Matthias Andree reports:
When using Unicode databases (default in more recent bogofilter installations), upon encountering invalid input sequences, bogofilter or bogolexer could overrun a malloc()'d buffer, corrupting the heap, while converting character sets. Bogofilter would usually be processing untrusted data received from the network at that time. This problem was aggravated by an unrelated bug that made bogofilter process binary attachments as though they were text, and attempt charset conversion on them. Given the MIME default character set, US-ASCII, all input octets in the range 0x80...0xff were considered invalid input sequences and could trigger the heap corruption.