TITLE: DHCP client daemon
LFS VERSION: 3.3+ (should work on older versions if bootscrips are updated)
AUTHOR: D.J. Lucas <dj_me@swbell.net>
Last updated 08/20/2002
SYNOPSIS: How to setup a DHCP client daemon. This is used with most
xDSL and cable gateways/routers and for cable ISPs. Also
used, nowadays, on almost all corporate networks to set up
non mission critical servers and most workstations.
______________________________________________________________________________
THIS HINT WILL NO LONGER BE UPDATED BEYOND 08/20/2002.
ALL INFORMATION HAS BEEN INCLUDED IN THE BLFS BOOK AVAILIBLE AT:
http://beyond.linuxfromscratch.org/
Rest assured, you will find a much broader and more thorough collection
of easy to follow instructions and 'hints' for things to do with your shiny
new LFS sytem on the BLFS site. Check it out!
______________________________________________________________________________
HINT: After reading the dhcpd hints by Simon Perreault <nomis80@videotron.ca>
and Thinus Pollard <thinusp@olienhout.org.za>, I had an excelent grasp on the
configuration and use of dhcpcd. I, however, felt the need to expand on their
ideas and incorporate the use of the DHCP client into the network (ethnet on
previous versions) init script. This will both reduce the amount of scripts in
/etc/rc.d/init.d/, and also allow for an easy change back to a static network
should the need arise.
NOTE: Just a basic reminder....because this works on my system, dosen't mean
that it will work on yours, tho it should. Even though these are just script
changes, IT IS ALWAYS A GOOD IDEA TO BACKUP any existing files that will be
changed, deleted, modified, moved, etc.... Also, I do not go into any of
the advanced options for cable/*DSL modems that may be required by some ISPs.
These options are covered in the man-pages for the client, however, if you are
having problems with broadband modems, look for the -D, -H, -I, and -i switches.
Furthur more, cable modem users will probably not want to use the -k option in
the ifdown-eth0 script below. Change to this to killproc or ifconfig down
on cable modem setups to preserve the cache file. 'man dhcpcd' for more info.
How to install and configure the DHCP client daemon according to the LFS
standards.
1. Get the current version of dhcpcd at:
ftp://ftp.phystech.com/pub/dhcpcd-1.3.22-pl1.tar.gz
2. Obtain the LFS patch for DHCPCD.
http://www.lucasit.com/linux/dhcpcd-1.3.22-pl1.patch.bz2
3. Unpack the archive, patch, and install.
_______________________________________________________________________________
tar -zxf dhcpcd-1.3.22-pl1.tar.gz
bzip2 -d dhcpcd-1.3.22-pl1.patch.bz2 &&
cd dhcpcd-1.3.22-pl1 &&
patch -Np1 -i ../dhcpcd-1.3.22-pl1.patch &&
./configure --prefix="" --sysconfdir=/var/lib \
--mandir=/usr/share/man &&
make &&
make install
_______________________________________________________________________________
Command Explanation:
patch -Np1 -i ../dhcpcd-1.3.22-pl1.patch
The LFS patch changes the location of the cache, info, and pid files to be
both LFS freindly and FHS compliant. This was added to make sure that cleanfs
correctly removes the pid files when the system is restarted without being
properly shut down. FYI: Cache files are now kept in /var/cache, info files
are still kept in <ConfigDir> which is now, by default /var/lib/dhcpc, and
pid files are now stored correctly in /var/run. All three changes were made
to make the client FHS compliant.
--prefix=""
We do not want to install dhcpcd in prefix=/usr because in a lot of systems,
usr may be network mounted. Also, this is not a program that we'd have most
users access, so it makes sense to me to install in /sbin. However, it is
your LFS system, so do as you see fit.
--sysconfdir=/var/lib
Tells the install script to create the */dhcpc directory in /var/lib as this
is now the default location of <ConfigDir>.
--mandir=/usr/share/man
Tells the install script to place the man files in /usr/share/man.
_____________________________________________________________________________
4. Create the new device specific scripts.
If you are using a version of LFS prior to 3.3, see my old hint for the
scripts that will work. The old hint is availible here:
http://hints.linuxfromscratch.org/hints/old/dhcpcd.txt
We will create the $nework_devices/ifup-eth? and $network_devices/ifdown-eth?
files. This method proves to be a much easier method for new device types that
weren't accounted for in the base build. Also, we now will not mess with your
existing config files, so should you need to return to a static assigned IP,
we can simply delete the scripts we've created. In the following example, we
are creating a script for eth0, if you need for eth1, rename the files to
ifup-eth1 and ifdown-eth1 respectively, and edit for eth1.
______________________________________________________________________________
cat > /etc/sysconfig/network-devices/ifup-eth0 << "EOF"
#!/bin/sh
# Begin /etc/sysconfig/network-devices/ifup-eth0
source /etc/sysconfig/rc
source $rc_functions
echo "Bringing up the eth0 interface via DHCP..."
/sbin/dhcpcd eth0
evaluate_retval
# End /etc/sysconfig/network-devices/ifup-eth0
EOF
_______________________________________________________________________________
cat > /etc/sysconfig/network-devices/ifdown-eth0 << "EOF"
#!/bin/sh
# Begin /etc/sysconfig/network-devices/ifdown
source /etc/sysconfig/rc
source $rc_functions
echo "Bringing down the eth0 interface..."
/sbin/dhcpcd -k eth0
evaluate_retval
# End /etc/sysconfig/network-devices/ifdown
EOF
chmod 755 /etc/sysconfig/network-devices/ifup-eth0
chmod 755 /etc/sysconfig/network-devices/ifdown-eth0
______________________________________________________________________________
5. Bootscrpt handling.
Another issue has come to surface with the way LFS handles shutdowns and
restarts in the new scripts. By default, in runlevel 0 and runlevel 6, network
is called as K80, this causes a problem because dhcpcd is exacly as it's name
implies, a daemon. In both of these runlevels, sendsignals is run at K50,
before the network script has a chance to run, therefore, the dhcp client
daemon is killed by the time the script runs. I don't know if this is a good
idea or not, but it works for what little I have on my system. It seems to me
that a lot of network services that will possibly be installed will be daemons,
and that sendsignals should be run after the network is properly shut down.
I'd appreciate any feedback that anybody could give on this statement. Run
the following commands to eliminate this problem as I have.
------------------------------------------------------------------------------
mv /etc/rc.d/rc0.d/K80network /etc/rc.d/rc0.d/K45network
mv /etc/rc.d/rc6.d/K80network /etc/rc.d/rc6.d/K45network
______________________________________________________________________________
As you can see, this will run the network script at K45, prior to sendsignals
at K50, eliminating this problem. There may be other problems that surface if
you do not move network dependent Kills before K45 on some systems. This may
not be the ideal solution for all systems. It's your own LFS, check it out.
It has also been suggested that sendsignals should be moved to K70 (moving
everything currently behind it as well to allow for more room between the K40
and K50. This is up to you, again, it's your LFS system, do what you like!
_____________________________________________________________________________
6. Changes to /etc/sysconfig/network
Note: The dhcpcd client handles the route setup internally if a default gateway
is provided by the DHCP server. If a card configured via DHCP is given a
default gateway by the DHCP server, then that card is automatically put into
the routing table as a default route. If you should need to disable this
feature, see the manpages for dhcpcd and route (man dhcpcd or man route)
You will need to chage the /etc/sysconfig/network file if the default route
device is now being configured by DHCP. If your default route device is
now set up by DHCP, then you'll need to comment out (#) the "GATEWAY" and the
"GATEWAY_IF" variables. Make a backup copy and fix it with the following:
_______________________________________________________________________________
cd /etc/sysconfig
cp network network.bak
sed 's/GATEWAY/# GATEWAY/' network.bak > network
_______________________________________________________________________________
7. Before you reboot you need to test the system. Start by running ifconfig
with no options to see which devices are enabled. Disable all devices except
lo by running:
_______________________________________________________________________________
ifconfig <device name> down
_______________________________________________________________________________
Now start the network using the following command:
_____________________________________________________________________________
/etc/rc.d/init.d/network start
_____________________________________________________________________________
Look for any failures and try to track them down. If you have no failures,
run ifconfig and make sure everything looks okay. Also check that your
routing is set up correctly by running route with no argurments. If
everything looks good, then stop the network.
_____________________________________________________________________________
/etc/rc.d/init.d/network stop
_____________________________________________________________________________
Again look for any failures and track them down. Again run route and ifconfig,
and look for any interfaces that aren't supposed to be there. If there are no
interfaces, besides lo and any static cards, returned by ifconfig and there
are no routes set,congratulations. Reboot your system and continue on your
path with LFS.
If you do have failures that you cannot track down. Please feel free to
e-mail me <dj_me@swbell.net> or better yet, use the blfs-suport mailing list
<blfs-support@linuxfromscratch.org>. Be sure to include the full problem
description (as much of it as you understand), the exact error message, and
the text of all your relevant config files. Also, if you are not subscribed
to the mailing list, please request that you be CC'd on all replies to your
inquiry for help. If you do have problems, you've made a backup of everything,
so put the backups back where you got them from.
______________________________________________________________________________
8. Conclusion: "I hope this helps somebody out there ;)" -- Thinus Pollard
I'd like to thank Thinus Pollard and Simon Perreault for their versions of
this hint....which had my LFS system up and running in under an hour. I'd
also like to thank the following dedicated LFS users who took the time to track
down the problems in the previous versions of this hint.
Special thanks to: Robert Smol, Rich Jones, Phil Gendreau, Cli - corrected
dhcpcd -k $Device, Markku Tikkanen, Tijmen Stam - provided cable modem info,
Steve Hayashi, Wolfgang Scheicher - provided dhcpcd default gateway and routing
info, Carl Spelkens - corrected the runlevels before I even wrote this hint,
and Oliver Brakmann - provided the patch for correcting the locations of the
state, cache, and pid files.
The new scripts in 3.3+ have disolved quite a few of the issues encountered in
the previous hints. So a big thanks to Gerard and the entire LFS Devel crew!
Many many thank for a truely incredible project! As always, please send any
tips, suggestions, flames...etc, to: dj_me@swbell.net.
-- D.J. Lucas <dj_me@swbell.net>