[Patch v2 3/4] store the default dump target info when default target is a separate disk

[Dave Young]

On 03/19/14 at 09:09am, Vivek Goyal wrote:
> On Wed, Mar 19, 2014 at 10:15:28AM +0800, Baoquan He wrote:
> > On 03/18/14 at 04:20pm, Vivek Goyal wrote:
> > > On Tue, Mar 18, 2014 at 02:35:39PM +0800, Baoquan He wrote:
> > > > On 03/17/14 at 11:16am, Vivek Goyal wrote:
> > > >  
> > > > > > +
> > > > > > +        if [ "$_fstype" = "nfs" ];then
> > > > > > +            kdump_install_net "$_target"
> > > > > > +        else
> > > > > > +            _target=$(kdump_to_udev_name $_target)
> > > > > > +        fi
> > > > > 
> > > > > Ok, we seem to have this functionality to convert to persistent names
> > > > > both in mkdumprd as well as dracut-module-sh. How about coming up with
> > > > > a common function in kudmp-lib.sh and use that at both the places. This
> > > > > is a cleanup and can go in a separate patch.
> > > > 
> > > > Here I don't understand, could you say more or give an example?
> > > 
> > > We convert UUID/LABEL/dev to respective persistent device name. And
> > > use get_persistent_dev() function ultimately.
> > > 
> > > Now there are two implementations of get_persistent_dev(). One in
> > > mkdumprd() and other in dracut. I think mkdumprd version came from
> > > dracut version.
> > > 
> > > So point being, if there is any need to convert to persistent names,
> > > then that logic should be shared between mkdumprd and dracut-module.sh.
> > > 
> > > We should not be using two separate functions to do the converstion.
> > > 
> > > If we define common set of functions in kdump-lib.sh and use that
> > > at both the places, code will become simpler, easy to understand
> > > and less error prone.
> > > 
> > > I realize that's little more intrisive and bigger cleanup. So I will
> > > open a bug to keep track of this cleanup. You can take it up after
> > > your other changes have been committed. I don't want your changes to
> > > block behind this bigger change.
> > 
> > Dave took the oldest version of this function to mkdumprd, now Harald
> > has changed it several times. 
> > 
> > If define it in kdump-lib.sh, any change or modification to that of
> > dracut version will not benifit dracut-module-setup.sh any more. This is
> > my worry.
> 
> I don't like two different versions of get_persistent_dev(). I am
> wondering why do we need to use get_persistent_dev() in mkdumprd at all?
> 
> Will dracut not translate device names into persistent names
> automatically. If yes, then we can probably get rid of
> get_persistent_dev() logic from mkdumprd and use the one available from
> dracut in dracut-module-setup.sh.

Dracut will put the --mount fstab lines in initramfs without any change
except the fsck default value.

The code in mkdumprd is for creating udev dev name in initramfs /etc/fstab
Moving it to dracus.sh is doable via replacing the 1st field of --mount
with $(get_presistent_dev $dev) before appending the line to fstab file.

Thanks
Dave