Milter-greylist

On Wed, Nov 30, 2005 at 10:00:57PM +0000, Emmanuel Dreyfus wrote:
> Could you grep sendmail source for DRAC usage? Just to be sure we are
> to open the file forever with a shared lock...

The reference to drac is made through the .cf file, rather than the
sendmail program itself.  The .mc says something like this:
    LOCAL_CONFIG
    # dynamic relay authorization control map
    Kdrac btree -o /usr/local/etc/dracd
    ...
    SLocal_check_relay
    # allow recent POP/IMAP mail clients to relay
    R$*				$: $&{client_addr}
    R$+				$: $(drac $1 $: ? $)
    R?				$@ ?
    R$+				$@ $#OK

Running lsof on an idle sendmail daemon shows none of the map files
are held open (e.g., dracd.db, aliases.db).  They are only opened
when reference is made to a particular map.  The sendmail map.c
code acquires and releases the lock with the sequence
    lockfile(..., LOCK_SH);
    ...
    lockfile(..., LOCK_UN);
In turn, lockfile() uses BSD style flock() when available, or fcntl()
style locks otherwise.

The check_drac() function that I posted last night no longer holds
the map file open.  Instead, the sequence goes like this:
    - open the DB file with the O_SHLOCK flag set
    - attempt to get() the item as an existence check
    - close the file (implicitly clearing the shared lock)
So if there are systems which don't support O_SHLOCK or only have
fcntl() locks, there could be a possible hole, but this risk seems
negligible.  Moreover, since the DB is merely used as an existence
check, the content is not that significant.

Romain