Securing DSL
Without a well-thought out security strategy, “always on” DSL Internet connections can translate into “inherently vulnerable.”
BY RANDY DAY
[This article originally appeared in the January, 2000 issue of Information Security Magazine (see www.infosecuritymag.com). The hardcopy article includes a sidebar on firewall products.]
In his keynote speech at the DSLcon’99
convention in Amsterdam last November, Anthony Alles remarked, “As DSL becomes
more widely deployed, increasing numbers of DSL subscribers
have reported attacks on their computers,
sometimes leading to copying or destruction of
sensitive data.” For millions of would-be DSL
users, Alles’s comments cut to the heart of one of the biggest uncertainties
about broadband Internet connections: security.
The growth rate of DSL
has been phenomenal. From an installed base of a few thousand at the start of
1998, DSL rollouts in the United States alone will reach about 600,000 by the
end of 1999, according to DSL Prime. Though DSL is still a new kid on the
block, it is quickly becoming an accepted connection technology for business
and residential customers alike, with speeds up to and even slightly greater
than T1 speeds. T1 connections, with their rapid service response time
guarantees, are still preferred for mission-critical applications. But for most
other broadband applications, DSL is an attractive solution.
What
About Cable Modems?
Before we dive into the key issues involved in
securing DSL, let me explain the intended focus of this article. While much of
the following discussion is relevant to both DSL and cable modems, the emphasis
is on DSL for a number of reasons. First, DSL is targeted to both consumers and
businesses, while cable modems are more prevalent in consumer markets. Second,
many cable modem implementations have high download speeds but rather limited
upload speeds, which limits the usefulness for many business applications.
Third, cable modems, because of the shared medium, have special security issues
that are addressed (at least partially) by specific hardware solutions in cable
modems. And finally, focusing in on one broadband technology makes it easier to
use specific examples.
Even with this
narrowed focus, it would be impossible to cover all security-related DSL issues
in one article. Hence, the discussion here is limited to the security issues
involved in the DSL connection itself as well as different network topologies
used to hook up a DSL connection.
Security
Risks
A DSL (or cable modem) connection to the
Internet is a greater security risk than a plain old analog modem dialup
connection. For one, the bandwidth is greater, allowing the possibility of more
cracking to be done in the same period of time. More importantly, the
connection is usually “always on,” which makes the host a much easier (and
potentially more lucrative) target to find. Finally, depending on the DSL
connection type, there may be a more porous door between the client and the ISP.
Let’s look at some
traditional security risks to see how a DSL connection is susceptible to--as
well as which risks we can diminish through proper engineering of a DSL network
connection.
Eavesdropping. Eavesdropping risks
arise when an outsider is able to listen in on your
local network traffic. This
information-gathering may range from spying on file transfers around your
network to gathering information about the network configuration itself, such
as the names of users, computers and other system resources. Eavesdropping may
directly compromise your network (e.g. spying on network file transfers), or
may be the source of information that could aid further intrusion attempts
(e.g. knowing the system resources to attack).
In the most pathological extreme, it’s possible (and it’s happened) to
build a DSL-connection that forces all traffic, even intra-LAN local traffic,
over the DSL connection. With the
higher speeds offered by DSL connections, the casual user might not even notice
this situation. More sophisticated DSL connection techniques can dramatically
reduce the eavesdropping threat.
Probing. Probing is a term that arises from the
practice of examining a host and testing its response to various probes. A
typical probe attack involves an automated program that scans a host’s TCP/UDP
ports, looking for running services that respond to queries. The attacker can
then attempt to exploit weaknesses in any responding services. For example, the
infamous Internet Worm, developed by Robert Morris and unleashed Nov. 2, 1988,
used a number of techniques to breach system security. One mechanism it used
was to probe hosts, looking for running instances of the “finger” program; once
found, the worm would try to exploit an unknown buffer overrun bug in many versions
of the finger program, causing the worm’s code to be executed on the target
host.
Common probing attacks
today include probing for sendmail servers; spammers love to find sendmail
servers that will send mail from any submitting host. Another common probe is
for Microsoft Windows file and print servers, which are subsequently queried,
looking for unprotected files. A
DSL-connected LAN, with it’s permanent connection, is an attractive target for
a probing attack. Porous DSL
connections can leak a constant stream of information to outside probers,
giving them the first toehold your LAN security walls.
Impersonation. Impersonation can
take a number of different forms, but one example is IP
address stealing. For example, depending on your
network configuration, it may be possible for another DSL subscriber to “steal”
your IP address, causing all information addressed to your computer (including,
say, the results of your online banking statement request), to pass instead to
the attacker’s host. In reality, IP address stealing is usually an impractical
attack, but it highlights two fundamental principles for security on the
Internet: (1) Any security mechanism based on IP addresses is flawed; (2)
Assume that any plain text sent over the Internet is open to prying
eyes.
Unlike
a PPP dialup connection, or a routed T1 connection, a bridged DSL connection is
susceptible to IP address stealing, as we’ll see later.
Denial-of-service. DoS attacks don’t
attempt to obtain private information or to
delete or corrupt files on attacked hosts.
Rather, they attempt to make a host unusable by keeping it too busy performing
useless functions. Many simple DoS
attacks don’t work well over a dialup connection because it’s just not possible
to send enough traffic over an analogue modem to have a serious negative impact
on a modern PC. However, with a
high-speed DSL connection, that’s always on, it becomes possible to generate
serious DoS attacks.
Viruses. Virus attacks are familiar to many:
they occur when a file, placed on a host and executed, performs harmful
actions, including attempting to spread itself to other hosts on the network.
DSL connections, even
with a secure network topology as discussed later, do not have any innate
protection against virus attacks, and ISPs have started to recognize this
opportunity. US West, for example, is reportedly offering a service to
virus-check incoming e-mail with Trend Micro software, charging $1.50 a month
for this value-added service. Good virus protection should be a primary concern
for any network, especially a network with an Internet connection.
Connection
Types
The first factor influencing the security of a
DSL connection is the type of connection used by the provisioning ISP. For this
discussion, the underlying DSL physical layer technology--SDSL, ADSL, RADSL,
etc.--is immaterial. Regardless of the xDSL technology used, ISPs need to build
a network topology on top of the DSL bit pipe. Options include a bridged
connection, a routed connection, or a Point-to-Point Protocol (PPP) connection,
either PPPoA (PPP over ATM) or PPPoE (PPP over Ethernet).
Bridged connection. The connection type
with the least inherit security is a bridged connection. A bridged connection
is a DSL modem with an Ethernet interface that logically acts as an Ethernet
bridge between the client device and the ISP’s network. As such, the DSL modem
in the bridged case acts exactly like a standard Ethernet bridge between two Ethernet
segments: the client’s and the ISP’s.
A bridge is a Layer 2
device; a basic bridge passes all traffic from one Ethernet segment to the
other Ethernet segment. With a bridging DSL modem plugged into a hub on the
client site, potentially all traffic on that hub--including local PC-to-PC
traffic--would pass over the bridge onto the ISP’s network.
In practice, all
bridged DSL modems are “learning bridges.” A learning bridge listens to the
Ethernet traffic to determine which MAC addresses are on which side of the
bridge. A learning bridge does not pass traffic across the bridge if the
destination MAC is on the same side as the source host. Consequently, a
learning bridge would not pass local PC-to-PC traffic across the bridge,
thereby reducing the amount of unnecessary traffic on the DSL link and
improving the security of client-side LAN traffic.
Bridged connections of
this type are infamous for their susceptibility to eavesdropping. Ethernet
broadcast traffic passes across a learning bridge in both directions. If a DSL
bridging modem is plugged into a LAN with multiple Windows hosts, and if those
hosts have Windows networking turned on and file and print sharing enabled, and
if the ISP is not filtering Windows networking ports (ports 137, 138 and 139),
it is possible for the local hosts to appear in the “Network Neighborhood” of
other customers on the same Ethernet segment at the ISP.
A bridged connection
is also more susceptible to IP address stealing than other DSL connection
types. Imagine the following scenario: Customer “Bob” has a bridging DSL
connection to “AnyISP.” AnyISP has specified that its DSL customers should be
using DHCP to get their IP configuration information. Bob sets his PC up
correctly and obtains, via the DHCP protocol, the IP addresses 206.128.1.3.
Meanwhile, customer
“Eve,” who also uses AnyISP and happens to be on the same logical Ethernet
segment as Customer Bob, has decided (perhaps accidentally, perhaps
maliciously) to manually configure her PC with the IP address 206.128.1.3. When
traffic from the Internet comes into AnyISP addressed to Bob at 206.128.1.3,
AnyISP has to determine which MAC address to send the traffic to. AnyISP will
ARP for address 206.128.1.3, and if Eve replies first, then her MAC address
will be entered into AnyISP’s ARP table as the proper endpoint. So, Eve will
get Bob’s traffic. Of course, Bob will quickly figure out that he’s not getting
any packets and call AnyISP to complain. If Eve was particularly smart, she
could capture all of the traffic destined for Bob, read it, and then pass it on
direct to Bob’s MAC address. Bob might never find out that Eve is eavesdropping
on his traffic.
Routed connections. A routed connection
is a DSL connection in which Layer 3 IP routing is used to connect the customer
to the ISP. In other words, the DSL modem functions as an IP router. The DSL
modem has its own IP address in this scenario, one of subnet of addresses assigned to the customer For this reason,
routed connection types are rare for residential customers.
The security
characteristics of a routed connection are similar to those of the Internet at
large. Local traffic, even LAN broadcasts, is not passed over the router to the
ISP. A Windows Network will not show up on external “Network Neighborhood”
lists unless special efforts are made to advertise its existence (e.g., sending
information to an external WINS server). It’s also difficult for others to
steal an IP address, since the ISP generally knows which IP subnet ranges are
assigned to which customers and has his routers configured accordingly.
PPP connections. PPPoA and PPPoE are
mechanisms to build a Point-to-Point Protocol logical link over ATM or
Ethernet, respectively. The PPP was originally conceived for building an
endpoint-to-endpoint data link layer over a simple construct like a serial
line. As such, PPP is the standard data link layer for dial-up modems. PPPoA
and PPPoE extend the PPP mechanism to more complex or shared networking
constructs like ATM or Ethernet. Because they fit in with the existing administrative,
billing and security infrastructure built around PPP, PPPoA and PPPoE are
finding favor with many ISPs.
Their security
characteristics of PPPoA and PPPoE are similar. Typically, as part of the PPP
link establishment, some sort of ID/password is required. The ISP can use this
credential set for authorization and billing purposes. Like the routed
connection type, a Windows Network will not show up on external “Network
Neighborhood” lists unless special efforts are made to advertise its existence.
The IP address is typically assigned as part of the link establishment in such
a way that the ISP knows which IP addresses are where. This makes IP address
stealing a difficult task.
PPPoA has an
additional twist: Since ATM is usually not extended to the PC environment, the
DSL modem device itself establishes the PPP link. This means the DSL device has
an IP address and is one end of a PPP link. If the ISP is providing only one IP
address to the customer, the DSL modem must run a NAT server in order to
service the customer’s PC(s): all customer access must pass through the NAT
server. In addition, the NAT server must run in the DSL modem, meaning that
customer has no choice over the NAT implementation used.
Security
Techniques
There are three basic ways to hook up a LAN to a
DSL connection: a direct connection, a multihomed gateway host, and a turnkey
firewall solution. These connections also apply to a single PC hookup.
Direct connection. One mechanism for
sharing a DSL modem is to connect it directly to your local LAN hub and to use
routable (nonprivate) IP addresses for each of your hosts. This can work for
bridged connections, routed connections and PPPoE connections. Figure 1 depicts
a shared bridging DSL modem; the routing DSL modem is similar. Figure 2 depicts
the analogous shared hub for a PPPoE DSL connection.

Figure 1: Direct Hub Connection, Bridging DSL
Modem
While this is a conceptually simple method for
sharing a DSL connection, it requires a fair amount of diligence if you are
concerned about the security of the local hosts. For example, in the bridging
case, the only thing preventing your local LAN traffic from zipping around your
ISP’s LAN is an Ethernet learning bridge. While a bridge will not forward
traffic destined to local MAC addresses, it will forward multicasts and
broadcasts. A malicious cracker could make use of this information to target
your systems. If the ISP is using DHCP to assign IP addresses, it is possible
(but unlikely) that each PC will be assigned an IP number on different IP
subnets. If this happens, then even local PC to PC LAN traffic will be routed
over the DSL bridge, to the ISP’s router, before returning back over the DSL
line to the local LAN. This is highly
insecure to say nothing of slow and wasteful of the DSL link’s bandwidth.
In addition, if you
are using more broadcast-chatty protocols on your local LAN, like NetBEUI, you
are passing even more information onto the ISP’s system. A learning bridge is not a security device; it is a
traffic-limiting device, designed to keep your local LAN traffic from swamping
the rest of the network. Moreover, a configuration like this leaves every host
equally open to attacks from the outside. Each host has to be secured
independently and equally if you want to prevent a weak link in your system.
Finally, some security techniques,such as firewalls, are not available in a
configuration like this.

Figure 2: Direct Connection in PPPoE Environment
As Figure 2
demonstrates, each host in a PPPoE connection should be configured with
multiple logical interfaces--one for the PPPoE link, one for the local
Ethernet. The <dashed blue lines> in the figure represent the PPPoE link
to the ISP, which has an IP address but is physically passed over the single
Ethernet interface in each host. If only the PPPoE interface is configured,
intra-LAN PC-to-PC traffic will be forced to flow over the PPPoE link to the
ISP and back, which is both insecure and slow.
One alternative with a
direct connection is to run two different networking protocols, for example
TCP/IP and NetBEUI. NetBEUI, while it
is broadcast-chatty, is a non-routable protocol, so it will not be passed by
the ISP’s router. Each PC can then be configured
to disable Windows file and print sharing on the TCP/IP protocol, which
provides some protection. However, the
protection is incomplete, because it may still be possible for other clients of
your ISP (but not other Internet users beyond your ISP) to access your file and
print server over the NetBEUI protocol.
There is nothing
inherently wrong with the direct hub configuration: it is a perfectly valid
configuration, depending on your security needs.
However, most companies with a concern about the data on their local LAN would
not use a configuration like this.
Dual-homed host. In a dual-homed
environment, a PC is configured as a gateway between the local LAN and the DSL
modem. With a bridged or routed DSL connection, this gateway can be configured
as either a router or a NAT/proxy server. With a PPPoE DSL connection, this gateway
would have to be configured as a NAT/proxy server (because the PPPoE protocol
won’t pass over
the gateway).

Figure 3: Dual-Homed Gateway
Figure 3 depicts the
dual-homed host configuration with the gateway configured as a NAT/proxy
server. The dual-homed host solution completely isolates your local LAN traffic
from the ISP’s network. In addition, it provides a central point of control
over the Internet connection. Any probing attacks have to pass through the
gateway, where filters can be set to block various ports. Likewise, any traffic
going out to the Internet has to pass through the gateway host, where it can be
logged or filtered with the appropriate firewall software.
In addition, with a
NAT/proxy server on the gateway, there is a reasonable level of protection to
other hosts in the local LAN, which do not have to be watched/secured as
tightly. A stateful NAT server provides a good level of protection to the
interior LAN. By allowing most outgoing traffic, but blocking all incoming
ports unless explicitly opened, a NAT server is the basis for most firewall
implementations. With a NAT server blocking all incoming ports, most probing
attacks fail. A NAT server also allows multiple interior hosts to share a
single external routable IP address. This can reduce costs by eliminating the
need for multiple IP addresses from the ISP, and it increases security by
allowing private address space IP numbers (nonroutable IPs) on the interior
network. A proxy server operates in a similar fashion on the application layer.
With a dual-homed
gateway in a Windows environment, it is also possible to disable all Windows
networking on the outside interface, which blocks a common attack point. To
disable all Windows networking on the outside interface in NT, open the Control
Panel, Network, Bindings and unbind the NetBIOS, Server and Workstation
services from the WINS client (TCP/IP) for the outside Ethernet interface. This
eliminates all Windows networking file access through the NIC connected to the
DSL: you won’t have to worry about anyone making a Network Neighborhood
connection to shares on your gateway or interior hosts. Moreover, unbinding
file and printer services from the outside interface does not eliminate file
and printer sharing through the local LAN side of your gateway machine.
The dual-homed
solution also helps protects the LAN against denial-of-service attacks, since
DoS attacks are often stopped by the NAT/proxy server. It’s hard for DoS
attacks to proceed against unaddressable hosts on the local LAN.
One drawback to the
dual-homed environment is that the gateway host, as the gendarme
protecting the LAN, becomes an especially
attractive target for attackers. Remember, the firewall provided by the
dual-homed host is only as secure as the host it runs on. As such, the gateway
should be hosted by an operating system with security capabilities. For the
Windows family, this means using Windows NT/2000; Windows 9x would not make a
good gateway host.
Turnkey
Firewall Solutions
The turnkey firewall solution is a variation on
the dual-homed PC solution. However, instead of a PC with NAT/proxy software, a
turnkey firewall device (e.g., the SonicWALL device from Sonic Systems) is used
as the gateway device (see sidebar, p. XX).
The major advantage to the turnkey solution is the ease of implementation: The
device is just dropped in, with one side wired to the DSL and the other
connected to the local LAN. These devices usually provide NAT or proxy
capabilities as well as logging and reporting facilities. In addition, they
often include content filtering capabilities, which allow the blocking of Web
sites based on content. Turnkey solutions are usually robust (no moving parts),
and the hassle-factor is low.
The downside to a
turnkey solution is that they have a definite cost (a dual-homed host can often
be established with existing equipment), and they are nonextensible. However,
in a business environment, the equipment cost is usually minimal, while the
payoff--including the ease of implementation and added logging/reporting and
filtering capabilities--is high.
Conclusion
It’s clear that the level of security in DSL
depends on several factors: the type of connection, the network topology, host
and/or gateway configurations, and the use of third-party tools such as firewalls
and proxies. As ISPs have grown their DSL rollouts, they have struggled with
scalability problems with birdged DSL connection types. For example, US West is converting their DSL
offering from a bridged connection to a PPPoA connection. This, combined with a maturing DSL equipment
market probably means that future DSL connection devices will be more highly
integrated devices, perhaps combining DSL signaling capabilities with some sort
of turnkey firewalling capabilities.
DSL “modems”, as they reach wider audiences, will become more of a plug
and play appliance, and part of the feature set of that appliance will be some
base level of security for the DSL connection.
Randy
Day (randyday@tuketu.com) is the prime
motivating force behind Tuketu.com, providing network and software development
consultation and is currently based in Seattle, Wash.
(SIDEBAR
ON FIREWALLS TO COME)
(BOX)
FYI
DSL
Info
DSL
Prime [since they are mentioned in the article and it’s a good site]
DSL
Reports
www.dslreports.com
ADSL Forum
www.adsl.com
Avalon Trials
http://dsl.avalon.net
Computing Central Bandwidth Forum
http://computingcentral.com/forums/bandwidth
Dan Kegel’s ADSL Page
http://alumni.caltech.edu/~dank/isdn/adsl.htm
David
Fannin’s ADSL for Linux
http://metalab.unc.edu/LDP/HOWTO/mini/ADSL.html
The
DSL Source Book
Available free from Paradyne Corp:
www.paradyne.com/sourcebook_offer/sb_html.html
Everything DSL
www.everthingdsl.com
High Bandwidth Web Page
www.specialty.com/hiband/
Network World DSL Resources
www.nwfusion.com/dsl/
Randy Day’s xDSL Page
www.tuketu.com/dsl/xdsl.htm
Universal ADSL Working Group
www.uawg.com
xDSL Deployment Worldwide
http://conk.com/world/dsl
xDSL FAQ
http://homepage.interaceess.com/~jkristof/xdsl-faq.txt