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;
37 static inline bool arphdr_ok(struct sk_buff *skb)
39 return skb->len >= skb_network_offset(skb) + sizeof(struct arp_eth_header);
42 static inline int check_iphdr(struct sk_buff *skb)
44 unsigned int nh_ofs = skb_network_offset(skb);
47 if (skb->len < nh_ofs + sizeof(struct iphdr))
50 ip_len = ip_hdrlen(skb);
51 if (ip_len < sizeof(struct iphdr) || skb->len < nh_ofs + ip_len)
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.
58 if (!pskb_may_pull(skb, min(nh_ofs + ip_len + 20, skb->len)))
61 skb_set_transport_header(skb, nh_ofs + ip_len);
65 static inline bool tcphdr_ok(struct sk_buff *skb)
67 int th_ofs = skb_transport_offset(skb);
68 if (skb->len >= th_ofs + sizeof(struct tcphdr)) {
69 int tcp_len = tcp_hdrlen(skb);
70 return (tcp_len >= sizeof(struct tcphdr)
71 && skb->len >= th_ofs + tcp_len);
76 static inline bool udphdr_ok(struct sk_buff *skb)
78 return skb->len >= skb_transport_offset(skb) + sizeof(struct udphdr);
81 static inline bool icmphdr_ok(struct sk_buff *skb)
83 return skb->len >= skb_transport_offset(skb) + sizeof(struct icmphdr);
86 #define TCP_FLAGS_OFFSET 13
87 #define TCP_FLAG_MASK 0x3f
89 void flow_used(struct sw_flow *flow, struct sk_buff *skb)
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;
99 spin_lock_bh(&flow->lock);
100 flow->used = jiffies;
101 flow->packet_count++;
102 flow->byte_count += skb->len;
103 flow->tcp_flags |= tcp_flags;
104 spin_unlock_bh(&flow->lock);
107 struct sw_flow_actions *flow_actions_alloc(size_t n_actions)
109 struct sw_flow_actions *sfa;
111 if (n_actions > (PAGE_SIZE - sizeof *sfa) / sizeof(union odp_action))
112 return ERR_PTR(-EINVAL);
114 sfa = kmalloc(sizeof *sfa + n_actions * sizeof(union odp_action),
117 return ERR_PTR(-ENOMEM);
119 sfa->n_actions = n_actions;
124 /* Frees 'flow' immediately. */
125 static void flow_free(struct sw_flow *flow)
129 kfree(flow->sf_acts);
130 kmem_cache_free(flow_cache, flow);
133 void flow_free_tbl(struct tbl_node *node)
135 struct sw_flow *flow = flow_cast(node);
139 /* RCU callback used by flow_deferred_free. */
140 static void rcu_free_flow_callback(struct rcu_head *rcu)
142 struct sw_flow *flow = container_of(rcu, struct sw_flow, rcu);
146 /* Schedules 'flow' to be freed after the next RCU grace period.
147 * The caller must hold rcu_read_lock for this to be sensible. */
148 void flow_deferred_free(struct sw_flow *flow)
150 call_rcu(&flow->rcu, rcu_free_flow_callback);
153 /* RCU callback used by flow_deferred_free_acts. */
154 static void rcu_free_acts_callback(struct rcu_head *rcu)
156 struct sw_flow_actions *sf_acts = container_of(rcu,
157 struct sw_flow_actions, rcu);
161 /* Schedules 'sf_acts' to be freed after the next RCU grace period.
162 * The caller must hold rcu_read_lock for this to be sensible. */
163 void flow_deferred_free_acts(struct sw_flow_actions *sf_acts)
165 call_rcu(&sf_acts->rcu, rcu_free_acts_callback);
168 static void parse_vlan(struct sk_buff *skb, struct odp_flow_key *key)
171 __be16 eth_type; /* ETH_P_8021Q */
174 struct qtag_prefix *qp;
176 if (skb->len < sizeof(struct qtag_prefix) + sizeof(__be16))
179 qp = (struct qtag_prefix *) skb->data;
180 key->dl_vlan = qp->tci & htons(VLAN_VID_MASK);
181 key->dl_vlan_pcp = (ntohs(qp->tci) & VLAN_PCP_MASK) >> VLAN_PCP_SHIFT;
182 __skb_pull(skb, sizeof(struct qtag_prefix));
185 static __be16 parse_ethertype(struct sk_buff *skb)
187 struct llc_snap_hdr {
188 u8 dsap; /* Always 0xAA */
189 u8 ssap; /* Always 0xAA */
194 struct llc_snap_hdr *llc;
197 proto = *(__be16 *) skb->data;
198 __skb_pull(skb, sizeof(__be16));
200 if (ntohs(proto) >= ODP_DL_TYPE_ETH2_CUTOFF)
203 if (unlikely(skb->len < sizeof(struct llc_snap_hdr)))
204 return htons(ODP_DL_TYPE_NOT_ETH_TYPE);
206 llc = (struct llc_snap_hdr *) skb->data;
207 if (llc->dsap != LLC_SAP_SNAP ||
208 llc->ssap != LLC_SAP_SNAP ||
209 (llc->oui[0] | llc->oui[1] | llc->oui[2]) != 0)
210 return htons(ODP_DL_TYPE_NOT_ETH_TYPE);
212 __skb_pull(skb, sizeof(struct llc_snap_hdr));
213 return llc->ethertype;
217 * flow_extract - extracts a flow key from an Ethernet frame.
218 * @skb: sk_buff that contains the frame, with skb->data pointing to the
220 * @in_port: port number on which @skb was received.
221 * @key: output flow key
223 * The caller must ensure that skb->len >= ETH_HLEN.
225 * Returns 0 if successful, otherwise a negative errno value.
227 * Initializes @skb header pointers as follows:
229 * - skb->mac_header: the Ethernet header.
231 * - skb->network_header: just past the Ethernet header, or just past the
232 * VLAN header, to the first byte of the Ethernet payload.
234 * - skb->transport_header: If key->dl_type is ETH_P_IP on output, then just
235 * past the IPv4 header, if one is present and of a correct length,
236 * otherwise the same as skb->network_header. For other key->dl_type
237 * values it is left untouched.
239 * Sets OVS_CB(skb)->is_frag to %true if @skb is an IPv4 fragment, otherwise to
242 int flow_extract(struct sk_buff *skb, u16 in_port, struct odp_flow_key *key)
246 memset(key, 0, sizeof *key);
247 key->tun_id = OVS_CB(skb)->tun_id;
248 key->in_port = in_port;
249 key->dl_vlan = htons(ODP_VLAN_NONE);
250 OVS_CB(skb)->is_frag = false;
253 * We would really like to pull as many bytes as we could possibly
254 * want to parse into the linear data area. Currently that is:
258 * 60 max IP header with options
259 * 20 max TCP/UDP/ICMP header (don't care about options)
263 * But Xen only allocates 64 or 72 bytes for the linear data area in
264 * netback, which means that we would reallocate and copy the skb's
265 * linear data on every packet if we did that. So instead just pull 64
266 * bytes, which is always sufficient without IP options, and then check
267 * whether we need to pull more later when we look at the IP header.
269 if (!pskb_may_pull(skb, min(skb->len, 64u)))
272 skb_reset_mac_header(skb);
276 memcpy(key->dl_src, eth->h_source, ETH_ALEN);
277 memcpy(key->dl_dst, eth->h_dest, ETH_ALEN);
279 /* dl_type, dl_vlan, dl_vlan_pcp. */
280 __skb_pull(skb, 2 * ETH_ALEN);
281 if (eth->h_proto == htons(ETH_P_8021Q))
282 parse_vlan(skb, key);
283 key->dl_type = parse_ethertype(skb);
284 skb_reset_network_header(skb);
285 __skb_push(skb, skb->data - (unsigned char *)eth);
288 if (key->dl_type == htons(ETH_P_IP)) {
292 error = check_iphdr(skb);
293 if (unlikely(error)) {
294 if (error == -EINVAL) {
295 skb->transport_header = skb->network_header;
302 key->nw_src = nh->saddr;
303 key->nw_dst = nh->daddr;
304 key->nw_tos = nh->tos & ~INET_ECN_MASK;
305 key->nw_proto = nh->protocol;
307 /* Transport layer. */
308 if (!(nh->frag_off & htons(IP_MF | IP_OFFSET))) {
309 if (key->nw_proto == IPPROTO_TCP) {
310 if (tcphdr_ok(skb)) {
311 struct tcphdr *tcp = tcp_hdr(skb);
312 key->tp_src = tcp->source;
313 key->tp_dst = tcp->dest;
315 /* Avoid tricking other code into
316 * thinking that this packet has an L4
320 } else if (key->nw_proto == IPPROTO_UDP) {
321 if (udphdr_ok(skb)) {
322 struct udphdr *udp = udp_hdr(skb);
323 key->tp_src = udp->source;
324 key->tp_dst = udp->dest;
326 /* Avoid tricking other code into
327 * thinking that this packet has an L4
331 } else if (key->nw_proto == IPPROTO_ICMP) {
332 if (icmphdr_ok(skb)) {
333 struct icmphdr *icmp = icmp_hdr(skb);
334 /* The ICMP type and code fields use the 16-bit
335 * transport port fields, so we need to store them
336 * in 16-bit network byte order. */
337 key->tp_src = htons(icmp->type);
338 key->tp_dst = htons(icmp->code);
340 /* Avoid tricking other code into
341 * thinking that this packet has an L4
347 OVS_CB(skb)->is_frag = true;
349 } else if (key->dl_type == htons(ETH_P_ARP) && arphdr_ok(skb)) {
350 struct arp_eth_header *arp;
352 arp = (struct arp_eth_header *)skb_network_header(skb);
354 if (arp->ar_hrd == htons(ARPHRD_ETHER)
355 && arp->ar_pro == htons(ETH_P_IP)
356 && arp->ar_hln == ETH_ALEN
357 && arp->ar_pln == 4) {
359 /* We only match on the lower 8 bits of the opcode. */
360 if (ntohs(arp->ar_op) <= 0xff) {
361 key->nw_proto = ntohs(arp->ar_op);
364 if (key->nw_proto == ARPOP_REQUEST
365 || key->nw_proto == ARPOP_REPLY) {
366 memcpy(&key->nw_src, arp->ar_sip, sizeof(key->nw_src));
367 memcpy(&key->nw_dst, arp->ar_tip, sizeof(key->nw_dst));
374 u32 flow_hash(const struct odp_flow_key *key)
376 return jhash2((u32*)key, sizeof *key / sizeof(u32), hash_seed);
379 int flow_cmp(const struct tbl_node *node, void *key2_)
381 const struct odp_flow_key *key1 = &flow_cast(node)->key;
382 const struct odp_flow_key *key2 = key2_;
384 return !memcmp(key1, key2, sizeof(struct odp_flow_key));
387 /* Initializes the flow module.
388 * Returns zero if successful or a negative error code. */
391 flow_cache = kmem_cache_create("sw_flow", sizeof(struct sw_flow), 0,
393 if (flow_cache == NULL)
396 get_random_bytes(&hash_seed, sizeof hash_seed);
401 /* Uninitializes the flow module. */
404 kmem_cache_destroy(flow_cache);