Re: [PATCH v3] SUNRPC: set desired file system root before connecting local transports

["J. Bruce Fields" <bfields@fieldses.org>]

On Tue, Nov 06, 2012 at 07:40:35AM -0500, Christoph Hellwig wrote:
> On Tue, Nov 06, 2012 at 07:06:42AM -0500, J. Bruce Fields wrote:
> > On Tue, Nov 06, 2012 at 02:14:50PM +0400, Stanislav Kinsbursky wrote:
> > > 09.10.2012 23:35, J. Bruce Fields ??????????:
> > > >Cc'ing Eric since I seem to recall he suggested doing it this way?
> > > >
> > > >Seems OK to me, but maybe that swap_root should be in common code?  (Or
> > > >maybe we could use set_fs_root()?)
> > > >
> > > 
> > > This patch is not good since, as Eric mentioned, all kernel threads
> > > share same fs struct.
> > > We can swap whole fs struct. Or we can unshare fs struct
> > > (unshare_fs_struct() is exported) and swap root in this case.
> > > But this approach is to close to set_fs_root() logic, which is not
> > > exported and seems there are some valid reasons for it.
> > 
> > What are those reasons?
> > 
> > Googling found one previous thread:
> > 
> > 	http://thread.gmane.org/gmane.linux.kernel/1259986/focus=47687
> > 
> > There Trond requests an ACK from Al or Cristoph for the export, but I
> > don't see either an ACK or any objection.
> 
> I really don't think messing with current->fs for workqueue worker
> threads is a good idea, as the worker threads are shared by different
> workqueues and thus this can easily cause havoc for entirely unrelated
> subsystems.

So you're worried that a bug in the nfs code could modify the root and
then not restore it?

--b.

> 
> To do this properly you'll need to avoid current->fs references in the
> sunrpc code.
> 
> And just in case it wasn't clear: the hack in this iteration is even
> worse than the original.