On Wed, Jun 19, 2013 at 10:29:22AM -0300, Carlos Maiolino wrote:
> Hi Dave,
>
> > > @@ -594,9 +594,10 @@ xfs_setattr_nonsize(
> > > * The set-user-ID and set-group-ID bits of a file will be
> > > * cleared upon successful return from chown()
> > > */
> > > - if ((ip->i_d.di_mode & (S_ISUID|S_ISGID)) &&
> > > - !capable(CAP_FSETID))
> > > - ip->i_d.di_mode &= ~(S_ISUID|S_ISGID);
> > > + if (!S_ISDIR(inode->i_mode))
> > > + if ((ip->i_d.di_mode & (S_ISUID|S_ISGID)) &&
> > > + !capable(CAP_FSETID))
> > > + ip->i_d.di_mode &= ~(S_ISUID|S_ISGID);
> >
> > I'm not sure I understand why this is part of this patch - the ACL
> > path does not enter this code branch (ATTR_UID/GID) so it doesn't
> > affect ACL inheritence. So this is some other behavioural change?
> >
> My apologies to have not commented it.
>
> During my code surfing to understand the problem, and what places we revoked
> sgid, I found this one, and, based on chmod specifications, we should keep
> sgid
> on the directory while chmoding it, unless the user explicitly ask for sgid
> removal, otherwise, if chmoding a file, we remove sgid if this isn't specified
> in the new mode. So, I've added a check here to ensure the inode isn't a dir
> before remove the sgid bit.
Does notify_change() or inode_change_ok() handle this appropriately?
i.e. do we even need that code there?
> Should I remove it from the patch?
It's unrelated to the ACL problem, so put it in a separate patch
with it's own commit description ;)
Cheers,
Dave.
--
Dave Chinner
david@xxxxxxxxxxxxx
|