It is called Venturi Effect.
The increase in speed of the air surrounding your vehicle comes with a decrease in pressure. That explains too why a chimney works better in windy days.
The Venturi effect is explained by applying the Bernoulli Equation (say, the conservation of energy of a small piece of fluid that moves within the flow) between two points along a streamline (in this case, we would follow a piece of air in a tunnel wind)
$\frac{1}{2} \rho v^2 + \rho g h + p = \text{constant}$
The increase in the first summand when the flow gains speed to adapt itself to the shape of the car, is compensated by a decrease in the pressure $p$. Look what happens in this picture (wikipedia) when the flow changes speed to adapt to the shape of the tube:
(Image from wikipedia)
$ $
Note the similarity with the high school equation for the conservation of mechanical energy of a particle:
$\frac{1}{2} m v^2 + m g h = \text{constant}$
(Just change the mass of the particle for the mass of a fluid volume unit, i.e. density, and add an additional summand to accout for the pressure, and you have Bernouilli's equation)
Bernouilli's equation is meant for an incompressible flow (water) which here means that the numerical results would be approximate, but qualitatively the same effect happens.
A related, interesting fact, is that submarine propellers must be carefully designed, in order to avoid points in which water suffers much too rapid a speed increase. When that happens, pressure becomes so low in that points that vacuum bubbles appear. The power released by the implosion of that bubbles against the surface of the propeller, not only is noisy, but also may damage the propeller itself. The phenomenon is called cavitation.
(Image from wikipedia)
There are actually several different things going on. The reason you hear a single tone is because of resonance. The reason that it is usually a higher frequency tone has to do partly with the falloff in amplification with distance and partly (perhaps mainly) with the frequency response of the microphone/amplifier/loudspeaker.
Let's model the situation with $A$ representing the combination of the microphone, amplifier and loudspeaker, and the effect of the sound traveling through the air from the loudspeaker back to the microphone as $B$. In general $A(\omega)$ will be a function of the frequency $\omega$, and will be almost linear until the output signal gets to be large enough that it gets "clipped". That is, $A$ behaves as a complex multiplier where the magnitude of $A$ is the amplification and the angle of $A$ is the phase shift.
$B$ also behaves linearly, and depends mostly on the distance $r$ of the microphone from the loudspeaker. The signal will travel through the air at the speed of sound (~340m/s), and so there will be a delay of $\frac{r}{340\mathrm{m/s}}$, and $B$ will attenuate the signal by some factor, let's say $1/r^2$. Here's the public domain picture from wikimedia commons:
If the input is $x(t)$ and the output is $y(t)$ then we get $ y=A(x+By) $ or
$$
y(t) = \frac{A}{1-AB} x(t)\mathrm{.}
$$
Oscillations occur when $AB = 1$ (exactly). That means that the attenuation from $B$ is equal to the amplification by $A$ (including clipping) and the phase shift of traveling distance $r$ exactly cancels the phase shift from the amplifier.
Let's consider the simplest possible case where the microphone, amplifier and loudspeaker have no phase shift. Then the phase shift for the feedback is based completely on the distance between the loudspeaker and microphone. The speed of sound is about 340 m/s. That means that a 20Hz sound wave is about 17 meters long, so you'd need to be about 17 meters from the speaker to get 20Hz (and all the harmonics of 20Hz) to be in phase.
So if you put a microphone 17 meters from the speaker will you get 20Hz feedback? Probably not. You'll probably get one of the higher frequency harmonics. Why? Because the microphone, amplifier and loudspeaker are probably much better at amplifying midrange frequencies than they are at amplifying low and high frequencies.
Look at the frequency response for the Shure SM58 microphone (a popular and widely used voice microphone). It looks like this:
Most guitar amps look similar. dB is a logarithmic scale, usually $20 \log_{10}\frac{p}{p_\mathrm{ref}}$ so 5dB is about 1.8x and 10dB is 3.2x. Some harmonic in that range between about 3KHz and 8KHz is probably going to dominate.
Note that the phase shift from the amplifier matters a lot and we didn't incorporate it into our model. For fun I decided to try out some different distances on my home computer to see what would happen. Because of constraints due to wiring the farthest I could get my mike from the speaker was 1.8 meters. In multiple trials I got 383, 379 and 386 Hz. 1.8 meters corresponds to a minimum frequency of 189 Hz in our model. At 10cm I got about 3700 Hz (lowest harmonic at that distance in our model is around 3400) At 20cm I got about 2600 Hz and at 50cm I got about 400 Hz. The lowest harmonic at 50cm would be about 680Hz according to the model.
Best Answer
The car is behaving like a closed pipe, so you get a resonance set up. There's a Wikipedia article here, but for once the Wikipedia article isn't that great, so there's another better article here. I imagine you (like most of us) will at some point have discovered you can make a sound by blowing across the top of an opened bottle, and it's the same thing happening in your car with the open window acting like the opening in the bottle. Since your car is much bigger than a bottle the resonance frequency is uncomfortably low.
When you open a second window you get an air current flowing through the car and this destroys the resonance.