Re: [PATCH v3] mm,writeback: fix divide by zero in pos_ratio_polynom

[Andrew Morton <akpm@linux-foundation.org>]

On Wed, 30 Apr 2014 15:30:04 -0400 Rik van Riel <riel@redhat.com> wrote:

> On 04/30/2014 03:00 PM, Andrew Morton wrote:
> > On Wed, 30 Apr 2014 10:41:14 -0400 Rik van Riel <riel@redhat.com> wrote:
> >
> >> It is possible for "limit - setpoint + 1" to equal zero, leading to a
> >> divide by zero error. Blindly adding 1 to "limit - setpoint" is not
> >> working, so we need to actually test the divisor before calling div64.
> >>
> >> ...
> >>
> >> --- a/mm/page-writeback.c
> >> +++ b/mm/page-writeback.c
> >> @@ -598,10 +598,15 @@ static inline long long pos_ratio_polynom(unsigned long setpoint,
> >>   					  unsigned long limit)
> >>   {
> >>   	long long pos_ratio;
> >> +	long divisor;
> >>   	long x;
> >>
> >> +	divisor = limit - setpoint;
> >> +	if (!(s32)divisor)
> >> +		divisor = 1;	/* Avoid div-by-zero */
> >> +
> >>   	x = div_s64(((s64)setpoint - (s64)dirty) << RATELIMIT_CALC_SHIFT,
> >> -		    limit - setpoint + 1);
> >> +		    (s32)divisor);
> >
> > Doesn't this just paper over the bug one time in four billion?  The
> > other 3999999999 times, pos_ratio_polynom() returns an incorect result?
> >
> > If it is indeed the case that pos_ratio_polynom() callers are
> > legitimately passing a setpoint which is more than 2^32 less than limit
> > then it would be better to handle that input correctly.
> 
> The easy way would be by calling div64_s64 and div64_u64,
> which are 64 bit all the way through.
> 
> Any objections?

Sounds good to me.

> The inlined bits seem to be stubs calling the _rem variants
> of the functions, and discarding the remainder.

I was referring to pos_ratio_polynom().  The compiler will probably be
uninlining it anyway, but still...