Network Security - ISA 656 Routing Securityastavrou/courses/isa_656_F07/... · Network Security - ISA 656 Routing Security Angelos Stavrou December 4, 2007. ... Bellovin’s “Security

Network Security - ISA 656Routing Security

Angelos Stavrou

December 4, 2007

What is Routing Security?

Routing Security

What is RoutingSecurity?

History of RoutingSecurity

Why So LittleWork?

How is it Different?

The Enemy’s Goal?

Routing Protocols

Routing in theInternet

Inter-ISP Routing

Link-Cutting Attack(Bellovin andGansner)

Defenses

Conclusions

2 / 41

■ Bad guys play games with routing protocols.

■ Traffic is diverted.◆ Enemy can see the traffic.

◆ Enemy can easily modify the traffic.

◆ Enemy can drop the traffic.

■ Cryptography can mitigate the effects, but notstop them.

History of Routing Security

Routing Security

Why So LittleWork?

The Enemy’s Goal?

Routing Protocols

Inter-ISP Routing

Defenses

Conclusions

3 / 41

■ Radia Perlman’s dissertation: Network Layer

Protocols with Byzantine Robustness, 1988.

■ Bellovin’s “Security Problems in the TCP/IPProtocol Suite”.

■ More work starting around 1996.

■ Kent et al., 2000 (two papers).

Why So Little Work?

Routing Security

Why So LittleWork?

The Enemy’s Goal?

Routing Protocols

Inter-ISP Routing

Defenses

Conclusions

4 / 41

■ It’s a really hard problem.

■ Actually, getting routing to work well is hardenough.

■ It’s outside the scope of traditionalcommunications security.

Routing Security

Why So LittleWork?

The Enemy’s Goal?

Routing Protocols

Inter-ISP Routing

Defenses

Conclusions

5 / 41

■ Most communications security failures happenbecause of buggy code or broken protocols.

■ Routing security failures happen despite goodcode and functioning protocols. The problemis a dishonest participant.

■ Hop-by-hop authentication isn’t sufficient.

The Enemy’s Goal?

Routing Security

Why So LittleWork?

The Enemy’s Goal?

Routing Protocols

Inter-ISP Routing

Defenses

Conclusions

6 / 41

Host A

Good: A−>X−>Y−>B

Bad: A−>X−>Z−>Y−>B

Host B

But how can this happen?

Routing Protocols

Routing Security

Routing Protocols

Normal Behavior

But Z Can LieUsing a Tunnel forPacket Re-injection

Why is the ProblemHard?

Inter-ISP Routing

Defenses

Conclusions

7 / 41

■ Routers speak to each other.

■ They exchange topology information and costinformation.

■ Each router calculates the shortest path toeach destination.

■ Routers forward packets along locally shortestpath.

■ Attacker can lie to other routers.

Normal Behavior

Routing Security

Routing Protocols

Normal Behavior

Inter-ISP Routing

Defenses

Conclusions

8 / 41

X−>A: Z(5), Y(5), B(15)

Z Host B

Host A10

Y−>X, Y−>Z: B(10)Z−>X: Y(5), B(15)

But Z Can Lie

Routing Security

Routing Protocols

Normal Behavior

Inter-ISP Routing

Defenses

Conclusions

9 / 41

Z−>X: Y(5), B(3)

Z Host B

Host A10

Y−>X, Y−>Z: B(10)

X−>A: Z(5), Y(5), B(8)

Note that X is telling the truth as it knows it.

Using a Tunnel for Packet

Re-injectionRouting Security

Routing Protocols

Normal Behavior

Inter-ISP Routing

Defenses

Conclusions

10 / 41

Host A

Y Host BQ

Why is the Problem Hard?

Routing Security

Routing Protocols

Normal Behavior

Inter-ISP Routing

Defenses

Conclusions

11 / 41

■ X has no knowledge of Z’s real connectivity.

■ Even Y has no such knowledge.

■ The problem isn’t the link from X to Z; theproblem is the information being sent. (Notethat Z might be deceived by some otherneighbor Q.)

Routing in the Internet

Routing Security

Routing Protocols

Routing in theInternetRouting in theInternetOSPF (OpenShortest Path First)

Characteristics ofInternal NetworksHow Do You SecureOSPF?AddressAuthorizationCertificateExternal Routing viaBGP

POP Topology

Noteworthy Points

Inter-ISP Routing

Defenses

Conclusions

12 / 41

■ Two types, internal and external routing.

■ Internal (within ISP, company): primarilyOSPF.

■ External (between ISPs, and some customers):BGP.

■ Topology matters.

OSPF (Open Shortest Path First)

Routing Security

Routing Protocols

POP Topology

Noteworthy Points

Inter-ISP Routing

Defenses

Conclusions

13 / 41

■ Each node announces its own connectivity.Announcement includes link cost.

■ Each node re-announces all informationreceived from peers.

■ Every node learns the full map of the network.

■ Each node calculates the shortest path to alldestinations.

■ Note: limited to a few thousand nodes atmost.

Characteristics of Internal Networks

Routing Security

Routing Protocols

POP Topology

Noteworthy Points

Inter-ISP Routing

Defenses

Conclusions

14 / 41

■ Common management.

■ Common agreement on cost metrics.

■ Companies have less rich topologies, but lesscontrolled networks.

■ ISPs have very rich—but veryspecialized—topologies, but well-controllednetworks.

■ Often based on Ethernet and its descendants.

How Do You Secure OSPF?

Routing Security

Routing Protocols

POP Topology

Noteworthy Points

Inter-ISP Routing

Defenses

Conclusions

15 / 41

■ Simple link security is hard: multiple-accessnet.

■ Shared secrets guard against new machinesbeing plugged in, but not against anauthorized party being dishonest.

■ Solution: digitally sign each routing update(expensive!). List authorizations in certificate.

■ Experimental RFC by Murphy et al., 1997.

■ Note: everyone sees the whole map;monitoring station can note discrepancies fromreality. (But bad guys can send out differentannouncements in different directions.)

Address Authorization Certificate

Routing Security

Routing Protocols

POP Topology

Noteworthy Points

Inter-ISP Routing

Defenses

Conclusions

16 / 41

■ Each router has certain interfaces and hencedirect network reachability

■ Each router therefore has a certificate bindingits public key to its valid addresses

■ Note well: the CA has to know the properaddresses for each router

■ But that’s the norm in OSPF environments

External Routing via BGP

Routing Security

Routing Protocols

POP Topology

Noteworthy Points

Inter-ISP Routing

Defenses

Conclusions

17 / 41

■ No common management (hence no metricsbeyond hop count).

■ No shared trust.

■ Policy considerations: by intent, not all pathsare actually usable.

POP Topology

Routing Security

Routing Protocols

POP Topology

Noteworthy Points

Inter-ISP Routing

Defenses

Conclusions

18 / 41

access router

access router access router access router

Noteworthy Points

Routing Security

Routing Protocols

POP Topology

Noteworthy Points

Inter-ISP Routing

Defenses

Conclusions

19 / 41

■ A lot of attention to redundancy.

■ Rarely-used links (i.e., R1→R2)Link cost must be carefully chosen to avoidexternal hops.

■ May have intermediate level of routers tohandle fan-out.

Routing Security

Routing Protocols

POP Topology

Noteworthy Points

Inter-ISP Routing

Defenses

Conclusions

20 / 41

InterISP Routing

Routing Security

Routing Protocols

Inter-ISP Routing

Path Vectors

PoliciesLong Prefixes andLoop-Free Routing

Longer Prefix Attack

Filtering

Secure BGP (Kentet al.)

Problems withSBGP

Certificate Issuance

Certificate TreeAuthorizationCertificates

Signed Origin BGP

Problems withSOBGP

Happy Packets

Defenses

Conclusions21 / 41

InterISP Routing

Routing Security

Routing Protocols

Inter-ISP Routing

Path Vectors

Filtering

Problems withSBGP

Signed Origin BGP

Problems withSOBGP

Happy Packets

Defenses

Conclusions22 / 41

■ “Tier 1” ISPs are peers, and freely exchangetraffic.

■ Small ISPs buy service from big ISPs.

■ Different grades of service: link L-Z is forcustomer access, not transit. C→B goes viaL-Y-X-W, not L-Z-W.

■ A is multi-homed, but W-A-Z is not a legalpath, even for backup.

■ BGP is distance vector, based on ISP hops.Announcement is full path to origin, not justmetric.

Path Vectors

Routing Security

Routing Protocols

Inter-ISP Routing

Path Vectors

Filtering

Problems withSBGP

Signed Origin BGP

Problems withSOBGP

Happy Packets

Defenses

Conclusions23 / 41

■ Route advertisements contain a prefix and alist of ASs to traverse to reach that prefix

■ Example: if B owns address block 10.0/16, Lwould see 〈10.0/16, {Y,X,W,B}〉

■ ASs do not see paths filtered by upstreamnodes. Y sees 〈10.0/16, {X,W,B}〉 and〈10.0/16, {Z,W,B}〉; since only forwards theformer to L, L knows nothing of the path via Z

Policies

Routing Security

Routing Protocols

Inter-ISP Routing

Path Vectors

Filtering

Problems withSBGP

Signed Origin BGP

Problems withSOBGP

Happy Packets

Defenses

Conclusions24 / 41

■ ISPs have a great deal of freedom whenchoosing the “best” path

■ While hop count is one metric, local policies(i.e., for traffic engineering) count more

■ These policies — in general, not disclosedpublicly — affect with path neighbors will see

Long Prefixes and Loop-Free Routing

Routing Security

Routing Protocols

Inter-ISP Routing

Path Vectors

Filtering

Problems withSBGP

Signed Origin BGP

Problems withSOBGP

Happy Packets

Defenses

Conclusions25 / 41

■ Routers ignore advertisements with their ownAS number in the path

■ This is essential to provide loop-free paths

■ Routers use longest match on prefixes whencalculating a path

■ These two facts can be combined to form anattack

Routing Security

Routing Protocols

Inter-ISP Routing

Path Vectors

Filtering

Problems withSBGP

Signed Origin BGP

Problems withSOBGP

Happy Packets

Defenses

Conclusions26 / 41

■ Suppose B owns 10.0/16. Z sees 〈10.0/16,{W,B}〉

■ A advertises 〈10.0.0/17, {A,W}〉

■ Z will route packets for 10.0.0/17to A — ithas a longer prefix

■ W will never see that path, and hence won’tpass it to B — the path (falsely) contains W,so it will be rejected by W

Filtering

Routing Security

Routing Protocols

Inter-ISP Routing

Path Vectors

Filtering

Problems withSBGP

Signed Origin BGP

Problems withSOBGP

Happy Packets

Defenses

Conclusions27 / 41

■ ISPs can filter route advertisements from theircustomers.

■ Doesn’t always happen: AS7007 incident,spammers, etc.

■ Not feasible at peering links.

Secure BGP (Kent et al.)

Routing Security

Routing Protocols

Inter-ISP Routing

Path Vectors

Filtering

Problems withSBGP

Signed Origin BGP

Problems withSOBGP

Happy Packets

Defenses

Conclusions28 / 41

■ Each node signs its announcements.

■ That is, X will send {W}X , {Y }X , {Z}X .

■ W will send{B}W , {A}W , {X}W , {X : {Z}X}W .

■ Chain of accountability.

Problems with SBGP

Routing Security

Routing Protocols

Inter-ISP Routing

Path Vectors

Filtering

Problems withSBGP

Signed Origin BGP

Problems withSOBGP

Happy Packets

Defenses

Conclusions29 / 41

■ Lots of digital signatures to calculate andverify.◆ Can use cache

◆ Verification can be delayed

■ Calculation expense is greatest when topologyis changing—i.e., just when you want rapidrecovery. (About 120K routes. . . )

■ How to deal with route aggregation?

■ What about secure route withdrawals whenlink or node fails?

■ Dirty data on address ownership.

Routing Security

Routing Protocols

Inter-ISP Routing

Path Vectors

Filtering

Problems withSBGP

Signed Origin BGP

Problems withSOBGP

Happy Packets

Defenses

Conclusions30 / 41

■ Who issues prefix ownership certificates?

■ Address space comes from upstream ISP orRIRs

■ RIRs really are authoritative — hence they’re amonopoly

■ If an RIR makes a mistake, the prefix is off theair

■ Is this a risk worth taking?

Certificate Tree

Routing Security

Routing Protocols

Inter-ISP Routing

Path Vectors

Filtering

Problems withSBGP

Signed Origin BGP

Problems withSOBGP

Happy Packets

Defenses

Conclusions31 / 41

■ The RIRs (Regional Internet Registries) giveaddresses to big ISPs and big end users

■ Accordingly, the RIRs should issue certificates

■ (Really, it should be ICANN, but the politics ofthat are too painful)

■ Small ISPs and small customers get addressspace from their own ISPs

■ Every ISP is thus a certificate holder and acertificate issuer

■ These are authorization certificates, notidentity certificates

Authorization Certificates

Routing Security

Routing Protocols

Inter-ISP Routing

Path Vectors

Filtering

Problems withSBGP

Signed Origin BGP

Problems withSOBGP

Happy Packets

Defenses

Conclusions32 / 41

■ The identity of the certificate holder isirrelevant

■ What matters is the authorization: thecertificate contains IP address ranges

■ The signing party has its own certificate listinglarger ranges of IP addresses, and hence theright to delegate them

Signed Origin BGP

Routing Security

Routing Protocols

Inter-ISP Routing

Path Vectors

Filtering

Problems withSBGP

Signed Origin BGP

Problems withSOBGP

Happy Packets

Defenses

Conclusions33 / 41

■ Suppose only the origin was digitally signed:〈10.0/16, B〉

■ In addition, all polices are (securely) publishedin some database

■ Receiving node verifies origin, then comparesreceived path against all policies

■ Query: is the received path consistent withpolicies?

■ Advantage: many fewer signatures

Problems with SOBGP

Routing Security

Routing Protocols

Inter-ISP Routing

Path Vectors

Filtering

Problems withSBGP

Signed Origin BGP

Problems withSOBGP

Happy Packets

Defenses

Conclusions34 / 41

■ Sill have monopoly RIRs

■ ISPs don’t like to publish policies

■ Clever attackers can play games in the middleof the path

Happy Packets

Routing Security

Routing Protocols

Inter-ISP Routing

Path Vectors

Filtering

Problems withSBGP

Signed Origin BGP

Problems withSOBGP

Happy Packets

Defenses

Conclusions35 / 41

■ Philosophy: don’t worry too much aboutrouting security

■ Crucial metric: do packets reach theirdestination?

■ What about confidentiality? If it matters,encrypt end-to-end

■ But what about traffic analysis?

Link-Cutting Attack (Bellovin and

Gansner)Routing Security

Routing Protocols

Inter-ISP Routing

Is Link-CuttingFeasible?Sample Link-CuttingAttackCost of Link-CuttingAttacks on theBackbone

Defenses

Conclusions

36 / 41

■ Suppose that we have SBGP and SOSPF.

■ Suppose the enemy controls a few links ornodes. Can he or she force traffic to traversethose paths?

■ Yes. . .

Is Link-Cutting Feasible?

Routing Security

Routing Protocols

Inter-ISP Routing

Defenses

Conclusions

37 / 41

■ Attacker must have network map.Easy for OSPF; probably doable for BGP—see“Rocketfuel” paper.

■ Can attacker determine peering policy?Unclear.

■ How can links be cut?Backhoes? “Ping of death”? DDoS attack onlink bandwidth?

Sample Link-Cutting Attack

Routing Security

Routing Protocols

Inter-ISP Routing

Defenses

Conclusions

38 / 41

Wb0Wb1

Xa1Xa2

Ya2Ya3

Cost of Link-Cutting Attacks on the

BackboneRouting Security

Routing Protocols

Inter-ISP Routing

Defenses

Conclusions

39 / 41

0 20 40 60 80 100 120 140 160 180

Number of nodes

"cut-effort-full""cut-effort-reduced"

"cut-avg-full""cut-avg-reduced"

Defenses

Routing Security

Routing Protocols

Inter-ISP Routing

Defenses

Conclusions

40 / 41

■ Hard to defend against—routing protocols aredoing what they’re supposed to!

■ Keeping attacker from learning the map isprobably infeasible.

■ Feed routing data into IDS?

■ Link-level restoration is a good choice, but canbe expensive.

■ Others?

Conclusions

Routing Security

Routing Protocols

Inter-ISP Routing

Defenses

Conclusions

41 / 41

■ Routing security is a major challenge.

■ Mentioned specifically in White HouseCybersecurity document.

■ Lots of room for new ideas.

Network Security - ISA 656 Routing Securityastavrou/courses/isa_656_F07/... · Network Security - ISA 656 Routing Security Angelos Stavrou December 4, 2007. ... Bellovin’s “Security

Documents

CASE STUDY Internal Security Assessor (ISA) Training SSC ISA...

ISA 562 1 Access Control ISA 562 Internet Security Theory &....

OU INFORMATION SECURITY & RISK MANAGEMENT ISA – February.....

ISA 656 F07 Review

ISA 662 Internet Security Protocols Kerberos

ISA 662 Information System Security

Network Security - ISA 656 Viruses, Trojan Horses, and...

ISA Security Compliance Institute

ISA Server 2004 Security Hardening Guide

ISA 562 Summer 2008 1 Information Security Management CISSP....

ISA 562 1 Topic 9: Operations Security ISA 562 Internet...

Network Security - ISA 656 Wireless Security

Information Security Awareness (ISA) in Everyday School Life

Cisco ISA 3000 Industrial Security Appliance에 대한...

Microsoft® · Web viewInternet Security and Acceleration.....

SMART Manufacturing & Cyber Security: Foundational ISA/IEC.....