2 * Copyright (c) 2008, 2009, 2010, 2011, 2012 Nicira, Inc.
4 * Licensed under the Apache License, Version 2.0 (the "License");
5 * you may not use this file except in compliance with the License.
6 * You may obtain a copy of the License at:
8 * http://www.apache.org/licenses/LICENSE-2.0
10 * Unless required by applicable law or agreed to in writing, software
11 * distributed under the License is distributed on an "AS IS" BASIS,
12 * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
13 * See the License for the specific language governing permissions and
14 * limitations under the License.
19 #include <arpa/inet.h>
22 #include <sys/socket.h>
26 #include "classifier.h"
32 #include "ofp-actions.h"
35 #include "ofproto-provider.h"
36 #include "openflow/openflow.h"
38 #include "poll-loop.h"
42 VLOG_DEFINE_THIS_MODULE(in_band);
44 /* Priorities used in classifier for in-band rules. These values are higher
45 * than any that may be set with OpenFlow, and "18" kind of looks like "IB".
46 * The ordering of priorities is not important because all of the rules set up
47 * by in-band control have the same action. The only reason to use more than
48 * one priority is to make the kind of flow easier to see during debugging. */
50 /* One set per bridge. */
51 IBR_FROM_LOCAL_DHCP = 180000, /* (a) From local port, DHCP. */
52 IBR_TO_LOCAL_ARP, /* (b) To local port, ARP. */
53 IBR_FROM_LOCAL_ARP, /* (c) From local port, ARP. */
55 /* One set per unique next-hop MAC. */
56 IBR_TO_NEXT_HOP_ARP, /* (d) To remote MAC, ARP. */
57 IBR_FROM_NEXT_HOP_ARP, /* (e) From remote MAC, ARP. */
59 /* One set per unique remote IP address. */
60 IBR_TO_REMOTE_ARP, /* (f) To remote IP, ARP. */
61 IBR_FROM_REMOTE_ARP, /* (g) From remote IP, ARP. */
63 /* One set per unique remote (IP,port) pair. */
64 IBR_TO_REMOTE_TCP, /* (h) To remote IP, TCP port. */
65 IBR_FROM_REMOTE_TCP /* (i) From remote IP, TCP port. */
68 /* Track one remote IP and next hop information. */
69 struct in_band_remote {
70 struct sockaddr_in remote_addr; /* IP address, in network byte order. */
71 uint8_t remote_mac[ETH_ADDR_LEN]; /* Next-hop MAC, all-zeros if unknown. */
72 uint8_t last_remote_mac[ETH_ADDR_LEN]; /* Previous nonzero next-hop MAC. */
73 struct netdev *remote_netdev; /* Device to send to next-hop MAC. */
76 /* What to do to an in_band_rule. */
78 ADD, /* Add the rule to ofproto's flow table. */
79 DELETE /* Delete the rule from ofproto's flow table. */
82 /* A rule to add to or delete from ofproto's flow table. */
84 struct hmap_node hmap_node; /* In struct in_band's "rules" hmap. */
86 unsigned int priority;
91 struct ofproto *ofproto;
94 /* Remote information. */
95 time_t next_remote_refresh; /* Refresh timer. */
96 struct in_band_remote *remotes;
99 /* Local information. */
100 time_t next_local_refresh; /* Refresh timer. */
101 uint8_t local_mac[ETH_ADDR_LEN]; /* Current MAC. */
102 struct netdev *local_netdev; /* Local port's network device. */
105 struct hmap rules; /* Contains "struct in_band_rule"s. */
108 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(60, 60);
111 refresh_remote(struct in_band *ib, struct in_band_remote *r)
113 struct in_addr next_hop_inaddr;
117 /* Find the next-hop IP address. */
118 memset(r->remote_mac, 0, sizeof r->remote_mac);
119 retval = netdev_get_next_hop(ib->local_netdev, &r->remote_addr.sin_addr,
120 &next_hop_inaddr, &next_hop_dev);
122 VLOG_WARN("cannot find route for controller ("IP_FMT"): %s",
123 IP_ARGS(r->remote_addr.sin_addr.s_addr), strerror(retval));
126 if (!next_hop_inaddr.s_addr) {
127 next_hop_inaddr = r->remote_addr.sin_addr;
130 /* Open the next-hop network device. */
131 if (!r->remote_netdev
132 || strcmp(netdev_get_name(r->remote_netdev), next_hop_dev))
134 netdev_close(r->remote_netdev);
136 retval = netdev_open(next_hop_dev, "system", &r->remote_netdev);
138 VLOG_WARN_RL(&rl, "cannot open netdev %s (next hop "
139 "to controller "IP_FMT"): %s",
140 next_hop_dev, IP_ARGS(r->remote_addr.sin_addr.s_addr),
148 /* Look up the MAC address of the next-hop IP address. */
149 retval = netdev_arp_lookup(r->remote_netdev, next_hop_inaddr.s_addr,
152 VLOG_DBG_RL(&rl, "cannot look up remote MAC address ("IP_FMT"): %s",
153 IP_ARGS(next_hop_inaddr.s_addr), strerror(retval));
156 /* If we don't have a MAC address, then refresh quickly, since we probably
157 * will get a MAC address soon (via ARP). Otherwise, we can afford to wait
159 return eth_addr_is_zero(r->remote_mac) ? 1 : 10;
163 refresh_remotes(struct in_band *ib)
165 struct in_band_remote *r;
168 if (time_now() < ib->next_remote_refresh) {
173 ib->next_remote_refresh = TIME_MAX;
174 for (r = ib->remotes; r < &ib->remotes[ib->n_remotes]; r++) {
175 uint8_t old_remote_mac[ETH_ADDR_LEN];
179 memcpy(old_remote_mac, r->remote_mac, ETH_ADDR_LEN);
181 /* Refresh remote information. */
182 next_refresh = refresh_remote(ib, r) + time_now();
183 ib->next_remote_refresh = MIN(ib->next_remote_refresh, next_refresh);
185 /* If the MAC changed, log the changes. */
186 if (!eth_addr_equals(r->remote_mac, old_remote_mac)) {
188 if (!eth_addr_is_zero(r->remote_mac)
189 && !eth_addr_equals(r->last_remote_mac, r->remote_mac)) {
190 VLOG_DBG("remote MAC address changed from "ETH_ADDR_FMT
192 ETH_ADDR_ARGS(r->last_remote_mac),
193 ETH_ADDR_ARGS(r->remote_mac));
194 memcpy(r->last_remote_mac, r->remote_mac, ETH_ADDR_LEN);
202 /* Refreshes the MAC address of the local port into ib->local_mac, if it is due
203 * for a refresh. Returns true if anything changed, otherwise false. */
205 refresh_local(struct in_band *ib)
207 uint8_t ea[ETH_ADDR_LEN];
211 if (now < ib->next_local_refresh) {
214 ib->next_local_refresh = now + 1;
216 if (netdev_get_etheraddr(ib->local_netdev, ea)
217 || eth_addr_equals(ea, ib->local_mac)) {
221 memcpy(ib->local_mac, ea, ETH_ADDR_LEN);
225 /* Returns true if 'packet' should be sent to the local port regardless
226 * of the flow table. */
228 in_band_msg_in_hook(struct in_band *in_band, const struct flow *flow,
229 const struct ofpbuf *packet)
231 /* Regardless of how the flow table is configured, we want to be
232 * able to see replies to our DHCP requests. */
233 if (flow->dl_type == htons(ETH_TYPE_IP)
234 && flow->nw_proto == IPPROTO_UDP
235 && flow->tp_src == htons(DHCP_SERVER_PORT)
236 && flow->tp_dst == htons(DHCP_CLIENT_PORT)
238 struct dhcp_header *dhcp;
240 dhcp = ofpbuf_at(packet, (char *)packet->l7 - (char *)packet->data,
246 refresh_local(in_band);
247 if (!eth_addr_is_zero(in_band->local_mac)
248 && eth_addr_equals(dhcp->chaddr, in_band->local_mac)) {
256 /* Returns true if the rule that would match 'flow' with 'actions' is
257 * allowed to be set up in the datapath. */
259 in_band_rule_check(const struct flow *flow, uint32_t local_odp_port,
260 const struct nlattr *actions, size_t actions_len)
262 /* Don't allow flows that would prevent DHCP replies from being seen
263 * by the local port. */
264 if (flow->dl_type == htons(ETH_TYPE_IP)
265 && flow->nw_proto == IPPROTO_UDP
266 && flow->tp_src == htons(DHCP_SERVER_PORT)
267 && flow->tp_dst == htons(DHCP_CLIENT_PORT)) {
268 const struct nlattr *a;
271 NL_ATTR_FOR_EACH_UNSAFE (a, left, actions, actions_len) {
272 if (nl_attr_type(a) == OVS_ACTION_ATTR_OUTPUT
273 && nl_attr_get_u32(a) == local_odp_port) {
284 add_rule(struct in_band *ib, const struct match *match, unsigned int priority)
286 uint32_t hash = match_hash(match, 0);
287 struct in_band_rule *rule;
289 HMAP_FOR_EACH_WITH_HASH (rule, hmap_node, hash, &ib->rules) {
290 if (match_equal(&rule->match, match)) {
296 rule = xmalloc(sizeof *rule);
297 rule->match = *match;
298 rule->priority = priority;
300 hmap_insert(&ib->rules, &rule->hmap_node, hash);
304 update_rules(struct in_band *ib)
306 struct in_band_rule *ib_rule;
307 struct in_band_remote *r;
310 /* Mark all the existing rules for deletion. (Afterward we will re-add any
311 * rules that are still valid.) */
312 HMAP_FOR_EACH (ib_rule, hmap_node, &ib->rules) {
313 ib_rule->op = DELETE;
316 if (ib->n_remotes && !eth_addr_is_zero(ib->local_mac)) {
317 /* (a) Allow DHCP requests sent from the local port. */
318 match_init_catchall(&match);
319 match_set_in_port(&match, OFPP_LOCAL);
320 match_set_dl_type(&match, htons(ETH_TYPE_IP));
321 match_set_dl_src(&match, ib->local_mac);
322 match_set_nw_proto(&match, IPPROTO_UDP);
323 match_set_tp_src(&match, htons(DHCP_CLIENT_PORT));
324 match_set_tp_dst(&match, htons(DHCP_SERVER_PORT));
325 add_rule(ib, &match, IBR_FROM_LOCAL_DHCP);
327 /* (b) Allow ARP replies to the local port's MAC address. */
328 match_init_catchall(&match);
329 match_set_dl_type(&match, htons(ETH_TYPE_ARP));
330 match_set_dl_dst(&match, ib->local_mac);
331 match_set_nw_proto(&match, ARP_OP_REPLY);
332 add_rule(ib, &match, IBR_TO_LOCAL_ARP);
334 /* (c) Allow ARP requests from the local port's MAC address. */
335 match_init_catchall(&match);
336 match_set_dl_type(&match, htons(ETH_TYPE_ARP));
337 match_set_dl_src(&match, ib->local_mac);
338 match_set_nw_proto(&match, ARP_OP_REQUEST);
339 add_rule(ib, &match, IBR_FROM_LOCAL_ARP);
342 for (r = ib->remotes; r < &ib->remotes[ib->n_remotes]; r++) {
343 const uint8_t *remote_mac = r->remote_mac;
345 if (eth_addr_is_zero(remote_mac)) {
349 /* (d) Allow ARP replies to the next hop's MAC address. */
350 match_init_catchall(&match);
351 match_set_dl_type(&match, htons(ETH_TYPE_ARP));
352 match_set_dl_dst(&match, remote_mac);
353 match_set_nw_proto(&match, ARP_OP_REPLY);
354 add_rule(ib, &match, IBR_TO_NEXT_HOP_ARP);
356 /* (e) Allow ARP requests from the next hop's MAC address. */
357 match_init_catchall(&match);
358 match_set_dl_type(&match, htons(ETH_TYPE_ARP));
359 match_set_dl_src(&match, remote_mac);
360 match_set_nw_proto(&match, ARP_OP_REQUEST);
361 add_rule(ib, &match, IBR_FROM_NEXT_HOP_ARP);
364 for (r = ib->remotes; r < &ib->remotes[ib->n_remotes]; r++) {
365 const struct sockaddr_in *a = &r->remote_addr;
367 /* (f) Allow ARP replies containing the remote's IP address as a
369 match_init_catchall(&match);
370 match_set_dl_type(&match, htons(ETH_TYPE_ARP));
371 match_set_nw_proto(&match, ARP_OP_REPLY);
372 match_set_nw_dst(&match, a->sin_addr.s_addr);
373 add_rule(ib, &match, IBR_TO_REMOTE_ARP);
375 /* (g) Allow ARP requests containing the remote's IP address as a
377 match_init_catchall(&match);
378 match_set_dl_type(&match, htons(ETH_TYPE_ARP));
379 match_set_nw_proto(&match, ARP_OP_REQUEST);
380 match_set_nw_src(&match, a->sin_addr.s_addr);
381 add_rule(ib, &match, IBR_FROM_REMOTE_ARP);
383 /* (h) Allow TCP traffic to the remote's IP and port. */
384 match_init_catchall(&match);
385 match_set_dl_type(&match, htons(ETH_TYPE_IP));
386 match_set_nw_proto(&match, IPPROTO_TCP);
387 match_set_nw_dst(&match, a->sin_addr.s_addr);
388 match_set_tp_dst(&match, a->sin_port);
389 add_rule(ib, &match, IBR_TO_REMOTE_TCP);
391 /* (i) Allow TCP traffic from the remote's IP and port. */
392 match_init_catchall(&match);
393 match_set_dl_type(&match, htons(ETH_TYPE_IP));
394 match_set_nw_proto(&match, IPPROTO_TCP);
395 match_set_nw_src(&match, a->sin_addr.s_addr);
396 match_set_tp_src(&match, a->sin_port);
397 add_rule(ib, &match, IBR_FROM_REMOTE_TCP);
401 /* Updates the OpenFlow flow table for the current state of in-band control.
402 * Returns true ordinarily. Returns false if no remotes are configured on 'ib'
403 * and 'ib' doesn't have any rules left to remove from the OpenFlow flow
404 * table. Thus, a false return value means that the caller can destroy 'ib'
405 * without leaving extra flows hanging around in the flow table. */
407 in_band_run(struct in_band *ib)
409 uint64_t ofpacts_stub[128 / 8];
410 struct ofpbuf ofpacts;
412 struct in_band_rule *rule, *next;
414 ofpbuf_use_stub(&ofpacts, ofpacts_stub, sizeof ofpacts_stub);
416 if (ib->queue_id >= 0) {
417 ofpact_put_SET_QUEUE(&ofpacts)->queue_id = ib->queue_id;
419 ofpact_put_OUTPUT(&ofpacts)->port = OFPP_NORMAL;
426 HMAP_FOR_EACH_SAFE (rule, next, hmap_node, &ib->rules) {
429 ofproto_add_flow(ib->ofproto, &rule->match, rule->priority,
430 ofpacts.data, ofpacts.size);
434 if (ofproto_delete_flow(ib->ofproto,
435 &rule->match, rule->priority)) {
436 /* ofproto doesn't have the rule anymore so there's no reason
437 * for us to track it any longer. */
438 hmap_remove(&ib->rules, &rule->hmap_node);
445 ofpbuf_uninit(&ofpacts);
447 return ib->n_remotes || !hmap_is_empty(&ib->rules);
451 in_band_wait(struct in_band *in_band)
454 = MIN(in_band->next_remote_refresh, in_band->next_local_refresh);
455 poll_timer_wait_until(wakeup * 1000);
459 in_band_create(struct ofproto *ofproto, const char *local_name,
460 struct in_band **in_bandp)
462 struct in_band *in_band;
463 struct netdev *local_netdev;
467 error = netdev_open(local_name, "system", &local_netdev);
469 VLOG_ERR("failed to initialize in-band control: cannot open "
470 "datapath local port %s (%s)", local_name, strerror(error));
474 in_band = xzalloc(sizeof *in_band);
475 in_band->ofproto = ofproto;
476 in_band->queue_id = -1;
477 in_band->next_remote_refresh = TIME_MIN;
478 in_band->next_local_refresh = TIME_MIN;
479 in_band->local_netdev = local_netdev;
480 hmap_init(&in_band->rules);
488 in_band_destroy(struct in_band *ib)
491 struct in_band_rule *rule, *next;
493 HMAP_FOR_EACH_SAFE (rule, next, hmap_node, &ib->rules) {
494 hmap_remove(&ib->rules, &rule->hmap_node);
497 hmap_destroy(&ib->rules);
498 in_band_set_remotes(ib, NULL, 0);
499 netdev_close(ib->local_netdev);
505 any_addresses_changed(struct in_band *ib,
506 const struct sockaddr_in *addresses, size_t n)
510 if (n != ib->n_remotes) {
514 for (i = 0; i < n; i++) {
515 const struct sockaddr_in *old = &ib->remotes[i].remote_addr;
516 const struct sockaddr_in *new = &addresses[i];
518 if (old->sin_addr.s_addr != new->sin_addr.s_addr ||
519 old->sin_port != new->sin_port) {
528 in_band_set_remotes(struct in_band *ib,
529 const struct sockaddr_in *addresses, size_t n)
533 if (!any_addresses_changed(ib, addresses, n)) {
537 /* Clear old remotes. */
538 for (i = 0; i < ib->n_remotes; i++) {
539 netdev_close(ib->remotes[i].remote_netdev);
543 /* Set up new remotes. */
544 ib->remotes = n ? xzalloc(n * sizeof *ib->remotes) : NULL;
546 for (i = 0; i < n; i++) {
547 ib->remotes[i].remote_addr = addresses[i];
550 /* Force refresh in next call to in_band_run(). */
551 ib->next_remote_refresh = TIME_MIN;
554 /* Sets the OpenFlow queue used by flows set up by 'ib' to 'queue_id'. If
555 * 'queue_id' is negative, 'ib' will not set any queue (which is also the
558 in_band_set_queue(struct in_band *ib, int queue_id)
560 ib->queue_id = queue_id;