[pve-devel] [PATCH storage v6 1/7] plugin: allow volume import of iso, snippets, vztmpl and import

[Fiona Ebner]

Am 13.02.25 um 18:21 schrieb Fiona Ebner:
> Am 20.01.25 um 12:28 schrieb Filip Schauer:
>> Extend volume import functionality to support 'iso', 'snippets',
>> 'vztmpl', and 'import' types, in addition to the existing support for
>> 'images' and 'rootdir'. This is a prerequisite for the ability to move
>> ISOs, snippets and container templates between nodes.
>>
>> Existing behavior for importing VM disks and container volumes remains
>> unchanged.
>>
>> Signed-off-by: Filip Schauer <f.schauer at proxmox.com>
>> ---
>>  src/PVE/Storage/Plugin.pm | 72 ++++++++++++++++++++++++++-------------
>>  1 file changed, 48 insertions(+), 24 deletions(-)
>>
>> diff --git a/src/PVE/Storage/Plugin.pm b/src/PVE/Storage/Plugin.pm
>> index 65cf43f..b682362 100644
>> --- a/src/PVE/Storage/Plugin.pm
>> +++ b/src/PVE/Storage/Plugin.pm
>> @@ -1738,6 +1738,8 @@ sub volume_export_formats {
>>  	my $format = ($class->parse_volname($volname))[6];
>>  	my $size = file_size_info($file, undef, $format);
>>  
>> +	return ('raw+size') if !defined($format);
> 
> Can we rather make this explicitly check for vtype being one of the ones
> we care about? That reflects more precisely what we want to test for.
> I'm also thinking whether we want to explicitly include the vtype in the
> new export format name to avoid potential type confusion down the line.
> I.e. if a storage wants to export an iso, it should not be importable as
> a snippet on the other side. We can keep the current formats for
> backwards-compat, but the formats for the newly supported vtypes could
> then be $vtype+meta, similar to the backup+size type you add. This would
> be more explicit and less likely to be(come) exploitable. And we could
> write the vtype into the header of the stream itself for extra safety.
> Maybe in a way that allows for some forwards-compatibility, e.g. what if
> ISO files get notes too at some point.
> 
> Just a crude outline what the stream could be from the top of my head:
> number of meta-properties
> for each meta-property: length of property name, property name, length
> of property data, property data
> length of volume, volume data

Or more easily
length of JSON with metadata, JSON with metadata
length of volume, volume data

> 
> For compatibility/extensibility, the stream consumer can then handle all
> meta properties it knows and skip the rest (with some informational
> logging about what was skipped). Or we can add a marker to each property
> whether it is required or optional and mark all important properties
> that should rather lead to failure than skipping. Then outdated nodes
> might not be able to receive certain streams, but we have the
> flexibility to decide for each property.
> 
> I'm sorry, it would expand the scope of the series a bit more, but I do
> think this will save us some headaches and give us more flexibility in
> the future. And it seems to be a natural generalization of what you
> already do for backup+size.
> 
> @other devs: opinions?
>