diff options
Diffstat (limited to 'libpod')
-rw-r--r-- | libpod/kube.go | 112 | ||||
-rw-r--r-- | libpod/networking_linux.go | 8 | ||||
-rw-r--r-- | libpod/oci_conmon_exec_linux.go | 48 | ||||
-rw-r--r-- | libpod/rootless_cni_linux.go | 34 | ||||
-rw-r--r-- | libpod/util.go | 17 |
5 files changed, 181 insertions, 38 deletions
diff --git a/libpod/kube.go b/libpod/kube.go index 753c58099..b5197293e 100644 --- a/libpod/kube.go +++ b/libpod/kube.go @@ -171,9 +171,10 @@ func (p *Pod) podWithContainers(containers []*Container, ports []v1.ContainerPor deDupPodVolumes := make(map[string]*v1.Volume) first := true podContainers := make([]v1.Container, 0, len(containers)) + dnsInfo := v1.PodDNSConfig{} for _, ctr := range containers { if !ctr.IsInfra() { - ctr, volumes, err := containerToV1Container(ctr) + ctr, volumes, _, err := containerToV1Container(ctr) if err != nil { return nil, err } @@ -196,6 +197,22 @@ func (p *Pod) podWithContainers(containers []*Container, ports []v1.ContainerPor vol := vol deDupPodVolumes[vol.Name] = &vol } + } else { + _, _, infraDNS, err := containerToV1Container(ctr) + if err != nil { + return nil, err + } + if infraDNS != nil { + if servers := infraDNS.Nameservers; len(servers) > 0 { + dnsInfo.Nameservers = servers + } + if searches := infraDNS.Searches; len(searches) > 0 { + dnsInfo.Searches = searches + } + if options := infraDNS.Options; len(options) > 0 { + dnsInfo.Options = options + } + } } } podVolumes := make([]v1.Volume, 0, len(deDupPodVolumes)) @@ -203,10 +220,10 @@ func (p *Pod) podWithContainers(containers []*Container, ports []v1.ContainerPor podVolumes = append(podVolumes, *vol) } - return addContainersAndVolumesToPodObject(podContainers, podVolumes, p.Name()), nil + return addContainersAndVolumesToPodObject(podContainers, podVolumes, p.Name(), &dnsInfo), nil } -func addContainersAndVolumesToPodObject(containers []v1.Container, volumes []v1.Volume, podName string) *v1.Pod { +func addContainersAndVolumesToPodObject(containers []v1.Container, volumes []v1.Volume, podName string, dnsOptions *v1.PodDNSConfig) *v1.Pod { tm := v12.TypeMeta{ Kind: "Pod", APIVersion: "v1", @@ -228,6 +245,9 @@ func addContainersAndVolumesToPodObject(containers []v1.Container, volumes []v1. Containers: containers, Volumes: volumes, } + if dnsOptions != nil { + ps.DNSConfig = dnsOptions + } p := v1.Pod{ TypeMeta: tm, ObjectMeta: om, @@ -241,32 +261,65 @@ func addContainersAndVolumesToPodObject(containers []v1.Container, volumes []v1. func simplePodWithV1Containers(ctrs []*Container) (*v1.Pod, error) { kubeCtrs := make([]v1.Container, 0, len(ctrs)) kubeVolumes := make([]v1.Volume, 0) + podDNS := v1.PodDNSConfig{} for _, ctr := range ctrs { - kubeCtr, kubeVols, err := containerToV1Container(ctr) + kubeCtr, kubeVols, ctrDNS, err := containerToV1Container(ctr) if err != nil { return nil, err } kubeCtrs = append(kubeCtrs, kubeCtr) kubeVolumes = append(kubeVolumes, kubeVols...) - } - return addContainersAndVolumesToPodObject(kubeCtrs, kubeVolumes, strings.ReplaceAll(ctrs[0].Name(), "_", "")), nil + // Combine DNS information in sum'd structure + if ctrDNS != nil { + // nameservers + if servers := ctrDNS.Nameservers; servers != nil { + if podDNS.Nameservers == nil { + podDNS.Nameservers = make([]string, 0) + } + for _, s := range servers { + if !util.StringInSlice(s, podDNS.Nameservers) { // only append if it does not exist + podDNS.Nameservers = append(podDNS.Nameservers, s) + } + } + } + // search domains + if domains := ctrDNS.Searches; domains != nil { + if podDNS.Searches == nil { + podDNS.Searches = make([]string, 0) + } + for _, d := range domains { + if !util.StringInSlice(d, podDNS.Searches) { // only append if it does not exist + podDNS.Searches = append(podDNS.Searches, d) + } + } + } + // dns options + if options := ctrDNS.Options; options != nil { + if podDNS.Options == nil { + podDNS.Options = make([]v1.PodDNSConfigOption, 0) + } + podDNS.Options = append(podDNS.Options, options...) + } + } // end if ctrDNS + } + return addContainersAndVolumesToPodObject(kubeCtrs, kubeVolumes, strings.ReplaceAll(ctrs[0].Name(), "_", ""), &podDNS), nil } // containerToV1Container converts information we know about a libpod container // to a V1.Container specification. -func containerToV1Container(c *Container) (v1.Container, []v1.Volume, error) { +func containerToV1Container(c *Container) (v1.Container, []v1.Volume, *v1.PodDNSConfig, error) { kubeContainer := v1.Container{} kubeVolumes := []v1.Volume{} kubeSec, err := generateKubeSecurityContext(c) if err != nil { - return kubeContainer, kubeVolumes, err + return kubeContainer, kubeVolumes, nil, err } if len(c.config.Spec.Linux.Devices) > 0 { // TODO Enable when we can support devices and their names kubeContainer.VolumeDevices = generateKubeVolumeDeviceFromLinuxDevice(c.Spec().Linux.Devices) - return kubeContainer, kubeVolumes, errors.Wrapf(define.ErrNotImplemented, "linux devices") + return kubeContainer, kubeVolumes, nil, errors.Wrapf(define.ErrNotImplemented, "linux devices") } if len(c.config.UserVolumes) > 0 { @@ -274,7 +327,7 @@ func containerToV1Container(c *Container) (v1.Container, []v1.Volume, error) { // Volume names need to be coordinated "globally" in the kube files. volumeMounts, volumes, err := libpodMountsToKubeVolumeMounts(c) if err != nil { - return kubeContainer, kubeVolumes, err + return kubeContainer, kubeVolumes, nil, err } kubeContainer.VolumeMounts = volumeMounts kubeVolumes = append(kubeVolumes, volumes...) @@ -282,16 +335,16 @@ func containerToV1Container(c *Container) (v1.Container, []v1.Volume, error) { envVariables, err := libpodEnvVarsToKubeEnvVars(c.config.Spec.Process.Env) if err != nil { - return kubeContainer, kubeVolumes, err + return kubeContainer, kubeVolumes, nil, err } portmappings, err := c.PortMappings() if err != nil { - return kubeContainer, kubeVolumes, err + return kubeContainer, kubeVolumes, nil, err } ports, err := ocicniPortMappingToContainerPort(portmappings) if err != nil { - return kubeContainer, kubeVolumes, err + return kubeContainer, kubeVolumes, nil, err } containerCommands := c.Command() @@ -355,7 +408,38 @@ func containerToV1Container(c *Container) (v1.Container, []v1.Volume, error) { } } - return kubeContainer, kubeVolumes, nil + // Obtain the DNS entries from the container + dns := v1.PodDNSConfig{} + + // DNS servers + if servers := c.config.DNSServer; len(servers) > 0 { + dnsServers := make([]string, 0) + for _, server := range servers { + dnsServers = append(dnsServers, server.String()) + } + dns.Nameservers = dnsServers + } + + // DNS search domains + if searches := c.config.DNSSearch; len(searches) > 0 { + dns.Searches = searches + } + + // DNS options + if options := c.config.DNSOption; len(options) > 0 { + dnsOptions := make([]v1.PodDNSConfigOption, 0) + for _, option := range options { + // the option can be "k:v" or just "k", no delimiter is required + opts := strings.SplitN(option, ":", 2) + dnsOpt := v1.PodDNSConfigOption{ + Name: opts[0], + Value: &opts[1], + } + dnsOptions = append(dnsOptions, dnsOpt) + } + dns.Options = dnsOptions + } + return kubeContainer, kubeVolumes, &dns, nil } // ocicniPortMappingToContainerPort takes an ocicni portmapping and converts diff --git a/libpod/networking_linux.go b/libpod/networking_linux.go index ef2f034ab..737dbf935 100644 --- a/libpod/networking_linux.go +++ b/libpod/networking_linux.go @@ -977,7 +977,7 @@ func (c *Container) getContainerNetworkInfo() (*define.InspectNetworkSettings, e if c.state.NetNS == nil { // We still want to make dummy configurations for each CNI net // the container joined. - if len(networks) > 0 && !isDefault { + if len(networks) > 0 { settings.Networks = make(map[string]*define.InspectAdditionalNetwork, len(networks)) for _, net := range networks { cniNet := new(define.InspectAdditionalNetwork) @@ -998,7 +998,7 @@ func (c *Container) getContainerNetworkInfo() (*define.InspectNetworkSettings, e } // If we have CNI networks - handle that here - if len(networks) > 0 && !isDefault { + if len(networks) > 0 { if len(networks) != len(c.state.NetworkStatus) { return nil, errors.Wrapf(define.ErrInternal, "network inspection mismatch: asked to join %d CNI network(s) %v, but have information on %d network(s)", len(networks), networks, len(c.state.NetworkStatus)) } @@ -1028,7 +1028,9 @@ func (c *Container) getContainerNetworkInfo() (*define.InspectNetworkSettings, e settings.Networks[name] = addedNet } - return settings, nil + if !isDefault { + return settings, nil + } } // If not joining networks, we should have at most 1 result diff --git a/libpod/oci_conmon_exec_linux.go b/libpod/oci_conmon_exec_linux.go index dc5dd03df..faf86ea5b 100644 --- a/libpod/oci_conmon_exec_linux.go +++ b/libpod/oci_conmon_exec_linux.go @@ -126,20 +126,25 @@ func (r *ConmonOCIRuntime) ExecContainerHTTP(ctr *Container, sessionID string, o }() attachChan := make(chan error) + conmonPipeDataChan := make(chan conmonPipeData) go func() { // attachToExec is responsible for closing pipes - attachChan <- attachExecHTTP(ctr, sessionID, req, w, streams, pipes, detachKeys, options.Terminal, cancel, hijackDone, holdConnOpen) + attachChan <- attachExecHTTP(ctr, sessionID, req, w, streams, pipes, detachKeys, options.Terminal, cancel, hijackDone, holdConnOpen, execCmd, conmonPipeDataChan, ociLog) close(attachChan) }() - // Wait for conmon to succeed, when return. - if err := execCmd.Wait(); err != nil { - return -1, nil, errors.Wrapf(err, "cannot run conmon") - } + // NOTE: the channel is needed to communicate conmon's data. In case + // of an error, the error will be written on the hijacked http + // connection such that remote clients will receive the error. + pipeData := <-conmonPipeDataChan - pid, err := readConmonPipeData(pipes.syncPipe, ociLog) + return pipeData.pid, attachChan, pipeData.err +} - return pid, attachChan, err +// conmonPipeData contains the data when reading from conmon's pipe. +type conmonPipeData struct { + pid int + err error } // ExecContainerDetached executes a command in a running container, but does @@ -488,9 +493,16 @@ func (r *ConmonOCIRuntime) startExec(c *Container, sessionID string, options *Ex } // Attach to a container over HTTP -func attachExecHTTP(c *Container, sessionID string, r *http.Request, w http.ResponseWriter, streams *HTTPAttachStreams, pipes *execPipes, detachKeys []byte, isTerminal bool, cancel <-chan bool, hijackDone chan<- bool, holdConnOpen <-chan bool) (deferredErr error) { +func attachExecHTTP(c *Container, sessionID string, r *http.Request, w http.ResponseWriter, streams *HTTPAttachStreams, pipes *execPipes, detachKeys []byte, isTerminal bool, cancel <-chan bool, hijackDone chan<- bool, holdConnOpen <-chan bool, execCmd *exec.Cmd, conmonPipeDataChan chan<- conmonPipeData, ociLog string) (deferredErr error) { + // NOTE: As you may notice, the attach code is quite complex. + // Many things happen concurrently and yet are interdependent. + // If you ever change this function, make sure to write to the + // conmonPipeDataChan in case of an error. + if pipes == nil || pipes.startPipe == nil || pipes.attachPipe == nil { - return errors.Wrapf(define.ErrInvalidArg, "must provide a start and attach pipe to finish an exec attach") + err := errors.Wrapf(define.ErrInvalidArg, "must provide a start and attach pipe to finish an exec attach") + conmonPipeDataChan <- conmonPipeData{-1, err} + return err } defer func() { @@ -509,17 +521,20 @@ func attachExecHTTP(c *Container, sessionID string, r *http.Request, w http.Resp // set up the socket path, such that it is the correct length and location for exec sockPath, err := c.execAttachSocketPath(sessionID) if err != nil { + conmonPipeDataChan <- conmonPipeData{-1, err} return err } // 2: read from attachFd that the parent process has set up the console socket if _, err := readConmonPipeData(pipes.attachPipe, ""); err != nil { + conmonPipeDataChan <- conmonPipeData{-1, err} return err } // 2: then attach conn, err := openUnixSocket(sockPath) if err != nil { + conmonPipeDataChan <- conmonPipeData{-1, err} return errors.Wrapf(err, "failed to connect to container's attach socket: %v", sockPath) } defer func() { @@ -540,11 +555,13 @@ func attachExecHTTP(c *Container, sessionID string, r *http.Request, w http.Resp // Perform hijack hijacker, ok := w.(http.Hijacker) if !ok { + conmonPipeDataChan <- conmonPipeData{-1, err} return errors.Errorf("unable to hijack connection") } httpCon, httpBuf, err := hijacker.Hijack() if err != nil { + conmonPipeDataChan <- conmonPipeData{-1, err} return errors.Wrapf(err, "error hijacking connection") } @@ -555,10 +572,23 @@ func attachExecHTTP(c *Container, sessionID string, r *http.Request, w http.Resp // Force a flush after the header is written. if err := httpBuf.Flush(); err != nil { + conmonPipeDataChan <- conmonPipeData{-1, err} return errors.Wrapf(err, "error flushing HTTP hijack header") } go func() { + // Wait for conmon to succeed, when return. + if err := execCmd.Wait(); err != nil { + conmonPipeDataChan <- conmonPipeData{-1, err} + } else { + pid, err := readConmonPipeData(pipes.syncPipe, ociLog) + if err != nil { + hijackWriteError(err, c.ID(), isTerminal, httpBuf) + conmonPipeDataChan <- conmonPipeData{pid, err} + } else { + conmonPipeDataChan <- conmonPipeData{pid, err} + } + } // We need to hold the connection open until the complete exec // function has finished. This channel will be closed in a defer // in that function, so we can wait for it here. diff --git a/libpod/rootless_cni_linux.go b/libpod/rootless_cni_linux.go index 9a980750f..94ae062aa 100644 --- a/libpod/rootless_cni_linux.go +++ b/libpod/rootless_cni_linux.go @@ -25,7 +25,7 @@ import ( // Built from ../contrib/rootless-cni-infra. var rootlessCNIInfraImage = map[string]string{ - "amd64": "quay.io/libpod/rootless-cni-infra@sha256:304742d5d221211df4ec672807a5842ff11e3729c50bc424ea0cea858f69d7b7", // 3-amd64 + "amd64": "quay.io/libpod/rootless-cni-infra@sha256:adf352454666f7ce9ca3e1098448b5ee18f89c4516471ec99447ec9ece917f36", // 5-amd64 } const ( @@ -58,9 +58,33 @@ func AllocRootlessCNI(ctx context.Context, c *Container) (ns.NetNS, []*cnitypes. return nil, nil, err } k8sPodName := getCNIPodName(c) // passed to CNI as K8S_POD_NAME + ip := "" + if c.config.StaticIP != nil { + ip = c.config.StaticIP.String() + } + mac := "" + if c.config.StaticMAC != nil { + mac = c.config.StaticMAC.String() + } + aliases, err := c.runtime.state.GetAllNetworkAliases(c) + if err != nil { + return nil, nil, err + } + capArgs := "" + // add network aliases json encoded as capabilityArgs for cni + if len(aliases) > 0 { + capabilityArgs := make(map[string]interface{}) + capabilityArgs["aliases"] = aliases + b, err := json.Marshal(capabilityArgs) + if err != nil { + return nil, nil, err + } + capArgs = string(b) + } + cniResults := make([]*cnitypes.Result, len(networks)) for i, nw := range networks { - cniRes, err := rootlessCNIInfraCallAlloc(infra, c.ID(), nw, k8sPodName) + cniRes, err := rootlessCNIInfraCallAlloc(infra, c.ID(), nw, k8sPodName, ip, mac, capArgs) if err != nil { return nil, nil, err } @@ -137,11 +161,11 @@ func getCNIPodName(c *Container) string { return c.Name() } -func rootlessCNIInfraCallAlloc(infra *Container, id, nw, k8sPodName string) (*cnitypes.Result, error) { - logrus.Debugf("rootless CNI: alloc %q, %q, %q", id, nw, k8sPodName) +func rootlessCNIInfraCallAlloc(infra *Container, id, nw, k8sPodName, ip, mac, capArgs string) (*cnitypes.Result, error) { + logrus.Debugf("rootless CNI: alloc %q, %q, %q, %q, %q, %q", id, nw, k8sPodName, ip, mac, capArgs) var err error - _, err = rootlessCNIInfraExec(infra, "alloc", id, nw, k8sPodName) + _, err = rootlessCNIInfraExec(infra, "alloc", id, nw, k8sPodName, ip, mac, capArgs) if err != nil { return nil, err } diff --git a/libpod/util.go b/libpod/util.go index bf9bf2542..391208fb9 100644 --- a/libpod/util.go +++ b/libpod/util.go @@ -235,20 +235,16 @@ func checkDependencyContainer(depCtr, ctr *Container) error { return nil } -// hijackWriteErrorAndClose writes an error to a hijacked HTTP session and -// closes it. Intended to HTTPAttach function. -// If error is nil, it will not be written; we'll only close the connection. -func hijackWriteErrorAndClose(toWrite error, cid string, terminal bool, httpCon io.Closer, httpBuf *bufio.ReadWriter) { +// hijackWriteError writes an error to a hijacked HTTP session. +func hijackWriteError(toWrite error, cid string, terminal bool, httpBuf *bufio.ReadWriter) { if toWrite != nil { - errString := []byte(fmt.Sprintf("%v\n", toWrite)) + errString := []byte(fmt.Sprintf("Error: %v\n", toWrite)) if !terminal { // We need a header. header := makeHTTPAttachHeader(2, uint32(len(errString))) if _, err := httpBuf.Write(header); err != nil { logrus.Errorf("Error writing header for container %s attach connection error: %v", cid, err) } - // TODO: May want to return immediately here to avoid - // writing garbage to the socket? } if _, err := httpBuf.Write(errString); err != nil { logrus.Errorf("Error writing error to container %s HTTP attach connection: %v", cid, err) @@ -257,6 +253,13 @@ func hijackWriteErrorAndClose(toWrite error, cid string, terminal bool, httpCon logrus.Errorf("Error flushing HTTP buffer for container %s HTTP attach connection: %v", cid, err) } } +} + +// hijackWriteErrorAndClose writes an error to a hijacked HTTP session and +// closes it. Intended to HTTPAttach function. +// If error is nil, it will not be written; we'll only close the connection. +func hijackWriteErrorAndClose(toWrite error, cid string, terminal bool, httpCon io.Closer, httpBuf *bufio.ReadWriter) { + hijackWriteError(toWrite, cid, terminal, httpBuf) if err := httpCon.Close(); err != nil { logrus.Errorf("Error closing container %s HTTP attach connection: %v", cid, err) |