On Thu, Dec 04, 2025 at 08:45:34AM +0100, Stefano Brivio wrote:
> For non-local connections, we advertise the same window size as what
> the peer in turn advertises to us, and limit it to the buffer size
> reported via SO_SNDBUF.
> 
> That's not quite correct: in order to later avoid failures while
> queueing data to the socket, we need to limit the window to the
> available buffer size, not the total one.
> 
> Use the SIOCOUTQ ioctl and subtract the number of outbound queued
> bytes from the total buffer size, then clamp to this value.
> 
> Signed-off-by: Stefano Brivio <sbrivio@redhat.com>

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

> ---
>  README.md |  2 +-
>  tcp.c     | 18 ++++++++++++++++--
>  2 files changed, 17 insertions(+), 3 deletions(-)
> 
> diff --git a/README.md b/README.md
> index 897ae8b..8fdc0a3 100644
> --- a/README.md
> +++ b/README.md
> @@ -291,7 +291,7 @@ speeding up local connections, and usually requiring NAT. _pasta_:
>  * ✅ all capabilities dropped, other than `CAP_NET_BIND_SERVICE` (if granted)
>  * ✅ with default options, user, mount, IPC, UTS, PID namespaces are detached
>  * ✅ no external dependencies (other than a standard C library)
> -* ✅ restrictive seccomp profiles (33 syscalls allowed for _passt_, 43 for
> +* ✅ restrictive seccomp profiles (34 syscalls allowed for _passt_, 43 for
>    _pasta_ on x86_64)
>  * ✅ examples of [AppArmor](/passt/tree/contrib/apparmor) and
>    [SELinux](/passt/tree/contrib/selinux) profiles available
> diff --git a/tcp.c b/tcp.c
> index fa95f6b..863ccdb 100644
> --- a/tcp.c
> +++ b/tcp.c
> @@ -1031,6 +1031,8 @@ void tcp_fill_headers(const struct ctx *c, struct tcp_tap_conn *conn,
>   * @tinfo:	tcp_info from kernel, can be NULL if not pre-fetched
>   *
>   * Return: 1 if sequence or window were updated, 0 otherwise
> + *
> + * #syscalls ioctl
>   */
>  int tcp_update_seqack_wnd(const struct ctx *c, struct tcp_tap_conn *conn,
>  			  bool force_seq, struct tcp_info_linux *tinfo)
> @@ -1113,9 +1115,21 @@ int tcp_update_seqack_wnd(const struct ctx *c, struct tcp_tap_conn *conn,
>  	if ((conn->flags & LOCAL) || tcp_rtt_dst_low(conn)) {
>  		new_wnd_to_tap = tinfo->tcpi_snd_wnd;
>  	} else {
> +		uint32_t sendq;
> +		int limit;
> +
> +		if (ioctl(s, SIOCOUTQ, &sendq)) {
> +			debug_perror("SIOCOUTQ on socket %i, assuming 0", s);
> +			sendq = 0;
> +		}
>  		tcp_get_sndbuf(conn);
> -		new_wnd_to_tap = MIN((int)tinfo->tcpi_snd_wnd,
> -				     SNDBUF_GET(conn));
> +
> +		if ((int)sendq > SNDBUF_GET(conn)) /* Due to memory pressure? */
> +			limit = 0;
> +		else
> +			limit = SNDBUF_GET(conn) - (int)sendq;
> +
> +		new_wnd_to_tap = MIN((int)tinfo->tcpi_snd_wnd, limit);
>  	}
>  
>  	new_wnd_to_tap = MIN(new_wnd_to_tap, MAX_WINDOW);
> -- 
> 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