Make a new directory on your Windows computer and call it 'debian' or
something. Then download the latest version of the Debian installer for 'etch' and save
it there. Go to:
http://www.debian.org/releases/etch/debian-installer/.
Read the errata while you are on that page. One interesting
errata is
http://bugs.debian.org/cgi-bin/bugreport.cgi?bug=401435. Note that there are etch 4.0r7 i386 or etch 4.0r7 ia64 or etch 4.0r7 amd64 CDs available from this location but unfortunately I have only tested this setup using the i386 CD (32bit). By default it installs the Linux kernel version 2.6.18. |
If using RAID1, you should have both drives installed. We are going to erase the hard drive so make sure you don't have any data on it you might need. You may have to change the order of boot devices in your BIOS before we begin. Boot up the computer using the Installer CD or the Installer floppy #1. If you use the floppy to boot up, it will prompt you for remaining floppies. I recommend the CD-ROM. The instructions below pertain to the CD-ROM method in the default "ask as few questions as possible" mode. Set the correct date and (local) time in the BIOS before you start. When the system boots up to the Debian screen, simply press [Enter] at the boot: prompt. |
[ !! Choose Language] This determines the language of the installer and picks a keyboard. This installation has only been tested with English - English [Choose country, territory or area] Choose what is appropriate |
Unplug the ethernet cable. |
[! Select a keyboard layout] American English selects a standard qwerty keyboard. |
There will be a few screens of activity, then this will come up: [Configuring the network with DHCP] Hit [Cancel] because we want DHCP configuration to fail. |
Plug the ethernet cable back in. |
[!! Configure the network] Network autoconfiguration failed We wanted that to happen, simply press: [Continue] On the next screen, choose the default of: [Configure network manually] |
[!! Configure the network] Make sure Num Lock is on! [IP address:]
111.111.111.111
[Netmask:]
255.255.255.x
[Gateway:]
333.333.333.333
[Name server addresses:]
444.444.444.444 555.555.555.555
[Hostname:]
msa
[Domain name:]
example.com
|
For partitioning hard drives, I am going to use software RAID1. I suggest using a pair of 320GB or larger hard drives. Another possibility is using four drives and allocating an entire pair of drives to the mail store: /var/vmail. See: http://verchick.com/mecham/public_html/raid/raid1.html |
[! Configure time zone] [Select your time zone:] Simply choose what is appropriate. |
[!! Set up users and passwords] This will ask for root's password and allow you to create a "normal" user and a password for that user. Watch your [Num Lock] status. Use really good passwords and don't forget them. Please add one, and just one, normal user here. Create a user who's main purpose in life might be to hold root's mail. I suggest calling the user myroot or something similar. |
[Installing the base system] Wait.... |
[! Configure the package manager] [Use a network mirror?] Choose [Yes] [Debian archive mirror country:] Choose your country [Debian archive mirror:] Choose a mirror near you (mirrors.kernel.org works very well in the US) [HTTP proxy information] (configure if needed, otherwise leave unconfigured) |
[! Configuring popularity contest]
You decide if you would like participate. |
[Debian software selection]
[Choose software to install:] You only want to select 'Standard system' here (nothing else). Use the space bar to deselect 'Desktop environment' then simply [Tab] over and select [Continue]. |
[Configuring console data] IMPORTANT! choose "Don't touch keymap" |
[Configuring Exim v4 (exim4-config)]
[General type of mail configuration:] choose [no configuration at this time] [Really leave the mail system unconfigured?] [Yes] [Root and postmaster mail recipient:] The "normal" user we added earlier will display here. This is fine, so simply accept this. All of root's mail will be redirected to this "normal" user's mailbox. This is necessary because you typically cannot access root's mailbox remotely. |
[! Install the GRUB boot loader on a hard disk] [Install the GRUB boot loader to the master boot record?] If you would like the install the (recommended) GRUB boot loader choose [Yes] If you would like the install the LILO boot loader [Tab] over and select [Go Back] Then select the 'Install the LILO boot loader...' [Finish the installation] Remove the CD or floppy when prompted, then hit [Continue] This will reboot. |
Once the system is installed, login as root and issue the following command:
apt-get install ntpdate ssh vim
|
i
",
(short for "insert"). You can edit text pretty much as you would expect in
Write mode. You exit out of Write mode and return to Command mode by hitting
the [Esc] key. There are many commands that can be learned in Command mode but
we only need to learn two more in addition to "i
". Those commands
are ":
" (a colon) and "/
" (a forward slash). The
colon is used to enter the third mode, the Command line mode and the slash
enables the Search command. When you are in Command line mode, you will see a
colon at the bottom of the screen. Here is a list of commands we will use while
in Command line mode:
:q
quit (provided you have not made any changes) By the way,
the lower case q is used often in *nix as a way to exit a screen. :q!
exits vi and discards changes (great when you trashed the
file and just want to start over!) :wq
saves the changes and exits vi (write and quit) :w
saves the current changes but does not exit vi (write) G
The capital "G" Goes to the bottom of the page (very handy)
/text_to_search_for
moves the cursor to the first occurrence of
text_to_search_for
ntpdate clock.fmt.he.net
|
Enter the following command:
dpkg-reconfigure locales
[Configuring locales] You use [PgUp] [PgDn] [up-arrow] [down-arrow] [tab] and [spacebar] to navigate and select. The etch installer software installed en_US.UTF-8 UTF-8 on my system. I suggest you install the en_US ISO-8859-1 locale (in addition to any other ISO-8859-x locales you may require). If you need to change the locale, or add additional locales, use the [arrow] [spacebar] and [tab] keys. An UTF-8 locale should not be used as the default system LANG (set in /etc/environment or /etc/default/locale), SpamAssassin and amavisd-new may have problems if you do. However, you should keep the UTF-8 locale in addition to the ISO-8859-x file or Perl may complain. [Which locale should be the default in the system environment?] I suggest you do NOT choose [None], I suggest you choose [en_US] or other non UTF-8 locale (an ISO-8859-x locale). |
Our default language is currently an UTF-8 locale.
We want our system wide language to be an ISO-8859-x (non UTF-8) locale. You can
set the language in /etc/environment (if it exists, otherwise it is set in
/etc/default/locale). This file is read when we log in. We need to use a non
UTF-8 locale so characters will appear as we expect them to and to avoid other
problems. It is best to run amavisd-new in a non-UTF8 locale environment.
The 'dpgk-reconfigure locales' program previously automatically updated
/etc/environment, but it no longer does when using the etch version so we are
going edit it manually (it now updates /etc/default/locale). Make sure you have
installed a corresponding ISO-8859-x locale for the UTF-8 locale we are going
to change:
cat /etc/environment
If the above returns "No such file or directory", then the setting is in /etc/default/locale and you can skip editing this file, otherwise please continue.
vim /etc/environment
Change LANG from a UTF-8 setting:
LANG="en_US.UTF-8"
to a non UTF-8 setting:
LANG="en_US"
Save and exit the file: [Esc]:wq to 'write and quit' or [Esc]:q to quit without saving (so you can give it another try). Note: you can run the command locale to see the current settings. It is best to reboot after changing the /etc/environment file. Changes are not recognized until you at least log out, then back in. |
There is an errata dealing with tcp_window_scaling on Linux kernel 2.6.17 (and newer). http://kerneltrap.org/node/6723 http://marc.info/?l=postfix-users&m=117457942431349 You may want to consider what may happen (large files fail to transfer between systems) when there is a buggy router between you and someone else, and may wish to make this change to the system (you decide):
echo "net.ipv4.tcp_wmem = 4096 65536 65536" >>/etc/sysctl.conf
I am going to assume this may slow down communications between systems under certain circumstances. |
vim /root/.profile
and just below the line "fi" insert this entry:
export EDITOR=/usr/bin/vim.basic
Save and exit the file: [ESC]:wq |
vim /etc/apt/sources.list
At this point, the contents of the file may look something like this: # # deb cdrom:[Debian GNU/Linux testing _Etch_ - Official Snapshot i386 Binary-1 (20061111)]/ etch main deb cdrom:[Debian GNU/Linux testing _Etch_ - Official Snapshot i386 Binary-1 (20061111)]/ etch main deb http://mirrors.kernel.org/debian/ etch main deb-src http://mirrors.kernel.org/debian/ etch main deb http://security.debian.org/ etch/updates main deb-src http://security.debian.org/ etch/updates mainWe need to modify this file so the result will look something like this: (with only the http server unique to your particular system) deb http://mirrors.kernel.org/debian/ etch main non-free contrib deb-src http://mirrors.kernel.org/debian/ etch main deb http://security.debian.org/ etch/updates main non-free contrib deb-src http://security.debian.org/ etch/updates main deb http://volatile.debian.org/debian-volatile etch/volatile mainNote what I have done here. Any lines that use the cdrom
#deb cdrom:[Debian GNU/Linux testing _etch_
have been erased and the words "non-free" and "contrib" have been added. Debian Volatile has also been added. |
gpg --keyserver subkeys.pgp.net --recv-key BBE55AB3
|
apt-get update
|
apt-get upgrade
|
reboot
|
apt-get remove nfs-common pidentd portmap
|
If you are using a multi-processor machine, then use a multi-processor kernel.
To locate available smp kernels for etch, you could run:
apt-cache search linux-image | grep smp | grep linux-image
If you are running a 2.6.18 (etch) kernel, and have a dual core Intel system, you could for example use the 'linux-image-2.6-686-smp' kernel. You would pick the kernel that most closely matches your system (and your current kernel). To install it, you would simply run (for example): apt-get install linux-image-2.6-686-smp If you installed a new kernel, please reboot afterwards.
|
There is a problem I describe
here. Here I attempt to fix it:
apt-get install yaird
Make sure it shows you have a good initrd.img that matches the other stuff:
-rw-r--r-- 1 root root 70781 2007-05-09 16:14 config-2.6.18-4-686 drwxr-xr-x 2 root root 4096 2007-06-10 09:25 grub -rw------- 1 root root 1135592 2007-06-19 20:15 initrd.img-2.6.18-4-686 -rw-r--r-- 1 root root 4490534 2007-06-19 20:14 initrd.img-2.6.18-4-686-backup -rw-r--r-- 1 root root 4490534 2007-06-10 09:25 initrd.img-2.6.18-4-686-backup2 -rw-r--r-- 1 root root 722037 2007-05-09 22:14 System.map-2.6.18-4-686 -rw-r--r-- 1 root root 1261213 2007-05-09 22:14 vmlinuz-2.6.18-4-686Then reboot:
reboot
|
iptables -F
|
vi /etc/network/interfaces
And insert the following text in the blank line just below "iface lo inet loopback":
pre-up iptables-restore < /etc/firewall-rules
While you are at it, 2 lines down, change allow-hotplug eth0 to:
auto eth0
Save and exit the file, then reboot:
reboot
|
iptables -F
from the
console to clear out iptables. This will allow you another shot at it.
apt-get install ntp make gcc bison flex libc6-dev logcheck logcheck-database flip psmisc dpkg-dev
For more information about logcheck rules and patterns to include or ignore:
vi -R /usr/share/doc/logcheck-database/README.logcheck-database.gz
and to debug logcheck:
su -s /bin/bash -c "/usr/sbin/logcheck -otd" logcheck
|
apt-get install mysql-server
Out of the box MySQL is tuned for a system with very little memory. We need to allocate more memory for caching items. This will make a huge difference in performance. We are going to change all tables to InnoDB so we will adjust some InnoDB settings. If you currently have data in MySQL or you have made changes to /etc/mysql/my.cnf you should not perform these steps to replace your my.cnf file with mine! Also, I am assuming you have at the very least 1GB of physical memory installed (2GB recommended - 3GB even better):
cp /etc/mysql/my.cnf /etc/mysql/my.cnf-original
Add a password for user 'root' (make sure the hostname is correct):
mysql -u root
From the mysql> prompt, create the two passwords required for root
(bold text items need to be replaced with your personal
settings - you should have edited this document already):
SET PASSWORD FOR 'root'@'localhost' = PASSWORD('roots_password');
Make a mental note of the innodb_fast_shutdown setting. The default is "1". Quit mysql:
QUIT
Since my patch changed the size of innodb_log_file_size, we will have to create new logs. If innodb_fast_shutdown was something other than "1", edit /etc/mysql/my.cnf and add an entry "innodb_fast_shutdown = 1" and then restart mysql with "/etc/init.d/mysql restart". Once innodb_fast_shutdown = 1 then:
/etc/init.d/mysql stop
Make sure that it shuts down without errors (to ensure that there is no information for outstanding transactions in the logs). Then:
mv /var/lib/mysql/ib_logfile0 /var/lib/mysql/iblogfile0-old
It should show our two log files are now 48MB in size (50331648). I have set "innodb_buffer_pool_size = 192M" which will give much better performance than the default of 8M, and "innodb_log_file_size = 48M" which is 25% of the innodb_buffer_pool_size. If you have plenty of RAM, you can increase innodb_buffer_pool_size even more, but personally I would not set it to more than 25% of physical RAM. You may need the extra space for a polar bear or dung beetle. |
The next command goes on one line. Before you issue this command, read the
paragraph following it.
apt-get install apache-ssl libapache-mod-php4 php4 php4-common php4-mysql php4-gd php4-mcrypt mcrypt ca-certificates openssl
You will be asked questions. Sample answers follow. Be sure to use the full name for your state or province name and the host name must be the FQDN host name of this machine (however, the certificate created here will soon be replaced):
Country Name US
Actually, that was just practice. We are not going to use the certificate we just created.
Do not browse to the web server yet. At least, DO NOT install the certificate.
Now install Postfix:
apt-get install postfix postfix-pcre postfix-mysql libsasl2-modules-sql libsasl2-modules
Answer the questions:
General type of configuration? Internet Site
apt-get install courier-imap-ssl courier-pop-ssl courier-authlib-mysql
Create directories for web-based administration ? [No] |
Every client that connects to this server will need to be able to resolve the
hostname of the server. Add an entry to your hosts file
or add an A record to your DNS server so we can properly interact with the server.
We are going to be our own Certificate Authority and sign our own certificates.
These commands are dependent on /etc/ssl/openssl.cnf as supplied by Debian.
We start by making a small change to /etc/ssl/openssl.cnf. We make it
so by default our certificates are good for 10 years instead of 1:
sed -i 's/= 365\t/= 3653\t/' /etc/ssl/openssl.cnf
We will set up a common place to put our certificates:
cd /root
Create a Root Certificate:
openssl req -new -x509 -extensions v3_ca -keyout demoCA/private/cakey.pem -out cacert.pem -days 3653
Enter a passphrase when prompted. You will need this passphrase in the future. What I mean is: make it unique and don't ever loose it. You will be asked questions. Sample answers follow. Be sure to use the full name for your state or province name and the Common Name should be something that describes you as an authority (like Widgits Inc. RootCA):
Country Name US
This process produces two files as output: a private key in demoCA/private/cakey.pem and a root CA certificate in cacert.pem. Any and all key files we produce will need to be protected from unauthorized persons reading them, and must not be lost for the next 10 years. Also realize that the CA you created can sign any number of certificates (until it expires 10 years from now) so you only need to (or want to) create the CA once. We will copy our cert and our key to something more descriptive:
cp -i demoCA/private/cakey.pem demoCA/private/cakey.example.com.pem
The cacert.example.com.pem and cacert.example.com.crt are copies of our certificate and are the files that can be distributed and installed on the client's machines. Windows clients would use the .crt file. On my Windows 2000 system, double clicking this file would install it in Internet Explorer (which is exactly what want). Simply browsing to our server will give us the opportunity to install the web server certificate we will create (this will be the Common Name msa.example.com) but this is not the same as installing the CA certificate in the Trusted Root Certification Authorities store (seen as the Common Name you entered above). Just in case you are not familiar, in IE6 it's Tools->Internet Options->Content->Certificates->Trusted Root Certification Authorities. Outlook and Outlook Express use the same certificate store as Internet Explorer. In Mozilla Thunderbird it's Tools->Options->Privacy->Security->View Certificates->Authorities. In Firefox it's Tools->Options->Advanced->Encryption->View Certificates->Authorities->Import. If you go through this process more than once while testing, don't install duplicate certificates. Delete any old 'test' certificate you previously installed before adding your new one that replaces it. In my old version of The Bat! I add a new contact in the "Trusted Root CA" section of the address book and import the certificate from there. I suggest using WinSCP to transfer the cacert.example.com.crt certificate to your machine. I think the worst part of getting this server set up is getting the CA certificates installed on the clients. Sometimes it's worth it to buy a certificate from a well known commercial CA that is already in the Trusted Root Certification Authorities store. We are now going to create a request for a certificate from the CA (which is us - but could be a commercial CA if you like). Everyone that connects to us will connect to the hostname of this machine. The Secure Web server, Secure IMAP server, Secure POP server and Postfix Secure SMTP server will all be msa.example.com, so the Common Name MUST BE our FQDN hostname when we create the request. The Organization name needs to be the same as the one in the CA cert. Do not enter your email address, challenge password or an optional company name when generating the CSR:
openssl req -new -nodes -out req.pem
Country Name US
This process produces two files as output, a private key in privkey.pem and a certificate signing request in req.pem. These files should be kept. The private key is of course necessary for SSL encryption. We will make backup copies of these files with more descriptive names:
cp -i privkey.pem privkey.msa.example.com.pem
Sign the Certificate (you will be asked for the pass phrase):
openssl ca -out cert.pem -cert cacert.pem -infiles req.pem
This process updates the CA database and produces two files as output, a certificate in cert.pem and a copy of the certificate in demoCA/newcerts/ named xx.pem, where xx is the serial number. We will copy the cert to a more descriptive name. The certificate has both the encoded version and a human-readable version in the same file. We want to strip off the human-readable portion as follows:
mv -i cert.pem temp.cert.msa.example.com.pem
Postfix will want the cert and the key in two separate files, apache-ssl will want the two combined (but can use two separate files if configured to do so). Courier will want the two combined, with a Diffie-Hellman code block added. Sheesh, why can't we all just get along?
cat privkey.msa.example.com.pem cert.msa.example.com.pem >key-cert.pem
After those steps, you have four installable components (and some more descriptive backup copies): A private key in privkey.pem (with a copy in privkey.msa.example.com.pem) A certificate in cert.pem (with a copy in cert.msa.example.com.pem) A combined private key and certificate in key-cert.pem (with a copy in key-cert.msa.example.com.pem) A combined private key, certificate and DH code in key-cert-dh.msa.example.com.pem Now give a copy of the combined certificate to apache-ssl. Apache-ssl is currently using the /etc/apache-ssl/apache.pem certificate (the SSLCertificateFile setting in /etc/apache-ssl/httpd.conf). We will reconfigure it to use /etc/apache-ssl/key-cert.msa.example.com.pem :
/etc/init.d/apache-ssl stop
Give Postfix the files it needs and tell it where they are (and set a couple other TLS parameters). We also make a backup of main.cf before we modify it for the first time:
cp -i /etc/postfix/main.cf /etc/postfix/main.cf-24jul2007
Install certificates in Courier IMAP and configure imapd-ssl to use them (this changes the TLS_CERTFILE setting):
cp -i key-cert-dh.msa.example.com.pem /etc/courier/
And configure Courier POP3 to use the same certificates:
sed -i 's/pop3d.pem/key-cert-dh.msa.example.com.pem/' /etc/courier/pop3d-ssl
Running some of these commands again will result in overwriting keys and certificates. That may not be good. Some files will necessarily be overwritten if additional certificates are requested, signed and created. That is expected, and is the reason we make host-specific copies of everything as we go along. Just be careful not to overwrite any host-specific files we have created. And remember, only one Root Certificate Authority needs creation. Make a backup of the session, both on and off the system (transfer the directory via WinSCP).
cp -r /root/CA /root/CA-24jul2007
Note that if you create certificates for additional hosts and want to provide SSL for multiple hosts via the VirtualHost directive, it is my understanding you will need a separate IP address for each host: http://httpd.apache.org/docs/trunk/ssl/ssl_faq.html#vhosts. For example, I created a request for a new cert for mailzu.domain.tld. Then I signed it using our CA cert, stripped off the human readable portion, combined the key and certificate into a single file called key-cert.mailzu.domain.tld.pem and copied this certificate to the /etc/apache-ssl directory. Later we will install MailZu, which has a home directory of /var/www/mailzu. If you wanted to access /var/www/mailzu using a virtual host (using SSL) via https://mailzu.domain.tld, you would have to add another IP address. Here I show adding one to an existing interface that has the IP address 192.168.1.222 - (this is in /etc/network/interfaces): auto eth0 iface eth0 inet static address 192.168.1.222 netmask 255.255.255.0 network 192.168.1.0 broadcast 192.168.1.255 gateway 192.168.1.1 auto eth0:0 iface eth0:0 inet static address 192.168.1.223 netmask 255.255.255.0 network 192.168.1.0 broadcast 192.168.1.255 gateway 192.168.1.1Then I created a file called /etc/apache-ssl/vhosts.conf with the contents: <VirtualHost 192.168.1.223:443> DocumentRoot /var/www/mailzu ServerName mailzu.domain.tld SSLCertificateFile /etc/apache-ssl/key-cert.mailzu.domain.tld.pem </VirtualHost>I personally do not go this route (who has IP addresses to spare?), so I do not do this. I choose to access MailZu via https://msa.example.com/mailzu. This means I only have to deal with one host certificate (and one IP address). BTW, I reboot after making changes to /etc/network/interfaces. |
The next command is on one line. We will use sed to edit a file that will enable php in apache-ssl.
sed -i 's|#AddType application/x-httpd-php .php|AddType application/x-httpd-php .php|' /etc/apache-ssl/httpd.conf
If you do not get:
LoadModule php4_module /usr/lib/apache/1.3/libphp4.so
Then you need perform the next step:
apache-modconf apache-ssl enable mod_php4
Regardless, restart apache-ssl:
/etc/init.d/apache-ssl restart
Download a home page. This page describes some of the features of the mail system to new users. Those features are not yet installed but we can install this page. One it is in place, edit it (you may prefer the WinSCP editor) and do a search and replace on the items listed at the top of the page. No doubt once this entire system is installed you will do a lot more editing of this page:
cd /var/www
Browse to the server (first add the address to DNS or your hosts file):
https://msa.example.com
|
apt-get install phpmyadmin
At this point our https://msa.example.com/phpmyadmin program is open for abuse. We will secure it by making the URL obscure, by setting a user name and password and by optionally (but highly recommended) only allowing chosen IP addresses to access the URL. Obscure the URL by changing its name (which should have already been done by editing this page):
mv /var/www/phpmyadmin /var/www/phpmyadmiNx
With the current settings in /etc/apache-ssl/httpd.conf, our files used for access control (files like .htpasswd) will not be used unless we tell apache-ssl to use them. We will modify the provided access control file, then change /etc/apache-ssl/httpd.conf so it uses it. The current user name is 'admin'. Let's begin by obscuring the user name:
sed -i 's/admin/myadmin_username/' /etc/phpmyadmin/htpasswd.setup
Then, create a password for that user:
htpasswd -c /etc/phpmyadmin/htpasswd.setup myadmin_username
New password: myadmin_passwordRe-type new password: myadmin_password Now we direct apache-ssl to use this access file by placing directives in a configuration file we create. Once this is in place, attempts to browse to the /phpmyadmiNx directory will be met with a login dialog box. Note that our version of /etc/apache-ssl/http.conf contains the command "Include /etc/apache-ssl/conf.d" so that any config files we place in this directory will be read. I am also going to illustrate limiting access to one single workstation (yours of course). This is optional, but recommended. If you need to allow access to a network of machines, see http://httpd.apache.org/docs/1.3/mod/mod_access.html:
vi /etc/apache-ssl/conf.d/phpmyadmin.conf
and insert the following, making sure the IP address is the IP address of your workstation (as the mailserver sees it). Also make sure any comments start at the leftmost column. <Directory /var/www/phpmyadmiNx/> order deny,allow deny from all allow from 666.666.666.666 #allow from 666.666.666.666 allow from 192.168 AuthUserFile /etc/phpmyadmin/htpasswd.setup AuthGroupFile /dev/null AuthName "phpMyAdmin" AuthType Basic require valid-user </Directory>Restart apache-ssl:
/etc/init.d/apache-ssl restart
and browse to phpMyAdmin:
https://msa.example.com/phpmyadmiNx
The first login is the phpMyAdmin user name and password: myadmin_username myadmin_password The second login is your mysql login (probably root and roots_password). You can close the phpMyAdmin web page. |
We want to explicitly set our domain name and host name in Postfix so there is no
possibility Postfix finds something else:
postconf -e "mydomain = example.com"
At this point, if you issue this command:
postconf -n
Postfix should show main.cf is configured something like this: alias_database = hash:/etc/aliases alias_maps = hash:/etc/aliases append_dot_mydomain = no biff = no config_directory = /etc/postfix inet_interfaces = all mailbox_command = procmail -a "$EXTENSION" mailbox_size_limit = 0 mydestination = example.com, msa.example.com, localhost.example.com, localhost mydomain = example.com myhostname = msa.example.com mynetworks = 127.0.0.0/8 myorigin = example.com recipient_delimiter = + relayhost = smtp_tls_session_cache_database = btree:${queue_directory}/smtp_scache smtpd_banner = $myhostname ESMTP $mail_name (Debian/GNU) smtpd_tls_CAfile = /etc/postfix/ssl/cacert.example.com.pem smtpd_tls_cert_file = /etc/postfix/ssl/cert.msa.example.com.pem smtpd_tls_key_file = /etc/postfix/ssl/privkey.msa.example.com.pem smtpd_tls_received_header = yes smtpd_tls_session_cache_database = btree:${queue_directory}/smtpd_scache smtpd_use_tls = yesPlease note that 'postconf -n' does not show every setting we have in main.cf. Postfix should accept mail for the addresses listed in /etc/aliases and deliver it to a mailbox in /var/mail. You should test this. Configure a MUA to use this mail server as its outgoing SMTP server. I like to designate a MUA I'm not using at the moment for testing purposes. For test purposes I typically I have several different MUAs (Outlook, Outlook Express, Thunderbird, The Bat!) set up on the system I am sending mail from. Do not send mail from the command line (using sendmail or other means) when testing this system. In all tests we perform I expect you to send test messages from an external client (unless of course we are testing something like SquirrelMail). Now send a message to postmaster@example.com. See what /var/log/mail.log said about the transaction (look for errors). It should also show you the user the message was delivered to:
tail -50 /var/log/mail.log
In the /var/mail directory you should see the mbox of the user the message was delivered to. You can ' more /var/mail/user ' to read the contents of the mbox. Grab the Postfix source code (we need a few samples from it):
cd /usr/local/src
Make sure you answer "n" to "Overwrite?". Do the first command separately:
cp -i /usr/local/src/postfix-2.3.8/conf/* /etc/postfix
On Debian, Postfix runs chrooted. The LINUX2 script is used to copy files to the chroot jail. |
We just need the software in place at this time.
apt-get install squirrelmail squirrelmail-locales maildrop
|
Add our virtual user and group:
groupadd vmail -g 6060
And create the directory where our mail will be stored:
mkdir /var/vmail
Install subversion:
apt-get install subversion
Download revision 1 of the source code (from the SVN repository):
cd /var/www
Just like phpMyAdmin, we obscure the postfixadmin URL. Run these commands in sections - but don't manually change to a different directory during the process. I hope you have saved this document to your computer and have done a search and replace on the bold items. This document is for you to customize to your particular system. Read the instructions at the top of the page once you have opened it in a plain text editor (like WordPad):
mv postfixadmin postFixadminx
cd /var/www/postFixadminx cp -i DATABASE_MYSQL.TXT DATABASE_MYSQL.TXT~You should vi config.inc.php and browse through it to familiarize yourself with all the possible settings and to make sure your domain name was properly updated. Create a .htaccess password for the admin url (user name will be pfadmin_username). Assign the password pfadmin_password
cd /var/www/postFixadminx/admin
Now tell apache-ssl to use the file. We also limit access to our own workstation, but you can add more IP addresses (or networks) if needed. Access to https://msa.example.com/postFixadminx/admin/ is controlled by the IP address(es) of the client and the .htaccess user name and password. The super user is the only one who should have access to this URL.
vi /etc/apache-ssl/conf.d/postfixadmin.conf
and insert this phrase. Don't forget to edit the ip address if you have not already done so and remember that comments must left justified: <Directory /var/www/postFixadminx/admin/> order deny,allow deny from all allow from 666.666.666.666 #allow from 666.666.666.666 allow from 192.168 AuthUserFile /var/www/postFixadminx/admin/.htpasswd AuthGroupFile /dev/null AuthName "Postfix Admin" AuthType Basic require valid-user </Directory>Restart apache-ssl:
/etc/init.d/apache-ssl restart
When we add a user to postfixadmin it creates records in the MySQL database but it does not create a Maildir for that user. We can make that happen with a patch to postfixadmin and a couple bash scripts. We also apply a patch to create new MySQL data in lower case rather than mixed case. Another patch is a group of patches others have submitted to the project:
cd /var/www/postFixadminx/
There is also a small optional patch that allows admins to create quotas larger than the default maxquota.
cd /var/www/postFixadminx/
In order to allow the www-data user to run the scripts, you must run:
visudo
and on the bottom of the page add this stuff (notice our hostname):
www-data msa=NOPASSWD: /usr/sbin/maildirmake.sh
Set permissions:
cd /var/www/postFixadminx
Now you can browse to postfixadmin and run the setup script:
https://msa.example.com/postFixadminx/setup.php
Assuming everything is OK, click on the link to the admin section and get past the apache security with pfadmin_username and pfadmin_password . You will need to log into the /admin page with a user name of admin@domain.tld and a password of admin (note: domain.tld is literally domain.tld). All 'Super admins' are directed to the /admin page. Click on "New Domain" and add the primary domain (example.com). Leave the optional "Add default mail aliases:" and "Mail server is backup MX:" boxes unchecked. All domains you add should have either an A record or MX record in DNS in order for PostfixAdmin to consider them valid. Add any other domains you need. Only add domains that will store mail locally (we will talk about relay domains later). Now click on "New Admin" and add your email address and a strong password. You are not adding a mailbox here, you are adding yourself as an administrator. Once you are added, you need to promote yourself to a Super admin. To do so, click "Admin List", edit your login and check the "Super admin" checkbox. You do not need to select any domains when you are a Super admin. Only select a domain (or domains) when adding a normal admin. Now Logout and log back in using your user name. Please delete admin@domain.tld otherwise the whole world knows your Super admin login. Only Super admins have access to the /admin page. You can add one or more 'normal' admins at this time if you like. You actually need to add an additional normal admin login for yourself that you will use on a day to day basis. Normal admins get a different screen. The https://msa.example.com/postFixadminx/ will be used by normal administrators and access to this URL will be controlled by IP address of the client and the user name and password that is assigned to the administrator by us (using the tools at https://msa.example.com/postFixadminx/admin). Users can change their forwarding information and password at https://msa.example.com/postFixadminx/users; however, they will not need to once we add a SquirrelMail plugin. That's all we'll do here for the moment, but leave the current page open. Return to the PuTTY screen and rename setup.php:
cd /var/www/postFixadminx
Back at the browser window you will notice a link to Return to https://msa.example.com/postFixadminx. Clicking on this should bring you to the normal admin page where you (and possibly other domain admins) would log in to manage your domain(s). The first thing you should do is Add Mailbox for yourself. The "Active:" and "Create mailbox:" check boxes should of course be checked. Notice that Username: is only the local part; you choose the @domain from the drop down box. Create the username in lower case. This is your IMAP account. It is imperative that our maildirmake.sh script runs correctly when a mailbox is created in postfixadmin, otherwise Postfix will accept the message but will not be able to deliver it to a mailbox. Check that a mailbox exists, and a squirrelmail profile was created:
ls -l /var/vmail/example.com
Note that if you delete a mailbox from postfixadmin, as a safety precaution we will not automatically delete the user's maildir (or mail). You will have to manually remove it. You should add another regular user "test@example.com" to use when you need to test sending and receiving messages as a normal user (not related to any administrator accounts). At this point we still need to configure Postfix to send mail to our virtual mailboxes using the information stored in the postfixadmin databases. Begin by downloading and modifying Postfix data access configuration files:
cd /etc/postfix
We need to remove our domain name from $mydestination because our domain will soon be listed as a virtual mailbox domain - and you cannot have a domain in more than one address class:
cp /etc/postfix/main.cf /etc/postfix/main.cf-domain
Now tell Postfix to use our MySQL data files (and maildrop - which still needs configuration):
touch /etc/postfix/virtual
Note that I use proxy:mysql: here (proxymap(8)). When proxymap(8) is used, changes to the MySQL tables may not be recognized immediately. During testing you may want to remove proxy: because it may cause frustration and confusion. Alternately, reload Postfix so changes are recognized sooner. To configure maildrop, first get a new maildroprc file from me that contains instructions to deliver spam to the user's Spam folder:
cd /etc
Now we need to edit the maildrop entry in master.cf. We will also make a backup of the original master.cf:
cp -i /etc/postfix/master.cf /etc/postfix/master.cf-24jul2007
Locate the current maildrop transport and comment it out (as shown) then insert the new maildrop transport as shown: #maildrop unix - n n - - pipe # flags=ODRhu user=vmail argv=/usr/bin/maildrop -d ${recipient} maildrop unix - n n - - pipe flags=ODRhu user=vmail:daemon argv=/usr/bin/maildrop -w 90 -d ${user}@${nexthop} ${extension} ${recipient} ${user} ${nexthop}Then reload Postfix:
postfix stop
You should get a message that files differ; run LINUX2 to correct that. At this point, since we are sending mail to maildrop, maildrop also needs to read the MySQL data in order to determine the relationship between user@example.com and the maildir where mail is supposed to go for that user. Fortunately, both Courier and maildrop can use the same configuration file to store the settings. This file is /etc/courier/authmysqlrc. We need to make quite a few changes to this file to get it to work in our current environment, so rather than edit it, get a new one from me. Note that tabs must be used to separate the variable and the value. Also, read through this file to get an idea of what it contains:
cd /etc/courier
You also need to tell authdaemond to use the authmysqlrc file we just modified. We will use data in our MySQL postfix database to authenticate users for IMAP, POP and SASL. For remote clients to send mail we will use SASL with TLS. We can optionally use CRAM-MD5 for those clients (like my old version of The Bat!) that seem to have a broken TLS implementation. At least the password will be encrypted during its trip across the wire:
sed -i 's/authmodulelist="authpam"/authmodulelist="authmysql"/' authdaemonrc
We will enable CRAM-MD5 login mechanism for imapd (port 143):
cp -ip /etc/courier/imapd /etc/courier/imapd~
Let's simply start off clean, we need to make certain stuff works after a reboot anyway:
reboot
Once the system comes back up:
tail -f /var/log/mail.log
Now we will send a message through the system to see if it lands in our maildir. You should already have a MUA set up to use this server as its outgoing server. You should also now be able to configure it to connect to the IMAP server using your username (full email address) and password. Assuming you have successfully created a maildir for yourself, send a message to yourself and see if you get it. Tail the mail.log file as the message goes through. Also: grep fatal /var/log/mail.log Success should look like: Jun 10 11:21:06 msa postfix/pipe[11513]: 0FBC5240C2: to=<garyv@example.com>, relay=maildrop, delay=0.13, delays=0.07/0.02/0/0.04, dsn=2.0.0, status=sent (delivered via maildrop service) If you have an error, you must fix it before you continue. We have made a lot of changes to a lot of files. It is certainly possible something happened along the way that would prevent proper delivery of a message. You should have an understanding of the files involved. You just have to find the incorrect setting(s). Some familiarity with Postfix would really be handy. Basic stuff like familiarity with the mailq, postsuper, qshape and postqueue commands. Also remember that you have phpMyAdmin to browse the database (and possibly make manual changes). You should browse the postfix database to get familiar with the structure of the data. You will also want to set up the IMAP account for the test@example.com user and send a message to that address.
ls -al /var/vmail/example.com/test/new
If you can deliver mail to users, it would be a good idea to familiarize yourself with PostfixAdmin at this time. Add some aliases, add some domains. Play with the software. Postfix will no longer use /etc/aliases for our virtual domains so you will need to make aliases (or mailboxes) for root, abuse, postmaster, webmaster and logcheck @example.com. If you fail to make aliases or mailboxes for recipients of system generated mail, maildrop will bounce the messages and complain with an 'Invalid user specified.' error. Set up your MUA to retrieve mail from the mailserver via IMAP SSL on port 993. If you must use POP3, use POP3 SSL on port 995. POP3 clients will only be able to retrieve mail from the /new folder which means they will never see their Spam folder unless they also use SquirrelMail. If you must use standard POP (110) or IMAP (143), configure the client to use TLS (if it's an option - some clients use it automatically). Some clients may wish to use CRAM-MD5 to authenticate to port 110 or 143. If so, you will have to fire up phpMyAdmin and place a cleartext password in the 'clear' field for those users. Hopefully you are using my database schema with the 'clear' field added. If not, it's likely maildrop will think every user is invalid. Install the CA certificate we created earlier on the test client if you have not already done so. Not having the root certificate properly installed will cause all kinds of grief. If you need to debug pop3d/imapd/pop3d-ssl/imapd-ssl, edit those files in the /etc/courier directory and add DEBUG_LOGIN=2. Then of course restart any of those services as needed. This will give more details in mail.log. Also check /var/log/auth.log: /etc/init.d/courier-pop restart /etc/init.d/courier-pop-ssl restart /etc/init.d/courier-imap restart /etc/init.d/courier-imap-ssl restart There are three (contributed) scripts that come with PostfixAdmin in the ADDITIONS directory that are used to delete orphaned maildirs. These are mailboxes you deleted in PostfixAdmin but the files remain on the system. As mentioned earlier, I do not automatically delete maildirs when a user is removed. I would leave orphaned files alone for some length of time. One of the provided scripts would delete all your mail if you let it. That can't be good. Of the other two, cleanupdirs.pl appears to be Ok but I'm still not going to allow it do delete maildirs; I'm only going to use it to report orphaned ones. I'm also going to rename it maildircheck.
cp -i /var/www/postFixadminx/ADDITIONS/cleanupdirs.pl /usr/sbin/maildircheck
To test, you need to delete a test user via the PostfixAdmin interface. Then run:
maildircheck
If we deleted test@example.com, look and see (and make sure) that the mail directory was NOT deleted, for example, run ls -l /var/vmail/example.com/test . Then add the user back in. If you actually wanted to use this script to delete the directories, you would have to uncomment the 'rmtree' line. I hope the script works and you don't end up deleting every mailbox on the system. Now I have an optional script that will automatically delete messages left in each user's Spam folder that is older than 24 days. You could test it (at some point in the future) by temporarily modifying the number of days (-mtime 0 = 1 day).
cd /etc/cron.daily
|
vi /etc/postfix/sasl/smtpd.conf
and insert the following:
pwcheck_method: authdaemond
The first three lines will authenticate against encrypted passwords in the 'password' field of the mailbox table in the postfix database (when a client and the server use PLAIN or LOGIN). The remaining lines are for CRAM-MD5 logins that will instead look up the cleartext password from the 'clear' field. Look in /var/log/auth.log for error messages. I reduce the log_level: to 0 when done testing. Passwords you enter in PostfixAdmin are encrypted but you will have to manually add any cleartext passwords. This can be done with an SQL statement or via the phpMyAdmin interface. The cleartext passwords do not have the be the same as the encrypted passwords - the client simply has to use the correct password for the chosen mechanism. Make sure we are not using saslauthd (don't worry if /etc/init.d/saslauthd does not exist):
sed -i 's/START=yes/START=no/' /etc/default/saslauthd
Since Debian runs Postfix chrooted, Postfix will need to find /var/run/courier/authdaemon/socket in the chroot jail. We will do this by making a hard link to the existing file:
mkdir -p /var/spool/postfix/var/run/courier/authdaemon
We have to recreate this link to 'socket' prior to Postfix starting up so we will modify the init script:
vi /etc/init.d/postfix
and just after # Make sure that the chroot environment is set up correctly.
(around line 45) insert the following:
ln -f /var/run/courier/authdaemon/socket /var/spool/postfix/var/run/courier/authdaemon/socket
Now tell Postfix to use SASL (we make a backup copy of main.cf should you need to refer to it):
cp -i /etc/postfix/main.cf /etc/postfix/main.cf-before-sasl
If SASL auth stops working with "warning: SASL authentication failure: cannot connect to Courier authdaemond: Connection refused", a likely cause is Postfix can no longer write to the socket (due to running chrooted). Restarting postfix with the modified init script may solve the problem.
/etc/init.d/postfix restart
Now we need to edit master.cf. You may want to use the WinSCP editor for this. You want to replace everything between the two grayed out lines:
vi /etc/postfix/master.cf
smtp inet n - - - - smtpd -o smtpd_use_tls=no -o smtpd_sasl_auth_enable=no # -o receive_override_options=no_address_mappings # If they want to relay, make them use port 587 (submission) or port 465 (smtps) # If using submission port, configure client to use CRAM-MD5 submission inet n - - - - smtpd -o smtpd_use_tls=no -o smtpd_sasl_auth_enable=yes -o smtpd_client_restrictions=permit_sasl_authenticated,reject # -o receive_override_options=no_address_mappings # Outlook and OE (and many others) expect smtpd_tls_wrappermode, # so have them submit here (port 465): smtps inet n - - - - smtpd -o smtpd_enforce_tls=yes -o smtpd_tls_wrappermode=yes -o smtpd_sasl_auth_enable=yes -o smtpd_client_restrictions=permit_sasl_authenticated,reject # -o receive_override_options=no_address_mappings # We will use port 4650 for clients that use STARTTLS: 4650 inet n - - - - smtpd -o smtpd_enforce_tls=yes -o smtpd_sasl_auth_enable=yes -o smtpd_client_restrictions=permit_sasl_authenticated,reject # -o receive_override_options=no_address_mappings #628 inet n - - - - qmqpdThen of course:
/etc/init.d/postfix restart
FYI, at this point output from 'postconf -n' looks like: alias_database = hash:/etc/aliases alias_maps = hash:/etc/aliases append_dot_mydomain = no biff = no broken_sasl_auth_clients = yes config_directory = /etc/postfix inet_interfaces = all mailbox_command = procmail -a "$EXTENSION" mailbox_size_limit = 0 mydestination = msa.example.com, localhost.example.com, localhost mydomain = example.com myhostname = msa.example.com mynetworks = 127.0.0.0/8 myorigin = example.com recipient_delimiter = + relayhost = smtp_tls_session_cache_database = btree:${queue_directory}/smtp_scache smtpd_banner = $myhostname ESMTP $mail_name (Debian/GNU) smtpd_recipient_restrictions = permit_mynetworks, permit_sasl_authenticated, reject_unauth_destination smtpd_sasl_auth_enable = yes smtpd_sasl_authenticated_header = yes smtpd_sasl_local_domain = $myhostname smtpd_sasl_security_options = noanonymous smtpd_tls_CAfile = /etc/postfix/ssl/cacert.example.com.pem smtpd_tls_cert_file = /etc/postfix/ssl/cert.msa.example.com.pem smtpd_tls_key_file = /etc/postfix/ssl/privkey.msa.example.com.pem smtpd_tls_received_header = yes smtpd_tls_session_cache_database = btree:${queue_directory}/smtpd_scache smtpd_use_tls = yes virtual_alias_maps = mysql:/etc/postfix/mysql_virtual_alias_maps.cf, hash:/etc/postfix/virtual virtual_gid_maps = static:6060 virtual_mailbox_base = /var/vmail/ virtual_mailbox_domains = mysql:/etc/postfix/mysql_virtual_domains_maps.cf virtual_mailbox_maps = mysql:/etc/postfix/mysql_virtual_mailbox_maps.cf virtual_minimum_uid = 6060 virtual_transport = maildrop virtual_uid_maps = static:6060Please note that 'postconf -n' does not show every setting we have in main.cf. You should experiment with SASL and TLS. To help debug (if necessary) you can place "debug_peer_list=666.666.666.666" (the IP of the client sending the messages) and/or "smtpd_tls_loglevel=3" (read this) in main.cf. Remove those settings when you are satisfied. Configure an Outlook Express IMAP account to require authentication and use SMTPS (AKA SSMTP) port 465 for outgoing, and IMAPS (secure IMAP) port 993 for incoming. Similar to this. Most other clients will use 465 also. Clients that can't use a dedicated TLS port (Postfix offers this via smtpd_tls_wrappermode=yes) may use port 4650 and instead use STARTTLS. Port 587 will be used for clients unable to get TLS working and will instead use CRAM-MD5 to encrypt the cleartext password. There is an issue with Mozilla Thunderbird. It tries to use CRAM-MD5 before it tries PLAIN. This will result in a "SASL authentication failure: empty secret" warning unless you enter a cleartext password for these clients. If you don't wish to do that, you can either fix Thunderbird, or not use CRAM-MD5. To fix Thunderbird, go to Tools->Options->Advanced->General->Config Editor and then double click on mail.smtpserver.default.trySecAuth (which is set to true by default) in order to set it to false. |
Here we install amavisd-new, SpamAssassin and ClamAV, then add the clamav user
to the amavis group. We update SpamAssassin's rules. Then we enable
spam and virus scanning in amavisd-new:
apt-get update
unrar (rar) is not free. See
http://www.rarsoft.com/index.htm
gpasswd -a clamav amavis
Note: during the time amavisd-new is restarting, mail cannot be delivered to it. Also note that amavisd-new may not be able to use the UNIX socket at /var/run/clamav/clamd.ctl until clamd has fully loaded the virus definition database - which can take minutes. Once this server has mail flowing through it, during the time amavisd-new is reloading Postfix may complain "connect to localhost[127.0.0.1]: Connection refused". Postfix will defer this mail (for about 15 minutes). To speed things up, an impatient person may run 'postfix flush' to flush the deferred queue, but I would not. We will install a script that runs sa-update daily:
cd /usr/sbin
And apply a couple small patches to amavisd-new:
cd /usr/sbin
You only need to run this script once a day. Place an entry in your crontab (on the first available blank line):
crontab -e
Replace MM with a number between 0 and 59 and HH with a number between 0 and 23:
MM HH * * * /usr/sbin/sa-update.sh
Now we reinstall clamav from Volatile (you would use this to upgrade to new versions):
apt-get -t etch install clamav clamav-daemon clamav-freshclam
Now vi /etc/amavis/conf.d/50-user and insert the text below in the middle of the file (must be between "use strict;" and "1;"). You may prefer the WinSCP editor since vim will tend to comment out this text when you paste it in. If you are still in the editor since the last edit, you may need to hit the refresh icon in order to see the /etc/amavis directory. Edit @local_domains_maps and include all your domains there. Also take a look at @mynetworks to see if you need to modify it. You will probably want to temporarily leave your network out of @mynetworks during testing (so you can send spam to test recipients): # nice to have $log_level (1-5) available: $log_level = 0; # If sender matches ACL, turn debugging fully up, just for this one message #@debug_sender_maps = ( ["test\@$mydomain"] ); # explicitly set $mydomain and $myhostname: $mydomain = 'example.com'; $myhostname = 'msa.example.com'; # when amavisd-new sends notifications, they appear to come from here: $mailfrom_notify_admin = "postmaster\@$mydomain"; $mailfrom_notify_recip = "postmaster\@$mydomain"; $mailfrom_notify_spamadmin = "postmaster\@$mydomain"; $hdrfrom_notify_sender = "amavisd-new <postmaster\@$mydomain>"; # Set number of processes. Rough guide for dual processor, 1GB = 6, 2GB = 12, 4GB = 24 # you MUST also change maxproc for the smtp-amavis transport to match this number, e.g: # smtp-amavis unix - - n - 6 smtp $max_servers = 6; # We discard (and quarantine) viruses, discard (and quarantine) spam (>= kill_level), # bounce (and quarantine) banned files and pass bad headers: $final_virus_destiny = D_DISCARD; $final_banned_destiny = D_BOUNCE; $final_spam_destiny = D_DISCARD; $final_bad_header_destiny = D_PASS; # don't quarantine bad headers (no need since we pass them all): $bad_header_quarantine_to = undef; # We use plus addressing to place spam in user's Spam folder: $recipient_delimiter = '+'; # Spam gets the Subject line prepended with: $sa_spam_subject_tag = 'Spam> '; # We tag all headers (for 'local' domains) with X-Spam info: $sa_tag_level_deflt = undef; # This is the system default spam tag level that will be overridden by user's preferences in MySQL $sa_tag2_level_deflt = 6.31; ## For relay domains not set up in MySQL you can create a static domain wide (or individual) map: #@spam_tag2_level_maps = ( # { 'postmaster@example.net' => 99.0, # '.example.net' => 8.0, # '.example.org' => 6.0 }, # \$sa_tag2_level_deflt, # catchall default #); # The default is to not quarantine any spam (outside of what users get in their Spam folder), # so set default kill_level high. Users can choose their own kill_level however. kill_level # will trigger quarantining (to MailZu). $sa_kill_level_deflt = 9999; ## Once again, relay domains may want something different: #@spam_kill_level_maps = ( # { 'postmaster@example.net' => 99.0, # '.example.net' => 8.0, # '.example.org' => 10.0 }, # \$sa_kill_level_deflt, # catchall default #); ## And some relay domains may wish to quarantine up to a certain level, then discard: #@spam_quarantine_cutoff_level_maps = ( # { '.example.net' => 20.5, # '.example.org' => 25 }, # \$sa_quarantine_cutoff_level, # catchall default (currently undef) #); # We will quarantine viruses to /var/lib/amavis/virusmails (the default). # We will use a cron job to automatically delete these files older than 14 days from the quarantine. # We can use amavisd-release or MailZu to release quarantined messages. We warn the recipients # and expect them to contact us (via the MailZu interface) if they need a banned file released. # Each domain can have their own administrators. @virus_admin_maps = ({ '.example.com' => 'postmaster@example.com', '.example.net' => 'postmaster@example.net', '.' => 'postmaster@example.com', }); @banned_admin_maps = ({ '.example.com' => 'postmaster@example.com', '.example.net' => 'postmaster@example.net', '.' => 'postmaster@example.com', }); $warnbannedrecip = 1; $defang_banned = 1; # recipient's local address(es) will be rewritten to user+spam when spam exceeds tag2_level # and as a result will be delivered to their Spam folder (thanks to maildrop) @addr_extension_spam_maps = ('spam'); # list domains in an external file (created by local_domains.sh script): @local_domains_maps = ( read_hash("$MYHOME/local_domains") ); # Since we configured SQL, we can use penpals feature: $penpals_bonus_score = 5; $penpals_threshold_low = 1; $penpals_threshold_high = 18; # We are going to create policy banks that will notify us of internally created spam # but also let banned files out (provided they are compressed). @mynetworks = qw( 127.0.0.0/8 [::1] [FE80::]/10 [FEC0::]/10 10.0.0.0/8 172.16.0.0/12 192.168.0.0/16 ); $inet_socket_port = [10024, 10026]; ## If using Mailzu, use this instead: #$inet_socket_port = [10024, 10026, 9998]; $inet_socket_bind = '127.0.0.1'; ## If using Mailzu, use this instead: #$inet_socket_bind = undef; ## Interface to MailZu #$interface_policy{'9998'} = 'MAILZU'; #$policy_bank{'MAILZU'} = { # protocol => 'AM.PDP', # inet_acl => [qw( 127.0.0.1 [::1] 111.111.111.111 )], #}; # We create a custom set of banned rules for the MYNETS and TRUSTED policy # banks. See also the 'DEFAULT' $banned_filename_re settings in 20-debian_defaults %banned_rules = ( 'BLOCK_EXE' => new_RE( # block double extensions in names: qr'\.[^./]*\.(exe|vbs|pif|scr|bat|cmd|com|cpl|dll)\.?$'i, # allow any name or type (except viruses) within an archive: [ qr'^\.(Z|gz|bz2|rpm|cpio|tar|zip|rar|arc|arj|zoo)$' => 0], # blocks MS executable file(1) types, unless allowed above: qr'^\.(exe|exe-ms)$', ), 'DEFAULT' => $banned_filename_re, ); $policy_bank{'MYNETS'} = { # mail originating from @mynetworks spam_admin_maps => ["postmaster\@$mydomain"], # alert of internal spam final_spam_destiny => D_BOUNCE, # so the sender knows they are a spammer spam_kill_level_maps => [10.0], spam_dsn_cutoff_level_maps => [9999], banned_filename_maps => ['BLOCK_EXE'], }; $interface_policy{'10026'} = 'TRUSTED'; $policy_bank{'TRUSTED'} = { # mail originating from trusted senders spam_admin_maps => ["postmaster\@$mydomain"], # alert of internal spam final_spam_destiny => D_BOUNCE, # so the sender knows they are a spammer spam_kill_level_maps => [10.0], spam_dsn_cutoff_level_maps => [9999], banned_filename_maps => ['BLOCK_EXE'], }; # Here we set up access to MySQL data: @lookup_sql_dsn = ( ['DBI:mysql:amavis:localhost', 'amavis', 'amavis_password'] ); @storage_sql_dsn = @lookup_sql_dsn; # If using MailZu, store banned files and spam to MySQL if you want to give users the # ability to read those messages in the MailZu interface: #$banned_files_quarantine_method = 'sql:'; #$spam_quarantine_method = 'sql:'; # If using MailZu and you do not wish to quarantine spam to MySQL but instead want to # quarantine to /var/lib/amavis/virusmails, MailZu cannot have spam messages # compressed (which is the default), so you would have to change from the default to this: #$spam_quarantine_method = 'local:spam-%m'; # Note: If you quarantine items locally, you would also need to create a script delete # old quarantined items. Look to /etc/cron.daily/rmvirusquar for an example # required because we set msgs.time_iso to type TIMESTAMP (required by MailZu) $timestamp_fmt_mysql = 1; # specific to the amavisnewsql SquirrelMail plugin $sql_select_white_black_list = 'SELECT wb FROM wblist'. ' WHERE (rid=?) AND (wblist.email IN (%k))'. ' ORDER BY wblist.priority DESC'; #----------------------------------------------------------Now we will create the MySQL schema for amavisd-new. This schema is a combination of the recommended amavisd-new schema, and the schema provided with the amavisnewsql SquirrelMail Plugin:
cd
You would have been prompted for roots_password. Secure 50-user from prying eyes (to protect the MySQL password):
chmod 640 /etc/amavis/conf.d/50-user
We are going to create a script that will pull a list of our domains from a PostfixAdmin table:
vi /usr/sbin/local_domains.sh
Insert the following (this is two lines - line 2 may wrap)
#!/bin/bash
Save the file. Then:
chmod 700 /usr/sbin/local_domains.sh
Our domains should be listed in the file. We will add this script to PostfixAdmin so it's updated every time we add or remove domains. We need to add another entry to /etc/sudoers so www-data can run this script:
visudo
and insert at the bottom (noting once again the hostname of the server):
www-data msa=NOPASSWD: /usr/sbin/local_domains.sh
Now patch PostfixAdmin so it uses this script:
cd /var/www/postFixadminx/admin
You should spend a minute to convince yourself this works. Remove /var/lib/amavis/local_domains and log into PostfixAdmin and edit a domain (you have to log in as the super-user in order to edit domains). Insure /var/lib/amavis/local_domains was created as expected, then:
amavisd-new reload
Install a couple maintenance scripts to prevent the amavis database from growing forever (messages over 24 days are deleted). If it grows forever, you or I made a mistake here:
cd /usr/sbin
Now configure postfix to use amavisd-new. I will show NEW changes to master.cf in red. Don't forget to match maxproc for the smtp-amavis transport to $max_servers. Also notice there are different port number involved for the content_filter overrides. You may also want to use the WinSCP editor here:
vi /etc/postfix/master.cf
# ========================================================================== # service type private unpriv chroot wakeup maxproc command + args # (yes) (yes) (yes) (never) (100) # ========================================================================== smtp inet n - - - - smtpd -o smtpd_use_tls=no -o smtpd_sasl_auth_enable=no -o content_filter=smtp-amavis:[127.0.0.1]:10024 # -o receive_override_options=no_address_mappings # If they want to relay, make them use port 587 (submission) or port 465 (smtps) # If using submission port, configure client to use CRAM-MD5 submission inet n - - - - smtpd -o smtpd_use_tls=no -o smtpd_sasl_auth_enable=yes -o smtpd_client_restrictions=permit_sasl_authenticated,reject -o content_filter=smtp-amavis:[127.0.0.1]:10026 # -o receive_override_options=no_address_mappings # Outlook and OE (and many others) expect smtpd_tls_wrappermode, # so have them submit here (PORT 465): smtps inet n - - - - smtpd -o smtpd_enforce_tls=yes -o smtpd_tls_wrappermode=yes -o smtpd_sasl_auth_enable=yes -o smtpd_client_restrictions=permit_sasl_authenticated,reject -o content_filter=smtp-amavis:[127.0.0.1]:10026 # -o receive_override_options=no_address_mappings # We will use port 4650 for clients that use STARTTLS: 4650 inet n - - - - smtpd -o smtpd_enforce_tls=yes -o smtpd_sasl_auth_enable=yes -o smtpd_client_restrictions=permit_sasl_authenticated,reject -o content_filter=smtp-amavis:[127.0.0.1]:10026 # -o receive_override_options=no_address_mappings #628 inet n - - - - qmqpd pickup fifo n - - 60 1 pickup -o content_filter= [... other stuff is here, but does not need editing ...] # Insert at the bottom and adjust maxproc from 6 if needed (remembering that this must match $max_servers): # smtp-amavis unix - - n - 6 smtp -o smtp_data_done_timeout=1200 -o smtp_send_xforward_command=yes -o disable_dns_lookups=yes 127.0.0.1:10025 inet n - n - - smtpd -o content_filter= -o local_recipient_maps= -o relay_recipient_maps= -o smtpd_restriction_classes= -o smtpd_delay_reject=no -o smtpd_client_restrictions=permit_mynetworks,reject -o smtpd_helo_restrictions= -o smtpd_sender_restrictions= -o smtpd_recipient_restrictions=permit_mynetworks,reject -o smtpd_data_restrictions=reject_unauth_pipelining -o smtpd_end_of_data_restrictions= -o mynetworks=127.0.0.0/8 -o smtpd_error_sleep_time=0 -o smtpd_soft_error_limit=1001 -o smtpd_hard_error_limit=1000 -o smtpd_client_connection_count_limit=0 -o smtpd_client_connection_rate_limit=0 -o receive_override_options=no_header_body_checks,no_unknown_recipient_checksWhen finished editing:
postconf -e "recipient_delimiter = +"
Configure pyzor (to use a mirror):
pyzor discover
Pyzor Ping should show 'OK'. If not, then it's possible your firewall is blocking udp replies from 82.94.255.100. We need amavisd-nanny which can be used as a diagnostic program and amavisd-release (for amavisd-new version 2.4.2) which can be used to release quarantined messages:
cd /usr/local/src
SpamAssassin, by default, will automatically attempt to figure out which Received: headers were inserted by mail servers in your network, and which were not. However, to be safe it's best to manually configure the trust path. The IP addresses listed in internal_networks and trusted_networks should be the IP addresses (or network addresses) of hosts on you network. If you are behind a NAT box, this would include your internal network, your public network and the loopback interface. If mail is relayed to you from a trusted 3rd party (maybe you use something like Postini to filter your mail), then those servers would be added to trusted_networks (but not internal_networks).
vi /etc/spamassassin/local.cf
Here is an example of what should be inserted:
# explicitly set our internal_networks (might be the same or similar to mynetworks)
Always lint SpamAssassin after modifying or adding files:
spamassassin --lint
Once you have correctly configured these settings:
amavisd-new stop
Test by sending a message through and observing output from mail.log. It can take between 1 second and 30 seconds (or even longer) to process a message. If you get clamd errors, /etc/init.d/clamav-daemon stop , wait a moment, /etc/init.d/clamav-daemon start , and wait a minute before continuing. I quarantine viruses locally for 14 days, then delete them. If you would like to accomplish this task, grab the script from me:
cd /etc/cron.daily
|
Placing Bayes and AWL data in MySQL will put some load on the MySQL server, but as long as
you have been reasonably generous with innodb_buffer_pool_size and innodb_log_file_size you will
greatly improve Bayes performance (do these in sections):
cd /etc/spamassassin/
Enter roots_password to complete the process.
rm gv-bayes-awl.sql.txt
It should show our nspam (number of spam) count is 1, and --lint should be clean. We should also create our .spamassassin directory and user_prefs file:
su amavis -c 'spamassassin <sample-spam.txt'
We continue by adding an AWL and bayes_seen maintenance script (watch for errors):
cd /etc
Once again make sure amavisd-new is processing messages. Send at least one test message and then read the headers of the message. It should show X-Spam headers and such. It will not show any BAYES hits yet - it is required that Bayes learns at least 200 ham messages first. I use autolearn (with an occasional manual feeding to sa-learn). I do not cover using other means of feeding spam and ham to SpamAssassin in this HOWTO - see man sa-learn . The SquirrelMail Spam Buttons plugin allows users to mark messages as 'Spam' or 'Not Spam' and feed them to sa-learn as such, but in a site wide database one has to remember that "one man's garbage is another man's treasure" and "too many cooks spoil the pot". Plus, there is the addition server load to consider - sa-learn is CPU intensive. It would be a good idea to test clamd. After temporarily disabling your desktop AV program I would send a test message through with ONLY the eicar string in the body of the message (absolutely no whitespace before or after the string - it must start and end on the very first line). You can stop amavisd-new and debug it with 'amavisd-new debug' or 'amavisd-new sa-debug' or you can raise $log_level or temporarily set @debug_sender_maps. |
Tell apache-ssl to use squirrelmail:
echo "Include /etc/squirrelmail/apache.conf" >> /etc/apache-ssl/httpd.conf
I am going to set the SquirrelMail URL to https://msa.example.com/mail/, so:
cd /etc/squirrelmail/
The 'allow from' IP address above is the IP address of your computer (as the mailserver sees it). Download some plugins:
cd /usr/share/squirrelmail/plugins/
We did some configuration of amavisnewsql. We turned off its (broken) quarantine function (and because it is broken we send spam to either a folder or MailZu). Now we put some sanity constraints on what users can enter for tag2_level and kill_level and modify some of the text that users see. I also add the ability for the users to set spam_quarantine_cutoff_level which gives them the option to discard high scoring spam. New users are added to the database when they log into SquirrelMail and go to Options->SpamAssassin Configuration.
wget http://verchick.com/mecham/public_html/spam/amavisnewsql.patch1.txt
Now start configuration (hint: enter squ[Tab]). Navigate to the items below from the main menu:
squirrelmail-configure
1. Organization preferences.
You should be able to test the setup. If not, check /etc/squirrelmail/apache.conf
for IP address access to configtest.php:
https://msa.example.com/mail/src/configtest.php
Browse to your SquirrelMail site, then log in to SquirrelMail with your own account (and the test account): https://msa.example.com/mail/ If you get an error: "Preference file, /test@example.com.pref.tmp, could not be opened." it means this user would need to go to Options and Personal Information and enter their name and email address. When you add users in PostfixAdmin, the preferences file in /var/lib/squirrelmail/data/ should automatically be created via the maildirmake.sh script. At some point you will no doubt wish to replace the SquirrelMail logo with yours. Place it in /usr/share/squirrelmail/images/ and change the size as required in the "Organization preferences" page. |
cd
If you get an error: Error 202 while performing register, aborting., you may need to run the 'razor-admin register' command more than once. Don't worry about it if /etc/razor/razor-agent.conf does not exist. Here we disable logging and then give amavisd-new a copy of the razor files:
sed -i 's/= 3/= 0/' /root/.razor/razor-agent.conf
|
apt-get install pflogsumm
Then just make sure you have a mailbox (or alias) for root. Root will get a report each morning after 06:25. Hint: cat /etc/crontab. There is one problem with the report. Any email that gets sent to amavisd-new, which includes most mail that isn't rejected at the front door, also comes back from amavisd-new. This means Postfix sees the email twice. So the report lists them twice. For those that are interested, there are a couple scripts out there that pre-process the log files to prevent reporting amavis entries: http://web.tiscali.it/postfix/prepflog.html and http://www.caspergasper.com/spam.shtml |
apt-get install bind9
The result of the last two commands should show 'named' is running without error. Now we create a new /etc/resolv.conf with the IP address of the server first and the IP address of the former primary DNS second:
cp /etc/resolv.conf /etc/resolv.conf-old
Then restart bind9 and check for errors again:
/etc/init.d/bind9 restart
So far we have set up bind9 as a local caching only name server. Here we add an additional security measure that prevents unauthorized machines from using our name server:
vi /etc/bind/named.conf.options
On the line below "directory" we want to add a line that restricts use of our name server to the network our mailserver is on. Place a [Tab] in front of the entry so it lines up with the other entries. You can add more than one network here if you like. Place a ";" (semicolon) after each network. Note that if you actually want to allow other clients to connect to our name server, as explained in the notes above you would also have to modify IP tables to allow this.
allow-query {222.222.222.222/24;};
Then restart bind9 and check for errors again:
/etc/init.d/bind9 restart
Optionally configure bind9 as a forwarding server. Bind9 as we have it configured now will first query the root servers for hints when needed. I prefer to forward queries to another DNS instead. There are advantages and disadvantages in doing this, but I prefer it. There are two main reasons not to use forwarding. The first is that ClamAV uses DNS to check for updates. It's possible that if you are relying on the forwarding DNS server to check the ClamAV DNS server, the information may not be up to date. The other reason has to do with spamhaus.org's DSNBLs. If you use someone else's (your ISP's) DNS server, all the queries from that DNS server are counted toward the total number of connections to the spamhaus servers, which may result in you loosing that free service. It is absolutely imperative that any name servers listed here are known to work from our mailserver. These will almost certainly be the primary and secondary servers you previously configured in /etc/resolv.conf or your ISP's servers (not 127.0.0.1, and not the IP address of the local machine). They should point to real name servers and not a DNS proxy like your Linksys broadband router or other gateway device (unless that proxy does not allow proper access to real name servers outside your network - which is sometimes the case). Add the 'forwarders' entry just below the 'allow-query' entry we just made:
vi /etc/bind/named.conf.options
and add:
forwarders {444.444.444.444; 555.555.555.555;};
To never query the root servers, optionally add (personally I do add this):
forward only;
Then restart bind9 and check for errors again:
/etc/init.d/bind9 restart
We added a new user to /etc/passwd so give Postfix a copy:
LINUX2
Check that we can resolve names:
dig yahoo.com
In the ANSWER SECTION you should see some A records with IP addresses and near the bottom the SERVER: should be this server. |
A few of these lines wrap (considerably) so I will assume you are going to select the
entire sections. If you need to edit anything, paste it into Notepad
(with Word Wrap turned off) first.
postmap /etc/postfix/sender_access
Send another test message to make sure we did not break something. You can use sender_access to whitelist senders that would get rejected by either reject_non_fqdn_sender or reject_unknown_sender_domain. You can use rbl_sender_exceptions and/or rbl_client_exceptions to whitelist those who would otherwise get rejected by sbl-xbl.spamhaus.org. Note: you can reject more mail by substituting zen.spamhaus.org for sbl-xbl.spanhaus.org (zen additionally includes pbl.spamhaus.org which rejects mail from dynamic IP addresses) but either way, you may have to pay for spamhaus DNSBLs depending on a few factors. See http://www.spamhaus.org/organization/dnsblusage.html. The rbl_recipient_exceptions table is for those recipients that wish to 'opt out' of RBL checks (postmaster and abuse come to mind). These next settings are not related to UCE. If (and only if) the IP address you present to the world is not the IP address of this server (you are configured to run behind a NAT firewall or a proxy server) please configure proxy_interfaces (1.2.3.4 represents the public IP address):
postconf -e "proxy_interfaces = 1.2.3.4"
At some point in the future, if you have clients on the local network that are going to use this server as their outgoing SMTP server, you may need to configure $mynetworks. Clients in mynetworks will bypass the reject_unauth_destination restriction and therefore be allowed to relay to domains other than ours. You can use mynetworks or you can use SASL auth. I have not had you configure $mynetworks until now because it is useful during testing to not be a member in $mynetworks. Let's create a little report that runs each morning that shows us how much mail is in our queues. This could serve as an alarm if the queues are stacked up:
vi /etc/cron.d/qshape-cron
and insert these two lines:
PATH=/usr/sbin
Root should get a report tomorrow at 06:50. Then we have to shut logcheck up (this is on one line):
echo "^\w{3} [ :0-9]{11} [._[:alnum:]-]+ postfix/pickup\[[0-9]+\]: [[:alnum:]]+: uid=[0-4]+ from=<postfix>" >>/etc/logcheck/ignore.d.server/postfix
While we are at it, there are a few other logcheck messages we need to suppress:
echo "^\w{3} [ :0-9]{11} [._[:alnum:]-]+ postfix/[ls]mtp\[[0-9]+\]: [[:upper:][:digit:]]+: to=<[^[:space:]]+>,( orig_to=<[^[:space:]]+>,)* relay=[^[:space:]]+,( conn_use=[[:digit:]]+,)? delay=[.0-9]+,( delays=[.0-9/]+, dsn=[0-9.]+,)? status=sent \(250 [0-9.]+ Ok, id=[-0-9]+, BOUNCE\)$" >>/etc/logcheck/ignore.d.server/postfix
By default "message_size_limit = 10240000" which is 10MB. Later we will install MailZu. When we do, we will also need to change memory_limit in /etc/php4/apache/php.ini to match this. Let's set that now:
sed -i 's/memory_limit = 8M/memory_limit = 10M/' /etc/php4/apache/php.ini
Another setting that would need to be updated if we changed the limit is in /etc/clamav/clamd.conf. The setting is ArchiveMaxFileSize. Let's change the limit to 20MB, just to show how it might be done:
postconf -e "message_size_limit = 20480000"
|
There are a lot of good reasons not to use an autoresponder. One reason is: most of the mail
we receive is spam and the senders use forged addresses. You will either be sending notices
to senders who did not send you mail, or you will fill up your deferred queue with undeliverable
mail. There are often other problems like multiple deliveries or the potential for infinite loops.
The 'cache' field in the vacation table that stores email addresses of senders we have already
sent notices to can fill up and cause errors.
Use at your own risk. We will not use vacation.pl as supplied with PostfixAdmin. I just want to make sure
you are using version 3.2, so get a copy from me. My copy also makes one small change: the envelope
sender will be the null sender <>. Even though we have the latest version,
there may be some lingering issues with vacation.pl; see http://high5.net/forum/
apt-get install libdbi-perl libdbd-mysql-perl
We place an entry in /etc/postfix/transport that will send mail to the vacation script. This is a bogus (sub)domain name. It does not need to be set up in DNS:
vi /etc/postfix/transport
and insert:
autoreply.example.com vacation:
vi /etc/postfix/master.cf
and insert (just below the maildrop transport might be a good place): vacation unix - n n - - pipe flags=Rq user=vacation argv=/var/spool/vacation/vacation.pl -f ${sender} -- ${recipient}Then:
postmap /etc/postfix/transport
Now we configure PostfixAdmin to use vacation:
sed -i "s/vacation'] = 'NO/vacation'] = 'YES/" /var/www/postFixadminx/config.inc.php
To change their Auto Response settings, a user could log into: https://msa.example.com/postFixadminx/users However, we can avoid users logging into PostfixAdmin by setting up a PostfixAdmin plugin in SquirrelMail (jump ahead to the next section to set it up - then return here) https://msa.example.com/mail. Test by logging into your own account and setting yourself up. Remember that the database records the senders it has already seen (so only one notice is sent per sender). To clear out the cache and start over, return from your vacation by choosing Auto Response->Coming Back and then Auto Response->Going Away again. Use phpMyAdmin to browse the database as you make changes. Now that vacation is working, consider turning it off so no one can use it. First use https://msa.example.com/phpmyadmiNx to make sure there are no entries in the vacation table and no aliases that send mail to @autoreply.example.com, then:
sed -i "s/vacation'] = 'YES/vacation'] = 'NO/" /var/www/postFixadminx/config.inc.php
|
There are two versions of this plugin out there. The one I use may be more secure:
pear install MDB2-2.4.1
Then install the plugin (the order in which the plugins are listed may be different):
8. Plugins
Now users will not need to log into PostfixAdmin to change their password, forwarding or Out of Office. Now they have similar functions in SquirrelMail (on the Options page). https://msa.example.com/mail If you turn off vacation in PostfixAdmin, then you also need to turn off vacation in the SquirrelMail postfixadmin plugin (first use https://msa.example.com/phpmyadmiNx to make sure there are no entries in the vacation table and no aliases that send mail to @autoreply.example.com):
sed -i "s/vacation'] = 'YES/vacation'] = 'NO/" /var/www/postFixadminx/config.inc.php
To turn Vacation back on:
sed -i "s/vacation'] = 'NO/vacation'] = 'YES/" /var/www/postFixadminx/config.inc.php
|
This little section represents two days of experimentation and research just to get MailZu
running. All mail left in the MailZu quarantine longer than 24 days is deleted.
apt-get install php-mail-mime php4-imap
Continue installing libc-client without Maildir support? [YES]
apache-modconf apache-ssl enable mod_imap
We are going authenticate users using a couple imap functions (but we will actually use the imap functions to authenticate via POP3), so we loaded in the mod_imap module. With the encrypted passwords we have, POP3 was the only interface I could get to work. The IMAP interface would have worked were it not for the AUTH=CRAM-MD5 setting we placed in /etc/courier/imapd. It appears to confuse the imap_open function used to connect to the server. Use of POP3 is undocumented and requires a slight change in one line of code in IMAPAuth.class.php:
cd /var/www
Note that the configuration changes below will only work if you first apply my mzcpatch.txt patch.
The only thing the patch does is change the example domain names (because they conflict with this howto).
sed -i "s/'user'/'amavis'/" config.php
cd /etc/logrotate.d
In 50-user we need to change a few things.
vi /etc/amavis/conf.d/50-user
Change this:
$inet_socket_port = [10024, 10026];
To this (add 9998 to $inet_socket_port):
#$inet_socket_port = [10024, 10026];
Change:
$inet_socket_bind = '127.0.0.1';
To (listen on all interfaces):
#$inet_socket_bind = '127.0.0.1';
Then enable the policy bank; change: ## Interface to MailZu
Make sure this next entry contains our local IP address (where 111.111.111.111 is the IP address of this machine):
## Interface to MailZu
Enable quarantine to SQL; uncomment these two lines as shown:
# If using MailZu, store banned files and spam to MySQL if you want to give users the
Once finished editing, reload amavisd-new:
amavisd-new reload
Quiet down logcheck:
echo "amavis\[[0-9]+\]: \(rel-.{12}\) Quarantined message release:" >>/etc/logcheck/ignore.d.server/amavisd-new
This should be enough configuration for you to log in (as a regular user). https://msa.example.com/mailzu. Now you can start experimenting. Log into SquirrelMail and change the recipient's SpamAssassin settings to "All spam to quarantine". This particular policy will quarantine at a score of 8 or higher and there is no spam_quarantine_cutoff_level in force. Send a message containing the gtube string (which will insure the message scores over 12) and then wait for it to show up in the quarantine. Release the message and see if it is delivered. Double check your settings if you get an error. Review the "Amavisd-new Configuration" notes at vi /var/www/mailzu/docs/INSTALL . Make sure the hostname of the server resolves properly. If necessary, read the mailing list archives. Now you will want to vi config.php and familiarize yourself with all the possible settings - including verbose logging. You may wish to make changes. One thing you will need to do is to create an administrator or two. See this section: // Super Admins can do anything mail admins can plus // change settings $conf['auth']['s_admins'] = array ('user1', 'user2'); // Mail Admins can see other users' spam and attachments // and can perform any action on them $conf['auth']['m_admins'] = array ();You will want to spend some time sending banned files and spam to your test user and having the test user release some spam and request banned files get released. You will want to spend some time as the administrator, responding to this user's requests. As Jerry Seinfeld says: "Well.. good luck with all that!". |
When we create a new mailbox, the maildirmake.sh script enters the quota information in
a 'maildirsize' file located in the user's maildir.
When we edit a user's mailbox, the quotachange.sh script updates the quota information.
Each time a message comes or goes from a user's INBOX the 'maildirsize'
file is updated. Because we are using maildrop as a LDA (Local Delivery Agent),
Postfix has no way of knowing how many bytes of data the recipient's mailbox contains,
but maildrop can figure it out.
In the maildrop transport in master.cf is a flag: "-w 90". This tells maildrop to send a warning
message to the recipient when the size of their mailbox has reached 90% of their quota. You have to
create this message. Here is a sample we will install:
vi /etc/quotawarnmsg
and insert (and edit as needed): From: Mail Delivery System <Mailer-Daemon@example.com> Reply-To: postmaster@example.com To: "Valued Customer"; Subject: Mail quota warning Mime-Version: 1.0 Content-Type: text/plain; charset=iso-8859-1 Content-Transfer-Encoding: 7bit Your mailbox on the example.com server is now more than 90% full. To continue to receive mail you must remove some messages from your mailbox. If you should reach 100% of your quota, mail delivery to you will stop. If using an email client such as Microsoft Outlook Express or Mozilla Thunderbird (for example) to access your mail, moving messages to a local folder may reduce the number of messages on our server. Your quota is displayed in the web interface at https://msa.example.com/mail If you feel your quota is too restrictive, please contact postmaster@example.com Thank YouTo test this you would first empty out all mail from the test user. https://msa.example.com/mail would be handy for this task (and to test). Then log in to https://msa.example.com/postFixadminx and edit that user's mailbox, setting the quota to 1MB. Verify maildirsize was updated:
cat /var/vmail/example.com/test/maildirsize
My system shows:
1024000S
Find a file that will not get rejected by amavisd-new (.txt .doc .xls .html .pdf) with a size around 70KB that can be used as an attachment. We want to send a series of messages to the test user that contains this attachment. We want to trigger the quota warning message by sending a series of messages that will total over 900KB (use the Check Mail link in the SquirrelMail interface to refresh the quota usage data). Once you are over 90%, if you log out and log back in to SquirrelMail you will see a warning message. One thing the message suggests is that you empty your trash. That's nice, but the truth is trash does not count toward the quota. The recipient should also get our message with the Subject: "Mail quota warning". The recipient is only warned once every 24 hours (controlled by existence of 'quotawarn' file) so if you want to test again, you will have to remove the quotawarn file from the recipient's maildir:
find /var/vmail -name quotawarn
Once you are satisfied with your message, send another message that puts this user over the top of their quota. If you look in mail.log you will see that the message has not been sent to the user but instead has been placed in the deferred queue (which can also be confirmed with the mailq command):
Jul 1 10:20:11 msa postfix/pipe[7387]: 57FAA24156: to=<test@example.com>, relay=maildrop, delay=0.15,
delays=0.08/0.02/0/0.05, dsn=4.3.0, status=deferred (temporary failure. Command output:
maildrop: maildir over quota. )
So now this mail will be stuck in the deferred queue until the user either deletes some messages to make room for it, or you increase their quota. Logcheck will notify you of the delivery failure. If this user is unresponsive to requests to clean up their mailbox, all new mail for this user will get deferred (and eventually returned to the sender after 5 days - Postfix $maximal_queue_lifetime). If this is the case you may want to manually intervene and start rejecting mail to this user instead of holding on to it for them (don't be tempted to delete mail from the deferred queue when this machine goes into production - it's not your mail to delete). To do so, you could add this recipient to /etc/postfix/reject_over_quota with something like this (this is all on one line):
test@example.com REJECT The recipient's mailbox is full - your message was not delivered to recipients hosted on the example.com mail server
Then of course:
postmap /etc/postfix/reject_over_quota
The downside to this is that the entire message will get rejected. If the message is addressed to multiple recipients on your server, none will receive the message. This can't be fair to the sender, or the other recipients, so I would consider it only in extreme cases. In general, when mail is delayed, the sender is not notofied of the delay. If you would like any sender whose mail is delayed (deferred) for any reason, you can have Postfix notify them. He we give Postfix one hour to attempt delivery before sending a warning to the sender:
postconf -e "delay_warning_time = 1h"
|
Mailgraph produces daily, weekly, monthly and yearly graphs of received/sent and bounced/rejected mail.
apt-get install rrdtool mailgraph
It may take a minute for this to work, but eventually if you browse to https://msa.example.com/cgi-bin/mailgraph.cgi you should get some graphs. |
This nice little program can perform backups of our MySQL data. We will make local backups
(and keep 3 days worth). We will talk about backing up the entire system later.
apt-get install libxml-parser-perl
This places entries in root's crontab. Run crontab -e and see if you have any extra newline characters separating any entries (and edit them out). The details of what just happened can be found here: http://www.howtoforge.com/mysql_zrm_debian_sarge. I suggest you save all five pages of that HOWTO to your computer (and maybe read them too). This program will send you a couple (annoying) email messages each day. |
At this point, we have a little problem. The problem is with aliases.
Alias expansion occurs twice, once when a message
comes in the front door via one of the smtpd daemons, and once again when amavis
returns the message to the Postfix reinjection port (listening on port 10025).
So, address rewriting occurs both before amavisd-new, and after amavisd-new.
Why does this matter? For each user added via PostfixAdmin, there is an entry created in the
postfix.alias table in the form:
test@example.com test@example.com
If a user adds a forwarder (via either the SquirrelMail or PostfixAdmin interface - and if the PostfixAdmin interface is used, they choose "Deliver to the local mailbox."), the user's alias record may now look something like this:
test@example.com test@example.com,garyv@example.net
In this example, test@example.com is a real user on our system, and the garyv@example.net mailbox is hosted by someone else. So, the message will now be addressed to both test@example.com and garyv@example.net. After amavisd-new processes the message, two separate messages emerge, one to test@example.com, and one to garyv@example.net. It's the one addressed to test@example.com that is the problem. Since the reinjection port at 127.0.0.1:10025 is just another smtpd daemon (which uses the cleanup daemon to facilitate rewriting the address), Postfix addresses this message to both test@example.com and garyv@example.net (again). The result is, garyv@example.net gets two copies of the same message (test@example.com only gets one copy because the message was already addressed to test@example.com). When I first wrote this document, I disabled address rewriting before amavisd-new (I used "-o receive_override_options=no_address_mappings" on the Internet facing smtpd daemons). This could work if all mail came in addressed to "hard" users (users with actual mailboxes), but if (as the administrator) you want to create an alias something like this:
test.user@example.com test@example.com
where test@example.com has a mailbox (and is in the amavis database) and test.user@example.com is created by the admin merely as a convenient way to redirect mail to the actual account, this does not work well if address rewriting only occurs after amavisd-new. The reason is, test.user@example.com would end up using the default settings in amavisd.conf or 50-user rather than the settings test@example.com desired (the settings stored in the MySQL database). So, what can we do? One option would be to edit the 127.0.0.1:10025 smtpd daemon and change this override:
-o receive_override_options=no_header_body_checks,no_unknown_recipient_checks
to:
-o receive_override_options=no_address_mappings,no_header_body_checks,no_unknown_recipient_checks
but the problem is, as shown in the next "Acting as a relay server" section, there are situations where we don't actually want to completely disable address rewriting after amavisd-new, we may simply want to use a different virtual alias table (or no virtual alias table). To that end, we will add a new cleanup daemon that amavisd-new can use, then, for the moment we tell this cleanup daemon not to use a virtual alias table. vi /etc/postfix/master.cf and copy the current cleanup daemon, name the copy amavis-cleanup and add an override that disables the current virtual alias maps: cleanup unix n - - - 0 cleanup amavis-cleanup unix n - - - 0 cleanup -o virtual_alias_maps= qmgr fifo n - n 300 1 qmgrThen, on the reinjection port, tell Postfix to use this new cleanup daemon for mail that amavisd-new sends to it: 127.0.0.1:10025 inet n - n - - smtpd -o cleanup_service_name=amavis-cleanup -o content_filter= -o local_recipient_maps= <...>Save and exit the file, then:
/etc/init.d/postfix stop
You can get more creative with this (if you need to). I suggest reading: http://www.ijs.si/software/amavisd/README.postfix.html and http://www.ijs.si/software/amavisd/README.postfix.old Note, if the user uses the PostfixAdmin interface to change the Forwarding alias(es) and they choose the "Forward to given email addresses only." option (as opposed to the "Deliver to the local mailbox." option), the postfix.alias record would look more like this:
test@example.com garyv@example.net
In this example, garyv@example.net is still a remote mailbox. Keep in mind that remote mailboxes are not going to have their email address in the amavis MySQL database (unless you manually add them), so this type of recipient will end up using the default (static) settings in the 50-user or amavisd.conf files. |
You may have domains that are not going to store their mail on this server but instead
want this box to clean up their mail for them and then relay it on to them. You use
relay_domains and transport_maps to configure those domains you wish to relay to.
The biggest
issue you have to deal with in this scenario is how to reject mail to invalid users. The downstream
servers may know who their valid users are, but this server does not. Nonetheless, you must obtain this
information one way or another so you can reject mail to invalid users. But first we need to
get to the point where mail is routed to them. Note that the current mysql_virtual_domains_maps.cf
is replaced with one that is compatible with mysql_relay_domains_maps.cf:
cd /etc/postfix
We are not going to store transport map data in MySQL (if later you want to, I leave it up to you to figure out how to do it). Each domain you relay for will need an entry in the transport map. The left hand side is the domain and the right hand side is the host name or IP address of the next hop server (with square brackets used to turn off MX lookups). This is the general format of the contents of /etc/postfix/transport (you can vi /etc/postfix/transport and enter valid data for your situation):
example.net relay:[777.777.777.777]
Use https://msa.example.com/postFixadminx/admin/ to add the relay domain(s). When you add the domain(s), check the "Mail server is backup MX:". The domain is now a relay domain. You cannot add mailboxes to relay domains. You cannot create aliases for relay domains. Once the domains have been added, you can reload postfix, and because we have added a domain or domains to /var/lib/amavis/local_domains we should also reload amavisd-new:
postmap /etc/postfix/transport
So now we have a problem. Let's say we have added two domains, one (example.net) relays mail to a server running sendmail and the other (example.org) relays mail to a server running Microsoft Exchange 2000. Our server will accept mail addressed to any recipient in either domain. This particular sendmail server is configured to immediately reject mail to invalid users. When it rejects a message, Postfix will create a bounce notice and attempt to deliver it to the sender. If the sender is completely bogus, the message will sit in our deferred queue for days while delivery attempts are made. If the sender is faked but points to a real address, then we are spamming an innocent victim. This victim is getting "joe jobbed" - and we are facilitating it - and now we are a source of backscatter. If we send a message to the domain that forwards to the sendmail server, we can see from the bounce notice that the downstream server (at the hypothetical address of 777.777.777.777) rejected it: <testgarbage@example.net>: host msa.example.com[111.111.111.111] said: 550 <testgarbage@example.net>: Recipient address rejected: undeliverable address: host 777.777.777.777[777.777.777.777] said: 550 5.1.1 <testgarbage@example.net>... User unknown (in reply to RCPT TO command) (in reply to RCPT TO command)In this case (the case being the downstream server immediately rejects mail to invalid users) we can use either address verification or relay_recipient_maps. Basically, with address verification (reject_unverified_recipient), Postfix first checks the downstream server to see if it will accept a message to the recipient or not, prior to accepting the message. If the downstream server rejects the message (due to invalid address), so will Postfix (before it accepts the message). On the other hand, if you use relay_recipient_maps, relay_recipient_maps requires that all known good recipient addresses (for the domains listed in relay_domains) are in a lookup table. Mail addressed to a recipient whose domain is listed in relay_domains that is not also listed in the table defined in relay_recipient_maps is rejected. If we relay a message to the Exchange server, the particular Exchange server in our example accepts the message and later generates a bounce notice which it mails to the sender. We can only use relay_recipient_maps in this case where the downstream server does not immediately reject messages addressed to invalid users. Let's continue by first setting up address verification for example.net (the domain using the sendmail server). I will illustrate this mixed setup in which both reject_unverified_recipient and relay_recipient_maps will be utilized. Keep in mind that for your setup, you will have to discover which relay domains of yours will immediately reject mail to invalid users and which will not (if you want to use reject_unverified_recipient that is). Keep in mind that use of reject_unverified_recipient is a convenience over use of relay_recipient_maps, but the ideal situation would be exclusive use of relay_recipient_maps for all your relay domains. Also keep in mind that you could set up relay_recipient_maps in an SQL table and create a query for it in the same fashion we have done for other Postfix SQL tables, but regardless of the type of table used to store the data, the question is - who will maintain the adding and deleting of email addresses?
vi /etc/postfix/verify_domains
and insert the domain(s) that use server(s) that will immediately reject mail to unknown users:
example.net reject_unverified_recipient
then postmap it:
postmap /etc/postfix/verify_domains
vi /etc/postfix/main.cf , make smtpd_recipient_restrictions pretty like this and add the red item in the position shown. Read the beginning (half dozen lines or so) of http://www.postfix.org/postconf.5.html: smtpd_recipient_restrictions = permit_mynetworks, permit_sasl_authenticated, reject_unauth_destination, reject_unlisted_recipient, check_recipient_access hash:/etc/postfix/verify_domains, check_recipient_access hash:/etc/postfix/reject_over_quota, check_sender_access hash:/etc/postfix/rbl_sender_exceptions, check_client_access hash:/etc/postfix/rbl_client_exceptions, check_recipient_access hash:/etc/postfix/rbl_recipient_exceptions, reject_rbl_client sbl-xbl.spamhaus.orgAlso add these items to main.cf:
address_verify_map = btree:/var/lib/postfix/verify_cache
postfix reload and test the setup (by sending mail to invalid users in the relay domain and reading the log). When finished testing, uncomment the unverified_recipient_reject_code line. With the other domain (example.org on the Exchange server) we need to set up relay_recipient_maps. We will create /etc/postfix/relay_recipients. Once relay_recipient_maps is configured in main.cf, any recipient or recipient domain of a relay domain not listed in /etc/postfix/relay_recipients will get rejected. This means /etc/postfix/relay_recipients must contain every single recipient of every single relay domain this machine accepts mail for. However, domain wildcards can be used as placeholders. In our example case, example.net is already using address verification to reject mail to invalid users, so we can use a wildcard for that domain: @example.net. For example.org we can start out with a wildcard while we are in the process of gathering valid addresses:
vi /etc/postfix/relay_recipients
@example.net 1
but once we have all the address gathered, we need to get rid of the wildcard:
@example.net 1
Then of course:
postmap /etc/postfix/relay_recipients
The "1" after each entry could be something else, like "OK" or something. It's not important what the actual text is because it's not used for anything, but it must exist. To set up relay_recipient_maps:
postconf -e "relay_recipient_maps = hash:/etc/postfix/relay_recipients"
That's it for recipient validation for relay domains. I'm going to change the subject a little. Since there are no mailboxes for relay domains, we cannot log into SquirrelMail to change amavisd-new settings (actually, in a moment I explain how you can work around this). At the moment all relay domain recipients will use the static settings in amavisd.conf and/or the other configuration files in /etc/amavis/conf.d and elsewhere (settings like $sa_kill_level_deflt). You can configure per-user or per-domain static settings there, or you can manually add the policies and users to MySQL. To get additional information on the amavis SQL setup, I suggest you jump ahead to the top of page 2 and read the "Amavisd-new lookups" section, then return here. Each user in the amavis database (a user can also be an entire domain) must be linked to a policy. The policy can define a number of settings, such as spam_tag2_level and spam_kill_level for example. Each policy or user in the amavis database must have a unique id, so each time a policy or user is updated in the SquirrelMail interface (which can happen at any moment) the policy_id_seq and the users_id_seq are updated so the application can keep track of which number it used for the last policy or user. If you have just one or two relay users you need to add, you can cheat and put our test user to good use. In phpMyAdmin edit the amavis 'users' table for the test user, replacing all the fields that need to be replaced - like email, fullname and username (but leave the id alone). The next time the test user logs in and goes to the SpamAssassin Configuration page a new record will be created. You can edit this one also and repeat the process, logging the test user in and out of SquirrelMail each time you need to create a new record. If you need to enter a larger number of users or policies it would be easier to increment the users_id_seq (for example) to a number large enough to accommodate the number of records you are going to add - before you add them. Then make sure the id numbers in the records are both unique and within the allocated range of numbers. Here (as an example) we change users_id_seq from 15 to 19 in order to accommodate 3 new users (we left 16 unused in case someone uses it before we can change 15 to 19). Once users_id_seq is set to 19, the SquirrelMail plugin will use 20 as the next number, so 17 through 19 is ours: USE amavis; SELECT id FROM users_id_seq; +----+ | 15 | +----+ UPDATE `users_id_seq` SET `id` = '19'; INSERT INTO users (id, priority, policy_id, email, fullname, digest, username, retention) VALUES (17,7,6,'user1@example.net','User 1','WD','user1@example.net',14); INSERT INTO users (id, priority, policy_id, email, fullname, digest, username, retention) VALUES (18,7,6,'user2@example.net','User 2','WD','user2@example.net',14); INSERT INTO users (id, priority, policy_id, email, fullname, digest, username, retention) VALUES (19,7,6,'user3@example.net','User 3','WD','user3@example.net',14);The same concept would apply to policies. Now you have another problem to deal with. The next-hop server may not accept mail addressed to user+spam@example.net. If it does not, then you may have to create aliases. This can be done either on this server, or at the nexthop server. To do it on this server, I would manually create these aliases and place them in a special virtual alias table created just for the purpose. This table is used after amavisd-new has processed a message. For example, let's create /etc/postfix/amavis_virtual.
touch /etc/postfix/amavis_virtual
Then vi /etc/postfix/master.cf and modify the amavis-cleanup service (created in the previous "The alias issue" section) to use this table: cleanup unix n - - - 0 cleanup amavis-cleanup unix n - - - 0 cleanup -o virtual_alias_maps=hash:/etc/postfix/amavis_virtual qmgr fifo n - n 300 1 qmgrThe format of the data in this table would be something like:
garyv+spam@example.net garyv@example.net
A one to one mapping like this is ideal - especially because some recipients in the same domain may want to use 'plus addressing' and others may not. I will tell you that is is possible to use a single entry in a regexp map to rewrite all the addresses to a particular domain, but this breaks recipient validation. In other words, mail addressed to anyone+spam@example.net may not be rejected immediately. This requires a regexp (or pcre) map, so we would have to add this map type to our list of virtual alias maps (the override is all on one line):
touch /etc/postfix/regexp_amavis_virtual
And once again, add it to master.cf: cleanup unix n - - - 0 cleanup amavis-cleanup unix n - - - 0 cleanup -o virtual_alias_maps=hash:/etc/postfix/amavis_virtual,regexp:/etc/postfix/regexp_amavis_virtual qmgr fifo n - n 300 1 qmgrThen you could vi /etc/postfix/regexp_amavis_virtual and insert something to this effect:
/^(.*)\+spam@example\.net$/ $1@example.net
Then:
/etc/init.d/postfix stop
Don't forget to uncomment #unverified_recipient_reject_code = 550 once address verification is known to work as expected (assuming you are using reject_unverified_recipient via the verify_domains map). Then test, test, test (make sure you send some spam). You may need to configure @spam_kill_level_maps and/or @spam_tag2_level_maps in /etc/amavis/conf.g/50-user for the relay domains. While testing, you may need to temporarily remove your network from @mynetworks in /etc/amavis/conf.g/50-user so the MYNETS policy bank is not loaded for your client. If you are using Exchange, there are some HOWTOs out there that may help with pulling data out of Active Directory: http://verchick.com/mecham/public_html/spam/relay_recipients.html http://www-personal.umich.edu/~malth/gaptuning/postfix/ http://www.unixwiz.net/techtips/postfix-exchange-users.html http://postfix.state-of-mind.de/patrick.koetter/mailrelay/ http://verchick.com/mecham/public_html/spam/PostfixAddressExtract.vbs.txt Now that you get a picture of some of the issues related to relay_domains take a look at this post: http://www.freespamfilter.org/forum/viewtopic.php?t=1159 |