On Mon, Feb 29, 2016 at 09:17:24AM +0100, Andreas Gruenbacher wrote:
> Add richacl xattr handler implementing the xattr operations based on the
> get_richacl and set_richacl inode operations.
Given all the issues with Posix ACLs and selinux attributes these really
should be proper syscalls instead of abusing the xattr interface.
On Fri, Mar 11, 2016 at 06:17:35AM -0800, Christoph Hellwig wrote:
> On Mon, Feb 29, 2016 at 09:17:24AM +0100, Andreas Gruenbacher wrote:
> > Add richacl xattr handler implementing the xattr operations based on the
> > get_richacl and set_richacl inode operations.
>
> Given all the issues with Posix ACLs and selinux attributes these really
> should be proper syscalls instead of abusing the xattr interface.
What are those problems exactly?
--b.
On Fri, Mar 11, 2016 at 09:19:05AM -0500, J. Bruce Fields wrote:
> On Fri, Mar 11, 2016 at 06:17:35AM -0800, Christoph Hellwig wrote:
> > On Mon, Feb 29, 2016 at 09:17:24AM +0100, Andreas Gruenbacher wrote:
> > > Add richacl xattr handler implementing the xattr operations based on the
> > > get_richacl and set_richacl inode operations.
> >
> > Given all the issues with Posix ACLs and selinux attributes these really
> > should be proper syscalls instead of abusing the xattr interface.
>
> What are those problems exactly?
That people get confused between the attr used by the xattr syscall
interface and the attr used to store things on disk or the protocol.
This has happened every time we have non-native support, e.g. XFS, NFS,
CIFS, ntfs, etc. And it's only going to become worse.
On Tue, Mar 15, 2016 at 12:10:14AM -0700, Christoph Hellwig wrote:
> On Fri, Mar 11, 2016 at 09:19:05AM -0500, J. Bruce Fields wrote:
> > On Fri, Mar 11, 2016 at 06:17:35AM -0800, Christoph Hellwig wrote:
> > > On Mon, Feb 29, 2016 at 09:17:24AM +0100, Andreas Gruenbacher wrote:
> > > > Add richacl xattr handler implementing the xattr operations based on the
> > > > get_richacl and set_richacl inode operations.
> > >
> > > Given all the issues with Posix ACLs and selinux attributes these really
> > > should be proper syscalls instead of abusing the xattr interface.
> >
> > What are those problems exactly?
>
> That people get confused between the attr used by the xattr syscall
> interface and the attr used to store things on disk or the protocol.
> This has happened every time we have non-native support, e.g. XFS, NFS,
> CIFS, ntfs, etc. And it's only going to become worse.
How has that confusion caused problems in practice?
--b.
On Tue, Mar 15, 2016 at 05:05:26PM -0400, J. Bruce Fields wrote:
> > That people get confused between the attr used by the xattr syscall
> > interface and the attr used to store things on disk or the protocol.
> > This has happened every time we have non-native support, e.g. XFS, NFS,
> > CIFS, ntfs, etc. And it's only going to become worse.
>
> How has that confusion caused problems in practice?
We had all kinds of bugs in this area that were only slowly uncovered.
We also had all kind of privilegue escalations with (non-ACLs) xattrs
as people never grasped the way different free-form namespaces have
different permission checking.