Skip to content

Latest commit

 

History

History
304 lines (224 loc) · 11.7 KB

README.md

File metadata and controls

304 lines (224 loc) · 11.7 KB
Raw

Better-qmail-remote

This is a wrapper around qmail-remote that currently does the following jobs:

  • DKIM signing

  • add hashcash header if the hashcash tool is installed (and silently do not if it isn't)

  • avoid sending backscatter, by diverting bounces for mails with spamassassin headers with high-ish scores locally (it also checks non-bounces using a separate score)

It being a wrapper written in Perl (with tainting checks on) means that there is no need to patch Qmail, i.e. it works with the qmail package from Debian (which is actually netqmail) and there should be less risk of opening up a security hole.

It is a reworked version of the dkim script from qmailtoaster. I'm running it on some small systems, where the overhead of running a Perl wrapper is irrelevant (and where I can in fact also easily afford to run hashcash).

This, together with using qpsmtpd in place of qmail-smtpd, should be enough to make Qmail work acceptably/well in a modern setting. (The only things missing are encryption and delay notifications. Also, you'll want to set up SPF, perhaps you're interested in using tinydns-scm for this purpose--ah, I haven't published that yet. Feel free to email me.)

Installation

This is for Debian(-derived) systems, you will have to install packages differently on other systems and may need to use different paths.

Check out the source code:

cd /opt
git clone https://github.com/pflanze/better-qmail-remote.git
cd better-qmail-remote

Check out the latest release (unless you want the latest changes and fixes):

v=`git tag -l | grep '^v' | sort |tail -1`
# to check the signature
gpg --recv-key A54A1D7CA1F94C866AC81A1F0FA5B21104EDB072
git tag -v "$v"
# check it out as a branch
git checkout -b "$v" "$v"

Get the submodules:

git submodule init && git submodule update

Optional: run the tests in the functional-perl submodule (although the kind of test most likely to fail is rather unlikely to be relevant for better-qmail-remote):

cd functional-perl
./test.pl
# NOTE: 't/csvstreams' is known to fail, no problem for this project.

Install dependencies:

apt-get install libmail-dkim-perl  # or: cpan -i Mail::DKIM

Install hashcash if you'd like to use it (it will automatically be used when accessible from $PATH):

apt-get install hashcash

To use the backscatter avoidance feature (for an explanation what this is, see the section below): decide on a group name for users who will check the captured bounces, then set up the Maildirs using:

/opt/better-qmail-remote/setup-debounce yourgroupname

Activate the wrapper:

# on Debian:
dpkg-divert --local --divert /usr/sbin/qmail-remote.orig --rename /usr/sbin/qmail-remote
# otherwise just `mv /usr/sbin/qmail-remote{,.orig}`, but you'll 
# have to do it again after package upgrades.
ln -s /opt/better-qmail-remote/qmail-remote /usr/sbin/

Configuration

better-qmail-remote tries to use sane defaults, so the only files needed are:

  • the key(s), which are expected to be found under the glob /var/qmail/control/dkim/*.key. The file name without the dot and suffix is taken to be the selector name. For every key file, qmail-remote also looks for the following files and if present uses their contents (just their first line, usually) instead of the defaults:

    • $selector.method.txt: the 'c' value in the signature, default: relaxed/relaxed
    • $selector.algorithm.txt: the key algorithm ('a' value in the signature), default: rsa-sha256

    There is also a left-over from the old way how the original project better-qmail-remote is based on was configured, see the "XML config file structure" section.

  • the spamcheck script, which is installed in /etc/better-qmail-remote/sbin/ as part of running setup-debounce. This script is called with the path to a mail file, and must print at least an X-Spam-Status:.*score=(\S+) header. You need to make sure that the socketdir path in _spamc.bash is accessible to the qmailr user. Note that Qmail forgoes the system user database, hence adding qmailr to a group will have no effect on permissions. The suggestion is to create /var/qmail/spamd-socket-for-qmailr owned by the spamd user and the qmail group, with -rwxr-x--- (750) permissions (qmailr runs with the qmail group), and then configure spamd to use that directory.

Key generation

If you don't have actual DKIM keys yet, to create RSA keys, you can use the command line tool opendkim-genkey which you can get on Debian sytems via:

apt-get install opendkim-tools

Then run like:

opendkim-genkey  # for options see `man opendkim-genkey`
mv default.private /var/qmail/control/dkim/global.key 
# ^ or replace 'global' with whatever selector you want

If you want to create EdDSA keys, you can use the included script opendkim-genkey-eddsa. For details, have a look at the article How to use DKIM with Ed25519.

Test suite

Run the following from the better-qmail-remote directory to run the test suite. The test suite assumes that the hashcash tool is installed.

make test

Backscatter avoidance

Neither Qmail's own qmail-smtpd nor qpsmtpd check for the existence of an email account for the given address in an incoming mail (no surprise since local delivery can happen in ways that are configured outside the scope of these programs). Qmail hence sends a bounce if the email later turns out to be undeliverable. This is rather bad if the mail is a spam, since now innocent third parties (whose email address was misused for sending out the spam) are getting the bounce. This might lead to your server's future deliveries being penalized.

Better-qmail-remote checks whether the outgoing email has an existing 'X-Spam-Status' header (as it was added by spamassassin when it arrived through e.g. qpsmtpd), and if the score is high-ish (currently hard coded in qmail-remote, "if ($spamscore >= ", separately for the bounce case and other cases), delivers locally instead, currently directly into a Maildir (the path of which is defined in lib/Spambounce_config.pm).

There's a tool included, debounce, that is meant to be run periodically (manually) and which then iterates over the mails in this Maildir, asking for each mail whether it's spam or ham, and uses that information to train spamassassin on the original incoming mail. For the actual training, it calls program names 'spam' and 'ham' with the path as only argument. You need to provide scripts with these names, reachable through $PATH. See the examples in config-template/sbin/.

To set up the backscatter trap, the Maildir at /var/qmail/Maildir_spambounce needs to be created (currently the path is hard-coded in lib/Spambounce_config.pm; please tell if and how you would like to have it changeable). This is done by running the setup-debounce script as root.

NOTE: (some?) recent Linux kernels come with link protections (CONFIG_AUDIT ?) enabled, which leads to the script failing with an error message saying 'xlink' .. 'permission denied'. This is due to the (non-root) user not owning the mail files. This protection can be disabled like:

echo 0 > /proc/sys/fs/protected_hardlinks  # and echo 0 > /proc/sys/fs/protected_symlinks ?

(Presumably this could also be made selective per binary/script (SELinux or AppArmour)?)

The scripts bounce-original and doublebounce-original simply extract the original incoming mail from a (double) bounce. They are provided as utilities for special circumstances.

XML config file structure

The original dkim script from qmailtoaster had code to read an XML file that configures the DKIM keys to be used. This code was left in, but the author is not currently using it (instead better-qmail-remote prefers a different approach, see "Configuration" section above), so it's untested. Here's some principles I gleaned from it before the author of this project realized that he didn't need it:

  • missing settings will be merged from the global-node

  • domain-entry will also match its subdomains

  • create empty domain-node to omit signing (or specify "none" as id)

     <dkimsign>
   <!-- per default sign all mails using dkim -->
   <global algorithm="rsa-sha256" domain="/var/qmail/control/me" keyfile="/var/qmail/control/dkim/global.key" method="simple" selector="beta">
     <types id="dkim" />
   </global>

   <!-- use dkim + domainkey for example.com -->
   <example.com selector="beta2">
     <types id="dkim" />
     <types id="domainkey" method="nofws" />
   </example.com>

   <!-- no signing for example2.com -->
   <example2.com />
 </dkimsign>

Also have a look at contents of the file install.sh, which is also a left-over from the original project. Don't just run it, it will be out of date. NOTE: this has not been tested since making the changes to the original script. TODO: update or merge/drop install.sh.

spf-forward script

Also included is a script spf-forward that can be used as a replacement for Qmail's forward program, but does sender rewriting so that SPF checks don't break.

Its usage is: spf-forward newsender_pre newsender_post address(es), for example use with a line in a .qmail (see man dot-qmail) file like this:

 |/opt/better-qmail-remote/spf-forward forwarder- @example.com youremailaddressat@gmail.com

A mail with Return-Path: <foo@bar.com> will now be forwarded with a new sender, Return-Path: <forwarder-foo=bar.com@example.com>

Note that this doesn't follow generic SRS schemes: there's no code to handle bounces. The assumption is that you set up a .qmail-forwarder-default file that will forward bounces to an admin. It's then that admin's responsibility to figure out the new target address, or to inform the original sender of the mail about the failure (and remove the forwarding).

spf-forward-mailfile script

While spf-forward is for automatic use within Qmail's processing chain and would be tedious to use manually, there is a script bin/spf-forward-mailfile which wraps spf-forward to allow for forwarding of mails stored in files (as in Maildirs). It uses the EMAIL environment variable to construct the newsender_pre newsender_post parts automatically, blindly assuming that the user part of your email can be used with "-" and the mangled original sender email appended. If you want to see what it runs, set the VERBOSE env var to 1.

Hacking

If BETTER_QMAIL_REMOTE__DEBUG is set to a true value (for Perl, e.g. 1), as is done by the test suite, a file is written with the debugging output and the file path is written to stderr.

Links

Possibly useful links:

Mailing lists: