pires for a particular half-open connection, the host will re-
set the connection and release all resources allocated for it.
If a spoofed SYN packet contains the source address of a
reachable IP host S
, that host will receive the second mes-
sage of the three-way handshake generated by D. Not ex-
pecting a SYN+ACK without having requested a connec-
tion, S
will send a RST packet to D, and consequently cause
D to reset the connection. It is therefore in the interest of an
attacker to forgesource addresses that do not belong to hosts
that are reachable from the victim D.
If the attacker wants the denial of service condition to
last longer than the timeout period, he needs to continuously
keep requesting new connections from the victim machine.
The amount of CPU and network bandwidth required by an
attacker for a sustained attack is negligible.
The basis of the attack is that TCP/IP does not offer
strong authentication on its control packets. Furthermore,
there is a requirementfor an inappropriately burdensomeal-
location of memory and computation resources on the target
side.
3.2. Different Attack Modes
Typical SYN flooding attacks can vary several parame-
ters: the number of SYN packets per source address sent in a
batch (
batch-size), the delay between successive batches
(
delay), and the mode of source address allocation (
mode).
We consider only source addresses of hosts that are not
reachable from D, be it because the addresses are not yet al-
located, assigned, or the associated hosts are very slow in re-
sponse, virtually or physically disconnected, or down. We
classify three possible modes of source address allocation:
The attacker can be using a single address, a short list of ad-
dresses, or no list at all
.
Single address: The attack scripts published in the hacker
magazines Phrack [6] and 2600 [8] take as a parame-
ter a single spoofed address that is used as the source
address of all SYN packets. In the absence of any de-
fense, this mode of attack is as effective as the other
three modes described.
Short list: An attacker can generate a small pool of ad-
dresses and use them as source addresses to generate
SYN packets.
No list: The attacker can use a different, randomly gener-
ated source address for each successive batch of SYN
packets.
We list the “single address” mode separately, because it represents an
important special case of the “list of addresses” mode.
4. Solutions
In our opinion a good solution should have the following
characteristics:
independence of operating system and network stack
implementation of the protected end systems
no requirement for IP or TCP protocol modifications
capability to protect sets of machines, and not only a
single machine
no special hardware requirements
portability
extensibility
configurability
The countermeasures described in this section have been
proposed by others to date. None of these proposals pro-
vides all the characteristics we are looking for.
4.1. Configuration Optimization
There are several ways of reducing the likelihoodand ef-
fects of an attack that involve changes in the configurations
of end systems and routers.
4.1.1 System Configuration Improvements
To defend against the exhaustion of resources in the systems
under attack, an obvious approach is to increase the number
of resources devoted to half-open TCP connections, and to
reduce the timeouts. These measures have been suggested
by different sources [11], and can be summarized as:
1. Reduce the timeout period from the default to a short
time, e.g., 10 seconds. This helps in pruning half-
open connections from the TCP queue.
2. Significantly increase the length of the backlog queue
from the default (see Table 1). This makes the system
able to cope with more simultaneous half-open con-
nections than before.
3. Disable non-essential services, thus reducing the
number of ports that can be attacked.
These measures help in dealing with attacks, but also
have severe shortcomings:
1. Loweringthe timeouts may deny legitimate access for
machines to which the round trip times exceed the
timeout period.