This is an archived copy of the Xen.org mailing list, which we have preserved to ensure that existing links to archives are not broken. The live archive, which contains the latest emails, can be found at http://lists.xen.org/
Home Products Support Community News


RE: [Xen-API] How Pygrub work on VHD

On Mon, 2010-01-25 at 13:13 -0800, Dave Scott wrote:
> Hi Anthony,
> > Thanks for your quick response,
> > 
> > my understanding is driver (maybe in user space) behind /dev/xvda
> > translates raw disk image format to VHD image format on demand, when
> > pygrub works on /dev/xvda
> That's correct. When a vhd-based VDI is attached to a domain, 
> blktap(kernel-space) + tapdisk(user-space) do the translation from raw disk 
> block accesses to vhd read/writes.

What you are talking about is how VM accesses vhd-based disk image. What
I want to know is how pygrub grabs kernel and initrd from vhd-based disk
image, pygrub is running on dom0, there is no /dev/xvda, which is inside


> > What I'm doing is,
> > 
> > 1. Create a lvm vdi on iscsi SR,
> > 2. dd a vhd file to this vdi,
> > 3. attach this vdi to a (empty)PV vm as device 0(vbd),
> > 4. mark this vbd bootable,
> > 5. then start this vm
> Unfortunately this isn't going to work. The choice of whether to use blktap 
> (vhd-capable) or blkback (raw device only) is a function of the SR's 
> content_type. The 'iscsi' SR uses blkback :(
> To see what I mean, try something like this instead:
> 1. Create an 'lvmoiscsi' SR
> 2. create a VDI in the new SR
> 3. look inside the new LV -- it should have vhd metadata
> Are you trying to import disks in .vhd format? The most future-proof way of 
> doing this is to:
> * create a VDI using the API
> * hotplug the VDI into a VM (eg dom0 or a domU)
> * decode the .vhd data, write() it to the raw block device and use seek() to 
> preserve sparseness
> Simply dd'ing an existing .vhd is risky because XCP is expecting the .vhd to 
> have a particular, optimized layout. In particular:
> * extra space is left at the beginning of the file for later resizing
> * parent locators have a particular naming convention
> * blocks are carefully aligned for performance

I understand all you said, but the volume in lvmoiscsi SR seems have the
exact format as VHD file. I'll get back to you after some experimentals

- Anthony

xen-api mailing list