switch-root: use MS_REC for /run, unless we are soft-rebooting

There are applications that rely on mounts under /run surviving the
switch from initrd to rootfs, so use MS_REC unless we are soft
rebooting.

Follow-up for 7c764d4

Fixes systemd#28452

src/core/main.c

@@ -1884,7 +1884,8 @@ static int do_reexecute(
         if (switch_root_dir) {
                 r = switch_root(/* new_root= */ switch_root_dir,
                                 /* old_root_after= */ NULL,
-                                /* flags= */ objective == MANAGER_SWITCH_ROOT ? SWITCH_ROOT_DESTROY_OLD_ROOT : 0);
+                                /* flags= */ (objective == MANAGER_SWITCH_ROOT ? SWITCH_ROOT_DESTROY_OLD_ROOT : 0) |
+                                             (objective == MANAGER_SOFT_REBOOT ? SWITCH_ROOT_SOFT_REBOOT : 0));
                 if (r < 0)
                         log_error_errno(r, "Failed to switch root, trying to continue: %m");
         }

src/shared/switch-root.c

@@ -30,14 +30,19 @@ int switch_root(const char *new_root,
                 const char *old_root_after,   /* path below the new root, where to place the old root after the transition; may be NULL to unmount it */
                 SwitchRootFlags flags) {
 
+        /* Stuff mounted below /run we don't save on soft reboot, as it might have lost its relevance, i.e.
+         * credentials, removable media and such, we rather want that the new boot mounts this fresh.
+         * But on the switch from initrd we do use MS_REC, as it is expected that mounts set up in /run
+         * are maintained. */
+        unsigned long run_mount_flags = MS_BIND|(!FLAGS_SET(flags, SWITCH_ROOT_SOFT_REBOOT) ? MS_REC : 0);
         struct {
                 const char *path;
                 unsigned long mount_flags;
         } transfer_table[] = {
                 { "/dev",                                 MS_BIND|MS_REC }, /* Recursive, because we want to save the original /dev/shm + /dev/pts and similar */
                 { "/sys",                                 MS_BIND|MS_REC }, /* Similar, we want to retain various API VFS, or the cgroupv1 /sys/fs/cgroup/ tree */
                 { "/proc",                                MS_BIND|MS_REC }, /* Similar */
-                { "/run",                                 MS_BIND        }, /* Stuff mounted below this we don't save, as it might have lost its relevance, i.e. credentials, removable media and such, we rather want that the new boot mounts this fresh */
+                { "/run",                                 run_mount_flags}, /* Recursive except on soft reboot, see above */
                 { SYSTEM_CREDENTIALS_DIRECTORY,           MS_BIND        }, /* Credentials passed into the system should survive */
                 { ENCRYPTED_SYSTEM_CREDENTIALS_DIRECTORY, MS_BIND        }, /* Similar */
                 { "/run/host",                            MS_BIND|MS_REC }, /* Host supplied hierarchy should also survive */

src/shared/switch-root.h

@@ -6,6 +6,7 @@
 typedef enum SwitchRootFlags {
         SWITCH_ROOT_DESTROY_OLD_ROOT = 1 << 0, /* rm -rf old root when switching – under the condition that it is backed by non-persistent tmpfs/ramfs/… */
         SWITCH_ROOT_DONT_SYNC        = 1 << 1, /* don't call sync() immediately before switching root */
+        SWITCH_ROOT_SOFT_REBOOT      = 1 << 2, /* move /run without MS_REC */
 } SwitchRootFlags;
 
 int switch_root(const char *new_root, const char *old_root_after, SwitchRootFlags flags);