Re: [PATCH][RFC] splice support

[Jens Axboe <axboe@suse.de>]

On Thu, Mar 30 2006, Andrew Morton wrote:
> Jens Axboe <axboe@suse.de> wrote:
> >
> > > The one-at-a-time logic looks OK from a quick scan.  Do we have logic in
> >  > there to check that we're not overrunning i_size?  (See the pain
> >  > do_generic_mapping_read() goes through).
> > 
> >  do_splice_to() checks that, should I move that checking further down in
> >  case the file is truncated?
> 
> Again, see do_generic_mapping_read()'s ghastly tricks - it checks i_size
> after each readpage().
> 
> i_size can increase or decrease under our feet if we're not holding i_mutex
> (and we don't want to).  So userspace is being silly and the main things we
> need to care about here are to not leak uninitialised data and to not oops.
> A readpage() outside i_size will return either all-zeroes or some valid
> data which isn't actually within i_size any more, so I guess we're OK.

So from the looks of things, worst case is returning zeroes if we
'raced' with someone truncating the file. I'll just leave it as-is for
now.

-- 
Jens Axboe