[ovs.git] / datapath / actions.c

/*
 * Distributed under the terms of the GNU GPL version 2.
 * 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.
 */

/* Functions for executing flow actions. */

#include <linux/skbuff.h>
#include <linux/in.h>
#include <linux/ip.h>
#include <linux/tcp.h>
#include <linux/udp.h>
#include <linux/in6.h>
#include <linux/if_vlan.h>
#include <net/inet_ecn.h>
#include <net/ip.h>
#include <net/checksum.h>

#include "actions.h"
#include "datapath.h"
#include "openvswitch/datapath-protocol.h"
#include "vport.h"

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;
		unsigned headroom = max(min_headroom, skb_headroom(skb));

		nskb = skb_copy_expand(skb, headroom, skb_tailroom(skb), gfp);
		if (nskb) {
			set_skb_csum_bits(skb, nskb);
			kfree_skb(skb);
			return nskb;
		}
	} else {
		unsigned int hdr_len = (skb_transport_offset(skb)
					+ sizeof(struct tcphdr));
		if (pskb_may_pull(skb, min(hdr_len, skb->len)))
			return skb;
	}
	kfree_skb(skb);
	return NULL;
}

static void set_tunnel(struct sk_buff *skb, struct odp_flow_key *key,
		       __be32 tun_id)
{
	OVS_CB(skb)->tun_id = key->tun_id = tun_id;
}

static struct sk_buff *
vlan_pull_tag(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))
		return skb;

	if (OVS_CB(skb)->ip_summed == 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);

	eh = (struct ethhdr *)skb_pull(skb, VLAN_HLEN);

	skb->protocol = eh->h_proto;
	skb->mac_header += VLAN_HLEN;

	return skb;
}


static struct sk_buff *
modify_vlan_tci(struct datapath *dp, struct sk_buff *skb,
		struct odp_flow_key *key, const union odp_action *a,
		int n_actions, gfp_t gfp)
{
	u16 tci, mask;

	if (a->type == ODPAT_SET_VLAN_VID) {
		tci = ntohs(a->vlan_vid.vlan_vid);
		mask = VLAN_VID_MASK;
		key->dl_vlan = a->vlan_vid.vlan_vid;
	} else {
		tci = a->vlan_pcp.vlan_pcp << VLAN_PCP_SHIFT;
		mask = VLAN_PCP_MASK;
		key->dl_vlan_pcp = a->vlan_pcp.vlan_pcp;
	}

	skb = make_writable(skb, VLAN_HLEN, gfp);
	if (!skb)
		return ERR_PTR(-ENOMEM);

	if (skb->protocol == htons(ETH_P_8021Q)) {
		/* Modify vlan id, but maintain other TCI values */
		struct vlan_ethhdr *vh = vlan_eth_hdr(skb);
		__be16 old_tci = vh->h_vlan_TCI;

		vh->h_vlan_TCI = htons((ntohs(vh->h_vlan_TCI) & ~mask) | tci);

		if (OVS_CB(skb)->ip_summed == OVS_CSUM_COMPLETE) {
			__be16 diff[] = { ~old_tci, vh->h_vlan_TCI };

			skb->csum = ~csum_partial((char *)diff, sizeof(diff),
						~skb->csum);
		}
	} else {
		/* Add vlan header */

		/* 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. */
		vswitch_skb_checksum_setup(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)) {
			struct sk_buff *segs;

			segs = skb_gso_segment(skb, 0);
			kfree_skb(skb);
			if (unlikely(IS_ERR(segs)))
				return ERR_CAST(segs);

			do {
				struct sk_buff *nskb = segs->next;
				int err;

				segs->next = NULL;

				/* GSO can change the checksum type so update.*/
				compute_ip_summed(segs, true);

				segs = __vlan_put_tag(segs, tci);
				err = -ENOMEM;
				if (segs) {
					struct odp_flow_key segkey = *key;
					err = execute_actions(dp, segs,
							      &segkey, a + 1,
							      n_actions - 1,
							      gfp);
				}

				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);
		}

		/* 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, tci);
		if (!skb)
			return ERR_PTR(-ENOMEM);

		/* GSO doesn't fix up the hardware computed checksum so this
		 * will only be hit in the non-GSO case. */
		if (OVS_CB(skb)->ip_summed == OVS_CSUM_COMPLETE)
			skb->csum = csum_add(skb->csum, csum_partial(skb->data
						+ ETH_HLEN, VLAN_HLEN, 0));
	}

	return skb;
}

static struct sk_buff *strip_vlan(struct sk_buff *skb,
				  struct odp_flow_key *key, gfp_t gfp)
{
	skb = make_writable(skb, 0, gfp);
	if (skb) {
		vlan_pull_tag(skb);
		key->dl_vlan = htons(ODP_VLAN_NONE);
	}
	return skb;
}

static struct sk_buff *set_dl_addr(struct sk_buff *skb,
				   struct odp_flow_key *key,
				   const struct odp_action_dl_addr *a,
				   gfp_t gfp)
{
	skb = make_writable(skb, 0, gfp);
	if (skb) {
		struct ethhdr *eh = eth_hdr(skb);
		if (a->type == ODPAT_SET_DL_SRC) {
			memcpy(eh->h_source, a->dl_addr, ETH_ALEN);
			memcpy(key->dl_src, a->dl_addr, ETH_ALEN);
		} else {
			memcpy(eh->h_dest, a->dl_addr, ETH_ALEN);
			memcpy(key->dl_dst, a->dl_addr, ETH_ALEN);
		}
	}
	return skb;
}

/* Updates 'sum', which is a field in 'skb''s data, given that a 4-byte field
 * covered by the sum has been changed from 'from' to 'to'.  If set,
 * 'pseudohdr' indicates that the field is in the TCP or UDP pseudo-header.
 * Based on nf_proto_csum_replace4. */
static void update_csum(__sum16 *sum, struct sk_buff *skb,
			__be32 from, __be32 to, int pseudohdr)
{
	__be32 diff[] = { ~from, to };

	if (OVS_CB(skb)->ip_summed != OVS_CSUM_PARTIAL) {
		*sum = csum_fold(csum_partial((char *)diff, sizeof(diff),
				~csum_unfold(*sum)));
		if (OVS_CB(skb)->ip_summed == OVS_CSUM_COMPLETE && pseudohdr)
			skb->csum = ~csum_partial((char *)diff, sizeof(diff),
						~skb->csum);
	} else if (pseudohdr)
		*sum = ~csum_fold(csum_partial((char *)diff, sizeof(diff),
				csum_unfold(*sum)));
}

static struct sk_buff *set_nw_addr(struct sk_buff *skb,
				   struct odp_flow_key *key,
				   const struct odp_action_nw_addr *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);
		u32 *f = a->type == ODPAT_SET_NW_SRC ? &nh->saddr : &nh->daddr;
		u32 old = *f;
		u32 new = a->nw_addr;

		if (key->nw_proto == IPPROTO_TCP) {
			struct tcphdr *th = tcp_hdr(skb);
			update_csum(&th->check, skb, old, new, 1);
		} else if (key->nw_proto == IPPROTO_UDP) {
			struct udphdr *th = udp_hdr(skb);
			update_csum(&th->check, skb, old, new, 1);
		}
		update_csum(&nh->check, skb, old, new, 0);
		*f = new;

		if (a->type == ODPAT_SET_NW_SRC)
			key->nw_src = a->nw_addr;
		else
			key->nw_dst = a->nw_addr;
	}
	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;
		u8 new;

		/* Set the DSCP bits and preserve the ECN bits. */
		new = a->nw_tos | (nh->tos & INET_ECN_MASK);
		update_csum(&nh->check, skb, htons((uint16_t)old),
				htons((uint16_t)new), 0);
		*f = new;
		key->nw_tos = a->nw_tos;
	}
	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,
	    gfp_t gfp)
{
	int check_ofs;

	if (key->dl_type != htons(ETH_P_IP))
		return skb;

	if (key->nw_proto == IPPROTO_TCP)
		check_ofs = offsetof(struct tcphdr, check);
	else if (key->nw_proto == IPPROTO_UDP)
		check_ofs = offsetof(struct udphdr, check);
	else
		return skb;

	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*)(skb_transport_header(skb) + check_ofs), 
				skb, old, new, 0);
		*f = new;
		if (a->type == ODPAT_SET_TP_SRC)
			key->tp_src = a->tp_port;
		else
			key->tp_dst = a->tp_port;
	}
	return skb;
}

static inline unsigned packet_length(const struct sk_buff *skb)
{
	unsigned length = skb->len - ETH_HLEN;
	if (skb->protocol == htons(ETH_P_8021Q))
		length -= VLAN_HLEN;
	return length;
}

static void
do_output(struct datapath *dp, struct sk_buff *skb, int out_port)
{
	struct dp_port *p;
	int mtu;

	if (!skb)
		goto error;

	p = rcu_dereference(dp->ports[out_port]);
	if (!p)
		goto error;

	mtu = vport_get_mtu(p->vport);
	if (packet_length(skb) > mtu && !skb_is_gso(skb)) {
		printk(KERN_WARNING "%s: dropped over-mtu packet: %d > %d\n",
		       dp_name(dp), packet_length(skb), mtu);
		goto error;
	}

	vport_send(p->vport, skb);
	return;

error:
	kfree_skb(skb);
}

/* Never consumes 'skb'.  Returns a port that 'skb' should be sent to, -1 if
 * none.  */
static int output_group(struct datapath *dp, __u16 group,
			struct sk_buff *skb, gfp_t gfp)
{
	struct dp_port_group *g = rcu_dereference(dp->groups[group]);
	int prev_port = -1;
	int i;

	if (!g)
		return -1;
	for (i = 0; i < g->n_ports; i++) {
		struct dp_port *p = rcu_dereference(dp->ports[g->ports[i]]);
		if (!p || OVS_CB(skb)->dp_port == p)
			continue;
		if (prev_port != -1) {
			struct sk_buff *clone = skb_clone(skb, gfp);
			if (!clone)
				return -1;
			do_output(dp, clone, prev_port);
		}
		prev_port = p->port_no;
	}
	return prev_port;
}

static int
output_control(struct datapath *dp, struct sk_buff *skb, u32 arg, gfp_t gfp)
{
	skb = skb_clone(skb, gfp);
	if (!skb)
		return -ENOMEM;
	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 dp_port *dp_port)
{
	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(&dp_port->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,
		    const union odp_action *a, int n_actions,
		    gfp_t gfp)
{
	/* 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
	 * then freeing the original skbuff is wasteful.  So the following code
	 * is slightly obscure just to avoid that. */
	int prev_port = -1;
	u32 priority = skb->priority;
	int err;

	if (dp->sflow_probability) {
		struct dp_port *p = OVS_CB(skb)->dp_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);
		}
	}

	OVS_CB(skb)->tun_id = 0;

	for (; n_actions > 0; a++, n_actions--) {
		WARN_ON_ONCE(skb_shared(skb));
		if (prev_port != -1) {
			do_output(dp, skb_clone(skb, gfp), prev_port);
			prev_port = -1;
		}

		switch (a->type) {
		case ODPAT_OUTPUT:
			prev_port = a->output.port;
			break;

		case ODPAT_OUTPUT_GROUP:
			prev_port = output_group(dp, a->output_group.group,
						 skb, gfp);
			break;

		case ODPAT_CONTROLLER:
			err = output_control(dp, skb, a->controller.arg, gfp);
			if (err) {
				kfree_skb(skb);
				return err;
			}
			break;

		case ODPAT_SET_TUNNEL:
			set_tunnel(skb, key, a->tunnel.tun_id);
			break;

		case ODPAT_SET_VLAN_VID:
		case ODPAT_SET_VLAN_PCP:
			skb = modify_vlan_tci(dp, skb, key, a, n_actions, gfp);
			if (IS_ERR(skb))
				return PTR_ERR(skb);
			break;

		case ODPAT_STRIP_VLAN:
			skb = strip_vlan(skb, key, gfp);
			break;

		case ODPAT_SET_DL_SRC:
		case ODPAT_SET_DL_DST:
			skb = set_dl_addr(skb, key, &a->dl_addr, gfp);
			break;

		case ODPAT_SET_NW_SRC:
		case ODPAT_SET_NW_DST:
			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);
			break;

		case ODPAT_SET_PRIORITY:
			skb->priority = a->priority.priority;
			break;

		case ODPAT_POP_PRIORITY:
			skb->priority = priority;
			break;
		}
		if (!skb)
			return -ENOMEM;
	}
	if (prev_port != -1)
		do_output(dp, skb, prev_port);
	else
		kfree_skb(skb);
	return 0;
}
Commit	Line	Data
064af421 BP	1	/*
064af421 BP	2	* Distributed under the terms of the GNU GPL version 2.
00ba5f3b	3	* Copyright (c) 2007, 2008, 2009, 2010 Nicira Networks.
a14bc59f BP	4	*
	5	* Significant portions of this file may be copied from parts of the Linux
	6	* kernel, by Linus Torvalds and others.
064af421 BP	7	*/
	8
	9	/* Functions for executing flow actions. */
	10
	11	#include <linux/skbuff.h>
	12	#include <linux/in.h>
	13	#include <linux/ip.h>
	14	#include <linux/tcp.h>
	15	#include <linux/udp.h>
	16	#include <linux/in6.h>
	17	#include <linux/if_vlan.h>
f1193301	18	#include <net/inet_ecn.h>
064af421 BP	19	#include <net/ip.h>
064af421 BP	20	#include <net/checksum.h>
f2459fe7	21
064af421	22	#include "actions.h"
f2459fe7	23	#include "datapath.h"
064af421	24	#include "openvswitch/datapath-protocol.h"
f2459fe7	25	#include "vport.h"
064af421	26
0cd8a05e JG	27	static struct sk_buff *
0cd8a05e JG	28	make_writable(struct sk_buff *skb, unsigned min_headroom, gfp_t gfp)
064af421 BP	29	{
064af421 BP	30	if (skb_shared(skb) \|\| skb_cloned(skb)) {
0cd8a05e JG	31	struct sk_buff *nskb;
	32	unsigned headroom = max(min_headroom, skb_headroom(skb));
	33
	34	nskb = skb_copy_expand(skb, headroom, skb_tailroom(skb), gfp);
064af421	35	if (nskb) {
ff6402a9	36	set_skb_csum_bits(skb, nskb);
064af421 BP	37	kfree_skb(skb);
	38	return nskb;
	39	}
	40	} else {
	41	unsigned int hdr_len = (skb_transport_offset(skb)
	42	+ sizeof(struct tcphdr));
	43	if (pskb_may_pull(skb, min(hdr_len, skb->len)))
	44	return skb;
	45	}
	46	kfree_skb(skb);
	47	return NULL;
	48	}
	49
659586ef JG	50	static void set_tunnel(struct sk_buff skb, struct odp_flow_key key,
	51	__be32 tun_id)
	52	{
	53	OVS_CB(skb)->tun_id = key->tun_id = tun_id;
	54	}
064af421 BP	55
	56	static struct sk_buff *
	57	vlan_pull_tag(struct sk_buff *skb)
	58	{
	59	struct vlan_ethhdr *vh = vlan_eth_hdr(skb);
	60	struct ethhdr *eh;
	61
064af421 BP	62	/* Verify we were given a vlan packet */
	63	if (vh->h_vlan_proto != htons(ETH_P_8021Q))
	64	return skb;
	65
635c9298 JG	66	if (OVS_CB(skb)->ip_summed == OVS_CSUM_COMPLETE)
	67	skb->csum = csum_sub(skb->csum, csum_partial(skb->data
	68	+ ETH_HLEN, VLAN_HLEN, 0));
	69
064af421 BP	70	memmove(skb->data + VLAN_HLEN, skb->data, 2 * VLAN_ETH_ALEN);
	71
	72	eh = (struct ethhdr *)skb_pull(skb, VLAN_HLEN);
	73
	74	skb->protocol = eh->h_proto;
	75	skb->mac_header += VLAN_HLEN;
	76
	77	return skb;
	78	}
	79
	80
	81	static struct sk_buff *
	82	modify_vlan_tci(struct datapath dp, struct sk_buff skb,
	83	struct odp_flow_key key, const union odp_action a,
	84	int n_actions, gfp_t gfp)
	85	{
	86	u16 tci, mask;
	87
	88	if (a->type == ODPAT_SET_VLAN_VID) {
	89	tci = ntohs(a->vlan_vid.vlan_vid);
	90	mask = VLAN_VID_MASK;
11cdf5e6	91	key->dl_vlan = a->vlan_vid.vlan_vid;
064af421	92	} else {
d42c4f8d	93	tci = a->vlan_pcp.vlan_pcp << VLAN_PCP_SHIFT;
064af421	94	mask = VLAN_PCP_MASK;
11cdf5e6	95	key->dl_vlan_pcp = a->vlan_pcp.vlan_pcp;
064af421 BP	96	}
064af421 BP	97
0cd8a05e	98	skb = make_writable(skb, VLAN_HLEN, gfp);
064af421 BP	99	if (!skb)
	100	return ERR_PTR(-ENOMEM);
	101
	102	if (skb->protocol == htons(ETH_P_8021Q)) {
	103	/* Modify vlan id, but maintain other TCI values */
	104	struct vlan_ethhdr *vh = vlan_eth_hdr(skb);
635c9298 JG	105	__be16 old_tci = vh->h_vlan_TCI;
635c9298 JG	106
064af421	107	vh->h_vlan_TCI = htons((ntohs(vh->h_vlan_TCI) & ~mask) \| tci);
635c9298 JG	108
	109	if (OVS_CB(skb)->ip_summed == OVS_CSUM_COMPLETE) {
	110	__be16 diff[] = { ~old_tci, vh->h_vlan_TCI };
	111
	112	skb->csum = ~csum_partial((char *)diff, sizeof(diff),
	113	~skb->csum);
	114	}
064af421 BP	115	} else {
	116	/* Add vlan header */
	117
	118	/* Set up checksumming pointers for checksum-deferred packets
	119	* on Xen. Otherwise, dev_queue_xmit() will try to do this
	120	* when we send the packet out on the wire, and it will fail at
	121	* that point because skb_checksum_setup() will not look inside
	122	* an 802.1Q header. */
b2f460c7	123	vswitch_skb_checksum_setup(skb);
064af421 BP	124
	125	/* GSO is not implemented for packets with an 802.1Q header, so
	126	* we have to do segmentation before we add that header.
	127	*
	128	* GSO does work with hardware-accelerated VLAN tagging, but we
	129	* can't use hardware-accelerated VLAN tagging since it
	130	* requires the device to have a VLAN group configured (with
	131	* e.g. vconfig(8)) and we don't do that.
	132	*
	133	* Having to do this here may be a performance loss, since we
	134	* can't take advantage of TSO hardware support, although it
	135	* does not make a measurable network performance difference
	136	* for 1G Ethernet. Fixing that would require patching the
	137	* kernel (either to add GSO support to the VLAN protocol or to
	138	* support hardware-accelerated VLAN tagging without VLAN
	139	* groups configured). */
	140	if (skb_is_gso(skb)) {
	141	struct sk_buff *segs;
	142
	143	segs = skb_gso_segment(skb, 0);
	144	kfree_skb(skb);
	145	if (unlikely(IS_ERR(segs)))
	146	return ERR_CAST(segs);
	147
	148	do {
	149	struct sk_buff *nskb = segs->next;
	150	int err;
	151
	152	segs->next = NULL;
	153
635c9298 JG	154	/* GSO can change the checksum type so update.*/
	155	compute_ip_summed(segs, true);
	156
064af421 BP	157	segs = __vlan_put_tag(segs, tci);
	158	err = -ENOMEM;
	159	if (segs) {
	160	struct odp_flow_key segkey = *key;
	161	err = execute_actions(dp, segs,
	162	&segkey, a + 1,
	163	n_actions - 1,
	164	gfp);
	165	}
	166
	167	if (unlikely(err)) {
	168	while ((segs = nskb)) {
	169	nskb = segs->next;
	170	segs->next = NULL;
	171	kfree_skb(segs);
	172	}
	173	return ERR_PTR(err);
	174	}
	175
	176	segs = nskb;
	177	} while (segs->next);
	178
	179	skb = segs;
635c9298	180	compute_ip_summed(skb, true);
064af421 BP	181	}
	182
	183	/* The hardware-accelerated version of vlan_put_tag() works
	184	* only for a device that has a VLAN group configured (with
	185	* e.g. vconfig(8)), so call the software-only version
	186	* __vlan_put_tag() directly instead.
	187	*/
	188	skb = __vlan_put_tag(skb, tci);
	189	if (!skb)
	190	return ERR_PTR(-ENOMEM);
635c9298 JG	191
	192	/* GSO doesn't fix up the hardware computed checksum so this
	193	* will only be hit in the non-GSO case. */
	194	if (OVS_CB(skb)->ip_summed == OVS_CSUM_COMPLETE)
	195	skb->csum = csum_add(skb->csum, csum_partial(skb->data
	196	+ ETH_HLEN, VLAN_HLEN, 0));
064af421 BP	197	}
	198
	199	return skb;
	200	}
	201
	202	static struct sk_buff strip_vlan(struct sk_buff skb,
	203	struct odp_flow_key *key, gfp_t gfp)
	204	{
0cd8a05e	205	skb = make_writable(skb, 0, gfp);
064af421 BP	206	if (skb) {
	207	vlan_pull_tag(skb);
	208	key->dl_vlan = htons(ODP_VLAN_NONE);
	209	}
	210	return skb;
	211	}
	212
	213	static struct sk_buff set_dl_addr(struct sk_buff skb,
11cdf5e6	214	struct odp_flow_key *key,
064af421 BP	215	const struct odp_action_dl_addr *a,
	216	gfp_t gfp)
	217	{
0cd8a05e	218	skb = make_writable(skb, 0, gfp);
064af421 BP	219	if (skb) {
064af421 BP	220	struct ethhdr *eh = eth_hdr(skb);
11cdf5e6 JG	221	if (a->type == ODPAT_SET_DL_SRC) {
	222	memcpy(eh->h_source, a->dl_addr, ETH_ALEN);
	223	memcpy(key->dl_src, a->dl_addr, ETH_ALEN);
	224	} else {
	225	memcpy(eh->h_dest, a->dl_addr, ETH_ALEN);
	226	memcpy(key->dl_dst, a->dl_addr, ETH_ALEN);
	227	}
064af421 BP	228	}
	229	return skb;
	230	}
	231
	232	/* Updates 'sum', which is a field in 'skb''s data, given that a 4-byte field
	233	* covered by the sum has been changed from 'from' to 'to'. If set,
	234	* 'pseudohdr' indicates that the field is in the TCP or UDP pseudo-header.
	235	* Based on nf_proto_csum_replace4. */
	236	static void update_csum(__sum16 sum, struct sk_buff skb,
	237	__be32 from, __be32 to, int pseudohdr)
	238	{
	239	__be32 diff[] = { ~from, to };
1cdb82b9	240
635c9298	241	if (OVS_CB(skb)->ip_summed != OVS_CSUM_PARTIAL) {
064af421 BP	242	sum = csum_fold(csum_partial((char )diff, sizeof(diff),
064af421 BP	243	~csum_unfold(*sum)));
635c9298	244	if (OVS_CB(skb)->ip_summed == OVS_CSUM_COMPLETE && pseudohdr)
064af421 BP	245	skb->csum = ~csum_partial((char *)diff, sizeof(diff),
	246	~skb->csum);
	247	} else if (pseudohdr)
	248	sum = ~csum_fold(csum_partial((char )diff, sizeof(diff),
	249	csum_unfold(*sum)));
	250	}
	251
	252	static struct sk_buff set_nw_addr(struct sk_buff skb,
	253	struct odp_flow_key *key,
	254	const struct odp_action_nw_addr *a,
	255	gfp_t gfp)
	256	{
	257	if (key->dl_type != htons(ETH_P_IP))
	258	return skb;
	259
0cd8a05e	260	skb = make_writable(skb, 0, gfp);
064af421 BP	261	if (skb) {
	262	struct iphdr *nh = ip_hdr(skb);
	263	u32 *f = a->type == ODPAT_SET_NW_SRC ? &nh->saddr : &nh->daddr;
	264	u32 old = *f;
	265	u32 new = a->nw_addr;
	266
	267	if (key->nw_proto == IPPROTO_TCP) {
	268	struct tcphdr *th = tcp_hdr(skb);
	269	update_csum(&th->check, skb, old, new, 1);
	270	} else if (key->nw_proto == IPPROTO_UDP) {
	271	struct udphdr *th = udp_hdr(skb);
	272	update_csum(&th->check, skb, old, new, 1);
	273	}
	274	update_csum(&nh->check, skb, old, new, 0);
	275	*f = new;
11cdf5e6 JG	276
	277	if (a->type == ODPAT_SET_NW_SRC)
	278	key->nw_src = a->nw_addr;
	279	else
	280	key->nw_dst = a->nw_addr;
064af421 BP	281	}
	282	return skb;
	283	}
	284
959a2ecd JP	285	static struct sk_buff set_nw_tos(struct sk_buff skb,
	286	struct odp_flow_key *key,
	287	const struct odp_action_nw_tos *a,
	288	gfp_t gfp)
	289	{
	290	if (key->dl_type != htons(ETH_P_IP))
	291	return skb;
	292
	293	skb = make_writable(skb, 0, gfp);
	294	if (skb) {
	295	struct iphdr *nh = ip_hdr(skb);
	296	u8 *f = &nh->tos;
	297	u8 old = *f;
f1193301	298	u8 new;
959a2ecd	299
f1193301	300	/* Set the DSCP bits and preserve the ECN bits. */
3c5f6de3	301	new = a->nw_tos \| (nh->tos & INET_ECN_MASK);
959a2ecd JP	302	update_csum(&nh->check, skb, htons((uint16_t)old),
	303	htons((uint16_t)new), 0);
	304	*f = new;
11cdf5e6	305	key->nw_tos = a->nw_tos;
959a2ecd JP	306	}
	307	return skb;
	308	}
	309
064af421 BP	310	static struct sk_buff *
	311	set_tp_port(struct sk_buff skb, struct odp_flow_key key,
	312	const struct odp_action_tp_port *a,
	313	gfp_t gfp)
	314	{
	315	int check_ofs;
	316
	317	if (key->dl_type != htons(ETH_P_IP))
	318	return skb;
	319
	320	if (key->nw_proto == IPPROTO_TCP)
	321	check_ofs = offsetof(struct tcphdr, check);
	322	else if (key->nw_proto == IPPROTO_UDP)
	323	check_ofs = offsetof(struct udphdr, check);
	324	else
	325	return skb;
	326
0cd8a05e	327	skb = make_writable(skb, 0, gfp);
064af421 BP	328	if (skb) {
	329	struct udphdr *th = udp_hdr(skb);
	330	u16 *f = a->type == ODPAT_SET_TP_SRC ? &th->source : &th->dest;
	331	u16 old = *f;
	332	u16 new = a->tp_port;
985224ac	333	update_csum((u16*)(skb_transport_header(skb) + check_ofs),
00ba5f3b	334	skb, old, new, 0);
064af421	335	*f = new;
11cdf5e6 JG	336	if (a->type == ODPAT_SET_TP_SRC)
	337	key->tp_src = a->tp_port;
	338	else
	339	key->tp_dst = a->tp_port;
064af421 BP	340	}
	341	return skb;
	342	}
	343
	344	static inline unsigned packet_length(const struct sk_buff *skb)
	345	{
	346	unsigned length = skb->len - ETH_HLEN;
	347	if (skb->protocol == htons(ETH_P_8021Q))
	348	length -= VLAN_HLEN;
	349	return length;
	350	}
	351
064af421 BP	352	static void
	353	do_output(struct datapath dp, struct sk_buff skb, int out_port)
	354	{
f2459fe7 JG	355	struct dp_port *p;
f2459fe7 JG	356	int mtu;
064af421 BP	357
	358	if (!skb)
	359	goto error;
	360
f2459fe7	361	p = rcu_dereference(dp->ports[out_port]);
064af421 BP	362	if (!p)
	363	goto error;
	364
f2459fe7 JG	365	mtu = vport_get_mtu(p->vport);
	366	if (packet_length(skb) > mtu && !skb_is_gso(skb)) {
	367	printk(KERN_WARNING "%s: dropped over-mtu packet: %d > %d\n",
	368	dp_name(dp), packet_length(skb), mtu);
	369	goto error;
	370	}
	371
	372	vport_send(p->vport, skb);
064af421 BP	373	return;
	374
	375	error:
	376	kfree_skb(skb);
	377	}
	378
	379	/* Never consumes 'skb'. Returns a port that 'skb' should be sent to, -1 if
	380	* none. */
	381	static int output_group(struct datapath *dp, __u16 group,
	382	struct sk_buff *skb, gfp_t gfp)
	383	{
	384	struct dp_port_group *g = rcu_dereference(dp->groups[group]);
	385	int prev_port = -1;
	386	int i;
	387
	388	if (!g)
	389	return -1;
	390	for (i = 0; i < g->n_ports; i++) {
f2459fe7 JG	391	struct dp_port *p = rcu_dereference(dp->ports[g->ports[i]]);
f2459fe7 JG	392	if (!p \|\| OVS_CB(skb)->dp_port == p)
064af421 BP	393	continue;
	394	if (prev_port != -1) {
	395	struct sk_buff *clone = skb_clone(skb, gfp);
	396	if (!clone)
	397	return -1;
	398	do_output(dp, clone, prev_port);
	399	}
	400	prev_port = p->port_no;
	401	}
	402	return prev_port;
	403	}
	404
	405	static int
	406	output_control(struct datapath dp, struct sk_buff skb, u32 arg, gfp_t gfp)
	407	{
	408	skb = skb_clone(skb, gfp);
	409	if (!skb)
	410	return -ENOMEM;
	411	return dp_output_control(dp, skb, _ODPL_ACTION_NR, arg);
	412	}
	413
72b06300 BP	414	/* Send a copy of this packet up to the sFlow agent, along with extra
	415	* information about what happened to it. */
	416	static void sflow_sample(struct datapath dp, struct sk_buff skb,
56fd8edf	417	const union odp_action *a, int n_actions,
f2459fe7	418	gfp_t gfp, struct dp_port *dp_port)
72b06300 BP	419	{
	420	struct odp_sflow_sample_header *hdr;
	421	unsigned int actlen = n_actions * sizeof(union odp_action);
	422	unsigned int hdrlen = sizeof(struct odp_sflow_sample_header);
	423	struct sk_buff *nskb;
72b06300 BP	424
	425	nskb = skb_copy_expand(skb, actlen + hdrlen, 0, gfp);
	426	if (!nskb)
	427	return;
	428
	429	memcpy(__skb_push(nskb, actlen), a, actlen);
	430	hdr = (struct odp_sflow_sample_header*)__skb_push(nskb, hdrlen);
	431	hdr->n_actions = n_actions;
f2459fe7	432	hdr->sample_pool = atomic_read(&dp_port->sflow_pool);
72b06300 BP	433	dp_output_control(dp, nskb, _ODPL_SFLOW_NR, 0);
	434	}
	435
064af421 BP	436	/* Execute a list of actions against 'skb'. */
	437	int execute_actions(struct datapath dp, struct sk_buff skb,
	438	struct odp_flow_key *key,
	439	const union odp_action *a, int n_actions,
	440	gfp_t gfp)
	441	{
	442	/* Every output action needs a separate clone of 'skb', but the common
	443	* case is just a single output action, so that doing a clone and
	444	* then freeing the original skbuff is wasteful. So the following code
	445	* is slightly obscure just to avoid that. */
	446	int prev_port = -1;
c1c9c9c4	447	u32 priority = skb->priority;
a5225dd6	448	int err;
72b06300 BP	449
72b06300 BP	450	if (dp->sflow_probability) {
f2459fe7	451	struct dp_port *p = OVS_CB(skb)->dp_port;
56fd8edf BP	452	if (p) {
	453	atomic_inc(&p->sflow_pool);
	454	if (dp->sflow_probability == UINT_MAX \|\|
	455	net_random() < dp->sflow_probability)
	456	sflow_sample(dp, skb, a, n_actions, gfp, p);
	457	}
72b06300 BP	458	}
72b06300 BP	459
659586ef JG	460	OVS_CB(skb)->tun_id = 0;
659586ef JG	461
064af421 BP	462	for (; n_actions > 0; a++, n_actions--) {
	463	WARN_ON_ONCE(skb_shared(skb));
	464	if (prev_port != -1) {
	465	do_output(dp, skb_clone(skb, gfp), prev_port);
	466	prev_port = -1;
	467	}
	468
	469	switch (a->type) {
	470	case ODPAT_OUTPUT:
	471	prev_port = a->output.port;
	472	break;
	473
	474	case ODPAT_OUTPUT_GROUP:
	475	prev_port = output_group(dp, a->output_group.group,
	476	skb, gfp);
	477	break;
	478
	479	case ODPAT_CONTROLLER:
	480	err = output_control(dp, skb, a->controller.arg, gfp);
	481	if (err) {
	482	kfree_skb(skb);
	483	return err;
	484	}
	485	break;
	486
659586ef JG	487	case ODPAT_SET_TUNNEL:
	488	set_tunnel(skb, key, a->tunnel.tun_id);
	489	break;
	490
064af421 BP	491	case ODPAT_SET_VLAN_VID:
	492	case ODPAT_SET_VLAN_PCP:
	493	skb = modify_vlan_tci(dp, skb, key, a, n_actions, gfp);
	494	if (IS_ERR(skb))
	495	return PTR_ERR(skb);
	496	break;
	497
	498	case ODPAT_STRIP_VLAN:
	499	skb = strip_vlan(skb, key, gfp);
	500	break;
	501
	502	case ODPAT_SET_DL_SRC:
	503	case ODPAT_SET_DL_DST:
11cdf5e6	504	skb = set_dl_addr(skb, key, &a->dl_addr, gfp);
064af421 BP	505	break;
	506
	507	case ODPAT_SET_NW_SRC:
	508	case ODPAT_SET_NW_DST:
	509	skb = set_nw_addr(skb, key, &a->nw_addr, gfp);
	510	break;
	511
959a2ecd JP	512	case ODPAT_SET_NW_TOS:
	513	skb = set_nw_tos(skb, key, &a->nw_tos, gfp);
	514	break;
	515
064af421 BP	516	case ODPAT_SET_TP_SRC:
	517	case ODPAT_SET_TP_DST:
	518	skb = set_tp_port(skb, key, &a->tp_port, gfp);
	519	break;
c1c9c9c4 BP	520
	521	case ODPAT_SET_PRIORITY:
	522	skb->priority = a->priority.priority;
	523	break;
	524
	525	case ODPAT_POP_PRIORITY:
	526	skb->priority = priority;
	527	break;
064af421 BP	528	}
	529	if (!skb)
	530	return -ENOMEM;
	531	}
	532	if (prev_port != -1)
	533	do_output(dp, skb, prev_port);
	534	else
	535	kfree_skb(skb);
a5225dd6	536	return 0;
064af421	537	}