X-Git-Url: https://pintos-os.org/cgi-bin/gitweb.cgi?a=blobdiff_plain;f=datapath%2Factions.c;h=de98d990c2dff6f79ade48695983a8324e0ff321;hb=9cb8877cf7d1d86101a8f27829ad47ea0c8b1fe5;hp=3223c65bd66b75680caf790ac470ff9f498e6d58;hpb=7aec165dbc4690c8c2c703d142e2f017bb851d31;p=openvswitch

diff --git a/datapath/actions.c b/datapath/actions.c
index 3223c65b..de98d990 100644
--- a/datapath/actions.c
+++ b/datapath/actions.c
@@ -23,12 +23,13 @@
 #include "actions.h"
 #include "checksum.h"
 #include "datapath.h"
+#include "loop_counter.h"
 #include "openvswitch/datapath-protocol.h"
+#include "vlan.h"
 #include "vport.h"
 
 static int do_execute_actions(struct datapath *, struct sk_buff *,
-			      const struct sw_flow_key *,
-			      const struct nlattr *actions, u32 actions_len);
+			      struct sw_flow_actions *acts);
 
 static struct sk_buff *make_writable(struct sk_buff *skb, unsigned min_headroom)
 {
@@ -52,20 +53,28 @@ static struct sk_buff *make_writable(struct sk_buff *skb, unsigned min_headroom)
 	return NULL;
 }
 
-static struct sk_buff *vlan_pull_tag(struct sk_buff *skb)
+static struct sk_buff *strip_vlan(struct sk_buff *skb)
 {
-	struct vlan_ethhdr *vh = vlan_eth_hdr(skb);
 	struct ethhdr *eh;
 
-	/* Verify we were given a vlan packet */
-	if (vh->h_vlan_proto != htons(ETH_P_8021Q) || skb->len < VLAN_ETH_HLEN)
+	if (vlan_tx_tag_present(skb)) {
+		vlan_set_tci(skb, 0);
 		return skb;
+	}
+
+	if (unlikely(vlan_eth_hdr(skb)->h_vlan_proto != htons(ETH_P_8021Q) ||
+	    skb->len < VLAN_ETH_HLEN))
+		return skb;
+
+	skb = make_writable(skb, 0);
+	if (unlikely(!skb))
+		return NULL;
 
 	if (get_ip_summed(skb) == OVS_CSUM_COMPLETE)
 		skb->csum = csum_sub(skb->csum, csum_partial(skb->data
 					+ ETH_HLEN, VLAN_HLEN, 0));
 
-	memmove(skb->data + VLAN_HLEN, skb->data, 2 * VLAN_ETH_ALEN);
+	memmove(skb->data + VLAN_HLEN, skb->data, 2 * ETH_ALEN);
 
 	eh = (struct ethhdr *)skb_pull(skb, VLAN_HLEN);
 
@@ -75,167 +84,62 @@ static struct sk_buff *vlan_pull_tag(struct sk_buff *skb)
 	return skb;
 }
 
-static struct sk_buff *modify_vlan_tci(struct datapath *dp, struct sk_buff *skb,
-				       const struct sw_flow_key *key,
-				       const struct nlattr *a, u32 actions_len)
+static struct sk_buff *modify_vlan_tci(struct sk_buff *skb, __be16 tci)
 {
-	__be16 tci = nla_get_be16(a);
-
-	skb = make_writable(skb, VLAN_HLEN);
-	if (!skb)
-		return ERR_PTR(-ENOMEM);
+	struct vlan_ethhdr *vh;
+	__be16 old_tci;
 
-	if (skb->protocol == htons(ETH_P_8021Q)) {
-		/* Modify vlan id, but maintain other TCI values */
-		struct vlan_ethhdr *vh;
-		__be16 old_tci;
+	if (vlan_tx_tag_present(skb) || skb->protocol != htons(ETH_P_8021Q))
+		return __vlan_hwaccel_put_tag(skb, ntohs(tci));
 
-		if (skb->len < VLAN_ETH_HLEN)
-			return skb;
-
-		vh = vlan_eth_hdr(skb);
-		old_tci = vh->h_vlan_TCI;
-
-		vh->h_vlan_TCI = tci;
-
-		if (get_ip_summed(skb) == OVS_CSUM_COMPLETE) {
-			__be16 diff[] = { ~old_tci, vh->h_vlan_TCI };
-
-			skb->csum = ~csum_partial((char *)diff, sizeof(diff),
-						~skb->csum);
-		}
-	} else {
-		int err;
-
-		/* Add vlan header */
+	skb = make_writable(skb, 0);
+	if (unlikely(!skb))
+		return NULL;
 
-		/* Set up checksumming pointers for checksum-deferred packets
-		 * on Xen.  Otherwise, dev_queue_xmit() will try to do this
-		 * when we send the packet out on the wire, and it will fail at
-		 * that point because skb_checksum_setup() will not look inside
-		 * an 802.1Q header. */
-		err = vswitch_skb_checksum_setup(skb);
-		if (unlikely(err)) {
-			kfree_skb(skb);
-			return ERR_PTR(err);
-		}
+	if (unlikely(skb->len < VLAN_ETH_HLEN))
+		return skb;
 
-		/* GSO is not implemented for packets with an 802.1Q header, so
-		 * we have to do segmentation before we add that header.
-		 *
-		 * GSO does work with hardware-accelerated VLAN tagging, but we
-		 * can't use hardware-accelerated VLAN tagging since it
-		 * requires the device to have a VLAN group configured (with
-		 * e.g. vconfig(8)) and we don't do that.
-		 *
-		 * Having to do this here may be a performance loss, since we
-		 * can't take advantage of TSO hardware support, although it
-		 * does not make a measurable network performance difference
-		 * for 1G Ethernet.  Fixing that would require patching the
-		 * kernel (either to add GSO support to the VLAN protocol or to
-		 * support hardware-accelerated VLAN tagging without VLAN
-		 * groups configured). */
-		if (skb_is_gso(skb)) {
-			const struct nlattr *actions_left;
-			int actions_len_left;
-			struct sk_buff *segs;
-
-			segs = skb_gso_segment(skb, 0);
-			kfree_skb(skb);
-			if (IS_ERR(segs))
-				return ERR_CAST(segs);
-
-			actions_len_left = actions_len;
-			actions_left = nla_next(a, &actions_len_left);
-
-			do {
-				struct sk_buff *nskb = segs->next;
-
-				segs->next = NULL;
-
-				/* GSO can change the checksum type so update.*/
-				compute_ip_summed(segs, true);
-
-				segs = __vlan_put_tag(segs, ntohs(tci));
-				err = -ENOMEM;
-				if (segs) {
-					err = do_execute_actions(
-						dp, segs, key, actions_left,
-						actions_len_left);
-				}
-
-				if (unlikely(err)) {
-					while ((segs = nskb)) {
-						nskb = segs->next;
-						segs->next = NULL;
-						kfree_skb(segs);
-					}
-					return ERR_PTR(err);
-				}
-
-				segs = nskb;
-			} while (segs->next);
-
-			skb = segs;
-			compute_ip_summed(skb, true);
-		}
+	vh = vlan_eth_hdr(skb);
 
-		/* The hardware-accelerated version of vlan_put_tag() works
-		 * only for a device that has a VLAN group configured (with
-		 * e.g. vconfig(8)), so call the software-only version
-		 * __vlan_put_tag() directly instead.
-		 */
-		skb = __vlan_put_tag(skb, ntohs(tci));
-		if (!skb)
-			return ERR_PTR(-ENOMEM);
+	old_tci = vh->h_vlan_TCI;
+	vh->h_vlan_TCI = tci;
 
-		/* GSO doesn't fix up the hardware computed checksum so this
-		 * will only be hit in the non-GSO case. */
-		if (get_ip_summed(skb) == OVS_CSUM_COMPLETE)
-			skb->csum = csum_add(skb->csum, csum_partial(skb->data
-						+ ETH_HLEN, VLAN_HLEN, 0));
+	if (get_ip_summed(skb) == OVS_CSUM_COMPLETE) {
+		__be16 diff[] = { ~old_tci, vh->h_vlan_TCI };
+		skb->csum = ~csum_partial((char *)diff, sizeof(diff), ~skb->csum);
 	}
 
 	return skb;
 }
 
-static struct sk_buff *strip_vlan(struct sk_buff *skb)
-{
-	skb = make_writable(skb, 0);
-	if (skb)
-		vlan_pull_tag(skb);
-	return skb;
-}
-
-static bool is_ip(struct sk_buff *skb, const struct sw_flow_key *key)
+static bool is_ip(struct sk_buff *skb)
 {
-	return (key->dl_type == htons(ETH_P_IP) &&
+	return (OVS_CB(skb)->flow->key.eth.type == htons(ETH_P_IP) &&
 		skb->transport_header > skb->network_header);
 }
 
-static __sum16 *get_l4_checksum(struct sk_buff *skb, const struct sw_flow_key *key)
+static __sum16 *get_l4_checksum(struct sk_buff *skb)
 {
+	u8 nw_proto = OVS_CB(skb)->flow->key.ip.proto;
 	int transport_len = skb->len - skb_transport_offset(skb);
-	if (key->nw_proto == IPPROTO_TCP) {
+	if (nw_proto == IPPROTO_TCP) {
 		if (likely(transport_len >= sizeof(struct tcphdr)))
 			return &tcp_hdr(skb)->check;
-	} else if (key->nw_proto == IPPROTO_UDP) {
+	} else if (nw_proto == IPPROTO_UDP) {
 		if (likely(transport_len >= sizeof(struct udphdr)))
 			return &udp_hdr(skb)->check;
 	}
 	return NULL;
 }
 
-static struct sk_buff *set_nw_addr(struct sk_buff *skb,
-				   const struct sw_flow_key *key,
-				   const struct nlattr *a)
+static struct sk_buff *set_nw_addr(struct sk_buff *skb, const struct nlattr *a)
 {
 	__be32 new_nwaddr = nla_get_be32(a);
 	struct iphdr *nh;
 	__sum16 *check;
 	__be32 *nwaddr;
 
-	if (unlikely(!is_ip(skb, key)))
+	if (unlikely(!is_ip(skb)))
 		return skb;
 
 	skb = make_writable(skb, 0);
@@ -245,21 +149,21 @@ static struct sk_buff *set_nw_addr(struct sk_buff *skb,
 	nh = ip_hdr(skb);
 	nwaddr = nla_type(a) == ODP_ACTION_ATTR_SET_NW_SRC ? &nh->saddr : &nh->daddr;
 
-	check = get_l4_checksum(skb, key);
+	check = get_l4_checksum(skb);
 	if (likely(check))
 		inet_proto_csum_replace4(check, skb, *nwaddr, new_nwaddr, 1);
 	csum_replace4(&nh->check, *nwaddr, new_nwaddr);
 
+	skb_clear_rxhash(skb);
+
 	*nwaddr = new_nwaddr;
 
 	return skb;
 }
 
-static struct sk_buff *set_nw_tos(struct sk_buff *skb,
-				  const struct sw_flow_key *key,
-				  u8 nw_tos)
+static struct sk_buff *set_nw_tos(struct sk_buff *skb, u8 nw_tos)
 {
-	if (unlikely(!is_ip(skb, key)))
+	if (unlikely(!is_ip(skb)))
 		return skb;
 
 	skb = make_writable(skb, 0);
@@ -278,15 +182,13 @@ static struct sk_buff *set_nw_tos(struct sk_buff *skb,
 	return skb;
 }
 
-static struct sk_buff *set_tp_port(struct sk_buff *skb,
-				   const struct sw_flow_key *key,
-				   const struct nlattr *a)
+static struct sk_buff *set_tp_port(struct sk_buff *skb, const struct nlattr *a)
 {
 	struct udphdr *th;
 	__sum16 *check;
 	__be16 *port;
 
-	if (unlikely(!is_ip(skb, key)))
+	if (unlikely(!is_ip(skb)))
 		return skb;
 
 	skb = make_writable(skb, 0);
@@ -294,7 +196,7 @@ static struct sk_buff *set_tp_port(struct sk_buff *skb,
 		return NULL;
 
 	/* Must follow make_writable() since that can move the skb data. */
-	check = get_l4_checksum(skb, key);
+	check = get_l4_checksum(skb);
 	if (unlikely(!check))
 		return skb;
 
@@ -309,6 +211,7 @@ static struct sk_buff *set_tp_port(struct sk_buff *skb,
 	port = nla_type(a) == ODP_ACTION_ATTR_SET_TP_SRC ? &th->source : &th->dest;
 	inet_proto_csum_replace2(check, skb, *port, nla_get_be16(a), 0);
 	*port = nla_get_be16(a);
+	skb_clear_rxhash(skb);
 
 	return skb;
 }
@@ -318,17 +221,16 @@ static struct sk_buff *set_tp_port(struct sk_buff *skb,
  *
  * @skb: skbuff containing an Ethernet packet, with network header pointing
  * just past the Ethernet and optional 802.1Q header.
- * @key: flow key extracted from @skb by flow_extract()
  *
  * Returns true if @skb is an invalid Ethernet+IPv4 ARP packet: one with screwy
  * or truncated header fields or one whose inner and outer Ethernet address
  * differ.
  */
-static bool is_spoofed_arp(struct sk_buff *skb, const struct sw_flow_key *key)
+static bool is_spoofed_arp(struct sk_buff *skb)
 {
 	struct arp_eth_header *arp;
 
-	if (key->dl_type != htons(ETH_P_ARP))
+	if (OVS_CB(skb)->flow->key.eth.type != htons(ETH_P_ARP))
 		return false;
 
 	if (skb_network_offset(skb) + sizeof(struct arp_eth_header) > skb->len)
@@ -360,8 +262,7 @@ error:
 	kfree_skb(skb);
 }
 
-static int output_control(struct datapath *dp, struct sk_buff *skb, u64 arg,
-			  const struct sw_flow_key *key)
+static int output_control(struct datapath *dp, struct sk_buff *skb, u64 arg)
 {
 	struct dp_upcall_info upcall;
 
@@ -370,7 +271,7 @@ static int output_control(struct datapath *dp, struct sk_buff *skb, u64 arg,
 		return -ENOMEM;
 
 	upcall.cmd = ODP_PACKET_CMD_ACTION;
-	upcall.key = key;
+	upcall.key = &OVS_CB(skb)->flow->key;
 	upcall.userdata = arg;
 	upcall.sample_pool = 0;
 	upcall.actions = NULL;
@@ -380,8 +281,7 @@ static int output_control(struct datapath *dp, struct sk_buff *skb, u64 arg,
 
 /* Execute a list of actions against 'skb'. */
 static int do_execute_actions(struct datapath *dp, struct sk_buff *skb,
-			      const struct sw_flow_key *key,
-			      const struct nlattr *actions, u32 actions_len)
+			      struct sw_flow_actions *acts)
 {
 	/* Every output action needs a separate clone of 'skb', but the common
 	 * case is just a single output action, so that doing a clone and
@@ -392,7 +292,8 @@ static int do_execute_actions(struct datapath *dp, struct sk_buff *skb,
 	const struct nlattr *a;
 	int rem, err;
 
-	for (a = actions, rem = actions_len; rem > 0; a = nla_next(a, &rem)) {
+	for (a = acts->actions, rem = acts->actions_len; rem > 0;
+	     a = nla_next(a, &rem)) {
 		if (prev_port != -1) {
 			do_output(dp, skb_clone(skb, GFP_ATOMIC), prev_port);
 			prev_port = -1;
@@ -404,7 +305,7 @@ static int do_execute_actions(struct datapath *dp, struct sk_buff *skb,
 			break;
 
 		case ODP_ACTION_ATTR_CONTROLLER:
-			err = output_control(dp, skb, nla_get_u64(a), key);
+			err = output_control(dp, skb, nla_get_u64(a));
 			if (err) {
 				kfree_skb(skb);
 				return err;
@@ -416,9 +317,7 @@ static int do_execute_actions(struct datapath *dp, struct sk_buff *skb,
 			break;
 
 		case ODP_ACTION_ATTR_SET_DL_TCI:
-			skb = modify_vlan_tci(dp, skb, key, a, rem);
-			if (IS_ERR(skb))
-				return PTR_ERR(skb);
+			skb = modify_vlan_tci(skb, nla_get_be16(a));
 			break;
 
 		case ODP_ACTION_ATTR_STRIP_VLAN:
@@ -441,16 +340,16 @@ static int do_execute_actions(struct datapath *dp, struct sk_buff *skb,
 
 		case ODP_ACTION_ATTR_SET_NW_SRC:
 		case ODP_ACTION_ATTR_SET_NW_DST:
-			skb = set_nw_addr(skb, key, a);
+			skb = set_nw_addr(skb, a);
 			break;
 
 		case ODP_ACTION_ATTR_SET_NW_TOS:
-			skb = set_nw_tos(skb, key, nla_get_u8(a));
+			skb = set_nw_tos(skb, nla_get_u8(a));
 			break;
 
 		case ODP_ACTION_ATTR_SET_TP_SRC:
 		case ODP_ACTION_ATTR_SET_TP_DST:
-			skb = set_tp_port(skb, key, a);
+			skb = set_tp_port(skb, a);
 			break;
 
 		case ODP_ACTION_ATTR_SET_PRIORITY:
@@ -462,7 +361,7 @@ static int do_execute_actions(struct datapath *dp, struct sk_buff *skb,
 			break;
 
 		case ODP_ACTION_ATTR_DROP_SPOOFED_ARP:
-			if (unlikely(is_spoofed_arp(skb, key)))
+			if (unlikely(is_spoofed_arp(skb)))
 				goto exit;
 			break;
 		}
@@ -478,8 +377,7 @@ exit:
 }
 
 static void sflow_sample(struct datapath *dp, struct sk_buff *skb,
-			 const struct sw_flow_key *key,
-			 const struct nlattr *a, u32 actions_len)
+			 struct sw_flow_actions *acts)
 {
 	struct sk_buff *nskb;
 	struct vport *p = OVS_CB(skb)->vport;
@@ -497,23 +395,46 @@ static void sflow_sample(struct datapath *dp, struct sk_buff *skb,
 		return;
 
 	upcall.cmd = ODP_PACKET_CMD_SAMPLE;
-	upcall.key = key;
+	upcall.key = &OVS_CB(skb)->flow->key;
 	upcall.userdata = 0;
 	upcall.sample_pool = atomic_read(&p->sflow_pool);
-	upcall.actions = a;
-	upcall.actions_len = actions_len;
+	upcall.actions = acts->actions;
+	upcall.actions_len = acts->actions_len;
 	dp_upcall(dp, nskb, &upcall);
 }
 
 /* Execute a list of actions against 'skb'. */
-int execute_actions(struct datapath *dp, struct sk_buff *skb,
-		    const struct sw_flow_key *key,
-		    const struct nlattr *actions, u32 actions_len)
+int execute_actions(struct datapath *dp, struct sk_buff *skb)
 {
-	if (dp->sflow_probability)
-		sflow_sample(dp, skb, key, actions, actions_len);
+	struct sw_flow_actions *acts = rcu_dereference(OVS_CB(skb)->flow->sf_acts);
+	struct loop_counter *loop;
+	int error;
+
+	/* Check whether we've looped too much. */
+	loop = loop_get_counter();
+	if (unlikely(++loop->count > MAX_LOOPS))
+		loop->looping = true;
+	if (unlikely(loop->looping)) {
+		error = loop_suppress(dp, acts);
+		kfree_skb(skb);
+		goto out_loop;
+	}
 
+	/* Really execute actions. */
+	if (dp->sflow_probability)
+		sflow_sample(dp, skb, acts);
 	OVS_CB(skb)->tun_id = 0;
+	error = do_execute_actions(dp, skb, acts);
+
+	/* Check whether sub-actions looped too much. */
+	if (unlikely(loop->looping))
+		error = loop_suppress(dp, acts);
+
+out_loop:
+	/* Decrement loop counter. */
+	if (!--loop->count)
+		loop->looping = false;
+	loop_put_counter();
 
-	return do_execute_actions(dp, skb, key, actions, actions_len);
+	return error;
 }