Merge branch 'L4S-style-ce_threshold_ect1-marking'

Eric Dumazet says:

====================
net/sched: implement L4S style ce_threshold_ect1 marking

As suggested by Ingemar Johansson, Neal Cardwell, and others, fq_codel can be used
for Low Latency, Low Loss, Scalable Throughput (L4S) with a small change.

In ce_threshold_ect1 mode, only ECT(1) packets can be marked to CE if
their sojourn time is above the threshold.
====================

Signed-off-by: David S. Miller <davem@davemloft.net>

6 files changed, 47 insertions, 7 deletions

diff --git a/include/net/codel.h b/include/net/codel.h
index a6e428f80135..5e8b181b76b8 100644
--- a/include/net/codel.h
+++ b/include/net/codel.h
@@ -102,6 +102,7 @@ static inline u32 codel_time_to_us(codel_time_t val)
  * @interval:	width of moving time window
  * @mtu:	device mtu, or minimal queue backlog in bytes.
  * @ecn:	is Explicit Congestion Notification enabled
+ * @ce_threshold_ect1: if ce_threshold only marks ECT(1) packets
  */
 struct codel_params {
 	codel_time_t	target;
@@ -109,6 +110,7 @@ struct codel_params {
 	codel_time_t	interval;
 	u32		mtu;
 	bool		ecn;
+	bool		ce_threshold_ect1;
 };
 
 /**
diff --git a/include/net/codel_impl.h b/include/net/codel_impl.h
index d289b91dcd65..7af2c3eb3c43 100644
--- a/include/net/codel_impl.h
+++ b/include/net/codel_impl.h
@@ -54,6 +54,7 @@ static void codel_params_init(struct codel_params *params)
 	params->interval = MS2TIME(100);
 	params->target = MS2TIME(5);
 	params->ce_threshold = CODEL_DISABLED_THRESHOLD;
+	params->ce_threshold_ect1 = false;
 	params->ecn = false;
 }
 
@@ -246,9 +247,20 @@ static struct sk_buff *codel_dequeue(void *ctx,
 						    vars->rec_inv_sqrt);
 	}
 end:
-	if (skb && codel_time_after(vars->ldelay, params->ce_threshold) &&
-	    INET_ECN_set_ce(skb))
-		stats->ce_mark++;
+	if (skb && codel_time_after(vars->ldelay, params->ce_threshold)) {
+		bool set_ce = true;
+
+		if (params->ce_threshold_ect1) {
+			/* Note: if skb_get_dsfield() returns -1, following
+			 * gives INET_ECN_MASK, which is != INET_ECN_ECT_1.
+			 */
+			u8 ecn = skb_get_dsfield(skb) & INET_ECN_MASK;
+
+			set_ce = (ecn == INET_ECN_ECT_1);
+		}
+		if (set_ce && INET_ECN_set_ce(skb))
+			stats->ce_mark++;
+	}
 	return skb;
 }
 
diff --git a/include/net/inet_ecn.h b/include/net/inet_ecn.h
index ba77f47ef61e..ea32393464a2 100644
--- a/include/net/inet_ecn.h
+++ b/include/net/inet_ecn.h
@@ -188,6 +188,23 @@ static inline int INET_ECN_set_ce(struct sk_buff *skb)
 	return 0;
 }
 
+static inline int skb_get_dsfield(struct sk_buff *skb)
+{
+	switch (skb_protocol(skb, true)) {
+	case cpu_to_be16(ETH_P_IP):
+		if (!pskb_network_may_pull(skb, sizeof(struct iphdr)))
+			break;
+		return ipv4_get_dsfield(ip_hdr(skb));
+
+	case cpu_to_be16(ETH_P_IPV6):
+		if (!pskb_network_may_pull(skb, sizeof(struct ipv6hdr)))
+			break;
+		return ipv6_get_dsfield(ipv6_hdr(skb));
+	}
+
+	return -1;
+}
+
 static inline int INET_ECN_set_ect1(struct sk_buff *skb)
 {
 	switch (skb_protocol(skb, true)) {
diff --git a/include/uapi/linux/pkt_sched.h b/include/uapi/linux/pkt_sched.h
index ec88590b3198..6be9a84cccfa 100644
--- a/include/uapi/linux/pkt_sched.h
+++ b/include/uapi/linux/pkt_sched.h
@@ -840,6 +840,7 @@ enum {
 	TCA_FQ_CODEL_CE_THRESHOLD,
 	TCA_FQ_CODEL_DROP_BATCH_SIZE,
 	TCA_FQ_CODEL_MEMORY_LIMIT,
+	TCA_FQ_CODEL_CE_THRESHOLD_ECT1,
 	__TCA_FQ_CODEL_MAX
 };
 
diff --git a/net/mac80211/sta_info.c b/net/mac80211/sta_info.c
index 2b5acb37587f..a39830418434 100644
--- a/net/mac80211/sta_info.c
+++ b/net/mac80211/sta_info.c
@@ -513,6 +513,7 @@ struct sta_info *sta_info_alloc(struct ieee80211_sub_if_data *sdata,
 	sta->cparams.target = MS2TIME(20);
 	sta->cparams.interval = MS2TIME(100);
 	sta->cparams.ecn = true;
+	sta->cparams.ce_threshold_ect1 = false;
 
 	sta_dbg(sdata, "Allocated STA %pM\n", sta->sta.addr);
 
diff --git a/net/sched/sch_fq_codel.c b/net/sched/sch_fq_codel.c
index bb0cd6d3d2c2..033d65d06eb1 100644
--- a/net/sched/sch_fq_codel.c
+++ b/net/sched/sch_fq_codel.c
@@ -362,6 +362,7 @@ static const struct nla_policy fq_codel_policy[TCA_FQ_CODEL_MAX + 1] = {
 	[TCA_FQ_CODEL_CE_THRESHOLD] = { .type = NLA_U32 },
 	[TCA_FQ_CODEL_DROP_BATCH_SIZE] = { .type = NLA_U32 },
 	[TCA_FQ_CODEL_MEMORY_LIMIT] = { .type = NLA_U32 },
+	[TCA_FQ_CODEL_CE_THRESHOLD_ECT1] = { .type = NLA_U8 },
 };
 
 static int fq_codel_change(struct Qdisc *sch, struct nlattr *opt,
@@ -408,6 +409,9 @@ static int fq_codel_change(struct Qdisc *sch, struct nlattr *opt,
 		q->cparams.ce_threshold = (val * NSEC_PER_USEC) >> CODEL_SHIFT;
 	}
 
+	if (tb[TCA_FQ_CODEL_CE_THRESHOLD_ECT1])
+		q->cparams.ce_threshold_ect1 = !!nla_get_u8(tb[TCA_FQ_CODEL_CE_THRESHOLD_ECT1]);
+
 	if (tb[TCA_FQ_CODEL_INTERVAL]) {
 		u64 interval = nla_get_u32(tb[TCA_FQ_CODEL_INTERVAL]);
 
@@ -544,10 +548,13 @@ static int fq_codel_dump(struct Qdisc *sch, struct sk_buff *skb)
 			q->flows_cnt))
 		goto nla_put_failure;
 
-	if (q->cparams.ce_threshold != CODEL_DISABLED_THRESHOLD &&
-	    nla_put_u32(skb, TCA_FQ_CODEL_CE_THRESHOLD,
-			codel_time_to_us(q->cparams.ce_threshold)))
-		goto nla_put_failure;
+	if (q->cparams.ce_threshold != CODEL_DISABLED_THRESHOLD) {
+		if (nla_put_u32(skb, TCA_FQ_CODEL_CE_THRESHOLD,
+				codel_time_to_us(q->cparams.ce_threshold)))
+			goto nla_put_failure;
+		if (nla_put_u8(skb, TCA_FQ_CODEL_CE_THRESHOLD_ECT1, q->cparams.ce_threshold_ect1))
+			goto nla_put_failure;
+	}
 
 	return nla_nest_end(skb, opts);