You will want some way for the computers on your network to refer to one another by name, and also a way for people in the outside world to refer to your exposed hosts by name. There are several ways to go about doing this.
[ Note: if you have chosen not to implement a private network, go to section Fully Exposed Network, Hosted By ISP. ]
In this configuration, you have delegated responsibility for the primary DNS authority on your domain to the ISP. You still use DNS within your private network when hosts there want to talk to one another. You have given your ISP a list of the names and IP numbers of all exposed hosts. If you want one externally visible machine, for instance betty.example.com, to act both as web and FTP server, you should ask the ISP to make CNAME entries for www.example.com and ftp.example.com pointing to betty.example.com.
Set up DNS on your private network gateway machine. This can be done securely, and makes upgrading easier, should you later decide to host primary DNS authority for your domain.
I will assume that you have decided to host DNS from the machine dns.example.com, which is on the private network gateway, and an alias for fred.example.com at 192.168.2.1. Some small modifications have to be made to this configuration if this is not the case. I will not cover that in this HOWTO unless there is significant interest.
You will have to download and compile a recent
version of BIND, the Berkeley Internet Name Domain. It is available at
the
BIND web site. Next, you have to configure the daemon.
Create the following file, /etc/named.conf
:
options {
directory "/var/named";
listen-on { 192.168.1.1 };
};
zone "." {
type hint;
file "root.hints";
};
zone "0.0.127.in-addr.arpa" {
type master;
file "pz/127.0.0";
};
zone "1.168.192.in-addr.arpa" {
type master;
file "pz/1.168.192";
};
zone "example.com" {
type master;
notify no;
file "pz/example.com";
};
Note that we are declaring ourselves the master for the example.com domain. Meanwhile, our ISP is also declaring itself to be the master for the same domain. This is not a problem, as long as you are careful about the setup. All of the machines on the private network must use dns.example.com to perform their name resolution. They must not use the name resolvers of the ISP, as the ISP name server believes itself to be authoritative over your entire domain, but it doesn't know the IP numbers or names of any machines on your private network. Similarly, hosts on exposed IP numbers in your domain must use the ISP name server, not the private name server on dns.example.com.
The various files under /var/named
must now be
created.
The root.hints
file is exactly as described in the
BIND documentation, or in the
DNS HOWTO. At the time of this writing, the following is a valid
root.hints file:
H.ROOT-SERVERS.NET. 6d15h26m24s IN A 128.63.2.53
C.ROOT-SERVERS.NET. 6d15h26m24s IN A 192.33.4.12
G.ROOT-SERVERS.NET. 6d15h26m24s IN A 192.112.36.4
F.ROOT-SERVERS.NET. 6d15h26m24s IN A 192.5.5.241
B.ROOT-SERVERS.NET. 6d15h26m24s IN A 128.9.0.107
J.ROOT-SERVERS.NET. 6d15h26m24s IN A 198.41.0.10
K.ROOT-SERVERS.NET. 6d15h26m24s IN A 193.0.14.129
L.ROOT-SERVERS.NET. 6d15h26m24s IN A 198.32.64.12
M.ROOT-SERVERS.NET. 6d15h26m24s IN A 202.12.27.33
I.ROOT-SERVERS.NET. 6d15h26m24s IN A 192.36.148.17
E.ROOT-SERVERS.NET. 6d15h26m24s IN A 192.203.230.10
D.ROOT-SERVERS.NET. 6d15h26m24s IN A 128.8.10.90
A.ROOT-SERVERS.NET. 6d15h26m24s IN A 198.41.0.4
The pz/127.0.0
file is as follows:
$TTL 86400
@ IN SOA example.com. root.example.com. (
1 ; Serial
8H ; Refresh
2H ; Retry
1W ; Expire
1D) ; Minimum TTL
NS dns.example.com.
1 PTR localhost.
The pz/1.168.192
file is as follows:
$TTL 86400
@ IN SOA dns.example.com. root.dns.example.com. (
1 ; Serial
8H ; Refresh 8 hours
2H ; Retry 2 hours
1W ; Expire 1 week
1D ; Minimum 1 day
)
NS dns.example.com.
1 PTR fred.example.com.
PTR dns.example.com.
PTR mail.example.com.
2 PTR barney.example.com.
3 PTR wilma.example.com.
and so on, where you create one PTR record for each machine with an
interface on the private network. In this example, fred.example.com is
on IP number 192.168.1.1, and is pointed to by the dns.example.com and
mail.example.com aliases. The machine barney.example.com is on IP
number 192.168.1.2, and so on.
The pz/example.com
file is as follows:
$TTL 86400
@ IN SOA example.com. root.dns.example.com. (
1 ; Serial
8H ; Refresh 8 hours
2H ; Retry 2 hours
1W ; Expire 1 week
1D ; Minimum 1 day
)
NS dns.example.com.
IN A 192.168.1.1
IN MX 10 mail.example.com.
IN MX 20 <ISP mail machine IP>.
localhost A 127.0.0.1
fred A 192.168.1.1
A 10.1.1.9
dns CNAME fred
mail CNAME fred
barney A 192.168.1.2
wilma A 192.168.1.3
betty A 10.1.1.10
www CNAME betty
ftp CNAME betty
Note that we create entries for machines both within the private
network and on external IPs, since machines within the private network
will not query the ISP's name servers for a request on, say,
betty.example.com. We also provide both IP numbers for fred, the
private and external IP numbers.
One line in the ``options'' section of
/etc/named.conf
bears discussion:
listen-on { 192.168.1.2 };
This will prevent your named daemon from answering DNS requests on the
outside interface (all requests from the outside must go through the
ISP's name resolver, not yours).
[ Note: if you have chosen not to implement a private network, go to section Fully Exposed Network, Hosted By ISP. ]
In this configuration, you have decided that your private network is
fairly small and unlikely to change often. You have decided not to use
the centralized database of a DNS server, and instead to maintain the
host resolution separately on each machine. All machines should use
the ISP's DNS server for their host name resolution for machines
beyond the private network gateway. For name resolution on the private
network, a hosts table has to be created. For Linux, this means
entering the names and IP numbers of all of the machines on the
private network into the /etc/hosts
on each
machine. Any time a new machine is added, or a name or IP number is
changed, this file has to be updated on each Linux box.
As in section DNS Resolution on Private Network, ISP Handles Domain, the list of host names on exposed IP numbers must be sent to the ISP, and any aliases (such as for www and ftp names) should be specified so that a CNAME entry can be created by the ISP.
While you could set up named resolution on the exposed hosts, and private database resolution for the private network, I will not cover that case. If you're going to be running named for one service, you ought really to do it for both, just to simplify the configuration. In this section I will assume that the private network gateway machine is handling name resolution both for the private network and for outside requests.
At the time of this writing, under version 8.2.2 of the BIND package, there is no way for a single named daemon to produce different answers to requests, depending on which interface the request arrives on. We want name resolution to act differently if the query comes from the outside world, because IP numbers on the private network shouldn't be sent out, but have to be available in answer to requests from within the private network. There is some discussion of a new ``views'' keyword which may be added to BIND to fill this need at a later date, but until that happens, the solution is to run two named daemons with different configurations.
First, set up the private network domain name server as described in section DNS Resolution on Private Network, ISP Handles Domain. This will be the name resolver visible from within your private network.
Next, you have to set up DNS for your domain, as visible to hosts in the outside world. First, check with your provider to see if they will delegate reverse lookups of your IP numbers to them. While the original DNS standard didn't account for the possibility of controlling reverse DNS on subnets smaller than a class C network, a workaround has been developed which works with all compliant DNS clients, and has been outlined in RFC 2317. If your provider is willing to delegate control of reverse DNS on your IP block, you will have to determine from them the exact name of the in-addr pseudo-domain they have chosen to delegate to (the RFC does not offer a convention they recommend for everyday use), and you will have to register control for that pseudo-domain. I will assume that the provider has delegated control to you, and the name of the pseudo-domain is 8.1.1.10.in-addr.arpa. The provider would create CNAME entries of the form
8.1.1.10.in-addr.arpa. 2H IN CNAME 8.8.1.1.10.in-addr.arpa.
9.1.1.10.in-addr.arpa. 2H IN CNAME 9.8.1.1.10.in-addr.arpa.
10.1.1.10.in-addr.arpa. 2H IN CNAME 10.8.1.1.10.in-addr.arpa.
etc.
in their zone file for the 1.1.10.in-addr.arpa domain. The
configuration of your 8.1.1.10.in-addr.arpa zone file is given later
in this section.
If your provider is willing to delegate control of the reverse DNS to you, they will create CNAME entries in their reverse DNS zone table for those IP numbers you control, pointing to the corresponding records in your pseudo-domain, as shown above. If they are not willing to delegate control to you, you will have to ask them to update their reverse DNS entries any time you add, delete, or change the name of an externally visible host in your domain. If the reverse DNS table is not synchronized with your forward DNS entries, certain services may generate warnings, or refuse to handle requests issued by machines affected by the mismatch.
You now have to create a second named setup, this one to handle requests issued by machines outside the private network gateway. This setup lists only those hosts and IP numbers which are externally visible, and responds only to requests on the outside interface of the private network gateway machine.
First, create a second configuration file, for instance
/etc/named.ext.conf
for requests from the external
interface. In our example, it might be as follows:
options {
directory "/var/named";
listen-on { 10.1.1.9; };
};
zone "." {
type hint;
file "root.hints";
};
zone "0.0.127.in-addr.arpa" {
type master;
file "pz/127.0.0";
};
zone "8.1.1.10.in-addr.arpa" {
type master;
file "ext/8.1.1.10";
};
zone "example.com" {
type master;
notify no;
file "ext/example.com";
};
The root.hints
and pz/127.0.0
files, both under /var/named
are shared with the
other running daemon. The file ext/8.1.1.10
is as
follows:
$TTL 86400
@ IN SOA fred.example.com. root.fred.example.com. (
1 ; Serial
10800 ; Refresh 3 hours
3600 ; Retry 1 hour
3600000 ; Expire 1000 hours
86400 ) ; Minimum 24 hours
NS dns.example.com.
9 IN PTR fred.example.com.
PTR dns.example.com.
PTR mail.example.com.
10 IN PTR betty.example.com.
PTR www.example.com.
PTR ftp.example.com.
The file ext/example.com
contains the following:
$TTL 86400
@ IN SOA example.com. root.fred.example.com. (
10021 ; Serial
8H ; Refresh 8 hours
2H ; Retry 2 hours
1W ; Expire 1 week
1D ; Minimum 1 day
)
NS fred.example.com.
IN A 209.217.100.58
IN MX 10 mail.example.com.
IN MX 20 <ISP Mail Machine>.
localhost A 127.0.0.1
fred A 10.1.1.9
betty A 10.1.1.10
dns CNAME fred
mail CNAME fred
www CNAME betty
ftp CNAME betty
Start the two daemons on the private network gateway machine. Put the following into your network daemon initialization scripts:
/usr/sbin/named -u dnsuser -g dnsgroup /etc/named.conf
/usr/sbin/named -u dnsuser -g dnsgroup /etc/named.ext.conf
I've assumed here that you have created the unprivileged user
``dnsuser, and the corresponding unprivileged group ``dnsgroup''. If a
bug in bind turns up, which allows an attacker to execute code from
within named, the attacker will find himself restricted to
those operations available to the unprivileged user. The
/var/named
directory and the files within should not be
writable by ``dnsuser''.
The machines on the private network must have their name resolution configured to ask dns.example.com (at IP 192.168.1.1 in our example), while the externally visible machines can either query the network gateway's outside interface (at IP 10.1.1.9 in our example), or the ISP's DNS servers.
In this configuration, you have chosen to expose all of your hosts. You have a real IP number for each machine in your domain, and you've given your ISP the list of machine names and IP numbers. The ISP has given you at least one IP number for their DNS host(s). Your Linux boxes are now configured for name resolution in /etc/resolv.conf:
search example.com
nameserver <DNS host 1>
nameserver <DNS host 2>
Windows boxes are configured with the same parameters, in the network settings dialogues.
The configurations described in section Setting Up Name Resolution have MX records pointing to a machine ``mail.example.com''. The MX record with the lowest priority number following tells remote sites where to send email. Other MX records with higher priority numbers are used as backup email receivers. These backups will hold the mail for a certain period of time if the primary email receiver is not able to accept the messages for some reason. In the examples in that section, I have assumed that fred.example.com, under its alias of mail.example.com, is handling email for the domain. If you have chosen to let the ISP handle all of your email hosting, you should change those MX records to point to the appropriate ISP machines. Ask your ISP technical support representative what host names you should use for the MX records in the various files.
If you have chosen to do full electronic mail hosting for your domain, you'll have to take special actions for email coming from hosts on the private network, and for allowing transparent mail reading from anywhere within the private network. Unless you're careful, messages are likely to sit around for long times if they are waiting on one host, and the intended recipient is logged on another machine. For security reasons, I recommend that the incoming email not be accessible from the externally visible hosts (this might help to discourage a PHB who wants his desktop machine to be on a real IP, then wonders why he gets brought down by a ping of death twice a day). A transparent email sharing system on the private network fairly straight-forward in sendmail. If anybody wants to provide tested solutions for other mail handling daemons, I welcome additions.
In order that email delivered to one host be visible on all machines, the simplest solution is to export the mail spool directory with read-write privileges over the entire private network. The private network gateway machine will also act as mail collector and forwarder for the entire private network, and so must have root write privileges to the mail spool drive. The other clients may or may not squash root, at your discretion. My general security philosophy is not to grant privileges unless there is a clear reason for it, so I squash root on the mail spool network drive for all hosts except the private network gateway machine. This has the effect that root can only read his mail from that machine, but this is not a particularly serious handicap. Note that the mail spool drive can be a directory on the private network gateway machine, exported via NFS, or it can be a directory on one of the internal servers, exported to the entire private network. If the mail spool drive is resident on the private network gateway, there is no issue of squashing root for that machine. If it is on another server, then note that email will be undeliverable if that server, the gateway machine, or the network connecting them, is down.
For Windows machines on your private network, you may either set up a POP server on the mail spool host, or use samba to export the mail spool to those machines. The Windows machines should be configured to send and retrieve mail under a Linux username, such as [email protected], so that the email address host name is the bare domain name, not a machine name like barney.example.com. The outgoing SMTP host should be set to the private network gateway machine, which will be responsible for forwarding the mail and doing any address rewriting.
Next, you should configure sendmail to forward email from the machines
on the private network, rewriting the addresses if necessary. Obtain
the latest sources to sendmail from the
sendmail.org WWW site. Compile
the binaries, then go to the cf/domain
subdirectory
within the sendmail source tree, and create the following new file:
example.com.m4
divert(-1)
#
# Copyright (c) 1998 Sendmail, Inc. All rights reserved.
# Copyright (c) 1983 Eric P. Allman. All rights reserved.
# Copyright (c) 1988, 1993
# The Regents of the University of California. All rights reserved.
#
# By using this file, you agree to the terms and conditions set
# forth in the LICENSE file which can be found at the top level of
# the sendmail distribution.
#
#
#
# The following is a generic domain file. You should be able to
# use it anywhere. If you want to customize it, copy it to a file
# named with your domain and make the edits; then, copy the appropriate
# .mc files and change `DOMAIN(generic)' to reference your updated domain
# files.
#
divert(0)
define(`confFORWARD_PATH', `$z/.forward.$w+$h:$z/.forward+$h:$z/.forward.$w:$z/.forward')dnl
FEATURE(redirect)dnl
MASQUERADE_AS(example.com)dnl
FEATURE(masquerade_envelope)dnl
This defines the domain ``example.com''. Next, you have to create the
sendmail.cf
files which will be used on the mail
host (the private network gateway), and on the other Linux nodes on
the private network.
Create the following file in the sendmail source tree, under
cf/cf
:
example.master.m4
divert(-1)
#
# Copyright (c) 1998 Sendmail, Inc. All rights reserved.
# Copyright (c) 1983 Eric P. Allman. All rights reserved.
# Copyright (c) 1988, 1993
# The Regents of the University of California. All rights reserved.
#
# By using this file, you agree to the terms and conditions set
# forth in the LICENSE file which can be found at the top level of
# the sendmail distribution.
#
#
#
# This is the prototype file for a configuration that supports nothing
# but basic SMTP connections via TCP.
#
# You MUST change the `OSTYPE' macro to specify the operating system
# on which this will run; this will set the location of various
# support files for your operating system environment. You MAY
# create a domain file in ../domain and reference it by adding a
# `DOMAIN' macro after the `OSTYPE' macro. I recommend that you
# first copy this to another file name so that new sendmail releases
# will not trash your changes.
#
divert(0)dnl
OSTYPE(linux)dnl
DOMAIN(example.com)dnl
FEATURE(nouucp)
FEATURE(relay_entire_domain)
FEATURE(`virtusertable', `hash /etc/sendmail/virtusertable')dnl
FEATURE(`genericstable', `hash /etc/sendmail/genericstable')dnl
define(`confPRIVACY_FLAGS', ``noexpn,novrfy'')dnl
MAILER(local)
MAILER(smtp)
Cw fred.example.com
Cw example.com
In this example we have disabled the ``expn'' and ``vrfy''
commands. An attacker could troll for aliases with ``expn'', trying
names like ``staff'', ``allstaff'', ``office'', and so on, until he
hits an alias which expands out several usernames for him. He can then
try the usernames against certain weak passwords in hopes of getting
in (assuming he can get a login prompt - the security settings
described in section
Securing Your Domain
are set up so that no login prompt is available for off-site attackers).
The other file you should create will define the sendmail.cf for the
slave machines:
example.slave.m4
divert(-1)
#
# Copyright (c) 1998 Sendmail, Inc. All rights reserved.
# Copyright (c) 1983 Eric P. Allman. All rights reserved.
# Copyright (c) 1988, 1993
# The Regents of the University of California. All rights reserved.
#
# By using this file, you agree to the terms and conditions set
# forth in the LICENSE file which can be found at the top level of
# the sendmail distribution.
#
#
#
# This the prototype for a "null client" -- that is, a client that
# does nothing except forward all mail to a mail hub. IT IS NOT
# USABLE AS IS!!!
#
# To use this, you MUST use the nullclient feature with the name of
# the mail hub as its argument. You MUST also define an `OSTYPE' to
# define the location of the queue directories and the like.
# In addition, you MAY select the nocanonify feature. This causes
# addresses to be sent unqualified via the SMTP connection; normally
# they are qualified with the masquerade name, which defaults to the
# name of the hub machine.
# Other than these, it should never contain any other lines.
#
divert(0)dnl
OSTYPE(linux)
FEATURE(nullclient, fred.$m)
Cm example.com
You build the appropriate sendmail.cf files with the command:
make example.master.cf example.slave.cf
and then copy the files to the appropriate machines under the name
sendmail.cf
.
This configuration puts most of the sendmail configuration files under
the /etc/sendmail/
subdirectory. This configuration
causes sendmail to parse and use two special files,
virtusertable.db
and
genericstable.db
. To use these special files,
create their parent files. First,
virtusertable.src
:
[email protected] jpublic
[email protected] [email protected]
[email protected] root
[email protected] #phb#@hotmail.com
This maps the email addresses on incoming email to new
destinations. Mail sent to [email protected] is delivered
locally to the Linux account jpublic. Mail to [email protected] is
redirected to another email account, possibly in a different
domain. Mail to [email protected] is sent to root, and so on.
The other file is genericstable.src
:
jpublic [email protected]
janedoe [email protected]
whgiii [email protected]
This file renames the sender on outgoing email from locally-sourced
mail. While it clearly can't affect the return address for mail sent
directly from [email protected], it allows you to rewrite
the sender's email address from the internal usernames to whatever
email address convention you've chosen. Finally, create the following
Makefile
in /etc/sendmail/
:
all : genericstable.db virtusertable.db
virtusertable.db : virtusertable.src
makemap hash virtusertable < virtusertable.src
genericstable.db : genericstable.src
makemap hash genericstable < genericstable.src
Run make to create the hashed files which sendmail
can use, and remember to re-run make and restart
sendmail (or send it a SIGHUP) after any changes to either of
these ``.src'' files.
My experience is only with sendmail. If anybody would like to write this section, please contact me. Otherwise, I may, at some later time, try to provide details myself on such MTAs as Postfix, Exim, or smail. I'd really rather somebody wrote these sections who uses those programs.
You should set up your externally visible web server on a machine outside the private network, and not on the private network gateway machine, for security reasons. If the web server needs access to databases or other resources stored on the private network, the situation becomes more complicated, both from a network and a security standpoint. Such configurations are beyond the scope of this document.
The details of setting up the server itself can be found in the apache documentation, and in the Linux WWW HOWTO document.
Once again, your FTP host should be an externally visible machine, and not the private network gateway machine. Follow the setup directions which ship with your FTP daemon package. Be sure to download the most recent version of the daemon, as there are security vulnerabilities in some older versions of many daemons. If your FTP site does not require anonymous users to upload files, be sure to disable that feature in the daemon. I recommend that user (non-anonymous) FTP logins not be permitted on the FTP host, that you require your regular users to use scp, the secure shell remote copy command, for any file updating they may have to do on the FTP host. This is to help build secure habits in the users, and to protect against the ``hostile router'' problem described in section Securing Your Domain.
This is discussed in detail in section Configuring Your Firewall.