[PATCH 1/2] tcp: fix undo after RTO for BIC

This patch fixes BIC so that cwnd reductions made during RTOs can be
undone (just as they already can be undone when using the default/Reno
behavior).

When undoing cwnd reductions, BIC-derived congestion control modules
were restoring the cwnd from last_max_cwnd. There were two problems
with using last_max_cwnd to restore a cwnd during undo:

(a) last_max_cwnd was set to 0 on state transitions into TCP_CA_Loss
(by calling the module's reset() functions), so cwnd reductions from
RTOs could not be undone.

(b) when fast_covergence is enabled (which it is by default)
last_max_cwnd does not actually hold the value of snd_cwnd before the
loss; instead, it holds a scaled-down version of snd_cwnd.

This patch makes the following changes:

(1) upon undo, revert snd_cwnd to ca->loss_cwnd, which is already, as
the existing comment notes, the "congestion window at last loss"

(2) stop forgetting ca->loss_cwnd on TCP_CA_Loss events

(3) use ca->last_max_cwnd to check if we're in slow start

Signed-off-by: Neal Cardwell <ncardwell@google.com>
---
 net/ipv4/tcp_bic.c |   11 +++++++----
 1 files changed, 7 insertions(+), 4 deletions(-)

diff --git a/net/ipv4/tcp_bic.c b/net/ipv4/tcp_bic.c
index 6187eb4..f45e1c2 100644
--- a/net/ipv4/tcp_bic.c
+++ b/net/ipv4/tcp_bic.c

@@ -63,7 +63,6 @@ static inline void bictcp_reset(struct bictcp *ca)
 {
 	ca->cnt = 0;
 	ca->last_max_cwnd = 0;
-	ca->loss_cwnd = 0;
 	ca->last_cwnd = 0;
 	ca->last_time = 0;
 	ca->epoch_start = 0;

@@ -72,7 +71,11 @@ static inline void bictcp_reset(struct bictcp *ca)
 
 static void bictcp_init(struct sock *sk)
 {
-	bictcp_reset(inet_csk_ca(sk));
+	struct bictcp *ca = inet_csk_ca(sk);
+
+	bictcp_reset(ca);
+	ca->loss_cwnd = 0;
+
 	if (initial_ssthresh)
 		tcp_sk(sk)->snd_ssthresh = initial_ssthresh;
 }

@@ -127,7 +130,7 @@ static inline void bictcp_update(struct bictcp *ca, u32 cwnd)
 	}
 
 	/* if in slow start or link utilization is very low */
-	if (ca->loss_cwnd == 0) {
+	if (ca->last_max_cwnd == 0) {
 		if (ca->cnt > 20) /* increase cwnd 5% per RTT */
 			ca->cnt = 20;
 	}

@@ -185,7 +188,7 @@ static u32 bictcp_undo_cwnd(struct sock *sk)
 {
 	const struct tcp_sock *tp = tcp_sk(sk);
 	const struct bictcp *ca = inet_csk_ca(sk);
-	return max(tp->snd_cwnd, ca->last_max_cwnd);
+	return max(tp->snd_cwnd, ca->loss_cwnd);
 }
 
 static void bictcp_state(struct sock *sk, u8 new_state)

-- 
1.7.7.3

This patch fixes CUBIC so that cwnd reductions made during RTOs can be
undone (just as they already can be undone when using the default/Reno
behavior).

When undoing cwnd reductions, BIC-derived congestion control modules
were restoring the cwnd from last_max_cwnd. There were two problems
with using last_max_cwnd to restore a cwnd during undo:

(a) last_max_cwnd was set to 0 on state transitions into TCP_CA_Loss
(by calling the module's reset() functions), so cwnd reductions from
RTOs could not be undone.

(b) when fast_covergence is enabled (which it is by default)
last_max_cwnd does not actually hold the value of snd_cwnd before the
loss; instead, it holds a scaled-down version of snd_cwnd.

This patch makes the following changes:

(1) upon undo, revert snd_cwnd to ca->loss_cwnd, which is already, as
the existing comment notes, the "congestion window at last loss"

(2) stop forgetting ca->loss_cwnd on TCP_CA_Loss events

(3) use ca->last_max_cwnd to check if we're in slow start

Signed-off-by: Neal Cardwell <ncardwell@google.com>
---
 net/ipv4/tcp_cubic.c |   10 ++++++----
 1 files changed, 6 insertions(+), 4 deletions(-)

diff --git a/net/ipv4/tcp_cubic.c b/net/ipv4/tcp_cubic.c
index f376b05..a9077f4 100644
--- a/net/ipv4/tcp_cubic.c
+++ b/net/ipv4/tcp_cubic.c

@@ -107,7 +107,6 @@ static inline void bictcp_reset(struct bictcp *ca)
 {
 	ca->cnt = 0;
 	ca->last_max_cwnd = 0;
-	ca->loss_cwnd = 0;
 	ca->last_cwnd = 0;
 	ca->last_time = 0;
 	ca->bic_origin_point = 0;

@@ -142,7 +141,10 @@ static inline void bictcp_hystart_reset(struct sock *sk)
 
 static void bictcp_init(struct sock *sk)
 {
-	bictcp_reset(inet_csk_ca(sk));
+	struct bictcp *ca = inet_csk_ca(sk);
+
+	bictcp_reset(ca);
+	ca->loss_cwnd = 0;
 
 	if (hystart)
 		bictcp_hystart_reset(sk);

@@ -275,7 +277,7 @@ static inline void bictcp_update(struct bictcp *ca, u32 cwnd)
 	 * The initial growth of cubic function may be too conservative
 	 * when the available bandwidth is still unknown.
 	 */
-	if (ca->loss_cwnd == 0 && ca->cnt > 20)
+	if (ca->last_max_cwnd == 0 && ca->cnt > 20)
 		ca->cnt = 20;	/* increase cwnd 5% per RTT */
 
 	/* TCP Friendly */

@@ -342,7 +344,7 @@ static u32 bictcp_undo_cwnd(struct sock *sk)
 {
 	struct bictcp *ca = inet_csk_ca(sk);
 
-	return max(tcp_sk(sk)->snd_cwnd, ca->last_max_cwnd);
+	return max(tcp_sk(sk)->snd_cwnd, ca->loss_cwnd);
 }
 
 static void bictcp_state(struct sock *sk, u8 new_state)

-- 
1.7.7.3

On Wed, 18 Jan 2012 22:47:59 -0500
Neal Cardwell [off-list ref] wrote:

This patch fixes CUBIC so that cwnd reductions made during RTOs can be
undone (just as they already can be undone when using the default/Reno
behavior).

When undoing cwnd reductions, BIC-derived congestion control modules
were restoring the cwnd from last_max_cwnd. There were two problems
with using last_max_cwnd to restore a cwnd during undo:

(a) last_max_cwnd was set to 0 on state transitions into TCP_CA_Loss
(by calling the module's reset() functions), so cwnd reductions from
RTOs could not be undone.

(b) when fast_covergence is enabled (which it is by default)
last_max_cwnd does not actually hold the value of snd_cwnd before the
loss; instead, it holds a scaled-down version of snd_cwnd.

This patch makes the following changes:

(1) upon undo, revert snd_cwnd to ca->loss_cwnd, which is already, as
the existing comment notes, the "congestion window at last loss"

(2) stop forgetting ca->loss_cwnd on TCP_CA_Loss events

(3) use ca->last_max_cwnd to check if we're in slow start

Signed-off-by: Neal Cardwell <ncardwell@google.com>

I was unsure of this, so forwarded it to the author 
and received his confirmation.

Acked-by: Stephen Hemminger <redacted>
Acked-by: Sangtae Ha <redacted>

From: Neal Cardwell <ncardwell@google.com>
Date: Wed, 18 Jan 2012 22:47:58 -0500

This patch fixes BIC so that cwnd reductions made during RTOs can be
undone (just as they already can be undone when using the default/Reno
behavior).

When undoing cwnd reductions, BIC-derived congestion control modules
were restoring the cwnd from last_max_cwnd. There were two problems
with using last_max_cwnd to restore a cwnd during undo:

(a) last_max_cwnd was set to 0 on state transitions into TCP_CA_Loss
(by calling the module's reset() functions), so cwnd reductions from
RTOs could not be undone.

(b) when fast_covergence is enabled (which it is by default)
last_max_cwnd does not actually hold the value of snd_cwnd before the
loss; instead, it holds a scaled-down version of snd_cwnd.

This patch makes the following changes:

(1) upon undo, revert snd_cwnd to ca->loss_cwnd, which is already, as
the existing comment notes, the "congestion window at last loss"

(2) stop forgetting ca->loss_cwnd on TCP_CA_Loss events

(3) use ca->last_max_cwnd to check if we're in slow start

Signed-off-by: Neal Cardwell <ncardwell@google.com>

Applied.

From: Stephen Hemminger <redacted>
Date: Thu, 19 Jan 2012 18:49:21 -0800

On Wed, 18 Jan 2012 22:47:59 -0500
Neal Cardwell [off-list ref] wrote:

quoted

This patch fixes CUBIC so that cwnd reductions made during RTOs can be
undone (just as they already can be undone when using the default/Reno
behavior).

When undoing cwnd reductions, BIC-derived congestion control modules
were restoring the cwnd from last_max_cwnd. There were two problems
with using last_max_cwnd to restore a cwnd during undo:

(a) last_max_cwnd was set to 0 on state transitions into TCP_CA_Loss
(by calling the module's reset() functions), so cwnd reductions from
RTOs could not be undone.

(b) when fast_covergence is enabled (which it is by default)
last_max_cwnd does not actually hold the value of snd_cwnd before the
loss; instead, it holds a scaled-down version of snd_cwnd.

This patch makes the following changes:

(1) upon undo, revert snd_cwnd to ca->loss_cwnd, which is already, as
the existing comment notes, the "congestion window at last loss"

(2) stop forgetting ca->loss_cwnd on TCP_CA_Loss events

(3) use ca->last_max_cwnd to check if we're in slow start

Signed-off-by: Neal Cardwell <ncardwell@google.com>

I was unsure of this, so forwarded it to the author 
and received his confirmation.

Acked-by: Stephen Hemminger <redacted>
Acked-by: Sangtae Ha <redacted>

Applied.