rescan partitions returns EIO since 2.6.8

rescan partitions returns EIO since 2.6.8

[Andries.Brouwer]

Martin Pool changed the behaviour of the BLKRRPART ioctl in 2.6.8.
The effect is that one now gets an I/O error when first
partitioning an empty disk:

# sfdisk /dev/sda
Checking that no-one is using this disk right now ...
BLKRRPART: Input/output error

Ugly. I am tempted to go back to the state before his patch.

Why was this patch made? Just something random that seemed like
a good idea? Is there software that needs it?

Andries


The guilty patch (called "lost error code"):

	if (!get_capacity(disk) || !(state = check_partition(disk, bdev)))
-          return 0;
+          return -EIO;

Re: rescan partitions returns EIO since 2.6.8

[Martin Pool]

On  5 Dec 2004, Andries.Brouwer@cwi.nl wrote:
> Martin Pool changed the behaviour of the BLKRRPART ioctl in 2.6.8.
> The effect is that one now gets an I/O error when first
> partitioning an empty disk:

> # sfdisk /dev/sda
> Checking that no-one is using this disk right now ...
> BLKRRPART: Input/output error

To me it seems more correct that a request to read the partition table
should fail if the partition table can't be read.  I had some code
that did care to know the difference, but if you really want to roll
it back I won't object.

fdisk, cfdisk and parted all just give a warning in this case.  You
can tell sfdisk to ignore the error with --no-reread.

-- 
Martin 

Re: rescan partitions returns EIO since 2.6.8

[Andries Brouwer]

On Mon, Dec 06, 2004 at 11:47:22AM +1100, Martin Pool wrote:
> On  5 Dec 2004, Andries.Brouwer@cwi.nl wrote:
> > Martin Pool changed the behaviour of the BLKRRPART ioctl in 2.6.8.
> > The effect is that one now gets an I/O error when first
> > partitioning an empty disk:
> 
> > # sfdisk /dev/sda
> > Checking that no-one is using this disk right now ...
> > BLKRRPART: Input/output error
> 
> To me it seems more correct that a request to read the partition table
> should fail if the partition table can't be read.

I do not view BLKRRPART as a request to read the partition table.
It is a request to revalidate: "if the disk is in use, return EBUSY,
otherwise, discard any old information, read any new information".
If the disk is blank then there is no new information to read,
that is not an error, and certainly not an I/O error.

> if you really want to roll it back I won't object.

Yes, I think I want to - am getting worried mail from people
who very much dislike I/O errors on a brand-new disk.

Andries

Re: rescan partitions returns EIO since 2.6.8

[Martin Pool]

On  6 Dec 2004, Andries Brouwer <Andries.Brouwer@cwi.nl> wrote:
> On Mon, Dec 06, 2004 at 11:47:22AM +1100, Martin Pool wrote:
> > On  5 Dec 2004, Andries.Brouwer@cwi.nl wrote:
> > To me it seems more correct that a request to read the partition table
> > should fail if the partition table can't be read.
> 
> I do not view BLKRRPART as a request to read the partition table.
> It is a request to revalidate: "if the disk is in use, return EBUSY,
> otherwise, discard any old information, read any new information".
> If the disk is blank then there is no new information to read,
> that is not an error, and certainly not an I/O error.
> 
> > if you really want to roll it back I won't object.
> 
> Yes, I think I want to - am getting worried mail from people
> who very much dislike I/O errors on a brand-new disk.

OK, my mistake.  Please take it out.

-- 
Martin