USS "mounts" the filesystems before the "kernel" actually starts, so everything is 
available from the beginning.  In Linux and most UNIX variants, there's a chicken-egg 
problem. To boot, you need the
kernel image, and some files from /etc, /sbin,  and /bin.  If /etc is a separate 
filesystem, there's no place for inittab to be (yet), so you can't determine how to 
start up.  Getting to /lib is
another issue.  It's perfectly possible in the Intel world to have your disk drivers 
as modules, so how do you read your disk if the device drivers are on it?

The usual solution is to create a "temporary" root filesystem, usually on a ramdisk, 
to get things going (using the BIOS disk access drivers), load required modules from 
that, then mount the "real"
root filesystem over it at the appropriate time.

The kernel lives in two places:  /boot for the image, and /lib/modules for loadable 
modules. /boot can be separate, since that's where you actually BOOT from.  During the 
initial startup, the kernel
mounts the root fs (defined in zipl.conf) read only.  Once that happens, /etc/inittab 
is read, /sbin/init kicks off (there's another directory that has to be in root), and 
everything else follows.
The rc scripts actually do the module loading, and at some point the root fs is 
remounted rw.  You can get some sense of this from following the startup messages.

In VERY OLD versions of UNIX, the kernel image lived in /, along with /bin, /etc, and 
/lib.  That was your basic startup set. (/sbin came later.)  I didn't start seeing a 
separate /boot directory till
Intel versions (Xenix, ix386, and SCO), although I think Solaris also has a separate 
/boot. That seemed to be mainly because older BIOS versions couldn't boot from a 
partition bigger than 1024
cylinders.  As the disks got bigger, this became more of a problem.

I'm on the fence about this whole issue.  I've done it both ways, and I don't see what 
having a separate /boot buys you as long as you can boot the root fs.  I like the IDEA 
of sharing the root fs
between systems R/O, but I haven't seen a method that doesn't require a lot of effort 
and surgery on the startup process, and the risk of getting component versions 
seriously mixed up.

What WOULD help is for someone to fix zipl so you can select the kernel you want via 
LOADPARM.

> -----Original Message-----
> From: Chet Norris [mailto:[EMAIL PROTECTED]]
> Sent: Thursday, February 20, 2003 3:09 PM
> To: [EMAIL PROTECTED]
> Subject: Re: [LINUX-390] R/O Linux guest?
>
>
> Mybe were talking RedHats to SuSe's or something.
>     I don't understand why /boot (the target of zipl) can be separated
> from the load point (/) file structure and /etc can't be. USS
> seemed to
> make some sense in that the things that made the image unique could be
> split away from the base code. ie. /etc, /var, /dev (and /tmp)
>     A kernel upgrade would modify /lib, /dev and /boot. To
> roll out the
> new kernel without applying it individually in each image I would have
> to copy a new "/" file structure and a new /boot file structure and
> then copy my original /etc back in over top of it. Right?
>
> --- "Post, Mark K" <[EMAIL PROTECTED]> wrote:
> > /lib is, and must be, part of the root file system (unless you're
> > willing to
> > play the games that I am not).  I'm not going to replace my entire
> > root file
> > system just to upgrade a kernel.  So, no problem.
> >
> > Mark Post
> >
> > -----Original Message-----
> > From: Hall, Ken (ECSS) [mailto:[EMAIL PROTECTED]]
> > Sent: Thursday, February 20, 2003 12:58 PM
> > To: [EMAIL PROTECTED]
> > Subject: Re: R/O Linux guest?
> >
> >
> > What about /lib (and particularly /lib/modules)?  You can't just
> > switch
> > kernels without having the corresponding modules available.
> >
> > Not to mention /var, and all of the RPM database stuff.
> >
> > Splitting off /boot seems to be mainly a relic from the days when
> > Linux
> > wouldn't boot if the root filesystem was bigger than 500mb. (or
> > whatever) on
> > Intel boxen.  I haven't seen a good reason to do it
> > for a long time.  If it's too big, it's a waste of space,
> and if it's
> > too
> > small, you can't keep the multiple kernels you want there.
> >
> > > -----Original Message-----
> > > From: Post, Mark K [mailto:[EMAIL PROTECTED]]
> > > Sent: Thursday, February 20, 2003 12:13 PM
> > > To: [EMAIL PROTECTED]
> > > Subject: Re: [LINUX-390] R/O Linux guest?
> > >
> > >
> > > Having /boot separate allows you to decide which volume you
> > > want to IPL
> > > from.  It also allows you to have multiple IPL volumes
> > > available.  I also
> > > have a /boot1, /boot2.4, etc.  /root is root's home directory
> > > and it forces
> > > me to be careful with how much junk I put there.  If it were
> > > part of /, then
> > > I could conceivably fill it up by being careless.
> > >
> > > In my particular case, /usr, /opt, are shared read-only with
> > > other systems.
> > >
> > > I'm not sure what you mean by a kernel upgrade forcing me to
> > replace
> > > multiple minidisks.  Most of the stuff that would need to be
> > > upgraded along
> > > with the kernel typically lives in /usr.
> > >
> > > Mark Post
> > >
> > > -----Original Message-----
> > > From: Chet Norris [mailto:[EMAIL PROTECTED]]
> > > Sent: Thursday, February 20, 2003 7:24 AM
> > > To: [EMAIL PROTECTED]
> > > Subject: Re: R/O Linux guest?
> > >
> > >
> > > Per the below (03/12/02) response, what devices are Read-Only and
> > > shared? It seems to me that only /usr and /usr/src could be. Then
> > why
> > > separate /root and /boot? I know you had a good reason, and I'm in
> > the
> > > process of re-mapping my file structures. Also, doesn't a kernel
> > > upgrade force you to roll out multiple minidisk replacements?
> > > Too bad we can't map it the same as USS with a separate /etc
> > > per image.
> > >
> > > From Archives Date: Tue, 12 Mar 2002 18:38:59 -0800
> > > Mark Post wrote:
> > > >/boot, /var and /tmp do _not_ have to be on the root file system.
> > > >Mine aren't.  Unless you play some games, /bin, /dev, /etc, /lib,
> > > >and /sbin have to be part of the root file system.  Anything else
> > > >can be easily put on a different one.
> > > >~ > df
> > > >Filesystem           1k-blocks      Used Available Use%
> Mounted on
> > > >/dev/dasdb1              52284     35868     13720  72% /
> > > >/dev/dasdc1            1062992    388560    620436  39% /tmp
> > > >/dev/dasdd1            1417324   1337424      7904  99% /usr
> > > >/dev/dasde1             111572     50520     55296  48% /var
> > > >/dev/dasdf1             104596     73036     26164  74% /opt
> > > >/dev/dasdg1              10432      1756      8140  18% /boot
> > > >/dev/dasdh1              52284      4936     44652  10% /root
> > > >/dev/dasdi1              24384     12912     10216  56% /home
> > > >/dev/dasdj1             921228    773876    100556  89% /usr/src
> > >
> > > Mark Post
> > >
> > >
> > >
> > > =====
> > > Chet Norris
> > > Marriott International,Inc.
> > >
> > > __________________________________________________
> > > Do you Yahoo!?
> > > Yahoo! Mail Plus - Powerful. Affordable. Sign up now.
> > > http://mailplus.yahoo.com
> > >
>
>
> =====
> Chet Norris
> Marriott International,Inc.
>
> __________________________________________________
> Do you Yahoo!?
> Yahoo! Mail Plus - Powerful. Affordable. Sign up now.
> http://mailplus.yahoo.com
>

Reply via email to