[PATCH v1 net] net: Annotate sk->sk_write_space() for UDP SOCKMAP.

[PATCH v1 net] net: Annotate sk->sk_write_space() for UDP SOCKMAP.

[Kuniyuki Iwashima]

UDP TX skb->destructor() is sock_wfree(), and UDP holds lock_sock()
only for UDP_CORK / MSG_MORE sendmsg().

Otherwise, sk->sk_write_space() may be read locklessly while SOCKMAP
rewrites sk->sk_write_space().

Let's use WRITE_ONCE() and READ_ONCE() for sk->sk_write_space().

Note that the write side is annotated by commit 2ef2b20cf4e0
("net: annotate data-races around sk->sk_{data_ready,write_space}").

Fixes: 7b98cd42b049 ("bpf: sockmap: Add UDP support")
Signed-off-by: Kuniyuki Iwashima <kuniyu@google.com>
---
 net/core/sock.c | 8 ++++++--
 1 file changed, 6 insertions(+), 2 deletions(-)

diff --git a/net/core/sock.c b/net/core/sock.c
index b37b664b6eb9..d097025c116a 100644
--- a/net/core/sock.c
+++ b/net/core/sock.c
@@ -2676,8 +2676,12 @@ void sock_wfree(struct sk_buff *skb)
 	int old;
 
 	if (!sock_flag(sk, SOCK_USE_WRITE_QUEUE)) {
+		void (*sk_write_space)(struct sock *sk);
+
+		sk_write_space = READ_ONCE(sk->sk_write_space);
+
 		if (sock_flag(sk, SOCK_RCU_FREE) &&
-		    sk->sk_write_space == sock_def_write_space) {
+		    sk_write_space == sock_def_write_space) {
 			rcu_read_lock();
 			free = __refcount_sub_and_test(len, &sk->sk_wmem_alloc,
 						       &old);
@@ -2693,7 +2697,7 @@ void sock_wfree(struct sk_buff *skb)
 		 * after sk_write_space() call
 		 */
 		WARN_ON(refcount_sub_and_test(len - 1, &sk->sk_wmem_alloc));
-		sk->sk_write_space(sk);
+		sk_write_space(sk);
 		len = 1;
 	}
 	/*
-- 
2.54.0.823.g6e5bcc1fc9-goog

Re: [PATCH v1 net] net: Annotate sk->sk_write_space() for UDP SOCKMAP.

[Jakub Sitnicki]

On Fri, May 29, 2026 at 07:39 PM GMT, Kuniyuki Iwashima wrote:
> UDP TX skb->destructor() is sock_wfree(), and UDP holds lock_sock()
> only for UDP_CORK / MSG_MORE sendmsg().

Nit: lock_sock() serializes map updates via bpf() syscall. We also have
sockmap iterator programs which can insert UDP sockets into sockmap. On
this path, sock_map_update_elem, we grab only bh_lock_sock, so the race
against UDP_CORK / MSG_MORE sendmsg() path also exists.

> Otherwise, sk->sk_write_space() may be read locklessly while SOCKMAP
> rewrites sk->sk_write_space().
>
> Let's use WRITE_ONCE() and READ_ONCE() for sk->sk_write_space().
>
> Note that the write side is annotated by commit 2ef2b20cf4e0
> ("net: annotate data-races around sk->sk_{data_ready,write_space}").
>
> Fixes: 7b98cd42b049 ("bpf: sockmap: Add UDP support")
> Signed-off-by: Kuniyuki Iwashima <kuniyu@google.com>
> ---

Reviewed-by: Jakub Sitnicki <jakub@cloudflare.com>

Re: [PATCH v1 net] net: Annotate sk->sk_write_space() for UDP SOCKMAP.

[patchwork-bot+netdevbpf]

Hello:

This patch was applied to netdev/net.git (main)
by Jakub Kicinski <kuba@kernel.org>:

On Fri, 29 May 2026 19:39:23 +0000 you wrote:
> UDP TX skb->destructor() is sock_wfree(), and UDP holds lock_sock()
> only for UDP_CORK / MSG_MORE sendmsg().
> 
> Otherwise, sk->sk_write_space() may be read locklessly while SOCKMAP
> rewrites sk->sk_write_space().
> 
> Let's use WRITE_ONCE() and READ_ONCE() for sk->sk_write_space().
> 
> [...]

Here is the summary with links:
  - [v1,net] net: Annotate sk->sk_write_space() for UDP SOCKMAP.
    https://git.kernel.org/netdev/net/c/b748765019fe

You are awesome, thank you!
-- 
Deet-doot-dot, I am a bot.
https://korg.docs.kernel.org/patchwork/pwbot.html