Re: [PATCH 2/3] migrate, tcp: Migrate RFC7323 timestamp

[Stefano Brivio <sbrivio@redhat.com>]

On Wed, 19 Mar 2025 16:14:22 +1100
David Gibson <david@gibson.dropbear.id.au> wrote:

> Currently our migration of the state of TCP sockets omits the RFC7323
> timestamp.  In some circumstances that can result in data sent from the
> target machine not being received, because it is discarded on the peer due
> to PAWS checking.
> 
> Add code to dump and restore the timestamp across migration.
> 
> Link: https://bugs.passt.top/show_bug.cgi?id=115
> 
> Signed-off-by: David Gibson <david@gibson.dropbear.id.au>

Sneakily changed all "RFC7323" references to the actual name of the
document, "RFC 7323", and dropped:

> ---
>  tcp.c      | 59 ++++++++++++++++++++++++++++++++++++++++++++++++++++++
>  tcp_conn.h |  2 ++
>  2 files changed, 61 insertions(+)
> 
> diff --git a/tcp.c b/tcp.c
> index 163ddd60..56df1636 100644
> --- a/tcp.c
> +++ b/tcp.c
> @@ -2861,6 +2861,57 @@ static int tcp_flow_dump_mss(const struct tcp_tap_conn *conn,
>  	return 0;
>  }
>  
> +
> +/**
> + * tcp_flow_dump_timestamp() - Dump RFC7323 timestamp via TCP_TIMESTAMP
> + * @conn:	Pointer to the TCP connection structure
> + * @t:		Extended migration data (tcpi_options must be populated)
> + *
> + * Return: 0 on success, negative error code on failure
> + */
> +static int tcp_flow_dump_timestamp(const struct tcp_tap_conn *conn,
> +				   struct tcp_tap_transfer_ext *t)
> +{
> +	int val = 0;
> +
> +	if (t->tcpi_options & TCPI_OPT_TIMESTAMPS) {
> +		socklen_t sl = sizeof(val);
> +
> +		if (getsockopt(conn->sock, SOL_TCP, TCP_TIMESTAMP, &val, &sl)) {
> +			int rc = -errno;
> +			flow_perror(conn, "Getting RFC7323 timestamp");
> +			return rc;
> +		}
> +	}
> +
> +	t->timestamp = (uint32_t)val;
> +	return 0;
> +}
> +
> +/**
> + * tcp_flow_repair_timestamp() - Restore RFC7323 timestamp via TCP_TIMESTAMP
> + * @conn:	Pointer to the TCP connection structure
> + * @t:		Extended migration data
> + *
> + * Return: 0 on success, negative error code on failure
> + */
> +static int tcp_flow_repair_timestamp(const struct tcp_tap_conn *conn,
> +				   const struct tcp_tap_transfer_ext *t)
> +{
> +	int val = (int)t->timestamp;
> +
> +	if (t->tcpi_options & TCPI_OPT_TIMESTAMPS) {
> +		if (setsockopt(conn->sock, SOL_TCP, TCP_TIMESTAMP,
> +			       &val, sizeof(val))) {
> +			int rc = -errno;
> +			flow_perror(conn, "Setting RFC7323 timestamp");
> +			return rc;
> +		}
> +	}
> +
> +	return 0;
> +}
> +
>  /**
>   * tcp_flow_dump_wnd() - Dump current tcp_repair_window parameters
>   * @conn:	Pointer to the TCP connection structure
> @@ -3260,6 +3311,9 @@ int tcp_flow_migrate_source_ext(int fd, const struct tcp_tap_conn *conn)
>  	if ((rc = tcp_flow_dump_mss(conn, t)))
>  		goto fail;
>  
> +	if ((rc = tcp_flow_dump_timestamp(conn, t)))
> +		goto fail;
> +
>  	if ((rc = tcp_flow_dump_wnd(conn, t)))
>  		goto fail;
>  
> @@ -3305,6 +3359,7 @@ int tcp_flow_migrate_source_ext(int fd, const struct tcp_tap_conn *conn)
>  	t->notsent	= htonl(t->notsent);
>  	t->rcvq		= htonl(t->rcvq);
>  	t->mss		= htonl(t->mss);
> +	t->timestamp	= htonl(t->timestamp);
>  
>  	t->snd_wl1	= htonl(t->snd_wl1);
>  	t->snd_wnd	= htonl(t->snd_wnd);
> @@ -3519,6 +3574,7 @@ int tcp_flow_migrate_target_ext(struct ctx *c, struct tcp_tap_conn *conn, int fd
>  	t.notsent	= ntohl(t.notsent);
>  	t.rcvq		= ntohl(t.rcvq);
>  	t.mss		= ntohl(t.mss);
> +	t.timestamp	= ntohl(t.timestamp);
>  
>  	t.snd_wl1	= ntohl(t.snd_wl1);
>  	t.snd_wnd	= ntohl(t.snd_wnd);
> @@ -3561,6 +3617,9 @@ int tcp_flow_migrate_target_ext(struct ctx *c, struct tcp_tap_conn *conn, int fd
>  		/* We weren't able to create the socket, discard flow */
>  		goto fail;
>  
> +	if (tcp_flow_repair_timestamp(conn, &t))
> +		goto fail;
> +	

...this stray tab. Series applied.

>  	if (tcp_flow_select_queue(conn, TCP_SEND_QUEUE))
>  		goto fail;
>  
> diff --git a/tcp_conn.h b/tcp_conn.h
> index 9126a36f..c79f558b 100644
> --- a/tcp_conn.h
> +++ b/tcp_conn.h
> @@ -152,6 +152,7 @@ struct tcp_tap_transfer {
>   * @notsent:		Part of pending send queue that wasn't sent out yet
>   * @rcvq:		Length of pending receive queue
>   * @mss:		Socket-side MSS clamp
> + * @timestamp:		RFC7323 timestamp
>   * @snd_wl1:		Next sequence used in window probe (next sequence - 1)
>   * @snd_wnd:		Socket-side sending window
>   * @max_window:		Window clamp
> @@ -171,6 +172,7 @@ struct tcp_tap_transfer_ext {
>  	uint32_t	rcvq;
>  
>  	uint32_t	mss;
> +	uint32_t	timestamp;
>  
>  	/* We can't just use struct tcp_repair_window: we need network order */
>  	uint32_t	snd_wl1;

-- 
Stefano