Add a runtime config

This adds support for a runtime configuration file that can do:
- `mkdir` to make a directory at runtime, eg in `/var` or `/tmp`, to avoid workarounds
- `interface` that can create network interfaces in a container or move them
- `bindNS` that can bind mount namespaces of an `onboot` container to a file so a service can be started in that namespace.

It merges the `service` and `onboot` tools (in `init`) to avoid duplication. This also saves some size for
eg LCOW which did not use the `onboot` code in `runc`.

Signed-off-by: Justin Cormack <justin.cormack@docker.com>

pkg/containerd/Dockerfile

@@ -20,9 +20,6 @@ RUN git checkout $CONTAINERD_COMMIT
 RUN make binaries EXTRA_FLAGS="-buildmode pie" EXTRA_LDFLAGS="-extldflags \\\"-fno-PIC -static\\\""
 RUN cp bin/containerd bin/ctr bin/containerd-shim /usr/bin/
 
-ADD cmd /go/src/cmd
-RUN cd /go/src/cmd/service && ./skanky-vendor.sh $GOPATH/src/github.com/containerd/containerd
-RUN go-compile.sh /go/src/cmd/service
 RUN mkdir -p /etc/init.d && ln -s /usr/bin/service /etc/init.d/020-containerd
 
 WORKDIR /
@@ -31,7 +28,7 @@ COPY . .
 FROM scratch
 ENTRYPOINT []
 WORKDIR /
-COPY --from=alpine /usr/bin/containerd /usr/bin/ctr /usr/bin/containerd-shim /go/bin/service /usr/bin/
+COPY --from=alpine /usr/bin/containerd /usr/bin/ctr /usr/bin/containerd-shim /usr/bin/
 COPY --from=alpine /etc/containerd/config.toml /etc/containerd/
 COPY --from=alpine /usr/share/zoneinfo/UTC /etc/localtime
 COPY --from=alpine /etc/init.d/ /etc/init.d/

pkg/containerd/cmd/service/prepare.go

@@ -1,87 +0,0 @@
-package main
-
-// Please note this file is shared between pkg/runc and pkg/containerd
-// Update it in both places if you make changes
-
-import (
-	"fmt"
-	"os"
-	"path/filepath"
-	"syscall"
-)
-
-func prepare(path string) error {
-	// see if we are dealing with a read only or read write container
-	if _, err := os.Stat(filepath.Join(path, "lower")); err != nil {
-		if os.IsNotExist(err) {
-			return prepareRO(path)
-		}
-		return err
-	}
-	return prepareRW(path)
-}
-
-func prepareRO(path string) error {
-	// make rootfs a mount point, as runc doesn't like it much otherwise
-	rootfs := filepath.Join(path, "rootfs")
-	if err := syscall.Mount(rootfs, rootfs, "", syscall.MS_BIND, ""); err != nil {
-		return err
-	}
-	return nil
-}
-
-func prepareRW(path string) error {
-	// mount a tmpfs on tmp for upper and workdirs
-	// make it private as nothing else should be using this
-	tmp := filepath.Join(path, "tmp")
-	if err := syscall.Mount("tmpfs", tmp, "tmpfs", 0, "size=10%"); err != nil {
-		return err
-	}
-	// make it private as nothing else should be using this
-	if err := syscall.Mount("", tmp, "", syscall.MS_REMOUNT|syscall.MS_PRIVATE, ""); err != nil {
-		return err
-	}
-	upper := filepath.Join(tmp, "upper")
-	// make the mount points
-	if err := os.Mkdir(upper, 0755); err != nil {
-		return err
-	}
-	work := filepath.Join(tmp, "work")
-	if err := os.Mkdir(work, 0755); err != nil {
-		return err
-	}
-	lower := filepath.Join(path, "lower")
-	rootfs := filepath.Join(path, "rootfs")
-	opt := fmt.Sprintf("lowerdir=%s,upperdir=%s,workdir=%s", lower, upper, work)
-	if err := syscall.Mount("overlay", rootfs, "overlay", 0, opt); err != nil {
-		return err
-	}
-	return nil
-}
-
-// cleanup functions are best efforts only, mainly for rw onboot containers
-func cleanup(path string) {
-	// see if we are dealing with a read only or read write container
-	if _, err := os.Stat(filepath.Join(path, "lower")); err != nil {
-		cleanupRO(path)
-	} else {
-		cleanupRW(path)
-	}
-}
-
-func cleanupRO(path string) {
-	// remove the bind mount
-	rootfs := filepath.Join(path, "rootfs")
-	_ = syscall.Unmount(rootfs, 0)
-}
-
-func cleanupRW(path string) {
-	// remove the overlay mount
-	rootfs := filepath.Join(path, "rootfs")
-	_ = os.RemoveAll(rootfs)
-	_ = syscall.Unmount(rootfs, 0)
-	// remove the tmpfs
-	tmp := filepath.Join(path, "tmp")
-	_ = os.RemoveAll(tmp)
-	_ = syscall.Unmount(tmp, 0)
-}

pkg/containerd/cmd/service/skanky-vendor.sh

@@ -1,14 +0,0 @@
-#!/bin/sh
-#
-# We only need the containerd client and its transitive dependencies
-# and we conveniently have a checkout already. We actually prefer to
-# reuse containerd's vendoring for consistency anyway.
-
-set -eu
-ctrd=$1
-cp -r $ctrd/vendor/ vendor/
-# We need containerd itself of course
-mkdir -p vendor/github.com/containerd
-cp -r $ctrd vendor/github.com/containerd/containerd
-# Stop go finding nested vendorings
-rm -rf vendor/github.com/containerd/containerd/vendor

pkg/init/Dockerfile

@@ -1,17 +1,28 @@
-FROM linuxkit/alpine:87a0cd10449d72f374f950004467737dbf440630 AS build
-RUN apk add --no-cache --initdb alpine-baselayout make gcc musl-dev
+FROM linuxkit/alpine:0fd732eb9e99c4db0953ae8de23d95de340ab847 AS build
+RUN apk add --no-cache --initdb alpine-baselayout make gcc musl-dev git linux-headers
 
-ADD usermode-helper.c .
-RUN make usermode-helper
+ADD usermode-helper.c ./
+RUN LDFLAGS=-static CFLAGS=-Werror make usermode-helper
 
 RUN apk add --no-cache go musl-dev
 ENV GOPATH=/go PATH=$PATH:/go/bin
 
-COPY init.go /go/src/init/
-COPY vendor /go/src/init/vendor/
-RUN go-compile.sh /go/src/init/
+COPY cmd /go/src/cmd
+RUN go-compile.sh /go/src/cmd/init
 
-FROM linuxkit/alpine:87a0cd10449d72f374f950004467737dbf440630 AS mirror
+# checkout containerd for vendoring
+ENV GOPATH=/go PATH=$PATH:/go/bin
+# CONTAINERD_REPO and CONTAINERD_COMMIT are defined in linuxkit/alpine
+RUN mkdir -p $GOPATH/src/github.com/containerd && \
+  cd $GOPATH/src/github.com/containerd && \
+  git clone $CONTAINERD_REPO
+WORKDIR $GOPATH/src/github.com/containerd/containerd
+RUN git checkout $CONTAINERD_COMMIT
+
+RUN cd /go/src/cmd/service && ./skanky-vendor.sh $GOPATH/src/github.com/containerd/containerd
+RUN go-compile.sh /go/src/cmd/service
+
+FROM linuxkit/alpine:6ed3b299f5243acb6459b4993549c5045e4ad7f4 AS mirror
 RUN mkdir -p /out/etc/apk && cp -r /etc/apk/* /out/etc/apk/
 RUN apk add --no-cache --initdb -p /out alpine-baselayout busybox musl
 
@@ -23,6 +34,7 @@ ENTRYPOINT []
 CMD []
 WORKDIR /
 COPY --from=build /go/bin/init /
+COPY --from=build /go/bin/service /usr/bin/
 COPY --from=build usermode-helper /sbin/
 COPY --from=mirror /out/ /
 COPY etc etc/

pkg/init/Makefile

@@ -1,4 +1,5 @@
 IMAGE=init
-DEPS=init.go vendor.conf usermode-helper.c $(wildcard etc/*) $(wildcard etc/init.d/*)
+NETWORK=1
+DEPS=usermode-helper.c $(wildcard etc/*) $(wildcard etc/init.d/*) $(shell find cmd -type f)
 
 include ../package.mk

pkg/init/cmd/service/main.go

@@ -5,6 +5,7 @@ import (
 	"fmt"
 	"os"
 	"path/filepath"
+	"strings"
 
 	log "github.com/sirupsen/logrus"
 )
@@ -13,6 +14,9 @@ const (
 	defaultSocket     = "/run/containerd/containerd.sock"
 	defaultPath       = "/containers/services"
 	defaultContainerd = "/usr/bin/containerd"
+	installPath       = "/usr/bin/service"
+	onbootPath        = "/containers/onboot"
+	shutdownPath      = "/containers/onshutdown"
 )
 
 var (
@@ -66,8 +70,17 @@ func main() {
 
 	args := flag.Args()
 	if len(args) < 1 {
-		systemInitCmd(args)
-		os.Exit(0)
+		// check if called form startup scripts
+		command := os.Args[0]
+		switch {
+		case strings.Contains(command, "onboot"):
+			os.Exit(runcInit(onbootPath))
+		case strings.Contains(command, "onshutdown"):
+			os.Exit(runcInit(shutdownPath))
+		case strings.Contains(command, "containerd"):
+			systemInitCmd([]string{})
+			os.Exit(0)
+		}
 	}
 
 	switch args[0] {

pkg/init/cmd/service/prepare.go

@@ -0,0 +1,255 @@
+package main
+
+import (
+	"encoding/json"
+	"fmt"
+	"io/ioutil"
+	"log"
+	"os"
+	"path/filepath"
+	"syscall"
+
+	"github.com/vishvananda/netlink"
+)
+
+const (
+	wgPath      = "/usr/bin/wg"
+	nsenterPath = "/usr/bin/nsenter-net"
+)
+
+// Note these definitions are from moby/tool/src/moby/config.go and should be kept in sync
+
+// Runtime is the type of config processed at runtime, not used to build the OCI spec
+type Runtime struct {
+	Mkdir      []string    `yaml:"mkdir" json:"mkdir,omitempty"`
+	Interfaces []Interface `yaml:"interfaces" json:"interfaces,omitempty"`
+	BindNS     *Namespaces `yaml:"bindNS" json:"bindNS,omitempty"`
+}
+
+// Namespaces is the type for configuring paths to bind namespaces
+type Namespaces struct {
+	Cgroup string `yaml:"cgroup" json:"cgroup,omitempty"`
+	Ipc    string `yaml:"ipc" json:"ipc,omitempty"`
+	Mnt    string `yaml:"mnt" json:"mnt,omitempty"`
+	Net    string `yaml:"net" json:"net,omitempty"`
+	Pid    string `yaml:"pid" json:"pid,omitempty"`
+	User   string `yaml:"user" json:"user,omitempty"`
+	Uts    string `yaml:"uts" json:"uts,omitempty"`
+}
+
+// Interface is the runtime config for network interfaces
+type Interface struct {
+	Name         string `yaml:"name" json:"name,omitempty"`
+	Add          string `yaml:"add" json:"add,omitempty"`
+	Peer         string `yaml:"peer" json:"peer,omitempty"`
+	CreateInRoot bool   `yaml:"createInRoot" json:"createInRoot"`
+}
+
+func getRuntimeConfig(path string) Runtime {
+	var runtime Runtime
+	conf, err := ioutil.ReadFile(filepath.Join(path, "runtime.json"))
+	if err != nil {
+		// if it does not exist it is fine to return an empty runtime, to not do anything
+		if os.IsNotExist(err) {
+			return runtime
+		}
+		log.Fatalf("Cannot read runtime config: %v", err)
+	}
+	if err := json.Unmarshal(conf, &runtime); err != nil {
+		log.Fatalf("Cannot parse runtime config: %v", err)
+	}
+	return runtime
+}
+
+// prepareFilesystem sets up the mounts, before the container is created
+func prepareFilesystem(path string, runtime Runtime) error {
+	// execute the runtime config that should be done up front
+	for _, dir := range runtime.Mkdir {
+		// in future we may need to change the structure to set mode, ownership
+		var mode os.FileMode = 0755
+		err := os.MkdirAll(dir, mode)
+		if err != nil {
+			return fmt.Errorf("Cannot create directory %s: %v", dir, err)
+		}
+	}
+
+	// see if we are dealing with a read only or read write container
+	if _, err := os.Stat(filepath.Join(path, "lower")); err != nil {
+		if os.IsNotExist(err) {
+			return prepareRO(path)
+		}
+		return err
+	}
+	return prepareRW(path)
+}
+
+func prepareRO(path string) error {
+	// make rootfs a mount point, as runc doesn't like it much otherwise
+	rootfs := filepath.Join(path, "rootfs")
+	if err := syscall.Mount(rootfs, rootfs, "", syscall.MS_BIND, ""); err != nil {
+		return err
+	}
+	return nil
+}
+
+func prepareRW(path string) error {
+	// mount a tmpfs on tmp for upper and workdirs
+	// make it private as nothing else should be using this
+	tmp := filepath.Join(path, "tmp")
+	if err := syscall.Mount("tmpfs", tmp, "tmpfs", 0, "size=10%"); err != nil {
+		return err
+	}
+	// make it private as nothing else should be using this
+	if err := syscall.Mount("", tmp, "", syscall.MS_REMOUNT|syscall.MS_PRIVATE, ""); err != nil {
+		return err
+	}
+	upper := filepath.Join(tmp, "upper")
+	// make the mount points
+	if err := os.Mkdir(upper, 0755); err != nil {
+		return err
+	}
+	work := filepath.Join(tmp, "work")
+	if err := os.Mkdir(work, 0755); err != nil {
+		return err
+	}
+	lower := filepath.Join(path, "lower")
+	rootfs := filepath.Join(path, "rootfs")
+	opt := fmt.Sprintf("lowerdir=%s,upperdir=%s,workdir=%s", lower, upper, work)
+	if err := syscall.Mount("overlay", rootfs, "overlay", 0, opt); err != nil {
+		return err
+	}
+	return nil
+}
+
+// bind mount a namespace file
+func bindNS(ns string, path string, pid int) error {
+	if path == "" {
+		return nil
+	}
+	// the path and file need to exist for the bind to succeed, so try to create
+	dir := filepath.Dir(path)
+	if err := os.MkdirAll(dir, 0755); err != nil {
+		return fmt.Errorf("Cannot create leading directories %s for bind mount destination: %v", dir, err)
+	}
+	fi, err := os.Create(path)
+	if err != nil {
+		return fmt.Errorf("Cannot create a mount point for namespace bind at %s: %v", path, err)
+	}
+	if err := fi.Close(); err != nil {
+		return err
+	}
+	if err := syscall.Mount(fmt.Sprintf("/proc/%d/ns/%s", pid, ns), path, "", syscall.MS_BIND, ""); err != nil {
+		return fmt.Errorf("Failed to bind %s namespace at %s: %v", ns, path, err)
+	}
+	return nil
+}
+
+// prepareProcess sets up anything that needs to be done after the container process is created, but before it runs
+// for example networking
+func prepareProcess(pid int, runtime Runtime) error {
+	for _, iface := range runtime.Interfaces {
+		if iface.Name == "" {
+			return fmt.Errorf("Interface requires a name")
+		}
+
+		var link netlink.Link
+		var ns interface{} = netlink.NsPid(pid)
+		var move bool
+		var err error
+
+		if iface.Peer != "" && iface.Add == "" {
+			// must be a veth if specify peer
+			iface.Add = "veth"
+		}
+
+		// if create in root is set, create in root namespace first, then move
+		// also do the same for a veth pair
+		if iface.CreateInRoot || iface.Add == "veth" {
+			ns = nil
+			move = true
+		}
+
+		if iface.Add != "" {
+			switch iface.Add {
+			case "veth":
+				if iface.Peer == "" {
+					return fmt.Errorf("Creating a veth pair %s requires a peer to be set", iface.Name)
+				}
+				la := netlink.LinkAttrs{Name: iface.Name, Namespace: ns}
+				link = &netlink.Veth{LinkAttrs: la, PeerName: iface.Peer}
+			default:
+				// no special creation options needed
+				la := netlink.LinkAttrs{Name: iface.Name, Namespace: ns}
+				link = &netlink.GenericLink{la, iface.Add}
+			}
+			if err := netlink.LinkAdd(link); err != nil {
+				return fmt.Errorf("Link add %s of type %s failed: %v", iface.Name, iface.Add, err)
+			}
+			fmt.Fprintf(os.Stderr, "Created interface %s type %s\n", iface.Name, iface.Add)
+		} else {
+			// find existing interface
+			link, err = netlink.LinkByName(iface.Name)
+			if err != nil {
+				return fmt.Errorf("Cannot find interface %s: %v", iface.Name, err)
+			}
+			// then move into namespace
+			move = true
+		}
+		if move {
+			if err := netlink.LinkSetNsPid(link, int(pid)); err != nil {
+				return fmt.Errorf("Cannot move interface %s into namespace: %v", iface.Name, err)
+			}
+			fmt.Fprintf(os.Stderr, "Moved interface %s to pid %d\n", iface.Name, pid)
+		}
+	}
+
+	if runtime.BindNS != nil {
+		binds := []struct {
+			ns   string
+			path string
+		}{
+			{"cgroup", runtime.BindNS.Cgroup},
+			{"ipc", runtime.BindNS.Ipc},
+			{"mnt", runtime.BindNS.Mnt},
+			{"net", runtime.BindNS.Net},
+			{"pid", runtime.BindNS.Pid},
+			{"user", runtime.BindNS.User},
+			{"uts", runtime.BindNS.Uts},
+		}
+
+		for _, b := range binds {
+			if err := bindNS(b.ns, b.path, pid); err != nil {
+				return err
+			}
+		}
+	}
+
+	return nil
+}
+
+// cleanup functions are best efforts only, mainly for rw onboot containers
+func cleanup(path string) {
+	// see if we are dealing with a read only or read write container
+	if _, err := os.Stat(filepath.Join(path, "lower")); err != nil {
+		cleanupRO(path)
+	} else {
+		cleanupRW(path)
+	}
+}
+
+func cleanupRO(path string) {
+	// remove the bind mount
+	rootfs := filepath.Join(path, "rootfs")
+	_ = syscall.Unmount(rootfs, 0)
+}
+
+func cleanupRW(path string) {
+	// remove the overlay mount
+	rootfs := filepath.Join(path, "rootfs")
+	_ = os.RemoveAll(rootfs)
+	_ = syscall.Unmount(rootfs, 0)
+	// remove the tmpfs
+	tmp := filepath.Join(path, "tmp")
+	_ = os.RemoveAll(tmp)
+	_ = syscall.Unmount(tmp, 0)
+}

pkg/init/cmd/service/runc.go

@@ -0,0 +1,111 @@
+package main
+
+import (
+	"io/ioutil"
+	"log"
+	"os"
+	"os/exec"
+	"path/filepath"
+	"strconv"
+
+	"github.com/containerd/containerd/sys"
+)
+
+const (
+	runcBinary = "/usr/bin/runc"
+)
+
+func runcInit(rootPath string) int {
+	// do nothing if the path does not exist
+	if _, err := os.Stat(rootPath); err != nil && os.IsNotExist(err) {
+		return 0
+	}
+
+	// get files; note ReadDir already sorts them
+	files, err := ioutil.ReadDir(rootPath)
+	if err != nil {
+		log.Fatalf("Cannot read files in %s: %v", rootPath, err)
+	}
+
+	tmpdir, err := ioutil.TempDir("", filepath.Base(rootPath))
+	if err != nil {
+		log.Fatalf("Cannot create temporary directory: %v", err)
+	}
+
+	// need to set ourselves as a child subreaper or we cannot wait for runc as reparents to init
+	if err := sys.SetSubreaper(1); err != nil {
+		log.Fatalf("Cannot set as subreaper: %v", err)
+	}
+
+	status := 0
+
+	for _, file := range files {
+		name := file.Name()
+		path := filepath.Join(rootPath, name)
+
+		runtimeConfig := getRuntimeConfig(path)
+
+		if err := prepareFilesystem(path, runtimeConfig); err != nil {
+			log.Printf("Error preparing %s: %v", name, err)
+			status = 1
+			continue
+		}
+		pidfile := filepath.Join(tmpdir, name)
+		cmd := exec.Command(runcBinary, "create", "--bundle", path, "--pid-file", pidfile, name)
+		cmd.Stdout = os.Stdout
+		cmd.Stderr = os.Stderr
+		if err := cmd.Run(); err != nil {
+			log.Printf("Error creating %s: %v", name, err)
+			status = 1
+			// skip cleanup on error for debug
+			continue
+		}
+		pf, err := ioutil.ReadFile(pidfile)
+		if err != nil {
+			log.Printf("Cannot read pidfile: %v", err)
+			status = 1
+			continue
+		}
+		pid, err := strconv.Atoi(string(pf))
+		if err != nil {
+			log.Printf("Cannot parse pid from pidfile: %v", err)
+			status = 1
+			continue
+		}
+
+		if err := prepareProcess(pid, runtimeConfig); err != nil {
+			log.Printf("Cannot prepare process: %v", err)
+			status = 1
+			continue
+		}
+
+		waitFor := make(chan *os.ProcessState)
+		go func() {
+			// never errors in Unix
+			p, _ := os.FindProcess(pid)
+			state, err := p.Wait()
+			if err != nil {
+				log.Printf("Process wait error: %v", err)
+			}
+			waitFor <- state
+		}()
+
+		cmd = exec.Command(runcBinary, "start", name)
+		cmd.Stdout = os.Stdout
+		cmd.Stderr = os.Stderr
+		if err := cmd.Run(); err != nil {
+			log.Printf("Error starting %s: %v", name, err)
+			status = 1
+			continue
+		}
+
+		_ = <-waitFor
+
+		cleanup(path)
+		_ = os.Remove(pidfile)
+	}
+
+	_ = os.RemoveAll(tmpdir)
+
+	return status
+}

pkg/init/cmd/service/start.go

@@ -55,11 +55,15 @@ func startCmd(args []string) {
 	log.Debugf("Started %s pid %d", id, pid)
 }
 
-func start(service, sock, path, dumpSpec string) (string, uint32, string, error) {
-	rootfs := filepath.Join(path, service, "rootfs")
+func start(service, sock, basePath, dumpSpec string) (string, uint32, string, error) {
+	path := filepath.Join(basePath, service)
 
-	if err := prepare(filepath.Join(path, service)); err != nil {
-		return "", 0, "preparing rootfs", err
+	runtimeConfig := getRuntimeConfig(path)
+
+	rootfs := filepath.Join(path, "rootfs")
+
+	if err := prepareFilesystem(path, runtimeConfig); err != nil {
+		return "", 0, "preparing filesystem", err
 	}
 
 	client, err := containerd.New(sock)
@@ -70,7 +74,7 @@ func start(service, sock, path, dumpSpec string) (string, uint32, string, error)
 	ctx := namespaces.WithNamespace(context.Background(), "default")
 
 	var spec *specs.Spec
-	specf, err := os.Open(filepath.Join(path, service, "config.json"))
+	specf, err := os.Open(filepath.Join(path, "config.json"))
 	if err != nil {
 		return "", 0, "failed to read service spec", err
 	}
@@ -119,6 +123,10 @@ func start(service, sock, path, dumpSpec string) (string, uint32, string, error)
 		return "", 0, "failed to create task", err
 	}
 
+	if err := prepareProcess(int(task.Pid()), runtimeConfig); err != nil {
+		return "", 0, "preparing process", err
+	}
+
 	if err := task.Start(ctx); err != nil {
 		// Don't destroy the container here so it can be inspected for debugging.
 		return "", 0, "failed to start task", err

pkg/runc/Dockerfile

@@ -20,14 +20,12 @@ RUN git checkout $RUNC_COMMIT
 RUN make static BUILDTAGS="seccomp" EXTRA_FLAGS="-buildmode pie" EXTRA_LDFLAGS="-extldflags \\\"-fno-PIC -static\\\""
 RUN cp runc /usr/bin/
 
-ADD cmd /go/src/cmd
-RUN go-compile.sh /go/src/cmd/onboot
-RUN mkdir -p /etc/init.d && ln -s /usr/bin/onboot /etc/init.d/010-onboot
-RUN mkdir -p /etc/shutdown.d && ln -s /usr/bin/onboot /etc/shutdown.d/010-onshutdown
+RUN mkdir -p /etc/init.d && ln -s /usr/bin/service /etc/init.d/010-onboot
+RUN mkdir -p /etc/shutdown.d && ln -s /usr/bin/service /etc/shutdown.d/010-onshutdown
 
 FROM scratch
 WORKDIR /
 ENTRYPOINT []
-COPY --from=alpine /usr/bin/runc /go/bin/onboot /usr/bin/
+COPY --from=alpine /usr/bin/runc /usr/bin/
 COPY --from=alpine /etc/init.d/ /etc/init.d/
 COPY --from=alpine /etc/shutdown.d/ /etc/shutdown.d/

pkg/runc/Makefile

@@ -1,5 +1,4 @@
 IMAGE=runc
 NETWORK=1
-DEPS=$(wildcard cmd/onboot/*.go)
 
 include ../package.mk

pkg/runc/cmd/onboot/main.go

@@ -1,66 +0,0 @@
-package main
-
-import (
-	"io/ioutil"
-	"log"
-	"os"
-	"os/exec"
-	"path/filepath"
-	"strings"
-)
-
-const (
-	runcBinary   = "/usr/bin/runc"
-	onbootPath   = "/containers/onboot"
-	shutdownPath = "/containers/onshutdown"
-)
-
-func main() {
-	// try to work out how we are being called
-	command := os.Args[0]
-	if len(os.Args) > 1 {
-		command = os.Args[1]
-	}
-	var path = onbootPath
-	switch {
-	case strings.Contains(command, "boot"):
-		path = onbootPath
-	case strings.Contains(command, "shutdown"):
-		path = shutdownPath
-	}
-
-	// do nothing if the path does not exist
-	if _, err := os.Stat(path); err != nil && os.IsNotExist(err) {
-		os.Exit(0)
-	}
-
-	// get files; note ReadDir already sorts them
-	files, err := ioutil.ReadDir(path)
-	if err != nil {
-		log.Fatalf("Cannot read files in %s: %v", path, err)
-	}
-
-	status := 0
-
-	for _, file := range files {
-		name := file.Name()
-		fullPath := filepath.Join(path, name)
-		if err := prepare(fullPath); err != nil {
-			log.Printf("Error preparing %s: %v", name, err)
-			status = 1
-			continue
-		}
-		cmd := exec.Command(runcBinary, "run", "--bundle", fullPath, name)
-		cmd.Stdout = os.Stdout
-		cmd.Stderr = os.Stderr
-		if err := cmd.Run(); err != nil {
-			log.Printf("Error running %s: %v", name, err)
-			status = 1
-		} else {
-			// do not clean up on error to make debug easier
-			cleanup(fullPath)
-		}
-	}
-
-	os.Exit(status)
-}

pkg/runc/cmd/onboot/prepare.go

@@ -1,87 +0,0 @@
-package main
-
-// Please note this file is shared between pkg/runc and pkg/containerd
-// Update it in both places if you make changes
-
-import (
-	"fmt"
-	"os"
-	"path/filepath"
-	"syscall"
-)
-
-func prepare(path string) error {
-	// see if we are dealing with a read only or read write container
-	if _, err := os.Stat(filepath.Join(path, "lower")); err != nil {
-		if os.IsNotExist(err) {
-			return prepareRO(path)
-		}
-		return err
-	}
-	return prepareRW(path)
-}
-
-func prepareRO(path string) error {
-	// make rootfs a mount point, as runc doesn't like it much otherwise
-	rootfs := filepath.Join(path, "rootfs")
-	if err := syscall.Mount(rootfs, rootfs, "", syscall.MS_BIND, ""); err != nil {
-		return err
-	}
-	return nil
-}
-
-func prepareRW(path string) error {
-	// mount a tmpfs on tmp for upper and workdirs
-	// make it private as nothing else should be using this
-	tmp := filepath.Join(path, "tmp")
-	if err := syscall.Mount("tmpfs", tmp, "tmpfs", 0, "size=10%"); err != nil {
-		return err
-	}
-	// make it private as nothing else should be using this
-	if err := syscall.Mount("", tmp, "", syscall.MS_REMOUNT|syscall.MS_PRIVATE, ""); err != nil {
-		return err
-	}
-	upper := filepath.Join(tmp, "upper")
-	// make the mount points
-	if err := os.Mkdir(upper, 0755); err != nil {
-		return err
-	}
-	work := filepath.Join(tmp, "work")
-	if err := os.Mkdir(work, 0755); err != nil {
-		return err
-	}
-	lower := filepath.Join(path, "lower")
-	rootfs := filepath.Join(path, "rootfs")
-	opt := fmt.Sprintf("lowerdir=%s,upperdir=%s,workdir=%s", lower, upper, work)
-	if err := syscall.Mount("overlay", rootfs, "overlay", 0, opt); err != nil {
-		return err
-	}
-	return nil
-}
-
-// cleanup functions are best efforts only, mainly for rw onboot containers
-func cleanup(path string) {
-	// see if we are dealing with a read only or read write container
-	if _, err := os.Stat(filepath.Join(path, "lower")); err != nil {
-		cleanupRO(path)
-	} else {
-		cleanupRW(path)
-	}
-}
-
-func cleanupRO(path string) {
-	// remove the bind mount
-	rootfs := filepath.Join(path, "rootfs")
-	_ = syscall.Unmount(rootfs, 0)
-}
-
-func cleanupRW(path string) {
-	// remove the overlay mount
-	rootfs := filepath.Join(path, "rootfs")
-	_ = os.RemoveAll(rootfs)
-	_ = syscall.Unmount(rootfs, 0)
-	// remove the tmpfs
-	tmp := filepath.Join(path, "tmp")
-	_ = os.RemoveAll(tmp)
-	_ = syscall.Unmount(tmp, 0)
-}