Re: [PATCH v4 3/8] tcp: Rewind sequence when guest shrinks window to zero

[David Gibson <david@gibson.dropbear.id.au>]

[-- Attachment #1: Type: text/plain, Size: 5240 bytes --]

On Tue, Sep 09, 2025 at 08:16:50PM +0200, Stefano Brivio wrote:
> A window shrunk to zero means by definition that anything else that
> might be in flight is now out of window. Restart from the currently
> acknowledged sequence.
> 
> We need to do that both in tcp_tap_window_update(), where we already
> check for zero-window updates, as well as in tcp_data_from_tap(),
> because we might get one of those updates in a batch of packets that
> also contains a non-zero window update.
> 
> Suggested-by: Jon Maloy <jmaloy@redhat.com>
> Signed-off-by: Stefano Brivio <sbrivio@redhat.com>

Reviewed-by: David Gibson <david@gibson.dropbear.id.au>

Though a couple of documentation nits below.

> ---
>  tcp.c | 34 +++++++++++++++++++++++++---------
>  1 file changed, 25 insertions(+), 9 deletions(-)
> 
> diff --git a/tcp.c b/tcp.c
> index 86e08f1..12d42e0 100644
> --- a/tcp.c
> +++ b/tcp.c
> @@ -1268,19 +1268,25 @@ static void tcp_get_tap_ws(struct tcp_tap_conn *conn,
>  
>  /**
>   * tcp_tap_window_update() - Process an updated window from tap side
> + * @c:		Execution context
>   * @conn:	Connection pointer
>   * @wnd:	Window value, host order, unscaled
>   */
> -static void tcp_tap_window_update(struct tcp_tap_conn *conn, unsigned wnd)
> +static void tcp_tap_window_update(const struct ctx *c,
> +				  struct tcp_tap_conn *conn, unsigned wnd)
>  {
>  	wnd = MIN(MAX_WINDOW, wnd << conn->ws_from_tap);
>  
>  	/* Work-around for bug introduced in peer kernel code, commit
> -	 * e2142825c120 ("net: tcp: send zero-window ACK when no memory").
> -	 * We don't update if window shrank to zero.
> +	 * e2142825c120 ("net: tcp: send zero-window ACK when no memory"): don't
> +	 * update the window if it shrank to zero, so that we'll eventually
> +	 * retry to send data, but rewind the sequence as that obviously implies
> +	 * that no data beyond the updated window will ever be acknowledged.

As noted earlier "will ever be acknowledged" might be a bit
misleading.  Maybe "no data beyond the window will be acknowledged
until it is retransmitted".

>  	 */
> -	if (!wnd && SEQ_LT(conn->seq_ack_from_tap, conn->seq_to_tap))
> +	if (!wnd && SEQ_LT(conn->seq_ack_from_tap, conn->seq_to_tap)) {
> +		tcp_rewind_seq(c, conn);
>  		return;
> +	}
>  
>  	conn->wnd_from_tap = MIN(wnd >> conn->ws_from_tap, USHRT_MAX);
>  
> @@ -1709,7 +1715,8 @@ static int tcp_data_from_tap(const struct ctx *c, struct tcp_tap_conn *conn,
>  			tcp_timer_ctl(c, conn);
>  
>  			if (p->count == 1) {
> -				tcp_tap_window_update(conn, ntohs(th->window));
> +				tcp_tap_window_update(c, conn,
> +						      ntohs(th->window));
>  				return 1;
>  			}
>  
> @@ -1728,6 +1735,15 @@ static int tcp_data_from_tap(const struct ctx *c, struct tcp_tap_conn *conn,
>  				       ack_seq == max_ack_seq &&
>  				       ntohs(th->window) == max_ack_seq_wnd;
>  
> +				/* See tcp_tap_window_update() for details. On
> +				 * top of that, we also need to check here if a
> +				 * zero-window update is contained in a batch of
> +				 * packets that includes a non-zero window as
> +				 * well.

I'm not 100% convinced of this reasoning.  But at worst this should
result in some unnecessary but mostly harmless retransmits, and it
seems to fix the problem empirically, so I'm not suggesting changing
it at this time.

> +				 */
> +				if (!ntohs(th->window))
> +					tcp_rewind_seq(c, conn);
> +
>  				max_ack_seq_wnd = ntohs(th->window);
>  				max_ack_seq = ack_seq;
>  			}
> @@ -1791,7 +1807,7 @@ static int tcp_data_from_tap(const struct ctx *c, struct tcp_tap_conn *conn,
>  	if (ack && !tcp_sock_consume(conn, max_ack_seq))
>  		tcp_update_seqack_from_tap(c, conn, max_ack_seq);
>  
> -	tcp_tap_window_update(conn, max_ack_seq_wnd);
> +	tcp_tap_window_update(c, conn, max_ack_seq_wnd);
>  
>  	if (retr) {
>  		flow_trace(conn,
> @@ -1880,7 +1896,7 @@ static void tcp_conn_from_sock_finish(const struct ctx *c,
>  				      const struct tcphdr *th,
>  				      const char *opts, size_t optlen)
>  {
> -	tcp_tap_window_update(conn, ntohs(th->window));
> +	tcp_tap_window_update(c, conn, ntohs(th->window));
>  	tcp_get_tap_ws(conn, opts, optlen);
>  
>  	/* First value is not scaled */
> @@ -2085,7 +2101,7 @@ int tcp_tap_handler(const struct ctx *c, uint8_t pif, sa_family_t af,
>  		if (!th->ack)
>  			goto reset;
>  
> -		tcp_tap_window_update(conn, ntohs(th->window));
> +		tcp_tap_window_update(c, conn, ntohs(th->window));
>  
>  		tcp_data_from_sock(c, conn);
>  
> @@ -2097,7 +2113,7 @@ int tcp_tap_handler(const struct ctx *c, uint8_t pif, sa_family_t af,
>  	if (conn->events & TAP_FIN_RCVD) {
>  		tcp_sock_consume(conn, ntohl(th->ack_seq));
>  		tcp_update_seqack_from_tap(c, conn, ntohl(th->ack_seq));
> -		tcp_tap_window_update(conn, ntohs(th->window));
> +		tcp_tap_window_update(c, conn, ntohs(th->window));
>  		tcp_data_from_sock(c, conn);
>  
>  		if (conn->events & SOCK_FIN_RCVD &&
> -- 
> 2.43.0
> 

-- 
David Gibson (he or they)	| I'll have my music baroque, and my code
david AT gibson.dropbear.id.au	| minimalist, thank you, not the other way
				| around.
http://www.ozlabs.org/~dgibson

[-- Attachment #2: signature.asc --]
[-- Type: application/pgp-signature, Size: 833 bytes --]