2013/9/18 Anders Ingemann <[email protected]>

> On 18 September 2013 12:28, olivier sallou <[email protected]>
> wrote:
> >
> >
> >
> > 2013/9/18 Anders Ingemann <[email protected]>
> >>
> >> On 18 September 2013 08:31, olivier sallou <[email protected]>
> >> wrote:
> >> >
> >> >
> >> >
> >> > 2013/9/17 Anders Ingemann <[email protected]>
> >> >>
> >> >> On 17 September 2013 15:46, olivier sallou <[email protected]
> >
> >> >> wrote:
> >> >> >
> >> >> >
> >> >> >
> >> >> > 2013/9/17 Anders Ingemann <[email protected]>
> >> >> >>
> >> >> >> > The virtualbox provider in python branch worked fine at booting
> >> >> >> > the
> >> >> >> > first bootable disk with grub installed via the loopback.
> >> >> >> Honestly, I was never able to get it to work. I think it failed
> >> >> >> exactly because it was a loopback device. Can you send a manifest
> >> >> >> that
> >> >> >> reproduces your setup? Maybe I can work my way up from there.
> >> >> >> > did you use grub2 on host machine ? grub1 fails to install over
> >> >> >> > loopback
> >> >> >> > device and lead to boot error (can't find grub).
> >> >> >> Exactly, which is why I am using dmsetup to fake a real hdd, so
> grub
> >> >> >> installs without a hitch :-).
> >> >> >
> >> >> >
> >> >> > I am using grub2 on my computer, it manages correctly the install
> >> >> > over
> >> >> > loopback.
> >> >> > Being able to use grub1 would be perfect, but I do not know how to
> >> >> > make
> >> >> > it
> >> >> > worked. I tried many things to get grub working, but it only worked
> >> >> > when
> >> >> > I
> >> >> > used grub2   :-(
> >> >> >>
> >> >> >>
> >> >> >>
> >> >> >>
> >> >> >> > did you look also at your generated device.map and grub.cfg
> files
> >> >> >> > ?
> >> >> >> Yes, everything seems fine, I'll have another look though (the
> >> >> >> config
> >> >> >> is in the pastebin I provided)
> >> >> >>
> >> >> >> > I know that while trying to setup grub stuff etc... for kvm and
> >> >> >> > virtualbox I had issues with auto-generated grub.cfg using
> >> >> >> > loopback
> >> >> >> > devices
> >> >> >> > instead of using disk devices (see loopback keywork in
> grub.cfg).
> >> >> >> Yes, I circumvent that by making the fake hdd, although maybe
> there
> >> >> >> is
> >> >> >> some leftover in the grub.cfg... I'll check it out.
> >> >> >>
> >> >> >> > I also used disk device (/dev/sda for example) instead of disk
> >> >> >> > uuids
> >> >> >> > in
> >> >> >> > boot menu setup of grub (disk uuids may not be the same when
> >> >> >> > booting
> >> >> >> > on your
> >> >> >> > final host).
> >> >> >> Actually. They will be exactly the same, I even have problems when
> >> >> >> attaching two disks created by the same snapshot because VBox
> >> >> >> doesn't
> >> >> >> like duplicate UUIDs. At the very least the partition UUIDs will
> be
> >> >> >> the same (which is what I use in fstab).
> >> >> >>
> >> >> >> Anders
> >> >> >>
> >> >> >>
> >> >> >> On 17 September 2013 08:13, olivier sallou
> >> >> >> <[email protected]>
> >> >> >> wrote:
> >> >> >> >
> >> >> >> >
> >> >> >> >
> >> >> >> > 2013/9/16 Anders Ingemann <[email protected]>
> >> >> >> >>
> >> >> >> >> Hey guys
> >> >> >> >>
> >> >> >> >> I posted to [email protected], asking for help to figure out
> why
> >> >> >> >> my
> >> >> >> >> VirtualBox image is stuck at "GRUB loading.". I thought that it
> >> >> >> >> might
> >> >> >> >> be more relevant over there, but maybe you guys have some ideas
> >> >> >> >> as
> >> >> >> >> well.
> >> >> >> >
> >> >> >> > did you look also at your generated device.map and grub.cfg
> files
> >> >> >> > ?
> >> >> >> >
> >> >> >> > I know that while trying to setup grub stuff etc... for kvm and
> >> >> >> > virtualbox I
> >> >> >> > had issues with auto-generated grub.cfg using loopback devices
> >> >> >> > instead
> >> >> >> > of
> >> >> >> > using disk devices (see loopback keywork in grub.cfg).
> >> >> >> >
> >> >> >> > I also used disk device (/dev/sda for example) instead of disk
> >> >> >> > uuids
> >> >> >> > in
> >> >> >> > boot
> >> >> >> > menu setup of grub (disk uuids may not be the same when booting
> on
> >> >> >> > your
> >> >> >> > final host).
> >> >> >> >
> >> >> >> > Olivier
> >> >> >> >
> >> >> >> >>
> >> >> >> >> Here's the link to the mailing list:
> >> >> >> >>
> http://lists.gnu.org/archive/html/help-grub/2013-09/threads.html
> >> >> >> >> (not
> >> >> >> >> a direct link, mailman seems to take its time indexing my post)
> >> >> >> >> ...and here's what I wrote
> >> >> >> >> ---------------------------
> >> >> >> >> Hello everybody
> >> >> >> >>
> >> >> >> >> I am developing the debian bootstrapper "build-debian-cloud"
> >> >> >> >> (https://github.com/andsens/build-debian-cloud) and am having
> >> >> >> >> some
> >> >> >> >> trouble getting grub to boot my debian installation.
> >> >> >> >> Since I bootstrap from a host system (with chroot etc.) I had
> to
> >> >> >> >> use
> >> >> >> >> some workarounds to get grub installed onto a loopback device
> >> >> >> >>
> >> >> >> >> (
> http://ebroder.net/2009/08/04/installing-grub-onto-a-disk-image/).
> >> >> >> >>
> >> >> >> >> The setup consists of a vdi image partitioned with GPT into
> boot,
> >> >> >> >> root, swap (in that order).
> >> >> >> >> I am unable to get past the "GRUB loading." message when
> booting
> >> >> >> >> the
> >> >> >> >> image in VirtualBox.
> >> >> >> >>
> >> >> >> >> I have a hard time figuring out what is wrong with the setup,
> >> >> >> >> especially because the scenario is a bit off the beaten path
> >> >> >> >> (mounting
> >> >> >> >> vdi as an network block device, bootstrapping, using dmsetup to
> >> >> >> >> fake
> >> >> >> >> a
> >> >> >> >> real hdd etc.). So the usual "just run BootRepair" or
> "reinstall
> >> >> >> >> grub"
> >> >> >> >> won't really help to create a stable bootstrapping process.
> >> >> >> >> I have run Ubuntus Boot Repair system to check if anything was
> >> >> >> >> wrong,
> >> >> >> >> but I can't seem to find anything. The output is here:
> >> >> >> >> http://paste.ubuntu.com/6115737/
> >> >> >> >>
> >> >> >> >> The setup can be fully reproduced by cloning my repo and
> running
> >> >> >> >> `./build-debian-cloud manifests/virtualbox.manifest.json` (only
> >> >> >> >> tested
> >> >> >> >> on debian wheezy).
> >> >> >> >> Simply create a new virtual machine in VBox, attach the
> resulting
> >> >> >> >> image and boot it.
> >> >> >> >>
> >> >> >> >> I would appreciate any help you can offer
> >> >> >> >> Anders Ingemann
> >> >> >> >> ---------------------------
> >> >> >> >>
> >> >> >> >>
> >> >> >> >> --
> >> >> >> >> To UNSUBSCRIBE, email to [email protected]
> >> >> >> >> with a subject of "unsubscribe". Trouble? Contact
> >> >> >> >> [email protected]
> >> >> >> >> Archive:
> >> >> >> >>
> >> >> >> >>
> >> >> >> >>
> >> >> >> >>
> http://lists.debian.org/camcogxfilv2+wq3fxe87wgv6ihs8etfqbejqb4vuof99onr...@mail.gmail.com
> >> >> >> >>
> >> >> >> >
> >> >> >> >
> >> >> >> >
> >> >> >> > --
> >> >> >> >
> >> >> >> > gpg key id: 4096R/326D8438  (keyring.debian.org)
> >> >> >> >
> >> >> >> > Key fingerprint = 5FB4 6F83 D3B9 5204 6335  D26D 78DC 68DB 326D
> >> >> >> > 8438
> >> >> >
> >> >> >
> >> >> >
> >> >> >
> >> >> > --
> >> >> >
> >> >> > gpg key id: 4096R/326D8438  (keyring.debian.org)
> >> >> >
> >> >> > Key fingerprint = 5FB4 6F83 D3B9 5204 6335  D26D 78DC 68DB 326D
> 8438
> >> >>
> >> >> OK, having done a bit of research, I feel like I need to clear some
> >> >> stuff up, so we avoid any further confusion. I *think* I am getting
> >> >> this right, please correct me if I am wrong.
> >> >>
> >> >> > I am using grub2 on my computer, it manages correctly the install
> >> >> > over
> >> >> > loopback.
> >> >> > Being able to use grub1 would be perfect, but I do not know how to
> >> >> > make
> >> >> > it worked. I tried many things to get grub working, but it only
> >> >> > worked when
> >> >> > I used grub2   :-(
> >> >> This is the main point that got me confused. The bootstrapper doesn't
> >> >> install grub1 on virtualbox and never has. It has always been grub2,
> >> >> debian switched to grub2 quite a while ago (since squeeze:
> >> >> https://wiki.debian.org/Grub). I am a bit puzzled as to how you
> >> >> managed to install the bootloader since grub2 gets all confused about
> >> >> loopback devices, which is why I haven't been able to install grub2
> >> >> with neither your virtualbox version nor mine (yet :-) ).
> >> >> Are you sure you didn't accidentally test the wrong image after you
> >> >> bootstrapped?
> >> >
> >> >
> >> > Sure, I am currently working on a vagrant provider, which is an extend
> >> > of
> >> > virtualbox provider and it works fine in virtualbox (at least for vbox
> >> > part,
> >> > not vagrant...).
> >> >
> >> > I had a quick look, and the fact is I get it worked because I use
> grub2
> >> > 2.0.x from sid. Previous versions (1.98, 1.99) did not manage
> correctly
> >> > indeed the loopback interface. The version > 2.0 manages it correctly.
> >> >
> >> > If you are on ubuntu or wheezy,... you get older grub2 release and it
> >> > does
> >> > not work...
> >> > v2.x should be in testing but it seems some bugs prevent migration to
> >> > testing.
> >> >>
> >> >>
> >> >> There is one exception to the use of grub1. We need it on EC2 because
> >> >> of the paravirtualized (PV) nature of the system. Here the grub
> >> >> bootloader is not actually installed to the volume, but rather
> >> >> launched from outside the box. That special version of grub then
> looks
> >> >> into the volume and parses the menu.lst file.
> >> >> menu.lst is the way grub1 figured out where the inital ram disk
> >> >> (initrd) was. The EC2 PV bootloader hasn't been updated yet and can
> >> >> therefore not parse the newer grub2 grub.cfg file.
> >> >> To that end we create a special grub2 configuration generation script
> >> >> which mimicks the old menu.lst layout. This config is then outputted
> >> >> by `update-grub` to /boot/grub/grub.cfg, which my bootloader symlinks
> >> >> to /boot/grub/menu.lst where PV-Grub (
> http://wiki.xen.org/wiki/PvGrub)
> >> >> can find it.
> >> >> Bonus info: This is what Charles Plessy has been working on with the
> >> >> "pv-grub-menu" package. Once that is working and packaged we don't
> the
> >> >> special grub2 config files any longer, because the tool generates
> that
> >> >> file.
> >> >>
> >> >> Now... on virtualbox _we don't need that workaround_. VirtualBox
> >> >> generates a full environment for grub to actually load. Meaning vbox
> >> >> couldn't care less about what the grub.cfg or menu.lst looks like.
> >> >>
> >> >> OK, so that was a bit of a tangent. I also want to report back what I
> >> >> have figured out so far.
> >> >>
> >> >> I am fairly certain that my grub.cfg is wrong, it has a `set
> >> >> root='/dev/mapper/vdb'`, which surely is not what grub sees at boot,
> >> >> mostly because '/dev/mapper' is only used by kpartx and lvm. Then
> >> >> again, what would root be set to when you boot from an lvm
> partitioned
> >> >> volume???
> >> >> When I change that line to `set root='/dev/sda1'` it still does not
> >> >> work. But at this point I figured out that my partitioning may be a
> >> >> bit wrong, the boot partition started at sector 0 instead of the
> usual
> >> >> 1MiB mark. Also, I used MB for partition sizes not MiB, meaning my
> >> >> 1024MB volume suddenly grew to 1074MB.
> >> >>
> >> >> There is another issue about grub not being able to figure out where
> >> >> the partitions are located on the volume. I am pretty sure however
> >> >> that this can be fixed with a proper device.map (btw. the device.map
> >> >> file is not read at boot, it is used by the grub tools to figure out
> >> >> how to make the grub.cfg file work).
> >> >>
> >> >> I believe grub should map hd0 as /dev/sda at boot, however even when
> I
> >> >> try to set root to (hd0,gpt1) it still does not work.
> >> >>
> >> >> Again, if anybody finds an error with what I have written, please
> >> >> chime in, I would appreciate it.
> >> >>
> >> >>
> >> >> Anders Ingemann
> >> >
> >> >
> >> >
> >> >
> >> > --
> >> >
> >> > gpg key id: 4096R/326D8438  (keyring.debian.org)
> >> >
> >> > Key fingerprint = 5FB4 6F83 D3B9 5204 6335  D26D 78DC 68DB 326D 8438
> >>
> >> > Sure, I am currently working on a vagrant provider
> >> Aw maan, I was looking forward to doing that myself :-)
> >
> >
> > Vagrant provider almost work, I have a few updates to do but first tests
> are
> > fine.
> >
> > However Vagrant needs VirtualBox on host (not image) to *pack* the image
> in
> > a Vagrant box. VirtualBox is in contrib, not free (and contrib backports
> for
> > wheezy).
> >
> > Should we require virtualbox and launch commands to create the box, even
> if
> > in contrib, or should we generate the vmdk disk and give instructions to
> the
> > user to create the pack box ? (I have a shell script for this for the
> > moment)
> >
> >
> >
> >>
> >> Guess I'll just work on my other plugin idea then: Minimizing the
> >> image size by e.g. preventing apt from writing the apt-cache to the
> >> bootstrap volume (meaning the vdi won't be expanded).
> >
> >
> > For the moment (what I coded), the virtualbox provider create a raw disk,
> > and is converted to vdi only at the end of the process. So vdi will be
> the
> > expanded size, not a compact one.
> >>
> >>
> >> > I had a quick look, and the fact is I get it worked because I use
> grub2
> >> > 2.0.x from sid. Previous versions (1.98, 1.99) did not manage
> correctly
> >> > indeed the loopback interface. The version > 2.0 manages it correctly.
> >> Oooh, ok. That explains it!
> >>
> >> > v2.x should be in testing but it seems some bugs prevent migration to
> >> > testing.
> >> No matter, we'll just handle it depending on the grub version once
> >> it's in debian testing. It'll cut down on the weird hackery I am doing
> >> now :-)
> >
> >
> > I do not know however when it gonna move to testing. Issue is related to
> > dependencies, so it could last long...  :-(
> >
> > Olivier
> >
> >
> >
> > --
> >
> > gpg key id: 4096R/326D8438  (keyring.debian.org)
> >
> > Key fingerprint = 5FB4 6F83 D3B9 5204 6335  D26D 78DC 68DB 326D 8438
>
> > Should we require virtualbox and launch commands to create the box, even
> if in contrib,
> That may be a possibility, though I'd like to keep all utilities free.
> Are you sure we can't package with qemu tools only?
>

No. The base disk of the vagrant box is created with qemu etc... but a
vagrant box is created from a vmdk disk (ok), a metadata json file (ok too)
and an ovf file.
Vagrant provides a vagrant package command that extracts a VM from
virtualbox and create the box package.
I can manually create the box, but the issue for the moment is to create an
ovf file matching the disk identifiers and MAC address of the VM.

I gonna look however if I could use an ovf file extracted from a test with
virtualbox and use it as a template to avoid virtualBox need.

Olivier

>
> > or should we generate the vmdk disk and give instructions to the user to
> create the pack box ? (I have a shell script for this for the moment)
> No, definitely not. The entire philosophy for this bootstrapper is to
> avoid exactly those scenarios, no matter how convoluted a packaging
> process may be, you can always script it. Once you require manual
> steps the iterable nature of the tool goes away. A great advantage of
> this tool is that you can modify the script/add plugins, bootstrap,
> test and then repeat the whole process without any cumbersome manual
> steps. Also, because of the manifest everything is reproducible,
> manual steps remove that feature.
>
> > For the moment (what I coded), the virtualbox provider create a raw
> disk, and is converted to vdi only at the end of the process. So vdi will
> be the expanded size, not a compact one.
> In the WIP branch I use qemu-nbd to bootstrap onto a proper vdi volume :-)
>
> > I had a quick look, and the fact is I get it worked because I use grub2
> 2.0.x from sid. Previous versions (1.98, 1.99) did not manage correctly
> indeed the loopback interface. The version > 2.0 manages it correctly.
> Oooh, ok. That explains it!
>
> > I do not know however when it gonna move to testing. Issue is related to
> dependencies, so it could last long...  :-(
> Doesn't matter, the hack is scripted now, so it's not really an issue
> any longer.
>
> p.s.: You forgot to reply to the mailing list btw.
>



-- 

gpg key id: 4096R/326D8438  (keyring.debian.org)

Key fingerprint = 5FB4 6F83 D3B9 5204 6335  D26D 78DC 68DB 326D 8438

Reply via email to