[PATCH] shadow_copy2: mount_point is valid only for locally mountable file-system

[Michael Adam]

On 2016-10-21 at 09:53 +0200, Michael Adam wrote:
> On 2016-10-10 at 17:11 -0700, Jeremy Allison wrote:
> > On Fri, Sep 30, 2016 at 01:31:58PM +0530, Rajesh Joseph wrote:
> > > shadow_copy2 currently assumes that all the underlying file-system are
> > > mountable and
> > > are mounted on samba server. But this may not be true for all cases, e.g.
> > > GlusterFS is
> > > accessed via libgfapi and is not mounted.
> > > 
> > > This can lead to connection failure if such file-systems are used along
> > > with shadow_copy2.
> > > The attached patch assume "/" as the default mount-point in such cases
> > > instead of failing
> > > the connect.
> > > 
> > > This may not be a complete fix because shadow_copy2_find_mount_point can
> > > return wrong
> > > mount_point if by coincidence the "connectpath" is actually present in the
> > > local file-system.
> > > 
> > > I had a brief discussion on this with Michael, and we think that it might
> > > be good if VFS layer
> > > can tell if the underlying file-system is mountable or not.
> > > 
> > > It would be great to know what others think on this problem.
> > 
> > Yeah, setting it to "/" can also cause problems here:
> > 
> > 2613         } else {
> > 2614                 config->snapshot_basepath = talloc_asprintf(config, "%s/%s",
> > 2615                                 config->mount_point, config->snapdir);
> > 2616                 if (config->snapshot_basepath == NULL) {
> > 2617                         DEBUG(0, ("talloc_asprintf() failed\n"));
> > 2618                         errno = ENOMEM;
> > 2619                         return -1;
> > 2620                 }
> > 2621         }
> > 
> > Let me think some more about what "connectpath" should be in
> > this case.
> > 
> > We have an existing VFS function SMB_VFS_CONNECTPATH(), maybe
> > this should be expanded to have a different meaning when
> > we're dealing with a completely virtual underlying FS.
> 
> Hm,
> 
> the connectpath is already set to the path relative to the
> volume by vfs_glusterfs/config. (Starting with "/" for the
> volume root.)
> 
> The problem Rajesh was pointing out was that the path
> of the share might be set to something like
> "path = /usr/local" (signifying a subdir of the gluster volume)
> which conincides to be a local path too, and might be
> a mount point. So the find-mountpoint function would
> return this as the mountpoint, which is wrong.
> In gluster's case we would always have to end up with "/".
> If there was a flag inside the vfs that indicates whether
> the filsystem is a local / mounted or mountable filesystem
> or not, then shadow_copy2_find_mount_point and possibly
> other places could behave accordingly.

Hi Jeremy,

Any results from thinking more about it? :-)

Even though the glusterfs vfs module has a configurative
workaround, I would still like to conceptually adress
this...

Cheers - Michael


> But maybe there are better options.
> For now we have the config workaround of setting
> "shadow:mountpoint = /" if glusterfs vfs-module is used.
> 
> Cheers - Michael
> 


-------------- next part --------------
A non-text attachment was scrubbed...
Name: signature.asc
Type: application/pgp-signature
Size: 163 bytes
Desc: not available
URL: <http://lists.samba.org/pipermail/samba-technical/attachments/20170411/99777c5f/signature.sig>