On Mon, Jun 20, 2011 at 08:30:40PM -0400, Kenneth R Westerback wrote:
> I committed a fix to fdisk(8) today to un-break the -i and -e options
> on 4096-byte devices. To make a long story short, it had been working
> accidentally until I committed a 4.9 change to fdisk(8) to make
> it pay attention to the errors returned from MBR_read().
>
> However this has raised once more concerns about what is going to
> happen when 4096-byte sector devices become common, then the norm,
> and then the smallest disks available.
>
> There has been a lot of work done in the last few years to de-couple
> the internal kernel view of a disk (a series of 512 byte blocks)
> and the 'real world' view of potentially different sector size
> devices. With seemless translation being done behind the scenes.
>
> However, as today has shown, there may well be further unreconstructed
> code making invalid assumptions or currently silently working
> accidentally and waiting for the day when it can blow up your
> machine.
>
> So if you have such a device (disks >3TB are your best bet) I would
> be very interested in hearing how hard you have to work to break
> something. Of course the clever can also create vnd's with such
> large sectors but actual hardware is more convincing.
>
> Various filesystems (supported ones only please!), utilities such as
> fdisk, disklabel, newfs, fsck, dump, restore, etc. Anything which
> may do any serious disk i/o or is suspected of attempting raw, low
> level i/o.
It just occurred to me that softraid crypto would benefit quite a bit
from 4096 byte sectors. And... we can use softraid to force 4096 byte
sectors pretty trivially so it can be used a test harness as well in
lieu of 3096 byte sector disks.
>
> Ken
