Re: mount.nfs libmount support (v3)

[Karel Zak <kzak@redhat.com>]

On Mon, Mar 21, 2011 at 03:04:10PM -0400, Steve Dickson wrote:
> On 03/14/2011 09:58 AM, Karel Zak wrote:
> > This is the third version of the mount.nfs with libmount support. The code
> > depends on util-linux >= v2.19 (e.g. Fedora 15).
> > 
> > v3:
> >   - call umount_error() on mount(2) error
> > 
> > v2:
> >   - move nfs_umount_do_umnt() to network.c
> >   - use union nfs_sockaddr
> >   - add missing GPL address paragraph
> >   - use abstract function the storing and retrieval mount options
> >   - use "static const" for struct option
> >   - use "strncmp() == 0" everywhere
> > 
> > 
> Looking at unmounting a non-existent fs I get:
> 
> # umount.nfs /mnt/home
> umount.nfs: remote share not in 'host:dir' format
> umount.nfs: /mnt/home: not mounted

 hmm..

> The message is being generated because nfs_umount23() is
> is passed an invalid devname ('/mnt/home' in my case).
> The reason the message was not being generated before
> is because nfs_umount23() was not called when the devname
> did not exist in the mtab.

 right

> I also notice that the libmount code does indeed  know the devname 
> is not in the mtab. The lookup_umount_fs() call fails which is in 
> the mnt_context_prepare_umount() call... But, rightly so, you continue 
> to process the umount. Plus the umount(2) is done during the 
> mnt_context_do_umount() call which ends up generating the correct
> "not mounted" message.
> 
> So my question is, is there some why I can tell, using the libmount
> API, that the fs was not in either /proc/mounts or /etc/mtab?

 That's simple, entries from mtab or /proc have always defined some
 mount options (at least rw/ro). This heuristic is already used in
 umount_main()

      if (!mnt_context_is_lazy(cxt)) {
          if (opts) {
             ....
          } else
              /* strange, no entry in mtab or /proc not mounted */
              nfs_umount23(spec, "tcp,v3");
      }

 If I good understand the 'else nfs_umount23(spec, "tcp,v3")' should be
 removed from the code. If I remove this line, then:

   # umount.nfs /mnt/home
   umount.nfs: /mnt/home: not found


 There is also mnt_fs_is_kernel() to check that the FS description is
 from /proc, maybe I can also add mnt_fs_is_mtab() to the API.

> Basically I do not want to call nfs_umount23() but I do want
> to call mnt_context_do_umount() when the file system is not
> in the mtab, similar to how it worked in the previous code.

 Yep, the "umount.nfs server://foo" should be evaluated and converted
 to the valid FS description by libmount (mnt_context_prepare_umount),
 any other fallback solution is unnecessary.

 It seems we need v4 of the patch... 
 
 Thanks for review!

    Karel


-- 
 Karel Zak  <kzak@redhat.com>
 http://karelzak.blogspot.com