September 24th, 2003, 11:52
I. Background

The Address Resolution Protocol (ARP) is fundamental to the operation
of IP with a variety of network technologies, such as Ethernet and
WLAN. It is used to map IP addresses to MAC addresses, which enables
hosts on a local network segment to communicate with each other
directly. These mappings are stored in the system's ARP cache.

FreeBSD's ARP cache is implemented within the kernel routing table as
a set of routes for the address family in use that have the LLINFO
flag set. This is most commonly often AF_INET (for IPv4). Normally,
when a FreeBSD system receives an ARP request for a network address
configured on one of its interfaces from a system on a local network,
it adds a reciprocal ARP entry to the cache for the system from where
the request originated. Expiry timers are used to purge unused
entries from the ARP cache. A reference count is maintained for each
ARP entry. If the reciprocal ARP entry is not in use by an upper
layer protocol, the reference count will be zero.

II. Problem Description

Under certain circumstances, it is possible for an attacker to flood a
FreeBSD system with spoofed ARP requests, causing resource starvation
which eventually results in a system panic. (The critical condition
is that a route exists for the apparent source of the ARP request.
This is always the case if the system has a default route configured
for that protocol family.)

If a large number of ARP requests with different network protocol
addresses are sent in a small space of time, resource starvation can
result, as the arplookup() function does not delete unnecessary ARP
entries cached as the result of responding to an ARP request.

NOTE WELL: Other BSD-derived systems may also be affected, as the
affected code dates well back to the CSRG branches.

III. Impact

An attacker on the local network may be able to cause the system to
hang or crash. The attacker must have physical access to the shared
network medium. In the case of a wireless network obtaining this
access may be trivial. Networks where proxy ARP is used to direct
traffic between LANs may be particularly vulnerable to the attack,
as the spoofed ARP requests could be bounced through to the target
via routers implementing proxy ARP.

Because the attack operates at Layer 2, the use of strong encryption
technologies such as IPsec cannot protect a system against the attack.