Re: FBSD 5.0 diskless environment does not work!
On Tue, 11 Mar 2003, M. Warner Losh wrote: :In message: [EMAIL PROTECTED] :Hartmann, O. [EMAIL PROTECTED] writes: :: Can anyone help? Has someone a runnng diskless FBSD 5.0-R/5-CURRENT :: environment? : :I fixed a couple of bugs in the /etc/rc.d files that broke diskless :boots about a month or two so after 5.0-RELEASE. It would have :precluded diskless systems network from working most of the time. : :Warner : Dear Warner. I cvsupdate, build worl and already mergmastered the newest stuff from FreeBSD 5.0-CURRENT to avoid being outdated. Possibly I should repeat in short terms what I did when 'migrating' from the working 4.X diskless system to the 5.0 system. I backuped everything so I can switch back to the working stuff. I installed a clear, fresh copy of a FreeBSD 5.0-CURRENT into the root path of the diskless stations, e.g. for the X11 terminals this is /usr/diskless/xterm. I populated ROOT/conf/[base,default,IP. as well as ROOT/etc. The config looks like very primitive and I did the whole config from scratch, so there should nothing be left from the 4.0 config. DHCP (isc-dhcpd) works like it worked before, no changes. The diskless station offers a request via PXE (Intel fxp device), gets IP and pxeboot file, loads kernel, bootstraps kernel and then does the whole stuff as it would be a standalone machine. It does not touch the diskless scripts! I checked this by adding some echo lines which reflect that those scripts has been involved. It seems to me that after the kernel has been booted and the init process is walking through all the configurations the system does not recognize itself being diskless. I'm sorry, I have no glue how this gets figured out, maybe there should be a special marker in some of the loader files or in device.hints or within KERNEL config. If I can figure out what is to do I would probably be able to boot into diskless. If I understand what the rc_ng system is doing, the first script is initdiskless in the hierarchy. But it does not get touched anyway. Next I try is to figure out whether there is a kernel-switch which could be set by some loader.conf tags telling that this kernel is some diskless kernel ... Sorry bothering you and wasting your time. Oliver -- MfG O. Hartmann [EMAIL PROTECTED] -- Systemadministration des Institutes fuer Physik der Atmosphaere (IPA) -- Johannes Gutenberg Universitaet Mainz Becherweg 21 55099 Mainz Tel: +496131/3924662 (Maschinenraum) Tel: +496131/3924144 (Buero) FAX: +496131/3923532 To Unsubscribe: send mail to [EMAIL PROTECTED] with unsubscribe freebsd-current in the body of the message
Re: FBSD 5.0 diskless environment does not work!
On Wed, Mar 12, 2003 at 03:43:52PM +0100, Hartmann, O. wrote: Possibly I should repeat in short terms what I did when 'migrating' from the working 4.X diskless system to the 5.0 system. I think some of the directory layout needed for diskless nfs booting was frobbed by dillon a few months ago (I had some trouble updating my diskless machines past this change, although I think that was mostly because of a typo in his commit that went unnoticed/untested for a few months). Compare your previous installation to the comments in rc.diskless (and I presume, the manpage). Kris pgp0.pgp Description: PGP signature
Re: FBSD 5.0 diskless environment does not work!
On Wed, 12 Mar 2003, Kris Kennaway wrote:
Dear Kris.
At the momnent I do several stupid tasks on the faulty environment
and I got several new informations which are hopefully helpful.
I deleted all rc.* files in etc except rc and rc.subr because they
seem to be needed. I wish to use rc_ng and therefore configured
rc.conf (by default it does).
Within rc I found this line:
files=`rcorder -k ${os} -s nostart /etc/rc.d/* 2/dev/null`
It shows us the order in which the rc.d/-located scripts get passed
through. The first one is 'initdiskless', when I do
files=`rcorder -k FreeBSD -s nostart *
being within the rc.d of the diskless environment.
Just for fun I provided this file with an additional line at the first
place it should show me something like
echo Hello, this is initdiskless
and then tried to boot the diskless station again.
The boot process was normal, kernel got bootstrapped but this specific file
never got touched (in meanwhile I tried to set manually 'nfs.diskless_valid=1'
in boot/loader.conf.local).
I think something got wind up in rc or rc.subr that prevents initdiskless get
executed. I'm not competent to do the work (braindamaged, stupid guy I am...).
Maybe there is another tricky point on which the whole thing collapses which
is invisible to me.
I nice feature would be to have some 'knob' switching on/off debugging, maybe
this is possible or already realized in the shell? How to do the verbosity task?
Within an other experiment I tried the old way to pass through the init scripts,
means: using the old rc.*-styles.
Then I get another weird error message. These scripts obviously walk through
rc.diskless1
and rc.diskless2 but I got a failure about a missing device md0c, which means to me:
no RAM disk. And therefore all subsequent errors are relying on this because the
copy actions fail.
Bytheway, I do not know how to do a fast and 'out of the long bug report way' bug
report: while the newest 5.0-CURRENT stuff got rid of Kerberos IV, thenaming of the
kerberos/heimdal services did some changing and in /etc/defaults/rc.conf it has
to be
kadmind5_server=/usr/libexec/kadmind # path to kerberos 5 admin daemon
^
instead of
kadmind5_server=/usr/libexec/k5admind # path to kerberos 5 admin daemon
.
Best wishes and thanks for your quick response,
Oliver
:On Wed, Mar 12, 2003 at 03:43:52PM +0100, Hartmann, O. wrote:
:
: Possibly I should repeat in short terms what I did when 'migrating'
: from the working 4.X diskless system to the 5.0 system.
:
:I think some of the directory layout needed for diskless nfs booting
:was frobbed by dillon a few months ago (I had some trouble updating my
:diskless machines past this change, although I think that was mostly
:because of a typo in his commit that went unnoticed/untested for a few
:months). Compare your previous installation to the comments in
:rc.diskless (and I presume, the manpage).
:
:Kris
:
--
MfG
O. Hartmann
[EMAIL PROTECTED]
--
Systemadministration des Institutes fuer Physik der Atmosphaere (IPA)
--
Johannes Gutenberg Universitaet Mainz
Becherweg 21
55099 Mainz
Tel: +496131/3924662 (Maschinenraum)
Tel: +496131/3924144 (Buero)
FAX: +496131/3923532
To Unsubscribe: send mail to [EMAIL PROTECTED]
with unsubscribe freebsd-current in the body of the message
Re: FBSD 5.0 diskless environment does not work!
On Wed, 12 Mar 2003 19:56:47 +0100 (CET) Hartmann, O. [EMAIL PROTECTED] wrote: I nice feature would be to have some 'knob' switching on/off debugging, maybe this is possible or already realized in the shell? How to do the verbosity task? rc_debug=yes -- Mike Makonnen | GPG-KEY: http://www.identd.net/~mtm/mtm.asc [EMAIL PROTECTED] | Fingerprint: D228 1A6F C64E 120A A1C9 A3AA DAE1 E2AF DBCC 68B9 To Unsubscribe: send mail to [EMAIL PROTECTED] with unsubscribe freebsd-current in the body of the message
Re: FBSD 5.0 diskless environment does not work!
On Wed, 12 Mar 2003, Kris Kennaway wrote: :On Wed, Mar 12, 2003 at 03:43:52PM +0100, Hartmann, O. wrote: : : Possibly I should repeat in short terms what I did when 'migrating' : from the working 4.X diskless system to the 5.0 system. : :I think some of the directory layout needed for diskless nfs booting :was frobbed by dillon a few months ago (I had some trouble updating my :diskless machines past this change, although I think that was mostly :because of a typo in his commit that went unnoticed/untested for a few :months). Compare your previous installation to the comments in :rc.diskless (and I presume, the manpage). : :Kris : Dear Kris, It's me again. I tried now the old fashioned rc.-script way and that do some 'diskless' recognition. The first row after the kernel has been mounted its root from the server is filled with IP, gateway etc. The next row shows some formating information and reports that md0c (RAM disk?) has been formated. But then the next row shows another error: mount: /dev/md0c: No such file or directory Then a bunch of cp failures shows up and at the end I receive a dhcp.host-name error (but this can be ignored, I think) and kernel drops into /bin/sh as it does when boot fails (and it fails in this case, but I compared it with a normal disk-boot). I did a ls -l on /dev and saw /dev/md0c already there. It seems weird to me ... sorry. rc.d does not work, definitely not. Maybe some of the routines in rc or rc.subr skip over initdiskless or some other essential scripts. I will give up for today, switch back to the running config and images up to tommorrow. Will do tomorrow next investigations ... oliver -- MfG O. Hartmann [EMAIL PROTECTED] -- Systemadministration des Institutes fuer Physik der Atmosphaere (IPA) -- Johannes Gutenberg Universitaet Mainz Becherweg 21 55099 Mainz Tel: +496131/3924662 (Maschinenraum) Tel: +496131/3924144 (Buero) FAX: +496131/3923532 To Unsubscribe: send mail to [EMAIL PROTECTED] with unsubscribe freebsd-current in the body of the message
Re: FBSD 5.0 diskless environment does not work!
In message: [EMAIL PROTECTED] Hartmann, O. [EMAIL PROTECTED] writes: : Can anyone help? Has someone a runnng diskless FBSD 5.0-R/5-CURRENT : environment? I fixed a couple of bugs in the /etc/rc.d files that broke diskless boots about a month or two so after 5.0-RELEASE. It would have precluded diskless systems network from working most of the time. Warner To Unsubscribe: send mail to [EMAIL PROTECTED] with unsubscribe freebsd-current in the body of the message
FBSD 5.0 diskless environment does not work!
Hello. For a long time with FreeBSD 4.X we ran and still run a bunch of diskless stations, X11 Terminals, some special workstations and now a growing system with several nodes for usage as PVM. Now I want to switch to FreeBSD 5.0 and run into massive problems. I searched for similar problems in the archive and groups.google, but I did not find any appropriate. Hope someone can help. The environment is really simple. All diskless machines uses PXE to bootstrap. In FreeBSD 4.7 this environment works perfect (and ist is still working), but in FreeBSD 5.0 I get into trouble. I want first to describe the environment. Each class of machines has its own full FreeBSD-4/5 installation. Within ist root-path, there is a populated /etc and a populated /conf. /conf has three main entries: /conf/base, /conf/default and /conf/IP.SPECIFIC.ENTRY. As I understand the diskless boot process in FreeBSD 4.X, at first a memory disc is created, then the stuff in /etc/ is copied to that RAM disc, executed and then the process does a kind of hierarchy overlay by unifying /etc with /conf/base with /conf/default and at last with /conf/IP.SPECIFIC.ENTRY and executes then the on top remaining config files (sorry about my English). That means: the most specific system configs are in /conf/IP.SPECIFIC.ENTRY and get executed. This seems to have changed in 5.0 as there is only the /conf/default and /conf/IP.SPECIFIC.ENTRY left. /conf/base has gone. But nevertheless, this should work, but does not! When booting a FBSD 5.0 client without a populated /etc directory but with a well populated /conf/defaults/etc directory, it fails. It also fails when populating /etc and delete those files which are in /conf/default/etc. Putting all the stuff in /etc forces the client to do things it should do, but this is not the way it should be and its not the way as it is decribed (rarely) in the docs I found and as I expect this in FreeBSD 4.7 (where this scheme works like a charme). This seems to be the first problem I can not figure out. What is the hierarchy of /etc and /conf/defaults/etc in FBSD 5.0 now? How can I keep my config files in sync without copying all the stuff again from /conf/default/etc to /etc (which breaks the config scheme, I think)? When the kernel has been bootstraped and started , the first thing is it moans about a non writeable root partition. This also does not change when exporting the root partition as read/writeable on the server (doing a killall -HUP mountd ...). I also use the rc_ng=YES-scheme. The kernel has options NFS_ROOT and MD_ROOT and also options md for the memory disc compiled in. The only problem is that in the clients /dev directory is no appropriate md0* node, I can not create it via MAKEDEV all in this directory and I wonder whether this is needed. Can anyone help? Has someone a runnng diskless FBSD 5.0-R/5-CURRENT environment? Thanks in advance, Oliver -- MfG O. Hartmann [EMAIL PROTECTED] -- Systemadministration des Institutes fuer Physik der Atmosphaere (IPA) -- Johannes Gutenberg Universitaet Mainz Becherweg 21 55099 Mainz Tel: +496131/3924662 (Maschinenraum) Tel: +496131/3924144 (Buero) FAX: +496131/3923532 To Unsubscribe: send mail to [EMAIL PROTECTED] with unsubscribe freebsd-current in the body of the message
