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);
72 COVERAGE_DEFINE(subfacet_install_fail);
77 static struct rule_dpif *rule_dpif_lookup(struct ofproto_dpif *,
79 struct flow_wildcards *wc);
81 static void rule_get_stats(struct rule *, uint64_t *packets, uint64_t *bytes);
82 static void rule_invalidate(const struct rule_dpif *);
84 static void mirror_destroy(struct ofmirror *);
85 static void update_mirror_stats(struct ofproto_dpif *ofproto,
86 mirror_mask_t mirrors,
87 uint64_t packets, uint64_t bytes);
89 static void bundle_remove(struct ofport *);
90 static void bundle_update(struct ofbundle *);
91 static void bundle_destroy(struct ofbundle *);
92 static void bundle_del_port(struct ofport_dpif *);
93 static void bundle_run(struct ofbundle *);
94 static void bundle_wait(struct ofbundle *);
96 static void stp_run(struct ofproto_dpif *ofproto);
97 static void stp_wait(struct ofproto_dpif *ofproto);
98 static int set_stp_port(struct ofport *,
99 const struct ofproto_port_stp_settings *);
101 static void compose_slow_path(const struct ofproto_dpif *, const struct flow *,
102 enum slow_path_reason,
103 uint64_t *stub, size_t stub_size,
104 const struct nlattr **actionsp,
105 size_t *actions_lenp);
107 /* A subfacet (see "struct subfacet" below) has three possible installation
110 * - SF_NOT_INSTALLED: Not installed in the datapath. This will only be the
111 * case just after the subfacet is created, just before the subfacet is
112 * destroyed, or if the datapath returns an error when we try to install a
115 * - SF_FAST_PATH: The subfacet's actions are installed in the datapath.
117 * - SF_SLOW_PATH: An action that sends every packet for the subfacet through
118 * ofproto_dpif is installed in the datapath.
121 SF_NOT_INSTALLED, /* No datapath flow for this subfacet. */
122 SF_FAST_PATH, /* Full actions are installed. */
123 SF_SLOW_PATH, /* Send-to-userspace action is installed. */
126 /* A dpif flow and actions associated with a facet.
128 * See also the large comment on struct facet. */
131 struct hmap_node hmap_node; /* In struct ofproto_dpif 'subfacets' list. */
132 struct list list_node; /* In struct facet's 'facets' list. */
133 struct facet *facet; /* Owning facet. */
134 struct dpif_backer *backer; /* Owning backer. */
136 enum odp_key_fitness key_fitness;
140 long long int used; /* Time last used; time created if not used. */
141 long long int created; /* Time created. */
143 uint64_t dp_packet_count; /* Last known packet count in the datapath. */
144 uint64_t dp_byte_count; /* Last known byte count in the datapath. */
146 enum subfacet_path path; /* Installed in datapath? */
149 #define SUBFACET_DESTROY_MAX_BATCH 50
151 static struct subfacet *subfacet_create(struct facet *, struct flow_miss *miss,
153 static struct subfacet *subfacet_find(struct dpif_backer *,
154 const struct nlattr *key, size_t key_len,
156 static void subfacet_destroy(struct subfacet *);
157 static void subfacet_destroy__(struct subfacet *);
158 static void subfacet_destroy_batch(struct dpif_backer *,
159 struct subfacet **, int n);
160 static void subfacet_reset_dp_stats(struct subfacet *,
161 struct dpif_flow_stats *);
162 static void subfacet_update_stats(struct subfacet *,
163 const struct dpif_flow_stats *);
164 static int subfacet_install(struct subfacet *,
165 const struct ofpbuf *odp_actions,
166 struct dpif_flow_stats *);
167 static void subfacet_uninstall(struct subfacet *);
169 /* A unique, non-overlapping instantiation of an OpenFlow flow.
171 * A facet associates a "struct flow", which represents the Open vSwitch
172 * userspace idea of an exact-match flow, with one or more subfacets.
173 * While the facet is created based on an exact-match flow, it is stored
174 * within the ofproto based on the wildcards that could be expressed
175 * based on the flow table and other configuration. (See the 'wc'
176 * description in "struct xlate_out" for more details.)
178 * Each subfacet tracks the datapath's idea of the flow equivalent to
179 * the facet. When the kernel module (or other dpif implementation) and
180 * Open vSwitch userspace agree on the definition of a flow key, there
181 * is exactly one subfacet per facet. If the dpif implementation
182 * supports more-specific flow matching than userspace, however, a facet
183 * can have more than one subfacet. Examples include the dpif
184 * implementation not supporting the same wildcards as userspace or some
185 * distinction in flow that userspace simply doesn't understand.
187 * Flow expiration works in terms of subfacets, so a facet must have at
188 * least one subfacet or it will never expire, leaking memory. */
191 struct hmap_node hmap_node; /* In owning ofproto's 'facets' hmap. */
192 struct list list_node; /* In owning rule's 'facets' list. */
193 struct rule_dpif *rule; /* Owning rule. */
196 struct list subfacets;
197 long long int used; /* Time last used; time created if not used. */
200 struct flow flow; /* Flow of the creating subfacet. */
201 struct cls_rule cr; /* In 'ofproto_dpif's facets classifier. */
205 * - Do include packets and bytes sent "by hand", e.g. with
208 * - Do include packets and bytes that were obtained from the datapath
209 * when a subfacet's statistics were reset (e.g. dpif_flow_put() with
210 * DPIF_FP_ZERO_STATS).
212 * - Do not include packets or bytes that can be obtained from the
213 * datapath for any existing subfacet.
215 uint64_t packet_count; /* Number of packets received. */
216 uint64_t byte_count; /* Number of bytes received. */
218 /* Resubmit statistics. */
219 uint64_t prev_packet_count; /* Number of packets from last stats push. */
220 uint64_t prev_byte_count; /* Number of bytes from last stats push. */
221 long long int prev_used; /* Used time from last stats push. */
224 uint64_t accounted_bytes; /* Bytes processed by facet_account(). */
225 struct netflow_flow nf_flow; /* Per-flow NetFlow tracking data. */
226 uint8_t tcp_flags; /* TCP flags seen for this 'rule'. */
228 struct xlate_out xout;
230 /* Storage for a single subfacet, to reduce malloc() time and space
231 * overhead. (A facet always has at least one subfacet and in the common
232 * case has exactly one subfacet. However, 'one_subfacet' may not
233 * always be valid, since it could have been removed after newer
234 * subfacets were pushed onto the 'subfacets' list.) */
235 struct subfacet one_subfacet;
237 long long int learn_rl; /* Rate limiter for facet_learn(). */
240 static struct facet *facet_create(const struct flow_miss *, struct rule_dpif *,
242 struct dpif_flow_stats *);
243 static void facet_remove(struct facet *);
244 static void facet_free(struct facet *);
246 static struct facet *facet_find(struct ofproto_dpif *, const struct flow *);
247 static struct facet *facet_lookup_valid(struct ofproto_dpif *,
248 const struct flow *);
249 static bool facet_revalidate(struct facet *);
250 static bool facet_check_consistency(struct facet *);
252 static void facet_flush_stats(struct facet *);
254 static void facet_reset_counters(struct facet *);
255 static void facet_push_stats(struct facet *, bool may_learn);
256 static void facet_learn(struct facet *);
257 static void facet_account(struct facet *);
258 static void push_all_stats(void);
260 static bool facet_is_controller_flow(struct facet *);
262 /* Node in 'ofport_dpif''s 'priorities' map. Used to maintain a map from
263 * 'priority' (the datapath's term for QoS queue) to the dscp bits which all
264 * traffic egressing the 'ofport' with that priority should be marked with. */
265 struct priority_to_dscp {
266 struct hmap_node hmap_node; /* Node in 'ofport_dpif''s 'priorities' map. */
267 uint32_t priority; /* Priority of this queue (see struct flow). */
269 uint8_t dscp; /* DSCP bits to mark outgoing traffic with. */
272 /* Linux VLAN device support (e.g. "eth0.10" for VLAN 10.)
274 * This is deprecated. It is only for compatibility with broken device drivers
275 * in old versions of Linux that do not properly support VLANs when VLAN
276 * devices are not used. When broken device drivers are no longer in
277 * widespread use, we will delete these interfaces. */
278 struct vlan_splinter {
279 struct hmap_node realdev_vid_node;
280 struct hmap_node vlandev_node;
281 ofp_port_t realdev_ofp_port;
282 ofp_port_t vlandev_ofp_port;
286 static bool vsp_adjust_flow(const struct ofproto_dpif *, struct flow *);
287 static void vsp_remove(struct ofport_dpif *);
288 static void vsp_add(struct ofport_dpif *, ofp_port_t realdev_ofp_port, int vid);
290 static ofp_port_t odp_port_to_ofp_port(const struct ofproto_dpif *,
291 odp_port_t odp_port);
293 static struct ofport_dpif *
294 ofport_dpif_cast(const struct ofport *ofport)
296 return ofport ? CONTAINER_OF(ofport, struct ofport_dpif, up) : NULL;
299 static void port_run(struct ofport_dpif *);
300 static void port_run_fast(struct ofport_dpif *);
301 static void port_wait(struct ofport_dpif *);
302 static int set_bfd(struct ofport *, const struct smap *);
303 static int set_cfm(struct ofport *, const struct cfm_settings *);
304 static void ofport_clear_priorities(struct ofport_dpif *);
305 static void ofport_update_peer(struct ofport_dpif *);
306 static void run_fast_rl(void);
308 struct dpif_completion {
309 struct list list_node;
310 struct ofoperation *op;
313 /* Reasons that we might need to revalidate every facet, and corresponding
316 * A value of 0 means that there is no need to revalidate.
318 * It would be nice to have some cleaner way to integrate with coverage
319 * counters, but with only a few reasons I guess this is good enough for
321 enum revalidate_reason {
322 REV_RECONFIGURE = 1, /* Switch configuration changed. */
323 REV_STP, /* Spanning tree protocol port status change. */
324 REV_PORT_TOGGLED, /* Port enabled or disabled by CFM, LACP, ...*/
325 REV_FLOW_TABLE, /* Flow table changed. */
326 REV_INCONSISTENCY /* Facet self-check failed. */
328 COVERAGE_DEFINE(rev_reconfigure);
329 COVERAGE_DEFINE(rev_stp);
330 COVERAGE_DEFINE(rev_port_toggled);
331 COVERAGE_DEFINE(rev_flow_table);
332 COVERAGE_DEFINE(rev_inconsistency);
334 /* Drop keys are odp flow keys which have drop flows installed in the kernel.
335 * These are datapath flows which have no associated ofproto, if they did we
336 * would use facets. */
338 struct hmap_node hmap_node;
343 struct avg_subfacet_rates {
344 double add_rate; /* Moving average of new flows created per minute. */
345 double del_rate; /* Moving average of flows deleted per minute. */
348 /* All datapaths of a given type share a single dpif backer instance. */
353 struct timer next_expiration;
354 struct hmap odp_to_ofport_map; /* ODP port to ofport mapping. */
356 struct simap tnl_backers; /* Set of dpif ports backing tunnels. */
358 /* Facet revalidation flags applying to facets which use this backer. */
359 enum revalidate_reason need_revalidate; /* Revalidate every facet. */
360 struct tag_set revalidate_set; /* Revalidate only matching facets. */
362 struct hmap drop_keys; /* Set of dropped odp keys. */
363 bool recv_set_enable; /* Enables or disables receiving packets. */
365 struct hmap subfacets;
366 struct governor *governor;
368 /* Subfacet statistics.
370 * These keep track of the total number of subfacets added and deleted and
371 * flow life span. They are useful for computing the flow rates stats
372 * exposed via "ovs-appctl dpif/show". The goal is to learn about
373 * traffic patterns in ways that we can use later to improve Open vSwitch
374 * performance in new situations. */
375 long long int created; /* Time when it is created. */
376 unsigned max_n_subfacet; /* Maximum number of flows */
377 unsigned avg_n_subfacet; /* Average number of flows. */
378 long long int avg_subfacet_life; /* Average life span of subfacets. */
380 /* The average number of subfacets... */
381 struct avg_subfacet_rates hourly; /* ...over the last hour. */
382 struct avg_subfacet_rates daily; /* ...over the last day. */
383 struct avg_subfacet_rates lifetime; /* ...over the switch lifetime. */
384 long long int last_minute; /* Last time 'hourly' was updated. */
386 /* Number of subfacets added or deleted since 'last_minute'. */
387 unsigned subfacet_add_count;
388 unsigned subfacet_del_count;
390 /* Number of subfacets added or deleted from 'created' to 'last_minute.' */
391 unsigned long long int total_subfacet_add_count;
392 unsigned long long int total_subfacet_del_count;
395 /* All existing ofproto_backer instances, indexed by ofproto->up.type. */
396 static struct shash all_dpif_backers = SHASH_INITIALIZER(&all_dpif_backers);
398 static void drop_key_clear(struct dpif_backer *);
399 static struct ofport_dpif *
400 odp_port_to_ofport(const struct dpif_backer *, odp_port_t odp_port);
401 static void update_moving_averages(struct dpif_backer *backer);
403 /* Defer flow mod completion until "ovs-appctl ofproto/unclog"? (Useful only
404 * for debugging the asynchronous flow_mod implementation.) */
407 /* All existing ofproto_dpif instances, indexed by ->up.name. */
408 static struct hmap all_ofproto_dpifs = HMAP_INITIALIZER(&all_ofproto_dpifs);
410 static void ofproto_dpif_unixctl_init(void);
413 #define FLOW_MISS_MAX_BATCH 50
414 static int handle_upcalls(struct dpif_backer *, unsigned int max_batch);
416 /* Flow expiration. */
417 static int expire(struct dpif_backer *);
420 static void send_netflow_active_timeouts(struct ofproto_dpif *);
423 static int send_packet(const struct ofport_dpif *, struct ofpbuf *packet);
425 /* Global variables. */
426 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
428 /* Initial mappings of port to bridge mappings. */
429 static struct shash init_ofp_ports = SHASH_INITIALIZER(&init_ofp_ports);
431 /* Factory functions. */
434 init(const struct shash *iface_hints)
436 struct shash_node *node;
438 /* Make a local copy, since we don't own 'iface_hints' elements. */
439 SHASH_FOR_EACH(node, iface_hints) {
440 const struct iface_hint *orig_hint = node->data;
441 struct iface_hint *new_hint = xmalloc(sizeof *new_hint);
443 new_hint->br_name = xstrdup(orig_hint->br_name);
444 new_hint->br_type = xstrdup(orig_hint->br_type);
445 new_hint->ofp_port = orig_hint->ofp_port;
447 shash_add(&init_ofp_ports, node->name, new_hint);
452 enumerate_types(struct sset *types)
454 dp_enumerate_types(types);
458 enumerate_names(const char *type, struct sset *names)
460 struct ofproto_dpif *ofproto;
463 HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
464 if (strcmp(type, ofproto->up.type)) {
467 sset_add(names, ofproto->up.name);
474 del(const char *type, const char *name)
479 error = dpif_open(name, type, &dpif);
481 error = dpif_delete(dpif);
488 port_open_type(const char *datapath_type, const char *port_type)
490 return dpif_port_open_type(datapath_type, port_type);
493 /* Type functions. */
495 static struct ofproto_dpif *
496 lookup_ofproto_dpif_by_port_name(const char *name)
498 struct ofproto_dpif *ofproto;
500 HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
501 if (sset_contains(&ofproto->ports, name)) {
510 type_run(const char *type)
512 static long long int push_timer = LLONG_MIN;
513 struct dpif_backer *backer;
517 backer = shash_find_data(&all_dpif_backers, type);
519 /* This is not necessarily a problem, since backers are only
520 * created on demand. */
524 dpif_run(backer->dpif);
526 /* The most natural place to push facet statistics is when they're pulled
527 * from the datapath. However, when there are many flows in the datapath,
528 * this expensive operation can occur so frequently, that it reduces our
529 * ability to quickly set up flows. To reduce the cost, we push statistics
531 if (time_msec() > push_timer) {
532 push_timer = time_msec() + 2000;
536 /* If vswitchd started with other_config:flow_restore_wait set as "true",
537 * and the configuration has now changed to "false", enable receiving
538 * packets from the datapath. */
539 if (!backer->recv_set_enable && !ofproto_get_flow_restore_wait()) {
540 backer->recv_set_enable = true;
542 error = dpif_recv_set(backer->dpif, backer->recv_set_enable);
544 VLOG_ERR("Failed to enable receiving packets in dpif.");
547 dpif_flow_flush(backer->dpif);
548 backer->need_revalidate = REV_RECONFIGURE;
551 if (backer->need_revalidate
552 || !tag_set_is_empty(&backer->revalidate_set)) {
553 struct tag_set revalidate_set = backer->revalidate_set;
554 bool need_revalidate = backer->need_revalidate;
555 struct ofproto_dpif *ofproto;
556 struct simap_node *node;
557 struct simap tmp_backers;
559 /* Handle tunnel garbage collection. */
560 simap_init(&tmp_backers);
561 simap_swap(&backer->tnl_backers, &tmp_backers);
563 HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
564 struct ofport_dpif *iter;
566 if (backer != ofproto->backer) {
570 HMAP_FOR_EACH (iter, up.hmap_node, &ofproto->up.ports) {
571 char namebuf[NETDEV_VPORT_NAME_BUFSIZE];
574 if (!iter->tnl_port) {
578 dp_port = netdev_vport_get_dpif_port(iter->up.netdev,
579 namebuf, sizeof namebuf);
580 node = simap_find(&tmp_backers, dp_port);
582 simap_put(&backer->tnl_backers, dp_port, node->data);
583 simap_delete(&tmp_backers, node);
584 node = simap_find(&backer->tnl_backers, dp_port);
586 node = simap_find(&backer->tnl_backers, dp_port);
588 odp_port_t odp_port = ODPP_NONE;
590 if (!dpif_port_add(backer->dpif, iter->up.netdev,
592 simap_put(&backer->tnl_backers, dp_port,
593 odp_to_u32(odp_port));
594 node = simap_find(&backer->tnl_backers, dp_port);
599 iter->odp_port = node ? u32_to_odp(node->data) : ODPP_NONE;
600 if (tnl_port_reconfigure(&iter->up, iter->odp_port,
602 backer->need_revalidate = REV_RECONFIGURE;
607 SIMAP_FOR_EACH (node, &tmp_backers) {
608 dpif_port_del(backer->dpif, u32_to_odp(node->data));
610 simap_destroy(&tmp_backers);
612 switch (backer->need_revalidate) {
613 case REV_RECONFIGURE: COVERAGE_INC(rev_reconfigure); break;
614 case REV_STP: COVERAGE_INC(rev_stp); break;
615 case REV_PORT_TOGGLED: COVERAGE_INC(rev_port_toggled); break;
616 case REV_FLOW_TABLE: COVERAGE_INC(rev_flow_table); break;
617 case REV_INCONSISTENCY: COVERAGE_INC(rev_inconsistency); break;
620 if (backer->need_revalidate) {
621 /* Clear the drop_keys in case we should now be accepting some
622 * formerly dropped flows. */
623 drop_key_clear(backer);
626 /* Clear the revalidation flags. */
627 tag_set_init(&backer->revalidate_set);
628 backer->need_revalidate = 0;
630 HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
631 struct facet *facet, *next;
632 struct cls_cursor cursor;
634 if (ofproto->backer != backer) {
638 cls_cursor_init(&cursor, &ofproto->facets, NULL);
639 CLS_CURSOR_FOR_EACH_SAFE (facet, next, cr, &cursor) {
641 || tag_set_intersects(&revalidate_set, facet->xout.tags)) {
642 facet_revalidate(facet);
649 if (!backer->recv_set_enable) {
650 /* Wake up before a max of 1000ms. */
651 timer_set_duration(&backer->next_expiration, 1000);
652 } else if (timer_expired(&backer->next_expiration)) {
653 int delay = expire(backer);
654 timer_set_duration(&backer->next_expiration, delay);
657 /* Check for port changes in the dpif. */
658 while ((error = dpif_port_poll(backer->dpif, &devname)) == 0) {
659 struct ofproto_dpif *ofproto;
660 struct dpif_port port;
662 /* Don't report on the datapath's device. */
663 if (!strcmp(devname, dpif_base_name(backer->dpif))) {
667 HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node,
668 &all_ofproto_dpifs) {
669 if (simap_contains(&ofproto->backer->tnl_backers, devname)) {
674 ofproto = lookup_ofproto_dpif_by_port_name(devname);
675 if (dpif_port_query_by_name(backer->dpif, devname, &port)) {
676 /* The port was removed. If we know the datapath,
677 * report it through poll_set(). If we don't, it may be
678 * notifying us of a removal we initiated, so ignore it.
679 * If there's a pending ENOBUFS, let it stand, since
680 * everything will be reevaluated. */
681 if (ofproto && ofproto->port_poll_errno != ENOBUFS) {
682 sset_add(&ofproto->port_poll_set, devname);
683 ofproto->port_poll_errno = 0;
685 } else if (!ofproto) {
686 /* The port was added, but we don't know with which
687 * ofproto we should associate it. Delete it. */
688 dpif_port_del(backer->dpif, port.port_no);
690 dpif_port_destroy(&port);
696 if (error != EAGAIN) {
697 struct ofproto_dpif *ofproto;
699 /* There was some sort of error, so propagate it to all
700 * ofprotos that use this backer. */
701 HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node,
702 &all_ofproto_dpifs) {
703 if (ofproto->backer == backer) {
704 sset_clear(&ofproto->port_poll_set);
705 ofproto->port_poll_errno = error;
710 if (backer->governor) {
713 governor_run(backer->governor);
715 /* If the governor has shrunk to its minimum size and the number of
716 * subfacets has dwindled, then drop the governor entirely.
718 * For hysteresis, the number of subfacets to drop the governor is
719 * smaller than the number needed to trigger its creation. */
720 n_subfacets = hmap_count(&backer->subfacets);
721 if (n_subfacets * 4 < flow_eviction_threshold
722 && governor_is_idle(backer->governor)) {
723 governor_destroy(backer->governor);
724 backer->governor = NULL;
732 dpif_backer_run_fast(struct dpif_backer *backer, int max_batch)
736 /* If recv_set_enable is false, we should not handle upcalls. */
737 if (!backer->recv_set_enable) {
741 /* Handle one or more batches of upcalls, until there's nothing left to do
742 * or until we do a fixed total amount of work.
744 * We do work in batches because it can be much cheaper to set up a number
745 * of flows and fire off their patches all at once. We do multiple batches
746 * because in some cases handling a packet can cause another packet to be
747 * queued almost immediately as part of the return flow. Both
748 * optimizations can make major improvements on some benchmarks and
749 * presumably for real traffic as well. */
751 while (work < max_batch) {
752 int retval = handle_upcalls(backer, max_batch - work);
763 type_run_fast(const char *type)
765 struct dpif_backer *backer;
767 backer = shash_find_data(&all_dpif_backers, type);
769 /* This is not necessarily a problem, since backers are only
770 * created on demand. */
774 return dpif_backer_run_fast(backer, FLOW_MISS_MAX_BATCH);
780 static long long int port_rl = LLONG_MIN;
781 static unsigned int backer_rl = 0;
783 if (time_msec() >= port_rl) {
784 struct ofproto_dpif *ofproto;
785 struct ofport_dpif *ofport;
787 HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
789 HMAP_FOR_EACH (ofport, up.hmap_node, &ofproto->up.ports) {
790 port_run_fast(ofport);
793 port_rl = time_msec() + 200;
796 /* XXX: We have to be careful not to do too much work in this function. If
797 * we call dpif_backer_run_fast() too often, or with too large a batch,
798 * performance improves signifcantly, but at a cost. It's possible for the
799 * number of flows in the datapath to increase without bound, and for poll
800 * loops to take 10s of seconds. The correct solution to this problem,
801 * long term, is to separate flow miss handling into it's own thread so it
802 * isn't affected by revalidations, and expirations. Until then, this is
803 * the best we can do. */
804 if (++backer_rl >= 10) {
805 struct shash_node *node;
808 SHASH_FOR_EACH (node, &all_dpif_backers) {
809 dpif_backer_run_fast(node->data, 1);
815 type_wait(const char *type)
817 struct dpif_backer *backer;
819 backer = shash_find_data(&all_dpif_backers, type);
821 /* This is not necessarily a problem, since backers are only
822 * created on demand. */
826 if (backer->governor) {
827 governor_wait(backer->governor);
830 timer_wait(&backer->next_expiration);
833 /* Basic life-cycle. */
835 static int add_internal_flows(struct ofproto_dpif *);
837 static struct ofproto *
840 struct ofproto_dpif *ofproto = xmalloc(sizeof *ofproto);
845 dealloc(struct ofproto *ofproto_)
847 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
852 close_dpif_backer(struct dpif_backer *backer)
854 struct shash_node *node;
856 ovs_assert(backer->refcount > 0);
858 if (--backer->refcount) {
862 drop_key_clear(backer);
863 hmap_destroy(&backer->drop_keys);
865 simap_destroy(&backer->tnl_backers);
866 hmap_destroy(&backer->odp_to_ofport_map);
867 node = shash_find(&all_dpif_backers, backer->type);
869 shash_delete(&all_dpif_backers, node);
870 dpif_close(backer->dpif);
872 ovs_assert(hmap_is_empty(&backer->subfacets));
873 hmap_destroy(&backer->subfacets);
874 governor_destroy(backer->governor);
879 /* Datapath port slated for removal from datapath. */
881 struct list list_node;
886 open_dpif_backer(const char *type, struct dpif_backer **backerp)
888 struct dpif_backer *backer;
889 struct dpif_port_dump port_dump;
890 struct dpif_port port;
891 struct shash_node *node;
892 struct list garbage_list;
893 struct odp_garbage *garbage, *next;
899 backer = shash_find_data(&all_dpif_backers, type);
906 backer_name = xasprintf("ovs-%s", type);
908 /* Remove any existing datapaths, since we assume we're the only
909 * userspace controlling the datapath. */
911 dp_enumerate_names(type, &names);
912 SSET_FOR_EACH(name, &names) {
913 struct dpif *old_dpif;
915 /* Don't remove our backer if it exists. */
916 if (!strcmp(name, backer_name)) {
920 if (dpif_open(name, type, &old_dpif)) {
921 VLOG_WARN("couldn't open old datapath %s to remove it", name);
923 dpif_delete(old_dpif);
924 dpif_close(old_dpif);
927 sset_destroy(&names);
929 backer = xmalloc(sizeof *backer);
931 error = dpif_create_and_open(backer_name, type, &backer->dpif);
934 VLOG_ERR("failed to open datapath of type %s: %s", type,
940 backer->type = xstrdup(type);
941 backer->governor = NULL;
942 backer->refcount = 1;
943 hmap_init(&backer->odp_to_ofport_map);
944 hmap_init(&backer->drop_keys);
945 hmap_init(&backer->subfacets);
946 timer_set_duration(&backer->next_expiration, 1000);
947 backer->need_revalidate = 0;
948 simap_init(&backer->tnl_backers);
949 tag_set_init(&backer->revalidate_set);
950 backer->recv_set_enable = !ofproto_get_flow_restore_wait();
953 if (backer->recv_set_enable) {
954 dpif_flow_flush(backer->dpif);
957 /* Loop through the ports already on the datapath and remove any
958 * that we don't need anymore. */
959 list_init(&garbage_list);
960 dpif_port_dump_start(&port_dump, backer->dpif);
961 while (dpif_port_dump_next(&port_dump, &port)) {
962 node = shash_find(&init_ofp_ports, port.name);
963 if (!node && strcmp(port.name, dpif_base_name(backer->dpif))) {
964 garbage = xmalloc(sizeof *garbage);
965 garbage->odp_port = port.port_no;
966 list_push_front(&garbage_list, &garbage->list_node);
969 dpif_port_dump_done(&port_dump);
971 LIST_FOR_EACH_SAFE (garbage, next, list_node, &garbage_list) {
972 dpif_port_del(backer->dpif, garbage->odp_port);
973 list_remove(&garbage->list_node);
977 shash_add(&all_dpif_backers, type, backer);
979 error = dpif_recv_set(backer->dpif, backer->recv_set_enable);
981 VLOG_ERR("failed to listen on datapath of type %s: %s",
982 type, strerror(error));
983 close_dpif_backer(backer);
987 backer->max_n_subfacet = 0;
988 backer->created = time_msec();
989 backer->last_minute = backer->created;
990 memset(&backer->hourly, 0, sizeof backer->hourly);
991 memset(&backer->daily, 0, sizeof backer->daily);
992 memset(&backer->lifetime, 0, sizeof backer->lifetime);
993 backer->subfacet_add_count = 0;
994 backer->subfacet_del_count = 0;
995 backer->total_subfacet_add_count = 0;
996 backer->total_subfacet_del_count = 0;
997 backer->avg_n_subfacet = 0;
998 backer->avg_subfacet_life = 0;
1004 construct(struct ofproto *ofproto_)
1006 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1007 struct shash_node *node, *next;
1008 odp_port_t max_ports;
1012 error = open_dpif_backer(ofproto->up.type, &ofproto->backer);
1017 max_ports = dpif_get_max_ports(ofproto->backer->dpif);
1018 ofproto_init_max_ports(ofproto_, u16_to_ofp(MIN(odp_to_u32(max_ports),
1019 ofp_to_u16(OFPP_MAX))));
1021 ofproto->netflow = NULL;
1022 ofproto->sflow = NULL;
1023 ofproto->ipfix = NULL;
1024 ofproto->stp = NULL;
1025 hmap_init(&ofproto->bundles);
1026 ofproto->ml = mac_learning_create(MAC_ENTRY_DEFAULT_IDLE_TIME);
1027 for (i = 0; i < MAX_MIRRORS; i++) {
1028 ofproto->mirrors[i] = NULL;
1030 ofproto->has_bonded_bundles = false;
1032 classifier_init(&ofproto->facets);
1033 ofproto->consistency_rl = LLONG_MIN;
1035 for (i = 0; i < N_TABLES; i++) {
1036 struct table_dpif *table = &ofproto->tables[i];
1038 table->catchall_table = NULL;
1039 table->other_table = NULL;
1040 table->basis = random_uint32();
1043 list_init(&ofproto->completions);
1045 ofproto_dpif_unixctl_init();
1047 ofproto->has_mirrors = false;
1048 ofproto->has_bundle_action = false;
1050 hmap_init(&ofproto->vlandev_map);
1051 hmap_init(&ofproto->realdev_vid_map);
1053 sset_init(&ofproto->ports);
1054 sset_init(&ofproto->ghost_ports);
1055 sset_init(&ofproto->port_poll_set);
1056 ofproto->port_poll_errno = 0;
1058 SHASH_FOR_EACH_SAFE (node, next, &init_ofp_ports) {
1059 struct iface_hint *iface_hint = node->data;
1061 if (!strcmp(iface_hint->br_name, ofproto->up.name)) {
1062 /* Check if the datapath already has this port. */
1063 if (dpif_port_exists(ofproto->backer->dpif, node->name)) {
1064 sset_add(&ofproto->ports, node->name);
1067 free(iface_hint->br_name);
1068 free(iface_hint->br_type);
1070 shash_delete(&init_ofp_ports, node);
1074 hmap_insert(&all_ofproto_dpifs, &ofproto->all_ofproto_dpifs_node,
1075 hash_string(ofproto->up.name, 0));
1076 memset(&ofproto->stats, 0, sizeof ofproto->stats);
1078 ofproto_init_tables(ofproto_, N_TABLES);
1079 error = add_internal_flows(ofproto);
1080 ofproto->up.tables[TBL_INTERNAL].flags = OFTABLE_HIDDEN | OFTABLE_READONLY;
1083 ofproto->n_missed = 0;
1089 add_internal_flow(struct ofproto_dpif *ofproto, int id,
1090 const struct ofpbuf *ofpacts, struct rule_dpif **rulep)
1092 struct ofputil_flow_mod fm;
1095 match_init_catchall(&fm.match);
1097 match_set_reg(&fm.match, 0, id);
1098 fm.new_cookie = htonll(0);
1099 fm.cookie = htonll(0);
1100 fm.cookie_mask = htonll(0);
1101 fm.table_id = TBL_INTERNAL;
1102 fm.command = OFPFC_ADD;
1103 fm.idle_timeout = 0;
1104 fm.hard_timeout = 0;
1108 fm.ofpacts = ofpacts->data;
1109 fm.ofpacts_len = ofpacts->size;
1111 error = ofproto_flow_mod(&ofproto->up, &fm);
1113 VLOG_ERR_RL(&rl, "failed to add internal flow %d (%s)",
1114 id, ofperr_to_string(error));
1118 *rulep = rule_dpif_lookup_in_table(ofproto, &fm.match.flow, NULL,
1120 ovs_assert(*rulep != NULL);
1126 add_internal_flows(struct ofproto_dpif *ofproto)
1128 struct ofpact_controller *controller;
1129 uint64_t ofpacts_stub[128 / 8];
1130 struct ofpbuf ofpacts;
1134 ofpbuf_use_stack(&ofpacts, ofpacts_stub, sizeof ofpacts_stub);
1137 controller = ofpact_put_CONTROLLER(&ofpacts);
1138 controller->max_len = UINT16_MAX;
1139 controller->controller_id = 0;
1140 controller->reason = OFPR_NO_MATCH;
1141 ofpact_pad(&ofpacts);
1143 error = add_internal_flow(ofproto, id++, &ofpacts, &ofproto->miss_rule);
1148 ofpbuf_clear(&ofpacts);
1149 error = add_internal_flow(ofproto, id++, &ofpacts,
1150 &ofproto->no_packet_in_rule);
1155 error = add_internal_flow(ofproto, id++, &ofpacts,
1156 &ofproto->drop_frags_rule);
1161 complete_operations(struct ofproto_dpif *ofproto)
1163 struct dpif_completion *c, *next;
1165 LIST_FOR_EACH_SAFE (c, next, list_node, &ofproto->completions) {
1166 ofoperation_complete(c->op, 0);
1167 list_remove(&c->list_node);
1173 destruct(struct ofproto *ofproto_)
1175 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1176 struct rule_dpif *rule, *next_rule;
1177 struct oftable *table;
1180 ofproto->backer->need_revalidate = REV_RECONFIGURE;
1181 hmap_remove(&all_ofproto_dpifs, &ofproto->all_ofproto_dpifs_node);
1182 complete_operations(ofproto);
1184 OFPROTO_FOR_EACH_TABLE (table, &ofproto->up) {
1185 struct cls_cursor cursor;
1187 cls_cursor_init(&cursor, &table->cls, NULL);
1188 CLS_CURSOR_FOR_EACH_SAFE (rule, next_rule, up.cr, &cursor) {
1189 ofproto_rule_destroy(&rule->up);
1193 for (i = 0; i < MAX_MIRRORS; i++) {
1194 mirror_destroy(ofproto->mirrors[i]);
1197 netflow_destroy(ofproto->netflow);
1198 dpif_sflow_destroy(ofproto->sflow);
1199 hmap_destroy(&ofproto->bundles);
1200 mac_learning_destroy(ofproto->ml);
1202 classifier_destroy(&ofproto->facets);
1204 hmap_destroy(&ofproto->vlandev_map);
1205 hmap_destroy(&ofproto->realdev_vid_map);
1207 sset_destroy(&ofproto->ports);
1208 sset_destroy(&ofproto->ghost_ports);
1209 sset_destroy(&ofproto->port_poll_set);
1211 close_dpif_backer(ofproto->backer);
1215 run_fast(struct ofproto *ofproto_)
1217 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1218 struct ofport_dpif *ofport;
1220 /* Do not perform any periodic activity required by 'ofproto' while
1221 * waiting for flow restore to complete. */
1222 if (ofproto_get_flow_restore_wait()) {
1226 HMAP_FOR_EACH (ofport, up.hmap_node, &ofproto->up.ports) {
1227 port_run_fast(ofport);
1234 run(struct ofproto *ofproto_)
1236 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1237 struct ofport_dpif *ofport;
1238 struct ofbundle *bundle;
1242 complete_operations(ofproto);
1245 /* Do not perform any periodic activity below required by 'ofproto' while
1246 * waiting for flow restore to complete. */
1247 if (ofproto_get_flow_restore_wait()) {
1251 error = run_fast(ofproto_);
1256 if (ofproto->netflow) {
1257 if (netflow_run(ofproto->netflow)) {
1258 send_netflow_active_timeouts(ofproto);
1261 if (ofproto->sflow) {
1262 dpif_sflow_run(ofproto->sflow);
1265 HMAP_FOR_EACH (ofport, up.hmap_node, &ofproto->up.ports) {
1268 HMAP_FOR_EACH (bundle, hmap_node, &ofproto->bundles) {
1273 mac_learning_run(ofproto->ml, &ofproto->backer->revalidate_set);
1275 /* Check the consistency of a random facet, to aid debugging. */
1276 if (time_msec() >= ofproto->consistency_rl
1277 && !classifier_is_empty(&ofproto->facets)
1278 && !ofproto->backer->need_revalidate) {
1279 struct cls_table *table;
1280 struct cls_rule *cr;
1281 struct facet *facet;
1283 ofproto->consistency_rl = time_msec() + 250;
1285 table = CONTAINER_OF(hmap_random_node(&ofproto->facets.tables),
1286 struct cls_table, hmap_node);
1287 cr = CONTAINER_OF(hmap_random_node(&table->rules), struct cls_rule,
1289 facet = CONTAINER_OF(cr, struct facet, cr);
1291 if (!tag_set_intersects(&ofproto->backer->revalidate_set,
1292 facet->xout.tags)) {
1293 if (!facet_check_consistency(facet)) {
1294 ofproto->backer->need_revalidate = REV_INCONSISTENCY;
1303 wait(struct ofproto *ofproto_)
1305 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1306 struct ofport_dpif *ofport;
1307 struct ofbundle *bundle;
1309 if (!clogged && !list_is_empty(&ofproto->completions)) {
1310 poll_immediate_wake();
1313 if (ofproto_get_flow_restore_wait()) {
1317 dpif_wait(ofproto->backer->dpif);
1318 dpif_recv_wait(ofproto->backer->dpif);
1319 if (ofproto->sflow) {
1320 dpif_sflow_wait(ofproto->sflow);
1322 if (!tag_set_is_empty(&ofproto->backer->revalidate_set)) {
1323 poll_immediate_wake();
1325 HMAP_FOR_EACH (ofport, up.hmap_node, &ofproto->up.ports) {
1328 HMAP_FOR_EACH (bundle, hmap_node, &ofproto->bundles) {
1329 bundle_wait(bundle);
1331 if (ofproto->netflow) {
1332 netflow_wait(ofproto->netflow);
1334 mac_learning_wait(ofproto->ml);
1336 if (ofproto->backer->need_revalidate) {
1337 /* Shouldn't happen, but if it does just go around again. */
1338 VLOG_DBG_RL(&rl, "need revalidate in ofproto_wait_cb()");
1339 poll_immediate_wake();
1344 get_memory_usage(const struct ofproto *ofproto_, struct simap *usage)
1346 const struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1347 struct cls_cursor cursor;
1348 size_t n_subfacets = 0;
1349 struct facet *facet;
1351 simap_increase(usage, "facets", classifier_count(&ofproto->facets));
1353 cls_cursor_init(&cursor, &ofproto->facets, NULL);
1354 CLS_CURSOR_FOR_EACH (facet, cr, &cursor) {
1355 n_subfacets += list_size(&facet->subfacets);
1357 simap_increase(usage, "subfacets", n_subfacets);
1361 flush(struct ofproto *ofproto_)
1363 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1364 struct subfacet *subfacet, *next_subfacet;
1365 struct subfacet *batch[SUBFACET_DESTROY_MAX_BATCH];
1369 HMAP_FOR_EACH_SAFE (subfacet, next_subfacet, hmap_node,
1370 &ofproto->backer->subfacets) {
1371 if (ofproto_dpif_cast(subfacet->facet->rule->up.ofproto) != ofproto) {
1375 if (subfacet->path != SF_NOT_INSTALLED) {
1376 batch[n_batch++] = subfacet;
1377 if (n_batch >= SUBFACET_DESTROY_MAX_BATCH) {
1378 subfacet_destroy_batch(ofproto->backer, batch, n_batch);
1382 subfacet_destroy(subfacet);
1387 subfacet_destroy_batch(ofproto->backer, batch, n_batch);
1392 get_features(struct ofproto *ofproto_ OVS_UNUSED,
1393 bool *arp_match_ip, enum ofputil_action_bitmap *actions)
1395 *arp_match_ip = true;
1396 *actions = (OFPUTIL_A_OUTPUT |
1397 OFPUTIL_A_SET_VLAN_VID |
1398 OFPUTIL_A_SET_VLAN_PCP |
1399 OFPUTIL_A_STRIP_VLAN |
1400 OFPUTIL_A_SET_DL_SRC |
1401 OFPUTIL_A_SET_DL_DST |
1402 OFPUTIL_A_SET_NW_SRC |
1403 OFPUTIL_A_SET_NW_DST |
1404 OFPUTIL_A_SET_NW_TOS |
1405 OFPUTIL_A_SET_TP_SRC |
1406 OFPUTIL_A_SET_TP_DST |
1411 get_tables(struct ofproto *ofproto_, struct ofp12_table_stats *ots)
1413 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1414 struct dpif_dp_stats s;
1415 uint64_t n_miss, n_no_pkt_in, n_bytes, n_dropped_frags;
1418 strcpy(ots->name, "classifier");
1420 dpif_get_dp_stats(ofproto->backer->dpif, &s);
1421 rule_get_stats(&ofproto->miss_rule->up, &n_miss, &n_bytes);
1422 rule_get_stats(&ofproto->no_packet_in_rule->up, &n_no_pkt_in, &n_bytes);
1423 rule_get_stats(&ofproto->drop_frags_rule->up, &n_dropped_frags, &n_bytes);
1425 n_lookup = s.n_hit + s.n_missed - n_dropped_frags;
1426 ots->lookup_count = htonll(n_lookup);
1427 ots->matched_count = htonll(n_lookup - n_miss - n_no_pkt_in);
1430 static struct ofport *
1433 struct ofport_dpif *port = xmalloc(sizeof *port);
1438 port_dealloc(struct ofport *port_)
1440 struct ofport_dpif *port = ofport_dpif_cast(port_);
1445 port_construct(struct ofport *port_)
1447 struct ofport_dpif *port = ofport_dpif_cast(port_);
1448 struct ofproto_dpif *ofproto = ofproto_dpif_cast(port->up.ofproto);
1449 const struct netdev *netdev = port->up.netdev;
1450 char namebuf[NETDEV_VPORT_NAME_BUFSIZE];
1451 struct dpif_port dpif_port;
1454 ofproto->backer->need_revalidate = REV_RECONFIGURE;
1455 port->bundle = NULL;
1458 port->tag = tag_create_random();
1459 port->may_enable = true;
1460 port->stp_port = NULL;
1461 port->stp_state = STP_DISABLED;
1462 port->tnl_port = NULL;
1464 hmap_init(&port->priorities);
1465 port->realdev_ofp_port = 0;
1466 port->vlandev_vid = 0;
1467 port->carrier_seq = netdev_get_carrier_resets(netdev);
1469 if (netdev_vport_is_patch(netdev)) {
1470 /* By bailing out here, we don't submit the port to the sFlow module
1471 * to be considered for counter polling export. This is correct
1472 * because the patch port represents an interface that sFlow considers
1473 * to be "internal" to the switch as a whole, and therefore not an
1474 * candidate for counter polling. */
1475 port->odp_port = ODPP_NONE;
1476 ofport_update_peer(port);
1480 error = dpif_port_query_by_name(ofproto->backer->dpif,
1481 netdev_vport_get_dpif_port(netdev, namebuf,
1488 port->odp_port = dpif_port.port_no;
1490 if (netdev_get_tunnel_config(netdev)) {
1491 port->tnl_port = tnl_port_add(&port->up, port->odp_port);
1493 /* Sanity-check that a mapping doesn't already exist. This
1494 * shouldn't happen for non-tunnel ports. */
1495 if (odp_port_to_ofp_port(ofproto, port->odp_port) != OFPP_NONE) {
1496 VLOG_ERR("port %s already has an OpenFlow port number",
1498 dpif_port_destroy(&dpif_port);
1502 hmap_insert(&ofproto->backer->odp_to_ofport_map, &port->odp_port_node,
1503 hash_int(odp_to_u32(port->odp_port), 0));
1505 dpif_port_destroy(&dpif_port);
1507 if (ofproto->sflow) {
1508 dpif_sflow_add_port(ofproto->sflow, port_, port->odp_port);
1515 port_destruct(struct ofport *port_)
1517 struct ofport_dpif *port = ofport_dpif_cast(port_);
1518 struct ofproto_dpif *ofproto = ofproto_dpif_cast(port->up.ofproto);
1519 const char *devname = netdev_get_name(port->up.netdev);
1520 char namebuf[NETDEV_VPORT_NAME_BUFSIZE];
1521 const char *dp_port_name;
1523 dp_port_name = netdev_vport_get_dpif_port(port->up.netdev, namebuf,
1525 if (dpif_port_exists(ofproto->backer->dpif, dp_port_name)) {
1526 /* The underlying device is still there, so delete it. This
1527 * happens when the ofproto is being destroyed, since the caller
1528 * assumes that removal of attached ports will happen as part of
1530 if (!port->tnl_port) {
1531 dpif_port_del(ofproto->backer->dpif, port->odp_port);
1533 ofproto->backer->need_revalidate = REV_RECONFIGURE;
1537 port->peer->peer = NULL;
1541 if (port->odp_port != ODPP_NONE && !port->tnl_port) {
1542 hmap_remove(&ofproto->backer->odp_to_ofport_map, &port->odp_port_node);
1545 tnl_port_del(port->tnl_port);
1546 sset_find_and_delete(&ofproto->ports, devname);
1547 sset_find_and_delete(&ofproto->ghost_ports, devname);
1548 ofproto->backer->need_revalidate = REV_RECONFIGURE;
1549 bundle_remove(port_);
1550 set_cfm(port_, NULL);
1551 set_bfd(port_, NULL);
1552 if (ofproto->sflow) {
1553 dpif_sflow_del_port(ofproto->sflow, port->odp_port);
1556 ofport_clear_priorities(port);
1557 hmap_destroy(&port->priorities);
1561 port_modified(struct ofport *port_)
1563 struct ofport_dpif *port = ofport_dpif_cast(port_);
1565 if (port->bundle && port->bundle->bond) {
1566 bond_slave_set_netdev(port->bundle->bond, port, port->up.netdev);
1570 cfm_set_netdev(port->cfm, port->up.netdev);
1573 if (port->tnl_port && tnl_port_reconfigure(&port->up, port->odp_port,
1575 ofproto_dpif_cast(port->up.ofproto)->backer->need_revalidate = true;
1578 ofport_update_peer(port);
1582 port_reconfigured(struct ofport *port_, enum ofputil_port_config old_config)
1584 struct ofport_dpif *port = ofport_dpif_cast(port_);
1585 struct ofproto_dpif *ofproto = ofproto_dpif_cast(port->up.ofproto);
1586 enum ofputil_port_config changed = old_config ^ port->up.pp.config;
1588 if (changed & (OFPUTIL_PC_NO_RECV | OFPUTIL_PC_NO_RECV_STP |
1589 OFPUTIL_PC_NO_FWD | OFPUTIL_PC_NO_FLOOD |
1590 OFPUTIL_PC_NO_PACKET_IN)) {
1591 ofproto->backer->need_revalidate = REV_RECONFIGURE;
1593 if (changed & OFPUTIL_PC_NO_FLOOD && port->bundle) {
1594 bundle_update(port->bundle);
1600 set_sflow(struct ofproto *ofproto_,
1601 const struct ofproto_sflow_options *sflow_options)
1603 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1604 struct dpif_sflow *ds = ofproto->sflow;
1606 if (sflow_options) {
1608 struct ofport_dpif *ofport;
1610 ds = ofproto->sflow = dpif_sflow_create();
1611 HMAP_FOR_EACH (ofport, up.hmap_node, &ofproto->up.ports) {
1612 dpif_sflow_add_port(ds, &ofport->up, ofport->odp_port);
1614 ofproto->backer->need_revalidate = REV_RECONFIGURE;
1616 dpif_sflow_set_options(ds, sflow_options);
1619 dpif_sflow_destroy(ds);
1620 ofproto->backer->need_revalidate = REV_RECONFIGURE;
1621 ofproto->sflow = NULL;
1629 struct ofproto *ofproto_,
1630 const struct ofproto_ipfix_bridge_exporter_options *bridge_exporter_options,
1631 const struct ofproto_ipfix_flow_exporter_options *flow_exporters_options,
1632 size_t n_flow_exporters_options)
1634 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1635 struct dpif_ipfix *di = ofproto->ipfix;
1637 if (bridge_exporter_options || flow_exporters_options) {
1639 di = ofproto->ipfix = dpif_ipfix_create();
1641 dpif_ipfix_set_options(
1642 di, bridge_exporter_options, flow_exporters_options,
1643 n_flow_exporters_options);
1646 dpif_ipfix_destroy(di);
1647 ofproto->ipfix = NULL;
1654 set_cfm(struct ofport *ofport_, const struct cfm_settings *s)
1656 struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
1663 struct ofproto_dpif *ofproto;
1665 ofproto = ofproto_dpif_cast(ofport->up.ofproto);
1666 ofproto->backer->need_revalidate = REV_RECONFIGURE;
1667 ofport->cfm = cfm_create(ofport->up.netdev);
1670 if (cfm_configure(ofport->cfm, s)) {
1676 cfm_destroy(ofport->cfm);
1682 get_cfm_status(const struct ofport *ofport_,
1683 struct ofproto_cfm_status *status)
1685 struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
1688 status->faults = cfm_get_fault(ofport->cfm);
1689 status->remote_opstate = cfm_get_opup(ofport->cfm);
1690 status->health = cfm_get_health(ofport->cfm);
1691 cfm_get_remote_mpids(ofport->cfm, &status->rmps, &status->n_rmps);
1699 set_bfd(struct ofport *ofport_, const struct smap *cfg)
1701 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport_->ofproto);
1702 struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
1706 ofport->bfd = bfd_configure(old, netdev_get_name(ofport->up.netdev), cfg);
1707 if (ofport->bfd != old) {
1708 ofproto->backer->need_revalidate = REV_RECONFIGURE;
1715 get_bfd_status(struct ofport *ofport_, struct smap *smap)
1717 struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
1720 bfd_get_status(ofport->bfd, smap);
1727 /* Spanning Tree. */
1730 send_bpdu_cb(struct ofpbuf *pkt, int port_num, void *ofproto_)
1732 struct ofproto_dpif *ofproto = ofproto_;
1733 struct stp_port *sp = stp_get_port(ofproto->stp, port_num);
1734 struct ofport_dpif *ofport;
1736 ofport = stp_port_get_aux(sp);
1738 VLOG_WARN_RL(&rl, "%s: cannot send BPDU on unknown port %d",
1739 ofproto->up.name, port_num);
1741 struct eth_header *eth = pkt->l2;
1743 netdev_get_etheraddr(ofport->up.netdev, eth->eth_src);
1744 if (eth_addr_is_zero(eth->eth_src)) {
1745 VLOG_WARN_RL(&rl, "%s: cannot send BPDU on port %d "
1746 "with unknown MAC", ofproto->up.name, port_num);
1748 send_packet(ofport, pkt);
1754 /* Configures STP on 'ofproto_' using the settings defined in 's'. */
1756 set_stp(struct ofproto *ofproto_, const struct ofproto_stp_settings *s)
1758 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1760 /* Only revalidate flows if the configuration changed. */
1761 if (!s != !ofproto->stp) {
1762 ofproto->backer->need_revalidate = REV_RECONFIGURE;
1766 if (!ofproto->stp) {
1767 ofproto->stp = stp_create(ofproto_->name, s->system_id,
1768 send_bpdu_cb, ofproto);
1769 ofproto->stp_last_tick = time_msec();
1772 stp_set_bridge_id(ofproto->stp, s->system_id);
1773 stp_set_bridge_priority(ofproto->stp, s->priority);
1774 stp_set_hello_time(ofproto->stp, s->hello_time);
1775 stp_set_max_age(ofproto->stp, s->max_age);
1776 stp_set_forward_delay(ofproto->stp, s->fwd_delay);
1778 struct ofport *ofport;
1780 HMAP_FOR_EACH (ofport, hmap_node, &ofproto->up.ports) {
1781 set_stp_port(ofport, NULL);
1784 stp_destroy(ofproto->stp);
1785 ofproto->stp = NULL;
1792 get_stp_status(struct ofproto *ofproto_, struct ofproto_stp_status *s)
1794 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1798 s->bridge_id = stp_get_bridge_id(ofproto->stp);
1799 s->designated_root = stp_get_designated_root(ofproto->stp);
1800 s->root_path_cost = stp_get_root_path_cost(ofproto->stp);
1809 update_stp_port_state(struct ofport_dpif *ofport)
1811 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport->up.ofproto);
1812 enum stp_state state;
1814 /* Figure out new state. */
1815 state = ofport->stp_port ? stp_port_get_state(ofport->stp_port)
1819 if (ofport->stp_state != state) {
1820 enum ofputil_port_state of_state;
1823 VLOG_DBG_RL(&rl, "port %s: STP state changed from %s to %s",
1824 netdev_get_name(ofport->up.netdev),
1825 stp_state_name(ofport->stp_state),
1826 stp_state_name(state));
1827 if (stp_learn_in_state(ofport->stp_state)
1828 != stp_learn_in_state(state)) {
1829 /* xxx Learning action flows should also be flushed. */
1830 mac_learning_flush(ofproto->ml,
1831 &ofproto->backer->revalidate_set);
1833 fwd_change = stp_forward_in_state(ofport->stp_state)
1834 != stp_forward_in_state(state);
1836 ofproto->backer->need_revalidate = REV_STP;
1837 ofport->stp_state = state;
1838 ofport->stp_state_entered = time_msec();
1840 if (fwd_change && ofport->bundle) {
1841 bundle_update(ofport->bundle);
1844 /* Update the STP state bits in the OpenFlow port description. */
1845 of_state = ofport->up.pp.state & ~OFPUTIL_PS_STP_MASK;
1846 of_state |= (state == STP_LISTENING ? OFPUTIL_PS_STP_LISTEN
1847 : state == STP_LEARNING ? OFPUTIL_PS_STP_LEARN
1848 : state == STP_FORWARDING ? OFPUTIL_PS_STP_FORWARD
1849 : state == STP_BLOCKING ? OFPUTIL_PS_STP_BLOCK
1851 ofproto_port_set_state(&ofport->up, of_state);
1855 /* Configures STP on 'ofport_' using the settings defined in 's'. The
1856 * caller is responsible for assigning STP port numbers and ensuring
1857 * there are no duplicates. */
1859 set_stp_port(struct ofport *ofport_,
1860 const struct ofproto_port_stp_settings *s)
1862 struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
1863 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport->up.ofproto);
1864 struct stp_port *sp = ofport->stp_port;
1866 if (!s || !s->enable) {
1868 ofport->stp_port = NULL;
1869 stp_port_disable(sp);
1870 update_stp_port_state(ofport);
1873 } else if (sp && stp_port_no(sp) != s->port_num
1874 && ofport == stp_port_get_aux(sp)) {
1875 /* The port-id changed, so disable the old one if it's not
1876 * already in use by another port. */
1877 stp_port_disable(sp);
1880 sp = ofport->stp_port = stp_get_port(ofproto->stp, s->port_num);
1881 stp_port_enable(sp);
1883 stp_port_set_aux(sp, ofport);
1884 stp_port_set_priority(sp, s->priority);
1885 stp_port_set_path_cost(sp, s->path_cost);
1887 update_stp_port_state(ofport);
1893 get_stp_port_status(struct ofport *ofport_,
1894 struct ofproto_port_stp_status *s)
1896 struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
1897 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport->up.ofproto);
1898 struct stp_port *sp = ofport->stp_port;
1900 if (!ofproto->stp || !sp) {
1906 s->port_id = stp_port_get_id(sp);
1907 s->state = stp_port_get_state(sp);
1908 s->sec_in_state = (time_msec() - ofport->stp_state_entered) / 1000;
1909 s->role = stp_port_get_role(sp);
1910 stp_port_get_counts(sp, &s->tx_count, &s->rx_count, &s->error_count);
1916 stp_run(struct ofproto_dpif *ofproto)
1919 long long int now = time_msec();
1920 long long int elapsed = now - ofproto->stp_last_tick;
1921 struct stp_port *sp;
1924 stp_tick(ofproto->stp, MIN(INT_MAX, elapsed));
1925 ofproto->stp_last_tick = now;
1927 while (stp_get_changed_port(ofproto->stp, &sp)) {
1928 struct ofport_dpif *ofport = stp_port_get_aux(sp);
1931 update_stp_port_state(ofport);
1935 if (stp_check_and_reset_fdb_flush(ofproto->stp)) {
1936 mac_learning_flush(ofproto->ml, &ofproto->backer->revalidate_set);
1942 stp_wait(struct ofproto_dpif *ofproto)
1945 poll_timer_wait(1000);
1949 /* Returns true if STP should process 'flow'. Sets fields in 'wc' that
1950 * were used to make the determination.*/
1952 stp_should_process_flow(const struct flow *flow, struct flow_wildcards *wc)
1954 memset(&wc->masks.dl_dst, 0xff, sizeof wc->masks.dl_dst);
1955 return eth_addr_equals(flow->dl_dst, eth_addr_stp);
1959 stp_process_packet(const struct ofport_dpif *ofport,
1960 const struct ofpbuf *packet)
1962 struct ofpbuf payload = *packet;
1963 struct eth_header *eth = payload.data;
1964 struct stp_port *sp = ofport->stp_port;
1966 /* Sink packets on ports that have STP disabled when the bridge has
1968 if (!sp || stp_port_get_state(sp) == STP_DISABLED) {
1972 /* Trim off padding on payload. */
1973 if (payload.size > ntohs(eth->eth_type) + ETH_HEADER_LEN) {
1974 payload.size = ntohs(eth->eth_type) + ETH_HEADER_LEN;
1977 if (ofpbuf_try_pull(&payload, ETH_HEADER_LEN + LLC_HEADER_LEN)) {
1978 stp_received_bpdu(sp, payload.data, payload.size);
1983 ofproto_dpif_queue_to_priority(const struct ofproto_dpif *ofproto,
1984 uint32_t queue_id, uint32_t *priority)
1986 return dpif_queue_to_priority(ofproto->backer->dpif, queue_id, priority);
1989 static struct priority_to_dscp *
1990 get_priority(const struct ofport_dpif *ofport, uint32_t priority)
1992 struct priority_to_dscp *pdscp;
1995 hash = hash_int(priority, 0);
1996 HMAP_FOR_EACH_IN_BUCKET (pdscp, hmap_node, hash, &ofport->priorities) {
1997 if (pdscp->priority == priority) {
2005 ofproto_dpif_dscp_from_priority(const struct ofport_dpif *ofport,
2006 uint32_t priority, uint8_t *dscp)
2008 struct priority_to_dscp *pdscp = get_priority(ofport, priority);
2009 *dscp = pdscp ? pdscp->dscp : 0;
2010 return pdscp != NULL;
2014 ofport_clear_priorities(struct ofport_dpif *ofport)
2016 struct priority_to_dscp *pdscp, *next;
2018 HMAP_FOR_EACH_SAFE (pdscp, next, hmap_node, &ofport->priorities) {
2019 hmap_remove(&ofport->priorities, &pdscp->hmap_node);
2025 set_queues(struct ofport *ofport_,
2026 const struct ofproto_port_queue *qdscp_list,
2029 struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
2030 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport->up.ofproto);
2031 struct hmap new = HMAP_INITIALIZER(&new);
2034 for (i = 0; i < n_qdscp; i++) {
2035 struct priority_to_dscp *pdscp;
2039 dscp = (qdscp_list[i].dscp << 2) & IP_DSCP_MASK;
2040 if (dpif_queue_to_priority(ofproto->backer->dpif, qdscp_list[i].queue,
2045 pdscp = get_priority(ofport, priority);
2047 hmap_remove(&ofport->priorities, &pdscp->hmap_node);
2049 pdscp = xmalloc(sizeof *pdscp);
2050 pdscp->priority = priority;
2052 ofproto->backer->need_revalidate = REV_RECONFIGURE;
2055 if (pdscp->dscp != dscp) {
2057 ofproto->backer->need_revalidate = REV_RECONFIGURE;
2060 hmap_insert(&new, &pdscp->hmap_node, hash_int(pdscp->priority, 0));
2063 if (!hmap_is_empty(&ofport->priorities)) {
2064 ofport_clear_priorities(ofport);
2065 ofproto->backer->need_revalidate = REV_RECONFIGURE;
2068 hmap_swap(&new, &ofport->priorities);
2076 /* Expires all MAC learning entries associated with 'bundle' and forces its
2077 * ofproto to revalidate every flow.
2079 * Normally MAC learning entries are removed only from the ofproto associated
2080 * with 'bundle', but if 'all_ofprotos' is true, then the MAC learning entries
2081 * are removed from every ofproto. When patch ports and SLB bonds are in use
2082 * and a VM migration happens and the gratuitous ARPs are somehow lost, this
2083 * avoids a MAC_ENTRY_IDLE_TIME delay before the migrated VM can communicate
2084 * with the host from which it migrated. */
2086 bundle_flush_macs(struct ofbundle *bundle, bool all_ofprotos)
2088 struct ofproto_dpif *ofproto = bundle->ofproto;
2089 struct mac_learning *ml = ofproto->ml;
2090 struct mac_entry *mac, *next_mac;
2092 ofproto->backer->need_revalidate = REV_RECONFIGURE;
2093 LIST_FOR_EACH_SAFE (mac, next_mac, lru_node, &ml->lrus) {
2094 if (mac->port.p == bundle) {
2096 struct ofproto_dpif *o;
2098 HMAP_FOR_EACH (o, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
2100 struct mac_entry *e;
2102 e = mac_learning_lookup(o->ml, mac->mac, mac->vlan,
2105 mac_learning_expire(o->ml, e);
2111 mac_learning_expire(ml, mac);
2116 static struct ofbundle *
2117 bundle_lookup(const struct ofproto_dpif *ofproto, void *aux)
2119 struct ofbundle *bundle;
2121 HMAP_FOR_EACH_IN_BUCKET (bundle, hmap_node, hash_pointer(aux, 0),
2122 &ofproto->bundles) {
2123 if (bundle->aux == aux) {
2130 /* Looks up each of the 'n_auxes' pointers in 'auxes' as bundles and adds the
2131 * ones that are found to 'bundles'. */
2133 bundle_lookup_multiple(struct ofproto_dpif *ofproto,
2134 void **auxes, size_t n_auxes,
2135 struct hmapx *bundles)
2139 hmapx_init(bundles);
2140 for (i = 0; i < n_auxes; i++) {
2141 struct ofbundle *bundle = bundle_lookup(ofproto, auxes[i]);
2143 hmapx_add(bundles, bundle);
2149 bundle_update(struct ofbundle *bundle)
2151 struct ofport_dpif *port;
2153 bundle->floodable = true;
2154 LIST_FOR_EACH (port, bundle_node, &bundle->ports) {
2155 if (port->up.pp.config & OFPUTIL_PC_NO_FLOOD
2156 || !stp_forward_in_state(port->stp_state)) {
2157 bundle->floodable = false;
2164 bundle_del_port(struct ofport_dpif *port)
2166 struct ofbundle *bundle = port->bundle;
2168 bundle->ofproto->backer->need_revalidate = REV_RECONFIGURE;
2170 list_remove(&port->bundle_node);
2171 port->bundle = NULL;
2174 lacp_slave_unregister(bundle->lacp, port);
2177 bond_slave_unregister(bundle->bond, port);
2180 bundle_update(bundle);
2184 bundle_add_port(struct ofbundle *bundle, ofp_port_t ofp_port,
2185 struct lacp_slave_settings *lacp)
2187 struct ofport_dpif *port;
2189 port = get_ofp_port(bundle->ofproto, ofp_port);
2194 if (port->bundle != bundle) {
2195 bundle->ofproto->backer->need_revalidate = REV_RECONFIGURE;
2197 bundle_del_port(port);
2200 port->bundle = bundle;
2201 list_push_back(&bundle->ports, &port->bundle_node);
2202 if (port->up.pp.config & OFPUTIL_PC_NO_FLOOD
2203 || !stp_forward_in_state(port->stp_state)) {
2204 bundle->floodable = false;
2208 bundle->ofproto->backer->need_revalidate = REV_RECONFIGURE;
2209 lacp_slave_register(bundle->lacp, port, lacp);
2216 bundle_destroy(struct ofbundle *bundle)
2218 struct ofproto_dpif *ofproto;
2219 struct ofport_dpif *port, *next_port;
2226 ofproto = bundle->ofproto;
2227 for (i = 0; i < MAX_MIRRORS; i++) {
2228 struct ofmirror *m = ofproto->mirrors[i];
2230 if (m->out == bundle) {
2232 } else if (hmapx_find_and_delete(&m->srcs, bundle)
2233 || hmapx_find_and_delete(&m->dsts, bundle)) {
2234 ofproto->backer->need_revalidate = REV_RECONFIGURE;
2239 LIST_FOR_EACH_SAFE (port, next_port, bundle_node, &bundle->ports) {
2240 bundle_del_port(port);
2243 bundle_flush_macs(bundle, true);
2244 hmap_remove(&ofproto->bundles, &bundle->hmap_node);
2246 free(bundle->trunks);
2247 lacp_destroy(bundle->lacp);
2248 bond_destroy(bundle->bond);
2253 bundle_set(struct ofproto *ofproto_, void *aux,
2254 const struct ofproto_bundle_settings *s)
2256 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
2257 bool need_flush = false;
2258 struct ofport_dpif *port;
2259 struct ofbundle *bundle;
2260 unsigned long *trunks;
2266 bundle_destroy(bundle_lookup(ofproto, aux));
2270 ovs_assert(s->n_slaves == 1 || s->bond != NULL);
2271 ovs_assert((s->lacp != NULL) == (s->lacp_slaves != NULL));
2273 bundle = bundle_lookup(ofproto, aux);
2275 bundle = xmalloc(sizeof *bundle);
2277 bundle->ofproto = ofproto;
2278 hmap_insert(&ofproto->bundles, &bundle->hmap_node,
2279 hash_pointer(aux, 0));
2281 bundle->name = NULL;
2283 list_init(&bundle->ports);
2284 bundle->vlan_mode = PORT_VLAN_TRUNK;
2286 bundle->trunks = NULL;
2287 bundle->use_priority_tags = s->use_priority_tags;
2288 bundle->lacp = NULL;
2289 bundle->bond = NULL;
2291 bundle->floodable = true;
2293 bundle->src_mirrors = 0;
2294 bundle->dst_mirrors = 0;
2295 bundle->mirror_out = 0;
2298 if (!bundle->name || strcmp(s->name, bundle->name)) {
2300 bundle->name = xstrdup(s->name);
2305 if (!bundle->lacp) {
2306 ofproto->backer->need_revalidate = REV_RECONFIGURE;
2307 bundle->lacp = lacp_create();
2309 lacp_configure(bundle->lacp, s->lacp);
2311 lacp_destroy(bundle->lacp);
2312 bundle->lacp = NULL;
2315 /* Update set of ports. */
2317 for (i = 0; i < s->n_slaves; i++) {
2318 if (!bundle_add_port(bundle, s->slaves[i],
2319 s->lacp ? &s->lacp_slaves[i] : NULL)) {
2323 if (!ok || list_size(&bundle->ports) != s->n_slaves) {
2324 struct ofport_dpif *next_port;
2326 LIST_FOR_EACH_SAFE (port, next_port, bundle_node, &bundle->ports) {
2327 for (i = 0; i < s->n_slaves; i++) {
2328 if (s->slaves[i] == port->up.ofp_port) {
2333 bundle_del_port(port);
2337 ovs_assert(list_size(&bundle->ports) <= s->n_slaves);
2339 if (list_is_empty(&bundle->ports)) {
2340 bundle_destroy(bundle);
2344 /* Set VLAN tagging mode */
2345 if (s->vlan_mode != bundle->vlan_mode
2346 || s->use_priority_tags != bundle->use_priority_tags) {
2347 bundle->vlan_mode = s->vlan_mode;
2348 bundle->use_priority_tags = s->use_priority_tags;
2353 vlan = (s->vlan_mode == PORT_VLAN_TRUNK ? -1
2354 : s->vlan >= 0 && s->vlan <= 4095 ? s->vlan
2356 if (vlan != bundle->vlan) {
2357 bundle->vlan = vlan;
2361 /* Get trunked VLANs. */
2362 switch (s->vlan_mode) {
2363 case PORT_VLAN_ACCESS:
2367 case PORT_VLAN_TRUNK:
2368 trunks = CONST_CAST(unsigned long *, s->trunks);
2371 case PORT_VLAN_NATIVE_UNTAGGED:
2372 case PORT_VLAN_NATIVE_TAGGED:
2373 if (vlan != 0 && (!s->trunks
2374 || !bitmap_is_set(s->trunks, vlan)
2375 || bitmap_is_set(s->trunks, 0))) {
2376 /* Force trunking the native VLAN and prohibit trunking VLAN 0. */
2378 trunks = bitmap_clone(s->trunks, 4096);
2380 trunks = bitmap_allocate1(4096);
2382 bitmap_set1(trunks, vlan);
2383 bitmap_set0(trunks, 0);
2385 trunks = CONST_CAST(unsigned long *, s->trunks);
2392 if (!vlan_bitmap_equal(trunks, bundle->trunks)) {
2393 free(bundle->trunks);
2394 if (trunks == s->trunks) {
2395 bundle->trunks = vlan_bitmap_clone(trunks);
2397 bundle->trunks = trunks;
2402 if (trunks != s->trunks) {
2407 if (!list_is_short(&bundle->ports)) {
2408 bundle->ofproto->has_bonded_bundles = true;
2410 if (bond_reconfigure(bundle->bond, s->bond)) {
2411 ofproto->backer->need_revalidate = REV_RECONFIGURE;
2414 bundle->bond = bond_create(s->bond);
2415 ofproto->backer->need_revalidate = REV_RECONFIGURE;
2418 LIST_FOR_EACH (port, bundle_node, &bundle->ports) {
2419 bond_slave_register(bundle->bond, port, port->up.netdev);
2422 bond_destroy(bundle->bond);
2423 bundle->bond = NULL;
2426 /* If we changed something that would affect MAC learning, un-learn
2427 * everything on this port and force flow revalidation. */
2429 bundle_flush_macs(bundle, false);
2436 bundle_remove(struct ofport *port_)
2438 struct ofport_dpif *port = ofport_dpif_cast(port_);
2439 struct ofbundle *bundle = port->bundle;
2442 bundle_del_port(port);
2443 if (list_is_empty(&bundle->ports)) {
2444 bundle_destroy(bundle);
2445 } else if (list_is_short(&bundle->ports)) {
2446 bond_destroy(bundle->bond);
2447 bundle->bond = NULL;
2453 send_pdu_cb(void *port_, const void *pdu, size_t pdu_size)
2455 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 10);
2456 struct ofport_dpif *port = port_;
2457 uint8_t ea[ETH_ADDR_LEN];
2460 error = netdev_get_etheraddr(port->up.netdev, ea);
2462 struct ofpbuf packet;
2465 ofpbuf_init(&packet, 0);
2466 packet_pdu = eth_compose(&packet, eth_addr_lacp, ea, ETH_TYPE_LACP,
2468 memcpy(packet_pdu, pdu, pdu_size);
2470 send_packet(port, &packet);
2471 ofpbuf_uninit(&packet);
2473 VLOG_ERR_RL(&rl, "port %s: cannot obtain Ethernet address of iface "
2474 "%s (%s)", port->bundle->name,
2475 netdev_get_name(port->up.netdev), strerror(error));
2480 bundle_send_learning_packets(struct ofbundle *bundle)
2482 struct ofproto_dpif *ofproto = bundle->ofproto;
2483 int error, n_packets, n_errors;
2484 struct mac_entry *e;
2486 error = n_packets = n_errors = 0;
2487 LIST_FOR_EACH (e, lru_node, &ofproto->ml->lrus) {
2488 if (e->port.p != bundle) {
2489 struct ofpbuf *learning_packet;
2490 struct ofport_dpif *port;
2494 /* The assignment to "port" is unnecessary but makes "grep"ing for
2495 * struct ofport_dpif more effective. */
2496 learning_packet = bond_compose_learning_packet(bundle->bond,
2500 ret = send_packet(port, learning_packet);
2501 ofpbuf_delete(learning_packet);
2511 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
2512 VLOG_WARN_RL(&rl, "bond %s: %d errors sending %d gratuitous learning "
2513 "packets, last error was: %s",
2514 bundle->name, n_errors, n_packets, strerror(error));
2516 VLOG_DBG("bond %s: sent %d gratuitous learning packets",
2517 bundle->name, n_packets);
2522 bundle_run(struct ofbundle *bundle)
2525 lacp_run(bundle->lacp, send_pdu_cb);
2528 struct ofport_dpif *port;
2530 LIST_FOR_EACH (port, bundle_node, &bundle->ports) {
2531 bond_slave_set_may_enable(bundle->bond, port, port->may_enable);
2534 bond_run(bundle->bond, &bundle->ofproto->backer->revalidate_set,
2535 lacp_status(bundle->lacp));
2536 if (bond_should_send_learning_packets(bundle->bond)) {
2537 bundle_send_learning_packets(bundle);
2543 bundle_wait(struct ofbundle *bundle)
2546 lacp_wait(bundle->lacp);
2549 bond_wait(bundle->bond);
2556 mirror_scan(struct ofproto_dpif *ofproto)
2560 for (idx = 0; idx < MAX_MIRRORS; idx++) {
2561 if (!ofproto->mirrors[idx]) {
2568 static struct ofmirror *
2569 mirror_lookup(struct ofproto_dpif *ofproto, void *aux)
2573 for (i = 0; i < MAX_MIRRORS; i++) {
2574 struct ofmirror *mirror = ofproto->mirrors[i];
2575 if (mirror && mirror->aux == aux) {
2583 /* Update the 'dup_mirrors' member of each of the ofmirrors in 'ofproto'. */
2585 mirror_update_dups(struct ofproto_dpif *ofproto)
2589 for (i = 0; i < MAX_MIRRORS; i++) {
2590 struct ofmirror *m = ofproto->mirrors[i];
2593 m->dup_mirrors = MIRROR_MASK_C(1) << i;
2597 for (i = 0; i < MAX_MIRRORS; i++) {
2598 struct ofmirror *m1 = ofproto->mirrors[i];
2605 for (j = i + 1; j < MAX_MIRRORS; j++) {
2606 struct ofmirror *m2 = ofproto->mirrors[j];
2608 if (m2 && m1->out == m2->out && m1->out_vlan == m2->out_vlan) {
2609 m1->dup_mirrors |= MIRROR_MASK_C(1) << j;
2610 m2->dup_mirrors |= m1->dup_mirrors;
2617 mirror_set(struct ofproto *ofproto_, void *aux,
2618 const struct ofproto_mirror_settings *s)
2620 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
2621 mirror_mask_t mirror_bit;
2622 struct ofbundle *bundle;
2623 struct ofmirror *mirror;
2624 struct ofbundle *out;
2625 struct hmapx srcs; /* Contains "struct ofbundle *"s. */
2626 struct hmapx dsts; /* Contains "struct ofbundle *"s. */
2629 mirror = mirror_lookup(ofproto, aux);
2631 mirror_destroy(mirror);
2637 idx = mirror_scan(ofproto);
2639 VLOG_WARN("bridge %s: maximum of %d port mirrors reached, "
2641 ofproto->up.name, MAX_MIRRORS, s->name);
2645 mirror = ofproto->mirrors[idx] = xzalloc(sizeof *mirror);
2646 mirror->ofproto = ofproto;
2649 mirror->out_vlan = -1;
2650 mirror->name = NULL;
2653 if (!mirror->name || strcmp(s->name, mirror->name)) {
2655 mirror->name = xstrdup(s->name);
2658 /* Get the new configuration. */
2659 if (s->out_bundle) {
2660 out = bundle_lookup(ofproto, s->out_bundle);
2662 mirror_destroy(mirror);
2668 out_vlan = s->out_vlan;
2670 bundle_lookup_multiple(ofproto, s->srcs, s->n_srcs, &srcs);
2671 bundle_lookup_multiple(ofproto, s->dsts, s->n_dsts, &dsts);
2673 /* If the configuration has not changed, do nothing. */
2674 if (hmapx_equals(&srcs, &mirror->srcs)
2675 && hmapx_equals(&dsts, &mirror->dsts)
2676 && vlan_bitmap_equal(mirror->vlans, s->src_vlans)
2677 && mirror->out == out
2678 && mirror->out_vlan == out_vlan)
2680 hmapx_destroy(&srcs);
2681 hmapx_destroy(&dsts);
2685 hmapx_swap(&srcs, &mirror->srcs);
2686 hmapx_destroy(&srcs);
2688 hmapx_swap(&dsts, &mirror->dsts);
2689 hmapx_destroy(&dsts);
2691 free(mirror->vlans);
2692 mirror->vlans = vlan_bitmap_clone(s->src_vlans);
2695 mirror->out_vlan = out_vlan;
2697 /* Update bundles. */
2698 mirror_bit = MIRROR_MASK_C(1) << mirror->idx;
2699 HMAP_FOR_EACH (bundle, hmap_node, &mirror->ofproto->bundles) {
2700 if (hmapx_contains(&mirror->srcs, bundle)) {
2701 bundle->src_mirrors |= mirror_bit;
2703 bundle->src_mirrors &= ~mirror_bit;
2706 if (hmapx_contains(&mirror->dsts, bundle)) {
2707 bundle->dst_mirrors |= mirror_bit;
2709 bundle->dst_mirrors &= ~mirror_bit;
2712 if (mirror->out == bundle) {
2713 bundle->mirror_out |= mirror_bit;
2715 bundle->mirror_out &= ~mirror_bit;
2719 ofproto->backer->need_revalidate = REV_RECONFIGURE;
2720 ofproto->has_mirrors = true;
2721 mac_learning_flush(ofproto->ml,
2722 &ofproto->backer->revalidate_set);
2723 mirror_update_dups(ofproto);
2729 mirror_destroy(struct ofmirror *mirror)
2731 struct ofproto_dpif *ofproto;
2732 mirror_mask_t mirror_bit;
2733 struct ofbundle *bundle;
2740 ofproto = mirror->ofproto;
2741 ofproto->backer->need_revalidate = REV_RECONFIGURE;
2742 mac_learning_flush(ofproto->ml, &ofproto->backer->revalidate_set);
2744 mirror_bit = MIRROR_MASK_C(1) << mirror->idx;
2745 HMAP_FOR_EACH (bundle, hmap_node, &ofproto->bundles) {
2746 bundle->src_mirrors &= ~mirror_bit;
2747 bundle->dst_mirrors &= ~mirror_bit;
2748 bundle->mirror_out &= ~mirror_bit;
2751 hmapx_destroy(&mirror->srcs);
2752 hmapx_destroy(&mirror->dsts);
2753 free(mirror->vlans);
2755 ofproto->mirrors[mirror->idx] = NULL;
2759 mirror_update_dups(ofproto);
2761 ofproto->has_mirrors = false;
2762 for (i = 0; i < MAX_MIRRORS; i++) {
2763 if (ofproto->mirrors[i]) {
2764 ofproto->has_mirrors = true;
2771 mirror_get_stats(struct ofproto *ofproto_, void *aux,
2772 uint64_t *packets, uint64_t *bytes)
2774 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
2775 struct ofmirror *mirror = mirror_lookup(ofproto, aux);
2778 *packets = *bytes = UINT64_MAX;
2784 *packets = mirror->packet_count;
2785 *bytes = mirror->byte_count;
2791 set_flood_vlans(struct ofproto *ofproto_, unsigned long *flood_vlans)
2793 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
2794 if (mac_learning_set_flood_vlans(ofproto->ml, flood_vlans)) {
2795 mac_learning_flush(ofproto->ml, &ofproto->backer->revalidate_set);
2801 is_mirror_output_bundle(const struct ofproto *ofproto_, void *aux)
2803 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
2804 struct ofbundle *bundle = bundle_lookup(ofproto, aux);
2805 return bundle && bundle->mirror_out != 0;
2809 forward_bpdu_changed(struct ofproto *ofproto_)
2811 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
2812 ofproto->backer->need_revalidate = REV_RECONFIGURE;
2816 set_mac_table_config(struct ofproto *ofproto_, unsigned int idle_time,
2819 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
2820 mac_learning_set_idle_time(ofproto->ml, idle_time);
2821 mac_learning_set_max_entries(ofproto->ml, max_entries);
2826 struct ofport_dpif *
2827 get_ofp_port(const struct ofproto_dpif *ofproto, ofp_port_t ofp_port)
2829 struct ofport *ofport = ofproto_get_port(&ofproto->up, ofp_port);
2830 return ofport ? ofport_dpif_cast(ofport) : NULL;
2833 struct ofport_dpif *
2834 get_odp_port(const struct ofproto_dpif *ofproto, odp_port_t odp_port)
2836 struct ofport_dpif *port = odp_port_to_ofport(ofproto->backer, odp_port);
2837 return port && &ofproto->up == port->up.ofproto ? port : NULL;
2841 ofproto_port_from_dpif_port(struct ofproto_dpif *ofproto,
2842 struct ofproto_port *ofproto_port,
2843 struct dpif_port *dpif_port)
2845 ofproto_port->name = dpif_port->name;
2846 ofproto_port->type = dpif_port->type;
2847 ofproto_port->ofp_port = odp_port_to_ofp_port(ofproto, dpif_port->port_no);
2851 ofport_update_peer(struct ofport_dpif *ofport)
2853 const struct ofproto_dpif *ofproto;
2854 struct dpif_backer *backer;
2855 const char *peer_name;
2857 if (!netdev_vport_is_patch(ofport->up.netdev)) {
2861 backer = ofproto_dpif_cast(ofport->up.ofproto)->backer;
2862 backer->need_revalidate = true;
2865 ofport->peer->peer = NULL;
2866 ofport->peer = NULL;
2869 peer_name = netdev_vport_patch_peer(ofport->up.netdev);
2874 HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
2875 struct ofport *peer_ofport;
2876 struct ofport_dpif *peer;
2877 const char *peer_peer;
2879 if (ofproto->backer != backer) {
2883 peer_ofport = shash_find_data(&ofproto->up.port_by_name, peer_name);
2888 peer = ofport_dpif_cast(peer_ofport);
2889 peer_peer = netdev_vport_patch_peer(peer->up.netdev);
2890 if (peer_peer && !strcmp(netdev_get_name(ofport->up.netdev),
2892 ofport->peer = peer;
2893 ofport->peer->peer = ofport;
2901 port_run_fast(struct ofport_dpif *ofport)
2903 if (ofport->cfm && cfm_should_send_ccm(ofport->cfm)) {
2904 struct ofpbuf packet;
2906 ofpbuf_init(&packet, 0);
2907 cfm_compose_ccm(ofport->cfm, &packet, ofport->up.pp.hw_addr);
2908 send_packet(ofport, &packet);
2909 ofpbuf_uninit(&packet);
2912 if (ofport->bfd && bfd_should_send_packet(ofport->bfd)) {
2913 struct ofpbuf packet;
2915 ofpbuf_init(&packet, 0);
2916 bfd_put_packet(ofport->bfd, &packet, ofport->up.pp.hw_addr);
2917 send_packet(ofport, &packet);
2918 ofpbuf_uninit(&packet);
2923 port_run(struct ofport_dpif *ofport)
2925 long long int carrier_seq = netdev_get_carrier_resets(ofport->up.netdev);
2926 bool carrier_changed = carrier_seq != ofport->carrier_seq;
2927 bool enable = netdev_get_carrier(ofport->up.netdev);
2929 ofport->carrier_seq = carrier_seq;
2931 port_run_fast(ofport);
2934 int cfm_opup = cfm_get_opup(ofport->cfm);
2936 cfm_run(ofport->cfm);
2937 enable = enable && !cfm_get_fault(ofport->cfm);
2939 if (cfm_opup >= 0) {
2940 enable = enable && cfm_opup;
2945 bfd_run(ofport->bfd);
2946 enable = enable && bfd_forwarding(ofport->bfd);
2949 if (ofport->bundle) {
2950 enable = enable && lacp_slave_may_enable(ofport->bundle->lacp, ofport);
2951 if (carrier_changed) {
2952 lacp_slave_carrier_changed(ofport->bundle->lacp, ofport);
2956 if (ofport->may_enable != enable) {
2957 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport->up.ofproto);
2959 if (ofproto->has_bundle_action) {
2960 ofproto->backer->need_revalidate = REV_PORT_TOGGLED;
2964 ofport->may_enable = enable;
2968 port_wait(struct ofport_dpif *ofport)
2971 cfm_wait(ofport->cfm);
2975 bfd_wait(ofport->bfd);
2980 port_query_by_name(const struct ofproto *ofproto_, const char *devname,
2981 struct ofproto_port *ofproto_port)
2983 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
2984 struct dpif_port dpif_port;
2987 if (sset_contains(&ofproto->ghost_ports, devname)) {
2988 const char *type = netdev_get_type_from_name(devname);
2990 /* We may be called before ofproto->up.port_by_name is populated with
2991 * the appropriate ofport. For this reason, we must get the name and
2992 * type from the netdev layer directly. */
2994 const struct ofport *ofport;
2996 ofport = shash_find_data(&ofproto->up.port_by_name, devname);
2997 ofproto_port->ofp_port = ofport ? ofport->ofp_port : OFPP_NONE;
2998 ofproto_port->name = xstrdup(devname);
2999 ofproto_port->type = xstrdup(type);
3005 if (!sset_contains(&ofproto->ports, devname)) {
3008 error = dpif_port_query_by_name(ofproto->backer->dpif,
3009 devname, &dpif_port);
3011 ofproto_port_from_dpif_port(ofproto, ofproto_port, &dpif_port);
3017 port_add(struct ofproto *ofproto_, struct netdev *netdev)
3019 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
3020 const char *devname = netdev_get_name(netdev);
3021 char namebuf[NETDEV_VPORT_NAME_BUFSIZE];
3022 const char *dp_port_name;
3024 if (netdev_vport_is_patch(netdev)) {
3025 sset_add(&ofproto->ghost_ports, netdev_get_name(netdev));
3029 dp_port_name = netdev_vport_get_dpif_port(netdev, namebuf, sizeof namebuf);
3030 if (!dpif_port_exists(ofproto->backer->dpif, dp_port_name)) {
3031 odp_port_t port_no = ODPP_NONE;
3034 error = dpif_port_add(ofproto->backer->dpif, netdev, &port_no);
3038 if (netdev_get_tunnel_config(netdev)) {
3039 simap_put(&ofproto->backer->tnl_backers,
3040 dp_port_name, odp_to_u32(port_no));
3044 if (netdev_get_tunnel_config(netdev)) {
3045 sset_add(&ofproto->ghost_ports, devname);
3047 sset_add(&ofproto->ports, devname);
3053 port_del(struct ofproto *ofproto_, ofp_port_t ofp_port)
3055 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
3056 struct ofport_dpif *ofport = get_ofp_port(ofproto, ofp_port);
3063 sset_find_and_delete(&ofproto->ghost_ports,
3064 netdev_get_name(ofport->up.netdev));
3065 ofproto->backer->need_revalidate = REV_RECONFIGURE;
3066 if (!ofport->tnl_port) {
3067 error = dpif_port_del(ofproto->backer->dpif, ofport->odp_port);
3069 /* The caller is going to close ofport->up.netdev. If this is a
3070 * bonded port, then the bond is using that netdev, so remove it
3071 * from the bond. The client will need to reconfigure everything
3072 * after deleting ports, so then the slave will get re-added. */
3073 bundle_remove(&ofport->up);
3080 port_get_stats(const struct ofport *ofport_, struct netdev_stats *stats)
3082 struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
3087 error = netdev_get_stats(ofport->up.netdev, stats);
3089 if (!error && ofport_->ofp_port == OFPP_LOCAL) {
3090 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport->up.ofproto);
3092 /* ofproto->stats.tx_packets represents packets that we created
3093 * internally and sent to some port (e.g. packets sent with
3094 * send_packet()). Account for them as if they had come from
3095 * OFPP_LOCAL and got forwarded. */
3097 if (stats->rx_packets != UINT64_MAX) {
3098 stats->rx_packets += ofproto->stats.tx_packets;
3101 if (stats->rx_bytes != UINT64_MAX) {
3102 stats->rx_bytes += ofproto->stats.tx_bytes;
3105 /* ofproto->stats.rx_packets represents packets that were received on
3106 * some port and we processed internally and dropped (e.g. STP).
3107 * Account for them as if they had been forwarded to OFPP_LOCAL. */
3109 if (stats->tx_packets != UINT64_MAX) {
3110 stats->tx_packets += ofproto->stats.rx_packets;
3113 if (stats->tx_bytes != UINT64_MAX) {
3114 stats->tx_bytes += ofproto->stats.rx_bytes;
3121 struct port_dump_state {
3126 struct ofproto_port port;
3131 port_dump_start(const struct ofproto *ofproto_ OVS_UNUSED, void **statep)
3133 *statep = xzalloc(sizeof(struct port_dump_state));
3138 port_dump_next(const struct ofproto *ofproto_, void *state_,
3139 struct ofproto_port *port)
3141 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
3142 struct port_dump_state *state = state_;
3143 const struct sset *sset;
3144 struct sset_node *node;
3146 if (state->has_port) {
3147 ofproto_port_destroy(&state->port);
3148 state->has_port = false;
3150 sset = state->ghost ? &ofproto->ghost_ports : &ofproto->ports;
3151 while ((node = sset_at_position(sset, &state->bucket, &state->offset))) {
3154 error = port_query_by_name(ofproto_, node->name, &state->port);
3156 *port = state->port;
3157 state->has_port = true;
3159 } else if (error != ENODEV) {
3164 if (!state->ghost) {
3165 state->ghost = true;
3168 return port_dump_next(ofproto_, state_, port);
3175 port_dump_done(const struct ofproto *ofproto_ OVS_UNUSED, void *state_)
3177 struct port_dump_state *state = state_;
3179 if (state->has_port) {
3180 ofproto_port_destroy(&state->port);
3187 port_poll(const struct ofproto *ofproto_, char **devnamep)
3189 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
3191 if (ofproto->port_poll_errno) {
3192 int error = ofproto->port_poll_errno;
3193 ofproto->port_poll_errno = 0;
3197 if (sset_is_empty(&ofproto->port_poll_set)) {
3201 *devnamep = sset_pop(&ofproto->port_poll_set);
3206 port_poll_wait(const struct ofproto *ofproto_)
3208 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
3209 dpif_port_poll_wait(ofproto->backer->dpif);
3213 port_is_lacp_current(const struct ofport *ofport_)
3215 const struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
3216 return (ofport->bundle && ofport->bundle->lacp
3217 ? lacp_slave_is_current(ofport->bundle->lacp, ofport)
3221 /* Upcall handling. */
3223 /* Flow miss batching.
3225 * Some dpifs implement operations faster when you hand them off in a batch.
3226 * To allow batching, "struct flow_miss" queues the dpif-related work needed
3227 * for a given flow. Each "struct flow_miss" corresponds to sending one or
3228 * more packets, plus possibly installing the flow in the dpif.
3230 * So far we only batch the operations that affect flow setup time the most.
3231 * It's possible to batch more than that, but the benefit might be minimal. */
3233 struct hmap_node hmap_node;
3234 struct ofproto_dpif *ofproto;
3236 enum odp_key_fitness key_fitness;
3237 const struct nlattr *key;
3239 struct list packets;
3240 enum dpif_upcall_type upcall_type;
3243 struct flow_miss_op {
3244 struct dpif_op dpif_op;
3246 uint64_t slow_stub[128 / 8]; /* Buffer for compose_slow_path() */
3247 struct xlate_out xout;
3248 bool xout_garbage; /* 'xout' needs to be uninitialized? */
3250 struct ofpbuf mask; /* Flow mask for "put" ops. */
3251 struct odputil_keybuf maskbuf;
3253 /* If this is a "put" op, then a pointer to the subfacet that should
3254 * be marked as uninstalled if the operation fails. */
3255 struct subfacet *subfacet;
3258 /* Sends an OFPT_PACKET_IN message for 'packet' of type OFPR_NO_MATCH to each
3259 * OpenFlow controller as necessary according to their individual
3260 * configurations. */
3262 send_packet_in_miss(struct ofproto_dpif *ofproto, const struct ofpbuf *packet,
3263 const struct flow *flow)
3265 struct ofputil_packet_in pin;
3267 pin.packet = packet->data;
3268 pin.packet_len = packet->size;
3269 pin.reason = OFPR_NO_MATCH;
3270 pin.controller_id = 0;
3275 pin.send_len = 0; /* not used for flow table misses */
3277 flow_get_metadata(flow, &pin.fmd);
3279 connmgr_send_packet_in(ofproto->up.connmgr, &pin);
3282 static struct flow_miss *
3283 flow_miss_find(struct hmap *todo, const struct ofproto_dpif *ofproto,
3284 const struct flow *flow, uint32_t hash)
3286 struct flow_miss *miss;
3288 HMAP_FOR_EACH_WITH_HASH (miss, hmap_node, hash, todo) {
3289 if (miss->ofproto == ofproto && flow_equal(&miss->flow, flow)) {
3297 /* Partially Initializes 'op' as an "execute" operation for 'miss' and
3298 * 'packet'. The caller must initialize op->actions and op->actions_len. If
3299 * 'miss' is associated with a subfacet the caller must also initialize the
3300 * returned op->subfacet, and if anything needs to be freed after processing
3301 * the op, the caller must initialize op->garbage also. */
3303 init_flow_miss_execute_op(struct flow_miss *miss, struct ofpbuf *packet,
3304 struct flow_miss_op *op)
3306 if (miss->flow.in_port.ofp_port
3307 != vsp_realdev_to_vlandev(miss->ofproto, miss->flow.in_port.ofp_port,
3308 miss->flow.vlan_tci)) {
3309 /* This packet was received on a VLAN splinter port. We
3310 * added a VLAN to the packet to make the packet resemble
3311 * the flow, but the actions were composed assuming that
3312 * the packet contained no VLAN. So, we must remove the
3313 * VLAN header from the packet before trying to execute the
3315 eth_pop_vlan(packet);
3318 op->subfacet = NULL;
3319 op->xout_garbage = false;
3320 op->dpif_op.type = DPIF_OP_EXECUTE;
3321 op->dpif_op.u.execute.key = miss->key;
3322 op->dpif_op.u.execute.key_len = miss->key_len;
3323 op->dpif_op.u.execute.packet = packet;
3324 ofpbuf_use_stack(&op->mask, &op->maskbuf, sizeof op->maskbuf);
3327 /* Helper for handle_flow_miss_without_facet() and
3328 * handle_flow_miss_with_facet(). */
3330 handle_flow_miss_common(struct rule_dpif *rule,
3331 struct ofpbuf *packet, const struct flow *flow)
3333 struct ofproto_dpif *ofproto = ofproto_dpif_cast(rule->up.ofproto);
3335 if (rule->up.cr.priority == FAIL_OPEN_PRIORITY) {
3337 * Extra-special case for fail-open mode.
3339 * We are in fail-open mode and the packet matched the fail-open
3340 * rule, but we are connected to a controller too. We should send
3341 * the packet up to the controller in the hope that it will try to
3342 * set up a flow and thereby allow us to exit fail-open.
3344 * See the top-level comment in fail-open.c for more information.
3346 send_packet_in_miss(ofproto, packet, flow);
3350 /* Figures out whether a flow that missed in 'ofproto', whose details are in
3351 * 'miss' masked by 'wc', is likely to be worth tracking in detail in userspace
3352 * and (usually) installing a datapath flow. The answer is usually "yes" (a
3353 * return value of true). However, for short flows the cost of bookkeeping is
3354 * much higher than the benefits, so when the datapath holds a large number of
3355 * flows we impose some heuristics to decide which flows are likely to be worth
3358 flow_miss_should_make_facet(struct flow_miss *miss, struct flow_wildcards *wc)
3360 struct dpif_backer *backer = miss->ofproto->backer;
3363 if (!backer->governor) {
3366 n_subfacets = hmap_count(&backer->subfacets);
3367 if (n_subfacets * 2 <= flow_eviction_threshold) {
3371 backer->governor = governor_create();
3374 hash = flow_hash_in_wildcards(&miss->flow, wc, 0);
3375 return governor_should_install_flow(backer->governor, hash,
3376 list_size(&miss->packets));
3379 /* Handles 'miss' without creating a facet or subfacet or creating any datapath
3380 * flow. 'miss->flow' must have matched 'rule' and been xlated into 'xout'.
3381 * May add an "execute" operation to 'ops' and increment '*n_ops'. */
3383 handle_flow_miss_without_facet(struct rule_dpif *rule, struct xlate_out *xout,
3384 struct flow_miss *miss,
3385 struct flow_miss_op *ops, size_t *n_ops)
3387 struct ofpbuf *packet;
3389 LIST_FOR_EACH (packet, list_node, &miss->packets) {
3391 COVERAGE_INC(facet_suppress);
3393 handle_flow_miss_common(rule, packet, &miss->flow);
3396 struct xlate_in xin;
3398 xlate_in_init(&xin, miss->ofproto, &miss->flow, rule, 0, packet);
3399 xlate_actions_for_side_effects(&xin);
3402 if (xout->odp_actions.size) {
3403 struct flow_miss_op *op = &ops[*n_ops];
3404 struct dpif_execute *execute = &op->dpif_op.u.execute;
3406 init_flow_miss_execute_op(miss, packet, op);
3407 xlate_out_copy(&op->xout, xout);
3408 execute->actions = op->xout.odp_actions.data;
3409 execute->actions_len = op->xout.odp_actions.size;
3410 op->xout_garbage = true;
3417 /* Handles 'miss', which matches 'facet'. May add any required datapath
3418 * operations to 'ops', incrementing '*n_ops' for each new op.
3420 * All of the packets in 'miss' are considered to have arrived at time 'now'.
3421 * This is really important only for new facets: if we just called time_msec()
3422 * here, then the new subfacet or its packets could look (occasionally) as
3423 * though it was used some time after the facet was used. That can make a
3424 * one-packet flow look like it has a nonzero duration, which looks odd in
3425 * e.g. NetFlow statistics.
3427 * If non-null, 'stats' will be folded into 'facet'. */
3429 handle_flow_miss_with_facet(struct flow_miss *miss, struct facet *facet,
3430 long long int now, struct dpif_flow_stats *stats,
3431 struct flow_miss_op *ops, size_t *n_ops)
3433 struct ofproto_dpif *ofproto = ofproto_dpif_cast(facet->rule->up.ofproto);
3434 enum subfacet_path want_path;
3435 struct subfacet *subfacet;
3436 struct ofpbuf *packet;
3438 subfacet = subfacet_create(facet, miss, now);
3439 want_path = facet->xout.slow ? SF_SLOW_PATH : SF_FAST_PATH;
3441 subfacet_update_stats(subfacet, stats);
3444 LIST_FOR_EACH (packet, list_node, &miss->packets) {
3445 struct flow_miss_op *op = &ops[*n_ops];
3447 handle_flow_miss_common(facet->rule, packet, &miss->flow);
3449 if (want_path != SF_FAST_PATH) {
3450 struct xlate_in xin;
3452 xlate_in_init(&xin, ofproto, &miss->flow, facet->rule, 0, packet);
3453 xlate_actions_for_side_effects(&xin);
3456 if (facet->xout.odp_actions.size) {
3457 struct dpif_execute *execute = &op->dpif_op.u.execute;
3459 init_flow_miss_execute_op(miss, packet, op);
3460 execute->actions = facet->xout.odp_actions.data,
3461 execute->actions_len = facet->xout.odp_actions.size;
3466 if (miss->upcall_type == DPIF_UC_MISS || subfacet->path != want_path) {
3467 struct flow_miss_op *op = &ops[(*n_ops)++];
3468 struct dpif_flow_put *put = &op->dpif_op.u.flow_put;
3470 subfacet->path = want_path;
3472 ofpbuf_use_stack(&op->mask, &op->maskbuf, sizeof op->maskbuf);
3473 odp_flow_key_from_mask(&op->mask, &facet->xout.wc.masks,
3474 &miss->flow, UINT32_MAX);
3476 op->xout_garbage = false;
3477 op->dpif_op.type = DPIF_OP_FLOW_PUT;
3478 op->subfacet = subfacet;
3479 put->flags = DPIF_FP_CREATE | DPIF_FP_MODIFY;
3480 put->key = miss->key;
3481 put->key_len = miss->key_len;
3482 put->mask = op->mask.data;
3483 put->mask_len = op->mask.size;
3485 if (want_path == SF_FAST_PATH) {
3486 put->actions = facet->xout.odp_actions.data;
3487 put->actions_len = facet->xout.odp_actions.size;
3489 compose_slow_path(ofproto, &miss->flow, facet->xout.slow,
3490 op->slow_stub, sizeof op->slow_stub,
3491 &put->actions, &put->actions_len);
3497 /* Handles flow miss 'miss'. May add any required datapath operations
3498 * to 'ops', incrementing '*n_ops' for each new op. */
3500 handle_flow_miss(struct flow_miss *miss, struct flow_miss_op *ops,
3503 struct ofproto_dpif *ofproto = miss->ofproto;
3504 struct dpif_flow_stats stats__;
3505 struct dpif_flow_stats *stats = &stats__;
3506 struct ofpbuf *packet;
3507 struct facet *facet;
3511 memset(stats, 0, sizeof *stats);
3513 LIST_FOR_EACH (packet, list_node, &miss->packets) {
3514 stats->tcp_flags |= packet_get_tcp_flags(packet, &miss->flow);
3515 stats->n_bytes += packet->size;
3519 facet = facet_lookup_valid(ofproto, &miss->flow);
3521 struct flow_wildcards wc;
3522 struct rule_dpif *rule;
3523 struct xlate_out xout;
3524 struct xlate_in xin;
3526 flow_wildcards_init_catchall(&wc);
3527 rule = rule_dpif_lookup(ofproto, &miss->flow, &wc);
3528 rule_credit_stats(rule, stats);
3530 xlate_in_init(&xin, ofproto, &miss->flow, rule, stats->tcp_flags,
3532 xin.resubmit_stats = stats;
3533 xin.may_learn = true;
3534 xlate_actions(&xin, &xout);
3535 flow_wildcards_or(&xout.wc, &xout.wc, &wc);
3537 /* There does not exist a bijection between 'struct flow' and datapath
3538 * flow keys with fitness ODP_FIT_TO_LITTLE. This breaks a fundamental
3539 * assumption used throughout the facet and subfacet handling code.
3540 * Since we have to handle these misses in userspace anyway, we simply
3541 * skip facet creation, avoiding the problem altogether. */
3542 if (miss->key_fitness == ODP_FIT_TOO_LITTLE
3543 || !flow_miss_should_make_facet(miss, &xout.wc)) {
3544 handle_flow_miss_without_facet(rule, &xout, miss, ops, n_ops);
3548 facet = facet_create(miss, rule, &xout, stats);
3551 handle_flow_miss_with_facet(miss, facet, now, stats, ops, n_ops);
3554 static struct drop_key *
3555 drop_key_lookup(const struct dpif_backer *backer, const struct nlattr *key,
3558 struct drop_key *drop_key;
3560 HMAP_FOR_EACH_WITH_HASH (drop_key, hmap_node, hash_bytes(key, key_len, 0),
3561 &backer->drop_keys) {
3562 if (drop_key->key_len == key_len
3563 && !memcmp(drop_key->key, key, key_len)) {
3571 drop_key_clear(struct dpif_backer *backer)
3573 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 15);
3574 struct drop_key *drop_key, *next;
3576 HMAP_FOR_EACH_SAFE (drop_key, next, hmap_node, &backer->drop_keys) {
3579 error = dpif_flow_del(backer->dpif, drop_key->key, drop_key->key_len,
3581 if (error && !VLOG_DROP_WARN(&rl)) {
3582 struct ds ds = DS_EMPTY_INITIALIZER;
3583 odp_flow_key_format(drop_key->key, drop_key->key_len, &ds);
3584 VLOG_WARN("Failed to delete drop key (%s) (%s)", strerror(error),
3589 hmap_remove(&backer->drop_keys, &drop_key->hmap_node);
3590 free(drop_key->key);
3595 /* Given a datpath, packet, and flow metadata ('backer', 'packet', and 'key'
3596 * respectively), populates 'flow' with the result of odp_flow_key_to_flow().
3597 * Optionally, if nonnull, populates 'fitnessp' with the fitness of 'flow' as
3598 * returned by odp_flow_key_to_flow(). Also, optionally populates 'ofproto'
3599 * with the ofproto_dpif, and 'odp_in_port' with the datapath in_port, that
3600 * 'packet' ingressed.
3602 * If 'ofproto' is nonnull, requires 'flow''s in_port to exist. Otherwise sets
3603 * 'flow''s in_port to OFPP_NONE.
3605 * This function does post-processing on data returned from
3606 * odp_flow_key_to_flow() to help make VLAN splinters transparent to the rest
3607 * of the upcall processing logic. In particular, if the extracted in_port is
3608 * a VLAN splinter port, it replaces flow->in_port by the "real" port, sets
3609 * flow->vlan_tci correctly for the VLAN of the VLAN splinter port, and pushes
3610 * a VLAN header onto 'packet' (if it is nonnull).
3612 * Similarly, this function also includes some logic to help with tunnels. It
3613 * may modify 'flow' as necessary to make the tunneling implementation
3614 * transparent to the upcall processing logic.
3616 * Returns 0 if successful, ENODEV if the parsed flow has no associated ofport,
3617 * or some other positive errno if there are other problems. */
3619 ofproto_receive(const struct dpif_backer *backer, struct ofpbuf *packet,
3620 const struct nlattr *key, size_t key_len,
3621 struct flow *flow, enum odp_key_fitness *fitnessp,
3622 struct ofproto_dpif **ofproto, odp_port_t *odp_in_port)
3624 const struct ofport_dpif *port;
3625 enum odp_key_fitness fitness;
3628 fitness = odp_flow_key_to_flow(key, key_len, flow);
3629 if (fitness == ODP_FIT_ERROR) {
3635 *odp_in_port = flow->in_port.odp_port;
3638 port = (tnl_port_should_receive(flow)
3639 ? ofport_dpif_cast(tnl_port_receive(flow))
3640 : odp_port_to_ofport(backer, flow->in_port.odp_port));
3641 flow->in_port.ofp_port = port ? port->up.ofp_port : OFPP_NONE;
3646 /* XXX: Since the tunnel module is not scoped per backer, for a tunnel port
3647 * it's theoretically possible that we'll receive an ofport belonging to an
3648 * entirely different datapath. In practice, this can't happen because no
3649 * platforms has two separate datapaths which each support tunneling. */
3650 ovs_assert(ofproto_dpif_cast(port->up.ofproto)->backer == backer);
3652 if (vsp_adjust_flow(ofproto_dpif_cast(port->up.ofproto), flow)) {
3654 /* Make the packet resemble the flow, so that it gets sent to
3655 * an OpenFlow controller properly, so that it looks correct
3656 * for sFlow, and so that flow_extract() will get the correct
3657 * vlan_tci if it is called on 'packet'.
3659 * The allocated space inside 'packet' probably also contains
3660 * 'key', that is, both 'packet' and 'key' are probably part of
3661 * a struct dpif_upcall (see the large comment on that
3662 * structure definition), so pushing data on 'packet' is in
3663 * general not a good idea since it could overwrite 'key' or
3664 * free it as a side effect. However, it's OK in this special
3665 * case because we know that 'packet' is inside a Netlink
3666 * attribute: pushing 4 bytes will just overwrite the 4-byte
3667 * "struct nlattr", which is fine since we don't need that
3668 * header anymore. */
3669 eth_push_vlan(packet, flow->vlan_tci);
3671 /* We can't reproduce 'key' from 'flow'. */
3672 fitness = fitness == ODP_FIT_PERFECT ? ODP_FIT_TOO_MUCH : fitness;
3677 *ofproto = ofproto_dpif_cast(port->up.ofproto);
3682 *fitnessp = fitness;
3688 handle_miss_upcalls(struct dpif_backer *backer, struct dpif_upcall *upcalls,
3691 struct dpif_upcall *upcall;
3692 struct flow_miss *miss;
3693 struct flow_miss misses[FLOW_MISS_MAX_BATCH];
3694 struct flow_miss_op flow_miss_ops[FLOW_MISS_MAX_BATCH * 2];
3695 struct dpif_op *dpif_ops[FLOW_MISS_MAX_BATCH * 2];
3705 /* Construct the to-do list.
3707 * This just amounts to extracting the flow from each packet and sticking
3708 * the packets that have the same flow in the same "flow_miss" structure so
3709 * that we can process them together. */
3712 for (upcall = upcalls; upcall < &upcalls[n_upcalls]; upcall++) {
3713 struct flow_miss *miss = &misses[n_misses];
3714 struct flow_miss *existing_miss;
3715 struct ofproto_dpif *ofproto;
3716 odp_port_t odp_in_port;
3721 error = ofproto_receive(backer, upcall->packet, upcall->key,
3722 upcall->key_len, &flow, &miss->key_fitness,
3723 &ofproto, &odp_in_port);
3724 if (error == ENODEV) {
3725 struct drop_key *drop_key;
3727 /* Received packet on datapath port for which we couldn't
3728 * associate an ofproto. This can happen if a port is removed
3729 * while traffic is being received. Print a rate-limited message
3730 * in case it happens frequently. Install a drop flow so
3731 * that future packets of the flow are inexpensively dropped
3733 VLOG_INFO_RL(&rl, "received packet on unassociated datapath port "
3734 "%"PRIu32, odp_in_port);
3736 drop_key = drop_key_lookup(backer, upcall->key, upcall->key_len);
3738 drop_key = xmalloc(sizeof *drop_key);
3739 drop_key->key = xmemdup(upcall->key, upcall->key_len);
3740 drop_key->key_len = upcall->key_len;
3742 hmap_insert(&backer->drop_keys, &drop_key->hmap_node,
3743 hash_bytes(drop_key->key, drop_key->key_len, 0));
3744 dpif_flow_put(backer->dpif, DPIF_FP_CREATE | DPIF_FP_MODIFY,
3745 drop_key->key, drop_key->key_len,
3746 NULL, 0, NULL, 0, NULL);
3754 ofproto->n_missed++;
3755 flow_extract(upcall->packet, flow.skb_priority, flow.skb_mark,
3756 &flow.tunnel, &flow.in_port, &miss->flow);
3758 /* Add other packets to a to-do list. */
3759 hash = flow_hash(&miss->flow, 0);
3760 existing_miss = flow_miss_find(&todo, ofproto, &miss->flow, hash);
3761 if (!existing_miss) {
3762 hmap_insert(&todo, &miss->hmap_node, hash);
3763 miss->ofproto = ofproto;
3764 miss->key = upcall->key;
3765 miss->key_len = upcall->key_len;
3766 miss->upcall_type = upcall->type;
3767 list_init(&miss->packets);
3771 miss = existing_miss;
3773 list_push_back(&miss->packets, &upcall->packet->list_node);
3776 /* Process each element in the to-do list, constructing the set of
3777 * operations to batch. */
3779 HMAP_FOR_EACH (miss, hmap_node, &todo) {
3780 handle_flow_miss(miss, flow_miss_ops, &n_ops);
3782 ovs_assert(n_ops <= ARRAY_SIZE(flow_miss_ops));
3784 /* Execute batch. */
3785 for (i = 0; i < n_ops; i++) {
3786 dpif_ops[i] = &flow_miss_ops[i].dpif_op;
3788 dpif_operate(backer->dpif, dpif_ops, n_ops);
3790 for (i = 0; i < n_ops; i++) {
3791 if (dpif_ops[i]->error != 0
3792 && flow_miss_ops[i].dpif_op.type == DPIF_OP_FLOW_PUT
3793 && flow_miss_ops[i].subfacet) {
3794 struct subfacet *subfacet = flow_miss_ops[i].subfacet;
3796 COVERAGE_INC(subfacet_install_fail);
3798 subfacet->path = SF_NOT_INSTALLED;
3802 if (flow_miss_ops[i].xout_garbage) {
3803 xlate_out_uninit(&flow_miss_ops[i].xout);
3806 hmap_destroy(&todo);
3809 static enum { SFLOW_UPCALL, MISS_UPCALL, BAD_UPCALL, FLOW_SAMPLE_UPCALL,
3811 classify_upcall(const struct dpif_upcall *upcall)
3813 size_t userdata_len;
3814 union user_action_cookie cookie;
3816 /* First look at the upcall type. */
3817 switch (upcall->type) {
3818 case DPIF_UC_ACTION:
3824 case DPIF_N_UC_TYPES:
3826 VLOG_WARN_RL(&rl, "upcall has unexpected type %"PRIu32, upcall->type);
3830 /* "action" upcalls need a closer look. */
3831 if (!upcall->userdata) {
3832 VLOG_WARN_RL(&rl, "action upcall missing cookie");
3835 userdata_len = nl_attr_get_size(upcall->userdata);
3836 if (userdata_len < sizeof cookie.type
3837 || userdata_len > sizeof cookie) {
3838 VLOG_WARN_RL(&rl, "action upcall cookie has unexpected size %zu",
3842 memset(&cookie, 0, sizeof cookie);
3843 memcpy(&cookie, nl_attr_get(upcall->userdata), userdata_len);
3844 if (userdata_len == sizeof cookie.sflow
3845 && cookie.type == USER_ACTION_COOKIE_SFLOW) {
3846 return SFLOW_UPCALL;
3847 } else if (userdata_len == sizeof cookie.slow_path
3848 && cookie.type == USER_ACTION_COOKIE_SLOW_PATH) {
3850 } else if (userdata_len == sizeof cookie.flow_sample
3851 && cookie.type == USER_ACTION_COOKIE_FLOW_SAMPLE) {
3852 return FLOW_SAMPLE_UPCALL;
3853 } else if (userdata_len == sizeof cookie.ipfix
3854 && cookie.type == USER_ACTION_COOKIE_IPFIX) {
3855 return IPFIX_UPCALL;
3857 VLOG_WARN_RL(&rl, "invalid user cookie of type %"PRIu16
3858 " and size %zu", cookie.type, userdata_len);
3864 handle_sflow_upcall(struct dpif_backer *backer,
3865 const struct dpif_upcall *upcall)
3867 struct ofproto_dpif *ofproto;
3868 union user_action_cookie cookie;
3870 odp_port_t odp_in_port;
3872 if (ofproto_receive(backer, upcall->packet, upcall->key, upcall->key_len,
3873 &flow, NULL, &ofproto, &odp_in_port)
3874 || !ofproto->sflow) {
3878 memset(&cookie, 0, sizeof cookie);
3879 memcpy(&cookie, nl_attr_get(upcall->userdata), sizeof cookie.sflow);
3880 dpif_sflow_received(ofproto->sflow, upcall->packet, &flow,
3881 odp_in_port, &cookie);
3885 handle_flow_sample_upcall(struct dpif_backer *backer,
3886 const struct dpif_upcall *upcall)
3888 struct ofproto_dpif *ofproto;
3889 union user_action_cookie cookie;
3892 if (ofproto_receive(backer, upcall->packet, upcall->key, upcall->key_len,
3893 &flow, NULL, &ofproto, NULL)
3894 || !ofproto->ipfix) {
3898 memset(&cookie, 0, sizeof cookie);
3899 memcpy(&cookie, nl_attr_get(upcall->userdata), sizeof cookie.flow_sample);
3901 /* The flow reflects exactly the contents of the packet. Sample
3902 * the packet using it. */
3903 dpif_ipfix_flow_sample(ofproto->ipfix, upcall->packet, &flow,
3904 cookie.flow_sample.collector_set_id,
3905 cookie.flow_sample.probability,
3906 cookie.flow_sample.obs_domain_id,
3907 cookie.flow_sample.obs_point_id);
3911 handle_ipfix_upcall(struct dpif_backer *backer,
3912 const struct dpif_upcall *upcall)
3914 struct ofproto_dpif *ofproto;
3917 if (ofproto_receive(backer, upcall->packet, upcall->key, upcall->key_len,
3918 &flow, NULL, &ofproto, NULL)
3919 || !ofproto->ipfix) {
3923 /* The flow reflects exactly the contents of the packet. Sample
3924 * the packet using it. */
3925 dpif_ipfix_bridge_sample(ofproto->ipfix, upcall->packet, &flow);
3929 handle_upcalls(struct dpif_backer *backer, unsigned int max_batch)
3931 struct dpif_upcall misses[FLOW_MISS_MAX_BATCH];
3932 struct ofpbuf miss_bufs[FLOW_MISS_MAX_BATCH];
3933 uint64_t miss_buf_stubs[FLOW_MISS_MAX_BATCH][4096 / 8];
3938 ovs_assert(max_batch <= FLOW_MISS_MAX_BATCH);
3941 for (n_processed = 0; n_processed < max_batch; n_processed++) {
3942 struct dpif_upcall *upcall = &misses[n_misses];
3943 struct ofpbuf *buf = &miss_bufs[n_misses];
3946 ofpbuf_use_stub(buf, miss_buf_stubs[n_misses],
3947 sizeof miss_buf_stubs[n_misses]);
3948 error = dpif_recv(backer->dpif, upcall, buf);
3954 switch (classify_upcall(upcall)) {
3956 /* Handle it later. */
3961 handle_sflow_upcall(backer, upcall);
3965 case FLOW_SAMPLE_UPCALL:
3966 handle_flow_sample_upcall(backer, upcall);
3971 handle_ipfix_upcall(backer, upcall);
3981 /* Handle deferred MISS_UPCALL processing. */
3982 handle_miss_upcalls(backer, misses, n_misses);
3983 for (i = 0; i < n_misses; i++) {
3984 ofpbuf_uninit(&miss_bufs[i]);
3990 /* Flow expiration. */
3992 static int subfacet_max_idle(const struct dpif_backer *);
3993 static void update_stats(struct dpif_backer *);
3994 static void rule_expire(struct rule_dpif *);
3995 static void expire_subfacets(struct dpif_backer *, int dp_max_idle);
3997 /* This function is called periodically by run(). Its job is to collect
3998 * updates for the flows that have been installed into the datapath, most
3999 * importantly when they last were used, and then use that information to
4000 * expire flows that have not been used recently.
4002 * Returns the number of milliseconds after which it should be called again. */
4004 expire(struct dpif_backer *backer)
4006 struct ofproto_dpif *ofproto;
4010 /* Periodically clear out the drop keys in an effort to keep them
4011 * relatively few. */
4012 drop_key_clear(backer);
4014 /* Update stats for each flow in the backer. */
4015 update_stats(backer);
4017 n_subfacets = hmap_count(&backer->subfacets);
4019 struct subfacet *subfacet;
4020 long long int total, now;
4024 HMAP_FOR_EACH (subfacet, hmap_node, &backer->subfacets) {
4025 total += now - subfacet->created;
4027 backer->avg_subfacet_life += total / n_subfacets;
4029 backer->avg_subfacet_life /= 2;
4031 backer->avg_n_subfacet += n_subfacets;
4032 backer->avg_n_subfacet /= 2;
4034 backer->max_n_subfacet = MAX(backer->max_n_subfacet, n_subfacets);
4036 max_idle = subfacet_max_idle(backer);
4037 expire_subfacets(backer, max_idle);
4039 HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
4040 struct rule *rule, *next_rule;
4042 if (ofproto->backer != backer) {
4046 /* Expire OpenFlow flows whose idle_timeout or hard_timeout
4048 LIST_FOR_EACH_SAFE (rule, next_rule, expirable,
4049 &ofproto->up.expirable) {
4050 rule_expire(rule_dpif_cast(rule));
4053 /* All outstanding data in existing flows has been accounted, so it's a
4054 * good time to do bond rebalancing. */
4055 if (ofproto->has_bonded_bundles) {
4056 struct ofbundle *bundle;
4058 HMAP_FOR_EACH (bundle, hmap_node, &ofproto->bundles) {
4060 bond_rebalance(bundle->bond, &backer->revalidate_set);
4066 return MIN(max_idle, 1000);
4069 /* Updates flow table statistics given that the datapath just reported 'stats'
4070 * as 'subfacet''s statistics. */
4072 update_subfacet_stats(struct subfacet *subfacet,
4073 const struct dpif_flow_stats *stats)
4075 struct facet *facet = subfacet->facet;
4076 struct ofproto_dpif *ofproto = ofproto_dpif_cast(facet->rule->up.ofproto);
4077 struct dpif_flow_stats diff;
4079 diff.tcp_flags = stats->tcp_flags;
4080 diff.used = stats->used;
4082 if (stats->n_packets >= subfacet->dp_packet_count) {
4083 diff.n_packets = stats->n_packets - subfacet->dp_packet_count;
4085 VLOG_WARN_RL(&rl, "unexpected packet count from the datapath");
4089 if (stats->n_bytes >= subfacet->dp_byte_count) {
4090 diff.n_bytes = stats->n_bytes - subfacet->dp_byte_count;
4092 VLOG_WARN_RL(&rl, "unexpected byte count from datapath");
4096 ofproto->n_hit += diff.n_packets;
4097 subfacet->dp_packet_count = stats->n_packets;
4098 subfacet->dp_byte_count = stats->n_bytes;
4099 subfacet_update_stats(subfacet, &diff);
4101 if (facet->accounted_bytes < facet->byte_count) {
4103 facet_account(facet);
4104 facet->accounted_bytes = facet->byte_count;
4108 /* 'key' with length 'key_len' bytes is a flow in 'dpif' that we know nothing
4109 * about, or a flow that shouldn't be installed but was anyway. Delete it. */
4111 delete_unexpected_flow(struct dpif_backer *backer,
4112 const struct nlattr *key, size_t key_len)
4114 if (!VLOG_DROP_WARN(&rl)) {
4118 odp_flow_key_format(key, key_len, &s);
4119 VLOG_WARN("unexpected flow: %s", ds_cstr(&s));
4123 COVERAGE_INC(facet_unexpected);
4124 dpif_flow_del(backer->dpif, key, key_len, NULL);
4127 /* Update 'packet_count', 'byte_count', and 'used' members of installed facets.
4129 * This function also pushes statistics updates to rules which each facet
4130 * resubmits into. Generally these statistics will be accurate. However, if a
4131 * facet changes the rule it resubmits into at some time in between
4132 * update_stats() runs, it is possible that statistics accrued to the
4133 * old rule will be incorrectly attributed to the new rule. This could be
4134 * avoided by calling update_stats() whenever rules are created or
4135 * deleted. However, the performance impact of making so many calls to the
4136 * datapath do not justify the benefit of having perfectly accurate statistics.
4138 * In addition, this function maintains per ofproto flow hit counts. The patch
4139 * port is not treated specially. e.g. A packet ingress from br0 patched into
4140 * br1 will increase the hit count of br0 by 1, however, does not affect
4141 * the hit or miss counts of br1.
4144 update_stats(struct dpif_backer *backer)
4146 const struct dpif_flow_stats *stats;
4147 struct dpif_flow_dump dump;
4148 const struct nlattr *key, *mask;
4149 size_t key_len, mask_len;
4151 dpif_flow_dump_start(&dump, backer->dpif);
4152 while (dpif_flow_dump_next(&dump, &key, &key_len,
4153 &mask, &mask_len, NULL, NULL, &stats)) {
4154 struct subfacet *subfacet;
4157 key_hash = odp_flow_key_hash(key, key_len);
4158 subfacet = subfacet_find(backer, key, key_len, key_hash);
4159 switch (subfacet ? subfacet->path : SF_NOT_INSTALLED) {
4161 update_subfacet_stats(subfacet, stats);
4165 /* Stats are updated per-packet. */
4168 case SF_NOT_INSTALLED:
4170 delete_unexpected_flow(backer, key, key_len);
4175 dpif_flow_dump_done(&dump);
4177 update_moving_averages(backer);
4180 /* Calculates and returns the number of milliseconds of idle time after which
4181 * subfacets should expire from the datapath. When a subfacet expires, we fold
4182 * its statistics into its facet, and when a facet's last subfacet expires, we
4183 * fold its statistic into its rule. */
4185 subfacet_max_idle(const struct dpif_backer *backer)
4188 * Idle time histogram.
4190 * Most of the time a switch has a relatively small number of subfacets.
4191 * When this is the case we might as well keep statistics for all of them
4192 * in userspace and to cache them in the kernel datapath for performance as
4195 * As the number of subfacets increases, the memory required to maintain
4196 * statistics about them in userspace and in the kernel becomes
4197 * significant. However, with a large number of subfacets it is likely
4198 * that only a few of them are "heavy hitters" that consume a large amount
4199 * of bandwidth. At this point, only heavy hitters are worth caching in
4200 * the kernel and maintaining in userspaces; other subfacets we can
4203 * The technique used to compute the idle time is to build a histogram with
4204 * N_BUCKETS buckets whose width is BUCKET_WIDTH msecs each. Each subfacet
4205 * that is installed in the kernel gets dropped in the appropriate bucket.
4206 * After the histogram has been built, we compute the cutoff so that only
4207 * the most-recently-used 1% of subfacets (but at least
4208 * flow_eviction_threshold flows) are kept cached. At least
4209 * the most-recently-used bucket of subfacets is kept, so actually an
4210 * arbitrary number of subfacets can be kept in any given expiration run
4211 * (though the next run will delete most of those unless they receive
4214 * This requires a second pass through the subfacets, in addition to the
4215 * pass made by update_stats(), because the former function never looks at
4216 * uninstallable subfacets.
4218 enum { BUCKET_WIDTH = ROUND_UP(100, TIME_UPDATE_INTERVAL) };
4219 enum { N_BUCKETS = 5000 / BUCKET_WIDTH };
4220 int buckets[N_BUCKETS] = { 0 };
4221 int total, subtotal, bucket;
4222 struct subfacet *subfacet;
4226 total = hmap_count(&backer->subfacets);
4227 if (total <= flow_eviction_threshold) {
4228 return N_BUCKETS * BUCKET_WIDTH;
4231 /* Build histogram. */
4233 HMAP_FOR_EACH (subfacet, hmap_node, &backer->subfacets) {
4234 long long int idle = now - subfacet->used;
4235 int bucket = (idle <= 0 ? 0
4236 : idle >= BUCKET_WIDTH * N_BUCKETS ? N_BUCKETS - 1
4237 : (unsigned int) idle / BUCKET_WIDTH);
4241 /* Find the first bucket whose flows should be expired. */
4242 subtotal = bucket = 0;
4244 subtotal += buckets[bucket++];
4245 } while (bucket < N_BUCKETS &&
4246 subtotal < MAX(flow_eviction_threshold, total / 100));
4248 if (VLOG_IS_DBG_ENABLED()) {
4252 ds_put_cstr(&s, "keep");
4253 for (i = 0; i < N_BUCKETS; i++) {
4255 ds_put_cstr(&s, ", drop");
4258 ds_put_format(&s, " %d:%d", i * BUCKET_WIDTH, buckets[i]);
4261 VLOG_INFO("%s (msec:count)", ds_cstr(&s));
4265 return bucket * BUCKET_WIDTH;
4269 expire_subfacets(struct dpif_backer *backer, int dp_max_idle)
4271 /* Cutoff time for most flows. */
4272 long long int normal_cutoff = time_msec() - dp_max_idle;
4274 /* We really want to keep flows for special protocols around, so use a more
4275 * conservative cutoff. */
4276 long long int special_cutoff = time_msec() - 10000;
4278 struct subfacet *subfacet, *next_subfacet;
4279 struct subfacet *batch[SUBFACET_DESTROY_MAX_BATCH];
4283 HMAP_FOR_EACH_SAFE (subfacet, next_subfacet, hmap_node,
4284 &backer->subfacets) {
4285 long long int cutoff;
4287 cutoff = (subfacet->facet->xout.slow & (SLOW_CFM | SLOW_BFD | SLOW_LACP
4291 if (subfacet->used < cutoff) {
4292 if (subfacet->path != SF_NOT_INSTALLED) {
4293 batch[n_batch++] = subfacet;
4294 if (n_batch >= SUBFACET_DESTROY_MAX_BATCH) {
4295 subfacet_destroy_batch(backer, batch, n_batch);
4299 subfacet_destroy(subfacet);
4305 subfacet_destroy_batch(backer, batch, n_batch);
4309 /* If 'rule' is an OpenFlow rule, that has expired according to OpenFlow rules,
4310 * then delete it entirely. */
4312 rule_expire(struct rule_dpif *rule)
4314 struct facet *facet, *next_facet;
4318 if (rule->up.pending) {
4319 /* We'll have to expire it later. */
4323 /* Has 'rule' expired? */
4325 if (rule->up.hard_timeout
4326 && now > rule->up.modified + rule->up.hard_timeout * 1000) {
4327 reason = OFPRR_HARD_TIMEOUT;
4328 } else if (rule->up.idle_timeout
4329 && now > rule->up.used + rule->up.idle_timeout * 1000) {
4330 reason = OFPRR_IDLE_TIMEOUT;
4335 COVERAGE_INC(ofproto_dpif_expired);
4337 /* Update stats. (This is a no-op if the rule expired due to an idle
4338 * timeout, because that only happens when the rule has no facets left.) */
4339 LIST_FOR_EACH_SAFE (facet, next_facet, list_node, &rule->facets) {
4340 facet_remove(facet);
4343 /* Get rid of the rule. */
4344 ofproto_rule_expire(&rule->up, reason);
4349 /* Creates and returns a new facet based on 'miss'.
4351 * The caller must already have determined that no facet with an identical
4352 * 'miss->flow' exists in 'miss->ofproto'.
4354 * 'rule' and 'xout' must have been created based on 'miss'.
4356 * 'facet'' statistics are initialized based on 'stats'.
4358 * The facet will initially have no subfacets. The caller should create (at
4359 * least) one subfacet with subfacet_create(). */
4360 static struct facet *
4361 facet_create(const struct flow_miss *miss, struct rule_dpif *rule,
4362 struct xlate_out *xout, struct dpif_flow_stats *stats)
4364 struct ofproto_dpif *ofproto = miss->ofproto;
4365 struct facet *facet;
4368 facet = xzalloc(sizeof *facet);
4369 facet->packet_count = facet->prev_packet_count = stats->n_packets;
4370 facet->byte_count = facet->prev_byte_count = stats->n_bytes;
4371 facet->tcp_flags = stats->tcp_flags;
4372 facet->used = stats->used;
4373 facet->flow = miss->flow;
4374 facet->learn_rl = time_msec() + 500;
4377 list_push_back(&facet->rule->facets, &facet->list_node);
4378 list_init(&facet->subfacets);
4379 netflow_flow_init(&facet->nf_flow);
4380 netflow_flow_update_time(ofproto->netflow, &facet->nf_flow, facet->used);
4382 xlate_out_copy(&facet->xout, xout);
4384 match_init(&match, &facet->flow, &facet->xout.wc);
4385 cls_rule_init(&facet->cr, &match, OFP_DEFAULT_PRIORITY);
4386 classifier_insert(&ofproto->facets, &facet->cr);
4388 facet->nf_flow.output_iface = facet->xout.nf_output_iface;
4394 facet_free(struct facet *facet)
4397 xlate_out_uninit(&facet->xout);
4402 /* Executes, within 'ofproto', the 'n_actions' actions in 'actions' on
4403 * 'packet', which arrived on 'in_port'. */
4405 execute_odp_actions(struct ofproto_dpif *ofproto, const struct flow *flow,
4406 const struct nlattr *odp_actions, size_t actions_len,
4407 struct ofpbuf *packet)
4409 struct odputil_keybuf keybuf;
4413 ofpbuf_use_stack(&key, &keybuf, sizeof keybuf);
4414 odp_flow_key_from_flow(&key, flow,
4415 ofp_port_to_odp_port(ofproto, flow->in_port.ofp_port));
4417 error = dpif_execute(ofproto->backer->dpif, key.data, key.size,
4418 odp_actions, actions_len, packet);
4422 /* Remove 'facet' from its ofproto and free up the associated memory:
4424 * - If 'facet' was installed in the datapath, uninstalls it and updates its
4425 * rule's statistics, via subfacet_uninstall().
4427 * - Removes 'facet' from its rule and from ofproto->facets.
4430 facet_remove(struct facet *facet)
4432 struct ofproto_dpif *ofproto = ofproto_dpif_cast(facet->rule->up.ofproto);
4433 struct subfacet *subfacet, *next_subfacet;
4435 ovs_assert(!list_is_empty(&facet->subfacets));
4437 /* First uninstall all of the subfacets to get final statistics. */
4438 LIST_FOR_EACH (subfacet, list_node, &facet->subfacets) {
4439 subfacet_uninstall(subfacet);
4442 /* Flush the final stats to the rule.
4444 * This might require us to have at least one subfacet around so that we
4445 * can use its actions for accounting in facet_account(), which is why we
4446 * have uninstalled but not yet destroyed the subfacets. */
4447 facet_flush_stats(facet);
4449 /* Now we're really all done so destroy everything. */
4450 LIST_FOR_EACH_SAFE (subfacet, next_subfacet, list_node,
4451 &facet->subfacets) {
4452 subfacet_destroy__(subfacet);
4454 classifier_remove(&ofproto->facets, &facet->cr);
4455 cls_rule_destroy(&facet->cr);
4456 list_remove(&facet->list_node);
4460 /* Feed information from 'facet' back into the learning table to keep it in
4461 * sync with what is actually flowing through the datapath. */
4463 facet_learn(struct facet *facet)
4465 long long int now = time_msec();
4467 if (!facet->xout.has_fin_timeout && now < facet->learn_rl) {
4471 facet->learn_rl = now + 500;
4473 if (!facet->xout.has_learn
4474 && !facet->xout.has_normal
4475 && (!facet->xout.has_fin_timeout
4476 || !(facet->tcp_flags & (TCP_FIN | TCP_RST)))) {
4480 facet_push_stats(facet, true);
4484 facet_account(struct facet *facet)
4486 struct ofproto_dpif *ofproto = ofproto_dpif_cast(facet->rule->up.ofproto);
4487 const struct nlattr *a;
4492 if (!facet->xout.has_normal || !ofproto->has_bonded_bundles) {
4495 n_bytes = facet->byte_count - facet->accounted_bytes;
4497 /* This loop feeds byte counters to bond_account() for rebalancing to use
4498 * as a basis. We also need to track the actual VLAN on which the packet
4499 * is going to be sent to ensure that it matches the one passed to
4500 * bond_choose_output_slave(). (Otherwise, we will account to the wrong
4503 * We use the actions from an arbitrary subfacet because they should all
4504 * be equally valid for our purpose. */
4505 vlan_tci = facet->flow.vlan_tci;
4506 NL_ATTR_FOR_EACH_UNSAFE (a, left, facet->xout.odp_actions.data,
4507 facet->xout.odp_actions.size) {
4508 const struct ovs_action_push_vlan *vlan;
4509 struct ofport_dpif *port;
4511 switch (nl_attr_type(a)) {
4512 case OVS_ACTION_ATTR_OUTPUT:
4513 port = get_odp_port(ofproto, nl_attr_get_odp_port(a));
4514 if (port && port->bundle && port->bundle->bond) {
4515 bond_account(port->bundle->bond, &facet->flow,
4516 vlan_tci_to_vid(vlan_tci), n_bytes);
4520 case OVS_ACTION_ATTR_POP_VLAN:
4521 vlan_tci = htons(0);
4524 case OVS_ACTION_ATTR_PUSH_VLAN:
4525 vlan = nl_attr_get(a);
4526 vlan_tci = vlan->vlan_tci;
4532 /* Returns true if the only action for 'facet' is to send to the controller.
4533 * (We don't report NetFlow expiration messages for such facets because they
4534 * are just part of the control logic for the network, not real traffic). */
4536 facet_is_controller_flow(struct facet *facet)
4539 const struct rule *rule = &facet->rule->up;
4540 const struct ofpact *ofpacts = rule->ofpacts;
4541 size_t ofpacts_len = rule->ofpacts_len;
4543 if (ofpacts_len > 0 &&
4544 ofpacts->type == OFPACT_CONTROLLER &&
4545 ofpact_next(ofpacts) >= ofpact_end(ofpacts, ofpacts_len)) {
4552 /* Folds all of 'facet''s statistics into its rule. Also updates the
4553 * accounting ofhook and emits a NetFlow expiration if appropriate. All of
4554 * 'facet''s statistics in the datapath should have been zeroed and folded into
4555 * its packet and byte counts before this function is called. */
4557 facet_flush_stats(struct facet *facet)
4559 struct ofproto_dpif *ofproto = ofproto_dpif_cast(facet->rule->up.ofproto);
4560 struct subfacet *subfacet;
4562 LIST_FOR_EACH (subfacet, list_node, &facet->subfacets) {
4563 ovs_assert(!subfacet->dp_byte_count);
4564 ovs_assert(!subfacet->dp_packet_count);
4567 facet_push_stats(facet, false);
4568 if (facet->accounted_bytes < facet->byte_count) {
4569 facet_account(facet);
4570 facet->accounted_bytes = facet->byte_count;
4573 if (ofproto->netflow && !facet_is_controller_flow(facet)) {
4574 struct ofexpired expired;
4575 expired.flow = facet->flow;
4576 expired.packet_count = facet->packet_count;
4577 expired.byte_count = facet->byte_count;
4578 expired.used = facet->used;
4579 netflow_expire(ofproto->netflow, &facet->nf_flow, &expired);
4582 /* Reset counters to prevent double counting if 'facet' ever gets
4584 facet_reset_counters(facet);
4586 netflow_flow_clear(&facet->nf_flow);
4587 facet->tcp_flags = 0;
4590 /* Searches 'ofproto''s table of facets for one which would be responsible for
4591 * 'flow'. Returns it if found, otherwise a null pointer.
4593 * The returned facet might need revalidation; use facet_lookup_valid()
4594 * instead if that is important. */
4595 static struct facet *
4596 facet_find(struct ofproto_dpif *ofproto, const struct flow *flow)
4598 struct cls_rule *cr = classifier_lookup(&ofproto->facets, flow, NULL);
4599 return cr ? CONTAINER_OF(cr, struct facet, cr) : NULL;
4602 /* Searches 'ofproto''s table of facets for one capable that covers
4603 * 'flow'. Returns it if found, otherwise a null pointer.
4605 * The returned facet is guaranteed to be valid. */
4606 static struct facet *
4607 facet_lookup_valid(struct ofproto_dpif *ofproto, const struct flow *flow)
4609 struct facet *facet;
4611 facet = facet_find(ofproto, flow);
4613 && (ofproto->backer->need_revalidate
4614 || tag_set_intersects(&ofproto->backer->revalidate_set,
4616 && !facet_revalidate(facet)) {
4624 facet_check_consistency(struct facet *facet)
4626 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 15);
4628 struct ofproto_dpif *ofproto = ofproto_dpif_cast(facet->rule->up.ofproto);
4630 struct xlate_out xout;
4631 struct xlate_in xin;
4633 struct rule_dpif *rule;
4636 /* Check the rule for consistency. */
4637 rule = rule_dpif_lookup(ofproto, &facet->flow, NULL);
4638 if (rule != facet->rule) {
4639 if (!VLOG_DROP_WARN(&rl)) {
4640 struct ds s = DS_EMPTY_INITIALIZER;
4642 flow_format(&s, &facet->flow);
4643 ds_put_format(&s, ": facet associated with wrong rule (was "
4644 "table=%"PRIu8",", facet->rule->up.table_id);
4645 cls_rule_format(&facet->rule->up.cr, &s);
4646 ds_put_format(&s, ") (should have been table=%"PRIu8",",
4648 cls_rule_format(&rule->up.cr, &s);
4649 ds_put_char(&s, ')');
4651 VLOG_WARN("%s", ds_cstr(&s));
4657 /* Check the datapath actions for consistency. */
4658 xlate_in_init(&xin, ofproto, &facet->flow, rule, 0, NULL);
4659 xlate_actions(&xin, &xout);
4661 ok = ofpbuf_equal(&facet->xout.odp_actions, &xout.odp_actions)
4662 && facet->xout.slow == xout.slow;
4663 if (!ok && !VLOG_DROP_WARN(&rl)) {
4664 struct ds s = DS_EMPTY_INITIALIZER;
4666 flow_format(&s, &facet->flow);
4667 ds_put_cstr(&s, ": inconsistency in facet");
4669 if (!ofpbuf_equal(&facet->xout.odp_actions, &xout.odp_actions)) {
4670 ds_put_cstr(&s, " (actions were: ");
4671 format_odp_actions(&s, facet->xout.odp_actions.data,
4672 facet->xout.odp_actions.size);
4673 ds_put_cstr(&s, ") (correct actions: ");
4674 format_odp_actions(&s, xout.odp_actions.data,
4675 xout.odp_actions.size);
4676 ds_put_char(&s, ')');
4679 if (facet->xout.slow != xout.slow) {
4680 ds_put_format(&s, " slow path incorrect. should be %d", xout.slow);
4683 VLOG_WARN("%s", ds_cstr(&s));
4686 xlate_out_uninit(&xout);
4691 /* Re-searches the classifier for 'facet':
4693 * - If the rule found is different from 'facet''s current rule, moves
4694 * 'facet' to the new rule and recompiles its actions.
4696 * - If the rule found is the same as 'facet''s current rule, leaves 'facet'
4697 * where it is and recompiles its actions anyway.
4699 * - If any of 'facet''s subfacets correspond to a new flow according to
4700 * ofproto_receive(), 'facet' is removed.
4702 * Returns true if 'facet' is still valid. False if 'facet' was removed. */
4704 facet_revalidate(struct facet *facet)
4706 struct ofproto_dpif *ofproto = ofproto_dpif_cast(facet->rule->up.ofproto);
4707 struct rule_dpif *new_rule;
4708 struct subfacet *subfacet;
4709 struct flow_wildcards wc;
4710 struct xlate_out xout;
4711 struct xlate_in xin;
4713 COVERAGE_INC(facet_revalidate);
4715 /* Check that child subfacets still correspond to this facet. Tunnel
4716 * configuration changes could cause a subfacet's OpenFlow in_port to
4718 LIST_FOR_EACH (subfacet, list_node, &facet->subfacets) {
4719 struct ofproto_dpif *recv_ofproto;
4720 struct flow recv_flow;
4723 error = ofproto_receive(ofproto->backer, NULL, subfacet->key,
4724 subfacet->key_len, &recv_flow, NULL,
4725 &recv_ofproto, NULL);
4727 || recv_ofproto != ofproto
4728 || facet != facet_find(ofproto, &recv_flow)) {
4729 facet_remove(facet);
4734 flow_wildcards_init_catchall(&wc);
4735 new_rule = rule_dpif_lookup(ofproto, &facet->flow, &wc);
4737 /* Calculate new datapath actions.
4739 * We do not modify any 'facet' state yet, because we might need to, e.g.,
4740 * emit a NetFlow expiration and, if so, we need to have the old state
4741 * around to properly compose it. */
4742 xlate_in_init(&xin, ofproto, &facet->flow, new_rule, 0, NULL);
4743 xlate_actions(&xin, &xout);
4744 flow_wildcards_or(&xout.wc, &xout.wc, &wc);
4746 /* A facet's slow path reason should only change under dramatic
4747 * circumstances. Rather than try to update everything, it's simpler to
4748 * remove the facet and start over.
4750 * More importantly, if a facet's wildcards change, it will be relatively
4751 * difficult to figure out if its subfacets still belong to it, and if not
4752 * which facet they may belong to. Again, to avoid the complexity, we
4753 * simply give up instead. */
4754 if (facet->xout.slow != xout.slow
4755 || memcmp(&facet->xout.wc, &xout.wc, sizeof xout.wc)) {
4756 facet_remove(facet);
4757 xlate_out_uninit(&xout);
4761 if (!ofpbuf_equal(&facet->xout.odp_actions, &xout.odp_actions)) {
4762 LIST_FOR_EACH(subfacet, list_node, &facet->subfacets) {
4763 if (subfacet->path == SF_FAST_PATH) {
4764 struct dpif_flow_stats stats;
4766 subfacet_install(subfacet, &xout.odp_actions, &stats);
4767 subfacet_update_stats(subfacet, &stats);
4771 facet_flush_stats(facet);
4773 ofpbuf_clear(&facet->xout.odp_actions);
4774 ofpbuf_put(&facet->xout.odp_actions, xout.odp_actions.data,
4775 xout.odp_actions.size);
4778 /* Update 'facet' now that we've taken care of all the old state. */
4779 facet->xout.tags = xout.tags;
4780 facet->xout.slow = xout.slow;
4781 facet->xout.has_learn = xout.has_learn;
4782 facet->xout.has_normal = xout.has_normal;
4783 facet->xout.has_fin_timeout = xout.has_fin_timeout;
4784 facet->xout.nf_output_iface = xout.nf_output_iface;
4785 facet->xout.mirrors = xout.mirrors;
4786 facet->nf_flow.output_iface = facet->xout.nf_output_iface;
4788 if (facet->rule != new_rule) {
4789 COVERAGE_INC(facet_changed_rule);
4790 list_remove(&facet->list_node);
4791 list_push_back(&new_rule->facets, &facet->list_node);
4792 facet->rule = new_rule;
4793 facet->used = new_rule->up.created;
4794 facet->prev_used = facet->used;
4797 xlate_out_uninit(&xout);
4802 facet_reset_counters(struct facet *facet)
4804 facet->packet_count = 0;
4805 facet->byte_count = 0;
4806 facet->prev_packet_count = 0;
4807 facet->prev_byte_count = 0;
4808 facet->accounted_bytes = 0;
4812 facet_push_stats(struct facet *facet, bool may_learn)
4814 struct dpif_flow_stats stats;
4816 ovs_assert(facet->packet_count >= facet->prev_packet_count);
4817 ovs_assert(facet->byte_count >= facet->prev_byte_count);
4818 ovs_assert(facet->used >= facet->prev_used);
4820 stats.n_packets = facet->packet_count - facet->prev_packet_count;
4821 stats.n_bytes = facet->byte_count - facet->prev_byte_count;
4822 stats.used = facet->used;
4823 stats.tcp_flags = facet->tcp_flags;
4825 if (may_learn || stats.n_packets || facet->used > facet->prev_used) {
4826 struct ofproto_dpif *ofproto =
4827 ofproto_dpif_cast(facet->rule->up.ofproto);
4829 struct ofport_dpif *in_port;
4830 struct xlate_in xin;
4832 facet->prev_packet_count = facet->packet_count;
4833 facet->prev_byte_count = facet->byte_count;
4834 facet->prev_used = facet->used;
4836 in_port = get_ofp_port(ofproto, facet->flow.in_port.ofp_port);
4837 if (in_port && in_port->tnl_port) {
4838 netdev_vport_inc_rx(in_port->up.netdev, &stats);
4841 rule_credit_stats(facet->rule, &stats);
4842 netflow_flow_update_time(ofproto->netflow, &facet->nf_flow,
4844 netflow_flow_update_flags(&facet->nf_flow, facet->tcp_flags);
4845 update_mirror_stats(ofproto, facet->xout.mirrors, stats.n_packets,
4848 xlate_in_init(&xin, ofproto, &facet->flow, facet->rule,
4849 stats.tcp_flags, NULL);
4850 xin.resubmit_stats = &stats;
4851 xin.may_learn = may_learn;
4852 xlate_actions_for_side_effects(&xin);
4857 push_all_stats__(bool run_fast)
4859 static long long int rl = LLONG_MIN;
4860 struct ofproto_dpif *ofproto;
4862 if (time_msec() < rl) {
4866 HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
4867 struct cls_cursor cursor;
4868 struct facet *facet;
4870 cls_cursor_init(&cursor, &ofproto->facets, NULL);
4871 CLS_CURSOR_FOR_EACH (facet, cr, &cursor) {
4872 facet_push_stats(facet, false);
4879 rl = time_msec() + 100;
4883 push_all_stats(void)
4885 push_all_stats__(true);
4889 rule_credit_stats(struct rule_dpif *rule, const struct dpif_flow_stats *stats)
4891 rule->packet_count += stats->n_packets;
4892 rule->byte_count += stats->n_bytes;
4893 ofproto_rule_update_used(&rule->up, stats->used);
4898 static struct subfacet *
4899 subfacet_find(struct dpif_backer *backer, const struct nlattr *key,
4900 size_t key_len, uint32_t key_hash)
4902 struct subfacet *subfacet;
4904 HMAP_FOR_EACH_WITH_HASH (subfacet, hmap_node, key_hash,
4905 &backer->subfacets) {
4906 if (subfacet->key_len == key_len
4907 && !memcmp(key, subfacet->key, key_len)) {
4915 /* Searches 'facet' (within 'ofproto') for a subfacet with the specified
4916 * 'key_fitness', 'key', and 'key_len' members in 'miss'. Returns the
4917 * existing subfacet if there is one, otherwise creates and returns a
4919 static struct subfacet *
4920 subfacet_create(struct facet *facet, struct flow_miss *miss,
4923 struct dpif_backer *backer = miss->ofproto->backer;
4924 enum odp_key_fitness key_fitness = miss->key_fitness;
4925 const struct nlattr *key = miss->key;
4926 size_t key_len = miss->key_len;
4928 struct subfacet *subfacet;
4930 key_hash = odp_flow_key_hash(key, key_len);
4932 if (list_is_empty(&facet->subfacets)) {
4933 subfacet = &facet->one_subfacet;
4935 subfacet = subfacet_find(backer, key, key_len, key_hash);
4937 if (subfacet->facet == facet) {
4941 /* This shouldn't happen. */
4942 VLOG_ERR_RL(&rl, "subfacet with wrong facet");
4943 subfacet_destroy(subfacet);
4946 subfacet = xmalloc(sizeof *subfacet);
4949 hmap_insert(&backer->subfacets, &subfacet->hmap_node, key_hash);
4950 list_push_back(&facet->subfacets, &subfacet->list_node);
4951 subfacet->facet = facet;
4952 subfacet->key_fitness = key_fitness;
4953 subfacet->key = xmemdup(key, key_len);
4954 subfacet->key_len = key_len;
4955 subfacet->used = now;
4956 subfacet->created = now;
4957 subfacet->dp_packet_count = 0;
4958 subfacet->dp_byte_count = 0;
4959 subfacet->path = SF_NOT_INSTALLED;
4960 subfacet->backer = backer;
4962 backer->subfacet_add_count++;
4966 /* Uninstalls 'subfacet' from the datapath, if it is installed, removes it from
4967 * its facet within 'ofproto', and frees it. */
4969 subfacet_destroy__(struct subfacet *subfacet)
4971 struct facet *facet = subfacet->facet;
4972 struct ofproto_dpif *ofproto = ofproto_dpif_cast(facet->rule->up.ofproto);
4974 /* Update ofproto stats before uninstall the subfacet. */
4975 ofproto->backer->subfacet_del_count++;
4977 subfacet_uninstall(subfacet);
4978 hmap_remove(&subfacet->backer->subfacets, &subfacet->hmap_node);
4979 list_remove(&subfacet->list_node);
4980 free(subfacet->key);
4981 if (subfacet != &facet->one_subfacet) {
4986 /* Destroys 'subfacet', as with subfacet_destroy__(), and then if this was the
4987 * last remaining subfacet in its facet destroys the facet too. */
4989 subfacet_destroy(struct subfacet *subfacet)
4991 struct facet *facet = subfacet->facet;
4993 if (list_is_singleton(&facet->subfacets)) {
4994 /* facet_remove() needs at least one subfacet (it will remove it). */
4995 facet_remove(facet);
4997 subfacet_destroy__(subfacet);
5002 subfacet_destroy_batch(struct dpif_backer *backer,
5003 struct subfacet **subfacets, int n)
5005 struct dpif_op ops[SUBFACET_DESTROY_MAX_BATCH];
5006 struct dpif_op *opsp[SUBFACET_DESTROY_MAX_BATCH];
5007 struct dpif_flow_stats stats[SUBFACET_DESTROY_MAX_BATCH];
5010 for (i = 0; i < n; i++) {
5011 ops[i].type = DPIF_OP_FLOW_DEL;
5012 ops[i].u.flow_del.key = subfacets[i]->key;
5013 ops[i].u.flow_del.key_len = subfacets[i]->key_len;
5014 ops[i].u.flow_del.stats = &stats[i];
5018 dpif_operate(backer->dpif, opsp, n);
5019 for (i = 0; i < n; i++) {
5020 subfacet_reset_dp_stats(subfacets[i], &stats[i]);
5021 subfacets[i]->path = SF_NOT_INSTALLED;
5022 subfacet_destroy(subfacets[i]);
5027 /* Updates 'subfacet''s datapath flow, setting its actions to 'actions_len'
5028 * bytes of actions in 'actions'. If 'stats' is non-null, statistics counters
5029 * in the datapath will be zeroed and 'stats' will be updated with traffic new
5030 * since 'subfacet' was last updated.
5032 * Returns 0 if successful, otherwise a positive errno value. */
5034 subfacet_install(struct subfacet *subfacet, const struct ofpbuf *odp_actions,
5035 struct dpif_flow_stats *stats)
5037 struct facet *facet = subfacet->facet;
5038 struct ofproto_dpif *ofproto = ofproto_dpif_cast(facet->rule->up.ofproto);
5039 enum subfacet_path path = facet->xout.slow ? SF_SLOW_PATH : SF_FAST_PATH;
5040 const struct nlattr *actions = odp_actions->data;
5041 size_t actions_len = odp_actions->size;
5042 struct odputil_keybuf maskbuf;
5045 uint64_t slow_path_stub[128 / 8];
5046 enum dpif_flow_put_flags flags;
5049 flags = DPIF_FP_CREATE | DPIF_FP_MODIFY;
5051 flags |= DPIF_FP_ZERO_STATS;
5054 if (path == SF_SLOW_PATH) {
5055 compose_slow_path(ofproto, &facet->flow, facet->xout.slow,
5056 slow_path_stub, sizeof slow_path_stub,
5057 &actions, &actions_len);
5060 ofpbuf_use_stack(&mask, &maskbuf, sizeof maskbuf);
5061 odp_flow_key_from_mask(&mask, &facet->xout.wc.masks,
5062 &facet->flow, UINT32_MAX);
5064 ret = dpif_flow_put(subfacet->backer->dpif, flags, subfacet->key,
5065 subfacet->key_len, mask.data, mask.size,
5066 actions, actions_len, stats);
5069 subfacet_reset_dp_stats(subfacet, stats);
5073 COVERAGE_INC(subfacet_install_fail);
5075 subfacet->path = path;
5080 /* If 'subfacet' is installed in the datapath, uninstalls it. */
5082 subfacet_uninstall(struct subfacet *subfacet)
5084 if (subfacet->path != SF_NOT_INSTALLED) {
5085 struct rule_dpif *rule = subfacet->facet->rule;
5086 struct ofproto_dpif *ofproto = ofproto_dpif_cast(rule->up.ofproto);
5087 struct dpif_flow_stats stats;
5090 error = dpif_flow_del(ofproto->backer->dpif, subfacet->key,
5091 subfacet->key_len, &stats);
5092 subfacet_reset_dp_stats(subfacet, &stats);
5094 subfacet_update_stats(subfacet, &stats);
5096 subfacet->path = SF_NOT_INSTALLED;
5098 ovs_assert(subfacet->dp_packet_count == 0);
5099 ovs_assert(subfacet->dp_byte_count == 0);
5103 /* Resets 'subfacet''s datapath statistics counters. This should be called
5104 * when 'subfacet''s statistics are cleared in the datapath. If 'stats' is
5105 * non-null, it should contain the statistics returned by dpif when 'subfacet'
5106 * was reset in the datapath. 'stats' will be modified to include only
5107 * statistics new since 'subfacet' was last updated. */
5109 subfacet_reset_dp_stats(struct subfacet *subfacet,
5110 struct dpif_flow_stats *stats)
5113 && subfacet->dp_packet_count <= stats->n_packets
5114 && subfacet->dp_byte_count <= stats->n_bytes) {
5115 stats->n_packets -= subfacet->dp_packet_count;
5116 stats->n_bytes -= subfacet->dp_byte_count;
5119 subfacet->dp_packet_count = 0;
5120 subfacet->dp_byte_count = 0;
5123 /* Folds the statistics from 'stats' into the counters in 'subfacet'.
5125 * Because of the meaning of a subfacet's counters, it only makes sense to do
5126 * this if 'stats' are not tracked in the datapath, that is, if 'stats'
5127 * represents a packet that was sent by hand or if it represents statistics
5128 * that have been cleared out of the datapath. */
5130 subfacet_update_stats(struct subfacet *subfacet,
5131 const struct dpif_flow_stats *stats)
5133 if (stats->n_packets || stats->used > subfacet->used) {
5134 struct facet *facet = subfacet->facet;
5136 subfacet->used = MAX(subfacet->used, stats->used);
5137 facet->used = MAX(facet->used, stats->used);
5138 facet->packet_count += stats->n_packets;
5139 facet->byte_count += stats->n_bytes;
5140 facet->tcp_flags |= stats->tcp_flags;
5146 /* Lookup 'flow' in 'ofproto''s classifier. If 'wc' is non-null, sets
5147 * the fields that were relevant as part of the lookup. */
5148 static struct rule_dpif *
5149 rule_dpif_lookup(struct ofproto_dpif *ofproto, const struct flow *flow,
5150 struct flow_wildcards *wc)
5152 struct rule_dpif *rule;
5154 rule = rule_dpif_lookup_in_table(ofproto, flow, wc, 0);
5159 return rule_dpif_miss_rule(ofproto, flow);
5163 rule_dpif_lookup_in_table(struct ofproto_dpif *ofproto,
5164 const struct flow *flow, struct flow_wildcards *wc,
5167 struct cls_rule *cls_rule;
5168 struct classifier *cls;
5171 if (table_id >= N_TABLES) {
5176 wc->masks.nw_frag |= FLOW_NW_FRAG_MASK;
5179 cls = &ofproto->up.tables[table_id].cls;
5180 frag = (flow->nw_frag & FLOW_NW_FRAG_ANY) != 0;
5181 if (frag && ofproto->up.frag_handling == OFPC_FRAG_NORMAL) {
5182 /* We must pretend that transport ports are unavailable. */
5183 struct flow ofpc_normal_flow = *flow;
5184 ofpc_normal_flow.tp_src = htons(0);
5185 ofpc_normal_flow.tp_dst = htons(0);
5186 cls_rule = classifier_lookup(cls, &ofpc_normal_flow, wc);
5187 } else if (frag && ofproto->up.frag_handling == OFPC_FRAG_DROP) {
5188 cls_rule = &ofproto->drop_frags_rule->up.cr;
5190 flow_wildcards_init_exact(wc);
5193 cls_rule = classifier_lookup(cls, flow, wc);
5195 return rule_dpif_cast(rule_from_cls_rule(cls_rule));
5199 rule_dpif_miss_rule(struct ofproto_dpif *ofproto, const struct flow *flow)
5201 struct ofport_dpif *port;
5203 port = get_ofp_port(ofproto, flow->in_port.ofp_port);
5205 VLOG_WARN_RL(&rl, "packet-in on unknown OpenFlow port %"PRIu16,
5206 flow->in_port.ofp_port);
5207 return ofproto->miss_rule;
5210 if (port->up.pp.config & OFPUTIL_PC_NO_PACKET_IN) {
5211 return ofproto->no_packet_in_rule;
5213 return ofproto->miss_rule;
5217 complete_operation(struct rule_dpif *rule)
5219 struct ofproto_dpif *ofproto = ofproto_dpif_cast(rule->up.ofproto);
5221 rule_invalidate(rule);
5223 struct dpif_completion *c = xmalloc(sizeof *c);
5224 c->op = rule->up.pending;
5225 list_push_back(&ofproto->completions, &c->list_node);
5227 ofoperation_complete(rule->up.pending, 0);
5231 static struct rule *
5234 struct rule_dpif *rule = xmalloc(sizeof *rule);
5239 rule_dealloc(struct rule *rule_)
5241 struct rule_dpif *rule = rule_dpif_cast(rule_);
5246 rule_construct(struct rule *rule_)
5248 struct rule_dpif *rule = rule_dpif_cast(rule_);
5249 struct ofproto_dpif *ofproto = ofproto_dpif_cast(rule->up.ofproto);
5250 struct rule_dpif *victim;
5253 rule->packet_count = 0;
5254 rule->byte_count = 0;
5256 victim = rule_dpif_cast(ofoperation_get_victim(rule->up.pending));
5257 if (victim && !list_is_empty(&victim->facets)) {
5258 struct facet *facet;
5260 rule->facets = victim->facets;
5261 list_moved(&rule->facets);
5262 LIST_FOR_EACH (facet, list_node, &rule->facets) {
5263 /* XXX: We're only clearing our local counters here. It's possible
5264 * that quite a few packets are unaccounted for in the datapath
5265 * statistics. These will be accounted to the new rule instead of
5266 * cleared as required. This could be fixed by clearing out the
5267 * datapath statistics for this facet, but currently it doesn't
5269 facet_reset_counters(facet);
5273 /* Must avoid list_moved() in this case. */
5274 list_init(&rule->facets);
5277 table_id = rule->up.table_id;
5279 rule->tag = victim->tag;
5280 } else if (table_id == 0) {
5285 miniflow_expand(&rule->up.cr.match.flow, &flow);
5286 rule->tag = rule_calculate_tag(&flow, &rule->up.cr.match.mask,
5287 ofproto->tables[table_id].basis);
5290 complete_operation(rule);
5295 rule_destruct(struct rule *rule_)
5297 struct rule_dpif *rule = rule_dpif_cast(rule_);
5298 struct facet *facet, *next_facet;
5300 LIST_FOR_EACH_SAFE (facet, next_facet, list_node, &rule->facets) {
5301 facet_revalidate(facet);
5304 complete_operation(rule);
5308 rule_get_stats(struct rule *rule_, uint64_t *packets, uint64_t *bytes)
5310 struct rule_dpif *rule = rule_dpif_cast(rule_);
5312 /* push_all_stats() can handle flow misses which, when using the learn
5313 * action, can cause rules to be added and deleted. This can corrupt our
5314 * caller's datastructures which assume that rule_get_stats() doesn't have
5315 * an impact on the flow table. To be safe, we disable miss handling. */
5316 push_all_stats__(false);
5318 /* Start from historical data for 'rule' itself that are no longer tracked
5319 * in facets. This counts, for example, facets that have expired. */
5320 *packets = rule->packet_count;
5321 *bytes = rule->byte_count;
5325 rule_dpif_execute(struct rule_dpif *rule, const struct flow *flow,
5326 struct ofpbuf *packet)
5328 struct ofproto_dpif *ofproto = ofproto_dpif_cast(rule->up.ofproto);
5329 struct dpif_flow_stats stats;
5330 struct xlate_out xout;
5331 struct xlate_in xin;
5333 dpif_flow_stats_extract(flow, packet, time_msec(), &stats);
5334 rule_credit_stats(rule, &stats);
5336 xlate_in_init(&xin, ofproto, flow, rule, stats.tcp_flags, packet);
5337 xin.resubmit_stats = &stats;
5338 xlate_actions(&xin, &xout);
5340 execute_odp_actions(ofproto, flow, xout.odp_actions.data,
5341 xout.odp_actions.size, packet);
5343 xlate_out_uninit(&xout);
5347 rule_execute(struct rule *rule, const struct flow *flow,
5348 struct ofpbuf *packet)
5350 rule_dpif_execute(rule_dpif_cast(rule), flow, packet);
5351 ofpbuf_delete(packet);
5356 rule_modify_actions(struct rule *rule_)
5358 struct rule_dpif *rule = rule_dpif_cast(rule_);
5360 complete_operation(rule);
5363 /* Sends 'packet' out 'ofport'.
5364 * May modify 'packet'.
5365 * Returns 0 if successful, otherwise a positive errno value. */
5367 send_packet(const struct ofport_dpif *ofport, struct ofpbuf *packet)
5369 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport->up.ofproto);
5370 uint64_t odp_actions_stub[1024 / 8];
5371 struct ofpbuf key, odp_actions;
5372 struct dpif_flow_stats stats;
5373 struct odputil_keybuf keybuf;
5374 struct ofpact_output output;
5375 struct xlate_out xout;
5376 struct xlate_in xin;
5378 union flow_in_port in_port_;
5381 ofpbuf_use_stub(&odp_actions, odp_actions_stub, sizeof odp_actions_stub);
5382 ofpbuf_use_stack(&key, &keybuf, sizeof keybuf);
5384 /* Use OFPP_NONE as the in_port to avoid special packet processing. */
5385 in_port_.ofp_port = OFPP_NONE;
5386 flow_extract(packet, 0, 0, NULL, &in_port_, &flow);
5387 odp_flow_key_from_flow(&key, &flow, ofp_port_to_odp_port(ofproto,
5389 dpif_flow_stats_extract(&flow, packet, time_msec(), &stats);
5391 ofpact_init(&output.ofpact, OFPACT_OUTPUT, sizeof output);
5392 output.port = ofport->up.ofp_port;
5395 xlate_in_init(&xin, ofproto, &flow, NULL, 0, packet);
5396 xin.ofpacts_len = sizeof output;
5397 xin.ofpacts = &output.ofpact;
5398 xin.resubmit_stats = &stats;
5399 xlate_actions(&xin, &xout);
5401 error = dpif_execute(ofproto->backer->dpif,
5403 xout.odp_actions.data, xout.odp_actions.size,
5405 xlate_out_uninit(&xout);
5408 VLOG_WARN_RL(&rl, "%s: failed to send packet on port %s (%s)",
5409 ofproto->up.name, netdev_get_name(ofport->up.netdev),
5413 ofproto->stats.tx_packets++;
5414 ofproto->stats.tx_bytes += packet->size;
5418 /* Composes an ODP action for a "slow path" action for 'flow' within 'ofproto'.
5419 * The action will state 'slow' as the reason that the action is in the slow
5420 * path. (This is purely informational: it allows a human viewing "ovs-dpctl
5421 * dump-flows" output to see why a flow is in the slow path.)
5423 * The 'stub_size' bytes in 'stub' will be used to store the action.
5424 * 'stub_size' must be large enough for the action.
5426 * The action and its size will be stored in '*actionsp' and '*actions_lenp',
5429 compose_slow_path(const struct ofproto_dpif *ofproto, const struct flow *flow,
5430 enum slow_path_reason slow,
5431 uint64_t *stub, size_t stub_size,
5432 const struct nlattr **actionsp, size_t *actions_lenp)
5434 union user_action_cookie cookie;
5437 cookie.type = USER_ACTION_COOKIE_SLOW_PATH;
5438 cookie.slow_path.unused = 0;
5439 cookie.slow_path.reason = slow;
5441 ofpbuf_use_stack(&buf, stub, stub_size);
5442 if (slow & (SLOW_CFM | SLOW_BFD | SLOW_LACP | SLOW_STP)) {
5443 uint32_t pid = dpif_port_get_pid(ofproto->backer->dpif,
5445 odp_put_userspace_action(pid, &cookie, sizeof cookie.slow_path, &buf);
5447 put_userspace_action(ofproto, &buf, flow, &cookie,
5448 sizeof cookie.slow_path);
5450 *actionsp = buf.data;
5451 *actions_lenp = buf.size;
5455 put_userspace_action(const struct ofproto_dpif *ofproto,
5456 struct ofpbuf *odp_actions,
5457 const struct flow *flow,
5458 const union user_action_cookie *cookie,
5459 const size_t cookie_size)
5463 pid = dpif_port_get_pid(ofproto->backer->dpif,
5464 ofp_port_to_odp_port(ofproto,
5465 flow->in_port.ofp_port));
5467 return odp_put_userspace_action(pid, cookie, cookie_size, odp_actions);
5472 update_mirror_stats(struct ofproto_dpif *ofproto, mirror_mask_t mirrors,
5473 uint64_t packets, uint64_t bytes)
5479 for (; mirrors; mirrors = zero_rightmost_1bit(mirrors)) {
5482 m = ofproto->mirrors[mirror_mask_ffs(mirrors) - 1];
5485 /* In normal circumstances 'm' will not be NULL. However,
5486 * if mirrors are reconfigured, we can temporarily get out
5487 * of sync in facet_revalidate(). We could "correct" the
5488 * mirror list before reaching here, but doing that would
5489 * not properly account the traffic stats we've currently
5490 * accumulated for previous mirror configuration. */
5494 m->packet_count += packets;
5495 m->byte_count += bytes;
5500 /* Optimized flow revalidation.
5502 * It's a difficult problem, in general, to tell which facets need to have
5503 * their actions recalculated whenever the OpenFlow flow table changes. We
5504 * don't try to solve that general problem: for most kinds of OpenFlow flow
5505 * table changes, we recalculate the actions for every facet. This is
5506 * relatively expensive, but it's good enough if the OpenFlow flow table
5507 * doesn't change very often.
5509 * However, we can expect one particular kind of OpenFlow flow table change to
5510 * happen frequently: changes caused by MAC learning. To avoid wasting a lot
5511 * of CPU on revalidating every facet whenever MAC learning modifies the flow
5512 * table, we add a special case that applies to flow tables in which every rule
5513 * has the same form (that is, the same wildcards), except that the table is
5514 * also allowed to have a single "catch-all" flow that matches all packets. We
5515 * optimize this case by tagging all of the facets that resubmit into the table
5516 * and invalidating the same tag whenever a flow changes in that table. The
5517 * end result is that we revalidate just the facets that need it (and sometimes
5518 * a few more, but not all of the facets or even all of the facets that
5519 * resubmit to the table modified by MAC learning). */
5521 /* Calculates the tag to use for 'flow' and mask 'mask' when it is inserted
5522 * into an OpenFlow table with the given 'basis'. */
5524 rule_calculate_tag(const struct flow *flow, const struct minimask *mask,
5527 if (minimask_is_catchall(mask)) {
5530 uint32_t hash = flow_hash_in_minimask(flow, mask, secret);
5531 return tag_create_deterministic(hash);
5535 /* Following a change to OpenFlow table 'table_id' in 'ofproto', update the
5536 * taggability of that table.
5538 * This function must be called after *each* change to a flow table. If you
5539 * skip calling it on some changes then the pointer comparisons at the end can
5540 * be invalid if you get unlucky. For example, if a flow removal causes a
5541 * cls_table to be destroyed and then a flow insertion causes a cls_table with
5542 * different wildcards to be created with the same address, then this function
5543 * will incorrectly skip revalidation. */
5545 table_update_taggable(struct ofproto_dpif *ofproto, uint8_t table_id)
5547 struct table_dpif *table = &ofproto->tables[table_id];
5548 const struct oftable *oftable = &ofproto->up.tables[table_id];
5549 struct cls_table *catchall, *other;
5550 struct cls_table *t;
5552 catchall = other = NULL;
5554 switch (hmap_count(&oftable->cls.tables)) {
5556 /* We could tag this OpenFlow table but it would make the logic a
5557 * little harder and it's a corner case that doesn't seem worth it
5563 HMAP_FOR_EACH (t, hmap_node, &oftable->cls.tables) {
5564 if (cls_table_is_catchall(t)) {
5566 } else if (!other) {
5569 /* Indicate that we can't tag this by setting both tables to
5570 * NULL. (We know that 'catchall' is already NULL.) */
5577 /* Can't tag this table. */
5581 if (table->catchall_table != catchall || table->other_table != other) {
5582 table->catchall_table = catchall;
5583 table->other_table = other;
5584 ofproto->backer->need_revalidate = REV_FLOW_TABLE;
5588 /* Given 'rule' that has changed in some way (either it is a rule being
5589 * inserted, a rule being deleted, or a rule whose actions are being
5590 * modified), marks facets for revalidation to ensure that packets will be
5591 * forwarded correctly according to the new state of the flow table.
5593 * This function must be called after *each* change to a flow table. See
5594 * the comment on table_update_taggable() for more information. */
5596 rule_invalidate(const struct rule_dpif *rule)
5598 struct ofproto_dpif *ofproto = ofproto_dpif_cast(rule->up.ofproto);
5600 table_update_taggable(ofproto, rule->up.table_id);
5602 if (!ofproto->backer->need_revalidate) {
5603 struct table_dpif *table = &ofproto->tables[rule->up.table_id];
5605 if (table->other_table && rule->tag) {
5606 tag_set_add(&ofproto->backer->revalidate_set, rule->tag);
5608 ofproto->backer->need_revalidate = REV_FLOW_TABLE;
5614 set_frag_handling(struct ofproto *ofproto_,
5615 enum ofp_config_flags frag_handling)
5617 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
5618 if (frag_handling != OFPC_FRAG_REASM) {
5619 ofproto->backer->need_revalidate = REV_RECONFIGURE;
5627 packet_out(struct ofproto *ofproto_, struct ofpbuf *packet,
5628 const struct flow *flow,
5629 const struct ofpact *ofpacts, size_t ofpacts_len)
5631 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
5632 struct odputil_keybuf keybuf;
5633 struct dpif_flow_stats stats;
5634 struct xlate_out xout;
5635 struct xlate_in xin;
5639 ofpbuf_use_stack(&key, &keybuf, sizeof keybuf);
5640 odp_flow_key_from_flow(&key, flow,
5641 ofp_port_to_odp_port(ofproto,
5642 flow->in_port.ofp_port));
5644 dpif_flow_stats_extract(flow, packet, time_msec(), &stats);
5646 xlate_in_init(&xin, ofproto, flow, NULL, stats.tcp_flags, packet);
5647 xin.resubmit_stats = &stats;
5648 xin.ofpacts_len = ofpacts_len;
5649 xin.ofpacts = ofpacts;
5651 xlate_actions(&xin, &xout);
5652 dpif_execute(ofproto->backer->dpif, key.data, key.size,
5653 xout.odp_actions.data, xout.odp_actions.size, packet);
5654 xlate_out_uninit(&xout);
5662 set_netflow(struct ofproto *ofproto_,
5663 const struct netflow_options *netflow_options)
5665 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
5667 if (netflow_options) {
5668 if (!ofproto->netflow) {
5669 ofproto->netflow = netflow_create();
5670 ofproto->backer->need_revalidate = REV_RECONFIGURE;
5672 return netflow_set_options(ofproto->netflow, netflow_options);
5673 } else if (ofproto->netflow) {
5674 ofproto->backer->need_revalidate = REV_RECONFIGURE;
5675 netflow_destroy(ofproto->netflow);
5676 ofproto->netflow = NULL;
5683 get_netflow_ids(const struct ofproto *ofproto_,
5684 uint8_t *engine_type, uint8_t *engine_id)
5686 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
5688 dpif_get_netflow_ids(ofproto->backer->dpif, engine_type, engine_id);
5692 send_active_timeout(struct ofproto_dpif *ofproto, struct facet *facet)
5694 if (!facet_is_controller_flow(facet) &&
5695 netflow_active_timeout_expired(ofproto->netflow, &facet->nf_flow)) {
5696 struct subfacet *subfacet;
5697 struct ofexpired expired;
5699 LIST_FOR_EACH (subfacet, list_node, &facet->subfacets) {
5700 if (subfacet->path == SF_FAST_PATH) {
5701 struct dpif_flow_stats stats;
5703 subfacet_install(subfacet, &facet->xout.odp_actions,
5705 subfacet_update_stats(subfacet, &stats);
5709 expired.flow = facet->flow;
5710 expired.packet_count = facet->packet_count;
5711 expired.byte_count = facet->byte_count;
5712 expired.used = facet->used;
5713 netflow_expire(ofproto->netflow, &facet->nf_flow, &expired);
5718 send_netflow_active_timeouts(struct ofproto_dpif *ofproto)
5720 struct cls_cursor cursor;
5721 struct facet *facet;
5723 cls_cursor_init(&cursor, &ofproto->facets, NULL);
5724 CLS_CURSOR_FOR_EACH (facet, cr, &cursor) {
5725 send_active_timeout(ofproto, facet);
5729 static struct ofproto_dpif *
5730 ofproto_dpif_lookup(const char *name)
5732 struct ofproto_dpif *ofproto;
5734 HMAP_FOR_EACH_WITH_HASH (ofproto, all_ofproto_dpifs_node,
5735 hash_string(name, 0), &all_ofproto_dpifs) {
5736 if (!strcmp(ofproto->up.name, name)) {
5744 ofproto_unixctl_fdb_flush(struct unixctl_conn *conn, int argc,
5745 const char *argv[], void *aux OVS_UNUSED)
5747 struct ofproto_dpif *ofproto;
5750 ofproto = ofproto_dpif_lookup(argv[1]);
5752 unixctl_command_reply_error(conn, "no such bridge");
5755 mac_learning_flush(ofproto->ml, &ofproto->backer->revalidate_set);
5757 HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
5758 mac_learning_flush(ofproto->ml, &ofproto->backer->revalidate_set);
5762 unixctl_command_reply(conn, "table successfully flushed");
5766 ofproto_unixctl_fdb_show(struct unixctl_conn *conn, int argc OVS_UNUSED,
5767 const char *argv[], void *aux OVS_UNUSED)
5769 struct ds ds = DS_EMPTY_INITIALIZER;
5770 const struct ofproto_dpif *ofproto;
5771 const struct mac_entry *e;
5773 ofproto = ofproto_dpif_lookup(argv[1]);
5775 unixctl_command_reply_error(conn, "no such bridge");
5779 ds_put_cstr(&ds, " port VLAN MAC Age\n");
5780 LIST_FOR_EACH (e, lru_node, &ofproto->ml->lrus) {
5781 struct ofbundle *bundle = e->port.p;
5782 ds_put_format(&ds, "%5d %4d "ETH_ADDR_FMT" %3d\n",
5783 ofbundle_get_a_port(bundle)->odp_port,
5784 e->vlan, ETH_ADDR_ARGS(e->mac),
5785 mac_entry_age(ofproto->ml, e));
5787 unixctl_command_reply(conn, ds_cstr(&ds));
5792 struct xlate_out xout;
5793 struct xlate_in xin;
5799 trace_format_rule(struct ds *result, int level, const struct rule_dpif *rule)
5801 ds_put_char_multiple(result, '\t', level);
5803 ds_put_cstr(result, "No match\n");
5807 ds_put_format(result, "Rule: table=%"PRIu8" cookie=%#"PRIx64" ",
5808 rule ? rule->up.table_id : 0, ntohll(rule->up.flow_cookie));
5809 cls_rule_format(&rule->up.cr, result);
5810 ds_put_char(result, '\n');
5812 ds_put_char_multiple(result, '\t', level);
5813 ds_put_cstr(result, "OpenFlow ");
5814 ofpacts_format(rule->up.ofpacts, rule->up.ofpacts_len, result);
5815 ds_put_char(result, '\n');
5819 trace_format_flow(struct ds *result, int level, const char *title,
5820 struct trace_ctx *trace)
5822 ds_put_char_multiple(result, '\t', level);
5823 ds_put_format(result, "%s: ", title);
5824 if (flow_equal(&trace->xin.flow, &trace->flow)) {
5825 ds_put_cstr(result, "unchanged");
5827 flow_format(result, &trace->xin.flow);
5828 trace->flow = trace->xin.flow;
5830 ds_put_char(result, '\n');
5834 trace_format_regs(struct ds *result, int level, const char *title,
5835 struct trace_ctx *trace)
5839 ds_put_char_multiple(result, '\t', level);
5840 ds_put_format(result, "%s:", title);
5841 for (i = 0; i < FLOW_N_REGS; i++) {
5842 ds_put_format(result, " reg%zu=0x%"PRIx32, i, trace->flow.regs[i]);
5844 ds_put_char(result, '\n');
5848 trace_format_odp(struct ds *result, int level, const char *title,
5849 struct trace_ctx *trace)
5851 struct ofpbuf *odp_actions = &trace->xout.odp_actions;
5853 ds_put_char_multiple(result, '\t', level);
5854 ds_put_format(result, "%s: ", title);
5855 format_odp_actions(result, odp_actions->data, odp_actions->size);
5856 ds_put_char(result, '\n');
5860 trace_resubmit(struct xlate_in *xin, struct rule_dpif *rule, int recurse)
5862 struct trace_ctx *trace = CONTAINER_OF(xin, struct trace_ctx, xin);
5863 struct ds *result = trace->result;
5865 ds_put_char(result, '\n');
5866 trace_format_flow(result, recurse + 1, "Resubmitted flow", trace);
5867 trace_format_regs(result, recurse + 1, "Resubmitted regs", trace);
5868 trace_format_odp(result, recurse + 1, "Resubmitted odp", trace);
5869 trace_format_rule(result, recurse + 1, rule);
5873 trace_report(struct xlate_in *xin, const char *s, int recurse)
5875 struct trace_ctx *trace = CONTAINER_OF(xin, struct trace_ctx, xin);
5876 struct ds *result = trace->result;
5878 ds_put_char_multiple(result, '\t', recurse);
5879 ds_put_cstr(result, s);
5880 ds_put_char(result, '\n');
5884 ofproto_unixctl_trace(struct unixctl_conn *conn, int argc, const char *argv[],
5885 void *aux OVS_UNUSED)
5887 const struct dpif_backer *backer;
5888 struct ofproto_dpif *ofproto;
5889 struct ofpbuf odp_key;
5890 struct ofpbuf *packet;
5898 ofpbuf_init(&odp_key, 0);
5900 /* Handle "-generate" or a hex string as the last argument. */
5901 if (!strcmp(argv[argc - 1], "-generate")) {
5902 packet = ofpbuf_new(0);
5905 const char *error = eth_from_hex(argv[argc - 1], &packet);
5908 } else if (argc == 4) {
5909 /* The 3-argument form must end in "-generate' or a hex string. */
5910 unixctl_command_reply_error(conn, error);
5915 /* Parse the flow and determine whether a datapath or
5916 * bridge is specified. If function odp_flow_key_from_string()
5917 * returns 0, the flow is a odp_flow. If function
5918 * parse_ofp_exact_flow() returns 0, the flow is a br_flow. */
5919 if (!odp_flow_from_string(argv[argc - 1], NULL, &odp_key, NULL)) {
5920 /* If the odp_flow is the second argument,
5921 * the datapath name is the first argument. */
5923 const char *dp_type;
5924 if (!strncmp(argv[1], "ovs-", 4)) {
5925 dp_type = argv[1] + 4;
5929 backer = shash_find_data(&all_dpif_backers, dp_type);
5931 unixctl_command_reply_error(conn, "Cannot find datapath "
5936 /* No datapath name specified, so there should be only one
5938 struct shash_node *node;
5939 if (shash_count(&all_dpif_backers) != 1) {
5940 unixctl_command_reply_error(conn, "Must specify datapath "
5941 "name, there is more than one type of datapath");
5944 node = shash_first(&all_dpif_backers);
5945 backer = node->data;
5948 /* Extract the ofproto_dpif object from the ofproto_receive()
5950 if (ofproto_receive(backer, NULL, odp_key.data,
5951 odp_key.size, &flow, NULL, &ofproto, NULL)) {
5952 unixctl_command_reply_error(conn, "Invalid datapath flow");
5955 ds_put_format(&result, "Bridge: %s\n", ofproto->up.name);
5956 } else if (!parse_ofp_exact_flow(&flow, argv[argc - 1])) {
5958 unixctl_command_reply_error(conn, "Must specify bridge name");
5962 ofproto = ofproto_dpif_lookup(argv[1]);
5964 unixctl_command_reply_error(conn, "Unknown bridge name");
5968 unixctl_command_reply_error(conn, "Bad flow syntax");
5972 /* Generate a packet, if requested. */
5974 if (!packet->size) {
5975 flow_compose(packet, &flow);
5977 union flow_in_port in_port_;
5979 in_port_ = flow.in_port;
5980 ds_put_cstr(&result, "Packet: ");
5981 s = ofp_packet_to_string(packet->data, packet->size);
5982 ds_put_cstr(&result, s);
5985 /* Use the metadata from the flow and the packet argument
5986 * to reconstruct the flow. */
5987 flow_extract(packet, flow.skb_priority, flow.skb_mark, NULL,
5992 ofproto_trace(ofproto, &flow, packet, &result);
5993 unixctl_command_reply(conn, ds_cstr(&result));
5996 ds_destroy(&result);
5997 ofpbuf_delete(packet);
5998 ofpbuf_uninit(&odp_key);
6002 ofproto_trace(struct ofproto_dpif *ofproto, const struct flow *flow,
6003 const struct ofpbuf *packet, struct ds *ds)
6005 struct rule_dpif *rule;
6007 ds_put_cstr(ds, "Flow: ");
6008 flow_format(ds, flow);
6009 ds_put_char(ds, '\n');
6011 rule = rule_dpif_lookup(ofproto, flow, NULL);
6013 trace_format_rule(ds, 0, rule);
6014 if (rule == ofproto->miss_rule) {
6015 ds_put_cstr(ds, "\nNo match, flow generates \"packet in\"s.\n");
6016 } else if (rule == ofproto->no_packet_in_rule) {
6017 ds_put_cstr(ds, "\nNo match, packets dropped because "
6018 "OFPPC_NO_PACKET_IN is set on in_port.\n");
6019 } else if (rule == ofproto->drop_frags_rule) {
6020 ds_put_cstr(ds, "\nPackets dropped because they are IP fragments "
6021 "and the fragment handling mode is \"drop\".\n");
6025 uint64_t odp_actions_stub[1024 / 8];
6026 struct ofpbuf odp_actions;
6027 struct trace_ctx trace;
6031 tcp_flags = packet ? packet_get_tcp_flags(packet, flow) : 0;
6034 ofpbuf_use_stub(&odp_actions,
6035 odp_actions_stub, sizeof odp_actions_stub);
6036 xlate_in_init(&trace.xin, ofproto, flow, rule, tcp_flags, packet);
6037 trace.xin.resubmit_hook = trace_resubmit;
6038 trace.xin.report_hook = trace_report;
6040 xlate_actions(&trace.xin, &trace.xout);
6042 ds_put_char(ds, '\n');
6043 trace_format_flow(ds, 0, "Final flow", &trace);
6045 match_init(&match, flow, &trace.xout.wc);
6046 ds_put_cstr(ds, "Relevant fields: ");
6047 match_format(&match, ds, OFP_DEFAULT_PRIORITY);
6048 ds_put_char(ds, '\n');
6050 ds_put_cstr(ds, "Datapath actions: ");
6051 format_odp_actions(ds, trace.xout.odp_actions.data,
6052 trace.xout.odp_actions.size);
6054 if (trace.xout.slow) {
6055 ds_put_cstr(ds, "\nThis flow is handled by the userspace "
6056 "slow path because it:");
6057 switch (trace.xout.slow) {
6059 ds_put_cstr(ds, "\n\t- Consists of CFM packets.");
6062 ds_put_cstr(ds, "\n\t- Consists of LACP packets.");
6065 ds_put_cstr(ds, "\n\t- Consists of STP packets.");
6068 ds_put_cstr(ds, "\n\t- Consists of BFD packets.");
6070 case SLOW_CONTROLLER:
6071 ds_put_cstr(ds, "\n\t- Sends \"packet-in\" messages "
6072 "to the OpenFlow controller.");
6079 xlate_out_uninit(&trace.xout);
6084 ofproto_dpif_clog(struct unixctl_conn *conn OVS_UNUSED, int argc OVS_UNUSED,
6085 const char *argv[] OVS_UNUSED, void *aux OVS_UNUSED)
6088 unixctl_command_reply(conn, NULL);
6092 ofproto_dpif_unclog(struct unixctl_conn *conn OVS_UNUSED, int argc OVS_UNUSED,
6093 const char *argv[] OVS_UNUSED, void *aux OVS_UNUSED)
6096 unixctl_command_reply(conn, NULL);
6099 /* Runs a self-check of flow translations in 'ofproto'. Appends a message to
6100 * 'reply' describing the results. */
6102 ofproto_dpif_self_check__(struct ofproto_dpif *ofproto, struct ds *reply)
6104 struct cls_cursor cursor;
6105 struct facet *facet;
6109 cls_cursor_init(&cursor, &ofproto->facets, NULL);
6110 CLS_CURSOR_FOR_EACH (facet, cr, &cursor) {
6111 if (!facet_check_consistency(facet)) {
6116 ofproto->backer->need_revalidate = REV_INCONSISTENCY;
6120 ds_put_format(reply, "%s: self-check failed (%d errors)\n",
6121 ofproto->up.name, errors);
6123 ds_put_format(reply, "%s: self-check passed\n", ofproto->up.name);
6128 ofproto_dpif_self_check(struct unixctl_conn *conn,
6129 int argc, const char *argv[], void *aux OVS_UNUSED)
6131 struct ds reply = DS_EMPTY_INITIALIZER;
6132 struct ofproto_dpif *ofproto;
6135 ofproto = ofproto_dpif_lookup(argv[1]);
6137 unixctl_command_reply_error(conn, "Unknown ofproto (use "
6138 "ofproto/list for help)");
6141 ofproto_dpif_self_check__(ofproto, &reply);
6143 HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
6144 ofproto_dpif_self_check__(ofproto, &reply);
6148 unixctl_command_reply(conn, ds_cstr(&reply));
6152 /* Store the current ofprotos in 'ofproto_shash'. Returns a sorted list
6153 * of the 'ofproto_shash' nodes. It is the responsibility of the caller
6154 * to destroy 'ofproto_shash' and free the returned value. */
6155 static const struct shash_node **
6156 get_ofprotos(struct shash *ofproto_shash)
6158 const struct ofproto_dpif *ofproto;
6160 HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
6161 char *name = xasprintf("%s@%s", ofproto->up.type, ofproto->up.name);
6162 shash_add_nocopy(ofproto_shash, name, ofproto);
6165 return shash_sort(ofproto_shash);
6169 ofproto_unixctl_dpif_dump_dps(struct unixctl_conn *conn, int argc OVS_UNUSED,
6170 const char *argv[] OVS_UNUSED,
6171 void *aux OVS_UNUSED)
6173 struct ds ds = DS_EMPTY_INITIALIZER;
6174 struct shash ofproto_shash;
6175 const struct shash_node **sorted_ofprotos;
6178 shash_init(&ofproto_shash);
6179 sorted_ofprotos = get_ofprotos(&ofproto_shash);
6180 for (i = 0; i < shash_count(&ofproto_shash); i++) {
6181 const struct shash_node *node = sorted_ofprotos[i];
6182 ds_put_format(&ds, "%s\n", node->name);
6185 shash_destroy(&ofproto_shash);
6186 free(sorted_ofprotos);
6188 unixctl_command_reply(conn, ds_cstr(&ds));
6193 show_dp_rates(struct ds *ds, const char *heading,
6194 const struct avg_subfacet_rates *rates)
6196 ds_put_format(ds, "%s add rate: %5.3f/min, del rate: %5.3f/min\n",
6197 heading, rates->add_rate, rates->del_rate);
6201 dpif_show_backer(const struct dpif_backer *backer, struct ds *ds)
6203 const struct shash_node **ofprotos;
6204 struct ofproto_dpif *ofproto;
6205 struct shash ofproto_shash;
6206 uint64_t n_hit, n_missed;
6207 long long int minutes;
6210 n_hit = n_missed = 0;
6211 HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
6212 if (ofproto->backer == backer) {
6213 n_missed += ofproto->n_missed;
6214 n_hit += ofproto->n_hit;
6218 ds_put_format(ds, "%s: hit:%"PRIu64" missed:%"PRIu64"\n",
6219 dpif_name(backer->dpif), n_hit, n_missed);
6220 ds_put_format(ds, "\tflows: cur: %zu, avg: %u, max: %u,"
6221 " life span: %lldms\n", hmap_count(&backer->subfacets),
6222 backer->avg_n_subfacet, backer->max_n_subfacet,
6223 backer->avg_subfacet_life);
6225 minutes = (time_msec() - backer->created) / (1000 * 60);
6226 if (minutes >= 60) {
6227 show_dp_rates(ds, "\thourly avg:", &backer->hourly);
6229 if (minutes >= 60 * 24) {
6230 show_dp_rates(ds, "\tdaily avg:", &backer->daily);
6232 show_dp_rates(ds, "\toverall avg:", &backer->lifetime);
6234 shash_init(&ofproto_shash);
6235 ofprotos = get_ofprotos(&ofproto_shash);
6236 for (i = 0; i < shash_count(&ofproto_shash); i++) {
6237 struct ofproto_dpif *ofproto = ofprotos[i]->data;
6238 const struct shash_node **ports;
6241 if (ofproto->backer != backer) {
6245 ds_put_format(ds, "\t%s: hit:%"PRIu64" missed:%"PRIu64"\n",
6246 ofproto->up.name, ofproto->n_hit, ofproto->n_missed);
6248 ports = shash_sort(&ofproto->up.port_by_name);
6249 for (j = 0; j < shash_count(&ofproto->up.port_by_name); j++) {
6250 const struct shash_node *node = ports[j];
6251 struct ofport *ofport = node->data;
6253 odp_port_t odp_port;
6255 ds_put_format(ds, "\t\t%s %u/", netdev_get_name(ofport->netdev),
6258 odp_port = ofp_port_to_odp_port(ofproto, ofport->ofp_port);
6259 if (odp_port != ODPP_NONE) {
6260 ds_put_format(ds, "%"PRIu32":", odp_port);
6262 ds_put_cstr(ds, "none:");
6265 ds_put_format(ds, " (%s", netdev_get_type(ofport->netdev));
6268 if (!netdev_get_config(ofport->netdev, &config)) {
6269 const struct smap_node **nodes;
6272 nodes = smap_sort(&config);
6273 for (i = 0; i < smap_count(&config); i++) {
6274 const struct smap_node *node = nodes[i];
6275 ds_put_format(ds, "%c %s=%s", i ? ',' : ':',
6276 node->key, node->value);
6280 smap_destroy(&config);
6282 ds_put_char(ds, ')');
6283 ds_put_char(ds, '\n');
6287 shash_destroy(&ofproto_shash);
6292 ofproto_unixctl_dpif_show(struct unixctl_conn *conn, int argc OVS_UNUSED,
6293 const char *argv[] OVS_UNUSED, void *aux OVS_UNUSED)
6295 struct ds ds = DS_EMPTY_INITIALIZER;
6296 const struct shash_node **backers;
6299 backers = shash_sort(&all_dpif_backers);
6300 for (i = 0; i < shash_count(&all_dpif_backers); i++) {
6301 dpif_show_backer(backers[i]->data, &ds);
6305 unixctl_command_reply(conn, ds_cstr(&ds));
6309 /* Dump the megaflow (facet) cache. This is useful to check the
6310 * correctness of flow wildcarding, since the same mechanism is used for
6311 * both xlate caching and kernel wildcarding.
6313 * It's important to note that in the output the flow description uses
6314 * OpenFlow (OFP) ports, but the actions use datapath (ODP) ports.
6316 * This command is only needed for advanced debugging, so it's not
6317 * documented in the man page. */
6319 ofproto_unixctl_dpif_dump_megaflows(struct unixctl_conn *conn,
6320 int argc OVS_UNUSED, const char *argv[],
6321 void *aux OVS_UNUSED)
6323 struct ds ds = DS_EMPTY_INITIALIZER;
6324 const struct ofproto_dpif *ofproto;
6325 long long int now = time_msec();
6326 struct cls_cursor cursor;
6327 struct facet *facet;
6329 ofproto = ofproto_dpif_lookup(argv[1]);
6331 unixctl_command_reply_error(conn, "no such bridge");
6335 cls_cursor_init(&cursor, &ofproto->facets, NULL);
6336 CLS_CURSOR_FOR_EACH (facet, cr, &cursor) {
6337 cls_rule_format(&facet->cr, &ds);
6338 ds_put_cstr(&ds, ", ");
6339 ds_put_format(&ds, "n_subfacets:%zu, ", list_size(&facet->subfacets));
6340 ds_put_format(&ds, "used:%.3fs, ", (now - facet->used) / 1000.0);
6341 ds_put_cstr(&ds, "Datapath actions: ");
6342 if (facet->xout.slow) {
6343 uint64_t slow_path_stub[128 / 8];
6344 const struct nlattr *actions;
6347 compose_slow_path(ofproto, &facet->flow, facet->xout.slow,
6348 slow_path_stub, sizeof slow_path_stub,
6349 &actions, &actions_len);
6350 format_odp_actions(&ds, actions, actions_len);
6352 format_odp_actions(&ds, facet->xout.odp_actions.data,
6353 facet->xout.odp_actions.size);
6355 ds_put_cstr(&ds, "\n");
6358 ds_chomp(&ds, '\n');
6359 unixctl_command_reply(conn, ds_cstr(&ds));
6364 ofproto_unixctl_dpif_dump_flows(struct unixctl_conn *conn,
6365 int argc OVS_UNUSED, const char *argv[],
6366 void *aux OVS_UNUSED)
6368 struct ds ds = DS_EMPTY_INITIALIZER;
6369 const struct ofproto_dpif *ofproto;
6370 struct subfacet *subfacet;
6372 ofproto = ofproto_dpif_lookup(argv[1]);
6374 unixctl_command_reply_error(conn, "no such bridge");
6378 update_stats(ofproto->backer);
6380 HMAP_FOR_EACH (subfacet, hmap_node, &ofproto->backer->subfacets) {
6381 struct facet *facet = subfacet->facet;
6383 if (ofproto_dpif_cast(facet->rule->up.ofproto) != ofproto) {
6387 odp_flow_key_format(subfacet->key, subfacet->key_len, &ds);
6389 ds_put_format(&ds, ", packets:%"PRIu64", bytes:%"PRIu64", used:",
6390 subfacet->dp_packet_count, subfacet->dp_byte_count);
6391 if (subfacet->used) {
6392 ds_put_format(&ds, "%.3fs",
6393 (time_msec() - subfacet->used) / 1000.0);
6395 ds_put_format(&ds, "never");
6397 if (subfacet->facet->tcp_flags) {
6398 ds_put_cstr(&ds, ", flags:");
6399 packet_format_tcp_flags(&ds, subfacet->facet->tcp_flags);
6402 ds_put_cstr(&ds, ", actions:");
6403 if (facet->xout.slow) {
6404 uint64_t slow_path_stub[128 / 8];
6405 const struct nlattr *actions;
6408 compose_slow_path(ofproto, &facet->flow, facet->xout.slow,
6409 slow_path_stub, sizeof slow_path_stub,
6410 &actions, &actions_len);
6411 format_odp_actions(&ds, actions, actions_len);
6413 format_odp_actions(&ds, facet->xout.odp_actions.data,
6414 facet->xout.odp_actions.size);
6416 ds_put_char(&ds, '\n');
6419 unixctl_command_reply(conn, ds_cstr(&ds));
6424 ofproto_unixctl_dpif_del_flows(struct unixctl_conn *conn,
6425 int argc OVS_UNUSED, const char *argv[],
6426 void *aux OVS_UNUSED)
6428 struct ds ds = DS_EMPTY_INITIALIZER;
6429 struct ofproto_dpif *ofproto;
6431 ofproto = ofproto_dpif_lookup(argv[1]);
6433 unixctl_command_reply_error(conn, "no such bridge");
6437 flush(&ofproto->up);
6439 unixctl_command_reply(conn, ds_cstr(&ds));
6444 ofproto_dpif_unixctl_init(void)
6446 static bool registered;
6452 unixctl_command_register(
6454 "[dp_name]|bridge odp_flow|br_flow [-generate|packet]",
6455 1, 3, ofproto_unixctl_trace, NULL);
6456 unixctl_command_register("fdb/flush", "[bridge]", 0, 1,
6457 ofproto_unixctl_fdb_flush, NULL);
6458 unixctl_command_register("fdb/show", "bridge", 1, 1,
6459 ofproto_unixctl_fdb_show, NULL);
6460 unixctl_command_register("ofproto/clog", "", 0, 0,
6461 ofproto_dpif_clog, NULL);
6462 unixctl_command_register("ofproto/unclog", "", 0, 0,
6463 ofproto_dpif_unclog, NULL);
6464 unixctl_command_register("ofproto/self-check", "[bridge]", 0, 1,
6465 ofproto_dpif_self_check, NULL);
6466 unixctl_command_register("dpif/dump-dps", "", 0, 0,
6467 ofproto_unixctl_dpif_dump_dps, NULL);
6468 unixctl_command_register("dpif/show", "", 0, 0, ofproto_unixctl_dpif_show,
6470 unixctl_command_register("dpif/dump-flows", "bridge", 1, 1,
6471 ofproto_unixctl_dpif_dump_flows, NULL);
6472 unixctl_command_register("dpif/del-flows", "bridge", 1, 1,
6473 ofproto_unixctl_dpif_del_flows, NULL);
6474 unixctl_command_register("dpif/dump-megaflows", "bridge", 1, 1,
6475 ofproto_unixctl_dpif_dump_megaflows, NULL);
6478 /* Linux VLAN device support (e.g. "eth0.10" for VLAN 10.)
6480 * This is deprecated. It is only for compatibility with broken device drivers
6481 * in old versions of Linux that do not properly support VLANs when VLAN
6482 * devices are not used. When broken device drivers are no longer in
6483 * widespread use, we will delete these interfaces. */
6486 set_realdev(struct ofport *ofport_, ofp_port_t realdev_ofp_port, int vid)
6488 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport_->ofproto);
6489 struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
6491 if (realdev_ofp_port == ofport->realdev_ofp_port
6492 && vid == ofport->vlandev_vid) {
6496 ofproto->backer->need_revalidate = REV_RECONFIGURE;
6498 if (ofport->realdev_ofp_port) {
6501 if (realdev_ofp_port && ofport->bundle) {
6502 /* vlandevs are enslaved to their realdevs, so they are not allowed to
6503 * themselves be part of a bundle. */
6504 bundle_set(ofport->up.ofproto, ofport->bundle, NULL);
6507 ofport->realdev_ofp_port = realdev_ofp_port;
6508 ofport->vlandev_vid = vid;
6510 if (realdev_ofp_port) {
6511 vsp_add(ofport, realdev_ofp_port, vid);
6518 hash_realdev_vid(ofp_port_t realdev_ofp_port, int vid)
6520 return hash_2words(ofp_to_u16(realdev_ofp_port), vid);
6523 /* Returns the OFP port number of the Linux VLAN device that corresponds to
6524 * 'vlan_tci' on the network device with port number 'realdev_ofp_port' in
6525 * 'struct ofport_dpif'. For example, given 'realdev_ofp_port' of eth0 and
6526 * 'vlan_tci' 9, it would return the port number of eth0.9.
6528 * Unless VLAN splinters are enabled for port 'realdev_ofp_port', this
6529 * function just returns its 'realdev_ofp_port' argument. */
6531 vsp_realdev_to_vlandev(const struct ofproto_dpif *ofproto,
6532 ofp_port_t realdev_ofp_port, ovs_be16 vlan_tci)
6534 if (!hmap_is_empty(&ofproto->realdev_vid_map)) {
6535 int vid = vlan_tci_to_vid(vlan_tci);
6536 const struct vlan_splinter *vsp;
6538 HMAP_FOR_EACH_WITH_HASH (vsp, realdev_vid_node,
6539 hash_realdev_vid(realdev_ofp_port, vid),
6540 &ofproto->realdev_vid_map) {
6541 if (vsp->realdev_ofp_port == realdev_ofp_port
6542 && vsp->vid == vid) {
6543 return vsp->vlandev_ofp_port;
6547 return realdev_ofp_port;
6550 static struct vlan_splinter *
6551 vlandev_find(const struct ofproto_dpif *ofproto, ofp_port_t vlandev_ofp_port)
6553 struct vlan_splinter *vsp;
6555 HMAP_FOR_EACH_WITH_HASH (vsp, vlandev_node,
6556 hash_int(ofp_to_u16(vlandev_ofp_port), 0),
6557 &ofproto->vlandev_map) {
6558 if (vsp->vlandev_ofp_port == vlandev_ofp_port) {
6566 /* Returns the OpenFlow port number of the "real" device underlying the Linux
6567 * VLAN device with OpenFlow port number 'vlandev_ofp_port' and stores the
6568 * VLAN VID of the Linux VLAN device in '*vid'. For example, given
6569 * 'vlandev_ofp_port' of eth0.9, it would return the OpenFlow port number of
6570 * eth0 and store 9 in '*vid'.
6572 * Returns 0 and does not modify '*vid' if 'vlandev_ofp_port' is not a Linux
6573 * VLAN device. Unless VLAN splinters are enabled, this is what this function
6576 vsp_vlandev_to_realdev(const struct ofproto_dpif *ofproto,
6577 ofp_port_t vlandev_ofp_port, int *vid)
6579 if (!hmap_is_empty(&ofproto->vlandev_map)) {
6580 const struct vlan_splinter *vsp;
6582 vsp = vlandev_find(ofproto, vlandev_ofp_port);
6587 return vsp->realdev_ofp_port;
6593 /* Given 'flow', a flow representing a packet received on 'ofproto', checks
6594 * whether 'flow->in_port' represents a Linux VLAN device. If so, changes
6595 * 'flow->in_port' to the "real" device backing the VLAN device, sets
6596 * 'flow->vlan_tci' to the VLAN VID, and returns true. Otherwise (which is
6597 * always the case unless VLAN splinters are enabled), returns false without
6598 * making any changes. */
6600 vsp_adjust_flow(const struct ofproto_dpif *ofproto, struct flow *flow)
6605 realdev = vsp_vlandev_to_realdev(ofproto, flow->in_port.ofp_port, &vid);
6610 /* Cause the flow to be processed as if it came in on the real device with
6611 * the VLAN device's VLAN ID. */
6612 flow->in_port.ofp_port = realdev;
6613 flow->vlan_tci = htons((vid & VLAN_VID_MASK) | VLAN_CFI);
6618 vsp_remove(struct ofport_dpif *port)
6620 struct ofproto_dpif *ofproto = ofproto_dpif_cast(port->up.ofproto);
6621 struct vlan_splinter *vsp;
6623 vsp = vlandev_find(ofproto, port->up.ofp_port);
6625 hmap_remove(&ofproto->vlandev_map, &vsp->vlandev_node);
6626 hmap_remove(&ofproto->realdev_vid_map, &vsp->realdev_vid_node);
6629 port->realdev_ofp_port = 0;
6631 VLOG_ERR("missing vlan device record");
6636 vsp_add(struct ofport_dpif *port, ofp_port_t realdev_ofp_port, int vid)
6638 struct ofproto_dpif *ofproto = ofproto_dpif_cast(port->up.ofproto);
6640 if (!vsp_vlandev_to_realdev(ofproto, port->up.ofp_port, NULL)
6641 && (vsp_realdev_to_vlandev(ofproto, realdev_ofp_port, htons(vid))
6642 == realdev_ofp_port)) {
6643 struct vlan_splinter *vsp;
6645 vsp = xmalloc(sizeof *vsp);
6646 hmap_insert(&ofproto->vlandev_map, &vsp->vlandev_node,
6647 hash_int(ofp_to_u16(port->up.ofp_port), 0));
6648 hmap_insert(&ofproto->realdev_vid_map, &vsp->realdev_vid_node,
6649 hash_realdev_vid(realdev_ofp_port, vid));
6650 vsp->realdev_ofp_port = realdev_ofp_port;
6651 vsp->vlandev_ofp_port = port->up.ofp_port;
6654 port->realdev_ofp_port = realdev_ofp_port;
6656 VLOG_ERR("duplicate vlan device record");
6661 ofp_port_to_odp_port(const struct ofproto_dpif *ofproto, ofp_port_t ofp_port)
6663 const struct ofport_dpif *ofport = get_ofp_port(ofproto, ofp_port);
6664 return ofport ? ofport->odp_port : ODPP_NONE;
6667 static struct ofport_dpif *
6668 odp_port_to_ofport(const struct dpif_backer *backer, odp_port_t odp_port)
6670 struct ofport_dpif *port;
6672 HMAP_FOR_EACH_IN_BUCKET (port, odp_port_node,
6673 hash_int(odp_to_u32(odp_port), 0),
6674 &backer->odp_to_ofport_map) {
6675 if (port->odp_port == odp_port) {
6684 odp_port_to_ofp_port(const struct ofproto_dpif *ofproto, odp_port_t odp_port)
6686 struct ofport_dpif *port;
6688 port = odp_port_to_ofport(ofproto->backer, odp_port);
6689 if (port && &ofproto->up == port->up.ofproto) {
6690 return port->up.ofp_port;
6696 /* Compute exponentially weighted moving average, adding 'new' as the newest,
6697 * most heavily weighted element. 'base' designates the rate of decay: after
6698 * 'base' further updates, 'new''s weight in the EWMA decays to about 1/e
6701 exp_mavg(double *avg, int base, double new)
6703 *avg = (*avg * (base - 1) + new) / base;
6707 update_moving_averages(struct dpif_backer *backer)
6709 const int min_ms = 60 * 1000; /* milliseconds in one minute. */
6710 long long int minutes = (time_msec() - backer->created) / min_ms;
6713 backer->lifetime.add_rate = (double) backer->total_subfacet_add_count
6715 backer->lifetime.del_rate = (double) backer->total_subfacet_del_count
6718 backer->lifetime.add_rate = 0.0;
6719 backer->lifetime.del_rate = 0.0;
6722 /* Update hourly averages on the minute boundaries. */
6723 if (time_msec() - backer->last_minute >= min_ms) {
6724 exp_mavg(&backer->hourly.add_rate, 60, backer->subfacet_add_count);
6725 exp_mavg(&backer->hourly.del_rate, 60, backer->subfacet_del_count);
6727 /* Update daily averages on the hour boundaries. */
6728 if ((backer->last_minute - backer->created) / min_ms % 60 == 59) {
6729 exp_mavg(&backer->daily.add_rate, 24, backer->hourly.add_rate);
6730 exp_mavg(&backer->daily.del_rate, 24, backer->hourly.del_rate);
6733 backer->total_subfacet_add_count += backer->subfacet_add_count;
6734 backer->total_subfacet_del_count += backer->subfacet_del_count;
6735 backer->subfacet_add_count = 0;
6736 backer->subfacet_del_count = 0;
6737 backer->last_minute += min_ms;
6741 const struct ofproto_class ofproto_dpif_class = {
6776 port_is_lacp_current,
6777 NULL, /* rule_choose_table */
6784 rule_modify_actions,
6798 get_stp_port_status,
6805 is_mirror_output_bundle,
6806 forward_bpdu_changed,
6807 set_mac_table_config,