Joel Holdsworth <joel@airwebreathe.org.uk> writes:
Hi Folks,
I've just joined this ML, and I'm hoping someone can give me some advice.
I've been experimenting with Kingston DT5000 hardware encrypted usb pen drives, and I'm hoping to write a driver.
The DT5000 drives are probably most useful in corporate settings, where sensitive need to be protected. They might also be useful for machines that have sensitive information that must be left unattended. They also have excellent tamper evident characteristics.
The device is a composite device containing an ejectable mass storage device, and a read only virtual cdrom. The cdrom contains windows software which autoruns and pops up a dialog that prompts the user to give the password, this is then conded and sent to the device via USB vendor control packets.
And after this the password application quits and the device is unlocked and works as a normal usb-storage device until it is unplugged? Or ejected? Or are all storage transactions encrypted?
I've got quite far hacking around with the device with pyusb and usbmon, and now I'm trying to gather information for a driver, so I have a few questions:
First, via a series of messages the host PC uses Elliptic Curve Diffie Helman to establish a shared secret with the DT5000. The shared secret is then used to encrypt the password hash. I couldn't find any elliptic curve code in the kernel. Is this correct? Perhaps there are patent concerns? are there any plans to add it in?
For a device like this, how would the user space interface look? Would one just make a sysfs file into which the the password plaintext is written? Or is it more normal to enter it via some userspace application that hashes and encrypts the password with the shared secret. Does dm_crypt, or IPSEC stuff offer a pattern to follow here? Or are there any very simple examples of password locking devices?
udisks offer a D-Bus API and a 'udisksctl lock|unlock' command using it: http://www.freedesktop.org/wiki/Software/udisks/ But as usual with fdo projects, this might already be obsolete for all I know...
I'm trying to figure out where the crypto work should be done. If the kernel did all the crypto, I guess the password interface would be very simple. If the task is split with user space then this is going to expose all this password hand-shake protocol to userspace - not such a slick interface I would say.
Also, the device contains a few user programmable fields: For example the owner's name, address, and a hint for the password etc. These things are read-only when the device is locked, but become writable when the device is unlocked. I suppose this information would be communicated via sysfs. Is the pattern generally to have a separate file for each one?
Also are there any general comments that I should take on board as I embark on this?
If the device can work with the generic usb-storage driver after a device specific initialization, then I would propose using a libusb based userspace application to initialize it. This application could be autorun by udev when the device is detected. A kernel driver is not required. Doing it in userspace has quite a few advantages. Like: Most of your questions becomes void :-) And you are free to use whatever crypto library you need. And the choices you have to make, like how to get the password from the user, are not carved in stone as a kernel userspace API. See also http://linuxdriverproject.org/mediawiki/index.php/No_Linux_Driver_Needed Bjørn