It was found that curl can be fooled into writing a zero byte out of bounds.

This bug can trigger when curl is told to work on an FTP URL, with the setting
to only issue a single CWD command (`--ftp-method singlecwd` or the libcurl
alternative `CURLOPT_FTP_FILEMETHOD`).

curl then URL-decodes the given path, calls strlen() on the result and deducts
the length of the file name part to find the end of the directory within the
buffer. It then writes a zero byte on that index, in a buffer allocated on the
heap.

If the directory part of the URL contains a "%00" sequence, the directory
length might end up shorter than the file name path, making the calculation
`size_t index = directory_len - filepart_len` end up with a huge index
variable for where the zero byte gets stored: `heap_buffer[index] = 0`. On
several architectures that huge index will wrap and work as a negative value,
thus overwriting memory *before* the intended heap buffer.

By using different file part lengths and putting %00 in different places in
the URL, an attacker that can control what paths a curl-using application uses
can write that zero byte on different indexes.