X-Git-Url: http://git.onelab.eu/?a=blobdiff_plain;f=datapath%2Factions.c;h=fccacf217cfc2473e5b00a89313a51844aeabd1b;hb=26c3f94a553be66dd037159148f30b47ef2c6f4b;hp=8a3e8abb549e1302bdbed626bd889728bf2854f5;hpb=39fb08818bbd9c438dbf23caa89937c663451b5a;p=sliver-openvswitch.git

diff --git a/datapath/actions.c b/datapath/actions.c
index 8a3e8abb5..fccacf217 100644
--- a/datapath/actions.c
+++ b/datapath/actions.c
@@ -1,6 +1,6 @@
 /*
  * Distributed under the terms of the GNU GPL version 2.
- * Copyright (c) 2007, 2008, 2009 Nicira Networks.
+ * Copyright (c) 2007, 2008, 2009, 2010 Nicira Networks.
  *
  * Significant portions of this file may be copied from parts of the Linux
  * kernel, by Linus Torvalds and others.
@@ -22,12 +22,28 @@
 #include "actions.h"
 #include "openvswitch/datapath-protocol.h"
 
-struct sk_buff *
-make_writable(struct sk_buff *skb, gfp_t gfp)
+static struct sk_buff *
+make_writable(struct sk_buff *skb, unsigned min_headroom, gfp_t gfp)
 {
 	if (skb_shared(skb) || skb_cloned(skb)) {
-		struct sk_buff *nskb = skb_copy(skb, gfp);
+		struct sk_buff *nskb;
+		unsigned headroom = max(min_headroom, skb_headroom(skb));
+
+		nskb = skb_copy_expand(skb, headroom, skb_tailroom(skb), gfp);
 		if (nskb) {
+#if LINUX_VERSION_CODE < KERNEL_VERSION(2,6,24)
+			/* Before 2.6.24 these fields were not copied when
+			 * doing an skb_copy_expand. */
+			nskb->ip_summed = skb->ip_summed;
+			nskb->csum = skb->csum;
+#endif
+#if defined(CONFIG_XEN) && defined(HAVE_PROTO_DATA_VALID)
+			/* These fields are copied in skb_clone but not in
+			 * skb_copy or related functions.  We need to manually
+			 * copy them over here. */
+			nskb->proto_data_valid = skb->proto_data_valid;
+			nskb->proto_csum_blank = skb->proto_csum_blank;
+#endif
 			kfree_skb(skb);
 			return nskb;
 		}
@@ -80,7 +96,7 @@ modify_vlan_tci(struct datapath *dp, struct sk_buff *skb,
 		mask = VLAN_PCP_MASK;
 	}
 
-	skb = make_writable(skb, gfp);
+	skb = make_writable(skb, VLAN_HLEN, gfp);
 	if (!skb)
 		return ERR_PTR(-ENOMEM);
 
@@ -168,7 +184,7 @@ modify_vlan_tci(struct datapath *dp, struct sk_buff *skb,
 static struct sk_buff *strip_vlan(struct sk_buff *skb,
 				  struct odp_flow_key *key, gfp_t gfp)
 {
-	skb = make_writable(skb, gfp);
+	skb = make_writable(skb, 0, gfp);
 	if (skb) {
 		vlan_pull_tag(skb);
 		key->dl_vlan = htons(ODP_VLAN_NONE);
@@ -180,7 +196,7 @@ static struct sk_buff *set_dl_addr(struct sk_buff *skb,
 				   const struct odp_action_dl_addr *a,
 				   gfp_t gfp)
 {
-	skb = make_writable(skb, gfp);
+	skb = make_writable(skb, 0, gfp);
 	if (skb) {
 		struct ethhdr *eh = eth_hdr(skb);
 		memcpy(a->type == ODPAT_SET_DL_SRC ? eh->h_source : eh->h_dest,
@@ -197,10 +213,43 @@ static void update_csum(__sum16 *sum, struct sk_buff *skb,
 			__be32 from, __be32 to, int pseudohdr)
 {
 	__be32 diff[] = { ~from, to };
-	if (skb->ip_summed != CHECKSUM_PARTIAL) {
+
+/* On older kernels, CHECKSUM_PARTIAL and CHECKSUM_COMPLETE are both defined
+ * as CHECKSUM_HW.  However, we can make some inferences so that we can update
+ * the checksums appropriately. */
+	enum {
+		CSUM_PARTIAL,	/* Partial checksum, skb->csum undefined. */
+		CSUM_PACKET,	/* In-packet checksum, skb->csum undefined. */
+		CSUM_COMPLETE,	/* In-packet checksum, skb->csum valid. */
+	} csum_type;
+
+	csum_type = CSUM_PACKET;
+#ifndef CHECKSUM_HW
+	/* Newer kernel, just map between kernel types and ours. */
+	if (skb->ip_summed == CHECKSUM_PARTIAL)
+		csum_type = CSUM_PARTIAL;
+	else if (skb->ip_summed == CHECKSUM_COMPLETE)
+		csum_type = CSUM_COMPLETE;
+#else
+	/* In theory this could be either CHECKSUM_PARTIAL or CHECKSUM_COMPLETE.
+	 * However, we should only get CHECKSUM_PARTIAL packets from Xen, which
+	 * uses some special fields to represent this (see below).  Since we
+	 * can only make one type work, pick the one that actually happens in
+	 * practice. */
+	if (skb->ip_summed == CHECKSUM_HW)
+		csum_type = CSUM_COMPLETE;
+#endif
+#if defined(CONFIG_XEN) && defined(HAVE_PROTO_DATA_VALID)
+	/* Xen has a special way of representing CHECKSUM_PARTIAL on older
+	 * kernels. */
+	if (skb->proto_csum_blank)
+		csum_type = CSUM_PARTIAL;
+#endif
+
+	if (csum_type != CSUM_PARTIAL) {
 		*sum = csum_fold(csum_partial((char *)diff, sizeof(diff),
 				~csum_unfold(*sum)));
-		if (skb->ip_summed == CHECKSUM_COMPLETE && pseudohdr)
+		if (csum_type == CSUM_COMPLETE && pseudohdr)
 			skb->csum = ~csum_partial((char *)diff, sizeof(diff),
 						~skb->csum);
 	} else if (pseudohdr)
@@ -216,7 +265,7 @@ static struct sk_buff *set_nw_addr(struct sk_buff *skb,
 	if (key->dl_type != htons(ETH_P_IP))
 		return skb;
 
-	skb = make_writable(skb, gfp);
+	skb = make_writable(skb, 0, gfp);
 	if (skb) {
 		struct iphdr *nh = ip_hdr(skb);
 		u32 *f = a->type == ODPAT_SET_NW_SRC ? &nh->saddr : &nh->daddr;
@@ -236,6 +285,30 @@ static struct sk_buff *set_nw_addr(struct sk_buff *skb,
 	return skb;
 }
 
+static struct sk_buff *set_nw_tos(struct sk_buff *skb,
+				   struct odp_flow_key *key,
+				   const struct odp_action_nw_tos *a,
+				   gfp_t gfp)
+{
+	if (key->dl_type != htons(ETH_P_IP))
+		return skb;
+
+	skb = make_writable(skb, 0, gfp);
+	if (skb) {
+		struct iphdr *nh = ip_hdr(skb);
+		u8 *f = &nh->tos;
+		u8 old = *f;
+
+		/* We only set the lower 6 bits. */
+		u8 new = (a->nw_tos & 0x3f) | (nh->tos & 0xc0);
+
+		update_csum(&nh->check, skb, htons((uint16_t)old),
+				htons((uint16_t)new), 0);
+		*f = new;
+	}
+	return skb;
+}
+
 static struct sk_buff *
 set_tp_port(struct sk_buff *skb, struct odp_flow_key *key,
 	    const struct odp_action_tp_port *a,
@@ -253,14 +326,14 @@ set_tp_port(struct sk_buff *skb, struct odp_flow_key *key,
 	else
 		return skb;
 
-	skb = make_writable(skb, gfp);
+	skb = make_writable(skb, 0, gfp);
 	if (skb) {
 		struct udphdr *th = udp_hdr(skb);
 		u16 *f = a->type == ODPAT_SET_TP_SRC ? &th->source : &th->dest;
 		u16 old = *f;
 		u16 new = a->tp_port;
-		update_csum((u16*)((u8*)skb->data + check_ofs),
-			    skb, old, new, 1);
+		update_csum((u16*)(skb_transport_header(skb) + check_ofs), 
+				skb, old, new, 0);
 		*f = new;
 	}
 	return skb;
@@ -286,6 +359,7 @@ int dp_xmit_skb(struct sk_buff *skb)
 		return -E2BIG;
 	}
 
+	forward_ip_summed(skb);
 	dev_queue_xmit(skb);
 
 	return len;
@@ -307,7 +381,7 @@ do_output(struct datapath *dp, struct sk_buff *skb, int out_port)
 	dev = skb->dev = p->dev;
 	if (is_dp_dev(dev))
 		dp_dev_recv(dev, skb);
-        else
+	else
 		dp_xmit_skb(skb);
 	return;
 
@@ -350,6 +424,28 @@ output_control(struct datapath *dp, struct sk_buff *skb, u32 arg, gfp_t gfp)
 	return dp_output_control(dp, skb, _ODPL_ACTION_NR, arg);
 }
 
+/* Send a copy of this packet up to the sFlow agent, along with extra
+ * information about what happened to it. */
+static void sflow_sample(struct datapath *dp, struct sk_buff *skb,
+			 const union odp_action *a, int n_actions,
+			 gfp_t gfp, struct net_bridge_port *nbp)
+{
+	struct odp_sflow_sample_header *hdr;
+	unsigned int actlen = n_actions * sizeof(union odp_action);
+	unsigned int hdrlen = sizeof(struct odp_sflow_sample_header);
+	struct sk_buff *nskb;
+
+	nskb = skb_copy_expand(skb, actlen + hdrlen, 0, gfp);
+	if (!nskb)
+		return;
+
+	memcpy(__skb_push(nskb, actlen), a, actlen);
+	hdr = (struct odp_sflow_sample_header*)__skb_push(nskb, hdrlen);
+	hdr->n_actions = n_actions;
+	hdr->sample_pool = atomic_read(&nbp->sflow_pool);
+	dp_output_control(dp, nskb, _ODPL_SFLOW_NR, 0);
+}
+
 /* Execute a list of actions against 'skb'. */
 int execute_actions(struct datapath *dp, struct sk_buff *skb,
 		    struct odp_flow_key *key,
@@ -362,6 +458,17 @@ int execute_actions(struct datapath *dp, struct sk_buff *skb,
 	 * is slightly obscure just to avoid that. */
 	int prev_port = -1;
 	int err;
+
+	if (dp->sflow_probability) {
+		struct net_bridge_port *p = skb->dev->br_port;
+		if (p) {
+			atomic_inc(&p->sflow_pool);
+			if (dp->sflow_probability == UINT_MAX ||
+			    net_random() < dp->sflow_probability)
+				sflow_sample(dp, skb, a, n_actions, gfp, p);
+		}
+	}
+
 	for (; n_actions > 0; a++, n_actions--) {
 		WARN_ON_ONCE(skb_shared(skb));
 		if (prev_port != -1) {
@@ -408,6 +515,10 @@ int execute_actions(struct datapath *dp, struct sk_buff *skb,
 			skb = set_nw_addr(skb, key, &a->nw_addr, gfp);
 			break;
 
+		case ODPAT_SET_NW_TOS:
+			skb = set_nw_tos(skb, key, &a->nw_tos, gfp);
+			break;
+
 		case ODPAT_SET_TP_SRC:
 		case ODPAT_SET_TP_DST:
 			skb = set_tp_port(skb, key, &a->tp_port, gfp);