Re: [libvirt] Consultation of the patch "util: Prepare helpers for unpriv_sgio setting"

>>>>> I added a disk to virtio-scsi bus through the VM's XML file like >>>>> this: >>>>> <disk type='block' device='disk'> >>>>> <driver name='qemu' type='raw' io='threads'/> >>>>> <source dev='/mnt/test.raw'/> >>>>> <target dev='sdb' bus='scsi'/> >>>>> <shareable/> >>>>> </disk> >>>>> >>>>> Then when I started the VM, I met the error. >>>>> [root@build SOURCES]# virsh start pan64ga >>>>> error: Failed to start domain pan64ga >>>>> error: Unable to get minor number of device '/mnt/test.raw': Invalid >>>>> argument

>> It's not reasonable. As the disk source you use is /mnt/test.raw, I >> believe it's not a block device. However, you used disk type='block'. >> Libvirt honors what you specified, and tries to grab the device IDs >> of the disk source (it expects the disk source is block, but it's not). > > This inspires me thinking of we should validate the disk source type > somewhere (when parsing or in the driver, as it's general for all > drivers). > Any opinions on should we validate whether the disk type and the specified source matches? Just realized that validating when parsing doesn't make sense, because the validation can imply the checking of whether the disk source existing or not, however, for a inactive domain, it's normal that its disk source may not existing. So the only way to do the validation is inside driver. But does it imply we will have regression if doing so? As yuxh said, the guest can actually be started successfully with block disk has a plain file source, not sure if it's expected result of qemu though.