Received-SPF: pass (google.com: domain of linux-kernel+bounces-3414-linux.lists.archive=gmail.com@vger.kernel.org designates 147.75.80.249 as permitted sender) client-ip=147.75.80.249;
Precedence: bulk
MIME-Version: 1.0
References: <20231218084252.7644-1-zhangyiqun@phytium.com.cn>
In-Reply-To: <20231218084252.7644-1-zhangyiqun@phytium.com.cn>
From: Eric Dumazet <edumazet@google.com>
Date: Mon, 18 Dec 2023 12:07:50 +0100
Message-ID: <CANn89i+t9t5ca=r6ZKw7s-HrxzgJjCB6hmWLccKmmxg8H=HUUA@mail.gmail.com>
Subject: Re: [PATCH] xfrm: Use spin_lock_bh() in xfrm_input()
To: Zhang Yiqun <zhangyiqun@phytium.com.cn>
Cc: steffen.klassert@secunet.com, herbert@gondor.apana.org.au, 
	davem@davemloft.net, kuba@kernel.org, pabeni@redhat.com, 
	netdev@vger.kernel.org, linux-kernel@vger.kernel.org
Content-Type: text/plain; charset="UTF-8"
Content-Transfer-Encoding: quoted-printable

On Mon, Dec 18, 2023 at 9:43=E2=80=AFAM Zhang Yiqun <zhangyiqun@phytium.com=
.cn> wrote:
>
> This patch is to change spin_lock() into spin_lock_bh(), which can
> disable bottem half in calling. If we leave this as spin_lock(),
> it may stuck in a deadlock, because the callback in bottem half in
> crypto driver will also call xfrm_input() again.
>
> Signed-off-by: Zhang Yiqun <zhangyiqun@phytium.com.cn>

When was the bug added ?
We need a FIxes: tag.

Also a stack trace to show the deadlock (or lockdep complaint ) would
be needed as well.

> ---
>  net/xfrm/xfrm_input.c | 4 ++--
>  1 file changed, 2 insertions(+), 2 deletions(-)
>
> diff --git a/net/xfrm/xfrm_input.c b/net/xfrm/xfrm_input.c
> index bd4ce21d76d7..f4cd46d73b1e 100644
> --- a/net/xfrm/xfrm_input.c
> +++ b/net/xfrm/xfrm_input.c
> @@ -581,7 +581,7 @@ int xfrm_input(struct sk_buff *skb, int nexthdr, __be=
32 spi, int encap_type)
>                 }
>
>  lock:
> -               spin_lock(&x->lock);
> +               spin_lock_bh(&x->lock);
>
>                 if (unlikely(x->km.state !=3D XFRM_STATE_VALID)) {
>                         if (x->km.state =3D=3D XFRM_STATE_ACQ)
> @@ -607,7 +607,7 @@ int xfrm_input(struct sk_buff *skb, int nexthdr, __be=
32 spi, int encap_type)
>                         goto drop_unlock;
>                 }
>
> -               spin_unlock(&x->lock);
> +               spin_unlock_bh(&x->lock);
>
>                 if (xfrm_tunnel_check(skb, x, family)) {
>                         XFRM_INC_STATS(net, LINUX_MIB_XFRMINSTATEMODEERRO=
R);

This patch is not correct anyway.
There are five places in xfrm_input() where x->lock is either locked
or unlocked.

Please tell us how this was tested.
Thanks.