Skip to content
New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

Sign up for GitHub

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

core/mount: use ptrace instead of go:linkname #10611

Merged
merged 2 commits into from
Aug 30, 2024
Merged

core/mount: use ptrace instead of go:linkname #10611

Show file tree
Hide file tree
Changes from all commits
Commits
Show all changes
2 commits
Select commit Hold shift + click to select a range
bcdf507
core/mount: add benchmark test for GetUsernsFD
fuweid Aug 19, 2024
3cd8f97
core/mount: use ptrace instead of go:linkname
fuweid Aug 19, 2024
File filter

Filter by extension

Filter by extension
Conversations
Failed to load comments.
Loading
Jump to
Jump to file
Failed to load files.
Loading
Diff view
Diff view
100 changes: 0 additions & 100 deletions core/mount/mount_idmapped_linux.go
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -19,15 +19,11 @@ package mount
import (
"fmt"
"os"
"runtime"
"strconv"
"strings"
"sync"
"syscall"

"golang.org/x/sys/unix"

"github.com/containerd/containerd/v2/pkg/sys"
)

// TODO: Support multiple mappings in future
Expand Down Expand Up @@ -106,99 +102,3 @@ func GetUsernsFD(uidmap, gidmap string) (_usernsFD *os.File, _ error) {
}
return getUsernsFD(uidMaps, gidMaps)
}

func getUsernsFD(uidMaps, gidMaps []syscall.SysProcIDMap) (_usernsFD *os.File, retErr error) {
runtime.LockOSThread()
defer runtime.UnlockOSThread()

pid, pidfd, errno := sys.ForkUserns()
if errno != 0 {
return nil, errno
}

pidFD := os.NewFile(pidfd, "pidfd")
defer func() {
unix.PidfdSendSignal(int(pidFD.Fd()), unix.SIGKILL, nil, 0)

pidfdWaitid(pidFD)

pidFD.Close()
}()

// NOTE:
//
// The usernsFD will hold the userns reference in kernel. Even if the
// child process is reaped, the usernsFD is still valid.
usernsFD, err := os.Open(fmt.Sprintf("/proc/%d/ns/user", pid))
if err != nil {
return nil, fmt.Errorf("failed to get userns file descriptor for /proc/%d/user/ns: %w", pid, err)
}
defer func() {
if retErr != nil {
usernsFD.Close()
}
}()

uidmapFile, err := os.OpenFile(fmt.Sprintf("/proc/%d/%s", pid, "uid_map"), os.O_WRONLY, 0600)
if err != nil {
return nil, fmt.Errorf("failed to open /proc/%d/uid_map: %w", pid, err)
}
defer uidmapFile.Close()

gidmapFile, err := os.OpenFile(fmt.Sprintf("/proc/%d/%s", pid, "gid_map"), os.O_WRONLY, 0600)
if err != nil {
return nil, fmt.Errorf("failed to open /proc/%d/gid_map: %w", pid, err)
}
defer gidmapFile.Close()

testHookKillChildBeforePidfdSendSignal(pid, pidFD)

// Ensure the child process is still alive. If the err is ESRCH, we
// should return error because we can't guarantee the usernsFD and
// u[g]idmapFile are valid. It's safe to return error and retry.
if err := unix.PidfdSendSignal(int(pidFD.Fd()), 0, nil, 0); err != nil {
return nil, fmt.Errorf("failed to ensure child process is alive: %w", err)
}

testHookKillChildAfterPidfdSendSignal(pid, pidFD)

// NOTE:
//
// The u[g]id_map file descriptor is still valid if the child process
// is reaped.
writeMappings := func(f *os.File, idmap []syscall.SysProcIDMap) error {
mappings := ""
for _, m := range idmap {
mappings = fmt.Sprintf("%s%d %d %d\n", mappings, m.ContainerID, m.HostID, m.Size)
}

_, err := f.Write([]byte(mappings))
if err1 := f.Close(); err1 != nil && err == nil {
err = err1
}
return err
}

if err := writeMappings(uidmapFile, uidMaps); err != nil {
return nil, fmt.Errorf("failed to write uid_map: %w", err)
}

if err := writeMappings(gidmapFile, gidMaps); err != nil {
return nil, fmt.Errorf("failed to write gid_map: %w", err)
}
return usernsFD, nil
}

func pidfdWaitid(pidFD *os.File) error {
return sys.IgnoringEINTR(func() error {
return unix.Waitid(unix.P_PIDFD, int(pidFD.Fd()), nil, unix.WEXITED, nil)
})
}

var (
testHookLock sync.Mutex

testHookKillChildBeforePidfdSendSignal = func(_pid uintptr, _pidFD *os.File) {}

testHookKillChildAfterPidfdSendSignal = func(_pid uintptr, _pidFD *os.File) {}
)
154 changes: 46 additions & 108 deletions core/mount/mount_idmapped_linux_test.go
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -20,15 +20,43 @@ import (
"fmt"
"os"
"path/filepath"
"sync"
"syscall"
"testing"

kernel "github.com/containerd/containerd/v2/pkg/kernelversion"
"github.com/containerd/continuity/testutil"
"github.com/stretchr/testify/require"
"golang.org/x/sys/unix"
)

func BenchmarkBatchRunGetUsernsFD_Concurrent1(b *testing.B) {
for range b.N {
benchmarkBatchRunGetUsernsFD(1)
}
}

func BenchmarkBatchRunGetUsernsFD_Concurrent10(b *testing.B) {
for range b.N {
benchmarkBatchRunGetUsernsFD(10)
}
}

func benchmarkBatchRunGetUsernsFD(n int) {
var wg sync.WaitGroup
wg.Add(n)
for i := 0; i < n; i++ {
go func() {
defer wg.Done()
fd, err := getUsernsFD(testUIDMaps, testGIDMaps)
if err != nil {
panic(err)
}
fd.Close()
}()
}
wg.Wait()
}

var (
testUIDMaps = []syscall.SysProcIDMap{
{ContainerID: 1000, HostID: 0, Size: 100},
Expand All @@ -43,7 +71,7 @@ var (
}
)

func TestGetUsernsFD(t *testing.T) {
func TestIdmappedMount(t *testing.T) {
testutil.RequiresRoot(t)

k512 := kernel.KernelVersion{Kernel: 5, Major: 12}
Expand All @@ -53,15 +81,12 @@ func TestGetUsernsFD(t *testing.T) {
t.Skip("GetUsernsFD requires kernel >= 5.12")
}

t.Run("basic", testGetUsernsFDBasic)

t.Run("when kill child process before write u[g]id maps", testGetUsernsFDKillChildWhenWriteUGIDMaps)

t.Run("when kill child process after open u[g]id_map file", testGetUsernsFDKillChildAfterOpenUGIDMapFiles)
t.Run("GetUsernsFD", testGetUsernsFD)

t.Run("IDMapMount", testIDMapMount)
}

func testGetUsernsFDBasic(t *testing.T) {
func testGetUsernsFD(t *testing.T) {
for idx, tc := range []struct {
uidMaps string
gidMaps string
Expand Down Expand Up @@ -99,108 +124,21 @@ func testGetUsernsFDBasic(t *testing.T) {
}
}

func testGetUsernsFDKillChildWhenWriteUGIDMaps(t *testing.T) {
hookFunc := func(reap bool) func(uintptr, *os.File) {
return func(_pid uintptr, pidFD *os.File) {
err := unix.PidfdSendSignal(int(pidFD.Fd()), unix.SIGKILL, nil, 0)
require.NoError(t, err)

if reap {
pidfdWaitid(pidFD)
}
}
}

for _, tcReap := range []bool{true, false} {
t.Run(fmt.Sprintf("#reap=%v", tcReap), func(t *testing.T) {
updateTestHookKillForGetUsernsFD(t, nil, hookFunc(tcReap))

usernsFD, err := getUsernsFD(testUIDMaps, testGIDMaps)
require.NoError(t, err)
defer usernsFD.Close()

srcDir, checkFunc := initIDMappedChecker(t, testUIDMaps, testGIDMaps)
destDir := t.TempDir()
defer func() {
require.NoError(t, UnmountAll(destDir, 0))
}()

err = IDMapMount(srcDir, destDir, int(usernsFD.Fd()))
usernsFD.Close()
require.NoError(t, err)
checkFunc(destDir)
})
}

}

func testGetUsernsFDKillChildAfterOpenUGIDMapFiles(t *testing.T) {
hookFunc := func(reap bool) func(uintptr, *os.File) {
return func(_pid uintptr, pidFD *os.File) {
err := unix.PidfdSendSignal(int(pidFD.Fd()), unix.SIGKILL, nil, 0)
require.NoError(t, err)

if reap {
pidfdWaitid(pidFD)
}
}
}

for _, tc := range []struct {
reap bool
expected error
}{
{
reap: false,
expected: nil,
},
{
reap: true,
expected: syscall.ESRCH,
},
} {
t.Run(fmt.Sprintf("#reap=%v", tc.reap), func(t *testing.T) {
updateTestHookKillForGetUsernsFD(t, hookFunc(tc.reap), nil)

usernsFD, err := getUsernsFD(testUIDMaps, testGIDMaps)
if tc.expected != nil {
require.Error(t, tc.expected, err)
return
}

require.NoError(t, err)
defer usernsFD.Close()
func testIDMapMount(t *testing.T) {
usernsFD, err := getUsernsFD(testUIDMaps, testGIDMaps)
require.NoError(t, err)
defer usernsFD.Close()

srcDir, checkFunc := initIDMappedChecker(t, testUIDMaps, testGIDMaps)
destDir := t.TempDir()
defer func() {
require.NoError(t, UnmountAll(destDir, 0))
}()
srcDir, checkFunc := initIDMappedChecker(t, testUIDMaps, testGIDMaps)
destDir := t.TempDir()
defer func() {
require.NoError(t, UnmountAll(destDir, 0))
}()

err = IDMapMount(srcDir, destDir, int(usernsFD.Fd()))
usernsFD.Close()
require.NoError(t, err)
checkFunc(destDir)
})
}
}

func updateTestHookKillForGetUsernsFD(t *testing.T, newBeforeFunc, newAfterFunc func(uintptr, *os.File)) {
testHookLock.Lock()

oldBefore := testHookKillChildBeforePidfdSendSignal
oldAfter := testHookKillChildAfterPidfdSendSignal
t.Cleanup(func() {
testHookKillChildBeforePidfdSendSignal = oldBefore
testHookKillChildAfterPidfdSendSignal = oldAfter
testHookLock.Unlock()
})
if newBeforeFunc != nil {
testHookKillChildBeforePidfdSendSignal = newBeforeFunc
}
if newAfterFunc != nil {
testHookKillChildAfterPidfdSendSignal = newAfterFunc
}
err = IDMapMount(srcDir, destDir, int(usernsFD.Fd()))
usernsFD.Close()
require.NoError(t, err)
checkFunc(destDir)
}

func initIDMappedChecker(t *testing.T, uidMaps, gidMaps []syscall.SysProcIDMap) (_srcDir string, _verifyFunc func(destDir string)) {
Expand Down
85 changes: 85 additions & 0 deletions core/mount/mount_idmapped_utils_linux.go
View file
Open in desktop
Original file line number Diff line number Diff line change
@@ -0,0 +1,85 @@
//go:build go1.23 && linux

/*
Copyright The containerd Authors.

Licensed under the Apache License, Version 2.0 (the "License");
you may not use this file except in compliance with the License.
You may obtain a copy of the License at

http://www.apache.org/licenses/LICENSE-2.0

Unless required by applicable law or agreed to in writing, software
distributed under the License is distributed on an "AS IS" BASIS,
WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
See the License for the specific language governing permissions and
limitations under the License.
*/

package mount

import (
"fmt"
"os"
"syscall"

"github.com/containerd/containerd/v2/pkg/sys"

"golang.org/x/sys/unix"
)

// getUsernsFD returns pinnable user namespace's file descriptor.
//
// NOTE: The GO runtime uses pidfd to handle subprocess since go1.23. However,
// it has double close issue tracked by [1]. We can't use pidfd directly and
// the GO runtime doesn't export interface to show if it's using pidfd or not.
// So, we call `sys.SupportsPidFD` first and then use `os.Process` directly.
//
// [1]: https://github.com/golang/go/issues/68984
func getUsernsFD(uidMaps, gidMaps []syscall.SysProcIDMap) (_ *os.File, retErr error) {
if !sys.SupportsPidFD() {
return nil, fmt.Errorf("failed to prevent pid reused issue because pidfd isn't supported")
}

proc, err := os.StartProcess("/proc/self/exe", []string{"containerd[getUsernsFD]"}, &os.ProcAttr{
Sys: &syscall.SysProcAttr{
Cloneflags: unix.CLONE_NEWUSER,
UidMappings: uidMaps,
GidMappings: gidMaps,
// NOTE: It's reexec but it's not heavy because subprocess
// be in PTRACE_TRACEME mode before performing execve.
Ptrace: true,
Pdeathsig: syscall.SIGKILL,
},
})
if err != nil {
return nil, fmt.Errorf("failed to start noop process for unshare: %w", err)
}

defer func() {
proc.Kill()
proc.Wait()
}()

// NOTE:
//
// The usernsFD will hold the userns reference in kernel. Even if the
// child process is reaped, the usernsFD is still valid.
usernsFD, err := os.Open(fmt.Sprintf("/proc/%d/ns/user", proc.Pid))
if err != nil {
return nil, fmt.Errorf("failed to get userns file descriptor for /proc/%d/user/ns: %w", proc.Pid, err)
}
defer func() {
if retErr != nil {
usernsFD.Close()
}
}()

// Ensure the child process is still alive. If the err is ESRCH, we
// should return error because we can't guarantee the usernsFD and
// u[g]idmapFile are valid. It's safe to return error and retry.
if err := proc.Signal(syscall.Signal(0)); err != nil {
return nil, fmt.Errorf("failed to ensure child process is alive: %w", err)
mikebrow marked this conversation as resolved.
Show resolved Hide resolved
}
return usernsFD, nil
}
Loading
Loading