Re: Linux 6.19.4 - Oops, regression

[Kris Karas (Bug Reporting)]

Greg KH wrote:
> Ick, not good.  Can you do 'git bisect' to find the problem commit?  As
> you have a pretty reliable reproducer it should go pretty fast.

Bisect complete (from 6.19.3 to 6.19.4)

> e308d4e35ce1e26cd67070a7035ad265662ab9e5 is the first bad commit
> commit e308d4e35ce1e26cd67070a7035ad265662ab9e5
> Author: Pablo Neira Ayuso <pablo@xxxxxxxxxxxxx>
> Date:   Wed Jan 21 01:08:45 2026 +0100
>
>     netfilter: nft_set_rbtree: translate rbtree to array for binary search
>     
>     [ Upstream commit 7e43e0a1141deec651a60109dab3690854107298 ]
>     
>     The rbtree can temporarily store overlapping inactive elements during
>     the transaction processing, leading to false negative lookups.
[snip]

Kris

P.S.  I ran the bisect from 7.0-rc1, so many reboots there, and
      it is fine.  The bug only seems to affect 6.x