[PATCH rhel6-branch] Increment MD container child counter even if its volumes are ignored (#1120640)

David Lehman dlehman at redhat.com
Fri Jul 18 15:13:42 UTC 2014


On 07/18/2014 09:00 AM, Artur Paszkiewicz wrote:
> On 07/18/2014 02:50 PM, David Lehman wrote:
>> On 07/18/2014 03:21 AM, Artur Paszkiewicz wrote:
>>> On 07/18/2014 09:49 AM, Vratislav Podzimek wrote:
>>>> On Fri, 2014-07-18 at 09:32 +0200, Artur Paszkiewicz wrote:
>>>>> On 07/17/2014 05:30 PM, David Lehman wrote:
>>>>>> On 07/17/2014 06:39 AM, Artur Paszkiewicz wrote:
>>>>>>> Volumes which are not selected in Specialized Storage Devices are
>>>>>>> ignored in addUdevDevice. This causes that their parent container
>>>>>>> won't have its child counter incremented, which can later cause an
>>>>>>> incorrect unusedRaidMembersWarning.
>>>>>>
>>>>>> Why would you not select all members/components of a container you plan to use?
>>>>>
>>>>> I may not want to use some particular container for installation. There
>>>>> can be multiple containers on different disks. This results in a
>>>>> misleading warning, that "disks contain BIOS RAID metadata, but are not
>>>>> part of any recognized BIOS RAID sets". This is not true - the RAID
>>>>> arrays are assembled and recognized, just not selected in the installer.
>>>> But only if you select some devices from the container for installation,
>>>> right? That's seems like a situation dangerous enough to show a warning
>>>> even if a bit misleading.
>>>
>>> No, this happens when I don't select any devices (BIOS RAID sets) from
>>> the container. I can't see why this could be dangerous. I'm not
>>> selecting any of the component disks of the container, in fact they are
>>
>> But what if you were selecting one of them? This patch would disable a valid warning.
>>
>> My take is that we shouldn't be showing a warning about a fwraid array on unselected disks. Rather than lie about what we've found, we should be more careful about what conditions we warn under.
>
> I see it this way, that this patch does not really disable the warning,
> but it tries to workaround a problem with MD RAID arrays detection:
> - a MD container is never ignored
> - a MD volume which belongs to a container is ignored when it is not
>    selected, and its container's child counter won't be incremented
>
> The warning is displayed for containers having kids == 0, so I get a
> warning about unrecognized BIOS RAID sets, even though they are
> recognized, just not selected for use.

This is still a bit of a hack. A device with kids > 0 is expected to be 
the parent of some other device. I don't know if that would cause any 
problems in the branch you are operating on, but it most likely would in 
the newer branches.

Normally if a fwraid array is not selected, its member devices will be 
ignored along with it. As a result, no warning about unused raid members 
is shown. I'm not sure why the container has any children if all of its 
member disks are ignored. I still think this is the root of the problem. 
Whatever mechanism we use for non-intel fwraid should be the mechanism 
used for intel fwraid as well.

David


More information about the anaconda-patches mailing list