Re: [PATCH] afs: proc cells and rootcell are writeable

[Ingo Molnar <mingo@kernel.org>]

* Geert Uytterhoeven <geert@linux-m68k.org> wrote:

> On Sun, Jan 26, 2014 at 9:25 PM, Ingo Molnar <mingo@kernel.org> wrote:
> > * Ingo Molnar <mingo@kernel.org> wrote:
> >> * Linus Torvalds <torvalds@linux-foundation.org> wrote:
> >> > On Sun, Jan 26, 2014 at 4:27 AM, David Howells <dhowells@redhat.com> wrote:
> >> > > -       p = proc_create("cells", 0, proc_afs, &afs_proc_cells_fops);
> >> > > +       p = proc_create("cells", S_IFREG | S_IRUGO | S_IWUSR, proc_afs, &afs_proc_cells_fops);
> >> > > -       p = proc_create("rootcell", 0, proc_afs, &afs_proc_rootcell_fops);
> >> > > +       p = proc_create("rootcell", S_IFREG | S_IRUGO | S_IWUSR, proc_afs, &afs_proc_rootcell_fops);
> >> >
> >> > So the S_IFREG isn't necessary.
> >> >
> >> > And quite frankly, I personally think S_IRUGO | S_IWUSR is _less_
> >> > readable than 0644. It's damn hard to parse those random letter
> >> > combinations, and at least I have to really think about it, in a way
> >> > that the octal representation does *not* make me go "I have to think
> >> > about that".
> >> >
> >> > So my personal preference would be to just see that simple 0644 in
> >> > proc_create. Hmm?
> >>
> >> Perhaps we could also generate the most common variants as:
> >>
> >>  #define PERM__rw_r__r__              0644
> >>  #define PERM__r________              0400
> >>  #define PERM__r__r__r__              0444
> >>  #define PERM__r_xr_xr_x              0555
> 
> I like it (also without the PERM prefix, cfr. Alexey's old patch).
> 
> >> or something similar, more or less matching the output of 'ls -l'?
> >
> > Another variant of this would be to do the following macro:
> >
> >         PERM(R_X, R_X, R_X)
> >         PERM(R__, R__, R__)
> >         PERM(RW_, R__, R__)
> 
> IMHO, this is again less outstanding.
> 
> > With the advantage of separating the groups better and reducing the
> > number of constants needed.
> 
> Only a limited number of combinations is in active use, right?

Correct - and in fact that kind of limitation is also a security 
feature: using patterns _outside_ of the typical, already defined 
group of permission patterns would in itself be a 'is that really 
justified?' red flag during review.

I'm fine with Alexey's shorter variant as well.

Would someone be interested in sending a real patch for it, defining a 
usable set of initial flags such as 0644, 0444, 0555 and 0600?

  comet:~/tip> for N in $(git grep -E '\.\<mode\>.*=.*0' arch/x86/ kernel/ | cut -d: -f2-); do echo $N; done | sort | grep ^0[0-7] | cut -c1-4 | uniq -c | sort -n
      1 0200
      1 0666
      5 0600
     15 0555
     16 0444
    148 0644

I'd definitely convert most of kernel/ and arch/x86/ to use them.

Thanks,

	Ingo