too much partition clearing

[Chris Lumens]

For reference:

https://bugzilla.redhat.com/bugzilla/show_bug.cgi?id=232936

We're just running the partition clearing code too often.  Every time
through openDevices, we are clearing the disks out.  This happens even
after partitions have been created.  And since it's kind of hard to make
physical volumes or RAID members on partitions that don't exist, we
explode.

One fix could be the addition of some sort of flag that gets set the
first time we clear partitions, and then is checked every time.  Yuck.
Also, openDevices is kind of a mess and needs to be doing less stuff on
each of the many times it's called.  It especially doesn't need to do
something destructive like clearing out the partitions every time.

I've moved the clearpart stuff out into its own method and call it from
turnOnFilesystems and when we do the early swap stuff.  I know this is
really subtle code so I may have messed something up.  However, basic
tests I've performed here today seem okay so far.  The patch is
attached.  Comments welcome.

- Chris

[Jeremy Katz]

On Tue, 2007-03-27 at 17:11 -0400, Chris Lumens wrote:
> I've moved the clearpart stuff out into its own method and call it from
> turnOnFilesystems and when we do the early swap stuff.  I know this is
> really subtle code so I may have messed something up.  However, basic
> tests I've performed here today seem okay so far.  The patch is
> attached.  Comments welcome.

While a little scary, I think this is probably the right thing to do.  I
expect that some of the weird cases like dmraid or mpath are where it's
most likely to blow up

Jeremy

[Chris Lumens]

> > I've moved the clearpart stuff out into its own method and call it from
> > turnOnFilesystems and when we do the early swap stuff.  I know this is
> > really subtle code so I may have messed something up.  However, basic
> > tests I've performed here today seem okay so far.  The patch is
> > attached.  Comments welcome.
>
> While a little scary, I think this is probably the right thing to do.  I
> expect that some of the weird cases like dmraid or mpath are where it's
> most likely to blow up

I'm pretty worried about it as well.  This code's basically a minefield
of special cases and side effects.  But I've committed it.  I'll try to
get some testing from the QA guys on the dmrain and mpath stuff,
assuming they have that capability.

- Chris