public inbox for passt-dev@passt.top
 help / color / mirror / code / Atom feed
* [PATCH v2] tcp: Limit fast retransmit to one per loss event
@ 2026-11-20  7:43 Anshu Kumari
  2026-11-20  9:07 ` Stefano Brivio
  0 siblings, 1 reply; 2+ messages in thread
From: Anshu Kumari @ 2026-11-20  7:43 UTC (permalink / raw)
  To: ada.urbaniec+bugzilla, passt-dev, sbrivio; +Cc: anskuma

Every duplicate ACK unconditionally triggered a fast retransmit,
even when the ACK sequence hadn't advanced.

Add a dup_acks counter (as a union with retries) to
tcp_tap_conn and tcp_tap_transfer that gets incremented on the
first fast retransmit and cleared when the ACK sequence
advances.

This is a quick fix to stop the amplification storm. Full
compliance with RFC 5681 Section 3.2 (three duplicate ACKs)
is left to a later patch.

Link: https://bugs.passt.top/show_bug.cgi?id=224
Signed-off-by: Anshu Kumari <anskuma@redhat.com>
---
v2:
  - added union for retries and dup_acks.
  - removed retransmitting flag.
---
 tcp.c      | 12 ++++++++----
 tcp_conn.h | 16 ++++++++++++----
 2 files changed, 20 insertions(+), 8 deletions(-)

diff --git a/tcp.c b/tcp.c
index 95f4f3e1..d205cd41 100644
--- a/tcp.c
+++ b/tcp.c
@@ -242,7 +242,7 @@
  *     - consume buffer by difference between new ack_seq and @seq_ack_from_tap
  *     - update @seq_ack_from_tap from ack_seq in header
  *     - on two duplicated ACKs, reset @seq_to_tap to @seq_ack_from_tap, and
- *       resend with steps listed above
+ *       resend with steps listed above, once per loss event
  *
  * - from tap/guest to socket:
  *   - on packet from tap/guest:
@@ -1263,7 +1263,7 @@ static void tcp_update_seqack_from_tap(const struct ctx *c,
 		if (SEQ_LT(seq, conn->seq_to_tap))
 			conn_flag(c, conn, ACK_FROM_TAP_DUE, now);
 
-		conn->retries = 0;
+		conn->dup_acks = 0;
 		conn->seq_ack_from_tap = seq;
 	}
 }
@@ -2082,7 +2082,9 @@ static int tcp_data_from_tap(const struct ctx *c, struct tcp_tap_conn *conn,
 
 	tcp_tap_window_update(c, conn, max_ack_seq_wnd, now);
 
-	if (retr) {
+	if (retr && !conn->dup_acks) {
+		conn->dup_acks++;
+
 		flow_trace(conn,
 			   "fast re-transmit, ACK: %u, previous sequence: %u",
 			   conn->seq_ack_from_tap, conn->seq_to_tap);
@@ -2443,7 +2445,9 @@ int tcp_tap_handler(const struct ctx *c, uint8_t pif, sa_family_t af,
 			tcp_update_seqack_from_tap(c, conn, ntohl(th->ack_seq),
 						   now);
 
-		if (retr) {
+		if (retr && !conn->dup_acks) {
+			conn->dup_acks++;
+
 			flow_trace(conn,
 				   "fast re-transmit, ACK: %u, previous sequence: %u",
 				   ntohl(th->ack_seq), conn->seq_to_tap);
diff --git a/tcp_conn.h b/tcp_conn.h
index 3173ddb9..ba175a3e 100644
--- a/tcp_conn.h
+++ b/tcp_conn.h
@@ -16,7 +16,8 @@
 /**
  * struct tcp_tap_conn - Descriptor for a TCP connection (not spliced)
  * @f:			Generic flow information
- * @retries:		Number of retries occurred due to timeouts
+ * @retries:		Number of retries due to timeouts
+ * @dup_acks:		Number of duplicate ACKs (union with @retries)
  * @ws_from_tap:	Window scaling factor advertised from tap/guest
  * @ws_to_tap:		Window scaling factor advertised to tap/guest
  * @tap_mss:		MSS advertised by tap/guest, rounded to 2 ^ TCP_MSS_BITS
@@ -41,7 +42,10 @@ struct tcp_tap_conn {
 	struct flow_common f;
 
 #define TCP_RETRIES_BITS		3
-	unsigned int	retries		:TCP_RETRIES_BITS;
+	union {
+		unsigned int	retries		:TCP_RETRIES_BITS;
+		unsigned int	dup_acks	:TCP_RETRIES_BITS;
+	};
 #define TCP_MAX_RETRIES			MAX_FROM_BITS(TCP_RETRIES_BITS)
 
 #define TCP_WS_BITS			4	/* RFC 7323 */
@@ -122,7 +126,8 @@ struct tcp_tap_conn {
  * struct tcp_tap_transfer - Migrated TCP data, flow table part, network order
  * @pif:		Interfaces for each side of the flow
  * @side:		Addresses and ports for each side of the flow
- * @retries:		Number of retries occurred due to timeouts
+ * @retries:		Number of retries due to timeouts
+ * @dup_acks:		Number of duplicate ACKs (union with @retries)
  * @ws_from_tap:	Window scaling factor advertised from tap/guest
  * @ws_to_tap:		Window scaling factor advertised to tap/guest
  * @events:		Connection events, implying connection states
@@ -142,7 +147,10 @@ struct tcp_tap_transfer {
 	uint8_t		pif[SIDES];
 	struct flowside	side[SIDES];
 
-	uint8_t		retries;
+	union {
+		uint8_t		retries;
+		uint8_t		dup_acks;
+	};
 	uint8_t		ws_from_tap;
 	uint8_t		ws_to_tap;
 	uint8_t		events;
-- 
2.55.0


^ permalink raw reply	[flat|nested] 2+ messages in thread

* Re: [PATCH v2] tcp: Limit fast retransmit to one per loss event
  2026-11-20  7:43 [PATCH v2] tcp: Limit fast retransmit to one per loss event Anshu Kumari
@ 2026-11-20  9:07 ` Stefano Brivio
  0 siblings, 0 replies; 2+ messages in thread
From: Stefano Brivio @ 2026-11-20  9:07 UTC (permalink / raw)
  To: Anshu Kumari; +Cc: ada.urbaniec+bugzilla, passt-dev

On Fri, 20 Nov 2026 13:13:04 +0530
Anshu Kumari <anskuma@redhat.com> wrote:

> Every duplicate ACK unconditionally triggered a fast retransmit,
> even when the ACK sequence hadn't advanced.
> 
> Add a dup_acks counter (as a union with retries) to
> tcp_tap_conn and tcp_tap_transfer that gets incremented on the
> first fast retransmit and cleared when the ACK sequence
> advances.

I still can't understand this part.  Why do we need to do this in this
patch? That is, in detail:

> 
> This is a quick fix to stop the amplification storm. Full
> compliance with RFC 5681 Section 3.2 (three duplicate ACKs)
> is left to a later patch.
> 
> Link: https://bugs.passt.top/show_bug.cgi?id=224
> Signed-off-by: Anshu Kumari <anskuma@redhat.com>
> ---
> v2:
>   - added union for retries and dup_acks.
>   - removed retransmitting flag.
> ---
>  tcp.c      | 12 ++++++++----
>  tcp_conn.h | 16 ++++++++++++----
>  2 files changed, 20 insertions(+), 8 deletions(-)
> 
> diff --git a/tcp.c b/tcp.c
> index 95f4f3e1..d205cd41 100644
> --- a/tcp.c
> +++ b/tcp.c
> @@ -242,7 +242,7 @@
>   *     - consume buffer by difference between new ack_seq and @seq_ack_from_tap
>   *     - update @seq_ack_from_tap from ack_seq in header
>   *     - on two duplicated ACKs, reset @seq_to_tap to @seq_ack_from_tap, and
> - *       resend with steps listed above
> + *       resend with steps listed above, once per loss event
>   *
>   * - from tap/guest to socket:
>   *   - on packet from tap/guest:
> @@ -1263,7 +1263,7 @@ static void tcp_update_seqack_from_tap(const struct ctx *c,
>  		if (SEQ_LT(seq, conn->seq_to_tap))
>  			conn_flag(c, conn, ACK_FROM_TAP_DUE, now);
>  
> -		conn->retries = 0;
> +		conn->dup_acks = 0;

We're not actually counting them, so this isn't entirely true (and in
any case not needed in this patch).

>  		conn->seq_ack_from_tap = seq;
>  	}
>  }
> @@ -2082,7 +2082,9 @@ static int tcp_data_from_tap(const struct ctx *c, struct tcp_tap_conn *conn,
>  
>  	tcp_tap_window_update(c, conn, max_ack_seq_wnd, now);
>  
> -	if (retr) {
> +	if (retr && !conn->dup_acks) {

This isn't about the number of duplicate ACKs.  What I was suggesting
in my draft patch on https://bugs.passt.top/show_bug.cgi?id=224 is:

	if (retr && !conn->retries)

meaning: we have a request for fast retransmit, and we didn't
retransmit yet.

> +		conn->dup_acks++;

This isn't true either.  We can't count them yet, with this patch.  I
think this should be conn->retries++ (or conn->retries = 1).

> +
>  		flow_trace(conn,
>  			   "fast re-transmit, ACK: %u, previous sequence: %u",
>  			   conn->seq_ack_from_tap, conn->seq_to_tap);
> @@ -2443,7 +2445,9 @@ int tcp_tap_handler(const struct ctx *c, uint8_t pif, sa_family_t af,
>  			tcp_update_seqack_from_tap(c, conn, ntohl(th->ack_seq),
>  						   now);
>  
> -		if (retr) {
> +		if (retr && !conn->dup_acks) {

Same as above, I think this should be if (retr && !conn->retries).

> +			conn->dup_acks++;

Same as above, I think this should be conn->retries++.

> +
>  			flow_trace(conn,
>  				   "fast re-transmit, ACK: %u, previous sequence: %u",
>  				   ntohl(th->ack_seq), conn->seq_to_tap);
> diff --git a/tcp_conn.h b/tcp_conn.h
> index 3173ddb9..ba175a3e 100644
> --- a/tcp_conn.h
> +++ b/tcp_conn.h
> @@ -16,7 +16,8 @@
>  /**
>   * struct tcp_tap_conn - Descriptor for a TCP connection (not spliced)
>   * @f:			Generic flow information
> - * @retries:		Number of retries occurred due to timeouts
> + * @retries:		Number of retries due to timeouts
> + * @dup_acks:		Number of duplicate ACKs (union with @retries)

Not true, at the moment.  This is simply @retries with this patch.
This change is not needed.

>   * @ws_from_tap:	Window scaling factor advertised from tap/guest
>   * @ws_to_tap:		Window scaling factor advertised to tap/guest
>   * @tap_mss:		MSS advertised by tap/guest, rounded to 2 ^ TCP_MSS_BITS
> @@ -41,7 +42,10 @@ struct tcp_tap_conn {
>  	struct flow_common f;
>  
>  #define TCP_RETRIES_BITS		3
> -	unsigned int	retries		:TCP_RETRIES_BITS;
> +	union {
> +		unsigned int	retries		:TCP_RETRIES_BITS;
> +		unsigned int	dup_acks	:TCP_RETRIES_BITS;
> +	};

Same here, I don't think this part is needed at this stage.

>  #define TCP_MAX_RETRIES			MAX_FROM_BITS(TCP_RETRIES_BITS)
>  
>  #define TCP_WS_BITS			4	/* RFC 7323 */
> @@ -122,7 +126,8 @@ struct tcp_tap_conn {
>   * struct tcp_tap_transfer - Migrated TCP data, flow table part, network order
>   * @pif:		Interfaces for each side of the flow
>   * @side:		Addresses and ports for each side of the flow
> - * @retries:		Number of retries occurred due to timeouts
> + * @retries:		Number of retries due to timeouts
> + * @dup_acks:		Number of duplicate ACKs (union with @retries)

Same here (not needed).

>   * @ws_from_tap:	Window scaling factor advertised from tap/guest
>   * @ws_to_tap:		Window scaling factor advertised to tap/guest
>   * @events:		Connection events, implying connection states
> @@ -142,7 +147,10 @@ struct tcp_tap_transfer {
>  	uint8_t		pif[SIDES];
>  	struct flowside	side[SIDES];
>  
> -	uint8_t		retries;
> +	union {
> +		uint8_t		retries;
> +		uint8_t		dup_acks;
> +	};

And here (not needed).

>  	uint8_t		ws_from_tap;
>  	uint8_t		ws_to_tap;
>  	uint8_t		events;

-- 
Stefano


^ permalink raw reply	[flat|nested] 2+ messages in thread

end of thread, other threads:[~2026-11-20  9:07 UTC | newest]

Thread overview: 2+ messages (download: mbox.gz / follow: Atom feed)
-- links below jump to the message on this page --
2026-11-20  7:43 [PATCH v2] tcp: Limit fast retransmit to one per loss event Anshu Kumari
2026-11-20  9:07 ` Stefano Brivio

Code repositories for project(s) associated with this public inbox

	https://passt.top/passt

This is a public inbox, see mirroring instructions
for how to clone and mirror all data and code used for this inbox;
as well as URLs for IMAP folder(s).