2 * Distributed under the terms of the GNU GPL version 2.
3 * Copyright (c) 2007, 2008, 2009 Nicira Networks.
5 * Significant portions of this file may be copied from parts of the Linux
6 * kernel, by Linus Torvalds and others.
10 #include <linux/netdevice.h>
11 #include <linux/etherdevice.h>
12 #include <linux/if_ether.h>
13 #include <linux/if_vlan.h>
14 #include <net/llc_pdu.h>
15 #include <linux/kernel.h>
16 #include <linux/jiffies.h>
17 #include <linux/llc.h>
18 #include <linux/module.h>
20 #include <linux/rcupdate.h>
21 #include <linux/if_arp.h>
22 #include <linux/if_ether.h>
24 #include <linux/tcp.h>
25 #include <linux/udp.h>
26 #include <linux/icmp.h>
31 struct kmem_cache *flow_cache;
35 __be16 ar_hrd; /* format of hardware address */
36 __be16 ar_pro; /* format of protocol address */
37 unsigned char ar_hln; /* length of hardware address */
38 unsigned char ar_pln; /* length of protocol address */
39 __be16 ar_op; /* ARP opcode (command) */
41 /* Ethernet+IPv4 specific members. */
42 unsigned char ar_sha[ETH_ALEN]; /* sender hardware address */
43 unsigned char ar_sip[4]; /* sender IP address */
44 unsigned char ar_tha[ETH_ALEN]; /* target hardware address */
45 unsigned char ar_tip[4]; /* target IP address */
46 } __attribute__((packed));
48 static inline int arphdr_ok(struct sk_buff *skb)
50 int nh_ofs = skb_network_offset(skb);
51 return pskb_may_pull(skb, nh_ofs + sizeof(struct arp_eth_header));
54 static inline int iphdr_ok(struct sk_buff *skb)
56 int nh_ofs = skb_network_offset(skb);
57 if (skb->len >= nh_ofs + sizeof(struct iphdr)) {
58 int ip_len = ip_hdrlen(skb);
59 return (ip_len >= sizeof(struct iphdr)
60 && pskb_may_pull(skb, nh_ofs + ip_len));
65 static inline int tcphdr_ok(struct sk_buff *skb)
67 int th_ofs = skb_transport_offset(skb);
68 if (pskb_may_pull(skb, 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 int udphdr_ok(struct sk_buff *skb)
78 int th_ofs = skb_transport_offset(skb);
79 return pskb_may_pull(skb, th_ofs + sizeof(struct udphdr));
82 static inline int icmphdr_ok(struct sk_buff *skb)
84 int th_ofs = skb_transport_offset(skb);
85 return pskb_may_pull(skb, th_ofs + sizeof(struct icmphdr));
88 #define TCP_FLAGS_OFFSET 13
89 #define TCP_FLAG_MASK 0x3f
91 static inline struct ovs_tcphdr *ovs_tcp_hdr(const struct sk_buff *skb)
93 return (struct ovs_tcphdr *)skb_transport_header(skb);
96 void flow_used(struct sw_flow *flow, struct sk_buff *skb)
101 if (flow->key.dl_type == htons(ETH_P_IP) && iphdr_ok(skb)) {
102 struct iphdr *nh = ip_hdr(skb);
103 flow->ip_tos = nh->tos;
104 if (flow->key.nw_proto == IPPROTO_TCP && tcphdr_ok(skb)) {
105 u8 *tcp = (u8 *)tcp_hdr(skb);
106 tcp_flags = *(tcp + TCP_FLAGS_OFFSET) & TCP_FLAG_MASK;
110 spin_lock_irqsave(&flow->lock, flags);
111 getnstimeofday(&flow->used);
112 flow->packet_count++;
113 flow->byte_count += skb->len;
114 flow->tcp_flags |= tcp_flags;
115 spin_unlock_irqrestore(&flow->lock, flags);
118 struct sw_flow_actions *flow_actions_alloc(size_t n_actions)
120 struct sw_flow_actions *sfa;
122 if (n_actions > (PAGE_SIZE - sizeof *sfa) / sizeof(union odp_action))
123 return ERR_PTR(-EINVAL);
125 sfa = kmalloc(sizeof *sfa + n_actions * sizeof(union odp_action),
128 return ERR_PTR(-ENOMEM);
130 sfa->n_actions = n_actions;
135 /* Frees 'flow' immediately. */
136 void flow_free(struct sw_flow *flow)
140 kfree(flow->sf_acts);
141 kmem_cache_free(flow_cache, flow);
144 /* RCU callback used by flow_deferred_free. */
145 static void rcu_free_flow_callback(struct rcu_head *rcu)
147 struct sw_flow *flow = container_of(rcu, struct sw_flow, rcu);
151 /* Schedules 'flow' to be freed after the next RCU grace period.
152 * The caller must hold rcu_read_lock for this to be sensible. */
153 void flow_deferred_free(struct sw_flow *flow)
155 call_rcu(&flow->rcu, rcu_free_flow_callback);
158 /* RCU callback used by flow_deferred_free_acts. */
159 static void rcu_free_acts_callback(struct rcu_head *rcu)
161 struct sw_flow_actions *sf_acts = container_of(rcu,
162 struct sw_flow_actions, rcu);
166 /* Schedules 'sf_acts' to be freed after the next RCU grace period.
167 * The caller must hold rcu_read_lock for this to be sensible. */
168 void flow_deferred_free_acts(struct sw_flow_actions *sf_acts)
170 call_rcu(&sf_acts->rcu, rcu_free_acts_callback);
173 #define SNAP_OUI_LEN 3
178 u8 dsap; /* Always 0xAA */
179 u8 ssap; /* Always 0xAA */
181 u8 oui[SNAP_OUI_LEN];
183 } __attribute__ ((packed));
185 static int is_snap(const struct eth_snap_hdr *esh)
187 return (esh->dsap == LLC_SAP_SNAP
188 && esh->ssap == LLC_SAP_SNAP
189 && !memcmp(esh->oui, "\0\0\0", 3));
192 /* Parses the Ethernet frame in 'skb', which was received on 'in_port',
193 * and initializes 'key' to match. Returns 1 if 'skb' contains an IP
194 * fragment, 0 otherwise. */
195 int flow_extract(struct sk_buff *skb, u16 in_port, struct odp_flow_key *key)
198 struct eth_snap_hdr *esh;
202 memset(key, 0, sizeof *key);
203 key->dl_vlan = htons(ODP_VLAN_NONE);
204 key->in_port = in_port;
206 if (skb->len < sizeof *eth)
208 if (!pskb_may_pull(skb, skb->len >= 64 ? 64 : skb->len)) {
212 skb_reset_mac_header(skb);
214 esh = (struct eth_snap_hdr *) eth;
215 nh_ofs = sizeof *eth;
216 if (likely(ntohs(eth->h_proto) >= ODP_DL_TYPE_ETH2_CUTOFF))
217 key->dl_type = eth->h_proto;
218 else if (skb->len >= sizeof *esh && is_snap(esh)) {
219 key->dl_type = esh->ethertype;
220 nh_ofs = sizeof *esh;
222 key->dl_type = htons(ODP_DL_TYPE_NOT_ETH_TYPE);
223 if (skb->len >= nh_ofs + sizeof(struct llc_pdu_un)) {
224 nh_ofs += sizeof(struct llc_pdu_un);
228 /* Check for a VLAN tag */
229 if (key->dl_type == htons(ETH_P_8021Q) &&
230 skb->len >= nh_ofs + sizeof(struct vlan_hdr)) {
231 struct vlan_hdr *vh = (struct vlan_hdr*)(skb->data + nh_ofs);
232 key->dl_type = vh->h_vlan_encapsulated_proto;
233 key->dl_vlan = vh->h_vlan_TCI & htons(VLAN_VID_MASK);
234 key->dl_vlan_pcp = (ntohs(vh->h_vlan_TCI) & 0xe000) >> 13;
235 nh_ofs += sizeof(struct vlan_hdr);
237 memcpy(key->dl_src, eth->h_source, ETH_ALEN);
238 memcpy(key->dl_dst, eth->h_dest, ETH_ALEN);
239 skb_set_network_header(skb, nh_ofs);
242 if (key->dl_type == htons(ETH_P_IP) && iphdr_ok(skb)) {
243 struct iphdr *nh = ip_hdr(skb);
244 int th_ofs = nh_ofs + nh->ihl * 4;
245 key->nw_src = nh->saddr;
246 key->nw_dst = nh->daddr;
247 key->nw_proto = nh->protocol;
248 skb_set_transport_header(skb, th_ofs);
250 /* Transport layer. */
251 if (!(nh->frag_off & htons(IP_MF | IP_OFFSET))) {
252 if (key->nw_proto == IPPROTO_TCP) {
253 if (tcphdr_ok(skb)) {
254 struct tcphdr *tcp = tcp_hdr(skb);
255 key->tp_src = tcp->source;
256 key->tp_dst = tcp->dest;
258 /* Avoid tricking other code into
259 * thinking that this packet has an L4
263 } else if (key->nw_proto == IPPROTO_UDP) {
264 if (udphdr_ok(skb)) {
265 struct udphdr *udp = udp_hdr(skb);
266 key->tp_src = udp->source;
267 key->tp_dst = udp->dest;
269 /* Avoid tricking other code into
270 * thinking that this packet has an L4
274 } else if (key->nw_proto == IPPROTO_ICMP) {
275 if (icmphdr_ok(skb)) {
276 struct icmphdr *icmp = icmp_hdr(skb);
277 /* The ICMP type and code fields use the 16-bit
278 * transport port fields, so we need to store them
279 * in 16-bit network byte order. */
280 key->tp_src = htons(icmp->type);
281 key->tp_dst = htons(icmp->code);
283 /* Avoid tricking other code into
284 * thinking that this packet has an L4
292 } else if (key->dl_type == htons(ETH_P_ARP) && arphdr_ok(skb)) {
293 struct arp_eth_header *arp;
295 arp = (struct arp_eth_header *)skb_network_header(skb);
297 if (arp->ar_hrd == htons(1)
298 && arp->ar_pro == htons(ETH_P_IP)
299 && arp->ar_hln == ETH_ALEN
300 && arp->ar_pln == 4) {
302 /* We only match on the lower 8 bits of the opcode. */
303 if (ntohs(arp->ar_op) <= 0xff) {
304 key->nw_proto = ntohs(arp->ar_op);
307 if (key->nw_proto == ARPOP_REQUEST
308 || key->nw_proto == ARPOP_REPLY) {
309 memcpy(&key->nw_src, arp->ar_sip, sizeof(key->nw_src));
310 memcpy(&key->nw_dst, arp->ar_tip, sizeof(key->nw_dst));
314 skb_reset_transport_header(skb);
319 /* Initializes the flow module.
320 * Returns zero if successful or a negative error code. */
323 flow_cache = kmem_cache_create("sw_flow", sizeof(struct sw_flow), 0,
325 if (flow_cache == NULL)
331 /* Uninitializes the flow module. */
334 kmem_cache_destroy(flow_cache);