[parted-devel] [PATCH] unneeded stats...

[Joel Granados]

On Mon, Feb 23, 2009 at 05:46:17PM +0100, Karel Zak wrote:
> On Mon, Feb 23, 2009 at 04:10:58PM +0000, Bryn M. Reeves wrote:
> > Joel Granados wrote:
> > > On Mon, Feb 23, 2009 at 03:50:02PM +0000, Bryn M. Reeves wrote:
> > >> Joel Granados Moreno wrote:
> > >>> Hello List
> > >>>
> > >>> Here is the reviewed patch for the unneeded stats that we are doing when
> > >>> looking for mounted devices.  It is modified to not search for "server
> > >>> strings" (//servername/path/to/somewhere)
> > >>>
> > >>> Review appreciated
> > >> There's a slight gotcha with this approach in that the kernel will  
> > >> honour the names that userspace gives it in a mount syscall. E.g. if I  
> > >> mount /dev/sda then that's what appears in /proc/mounts but if I mount  
> > >> //dev/sda, then that's the string that will appear.
> > >>
> > >> E.g.:
> > >>
> > >> # mount //dev/mapper/cyan /mnt
> > >> # grep cyan /proc/mounts
> > >> //dev/mapper/cyan /mnt ext3 rw,data=ordered 0 0
> > > 
> > > I tested for this situation in my fedora10 environment  With the latest
> > > parted.  And if I do the mount with "//" /proc/mounts will show it with
> > 
> > Yep, just confirmed that on my f10 box so this has changed somewhere 
> > between RHEL5's 2.6.18 (where I tested previously) and current upstream.
> 
>  The current mount(8) always canonicalize all paths, so it never call
>  mount(2) syscall with '//'. (RHEL5 != current:-) The more important
>  thing is kernel behaviour. You cannot rely on mount(8) -- there is
>  more ways how to mount devices.

ack.

> 
>  I think the best way how to resolve this problem is to call stat(2)
>  and ignore everything from /proc/mounts what is not S_IFBLK. It seems
>  that paths are unreliable...

Well, I still see the use for the first version of the patch, the one that did not
check for "//".  I mean, its still not relevant to check for "none" and
"tmpfs" as devices (and all strings that don't have "/" at the beginning).

Note that we are already stating and checking if they are S_IFBLK.  This
comes after we make sure that we are not looking for none paths.  So
when we find server mounts they *will* be ignored.

Regards

> 
>     Karel
> 
> -- 
>  Karel Zak  <kzak at redhat.com>

Thx for the review

-- 
Joel Andres Granados
Brno, Czech Republic, Red Hat.