func setUser(username string) error {
if username == "" {
return errors.New("-username argument missing")
}
newUser, err := user.Lookup(username)
if err != nil {
return err
}
uid, err := strconv.Atoi(newUser.Uid)
if err != nil {
return err
}
gid, err := strconv.Atoi(newUser.Gid)
if err != nil {
return err
}
if uid == 0 {
return errors.New("Do not run the Dominator as root")
os.Exit(1)
}
if err := syscall.Setresgid(gid, gid, gid); err != nil {
return err
}
return syscall.Setresuid(uid, uid, uid)
}
// DropRootTo calls setresgid(2) and setresuid(2) to permenently
// revoke root privileges. The USER, LOGNAME and HOME environmental
// variables are updated to match.
func DropRootTo(u *user.User) (err error) {
var uid, gid int
if gid, err = strconv.Atoi(u.Gid); err != nil {
return
}
if uid, err = strconv.Atoi(u.Uid); err != nil {
return
}
if err = syscall.Setresgid(gid, gid, gid); err != nil {
return
}
if err = syscall.Setresuid(uid, uid, uid); err != nil {
return
}
// based on what Avahi does.
os.Setenv("USER", u.Name)
os.Setenv("LOGNAME", u.Name)
os.Setenv("HOME", u.HomeDir)
return nil
}
func Setresuid(ruid, euid, suid int) error {
return syscall.Setresuid(ruid, euid, suid)
}
func stage1() int {
uuid, err := types.NewUUID(flag.Arg(0))
if err != nil {
log.Print("UUID is missing or malformed\n")
return 1
}
root := "."
p, err := stage1commontypes.LoadPod(root, uuid)
if err != nil {
log.PrintE("can't load pod", err)
return 1
}
// Sanity checks
if len(p.Manifest.Apps) != 1 {
log.Printf("flavor %q only supports 1 application per Pod for now", flavor)
return 1
}
ra := p.Manifest.Apps[0]
imgName := p.AppNameToImageName(ra.Name)
args := ra.App.Exec
if len(args) == 0 {
log.Printf(`image %q has an empty "exec" (try --exec=BINARY)`, imgName)
return 1
}
lfd, err := common.GetRktLockFD()
if err != nil {
log.PrintE("can't get rkt lock fd", err)
return 1
}
// set close-on-exec flag on RKT_LOCK_FD so it gets correctly closed after execution is finished
if err := sys.CloseOnExec(lfd, true); err != nil {
log.PrintE("can't set FD_CLOEXEC on rkt lock", err)
return 1
}
workDir := "/"
if ra.App.WorkingDirectory != "" {
workDir = ra.App.WorkingDirectory
}
env := []string{"PATH=/usr/local/sbin:/usr/local/bin:/usr/sbin:/usr/bin:/sbin:/bin"}
for _, e := range ra.App.Environment {
env = append(env, e.Name+"="+e.Value)
}
rfs := filepath.Join(common.AppPath(p.Root, ra.Name), "rootfs")
if err := copyResolv(p); err != nil {
log.PrintE("can't copy /etc/resolv.conf", err)
return 1
}
argFlyMounts, err := evaluateMounts(rfs, string(ra.Name), p)
if err != nil {
log.PrintE("can't evaluate mounts", err)
return 1
}
effectiveMounts := append(
[]flyMount{
{"", "", "/dev", "none", syscall.MS_REC | syscall.MS_SHARED},
{"/dev", rfs, "/dev", "none", syscall.MS_BIND | syscall.MS_REC},
{"", "", "/proc", "none", syscall.MS_REC | syscall.MS_SHARED},
{"/proc", rfs, "/proc", "none", syscall.MS_BIND | syscall.MS_REC},
{"", "", "/sys", "none", syscall.MS_REC | syscall.MS_SHARED},
{"/sys", rfs, "/sys", "none", syscall.MS_BIND | syscall.MS_REC},
{"tmpfs", rfs, "/tmp", "tmpfs", 0},
},
argFlyMounts...,
)
for _, mount := range effectiveMounts {
var (
err error
hostPathInfo os.FileInfo
targetPathInfo os.FileInfo
)
if strings.HasPrefix(mount.HostPath, "/") {
if hostPathInfo, err = os.Stat(mount.HostPath); err != nil {
log.PrintE(fmt.Sprintf("stat of host path %s", mount.HostPath), err)
return 1
}
} else {
hostPathInfo = nil
}
absTargetPath := filepath.Join(mount.TargetPrefixPath, mount.RelTargetPath)
if targetPathInfo, err = os.Stat(absTargetPath); err != nil && !os.IsNotExist(err) {
log.PrintE(fmt.Sprintf("stat of target path %s", absTargetPath), err)
return 1
}
switch {
case targetPathInfo == nil:
absTargetPathParent, _ := filepath.Split(absTargetPath)
if err := os.MkdirAll(absTargetPathParent, 0755); err != nil {
log.PrintE(fmt.Sprintf("can't create directory %q", absTargetPath), err)
return 1
}
switch {
case hostPathInfo == nil || hostPathInfo.IsDir():
if err := os.Mkdir(absTargetPath, 0755); err != nil {
log.PrintE(fmt.Sprintf("can't create directory %q", absTargetPath), err)
return 1
}
case !hostPathInfo.IsDir():
file, err := os.OpenFile(absTargetPath, os.O_CREATE, 0700)
if err != nil {
log.PrintE(fmt.Sprintf("can't create file %q", absTargetPath), err)
return 1
}
file.Close()
}
case hostPathInfo != nil:
switch {
case hostPathInfo.IsDir() && !targetPathInfo.IsDir():
log.Printf("can't mount because %q is a directory while %q is not", mount.HostPath, absTargetPath)
return 1
case !hostPathInfo.IsDir() && targetPathInfo.IsDir():
log.Printf("can't mount because %q is not a directory while %q is", mount.HostPath, absTargetPath)
return 1
}
}
if err := syscall.Mount(mount.HostPath, absTargetPath, mount.Fs, mount.Flags, ""); err != nil {
log.PrintE(fmt.Sprintf("can't mount %q on %q with flags %v", mount.HostPath, absTargetPath, mount.Flags), err)
return 1
}
}
if err = stage1common.WritePid(os.Getpid(), "pid"); err != nil {
log.Error(err)
return 1
}
var uidResolver, gidResolver user.Resolver
var uid, gid int
uidResolver, err = user.NumericIDs(ra.App.User)
if err != nil {
uidResolver, err = user.IDsFromStat(rfs, ra.App.User, nil)
}
if err != nil { // give up
log.PrintE(fmt.Sprintf("invalid user %q", ra.App.User), err)
return 1
}
if uid, _, err = uidResolver.IDs(); err != nil {
log.PrintE(fmt.Sprintf("failed to configure user %q", ra.App.User), err)
return 1
}
gidResolver, err = user.NumericIDs(ra.App.Group)
if err != nil {
gidResolver, err = user.IDsFromStat(rfs, ra.App.Group, nil)
}
if err != nil { // give up
log.PrintE(fmt.Sprintf("invalid group %q", ra.App.Group), err)
return 1
}
if _, gid, err = gidResolver.IDs(); err != nil {
log.PrintE(fmt.Sprintf("failed to configure group %q", ra.App.Group), err)
return 1
}
diag.Printf("chroot to %q", rfs)
if err := syscall.Chroot(rfs); err != nil {
log.PrintE("can't chroot", err)
return 1
}
if err := os.Chdir(workDir); err != nil {
log.PrintE(fmt.Sprintf("can't change to working directory %q", workDir), err)
return 1
}
// lock the current goroutine to its current OS thread.
// This will force the subsequent syscalls to be executed in the same OS thread as Setresuid, and Setresgid,
// see https://github.com/golang/go/issues/1435#issuecomment-66054163.
runtime.LockOSThread()
diag.Printf("setting uid %d gid %d", uid, gid)
if err := syscall.Setresgid(gid, gid, gid); err != nil {
log.PrintE(fmt.Sprintf("can't set gid %d", gid), err)
return 1
}
if err := syscall.Setresuid(uid, uid, uid); err != nil {
log.PrintE(fmt.Sprintf("can't set uid %d", uid), err)
return 1
}
diag.Printf("execing %q in %q", args, rfs)
err = stage1common.WithClearedCloExec(lfd, func() error {
return syscall.Exec(args[0], args, env)
})
if err != nil {
log.PrintE(fmt.Sprintf("can't execute %q", args[0]), err)
return 1
}
return 0
}
func stage1(rp *stage1commontypes.RuntimePod) int {
uuid, err := types.NewUUID(flag.Arg(0))
if err != nil {
log.Print("UUID is missing or malformed\n")
return 254
}
root := "."
p, err := stage1commontypes.LoadPod(root, uuid, rp)
if err != nil {
log.PrintE("can't load pod", err)
return 254
}
if err := p.SaveRuntime(); err != nil {
log.FatalE("failed to save runtime parameters", err)
}
// Sanity checks
if len(p.Manifest.Apps) != 1 {
log.Printf("flavor %q only supports 1 application per Pod for now", flavor)
return 254
}
ra := p.Manifest.Apps[0]
imgName := p.AppNameToImageName(ra.Name)
args := ra.App.Exec
if len(args) == 0 {
log.Printf(`image %q has an empty "exec" (try --exec=BINARY)`, imgName)
return 254
}
lfd, err := common.GetRktLockFD()
if err != nil {
log.PrintE("can't get rkt lock fd", err)
return 254
}
// set close-on-exec flag on RKT_LOCK_FD so it gets correctly closed after execution is finished
if err := sys.CloseOnExec(lfd, true); err != nil {
log.PrintE("can't set FD_CLOEXEC on rkt lock", err)
return 254
}
workDir := "/"
if ra.App.WorkingDirectory != "" {
workDir = ra.App.WorkingDirectory
}
env := []string{"PATH=/usr/local/sbin:/usr/local/bin:/usr/sbin:/usr/bin:/sbin:/bin"}
for _, e := range ra.App.Environment {
env = append(env, e.Name+"="+e.Value)
}
rfs := filepath.Join(common.AppPath(p.Root, ra.Name), "rootfs")
argFlyMounts, err := evaluateMounts(rfs, string(ra.Name), p)
if err != nil {
log.PrintE("can't evaluate mounts", err)
return 254
}
effectiveMounts := append(
[]flyMount{
{"", "", "/dev", "none", syscall.MS_REC | syscall.MS_SHARED},
{"/dev", rfs, "/dev", "none", syscall.MS_BIND | syscall.MS_REC},
{"", "", "/proc", "none", syscall.MS_REC | syscall.MS_SHARED},
{"/proc", rfs, "/proc", "none", syscall.MS_BIND | syscall.MS_REC},
{"", "", "/sys", "none", syscall.MS_REC | syscall.MS_SHARED},
{"/sys", rfs, "/sys", "none", syscall.MS_BIND | syscall.MS_REC},
{"tmpfs", rfs, "/tmp", "tmpfs", 0},
},
argFlyMounts...,
)
/* Process DNS config files
*
* /etc/resolv.conf: four modes
* 'host' - bind-mount host's file
* 'stage0' - bind-mount the file created by stage0
* 'default' - do nothing (we would respect CNI if fly had networking)
* 'none' - do nothing
*/
switch p.ResolvConfMode {
case "host":
effectiveMounts = append(effectiveMounts,
flyMount{"/etc/resolv.conf", rfs, "/etc/resolv.conf", "none", syscall.MS_BIND | syscall.MS_RDONLY})
case "stage0":
if err := copyResolv(p); err != nil {
log.PrintE("can't copy /etc/resolv.conf", err)
return 254
}
}
/*
* /etc/hosts: three modes:
* 'host' - bind-mount hosts's file
* 'stage0' - bind mount the file created by stage1
* 'default' - create a stub /etc/hosts if needed
*/
switch p.EtcHostsMode {
case "host":
effectiveMounts = append(effectiveMounts,
flyMount{"/etc/hosts", rfs, "/etc/hosts", "none", syscall.MS_BIND | syscall.MS_RDONLY})
case "stage0":
effectiveMounts = append(effectiveMounts, flyMount{
filepath.Join(common.Stage1RootfsPath(p.Root), "etc", "rkt-hosts"),
rfs,
"/etc/hosts",
"none",
syscall.MS_BIND | syscall.MS_RDONLY})
case "default":
stage2HostsPath := filepath.Join(common.AppRootfsPath(p.Root, ra.Name), "etc", "hosts")
if _, err := os.Stat(stage2HostsPath); err != nil && os.IsNotExist(err) {
fallbackHosts := []byte("127.0.0.1 localhost localdomain\n")
ioutil.WriteFile(stage2HostsPath, fallbackHosts, 0644)
}
}
for _, mount := range effectiveMounts {
diag.Printf("Processing %+v", mount)
var (
err error
hostPathInfo os.FileInfo
targetPathInfo os.FileInfo
)
if strings.HasPrefix(mount.HostPath, "/") {
if hostPathInfo, err = os.Stat(mount.HostPath); err != nil {
log.PrintE(fmt.Sprintf("stat of host path %s", mount.HostPath), err)
return 254
}
} else {
hostPathInfo = nil
}
absTargetPath := filepath.Join(mount.TargetPrefixPath, mount.RelTargetPath)
if targetPathInfo, err = os.Stat(absTargetPath); err != nil && !os.IsNotExist(err) {
log.PrintE(fmt.Sprintf("stat of target path %s", absTargetPath), err)
return 254
}
switch {
case (mount.Flags & syscall.MS_REMOUNT) != 0:
{
diag.Printf("don't attempt to create files for remount of %q", absTargetPath)
}
case targetPathInfo == nil:
absTargetPathParent, _ := filepath.Split(absTargetPath)
if err := os.MkdirAll(absTargetPathParent, 0755); err != nil {
log.PrintE(fmt.Sprintf("can't create directory %q", absTargetPath), err)
return 254
}
switch {
case hostPathInfo == nil || hostPathInfo.IsDir():
if err := os.Mkdir(absTargetPath, 0755); err != nil {
log.PrintE(fmt.Sprintf("can't create directory %q", absTargetPath), err)
return 254
}
case !hostPathInfo.IsDir():
file, err := os.OpenFile(absTargetPath, os.O_CREATE, 0700)
if err != nil {
log.PrintE(fmt.Sprintf("can't create file %q", absTargetPath), err)
return 254
}
file.Close()
}
case hostPathInfo != nil:
switch {
case hostPathInfo.IsDir() && !targetPathInfo.IsDir():
log.Printf("can't mount because %q is a directory while %q is not", mount.HostPath, absTargetPath)
return 254
case !hostPathInfo.IsDir() && targetPathInfo.IsDir():
log.Printf("can't mount because %q is not a directory while %q is", mount.HostPath, absTargetPath)
return 254
}
}
if err := syscall.Mount(mount.HostPath, absTargetPath, mount.Fs, mount.Flags, ""); err != nil {
log.PrintE(fmt.Sprintf("can't mount %q on %q with flags %v", mount.HostPath, absTargetPath, mount.Flags), err)
return 254
}
}
if err = stage1common.WritePid(os.Getpid(), "pid"); err != nil {
log.Error(err)
return 254
}
var uidResolver, gidResolver user.Resolver
var uid, gid int
uidResolver, err = user.NumericIDs(ra.App.User)
if err != nil {
uidResolver, err = user.IDsFromStat(rfs, ra.App.User, nil)
}
if err != nil { // give up
log.PrintE(fmt.Sprintf("invalid user %q", ra.App.User), err)
return 254
}
if uid, _, err = uidResolver.IDs(); err != nil {
log.PrintE(fmt.Sprintf("failed to configure user %q", ra.App.User), err)
return 254
}
gidResolver, err = user.NumericIDs(ra.App.Group)
if err != nil {
gidResolver, err = user.IDsFromStat(rfs, ra.App.Group, nil)
}
if err != nil { // give up
log.PrintE(fmt.Sprintf("invalid group %q", ra.App.Group), err)
return 254
}
if _, gid, err = gidResolver.IDs(); err != nil {
log.PrintE(fmt.Sprintf("failed to configure group %q", ra.App.Group), err)
return 254
}
diag.Printf("chroot to %q", rfs)
if err := syscall.Chroot(rfs); err != nil {
log.PrintE("can't chroot", err)
return 254
}
if err := os.Chdir(workDir); err != nil {
log.PrintE(fmt.Sprintf("can't change to working directory %q", workDir), err)
return 254
}
// lock the current goroutine to its current OS thread.
// This will force the subsequent syscalls to be executed in the same OS thread as Setresuid, and Setresgid,
// see https://github.com/golang/go/issues/1435#issuecomment-66054163.
runtime.LockOSThread()
diag.Printf("setting uid %d gid %d", uid, gid)
if err := syscall.Setresgid(gid, gid, gid); err != nil {
log.PrintE(fmt.Sprintf("can't set gid %d", gid), err)
return 254
}
if err := syscall.Setresuid(uid, uid, uid); err != nil {
log.PrintE(fmt.Sprintf("can't set uid %d", uid), err)
return 254
}
diag.Printf("execing %q in %q", args, rfs)
err = stage1common.WithClearedCloExec(lfd, func() error {
return syscall.Exec(args[0], args, env)
})
if err != nil {
log.PrintE(fmt.Sprintf("can't execute %q", args[0]), err)
return 254
}
return 0
}
func doChild(rootdir string, progfile string, plan plan, stdin_fd, stdout_fd, stderr_fd [2]int) (error, string) {
if os.Getpid() != 1 {
return errors.New("not cloned?"), ""
}
if err := syscall.Dup2(stdin_fd[0], 0); err != nil {
return err, "dup2 failed"
}
syscall.Close(stdin_fd[0])
syscall.Close(stdin_fd[1])
if err := syscall.Dup2(stdout_fd[1], 1); err != nil {
return err, "dup2 failed"
}
syscall.Close(stdout_fd[0])
syscall.Close(stdout_fd[1])
if err := syscall.Dup2(stderr_fd[1], 2); err != nil {
return err, "dup2 failed"
}
syscall.Close(stderr_fd[0])
syscall.Close(stderr_fd[1])
if _, _, err := syscall.Syscall(syscall.SYS_PRCTL, syscall.PR_SET_PDEATHSIG, uintptr(syscall.SIGKILL), 0); err != 0 {
return error(err), "PR_SET_PDEATHSIG failed"
}
if err := syscall.Sethostname([]byte("poe-sandbox")); err != nil {
return err, "sethostname failed"
}
mounts := []mountPoint{
{"none", "/", "", syscall.MS_PRIVATE | syscall.MS_REC, ""},
{rootdir, rootdir, "bind", syscall.MS_BIND | syscall.MS_REC, ""},
// { "none", rootdir + "/proc", "proc", syscall.MS_NOSUID | syscall.MS_NOEXEC | syscall.MS_NODEV, "" },
// { "none", rootdir + "/dev", "devtmpfs", syscall.MS_NOSUID | syscall.MS_NOEXEC, "" },
// { "none", rootdir + "/dev/shm", "tmpfs", syscall.MS_NOSUID | syscall.MS_NODEV, "" },
}
for _, point := range mounts {
if err := syscall.Mount(point.source, point.target, point.fstype, point.flags, point.data); err != nil {
return err, fmt.Sprintf("mount '%s' on '%s' (%s) failed", point.source, point.target, point.fstype)
}
}
if err := syscall.Chroot(rootdir); err != nil {
return err, "chroot failed"
}
if _, err := syscall.Setsid(); err != nil {
return err, "setsid failed"
}
pw, err := user.Lookup(username)
if err != nil {
return err, "getpwnam failed"
}
uid, err := strconv.Atoi(pw.Uid)
if err != nil {
return err, "atoi error"
}
gid, err := strconv.Atoi(pw.Gid)
if err != nil {
return err, "atoi error"
}
// TODO: initgroups
if err := syscall.Setresgid(gid, gid, gid); err != nil {
return err, "setresgid failed"
}
if err := syscall.Setresuid(uid, uid, uid); err != nil {
return err, "setresuid failed"
}
if err := syscall.Chdir("/tmp"); err != nil {
return err, "chdir failed"
}
// stop
os.Stdin.Read(make([]byte, 1))
// be traced
if _, _, err := syscall.Syscall(syscall.SYS_PRCTL, PR_SET_NO_NEW_PRIVS, 1, 0); err != 0 {
return error(err), "PR_SET_NO_NEW_PRIVS failed"
}
if err := SetupSeccomp(); err != nil {
return err, "seccomp fail"
}
envp := []string{
"PATH=/opt/bin:/usr/bin",
"USER="******"LOGNAME=" + username,
}
cmdl := make([]string, 0, len(plan.Compiler.Command)+len(plan.Extra)-1)
for _, arg := range plan.Compiler.Command {
if arg == "PROGRAM" {
cmdl = append(cmdl, progfile)
} else if arg == "EXTRA" {
cmdl = append(cmdl, plan.Extra...)
} else {
cmdl = append(cmdl, arg)
}
}
if err := syscall.Exec(cmdl[0], cmdl, envp); err != nil {
return err, "execve failed"
}
return errors.New("unreachable"), ""
}
func main() {
runtime.GOMAXPROCS(8)
dec := json.NewDecoder(os.Stdin)
var plan plan
if err := dec.Decode(&plan); err != nil {
panic(fmt.Sprintf("failed to parse plan: %s", err))
}
//fmt.Fprintf(os.Stderr, "plan: %+v\n", plan)
runtime.LockOSThread()
if err := syscall.Setresuid(0, 0, 0); err != nil {
poePanic(err, "setuid failed")
}
if err := InitializeSystemdBus(); err != nil {
poePanic(err, "failed to connect to systemd")
}
runtime.UnlockOSThread()
rootdir, errx := PlaygroundCreate(plan.Base, plan.Compiler.Overlay)
if errx != nil {
poePanic(errx, "playground_create failed")
}
progfile, errx := PlaygroundCopy(rootdir, plan.Source)
if errx != nil {
poePanic(errx, "playground_copy failed")
}
var stdin_fd, stdout_fd, stderr_fd [2]int
if err := syscall.Pipe2(stdin_fd[:], 0); err != nil {
poePanic(err, "pipe2 failed")
}
if err := syscall.Pipe2(stdout_fd[:], syscall.O_DIRECT); err != nil {
poePanic(err, "pipe2 failed")
}
if err := syscall.Pipe2(stderr_fd[:], syscall.O_DIRECT); err != nil {
poePanic(err, "pipe2 failed")
}
pid_, _, err := syscall.Syscall(syscall.SYS_CLONE, uintptr(syscall.SIGCHLD|syscall.CLONE_NEWIPC|syscall.CLONE_NEWNS|syscall.CLONE_NEWPID|syscall.CLONE_NEWUTS|syscall.CLONE_NEWNET), 0, 0)
pid := int(pid_)
if err != 0 {
poePanic(error(err), "clone failed")
} else if pid == 0 {
runtime.LockOSThread()
if err, msg := doChild(rootdir, progfile, plan, stdin_fd, stdout_fd, stderr_fd); err != nil {
fmt.Fprintf(os.Stderr, "%s (%s)", msg, err.Error())
os.Exit(127)
}
// unreachable
} else {
res := doParent(pid, stdin_fd, stdout_fd, stderr_fd)
cleanup()
var buf bytes.Buffer
binary.Write(&buf, binary.LittleEndian, int32(res.result))
binary.Write(&buf, binary.LittleEndian, int32(res.status))
if _, err := os.Stderr.Write(buf.Bytes()); err != nil {
poePanic(err, "stderr write failed")
}
if _, err := os.Stderr.Write([]byte(res.msg)); err != nil {
poePanic(err, "stderr write failed")
}
os.Exit(0)
// unreachable
}
}