2 * Copyright (c) 2009, 2010, 2011, 2012, 2013 Nicira, Inc.
4 * Licensed under the Apache License, Version 2.0 (the "License");
5 * you may not use this file except in compliance with the License.
6 * You may obtain a copy of the License at:
8 * http://www.apache.org/licenses/LICENSE-2.0
10 * Unless required by applicable law or agreed to in writing, software
11 * distributed under the License is distributed on an "AS IS" BASIS,
12 * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
13 * See the License for the specific language governing permissions and
14 * limitations under the License.
19 #include "ofproto/ofproto-dpif.h"
20 #include "ofproto/ofproto-provider.h"
27 #include "byte-order.h"
32 #include "dynamic-string.h"
33 #include "fail-open.h"
37 #include "mac-learning.h"
38 #include "meta-flow.h"
39 #include "multipath.h"
40 #include "netdev-vport.h"
45 #include "odp-execute.h"
48 #include "ofp-actions.h"
49 #include "ofp-parse.h"
50 #include "ofp-print.h"
51 #include "ofproto-dpif-governor.h"
52 #include "ofproto-dpif-ipfix.h"
53 #include "ofproto-dpif-sflow.h"
54 #include "ofproto-dpif-xlate.h"
55 #include "poll-loop.h"
60 #include "unaligned.h"
62 #include "vlan-bitmap.h"
65 VLOG_DEFINE_THIS_MODULE(ofproto_dpif);
67 COVERAGE_DEFINE(ofproto_dpif_expired);
68 COVERAGE_DEFINE(facet_changed_rule);
69 COVERAGE_DEFINE(facet_revalidate);
70 COVERAGE_DEFINE(facet_unexpected);
71 COVERAGE_DEFINE(facet_suppress);
76 static struct rule_dpif *rule_dpif_lookup(struct ofproto_dpif *,
78 struct flow_wildcards *wc);
80 static void rule_get_stats(struct rule *, uint64_t *packets, uint64_t *bytes);
81 static void rule_invalidate(const struct rule_dpif *);
83 static void mirror_destroy(struct ofmirror *);
84 static void update_mirror_stats(struct ofproto_dpif *ofproto,
85 mirror_mask_t mirrors,
86 uint64_t packets, uint64_t bytes);
88 static void bundle_remove(struct ofport *);
89 static void bundle_update(struct ofbundle *);
90 static void bundle_destroy(struct ofbundle *);
91 static void bundle_del_port(struct ofport_dpif *);
92 static void bundle_run(struct ofbundle *);
93 static void bundle_wait(struct ofbundle *);
95 static void stp_run(struct ofproto_dpif *ofproto);
96 static void stp_wait(struct ofproto_dpif *ofproto);
97 static int set_stp_port(struct ofport *,
98 const struct ofproto_port_stp_settings *);
100 static void compose_slow_path(const struct ofproto_dpif *, const struct flow *,
101 enum slow_path_reason,
102 uint64_t *stub, size_t stub_size,
103 const struct nlattr **actionsp,
104 size_t *actions_lenp);
106 /* A subfacet (see "struct subfacet" below) has three possible installation
109 * - SF_NOT_INSTALLED: Not installed in the datapath. This will only be the
110 * case just after the subfacet is created, just before the subfacet is
111 * destroyed, or if the datapath returns an error when we try to install a
114 * - SF_FAST_PATH: The subfacet's actions are installed in the datapath.
116 * - SF_SLOW_PATH: An action that sends every packet for the subfacet through
117 * ofproto_dpif is installed in the datapath.
120 SF_NOT_INSTALLED, /* No datapath flow for this subfacet. */
121 SF_FAST_PATH, /* Full actions are installed. */
122 SF_SLOW_PATH, /* Send-to-userspace action is installed. */
125 /* A dpif flow and actions associated with a facet.
127 * See also the large comment on struct facet. */
130 struct hmap_node hmap_node; /* In struct ofproto_dpif 'subfacets' list. */
131 struct list list_node; /* In struct facet's 'facets' list. */
132 struct facet *facet; /* Owning facet. */
133 struct dpif_backer *backer; /* Owning backer. */
135 enum odp_key_fitness key_fitness;
139 long long int used; /* Time last used; time created if not used. */
140 long long int created; /* Time created. */
142 uint64_t dp_packet_count; /* Last known packet count in the datapath. */
143 uint64_t dp_byte_count; /* Last known byte count in the datapath. */
145 enum subfacet_path path; /* Installed in datapath? */
148 #define SUBFACET_DESTROY_MAX_BATCH 50
150 static struct subfacet *subfacet_create(struct facet *, struct flow_miss *miss,
152 static struct subfacet *subfacet_find(struct dpif_backer *,
153 const struct nlattr *key, size_t key_len,
155 static void subfacet_destroy(struct subfacet *);
156 static void subfacet_destroy__(struct subfacet *);
157 static void subfacet_destroy_batch(struct dpif_backer *,
158 struct subfacet **, int n);
159 static void subfacet_reset_dp_stats(struct subfacet *,
160 struct dpif_flow_stats *);
161 static void subfacet_update_stats(struct subfacet *,
162 const struct dpif_flow_stats *);
163 static int subfacet_install(struct subfacet *,
164 const struct ofpbuf *odp_actions,
165 struct dpif_flow_stats *);
166 static void subfacet_uninstall(struct subfacet *);
168 /* A unique, non-overlapping instantiation of an OpenFlow flow.
170 * A facet associates a "struct flow", which represents the Open vSwitch
171 * userspace idea of an exact-match flow, with one or more subfacets.
172 * While the facet is created based on an exact-match flow, it is stored
173 * within the ofproto based on the wildcards that could be expressed
174 * based on the flow table and other configuration. (See the 'wc'
175 * description in "struct xlate_out" for more details.)
177 * Each subfacet tracks the datapath's idea of the flow equivalent to
178 * the facet. When the kernel module (or other dpif implementation) and
179 * Open vSwitch userspace agree on the definition of a flow key, there
180 * is exactly one subfacet per facet. If the dpif implementation
181 * supports more-specific flow matching than userspace, however, a facet
182 * can have more than one subfacet. Examples include the dpif
183 * implementation not supporting the same wildcards as userspace or some
184 * distinction in flow that userspace simply doesn't understand.
186 * Flow expiration works in terms of subfacets, so a facet must have at
187 * least one subfacet or it will never expire, leaking memory. */
190 struct hmap_node hmap_node; /* In owning ofproto's 'facets' hmap. */
191 struct list list_node; /* In owning rule's 'facets' list. */
192 struct rule_dpif *rule; /* Owning rule. */
195 struct list subfacets;
196 long long int used; /* Time last used; time created if not used. */
199 struct flow flow; /* Flow of the creating subfacet. */
200 struct cls_rule cr; /* In 'ofproto_dpif's facets classifier. */
204 * - Do include packets and bytes sent "by hand", e.g. with
207 * - Do include packets and bytes that were obtained from the datapath
208 * when a subfacet's statistics were reset (e.g. dpif_flow_put() with
209 * DPIF_FP_ZERO_STATS).
211 * - Do not include packets or bytes that can be obtained from the
212 * datapath for any existing subfacet.
214 uint64_t packet_count; /* Number of packets received. */
215 uint64_t byte_count; /* Number of bytes received. */
217 /* Resubmit statistics. */
218 uint64_t prev_packet_count; /* Number of packets from last stats push. */
219 uint64_t prev_byte_count; /* Number of bytes from last stats push. */
220 long long int prev_used; /* Used time from last stats push. */
223 uint64_t accounted_bytes; /* Bytes processed by facet_account(). */
224 struct netflow_flow nf_flow; /* Per-flow NetFlow tracking data. */
225 uint8_t tcp_flags; /* TCP flags seen for this 'rule'. */
227 struct xlate_out xout;
229 /* Storage for a single subfacet, to reduce malloc() time and space
230 * overhead. (A facet always has at least one subfacet and in the common
231 * case has exactly one subfacet. However, 'one_subfacet' may not
232 * always be valid, since it could have been removed after newer
233 * subfacets were pushed onto the 'subfacets' list.) */
234 struct subfacet one_subfacet;
236 long long int learn_rl; /* Rate limiter for facet_learn(). */
239 static struct facet *facet_create(const struct flow_miss *, struct rule_dpif *,
241 struct dpif_flow_stats *);
242 static void facet_remove(struct facet *);
243 static void facet_free(struct facet *);
245 static struct facet *facet_find(struct ofproto_dpif *, const struct flow *);
246 static struct facet *facet_lookup_valid(struct ofproto_dpif *,
247 const struct flow *);
248 static bool facet_revalidate(struct facet *);
249 static bool facet_check_consistency(struct facet *);
251 static void facet_flush_stats(struct facet *);
253 static void facet_reset_counters(struct facet *);
254 static void facet_push_stats(struct facet *, bool may_learn);
255 static void facet_learn(struct facet *);
256 static void facet_account(struct facet *);
257 static void push_all_stats(void);
259 static bool facet_is_controller_flow(struct facet *);
261 /* Linux VLAN device support (e.g. "eth0.10" for VLAN 10.)
263 * This is deprecated. It is only for compatibility with broken device drivers
264 * in old versions of Linux that do not properly support VLANs when VLAN
265 * devices are not used. When broken device drivers are no longer in
266 * widespread use, we will delete these interfaces. */
267 struct vlan_splinter {
268 struct hmap_node realdev_vid_node;
269 struct hmap_node vlandev_node;
270 uint16_t realdev_ofp_port;
271 uint16_t vlandev_ofp_port;
275 static bool vsp_adjust_flow(const struct ofproto_dpif *, struct flow *);
276 static void vsp_remove(struct ofport_dpif *);
277 static void vsp_add(struct ofport_dpif *, uint16_t realdev_ofp_port, int vid);
279 static uint16_t odp_port_to_ofp_port(const struct ofproto_dpif *,
282 static struct ofport_dpif *
283 ofport_dpif_cast(const struct ofport *ofport)
285 return ofport ? CONTAINER_OF(ofport, struct ofport_dpif, up) : NULL;
288 static void port_run(struct ofport_dpif *);
289 static void port_run_fast(struct ofport_dpif *);
290 static void port_wait(struct ofport_dpif *);
291 static int set_bfd(struct ofport *, const struct smap *);
292 static int set_cfm(struct ofport *, const struct cfm_settings *);
293 static void ofport_clear_priorities(struct ofport_dpif *);
294 static void run_fast_rl(void);
296 struct dpif_completion {
297 struct list list_node;
298 struct ofoperation *op;
301 COVERAGE_DEFINE(rev_reconfigure);
302 COVERAGE_DEFINE(rev_stp);
303 COVERAGE_DEFINE(rev_port_toggled);
304 COVERAGE_DEFINE(rev_flow_table);
305 COVERAGE_DEFINE(rev_inconsistency);
307 /* Drop keys are odp flow keys which have drop flows installed in the kernel.
308 * These are datapath flows which have no associated ofproto, if they did we
309 * would use facets. */
311 struct hmap_node hmap_node;
316 /* All existing ofproto_backer instances, indexed by ofproto->up.type. */
317 static struct shash all_dpif_backers = SHASH_INITIALIZER(&all_dpif_backers);
319 static void drop_key_clear(struct dpif_backer *);
320 static struct ofport_dpif *
321 odp_port_to_ofport(const struct dpif_backer *, uint32_t odp_port);
322 static void update_moving_averages(struct dpif_backer *backer);
324 /* Defer flow mod completion until "ovs-appctl ofproto/unclog"? (Useful only
325 * for debugging the asynchronous flow_mod implementation.) */
328 /* All existing ofproto_dpif instances, indexed by ->up.name. */
329 static struct hmap all_ofproto_dpifs = HMAP_INITIALIZER(&all_ofproto_dpifs);
331 static void ofproto_dpif_unixctl_init(void);
334 #define FLOW_MISS_MAX_BATCH 50
335 static int handle_upcalls(struct dpif_backer *, unsigned int max_batch);
337 /* Flow expiration. */
338 static int expire(struct dpif_backer *);
341 static void send_netflow_active_timeouts(struct ofproto_dpif *);
344 static int send_packet(const struct ofport_dpif *, struct ofpbuf *packet);
346 /* Global variables. */
347 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
349 /* Initial mappings of port to bridge mappings. */
350 static struct shash init_ofp_ports = SHASH_INITIALIZER(&init_ofp_ports);
352 /* Factory functions. */
355 init(const struct shash *iface_hints)
357 struct shash_node *node;
359 /* Make a local copy, since we don't own 'iface_hints' elements. */
360 SHASH_FOR_EACH(node, iface_hints) {
361 const struct iface_hint *orig_hint = node->data;
362 struct iface_hint *new_hint = xmalloc(sizeof *new_hint);
364 new_hint->br_name = xstrdup(orig_hint->br_name);
365 new_hint->br_type = xstrdup(orig_hint->br_type);
366 new_hint->ofp_port = orig_hint->ofp_port;
368 shash_add(&init_ofp_ports, node->name, new_hint);
373 enumerate_types(struct sset *types)
375 dp_enumerate_types(types);
379 enumerate_names(const char *type, struct sset *names)
381 struct ofproto_dpif *ofproto;
384 HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
385 if (strcmp(type, ofproto->up.type)) {
388 sset_add(names, ofproto->up.name);
395 del(const char *type, const char *name)
400 error = dpif_open(name, type, &dpif);
402 error = dpif_delete(dpif);
409 port_open_type(const char *datapath_type, const char *port_type)
411 return dpif_port_open_type(datapath_type, port_type);
414 /* Type functions. */
416 static struct ofproto_dpif *
417 lookup_ofproto_dpif_by_port_name(const char *name)
419 struct ofproto_dpif *ofproto;
421 HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
422 if (sset_contains(&ofproto->ports, name)) {
431 type_run(const char *type)
433 static long long int push_timer = LLONG_MIN;
434 struct dpif_backer *backer;
438 backer = shash_find_data(&all_dpif_backers, type);
440 /* This is not necessarily a problem, since backers are only
441 * created on demand. */
445 dpif_run(backer->dpif);
447 /* The most natural place to push facet statistics is when they're pulled
448 * from the datapath. However, when there are many flows in the datapath,
449 * this expensive operation can occur so frequently, that it reduces our
450 * ability to quickly set up flows. To reduce the cost, we push statistics
452 if (time_msec() > push_timer) {
453 push_timer = time_msec() + 2000;
457 /* If vswitchd started with other_config:flow_restore_wait set as "true",
458 * and the configuration has now changed to "false", enable receiving
459 * packets from the datapath. */
460 if (!backer->recv_set_enable && !ofproto_get_flow_restore_wait()) {
461 backer->recv_set_enable = true;
463 error = dpif_recv_set(backer->dpif, backer->recv_set_enable);
465 VLOG_ERR("Failed to enable receiving packets in dpif.");
468 dpif_flow_flush(backer->dpif);
469 backer->need_revalidate = REV_RECONFIGURE;
472 if (backer->need_revalidate
473 || !tag_set_is_empty(&backer->revalidate_set)) {
474 struct tag_set revalidate_set = backer->revalidate_set;
475 bool need_revalidate = backer->need_revalidate;
476 struct ofproto_dpif *ofproto;
477 struct simap_node *node;
478 struct simap tmp_backers;
480 /* Handle tunnel garbage collection. */
481 simap_init(&tmp_backers);
482 simap_swap(&backer->tnl_backers, &tmp_backers);
484 HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
485 struct ofport_dpif *iter;
487 if (backer != ofproto->backer) {
491 HMAP_FOR_EACH (iter, up.hmap_node, &ofproto->up.ports) {
492 char namebuf[NETDEV_VPORT_NAME_BUFSIZE];
495 if (!iter->tnl_port) {
499 dp_port = netdev_vport_get_dpif_port(iter->up.netdev,
500 namebuf, sizeof namebuf);
501 node = simap_find(&tmp_backers, dp_port);
503 simap_put(&backer->tnl_backers, dp_port, node->data);
504 simap_delete(&tmp_backers, node);
505 node = simap_find(&backer->tnl_backers, dp_port);
507 node = simap_find(&backer->tnl_backers, dp_port);
509 uint32_t odp_port = UINT32_MAX;
511 if (!dpif_port_add(backer->dpif, iter->up.netdev,
513 simap_put(&backer->tnl_backers, dp_port, odp_port);
514 node = simap_find(&backer->tnl_backers, dp_port);
519 iter->odp_port = node ? node->data : OVSP_NONE;
520 if (tnl_port_reconfigure(&iter->up, iter->odp_port,
522 backer->need_revalidate = REV_RECONFIGURE;
527 SIMAP_FOR_EACH (node, &tmp_backers) {
528 dpif_port_del(backer->dpif, node->data);
530 simap_destroy(&tmp_backers);
532 switch (backer->need_revalidate) {
533 case REV_RECONFIGURE: COVERAGE_INC(rev_reconfigure); break;
534 case REV_STP: COVERAGE_INC(rev_stp); break;
535 case REV_PORT_TOGGLED: COVERAGE_INC(rev_port_toggled); break;
536 case REV_FLOW_TABLE: COVERAGE_INC(rev_flow_table); break;
537 case REV_INCONSISTENCY: COVERAGE_INC(rev_inconsistency); break;
540 if (backer->need_revalidate) {
541 /* Clear the drop_keys in case we should now be accepting some
542 * formerly dropped flows. */
543 drop_key_clear(backer);
546 /* Clear the revalidation flags. */
547 tag_set_init(&backer->revalidate_set);
548 backer->need_revalidate = 0;
550 HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
551 struct facet *facet, *next;
552 struct cls_cursor cursor;
554 if (ofproto->backer != backer) {
558 cls_cursor_init(&cursor, &ofproto->facets, NULL);
559 CLS_CURSOR_FOR_EACH_SAFE (facet, next, cr, &cursor) {
561 || tag_set_intersects(&revalidate_set, facet->xout.tags)) {
562 facet_revalidate(facet);
569 if (!backer->recv_set_enable) {
570 /* Wake up before a max of 1000ms. */
571 timer_set_duration(&backer->next_expiration, 1000);
572 } else if (timer_expired(&backer->next_expiration)) {
573 int delay = expire(backer);
574 timer_set_duration(&backer->next_expiration, delay);
577 /* Check for port changes in the dpif. */
578 while ((error = dpif_port_poll(backer->dpif, &devname)) == 0) {
579 struct ofproto_dpif *ofproto;
580 struct dpif_port port;
582 /* Don't report on the datapath's device. */
583 if (!strcmp(devname, dpif_base_name(backer->dpif))) {
587 HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node,
588 &all_ofproto_dpifs) {
589 if (simap_contains(&ofproto->backer->tnl_backers, devname)) {
594 ofproto = lookup_ofproto_dpif_by_port_name(devname);
595 if (dpif_port_query_by_name(backer->dpif, devname, &port)) {
596 /* The port was removed. If we know the datapath,
597 * report it through poll_set(). If we don't, it may be
598 * notifying us of a removal we initiated, so ignore it.
599 * If there's a pending ENOBUFS, let it stand, since
600 * everything will be reevaluated. */
601 if (ofproto && ofproto->port_poll_errno != ENOBUFS) {
602 sset_add(&ofproto->port_poll_set, devname);
603 ofproto->port_poll_errno = 0;
605 } else if (!ofproto) {
606 /* The port was added, but we don't know with which
607 * ofproto we should associate it. Delete it. */
608 dpif_port_del(backer->dpif, port.port_no);
610 dpif_port_destroy(&port);
616 if (error != EAGAIN) {
617 struct ofproto_dpif *ofproto;
619 /* There was some sort of error, so propagate it to all
620 * ofprotos that use this backer. */
621 HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node,
622 &all_ofproto_dpifs) {
623 if (ofproto->backer == backer) {
624 sset_clear(&ofproto->port_poll_set);
625 ofproto->port_poll_errno = error;
630 if (backer->governor) {
633 governor_run(backer->governor);
635 /* If the governor has shrunk to its minimum size and the number of
636 * subfacets has dwindled, then drop the governor entirely.
638 * For hysteresis, the number of subfacets to drop the governor is
639 * smaller than the number needed to trigger its creation. */
640 n_subfacets = hmap_count(&backer->subfacets);
641 if (n_subfacets * 4 < flow_eviction_threshold
642 && governor_is_idle(backer->governor)) {
643 governor_destroy(backer->governor);
644 backer->governor = NULL;
652 dpif_backer_run_fast(struct dpif_backer *backer, int max_batch)
656 /* If recv_set_enable is false, we should not handle upcalls. */
657 if (!backer->recv_set_enable) {
661 /* Handle one or more batches of upcalls, until there's nothing left to do
662 * or until we do a fixed total amount of work.
664 * We do work in batches because it can be much cheaper to set up a number
665 * of flows and fire off their patches all at once. We do multiple batches
666 * because in some cases handling a packet can cause another packet to be
667 * queued almost immediately as part of the return flow. Both
668 * optimizations can make major improvements on some benchmarks and
669 * presumably for real traffic as well. */
671 while (work < max_batch) {
672 int retval = handle_upcalls(backer, max_batch - work);
683 type_run_fast(const char *type)
685 struct dpif_backer *backer;
687 backer = shash_find_data(&all_dpif_backers, type);
689 /* This is not necessarily a problem, since backers are only
690 * created on demand. */
694 return dpif_backer_run_fast(backer, FLOW_MISS_MAX_BATCH);
700 static long long int port_rl = LLONG_MIN;
701 static unsigned int backer_rl = 0;
703 if (time_msec() >= port_rl) {
704 struct ofproto_dpif *ofproto;
705 struct ofport_dpif *ofport;
707 HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
709 HMAP_FOR_EACH (ofport, up.hmap_node, &ofproto->up.ports) {
710 port_run_fast(ofport);
713 port_rl = time_msec() + 200;
716 /* XXX: We have to be careful not to do too much work in this function. If
717 * we call dpif_backer_run_fast() too often, or with too large a batch,
718 * performance improves signifcantly, but at a cost. It's possible for the
719 * number of flows in the datapath to increase without bound, and for poll
720 * loops to take 10s of seconds. The correct solution to this problem,
721 * long term, is to separate flow miss handling into it's own thread so it
722 * isn't affected by revalidations, and expirations. Until then, this is
723 * the best we can do. */
724 if (++backer_rl >= 10) {
725 struct shash_node *node;
728 SHASH_FOR_EACH (node, &all_dpif_backers) {
729 dpif_backer_run_fast(node->data, 1);
735 type_wait(const char *type)
737 struct dpif_backer *backer;
739 backer = shash_find_data(&all_dpif_backers, type);
741 /* This is not necessarily a problem, since backers are only
742 * created on demand. */
746 if (backer->governor) {
747 governor_wait(backer->governor);
750 timer_wait(&backer->next_expiration);
753 /* Basic life-cycle. */
755 static int add_internal_flows(struct ofproto_dpif *);
757 static struct ofproto *
760 struct ofproto_dpif *ofproto = xmalloc(sizeof *ofproto);
765 dealloc(struct ofproto *ofproto_)
767 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
772 close_dpif_backer(struct dpif_backer *backer)
774 struct shash_node *node;
776 ovs_assert(backer->refcount > 0);
778 if (--backer->refcount) {
782 drop_key_clear(backer);
783 hmap_destroy(&backer->drop_keys);
785 simap_destroy(&backer->tnl_backers);
786 hmap_destroy(&backer->odp_to_ofport_map);
787 node = shash_find(&all_dpif_backers, backer->type);
789 shash_delete(&all_dpif_backers, node);
790 dpif_close(backer->dpif);
792 ovs_assert(hmap_is_empty(&backer->subfacets));
793 hmap_destroy(&backer->subfacets);
794 governor_destroy(backer->governor);
799 /* Datapath port slated for removal from datapath. */
801 struct list list_node;
806 open_dpif_backer(const char *type, struct dpif_backer **backerp)
808 struct dpif_backer *backer;
809 struct dpif_port_dump port_dump;
810 struct dpif_port port;
811 struct shash_node *node;
812 struct list garbage_list;
813 struct odp_garbage *garbage, *next;
819 backer = shash_find_data(&all_dpif_backers, type);
826 backer_name = xasprintf("ovs-%s", type);
828 /* Remove any existing datapaths, since we assume we're the only
829 * userspace controlling the datapath. */
831 dp_enumerate_names(type, &names);
832 SSET_FOR_EACH(name, &names) {
833 struct dpif *old_dpif;
835 /* Don't remove our backer if it exists. */
836 if (!strcmp(name, backer_name)) {
840 if (dpif_open(name, type, &old_dpif)) {
841 VLOG_WARN("couldn't open old datapath %s to remove it", name);
843 dpif_delete(old_dpif);
844 dpif_close(old_dpif);
847 sset_destroy(&names);
849 backer = xmalloc(sizeof *backer);
851 error = dpif_create_and_open(backer_name, type, &backer->dpif);
854 VLOG_ERR("failed to open datapath of type %s: %s", type,
860 backer->type = xstrdup(type);
861 backer->governor = NULL;
862 backer->refcount = 1;
863 hmap_init(&backer->odp_to_ofport_map);
864 hmap_init(&backer->drop_keys);
865 hmap_init(&backer->subfacets);
866 timer_set_duration(&backer->next_expiration, 1000);
867 backer->need_revalidate = 0;
868 simap_init(&backer->tnl_backers);
869 tag_set_init(&backer->revalidate_set);
870 backer->recv_set_enable = !ofproto_get_flow_restore_wait();
873 if (backer->recv_set_enable) {
874 dpif_flow_flush(backer->dpif);
877 /* Loop through the ports already on the datapath and remove any
878 * that we don't need anymore. */
879 list_init(&garbage_list);
880 dpif_port_dump_start(&port_dump, backer->dpif);
881 while (dpif_port_dump_next(&port_dump, &port)) {
882 node = shash_find(&init_ofp_ports, port.name);
883 if (!node && strcmp(port.name, dpif_base_name(backer->dpif))) {
884 garbage = xmalloc(sizeof *garbage);
885 garbage->odp_port = port.port_no;
886 list_push_front(&garbage_list, &garbage->list_node);
889 dpif_port_dump_done(&port_dump);
891 LIST_FOR_EACH_SAFE (garbage, next, list_node, &garbage_list) {
892 dpif_port_del(backer->dpif, garbage->odp_port);
893 list_remove(&garbage->list_node);
897 shash_add(&all_dpif_backers, type, backer);
899 error = dpif_recv_set(backer->dpif, backer->recv_set_enable);
901 VLOG_ERR("failed to listen on datapath of type %s: %s",
902 type, strerror(error));
903 close_dpif_backer(backer);
907 backer->max_n_subfacet = 0;
908 backer->created = time_msec();
909 backer->last_minute = backer->created;
910 memset(&backer->hourly, 0, sizeof backer->hourly);
911 memset(&backer->daily, 0, sizeof backer->daily);
912 memset(&backer->lifetime, 0, sizeof backer->lifetime);
913 backer->subfacet_add_count = 0;
914 backer->subfacet_del_count = 0;
915 backer->total_subfacet_add_count = 0;
916 backer->total_subfacet_del_count = 0;
917 backer->avg_n_subfacet = 0;
918 backer->avg_subfacet_life = 0;
924 construct(struct ofproto *ofproto_)
926 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
927 struct shash_node *node, *next;
932 error = open_dpif_backer(ofproto->up.type, &ofproto->backer);
937 max_ports = dpif_get_max_ports(ofproto->backer->dpif);
938 ofproto_init_max_ports(ofproto_, MIN(max_ports, OFPP_MAX));
940 ofproto->netflow = NULL;
941 ofproto->sflow = NULL;
942 ofproto->ipfix = NULL;
944 hmap_init(&ofproto->bundles);
945 ofproto->ml = mac_learning_create(MAC_ENTRY_DEFAULT_IDLE_TIME);
946 for (i = 0; i < MAX_MIRRORS; i++) {
947 ofproto->mirrors[i] = NULL;
949 ofproto->has_bonded_bundles = false;
951 classifier_init(&ofproto->facets);
952 ofproto->consistency_rl = LLONG_MIN;
954 for (i = 0; i < N_TABLES; i++) {
955 struct table_dpif *table = &ofproto->tables[i];
957 table->catchall_table = NULL;
958 table->other_table = NULL;
959 table->basis = random_uint32();
962 list_init(&ofproto->completions);
964 ofproto_dpif_unixctl_init();
966 ofproto->has_mirrors = false;
967 ofproto->has_bundle_action = false;
969 hmap_init(&ofproto->vlandev_map);
970 hmap_init(&ofproto->realdev_vid_map);
972 sset_init(&ofproto->ports);
973 sset_init(&ofproto->ghost_ports);
974 sset_init(&ofproto->port_poll_set);
975 ofproto->port_poll_errno = 0;
977 SHASH_FOR_EACH_SAFE (node, next, &init_ofp_ports) {
978 struct iface_hint *iface_hint = node->data;
980 if (!strcmp(iface_hint->br_name, ofproto->up.name)) {
981 /* Check if the datapath already has this port. */
982 if (dpif_port_exists(ofproto->backer->dpif, node->name)) {
983 sset_add(&ofproto->ports, node->name);
986 free(iface_hint->br_name);
987 free(iface_hint->br_type);
989 shash_delete(&init_ofp_ports, node);
993 hmap_insert(&all_ofproto_dpifs, &ofproto->all_ofproto_dpifs_node,
994 hash_string(ofproto->up.name, 0));
995 memset(&ofproto->stats, 0, sizeof ofproto->stats);
997 ofproto_init_tables(ofproto_, N_TABLES);
998 error = add_internal_flows(ofproto);
999 ofproto->up.tables[TBL_INTERNAL].flags = OFTABLE_HIDDEN | OFTABLE_READONLY;
1002 ofproto->n_missed = 0;
1008 add_internal_flow(struct ofproto_dpif *ofproto, int id,
1009 const struct ofpbuf *ofpacts, struct rule_dpif **rulep)
1011 struct ofputil_flow_mod fm;
1014 match_init_catchall(&fm.match);
1016 match_set_reg(&fm.match, 0, id);
1017 fm.new_cookie = htonll(0);
1018 fm.cookie = htonll(0);
1019 fm.cookie_mask = htonll(0);
1020 fm.table_id = TBL_INTERNAL;
1021 fm.command = OFPFC_ADD;
1022 fm.idle_timeout = 0;
1023 fm.hard_timeout = 0;
1027 fm.ofpacts = ofpacts->data;
1028 fm.ofpacts_len = ofpacts->size;
1030 error = ofproto_flow_mod(&ofproto->up, &fm);
1032 VLOG_ERR_RL(&rl, "failed to add internal flow %d (%s)",
1033 id, ofperr_to_string(error));
1037 *rulep = rule_dpif_lookup_in_table(ofproto, &fm.match.flow, NULL,
1039 ovs_assert(*rulep != NULL);
1045 add_internal_flows(struct ofproto_dpif *ofproto)
1047 struct ofpact_controller *controller;
1048 uint64_t ofpacts_stub[128 / 8];
1049 struct ofpbuf ofpacts;
1053 ofpbuf_use_stack(&ofpacts, ofpacts_stub, sizeof ofpacts_stub);
1056 controller = ofpact_put_CONTROLLER(&ofpacts);
1057 controller->max_len = UINT16_MAX;
1058 controller->controller_id = 0;
1059 controller->reason = OFPR_NO_MATCH;
1060 ofpact_pad(&ofpacts);
1062 error = add_internal_flow(ofproto, id++, &ofpacts, &ofproto->miss_rule);
1067 ofpbuf_clear(&ofpacts);
1068 error = add_internal_flow(ofproto, id++, &ofpacts,
1069 &ofproto->no_packet_in_rule);
1074 error = add_internal_flow(ofproto, id++, &ofpacts,
1075 &ofproto->drop_frags_rule);
1080 complete_operations(struct ofproto_dpif *ofproto)
1082 struct dpif_completion *c, *next;
1084 LIST_FOR_EACH_SAFE (c, next, list_node, &ofproto->completions) {
1085 ofoperation_complete(c->op, 0);
1086 list_remove(&c->list_node);
1092 destruct(struct ofproto *ofproto_)
1094 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1095 struct rule_dpif *rule, *next_rule;
1096 struct oftable *table;
1099 hmap_remove(&all_ofproto_dpifs, &ofproto->all_ofproto_dpifs_node);
1100 complete_operations(ofproto);
1102 OFPROTO_FOR_EACH_TABLE (table, &ofproto->up) {
1103 struct cls_cursor cursor;
1105 cls_cursor_init(&cursor, &table->cls, NULL);
1106 CLS_CURSOR_FOR_EACH_SAFE (rule, next_rule, up.cr, &cursor) {
1107 ofproto_rule_destroy(&rule->up);
1111 for (i = 0; i < MAX_MIRRORS; i++) {
1112 mirror_destroy(ofproto->mirrors[i]);
1115 netflow_destroy(ofproto->netflow);
1116 dpif_sflow_destroy(ofproto->sflow);
1117 hmap_destroy(&ofproto->bundles);
1118 mac_learning_destroy(ofproto->ml);
1120 classifier_destroy(&ofproto->facets);
1122 hmap_destroy(&ofproto->vlandev_map);
1123 hmap_destroy(&ofproto->realdev_vid_map);
1125 sset_destroy(&ofproto->ports);
1126 sset_destroy(&ofproto->ghost_ports);
1127 sset_destroy(&ofproto->port_poll_set);
1129 close_dpif_backer(ofproto->backer);
1133 run_fast(struct ofproto *ofproto_)
1135 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1136 struct ofport_dpif *ofport;
1138 /* Do not perform any periodic activity required by 'ofproto' while
1139 * waiting for flow restore to complete. */
1140 if (ofproto_get_flow_restore_wait()) {
1144 HMAP_FOR_EACH (ofport, up.hmap_node, &ofproto->up.ports) {
1145 port_run_fast(ofport);
1152 run(struct ofproto *ofproto_)
1154 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1155 struct ofport_dpif *ofport;
1156 struct ofbundle *bundle;
1160 complete_operations(ofproto);
1163 /* Do not perform any periodic activity below required by 'ofproto' while
1164 * waiting for flow restore to complete. */
1165 if (ofproto_get_flow_restore_wait()) {
1169 error = run_fast(ofproto_);
1174 if (ofproto->netflow) {
1175 if (netflow_run(ofproto->netflow)) {
1176 send_netflow_active_timeouts(ofproto);
1179 if (ofproto->sflow) {
1180 dpif_sflow_run(ofproto->sflow);
1183 HMAP_FOR_EACH (ofport, up.hmap_node, &ofproto->up.ports) {
1186 HMAP_FOR_EACH (bundle, hmap_node, &ofproto->bundles) {
1191 mac_learning_run(ofproto->ml, &ofproto->backer->revalidate_set);
1193 /* Check the consistency of a random facet, to aid debugging. */
1194 if (time_msec() >= ofproto->consistency_rl
1195 && !classifier_is_empty(&ofproto->facets)
1196 && !ofproto->backer->need_revalidate) {
1197 struct cls_table *table;
1198 struct cls_rule *cr;
1199 struct facet *facet;
1201 ofproto->consistency_rl = time_msec() + 250;
1203 table = CONTAINER_OF(hmap_random_node(&ofproto->facets.tables),
1204 struct cls_table, hmap_node);
1205 cr = CONTAINER_OF(hmap_random_node(&table->rules), struct cls_rule,
1207 facet = CONTAINER_OF(cr, struct facet, cr);
1209 if (!tag_set_intersects(&ofproto->backer->revalidate_set,
1210 facet->xout.tags)) {
1211 if (!facet_check_consistency(facet)) {
1212 ofproto->backer->need_revalidate = REV_INCONSISTENCY;
1221 wait(struct ofproto *ofproto_)
1223 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1224 struct ofport_dpif *ofport;
1225 struct ofbundle *bundle;
1227 if (!clogged && !list_is_empty(&ofproto->completions)) {
1228 poll_immediate_wake();
1231 if (ofproto_get_flow_restore_wait()) {
1235 dpif_wait(ofproto->backer->dpif);
1236 dpif_recv_wait(ofproto->backer->dpif);
1237 if (ofproto->sflow) {
1238 dpif_sflow_wait(ofproto->sflow);
1240 if (!tag_set_is_empty(&ofproto->backer->revalidate_set)) {
1241 poll_immediate_wake();
1243 HMAP_FOR_EACH (ofport, up.hmap_node, &ofproto->up.ports) {
1246 HMAP_FOR_EACH (bundle, hmap_node, &ofproto->bundles) {
1247 bundle_wait(bundle);
1249 if (ofproto->netflow) {
1250 netflow_wait(ofproto->netflow);
1252 mac_learning_wait(ofproto->ml);
1254 if (ofproto->backer->need_revalidate) {
1255 /* Shouldn't happen, but if it does just go around again. */
1256 VLOG_DBG_RL(&rl, "need revalidate in ofproto_wait_cb()");
1257 poll_immediate_wake();
1262 get_memory_usage(const struct ofproto *ofproto_, struct simap *usage)
1264 const struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1265 struct cls_cursor cursor;
1266 size_t n_subfacets = 0;
1267 struct facet *facet;
1269 simap_increase(usage, "facets", classifier_count(&ofproto->facets));
1271 cls_cursor_init(&cursor, &ofproto->facets, NULL);
1272 CLS_CURSOR_FOR_EACH (facet, cr, &cursor) {
1273 n_subfacets += list_size(&facet->subfacets);
1275 simap_increase(usage, "subfacets", n_subfacets);
1279 flush(struct ofproto *ofproto_)
1281 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1282 struct subfacet *subfacet, *next_subfacet;
1283 struct subfacet *batch[SUBFACET_DESTROY_MAX_BATCH];
1287 HMAP_FOR_EACH_SAFE (subfacet, next_subfacet, hmap_node,
1288 &ofproto->backer->subfacets) {
1289 if (ofproto_dpif_cast(subfacet->facet->rule->up.ofproto) != ofproto) {
1293 if (subfacet->path != SF_NOT_INSTALLED) {
1294 batch[n_batch++] = subfacet;
1295 if (n_batch >= SUBFACET_DESTROY_MAX_BATCH) {
1296 subfacet_destroy_batch(ofproto->backer, batch, n_batch);
1300 subfacet_destroy(subfacet);
1305 subfacet_destroy_batch(ofproto->backer, batch, n_batch);
1310 get_features(struct ofproto *ofproto_ OVS_UNUSED,
1311 bool *arp_match_ip, enum ofputil_action_bitmap *actions)
1313 *arp_match_ip = true;
1314 *actions = (OFPUTIL_A_OUTPUT |
1315 OFPUTIL_A_SET_VLAN_VID |
1316 OFPUTIL_A_SET_VLAN_PCP |
1317 OFPUTIL_A_STRIP_VLAN |
1318 OFPUTIL_A_SET_DL_SRC |
1319 OFPUTIL_A_SET_DL_DST |
1320 OFPUTIL_A_SET_NW_SRC |
1321 OFPUTIL_A_SET_NW_DST |
1322 OFPUTIL_A_SET_NW_TOS |
1323 OFPUTIL_A_SET_TP_SRC |
1324 OFPUTIL_A_SET_TP_DST |
1329 get_tables(struct ofproto *ofproto_, struct ofp12_table_stats *ots)
1331 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1332 struct dpif_dp_stats s;
1333 uint64_t n_miss, n_no_pkt_in, n_bytes, n_dropped_frags;
1336 strcpy(ots->name, "classifier");
1338 dpif_get_dp_stats(ofproto->backer->dpif, &s);
1339 rule_get_stats(&ofproto->miss_rule->up, &n_miss, &n_bytes);
1340 rule_get_stats(&ofproto->no_packet_in_rule->up, &n_no_pkt_in, &n_bytes);
1341 rule_get_stats(&ofproto->drop_frags_rule->up, &n_dropped_frags, &n_bytes);
1343 n_lookup = s.n_hit + s.n_missed - n_dropped_frags;
1344 ots->lookup_count = htonll(n_lookup);
1345 ots->matched_count = htonll(n_lookup - n_miss - n_no_pkt_in);
1348 static struct ofport *
1351 struct ofport_dpif *port = xmalloc(sizeof *port);
1356 port_dealloc(struct ofport *port_)
1358 struct ofport_dpif *port = ofport_dpif_cast(port_);
1363 port_construct(struct ofport *port_)
1365 struct ofport_dpif *port = ofport_dpif_cast(port_);
1366 struct ofproto_dpif *ofproto = ofproto_dpif_cast(port->up.ofproto);
1367 const struct netdev *netdev = port->up.netdev;
1368 char namebuf[NETDEV_VPORT_NAME_BUFSIZE];
1369 struct dpif_port dpif_port;
1372 ofproto->backer->need_revalidate = REV_RECONFIGURE;
1373 port->bundle = NULL;
1376 port->tag = tag_create_random();
1377 port->may_enable = true;
1378 port->stp_port = NULL;
1379 port->stp_state = STP_DISABLED;
1380 port->tnl_port = NULL;
1381 hmap_init(&port->priorities);
1382 port->realdev_ofp_port = 0;
1383 port->vlandev_vid = 0;
1384 port->carrier_seq = netdev_get_carrier_resets(netdev);
1386 if (netdev_vport_is_patch(netdev)) {
1387 /* By bailing out here, we don't submit the port to the sFlow module
1388 * to be considered for counter polling export. This is correct
1389 * because the patch port represents an interface that sFlow considers
1390 * to be "internal" to the switch as a whole, and therefore not an
1391 * candidate for counter polling. */
1392 port->odp_port = OVSP_NONE;
1396 error = dpif_port_query_by_name(ofproto->backer->dpif,
1397 netdev_vport_get_dpif_port(netdev, namebuf,
1404 port->odp_port = dpif_port.port_no;
1406 if (netdev_get_tunnel_config(netdev)) {
1407 port->tnl_port = tnl_port_add(&port->up, port->odp_port);
1409 /* Sanity-check that a mapping doesn't already exist. This
1410 * shouldn't happen for non-tunnel ports. */
1411 if (odp_port_to_ofp_port(ofproto, port->odp_port) != OFPP_NONE) {
1412 VLOG_ERR("port %s already has an OpenFlow port number",
1414 dpif_port_destroy(&dpif_port);
1418 hmap_insert(&ofproto->backer->odp_to_ofport_map, &port->odp_port_node,
1419 hash_int(port->odp_port, 0));
1421 dpif_port_destroy(&dpif_port);
1423 if (ofproto->sflow) {
1424 dpif_sflow_add_port(ofproto->sflow, port_, port->odp_port);
1431 port_destruct(struct ofport *port_)
1433 struct ofport_dpif *port = ofport_dpif_cast(port_);
1434 struct ofproto_dpif *ofproto = ofproto_dpif_cast(port->up.ofproto);
1435 const char *devname = netdev_get_name(port->up.netdev);
1436 char namebuf[NETDEV_VPORT_NAME_BUFSIZE];
1437 const char *dp_port_name;
1439 dp_port_name = netdev_vport_get_dpif_port(port->up.netdev, namebuf,
1441 if (dpif_port_exists(ofproto->backer->dpif, dp_port_name)) {
1442 /* The underlying device is still there, so delete it. This
1443 * happens when the ofproto is being destroyed, since the caller
1444 * assumes that removal of attached ports will happen as part of
1446 if (!port->tnl_port) {
1447 dpif_port_del(ofproto->backer->dpif, port->odp_port);
1449 ofproto->backer->need_revalidate = REV_RECONFIGURE;
1452 if (port->odp_port != OVSP_NONE && !port->tnl_port) {
1453 hmap_remove(&ofproto->backer->odp_to_ofport_map, &port->odp_port_node);
1456 tnl_port_del(port->tnl_port);
1457 sset_find_and_delete(&ofproto->ports, devname);
1458 sset_find_and_delete(&ofproto->ghost_ports, devname);
1459 ofproto->backer->need_revalidate = REV_RECONFIGURE;
1460 bundle_remove(port_);
1461 set_cfm(port_, NULL);
1462 set_bfd(port_, NULL);
1463 if (ofproto->sflow) {
1464 dpif_sflow_del_port(ofproto->sflow, port->odp_port);
1467 ofport_clear_priorities(port);
1468 hmap_destroy(&port->priorities);
1472 port_modified(struct ofport *port_)
1474 struct ofport_dpif *port = ofport_dpif_cast(port_);
1476 if (port->bundle && port->bundle->bond) {
1477 bond_slave_set_netdev(port->bundle->bond, port, port->up.netdev);
1481 cfm_set_netdev(port->cfm, port->up.netdev);
1486 port_reconfigured(struct ofport *port_, enum ofputil_port_config old_config)
1488 struct ofport_dpif *port = ofport_dpif_cast(port_);
1489 struct ofproto_dpif *ofproto = ofproto_dpif_cast(port->up.ofproto);
1490 enum ofputil_port_config changed = old_config ^ port->up.pp.config;
1492 if (changed & (OFPUTIL_PC_NO_RECV | OFPUTIL_PC_NO_RECV_STP |
1493 OFPUTIL_PC_NO_FWD | OFPUTIL_PC_NO_FLOOD |
1494 OFPUTIL_PC_NO_PACKET_IN)) {
1495 ofproto->backer->need_revalidate = REV_RECONFIGURE;
1497 if (changed & OFPUTIL_PC_NO_FLOOD && port->bundle) {
1498 bundle_update(port->bundle);
1504 set_sflow(struct ofproto *ofproto_,
1505 const struct ofproto_sflow_options *sflow_options)
1507 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1508 struct dpif_sflow *ds = ofproto->sflow;
1510 if (sflow_options) {
1512 struct ofport_dpif *ofport;
1514 ds = ofproto->sflow = dpif_sflow_create();
1515 HMAP_FOR_EACH (ofport, up.hmap_node, &ofproto->up.ports) {
1516 dpif_sflow_add_port(ds, &ofport->up, ofport->odp_port);
1518 ofproto->backer->need_revalidate = REV_RECONFIGURE;
1520 dpif_sflow_set_options(ds, sflow_options);
1523 dpif_sflow_destroy(ds);
1524 ofproto->backer->need_revalidate = REV_RECONFIGURE;
1525 ofproto->sflow = NULL;
1533 struct ofproto *ofproto_,
1534 const struct ofproto_ipfix_bridge_exporter_options *bridge_exporter_options,
1535 const struct ofproto_ipfix_flow_exporter_options *flow_exporters_options,
1536 size_t n_flow_exporters_options)
1538 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1539 struct dpif_ipfix *di = ofproto->ipfix;
1541 if (bridge_exporter_options || flow_exporters_options) {
1543 di = ofproto->ipfix = dpif_ipfix_create();
1545 dpif_ipfix_set_options(
1546 di, bridge_exporter_options, flow_exporters_options,
1547 n_flow_exporters_options);
1550 dpif_ipfix_destroy(di);
1551 ofproto->ipfix = NULL;
1558 set_cfm(struct ofport *ofport_, const struct cfm_settings *s)
1560 struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
1567 struct ofproto_dpif *ofproto;
1569 ofproto = ofproto_dpif_cast(ofport->up.ofproto);
1570 ofproto->backer->need_revalidate = REV_RECONFIGURE;
1571 ofport->cfm = cfm_create(ofport->up.netdev);
1574 if (cfm_configure(ofport->cfm, s)) {
1580 cfm_destroy(ofport->cfm);
1586 get_cfm_status(const struct ofport *ofport_,
1587 struct ofproto_cfm_status *status)
1589 struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
1592 status->faults = cfm_get_fault(ofport->cfm);
1593 status->remote_opstate = cfm_get_opup(ofport->cfm);
1594 status->health = cfm_get_health(ofport->cfm);
1595 cfm_get_remote_mpids(ofport->cfm, &status->rmps, &status->n_rmps);
1603 set_bfd(struct ofport *ofport_, const struct smap *cfg)
1605 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport_->ofproto);
1606 struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
1610 ofport->bfd = bfd_configure(old, netdev_get_name(ofport->up.netdev), cfg);
1611 if (ofport->bfd != old) {
1612 ofproto->backer->need_revalidate = REV_RECONFIGURE;
1619 get_bfd_status(struct ofport *ofport_, struct smap *smap)
1621 struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
1624 bfd_get_status(ofport->bfd, smap);
1631 /* Spanning Tree. */
1634 send_bpdu_cb(struct ofpbuf *pkt, int port_num, void *ofproto_)
1636 struct ofproto_dpif *ofproto = ofproto_;
1637 struct stp_port *sp = stp_get_port(ofproto->stp, port_num);
1638 struct ofport_dpif *ofport;
1640 ofport = stp_port_get_aux(sp);
1642 VLOG_WARN_RL(&rl, "%s: cannot send BPDU on unknown port %d",
1643 ofproto->up.name, port_num);
1645 struct eth_header *eth = pkt->l2;
1647 netdev_get_etheraddr(ofport->up.netdev, eth->eth_src);
1648 if (eth_addr_is_zero(eth->eth_src)) {
1649 VLOG_WARN_RL(&rl, "%s: cannot send BPDU on port %d "
1650 "with unknown MAC", ofproto->up.name, port_num);
1652 send_packet(ofport, pkt);
1658 /* Configures STP on 'ofproto_' using the settings defined in 's'. */
1660 set_stp(struct ofproto *ofproto_, const struct ofproto_stp_settings *s)
1662 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1664 /* Only revalidate flows if the configuration changed. */
1665 if (!s != !ofproto->stp) {
1666 ofproto->backer->need_revalidate = REV_RECONFIGURE;
1670 if (!ofproto->stp) {
1671 ofproto->stp = stp_create(ofproto_->name, s->system_id,
1672 send_bpdu_cb, ofproto);
1673 ofproto->stp_last_tick = time_msec();
1676 stp_set_bridge_id(ofproto->stp, s->system_id);
1677 stp_set_bridge_priority(ofproto->stp, s->priority);
1678 stp_set_hello_time(ofproto->stp, s->hello_time);
1679 stp_set_max_age(ofproto->stp, s->max_age);
1680 stp_set_forward_delay(ofproto->stp, s->fwd_delay);
1682 struct ofport *ofport;
1684 HMAP_FOR_EACH (ofport, hmap_node, &ofproto->up.ports) {
1685 set_stp_port(ofport, NULL);
1688 stp_destroy(ofproto->stp);
1689 ofproto->stp = NULL;
1696 get_stp_status(struct ofproto *ofproto_, struct ofproto_stp_status *s)
1698 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1702 s->bridge_id = stp_get_bridge_id(ofproto->stp);
1703 s->designated_root = stp_get_designated_root(ofproto->stp);
1704 s->root_path_cost = stp_get_root_path_cost(ofproto->stp);
1713 update_stp_port_state(struct ofport_dpif *ofport)
1715 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport->up.ofproto);
1716 enum stp_state state;
1718 /* Figure out new state. */
1719 state = ofport->stp_port ? stp_port_get_state(ofport->stp_port)
1723 if (ofport->stp_state != state) {
1724 enum ofputil_port_state of_state;
1727 VLOG_DBG_RL(&rl, "port %s: STP state changed from %s to %s",
1728 netdev_get_name(ofport->up.netdev),
1729 stp_state_name(ofport->stp_state),
1730 stp_state_name(state));
1731 if (stp_learn_in_state(ofport->stp_state)
1732 != stp_learn_in_state(state)) {
1733 /* xxx Learning action flows should also be flushed. */
1734 mac_learning_flush(ofproto->ml,
1735 &ofproto->backer->revalidate_set);
1737 fwd_change = stp_forward_in_state(ofport->stp_state)
1738 != stp_forward_in_state(state);
1740 ofproto->backer->need_revalidate = REV_STP;
1741 ofport->stp_state = state;
1742 ofport->stp_state_entered = time_msec();
1744 if (fwd_change && ofport->bundle) {
1745 bundle_update(ofport->bundle);
1748 /* Update the STP state bits in the OpenFlow port description. */
1749 of_state = ofport->up.pp.state & ~OFPUTIL_PS_STP_MASK;
1750 of_state |= (state == STP_LISTENING ? OFPUTIL_PS_STP_LISTEN
1751 : state == STP_LEARNING ? OFPUTIL_PS_STP_LEARN
1752 : state == STP_FORWARDING ? OFPUTIL_PS_STP_FORWARD
1753 : state == STP_BLOCKING ? OFPUTIL_PS_STP_BLOCK
1755 ofproto_port_set_state(&ofport->up, of_state);
1759 /* Configures STP on 'ofport_' using the settings defined in 's'. The
1760 * caller is responsible for assigning STP port numbers and ensuring
1761 * there are no duplicates. */
1763 set_stp_port(struct ofport *ofport_,
1764 const struct ofproto_port_stp_settings *s)
1766 struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
1767 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport->up.ofproto);
1768 struct stp_port *sp = ofport->stp_port;
1770 if (!s || !s->enable) {
1772 ofport->stp_port = NULL;
1773 stp_port_disable(sp);
1774 update_stp_port_state(ofport);
1777 } else if (sp && stp_port_no(sp) != s->port_num
1778 && ofport == stp_port_get_aux(sp)) {
1779 /* The port-id changed, so disable the old one if it's not
1780 * already in use by another port. */
1781 stp_port_disable(sp);
1784 sp = ofport->stp_port = stp_get_port(ofproto->stp, s->port_num);
1785 stp_port_enable(sp);
1787 stp_port_set_aux(sp, ofport);
1788 stp_port_set_priority(sp, s->priority);
1789 stp_port_set_path_cost(sp, s->path_cost);
1791 update_stp_port_state(ofport);
1797 get_stp_port_status(struct ofport *ofport_,
1798 struct ofproto_port_stp_status *s)
1800 struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
1801 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport->up.ofproto);
1802 struct stp_port *sp = ofport->stp_port;
1804 if (!ofproto->stp || !sp) {
1810 s->port_id = stp_port_get_id(sp);
1811 s->state = stp_port_get_state(sp);
1812 s->sec_in_state = (time_msec() - ofport->stp_state_entered) / 1000;
1813 s->role = stp_port_get_role(sp);
1814 stp_port_get_counts(sp, &s->tx_count, &s->rx_count, &s->error_count);
1820 stp_run(struct ofproto_dpif *ofproto)
1823 long long int now = time_msec();
1824 long long int elapsed = now - ofproto->stp_last_tick;
1825 struct stp_port *sp;
1828 stp_tick(ofproto->stp, MIN(INT_MAX, elapsed));
1829 ofproto->stp_last_tick = now;
1831 while (stp_get_changed_port(ofproto->stp, &sp)) {
1832 struct ofport_dpif *ofport = stp_port_get_aux(sp);
1835 update_stp_port_state(ofport);
1839 if (stp_check_and_reset_fdb_flush(ofproto->stp)) {
1840 mac_learning_flush(ofproto->ml, &ofproto->backer->revalidate_set);
1846 stp_wait(struct ofproto_dpif *ofproto)
1849 poll_timer_wait(1000);
1853 /* Returns true if STP should process 'flow'. */
1855 stp_should_process_flow(const struct flow *flow)
1857 return eth_addr_equals(flow->dl_dst, eth_addr_stp);
1861 stp_process_packet(const struct ofport_dpif *ofport,
1862 const struct ofpbuf *packet)
1864 struct ofpbuf payload = *packet;
1865 struct eth_header *eth = payload.data;
1866 struct stp_port *sp = ofport->stp_port;
1868 /* Sink packets on ports that have STP disabled when the bridge has
1870 if (!sp || stp_port_get_state(sp) == STP_DISABLED) {
1874 /* Trim off padding on payload. */
1875 if (payload.size > ntohs(eth->eth_type) + ETH_HEADER_LEN) {
1876 payload.size = ntohs(eth->eth_type) + ETH_HEADER_LEN;
1879 if (ofpbuf_try_pull(&payload, ETH_HEADER_LEN + LLC_HEADER_LEN)) {
1880 stp_received_bpdu(sp, payload.data, payload.size);
1884 struct priority_to_dscp *
1885 get_priority(const struct ofport_dpif *ofport, uint32_t priority)
1887 struct priority_to_dscp *pdscp;
1890 hash = hash_int(priority, 0);
1891 HMAP_FOR_EACH_IN_BUCKET (pdscp, hmap_node, hash, &ofport->priorities) {
1892 if (pdscp->priority == priority) {
1900 ofport_clear_priorities(struct ofport_dpif *ofport)
1902 struct priority_to_dscp *pdscp, *next;
1904 HMAP_FOR_EACH_SAFE (pdscp, next, hmap_node, &ofport->priorities) {
1905 hmap_remove(&ofport->priorities, &pdscp->hmap_node);
1911 set_queues(struct ofport *ofport_,
1912 const struct ofproto_port_queue *qdscp_list,
1915 struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
1916 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport->up.ofproto);
1917 struct hmap new = HMAP_INITIALIZER(&new);
1920 for (i = 0; i < n_qdscp; i++) {
1921 struct priority_to_dscp *pdscp;
1925 dscp = (qdscp_list[i].dscp << 2) & IP_DSCP_MASK;
1926 if (dpif_queue_to_priority(ofproto->backer->dpif, qdscp_list[i].queue,
1931 pdscp = get_priority(ofport, priority);
1933 hmap_remove(&ofport->priorities, &pdscp->hmap_node);
1935 pdscp = xmalloc(sizeof *pdscp);
1936 pdscp->priority = priority;
1938 ofproto->backer->need_revalidate = REV_RECONFIGURE;
1941 if (pdscp->dscp != dscp) {
1943 ofproto->backer->need_revalidate = REV_RECONFIGURE;
1946 hmap_insert(&new, &pdscp->hmap_node, hash_int(pdscp->priority, 0));
1949 if (!hmap_is_empty(&ofport->priorities)) {
1950 ofport_clear_priorities(ofport);
1951 ofproto->backer->need_revalidate = REV_RECONFIGURE;
1954 hmap_swap(&new, &ofport->priorities);
1962 /* Expires all MAC learning entries associated with 'bundle' and forces its
1963 * ofproto to revalidate every flow.
1965 * Normally MAC learning entries are removed only from the ofproto associated
1966 * with 'bundle', but if 'all_ofprotos' is true, then the MAC learning entries
1967 * are removed from every ofproto. When patch ports and SLB bonds are in use
1968 * and a VM migration happens and the gratuitous ARPs are somehow lost, this
1969 * avoids a MAC_ENTRY_IDLE_TIME delay before the migrated VM can communicate
1970 * with the host from which it migrated. */
1972 bundle_flush_macs(struct ofbundle *bundle, bool all_ofprotos)
1974 struct ofproto_dpif *ofproto = bundle->ofproto;
1975 struct mac_learning *ml = ofproto->ml;
1976 struct mac_entry *mac, *next_mac;
1978 ofproto->backer->need_revalidate = REV_RECONFIGURE;
1979 LIST_FOR_EACH_SAFE (mac, next_mac, lru_node, &ml->lrus) {
1980 if (mac->port.p == bundle) {
1982 struct ofproto_dpif *o;
1984 HMAP_FOR_EACH (o, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
1986 struct mac_entry *e;
1988 e = mac_learning_lookup(o->ml, mac->mac, mac->vlan,
1991 mac_learning_expire(o->ml, e);
1997 mac_learning_expire(ml, mac);
2002 static struct ofbundle *
2003 bundle_lookup(const struct ofproto_dpif *ofproto, void *aux)
2005 struct ofbundle *bundle;
2007 HMAP_FOR_EACH_IN_BUCKET (bundle, hmap_node, hash_pointer(aux, 0),
2008 &ofproto->bundles) {
2009 if (bundle->aux == aux) {
2016 /* Looks up each of the 'n_auxes' pointers in 'auxes' as bundles and adds the
2017 * ones that are found to 'bundles'. */
2019 bundle_lookup_multiple(struct ofproto_dpif *ofproto,
2020 void **auxes, size_t n_auxes,
2021 struct hmapx *bundles)
2025 hmapx_init(bundles);
2026 for (i = 0; i < n_auxes; i++) {
2027 struct ofbundle *bundle = bundle_lookup(ofproto, auxes[i]);
2029 hmapx_add(bundles, bundle);
2035 bundle_update(struct ofbundle *bundle)
2037 struct ofport_dpif *port;
2039 bundle->floodable = true;
2040 LIST_FOR_EACH (port, bundle_node, &bundle->ports) {
2041 if (port->up.pp.config & OFPUTIL_PC_NO_FLOOD
2042 || !stp_forward_in_state(port->stp_state)) {
2043 bundle->floodable = false;
2050 bundle_del_port(struct ofport_dpif *port)
2052 struct ofbundle *bundle = port->bundle;
2054 bundle->ofproto->backer->need_revalidate = REV_RECONFIGURE;
2056 list_remove(&port->bundle_node);
2057 port->bundle = NULL;
2060 lacp_slave_unregister(bundle->lacp, port);
2063 bond_slave_unregister(bundle->bond, port);
2066 bundle_update(bundle);
2070 bundle_add_port(struct ofbundle *bundle, uint16_t ofp_port,
2071 struct lacp_slave_settings *lacp)
2073 struct ofport_dpif *port;
2075 port = get_ofp_port(bundle->ofproto, ofp_port);
2080 if (port->bundle != bundle) {
2081 bundle->ofproto->backer->need_revalidate = REV_RECONFIGURE;
2083 bundle_del_port(port);
2086 port->bundle = bundle;
2087 list_push_back(&bundle->ports, &port->bundle_node);
2088 if (port->up.pp.config & OFPUTIL_PC_NO_FLOOD
2089 || !stp_forward_in_state(port->stp_state)) {
2090 bundle->floodable = false;
2094 bundle->ofproto->backer->need_revalidate = REV_RECONFIGURE;
2095 lacp_slave_register(bundle->lacp, port, lacp);
2102 bundle_destroy(struct ofbundle *bundle)
2104 struct ofproto_dpif *ofproto;
2105 struct ofport_dpif *port, *next_port;
2112 ofproto = bundle->ofproto;
2113 for (i = 0; i < MAX_MIRRORS; i++) {
2114 struct ofmirror *m = ofproto->mirrors[i];
2116 if (m->out == bundle) {
2118 } else if (hmapx_find_and_delete(&m->srcs, bundle)
2119 || hmapx_find_and_delete(&m->dsts, bundle)) {
2120 ofproto->backer->need_revalidate = REV_RECONFIGURE;
2125 LIST_FOR_EACH_SAFE (port, next_port, bundle_node, &bundle->ports) {
2126 bundle_del_port(port);
2129 bundle_flush_macs(bundle, true);
2130 hmap_remove(&ofproto->bundles, &bundle->hmap_node);
2132 free(bundle->trunks);
2133 lacp_destroy(bundle->lacp);
2134 bond_destroy(bundle->bond);
2139 bundle_set(struct ofproto *ofproto_, void *aux,
2140 const struct ofproto_bundle_settings *s)
2142 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
2143 bool need_flush = false;
2144 struct ofport_dpif *port;
2145 struct ofbundle *bundle;
2146 unsigned long *trunks;
2152 bundle_destroy(bundle_lookup(ofproto, aux));
2156 ovs_assert(s->n_slaves == 1 || s->bond != NULL);
2157 ovs_assert((s->lacp != NULL) == (s->lacp_slaves != NULL));
2159 bundle = bundle_lookup(ofproto, aux);
2161 bundle = xmalloc(sizeof *bundle);
2163 bundle->ofproto = ofproto;
2164 hmap_insert(&ofproto->bundles, &bundle->hmap_node,
2165 hash_pointer(aux, 0));
2167 bundle->name = NULL;
2169 list_init(&bundle->ports);
2170 bundle->vlan_mode = PORT_VLAN_TRUNK;
2172 bundle->trunks = NULL;
2173 bundle->use_priority_tags = s->use_priority_tags;
2174 bundle->lacp = NULL;
2175 bundle->bond = NULL;
2177 bundle->floodable = true;
2179 bundle->src_mirrors = 0;
2180 bundle->dst_mirrors = 0;
2181 bundle->mirror_out = 0;
2184 if (!bundle->name || strcmp(s->name, bundle->name)) {
2186 bundle->name = xstrdup(s->name);
2191 if (!bundle->lacp) {
2192 ofproto->backer->need_revalidate = REV_RECONFIGURE;
2193 bundle->lacp = lacp_create();
2195 lacp_configure(bundle->lacp, s->lacp);
2197 lacp_destroy(bundle->lacp);
2198 bundle->lacp = NULL;
2201 /* Update set of ports. */
2203 for (i = 0; i < s->n_slaves; i++) {
2204 if (!bundle_add_port(bundle, s->slaves[i],
2205 s->lacp ? &s->lacp_slaves[i] : NULL)) {
2209 if (!ok || list_size(&bundle->ports) != s->n_slaves) {
2210 struct ofport_dpif *next_port;
2212 LIST_FOR_EACH_SAFE (port, next_port, bundle_node, &bundle->ports) {
2213 for (i = 0; i < s->n_slaves; i++) {
2214 if (s->slaves[i] == port->up.ofp_port) {
2219 bundle_del_port(port);
2223 ovs_assert(list_size(&bundle->ports) <= s->n_slaves);
2225 if (list_is_empty(&bundle->ports)) {
2226 bundle_destroy(bundle);
2230 /* Set VLAN tagging mode */
2231 if (s->vlan_mode != bundle->vlan_mode
2232 || s->use_priority_tags != bundle->use_priority_tags) {
2233 bundle->vlan_mode = s->vlan_mode;
2234 bundle->use_priority_tags = s->use_priority_tags;
2239 vlan = (s->vlan_mode == PORT_VLAN_TRUNK ? -1
2240 : s->vlan >= 0 && s->vlan <= 4095 ? s->vlan
2242 if (vlan != bundle->vlan) {
2243 bundle->vlan = vlan;
2247 /* Get trunked VLANs. */
2248 switch (s->vlan_mode) {
2249 case PORT_VLAN_ACCESS:
2253 case PORT_VLAN_TRUNK:
2254 trunks = CONST_CAST(unsigned long *, s->trunks);
2257 case PORT_VLAN_NATIVE_UNTAGGED:
2258 case PORT_VLAN_NATIVE_TAGGED:
2259 if (vlan != 0 && (!s->trunks
2260 || !bitmap_is_set(s->trunks, vlan)
2261 || bitmap_is_set(s->trunks, 0))) {
2262 /* Force trunking the native VLAN and prohibit trunking VLAN 0. */
2264 trunks = bitmap_clone(s->trunks, 4096);
2266 trunks = bitmap_allocate1(4096);
2268 bitmap_set1(trunks, vlan);
2269 bitmap_set0(trunks, 0);
2271 trunks = CONST_CAST(unsigned long *, s->trunks);
2278 if (!vlan_bitmap_equal(trunks, bundle->trunks)) {
2279 free(bundle->trunks);
2280 if (trunks == s->trunks) {
2281 bundle->trunks = vlan_bitmap_clone(trunks);
2283 bundle->trunks = trunks;
2288 if (trunks != s->trunks) {
2293 if (!list_is_short(&bundle->ports)) {
2294 bundle->ofproto->has_bonded_bundles = true;
2296 if (bond_reconfigure(bundle->bond, s->bond)) {
2297 ofproto->backer->need_revalidate = REV_RECONFIGURE;
2300 bundle->bond = bond_create(s->bond);
2301 ofproto->backer->need_revalidate = REV_RECONFIGURE;
2304 LIST_FOR_EACH (port, bundle_node, &bundle->ports) {
2305 bond_slave_register(bundle->bond, port, port->up.netdev);
2308 bond_destroy(bundle->bond);
2309 bundle->bond = NULL;
2312 /* If we changed something that would affect MAC learning, un-learn
2313 * everything on this port and force flow revalidation. */
2315 bundle_flush_macs(bundle, false);
2322 bundle_remove(struct ofport *port_)
2324 struct ofport_dpif *port = ofport_dpif_cast(port_);
2325 struct ofbundle *bundle = port->bundle;
2328 bundle_del_port(port);
2329 if (list_is_empty(&bundle->ports)) {
2330 bundle_destroy(bundle);
2331 } else if (list_is_short(&bundle->ports)) {
2332 bond_destroy(bundle->bond);
2333 bundle->bond = NULL;
2339 send_pdu_cb(void *port_, const void *pdu, size_t pdu_size)
2341 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 10);
2342 struct ofport_dpif *port = port_;
2343 uint8_t ea[ETH_ADDR_LEN];
2346 error = netdev_get_etheraddr(port->up.netdev, ea);
2348 struct ofpbuf packet;
2351 ofpbuf_init(&packet, 0);
2352 packet_pdu = eth_compose(&packet, eth_addr_lacp, ea, ETH_TYPE_LACP,
2354 memcpy(packet_pdu, pdu, pdu_size);
2356 send_packet(port, &packet);
2357 ofpbuf_uninit(&packet);
2359 VLOG_ERR_RL(&rl, "port %s: cannot obtain Ethernet address of iface "
2360 "%s (%s)", port->bundle->name,
2361 netdev_get_name(port->up.netdev), strerror(error));
2366 bundle_send_learning_packets(struct ofbundle *bundle)
2368 struct ofproto_dpif *ofproto = bundle->ofproto;
2369 int error, n_packets, n_errors;
2370 struct mac_entry *e;
2372 error = n_packets = n_errors = 0;
2373 LIST_FOR_EACH (e, lru_node, &ofproto->ml->lrus) {
2374 if (e->port.p != bundle) {
2375 struct ofpbuf *learning_packet;
2376 struct ofport_dpif *port;
2380 /* The assignment to "port" is unnecessary but makes "grep"ing for
2381 * struct ofport_dpif more effective. */
2382 learning_packet = bond_compose_learning_packet(bundle->bond,
2386 ret = send_packet(port, learning_packet);
2387 ofpbuf_delete(learning_packet);
2397 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
2398 VLOG_WARN_RL(&rl, "bond %s: %d errors sending %d gratuitous learning "
2399 "packets, last error was: %s",
2400 bundle->name, n_errors, n_packets, strerror(error));
2402 VLOG_DBG("bond %s: sent %d gratuitous learning packets",
2403 bundle->name, n_packets);
2408 bundle_run(struct ofbundle *bundle)
2411 lacp_run(bundle->lacp, send_pdu_cb);
2414 struct ofport_dpif *port;
2416 LIST_FOR_EACH (port, bundle_node, &bundle->ports) {
2417 bond_slave_set_may_enable(bundle->bond, port, port->may_enable);
2420 bond_run(bundle->bond, &bundle->ofproto->backer->revalidate_set,
2421 lacp_status(bundle->lacp));
2422 if (bond_should_send_learning_packets(bundle->bond)) {
2423 bundle_send_learning_packets(bundle);
2429 bundle_wait(struct ofbundle *bundle)
2432 lacp_wait(bundle->lacp);
2435 bond_wait(bundle->bond);
2442 mirror_scan(struct ofproto_dpif *ofproto)
2446 for (idx = 0; idx < MAX_MIRRORS; idx++) {
2447 if (!ofproto->mirrors[idx]) {
2454 static struct ofmirror *
2455 mirror_lookup(struct ofproto_dpif *ofproto, void *aux)
2459 for (i = 0; i < MAX_MIRRORS; i++) {
2460 struct ofmirror *mirror = ofproto->mirrors[i];
2461 if (mirror && mirror->aux == aux) {
2469 /* Update the 'dup_mirrors' member of each of the ofmirrors in 'ofproto'. */
2471 mirror_update_dups(struct ofproto_dpif *ofproto)
2475 for (i = 0; i < MAX_MIRRORS; i++) {
2476 struct ofmirror *m = ofproto->mirrors[i];
2479 m->dup_mirrors = MIRROR_MASK_C(1) << i;
2483 for (i = 0; i < MAX_MIRRORS; i++) {
2484 struct ofmirror *m1 = ofproto->mirrors[i];
2491 for (j = i + 1; j < MAX_MIRRORS; j++) {
2492 struct ofmirror *m2 = ofproto->mirrors[j];
2494 if (m2 && m1->out == m2->out && m1->out_vlan == m2->out_vlan) {
2495 m1->dup_mirrors |= MIRROR_MASK_C(1) << j;
2496 m2->dup_mirrors |= m1->dup_mirrors;
2503 mirror_set(struct ofproto *ofproto_, void *aux,
2504 const struct ofproto_mirror_settings *s)
2506 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
2507 mirror_mask_t mirror_bit;
2508 struct ofbundle *bundle;
2509 struct ofmirror *mirror;
2510 struct ofbundle *out;
2511 struct hmapx srcs; /* Contains "struct ofbundle *"s. */
2512 struct hmapx dsts; /* Contains "struct ofbundle *"s. */
2515 mirror = mirror_lookup(ofproto, aux);
2517 mirror_destroy(mirror);
2523 idx = mirror_scan(ofproto);
2525 VLOG_WARN("bridge %s: maximum of %d port mirrors reached, "
2527 ofproto->up.name, MAX_MIRRORS, s->name);
2531 mirror = ofproto->mirrors[idx] = xzalloc(sizeof *mirror);
2532 mirror->ofproto = ofproto;
2535 mirror->out_vlan = -1;
2536 mirror->name = NULL;
2539 if (!mirror->name || strcmp(s->name, mirror->name)) {
2541 mirror->name = xstrdup(s->name);
2544 /* Get the new configuration. */
2545 if (s->out_bundle) {
2546 out = bundle_lookup(ofproto, s->out_bundle);
2548 mirror_destroy(mirror);
2554 out_vlan = s->out_vlan;
2556 bundle_lookup_multiple(ofproto, s->srcs, s->n_srcs, &srcs);
2557 bundle_lookup_multiple(ofproto, s->dsts, s->n_dsts, &dsts);
2559 /* If the configuration has not changed, do nothing. */
2560 if (hmapx_equals(&srcs, &mirror->srcs)
2561 && hmapx_equals(&dsts, &mirror->dsts)
2562 && vlan_bitmap_equal(mirror->vlans, s->src_vlans)
2563 && mirror->out == out
2564 && mirror->out_vlan == out_vlan)
2566 hmapx_destroy(&srcs);
2567 hmapx_destroy(&dsts);
2571 hmapx_swap(&srcs, &mirror->srcs);
2572 hmapx_destroy(&srcs);
2574 hmapx_swap(&dsts, &mirror->dsts);
2575 hmapx_destroy(&dsts);
2577 free(mirror->vlans);
2578 mirror->vlans = vlan_bitmap_clone(s->src_vlans);
2581 mirror->out_vlan = out_vlan;
2583 /* Update bundles. */
2584 mirror_bit = MIRROR_MASK_C(1) << mirror->idx;
2585 HMAP_FOR_EACH (bundle, hmap_node, &mirror->ofproto->bundles) {
2586 if (hmapx_contains(&mirror->srcs, bundle)) {
2587 bundle->src_mirrors |= mirror_bit;
2589 bundle->src_mirrors &= ~mirror_bit;
2592 if (hmapx_contains(&mirror->dsts, bundle)) {
2593 bundle->dst_mirrors |= mirror_bit;
2595 bundle->dst_mirrors &= ~mirror_bit;
2598 if (mirror->out == bundle) {
2599 bundle->mirror_out |= mirror_bit;
2601 bundle->mirror_out &= ~mirror_bit;
2605 ofproto->backer->need_revalidate = REV_RECONFIGURE;
2606 ofproto->has_mirrors = true;
2607 mac_learning_flush(ofproto->ml,
2608 &ofproto->backer->revalidate_set);
2609 mirror_update_dups(ofproto);
2615 mirror_destroy(struct ofmirror *mirror)
2617 struct ofproto_dpif *ofproto;
2618 mirror_mask_t mirror_bit;
2619 struct ofbundle *bundle;
2626 ofproto = mirror->ofproto;
2627 ofproto->backer->need_revalidate = REV_RECONFIGURE;
2628 mac_learning_flush(ofproto->ml, &ofproto->backer->revalidate_set);
2630 mirror_bit = MIRROR_MASK_C(1) << mirror->idx;
2631 HMAP_FOR_EACH (bundle, hmap_node, &ofproto->bundles) {
2632 bundle->src_mirrors &= ~mirror_bit;
2633 bundle->dst_mirrors &= ~mirror_bit;
2634 bundle->mirror_out &= ~mirror_bit;
2637 hmapx_destroy(&mirror->srcs);
2638 hmapx_destroy(&mirror->dsts);
2639 free(mirror->vlans);
2641 ofproto->mirrors[mirror->idx] = NULL;
2645 mirror_update_dups(ofproto);
2647 ofproto->has_mirrors = false;
2648 for (i = 0; i < MAX_MIRRORS; i++) {
2649 if (ofproto->mirrors[i]) {
2650 ofproto->has_mirrors = true;
2657 mirror_get_stats(struct ofproto *ofproto_, void *aux,
2658 uint64_t *packets, uint64_t *bytes)
2660 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
2661 struct ofmirror *mirror = mirror_lookup(ofproto, aux);
2664 *packets = *bytes = UINT64_MAX;
2670 *packets = mirror->packet_count;
2671 *bytes = mirror->byte_count;
2677 set_flood_vlans(struct ofproto *ofproto_, unsigned long *flood_vlans)
2679 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
2680 if (mac_learning_set_flood_vlans(ofproto->ml, flood_vlans)) {
2681 mac_learning_flush(ofproto->ml, &ofproto->backer->revalidate_set);
2687 is_mirror_output_bundle(const struct ofproto *ofproto_, void *aux)
2689 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
2690 struct ofbundle *bundle = bundle_lookup(ofproto, aux);
2691 return bundle && bundle->mirror_out != 0;
2695 forward_bpdu_changed(struct ofproto *ofproto_)
2697 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
2698 ofproto->backer->need_revalidate = REV_RECONFIGURE;
2702 set_mac_table_config(struct ofproto *ofproto_, unsigned int idle_time,
2705 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
2706 mac_learning_set_idle_time(ofproto->ml, idle_time);
2707 mac_learning_set_max_entries(ofproto->ml, max_entries);
2712 struct ofport_dpif *
2713 get_ofp_port(const struct ofproto_dpif *ofproto, uint16_t ofp_port)
2715 struct ofport *ofport = ofproto_get_port(&ofproto->up, ofp_port);
2716 return ofport ? ofport_dpif_cast(ofport) : NULL;
2719 struct ofport_dpif *
2720 get_odp_port(const struct ofproto_dpif *ofproto, uint32_t odp_port)
2722 struct ofport_dpif *port = odp_port_to_ofport(ofproto->backer, odp_port);
2723 return port && &ofproto->up == port->up.ofproto ? port : NULL;
2727 ofproto_port_from_dpif_port(struct ofproto_dpif *ofproto,
2728 struct ofproto_port *ofproto_port,
2729 struct dpif_port *dpif_port)
2731 ofproto_port->name = dpif_port->name;
2732 ofproto_port->type = dpif_port->type;
2733 ofproto_port->ofp_port = odp_port_to_ofp_port(ofproto, dpif_port->port_no);
2736 struct ofport_dpif *
2737 ofport_get_peer(const struct ofport_dpif *ofport_dpif)
2739 const struct ofproto_dpif *ofproto;
2740 const struct dpif_backer *backer;
2743 peer = netdev_vport_patch_peer(ofport_dpif->up.netdev);
2748 backer = ofproto_dpif_cast(ofport_dpif->up.ofproto)->backer;
2749 HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
2750 struct ofport *ofport;
2752 if (ofproto->backer != backer) {
2756 ofport = shash_find_data(&ofproto->up.port_by_name, peer);
2758 return ofport_dpif_cast(ofport);
2765 port_run_fast(struct ofport_dpif *ofport)
2767 if (ofport->cfm && cfm_should_send_ccm(ofport->cfm)) {
2768 struct ofpbuf packet;
2770 ofpbuf_init(&packet, 0);
2771 cfm_compose_ccm(ofport->cfm, &packet, ofport->up.pp.hw_addr);
2772 send_packet(ofport, &packet);
2773 ofpbuf_uninit(&packet);
2776 if (ofport->bfd && bfd_should_send_packet(ofport->bfd)) {
2777 struct ofpbuf packet;
2779 ofpbuf_init(&packet, 0);
2780 bfd_put_packet(ofport->bfd, &packet, ofport->up.pp.hw_addr);
2781 send_packet(ofport, &packet);
2782 ofpbuf_uninit(&packet);
2787 port_run(struct ofport_dpif *ofport)
2789 long long int carrier_seq = netdev_get_carrier_resets(ofport->up.netdev);
2790 bool carrier_changed = carrier_seq != ofport->carrier_seq;
2791 bool enable = netdev_get_carrier(ofport->up.netdev);
2793 ofport->carrier_seq = carrier_seq;
2795 port_run_fast(ofport);
2797 if (ofport->tnl_port
2798 && tnl_port_reconfigure(&ofport->up, ofport->odp_port,
2799 &ofport->tnl_port)) {
2800 ofproto_dpif_cast(ofport->up.ofproto)->backer->need_revalidate = true;
2804 int cfm_opup = cfm_get_opup(ofport->cfm);
2806 cfm_run(ofport->cfm);
2807 enable = enable && !cfm_get_fault(ofport->cfm);
2809 if (cfm_opup >= 0) {
2810 enable = enable && cfm_opup;
2815 bfd_run(ofport->bfd);
2816 enable = enable && bfd_forwarding(ofport->bfd);
2819 if (ofport->bundle) {
2820 enable = enable && lacp_slave_may_enable(ofport->bundle->lacp, ofport);
2821 if (carrier_changed) {
2822 lacp_slave_carrier_changed(ofport->bundle->lacp, ofport);
2826 if (ofport->may_enable != enable) {
2827 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport->up.ofproto);
2829 if (ofproto->has_bundle_action) {
2830 ofproto->backer->need_revalidate = REV_PORT_TOGGLED;
2834 ofport->may_enable = enable;
2838 port_wait(struct ofport_dpif *ofport)
2841 cfm_wait(ofport->cfm);
2845 bfd_wait(ofport->bfd);
2850 port_query_by_name(const struct ofproto *ofproto_, const char *devname,
2851 struct ofproto_port *ofproto_port)
2853 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
2854 struct dpif_port dpif_port;
2857 if (sset_contains(&ofproto->ghost_ports, devname)) {
2858 const char *type = netdev_get_type_from_name(devname);
2860 /* We may be called before ofproto->up.port_by_name is populated with
2861 * the appropriate ofport. For this reason, we must get the name and
2862 * type from the netdev layer directly. */
2864 const struct ofport *ofport;
2866 ofport = shash_find_data(&ofproto->up.port_by_name, devname);
2867 ofproto_port->ofp_port = ofport ? ofport->ofp_port : OFPP_NONE;
2868 ofproto_port->name = xstrdup(devname);
2869 ofproto_port->type = xstrdup(type);
2875 if (!sset_contains(&ofproto->ports, devname)) {
2878 error = dpif_port_query_by_name(ofproto->backer->dpif,
2879 devname, &dpif_port);
2881 ofproto_port_from_dpif_port(ofproto, ofproto_port, &dpif_port);
2887 port_add(struct ofproto *ofproto_, struct netdev *netdev)
2889 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
2890 const char *devname = netdev_get_name(netdev);
2891 char namebuf[NETDEV_VPORT_NAME_BUFSIZE];
2892 const char *dp_port_name;
2894 if (netdev_vport_is_patch(netdev)) {
2895 sset_add(&ofproto->ghost_ports, netdev_get_name(netdev));
2899 dp_port_name = netdev_vport_get_dpif_port(netdev, namebuf, sizeof namebuf);
2900 if (!dpif_port_exists(ofproto->backer->dpif, dp_port_name)) {
2901 uint32_t port_no = UINT32_MAX;
2904 error = dpif_port_add(ofproto->backer->dpif, netdev, &port_no);
2908 if (netdev_get_tunnel_config(netdev)) {
2909 simap_put(&ofproto->backer->tnl_backers, dp_port_name, port_no);
2913 if (netdev_get_tunnel_config(netdev)) {
2914 sset_add(&ofproto->ghost_ports, devname);
2916 sset_add(&ofproto->ports, devname);
2922 port_del(struct ofproto *ofproto_, uint16_t ofp_port)
2924 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
2925 struct ofport_dpif *ofport = get_ofp_port(ofproto, ofp_port);
2932 sset_find_and_delete(&ofproto->ghost_ports,
2933 netdev_get_name(ofport->up.netdev));
2934 ofproto->backer->need_revalidate = REV_RECONFIGURE;
2935 if (!ofport->tnl_port) {
2936 error = dpif_port_del(ofproto->backer->dpif, ofport->odp_port);
2938 /* The caller is going to close ofport->up.netdev. If this is a
2939 * bonded port, then the bond is using that netdev, so remove it
2940 * from the bond. The client will need to reconfigure everything
2941 * after deleting ports, so then the slave will get re-added. */
2942 bundle_remove(&ofport->up);
2949 port_get_stats(const struct ofport *ofport_, struct netdev_stats *stats)
2951 struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
2956 error = netdev_get_stats(ofport->up.netdev, stats);
2958 if (!error && ofport_->ofp_port == OFPP_LOCAL) {
2959 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport->up.ofproto);
2961 /* ofproto->stats.tx_packets represents packets that we created
2962 * internally and sent to some port (e.g. packets sent with
2963 * send_packet()). Account for them as if they had come from
2964 * OFPP_LOCAL and got forwarded. */
2966 if (stats->rx_packets != UINT64_MAX) {
2967 stats->rx_packets += ofproto->stats.tx_packets;
2970 if (stats->rx_bytes != UINT64_MAX) {
2971 stats->rx_bytes += ofproto->stats.tx_bytes;
2974 /* ofproto->stats.rx_packets represents packets that were received on
2975 * some port and we processed internally and dropped (e.g. STP).
2976 * Account for them as if they had been forwarded to OFPP_LOCAL. */
2978 if (stats->tx_packets != UINT64_MAX) {
2979 stats->tx_packets += ofproto->stats.rx_packets;
2982 if (stats->tx_bytes != UINT64_MAX) {
2983 stats->tx_bytes += ofproto->stats.rx_bytes;
2990 struct port_dump_state {
2995 struct ofproto_port port;
3000 port_dump_start(const struct ofproto *ofproto_ OVS_UNUSED, void **statep)
3002 *statep = xzalloc(sizeof(struct port_dump_state));
3007 port_dump_next(const struct ofproto *ofproto_, void *state_,
3008 struct ofproto_port *port)
3010 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
3011 struct port_dump_state *state = state_;
3012 const struct sset *sset;
3013 struct sset_node *node;
3015 if (state->has_port) {
3016 ofproto_port_destroy(&state->port);
3017 state->has_port = false;
3019 sset = state->ghost ? &ofproto->ghost_ports : &ofproto->ports;
3020 while ((node = sset_at_position(sset, &state->bucket, &state->offset))) {
3023 error = port_query_by_name(ofproto_, node->name, &state->port);
3025 *port = state->port;
3026 state->has_port = true;
3028 } else if (error != ENODEV) {
3033 if (!state->ghost) {
3034 state->ghost = true;
3037 return port_dump_next(ofproto_, state_, port);
3044 port_dump_done(const struct ofproto *ofproto_ OVS_UNUSED, void *state_)
3046 struct port_dump_state *state = state_;
3048 if (state->has_port) {
3049 ofproto_port_destroy(&state->port);
3056 port_poll(const struct ofproto *ofproto_, char **devnamep)
3058 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
3060 if (ofproto->port_poll_errno) {
3061 int error = ofproto->port_poll_errno;
3062 ofproto->port_poll_errno = 0;
3066 if (sset_is_empty(&ofproto->port_poll_set)) {
3070 *devnamep = sset_pop(&ofproto->port_poll_set);
3075 port_poll_wait(const struct ofproto *ofproto_)
3077 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
3078 dpif_port_poll_wait(ofproto->backer->dpif);
3082 port_is_lacp_current(const struct ofport *ofport_)
3084 const struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
3085 return (ofport->bundle && ofport->bundle->lacp
3086 ? lacp_slave_is_current(ofport->bundle->lacp, ofport)
3090 /* Upcall handling. */
3092 /* Flow miss batching.
3094 * Some dpifs implement operations faster when you hand them off in a batch.
3095 * To allow batching, "struct flow_miss" queues the dpif-related work needed
3096 * for a given flow. Each "struct flow_miss" corresponds to sending one or
3097 * more packets, plus possibly installing the flow in the dpif.
3099 * So far we only batch the operations that affect flow setup time the most.
3100 * It's possible to batch more than that, but the benefit might be minimal. */
3102 struct hmap_node hmap_node;
3103 struct ofproto_dpif *ofproto;
3105 enum odp_key_fitness key_fitness;
3106 const struct nlattr *key;
3108 struct list packets;
3109 enum dpif_upcall_type upcall_type;
3112 struct flow_miss_op {
3113 struct dpif_op dpif_op;
3115 uint64_t slow_stub[128 / 8]; /* Buffer for compose_slow_path() */
3116 struct xlate_out xout;
3117 bool xout_garbage; /* 'xout' needs to be uninitialized? */
3120 /* Sends an OFPT_PACKET_IN message for 'packet' of type OFPR_NO_MATCH to each
3121 * OpenFlow controller as necessary according to their individual
3122 * configurations. */
3124 send_packet_in_miss(struct ofproto_dpif *ofproto, const struct ofpbuf *packet,
3125 const struct flow *flow)
3127 struct ofputil_packet_in pin;
3129 pin.packet = packet->data;
3130 pin.packet_len = packet->size;
3131 pin.reason = OFPR_NO_MATCH;
3132 pin.controller_id = 0;
3137 pin.send_len = 0; /* not used for flow table misses */
3139 flow_get_metadata(flow, &pin.fmd);
3141 connmgr_send_packet_in(ofproto->up.connmgr, &pin);
3144 enum slow_path_reason
3145 process_special(struct ofproto_dpif *ofproto, const struct flow *flow,
3146 const struct ofport_dpif *ofport, const struct ofpbuf *packet)
3150 } else if (ofport->cfm && cfm_should_process_flow(ofport->cfm, flow)) {
3152 cfm_process_heartbeat(ofport->cfm, packet);
3155 } else if (ofport->bfd && bfd_should_process_flow(flow)) {
3157 bfd_process_packet(ofport->bfd, flow, packet);
3160 } else if (ofport->bundle && ofport->bundle->lacp
3161 && flow->dl_type == htons(ETH_TYPE_LACP)) {
3163 lacp_process_packet(ofport->bundle->lacp, ofport, packet);
3166 } else if (ofproto->stp && stp_should_process_flow(flow)) {
3168 stp_process_packet(ofport, packet);
3176 static struct flow_miss *
3177 flow_miss_find(struct hmap *todo, const struct ofproto_dpif *ofproto,
3178 const struct flow *flow, uint32_t hash)
3180 struct flow_miss *miss;
3182 HMAP_FOR_EACH_WITH_HASH (miss, hmap_node, hash, todo) {
3183 if (miss->ofproto == ofproto && flow_equal(&miss->flow, flow)) {
3191 /* Partially Initializes 'op' as an "execute" operation for 'miss' and
3192 * 'packet'. The caller must initialize op->actions and op->actions_len. If
3193 * 'miss' is associated with a subfacet the caller must also initialize the
3194 * returned op->subfacet, and if anything needs to be freed after processing
3195 * the op, the caller must initialize op->garbage also. */
3197 init_flow_miss_execute_op(struct flow_miss *miss, struct ofpbuf *packet,
3198 struct flow_miss_op *op)
3200 if (miss->flow.in_port
3201 != vsp_realdev_to_vlandev(miss->ofproto, miss->flow.in_port,
3202 miss->flow.vlan_tci)) {
3203 /* This packet was received on a VLAN splinter port. We
3204 * added a VLAN to the packet to make the packet resemble
3205 * the flow, but the actions were composed assuming that
3206 * the packet contained no VLAN. So, we must remove the
3207 * VLAN header from the packet before trying to execute the
3209 eth_pop_vlan(packet);
3212 op->xout_garbage = false;
3213 op->dpif_op.type = DPIF_OP_EXECUTE;
3214 op->dpif_op.u.execute.key = miss->key;
3215 op->dpif_op.u.execute.key_len = miss->key_len;
3216 op->dpif_op.u.execute.packet = packet;
3219 /* Helper for handle_flow_miss_without_facet() and
3220 * handle_flow_miss_with_facet(). */
3222 handle_flow_miss_common(struct rule_dpif *rule,
3223 struct ofpbuf *packet, const struct flow *flow)
3225 struct ofproto_dpif *ofproto = ofproto_dpif_cast(rule->up.ofproto);
3227 if (rule->up.cr.priority == FAIL_OPEN_PRIORITY) {
3229 * Extra-special case for fail-open mode.
3231 * We are in fail-open mode and the packet matched the fail-open
3232 * rule, but we are connected to a controller too. We should send
3233 * the packet up to the controller in the hope that it will try to
3234 * set up a flow and thereby allow us to exit fail-open.
3236 * See the top-level comment in fail-open.c for more information.
3238 send_packet_in_miss(ofproto, packet, flow);
3242 /* Figures out whether a flow that missed in 'ofproto', whose details are in
3243 * 'miss' masked by 'wc', is likely to be worth tracking in detail in userspace
3244 * and (usually) installing a datapath flow. The answer is usually "yes" (a
3245 * return value of true). However, for short flows the cost of bookkeeping is
3246 * much higher than the benefits, so when the datapath holds a large number of
3247 * flows we impose some heuristics to decide which flows are likely to be worth
3250 flow_miss_should_make_facet(struct flow_miss *miss, struct flow_wildcards *wc)
3252 struct dpif_backer *backer = miss->ofproto->backer;
3255 if (!backer->governor) {
3258 n_subfacets = hmap_count(&backer->subfacets);
3259 if (n_subfacets * 2 <= flow_eviction_threshold) {
3263 backer->governor = governor_create();
3266 hash = flow_hash_in_wildcards(&miss->flow, wc, 0);
3267 return governor_should_install_flow(backer->governor, hash,
3268 list_size(&miss->packets));
3271 /* Handles 'miss' without creating a facet or subfacet or creating any datapath
3272 * flow. 'miss->flow' must have matched 'rule' and been xlated into 'xout'.
3273 * May add an "execute" operation to 'ops' and increment '*n_ops'. */
3275 handle_flow_miss_without_facet(struct rule_dpif *rule, struct xlate_out *xout,
3276 struct flow_miss *miss,
3277 struct flow_miss_op *ops, size_t *n_ops)
3279 struct ofpbuf *packet;
3281 LIST_FOR_EACH (packet, list_node, &miss->packets) {
3283 COVERAGE_INC(facet_suppress);
3285 handle_flow_miss_common(rule, packet, &miss->flow);
3288 struct xlate_in xin;
3290 xlate_in_init(&xin, miss->ofproto, &miss->flow, rule, 0, packet);
3291 xlate_actions_for_side_effects(&xin);
3294 if (xout->odp_actions.size) {
3295 struct flow_miss_op *op = &ops[*n_ops];
3296 struct dpif_execute *execute = &op->dpif_op.u.execute;
3298 init_flow_miss_execute_op(miss, packet, op);
3299 xlate_out_copy(&op->xout, xout);
3300 execute->actions = op->xout.odp_actions.data;
3301 execute->actions_len = op->xout.odp_actions.size;
3302 op->xout_garbage = true;
3309 /* Handles 'miss', which matches 'facet'. May add any required datapath
3310 * operations to 'ops', incrementing '*n_ops' for each new op.
3312 * All of the packets in 'miss' are considered to have arrived at time 'now'.
3313 * This is really important only for new facets: if we just called time_msec()
3314 * here, then the new subfacet or its packets could look (occasionally) as
3315 * though it was used some time after the facet was used. That can make a
3316 * one-packet flow look like it has a nonzero duration, which looks odd in
3317 * e.g. NetFlow statistics.
3319 * If non-null, 'stats' will be folded into 'facet'. */
3321 handle_flow_miss_with_facet(struct flow_miss *miss, struct facet *facet,
3322 long long int now, struct dpif_flow_stats *stats,
3323 struct flow_miss_op *ops, size_t *n_ops)
3325 struct ofproto_dpif *ofproto = ofproto_dpif_cast(facet->rule->up.ofproto);
3326 enum subfacet_path want_path;
3327 struct subfacet *subfacet;
3328 struct ofpbuf *packet;
3330 subfacet = subfacet_create(facet, miss, now);
3331 want_path = facet->xout.slow ? SF_SLOW_PATH : SF_FAST_PATH;
3333 subfacet_update_stats(subfacet, stats);
3336 LIST_FOR_EACH (packet, list_node, &miss->packets) {
3337 struct flow_miss_op *op = &ops[*n_ops];
3339 handle_flow_miss_common(facet->rule, packet, &miss->flow);
3341 if (want_path != SF_FAST_PATH) {
3342 struct xlate_in xin;
3344 xlate_in_init(&xin, ofproto, &miss->flow, facet->rule, 0, packet);
3345 xlate_actions_for_side_effects(&xin);
3348 if (facet->xout.odp_actions.size) {
3349 struct dpif_execute *execute = &op->dpif_op.u.execute;
3351 init_flow_miss_execute_op(miss, packet, op);
3352 execute->actions = facet->xout.odp_actions.data,
3353 execute->actions_len = facet->xout.odp_actions.size;
3358 if (miss->upcall_type == DPIF_UC_MISS || subfacet->path != want_path) {
3359 struct flow_miss_op *op = &ops[(*n_ops)++];
3360 struct dpif_flow_put *put = &op->dpif_op.u.flow_put;
3362 subfacet->path = want_path;
3364 op->xout_garbage = false;
3365 op->dpif_op.type = DPIF_OP_FLOW_PUT;
3366 put->flags = DPIF_FP_CREATE | DPIF_FP_MODIFY;
3367 put->key = miss->key;
3368 put->key_len = miss->key_len;
3369 if (want_path == SF_FAST_PATH) {
3370 put->actions = facet->xout.odp_actions.data;
3371 put->actions_len = facet->xout.odp_actions.size;
3373 compose_slow_path(ofproto, &miss->flow, facet->xout.slow,
3374 op->slow_stub, sizeof op->slow_stub,
3375 &put->actions, &put->actions_len);
3381 /* Handles flow miss 'miss'. May add any required datapath operations
3382 * to 'ops', incrementing '*n_ops' for each new op. */
3384 handle_flow_miss(struct flow_miss *miss, struct flow_miss_op *ops,
3387 struct ofproto_dpif *ofproto = miss->ofproto;
3388 struct dpif_flow_stats stats__;
3389 struct dpif_flow_stats *stats = &stats__;
3390 struct ofpbuf *packet;
3391 struct facet *facet;
3395 memset(stats, 0, sizeof *stats);
3397 LIST_FOR_EACH (packet, list_node, &miss->packets) {
3398 stats->tcp_flags |= packet_get_tcp_flags(packet, &miss->flow);
3399 stats->n_bytes += packet->size;
3403 facet = facet_lookup_valid(ofproto, &miss->flow);
3405 struct flow_wildcards wc;
3406 struct rule_dpif *rule;
3407 struct xlate_out xout;
3408 struct xlate_in xin;
3410 flow_wildcards_init_catchall(&wc);
3411 rule = rule_dpif_lookup(ofproto, &miss->flow, &wc);
3412 rule_credit_stats(rule, stats);
3414 xlate_in_init(&xin, ofproto, &miss->flow, rule, stats->tcp_flags,
3416 xin.resubmit_stats = stats;
3417 xin.may_learn = true;
3418 xlate_actions(&xin, &xout);
3419 flow_wildcards_or(&xout.wc, &xout.wc, &wc);
3421 /* There does not exist a bijection between 'struct flow' and datapath
3422 * flow keys with fitness ODP_FIT_TO_LITTLE. This breaks a fundamental
3423 * assumption used throughout the facet and subfacet handling code.
3424 * Since we have to handle these misses in userspace anyway, we simply
3425 * skip facet creation, avoiding the problem altogether. */
3426 if (miss->key_fitness == ODP_FIT_TOO_LITTLE
3427 || !flow_miss_should_make_facet(miss, &xout.wc)) {
3428 handle_flow_miss_without_facet(rule, &xout, miss, ops, n_ops);
3432 facet = facet_create(miss, rule, &xout, stats);
3435 handle_flow_miss_with_facet(miss, facet, now, stats, ops, n_ops);
3438 static struct drop_key *
3439 drop_key_lookup(const struct dpif_backer *backer, const struct nlattr *key,
3442 struct drop_key *drop_key;
3444 HMAP_FOR_EACH_WITH_HASH (drop_key, hmap_node, hash_bytes(key, key_len, 0),
3445 &backer->drop_keys) {
3446 if (drop_key->key_len == key_len
3447 && !memcmp(drop_key->key, key, key_len)) {
3455 drop_key_clear(struct dpif_backer *backer)
3457 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 15);
3458 struct drop_key *drop_key, *next;
3460 HMAP_FOR_EACH_SAFE (drop_key, next, hmap_node, &backer->drop_keys) {
3463 error = dpif_flow_del(backer->dpif, drop_key->key, drop_key->key_len,
3465 if (error && !VLOG_DROP_WARN(&rl)) {
3466 struct ds ds = DS_EMPTY_INITIALIZER;
3467 odp_flow_key_format(drop_key->key, drop_key->key_len, &ds);
3468 VLOG_WARN("Failed to delete drop key (%s) (%s)", strerror(error),
3473 hmap_remove(&backer->drop_keys, &drop_key->hmap_node);
3474 free(drop_key->key);
3479 /* Given a datpath, packet, and flow metadata ('backer', 'packet', and 'key'
3480 * respectively), populates 'flow' with the result of odp_flow_key_to_flow().
3481 * Optionally, if nonnull, populates 'fitnessp' with the fitness of 'flow' as
3482 * returned by odp_flow_key_to_flow(). Also, optionally populates 'ofproto'
3483 * with the ofproto_dpif, and 'odp_in_port' with the datapath in_port, that
3484 * 'packet' ingressed.
3486 * If 'ofproto' is nonnull, requires 'flow''s in_port to exist. Otherwise sets
3487 * 'flow''s in_port to OFPP_NONE.
3489 * This function does post-processing on data returned from
3490 * odp_flow_key_to_flow() to help make VLAN splinters transparent to the rest
3491 * of the upcall processing logic. In particular, if the extracted in_port is
3492 * a VLAN splinter port, it replaces flow->in_port by the "real" port, sets
3493 * flow->vlan_tci correctly for the VLAN of the VLAN splinter port, and pushes
3494 * a VLAN header onto 'packet' (if it is nonnull).
3496 * Similarly, this function also includes some logic to help with tunnels. It
3497 * may modify 'flow' as necessary to make the tunneling implementation
3498 * transparent to the upcall processing logic.
3500 * Returns 0 if successful, ENODEV if the parsed flow has no associated ofport,
3501 * or some other positive errno if there are other problems. */
3503 ofproto_receive(const struct dpif_backer *backer, struct ofpbuf *packet,
3504 const struct nlattr *key, size_t key_len,
3505 struct flow *flow, enum odp_key_fitness *fitnessp,
3506 struct ofproto_dpif **ofproto, uint32_t *odp_in_port)
3508 const struct ofport_dpif *port;
3509 enum odp_key_fitness fitness;
3512 fitness = odp_flow_key_to_flow(key, key_len, flow);
3513 if (fitness == ODP_FIT_ERROR) {
3519 *odp_in_port = flow->in_port;
3522 port = (tnl_port_should_receive(flow)
3523 ? ofport_dpif_cast(tnl_port_receive(flow))
3524 : odp_port_to_ofport(backer, flow->in_port));
3525 flow->in_port = port ? port->up.ofp_port : OFPP_NONE;
3530 /* XXX: Since the tunnel module is not scoped per backer, for a tunnel port
3531 * it's theoretically possible that we'll receive an ofport belonging to an
3532 * entirely different datapath. In practice, this can't happen because no
3533 * platforms has two separate datapaths which each support tunneling. */
3534 ovs_assert(ofproto_dpif_cast(port->up.ofproto)->backer == backer);
3536 if (vsp_adjust_flow(ofproto_dpif_cast(port->up.ofproto), flow)) {
3538 /* Make the packet resemble the flow, so that it gets sent to
3539 * an OpenFlow controller properly, so that it looks correct
3540 * for sFlow, and so that flow_extract() will get the correct
3541 * vlan_tci if it is called on 'packet'.
3543 * The allocated space inside 'packet' probably also contains
3544 * 'key', that is, both 'packet' and 'key' are probably part of
3545 * a struct dpif_upcall (see the large comment on that
3546 * structure definition), so pushing data on 'packet' is in
3547 * general not a good idea since it could overwrite 'key' or
3548 * free it as a side effect. However, it's OK in this special
3549 * case because we know that 'packet' is inside a Netlink
3550 * attribute: pushing 4 bytes will just overwrite the 4-byte
3551 * "struct nlattr", which is fine since we don't need that
3552 * header anymore. */
3553 eth_push_vlan(packet, flow->vlan_tci);
3555 /* We can't reproduce 'key' from 'flow'. */
3556 fitness = fitness == ODP_FIT_PERFECT ? ODP_FIT_TOO_MUCH : fitness;
3561 *ofproto = ofproto_dpif_cast(port->up.ofproto);
3566 *fitnessp = fitness;
3572 handle_miss_upcalls(struct dpif_backer *backer, struct dpif_upcall *upcalls,
3575 struct dpif_upcall *upcall;
3576 struct flow_miss *miss;
3577 struct flow_miss misses[FLOW_MISS_MAX_BATCH];
3578 struct flow_miss_op flow_miss_ops[FLOW_MISS_MAX_BATCH * 2];
3579 struct dpif_op *dpif_ops[FLOW_MISS_MAX_BATCH * 2];
3589 /* Construct the to-do list.
3591 * This just amounts to extracting the flow from each packet and sticking
3592 * the packets that have the same flow in the same "flow_miss" structure so
3593 * that we can process them together. */
3596 for (upcall = upcalls; upcall < &upcalls[n_upcalls]; upcall++) {
3597 struct flow_miss *miss = &misses[n_misses];
3598 struct flow_miss *existing_miss;
3599 struct ofproto_dpif *ofproto;
3600 uint32_t odp_in_port;
3605 error = ofproto_receive(backer, upcall->packet, upcall->key,
3606 upcall->key_len, &flow, &miss->key_fitness,
3607 &ofproto, &odp_in_port);
3608 if (error == ENODEV) {
3609 struct drop_key *drop_key;
3611 /* Received packet on datapath port for which we couldn't
3612 * associate an ofproto. This can happen if a port is removed
3613 * while traffic is being received. Print a rate-limited message
3614 * in case it happens frequently. Install a drop flow so
3615 * that future packets of the flow are inexpensively dropped
3617 VLOG_INFO_RL(&rl, "received packet on unassociated datapath port "
3618 "%"PRIu32, odp_in_port);
3620 drop_key = drop_key_lookup(backer, upcall->key, upcall->key_len);
3622 drop_key = xmalloc(sizeof *drop_key);
3623 drop_key->key = xmemdup(upcall->key, upcall->key_len);
3624 drop_key->key_len = upcall->key_len;
3626 hmap_insert(&backer->drop_keys, &drop_key->hmap_node,
3627 hash_bytes(drop_key->key, drop_key->key_len, 0));
3628 dpif_flow_put(backer->dpif, DPIF_FP_CREATE | DPIF_FP_MODIFY,
3629 drop_key->key, drop_key->key_len, NULL, 0, NULL);
3637 ofproto->n_missed++;
3638 flow_extract(upcall->packet, flow.skb_priority, flow.skb_mark,
3639 &flow.tunnel, flow.in_port, &miss->flow);
3641 /* Add other packets to a to-do list. */
3642 hash = flow_hash(&miss->flow, 0);
3643 existing_miss = flow_miss_find(&todo, ofproto, &miss->flow, hash);
3644 if (!existing_miss) {
3645 hmap_insert(&todo, &miss->hmap_node, hash);
3646 miss->ofproto = ofproto;
3647 miss->key = upcall->key;
3648 miss->key_len = upcall->key_len;
3649 miss->upcall_type = upcall->type;
3650 list_init(&miss->packets);
3654 miss = existing_miss;
3656 list_push_back(&miss->packets, &upcall->packet->list_node);
3659 /* Process each element in the to-do list, constructing the set of
3660 * operations to batch. */
3662 HMAP_FOR_EACH (miss, hmap_node, &todo) {
3663 handle_flow_miss(miss, flow_miss_ops, &n_ops);
3665 ovs_assert(n_ops <= ARRAY_SIZE(flow_miss_ops));
3667 /* Execute batch. */
3668 for (i = 0; i < n_ops; i++) {
3669 dpif_ops[i] = &flow_miss_ops[i].dpif_op;
3671 dpif_operate(backer->dpif, dpif_ops, n_ops);
3674 for (i = 0; i < n_ops; i++) {
3675 if (flow_miss_ops[i].xout_garbage) {
3676 xlate_out_uninit(&flow_miss_ops[i].xout);
3679 hmap_destroy(&todo);
3682 static enum { SFLOW_UPCALL, MISS_UPCALL, BAD_UPCALL, FLOW_SAMPLE_UPCALL,
3684 classify_upcall(const struct dpif_upcall *upcall)
3686 size_t userdata_len;
3687 union user_action_cookie cookie;
3689 /* First look at the upcall type. */
3690 switch (upcall->type) {
3691 case DPIF_UC_ACTION:
3697 case DPIF_N_UC_TYPES:
3699 VLOG_WARN_RL(&rl, "upcall has unexpected type %"PRIu32, upcall->type);
3703 /* "action" upcalls need a closer look. */
3704 if (!upcall->userdata) {
3705 VLOG_WARN_RL(&rl, "action upcall missing cookie");
3708 userdata_len = nl_attr_get_size(upcall->userdata);
3709 if (userdata_len < sizeof cookie.type
3710 || userdata_len > sizeof cookie) {
3711 VLOG_WARN_RL(&rl, "action upcall cookie has unexpected size %zu",
3715 memset(&cookie, 0, sizeof cookie);
3716 memcpy(&cookie, nl_attr_get(upcall->userdata), userdata_len);
3717 if (userdata_len == sizeof cookie.sflow
3718 && cookie.type == USER_ACTION_COOKIE_SFLOW) {
3719 return SFLOW_UPCALL;
3720 } else if (userdata_len == sizeof cookie.slow_path
3721 && cookie.type == USER_ACTION_COOKIE_SLOW_PATH) {
3723 } else if (userdata_len == sizeof cookie.flow_sample
3724 && cookie.type == USER_ACTION_COOKIE_FLOW_SAMPLE) {
3725 return FLOW_SAMPLE_UPCALL;
3726 } else if (userdata_len == sizeof cookie.ipfix
3727 && cookie.type == USER_ACTION_COOKIE_IPFIX) {
3728 return IPFIX_UPCALL;
3730 VLOG_WARN_RL(&rl, "invalid user cookie of type %"PRIu16
3731 " and size %zu", cookie.type, userdata_len);
3737 handle_sflow_upcall(struct dpif_backer *backer,
3738 const struct dpif_upcall *upcall)
3740 struct ofproto_dpif *ofproto;
3741 union user_action_cookie cookie;
3743 uint32_t odp_in_port;
3745 if (ofproto_receive(backer, upcall->packet, upcall->key, upcall->key_len,
3746 &flow, NULL, &ofproto, &odp_in_port)
3747 || !ofproto->sflow) {
3751 memset(&cookie, 0, sizeof cookie);
3752 memcpy(&cookie, nl_attr_get(upcall->userdata), sizeof cookie.sflow);
3753 dpif_sflow_received(ofproto->sflow, upcall->packet, &flow,
3754 odp_in_port, &cookie);
3758 handle_flow_sample_upcall(struct dpif_backer *backer,
3759 const struct dpif_upcall *upcall)
3761 struct ofproto_dpif *ofproto;
3762 union user_action_cookie cookie;
3765 if (ofproto_receive(backer, upcall->packet, upcall->key, upcall->key_len,
3766 &flow, NULL, &ofproto, NULL)
3767 || !ofproto->ipfix) {
3771 memset(&cookie, 0, sizeof cookie);
3772 memcpy(&cookie, nl_attr_get(upcall->userdata), sizeof cookie.flow_sample);
3774 /* The flow reflects exactly the contents of the packet. Sample
3775 * the packet using it. */
3776 dpif_ipfix_flow_sample(ofproto->ipfix, upcall->packet, &flow,
3777 cookie.flow_sample.collector_set_id,
3778 cookie.flow_sample.probability,
3779 cookie.flow_sample.obs_domain_id,
3780 cookie.flow_sample.obs_point_id);
3784 handle_ipfix_upcall(struct dpif_backer *backer,
3785 const struct dpif_upcall *upcall)
3787 struct ofproto_dpif *ofproto;
3790 if (ofproto_receive(backer, upcall->packet, upcall->key, upcall->key_len,
3791 &flow, NULL, &ofproto, NULL)
3792 || !ofproto->ipfix) {
3796 /* The flow reflects exactly the contents of the packet. Sample
3797 * the packet using it. */
3798 dpif_ipfix_bridge_sample(ofproto->ipfix, upcall->packet, &flow);
3802 handle_upcalls(struct dpif_backer *backer, unsigned int max_batch)
3804 struct dpif_upcall misses[FLOW_MISS_MAX_BATCH];
3805 struct ofpbuf miss_bufs[FLOW_MISS_MAX_BATCH];
3806 uint64_t miss_buf_stubs[FLOW_MISS_MAX_BATCH][4096 / 8];
3811 ovs_assert(max_batch <= FLOW_MISS_MAX_BATCH);
3814 for (n_processed = 0; n_processed < max_batch; n_processed++) {
3815 struct dpif_upcall *upcall = &misses[n_misses];
3816 struct ofpbuf *buf = &miss_bufs[n_misses];
3819 ofpbuf_use_stub(buf, miss_buf_stubs[n_misses],
3820 sizeof miss_buf_stubs[n_misses]);
3821 error = dpif_recv(backer->dpif, upcall, buf);
3827 switch (classify_upcall(upcall)) {
3829 /* Handle it later. */
3834 handle_sflow_upcall(backer, upcall);
3838 case FLOW_SAMPLE_UPCALL:
3839 handle_flow_sample_upcall(backer, upcall);
3844 handle_ipfix_upcall(backer, upcall);
3854 /* Handle deferred MISS_UPCALL processing. */
3855 handle_miss_upcalls(backer, misses, n_misses);
3856 for (i = 0; i < n_misses; i++) {
3857 ofpbuf_uninit(&miss_bufs[i]);
3863 /* Flow expiration. */
3865 static int subfacet_max_idle(const struct dpif_backer *);
3866 static void update_stats(struct dpif_backer *);
3867 static void rule_expire(struct rule_dpif *);
3868 static void expire_subfacets(struct dpif_backer *, int dp_max_idle);
3870 /* This function is called periodically by run(). Its job is to collect
3871 * updates for the flows that have been installed into the datapath, most
3872 * importantly when they last were used, and then use that information to
3873 * expire flows that have not been used recently.
3875 * Returns the number of milliseconds after which it should be called again. */
3877 expire(struct dpif_backer *backer)
3879 struct ofproto_dpif *ofproto;
3883 /* Periodically clear out the drop keys in an effort to keep them
3884 * relatively few. */
3885 drop_key_clear(backer);
3887 /* Update stats for each flow in the backer. */
3888 update_stats(backer);
3890 n_subfacets = hmap_count(&backer->subfacets);
3892 struct subfacet *subfacet;
3893 long long int total, now;
3897 HMAP_FOR_EACH (subfacet, hmap_node, &backer->subfacets) {
3898 total += now - subfacet->created;
3900 backer->avg_subfacet_life += total / n_subfacets;
3902 backer->avg_subfacet_life /= 2;
3904 backer->avg_n_subfacet += n_subfacets;
3905 backer->avg_n_subfacet /= 2;
3907 backer->max_n_subfacet = MAX(backer->max_n_subfacet, n_subfacets);
3909 max_idle = subfacet_max_idle(backer);
3910 expire_subfacets(backer, max_idle);
3912 HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
3913 struct rule *rule, *next_rule;
3915 if (ofproto->backer != backer) {
3919 /* Expire OpenFlow flows whose idle_timeout or hard_timeout
3921 LIST_FOR_EACH_SAFE (rule, next_rule, expirable,
3922 &ofproto->up.expirable) {
3923 rule_expire(rule_dpif_cast(rule));
3926 /* All outstanding data in existing flows has been accounted, so it's a
3927 * good time to do bond rebalancing. */
3928 if (ofproto->has_bonded_bundles) {
3929 struct ofbundle *bundle;
3931 HMAP_FOR_EACH (bundle, hmap_node, &ofproto->bundles) {
3933 bond_rebalance(bundle->bond, &backer->revalidate_set);
3939 return MIN(max_idle, 1000);
3942 /* Updates flow table statistics given that the datapath just reported 'stats'
3943 * as 'subfacet''s statistics. */
3945 update_subfacet_stats(struct subfacet *subfacet,
3946 const struct dpif_flow_stats *stats)
3948 struct facet *facet = subfacet->facet;
3949 struct ofproto_dpif *ofproto = ofproto_dpif_cast(facet->rule->up.ofproto);
3950 struct dpif_flow_stats diff;
3952 diff.tcp_flags = stats->tcp_flags;
3953 diff.used = stats->used;
3955 if (stats->n_packets >= subfacet->dp_packet_count) {
3956 diff.n_packets = stats->n_packets - subfacet->dp_packet_count;
3958 VLOG_WARN_RL(&rl, "unexpected packet count from the datapath");
3962 if (stats->n_bytes >= subfacet->dp_byte_count) {
3963 diff.n_bytes = stats->n_bytes - subfacet->dp_byte_count;
3965 VLOG_WARN_RL(&rl, "unexpected byte count from datapath");
3969 ofproto->n_hit += diff.n_packets;
3970 subfacet->dp_packet_count = stats->n_packets;
3971 subfacet->dp_byte_count = stats->n_bytes;
3972 subfacet_update_stats(subfacet, &diff);
3974 if (facet->accounted_bytes < facet->byte_count) {
3976 facet_account(facet);
3977 facet->accounted_bytes = facet->byte_count;
3981 /* 'key' with length 'key_len' bytes is a flow in 'dpif' that we know nothing
3982 * about, or a flow that shouldn't be installed but was anyway. Delete it. */
3984 delete_unexpected_flow(struct dpif_backer *backer,
3985 const struct nlattr *key, size_t key_len)
3987 if (!VLOG_DROP_WARN(&rl)) {
3991 odp_flow_key_format(key, key_len, &s);
3992 VLOG_WARN("unexpected flow: %s", ds_cstr(&s));
3996 COVERAGE_INC(facet_unexpected);
3997 dpif_flow_del(backer->dpif, key, key_len, NULL);
4000 /* Update 'packet_count', 'byte_count', and 'used' members of installed facets.
4002 * This function also pushes statistics updates to rules which each facet
4003 * resubmits into. Generally these statistics will be accurate. However, if a
4004 * facet changes the rule it resubmits into at some time in between
4005 * update_stats() runs, it is possible that statistics accrued to the
4006 * old rule will be incorrectly attributed to the new rule. This could be
4007 * avoided by calling update_stats() whenever rules are created or
4008 * deleted. However, the performance impact of making so many calls to the
4009 * datapath do not justify the benefit of having perfectly accurate statistics.
4011 * In addition, this function maintains per ofproto flow hit counts. The patch
4012 * port is not treated specially. e.g. A packet ingress from br0 patched into
4013 * br1 will increase the hit count of br0 by 1, however, does not affect
4014 * the hit or miss counts of br1.
4017 update_stats(struct dpif_backer *backer)
4019 const struct dpif_flow_stats *stats;
4020 struct dpif_flow_dump dump;
4021 const struct nlattr *key;
4024 dpif_flow_dump_start(&dump, backer->dpif);
4025 while (dpif_flow_dump_next(&dump, &key, &key_len, NULL, NULL, &stats)) {
4026 struct subfacet *subfacet;
4029 key_hash = odp_flow_key_hash(key, key_len);
4030 subfacet = subfacet_find(backer, key, key_len, key_hash);
4031 switch (subfacet ? subfacet->path : SF_NOT_INSTALLED) {
4033 update_subfacet_stats(subfacet, stats);
4037 /* Stats are updated per-packet. */
4040 case SF_NOT_INSTALLED:
4042 delete_unexpected_flow(backer, key, key_len);
4047 dpif_flow_dump_done(&dump);
4049 update_moving_averages(backer);
4052 /* Calculates and returns the number of milliseconds of idle time after which
4053 * subfacets should expire from the datapath. When a subfacet expires, we fold
4054 * its statistics into its facet, and when a facet's last subfacet expires, we
4055 * fold its statistic into its rule. */
4057 subfacet_max_idle(const struct dpif_backer *backer)
4060 * Idle time histogram.
4062 * Most of the time a switch has a relatively small number of subfacets.
4063 * When this is the case we might as well keep statistics for all of them
4064 * in userspace and to cache them in the kernel datapath for performance as
4067 * As the number of subfacets increases, the memory required to maintain
4068 * statistics about them in userspace and in the kernel becomes
4069 * significant. However, with a large number of subfacets it is likely
4070 * that only a few of them are "heavy hitters" that consume a large amount
4071 * of bandwidth. At this point, only heavy hitters are worth caching in
4072 * the kernel and maintaining in userspaces; other subfacets we can
4075 * The technique used to compute the idle time is to build a histogram with
4076 * N_BUCKETS buckets whose width is BUCKET_WIDTH msecs each. Each subfacet
4077 * that is installed in the kernel gets dropped in the appropriate bucket.
4078 * After the histogram has been built, we compute the cutoff so that only
4079 * the most-recently-used 1% of subfacets (but at least
4080 * flow_eviction_threshold flows) are kept cached. At least
4081 * the most-recently-used bucket of subfacets is kept, so actually an
4082 * arbitrary number of subfacets can be kept in any given expiration run
4083 * (though the next run will delete most of those unless they receive
4086 * This requires a second pass through the subfacets, in addition to the
4087 * pass made by update_stats(), because the former function never looks at
4088 * uninstallable subfacets.
4090 enum { BUCKET_WIDTH = ROUND_UP(100, TIME_UPDATE_INTERVAL) };
4091 enum { N_BUCKETS = 5000 / BUCKET_WIDTH };
4092 int buckets[N_BUCKETS] = { 0 };
4093 int total, subtotal, bucket;
4094 struct subfacet *subfacet;
4098 total = hmap_count(&backer->subfacets);
4099 if (total <= flow_eviction_threshold) {
4100 return N_BUCKETS * BUCKET_WIDTH;
4103 /* Build histogram. */
4105 HMAP_FOR_EACH (subfacet, hmap_node, &backer->subfacets) {
4106 long long int idle = now - subfacet->used;
4107 int bucket = (idle <= 0 ? 0
4108 : idle >= BUCKET_WIDTH * N_BUCKETS ? N_BUCKETS - 1
4109 : (unsigned int) idle / BUCKET_WIDTH);
4113 /* Find the first bucket whose flows should be expired. */
4114 subtotal = bucket = 0;
4116 subtotal += buckets[bucket++];
4117 } while (bucket < N_BUCKETS &&
4118 subtotal < MAX(flow_eviction_threshold, total / 100));
4120 if (VLOG_IS_DBG_ENABLED()) {
4124 ds_put_cstr(&s, "keep");
4125 for (i = 0; i < N_BUCKETS; i++) {
4127 ds_put_cstr(&s, ", drop");
4130 ds_put_format(&s, " %d:%d", i * BUCKET_WIDTH, buckets[i]);
4133 VLOG_INFO("%s (msec:count)", ds_cstr(&s));
4137 return bucket * BUCKET_WIDTH;
4141 expire_subfacets(struct dpif_backer *backer, int dp_max_idle)
4143 /* Cutoff time for most flows. */
4144 long long int normal_cutoff = time_msec() - dp_max_idle;
4146 /* We really want to keep flows for special protocols around, so use a more
4147 * conservative cutoff. */
4148 long long int special_cutoff = time_msec() - 10000;
4150 struct subfacet *subfacet, *next_subfacet;
4151 struct subfacet *batch[SUBFACET_DESTROY_MAX_BATCH];
4155 HMAP_FOR_EACH_SAFE (subfacet, next_subfacet, hmap_node,
4156 &backer->subfacets) {
4157 long long int cutoff;
4159 cutoff = (subfacet->facet->xout.slow & (SLOW_CFM | SLOW_BFD | SLOW_LACP
4163 if (subfacet->used < cutoff) {
4164 if (subfacet->path != SF_NOT_INSTALLED) {
4165 batch[n_batch++] = subfacet;
4166 if (n_batch >= SUBFACET_DESTROY_MAX_BATCH) {
4167 subfacet_destroy_batch(backer, batch, n_batch);
4171 subfacet_destroy(subfacet);
4177 subfacet_destroy_batch(backer, batch, n_batch);
4181 /* If 'rule' is an OpenFlow rule, that has expired according to OpenFlow rules,
4182 * then delete it entirely. */
4184 rule_expire(struct rule_dpif *rule)
4186 struct facet *facet, *next_facet;
4190 if (rule->up.pending) {
4191 /* We'll have to expire it later. */
4195 /* Has 'rule' expired? */
4197 if (rule->up.hard_timeout
4198 && now > rule->up.modified + rule->up.hard_timeout * 1000) {
4199 reason = OFPRR_HARD_TIMEOUT;
4200 } else if (rule->up.idle_timeout
4201 && now > rule->up.used + rule->up.idle_timeout * 1000) {
4202 reason = OFPRR_IDLE_TIMEOUT;
4207 COVERAGE_INC(ofproto_dpif_expired);
4209 /* Update stats. (This is a no-op if the rule expired due to an idle
4210 * timeout, because that only happens when the rule has no facets left.) */
4211 LIST_FOR_EACH_SAFE (facet, next_facet, list_node, &rule->facets) {
4212 facet_remove(facet);
4215 /* Get rid of the rule. */
4216 ofproto_rule_expire(&rule->up, reason);
4221 /* Creates and returns a new facet based on 'miss'.
4223 * The caller must already have determined that no facet with an identical
4224 * 'miss->flow' exists in 'miss->ofproto'.
4226 * 'rule' and 'xout' must have been created based on 'miss'.
4228 * 'facet'' statistics are initialized based on 'stats'.
4230 * The facet will initially have no subfacets. The caller should create (at
4231 * least) one subfacet with subfacet_create(). */
4232 static struct facet *
4233 facet_create(const struct flow_miss *miss, struct rule_dpif *rule,
4234 struct xlate_out *xout, struct dpif_flow_stats *stats)
4236 struct ofproto_dpif *ofproto = miss->ofproto;
4237 struct facet *facet;
4240 facet = xzalloc(sizeof *facet);
4241 facet->packet_count = facet->prev_packet_count = stats->n_packets;
4242 facet->byte_count = facet->prev_byte_count = stats->n_bytes;
4243 facet->tcp_flags = stats->tcp_flags;
4244 facet->used = stats->used;
4245 facet->flow = miss->flow;
4246 facet->learn_rl = time_msec() + 500;
4249 list_push_back(&facet->rule->facets, &facet->list_node);
4250 list_init(&facet->subfacets);
4251 netflow_flow_init(&facet->nf_flow);
4252 netflow_flow_update_time(ofproto->netflow, &facet->nf_flow, facet->used);
4254 xlate_out_copy(&facet->xout, xout);
4256 match_init(&match, &facet->flow, &facet->xout.wc);
4257 cls_rule_init(&facet->cr, &match, OFP_DEFAULT_PRIORITY);
4258 classifier_insert(&ofproto->facets, &facet->cr);
4260 facet->nf_flow.output_iface = facet->xout.nf_output_iface;
4266 facet_free(struct facet *facet)
4269 xlate_out_uninit(&facet->xout);
4274 /* Executes, within 'ofproto', the 'n_actions' actions in 'actions' on
4275 * 'packet', which arrived on 'in_port'. */
4277 execute_odp_actions(struct ofproto_dpif *ofproto, const struct flow *flow,
4278 const struct nlattr *odp_actions, size_t actions_len,
4279 struct ofpbuf *packet)
4281 struct odputil_keybuf keybuf;
4285 ofpbuf_use_stack(&key, &keybuf, sizeof keybuf);
4286 odp_flow_key_from_flow(&key, flow,
4287 ofp_port_to_odp_port(ofproto, flow->in_port));
4289 error = dpif_execute(ofproto->backer->dpif, key.data, key.size,
4290 odp_actions, actions_len, packet);
4294 /* Remove 'facet' from its ofproto and free up the associated memory:
4296 * - If 'facet' was installed in the datapath, uninstalls it and updates its
4297 * rule's statistics, via subfacet_uninstall().
4299 * - Removes 'facet' from its rule and from ofproto->facets.
4302 facet_remove(struct facet *facet)
4304 struct ofproto_dpif *ofproto = ofproto_dpif_cast(facet->rule->up.ofproto);
4305 struct subfacet *subfacet, *next_subfacet;
4307 ovs_assert(!list_is_empty(&facet->subfacets));
4309 /* First uninstall all of the subfacets to get final statistics. */
4310 LIST_FOR_EACH (subfacet, list_node, &facet->subfacets) {
4311 subfacet_uninstall(subfacet);
4314 /* Flush the final stats to the rule.
4316 * This might require us to have at least one subfacet around so that we
4317 * can use its actions for accounting in facet_account(), which is why we
4318 * have uninstalled but not yet destroyed the subfacets. */
4319 facet_flush_stats(facet);
4321 /* Now we're really all done so destroy everything. */
4322 LIST_FOR_EACH_SAFE (subfacet, next_subfacet, list_node,
4323 &facet->subfacets) {
4324 subfacet_destroy__(subfacet);
4326 classifier_remove(&ofproto->facets, &facet->cr);
4327 cls_rule_destroy(&facet->cr);
4328 list_remove(&facet->list_node);
4332 /* Feed information from 'facet' back into the learning table to keep it in
4333 * sync with what is actually flowing through the datapath. */
4335 facet_learn(struct facet *facet)
4337 long long int now = time_msec();
4339 if (!facet->xout.has_fin_timeout && now < facet->learn_rl) {
4343 facet->learn_rl = now + 500;
4345 if (!facet->xout.has_learn
4346 && !facet->xout.has_normal
4347 && (!facet->xout.has_fin_timeout
4348 || !(facet->tcp_flags & (TCP_FIN | TCP_RST)))) {
4352 facet_push_stats(facet, true);
4356 facet_account(struct facet *facet)
4358 struct ofproto_dpif *ofproto = ofproto_dpif_cast(facet->rule->up.ofproto);
4359 const struct nlattr *a;
4364 if (!facet->xout.has_normal || !ofproto->has_bonded_bundles) {
4367 n_bytes = facet->byte_count - facet->accounted_bytes;
4369 /* This loop feeds byte counters to bond_account() for rebalancing to use
4370 * as a basis. We also need to track the actual VLAN on which the packet
4371 * is going to be sent to ensure that it matches the one passed to
4372 * bond_choose_output_slave(). (Otherwise, we will account to the wrong
4375 * We use the actions from an arbitrary subfacet because they should all
4376 * be equally valid for our purpose. */
4377 vlan_tci = facet->flow.vlan_tci;
4378 NL_ATTR_FOR_EACH_UNSAFE (a, left, facet->xout.odp_actions.data,
4379 facet->xout.odp_actions.size) {
4380 const struct ovs_action_push_vlan *vlan;
4381 struct ofport_dpif *port;
4383 switch (nl_attr_type(a)) {
4384 case OVS_ACTION_ATTR_OUTPUT:
4385 port = get_odp_port(ofproto, nl_attr_get_u32(a));
4386 if (port && port->bundle && port->bundle->bond) {
4387 bond_account(port->bundle->bond, &facet->flow,
4388 vlan_tci_to_vid(vlan_tci), n_bytes);
4392 case OVS_ACTION_ATTR_POP_VLAN:
4393 vlan_tci = htons(0);
4396 case OVS_ACTION_ATTR_PUSH_VLAN:
4397 vlan = nl_attr_get(a);
4398 vlan_tci = vlan->vlan_tci;
4404 /* Returns true if the only action for 'facet' is to send to the controller.
4405 * (We don't report NetFlow expiration messages for such facets because they
4406 * are just part of the control logic for the network, not real traffic). */
4408 facet_is_controller_flow(struct facet *facet)
4411 const struct rule *rule = &facet->rule->up;
4412 const struct ofpact *ofpacts = rule->ofpacts;
4413 size_t ofpacts_len = rule->ofpacts_len;
4415 if (ofpacts_len > 0 &&
4416 ofpacts->type == OFPACT_CONTROLLER &&
4417 ofpact_next(ofpacts) >= ofpact_end(ofpacts, ofpacts_len)) {
4424 /* Folds all of 'facet''s statistics into its rule. Also updates the
4425 * accounting ofhook and emits a NetFlow expiration if appropriate. All of
4426 * 'facet''s statistics in the datapath should have been zeroed and folded into
4427 * its packet and byte counts before this function is called. */
4429 facet_flush_stats(struct facet *facet)
4431 struct ofproto_dpif *ofproto = ofproto_dpif_cast(facet->rule->up.ofproto);
4432 struct subfacet *subfacet;
4434 LIST_FOR_EACH (subfacet, list_node, &facet->subfacets) {
4435 ovs_assert(!subfacet->dp_byte_count);
4436 ovs_assert(!subfacet->dp_packet_count);
4439 facet_push_stats(facet, false);
4440 if (facet->accounted_bytes < facet->byte_count) {
4441 facet_account(facet);
4442 facet->accounted_bytes = facet->byte_count;
4445 if (ofproto->netflow && !facet_is_controller_flow(facet)) {
4446 struct ofexpired expired;
4447 expired.flow = facet->flow;
4448 expired.packet_count = facet->packet_count;
4449 expired.byte_count = facet->byte_count;
4450 expired.used = facet->used;
4451 netflow_expire(ofproto->netflow, &facet->nf_flow, &expired);
4454 /* Reset counters to prevent double counting if 'facet' ever gets
4456 facet_reset_counters(facet);
4458 netflow_flow_clear(&facet->nf_flow);
4459 facet->tcp_flags = 0;
4462 /* Searches 'ofproto''s table of facets for one which would be responsible for
4463 * 'flow'. Returns it if found, otherwise a null pointer.
4465 * The returned facet might need revalidation; use facet_lookup_valid()
4466 * instead if that is important. */
4467 static struct facet *
4468 facet_find(struct ofproto_dpif *ofproto, const struct flow *flow)
4470 struct cls_rule *cr = classifier_lookup(&ofproto->facets, flow, NULL);
4471 return cr ? CONTAINER_OF(cr, struct facet, cr) : NULL;
4474 /* Searches 'ofproto''s table of facets for one capable that covers
4475 * 'flow'. Returns it if found, otherwise a null pointer.
4477 * The returned facet is guaranteed to be valid. */
4478 static struct facet *
4479 facet_lookup_valid(struct ofproto_dpif *ofproto, const struct flow *flow)
4481 struct facet *facet;
4483 facet = facet_find(ofproto, flow);
4485 && (ofproto->backer->need_revalidate
4486 || tag_set_intersects(&ofproto->backer->revalidate_set,
4488 && !facet_revalidate(facet)) {
4496 facet_check_consistency(struct facet *facet)
4498 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 15);
4500 struct ofproto_dpif *ofproto = ofproto_dpif_cast(facet->rule->up.ofproto);
4502 struct xlate_out xout;
4503 struct xlate_in xin;
4505 struct rule_dpif *rule;
4508 /* Check the rule for consistency. */
4509 rule = rule_dpif_lookup(ofproto, &facet->flow, NULL);
4510 if (rule != facet->rule) {
4511 if (!VLOG_DROP_WARN(&rl)) {
4512 struct ds s = DS_EMPTY_INITIALIZER;
4514 flow_format(&s, &facet->flow);
4515 ds_put_format(&s, ": facet associated with wrong rule (was "
4516 "table=%"PRIu8",", facet->rule->up.table_id);
4517 cls_rule_format(&facet->rule->up.cr, &s);
4518 ds_put_format(&s, ") (should have been table=%"PRIu8",",
4520 cls_rule_format(&rule->up.cr, &s);
4521 ds_put_char(&s, ')');
4523 VLOG_WARN("%s", ds_cstr(&s));
4529 /* Check the datapath actions for consistency. */
4530 xlate_in_init(&xin, ofproto, &facet->flow, rule, 0, NULL);
4531 xlate_actions(&xin, &xout);
4533 ok = ofpbuf_equal(&facet->xout.odp_actions, &xout.odp_actions)
4534 && facet->xout.slow == xout.slow;
4535 if (!ok && !VLOG_DROP_WARN(&rl)) {
4536 struct ds s = DS_EMPTY_INITIALIZER;
4538 flow_format(&s, &facet->flow);
4539 ds_put_cstr(&s, ": inconsistency in facet");
4541 if (!ofpbuf_equal(&facet->xout.odp_actions, &xout.odp_actions)) {
4542 ds_put_cstr(&s, " (actions were: ");
4543 format_odp_actions(&s, facet->xout.odp_actions.data,
4544 facet->xout.odp_actions.size);
4545 ds_put_cstr(&s, ") (correct actions: ");
4546 format_odp_actions(&s, xout.odp_actions.data,
4547 xout.odp_actions.size);
4548 ds_put_char(&s, ')');
4551 if (facet->xout.slow != xout.slow) {
4552 ds_put_format(&s, " slow path incorrect. should be %d", xout.slow);
4555 VLOG_WARN("%s", ds_cstr(&s));
4558 xlate_out_uninit(&xout);
4563 /* Re-searches the classifier for 'facet':
4565 * - If the rule found is different from 'facet''s current rule, moves
4566 * 'facet' to the new rule and recompiles its actions.
4568 * - If the rule found is the same as 'facet''s current rule, leaves 'facet'
4569 * where it is and recompiles its actions anyway.
4571 * - If any of 'facet''s subfacets correspond to a new flow according to
4572 * ofproto_receive(), 'facet' is removed.
4574 * Returns true if 'facet' is still valid. False if 'facet' was removed. */
4576 facet_revalidate(struct facet *facet)
4578 struct ofproto_dpif *ofproto = ofproto_dpif_cast(facet->rule->up.ofproto);
4579 struct rule_dpif *new_rule;
4580 struct subfacet *subfacet;
4581 struct flow_wildcards wc;
4582 struct xlate_out xout;
4583 struct xlate_in xin;
4585 COVERAGE_INC(facet_revalidate);
4587 /* Check that child subfacets still correspond to this facet. Tunnel
4588 * configuration changes could cause a subfacet's OpenFlow in_port to
4590 LIST_FOR_EACH (subfacet, list_node, &facet->subfacets) {
4591 struct ofproto_dpif *recv_ofproto;
4592 struct flow recv_flow;
4595 error = ofproto_receive(ofproto->backer, NULL, subfacet->key,
4596 subfacet->key_len, &recv_flow, NULL,
4597 &recv_ofproto, NULL);
4599 || recv_ofproto != ofproto
4600 || facet != facet_find(ofproto, &recv_flow)) {
4601 facet_remove(facet);
4606 flow_wildcards_init_catchall(&wc);
4607 new_rule = rule_dpif_lookup(ofproto, &facet->flow, &wc);
4609 /* Calculate new datapath actions.
4611 * We do not modify any 'facet' state yet, because we might need to, e.g.,
4612 * emit a NetFlow expiration and, if so, we need to have the old state
4613 * around to properly compose it. */
4614 xlate_in_init(&xin, ofproto, &facet->flow, new_rule, 0, NULL);
4615 xlate_actions(&xin, &xout);
4616 flow_wildcards_or(&xout.wc, &xout.wc, &wc);
4618 /* A facet's slow path reason should only change under dramatic
4619 * circumstances. Rather than try to update everything, it's simpler to
4620 * remove the facet and start over.
4622 * More importantly, if a facet's wildcards change, it will be relatively
4623 * difficult to figure out if its subfacets still belong to it, and if not
4624 * which facet they may belong to. Again, to avoid the complexity, we
4625 * simply give up instead. */
4626 if (facet->xout.slow != xout.slow
4627 || memcmp(&facet->xout.wc, &xout.wc, sizeof xout.wc)) {
4628 facet_remove(facet);
4629 xlate_out_uninit(&xout);
4633 if (!ofpbuf_equal(&facet->xout.odp_actions, &xout.odp_actions)) {
4634 LIST_FOR_EACH(subfacet, list_node, &facet->subfacets) {
4635 if (subfacet->path == SF_FAST_PATH) {
4636 struct dpif_flow_stats stats;
4638 subfacet_install(subfacet, &xout.odp_actions, &stats);
4639 subfacet_update_stats(subfacet, &stats);
4643 facet_flush_stats(facet);
4645 ofpbuf_clear(&facet->xout.odp_actions);
4646 ofpbuf_put(&facet->xout.odp_actions, xout.odp_actions.data,
4647 xout.odp_actions.size);
4650 /* Update 'facet' now that we've taken care of all the old state. */
4651 facet->xout.tags = xout.tags;
4652 facet->xout.slow = xout.slow;
4653 facet->xout.has_learn = xout.has_learn;
4654 facet->xout.has_normal = xout.has_normal;
4655 facet->xout.has_fin_timeout = xout.has_fin_timeout;
4656 facet->xout.nf_output_iface = xout.nf_output_iface;
4657 facet->xout.mirrors = xout.mirrors;
4658 facet->nf_flow.output_iface = facet->xout.nf_output_iface;
4660 if (facet->rule != new_rule) {
4661 COVERAGE_INC(facet_changed_rule);
4662 list_remove(&facet->list_node);
4663 list_push_back(&new_rule->facets, &facet->list_node);
4664 facet->rule = new_rule;
4665 facet->used = new_rule->up.created;
4666 facet->prev_used = facet->used;
4669 xlate_out_uninit(&xout);
4674 facet_reset_counters(struct facet *facet)
4676 facet->packet_count = 0;
4677 facet->byte_count = 0;
4678 facet->prev_packet_count = 0;
4679 facet->prev_byte_count = 0;
4680 facet->accounted_bytes = 0;
4684 facet_push_stats(struct facet *facet, bool may_learn)
4686 struct dpif_flow_stats stats;
4688 ovs_assert(facet->packet_count >= facet->prev_packet_count);
4689 ovs_assert(facet->byte_count >= facet->prev_byte_count);
4690 ovs_assert(facet->used >= facet->prev_used);
4692 stats.n_packets = facet->packet_count - facet->prev_packet_count;
4693 stats.n_bytes = facet->byte_count - facet->prev_byte_count;
4694 stats.used = facet->used;
4695 stats.tcp_flags = facet->tcp_flags;
4697 if (may_learn || stats.n_packets || facet->used > facet->prev_used) {
4698 struct ofproto_dpif *ofproto =
4699 ofproto_dpif_cast(facet->rule->up.ofproto);
4701 struct ofport_dpif *in_port;
4702 struct xlate_in xin;
4704 facet->prev_packet_count = facet->packet_count;
4705 facet->prev_byte_count = facet->byte_count;
4706 facet->prev_used = facet->used;
4708 in_port = get_ofp_port(ofproto, facet->flow.in_port);
4709 if (in_port && in_port->tnl_port) {
4710 netdev_vport_inc_rx(in_port->up.netdev, &stats);
4713 rule_credit_stats(facet->rule, &stats);
4714 netflow_flow_update_time(ofproto->netflow, &facet->nf_flow,
4716 netflow_flow_update_flags(&facet->nf_flow, facet->tcp_flags);
4717 update_mirror_stats(ofproto, facet->xout.mirrors, stats.n_packets,
4720 xlate_in_init(&xin, ofproto, &facet->flow, facet->rule,
4721 stats.tcp_flags, NULL);
4722 xin.resubmit_stats = &stats;
4723 xin.may_learn = may_learn;
4724 xlate_actions_for_side_effects(&xin);
4729 push_all_stats__(bool run_fast)
4731 static long long int rl = LLONG_MIN;
4732 struct ofproto_dpif *ofproto;
4734 if (time_msec() < rl) {
4738 HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
4739 struct cls_cursor cursor;
4740 struct facet *facet;
4742 cls_cursor_init(&cursor, &ofproto->facets, NULL);
4743 CLS_CURSOR_FOR_EACH (facet, cr, &cursor) {
4744 facet_push_stats(facet, false);
4751 rl = time_msec() + 100;
4755 push_all_stats(void)
4757 push_all_stats__(true);
4761 rule_credit_stats(struct rule_dpif *rule, const struct dpif_flow_stats *stats)
4763 rule->packet_count += stats->n_packets;
4764 rule->byte_count += stats->n_bytes;
4765 ofproto_rule_update_used(&rule->up, stats->used);
4770 static struct subfacet *
4771 subfacet_find(struct dpif_backer *backer, const struct nlattr *key,
4772 size_t key_len, uint32_t key_hash)
4774 struct subfacet *subfacet;
4776 HMAP_FOR_EACH_WITH_HASH (subfacet, hmap_node, key_hash,
4777 &backer->subfacets) {
4778 if (subfacet->key_len == key_len
4779 && !memcmp(key, subfacet->key, key_len)) {
4787 /* Searches 'facet' (within 'ofproto') for a subfacet with the specified
4788 * 'key_fitness', 'key', and 'key_len' members in 'miss'. Returns the
4789 * existing subfacet if there is one, otherwise creates and returns a
4791 static struct subfacet *
4792 subfacet_create(struct facet *facet, struct flow_miss *miss,
4795 struct dpif_backer *backer = miss->ofproto->backer;
4796 enum odp_key_fitness key_fitness = miss->key_fitness;
4797 const struct nlattr *key = miss->key;
4798 size_t key_len = miss->key_len;
4800 struct subfacet *subfacet;
4802 key_hash = odp_flow_key_hash(key, key_len);
4804 if (list_is_empty(&facet->subfacets)) {
4805 subfacet = &facet->one_subfacet;
4807 subfacet = subfacet_find(backer, key, key_len, key_hash);
4809 if (subfacet->facet == facet) {
4813 /* This shouldn't happen. */
4814 VLOG_ERR_RL(&rl, "subfacet with wrong facet");
4815 subfacet_destroy(subfacet);
4818 subfacet = xmalloc(sizeof *subfacet);
4821 hmap_insert(&backer->subfacets, &subfacet->hmap_node, key_hash);
4822 list_push_back(&facet->subfacets, &subfacet->list_node);
4823 subfacet->facet = facet;
4824 subfacet->key_fitness = key_fitness;
4825 subfacet->key = xmemdup(key, key_len);
4826 subfacet->key_len = key_len;
4827 subfacet->used = now;
4828 subfacet->created = now;
4829 subfacet->dp_packet_count = 0;
4830 subfacet->dp_byte_count = 0;
4831 subfacet->path = SF_NOT_INSTALLED;
4832 subfacet->backer = backer;
4834 backer->subfacet_add_count++;
4838 /* Uninstalls 'subfacet' from the datapath, if it is installed, removes it from
4839 * its facet within 'ofproto', and frees it. */
4841 subfacet_destroy__(struct subfacet *subfacet)
4843 struct facet *facet = subfacet->facet;
4844 struct ofproto_dpif *ofproto = ofproto_dpif_cast(facet->rule->up.ofproto);
4846 /* Update ofproto stats before uninstall the subfacet. */
4847 ofproto->backer->subfacet_del_count++;
4849 subfacet_uninstall(subfacet);
4850 hmap_remove(&subfacet->backer->subfacets, &subfacet->hmap_node);
4851 list_remove(&subfacet->list_node);
4852 free(subfacet->key);
4853 if (subfacet != &facet->one_subfacet) {
4858 /* Destroys 'subfacet', as with subfacet_destroy__(), and then if this was the
4859 * last remaining subfacet in its facet destroys the facet too. */
4861 subfacet_destroy(struct subfacet *subfacet)
4863 struct facet *facet = subfacet->facet;
4865 if (list_is_singleton(&facet->subfacets)) {
4866 /* facet_remove() needs at least one subfacet (it will remove it). */
4867 facet_remove(facet);
4869 subfacet_destroy__(subfacet);
4874 subfacet_destroy_batch(struct dpif_backer *backer,
4875 struct subfacet **subfacets, int n)
4877 struct dpif_op ops[SUBFACET_DESTROY_MAX_BATCH];
4878 struct dpif_op *opsp[SUBFACET_DESTROY_MAX_BATCH];
4879 struct dpif_flow_stats stats[SUBFACET_DESTROY_MAX_BATCH];
4882 for (i = 0; i < n; i++) {
4883 ops[i].type = DPIF_OP_FLOW_DEL;
4884 ops[i].u.flow_del.key = subfacets[i]->key;
4885 ops[i].u.flow_del.key_len = subfacets[i]->key_len;
4886 ops[i].u.flow_del.stats = &stats[i];
4890 dpif_operate(backer->dpif, opsp, n);
4891 for (i = 0; i < n; i++) {
4892 subfacet_reset_dp_stats(subfacets[i], &stats[i]);
4893 subfacets[i]->path = SF_NOT_INSTALLED;
4894 subfacet_destroy(subfacets[i]);
4899 /* Updates 'subfacet''s datapath flow, setting its actions to 'actions_len'
4900 * bytes of actions in 'actions'. If 'stats' is non-null, statistics counters
4901 * in the datapath will be zeroed and 'stats' will be updated with traffic new
4902 * since 'subfacet' was last updated.
4904 * Returns 0 if successful, otherwise a positive errno value. */
4906 subfacet_install(struct subfacet *subfacet, const struct ofpbuf *odp_actions,
4907 struct dpif_flow_stats *stats)
4909 struct facet *facet = subfacet->facet;
4910 struct ofproto_dpif *ofproto = ofproto_dpif_cast(facet->rule->up.ofproto);
4911 enum subfacet_path path = facet->xout.slow ? SF_SLOW_PATH : SF_FAST_PATH;
4912 const struct nlattr *actions = odp_actions->data;
4913 size_t actions_len = odp_actions->size;
4915 uint64_t slow_path_stub[128 / 8];
4916 enum dpif_flow_put_flags flags;
4919 flags = DPIF_FP_CREATE | DPIF_FP_MODIFY;
4921 flags |= DPIF_FP_ZERO_STATS;
4924 if (path == SF_SLOW_PATH) {
4925 compose_slow_path(ofproto, &facet->flow, facet->xout.slow,
4926 slow_path_stub, sizeof slow_path_stub,
4927 &actions, &actions_len);
4930 ret = dpif_flow_put(subfacet->backer->dpif, flags, subfacet->key,
4931 subfacet->key_len, actions, actions_len, stats);
4934 subfacet_reset_dp_stats(subfacet, stats);
4938 subfacet->path = path;
4943 /* If 'subfacet' is installed in the datapath, uninstalls it. */
4945 subfacet_uninstall(struct subfacet *subfacet)
4947 if (subfacet->path != SF_NOT_INSTALLED) {
4948 struct rule_dpif *rule = subfacet->facet->rule;
4949 struct ofproto_dpif *ofproto = ofproto_dpif_cast(rule->up.ofproto);
4950 struct dpif_flow_stats stats;
4953 error = dpif_flow_del(ofproto->backer->dpif, subfacet->key,
4954 subfacet->key_len, &stats);
4955 subfacet_reset_dp_stats(subfacet, &stats);
4957 subfacet_update_stats(subfacet, &stats);
4959 subfacet->path = SF_NOT_INSTALLED;
4961 ovs_assert(subfacet->dp_packet_count == 0);
4962 ovs_assert(subfacet->dp_byte_count == 0);
4966 /* Resets 'subfacet''s datapath statistics counters. This should be called
4967 * when 'subfacet''s statistics are cleared in the datapath. If 'stats' is
4968 * non-null, it should contain the statistics returned by dpif when 'subfacet'
4969 * was reset in the datapath. 'stats' will be modified to include only
4970 * statistics new since 'subfacet' was last updated. */
4972 subfacet_reset_dp_stats(struct subfacet *subfacet,
4973 struct dpif_flow_stats *stats)
4976 && subfacet->dp_packet_count <= stats->n_packets
4977 && subfacet->dp_byte_count <= stats->n_bytes) {
4978 stats->n_packets -= subfacet->dp_packet_count;
4979 stats->n_bytes -= subfacet->dp_byte_count;
4982 subfacet->dp_packet_count = 0;
4983 subfacet->dp_byte_count = 0;
4986 /* Folds the statistics from 'stats' into the counters in 'subfacet'.
4988 * Because of the meaning of a subfacet's counters, it only makes sense to do
4989 * this if 'stats' are not tracked in the datapath, that is, if 'stats'
4990 * represents a packet that was sent by hand or if it represents statistics
4991 * that have been cleared out of the datapath. */
4993 subfacet_update_stats(struct subfacet *subfacet,
4994 const struct dpif_flow_stats *stats)
4996 if (stats->n_packets || stats->used > subfacet->used) {
4997 struct facet *facet = subfacet->facet;
4999 subfacet->used = MAX(subfacet->used, stats->used);
5000 facet->used = MAX(facet->used, stats->used);
5001 facet->packet_count += stats->n_packets;
5002 facet->byte_count += stats->n_bytes;
5003 facet->tcp_flags |= stats->tcp_flags;
5009 /* Lookup 'flow' in 'ofproto''s classifier. If 'wc' is non-null, sets
5010 * the fields that were relevant as part of the lookup. */
5011 static struct rule_dpif *
5012 rule_dpif_lookup(struct ofproto_dpif *ofproto, const struct flow *flow,
5013 struct flow_wildcards *wc)
5015 struct rule_dpif *rule;
5017 rule = rule_dpif_lookup_in_table(ofproto, flow, wc, 0);
5022 return rule_dpif_miss_rule(ofproto, flow);
5026 rule_dpif_lookup_in_table(struct ofproto_dpif *ofproto,
5027 const struct flow *flow, struct flow_wildcards *wc,
5030 struct cls_rule *cls_rule;
5031 struct classifier *cls;
5034 if (table_id >= N_TABLES) {
5038 cls = &ofproto->up.tables[table_id].cls;
5039 frag = (flow->nw_frag & FLOW_NW_FRAG_ANY) != 0;
5040 if (frag && ofproto->up.frag_handling == OFPC_FRAG_NORMAL) {
5041 /* We must pretend that transport ports are unavailable. */
5042 struct flow ofpc_normal_flow = *flow;
5043 ofpc_normal_flow.tp_src = htons(0);
5044 ofpc_normal_flow.tp_dst = htons(0);
5045 cls_rule = classifier_lookup(cls, &ofpc_normal_flow, wc);
5046 } else if (frag && ofproto->up.frag_handling == OFPC_FRAG_DROP) {
5047 cls_rule = &ofproto->drop_frags_rule->up.cr;
5049 flow_wildcards_init_exact(wc);
5052 cls_rule = classifier_lookup(cls, flow, wc);
5054 return rule_dpif_cast(rule_from_cls_rule(cls_rule));
5058 rule_dpif_miss_rule(struct ofproto_dpif *ofproto, const struct flow *flow)
5060 struct ofport_dpif *port;
5062 port = get_ofp_port(ofproto, flow->in_port);
5064 VLOG_WARN_RL(&rl, "packet-in on unknown port %"PRIu16, flow->in_port);
5065 return ofproto->miss_rule;
5068 if (port->up.pp.config & OFPUTIL_PC_NO_PACKET_IN) {
5069 return ofproto->no_packet_in_rule;
5071 return ofproto->miss_rule;
5075 complete_operation(struct rule_dpif *rule)
5077 struct ofproto_dpif *ofproto = ofproto_dpif_cast(rule->up.ofproto);
5079 rule_invalidate(rule);
5081 struct dpif_completion *c = xmalloc(sizeof *c);
5082 c->op = rule->up.pending;
5083 list_push_back(&ofproto->completions, &c->list_node);
5085 ofoperation_complete(rule->up.pending, 0);
5089 static struct rule *
5092 struct rule_dpif *rule = xmalloc(sizeof *rule);
5097 rule_dealloc(struct rule *rule_)
5099 struct rule_dpif *rule = rule_dpif_cast(rule_);
5104 rule_construct(struct rule *rule_)
5106 struct rule_dpif *rule = rule_dpif_cast(rule_);
5107 struct ofproto_dpif *ofproto = ofproto_dpif_cast(rule->up.ofproto);
5108 struct rule_dpif *victim;
5111 rule->packet_count = 0;
5112 rule->byte_count = 0;
5114 victim = rule_dpif_cast(ofoperation_get_victim(rule->up.pending));
5115 if (victim && !list_is_empty(&victim->facets)) {
5116 struct facet *facet;
5118 rule->facets = victim->facets;
5119 list_moved(&rule->facets);
5120 LIST_FOR_EACH (facet, list_node, &rule->facets) {
5121 /* XXX: We're only clearing our local counters here. It's possible
5122 * that quite a few packets are unaccounted for in the datapath
5123 * statistics. These will be accounted to the new rule instead of
5124 * cleared as required. This could be fixed by clearing out the
5125 * datapath statistics for this facet, but currently it doesn't
5127 facet_reset_counters(facet);
5131 /* Must avoid list_moved() in this case. */
5132 list_init(&rule->facets);
5135 table_id = rule->up.table_id;
5137 rule->tag = victim->tag;
5138 } else if (table_id == 0) {
5143 miniflow_expand(&rule->up.cr.match.flow, &flow);
5144 rule->tag = rule_calculate_tag(&flow, &rule->up.cr.match.mask,
5145 ofproto->tables[table_id].basis);
5148 complete_operation(rule);
5153 rule_destruct(struct rule *rule_)
5155 struct rule_dpif *rule = rule_dpif_cast(rule_);
5156 struct facet *facet, *next_facet;
5158 LIST_FOR_EACH_SAFE (facet, next_facet, list_node, &rule->facets) {
5159 facet_revalidate(facet);
5162 complete_operation(rule);
5166 rule_get_stats(struct rule *rule_, uint64_t *packets, uint64_t *bytes)
5168 struct rule_dpif *rule = rule_dpif_cast(rule_);
5170 /* push_all_stats() can handle flow misses which, when using the learn
5171 * action, can cause rules to be added and deleted. This can corrupt our
5172 * caller's datastructures which assume that rule_get_stats() doesn't have
5173 * an impact on the flow table. To be safe, we disable miss handling. */
5174 push_all_stats__(false);
5176 /* Start from historical data for 'rule' itself that are no longer tracked
5177 * in facets. This counts, for example, facets that have expired. */
5178 *packets = rule->packet_count;
5179 *bytes = rule->byte_count;
5183 rule_dpif_execute(struct rule_dpif *rule, const struct flow *flow,
5184 struct ofpbuf *packet)
5186 struct ofproto_dpif *ofproto = ofproto_dpif_cast(rule->up.ofproto);
5187 struct dpif_flow_stats stats;
5188 struct xlate_out xout;
5189 struct xlate_in xin;
5191 dpif_flow_stats_extract(flow, packet, time_msec(), &stats);
5192 rule_credit_stats(rule, &stats);
5194 xlate_in_init(&xin, ofproto, flow, rule, stats.tcp_flags, packet);
5195 xin.resubmit_stats = &stats;
5196 xlate_actions(&xin, &xout);
5198 execute_odp_actions(ofproto, flow, xout.odp_actions.data,
5199 xout.odp_actions.size, packet);
5201 xlate_out_uninit(&xout);
5205 rule_execute(struct rule *rule, const struct flow *flow,
5206 struct ofpbuf *packet)
5208 rule_dpif_execute(rule_dpif_cast(rule), flow, packet);
5209 ofpbuf_delete(packet);
5214 rule_modify_actions(struct rule *rule_)
5216 struct rule_dpif *rule = rule_dpif_cast(rule_);
5218 complete_operation(rule);
5221 /* Sends 'packet' out 'ofport'.
5222 * May modify 'packet'.
5223 * Returns 0 if successful, otherwise a positive errno value. */
5225 send_packet(const struct ofport_dpif *ofport, struct ofpbuf *packet)
5227 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport->up.ofproto);
5228 uint64_t odp_actions_stub[1024 / 8];
5229 struct ofpbuf key, odp_actions;
5230 struct dpif_flow_stats stats;
5231 struct odputil_keybuf keybuf;
5232 struct ofpact_output output;
5233 struct xlate_out xout;
5234 struct xlate_in xin;
5238 ofpbuf_use_stub(&odp_actions, odp_actions_stub, sizeof odp_actions_stub);
5239 ofpbuf_use_stack(&key, &keybuf, sizeof keybuf);
5241 /* Use OFPP_NONE as the in_port to avoid special packet processing. */
5242 flow_extract(packet, 0, 0, NULL, OFPP_NONE, &flow);
5243 odp_flow_key_from_flow(&key, &flow, ofp_port_to_odp_port(ofproto,
5245 dpif_flow_stats_extract(&flow, packet, time_msec(), &stats);
5247 ofpact_init(&output.ofpact, OFPACT_OUTPUT, sizeof output);
5248 output.port = ofport->up.ofp_port;
5251 xlate_in_init(&xin, ofproto, &flow, NULL, 0, packet);
5252 xin.ofpacts_len = sizeof output;
5253 xin.ofpacts = &output.ofpact;
5254 xin.resubmit_stats = &stats;
5255 xlate_actions(&xin, &xout);
5257 error = dpif_execute(ofproto->backer->dpif,
5259 xout.odp_actions.data, xout.odp_actions.size,
5261 xlate_out_uninit(&xout);
5264 VLOG_WARN_RL(&rl, "%s: failed to send packet on port %s (%s)",
5265 ofproto->up.name, netdev_get_name(ofport->up.netdev),
5269 ofproto->stats.tx_packets++;
5270 ofproto->stats.tx_bytes += packet->size;
5274 /* Composes an ODP action for a "slow path" action for 'flow' within 'ofproto'.
5275 * The action will state 'slow' as the reason that the action is in the slow
5276 * path. (This is purely informational: it allows a human viewing "ovs-dpctl
5277 * dump-flows" output to see why a flow is in the slow path.)
5279 * The 'stub_size' bytes in 'stub' will be used to store the action.
5280 * 'stub_size' must be large enough for the action.
5282 * The action and its size will be stored in '*actionsp' and '*actions_lenp',
5285 compose_slow_path(const struct ofproto_dpif *ofproto, const struct flow *flow,
5286 enum slow_path_reason slow,
5287 uint64_t *stub, size_t stub_size,
5288 const struct nlattr **actionsp, size_t *actions_lenp)
5290 union user_action_cookie cookie;
5293 cookie.type = USER_ACTION_COOKIE_SLOW_PATH;
5294 cookie.slow_path.unused = 0;
5295 cookie.slow_path.reason = slow;
5297 ofpbuf_use_stack(&buf, stub, stub_size);
5298 if (slow & (SLOW_CFM | SLOW_BFD | SLOW_LACP | SLOW_STP)) {
5299 uint32_t pid = dpif_port_get_pid(ofproto->backer->dpif, UINT32_MAX);
5300 odp_put_userspace_action(pid, &cookie, sizeof cookie.slow_path, &buf);
5302 put_userspace_action(ofproto, &buf, flow, &cookie,
5303 sizeof cookie.slow_path);
5305 *actionsp = buf.data;
5306 *actions_lenp = buf.size;
5310 put_userspace_action(const struct ofproto_dpif *ofproto,
5311 struct ofpbuf *odp_actions,
5312 const struct flow *flow,
5313 const union user_action_cookie *cookie,
5314 const size_t cookie_size)
5318 pid = dpif_port_get_pid(ofproto->backer->dpif,
5319 ofp_port_to_odp_port(ofproto, flow->in_port));
5321 return odp_put_userspace_action(pid, cookie, cookie_size, odp_actions);
5326 update_mirror_stats(struct ofproto_dpif *ofproto, mirror_mask_t mirrors,
5327 uint64_t packets, uint64_t bytes)
5333 for (; mirrors; mirrors = zero_rightmost_1bit(mirrors)) {
5336 m = ofproto->mirrors[mirror_mask_ffs(mirrors) - 1];
5339 /* In normal circumstances 'm' will not be NULL. However,
5340 * if mirrors are reconfigured, we can temporarily get out
5341 * of sync in facet_revalidate(). We could "correct" the
5342 * mirror list before reaching here, but doing that would
5343 * not properly account the traffic stats we've currently
5344 * accumulated for previous mirror configuration. */
5348 m->packet_count += packets;
5349 m->byte_count += bytes;
5354 /* Optimized flow revalidation.
5356 * It's a difficult problem, in general, to tell which facets need to have
5357 * their actions recalculated whenever the OpenFlow flow table changes. We
5358 * don't try to solve that general problem: for most kinds of OpenFlow flow
5359 * table changes, we recalculate the actions for every facet. This is
5360 * relatively expensive, but it's good enough if the OpenFlow flow table
5361 * doesn't change very often.
5363 * However, we can expect one particular kind of OpenFlow flow table change to
5364 * happen frequently: changes caused by MAC learning. To avoid wasting a lot
5365 * of CPU on revalidating every facet whenever MAC learning modifies the flow
5366 * table, we add a special case that applies to flow tables in which every rule
5367 * has the same form (that is, the same wildcards), except that the table is
5368 * also allowed to have a single "catch-all" flow that matches all packets. We
5369 * optimize this case by tagging all of the facets that resubmit into the table
5370 * and invalidating the same tag whenever a flow changes in that table. The
5371 * end result is that we revalidate just the facets that need it (and sometimes
5372 * a few more, but not all of the facets or even all of the facets that
5373 * resubmit to the table modified by MAC learning). */
5375 /* Calculates the tag to use for 'flow' and mask 'mask' when it is inserted
5376 * into an OpenFlow table with the given 'basis'. */
5378 rule_calculate_tag(const struct flow *flow, const struct minimask *mask,
5381 if (minimask_is_catchall(mask)) {
5384 uint32_t hash = flow_hash_in_minimask(flow, mask, secret);
5385 return tag_create_deterministic(hash);
5389 /* Following a change to OpenFlow table 'table_id' in 'ofproto', update the
5390 * taggability of that table.
5392 * This function must be called after *each* change to a flow table. If you
5393 * skip calling it on some changes then the pointer comparisons at the end can
5394 * be invalid if you get unlucky. For example, if a flow removal causes a
5395 * cls_table to be destroyed and then a flow insertion causes a cls_table with
5396 * different wildcards to be created with the same address, then this function
5397 * will incorrectly skip revalidation. */
5399 table_update_taggable(struct ofproto_dpif *ofproto, uint8_t table_id)
5401 struct table_dpif *table = &ofproto->tables[table_id];
5402 const struct oftable *oftable = &ofproto->up.tables[table_id];
5403 struct cls_table *catchall, *other;
5404 struct cls_table *t;
5406 catchall = other = NULL;
5408 switch (hmap_count(&oftable->cls.tables)) {
5410 /* We could tag this OpenFlow table but it would make the logic a
5411 * little harder and it's a corner case that doesn't seem worth it
5417 HMAP_FOR_EACH (t, hmap_node, &oftable->cls.tables) {
5418 if (cls_table_is_catchall(t)) {
5420 } else if (!other) {
5423 /* Indicate that we can't tag this by setting both tables to
5424 * NULL. (We know that 'catchall' is already NULL.) */
5431 /* Can't tag this table. */
5435 if (table->catchall_table != catchall || table->other_table != other) {
5436 table->catchall_table = catchall;
5437 table->other_table = other;
5438 ofproto->backer->need_revalidate = REV_FLOW_TABLE;
5442 /* Given 'rule' that has changed in some way (either it is a rule being
5443 * inserted, a rule being deleted, or a rule whose actions are being
5444 * modified), marks facets for revalidation to ensure that packets will be
5445 * forwarded correctly according to the new state of the flow table.
5447 * This function must be called after *each* change to a flow table. See
5448 * the comment on table_update_taggable() for more information. */
5450 rule_invalidate(const struct rule_dpif *rule)
5452 struct ofproto_dpif *ofproto = ofproto_dpif_cast(rule->up.ofproto);
5454 table_update_taggable(ofproto, rule->up.table_id);
5456 if (!ofproto->backer->need_revalidate) {
5457 struct table_dpif *table = &ofproto->tables[rule->up.table_id];
5459 if (table->other_table && rule->tag) {
5460 tag_set_add(&ofproto->backer->revalidate_set, rule->tag);
5462 ofproto->backer->need_revalidate = REV_FLOW_TABLE;
5468 set_frag_handling(struct ofproto *ofproto_,
5469 enum ofp_config_flags frag_handling)
5471 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
5472 if (frag_handling != OFPC_FRAG_REASM) {
5473 ofproto->backer->need_revalidate = REV_RECONFIGURE;
5481 packet_out(struct ofproto *ofproto_, struct ofpbuf *packet,
5482 const struct flow *flow,
5483 const struct ofpact *ofpacts, size_t ofpacts_len)
5485 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
5486 struct odputil_keybuf keybuf;
5487 struct dpif_flow_stats stats;
5488 struct xlate_out xout;
5489 struct xlate_in xin;
5493 ofpbuf_use_stack(&key, &keybuf, sizeof keybuf);
5494 odp_flow_key_from_flow(&key, flow,
5495 ofp_port_to_odp_port(ofproto, flow->in_port));
5497 dpif_flow_stats_extract(flow, packet, time_msec(), &stats);
5499 xlate_in_init(&xin, ofproto, flow, NULL, stats.tcp_flags, packet);
5500 xin.resubmit_stats = &stats;
5501 xin.ofpacts_len = ofpacts_len;
5502 xin.ofpacts = ofpacts;
5504 xlate_actions(&xin, &xout);
5505 dpif_execute(ofproto->backer->dpif, key.data, key.size,
5506 xout.odp_actions.data, xout.odp_actions.size, packet);
5507 xlate_out_uninit(&xout);
5515 set_netflow(struct ofproto *ofproto_,
5516 const struct netflow_options *netflow_options)
5518 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
5520 if (netflow_options) {
5521 if (!ofproto->netflow) {
5522 ofproto->netflow = netflow_create();
5524 return netflow_set_options(ofproto->netflow, netflow_options);
5526 netflow_destroy(ofproto->netflow);
5527 ofproto->netflow = NULL;
5533 get_netflow_ids(const struct ofproto *ofproto_,
5534 uint8_t *engine_type, uint8_t *engine_id)
5536 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
5538 dpif_get_netflow_ids(ofproto->backer->dpif, engine_type, engine_id);
5542 send_active_timeout(struct ofproto_dpif *ofproto, struct facet *facet)
5544 if (!facet_is_controller_flow(facet) &&
5545 netflow_active_timeout_expired(ofproto->netflow, &facet->nf_flow)) {
5546 struct subfacet *subfacet;
5547 struct ofexpired expired;
5549 LIST_FOR_EACH (subfacet, list_node, &facet->subfacets) {
5550 if (subfacet->path == SF_FAST_PATH) {
5551 struct dpif_flow_stats stats;
5553 subfacet_install(subfacet, &facet->xout.odp_actions,
5555 subfacet_update_stats(subfacet, &stats);
5559 expired.flow = facet->flow;
5560 expired.packet_count = facet->packet_count;
5561 expired.byte_count = facet->byte_count;
5562 expired.used = facet->used;
5563 netflow_expire(ofproto->netflow, &facet->nf_flow, &expired);
5568 send_netflow_active_timeouts(struct ofproto_dpif *ofproto)
5570 struct cls_cursor cursor;
5571 struct facet *facet;
5573 cls_cursor_init(&cursor, &ofproto->facets, NULL);
5574 CLS_CURSOR_FOR_EACH (facet, cr, &cursor) {
5575 send_active_timeout(ofproto, facet);
5579 static struct ofproto_dpif *
5580 ofproto_dpif_lookup(const char *name)
5582 struct ofproto_dpif *ofproto;
5584 HMAP_FOR_EACH_WITH_HASH (ofproto, all_ofproto_dpifs_node,
5585 hash_string(name, 0), &all_ofproto_dpifs) {
5586 if (!strcmp(ofproto->up.name, name)) {
5594 ofproto_unixctl_fdb_flush(struct unixctl_conn *conn, int argc,
5595 const char *argv[], void *aux OVS_UNUSED)
5597 struct ofproto_dpif *ofproto;
5600 ofproto = ofproto_dpif_lookup(argv[1]);
5602 unixctl_command_reply_error(conn, "no such bridge");
5605 mac_learning_flush(ofproto->ml, &ofproto->backer->revalidate_set);
5607 HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
5608 mac_learning_flush(ofproto->ml, &ofproto->backer->revalidate_set);
5612 unixctl_command_reply(conn, "table successfully flushed");
5616 ofproto_unixctl_fdb_show(struct unixctl_conn *conn, int argc OVS_UNUSED,
5617 const char *argv[], void *aux OVS_UNUSED)
5619 struct ds ds = DS_EMPTY_INITIALIZER;
5620 const struct ofproto_dpif *ofproto;
5621 const struct mac_entry *e;
5623 ofproto = ofproto_dpif_lookup(argv[1]);
5625 unixctl_command_reply_error(conn, "no such bridge");
5629 ds_put_cstr(&ds, " port VLAN MAC Age\n");
5630 LIST_FOR_EACH (e, lru_node, &ofproto->ml->lrus) {
5631 struct ofbundle *bundle = e->port.p;
5632 ds_put_format(&ds, "%5d %4d "ETH_ADDR_FMT" %3d\n",
5633 ofbundle_get_a_port(bundle)->odp_port,
5634 e->vlan, ETH_ADDR_ARGS(e->mac),
5635 mac_entry_age(ofproto->ml, e));
5637 unixctl_command_reply(conn, ds_cstr(&ds));
5642 struct xlate_out xout;
5643 struct xlate_in xin;
5649 trace_format_rule(struct ds *result, int level, const struct rule_dpif *rule)
5651 ds_put_char_multiple(result, '\t', level);
5653 ds_put_cstr(result, "No match\n");
5657 ds_put_format(result, "Rule: table=%"PRIu8" cookie=%#"PRIx64" ",
5658 rule ? rule->up.table_id : 0, ntohll(rule->up.flow_cookie));
5659 cls_rule_format(&rule->up.cr, result);
5660 ds_put_char(result, '\n');
5662 ds_put_char_multiple(result, '\t', level);
5663 ds_put_cstr(result, "OpenFlow ");
5664 ofpacts_format(rule->up.ofpacts, rule->up.ofpacts_len, result);
5665 ds_put_char(result, '\n');
5669 trace_format_flow(struct ds *result, int level, const char *title,
5670 struct trace_ctx *trace)
5672 ds_put_char_multiple(result, '\t', level);
5673 ds_put_format(result, "%s: ", title);
5674 if (flow_equal(&trace->xin.flow, &trace->flow)) {
5675 ds_put_cstr(result, "unchanged");
5677 flow_format(result, &trace->xin.flow);
5678 trace->flow = trace->xin.flow;
5680 ds_put_char(result, '\n');
5684 trace_format_regs(struct ds *result, int level, const char *title,
5685 struct trace_ctx *trace)
5689 ds_put_char_multiple(result, '\t', level);
5690 ds_put_format(result, "%s:", title);
5691 for (i = 0; i < FLOW_N_REGS; i++) {
5692 ds_put_format(result, " reg%zu=0x%"PRIx32, i, trace->flow.regs[i]);
5694 ds_put_char(result, '\n');
5698 trace_format_odp(struct ds *result, int level, const char *title,
5699 struct trace_ctx *trace)
5701 struct ofpbuf *odp_actions = &trace->xout.odp_actions;
5703 ds_put_char_multiple(result, '\t', level);
5704 ds_put_format(result, "%s: ", title);
5705 format_odp_actions(result, odp_actions->data, odp_actions->size);
5706 ds_put_char(result, '\n');
5710 trace_resubmit(struct xlate_in *xin, struct rule_dpif *rule, int recurse)
5712 struct trace_ctx *trace = CONTAINER_OF(xin, struct trace_ctx, xin);
5713 struct ds *result = trace->result;
5715 ds_put_char(result, '\n');
5716 trace_format_flow(result, recurse + 1, "Resubmitted flow", trace);
5717 trace_format_regs(result, recurse + 1, "Resubmitted regs", trace);
5718 trace_format_odp(result, recurse + 1, "Resubmitted odp", trace);
5719 trace_format_rule(result, recurse + 1, rule);
5723 trace_report(struct xlate_in *xin, const char *s, int recurse)
5725 struct trace_ctx *trace = CONTAINER_OF(xin, struct trace_ctx, xin);
5726 struct ds *result = trace->result;
5728 ds_put_char_multiple(result, '\t', recurse);
5729 ds_put_cstr(result, s);
5730 ds_put_char(result, '\n');
5734 ofproto_unixctl_trace(struct unixctl_conn *conn, int argc, const char *argv[],
5735 void *aux OVS_UNUSED)
5737 const struct dpif_backer *backer;
5738 struct ofproto_dpif *ofproto;
5739 struct ofpbuf odp_key;
5740 struct ofpbuf *packet;
5748 ofpbuf_init(&odp_key, 0);
5750 /* Handle "-generate" or a hex string as the last argument. */
5751 if (!strcmp(argv[argc - 1], "-generate")) {
5752 packet = ofpbuf_new(0);
5755 const char *error = eth_from_hex(argv[argc - 1], &packet);
5758 } else if (argc == 4) {
5759 /* The 3-argument form must end in "-generate' or a hex string. */
5760 unixctl_command_reply_error(conn, error);
5765 /* Parse the flow and determine whether a datapath or
5766 * bridge is specified. If function odp_flow_key_from_string()
5767 * returns 0, the flow is a odp_flow. If function
5768 * parse_ofp_exact_flow() returns 0, the flow is a br_flow. */
5769 if (!odp_flow_key_from_string(argv[argc - 1], NULL, &odp_key)) {
5770 /* If the odp_flow is the second argument,
5771 * the datapath name is the first argument. */
5773 const char *dp_type;
5774 if (!strncmp(argv[1], "ovs-", 4)) {
5775 dp_type = argv[1] + 4;
5779 backer = shash_find_data(&all_dpif_backers, dp_type);
5781 unixctl_command_reply_error(conn, "Cannot find datapath "
5786 /* No datapath name specified, so there should be only one
5788 struct shash_node *node;
5789 if (shash_count(&all_dpif_backers) != 1) {
5790 unixctl_command_reply_error(conn, "Must specify datapath "
5791 "name, there is more than one type of datapath");
5794 node = shash_first(&all_dpif_backers);
5795 backer = node->data;
5798 /* Extract the ofproto_dpif object from the ofproto_receive()
5800 if (ofproto_receive(backer, NULL, odp_key.data,
5801 odp_key.size, &flow, NULL, &ofproto, NULL)) {
5802 unixctl_command_reply_error(conn, "Invalid datapath flow");
5805 ds_put_format(&result, "Bridge: %s\n", ofproto->up.name);
5806 } else if (!parse_ofp_exact_flow(&flow, argv[argc - 1])) {
5808 unixctl_command_reply_error(conn, "Must specify bridge name");
5812 ofproto = ofproto_dpif_lookup(argv[1]);
5814 unixctl_command_reply_error(conn, "Unknown bridge name");
5818 unixctl_command_reply_error(conn, "Bad flow syntax");
5822 /* Generate a packet, if requested. */
5824 if (!packet->size) {
5825 flow_compose(packet, &flow);
5827 ds_put_cstr(&result, "Packet: ");
5828 s = ofp_packet_to_string(packet->data, packet->size);
5829 ds_put_cstr(&result, s);
5832 /* Use the metadata from the flow and the packet argument
5833 * to reconstruct the flow. */
5834 flow_extract(packet, flow.skb_priority, flow.skb_mark, NULL,
5835 flow.in_port, &flow);
5839 ofproto_trace(ofproto, &flow, packet, &result);
5840 unixctl_command_reply(conn, ds_cstr(&result));
5843 ds_destroy(&result);
5844 ofpbuf_delete(packet);
5845 ofpbuf_uninit(&odp_key);
5849 ofproto_trace(struct ofproto_dpif *ofproto, const struct flow *flow,
5850 const struct ofpbuf *packet, struct ds *ds)
5852 struct rule_dpif *rule;
5854 ds_put_cstr(ds, "Flow: ");
5855 flow_format(ds, flow);
5856 ds_put_char(ds, '\n');
5858 rule = rule_dpif_lookup(ofproto, flow, NULL);
5860 trace_format_rule(ds, 0, rule);
5861 if (rule == ofproto->miss_rule) {
5862 ds_put_cstr(ds, "\nNo match, flow generates \"packet in\"s.\n");
5863 } else if (rule == ofproto->no_packet_in_rule) {
5864 ds_put_cstr(ds, "\nNo match, packets dropped because "
5865 "OFPPC_NO_PACKET_IN is set on in_port.\n");
5866 } else if (rule == ofproto->drop_frags_rule) {
5867 ds_put_cstr(ds, "\nPackets dropped because they are IP fragments "
5868 "and the fragment handling mode is \"drop\".\n");
5872 uint64_t odp_actions_stub[1024 / 8];
5873 struct ofpbuf odp_actions;
5874 struct trace_ctx trace;
5878 tcp_flags = packet ? packet_get_tcp_flags(packet, flow) : 0;
5881 ofpbuf_use_stub(&odp_actions,
5882 odp_actions_stub, sizeof odp_actions_stub);
5883 xlate_in_init(&trace.xin, ofproto, flow, rule, tcp_flags, packet);
5884 trace.xin.resubmit_hook = trace_resubmit;
5885 trace.xin.report_hook = trace_report;
5887 xlate_actions(&trace.xin, &trace.xout);
5889 ds_put_char(ds, '\n');
5890 trace_format_flow(ds, 0, "Final flow", &trace);
5892 match_init(&match, flow, &trace.xout.wc);
5893 ds_put_cstr(ds, "Relevant fields: ");
5894 match_format(&match, ds, OFP_DEFAULT_PRIORITY);
5895 ds_put_char(ds, '\n');
5897 ds_put_cstr(ds, "Datapath actions: ");
5898 format_odp_actions(ds, trace.xout.odp_actions.data,
5899 trace.xout.odp_actions.size);
5901 if (trace.xout.slow) {
5902 ds_put_cstr(ds, "\nThis flow is handled by the userspace "
5903 "slow path because it:");
5904 switch (trace.xout.slow) {
5906 ds_put_cstr(ds, "\n\t- Consists of CFM packets.");
5909 ds_put_cstr(ds, "\n\t- Consists of LACP packets.");
5912 ds_put_cstr(ds, "\n\t- Consists of STP packets.");
5915 ds_put_cstr(ds, "\n\t- Consists of BFD packets.");
5917 case SLOW_CONTROLLER:
5918 ds_put_cstr(ds, "\n\t- Sends \"packet-in\" messages "
5919 "to the OpenFlow controller.");
5926 xlate_out_uninit(&trace.xout);
5931 ofproto_dpif_clog(struct unixctl_conn *conn OVS_UNUSED, int argc OVS_UNUSED,
5932 const char *argv[] OVS_UNUSED, void *aux OVS_UNUSED)
5935 unixctl_command_reply(conn, NULL);
5939 ofproto_dpif_unclog(struct unixctl_conn *conn OVS_UNUSED, int argc OVS_UNUSED,
5940 const char *argv[] OVS_UNUSED, void *aux OVS_UNUSED)
5943 unixctl_command_reply(conn, NULL);
5946 /* Runs a self-check of flow translations in 'ofproto'. Appends a message to
5947 * 'reply' describing the results. */
5949 ofproto_dpif_self_check__(struct ofproto_dpif *ofproto, struct ds *reply)
5951 struct cls_cursor cursor;
5952 struct facet *facet;
5956 cls_cursor_init(&cursor, &ofproto->facets, NULL);
5957 CLS_CURSOR_FOR_EACH (facet, cr, &cursor) {
5958 if (!facet_check_consistency(facet)) {
5963 ofproto->backer->need_revalidate = REV_INCONSISTENCY;
5967 ds_put_format(reply, "%s: self-check failed (%d errors)\n",
5968 ofproto->up.name, errors);
5970 ds_put_format(reply, "%s: self-check passed\n", ofproto->up.name);
5975 ofproto_dpif_self_check(struct unixctl_conn *conn,
5976 int argc, const char *argv[], void *aux OVS_UNUSED)
5978 struct ds reply = DS_EMPTY_INITIALIZER;
5979 struct ofproto_dpif *ofproto;
5982 ofproto = ofproto_dpif_lookup(argv[1]);
5984 unixctl_command_reply_error(conn, "Unknown ofproto (use "
5985 "ofproto/list for help)");
5988 ofproto_dpif_self_check__(ofproto, &reply);
5990 HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
5991 ofproto_dpif_self_check__(ofproto, &reply);
5995 unixctl_command_reply(conn, ds_cstr(&reply));
5999 /* Store the current ofprotos in 'ofproto_shash'. Returns a sorted list
6000 * of the 'ofproto_shash' nodes. It is the responsibility of the caller
6001 * to destroy 'ofproto_shash' and free the returned value. */
6002 static const struct shash_node **
6003 get_ofprotos(struct shash *ofproto_shash)
6005 const struct ofproto_dpif *ofproto;
6007 HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
6008 char *name = xasprintf("%s@%s", ofproto->up.type, ofproto->up.name);
6009 shash_add_nocopy(ofproto_shash, name, ofproto);
6012 return shash_sort(ofproto_shash);
6016 ofproto_unixctl_dpif_dump_dps(struct unixctl_conn *conn, int argc OVS_UNUSED,
6017 const char *argv[] OVS_UNUSED,
6018 void *aux OVS_UNUSED)
6020 struct ds ds = DS_EMPTY_INITIALIZER;
6021 struct shash ofproto_shash;
6022 const struct shash_node **sorted_ofprotos;
6025 shash_init(&ofproto_shash);
6026 sorted_ofprotos = get_ofprotos(&ofproto_shash);
6027 for (i = 0; i < shash_count(&ofproto_shash); i++) {
6028 const struct shash_node *node = sorted_ofprotos[i];
6029 ds_put_format(&ds, "%s\n", node->name);
6032 shash_destroy(&ofproto_shash);
6033 free(sorted_ofprotos);
6035 unixctl_command_reply(conn, ds_cstr(&ds));
6040 show_dp_rates(struct ds *ds, const char *heading,
6041 const struct avg_subfacet_rates *rates)
6043 ds_put_format(ds, "%s add rate: %5.3f/min, del rate: %5.3f/min\n",
6044 heading, rates->add_rate, rates->del_rate);
6048 dpif_show_backer(const struct dpif_backer *backer, struct ds *ds)
6050 const struct shash_node **ofprotos;
6051 struct ofproto_dpif *ofproto;
6052 struct shash ofproto_shash;
6053 uint64_t n_hit, n_missed;
6054 long long int minutes;
6057 n_hit = n_missed = 0;
6058 HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
6059 if (ofproto->backer == backer) {
6060 n_missed += ofproto->n_missed;
6061 n_hit += ofproto->n_hit;
6065 ds_put_format(ds, "%s: hit:%"PRIu64" missed:%"PRIu64"\n",
6066 dpif_name(backer->dpif), n_hit, n_missed);
6067 ds_put_format(ds, "\tflows: cur: %zu, avg: %u, max: %u,"
6068 " life span: %lldms\n", hmap_count(&backer->subfacets),
6069 backer->avg_n_subfacet, backer->max_n_subfacet,
6070 backer->avg_subfacet_life);
6072 minutes = (time_msec() - backer->created) / (1000 * 60);
6073 if (minutes >= 60) {
6074 show_dp_rates(ds, "\thourly avg:", &backer->hourly);
6076 if (minutes >= 60 * 24) {
6077 show_dp_rates(ds, "\tdaily avg:", &backer->daily);
6079 show_dp_rates(ds, "\toverall avg:", &backer->lifetime);
6081 shash_init(&ofproto_shash);
6082 ofprotos = get_ofprotos(&ofproto_shash);
6083 for (i = 0; i < shash_count(&ofproto_shash); i++) {
6084 struct ofproto_dpif *ofproto = ofprotos[i]->data;
6085 const struct shash_node **ports;
6088 if (ofproto->backer != backer) {
6092 ds_put_format(ds, "\t%s: hit:%"PRIu64" missed:%"PRIu64"\n",
6093 ofproto->up.name, ofproto->n_hit, ofproto->n_missed);
6095 ports = shash_sort(&ofproto->up.port_by_name);
6096 for (j = 0; j < shash_count(&ofproto->up.port_by_name); j++) {
6097 const struct shash_node *node = ports[j];
6098 struct ofport *ofport = node->data;
6102 ds_put_format(ds, "\t\t%s %u/", netdev_get_name(ofport->netdev),
6105 odp_port = ofp_port_to_odp_port(ofproto, ofport->ofp_port);
6106 if (odp_port != OVSP_NONE) {
6107 ds_put_format(ds, "%"PRIu32":", odp_port);
6109 ds_put_cstr(ds, "none:");
6112 ds_put_format(ds, " (%s", netdev_get_type(ofport->netdev));
6115 if (!netdev_get_config(ofport->netdev, &config)) {
6116 const struct smap_node **nodes;
6119 nodes = smap_sort(&config);
6120 for (i = 0; i < smap_count(&config); i++) {
6121 const struct smap_node *node = nodes[i];
6122 ds_put_format(ds, "%c %s=%s", i ? ',' : ':',
6123 node->key, node->value);
6127 smap_destroy(&config);
6129 ds_put_char(ds, ')');
6130 ds_put_char(ds, '\n');
6134 shash_destroy(&ofproto_shash);
6139 ofproto_unixctl_dpif_show(struct unixctl_conn *conn, int argc OVS_UNUSED,
6140 const char *argv[] OVS_UNUSED, void *aux OVS_UNUSED)
6142 struct ds ds = DS_EMPTY_INITIALIZER;
6143 const struct shash_node **backers;
6146 backers = shash_sort(&all_dpif_backers);
6147 for (i = 0; i < shash_count(&all_dpif_backers); i++) {
6148 dpif_show_backer(backers[i]->data, &ds);
6152 unixctl_command_reply(conn, ds_cstr(&ds));
6156 /* Dump the megaflow (facet) cache. This is useful to check the
6157 * correctness of flow wildcarding, since the same mechanism is used for
6158 * both xlate caching and kernel wildcarding.
6160 * It's important to note that in the output the flow description uses
6161 * OpenFlow (OFP) ports, but the actions use datapath (ODP) ports.
6163 * This command is only needed for advanced debugging, so it's not
6164 * documented in the man page. */
6166 ofproto_unixctl_dpif_dump_megaflows(struct unixctl_conn *conn,
6167 int argc OVS_UNUSED, const char *argv[],
6168 void *aux OVS_UNUSED)
6170 struct ds ds = DS_EMPTY_INITIALIZER;
6171 const struct ofproto_dpif *ofproto;
6172 long long int now = time_msec();
6173 struct cls_cursor cursor;
6174 struct facet *facet;
6176 ofproto = ofproto_dpif_lookup(argv[1]);
6178 unixctl_command_reply_error(conn, "no such bridge");
6182 cls_cursor_init(&cursor, &ofproto->facets, NULL);
6183 CLS_CURSOR_FOR_EACH (facet, cr, &cursor) {
6184 cls_rule_format(&facet->cr, &ds);
6185 ds_put_cstr(&ds, ", ");
6186 ds_put_format(&ds, "n_subfacets:%zu, ", list_size(&facet->subfacets));
6187 ds_put_format(&ds, "used:%.3fs, ", (now - facet->used) / 1000.0);
6188 ds_put_cstr(&ds, "Datapath actions: ");
6189 if (facet->xout.slow) {
6190 uint64_t slow_path_stub[128 / 8];
6191 const struct nlattr *actions;
6194 compose_slow_path(ofproto, &facet->flow, facet->xout.slow,
6195 slow_path_stub, sizeof slow_path_stub,
6196 &actions, &actions_len);
6197 format_odp_actions(&ds, actions, actions_len);
6199 format_odp_actions(&ds, facet->xout.odp_actions.data,
6200 facet->xout.odp_actions.size);
6202 ds_put_cstr(&ds, "\n");
6205 ds_chomp(&ds, '\n');
6206 unixctl_command_reply(conn, ds_cstr(&ds));
6211 ofproto_unixctl_dpif_dump_flows(struct unixctl_conn *conn,
6212 int argc OVS_UNUSED, const char *argv[],
6213 void *aux OVS_UNUSED)
6215 struct ds ds = DS_EMPTY_INITIALIZER;
6216 const struct ofproto_dpif *ofproto;
6217 struct subfacet *subfacet;
6219 ofproto = ofproto_dpif_lookup(argv[1]);
6221 unixctl_command_reply_error(conn, "no such bridge");
6225 update_stats(ofproto->backer);
6227 HMAP_FOR_EACH (subfacet, hmap_node, &ofproto->backer->subfacets) {
6228 struct facet *facet = subfacet->facet;
6230 if (ofproto_dpif_cast(facet->rule->up.ofproto) != ofproto) {
6234 odp_flow_key_format(subfacet->key, subfacet->key_len, &ds);
6236 ds_put_format(&ds, ", packets:%"PRIu64", bytes:%"PRIu64", used:",
6237 subfacet->dp_packet_count, subfacet->dp_byte_count);
6238 if (subfacet->used) {
6239 ds_put_format(&ds, "%.3fs",
6240 (time_msec() - subfacet->used) / 1000.0);
6242 ds_put_format(&ds, "never");
6244 if (subfacet->facet->tcp_flags) {
6245 ds_put_cstr(&ds, ", flags:");
6246 packet_format_tcp_flags(&ds, subfacet->facet->tcp_flags);
6249 ds_put_cstr(&ds, ", actions:");
6250 if (facet->xout.slow) {
6251 uint64_t slow_path_stub[128 / 8];
6252 const struct nlattr *actions;
6255 compose_slow_path(ofproto, &facet->flow, facet->xout.slow,
6256 slow_path_stub, sizeof slow_path_stub,
6257 &actions, &actions_len);
6258 format_odp_actions(&ds, actions, actions_len);
6260 format_odp_actions(&ds, facet->xout.odp_actions.data,
6261 facet->xout.odp_actions.size);
6263 ds_put_char(&ds, '\n');
6266 unixctl_command_reply(conn, ds_cstr(&ds));
6271 ofproto_unixctl_dpif_del_flows(struct unixctl_conn *conn,
6272 int argc OVS_UNUSED, const char *argv[],
6273 void *aux OVS_UNUSED)
6275 struct ds ds = DS_EMPTY_INITIALIZER;
6276 struct ofproto_dpif *ofproto;
6278 ofproto = ofproto_dpif_lookup(argv[1]);
6280 unixctl_command_reply_error(conn, "no such bridge");
6284 flush(&ofproto->up);
6286 unixctl_command_reply(conn, ds_cstr(&ds));
6291 ofproto_dpif_unixctl_init(void)
6293 static bool registered;
6299 unixctl_command_register(
6301 "[dp_name]|bridge odp_flow|br_flow [-generate|packet]",
6302 1, 3, ofproto_unixctl_trace, NULL);
6303 unixctl_command_register("fdb/flush", "[bridge]", 0, 1,
6304 ofproto_unixctl_fdb_flush, NULL);
6305 unixctl_command_register("fdb/show", "bridge", 1, 1,
6306 ofproto_unixctl_fdb_show, NULL);
6307 unixctl_command_register("ofproto/clog", "", 0, 0,
6308 ofproto_dpif_clog, NULL);
6309 unixctl_command_register("ofproto/unclog", "", 0, 0,
6310 ofproto_dpif_unclog, NULL);
6311 unixctl_command_register("ofproto/self-check", "[bridge]", 0, 1,
6312 ofproto_dpif_self_check, NULL);
6313 unixctl_command_register("dpif/dump-dps", "", 0, 0,
6314 ofproto_unixctl_dpif_dump_dps, NULL);
6315 unixctl_command_register("dpif/show", "", 0, 0, ofproto_unixctl_dpif_show,
6317 unixctl_command_register("dpif/dump-flows", "bridge", 1, 1,
6318 ofproto_unixctl_dpif_dump_flows, NULL);
6319 unixctl_command_register("dpif/del-flows", "bridge", 1, 1,
6320 ofproto_unixctl_dpif_del_flows, NULL);
6321 unixctl_command_register("dpif/dump-megaflows", "bridge", 1, 1,
6322 ofproto_unixctl_dpif_dump_megaflows, NULL);
6325 /* Linux VLAN device support (e.g. "eth0.10" for VLAN 10.)
6327 * This is deprecated. It is only for compatibility with broken device drivers
6328 * in old versions of Linux that do not properly support VLANs when VLAN
6329 * devices are not used. When broken device drivers are no longer in
6330 * widespread use, we will delete these interfaces. */
6333 set_realdev(struct ofport *ofport_, uint16_t realdev_ofp_port, int vid)
6335 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport_->ofproto);
6336 struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
6338 if (realdev_ofp_port == ofport->realdev_ofp_port
6339 && vid == ofport->vlandev_vid) {
6343 ofproto->backer->need_revalidate = REV_RECONFIGURE;
6345 if (ofport->realdev_ofp_port) {
6348 if (realdev_ofp_port && ofport->bundle) {
6349 /* vlandevs are enslaved to their realdevs, so they are not allowed to
6350 * themselves be part of a bundle. */
6351 bundle_set(ofport->up.ofproto, ofport->bundle, NULL);
6354 ofport->realdev_ofp_port = realdev_ofp_port;
6355 ofport->vlandev_vid = vid;
6357 if (realdev_ofp_port) {
6358 vsp_add(ofport, realdev_ofp_port, vid);
6365 hash_realdev_vid(uint16_t realdev_ofp_port, int vid)
6367 return hash_2words(realdev_ofp_port, vid);
6370 /* Returns the OFP port number of the Linux VLAN device that corresponds to
6371 * 'vlan_tci' on the network device with port number 'realdev_ofp_port' in
6372 * 'struct ofport_dpif'. For example, given 'realdev_ofp_port' of eth0 and
6373 * 'vlan_tci' 9, it would return the port number of eth0.9.
6375 * Unless VLAN splinters are enabled for port 'realdev_ofp_port', this
6376 * function just returns its 'realdev_ofp_port' argument. */
6378 vsp_realdev_to_vlandev(const struct ofproto_dpif *ofproto,
6379 uint16_t realdev_ofp_port, ovs_be16 vlan_tci)
6381 if (!hmap_is_empty(&ofproto->realdev_vid_map)) {
6382 int vid = vlan_tci_to_vid(vlan_tci);
6383 const struct vlan_splinter *vsp;
6385 HMAP_FOR_EACH_WITH_HASH (vsp, realdev_vid_node,
6386 hash_realdev_vid(realdev_ofp_port, vid),
6387 &ofproto->realdev_vid_map) {
6388 if (vsp->realdev_ofp_port == realdev_ofp_port
6389 && vsp->vid == vid) {
6390 return vsp->vlandev_ofp_port;
6394 return realdev_ofp_port;
6397 static struct vlan_splinter *
6398 vlandev_find(const struct ofproto_dpif *ofproto, uint16_t vlandev_ofp_port)
6400 struct vlan_splinter *vsp;
6402 HMAP_FOR_EACH_WITH_HASH (vsp, vlandev_node, hash_int(vlandev_ofp_port, 0),
6403 &ofproto->vlandev_map) {
6404 if (vsp->vlandev_ofp_port == vlandev_ofp_port) {
6412 /* Returns the OpenFlow port number of the "real" device underlying the Linux
6413 * VLAN device with OpenFlow port number 'vlandev_ofp_port' and stores the
6414 * VLAN VID of the Linux VLAN device in '*vid'. For example, given
6415 * 'vlandev_ofp_port' of eth0.9, it would return the OpenFlow port number of
6416 * eth0 and store 9 in '*vid'.
6418 * Returns 0 and does not modify '*vid' if 'vlandev_ofp_port' is not a Linux
6419 * VLAN device. Unless VLAN splinters are enabled, this is what this function
6422 vsp_vlandev_to_realdev(const struct ofproto_dpif *ofproto,
6423 uint16_t vlandev_ofp_port, int *vid)
6425 if (!hmap_is_empty(&ofproto->vlandev_map)) {
6426 const struct vlan_splinter *vsp;
6428 vsp = vlandev_find(ofproto, vlandev_ofp_port);
6433 return vsp->realdev_ofp_port;
6439 /* Given 'flow', a flow representing a packet received on 'ofproto', checks
6440 * whether 'flow->in_port' represents a Linux VLAN device. If so, changes
6441 * 'flow->in_port' to the "real" device backing the VLAN device, sets
6442 * 'flow->vlan_tci' to the VLAN VID, and returns true. Otherwise (which is
6443 * always the case unless VLAN splinters are enabled), returns false without
6444 * making any changes. */
6446 vsp_adjust_flow(const struct ofproto_dpif *ofproto, struct flow *flow)
6451 realdev = vsp_vlandev_to_realdev(ofproto, flow->in_port, &vid);
6456 /* Cause the flow to be processed as if it came in on the real device with
6457 * the VLAN device's VLAN ID. */
6458 flow->in_port = realdev;
6459 flow->vlan_tci = htons((vid & VLAN_VID_MASK) | VLAN_CFI);
6464 vsp_remove(struct ofport_dpif *port)
6466 struct ofproto_dpif *ofproto = ofproto_dpif_cast(port->up.ofproto);
6467 struct vlan_splinter *vsp;
6469 vsp = vlandev_find(ofproto, port->up.ofp_port);
6471 hmap_remove(&ofproto->vlandev_map, &vsp->vlandev_node);
6472 hmap_remove(&ofproto->realdev_vid_map, &vsp->realdev_vid_node);
6475 port->realdev_ofp_port = 0;
6477 VLOG_ERR("missing vlan device record");
6482 vsp_add(struct ofport_dpif *port, uint16_t realdev_ofp_port, int vid)
6484 struct ofproto_dpif *ofproto = ofproto_dpif_cast(port->up.ofproto);
6486 if (!vsp_vlandev_to_realdev(ofproto, port->up.ofp_port, NULL)
6487 && (vsp_realdev_to_vlandev(ofproto, realdev_ofp_port, htons(vid))
6488 == realdev_ofp_port)) {
6489 struct vlan_splinter *vsp;
6491 vsp = xmalloc(sizeof *vsp);
6492 hmap_insert(&ofproto->vlandev_map, &vsp->vlandev_node,
6493 hash_int(port->up.ofp_port, 0));
6494 hmap_insert(&ofproto->realdev_vid_map, &vsp->realdev_vid_node,
6495 hash_realdev_vid(realdev_ofp_port, vid));
6496 vsp->realdev_ofp_port = realdev_ofp_port;
6497 vsp->vlandev_ofp_port = port->up.ofp_port;
6500 port->realdev_ofp_port = realdev_ofp_port;
6502 VLOG_ERR("duplicate vlan device record");
6507 ofp_port_to_odp_port(const struct ofproto_dpif *ofproto, uint16_t ofp_port)
6509 const struct ofport_dpif *ofport = get_ofp_port(ofproto, ofp_port);
6510 return ofport ? ofport->odp_port : OVSP_NONE;
6513 static struct ofport_dpif *
6514 odp_port_to_ofport(const struct dpif_backer *backer, uint32_t odp_port)
6516 struct ofport_dpif *port;
6518 HMAP_FOR_EACH_IN_BUCKET (port, odp_port_node,
6519 hash_int(odp_port, 0),
6520 &backer->odp_to_ofport_map) {
6521 if (port->odp_port == odp_port) {
6530 odp_port_to_ofp_port(const struct ofproto_dpif *ofproto, uint32_t odp_port)
6532 struct ofport_dpif *port;
6534 port = odp_port_to_ofport(ofproto->backer, odp_port);
6535 if (port && &ofproto->up == port->up.ofproto) {
6536 return port->up.ofp_port;
6542 /* Compute exponentially weighted moving average, adding 'new' as the newest,
6543 * most heavily weighted element. 'base' designates the rate of decay: after
6544 * 'base' further updates, 'new''s weight in the EWMA decays to about 1/e
6547 exp_mavg(double *avg, int base, double new)
6549 *avg = (*avg * (base - 1) + new) / base;
6553 update_moving_averages(struct dpif_backer *backer)
6555 const int min_ms = 60 * 1000; /* milliseconds in one minute. */
6556 long long int minutes = (time_msec() - backer->created) / min_ms;
6559 backer->lifetime.add_rate = (double) backer->total_subfacet_add_count
6561 backer->lifetime.del_rate = (double) backer->total_subfacet_del_count
6564 backer->lifetime.add_rate = 0.0;
6565 backer->lifetime.del_rate = 0.0;
6568 /* Update hourly averages on the minute boundaries. */
6569 if (time_msec() - backer->last_minute >= min_ms) {
6570 exp_mavg(&backer->hourly.add_rate, 60, backer->subfacet_add_count);
6571 exp_mavg(&backer->hourly.del_rate, 60, backer->subfacet_del_count);
6573 /* Update daily averages on the hour boundaries. */
6574 if ((backer->last_minute - backer->created) / min_ms % 60 == 59) {
6575 exp_mavg(&backer->daily.add_rate, 24, backer->hourly.add_rate);
6576 exp_mavg(&backer->daily.del_rate, 24, backer->hourly.del_rate);
6579 backer->total_subfacet_add_count += backer->subfacet_add_count;
6580 backer->total_subfacet_del_count += backer->subfacet_del_count;
6581 backer->subfacet_add_count = 0;
6582 backer->subfacet_del_count = 0;
6583 backer->last_minute += min_ms;
6587 const struct ofproto_class ofproto_dpif_class = {
6622 port_is_lacp_current,
6623 NULL, /* rule_choose_table */
6630 rule_modify_actions,
6644 get_stp_port_status,
6651 is_mirror_output_bundle,
6652 forward_bpdu_changed,
6653 set_mac_table_config,