loopdev regression with v2.23

loopdev regression with v2.23

[Dave Reisner]

Hi,

I have a bug report that points out that v2.23 has disallowed losetup of
a block device with an offset:

https://bugs.archlinux.org/task/35193

As I also note in the bug report, I've bisected this to commit 293714c0d15.

I'm not clear on what the proper fix is here -- "expected_size" is set
to 0 since the fstat call is performed on a special file which has a 0
size. Is the right thing to do simply return 0 if the size of the
"backing file" is 0?

There's also the problem of the error message -- the fabricated ERANGE
return fails to also set errno=ERANGE before returning.

Cheers,
Dave

Re: loopdev regression with v2.23

[Karel Zak]

On Sat, May 11, 2013 at 07:39:10PM -0400, Dave Reisner wrote:
> I have a bug report that points out that v2.23 has disallowed losetup of
> a block device with an offset:
> 
> https://bugs.archlinux.org/task/35193
> 
> As I also note in the bug report, I've bisected this to commit 293714c0d15.
> 
> I'm not clear on what the proper fix is here -- "expected_size" is set
> to 0 since the fstat call is performed on a special file which has a 0
> size. Is the right thing to do simply return 0 if the size of the
> "backing file" is 0?

The right thing is to use size of the block device if the backing file
is a block device. Fixed. Git pull.

All this is workaround for kernel bug (already fixed in v3.9). I hope
that one day we will use KERNEL_VERSION macro to exclude all this
painful thing :-)

Thanks!

    Karel

-- 
 Karel Zak  <kzak@redhat.com>
 http://karelzak.blogspot.com