[6947] in testers
Re: prevert.mit.edu: failed Linux 9.4 update
daemon@ATHENA.MIT.EDU (Greg Hudson)
Thu May 5 13:04:00 2005
From: Greg Hudson <ghudson@MIT.EDU>
To: "andrew m. boardman" <amb@mit.edu>
Cc: testers@mit.edu
In-Reply-To: <200505042306.j44N628k019336@pothole.mit.edu>
Content-Type: text/plain
Content-Transfer-Encoding: 7bit
Date: Thu, 05 May 2005 13:03:48 -0400
Message-Id: <1115312628.3774.15.camel@egyptian-gods.mit.edu>
Mime-Version: 1.0
On Wed, 2005-05-04 at 19:06 -0400, andrew m. boardman wrote:
> >I don't know why anaconda upgrades wouldn't run into the same problem,
> >partly because I don't know what's cleaning out /dev.
>
> The postinstall script for udev starts it as soon as it's installed,
> preumably being mounted over the the original device directory. It's
> tolerant of having devices manually created in it, though, which leads to
> some possibilities for temporary workarounds during the install.
I don't think udev mounts anything over the /dev directory
(although /dev/pts appears to be a mountpoint). I think the hotplug
system invokes udev, which grovels over a nest of config files and
information under /sys (which appears to be a kernel filesystem even
though it's not listed in the mount table) and creates stuff under /dev
to match the device.
I think all the devices are disappearing from /dev during the update
simply because the "dev" package is being removed. (During an anaconda
update, the miniroot's /dev is bind-mounted into /mnt/sysimage/dev; if
the miniroot is mounted read-only, then that would prevent devices from
going away. I don't know if the miniroot is actually read-only,
though.)
However, something must also be cleaning out /dev at boot time, since
the /dev/mouse symlink is present after an update but absent after the
ensuing reboot.
