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
|
//go:build linux || freebsd
// +build linux freebsd
package libpod
import (
"fmt"
"regexp"
"github.com/containers/common/libnetwork/types"
"github.com/containers/common/pkg/machine"
"github.com/containers/podman/v4/libpod/define"
"github.com/containers/podman/v4/pkg/namespaces"
"github.com/sirupsen/logrus"
)
// convertPortMappings will remove the HostIP part from the ports when running inside podman machine.
// This is need because a HostIP of 127.0.0.1 would now allow the gvproxy forwarder to reach to open ports.
// For machine the HostIP must only be used by gvproxy and never in the VM.
func (c *Container) convertPortMappings() []types.PortMapping {
if !machine.IsGvProxyBased() || len(c.config.PortMappings) == 0 {
return c.config.PortMappings
}
// if we run in a machine VM we have to ignore the host IP part
newPorts := make([]types.PortMapping, 0, len(c.config.PortMappings))
for _, port := range c.config.PortMappings {
port.HostIP = ""
newPorts = append(newPorts, port)
}
return newPorts
}
func (c *Container) getNetworkOptions(networkOpts map[string]types.PerNetworkOptions) types.NetworkOptions {
opts := types.NetworkOptions{
ContainerID: c.config.ID,
ContainerName: getCNIPodName(c),
}
opts.PortMappings = c.convertPortMappings()
// If the container requested special network options use this instead of the config.
// This is the case for container restore or network reload.
if c.perNetworkOpts != nil {
opts.Networks = c.perNetworkOpts
} else {
opts.Networks = networkOpts
}
return opts
}
// setUpNetwork will set up the the networks, on error it will also tear down the cni
// networks. If rootless it will join/create the rootless network namespace.
func (r *Runtime) setUpNetwork(ns string, opts types.NetworkOptions) (map[string]types.StatusBlock, error) {
rootlessNetNS, err := r.GetRootlessNetNs(true)
if err != nil {
return nil, err
}
var results map[string]types.StatusBlock
setUpPod := func() error {
results, err = r.network.Setup(ns, types.SetupOptions{NetworkOptions: opts})
return err
}
// rootlessNetNS is nil if we are root
if rootlessNetNS != nil {
// execute the setup in the rootless net ns
err = rootlessNetNS.Do(setUpPod)
rootlessNetNS.Lock.Unlock()
} else {
err = setUpPod()
}
return results, err
}
// getCNIPodName return the pod name (hostname) used by CNI and the dnsname plugin.
// If we are in the pod network namespace use the pod name otherwise the container name
func getCNIPodName(c *Container) string {
if c.config.NetMode.IsPod() || c.IsInfra() {
pod, err := c.runtime.state.Pod(c.PodID())
if err == nil {
return pod.Name()
}
}
return c.Name()
}
// Tear down a container's network configuration and joins the
// rootless net ns as rootless user
func (r *Runtime) teardownNetwork(ns string, opts types.NetworkOptions) error {
rootlessNetNS, err := r.GetRootlessNetNs(false)
if err != nil {
return err
}
tearDownPod := func() error {
if err := r.network.Teardown(ns, types.TeardownOptions{NetworkOptions: opts}); err != nil {
return fmt.Errorf("tearing down network namespace configuration for container %s: %w", opts.ContainerID, err)
}
return nil
}
// rootlessNetNS is nil if we are root
if rootlessNetNS != nil {
// execute the cni setup in the rootless net ns
err = rootlessNetNS.Do(tearDownPod)
if cerr := rootlessNetNS.Cleanup(r); cerr != nil {
logrus.WithError(err).Error("failed to clean up rootless netns")
}
rootlessNetNS.Lock.Unlock()
} else {
err = tearDownPod()
}
return err
}
// Tear down a container's CNI network configuration, but do not tear down the
// namespace itself.
func (r *Runtime) teardownCNI(ctr *Container) error {
if ctr.state.NetNS == nil {
// The container has no network namespace, we're set
return nil
}
logrus.Debugf("Tearing down network namespace at %s for container %s", ctr.state.NetNS.Path(), ctr.ID())
networks, err := ctr.networks()
if err != nil {
return err
}
if !ctr.config.NetMode.IsSlirp4netns() && len(networks) > 0 {
netOpts := ctr.getNetworkOptions(networks)
return r.teardownNetwork(ctr.state.NetNS.Path(), netOpts)
}
return nil
}
// isBridgeNetMode checks if the given network mode is bridge.
// It returns nil when it is set to bridge and an error otherwise.
func isBridgeNetMode(n namespaces.NetworkMode) error {
if !n.IsBridge() {
return fmt.Errorf("%q is not supported: %w", n, define.ErrNetworkModeInvalid)
}
return nil
}
// Reload only works with containers with a configured network.
// It will tear down, and then reconfigure, the network of the container.
// This is mainly used when a reload of firewall rules wipes out existing
// firewall configuration.
// Efforts will be made to preserve MAC and IP addresses, but this only works if
// the container only joined a single CNI network, and was only assigned a
// single MAC or IP.
// Only works on root containers at present, though in the future we could
// extend this to stop + restart slirp4netns
func (r *Runtime) reloadContainerNetwork(ctr *Container) (map[string]types.StatusBlock, error) {
if ctr.state.NetNS == nil {
return nil, fmt.Errorf("container %s network is not configured, refusing to reload: %w", ctr.ID(), define.ErrCtrStateInvalid)
}
if err := isBridgeNetMode(ctr.config.NetMode); err != nil {
return nil, err
}
logrus.Infof("Going to reload container %s network", ctr.ID())
err := r.teardownCNI(ctr)
if err != nil {
// teardownCNI will error if the iptables rules do not exists and this is the case after
// a firewall reload. The purpose of network reload is to recreate the rules if they do
// not exists so we should not log this specific error as error. This would confuse users otherwise.
// iptables-legacy and iptables-nft will create different errors make sure to match both.
b, rerr := regexp.MatchString("Couldn't load target `CNI-[a-f0-9]{24}':No such file or directory|Chain 'CNI-[a-f0-9]{24}' does not exist", err.Error())
if rerr == nil && !b {
logrus.Error(err)
} else {
logrus.Info(err)
}
}
networkOpts, err := ctr.networks()
if err != nil {
return nil, err
}
// Set the same network settings as before..
netStatus := ctr.getNetworkStatus()
for network, perNetOpts := range networkOpts {
for name, netInt := range netStatus[network].Interfaces {
perNetOpts.InterfaceName = name
perNetOpts.StaticMAC = netInt.MacAddress
for _, netAddress := range netInt.Subnets {
perNetOpts.StaticIPs = append(perNetOpts.StaticIPs, netAddress.IPNet.IP)
}
// Normally interfaces have a length of 1, only for some special cni configs we could get more.
// For now just use the first interface to get the ips this should be good enough for most cases.
break
}
networkOpts[network] = perNetOpts
}
ctr.perNetworkOpts = networkOpts
return r.configureNetNS(ctr, ctr.state.NetNS)
}
|