2 * Copyright (c) 2009, 2010, 2011, 2012, 2013, 2014 Nicira, Inc.
3 * Copyright (c) 2013 Simon Horman
5 * Licensed under the Apache License, Version 2.0 (the "License");
6 * you may not use this file except in compliance with the License.
7 * You may obtain a copy of the License at:
9 * http://www.apache.org/licenses/LICENSE-2.0
11 * Unless required by applicable law or agreed to in writing, software
12 * distributed under the License is distributed on an "AS IS" BASIS,
13 * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
14 * See the License for the specific language governing permissions and
15 * limitations under the License.
19 #include "odp-execute.h"
20 #include <arpa/inet.h>
21 #include <netinet/in.h>
22 #include <netinet/ip6.h>
29 #include "odp-netlink.h"
31 #include "packet-dpif.h"
34 #include "unaligned.h"
37 /* Masked copy of an ethernet address. 'src' is already properly masked. */
39 ether_addr_copy_masked(uint8_t *dst, const uint8_t *src,
44 for (i = 0; i < ETH_ADDR_LEN; i++) {
45 dst[i] = src[i] | (dst[i] & ~mask[i]);
50 odp_eth_set_addrs(struct ofpbuf *packet, const struct ovs_key_ethernet *key,
51 const struct ovs_key_ethernet *mask)
53 struct eth_header *eh = ofpbuf_l2(packet);
57 memcpy(eh->eth_src, key->eth_src, sizeof eh->eth_src);
58 memcpy(eh->eth_dst, key->eth_dst, sizeof eh->eth_dst);
60 ether_addr_copy_masked(eh->eth_src, key->eth_src, mask->eth_src);
61 ether_addr_copy_masked(eh->eth_dst, key->eth_dst, mask->eth_dst);
67 odp_set_ipv4(struct ofpbuf *packet, const struct ovs_key_ipv4 *key,
68 const struct ovs_key_ipv4 *mask)
70 struct ip_header *nh = ofpbuf_l3(packet);
74 key->ipv4_src | (get_16aligned_be32(&nh->ip_src) & ~mask->ipv4_src),
75 key->ipv4_dst | (get_16aligned_be32(&nh->ip_dst) & ~mask->ipv4_dst),
76 key->ipv4_tos | (nh->ip_tos & ~mask->ipv4_tos),
77 key->ipv4_ttl | (nh->ip_ttl & ~mask->ipv4_ttl));
80 static const ovs_be32 *
81 mask_ipv6_addr(const ovs_16aligned_be32 *old, const ovs_be32 *addr,
82 const ovs_be32 *mask, ovs_be32 *masked)
84 for (int i = 0; i < 4; i++) {
85 masked[i] = addr[i] | (get_16aligned_be32(&old[i]) & ~mask[i]);
92 odp_set_ipv6(struct ofpbuf *packet, const struct ovs_key_ipv6 *key,
93 const struct ovs_key_ipv6 *mask)
95 struct ovs_16aligned_ip6_hdr *nh = ofpbuf_l3(packet);
96 ovs_be32 sbuf[4], dbuf[4];
97 uint8_t old_tc = ntohl(get_16aligned_be32(&nh->ip6_flow)) >> 20;
98 ovs_be32 old_fl = get_16aligned_be32(&nh->ip6_flow) & htonl(0xfffff);
103 mask_ipv6_addr(nh->ip6_src.be32, key->ipv6_src, mask->ipv6_src, sbuf),
104 mask_ipv6_addr(nh->ip6_dst.be32, key->ipv6_dst, mask->ipv6_dst, dbuf),
105 key->ipv6_tclass | (old_tc & ~mask->ipv6_tclass),
106 key->ipv6_label | (old_fl & ~mask->ipv6_label),
107 key->ipv6_hlimit | (nh->ip6_hlim & ~mask->ipv6_hlimit));
111 odp_set_tcp(struct ofpbuf *packet, const struct ovs_key_tcp *key,
112 const struct ovs_key_tcp *mask)
114 struct tcp_header *th = ofpbuf_l4(packet);
116 if (OVS_LIKELY(th && ofpbuf_get_tcp_payload(packet))) {
117 packet_set_tcp_port(packet,
118 key->tcp_src | (th->tcp_src & ~mask->tcp_src),
119 key->tcp_dst | (th->tcp_dst & ~mask->tcp_dst));
124 odp_set_udp(struct ofpbuf *packet, const struct ovs_key_udp *key,
125 const struct ovs_key_udp *mask)
127 struct udp_header *uh = ofpbuf_l4(packet);
129 if (OVS_LIKELY(uh && ofpbuf_get_udp_payload(packet))) {
130 packet_set_udp_port(packet,
131 key->udp_src | (uh->udp_src & ~mask->udp_src),
132 key->udp_dst | (uh->udp_dst & ~mask->udp_dst));
137 odp_set_sctp(struct ofpbuf *packet, const struct ovs_key_sctp *key,
138 const struct ovs_key_sctp *mask)
140 struct sctp_header *sh = ofpbuf_l4(packet);
142 if (OVS_LIKELY(sh && ofpbuf_get_sctp_payload(packet))) {
143 packet_set_sctp_port(packet,
144 key->sctp_src | (sh->sctp_src & ~mask->sctp_src),
145 key->sctp_dst | (sh->sctp_dst & ~mask->sctp_dst));
150 odp_set_tunnel_action(const struct nlattr *a, struct flow_tnl *tun_key)
152 enum odp_key_fitness fitness;
154 fitness = odp_tun_key_from_attr(a, tun_key);
155 ovs_assert(fitness != ODP_FIT_ERROR);
159 set_arp(struct ofpbuf *packet, const struct ovs_key_arp *key,
160 const struct ovs_key_arp *mask)
162 struct arp_eth_header *arp = ofpbuf_l3(packet);
165 arp->ar_op = key->arp_op;
166 memcpy(arp->ar_sha, key->arp_sha, ETH_ADDR_LEN);
167 put_16aligned_be32(&arp->ar_spa, key->arp_sip);
168 memcpy(arp->ar_tha, key->arp_tha, ETH_ADDR_LEN);
169 put_16aligned_be32(&arp->ar_tpa, key->arp_tip);
171 ovs_be32 ar_spa = get_16aligned_be32(&arp->ar_spa);
172 ovs_be32 ar_tpa = get_16aligned_be32(&arp->ar_tpa);
174 arp->ar_op = key->arp_op | (arp->ar_op & ~mask->arp_op);
175 ether_addr_copy_masked(arp->ar_sha, key->arp_sha, mask->arp_sha);
176 put_16aligned_be32(&arp->ar_spa,
177 key->arp_sip | (ar_spa & ~mask->arp_sip));
178 ether_addr_copy_masked(arp->ar_tha, key->arp_tha, mask->arp_tha);
179 put_16aligned_be32(&arp->ar_tpa,
180 key->arp_tip | (ar_tpa & ~mask->arp_tip));
185 odp_execute_set_action(struct dpif_packet *packet, const struct nlattr *a)
187 enum ovs_key_attr type = nl_attr_type(a);
188 const struct ovs_key_ipv4 *ipv4_key;
189 const struct ovs_key_ipv6 *ipv6_key;
190 struct pkt_metadata *md = &packet->md;
193 case OVS_KEY_ATTR_PRIORITY:
194 md->skb_priority = nl_attr_get_u32(a);
197 case OVS_KEY_ATTR_TUNNEL:
198 odp_set_tunnel_action(a, &md->tunnel);
201 case OVS_KEY_ATTR_SKB_MARK:
202 md->pkt_mark = nl_attr_get_u32(a);
205 case OVS_KEY_ATTR_ETHERNET:
206 odp_eth_set_addrs(&packet->ofpbuf, nl_attr_get(a), NULL);
209 case OVS_KEY_ATTR_IPV4:
210 ipv4_key = nl_attr_get_unspec(a, sizeof(struct ovs_key_ipv4));
211 packet_set_ipv4(&packet->ofpbuf, ipv4_key->ipv4_src,
212 ipv4_key->ipv4_dst, ipv4_key->ipv4_tos,
216 case OVS_KEY_ATTR_IPV6:
217 ipv6_key = nl_attr_get_unspec(a, sizeof(struct ovs_key_ipv6));
218 packet_set_ipv6(&packet->ofpbuf, ipv6_key->ipv6_proto,
219 ipv6_key->ipv6_src, ipv6_key->ipv6_dst,
220 ipv6_key->ipv6_tclass, ipv6_key->ipv6_label,
221 ipv6_key->ipv6_hlimit);
224 case OVS_KEY_ATTR_TCP:
225 if (OVS_LIKELY(ofpbuf_get_tcp_payload(&packet->ofpbuf))) {
226 const struct ovs_key_tcp *tcp_key
227 = nl_attr_get_unspec(a, sizeof(struct ovs_key_tcp));
229 packet_set_tcp_port(&packet->ofpbuf, tcp_key->tcp_src,
234 case OVS_KEY_ATTR_UDP:
235 if (OVS_LIKELY(ofpbuf_get_udp_payload(&packet->ofpbuf))) {
236 const struct ovs_key_udp *udp_key
237 = nl_attr_get_unspec(a, sizeof(struct ovs_key_udp));
239 packet_set_udp_port(&packet->ofpbuf, udp_key->udp_src,
244 case OVS_KEY_ATTR_SCTP:
245 if (OVS_LIKELY(ofpbuf_get_sctp_payload(&packet->ofpbuf))) {
246 const struct ovs_key_sctp *sctp_key
247 = nl_attr_get_unspec(a, sizeof(struct ovs_key_sctp));
249 packet_set_sctp_port(&packet->ofpbuf, sctp_key->sctp_src,
254 case OVS_KEY_ATTR_MPLS:
255 set_mpls_lse(&packet->ofpbuf, nl_attr_get_be32(a));
258 case OVS_KEY_ATTR_ARP:
259 set_arp(&packet->ofpbuf, nl_attr_get(a), NULL);
262 case OVS_KEY_ATTR_DP_HASH:
263 md->dp_hash = nl_attr_get_u32(a);
264 dpif_packet_set_dp_hash(packet, md->dp_hash);
267 case OVS_KEY_ATTR_RECIRC_ID:
268 md->recirc_id = nl_attr_get_u32(a);
271 case OVS_KEY_ATTR_UNSPEC:
272 case OVS_KEY_ATTR_ENCAP:
273 case OVS_KEY_ATTR_ETHERTYPE:
274 case OVS_KEY_ATTR_IN_PORT:
275 case OVS_KEY_ATTR_VLAN:
276 case OVS_KEY_ATTR_ICMP:
277 case OVS_KEY_ATTR_ICMPV6:
278 case OVS_KEY_ATTR_ND:
279 case OVS_KEY_ATTR_TCP_FLAGS:
280 case __OVS_KEY_ATTR_MAX:
286 #define get_mask(a, type) ((const type *)(const void *)(a + 1) + 1)
289 odp_execute_masked_set_action(struct dpif_packet *packet,
290 const struct nlattr *a)
292 struct pkt_metadata *md = &packet->md;
293 enum ovs_key_attr type = nl_attr_type(a);
297 case OVS_KEY_ATTR_PRIORITY:
298 md->skb_priority = nl_attr_get_u32(a)
299 | (md->skb_priority & ~*get_mask(a, uint32_t));
302 case OVS_KEY_ATTR_SKB_MARK:
303 md->pkt_mark = nl_attr_get_u32(a)
304 | (md->pkt_mark & ~*get_mask(a, uint32_t));
307 case OVS_KEY_ATTR_ETHERNET:
308 odp_eth_set_addrs(&packet->ofpbuf, nl_attr_get(a),
309 get_mask(a, struct ovs_key_ethernet));
312 case OVS_KEY_ATTR_IPV4:
313 odp_set_ipv4(&packet->ofpbuf, nl_attr_get(a),
314 get_mask(a, struct ovs_key_ipv4));
317 case OVS_KEY_ATTR_IPV6:
318 odp_set_ipv6(&packet->ofpbuf, nl_attr_get(a),
319 get_mask(a, struct ovs_key_ipv6));
322 case OVS_KEY_ATTR_TCP:
323 odp_set_tcp(&packet->ofpbuf, nl_attr_get(a),
324 get_mask(a, struct ovs_key_tcp));
327 case OVS_KEY_ATTR_UDP:
328 odp_set_udp(&packet->ofpbuf, nl_attr_get(a),
329 get_mask(a, struct ovs_key_udp));
332 case OVS_KEY_ATTR_SCTP:
333 odp_set_sctp(&packet->ofpbuf, nl_attr_get(a),
334 get_mask(a, struct ovs_key_sctp));
337 case OVS_KEY_ATTR_MPLS:
338 mh = ofpbuf_l2_5(&packet->ofpbuf);
340 put_16aligned_be32(&mh->mpls_lse, nl_attr_get_be32(a)
341 | (get_16aligned_be32(&mh->mpls_lse)
342 & ~*get_mask(a, ovs_be32)));
346 case OVS_KEY_ATTR_ARP:
347 set_arp(&packet->ofpbuf, nl_attr_get(a),
348 get_mask(a, struct ovs_key_arp));
351 case OVS_KEY_ATTR_DP_HASH:
352 md->dp_hash = nl_attr_get_u32(a)
353 | (dpif_packet_get_dp_hash(packet) & ~*get_mask(a, uint32_t));
354 dpif_packet_set_dp_hash(packet, md->dp_hash);
357 case OVS_KEY_ATTR_RECIRC_ID:
358 md->recirc_id = nl_attr_get_u32(a)
359 | (md->recirc_id & ~*get_mask(a, uint32_t));
362 case OVS_KEY_ATTR_TUNNEL: /* Masked data not supported for tunnel. */
363 case OVS_KEY_ATTR_UNSPEC:
364 case OVS_KEY_ATTR_ENCAP:
365 case OVS_KEY_ATTR_ETHERTYPE:
366 case OVS_KEY_ATTR_IN_PORT:
367 case OVS_KEY_ATTR_VLAN:
368 case OVS_KEY_ATTR_ICMP:
369 case OVS_KEY_ATTR_ICMPV6:
370 case OVS_KEY_ATTR_ND:
371 case OVS_KEY_ATTR_TCP_FLAGS:
372 case __OVS_KEY_ATTR_MAX:
379 odp_execute_sample(void *dp, struct dpif_packet *packet, bool steal,
380 const struct nlattr *action,
381 odp_execute_cb dp_execute_action)
383 const struct nlattr *subactions = NULL;
384 const struct nlattr *a;
387 NL_NESTED_FOR_EACH_UNSAFE (a, left, action) {
388 int type = nl_attr_type(a);
390 switch ((enum ovs_sample_attr) type) {
391 case OVS_SAMPLE_ATTR_PROBABILITY:
392 if (random_uint32() >= nl_attr_get_u32(a)) {
394 dpif_packet_delete(packet);
400 case OVS_SAMPLE_ATTR_ACTIONS:
404 case OVS_SAMPLE_ATTR_UNSPEC:
405 case __OVS_SAMPLE_ATTR_MAX:
411 odp_execute_actions(dp, &packet, 1, steal, nl_attr_get(subactions),
412 nl_attr_get_size(subactions), dp_execute_action);
416 odp_execute_actions(void *dp, struct dpif_packet **packets, int cnt, bool steal,
417 const struct nlattr *actions, size_t actions_len,
418 odp_execute_cb dp_execute_action)
420 const struct nlattr *a;
424 NL_ATTR_FOR_EACH_UNSAFE (a, left, actions, actions_len) {
425 int type = nl_attr_type(a);
426 bool last_action = (left <= NLA_ALIGN(a->nla_len));
428 switch ((enum ovs_action_attr) type) {
429 /* These only make sense in the context of a datapath. */
430 case OVS_ACTION_ATTR_OUTPUT:
431 case OVS_ACTION_ATTR_TUNNEL_PUSH:
432 case OVS_ACTION_ATTR_TUNNEL_POP:
433 case OVS_ACTION_ATTR_USERSPACE:
434 case OVS_ACTION_ATTR_RECIRC:
435 if (dp_execute_action) {
436 /* Allow 'dp_execute_action' to steal the packet data if we do
437 * not need it any more. */
438 bool may_steal = steal && last_action;
440 dp_execute_action(dp, packets, cnt, a, may_steal);
443 /* We do not need to free the packets. dp_execute_actions()
450 case OVS_ACTION_ATTR_HASH: {
451 const struct ovs_action_hash *hash_act = nl_attr_get(a);
453 /* Calculate a hash value directly. This might not match the
454 * value computed by the datapath, but it is much less expensive,
455 * and the current use case (bonding) does not require a strict
456 * match to work properly. */
457 if (hash_act->hash_alg == OVS_HASH_ALG_L4) {
461 for (i = 0; i < cnt; i++) {
462 flow_extract(&packets[i]->ofpbuf, &packets[i]->md, &flow);
463 hash = flow_hash_5tuple(&flow, hash_act->hash_basis);
465 /* We also store the hash value with each packet */
466 dpif_packet_set_dp_hash(packets[i], hash ? hash : 1);
469 /* Assert on unknown hash algorithm. */
475 case OVS_ACTION_ATTR_PUSH_VLAN: {
476 const struct ovs_action_push_vlan *vlan = nl_attr_get(a);
478 for (i = 0; i < cnt; i++) {
479 struct ofpbuf *buf = &packets[i]->ofpbuf;
481 eth_push_vlan(buf, htons(ETH_TYPE_VLAN), vlan->vlan_tci);
486 case OVS_ACTION_ATTR_POP_VLAN:
487 for (i = 0; i < cnt; i++) {
488 struct ofpbuf *buf = &packets[i]->ofpbuf;
494 case OVS_ACTION_ATTR_PUSH_MPLS: {
495 const struct ovs_action_push_mpls *mpls = nl_attr_get(a);
497 for (i = 0; i < cnt; i++) {
498 struct ofpbuf *buf = &packets[i]->ofpbuf;
500 push_mpls(buf, mpls->mpls_ethertype, mpls->mpls_lse);
505 case OVS_ACTION_ATTR_POP_MPLS:
506 for (i = 0; i < cnt; i++) {
507 struct ofpbuf *buf = &packets[i]->ofpbuf;
509 pop_mpls(buf, nl_attr_get_be16(a));
513 case OVS_ACTION_ATTR_SET:
514 for (i = 0; i < cnt; i++) {
515 odp_execute_set_action(packets[i], nl_attr_get(a));
519 case OVS_ACTION_ATTR_SET_MASKED:
520 for (i = 0; i < cnt; i++) {
521 odp_execute_masked_set_action(packets[i], nl_attr_get(a));
525 case OVS_ACTION_ATTR_SAMPLE:
526 for (i = 0; i < cnt; i++) {
527 odp_execute_sample(dp, packets[i], steal && last_action, a,
532 /* We do not need to free the packets. odp_execute_sample() has
538 case OVS_ACTION_ATTR_UNSPEC:
539 case __OVS_ACTION_ATTR_MAX:
545 for (i = 0; i < cnt; i++) {
546 dpif_packet_delete(packets[i]);