Received-SPF: pass (google.com: domain of linux-kernel-owner@vger.kernel.org designates 2620:137:e000::1:20 as permitted sender) client-ip=2620:137:e000::1:20;
MIME-Version: 1.0
References: <20230531141556.1637341-1-lee@kernel.org>
In-Reply-To: <20230531141556.1637341-1-lee@kernel.org>
From:   Jamal Hadi Salim <jhs@mojatatu.com>
Date:   Wed, 31 May 2023 11:05:25 -0400
Message-ID: <CAM0EoMnw_4Z6tGC8=JkGKuCFTkJHt5JOFj56+_Z6GDExAMbugQ@mail.gmail.com>
Subject: Re: [PATCH 1/1] net/sched: cls_u32: Fix reference counter leak
 leading to overflow
To:     Lee Jones <lee@kernel.org>
Cc:     xiyou.wangcong@gmail.com, jiri@resnulli.us, davem@davemloft.net,
        edumazet@google.com, kuba@kernel.org, pabeni@redhat.com,
        linux-kernel@vger.kernel.org, netdev@vger.kernel.org,
        stable@kernel.org
Content-Type: text/plain; charset="UTF-8"
Content-Transfer-Encoding: quoted-printable
Precedence: bulk

On Wed, May 31, 2023 at 10:16=E2=80=AFAM Lee Jones <lee@kernel.org> wrote:
>
> In the event of a failure in tcf_change_indev(), u32_set_parms() will
> immediately return without decrementing the recently incremented
> reference counter.  If this happens enough times, the counter will
> rollover and the reference freed, leading to a double free which can be
> used to do 'bad things'.
>
> Cc: stable@kernel.org # v4.14+
> Signed-off-by: Lee Jones <lee@kernel.org>
> ---
>  net/sched/cls_u32.c | 5 ++++-
>  1 file changed, 4 insertions(+), 1 deletion(-)
>
> diff --git a/net/sched/cls_u32.c b/net/sched/cls_u32.c
> index 4e2e269f121f8..fad61ca5e90bf 100644
> --- a/net/sched/cls_u32.c
> +++ b/net/sched/cls_u32.c
> @@ -762,8 +762,11 @@ static int u32_set_parms(struct net *net, struct tcf=
_proto *tp,
>         if (tb[TCA_U32_INDEV]) {
>                 int ret;
>                 ret =3D tcf_change_indev(net, tb[TCA_U32_INDEV], extack);
> -               if (ret < 0)
> +               if (ret < 0) {
> +                       if (tb[TCA_U32_LINK])
> +                               n->ht_down->refcnt--;
>                         return -EINVAL;
> +               }
>                 n->ifindex =3D ret;
>         }
>         return 0;

The spirit of the patch looks right I dont think this fully solves the
issue you state.
My suggestion: Move the if (tb[TCA_U32_INDEV])  above the if
(tb[TCA_U32_LINK]) {
Did you see this in practice or you found it by eyeballing the code?
Can you also add a tdc test for it? There are simple ways to create
the scenario.

cheers,
jamal
> 2.41.0.rc0.172.g3f132b7071-goog
>