pvops xen_blkfront does not enforce device names

pvops xen_blkfront does not enforce device names

[Olaf Hering]

Why does xen_blkfront not enforce kernel device names from domU.cfg?
In my PV domU.cfg I still have something like this:
disk=[ 'file:/path,hda,w' ]

With pvops and xen_blkfront I get xvda.
With xenlinux and xenblk I get hda.

Olaf

_______________________________________________
Xen-devel mailing list
Xen-devel@lists.xen.org
http://lists.xen.org/xen-devel

Re: pvops xen_blkfront does not enforce device names

[Konrad Rzeszutek Wilk]

On Tue, Apr 26, 2016 at 03:28:45PM +0200, Olaf Hering wrote:
> Why does xen_blkfront not enforce kernel device names from domU.cfg?
> In my PV domU.cfg I still have something like this:
> disk=[ 'file:/path,hda,w' ]
> 
> With pvops and xen_blkfront I get xvda.
> With xenlinux and xenblk I get hda.

It has no business subverting the hda device names which
are specifically for legacy devices.

[We got smacked down at some point for having done this]

> 
> Olaf
> 
> _______________________________________________
> Xen-devel mailing list
> Xen-devel@lists.xen.org
> http://lists.xen.org/xen-devel

_______________________________________________
Xen-devel mailing list
Xen-devel@lists.xen.org
http://lists.xen.org/xen-devel

Re: pvops xen_blkfront does not enforce device names

[Jan Beulich]

>>> On 29.04.16 at 08:45, <konrad.wilk@oracle.com> wrote:
> On Tue, Apr 26, 2016 at 03:28:45PM +0200, Olaf Hering wrote:
>> Why does xen_blkfront not enforce kernel device names from domU.cfg?
>> In my PV domU.cfg I still have something like this:
>> disk=[ 'file:/path,hda,w' ]
>> 
>> With pvops and xen_blkfront I get xvda.
>> With xenlinux and xenblk I get hda.
> 
> It has no business subverting the hda device names which
> are specifically for legacy devices.

That's one view. The other is that people moving from XenoLinux to
pv-ops are in trouble without it. How did you (or Citrix, if anyone of
them looks) deal with this when doing the transition?

> [We got smacked down at some point for having done this]

In which way?

Jan


_______________________________________________
Xen-devel mailing list
Xen-devel@lists.xen.org
http://lists.xen.org/xen-devel

Re: pvops xen_blkfront does not enforce device names

[Konrad Rzeszutek Wilk]

On Fri, Apr 29, 2016 at 01:29:17AM -0600, Jan Beulich wrote:
> >>> On 29.04.16 at 08:45, <konrad.wilk@oracle.com> wrote:
> > On Tue, Apr 26, 2016 at 03:28:45PM +0200, Olaf Hering wrote:
> >> Why does xen_blkfront not enforce kernel device names from domU.cfg?
> >> In my PV domU.cfg I still have something like this:
> >> disk=[ 'file:/path,hda,w' ]
> >> 
> >> With pvops and xen_blkfront I get xvda.
> >> With xenlinux and xenblk I get hda.
> > 
> > It has no business subverting the hda device names which
> > are specifically for legacy devices.
> 
> That's one view. The other is that people moving from XenoLinux to
> pv-ops are in trouble without it. How did you (or Citrix, if anyone of
> them looks) deal with this when doing the transition?

We had transition folks to use 'UUID' or volume group names in their 'root='
entry so never had an issue with this.
> 
> > [We got smacked down at some point for having done this]
> 
> In which way?

For usurping the 'hda'. Stefano may remember the details more..
> 
> Jan
> 

_______________________________________________
Xen-devel mailing list
Xen-devel@lists.xen.org
http://lists.xen.org/xen-devel

Re: pvops xen_blkfront does not enforce device names

[Stefano Stabellini]

On Fri, 29 Apr 2016, Konrad Rzeszutek Wilk wrote:
> On Fri, Apr 29, 2016 at 01:29:17AM -0600, Jan Beulich wrote:
> > >>> On 29.04.16 at 08:45, <konrad.wilk@oracle.com> wrote:
> > > On Tue, Apr 26, 2016 at 03:28:45PM +0200, Olaf Hering wrote:
> > >> Why does xen_blkfront not enforce kernel device names from domU.cfg?
> > >> In my PV domU.cfg I still have something like this:
> > >> disk=[ 'file:/path,hda,w' ]
> > >> 
> > >> With pvops and xen_blkfront I get xvda.
> > >> With xenlinux and xenblk I get hda.
> > > 
> > > It has no business subverting the hda device names which
> > > are specifically for legacy devices.
> > 
> > That's one view. The other is that people moving from XenoLinux to
> > pv-ops are in trouble without it. How did you (or Citrix, if anyone of
> > them looks) deal with this when doing the transition?
> 
> We had transition folks to use 'UUID' or volume group names in their 'root='
> entry so never had an issue with this.
> > 
> > > [We got smacked down at some point for having done this]
> > 
> > In which way?
> 
> For usurping the 'hda'. Stefano may remember the details more..

The argument goes as follow: 'hda' as 'xvda' or 'sda' are Linux specific
names. A different operating system could name its devices 'goofy', or
any other made up names. The Xen toolstack has no business in choosing
the naming scheme for the guest kernel to use. Moreover xl can do
nothing to enforce it: what if the OS stubbornly keeps calling its
devices 'goofy' even if according to xl should be 'hda'? It is just an
internal OS representation, so the OS can do as it wishes.

Linux doesn't want to call PV disks anything else than xvd*. After all
they are PV disks, aren't they? And Xen cannot do anything about it.

_______________________________________________
Xen-devel mailing list
Xen-devel@lists.xen.org
http://lists.xen.org/xen-devel

Re: pvops xen_blkfront does not enforce device names

[David Vrabel]

On 29/04/16 08:29, Jan Beulich wrote:
>>>> On 29.04.16 at 08:45, <konrad.wilk@oracle.com> wrote:
>> On Tue, Apr 26, 2016 at 03:28:45PM +0200, Olaf Hering wrote:
>>> Why does xen_blkfront not enforce kernel device names from domU.cfg?
>>> In my PV domU.cfg I still have something like this:
>>> disk=[ 'file:/path,hda,w' ]
>>>
>>> With pvops and xen_blkfront I get xvda.
>>> With xenlinux and xenblk I get hda.
>>
>> It has no business subverting the hda device names which
>> are specifically for legacy devices.
> 
> That's one view. The other is that people moving from XenoLinux to
> pv-ops are in trouble without it. How did you (or Citrix, if anyone of
> them looks) deal with this when doing the transition?

XenServer always named its PV disks as xvda etc.

David

_______________________________________________
Xen-devel mailing list
Xen-devel@lists.xen.org
http://lists.xen.org/xen-devel

Re: pvops xen_blkfront does not enforce device names

[Andrew Cooper]

On 29/04/16 10:39, David Vrabel wrote:
> On 29/04/16 08:29, Jan Beulich wrote:
>>>>> On 29.04.16 at 08:45, <konrad.wilk@oracle.com> wrote:
>>> On Tue, Apr 26, 2016 at 03:28:45PM +0200, Olaf Hering wrote:
>>>> Why does xen_blkfront not enforce kernel device names from domU.cfg?
>>>> In my PV domU.cfg I still have something like this:
>>>> disk=[ 'file:/path,hda,w' ]
>>>>
>>>> With pvops and xen_blkfront I get xvda.
>>>> With xenlinux and xenblk I get hda.
>>> It has no business subverting the hda device names which
>>> are specifically for legacy devices.
>> That's one view. The other is that people moving from XenoLinux to
>> pv-ops are in trouble without it. How did you (or Citrix, if anyone of
>> them looks) deal with this when doing the transition?
> XenServer always named its PV disks as xvda etc.

The toolstack has no business trying to control how the guest chooses to
name its block devices.  The device name is private information to the
guest.

It should continue to function in exactly the same way even if the guest
admin edits their udev rules to rename it to something crazy like "&&&&".

~Andrew

_______________________________________________
Xen-devel mailing list
Xen-devel@lists.xen.org
http://lists.xen.org/xen-devel

Re: pvops xen_blkfront does not enforce device names

[Olaf Hering]

On Fri, Apr 29, Konrad Rzeszutek Wilk wrote:

> On Tue, Apr 26, 2016 at 03:28:45PM +0200, Olaf Hering wrote:
> > Why does xen_blkfront not enforce kernel device names from domU.cfg?
> > In my PV domU.cfg I still have something like this:
> > disk=[ 'file:/path,hda,w' ]
> > 
> > With pvops and xen_blkfront I get xvda.
> > With xenlinux and xenblk I get hda.
> 
> It has no business subverting the hda device names which
> are specifically for legacy devices.

Old IDE drives are not SCSI either, but now they appear as 'sd' since a
while.

'hd' or 'sd' or 'mmc' or 'xvd' are just names. They do not represent
hardware as such. Otherwise the interface into the blocklayer to request
major/minor/name would be like 'hey, I'm the IDE/SCSI/MMC/whatever
driver, give me something'.

Anyway, since this is now set in stone, any VM out there which still is
configured to use /dev/hd* has to be adjusted to /dev/xvd* upfront when
going from xenlinux to pvops.

Olaf

_______________________________________________
Xen-devel mailing list
Xen-devel@lists.xen.org
http://lists.xen.org/xen-devel

Re: pvops xen_blkfront does not enforce device names

[Konrad Rzeszutek Wilk]

On Fri, Apr 29, 2016 at 01:27:35PM +0200, Olaf Hering wrote:
> On Fri, Apr 29, Konrad Rzeszutek Wilk wrote:
> 
> > On Tue, Apr 26, 2016 at 03:28:45PM +0200, Olaf Hering wrote:
> > > Why does xen_blkfront not enforce kernel device names from domU.cfg?
> > > In my PV domU.cfg I still have something like this:
> > > disk=[ 'file:/path,hda,w' ]
> > > 
> > > With pvops and xen_blkfront I get xvda.
> > > With xenlinux and xenblk I get hda.
> > 
> > It has no business subverting the hda device names which
> > are specifically for legacy devices.
> 
> Old IDE drives are not SCSI either, but now they appear as 'sd' since a
> while.
> 
> 'hd' or 'sd' or 'mmc' or 'xvd' are just names. They do not represent
> hardware as such. Otherwise the interface into the blocklayer to request
> major/minor/name would be like 'hey, I'm the IDE/SCSI/MMC/whatever
> driver, give me something'.
> 
> Anyway, since this is now set in stone, any VM out there which still is
> configured to use /dev/hd* has to be adjusted to /dev/xvd* upfront when
> going from xenlinux to pvops.

You may want to adjust it to use UUID. It really makes it easier in case
they also want to swap the disks arounds, etc.
> 
> Olaf

_______________________________________________
Xen-devel mailing list
Xen-devel@lists.xen.org
http://lists.xen.org/xen-devel