Merge pull request #3959 from giuseppe/rootless-use-systemd-scope

rootless: automatically create a systemd scope
This commit is contained in:
OpenShift Merge Robot
2019-09-12 19:04:07 +02:00
committed by GitHub
5 changed files with 187 additions and 35 deletions

View File

@ -5,9 +5,12 @@ package main
import (
"context"
"fmt"
"io/ioutil"
"log/syslog"
"os"
"runtime/pprof"
"strconv"
"strings"
"syscall"
@ -18,6 +21,7 @@ import (
"github.com/containers/libpod/pkg/rootless"
"github.com/containers/libpod/pkg/tracing"
"github.com/containers/libpod/pkg/util"
"github.com/containers/libpod/utils"
"github.com/opentracing/opentracing-go"
"github.com/pkg/errors"
"github.com/sirupsen/logrus"
@ -119,7 +123,29 @@ func profileOff(cmd *cobra.Command) error {
return nil
}
func movePauseProcessToScope() error {
pausePidPath, err := util.GetRootlessPauseProcessPidPath()
if err != nil {
return errors.Wrapf(err, "could not get pause process pid file path")
}
data, err := ioutil.ReadFile(pausePidPath)
if err != nil {
return errors.Wrapf(err, "cannot read pause pid file")
}
pid, err := strconv.ParseUint(string(data), 10, 0)
if err != nil {
return errors.Wrapf(err, "cannot parse pid file %s", pausePidPath)
}
return utils.RunUnderSystemdScope(int(pid), "user.slice", "podman-pause.scope")
}
func setupRootless(cmd *cobra.Command, args []string) error {
if !rootless.IsRootless() {
return nil
}
matches, err := rootless.ConfigurationMatches()
if err != nil {
return err
@ -128,9 +154,6 @@ func setupRootless(cmd *cobra.Command, args []string) error {
logrus.Warningf("the current user namespace doesn't match the configuration in /etc/subuid or /etc/subgid")
logrus.Warningf("you can use `%s system migrate` to recreate the user namespace and restart the containers", os.Args[0])
}
if os.Geteuid() == 0 || cmd == _searchCommand || cmd == _versionCommand || cmd == _mountCommand || cmd == _migrateCommand || strings.HasPrefix(cmd.Use, "help") {
return nil
}
podmanCmd := cliconfig.PodmanCommand{
Command: cmd,
@ -139,6 +162,39 @@ func setupRootless(cmd *cobra.Command, args []string) error {
Remote: remoteclient,
}
runtime, err := libpodruntime.GetRuntime(getContext(), &podmanCmd)
if err != nil {
return errors.Wrapf(err, "could not get runtime")
}
defer runtime.DeferredShutdown(false)
// do it only after podman has already re-execed and running with uid==0.
if os.Geteuid() == 0 {
ownsCgroup, err := cgroups.UserOwnsCurrentSystemdCgroup()
if err != nil {
return err
}
if !ownsCgroup {
unitName := fmt.Sprintf("podman-%d.scope", os.Getpid())
if err := utils.RunUnderSystemdScope(os.Getpid(), "user.slice", unitName); err != nil {
conf, err := runtime.GetConfig()
if err != nil {
return err
}
if conf.CgroupManager == libpod.SystemdCgroupsManager {
logrus.Warnf("Failed to add podman to systemd sandbox cgroup: %v", err)
} else {
logrus.Debugf("Failed to add podman to systemd sandbox cgroup: %v", err)
}
}
}
}
if os.Geteuid() == 0 || cmd == _searchCommand || cmd == _versionCommand || cmd == _mountCommand || cmd == _migrateCommand || strings.HasPrefix(cmd.Use, "help") {
return nil
}
pausePidPath, err := util.GetRootlessPauseProcessPidPath()
if err != nil {
return errors.Wrapf(err, "could not get pause process pid file path")
@ -158,13 +214,6 @@ func setupRootless(cmd *cobra.Command, args []string) error {
}
// if there is no pid file, try to join existing containers, and create a pause process.
runtime, err := libpodruntime.GetRuntime(getContext(), &podmanCmd)
if err != nil {
return errors.Wrapf(err, "could not get runtime")
}
defer runtime.DeferredShutdown(false)
ctrs, err := runtime.GetRunningContainers()
if err != nil {
logrus.Errorf(err.Error())
@ -177,6 +226,17 @@ func setupRootless(cmd *cobra.Command, args []string) error {
}
became, ret, err := rootless.TryJoinFromFilePaths(pausePidPath, true, paths)
if err := movePauseProcessToScope(); err != nil {
conf, err := runtime.GetConfig()
if err != nil {
return err
}
if conf.CgroupManager == libpod.SystemdCgroupsManager {
logrus.Warnf("Failed to add pause process to systemd sandbox cgroup: %v", err)
} else {
logrus.Debugf("Failed to add pause process to systemd sandbox cgroup: %v", err)
}
}
if err != nil {
logrus.Errorf(err.Error())
os.Exit(1)

View File

@ -21,6 +21,7 @@ import (
"github.com/containers/libpod/pkg/cgroups"
"github.com/containers/libpod/pkg/errorhandling"
"github.com/containers/libpod/pkg/lookup"
"github.com/containers/libpod/pkg/rootless"
"github.com/containers/libpod/pkg/util"
"github.com/containers/libpod/utils"
"github.com/coreos/go-systemd/activation"
@ -359,11 +360,21 @@ func startCommandGivenSelinux(cmd *exec.Cmd) error {
// moveConmonToCgroupAndSignal gets a container's cgroupParent and moves the conmon process to that cgroup
// it then signals for conmon to start by sending nonse data down the start fd
func (r *OCIRuntime) moveConmonToCgroupAndSignal(ctr *Container, cmd *exec.Cmd, startFd *os.File, uuid string) error {
mustCreateCgroup := true
// If cgroup creation is disabled - just signal.
if ctr.config.NoCgroups {
return writeConmonPipeData(startFd)
mustCreateCgroup = false
}
if rootless.IsRootless() {
ownsCgroup, err := cgroups.UserOwnsCurrentSystemdCgroup()
if err != nil {
return err
}
mustCreateCgroup = !ownsCgroup
}
if mustCreateCgroup {
cgroupParent := ctr.CgroupParent()
if r.cgroupManager == SystemdCgroupsManager {
unitName := createUnitName("libpod-conmon", ctr.ID())
@ -391,6 +402,7 @@ func (r *OCIRuntime) moveConmonToCgroupAndSignal(ctr *Container, cmd *exec.Cmd,
}
}
}
}
/* We set the cgroup, now the child can start creating children */
if err := writeConmonPipeData(startFd); err != nil {

View File

@ -3,8 +3,15 @@
package cgroups
import (
"bufio"
"fmt"
"os"
"path/filepath"
"strings"
"sync"
"syscall"
"github.com/pkg/errors"
)
var (
@ -25,3 +32,58 @@ func IsCgroup2UnifiedMode() (bool, error) {
})
return isUnified, isUnifiedErr
}
// UserOwnsCurrentSystemdCgroup checks whether the current EUID owns the
// current cgroup.
func UserOwnsCurrentSystemdCgroup() (bool, error) {
uid := os.Geteuid()
cgroup2, err := IsCgroup2UnifiedMode()
if err != nil {
return false, err
}
f, err := os.Open("/proc/self/cgroup")
if err != nil {
return false, errors.Wrapf(err, "open file /proc/self/cgroup")
}
defer f.Close()
scanner := bufio.NewScanner(f)
for scanner.Scan() {
line := scanner.Text()
parts := strings.SplitN(line, ":", 3)
if len(parts) < 3 {
continue
}
var cgroupPath string
if cgroup2 {
cgroupPath = filepath.Join(cgroupRoot, parts[2])
} else {
if parts[1] != "name=systemd" {
continue
}
cgroupPath = filepath.Join(cgroupRoot, "systemd", parts[2])
}
st, err := os.Stat(cgroupPath)
if err != nil {
return false, err
}
s := st.Sys()
if s == nil {
return false, fmt.Errorf("error stat cgroup path %s", cgroupPath)
}
if int(s.(*syscall.Stat_t).Uid) != uid {
return false, nil
}
}
if err := scanner.Err(); err != nil {
return false, errors.Wrapf(err, "parsing file /proc/self/cgroup")
}
return true, nil
}

View File

@ -6,3 +6,9 @@ package cgroups
func IsCgroup2UnifiedMode() (bool, error) {
return false, nil
}
// UserOwnsCurrentSystemdCgroup checks whether the current EUID owns the
// current cgroup.
func UserOwnsCurrentSystemdCgroup() (bool, error) {
return false, nil
}

View File

@ -3,6 +3,8 @@
package utils
import (
"github.com/containers/libpod/pkg/cgroups"
"github.com/containers/libpod/pkg/rootless"
systemdDbus "github.com/coreos/go-systemd/dbus"
"github.com/godbus/dbus"
)
@ -10,10 +12,20 @@ import (
// RunUnderSystemdScope adds the specified pid to a systemd scope
func RunUnderSystemdScope(pid int, slice string, unitName string) error {
var properties []systemdDbus.Property
conn, err := systemdDbus.New()
var conn *systemdDbus.Conn
var err error
if rootless.IsRootless() {
conn, err = cgroups.GetUserConnection(rootless.GetRootlessUID())
if err != nil {
return err
}
} else {
conn, err = systemdDbus.New()
if err != nil {
return err
}
}
properties = append(properties, systemdDbus.PropSlice(slice))
properties = append(properties, newProp("PIDs", []uint32{uint32(pid)}))
properties = append(properties, newProp("Delegate", true))