Re: FAI in italian
> The present documentation of FAI covers about sixty pages of written > text. A single book describing a less advanced product like Microsoft > RIS and MSI packaging never encompass less than 500 pages. Even though > FAI for Debian GNU/Linux is a lot more flexible and configurable, the > documentation only describes the default setup. Hmm, I'm one of the experience Linux people that decided to use FAI. Before this, I used Slackware, RedHat, Mandrake and played from Time to time with Knoppix and was forced to look after SuSE installations at customers. I made my own Mandrake Mass-Installation CD and Mandrake Network-Installation Setup. I made my own RPMs and am now making my own DEBs to customize things. For me the fact that FAI "only" had 60 pages was a plus. I don't want to read reading 500 pages where 440 pages do repeating mostly things that I already know. It would be tedious to filter out the relevant information. Therefore I'd suggest that try to earn some money and use your documentation and ask O'Reilly or some other publisher if they help you to create a book for it, e.g "Mass Linux Installation for the Masses" :-)
Re: fun with FAI and Debian sid
> Why not install Sarge with FAI and then upgrade? That would probably be > a lot smoother. Probably. However, I don't have the necessary *.deb's in my local Debian repository. Also I guess that this would then not be a 100% *F*ully *a*utomated *i*nstallation. And to get it 100% FAI, I'd have to invest an equal amount of time as I had to invest now. Greetings, Holger
fun with FAI and Debian sid
Hi all!
I had the plan to install systems via FAI with Debian SID and run into a
couple of problems.
One was because of the abysmal debootstrap, which is completely broken for
"sid". People on #fai helped me and these are the steps I needed to let it
run smoothly:
debootstrap Devices
---
Get debootstrap from Debian stable, extract (e.g. via "mc")
the /usr/lib/debootstrap/devices.tar.gz. Replace the devices.tar.gz from
sid's debootstrap with this one. Otherwise you will only have about 80
devices in $NFSROOT/dev and FAI could not even open tty2, tty3 and tty4.
debootstrap Dependencies
Then you need lots of options in your /etc/fai/make-fai-nfsroot.conf to help
the braindead dependencies of debootstrap. I use
FAI_DEBOOTSTRAP_OPTS="--exclude=pcmcia-cs,ppp,pppconfig,pppoe,
pppoeconf,dhcp-client,exim4,exim4-base,exim4-config,
exim4-daemon-light,mailx,at,fdutils,info,modconf,libident,
logrotate,exim,man-db --include=libstdc++6,gcc-4.0-base,
libsigc++-1.2-5c2,libgdbm3,gettext-base,netkit-inetd,
aptitude,libssl0.9.7,libstdc++5,libdb4.2,libopencdk8,
laptop-detect,dmidecode,gcc-3.3-base,groff-base"
(all in one long line). The line was created incrementally, maybe there's room
to optimize it. What was weird is that I had to --exclude man-db, otherwise
it made an error, saying it groff-base was missing. But chroot'ing into
$NFSROOT and a 'dpkg -l "groff*"' revealed that groff-base was there ...
As I said, debootstrap is an abyssmal tool.
Local apt mirror
I have a local mirror which I create with reprepro. That's all fine, but the
new apt in debootstrap wants to check GnuPG signatures and barks if the
signatures are wrong.
So I changed /usr/sbin/make-fai-bootstrap:
create_base() {
if [ "$FAI_DEBOOTSTRAP" ]; then
call_with_stamp call_debootstrap $FAI_DEBOOTSTRAP
$ROOTCMD apt-get clean
+ mkdir -p $NFSROOT/etc/apt/apt.conf.d
+ echo 'APT::Get::AllowUnauthenticated "true";' \
+ >$NFSROOT/etc/apt/apt.conf.d/fai-unauthori
rm -f $NFSROOT/etc/resolv.conf
echo "Creating base.tgz"
tar -l -C $NFSROOT -cf - --exclude var/tmp/base.tgz . | \
gzip > $NFSROOT/var/tmp/base.tgz
else
die "\$FAI_DEBOOTSTRAP not defined."
fi
}
Otherwise, you'd have to use '-y --force-yes' all over the way to get your
unsigned (mirrored or local) *.debs installed.
BTW: make-fai-nfsroot has a little quirk: tar should not use -l but
--one-file-system. tar from Debian "sid" spits out a warning if you use -l.
feature request: --one-file-system
While doing "make-fai-nfsroot": ... Creating base.tgz tar: Semantics of -l option will change in the future releases. tar: Please use --one-file-system option instead. ...
fai-mirror has problems with variables
# export kernelversion=2.6.10
# export kernelsubversion=1-686
# fai-mirror -cDEFAULT -v .
Getting package information
Downloading packages for classes: DEFAULT
Reading config file /etc/fai/install_packages.conf
install_packages: read config
file /usr/local/share/fai/package_config/DEFAULT
install_packages: executing apt-get -qq -d -o
Aptitude::Log=/dev/null -o Dir::State::status=./aptcache/statefile
-o Dir::Cache=./aptcache/var/cache/apt -o
Dir::State=./aptcache/var/cache/apt -o
Dir::Cache::Archives=./aptcache/var/cache/apt/archives -o
Dir::Etc::sourcelist=./aptcache/etc/apt/sources.list -o
Dir::State::Lists=./aptcache/var/lib/apt/lists/ -y --force-yes
--fix-missing install file hdparm less module-init-tools
install_packages: executing true
install_packages: executing true
install_packages: executing true
install_packages: executing true
Adding packages from variable $addpackages:
kernel-image-${kernelversion}-${kernelsubversion}E: Regex compilation
error - Invalid preceding regular expression
kernel-image-${kernelversion}-${kernelsubversion}E: Regex compilation
error - Invalid preceding regular expression
kernel-image-${kernelversion}-${kernelsubversion}E: Regex compilation
error - Invalid preceding regular expression
...
fai-mirror is missing a dependency
Hi ! I just tried fai 2.7 and issued the "fai-mirror" command. It did not work out-of-the-box, because the fai deb doesn't include "libapt-pkg-perl" as a dependency.
Re: Newbie problem: IP address getting set to hostname
> hostname gets set to its IP address. Is anyone aware what could be > causing this? Yes, and if you would have checked the mailing list archive, you would know that. See the thread starting 24th January with "fai trick: weird hostname". Archive: http://www.uni-koeln.de/bin2/maillist/linux-fai?step=50
Re: dhcp doesn't seem to set the hostname on a client
> I have one problem with the FAI clients: They boot perfectly using DHCP > (with kernel append ip=dhcp), but the hostname is set to the IP. See the thread "fai trick: fix weird hostname" just some days ago in the mailing list archive. It includes a way how I solved this.
Re: screen ant
> If not, does anyone have the faintest idea of what I might need to > examine to debug this? I simply have no idea what could be causing this > or where to start looking. I'd "apt-get strace" and then run your program in both environments under the auspiece of strace. Then compare the result, e.g. do something like this: $ strace -ooutput -efile -ff $ grep open output* on both systems and find out if it was unable to open some file.
Re: fai trick: fix weird hostname
> hi holger,
>
> Did I get you right - you cannot modify die DHCP server's
> configuration?
...
> host hostnameXYZ {
> hardware ethernet 00:02:B3:ff:ff:ff;
> fixed-address 192.168.X,Y;
> }
No, you didn't get me right. I don't WANT to assign hostnames. Those IP
addresses are from a dynamic range with a netmask 255.255.0.0 and I don't
want to assign 65536 different hostnames.
The IP address is used today to create a server that is shipped tomorrow.
And in a week the IP address is used to repair a handheld data entry
system. And next month it might be used for another thing.
So having hostnames, even artifical ones like ip16-145.mn-solutions.de is
not what I want.
Instead, for me it's ok if all 16 servers I'm configuring get the same
hostname. They are installed into some customers network and they don't
care about hostname or, if they care, assign their own name.
Re: fai trick: fix weird hostname
> In fai 2.6.6 you can set the hostname by appending following parameter > to the kernel: > > HOST=demohost > > This will also set the hostname. For example. the fai CD contains > following boot entry for grub: I tried that first, but it did not work. The code line in /usr/lib/fai/sbin/rcS_fai reads: - if [ $HOSTNAME = '(none)' -a -n "$HOST" ]; then - and therefore HOST= on the kernel command line only works if HOSTNAME had the previous value '(none)'. But in my setup this is not true, the DHCP client set it to something numeric: # cd /tmp/fai # grep 172.16.1.145 * -- boot.log:IPADDR='172.16.1.145' dhclient.log:bount to 172.16.1.145 -- renewal in 3600 seconds. dmesg.log: variables.sh: HOSTNAME='172.16.1.145' variables.sh: IPADDR='172.16.1.145' variables.sh: sndhostname='172.16.1.145' -- I'm using ISC DHCP Client V3.0.1.
fai trick: fix weird hostname
Sometims I have to setup similar servers for customers. They don't end up
in our local LAN, so they don't get local IP adresses. Instead, they get
some short-lived dynamic IP address via DHCP.
There is no hostname in our local DNS for those DHCP addresses, so FAI
assigns the IP address (!) to HOSTNAME. But '172.16.1.241' is a weird
hostname. This is later used in /etc/hostname, /etc/hosts and probably at
other places.
The following hook fixes this behavior:
/usr/local/fai/hooks/setup.DEFAULT.source
echo $HOSTNAME | grep -q ^[0-9] && {
export HOSTNAME=linux
hostname $HOSTNAME
}
qemu and fai?
I just apt-get (apt-got?) qemu and was able to run the demo linux image. How would I now use qemu to make fai installation tests? Is there any howto somewhere? I don't know the exact virtualization level of QEmu, e.g. can I directly run the pxelinux image? Or should I try to get it root-mounting the FAI nfs image?
Re: more than 2TB?
> See above. I'm maintaining a FAI setup tree for production use and > won't take the risk to break something. You can have several, independend FAI setup tree! Several ways are possible *) use the -c option to make-fai-nfsroot and a completely different fai fai config and nfsroot directory *) use a different machine for the test-FAI and let your DHCP-Server serve a different root-path for those machines that you test with FAI.
Bug with mixing "PACKAGE install" and "PACKAGE aptitude"
I read in the changelog that FAI can also install via aptitude. So I changed one of my files in package_config from "PACKAGE install" to "PACKAGE aptiude". In software.log however I saw that all files have been installed via aptitude. -- M&N Solutions GmbH http://www.mn-solutions.de Holger Schurig Dieselstr. 18 61191 Rosbach v.d.Höhe Tel: (+49) 6003 9141 0 Fax: (+49) 6003 9141 49
Question regarding file modes
man fai says:
-m user,group,mode
Set user, group and mode for all copied files (mode as octal
number, user and group
numeric id or name). If not specified, use file file-mode or
data of source file.
however, it should be 'file-modes', not 'file-mode'.
However, I have a file /usr/local/share/fai/files/etc/quagga/zebra.conf/GATEWAY
which contains
--
quagga quagga 640
--
After install, shell.log says just
--
reading /fai/files/etc/quagga/zebra.conf/file-modes
--
without any problems. Unfortunately, when I later boot, the file has the wrong
owner:
--
-rw-r- 1 root root 106 Nov 12 13:44 zebra.conf
--
The fcopy took place very late in the installation, via
scripts/LAST/S80-fcopy-recursive:
--
#!/bin/sh
fcopy -B -r /
exit 0
--
So, I guess the error is because of fcopy's name2num function:
--
sub name2num {
# convert names to numeric uid, gid
my ($user,$group) = @_;
my $uid = ($user =~ /^\d+$/) ? $user : getpwnam $user;
my $gid = ($group =~ /^\d+$/) ? $group : getgrnam $group;
warn "name2id $user = $uid ; $group = $gid\n" if $debug;
return ($uid,$gid);
}
--
which uses getpwnam / getgrnam. This is wrong, because the uid and gid should
be taken from $target/etc/passwd or $target/etc/group, but not from /etc/passwd
or /etc/group.
--
M&N Solutions GmbH http://www.mn-solutions.de
Holger Schurig
Dieselstr. 18
61191 Rosbach v.d.Höhe
Tel: (+49) 6003 9141 0 Fax: (+49) 6003 9141 49
sample on how to do debconf integration
Hi ! Recently, the new debian-installer got auto-installation support via debconf-preseeding. Currently, FAI doesn't do anything with debconf. I want to try this, so I created a hooks/mirror.LAST file which contains this - #!/bin/bash echo "Task mirror" for cl in $classes; do echo " trying debconf/$cl" if [ -f "/fai/debconf/$cl" ]; then $ROOTCMD debconf-set-selections <"/fai/debconf/$cl" && echo " loaded debconf/$cl" fi done unset cl - and in debconf/ I can now have files named like classes, e.g debconf/X11_BASE: - xserver-xfree86 xserver-xfree86/autodetect_mouseboolean true xserver-xfree86 xserver-xfree86/autodetect_monitor boolean true xserver-xfree86 xserver-xfree86/config/monitor/lcd boolean false - because task 'mirror' runs before task 'instsoft', the debconf should already contain the needed info for the postinst script. However, I'm not an expert on Debian yet, so I'm not sure if this is the right approach. If it is so simple, then why isn't it in FAI already? :-) So, any comments about this?
Re: change root password for clients installation
On Friday 05 November 2004 11:55, Gerald Houart wrote:
> perhaps a stupid question but i was not able to modify root password
> for clients installation!
>
> when i do so with an existing password (from password or shadow file)
> on another debian linux computer it is not usable after installation
> what kind of encryption is used to generate root password in fai? how
> to generate a password to insert in DEFAULT.var ?
>
I have this script:
scripts/DEFAULT/S10-etc-passwd:
--
#!/bin/sh
$ROOTCMD shadowconfig on
perl -pi -e 's{^root::}{root:$ENV{rootpw}:}' $target/etc/shadow
--
and in classes/.var I have:
--
rootpw='$1$DW59daXn$EfS4ThGSF44.JGaPybPnD/'
--
I optained the hash by logging in normally, using the normal "passwd" command
and then I copied the md5-hash from /etc/shadow.
Re: fcopy: "file-modes" for directories?
> for directories instead of plain files? Is there a reason why this > would not make sense, or am I just the first one who could use this Maybe it makes sense, but maybe you can help yourself with this: install -m 0700 -d $target/some/dir ? fcopy /some/dir/somefile
Re: TFTP Timeout
> aracne:/# in.tftpd & > [1] 399 > aracne:/# tail -1 /var/log/messages > Oct 27 17:11:16 aracne in.tftpd[399]: recvfrom: Socket operation on > non-socket > [1]+ Exit 74 in.tftpd > aracne:/# grep tftp /etc/inetd.conf > tftp dgram udp waitroot /usr/sbin/tcpd When you enter in.tftp into /etc/inetd.conf, then you send a HUP to inetd and let it call it. In this case, you don't call it manually. Also, the tftp entry you grepped is without relevance. It calls tcpd, which is from tcpwrappers. But it doesn't call in.tftpd in turn. So maybe you uncomment that line. Please post you next message only after you read man in.tftpd man inetd man inetd.conf You should also read section 3.2 in the FAI manual, and do it as described there. Whenever you do in.tftpd & you do something wrong. PERIOD.
Re: TFTP Timeout
> Oct 27 15:58:00 aracne in.tftpd[351]: recvfrom: Socket operation on > non-socket in.tftpd is written so that it is normally run from inetd, via /etc/inetd.conf. When you call in.tftpd from the command line, you have to add the -l command line switch, as "man in.tftpd" will tell you: OPTIONS -l Run the server in standalone (listen) mode, rather than run from inetd. ... When called via inetd, the stdin is not a terminal, but a socket (the one that inetd opened), so in.tftpd can do socket operations on stdin. When you call it from commandline, without -l, then stdin is a tty. And tty's won't allow socket operations.
Re: distributing package updates
> How do those of you that installed a sizable number of Debian boxes > manage your package updates? I'm trying to figure out how to get around > apt-get needing you to sit there and answer the same questions for each > package on each box in a group of identically configured machines. I'm not exactly in love with cfengine2, but you can use it for centralized updates.
bug in /usr/share/perl5/Debian/Fai.pm
When you look at this excerpt ...
sub read_all_info () {
read_disk_info;
read_memory_info;
read_kernel_messages;
}
... and you know perl, then you would realize that you should rewrite this to
sub read_all_info () {
read_disk_info();
read_memory_info();
read_kernel_messages();
}
to get it warning-free.
"perl -w /usr/share/doc/fai/examples/advanced/class/05nic" will show you the warnings.
--
M&N Solutions GmbH http://www.mn-solutions.de
Holger Schurig
Dieselstr. 18
61191 Rosbach v.d.Höhe
Tel: (+49) 6003 9141 0 Fax: (+49) 6003 9141 49
Duplicated FAI-Classes ?
During installation, my /tmp/fai/FAI_CLASSES looks like: DEFAULT LINUX I386 MNWH PART_ONE BOOT_MBR BOOT_LILO NET_BASE MNWH PART_ONE BOOT_MBR BOOT_LILO NET_BASE mnwh LAST how does this come? My 02alias is the usual one: uname -s | tr '[:lower:]' '[:upper:]' [ -x "`which dpkg`" ] && dpkg --print-installation-architecture | tr /a-z/ /A-Z/ case $HOSTNAME in mnwh) catnc mnwh ;; esac and mnwh is also simple: MNWH PART_ONE BOOT_MBR BOOT_LILO NET_BASE
idea for faimond
Currently, faimond just display things that it got, once could also do netcat -p 4711 -l instead. However, we could add some little protocol into the stream to do this: * log start/stops of tasks as now * transfer the *.log files to the server, without unsafe protocols like rsh * do the chroot-thingy via faimond This way, one central instance (faimond) would have write rights to FAI's configuration resp. logging space and not several clients in your network. -- M&N Solutions GmbH http://www.mn-solutions.de Holger Schurig Dieselstr. 18 61191 Rosbach v.d.Höhe Tel: (+49) 6003 9141 0 Fax: (+49) 6003 9141 49
