Re: [PATCH 2/2][RFC][BUG] msync: updating ctime and mtime at syncing

[Peter Zijlstra <a.p.zijlstra@chello.nl>]

On Sat, 2008-01-12 at 15:38 +0300, Anton Salikhmetov wrote:
> 2008/1/12, Peter Zijlstra <a.p.zijlstra@chello.nl>:
> >
> > On Sat, 2008-01-12 at 10:36 +0100, Peter Zijlstra wrote:
> > > On Fri, 2008-01-11 at 03:44 +0300, Anton Salikhmetov wrote:
> > >
> > > > +/*
> > > > + * Update the ctime and mtime stamps after checking if they are to be updated.
> > > > + */
> > > > +void mapped_file_update_time(struct file *file)
> > > > +{
> > > > +   if (test_and_clear_bit(AS_MCTIME, &file->f_mapping->flags)) {
> > > > +           get_file(file);
> > > > +           file_update_time(file);
> > > > +           fput(file);
> > > > +   }
> > > > +}
> > > > +
> > >
> > > I don't think you need the get/put file stuff here, because
> >
> > BTW, the reason for me noticing this is that if it would be needed there
> > is a race condition right there, who is to say that the file pointer
> > you're deref'ing in your test condition isn't a dead one already.
> 
> So, in your opinion, is it at all needed here to play with the file reference
> counter? May I drop the get_file() and fput() calls from the
> sys_msync() function?

No, the ones in sys_msync() around calling do_fsync() are most
definately needed because we release mmap_sem there.

What I'm saying is that you can remove the get_file()/fput() calls from
your new mapped_file_update_time() function.