mirror of
https://github.com/containers/podman.git
synced 2025-10-11 16:26:00 +08:00
use rootless netns from c/common
Use the new rootlessnetns logic from c/common, drop the podman code here and make use of the new much simpler API. ref: https://github.com/containers/common/pull/1761 [NO NEW TESTS NEEDED] Signed-off-by: Paul Holzinger <pholzing@redhat.com>
This commit is contained in:
0
utils/testdata/cgroup.empty
vendored
0
utils/testdata/cgroup.empty
vendored
1
utils/testdata/cgroup.other
vendored
1
utils/testdata/cgroup.other
vendored
@ -1 +0,0 @@
|
||||
0::/other
|
1
utils/testdata/cgroup.root
vendored
1
utils/testdata/cgroup.root
vendored
@ -1 +0,0 @@
|
||||
0::/
|
119
utils/utils.go
119
utils/utils.go
@ -2,20 +2,16 @@ package utils
|
||||
|
||||
import (
|
||||
"bytes"
|
||||
"crypto/rand"
|
||||
"fmt"
|
||||
"io"
|
||||
"os"
|
||||
"os/exec"
|
||||
"strconv"
|
||||
"strings"
|
||||
"sync"
|
||||
"time"
|
||||
|
||||
"github.com/containers/common/pkg/cgroups"
|
||||
"github.com/containers/storage/pkg/archive"
|
||||
"github.com/containers/storage/pkg/chrootarchive"
|
||||
"github.com/godbus/dbus/v5"
|
||||
"github.com/sirupsen/logrus"
|
||||
"github.com/vbauerster/mpb/v8"
|
||||
"github.com/vbauerster/mpb/v8/decor"
|
||||
@ -133,121 +129,6 @@ func RemoveScientificNotationFromFloat(x float64) (float64, error) {
|
||||
return result, nil
|
||||
}
|
||||
|
||||
var (
|
||||
runsOnSystemdOnce sync.Once
|
||||
runsOnSystemd bool
|
||||
)
|
||||
|
||||
// RunsOnSystemd returns whether the system is using systemd
|
||||
func RunsOnSystemd() bool {
|
||||
runsOnSystemdOnce.Do(func() {
|
||||
// per sd_booted(3), check for this dir
|
||||
fd, err := os.Stat("/run/systemd/system")
|
||||
runsOnSystemd = err == nil && fd.IsDir()
|
||||
})
|
||||
return runsOnSystemd
|
||||
}
|
||||
|
||||
func moveProcessPIDFileToScope(pidPath, slice, scope string) error {
|
||||
data, err := os.ReadFile(pidPath)
|
||||
if err != nil {
|
||||
// do not raise an error if the file doesn't exist
|
||||
if os.IsNotExist(err) {
|
||||
return nil
|
||||
}
|
||||
return fmt.Errorf("cannot read pid file: %w", err)
|
||||
}
|
||||
pid, err := strconv.ParseUint(string(data), 10, 0)
|
||||
if err != nil {
|
||||
return fmt.Errorf("cannot parse pid file %s: %w", pidPath, err)
|
||||
}
|
||||
|
||||
return moveProcessToScope(int(pid), slice, scope)
|
||||
}
|
||||
|
||||
func moveProcessToScope(pid int, slice, scope string) error {
|
||||
err := RunUnderSystemdScope(pid, slice, scope)
|
||||
// If the PID is not valid anymore, do not return an error.
|
||||
if dbusErr, ok := err.(dbus.Error); ok {
|
||||
if dbusErr.Name == "org.freedesktop.DBus.Error.UnixProcessIdUnknown" {
|
||||
return nil
|
||||
}
|
||||
}
|
||||
return err
|
||||
}
|
||||
|
||||
// MoveRootlessNetnsSlirpProcessToUserSlice moves the slirp4netns process for the rootless netns
|
||||
// into a different scope so that systemd does not kill it with a container.
|
||||
func MoveRootlessNetnsSlirpProcessToUserSlice(pid int) error {
|
||||
randBytes := make([]byte, 4)
|
||||
_, err := rand.Read(randBytes)
|
||||
if err != nil {
|
||||
return err
|
||||
}
|
||||
return moveProcessToScope(pid, "user.slice", fmt.Sprintf("rootless-netns-%x.scope", randBytes))
|
||||
}
|
||||
|
||||
// MovePauseProcessToScope moves the pause process used for rootless mode to keep the namespaces alive to
|
||||
// a separate scope.
|
||||
func MovePauseProcessToScope(pausePidPath string) {
|
||||
var err error
|
||||
|
||||
for i := 0; i < 10; i++ {
|
||||
randBytes := make([]byte, 4)
|
||||
_, err = rand.Read(randBytes)
|
||||
if err != nil {
|
||||
logrus.Errorf("failed to read random bytes: %v", err)
|
||||
continue
|
||||
}
|
||||
err = moveProcessPIDFileToScope(pausePidPath, "user.slice", fmt.Sprintf("podman-pause-%x.scope", randBytes))
|
||||
if err == nil {
|
||||
return
|
||||
}
|
||||
}
|
||||
|
||||
if err != nil {
|
||||
unified, err2 := cgroups.IsCgroup2UnifiedMode()
|
||||
if err2 != nil {
|
||||
logrus.Warnf("Failed to detect if running with cgroup unified: %v", err)
|
||||
}
|
||||
if RunsOnSystemd() && unified {
|
||||
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)
|
||||
}
|
||||
}
|
||||
}
|
||||
|
||||
var (
|
||||
maybeMoveToSubCgroupSync sync.Once
|
||||
maybeMoveToSubCgroupSyncErr error
|
||||
)
|
||||
|
||||
// MaybeMoveToSubCgroup moves the current process in a sub cgroup when
|
||||
// it is running in the root cgroup on a system that uses cgroupv2.
|
||||
func MaybeMoveToSubCgroup() error {
|
||||
maybeMoveToSubCgroupSync.Do(func() {
|
||||
unifiedMode, err := cgroups.IsCgroup2UnifiedMode()
|
||||
if err != nil {
|
||||
maybeMoveToSubCgroupSyncErr = err
|
||||
return
|
||||
}
|
||||
if !unifiedMode {
|
||||
maybeMoveToSubCgroupSyncErr = nil
|
||||
return
|
||||
}
|
||||
cgroup, err := GetOwnCgroup()
|
||||
if err != nil {
|
||||
maybeMoveToSubCgroupSyncErr = err
|
||||
return
|
||||
}
|
||||
if cgroup == "/" {
|
||||
maybeMoveToSubCgroupSyncErr = MoveUnderCgroupSubtree("init")
|
||||
}
|
||||
})
|
||||
return maybeMoveToSubCgroupSyncErr
|
||||
}
|
||||
|
||||
// GuardedRemoveAll functions much like os.RemoveAll but
|
||||
// will not delete certain catastrophic paths.
|
||||
func GuardedRemoveAll(path string) error {
|
||||
|
@ -1,205 +0,0 @@
|
||||
//go:build linux || darwin || freebsd
|
||||
// +build linux darwin freebsd
|
||||
|
||||
package utils
|
||||
|
||||
import (
|
||||
"bufio"
|
||||
"bytes"
|
||||
"context"
|
||||
"fmt"
|
||||
"os"
|
||||
"path/filepath"
|
||||
"strings"
|
||||
|
||||
"github.com/containers/common/pkg/cgroups"
|
||||
"github.com/containers/podman/v4/pkg/rootless"
|
||||
systemdDbus "github.com/coreos/go-systemd/v22/dbus"
|
||||
"github.com/godbus/dbus/v5"
|
||||
"github.com/sirupsen/logrus"
|
||||
)
|
||||
|
||||
// RunUnderSystemdScope adds the specified pid to a systemd scope
|
||||
func RunUnderSystemdScope(pid int, slice string, unitName string) error {
|
||||
var properties []systemdDbus.Property
|
||||
var conn *systemdDbus.Conn
|
||||
var err error
|
||||
|
||||
if rootless.IsRootless() {
|
||||
conn, err = cgroups.UserConnection(rootless.GetRootlessUID())
|
||||
if err != nil {
|
||||
return err
|
||||
}
|
||||
} else {
|
||||
conn, err = systemdDbus.NewWithContext(context.Background())
|
||||
if err != nil {
|
||||
return err
|
||||
}
|
||||
}
|
||||
defer conn.Close()
|
||||
properties = append(properties, systemdDbus.PropSlice(slice))
|
||||
properties = append(properties, newProp("PIDs", []uint32{uint32(pid)}))
|
||||
properties = append(properties, newProp("Delegate", true))
|
||||
properties = append(properties, newProp("DefaultDependencies", false))
|
||||
ch := make(chan string)
|
||||
_, err = conn.StartTransientUnitContext(context.Background(), unitName, "replace", properties, ch)
|
||||
if err != nil {
|
||||
// On errors check if the cgroup already exists, if it does move the process there
|
||||
if props, err := conn.GetUnitTypePropertiesContext(context.Background(), unitName, "Scope"); err == nil {
|
||||
if cgroup, ok := props["ControlGroup"].(string); ok && cgroup != "" {
|
||||
if err := MoveUnderCgroup(cgroup, "", []uint32{uint32(pid)}); err == nil {
|
||||
return nil
|
||||
}
|
||||
// On errors return the original error message we got from StartTransientUnit.
|
||||
}
|
||||
}
|
||||
return err
|
||||
}
|
||||
|
||||
// Block until job is started
|
||||
<-ch
|
||||
|
||||
return nil
|
||||
}
|
||||
|
||||
func getCgroupProcess(procFile string, allowRoot bool) (string, error) {
|
||||
f, err := os.Open(procFile)
|
||||
if err != nil {
|
||||
return "", err
|
||||
}
|
||||
defer f.Close()
|
||||
|
||||
scanner := bufio.NewScanner(f)
|
||||
cgroup := ""
|
||||
for scanner.Scan() {
|
||||
line := scanner.Text()
|
||||
parts := strings.SplitN(line, ":", 3)
|
||||
if len(parts) != 3 {
|
||||
return "", fmt.Errorf("cannot parse cgroup line %q", line)
|
||||
}
|
||||
if strings.HasPrefix(line, "0::") {
|
||||
cgroup = line[3:]
|
||||
break
|
||||
}
|
||||
if len(parts[2]) > len(cgroup) {
|
||||
cgroup = parts[2]
|
||||
}
|
||||
}
|
||||
if len(cgroup) == 0 || (!allowRoot && cgroup == "/") {
|
||||
return "", fmt.Errorf("could not find cgroup mount in %q", procFile)
|
||||
}
|
||||
return cgroup, nil
|
||||
}
|
||||
|
||||
// GetOwnCgroup returns the cgroup for the current process.
|
||||
func GetOwnCgroup() (string, error) {
|
||||
return getCgroupProcess("/proc/self/cgroup", true)
|
||||
}
|
||||
|
||||
func GetOwnCgroupDisallowRoot() (string, error) {
|
||||
return getCgroupProcess("/proc/self/cgroup", false)
|
||||
}
|
||||
|
||||
// GetCgroupProcess returns the cgroup for the specified process process.
|
||||
func GetCgroupProcess(pid int) (string, error) {
|
||||
return getCgroupProcess(fmt.Sprintf("/proc/%d/cgroup", pid), true)
|
||||
}
|
||||
|
||||
// MoveUnderCgroupSubtree moves the PID under a cgroup subtree.
|
||||
func MoveUnderCgroupSubtree(subtree string) error {
|
||||
return MoveUnderCgroup("", subtree, nil)
|
||||
}
|
||||
|
||||
// MoveUnderCgroup moves a group of processes to a new cgroup.
|
||||
// If cgroup is the empty string, then the current calling process cgroup is used.
|
||||
// If processes is empty, then the processes from the current cgroup are moved.
|
||||
func MoveUnderCgroup(cgroup, subtree string, processes []uint32) error {
|
||||
procFile := "/proc/self/cgroup"
|
||||
f, err := os.Open(procFile)
|
||||
if err != nil {
|
||||
return err
|
||||
}
|
||||
defer f.Close()
|
||||
|
||||
unifiedMode, err := cgroups.IsCgroup2UnifiedMode()
|
||||
if err != nil {
|
||||
return err
|
||||
}
|
||||
|
||||
scanner := bufio.NewScanner(f)
|
||||
for scanner.Scan() {
|
||||
line := scanner.Text()
|
||||
parts := strings.SplitN(line, ":", 3)
|
||||
if len(parts) != 3 {
|
||||
return fmt.Errorf("cannot parse cgroup line %q", line)
|
||||
}
|
||||
|
||||
// root cgroup, skip it
|
||||
if parts[2] == "/" && !(unifiedMode && parts[1] == "") {
|
||||
continue
|
||||
}
|
||||
|
||||
cgroupRoot := "/sys/fs/cgroup"
|
||||
// Special case the unified mount on hybrid cgroup and named hierarchies.
|
||||
// This works on Fedora 31, but we should really parse the mounts to see
|
||||
// where the cgroup hierarchy is mounted.
|
||||
if parts[1] == "" && !unifiedMode {
|
||||
// If it is not using unified mode, the cgroup v2 hierarchy is
|
||||
// usually mounted under /sys/fs/cgroup/unified
|
||||
cgroupRoot = filepath.Join(cgroupRoot, "unified")
|
||||
|
||||
// Ignore the unified mount if it doesn't exist
|
||||
if _, err := os.Stat(cgroupRoot); err != nil && os.IsNotExist(err) {
|
||||
continue
|
||||
}
|
||||
} else if parts[1] != "" {
|
||||
// Assume the controller is mounted at /sys/fs/cgroup/$CONTROLLER.
|
||||
controller := strings.TrimPrefix(parts[1], "name=")
|
||||
cgroupRoot = filepath.Join(cgroupRoot, controller)
|
||||
}
|
||||
|
||||
parentCgroup := cgroup
|
||||
if parentCgroup == "" {
|
||||
parentCgroup = parts[2]
|
||||
}
|
||||
newCgroup := filepath.Join(cgroupRoot, parentCgroup, subtree)
|
||||
if err := os.MkdirAll(newCgroup, 0755); err != nil && !os.IsExist(err) {
|
||||
return err
|
||||
}
|
||||
|
||||
f, err := os.OpenFile(filepath.Join(newCgroup, "cgroup.procs"), os.O_RDWR, 0755)
|
||||
if err != nil {
|
||||
return err
|
||||
}
|
||||
defer f.Close()
|
||||
|
||||
if len(processes) > 0 {
|
||||
for _, pid := range processes {
|
||||
if _, err := f.WriteString(fmt.Sprintf("%d\n", pid)); err != nil {
|
||||
logrus.Debugf("Cannot move process %d to cgroup %q: %v", pid, newCgroup, err)
|
||||
}
|
||||
}
|
||||
} else {
|
||||
processesData, err := os.ReadFile(filepath.Join(cgroupRoot, parts[2], "cgroup.procs"))
|
||||
if err != nil {
|
||||
return err
|
||||
}
|
||||
for _, pid := range bytes.Split(processesData, []byte("\n")) {
|
||||
if len(pid) == 0 {
|
||||
continue
|
||||
}
|
||||
if _, err := f.Write(pid); err != nil {
|
||||
logrus.Debugf("Cannot move process %s to cgroup %q: %v", string(pid), newCgroup, err)
|
||||
}
|
||||
}
|
||||
}
|
||||
}
|
||||
return nil
|
||||
}
|
||||
|
||||
func newProp(name string, units interface{}) systemdDbus.Property {
|
||||
return systemdDbus.Property{
|
||||
Name: name,
|
||||
Value: dbus.MakeVariant(units),
|
||||
}
|
||||
}
|
@ -1,26 +0,0 @@
|
||||
//go:build linux || darwin || freebsd
|
||||
// +build linux darwin freebsd
|
||||
|
||||
package utils
|
||||
|
||||
import (
|
||||
"testing"
|
||||
|
||||
"github.com/stretchr/testify/assert"
|
||||
)
|
||||
|
||||
func TestCgroupProcess(t *testing.T) {
|
||||
val, err := getCgroupProcess("testdata/cgroup.root", true)
|
||||
assert.Nil(t, err)
|
||||
assert.Equal(t, "/", val)
|
||||
|
||||
_, err = getCgroupProcess("testdata/cgroup.root", false)
|
||||
assert.NotNil(t, err)
|
||||
|
||||
val, err = getCgroupProcess("testdata/cgroup.other", true)
|
||||
assert.Nil(t, err)
|
||||
assert.Equal(t, "/other", val)
|
||||
|
||||
_, err = getCgroupProcess("testdata/cgroup.empty", true)
|
||||
assert.NotNil(t, err)
|
||||
}
|
@ -1,26 +0,0 @@
|
||||
//go:build windows
|
||||
// +build windows
|
||||
|
||||
package utils
|
||||
|
||||
import "errors"
|
||||
|
||||
func RunUnderSystemdScope(pid int, slice string, unitName string) error {
|
||||
return errors.New("not implemented for windows")
|
||||
}
|
||||
|
||||
func MoveUnderCgroupSubtree(subtree string) error {
|
||||
return errors.New("not implemented for windows")
|
||||
}
|
||||
|
||||
func GetOwnCgroup() (string, error) {
|
||||
return "", errors.New("not implemented for windows")
|
||||
}
|
||||
|
||||
func GetOwnCgroupDisallowRoot() (string, error) {
|
||||
return "", errors.New("not implemented for windows")
|
||||
}
|
||||
|
||||
func GetCgroupProcess(pid int) (string, error) {
|
||||
return "", errors.New("not implemented for windows")
|
||||
}
|
Reference in New Issue
Block a user