[PATCH net-next] netem: add ECN capability

From: Eric Dumazet <edumazet@google.com>

Add ECN (Explicit Congestion Notification) marking capability to netem

tc qdisc add dev eth0 root netem drop 0.5 ecn

Instead of dropping packets, try to ECN mark them.

Signed-off-by: Eric Dumazet <edumazet@google.com>
Cc: Neal Cardwell <ncardwell@google.com>
Cc: Tom Herbert <redacted>
Cc: Hagen Paul Pfeifer <redacted>
Cc: Stephen Hemminger <redacted>
---
 include/linux/pkt_sched.h |    1 +
 net/sched/sch_netem.c     |   18 +++++++++++++++---
 2 files changed, 16 insertions(+), 3 deletions(-)

diff --git a/include/linux/pkt_sched.h b/include/linux/pkt_sched.h
index 410b33d..ffe975c 100644
--- a/include/linux/pkt_sched.h
+++ b/include/linux/pkt_sched.h

@@ -509,6 +509,7 @@ enum {
 	TCA_NETEM_CORRUPT,
 	TCA_NETEM_LOSS,
 	TCA_NETEM_RATE,
+	TCA_NETEM_ECN,
 	__TCA_NETEM_MAX,
 };
 

diff --git a/net/sched/sch_netem.c b/net/sched/sch_netem.c
index 1109731..231cd11 100644
--- a/net/sched/sch_netem.c
+++ b/net/sched/sch_netem.c

@@ -26,6 +26,7 @@
 
 #include <net/netlink.h>
 #include <net/pkt_sched.h>
+#include <net/inet_ecn.h>
 
 #define VERSION "1.3"
 

@@ -78,6 +79,7 @@ struct netem_sched_data {
 	psched_tdiff_t jitter;
 
 	u32 loss;
+	u32 ecn;
 	u32 limit;
 	u32 counter;
 	u32 gap;

@@ -374,9 +376,12 @@ static int netem_enqueue(struct sk_buff *skb, struct Qdisc *sch)
 		++count;
 
 	/* Drop packet? */
-	if (loss_event(q))
-		--count;
-
+	if (loss_event(q)) {
+		if (q->ecn && INET_ECN_set_ce(skb))
+			sch->qstats.drops++; /* mark packet */
+		else
+			--count;
+	}
 	if (count == 0) {
 		sch->qstats.drops++;
 		kfree_skb(skb);

@@ -706,6 +711,7 @@ static const struct nla_policy netem_policy[TCA_NETEM_MAX + 1] = {
 	[TCA_NETEM_CORRUPT]	= { .len = sizeof(struct tc_netem_corrupt) },
 	[TCA_NETEM_RATE]	= { .len = sizeof(struct tc_netem_rate) },
 	[TCA_NETEM_LOSS]	= { .type = NLA_NESTED },
+	[TCA_NETEM_ECN]		= { .type = NLA_U32 },
 };
 
 static int parse_attr(struct nlattr *tb[], int maxtype, struct nlattr *nla,

@@ -776,6 +782,9 @@ static int netem_change(struct Qdisc *sch, struct nlattr *opt)
 	if (tb[TCA_NETEM_RATE])
 		get_rate(sch, tb[TCA_NETEM_RATE]);
 
+	if (tb[TCA_NETEM_ECN])
+		q->ecn = nla_get_u32(tb[TCA_NETEM_ECN]);
+
 	q->loss_model = CLG_RANDOM;
 	if (tb[TCA_NETEM_LOSS])
 		ret = get_loss_clg(sch, tb[TCA_NETEM_LOSS]);

@@ -902,6 +911,9 @@ static int netem_dump(struct Qdisc *sch, struct sk_buff *skb)
 	if (nla_put(skb, TCA_NETEM_RATE, sizeof(rate), &rate))
 		goto nla_put_failure;
 
+	if (q->ecn && nla_put_u32(skb, TCA_NETEM_ECN, q->ecn))
+		goto nla_put_failure;
+
 	if (dump_loss_model(q, skb) != 0)
 		goto nla_put_failure;
 

Nice!


* Eric Dumazet | 2012-05-01 11:11:05 [+0200]:

	u32 loss;
+	u32 ecn;

Why no bool here Eric?

This netem API, what a heck ... Ideal we would extend TCA_NETEM_LOSS
somehow[tm], but yes ... ;(

Acked-by: Hagen Paul Pfeifer <redacted>

On Tue, May 1, 2012 at 2:28 AM, Hagen Paul Pfeifer [off-list ref] wrote:

Nice!

Applause as well! While this is something of a fork of this
discussion, I would rather like to resurrect a variant of the old
'time in queue' patch...

http://patchwork.ozlabs.org/patch/125329/

As best as I recall we'd run into trouble with combining things like
netem and complex qdiscs, because we had basically overrun the size of
the control block.

And then there was a long discussion of all that and I forget if any
conclusion was reached.

An alternative to reducing the length of the control block was to find
some way to leverage this sanely in the qdisc layer...

#define HAVE_HW_TIME_STAMP

/**
 * struct skb_shared_hwtstamps - hardware time stamps
 * @hwtstamp:   hardware time stamp transformed into duration
 *              since arbitrary point in time
 * @syststamp:  hwtstamp transformed to system time base
 *


-- 
Dave Täht
SKYPE: davetaht
US Tel: 1-239-829-5608
http://www.bufferbloat.net

From: Eric Dumazet <redacted>
Date: Tue, 01 May 2012 11:11:05 +0200

From: Eric Dumazet <edumazet@google.com>

Add ECN (Explicit Congestion Notification) marking capability to netem

tc qdisc add dev eth0 root netem drop 0.5 ecn

Instead of dropping packets, try to ECN mark them.

Signed-off-by: Eric Dumazet <edumazet@google.com>

Applied.

On Tue, 01 May 2012 11:11:05 +0200
Eric Dumazet [off-list ref] wrote:

The concept is fine, but a couple of questions.
 1. Why a whole u32 for boolean?
 2. The logic in this part of netem is setup to handle case of random duplication
    combined with random loss. With ecn option set, will this code correctly
    handled a duplication combined with a loss and send one packet?
    It looks like the new code would change that behaviour.

On Tue, 01 May 2012 09:40:51 -0400 (EDT)
David Miller [off-list ref] wrote:

From: Eric Dumazet <redacted>
Date: Tue, 01 May 2012 11:11:05 +0200

quoted

From: Eric Dumazet <edumazet@google.com>

Add ECN (Explicit Congestion Notification) marking capability to netem

tc qdisc add dev eth0 root netem drop 0.5 ecn

Instead of dropping packets, try to ECN mark them.

Signed-off-by: Eric Dumazet <edumazet@google.com>

Applied.

At least give a day of review, rather than premature acceptance

On Tue, 2012-05-01 at 09:59 -0700, Stephen Hemminger wrote:

The concept is fine, but a couple of questions.
 1. Why a whole u32 for boolean?

a boolean in this structure wont save space, and this file is full of
u32. Why bother ?

IMHO boolean are fine for function arguments, but in a structure, not
very helpful.

 2. The logic in this part of netem is setup to handle case of random duplication
    combined with random loss. With ecn option set, will this code correctly
    handled a duplication combined with a loss and send one packet?
    It looks like the new code would change that behaviour.

Hmm, I'll take a look.

From: Stephen Hemminger <redacted>
Date: Tue, 1 May 2012 09:59:38 -0700

On Tue, 01 May 2012 09:40:51 -0400 (EDT)
David Miller [off-list ref] wrote:

quoted

From: Eric Dumazet <redacted>
Date: Tue, 01 May 2012 11:11:05 +0200

quoted

From: Eric Dumazet <edumazet@google.com>

Add ECN (Explicit Congestion Notification) marking capability to netem

tc qdisc add dev eth0 root netem drop 0.5 ecn

Instead of dropping packets, try to ECN mark them.

Signed-off-by: Eric Dumazet <edumazet@google.com>

Applied.

At least give a day of review, rather than premature acceptance

Yeah, because Eric Dumazet is going to fall off the face of the
planet and not fix whatever problems you have with his work.

Get real Stephen.

On Tue, May 1, 2012 at 10:17 AM, David Miller [off-list ref] wrote:

From: Stephen Hemminger <redacted>
Date: Tue, 1 May 2012 09:59:38 -0700

quoted

On Tue, 01 May 2012 09:40:51 -0400 (EDT)
David Miller [off-list ref] wrote:

quoted

From: Eric Dumazet <redacted>
Date: Tue, 01 May 2012 11:11:05 +0200

quoted

From: Eric Dumazet <edumazet@google.com>

Add ECN (Explicit Congestion Notification) marking capability to netem

tc qdisc add dev eth0 root netem drop 0.5 ecn

Instead of dropping packets, try to ECN mark them.

Signed-off-by: Eric Dumazet <edumazet@google.com>

Applied.

At least give a day of review, rather than premature acceptance

Yeah, because Eric Dumazet is going to fall off the face of the
planet and not fix whatever problems you have with his work.

Get real Stephen.

I'd be more concerned about wayward buses, or a team of elite
ninja's hired by outraged BSD and windows hackers sent to slow
eric down.

As I think about it, I think he could handle the ninja problem, tho.



-- 
Dave Täht
SKYPE: davetaht
US Tel: 1-239-829-5608
http://www.bufferbloat.net

From: Dave Taht <redacted>
Date: Tue, 1 May 2012 11:49:51 -0700

As I think about it, I think he could handle the ninja problem, tho.

Eric is the Networking Ninja.