Hi Martin, thanks for the review!
Am 06.01.2015 um 08:23 schrieb Martin Pitt:
> Michael Biebl [2015-01-06 1:34 +0100]:
>> +static bool is_mounted(const char *dev_path) {
>> + _cleanup_endmntent_ FILE *mtab = NULL;
>> + struct mntent *me = NULL;
>> +
>> + mtab = setmntent("/etc/mtab", "r");
>
> Can we rely on /etc/mtab at this point? Should we rather look at
> /proc/mounts here?
/proc is an API fs and mounted pretty early during boot by systemd, so
/proc/mounts is available when the generators run. /etc/mtab should be a
symlink to /proc/mounts nowadays so it *should* be available.
But you are right, there might be cases when /etc/fstab is still a file
and possibly out-of-date, so using /proc/mounts might be safer.
I thought about that as well, but...
> Also, I think there is a simpler way to check if a dir is a
> mountpoint: s1 = stat(dir), s2 = stat(parent(dir))
> (i. e. dir + "/.."), and check
>
> is_mount = (st.st_dev != st2.st_dev) ||
> (st.st_dev == st2.st_dev && st.st_ino == st2.st_ino);
.. this seems simpler indeed. So let's just use that. Thanks for the hint!
> That's the approach that /bin/mountpoint uses, and it avoids relying
> on having /etc/mtab, /proc mounted, and doing the iteration.
>
> The first check is obvious; I think the second check is only needed
> for bind mounts, which we could probably even ignore here (but then
> again it doesn't hurt to keep it).
Nod.
>> @@ -83,6 +103,11 @@ int main(int argc, char *argv[]) {
>> !path_equal(me->mnt_dir, "/usr"))
>> continue;
>>
>> + /* Don't remount /usr if it hasn't been mounted by the
>> initrd */
>> + if (path_equal(me->mnt_dir, "/usr") &&
>> + !is_mounted("/usr"))
>> + continue;
>> +
>
> That looks like a good approach to me indeed.
>
Will push an updated patch including your suggestion.
Michael
--
Why is it that all of the instruments seeking intelligent life in the
universe are pointed away from Earth?
signature.asc
Description: OpenPGP digital signature
