summaryrefslogtreecommitdiff
path: root/libpod/networking.go
blob: 092ce2a3fa15edb50b30d1d046af6ec5487d8e14 (plain)
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
178
179
180
181
182
183
184
185
186
187
188
189
190
191
192
193
194
195
196
197
198
199
200
201
202
203
204
205
206
207
208
209
package libpod

import (
	"crypto/rand"
	"fmt"
	"net"
	"os"
	"path/filepath"
	"strings"
	"syscall"

	cnitypes "github.com/containernetworking/cni/pkg/types/current"
	"github.com/containernetworking/plugins/pkg/ns"
	"github.com/cri-o/ocicni/pkg/ocicni"
	"github.com/pkg/errors"
	"github.com/projectatomic/libpod/utils"
	"github.com/sirupsen/logrus"
	"golang.org/x/sys/unix"
)

// Get an OCICNI network config
func getPodNetwork(id, name, nsPath string, ports []ocicni.PortMapping) ocicni.PodNetwork {
	return ocicni.PodNetwork{
		Name:         name,
		Namespace:    name, // TODO is there something else we should put here? We don't know about Kube namespaces
		ID:           id,
		NetNS:        nsPath,
		PortMappings: ports,
	}
}

// Create and configure a new network namespace for a container
func (r *Runtime) configureNetNS(ctr *Container, ctrNS ns.NetNS) (err error) {
	podNetwork := getPodNetwork(ctr.ID(), ctr.Name(), ctrNS.Path(), ctr.config.PortMappings)

	result, err := r.netPlugin.SetUpPod(podNetwork)
	if err != nil {
		return errors.Wrapf(err, "error configuring network namespace for container %s", ctr.ID())
	}
	defer func() {
		if err != nil {
			if err2 := r.netPlugin.TearDownPod(podNetwork); err2 != nil {
				logrus.Errorf("Error tearing down partially created network namespace for container %s: %v", ctr.ID(), err2)
			}
		}
	}()

	logrus.Debugf("Response from CNI plugins: %v", result.String())

	resultStruct, err := cnitypes.GetResult(result)
	if err != nil {
		return errors.Wrapf(err, "error parsing result from CNI plugins")
	}

	ctr.state.NetNS = ctrNS
	ctr.state.IPs = resultStruct.IPs
	ctr.state.Routes = resultStruct.Routes

	// We need to temporarily use iptables to allow the container
	// to resolve DNS until this issue is fixed upstream.
	// https://github.com/containernetworking/plugins/pull/75
	if resultStruct.IPs != nil {
		for _, ip := range resultStruct.IPs {
			iptablesDNS("-I", ip.Address.IP.String())
		}
	}
	return nil
}

// Create and configure a new network namespace for a container
func (r *Runtime) createNetNS(ctr *Container) (err error) {
	ctrNS, err := ns.NewNS()
	if err != nil {
		return errors.Wrapf(err, "error creating network namespace for container %s", ctr.ID())
	}
	defer func() {
		if err != nil {
			if err2 := ctrNS.Close(); err2 != nil {
				logrus.Errorf("Error closing partially created network namespace for container %s: %v", ctr.ID(), err2)
			}
		}
	}()

	logrus.Debugf("Made network namespace at %s for container %s", ctrNS.Path(), ctr.ID())
	return r.configureNetNS(ctr, ctrNS)
}

// Configure the network namespace using the container process
func (r *Runtime) setupNetNS(ctr *Container) (err error) {
	nsProcess := fmt.Sprintf("/proc/%d/ns/net", ctr.state.PID)

	b := make([]byte, 16)

	if _, err := rand.Reader.Read(b); err != nil {
		return errors.Wrapf(err, "failed to generate random netns name")
	}

	nsPath := fmt.Sprintf("/var/run/netns/cni-%x-%x-%x-%x-%x", b[0:4], b[4:6], b[6:8], b[8:10], b[10:])

	if err := os.MkdirAll(filepath.Dir(nsPath), 0711); err != nil {
		return errors.Wrapf(err, "cannot create %s", filepath.Dir(nsPath))
	}

	mountPointFd, err := os.Create(nsPath)
	if err != nil {
		return errors.Wrapf(err, "cannot open %s", nsPath)
	}
	mountPointFd.Close()

	if err := unix.Mount(nsProcess, nsPath, "none", unix.MS_BIND, ""); err != nil {
		return errors.Wrapf(err, "cannot mount %s", nsPath)
	}

	netNS, err := ns.GetNS(nsPath)
	if err != nil {
		return err
	}
	return r.configureNetNS(ctr, netNS)
}

// iptablesDNS accepts an arg (-I|-D) and IP address of the container and then
// generates an iptables command to either add or subtract the needed rule
func iptablesDNS(arg, ip string) error {
	iptablesCmd := []string{"-t", "filter", arg, "FORWARD", "-s", ip, "!", "-o", ip, "-j", "ACCEPT"}
	logrus.Debug("Running iptables command: ", strings.Join(iptablesCmd, " "))
	_, err := utils.ExecCmd("iptables", iptablesCmd...)
	if err != nil {
		logrus.Error(err)
	}
	return err
}

// Join an existing network namespace
func joinNetNS(path string) (ns.NetNS, error) {
	ns, err := ns.GetNS(path)
	if err != nil {
		return nil, errors.Wrapf(err, "error retrieving network namespace at %s", path)
	}

	return ns, nil
}

// Get a container's IP address
func (r *Runtime) getContainerIP(ctr *Container) (net.IP, error) {
	if ctr.state.NetNS == nil {
		return nil, errors.Wrapf(ErrInvalidArg, "container %s has no network namespace, cannot get IP", ctr.ID())
	}

	podNetwork := getPodNetwork(ctr.ID(), ctr.Name(), ctr.state.NetNS.Path(), ctr.config.PortMappings)

	ipStr, err := r.netPlugin.GetPodNetworkStatus(podNetwork)
	if err != nil {
		return nil, errors.Wrapf(err, "error retrieving network status of container %s", ctr.ID())
	}

	ip := net.ParseIP(ipStr)
	if ip == nil {
		return nil, errors.Wrapf(ErrInternal, "error parsing IP address %s for container %s", ipStr, ctr.ID())
	}

	return ip, nil
}

// Tear down a network namespace
func (r *Runtime) teardownNetNS(ctr *Container) error {
	if ctr.state.NetNS == nil {
		// The container has no network namespace, we're set
		return nil
	}

	// Because we are using iptables to allow the container to resolve DNS
	// on per IP address, we also need to try to remove the iptables rule
	// on cleanup. Remove when https://github.com/containernetworking/plugins/pull/75
	// is merged.
	for _, ip := range ctr.state.IPs {
		iptablesDNS("-D", ip.Address.IP.String())
	}

	logrus.Debugf("Tearing down network namespace at %s for container %s", ctr.state.NetNS.Path(), ctr.ID())

	podNetwork := getPodNetwork(ctr.ID(), ctr.Name(), ctr.state.NetNS.Path(), ctr.config.PortMappings)

	// The network may have already been torn down, so don't fail here, just log
	if err := r.netPlugin.TearDownPod(podNetwork); err != nil {
		logrus.Errorf("Failed to tear down network namespace for container %s: %v", ctr.ID(), err)
	}

	nsPath := ctr.state.NetNS.Path()

	if err := ctr.state.NetNS.Close(); err != nil {
		return errors.Wrapf(err, "error closing network namespace for container %s", ctr.ID())
	}

	// We need to unconditionally try to unmount/remove the namespace
	// because we may be in a separate process from the one that created the
	// namespace, and Close() will only do that if it is the same process.
	if err := unix.Unmount(nsPath, unix.MNT_DETACH); err != nil {
		if err != syscall.EINVAL && err != syscall.ENOENT {
			return errors.Wrapf(err, "error unmounting network namespace %s for container %s", nsPath, ctr.ID())
		}
	}
	if err := os.RemoveAll(nsPath); err != nil && !os.IsNotExist(err) {
		return errors.Wrapf(err, "error removing network namespace %s for container %s", nsPath, ctr.ID())
	}

	ctr.state.NetNS = nil

	return nil
}