[parted-devel] [PATCH] libparted: Fixed bug in initializing and re-reading partition table of FBA devices

[Nageswara R Sastry]

On 10/26/2013 1:34 AM, Phillip Susi wrote:

Thanks for the reply.

> Are you saying that this kind of disk can not really be partitioned (
> since the kernel always reports exactly one partition for the whole
> disk )?  If so then it does not make much sense to run parted in the
> first place.
Yes, this kind of disk can not be partitioned and having one partition
for the whole disk.
Parted tool's DASD code correctly identifies this kind of disk and the
partition. This helps in successful installation of an Operating System
on the top of FBA disk(s).

> It looks like your patch asks the kernel to throw out the partition
> table that parted has just installed with BLKPG and go back to that
> single dummy partition.
FBA disk does not contain a partition table to be thrown/created by
the BLKPG ioctl. With out BLKRRPART ioctl code, kernel is not aware of
the implicit/single partition available with FBA disk.

*With out patch:*
FBA disk having an implicit partition will not be visible to kernel.

For example: before running parted tool, the /dev entries looks like,
Assuming FBA disk is 'dasda'.
 # ls /dev/dasda*
   /dev/dasda  /dev/dasda1

After running parted tool, the /dev entries looks like,
 # ls /dev/dasda*
   /dev/dasda

*With patch:*
FBA disk having an implicit partition will be visible to kernel.

For example: before and after running parted tool, the /dev/ entries
looks like,
Assuming FBA disk is 'dasda'.
 # ls /dev/dasda*
   /dev/dasda  /dev/dasda1


> Also if you are going to retry the ioctl a few times, you should have
> a small delay between each attempt.
>From my colleagues learned that, time delay for all the iterations together
should not cross '2 seconds'. Is that acceptable? or do you have a different
number.

Thank you.

Regards
R.Nageswara Sastry