Problem:

In the function: Format_description_log_event::write(..)

memcpy((char *)buff + ST_COMMON_HEADER_LEN_OFFSET + 1, &post_header_len[0],
        Binary_log_event::LOG_EVENT_TYPES);

This works fine, if the server generating the FDE is the same as the server
writing it out (hence post_header_len.size == Binary_log_event::LOG_EVENT_TYPES).

However, when we rotate relay logs, we use this function to write out the FDE
by using the event obtained from the master directly. If the master is on an
older version of MySQL with not as many event types, then this memcpy will
read beyond its buffer causing an ASAN failure.

Analysis:

Different versions of mysqld have different number of binlog event types. The
Format_description_event log has an post_header_len whose size is based on the number
of events supported by the mysqld version. So a 5.6 mysqld generates a FDE
with 35 types while 5.7 mysqld generates a FDE with 38 types. When 5.7 mysqld
attempts to read a 5.6 generated FDE and then rewrite out the 5.6 FDE, it
uses its own value of LOG_EVENT_TYPES (38) to determine the length of the
array, which is actually only 35 types.

The copy accesses undefined memory to generate the FDE.

Fix:
if the post_header_len is greater than Binary_log_event::LOG_EVENT_TYPES, use
the sizeof post_header_len vector while writing post_header_len vector,
otherwise use Binary_log_event::LOG_EVENT_TYPES.

P.S: This patch also corrects another issue which happens while decoding the
post_header_len array in FDE.

While decoding post_header_len we use the variable number_of_event_types to
determine how many bytes to read from the buffer, now at this point the
number_of_event_types holds the value which is one more than the actual value.
This problem happens because we don't consider the "BINLOG_CHECKSUM_ALG_DESC_LEN"
before decoding post_header_len. In 8.0 this issue has already been
corrected(WL#11567) so this part of change is just for 5.7.

RB: 21849