rootless: add management for the userNS

When running podman as non root user always create an userNS and let
the OCI runtime use it.

Signed-off-by: Giuseppe Scrivano <gscrivan@redhat.com>

Closes: #936
Approved by: rhatdan

2 files changed, 273 insertions, 0 deletions

diff --git a/pkg/rootless/rootless.go b/pkg/rootless/rootless.go
new file mode 100644
index 000000000..737fc91c7
--- /dev/null
+++ b/pkg/rootless/rootless.go
@@ -0,0 +1,145 @@
+package rootless
+
+import (
+	"fmt"
+	"io/ioutil"
+	"os"
+	"os/exec"
+	gosignal "os/signal"
+	"runtime"
+	"syscall"
+
+	"github.com/containers/storage/pkg/idtools"
+	"github.com/docker/docker/pkg/signal"
+	"github.com/pkg/errors"
+)
+
+/*
+extern int reexec_in_user_namespace(int ready);
+extern int reexec_in_user_namespace_wait(int pid);
+*/
+import "C"
+
+func runInUser() error {
+	os.Setenv("_LIBPOD_USERNS_CONFIGURED", "done")
+	return nil
+}
+
+// IsRootless tells us if we are running in rootless mode
+func IsRootless() bool {
+	return os.Getuid() != 0 || os.Getenv("_LIBPOD_USERNS_CONFIGURED") != ""
+}
+
+func tryMappingTool(tool string, pid int, hostID int, mappings []idtools.IDMap) error {
+	path, err := exec.LookPath(tool)
+	if err != nil {
+		return err
+	}
+
+	appendTriplet := func(l []string, a, b, c int) []string {
+		return append(l, fmt.Sprintf("%d", a), fmt.Sprintf("%d", b), fmt.Sprintf("%d", c))
+	}
+
+	args := []string{path, fmt.Sprintf("%d", pid)}
+	args = appendTriplet(args, 0, hostID, 1)
+	if mappings != nil {
+		for _, i := range mappings {
+			args = appendTriplet(args, i.ContainerID+1, i.HostID, i.Size)
+		}
+	}
+	cmd := exec.Cmd{
+		Path: path,
+		Args: args,
+	}
+	return cmd.Run()
+}
+
+// BecomeRootInUserNS re-exec podman in a new userNS
+func BecomeRootInUserNS() (bool, error) {
+
+	if os.Getuid() == 0 || os.Getenv("_LIBPOD_USERNS_CONFIGURED") != "" {
+		if os.Getenv("_LIBPOD_USERNS_CONFIGURED") == "init" {
+			return false, runInUser()
+		}
+		return false, nil
+	}
+
+	runtime.LockOSThread()
+	defer runtime.UnlockOSThread()
+
+	r, w, err := os.Pipe()
+	if err != nil {
+		return false, err
+	}
+	defer r.Close()
+	defer w.Close()
+
+	pidC := C.reexec_in_user_namespace(C.int(r.Fd()))
+	pid := int(pidC)
+	if pid < 0 {
+		return false, errors.Errorf("cannot re-exec process")
+	}
+
+	setgroups := fmt.Sprintf("/proc/%d/setgroups", pid)
+	err = ioutil.WriteFile(setgroups, []byte("deny\n"), 0666)
+	if err != nil {
+		return false, errors.Wrapf(err, "cannot write setgroups file")
+	}
+
+	var uids, gids []idtools.IDMap
+	username := os.Getenv("USER")
+	mappings, err := idtools.NewIDMappings(username, username)
+	if err == nil {
+		uids = mappings.UIDs()
+		gids = mappings.GIDs()
+	}
+
+	uidsMapped := false
+	if mappings != nil && uids != nil {
+		uidsMapped = tryMappingTool("newuidmap", pid, os.Getuid(), uids) == nil
+	}
+	if !uidsMapped {
+		uidMap := fmt.Sprintf("/proc/%d/uid_map", pid)
+		err = ioutil.WriteFile(uidMap, []byte(fmt.Sprintf("%d %d 1\n", 0, os.Getuid())), 0666)
+		if err != nil {
+			return false, errors.Wrapf(err, "cannot write uid_map")
+		}
+	}
+
+	gidsMapped := false
+	if mappings != nil && gids != nil {
+		gidsMapped = tryMappingTool("newgidmap", pid, os.Getgid(), gids) == nil
+	}
+	if !gidsMapped {
+		gidMap := fmt.Sprintf("/proc/%d/gid_map", pid)
+		err = ioutil.WriteFile(gidMap, []byte(fmt.Sprintf("%d %d 1\n", 0, os.Getgid())), 0666)
+		if err != nil {
+			return false, errors.Wrapf(err, "cannot write gid_map")
+		}
+	}
+
+	_, err = w.Write([]byte("1"))
+	if err != nil {
+		return false, errors.Wrapf(err, "write to sync pipe")
+	}
+
+	c := make(chan os.Signal, 1)
+
+	gosignal.Notify(c)
+	defer gosignal.Reset()
+	go func() {
+		for s := range c {
+			if s == signal.SIGCHLD || s == signal.SIGPIPE {
+				continue
+			}
+
+			syscall.Kill(int(pidC), s.(syscall.Signal))
+		}
+	}()
+
+	if C.reexec_in_user_namespace_wait(pidC) < 0 {
+		return false, errors.Wrapf(err, "error waiting for the re-exec process")
+	}
+
+	return true, nil
+}
diff --git a/pkg/rootless/rootless_linux.c b/pkg/rootless/rootless_linux.c
new file mode 100644
index 000000000..f107af7c7
--- /dev/null
+++ b/pkg/rootless/rootless_linux.c
@@ -0,0 +1,128 @@
+#define _GNU_SOURCE
+#include <sched.h>
+#include <stdio.h>
+#include <unistd.h>
+#include <sys/syscall.h>
+#include <stdlib.h>
+#include <errno.h>
+#include <sys/stat.h>
+#include <limits.h>
+#include <sys/types.h>
+#include <signal.h>
+#include <fcntl.h>
+#include <sys/wait.h>
+
+static int
+syscall_clone (unsigned long flags, void *child_stack)
+{
+  return (int) syscall (__NR_clone, flags, child_stack);
+}
+
+static char **
+get_cmd_line_args (pid_t pid)
+{
+  int fd;
+  char path[PATH_MAX];
+  char *buffer;
+  size_t allocated;
+  size_t used = 0;
+  int ret;
+  int i, argc = 0;
+  char **argv;
+
+  sprintf (path, "/proc/%d/cmdline", pid);
+  fd = open (path, O_RDONLY);
+  if (fd < 0)
+    return NULL;
+
+  allocated = 512;
+  buffer = malloc (allocated);
+  if (buffer == NULL)
+    return NULL;
+  for (;;)
+    {
+      do
+        ret = read (fd, buffer + used, allocated - used);
+      while (ret < 0 && errno == EINTR);
+      if (ret < 0)
+        return NULL;
+
+      if (ret == 0)
+        break;
+
+      used += ret;
+      if (allocated == used)
+        {
+          allocated += 512;
+          buffer = realloc (buffer, allocated);
+          if (buffer == NULL)
+            return NULL;
+        }
+    }
+  close (fd);
+
+  for (i = 0; i < used; i++)
+    if (buffer[i] == '\0')
+      argc++;
+
+  argv = malloc (sizeof (char *) * (argc + 1));
+  argc = 0;
+
+  argv[argc++] = buffer;
+  for (i = 0; i < used - 1; i++)
+    if (buffer[i] == '\0')
+      argv[argc++] = buffer + i + 1;
+
+  argv[argc] = NULL;
+
+  return argv;
+}
+
+int
+reexec_in_user_namespace(int ready)
+{
+  int ret;
+  pid_t pid;
+  char b;
+  pid_t ppid = getpid ();
+  char **argv;
+
+  pid = syscall_clone (CLONE_NEWUSER|SIGCHLD, NULL);
+  if (pid)
+    return pid;
+
+  argv = get_cmd_line_args (ppid);
+
+  setenv ("_LIBPOD_USERNS_CONFIGURED", "init", 1);
+
+  do
+    ret = read (ready, &b, 1) < 0;
+  while (ret < 0 && errno == EINTR);
+  if (ret < 0)
+    _exit (1);
+  close (ready);
+
+  execv (argv[0], argv);
+
+  _exit (1);
+}
+
+int
+reexec_in_user_namespace_wait (int pid)
+{
+  pid_t p;
+  int status;
+
+  do
+    p = waitpid (pid, &status, 0);
+  while (p < 0 && errno == EINTR);
+
+  if (p < 0)
+    return -1;
+
+  if (WIFEXITED (status))
+    return WEXITSTATUS (status);
+  if (WIFSIGNALED (status))
+    return 128 + WTERMSIG (status);
+  return -1;
+}