You need entropy in a low-entropy environment and can prompt the user to type in order to collect it.
On Unix, read directly from the controlling terminal (/dev/tty). On Windows, process all keyboard events. Mix into an entropy pool the key pressed, along with the timestamp at which each one was processed. Estimate entropy based upon your operating environment; see the considerations in Recipe 11.19.
There can be a reasonable amount of entropy in key presses. The entropy comes not simply from which key is pressed, but from when each key is pressed. In fact, measuring which key is pressed can have very little entropy in it, particularly in an embedded environment where there are only a few keys. Most of the entropy will come from the exact timing of the key press.
The basic methodology is to mix the character pressed, along with a timestamp, into the entropy pool. We will provide an example implementation in this section, where that operation is merely hashing the data into a running SHA1 context. If you can easily get information on both key presses and key releases (as in an event-driven system like Windows), we strongly recommend that you mix such information in as well.
The big issue is in estimating the amount of entropy in each key press. The first worry is what happens if the user holds down a key. The keyboard repeat may be so predictable that all entropy is lost. That is easy to thwart, though. You simply do ...