[Gluster-devel] Problems about acl_get_file used in posix_pacl_get

[Niels de Vos]

On Fri, Aug 17, 2018 at 07:11:12PM +0800, Kinglong Mee wrote:
> Hi Niels,
> 
> On 2018/8/17 18:14, Niels de Vos wrote:
> > On Fri, Aug 17, 2018 at 05:22:17PM +0800, Kinglong Mee wrote:
> >> Hi Niels,
> >>
> >> On 2018/8/17 17:13, Niels de Vos wrote:
> >>> On Fri, Aug 17, 2018 at 03:04:43PM +0800, Kinglong Mee wrote:
> >>>> Hello folks,
> >>>>
> >>>> nfs-ganesha using the new gfapi named glfs_h_acl_set/glfs_h_acl_get,
> >>>> at xlator posix, glusterfsd calls acl_get_file/acl_set_file (libacl functions) to process xattrs.
> >>>>
> >>>> By default, sys_lsetxattr/sys_llistxattr/sys_lgetxattr/sys_lremovexattr are used to process xattrs.
> >>>> But, unfortunately, those two functions do syscall by getxattr/setxattr.
> >>>> I don't think that is we want.
> >>>>
> >>>> Is it a known problem ?
> >>>
> >>> There should not be a problem for libacl to use syscalls directly. The
> >>> Gluster sources use sys_<syscall> so that there can be wrappers for the
> >>> differences between OS's. In the end, these sys_<syscall> functions will
> >>> mostly call the <syscall> with (adapted) arguments.
> >>>
> >>> I do not know what problem you are facing, but I can imagine that there
> >>> is a 'getxattr' symbol in the executable image that gets called by
> >>> libacl, instead of the 'getxattr' syscall. This will likely result in
> >>> very strange behaviour, if not segfaults.
> >>
> >> Sorry for my unclear description.
> >> The real problem here is libacl gets/sets xattrs by getxattr/setxattr which follow symbolic links,
> >> but, posix xlator get/set xattrs by sys_l*xattr which do not follow symbolic links.
> > 
> > Permission checking is done by the kernel. I do not think setting ACLs
> > on a symlink makes much sense. More liberal permissions on the symlink
> > will not help with accessing the contents, and restricting permissions
> > on a symlink still give the user to access the contents through its real
> > filename.
> > 
> > Is there a reason that having ACLs on a symlink can have benefits?
> 
> Sorry, i don't know.
> 
> Md-cache supports caching GF_POSIX_ACL_ACCESS/GF_POSIX_ACL_DEFAULT right now,
> but posix_xattr_fill (call _posix_xattr_get_set) does not fill those two xattrs.
> 
> After I adds the posix_pacl_get to _posix_xattr_get_set,
> there are some problems for symlink files.
> So that, I find the different between acl_get_file and sys_llistxattr/sys_lgetxattr.

Because (Linux) filesystems do not have ACLs on symlinks, I think
md-cache should also not fetch/cache ACLs on symlinks.

Poornima, do you have an opinion about this?

Niels