8:18 a.m.
On 1/26/19 10:31 AM, John Ferlan wrote:
On 1/26/19 6:37 AM, Daniel Henrique Barboza wrote:
>
> On 1/25/19 8:27 PM, John Ferlan wrote:
>> On 1/18/19 1:59 PM, Daniel Henrique Barboza wrote:
>>> In a case where we want to hotplug the following disk:
>>>
>>> <disk type='file' device='disk'>
>>> (...)
>>> <address type='drive' controller='0' bus='0'
target='0' unit='0'/>
>>> </disk>
>>>
>>> In a QEMU guest that has a single OS disk, as follows:
>>>
>>> <disk type='file' device='disk'>
>>> (...)
>>> <address type='drive' controller='0' bus='0'
target='0' unit='0'/>
>>> </disk>
>>>
>>> What happens is that the existing guest disk will receive the ID
>>> 'scsi0-0-0-0' due to how Libvirt calculate the alias based on
>>> the address in qemu_alias.c, qemuAssignDeviceDiskAlias. When hotplugging
>>> a disk that happens to have the same address, Libvirt will calculate
>>> the same ID to it and attempt to device_add. QEMU will refuse it:
>>>
>>> $ virsh attach-device dhb hp-disk-dup.xml
>>> error: Failed to attach device from hp-disk-dup.xml
>>> error: internal error: unable to execute QEMU command 'device_add':
>>> Duplicate ID 'scsi0-0-0-0' for device
>>>
>>> And Libvirt follows it up with a cleanup code in
>>> qemuDomainAttachDiskGeneric
>>> that ends up removing what supposedly is a faulty hotplugged disk
>>> but, in
>>> this case, ends up being the original guest disk. This happens
>>> because Libvirt
>>> doesn't differentiate the error received by QMP device_add.
>>>
>>> An argument can be made for how QMP device_add should provide a
>>> different
>>> error code for this scenario. A quicker way to solve the problem is
>>> presented in this patch: let us check the address of disk to be
>>> attached and
>>> see if there is already a disk with the same address in the VM
>>> definition.
>>> In this case, error out without calling device_add.
>>>
>>> After this patch, this is the result of the previous attach-device call:
>>>
>>> $ ./run tools/virsh attach-device dhb ~/hp-disk-dup.xml
>>> error: Failed to attach device from /home/danielhb/hp-disk-dup.xml
>>> error: operation failed: attached disk address conflicts with
>>> existing disk 'scsi0-0-0-0'
>>>
>>> Reported-by: Srikanth Aithal <bssrikanth(a)in.ibm.com>
>>> Signed-off-by: Daniel Henrique Barboza <danielhb413(a)gmail.com>
>>> ---
>>> src/qemu/qemu_hotplug.c | 39 +++++++++++++++++++++++++++++++++++++++
>>> 1 file changed, 39 insertions(+)
>>>
>>> diff --git a/src/qemu/qemu_hotplug.c b/src/qemu/qemu_hotplug.c
>>> index a1c3ca999b..4e6703f0b8 100644
>>> --- a/src/qemu/qemu_hotplug.c
>>> +++ b/src/qemu/qemu_hotplug.c
>>> @@ -875,6 +875,36 @@ qemuDomainChangeEjectableMedia(virQEMUDriverPtr
>>> driver,
>>> }
>>> +/**
>>> + * qemuDomainFindDiskByAddress:
>>> + *
>>> + * Returns an existing disk in the VM definition that matches a given
>>> + * bus/controller/unit/target set, NULL in no match was found. */
>>> +static virDomainDiskDefPtr
>>> +qemuDomainFindDiskByAddress(virDomainDefPtr def,
>>> + virDomainDeviceInfo info)
>>> +{
>>> + virDomainDeviceInfo vm_info;
>> Prefer to see something like devinfo or diskinfo - what you look at
>> isn't a vm (virtual machine)!
> Ok!
>
>>> + int idx;
>>> +
>>> + for (idx = 0; idx < def->ndisks; idx++) {
>>> + vm_info = def->disks[idx]->info;> + if
>>> ((vm_info.addr.drive.bus == info.addr.drive.bus) &&
>>> + (vm_info.addr.drive.controller ==
>>> info.addr.drive.controller) &&
>>> + (vm_info.addr.drive.unit == info.addr.drive.unit)) {
>> This would seem to assuming something about the address type wouldn't
>> it? A _virDomainDeviceInfo is structure that uses it's @type in order
>> to determine what type of address is being used...
> Yeah, this function is being called under a conditional inside
> qemuDomainAttachDiskGeneric:
>
qemuDomainAttachDiskGeneric has 3 callers:
qemuDomainAttachVirtioDiskDevice
qemuDomainAttachSCSIDisk
qemuDomainAttachUSBMassStorageDevice
So why wouldn't the same possibility exist for other address types? A
very generically named function qemuDomainFindDiskByAddress is doing
something that's SCSI specific.
Still prior to that those 3 functions above are called by
qemuDomainAttachDeviceDiskLiveInternal
Within that context there's a call to virDomainDiskDefCheckDuplicateInfo
which perhaps may be place to check that if the "to be added" disk has
an address e.g. call virDomainDeviceInfoAddressIsEqual and if it returns
true, then you have an error.
That's a good point. Perhaps there is a way to do this verification
regardless of address type.
Still this is something that virDomainDefValidate processing does via
virDomainDefCheckDuplicateDiskInfo and I'm wondering why
qemuDomainAttachDeviceLiveAndConfig doesn't do during it's processing of
virDomainDeviceDefParse, but I don't have the cycles right now to
investigate
Fair enough. I'll investigate this further before sending a v3. Perhaps
an RFC
patch to explore the idea might be a good idea too.
DHB
John
> + if ((disk->info.type == VIR_DOMAIN_DEVICE_ADDRESS_TYPE_DRIVE) &&
> + (vm_disk = qemuDomainFindDiskByAddress(vm->def, disk->info))) {
>
> It would be clearer to simply move this conditional inside the function
> I guess.
>
>
>
>> Look at tests/qemuxml2xmloutdata/disk-virtio.xml and wonder what would
>> happen...
>>
>> Suggestion - look for virDomainDeviceInfoAddressIsEqual and maybe even
>> virDomainDefHasDeviceAddress
>>
>> Perhaps even search through domain_conf.c for 'ndisks' and see how other
>> iterations are done.
> I'll take a look into those.
>
> Just to make my intentions clear: in v1 I just made an alias check
> after qemuAssingDeviceDiskAlias to see if the calculated alias already
> exists
> in the domain def. With this new patch the idea was to check for the same
> address instead - giving that the alias code uses the address to
> calculate, the
> effect would be the same as v1 with a plus of avoiding a collision with
> user
> created aliases. The logic of address checking was inspired by how the
> address
> is used inside qemuAssignDeviceDiskAlias.
>
>
>
> Thanks,
>
>
> DHB
>
>
>
>
>> In a way I'm wondering how we got this far where XML with a duplicated
>> address was accepted. Of course I haven't thought/looked that hard at
>> the hotplug path lately either.
>>
>> John
>>
>>
>>> + /* Address does not have target to compare. We have
>>> + * a match. */
>>> + if (!info.addr.drive.target)
>>> + return def->disks[idx];
>>> + else if (vm_info.addr.drive.target &&
>>> + vm_info.addr.drive.target ==
>>> info.addr.drive.target)
>>> + return def->disks[idx];
>>> + }
>>> + }
>>> + return NULL;
>>> +}
>>> +
>>> +
>>> /**
>>> * qemuDomainAttachDiskGeneric:
>>> *
>>> @@ -888,12 +918,21 @@ qemuDomainAttachDiskGeneric(virQEMUDriverPtr
>>> driver,
>>> int ret = -1;
>>> qemuDomainObjPrivatePtr priv = vm->privateData;
>>> qemuHotplugDiskSourceDataPtr diskdata = NULL;
>>> + virDomainDiskDefPtr vm_disk = NULL;
>>> char *devstr = NULL;
>>> virQEMUDriverConfigPtr cfg = virQEMUDriverGetConfig(driver);
>>> if (qemuHotplugPrepareDiskAccess(driver, vm, disk, NULL,
>>> false) < 0)
>>> goto cleanup;
>>> + if ((disk->info.type == VIR_DOMAIN_DEVICE_ADDRESS_TYPE_DRIVE)
&&
>>> + (vm_disk = qemuDomainFindDiskByAddress(vm->def, disk->info)))
{
>>> + virReportError(VIR_ERR_OPERATION_FAILED,
>>> + _("attached disk address conflicts with
>>> existing "
>>> + "disk '%s'"),
vm_disk->info.alias);
>>> + goto error;
>>> + }
>>> +
>>> if (qemuAssignDeviceDiskAlias(vm->def, disk, priv->qemuCaps)
< 0)
>>> goto error;
>>>