[mirror_ovs.git] / datapath / flow.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.
 */

#include "flow.h"
#include "datapath.h"
#include <linux/netdevice.h>
#include <linux/etherdevice.h>
#include <linux/if_ether.h>
#include <linux/if_vlan.h>
#include <net/llc_pdu.h>
#include <linux/kernel.h>
#include <linux/jhash.h>
#include <linux/jiffies.h>
#include <linux/llc.h>
#include <linux/module.h>
#include <linux/in.h>
#include <linux/rcupdate.h>
#include <linux/if_arp.h>
#include <linux/if_ether.h>
#include <linux/ip.h>
#include <linux/tcp.h>
#include <linux/udp.h>
#include <linux/icmp.h>
#include <net/inet_ecn.h>
#include <net/ip.h>

#include "compat.h"

struct kmem_cache *flow_cache;
static unsigned int hash_seed;

static inline bool arphdr_ok(struct sk_buff *skb)
{
	return skb->len >= skb_network_offset(skb) + sizeof(struct arp_eth_header);
}

static inline int check_iphdr(struct sk_buff *skb)
{
	unsigned int nh_ofs = skb_network_offset(skb);
	unsigned int ip_len;

	if (skb->len < nh_ofs + sizeof(struct iphdr))
		return -EINVAL;

	ip_len = ip_hdrlen(skb);
	if (ip_len < sizeof(struct iphdr) || skb->len < nh_ofs + ip_len)
		return -EINVAL;

	/*
	 * Pull enough header bytes to account for the IP header plus the
	 * longest transport header that we parse, currently 20 bytes for TCP.
	 */
	if (!pskb_may_pull(skb, min(nh_ofs + ip_len + 20, skb->len)))
		return -ENOMEM;

	skb_set_transport_header(skb, nh_ofs + ip_len);
	return 0;
}

static inline bool tcphdr_ok(struct sk_buff *skb)
{
	int th_ofs = skb_transport_offset(skb);
	if (skb->len >= th_ofs + sizeof(struct tcphdr)) {
		int tcp_len = tcp_hdrlen(skb);
		return (tcp_len >= sizeof(struct tcphdr)
			&& skb->len >= th_ofs + tcp_len);
	}
	return false;
}

static inline bool udphdr_ok(struct sk_buff *skb)
{
	return skb->len >= skb_transport_offset(skb) + sizeof(struct udphdr);
}

static inline bool icmphdr_ok(struct sk_buff *skb)
{
	return skb->len >= skb_transport_offset(skb) + sizeof(struct icmphdr);
}

#define TCP_FLAGS_OFFSET 13
#define TCP_FLAG_MASK 0x3f

void flow_used(struct sw_flow *flow, struct sk_buff *skb)
{
	u8 tcp_flags = 0;

	if (flow->key.dl_type == htons(ETH_P_IP) &&
	    flow->key.nw_proto == IPPROTO_TCP) {
		u8 *tcp = (u8 *)tcp_hdr(skb);
		tcp_flags = *(tcp + TCP_FLAGS_OFFSET) & TCP_FLAG_MASK;
	}

	spin_lock_bh(&flow->lock);
	flow->used = jiffies;
	flow->packet_count++;
	flow->byte_count += skb->len;
	flow->tcp_flags |= tcp_flags;
	spin_unlock_bh(&flow->lock);
}

struct sw_flow_actions *flow_actions_alloc(size_t n_actions)
{
	struct sw_flow_actions *sfa;

	/* At least DP_MAX_PORTS actions are required to be able to flood a
	 * packet to every port.  Factor of 2 allows for setting VLAN tags,
	 * etc. */
	if (n_actions > 2 * DP_MAX_PORTS)
		return ERR_PTR(-EINVAL);

	sfa = kmalloc(sizeof *sfa + n_actions * sizeof(union odp_action),
		      GFP_KERNEL);
	if (!sfa)
		return ERR_PTR(-ENOMEM);

	sfa->n_actions = n_actions;
	return sfa;
}

struct sw_flow *flow_alloc(void)
{
	struct sw_flow *flow;

	flow = kmem_cache_alloc(flow_cache, GFP_KERNEL);
	if (!flow)
		return ERR_PTR(-ENOMEM);

	spin_lock_init(&flow->lock);
	atomic_set(&flow->refcnt, 1);
	flow->dead = false;

	return flow;
}

void flow_free_tbl(struct tbl_node *node)
{
	struct sw_flow *flow = flow_cast(node);

	flow->dead = true;
	flow_put(flow);
}

/* RCU callback used by flow_deferred_free. */
static void rcu_free_flow_callback(struct rcu_head *rcu)
{
	struct sw_flow *flow = container_of(rcu, struct sw_flow, rcu);

	flow->dead = true;
	flow_put(flow);
}

/* Schedules 'flow' to be freed after the next RCU grace period.
 * The caller must hold rcu_read_lock for this to be sensible. */
void flow_deferred_free(struct sw_flow *flow)
{
	call_rcu(&flow->rcu, rcu_free_flow_callback);
}

void flow_hold(struct sw_flow *flow)
{
	atomic_inc(&flow->refcnt);
}

void flow_put(struct sw_flow *flow)
{
	if (unlikely(!flow))
		return;

	if (atomic_dec_and_test(&flow->refcnt)) {
		kfree(flow->sf_acts);
		kmem_cache_free(flow_cache, flow);
	}
}

/* RCU callback used by flow_deferred_free_acts. */
static void rcu_free_acts_callback(struct rcu_head *rcu)
{
	struct sw_flow_actions *sf_acts = container_of(rcu,
			struct sw_flow_actions, rcu);
	kfree(sf_acts);
}

/* Schedules 'sf_acts' to be freed after the next RCU grace period.
 * The caller must hold rcu_read_lock for this to be sensible. */
void flow_deferred_free_acts(struct sw_flow_actions *sf_acts)
{
	call_rcu(&sf_acts->rcu, rcu_free_acts_callback);
}

static void parse_vlan(struct sk_buff *skb, struct odp_flow_key *key)
{
	struct qtag_prefix {
		__be16 eth_type; /* ETH_P_8021Q */
		__be16 tci;
	};
	struct qtag_prefix *qp;

	if (skb->len < sizeof(struct qtag_prefix) + sizeof(__be16))
		return;

	qp = (struct qtag_prefix *) skb->data;
	key->dl_vlan = qp->tci & htons(VLAN_VID_MASK);
	key->dl_vlan_pcp = (ntohs(qp->tci) & VLAN_PCP_MASK) >> VLAN_PCP_SHIFT;
	__skb_pull(skb, sizeof(struct qtag_prefix));
}

static __be16 parse_ethertype(struct sk_buff *skb)
{
	struct llc_snap_hdr {
		u8  dsap;  /* Always 0xAA */
		u8  ssap;  /* Always 0xAA */
		u8  ctrl;
		u8  oui[3];
		u16 ethertype;
	};
	struct llc_snap_hdr *llc;
	__be16 proto;

	proto = *(__be16 *) skb->data;
	__skb_pull(skb, sizeof(__be16));

	if (ntohs(proto) >= ODP_DL_TYPE_ETH2_CUTOFF)
		return proto;

	if (unlikely(skb->len < sizeof(struct llc_snap_hdr)))
		return htons(ODP_DL_TYPE_NOT_ETH_TYPE);

	llc = (struct llc_snap_hdr *) skb->data;
	if (llc->dsap != LLC_SAP_SNAP ||
	    llc->ssap != LLC_SAP_SNAP ||
	    (llc->oui[0] | llc->oui[1] | llc->oui[2]) != 0)
		return htons(ODP_DL_TYPE_NOT_ETH_TYPE);

	__skb_pull(skb, sizeof(struct llc_snap_hdr));
	return llc->ethertype;
}

/**
 * flow_extract - extracts a flow key from an Ethernet frame.
 * @skb: sk_buff that contains the frame, with skb->data pointing to the
 * Ethernet header
 * @in_port: port number on which @skb was received.
 * @key: output flow key
 *
 * The caller must ensure that skb->len >= ETH_HLEN.
 *
 * Returns 0 if successful, otherwise a negative errno value.
 *
 * Initializes @skb header pointers as follows:
 *
 *    - skb->mac_header: the Ethernet header.
 *
 *    - skb->network_header: just past the Ethernet header, or just past the
 *      VLAN header, to the first byte of the Ethernet payload.
 *
 *    - skb->transport_header: If key->dl_type is ETH_P_IP on output, then just
 *      past the IPv4 header, if one is present and of a correct length,
 *      otherwise the same as skb->network_header.  For other key->dl_type
 *      values it is left untouched.
 *
 * Sets OVS_CB(skb)->is_frag to %true if @skb is an IPv4 fragment, otherwise to
 * %false.
 */
int flow_extract(struct sk_buff *skb, u16 in_port, struct odp_flow_key *key,
		 bool *is_frag)
{
	struct ethhdr *eth;

	memset(key, 0, sizeof *key);
	key->tun_id = OVS_CB(skb)->tun_id;
	key->in_port = in_port;
	key->dl_vlan = htons(ODP_VLAN_NONE);
	*is_frag = false;

	/*
	 * We would really like to pull as many bytes as we could possibly
	 * want to parse into the linear data area.  Currently that is:
	 *
	 *    14     Ethernet header
	 *     4     VLAN header
	 *    60     max IP header with options
	 *    20     max TCP/UDP/ICMP header (don't care about options)
	 *    --
	 *    98
	 *
	 * But Xen only allocates 64 or 72 bytes for the linear data area in
	 * netback, which means that we would reallocate and copy the skb's
	 * linear data on every packet if we did that.  So instead just pull 64
	 * bytes, which is always sufficient without IP options, and then check
	 * whether we need to pull more later when we look at the IP header.
	 */
	if (!pskb_may_pull(skb, min(skb->len, 64u)))
		return -ENOMEM;

	skb_reset_mac_header(skb);

	/* Link layer. */
	eth = eth_hdr(skb);
	memcpy(key->dl_src, eth->h_source, ETH_ALEN);
	memcpy(key->dl_dst, eth->h_dest, ETH_ALEN);

	/* dl_type, dl_vlan, dl_vlan_pcp. */
	__skb_pull(skb, 2 * ETH_ALEN);
	if (eth->h_proto == htons(ETH_P_8021Q))
		parse_vlan(skb, key);
	key->dl_type = parse_ethertype(skb);
	skb_reset_network_header(skb);
	__skb_push(skb, skb->data - (unsigned char *)eth);

	/* Network layer. */
	if (key->dl_type == htons(ETH_P_IP)) {
		struct iphdr *nh;
		int error;

		error = check_iphdr(skb);
		if (unlikely(error)) {
			if (error == -EINVAL) {
				skb->transport_header = skb->network_header;
				return 0;
			}
			return error;
		}

		nh = ip_hdr(skb);
		key->nw_src = nh->saddr;
		key->nw_dst = nh->daddr;
		key->nw_tos = nh->tos & ~INET_ECN_MASK;
		key->nw_proto = nh->protocol;

		/* Transport layer. */
		if (!(nh->frag_off & htons(IP_MF | IP_OFFSET))) {
			if (key->nw_proto == IPPROTO_TCP) {
				if (tcphdr_ok(skb)) {
					struct tcphdr *tcp = tcp_hdr(skb);
					key->tp_src = tcp->source;
					key->tp_dst = tcp->dest;
				}
			} else if (key->nw_proto == IPPROTO_UDP) {
				if (udphdr_ok(skb)) {
					struct udphdr *udp = udp_hdr(skb);
					key->tp_src = udp->source;
					key->tp_dst = udp->dest;
				}
			} else if (key->nw_proto == IPPROTO_ICMP) {
				if (icmphdr_ok(skb)) {
					struct icmphdr *icmp = icmp_hdr(skb);
					/* The ICMP type and code fields use the 16-bit
					 * transport port fields, so we need to store them
					 * in 16-bit network byte order. */
					key->tp_src = htons(icmp->type);
					key->tp_dst = htons(icmp->code);
				}
			}
		} else
			*is_frag = true;

	} else if (key->dl_type == htons(ETH_P_ARP) && arphdr_ok(skb)) {
		struct arp_eth_header *arp;

		arp = (struct arp_eth_header *)skb_network_header(skb);

		if (arp->ar_hrd == htons(ARPHRD_ETHER)
				&& arp->ar_pro == htons(ETH_P_IP)
				&& arp->ar_hln == ETH_ALEN
				&& arp->ar_pln == 4) {

			/* We only match on the lower 8 bits of the opcode. */
			if (ntohs(arp->ar_op) <= 0xff)
				key->nw_proto = ntohs(arp->ar_op);

			if (key->nw_proto == ARPOP_REQUEST
					|| key->nw_proto == ARPOP_REPLY) {
				memcpy(&key->nw_src, arp->ar_sip, sizeof(key->nw_src));
				memcpy(&key->nw_dst, arp->ar_tip, sizeof(key->nw_dst));
			}
		}
	}
	return 0;
}

u32 flow_hash(const struct odp_flow_key *key)
{
	return jhash2((u32*)key, sizeof *key / sizeof(u32), hash_seed);
}

int flow_cmp(const struct tbl_node *node, void *key2_)
{
	const struct odp_flow_key *key1 = &flow_cast(node)->key;
	const struct odp_flow_key *key2 = key2_;

	return !memcmp(key1, key2, sizeof(struct odp_flow_key));
}

/* Initializes the flow module.
 * Returns zero if successful or a negative error code. */
int flow_init(void)
{
	flow_cache = kmem_cache_create("sw_flow", sizeof(struct sw_flow), 0,
					0, NULL);
	if (flow_cache == NULL)
		return -ENOMEM;

	get_random_bytes(&hash_seed, sizeof hash_seed);

	return 0;
}

/* Uninitializes the flow module. */
void flow_exit(void)
{
	kmem_cache_destroy(flow_cache);
}
Commit	Line	Data
064af421 BP	1	/*
064af421 BP	2	* Distributed under the terms of the GNU GPL version 2.
834377ea	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	#include "flow.h"
f5e86186	10	#include "datapath.h"
064af421 BP	11	#include <linux/netdevice.h>
	12	#include <linux/etherdevice.h>
	13	#include <linux/if_ether.h>
	14	#include <linux/if_vlan.h>
	15	#include <net/llc_pdu.h>
	16	#include <linux/kernel.h>
8d5ebd83	17	#include <linux/jhash.h>
064af421 BP	18	#include <linux/jiffies.h>
	19	#include <linux/llc.h>
	20	#include <linux/module.h>
	21	#include <linux/in.h>
	22	#include <linux/rcupdate.h>
a26ef517	23	#include <linux/if_arp.h>
064af421 BP	24	#include <linux/if_ether.h>
	25	#include <linux/ip.h>
	26	#include <linux/tcp.h>
	27	#include <linux/udp.h>
	28	#include <linux/icmp.h>
3c5f6de3	29	#include <net/inet_ecn.h>
064af421 BP	30	#include <net/ip.h>
	31
	32	#include "compat.h"
	33
	34	struct kmem_cache *flow_cache;
8d5ebd83	35	static unsigned int hash_seed;
064af421	36
e819fb47	37	static inline bool arphdr_ok(struct sk_buff *skb)
a26ef517	38	{
7d0ab001	39	return skb->len >= skb_network_offset(skb) + sizeof(struct arp_eth_header);
a26ef517 JP	40	}
a26ef517 JP	41
4c1ad233	42	static inline int check_iphdr(struct sk_buff *skb)
064af421	43	{
4c1ad233 BP	44	unsigned int nh_ofs = skb_network_offset(skb);
	45	unsigned int ip_len;
	46
	47	if (skb->len < nh_ofs + sizeof(struct iphdr))
	48	return -EINVAL;
	49
	50	ip_len = ip_hdrlen(skb);
	51	if (ip_len < sizeof(struct iphdr) \|\| skb->len < nh_ofs + ip_len)
	52	return -EINVAL;
	53
	54	/*
	55	* Pull enough header bytes to account for the IP header plus the
	56	* longest transport header that we parse, currently 20 bytes for TCP.
	57	*/
	58	if (!pskb_may_pull(skb, min(nh_ofs + ip_len + 20, skb->len)))
	59	return -ENOMEM;
	60
	61	skb_set_transport_header(skb, nh_ofs + ip_len);
	62	return 0;
064af421 BP	63	}
064af421 BP	64
e819fb47	65	static inline bool tcphdr_ok(struct sk_buff *skb)
064af421 BP	66	{
064af421 BP	67	int th_ofs = skb_transport_offset(skb);
7d0ab001	68	if (skb->len >= th_ofs + sizeof(struct tcphdr)) {
064af421 BP	69	int tcp_len = tcp_hdrlen(skb);
	70	return (tcp_len >= sizeof(struct tcphdr)
	71	&& skb->len >= th_ofs + tcp_len);
	72	}
e819fb47	73	return false;
064af421 BP	74	}
064af421 BP	75
e819fb47	76	static inline bool udphdr_ok(struct sk_buff *skb)
064af421	77	{
7d0ab001	78	return skb->len >= skb_transport_offset(skb) + sizeof(struct udphdr);
064af421 BP	79	}
064af421 BP	80
e819fb47	81	static inline bool icmphdr_ok(struct sk_buff *skb)
064af421	82	{
7d0ab001	83	return skb->len >= skb_transport_offset(skb) + sizeof(struct icmphdr);
064af421 BP	84	}
	85
	86	#define TCP_FLAGS_OFFSET 13
	87	#define TCP_FLAG_MASK 0x3f
	88
064af421 BP	89	void flow_used(struct sw_flow flow, struct sk_buff skb)
064af421 BP	90	{
064af421 BP	91	u8 tcp_flags = 0;
064af421 BP	92
abfec865 BP	93	if (flow->key.dl_type == htons(ETH_P_IP) &&
	94	flow->key.nw_proto == IPPROTO_TCP) {
	95	u8 tcp = (u8 )tcp_hdr(skb);
	96	tcp_flags = *(tcp + TCP_FLAGS_OFFSET) & TCP_FLAG_MASK;
064af421 BP	97	}
064af421 BP	98
f2459fe7	99	spin_lock_bh(&flow->lock);
6bfafa55	100	flow->used = jiffies;
064af421 BP	101	flow->packet_count++;
	102	flow->byte_count += skb->len;
	103	flow->tcp_flags \|= tcp_flags;
f2459fe7	104	spin_unlock_bh(&flow->lock);
064af421 BP	105	}
	106
	107	struct sw_flow_actions *flow_actions_alloc(size_t n_actions)
	108	{
	109	struct sw_flow_actions *sfa;
	110
722d19c5 BP	111	/* At least DP_MAX_PORTS actions are required to be able to flood a
	112	* packet to every port. Factor of 2 allows for setting VLAN tags,
	113	* etc. */
	114	if (n_actions > 2 * DP_MAX_PORTS)
064af421 BP	115	return ERR_PTR(-EINVAL);
	116
	117	sfa = kmalloc(sizeof sfa + n_actions sizeof(union odp_action),
	118	GFP_KERNEL);
	119	if (!sfa)
	120	return ERR_PTR(-ENOMEM);
	121
	122	sfa->n_actions = n_actions;
	123	return sfa;
	124	}
	125
560e8022 JG	126	struct sw_flow *flow_alloc(void)
	127	{
	128	struct sw_flow *flow;
	129
	130	flow = kmem_cache_alloc(flow_cache, GFP_KERNEL);
	131	if (!flow)
	132	return ERR_PTR(-ENOMEM);
	133
	134	spin_lock_init(&flow->lock);
fb8c9347 JG	135	atomic_set(&flow->refcnt, 1);
fb8c9347 JG	136	flow->dead = false;
064af421	137
560e8022 JG	138	return flow;
	139	}
	140
8d5ebd83 JG	141	void flow_free_tbl(struct tbl_node *node)
	142	{
	143	struct sw_flow *flow = flow_cast(node);
fb8c9347 JG	144
	145	flow->dead = true;
	146	flow_put(flow);
8d5ebd83 JG	147	}
8d5ebd83 JG	148
064af421 BP	149	/* RCU callback used by flow_deferred_free. */
	150	static void rcu_free_flow_callback(struct rcu_head *rcu)
	151	{
	152	struct sw_flow *flow = container_of(rcu, struct sw_flow, rcu);
fb8c9347 JG	153
	154	flow->dead = true;
	155	flow_put(flow);
064af421 BP	156	}
	157
	158	/* Schedules 'flow' to be freed after the next RCU grace period.
	159	* The caller must hold rcu_read_lock for this to be sensible. */
	160	void flow_deferred_free(struct sw_flow *flow)
	161	{
	162	call_rcu(&flow->rcu, rcu_free_flow_callback);
	163	}
	164
fb8c9347 JG	165	void flow_hold(struct sw_flow *flow)
	166	{
	167	atomic_inc(&flow->refcnt);
	168	}
	169
	170	void flow_put(struct sw_flow *flow)
	171	{
	172	if (unlikely(!flow))
	173	return;
	174
	175	if (atomic_dec_and_test(&flow->refcnt)) {
	176	kfree(flow->sf_acts);
	177	kmem_cache_free(flow_cache, flow);
	178	}
	179	}
	180
064af421 BP	181	/* RCU callback used by flow_deferred_free_acts. */
	182	static void rcu_free_acts_callback(struct rcu_head *rcu)
	183	{
d295e8e9	184	struct sw_flow_actions *sf_acts = container_of(rcu,
064af421 BP	185	struct sw_flow_actions, rcu);
	186	kfree(sf_acts);
	187	}
	188
	189	/* Schedules 'sf_acts' to be freed after the next RCU grace period.
	190	* The caller must hold rcu_read_lock for this to be sensible. */
	191	void flow_deferred_free_acts(struct sw_flow_actions *sf_acts)
	192	{
	193	call_rcu(&sf_acts->rcu, rcu_free_acts_callback);
	194	}
	195
50f06e16	196	static void parse_vlan(struct sk_buff skb, struct odp_flow_key key)
064af421	197	{
50f06e16 BP	198	struct qtag_prefix {
	199	__be16 eth_type; /* ETH_P_8021Q */
	200	__be16 tci;
	201	};
	202	struct qtag_prefix *qp;
	203
	204	if (skb->len < sizeof(struct qtag_prefix) + sizeof(__be16))
	205	return;
	206
	207	qp = (struct qtag_prefix *) skb->data;
	208	key->dl_vlan = qp->tci & htons(VLAN_VID_MASK);
	209	key->dl_vlan_pcp = (ntohs(qp->tci) & VLAN_PCP_MASK) >> VLAN_PCP_SHIFT;
	210	__skb_pull(skb, sizeof(struct qtag_prefix));
	211	}
	212
	213	static __be16 parse_ethertype(struct sk_buff *skb)
064af421	214	{
50f06e16 BP	215	struct llc_snap_hdr {
	216	u8 dsap; /* Always 0xAA */
	217	u8 ssap; /* Always 0xAA */
	218	u8 ctrl;
	219	u8 oui[3];
	220	u16 ethertype;
	221	};
	222	struct llc_snap_hdr *llc;
	223	__be16 proto;
	224
	225	proto = (__be16 ) skb->data;
	226	__skb_pull(skb, sizeof(__be16));
	227
	228	if (ntohs(proto) >= ODP_DL_TYPE_ETH2_CUTOFF)
	229	return proto;
	230
	231	if (unlikely(skb->len < sizeof(struct llc_snap_hdr)))
	232	return htons(ODP_DL_TYPE_NOT_ETH_TYPE);
	233
	234	llc = (struct llc_snap_hdr *) skb->data;
	235	if (llc->dsap != LLC_SAP_SNAP \|\|
	236	llc->ssap != LLC_SAP_SNAP \|\|
	237	(llc->oui[0] \| llc->oui[1] \| llc->oui[2]) != 0)
	238	return htons(ODP_DL_TYPE_NOT_ETH_TYPE);
	239
	240	__skb_pull(skb, sizeof(struct llc_snap_hdr));
	241	return llc->ethertype;
064af421 BP	242	}
064af421 BP	243
a31e0e31 BP	244	/**
	245	* flow_extract - extracts a flow key from an Ethernet frame.
	246	* @skb: sk_buff that contains the frame, with skb->data pointing to the
	247	* Ethernet header
	248	* @in_port: port number on which @skb was received.
	249	* @key: output flow key
	250	*
	251	* The caller must ensure that skb->len >= ETH_HLEN.
	252	*
4c1ad233 BP	253	* Returns 0 if successful, otherwise a negative errno value.
4c1ad233 BP	254	*
59a18f80 BP	255	* Initializes @skb header pointers as follows:
	256	*
	257	* - skb->mac_header: the Ethernet header.
	258	*
	259	* - skb->network_header: just past the Ethernet header, or just past the
	260	* VLAN header, to the first byte of the Ethernet payload.
	261	*
	262	* - skb->transport_header: If key->dl_type is ETH_P_IP on output, then just
	263	* past the IPv4 header, if one is present and of a correct length,
	264	* otherwise the same as skb->network_header. For other key->dl_type
	265	* values it is left untouched.
	266	*
769f8ccd BP	267	* Sets OVS_CB(skb)->is_frag to %true if @skb is an IPv4 fragment, otherwise to
769f8ccd BP	268	* %false.
a31e0e31	269	*/
b7a31ec1 JG	270	int flow_extract(struct sk_buff skb, u16 in_port, struct odp_flow_key key,
b7a31ec1 JG	271	bool *is_frag)
064af421 BP	272	{
064af421 BP	273	struct ethhdr *eth;
064af421 BP	274
064af421 BP	275	memset(key, 0, sizeof *key);
659586ef	276	key->tun_id = OVS_CB(skb)->tun_id;
064af421	277	key->in_port = in_port;
659586ef	278	key->dl_vlan = htons(ODP_VLAN_NONE);
b7a31ec1	279	*is_frag = false;
064af421	280
4c1ad233 BP	281	/*
	282	* We would really like to pull as many bytes as we could possibly
	283	* want to parse into the linear data area. Currently that is:
	284	*
	285	* 14 Ethernet header
	286	* 4 VLAN header
	287	* 60 max IP header with options
	288	* 20 max TCP/UDP/ICMP header (don't care about options)
	289	* --
	290	* 98
	291	*
	292	* But Xen only allocates 64 or 72 bytes for the linear data area in
	293	* netback, which means that we would reallocate and copy the skb's
	294	* linear data on every packet if we did that. So instead just pull 64
	295	* bytes, which is always sufficient without IP options, and then check
	296	* whether we need to pull more later when we look at the IP header.
	297	*/
d9fce1ca	298	if (!pskb_may_pull(skb, min(skb->len, 64u)))
4c1ad233	299	return -ENOMEM;
064af421 BP	300
064af421 BP	301	skb_reset_mac_header(skb);
064af421	302
50f06e16 BP	303	/* Link layer. */
50f06e16 BP	304	eth = eth_hdr(skb);
064af421 BP	305	memcpy(key->dl_src, eth->h_source, ETH_ALEN);
064af421 BP	306	memcpy(key->dl_dst, eth->h_dest, ETH_ALEN);
50f06e16 BP	307
	308	/* dl_type, dl_vlan, dl_vlan_pcp. */
	309	__skb_pull(skb, 2 * ETH_ALEN);
	310	if (eth->h_proto == htons(ETH_P_8021Q))
	311	parse_vlan(skb, key);
	312	key->dl_type = parse_ethertype(skb);
	313	skb_reset_network_header(skb);
	314	__skb_push(skb, skb->data - (unsigned char *)eth);
064af421 BP	315
064af421 BP	316	/* Network layer. */
4c1ad233 BP	317	if (key->dl_type == htons(ETH_P_IP)) {
	318	struct iphdr *nh;
	319	int error;
	320
	321	error = check_iphdr(skb);
	322	if (unlikely(error)) {
	323	if (error == -EINVAL) {
	324	skb->transport_header = skb->network_header;
	325	return 0;
	326	}
	327	return error;
	328	}
	329
	330	nh = ip_hdr(skb);
064af421 BP	331	key->nw_src = nh->saddr;
064af421 BP	332	key->nw_dst = nh->daddr;
f5e86186	333	key->nw_tos = nh->tos & ~INET_ECN_MASK;
064af421	334	key->nw_proto = nh->protocol;
064af421 BP	335
	336	/* Transport layer. */
	337	if (!(nh->frag_off & htons(IP_MF \| IP_OFFSET))) {
	338	if (key->nw_proto == IPPROTO_TCP) {
	339	if (tcphdr_ok(skb)) {
	340	struct tcphdr *tcp = tcp_hdr(skb);
	341	key->tp_src = tcp->source;
	342	key->tp_dst = tcp->dest;
064af421 BP	343	}
	344	} else if (key->nw_proto == IPPROTO_UDP) {
	345	if (udphdr_ok(skb)) {
	346	struct udphdr *udp = udp_hdr(skb);
	347	key->tp_src = udp->source;
	348	key->tp_dst = udp->dest;
064af421 BP	349	}
	350	} else if (key->nw_proto == IPPROTO_ICMP) {
	351	if (icmphdr_ok(skb)) {
	352	struct icmphdr *icmp = icmp_hdr(skb);
	353	/* The ICMP type and code fields use the 16-bit
	354	* transport port fields, so we need to store them
	355	* in 16-bit network byte order. */
	356	key->tp_src = htons(icmp->type);
	357	key->tp_dst = htons(icmp->code);
064af421 BP	358	}
064af421 BP	359	}
b7a31ec1 JG	360	} else
	361	*is_frag = true;
	362
a26ef517 JP	363	} else if (key->dl_type == htons(ETH_P_ARP) && arphdr_ok(skb)) {
	364	struct arp_eth_header *arp;
	365
	366	arp = (struct arp_eth_header *)skb_network_header(skb);
	367
f5e86186	368	if (arp->ar_hrd == htons(ARPHRD_ETHER)
de3f65ea JP	369	&& arp->ar_pro == htons(ETH_P_IP)
	370	&& arp->ar_hln == ETH_ALEN
	371	&& arp->ar_pln == 4) {
	372
	373	/* We only match on the lower 8 bits of the opcode. */
b7a31ec1	374	if (ntohs(arp->ar_op) <= 0xff)
de3f65ea	375	key->nw_proto = ntohs(arp->ar_op);
de3f65ea	376
d295e8e9	377	if (key->nw_proto == ARPOP_REQUEST
de3f65ea JP	378	\|\| key->nw_proto == ARPOP_REPLY) {
	379	memcpy(&key->nw_src, arp->ar_sip, sizeof(key->nw_src));
	380	memcpy(&key->nw_dst, arp->ar_tip, sizeof(key->nw_dst));
	381	}
	382	}
064af421	383	}
769f8ccd	384	return 0;
064af421 BP	385	}
064af421 BP	386
8d5ebd83 JG	387	u32 flow_hash(const struct odp_flow_key *key)
	388	{
	389	return jhash2((u32)key, sizeof key / sizeof(u32), hash_seed);
	390	}
	391
	392	int flow_cmp(const struct tbl_node node, void key2_)
	393	{
	394	const struct odp_flow_key *key1 = &flow_cast(node)->key;
	395	const struct odp_flow_key *key2 = key2_;
	396
	397	return !memcmp(key1, key2, sizeof(struct odp_flow_key));
	398	}
	399
064af421 BP	400	/* Initializes the flow module.
	401	* Returns zero if successful or a negative error code. */
	402	int flow_init(void)
	403	{
	404	flow_cache = kmem_cache_create("sw_flow", sizeof(struct sw_flow), 0,
	405	0, NULL);
	406	if (flow_cache == NULL)
	407	return -ENOMEM;
	408
8d5ebd83 JG	409	get_random_bytes(&hash_seed, sizeof hash_seed);
8d5ebd83 JG	410
064af421 BP	411	return 0;
	412	}
	413
	414	/* Uninitializes the flow module. */
	415	void flow_exit(void)
	416	{
	417	kmem_cache_destroy(flow_cache);
	418	}