[PATCH qemu-server v6 10/13] network: sdn: push the DHCP mappings of a guest when it resumes

[Hannes Laimer <h.laimer@proxmox.com>]

A mapping edit and a migration of the guest exclude each other, but an
edit through a node not yet upgraded takes no lock and writes no
reservation, so it can land while the guest is migrated. Every resume
now pushes the mapping of each NIC of the guest on the local node
before the guest continues, and the last step of a live migration
resumes the target after the config moved, so the target serves what
the cache vouches for. The push asks the IPAM only when the cache
cannot vouch for the address. The push does not wait long for a push
holding the reservation file. A failed request fails the task after
the guest continued, naming the IPAM's error. An IPAM that does not
answer holds the resume up to the request's timeout first, and a slow
one for every request the lookup makes. That beats serving the guest a
stale address.

Signed-off-by: Hannes Laimer <h.laimer@proxmox.com>
---
 src/PVE/QemuServer/Network.pm  | 18 ++++++++++++++++++
 src/PVE/QemuServer/RunState.pm |  8 ++++++++
 2 files changed, 26 insertions(+)

diff --git a/src/PVE/QemuServer/Network.pm b/src/PVE/QemuServer/Network.pm
index b4893c015646..7ae448ea1369 100644
--- a/src/PVE/QemuServer/Network.pm
+++ b/src/PVE/QemuServer/Network.pm
@@ -272,6 +272,24 @@ sub add_random_macs {
     }
 }
 
+# a resume pushes the current mapping of each NIC. The target of a live
+# migration resumes after the config moved and serves the current addresses
+# from then on. The push does not wait long for a push holding the
+# reservation file
+sub add_nets_dhcp_mapping {
+    my ($conf) = @_;
+
+    my @errors;
+    for my $opt (keys %$conf) {
+        next if $opt !~ m/^net\d+$/;
+        my $net = parse_net($conf->{$opt}, 1) or next;
+        next if !$net->{bridge} || !$net->{macaddr};
+        eval { PVE::Network::SDN::Dhcp::update_mapping($net->{bridge}, $net->{macaddr}, 1) };
+        push @errors, $@ if $@;
+    }
+    die join('', @errors) if @errors;
+}
+
 sub add_nets_bridge_fdb {
     my ($conf, $vmid) = @_;
 
diff --git a/src/PVE/QemuServer/RunState.pm b/src/PVE/QemuServer/RunState.pm
index 889ddd28287b..cf1e1873c39d 100644
--- a/src/PVE/QemuServer/RunState.pm
+++ b/src/PVE/QemuServer/RunState.pm
@@ -181,8 +181,16 @@ sub vm_resume($vmid, $skiplock = undef, $nocheck = undef) {
 
             PVE::QemuServer::Network::add_nets_bridge_fdb($conf, $vmid)
                 if $resume_cmd eq 'cont';
+            # a failed push is raised after the guest continued, so the task
+            # says what went wrong and the guest does not stay paused for it
+            my $dhcp_err;
+            if ($resume_cmd eq 'cont') {
+                eval { PVE::QemuServer::Network::add_nets_dhcp_mapping($conf) };
+                $dhcp_err = $@;
+            }
 
             mon_cmd($vmid, $resume_cmd);
+            die $dhcp_err if $dhcp_err;
         },
     );
 }
-- 
2.47.3