Re: devices missing?
This goes mainly to Thomas, but I CC it to the list for archive
purposes:
On Tue, Jan 04, 2005 at 02:56:36PM +0100, Steffen Grunewald wrote:
> > > and mount /tmp/target/proc before/after mountdisks - why doesn't
> > > mountdisks do that although there's a /proc entry in /tmp/fai/fstab ?)
> > Currently /tmp/target/proc is mounted in the task updatebase. But I
> > will move it to task extrbase which makes more sense.
> Why don't you modify mount2dir instead? (second case: /dev/* ->
> /dev/*|none)
With umount /proc deactivated in subroutines-linux, the problem will not
be fixed completely - it's also necessary to stop and start the udev
init script somewhere in between.
Watching the output of "watch -d -n5 mount" I found that udev comes into
action about 5 minutes into my install (just in the middle of instsoft),
so the best place to do this would be a hook to configure. I've been
typing a fix when my file server stopped responding ...
I started a fresh installation on a "virgin" disk - which failed (sfdisk
doesn't like this). Had to fdisk, "w" and exit... and of course one
should never add lines below the "exit 0" line of a script.
Still there are lots of error messages from software that tries to
access /dev/{,u}random and the like (before the hook is executed)...
so that last word isn't spoken yet.
udev 0.050-3 has been uploaded to unstable on Jan 1, perhaps it will
cure the problems? Otherwise I'll file a bug against it... (I'll try and
get the unstable package to local storage tomorrow)
Cheers,
Steffen
Re: devices missing?
On Tue, Jan 04, 2005 at 02:42:31PM +0100, Thomas Lange wrote:
> > On Tue, 4 Jan 2005 14:11:24 +0100, Steffen Grunewald <[EMAIL
> > PROTECTED]> said:
>
> >> > /usr/sbin/mkinitrd: device /dev/hda1 is not a block device
> >> So everything boils down how to avoid the /tmp/target/.dev tree being
> >> mounted... who knows how?
>
> > and mount /tmp/target/proc before/after mountdisks - why doesn't
> > mountdisks do that although there's a /proc entry in /tmp/fai/fstab ?)
> Currently /tmp/target/proc is mounted in the task updatebase. But I
> will move it to task extrbase which makes more sense.
Why don't you modify mount2dir instead? (second case: /dev/* ->
/dev/*|none)
That's how I do it now, and it seems to work.
/tmp/target/proc seems never to be mounted on my machine, or at least it
gets unmounted before faiend ... where I have the chance to look at the
system (fai-chboot -FIv name for now)
This is for version 1.3 of /usr/lib/fai/mount2dir. I'm a bit late again.
BTW, when you start to modify mount2dir: in the sysinfo case, there
might be a $mountpoint of "swap" instead of "none" which also should be
skipped, else one gets error messages while trying to mount
/tmp/targetswap ...
> From the udev startup script:
>
> # if you don't like this, remove /.dev/.
> [ -d /.dev ] && mount -n --bind /dev /.dev
>
> So maybe a hook which removes /.dev will help. But which package is
> creating tmp/target/.dev ?
the udevd daemon.
Since hooks can only be placed in front of a task, it's kind of
futile...
Steffen
--
Steffen Grunewald * * * Merlin cluster admin (http://pandora.aei.mpg.de)
Albert-Einstein-Institut (MPI Gravitationsphysik, http://www.aei.mpg.de)
Science Park Golm, Am Mühlenberg 1, 14476 Potsdam, Germany
e-mail: steffen.grunewald(*)aei.mpg.de * +49-331-567-{fon:7233,fax:7298}
Re: devices missing?
> On Tue, 4 Jan 2005 14:11:24 +0100, Steffen Grunewald <[EMAIL PROTECTED]> > said: >> > /usr/sbin/mkinitrd: device /dev/hda1 is not a block device >> So everything boils down how to avoid the /tmp/target/.dev tree being >> mounted... who knows how? > and mount /tmp/target/proc before/after mountdisks - why doesn't > mountdisks do that although there's a /proc entry in /tmp/fai/fstab ?) Currently /tmp/target/proc is mounted in the task updatebase. But I will move it to task extrbase which makes more sense. >From the udev startup script: # if you don't like this, remove /.dev/. [ -d /.dev ] && mount -n --bind /dev /.dev So maybe a hook which removes /.dev will help. But which package is creating tmp/target/.dev ? -- regards Thomas
Re: devices missing?
Hi,
talking to myself again...
> > /usr/sbin/mkinitrd: device /dev/hda1 is not a block device
> So everything boils down how to avoid the /tmp/target/.dev tree being
> mounted... who knows how?
Tracked this down (getting this to run is not my major task, so I only
manage to devote a few minutes per attempt to it). The package
responsible is udev, which is in the dependencies for hal, which is
required by gnome-volume-manager, which is included by gnome.
udev underwent two changes in the corresponding time range (Dec 21:
fine, now: :-(((), and in fact the previous installation used version
0.048-2, now I have 0.050-2. In the Debian changelog, there are some
indications that something important-for-FAI has changed.
Trying to restart /etc/init.d/udev in the chrooted environment would not
help unless /proc has been mounted before (this seems to be the
difference). In a "normal" context this would of course be the case; but
in the FAI case I cannot see where to mount /tmp/target/proc ...
(I can't do this using chroot mount /proc in a hook just after mountdisks
since there is no chroot environment -namely sh- avaliable yet. I could try
and mount /tmp/target/proc before/after mountdisks - why doesn't
mountdisks do that although there's a /proc entry in /tmp/fai/fstab ?)
Still stuck, even tough I now know why I don't know how to get out.
S
--
Steffen Grunewald * * * Merlin cluster admin (http://pandora.aei.mpg.de)
Albert-Einstein-Institut (MPI Gravitationsphysik, http://www.aei.mpg.de)
Science Park Golm, Am Mühlenberg 1, 14476 Potsdam, Germany
e-mail: steffen.grunewald(*)aei.mpg.de * +49-331-567-{fon:7233,fax:7298}
Re: devices missing?
On Tue, Jan 04, 2005 at 11:42:45AM +0100, Steffen Grunewald wrote:
> Hi,
>
> did a fresh installation (with unchanged configuration, just to update
> the set of packages on the machine), and ended up with an unbootable
> system. (The last successful attempt dates back to 20041221; I'm using
> sarge.)
>
> Looking through the log files, and trying to re-do the failed kernel
> installation step, I found that the message
>
> /usr/sbin/mkinitrd: device /dev/hda1 is not a block device
>
> has been caused by an almost empty /dev/ directory below /tmp/target:
>
> xxx:/dev# ls
> MAKEDEV fd loop net nvram pts shm stderr stdout
> core initctl mouse null pppscsi sndstat stdin xconsole
Since a pure MAKEDEV also didn't work from the command line, I digged
somewhat deeper and found that the /.dev directory would be used if it
is mounted. (This seems to be a 2.6 feature which I don't know how to
handle yet, so any hints are appreciated.)
Unmounting /.dev would make the devices in the correct place.
So everything boils down how to avoid the /tmp/target/.dev tree being
mounted... who knows how?
(Note that I didn't change anything. The vmlinuz-install is dated Sep
11, and there have been no significant changes to neither the /fai tree
nor the setup of the server while I was on vacation. So it must be a new
feature of one or more upgraded packages...)
CHeers,
Steffen
--
Steffen Grunewald * * * Merlin cluster admin (http://pandora.aei.mpg.de)
Albert-Einstein-Institut (MPI Gravitationsphysik, http://www.aei.mpg.de)
Science Park Golm, Am Mühlenberg 1, 14476 Potsdam, Germany
e-mail: steffen.grunewald(*)aei.mpg.de * +49-331-567-{fon:7233,fax:7298}
devices missing?
Hi,
did a fresh installation (with unchanged configuration, just to update
the set of packages on the machine), and ended up with an unbootable
system. (The last successful attempt dates back to 20041221; I'm using
sarge.)
Looking through the log files, and trying to re-do the failed kernel
installation step, I found that the message
/usr/sbin/mkinitrd: device /dev/hda1 is not a block device
has been caused by an almost empty /dev/ directory below /tmp/target:
xxx:/dev# ls
MAKEDEV fd loop net nvram pts shm stderr stdout
core initctl mouse null pppscsi sndstat stdin xconsole
Obviously, there have been some changes somewhere. makedev (now 2.3.1-75)
has a timestamp dating back to August, so it's not to be blamed in the
first place... Any ideas? (Of course I can run MAKEDEV with the
appropriate options, but this should have been done by some basic
installation step before.)
I also get a lot of error messages from SSL related packages (apache,
ssh...) complaining about the PRNG. This might be bugs or features,
something I have to find out...
Cheers,
Steffen
--
Steffen Grunewald * * * Merlin cluster admin (http://pandora.aei.mpg.de)
Albert-Einstein-Institut (MPI Gravitationsphysik, http://www.aei.mpg.de)
Science Park Golm, Am Mühlenberg 1, 14476 Potsdam, Germany
e-mail: steffen.grunewald(*)aei.mpg.de * +49-331-567-{fon:7233,fax:7298}
