Re: [PATCH net-next v7 3/4] virtio/vsock: fix logic which reduces credit update messages

["Michael S. Tsirkin" <mst@redhat.com>]

On Thu, Dec 07, 2023 at 12:52:51AM +0300, Arseniy Krasnov wrote:
> 
> 
> On 07.12.2023 00:53, Michael S. Tsirkin wrote:
> > On Thu, Dec 07, 2023 at 12:18:48AM +0300, Arseniy Krasnov wrote:
> >> Add one more condition for sending credit update during dequeue from
> >> stream socket: when number of bytes in the rx queue is smaller than
> >> SO_RCVLOWAT value of the socket. This is actual for non-default value
> >> of SO_RCVLOWAT (e.g. not 1) - idea is to "kick" peer to continue data
> >> transmission, because we need at least SO_RCVLOWAT bytes in our rx
> >> queue to wake up user for reading data (in corner case it is also
> >> possible to stuck both tx and rx sides, this is why 'Fixes' is used).
> >> Also handle case when 'fwd_cnt' wraps, while 'last_fwd_cnt' is still
> >> not.
> >>
> >> Fixes: b89d882dc9fc ("vsock/virtio: reduce credit update messages")
> >> Signed-off-by: Arseniy Krasnov <avkrasnov@salutedevices.com>
> >> ---
> >>  Changelog:
> >>  v6 -> v7:
> >>   * Handle wrap of 'fwd_cnt'.
> >>   * Do to send credit update when 'fwd_cnt' == 'last_fwd_cnt'.
> >>
> >>  net/vmw_vsock/virtio_transport_common.c | 18 +++++++++++++++---
> >>  1 file changed, 15 insertions(+), 3 deletions(-)
> >>
> >> diff --git a/net/vmw_vsock/virtio_transport_common.c b/net/vmw_vsock/virtio_transport_common.c
> >> index e137d740804e..39f8660d825d 100644
> >> --- a/net/vmw_vsock/virtio_transport_common.c
> >> +++ b/net/vmw_vsock/virtio_transport_common.c
> >> @@ -558,6 +558,8 @@ virtio_transport_stream_do_dequeue(struct vsock_sock *vsk,
> >>  	struct virtio_vsock_sock *vvs = vsk->trans;
> >>  	size_t bytes, total = 0;
> >>  	struct sk_buff *skb;
> >> +	u32 fwd_cnt_delta;
> >> +	bool low_rx_bytes;
> >>  	int err = -EFAULT;
> >>  	u32 free_space;
> >>  
> >> @@ -601,7 +603,15 @@ virtio_transport_stream_do_dequeue(struct vsock_sock *vsk,
> >>  		}
> >>  	}
> >>  
> >> -	free_space = vvs->buf_alloc - (vvs->fwd_cnt - vvs->last_fwd_cnt);
> >> +	/* Handle wrap of 'fwd_cnt'. */
> >> +	if (vvs->fwd_cnt < vvs->last_fwd_cnt)
> >> +		fwd_cnt_delta = vvs->fwd_cnt + (U32_MAX - vvs->last_fwd_cnt);
> > 
> > Are you sure there's no off by one here? for example if fwd_cnt is 0
> > and last_fwd_cnt is 0xfffffffff then apparently delta is 0.
> 
> Seems yes, I need +1 here

And then you will get a nop, because assigning U32_MAX + 1 to u32
gives you 0. Adding () does nothing to change the result,
+ and - are commutative.


> > 
> > 
> >> +	else
> >> +		fwd_cnt_delta = vvs->fwd_cnt - vvs->last_fwd_cnt;
> > 
> > I actually don't see what is wrong with just
> > 	fwd_cnt_delta = vvs->fwd_cnt - vvs->last_fwd_cnt
> > 32 bit unsigned math will I think handle wrap around correctly.
> > 
> > And given buf_alloc is also u32 - I don't see where the bug is in
> > the original code.
> 
> I think problem is when fwd_cnt wraps, while last_fwd_cnt is not. In this
> case fwd_cnt_delta will be too big, so we won't send credit update which
> leads to stall for sender
> 
> Thanks, Arseniy

Care coming up with an example?


> > 
> > 
> >> +
> >> +	free_space = vvs->buf_alloc - fwd_cnt_delta;
> >> +	low_rx_bytes = (vvs->rx_bytes <
> >> +			sock_rcvlowat(sk_vsock(vsk), 0, INT_MAX));
> >>  
> >>  	spin_unlock_bh(&vvs->rx_lock);
> >>  
> >> @@ -611,9 +621,11 @@ virtio_transport_stream_do_dequeue(struct vsock_sock *vsk,
> >>  	 * too high causes extra messages. Too low causes transmitter
> >>  	 * stalls. As stalls are in theory more expensive than extra
> >>  	 * messages, we set the limit to a high value. TODO: experiment
> >> -	 * with different values.
> >> +	 * with different values. Also send credit update message when
> >> +	 * number of bytes in rx queue is not enough to wake up reader.
> >>  	 */
> >> -	if (free_space < VIRTIO_VSOCK_MAX_PKT_BUF_SIZE)
> >> +	if (fwd_cnt_delta &&
> >> +	    (free_space < VIRTIO_VSOCK_MAX_PKT_BUF_SIZE || low_rx_bytes))
> >>  		virtio_transport_send_credit_update(vsk);
> >>  
> >>  	return total;
> >> -- 
> >> 2.25.1
> >