This is a `shim' mailer.  It's designed to be put in front of your
world-facing MX host(s) when you're getting hammered by user-unknown
connections, typically backscatter from spam runs.  It is designed to
be very lightweight; it deals with all the hard parts of SMTP by
handing them off to some other ("backend") machine.  It speaks SMTP to
the client until a valid username is received, at which point it
establishes a connection to the backend machine's SMTP server and
simply passes protocol between the two.

Of course, it has to be able to tell what addresses are valid.  This
test is isolated in a separate file (validaddr.c).  The code as
distributed expects to find a list of local-parts in
/etc/mailshim-userlist, and requires that the addresses be
@example.net; you presumably will want to change at least the latter.
If your backend machine is prepared to handle mail for multiple
domains, you can insert a single mailshim in front of them all by
making this code more elaborate.

It is not critical that mailshim be able to tell _exactly_ what
addresses are valid.  If mailshim thinks a valid address is invalid,
it will reject mail to that address, so don't do this; but if mailshim
thinks an invalid address is valid, all that will happen is that the
backend mailserver will have to reject it.  As long as mailshim cuts
the huge numbers of unknown-user connections to your backend mailserver
down to something manageable, it will be doing its job.

This does have the potential disadvantage that certain anti-spam
measures taken at SMTP time, such as ident checks, and some other
measures like SMTP AUTH that are not normally thought of as anti-spam,
cannot be done with mailshim as distributed.  If you're a decent C
coder, you may be able to hack such things in, but otherwise, you'll
have to choose between them and mailshim.

One thing that mailshim _does_ do is DNSBL checks.  This too is pulled
out into a separate module; see checkip.c.  Another thing is banner
delays; see the DEF_SLEEP define and the "xctl sbb" local command.

Building: Compile the C files and link the results together.  Which C
files you want depends on which program you're trying to build.  While
the Makefile probably won't work for you (it refers to an include file
that isn't in the package), it will tell you which source files go to
make up which program (strictly, which object files, but they are
1-to-1 with source files).  This program's home machine is a NetBSD
system; I think I've mostly managed to avoid nonportabilities, except
for a few gccisms (I see a nested function, for example).  If you have
trouble, let me know and I'll see what I can do to help.

README.linux is a document collecting some findings from someone who
worked on getting it to behave on Linux.

Copyright licensing: this program is in the public domain.  Anyone may
use it in any way for any purpose (though I would appreciate credit
where it is due).  If it breaks you get to keep both pieces.

If you find bugs, I'd like to hear about them, especially if you also
fix them.  If you make improvements you think are worth rolling back
into the main distribution (and don't mind releasing into the public
domain), I'm also interested in those.

/~\ The ASCII				der Mouse
\ / Ribbon Campaign
 X  Against HTML	       mouse@rodents.montreal.qc.ca
/ \ Email!	     7D C8 61 52 5D E7 2D 39  4E F1 31 3E E8 B3 27 4B