On Tue, Jul 22, 2025 at 04:11:36AM +0200, Pablo Neira Ayuso wrote: > > This function is called from abort path too. I suspect total_jump_count > for this table will not be OK in such case. And this selftest does cover > many cases. > Your suspicion was correct. It took a bit to run over the abort path that used validation (I think it would just be userland sending abort), but it looked like it updated the total before the roll back. My solution was to send in the abort state, and when action != NFNL_ABORT_VALIDATE, we don't update the totals. Another alternative would be to just run validate again at the end of the abort (after the rollback), but that seemed wrong to do the costly validate twice. See v6 patch (just sent) that also addresses your other feedback. Thanks! SB
