Date: Sat, 14 Mar 1998 09:45:14 -0500 (EST) From: Robert Watson <robert@cyrus.watson.org> To: Mike Smith <mike@smith.net.au> Cc: Doug White <dwhite@resnet.uoregon.edu>, "Jordan K. Hubbard" <jkh@time.cdrom.com>, Aleksey Zvyagin <zal@rest.ru>, freebsd-stable@FreeBSD.ORG, freebsd-questions@FreeBSD.ORG Subject: Re: Help! Upgrade 2.2.5-RELEASE to 2.2-STABLE. Message-ID: <Pine.BSF.3.96.980314093642.27517C-100000@trojanhorse.pr.watson.org> In-Reply-To: <199803140337.TAA03072@dingo.cdrom.com>
next in thread | previous in thread | raw e-mail | index | archive | help
On Fri, 13 Mar 1998, Mike Smith wrote: > > > P.S. As has been covered in *great* detail this last week, and as is > > > also available via the mailing list archives, edit /dev/wd0a to be the > > > fully-slice-qualified name, like the other entries. e.g. /dev/wd0s1a. > > > > I think I'll have to join Paul in questioning whether this breakage is > > going to make 2.2.6 a support nightmare. I don't want this to become > > another ``unknown login class root''-style hailstorm. :-( > > If you update /sbin/mount at the same time, it's a no-op. I got bitten > by the two different "dedicated" disk types issue though; that hurt. > The fix is still being tested in -current, but I haven't heard any > complaints about it, so I was planning to bring it over tonight. > > I think it was my turn to be bitten by a "this is simple and innocouous > and will reduce world ugliness" change. 8( In fact, there are situations where the upgrade of /sbin/mount and the kernel at the same time on a "sliced" machine *WILL NOT WORK*. I just got bitten by this about 4 hours ago -- or rather, 30 minutes ago. My main server machines reboot at a nasty time on a Saturday morning (nasty EST, that is :). I upgraded all of them to a recent (Mar 9) -STABLE kernel and mount. The main server machine did not come up. Reason? Because the slice arrangement was like this, left over from the days before >1024 cylinder boot: /dev/wd0s1a on / /dev/wd0s2e on /usr /dev/wd0s2f on /homea /dev/wd2s1a on /homeb /dev/sd0a on /homec /dev/sd1s1e on /home /dev/wd0s2h on /usr/var/mail That is, there are two slices on wd0, the boot device. The a partition (wd0a) completely fills wd0s1. The other three partitions on wd0 are in wd0s2. So the unfortunate news is that /dev/wd0s1a DID NOT EXIST. The 2.0.5 install floppy that originally built the machine never created the slice entries in /dev, and therefore the remount from root_device to /dev/wd0s1a failed. Talk about sucky. The root_device is read-only by definition, and now one is at a single-user shell but cannot write to the device to create a device node, and cannot remount to make it writable. :) While this is definitely fixable, it is fairly sucky. :) So really the text in the FAQ needs also to do this: Step 2: Verify that all the compatibility slice entries in your /etc/fstab also have standard /dev slice entries. If necessary, create them using MAKEDEV *before* rebooting or modifying fstab (do not caught with your pants down if there is power loss). And maybe references to how to use MAKEDEV. All of our -stable test machines came up fine after the change, so I went ahead and rolled it onto our main servers. Big mistake. :) Would really have sucked if I wasn't local -- which I am not 3 months of the year. Sorry to whine, but it was a generally unpleasent experience. I should have checked more carefully, but to be honest, I probably never thought to create the slice /dev entries later because they were never used. I have upgraded the machine from 2.0.5 up through 2.2.2 before switching to -STABLE and never had a problem. :) Robert N Watson Carnegie Mellon University http://www.cmu.edu/ SafePort Network Services http://www.safeport.com/ robert@fledge.watson.org http://www.watson.org/~robert/ To Unsubscribe: send mail to majordomo@FreeBSD.org with "unsubscribe freebsd-questions" in the body of the message
Want to link to this message? Use this URL: <https://mail-archive.FreeBSD.org/cgi/mid.cgi?Pine.BSF.3.96.980314093642.27517C-100000>