Date: Fri, 18 Jul 2003 11:20:30 -0400 (EDT) From: Matt Loschert <loschert@servint.com> To: Valentin Nechayev <netch@lucky.net> Cc: hackers@freebsd.org Subject: Re: complicated downgrade Message-ID: <20030718111819.V30205@schnell.net> In-Reply-To: <20030718091248.GO76126@lucky.net> References: <20030718091248.GO76126@lucky.net>
next in thread | previous in thread | raw e-mail | index | archive | help
On Fri, 18 Jul 2003, Valentin Nechayev wrote:
> (Cc'ed to phk@ as to main GEOM and DEVFS developer; see corresponding
> questions below.)
>
> Hi,
>
> I need to downgrade a remote FreeBSD system from 5.1-release to 4.8-release
> remotely without any local help (except possible hitting Reset).
> Don't ask why the collocation provider is too ugly and too far from me; it's
> given and unchangeable. This system never was 4.* (began from 5.0-DP2).
>
> I suppose the following algorith to have chance to succeed.
> If anybody can check it and fix, please help.
>
> 1. Download 4.8-release and untar it to the separate tree in file system.
>
> 2. Remove all unneeded for this system (including modules, fore_dlnd,
> mount_portalfs and other unused programs) from directories placed
> on root fs; this is due to its small size (~64M).
>
> 3. Place directories for 4.8 as /bin4, /sbin4, /etc4, /boot4, /usr/lib4, etc.
>
> 4. Place kernel for this system as /kernel; add /kernel.GENERIC also.
>
> 5. Run MAKEDEV in /dev4 for all standard entries and entries for specific
> disks (/dev/ar0s1a, etc.)
>
> 6. Add minimal site-specific contents to /etc4, enough to boot and run sshd
> to allow admin to enter. It includes master.passwd (with admin entry),
> /etc/ssh/* (copied keys and configs), /etc/resolv.conf, /etc/rc.conf,
> /etc/fstab, /etc/group (allowing su); what else?
>
> 7. (The most horrible step.)
>
> Goal is to have /dev on disk (not DEVFS!) appropriate for booting 4.8.
> Two alternatives are possible:
>
> 7a.1. Patch kernel to disable GEOM's protection against writing to
> devices which is mounted or have mounted subparts.
> (Here is most foggy place: I don't know how to do it in such way
> as to disable rereading of disk structure on this place.)
> 7a.2. Reboot with patched kernel.
> 7a.3. Use a binary editor and renames root directory entries directly:
> /dev4 to /dev, /dev to /dev5. (If entry name length matters,
> use, e.g., de4 instead of dev4.)
> 7a.4. Immediately reboot as to prevent namei subsystem damage.
>
> Result of all this is that devfs is mounted not over empty directory
> (or only with standard entries, without local specific disks), but over
> full working one.
>
> Alternative variant for step 7:
>
> 7b. add code to /sbin/init before mounting devfs: remount root to rw,
> copy entries to it (/bin/cp -Rp). It will work if mount() allows
> remounting to rw without correct /dev entry (this may have problems,
> as I saw on 4.8). Reboot to run this code from init.
>
> 8. Disable all processes except sshd and run the following (saying generally):
>
> for D in /bin /sbin /etc /boot /usr/bin /usr/sbin /usr/lib /usr/libexec \
> /usr/libdata /usr/share /usr/local /var/db
> do
> mv ${D} ${D}5
> mv ${D}4 {D}
> done
>
> 9. Reboot and hope 4.8 will load and run.
>
> I attached some local outputs to check for unexpectedness.
>
>
> -netch-
This is probably obvious, but whatever you do, I would suggest building a
crash box to locally test the procedure on beforehand. I made the mistake
of not perfectly testing a remote upgrade once upon a time, and will never
do it again.
- Matt
--
Matt Loschert - Software Engineer | email: loschert@servint.com |
ServInt Internet Services | web: http://www.servint.net/ |
McLean, Virginia USA | phone: (703) 847-1381 |
Want to link to this message? Use this URL: <https://mail-archive.FreeBSD.org/cgi/mid.cgi?20030718111819.V30205>