Re: chmod changing ctime even if no mode change

[Pádraig Brady]

On 10/13/2013 01:43 PM, Bernhard Voelker wrote:
> In a new openSUSE bug [1], a user complains that chmod changes a
> file's change status timestamp although chmod did not have to change
> any of the permission bits, e.g.
> 
>   $ touch test
>   $ chmod 700 test
>   $ stat test
>   $ chmod 700 test
>   $ stat test
> 
> The effect for the user is that backup/restore takes much longer
> as backups are also triggered although it would not be necessary.
> (They use a cron job for "fixing" the permissions in a shared
> directory).
> 
> Apart from that, POSIX [2] says:
> 
>   Upon successfully changing the file mode bits of a file, the chmod
>   utility shall mark for update the last file status change timestamp
>   of the file.
> 
> I'm somehow tempted to read that as "but don't do it if nothing has
> changed". Fixing the code accordingly wouldn't be too hard ... _but_
> that would stand in contrast to commit f8e66794 [3] which (probably
> precautionarily?) did it the other way round back in 2000:
> 
>   Perform the chmod even if the
>   file mode permission bits are the same as those that should be set.
>   Omitting the chmod call would be alright with minimal 1003.1e DS17
>   ACLs, but eventually there will be other permissions in addition to
>   rwx.  E.g., add and delete for directories, and something analogous
>   to NT's take ownership permission.
> 
> Does anyone know of a real situation where omitting the actual chmod(3)
> call would cause problems?  Otherwise, I'd say that current chmod(1)
> violates POSIX.
> 
> WDYT?
> 
> [1] openSUSE bug:
>     https://bugzilla.novell.com/show_bug.cgi?id=845245
> 
> [2] POSIX spec:
>     http://pubs.opengroup.org/onlinepubs/9699919799/utilities/chmod.html
> 
> [3] commit f8e66794:
>     http://git.sv.gnu.org/cgit/coreutils.git/commit/?id=f8e66794
> 
> Have a nice day,
> Berny

This is a tricky one, previously discussed at:
http://lists.gnu.org/archive/html/bug-coreutils/2010-01/msg00303.html

Pádraig.