2 * Distributed under the terms of the GNU GPL version 2.
3 * Copyright (c) 2007, 2008, 2009, 2010, 2011 Nicira Networks.
5 * Significant portions of this file may be copied from parts of the Linux
6 * kernel, by Linus Torvalds and others.
9 /* Functions for executing flow actions. */
11 #define pr_fmt(fmt) KBUILD_MODNAME ": " fmt
13 #include <linux/skbuff.h>
16 #include <linux/openvswitch.h>
17 #include <linux/tcp.h>
18 #include <linux/udp.h>
19 #include <linux/in6.h>
20 #include <linux/if_arp.h>
21 #include <linux/if_vlan.h>
22 #include <net/inet_ecn.h>
24 #include <net/checksum.h>
32 static int do_execute_actions(struct datapath *dp, struct sk_buff *skb,
33 const struct nlattr *attr, int len, bool keep_skb);
35 static int make_writable(struct sk_buff *skb, int write_len)
37 if (!skb_cloned(skb) || skb_clone_writable(skb, write_len))
40 return pskb_expand_head(skb, 0, 0, GFP_ATOMIC);
43 /* remove VLAN header from packet and update csum accrodingly. */
44 static int __pop_vlan_tci(struct sk_buff *skb, __be16 *current_tci)
47 struct vlan_ethhdr *veth;
50 err = make_writable(skb, VLAN_ETH_HLEN);
54 if (get_ip_summed(skb) == OVS_CSUM_COMPLETE)
55 skb->csum = csum_sub(skb->csum, csum_partial(skb->data
56 + ETH_HLEN, VLAN_HLEN, 0));
58 veth = (struct vlan_ethhdr *) skb->data;
59 *current_tci = veth->h_vlan_TCI;
61 memmove(skb->data + VLAN_HLEN, skb->data, 2 * ETH_ALEN);
63 eh = (struct ethhdr *)__skb_pull(skb, VLAN_HLEN);
65 skb->protocol = eh->h_proto;
66 skb->mac_header += VLAN_HLEN;
71 static int pop_vlan(struct sk_buff *skb)
76 if (likely(vlan_tx_tag_present(skb))) {
79 if (unlikely(skb->protocol != htons(ETH_P_8021Q) ||
80 skb->len < VLAN_ETH_HLEN))
83 err = __pop_vlan_tci(skb, &tci);
87 /* move next vlan tag to hw accel tag */
88 if (likely(skb->protocol != htons(ETH_P_8021Q) ||
89 skb->len < VLAN_ETH_HLEN))
92 err = __pop_vlan_tci(skb, &tci);
96 __vlan_hwaccel_put_tag(skb, ntohs(tci));
100 static int push_vlan(struct sk_buff *skb, const struct ovs_key_8021q *q_key)
102 if (unlikely(vlan_tx_tag_present(skb))) {
105 /* push down current VLAN tag */
106 current_tag = vlan_tx_tag_get(skb);
108 if (!__vlan_put_tag(skb, current_tag))
111 if (get_ip_summed(skb) == OVS_CSUM_COMPLETE)
112 skb->csum = csum_add(skb->csum, csum_partial(skb->data
113 + ETH_HLEN, VLAN_HLEN, 0));
116 __vlan_hwaccel_put_tag(skb, ntohs(q_key->q_tci));
120 static int set_eth_addr(struct sk_buff *skb,
121 const struct ovs_key_ethernet *eth_key)
124 err = make_writable(skb, ETH_HLEN);
128 memcpy(eth_hdr(skb)->h_source, eth_key->eth_src, ETH_HLEN);
129 memcpy(eth_hdr(skb)->h_dest, eth_key->eth_dst, ETH_HLEN);
134 static void set_ip_addr(struct sk_buff *skb, struct iphdr *nh,
135 __be32 *addr, __be32 new_addr)
137 int transport_len = skb->len - skb_transport_offset(skb);
139 if (nh->protocol == IPPROTO_TCP) {
140 if (likely(transport_len >= sizeof(struct tcphdr)))
141 inet_proto_csum_replace4(&tcp_hdr(skb)->check, skb,
143 } else if (nh->protocol == IPPROTO_UDP) {
144 if (likely(transport_len >= sizeof(struct udphdr)))
145 inet_proto_csum_replace4(&udp_hdr(skb)->check, skb,
149 csum_replace4(&nh->check, *addr, new_addr);
150 skb_clear_rxhash(skb);
154 static void set_ip_tos(struct sk_buff *skb, struct iphdr *nh, u8 new_tos)
158 /* Set the DSCP bits and preserve the ECN bits. */
160 new = new_tos | (nh->tos & INET_ECN_MASK);
161 csum_replace2(&nh->check, htons(old), htons(new));
165 static int set_ipv4(struct sk_buff *skb, const struct ovs_key_ipv4 *ipv4_key)
170 err = make_writable(skb, skb_network_offset(skb) +
171 sizeof(struct iphdr));
177 if (ipv4_key->ipv4_src != nh->saddr)
178 set_ip_addr(skb, nh, &nh->saddr, ipv4_key->ipv4_src);
180 if (ipv4_key->ipv4_dst != nh->daddr)
181 set_ip_addr(skb, nh, &nh->daddr, ipv4_key->ipv4_dst);
183 if (ipv4_key->ipv4_tos != nh->tos)
184 set_ip_tos(skb, nh, ipv4_key->ipv4_tos);
189 /* Must follow make_writable() since that can move the skb data. */
190 static void set_tp_port(struct sk_buff *skb, __be16 *port,
191 __be16 new_port, __sum16 *check)
193 inet_proto_csum_replace2(check, skb, *port, new_port, 0);
195 skb_clear_rxhash(skb);
198 static int set_udp_port(struct sk_buff *skb,
199 const struct ovs_key_udp *udp_port_key)
204 err = make_writable(skb, skb_transport_offset(skb) +
205 sizeof(struct udphdr));
210 if (udp_port_key->udp_src != uh->source)
211 set_tp_port(skb, &uh->source, udp_port_key->udp_src, &uh->check);
213 if (udp_port_key->udp_dst != uh->dest)
214 set_tp_port(skb, &uh->dest, udp_port_key->udp_dst, &uh->check);
219 static int set_tcp_port(struct sk_buff *skb,
220 const struct ovs_key_tcp *tcp_port_key)
225 err = make_writable(skb, skb_transport_offset(skb) +
226 sizeof(struct tcphdr));
231 if (tcp_port_key->tcp_src != th->source)
232 set_tp_port(skb, &th->source, tcp_port_key->tcp_src, &th->check);
234 if (tcp_port_key->tcp_dst != th->dest)
235 set_tp_port(skb, &th->dest, tcp_port_key->tcp_dst, &th->check);
240 static int do_output(struct datapath *dp, struct sk_buff *skb, int out_port)
247 vport = rcu_dereference(dp->ports[out_port]);
248 if (unlikely(!vport)) {
253 vport_send(vport, skb);
257 static int output_userspace(struct datapath *dp, struct sk_buff *skb,
258 const struct nlattr *attr)
260 struct dp_upcall_info upcall;
261 const struct nlattr *a;
264 upcall.cmd = OVS_PACKET_CMD_ACTION;
265 upcall.key = &OVS_CB(skb)->flow->key;
266 upcall.userdata = NULL;
269 for (a = nla_data(attr), rem = nla_len(attr); rem > 0;
270 a = nla_next(a, &rem)) {
271 switch (nla_type(a)) {
272 case OVS_USERSPACE_ATTR_USERDATA:
276 case OVS_USERSPACE_ATTR_PID:
277 upcall.pid = nla_get_u32(a);
282 return dp_upcall(dp, skb, &upcall);
285 static int sample(struct datapath *dp, struct sk_buff *skb,
286 const struct nlattr *attr)
288 const struct nlattr *acts_list = NULL;
289 const struct nlattr *a;
292 for (a = nla_data(attr), rem = nla_len(attr); rem > 0;
293 a = nla_next(a, &rem)) {
294 switch (nla_type(a)) {
295 case OVS_SAMPLE_ATTR_PROBABILITY:
296 if (net_random() >= nla_get_u32(a))
300 case OVS_SAMPLE_ATTR_ACTIONS:
306 return do_execute_actions(dp, skb, nla_data(acts_list),
307 nla_len(acts_list), true);
310 static int execute_set_action(struct sk_buff *skb,
311 const struct nlattr *nested_attr)
315 switch (nla_type(nested_attr)) {
316 case OVS_KEY_ATTR_PRIORITY:
317 skb->priority = nla_get_u32(nested_attr);
320 case OVS_KEY_ATTR_TUN_ID:
321 OVS_CB(skb)->tun_id = nla_get_be64(nested_attr);
324 case OVS_KEY_ATTR_ETHERNET:
325 err = set_eth_addr(skb, nla_data(nested_attr));
328 case OVS_KEY_ATTR_IPV4:
329 err = set_ipv4(skb, nla_data(nested_attr));
332 case OVS_KEY_ATTR_TCP:
333 err = set_tcp_port(skb, nla_data(nested_attr));
336 case OVS_KEY_ATTR_UDP:
337 err = set_udp_port(skb, nla_data(nested_attr));
344 /* Execute a list of actions against 'skb'. */
345 static int do_execute_actions(struct datapath *dp, struct sk_buff *skb,
346 const struct nlattr *attr, int len, bool keep_skb)
348 /* Every output action needs a separate clone of 'skb', but the common
349 * case is just a single output action, so that doing a clone and
350 * then freeing the original skbuff is wasteful. So the following code
351 * is slightly obscure just to avoid that. */
353 const struct nlattr *a;
356 for (a = attr, rem = len; rem > 0;
357 a = nla_next(a, &rem)) {
360 if (prev_port != -1) {
361 do_output(dp, skb_clone(skb, GFP_ATOMIC), prev_port);
365 switch (nla_type(a)) {
366 case OVS_ACTION_ATTR_OUTPUT:
367 prev_port = nla_get_u32(a);
370 case OVS_ACTION_ATTR_USERSPACE:
371 output_userspace(dp, skb, a);
374 case OVS_ACTION_ATTR_PUSH:
375 /* Only supported push action is on vlan tag. */
376 err = push_vlan(skb, nla_data(nla_data(a)));
377 if (unlikely(err)) /* skb already freed. */
381 case OVS_ACTION_ATTR_POP:
382 /* Only supported pop action is on vlan tag. */
386 case OVS_ACTION_ATTR_SET:
387 err = execute_set_action(skb, nla_data(a));
390 case OVS_ACTION_ATTR_SAMPLE:
391 err = sample(dp, skb, a);
401 if (prev_port != -1) {
403 skb = skb_clone(skb, GFP_ATOMIC);
405 do_output(dp, skb, prev_port);
406 } else if (!keep_skb)
412 /* We limit the number of times that we pass into execute_actions()
413 * to avoid blowing out the stack in the event that we have a loop. */
416 struct loop_counter {
417 u8 count; /* Count. */
418 bool looping; /* Loop detected? */
421 static DEFINE_PER_CPU(struct loop_counter, loop_counters);
423 static int loop_suppress(struct datapath *dp, struct sw_flow_actions *actions)
426 pr_warn("%s: flow looped %d times, dropping\n",
427 dp_name(dp), MAX_LOOPS);
428 actions->actions_len = 0;
432 /* Execute a list of actions against 'skb'. */
433 int execute_actions(struct datapath *dp, struct sk_buff *skb)
435 struct sw_flow_actions *acts = rcu_dereference(OVS_CB(skb)->flow->sf_acts);
436 struct loop_counter *loop;
439 /* Check whether we've looped too much. */
440 loop = &__get_cpu_var(loop_counters);
441 if (unlikely(++loop->count > MAX_LOOPS))
442 loop->looping = true;
443 if (unlikely(loop->looping)) {
444 error = loop_suppress(dp, acts);
449 OVS_CB(skb)->tun_id = 0;
450 error = do_execute_actions(dp, skb, acts->actions,
451 acts->actions_len, false);
453 /* Check whether sub-actions looped too much. */
454 if (unlikely(loop->looping))
455 error = loop_suppress(dp, acts);
458 /* Decrement loop counter. */
460 loop->looping = false;