Re: [PATCH net v2 2/5] vsock/virtio: Streamline socket reset on transport/PM event

[sashiko-bot@kernel.org]

Thank you for your contribution! Sashiko AI review found 1 potential issue(s) to consider:
- [Medium] vsock: lost peer_shutdown update in lockless reset
--

commit c52d0f36e0857217fee2a98454aaab9edf069cc3
Author: Michal Luczaj <mhal@rbox.co>

vsock/virtio: Streamline socket reset on transport/PM event

This commit streamlines socket resets on transport or power management
events by setting SHUTDOWN_MASK in virtio_vsock_reset_sock(). It aligns
the virtio transport with similar behavior in vhost and VMCI.

> diff --git a/net/vmw_vsock/virtio_transport.c b/net/vmw_vsock/virtio_transport.c
> --- a/net/vmw_vsock/virtio_transport.c
> +++ b/net/vmw_vsock/virtio_transport.c

[ ... ]

>  static void virtio_vsock_reset_sock(struct sock *sk)
>  {
> +	struct vsock_sock *vsk = vsock_sk(sk);
> +
>  	/* vmci_transport.c doesn't take sk_lock here either.  At least we're
>  	 * under vsock_table_lock so the sock cannot disappear while we're
>  	 * executing.
>  	 */
>  
> +	vsk->peer_shutdown = SHUTDOWN_MASK;

[Severity: Medium]
Could this lockless assignment race with the lock-protected read-modify-write
in the RX path?

Looking at virtio_transport_recv_connected(), it processes
VIRTIO_VSOCK_OP_SHUTDOWN while holding sk_lock. It reads vsk->peer_shutdown
via READ_ONCE(), applies a bitwise OR (such as peer_shutdown |= RCV_SHUTDOWN),
and writes it back with WRITE_ONCE().

If the lockless assignment here occurs between the RX thread's read and
write, could the RX thread overwrite SHUTDOWN_MASK with its local value,
dropping the SEND_SHUTDOWN state?

If this lost update occurs, it seems the socket would lose its full shutdown
state. Would this cause vsock_poll() to miss setting EPOLLHUP and EPOLLRDHUP
flags, breaking API expectations?

Additionally, might this plain write trigger KCSAN data race warnings against
concurrent READ_ONCE() calls?

>  	sk->sk_state = TCP_CLOSE;
>  	sk->sk_err = ECONNRESET;
>  	sk_error_report(sk);
>  }

-- 
Sashiko AI review · https://sashiko.dev/#/patchset/20260915-vsock-connect-reset-closing-v2-0-a1d9abb472f7@rbox.co?part=2