src/shared/switch-root.c

   1 /* SPDX-License-Identifier: LGPL-2.1+ */
   2 /***
   3   This file is part of systemd.
   4
   5   Copyright 2012 Harald Hoyer, Lennart Poettering
   6 ***/
   7
   8 #include <errno.h>
   9 #include <fcntl.h>
  10 #include <limits.h>
  11 #include <stdbool.h>
  12 #include <stdio.h>
  13 #include <sys/mount.h>
  14 #include <sys/stat.h>
  15 #include <unistd.h>
  16
  17 #include "base-filesystem.h"
  18 #include "fd-util.h"
  19 #include "fs-util.h"
  20 #include "log.h"
  21 #include "missing.h"
  22 #include "mkdir.h"
  23 #include "mount-util.h"
  24 #include "path-util.h"
  25 #include "rm-rf.h"
  26 #include "stdio-util.h"
  27 #include "string-util.h"
  28 #include "strv.h"
  29 #include "switch-root.h"
  30 #include "user-util.h"
  31 #include "util.h"
  32
  33 int switch_root(const char *new_root,
  34                 const char *old_root_after, /* path below the new root, where to place the old root after the transition */
  35                 bool unmount_old_root,
  36                 unsigned long mount_flags) {  /* MS_MOVE or MS_BIND */
  37
  38         _cleanup_free_ char *resolved_old_root_after = NULL;
  39         _cleanup_close_ int old_root_fd = -1;
  40         bool old_root_remove;
  41         const char *i;
  42         int r;
  43
  44         assert(new_root);
  45         assert(old_root_after);
  46
  47         if (path_equal(new_root, "/"))
  48                 return 0;
  49
  50         /* Check if we shall remove the contents of the old root */
  51         old_root_remove = in_initrd();
  52         if (old_root_remove) {
  53                 old_root_fd = open("/", O_RDONLY|O_NONBLOCK|O_CLOEXEC|O_NOCTTY|O_DIRECTORY);
  54                 if (old_root_fd < 0)
  55                         return log_error_errno(errno, "Failed to open root directory: %m");
  56         }
  57
  58         /* Determine where we shall place the old root after the transition */
  59         r = chase_symlinks(old_root_after, new_root, CHASE_PREFIX_ROOT|CHASE_NONEXISTENT, &resolved_old_root_after);
  60         if (r < 0)
  61                 return log_error_errno(r, "Failed to resolve %s/%s: %m", new_root, old_root_after);
  62         if (r == 0) /* Doesn't exist yet. Let's create it */
  63                 (void) mkdir_p_label(resolved_old_root_after, 0755);
  64
  65         /* Work-around for kernel design: the kernel refuses MS_MOVE if any file systems are mounted MS_SHARED. Hence
  66          * remount them MS_PRIVATE here as a work-around.
  67          *
  68          * https://bugzilla.redhat.com/show_bug.cgi?id=847418 */
  69         if (mount(NULL, "/", NULL, MS_REC|MS_PRIVATE, NULL) < 0)
  70                 return log_error_errno(errno, "Failed to set \"/\" mount propagation to private: %m");
  71
  72         FOREACH_STRING(i, "/sys", "/dev", "/run", "/proc") {
  73                 _cleanup_free_ char *chased = NULL;
  74
  75                 r = chase_symlinks(i, new_root, CHASE_PREFIX_ROOT|CHASE_NONEXISTENT, &chased);
  76                 if (r < 0)
  77                         return log_error_errno(r, "Failed to resolve %s/%s: %m", new_root, i);
  78                 if (r > 0) {
  79                         /* Already exists. Let's see if it is a mount point already. */
  80                         r = path_is_mount_point(chased, NULL, 0);
  81                         if (r < 0)
  82                                 return log_error_errno(r, "Failed to determine whether %s is a mount point: %m", chased);
  83                         if (r > 0) /* If it is already mounted, then do nothing */
  84                                 continue;
  85                 } else
  86                          /* Doesn't exist yet? */
  87                         (void) mkdir_p_label(chased, 0755);
  88
  89                 if (mount(i, chased, NULL, mount_flags, NULL) < 0)
  90                         return log_error_errno(r, "Failed to mount %s to %s: %m", i, chased);
  91         }
  92
  93         /* Do not fail if base_filesystem_create() fails. Not all switch roots are like base_filesystem_create() wants
  94          * them to look like. They might even boot, if they are RO and don't have the FS layout. Just ignore the error
  95          * and switch_root() nevertheless. */
  96         (void) base_filesystem_create(new_root, UID_INVALID, GID_INVALID);
  97
  98         if (chdir(new_root) < 0)
  99                 return log_error_errno(errno, "Failed to change directory to %s: %m", new_root);
 100
 101         /* We first try a pivot_root() so that we can umount the old root dir. In many cases (i.e. where rootfs is /),
 102          * that's not possible however, and hence we simply overmount root */
 103         if (pivot_root(new_root, resolved_old_root_after) >= 0) {
 104
 105                 /* Immediately get rid of the old root, if detach_oldroot is set.
 106                  * Since we are running off it we need to do this lazily. */
 107                 if (unmount_old_root) {
 108                         r = umount_recursive(old_root_after, MNT_DETACH);
 109                         if (r < 0)
 110                                 log_warning_errno(r, "Failed to unmount old root directory tree, ignoring: %m");
 111                 }
 112
 113         } else if (mount(new_root, "/", NULL, MS_MOVE, NULL) < 0)
 114                 return log_error_errno(errno, "Failed to move %s to /: %m", new_root);
 115
 116         if (chroot(".") < 0)
 117                 return log_error_errno(errno, "Failed to change root: %m");
 118
 119         if (chdir("/") < 0)
 120                 return log_error_errno(errno, "Failed to change directory: %m");
 121
 122         if (old_root_fd >= 0) {
 123                 struct stat rb;
 124
 125                 if (fstat(old_root_fd, &rb) < 0)
 126                         log_warning_errno(errno, "Failed to stat old root directory, leaving: %m");
 127                 else {
 128                         (void) rm_rf_children(old_root_fd, 0, &rb);
 129                         old_root_fd = -1;
 130                 }
 131         }
 132
 133         return 0;
 134 }