[pve-devel] [PATCH qemu-server 2/2] map cloudinit disk to new vmid on restore

[Thomas Lamprecht]

On 5/13/19 2:01 PM, Mira Limbeck wrote:
> since the restore is now working but does not map the disk to the new
> vmid, this patch solves that. it allocates the new disk with vdisk_alloc
> to get the new disk name.
> the regex to check if it could be a cloudinit config line is limited to
> ide|sata|scsi because those are the only ones supported for cdrom
> drives. the following 'parse_drive' is so we can use
> 'drive_is_cloudinit' instead of having to use a custom regex for the
> is_cloudinit check.
> the format on restore is kept if the underlying storage supports it,
> otherwise the default for that storage is used.
> 
> this should fix #1807 completely. the restore error was already resolved
> with commit 7e8ab2a, but the vmid of the disk might not have matched the new
> one.

looks OK in general, some minor comment inline.

> 
> Signed-off-by: Mira Limbeck <m.limbeck at proxmox.com>
> ---
>  PVE/QemuServer.pm | 29 ++++++++++++++++++++++++++---
>  1 file changed, 26 insertions(+), 3 deletions(-)
> 
> diff --git a/PVE/QemuServer.pm b/PVE/QemuServer.pm
> index 9d560ec..06329bd 100644
> --- a/PVE/QemuServer.pm
> +++ b/PVE/QemuServer.pm
> @@ -6264,6 +6264,22 @@ sub restore_vma_archive {
>  		$storage_limits{$storeid} = $bwlimit;
>  
>  		$virtdev_hash->{$virtdev} = $devinfo->{$devname};
> +	    } elsif ($line =~ m/^((?:ide|sata|scsi)\d+):\s*(.*)\s*$/) {
> +		my $virtdev = $1;
> +		my $drive = parse_drive($virtdev, $2);
> +		if (drive_is_cloudinit($drive)) {
> +		    my ($storeid, $volname) = PVE::Storage::parse_volume_id($drive->{file});
> +		    my $scfg = PVE::Storage::storage_config($cfg, $storeid);
> +		    my $format = qemu_img_format($scfg, $volname); # has 'raw' fallback
> +
> +		    my $d = {};
> +		    $d->{format} = $format;
> +		    $d->{storeid} = $opts->{storage} // $storeid;
> +		    $d->{size} = $PVE::QemuServer::Cloudinit::CLOUDINIT_DISK_SIZE;

do we need to set it, or wouldn't commit_ci just set it for us once the VM gets
actually started again? 

> +		    $d->{file} = $drive->{file}; # to make drive_is_cloudinit check possible
> +		    $d->{name} = "vm-$vmid-cloudinit";

maybe a direct initialization alá

my $d = {
    format => $format,
    storeid => $opts->{storage} // $storeid,
    ...,
    name => "vm-$vmid-cloudinit",
};

could look a bit cleaner, but naturally just IMO :-)

> +		    $virtdev_hash->{$virtdev} = $d;
> +		}
>  	    }
>  	}
>  
> @@ -6334,8 +6350,12 @@ sub restore_vma_archive {
>  	    my $supported = grep { $_ eq $d->{format} } @$validFormats;
>  	    $d->{format} = $defFormat if !$supported;
>  
> +	    my $name = $d->{name} // undef;

// is a check for defined-ness, so above is a no-op?

> +	    if ($name && $d->{format} ne 'raw') {
> +		$name .= ".$d->{format}";
> +	    }
>  	    my $volid = PVE::Storage::vdisk_alloc($cfg, $storeid, $vmid,
> -						  $d->{format}, undef, $alloc_size);
> +						  $d->{format}, $name, $alloc_size);
>  	    print STDERR "new volume ID is '$volid'\n";
>  	    $d->{volid} = $volid;
>  	    my $path = PVE::Storage::path($cfg, $volid);
> @@ -6347,9 +6367,12 @@ sub restore_vma_archive {
>  		$write_zeros = 0;
>  	    }
>  
> -	    print $fifofh "${map_opts}format=$d->{format}:${write_zeros}:$d->{devname}=$path\n";
> +	    my $is_cloudinit = defined($d->{file}) && drive_is_cloudinit($d);
> +	    if (!$is_cloudinit) {
> +		print $fifofh "${map_opts}format=$d->{format}:${write_zeros}:$d->{devname}=$path\n";
>  
> -	    print "map '$d->{devname}' to '$path' (write zeros = ${write_zeros})\n";
> +		print "map '$d->{devname}' to '$path' (write zeros = ${write_zeros})\n";
> +	    }
>  	    $map->{$virtdev} = $volid;
>  	}
>  
>