 16, 2016 Q159 – Q161 Added Mar 16, 2016 Q162 – Q163 Added Mar 25, 2016 Q164 Added May
Page 33: Cyber Grand Challenge - MIT Lincoln Laboratory · Q157 Added Jan 27, 2016 Q158 Added Mar 16, 2016 Q159 – Q161 Added Mar 16, 2016 Q162 – Q163 Added Mar 25, 2016 Q164 Added May
Page 34: Cyber Grand Challenge - MIT Lincoln Laboratory · Q157 Added Jan 27, 2016 Q158 Added Mar 16, 2016 Q159 – Q161 Added Mar 16, 2016 Q162 – Q163 Added Mar 25, 2016 Q164 Added May
Page 35: Cyber Grand Challenge - MIT Lincoln Laboratory · Q157 Added Jan 27, 2016 Q158 Added Mar 16, 2016 Q159 – Q161 Added Mar 16, 2016 Q162 – Q163 Added Mar 25, 2016 Q164 Added May
Page 36: Cyber Grand Challenge - MIT Lincoln Laboratory · Q157 Added Jan 27, 2016 Q158 Added Mar 16, 2016 Q159 – Q161 Added Mar 16, 2016 Q162 – Q163 Added Mar 25, 2016 Q164 Added May
Page 37: Cyber Grand Challenge - MIT Lincoln Laboratory · Q157 Added Jan 27, 2016 Q158 Added Mar 16, 2016 Q159 – Q161 Added Mar 16, 2016 Q162 – Q163 Added Mar 25, 2016 Q164 Added May

Q112:WhatisthemethodologytofindmysubmissionsinthereleasedSE‐2scoringdata?A112:RevisedinFAQv16:ReplacementBinaryscoresandPoVscoreshavebeenreleasedasseparate.csvfilesatcgc.darpa.mil;perthebundledinstructionstheSHA256ofeachRBandPoVisusedtocalculatetheRBandPoVidentifiersusedinthese.csvfiles.Pleasenotethattheseidentifiersarenotprovidedbythesubmission

Page 38: Cyber Grand Challenge - MIT Lincoln Laboratory · Q157 Added Jan 27, 2016 Q158 Added Mar 16, 2016 Q159 – Q161 Added Mar 16, 2016 Q162 – Q163 Added Mar 25, 2016 Q164 Added May

verificationscript[1];theseidentifiersshouldbecalculatedseparatelyusingthesolutionpackageverificationscript[2].[1]https://github.com/CyberGrandChallenge/cgc‐release‐documentation/blob/master/scripts/cqe_submission_verification.py[2]https://github.com/CyberGrandChallenge/cgc‐releasedocumentation/blob/master/scripts/cqe_verify_solution_package.pyQ111:DuringtheCQE,whatkindofhumaninvolvementisallowed?AlthoughwehavereadSection4(FullAutomationRequirement)oftheCGCRules,itremainsunclearwhat"humanassistanceincyberreasoningprocesses"means,exactly.Onascaleofwhatweimaginetobemostreasonabletotheleastreasonable,hereareafewexamplesaboutwhichwearecurious(ofcourse,specificexamplessuchassplittinguppcapsarethingswewillhandleaswehavethoughtofthem,thesearejustexampleswehaveencounteredtohelpmotivateourquestions):*powerorhardwarefailureandwemanuallyrebootservers;*kernelpanicorOOMandwemanuallyrebootservers;*launchingVMs/scripts(suchasforfuzzingorsubmittingbinaries)byhandatthestartoftheCQE;*largepcapcausesustomanuallysplititupbeforefeedingittoourCRS;*trivialbug(suchasmisnamedpythonvariable)inourCRS,sowefixthebugandrestarttheCRS;*binaryusesbehaviorwedidn'taccountfor(suchaswritingtoSTDIN)andwepatchoursystemtoaccountforthis;*buginourCRScausesustomissthefaultingaddressforacrash,sowefixthebugandrerunthatportionoftheCRS;*crashisnotreproducible,sowemanuallyfixitup*CRScannotfindcrash,sowemanuallydoitThefirstthreewecurrentlydeemas"inscope".Thelasttwowecurrentlydeem"outofscope".Thefourinthemiddleweareveryunsureof.A111:DARPA/CGCbelievesthattheissueofhumaninterferenceincyberreasoningprocessesisclear‐cutandcaneasilybedifferentiatedbyanyofourexpertcompetitorsinthisfield.TheDARPACompetitionAgreementgovernshumaninterferenceincyberreasoningfunctions:anyfunctionthatinteractswithChallengeSetsandmakesdecisionsabouthowtoformulateReplacementBinariesandPoV’s.Achangetoareasoningfunctionwillchangetheoutcomeofasystem’sperformanceinanyAreaofExcellence[1].WhenahumanperformsacyberreasoningprocessonthecontentsoftheChallengeBundle,theresultingknowledgeisfruitofthepoisonoustree.Aftergainingsuchknowledgefromanysource,ahumanmaynotmodifytheCRS.

Page 39: Cyber Grand Challenge - MIT Lincoln Laboratory · Q157 Added Jan 27, 2016 Q158 Added Mar 16, 2016 Q159 – Q161 Added Mar 16, 2016 Q162 – Q163 Added Mar 25, 2016 Q164 Added May

Below,thistestisappliedtothelistofexamplesprovidedinthequestion:

1> Action:powerorhardwarefailureandwemanuallyrebootserverso Humandidnotpossesstaintedknowledge.Nointegrityissue.

2> Action:KernelpanicorOOMandwemanuallyrebootservers

o Humandidnotpossesstaintedknowledge.Nointegrityissue.

3> Action:LaunchingVMs/scripts(suchasforfuzzingorsubmittingbinaries)byhandatthestartoftheCQE

o Humandidnotpossesstaintedknowledge.Nointegrityissue.

4> Action:Manuallyun‐packagingChallengeBundleandfeedingcontentsintoCRS.

o Humandidnotpossesstaintedknowledge.Nointegrityissue.

5> Action:ManuallypackagingresultsfromtheCRSforsubmissiontoDARPA.o Humandidnotpossesstaintedknowledge.Nointegrityissue.

6> Action:LargepcapcausesustomanuallysplititupbeforefeedingittoourCRS

o Humandidnotpossesstaintedknowledge.Nointegrityissue.

7> Action:Trivialbug(suchasmisnamedpythonvariable)inourCRS,sowefixthebugandrestarttheCRS

o ThisexampledescribesanunanticipatedsystembugthatoccurredduetoautomatedprocessingoftheChallengeBundle: IftheCRSsimplycrashesoninputandthebugcanbefixed

throughCRScodeinspection,notaintedknowledgewasinvolved;nointegrityviolation.

IfahumanexaminestheChallengeBundleandperformsacyberreasoningprocesstoarriveattheknowledgeusedtopatchthebug,anintegrityviolationhasoccurred.TheknowledgeusedtopatchthebugderivedfromhumanexaminationoftheChallengeBundle,andisthereforefruitofthepoisonoustree.

Forexample,ifaCRScontainsaflawthatisdesignedtocrashonthefirstCBprocessed,itisnotanintegrityviolationtousethecrashdumptoidentifythisflawaslongasinspectionofthecrashdumpdoesnotyieldtaintedcyberreasoningknowledgegainedfromtheChallengeBundle.HoweverwereateamtoexaminetheChallengeBundleinanattemptto“optimallyfix”thisflaw,perhapsbyinsertingnewPoVseedcases,anintegrityviolationwouldhaveoccurred.

Page 40: Cyber Grand Challenge - MIT Lincoln Laboratory · Q157 Added Jan 27, 2016 Q158 Added Mar 16, 2016 Q159 – Q161 Added Mar 16, 2016 Q162 – Q163 Added Mar 25, 2016 Q164 Added May

8> Action:Binaryusesbehaviorwedidn'taccountfor(suchaswritingtoSTDIN)andwepatchoursystemtoaccountforthis

o ChallengeSetswhosespecificationisunanticipatedbyteamsmaycausefaultsandshouldbehandledpercase7,above.

9> Action:BuginourCRScausesustomissthefaultingaddressforacrash,sowefixthebugandrerunthatportionoftheCRS

o Inthishypotheticalscenario:

a. Abugexistsinthedynamicanalysis/concreteinputportionofaCRS.

b. ThisbugcausesafalsenegativeinwhichtheCRSdoesnotidentifyacrash.

c. Ahumanusesadebuggertoidentifythefalsenegative.d. Humanusesbugreportfrom9.ctofixtheCRSbugandrestarts

CRS.Anintegrityviolationoccurredinstep9.cwhenahumanperformedacyberreasoningprocessontheChallengeBundle,andthenusedfruitofthispoisonoustreetoidentifyaflawandformulateafixintheCRS.

10> Action:Crashisnotreproducible,sowemanuallyfixitupo Inthishypotheticalscenario:

a. AhumanexaminestheoutputofaCRSinadebuggerortest

harnessbytestingtheCRS‐generatedPoVwiththeCBprovidedintheChallengeBundle.

b. Ahumandetermines,throughiterativetestingandanalysisusingthedatacomponentsin10.a,ahigherreliabilityformulationforthePoV.

c. Usingtheknowledgefrom10.b,ahumanreprogramstheCRStoemitanewlyformulatedPoV.

DuetotheuseoftheCBprovidedbytheChallengeBundlein10.a,thechangemadein10.cisfruitofthepoisonoustree;thisisanintegrityviolation.

11> Action:CRScannotfindcrash,sowemanuallydoit

o HumanPoVgenerationrequirescyberreasoning;thecreationofPoVsisanAreaofExcellence[1];integrityviolation.

[1]AreasofExcellencearedescribedintheCGCRules,Section1.3

Page 41: Cyber Grand Challenge - MIT Lincoln Laboratory · Q157 Added Jan 27, 2016 Q158 Added Mar 16, 2016 Q159 – Q161 Added Mar 16, 2016 Q162 – Q163 Added Mar 25, 2016 Q164 Added May

Q110:SomeCBsinSE1containedvulnerabilitiesthatwerenotprovenbythereferencePoVset.ACRSthatpatchesaflawprovenbyareferencePoVcanconsistentlyincreasetheReferencecomponentofitsSecurityscore.Ontheotherhand,aCRSthatpatchesaflawnottestedbythereferencePoVsetcanonlyimprovetheConsensuscomponentofitsSecurityscore.Thisinconsistencyappearstoleadtoacompetitionthatmaynotconsistentlydistinguisheffectivepatching.Whycan'ttheDARPAscoringteamdirectlyaddCRS‐discoveredPoVstothelistofreferencePoVsforthepurposeofSecurityscoring?

A110:InordertomitigateAttack3,below.

Q109:WhydoesthescoringsystemdescribedinA58onlyallowforonePoVperCB?

A109:InordertomitigateAttack1andAttack2below.

Q108:Whywasn'taPoVweightedbythenumberofRBsitprovesvulnerable?

A108:InordertomitigateAttack2below.

Q107:Whywasn'tanRB'sSecurityscoreweightedbythetotalnumberofPoVsitdefendsagainst?

A107:InordertomitigateAttack1&Attack4below.

CQEThreatModeling

BelowisalistofmitigatedattacksagainsttheintegrityoftheCGCcompetition.DARPAhasmadeeveryefforttoincreasecompetitorconfidenceincompetitionresultsbymitigatingattacksagainstcompetitionintegrity.CQEscoringistheresultofextensivethreatmodeling.Asubsetofthisthreatmodelingexerciseisincludedinthelistofmitigatedattacksbelowthatareprovidedasareferencepointforanswersaboutspecificscoringdesigndecisions:

Attack1:PoVStuffing

ThisattackrequiresascoringsysteminwhichaCRSmaysubmitandreceivecreditformultiplePoVsthatevaluateasingleCB.Tomountthisattack,aCRSusesknowledgeofaflawtogenerateagreatnumberofPoVsthatproveasingleflawinaCB.Forexample,inthecaseofasimplebufferoverflow,aCRScouldsubmitthousandsofPoVsthatoverflowthesamebufferwithdifferentcontents.ThisattackcannotbeeasilydistinguishedfromaCRSthathasaccomplishedthemoredifficultfeatofgeneratingmultiplePoVsthatprovemultipleflaws.AttemptstodeterminewhenaPoVprovesa"uniqueflaw"requiressubjectivejudgmentsonbehalfoftheDARPA/CGCscoringteam;seeA9.

Page 42: Cyber Grand Challenge - MIT Lincoln Laboratory · Q157 Added Jan 27, 2016 Q158 Added Mar 16, 2016 Q159 – Q161 Added Mar 16, 2016 Q162 – Q163 Added Mar 25, 2016 Q164 Added May

Attack2:SockPuppetCrashDummies

Inthisattack,aroguecompetitor,inviolationoftheDARPAcompetitionagreement,standsupmultiple"sockpuppet"teams.Asetofthesesockpuppetteamsintentionallyintroduce"keyed"flawsintheirreplacementbinaries(RBs)thatrequireuniqueknowledge(a"key")toprove.InviolationoftheDARPAcompetitionagreement,thesockpuppetssharethisknowledgewiththeroguecompetitor.ThisroguecompetitorprovidesconsensusevaluationPoVsforthe"keyed"flaws,therebyimprovingitsEvaluationscorewithoutactuallyfindinglegitimatesoftwareflaws.Manysubtlevariationsofthissimpleattackexist.

Attack3:ReferenceCollusion

Inthisattack,aroguecompetitor,inviolationoftheDARPAcompetitionagreement,standsupmultiple"sockpuppet"teams.InviolationoftheDARPAcompetitionagreement,thesockpuppetscolludewiththeroguecompetitor,onlyintroducingPoVsthatproveflawsknowntobepatchedbytheroguecompetitor.ThisattackallowsSecurityscoringtobedominatedbyflawschosenbytheroguecompetitor.

Attack4:ConsensusCollusion

Inthisattack,aroguecompetitor,inviolationoftheDARPAcompetitionagreement,standsupmultiple"sockpuppet"teams.EachofthesesockpuppetsintroducesavalidsetofPoVscontaininga"key".InviolationoftheCGCRules,thesockpuppetscolludewiththeroguecompetitor,sharingthe"key"value.TheroguecompetitorisabletoartificiallyimproveitsSecurityscorewithoutactuallymitigatingaflawbysubmittingRBsthatblockanyinputcontainingthe"key”.Manysubtlevariationsofthissimpleattackexist.

Q106:WilladditionalChallengeBinaryspecificationsbereleasedthatexceedthespecificityoftheexistingdocumentation[1]?A106:No.CyberReasoningSystemsshouldreasonaboutsoftwarepertheCGCRules[2].[1]https://github.com/CyberGrandChallenge/cgc‐release‐documentation[2]https://cgc.darpa.mil/documents.aspxQ105:HowwillDARPA/CGCprovethatallPoVsscoredduringCQEcouldcauseunlimitedexecutionofarbitrarycode?A105:DARPA/CGCdoesnotassertorintendtoprovethis.Softwarevulnerabilitiescanleadtoavarietyofeffectsrangingfromdenialofservicetocodeexecution;DARPA/CGCwillacceptProofofsuchVulnerabilitiesinSE2andCQEperA28.

Page 43: Cyber Grand Challenge - MIT Lincoln Laboratory · Q157 Added Jan 27, 2016 Q158 Added Mar 16, 2016 Q159 – Q161 Added Mar 16, 2016 Q162 – Q163 Added Mar 25, 2016 Q164 Added May

Q104:We'reunsurethatourapproachwillworkontheDARPA/CGCscoringapparatus.Canyouguaranteethesuccessofourapproach?A104:No.DARPA/CGCrecommendsthatadventurousorriskyuseofhardwareorsoftwarefeaturesbeevaluatedthoroughlyduringScoredEvents.Commercialsecuritysoftwareisoftenrequiredtooperateinuncertaincomputingenvironmentsandreliesonhardwareinterrogation(suchasCPUID)andsoftwareinterrogation(suchasO/Sversionnumbers)toproceed.Q103:DidDARPA/CGCchangethewaymemoryusageisquantifiedormeasuredinDECREE?A103:No.Memoryusagemeasurementhasremainedconsistentwiththeformulareviewedduringpubliccomment,publishedinA58anddetailedintheDECREEdocumentation[1].Inanattempttoassistcompetitorswithestimatingthisquantity,DARPA/CGCbuiltandcontinuestoupdateawalkthroughtoassistcompetitorsinmakingtheirownmeasurements.OneLinuxquirkthatcouldcausemeasurementproblemsforcompetitorsisthatchildprocessesareassignedamaximumRSSvalueinheritedfromtheirparent.Thiscausesestimationproblemswhenalargeparentprocesslaunchesasmallchildprocess.Inordertoallowcompetitorstomakehigher‐fidelityestimates,thisquirkwasdisabledforCGCprocessesinDECREE[2].DARPA/CGCwillcontinuetoworktocreatethemosthighfidelityimplementationofthescoringmethodologypublishedinA58aspossibleandprovideassistancetocompetitorstomakeaccuratemeasurements.[1]https://github.com/CyberGrandChallenge/cgc‐release‐documentation/blob/master/CQE%20Scoring.pdf[2]https://github.com/CyberGrandChallenge/linux‐source‐3.13.2‐cgc/blob/master/fs/binfmt_cgc.c#L300Q102:InSE1,thereleasedreferencesecurityscoresarealleither0.0or1.0.Hasthisbeenconfirmed?A102:Yes.Q101:WhiletraditionallyCapturetheFlagcontestsandDARPAChallengeshavenevergivencompetitorsaccesstothescoringsystemandassociatedsensorspriortocompetition,couldCyberGrandChallengebecomethefirsttoallowopensourceaccesstothescoringsystem?A101:DARPA/CGCwillnotallowcompetitoraccesstothescoringmeasurementcodeanditsassociatedsensorsduetothecurrentlackoftechnologycapableofmakingstrongintegritypromisesaboutexecutablecodesubjecttoadversarialintrospection.DARPA/CGCwillcontinuetouseanopen,competitor‐reviewedscoringalgorithmdescribedinthisFAQ(A58/A59),allowperiodiccompetitor

Page 44: Cyber Grand Challenge - MIT Lincoln Laboratory · Q157 Added Jan 27, 2016 Q158 Added Mar 16, 2016 Q159 – Q161 Added Mar 16, 2016 Q162 – Q163 Added Mar 25, 2016 Q164 Added May

accesstothescoringsystemthroughthescheduledScoredEvents,andprovidemeasurementassistancetocompetitorsthroughDECREEreleases.Q100:CanthescoringmethodologybeperiodicallyupdatedsimilartotheDECREEreleases?A100:No.Allcompetitorshavehadequalaccessandabilitytocontributetothescoringalgorithmduringitsperiodofpubliccomment.Changingthescoringsystemmid‐competitionisdisruptivetotheoverallcompetitorpoolandwilldegradecompetitorconfidence.Competitorsshouldbeassuredthatscoringfundamentalswillnotbealtered.Q99:We'reusingtheoperatingsystemtomeasureperformanceandtheOSseemstomisssignificantincreasesordecreasesinperformancetiminginourreplacementCBs.WhatdoesDARPAsuggest?A99:DARPA/CGCachievedrepeatable,highprecisionperformancemeasurementsusinguniformhardwareandhardware‐supportedmeasurements.ThismeasurementapparatusbuildsontheworkofLevinthal[1‐2],Duetal[3‐4],Weaveretal[5‐6],andZapanuksetal[7].ToassistcompetitorsinachievingtimingmeasurementsthatmorecloselytracktheDARPA/CGCmeasurementapparatus,DECREEhasbeenupdatedtoprovidehigherprecisiontimingapproximationstothecompetitorsthantheprevious10msresolutionclock.Pleaseseehttps://github.com/CyberGrandChallenge/cgc‐release‐documentation/blob/master/walk‐throughs/scoring‐cbs.mdDARPA/CGCperformed100testingrunsoftheScoredEvent1challengebundleusingboththeDARPAperformancetimingapparatusandthetask.clockbasedtoolreleasedinDECREE(above).Theresultsofthisexperimentshowthatforagivenchallengebinary,ExecTimeOverheadpossessesameandifferencefromDARPA'sscoringcalculationof0.5%witha95%confidenceintervalof(0.441%,0.576%).ThisexperimentwasperformedonaHaswellmicroarchitecturewith25MBofcacheanda2.3Ghzbaseclock.[1]Levinthal,D.PerformanceAnalysisGuideforIntelCorei7ProcessorandIntelXeon5500processors2009[2]Levinthal,D.CycleAccountingAnalysisonIntelCore2Processors2009[3]Du,J.;Sehrawat,N.&Zwaenepoel,W.PerformanceProfilingofVirtualMachines,SIGPLANNot.,ACM,2011,46,3‐14[4]Du,J.;Sehrawat,N.&Zwaenepoel,W.PerformanceProfilinginaVirtualizedEnvironmentProc.HotCloud2010,USENIX,2010[5]Weaver,V.M.;Terpstra,D.&Moore,S.Non‐DeterminismandOvercountonModernHardwarePerformanceCounterImplementationsInternationalSymposiumonPerformanceAnalysisofSystemsandSoftware,2013,215‐224

Page 45: Cyber Grand Challenge - MIT Lincoln Laboratory · Q157 Added Jan 27, 2016 Q158 Added Mar 16, 2016 Q159 – Q161 Added Mar 16, 2016 Q162 – Q163 Added Mar 25, 2016 Q164 Added May

[6]Weaver,V.M.&McKee,S.A.CanHardwarePerformanceCountersbeTrusted?InternationalSymposiumonWorkloadCharacterization,2008,141‐150[7]Zaparanuks,D.;Jovic,M.&Hauswirth,M.AccuracyofPerformanceCounterMeasurementsIEEESymposiumonPerformanceAnalysisofSystemsandSoftware,2009,23‐32Q98:We'reconcernedthatDARPA'sperformancemeasurementsarenotrepeatable.Aretheyrepeatable?A98:Yes.Inaprocessknownascontinuousintegrationtesting,theSE1/SE2/CQEscoringprocessanditsassociatedunittestsarecontinuallyre‐testedandcomparedagainsthand‐confirmedperformancemeasurements.Thistestingcorpuswillgrowtoincludeallscoredevents.Theseunittestsincludeallcompetitorsubmissionstodate.SeeA99.Q97:PerthelimitedIPrightsdescribedinA81andtheCGCRulesisthereaconflictofinterestissuewiththeCQEtechnicalpaper;specifically,willthesetechnicalpapersbeusedtomanipulatethecompetitiontodefeatmydescribedapproach?A97:No.DARPA/CGCwascreatedwiththeintentionofbringingthegameofCapturetheFlagtoautomatedsystems.CTFcontestsasawholearedesignedtobeobjectivetestsofskill‐alevelplayingfieldinwhichcompetitorsfromaroundtheglobearetestedsolelyontheirabilitytosolvedifficultreverseengineeringproblems.DARPA/CGCisaCTFcompetitioninbothletterandspirit.Specifically:

DARPA/CGCwillnevertakeanyactiontointentionallydemoteorpromoteanycompetitororsetofcompetitors.

Constructionofthechallengehasbeenundertakeninanefforttomirrorrealworldchallenges.

PertheIPrightsspecifiedintheCGCRulesandA81,neithertheCQEnortheCFEtechnicalreportswillbesharedwithChallengeBinaryauthorperformerteams.

Q96:RegardingtheCQEtechnicalpaper,Iassumeyouwantanoverviewofthetools,approaches,gametheoryandprogramanalysis,butnotafulldesigndocumentanddiscussionofeveryalgorithmandheuristicused.Correct?A96:Yes

Page 46: Cyber Grand Challenge - MIT Lincoln Laboratory · Q157 Added Jan 27, 2016 Q158 Added Mar 16, 2016 Q159 – Q161 Added Mar 16, 2016 Q162 – Q163 Added Mar 25, 2016 Q164 Added May

Q95:WhathappenswhenDARPAconfirmsacceptanceofourCQEtechnicalpaper?A95:Persection3.1.3oftheCGCrules,CQEtechnicalpaperacceptanceisrequiredinordertoadvanceasafinalistandreceiveCQEprizesinaccordancewithsection3.1.4oftheRules.Q94:Inourtesting,someofourPoVsdon'tseemtoworkreliablyagainstsomeCBs.Whatshouldwedo?A94:DARPAhasdevelopedahigh‐reliabilityPoVharnessforscoringSE1/SE2/CQEthatmaymitigatethisissuesomewhatduringeventscoring.InCFE,PoVreliabilitywillbetheresponsibilityofcompetitors.Q93:Section4oftheCGCRulesrequestsnotificationofpotentialthreatstotheintegrityoftheCGCcompetition.WeareawareofamethodologytocauseatotalmappingofLinuxprocessmemorythatdoesnotcauseLinuxtoreportthememoryasallocatedorconsumed.WillthisthreatentheintegrityoftheCGCcompetition?<METHODOLOGYREDACTED>A93:No.Pleasenotethefollowing:1>DECREEisnotLinux2>DECREEisaspecification3>AreferenceimplementationofDECREEatopLinuxhasbeenreleasedongithub4>DECREEanditsassociatedreferenceimplementationsdonotsupportthereportedshenanigans5>DARPAappreciatescompetitorsupportinconductingahighintegritycompetitionQ92:ThereferencesecurityscoresforreplacementbinariesforCBe7cd3901inScoredEvent1appearedtobeincorrect.CanDARPAconfirmthosescores?A92:AnerrorintheSE1scoringsystemincorrectlyscoredthereferencePoVfore7cd3901asfailingtoprovevulnerability.ThisresultedininflatedreferencesecurityscoresforseveralreplacementCBs.Correctedscoreshavenowbeenpostedatcgc.darpa.mil.DARPAwouldliketothanktheFuzzBombTeamforreportingandhelpingdiagnosethisissue.TheSE2scoringsystemwillincorporatealllessonslearnedfromSE1.CompetitorparticipationandfeedbackcontinuestobeinvaluableontheroadtoCQE.Q91:Willthetest_eventbucketsbemaintainedforteamtestingduringtheevents?A91:Yes,thetest_eventbucketswillexistbeforeandduringscoredevents.CompetitorsareadvisedthesebucketsareatemporaryspaceandfilesmaybedeletedbyDARPAforresourceconsumptionreasonsatanytime.

Page 47: Cyber Grand Challenge - MIT Lincoln Laboratory · Q157 Added Jan 27, 2016 Q158 Added Mar 16, 2016 Q159 – Q161 Added Mar 16, 2016 Q162 – Q163 Added Mar 25, 2016 Q164 Added May

Q90:WilllibcgcimplementallLLVMintrinsics?A90:No.PriortoCQE,DARPAwillreleasetheversionoflibcgctobeusedduringCQE.Q89:Willthesourcecodetothenewchallengesincludedinthescoredeventsbereleased?A89:Yes,sourcecodewillbereleasedpriortothenextevent.Q88:TheSecurity:ReferencescorereleasedintheSE1scoringappearstohaveamaximumvalueof1.Whyisthis?A88:PerA59,theReferencecomponentoftheSecurityscorehasamaximumvalueof1.Q87:WillDARPAprovidethetotalscoresforthetoprankedteamsinScoredEvent1?A87:No.SeeQ/A68.Q86:ScoredEvent1includedpreviouslyreleased/examinedChallengeBinaries.HowwilltheseChallengeBinariesbetreatedinfutureCGCevents?A86:Definitions:‐ Previouslyreleased&examinedChallengeBinarieswillbereferredtoasOld

CBs‐ ChallengeBinariesreleasedforthefirsttimeduringacompetitioneventwillbe

referredtoasNewCBs.

ScoredEvent2:AmixofOldCBsandNewCBs.OnlyNewCBswillbeconsideredforrankingpurposes.

CQE:ScoringusingNewCBsonly. CFETrials:OldCBsandNewCBsmaybeused. CFE:ScoringusingNewCBsonly.

PleasenotethatCQE&CFEaretheonlyCGCeventsthatinvolveprizeauthority.Q85:Shouldteamsexpectpacketlossinthesamplepacketcapturesreleasedaspartoftheeventchallengebundle?A85:Yes.DARPAisevaluatingthepacketlossrateforScoredEvent2.

Page 48: Cyber Grand Challenge - MIT Lincoln Laboratory · Q157 Added Jan 27, 2016 Q158 Added Mar 16, 2016 Q159 – Q161 Added Mar 16, 2016 Q162 – Q163 Added Mar 25, 2016 Q164 Added May

Q84:WeresomePoVsubmissionsnotscoredinScoredEvent1?A84:No.AllsubmittedPoVswerescored.InafewcasestheCGCteamhadtofixtheXMLDTDpathofsubmittedPoVsoraddmissing<xml>tags.Pleaserunthescriptfoundathttps://github.com/CyberGrandChallenge/Event‐FAQ/blob/master/cqe‐verify‐solution‐package.pytocalculatethehashesusedinthescoringrelease.Infutureevents,theCGCteamwillnotmanuallycorrectmalformedsubmissions.Q83:I’mtryingtoreproduceresultsonDECREEusingmixedcomponentsfromanoldreleaseandanewrelease.HowshouldIproceed?A83:DECREEanditsassociatedtestsuitesdonotmaintainbackwardscompatibility.CompetitorsshouldusethelatestVMandtoolsuitereleaseavailable.Q82:Arechallengebinarycreatorsallowedtouseinlineassembler?A82:TheguidelinesprovidedtoCBauthorscanbefoundathttps://github.com/CyberGrandChallenge/cgc‐release‐documentation/blob/master/walk‐throughs/submitting‐a‐cb.mdQ81:WhatisthedistributionoftheCQETechnicalPaperrequiredforteamstoadvancetoCFE?A81:ThisquestionisansweredindetailintheCGCRules,Section5,IntellectualProperty;technicalpapersareprivatecommunicationstoDARPA.DARPAwillobservealllimitationsdetailedinSection5.DARPAwillnotallowaccesstotechnicalpapersbyanypartynotlistedinSection5.Competitorswishingtouseemailencryptionwhensendingtechnicalpapersmayrequestasignedemailfromcybergrandchallenge@darpa.mil;theS/MIMEpublickeyassociatedwiththisemailhasthefingerprint1E48B8A5EFC682F92C867BB6D01A3455.Q80:WilltheCBssettleonastandardlibcreplacementthatwecanrelyuponforanalysis?A80:No.SeeinstructionsprovidedtoCBauthors:https://github.com/CyberGrandChallenge/cgc‐release‐documentation/blob/master/walk‐throughs/submitting‐a‐cb.mdScoringTie‐Breakercomments(AnswersRound1,Q78‐79): Q79:CouldDARPAusethetimeoflastsubmissionasasimpletiebreaker? A79:DARPAdoesnotwishtopenalizeinternationalcompetitorswhomay: resideinlocationswithincreasedtransittimetotheCQEservershaveaccess toslow,lowerbandwidthconnectionsresideinaninconvenienttimezone relativetothestartofCQE.

Page 49: Cyber Grand Challenge - MIT Lincoln Laboratory · Q157 Added Jan 27, 2016 Q158 Added Mar 16, 2016 Q159 – Q161 Added Mar 16, 2016 Q162 – Q163 Added Mar 25, 2016 Q164 Added May

Q78:CouldthetiebreakeralgorithmfirstremovetheProof‐of‐ Vulnerabilityscoretoincentivizepurelydefensivesolutions? A78:TheSE1tiebreakeralgorithmincreasestheimportanceofthe Defensivescoreinthefirstfourofitsfivecomputationrounds.Q77:MayIsubmitaPoVofarbitrarysize?A77:No.PoVsarelimitedtonogreaterthan10x10242bytes.Q76:WhatconstitutesatestcasewhencalculatingtheretainedfunctionalityscoreinCQE?A76:AtestcaseisasingleservicepollwithinasingleTCPconnection.NoTCPconnectionwillcontainmorethanoneservicepoll.Individualservicepollswilleitherpassorfail.ReplacementChallengebinarieswillbetestedbyatleast1000individualservicepolls.TheseresultswillbeusedtocalculateretainedfunctionalityperA59.Q75:HowwillDARPAcommunicatewiththeteamsduringtheScoredEventsandCQE?A75:Competitorscanemailquestionstocybergrandchallenge@darpa.milduringtheevents.AneventFAQwillbeupdatedasneededongithubathttps://github.com/CyberGrandChallenge/Event‐FAQinordertodisseminateinformationtoallteamsperA70.TheeventFAQwillbeincorporatedintothisdocumentafterthecompletionofeachevent.Q74:IamaforeignnationalwhoiseligibletoparticipatepertheCGCRules.IhavecreatedaUS‐basedLLCwithaUS‐basedRegisteredAgenttoserveastheEntrantformyCGCteam;thisLLCisalsoeligibletoparticipatepertheCGCRules.IsthisapproachcompliantwiththeCGCRules?A74:Yes.Q73:WhathappenswhenaconnectionismadetoaDECREEservice?A73:inetd‐style.Anewinstanceiscreatedtohandlethenewconnection.Thisnewinstanceistorndownaftertheconnectionterminates.Q72:WhattypesofconnectionswillbemadeduringCQEscoring?A72:MultipleconnectionswillbemadefromServicePollers.MultipleconnectionswillalsobemadefromProofofVulnerabilitymodules.ServicePollsandPoVmoduleswillnevershareconnections.

Page 50: Cyber Grand Challenge - MIT Lincoln Laboratory · Q157 Added Jan 27, 2016 Q158 Added Mar 16, 2016 Q159 – Q161 Added Mar 16, 2016 Q162 – Q163 Added Mar 25, 2016 Q164 Added May

Q71:WhattypesofconnectionswillbemadeduringCFEscoring?A71:Multipleconnectionswillbemadefromservicepollers.Multipleconnectionswillalsobemadefromlogicbuiltbycompetitors.Servicepollsandcompetitorlogicwillnevershareconnections.Q70:WhatotheraccesstoCyberGrandChallengeisavailabletocompetitorsoutsideofthecybergrandchallenge@darpa.milemailboxandtheFAQresponses?A70:Intheinterestsofconductingafairandequitableglobalcompetition,accesstochallengeinformationismadeavailableelectronicallytoallcompetitors.Allcompetitorswhethernextdoororacrosstheglobe,maysubmitquestionsthroughthemailbox,andresponseswillbecommunicatedthroughthisFAQ.Q69:AreCFEfinalistsrequiredtobringhardwaretocompeteinCFE?A69:No.Finalistswillhavetheoptionofeither:1. BringingacompetitionsystemtoCFEinaccordancewithA31,or2. CompetinginCFEonaDARPA‐providedcomputecloudinstanceafterhaving

acceptedtheDARPACloudAgreement.EachDARPA‐providedcomputecloudinstancewillbeontheorderofhundredsofx86‐64cores.FurtherdetailsregardingtheCloudAgreementandsystemspecificationswillbereleasedatalaterdate.Q68:WhatinformationwillbereleasedtocompetitorsafterScoredEvent#1?A68:PleasenotethatinformationreleaseafterScoredEventswillbeentirelydifferentfromthepost‐CQEinformationreleaseaddressedinA25.AfterScoredEvent#1,thefollowinginformationwillbereleasedpublicly:‐ Thenamesoftheseventop‐scoringteamsinrankorder.‐ AlistofSHA‐256hashesforsubmittedChallengeBinariesandtheir

associatedscoresandcorrespondingreferenceCBname.‐ AlistofSHA‐256hashesforPoVsandtheirassociatedscoresand

correspondingreferenceCBname.Pleasenote,thesereleasedhashlistswillnotcorrelatescoredsubmissionstoteams.CompetitorswillberequiredtocalculateSHA‐256hashesoftheirsubmittedinputsinordertodeterminetheirscores.

Page 51: Cyber Grand Challenge - MIT Lincoln Laboratory · Q157 Added Jan 27, 2016 Q158 Added Mar 16, 2016 Q159 – Q161 Added Mar 16, 2016 Q162 – Q163 Added Mar 25, 2016 Q164 Added May

Q67:HowwillrankingoccurinScoredEvent#1?A67:Multiplesubmissionsmaybescored;hashlistinformationonmultiplesubmissionswillbeavailableviathehashlistformat(A68).Rankswillbedeterminedusingthescoreassignedtoeachteam’sfinalsubmission.Q66:WhatwillCQEChallengeBundlecontain?A66:AtthebeginningofCQE,competitorswillgainaccesstoCQEChallengeBundle(bundlewillcontainacollectionofReferenceCBs,aswellassomepcaprecordingsofsomeservicepollinteractionsbetweenServicePollersandtheseReferenceCBs).Theseservicepollinteractionsamples,wherepresent,arenotguaranteedtobecomplete.Q65:WhatwillScoredEventChallengeBundlescontain?A65:ScoredeventsareintendedtoprovidetechnicalpreparationforCQE;thereforetheScoredEventBundleswillmirrortheformatoftheCQEChallengeBundletothegreatestextentpossible.CompetitorsshouldnotethattheCQEBundlewillbemuchlargerthantheScoredEventBundles.TheseScoredEventBundlesmayalsore‐usepreviouslyreleasedCBs.Q64:WhatisDECREE?A64:DECREEisanopen‐sourceextensionbuiltatoptheLinuxoperatingsystem.Constructedfromthegroundupasaplatformforoperatingsmall,isolatedsoftwaretestsamplesthatareincompatiblewithanyothersoftwareintheworld—DECREEaimstoprovideasaferesearchandexperimentationenvironmentfortheCyberGrandChallenge.DECREEbinariesandsourceareavailable:http://repo.cybergrandchallenge.com/http://github.com/cybergrandchallenge/Q63:HowshouldissuesinDECREEbereported?A63:[email protected]:WillalladvancedapplicationdefensesthatpreventarbitrarycodefromrunningincreasethesecurityscoreinCQE?A62:No.CGCscoringdoesnotrequirearbitrarycodeexecution,thereforemechanismswhichfrustratearbitrarycodeexecutionwillnotnecessarilypreventscoringevents.InCQE,competitorshavetheopportunitytomitigatedenialofserviceflaws.SeealsoQ4.

Page 52: Cyber Grand Challenge - MIT Lincoln Laboratory · Q157 Added Jan 27, 2016 Q158 Added Mar 16, 2016 Q159 – Q161 Added Mar 16, 2016 Q162 – Q163 Added Mar 25, 2016 Q164 Added May

Q61:WilltheReferencePatchedCBperformdifferentlythantheOriginalCB?A61:AdiversegroupofsoftwareauthorsarebuildingalargecorpusofCBsforCGCincorporatingmanyclassesofvulnerabilities.TheseCBauthorsarerequiredtoprovideasingleReferencePatchedCBthatpassesthesamefunctionalitytestsuiteastheOriginalCBandisnotsusceptibletoanyofthereferencePoVs.Q60:HowdoestheInterProcessCommunication(IPC)workinChallengeBinaries(CBs)?A60:DECREEprecludescommunicationviasharedmemory,network,orpersistentstoragebetweendifferentCBsaswellasdifferentconnectionsservicedbythesameCB.InordertoofferarichCBportfoliowithbroadCWEcoverageincludingconcurrencyissues,DARPAallowsfortheuseofaCGCIPCmechanismwithinasingleCB,whichworksasfollows.EachCBmaybecomposedofmultiplebinariesrunningindistinctprocesses.TheCGCcompetitionframeworkwilllaunchallofthebinariesassociatedwiththechallenge.Eachoftheseprocesseswillbepre‐connectedwithfiledescriptorstocommunicatewiththeothersviareceive()andtransmit()systemcalls(seeFigure1)inamannerdeterminedbytheCBauthor(seeFigure2).ExampleIPCCBswillbeprovidedpriortothefirstCQEScoredEvent.

Figure1:Filedescriptorconnections

Figure2:ExampleuseofIPCinter‐connections

Q59:WhatisthescoringmethodforCQE?A59:DARPAheldaperiodofpubliccommentfortheCFEandCQEscoringmethods,andfeedbackfromtheCGCcommunitywasreceivedandreviewedbyDARPA.UpdatedscoringmethodsforCQEandCFEwerereleasedonMonday,March10,2014.ThesescoringmethodsareconsistentwiththeCyberGrandChallengeRules.SeealsoQ/A26,27,and30.

Page 53: Cyber Grand Challenge - MIT Lincoln Laboratory · Q157 Added Jan 27, 2016 Q158 Added Mar 16, 2016 Q159 – Q161 Added Mar 16, 2016 Q162 – Q163 Added Mar 25, 2016 Q164 Added May

CQEscoringistheproductofthreeassessedquantities:Availabilityscore,Securityscore,andEvaluationscore.ThesescoresmaptotheAreasofExcellence(AoE)locatedintheCyberGrandChallengeRulesasfollows:Availability:AreaofExcellence4Security:AreaofExcellence2Evaluation:AreaofExcellence3DuringCQE,theCBsdistributedbyDARPAshallbereferredtoas“referenceCBs”,nottobeconfusedwithreplacementCBssubmittedbycompetitors.PoVsusedbyDARPAforscoringwillbereferredtoas“referencePoVs”.PoVssubmittedbycompetitorsystemstoDARPAduringCQEwillbereferredtoas“submittedPoVs”.DARPAwillscoreonereplacementCBandonesubmittedPoVperreferenceChallengeBinary.CQEScoreswillbeassessedperreferenceChallengeBinary(“CBscore”).EachCBscorewillbecalculatedasfollows:

Availability*Security*Evaluation

Availability:Thisquantityshallvaryasamulti‐stepfunctionbetween0and1,with1beingaperfectscore.PerformanceandRetainedFunctionalitywillbemeasured,withAvailabilitybeingsettotheminimumofthesequantities.‐ PerformanceofasubmittedCBwilldecrementbasedonthegreatestmeasured

increaseinsystemresourceutilization.Thesemeasuredincreasesincludefilesize,executiontime,andmemoryusage.Foreachofthesemeasuredquantities,amaximumacceptableincreaseisindicatedbelow,afterwhichPerformancebeginstodecrement: Filesize:+40% Executiontime:+10% Memoryusage:+10%

‐ RetainedFunctionalitywillbethepercentageoftestcasesthereplacementCB

passes.Competitorsareadvisedthatuseofamulti‐stepfunctionimposesafaster‐than‐linearAvailabilitydropoffbasedondamagetoPerformanceorRetainedFunctionality.Security:Thisquantityshallbedeterminedusingtwoquantities,ReferenceandConsensus:‐ Reference:ThenumberofreferencePoVswhichdonotprovevulnerabilityin

thereplacementCB,dividedbythenumberofreferencePoVs

Page 54: Cyber Grand Challenge - MIT Lincoln Laboratory · Q157 Added Jan 27, 2016 Q158 Added Mar 16, 2016 Q159 – Q161 Added Mar 16, 2016 Q162 – Q163 Added Mar 25, 2016 Q164 Added May

‐ Consensus:Thisquantitywillbesetto0or1:• 0:AnysubmittedPoVprovedvulnerabilityinthereplacementCB• 1:NosubmittedPoVprovedvulnerabilityinthereplacementCB

IfReferenceiszero,Securitywillbesettozero.IfReferenceisnonzero,Securitywillbecalculatedasfollows: 1+(Reference+Consensus)/2Evaluation:Thisquantitywillbesetto1or2:‐ 1:ThePoVemittedbythisCRSdidnotprovevulnerabilityinthereferenceCB‐ 2:ThePoVemittedbythisCRSdidprovevulnerabilityinthereferenceCBACRS’stotalscoreattheendofCQEshallbethesumofthatCRS’sCBscores.Q58:WhatisthescoringmethodforCFE?A58:DARPAheldaperiodofpubliccommentfortheCFEandCQEscoringmethods,andfeedbackfromtheCGCcommunitywasreceivedandreviewedbyDARPA.UpdatedscoringmethodsforCQEandCFEwerereleasedonMonday,March10,2014.ThesescoringmethodsareconsistentwiththeCyberGrandChallengeRules.SeealsoQ/A26,27,and30.CFEscoringistheproductofthreeassessedquantities:Availabilityscore,Securityscore,andEvaluationscore.ThesescoresmaptotheAreasofExcellence(AoE)locatedintheCyberGrandChallengeRulesasfollows:Availability:AreaofExcellence4Security:AoE2and5Evaluation:AreaofExcellence3CFEscoringwilltakeplaceovermanyrounds.ScoreswillbeassessedperChallengeBinaryperround(“CBroundscore”);itfollowsthatthesumoftheCBroundscoresforasingleCRSwillbethetotalscoreforthatsystem.EachCBroundscorewillbecalculatedasfollows:

Availability*Security*EvaluationAvailability:Thisquantityshallvaryasamulti‐stepfunctionbetween0and1,with1beingaperfectscore.PerformanceandRetainedFunctionalitywillbemeasured,withAvailabilitybeingsettotheminimumofthesequantities.‐ PerformanceofasubmittedCBwilldecrementbasedonthegreatestmeasured

increaseinsystemresourceutilization.Thesemeasuredincreasesincludefilesize,executiontime,andmemoryusage.Foreachofthesemeasuredquantities,amaximumacceptableincreaseisindicatedbelow,afterwhichPerformancebeginstodecrement:

Page 55: Cyber Grand Challenge - MIT Lincoln Laboratory · Q157 Added Jan 27, 2016 Q158 Added Mar 16, 2016 Q159 – Q161 Added Mar 16, 2016 Q162 – Q163 Added Mar 25, 2016 Q164 Added May

Filesize:+20% Executiontime:+5% Memoryusage:+5%

‐ RetainedFunctionalitywillbethepercentageofnetworktestcasesthe

replacementCBpasses. Competitorsareadvisedthatuseofamulti‐stepfunctionimposesafaster‐than‐

linearAvailabilitydropoffbasedondamagetoPerformanceorRetainedFunctionality.

Security:Thisquantitywillbesetto1or2:‐ 1:AtleastonePoVfromacompetitorprovedvulnerabilityinthisCBduringthis

round‐ 2:NoPoVsfromcompetitorsprovedvulnerabilityinthisCBduringthisroundEvaluation:Thisquantitywillvarybetween1and2:‐ 1:NoPoVsemittedbythisCRSprovedvulnerabilitiesinthisCBserviceon

competitorsystems‐ 1+(x/(N‐1))PoVsemittedbythisCRSforthisCBserviceprovedvulnerabilityin

xcompetitorsystems,whereNisthenumberofCRSsparticipatinginCFE.ACRS’stotalscoreattheendofCFEshallbethesumofthatCRS’sCBroundscores.Q57:WhatistheAPItotheCGCenvironment?A57:ThefollowingClanguagefunctionprototypesareprovided:

void _terminate(unsigned int status); int allocate(size_t length, int prot, void **addr); int deallocate(void *addr, size_t length); int fdwait(int nfds, fd_set *readfds, fd_set *writefds,

struct timeval *timeout, int *readyfds); int random(void *buf, size_t count, size_t *rnd_bytes); int receive(int fd, void *buf, size_t count, size_t

*rx_bytes); int transmit(int fd, const void *buf, size_t count, size_t

*tx_bytes);ThesefunctionprototypesarenotionalandmaybeimprovedduetofeedbackpriortoCGCkickoff.Q56:Canforeignnationalsparticipateinthischallenge?

Page 56: Cyber Grand Challenge - MIT Lincoln Laboratory · Q157 Added Jan 27, 2016 Q158 Added Mar 16, 2016 Q159 – Q161 Added Mar 16, 2016 Q162 – Q163 Added Mar 25, 2016 Q164 Added May

A56:ThisquestionisaddressedintheCGCRulesSection2andSection6.ForeignnationalsmayparticipateinCyberGrandChallengewithinateamwhichconformstotheCGCRules.Q55:DARPA‐BAA‐14‐05mentionsDARPA‐BAA‐14‐03,whichdescribesthearchitectureframework.WhereisDARPA‐BAA‐14‐03?A55:DARPAanticipatesDARPA‐BAA‐14‐03tobepublishedinthenearfuture.Q54:DoesDARPAhaveacompletegovernmentteamorarethereopportunitiesforCGCsupportindevelopment,judging,operating,etc.?A54:DARPAanticipatesasecondBAAwithotheropportunitieswithinthischallenge.Q53:Canforeignteamsapplyforthefundingalsoorcanteamshaveforeignmembers?A53:ReviewtheeligibilitysectionofDARPA‐BAA‐14‐05(3.1.4)andtheRules(2.1).Q52:Isthis6.1or6.2money?A52:DARPAanticipates6.2.fundsforawardsunderDARPA‐BAA‐14‐05andDARPA‐BAA‐14‐03.Q51:Doesfundamentalversusnon‐fundamentalaffectdesirability?A51:SeeDARPA‐BAA‐14‐05section2.2.Q50:Arethereanyrestrictionsonforeignsubcontractors?Ifso,whataretherestrictions?A50:Seesection3.1.3ofDARPA‐BAA‐14‐05.Q49:Willtheproposalevaluationsfavorsmallbusiness,orisitalevelplayingfieldbasedonmerit?A49:Seesection5ofDARPA‐BAA‐14‐05.Allproposalsareevaluatedonthesamecriteria.Q48:ArethedeliverablesandpaymentpercentagesinDARPA‐BAA‐14‐05fixed,orcanweproposealternatives?A48:Theyarenotional,notfixed.Youcanproposealternatives.

Page 57: Cyber Grand Challenge - MIT Lincoln Laboratory · Q157 Added Jan 27, 2016 Q158 Added Mar 16, 2016 Q159 – Q161 Added Mar 16, 2016 Q162 – Q163 Added Mar 25, 2016 Q164 Added May

Q47:Canyouclarifythelengthoftheperiodsofperformanceforthebaseandoptionperiods?A47:UnderDARPA‐BAA‐14‐05,eachperiodofperformanceis12months.ThescheduleinDARPA‐BAA‐14‐05isnotional.Planforallactivitiestotakeplacewithintwo12monthphases.Q46:IsitpossibletocombinewithanothergroupaftertheCQE?A46:Yes.Q45:Cananorganizationhavetwoteams,oneforOpentrackandoneforProposaltrack?A45:ThisisexcludedintheRules.Teamsareintendedtobewhollyseparate.

Q44:IfIsubmitaproposaltotheCompetitionBAA(DARPA‐BAA‐14‐05)anddonotgetselected,canIsubmittotheArchitectureBAA(DARPA‐BAA‐14‐03)?A44:There’snothingtopreventyoufromsubmittingtoboth,butyoucannotbeselectedforawardunderboth.IntheeventthataproposersubmitsanotherwiseselectableproposaltobothDARPA‐BAA‐14‐05andDARPA‐BAA‐14‐03,thedecisionastowhichproposaltoconsiderforawardisatthediscretionoftheGovernment.Q43:MustwedeliveraworkingspreadsheetaspartoftheproposalforDARPA‐BAA‐14‐05oristhatjustDARPA’spreference?Yousaiditwouldbe“helpful”versus“required?”A43:Persection4.2.1.2ofDARPA‐BAA‐14‐05,thecostproposalshouldincludeaspreadsheetfile(.xlsorequivalentformat)thatprovidesformulatraceabilityamongallcomponentsofthecostproposal.Thespreadsheetfilemustbeincludedasaseparatecomponentofthefullproposalpackage.Q42:CanwetalktotheContractingOfficerbeforeaproposalissubmitted?A42:ReferenceSection7ofDARPA‐BAA‐14‐05,questionsshouldbesubmittedtoCGC‐[email protected]:AretheretwoBAA’santicipatedforthisprogram,theArchitectureBAA(DARPA‐BAA‐14‐03)andtheCompetitionBAA(DARPA‐BAA‐14‐05)?A41:Yes.

Page 58: Cyber Grand Challenge - MIT Lincoln Laboratory · Q157 Added Jan 27, 2016 Q158 Added Mar 16, 2016 Q159 – Q161 Added Mar 16, 2016 Q162 – Q163 Added Mar 25, 2016 Q164 Added May

Q40:WhatistheeligibilityforusinganOTforprototypes(845)?A40:SeeDARPA’scontractmanagementwebsite(http://www.darpa.mil/Opportunities/Contract_Management/Other_Transactions_and_Technology_Investment_Agreements.aspx)forinformationregardingOTforPrototypeawards.Q39:IstheelectronicsubmittalsystemsimilartoT‐FIMS?A39:Yes.Q38:Couldtheamountsoftheprojectbelargerifanentitysuppliedacostsharebeyondthe$750k?A38:Yes.Q37:WithregardtoSection4.2.1.2.3ofDARPA‐BAA‐14‐05,wherearegovernmentratesandDefenseContractAuditAgency(DCAA)ratesdefined?A37:FARPart42discussesproceduresforestablishingforwardpricingrates.InformationisalsoavailableontheDefenseContractManagementAgency’s(DCMA)Websitehttp://guidebook.dcma.mil/41/.YoudonothavetohaveDCMAapprovedratestoproposeandreceiveanawardunderDARPA‐BAA‐14‐05.Section4.2.1.2.3requiresaproposertojustifyitsproposeddirectlaborratesandprovidesseveralexamplesofhowthatcanbeaccomplished.Q36:WithregardtoSection4.2.1.1.1ofDARPA‐BAA‐14‐05,wherearethetypesofbusinessesdescribed?A36:BusinesssizesaredefinedbytheSmallBusinessAdministration(http://www.sba.gov/content/table‐small‐business‐size‐standards).AdefinitionofHBCUandMinorityInstitutionscanbefoundinDFARS252.226‐7000(http://www.acq.osd.mil/dpap/dars/dfars/html/current/252226.htm#252.226‐7000).Q35:Istherealimittothenumberofteamsawardedortotalamountofgrants?A35:NograntswillbeawardedunderDARPA‐BAA‐14‐05,onlyFirm‐Fixed‐PriceProcurementContractsandOtherTransactions.UnderDARPA‐BAA‐14‐05,DARPAanticipatesmultipleawardsof$750,000perphaseofatwo‐phaseeffort;however,pertheBAA,thenumber/amountofawardswilldependonthequalityoftheproposalsreceivedandtheavailabilityoffunds.

Page 59: Cyber Grand Challenge - MIT Lincoln Laboratory · Q157 Added Jan 27, 2016 Q158 Added Mar 16, 2016 Q159 – Q161 Added Mar 16, 2016 Q162 – Q163 Added Mar 25, 2016 Q164 Added May

Q34:Willacceptedproposalsbecomepublic?A34.DARPAwillnotpublishawardedproposalsunderDARPA‐BAA‐14‐05.Persection4.2.2oftheBAA,DARPAtreatsproposalsassourceselectioninformation(seeFAR2.101and3.104)andprotectsthemassuch,usingsecurehandlinganddestructionprocedures.Q33:DuringCFE,howwillaCRSmonitorandmodifytraffictoanetworkedhost?A33:Monitor:DuringCFE,eachcompetitorCRSwillreceivearead‐onlystreamofallCompetitorCRSnetworktrafficdirectedtowarditsnetworkhostovertheCFEnetwork.Modify:CompetitorsystemswillbeprovidedwithaccesstoaDARPA‐managednetworkappliancewithinthecompetitionframeworkwhichwillallowfortrafficmodificationbetweentheCFEnetworkandthenetworkhostdefendedbytheCRS.ThemanagedappliancewillusefiltersprovidedbytheCRSthroughtheCompetitionFrameworkAPI.Filtersneednotmodifytraffic;dependingontheirformulationtheymaymodifytraffic,alert,ortakenoaction.LikeChallengeBinaries,filtersprovidedbyaCRSwillbedistributedtoallcompetitorsystemsforpurposesofconsensusevaluation(Shannon'sMaxim).AlertsgeneratedonthemanagedappliancebyfieldedfilterswillbecommunicatedbacktotheCRSthroughtheCompetitionFrameworkAPI.TheseCRS‐providedfilterswillconformtoasubsetoftheopensourceSnortrulesyntax;thissubsetwillbepublishedbyDARPApriortokickoffofCyberGrandChallenge.Q32:HowwillCRShardwareinterfacewiththeCQEevent?A32:CQEisanInternet‐accessibleeventthatwillnotrequireaCRStobephysicallypresentataneventlocation.Section3.1oftheCGCRulesprovidesinformationontheCQECorpusandworkflow.Competitorsshouldnotetherequirementsofthepost‐CQESiteVisitfromtheCGCRules,Section3.1.3.2.Q31:HowwillCRShardwareinterfacewiththecompetitionframeworkduringCFE?A31:ACRSusedinCFEmustfitentirelyinasingle,standard19"42Urack.CFEwilltakeplaceataphysicallocationtobespecifiedbyDARPA.CompetitorsystemsmustbephysicallypresentattheCFEComputeLocationinordertointerfacedirectlywiththecompetitionframework.Furtherdetailsaboutavailablepower,cooling,andinfrastructurewillbepublishedasCyberGrandChallengeprogresses.

Page 60: Cyber Grand Challenge - MIT Lincoln Laboratory · Q157 Added Jan 27, 2016 Q158 Added Mar 16, 2016 Q159 – Q161 Added Mar 16, 2016 Q162 – Q163 Added Mar 25, 2016 Q164 Added May

Q30:WhenwillCQEandCFEscoringbefinalized?A30:DARPAiscurrentlyholdingaperiodofpubliccommentfortheCFEandCQEscoringmethods.Feedbackshouldbeaddressedtocybergrandchallenge@darpa.mil.ScoringwillbeopenforaperiodofpubliccommentconcludingonFriday,February14,2014.AfinalscoringalgorithmforCQEandCFEwillbereleasedonMonday,March10,2014.Likethecurrentproposedscoringalgorithms,thefinalscoringmethodswillbeconsistentwiththeCyberGrandChallengeRules.SeealsoQ/A26and27.Q29:WhatconstitutesaProofofVulnerabilityinCFE?A29:ProofofVulnerability(PoV)maybedemonstratedduringCFEbyinteractingwiththeCompetitionFrameworkviatwodifferentmethods.Method1:RegistersToprovethataremoteCBisvulnerable,aMethod1PoVshallredirectexecutiontoaninstructionpointervaluenegotiatedwiththeCompetitionFrameworkAPI.Executionmustfaultatthisvalue.Attimeoffault,onegeneralpurposeregistermustalsobesettoasecondvaluenegotiatedwiththeCompetitionFrameworkAPI.Thisredirectionwillcreateaneffectdetectablebyinstrumentationoperatedonthecompetitionframework.Method2:MemoryToprovethataremoteCBisvulnerable,aMethod2PoVwillretrieveavaluepresentinthememoryofacompetitorCB.ThisvaluemustberetrievedfromamemoryrangespecifiedbytheCompetitionFrameworkAPI,mustbeofasizespecifiedbytheCompetitionFrameworkAPI,andmustbesubmittedthroughtheCompetitionFrameworkAPI.Thissubmittalwillcreateaneffectdetectablebyinstrumentationoperatedonthecompetitionframework.Q28:WhatconstitutesaProofofVulnerabilityinCQE?A28:ACQEProofofVulnerabilitywillcauseaCBtocrashinamannerconsistentwithmemorycorruptionflawswithsecurityimpact,e.g.,asegmentationfaultorillegalinstructionfault.

Page 61: Cyber Grand Challenge - MIT Lincoln Laboratory · Q157 Added Jan 27, 2016 Q158 Added Mar 16, 2016 Q159 – Q161 Added Mar 16, 2016 Q162 – Q163 Added Mar 25, 2016 Q164 Added May

Q27:WhatisthescoringmethodforCQE?A27:ThebelowscoringmethodcandidatewillbeopenforpubliccommentuntilFriday,February14,2014.CQEscoringistheproductofthreeassessedquantities:Availabilityscore,SecurityScore,andEvaluationscore.ThesescoresmaptotheAreasofExcellence(AoE)locatedintheCyberGrandChallengeRulesasfollows:Availability:AreaofExcellence4Security:AreaofExcellence2Evaluation:AreaofExcellence3DuringCQE,theCBsdistributedbyDARPAshallbereferredtoas“referenceCBs”,nottobeconfusedwithreplacementCBssubmittedbycompetitors.PoVsusedbyDARPAforscoringwillbereferredtoas“referencePoVs”.PoVssubmittedbycompetitorsystemstoDARPAduringCQEwillbereferredtoas“submittedPoVs”.CQEScoreswillbeassessedperChallengeBinary(“CBscore”).EachCBscorewillbecalculatedasfollows:

Availability*Security*EvaluationAvailability:Thisquantityshallvaryasamulti‐stepfunctionbetween0and1,with1beingaperfectscore.Performanceandretainedfunctionalitywillbemeasured,withAvailabilitybeingsettotheminimumofthesequantities.CompetitorsareadvisedthatslowingdownthefunctionofareplacementCBwillresultinafaster‐than‐linearAvailabilityscoredropoff.Security:Thisquantitywillbecalculatedasfollows:1+(Reference+Consensus)/2‐ Reference:ThenumberofreferencePoVswhichdonotprovevulnerabilityin

thereplacementCB,dividedbythenumberofreferencePoVs‐ Consensus:

Thisquantitywillbesetto0or1:o 0:AnysubmittedPoVprovedvulnerabilityinthereplacementCBo 1:NosubmittedPoVprovedvulnerabilityinthereplacementCB

Evaluation:Thisquantitywillbesetto1or2:‐ 1:APoVemittedbythisCRSdidnotprovevulnerabilityinthereferenceCB‐ 2:APoVemittedbythisCRSdidprovevulnerabilityinthereferenceCBACRS’stotalscoreattheendofCQEshallbethesumofthatCRS’sCBscores.

Page 62: Cyber Grand Challenge - MIT Lincoln Laboratory · Q157 Added Jan 27, 2016 Q158 Added Mar 16, 2016 Q159 – Q161 Added Mar 16, 2016 Q162 – Q163 Added Mar 25, 2016 Q164 Added May

Q26:WhatisthescoringmethodforCFE?ThisscoringmethodcandidatewillbeopenforpubliccommentuntilFriday,February14,2014.A26:CFEscoringistheproductofthreeassessedquantities:Availabilityscore,SecurityScore,andEvaluationscore.ThesescoresmaptotheAreasofExcellence(AoE)locatedintheCyberGrandChallengeRulesasfollows:Availability:AreaofExcellence4Security:AoE2,5Evaluation:AreaofExcellence3CFEscoringwilltakeplaceovermanyrounds.ScoreswillbeassessedperChallengeBinaryperround(“CBroundscore”);itfollowsthatthesumoftheCBroundscoresforasingleCRSwillbethetotalscoreforthatsystem.EachCBroundscorewillbecalculatedasfollows:

Availability*Security*EvaluationAvailability:Thisquantityshallvaryasamulti‐stepfunctionbetween0and1,with1beingaperfectscore.Performanceandretainedfunctionalitywillbemeasured,withAvailabilitybeingsettotheminimumofthesequantities.CompetitorsareadvisedthatslowingdownthefunctionofareplacementCBwillresultinafaster‐than‐linearAvailabilityscoredropoff.Security:Thisquantitywillbesetto1or2:‐ 1:AtleastonePoVfromacompetitorprovedvulnerabilityinthisCBduringthis

round‐ 2:NoPoVsfromcompetitorsprovedvulnerabilityinthisCBduringthisroundEvaluation:Thisquantitywillvarybetween1and2:‐ 1:NoPoVsemittedbythisCRSprovedvulnerabilitiesinthisCBserviceon

competitorsystems‐ 1+(x/(N‐1))PoVsemittedbythisCRSforthisCBserviceprovedvulnerabilityin

xcompetitorsystems,whereNisthenumberofCRSsparticipatinginCFE.ACRS’stotalscoreattheendofCFEshallbethesumofthatCRS’sCBroundscores.Q25:WhatwillbepubliclyreleasedPost‐CQE?A25:DARPAintendstoreleasethefollowingitemspost‐CQE:

‐ ReferenceCBs(initialCorpusdistributedforCQE)‐ PoVs,includingbothreferencePoVsandPoVsgatheredduringtheCQE

Page 63: Cyber Grand Challenge - MIT Lincoln Laboratory · Q157 Added Jan 27, 2016 Q158 Added Mar 16, 2016 Q159 – Q161 Added Mar 16, 2016 Q162 – Q163 Added Mar 25, 2016 Q164 Added May

‐ ReplacementCBsfromtheCQE,includingreferencepatchedCBs‐ PCAPoftrafficusedduringCQEevaluation‐ ReferenceservicepollersforeachCB‐ ReferenceCBsourcecode‐ AdetailedlistofscoresforeachCBforeachfinalist‐ Teamrankings(includingOpenTrackandProposalTrack)

DARPAmaymodifythislistofintendeddeliverablesatitssolediscretion.Q24:Whatinformationaboutchallengebinarieswillbeprovidedaheadoftime(e.g.,sampleinputandresponse;interactionprotocol,APIforservice,etc.)?A24:DARPAwillprovideaninterfacedocumentdetailingthemethodsCBswillusetointerfacewiththeirexecutionenvironment.Q23:Whatwillweknowaboutchallengenetworkconfiguration(e.g.,addressranges)beforethefinalevent?A23:TheCFEnetworktopologywillbeknownpriortoCFE.Inaddition,competitorswillhavetheopportunitytotesttechnologyinteroperabilityduringCFETrials.Q22:Willtheexecutionenvironmentbeprovidedtotheteams?A22:Asampleenvironmentwillbeprovidedpriortotheprogramcommencing(proposaltrackawardshavebeenfinalizedandopentrackteamshavebeenregistered/accepted)intheformofavirtualmachine.Q21:WillsampleinputsbeprovidedwithsomeofthechallengebinariesintheCQEcorpus?A21:Yes.Q20:CansecurereplacementCBsbesubmittedbyaCRSthroughoutCFE?A20:Yes.Q19:WhatistheimpactofsubmittingareplacementCB?A19:ThesubmissionofsecurereplacementsmayberatelimitedbytheCompetitionFrameworkAPI,andfieldingareplacementCBmayimpactserviceavailability.

Page 64: Cyber Grand Challenge - MIT Lincoln Laboratory · Q157 Added Jan 27, 2016 Q158 Added Mar 16, 2016 Q159 – Q161 Added Mar 16, 2016 Q162 – Q163 Added Mar 25, 2016 Q164 Added May

Q18:Aretherenetworkingconstraintsonpatching?Reachingouttoremoteservers?MayCBscommunicatewiththeCRSwhileexecutingonthenetworkhost?A18:DuringCFE,ChallengeBinarieswillnothavetheabilitytoinitiateconnections.Q17:DuringCFE,fornetworkdefense,willexistingtoolsforscanninganddefending(TCP/UDP/NMAP,wireshark,snort,etc.)work,ormustwedevelopnewtools?Doyouexpecttheteamstodevelopprogramanalysistoolsthemselvesoruseoff‐the‐shelfones?A17:DARPAwillnotdictatewhatautomatedapproachesareacceptablewithinaCRS.Q16:DuringCFE,whatinformation(datasources)willourCRShaveaccessto?SpecificallywillourCRShaveaccesstocrashlogs,coredumps,andfullnetworktrafficfeed?A16:DuringCFE,aCRSwillhaveaccesstoareadonlynetworktap.DuringCFE,aCRSwillhavetheabilitytorequestsomeCBstatusinformationthroughtheCompetitionFrameworkAPI.DatasourcesautomaticallygeneratedbyaCRSinternallywillnotbedictatedbyDARPA.Q15:DuringCFE,howmanynetworkedhostswillcompetitorsberesponsibleformonitoring/protecting?A15:One.Q14:DuringCFE,willcompetitorshaveaccesstothenetworkhost?A14:ACRSwillhavetheabilitytoquerytheCompetitionFrameworkAPIforsomeCBstatusinformation.ACRSwillhavetheabilitytofieldreplacementCBsthroughtheCompetitionFrameworkAPI.Q13:DuringCFE,willyoubeissuingnewbinariestoteamsaftercompetitionstart,orwillyougiveallbinariestoteamsbeforestart?A13:DuringCFE,aCRSwillbenotifiedthataCBisavailablethroughtheCompetitionFrameworkAPI.Q12:WhatprogramminglanguageswillCBsbewrittenin?A12:TheCfamilyoflanguages.

Page 65: Cyber Grand Challenge - MIT Lincoln Laboratory · Q157 Added Jan 27, 2016 Q158 Added Mar 16, 2016 Q159 – Q161 Added Mar 16, 2016 Q162 – Q163 Added Mar 25, 2016 Q164 Added May

Q11:DoestheU.S.GovernmentassertanyintellectualpropertyrightstoCRSsourcecodedevelopedbyopentrackcompetitors?A11:No.Q10:WhattypeofsecurityvulnerabilitieswillCGCaddress?A10:CGCChallengeBinariesshallcontaintraditionalmemorycorruptionflaws.AsubsetofrelevantflawtypesdrawnfromtheMITRECommonWeaknessEnumerationentriesasfoundonhttp://cwe.mitre.org/follows;teamsareencouragedtomakeuseofthislistasastartingpoint,notareference.CWE‐120:BufferCopywithoutCheckingSizeofInput('ClassicBufferOverflow')CWE‐121:Stack‐basedBufferOverflowCWE‐122:Heap‐basedBufferOverflowCWE‐123:Write‐what‐whereConditionCWE‐124:BufferUnderwrite('BufferUnderflow')CWE‐128:Wrap‐aroundErrorCWE‐129:ImproperValidationofArrayIndexCWE‐130:ImproperHandlingofLengthParameterInconsistencyCWE‐131:IncorrectCalculationofBufferSizeCWE‐134:UncontrolledFormatStringCWE‐135:IncorrectCalculationofMulti‐ByteStringLengthCWE‐147:ImproperNeutralizationofInputTerminatorsCWE‐158:ImproperNeutralizationofNullByteorNULCharacterCWE‐170:ImproperNullTerminationCWE‐190:IntegerOverfloworWraparoundCWE‐191:IntegerUnderflow(WraporWraparound)CWE‐193:Off‐by‐oneErrorCWE‐194:UnexpectedSignExtensionCWE‐195:SignedtoUnsignedConversionErrorCWE‐196:UnsignedtoSignedConversionErrorCWE‐401:ImproperReleaseofMemoryBeforeRemovingLastReferenceCWE‐409:ImproperHandlingofHighlyCompressedData(DataAmplification)CWE‐415:DoubleFreeCWE‐416:UseAfterFreeCWE‐457:UseofUninitializedVariableCWE‐466:ReturnofpointervalueoutsideofexpectedrangeCWE‐467:Useofsizeof()onaPointerTypeCWE‐468:IncorrectPointerScalingCWE‐469:UseofPointerSubtractiontoDetermineSizeCWE‐763:ReleaseofInvalidPointerorReferenceCWE‐786:AccessofMemoryLocationBeforeStartofBufferCWE‐787:Out‐of‐boundsWriteCWE‐788:AccessofMemoryLocationAfterEndofBufferCWE‐805:BufferAccesswithIncorrectLengthValue

Page 66: Cyber Grand Challenge - MIT Lincoln Laboratory · Q157 Added Jan 27, 2016 Q158 Added Mar 16, 2016 Q159 – Q161 Added Mar 16, 2016 Q162 – Q163 Added Mar 25, 2016 Q164 Added May

CWE‐806:BufferAccessUsingSizeofSourceBufferCWE‐822:UntrustedPointerDereferenceCWE‐823:UseofOut‐of‐rangePointerOffsetCWE‐824:AccessofUninitializedPointerCWE‐825:ExpiredPointerDereferenceQ9:WhatconstitutesasoftwareflawinCyberGrandChallenge?A9:DARPACGCwillnotprovideaformaldefinitionofasoftwareflaw;thisquestionliesoutsidethescopeofthechallenge.TheCGCwilloperateinthetraditionofexistingcybercompetitions:aflawisprovenwhenaninputdeliveredfromthenetworktoaflawedsoftwareprogram(CB)createsaneffectdetectablebyinstrumentationoperatedbythecompetitionframework.CGCChallengeBinarieswillcontainmemorycorruptionflawsrepresentativeofflawscategorizedbytheMITRECWE1,however,CompetitorSystemsmayproveanysoftwareflawtheydiscoverthroughautomatedreasoning.AlistofrepresentativeCWEcategorieswillbereleasedpriortothekickoffofCyberGrandChallenge.Q8:WhatplatformwillCGCrunon?A8:CGCChallengeBinaries(CBs)willbeincompatiblewithanyknownOSarchitecture.CBswillruninanenvironmentcustombuiltforthecompetition.Knowledgeoftheoperatingsystemwillnotbeinscopeforthecompetition;rather,CGCrequiresacompetitionsystemtoreasonaboutthefunctionofcompiledbinariesreceivinginputsfromthenetwork.CBswillnotconformtoanycurrentlyknownapplicationlayerprotocols.CBprotocolknowledgemustbegeneratedautomaticallybycompetitionsystemsduringCGCeventsthroughaprocessofautomatedreasoningaboutsoftware.TheseconstraintswillensurethatallknowledgeinusebycompetitionsystemsduringCGCeventsisgeneratedviaautomaticprocesses.Q7:WhatCPUarchitecturewillCGCrunon?A7:Forthepurposeofmaximizingaccessibilityandparticipation:Intelx86,32‐bit.Q6:Whatcompilerwillbeusedtobuildthebinaries?A6:CGCwilldistributeareferencecompilertoolchainpriortochallengekickoff.However,challengebinariesmaybeproducedbyanycompilerincludingthereferencecompiler.

1http://cwe.mitre.org/

Page 67: Cyber Grand Challenge - MIT Lincoln Laboratory · Q157 Added Jan 27, 2016 Q158 Added Mar 16, 2016 Q159 – Q161 Added Mar 16, 2016 Q162 – Q163 Added Mar 25, 2016 Q164 Added May

Q5:Duringthefinalevent,whathappenswhenmyCompetitionSystemfieldsanewChallengeBinary?A5:DuringCFE,inordertoenactdefenses,aCRSmaychoosetoreplaceaCBwithanewlysecuredversion.TofieldareplacementCB,aCRSmustsubmitthereplacementthroughanautomatedAPIoperatedbythecompetitionframework.ThecompetitionframeworkwilldeploythereplacementbinaryonbehalfoftheCRStoitsnetworkedhost.Additionally,thecompetitionframeworkwillmakeacopyofthereplacementCBavailabletoallcompetitorsystemsforthepurposesofconsensusevaluation(Shannon’sMaxim).Oncedeployed,replacementCBswillberequiredtofunctionasself‐containedreplacementswithoutcustomdependencies,libraries,etc.Q4:I'minterestedinadvancedapplicationdefenses.WillthesebepartofCGC?A4:DuringCFE,systemswillhavetheabilitytodeploynetworkdefensesaswellasapplicationdefenses.Todeployapplicationdefenses,competitionsystemsmayanalyzeCBsandfieldsecurereplacements.DuetothecompetitivenatureofCGC,DARPAexpectsthatcompetitorswillfieldmanyapproachesofvaryingtype,advancement,andefficacy.Q3:WhatlimitationsareimposedonreplacementCBsduringCFE?A3:DuringCFE,thecompetitionframeworkwillmonitortheavailabilityandcorrectfunctionofeachCB.IfaCRSdeploysreplacementCBsthatdegradeCBfunctionbyimpactingperformance,correctnessofCBresponses,ortheabilitytoservicenetworkrequests,anegativeimpactonscoringisexpected.SimilarconstraintswillbeimposedonreplacementCBsduringCQEscoring.Q2:IntheCGCRules,AreaofExcellence2specifiesAutonomousPatching.DoesthismeanaCyberReasoningSystem(CRS)isrequiredtoisolateandremoveflaws,ormayaCRSfieldanysecurereplacementChallengeBinary(CB)?A2:DuringtheCGCQualificationEvent(CQE)andFinalEvent(CFE),CBswillbeevaluatedbasedonavailability,correctfunction,andthemitigationofflaws,asdescribedintheCGCRulesandthisFAQ.NospecificrequirementsareimposedontheformulationmethodforsecurereplacementCBs.Q1:AreyouplanninganIndustryDayforcompetitors?A1:TwoCompetitorDaysessionsareplanned,oneontheEastCoast,andoneontheWestCoast.

‐TheEastCoastCompetitorDaysarecurrentlyscheduledforDecember3and4,2013attheDARPAConferenceCenter,675NorthRandolphStreet,

Page 68: Cyber Grand Challenge - MIT Lincoln Laboratory · Q157 Added Jan 27, 2016 Q158 Added Mar 16, 2016 Q159 – Q161 Added Mar 16, 2016 Q162 – Q163 Added Mar 25, 2016 Q164 Added May

Arlington,VA22203.Note:theseconddaywillbearepeatofthefirstdaytoaccommodateregisteredattendees.Availabilityisonafirst‐come‐first‐servedbasis.AllregistrationswillbefortheDecember3sessionuntilcapacityisreached;atthatpoint,registrationswillbefortheDecember4session.Pleasevisithttp://www.sa‐meetings.com/darpacgccompetitordayformoreinformationandtoregister.‐TheWestCoastCompetitorDayiscurrentlyscheduledforDecember9,2013attheWestinSt.Francis,335PowellSt,SanFrancisco,CA.Availabilityisonafirst‐come‐first‐servedbasis.Pleasevisithttp://www.sa‐meetings.com/darpacgccompetitordaywestformoreinformationandtoregister.