next up previous
Next: Deauthentication Attack Up: Practical Attacks and Defenses Previous: Practical Attacks and Defenses
Whole Paper:Single Page Version

802.11 Attack Infrastructure

From a purely practical perspective, a key engineering question is, ``Can an attack be generated with commodity hardware?'' While theoretical vulnerabilities are clearly important, we feel that attacks with software implementations represent a qualitatively greater threat since they are available to a dramatically expanded set of potential attackers.

At first glance this appears to be a trivial problem since all 802.11 Network Interface Cards (NIC) are inherently able to generate arbitrary frames. However, in practice, all 802.11(a,b) devices we are aware of implement key MAC functions in firmware and moderate access to the radio through a constrained interface. The implementation of this firmware, in turn, dictates the limits of how a NIC can be used by an attacker. Indeed, in reviewing preprints of this paper, several 802.11 experts declared the virtual carrier-sense attack infeasible in practice due to such limitations.

In testing a wide variety of 802.11 NICs we have found that most allow the generation of management frames necessary to exploit the identity attacks described earlier - typically using semi-documented or undocumented modes of operation, such as HostAP and HostBSS mode in Intersil firmware. However, these same devices do not typically allow the generation of any control frames, permit other key fields (such as Duration and FCS) to be specified by the host, or allow reserved or illegal field values to be transmitted. Instead, the firmware overwrites these fields with appropriate values after the host requests that queued data be transmitted. While it might be possible to reverse-engineer the firmware to remove this limitation, we believe the effort to do so would be considerable. Instead, we have developed an alternative mechanism to sidestep the limitations imposed by the firmware interface. To understand our approach it is first necessary to understand the architecture of existing 802.11 products.

Figure 3: A block diagram depicting how the ``aux port'' can be used to circumvent the limitations imposed by the firmware. By using this raw memory interface, the host can transform ``normal'' packets into arbitrary 802.11 frames as they are transmitted.
$>32$

Most commodity 802.11 devices, including those using Intersil Prism, Lucent/Agere/Orinoco/Proxim Hermes and Cisco Aironet chipsets are based on an initial MAC design originated by Choice Microsystems (since acquired by Intersil). In this architecture, all low-level functions - including frame transmission, scheduling, acknowledgement, and fragmentation - are implemented in firmware while the host is simply responsible for managing data transfer to and from the device. Data transfer is achieved through a firmware-implemented ``Buffer Access Path'' (BAP) that shields the driver writer from the details of NIC memory management and synchronization. While the BAP interface will typically accept raw 802.11 frames, these packets are then further interpreted by concurrent firmware processes. As a result, only a subset of potential frames can be successfully transmitted by the host.

However, Choice-based MACs also provide an unbuffered, unsychronized raw memory access interface for debug purposes - typically called the ``aux port''. By properly configuring the host and NIC, it is possible to write a frame via the BAP interface, locate it in the NIC's SRAM, request a transmission, and then modify the packet via the aux port - after the firmware has processed it, but before it is actually transmitted. This process is depicted in Figure 3. To synchronize the host and NIC, a simple barrier can be implemented by spinning on an 802.11 header field (such as duration) that is overwritten by the firmware. Alternatively, the host can continuously overwrite if synchronization is unnecessary. In practice, this ``data race'' approach, while undeniably ugly, is both reliable and permits the generation of arbitrary 802.11 MAC frames. Using this method we are able to implement any of the attacks previously described using off-the-shelf hardware. We believe we are the first to demonstrate this capability using commodity equipment.

Figure 4: iPAQ H3600 with Dlink DWL-650 card, running Swat attack testing tool. Individual clients and AP's are identified either using MAC address or by passively monitoring the channel and extracting destination IP addresses and DNS names.
$>32$

Our prototype, called Swat, consists of an iPAQ H3600 Pocket PC, running Familiar Linux, with a DLink DWL-650 PCMCIA 802.11 interface mounted in a standard PC Card sleeve. The entire device weighs approximately 375g (a bit over 12 oz) and is easily concealed in a coat pocket. More modern Pocket PCs, such as the Toshiba e740/e750 and the HP iPAQ 5450, include integral 802.11 functionally and could accomplish the same feats with roughly half the size and weight.

To experiment with denial-of-service attacks we have built a demonstration application that passively monitors wireless channels for APs and clients. Individual clients are identified initially by their MAC address, but as they generate traffic, a custom DNS resolver and a slightly modified version of dsniff [Son] is used to isolate better identifiers (e.g., userids, DNS address of IMAP server, etc). These identifiers can be used to select individual hosts for attack, or all hosts may be attacked en masse. The application and the actual device are pictured in Figure 4.

In the remainder of this section, we analyze the impact of the deauthentication attack and a preliminary defense mechanism, followed by a similar examination of the virtual carrier-sense attack and defense.


next up previous
Next: Deauthentication Attack Up: Practical Attacks and Defenses Previous: Practical Attacks and Defenses
Whole Paper:Single Page Version

John Bellado 2003-05-16
In Proceedings of the USENIX Security Symposium, Aug 2003