[systemd-devel] Failure to remount / ro (Was: Re: [PATCH 2/4] Adding unmount functions to be used in shutdown)

[Michael Biebl]

2010/10/24 Gustavo Sverzut Barbieri <barbieri at profusion.mobi>:
> On Sun, Oct 24, 2010 at 2:01 PM, Michael Biebl <mbiebl at gmail.com> wrote:
>> 2010/10/24 Gustavo Sverzut Barbieri <barbieri at profusion.mobi>:
>>> On Sun, Oct 24, 2010 at 12:41 PM, Michael Biebl <mbiebl at gmail.com> wrote:
>>>> Together with fidencio I tracked down the bug which caused / *not* to
>>>> be remounted ro on shutdown.
>>>>
>>>> The relevant code is in src/umount.c
>>>>
>>>> 2010/10/7  <fidencio at profusion.mobi>:
>>>>> +int umount_all(void) {
>>>>> +        int r;
>>>>> +        LIST_HEAD(MountPoint, mp_list_head);
>>>>> +
>>>>> +        LIST_HEAD_INIT(MountPoint, mp_list_head);
>>>>> +
>>>>> +        r = mount_points_list_get(&mp_list_head);
>>>>> +        if (r < 0)
>>>>> +                goto end;
>>>>> +
>>>>> +        r = mount_points_list_umount(&mp_list_head);
>>>>> +        if (r <= 0)
>>>>> +                goto end;
>>>>
>>>> r is the number of mount points which failed to be unmounted.
>>>> We always skip / in mount_points_list_umount though,
>>>> I.e. if there is no failed unmount attempt, we will never try to remount / ro.
>>>>
>>>> Maybe we should just call mount_points_list_remount_read_only unconditionally.
>>>> or increment the failed counter in mount_points_list_umount if we skip "/"
>>>
>>> I guess either we run it unconditionally or we just call the remount
>>> of / to "ro" explicitly if r == 0 (clear I guess)
>>>
>>>
>>>
>>>> +static int mount_points_list_umount(MountPoint **mount_point_list_head) {
>>>> +        MountPoint *mp, *mp_next;
>>>> +        int failed = 0;
>>>> +
>>>> +        LIST_FOREACH_SAFE(mount_point, mp, mp_next, *mount_point_list_head) {
>>>> +                if (streq(mp->path, "/"))
>>>> +                        continue;
>>>>                          ^ add failed++ to make sure / is remounted-ro?
>>>
>>> it would be a bit misleading.
>>
>> Well, not really. Skipping / is just an optimisation as we know that
>> unmounting it would fail anyway.
>
> sure, but I guess just making it explicit on the caller side is good
> as well, instead of returning a failed state. We can also more easily
> know what to return to main loop as r == 0 -> remount / ro -> return
> nothing else to do.

If r == 0, it can also mean though, that "/" was not in
**mount_point_list_head, which makes the return code of
mount_points_list_umount ambiguous.

Also, incremeting n_failed is a very simple fix, whereas handling it
it umount_all() adds more special cases and duplication.



-- 
Why is it that all of the instruments seeking intelligent life in the
universe are pointed away from Earth?