rootless: join namespace immediately when possible

add a shortcut for joining immediately the namespace so we don't need
to re-exec Podman.

With the pause process simplificaton, we can now attempt to join the
namespaces as soon as Podman starts (and before the Go runtime kicks
in), so that we don't need to re-exec and use just one process.

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

1 files changed, 162 insertions, 10 deletions

diff --git a/pkg/rootless/rootless_linux.c b/pkg/rootless/rootless_linux.c
index b87deb86e..a08cfd36a 100644
--- a/pkg/rootless/rootless_linux.c
+++ b/pkg/rootless/rootless_linux.c
@@ -39,6 +39,7 @@ static const char *_unprivileged_user_namespaces = "/proc/sys/kernel/unprivilege
 
 static int open_files_max_fd;
 fd_set open_files_set;
+static uid_t rootless_uid_init;
 
 static int
 syscall_setresuid (uid_t ruid, uid_t euid, uid_t suid)
@@ -52,6 +53,12 @@ syscall_setresgid (gid_t rgid, gid_t egid, gid_t sgid)
   return (int) syscall (__NR_setresgid, rgid, egid, sgid);
 }
 
+uid_t
+rootless_uid ()
+{
+  return rootless_uid_init;
+}
+
 static void
 do_pause ()
 {
@@ -72,7 +79,10 @@ get_cmd_line_args (pid_t pid)
   int i, argc = 0;
   char **argv;
 
-  sprintf (path, "/proc/%d/cmdline", pid);
+  if (pid)
+    sprintf (path, "/proc/%d/cmdline", pid);
+  else
+    strcpy (path, "/proc/self/cmdline");
   fd = open (path, O_RDONLY);
   if (fd < 0)
     return NULL;
@@ -87,7 +97,10 @@ get_cmd_line_args (pid_t pid)
         ret = read (fd, buffer + used, allocated - used);
       while (ret < 0 && errno == EINTR);
       if (ret < 0)
-        return NULL;
+        {
+          free (buffer);
+          return NULL;
+        }
 
       if (ret == 0)
         break;
@@ -97,11 +110,12 @@ get_cmd_line_args (pid_t pid)
         {
           allocated += 512;
           char *tmp = realloc (buffer, allocated);
-          if (buffer == NULL) {
-		  free(buffer);
-		  return NULL;
-	  }
-	  buffer=tmp;
+          if (buffer == NULL)
+            {
+              free (buffer);
+              return NULL;
+            }
+	  buffer = tmp;
         }
     }
   close (fd);
@@ -110,11 +124,17 @@ get_cmd_line_args (pid_t pid)
     if (buffer[i] == '\0')
       argc++;
   if (argc == 0)
-    return NULL;
+    {
+      free (buffer);
+      return NULL;
+    }
 
   argv = malloc (sizeof (char *) * (argc + 1));
   if (argv == NULL)
-    return NULL;
+    {
+      free (buffer);
+      return NULL;
+    }
   argc = 0;
 
   argv[argc++] = buffer;
@@ -127,6 +147,40 @@ get_cmd_line_args (pid_t pid)
   return argv;
 }
 
+static bool
+can_use_shortcut ()
+{
+  int argc;
+  char **argv;
+  bool ret = true;
+
+#ifdef DISABLE_JOIN_SHORTCUT
+  return false;
+#endif
+
+  argv = get_cmd_line_args (0);
+  if (argv == NULL)
+    return NULL;
+
+  for (argc = 0; argv[argc]; argc++)
+    {
+      if (argc == 0 || argv[argc][0] == '-')
+        continue;
+
+      if (strcmp (argv[argc], "mount") == 0
+          || strcmp (argv[argc], "search") == 0
+          || strcmp (argv[argc], "system") == 0)
+        {
+          ret = false;
+          break;
+        }
+    }
+
+  free (argv[0]);
+  free (argv);
+  return ret;
+}
+
 static void __attribute__((constructor)) init()
 {
   const char *xdg_runtime_dir;
@@ -159,6 +213,104 @@ static void __attribute__((constructor)) init()
         }
       closedir (d);
     }
+
+  /* Shortcut.  If we are able to join the pause pid file, do it now so we don't
+     need to re-exec.  */
+  xdg_runtime_dir = getenv ("XDG_RUNTIME_DIR");
+  if (xdg_runtime_dir && xdg_runtime_dir[0] && can_use_shortcut ())
+    {
+      int r;
+      int fd;
+      long pid;
+      char buf[12];
+      uid_t uid;
+      char path[PATH_MAX];
+      const char *const suffix = "/libpod/pause.pid";
+      char *cwd = getcwd (NULL, 0);
+
+      if (cwd == NULL)
+        {
+          fprintf (stderr, "error getting current working directory: %s\n", strerror (errno));
+          _exit (EXIT_FAILURE);
+        }
+
+      if (strlen (xdg_runtime_dir) >= PATH_MAX - strlen (suffix))
+        {
+          fprintf (stderr, "invalid value for XDG_RUNTIME_DIR: %s", strerror (ENAMETOOLONG));
+          exit (EXIT_FAILURE);
+        }
+
+      sprintf (path, "%s%s", xdg_runtime_dir, suffix);
+      fd = open (path, O_RDONLY);
+      if (fd < 0)
+        {
+          free (cwd);
+          return;
+        }
+
+      r = read (fd, buf, sizeof (buf));
+      close (fd);
+      if (r < 0)
+        {
+          free (cwd);
+          return;
+        }
+      pid = strtol (buf, NULL, 10);
+      if (pid == LONG_MAX)
+        {
+          free (cwd);
+          return;
+        }
+
+      uid = geteuid ();
+
+      sprintf (path, "/proc/%d/ns/user", pid);
+      fd = open (path, O_RDONLY);
+      if (fd < 0 || setns (fd, 0) < 0)
+        {
+          free (cwd);
+          return;
+        }
+      close (fd);
+
+      /* Errors here cannot be ignored as we already joined a ns.  */
+      sprintf (path, "/proc/%d/ns/mnt", pid);
+      fd = open (path, O_RDONLY);
+      if (fd < 0)
+        {
+          fprintf (stderr, "cannot open %s: %s", path, strerror (errno));
+          exit (EXIT_FAILURE);
+        }
+
+      r = setns (fd, 0);
+      if (r < 0)
+        {
+          fprintf (stderr, "cannot join mount namespace for %d: %s", pid, strerror (errno));
+          exit (EXIT_FAILURE);
+        }
+      close (fd);
+
+      if (syscall_setresgid (0, 0, 0) < 0)
+        {
+          fprintf (stderr, "cannot setresgid: %s\n", strerror (errno));
+          _exit (EXIT_FAILURE);
+        }
+
+      if (syscall_setresuid (0, 0, 0) < 0)
+        {
+          fprintf (stderr, "cannot setresuid: %s\n", strerror (errno));
+          _exit (EXIT_FAILURE);
+        }
+
+      if (chdir (cwd) < 0)
+        {
+          fprintf (stderr, "cannot chdir: %s\n", strerror (errno));
+          _exit (EXIT_FAILURE);
+        }
+
+      free (cwd);
+      rootless_uid_init = uid;
+    }
 }
 
 static int
@@ -344,7 +496,7 @@ reexec_userns_join (int userns, int mountns, char *pause_pid_file_path)
       fprintf (stderr, "cannot setns: %s\n", strerror (errno));
       _exit (EXIT_FAILURE);
     }
-  close (userns);
+  close (mountns);
 
   if (syscall_setresgid (0, 0, 0) < 0)
     {