A single filter process can handle any number of connections simultaneously.
All the filtering callbacks must therefore be reentrant and they
must use appropriate external synchronization methods to access
global data. Because a one-to-one correspondence between the threads
and connections (N connections mapped on to M threads, where M is
less than or equal to N) does not exist, you must access connection-specific
data through the handles provided by the Milter library. You must
not rely on the thread-specific data blocks supplied by the library
to store data blocks (for example, pthread_getspecific()) to store connection-specific data.
For more information on setting and getting connection-specific
pointers, see “The
smfi_setpriv() API” and “The
smfi_getpriv() API”, respectively.
