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;
From:   Hawkins Jiawei <yin31149@gmail.com>
To:     dhowells@redhat.com, "David S. Miller" <davem@davemloft.net>,
        Eric Dumazet <edumazet@google.com>,
        Jakub Kicinski <kuba@kernel.org>,
        Paolo Abeni <pabeni@redhat.com>,
        Alexei Starovoitov <ast@kernel.org>,
        Daniel Borkmann <daniel@iogearbox.net>,
        Andrii Nakryiko <andrii@kernel.org>,
        Martin KaFai Lau <kafai@fb.com>,
        Song Liu <songliubraving@fb.com>, Yonghong Song <yhs@fb.com>,
        John Fastabend <john.fastabend@gmail.com>,
        KP Singh <kpsingh@kernel.org>
Cc:     linux-kernel@vger.kernel.org, netdev@vger.kernel.org,
        yin31149@gmail.com, linux-kernel-mentees@lists.linuxfoundation.org,
        bpf@vger.kernel.org
Subject: Re: [PATCH] net: Fix suspicious RCU usage in bpf_sk_reuseport_detach()
Date:   Tue, 16 Aug 2022 18:34:52 +0800
Message-Id: <20220816103452.479281-1-yin31149@gmail.com>
In-Reply-To: <166064248071.3502205.10036394558814861778.stgit@warthog.procyon.org.uk>
References: <166064248071.3502205.10036394558814861778.stgit@warthog.procyon.org.uk>
MIME-Version: 1.0
Content-Transfer-Encoding: 8bit
Precedence: bulk

On Tue, 16 Aug 2022 at 17:34, David Howells <dhowells@redhat.com> wrote:
>
> Fix this by adding a new helper, __locked_read_sk_user_data_with_flags()
> that checks to see if sk->sk_callback_lock() is held and use that here
> instead.
Hi, I wonder if we make this more geniric, for I think maybe the future
code who use __rcu_dereference_sk_user_data_with_flags() may
also meet this bug.

To be more specific, maybe we can refactor
__rcu_dereference_sk_user_data_with_flags() to
__rcu_dereference_sk_user_data_with_flags_check(), like
rcu_dereference() and rcu_dereference_check(). Maybe:

diff --git a/include/net/sock.h b/include/net/sock.h
index 05a1bbdf5805..cf123954eab9 100644
--- a/include/net/sock.h
+++ b/include/net/sock.h
@@ -578,18 +578,27 @@ static inline bool sk_user_data_is_nocopy(const struct sock *sk)
 #define __sk_user_data(sk) ((*((void __rcu **)&(sk)->sk_user_data)))
 
 /**
- * __rcu_dereference_sk_user_data_with_flags - return the pointer
- * only if argument flags all has been set in sk_user_data. Otherwise
- * return NULL
+ * __rcu_dereference_sk_user_data_with_flags_check - return the pointer
+ * only if argument flags all has been set in sk_user_data, with debug
+ * checking. Otherwise return NULL
  *
- * @sk: socket
- * @flags: flag bits
+ * Do __rcu_dereference_sk_user_data_with_flags(), but check that the
+ * conditions under which the rcu dereference will take place are correct,
+ * which is a bit like rcu_dereference_check() and rcu_derefence().
+ *
+ * @sk		: socket
+ * @flags	: flag bits
+ * @condition	: the conditions under which the rcu dereference will
+ * take place
  */
 static inline void *
-__rcu_dereference_sk_user_data_with_flags(const struct sock *sk,
-					  uintptr_t flags)
+__rcu_dereference_sk_user_data_with_flags_check(const struct sock *sk,
+						uintptr_t flags, bool condition)
 {
-	uintptr_t sk_user_data = (uintptr_t)rcu_dereference(__sk_user_data(sk));
+	uintptr_t sk_user_data;
+
+	sk_user_data = (uintptr_t)rcu_dereference_check(__sk_user_data(sk),
+							condition);
 
 	WARN_ON_ONCE(flags & SK_USER_DATA_PTRMASK);
 
@@ -598,6 +607,8 @@ __rcu_dereference_sk_user_data_with_flags(const struct sock *sk,
 	return NULL;
 }
 
+#define __rcu_dereference_sk_user_data_with_flags(sk, flags) \
+	__rcu_dereference_sk_user_data_with_flags_check(sk, flags, 0)
 #define rcu_dereference_sk_user_data(sk)				\
 	__rcu_dereference_sk_user_data_with_flags(sk, 0)
 #define __rcu_assign_sk_user_data_with_flags(sk, ptr, flags)		\

> +/**
> + * __locked_read_sk_user_data_with_flags - return the pointer
> + * only if argument flags all has been set in sk_user_data. Otherwise
> + * return NULL
> + *
> +               (uintptr_t)rcu_dereference_check(__sk_user_data(sk),
> +                                                lockdep_is_held(&sk->sk_callback_lock));

> diff --git a/kernel/bpf/reuseport_array.c b/kernel/bpf/reuseport_array.c
> index 85fa9dbfa8bf..82c61612f382 100644
> --- a/kernel/bpf/reuseport_array.c
> +++ b/kernel/bpf/reuseport_array.c
> @@ -24,7 +24,7 @@ void bpf_sk_reuseport_detach(struct sock *sk)
>         struct sock __rcu **socks;
> 
>         write_lock_bh(&sk->sk_callback_lock);
> -       socks = __rcu_dereference_sk_user_data_with_flags(sk, SK_USER_DATA_BPF);
> +       socks = __locked_read_sk_user_data_with_flags(sk, SK_USER_DATA_BPF);
>         if (socks) {
>                 WRITE_ONCE(sk->sk_user_data, NULL);
>                 /*
Then, as you point out, we can pass
condition(lockdep_is_held(&sk->sk_callback_lock)) to
__rcu_dereference_sk_user_data_with_flags_check() in order to
make compiler happy as below:

diff --git a/kernel/bpf/reuseport_array.c b/kernel/bpf/reuseport_array.c
index 85fa9dbfa8bf..a772610987c5 100644
--- a/kernel/bpf/reuseport_array.c
+++ b/kernel/bpf/reuseport_array.c
@@ -24,7 +24,10 @@ void bpf_sk_reuseport_detach(struct sock *sk)
 	struct sock __rcu **socks;
 
 	write_lock_bh(&sk->sk_callback_lock);
-	socks = __rcu_dereference_sk_user_data_with_flags(sk, SK_USER_DATA_BPF);
+	socks = __rcu_dereference_sk_user_data_with_flags_check(
+			sk, SK_USER_DATA_BPF,
+			lockdep_is_held(&sk->sk_callback_lock));
+
 	if (socks) {
 		WRITE_ONCE(sk->sk_user_data, NULL);
 		/*