RE: "/etc/init.d/boot" relies on "/usr/." being mounted...
-----BEGIN PGP SIGNED MESSAGE-----
>>>>> "Michael" == Michael Meskes <meskes@topsystem.de> writes:
>-----Original Message-----
>From: Karl M. Hegbloom [SMTP:karlheg@inetarena.com]
>Sent: Tuesday, April 22, 1997 6:35 AM
>To: debian-devel@lists.debian.org
>Cc: Die Adresse des Empfängers ist unbekannt.
>Subject: RE: "/etc/init.d/boot" relies on "/usr/." being mounted...
>
>>>>>> "Michael" == Michael Meskes <meskes@topsystem.de> writes:
>
> Michael> Is there a scenario where /usr is not mounted but /proc
> Michael> is? /etc/init.d/kerneld does check whether /proc is
> Michael> mounted.
>
> If the transnames patch has been applied to your kernel, then yes:
Since "/proc" must be mounted to make "/proc/nametrans" visible.
Michael> Eh, what is the transnames patch?
It makes it so you can mount the root filesystem of a server as the
root filesystem of a peer machine on a LAN. They share the same
filesystem, except for certain machine specific configuration files
and/or directories, which are distinguished by a suffix attached to
the filename.
Essentially, the patch hooks into the virtual filesystem layer, [1]
and does a name translation on filenames with a suffix that specifies
which machine should 'see' the file, based on kernel type, hostname,
machine architecture, or OS type. In this way, several machines can
share the same set of harddrives, and most of the configuration files.
They are close twins of one another; almost exact clones.
You can find it at: (binocular for 'trans')
http://www.linuxhq.com/upatch20.html
There is a README in the patch that explains it fairly well. [2]
Each machine has its own "/etc/fstab", "/tmp", and sets of
"/etc/cron.*" directories, amoung others. For example, I have several
/tmp directories. One is "/tmp#host=cherryflower#" and the other is
"/tmp#ktype=server#". As a failsafe, I've done `ln -s
"/tmp#ktype=server#" /tmp` on the server, so that it will run even
without the transnaming patch applied. With transnames, though, that
symlink is seen as actually being "/tmp", and access to it gets you
"/tmp#ktype=server#". (That's part of the patch.)
The second machine, connected via ethernet, I boot from a floppy disk
that has its kernel configured to use the `bootp` protocol. The
"/etc/bootptab" is set up so that its root filesystem is "/", mounted
with nfs. It mounts the '/' of the server, and because of transnames,
it sees the machine specific files meant for it, as well as all of the
other files in the system.
On both machines, which share the harddrives of the server, an `ls`
will turn up all three of the "/tmp" directories, listing:
drwxrwxrwt 8 root adm 2048 Apr 22 09:45 /tmp
drwxrwxrwt 2 root adm 1024 Nov 28 05:40 /tmp#host=cherryflower#
drwxrwxrwt 4 root adm 1024 Feb 3 22:22 /tmp#ktype=server#
From bittersweet, `ls /tmp` shows the contents of the server one,
while `ls /tmp` from cherryflower shows its "/tmp" directory, since
the kernel translates the filename. From either machine, given that
access permissions will allow it, one can access either of the two
/tmp directories by listing it explicitly on the commandline, using
quotation marks so the shell doesn't see the hash character as a
comment mark.
I have not tried it yet, but I believe that it will be possible to
boot a Lose95 machine that sits on a LAN using a floppy, and be able
to mount the "/" of a properly configured server, thus making a
temporary Linux workstation, complete with X-Server. The
"/etc/fstab#host=tempo#" would not need to contain entries for that
box's local hard drives, only for the nfs mounted "/" directory, and
perhaps a local CD-W/ROM...
I think that on a fast LAN, it will be quite useable... I've very
little experience with ethernet at this point, so I cannot really say
what it will be like. There are, no doubt, others on this planet who
have actually done empirical studies of the applications of the
transnames patch. :-) [3]
Footnotes:
[1] It becomes a make-time configuration option, so you can compile
either with or without transnameing enabled.
[2] Better than I can in a few short paragraphs within a few short
minutes. I have not studied this in depth; I got the patch on the
net, and haven't played with it a whole lot yet. I have much to do.
This is my first year at this. I believe this work was done in
Stuttgart at the college there.
[3] I would be interested in hearing about that! Soon I will look
for research reports on it...
- --
Karl M. Hegbloom <karlheg@inetarena.com>
http://www.inetarena.com/~karlheg
Portland, OR USA
Debian GNU 1.2 Linux 2.0.30t
Have you seen the Emperor's new red hat?
-----BEGIN PGP SIGNATURE-----
Version: 2.6.3
Charset: noconv
Comment: Processed by Mailcrypt 3.4, an Emacs/PGP interface
iQB1AwUBM1z90vzDQLi9TTotAQFnhgMAlq5pNLDJmH6E2xTC5uGuBLmw0m5rMIBY
U812XPUx0DGmMLzxBiWfh/hlDqfbREHEmncU4PelcvDBSBUKVXvWSFCu7/Xw2DKS
vDq+m7w0XEtqU7OfdPyRldLB0JUxu9DS
=M63H
-----END PGP SIGNATURE-----
--
TO UNSUBSCRIBE FROM THIS MAILING LIST: e-mail the word "unsubscribe" to
debian-devel-request@lists.debian.org .
Trouble? e-mail to templin@bucknell.edu .
Reply to: