Hole punching and mmap races

[Jan Kara]

On Fri 18-05-12 20:12:10, Dave Chinner wrote:
> On Fri, May 18, 2012 at 01:28:29AM +0200, Jan Kara wrote:
> > On Thu 17-05-12 17:43:08, Dave Chinner wrote:
> > > On Wed, May 16, 2012 at 03:04:45PM +0200, Jan Kara wrote:
> > > > On Wed 16-05-12 12:14:23, Dave Chinner wrote:
> > > IIRC, it's a rare case (that I consider insane, BTW):  read from a
> > > file with into a buffer that is a mmap()d region of the same file
> > > that has not been faulted in yet.....
> >   With punch hole, the race is less insane - just punching hole in the area
> > which is accessed via mmap could race in a bad way AFAICS.
> 
> Seems the simple answer to me is to prevent page faults while hole
> punching, then....
  Yes, that's what I was suggesting in the beginning :) And I was asking
whether people are OK with another lock in the page fault path (in
particular in ->page_mkwrite) or whether someone has a better idea (e.g.
taking mmap_sem in the hole punching path seems possible but I'm not sure
whether that would be considered acceptable abuse).

								Honza
-- 
Jan Kara <jack at suse.cz>
SUSE Labs, CR