questions on backing chain file, block lun, active commit

[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index]

 



While investigating active commit, I noticed that our code for probing
backing chains currently guesses at type="file" vs. type="block" solely
based on stat results (regular file vs. block device).  Is a block
device allowed to be used as <disk type='block' device='lun'> at the
same time it has qcow2 format, or does use of device='lun' enforce that
the block device is used with raw format?  My worry is that if a lun
device can be used with qcow2 format, then it can have a non-block
backing file, and an active commit would convert <disk type='block'
device='lun'> into <disk type='file' device='lun'> which is not valid.

On a related vein, that means that an active commit currently auto-picks
whether the <disk> will be listed as type='file' or type='block' solely
on the stat results.  Elsewhere, we allow type='file' even for block
devices, where a noticeable difference is how virDomainGetBlockInfo()
reports allocation (for type='file', allocation is based on how sparse
the file is, but a block device is not sparse; for type='block',
allocation is based on asking qemu the maximum used cluster).  Should we
be auto-converting to type='block' in other cases where we have stat
results?  For example, I posted a patch today to let the user explicitly
request that virsh blockcopy to a block device will use type='block'
because the existing code has always used type='file', but if we were
doing things automatically, then the user would automatically get
type='block' for block device destinations instead of having to request
it; but has no way to forcefully list a file even when the destination
is a block (at least, not until I implement virDomainBlockCopy() that
takes an XML <disk> description).

One drawback of autoconversion is the XML changes - with type='file',
the host resource is located at xpath disk/source/@file, but with
type='block', it is at xpath disk/source/@dev.  If we ever convert a
user's type='file' into a block device, but the user isn't expecting the
conversion, then they won't be able to find the source file name in the
output XML.  So on that ground, autoprobing type='block' for active
commit is fishy; we really should be honoring the user's input for
<backingStore> rather than probing it ourselves every time.  Until that
point, and given that I just proposed a patch to turn on type='block'
for blockcopy, where type='file' is used in all other cases, should we
have the same sort of flag for active commit?

-- 
Eric Blake   eblake redhat com    +1-919-301-3266
Libvirt virtualization library http://libvirt.org

Attachment: signature.asc
Description: OpenPGP digital signature

--
libvir-list mailing list
libvir-list@xxxxxxxxxx
https://www.redhat.com/mailman/listinfo/libvir-list

[Index of Archives]     [Virt Tools]     [Libvirt Users]     [Lib OS Info]     [Fedora Users]     [Fedora Desktop]     [Fedora SELinux]     [Big List of Linux Books]     [Yosemite News]     [KDE Users]     [Fedora Tools]