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.
9 /* Functions for executing flow actions. */
11 #include <linux/skbuff.h>
14 #include <linux/tcp.h>
15 #include <linux/udp.h>
16 #include <linux/in6.h>
17 #include <linux/if_arp.h>
18 #include <linux/if_vlan.h>
19 #include <net/inet_ecn.h>
21 #include <net/checksum.h>
25 #include "openvswitch/datapath-protocol.h"
28 static struct sk_buff *make_writable(struct sk_buff *skb, unsigned min_headroom, gfp_t gfp)
30 if (skb_cloned(skb)) {
32 unsigned headroom = max(min_headroom, skb_headroom(skb));
34 nskb = skb_copy_expand(skb, headroom, skb_tailroom(skb), gfp);
36 set_skb_csum_bits(skb, nskb);
41 unsigned int hdr_len = (skb_transport_offset(skb)
42 + sizeof(struct tcphdr));
43 if (pskb_may_pull(skb, min(hdr_len, skb->len)))
50 static struct sk_buff *vlan_pull_tag(struct sk_buff *skb)
52 struct vlan_ethhdr *vh = vlan_eth_hdr(skb);
55 /* Verify we were given a vlan packet */
56 if (vh->h_vlan_proto != htons(ETH_P_8021Q))
59 if (OVS_CB(skb)->ip_summed == OVS_CSUM_COMPLETE)
60 skb->csum = csum_sub(skb->csum, csum_partial(skb->data
61 + ETH_HLEN, VLAN_HLEN, 0));
63 memmove(skb->data + VLAN_HLEN, skb->data, 2 * VLAN_ETH_ALEN);
65 eh = (struct ethhdr *)skb_pull(skb, VLAN_HLEN);
67 skb->protocol = eh->h_proto;
68 skb->mac_header += VLAN_HLEN;
73 static struct sk_buff *modify_vlan_tci(struct datapath *dp, struct sk_buff *skb,
74 const struct odp_flow_key *key,
75 const union odp_action *a, int n_actions,
80 if (a->type == ODPAT_SET_VLAN_VID) {
81 tci = ntohs(a->vlan_vid.vlan_vid);
84 tci = a->vlan_pcp.vlan_pcp << VLAN_PCP_SHIFT;
88 skb = make_writable(skb, VLAN_HLEN, gfp);
90 return ERR_PTR(-ENOMEM);
92 if (skb->protocol == htons(ETH_P_8021Q)) {
93 /* Modify vlan id, but maintain other TCI values */
94 struct vlan_ethhdr *vh = vlan_eth_hdr(skb);
95 __be16 old_tci = vh->h_vlan_TCI;
97 vh->h_vlan_TCI = htons((ntohs(vh->h_vlan_TCI) & ~mask) | tci);
99 if (OVS_CB(skb)->ip_summed == OVS_CSUM_COMPLETE) {
100 __be16 diff[] = { ~old_tci, vh->h_vlan_TCI };
102 skb->csum = ~csum_partial((char *)diff, sizeof(diff),
108 /* Add vlan header */
110 /* Set up checksumming pointers for checksum-deferred packets
111 * on Xen. Otherwise, dev_queue_xmit() will try to do this
112 * when we send the packet out on the wire, and it will fail at
113 * that point because skb_checksum_setup() will not look inside
114 * an 802.1Q header. */
115 err = vswitch_skb_checksum_setup(skb);
121 /* GSO is not implemented for packets with an 802.1Q header, so
122 * we have to do segmentation before we add that header.
124 * GSO does work with hardware-accelerated VLAN tagging, but we
125 * can't use hardware-accelerated VLAN tagging since it
126 * requires the device to have a VLAN group configured (with
127 * e.g. vconfig(8)) and we don't do that.
129 * Having to do this here may be a performance loss, since we
130 * can't take advantage of TSO hardware support, although it
131 * does not make a measurable network performance difference
132 * for 1G Ethernet. Fixing that would require patching the
133 * kernel (either to add GSO support to the VLAN protocol or to
134 * support hardware-accelerated VLAN tagging without VLAN
135 * groups configured). */
136 if (skb_is_gso(skb)) {
137 struct sk_buff *segs;
139 segs = skb_gso_segment(skb, 0);
141 if (unlikely(IS_ERR(segs)))
142 return ERR_CAST(segs);
145 struct sk_buff *nskb = segs->next;
150 /* GSO can change the checksum type so update.*/
151 compute_ip_summed(segs, true);
153 segs = __vlan_put_tag(segs, tci);
156 err = execute_actions(dp, segs,
163 while ((segs = nskb)) {
172 } while (segs->next);
175 compute_ip_summed(skb, true);
178 /* The hardware-accelerated version of vlan_put_tag() works
179 * only for a device that has a VLAN group configured (with
180 * e.g. vconfig(8)), so call the software-only version
181 * __vlan_put_tag() directly instead.
183 skb = __vlan_put_tag(skb, tci);
185 return ERR_PTR(-ENOMEM);
187 /* GSO doesn't fix up the hardware computed checksum so this
188 * will only be hit in the non-GSO case. */
189 if (OVS_CB(skb)->ip_summed == OVS_CSUM_COMPLETE)
190 skb->csum = csum_add(skb->csum, csum_partial(skb->data
191 + ETH_HLEN, VLAN_HLEN, 0));
197 static struct sk_buff *strip_vlan(struct sk_buff *skb, gfp_t gfp)
199 skb = make_writable(skb, 0, gfp);
206 static struct sk_buff *set_dl_addr(struct sk_buff *skb,
207 const struct odp_action_dl_addr *a,
210 skb = make_writable(skb, 0, gfp);
212 struct ethhdr *eh = eth_hdr(skb);
213 if (a->type == ODPAT_SET_DL_SRC)
214 memcpy(eh->h_source, a->dl_addr, ETH_ALEN);
216 memcpy(eh->h_dest, a->dl_addr, ETH_ALEN);
221 /* Updates 'sum', which is a field in 'skb''s data, given that a 4-byte field
222 * covered by the sum has been changed from 'from' to 'to'. If set,
223 * 'pseudohdr' indicates that the field is in the TCP or UDP pseudo-header.
224 * Based on nf_proto_csum_replace4. */
225 static void update_csum(__sum16 *sum, struct sk_buff *skb,
226 __be32 from, __be32 to, int pseudohdr)
228 __be32 diff[] = { ~from, to };
230 if (OVS_CB(skb)->ip_summed != OVS_CSUM_PARTIAL) {
231 *sum = csum_fold(csum_partial((char *)diff, sizeof(diff),
232 ~csum_unfold(*sum)));
233 if (OVS_CB(skb)->ip_summed == OVS_CSUM_COMPLETE && pseudohdr)
234 skb->csum = ~csum_partial((char *)diff, sizeof(diff),
236 } else if (pseudohdr)
237 *sum = ~csum_fold(csum_partial((char *)diff, sizeof(diff),
241 static struct sk_buff *set_nw_addr(struct sk_buff *skb,
242 const struct odp_flow_key *key,
243 const struct odp_action_nw_addr *a,
246 if (key->dl_type != htons(ETH_P_IP))
249 skb = make_writable(skb, 0, gfp);
251 struct iphdr *nh = ip_hdr(skb);
252 u32 *f = a->type == ODPAT_SET_NW_SRC ? &nh->saddr : &nh->daddr;
254 u32 new = a->nw_addr;
256 if (key->nw_proto == IPPROTO_TCP) {
257 struct tcphdr *th = tcp_hdr(skb);
258 update_csum(&th->check, skb, old, new, 1);
259 } else if (key->nw_proto == IPPROTO_UDP) {
260 struct udphdr *th = udp_hdr(skb);
261 update_csum(&th->check, skb, old, new, 1);
263 update_csum(&nh->check, skb, old, new, 0);
269 static struct sk_buff *set_nw_tos(struct sk_buff *skb,
270 const struct odp_flow_key *key,
271 const struct odp_action_nw_tos *a,
274 if (key->dl_type != htons(ETH_P_IP))
277 skb = make_writable(skb, 0, gfp);
279 struct iphdr *nh = ip_hdr(skb);
284 /* Set the DSCP bits and preserve the ECN bits. */
285 new = a->nw_tos | (nh->tos & INET_ECN_MASK);
286 update_csum(&nh->check, skb, htons((uint16_t)old),
287 htons((uint16_t)new), 0);
293 static struct sk_buff *set_tp_port(struct sk_buff *skb,
294 const struct odp_flow_key *key,
295 const struct odp_action_tp_port *a, gfp_t gfp)
299 if (key->dl_type != htons(ETH_P_IP))
302 if (key->nw_proto == IPPROTO_TCP)
303 check_ofs = offsetof(struct tcphdr, check);
304 else if (key->nw_proto == IPPROTO_UDP)
305 check_ofs = offsetof(struct udphdr, check);
309 skb = make_writable(skb, 0, gfp);
311 struct udphdr *th = udp_hdr(skb);
312 u16 *f = a->type == ODPAT_SET_TP_SRC ? &th->source : &th->dest;
314 u16 new = a->tp_port;
315 update_csum((u16*)(skb_transport_header(skb) + check_ofs),
323 * is_spoofed_arp - check for invalid ARP packet
325 * @skb: skbuff containing an Ethernet packet, with network header pointing
326 * just past the Ethernet and optional 802.1Q header.
327 * @key: flow key extracted from @skb by flow_extract()
329 * Returns true if @skb is an invalid Ethernet+IPv4 ARP packet: one with screwy
330 * or truncated header fields or one whose inner and outer Ethernet address
333 static bool is_spoofed_arp(struct sk_buff *skb, const struct odp_flow_key *key)
335 struct arp_eth_header *arp;
337 if (key->dl_type != htons(ETH_P_ARP))
340 if (skb_network_offset(skb) + sizeof(struct arp_eth_header) > skb->len)
343 arp = (struct arp_eth_header *)skb_network_header(skb);
344 return (arp->ar_hrd != htons(ARPHRD_ETHER) ||
345 arp->ar_pro != htons(ETH_P_IP) ||
346 arp->ar_hln != ETH_ALEN ||
348 compare_ether_addr(arp->ar_sha, eth_hdr(skb)->h_source));
351 static void do_output(struct datapath *dp, struct sk_buff *skb, int out_port)
358 p = rcu_dereference(dp->ports[out_port]);
362 vport_send(p->vport, skb);
369 /* Never consumes 'skb'. Returns a port that 'skb' should be sent to, -1 if
371 static int output_group(struct datapath *dp, __u16 group,
372 struct sk_buff *skb, gfp_t gfp)
374 struct dp_port_group *g = rcu_dereference(dp->groups[group]);
380 for (i = 0; i < g->n_ports; i++) {
381 struct dp_port *p = rcu_dereference(dp->ports[g->ports[i]]);
382 if (!p || OVS_CB(skb)->dp_port == p)
384 if (prev_port != -1) {
385 struct sk_buff *clone = skb_clone(skb, gfp);
388 do_output(dp, clone, prev_port);
390 prev_port = p->port_no;
395 static int output_control(struct datapath *dp, struct sk_buff *skb, u32 arg,
398 skb = skb_clone(skb, gfp);
401 return dp_output_control(dp, skb, _ODPL_ACTION_NR, arg);
404 /* Send a copy of this packet up to the sFlow agent, along with extra
405 * information about what happened to it. */
406 static void sflow_sample(struct datapath *dp, struct sk_buff *skb,
407 const union odp_action *a, int n_actions,
408 gfp_t gfp, struct dp_port *dp_port)
410 struct odp_sflow_sample_header *hdr;
411 unsigned int actlen = n_actions * sizeof(union odp_action);
412 unsigned int hdrlen = sizeof(struct odp_sflow_sample_header);
413 struct sk_buff *nskb;
415 nskb = skb_copy_expand(skb, actlen + hdrlen, 0, gfp);
419 memcpy(__skb_push(nskb, actlen), a, actlen);
420 hdr = (struct odp_sflow_sample_header*)__skb_push(nskb, hdrlen);
421 hdr->n_actions = n_actions;
422 hdr->sample_pool = atomic_read(&dp_port->sflow_pool);
423 dp_output_control(dp, nskb, _ODPL_SFLOW_NR, 0);
426 /* Execute a list of actions against 'skb'. */
427 int execute_actions(struct datapath *dp, struct sk_buff *skb,
428 const struct odp_flow_key *key,
429 const union odp_action *a, int n_actions,
432 /* Every output action needs a separate clone of 'skb', but the common
433 * case is just a single output action, so that doing a clone and
434 * then freeing the original skbuff is wasteful. So the following code
435 * is slightly obscure just to avoid that. */
437 u32 priority = skb->priority;
440 if (dp->sflow_probability) {
441 struct dp_port *p = OVS_CB(skb)->dp_port;
443 atomic_inc(&p->sflow_pool);
444 if (dp->sflow_probability == UINT_MAX ||
445 net_random() < dp->sflow_probability)
446 sflow_sample(dp, skb, a, n_actions, gfp, p);
450 OVS_CB(skb)->tun_id = 0;
452 for (; n_actions > 0; a++, n_actions--) {
453 if (prev_port != -1) {
454 do_output(dp, skb_clone(skb, gfp), prev_port);
460 prev_port = a->output.port;
463 case ODPAT_OUTPUT_GROUP:
464 prev_port = output_group(dp, a->output_group.group,
468 case ODPAT_CONTROLLER:
469 err = output_control(dp, skb, a->controller.arg, gfp);
476 case ODPAT_SET_TUNNEL:
477 OVS_CB(skb)->tun_id = a->tunnel.tun_id;
480 case ODPAT_SET_VLAN_VID:
481 case ODPAT_SET_VLAN_PCP:
482 skb = modify_vlan_tci(dp, skb, key, a, n_actions, gfp);
487 case ODPAT_STRIP_VLAN:
488 skb = strip_vlan(skb, gfp);
491 case ODPAT_SET_DL_SRC:
492 case ODPAT_SET_DL_DST:
493 skb = set_dl_addr(skb, &a->dl_addr, gfp);
496 case ODPAT_SET_NW_SRC:
497 case ODPAT_SET_NW_DST:
498 skb = set_nw_addr(skb, key, &a->nw_addr, gfp);
501 case ODPAT_SET_NW_TOS:
502 skb = set_nw_tos(skb, key, &a->nw_tos, gfp);
505 case ODPAT_SET_TP_SRC:
506 case ODPAT_SET_TP_DST:
507 skb = set_tp_port(skb, key, &a->tp_port, gfp);
510 case ODPAT_SET_PRIORITY:
511 skb->priority = a->priority.priority;
514 case ODPAT_POP_PRIORITY:
515 skb->priority = priority;
518 case ODPAT_DROP_SPOOFED_ARP:
519 if (unlikely(is_spoofed_arp(skb, key)))
528 do_output(dp, skb, prev_port);