Using the --dump-slave option of mysqldump stops the replication at the beginning of the dump and restarts it once its done but if the dump fails for any reason (if the database does not exists or the user doesnt have enough permission on it for example), the replication is not restarted at the end of the process.
To reproduce, use a user with the REPLICATION SLAVE grant (and check that the replication is running prior) :
The slave replication is now stopped (Slave_SQL_Running = No).
ps: i havent tested if a similar issue exists with --master-data.
Debian Wheezy amd64