Merge pull request #19 from alibaba/update_2019_0213

Update runc 2019 0213

Dockerfile

@@ -45,10 +45,14 @@ RUN cd /tmp \
     && rm -rf /tmp/bats
 
 # install criu
+# For CRIU 3.11 one patch is needed for the test 'checkpoint and restore with container specific CRIU config'
+# This should be no longer necessary with CRIU 3.12
+# See https://github.com/opencontainers/runc/pull/1933
 ENV CRIU_VERSION v3.11
 RUN mkdir -p /usr/src/criu \
     && curl -sSL https://github.com/checkpoint-restore/criu/archive/${CRIU_VERSION}.tar.gz | tar -v -C /usr/src/criu/ -xz --strip-components=1 \
     && cd /usr/src/criu \
+    && curl https://github.com/checkpoint-restore/criu/commit/bb0b2f2635d71e549851b7c626a1464e42a3b5c7.patch | patch -p1 \
     && make install-criu \
     && rm -rf /usr/src/criu
 

docs/checkpoint-restore.md

@@ -0,0 +1,50 @@
+# Checkpoint and Restore #
+
+For a basic description about checkpointing and restoring containers with
+`runc` please see [runc-checkpoint(8)](../man/runc-checkpoint.8.md) and
+[runc-restore(8)](../man/runc-restore.8.md).
+
+## Checkpoint/Restore Annotations ##
+
+In addition to specifying options on the command-line like it is described
+in the man-pages (see above), it is also possible to influence CRIU's
+behaviour using CRIU configuration files. For details about CRIU's
+configuration file support please see [CRIU's wiki](https://criu.org/Configuration_files).
+
+In addition to CRIU's default configuration files `runc` tells CRIU to
+also evaluate the file `/etc/criu/runc.conf`. Using the annotation
+`org.criu.config` it is, however, possible to change this additional
+CRIU configuration file.
+
+If the annotation `org.criu.config` is set to an empty string `runc`
+will not pass any additional configuration file to CRIU. With an empty
+string it is therefore possible to disable the additional CRIU configuration
+file. This can be used to make sure that no additional configuration file
+changes CRIU's behaviour accidentally.
+
+If the annotation `org.criu.config` is set to a non-empty string `runc` will
+pass that string to CRIU to be evaluated as an additional configuration file.
+If CRIU cannot open this additional configuration file, it will ignore this
+file and continue.
+
+### Annotation Example to disable additional CRIU configuration file ###
+
+```
+{
+	"ociVersion": "1.0.0",
+	"annotations": {
+		"org.criu.config": ""
+	},
+	"process": {
+```
+
+### Annotation Example to set a specific CRIU configuration file ###
+
+```
+{
+	"ociVersion": "1.0.0",
+	"annotations": {
+		"org.criu.config": "/etc/special-runc-criu-options"
+	},
+	"process": {
+```

libcontainer/cgroups/systemd/apply_systemd.go

@@ -5,6 +5,7 @@ package systemd
 import (
 	"errors"
 	"fmt"
+	"io/ioutil"
 	"math"
 	"os"
 	"path/filepath"
@@ -590,6 +591,15 @@ func setKernelMemory(c *configs.Cgroup) error {
 	if err := os.MkdirAll(path, 0755); err != nil {
 		return err
 	}
+	// do not try to enable the kernel memory if we already have
+	// tasks in the cgroup.
+	content, err := ioutil.ReadFile(filepath.Join(path, "tasks"))
+	if err != nil {
+		return err
+	}
+	if len(content) > 0 {
+		return nil
+	}
 	return fs.EnableKernelMemoryAccounting(path)
 }
 

libcontainer/cgroups/utils.go

@@ -14,6 +14,7 @@ import (
 	"time"
 
 	units "github.com/docker/go-units"
+	"golang.org/x/sys/unix"
 )
 
 const (
@@ -463,11 +464,40 @@ func WriteCgroupProc(dir string, pid int) error {
 		return fmt.Errorf("no such directory for %s", CgroupProcesses)
 	}
 
-	// Don't attach any pid to the cgroup if -1 is specified as a pid
-	if pid != -1 {
-		if err := ioutil.WriteFile(filepath.Join(dir, CgroupProcesses), []byte(strconv.Itoa(pid)), 0700); err != nil {
-			return fmt.Errorf("failed to write %v to %v: %v", pid, CgroupProcesses, err)
+	// Dont attach any pid to the cgroup if -1 is specified as a pid
+	if pid == -1 {
+		return nil
+	}
+
+	cgroupProcessesFile, err := os.OpenFile(filepath.Join(dir, CgroupProcesses), os.O_WRONLY|os.O_CREATE|os.O_TRUNC, 0700)
+	if err != nil {
+		return fmt.Errorf("failed to write %v to %v: %v", pid, CgroupProcesses, err)
+	}
+	defer cgroupProcessesFile.Close()
+
+	for i := 0; i < 5; i++ {
+		_, err = cgroupProcessesFile.WriteString(strconv.Itoa(pid))
+		if err == nil {
+			return nil
 		}
+
+		// EINVAL might mean that the task being added to cgroup.procs is in state
+		// TASK_NEW. We should attempt to do so again.
+		if isEINVAL(err) {
+			time.Sleep(30 * time.Millisecond)
+			continue
+		}
+
+		return fmt.Errorf("failed to write %v to %v: %v", pid, CgroupProcesses, err)
+	}
+	return err
+}
+
+func isEINVAL(err error) bool {
+	switch err := err.(type) {
+	case *os.PathError:
+		return err.Err == unix.EINVAL
+	default:
+		return false
 	}
-	return nil
 }

libcontainer/container_linux.go

@@ -875,6 +875,32 @@ func waitForCriuLazyServer(r *os.File, status string) error {
 	return nil
 }
 
+func (c *linuxContainer) handleCriuConfigurationFile(rpcOpts *criurpc.CriuOpts) {
+	// CRIU will evaluate a configuration starting with release 3.11.
+	// Settings in the configuration file will overwrite RPC settings.
+	// Look for annotations. The annotation 'org.criu.config'
+	// specifies if CRIU should use a different, container specific
+	// configuration file.
+	_, annotations := utils.Annotations(c.config.Labels)
+	configFile, exists := annotations["org.criu.config"]
+	if exists {
+		// If the annotation 'org.criu.config' exists and is set
+		// to a non-empty string, tell CRIU to use that as a
+		// configuration file. If the file does not exist, CRIU
+		// will just ignore it.
+		if configFile != "" {
+			rpcOpts.ConfigFile = proto.String(configFile)
+		}
+		// If 'org.criu.config' exists and is set to an empty
+		// string, a runc specific CRIU configuration file will
+		// be not set at all.
+	} else {
+		// If the mentioned annotation has not been found, specify
+		// a default CRIU configuration file.
+		rpcOpts.ConfigFile = proto.String("/etc/criu/runc.conf")
+	}
+}
+
 func (c *linuxContainer) Checkpoint(criuOpts *CriuOpts) error {
 	c.m.Lock()
 	defer c.m.Unlock()
@@ -940,6 +966,8 @@ func (c *linuxContainer) Checkpoint(criuOpts *CriuOpts) error {
 		LazyPages:       proto.Bool(criuOpts.LazyPages),
 	}
 
+	c.handleCriuConfigurationFile(&rpcOpts)
+
 	// If the container is running in a network namespace and has
 	// a path to the network namespace configured, we will dump
 	// that network namespace as an external namespace and we
@@ -1190,6 +1218,8 @@ func (c *linuxContainer) Restore(process *Process, criuOpts *CriuOpts) error {
 		},
 	}
 
+	c.handleCriuConfigurationFile(req.Opts)
+
 	// Same as during checkpointing. If the container has a specific network namespace
 	// assigned to it, this now expects that the checkpoint will be restored in a
 	// already created network namespace.