[pve-devel] [RFC v10 qemu-server/manager] API for disk import and OVF
Fabian Grünbichler
f.gruenbichler at proxmox.com
Tue Jan 18 11:19:41 CET 2022
On January 18, 2022 10:08 am, Fabian Ebner wrote:
> Am 17.01.22 um 16:43 schrieb Fabian Grünbichler:
>> On January 13, 2022 11:08 am, Fabian Ebner wrote:
>>> Extend qm importdisk/importovf functionality to the API.
>>>
>>>
>>> Used Dominic's latest version[0] as a starting point. GUI part still
>>> needs to be rebased/updated, so it's not included here.
>>>
>>>
>>> Changes from v9:
>>>
>>> * Split patch into smaller parts
>>>
>>> * Some minor (style) fixes/improvements (see individual patches)
>>>
>>> * Drop $manifest_only parameter for parse_ovf
>>>
>>> Instead, untaint the path when calling file_size_info, which makes
>>> the call also work via API which uses the -T switch. If we do want
>>> to keep $manifest_only, I'd argue that it should also skip the
>>> file existence check and not only the file size check. Opinions?
>>>
>>> * Re-use do_import function
>>>
>>> Rather than duplicating most of it. The down side is the need to
>>> add a new parameter for skipping configuration update. But I
>>> suppose the plan is to have qm import switch to the new API at
>>> some point, and then do_import can be simplified.
>>>
>>> * Instead of adding an import-sources parameter to the API, use a new
>>> import-from property for the disk, that's only available with
>>> import/alloc-enabled API endpoints via its own version of the schema
>>>
>>> Avoids the split across regular drive key parameters and
>>> 'import_soruces', which avoids quite a bit of cross-checking
>>> between the two and parsing/passing around the latter.
>>>
>>> The big downsides are:
>>> * Schema handling is a bit messy.
>>> * Need to special case print_drive, because we do intermediate
>>> parse/print to cleanup drive paths. At a first glance, this
>>> seems not too easy to avoid without complicating things elsewehere.
>>> * Using the import-aware parse_drive in parse_volume, because that
>>> is used via the foreach_volume iterators handling the parameters
>>> of the import-enabled endpoints. Could be avoided by using for
>>> loops with the import-aware parse_drive instead of
>>> foreach_volume.
>>>
>>> Counter-arguments for using a single schema (citing Fabian G.):
>>> * docs/schema dump/api docs: shouldn't look like you can put that
>>> everywhere where we use the config schema
>>> * shouldn't have nasty side-effects if someone puts it into the
>>> config
>>>
>>>
>>> After all, the 'import-from' disk property approach turned out to be
>>> a uglier than I hoped it would.
>>>
>>> My problem with the 'import-sources' API parameter approach (see [0]
>>> for details) is that it requires specifying both
>>> scsi0: <target storage>:-1,<properties>
>>> import-sources: scsi0=/path/or/volid/for/source
>>> leading to a not ideal user interface and parameter handling needing
>>> cross-checks to verify that the right combination is specified, and
>>> passing both around at the same time.
>>>
>>> Another approach would be adding a new special volid syntax using
>>> my $IMPORT_DISK_RE = qr!^(([^/:\s]+):)import:(.*)$!;
>>> allowing for e.g.
>>> qm set 126 -scsi1 rbdkvm:import:myzpool:vm-114-disk-0,aio=native
>>> qm set 126 -scsi2 rbdkvm:import:/dev/zvol/myzpool/vm-114-disk-1,backup=0
>>> Yes, it's a hack, but it would avoid the pain points from both other
>>> approaches and be very simple. See the end of the mail for a POC.
>>
>> the biggest down-side is that this 'invades' the storage plugin-owned
>> volume ID namespace (further than the pre-existing, documented new disk
>> syntax) - a volume ID is defined as anything starting with a storage
>> identifier, followed by ':', followed by an arbitrary string (where the
>> semantics of that string are up to the plugin).
>>
>> while I don't think there is a storage plugin out there that somehow
>> starts its volume IDs with 'import:' by default, it still doesn't feel
>> very nice.. there might be plugins that are not very strict in their
>> parsing that might accept such an 'import volid' as regular volid, and
>> parse the latter part (which can be a regular volid after all!) instead
>> of the full thing, which would make code paths that should not accept
>> import volids accept them and just use the source without importing,
>> with potentially disastrous consequences :-/
>>
>
> Ok, good point!
>
>> that being said - other then this bigger conceptual question I only
>> found nits/fixup/followup material - so if we want to take the current
>> RFC I'd give this a more in-depth test spin in the next few days and
>> apply with the small stuff adressed ;)
>>
>
> Fine by me. I could also re-spin with the import-sources API parameter
> if that's preferred. If we go with the import-from approach, besides
> addressing your nits, I'd also not make parse_volume unconditionally
> parse with the extended schema, but replace the foreach_volume iterators
> in the appropriate places.
I actually prefer the import-from approach to the old one, even with the
duplicate schema downside - it makes it more explicit where we want to
allow 'magically' allocating new volumes, which was always a bit
confusing when you didn't have a clear picture of the code in your
head..
>
>>>
>>>
>>> [0]: https://lists.proxmox.com/pipermail/pve-devel/2021-June/048564.html
>>>
>>>
>>> pve-manager:
>>>
>>> Fabian Ebner (1):
>>> api: nodes: add readovf endpoint
>>>
>>> PVE/API2/Nodes.pm | 7 +++++++
>>> 1 file changed, 7 insertions(+)
>>>
>>>
>>> qemu-server:
>>>
>>> Dominic Jäger (1):
>>> api: support VM disk import
>>>
>>> Fabian Ebner (6):
>>> schema: add pve-volume-id-or-absolute-path
>>> parse ovf: untaint path when calling file_size_info
>>> api: add endpoint for parsing .ovf files
>>> image convert: allow block device as source
>>> schema: drive: use separate schema when disk allocation is possible
>>> api: create disks: factor out common part from if/else
>>>
>>> PVE/API2/Qemu.pm | 86 +++++++++++++++++++++++----------
>>> PVE/API2/Qemu/Makefile | 2 +-
>>> PVE/API2/Qemu/OVF.pm | 55 +++++++++++++++++++++
>>> PVE/QemuConfig.pm | 2 +-
>>> PVE/QemuServer.pm | 57 +++++++++++++++++++---
>>> PVE/QemuServer/Drive.pm | 92 +++++++++++++++++++++++++++---------
>>> PVE/QemuServer/ImportDisk.pm | 2 +-
>>> PVE/QemuServer/OVF.pm | 9 ++--
>>> 8 files changed, 242 insertions(+), 63 deletions(-)
>>> create mode 100644 PVE/API2/Qemu/OVF.pm
>>>
>>> --
>>> 2.30.2
>>>
>>> diff --git a/PVE/API2/Qemu.pm b/PVE/API2/Qemu.pm
>>> index 6992f6f..6a22899 100644
>>> --- a/PVE/API2/Qemu.pm
>>> +++ b/PVE/API2/Qemu.pm
>>> @@ -21,8 +21,9 @@ use PVE::ReplicationConfig;
>>> use PVE::GuestHelpers;
>>> use PVE::QemuConfig;
>>> use PVE::QemuServer;
>>> -use PVE::QemuServer::Drive;
>>> use PVE::QemuServer::CPUConfig;
>>> +use PVE::QemuServer::Drive;
>>> +use PVE::QemuServer::ImportDisk;
>>> use PVE::QemuServer::Monitor qw(mon_cmd);
>>> use PVE::QemuServer::Machine;
>>> use PVE::QemuMigrate;
>>> @@ -64,6 +65,7 @@ my $resolve_cdrom_alias = sub {
>>> };
>>>
>>> my $NEW_DISK_RE = qr!^(([^/:\s]+):)?(\d+(\.\d+)?)$!;
>>> +my $IMPORT_DISK_RE = qr!^(([^/:\s]+):)import:(.*)$!;
>>> my $check_storage_access = sub {
>>> my ($rpcenv, $authuser, $storecfg, $vmid, $settings, $default_storage) = @_;
>>>
>>> @@ -86,6 +88,9 @@ my $check_storage_access = sub {
>>> my $scfg = PVE::Storage::storage_config($storecfg, $storeid);
>>> raise_param_exc({ storage => "storage '$storeid' does not support vm images"})
>>> if !$scfg->{content}->{images};
>>> + } elsif (!$isCDROM && ($volid =~ $IMPORT_DISK_RE)) {
>>> + warn "check access matched: $2 and $3\n";
>>> + warn "TODO implement import access check!\n";
>>> } else {
>>> PVE::Storage::check_volume_access($rpcenv, $authuser, $storecfg, $vmid, $volid);
>>> }
>>> @@ -212,6 +217,29 @@ my $create_disks = sub {
>>> $disk->{size} = PVE::Tools::convert_size($size, 'kb' => 'b');
>>> delete $disk->{format}; # no longer needed
>>> $res->{$ds} = PVE::QemuServer::print_drive($disk);
>>> + } elsif ($volid =~ $IMPORT_DISK_RE) {
>>> + my ($storeid, $source) = ($2, $3);
>>> +
>>> + $source = PVE::Storage::abs_filesystem_path($storecfg, $source, 1);
>>> + my $src_size = PVE::Storage::file_size_info($source);
>>> + die "Could not get file size of $source" if !defined($src_size);
>>> +
>>> + my (undef, $dst_volid) = PVE::QemuServer::ImportDisk::do_import(
>>> + $source,
>>> + $vmid,
>>> + $storeid,
>>> + {
>>> + drive_name => $ds,
>>> + format => $disk->{format},
>>> + 'skip-config-update' => 1,
>>> + },
>>> + );
>>> +
>>> + push @$vollist, $dst_volid;
>>> + $disk->{file} = $dst_volid;
>>> + $disk->{size} = $src_size;
>>> + delete $disk->{format}; # no longer needed
>>> + $res->{$ds} = PVE::QemuServer::print_drive($disk);
>>> } else {
>>>
>>> PVE::Storage::check_volume_access($rpcenv, $authuser, $storecfg, $vmid, $volid);
>>> diff --git a/PVE/QemuServer/ImportDisk.pm b/PVE/QemuServer/ImportDisk.pm
>>> index 51ad52e..7557cac 100755
>>> --- a/PVE/QemuServer/ImportDisk.pm
>>> +++ b/PVE/QemuServer/ImportDisk.pm
>>> @@ -71,7 +71,7 @@ sub do_import {
>>> PVE::Storage::activate_volumes($storecfg, [$dst_volid]);
>>> PVE::QemuServer::qemu_img_convert($src_path, $dst_volid, $src_size, undef, $zeroinit);
>>> PVE::Storage::deactivate_volumes($storecfg, [$dst_volid]);
>>> - PVE::QemuConfig->lock_config($vmid, $create_drive);
>>> + PVE::QemuConfig->lock_config($vmid, $create_drive) if !$params->{'skip-config-update'};
>>> };
>>> if (my $err = $@) {
>>> eval { PVE::Storage::vdisk_free($storecfg, $dst_volid) };
>>> --
>>> 2.30.2
>>>
>>>
>>> _______________________________________________
>>> pve-devel mailing list
>>> pve-devel at lists.proxmox.com
>>> https://lists.proxmox.com/cgi-bin/mailman/listinfo/pve-devel
>>>
>>
>>
>> _______________________________________________
>> pve-devel mailing list
>> pve-devel at lists.proxmox.com
>> https://lists.proxmox.com/cgi-bin/mailman/listinfo/pve-devel
>
More information about the pve-devel
mailing list