Re: bdi_min_ratio never shrinks, ultimately preventing validsetting of min_ratio

[Andrew Morton]

On Thu, 02 Jun 2011 23:25:58 +0200
Peter Zijlstra <peterz@xxxxxxxxxxxxx> wrote:

> On Thu, 2011-06-02 at 13:32 -0500, lkml@xxxxxxxxxxx wrote:
> > > > There is no place in this listing where the value is decremented by the
> > > > respective bdi's min_ratio when a bdi is torn down. 
> > > 
> > > There is, adding a negative number is equal to a subtraction.
> > > 
> > >                 min_ratio -= bdi->min_ratio;
> > >                 if (bdi_min_ratio + min_ratio < 100) {
> > >                         bdi_min_ratio += min_ratio;
> > >                         bdi->min_ratio += min_ratio;
> > >                 }
> > > 
> > > is the relevant piece, note that bdi->min_ratio is the current setting,
> > > this makes min_ratio the difference between the new and old setting, and
> > > adding this to both bdi_min_ratio (the global sum) and bdi->min_ratio
> > > dtrt regardless if the new value is larger or smaller than the old
> > > value.
> > 
> > This accounts for the repeated setting of min_ratio on the same bdi.  But
> > does bdi_set_min_ratio() get entered with a min_ratio of 0 on bdi removal?
> > If not, we leak the non-zero min_ratio of a removed bdi.
> 
> That does not appear to be the case, good catch. Would you be bitten by
> that particular scenario? If so, does the below cure things for you?
> 
> ---
>  mm/backing-dev.c |    1 +
>  1 files changed, 1 insertions(+), 0 deletions(-)
> 
> diff --git a/mm/backing-dev.c b/mm/backing-dev.c
> index f032e6e..e56fe35 100644
> --- a/mm/backing-dev.c
> +++ b/mm/backing-dev.c
> @@ -606,6 +606,7 @@ static void bdi_prune_sb(struct backing_dev_info *bdi)
>  void bdi_unregister(struct backing_dev_info *bdi)
>  {
>  	if (bdi->dev) {
> +		bdi_set_min_ratio(bdi, 0);
>  		trace_writeback_bdi_unregister(bdi);
>  		bdi_prune_sb(bdi);
>  		del_timer_sync(&bdi->wb.wakeup_timer);

I grabbed this, wrote a changelog and stuck your signed-off-by on it. 
Vito, it would be great if you are able to test this please.

I also added a cc:stable but I didn't work out how far back in time it
goes.  A long way, I think?

--
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
the body of a message to majordomo@xxxxxxxxxxxxxxx
More majordomo info at  http://vger.kernel.org/majordomo-info.html
Please read the FAQ at  http://www.tux.org/lkml/