Re: [PATCH 2.4] Fix rlimit check in precheck_file_write()

[Jason Uhlenkott <jasonuhl@sgi.com>]

On Thu, Dec 23, 2004 at 11:29:33AM +0000, Alan Cox wrote:
> On Mer, 2004-12-22 at 21:57, Jason Uhlenkott wrote:
> > Remove a broken assumption that rlimits are 32 bits which can cause
> > nasty things to happen on 64-bit machines if we try to write past the
> > 2^32-1th character of a file and a larger file size limit exists.
> > 
> > Signed-off-by: Jason Uhlenkott <jasonuhl@sgi.com>
> > 
> > --- linux-2.4.29-pre3.orig/mm/filemap.c	2004-11-17 03:54:22.000000000 -0800
> > +++ linux-2.4.29-pre3/mm/filemap.c	2004-12-22 13:41:46.000000000 -0800
> > @@ -3088,9 +3088,9 @@
> >  			send_sig(SIGXFSZ, current, 0);
> >  			goto out;
> >  		}
> > -		if (pos > 0xFFFFFFFFULL || *count > limit - (u32)pos) {
> > +		if (*count > limit - pos) {
> >  			/* send_sig(SIGXFSZ, current, 0); */
> > -			*count = limit - (u32)pos;
> > +			*count = limit - pos;
> >  		}
> 
> Are you sure this is safe for all conceivable 32bit cases as well as
> your 64bit one ? I don't think it is looking at the overflow cases in
> the if that you removed checking of.

We can't overflow -- immediately above this, we bail out if pos >= limit.