[libvirt] [PATCH] qemu: blockCopy: Allow shallow block copy into a raw image
The documentation states that for shallow block copy the image has to have the same guest visible content as backing file of the current image. This condition can be achieved also with a raw file (or a qcow without a backing file) so remove the condition that would disallow it. (This patch additionally fixes crash described in https://bugzilla.redhat.com/show_bug.cgi?id=1215569 since it removes the code) --- src/qemu/qemu_driver.c | 10 ---------- 1 file changed, 10 deletions(-) diff --git a/src/qemu/qemu_driver.c b/src/qemu/qemu_driver.c index 70bf7aa..f979d33 100644 --- a/src/qemu/qemu_driver.c +++ b/src/qemu/qemu_driver.c @@ -16815,16 +16815,6 @@ qemuDomainBlockCopyCommon(virDomainObjPtr vm, if (qemuDomainDetermineDiskChain(driver, vm, disk, false, true) < 0) goto endjob; - if ((flags & VIR_DOMAIN_BLOCK_COPY_SHALLOW) && - mirror->format == VIR_STORAGE_FILE_RAW && - disk->src->backingStore->path) { - virReportError(VIR_ERR_CONFIG_UNSUPPORTED, - _("disk '%s' has backing file, so raw shallow copy " - "is not possible"), - disk->dst); - goto endjob; - } - /* Prepare the destination file. */ /* XXX Allow non-file mirror destinations */ if (!virStorageSourceIsLocalStorage(mirror)) { -- 2.3.5
On 04/27/2015 08:01 PM, Peter Krempa wrote:
The documentation states that for shallow block copy the image has to have the same guest visible content as backing file of the current image. This condition can be achieved also with a raw file (or a qcow without a backing file) so remove the condition that would disallow it.
(This patch additionally fixes crash described in https://bugzilla.redhat.com/show_bug.cgi?id=1215569 since it removes the code) --- src/qemu/qemu_driver.c | 10 ---------- 1 file changed, 10 deletions(-)
diff --git a/src/qemu/qemu_driver.c b/src/qemu/qemu_driver.c index 70bf7aa..f979d33 100644 --- a/src/qemu/qemu_driver.c +++ b/src/qemu/qemu_driver.c @@ -16815,16 +16815,6 @@ qemuDomainBlockCopyCommon(virDomainObjPtr vm, if (qemuDomainDetermineDiskChain(driver, vm, disk, false, true) < 0) goto endjob;
- if ((flags & VIR_DOMAIN_BLOCK_COPY_SHALLOW) && - mirror->format == VIR_STORAGE_FILE_RAW && - disk->src->backingStore->path) { - virReportError(VIR_ERR_CONFIG_UNSUPPORTED, - _("disk '%s' has backing file, so raw shallow copy " - "is not possible"), - disk->dst); - goto endjob; - } -
Although a shallow blockcopy of file without backing file is semantically correct, but still feel a little weird. And, a shallow blockcommit of file without backing file will failed with error "error: invalid argument: top '/var/lib/libvirt/images/raw.img' in chain for 'vda' has no backing file" Should libvirt post error when try a shallow blockcopy of file without backing file, just as shallow blockcommit?
/* Prepare the destination file. */ /* XXX Allow non-file mirror destinations */ if (!virStorageSourceIsLocalStorage(mirror)) {
-- Regards shyu
On Mon, Apr 27, 2015 at 22:11:40 +0800, Shanzhi Yu wrote:
On 04/27/2015 08:01 PM, Peter Krempa wrote:
The documentation states that for shallow block copy the image has to have the same guest visible content as backing file of the current image. This condition can be achieved also with a raw file (or a qcow without a backing file) so remove the condition that would disallow it.
(This patch additionally fixes crash described in https://bugzilla.redhat.com/show_bug.cgi?id=1215569 since it removes the code) --- src/qemu/qemu_driver.c | 10 ---------- 1 file changed, 10 deletions(-)
diff --git a/src/qemu/qemu_driver.c b/src/qemu/qemu_driver.c index 70bf7aa..f979d33 100644 --- a/src/qemu/qemu_driver.c +++ b/src/qemu/qemu_driver.c @@ -16815,16 +16815,6 @@ qemuDomainBlockCopyCommon(virDomainObjPtr vm, if (qemuDomainDetermineDiskChain(driver, vm, disk, false, true) < 0) goto endjob;
- if ((flags & VIR_DOMAIN_BLOCK_COPY_SHALLOW) && - mirror->format == VIR_STORAGE_FILE_RAW && - disk->src->backingStore->path) { - virReportError(VIR_ERR_CONFIG_UNSUPPORTED, - _("disk '%s' has backing file, so raw shallow copy " - "is not possible"), - disk->dst); - goto endjob; - } -
Although a shallow blockcopy of file without backing file is semantically correct, but still feel a little weird. And, a shallow blockcommit of file without backing file will failed with error "error: invalid argument: top '/var/lib/libvirt/images/raw.img' in chain for 'vda' has no backing file"
Should libvirt post error when try a shallow blockcopy of file without backing file, just as shallow blockcommit?
I cannot reproduce the error above, could you please post steps to do that? or perhaps debug log from libvirt? When copying from a single layer image with the --shallow flag I'm able to successfully copy both into a raw and a qcow2 file.
/* Prepare the destination file. */ /* XXX Allow non-file mirror destinations */ if (!virStorageSourceIsLocalStorage(mirror)) {
Peter
On 04/27/2015 11:38 PM, Peter Krempa wrote:
On Mon, Apr 27, 2015 at 22:11:40 +0800, Shanzhi Yu wrote:
On 04/27/2015 08:01 PM, Peter Krempa wrote:
The documentation states that for shallow block copy the image has to have the same guest visible content as backing file of the current image. This condition can be achieved also with a raw file (or a qcow without a backing file) so remove the condition that would disallow it.
(This patch additionally fixes crash described in https://bugzilla.redhat.com/show_bug.cgi?id=1215569 since it removes the code) --- src/qemu/qemu_driver.c | 10 ---------- 1 file changed, 10 deletions(-)
diff --git a/src/qemu/qemu_driver.c b/src/qemu/qemu_driver.c index 70bf7aa..f979d33 100644 --- a/src/qemu/qemu_driver.c +++ b/src/qemu/qemu_driver.c @@ -16815,16 +16815,6 @@ qemuDomainBlockCopyCommon(virDomainObjPtr vm, if (qemuDomainDetermineDiskChain(driver, vm, disk, false, true) < 0) goto endjob;
- if ((flags & VIR_DOMAIN_BLOCK_COPY_SHALLOW) && - mirror->format == VIR_STORAGE_FILE_RAW && - disk->src->backingStore->path) { - virReportError(VIR_ERR_CONFIG_UNSUPPORTED, - _("disk '%s' has backing file, so raw shallow copy " - "is not possible"), - disk->dst); - goto endjob; - } - Although a shallow blockcopy of file without backing file is semantically correct, but still feel a little weird. And, a shallow blockcommit of file without backing file will failed with error "error: invalid argument: top '/var/lib/libvirt/images/raw.img' in chain for 'vda' has no backing file"
Should libvirt post error when try a shallow blockcopy of file without backing file, just as shallow blockcommit?
I cannot reproduce the error above, could you please post steps to do that? or perhaps debug log from libvirt?
It is easy to reproduce. Suppose live guest xml looks like: # virsh dumpxml vm1|grep disk -A 8 <disk type='file' device='disk'> <driver name='qemu' type='qcow2'/> <source file='/var/lib/libvirt/images/raw.1430143276'/> <backingStore type='file' index='1'> <format type='raw'/> <source file='/var/lib/libvirt/images/raw.img'/> <backingStore/> </backingStore> <target dev='vda' bus='virtio'/> <alias name='virtio-disk0'/> <address type='pci' domain='0x0000' bus='0x00' slot='0x03' function='0x0'/> </disk> Then try a shallow blockcommit of file has no backing file # virsh blockcommit vm1 vda --top vda[1] --shallow error: invalid argument: top '/var/lib/libvirt/images/raw.img' in chain for 'vda' has no backing file
When copying from a single layer image with the --shallow flag I'm able to successfully copy both into a raw and a qcow2 file.
I also can do it. But a shallow block copy of file which has backing file into a raw file does not make any sense, right?
/* Prepare the destination file. */ /* XXX Allow non-file mirror destinations */ if (!virStorageSourceIsLocalStorage(mirror)) {
Peter
-- Regards shyu
On 04/27/2015 10:08 PM, Shanzhi Yu wrote:
Should libvirt post error when try a shallow blockcopy of file without backing file, just as shallow blockcommit? I cannot reproduce the error above, could you please post steps to do that? or perhaps debug log from libvirt?
It is easy to reproduce. Suppose live guest xml looks like:
# virsh dumpxml vm1|grep disk -A 8 <disk type='file' device='disk'> <driver name='qemu' type='qcow2'/> <source file='/var/lib/libvirt/images/raw.1430143276'/> <backingStore type='file' index='1'> <format type='raw'/> <source file='/var/lib/libvirt/images/raw.img'/> <backingStore/> </backingStore> <target dev='vda' bus='virtio'/> <alias name='virtio-disk0'/> <address type='pci' domain='0x0000' bus='0x00' slot='0x03' function='0x0'/> </disk>
Then try a shallow blockcommit of file has no backing file
But a shallow block copy of file which has backing file into a raw file does not make any sense, right?
A shallow block copy of: A <- B into a pre-existing file C, where C starts life with contents identical to A, makes sense. A shallow block copy where _libvirt_ creates C does not make sense. The key is that the moment you use the REUSE_EXTERNAL flag, _you_ are now responsible for providing a valid destination (and anything is supported; libvirt should not get in your way). -- Eric Blake eblake redhat com +1-919-301-3266 Libvirt virtualization library http://libvirt.org
On 04/27/2015 06:01 AM, Peter Krempa wrote:
The documentation states that for shallow block copy the image has to have the same guest visible content as backing file of the current image. This condition can be achieved also with a raw file (or a qcow without a backing file) so remove the condition that would disallow it.
(This patch additionally fixes crash described in https://bugzilla.redhat.com/show_bug.cgi?id=1215569 since it removes the code) --- src/qemu/qemu_driver.c | 10 ---------- 1 file changed, 10 deletions(-)
This removes a bit of a safety valve that tries to prevent users from accidentally creating corrupt images when they use the API incorrectly, but as the safety valve didn't always work and we could actually crash, and as the documentation has already warned the user about using it correctly, I can probably live with it.
diff --git a/src/qemu/qemu_driver.c b/src/qemu/qemu_driver.c index 70bf7aa..f979d33 100644 --- a/src/qemu/qemu_driver.c +++ b/src/qemu/qemu_driver.c @@ -16815,16 +16815,6 @@ qemuDomainBlockCopyCommon(virDomainObjPtr vm, if (qemuDomainDetermineDiskChain(driver, vm, disk, false, true) < 0) goto endjob;
- if ((flags & VIR_DOMAIN_BLOCK_COPY_SHALLOW) && - mirror->format == VIR_STORAGE_FILE_RAW && - disk->src->backingStore->path) { - virReportError(VIR_ERR_CONFIG_UNSUPPORTED, - _("disk '%s' has backing file, so raw shallow copy " - "is not possible"), - disk->dst); - goto endjob; - }
However, one thing I'm not sure about: a shallow block copy with the REUSE_EXTERNAL flag always makes sense (raw or not, because the user of that flag knows what they are doing). But a shallow block copy without reusing an external file does not make sense - if libvirt is creating the destination file, then we KNOW that the destination file does not match the guest contents (except for the corner case where the guest is not using the block and therefore it remains in the default state of all 0s because it is unwritten). I'm not convinced yet that this is the right patch. -- Eric Blake eblake redhat com +1-919-301-3266 Libvirt virtualization library http://libvirt.org
participants (3)
-
Eric Blake -
Peter Krempa -
Shanzhi Yu