]> git.proxmox.com Git - ovs.git/blob - ofproto/ofproto-dpif.c
dpif-netdev: Add clone action
[ovs.git] / ofproto / ofproto-dpif.c
1 /*
2 * Copyright (c) 2009, 2010, 2011, 2012, 2013, 2014, 2015, 2016 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 <errno.h>
19
20 #include "bfd.h"
21 #include "bond.h"
22 #include "bundle.h"
23 #include "byte-order.h"
24 #include "connectivity.h"
25 #include "connmgr.h"
26 #include "coverage.h"
27 #include "cfm.h"
28 #include "ct-dpif.h"
29 #include "fail-open.h"
30 #include "guarded-list.h"
31 #include "hmapx.h"
32 #include "lacp.h"
33 #include "learn.h"
34 #include "mac-learning.h"
35 #include "mcast-snooping.h"
36 #include "multipath.h"
37 #include "netdev-vport.h"
38 #include "netdev.h"
39 #include "netlink.h"
40 #include "nx-match.h"
41 #include "odp-util.h"
42 #include "odp-execute.h"
43 #include "ofproto/ofproto-dpif.h"
44 #include "ofproto/ofproto-provider.h"
45 #include "ofproto-dpif-ipfix.h"
46 #include "ofproto-dpif-mirror.h"
47 #include "ofproto-dpif-monitor.h"
48 #include "ofproto-dpif-rid.h"
49 #include "ofproto-dpif-sflow.h"
50 #include "ofproto-dpif-trace.h"
51 #include "ofproto-dpif-upcall.h"
52 #include "ofproto-dpif-xlate.h"
53 #include "ofproto-dpif-xlate-cache.h"
54 #include "openvswitch/ofp-actions.h"
55 #include "openvswitch/dynamic-string.h"
56 #include "openvswitch/meta-flow.h"
57 #include "openvswitch/ofp-parse.h"
58 #include "openvswitch/ofp-print.h"
59 #include "openvswitch/ofp-util.h"
60 #include "openvswitch/ofpbuf.h"
61 #include "openvswitch/vlog.h"
62 #include "ovs-lldp.h"
63 #include "ovs-rcu.h"
64 #include "ovs-router.h"
65 #include "poll-loop.h"
66 #include "seq.h"
67 #include "simap.h"
68 #include "smap.h"
69 #include "timer.h"
70 #include "tunnel.h"
71 #include "unaligned.h"
72 #include "unixctl.h"
73 #include "util.h"
74 #include "vlan-bitmap.h"
75
76 VLOG_DEFINE_THIS_MODULE(ofproto_dpif);
77
78 COVERAGE_DEFINE(ofproto_dpif_expired);
79 COVERAGE_DEFINE(packet_in_overflow);
80
81 struct flow_miss;
82
83 static void rule_get_stats(struct rule *, uint64_t *packets, uint64_t *bytes,
84 long long int *used);
85 static struct rule_dpif *rule_dpif_cast(const struct rule *);
86 static void rule_expire(struct rule_dpif *, long long now);
87
88 struct ofbundle {
89 struct hmap_node hmap_node; /* In struct ofproto's "bundles" hmap. */
90 struct ofproto_dpif *ofproto; /* Owning ofproto. */
91 void *aux; /* Key supplied by ofproto's client. */
92 char *name; /* Identifier for log messages. */
93
94 /* Configuration. */
95 struct ovs_list ports; /* Contains "struct ofport"s. */
96 enum port_vlan_mode vlan_mode; /* VLAN mode */
97 int vlan; /* -1=trunk port, else a 12-bit VLAN ID. */
98 unsigned long *trunks; /* Bitmap of trunked VLANs, if 'vlan' == -1.
99 * NULL if all VLANs are trunked. */
100 struct lacp *lacp; /* LACP if LACP is enabled, otherwise NULL. */
101 struct bond *bond; /* Nonnull iff more than one port. */
102 bool use_priority_tags; /* Use 802.1p tag for frames in VLAN 0? */
103
104 bool protected; /* Protected port mode */
105
106 /* Status. */
107 bool floodable; /* True if no port has OFPUTIL_PC_NO_FLOOD set. */
108 };
109
110 static void bundle_remove(struct ofport *);
111 static void bundle_update(struct ofbundle *);
112 static void bundle_destroy(struct ofbundle *);
113 static void bundle_del_port(struct ofport_dpif *);
114 static void bundle_run(struct ofbundle *);
115 static void bundle_wait(struct ofbundle *);
116 static void bundle_flush_macs(struct ofbundle *, bool);
117 static void bundle_move(struct ofbundle *, struct ofbundle *);
118
119 static void stp_run(struct ofproto_dpif *ofproto);
120 static void stp_wait(struct ofproto_dpif *ofproto);
121 static int set_stp_port(struct ofport *,
122 const struct ofproto_port_stp_settings *);
123
124 static void rstp_run(struct ofproto_dpif *ofproto);
125 static void set_rstp_port(struct ofport *,
126 const struct ofproto_port_rstp_settings *);
127
128 struct ofport_dpif {
129 struct hmap_node odp_port_node; /* In dpif_backer's "odp_to_ofport_map". */
130 struct ofport up;
131
132 odp_port_t odp_port;
133 struct ofbundle *bundle; /* Bundle that contains this port, if any. */
134 struct ovs_list bundle_node;/* In struct ofbundle's "ports" list. */
135 struct cfm *cfm; /* Connectivity Fault Management, if any. */
136 struct bfd *bfd; /* BFD, if any. */
137 struct lldp *lldp; /* lldp, if any. */
138 bool may_enable; /* May be enabled in bonds. */
139 bool is_tunnel; /* This port is a tunnel. */
140 bool is_layer3; /* This is a layer 3 port. */
141 long long int carrier_seq; /* Carrier status changes. */
142 struct ofport_dpif *peer; /* Peer if patch port. */
143
144 /* Spanning tree. */
145 struct stp_port *stp_port; /* Spanning Tree Protocol, if any. */
146 enum stp_state stp_state; /* Always STP_DISABLED if STP not in use. */
147 long long int stp_state_entered;
148
149 /* Rapid Spanning Tree. */
150 struct rstp_port *rstp_port; /* Rapid Spanning Tree Protocol, if any. */
151 enum rstp_state rstp_state; /* Always RSTP_DISABLED if RSTP not in use. */
152
153 /* Queue to DSCP mapping. */
154 struct ofproto_port_queue *qdscp;
155 size_t n_qdscp;
156 };
157
158 static odp_port_t ofp_port_to_odp_port(const struct ofproto_dpif *,
159 ofp_port_t);
160
161 static ofp_port_t odp_port_to_ofp_port(const struct ofproto_dpif *,
162 odp_port_t);
163
164 static struct ofport_dpif *
165 ofport_dpif_cast(const struct ofport *ofport)
166 {
167 return ofport ? CONTAINER_OF(ofport, struct ofport_dpif, up) : NULL;
168 }
169
170 static void port_run(struct ofport_dpif *);
171 static int set_bfd(struct ofport *, const struct smap *);
172 static int set_cfm(struct ofport *, const struct cfm_settings *);
173 static int set_lldp(struct ofport *ofport_, const struct smap *cfg);
174 static void ofport_update_peer(struct ofport_dpif *);
175
176 COVERAGE_DEFINE(rev_reconfigure);
177 COVERAGE_DEFINE(rev_stp);
178 COVERAGE_DEFINE(rev_rstp);
179 COVERAGE_DEFINE(rev_bond);
180 COVERAGE_DEFINE(rev_port_toggled);
181 COVERAGE_DEFINE(rev_flow_table);
182 COVERAGE_DEFINE(rev_mac_learning);
183 COVERAGE_DEFINE(rev_mcast_snooping);
184
185 /* All existing ofproto_backer instances, indexed by ofproto->up.type. */
186 struct shash all_dpif_backers = SHASH_INITIALIZER(&all_dpif_backers);
187
188 /* All existing ofproto_dpif instances, indexed by ->up.name. */
189 struct hmap all_ofproto_dpifs = HMAP_INITIALIZER(&all_ofproto_dpifs);
190
191 static bool ofproto_use_tnl_push_pop = true;
192 static void ofproto_unixctl_init(void);
193
194 static inline struct ofproto_dpif *
195 ofproto_dpif_cast(const struct ofproto *ofproto)
196 {
197 ovs_assert(ofproto->ofproto_class == &ofproto_dpif_class);
198 return CONTAINER_OF(ofproto, struct ofproto_dpif, up);
199 }
200
201 /* Global variables. */
202 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
203
204 /* Initial mappings of port to bridge mappings. */
205 static struct shash init_ofp_ports = SHASH_INITIALIZER(&init_ofp_ports);
206
207 /* Initialize 'ofm' for a learn action. If the rule already existed, reference
208 * to that rule is taken, otherwise a new rule is created. 'ofm' keeps the
209 * rule reference in both cases. */
210 enum ofperr
211 ofproto_dpif_flow_mod_init_for_learn(struct ofproto_dpif *ofproto,
212 const struct ofputil_flow_mod *fm,
213 struct ofproto_flow_mod *ofm)
214 {
215 /* This will not take the global 'ofproto_mutex'. */
216 return ofproto_flow_mod_init_for_learn(&ofproto->up, fm, ofm);
217 }
218
219 /* Appends 'am' to the queue of asynchronous messages to be sent to the
220 * controller. Takes ownership of 'am' and any data it points to. */
221 void
222 ofproto_dpif_send_async_msg(struct ofproto_dpif *ofproto,
223 struct ofproto_async_msg *am)
224 {
225 if (!guarded_list_push_back(&ofproto->ams, &am->list_node, 1024)) {
226 COVERAGE_INC(packet_in_overflow);
227 ofproto_async_msg_free(am);
228 }
229
230 /* Wakes up main thread for packet-in I/O. */
231 seq_change(ofproto->ams_seq);
232 }
233 \f
234 /* Factory functions. */
235
236 static void
237 init(const struct shash *iface_hints)
238 {
239 struct shash_node *node;
240
241 /* Make a local copy, since we don't own 'iface_hints' elements. */
242 SHASH_FOR_EACH(node, iface_hints) {
243 const struct iface_hint *orig_hint = node->data;
244 struct iface_hint *new_hint = xmalloc(sizeof *new_hint);
245
246 new_hint->br_name = xstrdup(orig_hint->br_name);
247 new_hint->br_type = xstrdup(orig_hint->br_type);
248 new_hint->ofp_port = orig_hint->ofp_port;
249
250 shash_add(&init_ofp_ports, node->name, new_hint);
251 }
252
253 ofproto_unixctl_init();
254 ofproto_dpif_trace_init();
255 udpif_init();
256 }
257
258 static void
259 enumerate_types(struct sset *types)
260 {
261 dp_enumerate_types(types);
262 }
263
264 static int
265 enumerate_names(const char *type, struct sset *names)
266 {
267 struct ofproto_dpif *ofproto;
268
269 sset_clear(names);
270 HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
271 if (strcmp(type, ofproto->up.type)) {
272 continue;
273 }
274 sset_add(names, ofproto->up.name);
275 }
276
277 return 0;
278 }
279
280 static int
281 del(const char *type, const char *name)
282 {
283 struct dpif *dpif;
284 int error;
285
286 error = dpif_open(name, type, &dpif);
287 if (!error) {
288 error = dpif_delete(dpif);
289 dpif_close(dpif);
290 }
291 return error;
292 }
293 \f
294 static const char *
295 port_open_type(const char *datapath_type, const char *port_type)
296 {
297 return dpif_port_open_type(datapath_type, port_type);
298 }
299
300 /* Type functions. */
301
302 static void process_dpif_port_changes(struct dpif_backer *);
303 static void process_dpif_all_ports_changed(struct dpif_backer *);
304 static void process_dpif_port_change(struct dpif_backer *,
305 const char *devname);
306 static void process_dpif_port_error(struct dpif_backer *, int error);
307
308 static struct ofproto_dpif *
309 lookup_ofproto_dpif_by_port_name(const char *name)
310 {
311 struct ofproto_dpif *ofproto;
312
313 HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
314 if (sset_contains(&ofproto->ports, name)) {
315 return ofproto;
316 }
317 }
318
319 return NULL;
320 }
321
322 static int
323 type_run(const char *type)
324 {
325 struct dpif_backer *backer;
326
327 backer = shash_find_data(&all_dpif_backers, type);
328 if (!backer) {
329 /* This is not necessarily a problem, since backers are only
330 * created on demand. */
331 return 0;
332 }
333
334 /* This must be called before dpif_run() */
335 dpif_poll_threads_set(backer->dpif, pmd_cpu_mask);
336
337 if (dpif_run(backer->dpif)) {
338 backer->need_revalidate = REV_RECONFIGURE;
339 }
340
341 udpif_run(backer->udpif);
342
343 /* If vswitchd started with other_config:flow_restore_wait set as "true",
344 * and the configuration has now changed to "false", enable receiving
345 * packets from the datapath. */
346 if (!backer->recv_set_enable && !ofproto_get_flow_restore_wait()) {
347 int error;
348
349 backer->recv_set_enable = true;
350
351 error = dpif_recv_set(backer->dpif, backer->recv_set_enable);
352 if (error) {
353 VLOG_ERR("Failed to enable receiving packets in dpif.");
354 return error;
355 }
356 dpif_flow_flush(backer->dpif);
357 backer->need_revalidate = REV_RECONFIGURE;
358 }
359
360 if (backer->recv_set_enable) {
361 udpif_set_threads(backer->udpif, n_handlers, n_revalidators);
362 }
363
364 if (backer->need_revalidate) {
365 struct ofproto_dpif *ofproto;
366 struct simap_node *node;
367 struct simap tmp_backers;
368
369 /* Handle tunnel garbage collection. */
370 simap_init(&tmp_backers);
371 simap_swap(&backer->tnl_backers, &tmp_backers);
372
373 HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
374 struct ofport_dpif *iter;
375
376 if (backer != ofproto->backer) {
377 continue;
378 }
379
380 HMAP_FOR_EACH (iter, up.hmap_node, &ofproto->up.ports) {
381 char namebuf[NETDEV_VPORT_NAME_BUFSIZE];
382 const char *dp_port;
383
384 if (!iter->is_tunnel) {
385 continue;
386 }
387
388 dp_port = netdev_vport_get_dpif_port(iter->up.netdev,
389 namebuf, sizeof namebuf);
390 node = simap_find(&tmp_backers, dp_port);
391 if (node) {
392 simap_put(&backer->tnl_backers, dp_port, node->data);
393 simap_delete(&tmp_backers, node);
394 node = simap_find(&backer->tnl_backers, dp_port);
395 } else {
396 node = simap_find(&backer->tnl_backers, dp_port);
397 if (!node) {
398 odp_port_t odp_port = ODPP_NONE;
399
400 if (!dpif_port_add(backer->dpif, iter->up.netdev,
401 &odp_port)) {
402 simap_put(&backer->tnl_backers, dp_port,
403 odp_to_u32(odp_port));
404 node = simap_find(&backer->tnl_backers, dp_port);
405 }
406 }
407 }
408
409 iter->odp_port = node ? u32_to_odp(node->data) : ODPP_NONE;
410 if (tnl_port_reconfigure(iter, iter->up.netdev,
411 iter->odp_port,
412 ovs_native_tunneling_is_on(ofproto), dp_port)) {
413 backer->need_revalidate = REV_RECONFIGURE;
414 }
415 }
416 }
417
418 SIMAP_FOR_EACH (node, &tmp_backers) {
419 dpif_port_del(backer->dpif, u32_to_odp(node->data));
420 }
421 simap_destroy(&tmp_backers);
422
423 switch (backer->need_revalidate) {
424 case REV_RECONFIGURE: COVERAGE_INC(rev_reconfigure); break;
425 case REV_STP: COVERAGE_INC(rev_stp); break;
426 case REV_RSTP: COVERAGE_INC(rev_rstp); break;
427 case REV_BOND: COVERAGE_INC(rev_bond); break;
428 case REV_PORT_TOGGLED: COVERAGE_INC(rev_port_toggled); break;
429 case REV_FLOW_TABLE: COVERAGE_INC(rev_flow_table); break;
430 case REV_MAC_LEARNING: COVERAGE_INC(rev_mac_learning); break;
431 case REV_MCAST_SNOOPING: COVERAGE_INC(rev_mcast_snooping); break;
432 }
433 backer->need_revalidate = 0;
434
435 HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
436 struct ofport_dpif *ofport;
437 struct ofbundle *bundle;
438
439 if (ofproto->backer != backer) {
440 continue;
441 }
442
443 xlate_txn_start();
444 xlate_ofproto_set(ofproto, ofproto->up.name,
445 ofproto->backer->dpif, ofproto->ml,
446 ofproto->stp, ofproto->rstp, ofproto->ms,
447 ofproto->mbridge, ofproto->sflow, ofproto->ipfix,
448 ofproto->netflow,
449 ofproto->up.forward_bpdu,
450 connmgr_has_in_band(ofproto->up.connmgr),
451 &ofproto->backer->support);
452
453 HMAP_FOR_EACH (bundle, hmap_node, &ofproto->bundles) {
454 xlate_bundle_set(ofproto, bundle, bundle->name,
455 bundle->vlan_mode, bundle->vlan,
456 bundle->trunks, bundle->use_priority_tags,
457 bundle->bond, bundle->lacp,
458 bundle->floodable, bundle->protected);
459 }
460
461 HMAP_FOR_EACH (ofport, up.hmap_node, &ofproto->up.ports) {
462 int stp_port = ofport->stp_port
463 ? stp_port_no(ofport->stp_port)
464 : -1;
465 xlate_ofport_set(ofproto, ofport->bundle, ofport,
466 ofport->up.ofp_port, ofport->odp_port,
467 ofport->up.netdev, ofport->cfm, ofport->bfd,
468 ofport->lldp, ofport->peer, stp_port,
469 ofport->rstp_port, ofport->qdscp,
470 ofport->n_qdscp, ofport->up.pp.config,
471 ofport->up.pp.state, ofport->is_tunnel,
472 ofport->may_enable);
473 }
474 xlate_txn_commit();
475 }
476
477 udpif_revalidate(backer->udpif);
478 }
479
480 process_dpif_port_changes(backer);
481
482 return 0;
483 }
484
485 /* Check for and handle port changes in 'backer''s dpif. */
486 static void
487 process_dpif_port_changes(struct dpif_backer *backer)
488 {
489 for (;;) {
490 char *devname;
491 int error;
492
493 error = dpif_port_poll(backer->dpif, &devname);
494 switch (error) {
495 case EAGAIN:
496 return;
497
498 case ENOBUFS:
499 process_dpif_all_ports_changed(backer);
500 break;
501
502 case 0:
503 process_dpif_port_change(backer, devname);
504 free(devname);
505 break;
506
507 default:
508 process_dpif_port_error(backer, error);
509 break;
510 }
511 }
512 }
513
514 static void
515 process_dpif_all_ports_changed(struct dpif_backer *backer)
516 {
517 struct ofproto_dpif *ofproto;
518 struct dpif_port dpif_port;
519 struct dpif_port_dump dump;
520 struct sset devnames;
521 const char *devname;
522
523 sset_init(&devnames);
524 HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
525 if (ofproto->backer == backer) {
526 struct ofport *ofport;
527
528 HMAP_FOR_EACH (ofport, hmap_node, &ofproto->up.ports) {
529 sset_add(&devnames, netdev_get_name(ofport->netdev));
530 }
531 }
532 }
533 DPIF_PORT_FOR_EACH (&dpif_port, &dump, backer->dpif) {
534 sset_add(&devnames, dpif_port.name);
535 }
536
537 SSET_FOR_EACH (devname, &devnames) {
538 process_dpif_port_change(backer, devname);
539 }
540 sset_destroy(&devnames);
541 }
542
543 static void
544 process_dpif_port_change(struct dpif_backer *backer, const char *devname)
545 {
546 struct ofproto_dpif *ofproto;
547 struct dpif_port port;
548
549 /* Don't report on the datapath's device. */
550 if (!strcmp(devname, dpif_base_name(backer->dpif))) {
551 return;
552 }
553
554 HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node,
555 &all_ofproto_dpifs) {
556 if (simap_contains(&ofproto->backer->tnl_backers, devname)) {
557 return;
558 }
559 }
560
561 ofproto = lookup_ofproto_dpif_by_port_name(devname);
562 if (dpif_port_query_by_name(backer->dpif, devname, &port)) {
563 /* The port was removed. If we know the datapath,
564 * report it through poll_set(). If we don't, it may be
565 * notifying us of a removal we initiated, so ignore it.
566 * If there's a pending ENOBUFS, let it stand, since
567 * everything will be reevaluated. */
568 if (ofproto && ofproto->port_poll_errno != ENOBUFS) {
569 sset_add(&ofproto->port_poll_set, devname);
570 ofproto->port_poll_errno = 0;
571 }
572 } else if (!ofproto) {
573 /* The port was added, but we don't know with which
574 * ofproto we should associate it. Delete it. */
575 dpif_port_del(backer->dpif, port.port_no);
576 } else {
577 struct ofport_dpif *ofport;
578
579 ofport = ofport_dpif_cast(shash_find_data(
580 &ofproto->up.port_by_name, devname));
581 if (ofport
582 && ofport->odp_port != port.port_no
583 && !odp_port_to_ofport(backer, port.port_no))
584 {
585 /* 'ofport''s datapath port number has changed from
586 * 'ofport->odp_port' to 'port.port_no'. Update our internal data
587 * structures to match. */
588 ovs_rwlock_wrlock(&backer->odp_to_ofport_lock);
589 hmap_remove(&backer->odp_to_ofport_map, &ofport->odp_port_node);
590 ofport->odp_port = port.port_no;
591 hmap_insert(&backer->odp_to_ofport_map, &ofport->odp_port_node,
592 hash_odp_port(port.port_no));
593 ovs_rwlock_unlock(&backer->odp_to_ofport_lock);
594 backer->need_revalidate = REV_RECONFIGURE;
595 }
596 }
597 dpif_port_destroy(&port);
598 }
599
600 /* Propagate 'error' to all ofprotos based on 'backer'. */
601 static void
602 process_dpif_port_error(struct dpif_backer *backer, int error)
603 {
604 struct ofproto_dpif *ofproto;
605
606 HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
607 if (ofproto->backer == backer) {
608 sset_clear(&ofproto->port_poll_set);
609 ofproto->port_poll_errno = error;
610 }
611 }
612 }
613
614 static void
615 type_wait(const char *type)
616 {
617 struct dpif_backer *backer;
618
619 backer = shash_find_data(&all_dpif_backers, type);
620 if (!backer) {
621 /* This is not necessarily a problem, since backers are only
622 * created on demand. */
623 return;
624 }
625
626 dpif_wait(backer->dpif);
627 }
628 \f
629 /* Basic life-cycle. */
630
631 static int add_internal_flows(struct ofproto_dpif *);
632
633 static struct ofproto *
634 alloc(void)
635 {
636 struct ofproto_dpif *ofproto = xzalloc(sizeof *ofproto);
637 return &ofproto->up;
638 }
639
640 static void
641 dealloc(struct ofproto *ofproto_)
642 {
643 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
644 free(ofproto);
645 }
646
647 static void
648 close_dpif_backer(struct dpif_backer *backer)
649 {
650 ovs_assert(backer->refcount > 0);
651
652 if (--backer->refcount) {
653 return;
654 }
655
656 udpif_destroy(backer->udpif);
657
658 simap_destroy(&backer->tnl_backers);
659 ovs_rwlock_destroy(&backer->odp_to_ofport_lock);
660 hmap_destroy(&backer->odp_to_ofport_map);
661 shash_find_and_delete(&all_dpif_backers, backer->type);
662 free(backer->type);
663 free(backer->dp_version_string);
664 dpif_close(backer->dpif);
665 free(backer);
666 }
667
668 /* Datapath port slated for removal from datapath. */
669 struct odp_garbage {
670 struct ovs_list list_node;
671 odp_port_t odp_port;
672 };
673
674 static bool check_variable_length_userdata(struct dpif_backer *backer);
675 static void check_support(struct dpif_backer *backer);
676
677 static int
678 open_dpif_backer(const char *type, struct dpif_backer **backerp)
679 {
680 struct dpif_backer *backer;
681 struct dpif_port_dump port_dump;
682 struct dpif_port port;
683 struct shash_node *node;
684 struct ovs_list garbage_list;
685 struct odp_garbage *garbage;
686
687 struct sset names;
688 char *backer_name;
689 const char *name;
690 int error;
691
692 backer = shash_find_data(&all_dpif_backers, type);
693 if (backer) {
694 backer->refcount++;
695 *backerp = backer;
696 return 0;
697 }
698
699 backer_name = xasprintf("ovs-%s", type);
700
701 /* Remove any existing datapaths, since we assume we're the only
702 * userspace controlling the datapath. */
703 sset_init(&names);
704 dp_enumerate_names(type, &names);
705 SSET_FOR_EACH(name, &names) {
706 struct dpif *old_dpif;
707
708 /* Don't remove our backer if it exists. */
709 if (!strcmp(name, backer_name)) {
710 continue;
711 }
712
713 if (dpif_open(name, type, &old_dpif)) {
714 VLOG_WARN("couldn't open old datapath %s to remove it", name);
715 } else {
716 dpif_delete(old_dpif);
717 dpif_close(old_dpif);
718 }
719 }
720 sset_destroy(&names);
721
722 backer = xmalloc(sizeof *backer);
723
724 error = dpif_create_and_open(backer_name, type, &backer->dpif);
725 free(backer_name);
726 if (error) {
727 VLOG_ERR("failed to open datapath of type %s: %s", type,
728 ovs_strerror(error));
729 free(backer);
730 return error;
731 }
732 backer->udpif = udpif_create(backer, backer->dpif);
733
734 backer->type = xstrdup(type);
735 backer->refcount = 1;
736 hmap_init(&backer->odp_to_ofport_map);
737 ovs_rwlock_init(&backer->odp_to_ofport_lock);
738 backer->need_revalidate = 0;
739 simap_init(&backer->tnl_backers);
740 backer->recv_set_enable = !ofproto_get_flow_restore_wait();
741 *backerp = backer;
742
743 if (backer->recv_set_enable) {
744 dpif_flow_flush(backer->dpif);
745 }
746
747 /* Loop through the ports already on the datapath and remove any
748 * that we don't need anymore. */
749 ovs_list_init(&garbage_list);
750 dpif_port_dump_start(&port_dump, backer->dpif);
751 while (dpif_port_dump_next(&port_dump, &port)) {
752 node = shash_find(&init_ofp_ports, port.name);
753 if (!node && strcmp(port.name, dpif_base_name(backer->dpif))) {
754 garbage = xmalloc(sizeof *garbage);
755 garbage->odp_port = port.port_no;
756 ovs_list_push_front(&garbage_list, &garbage->list_node);
757 }
758 }
759 dpif_port_dump_done(&port_dump);
760
761 LIST_FOR_EACH_POP (garbage, list_node, &garbage_list) {
762 dpif_port_del(backer->dpif, garbage->odp_port);
763 free(garbage);
764 }
765
766 shash_add(&all_dpif_backers, type, backer);
767
768 check_support(backer);
769 atomic_count_init(&backer->tnl_count, 0);
770
771 error = dpif_recv_set(backer->dpif, backer->recv_set_enable);
772 if (error) {
773 VLOG_ERR("failed to listen on datapath of type %s: %s",
774 type, ovs_strerror(error));
775 close_dpif_backer(backer);
776 return error;
777 }
778
779 if (backer->recv_set_enable) {
780 udpif_set_threads(backer->udpif, n_handlers, n_revalidators);
781 }
782
783 /* This check fails if performed before udpif threads have been set,
784 * as the kernel module checks that the 'pid' in userspace action
785 * is non-zero. */
786 backer->support.variable_length_userdata
787 = check_variable_length_userdata(backer);
788 backer->dp_version_string = dpif_get_dp_version(backer->dpif);
789
790 return error;
791 }
792
793 bool
794 ovs_native_tunneling_is_on(struct ofproto_dpif *ofproto)
795 {
796 return ofproto_use_tnl_push_pop && ofproto->backer->support.tnl_push_pop &&
797 atomic_count_get(&ofproto->backer->tnl_count);
798 }
799
800 /* Tests whether 'backer''s datapath supports recirculation. Only newer
801 * datapaths support OVS_KEY_ATTR_RECIRC_ID in keys. We need to disable some
802 * features on older datapaths that don't support this feature.
803 *
804 * Returns false if 'backer' definitely does not support recirculation, true if
805 * it seems to support recirculation or if at least the error we get is
806 * ambiguous. */
807 static bool
808 check_recirc(struct dpif_backer *backer)
809 {
810 struct flow flow;
811 struct odputil_keybuf keybuf;
812 struct ofpbuf key;
813 bool enable_recirc;
814 struct odp_flow_key_parms odp_parms = {
815 .flow = &flow,
816 .support = {
817 .recirc = true,
818 },
819 };
820
821 memset(&flow, 0, sizeof flow);
822 flow.recirc_id = 1;
823 flow.dp_hash = 1;
824
825 ofpbuf_use_stack(&key, &keybuf, sizeof keybuf);
826 odp_flow_key_from_flow(&odp_parms, &key);
827 enable_recirc = dpif_probe_feature(backer->dpif, "recirculation", &key,
828 NULL);
829
830 if (enable_recirc) {
831 VLOG_INFO("%s: Datapath supports recirculation",
832 dpif_name(backer->dpif));
833 } else {
834 VLOG_INFO("%s: Datapath does not support recirculation",
835 dpif_name(backer->dpif));
836 }
837
838 return enable_recirc;
839 }
840
841 /* Tests whether 'dpif' supports unique flow ids. We can skip serializing
842 * some flow attributes for datapaths that support this feature.
843 *
844 * Returns true if 'dpif' supports UFID for flow operations.
845 * Returns false if 'dpif' does not support UFID. */
846 static bool
847 check_ufid(struct dpif_backer *backer)
848 {
849 struct flow flow;
850 struct odputil_keybuf keybuf;
851 struct ofpbuf key;
852 ovs_u128 ufid;
853 bool enable_ufid;
854 struct odp_flow_key_parms odp_parms = {
855 .flow = &flow,
856 };
857
858 memset(&flow, 0, sizeof flow);
859 flow.dl_type = htons(0x1234);
860
861 ofpbuf_use_stack(&key, &keybuf, sizeof keybuf);
862 odp_flow_key_from_flow(&odp_parms, &key);
863 dpif_flow_hash(backer->dpif, key.data, key.size, &ufid);
864
865 enable_ufid = dpif_probe_feature(backer->dpif, "UFID", &key, &ufid);
866
867 if (enable_ufid) {
868 VLOG_INFO("%s: Datapath supports unique flow ids",
869 dpif_name(backer->dpif));
870 } else {
871 VLOG_INFO("%s: Datapath does not support unique flow ids",
872 dpif_name(backer->dpif));
873 }
874 return enable_ufid;
875 }
876
877 /* Tests whether 'backer''s datapath supports variable-length
878 * OVS_USERSPACE_ATTR_USERDATA in OVS_ACTION_ATTR_USERSPACE actions. We need
879 * to disable some features on older datapaths that don't support this
880 * feature.
881 *
882 * Returns false if 'backer' definitely does not support variable-length
883 * userdata, true if it seems to support them or if at least the error we get
884 * is ambiguous. */
885 static bool
886 check_variable_length_userdata(struct dpif_backer *backer)
887 {
888 struct eth_header *eth;
889 struct ofpbuf actions;
890 struct dpif_execute execute;
891 struct dp_packet packet;
892 struct flow flow;
893 size_t start;
894 int error;
895
896 /* Compose a userspace action that will cause an ERANGE error on older
897 * datapaths that don't support variable-length userdata.
898 *
899 * We really test for using userdata longer than 8 bytes, but older
900 * datapaths accepted these, silently truncating the userdata to 8 bytes.
901 * The same older datapaths rejected userdata shorter than 8 bytes, so we
902 * test for that instead as a proxy for longer userdata support. */
903 ofpbuf_init(&actions, 64);
904 start = nl_msg_start_nested(&actions, OVS_ACTION_ATTR_USERSPACE);
905 nl_msg_put_u32(&actions, OVS_USERSPACE_ATTR_PID,
906 dpif_port_get_pid(backer->dpif, ODPP_NONE, 0));
907 nl_msg_put_unspec_zero(&actions, OVS_USERSPACE_ATTR_USERDATA, 4);
908 nl_msg_end_nested(&actions, start);
909
910 /* Compose a dummy ethernet packet. */
911 dp_packet_init(&packet, ETH_HEADER_LEN);
912 eth = dp_packet_put_zeros(&packet, ETH_HEADER_LEN);
913 eth->eth_type = htons(0x1234);
914
915 flow_extract(&packet, &flow);
916
917 /* Execute the actions. On older datapaths this fails with ERANGE, on
918 * newer datapaths it succeeds. */
919 execute.actions = actions.data;
920 execute.actions_len = actions.size;
921 execute.packet = &packet;
922 execute.flow = &flow;
923 execute.needs_help = false;
924 execute.probe = true;
925 execute.mtu = 0;
926
927 error = dpif_execute(backer->dpif, &execute);
928
929 dp_packet_uninit(&packet);
930 ofpbuf_uninit(&actions);
931
932 switch (error) {
933 case 0:
934 return true;
935
936 case ERANGE:
937 /* Variable-length userdata is not supported. */
938 VLOG_WARN("%s: datapath does not support variable-length userdata "
939 "feature (needs Linux 3.10+ or kernel module from OVS "
940 "1..11+). The NXAST_SAMPLE action will be ignored.",
941 dpif_name(backer->dpif));
942 return false;
943
944 default:
945 /* Something odd happened. We're not sure whether variable-length
946 * userdata is supported. Default to "yes". */
947 VLOG_WARN("%s: variable-length userdata feature probe failed (%s)",
948 dpif_name(backer->dpif), ovs_strerror(error));
949 return true;
950 }
951 }
952
953 /* Tests the MPLS label stack depth supported by 'backer''s datapath.
954 *
955 * Returns the number of elements in a struct flow's mpls_lse field
956 * if the datapath supports at least that many entries in an
957 * MPLS label stack.
958 * Otherwise returns the number of MPLS push actions supported by
959 * the datapath. */
960 static size_t
961 check_max_mpls_depth(struct dpif_backer *backer)
962 {
963 struct flow flow;
964 int n;
965
966 for (n = 0; n < FLOW_MAX_MPLS_LABELS; n++) {
967 struct odputil_keybuf keybuf;
968 struct ofpbuf key;
969 struct odp_flow_key_parms odp_parms = {
970 .flow = &flow,
971 };
972
973 memset(&flow, 0, sizeof flow);
974 flow.dl_type = htons(ETH_TYPE_MPLS);
975 flow_set_mpls_bos(&flow, n, 1);
976
977 ofpbuf_use_stack(&key, &keybuf, sizeof keybuf);
978 odp_flow_key_from_flow(&odp_parms, &key);
979 if (!dpif_probe_feature(backer->dpif, "MPLS", &key, NULL)) {
980 break;
981 }
982 }
983
984 VLOG_INFO("%s: MPLS label stack length probed as %d",
985 dpif_name(backer->dpif), n);
986 return n;
987 }
988
989 /* Tests whether 'backer''s datapath supports masked data in
990 * OVS_ACTION_ATTR_SET actions. We need to disable some features on older
991 * datapaths that don't support this feature. */
992 static bool
993 check_masked_set_action(struct dpif_backer *backer)
994 {
995 struct eth_header *eth;
996 struct ofpbuf actions;
997 struct dpif_execute execute;
998 struct dp_packet packet;
999 struct flow flow;
1000 int error;
1001 struct ovs_key_ethernet key, mask;
1002
1003 /* Compose a set action that will cause an EINVAL error on older
1004 * datapaths that don't support masked set actions.
1005 * Avoid using a full mask, as it could be translated to a non-masked
1006 * set action instead. */
1007 ofpbuf_init(&actions, 64);
1008 memset(&key, 0x53, sizeof key);
1009 memset(&mask, 0x7f, sizeof mask);
1010 commit_masked_set_action(&actions, OVS_KEY_ATTR_ETHERNET, &key, &mask,
1011 sizeof key);
1012
1013 /* Compose a dummy ethernet packet. */
1014 dp_packet_init(&packet, ETH_HEADER_LEN);
1015 eth = dp_packet_put_zeros(&packet, ETH_HEADER_LEN);
1016 eth->eth_type = htons(0x1234);
1017
1018 flow_extract(&packet, &flow);
1019
1020 /* Execute the actions. On older datapaths this fails with EINVAL, on
1021 * newer datapaths it succeeds. */
1022 execute.actions = actions.data;
1023 execute.actions_len = actions.size;
1024 execute.packet = &packet;
1025 execute.flow = &flow;
1026 execute.needs_help = false;
1027 execute.probe = true;
1028 execute.mtu = 0;
1029
1030 error = dpif_execute(backer->dpif, &execute);
1031
1032 dp_packet_uninit(&packet);
1033 ofpbuf_uninit(&actions);
1034
1035 if (error) {
1036 /* Masked set action is not supported. */
1037 VLOG_INFO("%s: datapath does not support masked set action feature.",
1038 dpif_name(backer->dpif));
1039 }
1040 return !error;
1041 }
1042
1043 /* Tests whether 'backer''s datapath supports truncation of a packet in
1044 * OVS_ACTION_ATTR_TRUNC. We need to disable some features on older
1045 * datapaths that don't support this feature. */
1046 static bool
1047 check_trunc_action(struct dpif_backer *backer)
1048 {
1049 struct eth_header *eth;
1050 struct ofpbuf actions;
1051 struct dpif_execute execute;
1052 struct dp_packet packet;
1053 struct ovs_action_trunc *trunc;
1054 struct flow flow;
1055 int error;
1056
1057 /* Compose an action with output(port:1,
1058 * max_len:OVS_ACTION_OUTPUT_MIN + 1).
1059 * This translates to one truncate action and one output action. */
1060 ofpbuf_init(&actions, 64);
1061 trunc = nl_msg_put_unspec_uninit(&actions,
1062 OVS_ACTION_ATTR_TRUNC, sizeof *trunc);
1063
1064 trunc->max_len = ETH_HEADER_LEN + 1;
1065 nl_msg_put_odp_port(&actions, OVS_ACTION_ATTR_OUTPUT, u32_to_odp(1));
1066
1067 /* Compose a dummy Ethernet packet. */
1068 dp_packet_init(&packet, ETH_HEADER_LEN);
1069 eth = dp_packet_put_zeros(&packet, ETH_HEADER_LEN);
1070 eth->eth_type = htons(0x1234);
1071
1072 flow_extract(&packet, &flow);
1073
1074 /* Execute the actions. On older datapaths this fails with EINVAL, on
1075 * newer datapaths it succeeds. */
1076 execute.actions = actions.data;
1077 execute.actions_len = actions.size;
1078 execute.packet = &packet;
1079 execute.flow = &flow;
1080 execute.needs_help = false;
1081 execute.probe = true;
1082 execute.mtu = 0;
1083
1084 error = dpif_execute(backer->dpif, &execute);
1085
1086 dp_packet_uninit(&packet);
1087 ofpbuf_uninit(&actions);
1088
1089 if (error) {
1090 VLOG_INFO("%s: Datapath does not support truncate action",
1091 dpif_name(backer->dpif));
1092 } else {
1093 VLOG_INFO("%s: Datapath supports truncate action",
1094 dpif_name(backer->dpif));
1095 }
1096
1097 return !error;
1098 }
1099
1100 /* Tests whether 'backer''s datapath supports the clone action
1101 * OVS_ACTION_ATTR_CLONE. */
1102 static bool
1103 check_clone(struct dpif_backer *backer)
1104 {
1105 struct dpif_execute execute;
1106 struct eth_header *eth;
1107 struct flow flow;
1108 struct dp_packet packet;
1109 struct ofpbuf actions;
1110 size_t clone_start;
1111 int error;
1112
1113 /* Compose clone with an empty action list.
1114 * and check if datapath can decode the message. */
1115 ofpbuf_init(&actions, 64);
1116 clone_start = nl_msg_start_nested(&actions, OVS_ACTION_ATTR_CLONE);
1117 nl_msg_end_nested(&actions, clone_start);
1118
1119 /* Compose a dummy Ethernet packet. */
1120 dp_packet_init(&packet, ETH_HEADER_LEN);
1121 eth = dp_packet_put_zeros(&packet, ETH_HEADER_LEN);
1122 eth->eth_type = htons(0x1234);
1123
1124 flow_extract(&packet, &flow);
1125
1126 /* Execute the actions. On older datapaths this fails with EINVAL, on
1127 * newer datapaths it succeeds. */
1128 execute.actions = actions.data;
1129 execute.actions_len = actions.size;
1130 execute.packet = &packet;
1131 execute.flow = &flow;
1132 execute.needs_help = false;
1133 execute.probe = true;
1134 execute.mtu = 0;
1135
1136 error = dpif_execute(backer->dpif, &execute);
1137
1138 dp_packet_uninit(&packet);
1139 ofpbuf_uninit(&actions);
1140
1141 if (error) {
1142 VLOG_INFO("%s: Datapath does not support clone action",
1143 dpif_name(backer->dpif));
1144 } else {
1145 VLOG_INFO("%s: Datapath supports clone action",
1146 dpif_name(backer->dpif));
1147 }
1148
1149 return !error;
1150 }
1151
1152 #define CHECK_FEATURE__(NAME, SUPPORT, FIELD, VALUE) \
1153 static bool \
1154 check_##NAME(struct dpif_backer *backer) \
1155 { \
1156 struct flow flow; \
1157 struct odputil_keybuf keybuf; \
1158 struct ofpbuf key; \
1159 bool enable; \
1160 struct odp_flow_key_parms odp_parms = { \
1161 .flow = &flow, \
1162 .support = { \
1163 .SUPPORT = true, \
1164 }, \
1165 }; \
1166 \
1167 memset(&flow, 0, sizeof flow); \
1168 flow.FIELD = VALUE; \
1169 \
1170 ofpbuf_use_stack(&key, &keybuf, sizeof keybuf); \
1171 odp_flow_key_from_flow(&odp_parms, &key); \
1172 enable = dpif_probe_feature(backer->dpif, #NAME, &key, NULL); \
1173 \
1174 if (enable) { \
1175 VLOG_INFO("%s: Datapath supports "#NAME, dpif_name(backer->dpif)); \
1176 } else { \
1177 VLOG_INFO("%s: Datapath does not support "#NAME, \
1178 dpif_name(backer->dpif)); \
1179 } \
1180 \
1181 return enable; \
1182 }
1183 #define CHECK_FEATURE(FIELD) CHECK_FEATURE__(FIELD, FIELD, FIELD, 1)
1184
1185 CHECK_FEATURE(ct_state)
1186 CHECK_FEATURE(ct_zone)
1187 CHECK_FEATURE(ct_mark)
1188 CHECK_FEATURE__(ct_label, ct_label, ct_label.u64.lo, 1)
1189 CHECK_FEATURE__(ct_state_nat, ct_state, ct_state, CS_TRACKED|CS_SRC_NAT)
1190
1191 #undef CHECK_FEATURE
1192 #undef CHECK_FEATURE__
1193
1194 static void
1195 check_support(struct dpif_backer *backer)
1196 {
1197 /* This feature needs to be tested after udpif threads are set. */
1198 backer->support.variable_length_userdata = false;
1199
1200 /* Actions. */
1201 backer->support.odp.recirc = check_recirc(backer);
1202 backer->support.odp.max_mpls_depth = check_max_mpls_depth(backer);
1203 backer->support.masked_set_action = check_masked_set_action(backer);
1204 backer->support.trunc = check_trunc_action(backer);
1205 backer->support.ufid = check_ufid(backer);
1206 backer->support.tnl_push_pop = dpif_supports_tnl_push_pop(backer->dpif);
1207 backer->support.clone = check_clone(backer);
1208
1209 /* Flow fields. */
1210 backer->support.odp.ct_state = check_ct_state(backer);
1211 backer->support.odp.ct_zone = check_ct_zone(backer);
1212 backer->support.odp.ct_mark = check_ct_mark(backer);
1213 backer->support.odp.ct_label = check_ct_label(backer);
1214
1215 backer->support.odp.ct_state_nat = check_ct_state_nat(backer);
1216 }
1217
1218 static int
1219 construct(struct ofproto *ofproto_)
1220 {
1221 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1222 struct shash_node *node, *next;
1223 int error;
1224
1225 /* Tunnel module can get used right after the udpif threads are running. */
1226 ofproto_tunnel_init();
1227
1228 error = open_dpif_backer(ofproto->up.type, &ofproto->backer);
1229 if (error) {
1230 return error;
1231 }
1232
1233 uuid_generate(&ofproto->uuid);
1234 atomic_init(&ofproto->tables_version, OVS_VERSION_MIN);
1235 ofproto->netflow = NULL;
1236 ofproto->sflow = NULL;
1237 ofproto->ipfix = NULL;
1238 ofproto->stp = NULL;
1239 ofproto->rstp = NULL;
1240 ofproto->dump_seq = 0;
1241 hmap_init(&ofproto->bundles);
1242 ofproto->ml = mac_learning_create(MAC_ENTRY_DEFAULT_IDLE_TIME);
1243 ofproto->ms = NULL;
1244 ofproto->mbridge = mbridge_create();
1245 ofproto->has_bonded_bundles = false;
1246 ofproto->lacp_enabled = false;
1247 ovs_mutex_init_adaptive(&ofproto->stats_mutex);
1248
1249 guarded_list_init(&ofproto->ams);
1250
1251 sset_init(&ofproto->ports);
1252 sset_init(&ofproto->ghost_ports);
1253 sset_init(&ofproto->port_poll_set);
1254 ofproto->port_poll_errno = 0;
1255 ofproto->change_seq = 0;
1256 ofproto->ams_seq = seq_create();
1257 ofproto->ams_seqno = seq_read(ofproto->ams_seq);
1258
1259
1260 SHASH_FOR_EACH_SAFE (node, next, &init_ofp_ports) {
1261 struct iface_hint *iface_hint = node->data;
1262
1263 if (!strcmp(iface_hint->br_name, ofproto->up.name)) {
1264 /* Check if the datapath already has this port. */
1265 if (dpif_port_exists(ofproto->backer->dpif, node->name)) {
1266 sset_add(&ofproto->ports, node->name);
1267 }
1268
1269 free(iface_hint->br_name);
1270 free(iface_hint->br_type);
1271 free(iface_hint);
1272 shash_delete(&init_ofp_ports, node);
1273 }
1274 }
1275
1276 hmap_insert(&all_ofproto_dpifs, &ofproto->all_ofproto_dpifs_node,
1277 hash_string(ofproto->up.name, 0));
1278 memset(&ofproto->stats, 0, sizeof ofproto->stats);
1279
1280 ofproto_init_tables(ofproto_, N_TABLES);
1281 error = add_internal_flows(ofproto);
1282
1283 ofproto->up.tables[TBL_INTERNAL].flags = OFTABLE_HIDDEN | OFTABLE_READONLY;
1284
1285 return error;
1286 }
1287
1288 static int
1289 add_internal_miss_flow(struct ofproto_dpif *ofproto, int id,
1290 const struct ofpbuf *ofpacts, struct rule_dpif **rulep)
1291 {
1292 struct match match;
1293 int error;
1294 struct rule *rule;
1295
1296 match_init_catchall(&match);
1297 match_set_reg(&match, 0, id);
1298
1299 error = ofproto_dpif_add_internal_flow(ofproto, &match, 0, 0, ofpacts,
1300 &rule);
1301 *rulep = error ? NULL : rule_dpif_cast(rule);
1302
1303 return error;
1304 }
1305
1306 static int
1307 add_internal_flows(struct ofproto_dpif *ofproto)
1308 {
1309 struct ofpact_controller *controller;
1310 uint64_t ofpacts_stub[128 / 8];
1311 struct ofpbuf ofpacts;
1312 struct rule *unused_rulep OVS_UNUSED;
1313 struct match match;
1314 int error;
1315 int id;
1316
1317 ofpbuf_use_stack(&ofpacts, ofpacts_stub, sizeof ofpacts_stub);
1318 id = 1;
1319
1320 controller = ofpact_put_CONTROLLER(&ofpacts);
1321 controller->max_len = UINT16_MAX;
1322 controller->controller_id = 0;
1323 controller->reason = OFPR_IMPLICIT_MISS;
1324 ofpact_finish_CONTROLLER(&ofpacts, &controller);
1325
1326 error = add_internal_miss_flow(ofproto, id++, &ofpacts,
1327 &ofproto->miss_rule);
1328 if (error) {
1329 return error;
1330 }
1331
1332 ofpbuf_clear(&ofpacts);
1333 error = add_internal_miss_flow(ofproto, id++, &ofpacts,
1334 &ofproto->no_packet_in_rule);
1335 if (error) {
1336 return error;
1337 }
1338
1339 error = add_internal_miss_flow(ofproto, id++, &ofpacts,
1340 &ofproto->drop_frags_rule);
1341 if (error) {
1342 return error;
1343 }
1344
1345 /* Drop any run away non-recirc rule lookups. Recirc_id has to be
1346 * zero when reaching this rule.
1347 *
1348 * (priority=2), recirc_id=0, actions=drop
1349 */
1350 ofpbuf_clear(&ofpacts);
1351 match_init_catchall(&match);
1352 match_set_recirc_id(&match, 0);
1353 error = ofproto_dpif_add_internal_flow(ofproto, &match, 2, 0, &ofpacts,
1354 &unused_rulep);
1355 return error;
1356 }
1357
1358 static void
1359 destruct(struct ofproto *ofproto_)
1360 {
1361 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1362 struct ofproto_async_msg *am;
1363 struct rule_dpif *rule;
1364 struct oftable *table;
1365 struct ovs_list ams;
1366
1367 ofproto->backer->need_revalidate = REV_RECONFIGURE;
1368 xlate_txn_start();
1369 xlate_remove_ofproto(ofproto);
1370 xlate_txn_commit();
1371
1372 /* Ensure that the upcall processing threads have no remaining references
1373 * to the ofproto or anything in it. */
1374 udpif_synchronize(ofproto->backer->udpif);
1375
1376 hmap_remove(&all_ofproto_dpifs, &ofproto->all_ofproto_dpifs_node);
1377
1378 OFPROTO_FOR_EACH_TABLE (table, &ofproto->up) {
1379 CLS_FOR_EACH (rule, up.cr, &table->cls) {
1380 ofproto_rule_delete(&ofproto->up, &rule->up);
1381 }
1382 }
1383 ofproto_group_delete_all(&ofproto->up);
1384
1385 guarded_list_pop_all(&ofproto->ams, &ams);
1386 LIST_FOR_EACH_POP (am, list_node, &ams) {
1387 ofproto_async_msg_free(am);
1388 }
1389 guarded_list_destroy(&ofproto->ams);
1390
1391 recirc_free_ofproto(ofproto, ofproto->up.name);
1392
1393 mbridge_unref(ofproto->mbridge);
1394
1395 netflow_unref(ofproto->netflow);
1396 dpif_sflow_unref(ofproto->sflow);
1397 dpif_ipfix_unref(ofproto->ipfix);
1398 hmap_destroy(&ofproto->bundles);
1399 mac_learning_unref(ofproto->ml);
1400 mcast_snooping_unref(ofproto->ms);
1401
1402 sset_destroy(&ofproto->ports);
1403 sset_destroy(&ofproto->ghost_ports);
1404 sset_destroy(&ofproto->port_poll_set);
1405
1406 ovs_mutex_destroy(&ofproto->stats_mutex);
1407
1408 seq_destroy(ofproto->ams_seq);
1409
1410 close_dpif_backer(ofproto->backer);
1411 }
1412
1413 static int
1414 run(struct ofproto *ofproto_)
1415 {
1416 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1417 uint64_t new_seq, new_dump_seq;
1418
1419 if (mbridge_need_revalidate(ofproto->mbridge)) {
1420 ofproto->backer->need_revalidate = REV_RECONFIGURE;
1421 ovs_rwlock_wrlock(&ofproto->ml->rwlock);
1422 mac_learning_flush(ofproto->ml);
1423 ovs_rwlock_unlock(&ofproto->ml->rwlock);
1424 mcast_snooping_mdb_flush(ofproto->ms);
1425 }
1426
1427 /* Always updates the ofproto->ams_seqno to avoid frequent wakeup during
1428 * flow restore. Even though nothing is processed during flow restore,
1429 * all queued 'ams' will be handled immediately when flow restore
1430 * completes. */
1431 ofproto->ams_seqno = seq_read(ofproto->ams_seq);
1432
1433 /* Do not perform any periodic activity required by 'ofproto' while
1434 * waiting for flow restore to complete. */
1435 if (!ofproto_get_flow_restore_wait()) {
1436 struct ofproto_async_msg *am;
1437 struct ovs_list ams;
1438
1439 guarded_list_pop_all(&ofproto->ams, &ams);
1440 LIST_FOR_EACH_POP (am, list_node, &ams) {
1441 connmgr_send_async_msg(ofproto->up.connmgr, am);
1442 ofproto_async_msg_free(am);
1443 }
1444 }
1445
1446 if (ofproto->netflow) {
1447 netflow_run(ofproto->netflow);
1448 }
1449 if (ofproto->sflow) {
1450 dpif_sflow_run(ofproto->sflow);
1451 }
1452 if (ofproto->ipfix) {
1453 dpif_ipfix_run(ofproto->ipfix);
1454 }
1455
1456 new_seq = seq_read(connectivity_seq_get());
1457 if (ofproto->change_seq != new_seq) {
1458 struct ofport_dpif *ofport;
1459
1460 HMAP_FOR_EACH (ofport, up.hmap_node, &ofproto->up.ports) {
1461 port_run(ofport);
1462 }
1463
1464 ofproto->change_seq = new_seq;
1465 }
1466 if (ofproto->lacp_enabled || ofproto->has_bonded_bundles) {
1467 struct ofbundle *bundle;
1468
1469 HMAP_FOR_EACH (bundle, hmap_node, &ofproto->bundles) {
1470 bundle_run(bundle);
1471 }
1472 }
1473
1474 stp_run(ofproto);
1475 rstp_run(ofproto);
1476 ovs_rwlock_wrlock(&ofproto->ml->rwlock);
1477 if (mac_learning_run(ofproto->ml)) {
1478 ofproto->backer->need_revalidate = REV_MAC_LEARNING;
1479 }
1480 ovs_rwlock_unlock(&ofproto->ml->rwlock);
1481
1482 if (mcast_snooping_run(ofproto->ms)) {
1483 ofproto->backer->need_revalidate = REV_MCAST_SNOOPING;
1484 }
1485
1486 new_dump_seq = seq_read(udpif_dump_seq(ofproto->backer->udpif));
1487 if (ofproto->dump_seq != new_dump_seq) {
1488 struct rule *rule, *next_rule;
1489 long long now = time_msec();
1490
1491 /* We know stats are relatively fresh, so now is a good time to do some
1492 * periodic work. */
1493 ofproto->dump_seq = new_dump_seq;
1494
1495 /* Expire OpenFlow flows whose idle_timeout or hard_timeout
1496 * has passed. */
1497 ovs_mutex_lock(&ofproto_mutex);
1498 LIST_FOR_EACH_SAFE (rule, next_rule, expirable,
1499 &ofproto->up.expirable) {
1500 rule_expire(rule_dpif_cast(rule), now);
1501 }
1502 ovs_mutex_unlock(&ofproto_mutex);
1503
1504 /* All outstanding data in existing flows has been accounted, so it's a
1505 * good time to do bond rebalancing. */
1506 if (ofproto->has_bonded_bundles) {
1507 struct ofbundle *bundle;
1508
1509 HMAP_FOR_EACH (bundle, hmap_node, &ofproto->bundles) {
1510 if (bundle->bond) {
1511 bond_rebalance(bundle->bond);
1512 }
1513 }
1514 }
1515 }
1516 return 0;
1517 }
1518
1519 static void
1520 ofproto_dpif_wait(struct ofproto *ofproto_)
1521 {
1522 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1523
1524 if (ofproto_get_flow_restore_wait()) {
1525 return;
1526 }
1527
1528 if (ofproto->sflow) {
1529 dpif_sflow_wait(ofproto->sflow);
1530 }
1531 if (ofproto->ipfix) {
1532 dpif_ipfix_wait(ofproto->ipfix);
1533 }
1534 if (ofproto->lacp_enabled || ofproto->has_bonded_bundles) {
1535 struct ofbundle *bundle;
1536
1537 HMAP_FOR_EACH (bundle, hmap_node, &ofproto->bundles) {
1538 bundle_wait(bundle);
1539 }
1540 }
1541 if (ofproto->netflow) {
1542 netflow_wait(ofproto->netflow);
1543 }
1544 ovs_rwlock_rdlock(&ofproto->ml->rwlock);
1545 mac_learning_wait(ofproto->ml);
1546 ovs_rwlock_unlock(&ofproto->ml->rwlock);
1547 mcast_snooping_wait(ofproto->ms);
1548 stp_wait(ofproto);
1549 if (ofproto->backer->need_revalidate) {
1550 poll_immediate_wake();
1551 }
1552
1553 seq_wait(udpif_dump_seq(ofproto->backer->udpif), ofproto->dump_seq);
1554 seq_wait(ofproto->ams_seq, ofproto->ams_seqno);
1555 }
1556
1557 static void
1558 type_get_memory_usage(const char *type, struct simap *usage)
1559 {
1560 struct dpif_backer *backer;
1561
1562 backer = shash_find_data(&all_dpif_backers, type);
1563 if (backer) {
1564 udpif_get_memory_usage(backer->udpif, usage);
1565 }
1566 }
1567
1568 static void
1569 flush(struct ofproto *ofproto_)
1570 {
1571 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1572 struct dpif_backer *backer = ofproto->backer;
1573
1574 if (backer) {
1575 udpif_flush(backer->udpif);
1576 }
1577 }
1578
1579 static void
1580 query_tables(struct ofproto *ofproto,
1581 struct ofputil_table_features *features,
1582 struct ofputil_table_stats *stats)
1583 {
1584 strcpy(features->name, "classifier");
1585
1586 if (stats) {
1587 int i;
1588
1589 for (i = 0; i < ofproto->n_tables; i++) {
1590 unsigned long missed, matched;
1591
1592 atomic_read_relaxed(&ofproto->tables[i].n_matched, &matched);
1593 atomic_read_relaxed(&ofproto->tables[i].n_missed, &missed);
1594
1595 stats[i].matched_count = matched;
1596 stats[i].lookup_count = matched + missed;
1597 }
1598 }
1599 }
1600
1601 static void
1602 set_tables_version(struct ofproto *ofproto_, ovs_version_t version)
1603 {
1604 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1605
1606 /* Use memory_order_release to signify that any prior memory accesses can
1607 * not be reordered to happen after this atomic store. This makes sure the
1608 * new version is properly set up when the readers can read this 'version'
1609 * value. */
1610 atomic_store_explicit(&ofproto->tables_version, version,
1611 memory_order_release);
1612 /* 'need_revalidate' can be reordered to happen before the atomic_store
1613 * above, but it does not matter as this variable is not accessed by other
1614 * threads. */
1615 ofproto->backer->need_revalidate = REV_FLOW_TABLE;
1616 }
1617
1618 static struct ofport *
1619 port_alloc(void)
1620 {
1621 struct ofport_dpif *port = xzalloc(sizeof *port);
1622 return &port->up;
1623 }
1624
1625 static void
1626 port_dealloc(struct ofport *port_)
1627 {
1628 struct ofport_dpif *port = ofport_dpif_cast(port_);
1629 free(port);
1630 }
1631
1632 static int
1633 port_construct(struct ofport *port_)
1634 {
1635 struct ofport_dpif *port = ofport_dpif_cast(port_);
1636 struct ofproto_dpif *ofproto = ofproto_dpif_cast(port->up.ofproto);
1637 const struct netdev *netdev = port->up.netdev;
1638 char namebuf[NETDEV_VPORT_NAME_BUFSIZE];
1639 const char *dp_port_name;
1640 struct dpif_port dpif_port;
1641 int error;
1642
1643 ofproto->backer->need_revalidate = REV_RECONFIGURE;
1644 port->bundle = NULL;
1645 port->cfm = NULL;
1646 port->bfd = NULL;
1647 port->lldp = NULL;
1648 port->may_enable = false;
1649 port->stp_port = NULL;
1650 port->stp_state = STP_DISABLED;
1651 port->rstp_port = NULL;
1652 port->rstp_state = RSTP_DISABLED;
1653 port->is_tunnel = false;
1654 port->peer = NULL;
1655 port->qdscp = NULL;
1656 port->n_qdscp = 0;
1657 port->carrier_seq = netdev_get_carrier_resets(netdev);
1658 port->is_layer3 = netdev_vport_is_layer3(netdev);
1659
1660 if (netdev_vport_is_patch(netdev)) {
1661 /* By bailing out here, we don't submit the port to the sFlow module
1662 * to be considered for counter polling export. This is correct
1663 * because the patch port represents an interface that sFlow considers
1664 * to be "internal" to the switch as a whole, and therefore not a
1665 * candidate for counter polling. */
1666 port->odp_port = ODPP_NONE;
1667 ofport_update_peer(port);
1668 return 0;
1669 }
1670
1671 dp_port_name = netdev_vport_get_dpif_port(netdev, namebuf, sizeof namebuf);
1672 error = dpif_port_query_by_name(ofproto->backer->dpif, dp_port_name,
1673 &dpif_port);
1674 if (error) {
1675 return error;
1676 }
1677
1678 port->odp_port = dpif_port.port_no;
1679
1680 if (netdev_get_tunnel_config(netdev)) {
1681 atomic_count_inc(&ofproto->backer->tnl_count);
1682 error = tnl_port_add(port, port->up.netdev, port->odp_port,
1683 ovs_native_tunneling_is_on(ofproto), dp_port_name);
1684 if (error) {
1685 atomic_count_dec(&ofproto->backer->tnl_count);
1686 dpif_port_destroy(&dpif_port);
1687 return error;
1688 }
1689
1690 port->is_tunnel = true;
1691 if (ofproto->ipfix) {
1692 dpif_ipfix_add_tunnel_port(ofproto->ipfix, port_, port->odp_port);
1693 }
1694 } else {
1695 /* Sanity-check that a mapping doesn't already exist. This
1696 * shouldn't happen for non-tunnel ports. */
1697 if (odp_port_to_ofp_port(ofproto, port->odp_port) != OFPP_NONE) {
1698 VLOG_ERR("port %s already has an OpenFlow port number",
1699 dpif_port.name);
1700 dpif_port_destroy(&dpif_port);
1701 return EBUSY;
1702 }
1703
1704 ovs_rwlock_wrlock(&ofproto->backer->odp_to_ofport_lock);
1705 hmap_insert(&ofproto->backer->odp_to_ofport_map, &port->odp_port_node,
1706 hash_odp_port(port->odp_port));
1707 ovs_rwlock_unlock(&ofproto->backer->odp_to_ofport_lock);
1708 }
1709 dpif_port_destroy(&dpif_port);
1710
1711 if (ofproto->sflow) {
1712 dpif_sflow_add_port(ofproto->sflow, port_, port->odp_port);
1713 }
1714
1715 return 0;
1716 }
1717
1718 static void
1719 port_destruct(struct ofport *port_, bool del)
1720 {
1721 struct ofport_dpif *port = ofport_dpif_cast(port_);
1722 struct ofproto_dpif *ofproto = ofproto_dpif_cast(port->up.ofproto);
1723 const char *devname = netdev_get_name(port->up.netdev);
1724 char namebuf[NETDEV_VPORT_NAME_BUFSIZE];
1725 const char *dp_port_name;
1726
1727 ofproto->backer->need_revalidate = REV_RECONFIGURE;
1728 xlate_txn_start();
1729 xlate_ofport_remove(port);
1730 xlate_txn_commit();
1731
1732 dp_port_name = netdev_vport_get_dpif_port(port->up.netdev, namebuf,
1733 sizeof namebuf);
1734 if (del && dpif_port_exists(ofproto->backer->dpif, dp_port_name)) {
1735 /* The underlying device is still there, so delete it. This
1736 * happens when the ofproto is being destroyed, since the caller
1737 * assumes that removal of attached ports will happen as part of
1738 * destruction. */
1739 if (!port->is_tunnel) {
1740 dpif_port_del(ofproto->backer->dpif, port->odp_port);
1741 }
1742 }
1743
1744 if (port->peer) {
1745 port->peer->peer = NULL;
1746 port->peer = NULL;
1747 }
1748
1749 if (port->odp_port != ODPP_NONE && !port->is_tunnel) {
1750 ovs_rwlock_wrlock(&ofproto->backer->odp_to_ofport_lock);
1751 hmap_remove(&ofproto->backer->odp_to_ofport_map, &port->odp_port_node);
1752 ovs_rwlock_unlock(&ofproto->backer->odp_to_ofport_lock);
1753 }
1754
1755 if (port->is_tunnel) {
1756 atomic_count_dec(&ofproto->backer->tnl_count);
1757 }
1758
1759 if (port->is_tunnel && ofproto->ipfix) {
1760 dpif_ipfix_del_tunnel_port(ofproto->ipfix, port->odp_port);
1761 }
1762
1763 tnl_port_del(port);
1764 sset_find_and_delete(&ofproto->ports, devname);
1765 sset_find_and_delete(&ofproto->ghost_ports, devname);
1766 bundle_remove(port_);
1767 set_cfm(port_, NULL);
1768 set_bfd(port_, NULL);
1769 set_lldp(port_, NULL);
1770 if (port->stp_port) {
1771 stp_port_disable(port->stp_port);
1772 }
1773 set_rstp_port(port_, NULL);
1774 if (ofproto->sflow) {
1775 dpif_sflow_del_port(ofproto->sflow, port->odp_port);
1776 }
1777
1778 free(port->qdscp);
1779 }
1780
1781 static void
1782 port_modified(struct ofport *port_)
1783 {
1784 struct ofport_dpif *port = ofport_dpif_cast(port_);
1785 char namebuf[NETDEV_VPORT_NAME_BUFSIZE];
1786 const char *dp_port_name;
1787 struct netdev *netdev = port->up.netdev;
1788
1789 if (port->bundle && port->bundle->bond) {
1790 bond_slave_set_netdev(port->bundle->bond, port, netdev);
1791 }
1792
1793 if (port->cfm) {
1794 cfm_set_netdev(port->cfm, netdev);
1795 }
1796
1797 if (port->bfd) {
1798 bfd_set_netdev(port->bfd, netdev);
1799 }
1800
1801 ofproto_dpif_monitor_port_update(port, port->bfd, port->cfm,
1802 port->lldp, &port->up.pp.hw_addr);
1803
1804 dp_port_name = netdev_vport_get_dpif_port(netdev, namebuf, sizeof namebuf);
1805
1806 if (port->is_tunnel) {
1807 struct ofproto_dpif *ofproto = ofproto_dpif_cast(port->up.ofproto);
1808
1809 if (tnl_port_reconfigure(port, netdev, port->odp_port,
1810 ovs_native_tunneling_is_on(ofproto),
1811 dp_port_name)) {
1812 ofproto->backer->need_revalidate = REV_RECONFIGURE;
1813 }
1814 }
1815
1816 ofport_update_peer(port);
1817 }
1818
1819 static void
1820 port_reconfigured(struct ofport *port_, enum ofputil_port_config old_config)
1821 {
1822 struct ofport_dpif *port = ofport_dpif_cast(port_);
1823 struct ofproto_dpif *ofproto = ofproto_dpif_cast(port->up.ofproto);
1824 enum ofputil_port_config changed = old_config ^ port->up.pp.config;
1825
1826 if (changed & (OFPUTIL_PC_NO_RECV | OFPUTIL_PC_NO_RECV_STP |
1827 OFPUTIL_PC_NO_FWD | OFPUTIL_PC_NO_FLOOD |
1828 OFPUTIL_PC_NO_PACKET_IN)) {
1829 ofproto->backer->need_revalidate = REV_RECONFIGURE;
1830
1831 if (changed & OFPUTIL_PC_NO_FLOOD && port->bundle) {
1832 bundle_update(port->bundle);
1833 }
1834 }
1835 }
1836
1837 static int
1838 set_sflow(struct ofproto *ofproto_,
1839 const struct ofproto_sflow_options *sflow_options)
1840 {
1841 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1842 struct dpif_sflow *ds = ofproto->sflow;
1843
1844 if (sflow_options) {
1845 uint32_t old_probability = ds ? dpif_sflow_get_probability(ds) : 0;
1846 if (!ds) {
1847 struct ofport_dpif *ofport;
1848
1849 ds = ofproto->sflow = dpif_sflow_create();
1850 HMAP_FOR_EACH (ofport, up.hmap_node, &ofproto->up.ports) {
1851 dpif_sflow_add_port(ds, &ofport->up, ofport->odp_port);
1852 }
1853 }
1854 dpif_sflow_set_options(ds, sflow_options);
1855 if (dpif_sflow_get_probability(ds) != old_probability) {
1856 ofproto->backer->need_revalidate = REV_RECONFIGURE;
1857 }
1858 } else {
1859 if (ds) {
1860 dpif_sflow_unref(ds);
1861 ofproto->backer->need_revalidate = REV_RECONFIGURE;
1862 ofproto->sflow = NULL;
1863 }
1864 }
1865 return 0;
1866 }
1867
1868 static int
1869 set_ipfix(
1870 struct ofproto *ofproto_,
1871 const struct ofproto_ipfix_bridge_exporter_options *bridge_exporter_options,
1872 const struct ofproto_ipfix_flow_exporter_options *flow_exporters_options,
1873 size_t n_flow_exporters_options)
1874 {
1875 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1876 struct dpif_ipfix *di = ofproto->ipfix;
1877 bool has_options = bridge_exporter_options || flow_exporters_options;
1878 bool new_di = false;
1879
1880 if (has_options && !di) {
1881 di = ofproto->ipfix = dpif_ipfix_create();
1882 new_di = true;
1883 }
1884
1885 if (di) {
1886 /* Call set_options in any case to cleanly flush the flow
1887 * caches in the last exporters that are to be destroyed. */
1888 dpif_ipfix_set_options(
1889 di, bridge_exporter_options, flow_exporters_options,
1890 n_flow_exporters_options);
1891
1892 /* Add tunnel ports only when a new ipfix created */
1893 if (new_di == true) {
1894 struct ofport_dpif *ofport;
1895 HMAP_FOR_EACH (ofport, up.hmap_node, &ofproto->up.ports) {
1896 if (ofport->is_tunnel == true) {
1897 dpif_ipfix_add_tunnel_port(di, &ofport->up, ofport->odp_port);
1898 }
1899 }
1900 }
1901
1902 if (!has_options) {
1903 dpif_ipfix_unref(di);
1904 ofproto->ipfix = NULL;
1905 }
1906 }
1907
1908 return 0;
1909 }
1910
1911 static int
1912 get_ipfix_stats(const struct ofproto *ofproto_,
1913 bool bridge_ipfix,
1914 struct ovs_list *replies)
1915 {
1916 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1917 struct dpif_ipfix *di = ofproto->ipfix;
1918
1919 if (!di) {
1920 return OFPERR_NXST_NOT_CONFIGURED;
1921 }
1922
1923 return dpif_ipfix_get_stats(di, bridge_ipfix, replies);
1924 }
1925
1926 static int
1927 set_cfm(struct ofport *ofport_, const struct cfm_settings *s)
1928 {
1929 struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
1930 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport->up.ofproto);
1931 struct cfm *old = ofport->cfm;
1932 int error = 0;
1933
1934 if (s) {
1935 if (!ofport->cfm) {
1936 ofport->cfm = cfm_create(ofport->up.netdev);
1937 }
1938
1939 if (cfm_configure(ofport->cfm, s)) {
1940 error = 0;
1941 goto out;
1942 }
1943
1944 error = EINVAL;
1945 }
1946 cfm_unref(ofport->cfm);
1947 ofport->cfm = NULL;
1948 out:
1949 if (ofport->cfm != old) {
1950 ofproto->backer->need_revalidate = REV_RECONFIGURE;
1951 }
1952 ofproto_dpif_monitor_port_update(ofport, ofport->bfd, ofport->cfm,
1953 ofport->lldp, &ofport->up.pp.hw_addr);
1954 return error;
1955 }
1956
1957 static bool
1958 cfm_status_changed(struct ofport *ofport_)
1959 {
1960 struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
1961
1962 return ofport->cfm ? cfm_check_status_change(ofport->cfm) : true;
1963 }
1964
1965 static int
1966 get_cfm_status(const struct ofport *ofport_,
1967 struct cfm_status *status)
1968 {
1969 struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
1970 int ret = 0;
1971
1972 if (ofport->cfm) {
1973 cfm_get_status(ofport->cfm, status);
1974 } else {
1975 ret = ENOENT;
1976 }
1977
1978 return ret;
1979 }
1980
1981 static int
1982 set_bfd(struct ofport *ofport_, const struct smap *cfg)
1983 {
1984 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport_->ofproto);
1985 struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
1986 struct bfd *old;
1987
1988 old = ofport->bfd;
1989 ofport->bfd = bfd_configure(old, netdev_get_name(ofport->up.netdev),
1990 cfg, ofport->up.netdev);
1991 if (ofport->bfd != old) {
1992 ofproto->backer->need_revalidate = REV_RECONFIGURE;
1993 }
1994 ofproto_dpif_monitor_port_update(ofport, ofport->bfd, ofport->cfm,
1995 ofport->lldp, &ofport->up.pp.hw_addr);
1996 return 0;
1997 }
1998
1999 static bool
2000 bfd_status_changed(struct ofport *ofport_)
2001 {
2002 struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
2003
2004 return ofport->bfd ? bfd_check_status_change(ofport->bfd) : true;
2005 }
2006
2007 static int
2008 get_bfd_status(struct ofport *ofport_, struct smap *smap)
2009 {
2010 struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
2011 int ret = 0;
2012
2013 if (ofport->bfd) {
2014 bfd_get_status(ofport->bfd, smap);
2015 } else {
2016 ret = ENOENT;
2017 }
2018
2019 return ret;
2020 }
2021
2022 static int
2023 set_lldp(struct ofport *ofport_,
2024 const struct smap *cfg)
2025 {
2026 struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
2027 int error = 0;
2028
2029 if (cfg) {
2030 if (!ofport->lldp) {
2031 struct ofproto_dpif *ofproto;
2032
2033 ofproto = ofproto_dpif_cast(ofport->up.ofproto);
2034 ofproto->backer->need_revalidate = REV_RECONFIGURE;
2035 ofport->lldp = lldp_create(ofport->up.netdev, ofport_->mtu, cfg);
2036 }
2037
2038 if (!lldp_configure(ofport->lldp, cfg)) {
2039 error = EINVAL;
2040 }
2041 }
2042 if (error) {
2043 lldp_unref(ofport->lldp);
2044 ofport->lldp = NULL;
2045 }
2046
2047 ofproto_dpif_monitor_port_update(ofport,
2048 ofport->bfd,
2049 ofport->cfm,
2050 ofport->lldp,
2051 &ofport->up.pp.hw_addr);
2052 return error;
2053 }
2054
2055 static bool
2056 get_lldp_status(const struct ofport *ofport_,
2057 struct lldp_status *status OVS_UNUSED)
2058 {
2059 struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
2060
2061 return ofport->lldp ? true : false;
2062 }
2063
2064 static int
2065 set_aa(struct ofproto *ofproto OVS_UNUSED,
2066 const struct aa_settings *s)
2067 {
2068 return aa_configure(s);
2069 }
2070
2071 static int
2072 aa_mapping_set(struct ofproto *ofproto_ OVS_UNUSED, void *aux,
2073 const struct aa_mapping_settings *s)
2074 {
2075 return aa_mapping_register(aux, s);
2076 }
2077
2078 static int
2079 aa_mapping_unset(struct ofproto *ofproto OVS_UNUSED, void *aux)
2080 {
2081 return aa_mapping_unregister(aux);
2082 }
2083
2084 static int
2085 aa_vlan_get_queued(struct ofproto *ofproto OVS_UNUSED, struct ovs_list *list)
2086 {
2087 return aa_get_vlan_queued(list);
2088 }
2089
2090 static unsigned int
2091 aa_vlan_get_queue_size(struct ofproto *ofproto OVS_UNUSED)
2092 {
2093 return aa_get_vlan_queue_size();
2094 }
2095
2096 \f
2097 /* Spanning Tree. */
2098
2099 /* Called while rstp_mutex is held. */
2100 static void
2101 rstp_send_bpdu_cb(struct dp_packet *pkt, void *ofport_, void *ofproto_)
2102 {
2103 struct ofproto_dpif *ofproto = ofproto_;
2104 struct ofport_dpif *ofport = ofport_;
2105 struct eth_header *eth = dp_packet_l2(pkt);
2106
2107 netdev_get_etheraddr(ofport->up.netdev, &eth->eth_src);
2108 if (eth_addr_is_zero(eth->eth_src)) {
2109 VLOG_WARN_RL(&rl, "%s port %d: cannot send RSTP BPDU on a port which "
2110 "does not have a configured source MAC address.",
2111 ofproto->up.name, ofp_to_u16(ofport->up.ofp_port));
2112 } else {
2113 ofproto_dpif_send_packet(ofport, false, pkt);
2114 }
2115 dp_packet_delete(pkt);
2116 }
2117
2118 static void
2119 send_bpdu_cb(struct dp_packet *pkt, int port_num, void *ofproto_)
2120 {
2121 struct ofproto_dpif *ofproto = ofproto_;
2122 struct stp_port *sp = stp_get_port(ofproto->stp, port_num);
2123 struct ofport_dpif *ofport;
2124
2125 ofport = stp_port_get_aux(sp);
2126 if (!ofport) {
2127 VLOG_WARN_RL(&rl, "%s: cannot send BPDU on unknown port %d",
2128 ofproto->up.name, port_num);
2129 } else {
2130 struct eth_header *eth = dp_packet_l2(pkt);
2131
2132 netdev_get_etheraddr(ofport->up.netdev, &eth->eth_src);
2133 if (eth_addr_is_zero(eth->eth_src)) {
2134 VLOG_WARN_RL(&rl, "%s: cannot send BPDU on port %d "
2135 "with unknown MAC", ofproto->up.name, port_num);
2136 } else {
2137 ofproto_dpif_send_packet(ofport, false, pkt);
2138 }
2139 }
2140 dp_packet_delete(pkt);
2141 }
2142
2143 /* Configure RSTP on 'ofproto_' using the settings defined in 's'. */
2144 static void
2145 set_rstp(struct ofproto *ofproto_, const struct ofproto_rstp_settings *s)
2146 {
2147 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
2148
2149 /* Only revalidate flows if the configuration changed. */
2150 if (!s != !ofproto->rstp) {
2151 ofproto->backer->need_revalidate = REV_RECONFIGURE;
2152 }
2153
2154 if (s) {
2155 if (!ofproto->rstp) {
2156 ofproto->rstp = rstp_create(ofproto_->name, s->address,
2157 rstp_send_bpdu_cb, ofproto);
2158 ofproto->rstp_last_tick = time_msec();
2159 }
2160 rstp_set_bridge_address(ofproto->rstp, s->address);
2161 rstp_set_bridge_priority(ofproto->rstp, s->priority);
2162 rstp_set_bridge_ageing_time(ofproto->rstp, s->ageing_time);
2163 rstp_set_bridge_force_protocol_version(ofproto->rstp,
2164 s->force_protocol_version);
2165 rstp_set_bridge_max_age(ofproto->rstp, s->bridge_max_age);
2166 rstp_set_bridge_forward_delay(ofproto->rstp, s->bridge_forward_delay);
2167 rstp_set_bridge_transmit_hold_count(ofproto->rstp,
2168 s->transmit_hold_count);
2169 } else {
2170 struct ofport *ofport;
2171 HMAP_FOR_EACH (ofport, hmap_node, &ofproto->up.ports) {
2172 set_rstp_port(ofport, NULL);
2173 }
2174 rstp_unref(ofproto->rstp);
2175 ofproto->rstp = NULL;
2176 }
2177 }
2178
2179 static void
2180 get_rstp_status(struct ofproto *ofproto_, struct ofproto_rstp_status *s)
2181 {
2182 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
2183
2184 if (ofproto->rstp) {
2185 s->enabled = true;
2186 s->root_id = rstp_get_root_id(ofproto->rstp);
2187 s->bridge_id = rstp_get_bridge_id(ofproto->rstp);
2188 s->designated_id = rstp_get_designated_id(ofproto->rstp);
2189 s->root_path_cost = rstp_get_root_path_cost(ofproto->rstp);
2190 s->designated_port_id = rstp_get_designated_port_id(ofproto->rstp);
2191 s->bridge_port_id = rstp_get_bridge_port_id(ofproto->rstp);
2192 } else {
2193 s->enabled = false;
2194 }
2195 }
2196
2197 static void
2198 update_rstp_port_state(struct ofport_dpif *ofport)
2199 {
2200 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport->up.ofproto);
2201 enum rstp_state state;
2202
2203 /* Figure out new state. */
2204 state = ofport->rstp_port ? rstp_port_get_state(ofport->rstp_port)
2205 : RSTP_DISABLED;
2206
2207 /* Update state. */
2208 if (ofport->rstp_state != state) {
2209 enum ofputil_port_state of_state;
2210 bool fwd_change;
2211
2212 VLOG_DBG("port %s: RSTP state changed from %s to %s",
2213 netdev_get_name(ofport->up.netdev),
2214 rstp_state_name(ofport->rstp_state),
2215 rstp_state_name(state));
2216
2217 if (rstp_learn_in_state(ofport->rstp_state)
2218 != rstp_learn_in_state(state)) {
2219 /* XXX: Learning action flows should also be flushed. */
2220 if (ofport->bundle) {
2221 if (!rstp_shift_root_learned_address(ofproto->rstp)
2222 || rstp_get_old_root_aux(ofproto->rstp) != ofport) {
2223 bundle_flush_macs(ofport->bundle, false);
2224 }
2225 }
2226 }
2227 fwd_change = rstp_forward_in_state(ofport->rstp_state)
2228 != rstp_forward_in_state(state);
2229
2230 ofproto->backer->need_revalidate = REV_RSTP;
2231 ofport->rstp_state = state;
2232
2233 if (fwd_change && ofport->bundle) {
2234 bundle_update(ofport->bundle);
2235 }
2236
2237 /* Update the RSTP state bits in the OpenFlow port description. */
2238 of_state = ofport->up.pp.state & ~OFPUTIL_PS_STP_MASK;
2239 of_state |= (state == RSTP_LEARNING ? OFPUTIL_PS_STP_LEARN
2240 : state == RSTP_FORWARDING ? OFPUTIL_PS_STP_FORWARD
2241 : state == RSTP_DISCARDING ? OFPUTIL_PS_STP_LISTEN
2242 : 0);
2243 ofproto_port_set_state(&ofport->up, of_state);
2244 }
2245 }
2246
2247 static void
2248 rstp_run(struct ofproto_dpif *ofproto)
2249 {
2250 if (ofproto->rstp) {
2251 long long int now = time_msec();
2252 long long int elapsed = now - ofproto->rstp_last_tick;
2253 struct rstp_port *rp;
2254 struct ofport_dpif *ofport;
2255
2256 /* Every second, decrease the values of the timers. */
2257 if (elapsed >= 1000) {
2258 rstp_tick_timers(ofproto->rstp);
2259 ofproto->rstp_last_tick = now;
2260 }
2261 rp = NULL;
2262 while ((ofport = rstp_get_next_changed_port_aux(ofproto->rstp, &rp))) {
2263 update_rstp_port_state(ofport);
2264 }
2265 rp = NULL;
2266 ofport = NULL;
2267 /* FIXME: This check should be done on-event (i.e., when setting
2268 * p->fdb_flush) and not periodically.
2269 */
2270 while ((ofport = rstp_check_and_reset_fdb_flush(ofproto->rstp, &rp))) {
2271 if (!rstp_shift_root_learned_address(ofproto->rstp)
2272 || rstp_get_old_root_aux(ofproto->rstp) != ofport) {
2273 bundle_flush_macs(ofport->bundle, false);
2274 }
2275 }
2276
2277 if (rstp_shift_root_learned_address(ofproto->rstp)) {
2278 struct ofport_dpif *old_root_aux =
2279 (struct ofport_dpif *)rstp_get_old_root_aux(ofproto->rstp);
2280 struct ofport_dpif *new_root_aux =
2281 (struct ofport_dpif *)rstp_get_new_root_aux(ofproto->rstp);
2282 if (old_root_aux != NULL && new_root_aux != NULL) {
2283 bundle_move(old_root_aux->bundle, new_root_aux->bundle);
2284 rstp_reset_root_changed(ofproto->rstp);
2285 }
2286 }
2287 }
2288 }
2289
2290 /* Configures STP on 'ofproto_' using the settings defined in 's'. */
2291 static int
2292 set_stp(struct ofproto *ofproto_, const struct ofproto_stp_settings *s)
2293 {
2294 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
2295
2296 /* Only revalidate flows if the configuration changed. */
2297 if (!s != !ofproto->stp) {
2298 ofproto->backer->need_revalidate = REV_RECONFIGURE;
2299 }
2300
2301 if (s) {
2302 if (!ofproto->stp) {
2303 ofproto->stp = stp_create(ofproto_->name, s->system_id,
2304 send_bpdu_cb, ofproto);
2305 ofproto->stp_last_tick = time_msec();
2306 }
2307
2308 stp_set_bridge_id(ofproto->stp, s->system_id);
2309 stp_set_bridge_priority(ofproto->stp, s->priority);
2310 stp_set_hello_time(ofproto->stp, s->hello_time);
2311 stp_set_max_age(ofproto->stp, s->max_age);
2312 stp_set_forward_delay(ofproto->stp, s->fwd_delay);
2313 } else {
2314 struct ofport *ofport;
2315
2316 HMAP_FOR_EACH (ofport, hmap_node, &ofproto->up.ports) {
2317 set_stp_port(ofport, NULL);
2318 }
2319
2320 stp_unref(ofproto->stp);
2321 ofproto->stp = NULL;
2322 }
2323
2324 return 0;
2325 }
2326
2327 static int
2328 get_stp_status(struct ofproto *ofproto_, struct ofproto_stp_status *s)
2329 {
2330 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
2331
2332 if (ofproto->stp) {
2333 s->enabled = true;
2334 s->bridge_id = stp_get_bridge_id(ofproto->stp);
2335 s->designated_root = stp_get_designated_root(ofproto->stp);
2336 s->root_path_cost = stp_get_root_path_cost(ofproto->stp);
2337 } else {
2338 s->enabled = false;
2339 }
2340
2341 return 0;
2342 }
2343
2344 static void
2345 update_stp_port_state(struct ofport_dpif *ofport)
2346 {
2347 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport->up.ofproto);
2348 enum stp_state state;
2349
2350 /* Figure out new state. */
2351 state = ofport->stp_port ? stp_port_get_state(ofport->stp_port)
2352 : STP_DISABLED;
2353
2354 /* Update state. */
2355 if (ofport->stp_state != state) {
2356 enum ofputil_port_state of_state;
2357 bool fwd_change;
2358
2359 VLOG_DBG("port %s: STP state changed from %s to %s",
2360 netdev_get_name(ofport->up.netdev),
2361 stp_state_name(ofport->stp_state),
2362 stp_state_name(state));
2363 if (stp_learn_in_state(ofport->stp_state)
2364 != stp_learn_in_state(state)) {
2365 /* xxx Learning action flows should also be flushed. */
2366 ovs_rwlock_wrlock(&ofproto->ml->rwlock);
2367 mac_learning_flush(ofproto->ml);
2368 ovs_rwlock_unlock(&ofproto->ml->rwlock);
2369 mcast_snooping_mdb_flush(ofproto->ms);
2370 }
2371 fwd_change = stp_forward_in_state(ofport->stp_state)
2372 != stp_forward_in_state(state);
2373
2374 ofproto->backer->need_revalidate = REV_STP;
2375 ofport->stp_state = state;
2376 ofport->stp_state_entered = time_msec();
2377
2378 if (fwd_change && ofport->bundle) {
2379 bundle_update(ofport->bundle);
2380 }
2381
2382 /* Update the STP state bits in the OpenFlow port description. */
2383 of_state = ofport->up.pp.state & ~OFPUTIL_PS_STP_MASK;
2384 of_state |= (state == STP_LISTENING ? OFPUTIL_PS_STP_LISTEN
2385 : state == STP_LEARNING ? OFPUTIL_PS_STP_LEARN
2386 : state == STP_FORWARDING ? OFPUTIL_PS_STP_FORWARD
2387 : state == STP_BLOCKING ? OFPUTIL_PS_STP_BLOCK
2388 : 0);
2389 ofproto_port_set_state(&ofport->up, of_state);
2390 }
2391 }
2392
2393 /* Configures STP on 'ofport_' using the settings defined in 's'. The
2394 * caller is responsible for assigning STP port numbers and ensuring
2395 * there are no duplicates. */
2396 static int
2397 set_stp_port(struct ofport *ofport_,
2398 const struct ofproto_port_stp_settings *s)
2399 {
2400 struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
2401 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport->up.ofproto);
2402 struct stp_port *sp = ofport->stp_port;
2403
2404 if (!s || !s->enable) {
2405 if (sp) {
2406 ofport->stp_port = NULL;
2407 stp_port_disable(sp);
2408 update_stp_port_state(ofport);
2409 }
2410 return 0;
2411 } else if (sp && stp_port_no(sp) != s->port_num
2412 && ofport == stp_port_get_aux(sp)) {
2413 /* The port-id changed, so disable the old one if it's not
2414 * already in use by another port. */
2415 stp_port_disable(sp);
2416 }
2417
2418 sp = ofport->stp_port = stp_get_port(ofproto->stp, s->port_num);
2419
2420 /* Set name before enabling the port so that debugging messages can print
2421 * the name. */
2422 stp_port_set_name(sp, netdev_get_name(ofport->up.netdev));
2423 stp_port_enable(sp);
2424
2425 stp_port_set_aux(sp, ofport);
2426 stp_port_set_priority(sp, s->priority);
2427 stp_port_set_path_cost(sp, s->path_cost);
2428
2429 update_stp_port_state(ofport);
2430
2431 return 0;
2432 }
2433
2434 static int
2435 get_stp_port_status(struct ofport *ofport_,
2436 struct ofproto_port_stp_status *s)
2437 {
2438 struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
2439 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport->up.ofproto);
2440 struct stp_port *sp = ofport->stp_port;
2441
2442 if (!ofproto->stp || !sp) {
2443 s->enabled = false;
2444 return 0;
2445 }
2446
2447 s->enabled = true;
2448 s->port_id = stp_port_get_id(sp);
2449 s->state = stp_port_get_state(sp);
2450 s->sec_in_state = (time_msec() - ofport->stp_state_entered) / 1000;
2451 s->role = stp_port_get_role(sp);
2452
2453 return 0;
2454 }
2455
2456 static int
2457 get_stp_port_stats(struct ofport *ofport_,
2458 struct ofproto_port_stp_stats *s)
2459 {
2460 struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
2461 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport->up.ofproto);
2462 struct stp_port *sp = ofport->stp_port;
2463
2464 if (!ofproto->stp || !sp) {
2465 s->enabled = false;
2466 return 0;
2467 }
2468
2469 s->enabled = true;
2470 stp_port_get_counts(sp, &s->tx_count, &s->rx_count, &s->error_count);
2471
2472 return 0;
2473 }
2474
2475 static void
2476 stp_run(struct ofproto_dpif *ofproto)
2477 {
2478 if (ofproto->stp) {
2479 long long int now = time_msec();
2480 long long int elapsed = now - ofproto->stp_last_tick;
2481 struct stp_port *sp;
2482
2483 if (elapsed > 0) {
2484 stp_tick(ofproto->stp, MIN(INT_MAX, elapsed));
2485 ofproto->stp_last_tick = now;
2486 }
2487 while (stp_get_changed_port(ofproto->stp, &sp)) {
2488 struct ofport_dpif *ofport = stp_port_get_aux(sp);
2489
2490 if (ofport) {
2491 update_stp_port_state(ofport);
2492 }
2493 }
2494
2495 if (stp_check_and_reset_fdb_flush(ofproto->stp)) {
2496 ovs_rwlock_wrlock(&ofproto->ml->rwlock);
2497 mac_learning_flush(ofproto->ml);
2498 ovs_rwlock_unlock(&ofproto->ml->rwlock);
2499 mcast_snooping_mdb_flush(ofproto->ms);
2500 }
2501 }
2502 }
2503
2504 static void
2505 stp_wait(struct ofproto_dpif *ofproto)
2506 {
2507 if (ofproto->stp) {
2508 poll_timer_wait(1000);
2509 }
2510 }
2511
2512 /* Configures RSTP on 'ofport_' using the settings defined in 's'. The
2513 * caller is responsible for assigning RSTP port numbers and ensuring
2514 * there are no duplicates. */
2515 static void
2516 set_rstp_port(struct ofport *ofport_,
2517 const struct ofproto_port_rstp_settings *s)
2518 {
2519 struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
2520 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport->up.ofproto);
2521 struct rstp_port *rp = ofport->rstp_port;
2522
2523 if (!s || !s->enable) {
2524 if (rp) {
2525 rstp_port_set_aux(rp, NULL);
2526 rstp_port_set_state(rp, RSTP_DISABLED);
2527 rstp_port_set_mac_operational(rp, false);
2528 ofport->rstp_port = NULL;
2529 rstp_port_unref(rp);
2530 update_rstp_port_state(ofport);
2531 }
2532 return;
2533 }
2534
2535 /* Check if need to add a new port. */
2536 if (!rp) {
2537 rp = ofport->rstp_port = rstp_add_port(ofproto->rstp);
2538 }
2539
2540 rstp_port_set(rp, s->port_num, s->priority, s->path_cost,
2541 s->admin_edge_port, s->auto_edge,
2542 s->admin_p2p_mac_state, s->admin_port_state, s->mcheck,
2543 ofport);
2544 update_rstp_port_state(ofport);
2545 /* Synchronize operational status. */
2546 rstp_port_set_mac_operational(rp, ofport->may_enable);
2547 }
2548
2549 static void
2550 get_rstp_port_status(struct ofport *ofport_,
2551 struct ofproto_port_rstp_status *s)
2552 {
2553 struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
2554 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport->up.ofproto);
2555 struct rstp_port *rp = ofport->rstp_port;
2556
2557 if (!ofproto->rstp || !rp) {
2558 s->enabled = false;
2559 return;
2560 }
2561
2562 s->enabled = true;
2563 rstp_port_get_status(rp, &s->port_id, &s->state, &s->role,
2564 &s->designated_bridge_id, &s->designated_port_id,
2565 &s->designated_path_cost, &s->tx_count,
2566 &s->rx_count, &s->error_count, &s->uptime);
2567 }
2568
2569 \f
2570 static int
2571 set_queues(struct ofport *ofport_, const struct ofproto_port_queue *qdscp,
2572 size_t n_qdscp)
2573 {
2574 struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
2575 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport->up.ofproto);
2576
2577 if (ofport->n_qdscp != n_qdscp
2578 || (n_qdscp && memcmp(ofport->qdscp, qdscp,
2579 n_qdscp * sizeof *qdscp))) {
2580 ofproto->backer->need_revalidate = REV_RECONFIGURE;
2581 free(ofport->qdscp);
2582 ofport->qdscp = n_qdscp
2583 ? xmemdup(qdscp, n_qdscp * sizeof *qdscp)
2584 : NULL;
2585 ofport->n_qdscp = n_qdscp;
2586 }
2587
2588 return 0;
2589 }
2590 \f
2591 /* Bundles. */
2592
2593 /* Expires all MAC learning entries associated with 'bundle' and forces its
2594 * ofproto to revalidate every flow.
2595 *
2596 * Normally MAC learning entries are removed only from the ofproto associated
2597 * with 'bundle', but if 'all_ofprotos' is true, then the MAC learning entries
2598 * are removed from every ofproto. When patch ports and SLB bonds are in use
2599 * and a VM migration happens and the gratuitous ARPs are somehow lost, this
2600 * avoids a MAC_ENTRY_IDLE_TIME delay before the migrated VM can communicate
2601 * with the host from which it migrated. */
2602 static void
2603 bundle_flush_macs(struct ofbundle *bundle, bool all_ofprotos)
2604 {
2605 struct ofproto_dpif *ofproto = bundle->ofproto;
2606 struct mac_learning *ml = ofproto->ml;
2607 struct mac_entry *mac, *next_mac;
2608
2609 ofproto->backer->need_revalidate = REV_RECONFIGURE;
2610 ovs_rwlock_wrlock(&ml->rwlock);
2611 LIST_FOR_EACH_SAFE (mac, next_mac, lru_node, &ml->lrus) {
2612 if (mac_entry_get_port(ml, mac) == bundle) {
2613 if (all_ofprotos) {
2614 struct ofproto_dpif *o;
2615
2616 HMAP_FOR_EACH (o, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
2617 if (o != ofproto) {
2618 struct mac_entry *e;
2619
2620 ovs_rwlock_wrlock(&o->ml->rwlock);
2621 e = mac_learning_lookup(o->ml, mac->mac, mac->vlan);
2622 if (e) {
2623 mac_learning_expire(o->ml, e);
2624 }
2625 ovs_rwlock_unlock(&o->ml->rwlock);
2626 }
2627 }
2628 }
2629
2630 mac_learning_expire(ml, mac);
2631 }
2632 }
2633 ovs_rwlock_unlock(&ml->rwlock);
2634 }
2635
2636 static void
2637 bundle_move(struct ofbundle *old, struct ofbundle *new)
2638 {
2639 struct ofproto_dpif *ofproto = old->ofproto;
2640 struct mac_learning *ml = ofproto->ml;
2641 struct mac_entry *mac, *next_mac;
2642
2643 ovs_assert(new->ofproto == old->ofproto);
2644
2645 ofproto->backer->need_revalidate = REV_RECONFIGURE;
2646 ovs_rwlock_wrlock(&ml->rwlock);
2647 LIST_FOR_EACH_SAFE (mac, next_mac, lru_node, &ml->lrus) {
2648 if (mac_entry_get_port(ml, mac) == old) {
2649 mac_entry_set_port(ml, mac, new);
2650 }
2651 }
2652 ovs_rwlock_unlock(&ml->rwlock);
2653 }
2654
2655 static struct ofbundle *
2656 bundle_lookup(const struct ofproto_dpif *ofproto, void *aux)
2657 {
2658 struct ofbundle *bundle;
2659
2660 HMAP_FOR_EACH_IN_BUCKET (bundle, hmap_node, hash_pointer(aux, 0),
2661 &ofproto->bundles) {
2662 if (bundle->aux == aux) {
2663 return bundle;
2664 }
2665 }
2666 return NULL;
2667 }
2668
2669 static void
2670 bundle_update(struct ofbundle *bundle)
2671 {
2672 struct ofport_dpif *port;
2673
2674 bundle->floodable = true;
2675 LIST_FOR_EACH (port, bundle_node, &bundle->ports) {
2676 if (port->up.pp.config & OFPUTIL_PC_NO_FLOOD
2677 || port->is_layer3
2678 || (bundle->ofproto->stp && !stp_forward_in_state(port->stp_state))
2679 || (bundle->ofproto->rstp && !rstp_forward_in_state(port->rstp_state))) {
2680 bundle->floodable = false;
2681 break;
2682 }
2683 }
2684 }
2685
2686 static void
2687 bundle_del_port(struct ofport_dpif *port)
2688 {
2689 struct ofbundle *bundle = port->bundle;
2690
2691 bundle->ofproto->backer->need_revalidate = REV_RECONFIGURE;
2692
2693 ovs_list_remove(&port->bundle_node);
2694 port->bundle = NULL;
2695
2696 if (bundle->lacp) {
2697 lacp_slave_unregister(bundle->lacp, port);
2698 }
2699 if (bundle->bond) {
2700 bond_slave_unregister(bundle->bond, port);
2701 }
2702
2703 bundle_update(bundle);
2704 }
2705
2706 static bool
2707 bundle_add_port(struct ofbundle *bundle, ofp_port_t ofp_port,
2708 struct lacp_slave_settings *lacp)
2709 {
2710 struct ofport_dpif *port;
2711
2712 port = ofp_port_to_ofport(bundle->ofproto, ofp_port);
2713 if (!port) {
2714 return false;
2715 }
2716
2717 if (port->bundle != bundle) {
2718 bundle->ofproto->backer->need_revalidate = REV_RECONFIGURE;
2719 if (port->bundle) {
2720 bundle_remove(&port->up);
2721 }
2722
2723 port->bundle = bundle;
2724 ovs_list_push_back(&bundle->ports, &port->bundle_node);
2725 if (port->up.pp.config & OFPUTIL_PC_NO_FLOOD
2726 || port->is_layer3
2727 || (bundle->ofproto->stp && !stp_forward_in_state(port->stp_state))
2728 || (bundle->ofproto->rstp && !rstp_forward_in_state(port->rstp_state))) {
2729 bundle->floodable = false;
2730 }
2731 }
2732 if (lacp) {
2733 bundle->ofproto->backer->need_revalidate = REV_RECONFIGURE;
2734 lacp_slave_register(bundle->lacp, port, lacp);
2735 }
2736
2737 return true;
2738 }
2739
2740 static void
2741 bundle_destroy(struct ofbundle *bundle)
2742 {
2743 struct ofproto_dpif *ofproto;
2744 struct ofport_dpif *port, *next_port;
2745
2746 if (!bundle) {
2747 return;
2748 }
2749
2750 ofproto = bundle->ofproto;
2751 mbridge_unregister_bundle(ofproto->mbridge, bundle);
2752
2753 xlate_txn_start();
2754 xlate_bundle_remove(bundle);
2755 xlate_txn_commit();
2756
2757 LIST_FOR_EACH_SAFE (port, next_port, bundle_node, &bundle->ports) {
2758 bundle_del_port(port);
2759 }
2760
2761 bundle_flush_macs(bundle, true);
2762 hmap_remove(&ofproto->bundles, &bundle->hmap_node);
2763 free(bundle->name);
2764 free(bundle->trunks);
2765 lacp_unref(bundle->lacp);
2766 bond_unref(bundle->bond);
2767 free(bundle);
2768 }
2769
2770 static int
2771 bundle_set(struct ofproto *ofproto_, void *aux,
2772 const struct ofproto_bundle_settings *s)
2773 {
2774 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
2775 bool need_flush = false;
2776 struct ofport_dpif *port;
2777 struct ofbundle *bundle;
2778 unsigned long *trunks;
2779 int vlan;
2780 size_t i;
2781 bool ok;
2782
2783 if (!s) {
2784 bundle_destroy(bundle_lookup(ofproto, aux));
2785 return 0;
2786 }
2787
2788 ovs_assert(s->n_slaves == 1 || s->bond != NULL);
2789 ovs_assert((s->lacp != NULL) == (s->lacp_slaves != NULL));
2790
2791 bundle = bundle_lookup(ofproto, aux);
2792 if (!bundle) {
2793 bundle = xmalloc(sizeof *bundle);
2794
2795 bundle->ofproto = ofproto;
2796 hmap_insert(&ofproto->bundles, &bundle->hmap_node,
2797 hash_pointer(aux, 0));
2798 bundle->aux = aux;
2799 bundle->name = NULL;
2800
2801 ovs_list_init(&bundle->ports);
2802 bundle->vlan_mode = PORT_VLAN_TRUNK;
2803 bundle->vlan = -1;
2804 bundle->trunks = NULL;
2805 bundle->use_priority_tags = s->use_priority_tags;
2806 bundle->lacp = NULL;
2807 bundle->bond = NULL;
2808
2809 bundle->floodable = true;
2810 bundle->protected = false;
2811 mbridge_register_bundle(ofproto->mbridge, bundle);
2812 }
2813
2814 if (!bundle->name || strcmp(s->name, bundle->name)) {
2815 free(bundle->name);
2816 bundle->name = xstrdup(s->name);
2817 }
2818
2819 /* LACP. */
2820 if (s->lacp) {
2821 ofproto->lacp_enabled = true;
2822 if (!bundle->lacp) {
2823 ofproto->backer->need_revalidate = REV_RECONFIGURE;
2824 bundle->lacp = lacp_create();
2825 }
2826 lacp_configure(bundle->lacp, s->lacp);
2827 } else {
2828 lacp_unref(bundle->lacp);
2829 bundle->lacp = NULL;
2830 }
2831
2832 /* Update set of ports. */
2833 ok = true;
2834 for (i = 0; i < s->n_slaves; i++) {
2835 if (!bundle_add_port(bundle, s->slaves[i],
2836 s->lacp ? &s->lacp_slaves[i] : NULL)) {
2837 ok = false;
2838 }
2839 }
2840 if (!ok || ovs_list_size(&bundle->ports) != s->n_slaves) {
2841 struct ofport_dpif *next_port;
2842
2843 LIST_FOR_EACH_SAFE (port, next_port, bundle_node, &bundle->ports) {
2844 for (i = 0; i < s->n_slaves; i++) {
2845 if (s->slaves[i] == port->up.ofp_port) {
2846 goto found;
2847 }
2848 }
2849
2850 bundle_del_port(port);
2851 found: ;
2852 }
2853 }
2854 ovs_assert(ovs_list_size(&bundle->ports) <= s->n_slaves);
2855
2856 if (ovs_list_is_empty(&bundle->ports)) {
2857 bundle_destroy(bundle);
2858 return EINVAL;
2859 }
2860
2861 /* Set VLAN tagging mode */
2862 if (s->vlan_mode != bundle->vlan_mode
2863 || s->use_priority_tags != bundle->use_priority_tags) {
2864 bundle->vlan_mode = s->vlan_mode;
2865 bundle->use_priority_tags = s->use_priority_tags;
2866 need_flush = true;
2867 }
2868
2869 /* Set VLAN tag. */
2870 vlan = (s->vlan_mode == PORT_VLAN_TRUNK ? -1
2871 : s->vlan >= 0 && s->vlan <= 4095 ? s->vlan
2872 : 0);
2873 if (vlan != bundle->vlan) {
2874 bundle->vlan = vlan;
2875 need_flush = true;
2876 }
2877
2878 /* Get trunked VLANs. */
2879 switch (s->vlan_mode) {
2880 case PORT_VLAN_ACCESS:
2881 trunks = NULL;
2882 break;
2883
2884 case PORT_VLAN_TRUNK:
2885 trunks = CONST_CAST(unsigned long *, s->trunks);
2886 break;
2887
2888 case PORT_VLAN_NATIVE_UNTAGGED:
2889 case PORT_VLAN_NATIVE_TAGGED:
2890 if (vlan != 0 && (!s->trunks
2891 || !bitmap_is_set(s->trunks, vlan)
2892 || bitmap_is_set(s->trunks, 0))) {
2893 /* Force trunking the native VLAN and prohibit trunking VLAN 0. */
2894 if (s->trunks) {
2895 trunks = bitmap_clone(s->trunks, 4096);
2896 } else {
2897 trunks = bitmap_allocate1(4096);
2898 }
2899 bitmap_set1(trunks, vlan);
2900 bitmap_set0(trunks, 0);
2901 } else {
2902 trunks = CONST_CAST(unsigned long *, s->trunks);
2903 }
2904 break;
2905
2906 default:
2907 OVS_NOT_REACHED();
2908 }
2909 if (!vlan_bitmap_equal(trunks, bundle->trunks)) {
2910 free(bundle->trunks);
2911 if (trunks == s->trunks) {
2912 bundle->trunks = vlan_bitmap_clone(trunks);
2913 } else {
2914 bundle->trunks = trunks;
2915 trunks = NULL;
2916 }
2917 need_flush = true;
2918 }
2919 if (trunks != s->trunks) {
2920 free(trunks);
2921 }
2922
2923 /* Bonding. */
2924 if (!ovs_list_is_short(&bundle->ports)) {
2925 bundle->ofproto->has_bonded_bundles = true;
2926 if (bundle->bond) {
2927 if (bond_reconfigure(bundle->bond, s->bond)) {
2928 ofproto->backer->need_revalidate = REV_RECONFIGURE;
2929 }
2930 } else {
2931 bundle->bond = bond_create(s->bond, ofproto);
2932 ofproto->backer->need_revalidate = REV_RECONFIGURE;
2933 }
2934
2935 LIST_FOR_EACH (port, bundle_node, &bundle->ports) {
2936 bond_slave_register(bundle->bond, port,
2937 port->up.ofp_port, port->up.netdev);
2938 }
2939 } else {
2940 bond_unref(bundle->bond);
2941 bundle->bond = NULL;
2942 }
2943
2944 /* Set proteced port mode */
2945 if (s->protected != bundle->protected) {
2946 bundle->protected = s->protected;
2947 need_flush = true;
2948 }
2949
2950 /* If we changed something that would affect MAC learning, un-learn
2951 * everything on this port and force flow revalidation. */
2952 if (need_flush) {
2953 bundle_flush_macs(bundle, false);
2954 }
2955
2956 return 0;
2957 }
2958
2959 static void
2960 bundle_remove(struct ofport *port_)
2961 {
2962 struct ofport_dpif *port = ofport_dpif_cast(port_);
2963 struct ofbundle *bundle = port->bundle;
2964
2965 if (bundle) {
2966 bundle_del_port(port);
2967 if (ovs_list_is_empty(&bundle->ports)) {
2968 bundle_destroy(bundle);
2969 } else if (ovs_list_is_short(&bundle->ports)) {
2970 bond_unref(bundle->bond);
2971 bundle->bond = NULL;
2972 }
2973 }
2974 }
2975
2976 static void
2977 send_pdu_cb(void *port_, const void *pdu, size_t pdu_size)
2978 {
2979 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 10);
2980 struct ofport_dpif *port = port_;
2981 struct eth_addr ea;
2982 int error;
2983
2984 error = netdev_get_etheraddr(port->up.netdev, &ea);
2985 if (!error) {
2986 struct dp_packet packet;
2987 void *packet_pdu;
2988
2989 dp_packet_init(&packet, 0);
2990 packet_pdu = eth_compose(&packet, eth_addr_lacp, ea, ETH_TYPE_LACP,
2991 pdu_size);
2992 memcpy(packet_pdu, pdu, pdu_size);
2993
2994 ofproto_dpif_send_packet(port, false, &packet);
2995 dp_packet_uninit(&packet);
2996 } else {
2997 VLOG_ERR_RL(&rl, "port %s: cannot obtain Ethernet address of iface "
2998 "%s (%s)", port->bundle->name,
2999 netdev_get_name(port->up.netdev), ovs_strerror(error));
3000 }
3001 }
3002
3003 static void
3004 bundle_send_learning_packets(struct ofbundle *bundle)
3005 {
3006 struct ofproto_dpif *ofproto = bundle->ofproto;
3007 int error, n_packets, n_errors;
3008 struct mac_entry *e;
3009 struct pkt_list {
3010 struct ovs_list list_node;
3011 struct ofport_dpif *port;
3012 struct dp_packet *pkt;
3013 } *pkt_node;
3014 struct ovs_list packets;
3015
3016 ovs_list_init(&packets);
3017 ovs_rwlock_rdlock(&ofproto->ml->rwlock);
3018 LIST_FOR_EACH (e, lru_node, &ofproto->ml->lrus) {
3019 if (mac_entry_get_port(ofproto->ml, e) != bundle) {
3020 pkt_node = xmalloc(sizeof *pkt_node);
3021 pkt_node->pkt = bond_compose_learning_packet(bundle->bond,
3022 e->mac, e->vlan,
3023 (void **)&pkt_node->port);
3024 ovs_list_push_back(&packets, &pkt_node->list_node);
3025 }
3026 }
3027 ovs_rwlock_unlock(&ofproto->ml->rwlock);
3028
3029 error = n_packets = n_errors = 0;
3030 LIST_FOR_EACH_POP (pkt_node, list_node, &packets) {
3031 int ret;
3032
3033 ret = ofproto_dpif_send_packet(pkt_node->port, false, pkt_node->pkt);
3034 dp_packet_delete(pkt_node->pkt);
3035 free(pkt_node);
3036 if (ret) {
3037 error = ret;
3038 n_errors++;
3039 }
3040 n_packets++;
3041 }
3042
3043 if (n_errors) {
3044 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
3045 VLOG_WARN_RL(&rl, "bond %s: %d errors sending %d gratuitous learning "
3046 "packets, last error was: %s",
3047 bundle->name, n_errors, n_packets, ovs_strerror(error));
3048 } else {
3049 VLOG_DBG("bond %s: sent %d gratuitous learning packets",
3050 bundle->name, n_packets);
3051 }
3052 }
3053
3054 static void
3055 bundle_run(struct ofbundle *bundle)
3056 {
3057 if (bundle->lacp) {
3058 lacp_run(bundle->lacp, send_pdu_cb);
3059 }
3060 if (bundle->bond) {
3061 struct ofport_dpif *port;
3062
3063 LIST_FOR_EACH (port, bundle_node, &bundle->ports) {
3064 bond_slave_set_may_enable(bundle->bond, port, port->may_enable);
3065 }
3066
3067 if (bond_run(bundle->bond, lacp_status(bundle->lacp))) {
3068 bundle->ofproto->backer->need_revalidate = REV_BOND;
3069 }
3070
3071 if (bond_should_send_learning_packets(bundle->bond)) {
3072 bundle_send_learning_packets(bundle);
3073 }
3074 }
3075 }
3076
3077 static void
3078 bundle_wait(struct ofbundle *bundle)
3079 {
3080 if (bundle->lacp) {
3081 lacp_wait(bundle->lacp);
3082 }
3083 if (bundle->bond) {
3084 bond_wait(bundle->bond);
3085 }
3086 }
3087 \f
3088 /* Mirrors. */
3089
3090 static int
3091 mirror_set__(struct ofproto *ofproto_, void *aux,
3092 const struct ofproto_mirror_settings *s)
3093 {
3094 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
3095 struct ofbundle **srcs, **dsts;
3096 int error;
3097 size_t i;
3098
3099 if (!s) {
3100 mirror_destroy(ofproto->mbridge, aux);
3101 return 0;
3102 }
3103
3104 srcs = xmalloc(s->n_srcs * sizeof *srcs);
3105 dsts = xmalloc(s->n_dsts * sizeof *dsts);
3106
3107 for (i = 0; i < s->n_srcs; i++) {
3108 srcs[i] = bundle_lookup(ofproto, s->srcs[i]);
3109 }
3110
3111 for (i = 0; i < s->n_dsts; i++) {
3112 dsts[i] = bundle_lookup(ofproto, s->dsts[i]);
3113 }
3114
3115 error = mirror_set(ofproto->mbridge, aux, s->name, srcs, s->n_srcs, dsts,
3116 s->n_dsts, s->src_vlans,
3117 bundle_lookup(ofproto, s->out_bundle),
3118 s->snaplen, s->out_vlan);
3119 free(srcs);
3120 free(dsts);
3121 return error;
3122 }
3123
3124 static int
3125 mirror_get_stats__(struct ofproto *ofproto, void *aux,
3126 uint64_t *packets, uint64_t *bytes)
3127 {
3128 return mirror_get_stats(ofproto_dpif_cast(ofproto)->mbridge, aux, packets,
3129 bytes);
3130 }
3131
3132 static int
3133 set_flood_vlans(struct ofproto *ofproto_, unsigned long *flood_vlans)
3134 {
3135 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
3136 ovs_rwlock_wrlock(&ofproto->ml->rwlock);
3137 if (mac_learning_set_flood_vlans(ofproto->ml, flood_vlans)) {
3138 mac_learning_flush(ofproto->ml);
3139 }
3140 ovs_rwlock_unlock(&ofproto->ml->rwlock);
3141 return 0;
3142 }
3143
3144 static bool
3145 is_mirror_output_bundle(const struct ofproto *ofproto_, void *aux)
3146 {
3147 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
3148 struct ofbundle *bundle = bundle_lookup(ofproto, aux);
3149 return bundle && mirror_bundle_out(ofproto->mbridge, bundle) != 0;
3150 }
3151
3152 static void
3153 forward_bpdu_changed(struct ofproto *ofproto_)
3154 {
3155 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
3156 ofproto->backer->need_revalidate = REV_RECONFIGURE;
3157 }
3158
3159 static void
3160 set_mac_table_config(struct ofproto *ofproto_, unsigned int idle_time,
3161 size_t max_entries)
3162 {
3163 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
3164 ovs_rwlock_wrlock(&ofproto->ml->rwlock);
3165 mac_learning_set_idle_time(ofproto->ml, idle_time);
3166 mac_learning_set_max_entries(ofproto->ml, max_entries);
3167 ovs_rwlock_unlock(&ofproto->ml->rwlock);
3168 }
3169
3170 /* Configures multicast snooping on 'ofport' using the settings
3171 * defined in 's'. */
3172 static int
3173 set_mcast_snooping(struct ofproto *ofproto_,
3174 const struct ofproto_mcast_snooping_settings *s)
3175 {
3176 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
3177
3178 /* Only revalidate flows if the configuration changed. */
3179 if (!s != !ofproto->ms) {
3180 ofproto->backer->need_revalidate = REV_RECONFIGURE;
3181 }
3182
3183 if (s) {
3184 if (!ofproto->ms) {
3185 ofproto->ms = mcast_snooping_create();
3186 }
3187
3188 ovs_rwlock_wrlock(&ofproto->ms->rwlock);
3189 mcast_snooping_set_idle_time(ofproto->ms, s->idle_time);
3190 mcast_snooping_set_max_entries(ofproto->ms, s->max_entries);
3191 if (mcast_snooping_set_flood_unreg(ofproto->ms, s->flood_unreg)) {
3192 ofproto->backer->need_revalidate = REV_RECONFIGURE;
3193 }
3194 ovs_rwlock_unlock(&ofproto->ms->rwlock);
3195 } else {
3196 mcast_snooping_unref(ofproto->ms);
3197 ofproto->ms = NULL;
3198 }
3199
3200 return 0;
3201 }
3202
3203 /* Configures multicast snooping port's flood settings on 'ofproto'. */
3204 static int
3205 set_mcast_snooping_port(struct ofproto *ofproto_, void *aux,
3206 const struct ofproto_mcast_snooping_port_settings *s)
3207 {
3208 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
3209 struct ofbundle *bundle = bundle_lookup(ofproto, aux);
3210
3211 if (ofproto->ms && s) {
3212 ovs_rwlock_wrlock(&ofproto->ms->rwlock);
3213 mcast_snooping_set_port_flood(ofproto->ms, bundle, s->flood);
3214 mcast_snooping_set_port_flood_reports(ofproto->ms, bundle,
3215 s->flood_reports);
3216 ovs_rwlock_unlock(&ofproto->ms->rwlock);
3217 }
3218 return 0;
3219 }
3220
3221 \f
3222 /* Ports. */
3223
3224 struct ofport_dpif *
3225 ofp_port_to_ofport(const struct ofproto_dpif *ofproto, ofp_port_t ofp_port)
3226 {
3227 struct ofport *ofport = ofproto_get_port(&ofproto->up, ofp_port);
3228 return ofport ? ofport_dpif_cast(ofport) : NULL;
3229 }
3230
3231 static void
3232 ofproto_port_from_dpif_port(struct ofproto_dpif *ofproto,
3233 struct ofproto_port *ofproto_port,
3234 struct dpif_port *dpif_port)
3235 {
3236 ofproto_port->name = dpif_port->name;
3237 ofproto_port->type = dpif_port->type;
3238 ofproto_port->ofp_port = odp_port_to_ofp_port(ofproto, dpif_port->port_no);
3239 }
3240
3241 static void
3242 ofport_update_peer(struct ofport_dpif *ofport)
3243 {
3244 const struct ofproto_dpif *ofproto;
3245 struct dpif_backer *backer;
3246 char *peer_name;
3247
3248 if (!netdev_vport_is_patch(ofport->up.netdev)) {
3249 return;
3250 }
3251
3252 backer = ofproto_dpif_cast(ofport->up.ofproto)->backer;
3253 backer->need_revalidate = REV_RECONFIGURE;
3254
3255 if (ofport->peer) {
3256 ofport->peer->peer = NULL;
3257 ofport->peer = NULL;
3258 }
3259
3260 peer_name = netdev_vport_patch_peer(ofport->up.netdev);
3261 if (!peer_name) {
3262 return;
3263 }
3264
3265 HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
3266 struct ofport *peer_ofport;
3267 struct ofport_dpif *peer;
3268 char *peer_peer;
3269
3270 if (ofproto->backer != backer) {
3271 continue;
3272 }
3273
3274 peer_ofport = shash_find_data(&ofproto->up.port_by_name, peer_name);
3275 if (!peer_ofport) {
3276 continue;
3277 }
3278
3279 peer = ofport_dpif_cast(peer_ofport);
3280 peer_peer = netdev_vport_patch_peer(peer->up.netdev);
3281 if (peer_peer && !strcmp(netdev_get_name(ofport->up.netdev),
3282 peer_peer)) {
3283 ofport->peer = peer;
3284 ofport->peer->peer = ofport;
3285 }
3286 free(peer_peer);
3287
3288 break;
3289 }
3290 free(peer_name);
3291 }
3292
3293 static void
3294 port_run(struct ofport_dpif *ofport)
3295 {
3296 long long int carrier_seq = netdev_get_carrier_resets(ofport->up.netdev);
3297 bool carrier_changed = carrier_seq != ofport->carrier_seq;
3298 bool enable = netdev_get_carrier(ofport->up.netdev);
3299 bool cfm_enable = false;
3300 bool bfd_enable = false;
3301
3302 ofport->carrier_seq = carrier_seq;
3303
3304 if (ofport->cfm) {
3305 int cfm_opup = cfm_get_opup(ofport->cfm);
3306
3307 cfm_enable = !cfm_get_fault(ofport->cfm);
3308
3309 if (cfm_opup >= 0) {
3310 cfm_enable = cfm_enable && cfm_opup;
3311 }
3312 }
3313
3314 if (ofport->bfd) {
3315 bfd_enable = bfd_forwarding(ofport->bfd);
3316 }
3317
3318 if (ofport->bfd || ofport->cfm) {
3319 enable = enable && (cfm_enable || bfd_enable);
3320 }
3321
3322 if (ofport->bundle) {
3323 enable = enable && lacp_slave_may_enable(ofport->bundle->lacp, ofport);
3324 if (carrier_changed) {
3325 lacp_slave_carrier_changed(ofport->bundle->lacp, ofport);
3326 }
3327 }
3328
3329 if (ofport->may_enable != enable) {
3330 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport->up.ofproto);
3331
3332 ofproto->backer->need_revalidate = REV_PORT_TOGGLED;
3333
3334 if (ofport->rstp_port) {
3335 rstp_port_set_mac_operational(ofport->rstp_port, enable);
3336 }
3337 }
3338
3339 ofport->may_enable = enable;
3340 }
3341
3342 static int
3343 port_query_by_name(const struct ofproto *ofproto_, const char *devname,
3344 struct ofproto_port *ofproto_port)
3345 {
3346 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
3347 struct dpif_port dpif_port;
3348 int error;
3349
3350 if (sset_contains(&ofproto->ghost_ports, devname)) {
3351 const char *type = netdev_get_type_from_name(devname);
3352
3353 /* We may be called before ofproto->up.port_by_name is populated with
3354 * the appropriate ofport. For this reason, we must get the name and
3355 * type from the netdev layer directly. */
3356 if (type) {
3357 const struct ofport *ofport;
3358
3359 ofport = shash_find_data(&ofproto->up.port_by_name, devname);
3360 ofproto_port->ofp_port = ofport ? ofport->ofp_port : OFPP_NONE;
3361 ofproto_port->name = xstrdup(devname);
3362 ofproto_port->type = xstrdup(type);
3363 return 0;
3364 }
3365 return ENODEV;
3366 }
3367
3368 if (!sset_contains(&ofproto->ports, devname)) {
3369 return ENODEV;
3370 }
3371 error = dpif_port_query_by_name(ofproto->backer->dpif,
3372 devname, &dpif_port);
3373 if (!error) {
3374 ofproto_port_from_dpif_port(ofproto, ofproto_port, &dpif_port);
3375 }
3376 return error;
3377 }
3378
3379 static int
3380 port_add(struct ofproto *ofproto_, struct netdev *netdev)
3381 {
3382 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
3383 const char *devname = netdev_get_name(netdev);
3384 char namebuf[NETDEV_VPORT_NAME_BUFSIZE];
3385 const char *dp_port_name;
3386
3387 if (netdev_vport_is_patch(netdev)) {
3388 sset_add(&ofproto->ghost_ports, netdev_get_name(netdev));
3389 return 0;
3390 }
3391
3392 dp_port_name = netdev_vport_get_dpif_port(netdev, namebuf, sizeof namebuf);
3393 if (!dpif_port_exists(ofproto->backer->dpif, dp_port_name)) {
3394 odp_port_t port_no = ODPP_NONE;
3395 int error;
3396
3397 error = dpif_port_add(ofproto->backer->dpif, netdev, &port_no);
3398 if (error) {
3399 return error;
3400 }
3401 if (netdev_get_tunnel_config(netdev)) {
3402 simap_put(&ofproto->backer->tnl_backers,
3403 dp_port_name, odp_to_u32(port_no));
3404 }
3405 }
3406
3407 if (netdev_get_tunnel_config(netdev)) {
3408 sset_add(&ofproto->ghost_ports, devname);
3409 } else {
3410 sset_add(&ofproto->ports, devname);
3411 }
3412 return 0;
3413 }
3414
3415 static int
3416 port_del(struct ofproto *ofproto_, ofp_port_t ofp_port)
3417 {
3418 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
3419 struct ofport_dpif *ofport = ofp_port_to_ofport(ofproto, ofp_port);
3420 int error = 0;
3421
3422 if (!ofport) {
3423 return 0;
3424 }
3425
3426 sset_find_and_delete(&ofproto->ghost_ports,
3427 netdev_get_name(ofport->up.netdev));
3428 ofproto->backer->need_revalidate = REV_RECONFIGURE;
3429 if (!ofport->is_tunnel && !netdev_vport_is_patch(ofport->up.netdev)) {
3430 error = dpif_port_del(ofproto->backer->dpif, ofport->odp_port);
3431 if (!error) {
3432 /* The caller is going to close ofport->up.netdev. If this is a
3433 * bonded port, then the bond is using that netdev, so remove it
3434 * from the bond. The client will need to reconfigure everything
3435 * after deleting ports, so then the slave will get re-added. */
3436 bundle_remove(&ofport->up);
3437 }
3438 }
3439 return error;
3440 }
3441
3442 static int
3443 port_set_config(const struct ofport *ofport_, const struct smap *cfg)
3444 {
3445 struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
3446 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport->up.ofproto);
3447
3448 if (sset_contains(&ofproto->ghost_ports,
3449 netdev_get_name(ofport->up.netdev))) {
3450 return 0;
3451 }
3452
3453 return dpif_port_set_config(ofproto->backer->dpif, ofport->odp_port, cfg);
3454 }
3455
3456 static int
3457 port_get_stats(const struct ofport *ofport_, struct netdev_stats *stats)
3458 {
3459 struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
3460 int error;
3461
3462 error = netdev_get_stats(ofport->up.netdev, stats);
3463
3464 if (!error && ofport_->ofp_port == OFPP_LOCAL) {
3465 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport->up.ofproto);
3466
3467 ovs_mutex_lock(&ofproto->stats_mutex);
3468 /* ofproto->stats.tx_packets represents packets that we created
3469 * internally and sent to some port (e.g. packets sent with
3470 * ofproto_dpif_send_packet()). Account for them as if they had
3471 * come from OFPP_LOCAL and got forwarded. */
3472
3473 if (stats->rx_packets != UINT64_MAX) {
3474 stats->rx_packets += ofproto->stats.tx_packets;
3475 }
3476
3477 if (stats->rx_bytes != UINT64_MAX) {
3478 stats->rx_bytes += ofproto->stats.tx_bytes;
3479 }
3480
3481 /* ofproto->stats.rx_packets represents packets that were received on
3482 * some port and we processed internally and dropped (e.g. STP).
3483 * Account for them as if they had been forwarded to OFPP_LOCAL. */
3484
3485 if (stats->tx_packets != UINT64_MAX) {
3486 stats->tx_packets += ofproto->stats.rx_packets;
3487 }
3488
3489 if (stats->tx_bytes != UINT64_MAX) {
3490 stats->tx_bytes += ofproto->stats.rx_bytes;
3491 }
3492 ovs_mutex_unlock(&ofproto->stats_mutex);
3493 }
3494
3495 return error;
3496 }
3497
3498 static int
3499 port_get_lacp_stats(const struct ofport *ofport_, struct lacp_slave_stats *stats)
3500 {
3501 struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
3502 if (ofport->bundle && ofport->bundle->lacp) {
3503 if (lacp_get_slave_stats(ofport->bundle->lacp, ofport, stats)) {
3504 return 0;
3505 }
3506 }
3507 return -1;
3508 }
3509
3510 struct port_dump_state {
3511 struct sset_position pos;
3512 bool ghost;
3513
3514 struct ofproto_port port;
3515 bool has_port;
3516 };
3517
3518 static int
3519 port_dump_start(const struct ofproto *ofproto_ OVS_UNUSED, void **statep)
3520 {
3521 *statep = xzalloc(sizeof(struct port_dump_state));
3522 return 0;
3523 }
3524
3525 static int
3526 port_dump_next(const struct ofproto *ofproto_, void *state_,
3527 struct ofproto_port *port)
3528 {
3529 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
3530 struct port_dump_state *state = state_;
3531 const struct sset *sset;
3532 struct sset_node *node;
3533
3534 if (state->has_port) {
3535 ofproto_port_destroy(&state->port);
3536 state->has_port = false;
3537 }
3538 sset = state->ghost ? &ofproto->ghost_ports : &ofproto->ports;
3539 while ((node = sset_at_position(sset, &state->pos))) {
3540 int error;
3541
3542 error = port_query_by_name(ofproto_, node->name, &state->port);
3543 if (!error) {
3544 *port = state->port;
3545 state->has_port = true;
3546 return 0;
3547 } else if (error != ENODEV) {
3548 return error;
3549 }
3550 }
3551
3552 if (!state->ghost) {
3553 state->ghost = true;
3554 memset(&state->pos, 0, sizeof state->pos);
3555 return port_dump_next(ofproto_, state_, port);
3556 }
3557
3558 return EOF;
3559 }
3560
3561 static int
3562 port_dump_done(const struct ofproto *ofproto_ OVS_UNUSED, void *state_)
3563 {
3564 struct port_dump_state *state = state_;
3565
3566 if (state->has_port) {
3567 ofproto_port_destroy(&state->port);
3568 }
3569 free(state);
3570 return 0;
3571 }
3572
3573 static int
3574 port_poll(const struct ofproto *ofproto_, char **devnamep)
3575 {
3576 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
3577
3578 if (ofproto->port_poll_errno) {
3579 int error = ofproto->port_poll_errno;
3580 ofproto->port_poll_errno = 0;
3581 return error;
3582 }
3583
3584 if (sset_is_empty(&ofproto->port_poll_set)) {
3585 return EAGAIN;
3586 }
3587
3588 *devnamep = sset_pop(&ofproto->port_poll_set);
3589 return 0;
3590 }
3591
3592 static void
3593 port_poll_wait(const struct ofproto *ofproto_)
3594 {
3595 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
3596 dpif_port_poll_wait(ofproto->backer->dpif);
3597 }
3598
3599 static int
3600 port_is_lacp_current(const struct ofport *ofport_)
3601 {
3602 const struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
3603 return (ofport->bundle && ofport->bundle->lacp
3604 ? lacp_slave_is_current(ofport->bundle->lacp, ofport)
3605 : -1);
3606 }
3607 \f
3608 /* If 'rule' is an OpenFlow rule, that has expired according to OpenFlow rules,
3609 * then delete it entirely. */
3610 static void
3611 rule_expire(struct rule_dpif *rule, long long now)
3612 OVS_REQUIRES(ofproto_mutex)
3613 {
3614 uint16_t hard_timeout, idle_timeout;
3615 int reason = -1;
3616
3617 hard_timeout = rule->up.hard_timeout;
3618 idle_timeout = rule->up.idle_timeout;
3619
3620 /* Has 'rule' expired? */
3621 if (hard_timeout) {
3622 long long int modified;
3623
3624 ovs_mutex_lock(&rule->up.mutex);
3625 modified = rule->up.modified;
3626 ovs_mutex_unlock(&rule->up.mutex);
3627
3628 if (now > modified + hard_timeout * 1000) {
3629 reason = OFPRR_HARD_TIMEOUT;
3630 }
3631 }
3632
3633 if (reason < 0 && idle_timeout) {
3634 long long int used;
3635
3636 ovs_mutex_lock(&rule->stats_mutex);
3637 used = rule->stats.used;
3638 ovs_mutex_unlock(&rule->stats_mutex);
3639
3640 if (now > used + idle_timeout * 1000) {
3641 reason = OFPRR_IDLE_TIMEOUT;
3642 }
3643 }
3644
3645 if (reason >= 0) {
3646 COVERAGE_INC(ofproto_dpif_expired);
3647 ofproto_rule_expire(&rule->up, reason);
3648 }
3649 }
3650
3651 static void
3652 ofproto_dpif_set_packet_odp_port(const struct ofproto_dpif *ofproto,
3653 ofp_port_t in_port, struct dp_packet *packet)
3654 {
3655 if (in_port == OFPP_NONE) {
3656 in_port = OFPP_LOCAL;
3657 }
3658 packet->md.in_port.odp_port = ofp_port_to_odp_port(ofproto, in_port);
3659 }
3660
3661 int
3662 ofproto_dpif_execute_actions__(struct ofproto_dpif *ofproto,
3663 ovs_version_t version, const struct flow *flow,
3664 struct rule_dpif *rule,
3665 const struct ofpact *ofpacts, size_t ofpacts_len,
3666 int depth, int resubmits,
3667 struct dp_packet *packet)
3668 {
3669 struct dpif_flow_stats stats;
3670 struct xlate_out xout;
3671 struct xlate_in xin;
3672 struct dpif_execute execute;
3673 int error;
3674
3675 ovs_assert((rule != NULL) != (ofpacts != NULL));
3676
3677 dpif_flow_stats_extract(flow, packet, time_msec(), &stats);
3678
3679 if (rule) {
3680 rule_dpif_credit_stats(rule, &stats);
3681 }
3682
3683 uint64_t odp_actions_stub[1024 / 8];
3684 struct ofpbuf odp_actions = OFPBUF_STUB_INITIALIZER(odp_actions_stub);
3685 xlate_in_init(&xin, ofproto, version, flow, flow->in_port.ofp_port, rule,
3686 stats.tcp_flags, packet, NULL, &odp_actions);
3687 xin.ofpacts = ofpacts;
3688 xin.ofpacts_len = ofpacts_len;
3689 xin.resubmit_stats = &stats;
3690 xin.depth = depth;
3691 xin.resubmits = resubmits;
3692 if (xlate_actions(&xin, &xout) != XLATE_OK) {
3693 error = EINVAL;
3694 goto out;
3695 }
3696
3697 execute.actions = odp_actions.data;
3698 execute.actions_len = odp_actions.size;
3699
3700 pkt_metadata_from_flow(&packet->md, flow);
3701 execute.packet = packet;
3702 execute.flow = flow;
3703 execute.needs_help = (xout.slow & SLOW_ACTION) != 0;
3704 execute.probe = false;
3705 execute.mtu = 0;
3706
3707 /* Fix up in_port. */
3708 ofproto_dpif_set_packet_odp_port(ofproto, flow->in_port.ofp_port, packet);
3709
3710 error = dpif_execute(ofproto->backer->dpif, &execute);
3711 out:
3712 xlate_out_uninit(&xout);
3713 ofpbuf_uninit(&odp_actions);
3714
3715 return error;
3716 }
3717
3718 /* Executes, within 'ofproto', the actions in 'rule' or 'ofpacts' on 'packet'.
3719 * 'flow' must reflect the data in 'packet'. */
3720 int
3721 ofproto_dpif_execute_actions(struct ofproto_dpif *ofproto,
3722 ovs_version_t version, const struct flow *flow,
3723 struct rule_dpif *rule,
3724 const struct ofpact *ofpacts, size_t ofpacts_len,
3725 struct dp_packet *packet)
3726 {
3727 return ofproto_dpif_execute_actions__(ofproto, version, flow, rule,
3728 ofpacts, ofpacts_len, 0, 0, packet);
3729 }
3730
3731 static void
3732 rule_dpif_credit_stats__(struct rule_dpif *rule,
3733 const struct dpif_flow_stats *stats,
3734 bool credit_counts)
3735 OVS_REQUIRES(rule->stats_mutex)
3736 {
3737 if (credit_counts) {
3738 rule->stats.n_packets += stats->n_packets;
3739 rule->stats.n_bytes += stats->n_bytes;
3740 }
3741 rule->stats.used = MAX(rule->stats.used, stats->used);
3742 }
3743
3744 void
3745 rule_dpif_credit_stats(struct rule_dpif *rule,
3746 const struct dpif_flow_stats *stats)
3747 {
3748 ovs_mutex_lock(&rule->stats_mutex);
3749 if (OVS_UNLIKELY(rule->new_rule)) {
3750 ovs_mutex_lock(&rule->new_rule->stats_mutex);
3751 rule_dpif_credit_stats__(rule->new_rule, stats, rule->forward_counts);
3752 ovs_mutex_unlock(&rule->new_rule->stats_mutex);
3753 } else {
3754 rule_dpif_credit_stats__(rule, stats, true);
3755 }
3756 ovs_mutex_unlock(&rule->stats_mutex);
3757 }
3758
3759 /* Sets 'rule''s recirculation id. */
3760 static void
3761 rule_dpif_set_recirc_id(struct rule_dpif *rule, uint32_t id)
3762 OVS_REQUIRES(rule->up.mutex)
3763 {
3764 ovs_assert(!rule->recirc_id || rule->recirc_id == id);
3765 if (rule->recirc_id == id) {
3766 /* Release the new reference to the same id. */
3767 recirc_free_id(id);
3768 } else {
3769 rule->recirc_id = id;
3770 }
3771 }
3772
3773 /* Sets 'rule''s recirculation id. */
3774 void
3775 rule_set_recirc_id(struct rule *rule_, uint32_t id)
3776 {
3777 struct rule_dpif *rule = rule_dpif_cast(rule_);
3778
3779 ovs_mutex_lock(&rule->up.mutex);
3780 rule_dpif_set_recirc_id(rule, id);
3781 ovs_mutex_unlock(&rule->up.mutex);
3782 }
3783
3784 ovs_version_t
3785 ofproto_dpif_get_tables_version(struct ofproto_dpif *ofproto)
3786 {
3787 ovs_version_t version;
3788
3789 /* Use memory_order_acquire to signify that any following memory accesses
3790 * can not be reordered to happen before this atomic read. This makes sure
3791 * all following reads relate to this or a newer version, but never to an
3792 * older version. */
3793 atomic_read_explicit(&ofproto->tables_version, &version,
3794 memory_order_acquire);
3795 return version;
3796 }
3797
3798 /* The returned rule (if any) is valid at least until the next RCU quiescent
3799 * period. If the rule needs to stay around longer, the caller should take
3800 * a reference.
3801 *
3802 * 'flow' is non-const to allow for temporary modifications during the lookup.
3803 * Any changes are restored before returning. */
3804 static struct rule_dpif *
3805 rule_dpif_lookup_in_table(struct ofproto_dpif *ofproto, ovs_version_t version,
3806 uint8_t table_id, struct flow *flow,
3807 struct flow_wildcards *wc)
3808 {
3809 struct classifier *cls = &ofproto->up.tables[table_id].cls;
3810 return rule_dpif_cast(rule_from_cls_rule(classifier_lookup(cls, version,
3811 flow, wc)));
3812 }
3813
3814 void
3815 ofproto_dpif_credit_table_stats(struct ofproto_dpif *ofproto, uint8_t table_id,
3816 uint64_t n_matches, uint64_t n_misses)
3817 {
3818 struct oftable *tbl = &ofproto->up.tables[table_id];
3819 unsigned long orig;
3820
3821 if (n_matches) {
3822 atomic_add_relaxed(&tbl->n_matched, n_matches, &orig);
3823 }
3824 if (n_misses) {
3825 atomic_add_relaxed(&tbl->n_missed, n_misses, &orig);
3826 }
3827 }
3828
3829 /* Look up 'flow' in 'ofproto''s classifier version 'version', starting from
3830 * table '*table_id'. Returns the rule that was found, which may be one of the
3831 * special rules according to packet miss hadling. If 'may_packet_in' is
3832 * false, returning of the miss_rule (which issues packet ins for the
3833 * controller) is avoided. Updates 'wc', if nonnull, to reflect the fields
3834 * that were used during the lookup.
3835 *
3836 * If 'honor_table_miss' is true, the first lookup occurs in '*table_id', but
3837 * if none is found then the table miss configuration for that table is
3838 * honored, which can result in additional lookups in other OpenFlow tables.
3839 * In this case the function updates '*table_id' to reflect the final OpenFlow
3840 * table that was searched.
3841 *
3842 * If 'honor_table_miss' is false, then only one table lookup occurs, in
3843 * '*table_id'.
3844 *
3845 * The rule is returned in '*rule', which is valid at least until the next
3846 * RCU quiescent period. If the '*rule' needs to stay around longer, the
3847 * caller must take a reference.
3848 *
3849 * 'in_port' allows the lookup to take place as if the in port had the value
3850 * 'in_port'. This is needed for resubmit action support.
3851 *
3852 * 'flow' is non-const to allow for temporary modifications during the lookup.
3853 * Any changes are restored before returning. */
3854 struct rule_dpif *
3855 rule_dpif_lookup_from_table(struct ofproto_dpif *ofproto,
3856 ovs_version_t version, struct flow *flow,
3857 struct flow_wildcards *wc,
3858 const struct dpif_flow_stats *stats,
3859 uint8_t *table_id, ofp_port_t in_port,
3860 bool may_packet_in, bool honor_table_miss,
3861 struct xlate_cache *xcache)
3862 {
3863 ovs_be16 old_tp_src = flow->tp_src, old_tp_dst = flow->tp_dst;
3864 ofp_port_t old_in_port = flow->in_port.ofp_port;
3865 enum ofputil_table_miss miss_config;
3866 struct rule_dpif *rule;
3867 uint8_t next_id;
3868
3869 /* We always unwildcard nw_frag (for IP), so they
3870 * need not be unwildcarded here. */
3871 if (flow->nw_frag & FLOW_NW_FRAG_ANY
3872 && ofproto->up.frag_handling != OFPUTIL_FRAG_NX_MATCH) {
3873 if (ofproto->up.frag_handling == OFPUTIL_FRAG_NORMAL) {
3874 /* We must pretend that transport ports are unavailable. */
3875 flow->tp_src = htons(0);
3876 flow->tp_dst = htons(0);
3877 } else {
3878 /* Must be OFPUTIL_FRAG_DROP (we don't have OFPUTIL_FRAG_REASM).
3879 * Use the drop_frags_rule (which cannot disappear). */
3880 rule = ofproto->drop_frags_rule;
3881 if (stats) {
3882 struct oftable *tbl = &ofproto->up.tables[*table_id];
3883 unsigned long orig;
3884
3885 atomic_add_relaxed(&tbl->n_matched, stats->n_packets, &orig);
3886 }
3887 if (xcache) {
3888 struct xc_entry *entry;
3889
3890 entry = xlate_cache_add_entry(xcache, XC_TABLE);
3891 entry->table.ofproto = ofproto;
3892 entry->table.id = *table_id;
3893 entry->table.match = true;
3894 }
3895 return rule;
3896 }
3897 }
3898
3899 /* Look up a flow with 'in_port' as the input port. Then restore the
3900 * original input port (otherwise OFPP_NORMAL and OFPP_IN_PORT will
3901 * have surprising behavior). */
3902 flow->in_port.ofp_port = in_port;
3903
3904 /* Our current implementation depends on n_tables == N_TABLES, and
3905 * TBL_INTERNAL being the last table. */
3906 BUILD_ASSERT_DECL(N_TABLES == TBL_INTERNAL + 1);
3907
3908 miss_config = OFPUTIL_TABLE_MISS_CONTINUE;
3909
3910 for (next_id = *table_id;
3911 next_id < ofproto->up.n_tables;
3912 next_id++, next_id += (next_id == TBL_INTERNAL))
3913 {
3914 *table_id = next_id;
3915 rule = rule_dpif_lookup_in_table(ofproto, version, next_id, flow, wc);
3916 if (stats) {
3917 struct oftable *tbl = &ofproto->up.tables[next_id];
3918 unsigned long orig;
3919
3920 atomic_add_relaxed(rule ? &tbl->n_matched : &tbl->n_missed,
3921 stats->n_packets, &orig);
3922 }
3923 if (xcache) {
3924 struct xc_entry *entry;
3925
3926 entry = xlate_cache_add_entry(xcache, XC_TABLE);
3927 entry->table.ofproto = ofproto;
3928 entry->table.id = next_id;
3929 entry->table.match = (rule != NULL);
3930 }
3931 if (rule) {
3932 goto out; /* Match. */
3933 }
3934 if (honor_table_miss) {
3935 miss_config = ofproto_table_get_miss_config(&ofproto->up,
3936 *table_id);
3937 if (miss_config == OFPUTIL_TABLE_MISS_CONTINUE) {
3938 continue;
3939 }
3940 }
3941 break;
3942 }
3943 /* Miss. */
3944 rule = ofproto->no_packet_in_rule;
3945 if (may_packet_in) {
3946 if (miss_config == OFPUTIL_TABLE_MISS_CONTINUE
3947 || miss_config == OFPUTIL_TABLE_MISS_CONTROLLER) {
3948 struct ofport_dpif *port;
3949
3950 port = ofp_port_to_ofport(ofproto, old_in_port);
3951 if (!port) {
3952 VLOG_WARN_RL(&rl, "packet-in on unknown OpenFlow port %"PRIu32,
3953 old_in_port);
3954 } else if (!(port->up.pp.config & OFPUTIL_PC_NO_PACKET_IN)) {
3955 rule = ofproto->miss_rule;
3956 }
3957 } else if (miss_config == OFPUTIL_TABLE_MISS_DEFAULT &&
3958 connmgr_wants_packet_in_on_miss(ofproto->up.connmgr)) {
3959 rule = ofproto->miss_rule;
3960 }
3961 }
3962 out:
3963 /* Restore port numbers, as they may have been modified above. */
3964 flow->tp_src = old_tp_src;
3965 flow->tp_dst = old_tp_dst;
3966 /* Restore the old in port. */
3967 flow->in_port.ofp_port = old_in_port;
3968
3969 return rule;
3970 }
3971
3972 static struct rule_dpif *rule_dpif_cast(const struct rule *rule)
3973 {
3974 return rule ? CONTAINER_OF(rule, struct rule_dpif, up) : NULL;
3975 }
3976
3977 static struct rule *
3978 rule_alloc(void)
3979 {
3980 struct rule_dpif *rule = xzalloc(sizeof *rule);
3981 return &rule->up;
3982 }
3983
3984 static void
3985 rule_dealloc(struct rule *rule_)
3986 {
3987 struct rule_dpif *rule = rule_dpif_cast(rule_);
3988 free(rule);
3989 }
3990
3991 static enum ofperr
3992 check_mask(struct ofproto_dpif *ofproto, const struct miniflow *flow)
3993 {
3994 const struct odp_support *support;
3995 uint16_t ct_state, ct_zone;
3996 ovs_u128 ct_label;
3997 uint32_t ct_mark;
3998
3999 support = &ofproto->backer->support.odp;
4000 ct_state = MINIFLOW_GET_U16(flow, ct_state);
4001 if (support->ct_state && support->ct_zone && support->ct_mark
4002 && support->ct_label && support->ct_state_nat) {
4003 return ct_state & CS_UNSUPPORTED_MASK ? OFPERR_OFPBMC_BAD_MASK : 0;
4004 }
4005
4006 ct_zone = MINIFLOW_GET_U16(flow, ct_zone);
4007 ct_mark = MINIFLOW_GET_U32(flow, ct_mark);
4008 ct_label = MINIFLOW_GET_U128(flow, ct_label);
4009
4010 if ((ct_state && !support->ct_state)
4011 || (ct_state & CS_UNSUPPORTED_MASK)
4012 || ((ct_state & (CS_SRC_NAT | CS_DST_NAT)) && !support->ct_state_nat)
4013 || (ct_zone && !support->ct_zone)
4014 || (ct_mark && !support->ct_mark)
4015 || (!ovs_u128_is_zero(ct_label) && !support->ct_label)) {
4016 return OFPERR_OFPBMC_BAD_MASK;
4017 }
4018
4019 return 0;
4020 }
4021
4022 static void
4023 report_unsupported_ct(const char *detail)
4024 {
4025 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
4026 VLOG_WARN_RL(&rl, "Rejecting ct action because datapath does not support "
4027 "ct action%s%s (your kernel module may be out of date)",
4028 detail ? " " : "",
4029 detail ? detail : "");
4030 }
4031
4032 static enum ofperr
4033 check_actions(const struct ofproto_dpif *ofproto,
4034 const struct rule_actions *const actions)
4035 {
4036 const struct ofpact *ofpact;
4037
4038 OFPACT_FOR_EACH (ofpact, actions->ofpacts, actions->ofpacts_len) {
4039 const struct odp_support *support;
4040 const struct ofpact_conntrack *ct;
4041 const struct ofpact *a;
4042
4043 if (ofpact->type != OFPACT_CT) {
4044 continue;
4045 }
4046
4047 ct = CONTAINER_OF(ofpact, struct ofpact_conntrack, ofpact);
4048 support = &ofproto->backer->support.odp;
4049
4050 if (!support->ct_state) {
4051 report_unsupported_ct(NULL);
4052 return OFPERR_OFPBAC_BAD_TYPE;
4053 }
4054 if ((ct->zone_imm || ct->zone_src.field) && !support->ct_zone) {
4055 report_unsupported_ct("zone");
4056 return OFPERR_OFPBAC_BAD_ARGUMENT;
4057 }
4058
4059 OFPACT_FOR_EACH(a, ct->actions, ofpact_ct_get_action_len(ct)) {
4060 const struct mf_field *dst = ofpact_get_mf_dst(a);
4061
4062 if (a->type == OFPACT_NAT && !support->ct_state_nat) {
4063 /* The backer doesn't seem to support the NAT bits in
4064 * 'ct_state': assume that it doesn't support the NAT
4065 * action. */
4066 report_unsupported_ct("nat");
4067 return OFPERR_OFPBAC_BAD_TYPE;
4068 }
4069 if (dst && ((dst->id == MFF_CT_MARK && !support->ct_mark)
4070 || (dst->id == MFF_CT_LABEL && !support->ct_label))) {
4071 report_unsupported_ct("setting mark and/or label");
4072 return OFPERR_OFPBAC_BAD_SET_ARGUMENT;
4073 }
4074 }
4075 }
4076
4077 return 0;
4078 }
4079
4080 static enum ofperr
4081 rule_check(struct rule *rule)
4082 {
4083 struct ofproto_dpif *ofproto = ofproto_dpif_cast(rule->ofproto);
4084 enum ofperr err;
4085
4086 err = check_mask(ofproto, &rule->cr.match.mask->masks);
4087 if (err) {
4088 return err;
4089 }
4090 return check_actions(ofproto, rule->actions);
4091 }
4092
4093 static enum ofperr
4094 rule_construct(struct rule *rule_)
4095 OVS_NO_THREAD_SAFETY_ANALYSIS
4096 {
4097 struct rule_dpif *rule = rule_dpif_cast(rule_);
4098 int error;
4099
4100 error = rule_check(rule_);
4101 if (error) {
4102 return error;
4103 }
4104
4105 ovs_mutex_init_adaptive(&rule->stats_mutex);
4106 rule->stats.n_packets = 0;
4107 rule->stats.n_bytes = 0;
4108 rule->stats.used = rule->up.modified;
4109 rule->recirc_id = 0;
4110 rule->new_rule = NULL;
4111 rule->forward_counts = false;
4112
4113 return 0;
4114 }
4115
4116 static void
4117 rule_insert(struct rule *rule_, struct rule *old_rule_, bool forward_counts)
4118 OVS_REQUIRES(ofproto_mutex)
4119 {
4120 struct rule_dpif *rule = rule_dpif_cast(rule_);
4121
4122 if (old_rule_) {
4123 struct rule_dpif *old_rule = rule_dpif_cast(old_rule_);
4124
4125 ovs_assert(!old_rule->new_rule);
4126
4127 /* Take a reference to the new rule, and refer all stats updates from
4128 * the old rule to the new rule. */
4129 ofproto_rule_ref(&rule->up);
4130
4131 ovs_mutex_lock(&old_rule->stats_mutex);
4132 ovs_mutex_lock(&rule->stats_mutex);
4133 old_rule->new_rule = rule; /* Forward future stats. */
4134 old_rule->forward_counts = forward_counts;
4135
4136 if (forward_counts) {
4137 rule->stats = old_rule->stats; /* Transfer stats to the new
4138 * rule. */
4139 } else {
4140 /* Used timestamp must be forwarded whenever a rule is modified. */
4141 rule->stats.used = old_rule->stats.used;
4142 }
4143 ovs_mutex_unlock(&rule->stats_mutex);
4144 ovs_mutex_unlock(&old_rule->stats_mutex);
4145 }
4146 }
4147
4148 static void
4149 rule_destruct(struct rule *rule_)
4150 OVS_NO_THREAD_SAFETY_ANALYSIS
4151 {
4152 struct rule_dpif *rule = rule_dpif_cast(rule_);
4153
4154 ovs_mutex_destroy(&rule->stats_mutex);
4155 /* Release reference to the new rule, if any. */
4156 if (rule->new_rule) {
4157 ofproto_rule_unref(&rule->new_rule->up);
4158 }
4159 if (rule->recirc_id) {
4160 recirc_free_id(rule->recirc_id);
4161 }
4162 }
4163
4164 static void
4165 rule_get_stats(struct rule *rule_, uint64_t *packets, uint64_t *bytes,
4166 long long int *used)
4167 {
4168 struct rule_dpif *rule = rule_dpif_cast(rule_);
4169
4170 ovs_mutex_lock(&rule->stats_mutex);
4171 if (OVS_UNLIKELY(rule->new_rule)) {
4172 rule_get_stats(&rule->new_rule->up, packets, bytes, used);
4173 } else {
4174 *packets = rule->stats.n_packets;
4175 *bytes = rule->stats.n_bytes;
4176 *used = rule->stats.used;
4177 }
4178 ovs_mutex_unlock(&rule->stats_mutex);
4179 }
4180
4181 struct ofproto_dpif_packet_out {
4182 struct xlate_cache xcache;
4183 struct ofpbuf odp_actions;
4184 struct recirc_refs rr;
4185 bool needs_help;
4186 };
4187
4188
4189 static struct ofproto_dpif_packet_out *
4190 ofproto_dpif_packet_out_new(void)
4191 {
4192 struct ofproto_dpif_packet_out *aux = xmalloc(sizeof *aux);
4193 xlate_cache_init(&aux->xcache);
4194 ofpbuf_init(&aux->odp_actions, 64);
4195 aux->rr = RECIRC_REFS_EMPTY_INITIALIZER;
4196 aux->needs_help = false;
4197
4198 return aux;
4199 }
4200
4201 static void
4202 ofproto_dpif_packet_out_delete(struct ofproto_dpif_packet_out *aux)
4203 {
4204 if (aux) {
4205 xlate_cache_uninit(&aux->xcache);
4206 ofpbuf_uninit(&aux->odp_actions);
4207 recirc_refs_unref(&aux->rr);
4208 free(aux);
4209 }
4210 }
4211
4212 static enum ofperr
4213 packet_xlate(struct ofproto *ofproto_, struct ofproto_packet_out *opo)
4214 OVS_REQUIRES(ofproto_mutex)
4215 {
4216 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
4217 struct xlate_out xout;
4218 struct xlate_in xin;
4219 enum ofperr error = 0;
4220
4221 struct ofproto_dpif_packet_out *aux = ofproto_dpif_packet_out_new();
4222
4223 xlate_in_init(&xin, ofproto, opo->version, opo->flow,
4224 opo->flow->in_port.ofp_port, NULL, 0, opo->packet, NULL,
4225 &aux->odp_actions);
4226 xin.ofpacts = opo->ofpacts;
4227 xin.ofpacts_len = opo->ofpacts_len;
4228 /* No learning or stats, but collect side effects to xcache. */
4229 xin.allow_side_effects = false;
4230 xin.resubmit_stats = NULL;
4231 xin.xcache = &aux->xcache;
4232
4233 if (xlate_actions(&xin, &xout) != XLATE_OK) {
4234 error = OFPERR_OFPFMFC_UNKNOWN; /* Error processing actions. */
4235 goto error_out;
4236 } else {
4237 /* Prepare learn actions. */
4238 struct xc_entry *entry;
4239 struct ofpbuf entries = aux->xcache.entries;
4240
4241 XC_ENTRY_FOR_EACH (entry, &entries) {
4242 if (entry->type == XC_LEARN) {
4243 struct ofproto_flow_mod *ofm = entry->learn.ofm;
4244
4245 error = ofproto_flow_mod_learn_refresh(ofm);
4246 if (error) {
4247 goto error_out;
4248 }
4249 struct rule *rule = ofm->temp_rule;
4250 ofm->learn_adds_rule = (rule->state == RULE_INITIALIZED);
4251 if (ofm->learn_adds_rule) {
4252 /* If learning on a different bridge, must use its next
4253 * version number. */
4254 ofm->version = (rule->ofproto == ofproto_)
4255 ? opo->version : rule->ofproto->tables_version + 1;
4256 error = ofproto_flow_mod_learn_start(ofm);
4257 if (error) {
4258 goto error_out;
4259 }
4260 }
4261 }
4262 }
4263
4264 /* Success. */
4265 aux->needs_help = (xout.slow & SLOW_ACTION) != 0;
4266 recirc_refs_swap(&aux->rr, &xout.recircs); /* Hold recirc refs. */
4267 }
4268 xlate_out_uninit(&xout);
4269 opo->aux = aux;
4270 return 0;
4271
4272 error_out:
4273 xlate_out_uninit(&xout);
4274 ofproto_dpif_packet_out_delete(aux);
4275 opo->aux = NULL;
4276 return error;
4277 }
4278
4279 static void
4280 packet_xlate_revert(struct ofproto *ofproto OVS_UNUSED,
4281 struct ofproto_packet_out *opo)
4282 OVS_REQUIRES(ofproto_mutex)
4283 {
4284 struct ofproto_dpif_packet_out *aux = opo->aux;
4285 ovs_assert(aux);
4286
4287 /* Revert the learned flows. */
4288 struct xc_entry *entry;
4289 struct ofpbuf entries = aux->xcache.entries;
4290
4291 XC_ENTRY_FOR_EACH (entry, &entries) {
4292 if (entry->type == XC_LEARN && entry->learn.ofm->learn_adds_rule) {
4293 ofproto_flow_mod_learn_revert(entry->learn.ofm);
4294 }
4295 }
4296
4297 ofproto_dpif_packet_out_delete(aux);
4298 opo->aux = NULL;
4299 }
4300
4301 /* Push stats and perform side effects of flow translation. */
4302 static void
4303 ofproto_dpif_xcache_execute(struct ofproto_dpif *ofproto,
4304 struct xlate_cache *xcache,
4305 const struct dpif_flow_stats *stats)
4306 OVS_REQUIRES(ofproto_mutex)
4307 {
4308 struct xc_entry *entry;
4309 struct ofpbuf entries = xcache->entries;
4310
4311 XC_ENTRY_FOR_EACH (entry, &entries) {
4312 switch (entry->type) {
4313 case XC_LEARN:
4314 /* Finish the learned flows. */
4315 if (entry->learn.ofm->learn_adds_rule) {
4316 ofproto_flow_mod_learn_finish(entry->learn.ofm, &ofproto->up);
4317 }
4318 break;
4319 case XC_FIN_TIMEOUT:
4320 if (stats->tcp_flags & (TCP_FIN | TCP_RST)) {
4321 /* 'ofproto_mutex' already held */
4322 ofproto_rule_reduce_timeouts__(&entry->fin.rule->up,
4323 entry->fin.idle,
4324 entry->fin.hard);
4325 }
4326 break;
4327 /* All the rest can be dealt with by the xlate layer. */
4328 case XC_TABLE:
4329 case XC_RULE:
4330 case XC_BOND:
4331 case XC_NETDEV:
4332 case XC_NETFLOW:
4333 case XC_MIRROR:
4334 case XC_NORMAL:
4335 case XC_GROUP:
4336 case XC_TNL_NEIGH:
4337 case XC_CONTROLLER:
4338 xlate_push_stats_entry(entry, stats);
4339 break;
4340 default:
4341 OVS_NOT_REACHED();
4342 }
4343 }
4344 }
4345
4346 static void
4347 packet_execute(struct ofproto *ofproto_, struct ofproto_packet_out *opo)
4348 OVS_REQUIRES(ofproto_mutex)
4349 {
4350 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
4351 struct dpif_flow_stats stats;
4352 struct dpif_execute execute;
4353
4354 struct ofproto_dpif_packet_out *aux = opo->aux;
4355 ovs_assert(aux);
4356
4357 /* Run the side effects from the xcache. */
4358 dpif_flow_stats_extract(opo->flow, opo->packet, time_msec(), &stats);
4359 ofproto_dpif_xcache_execute(ofproto, &aux->xcache, &stats);
4360
4361 execute.actions = aux->odp_actions.data;
4362 execute.actions_len = aux->odp_actions.size;
4363
4364 pkt_metadata_from_flow(&opo->packet->md, opo->flow);
4365 execute.packet = opo->packet;
4366 execute.flow = opo->flow;
4367 execute.needs_help = aux->needs_help;
4368 execute.probe = false;
4369 execute.mtu = 0;
4370
4371 /* Fix up in_port. */
4372 ofproto_dpif_set_packet_odp_port(ofproto, opo->flow->in_port.ofp_port,
4373 opo->packet);
4374
4375 dpif_execute(ofproto->backer->dpif, &execute);
4376 ofproto_dpif_packet_out_delete(aux);
4377 opo->aux = NULL;
4378 }
4379
4380 static struct group_dpif *group_dpif_cast(const struct ofgroup *group)
4381 {
4382 return group ? CONTAINER_OF(group, struct group_dpif, up) : NULL;
4383 }
4384
4385 static struct ofgroup *
4386 group_alloc(void)
4387 {
4388 struct group_dpif *group = xzalloc(sizeof *group);
4389 return &group->up;
4390 }
4391
4392 static void
4393 group_dealloc(struct ofgroup *group_)
4394 {
4395 struct group_dpif *group = group_dpif_cast(group_);
4396 free(group);
4397 }
4398
4399 static void
4400 group_construct_stats(struct group_dpif *group)
4401 OVS_REQUIRES(group->stats_mutex)
4402 {
4403 group->packet_count = 0;
4404 group->byte_count = 0;
4405
4406 struct ofputil_bucket *bucket;
4407 LIST_FOR_EACH (bucket, list_node, &group->up.buckets) {
4408 bucket->stats.packet_count = 0;
4409 bucket->stats.byte_count = 0;
4410 }
4411 }
4412
4413 void
4414 group_dpif_credit_stats(struct group_dpif *group,
4415 struct ofputil_bucket *bucket,
4416 const struct dpif_flow_stats *stats)
4417 {
4418 ovs_mutex_lock(&group->stats_mutex);
4419 group->packet_count += stats->n_packets;
4420 group->byte_count += stats->n_bytes;
4421 if (bucket) {
4422 bucket->stats.packet_count += stats->n_packets;
4423 bucket->stats.byte_count += stats->n_bytes;
4424 } else { /* Credit to all buckets */
4425 struct ofputil_bucket *bucket;
4426 LIST_FOR_EACH (bucket, list_node, &group->up.buckets) {
4427 bucket->stats.packet_count += stats->n_packets;
4428 bucket->stats.byte_count += stats->n_bytes;
4429 }
4430 }
4431 ovs_mutex_unlock(&group->stats_mutex);
4432 }
4433
4434 static enum ofperr
4435 group_construct(struct ofgroup *group_)
4436 {
4437 struct group_dpif *group = group_dpif_cast(group_);
4438
4439 ovs_mutex_init_adaptive(&group->stats_mutex);
4440 ovs_mutex_lock(&group->stats_mutex);
4441 group_construct_stats(group);
4442 ovs_mutex_unlock(&group->stats_mutex);
4443 return 0;
4444 }
4445
4446 static void
4447 group_destruct(struct ofgroup *group_)
4448 {
4449 struct group_dpif *group = group_dpif_cast(group_);
4450 ovs_mutex_destroy(&group->stats_mutex);
4451 }
4452
4453 static enum ofperr
4454 group_get_stats(const struct ofgroup *group_, struct ofputil_group_stats *ogs)
4455 {
4456 struct group_dpif *group = group_dpif_cast(group_);
4457
4458 ovs_mutex_lock(&group->stats_mutex);
4459 ogs->packet_count = group->packet_count;
4460 ogs->byte_count = group->byte_count;
4461
4462 struct bucket_counter *bucket_stats = ogs->bucket_stats;
4463 struct ofputil_bucket *bucket;
4464 LIST_FOR_EACH (bucket, list_node, &group->up.buckets) {
4465 bucket_stats->packet_count = bucket->stats.packet_count;
4466 bucket_stats->byte_count = bucket->stats.byte_count;
4467 bucket_stats++;
4468 }
4469 ovs_mutex_unlock(&group->stats_mutex);
4470
4471 return 0;
4472 }
4473
4474 /* If the group exists, this function increments the groups's reference count.
4475 *
4476 * Make sure to call ofproto_group_unref() after no longer needing to maintain
4477 * a reference to the group. */
4478 struct group_dpif *
4479 group_dpif_lookup(struct ofproto_dpif *ofproto, uint32_t group_id,
4480 ovs_version_t version, bool take_ref)
4481 {
4482 struct ofgroup *ofgroup = ofproto_group_lookup(&ofproto->up, group_id,
4483 version, take_ref);
4484 return ofgroup ? group_dpif_cast(ofgroup) : NULL;
4485 }
4486 \f
4487 /* Sends 'packet' out 'ofport'. If 'port' is a tunnel and that tunnel type
4488 * supports a notion of an OAM flag, sets it if 'oam' is true.
4489 * May modify 'packet'.
4490 * Returns 0 if successful, otherwise a positive errno value. */
4491 int
4492 ofproto_dpif_send_packet(const struct ofport_dpif *ofport, bool oam,
4493 struct dp_packet *packet)
4494 {
4495 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport->up.ofproto);
4496 int error;
4497
4498 error = xlate_send_packet(ofport, oam, packet);
4499
4500 ovs_mutex_lock(&ofproto->stats_mutex);
4501 ofproto->stats.tx_packets++;
4502 ofproto->stats.tx_bytes += dp_packet_size(packet);
4503 ovs_mutex_unlock(&ofproto->stats_mutex);
4504 return error;
4505 }
4506 \f
4507 /* Return the version string of the datapath that backs up
4508 * this 'ofproto'.
4509 */
4510 static const char *
4511 get_datapath_version(const struct ofproto *ofproto_)
4512 {
4513 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
4514
4515 return ofproto->backer->dp_version_string;
4516 }
4517
4518 static void
4519 ct_flush(const struct ofproto *ofproto_, const uint16_t *zone)
4520 {
4521 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
4522
4523 ct_dpif_flush(ofproto->backer->dpif, zone);
4524 }
4525
4526 static bool
4527 set_frag_handling(struct ofproto *ofproto_,
4528 enum ofputil_frag_handling frag_handling)
4529 {
4530 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
4531 if (frag_handling != OFPUTIL_FRAG_REASM) {
4532 ofproto->backer->need_revalidate = REV_RECONFIGURE;
4533 return true;
4534 } else {
4535 return false;
4536 }
4537 }
4538
4539 static enum ofperr
4540 nxt_resume(struct ofproto *ofproto_,
4541 const struct ofputil_packet_in_private *pin)
4542 {
4543 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
4544
4545 /* Translate pin into datapath actions. */
4546 uint64_t odp_actions_stub[1024 / 8];
4547 struct ofpbuf odp_actions = OFPBUF_STUB_INITIALIZER(odp_actions_stub);
4548 enum slow_path_reason slow;
4549 enum ofperr error = xlate_resume(ofproto, pin, &odp_actions, &slow);
4550
4551 /* Steal 'pin->packet' and put it into a dp_packet. */
4552 struct dp_packet packet;
4553 dp_packet_init(&packet, pin->public.packet_len);
4554 dp_packet_put(&packet, pin->public.packet, pin->public.packet_len);
4555
4556 pkt_metadata_from_flow(&packet.md, &pin->public.flow_metadata.flow);
4557
4558 /* Fix up in_port. */
4559 ofproto_dpif_set_packet_odp_port(ofproto,
4560 pin->public.flow_metadata.flow.in_port.ofp_port,
4561 &packet);
4562
4563 struct flow headers;
4564 flow_extract(&packet, &headers);
4565
4566 /* Execute the datapath actions on the packet. */
4567 struct dpif_execute execute = {
4568 .actions = odp_actions.data,
4569 .actions_len = odp_actions.size,
4570 .needs_help = (slow & SLOW_ACTION) != 0,
4571 .packet = &packet,
4572 .flow = &headers,
4573 };
4574 dpif_execute(ofproto->backer->dpif, &execute);
4575
4576 /* Clean up. */
4577 ofpbuf_uninit(&odp_actions);
4578 dp_packet_uninit(&packet);
4579
4580 return error;
4581 }
4582 \f
4583 /* NetFlow. */
4584
4585 static int
4586 set_netflow(struct ofproto *ofproto_,
4587 const struct netflow_options *netflow_options)
4588 {
4589 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
4590
4591 if (netflow_options) {
4592 if (!ofproto->netflow) {
4593 ofproto->netflow = netflow_create();
4594 ofproto->backer->need_revalidate = REV_RECONFIGURE;
4595 }
4596 return netflow_set_options(ofproto->netflow, netflow_options);
4597 } else if (ofproto->netflow) {
4598 ofproto->backer->need_revalidate = REV_RECONFIGURE;
4599 netflow_unref(ofproto->netflow);
4600 ofproto->netflow = NULL;
4601 }
4602
4603 return 0;
4604 }
4605
4606 static void
4607 get_netflow_ids(const struct ofproto *ofproto_,
4608 uint8_t *engine_type, uint8_t *engine_id)
4609 {
4610 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
4611
4612 dpif_get_netflow_ids(ofproto->backer->dpif, engine_type, engine_id);
4613 }
4614 \f
4615 struct ofproto_dpif *
4616 ofproto_dpif_lookup(const char *name)
4617 {
4618 struct ofproto_dpif *ofproto;
4619
4620 HMAP_FOR_EACH_WITH_HASH (ofproto, all_ofproto_dpifs_node,
4621 hash_string(name, 0), &all_ofproto_dpifs) {
4622 if (!strcmp(ofproto->up.name, name)) {
4623 return ofproto;
4624 }
4625 }
4626 return NULL;
4627 }
4628
4629 static void
4630 ofproto_unixctl_fdb_flush(struct unixctl_conn *conn, int argc,
4631 const char *argv[], void *aux OVS_UNUSED)
4632 {
4633 struct ofproto_dpif *ofproto;
4634
4635 if (argc > 1) {
4636 ofproto = ofproto_dpif_lookup(argv[1]);
4637 if (!ofproto) {
4638 unixctl_command_reply_error(conn, "no such bridge");
4639 return;
4640 }
4641 ovs_rwlock_wrlock(&ofproto->ml->rwlock);
4642 mac_learning_flush(ofproto->ml);
4643 ovs_rwlock_unlock(&ofproto->ml->rwlock);
4644 } else {
4645 HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
4646 ovs_rwlock_wrlock(&ofproto->ml->rwlock);
4647 mac_learning_flush(ofproto->ml);
4648 ovs_rwlock_unlock(&ofproto->ml->rwlock);
4649 }
4650 }
4651
4652 unixctl_command_reply(conn, "table successfully flushed");
4653 }
4654
4655 static void
4656 ofproto_unixctl_mcast_snooping_flush(struct unixctl_conn *conn, int argc,
4657 const char *argv[], void *aux OVS_UNUSED)
4658 {
4659 struct ofproto_dpif *ofproto;
4660
4661 if (argc > 1) {
4662 ofproto = ofproto_dpif_lookup(argv[1]);
4663 if (!ofproto) {
4664 unixctl_command_reply_error(conn, "no such bridge");
4665 return;
4666 }
4667
4668 if (!mcast_snooping_enabled(ofproto->ms)) {
4669 unixctl_command_reply_error(conn, "multicast snooping is disabled");
4670 return;
4671 }
4672 mcast_snooping_mdb_flush(ofproto->ms);
4673 } else {
4674 HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
4675 if (!mcast_snooping_enabled(ofproto->ms)) {
4676 continue;
4677 }
4678 mcast_snooping_mdb_flush(ofproto->ms);
4679 }
4680 }
4681
4682 unixctl_command_reply(conn, "table successfully flushed");
4683 }
4684
4685 static struct ofport_dpif *
4686 ofbundle_get_a_port(const struct ofbundle *bundle)
4687 {
4688 return CONTAINER_OF(ovs_list_front(&bundle->ports), struct ofport_dpif,
4689 bundle_node);
4690 }
4691
4692 static void
4693 ofproto_unixctl_fdb_show(struct unixctl_conn *conn, int argc OVS_UNUSED,
4694 const char *argv[], void *aux OVS_UNUSED)
4695 {
4696 struct ds ds = DS_EMPTY_INITIALIZER;
4697 const struct ofproto_dpif *ofproto;
4698 const struct mac_entry *e;
4699
4700 ofproto = ofproto_dpif_lookup(argv[1]);
4701 if (!ofproto) {
4702 unixctl_command_reply_error(conn, "no such bridge");
4703 return;
4704 }
4705
4706 ds_put_cstr(&ds, " port VLAN MAC Age\n");
4707 ovs_rwlock_rdlock(&ofproto->ml->rwlock);
4708 LIST_FOR_EACH (e, lru_node, &ofproto->ml->lrus) {
4709 struct ofbundle *bundle = mac_entry_get_port(ofproto->ml, e);
4710 char name[OFP_MAX_PORT_NAME_LEN];
4711
4712 ofputil_port_to_string(ofbundle_get_a_port(bundle)->up.ofp_port,
4713 name, sizeof name);
4714 ds_put_format(&ds, "%5s %4d "ETH_ADDR_FMT" %3d\n",
4715 name, e->vlan, ETH_ADDR_ARGS(e->mac),
4716 mac_entry_age(ofproto->ml, e));
4717 }
4718 ovs_rwlock_unlock(&ofproto->ml->rwlock);
4719 unixctl_command_reply(conn, ds_cstr(&ds));
4720 ds_destroy(&ds);
4721 }
4722
4723 static void
4724 ofproto_unixctl_mcast_snooping_show(struct unixctl_conn *conn,
4725 int argc OVS_UNUSED,
4726 const char *argv[],
4727 void *aux OVS_UNUSED)
4728 {
4729 struct ds ds = DS_EMPTY_INITIALIZER;
4730 const struct ofproto_dpif *ofproto;
4731 const struct ofbundle *bundle;
4732 const struct mcast_group *grp;
4733 struct mcast_group_bundle *b;
4734 struct mcast_mrouter_bundle *mrouter;
4735
4736 ofproto = ofproto_dpif_lookup(argv[1]);
4737 if (!ofproto) {
4738 unixctl_command_reply_error(conn, "no such bridge");
4739 return;
4740 }
4741
4742 if (!mcast_snooping_enabled(ofproto->ms)) {
4743 unixctl_command_reply_error(conn, "multicast snooping is disabled");
4744 return;
4745 }
4746
4747 ds_put_cstr(&ds, " port VLAN GROUP Age\n");
4748 ovs_rwlock_rdlock(&ofproto->ms->rwlock);
4749 LIST_FOR_EACH (grp, group_node, &ofproto->ms->group_lru) {
4750 LIST_FOR_EACH(b, bundle_node, &grp->bundle_lru) {
4751 char name[OFP_MAX_PORT_NAME_LEN];
4752
4753 bundle = b->port;
4754 ofputil_port_to_string(ofbundle_get_a_port(bundle)->up.ofp_port,
4755 name, sizeof name);
4756 ds_put_format(&ds, "%5s %4d ", name, grp->vlan);
4757 ipv6_format_mapped(&grp->addr, &ds);
4758 ds_put_format(&ds, " %3d\n",
4759 mcast_bundle_age(ofproto->ms, b));
4760 }
4761 }
4762
4763 /* ports connected to multicast routers */
4764 LIST_FOR_EACH(mrouter, mrouter_node, &ofproto->ms->mrouter_lru) {
4765 char name[OFP_MAX_PORT_NAME_LEN];
4766
4767 bundle = mrouter->port;
4768 ofputil_port_to_string(ofbundle_get_a_port(bundle)->up.ofp_port,
4769 name, sizeof name);
4770 ds_put_format(&ds, "%5s %4d querier %3d\n",
4771 name, mrouter->vlan,
4772 mcast_mrouter_age(ofproto->ms, mrouter));
4773 }
4774 ovs_rwlock_unlock(&ofproto->ms->rwlock);
4775 unixctl_command_reply(conn, ds_cstr(&ds));
4776 ds_destroy(&ds);
4777 }
4778
4779 /* Store the current ofprotos in 'ofproto_shash'. Returns a sorted list
4780 * of the 'ofproto_shash' nodes. It is the responsibility of the caller
4781 * to destroy 'ofproto_shash' and free the returned value. */
4782 static const struct shash_node **
4783 get_ofprotos(struct shash *ofproto_shash)
4784 {
4785 const struct ofproto_dpif *ofproto;
4786
4787 HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
4788 char *name = xasprintf("%s@%s", ofproto->up.type, ofproto->up.name);
4789 shash_add_nocopy(ofproto_shash, name, ofproto);
4790 }
4791
4792 return shash_sort(ofproto_shash);
4793 }
4794
4795 static void
4796 ofproto_unixctl_dpif_dump_dps(struct unixctl_conn *conn, int argc OVS_UNUSED,
4797 const char *argv[] OVS_UNUSED,
4798 void *aux OVS_UNUSED)
4799 {
4800 struct ds ds = DS_EMPTY_INITIALIZER;
4801 struct shash ofproto_shash;
4802 const struct shash_node **sorted_ofprotos;
4803 int i;
4804
4805 shash_init(&ofproto_shash);
4806 sorted_ofprotos = get_ofprotos(&ofproto_shash);
4807 for (i = 0; i < shash_count(&ofproto_shash); i++) {
4808 const struct shash_node *node = sorted_ofprotos[i];
4809 ds_put_format(&ds, "%s\n", node->name);
4810 }
4811
4812 shash_destroy(&ofproto_shash);
4813 free(sorted_ofprotos);
4814
4815 unixctl_command_reply(conn, ds_cstr(&ds));
4816 ds_destroy(&ds);
4817 }
4818
4819 static void
4820 dpif_show_backer(const struct dpif_backer *backer, struct ds *ds)
4821 {
4822 const struct shash_node **ofprotos;
4823 struct dpif_dp_stats dp_stats;
4824 struct shash ofproto_shash;
4825 size_t i;
4826
4827 dpif_get_dp_stats(backer->dpif, &dp_stats);
4828
4829 ds_put_format(ds, "%s: hit:%"PRIu64" missed:%"PRIu64"\n",
4830 dpif_name(backer->dpif), dp_stats.n_hit, dp_stats.n_missed);
4831
4832 shash_init(&ofproto_shash);
4833 ofprotos = get_ofprotos(&ofproto_shash);
4834 for (i = 0; i < shash_count(&ofproto_shash); i++) {
4835 struct ofproto_dpif *ofproto = ofprotos[i]->data;
4836 const struct shash_node **ports;
4837 size_t j;
4838
4839 if (ofproto->backer != backer) {
4840 continue;
4841 }
4842
4843 ds_put_format(ds, "\t%s:\n", ofproto->up.name);
4844
4845 ports = shash_sort(&ofproto->up.port_by_name);
4846 for (j = 0; j < shash_count(&ofproto->up.port_by_name); j++) {
4847 const struct shash_node *node = ports[j];
4848 struct ofport *ofport = node->data;
4849 struct smap config;
4850 odp_port_t odp_port;
4851
4852 ds_put_format(ds, "\t\t%s %u/", netdev_get_name(ofport->netdev),
4853 ofport->ofp_port);
4854
4855 odp_port = ofp_port_to_odp_port(ofproto, ofport->ofp_port);
4856 if (odp_port != ODPP_NONE) {
4857 ds_put_format(ds, "%"PRIu32":", odp_port);
4858 } else {
4859 ds_put_cstr(ds, "none:");
4860 }
4861
4862 ds_put_format(ds, " (%s", netdev_get_type(ofport->netdev));
4863
4864 smap_init(&config);
4865 if (!netdev_get_config(ofport->netdev, &config)) {
4866 const struct smap_node **nodes;
4867 size_t i;
4868
4869 nodes = smap_sort(&config);
4870 for (i = 0; i < smap_count(&config); i++) {
4871 const struct smap_node *node = nodes[i];
4872 ds_put_format(ds, "%c %s=%s", i ? ',' : ':',
4873 node->key, node->value);
4874 }
4875 free(nodes);
4876 }
4877 smap_destroy(&config);
4878
4879 ds_put_char(ds, ')');
4880 ds_put_char(ds, '\n');
4881 }
4882 free(ports);
4883 }
4884 shash_destroy(&ofproto_shash);
4885 free(ofprotos);
4886 }
4887
4888 static void
4889 ofproto_unixctl_dpif_show(struct unixctl_conn *conn, int argc OVS_UNUSED,
4890 const char *argv[] OVS_UNUSED, void *aux OVS_UNUSED)
4891 {
4892 struct ds ds = DS_EMPTY_INITIALIZER;
4893 const struct shash_node **backers;
4894 int i;
4895
4896 backers = shash_sort(&all_dpif_backers);
4897 for (i = 0; i < shash_count(&all_dpif_backers); i++) {
4898 dpif_show_backer(backers[i]->data, &ds);
4899 }
4900 free(backers);
4901
4902 unixctl_command_reply(conn, ds_cstr(&ds));
4903 ds_destroy(&ds);
4904 }
4905
4906 static void
4907 ofproto_unixctl_dpif_dump_flows(struct unixctl_conn *conn,
4908 int argc OVS_UNUSED, const char *argv[],
4909 void *aux OVS_UNUSED)
4910 {
4911 const struct ofproto_dpif *ofproto;
4912
4913 struct ds ds = DS_EMPTY_INITIALIZER;
4914 bool verbosity = false;
4915
4916 struct dpif_port dpif_port;
4917 struct dpif_port_dump port_dump;
4918 struct hmap portno_names;
4919
4920 struct dpif_flow_dump *flow_dump;
4921 struct dpif_flow_dump_thread *flow_dump_thread;
4922 struct dpif_flow f;
4923 int error;
4924
4925 ofproto = ofproto_dpif_lookup(argv[argc - 1]);
4926 if (!ofproto) {
4927 unixctl_command_reply_error(conn, "no such bridge");
4928 return;
4929 }
4930
4931 if (argc > 2 && !strcmp(argv[1], "-m")) {
4932 verbosity = true;
4933 }
4934
4935 hmap_init(&portno_names);
4936 DPIF_PORT_FOR_EACH (&dpif_port, &port_dump, ofproto->backer->dpif) {
4937 odp_portno_names_set(&portno_names, dpif_port.port_no, dpif_port.name);
4938 }
4939
4940 ds_init(&ds);
4941 flow_dump = dpif_flow_dump_create(ofproto->backer->dpif, false);
4942 flow_dump_thread = dpif_flow_dump_thread_create(flow_dump);
4943 while (dpif_flow_dump_next(flow_dump_thread, &f, 1)) {
4944 struct flow flow;
4945
4946 if (odp_flow_key_to_flow(f.key, f.key_len, &flow) == ODP_FIT_ERROR
4947 || xlate_lookup_ofproto(ofproto->backer, &flow, NULL) != ofproto) {
4948 continue;
4949 }
4950
4951 if (verbosity) {
4952 odp_format_ufid(&f.ufid, &ds);
4953 ds_put_cstr(&ds, " ");
4954 }
4955 odp_flow_format(f.key, f.key_len, f.mask, f.mask_len,
4956 &portno_names, &ds, verbosity);
4957 ds_put_cstr(&ds, ", ");
4958 dpif_flow_stats_format(&f.stats, &ds);
4959 ds_put_cstr(&ds, ", actions:");
4960 format_odp_actions(&ds, f.actions, f.actions_len);
4961 ds_put_char(&ds, '\n');
4962 }
4963 dpif_flow_dump_thread_destroy(flow_dump_thread);
4964 error = dpif_flow_dump_destroy(flow_dump);
4965
4966 if (error) {
4967 ds_clear(&ds);
4968 ds_put_format(&ds, "dpif/dump_flows failed: %s", ovs_strerror(errno));
4969 unixctl_command_reply_error(conn, ds_cstr(&ds));
4970 } else {
4971 unixctl_command_reply(conn, ds_cstr(&ds));
4972 }
4973 odp_portno_names_destroy(&portno_names);
4974 hmap_destroy(&portno_names);
4975 ds_destroy(&ds);
4976 }
4977
4978 static void
4979 ofproto_revalidate_all_backers(void)
4980 {
4981 const struct shash_node **backers;
4982 int i;
4983
4984 backers = shash_sort(&all_dpif_backers);
4985 for (i = 0; i < shash_count(&all_dpif_backers); i++) {
4986 struct dpif_backer *backer = backers[i]->data;
4987 backer->need_revalidate = REV_RECONFIGURE;
4988 }
4989 free(backers);
4990 }
4991
4992 static void
4993 disable_tnl_push_pop(struct unixctl_conn *conn OVS_UNUSED, int argc OVS_UNUSED,
4994 const char *argv[], void *aux OVS_UNUSED)
4995 {
4996 if (!strcasecmp(argv[1], "off")) {
4997 ofproto_use_tnl_push_pop = false;
4998 unixctl_command_reply(conn, "Tunnel push-pop off");
4999 ofproto_revalidate_all_backers();
5000 } else if (!strcasecmp(argv[1], "on")) {
5001 ofproto_use_tnl_push_pop = true;
5002 unixctl_command_reply(conn, "Tunnel push-pop on");
5003 ofproto_revalidate_all_backers();
5004 } else {
5005 unixctl_command_reply_error(conn, "Invalid argument");
5006 }
5007 }
5008
5009 static void
5010 disable_datapath_truncate(struct unixctl_conn *conn OVS_UNUSED,
5011 int argc OVS_UNUSED,
5012 const char *argv[] OVS_UNUSED,
5013 void *aux OVS_UNUSED)
5014 {
5015 const struct shash_node **backers;
5016 int i;
5017
5018 backers = shash_sort(&all_dpif_backers);
5019 for (i = 0; i < shash_count(&all_dpif_backers); i++) {
5020 struct dpif_backer *backer = backers[i]->data;
5021 backer->support.trunc = false;
5022 }
5023 free(backers);
5024 unixctl_command_reply(conn, "Datapath truncate action diabled");
5025 }
5026
5027 static void
5028 ofproto_unixctl_init(void)
5029 {
5030 static bool registered;
5031 if (registered) {
5032 return;
5033 }
5034 registered = true;
5035
5036 unixctl_command_register("fdb/flush", "[bridge]", 0, 1,
5037 ofproto_unixctl_fdb_flush, NULL);
5038 unixctl_command_register("fdb/show", "bridge", 1, 1,
5039 ofproto_unixctl_fdb_show, NULL);
5040 unixctl_command_register("mdb/flush", "[bridge]", 0, 1,
5041 ofproto_unixctl_mcast_snooping_flush, NULL);
5042 unixctl_command_register("mdb/show", "bridge", 1, 1,
5043 ofproto_unixctl_mcast_snooping_show, NULL);
5044 unixctl_command_register("dpif/dump-dps", "", 0, 0,
5045 ofproto_unixctl_dpif_dump_dps, NULL);
5046 unixctl_command_register("dpif/show", "", 0, 0, ofproto_unixctl_dpif_show,
5047 NULL);
5048 unixctl_command_register("dpif/dump-flows", "[-m] bridge", 1, 2,
5049 ofproto_unixctl_dpif_dump_flows, NULL);
5050
5051 unixctl_command_register("ofproto/tnl-push-pop", "[on]|[off]", 1, 1,
5052 disable_tnl_push_pop, NULL);
5053
5054 unixctl_command_register("dpif/disable-truncate", "", 0, 0,
5055 disable_datapath_truncate, NULL);
5056 }
5057 \f
5058 static odp_port_t
5059 ofp_port_to_odp_port(const struct ofproto_dpif *ofproto, ofp_port_t ofp_port)
5060 {
5061 const struct ofport_dpif *ofport = ofp_port_to_ofport(ofproto, ofp_port);
5062 return ofport ? ofport->odp_port : ODPP_NONE;
5063 }
5064
5065 struct ofport_dpif *
5066 odp_port_to_ofport(const struct dpif_backer *backer, odp_port_t odp_port)
5067 {
5068 struct ofport_dpif *port;
5069
5070 ovs_rwlock_rdlock(&backer->odp_to_ofport_lock);
5071 HMAP_FOR_EACH_IN_BUCKET (port, odp_port_node, hash_odp_port(odp_port),
5072 &backer->odp_to_ofport_map) {
5073 if (port->odp_port == odp_port) {
5074 ovs_rwlock_unlock(&backer->odp_to_ofport_lock);
5075 return port;
5076 }
5077 }
5078
5079 ovs_rwlock_unlock(&backer->odp_to_ofport_lock);
5080 return NULL;
5081 }
5082
5083 static ofp_port_t
5084 odp_port_to_ofp_port(const struct ofproto_dpif *ofproto, odp_port_t odp_port)
5085 {
5086 struct ofport_dpif *port;
5087
5088 port = odp_port_to_ofport(ofproto->backer, odp_port);
5089 if (port && &ofproto->up == port->up.ofproto) {
5090 return port->up.ofp_port;
5091 } else {
5092 return OFPP_NONE;
5093 }
5094 }
5095
5096 int
5097 ofproto_dpif_add_internal_flow(struct ofproto_dpif *ofproto,
5098 const struct match *match, int priority,
5099 uint16_t idle_timeout,
5100 const struct ofpbuf *ofpacts,
5101 struct rule **rulep)
5102 {
5103 struct ofputil_flow_mod fm;
5104 struct rule_dpif *rule;
5105 int error;
5106
5107 fm = (struct ofputil_flow_mod) {
5108 .buffer_id = UINT32_MAX,
5109 .match = *match,
5110 .priority = priority,
5111 .table_id = TBL_INTERNAL,
5112 .command = OFPFC_ADD,
5113 .idle_timeout = idle_timeout,
5114 .flags = OFPUTIL_FF_HIDDEN_FIELDS | OFPUTIL_FF_NO_READONLY,
5115 .ofpacts = ofpacts->data,
5116 .ofpacts_len = ofpacts->size,
5117 };
5118
5119 error = ofproto_flow_mod(&ofproto->up, &fm);
5120 if (error) {
5121 VLOG_ERR_RL(&rl, "failed to add internal flow (%s)",
5122 ofperr_to_string(error));
5123 *rulep = NULL;
5124 return error;
5125 }
5126
5127 rule = rule_dpif_lookup_in_table(ofproto,
5128 ofproto_dpif_get_tables_version(ofproto),
5129 TBL_INTERNAL, &fm.match.flow,
5130 &fm.match.wc);
5131 if (rule) {
5132 *rulep = &rule->up;
5133 } else {
5134 OVS_NOT_REACHED();
5135 }
5136 return 0;
5137 }
5138
5139 int
5140 ofproto_dpif_delete_internal_flow(struct ofproto_dpif *ofproto,
5141 struct match *match, int priority)
5142 {
5143 struct ofputil_flow_mod fm;
5144 int error;
5145
5146 fm = (struct ofputil_flow_mod) {
5147 .buffer_id = UINT32_MAX,
5148 .match = *match,
5149 .priority = priority,
5150 .table_id = TBL_INTERNAL,
5151 .flags = OFPUTIL_FF_HIDDEN_FIELDS | OFPUTIL_FF_NO_READONLY,
5152 .command = OFPFC_DELETE_STRICT,
5153 };
5154
5155 error = ofproto_flow_mod(&ofproto->up, &fm);
5156 if (error) {
5157 VLOG_ERR_RL(&rl, "failed to delete internal flow (%s)",
5158 ofperr_to_string(error));
5159 return error;
5160 }
5161
5162 return 0;
5163 }
5164
5165 const struct ofproto_class ofproto_dpif_class = {
5166 init,
5167 enumerate_types,
5168 enumerate_names,
5169 del,
5170 port_open_type,
5171 type_run,
5172 type_wait,
5173 alloc,
5174 construct,
5175 destruct,
5176 dealloc,
5177 run,
5178 ofproto_dpif_wait,
5179 NULL, /* get_memory_usage. */
5180 type_get_memory_usage,
5181 flush,
5182 query_tables,
5183 set_tables_version,
5184 port_alloc,
5185 port_construct,
5186 port_destruct,
5187 port_dealloc,
5188 port_modified,
5189 port_reconfigured,
5190 port_query_by_name,
5191 port_add,
5192 port_del,
5193 port_set_config,
5194 port_get_stats,
5195 port_dump_start,
5196 port_dump_next,
5197 port_dump_done,
5198 port_poll,
5199 port_poll_wait,
5200 port_is_lacp_current,
5201 port_get_lacp_stats,
5202 NULL, /* rule_choose_table */
5203 rule_alloc,
5204 rule_construct,
5205 rule_insert,
5206 NULL, /* rule_delete */
5207 rule_destruct,
5208 rule_dealloc,
5209 rule_get_stats,
5210 packet_xlate,
5211 packet_xlate_revert,
5212 packet_execute,
5213 set_frag_handling,
5214 nxt_resume,
5215 set_netflow,
5216 get_netflow_ids,
5217 set_sflow,
5218 set_ipfix,
5219 get_ipfix_stats,
5220 set_cfm,
5221 cfm_status_changed,
5222 get_cfm_status,
5223 set_lldp,
5224 get_lldp_status,
5225 set_aa,
5226 aa_mapping_set,
5227 aa_mapping_unset,
5228 aa_vlan_get_queued,
5229 aa_vlan_get_queue_size,
5230 set_bfd,
5231 bfd_status_changed,
5232 get_bfd_status,
5233 set_stp,
5234 get_stp_status,
5235 set_stp_port,
5236 get_stp_port_status,
5237 get_stp_port_stats,
5238 set_rstp,
5239 get_rstp_status,
5240 set_rstp_port,
5241 get_rstp_port_status,
5242 set_queues,
5243 bundle_set,
5244 bundle_remove,
5245 mirror_set__,
5246 mirror_get_stats__,
5247 set_flood_vlans,
5248 is_mirror_output_bundle,
5249 forward_bpdu_changed,
5250 set_mac_table_config,
5251 set_mcast_snooping,
5252 set_mcast_snooping_port,
5253 NULL, /* meter_get_features */
5254 NULL, /* meter_set */
5255 NULL, /* meter_get */
5256 NULL, /* meter_del */
5257 group_alloc, /* group_alloc */
5258 group_construct, /* group_construct */
5259 group_destruct, /* group_destruct */
5260 group_dealloc, /* group_dealloc */
5261 NULL, /* group_modify */
5262 group_get_stats, /* group_get_stats */
5263 get_datapath_version, /* get_datapath_version */
5264 ct_flush, /* ct_flush */
5265 };