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

Re: [PATCH 1/2] Put back the deepcopy of the device stack in ActionDestroyFormat.



On Mon, 15 Feb 2010 19:01:59 -0600, Dave Lehman wrote:
> On Sat, 2010-02-13 at 08:12 +0100, Hans de Goede wrote:
>  > Hi,
>  >
>  > On 02/12/2010 09:28 PM, David Lehman wrote:
>  > > This is needed to make preexisting LUKS LVM PVs work correctly.
>  > >
>  >
>  > Hmm, iirc there were several issues with this approach, doesn't
>  > the need for this indicate a deviceaction sorting problem and
>  > wouldn't it be better to fix that instead?
>
> I don't recall the previous approach causing problems. Mostly I just
> wished it wasn't necessary. It isn't related to action sorting, but I
> will admit that I don't totally understand why it happens in certain
> situations and not in others.
>
> Since we're waiting on beta1 to clear anyway I'll take another look at
> this and see if there's a better solution.
>

IIRC the problem is that when we do the deepcopy, the (not deep copied)
partedDisk references in various places point to the partedDisk instance
that gets modified as we do partitioning. And when we reset these
partedDisk references to point to the origPartedDisk's before executing
actions, this only affects all the current formats / devices, not the
deepcopies.

So the destroy action gets executed on a device which (in some cases)
points to the state the partition table will be in at the end of partitioning,
instead of in the state it was in when we started, which may lead to us
operating on the wrong device node ie /dev/sda5 instead of /dev/sda6.

This is all very much IIRC, I was debugging some stuff where I suspected
the above issue and then your patch to remove the deepcopy came along and
I thought good riddance, and stopped worrying about this.

Regards,

Hans


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