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

Re: anaconda [parted] always overwrites the first and last 9k of a disk



Hi,

On 02/18/2010 04:46 PM, Eric Sandeen wrote:
Hans de Goede wrote:
Hi,

While investigating bug 561976 (fedora) / bug 566181 (rhel) I've found out
that anaconda always overwrites the first and last 9k of a disk (unless
the disk is completely unused by the installation).

This is caused by a bug in libparted's ped_disk_duplicate() function which
anaconda uses through pyparted (and parted itself does not use).

I'm using this wide "broadcast" mechanism to make people aware of this bug,
as it might very well explain some issues being seen with LVM too, if a
PV partition is at the end of the disk, this bug might corrupt the LVM
metadata.

Note this bug is present in older (F-12 and F-11) parted versions too! We
may not be hitting it in F-11 (not sure if we are using disk duplication
there), but we are definitely hitting this in F-12 too.

Under what circumstances does it do this?  In some circumstances this may
be correct; if parted is changing what the disk "is" it should probably be
wiping out old signatures that tools like udev or blkid may find... for
example old md superblocks at the end of the disk...


As discussed on irc, it was doing this under very wrong circumstances, for
example when only changing the type of a partition.

But it only did this when used in the way anaconda uses it, the cmdline utility is
"safe".

Regards,

Hans


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