Created attachment 335857 [details] Screenshot Description of problem: In F10, when installing in a KVM guest, the guest disks would show up in anaconda as QEMU disks. In rawhide, they show up as UNKNOWN (see attached screenshot). Version-Release number of selected component (if applicable): libvirt.i586 0.6.1-5.fc11 rawhide qemu.i586 2:0.10-0.9.kvm20090310git.fc11 rawhide virt-manager.i586 0.7.0-1.fc11 rawhide How reproducible: Everytime Additional info: <disk type='block' device='disk'> <source dev='/dev/mapper/VolGroup00-vguest1--disk0'/> <target dev='vda' bus='virtio'/> </disk>
Created attachment 335858 [details] Screenshot (shows IDE, SCSI and VirtIO disks) Attached screenshot shows an install with 3 disks (IDE, SCSI and VirtIO). Seems like just the virtio disks showing up as UNKNOWN
The virt host is F10 and included the following packages: * kvm-74-10.fc10.x86_64 * virt-manager-0.6.0-5.fc10.x86_64 * libvirt-0.5.1-2.fc10.x86_64 * kernel-2.6.27.19-170.2.35.fc10.x86_64
Verified this... QEMU does not get involved. Anaconda gets this information from sysfs. virtio devices does not export model information to sysfs. So this is a kernel problem. Whenever I get some cycles, I plan on implementing it. We can leave this bug as-is, to keep track of it.
Looks like these strings are coming from parted The way it works for Xen is: } else if (dev_major == XVD_MAJOR && (dev_minor % 0x10 == 0)) { dev->type = PED_DEVICE_XVD; ... case PED_DEVICE_XVD: if (!init_generic (dev, _("Xen Virtual Block Device"))) goto error_free_arch_specific; break; ... char* transport[14] = {... "xvd"}; See e.g. http://git.debian.org/?p=parted/parted.git;a=blob;f=libparted/arch/linux.c;h=6e47107610c67c787108a865fba9291b3632febc;hb=HEAD#l560 Moving to parted
Just chatting to jgranado on irc - virtio_blk doesn't have an assigned major: major = register_blkdev(0, "virtblk");
Only thing I could come up with that would be rather consistent is to use /sys/block/vda/device/modalias and grep for virtio. any suggestions?
OTOH /proc/devices and grep for virtblk is probably better. that is what is currently done with device-mapper devs.
this will be fixed in the next version of parted 1.8.8-15.
parted 1.8.8-16 is the next. sorry.
Cool, you went with the /proc/devices option - thanks. For reference, patch is this one: http://cvs.fedoraproject.org/viewvc/rpms/parted/devel/parted-1.8.8-virtio.patch?revision=1.1&view=markup
Hi Joel, Thanks for doing that. Patch looks perfect. If you post upstream, I'll ack/apply it. Bonus points if you can write or outline a test to exercise either part of the new code ;-)