Re: [net-next,V5,06/15] devlink: Decouple rate storage from associated devlink object

[Cosmin Ratiu <cratiu@nvidia.com>]

On Wed, 2026-01-21 at 19:39 -0800, Jakub Kicinski wrote:
> > @@ -220,6 +248,9 @@ devlink_nl_rate_get_dump_one(struct sk_buff
> > *msg, struct devlink *devlink,
> >  			idx++;
> >  			continue;
> >  		}
> > +		if (devlink_rate->devlink != devlink)
> > +			continue;
> 
> When this filter skips a non-matching entry, idx is not incremented.
> However, state->idx is used to track position across netlink dump
> resumptions and expects idx to reflect the actual position in the
> list.
> 
> Once cross-device rate nodes are enabled (later in this series), the
> list
> can contain entries from multiple devlinks. If the dump requires
> multiple
> netlink messages and non-matching entries exist between matching
> ones,
> the idx tracking becomes inconsistent with the list position. This
> can
> cause entries to be reported multiple times on dump resumption.
> 
> Should this continue statement also increment idx to maintain correct
> position tracking?

Another useful finding. I suspect the code might work as-is, since
state->idx tracks "the next pending item for 'devlink'" from the list,
but it's quite hard to reason about that and it's probably better to do
as the tool suggests, and track the list position as now. Will do so in
the next version.

Cosmin.