Fixed hns clean only in case of reboot

Signed-off-by: Roberto Bonafiglia <[email protected]>

Dockerfile.windows

@@ -56,6 +56,7 @@ RUN mkdir -p charts
 
 # We use the containerd-shim-runhcs-v1.exe binary from upstream, as it apparently can't be cross-built on Linux
 COPY Dockerfile ./
+COPY windows/node-calico.ps1 rancher/
 RUN CONTAINERD_VERSION=$(grep "rancher/hardened-containerd" Dockerfile | grep ':v' | cut -d '=' -f 2- | grep -oE "([0-9]+)\.([0-9]+)\.([0-9]+)") \
  && curl -sLO https://github.com/containerd/containerd/releases/download/v${CONTAINERD_VERSION}/containerd-${CONTAINERD_VERSION}-windows-amd64.tar.gz \
  && curl -sLO https://github.com/containerd/containerd/releases/download/v${CONTAINERD_VERSION}/containerd-${CONTAINERD_VERSION}-windows-amd64.tar.gz.sha256sum \

pkg/windows/calico.go

@@ -302,7 +302,9 @@ func (c *Calico) Start(ctx context.Context) error {
 
 // generateCalicoNetworks creates the overlay networks for internode networking
 func (c *Calico) generateCalicoNetworks() error {
-	if err := deleteAllNetworks(); err != nil {
+	cmd := exec.Command("powershell.exe", "C:\\var\\lib\\rancher\\rke2\\bin\\node-calico.ps1")
+	if err := cmd.Run(); err != nil {
+	//if err := deleteAllNetworks(); err != nil {
 		return errors.Wrapf(err, "failed to delete all networks before bootstrapping calico")
 	}
 

pkg/windows/utils.go

@@ -34,6 +34,12 @@ var (
 // createHnsNetwork creates the network that will connect nodes and returns its managementIP
 func createHnsNetwork(backend string, networkAdapter string) (string, error) {
 	var network hcsshim.HNSNetwork
+        // Check if the interface already exists
+	hcsnetwork, err := hcsshim.GetHNSNetworkByName(CalicoHnsNetworkName)
+	if err == nil {
+		return hcsnetwork.ManagementIP, nil
+	}
+
 	if backend == "vxlan" {
 		// Ignoring the return because both true and false without an error represent that the firewall rule was created or already exists
 		if _, err := wapi.FirewallRuleAdd("OverlayTraffic4789UDP", "Overlay network traffic UDP", "", "4789", wapi.NET_FW_IP_PROTOCOL_UDP, wapi.NET_FW_PROFILE2_ALL); err != nil {

windows/node-calico.ps1

@@ -0,0 +1,96 @@
+$softwareRegistryKey = "HKLM:\Software\Tigera"
+$calicoRegistryKey = $softwareRegistryKey + "\Calico"
+
+function Get-LastBootTime()
+{   
+    $bootTime = (Get-CimInstance win32_operatingsystem | select @{LABEL='LastBootUpTime';EXPRESSION={$_.lastbootuptime}}).LastBootUpTime
+    if (($bootTime -EQ $null) -OR ($bootTime.length -EQ 0))
+    {
+        throw "Failed to get last boot time"
+    } 
+
+    # This function is used in conjunction with Get-StoredLastBootTime, which
+    # returns a string, so convert the datetime value to a string using the "general" standard format.
+    return $bootTime.ToString("G")
+}
+
+function Get-StoredLastBootTime()
+{
+    try
+    {
+        return (Get-ItemProperty $calicoRegistryKey -ErrorAction Ignore).LastBootTime
+    }
+    catch
+    {
+        $PSItem.Exception.Message
+    }
+}
+
+function ensureRegistryKey()
+{
+    if (! (Test-Path $softwareRegistryKey))
+    {
+        New-Item $softwareRegistryKey
+    }
+    if (! (Test-Path $calicoRegistryKey))
+    {
+        New-Item $calicoRegistryKey
+    }
+}
+
+function Set-StoredLastBootTime($lastBootTime)
+{
+    ensureRegistryKey
+
+    return Set-ItemProperty $calicoRegistryKey -Name LastBootTime -Value $lastBootTime
+}
+
+$lastBootTime = Get-LastBootTime
+
+# Check if the node has been rebooted.  If so, the HNS networks will be in unknown state so we need to
+# clean them up and recreate them.
+$prevLastBootTime = Get-StoredLastBootTime
+Write-Host "StoredLastBootTime $prevLastBootTime, CurrentLastBootTime $lastBootTime"
+if ($prevLastBootTime -NE $lastBootTime)
+{
+     if ((Get-HNSNetwork | ? Type -NE nat))
+     {
+        Write-Host "First time Calico has run since boot up, cleaning out any old network state."
+        Get-HNSNetwork | ? Type -NE nat | Remove-HNSNetwork
+        do
+        {
+            Write-Host "Waiting for network deletion to complete."
+            Start-Sleep 1
+        } while ((Get-HNSNetwork | ? Type -NE nat))
+     }
+     # After deletion of all hns networks, wait for an interface to have an IP that is not a 169.254.0.0/16 (or 127.0.0.0/8) address,
+     # before creation of External network.
+     $isValidIP = $false
+     $IPRegEx1='(^127\.0\.0\.)'
+     $IPRegEx2='(^169\.254\.)'
+     while(!($isValidIP) -AND ($timeout -gt 0))
+     {
+         $IPAddress = (Get-NetIPAddress -AddressFamily IPv4).IPAddress
+         Write-Host "`nTimeout Remaining: $timeout sec"
+         Write-Host "List of IP Address before initialising Calico: $IPAddress"
+         Foreach ($ip in $IPAddress)
+         {
+             if (($ip -NotMatch $IPRegEx1) -AND ($ip -NotMatch $IPRegEx2))
+             {
+                 $isValidIP = $true
+                 Write-Host "`nFound valid IP: $ip"
+                 break
+             }
+         }
+         if (!($isValidIP))
+         {
+             Start-Sleep -s 5
+             $timeout = $timeout - 5
+         }
+     }
+}
+
+Set-StoredLastBootTime $lastBootTime
+$Stored = Get-StoredLastBootTime
+Write-Host "Stored new lastBootTime $Stored"
+