The code for decoding a BSON stream does not properly catch all invalid sizes. We check that a document size isn't > 16MB, but not that it isn't less than 5. So when we have an invalid bson document with size < 5, and we hit the line

	_, err = io.ReadAtLeast(bs.Stream, into[4:int(bsonSize)], int(bsonSize-4))

we get a

panic: runtime error: slice bounds out of range