Reading a float from a packet

A received packet has a 4-byte float starting at buf[2]. Which is the correct, portable way to read it on a Cortex-M?

uint8_t buf[16];
float f;
/* option A */ f = *(float *)&buf[2];