Merge pull request #109826 from danwinship/multi-load-balancer

fix kube-proxy bug with multiple LB IPs and source ranges
2025-07-19 09:52:49 +00:00 · 2022-05-06 03:09:15 -07:00 · 2022-05-06 03:09:15 -07:00 · 2b3508e0f1
commit 2b3508e0f1
parent 6585b9da58 813aca47af
2 changed files with 118 additions and 54 deletions
--- a/pkg/proxy/iptables/proxier.go
+++ b/pkg/proxy/iptables/proxier.go
@ -1243,6 +1243,46 @@ func (proxier *Proxier) syncProxyRules() {
 				// The firewall chain will jump to the "external destination"
 				// chain.
 				nextChain = svcInfo.firewallChainName
+
+				// The service firewall rules are created based on the
+				// loadBalancerSourceRanges field.  This only works for
+				// VIP-like loadbalancers that preserve source IPs.  For
+				// loadbalancers which direct traffic to service NodePort, the
+				// firewall rules will not apply.
+				args = append(args[:0],
+					"-A", string(nextChain),
+					"-m", "comment", "--comment", fmt.Sprintf(`"%s loadbalancer IP"`, svcNameString),
+				)
+
+				// firewall filter based on each source range
+				allowFromNode := false
+				for _, src := range svcInfo.LoadBalancerSourceRanges() {
+					proxier.natRules.Write(args, "-s", src, "-j", string(externalTrafficChain))
+					_, cidr, err := netutils.ParseCIDRSloppy(src)
+					if err != nil {
+						klog.ErrorS(err, "Error parsing CIDR in LoadBalancerSourceRanges, dropping it", "cidr", cidr)
+					} else if cidr.Contains(proxier.nodeIP) {
+						allowFromNode = true
+					}
+				}
+				// For VIP-like LBs, the VIP is often added as a local
+				// address (via an IP route rule).  In that case, a request
+				// from a node to the VIP will not hit the loadbalancer but
+				// will loop back with the source IP set to the VIP.  We
+				// need the following rules to allow requests from this node.
+				if allowFromNode {
+					for _, lbip := range svcInfo.LoadBalancerIPStrings() {
+						proxier.natRules.Write(
+							args,
+							"-s", lbip,
+							"-j", string(externalTrafficChain))
+					}
+				}
+
+				// If the packet was able to reach the end of firewall chain,
+				// then it did not get DNATed.  It means the packet cannot go
+				// thru the firewall, then mark it for DROP.
+				proxier.natRules.Write(args, "-j", string(KubeMarkDropChain))
 			}

 			for _, lbip := range svcInfo.LoadBalancerIPStrings() {
@ -1254,44 +1294,6 @@ func (proxier *Proxier) syncProxyRules() {
 					"--dport", strconv.Itoa(svcInfo.Port()),
 					"-j", string(nextChain))

-				// The service firewall rules are created based on the
-				// loadBalancerSourceRanges field.  This only works for
-				// VIP-like loadbalancers that preserve source IPs.  For
-				// loadbalancers which direct traffic to service NodePort, the
-				// firewall rules will not apply.
-				if len(svcInfo.LoadBalancerSourceRanges()) > 0 {
-					args = append(args[:0],
-						"-A", string(nextChain),
-						"-m", "comment", "--comment", fmt.Sprintf(`"%s loadbalancer IP"`, svcNameString),
-					)
-
-					// firewall filter based on each source range
-					allowFromNode := false
-					for _, src := range svcInfo.LoadBalancerSourceRanges() {
-						proxier.natRules.Write(args, "-s", src, "-j", string(externalTrafficChain))
-						_, cidr, err := netutils.ParseCIDRSloppy(src)
-						if err != nil {
-							klog.ErrorS(err, "Error parsing CIDR in LoadBalancerSourceRanges, dropping it", "cidr", cidr)
-						} else if cidr.Contains(proxier.nodeIP) {
-							allowFromNode = true
-						}
-					}
-					// For VIP-like LBs, the VIP is often added as a local
-					// address (via an IP route rule).  In that case, a request
-					// from a node to the VIP will not hit the loadbalancer but
-					// will loop back with the source IP set to the VIP.  We
-					// need the following rule to allow requests from this node.
-					if allowFromNode {
-						proxier.natRules.Write(
-							args,
-							"-s", lbip,
-							"-j", string(externalTrafficChain))
-					}
-					// If the packet was able to reach the end of firewall chain,
-					// then it did not get DNATed.  It means the packet cannot go
-					// thru the firewall, then mark it for DROP.
-					proxier.natRules.Write(args, "-j", string(KubeMarkDropChain))
-				}
 			}
 		} else {
 			// No endpoints.
--- a/pkg/proxy/iptables/proxier_test.go
+++ b/pkg/proxy/iptables/proxier_test.go
@ -420,6 +420,7 @@ func NewFakeProxier(ipt utiliptables.Interface) *Proxier {
 		filterRules:              utilproxy.LineBuffer{},
 		natChains:                utilproxy.LineBuffer{},
 		natRules:                 utilproxy.LineBuffer{},
+		nodeIP:                   netutils.ParseIPSloppy(testNodeIP),
 		nodePortAddresses:        make([]string, 0),
 		networkInterfacer:        networkInterfacer,
 	}
@ -2369,7 +2370,8 @@ func TestLoadBalancer(t *testing.T) {
 	svcIP := "172.30.0.41"
 	svcPort := 80
 	svcNodePort := 3001
-	svcLBIP := "1.2.3.4"
+	svcLBIP1 := "1.2.3.4"
+	svcLBIP2 := "5.6.7.8"
 	svcPortName := proxy.ServicePortName{
 		NamespacedName: makeNSN("ns1", "svc1"),
 		Port:           "p80",
@ -2386,12 +2388,16 @@ func TestLoadBalancer(t *testing.T) {
 				Protocol: v1.ProtocolTCP,
 				NodePort: int32(svcNodePort),
 			}}
-			svc.Status.LoadBalancer.Ingress = []v1.LoadBalancerIngress{{
-				IP: svcLBIP,
-			}}
-			// Also ensure that invalid LoadBalancerSourceRanges will not result
-			// in a crash.
-			svc.Spec.LoadBalancerSourceRanges = []string{" 203.0.113.0/25"}
+			svc.Status.LoadBalancer.Ingress = []v1.LoadBalancerIngress{
+				{IP: svcLBIP1},
+				{IP: svcLBIP2},
+			}
+			svc.Spec.LoadBalancerSourceRanges = []string{
+				"192.168.0.0/24",
+
+				// Regression test that excess whitespace gets ignored
+				" 203.0.113.0/25",
+			}
 		}),
 	)

@ -2435,10 +2441,14 @@ func TestLoadBalancer(t *testing.T) {
 		-A KUBE-NODEPORTS -m comment --comment ns1/svc1:p80 -m tcp -p tcp --dport 3001 -j KUBE-EXT-XPGD46QRK7WJZT7O
 		-A KUBE-SERVICES -m comment --comment "ns1/svc1:p80 cluster IP" -m tcp -p tcp -d 172.30.0.41 --dport 80 -j KUBE-SVC-XPGD46QRK7WJZT7O
 		-A KUBE-SERVICES -m comment --comment "ns1/svc1:p80 loadbalancer IP" -m tcp -p tcp -d 1.2.3.4 --dport 80 -j KUBE-FW-XPGD46QRK7WJZT7O
+		-A KUBE-SERVICES -m comment --comment "ns1/svc1:p80 loadbalancer IP" -m tcp -p tcp -d 5.6.7.8 --dport 80 -j KUBE-FW-XPGD46QRK7WJZT7O
 		-A KUBE-SERVICES -m comment --comment "kubernetes service nodeports; NOTE: this must be the last rule in this chain" -m addrtype --dst-type LOCAL -j KUBE-NODEPORTS
 		-A KUBE-EXT-XPGD46QRK7WJZT7O -m comment --comment "masquerade traffic for ns1/svc1:p80 external destinations" -j KUBE-MARK-MASQ
 		-A KUBE-EXT-XPGD46QRK7WJZT7O -j KUBE-SVC-XPGD46QRK7WJZT7O
+		-A KUBE-FW-XPGD46QRK7WJZT7O -m comment --comment "ns1/svc1:p80 loadbalancer IP" -s 192.168.0.0/24 -j KUBE-EXT-XPGD46QRK7WJZT7O
 		-A KUBE-FW-XPGD46QRK7WJZT7O -m comment --comment "ns1/svc1:p80 loadbalancer IP" -s 203.0.113.0/25 -j KUBE-EXT-XPGD46QRK7WJZT7O
+		-A KUBE-FW-XPGD46QRK7WJZT7O -m comment --comment "ns1/svc1:p80 loadbalancer IP" -s 1.2.3.4 -j KUBE-EXT-XPGD46QRK7WJZT7O
+		-A KUBE-FW-XPGD46QRK7WJZT7O -m comment --comment "ns1/svc1:p80 loadbalancer IP" -s 5.6.7.8 -j KUBE-EXT-XPGD46QRK7WJZT7O
 		-A KUBE-FW-XPGD46QRK7WJZT7O -m comment --comment "ns1/svc1:p80 loadbalancer IP" -j KUBE-MARK-DROP
 		-A KUBE-MARK-MASQ -j MARK --or-mark 0x4000
 		-A KUBE-POSTROUTING -m mark ! --mark 0x4000/0x4000 -j RETURN
@ -2479,34 +2489,86 @@ func TestLoadBalancer(t *testing.T) {
 			masq:     true,
 		},
 		{
-			name:     "accepted external to LB",
+			name:     "accepted external to LB1",
 			sourceIP: testExternalClient,
-			destIP:   svcLBIP,
+			destIP:   svcLBIP1,
 			destPort: svcPort,
 			output:   fmt.Sprintf("%s:%d", epIP, svcPort),
 			masq:     true,
 		},
 		{
-			name:     "blocked external to LB",
+			name:     "accepted external to LB2",
+			sourceIP: testExternalClient,
+			destIP:   svcLBIP2,
+			destPort: svcPort,
+			output:   fmt.Sprintf("%s:%d", epIP, svcPort),
+			masq:     true,
+		},
+		{
+			name:     "blocked external to LB1",
 			sourceIP: testExternalClientBlocked,
-			destIP:   svcLBIP,
+			destIP:   svcLBIP1,
 			destPort: svcPort,
 			output:   "DROP",
 		},
 		{
-			name:     "pod to LB (blocked by LoadBalancerSourceRanges)",
+			name:     "blocked external to LB2",
+			sourceIP: testExternalClientBlocked,
+			destIP:   svcLBIP2,
+			destPort: svcPort,
+			output:   "DROP",
+		},
+		{
+			name:     "pod to LB1 (blocked by LoadBalancerSourceRanges)",
 			sourceIP: "10.0.0.2",
-			destIP:   svcLBIP,
+			destIP:   svcLBIP1,
 			destPort: svcPort,
 			output:   "DROP",
 		},
 		{
-			name:     "node to LB (blocked by LoadBalancerSourceRanges)",
-			sourceIP: testNodeIP,
-			destIP:   svcLBIP,
+			name:     "pod to LB2 (blocked by LoadBalancerSourceRanges)",
+			sourceIP: "10.0.0.2",
+			destIP:   svcLBIP2,
 			destPort: svcPort,
 			output:   "DROP",
 		},
+		{
+			name:     "node to LB1 (allowed by LoadBalancerSourceRanges)",
+			sourceIP: testNodeIP,
+			destIP:   svcLBIP1,
+			destPort: svcPort,
+			output:   fmt.Sprintf("%s:%d", epIP, svcPort),
+			masq:     true,
+		},
+		{
+			name:     "node to LB2 (allowed by LoadBalancerSourceRanges)",
+			sourceIP: testNodeIP,
+			destIP:   svcLBIP2,
+			destPort: svcPort,
+			output:   fmt.Sprintf("%s:%d", epIP, svcPort),
+			masq:     true,
+		},
+
+		// The LB rules assume that when you connect from a node to a LB IP, that
+		// something external to kube-proxy will cause the connection to be
+		// SNATted to the LB IP, so if the LoadBalancerSourceRanges include the
+		// node IP, then we add a rule allowing traffic from the LB IP as well...
+		{
+			name:     "same node to LB1, SNATted to LB1 (implicitly allowed)",
+			sourceIP: svcLBIP1,
+			destIP:   svcLBIP1,
+			destPort: svcPort,
+			output:   fmt.Sprintf("%s:%d", epIP, svcPort),
+			masq:     true,
+		},
+		{
+			name:     "same node to LB2, SNATted to LB2 (implicitly allowed)",
+			sourceIP: svcLBIP2,
+			destIP:   svcLBIP2,
+			destPort: svcPort,
+			output:   fmt.Sprintf("%s:%d", epIP, svcPort),
+			masq:     true,
+		},
 	})
 }