ECN marking implenetation for dummynet.

Changes include both DCTCP and RFC 3168 ECN marking methodology.

DCTCP draft: http://tools.ietf.org/html/draft-bensley-tcpm-dctcp-00

Submitted by:	Midori Kato (aoimidori27@gmail.com)
Worked with:	Lars Eggert (lars@netapp.com)
Reviewed by:	luigi, hiren

7 files changed, 108 insertions, 15 deletions

diff --git a/sbin/ipfw/dummynet.c b/sbin/ipfw/dummynet.c
index 28dc2c7..cb62853 100644
--- a/sbin/ipfw/dummynet.c
+++ b/sbin/ipfw/dummynet.c
@@ -56,6 +56,7 @@ static struct _s_x dummynet_params[] = {
 	{ "sched_mask",		TOK_SCHED_MASK },
 	{ "flow_mask",		TOK_FLOW_MASK },
 	{ "droptail",		TOK_DROPTAIL },
+	{ "ecn",		TOK_ECN },
 	{ "red",		TOK_RED },
 	{ "gred",		TOK_GRED },
 	{ "bw",			TOK_BW },
@@ -239,7 +240,7 @@ print_flowset_parms(struct dn_fs *fs, char *prefix)
 	else
 		plr[0] = '\0';
 
-	if (fs->flags & DN_IS_RED)	/* RED parameters */
+	if (fs->flags & DN_IS_RED) {	/* RED parameters */
 		sprintf(red,
 		    "\n\t %cRED w_q %f min_th %d max_th %d max_p %f",
 		    (fs->flags & DN_IS_GENTLE_RED) ? 'G' : ' ',
@@ -247,7 +248,9 @@ print_flowset_parms(struct dn_fs *fs, char *prefix)
 		    fs->min_th,
 		    fs->max_th,
 		    1.0 * fs->max_p / (double)(1 << SCALE_RED));
-	else
+		if (fs->flags & DN_IS_ECN)
+			strncat(red, " (ecn)", 6);
+	} else
 		sprintf(red, "droptail");
 
 	if (prefix[0]) {
@@ -1046,13 +1049,17 @@ end_mask:
 			}
 			if ((end = strsep(&av[0], "/"))) {
 			    double max_p = strtod(end, NULL);
-			    if (max_p > 1 || max_p <= 0)
-				errx(EX_DATAERR, "0 < max_p <= 1");
+			    if (max_p > 1 || max_p < 0)
+				errx(EX_DATAERR, "0 <= max_p <= 1");
 			    fs->max_p = (int)(max_p * (1 << SCALE_RED));
 			}
 			ac--; av++;
 			break;
 
+		case TOK_ECN:
+			fs->flags |= DN_IS_ECN;
+			break;
+
 		case TOK_DROPTAIL:
 			NEED(fs, "droptail is only for flowsets");
 			fs->flags &= ~(DN_IS_RED|DN_IS_GENTLE_RED);
@@ -1175,13 +1182,20 @@ end_mask:
 			errx(EX_DATAERR, "2 <= queue size <= %ld", limit);
 	    }
 
+	    if ((fs->flags & DN_IS_ECN) && !(fs->flags & DN_IS_RED))
+		errx(EX_USAGE, "enable red/gred for ECN");
+
 	    if (fs->flags & DN_IS_RED) {
 		size_t len;
 		int lookup_depth, avg_pkt_size;
 
-		if (fs->min_th >= fs->max_th)
+		if (!(fs->flags & DN_IS_ECN) && (fs->min_th >= fs->max_th))
 		    errx(EX_DATAERR, "min_th %d must be < than max_th %d",
 			fs->min_th, fs->max_th);
+		else if ((fs->flags & DN_IS_ECN) && (fs->min_th > fs->max_th))
+		    errx(EX_DATAERR, "min_th %d must be =< than max_th %d",
+			fs->min_th, fs->max_th);
+
 		if (fs->max_th == 0)
 		    errx(EX_DATAERR, "max_th must be > 0");
 
diff --git a/sbin/ipfw/ipfw.8 b/sbin/ipfw/ipfw.8
index bc8d819..b419371 100644
--- a/sbin/ipfw/ipfw.8
+++ b/sbin/ipfw/ipfw.8
@@ -1,7 +1,7 @@
 .\"
 .\" $FreeBSD$
 .\"
-.Dd October 25, 2012
+.Dd May 31, 2014
 .Dt IPFW 8
 .Os
 .Sh NAME
@@ -2441,22 +2441,23 @@ and
 control the maximum lengths that can be specified.
 .Pp
 .It Cm red | gred Ar w_q Ns / Ns Ar min_th Ns / Ns Ar max_th Ns / Ns Ar max_p
+[ecn]
 Make use of the RED (Random Early Detection) queue management algorithm.
 .Ar w_q
 and
 .Ar max_p
 are floating
-point numbers between 0 and 1 (0 not included), while
+point numbers between 0 and 1 (inclusive), while
 .Ar min_th
 and
 .Ar max_th
 are integer numbers specifying thresholds for queue management
 (thresholds are computed in bytes if the queue has been defined
 in bytes, in slots otherwise).
-The
+The two parameters can also be of the same value if needed. The
 .Nm dummynet
-also supports the gentle RED variant (gred).
-Three
+also supports the gentle RED variant (gred) and ECN (Explicit Congestion
+Notification) as optional. Three
 .Xr sysctl 8
 variables can be used to control the RED behaviour:
 .Bl -tag -width indent
diff --git a/sbin/ipfw/ipfw2.h b/sbin/ipfw/ipfw2.h
index 6e895b8..2301c40 100644
--- a/sbin/ipfw/ipfw2.h
+++ b/sbin/ipfw/ipfw2.h
@@ -165,6 +165,7 @@ enum tokens {
 	TOK_BURST,
 	TOK_RED,
 	TOK_GRED,
+	TOK_ECN,
 	TOK_DROPTAIL,
 	TOK_PROTO,
 	/* dummynet tokens */
diff --git a/sys/netinet/ip_dummynet.h b/sys/netinet/ip_dummynet.h
index 1c09197..202f1e2 100644
--- a/sys/netinet/ip_dummynet.h
+++ b/sys/netinet/ip_dummynet.h
@@ -104,6 +104,7 @@ enum {	/* user flags */
 	DN_HAS_PROFILE	= 0x0010,	/* a link has a profile */
 	DN_IS_RED	= 0x0020,
 	DN_IS_GENTLE_RED= 0x0040,
+	DN_IS_ECN	= 0x0080,
 	DN_PIPE_CMD	= 0x1000,	/* pipe config... */
 };
 
diff --git a/sys/netpfil/ipfw/ip_dn_glue.c b/sys/netpfil/ipfw/ip_dn_glue.c
index 7d7e695..095758f 100644
--- a/sys/netpfil/ipfw/ip_dn_glue.c
+++ b/sys/netpfil/ipfw/ip_dn_glue.c
@@ -83,6 +83,7 @@ struct dn_flow_set {
 #define DNOLD_QSIZE_IS_BYTES   0x0008  /* queue size is measured in bytes */
 #define DNOLD_NOERROR      0x0010  /* do not report ENOBUFS on drops  */
 #define DNOLD_HAS_PROFILE      0x0020  /* the pipe has a delay profile. */
+#define DNOLD_IS_ECN       0x0040
 #define DNOLD_IS_PIPE      0x4000
 #define DNOLD_IS_QUEUE     0x8000
 
@@ -338,6 +339,8 @@ convertflags2new(int src)
 		dst |= DN_IS_RED;
 	if (src & DNOLD_IS_GENTLE_RED)
 		dst |= DN_IS_GENTLE_RED;
+	if (src & DNOLD_IS_ECN)
+		dst |= DN_IS_ECN;
 	if (src & DNOLD_HAS_PROFILE)
 		dst |= DN_HAS_PROFILE;
 
diff --git a/sys/netpfil/ipfw/ip_dn_io.c b/sys/netpfil/ipfw/ip_dn_io.c
index 1748194..df298ae 100644
--- a/sys/netpfil/ipfw/ip_dn_io.c
+++ b/sys/netpfil/ipfw/ip_dn_io.c
@@ -367,6 +367,8 @@ red_drops (struct dn_queue *q, int len)
 		return (0);	/* accept packet */
 	}
 	if (q->avg >= fs->max_th) {	/* average queue >=  max threshold */
+		if (fs->fs.flags & DN_IS_ECN)
+			return (1);
 		if (fs->fs.flags & DN_IS_GENTLE_RED) {
 			/*
 			 * According to Gentle-RED, if avg is greater than
@@ -382,6 +384,8 @@ red_drops (struct dn_queue *q, int len)
 			return (1);
 		}
 	} else if (q->avg > fs->min_th) {
+		if (fs->fs.flags & DN_IS_ECN)
+			return (1);
 		/*
 		 * We compute p_b using the linear dropping function
 		 *	 p_b = c_1 * avg - c_2
@@ -414,6 +418,70 @@ red_drops (struct dn_queue *q, int len)
 }
 
 /*
+ * ECN/ECT Processing (partially adopted from altq)
+ */
+static int
+ecn_mark(struct mbuf* m)
+{
+	struct ip *ip;
+	ip = mtod(m, struct ip *);
+
+	switch (ip->ip_v) {
+	case IPVERSION:
+	{
+		u_int8_t otos;
+		int sum;
+
+		if ((ip->ip_tos & IPTOS_ECN_MASK) == IPTOS_ECN_NOTECT)
+			return (0);	/* not-ECT */
+		if ((ip->ip_tos & IPTOS_ECN_MASK) == IPTOS_ECN_CE)
+			return (1);	/* already marked */
+
+		/*
+		 * ecn-capable but not marked,
+		 * mark CE and update checksum
+		 */
+		otos = ip->ip_tos;
+		ip->ip_tos |= IPTOS_ECN_CE;
+		/*
+		 * update checksum (from RFC1624)
+		 *	   HC' = ~(~HC + ~m + m')
+		 */
+		sum = ~ntohs(ip->ip_sum) & 0xffff;
+		sum += (~otos & 0xffff) + ip->ip_tos;
+		sum = (sum >> 16) + (sum & 0xffff);
+		sum += (sum >> 16);  /* add carry */
+		ip->ip_sum = htons(~sum & 0xffff);
+		return (1);
+	}
+#ifdef INET6
+	case (IPV6_VERSION >> 4):
+	{
+		struct ip6_hdr *ip6 = mtod(m, struct ip6_hdr *);
+		u_int32_t flowlabel;
+
+		flowlabel = ntohl(ip6->ip6_flow);
+		if ((flowlabel >> 28) != 6)
+			return (0);	/* version mismatch! */
+		if ((flowlabel & (IPTOS_ECN_MASK << 20)) ==
+		    (IPTOS_ECN_NOTECT << 20))
+			return (0);	/* not-ECT */
+		if ((flowlabel & (IPTOS_ECN_MASK << 20)) ==
+		    (IPTOS_ECN_CE << 20))
+			return (1);	/* already marked */
+		/*
+		 * ecn-capable but not marked, mark CE
+		 */
+		flowlabel |= (IPTOS_ECN_CE << 20);
+		ip6->ip6_flow = htonl(flowlabel);
+		return (1);
+	}
+#endif
+	}
+	return (0);
+}
+
+/*
  * Enqueue a packet in q, subject to space and queue management policy
  * (whose parameters are in q->fs).
  * Update stats for the queue and the scheduler.
@@ -444,8 +512,10 @@ dn_enqueue(struct dn_queue *q, struct mbuf* m, int drop)
 		goto drop;
 	if (f->plr && random() < f->plr)
 		goto drop;
-	if (f->flags & DN_IS_RED && red_drops(q, m->m_pkthdr.len))
-		goto drop;
+	if (f->flags & DN_IS_RED && red_drops(q, m->m_pkthdr.len)) {
+		if (!(f->flags & DN_IS_ECN) || !ecn_mark(m))
+			goto drop;
+	}
 	if (f->flags & DN_QSIZE_BYTES) {
 		if (q->ni.len_bytes > f->qsize)
 			goto drop;
@@ -457,14 +527,14 @@ dn_enqueue(struct dn_queue *q, struct mbuf* m, int drop)
 	q->ni.len_bytes += len;
 	ni->length++;
 	ni->len_bytes += len;
-	return 0;
+	return (0);
 
 drop:
 	io_pkt_drop++;
 	q->ni.drops++;
 	ni->drops++;
 	FREE_PKT(m);
-	return 1;
+	return (1);
 }
 
 /*
diff --git a/sys/netpfil/ipfw/ip_dummynet.c b/sys/netpfil/ipfw/ip_dummynet.c
index 453c0f1..5268686 100644
--- a/sys/netpfil/ipfw/ip_dummynet.c
+++ b/sys/netpfil/ipfw/ip_dummynet.c
@@ -1071,7 +1071,10 @@ config_red(struct dn_fsk *fs)
 	fs->min_th = SCALE(fs->fs.min_th);
 	fs->max_th = SCALE(fs->fs.max_th);
 
-	fs->c_1 = fs->max_p / (fs->fs.max_th - fs->fs.min_th);
+	if (fs->fs.max_th == fs->fs.min_th)
+		fs->c_1 = fs->max_p;
+	else
+		fs->c_1 = SCALE((int64_t)(fs->max_p)) / (fs->fs.max_th - fs->fs.min_th);
 	fs->c_2 = SCALE_MUL(fs->c_1, SCALE(fs->fs.min_th));
 
 	if (fs->fs.flags & DN_IS_GENTLE_RED) {