Two questions, probably the same answer. I am just trying to
understand why you are trying to do this, and what practical use this
has. Maybe I will have a use for this in the future.
What is the goal in doing this?
Are usb thumb drives not an option?
Lewis
On Aug 19, 2009, at 7:56 AM, Joshua Kinard wrote:
Hmm, I don't think qcow format is going to work with that method.
At least when I tried it, the bochs bios didn't like what it was
looking at. But it's hard to tell if that was bochs, or ntldr
giving me a fit.
I do know that RAW images work fine, though, and I just tested it
with squashfs by running mksquashfs on the base image, and mounting
that via loop2 to my base/ subfolder. Then re-created the scratch
area, ran dmsetup, and booted. Loaded up great, did some changes,
shutdown, unloaded the mapper device, unloaded the scratch loop,
unmounted my tmpfs, and re-created that segment again, and Windows
came back in pristine condition. So that pretty much means my next
battle is Debian's package manager not liking my local repository (I
can't connect my Xen system to the internet, so I have to use such a
setup). As that's how the Xen LiveCD scripts are doing things, and
assuming everything can be fetched online.
Some things I have noticed, though:
- After detaching a loopback device, there seems to be a delay in
when the kernel actually notices that it's detached. If one does
losetup -d <loop> followed by losetup <loop> <new img>, it seems
like in a few cases, the kernel still thought it was looking at the
older loop attachment. I ran into this when using loopback to look
at the disk at the disk level, and then passing -o 32256 to losetup
to look strictly at the first/primary NTFS partition (while trying
to clone it to a smaller disk image, which I finally gave up on
doing).
- Haven't found the cause just yet, but dmsetup sometimes doesn't
like looking at loopback devices, and possibly per the first item,
refuses to create the mapped device, citing a resource is busy.
Reboot fixes this, and I suspect both of these are caused by all the
experimenting I'm doing with switching between the various
loopbacks, I probably confuse the kernel. From a scripted
standpoint, it should be a non-issue, though.
- Only tested this entire idea/setup against Windows XP. It seems
to be fine, and that generally means any other OS as a domU should
work fine as well (because NTFS is by far the pickiest, most
temperamental FS known). You get way more options with Linux/Other
UNIXes as far as tricking them into thinking they're in a r/w world,
but this can be extended to encapsulate quite a few domU's.
However, the only limitation I can see is the number of loopback
interfaces. I use three to setup the Windows domU: /dev/loop2 for
the squashfs mount, /dev/loop0 for the windows image inside the
squashfs mount, and /dev/loop1 for the scratch area. I think more
loop devices can be easily created, but I've not experimented that
far just yet...
That's all for now...I've got to install my network scanner into
this Windows image and re-clean it and run these tests all over again.
Cheers!
--J
From: Thiago Camargo Martins Cordeiro [thiagocmartinsc@xxxxxxxxx]
Sent: Tuesday, August 18, 2009 11:29 PM
To: Joshua Kinard
Cc: xen-users@xxxxxxxxxxxxxxxxxxx
Subject: Re: [Xen-users] Bootable LiveDVD w/ Xen that boots Windows
Image?
This procedure will be in the next release of the Xen Live CD!!
qcow2 and LVM CoW snapshot for saving storage space while providing
fast domU creation from RW snapshots.... good for Eucalyptus, I guess!
2009/8/18 Joshua Kinard <joshua.kinard@xxxxxxxxxxxxx>
Think I cracked it. Don't even need to deal with qemu's qcow stuff
(so far). Still more tests pending (including using read-only drive
mounts). But assume the following steps in a dom0:
- mkdir /windows
- cd /windows
- dd if=/dev/zero of=xp.img bs=1024k count=1 seek=7168
- touch xp.cfg
- nano xp.cfg
- <edit xp.cfg, setup as desired>
- xm create xp.cfg
- <install WinXP, patch, delete garbage, defrag, zero-out free
space, then shutdown>
- xm list (verify domU is actually dead)
Now after this, xp.img will contain a baseline WinXP install. In my
specific case, I compacted everything down to ~2GB total installed
(inside the domU that is) by using straight qemu-emulated devices
(not the GPLPV stuff), no page file or hibernation, stripping nearly
every misc utility, and a bunch of other stuff, but NO NTFS
compression (for now, at least).
Next:
- dd if=/dev/zero of=xp_scratch.img bs=1024k count=1 seek=2048
- losetup /dev/loop0 xp.img
- losetup /dev/loop1 xp_scratch.img
- echo 0 $(blockdev --getsize /dev/loop0) snapshot /dev/loop0 /dev/
loop1 n 64 | dmsetup create xpcow
This maps our baseline WinXP image (xp.img) to /dev/loop0 and a
scratch area to /dev/loop1 (I don't even know if this is needed just
yet). Needed because dmsetup only works with block devices. Using
dmsetup, a snapshot is created, so that all writes to loop0 get
redirected to loop1, and creates a device, /dev/mapper/xpcow.
Edit xp.cfg to use phy:/dev/mapper/xpcow,hda,w, and boot Windows:
- xm create xp.cfg
In Windows, create a dummy file on the desktop or three, then
shutdown. Re-start the domU and back at the desktop, we see that
the newly-created files still exist, so shutdown again. Now:
- dmsetup remove /dev/mapper/xpcow
- losetup -d /dev/loop1
- dd if=/dev/zero of=xp_scratch2.img bs=1024k count=1 seek=2048
- losetup /dev/loop1 xp_scratch2.img
- echo 0 $(blockdev --getsize /dev/loop0) snapshot /dev/loop0 /dev/
loop1 n 64 | dmsetup create xpcow
- xm create xp.cfg
Back at the desktop, the newly-created files are now gone!
I did it this way as a manual test, because I'm veryifying a few
things:
a. Simulating the effect of creating the scratch image in a /tmpfs
mount, while the baseline image will be on a physical, read-only
medium. By removing the mapped device and changing out the scratch
file, I was seeing whether the XP image would indeed revert to the
baseline state, to simulate a reboot in which /tmpfs is cleared.
Supposedly, the 'n' parameter to dmsetup does this anyways, but I
wanted to be doubly sure.
b. Need to scriptify the whole mess, since the baseline image will
already be pre-configured, the generation of the scratch image,
setup of the loop devices, and the device-mapping will be on-the-fly
when some future CD Image I put together boots up.
c. Avoids all the problems I see on Google about using blktap with
qcow with gplpv and such. I'm probably not grasping some of it, but
I think those references are aiming for different uses. So far,
this setup mostly relies wholly on the Linux dom0 to piece
everything together, leaving the WinXP domU utterly oblivious to
what is going on. And given Windows' nature, that's exactly what we
want.
Performance isn't an issue for me in this, so I'm sure on a CD, this
will be slower than running Oracle on an 8086 (pretend for a moment
that this is actually possible), but if it works, and WinXP doesn't
complain, then all is good.
Thoughts?
________________________________________
From: xen-users-bounces@xxxxxxxxxxxxxxxxxxx [xen-users-bounces@xxxxxxxxxxxxxxxxxxx
] On Behalf Of Joshua Kinard [joshua.kinard@xxxxxxxxxxxxx]
Sent: Monday, August 17, 2009 9:50 AM
To: xen-users@xxxxxxxxxxxxxxxxxxx
Subject: RE: [Xen-users] Bootable LiveDVD w/ Xen that boots Windows
Image?
Hmm, block-level copy-on-write? If qcow supports this out-of-the-
box, then I technically don't need any kind of UnionFS? Just
someway to specify to Xen a read-only image with a qcow2 image in a
writable zone?
Going to dredge through some EU software patent I dug up from
Microsoft. Supposed to document the /MININT switch to the NT kernel
that supposedly makes Windows write all registry changes to volatile
memory instead of to the registry hives. It's mostly used in the
WinPE environment, but it looks like BartPE leverages it too. Seems
I need more to it, though, than just modifying boot.ini to pass it,
as once I started to boot Windows from a read-only drive mount
(after discovering Xen/qemu don't properly honor the mode bit in the
Xen config for r or r/w, and even file-level permissions), Windows
crashed with UNMOUNTABLE_BOOT_VOLUME as its error. But it
highlights the capability is there....just not easy.
USB is out, unfortunately, too. For security reasons, we ban USB/
Firewire drives here, with the exception of CD Burners. Ditto on
memory cards (SD, MMC, etc..). I've got very narrow operating
parameters to work in, which is why this is proving to be quite a
fun challenge.
Thanks!,
--J
________________________________________
From: Fajar A. Nugraha [fajar@xxxxxxxxx]
Sent: Friday, August 14, 2009 4:53 AM
To: Joshua Kinard
Cc: xen-users@xxxxxxxxxxxxxxxxxxx
Subject: Re: [Xen-users] Bootable LiveDVD w/ Xen that boots Windows
Image?
On Fri, Aug 14, 2009 at 1:37 AM, Joshua
Kinard<joshua.kinard@xxxxxxxxxxxxx> wrote:
> Hmm, I didn't think of it that way.
>
> The way I read up on UnionFS and aufs' functionality, was they
could essentially merge two files virtually,
Merge two directories together, to be exact. Not merge files.
> i.e., the kernel module would be able to look at the operation
coming in and route it to the proper descriptor (i.e., read() --> /
livecd/windows/winxp.img OR write() --> /tmp/winxp.img, with /tmp
being in tmpfs). I guess it's not as granular as that it seems.
Would be a neat trick, but I imagine it'd be complex as anything for
a kernel module to have to keep track of which files have variants
loaded in the writeable union area.
What you describe is essentially block-level copy-on-write, which is
what qcow or zfs does. Aufs/unionfs does this per file.
> As for the application, it's a complex network security scanner,
made by eEye Digital Security, called "Retina". We just don't want
to setup and baby sit Windows installations on our Unix networks
strictly for this one app, so I figured if I can get it to run off
of a CD, we can just park some diskless hardware in a closet and
pull it out whenever we need to do network testing and such.
If it were me I'd simply setup a Windows domU on any server with
enough disk and RAM, make a "template" of the "good" installation
(preferably zfs or LVM snapshot), start it only when it's needed, shut
it down afterwards, and (if necessary) rollback to the "good"
template. That is assuming that all host/network tested reachable from
my datacenter (either with vlans or routing). No need to add the
complexity of a live DVD/USB.
If portability is a requirement, and you're absolutely sure you'll
always have VT-capable host handy, then using live USB is much
perefered than DVD due to performance and complexity reasons. But hey,
that's just me :D
Let us know if you find a solution that works.
--
Fajar
_______________________________________________
Xen-users mailing list
Xen-users@xxxxxxxxxxxxxxxxxxx
http://lists.xensource.com/xen-users
_______________________________________________
Xen-users mailing list
Xen-users@xxxxxxxxxxxxxxxxxxx
http://lists.xensource.com/xen-users
_______________________________________________
Xen-users mailing list
Xen-users@xxxxxxxxxxxxxxxxxxx
http://lists.xensource.com/xen-users
_______________________________________________
Xen-users mailing list
Xen-users@xxxxxxxxxxxxxxxxxxx
http://lists.xensource.com/xen-users
|
Home