Let's say we have a uint8, which can only have 8 bits. That means the largest number we can store is binary 11111111 (or in decimal, 2^8 - 1 = 255).
If we'd call:
uint8 number = 255;
number++;We've cause an overflow — so number is counterintuitively now equal to 0 even though we increased it. (If you add 1 to binary 11111111, it resets back to 00000000).
An underflow is similar, where if you subtract 1 from a uint8 that equals 0, it will now equal 255 (because uints are unsigned, and cannot be negative).
It seems unlikely that a uint256 would overflow as 2^256 is a really big number, but it's still good to put protections in our contract so that our dApps are safe from unexpected behaviour.
To prevent overflows and underflows, we can use SafeMath library.