Re: [PATCH] handshake: set hr_sk field earlier in handshake_req_submit

[Chuck Lever III <chuck.lever@oracle.com>]

> On Mar 28, 2023, at 2:10 PM, Jeff Layton <jlayton@kernel.org> wrote:
> 
> Otherwise, the request is always hashed with the socket set to a NULL
> pointer.

Sounds plausible. I will need to check the clean-up paths too.


> Signed-off-by: Jeff Layton <jlayton@kernel.org>
> ---
> net/handshake/request.c | 2 +-
> 1 file changed, 1 insertion(+), 1 deletion(-)
> 
> I think this is the problem I was hitting earlier. Chuck, feel free to
> fold this into the original patch that adds this code.
> 
> diff --git a/net/handshake/request.c b/net/handshake/request.c
> index 54ba304b5bef..44a097210e59 100644
> --- a/net/handshake/request.c
> +++ b/net/handshake/request.c
> @@ -207,6 +207,7 @@ int handshake_req_submit(struct socket *sock, struct handshake_req *req,
> 	if (READ_ONCE(hn->hn_pending) >= hn->hn_pending_max)
> 		goto out_err;
> 
> +	req->hr_sk = sk;
> 	req->hr_odestruct = sk->sk_destruct;
> 	sk->sk_destruct = handshake_sk_destruct;
> 	spin_lock(&hn->hn_lock);
> @@ -230,7 +231,6 @@ int handshake_req_submit(struct socket *sock, struct handshake_req *req,
> 
> 	/* Prevent socket release while a handshake request is pending */
> 	sock_hold(sk);
> -	req->hr_sk = sk;
> 
> 	trace_handshake_submit(net, req, sk);
> 	return 0;
> -- 
> 2.39.2
> 

--
Chuck Lever