Robert Lychev, Georgia Tech, Sharon Goldberg, Boston University

Transcription

Robert Lychev, Georgia Tech, Sharon Goldberg, Boston University
ARE WE THERE YET? QUANTIFYING BGP SECURITY IN PARTIAL DEPLOYMENT
Robert Lychev
Sharon Goldberg
Michael Schapira
Boston University
Hebrew University
Georgia Tech
Abstract
The Border Gateway Protocol (BGP) sets up routes between the smaller
networks that make up the Internet. In the protocol, each autonomous
system (AS) --- an independent network administrated by a single profitseeking entity --- maintains a list of possible routes to various
VZW:
ISP 1:
Level3, VZW, 22394
66.174.161.0/24
(Level3, VZW, 22394, Prefix)
ISP 1
after securing just one more node AS 31283,
AS 2905 switches to a route through attacker AS 3340
reachability information
traffic flow
Level 3
Verizon
Wireless
China
Telecom
VZW:
(22394, Prefix)
22394
Level3: (VZW, 22394, Prefix)
VZW:
malicious router cannot announce a direct
route to 22394, since 22394 never said
(22394, Prefix)
ChinaTel:
(22394, Prefix)
The Internet has about 40K ASes with
competing interests, so not all of them
can be secured at the same time!
before AS 31283 deploys S-BGP
after AS 31283 deploys S-BGP
and when security is second or third
…
Who must be secured first?
What are our objectives
and obstacles?
Background and Motivation
reachability
information traffic flow
(22394, Prefix)
Level3: (VZW, 22394, Prefix)
destinations. Such reachability information is disseminated globally as
each AS selects its most preferred routes to the corresponding
destinations and advertises them to its neighboring ASes. Despite its
crucial role, BGP is notoriously vulnerable to serious problems, including
propagation of bogus routing information due to attacks or
misconfigurations and network instabilities in the form of persistent
routing oscillations.
The S*BGP protocols (Secure BGP, secure origin BGP, BGPsec, etc)
were proposed to address these issues. However, even if S*BGP
protocols were to ever reach complete Internet-wide deployment, they
are bound to go through a phase of incremental deployment, where ASes
decide if and when to deploy such protocols. In this initial phase, secure
and non-secure versions of BGP will have to coexist. This coexistence of
BGP and S*BGP gives rise to many unanswered questions about BGP
security under partial deployment (i.e. when some ASes have deployed
S*BGP, but others have not).
By combining theoretical analyses with multiple large-scale
simulations on empirical data, we address many important questions such
as
``What security guarantees are achievable under partial deployment and
how to quantify them?'',
``What new vulnerabilities are introduced into the interdomain routing
system?'',
``Under what conditions is S*BGP guaranteed to converge even in
presence of attackers?'',
``Which ASes are the most important to secure?'', and many more.
By word of mouth smaller networks such as ATT, BU, Verizon,
etc., use BGP to set up routes to various destination IP prefixes.
This is what glues the Internet together!
As security is given higher priority, it affects
more ASes, but when security is second or first
…
with a blackhole attack,
AS 4559 attracts AS 111
that used a secure route
before the attack
HPC Provider
in Texas
Our objective is to minimize the amount of traffic that an attacker
attracts with its attack.
peers
exchange
traffic for free
Theorem: The problem of minimizing the number of ASes that a
particular attacker attracts with its attack is NP-hard .
*Remark: In reality, our job is harder because we do not know
where the attacker is.
ISP 1
Boston University
Inconsistent prioritization of security
across secure ASes can lead to BGP Wedgies
Level 3
Verizon
Wireless
China
Telecom
Telecom
provider in
southeast Asia
and unpredictable network behavior
VZW, 22394
66.174.161.0/24
Level3, VZW, 22394
66.174.161.0/24
that is difficult to debug.
22394
before the blackhole attack
after the blackhole attack
22394
66.174.161.0/24
When selecting a favorite route, every node must consider: local
preferences (business relationship with its neighbor on that route)
and route length. Priority for security can be considered before
local preferences (first), after route length (third), and in between
(second).
Unfortunately, to attract/intercept traffic, a malicious AS can
pretend to connect to a valid origin directly. A similar incident
happened in April 2010 involving China Telecom and some US
government sites.
Swedish ISP places
security below local
preferences
$
customer routes
are preferred
to provider routes
Level3, VZW, 22394
66.174.161.0/24
Theorem: security non-monotonicity and protocol
downgrades do not occur when security is 3rd and 1st
respectively, for networks that satisfy well-known
properties such as GR conditions [1].
customer pays its provider
reachability information
traffic flow
ChinaTel, 22394
66.174.161.0/24
peer routes are preferred to provider routes
Empirically, non-monotonicity and protocol downgrades do not
seem to pose a major issue in general. This suggests that
higher prioritization of security may result in more benefit to the
Internet overall.
Norwegian ISP prioritizes
security above
everything
Level 3
China
Telecom
Verizon
Wireless
22394
Average Fraction of Sources
ISP 1
Sec 1st
Sec 2nd Sec 3rd
Sec 1st
Sec 2nd
Sec 1st
Sec 2nd
Sec 3rd
does not deploy S-BGP
Using Public Key Cryptography, S-BGP prevents ASes from
announcing prefixes they do not own as well as routes that were not
announced to them. This requires Public Key Cryptography (PKI).
When top 12 tier-2 ASes and their biggest Stubs are secured, ~3000 ASes in total.
Theorem: when all secure ASes prioritize security the same
way, S-BGP converges to a unique stable state, for
networks that satisfy well-known properties such as GR
conditions [1].
References:
[1] L. Gao and J. Rexford. Stable Internet routing without global coordination. Trans. On Networking, 2001.
[2] On Blind Mice and the Elephant, An Aqua Lab Project,
www.aqualab.cs.northwestern.edu/projects/BlindMice.html.