USB capture setup
This page is about capturing raw USB traffic, e.g. the packets a USB mouse will generate on the Universal Serial Bus.
Table of contents
USB attached network interfaces
A special case are network interfaces connected to a host computer through an USB cable. The operating system "converts" the raw USB packets into the network traffic (e.g. Ethernet packets) and provides a network interface that looks like an ordinary network interface. So you can capture from:
- the USB device for raw USB traffic (if supported)
- the network device for "normal" network packets
The USB bus will add additional overhead, so the raw USB traffic will have higher volume than the network traffic, even if the only active USB devices on the system are network adapters. (If there are other active USB devices, the raw USB traffic will include traffic to and from those devices, so it will obviously have higher volume than Ethernet traffic.)
Capturing USB traffic on Linux is possible since Wireshark 1.2.0, libpcap 1.0.0, and Linux 2.6.11, using the Linux usbmon interface. To dump USB traffic on Linux, you need the usbmon kernel module. If it is not loaded yet, run this command as root:
With Linux kernels prior to 2.6.23, you will also need to run this command as root:
mount -t debugfs none /sys/kernel/debug
and, with those kernels, the usbmon mechanism's protocol limits the total amount of data captured for each raw USB block to about 30 bytes. With a 2.6.23 or later kernel, and libpcap 1.1.0 and later, that size limitation is removed.
In libpcap 1.1.0 and later, the devices on which you can capture are named usbmonX, where X is the USB bus number. On Linux 2.6.22 and later, the special "usbmon0" interface receives events from all USB buses. In libpcap 1.0.x, the devices were named usbX.
You can also capture and debug USB traffic on a virtual Windows machine under VirtualBox. In some ways this is more convenient than working with a separate Windows box.
In this example, an embedded Linux device running g_ether (RNDIS ethernet gadget) connects to Windows. e.g. an NSLU2 with a USB slave modification http://www.nslu2-linux.org/wiki/HowTo/AddDeviceSideUSBPort but it should work for almost any USB device.
With this method, Linux recognises the USB device (i.e. >lsusb will still show them), but VirtualBox hooks it into Windows but Wireshark on linux still gets to snoop on all the packets.
1. Install a VirtualBox Windows guest on your Linux host. Start up the virtual Windows session.
2. Plug-in the embedded slave device via a USB cable, which itself should be either a device Windows already knows about (or in this case it was running a valid g_ether gadget stack and needed a .inf file)
3. Run >lsusb and take a note of which bus the device connects.
- "Bus 003 Device 003: ID 0525:a4a2 Netchip Technology, Inc. Linux-USB Ethernet/RNDIS Gadget"
4. On linux side,run >ifconfig usb0 down - this prevents both the linux system and the windows system from fighting over the device
5. On the Windows virtual machine, on VirtualBox menus click the checkbox
[Devices]->Usb devices>[x]Your device
- to let windows see the USB device.
6. Now Windows should recognise the device and proceed with the "plug-and-pray" session for driver initialisation.
I worked from the instructions on http://docwiki.gumstix.org/index.php/Windows_XP_usbnet to install the driver.
7. In this example, I had to set up the networking options for IP address, Gateway etc on Windows to match the IP network on the gadget but for other USB device types there will be no extra setup. In any case this is just normal Windows behavior.
8. On Linux, startup Wireshark and using the Bus number given earlier from >lsusb command to sniff for packets.
Hints for developing something like a Windows native "USBPcap": a kernel mode filter device driver has to be written. An older Driver Development Kit (DDK) is available which at least can compile kernel mode binaries. The most important functions to install the filter driver are CreateService() and SetupDiSetDeviceRegistryProperty() function with SPDRP_LOWERFILTERS parameter.
Why was the note about inaccurate time stamps removed?!? - UlfLamping
The timestamps should be ok now since libpcap worksaround the issue by explicitly calling gettimeofday()- ronnie
Well, the inaccuracies I had in mind was about the "delta" involved between the data is received from the USB device and actually timestamped from the kernel. This delta will be substantially lower for e.g. PCI based nic's than for USB ones - and should be mentioned. Or am I just wrong on this topic and this can be ignored - which should be mentioned then too? - UlfLamping
There's "capturing on USB-attached networking interfaces" and there's "capturing USB traffic"; this page is for the latter, but it sounds as if the time stamp delta is an issue for the former. - Guy Harris