[Libguestfs] guestmount failing to mount a qcow image

[Pádraig Brady]

On 12/08/2011 05:43 PM, Richard W.M. Jones wrote:
> On Thu, Dec 08, 2011 at 04:49:08PM +0000, Pádraig Brady wrote:
>> # guestmount --rw -a /tmp/disk -i /tmp/tmplFlBCc
>> libguestfs: error: mount_options: mount_options_stub: /dev/root: No
>> such file or directory
> 
> This image works OK with libguestfs 1.14/1.15.  I suspect the commit
> you need to make it work is:
> 
>   commit 917f947590c92318fee2545ba88245d0de012e31
>   Author: Richard W.M. Jones <rjones at redhat.com>
>   Date:   Fri Sep 16 15:26:20 2011 +0100
> 
>     inspection: Handle /dev/root in /etc/fstab.
>     
>     This means "the device that holds /etc/fstab", so map it correctly.
>     
>     This fixes support for ttylinux and also some other guests that use
>     /dev/root instead of a real device name.
> 
> The bigger issue here for OpenStack is how to recover gracefully when
> inspection fails.  Even though guestmount -i doesn't work on the older
> libguestfs, you could still have used explicit -m option(s).

Yep that works:  `guestmount -a annoying.qcow2 -m /dev/sda mount_dir/`

The awkward thing about the failure mode is that it warns,
but returns EXIT_SUCCESS, leaving the dir "half mounted"
(and then fusermount -u gives an EBUSY error (hmm I might use -z))
I'm tempted to assume an error if there is any output to stderr.

> You would have to know somehow which filesystem(s) are available and which
> you want to modify.  That is information you can get from the API
> (g.list_filesystems) or using virt-filesystems.  It's reasonable, if
> there is only one filesystem in the image, to just mount that one.
> What's less clear is what should happen if there are multiple
> filesystems: that information would have to be passed down from the
> user through OpenStack.

There is a 'partition' variable passed through the api,
I guess it could be 0 = /dev/sda, 1..N = /dev/sdaN, -1 => -i

I'll sort something out.

cheers,
Pádraig.