Commit 198bc90e0e73 ("tcp: make sure init the accept_queue's spinlocks
once") moved the TCP accept_queue spinlock initialization from a common
core function to two places for two cases: the common accept callback,
and the socket create callback.
For the second case, only AF_INET (IPv4) was considered. This results
in a lockdep error when accepting an incoming IPv6 TCP connection.
INFO: trying to register non-static key.
The code is fine but needs lockdep annotation, or maybe
you didn't initialize this object before use?
turning off the locking correctness validator.
Call trace:
... <stack dump> ...
register_lock_class (kernel/locking/lockdep.c:977 kernel/locking/lockdep.c:1289)
__lock_acquire (kernel/locking/lockdep.c:5014)
lock_acquire (./arch/arm64/include/asm/percpu.h:40 kernel/locking/lockdep.c:467 kernel/locking/lockdep.c:5756 kernel/locking/lockdep.c:5719)
_raw_spin_lock (./include/linux/spinlock_api_smp.h:134 kernel/locking/spinlock.c:154)
inet_csk_complete_hashdance (net/ipv4/inet_connection_sock.c:1303 net/ipv4/inet_connection_sock.c:1355)
tcp_check_req (net/ipv4/tcp_minisocks.c:653)
tcp_v6_rcv (net/ipv6/tcp_ipv6.c:1837)
ip6_protocol_deliver_rcu (net/ipv6/ip6_input.c:438)
ip6_input_finish (./include/linux/rcupdate.h:779 net/ipv6/ip6_input.c:484)
ip6_input (./include/linux/netfilter.h:314 ./include/linux/netfilter.h:308 net/ipv6/ip6_input.c:492)
ip6_sublist_rcv_finish (net/ipv6/ip6_input.c:86 (discriminator 3))
ip6_sublist_rcv (net/ipv6/ip6_input.c:317)
ipv6_list_rcv (net/ipv6/ip6_input.c:326)
__netif_receive_skb_list_core (net/core/dev.c:5577 net/core/dev.c:5625)
netif_receive_skb_list_internal (net/core/dev.c:5679 net/core/dev.c:5768)
napi_complete_done (./include/linux/list.h:37 (discriminator 2) ./include/net/gro.h:440 (discriminator 2) ./include/net/gro.h:435 (discriminator 2) net/core/dev.c:6108 (discriminator 2))
... <device callback> ...
Fix this by adding the appropriate code to AF_INET6 (IPv6) socket create
callback, mirroring what was added for AF_INET.
Fixes: 198bc90e0e73 ("tcp: make sure init the accept_queue's spinlocks once")
Signed-off-by: Chen-Yu Tsai <[email protected]>
---
net/ipv6/af_inet6.c | 3 +++
1 file changed, 3 insertions(+)
diff --git a/net/ipv6/af_inet6.c b/net/ipv6/af_inet6.c
index 13a1833a4df5..959bfd9f6344 100644
--- a/net/ipv6/af_inet6.c
+++ b/net/ipv6/af_inet6.c
@@ -199,6 +199,9 @@ static int inet6_create(struct net *net, struct socket *sock, int protocol,
if (INET_PROTOSW_REUSE & answer_flags)
sk->sk_reuse = SK_CAN_REUSE;
+ if (INET_PROTOSW_ICSK & answer_flags)
+ inet_init_csk_locks(sk);
+
inet = inet_sk(sk);
inet_assign_bit(IS_ICSK, sk, INET_PROTOSW_ICSK & answer_flags);
--
2.43.0.429.g432eaa2c6b-goog
Hi Chen-Yu Tsai:
I have send the fixed patch.
https://lore.kernel.org/all/[email protected]/
Thank you
Zhengchao Shao
On 2024/1/22 18:23, Chen-Yu Tsai wrote:
> Commit 198bc90e0e73 ("tcp: make sure init the accept_queue's spinlocks
> once") moved the TCP accept_queue spinlock initialization from a common
> core function to two places for two cases: the common accept callback,
> and the socket create callback.
>
> For the second case, only AF_INET (IPv4) was considered. This results
> in a lockdep error when accepting an incoming IPv6 TCP connection.
>
> INFO: trying to register non-static key.
> The code is fine but needs lockdep annotation, or maybe
> you didn't initialize this object before use?
> turning off the locking correctness validator.
> Call trace:
> ... <stack dump> ...
> register_lock_class (kernel/locking/lockdep.c:977 kernel/locking/lockdep.c:1289)
> __lock_acquire (kernel/locking/lockdep.c:5014)
> lock_acquire (./arch/arm64/include/asm/percpu.h:40 kernel/locking/lockdep.c:467 kernel/locking/lockdep.c:5756 kernel/locking/lockdep.c:5719)
> _raw_spin_lock (./include/linux/spinlock_api_smp.h:134 kernel/locking/spinlock.c:154)
> inet_csk_complete_hashdance (net/ipv4/inet_connection_sock.c:1303 net/ipv4/inet_connection_sock.c:1355)
> tcp_check_req (net/ipv4/tcp_minisocks.c:653)
> tcp_v6_rcv (net/ipv6/tcp_ipv6.c:1837)
> ip6_protocol_deliver_rcu (net/ipv6/ip6_input.c:438)
> ip6_input_finish (./include/linux/rcupdate.h:779 net/ipv6/ip6_input.c:484)
> ip6_input (./include/linux/netfilter.h:314 ./include/linux/netfilter.h:308 net/ipv6/ip6_input.c:492)
> ip6_sublist_rcv_finish (net/ipv6/ip6_input.c:86 (discriminator 3))
> ip6_sublist_rcv (net/ipv6/ip6_input.c:317)
> ipv6_list_rcv (net/ipv6/ip6_input.c:326)
> __netif_receive_skb_list_core (net/core/dev.c:5577 net/core/dev.c:5625)
> netif_receive_skb_list_internal (net/core/dev.c:5679 net/core/dev.c:5768)
> napi_complete_done (./include/linux/list.h:37 (discriminator 2) ./include/net/gro.h:440 (discriminator 2) ./include/net/gro.h:435 (discriminator 2) net/core/dev.c:6108 (discriminator 2))
> ... <device callback> ...
>
> Fix this by adding the appropriate code to AF_INET6 (IPv6) socket create
> callback, mirroring what was added for AF_INET.
>
> Fixes: 198bc90e0e73 ("tcp: make sure init the accept_queue's spinlocks once")
> Signed-off-by: Chen-Yu Tsai <[email protected]>
> ---
> net/ipv6/af_inet6.c | 3 +++
> 1 file changed, 3 insertions(+)
>
> diff --git a/net/ipv6/af_inet6.c b/net/ipv6/af_inet6.c
> index 13a1833a4df5..959bfd9f6344 100644
> --- a/net/ipv6/af_inet6.c
> +++ b/net/ipv6/af_inet6.c
> @@ -199,6 +199,9 @@ static int inet6_create(struct net *net, struct socket *sock, int protocol,
> if (INET_PROTOSW_REUSE & answer_flags)
> sk->sk_reuse = SK_CAN_REUSE;
>
> + if (INET_PROTOSW_ICSK & answer_flags)
> + inet_init_csk_locks(sk);
> +
> inet = inet_sk(sk);
> inet_assign_bit(IS_ICSK, sk, INET_PROTOSW_ICSK & answer_flags);
>