Fix for sysbox issue #544

libcontainer/cgroups/fscommon/open.go → libcontainer/cgroups/file.go

@@ -1,6 +1,9 @@
-package fscommon
+package cgroups
 
 import (
+	"bytes"
+	"fmt"
+	"io/ioutil"
 	"os"
 	"strings"
 	"sync"
@@ -11,6 +14,88 @@ import (
 	"golang.org/x/sys/unix"
 )
 
+// OpenFile opens a cgroup file in a given dir with given flags.
+// It is supposed to be used for cgroup files only.
+func OpenFile(dir, file string, flags int) (*os.File, error) {
+	if dir == "" {
+		return nil, errors.Errorf("no directory specified for %s", file)
+	}
+	return openFile(dir, file, flags)
+}
+
+// ReadFile reads data from a cgroup file in dir.
+// It is supposed to be used for cgroup files only.
+func ReadFile(dir, file string) (string, error) {
+	fd, err := OpenFile(dir, file, unix.O_RDONLY)
+	if err != nil {
+		return "", err
+	}
+	defer fd.Close()
+	var buf bytes.Buffer
+
+	_, err = buf.ReadFrom(fd)
+	return buf.String(), err
+}
+
+// WriteFile writes data to a cgroup file in dir.
+// It is supposed to be used for cgroup files only.
+func WriteFile(dir, file, data string) error {
+	fd, err := OpenFile(dir, file, unix.O_WRONLY)
+	if err != nil {
+		return err
+	}
+	defer fd.Close()
+	if err := retryingWriteFile(fd, data); err != nil {
+		return errors.Wrapf(err, "failed to write %q", data)
+	}
+	return nil
+}
+
+func CopyFile(source, dest string) error {
+	var (
+		srcF *os.File
+		dstF *os.File
+		data []byte
+		err  error
+	)
+
+	srcF, err = os.Open(source)
+	if err != nil {
+		return fmt.Errorf("failed to open %s: %s", source, err)
+	}
+	defer srcF.Close()
+
+	dstF, err = os.Open(dest)
+	if err != nil {
+		dstF.Close()
+		return fmt.Errorf("failed to open %s: %s", dest, err)
+	}
+	defer dstF.Close()
+
+	data, err = ioutil.ReadFile(source)
+	if err != nil {
+		return fmt.Errorf("failed to read %s: %s", source, err)
+	}
+
+	err = ioutil.WriteFile(dest, data, 0)
+	if err != nil {
+		return fmt.Errorf("failed to read %s: %s", dest, err)
+	}
+
+	return nil
+}
+
+func retryingWriteFile(fd *os.File, data string) error {
+	for {
+		_, err := fd.Write([]byte(data))
+		if errors.Is(err, unix.EINTR) {
+			logrus.Infof("interrupted while writing %s to %s", data, fd.Name())
+			continue
+		}
+		return err
+	}
+}
+
 const (
 	cgroupfsDir    = "/sys/fs/cgroup"
 	cgroupfsPrefix = cgroupfsDir + "/"
@@ -61,10 +146,7 @@ func prepareOpenat2() error {
 
 // OpenFile opens a cgroup file in a given dir with given flags.
 // It is supposed to be used for cgroup files only.
-func OpenFile(dir, file string, flags int) (*os.File, error) {
-	if dir == "" {
-		return nil, errors.Errorf("no directory specified for %s", file)
-	}
+func openFile(dir, file string, flags int) (*os.File, error) {
 	mode := os.FileMode(0)
 	if TestMode && flags&os.O_WRONLY != 0 {
 		// "emulate" cgroup fs for unit tests

libcontainer/cgroups/fscommon/fscommon_test.go → libcontainer/cgroups/file_test.go

@@ -1,6 +1,6 @@
 // +build linux
 
-package fscommon
+package cgroups
 
 import (
 	"fmt"

libcontainer/cgroups/fs/fs.go

@@ -32,6 +32,7 @@ var (
 		&NetPrioGroup{},
 		&PerfEventGroup{},
 		&FreezerGroup{},
+		&RdmaGroup{},
 		&NameGroup{GroupName: "name=systemd", Join: true},
 	}
 	HugePageSizes, _ = cgroups.GetHugePageSize()

libcontainer/cgroups/fs/rdma.go

@@ -0,0 +1,41 @@
+package fs
+
+import (
+	"fmt"
+	"os"
+
+	"github.com/opencontainers/runc/libcontainer/cgroups"
+	"github.com/opencontainers/runc/libcontainer/cgroups/fscommon"
+	"github.com/opencontainers/runc/libcontainer/configs"
+)
+
+type RdmaGroup struct{}
+
+func (s *RdmaGroup) Name() string {
+	return "rdma"
+}
+
+func (s *RdmaGroup) Apply(path string, d *cgroupData) error {
+	return join(path, d.pid)
+}
+
+func (s *RdmaGroup) Set(path string, cgroup *configs.Cgroup) error {
+	return fscommon.RdmaSet(path, cgroup)
+}
+
+func (s *RdmaGroup) GetStats(path string, stats *cgroups.Stats) error {
+	return fscommon.RdmaGetStats(path, stats)
+}
+
+func (s *RdmaGroup) Clone(source, dest string) error {
+
+	if err := fscommon.WriteFile(source, "cgroup.clone_children", "1"); err != nil {
+		return err
+	}
+
+	if err := os.MkdirAll(dest, 0755); err != nil {
+		return fmt.Errorf("Failed to create cgroup %s", dest)
+	}
+
+	return nil
+}

libcontainer/cgroups/fs/util_test.go

@@ -13,12 +13,13 @@ import (
 	"path/filepath"
 	"testing"
 
+	"github.com/opencontainers/runc/libcontainer/cgroups"
 	"github.com/opencontainers/runc/libcontainer/cgroups/fscommon"
 	"github.com/opencontainers/runc/libcontainer/configs"
 )
 
 func init() {
-	fscommon.TestMode = true
+	cgroups.TestMode = true
 }
 
 type cgroupTestUtil struct {

libcontainer/cgroups/fs2/fs2.go

@@ -143,6 +143,10 @@ func (m *manager) GetStats() (*cgroups.Stats, error) {
 			errs = append(errs, err)
 		}
 	}
+	// rdma (since kernel 4.11)
+	if err := fscommon.RdmaGetStats(m.dirPath, st); err != nil && !os.IsNotExist(err) {
+		errs = append(errs, err)
+	}
 	if len(errs) > 0 && !m.rootless {
 		return st, errors.Errorf("error while statting cgroup v2: %+v", errs)
 	}
@@ -204,6 +208,10 @@ func (m *manager) Set(container *configs.Config) error {
 	if err := setHugeTlb(m.dirPath, container.Cgroups); err != nil {
 		return err
 	}
+	// rdma (since kernel 4.11)
+	if err := fscommon.RdmaSet(m.dirPath, container.Cgroups); err != nil {
+		return err
+	}
 	// freezer (since kernel 5.2, pseudo-controller)
 	if err := setFreezer(m.dirPath, container.Cgroups.Freezer); err != nil {
 		return err