SYSTEMD: Give us a option for upstart
Orcan Ogetbil
oget.fedora at gmail.com
Tue Jun 14 14:32:47 UTC 2011
On Tue, Jun 14, 2011 at 10:10 AM, Jeff Spaleta wrote:
> On Tue, Jun 14, 2011 at 5:32 AM, Reindl Harald wrote:
>> i think this would be a good idea
>>
>> PHP (my main language) is fighting with traling slash or not troubles
>> over all the years, but there is nothing to stop the boot-process and
>> systemd is a very different level of software
>
> Let's be clear...the bug was actually in mount...not systemd. And the
> fix has been committed for the mount binary according the bug ticket
> against the utils-linux package. So the problem has been solved very
> quickly after the _correct_ developers were notified via the
> established bug tracking mechanism.
>
> And its a weird bug for mount...not what I would have expected.
>
> Simple test outside of systemd for everyone falling along using a ntfs
> disk I just happened to have.
> First using an entry like this without a slash
> /dev/sdb1 /mnt ntfs defaults 0 0
>
> mount /mnt and mount /mnt/ both work and the drive mounts
>
> Second using an entry like this with a slash
> /dev/sdb1 /mnt/ ntfs defaults 0 0
>
> mount /mnt fails to mount with an error and mount /mnt/ succeeds
>
>
> What my very simple test shows is that this is totally inconsistent
> behavior on the part of mount in handling trailing slashes or the lack
> thereof. There's no good reason why that 1 failure should be
> happening especially when clearly the mount binary is internally
> manipulating trailing slashes in some cases. If it wasn't then I
> should have gotten a failure in the first case.
>
> Reindi,
> If you want to be passionate and be upset about system breakage, that
> is absolutely your right to do so. But I caution you that you are not
> channeling your passion effectively. I hope in the future you budget
> some time as a volunteer to be involved in the pre-release testing so
> you can help catch problems prior to release. I also hope you learn
> to be less aggressive when discussing issues with people with whom you
> don't have an established working relationship.
>
> And please, avoid prejudicing a new component of the software stack
> when things like this happen. New code typically does a better job at
> tickling implicit assumptions than experienced sysadmins and testers.
> In this case, mount is broken, and has been broken for years, and
> we've all been living with that brokenness and not realizing it
> because we've conditioned ourselves to interact with mount in a way
> that avoids the breakage. Please, lay the blame at the feet of the
> correct piece of software. In this case the mount binary is behaving
> inconsistently and has undocumented quirks that have gone unfixed for
> YEARS until this bug was filed and fixed. FIXED...I can't stress that
> enough...the fix has already been committed and we are just waiting
> for packages now.
>
> All systemd was doing was breaking an _undocumented_ _implicit_
> _assumption_ that the mount command was using to map mount cmdline
> mountpoints to fstab entry mountpoints. Mount was assuming that when
> an fstab entry had a trailing slash then the mount cmdline mntpoint
> argument would also have a trailing slash and mount was failing when
> the trailing slash was missing in the cmdline argument. Is there a
> good reason for mount to do this? I can't think of one so far noone
> has defended mount's behavior in this regard. And as far as I know its
> not a documented behavior of mount. And since its not documented there
> was no reason that anyone (including the systemd authors) could know
> that stripping the trailing slash when parsing the fstab entry would
> cause mount to fail. Doubly so when the slash is missing mount
> processes cmndline mountpoints with trailing slashes without issue.
>
I understand the inconsistency and it is indeed a bug in mount.
Nevertheless you are missing the point. If X worked before (X=mounting
at boot with fstab containing trailing slashes), and stops working now
because of the change Y I made, I am responsible for fixing X or Y.
The question of 'which one contains the bug' is irrelevant for the
user. Some folks think that this is a corner case and it is easy to
miss. I think that this is a fundamental mistake and this should be
one of the first things a programmer should learn. It pretty much
compares a physicist forgetting F=ma. Well, we all do mistakes.
Unfortunately, it caused problems for at least a couple people.
Hopefully the programmer learned his lesson.
Orcan
More information about the devel
mailing list