Support matching and modifying IP ECN bits.

Signed-off-by: Justin Pettit <[email protected]>
Acked-by: Jesse Gross <[email protected]>

NEWS

@@ -2,6 +2,8 @@ post-v1.3.0
 ------------------------
     - OpenFlow:
        - Added ability to match on IPv6 flow label through NXM.
+       - Added ability to match on ECN bits in IPv4 and IPv6 through NXM.
+       - Added ability to modify ECN bits in IPv4.
     - ovs-appctl:
       - New "fdb/flush" command to flush bridge's MAC learning table.
 

datapath/actions.c

@@ -19,9 +19,9 @@
 #include <linux/in6.h>
 #include <linux/if_arp.h>
 #include <linux/if_vlan.h>
-#include <net/inet_ecn.h>
 #include <net/ip.h>
 #include <net/checksum.h>
+#include <net/dsfield.h>
 
 #include "actions.h"
 #include "checksum.h"
@@ -151,17 +151,6 @@ static void set_ip_addr(struct sk_buff *skb, struct iphdr *nh,
 	*addr = new_addr;
 }
 
-static void set_ip_tos(struct sk_buff *skb, struct iphdr *nh, u8 new_tos)
-{
-	u8 old, new;
-
-	/* Set the DSCP bits and preserve the ECN bits. */
-	old = nh->tos;
-	new = new_tos | (nh->tos & INET_ECN_MASK);
-	csum_replace2(&nh->check, htons(old), htons(new));
-	nh->tos = new;
-}
-
 static int set_ipv4(struct sk_buff *skb, const struct ovs_key_ipv4 *ipv4_key)
 {
 	struct iphdr *nh;
@@ -181,7 +170,7 @@ static int set_ipv4(struct sk_buff *skb, const struct ovs_key_ipv4 *ipv4_key)
 		set_ip_addr(skb, nh, &nh->daddr, ipv4_key->ipv4_dst);
 
 	if (ipv4_key->ipv4_tos != nh->tos)
-		set_ip_tos(skb, nh, ipv4_key->ipv4_tos);
+		ipv4_change_dsfield(nh, 0, ipv4_key->ipv4_tos);
 
 	return 0;
 }

datapath/datapath.c

@@ -41,7 +41,6 @@
 #include <linux/openvswitch.h>
 #include <linux/rculist.h>
 #include <linux/dmi.h>
-#include <net/inet_ecn.h>
 #include <net/genetlink.h>
 
 #include "checksum.h"
@@ -589,9 +588,6 @@ static int validate_action_key(const struct nlattr *a,
 		if (ipv4_key->ipv4_proto != flow_key->ip.proto)
 			return -EINVAL;
 
-		if (ipv4_key->ipv4_tos & INET_ECN_MASK)
-			return -EINVAL;
-
 		if (ipv4_key->ipv4_frag != flow_key->ip.frag)
 			return -EINVAL;
 

datapath/flow.c

@@ -30,7 +30,6 @@
 #include <linux/icmp.h>
 #include <linux/icmpv6.h>
 #include <linux/rculist.h>
-#include <net/inet_ecn.h>
 #include <net/ip.h>
 #include <net/ipv6.h>
 #include <net/ndisc.h>
@@ -200,7 +199,7 @@ static int parse_ipv6hdr(struct sk_buff *skb, struct sw_flow_key *key,
 	payload_ofs = (u8 *)(nh + 1) - skb->data;
 
 	key->ip.proto = NEXTHDR_NONE;
-	key->ip.tos = ipv6_get_dsfield(nh) & ~INET_ECN_MASK;
+	key->ip.tos = ipv6_get_dsfield(nh);
 	key->ipv6.label = *(__be32 *)nh & htonl(IPV6_FLOWINFO_FLOWLABEL);
 	ipv6_addr_copy(&key->ipv6.addr.src, &nh->saddr);
 	ipv6_addr_copy(&key->ipv6.addr.dst, &nh->daddr);
@@ -689,7 +688,7 @@ int flow_extract(struct sk_buff *skb, u16 in_port, struct sw_flow_key *key,
 		key->ipv4.addr.dst = nh->daddr;
 
 		key->ip.proto = nh->protocol;
-		key->ip.tos = nh->tos & ~INET_ECN_MASK;
+		key->ip.tos = nh->tos;
 
 		offset = nh->frag_off & htons(IP_OFFSET);
 		if (offset) {
@@ -958,8 +957,6 @@ int flow_from_nlattrs(struct sw_flow_key *swkey, int *key_lenp,
 			if (swkey->eth.type != htons(ETH_P_IP))
 				goto invalid;
 			ipv4_key = nla_data(nla);
-			if (ipv4_key->ipv4_tos & INET_ECN_MASK)
-				goto invalid;
 			if (ipv4_key->ipv4_frag > OVS_FRAG_TYPE_MAX)
 				goto invalid;
 			swkey->ip.proto = ipv4_key->ipv4_proto;
@@ -974,8 +971,6 @@ int flow_from_nlattrs(struct sw_flow_key *swkey, int *key_lenp,
 			if (swkey->eth.type != htons(ETH_P_IPV6))
 				goto invalid;
 			ipv6_key = nla_data(nla);
-			if (ipv6_key->ipv6_tos & INET_ECN_MASK)
-				goto invalid;
 			if (ipv6_key->ipv6_frag > OVS_FRAG_TYPE_MAX)
 				goto invalid;
 			swkey->ipv6.label = ipv6_key->ipv6_label;
@@ -1249,7 +1244,7 @@ int flow_to_nlattrs(const struct sw_flow_key *swkey, struct sk_buff *skb)
 		ipv4_key->ipv4_src = swkey->ipv4.addr.src;
 		ipv4_key->ipv4_dst = swkey->ipv4.addr.dst;
 		ipv4_key->ipv4_proto = swkey->ip.proto;
-		ipv4_key->ipv4_tos = swkey->ip.tos & ~INET_ECN_MASK;
+		ipv4_key->ipv4_tos = swkey->ip.tos;
 		ipv4_key->ipv4_frag = swkey->ip.frag;
 	} else if (swkey->eth.type == htons(ETH_P_IPV6)) {
 		struct ovs_key_ipv6 *ipv6_key;
@@ -1265,7 +1260,7 @@ int flow_to_nlattrs(const struct sw_flow_key *swkey, struct sk_buff *skb)
 				sizeof(ipv6_key->ipv6_dst));
 		ipv6_key->ipv6_label = swkey->ipv6.label;
 		ipv6_key->ipv6_proto = swkey->ip.proto;
-		ipv6_key->ipv6_tos = swkey->ip.tos & ~INET_ECN_MASK;
+		ipv6_key->ipv6_tos = swkey->ip.tos;
 		ipv6_key->ipv6_frag = swkey->ip.frag;
 	} else if (swkey->eth.type == htons(ETH_P_ARP)) {
 		struct ovs_key_arp *arp_key;

datapath/flow.h

@@ -44,7 +44,7 @@ struct sw_flow_key {
 	} eth;
 	struct {
 		u8     proto;		/* IP protocol or lower 8 bits of ARP opcode. */
-		u8     tos;         /* IP ToS DSCP in high 6 bits. */
+		u8     tos; 		/* IP ToS. */
 		u8     frag;        /* One of OVS_FRAG_TYPE_*. */
 	} ip;
 	union {

include/openflow/nicira-ext.h

@@ -1623,6 +1623,15 @@ OFP_ASSERT(sizeof(struct nx_action_output_reg) == 24);
  * Masking: Not maskable. */
 #define NXM_NX_IPV6_LABEL  NXM_HEADER  (0x0001, 27, 4)
 
+/* The ECN of the IP header.
+ *
+ * Prereqs: NXM_OF_ETH_TYPE must be either 0x0800 or 0x86dd.
+ *
+ * Format: ECN in the low-order 2 bits.
+ *
+ * Masking: Not maskable. */
+#define NXM_NX_IP_ECN      NXM_HEADER  (0x0001, 28, 1)
+
 /* ## --------------------- ## */
 /* ## Requests and replies. ## */
 /* ## --------------------- ## */

lib/classifier.c

@@ -318,11 +318,19 @@ cls_rule_set_nw_dst_masked(struct cls_rule *rule, ovs_be32 ip, ovs_be32 mask)
 }
 
 void
-cls_rule_set_nw_tos(struct cls_rule *rule, uint8_t nw_tos)
+cls_rule_set_nw_dscp(struct cls_rule *rule, uint8_t nw_dscp)
 {
     rule->wc.tos_mask |= IP_DSCP_MASK;
     rule->flow.tos &= ~IP_DSCP_MASK;
-    rule->flow.tos |= nw_tos & IP_DSCP_MASK;
+    rule->flow.tos |= nw_dscp & IP_DSCP_MASK;
+}
+
+void
+cls_rule_set_nw_ecn(struct cls_rule *rule, uint8_t nw_ecn)
+{
+    rule->wc.tos_mask |= IP_ECN_MASK;
+    rule->flow.tos &= ~IP_ECN_MASK;
+    rule->flow.tos |= nw_ecn & IP_ECN_MASK;
 }
 
 void
@@ -620,6 +628,9 @@ cls_rule_format(const struct cls_rule *rule, struct ds *s)
     if (wc->tos_mask & IP_DSCP_MASK) {
         ds_put_format(s, "nw_tos=%"PRIu8",", f->tos & IP_DSCP_MASK);
     }
+    if (wc->tos_mask & IP_ECN_MASK) {
+        ds_put_format(s, "nw_ecn=%"PRIu8",", f->tos & IP_ECN_MASK);
+    }
     switch (wc->frag_mask) {
     case FLOW_FRAG_ANY | FLOW_FRAG_LATER:
         ds_put_format(s, "frag=%s,",

lib/classifier.h

@@ -116,7 +116,8 @@ void cls_rule_set_nw_src(struct cls_rule *, ovs_be32);
 bool cls_rule_set_nw_src_masked(struct cls_rule *, ovs_be32 ip, ovs_be32 mask);
 void cls_rule_set_nw_dst(struct cls_rule *, ovs_be32);
 bool cls_rule_set_nw_dst_masked(struct cls_rule *, ovs_be32 ip, ovs_be32 mask);
-void cls_rule_set_nw_tos(struct cls_rule *, uint8_t);
+void cls_rule_set_nw_dscp(struct cls_rule *, uint8_t);
+void cls_rule_set_nw_ecn(struct cls_rule *, uint8_t);
 void cls_rule_set_frag(struct cls_rule *, uint8_t frag);
 void cls_rule_set_frag_masked(struct cls_rule *, uint8_t frag, uint8_t mask);
 void cls_rule_set_icmp_type(struct cls_rule *, uint8_t);

lib/dpif-netdev.c

@@ -1086,12 +1086,9 @@ dp_netdev_set_ip_tos(struct ip_header *nh, uint8_t new_tos)
 {
     uint8_t *field = &nh->ip_tos;
 
-    /* Set the DSCP bits and preserve the ECN bits. */
-    uint8_t new = new_tos | (nh->ip_tos & IP_ECN_MASK);
-
     nh->ip_csum = recalc_csum16(nh->ip_csum, htons((uint16_t)*field),
-			            htons((uint16_t) new));
-    *field = new;
+			            htons((uint16_t) new_tos));
+    *field = new_tos;
 }
 
 static void

lib/flow.c

@@ -148,7 +148,7 @@ parse_ipv6(struct ofpbuf *packet, struct flow *flow)
     flow->ipv6_dst = nh->ip6_dst;
 
     tc_flow = get_unaligned_be32(&nh->ip6_flow);
-    flow->tos = (ntohl(tc_flow) >> 4) & IP_DSCP_MASK;
+    flow->tos = ntohl(tc_flow) >> 4;
     flow->ipv6_label = tc_flow & htonl(IPV6_LABEL_MASK);
     flow->nw_proto = IPPROTO_NONE;
 
@@ -369,7 +369,7 @@ flow_extract(struct ofpbuf *packet, uint32_t priority, ovs_be64 tun_id,
             flow->nw_dst = get_unaligned_be32(&nh->ip_dst);
             flow->nw_proto = nh->ip_proto;
 
-            flow->tos = nh->ip_tos & IP_DSCP_MASK;
+            flow->tos = nh->ip_tos;
             if (IP_IS_FRAGMENT(nh->ip_frag_off)) {
                 flow->frag = FLOW_FRAG_ANY;
                 if (nh->ip_frag_off & htons(IP_FRAG_OFF_MASK)) {
@@ -525,19 +525,15 @@ flow_format(struct ds *ds, const struct flow *flow)
                   ntohs(flow->dl_type));
 
     if (flow->dl_type == htons(ETH_TYPE_IPV6)) {
-        ds_put_format(ds, " label%#"PRIx32" proto%"PRIu8" tos%"PRIu8" ipv6",
-                      ntohl(flow->ipv6_label), flow->nw_proto,
-                      flow->tos & IP_DSCP_MASK);
+        ds_put_format(ds, " label%#"PRIx32" proto%"PRIu8" tos%#"PRIx8" ipv6",
+                      ntohl(flow->ipv6_label), flow->nw_proto, flow->tos);
         print_ipv6_addr(ds, &flow->ipv6_src);
         ds_put_cstr(ds, "->");
         print_ipv6_addr(ds, &flow->ipv6_dst);
 
     } else {
-        ds_put_format(ds, " proto%"PRIu8
-                          " tos%"PRIu8
-                          " ip"IP_FMT"->"IP_FMT,
-                      flow->nw_proto,
-                      flow->tos & IP_DSCP_MASK,
+        ds_put_format(ds, " proto%"PRIu8" tos%#"PRIx8" ip"IP_FMT"->"IP_FMT,
+                      flow->nw_proto, flow->tos,
                       IP_ARGS(&flow->nw_src),
                       IP_ARGS(&flow->nw_dst));
     }
@@ -1016,7 +1012,7 @@ flow_compose(struct ofpbuf *b, const struct flow *flow)
 
         b->l3 = ip = ofpbuf_put_zeros(b, sizeof *ip);
         ip->ip_ihl_ver = IP_IHL_VER(5, 4);
-        ip->ip_tos = flow->tos & IP_DSCP_MASK;
+        ip->ip_tos = flow->tos;
         ip->ip_proto = flow->nw_proto;
         ip->ip_src = flow->nw_src;
         ip->ip_dst = flow->nw_dst;