Merge pull request #7460 from AkihiroSuda/allow-rootless-cni

rootless: support `podman network create` (CNI-in-slirp4netns)

cmd/podman/networks/create.go

@@ -21,9 +21,6 @@ var (
 		RunE:    networkCreate,
 		Args:    cobra.MaximumNArgs(1),
 		Example: `podman network create podman1`,
-		Annotations: map[string]string{
-			registry.ParentNSRequired: "",
-		},
 	}
 )
 

cmd/podman/networks/inspect.go

@@ -22,9 +22,6 @@ var (
 		RunE:    networkInspect,
 		Example: `podman network inspect podman`,
 		Args:    cobra.MinimumNArgs(1),
-		Annotations: map[string]string{
-			registry.ParentNSRequired: "",
-		},
 	}
 )
 

cmd/podman/networks/list.go

@@ -25,9 +25,6 @@ var (
 		Long:    networklistDescription,
 		RunE:    networkList,
 		Example: `podman network list`,
-		Annotations: map[string]string{
-			registry.ParentNSRequired: "",
-		},
 	}
 )
 

cmd/podman/networks/rm.go

@@ -19,9 +19,6 @@ var (
 		RunE:    networkRm,
 		Example: `podman network rm podman`,
 		Args:    cobra.MinimumNArgs(1),
-		Annotations: map[string]string{
-			registry.ParentNSRequired: "",
-		},
 	}
 )
 

contrib/rootless-cni-infra/Containerfile

@@ -0,0 +1,35 @@
+ARG GOLANG_VERSION=1.15
+ARG ALPINE_VERSION=3.12
+ARG CNI_VERSION=v0.8.0
+ARG CNI_PLUGINS_VERSION=v0.8.7
+# Aug 20, 2020
+ARG DNSNAME_VESION=78b4da7bbfc51c27366da630e1df1c4f2e8b1b5b
+
+FROM golang:${GOLANG_VERSION}-alpine${ALPINE_VERSION} AS golang-base
+RUN apk add --no-cache git
+
+FROM golang-base AS cnitool
+RUN git clone https://github.com/containernetworking/cni /go/src/github.com/containernetworking/cni
+WORKDIR /go/src/github.com/containernetworking/cni
+ARG CNI_VERSION
+RUN git checkout ${CNI_VERSION}
+RUN go build -o /cnitool ./cnitool
+
+FROM golang-base AS dnsname
+RUN git clone https://github.com/containers/dnsname /go/src/github.com/containers/dnsname
+WORKDIR /go/src/github.com/containers/dnsname
+ARG DNSNAME_VERSION
+RUN git checkout ${DNSNAME_VERSION}
+RUN go build -o /dnsname ./plugins/meta/dnsname
+
+FROM alpine:${ALPINE_VERSION}
+RUN apk add --no-cache curl dnsmasq iptables ip6tables iproute2
+ARG TARGETARCH
+ARG CNI_PLUGINS_VERSION
+RUN mkdir -p /opt/cni/bin && \
+    curl -fsSL https://github.com/containernetworking/plugins/releases/download/${CNI_PLUGINS_VERSION}/cni-plugins-linux-${TARGETARCH}-${CNI_PLUGINS_VERSION}.tgz | tar xz -C /opt/cni/bin
+COPY --from=cnitool /cnitool /usr/local/bin
+COPY --from=dnsname /dnsname /opt/cni/bin
+COPY rootless-cni-infra /usr/local/bin
+ENV CNI_PATH=/opt/cni/bin
+CMD ["sleep", "infinity"]

contrib/rootless-cni-infra/README.md

@@ -0,0 +1,22 @@
+# rootless-cni-infra
+
+Infra container for CNI-in-slirp4netns.
+
+## How it works
+
+When a CNI network is specified for `podman run` in rootless mode, Podman launches the `rootless-cni-infra` container to execute CNI plugins inside slirp4netns.
+
+The infra container is created per user, by executing an equivalent of:
+`podman run -d --name rootless-cni-infra --pid=host --privileged -v $HOME/.config/cni/net.d:/etc/cni/net.d rootless-cni-infra`.
+The infra container is automatically deleted when no CNI network is in use.
+
+Podman then allocates a CNI netns in the infra container, by executing an equivalent of:
+`podman exec rootless-cni-infra rootless-cni-infra alloc $CONTAINER_ID $NETWORK_NAME $POD_NAME`.
+
+The allocated netns is deallocated when the container is being removed, by executing an equivalent of:
+`podman exec rootless-cni-infra rootless-cni-infra dealloc $CONTAINER_ID $NETWORK_NAME`.
+
+## Directory layout
+
+* `/run/rootless-cni-infra/${CONTAINER_ID}/pid`: PID of the `sleep infinity` process that corresponds to the allocated netns
+* `/run/rootless-cni-infra/${CONTAINER_ID}/attached/${NETWORK_NAME}`: CNI result

contrib/rootless-cni-infra/rootless-cni-infra

@@ -0,0 +1,147 @@
+#!/bin/sh
+set -eu
+
+ARG0="$0"
+VERSION="0.1.0"
+BASE="/run/rootless-cni-infra"
+
+# CLI subcommand: "alloc $CONTAINER_ID $NETWORK_NAME $POD_NAME"
+cmd_entrypoint_alloc() {
+	if [ "$#" -ne 3 ]; then
+		echo >&2 "Usage: $ARG0 alloc CONTAINER_ID NETWORK_NAME POD_NAME"
+		exit 1
+	fi
+
+	ID="$1"
+	NET="$2"
+	K8S_POD_NAME="$3"
+
+	dir="${BASE}/${ID}"
+	mkdir -p "${dir}/attached"
+
+	pid=""
+	if [ -f "${dir}/pid" ]; then
+		pid=$(cat "${dir}/pid")
+	else
+		unshare -n sleep infinity &
+		pid="$!"
+		echo "${pid}" >"${dir}/pid"
+		nsenter -t "${pid}" -n ip link set lo up
+	fi
+	CNI_ARGS="IgnoreUnknown=1;K8S_POD_NAME=${K8S_POD_NAME}"
+	nwcount=$(find "${dir}/attached" -type f | wc -l)
+	CNI_IFNAME="eth${nwcount}"
+	export CNI_ARGS CNI_IFNAME
+	cnitool add "${NET}" "/proc/${pid}/ns/net" >"${dir}/attached/${NET}"
+
+	# return the result
+	ns="/proc/${pid}/ns/net"
+	echo "{\"ns\":\"${ns}\"}"
+}
+
+# CLI subcommand: "dealloc $CONTAINER_ID $NETWORK_NAME"
+cmd_entrypoint_dealloc() {
+	if [ "$#" -ne 2 ]; then
+		echo >&2 "Usage: $ARG0 dealloc CONTAINER_ID NETWORK_NAME"
+		exit 1
+	fi
+
+	ID=$1
+	NET=$2
+
+	dir="${BASE}/${ID}"
+	if [ ! -f "${dir}/pid" ]; then
+		exit 0
+	fi
+	pid=$(cat "${dir}/pid")
+	cnitool del "${NET}" "/proc/${pid}/ns/net"
+	rm -f "${dir}/attached/${NET}"
+
+	nwcount=$(find "${dir}/attached" -type f | wc -l)
+	if [ "${nwcount}" = 0 ]; then
+		kill -9 "${pid}"
+		rm -rf "${dir}"
+	fi
+
+	# return empty json
+	echo "{}"
+}
+
+# CLI subcommand: "is-idle"
+cmd_entrypoint_is_idle() {
+	if [ ! -d ${BASE} ]; then
+		echo '{"idle": true}'
+	elif [ -z "$(ls -1 ${BASE})" ]; then
+		echo '{"idle": true}'
+	else
+		echo '{"idle": false}'
+	fi
+}
+
+# CLI subcommand: "print-cni-result $CONTAINER_ID $NETWORK_NAME"
+cmd_entrypoint_print_cni_result() {
+	if [ "$#" -ne 2 ]; then
+		echo >&2 "Usage: $ARG0 print-cni-result CONTAINER_ID NETWORK_NAME"
+		exit 1
+	fi
+
+	ID=$1
+	NET=$2
+
+	# the result shall be CNI JSON
+	cat "${BASE}/${ID}/attached/${NET}"
+}
+
+# CLI subcommand: "print-netns-path $CONTAINER_ID"
+cmd_entrypoint_print_netns_path() {
+	if [ "$#" -ne 1 ]; then
+		echo >&2 "Usage: $ARG0 print-netns-path CONTAINER_ID"
+		exit 1
+	fi
+
+	ID=$1
+
+	pid=$(cat "${BASE}/${ID}/pid")
+	path="/proc/${pid}/ns/net"
+
+	# return the result
+	echo "{\"path\":\"${path}\"}"
+}
+
+# CLI subcommand: "help"
+cmd_entrypoint_help() {
+	echo "Usage: ${ARG0} COMMAND"
+	echo
+	echo "Rootless CNI Infra container"
+	echo
+	echo "Commands:"
+	echo "  alloc             Allocate a netns"
+	echo "  dealloc           Deallocate a netns"
+	echo "  is-idle           Print whether the infra container is idle"
+	echo "  print-cni-result  Print CNI result"
+	echo "  print-netns-path  Print netns path"
+	echo "  help              Print help"
+	echo "  version           Print version"
+}
+
+# CLI subcommand: "version"
+cmd_entrypoint_version() {
+	echo "{\"version\": \"${VERSION}\"}"
+}
+
+# parse args
+command="${1:-}"
+if [ -z "$command" ]; then
+	echo >&2 "No command was specified. Run \`${ARG0} help\` to see the usage."
+	exit 1
+fi
+
+command_func=$(echo "cmd_entrypoint_${command}" | sed -e "s/-/_/g")
+if ! command -v "${command_func}" >/dev/null 2>&1; then
+	echo >&2 "Unknown command: ${command}. Run \`${ARG0} help\` to see the usage."
+	exit 1
+fi
+
+# start the command func
+shift
+"${command_func}" "$@"

libpod/container_internal.go

@@ -957,8 +957,10 @@ func (c *Container) completeNetworkSetup() error {
 	if err := c.syncContainer(); err != nil {
 		return err
 	}
-	if c.config.NetMode.IsSlirp4netns() {
+	if rootless.IsRootless() {
 		return c.runtime.setupRootlessNetNS(c)
+	} else if c.config.NetMode.IsSlirp4netns() {
+		return c.runtime.setupSlirp4netns(c)
 	}
 	if err := c.runtime.setupNetNS(c); err != nil {
 		return err

libpod/container_internal_linux.go

@@ -84,7 +84,11 @@ func (c *Container) prepare() error {
 		// Set up network namespace if not already set up
 		noNetNS := c.state.NetNS == nil
 		if c.config.CreateNetNS && noNetNS && !c.config.PostConfigureNetNS {
-			netNS, networkStatus, createNetNSErr = c.runtime.createNetNS(c)
+			if rootless.IsRootless() && len(c.config.Networks) > 0 {
+				netNS, networkStatus, createNetNSErr = AllocRootlessCNI(context.Background(), c)
+			} else {
+				netNS, networkStatus, createNetNSErr = c.runtime.createNetNS(c)
+			}
 			if createNetNSErr != nil {
 				return
 			}
@@ -98,8 +102,12 @@ func (c *Container) prepare() error {
 		}
 
 		// handle rootless network namespace setup
-		if noNetNS && c.config.NetMode.IsSlirp4netns() && !c.config.PostConfigureNetNS {
-			createNetNSErr = c.runtime.setupRootlessNetNS(c)
+		if noNetNS && !c.config.PostConfigureNetNS {
+			if rootless.IsRootless() {
+				createNetNSErr = c.runtime.setupRootlessNetNS(c)
+			} else if c.config.NetMode.IsSlirp4netns() {
+				createNetNSErr = c.runtime.setupSlirp4netns(c)
+			}
 		}
 	}()
 	// Mount storage if not mounted

libpod/container_validate.go

@@ -2,7 +2,6 @@ package libpod
 
 import (
 	"github.com/containers/podman/v2/libpod/define"
-	"github.com/containers/podman/v2/pkg/rootless"
 	spec "github.com/opencontainers/runtime-spec/specs-go"
 	"github.com/pkg/errors"
 )
@@ -68,16 +67,6 @@ func (c *Container) validate() error {
 		}
 	}
 
-	// Rootless has some requirements, compared to networks.
-	if rootless.IsRootless() {
-		if len(c.config.Networks) > 0 {
-			return errors.Wrapf(define.ErrInvalidArg, "cannot join CNI networks if running rootless")
-		}
-
-		// TODO: Should we make sure network mode is set to Slirp if set
-		// at all?
-	}
-
 	// Can only set static IP or MAC is creating a network namespace.
 	if !c.config.CreateNetNS && (c.config.StaticIP != nil || c.config.StaticMAC != nil) {
 		return errors.Wrapf(define.ErrInvalidArg, "cannot set static IP or MAC address if not creating a network namespace")

libpod/networking_linux.go

@@ -4,6 +4,7 @@ package libpod
 
 import (
 	"bytes"
+	"context"
 	"crypto/rand"
 	"fmt"
 	"io"
@@ -208,6 +209,20 @@ func checkSlirpFlags(path string) (*slirpFeatures, error) {
 
 // Configure the network namespace for a rootless container
 func (r *Runtime) setupRootlessNetNS(ctr *Container) error {
+	if ctr.config.NetMode.IsSlirp4netns() {
+		return r.setupSlirp4netns(ctr)
+	}
+	if len(ctr.config.Networks) > 0 {
+		// set up port forwarder for CNI-in-slirp4netns
+		netnsPath := ctr.state.NetNS.Path()
+		// TODO: support slirp4netns port forwarder as well
+		return r.setupRootlessPortMappingViaRLK(ctr, netnsPath)
+	}
+	return nil
+}
+
+// setupSlirp4netns can be called in rootful as well as in rootless
+func (r *Runtime) setupSlirp4netns(ctr *Container) error {
 	path := r.config.Engine.NetworkCmdPath
 
 	if path == "" {
@@ -711,7 +726,7 @@ func (r *Runtime) teardownNetNS(ctr *Container) error {
 
 	logrus.Debugf("Tearing down network namespace at %s for container %s", ctr.state.NetNS.Path(), ctr.ID())
 
-	// rootless containers do not use the CNI plugin
+	// rootless containers do not use the CNI plugin directly
 	if !rootless.IsRootless() && !ctr.config.NetMode.IsSlirp4netns() {
 		var requestedIP net.IP
 		if ctr.requestedIP != nil {
@@ -738,6 +753,13 @@ func (r *Runtime) teardownNetNS(ctr *Container) error {
 		}
 	}
 
+	// CNI-in-slirp4netns
+	if rootless.IsRootless() && len(ctr.config.Networks) != 0 {
+		if err := DeallocRootlessCNI(context.Background(), ctr); err != nil {
+			return errors.Wrapf(err, "error tearing down CNI-in-slirp4netns for container %s", ctr.ID())
+		}
+	}
+
 	// First unmount the namespace
 	if err := netns.UnmountNS(ctr.state.NetNS); err != nil {
 		return errors.Wrapf(err, "error unmounting network namespace for container %s", ctr.ID())

libpod/networking_unsupported.go

@@ -8,6 +8,10 @@ func (r *Runtime) setupRootlessNetNS(ctr *Container) error {
 	return define.ErrNotImplemented
 }
 
+func (r *Runtime) setupSlirp4netns(ctr *Container) error {
+	return define.ErrNotImplemented
+}
+
 func (r *Runtime) setupNetNS(ctr *Container) error {
 	return define.ErrNotImplemented
 }

libpod/oci_conmon_linux.go

@@ -1086,7 +1086,7 @@ func (r *ConmonOCIRuntime) createOCIContainer(ctr *Container, restoreOptions *Co
 	cmd.ExtraFiles = append(cmd.ExtraFiles, childSyncPipe, childStartPipe)
 	cmd.ExtraFiles = append(cmd.ExtraFiles, envFiles...)
 
-	if r.reservePorts && !ctr.config.NetMode.IsSlirp4netns() {
+	if r.reservePorts && !rootless.IsRootless() && !ctr.config.NetMode.IsSlirp4netns() {
 		ports, err := bindPorts(ctr.config.PortMappings)
 		if err != nil {
 			return err
@@ -1098,7 +1098,7 @@ func (r *ConmonOCIRuntime) createOCIContainer(ctr *Container, restoreOptions *Co
 		cmd.ExtraFiles = append(cmd.ExtraFiles, ports...)
 	}
 
-	if ctr.config.NetMode.IsSlirp4netns() {
+	if ctr.config.NetMode.IsSlirp4netns() || rootless.IsRootless() {
 		if ctr.config.PostConfigureNetNS {
 			havePortMapping := len(ctr.Config().PortMappings) > 0
 			if havePortMapping {