struct nf_conn)->timeout can be read/written locklessly,
add READ_ONCE()/WRITE_ONCE() to prevent load/store tearing.

The patch 'commit 802a7dc5cf1b ("netfilter: conntrack: annotate
data-races around ct->timeout")'fixed it, but there was a
missing part that this patch completes it.

Fixes: 802a7dc5cf1b ("netfilter: conntrack: annotate data-races around ct->timeout")
Signed-off-by: Chenguang Zhao <zhaochenguang@kylinos.cn>
---
 net/netfilter/nf_conntrack_core.c | 2 +-
 1 file changed, 1 insertion(+), 1 deletion(-)

diff --git a/net/netfilter/nf_conntrack_core.c b/net/netfilter/nf_conntrack_core.c
index 344f88295976..df4426adc9c8 100644
--- a/net/netfilter/nf_conntrack_core.c
+++ b/net/netfilter/nf_conntrack_core.c
@@ -1297,7 +1297,7 @@ __nf_conntrack_confirm(struct sk_buff *skb)
 	/* Timeout is relative to confirmation time, not original
 	   setting time, otherwise we'd get timer wrap in
 	   weird delay cases. */
-	ct->timeout += nfct_time_stamp;
+	WRITE_ONCE(ct->timeout, READ_ONCE(ct->timeout) + nfct_time_stamp);
 
 	__nf_conntrack_insert_prepare(ct);
 
-- 
2.25.1

Chenguang Zhao <zhaochenguang@kylinos.cn> wrote:
> struct nf_conn)->timeout can be read/written locklessly,
> add READ_ONCE()/WRITE_ONCE() to prevent load/store tearing.

Do you have a KCSAN splat or similar?

> The patch 'commit 802a7dc5cf1b ("netfilter: conntrack: annotate
> data-races around ct->timeout")'fixed it, but there was a
> missing part that this patch completes it.

I'm no longer sure this was missing.

> diff --git a/net/netfilter/nf_conntrack_core.c b/net/netfilter/nf_conntrack_core.c
> index 344f88295976..df4426adc9c8 100644
> --- a/net/netfilter/nf_conntrack_core.c
> +++ b/net/netfilter/nf_conntrack_core.c
> @@ -1297,7 +1297,7 @@ __nf_conntrack_confirm(struct sk_buff *skb)
>  	/* Timeout is relative to confirmation time, not original
>  	   setting time, otherwise we'd get timer wrap in
>  	   weird delay cases. */
> -	ct->timeout += nfct_time_stamp;
> +	WRITE_ONCE(ct->timeout, READ_ONCE(ct->timeout) + nfct_time_stamp);

Here we hold the bucket insert locks for ct, so I don't see
how we can have concurrent modification here.