firewall.go

/*
Copyright 2016 The Kubernetes Authors.

Licensed under the Apache License, Version 2.0 (the "License");
you may not use this file except in compliance with the License.
You may obtain a copy of the License at

    http://www.apache.org/licenses/LICENSE-2.0

Unless required by applicable law or agreed to in writing, software
distributed under the License is distributed on an "AS IS" BASIS,
WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
See the License for the specific language governing permissions and
limitations under the License.
*/

package model

import (
	"fmt"
	"strconv"

	"k8s.io/kops/pkg/apis/kops"
	"k8s.io/kops/upup/pkg/fi"
	"k8s.io/kops/upup/pkg/fi/cloudup/awstasks"

	"k8s.io/klog"
)

type Protocol int

const (
	ProtocolIPIP Protocol = 4
)

// FirewallModelBuilder configures firewall network objects
type FirewallModelBuilder struct {
	*KopsModelContext
	Lifecycle *fi.Lifecycle
}

var _ fi.ModelBuilder = &FirewallModelBuilder{}

func (b *FirewallModelBuilder) Build(c *fi.ModelBuilderContext) error {
	nodeGroups, err := b.buildNodeRules(c)
	if err != nil {
		return err
	}

	masterGroups, err := b.buildMasterRules(c, nodeGroups)
	if err != nil {
		return err
	}

	// We _should_ block per port... but:
	// * It causes e2e tests to break
	// * Users expect to be able to reach pods
	// * If users are running an overlay, we punch a hole in it anyway
	// b.applyNodeToMasterAllowSpecificPorts(c)
	b.applyNodeToMasterBlockSpecificPorts(c, nodeGroups, masterGroups)

	return nil
}

func (b *FirewallModelBuilder) buildNodeRules(c *fi.ModelBuilderContext) ([]SecurityGroupInfo, error) {

	nodeGroups, err := b.GetSecurityGroups(kops.InstanceGroupRoleNode)
	if err != nil {
		return nil, err
	}

	for _, group := range nodeGroups {
		group.Task.Lifecycle = b.Lifecycle
		c.AddTask(group.Task)
	}

	for _, src := range nodeGroups {
		// Allow full egress
		{
			t := &awstasks.SecurityGroupRule{
				Name:          s("node-egress" + src.Suffix),
				Lifecycle:     b.Lifecycle,
				SecurityGroup: src.Task,
				Egress:        fi.Bool(true),
				CIDR:          s("0.0.0.0/0"),
			}
			c.AddTask(t)
		}

		// Nodes can talk to nodes
		for _, dest := range nodeGroups {
			suffix := JoinSuffixes(src, dest)

			t := &awstasks.SecurityGroupRule{
				Name:          s("all-node-to-node" + suffix),
				Lifecycle:     b.Lifecycle,
				SecurityGroup: dest.Task,
				SourceGroup:   src.Task,
			}
			c.AddTask(t)
		}

	}

	return nodeGroups, nil
}

/*
This is dead code, but hopefully one day we can open specific ports only, for better security

func (b *FirewallModelBuilder) applyNodeToMasterAllowSpecificPorts(c *fi.ModelBuilderContext) {
	// TODO: We need to remove the ALL rule
	//W1229 12:32:22.300132    9003 executor.go:109] error running task "SecurityGroupRule/node-to-master-443" (9m58s remaining to succeed): error creating SecurityGroupIngress: InvalidPermission.Duplicate: the specified rule "peer: sg-f6b1a68b, ALL, ALLOW" already exists
	//status code: 400, request id: 6a69627f-9a26-4bd0-b294-a9a96f89bc46

	udpPorts := []int64{}
	tcpPorts := []int64{}
	protocols := []Protocol{}

	// allow access to API
	tcpPorts = append(tcpPorts, 443)

	// allow cadvisor
	tcpPorts = append(tcpPorts, 4194)

	// kubelet read-only used by heapster
	tcpPorts = append(tcpPorts, 10255)

	if b.Cluster.Spec.Networking != nil {
		if b.Cluster.Spec.Networking.Kopeio != nil {
			// VXLAN over UDP
			udpPorts = append(udpPorts, 4789)
		}

		if b.Cluster.Spec.Networking.Weave != nil {
			udpPorts = append(udpPorts, 6783)
			tcpPorts = append(tcpPorts, 6783)
			udpPorts = append(udpPorts, 6784)
		}

		if b.Cluster.Spec.Networking.Flannel != nil {
			switch b.Cluster.Spec.Networking.Flannel.Backend {
			case "", "udp":
				udpPorts = append(udpPorts, 8285)
			case "vxlan":
				udpPorts = append(udpPorts, 8472)
			default:
				klog.Warningf("unknown flannel networking backend %q", b.Cluster.Spec.Networking.Flannel.Backend)
			}
		}

		if b.Cluster.Spec.Networking.Calico != nil {
			// Calico needs to access etcd
			// TODO: Remove, replace with etcd in calico manifest
			// https://coreos.com/etcd/docs/latest/v2/configuration.html
			klog.Warningf("Opening etcd port on masters for access from the nodes, for calico.  This is unsafe in untrusted environments.")
			tcpPorts = append(tcpPorts, 4001)
			tcpPorts = append(tcpPorts, 179)
			protocols = append(protocols, ProtocolIPIP)
		}

		if b.Cluster.Spec.Networking.Romana != nil {
			// Romana needs to access etcd
			klog.Warningf("Opening etcd port on masters for access from the nodes, for romana.  This is unsafe in untrusted environments.")
			tcpPorts = append(tcpPorts, 4001)
			tcpPorts = append(tcpPorts, 9600)
		}

		if b.Cluster.Spec.Networking.Cilium != nil {
			// Cilium needs to access etcd
			klog.Warningf("Opening etcd port on masters for access from the nodes, for Cilium.  This is unsafe in untrusted environments.")
			tcpPorts = append(tcpPorts, 4001)
		}

		if b.Cluster.Spec.Networking.Kuberouter != nil {
			protocols = append(protocols, ProtocolIPIP)
		}
	}

	for _, udpPort := range udpPorts {
		t := &awstasks.SecurityGroupRule{
			Name:          s(fmt.Sprintf("node-to-master-udp-%d", udpPort)),
			Lifecycle:     b.Lifecycle,
			SecurityGroup: b.LinkToSecurityGroup(kops.InstanceGroupRoleMaster),
			SourceGroup:   b.LinkToSecurityGroup(kops.InstanceGroupRoleNode),
			FromPort:      i64(udpPort),
			ToPort:        i64(udpPort),
			Protocol:      s("udp"),
		}
		c.AddTask(t)
	}
	for _, tcpPort := range tcpPorts {
		t := &awstasks.SecurityGroupRule{
			Name:          s(fmt.Sprintf("node-to-master-tcp-%d", tcpPort)),
			Lifecycle:     b.Lifecycle,
			SecurityGroup: b.LinkToSecurityGroup(kops.InstanceGroupRoleMaster),
			SourceGroup:   b.LinkToSecurityGroup(kops.InstanceGroupRoleNode),
			FromPort:      i64(tcpPort),
			ToPort:        i64(tcpPort),
			Protocol:      s("tcp"),
		}
		c.AddTask(t)
	}
	for _, protocol := range protocols {
		awsName := strconv.Itoa(int(protocol))
		name := awsName
		switch protocol {
		case ProtocolIPIP:
			name = "ipip"
		default:
			klog.Warningf("unknown protocol %q - naming by number", awsName)
		}

		t := &awstasks.SecurityGroupRule{
			Name:          s("node-to-master-protocol-" + name),
			Lifecycle:     b.Lifecycle,
			SecurityGroup: b.LinkToSecurityGroup(kops.InstanceGroupRoleMaster),
			SourceGroup:   b.LinkToSecurityGroup(kops.InstanceGroupRoleNode),
			Protocol:      s(awsName),
		}
		c.AddTask(t)
	}
}
*/

func (b *FirewallModelBuilder) applyNodeToMasterBlockSpecificPorts(c *fi.ModelBuilderContext, nodeGroups []SecurityGroupInfo, masterGroups []SecurityGroupInfo) {
	type portRange struct {
		From int
		To   int
	}

	// TODO: Make less hacky
	// TODO: Fix management - we need a wildcard matcher now
	tcpBlocked := make(map[int]bool)

	// Don't allow nodes to access etcd client port
	tcpBlocked[4001] = true
	tcpBlocked[4002] = true

	// Don't allow nodes to access etcd peer port
	tcpBlocked[2380] = true
	tcpBlocked[2381] = true

	udpRanges := []portRange{{From: 1, To: 65535}}
	protocols := []Protocol{}

	if b.Cluster.Spec.Networking.Calico != nil {
		// Calico needs to access etcd
		// TODO: Remove, replace with etcd in calico manifest
		klog.Warningf("Opening etcd port on masters for access from the nodes, for calico.  This is unsafe in untrusted environments.")
		tcpBlocked[4001] = false
		protocols = append(protocols, ProtocolIPIP)
	}

	if b.Cluster.Spec.Networking.Romana != nil {
		// Romana needs to access etcd
		klog.Warningf("Opening etcd port on masters for access from the nodes, for romana.  This is unsafe in untrusted environments.")
		tcpBlocked[4001] = false
		protocols = append(protocols, ProtocolIPIP)
	}

	if b.Cluster.Spec.Networking.Kuberouter != nil {
		protocols = append(protocols, ProtocolIPIP)
	}

	tcpRanges := []portRange{
		{From: 1, To: 0},
	}
	for port := 1; port < 65536; port++ {
		previous := &tcpRanges[len(tcpRanges)-1]
		if !tcpBlocked[port] {
			if (previous.To + 1) == port {
				previous.To = port
			} else {
				tcpRanges = append(tcpRanges, portRange{From: port, To: port})
			}
		}
	}

	for _, masterGroup := range masterGroups {
		for _, nodeGroup := range nodeGroups {
			suffix := JoinSuffixes(nodeGroup, masterGroup)

			for _, r := range udpRanges {
				t := &awstasks.SecurityGroupRule{
					Name:          s(fmt.Sprintf("node-to-master-udp-%d-%d%s", r.From, r.To, suffix)),
					Lifecycle:     b.Lifecycle,
					SecurityGroup: masterGroup.Task,
					SourceGroup:   nodeGroup.Task,
					FromPort:      i64(int64(r.From)),
					ToPort:        i64(int64(r.To)),
					Protocol:      s("udp"),
				}
				c.AddTask(t)
			}
			for _, r := range tcpRanges {
				t := &awstasks.SecurityGroupRule{
					Name:          s(fmt.Sprintf("node-to-master-tcp-%d-%d%s", r.From, r.To, suffix)),
					Lifecycle:     b.Lifecycle,
					SecurityGroup: masterGroup.Task,
					SourceGroup:   nodeGroup.Task,
					FromPort:      i64(int64(r.From)),
					ToPort:        i64(int64(r.To)),
					Protocol:      s("tcp"),
				}
				c.AddTask(t)
			}
			for _, protocol := range protocols {
				awsName := strconv.Itoa(int(protocol))
				name := awsName
				switch protocol {
				case ProtocolIPIP:
					name = "ipip"
				default:
					klog.Warningf("unknown protocol %q - naming by number", awsName)
				}

				t := &awstasks.SecurityGroupRule{
					Name:          s(fmt.Sprintf("node-to-master-protocol-%s%s", name, suffix)),
					Lifecycle:     b.Lifecycle,
					SecurityGroup: masterGroup.Task,
					SourceGroup:   nodeGroup.Task,
					Protocol:      s(awsName),
				}
				c.AddTask(t)
			}
		}
	}

	// For AmazonVPC networking, pods running in Nodes could need to reach pods in master/s
	if b.Cluster.Spec.Networking != nil && b.Cluster.Spec.Networking.AmazonVPC != nil {
		// Nodes can talk to masters
		for _, src := range nodeGroups {
			for _, dest := range masterGroups {
				suffix := JoinSuffixes(src, dest)

				t := &awstasks.SecurityGroupRule{
					Name:          s("all-nodes-to-master" + suffix),
					Lifecycle:     b.Lifecycle,
					SecurityGroup: dest.Task,
					SourceGroup:   src.Task,
				}
				c.AddTask(t)
			}
		}
	}

}

func (b *FirewallModelBuilder) buildMasterRules(c *fi.ModelBuilderContext, nodeGroups []SecurityGroupInfo) ([]SecurityGroupInfo, error) {
	masterGroups, err := b.GetSecurityGroups(kops.InstanceGroupRoleMaster)
	if err != nil {
		return nil, err
	}

	for _, group := range masterGroups {
		group.Task.Lifecycle = b.Lifecycle
		c.AddTask(group.Task)
	}

	for _, src := range masterGroups {
		// Allow full egress
		{
			t := &awstasks.SecurityGroupRule{
				Name:          s("master-egress" + src.Suffix),
				Lifecycle:     b.Lifecycle,
				SecurityGroup: src.Task,
				Egress:        fi.Bool(true),
				CIDR:          s("0.0.0.0/0"),
			}
			c.AddTask(t)
		}

		// Masters can talk to masters
		for _, dest := range masterGroups {
			suffix := JoinSuffixes(src, dest)

			t := &awstasks.SecurityGroupRule{
				Name:          s("all-master-to-master" + suffix),
				Lifecycle:     b.Lifecycle,
				SecurityGroup: dest.Task,
				SourceGroup:   src.Task,
			}
			c.AddTask(t)
		}

		// Masters can talk to nodes
		for _, dest := range nodeGroups {
			suffix := JoinSuffixes(src, dest)

			t := &awstasks.SecurityGroupRule{
				Name:          s("all-master-to-node" + suffix),
				Lifecycle:     b.Lifecycle,
				SecurityGroup: dest.Task,
				SourceGroup:   src.Task,
			}
			c.AddTask(t)
		}
	}

	return masterGroups, nil
}

type SecurityGroupInfo struct {
	Name   string
	Suffix string
	Task   *awstasks.SecurityGroup
}

func (b *KopsModelContext) GetSecurityGroups(role kops.InstanceGroupRole) ([]SecurityGroupInfo, error) {
	var baseGroup *awstasks.SecurityGroup
	if role == kops.InstanceGroupRoleMaster {
		name := b.SecurityGroupName(role)
		baseGroup = &awstasks.SecurityGroup{
			Name:        s(name),
			VPC:         b.LinkToVPC(),
			Description: s("Security group for masters"),
			RemoveExtraRules: []string{
				"port=22",   // SSH
				"port=443",  // k8s api
				"port=2380", // etcd main peer
				"port=2381", // etcd events peer
				"port=4001", // etcd main
				"port=4002", // etcd events
				"port=4789", // VXLAN
				"port=179",  // Calico

				// TODO: UDP vs TCP
				// TODO: Protocol 4 for calico
			},
		}
		baseGroup.Tags = b.CloudTags(name, false)
	} else if role == kops.InstanceGroupRoleNode {
		name := b.SecurityGroupName(role)
		baseGroup = &awstasks.SecurityGroup{
			Name:             s(name),
			VPC:              b.LinkToVPC(),
			Description:      s("Security group for nodes"),
			RemoveExtraRules: []string{"port=22"},
		}
		baseGroup.Tags = b.CloudTags(name, false)
	} else if role == kops.InstanceGroupRoleBastion {
		name := b.SecurityGroupName(role)
		baseGroup = &awstasks.SecurityGroup{
			Name:             s(name),
			VPC:              b.LinkToVPC(),
			Description:      s("Security group for bastion"),
			RemoveExtraRules: []string{"port=22"},
		}
		baseGroup.Tags = b.CloudTags(name, false)
	} else {
		return nil, fmt.Errorf("not a supported security group type")
	}

	var groups []SecurityGroupInfo

	done := make(map[string]bool)

	// Build groups that specify a SecurityGroupOverride
	allOverrides := true
	for _, ig := range b.InstanceGroups {
		if ig.Spec.Role != role {
			continue
		}

		if ig.Spec.SecurityGroupOverride == nil {
			allOverrides = false
			continue
		}

		name := fi.StringValue(ig.Spec.SecurityGroupOverride)

		// De-duplicate security groups
		if done[name] {
			continue
		}
		done[name] = true

		sgName := fmt.Sprintf("%v-%v", fi.StringValue(ig.Spec.SecurityGroupOverride), role)
		t := &awstasks.SecurityGroup{
			Name:        &sgName,
			ID:          ig.Spec.SecurityGroupOverride,
			VPC:         b.LinkToVPC(),
			Shared:      fi.Bool(true),
			Description: baseGroup.Description,
		}
		// Because the SecurityGroup is shared, we don't set RemoveExtraRules
		// This does mean we don't check them.  We might want to revisit this in future.

		suffix := "-" + name

		groups = append(groups, SecurityGroupInfo{
			Name:   name,
			Suffix: suffix,
			Task:   t,
		})
	}

	// Add the default SecurityGroup, if any InstanceGroups are using the default
	if !allOverrides {
		groups = append(groups, SecurityGroupInfo{
			Name: fi.StringValue(baseGroup.Name),
			Task: baseGroup,
		})
	}

	return groups, nil
}

// JoinSuffixes constructs a suffix for traffic from the src to the dest group
// We have to avoid ambiguity in the case where one has a suffix and the other does not,
// where normally l.Suffix + r.Suffix would equal r.Suffix + l.Suffix
func JoinSuffixes(src SecurityGroupInfo, dest SecurityGroupInfo) string {
	if src.Suffix == "" && dest.Suffix == "" {
		return ""
	}

	s := src.Suffix
	if s == "" {
		s = "-default"
	}

	d := dest.Suffix
	if d == "" {
		d = "-default"
	}

	return s + d
}