summaryrefslogtreecommitdiff
path: root/pkg/rootless
diff options
context:
space:
mode:
Diffstat (limited to 'pkg/rootless')
-rw-r--r--pkg/rootless/rootless_linux.c63
-rw-r--r--pkg/rootless/rootless_linux.go113
2 files changed, 50 insertions, 126 deletions
diff --git a/pkg/rootless/rootless_linux.c b/pkg/rootless/rootless_linux.c
index da52a7217..716db81dc 100644
--- a/pkg/rootless/rootless_linux.c
+++ b/pkg/rootless/rootless_linux.c
@@ -535,13 +535,41 @@ create_pause_process (const char *pause_pid_file_path, char **argv)
}
}
+static int
+open_namespace (int pid_to_join, const char *ns_file)
+{
+ char ns_path[PATH_MAX];
+ int ret;
+
+ ret = snprintf (ns_path, PATH_MAX, "/proc/%d/ns/%s", pid_to_join, ns_file);
+ if (ret == PATH_MAX)
+ {
+ fprintf (stderr, "internal error: namespace path too long\n");
+ return -1;
+ }
+
+ return open (ns_path, O_CLOEXEC | O_RDONLY);
+}
+
+static void
+join_namespace_or_die (const char *name, int ns_fd)
+{
+ if (setns (ns_fd, 0) < 0)
+ {
+ fprintf (stderr, "cannot set %s namespace\n", name);
+ _exit (EXIT_FAILURE);
+ }
+}
+
int
-reexec_userns_join (int userns, int mountns, char *pause_pid_file_path)
+reexec_userns_join (int pid_to_join, char *pause_pid_file_path)
{
char uid[16];
char gid[16];
char **argv;
int pid;
+ int mnt_ns = -1;
+ int user_ns = -1;
char *cwd = getcwd (NULL, 0);
sigset_t sigset, oldsigset;
@@ -561,14 +589,28 @@ reexec_userns_join (int userns, int mountns, char *pause_pid_file_path)
_exit (EXIT_FAILURE);
}
+ user_ns = open_namespace (pid_to_join, "user");
+ if (user_ns < 0)
+ return user_ns;
+ mnt_ns = open_namespace (pid_to_join, "mnt");
+ if (mnt_ns < 0)
+ {
+ close (user_ns);
+ return mnt_ns;
+ }
+
pid = fork ();
if (pid < 0)
fprintf (stderr, "cannot fork: %s\n", strerror (errno));
if (pid)
{
- /* We passed down these fds, close them. */
int f;
+
+ /* We passed down these fds, close them. */
+ close (user_ns);
+ close (mnt_ns);
+
for (f = 3; f < open_files_max_fd; f++)
if (open_files_set == NULL || FD_ISSET (f % FD_SETSIZE, &(open_files_set[f / FD_SETSIZE])))
close (f);
@@ -606,19 +648,10 @@ reexec_userns_join (int userns, int mountns, char *pause_pid_file_path)
_exit (EXIT_FAILURE);
}
- if (setns (userns, 0) < 0)
- {
- fprintf (stderr, "cannot setns: %s\n", strerror (errno));
- _exit (EXIT_FAILURE);
- }
- close (userns);
-
- if (mountns >= 0 && setns (mountns, 0) < 0)
- {
- fprintf (stderr, "cannot setns: %s\n", strerror (errno));
- _exit (EXIT_FAILURE);
- }
- close (mountns);
+ join_namespace_or_die ("user", user_ns);
+ join_namespace_or_die ("mnt", mnt_ns);
+ close (user_ns);
+ close (mnt_ns);
if (syscall_setresgid (0, 0, 0) < 0)
{
diff --git a/pkg/rootless/rootless_linux.go b/pkg/rootless/rootless_linux.go
index 5ddfab7ad..3de136f12 100644
--- a/pkg/rootless/rootless_linux.go
+++ b/pkg/rootless/rootless_linux.go
@@ -31,7 +31,7 @@ extern uid_t rootless_uid();
extern uid_t rootless_gid();
extern int reexec_in_user_namespace(int ready, char *pause_pid_file_path, char *file_to_read, int fd);
extern int reexec_in_user_namespace_wait(int pid, int options);
-extern int reexec_userns_join(int userns, int mountns, char *pause_pid_file_path);
+extern int reexec_userns_join(int pid, char *pause_pid_file_path);
*/
import "C"
@@ -124,91 +124,6 @@ func tryMappingTool(tool string, pid int, hostID int, mappings []idtools.IDMap)
return nil
}
-func readUserNs(path string) (string, error) {
- b := make([]byte, 256)
- _, err := unix.Readlink(path, b)
- if err != nil {
- return "", err
- }
- return string(b), nil
-}
-
-func readUserNsFd(fd uintptr) (string, error) {
- return readUserNs(fmt.Sprintf("/proc/self/fd/%d", fd))
-}
-
-func getParentUserNs(fd uintptr) (uintptr, error) {
- const nsGetParent = 0xb702
- ret, _, errno := unix.Syscall(unix.SYS_IOCTL, fd, uintptr(nsGetParent), 0)
- if errno != 0 {
- return 0, errno
- }
- return (uintptr)(unsafe.Pointer(ret)), nil
-}
-
-// getUserNSFirstChild returns an open FD for the first direct child user namespace that created the process
-// Each container creates a new user namespace where the runtime runs. The current process in the container
-// might have created new user namespaces that are child of the initial namespace we created.
-// This function finds the initial namespace created for the container that is a child of the current namespace.
-//
-// current ns
-// / \
-// TARGET -> a [other containers]
-// /
-// b
-// /
-// NS READ USING THE PID -> c
-func getUserNSFirstChild(fd uintptr) (*os.File, error) {
- currentNS, err := readUserNs("/proc/self/ns/user")
- if err != nil {
- return nil, err
- }
-
- ns, err := readUserNsFd(fd)
- if err != nil {
- return nil, errors.Wrapf(err, "cannot read user namespace")
- }
- if ns == currentNS {
- return nil, errors.New("process running in the same user namespace")
- }
-
- for {
- nextFd, err := getParentUserNs(fd)
- if err != nil {
- if err == unix.ENOTTY {
- return os.NewFile(fd, "userns child"), nil
- }
- return nil, errors.Wrapf(err, "cannot get parent user namespace")
- }
-
- ns, err = readUserNsFd(nextFd)
- if err != nil {
- return nil, errors.Wrapf(err, "cannot read user namespace")
- }
-
- if ns == currentNS {
- if err := unix.Close(int(nextFd)); err != nil {
- return nil, err
- }
-
- // Drop O_CLOEXEC for the fd.
- _, _, errno := unix.Syscall(unix.SYS_FCNTL, fd, unix.F_SETFD, 0)
- if errno != 0 {
- if err := unix.Close(int(fd)); err != nil {
- logrus.Errorf("failed to close file descriptor %d", fd)
- }
- return nil, errno
- }
-
- return os.NewFile(fd, "userns child"), nil
- }
- if err := unix.Close(int(fd)); err != nil {
- return nil, err
- }
- fd = nextFd
- }
-}
-
// joinUserAndMountNS re-exec podman in a new userNS and join the user and mount
// namespace of the specified PID without looking up its parent. Useful to join directly
// the conmon process.
@@ -220,31 +135,7 @@ func joinUserAndMountNS(pid uint, pausePid string) (bool, int, error) {
cPausePid := C.CString(pausePid)
defer C.free(unsafe.Pointer(cPausePid))
- userNS, err := os.Open(fmt.Sprintf("/proc/%d/ns/user", pid))
- if err != nil {
- return false, -1, err
- }
- defer func() {
- if err := userNS.Close(); err != nil {
- logrus.Errorf("unable to close namespace: %q", err)
- }
- }()
-
- mountNS, err := os.Open(fmt.Sprintf("/proc/%d/ns/mnt", pid))
- if err != nil {
- return false, -1, err
- }
- defer func() {
- if err := mountNS.Close(); err != nil {
- logrus.Errorf("unable to close namespace: %q", err)
- }
- }()
-
- fd, err := getUserNSFirstChild(userNS.Fd())
- if err != nil {
- return false, -1, err
- }
- pidC := C.reexec_userns_join(C.int(fd.Fd()), C.int(mountNS.Fd()), cPausePid)
+ pidC := C.reexec_userns_join(C.int(pid), cPausePid)
if int(pidC) < 0 {
return false, -1, errors.Errorf("cannot re-exec process")
}
generated by cgit v1.2.3-54-g00ecf (git 2.39.0) at 2025-10-04 06:03:19 +0000