Re: [PATCH net 01/10] netfilter: flowtable: publish HW_DEAD after worker is done

[Pablo Neira Ayuso <pablo@netfilter.org>]

On Thu, Sep 17, 2026 at 07:04:06PM -0700, Jakub Kicinski wrote:
> This is an AI-generated review of your patch. The human sending this
> email has considered the AI review valid, or at least plausible.
> ---
> netfilter: flowtable: publish HW_DEAD after worker is done
> 
> The patch moves the NF_FLOW_HW_DEAD publication out of
> flow_offload_work_del() into the common epilogue of
> flow_offload_work_handler(), so that it happens after
> clear_bit(NF_FLOW_HW_PENDING) rather than before it.  The stated goal is
> to make setting HW_DEAD the last flow access performed by destroy work,
> so garbage collection cannot free the flow while the worker is still
> touching it.
> 
> > diff --git a/net/netfilter/nf_flow_table_offload.c b/net/netfilter/nf_flow_table_offload.c
> > index 801a3dd9ceea3..6757fd89c1f1a 100644
> > --- a/net/netfilter/nf_flow_table_offload.c
> > +++ b/net/netfilter/nf_flow_table_offload.c
> > @@ -995,7 +995,6 @@ static void flow_offload_work_del(struct flow_offload_work *offload)
> >  	flow_offload_tuple_del(offload, FLOW_OFFLOAD_DIR_ORIGINAL);
> >  	if (test_bit(NF_FLOW_HW_BIDIRECTIONAL, &offload->flow->flags))
> >  		flow_offload_tuple_del(offload, FLOW_OFFLOAD_DIR_REPLY);
> > -	set_bit(NF_FLOW_HW_DEAD, &offload->flow->flags);
> >  }
> >  
> >  static void flow_offload_tuple_stats(struct flow_offload_work *offload,
> > @@ -1059,6 +1058,12 @@ static void flow_offload_work_handler(struct work_struct *work)
> >  	}
> >  
> >  	clear_bit(NF_FLOW_HW_PENDING, &offload->flow->flags);
> > +	if (offload->cmd == FLOW_CLS_DESTROY) {
> > +		/* Publish after the worker's last flow access. */
> > +		smp_mb__before_atomic();
> > +		set_bit(NF_FLOW_HW_DEAD, &offload->flow->flags);
> > +	}
> > +
> 
> [Severity: High]
> Does this reordering close the whole use-after-free, or only the
> FLOW_CLS_DESTROY part of it?

This describes a pre-existing issue. I posted patches for this I have
to pick up.