Merge pull request #20 from alibaba/update_2019_0308

Update 2019 0308

.travis.yml

@@ -1,3 +1,4 @@
+dist: xenial
 language: go
 go:
   - 1.9.x
@@ -21,9 +22,8 @@ env:
     - BUILDTAGS="seccomp apparmor selinux ambient"
 
 before_install:
-  - echo "deb http://archive.ubuntu.com/ubuntu trusty-backports main restricted universe multiverse" | sudo tee -a /etc/apt/sources.list
   - sudo apt-get -qq update
-  - sudo apt-get install -y libseccomp-dev/trusty-backports
+  - sudo apt-get install -y libseccomp-dev
   - go get -u golang.org/x/lint/golint
   - go get -u github.com/vbatts/git-validation
   - env | grep TRAVIS_

README.md

@@ -16,10 +16,9 @@ This means that `runc` 1.0.0 should implement the 1.0 version of the specificati
 
 You can find official releases of `runc` on the [release](https://github.com/opencontainers/runc/releases) page.
 
-### Security
+## Security
 
-If you wish to report a security issue, please disclose the issue responsibly
-to security@opencontainers.org.
+Reporting process and disclosure communications are outlined in [/org/security](https://github.com/opencontainers/org/blob/master/security/)
 
 ## Building
 

exec.go

@@ -89,6 +89,10 @@ following will output a list of processes running in the container:
 			Usage:  "disable the use of the subreaper used to reap reparented processes",
 			Hidden: true,
 		},
+		cli.IntFlag{
+			Name:  "preserve-fds",
+			Usage: "Pass N additional file descriptors to the container (stdio + $LISTEN_FDS + N in total)",
+		},
 	},
 	Action: func(context *cli.Context) error {
 		if err := checkArgs(context, 1, minArgs); err != nil {
@@ -141,6 +145,7 @@ func execProcess(context *cli.Context) (int, error) {
 		pidFile:         context.String("pid-file"),
 		action:          CT_ACT_RUN,
 		init:            false,
+		preserveFDs:     context.Int("preserve-fds"),
 	}
 	return r.run(p)
 }

libcontainer/cgroups/systemd/apply_systemd.go

@@ -72,13 +72,11 @@ const (
 )
 
 var (
-	connLock                        sync.Mutex
-	theConn                         *systemdDbus.Conn
-	hasStartTransientUnit           bool
-	hasStartTransientSliceUnit      bool
-	hasTransientDefaultDependencies bool
-	hasDelegateScope                bool
-	hasDelegateSlice                bool
+	connLock                   sync.Mutex
+	theConn                    *systemdDbus.Conn
+	hasStartTransientUnit      bool
+	hasStartTransientSliceUnit bool
+	hasDelegateSlice           bool
 )
 
 func newProp(name string, units interface{}) systemdDbus.Property {
@@ -116,53 +114,6 @@ func UseSystemd() bool {
 			}
 		}
 
-		// Ensure the scope name we use doesn't exist. Use the Pid to
-		// avoid collisions between multiple libcontainer users on a
-		// single host.
-		scope := fmt.Sprintf("libcontainer-%d-systemd-test-default-dependencies.scope", os.Getpid())
-		testScopeExists := true
-		for i := 0; i <= testScopeWait; i++ {
-			if _, err := theConn.StopUnit(scope, "replace", nil); err != nil {
-				if dbusError, ok := err.(dbus.Error); ok {
-					if strings.Contains(dbusError.Name, "org.freedesktop.systemd1.NoSuchUnit") {
-						testScopeExists = false
-						break
-					}
-				}
-			}
-			time.Sleep(time.Millisecond)
-		}
-
-		// Bail out if we can't kill this scope without testing for DefaultDependencies
-		if testScopeExists {
-			return hasStartTransientUnit
-		}
-
-		// Assume StartTransientUnit on a scope allows DefaultDependencies
-		hasTransientDefaultDependencies = true
-		ddf := newProp("DefaultDependencies", false)
-		if _, err := theConn.StartTransientUnit(scope, "replace", []systemdDbus.Property{ddf}, nil); err != nil {
-			if dbusError, ok := err.(dbus.Error); ok {
-				if strings.Contains(dbusError.Name, "org.freedesktop.DBus.Error.PropertyReadOnly") {
-					hasTransientDefaultDependencies = false
-				}
-			}
-		}
-
-		// Not critical because of the stop unit logic above.
-		theConn.StopUnit(scope, "replace", nil)
-
-		// Assume StartTransientUnit on a scope allows Delegate
-		hasDelegateScope = true
-		dlScope := newProp("Delegate", true)
-		if _, err := theConn.StartTransientUnit(scope, "replace", []systemdDbus.Property{dlScope}, nil); err != nil {
-			if dbusError, ok := err.(dbus.Error); ok {
-				if strings.Contains(dbusError.Name, "org.freedesktop.DBus.Error.PropertyReadOnly") {
-					hasDelegateScope = false
-				}
-			}
-		}
-
 		// Assume we have the ability to start a transient unit as a slice
 		// This was broken until systemd v229, but has been back-ported on RHEL environments >= 219
 		// For details, see: https://bugzilla.redhat.com/show_bug.cgi?id=1370299
@@ -207,7 +158,6 @@ func UseSystemd() bool {
 		}
 
 		// Not critical because of the stop unit logic above.
-		theConn.StopUnit(scope, "replace", nil)
 		theConn.StopUnit(slice, "replace", nil)
 	}
 	return hasStartTransientUnit
@@ -268,9 +218,8 @@ func (m *Manager) Apply(pid int) error {
 			properties = append(properties, newProp("Delegate", true))
 		}
 	} else {
-		if hasDelegateScope {
-			properties = append(properties, newProp("Delegate", true))
-		}
+		// Assume scopes always support delegation.
+		properties = append(properties, newProp("Delegate", true))
 	}
 
 	// Always enable accounting, this gets us the same behaviour as the fs implementation,
@@ -280,10 +229,9 @@ func (m *Manager) Apply(pid int) error {
 		newProp("CPUAccounting", true),
 		newProp("BlockIOAccounting", true))
 
-	if hasTransientDefaultDependencies {
-		properties = append(properties,
-			newProp("DefaultDependencies", false))
-	}
+	// Assume DefaultDependencies= will always work (the check for it was previously broken.)
+	properties = append(properties,
+		newProp("DefaultDependencies", false))
 
 	if c.Resources.Memory != 0 {
 		properties = append(properties,

libcontainer/container_linux.go

@@ -19,14 +19,15 @@ import (
 	"syscall" // only for SysProcAttr and Signal
 	"time"
 
+	"github.com/cyphar/filepath-securejoin"
 	"github.com/opencontainers/runc/libcontainer/cgroups"
 	"github.com/opencontainers/runc/libcontainer/configs"
-	"github.com/opencontainers/runc/libcontainer/criurpc"
 	"github.com/opencontainers/runc/libcontainer/intelrdt"
 	"github.com/opencontainers/runc/libcontainer/system"
 	"github.com/opencontainers/runc/libcontainer/utils"
 	"github.com/opencontainers/runtime-spec/specs-go"
 
+	criurpc "github.com/checkpoint-restore/go-criu/rpc"
 	"github.com/golang/protobuf/proto"
 	"github.com/sirupsen/logrus"
 	"github.com/vishvananda/netlink/nl"
@@ -481,6 +482,7 @@ func (c *linuxContainer) commandTemplate(p *Process, childPipe *os.File) (*exec.
 	cmd.ExtraFiles = append(cmd.ExtraFiles, childPipe)
 	cmd.Env = append(cmd.Env,
 		fmt.Sprintf("_LIBCONTAINER_INITPIPE=%d", stdioFdCount+len(cmd.ExtraFiles)-1),
+		fmt.Sprintf("_LIBCONTAINER_STATEDIR=%s", c.root),
 	)
 	// NOTE: when running a container with no PID namespace and the parent process spawning the container is
 	// PID1 the pdeathsig is being delivered to the container's init process by the kernel for some reason
@@ -1139,6 +1141,75 @@ func (c *linuxContainer) restoreNetwork(req *criurpc.CriuReq, criuOpts *CriuOpts
 	}
 }
 
+// makeCriuRestoreMountpoints makes the actual mountpoints for the
+// restore using CRIU. This function is inspired from the code in
+// rootfs_linux.go
+func (c *linuxContainer) makeCriuRestoreMountpoints(m *configs.Mount) error {
+	switch m.Device {
+	case "cgroup":
+		// Do nothing for cgroup, CRIU should handle it
+	case "bind":
+		// The prepareBindMount() function checks if source
+		// exists. So it cannot be used for other filesystem types.
+		if err := prepareBindMount(m, c.config.Rootfs); err != nil {
+			return err
+		}
+	default:
+		// for all other file-systems just create the mountpoints
+		dest, err := securejoin.SecureJoin(c.config.Rootfs, m.Destination)
+		if err != nil {
+			return err
+		}
+		if err := checkMountDestination(c.config.Rootfs, dest); err != nil {
+			return err
+		}
+		m.Destination = dest
+		if err := os.MkdirAll(dest, 0755); err != nil {
+			return err
+		}
+	}
+	return nil
+}
+
+// isPathInPrefixList is a small function for CRIU restore to make sure
+// mountpoints, which are on a tmpfs, are not created in the roofs
+func isPathInPrefixList(path string, prefix []string) bool {
+	for _, p := range prefix {
+		if strings.HasPrefix(path, p+"/") {
+			return false
+		}
+	}
+	return true
+}
+
+// prepareCriuRestoreMounts tries to set up the rootfs of the
+// container to be restored in the same way runc does it for
+// initial container creation. Even for a read-only rootfs container
+// runc modifies the rootfs to add mountpoints which do not exist.
+// This function also creates missing mountpoints as long as they
+// are not on top of a tmpfs, as CRIU will restore tmpfs content anyway.
+func (c *linuxContainer) prepareCriuRestoreMounts(mounts []*configs.Mount) error {
+	// First get a list of a all tmpfs mounts
+	tmpfs := []string{}
+	for _, m := range mounts {
+		switch m.Device {
+		case "tmpfs":
+			tmpfs = append(tmpfs, m.Destination)
+		}
+	}
+	// Now go through all mounts and create the mountpoints
+	// if the mountpoints are not on a tmpfs, as CRIU will
+	// restore the complete tmpfs content from its checkpoint.
+	for _, m := range mounts {
+		if isPathInPrefixList(m.Destination, tmpfs) {
+			if err := c.makeCriuRestoreMountpoints(m); err != nil {
+				return err
+			}
+		}
+	}
+	return nil
+}
+
 func (c *linuxContainer) Restore(process *Process, criuOpts *CriuOpts) error {
 	c.m.Lock()
 	defer c.m.Unlock()
@@ -1252,6 +1323,12 @@ func (c *linuxContainer) Restore(process *Process, criuOpts *CriuOpts) error {
 		}
 	}
 
+	// This will modify the rootfs of the container in the same way runc
+	// modifies the container during initial creation.
+	if err := c.prepareCriuRestoreMounts(c.config.Mounts); err != nil {
+		return err
+	}
+
 	for _, m := range c.config.Mounts {
 		switch m.Device {
 		case "bind":