2 * Distributed under the terms of the GNU GPL version 2.
3 * Copyright (c) 2007, 2008, 2009, 2010 Nicira Networks.
5 * Significant portions of this file may be copied from parts of the Linux
6 * kernel, by Linus Torvalds and others.
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>
17 #include <linux/jhash.h>
18 #include <linux/jiffies.h>
19 #include <linux/llc.h>
20 #include <linux/module.h>
22 #include <linux/rcupdate.h>
23 #include <linux/if_arp.h>
24 #include <linux/if_ether.h>
26 #include <linux/tcp.h>
27 #include <linux/udp.h>
28 #include <linux/icmp.h>
29 #include <net/inet_ecn.h>
34 struct kmem_cache *flow_cache;
35 static unsigned int hash_seed;
39 __be16 ar_hrd; /* format of hardware address */
40 __be16 ar_pro; /* format of protocol address */
41 unsigned char ar_hln; /* length of hardware address */
42 unsigned char ar_pln; /* length of protocol address */
43 __be16 ar_op; /* ARP opcode (command) */
45 /* Ethernet+IPv4 specific members. */
46 unsigned char ar_sha[ETH_ALEN]; /* sender hardware address */
47 unsigned char ar_sip[4]; /* sender IP address */
48 unsigned char ar_tha[ETH_ALEN]; /* target hardware address */
49 unsigned char ar_tip[4]; /* target IP address */
50 } __attribute__((packed));
52 static inline bool arphdr_ok(struct sk_buff *skb)
54 int nh_ofs = skb_network_offset(skb);
55 return pskb_may_pull(skb, nh_ofs + sizeof(struct arp_eth_header));
58 static inline bool iphdr_ok(struct sk_buff *skb)
60 int nh_ofs = skb_network_offset(skb);
61 if (skb->len >= nh_ofs + sizeof(struct iphdr)) {
62 int ip_len = ip_hdrlen(skb);
63 return (ip_len >= sizeof(struct iphdr)
64 && pskb_may_pull(skb, nh_ofs + ip_len));
69 static inline bool tcphdr_ok(struct sk_buff *skb)
71 int th_ofs = skb_transport_offset(skb);
72 if (pskb_may_pull(skb, th_ofs + sizeof(struct tcphdr))) {
73 int tcp_len = tcp_hdrlen(skb);
74 return (tcp_len >= sizeof(struct tcphdr)
75 && skb->len >= th_ofs + tcp_len);
80 static inline bool udphdr_ok(struct sk_buff *skb)
82 int th_ofs = skb_transport_offset(skb);
83 return pskb_may_pull(skb, th_ofs + sizeof(struct udphdr));
86 static inline bool icmphdr_ok(struct sk_buff *skb)
88 int th_ofs = skb_transport_offset(skb);
89 return pskb_may_pull(skb, th_ofs + sizeof(struct icmphdr));
92 #define TCP_FLAGS_OFFSET 13
93 #define TCP_FLAG_MASK 0x3f
95 void flow_used(struct sw_flow *flow, struct sk_buff *skb)
99 if (flow->key.dl_type == htons(ETH_P_IP) &&
100 flow->key.nw_proto == IPPROTO_TCP) {
101 u8 *tcp = (u8 *)tcp_hdr(skb);
102 tcp_flags = *(tcp + TCP_FLAGS_OFFSET) & TCP_FLAG_MASK;
105 spin_lock_bh(&flow->lock);
106 flow->used = jiffies;
107 flow->packet_count++;
108 flow->byte_count += skb->len;
109 flow->tcp_flags |= tcp_flags;
110 spin_unlock_bh(&flow->lock);
113 struct sw_flow_actions *flow_actions_alloc(size_t n_actions)
115 struct sw_flow_actions *sfa;
117 if (n_actions > (PAGE_SIZE - sizeof *sfa) / sizeof(union odp_action))
118 return ERR_PTR(-EINVAL);
120 sfa = kmalloc(sizeof *sfa + n_actions * sizeof(union odp_action),
123 return ERR_PTR(-ENOMEM);
125 sfa->n_actions = n_actions;
130 /* Frees 'flow' immediately. */
131 static void flow_free(struct sw_flow *flow)
135 kfree(flow->sf_acts);
136 kmem_cache_free(flow_cache, flow);
139 void flow_free_tbl(struct tbl_node *node)
141 struct sw_flow *flow = flow_cast(node);
145 /* RCU callback used by flow_deferred_free. */
146 static void rcu_free_flow_callback(struct rcu_head *rcu)
148 struct sw_flow *flow = container_of(rcu, struct sw_flow, rcu);
152 /* Schedules 'flow' to be freed after the next RCU grace period.
153 * The caller must hold rcu_read_lock for this to be sensible. */
154 void flow_deferred_free(struct sw_flow *flow)
156 call_rcu(&flow->rcu, rcu_free_flow_callback);
159 /* RCU callback used by flow_deferred_free_acts. */
160 static void rcu_free_acts_callback(struct rcu_head *rcu)
162 struct sw_flow_actions *sf_acts = container_of(rcu,
163 struct sw_flow_actions, rcu);
167 /* Schedules 'sf_acts' to be freed after the next RCU grace period.
168 * The caller must hold rcu_read_lock for this to be sensible. */
169 void flow_deferred_free_acts(struct sw_flow_actions *sf_acts)
171 call_rcu(&sf_acts->rcu, rcu_free_acts_callback);
174 static void parse_vlan(struct sk_buff *skb, struct odp_flow_key *key)
177 __be16 eth_type; /* ETH_P_8021Q */
180 struct qtag_prefix *qp;
182 if (skb->len < sizeof(struct qtag_prefix) + sizeof(__be16))
185 qp = (struct qtag_prefix *) skb->data;
186 key->dl_vlan = qp->tci & htons(VLAN_VID_MASK);
187 key->dl_vlan_pcp = (ntohs(qp->tci) & VLAN_PCP_MASK) >> VLAN_PCP_SHIFT;
188 __skb_pull(skb, sizeof(struct qtag_prefix));
191 static __be16 parse_ethertype(struct sk_buff *skb)
193 struct llc_snap_hdr {
194 u8 dsap; /* Always 0xAA */
195 u8 ssap; /* Always 0xAA */
200 struct llc_snap_hdr *llc;
203 proto = *(__be16 *) skb->data;
204 __skb_pull(skb, sizeof(__be16));
206 if (ntohs(proto) >= ODP_DL_TYPE_ETH2_CUTOFF)
209 if (unlikely(skb->len < sizeof(struct llc_snap_hdr)))
210 return htons(ODP_DL_TYPE_NOT_ETH_TYPE);
212 llc = (struct llc_snap_hdr *) skb->data;
213 if (llc->dsap != LLC_SAP_SNAP ||
214 llc->ssap != LLC_SAP_SNAP ||
215 (llc->oui[0] | llc->oui[1] | llc->oui[2]) != 0)
216 return htons(ODP_DL_TYPE_NOT_ETH_TYPE);
218 __skb_pull(skb, sizeof(struct llc_snap_hdr));
219 return llc->ethertype;
223 * flow_extract - extracts a flow key from an Ethernet frame.
224 * @skb: sk_buff that contains the frame, with skb->data pointing to the
226 * @in_port: port number on which @skb was received.
227 * @key: output flow key
229 * The caller must ensure that skb->len >= ETH_HLEN.
231 * Sets OVS_CB(skb)->is_frag to %true if @skb is an IPv4 fragment, otherwise to
234 int flow_extract(struct sk_buff *skb, u16 in_port, struct odp_flow_key *key)
238 memset(key, 0, sizeof *key);
239 key->tun_id = OVS_CB(skb)->tun_id;
240 key->in_port = in_port;
241 key->dl_vlan = htons(ODP_VLAN_NONE);
242 OVS_CB(skb)->is_frag = false;
244 if (!pskb_may_pull(skb, min(skb->len, 64u)))
247 skb_reset_mac_header(skb);
251 memcpy(key->dl_src, eth->h_source, ETH_ALEN);
252 memcpy(key->dl_dst, eth->h_dest, ETH_ALEN);
254 /* dl_type, dl_vlan, dl_vlan_pcp. */
255 __skb_pull(skb, 2 * ETH_ALEN);
256 if (eth->h_proto == htons(ETH_P_8021Q))
257 parse_vlan(skb, key);
258 key->dl_type = parse_ethertype(skb);
259 skb_reset_network_header(skb);
260 __skb_push(skb, skb->data - (unsigned char *)eth);
263 if (key->dl_type == htons(ETH_P_IP) && iphdr_ok(skb)) {
264 struct iphdr *nh = ip_hdr(skb);
265 int th_ofs = skb_network_offset(skb) + nh->ihl * 4;
266 key->nw_src = nh->saddr;
267 key->nw_dst = nh->daddr;
268 key->nw_tos = nh->tos & ~INET_ECN_MASK;
269 key->nw_proto = nh->protocol;
270 skb_set_transport_header(skb, th_ofs);
272 /* Transport layer. */
273 if (!(nh->frag_off & htons(IP_MF | IP_OFFSET))) {
274 if (key->nw_proto == IPPROTO_TCP) {
275 if (tcphdr_ok(skb)) {
276 struct tcphdr *tcp = tcp_hdr(skb);
277 key->tp_src = tcp->source;
278 key->tp_dst = tcp->dest;
280 /* Avoid tricking other code into
281 * thinking that this packet has an L4
285 } else if (key->nw_proto == IPPROTO_UDP) {
286 if (udphdr_ok(skb)) {
287 struct udphdr *udp = udp_hdr(skb);
288 key->tp_src = udp->source;
289 key->tp_dst = udp->dest;
291 /* Avoid tricking other code into
292 * thinking that this packet has an L4
296 } else if (key->nw_proto == IPPROTO_ICMP) {
297 if (icmphdr_ok(skb)) {
298 struct icmphdr *icmp = icmp_hdr(skb);
299 /* The ICMP type and code fields use the 16-bit
300 * transport port fields, so we need to store them
301 * in 16-bit network byte order. */
302 key->tp_src = htons(icmp->type);
303 key->tp_dst = htons(icmp->code);
305 /* Avoid tricking other code into
306 * thinking that this packet has an L4
312 OVS_CB(skb)->is_frag = true;
314 } else if (key->dl_type == htons(ETH_P_ARP) && arphdr_ok(skb)) {
315 struct arp_eth_header *arp;
317 arp = (struct arp_eth_header *)skb_network_header(skb);
319 if (arp->ar_hrd == htons(ARPHRD_ETHER)
320 && arp->ar_pro == htons(ETH_P_IP)
321 && arp->ar_hln == ETH_ALEN
322 && arp->ar_pln == 4) {
324 /* We only match on the lower 8 bits of the opcode. */
325 if (ntohs(arp->ar_op) <= 0xff) {
326 key->nw_proto = ntohs(arp->ar_op);
329 if (key->nw_proto == ARPOP_REQUEST
330 || key->nw_proto == ARPOP_REPLY) {
331 memcpy(&key->nw_src, arp->ar_sip, sizeof(key->nw_src));
332 memcpy(&key->nw_dst, arp->ar_tip, sizeof(key->nw_dst));
336 skb_reset_transport_header(skb);
341 u32 flow_hash(const struct odp_flow_key *key)
343 return jhash2((u32*)key, sizeof *key / sizeof(u32), hash_seed);
346 int flow_cmp(const struct tbl_node *node, void *key2_)
348 const struct odp_flow_key *key1 = &flow_cast(node)->key;
349 const struct odp_flow_key *key2 = key2_;
351 return !memcmp(key1, key2, sizeof(struct odp_flow_key));
354 /* Initializes the flow module.
355 * Returns zero if successful or a negative error code. */
358 flow_cache = kmem_cache_create("sw_flow", sizeof(struct sw_flow), 0,
360 if (flow_cache == NULL)
363 get_random_bytes(&hash_seed, sizeof hash_seed);
368 /* Uninitializes the flow module. */
371 kmem_cache_destroy(flow_cache);