David O'Brien wrote:
> > 1)    Soft Updates enabled on a root partition.
> >
> >       This comes back to the old "you can't turn SU
> >       on or off, except via tunefs".  So even if you
> >       boot via CDROM, it's too late, if the CDROM
> >       kernel supports SU, since it's already on, and
> >       you can't substitute an async mount.
> 
> Why not?  The install kernel does not have SOFTUPDATES enabled.
> Can one not mount a parition aync if the softupdates flag was set via
> tunefs in the past [and the running kernel does not support them]?

No.  The tunefs flag takes precedence.  Look at the FFS
code.  I've submitted patches once in the past, which basically
converted the "update" of mounts into an unmount+remount,
but that was shot down as a "bad idea".

The problem is that you have to wait for the clock to drain,
since the device you use might be something like a vn, also
on a soft updates partition.  It gets messy.

The initial mount, even if taken care of, will not
get me where I need to be, as upgrades are more frequent
than initial installs.  I expect that the rest of the
FreeBSD community has this experience as well (this
bodes ill for the growth of FreeBSD adoption, but that
is a -advocacy topic).

> > 2)    Most of my upgrades are over a network, not
> >       booting off of CDROM.
> 
> How are you running sysinstall?  I do not know what you
> mean by "over the network".  To me that means you have
> made the two boot floppy set and used that -- the "network"
> part is where you are getting your distribution sets from.
> In this case, you *are* running sysinstal as init.  Perhaps
> I do not fully understand how you are doing this.

I have no floppy.  I have no CDROM.  I have no video
card or keyboard, so I can't invoke the Intel FXP BIOS
etherboot code.  I have only a serial console, a network
connection, and a hard disk.

--------------------------------------------------------
Terry's magic embedded system upgrade procedure, Mark I
To be run at the console, logged in as root ...
--------------------------------------------------------
echo "This crap should be accessible on the CD, but isn't"
ftp machine.installed.from.cdrom
        cd /stand
        lcd /tmp
        get sysinstall
        cd /dev
        get MAKEDEV
        get MAKEDEV.local
        quit
mount nfs.server.with.cdrom:/cdrom /mnt
cd /tmp
echo "Must use a fricking matching sysinstall..."
chmod 755 sysinstall
./sysinstall
        upgrade
        local disk = /mnt
        force the S.O.B. to install /src...
        exit
echo "non-boot sysinstall 'can't be bothered' with this..."
cp MAKEDEV /dev
cp MAKEDEV.local /dev
cp sysinstall /stand/sysinstall
cd /dev
sh MAKEDEV all
echo "What were they thinking when they changes this?!?"
cd /etc
cat >> pam.conf

# this is crap for ssh because ssh won't work otherwise
sshd ...
sshd ...
sshd ...
^D
sync
sync
sync
reboot
--------------------------------------------------------


Make sense now?  It even works for "downgrading" your
-CURRENT system to -STABLE, if you want to not crash
all the time, and realize the errors of your ways in
having installed -CURRENT...


> > 3)    The default in 4.3-RELEASE is to have the IDE
> >       write caching off.
> 
> If you submit a patch to add the proper entries to
> /boot/loader.conf in the MFSROOT image, I'd commit it.
> This would ensure the installation process always runs
> with write cashing on.

I wasn't sure that this was processed correctly from
the MFS image.  I will look at doing a patch; but it
won't help my own situation, since I don't really boot
the thing in order to upgrade.

-- Terry

To Unsubscribe: send mail to [EMAIL PROTECTED]
with "unsubscribe freebsd-hackers" in the body of the message

Reply via email to