Re: [RFC PATCH] /sys/block -> /sys/class/block (Fedora 3 & 4 testers wanted)

[Cornelia Huck <cornelia.huck@de.ibm.com>]

On Tue, 22 May 2007 17:32:55 -0700,
Greg KH <greg@kroah.com> wrote:

> On Tue, May 22, 2007 at 06:28:12PM +0200, Cornelia Huck wrote:
> > On Tue, 22 May 2007 10:25:01 +0200,
> > Cornelia Huck <cornelia.huck@de.ibm.com> wrote:
> > 
> > > I tried this on one of our internal drivers, which is based on FC 3 (or
> > > 4, can look this up). With s390 defconfig, it is unable to open the
> > > root device /dev/dasda1 (which is unsurprising, considering udev (063)
> > > decided to create /dev/dasda(1) as a character node instead of a block
> > > node).
> > 
> > Just to be clear, it's fsck that complains:
> > 
> > Checking filesystems  
> > Checking all file systems.  
> > [/sbin/fsck.ext3 (1) -- /] fsck.ext3 -a /dev/dasda1   
> > fsck.ext3: No such device or address while trying to open /dev/dasda1   
> > Possibly non-existent or swap device?  
> > [FAILED]   
> > 
> > (so that is after udev has been started and obviously dasda1 could be
> > accessed)
> 
> But /dev/dasda1 isn't present?  Or why would fsck be complaining here?

See above. /dev/dasda1 exists, but strangely as a character device,
which makes fsck unhappy.

> 
> > > When I look at the system, /sys/block/ and /sys/class/block/ look sane
> > > at first glance (working on a 3270 console is usally PITA...)
> > > 
> > > Even more surprisingly, the system comes up fine (once I added ptmx to
> > > makedev.d/) with CONFIG_SYSFS_DEPRECATED not set. /sys/block/
> > > and /sys/class/block/ look just like expected. (Unfortunately, our
> > > lsdasd tool breaks with this...)
> > > 
> > > I'll see if I can find out more.
> > 
> > I currently have the inkling that
> > 
> > lrwxrwxrwx  1 root root    0 May 22 15:59 block:dasda1-> ../../../class/block/dasda1
> > 
> > in /sys/block/dasda/ is the culprit.
> 
> Why?

See the paragraph just below, but it's more like a hunch currently.

> 
> > In all other versions I've tried (without CONFIG_SYSFS_DEPRECATED, and
> > on an older kernel with and without CONFIG_SYSFS_DEPRECATED), it's
> > always dasda1.
> 
> That's just a back symlink, the real device should still be there.

It is, but somehow udev seems confused.

> 
> Oh, and yes, you will probably have to have CONFIG_SYSFS_DEPRECATED
> enabaled to have a chance for these to work on Fedora 4 or 3.

Hmpf. That's just the problem :)

- enable CONFIG_SYSFS_DEPRECATED: fails as described
- disable CONFIG_SYSFS_DEPRECATED: works

If one of the two was to fail, I'd rather expect it the other way
around...

> 
> > I can continue investigating tomorrow, unless someone has a good idea :)
> 
> I don't, but any information you can find out would be greatly
> appreciated, especially as it seems like it is working, but something is
> still a little bit wrong.

It's not really according to my definition of "working" :) But yes,
I'll see what I can do.