[pve-devel] [PATCH V2 qemu-server] fix #1739: cloudinit: add cisshdeletehostkeys option

[Mira Limbeck]

Thank you for the patch.

I tried to test this patch with both a CentOS 8 and Ubuntu 20.04 VM and 
it never worked right.

I added another line for 'once|yes|on|true' where ssh_deletekeys is set 
to 'true' as that is missing in this patch. Then I tried it with CentOS 
8 and cloud-init 19.4.

In the beginning the keys were deleted and regenerated, but after a few 
tries no matter using either of 'once', 'yes', 'on' or 'true' did not 
change anything. Even with a new instance-id the keys were never deleted 
no regenerated.

In Ubuntu 20.04 with cloud-init 20.4 it was the other way around, even 
rebooting without any changes to the cloud-init config (no instance-id 
change) and cisshdeletehostkeys set to 'no' the keys were regenerated 
every time.

What did you use to test this patch (OS, cloud-init version, specific 
cloudimg)?


Some additional comments inline

On 2/7/21 3:18 PM, Alexandre Derumier wrote:
> This define behaviour of ssh server keys generation on cloudinit
> config change.
>
> different value:
>
> - once : only once at vmstart  (default value)
> - no|off|false : never generate ssh key
> - yes|on|true: always generate ssh key
>
> When value is defined to 'once', the value is rewriten to 'no'
> in vmconfig after vm start
>
> We need to change instance-id at each config change because
> network configuration is only done once in pre-init local service,
> and a lock is done on instance-id, and they are no way to change
> that in cloudinit vm config.
>
> The ssh host keys are regenerated by ssh module, once by instance.
> user could change that,and do only once,
> but in this same ssh module, the user public keys are also deployed.
>
> Signed-off-by: Alexandre Derumier <aderumier at odiso.com>
> ---
>   PVE/QemuServer.pm           |  9 ++++++++-
>   PVE/QemuServer/Cloudinit.pm | 10 +++++++++-
>   2 files changed, 17 insertions(+), 2 deletions(-)
>
> diff --git a/PVE/QemuServer.pm b/PVE/QemuServer.pm
> index f401baf..c899b35 100644
> --- a/PVE/QemuServer.pm
> +++ b/PVE/QemuServer.pm
> @@ -760,6 +760,13 @@ my $confdesc_cloudinit = {
>   	format => 'urlencoded',
>   	description => "cloud-init: Setup public SSH keys (one key per line, OpenSSH format).",
>       },
> +    cisshdeletehostkeys => {
> +	optional => 1,
> +	type => 'string',
> +	enum => [qw(once yes no on off true false)],
> +	default_key => 1,
> +	description => "cloud-init: Regenerate host SSH keys on config change.",
> +    },
>   };

As mentioned on the previous patch, I'm not sure if we want to add 
another cloud-init config to the global config namespace. We could reuse 
the option 'cicustom' for this.

Perhaps @Thomas could chime in regarding this?

>   
>   # what about other qemu settings ?
> @@ -4951,7 +4958,7 @@ sub vm_start_nolock {
>   	$conf = PVE::QemuConfig->load_config($vmid); # update/reload
>       }
>   
> -    PVE::QemuServer::Cloudinit::generate_cloudinitconfig($conf, $vmid);
> +    PVE::QemuServer::Cloudinit::generate_cloudinitconfig($conf, $vmid, 1);
>   
>       my $defaults = load_defaults();
>   
> diff --git a/PVE/QemuServer/Cloudinit.pm b/PVE/QemuServer/Cloudinit.pm
> index c464bf3..7b24f93 100644
> --- a/PVE/QemuServer/Cloudinit.pm
> +++ b/PVE/QemuServer/Cloudinit.pm
> @@ -135,6 +135,7 @@ sub cloudinit_userdata {
>   	foreach my $k (@$keys) {
>   	    $content .= "  - $k\n";
>   	}
> +	$content .= "ssh_deletekeys: false\n" if defined($conf->{cisshdeletehostkeys}) && $conf->{cisshdeletehostkeys} =~ m/^(off|no|false)/;

This is missing the case for 'once', 'on', 'true' and 'yes' where 
ssh_deletekeys is set to 'true'.

And there's an additional whitespace at the end of the line.

>   
>       }
>       $content .= "chpasswd:\n";
>       $content .= "  expire: False\n";
> @@ -554,9 +555,10 @@ my $cloudinit_methods = {
>   };
>   
>   sub generate_cloudinitconfig {
> -    my ($conf, $vmid) = @_;
> +    my ($conf, $vmid, $vmstart) = @_;
>   
>       my $format = get_cloudinit_format($conf);
> +    my $generated = undef;
>   
>       PVE::QemuConfig->foreach_volume($conf, sub {
>           my ($ds, $drive) = @_;
> @@ -569,7 +571,13 @@ sub generate_cloudinitconfig {
>   	    or die "missing cloudinit methods for format '$format'\n";
>   
>   	$generator->($conf, $vmid, $drive, $volname, $storeid);
> +	$generated = 1;
>       });
> +
> +    if ($vmstart && $generated && (!defined($conf->{cisshdeletehostkeys}) || $conf->{cisshdeletehostkeys} eq 'once')) {
> +	$conf->{cisshdeletehostkeys} = 'no';
> +	PVE::QemuConfig->write_config($vmid, $conf);
> +    }

This always sets 'cisshdeletehostkeys' to 'no' if it is not set at all 
in the config which changes the default. I'd rather not do that for all, 
but only if it is set explicitly to 'once'.

So that would then be:

if ($vmstart && $generated && defined($conf->{cisshdeletehostkeys}) && 
$conf->{cisshdeletehostkeys} eq 'once')

>   }
>   
>   sub dump_cloudinit_config {