Apache HTTP Server Version 2.0
Apache HTTP Server Version 2.0
This page could be summarized with the statement: don't configure Apache in such a way that it relies on DNS resolution for parsing of the configuration files. If Apache requires DNS resolution to parse the configuration files then your server may be subject to reliability problems (ie. it might not boot), or denial and theft of service attacks (including users able to steal hits from other users).
<VirtualHost www.abc.dom>
ServerAdmin [email protected]
DocumentRoot /www/abc
</VirtualHost>
In order for Apache to function properly, it absolutely needs
to have two pieces of information about each virtual host: the
ServerName and at least one
IP address that the server will bind and respond to. The above
example does not include the IP address, so Apache must use DNS
to find the address of www.abc.dom. If for some
reason DNS is not available at the time your server is parsing
its config file, then this virtual host will not be
configured. It won't be able to respond to any hits
to this virtual host (prior to Apache version 1.2 the server
would not even boot).
Suppose that www.abc.dom has address 10.0.0.1.
Then consider this configuration snippet:
<VirtualHost 10.0.0.1>
ServerAdmin [email protected]
DocumentRoot /www/abc
</VirtualHost>
This time Apache needs to use reverse DNS to find the
ServerName for this virtualhost. If that reverse
lookup fails then it will partially disable the virtualhost
(prior to Apache version 1.2 the server would not even boot).
If the virtual host is name-based then it will effectively be
totally disabled, but if it is IP-based then it will mostly
work. However, if Apache should ever have to generate a full
URL for the server which includes the server name, then it will
fail to generate a valid URL.
Here is a snippet that avoids both of these problems:
<VirtualHost 10.0.0.1>
ServerName www.abc.dom
ServerAdmin [email protected]
DocumentRoot /www/abc
</VirtualHost>
There are (at least) two forms that denial of service
can come in. If you are running a version of Apache prior to
version 1.2 then your server will not even boot if one of the
two DNS lookups mentioned above fails for any of your virtual
hosts. In some cases this DNS lookup may not even be under your
control; for example, if abc.dom is one of your
customers and they control their own DNS, they can force your
(pre-1.2) server to fail while booting simply by deleting the
www.abc.dom record.
Another form is far more insidious. Consider this configuration snippet:
<VirtualHost www.abc.dom>
ServerAdmin [email protected]
DocumentRoot /www/abc
</VirtualHost>
<VirtualHost www.def.dom>
ServerAdmin [email protected]
DocumentRoot /www/def
</VirtualHost>
Suppose that you've assigned 10.0.0.1 to
www.abc.dom and 10.0.0.2 to
www.def.dom. Furthermore, suppose that
def.dom has control of their own DNS. With this
config you have put def.dom into a position where
they can steal all traffic destined to abc.dom. To
do so, all they have to do is set www.def.dom to
10.0.0.1. Since they control their own DNS you can't stop them
from pointing the www.def.dom record wherever they
wish.
Requests coming in to 10.0.0.1 (including all those where
users typed in URLs of the form
http://www.abc.dom/whatever) will all be served by
the def.dom virtual host. To better understand why
this happens requires a more in-depth discussion of how Apache
matches up incoming requests with the virtual host that will
serve it. A rough document describing this is available.
The addition of name-based
virtual host support in Apache 1.1 requires Apache to know
the IP address(es) of the host that httpd is running on. To get
this address it uses either the global
ServerName
(if present) or calls the C function gethostname
(which should return the same as typing "hostname" at the
command prompt). Then it performs a DNS lookup on this address.
At present there is no way to avoid this lookup.
If you fear that this lookup might fail because your DNS
server is down then you can insert the hostname in
/etc/hosts (where you probably already have it so
that the machine can boot properly). Then ensure that your
machine is configured to use /etc/hosts in the
event that DNS fails. Depending on what OS you are using this
might be accomplished by editing /etc/resolv.conf,
or maybe /etc/nsswitch.conf.
If your server doesn't have to perform DNS for any other
reason then you might be able to get away with running Apache
with the HOSTRESORDER environment variable set to
"local". This all depends on what OS and resolver libraries you
are using. It also affects CGIs unless you use
mod_env to control the environment. It's best
to consult the man pages or FAQs for your OS.
VirtualHost
Listen
ServerName
<VirtualHost _default_:*>
server that has no pages to serveThe situation regarding DNS is highly undesirable. For Apache 1.2 we've attempted to make the server at least continue booting in the event of failed DNS, but it might not be the best we can do. In any event, requiring the use of explicit IP addresses in configuration files is highly undesirable in today's Internet where renumbering is a necessity.
A possible work around to the theft of service attack described above would be to perform a reverse DNS lookup on the IP address returned by the forward lookup and compare the two names -- in the event of a mismatch, the virtualhost would be disabled. This would require reverse DNS to be configured properly (which is something that most admins are familiar with because of the common use of "double-reverse" DNS lookups by FTP servers and TCP wrappers).
In any event, it doesn't seem possible to reliably boot a virtual-hosted web server when DNS has failed unless IP addresses are used. Partial solutions such as disabling portions of the configuration might be worse than not booting at all depending on what the webserver is supposed to accomplish.
As HTTP/1.1 is deployed and browsers and proxies start
issuing the Host header it will become possible to
avoid the use of IP-based virtual hosts entirely. In this case,
a webserver has no requirement to do DNS lookups during
configuration. But as of March 1997 these features have not
been deployed widely enough to be put into use on critical
webservers.