Interfaces to Input Entropy

The kernel exports a family of interfaces to facilitate feeding data into the entropy pool. They are called by the appropriate kernel subsystems or drivers. They are

void add_interrupt_randomness(int irq)
void add_keyboard_randomness(unsigned char scancode)
void add_mouse_randomness(__u32 mouse_data)

add_interrupt_randomness()is called by the interrupt system whenever an interrupt is received whose handler was registered with SA_SAMPLE_RANDOM. The parameter irq is the interrupt number. The random number generator uses the timing between interrupts as a source of noise. Note that not all devices are suitable for this; if the device generates interrupts deterministically (for example, the timer interrupt) or may be influenced by an outside attacker (for example, a network device) it should not feed the pool. An acceptable device is a hard disk, which generates interrupts at an unpredictable rate.

add_keyboard_randomness() uses the scancode and the timing between successive key presses to feed the entropy pool. Interestingly, the routine is smart enough to ignore autorepeat (when the user holds a key down) because both the scancode and timing interval would then be constant, contributing no entropy. The sole parameter is the scancode of the pressed key.

add_mouse_randomness() uses the mouse position as well as the timing between interrupts to feed the pool. The parameter mouse_data is the hardware-reported position of the mouse.

All three of these routines add the supplied data to the entropy pool, calculate an estimate of the entropy of the given data, and increment the entropy estimate by this amount.

All these exported interfaces use the internal function add_timer_randomness() to feed the pool. This function calculates the timing between successive events of the same time and adds the delay to the pool. For example, the timing between two successive disk interrupts is largely randomespecially when measured very precisely. Often, the low-order bits are electrical noise. After this function feeds the pool, it calculates how much randomness was present in the data. It does this by calculating the first-, second-, and third-order deltas from the previous timing and first- and second-order deltas. The greatest of these deltas, rounded down to 12 bits, is used as the entropy estimate.