]> git.proxmox.com Git - mirror_ovs.git/blob - ofproto/in-band.c
ofproto: Use ofputil_uninit_group_mod().
[mirror_ovs.git] / ofproto / in-band.c
1 /*
2 * Copyright (c) 2008, 2009, 2010, 2011, 2012, 2013, 2014, 2015 Nicira, Inc.
3 *
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:
7 *
8 * http://www.apache.org/licenses/LICENSE-2.0
9 *
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.
15 */
16
17 #include <config.h>
18 #include <arpa/inet.h>
19 #include <errno.h>
20 #include <inttypes.h>
21 #include <sys/socket.h>
22 #include <net/if.h>
23 #include <string.h>
24 #include <stdlib.h>
25 #include "classifier.h"
26 #include "dhcp.h"
27 #include "flow.h"
28 #include "in-band.h"
29 #include "netdev.h"
30 #include "netlink.h"
31 #include "odp-util.h"
32 #include "ofproto.h"
33 #include "ofproto-provider.h"
34 #include "openflow/openflow.h"
35 #include "openvswitch/ofp-actions.h"
36 #include "openvswitch/ofpbuf.h"
37 #include "openvswitch/vlog.h"
38 #include "packets.h"
39 #include "poll-loop.h"
40 #include "timeval.h"
41
42 VLOG_DEFINE_THIS_MODULE(in_band);
43
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. */
49 enum {
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. */
54
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. */
58
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. */
62
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. */
66 };
67
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 struct eth_addr remote_mac; /* Next-hop MAC, all-zeros if unknown. */
72 struct eth_addr last_remote_mac; /* Previous nonzero next-hop MAC. */
73 struct netdev *remote_netdev; /* Device to send to next-hop MAC. */
74 };
75
76 /* What to do to an in_band_rule. */
77 enum in_band_op {
78 ADD, /* Add the rule to ofproto's flow table. */
79 DEL /* Delete the rule from ofproto's flow table. */
80 };
81
82 /* A rule to add to or delete from ofproto's flow table. */
83 struct in_band_rule {
84 struct hmap_node hmap_node; /* In struct in_band's "rules" hmap. */
85 struct match match;
86 int priority;
87 enum in_band_op op;
88 };
89
90 struct in_band {
91 struct ofproto *ofproto;
92 int queue_id;
93
94 /* Remote information. */
95 time_t next_remote_refresh; /* Refresh timer. */
96 struct in_band_remote *remotes;
97 size_t n_remotes;
98
99 /* Local information. */
100 time_t next_local_refresh; /* Refresh timer. */
101 struct eth_addr local_mac; /* Current MAC. */
102 struct netdev *local_netdev; /* Local port's network device. */
103
104 /* Flow tracking. */
105 struct hmap rules; /* Contains "struct in_band_rule"s. */
106 };
107
108 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(60, 60);
109
110 static int
111 refresh_remote(struct in_band *ib, struct in_band_remote *r)
112 {
113 struct in_addr next_hop_inaddr;
114 char *next_hop_dev;
115 int retval;
116
117 /* Find the next-hop IP address. */
118 r->remote_mac = eth_addr_zero;
119 retval = netdev_get_next_hop(ib->local_netdev, &r->remote_addr.sin_addr,
120 &next_hop_inaddr, &next_hop_dev);
121 if (retval) {
122 VLOG_WARN("%s: cannot find route for controller ("IP_FMT"): %s",
123 ib->ofproto->name, IP_ARGS(r->remote_addr.sin_addr.s_addr),
124 ovs_strerror(retval));
125 return 1;
126 }
127 if (!next_hop_inaddr.s_addr) {
128 next_hop_inaddr = r->remote_addr.sin_addr;
129 }
130
131 /* Open the next-hop network device. */
132 if (!r->remote_netdev
133 || strcmp(netdev_get_name(r->remote_netdev), next_hop_dev))
134 {
135 netdev_close(r->remote_netdev);
136
137 retval = netdev_open(next_hop_dev, NULL, &r->remote_netdev);
138 if (retval) {
139 VLOG_WARN_RL(&rl, "%s: cannot open netdev %s (next hop "
140 "to controller "IP_FMT"): %s",
141 ib->ofproto->name, next_hop_dev,
142 IP_ARGS(r->remote_addr.sin_addr.s_addr),
143 ovs_strerror(retval));
144 free(next_hop_dev);
145 return 1;
146 }
147 }
148 free(next_hop_dev);
149
150 /* Look up the MAC address of the next-hop IP address. */
151 retval = netdev_arp_lookup(r->remote_netdev, next_hop_inaddr.s_addr,
152 &r->remote_mac);
153 if (retval) {
154 VLOG_DBG_RL(&rl, "%s: cannot look up remote MAC address ("IP_FMT"): %s",
155 ib->ofproto->name, IP_ARGS(next_hop_inaddr.s_addr),
156 ovs_strerror(retval));
157 }
158
159 /* If we don't have a MAC address, then refresh quickly, since we probably
160 * will get a MAC address soon (via ARP). Otherwise, we can afford to wait
161 * a little while. */
162 return eth_addr_is_zero(r->remote_mac) ? 1 : 10;
163 }
164
165 static bool
166 refresh_remotes(struct in_band *ib)
167 {
168 struct in_band_remote *r;
169 bool any_changes;
170
171 if (time_now() < ib->next_remote_refresh) {
172 return false;
173 }
174
175 any_changes = false;
176 ib->next_remote_refresh = TIME_MAX;
177 for (r = ib->remotes; r < &ib->remotes[ib->n_remotes]; r++) {
178 struct eth_addr old_remote_mac;
179 time_t next_refresh;
180
181 /* Save old MAC. */
182 old_remote_mac = r->remote_mac;
183
184 /* Refresh remote information. */
185 next_refresh = refresh_remote(ib, r) + time_now();
186 ib->next_remote_refresh = MIN(ib->next_remote_refresh, next_refresh);
187
188 /* If the MAC changed, log the changes. */
189 if (!eth_addr_equals(r->remote_mac, old_remote_mac)) {
190 any_changes = true;
191 if (!eth_addr_is_zero(r->remote_mac)
192 && !eth_addr_equals(r->last_remote_mac, r->remote_mac)) {
193 VLOG_DBG("%s: remote MAC address changed from "ETH_ADDR_FMT
194 " to "ETH_ADDR_FMT,
195 ib->ofproto->name,
196 ETH_ADDR_ARGS(r->last_remote_mac),
197 ETH_ADDR_ARGS(r->remote_mac));
198 r->last_remote_mac = r->remote_mac;
199 }
200 }
201 }
202
203 return any_changes;
204 }
205
206 /* Refreshes the MAC address of the local port into ib->local_mac, if it is due
207 * for a refresh. Returns true if anything changed, otherwise false. */
208 static bool
209 refresh_local(struct in_band *ib)
210 {
211 struct eth_addr ea;
212 time_t now;
213
214 now = time_now();
215 if (now < ib->next_local_refresh) {
216 return false;
217 }
218 ib->next_local_refresh = now + 1;
219
220 if (netdev_get_etheraddr(ib->local_netdev, &ea)
221 || eth_addr_equals(ea, ib->local_mac)) {
222 return false;
223 }
224
225 ib->local_mac = ea;
226 return true;
227 }
228
229 /* Returns true if packets in 'flow' should be directed to the local port.
230 * (This keeps the flow table from preventing DHCP replies from being seen by
231 * the local port.) */
232 bool
233 in_band_must_output_to_local_port(const struct flow *flow)
234 {
235 return (flow->dl_type == htons(ETH_TYPE_IP)
236 && flow->nw_proto == IPPROTO_UDP
237 && flow->tp_src == htons(DHCP_SERVER_PORT)
238 && flow->tp_dst == htons(DHCP_CLIENT_PORT));
239 }
240
241 /* Returns the number of in-band rules currently installed in the flow
242 * table. */
243 int
244 in_band_count_rules(const struct in_band *ib)
245 {
246 return hmap_count(&ib->rules);
247 }
248
249 static void
250 add_rule(struct in_band *ib, const struct match *match, int priority)
251 {
252 uint32_t hash = match_hash(match, 0);
253 struct in_band_rule *rule;
254
255 HMAP_FOR_EACH_WITH_HASH (rule, hmap_node, hash, &ib->rules) {
256 if (match_equal(&rule->match, match)) {
257 rule->op = ADD;
258 return;
259 }
260 }
261
262 rule = xmalloc(sizeof *rule);
263 rule->match = *match;
264 rule->priority = priority;
265 rule->op = ADD;
266 hmap_insert(&ib->rules, &rule->hmap_node, hash);
267 }
268
269 static void
270 update_rules(struct in_band *ib)
271 {
272 struct in_band_rule *ib_rule;
273 struct in_band_remote *r;
274 struct match match;
275
276 /* Mark all the existing rules for deletion. (Afterward we will re-add any
277 * rules that are still valid.) */
278 HMAP_FOR_EACH (ib_rule, hmap_node, &ib->rules) {
279 ib_rule->op = DEL;
280 }
281
282 if (ib->n_remotes && !eth_addr_is_zero(ib->local_mac)) {
283 /* (a) Allow DHCP requests sent from the local port. */
284 match_init_catchall(&match);
285 match_set_in_port(&match, OFPP_LOCAL);
286 match_set_dl_type(&match, htons(ETH_TYPE_IP));
287 match_set_dl_src(&match, ib->local_mac);
288 match_set_nw_proto(&match, IPPROTO_UDP);
289 match_set_tp_src(&match, htons(DHCP_CLIENT_PORT));
290 match_set_tp_dst(&match, htons(DHCP_SERVER_PORT));
291 add_rule(ib, &match, IBR_FROM_LOCAL_DHCP);
292
293 /* (b) Allow ARP replies to the local port's MAC address. */
294 match_init_catchall(&match);
295 match_set_dl_type(&match, htons(ETH_TYPE_ARP));
296 match_set_dl_dst(&match, ib->local_mac);
297 match_set_nw_proto(&match, ARP_OP_REPLY);
298 add_rule(ib, &match, IBR_TO_LOCAL_ARP);
299
300 /* (c) Allow ARP requests from the local port's MAC address. */
301 match_init_catchall(&match);
302 match_set_dl_type(&match, htons(ETH_TYPE_ARP));
303 match_set_dl_src(&match, ib->local_mac);
304 match_set_nw_proto(&match, ARP_OP_REQUEST);
305 add_rule(ib, &match, IBR_FROM_LOCAL_ARP);
306 }
307
308 for (r = ib->remotes; r < &ib->remotes[ib->n_remotes]; r++) {
309 if (eth_addr_is_zero(r->remote_mac)) {
310 continue;
311 }
312
313 /* (d) Allow ARP replies to the next hop's MAC address. */
314 match_init_catchall(&match);
315 match_set_dl_type(&match, htons(ETH_TYPE_ARP));
316 match_set_dl_dst(&match, r->remote_mac);
317 match_set_nw_proto(&match, ARP_OP_REPLY);
318 add_rule(ib, &match, IBR_TO_NEXT_HOP_ARP);
319
320 /* (e) Allow ARP requests from the next hop's MAC address. */
321 match_init_catchall(&match);
322 match_set_dl_type(&match, htons(ETH_TYPE_ARP));
323 match_set_dl_src(&match, r->remote_mac);
324 match_set_nw_proto(&match, ARP_OP_REQUEST);
325 add_rule(ib, &match, IBR_FROM_NEXT_HOP_ARP);
326 }
327
328 for (r = ib->remotes; r < &ib->remotes[ib->n_remotes]; r++) {
329 const struct sockaddr_in *a = &r->remote_addr;
330
331 /* (f) Allow ARP replies containing the remote's IP address as a
332 * target. */
333 match_init_catchall(&match);
334 match_set_dl_type(&match, htons(ETH_TYPE_ARP));
335 match_set_nw_proto(&match, ARP_OP_REPLY);
336 match_set_nw_dst(&match, a->sin_addr.s_addr);
337 add_rule(ib, &match, IBR_TO_REMOTE_ARP);
338
339 /* (g) Allow ARP requests containing the remote's IP address as a
340 * source. */
341 match_init_catchall(&match);
342 match_set_dl_type(&match, htons(ETH_TYPE_ARP));
343 match_set_nw_proto(&match, ARP_OP_REQUEST);
344 match_set_nw_src(&match, a->sin_addr.s_addr);
345 add_rule(ib, &match, IBR_FROM_REMOTE_ARP);
346
347 /* (h) Allow TCP traffic to the remote's IP and port. */
348 match_init_catchall(&match);
349 match_set_dl_type(&match, htons(ETH_TYPE_IP));
350 match_set_nw_proto(&match, IPPROTO_TCP);
351 match_set_nw_dst(&match, a->sin_addr.s_addr);
352 match_set_tp_dst(&match, a->sin_port);
353 add_rule(ib, &match, IBR_TO_REMOTE_TCP);
354
355 /* (i) Allow TCP traffic from the remote's IP and port. */
356 match_init_catchall(&match);
357 match_set_dl_type(&match, htons(ETH_TYPE_IP));
358 match_set_nw_proto(&match, IPPROTO_TCP);
359 match_set_nw_src(&match, a->sin_addr.s_addr);
360 match_set_tp_src(&match, a->sin_port);
361 add_rule(ib, &match, IBR_FROM_REMOTE_TCP);
362 }
363 }
364
365 /* Updates the OpenFlow flow table for the current state of in-band control.
366 * Returns true ordinarily. Returns false if no remotes are configured on 'ib'
367 * and 'ib' doesn't have any rules left to remove from the OpenFlow flow
368 * table. Thus, a false return value means that the caller can destroy 'ib'
369 * without leaving extra flows hanging around in the flow table. */
370 bool
371 in_band_run(struct in_band *ib)
372 {
373 uint64_t ofpacts_stub[128 / 8];
374 struct ofpbuf ofpacts;
375
376 struct in_band_rule *rule, *next;
377
378 ofpbuf_use_stub(&ofpacts, ofpacts_stub, sizeof ofpacts_stub);
379
380 if (ib->queue_id >= 0) {
381 ofpact_put_SET_QUEUE(&ofpacts)->queue_id = ib->queue_id;
382 }
383 ofpact_put_OUTPUT(&ofpacts)->port = OFPP_NORMAL;
384
385 refresh_local(ib);
386 refresh_remotes(ib);
387
388 update_rules(ib);
389
390 HMAP_FOR_EACH_SAFE (rule, next, hmap_node, &ib->rules) {
391 switch (rule->op) {
392 case ADD:
393 ofproto_add_flow(ib->ofproto, &rule->match, rule->priority,
394 ofpacts.data, ofpacts.size);
395 break;
396
397 case DEL:
398 ofproto_delete_flow(ib->ofproto, &rule->match, rule->priority);
399 hmap_remove(&ib->rules, &rule->hmap_node);
400 free(rule);
401 break;
402 }
403 }
404
405 ofpbuf_uninit(&ofpacts);
406
407 return ib->n_remotes || !hmap_is_empty(&ib->rules);
408 }
409
410 void
411 in_band_wait(struct in_band *in_band)
412 {
413 long long int wakeup
414 = MIN(in_band->next_remote_refresh, in_band->next_local_refresh);
415 poll_timer_wait_until(wakeup * 1000);
416 }
417
418 int
419 in_band_create(struct ofproto *ofproto, const char *local_name,
420 struct in_band **in_bandp)
421 {
422 struct in_band *in_band;
423 struct netdev *local_netdev;
424 int error;
425 const char *type = ofproto_port_open_type(ofproto->type, "internal");
426
427 *in_bandp = NULL;
428 error = netdev_open(local_name, type, &local_netdev);
429 if (error) {
430 VLOG_ERR("%s: failed to initialize in-band control: cannot open "
431 "datapath local port %s (%s)", ofproto->name,
432 local_name, ovs_strerror(error));
433 return error;
434 }
435
436 in_band = xzalloc(sizeof *in_band);
437 in_band->ofproto = ofproto;
438 in_band->queue_id = -1;
439 in_band->next_remote_refresh = TIME_MIN;
440 in_band->next_local_refresh = TIME_MIN;
441 in_band->local_netdev = local_netdev;
442 hmap_init(&in_band->rules);
443
444 *in_bandp = in_band;
445
446 return 0;
447 }
448
449 void
450 in_band_destroy(struct in_band *ib)
451 {
452 if (ib) {
453 struct in_band_rule *rule;
454
455 HMAP_FOR_EACH_POP (rule, hmap_node, &ib->rules) {
456 free(rule);
457 }
458 hmap_destroy(&ib->rules);
459 in_band_set_remotes(ib, NULL, 0);
460 netdev_close(ib->local_netdev);
461 free(ib);
462 }
463 }
464
465 static bool
466 any_addresses_changed(struct in_band *ib,
467 const struct sockaddr_in *addresses, size_t n)
468 {
469 size_t i;
470
471 if (n != ib->n_remotes) {
472 return true;
473 }
474
475 for (i = 0; i < n; i++) {
476 const struct sockaddr_in *old = &ib->remotes[i].remote_addr;
477 const struct sockaddr_in *new = &addresses[i];
478
479 if (old->sin_addr.s_addr != new->sin_addr.s_addr ||
480 old->sin_port != new->sin_port) {
481 return true;
482 }
483 }
484
485 return false;
486 }
487
488 void
489 in_band_set_remotes(struct in_band *ib,
490 const struct sockaddr_in *addresses, size_t n)
491 {
492 size_t i;
493
494 if (!any_addresses_changed(ib, addresses, n)) {
495 return;
496 }
497
498 /* Clear old remotes. */
499 for (i = 0; i < ib->n_remotes; i++) {
500 netdev_close(ib->remotes[i].remote_netdev);
501 }
502 free(ib->remotes);
503
504 /* Set up new remotes. */
505 ib->remotes = n ? xzalloc(n * sizeof *ib->remotes) : NULL;
506 ib->n_remotes = n;
507 for (i = 0; i < n; i++) {
508 ib->remotes[i].remote_addr = addresses[i];
509 }
510
511 /* Force refresh in next call to in_band_run(). */
512 ib->next_remote_refresh = TIME_MIN;
513 }
514
515 /* Sets the OpenFlow queue used by flows set up by 'ib' to 'queue_id'. If
516 * 'queue_id' is negative, 'ib' will not set any queue (which is also the
517 * default). */
518 void
519 in_band_set_queue(struct in_band *ib, int queue_id)
520 {
521 ib->queue_id = queue_id;
522 }
523