Internet Systems Consortium DHCP Distribution
July 14, 2004
You should read this file carefully before trying to install or use
the ISC DHCP Distribution.
TABLE OF CONTENTS
1 WHERE TO FIND DOCUMENTATION
2 RELEASE STATUS
3 BUILDING THE DHCP DISTRIBUTION
3.1 UNPACKING IT
3.2 CONFIGURING IT
3.2.1 DYNAMIC DNS UPDATES
3.2.2 LOCALLY DEFINED OPTIONS
3.3 BUILDING IT
4 INSTALLING THE DHCP DISTRIBUTION
5 USING THE DHCP DISTRIBUTION
5.1 FIREWALL RULES
5.2.1 IF_TR.H NOT FOUND
5.2.2 SO_ATTACH_FILTER UNDECLARED
5.2.3 PROTOCOL NOT CONFIGURED
5.2.6 IP BOOTP AGENT
5.2.7 MULTIPLE INTERFACES
6.1 HOW TO REPORT BUGS
WHERE TO FIND DOCUMENTATION
Documentation for this software includes this README file, the
RELNOTES file, and the manual pages, which are in the server, common,
client and relay subdirectories. The README file (this file) includes
late-breaking operational and system-specific information that you
should read even if you don't want to read the manual pages, and that
you should *certainly* read if you run into trouble. Internet
standards relating to the DHCP protocol are stored in the doc
subdirectory. You will have the best luck reading the manual pages if
you build this software and then install it, although you can read
them directly out of the distribution if you need to.
DHCP server documentation is in the dhcpd man page. Information about
the DHCP server lease database is in the dhcpd.leases man page.
Server configuration documentation is in the dhcpd.conf man page as
well as the dhcp-options man page. A sample DHCP server
configuration is in the file server/dhcpd.conf. The source for the
dhcpd, dhcpd.leases and dhcpd.conf man pages is in the server/ sub-
directory in the distribution. The source for the dhcp-options.5
man page is in the common/ subdirectory.
DHCP Client documentation is in the dhclient man page. DHCP client
configuration documentation is in the dhclient.conf man page and the
dhcp-options man page. The DHCP client configuration script is
documented in the dhclient-script man page. The format of the DHCP
client lease database is documented in the dhclient.leases man page.
The source for all these man pages is in the client/ subdirectory in
the distribution. In addition, the dhcp-options man page should be
referred to for information about DHCP options.
DHCP relay agent documentation is in the dhcrelay man page, the source
for which is distributed in the relay/ subdirectory.
To read installed manual pages, use the man command. Type "man page"
where page is the name of the manual page. This will only work if
you have installed the ISC DHCP distribution using the ``make install''
command (described later).
If you want to read manual pages that aren't installed, you can type
``nroff -man page |more'' where page is the filename of the
unformatted manual page. The filename of an unformatted manual page
is the name of the manual page, followed by '.', followed by some
number - 5 for documentation about files, and 8 for documentation
about programs. For example, to read the dhcp-options man page,
you would type ``nroff -man common/dhcp-options.5 |more'', assuming
your current working directory is the top level directory of the ISC
If you do not have the nroff command, you can type ``more catpage''
where catpage is the filename of the catted man page. Catted man
pages names are the name of the manual page followed by ".cat"
followed by 5 or 8, as with unformatted manual pages.
Please note that until you install the manual pages, the pathnames of
files to which they refer will not be correct for your operating
This is release version 3.0.1 of the ISC DHCP Distribution.
In this release, the server and relay agent are currently fully
functional on NetBSD, Linux systems with kernel version 2.2 or later,
FreeBSD, OpenBSD, BSD/OS, Digital Tru64 Unix and Solaris. The software
will also run on HP-UX, but only supports a single network interface.
Ports also exist for QNX, SCO, NeXTStep, and MacOS X, but are not in
wide use, with all that implies. We are not aware of an easy way to
get this software running on HP-UX.
The DHCP client currently only knows how to configure the network on
NetBSD, FreeBSD, OpenBSD, BSD/os, Linux, Solaris and NextStep. The
client depends on a system-dependent shell script to do network
configuration - support for other operating systems is simply a matter
of porting this shell script to the new platform.
If you are running the DHCP distribution on a machine which is a
firewall, or if there is a firewall between your DHCP server(s) and
DHCP clients, please read the section on firewalls which appears later
in this document.
If you wish to run the DHCP Distribution on Linux, please see the
Linux-specific notes later in this document. If you wish to run on an
SCO release, please see the SCO-specific notes later in this document.
You particularly need to read these notes if you intend to support
Windows 95 clients. If you are running a version of FreeBSD prior to
2.2, please read the note on FreeBSD. If you are running HP-UX or
Ultrix, please read the notes for those operating systems below. If
you are running NeXTSTEP, please see the notes on NeXTSTEP below.
If you start dhcpd and get a message, "no free bpf", that means you
need to configure the Berkeley Packet Filter into your operating
system kernel. On NetBSD, FreeBSD and BSD/os, type ``man bpf'' for
information. On Digital Unix, type ``man pfilt''.
BUILDING THE DHCP DISTRIBUTION
To build the DHCP Distribution, unpack the compressed tar file using
the tar utility and the gzip command - type something like:
zcat dhcp-3.0.1.tar.gz |tar xvf -
On BSD/OS, you have to type gzcat, not zcat, and you may run into
similar problems on other operating systems.
Now, cd to the dhcp-3.0.1 subdirectory that you've just created and
configure the source tree by typing:
If the configure utility can figure out what sort of system you're
running on, it will create a custom Makefile for you for that
system; otherwise, it will complain. If it can't figure out what
system you are using, that system is not supported - you are on
DYNAMIC DNS UPDATES
A fully-featured implementation of dynamic DNS updates is included in
this release. There are no build dependencies with any BIND version
- this version can and should just use the resolver in your C library.
There is documentation for the DDNS support in the dhcpd.conf manual
page - see the beginning of this document for information on finding
LOCALLY DEFINED OPTIONS
In previous versions of the DHCP server there was a mechanism whereby
options that were not known by the server could be configured using
a name made up of the option code number and an identifier:
"option-nnn" This is no longer supported, because it is not future-
proof. Instead, if you want to use an option that the server doesn't
know about, you must explicitly define it using the method described
in the dhcp-options man page under the DEFINING NEW OPTIONS heading.
Once you've run configure, just type ``make'', and after a while
you should have a dhcp server. If you get compile errors on one
of the supported systems mentioned earlier, please let us know.
If you get warnings, it's not likely to be a problem - the DHCP
server compiles completely warning-free on as many architectures
as we can manage, but there are a few for which this is difficult.
If you get errors on a system not mentioned above, you will need
to do some programming or debugging on your own to get the DHCP
INSTALLING THE DHCP DISTRIBUTION
Once you have successfully gotten the DHCP Distribution to build, you
can install it by typing ``make install''. If you already have an old
version of the DHCP Distribution installed, you may want to save it
before typing ``make install''.
USING THE DHCP DISTRIBUTION
If you are running the DHCP server or client on a computer that's also
acting as a firewall, you must be sure to allow DHCP packets through
the firewall. In particular, your firewall rules _must_ allow packets
from IP address 0.0.0.0 to IP address 255.255.255.255 from UDP port 68
to UDP port 67 through. They must also allow packets from your local
firewall's IP address and UDP port 67 through to any address your DHCP
server might serve on UDP port 68. Finally, packets from relay agents
on port 67 to the DHCP server on port 67, and vice versa, must be
We have noticed that on some systems where we are using a packet
filter, if you set up a firewall that blocks UDP port 67 and 68
entirely, packets sent through the packet filter will not be blocked.
However, unicast packets will be blocked. This can result in strange
behaviour, particularly on DHCP clients, where the initial packet
exchange is broadcast, but renewals are unicast - the client will
appear to be unable to renew until it starts broadcasting its
renewals, and then suddenly it'll work. The fix is to fix the
firewall rules as described above.
If you have a server that is connected to two networks, and you only
want to provide DHCP service on one of those networks (e.g., you are
using a cable modem and have set up a NAT router), if you don't write
any subnet declaration for the network you aren't supporting, the DHCP
server will ignore input on that network interface if it can. If it
can't, it will refuse to run - some operating systems do not have the
capability of supporting DHCP on machines with more than one
interface, and ironically this is the case even if you don't want to
provide DHCP service on one of those interfaces.
There are three big LINUX issues: the all-ones broadcast address,
Linux 2.1 ip_bootp_agent enabling, and operations with more than one
network interface. There are also two potential compilation/runtime
problems for Linux 2.1/2.2: the "SO_ATTACH_FILTER undeclared" problem
and the "protocol not configured" problem.
LINUX: SO_ATTACH_FILTER UNDECLARED
In addition, there is a minor issue that we will mention here because
this release is so close on the heels of the Linux 2.2 release: there
is a symlink in /usr/include that points at the linux asm headers. It
appears to be not uncommon that this link won't be updated correctly,
in which case you'll get the following error when you try to build:
lpf.c: In function `if_register_receive':
lpf.c:152: `SO_ATTACH_FILTER' undeclared (first use this function)
lpf.c:152: (Each undeclared identifier is reported only once
lpf.c:152: for each function it appears in.)
The line numbers may be different, of course. If you see this
header, your linux asm header link is probably bad, and you should
make sure it's pointing to correct linux source directory.
LINUX: PROTOCOL NOT CONFIGURED
One additional Linux 2.1/2.2 issue: if you get the following message,
it's because your kernel doesn't have the linux packetfilter or raw
packet socket configured:
Make sure CONFIG_PACKET (Packet socket) and CONFIG_FILTER (Socket
Filtering) are enabled in your kernel configuration
If this happens, you need to configure your Linux kernel to support
Socket Filtering and the Packet socket. You can do this by typing
``make config'', ``make menuconfig'' or ``make xconfig'', and then
enabling the Packet socket and Socket Filtering options that you'll
see displayed on the menu or in the questionnaire. You can also edit
your linux kernel .config file directly: set CONFIG_FILTER=y and
CONFIG_PACKET=y. If you do this, make sure you run ``make oldconfig''
afterwards, so that the changes you've made are propogated to the
kernel header files. After you've reconfigured, you need to type
``make'' to build a new Linux kernel, and then install it in the
appropriate place (probably /linux). Make sure to save a copy of your
If the preceding paragraph made no sense to you, ask your Linux
vendor/guru for help - please don't ask us.
If you set CONFIG_PACKET=m or CONFIG_FILTER=m, then you must tell the
kernel module loader to load the appropriate modules. If this doesn't
make sense to you, don't use CONFIG_whatever=m - use CONFIG_whatever=y.
Don't ask for help with this on the DHCP mailing list - it's a Linux
kernel issue. This is probably not a problem with the most recent
Linux 2.2.x kernels.
If you are running a recent version of Linux, this won't be a problem,
but on older versions of Linux (kernel versions prior to 2.2), there
is a potential problem with the broadcast address being sent
In order for dhcpd to work correctly with picky DHCP clients (e.g.,
Windows 95), it must be able to send packets with an IP destination
address of 255.255.255.255. Unfortunately, Linux changes an IP
destination of 255.255.255.255 into the local subnet broadcast address
(here, that's 22.214.171.124).
This isn't generally a problem on Linux 2.2 and later kernels, since
we completely bypass the Linux IP stack, but on old versions of Linux
2.1 and all versions of Linux prior to 2.1, it is a problem - pickier
DHCP clients connected to the same network as the ISC DHCP server or
ISC relay agent will not see messages from the DHCP server. It *is*
possible to run into trouble with this on Linux 2.2 and later if you
are running a verson of the DHCP server that was compiled on a Linux
2.0 system, though.
It is possible to work around this problem on some versions of Linux
by creating a host route from your network interface address to
255.255.255.255. The command you need to use to do this on Linux
varies from version to version. The easiest version is:
route add -host 255.255.255.255 dev eth0
On some older Linux systems, you will get an error if you try to do
this. On those systems, try adding the following entry to your
route add -host all-ones dev eth0
Another route that has worked for some users is:
route add -net 255.255.255.0 dev eth0
If you are not using eth0 as your network interface, you should
specify the network interface you *are* using in your route command.
LINUX: IP BOOTP AGENT
Some versions of the Linux 2.1 kernel apparently prevent dhcpd from
working unless you enable it by doing the following:
echo 1 >/proc/sys/net/ipv4/ip_bootp_agent
LINUX: MULTIPLE INTERFACES
Very old versions of the Linux kernel do not provide a networking API
that allows dhcpd to operate correctly if the system has more than one
broadcast network interface. However, Linux 2.0 kernels with version
numbers greater than or equal to 2.0.31 add an API feature: the
SO_BINDTODEVICE socket option. If SO_BINDTODEVICE is present, it is
possible for dhcpd to operate on Linux with more than one network
interface. In order to take advantage of this, you must be running a
2.0.31 or greater kernel, and you must have 2.0.31 or later system
headers installed *before* you build the DHCP Distribution.
We have heard reports that you must still add routes to 255.255.255.255
in order for the all-ones broadcast to work, even on 2.0.31 kernels.
In fact, you now need to add a route for each interface. Hopefully
the Linux kernel gurus will get this straight eventually.
Linux 2.1 and later kernels do not use SO_BINDTODEVICE or require the
broadcast address hack, but do support multiple interfaces, using the
Linux Packet Filter.
SCO has the same problem as Linux (described earlier). The thing is,
SCO *really* doesn't want to let you add a host route to the all-ones
On more recent versions of SCO, you can do this:
ifconfig net0 xxx.xxx.xxx.xxx netmask 0xNNNNNNNN broadcast 255.255.255.255
If this doesn't work, you can also try the following strange hack:
ifconfig net0 alias 10.1.1.1 netmask 126.96.36.199
Apparently this works because of an interaction between SCO's support
for network classes and the weird netmask. The 10.* network is just a
dummy that can generally be assumed to be safe. Don't ask why this
works. Just try it. If it works for you, great. SCO has added
support for doing DHCP in a more sensible way, but I have not had the
time or cause to implement them. If you are interested in this, and
are able to hack your way out of a wet paper back without assistance,
we'd appreciate it if you'd give it a try, but don't expect too much
support from us (sorry!).
HP-UX has the same problem with the all-ones broadcast address that
SCO and Linux have. One user reported that adding the following to
/etc/rc.config.d/netconf helped (you may have to modify this to suit
your local configuration):
Now that we have Ultrix packet filter support, the DHCP Distribution
on Ultrix should be pretty trouble-free. However, one thing you do
need to be aware of is that it now requires that the pfilt device be
configured into your kernel and present in /dev. If you type ``man
packetfilter'', you will get some information on how to configure your
kernel for the packet filter (if it isn't already) and how to make an
entry for it in /dev.
Versions of FreeBSD prior to 2.2 have a bug in BPF support in that the
ethernet driver swaps the ethertype field in the ethernet header
downstream from BPF, which corrupts the output packet. If you are
running a version of FreeBSD prior to 2.2, and you find that dhcpd
can't communicate with its clients, you should #define BROKEN_FREEBSD_BPF
in site.h and recompile.
Modern versions of FreeBSD include the ISC DHCP 3.0 client as part of
the base system, and the full distribution (for the DHCP server and
relay agent) is available from the Ports Collection in
/usr/ports/net/isc-dhcp3, or as a package on FreeBSD installation
The NeXTSTEP support uses the NeXTSTEP Berkeley Packet Filter
extension, which is not included in the base NextStep system. You
must install this extension in order to get dhcpd or dhclient to work.
One problem which has been observed and is not fixed in this
patchlevel has to do with using DLPI on Solaris machines. The symptom
of this problem is that the DHCP server never receives any requests.
This has been observed with Solaris 2.6 and Solaris 7 on Intel x86
systems, although it may occur with other systems as well. If you
encounter this symptom, and you are running the DHCP server on a
machine with a single broadcast network interface, you may wish to
edit the includes/site.h file and uncomment the #define USE_SOCKETS
line. Then type ``make clean; make''.
The DHCP client on Solaris will only work with DLPI. If you run it
and it just keeps saying it's sending DHCPREQUEST packets, but never
gets a response, you may be having DLPI trouble as described above.
If so, we have no solution to offer at this time. Also, because
Solaris requires you to "plumb" an interface before it can be detected
by the DHCP client, you must either specify the name(s) of the
interface(s) you want to configure on the command line, or must plumb
the interfaces prior to invoking the DHCP client. This can be done
with ``ifconfig iface plumb'', where iface is the name of the
interface (e.g., ``ifconfig hme0 plumb'').
It should be noted that Solaris versions from 2.6 onward include a
DHCP client that you can run with ``/sbin/ifconfig iface dhcp start''
rather than using the ISC DHCP client. The feature set of the Solaris
client is different (not necessarily better or worse) than that of the
ISC client, but in most cases it will be a lot easier for you to just
use that. Please do not ask for help in using the Solaris DHCP client
on Internet Systems Consortium mailing lists - that's why you're
paying Sun the big bucks. If you're having a problem with the
Solaris client interoperating with the ISC dhcp server, that's another
matter, but please check with Sun first.
The Internet Systems Consortium DHCP server is not a commercial
product, and is not supported by the ISC. However, it has attracted a
fairly sizable following on the Internet, which means that there are a
lot of knowledgable users who may be able to help you if you get
stuck. These people generally read the email@example.com mailing
If you are going to use dhcpd, you should probably subscribe to the
dhcp-server and dhcp-announce mailing lists. If you will be using
dhclient, you should subscribe to the dhcp-client mailing list.
If you need help, you should ask on the dhcp-server or dhcp-client
mailing list - whichever is appropriate to your application. Support
requests for the ISC DHCP client should go to firstname.lastname@example.org.
Support requests for the DHCP server should go to email@example.com.
If you are having trouble with a combination of the client and server,
send the request to firstname.lastname@example.org. Please do not cross-post to
both lists under any circumstances.
WHERE TO REPORT BUGS: If you want the act of sending in a bug report
to result in you getting help in the form of a fixed piece of
software, you are asking for help. Your bug report is helpful to us,
but fundamentally you are making a support request, so please use the
addresses described in the previous paragraphs. If you are _sure_ that
your problem is a bug, and not user error, or if your bug report
includes a patch, you can send it to email@example.com without
subscribing. This mailing list goes into a bug tracking system, so
you don't need to check periodically to see if we still remember the
bug - if you haven't been notified that the bug has been closed, we
still consider it a bug, and still have it in the system.
PLEASE DO NOT REPORT BUGS IN OLD SOFTWARE RELEASES! Fetch the latest
release and see if the bug is still in that version of the software,
and if it's not, _then_ report it. It's okay to report bugs in the
latest patchlevel of a major version that's not the most recent major
version, though - for example, if you're running 3.0pl2, you don't have
to upgrade to a 3.0.1rc (release candidate) before you can report bugs.
PLEASE DO NOT REPORT BUGS IF YOU ARE RUNNING A VERSION OF THE ISC
DHCP DISTRIBUTION THAT YOU DIDN'T GET FROM THE ISC! Free operating
system distributions are notorious for including outdated versions of
software, and also versions of software that were not compiled on your
particular version of the operating system. These versions
frequently do not work. Getting a source distribution from the ISC
and installing it frequently *does* work. Please try this *before*
asking for help.
PLEASE READ THIS README FILE CAREFULLY BEFORE REPORTING BUGS,
PARTICULARLY THE SECTION BELOW ON WHAT TO INCLUDE IN A BUG REPORT OR
PLEASE DO NOT SEND REQUESTS FOR SUPPORT DIRECTLY TO THE ENGINEERS WHO
WORK ON THE ISC DHCP DISTRIBUTION! *PARTICULARLY*, DO NOT SEND MAIL
TO THE ENGINEERS BECAUSE YOU AREN'T SURE TO WHOM YOU SHOULD SEND MAIL
- if you aren't sure, *ask* on the firstname.lastname@example.org or
email@example.com mailing list.
The number of people using the DHCP Distribution is sufficiently large
that if we take interrupts every time any one of those people runs
into trouble, we will never get any more coding done. If you send a
support request directly to any ISC or Nominum engineer, we will
forward it to the mailing list, or possibly ignore it, depending on
how much stress we are under at the time.
Please do not Cc: us on mail you send to these lists - we read both
mailing lists, so this just means we get two copies!
If your question can only be answered by one of the engineers, send it
to the appropriate public mailing list anyway - we will answer it
there. When we have time.
Please do not think "Oh, I don't want to bother the whole mailing list
with this question." If you are too embarrassed to ask publically,
get a support contract.
If you are concerned about bothering everybody on the list, that's
great, but that's what the list is there for. When you send mail to
one of the engineers, you are taking resources away from everybody on
the mailing list *anyway* - they just don't know it.
We're not writing this because we don't respect you - we really do
want to help you, and we appreciate your bug reports and comments.
But please use the mechanisms we have in place to provide you with
help, because otherwise you are almost certainly depriving someone
else of our help.
PLEASE DO NOT CALL US ON THE PHONE FOR HELP! Answering the phone
takes a lot more of our time and attention than answering email. If
you do call us on the phone, we will tell you to send email to the
mailing list or buy a support contract, so please don't waste your
time or ours. If you have a support contract, please use the support
channel mentioned in the support contract - otherwise you probably
won't get timely support unless you happen to ask an interesting
question and we happen to have some time to kill, because we can't
tell you're a support customer if you send mail to the public mailing
HOW TO REPORT BUGS OR REQUEST HELP
When you report bugs or ask for help, please provide us complete
information. A list of information we need follows. Please read it
carefully, and put all the information you can into your initial bug
report, so that we don't have to ask you any questions in order to
figure out your problem. If you need handholding support, please
consider contacting a commercial provider of the ISC DHCP
1. The specific operating system name and version of the
machine on which the DHCP server or client is running.
2. The specific operating system name and version of the
machine on which the client is running, if you are having
trouble getting a client working with the server.
3. If you're running Linux, the version number we care about is
the kernel version and maybe the library version, not the
distribution version - e.g., while we don't mind knowing
that you're running Redhat version mumble.foo, we must know
what kernel version you're running, and it helps if you can
tell us what version of the C library you're running,
although if you don't know that off the top of your head it
may be hard for you to figure it out, so don't go crazy
4. The specific version of the DHCP distribution you're
running, for example "2.0b1pl19", not "2.0".
5. Please explain the problem carefully, thinking through what
you're saying to ensure that you don't assume we know
something about your situation that we don't know.
6. Include your dhcpd.conf and dhcpd.leases file if they're not
huge (if they are huge, we may need them anyway, but don't
send them until you're asked). Huge means more than 100
7. Include a log of your server or client running until it
encounters the problem - for example, if you are having
trouble getting some client to get an address, restart the
server with the -d flag and then restart the client, and
send us what the server prints. Likewise, with the client,
include the output of the client as it fails to get an
address or otherwise does the wrong thing. Do not leave
out parts of the output that you think aren't interesting.
8. If the client or server is dumping core, please run the
debugger and get a stack trace, and include that in your
bug report. For example, if your debugger is gdb, do the
gdb dhcpd dhcpd.core
This assumes that it's the dhcp server you're debugging, and
that the core file is in dhcpd.core.
9. If you know that the problem is an actual bug, and you can
reproduce the bug, you can skip steps 6 through 8 and instead
capture a trace file using the -tf flag (see the man page for
details). If you do this, and there is anything in your
dhcp configuration that you are not willing to make public,
please send the trace file to firstname.lastname@example.org and NOT to
email@example.com, because the tracefile contains your entire
PLEASE DO NOT send queries about non-isc clients to the dhcp-client
mailing list. If you're asking about them on an ISC mailing list,
it's probably because you're using the ISC DHCP server, so ask there.
If you are having problems with a client whose executable is called
dhcpcd, this is _not_ the ISC DHCP client, and we probably can't help
you with it.
Please see http://www.isc.org/sw/dhcp/ for details on how to subscribe
to the ISC DHCP mailing lists.