md/raid devices don't show up in /proc/partitions in 2.6 :-(

md/raid devices don't show up in /proc/partitions in 2.6 :-(

[Neil Brown]

I just noticed that md devices do not show up in /proc/partitions in
2.6.

I realise that they don't actually have partitions and are just 'whole
devices', but other whole devices do appear in /proc/partitions (Along
with their partitions if any).

This seems to be a regression from 2.4 where md devices do appear in
/proc/partitions.

The cause appears to be a patch from 'torvalds' some 15 months ago
(in version 1.36 for drivers/block/genhd.c) which has the comment:

      Avoid confusion "mount" and "fsck" - don't show things like
      floppies and CD's in /proc/partitions.

It excluded devices that cannot be partitioned, and devices with zero
size from /proc/partitions.

The 'zero size' possibly makes sense (2.4 excludes those), but I would
like to register a vote against excluding devices without partitions,
as this excluded 'md' devices and I would really like them to be
included.


Is there really a good reason for this?  How badly does mount get
confused? and is that not the fault of mount?

NeilBrown

Re: md/raid devices don't show up in /proc/partitions in 2.6 :-(

[Lincoln Dale]

At 11:15 AM 27/11/2003, Neil Brown wrote:
>I just noticed that md devices do not show up in /proc/partitions in
>2.6.
[..]

this also caused no end of badness trying to migrate from LVM2 (Linux 2.4) 
to DM+LVM2 (Linux 2.6) where the major/minor numbers changed.

while the real fix is to educate lvmcreate_initrd to be more intelligent (i 
had to create a new one which includes 
mknod/devmap_mknod.sh/mkdir/sed/rm/cat/lvdisplay binaries), it is certainly 
a big trap for the unwary that happen to use LVM1/LVM2 as a root volume.


cheers,

lincoln.

Re: md/raid devices don't show up in /proc/partitions in 2.6 :-(

[Lars Marowsky-Bree]

On 2003-11-27T11:15:10,
   Neil Brown <neilb@cse.unsw.edu.au> said:

> I just noticed that md devices do not show up in /proc/partitions in
> 2.6.

Using /proc/partitions for finding the existing block devices /
partitions does seem to be kind of obsolete, as sysfs exports all of
this information too?

So, if anything, I'd drop the confusing redundancy and kill
/proc/partitions - the names there may have little resemblance to how
udev (et al) chose to name the device node in the filesystem anyway.


Sincerely,
    Lars Marowsky-Brée <lmb@suse.de>

-- 
High Availability & Clustering	      \ ever tried. ever failed. no matter.
SUSE Labs			      | try again. fail again. fail better.
Research & Development, SUSE LINUX AG \ 	-- Samuel Beckett