In this paper I have tried to include the following questions:

	- What is a denial of service attack?
	- Why would someone crash a system?
	- How can someone crash a system.
	- How do I protect a system against denial of service attacks?
	


.A. INTRODUCTION
~~~~~~~~~~~~~~~~

.A.1. WHAT IS A DENIAL OF SERVICE ATTACK?
-----------------------------------------

Denial of service is about without permission knocking off
services, for example through crashing the whole system. This
kind of attacks are easy to launch and it is hard to protect
a system against them. The basic problem is that Unix
assumes that users on the system or on other systems will be
well behaved. 

.A.2. WHY WOULD SOMEONE CRASH A SYSTEM?
---------------------------------------
 
.A.2.1. INTRODUCTION
--------------------

Why would someone crash a system? I can think of several reasons
that I have presentated more precisely in a section for each reason,
but for short:

	.1. Sub-cultural status.
	.2. To gain access.
	.3. Revenge.
	.4. Political reasons.
	.5. Economical reasons.
	.6. Nastiness.

I think that number one and six are the more common today, but that
number four and five will be the more common ones in the future.

.B. SOME BASIC TARGETS FOR AN ATTACK
~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~

.B.1. SWAP SPACE
----------------

Most systems have several hundred Mbytes of swap space to 
service client requests. The swap space is typical used
for forked child processes which have a short life time.
The swap space will therefore almost never in a normal
cause be used heavily. A denial of service could be based 
on a method that tries to fill up the swap space.

.B.2. BANDWIDTH
---------------

If the bandwidth is to high the network will be useless. Most
denial of service attack influence the bandwidth in some way.

.B.3. KERNEL TABLES
-------------------

It is trivial to overflow the kernel tables which will cause
serious problems on the system. Systems with write through
caches and small write buffers is especially sensitive.

Kernel memory allocation is also a target that is sensitive.
The kernel have a kernel map limit, if the system reach this
limit it can not allocate more kernel memory and must be rebooted.
The kernel memory is not only used for RAM, CPU:s, screens and so
on, it it also used for ordinaries processes. Meaning that any system
can be crashed and with a mean (or in some sense good) algorithm pretty
fast.

For Solaris 2.X it is measured and reported with the sar command
how much kernel memory the system is using, but for SunOS 4.X there
is no such command. Meaning that under SunOS 4.X you don't even can
get a warning. If you do use Solaris you should write sar -k 1 to
get the information. netstat -k can also be used and shows how much
memory the kernel have allocated in the subpaging.
 
.B.4. RAM
---------

A denial of service attack that allocates a large amount of RAM
can make a great deal of problems. NFS and mail servers are 
actually extremely sensitive because they do not need much
RAM and therefore often don't have much RAM. An attack at 
a NFS server is trivial. The normal NFS client will do a 
great deal of caching, but a NFS client can be anything 
including the program you wrote yourself...

.B.5. DISKS
-----------

A classic attack is to fill up the hard disk, but an attack at
the disks can be so much more. For example can an overloaded disk
be misused in many ways.

.B.6. CACHES
-------------

A denial of service attack involving caches can be based on a method
to block the cache or to avoid the cache.

These caches are found on Solaris 2.X:

Directory name lookup cache: Associates the name of a file with a vnode.

Inode cache: Cache information read from disk in case it is needed
again.

Rnode cache: Holds information about the NFS filesystem.

Buffer cache: Cache inode indirect blocks and cylinders to realed disk
I/O.

.B.7. INETD
-----------

Well once inetd crashed all other services running through inetd no
longer will work.

.C. ATTACKING FROM THE OUTSIDE
~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~


.C.1. TAKING ADVANTAGE OF FINGER
--------------------------------

Most fingerd installations support redirections to an other host.

Ex:

	$finger @[email protected]

finger will in the example go through system.one.com and on to
system.two.com. As far as system.two.com knows it is system.one.com
who is fingering. So this method can be used for hiding, but also
for a very dirty denial of service attack. Lock at this:

	$ finger @@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@host.we.attack

All those @ signs will get finger to finger host.we.attack again and
again and again... The effect on host.we.attack is powerful and
the result is high bandwidth, short free memory and a hard disk with
less free space, due to all child processes (compare with .D.5.).

The solution is to install a fingerd which don't support redirections,
for example GNU finger. You could also turn the finger service off,
but I think that is just a bit to much.

.C.2. UDP AND SUNOS 4.1.3.
--------------------------

SunOS 4.1.3. is known to boot if a packet with incorrect information
in the header is sent to it. This is the cause if the ip_options
indicate a wrong size of the packet.

The solution is to install the proper patch.


.C.5. ATTACKING WITH LYNX CLIENTS
---------------------------------

A World Wide Web server will fork an httpd process as a respond
to a request from a client, typical Netscape or Mosaic. The process
lasts for less than one second and the load will therefore never
show up if someone uses ps. In most causes it is therefore very
safe to launch a denial of service attack that makes use of 
multiple W3 clients, typical lynx clients. But note that the netstat
command could be used to detect the attack (thanks to Paul D. Robertson).

Some httpd:s (for example http-gw) will have problems besides the normal
high bandwidth, low memory... And the attack can in those causes get
the server to loop (compare with .C.6.)

.C.6. MALICIOUS USE OF telnet
-----------------------------

Study this little script:

Ex:

	while : ; do
	telnet system.we.attack &
	done

An attack using this script might eat some bandwidth, but it is
nothing compared to the finger method or most other methods. Well
the point is that some pretty common firewalls and httpd:s thinks
that the attack is a loop and turn them self down, until the
administrator sends kill -HUP. 

This is a simple high risk vulnerability that should be checked
and if present fixed.

.C.7. MALICIOUS USE OF telnet UNDER SOLARIS 2.4
------------------------------------------------

If the attacker makes a telnet connections to the Solaris 2.4 host and
quits using:

Ex:

	Control-}
	quit

then will inetd keep going "forever". Well a couple of hundred...

The solution is to install the proper patch.

.C.8. HOW TO DISABLE ACCOUNTS
-----------------------------

Some systems disable an account after N number of bad logins, or waits
N seconds. You can use this feature to lock out specific users from
the system.

.C.9. LINUX AND TCP TIME, DAYTIME
----------------------------------

Inetd under Linux is known to crash if to many SYN packets sends to
daytime (port 13) and/or time (port 37).

The solution is to install the proper patch.

.C.10. HOW TO DISABLE SERVICES
------------------------------

Most Unix systems disable a service after N sessions have been
open in a given time. Well most systems have a reasonable default
(lets say 800 - 1000), but not some SunOS systems that have the
default set to 48...

The solutions is to set the number to something reasonable.

.C.11. PARAGON OS BETA R1.4
---------------------------

If someone redirects an ICMP (Internet Control Message Protocol) packet
to a paragon OS beta R1.4 will the machine freeze up and must be
rebooted. An ICMP redirect tells the system to override routing
tables. Routers use this to tell the host that it is sending
to the wrong router. 

The solution is to install the proper patch.

.C.12. NOVELLS NETWARE FTP
--------------------------

Novells Netware FTP server is known to get short of memory if multiple
ftp sessions connects to it.

.C.13. ICMP REDIRECT ATTACKS
----------------------------

Gateways uses ICMP redirect to tell the system to override routing
tables, that is telling the system to take a better way. To be able
to misuse ICMP redirection we must know an existing connection
(well we could make one for ourself, but there is not much use for that). 
If we have found a connection we can send a route that
loses it connectivity or we could send false messages to the host
if the connection we have found don't use cryptation.  

Ex: (false messages to send)

	DESTINATION UNREACHABLE 
	TIME TO LIVE EXCEEDED
	PARAMETER PROBLEM
	PACKET TOO BIG

The effect of such messages is a reset of the connection.

The solution could be to turn ICMP redirects off, not much proper use
of the service.

.C.14. BROADCAST STORMS
-----------------------

This is a very popular method in networks there all of the hosts are
acting as gateways. 

There are many versions of the attack, but the basic method is to 
send a lot of packets to all hosts in the network with a destination 
that don't exist. Each host will try to forward each packet so 
the packets will bounce around for a long time. And if new packets 
keep coming the network will soon be in trouble.

Services that can be misused as tools in this kind of attack is for
example ping, finger and sendmail. But most services can be misused
in some way or another.

.C.15. EMAIL BOMBING AND SPAMMING
---------------------------------

In a email bombing attack the attacker will repeatedly send identical
email messages to an address. The effect on the target is high bandwidth,
a hard disk with less space and so on... Email spamming is about sending
mail to all (or rather many) of the users of a system. The point of
using spamming instead of bombing is that some users will try to
send a replay and if the address is false will the mail bounce back. In
that cause have one mail transformed to three mails. The effect on the
bandwidth is obvious.

There is no way to prevent email bombing or spamming. However have
a look at CERT:s paper "Email bombing and spamming".

.C.16. TIME AND KERBEROS
------------------------

If not the the source and target machine is closely aligned will the
ticket be rejected, that means that if not the protocol that set the 
time is protected it will be possible to set a kerberos server of
function.

.C.17. THE DOT DOT BUG
----------------------

Windows NT file sharing system is vulnerable to the under Windows 95
famous dot dot bug (dot dot like ..). Meaning that anyone can crash
the system. If someone sends a "DIR ..\" to the workstation will a
STOP messages appear on the screen on the Windows NT computer. Note that
it applies to version 3.50 and 3.51 for both workstation and server
version.

The solution is to install the proper patch.

.C.18. SUNOS KERNEL PANIC
-------------------------

Some SunOS systems (running TIS?) will get a kernel panic if a
getsockopt() is done after that a connection has been reset.

The solution could be to install Sun patch 100804.

.C.19. HOSTILE APPLETS
----------------------

A hostile applet is any applet that attempts to use your system
in an inappropriate manner. The problems in the java language
could be sorted in two main groups:

	1) Problems due to bugs.
	2) Problems due to features in the language.

In group one we have for example the java bytecode verifier bug, which
makes is possible for an applet to execute any command that the user
can execute. Meaning that all the attack methods described in .D.X. 
could be executed through an applet. The java bytecode verifier bug
was discovered in late March 1996 and no patch have yet been available
(correct me if I'am wrong!!!).

Note that two other bugs could be found in group one, but they
are both fixed in Netscape 2.01 and JDK 1.0.1.

Group two are more interesting and one large problem found is the
fact that java can connect to the ports. Meaning that all the methods
described in .C.X. can be performed by an applet. More information
and examples could be found at address:
	
	http://www.math.gatech.edu/~mladue/HostileArticle.html

If you need a high level of security you should use some sort of
firewall for protection against java. As a user you could have
java disable. 

.C.20. VIRUS
------------

Computer virus is written for the purpose of spreading and
destroying systems. Virus is still the most common and famous
denial of service attack method.

It is a misunderstanding that virus writing is hard. If you know
assembly language and have source code for a couple of virus it
is easy. Several automatic toolkits for virus construction could
also be found, for example:
	
	* Genvir.
	* VCS (Virus Construction Set).
	* VCL (Virus Construction Laboratory).
	* PS-MPC (Phalcon/Skism - Mass Produced Code Generator).
	* IVP (Instant Virus Production Kit).
	* G2 (G Squared).

PS-MPC and VCL is known to be the best and can help the novice programmer
to learn how to write virus.

An automatic tool called MtE could also be found. MtE will transform
virus to a polymorphic virus. The polymorphic engine of MtE is well
known and should easily be catch by any scanner.

.C.21. ANONYMOUS FTP ABUSE
--------------------------

If an anonymous FTP archive have a writable area it could be misused
for a denial of service attack similar with with .D.3. That is we can
fill up the hard disk.

Also can a host get temporarily unusable by massive numbers of
FTP requests.

For more information on how to protect an anonymous FTP site could
CERT:s "Anonymous FTP Abuses" be a good start.

.C.23. PING FLOODING
--------------------

I haven't tested how big the impact of a ping flooding attack is, but
it might be quite big.

Under Unix we could try something like: ping -s host
to send 64 bytes packets. 

If you have Windows 95, click the start button, select RUN, then type
in: PING -T -L 256 xxx.xxx.xxx.xx. Start about 15 sessions.