Bug 498137

dlehman brought this to my attention.  The reproducer is simple using pyparted and ipython.  Conditions:

1) msdos labeled disk with some partitions on it, but NOT starting at the beginning of the disk.  I started around cylinder 100 and created two partitions.

2) Make filesystems on the partitions, just for sanity.

3) Start ipython and run the following:


In [1]: import parted

In [2]: device = parted.getDevice("/dev/sdb")

In [3]: disk = parted.Disk(device)

In [4]: freespace = disk.getFreeSpaceRegions()

In [5]: freespace
Out[5]: [<parted.geometry.Geometry object at 0x1c019d0>]

In [6]: freespace[0].start
Out[6]: 63L

In [7]: disk.deleteAllPartitions()
Out[7]: True

In [8]: disk.commit()
Out[7]: True

In [9]: freespace = disk.getFreeSpaceRegions()

In [10]: freespace
Out[10]: [<parted.geometry.Geometry object at 0x1c01c10>]

In [11]: freespace[0].start
Out[11]: 0L


Problem:
The disk still has an msdos label, but when we delete all partitions, libparted internally updates the part_list with new freespace regions.  After all partitions are removed, the new freespace region starts at sector 0.  However, if we have any partitions on the disk but some room at the beginning free, libparted will report sector 63 as the first available sector for free space.

dlehman and I agree that regardless of the number of partitions on the disk, libparted should be consistently reporting the correct starting sector for free space.  This is disk label dependent.  msdos and pc98 (maybe?) should start at sector 63 since everything before that is reserved for the MBR.  Other disk labels will have other conditions.

What I have done is created a patch for parted that creates a new PedDiskOps function that is per disk label.  It's called 'get_start_sector'.  For msdos and pc98, I'm returning 63.  For all other disk labels, I am returning 0, which is consistent with what _disk_alloc_freespace() was doing before.

Now, how come libparted is doing the correct thing when there is actually a partition or two on the disk (the correct thing being returning sector 63 for the first available free space)?  The problem is buried in the msdos_alloc_metadata() function in libparted/labels/dos.c.  It seems that metadata allocation actually happens once all partitions are removed, but it's not set up correctly when part_list contains FREESPACE partitions.

So, I see two ways to correct this problem.  Hack around it using the patch I've provided.  Or figure out why msdos_alloc_metadata() is not setting up placeholders correctly with no partitions on the disk.