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-mirror.h"
54 #include "ofproto-dpif-sflow.h"
55 #include "ofproto-dpif-xlate.h"
56 #include "poll-loop.h"
61 #include "unaligned.h"
63 #include "vlan-bitmap.h"
66 VLOG_DEFINE_THIS_MODULE(ofproto_dpif);
68 COVERAGE_DEFINE(ofproto_dpif_expired);
69 COVERAGE_DEFINE(facet_changed_rule);
70 COVERAGE_DEFINE(facet_revalidate);
71 COVERAGE_DEFINE(facet_unexpected);
72 COVERAGE_DEFINE(facet_suppress);
73 COVERAGE_DEFINE(subfacet_install_fail);
78 static struct rule_dpif *rule_dpif_lookup(struct ofproto_dpif *,
80 struct flow_wildcards *wc);
82 static void rule_get_stats(struct rule *, uint64_t *packets, uint64_t *bytes);
83 static void rule_invalidate(const struct rule_dpif *);
85 static void bundle_remove(struct ofport *);
86 static void bundle_update(struct ofbundle *);
87 static void bundle_destroy(struct ofbundle *);
88 static void bundle_del_port(struct ofport_dpif *);
89 static void bundle_run(struct ofbundle *);
90 static void bundle_wait(struct ofbundle *);
92 static void stp_run(struct ofproto_dpif *ofproto);
93 static void stp_wait(struct ofproto_dpif *ofproto);
94 static int set_stp_port(struct ofport *,
95 const struct ofproto_port_stp_settings *);
97 static void compose_slow_path(const struct ofproto_dpif *, const struct flow *,
98 enum slow_path_reason,
99 uint64_t *stub, size_t stub_size,
100 const struct nlattr **actionsp,
101 size_t *actions_lenp);
103 /* A subfacet (see "struct subfacet" below) has three possible installation
106 * - SF_NOT_INSTALLED: Not installed in the datapath. This will only be the
107 * case just after the subfacet is created, just before the subfacet is
108 * destroyed, or if the datapath returns an error when we try to install a
111 * - SF_FAST_PATH: The subfacet's actions are installed in the datapath.
113 * - SF_SLOW_PATH: An action that sends every packet for the subfacet through
114 * ofproto_dpif is installed in the datapath.
117 SF_NOT_INSTALLED, /* No datapath flow for this subfacet. */
118 SF_FAST_PATH, /* Full actions are installed. */
119 SF_SLOW_PATH, /* Send-to-userspace action is installed. */
122 /* A dpif flow and actions associated with a facet.
124 * See also the large comment on struct facet. */
127 struct hmap_node hmap_node; /* In struct ofproto_dpif 'subfacets' list. */
128 struct list list_node; /* In struct facet's 'facets' list. */
129 struct facet *facet; /* Owning facet. */
130 struct dpif_backer *backer; /* Owning backer. */
132 enum odp_key_fitness key_fitness;
136 long long int used; /* Time last used; time created if not used. */
137 long long int created; /* Time created. */
139 uint64_t dp_packet_count; /* Last known packet count in the datapath. */
140 uint64_t dp_byte_count; /* Last known byte count in the datapath. */
142 enum subfacet_path path; /* Installed in datapath? */
145 #define SUBFACET_DESTROY_MAX_BATCH 50
147 static struct subfacet *subfacet_create(struct facet *, struct flow_miss *miss,
149 static struct subfacet *subfacet_find(struct dpif_backer *,
150 const struct nlattr *key, size_t key_len,
152 static void subfacet_destroy(struct subfacet *);
153 static void subfacet_destroy__(struct subfacet *);
154 static void subfacet_destroy_batch(struct dpif_backer *,
155 struct subfacet **, int n);
156 static void subfacet_reset_dp_stats(struct subfacet *,
157 struct dpif_flow_stats *);
158 static void subfacet_update_stats(struct subfacet *,
159 const struct dpif_flow_stats *);
160 static int subfacet_install(struct subfacet *,
161 const struct ofpbuf *odp_actions,
162 struct dpif_flow_stats *);
163 static void subfacet_uninstall(struct subfacet *);
165 /* A unique, non-overlapping instantiation of an OpenFlow flow.
167 * A facet associates a "struct flow", which represents the Open vSwitch
168 * userspace idea of an exact-match flow, with one or more subfacets.
169 * While the facet is created based on an exact-match flow, it is stored
170 * within the ofproto based on the wildcards that could be expressed
171 * based on the flow table and other configuration. (See the 'wc'
172 * description in "struct xlate_out" for more details.)
174 * Each subfacet tracks the datapath's idea of the flow equivalent to
175 * the facet. When the kernel module (or other dpif implementation) and
176 * Open vSwitch userspace agree on the definition of a flow key, there
177 * is exactly one subfacet per facet. If the dpif implementation
178 * supports more-specific flow matching than userspace, however, a facet
179 * can have more than one subfacet. Examples include the dpif
180 * implementation not supporting the same wildcards as userspace or some
181 * distinction in flow that userspace simply doesn't understand.
183 * Flow expiration works in terms of subfacets, so a facet must have at
184 * least one subfacet or it will never expire, leaking memory. */
187 struct hmap_node hmap_node; /* In owning ofproto's 'facets' hmap. */
188 struct list list_node; /* In owning rule's 'facets' list. */
189 struct rule_dpif *rule; /* Owning rule. */
192 struct list subfacets;
193 long long int used; /* Time last used; time created if not used. */
196 struct flow flow; /* Flow of the creating subfacet. */
197 struct cls_rule cr; /* In 'ofproto_dpif's facets classifier. */
201 * - Do include packets and bytes sent "by hand", e.g. with
204 * - Do include packets and bytes that were obtained from the datapath
205 * when a subfacet's statistics were reset (e.g. dpif_flow_put() with
206 * DPIF_FP_ZERO_STATS).
208 * - Do not include packets or bytes that can be obtained from the
209 * datapath for any existing subfacet.
211 uint64_t packet_count; /* Number of packets received. */
212 uint64_t byte_count; /* Number of bytes received. */
214 /* Resubmit statistics. */
215 uint64_t prev_packet_count; /* Number of packets from last stats push. */
216 uint64_t prev_byte_count; /* Number of bytes from last stats push. */
217 long long int prev_used; /* Used time from last stats push. */
220 uint64_t accounted_bytes; /* Bytes processed by facet_account(). */
221 struct netflow_flow nf_flow; /* Per-flow NetFlow tracking data. */
222 uint8_t tcp_flags; /* TCP flags seen for this 'rule'. */
224 struct xlate_out xout;
226 /* Storage for a single subfacet, to reduce malloc() time and space
227 * overhead. (A facet always has at least one subfacet and in the common
228 * case has exactly one subfacet. However, 'one_subfacet' may not
229 * always be valid, since it could have been removed after newer
230 * subfacets were pushed onto the 'subfacets' list.) */
231 struct subfacet one_subfacet;
233 long long int learn_rl; /* Rate limiter for facet_learn(). */
236 static struct facet *facet_create(const struct flow_miss *, struct rule_dpif *,
238 struct dpif_flow_stats *);
239 static void facet_remove(struct facet *);
240 static void facet_free(struct facet *);
242 static struct facet *facet_find(struct ofproto_dpif *, const struct flow *);
243 static struct facet *facet_lookup_valid(struct ofproto_dpif *,
244 const struct flow *);
245 static bool facet_revalidate(struct facet *);
246 static bool facet_check_consistency(struct facet *);
248 static void facet_flush_stats(struct facet *);
250 static void facet_reset_counters(struct facet *);
251 static void facet_push_stats(struct facet *, bool may_learn);
252 static void facet_learn(struct facet *);
253 static void facet_account(struct facet *);
254 static void push_all_stats(void);
256 static bool facet_is_controller_flow(struct facet *);
258 /* Node in 'ofport_dpif''s 'priorities' map. Used to maintain a map from
259 * 'priority' (the datapath's term for QoS queue) to the dscp bits which all
260 * traffic egressing the 'ofport' with that priority should be marked with. */
261 struct priority_to_dscp {
262 struct hmap_node hmap_node; /* Node in 'ofport_dpif''s 'priorities' map. */
263 uint32_t priority; /* Priority of this queue (see struct flow). */
265 uint8_t dscp; /* DSCP bits to mark outgoing traffic with. */
268 /* Linux VLAN device support (e.g. "eth0.10" for VLAN 10.)
270 * This is deprecated. It is only for compatibility with broken device drivers
271 * in old versions of Linux that do not properly support VLANs when VLAN
272 * devices are not used. When broken device drivers are no longer in
273 * widespread use, we will delete these interfaces. */
274 struct vlan_splinter {
275 struct hmap_node realdev_vid_node;
276 struct hmap_node vlandev_node;
277 ofp_port_t realdev_ofp_port;
278 ofp_port_t vlandev_ofp_port;
282 static bool vsp_adjust_flow(const struct ofproto_dpif *, struct flow *);
283 static void vsp_remove(struct ofport_dpif *);
284 static void vsp_add(struct ofport_dpif *, ofp_port_t realdev_ofp_port, int vid);
286 static ofp_port_t odp_port_to_ofp_port(const struct ofproto_dpif *,
287 odp_port_t odp_port);
289 static struct ofport_dpif *
290 ofport_dpif_cast(const struct ofport *ofport)
292 return ofport ? CONTAINER_OF(ofport, struct ofport_dpif, up) : NULL;
295 static void port_run(struct ofport_dpif *);
296 static void port_run_fast(struct ofport_dpif *);
297 static void port_wait(struct ofport_dpif *);
298 static int set_bfd(struct ofport *, const struct smap *);
299 static int set_cfm(struct ofport *, const struct cfm_settings *);
300 static void ofport_clear_priorities(struct ofport_dpif *);
301 static void ofport_update_peer(struct ofport_dpif *);
302 static void run_fast_rl(void);
304 struct dpif_completion {
305 struct list list_node;
306 struct ofoperation *op;
309 /* Reasons that we might need to revalidate every facet, and corresponding
312 * A value of 0 means that there is no need to revalidate.
314 * It would be nice to have some cleaner way to integrate with coverage
315 * counters, but with only a few reasons I guess this is good enough for
317 enum revalidate_reason {
318 REV_RECONFIGURE = 1, /* Switch configuration changed. */
319 REV_STP, /* Spanning tree protocol port status change. */
320 REV_PORT_TOGGLED, /* Port enabled or disabled by CFM, LACP, ...*/
321 REV_FLOW_TABLE, /* Flow table changed. */
322 REV_INCONSISTENCY /* Facet self-check failed. */
324 COVERAGE_DEFINE(rev_reconfigure);
325 COVERAGE_DEFINE(rev_stp);
326 COVERAGE_DEFINE(rev_port_toggled);
327 COVERAGE_DEFINE(rev_flow_table);
328 COVERAGE_DEFINE(rev_inconsistency);
330 /* Drop keys are odp flow keys which have drop flows installed in the kernel.
331 * These are datapath flows which have no associated ofproto, if they did we
332 * would use facets. */
334 struct hmap_node hmap_node;
339 struct avg_subfacet_rates {
340 double add_rate; /* Moving average of new flows created per minute. */
341 double del_rate; /* Moving average of flows deleted per minute. */
344 /* All datapaths of a given type share a single dpif backer instance. */
349 struct timer next_expiration;
350 struct hmap odp_to_ofport_map; /* ODP port to ofport mapping. */
352 struct simap tnl_backers; /* Set of dpif ports backing tunnels. */
354 /* Facet revalidation flags applying to facets which use this backer. */
355 enum revalidate_reason need_revalidate; /* Revalidate every facet. */
356 struct tag_set revalidate_set; /* Revalidate only matching facets. */
358 struct hmap drop_keys; /* Set of dropped odp keys. */
359 bool recv_set_enable; /* Enables or disables receiving packets. */
361 struct hmap subfacets;
362 struct governor *governor;
364 /* Subfacet statistics.
366 * These keep track of the total number of subfacets added and deleted and
367 * flow life span. They are useful for computing the flow rates stats
368 * exposed via "ovs-appctl dpif/show". The goal is to learn about
369 * traffic patterns in ways that we can use later to improve Open vSwitch
370 * performance in new situations. */
371 long long int created; /* Time when it is created. */
372 unsigned max_n_subfacet; /* Maximum number of flows */
373 unsigned avg_n_subfacet; /* Average number of flows. */
374 long long int avg_subfacet_life; /* Average life span of subfacets. */
376 /* The average number of subfacets... */
377 struct avg_subfacet_rates hourly; /* ...over the last hour. */
378 struct avg_subfacet_rates daily; /* ...over the last day. */
379 struct avg_subfacet_rates lifetime; /* ...over the switch lifetime. */
380 long long int last_minute; /* Last time 'hourly' was updated. */
382 /* Number of subfacets added or deleted since 'last_minute'. */
383 unsigned subfacet_add_count;
384 unsigned subfacet_del_count;
386 /* Number of subfacets added or deleted from 'created' to 'last_minute.' */
387 unsigned long long int total_subfacet_add_count;
388 unsigned long long int total_subfacet_del_count;
391 /* All existing ofproto_backer instances, indexed by ofproto->up.type. */
392 static struct shash all_dpif_backers = SHASH_INITIALIZER(&all_dpif_backers);
394 static void drop_key_clear(struct dpif_backer *);
395 static struct ofport_dpif *
396 odp_port_to_ofport(const struct dpif_backer *, odp_port_t odp_port);
397 static void update_moving_averages(struct dpif_backer *backer);
399 /* Defer flow mod completion until "ovs-appctl ofproto/unclog"? (Useful only
400 * for debugging the asynchronous flow_mod implementation.) */
403 /* By default, flows in the datapath are wildcarded (megaflows). They
404 * may be disabled with the "ovs-appctl dpif/disable-megaflows" command. */
405 static bool enable_megaflows = true;
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->is_tunnel) {
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, iter->up.netdev,
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,
935 ovs_strerror(error));
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, ovs_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 ofproto->mbridge = mbridge_create();
1028 ofproto->has_bonded_bundles = false;
1030 classifier_init(&ofproto->facets);
1031 ofproto->consistency_rl = LLONG_MIN;
1033 for (i = 0; i < N_TABLES; i++) {
1034 struct table_dpif *table = &ofproto->tables[i];
1036 table->catchall_table = NULL;
1037 table->other_table = NULL;
1038 table->basis = random_uint32();
1041 list_init(&ofproto->completions);
1043 ofproto_dpif_unixctl_init();
1045 hmap_init(&ofproto->vlandev_map);
1046 hmap_init(&ofproto->realdev_vid_map);
1048 sset_init(&ofproto->ports);
1049 sset_init(&ofproto->ghost_ports);
1050 sset_init(&ofproto->port_poll_set);
1051 ofproto->port_poll_errno = 0;
1053 SHASH_FOR_EACH_SAFE (node, next, &init_ofp_ports) {
1054 struct iface_hint *iface_hint = node->data;
1056 if (!strcmp(iface_hint->br_name, ofproto->up.name)) {
1057 /* Check if the datapath already has this port. */
1058 if (dpif_port_exists(ofproto->backer->dpif, node->name)) {
1059 sset_add(&ofproto->ports, node->name);
1062 free(iface_hint->br_name);
1063 free(iface_hint->br_type);
1065 shash_delete(&init_ofp_ports, node);
1069 hmap_insert(&all_ofproto_dpifs, &ofproto->all_ofproto_dpifs_node,
1070 hash_string(ofproto->up.name, 0));
1071 memset(&ofproto->stats, 0, sizeof ofproto->stats);
1073 ofproto_init_tables(ofproto_, N_TABLES);
1074 error = add_internal_flows(ofproto);
1075 ofproto->up.tables[TBL_INTERNAL].flags = OFTABLE_HIDDEN | OFTABLE_READONLY;
1078 ofproto->n_missed = 0;
1084 add_internal_flow(struct ofproto_dpif *ofproto, int id,
1085 const struct ofpbuf *ofpacts, struct rule_dpif **rulep)
1087 struct ofputil_flow_mod fm;
1090 match_init_catchall(&fm.match);
1092 match_set_reg(&fm.match, 0, id);
1093 fm.new_cookie = htonll(0);
1094 fm.cookie = htonll(0);
1095 fm.cookie_mask = htonll(0);
1096 fm.modify_cookie = false;
1097 fm.table_id = TBL_INTERNAL;
1098 fm.command = OFPFC_ADD;
1099 fm.idle_timeout = 0;
1100 fm.hard_timeout = 0;
1104 fm.ofpacts = ofpacts->data;
1105 fm.ofpacts_len = ofpacts->size;
1107 error = ofproto_flow_mod(&ofproto->up, &fm);
1109 VLOG_ERR_RL(&rl, "failed to add internal flow %d (%s)",
1110 id, ofperr_to_string(error));
1114 *rulep = rule_dpif_lookup_in_table(ofproto, &fm.match.flow, NULL,
1116 ovs_assert(*rulep != NULL);
1122 add_internal_flows(struct ofproto_dpif *ofproto)
1124 struct ofpact_controller *controller;
1125 uint64_t ofpacts_stub[128 / 8];
1126 struct ofpbuf ofpacts;
1130 ofpbuf_use_stack(&ofpacts, ofpacts_stub, sizeof ofpacts_stub);
1133 controller = ofpact_put_CONTROLLER(&ofpacts);
1134 controller->max_len = UINT16_MAX;
1135 controller->controller_id = 0;
1136 controller->reason = OFPR_NO_MATCH;
1137 ofpact_pad(&ofpacts);
1139 error = add_internal_flow(ofproto, id++, &ofpacts, &ofproto->miss_rule);
1144 ofpbuf_clear(&ofpacts);
1145 error = add_internal_flow(ofproto, id++, &ofpacts,
1146 &ofproto->no_packet_in_rule);
1151 error = add_internal_flow(ofproto, id++, &ofpacts,
1152 &ofproto->drop_frags_rule);
1157 complete_operations(struct ofproto_dpif *ofproto)
1159 struct dpif_completion *c, *next;
1161 LIST_FOR_EACH_SAFE (c, next, list_node, &ofproto->completions) {
1162 ofoperation_complete(c->op, 0);
1163 list_remove(&c->list_node);
1169 destruct(struct ofproto *ofproto_)
1171 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1172 struct rule_dpif *rule, *next_rule;
1173 struct oftable *table;
1175 ofproto->backer->need_revalidate = REV_RECONFIGURE;
1176 hmap_remove(&all_ofproto_dpifs, &ofproto->all_ofproto_dpifs_node);
1177 complete_operations(ofproto);
1179 OFPROTO_FOR_EACH_TABLE (table, &ofproto->up) {
1180 struct cls_cursor cursor;
1182 cls_cursor_init(&cursor, &table->cls, NULL);
1183 CLS_CURSOR_FOR_EACH_SAFE (rule, next_rule, up.cr, &cursor) {
1184 ofproto_rule_destroy(&rule->up);
1188 mbridge_unref(ofproto->mbridge);
1190 netflow_destroy(ofproto->netflow);
1191 dpif_sflow_unref(ofproto->sflow);
1192 hmap_destroy(&ofproto->bundles);
1193 mac_learning_unref(ofproto->ml);
1195 classifier_destroy(&ofproto->facets);
1197 hmap_destroy(&ofproto->vlandev_map);
1198 hmap_destroy(&ofproto->realdev_vid_map);
1200 sset_destroy(&ofproto->ports);
1201 sset_destroy(&ofproto->ghost_ports);
1202 sset_destroy(&ofproto->port_poll_set);
1204 close_dpif_backer(ofproto->backer);
1208 run_fast(struct ofproto *ofproto_)
1210 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1211 struct ofport_dpif *ofport;
1213 /* Do not perform any periodic activity required by 'ofproto' while
1214 * waiting for flow restore to complete. */
1215 if (ofproto_get_flow_restore_wait()) {
1219 HMAP_FOR_EACH (ofport, up.hmap_node, &ofproto->up.ports) {
1220 port_run_fast(ofport);
1227 run(struct ofproto *ofproto_)
1229 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1230 struct ofport_dpif *ofport;
1231 struct ofbundle *bundle;
1235 complete_operations(ofproto);
1238 if (mbridge_need_revalidate(ofproto->mbridge)) {
1239 ofproto->backer->need_revalidate = REV_RECONFIGURE;
1240 mac_learning_flush(ofproto->ml, NULL);
1243 /* Do not perform any periodic activity below required by 'ofproto' while
1244 * waiting for flow restore to complete. */
1245 if (ofproto_get_flow_restore_wait()) {
1249 error = run_fast(ofproto_);
1254 if (ofproto->netflow) {
1255 if (netflow_run(ofproto->netflow)) {
1256 send_netflow_active_timeouts(ofproto);
1259 if (ofproto->sflow) {
1260 dpif_sflow_run(ofproto->sflow);
1263 HMAP_FOR_EACH (ofport, up.hmap_node, &ofproto->up.ports) {
1266 HMAP_FOR_EACH (bundle, hmap_node, &ofproto->bundles) {
1271 mac_learning_run(ofproto->ml, &ofproto->backer->revalidate_set);
1273 /* Check the consistency of a random facet, to aid debugging. */
1274 if (time_msec() >= ofproto->consistency_rl
1275 && !classifier_is_empty(&ofproto->facets)
1276 && !ofproto->backer->need_revalidate) {
1277 struct cls_table *table;
1278 struct cls_rule *cr;
1279 struct facet *facet;
1281 ofproto->consistency_rl = time_msec() + 250;
1283 table = CONTAINER_OF(hmap_random_node(&ofproto->facets.tables),
1284 struct cls_table, hmap_node);
1285 cr = CONTAINER_OF(hmap_random_node(&table->rules), struct cls_rule,
1287 facet = CONTAINER_OF(cr, struct facet, cr);
1289 if (!tag_set_intersects(&ofproto->backer->revalidate_set,
1290 facet->xout.tags)) {
1291 if (!facet_check_consistency(facet)) {
1292 ofproto->backer->need_revalidate = REV_INCONSISTENCY;
1301 wait(struct ofproto *ofproto_)
1303 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1304 struct ofport_dpif *ofport;
1305 struct ofbundle *bundle;
1307 if (!clogged && !list_is_empty(&ofproto->completions)) {
1308 poll_immediate_wake();
1311 if (ofproto_get_flow_restore_wait()) {
1315 dpif_wait(ofproto->backer->dpif);
1316 dpif_recv_wait(ofproto->backer->dpif);
1317 if (ofproto->sflow) {
1318 dpif_sflow_wait(ofproto->sflow);
1320 if (!tag_set_is_empty(&ofproto->backer->revalidate_set)) {
1321 poll_immediate_wake();
1323 HMAP_FOR_EACH (ofport, up.hmap_node, &ofproto->up.ports) {
1326 HMAP_FOR_EACH (bundle, hmap_node, &ofproto->bundles) {
1327 bundle_wait(bundle);
1329 if (ofproto->netflow) {
1330 netflow_wait(ofproto->netflow);
1332 mac_learning_wait(ofproto->ml);
1334 if (ofproto->backer->need_revalidate) {
1335 /* Shouldn't happen, but if it does just go around again. */
1336 VLOG_DBG_RL(&rl, "need revalidate in ofproto_wait_cb()");
1337 poll_immediate_wake();
1342 get_memory_usage(const struct ofproto *ofproto_, struct simap *usage)
1344 const struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1345 struct cls_cursor cursor;
1346 size_t n_subfacets = 0;
1347 struct facet *facet;
1349 simap_increase(usage, "facets", classifier_count(&ofproto->facets));
1351 cls_cursor_init(&cursor, &ofproto->facets, NULL);
1352 CLS_CURSOR_FOR_EACH (facet, cr, &cursor) {
1353 n_subfacets += list_size(&facet->subfacets);
1355 simap_increase(usage, "subfacets", n_subfacets);
1359 flush(struct ofproto *ofproto_)
1361 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1362 struct subfacet *subfacet, *next_subfacet;
1363 struct subfacet *batch[SUBFACET_DESTROY_MAX_BATCH];
1367 HMAP_FOR_EACH_SAFE (subfacet, next_subfacet, hmap_node,
1368 &ofproto->backer->subfacets) {
1369 if (ofproto_dpif_cast(subfacet->facet->rule->up.ofproto) != ofproto) {
1373 if (subfacet->path != SF_NOT_INSTALLED) {
1374 batch[n_batch++] = subfacet;
1375 if (n_batch >= SUBFACET_DESTROY_MAX_BATCH) {
1376 subfacet_destroy_batch(ofproto->backer, batch, n_batch);
1380 subfacet_destroy(subfacet);
1385 subfacet_destroy_batch(ofproto->backer, batch, n_batch);
1390 get_features(struct ofproto *ofproto_ OVS_UNUSED,
1391 bool *arp_match_ip, enum ofputil_action_bitmap *actions)
1393 *arp_match_ip = true;
1394 *actions = (OFPUTIL_A_OUTPUT |
1395 OFPUTIL_A_SET_VLAN_VID |
1396 OFPUTIL_A_SET_VLAN_PCP |
1397 OFPUTIL_A_STRIP_VLAN |
1398 OFPUTIL_A_SET_DL_SRC |
1399 OFPUTIL_A_SET_DL_DST |
1400 OFPUTIL_A_SET_NW_SRC |
1401 OFPUTIL_A_SET_NW_DST |
1402 OFPUTIL_A_SET_NW_TOS |
1403 OFPUTIL_A_SET_TP_SRC |
1404 OFPUTIL_A_SET_TP_DST |
1409 get_tables(struct ofproto *ofproto_, struct ofp12_table_stats *ots)
1411 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1412 struct dpif_dp_stats s;
1413 uint64_t n_miss, n_no_pkt_in, n_bytes, n_dropped_frags;
1416 strcpy(ots->name, "classifier");
1418 dpif_get_dp_stats(ofproto->backer->dpif, &s);
1419 rule_get_stats(&ofproto->miss_rule->up, &n_miss, &n_bytes);
1420 rule_get_stats(&ofproto->no_packet_in_rule->up, &n_no_pkt_in, &n_bytes);
1421 rule_get_stats(&ofproto->drop_frags_rule->up, &n_dropped_frags, &n_bytes);
1423 n_lookup = s.n_hit + s.n_missed - n_dropped_frags;
1424 ots->lookup_count = htonll(n_lookup);
1425 ots->matched_count = htonll(n_lookup - n_miss - n_no_pkt_in);
1428 static struct ofport *
1431 struct ofport_dpif *port = xmalloc(sizeof *port);
1436 port_dealloc(struct ofport *port_)
1438 struct ofport_dpif *port = ofport_dpif_cast(port_);
1443 port_construct(struct ofport *port_)
1445 struct ofport_dpif *port = ofport_dpif_cast(port_);
1446 struct ofproto_dpif *ofproto = ofproto_dpif_cast(port->up.ofproto);
1447 const struct netdev *netdev = port->up.netdev;
1448 char namebuf[NETDEV_VPORT_NAME_BUFSIZE];
1449 struct dpif_port dpif_port;
1452 ofproto->backer->need_revalidate = REV_RECONFIGURE;
1453 port->bundle = NULL;
1456 port->tag = tag_create_random();
1457 port->may_enable = true;
1458 port->stp_port = NULL;
1459 port->stp_state = STP_DISABLED;
1460 port->is_tunnel = false;
1462 hmap_init(&port->priorities);
1463 port->realdev_ofp_port = 0;
1464 port->vlandev_vid = 0;
1465 port->carrier_seq = netdev_get_carrier_resets(netdev);
1467 if (netdev_vport_is_patch(netdev)) {
1468 /* By bailing out here, we don't submit the port to the sFlow module
1469 * to be considered for counter polling export. This is correct
1470 * because the patch port represents an interface that sFlow considers
1471 * to be "internal" to the switch as a whole, and therefore not an
1472 * candidate for counter polling. */
1473 port->odp_port = ODPP_NONE;
1474 ofport_update_peer(port);
1478 error = dpif_port_query_by_name(ofproto->backer->dpif,
1479 netdev_vport_get_dpif_port(netdev, namebuf,
1486 port->odp_port = dpif_port.port_no;
1488 if (netdev_get_tunnel_config(netdev)) {
1489 tnl_port_add(port, port->up.netdev, port->odp_port);
1490 port->is_tunnel = true;
1492 /* Sanity-check that a mapping doesn't already exist. This
1493 * shouldn't happen for non-tunnel ports. */
1494 if (odp_port_to_ofp_port(ofproto, port->odp_port) != OFPP_NONE) {
1495 VLOG_ERR("port %s already has an OpenFlow port number",
1497 dpif_port_destroy(&dpif_port);
1501 hmap_insert(&ofproto->backer->odp_to_ofport_map, &port->odp_port_node,
1502 hash_odp_port(port->odp_port));
1504 dpif_port_destroy(&dpif_port);
1506 if (ofproto->sflow) {
1507 dpif_sflow_add_port(ofproto->sflow, port_, port->odp_port);
1514 port_destruct(struct ofport *port_)
1516 struct ofport_dpif *port = ofport_dpif_cast(port_);
1517 struct ofproto_dpif *ofproto = ofproto_dpif_cast(port->up.ofproto);
1518 const char *devname = netdev_get_name(port->up.netdev);
1519 char namebuf[NETDEV_VPORT_NAME_BUFSIZE];
1520 const char *dp_port_name;
1522 ofproto->backer->need_revalidate = REV_RECONFIGURE;
1524 dp_port_name = netdev_vport_get_dpif_port(port->up.netdev, namebuf,
1526 if (dpif_port_exists(ofproto->backer->dpif, dp_port_name)) {
1527 /* The underlying device is still there, so delete it. This
1528 * happens when the ofproto is being destroyed, since the caller
1529 * assumes that removal of attached ports will happen as part of
1531 if (!port->is_tunnel) {
1532 dpif_port_del(ofproto->backer->dpif, port->odp_port);
1537 port->peer->peer = NULL;
1541 if (port->odp_port != ODPP_NONE && !port->is_tunnel) {
1542 hmap_remove(&ofproto->backer->odp_to_ofport_map, &port->odp_port_node);
1546 sset_find_and_delete(&ofproto->ports, devname);
1547 sset_find_and_delete(&ofproto->ghost_ports, devname);
1548 bundle_remove(port_);
1549 set_cfm(port_, NULL);
1550 set_bfd(port_, NULL);
1551 if (ofproto->sflow) {
1552 dpif_sflow_del_port(ofproto->sflow, port->odp_port);
1555 ofport_clear_priorities(port);
1556 hmap_destroy(&port->priorities);
1560 port_modified(struct ofport *port_)
1562 struct ofport_dpif *port = ofport_dpif_cast(port_);
1564 if (port->bundle && port->bundle->bond) {
1565 bond_slave_set_netdev(port->bundle->bond, port, port->up.netdev);
1569 cfm_set_netdev(port->cfm, port->up.netdev);
1572 if (port->is_tunnel && tnl_port_reconfigure(port, port->up.netdev,
1574 ofproto_dpif_cast(port->up.ofproto)->backer->need_revalidate =
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_unref(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_unref(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_unref(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) {
2131 bundle_update(struct ofbundle *bundle)
2133 struct ofport_dpif *port;
2135 bundle->floodable = true;
2136 LIST_FOR_EACH (port, bundle_node, &bundle->ports) {
2137 if (port->up.pp.config & OFPUTIL_PC_NO_FLOOD
2138 || !stp_forward_in_state(port->stp_state)) {
2139 bundle->floodable = false;
2146 bundle_del_port(struct ofport_dpif *port)
2148 struct ofbundle *bundle = port->bundle;
2150 bundle->ofproto->backer->need_revalidate = REV_RECONFIGURE;
2152 list_remove(&port->bundle_node);
2153 port->bundle = NULL;
2156 lacp_slave_unregister(bundle->lacp, port);
2159 bond_slave_unregister(bundle->bond, port);
2162 bundle_update(bundle);
2166 bundle_add_port(struct ofbundle *bundle, ofp_port_t ofp_port,
2167 struct lacp_slave_settings *lacp)
2169 struct ofport_dpif *port;
2171 port = get_ofp_port(bundle->ofproto, ofp_port);
2176 if (port->bundle != bundle) {
2177 bundle->ofproto->backer->need_revalidate = REV_RECONFIGURE;
2179 bundle_del_port(port);
2182 port->bundle = bundle;
2183 list_push_back(&bundle->ports, &port->bundle_node);
2184 if (port->up.pp.config & OFPUTIL_PC_NO_FLOOD
2185 || !stp_forward_in_state(port->stp_state)) {
2186 bundle->floodable = false;
2190 bundle->ofproto->backer->need_revalidate = REV_RECONFIGURE;
2191 lacp_slave_register(bundle->lacp, port, lacp);
2198 bundle_destroy(struct ofbundle *bundle)
2200 struct ofproto_dpif *ofproto;
2201 struct ofport_dpif *port, *next_port;
2207 ofproto = bundle->ofproto;
2208 mbridge_unregister_bundle(ofproto->mbridge, bundle->aux);
2210 LIST_FOR_EACH_SAFE (port, next_port, bundle_node, &bundle->ports) {
2211 bundle_del_port(port);
2214 bundle_flush_macs(bundle, true);
2215 hmap_remove(&ofproto->bundles, &bundle->hmap_node);
2217 free(bundle->trunks);
2218 lacp_unref(bundle->lacp);
2219 bond_unref(bundle->bond);
2224 bundle_set(struct ofproto *ofproto_, void *aux,
2225 const struct ofproto_bundle_settings *s)
2227 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
2228 bool need_flush = false;
2229 struct ofport_dpif *port;
2230 struct ofbundle *bundle;
2231 unsigned long *trunks;
2237 bundle_destroy(bundle_lookup(ofproto, aux));
2241 ovs_assert(s->n_slaves == 1 || s->bond != NULL);
2242 ovs_assert((s->lacp != NULL) == (s->lacp_slaves != NULL));
2244 bundle = bundle_lookup(ofproto, aux);
2246 bundle = xmalloc(sizeof *bundle);
2248 bundle->ofproto = ofproto;
2249 hmap_insert(&ofproto->bundles, &bundle->hmap_node,
2250 hash_pointer(aux, 0));
2252 bundle->name = NULL;
2254 list_init(&bundle->ports);
2255 bundle->vlan_mode = PORT_VLAN_TRUNK;
2257 bundle->trunks = NULL;
2258 bundle->use_priority_tags = s->use_priority_tags;
2259 bundle->lacp = NULL;
2260 bundle->bond = NULL;
2262 bundle->floodable = true;
2263 mbridge_register_bundle(ofproto->mbridge, bundle);
2266 if (!bundle->name || strcmp(s->name, bundle->name)) {
2268 bundle->name = xstrdup(s->name);
2273 if (!bundle->lacp) {
2274 ofproto->backer->need_revalidate = REV_RECONFIGURE;
2275 bundle->lacp = lacp_create();
2277 lacp_configure(bundle->lacp, s->lacp);
2279 lacp_unref(bundle->lacp);
2280 bundle->lacp = NULL;
2283 /* Update set of ports. */
2285 for (i = 0; i < s->n_slaves; i++) {
2286 if (!bundle_add_port(bundle, s->slaves[i],
2287 s->lacp ? &s->lacp_slaves[i] : NULL)) {
2291 if (!ok || list_size(&bundle->ports) != s->n_slaves) {
2292 struct ofport_dpif *next_port;
2294 LIST_FOR_EACH_SAFE (port, next_port, bundle_node, &bundle->ports) {
2295 for (i = 0; i < s->n_slaves; i++) {
2296 if (s->slaves[i] == port->up.ofp_port) {
2301 bundle_del_port(port);
2305 ovs_assert(list_size(&bundle->ports) <= s->n_slaves);
2307 if (list_is_empty(&bundle->ports)) {
2308 bundle_destroy(bundle);
2312 /* Set VLAN tagging mode */
2313 if (s->vlan_mode != bundle->vlan_mode
2314 || s->use_priority_tags != bundle->use_priority_tags) {
2315 bundle->vlan_mode = s->vlan_mode;
2316 bundle->use_priority_tags = s->use_priority_tags;
2321 vlan = (s->vlan_mode == PORT_VLAN_TRUNK ? -1
2322 : s->vlan >= 0 && s->vlan <= 4095 ? s->vlan
2324 if (vlan != bundle->vlan) {
2325 bundle->vlan = vlan;
2329 /* Get trunked VLANs. */
2330 switch (s->vlan_mode) {
2331 case PORT_VLAN_ACCESS:
2335 case PORT_VLAN_TRUNK:
2336 trunks = CONST_CAST(unsigned long *, s->trunks);
2339 case PORT_VLAN_NATIVE_UNTAGGED:
2340 case PORT_VLAN_NATIVE_TAGGED:
2341 if (vlan != 0 && (!s->trunks
2342 || !bitmap_is_set(s->trunks, vlan)
2343 || bitmap_is_set(s->trunks, 0))) {
2344 /* Force trunking the native VLAN and prohibit trunking VLAN 0. */
2346 trunks = bitmap_clone(s->trunks, 4096);
2348 trunks = bitmap_allocate1(4096);
2350 bitmap_set1(trunks, vlan);
2351 bitmap_set0(trunks, 0);
2353 trunks = CONST_CAST(unsigned long *, s->trunks);
2360 if (!vlan_bitmap_equal(trunks, bundle->trunks)) {
2361 free(bundle->trunks);
2362 if (trunks == s->trunks) {
2363 bundle->trunks = vlan_bitmap_clone(trunks);
2365 bundle->trunks = trunks;
2370 if (trunks != s->trunks) {
2375 if (!list_is_short(&bundle->ports)) {
2376 bundle->ofproto->has_bonded_bundles = true;
2378 if (bond_reconfigure(bundle->bond, s->bond)) {
2379 ofproto->backer->need_revalidate = REV_RECONFIGURE;
2382 bundle->bond = bond_create(s->bond);
2383 ofproto->backer->need_revalidate = REV_RECONFIGURE;
2386 LIST_FOR_EACH (port, bundle_node, &bundle->ports) {
2387 bond_slave_register(bundle->bond, port, port->up.netdev);
2390 bond_unref(bundle->bond);
2391 bundle->bond = NULL;
2394 /* If we changed something that would affect MAC learning, un-learn
2395 * everything on this port and force flow revalidation. */
2397 bundle_flush_macs(bundle, false);
2404 bundle_remove(struct ofport *port_)
2406 struct ofport_dpif *port = ofport_dpif_cast(port_);
2407 struct ofbundle *bundle = port->bundle;
2410 bundle_del_port(port);
2411 if (list_is_empty(&bundle->ports)) {
2412 bundle_destroy(bundle);
2413 } else if (list_is_short(&bundle->ports)) {
2414 bond_unref(bundle->bond);
2415 bundle->bond = NULL;
2421 send_pdu_cb(void *port_, const void *pdu, size_t pdu_size)
2423 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 10);
2424 struct ofport_dpif *port = port_;
2425 uint8_t ea[ETH_ADDR_LEN];
2428 error = netdev_get_etheraddr(port->up.netdev, ea);
2430 struct ofpbuf packet;
2433 ofpbuf_init(&packet, 0);
2434 packet_pdu = eth_compose(&packet, eth_addr_lacp, ea, ETH_TYPE_LACP,
2436 memcpy(packet_pdu, pdu, pdu_size);
2438 send_packet(port, &packet);
2439 ofpbuf_uninit(&packet);
2441 VLOG_ERR_RL(&rl, "port %s: cannot obtain Ethernet address of iface "
2442 "%s (%s)", port->bundle->name,
2443 netdev_get_name(port->up.netdev), ovs_strerror(error));
2448 bundle_send_learning_packets(struct ofbundle *bundle)
2450 struct ofproto_dpif *ofproto = bundle->ofproto;
2451 int error, n_packets, n_errors;
2452 struct mac_entry *e;
2454 error = n_packets = n_errors = 0;
2455 LIST_FOR_EACH (e, lru_node, &ofproto->ml->lrus) {
2456 if (e->port.p != bundle) {
2457 struct ofpbuf *learning_packet;
2458 struct ofport_dpif *port;
2462 /* The assignment to "port" is unnecessary but makes "grep"ing for
2463 * struct ofport_dpif more effective. */
2464 learning_packet = bond_compose_learning_packet(bundle->bond,
2468 ret = send_packet(port, learning_packet);
2469 ofpbuf_delete(learning_packet);
2479 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
2480 VLOG_WARN_RL(&rl, "bond %s: %d errors sending %d gratuitous learning "
2481 "packets, last error was: %s",
2482 bundle->name, n_errors, n_packets, ovs_strerror(error));
2484 VLOG_DBG("bond %s: sent %d gratuitous learning packets",
2485 bundle->name, n_packets);
2490 bundle_run(struct ofbundle *bundle)
2493 lacp_run(bundle->lacp, send_pdu_cb);
2496 struct ofport_dpif *port;
2498 LIST_FOR_EACH (port, bundle_node, &bundle->ports) {
2499 bond_slave_set_may_enable(bundle->bond, port, port->may_enable);
2502 bond_run(bundle->bond, &bundle->ofproto->backer->revalidate_set,
2503 lacp_status(bundle->lacp));
2504 if (bond_should_send_learning_packets(bundle->bond)) {
2505 bundle_send_learning_packets(bundle);
2511 bundle_wait(struct ofbundle *bundle)
2514 lacp_wait(bundle->lacp);
2517 bond_wait(bundle->bond);
2524 mirror_set__(struct ofproto *ofproto_, void *aux,
2525 const struct ofproto_mirror_settings *s)
2527 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
2528 struct ofbundle **srcs, **dsts;
2533 mirror_destroy(ofproto->mbridge, aux);
2537 srcs = xmalloc(s->n_srcs * sizeof *srcs);
2538 dsts = xmalloc(s->n_dsts * sizeof *dsts);
2540 for (i = 0; i < s->n_srcs; i++) {
2541 srcs[i] = bundle_lookup(ofproto, s->srcs[i]);
2544 for (i = 0; i < s->n_dsts; i++) {
2545 dsts[i] = bundle_lookup(ofproto, s->dsts[i]);
2548 error = mirror_set(ofproto->mbridge, aux, s->name, srcs, s->n_srcs, dsts,
2549 s->n_dsts, s->src_vlans,
2550 bundle_lookup(ofproto, s->out_bundle), s->out_vlan);
2557 mirror_get_stats__(struct ofproto *ofproto, void *aux,
2558 uint64_t *packets, uint64_t *bytes)
2561 return mirror_get_stats(ofproto_dpif_cast(ofproto)->mbridge, aux, packets,
2566 set_flood_vlans(struct ofproto *ofproto_, unsigned long *flood_vlans)
2568 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
2569 if (mac_learning_set_flood_vlans(ofproto->ml, flood_vlans)) {
2570 mac_learning_flush(ofproto->ml, &ofproto->backer->revalidate_set);
2576 is_mirror_output_bundle(const struct ofproto *ofproto_, void *aux)
2578 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
2579 struct ofbundle *bundle = bundle_lookup(ofproto, aux);
2580 return bundle && mirror_bundle_out(ofproto->mbridge, bundle) != 0;
2584 forward_bpdu_changed(struct ofproto *ofproto_)
2586 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
2587 ofproto->backer->need_revalidate = REV_RECONFIGURE;
2591 set_mac_table_config(struct ofproto *ofproto_, unsigned int idle_time,
2594 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
2595 mac_learning_set_idle_time(ofproto->ml, idle_time);
2596 mac_learning_set_max_entries(ofproto->ml, max_entries);
2601 struct ofport_dpif *
2602 get_ofp_port(const struct ofproto_dpif *ofproto, ofp_port_t ofp_port)
2604 struct ofport *ofport = ofproto_get_port(&ofproto->up, ofp_port);
2605 return ofport ? ofport_dpif_cast(ofport) : NULL;
2608 struct ofport_dpif *
2609 get_odp_port(const struct ofproto_dpif *ofproto, odp_port_t odp_port)
2611 struct ofport_dpif *port = odp_port_to_ofport(ofproto->backer, odp_port);
2612 return port && &ofproto->up == port->up.ofproto ? port : NULL;
2616 ofproto_port_from_dpif_port(struct ofproto_dpif *ofproto,
2617 struct ofproto_port *ofproto_port,
2618 struct dpif_port *dpif_port)
2620 ofproto_port->name = dpif_port->name;
2621 ofproto_port->type = dpif_port->type;
2622 ofproto_port->ofp_port = odp_port_to_ofp_port(ofproto, dpif_port->port_no);
2626 ofport_update_peer(struct ofport_dpif *ofport)
2628 const struct ofproto_dpif *ofproto;
2629 struct dpif_backer *backer;
2630 const char *peer_name;
2632 if (!netdev_vport_is_patch(ofport->up.netdev)) {
2636 backer = ofproto_dpif_cast(ofport->up.ofproto)->backer;
2637 backer->need_revalidate = REV_RECONFIGURE;
2640 ofport->peer->peer = NULL;
2641 ofport->peer = NULL;
2644 peer_name = netdev_vport_patch_peer(ofport->up.netdev);
2649 HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
2650 struct ofport *peer_ofport;
2651 struct ofport_dpif *peer;
2652 const char *peer_peer;
2654 if (ofproto->backer != backer) {
2658 peer_ofport = shash_find_data(&ofproto->up.port_by_name, peer_name);
2663 peer = ofport_dpif_cast(peer_ofport);
2664 peer_peer = netdev_vport_patch_peer(peer->up.netdev);
2665 if (peer_peer && !strcmp(netdev_get_name(ofport->up.netdev),
2667 ofport->peer = peer;
2668 ofport->peer->peer = ofport;
2676 port_run_fast(struct ofport_dpif *ofport)
2678 if (ofport->cfm && cfm_should_send_ccm(ofport->cfm)) {
2679 struct ofpbuf packet;
2681 ofpbuf_init(&packet, 0);
2682 cfm_compose_ccm(ofport->cfm, &packet, ofport->up.pp.hw_addr);
2683 send_packet(ofport, &packet);
2684 ofpbuf_uninit(&packet);
2687 if (ofport->bfd && bfd_should_send_packet(ofport->bfd)) {
2688 struct ofpbuf packet;
2690 ofpbuf_init(&packet, 0);
2691 bfd_put_packet(ofport->bfd, &packet, ofport->up.pp.hw_addr);
2692 send_packet(ofport, &packet);
2693 ofpbuf_uninit(&packet);
2698 port_run(struct ofport_dpif *ofport)
2700 long long int carrier_seq = netdev_get_carrier_resets(ofport->up.netdev);
2701 bool carrier_changed = carrier_seq != ofport->carrier_seq;
2702 bool enable = netdev_get_carrier(ofport->up.netdev);
2704 ofport->carrier_seq = carrier_seq;
2706 port_run_fast(ofport);
2709 int cfm_opup = cfm_get_opup(ofport->cfm);
2711 cfm_run(ofport->cfm);
2712 enable = enable && !cfm_get_fault(ofport->cfm);
2714 if (cfm_opup >= 0) {
2715 enable = enable && cfm_opup;
2720 bfd_run(ofport->bfd);
2721 enable = enable && bfd_forwarding(ofport->bfd);
2724 if (ofport->bundle) {
2725 enable = enable && lacp_slave_may_enable(ofport->bundle->lacp, ofport);
2726 if (carrier_changed) {
2727 lacp_slave_carrier_changed(ofport->bundle->lacp, ofport);
2731 if (ofport->may_enable != enable) {
2732 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport->up.ofproto);
2733 ofproto->backer->need_revalidate = REV_PORT_TOGGLED;
2736 ofport->may_enable = enable;
2740 port_wait(struct ofport_dpif *ofport)
2743 cfm_wait(ofport->cfm);
2747 bfd_wait(ofport->bfd);
2752 port_query_by_name(const struct ofproto *ofproto_, const char *devname,
2753 struct ofproto_port *ofproto_port)
2755 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
2756 struct dpif_port dpif_port;
2759 if (sset_contains(&ofproto->ghost_ports, devname)) {
2760 const char *type = netdev_get_type_from_name(devname);
2762 /* We may be called before ofproto->up.port_by_name is populated with
2763 * the appropriate ofport. For this reason, we must get the name and
2764 * type from the netdev layer directly. */
2766 const struct ofport *ofport;
2768 ofport = shash_find_data(&ofproto->up.port_by_name, devname);
2769 ofproto_port->ofp_port = ofport ? ofport->ofp_port : OFPP_NONE;
2770 ofproto_port->name = xstrdup(devname);
2771 ofproto_port->type = xstrdup(type);
2777 if (!sset_contains(&ofproto->ports, devname)) {
2780 error = dpif_port_query_by_name(ofproto->backer->dpif,
2781 devname, &dpif_port);
2783 ofproto_port_from_dpif_port(ofproto, ofproto_port, &dpif_port);
2789 port_add(struct ofproto *ofproto_, struct netdev *netdev)
2791 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
2792 const char *devname = netdev_get_name(netdev);
2793 char namebuf[NETDEV_VPORT_NAME_BUFSIZE];
2794 const char *dp_port_name;
2796 if (netdev_vport_is_patch(netdev)) {
2797 sset_add(&ofproto->ghost_ports, netdev_get_name(netdev));
2801 dp_port_name = netdev_vport_get_dpif_port(netdev, namebuf, sizeof namebuf);
2802 if (!dpif_port_exists(ofproto->backer->dpif, dp_port_name)) {
2803 odp_port_t port_no = ODPP_NONE;
2806 error = dpif_port_add(ofproto->backer->dpif, netdev, &port_no);
2810 if (netdev_get_tunnel_config(netdev)) {
2811 simap_put(&ofproto->backer->tnl_backers,
2812 dp_port_name, odp_to_u32(port_no));
2816 if (netdev_get_tunnel_config(netdev)) {
2817 sset_add(&ofproto->ghost_ports, devname);
2819 sset_add(&ofproto->ports, devname);
2825 port_del(struct ofproto *ofproto_, ofp_port_t ofp_port)
2827 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
2828 struct ofport_dpif *ofport = get_ofp_port(ofproto, ofp_port);
2835 sset_find_and_delete(&ofproto->ghost_ports,
2836 netdev_get_name(ofport->up.netdev));
2837 ofproto->backer->need_revalidate = REV_RECONFIGURE;
2838 if (!ofport->is_tunnel) {
2839 error = dpif_port_del(ofproto->backer->dpif, ofport->odp_port);
2841 /* The caller is going to close ofport->up.netdev. If this is a
2842 * bonded port, then the bond is using that netdev, so remove it
2843 * from the bond. The client will need to reconfigure everything
2844 * after deleting ports, so then the slave will get re-added. */
2845 bundle_remove(&ofport->up);
2852 port_get_stats(const struct ofport *ofport_, struct netdev_stats *stats)
2854 struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
2859 error = netdev_get_stats(ofport->up.netdev, stats);
2861 if (!error && ofport_->ofp_port == OFPP_LOCAL) {
2862 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport->up.ofproto);
2864 /* ofproto->stats.tx_packets represents packets that we created
2865 * internally and sent to some port (e.g. packets sent with
2866 * send_packet()). Account for them as if they had come from
2867 * OFPP_LOCAL and got forwarded. */
2869 if (stats->rx_packets != UINT64_MAX) {
2870 stats->rx_packets += ofproto->stats.tx_packets;
2873 if (stats->rx_bytes != UINT64_MAX) {
2874 stats->rx_bytes += ofproto->stats.tx_bytes;
2877 /* ofproto->stats.rx_packets represents packets that were received on
2878 * some port and we processed internally and dropped (e.g. STP).
2879 * Account for them as if they had been forwarded to OFPP_LOCAL. */
2881 if (stats->tx_packets != UINT64_MAX) {
2882 stats->tx_packets += ofproto->stats.rx_packets;
2885 if (stats->tx_bytes != UINT64_MAX) {
2886 stats->tx_bytes += ofproto->stats.rx_bytes;
2893 struct port_dump_state {
2898 struct ofproto_port port;
2903 port_dump_start(const struct ofproto *ofproto_ OVS_UNUSED, void **statep)
2905 *statep = xzalloc(sizeof(struct port_dump_state));
2910 port_dump_next(const struct ofproto *ofproto_, void *state_,
2911 struct ofproto_port *port)
2913 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
2914 struct port_dump_state *state = state_;
2915 const struct sset *sset;
2916 struct sset_node *node;
2918 if (state->has_port) {
2919 ofproto_port_destroy(&state->port);
2920 state->has_port = false;
2922 sset = state->ghost ? &ofproto->ghost_ports : &ofproto->ports;
2923 while ((node = sset_at_position(sset, &state->bucket, &state->offset))) {
2926 error = port_query_by_name(ofproto_, node->name, &state->port);
2928 *port = state->port;
2929 state->has_port = true;
2931 } else if (error != ENODEV) {
2936 if (!state->ghost) {
2937 state->ghost = true;
2940 return port_dump_next(ofproto_, state_, port);
2947 port_dump_done(const struct ofproto *ofproto_ OVS_UNUSED, void *state_)
2949 struct port_dump_state *state = state_;
2951 if (state->has_port) {
2952 ofproto_port_destroy(&state->port);
2959 port_poll(const struct ofproto *ofproto_, char **devnamep)
2961 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
2963 if (ofproto->port_poll_errno) {
2964 int error = ofproto->port_poll_errno;
2965 ofproto->port_poll_errno = 0;
2969 if (sset_is_empty(&ofproto->port_poll_set)) {
2973 *devnamep = sset_pop(&ofproto->port_poll_set);
2978 port_poll_wait(const struct ofproto *ofproto_)
2980 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
2981 dpif_port_poll_wait(ofproto->backer->dpif);
2985 port_is_lacp_current(const struct ofport *ofport_)
2987 const struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
2988 return (ofport->bundle && ofport->bundle->lacp
2989 ? lacp_slave_is_current(ofport->bundle->lacp, ofport)
2993 /* Upcall handling. */
2995 /* Flow miss batching.
2997 * Some dpifs implement operations faster when you hand them off in a batch.
2998 * To allow batching, "struct flow_miss" queues the dpif-related work needed
2999 * for a given flow. Each "struct flow_miss" corresponds to sending one or
3000 * more packets, plus possibly installing the flow in the dpif.
3002 * So far we only batch the operations that affect flow setup time the most.
3003 * It's possible to batch more than that, but the benefit might be minimal. */
3005 struct hmap_node hmap_node;
3006 struct ofproto_dpif *ofproto;
3008 enum odp_key_fitness key_fitness;
3009 const struct nlattr *key;
3011 struct list packets;
3012 enum dpif_upcall_type upcall_type;
3015 struct flow_miss_op {
3016 struct dpif_op dpif_op;
3018 uint64_t slow_stub[128 / 8]; /* Buffer for compose_slow_path() */
3019 struct xlate_out xout;
3020 bool xout_garbage; /* 'xout' needs to be uninitialized? */
3022 struct ofpbuf mask; /* Flow mask for "put" ops. */
3023 struct odputil_keybuf maskbuf;
3025 /* If this is a "put" op, then a pointer to the subfacet that should
3026 * be marked as uninstalled if the operation fails. */
3027 struct subfacet *subfacet;
3030 /* Sends an OFPT_PACKET_IN message for 'packet' of type OFPR_NO_MATCH to each
3031 * OpenFlow controller as necessary according to their individual
3032 * configurations. */
3034 send_packet_in_miss(struct ofproto_dpif *ofproto, const struct ofpbuf *packet,
3035 const struct flow *flow)
3037 struct ofputil_packet_in pin;
3039 pin.packet = packet->data;
3040 pin.packet_len = packet->size;
3041 pin.reason = OFPR_NO_MATCH;
3042 pin.controller_id = 0;
3047 pin.send_len = 0; /* not used for flow table misses */
3049 flow_get_metadata(flow, &pin.fmd);
3051 connmgr_send_packet_in(ofproto->up.connmgr, &pin);
3054 static struct flow_miss *
3055 flow_miss_find(struct hmap *todo, const struct ofproto_dpif *ofproto,
3056 const struct flow *flow, uint32_t hash)
3058 struct flow_miss *miss;
3060 HMAP_FOR_EACH_WITH_HASH (miss, hmap_node, hash, todo) {
3061 if (miss->ofproto == ofproto && flow_equal(&miss->flow, flow)) {
3069 /* Partially Initializes 'op' as an "execute" operation for 'miss' and
3070 * 'packet'. The caller must initialize op->actions and op->actions_len. If
3071 * 'miss' is associated with a subfacet the caller must also initialize the
3072 * returned op->subfacet, and if anything needs to be freed after processing
3073 * the op, the caller must initialize op->garbage also. */
3075 init_flow_miss_execute_op(struct flow_miss *miss, struct ofpbuf *packet,
3076 struct flow_miss_op *op)
3078 if (miss->flow.in_port.ofp_port
3079 != vsp_realdev_to_vlandev(miss->ofproto, miss->flow.in_port.ofp_port,
3080 miss->flow.vlan_tci)) {
3081 /* This packet was received on a VLAN splinter port. We
3082 * added a VLAN to the packet to make the packet resemble
3083 * the flow, but the actions were composed assuming that
3084 * the packet contained no VLAN. So, we must remove the
3085 * VLAN header from the packet before trying to execute the
3087 eth_pop_vlan(packet);
3090 op->subfacet = NULL;
3091 op->xout_garbage = false;
3092 op->dpif_op.type = DPIF_OP_EXECUTE;
3093 op->dpif_op.u.execute.key = miss->key;
3094 op->dpif_op.u.execute.key_len = miss->key_len;
3095 op->dpif_op.u.execute.packet = packet;
3096 ofpbuf_use_stack(&op->mask, &op->maskbuf, sizeof op->maskbuf);
3099 /* Helper for handle_flow_miss_without_facet() and
3100 * handle_flow_miss_with_facet(). */
3102 handle_flow_miss_common(struct rule_dpif *rule,
3103 struct ofpbuf *packet, const struct flow *flow)
3105 struct ofproto_dpif *ofproto = ofproto_dpif_cast(rule->up.ofproto);
3107 if (rule->up.cr.priority == FAIL_OPEN_PRIORITY) {
3109 * Extra-special case for fail-open mode.
3111 * We are in fail-open mode and the packet matched the fail-open
3112 * rule, but we are connected to a controller too. We should send
3113 * the packet up to the controller in the hope that it will try to
3114 * set up a flow and thereby allow us to exit fail-open.
3116 * See the top-level comment in fail-open.c for more information.
3118 send_packet_in_miss(ofproto, packet, flow);
3122 /* Figures out whether a flow that missed in 'ofproto', whose details are in
3123 * 'miss' masked by 'wc', is likely to be worth tracking in detail in userspace
3124 * and (usually) installing a datapath flow. The answer is usually "yes" (a
3125 * return value of true). However, for short flows the cost of bookkeeping is
3126 * much higher than the benefits, so when the datapath holds a large number of
3127 * flows we impose some heuristics to decide which flows are likely to be worth
3130 flow_miss_should_make_facet(struct flow_miss *miss, struct flow_wildcards *wc)
3132 struct dpif_backer *backer = miss->ofproto->backer;
3135 switch (flow_miss_model) {
3136 case OFPROTO_HANDLE_MISS_AUTO:
3138 case OFPROTO_HANDLE_MISS_WITH_FACETS:
3140 case OFPROTO_HANDLE_MISS_WITHOUT_FACETS:
3144 if (!backer->governor) {
3147 n_subfacets = hmap_count(&backer->subfacets);
3148 if (n_subfacets * 2 <= flow_eviction_threshold) {
3152 backer->governor = governor_create();
3155 hash = flow_hash_in_wildcards(&miss->flow, wc, 0);
3156 return governor_should_install_flow(backer->governor, hash,
3157 list_size(&miss->packets));
3160 /* Handles 'miss' without creating a facet or subfacet or creating any datapath
3161 * flow. 'miss->flow' must have matched 'rule' and been xlated into 'xout'.
3162 * May add an "execute" operation to 'ops' and increment '*n_ops'. */
3164 handle_flow_miss_without_facet(struct rule_dpif *rule, struct xlate_out *xout,
3165 struct flow_miss *miss,
3166 struct flow_miss_op *ops, size_t *n_ops)
3168 struct ofpbuf *packet;
3170 LIST_FOR_EACH (packet, list_node, &miss->packets) {
3172 COVERAGE_INC(facet_suppress);
3174 handle_flow_miss_common(rule, packet, &miss->flow);
3177 struct xlate_in xin;
3179 xlate_in_init(&xin, miss->ofproto, &miss->flow, rule, 0, packet);
3180 xlate_actions_for_side_effects(&xin);
3183 if (xout->odp_actions.size) {
3184 struct flow_miss_op *op = &ops[*n_ops];
3185 struct dpif_execute *execute = &op->dpif_op.u.execute;
3187 init_flow_miss_execute_op(miss, packet, op);
3188 xlate_out_copy(&op->xout, xout);
3189 execute->actions = op->xout.odp_actions.data;
3190 execute->actions_len = op->xout.odp_actions.size;
3191 op->xout_garbage = true;
3198 /* Handles 'miss', which matches 'facet'. May add any required datapath
3199 * operations to 'ops', incrementing '*n_ops' for each new op.
3201 * All of the packets in 'miss' are considered to have arrived at time 'now'.
3202 * This is really important only for new facets: if we just called time_msec()
3203 * here, then the new subfacet or its packets could look (occasionally) as
3204 * though it was used some time after the facet was used. That can make a
3205 * one-packet flow look like it has a nonzero duration, which looks odd in
3206 * e.g. NetFlow statistics.
3208 * If non-null, 'stats' will be folded into 'facet'. */
3210 handle_flow_miss_with_facet(struct flow_miss *miss, struct facet *facet,
3211 long long int now, struct dpif_flow_stats *stats,
3212 struct flow_miss_op *ops, size_t *n_ops)
3214 struct ofproto_dpif *ofproto = ofproto_dpif_cast(facet->rule->up.ofproto);
3215 enum subfacet_path want_path;
3216 struct subfacet *subfacet;
3217 struct ofpbuf *packet;
3219 subfacet = subfacet_create(facet, miss, now);
3220 want_path = facet->xout.slow ? SF_SLOW_PATH : SF_FAST_PATH;
3222 subfacet_update_stats(subfacet, stats);
3225 LIST_FOR_EACH (packet, list_node, &miss->packets) {
3226 struct flow_miss_op *op = &ops[*n_ops];
3228 handle_flow_miss_common(facet->rule, packet, &miss->flow);
3230 if (want_path != SF_FAST_PATH) {
3231 struct xlate_in xin;
3233 xlate_in_init(&xin, ofproto, &miss->flow, facet->rule, 0, packet);
3234 xlate_actions_for_side_effects(&xin);
3237 if (facet->xout.odp_actions.size) {
3238 struct dpif_execute *execute = &op->dpif_op.u.execute;
3240 init_flow_miss_execute_op(miss, packet, op);
3241 execute->actions = facet->xout.odp_actions.data,
3242 execute->actions_len = facet->xout.odp_actions.size;
3247 if (miss->upcall_type == DPIF_UC_MISS || subfacet->path != want_path) {
3248 struct flow_miss_op *op = &ops[(*n_ops)++];
3249 struct dpif_flow_put *put = &op->dpif_op.u.flow_put;
3251 subfacet->path = want_path;
3253 ofpbuf_use_stack(&op->mask, &op->maskbuf, sizeof op->maskbuf);
3254 if (enable_megaflows) {
3255 odp_flow_key_from_mask(&op->mask, &facet->xout.wc.masks,
3256 &miss->flow, UINT32_MAX);
3259 op->xout_garbage = false;
3260 op->dpif_op.type = DPIF_OP_FLOW_PUT;
3261 op->subfacet = subfacet;
3262 put->flags = DPIF_FP_CREATE | DPIF_FP_MODIFY;
3263 put->key = miss->key;
3264 put->key_len = miss->key_len;
3265 put->mask = op->mask.data;
3266 put->mask_len = op->mask.size;
3268 if (want_path == SF_FAST_PATH) {
3269 put->actions = facet->xout.odp_actions.data;
3270 put->actions_len = facet->xout.odp_actions.size;
3272 compose_slow_path(ofproto, &miss->flow, facet->xout.slow,
3273 op->slow_stub, sizeof op->slow_stub,
3274 &put->actions, &put->actions_len);
3280 /* Handles flow miss 'miss'. May add any required datapath operations
3281 * to 'ops', incrementing '*n_ops' for each new op. */
3283 handle_flow_miss(struct flow_miss *miss, struct flow_miss_op *ops,
3286 struct ofproto_dpif *ofproto = miss->ofproto;
3287 struct dpif_flow_stats stats__;
3288 struct dpif_flow_stats *stats = &stats__;
3289 struct ofpbuf *packet;
3290 struct facet *facet;
3294 memset(stats, 0, sizeof *stats);
3296 LIST_FOR_EACH (packet, list_node, &miss->packets) {
3297 stats->tcp_flags |= packet_get_tcp_flags(packet, &miss->flow);
3298 stats->n_bytes += packet->size;
3302 facet = facet_lookup_valid(ofproto, &miss->flow);
3304 struct flow_wildcards wc;
3305 struct rule_dpif *rule;
3306 struct xlate_out xout;
3307 struct xlate_in xin;
3309 flow_wildcards_init_catchall(&wc);
3310 rule = rule_dpif_lookup(ofproto, &miss->flow, &wc);
3311 rule_credit_stats(rule, stats);
3313 xlate_in_init(&xin, ofproto, &miss->flow, rule, stats->tcp_flags,
3315 xin.resubmit_stats = stats;
3316 xin.may_learn = true;
3317 xlate_actions(&xin, &xout);
3318 flow_wildcards_or(&xout.wc, &xout.wc, &wc);
3320 /* There does not exist a bijection between 'struct flow' and datapath
3321 * flow keys with fitness ODP_FIT_TO_LITTLE. This breaks a fundamental
3322 * assumption used throughout the facet and subfacet handling code.
3323 * Since we have to handle these misses in userspace anyway, we simply
3324 * skip facet creation, avoiding the problem altogether. */
3325 if (miss->key_fitness == ODP_FIT_TOO_LITTLE
3326 || !flow_miss_should_make_facet(miss, &xout.wc)) {
3327 handle_flow_miss_without_facet(rule, &xout, miss, ops, n_ops);
3331 facet = facet_create(miss, rule, &xout, stats);
3334 handle_flow_miss_with_facet(miss, facet, now, stats, ops, n_ops);
3337 static struct drop_key *
3338 drop_key_lookup(const struct dpif_backer *backer, const struct nlattr *key,
3341 struct drop_key *drop_key;
3343 HMAP_FOR_EACH_WITH_HASH (drop_key, hmap_node, hash_bytes(key, key_len, 0),
3344 &backer->drop_keys) {
3345 if (drop_key->key_len == key_len
3346 && !memcmp(drop_key->key, key, key_len)) {
3354 drop_key_clear(struct dpif_backer *backer)
3356 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 15);
3357 struct drop_key *drop_key, *next;
3359 HMAP_FOR_EACH_SAFE (drop_key, next, hmap_node, &backer->drop_keys) {
3362 error = dpif_flow_del(backer->dpif, drop_key->key, drop_key->key_len,
3364 if (error && !VLOG_DROP_WARN(&rl)) {
3365 struct ds ds = DS_EMPTY_INITIALIZER;
3366 odp_flow_key_format(drop_key->key, drop_key->key_len, &ds);
3367 VLOG_WARN("Failed to delete drop key (%s) (%s)",
3368 ovs_strerror(error), ds_cstr(&ds));
3372 hmap_remove(&backer->drop_keys, &drop_key->hmap_node);
3373 free(drop_key->key);
3378 /* Given a datpath, packet, and flow metadata ('backer', 'packet', and 'key'
3379 * respectively), populates 'flow' with the result of odp_flow_key_to_flow().
3380 * Optionally, if nonnull, populates 'fitnessp' with the fitness of 'flow' as
3381 * returned by odp_flow_key_to_flow(). Also, optionally populates 'ofproto'
3382 * with the ofproto_dpif, and 'odp_in_port' with the datapath in_port, that
3383 * 'packet' ingressed.
3385 * If 'ofproto' is nonnull, requires 'flow''s in_port to exist. Otherwise sets
3386 * 'flow''s in_port to OFPP_NONE.
3388 * This function does post-processing on data returned from
3389 * odp_flow_key_to_flow() to help make VLAN splinters transparent to the rest
3390 * of the upcall processing logic. In particular, if the extracted in_port is
3391 * a VLAN splinter port, it replaces flow->in_port by the "real" port, sets
3392 * flow->vlan_tci correctly for the VLAN of the VLAN splinter port, and pushes
3393 * a VLAN header onto 'packet' (if it is nonnull).
3395 * Similarly, this function also includes some logic to help with tunnels. It
3396 * may modify 'flow' as necessary to make the tunneling implementation
3397 * transparent to the upcall processing logic.
3399 * Returns 0 if successful, ENODEV if the parsed flow has no associated ofport,
3400 * or some other positive errno if there are other problems. */
3402 ofproto_receive(const struct dpif_backer *backer, struct ofpbuf *packet,
3403 const struct nlattr *key, size_t key_len,
3404 struct flow *flow, enum odp_key_fitness *fitnessp,
3405 struct ofproto_dpif **ofproto, odp_port_t *odp_in_port)
3407 const struct ofport_dpif *port;
3408 enum odp_key_fitness fitness;
3411 fitness = odp_flow_key_to_flow(key, key_len, flow);
3412 if (fitness == ODP_FIT_ERROR) {
3418 *odp_in_port = flow->in_port.odp_port;
3421 port = (tnl_port_should_receive(flow)
3422 ? tnl_port_receive(flow)
3423 : odp_port_to_ofport(backer, flow->in_port.odp_port));
3424 flow->in_port.ofp_port = port ? port->up.ofp_port : OFPP_NONE;
3429 /* XXX: Since the tunnel module is not scoped per backer, for a tunnel port
3430 * it's theoretically possible that we'll receive an ofport belonging to an
3431 * entirely different datapath. In practice, this can't happen because no
3432 * platforms has two separate datapaths which each support tunneling. */
3433 ovs_assert(ofproto_dpif_cast(port->up.ofproto)->backer == backer);
3435 if (vsp_adjust_flow(ofproto_dpif_cast(port->up.ofproto), flow)) {
3437 /* Make the packet resemble the flow, so that it gets sent to
3438 * an OpenFlow controller properly, so that it looks correct
3439 * for sFlow, and so that flow_extract() will get the correct
3440 * vlan_tci if it is called on 'packet'.
3442 * The allocated space inside 'packet' probably also contains
3443 * 'key', that is, both 'packet' and 'key' are probably part of
3444 * a struct dpif_upcall (see the large comment on that
3445 * structure definition), so pushing data on 'packet' is in
3446 * general not a good idea since it could overwrite 'key' or
3447 * free it as a side effect. However, it's OK in this special
3448 * case because we know that 'packet' is inside a Netlink
3449 * attribute: pushing 4 bytes will just overwrite the 4-byte
3450 * "struct nlattr", which is fine since we don't need that
3451 * header anymore. */
3452 eth_push_vlan(packet, flow->vlan_tci);
3454 /* We can't reproduce 'key' from 'flow'. */
3455 fitness = fitness == ODP_FIT_PERFECT ? ODP_FIT_TOO_MUCH : fitness;
3460 *ofproto = ofproto_dpif_cast(port->up.ofproto);
3465 *fitnessp = fitness;
3471 handle_miss_upcalls(struct dpif_backer *backer, struct dpif_upcall *upcalls,
3474 struct dpif_upcall *upcall;
3475 struct flow_miss *miss;
3476 struct flow_miss misses[FLOW_MISS_MAX_BATCH];
3477 struct flow_miss_op flow_miss_ops[FLOW_MISS_MAX_BATCH * 2];
3478 struct dpif_op *dpif_ops[FLOW_MISS_MAX_BATCH * 2];
3488 /* Construct the to-do list.
3490 * This just amounts to extracting the flow from each packet and sticking
3491 * the packets that have the same flow in the same "flow_miss" structure so
3492 * that we can process them together. */
3495 for (upcall = upcalls; upcall < &upcalls[n_upcalls]; upcall++) {
3496 struct flow_miss *miss = &misses[n_misses];
3497 struct flow_miss *existing_miss;
3498 struct ofproto_dpif *ofproto;
3499 odp_port_t odp_in_port;
3504 error = ofproto_receive(backer, upcall->packet, upcall->key,
3505 upcall->key_len, &flow, &miss->key_fitness,
3506 &ofproto, &odp_in_port);
3507 if (error == ENODEV) {
3508 struct drop_key *drop_key;
3510 /* Received packet on datapath port for which we couldn't
3511 * associate an ofproto. This can happen if a port is removed
3512 * while traffic is being received. Print a rate-limited message
3513 * in case it happens frequently. Install a drop flow so
3514 * that future packets of the flow are inexpensively dropped
3516 VLOG_INFO_RL(&rl, "received packet on unassociated datapath port "
3517 "%"PRIu32, odp_in_port);
3519 drop_key = drop_key_lookup(backer, upcall->key, upcall->key_len);
3521 drop_key = xmalloc(sizeof *drop_key);
3522 drop_key->key = xmemdup(upcall->key, upcall->key_len);
3523 drop_key->key_len = upcall->key_len;
3525 hmap_insert(&backer->drop_keys, &drop_key->hmap_node,
3526 hash_bytes(drop_key->key, drop_key->key_len, 0));
3527 dpif_flow_put(backer->dpif, DPIF_FP_CREATE | DPIF_FP_MODIFY,
3528 drop_key->key, drop_key->key_len,
3529 NULL, 0, NULL, 0, NULL);
3537 ofproto->n_missed++;
3538 flow_extract(upcall->packet, flow.skb_priority, flow.skb_mark,
3539 &flow.tunnel, &flow.in_port, &miss->flow);
3541 /* Add other packets to a to-do list. */
3542 hash = flow_hash(&miss->flow, 0);
3543 existing_miss = flow_miss_find(&todo, ofproto, &miss->flow, hash);
3544 if (!existing_miss) {
3545 hmap_insert(&todo, &miss->hmap_node, hash);
3546 miss->ofproto = ofproto;
3547 miss->key = upcall->key;
3548 miss->key_len = upcall->key_len;
3549 miss->upcall_type = upcall->type;
3550 list_init(&miss->packets);
3554 miss = existing_miss;
3556 list_push_back(&miss->packets, &upcall->packet->list_node);
3559 /* Process each element in the to-do list, constructing the set of
3560 * operations to batch. */
3562 HMAP_FOR_EACH (miss, hmap_node, &todo) {
3563 handle_flow_miss(miss, flow_miss_ops, &n_ops);
3565 ovs_assert(n_ops <= ARRAY_SIZE(flow_miss_ops));
3567 /* Execute batch. */
3568 for (i = 0; i < n_ops; i++) {
3569 dpif_ops[i] = &flow_miss_ops[i].dpif_op;
3571 dpif_operate(backer->dpif, dpif_ops, n_ops);
3573 for (i = 0; i < n_ops; i++) {
3574 if (dpif_ops[i]->error != 0
3575 && flow_miss_ops[i].dpif_op.type == DPIF_OP_FLOW_PUT
3576 && flow_miss_ops[i].subfacet) {
3577 struct subfacet *subfacet = flow_miss_ops[i].subfacet;
3579 COVERAGE_INC(subfacet_install_fail);
3581 subfacet->path = SF_NOT_INSTALLED;
3585 if (flow_miss_ops[i].xout_garbage) {
3586 xlate_out_uninit(&flow_miss_ops[i].xout);
3589 hmap_destroy(&todo);
3592 static enum { SFLOW_UPCALL, MISS_UPCALL, BAD_UPCALL, FLOW_SAMPLE_UPCALL,
3594 classify_upcall(const struct dpif_upcall *upcall)
3596 size_t userdata_len;
3597 union user_action_cookie cookie;
3599 /* First look at the upcall type. */
3600 switch (upcall->type) {
3601 case DPIF_UC_ACTION:
3607 case DPIF_N_UC_TYPES:
3609 VLOG_WARN_RL(&rl, "upcall has unexpected type %"PRIu32, upcall->type);
3613 /* "action" upcalls need a closer look. */
3614 if (!upcall->userdata) {
3615 VLOG_WARN_RL(&rl, "action upcall missing cookie");
3618 userdata_len = nl_attr_get_size(upcall->userdata);
3619 if (userdata_len < sizeof cookie.type
3620 || userdata_len > sizeof cookie) {
3621 VLOG_WARN_RL(&rl, "action upcall cookie has unexpected size %zu",
3625 memset(&cookie, 0, sizeof cookie);
3626 memcpy(&cookie, nl_attr_get(upcall->userdata), userdata_len);
3627 if (userdata_len == sizeof cookie.sflow
3628 && cookie.type == USER_ACTION_COOKIE_SFLOW) {
3629 return SFLOW_UPCALL;
3630 } else if (userdata_len == sizeof cookie.slow_path
3631 && cookie.type == USER_ACTION_COOKIE_SLOW_PATH) {
3633 } else if (userdata_len == sizeof cookie.flow_sample
3634 && cookie.type == USER_ACTION_COOKIE_FLOW_SAMPLE) {
3635 return FLOW_SAMPLE_UPCALL;
3636 } else if (userdata_len == sizeof cookie.ipfix
3637 && cookie.type == USER_ACTION_COOKIE_IPFIX) {
3638 return IPFIX_UPCALL;
3640 VLOG_WARN_RL(&rl, "invalid user cookie of type %"PRIu16
3641 " and size %zu", cookie.type, userdata_len);
3647 handle_sflow_upcall(struct dpif_backer *backer,
3648 const struct dpif_upcall *upcall)
3650 struct ofproto_dpif *ofproto;
3651 union user_action_cookie cookie;
3653 odp_port_t odp_in_port;
3655 if (ofproto_receive(backer, upcall->packet, upcall->key, upcall->key_len,
3656 &flow, NULL, &ofproto, &odp_in_port)
3657 || !ofproto->sflow) {
3661 memset(&cookie, 0, sizeof cookie);
3662 memcpy(&cookie, nl_attr_get(upcall->userdata), sizeof cookie.sflow);
3663 dpif_sflow_received(ofproto->sflow, upcall->packet, &flow,
3664 odp_in_port, &cookie);
3668 handle_flow_sample_upcall(struct dpif_backer *backer,
3669 const struct dpif_upcall *upcall)
3671 struct ofproto_dpif *ofproto;
3672 union user_action_cookie cookie;
3675 if (ofproto_receive(backer, upcall->packet, upcall->key, upcall->key_len,
3676 &flow, NULL, &ofproto, NULL)
3677 || !ofproto->ipfix) {
3681 memset(&cookie, 0, sizeof cookie);
3682 memcpy(&cookie, nl_attr_get(upcall->userdata), sizeof cookie.flow_sample);
3684 /* The flow reflects exactly the contents of the packet. Sample
3685 * the packet using it. */
3686 dpif_ipfix_flow_sample(ofproto->ipfix, upcall->packet, &flow,
3687 cookie.flow_sample.collector_set_id,
3688 cookie.flow_sample.probability,
3689 cookie.flow_sample.obs_domain_id,
3690 cookie.flow_sample.obs_point_id);
3694 handle_ipfix_upcall(struct dpif_backer *backer,
3695 const struct dpif_upcall *upcall)
3697 struct ofproto_dpif *ofproto;
3700 if (ofproto_receive(backer, upcall->packet, upcall->key, upcall->key_len,
3701 &flow, NULL, &ofproto, NULL)
3702 || !ofproto->ipfix) {
3706 /* The flow reflects exactly the contents of the packet. Sample
3707 * the packet using it. */
3708 dpif_ipfix_bridge_sample(ofproto->ipfix, upcall->packet, &flow);
3712 handle_upcalls(struct dpif_backer *backer, unsigned int max_batch)
3714 struct dpif_upcall misses[FLOW_MISS_MAX_BATCH];
3715 struct ofpbuf miss_bufs[FLOW_MISS_MAX_BATCH];
3716 uint64_t miss_buf_stubs[FLOW_MISS_MAX_BATCH][4096 / 8];
3721 ovs_assert(max_batch <= FLOW_MISS_MAX_BATCH);
3724 for (n_processed = 0; n_processed < max_batch; n_processed++) {
3725 struct dpif_upcall *upcall = &misses[n_misses];
3726 struct ofpbuf *buf = &miss_bufs[n_misses];
3729 ofpbuf_use_stub(buf, miss_buf_stubs[n_misses],
3730 sizeof miss_buf_stubs[n_misses]);
3731 error = dpif_recv(backer->dpif, upcall, buf);
3737 switch (classify_upcall(upcall)) {
3739 /* Handle it later. */
3744 handle_sflow_upcall(backer, upcall);
3748 case FLOW_SAMPLE_UPCALL:
3749 handle_flow_sample_upcall(backer, upcall);
3754 handle_ipfix_upcall(backer, upcall);
3764 /* Handle deferred MISS_UPCALL processing. */
3765 handle_miss_upcalls(backer, misses, n_misses);
3766 for (i = 0; i < n_misses; i++) {
3767 ofpbuf_uninit(&miss_bufs[i]);
3773 /* Flow expiration. */
3775 static int subfacet_max_idle(const struct dpif_backer *);
3776 static void update_stats(struct dpif_backer *);
3777 static void rule_expire(struct rule_dpif *);
3778 static void expire_subfacets(struct dpif_backer *, int dp_max_idle);
3780 /* This function is called periodically by run(). Its job is to collect
3781 * updates for the flows that have been installed into the datapath, most
3782 * importantly when they last were used, and then use that information to
3783 * expire flows that have not been used recently.
3785 * Returns the number of milliseconds after which it should be called again. */
3787 expire(struct dpif_backer *backer)
3789 struct ofproto_dpif *ofproto;
3793 /* Periodically clear out the drop keys in an effort to keep them
3794 * relatively few. */
3795 drop_key_clear(backer);
3797 /* Update stats for each flow in the backer. */
3798 update_stats(backer);
3800 n_subfacets = hmap_count(&backer->subfacets);
3802 struct subfacet *subfacet;
3803 long long int total, now;
3807 HMAP_FOR_EACH (subfacet, hmap_node, &backer->subfacets) {
3808 total += now - subfacet->created;
3810 backer->avg_subfacet_life += total / n_subfacets;
3812 backer->avg_subfacet_life /= 2;
3814 backer->avg_n_subfacet += n_subfacets;
3815 backer->avg_n_subfacet /= 2;
3817 backer->max_n_subfacet = MAX(backer->max_n_subfacet, n_subfacets);
3819 max_idle = subfacet_max_idle(backer);
3820 expire_subfacets(backer, max_idle);
3822 HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
3823 struct rule *rule, *next_rule;
3825 if (ofproto->backer != backer) {
3829 /* Expire OpenFlow flows whose idle_timeout or hard_timeout
3831 LIST_FOR_EACH_SAFE (rule, next_rule, expirable,
3832 &ofproto->up.expirable) {
3833 rule_expire(rule_dpif_cast(rule));
3836 /* All outstanding data in existing flows has been accounted, so it's a
3837 * good time to do bond rebalancing. */
3838 if (ofproto->has_bonded_bundles) {
3839 struct ofbundle *bundle;
3841 HMAP_FOR_EACH (bundle, hmap_node, &ofproto->bundles) {
3843 bond_rebalance(bundle->bond, &backer->revalidate_set);
3849 return MIN(max_idle, 1000);
3852 /* Updates flow table statistics given that the datapath just reported 'stats'
3853 * as 'subfacet''s statistics. */
3855 update_subfacet_stats(struct subfacet *subfacet,
3856 const struct dpif_flow_stats *stats)
3858 struct facet *facet = subfacet->facet;
3859 struct ofproto_dpif *ofproto = ofproto_dpif_cast(facet->rule->up.ofproto);
3860 struct dpif_flow_stats diff;
3862 diff.tcp_flags = stats->tcp_flags;
3863 diff.used = stats->used;
3865 if (stats->n_packets >= subfacet->dp_packet_count) {
3866 diff.n_packets = stats->n_packets - subfacet->dp_packet_count;
3868 VLOG_WARN_RL(&rl, "unexpected packet count from the datapath");
3872 if (stats->n_bytes >= subfacet->dp_byte_count) {
3873 diff.n_bytes = stats->n_bytes - subfacet->dp_byte_count;
3875 VLOG_WARN_RL(&rl, "unexpected byte count from datapath");
3879 ofproto->n_hit += diff.n_packets;
3880 subfacet->dp_packet_count = stats->n_packets;
3881 subfacet->dp_byte_count = stats->n_bytes;
3882 subfacet_update_stats(subfacet, &diff);
3884 if (facet->accounted_bytes < facet->byte_count) {
3886 facet_account(facet);
3887 facet->accounted_bytes = facet->byte_count;
3891 /* 'key' with length 'key_len' bytes is a flow in 'dpif' that we know nothing
3892 * about, or a flow that shouldn't be installed but was anyway. Delete it. */
3894 delete_unexpected_flow(struct dpif_backer *backer,
3895 const struct nlattr *key, size_t key_len)
3897 if (!VLOG_DROP_WARN(&rl)) {
3901 odp_flow_key_format(key, key_len, &s);
3902 VLOG_WARN("unexpected flow: %s", ds_cstr(&s));
3906 COVERAGE_INC(facet_unexpected);
3907 dpif_flow_del(backer->dpif, key, key_len, NULL);
3910 /* Update 'packet_count', 'byte_count', and 'used' members of installed facets.
3912 * This function also pushes statistics updates to rules which each facet
3913 * resubmits into. Generally these statistics will be accurate. However, if a
3914 * facet changes the rule it resubmits into at some time in between
3915 * update_stats() runs, it is possible that statistics accrued to the
3916 * old rule will be incorrectly attributed to the new rule. This could be
3917 * avoided by calling update_stats() whenever rules are created or
3918 * deleted. However, the performance impact of making so many calls to the
3919 * datapath do not justify the benefit of having perfectly accurate statistics.
3921 * In addition, this function maintains per ofproto flow hit counts. The patch
3922 * port is not treated specially. e.g. A packet ingress from br0 patched into
3923 * br1 will increase the hit count of br0 by 1, however, does not affect
3924 * the hit or miss counts of br1.
3927 update_stats(struct dpif_backer *backer)
3929 const struct dpif_flow_stats *stats;
3930 struct dpif_flow_dump dump;
3931 const struct nlattr *key, *mask;
3932 size_t key_len, mask_len;
3934 dpif_flow_dump_start(&dump, backer->dpif);
3935 while (dpif_flow_dump_next(&dump, &key, &key_len,
3936 &mask, &mask_len, NULL, NULL, &stats)) {
3937 struct subfacet *subfacet;
3940 key_hash = odp_flow_key_hash(key, key_len);
3941 subfacet = subfacet_find(backer, key, key_len, key_hash);
3942 switch (subfacet ? subfacet->path : SF_NOT_INSTALLED) {
3944 update_subfacet_stats(subfacet, stats);
3948 /* Stats are updated per-packet. */
3951 case SF_NOT_INSTALLED:
3953 delete_unexpected_flow(backer, key, key_len);
3958 dpif_flow_dump_done(&dump);
3960 update_moving_averages(backer);
3963 /* Calculates and returns the number of milliseconds of idle time after which
3964 * subfacets should expire from the datapath. When a subfacet expires, we fold
3965 * its statistics into its facet, and when a facet's last subfacet expires, we
3966 * fold its statistic into its rule. */
3968 subfacet_max_idle(const struct dpif_backer *backer)
3971 * Idle time histogram.
3973 * Most of the time a switch has a relatively small number of subfacets.
3974 * When this is the case we might as well keep statistics for all of them
3975 * in userspace and to cache them in the kernel datapath for performance as
3978 * As the number of subfacets increases, the memory required to maintain
3979 * statistics about them in userspace and in the kernel becomes
3980 * significant. However, with a large number of subfacets it is likely
3981 * that only a few of them are "heavy hitters" that consume a large amount
3982 * of bandwidth. At this point, only heavy hitters are worth caching in
3983 * the kernel and maintaining in userspaces; other subfacets we can
3986 * The technique used to compute the idle time is to build a histogram with
3987 * N_BUCKETS buckets whose width is BUCKET_WIDTH msecs each. Each subfacet
3988 * that is installed in the kernel gets dropped in the appropriate bucket.
3989 * After the histogram has been built, we compute the cutoff so that only
3990 * the most-recently-used 1% of subfacets (but at least
3991 * flow_eviction_threshold flows) are kept cached. At least
3992 * the most-recently-used bucket of subfacets is kept, so actually an
3993 * arbitrary number of subfacets can be kept in any given expiration run
3994 * (though the next run will delete most of those unless they receive
3997 * This requires a second pass through the subfacets, in addition to the
3998 * pass made by update_stats(), because the former function never looks at
3999 * uninstallable subfacets.
4001 enum { BUCKET_WIDTH = ROUND_UP(100, TIME_UPDATE_INTERVAL) };
4002 enum { N_BUCKETS = 5000 / BUCKET_WIDTH };
4003 int buckets[N_BUCKETS] = { 0 };
4004 int total, subtotal, bucket;
4005 struct subfacet *subfacet;
4009 total = hmap_count(&backer->subfacets);
4010 if (total <= flow_eviction_threshold) {
4011 return N_BUCKETS * BUCKET_WIDTH;
4014 /* Build histogram. */
4016 HMAP_FOR_EACH (subfacet, hmap_node, &backer->subfacets) {
4017 long long int idle = now - subfacet->used;
4018 int bucket = (idle <= 0 ? 0
4019 : idle >= BUCKET_WIDTH * N_BUCKETS ? N_BUCKETS - 1
4020 : (unsigned int) idle / BUCKET_WIDTH);
4024 /* Find the first bucket whose flows should be expired. */
4025 subtotal = bucket = 0;
4027 subtotal += buckets[bucket++];
4028 } while (bucket < N_BUCKETS &&
4029 subtotal < MAX(flow_eviction_threshold, total / 100));
4031 if (VLOG_IS_DBG_ENABLED()) {
4035 ds_put_cstr(&s, "keep");
4036 for (i = 0; i < N_BUCKETS; i++) {
4038 ds_put_cstr(&s, ", drop");
4041 ds_put_format(&s, " %d:%d", i * BUCKET_WIDTH, buckets[i]);
4044 VLOG_INFO("%s (msec:count)", ds_cstr(&s));
4048 return bucket * BUCKET_WIDTH;
4052 expire_subfacets(struct dpif_backer *backer, int dp_max_idle)
4054 /* Cutoff time for most flows. */
4055 long long int normal_cutoff = time_msec() - dp_max_idle;
4057 /* We really want to keep flows for special protocols around, so use a more
4058 * conservative cutoff. */
4059 long long int special_cutoff = time_msec() - 10000;
4061 struct subfacet *subfacet, *next_subfacet;
4062 struct subfacet *batch[SUBFACET_DESTROY_MAX_BATCH];
4066 HMAP_FOR_EACH_SAFE (subfacet, next_subfacet, hmap_node,
4067 &backer->subfacets) {
4068 long long int cutoff;
4070 cutoff = (subfacet->facet->xout.slow & (SLOW_CFM | SLOW_BFD | SLOW_LACP
4074 if (subfacet->used < cutoff) {
4075 if (subfacet->path != SF_NOT_INSTALLED) {
4076 batch[n_batch++] = subfacet;
4077 if (n_batch >= SUBFACET_DESTROY_MAX_BATCH) {
4078 subfacet_destroy_batch(backer, batch, n_batch);
4082 subfacet_destroy(subfacet);
4088 subfacet_destroy_batch(backer, batch, n_batch);
4092 /* If 'rule' is an OpenFlow rule, that has expired according to OpenFlow rules,
4093 * then delete it entirely. */
4095 rule_expire(struct rule_dpif *rule)
4097 struct facet *facet, *next_facet;
4101 if (rule->up.pending) {
4102 /* We'll have to expire it later. */
4106 /* Has 'rule' expired? */
4108 if (rule->up.hard_timeout
4109 && now > rule->up.modified + rule->up.hard_timeout * 1000) {
4110 reason = OFPRR_HARD_TIMEOUT;
4111 } else if (rule->up.idle_timeout
4112 && now > rule->up.used + rule->up.idle_timeout * 1000) {
4113 reason = OFPRR_IDLE_TIMEOUT;
4118 COVERAGE_INC(ofproto_dpif_expired);
4120 /* Update stats. (This is a no-op if the rule expired due to an idle
4121 * timeout, because that only happens when the rule has no facets left.) */
4122 LIST_FOR_EACH_SAFE (facet, next_facet, list_node, &rule->facets) {
4123 facet_remove(facet);
4126 /* Get rid of the rule. */
4127 ofproto_rule_expire(&rule->up, reason);
4132 /* Creates and returns a new facet based on 'miss'.
4134 * The caller must already have determined that no facet with an identical
4135 * 'miss->flow' exists in 'miss->ofproto'.
4137 * 'rule' and 'xout' must have been created based on 'miss'.
4139 * 'facet'' statistics are initialized based on 'stats'.
4141 * The facet will initially have no subfacets. The caller should create (at
4142 * least) one subfacet with subfacet_create(). */
4143 static struct facet *
4144 facet_create(const struct flow_miss *miss, struct rule_dpif *rule,
4145 struct xlate_out *xout, struct dpif_flow_stats *stats)
4147 struct ofproto_dpif *ofproto = miss->ofproto;
4148 struct facet *facet;
4151 facet = xzalloc(sizeof *facet);
4152 facet->packet_count = facet->prev_packet_count = stats->n_packets;
4153 facet->byte_count = facet->prev_byte_count = stats->n_bytes;
4154 facet->tcp_flags = stats->tcp_flags;
4155 facet->used = stats->used;
4156 facet->flow = miss->flow;
4157 facet->learn_rl = time_msec() + 500;
4160 list_push_back(&facet->rule->facets, &facet->list_node);
4161 list_init(&facet->subfacets);
4162 netflow_flow_init(&facet->nf_flow);
4163 netflow_flow_update_time(ofproto->netflow, &facet->nf_flow, facet->used);
4165 xlate_out_copy(&facet->xout, xout);
4167 match_init(&match, &facet->flow, &facet->xout.wc);
4168 cls_rule_init(&facet->cr, &match, OFP_DEFAULT_PRIORITY);
4169 classifier_insert(&ofproto->facets, &facet->cr);
4171 facet->nf_flow.output_iface = facet->xout.nf_output_iface;
4177 facet_free(struct facet *facet)
4180 xlate_out_uninit(&facet->xout);
4185 /* Executes, within 'ofproto', the 'n_actions' actions in 'actions' on
4186 * 'packet', which arrived on 'in_port'. */
4188 execute_odp_actions(struct ofproto_dpif *ofproto, const struct flow *flow,
4189 const struct nlattr *odp_actions, size_t actions_len,
4190 struct ofpbuf *packet)
4192 struct odputil_keybuf keybuf;
4196 ofpbuf_use_stack(&key, &keybuf, sizeof keybuf);
4197 odp_flow_key_from_flow(&key, flow,
4198 ofp_port_to_odp_port(ofproto, flow->in_port.ofp_port));
4200 error = dpif_execute(ofproto->backer->dpif, key.data, key.size,
4201 odp_actions, actions_len, packet);
4205 /* Remove 'facet' from its ofproto and free up the associated memory:
4207 * - If 'facet' was installed in the datapath, uninstalls it and updates its
4208 * rule's statistics, via subfacet_uninstall().
4210 * - Removes 'facet' from its rule and from ofproto->facets.
4213 facet_remove(struct facet *facet)
4215 struct ofproto_dpif *ofproto = ofproto_dpif_cast(facet->rule->up.ofproto);
4216 struct subfacet *subfacet, *next_subfacet;
4218 ovs_assert(!list_is_empty(&facet->subfacets));
4220 /* First uninstall all of the subfacets to get final statistics. */
4221 LIST_FOR_EACH (subfacet, list_node, &facet->subfacets) {
4222 subfacet_uninstall(subfacet);
4225 /* Flush the final stats to the rule.
4227 * This might require us to have at least one subfacet around so that we
4228 * can use its actions for accounting in facet_account(), which is why we
4229 * have uninstalled but not yet destroyed the subfacets. */
4230 facet_flush_stats(facet);
4232 /* Now we're really all done so destroy everything. */
4233 LIST_FOR_EACH_SAFE (subfacet, next_subfacet, list_node,
4234 &facet->subfacets) {
4235 subfacet_destroy__(subfacet);
4237 classifier_remove(&ofproto->facets, &facet->cr);
4238 cls_rule_destroy(&facet->cr);
4239 list_remove(&facet->list_node);
4243 /* Feed information from 'facet' back into the learning table to keep it in
4244 * sync with what is actually flowing through the datapath. */
4246 facet_learn(struct facet *facet)
4248 long long int now = time_msec();
4250 if (!facet->xout.has_fin_timeout && now < facet->learn_rl) {
4254 facet->learn_rl = now + 500;
4256 if (!facet->xout.has_learn
4257 && !facet->xout.has_normal
4258 && (!facet->xout.has_fin_timeout
4259 || !(facet->tcp_flags & (TCP_FIN | TCP_RST)))) {
4263 facet_push_stats(facet, true);
4267 facet_account(struct facet *facet)
4269 struct ofproto_dpif *ofproto = ofproto_dpif_cast(facet->rule->up.ofproto);
4270 const struct nlattr *a;
4275 if (!facet->xout.has_normal || !ofproto->has_bonded_bundles) {
4278 n_bytes = facet->byte_count - facet->accounted_bytes;
4280 /* This loop feeds byte counters to bond_account() for rebalancing to use
4281 * as a basis. We also need to track the actual VLAN on which the packet
4282 * is going to be sent to ensure that it matches the one passed to
4283 * bond_choose_output_slave(). (Otherwise, we will account to the wrong
4286 * We use the actions from an arbitrary subfacet because they should all
4287 * be equally valid for our purpose. */
4288 vlan_tci = facet->flow.vlan_tci;
4289 NL_ATTR_FOR_EACH_UNSAFE (a, left, facet->xout.odp_actions.data,
4290 facet->xout.odp_actions.size) {
4291 const struct ovs_action_push_vlan *vlan;
4292 struct ofport_dpif *port;
4294 switch (nl_attr_type(a)) {
4295 case OVS_ACTION_ATTR_OUTPUT:
4296 port = get_odp_port(ofproto, nl_attr_get_odp_port(a));
4297 if (port && port->bundle && port->bundle->bond) {
4298 bond_account(port->bundle->bond, &facet->flow,
4299 vlan_tci_to_vid(vlan_tci), n_bytes);
4303 case OVS_ACTION_ATTR_POP_VLAN:
4304 vlan_tci = htons(0);
4307 case OVS_ACTION_ATTR_PUSH_VLAN:
4308 vlan = nl_attr_get(a);
4309 vlan_tci = vlan->vlan_tci;
4315 /* Returns true if the only action for 'facet' is to send to the controller.
4316 * (We don't report NetFlow expiration messages for such facets because they
4317 * are just part of the control logic for the network, not real traffic). */
4319 facet_is_controller_flow(struct facet *facet)
4322 const struct rule *rule = &facet->rule->up;
4323 const struct ofpact *ofpacts = rule->ofpacts;
4324 size_t ofpacts_len = rule->ofpacts_len;
4326 if (ofpacts_len > 0 &&
4327 ofpacts->type == OFPACT_CONTROLLER &&
4328 ofpact_next(ofpacts) >= ofpact_end(ofpacts, ofpacts_len)) {
4335 /* Folds all of 'facet''s statistics into its rule. Also updates the
4336 * accounting ofhook and emits a NetFlow expiration if appropriate. All of
4337 * 'facet''s statistics in the datapath should have been zeroed and folded into
4338 * its packet and byte counts before this function is called. */
4340 facet_flush_stats(struct facet *facet)
4342 struct ofproto_dpif *ofproto = ofproto_dpif_cast(facet->rule->up.ofproto);
4343 struct subfacet *subfacet;
4345 LIST_FOR_EACH (subfacet, list_node, &facet->subfacets) {
4346 ovs_assert(!subfacet->dp_byte_count);
4347 ovs_assert(!subfacet->dp_packet_count);
4350 facet_push_stats(facet, false);
4351 if (facet->accounted_bytes < facet->byte_count) {
4352 facet_account(facet);
4353 facet->accounted_bytes = facet->byte_count;
4356 if (ofproto->netflow && !facet_is_controller_flow(facet)) {
4357 struct ofexpired expired;
4358 expired.flow = facet->flow;
4359 expired.packet_count = facet->packet_count;
4360 expired.byte_count = facet->byte_count;
4361 expired.used = facet->used;
4362 netflow_expire(ofproto->netflow, &facet->nf_flow, &expired);
4365 /* Reset counters to prevent double counting if 'facet' ever gets
4367 facet_reset_counters(facet);
4369 netflow_flow_clear(&facet->nf_flow);
4370 facet->tcp_flags = 0;
4373 /* Searches 'ofproto''s table of facets for one which would be responsible for
4374 * 'flow'. Returns it if found, otherwise a null pointer.
4376 * The returned facet might need revalidation; use facet_lookup_valid()
4377 * instead if that is important. */
4378 static struct facet *
4379 facet_find(struct ofproto_dpif *ofproto, const struct flow *flow)
4381 struct cls_rule *cr = classifier_lookup(&ofproto->facets, flow, NULL);
4382 return cr ? CONTAINER_OF(cr, struct facet, cr) : NULL;
4385 /* Searches 'ofproto''s table of facets for one capable that covers
4386 * 'flow'. Returns it if found, otherwise a null pointer.
4388 * The returned facet is guaranteed to be valid. */
4389 static struct facet *
4390 facet_lookup_valid(struct ofproto_dpif *ofproto, const struct flow *flow)
4392 struct facet *facet;
4394 facet = facet_find(ofproto, flow);
4396 && (ofproto->backer->need_revalidate
4397 || tag_set_intersects(&ofproto->backer->revalidate_set,
4399 && !facet_revalidate(facet)) {
4407 facet_check_consistency(struct facet *facet)
4409 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 15);
4411 struct ofproto_dpif *ofproto = ofproto_dpif_cast(facet->rule->up.ofproto);
4413 struct xlate_out xout;
4414 struct xlate_in xin;
4416 struct rule_dpif *rule;
4419 /* Check the rule for consistency. */
4420 rule = rule_dpif_lookup(ofproto, &facet->flow, NULL);
4421 if (rule != facet->rule) {
4422 if (!VLOG_DROP_WARN(&rl)) {
4423 struct ds s = DS_EMPTY_INITIALIZER;
4425 flow_format(&s, &facet->flow);
4426 ds_put_format(&s, ": facet associated with wrong rule (was "
4427 "table=%"PRIu8",", facet->rule->up.table_id);
4428 cls_rule_format(&facet->rule->up.cr, &s);
4429 ds_put_format(&s, ") (should have been table=%"PRIu8",",
4431 cls_rule_format(&rule->up.cr, &s);
4432 ds_put_char(&s, ')');
4434 VLOG_WARN("%s", ds_cstr(&s));
4440 /* Check the datapath actions for consistency. */
4441 xlate_in_init(&xin, ofproto, &facet->flow, rule, 0, NULL);
4442 xlate_actions(&xin, &xout);
4444 ok = ofpbuf_equal(&facet->xout.odp_actions, &xout.odp_actions)
4445 && facet->xout.slow == xout.slow;
4446 if (!ok && !VLOG_DROP_WARN(&rl)) {
4447 struct ds s = DS_EMPTY_INITIALIZER;
4449 flow_format(&s, &facet->flow);
4450 ds_put_cstr(&s, ": inconsistency in facet");
4452 if (!ofpbuf_equal(&facet->xout.odp_actions, &xout.odp_actions)) {
4453 ds_put_cstr(&s, " (actions were: ");
4454 format_odp_actions(&s, facet->xout.odp_actions.data,
4455 facet->xout.odp_actions.size);
4456 ds_put_cstr(&s, ") (correct actions: ");
4457 format_odp_actions(&s, xout.odp_actions.data,
4458 xout.odp_actions.size);
4459 ds_put_char(&s, ')');
4462 if (facet->xout.slow != xout.slow) {
4463 ds_put_format(&s, " slow path incorrect. should be %d", xout.slow);
4466 VLOG_WARN("%s", ds_cstr(&s));
4469 xlate_out_uninit(&xout);
4474 /* Re-searches the classifier for 'facet':
4476 * - If the rule found is different from 'facet''s current rule, moves
4477 * 'facet' to the new rule and recompiles its actions.
4479 * - If the rule found is the same as 'facet''s current rule, leaves 'facet'
4480 * where it is and recompiles its actions anyway.
4482 * - If any of 'facet''s subfacets correspond to a new flow according to
4483 * ofproto_receive(), 'facet' is removed.
4485 * Returns true if 'facet' is still valid. False if 'facet' was removed. */
4487 facet_revalidate(struct facet *facet)
4489 struct ofproto_dpif *ofproto = ofproto_dpif_cast(facet->rule->up.ofproto);
4490 struct rule_dpif *new_rule;
4491 struct subfacet *subfacet;
4492 struct flow_wildcards wc;
4493 struct xlate_out xout;
4494 struct xlate_in xin;
4496 COVERAGE_INC(facet_revalidate);
4498 /* Check that child subfacets still correspond to this facet. Tunnel
4499 * configuration changes could cause a subfacet's OpenFlow in_port to
4501 LIST_FOR_EACH (subfacet, list_node, &facet->subfacets) {
4502 struct ofproto_dpif *recv_ofproto;
4503 struct flow recv_flow;
4506 error = ofproto_receive(ofproto->backer, NULL, subfacet->key,
4507 subfacet->key_len, &recv_flow, NULL,
4508 &recv_ofproto, NULL);
4510 || recv_ofproto != ofproto
4511 || facet != facet_find(ofproto, &recv_flow)) {
4512 facet_remove(facet);
4517 flow_wildcards_init_catchall(&wc);
4518 new_rule = rule_dpif_lookup(ofproto, &facet->flow, &wc);
4520 /* Calculate new datapath actions.
4522 * We do not modify any 'facet' state yet, because we might need to, e.g.,
4523 * emit a NetFlow expiration and, if so, we need to have the old state
4524 * around to properly compose it. */
4525 xlate_in_init(&xin, ofproto, &facet->flow, new_rule, 0, NULL);
4526 xlate_actions(&xin, &xout);
4527 flow_wildcards_or(&xout.wc, &xout.wc, &wc);
4529 /* A facet's slow path reason should only change under dramatic
4530 * circumstances. Rather than try to update everything, it's simpler to
4531 * remove the facet and start over.
4533 * More importantly, if a facet's wildcards change, it will be relatively
4534 * difficult to figure out if its subfacets still belong to it, and if not
4535 * which facet they may belong to. Again, to avoid the complexity, we
4536 * simply give up instead. */
4537 if (facet->xout.slow != xout.slow
4538 || memcmp(&facet->xout.wc, &xout.wc, sizeof xout.wc)) {
4539 facet_remove(facet);
4540 xlate_out_uninit(&xout);
4544 if (!ofpbuf_equal(&facet->xout.odp_actions, &xout.odp_actions)) {
4545 LIST_FOR_EACH(subfacet, list_node, &facet->subfacets) {
4546 if (subfacet->path == SF_FAST_PATH) {
4547 struct dpif_flow_stats stats;
4549 subfacet_install(subfacet, &xout.odp_actions, &stats);
4550 subfacet_update_stats(subfacet, &stats);
4554 facet_flush_stats(facet);
4556 ofpbuf_clear(&facet->xout.odp_actions);
4557 ofpbuf_put(&facet->xout.odp_actions, xout.odp_actions.data,
4558 xout.odp_actions.size);
4561 /* Update 'facet' now that we've taken care of all the old state. */
4562 facet->xout.tags = xout.tags;
4563 facet->xout.slow = xout.slow;
4564 facet->xout.has_learn = xout.has_learn;
4565 facet->xout.has_normal = xout.has_normal;
4566 facet->xout.has_fin_timeout = xout.has_fin_timeout;
4567 facet->xout.nf_output_iface = xout.nf_output_iface;
4568 facet->xout.mirrors = xout.mirrors;
4569 facet->nf_flow.output_iface = facet->xout.nf_output_iface;
4571 if (facet->rule != new_rule) {
4572 COVERAGE_INC(facet_changed_rule);
4573 list_remove(&facet->list_node);
4574 list_push_back(&new_rule->facets, &facet->list_node);
4575 facet->rule = new_rule;
4576 facet->used = new_rule->up.created;
4577 facet->prev_used = facet->used;
4580 xlate_out_uninit(&xout);
4585 facet_reset_counters(struct facet *facet)
4587 facet->packet_count = 0;
4588 facet->byte_count = 0;
4589 facet->prev_packet_count = 0;
4590 facet->prev_byte_count = 0;
4591 facet->accounted_bytes = 0;
4595 facet_push_stats(struct facet *facet, bool may_learn)
4597 struct dpif_flow_stats stats;
4599 ovs_assert(facet->packet_count >= facet->prev_packet_count);
4600 ovs_assert(facet->byte_count >= facet->prev_byte_count);
4601 ovs_assert(facet->used >= facet->prev_used);
4603 stats.n_packets = facet->packet_count - facet->prev_packet_count;
4604 stats.n_bytes = facet->byte_count - facet->prev_byte_count;
4605 stats.used = facet->used;
4606 stats.tcp_flags = facet->tcp_flags;
4608 if (may_learn || stats.n_packets || facet->used > facet->prev_used) {
4609 struct ofproto_dpif *ofproto =
4610 ofproto_dpif_cast(facet->rule->up.ofproto);
4612 struct ofport_dpif *in_port;
4613 struct xlate_in xin;
4615 facet->prev_packet_count = facet->packet_count;
4616 facet->prev_byte_count = facet->byte_count;
4617 facet->prev_used = facet->used;
4619 in_port = get_ofp_port(ofproto, facet->flow.in_port.ofp_port);
4620 if (in_port && in_port->is_tunnel) {
4621 netdev_vport_inc_rx(in_port->up.netdev, &stats);
4624 rule_credit_stats(facet->rule, &stats);
4625 netflow_flow_update_time(ofproto->netflow, &facet->nf_flow,
4627 netflow_flow_update_flags(&facet->nf_flow, facet->tcp_flags);
4628 mirror_update_stats(ofproto->mbridge, facet->xout.mirrors,
4629 stats.n_packets, stats.n_bytes);
4631 xlate_in_init(&xin, ofproto, &facet->flow, facet->rule,
4632 stats.tcp_flags, NULL);
4633 xin.resubmit_stats = &stats;
4634 xin.may_learn = may_learn;
4635 xlate_actions_for_side_effects(&xin);
4640 push_all_stats__(bool run_fast)
4642 static long long int rl = LLONG_MIN;
4643 struct ofproto_dpif *ofproto;
4645 if (time_msec() < rl) {
4649 HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
4650 struct cls_cursor cursor;
4651 struct facet *facet;
4653 cls_cursor_init(&cursor, &ofproto->facets, NULL);
4654 CLS_CURSOR_FOR_EACH (facet, cr, &cursor) {
4655 facet_push_stats(facet, false);
4662 rl = time_msec() + 100;
4666 push_all_stats(void)
4668 push_all_stats__(true);
4672 rule_credit_stats(struct rule_dpif *rule, const struct dpif_flow_stats *stats)
4674 rule->packet_count += stats->n_packets;
4675 rule->byte_count += stats->n_bytes;
4676 ofproto_rule_update_used(&rule->up, stats->used);
4681 static struct subfacet *
4682 subfacet_find(struct dpif_backer *backer, const struct nlattr *key,
4683 size_t key_len, uint32_t key_hash)
4685 struct subfacet *subfacet;
4687 HMAP_FOR_EACH_WITH_HASH (subfacet, hmap_node, key_hash,
4688 &backer->subfacets) {
4689 if (subfacet->key_len == key_len
4690 && !memcmp(key, subfacet->key, key_len)) {
4698 /* Searches 'facet' (within 'ofproto') for a subfacet with the specified
4699 * 'key_fitness', 'key', and 'key_len' members in 'miss'. Returns the
4700 * existing subfacet if there is one, otherwise creates and returns a
4702 static struct subfacet *
4703 subfacet_create(struct facet *facet, struct flow_miss *miss,
4706 struct dpif_backer *backer = miss->ofproto->backer;
4707 enum odp_key_fitness key_fitness = miss->key_fitness;
4708 const struct nlattr *key = miss->key;
4709 size_t key_len = miss->key_len;
4711 struct subfacet *subfacet;
4713 key_hash = odp_flow_key_hash(key, key_len);
4715 if (list_is_empty(&facet->subfacets)) {
4716 subfacet = &facet->one_subfacet;
4718 subfacet = subfacet_find(backer, key, key_len, key_hash);
4720 if (subfacet->facet == facet) {
4724 /* This shouldn't happen. */
4725 VLOG_ERR_RL(&rl, "subfacet with wrong facet");
4726 subfacet_destroy(subfacet);
4729 subfacet = xmalloc(sizeof *subfacet);
4732 hmap_insert(&backer->subfacets, &subfacet->hmap_node, key_hash);
4733 list_push_back(&facet->subfacets, &subfacet->list_node);
4734 subfacet->facet = facet;
4735 subfacet->key_fitness = key_fitness;
4736 subfacet->key = xmemdup(key, key_len);
4737 subfacet->key_len = key_len;
4738 subfacet->used = now;
4739 subfacet->created = now;
4740 subfacet->dp_packet_count = 0;
4741 subfacet->dp_byte_count = 0;
4742 subfacet->path = SF_NOT_INSTALLED;
4743 subfacet->backer = backer;
4745 backer->subfacet_add_count++;
4749 /* Uninstalls 'subfacet' from the datapath, if it is installed, removes it from
4750 * its facet within 'ofproto', and frees it. */
4752 subfacet_destroy__(struct subfacet *subfacet)
4754 struct facet *facet = subfacet->facet;
4755 struct ofproto_dpif *ofproto = ofproto_dpif_cast(facet->rule->up.ofproto);
4757 /* Update ofproto stats before uninstall the subfacet. */
4758 ofproto->backer->subfacet_del_count++;
4760 subfacet_uninstall(subfacet);
4761 hmap_remove(&subfacet->backer->subfacets, &subfacet->hmap_node);
4762 list_remove(&subfacet->list_node);
4763 free(subfacet->key);
4764 if (subfacet != &facet->one_subfacet) {
4769 /* Destroys 'subfacet', as with subfacet_destroy__(), and then if this was the
4770 * last remaining subfacet in its facet destroys the facet too. */
4772 subfacet_destroy(struct subfacet *subfacet)
4774 struct facet *facet = subfacet->facet;
4776 if (list_is_singleton(&facet->subfacets)) {
4777 /* facet_remove() needs at least one subfacet (it will remove it). */
4778 facet_remove(facet);
4780 subfacet_destroy__(subfacet);
4785 subfacet_destroy_batch(struct dpif_backer *backer,
4786 struct subfacet **subfacets, int n)
4788 struct dpif_op ops[SUBFACET_DESTROY_MAX_BATCH];
4789 struct dpif_op *opsp[SUBFACET_DESTROY_MAX_BATCH];
4790 struct dpif_flow_stats stats[SUBFACET_DESTROY_MAX_BATCH];
4793 for (i = 0; i < n; i++) {
4794 ops[i].type = DPIF_OP_FLOW_DEL;
4795 ops[i].u.flow_del.key = subfacets[i]->key;
4796 ops[i].u.flow_del.key_len = subfacets[i]->key_len;
4797 ops[i].u.flow_del.stats = &stats[i];
4801 dpif_operate(backer->dpif, opsp, n);
4802 for (i = 0; i < n; i++) {
4803 subfacet_reset_dp_stats(subfacets[i], &stats[i]);
4804 subfacets[i]->path = SF_NOT_INSTALLED;
4805 subfacet_destroy(subfacets[i]);
4810 /* Updates 'subfacet''s datapath flow, setting its actions to 'actions_len'
4811 * bytes of actions in 'actions'. If 'stats' is non-null, statistics counters
4812 * in the datapath will be zeroed and 'stats' will be updated with traffic new
4813 * since 'subfacet' was last updated.
4815 * Returns 0 if successful, otherwise a positive errno value. */
4817 subfacet_install(struct subfacet *subfacet, const struct ofpbuf *odp_actions,
4818 struct dpif_flow_stats *stats)
4820 struct facet *facet = subfacet->facet;
4821 struct ofproto_dpif *ofproto = ofproto_dpif_cast(facet->rule->up.ofproto);
4822 enum subfacet_path path = facet->xout.slow ? SF_SLOW_PATH : SF_FAST_PATH;
4823 const struct nlattr *actions = odp_actions->data;
4824 size_t actions_len = odp_actions->size;
4825 struct odputil_keybuf maskbuf;
4828 uint64_t slow_path_stub[128 / 8];
4829 enum dpif_flow_put_flags flags;
4832 flags = DPIF_FP_CREATE | DPIF_FP_MODIFY;
4834 flags |= DPIF_FP_ZERO_STATS;
4837 if (path == SF_SLOW_PATH) {
4838 compose_slow_path(ofproto, &facet->flow, facet->xout.slow,
4839 slow_path_stub, sizeof slow_path_stub,
4840 &actions, &actions_len);
4843 ofpbuf_use_stack(&mask, &maskbuf, sizeof maskbuf);
4844 if (enable_megaflows) {
4845 odp_flow_key_from_mask(&mask, &facet->xout.wc.masks,
4846 &facet->flow, UINT32_MAX);
4849 ret = dpif_flow_put(subfacet->backer->dpif, flags, subfacet->key,
4850 subfacet->key_len, mask.data, mask.size,
4851 actions, actions_len, stats);
4854 subfacet_reset_dp_stats(subfacet, stats);
4858 COVERAGE_INC(subfacet_install_fail);
4860 subfacet->path = path;
4865 /* If 'subfacet' is installed in the datapath, uninstalls it. */
4867 subfacet_uninstall(struct subfacet *subfacet)
4869 if (subfacet->path != SF_NOT_INSTALLED) {
4870 struct rule_dpif *rule = subfacet->facet->rule;
4871 struct ofproto_dpif *ofproto = ofproto_dpif_cast(rule->up.ofproto);
4872 struct dpif_flow_stats stats;
4875 error = dpif_flow_del(ofproto->backer->dpif, subfacet->key,
4876 subfacet->key_len, &stats);
4877 subfacet_reset_dp_stats(subfacet, &stats);
4879 subfacet_update_stats(subfacet, &stats);
4881 subfacet->path = SF_NOT_INSTALLED;
4883 ovs_assert(subfacet->dp_packet_count == 0);
4884 ovs_assert(subfacet->dp_byte_count == 0);
4888 /* Resets 'subfacet''s datapath statistics counters. This should be called
4889 * when 'subfacet''s statistics are cleared in the datapath. If 'stats' is
4890 * non-null, it should contain the statistics returned by dpif when 'subfacet'
4891 * was reset in the datapath. 'stats' will be modified to include only
4892 * statistics new since 'subfacet' was last updated. */
4894 subfacet_reset_dp_stats(struct subfacet *subfacet,
4895 struct dpif_flow_stats *stats)
4898 && subfacet->dp_packet_count <= stats->n_packets
4899 && subfacet->dp_byte_count <= stats->n_bytes) {
4900 stats->n_packets -= subfacet->dp_packet_count;
4901 stats->n_bytes -= subfacet->dp_byte_count;
4904 subfacet->dp_packet_count = 0;
4905 subfacet->dp_byte_count = 0;
4908 /* Folds the statistics from 'stats' into the counters in 'subfacet'.
4910 * Because of the meaning of a subfacet's counters, it only makes sense to do
4911 * this if 'stats' are not tracked in the datapath, that is, if 'stats'
4912 * represents a packet that was sent by hand or if it represents statistics
4913 * that have been cleared out of the datapath. */
4915 subfacet_update_stats(struct subfacet *subfacet,
4916 const struct dpif_flow_stats *stats)
4918 if (stats->n_packets || stats->used > subfacet->used) {
4919 struct facet *facet = subfacet->facet;
4921 subfacet->used = MAX(subfacet->used, stats->used);
4922 facet->used = MAX(facet->used, stats->used);
4923 facet->packet_count += stats->n_packets;
4924 facet->byte_count += stats->n_bytes;
4925 facet->tcp_flags |= stats->tcp_flags;
4931 /* Lookup 'flow' in 'ofproto''s classifier. If 'wc' is non-null, sets
4932 * the fields that were relevant as part of the lookup. */
4933 static struct rule_dpif *
4934 rule_dpif_lookup(struct ofproto_dpif *ofproto, const struct flow *flow,
4935 struct flow_wildcards *wc)
4937 struct rule_dpif *rule;
4939 rule = rule_dpif_lookup_in_table(ofproto, flow, wc, 0);
4944 return rule_dpif_miss_rule(ofproto, flow);
4948 rule_dpif_lookup_in_table(struct ofproto_dpif *ofproto,
4949 const struct flow *flow, struct flow_wildcards *wc,
4952 struct cls_rule *cls_rule;
4953 struct classifier *cls;
4956 if (table_id >= N_TABLES) {
4961 memset(&wc->masks.dl_type, 0xff, sizeof wc->masks.dl_type);
4962 wc->masks.nw_frag |= FLOW_NW_FRAG_MASK;
4965 cls = &ofproto->up.tables[table_id].cls;
4966 frag = (flow->nw_frag & FLOW_NW_FRAG_ANY) != 0;
4967 if (frag && ofproto->up.frag_handling == OFPC_FRAG_NORMAL) {
4968 /* We must pretend that transport ports are unavailable. */
4969 struct flow ofpc_normal_flow = *flow;
4970 ofpc_normal_flow.tp_src = htons(0);
4971 ofpc_normal_flow.tp_dst = htons(0);
4972 cls_rule = classifier_lookup(cls, &ofpc_normal_flow, wc);
4973 } else if (frag && ofproto->up.frag_handling == OFPC_FRAG_DROP) {
4974 cls_rule = &ofproto->drop_frags_rule->up.cr;
4976 flow_wildcards_init_exact(wc);
4979 cls_rule = classifier_lookup(cls, flow, wc);
4981 return rule_dpif_cast(rule_from_cls_rule(cls_rule));
4985 rule_dpif_miss_rule(struct ofproto_dpif *ofproto, const struct flow *flow)
4987 struct ofport_dpif *port;
4989 port = get_ofp_port(ofproto, flow->in_port.ofp_port);
4991 VLOG_WARN_RL(&rl, "packet-in on unknown OpenFlow port %"PRIu16,
4992 flow->in_port.ofp_port);
4993 return ofproto->miss_rule;
4996 if (port->up.pp.config & OFPUTIL_PC_NO_PACKET_IN) {
4997 return ofproto->no_packet_in_rule;
4999 return ofproto->miss_rule;
5003 complete_operation(struct rule_dpif *rule)
5005 struct ofproto_dpif *ofproto = ofproto_dpif_cast(rule->up.ofproto);
5007 rule_invalidate(rule);
5009 struct dpif_completion *c = xmalloc(sizeof *c);
5010 c->op = rule->up.pending;
5011 list_push_back(&ofproto->completions, &c->list_node);
5013 ofoperation_complete(rule->up.pending, 0);
5017 static struct rule *
5020 struct rule_dpif *rule = xmalloc(sizeof *rule);
5025 rule_dealloc(struct rule *rule_)
5027 struct rule_dpif *rule = rule_dpif_cast(rule_);
5032 rule_construct(struct rule *rule_)
5034 struct rule_dpif *rule = rule_dpif_cast(rule_);
5035 struct ofproto_dpif *ofproto = ofproto_dpif_cast(rule->up.ofproto);
5036 struct rule_dpif *victim;
5039 rule->packet_count = 0;
5040 rule->byte_count = 0;
5042 victim = rule_dpif_cast(ofoperation_get_victim(rule->up.pending));
5043 if (victim && !list_is_empty(&victim->facets)) {
5044 struct facet *facet;
5046 rule->facets = victim->facets;
5047 list_moved(&rule->facets);
5048 LIST_FOR_EACH (facet, list_node, &rule->facets) {
5049 /* XXX: We're only clearing our local counters here. It's possible
5050 * that quite a few packets are unaccounted for in the datapath
5051 * statistics. These will be accounted to the new rule instead of
5052 * cleared as required. This could be fixed by clearing out the
5053 * datapath statistics for this facet, but currently it doesn't
5055 facet_reset_counters(facet);
5059 /* Must avoid list_moved() in this case. */
5060 list_init(&rule->facets);
5063 table_id = rule->up.table_id;
5065 rule->tag = victim->tag;
5066 } else if (table_id == 0) {
5071 miniflow_expand(&rule->up.cr.match.flow, &flow);
5072 rule->tag = rule_calculate_tag(&flow, &rule->up.cr.match.mask,
5073 ofproto->tables[table_id].basis);
5076 complete_operation(rule);
5081 rule_destruct(struct rule *rule_)
5083 struct rule_dpif *rule = rule_dpif_cast(rule_);
5084 struct facet *facet, *next_facet;
5086 LIST_FOR_EACH_SAFE (facet, next_facet, list_node, &rule->facets) {
5087 facet_revalidate(facet);
5090 complete_operation(rule);
5094 rule_get_stats(struct rule *rule_, uint64_t *packets, uint64_t *bytes)
5096 struct rule_dpif *rule = rule_dpif_cast(rule_);
5098 /* push_all_stats() can handle flow misses which, when using the learn
5099 * action, can cause rules to be added and deleted. This can corrupt our
5100 * caller's datastructures which assume that rule_get_stats() doesn't have
5101 * an impact on the flow table. To be safe, we disable miss handling. */
5102 push_all_stats__(false);
5104 /* Start from historical data for 'rule' itself that are no longer tracked
5105 * in facets. This counts, for example, facets that have expired. */
5106 *packets = rule->packet_count;
5107 *bytes = rule->byte_count;
5111 rule_dpif_execute(struct rule_dpif *rule, const struct flow *flow,
5112 struct ofpbuf *packet)
5114 struct ofproto_dpif *ofproto = ofproto_dpif_cast(rule->up.ofproto);
5115 struct dpif_flow_stats stats;
5116 struct xlate_out xout;
5117 struct xlate_in xin;
5119 dpif_flow_stats_extract(flow, packet, time_msec(), &stats);
5120 rule_credit_stats(rule, &stats);
5122 xlate_in_init(&xin, ofproto, flow, rule, stats.tcp_flags, packet);
5123 xin.resubmit_stats = &stats;
5124 xlate_actions(&xin, &xout);
5126 execute_odp_actions(ofproto, flow, xout.odp_actions.data,
5127 xout.odp_actions.size, packet);
5129 xlate_out_uninit(&xout);
5133 rule_execute(struct rule *rule, const struct flow *flow,
5134 struct ofpbuf *packet)
5136 rule_dpif_execute(rule_dpif_cast(rule), flow, packet);
5137 ofpbuf_delete(packet);
5142 rule_modify_actions(struct rule *rule_)
5144 struct rule_dpif *rule = rule_dpif_cast(rule_);
5146 complete_operation(rule);
5149 /* Sends 'packet' out 'ofport'.
5150 * May modify 'packet'.
5151 * Returns 0 if successful, otherwise a positive errno value. */
5153 send_packet(const struct ofport_dpif *ofport, struct ofpbuf *packet)
5155 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport->up.ofproto);
5156 uint64_t odp_actions_stub[1024 / 8];
5157 struct ofpbuf key, odp_actions;
5158 struct dpif_flow_stats stats;
5159 struct odputil_keybuf keybuf;
5160 struct ofpact_output output;
5161 struct xlate_out xout;
5162 struct xlate_in xin;
5164 union flow_in_port in_port_;
5167 ofpbuf_use_stub(&odp_actions, odp_actions_stub, sizeof odp_actions_stub);
5168 ofpbuf_use_stack(&key, &keybuf, sizeof keybuf);
5170 /* Use OFPP_NONE as the in_port to avoid special packet processing. */
5171 in_port_.ofp_port = OFPP_NONE;
5172 flow_extract(packet, 0, 0, NULL, &in_port_, &flow);
5173 odp_flow_key_from_flow(&key, &flow, ofp_port_to_odp_port(ofproto,
5175 dpif_flow_stats_extract(&flow, packet, time_msec(), &stats);
5177 ofpact_init(&output.ofpact, OFPACT_OUTPUT, sizeof output);
5178 output.port = ofport->up.ofp_port;
5181 xlate_in_init(&xin, ofproto, &flow, NULL, 0, packet);
5182 xin.ofpacts_len = sizeof output;
5183 xin.ofpacts = &output.ofpact;
5184 xin.resubmit_stats = &stats;
5185 xlate_actions(&xin, &xout);
5187 error = dpif_execute(ofproto->backer->dpif,
5189 xout.odp_actions.data, xout.odp_actions.size,
5191 xlate_out_uninit(&xout);
5194 VLOG_WARN_RL(&rl, "%s: failed to send packet on port %s (%s)",
5195 ofproto->up.name, netdev_get_name(ofport->up.netdev),
5196 ovs_strerror(error));
5199 ofproto->stats.tx_packets++;
5200 ofproto->stats.tx_bytes += packet->size;
5204 /* Composes an ODP action for a "slow path" action for 'flow' within 'ofproto'.
5205 * The action will state 'slow' as the reason that the action is in the slow
5206 * path. (This is purely informational: it allows a human viewing "ovs-dpctl
5207 * dump-flows" output to see why a flow is in the slow path.)
5209 * The 'stub_size' bytes in 'stub' will be used to store the action.
5210 * 'stub_size' must be large enough for the action.
5212 * The action and its size will be stored in '*actionsp' and '*actions_lenp',
5215 compose_slow_path(const struct ofproto_dpif *ofproto, const struct flow *flow,
5216 enum slow_path_reason slow,
5217 uint64_t *stub, size_t stub_size,
5218 const struct nlattr **actionsp, size_t *actions_lenp)
5220 union user_action_cookie cookie;
5223 cookie.type = USER_ACTION_COOKIE_SLOW_PATH;
5224 cookie.slow_path.unused = 0;
5225 cookie.slow_path.reason = slow;
5227 ofpbuf_use_stack(&buf, stub, stub_size);
5228 if (slow & (SLOW_CFM | SLOW_BFD | SLOW_LACP | SLOW_STP)) {
5229 uint32_t pid = dpif_port_get_pid(ofproto->backer->dpif,
5231 odp_put_userspace_action(pid, &cookie, sizeof cookie.slow_path, &buf);
5233 put_userspace_action(ofproto, &buf, flow, &cookie,
5234 sizeof cookie.slow_path);
5236 *actionsp = buf.data;
5237 *actions_lenp = buf.size;
5241 put_userspace_action(const struct ofproto_dpif *ofproto,
5242 struct ofpbuf *odp_actions,
5243 const struct flow *flow,
5244 const union user_action_cookie *cookie,
5245 const size_t cookie_size)
5249 pid = dpif_port_get_pid(ofproto->backer->dpif,
5250 ofp_port_to_odp_port(ofproto,
5251 flow->in_port.ofp_port));
5253 return odp_put_userspace_action(pid, cookie, cookie_size, odp_actions);
5257 calculate_flow_tag(struct ofproto_dpif *ofproto, const struct flow *flow,
5258 uint8_t table_id, struct rule_dpif *rule)
5260 if (table_id > 0 && table_id < N_TABLES) {
5261 struct table_dpif *table = &ofproto->tables[table_id];
5262 if (table->other_table) {
5263 return (rule && rule->tag
5265 : rule_calculate_tag(flow, &table->other_table->mask,
5273 /* Optimized flow revalidation.
5275 * It's a difficult problem, in general, to tell which facets need to have
5276 * their actions recalculated whenever the OpenFlow flow table changes. We
5277 * don't try to solve that general problem: for most kinds of OpenFlow flow
5278 * table changes, we recalculate the actions for every facet. This is
5279 * relatively expensive, but it's good enough if the OpenFlow flow table
5280 * doesn't change very often.
5282 * However, we can expect one particular kind of OpenFlow flow table change to
5283 * happen frequently: changes caused by MAC learning. To avoid wasting a lot
5284 * of CPU on revalidating every facet whenever MAC learning modifies the flow
5285 * table, we add a special case that applies to flow tables in which every rule
5286 * has the same form (that is, the same wildcards), except that the table is
5287 * also allowed to have a single "catch-all" flow that matches all packets. We
5288 * optimize this case by tagging all of the facets that resubmit into the table
5289 * and invalidating the same tag whenever a flow changes in that table. The
5290 * end result is that we revalidate just the facets that need it (and sometimes
5291 * a few more, but not all of the facets or even all of the facets that
5292 * resubmit to the table modified by MAC learning). */
5294 /* Calculates the tag to use for 'flow' and mask 'mask' when it is inserted
5295 * into an OpenFlow table with the given 'basis'. */
5297 rule_calculate_tag(const struct flow *flow, const struct minimask *mask,
5300 if (minimask_is_catchall(mask)) {
5303 uint32_t hash = flow_hash_in_minimask(flow, mask, secret);
5304 return tag_create_deterministic(hash);
5308 /* Following a change to OpenFlow table 'table_id' in 'ofproto', update the
5309 * taggability of that table.
5311 * This function must be called after *each* change to a flow table. If you
5312 * skip calling it on some changes then the pointer comparisons at the end can
5313 * be invalid if you get unlucky. For example, if a flow removal causes a
5314 * cls_table to be destroyed and then a flow insertion causes a cls_table with
5315 * different wildcards to be created with the same address, then this function
5316 * will incorrectly skip revalidation. */
5318 table_update_taggable(struct ofproto_dpif *ofproto, uint8_t table_id)
5320 struct table_dpif *table = &ofproto->tables[table_id];
5321 const struct oftable *oftable = &ofproto->up.tables[table_id];
5322 struct cls_table *catchall, *other;
5323 struct cls_table *t;
5325 catchall = other = NULL;
5327 switch (hmap_count(&oftable->cls.tables)) {
5329 /* We could tag this OpenFlow table but it would make the logic a
5330 * little harder and it's a corner case that doesn't seem worth it
5336 HMAP_FOR_EACH (t, hmap_node, &oftable->cls.tables) {
5337 if (cls_table_is_catchall(t)) {
5339 } else if (!other) {
5342 /* Indicate that we can't tag this by setting both tables to
5343 * NULL. (We know that 'catchall' is already NULL.) */
5350 /* Can't tag this table. */
5354 if (table->catchall_table != catchall || table->other_table != other) {
5355 table->catchall_table = catchall;
5356 table->other_table = other;
5357 ofproto->backer->need_revalidate = REV_FLOW_TABLE;
5361 /* Given 'rule' that has changed in some way (either it is a rule being
5362 * inserted, a rule being deleted, or a rule whose actions are being
5363 * modified), marks facets for revalidation to ensure that packets will be
5364 * forwarded correctly according to the new state of the flow table.
5366 * This function must be called after *each* change to a flow table. See
5367 * the comment on table_update_taggable() for more information. */
5369 rule_invalidate(const struct rule_dpif *rule)
5371 struct ofproto_dpif *ofproto = ofproto_dpif_cast(rule->up.ofproto);
5373 table_update_taggable(ofproto, rule->up.table_id);
5375 if (!ofproto->backer->need_revalidate) {
5376 struct table_dpif *table = &ofproto->tables[rule->up.table_id];
5378 if (table->other_table && rule->tag) {
5379 tag_set_add(&ofproto->backer->revalidate_set, rule->tag);
5381 ofproto->backer->need_revalidate = REV_FLOW_TABLE;
5387 set_frag_handling(struct ofproto *ofproto_,
5388 enum ofp_config_flags frag_handling)
5390 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
5391 if (frag_handling != OFPC_FRAG_REASM) {
5392 ofproto->backer->need_revalidate = REV_RECONFIGURE;
5400 packet_out(struct ofproto *ofproto_, struct ofpbuf *packet,
5401 const struct flow *flow,
5402 const struct ofpact *ofpacts, size_t ofpacts_len)
5404 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
5405 struct odputil_keybuf keybuf;
5406 struct dpif_flow_stats stats;
5407 struct xlate_out xout;
5408 struct xlate_in xin;
5412 ofpbuf_use_stack(&key, &keybuf, sizeof keybuf);
5413 odp_flow_key_from_flow(&key, flow,
5414 ofp_port_to_odp_port(ofproto,
5415 flow->in_port.ofp_port));
5417 dpif_flow_stats_extract(flow, packet, time_msec(), &stats);
5419 xlate_in_init(&xin, ofproto, flow, NULL, stats.tcp_flags, packet);
5420 xin.resubmit_stats = &stats;
5421 xin.ofpacts_len = ofpacts_len;
5422 xin.ofpacts = ofpacts;
5424 xlate_actions(&xin, &xout);
5425 dpif_execute(ofproto->backer->dpif, key.data, key.size,
5426 xout.odp_actions.data, xout.odp_actions.size, packet);
5427 xlate_out_uninit(&xout);
5435 set_netflow(struct ofproto *ofproto_,
5436 const struct netflow_options *netflow_options)
5438 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
5440 if (netflow_options) {
5441 if (!ofproto->netflow) {
5442 ofproto->netflow = netflow_create();
5443 ofproto->backer->need_revalidate = REV_RECONFIGURE;
5445 return netflow_set_options(ofproto->netflow, netflow_options);
5446 } else if (ofproto->netflow) {
5447 ofproto->backer->need_revalidate = REV_RECONFIGURE;
5448 netflow_destroy(ofproto->netflow);
5449 ofproto->netflow = NULL;
5456 get_netflow_ids(const struct ofproto *ofproto_,
5457 uint8_t *engine_type, uint8_t *engine_id)
5459 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
5461 dpif_get_netflow_ids(ofproto->backer->dpif, engine_type, engine_id);
5465 send_active_timeout(struct ofproto_dpif *ofproto, struct facet *facet)
5467 if (!facet_is_controller_flow(facet) &&
5468 netflow_active_timeout_expired(ofproto->netflow, &facet->nf_flow)) {
5469 struct subfacet *subfacet;
5470 struct ofexpired expired;
5472 LIST_FOR_EACH (subfacet, list_node, &facet->subfacets) {
5473 if (subfacet->path == SF_FAST_PATH) {
5474 struct dpif_flow_stats stats;
5476 subfacet_install(subfacet, &facet->xout.odp_actions,
5478 subfacet_update_stats(subfacet, &stats);
5482 expired.flow = facet->flow;
5483 expired.packet_count = facet->packet_count;
5484 expired.byte_count = facet->byte_count;
5485 expired.used = facet->used;
5486 netflow_expire(ofproto->netflow, &facet->nf_flow, &expired);
5491 send_netflow_active_timeouts(struct ofproto_dpif *ofproto)
5493 struct cls_cursor cursor;
5494 struct facet *facet;
5496 cls_cursor_init(&cursor, &ofproto->facets, NULL);
5497 CLS_CURSOR_FOR_EACH (facet, cr, &cursor) {
5498 send_active_timeout(ofproto, facet);
5502 static struct ofproto_dpif *
5503 ofproto_dpif_lookup(const char *name)
5505 struct ofproto_dpif *ofproto;
5507 HMAP_FOR_EACH_WITH_HASH (ofproto, all_ofproto_dpifs_node,
5508 hash_string(name, 0), &all_ofproto_dpifs) {
5509 if (!strcmp(ofproto->up.name, name)) {
5517 ofproto_unixctl_fdb_flush(struct unixctl_conn *conn, int argc,
5518 const char *argv[], void *aux OVS_UNUSED)
5520 struct ofproto_dpif *ofproto;
5523 ofproto = ofproto_dpif_lookup(argv[1]);
5525 unixctl_command_reply_error(conn, "no such bridge");
5528 mac_learning_flush(ofproto->ml, &ofproto->backer->revalidate_set);
5530 HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
5531 mac_learning_flush(ofproto->ml, &ofproto->backer->revalidate_set);
5535 unixctl_command_reply(conn, "table successfully flushed");
5539 ofproto_unixctl_fdb_show(struct unixctl_conn *conn, int argc OVS_UNUSED,
5540 const char *argv[], void *aux OVS_UNUSED)
5542 struct ds ds = DS_EMPTY_INITIALIZER;
5543 const struct ofproto_dpif *ofproto;
5544 const struct mac_entry *e;
5546 ofproto = ofproto_dpif_lookup(argv[1]);
5548 unixctl_command_reply_error(conn, "no such bridge");
5552 ds_put_cstr(&ds, " port VLAN MAC Age\n");
5553 LIST_FOR_EACH (e, lru_node, &ofproto->ml->lrus) {
5554 struct ofbundle *bundle = e->port.p;
5555 char name[OFP_MAX_PORT_NAME_LEN];
5557 ofputil_port_to_string(ofbundle_get_a_port(bundle)->up.ofp_port,
5559 ds_put_format(&ds, "%5s %4d "ETH_ADDR_FMT" %3d\n",
5560 name, e->vlan, ETH_ADDR_ARGS(e->mac),
5561 mac_entry_age(ofproto->ml, e));
5563 unixctl_command_reply(conn, ds_cstr(&ds));
5568 struct xlate_out xout;
5569 struct xlate_in xin;
5575 trace_format_rule(struct ds *result, int level, const struct rule_dpif *rule)
5577 ds_put_char_multiple(result, '\t', level);
5579 ds_put_cstr(result, "No match\n");
5583 ds_put_format(result, "Rule: table=%"PRIu8" cookie=%#"PRIx64" ",
5584 rule ? rule->up.table_id : 0, ntohll(rule->up.flow_cookie));
5585 cls_rule_format(&rule->up.cr, result);
5586 ds_put_char(result, '\n');
5588 ds_put_char_multiple(result, '\t', level);
5589 ds_put_cstr(result, "OpenFlow ");
5590 ofpacts_format(rule->up.ofpacts, rule->up.ofpacts_len, result);
5591 ds_put_char(result, '\n');
5595 trace_format_flow(struct ds *result, int level, const char *title,
5596 struct trace_ctx *trace)
5598 ds_put_char_multiple(result, '\t', level);
5599 ds_put_format(result, "%s: ", title);
5600 if (flow_equal(&trace->xin.flow, &trace->flow)) {
5601 ds_put_cstr(result, "unchanged");
5603 flow_format(result, &trace->xin.flow);
5604 trace->flow = trace->xin.flow;
5606 ds_put_char(result, '\n');
5610 trace_format_regs(struct ds *result, int level, const char *title,
5611 struct trace_ctx *trace)
5615 ds_put_char_multiple(result, '\t', level);
5616 ds_put_format(result, "%s:", title);
5617 for (i = 0; i < FLOW_N_REGS; i++) {
5618 ds_put_format(result, " reg%zu=0x%"PRIx32, i, trace->flow.regs[i]);
5620 ds_put_char(result, '\n');
5624 trace_format_odp(struct ds *result, int level, const char *title,
5625 struct trace_ctx *trace)
5627 struct ofpbuf *odp_actions = &trace->xout.odp_actions;
5629 ds_put_char_multiple(result, '\t', level);
5630 ds_put_format(result, "%s: ", title);
5631 format_odp_actions(result, odp_actions->data, odp_actions->size);
5632 ds_put_char(result, '\n');
5636 trace_resubmit(struct xlate_in *xin, struct rule_dpif *rule, int recurse)
5638 struct trace_ctx *trace = CONTAINER_OF(xin, struct trace_ctx, xin);
5639 struct ds *result = trace->result;
5641 ds_put_char(result, '\n');
5642 trace_format_flow(result, recurse + 1, "Resubmitted flow", trace);
5643 trace_format_regs(result, recurse + 1, "Resubmitted regs", trace);
5644 trace_format_odp(result, recurse + 1, "Resubmitted odp", trace);
5645 trace_format_rule(result, recurse + 1, rule);
5649 trace_report(struct xlate_in *xin, const char *s, int recurse)
5651 struct trace_ctx *trace = CONTAINER_OF(xin, struct trace_ctx, xin);
5652 struct ds *result = trace->result;
5654 ds_put_char_multiple(result, '\t', recurse);
5655 ds_put_cstr(result, s);
5656 ds_put_char(result, '\n');
5660 ofproto_unixctl_trace(struct unixctl_conn *conn, int argc, const char *argv[],
5661 void *aux OVS_UNUSED)
5663 const struct dpif_backer *backer;
5664 struct ofproto_dpif *ofproto;
5665 struct ofpbuf odp_key, odp_mask;
5666 struct ofpbuf *packet;
5674 ofpbuf_init(&odp_key, 0);
5675 ofpbuf_init(&odp_mask, 0);
5677 /* Handle "-generate" or a hex string as the last argument. */
5678 if (!strcmp(argv[argc - 1], "-generate")) {
5679 packet = ofpbuf_new(0);
5682 const char *error = eth_from_hex(argv[argc - 1], &packet);
5685 } else if (argc == 4) {
5686 /* The 3-argument form must end in "-generate' or a hex string. */
5687 unixctl_command_reply_error(conn, error);
5692 /* Parse the flow and determine whether a datapath or
5693 * bridge is specified. If function odp_flow_key_from_string()
5694 * returns 0, the flow is a odp_flow. If function
5695 * parse_ofp_exact_flow() returns 0, the flow is a br_flow. */
5696 if (!odp_flow_from_string(argv[argc - 1], NULL, &odp_key, &odp_mask)) {
5697 /* If the odp_flow is the second argument,
5698 * the datapath name is the first argument. */
5700 const char *dp_type;
5701 if (!strncmp(argv[1], "ovs-", 4)) {
5702 dp_type = argv[1] + 4;
5706 backer = shash_find_data(&all_dpif_backers, dp_type);
5708 unixctl_command_reply_error(conn, "Cannot find datapath "
5713 /* No datapath name specified, so there should be only one
5715 struct shash_node *node;
5716 if (shash_count(&all_dpif_backers) != 1) {
5717 unixctl_command_reply_error(conn, "Must specify datapath "
5718 "name, there is more than one type of datapath");
5721 node = shash_first(&all_dpif_backers);
5722 backer = node->data;
5725 /* Extract the ofproto_dpif object from the ofproto_receive()
5727 if (ofproto_receive(backer, NULL, odp_key.data,
5728 odp_key.size, &flow, NULL, &ofproto, NULL)) {
5729 unixctl_command_reply_error(conn, "Invalid datapath flow");
5732 ds_put_format(&result, "Bridge: %s\n", ofproto->up.name);
5733 } else if (!parse_ofp_exact_flow(&flow, argv[argc - 1])) {
5735 unixctl_command_reply_error(conn, "Must specify bridge name");
5739 ofproto = ofproto_dpif_lookup(argv[1]);
5741 unixctl_command_reply_error(conn, "Unknown bridge name");
5745 unixctl_command_reply_error(conn, "Bad flow syntax");
5749 /* Generate a packet, if requested. */
5751 if (!packet->size) {
5752 flow_compose(packet, &flow);
5754 union flow_in_port in_port_;
5756 in_port_ = flow.in_port;
5757 ds_put_cstr(&result, "Packet: ");
5758 s = ofp_packet_to_string(packet->data, packet->size);
5759 ds_put_cstr(&result, s);
5762 /* Use the metadata from the flow and the packet argument
5763 * to reconstruct the flow. */
5764 flow_extract(packet, flow.skb_priority, flow.skb_mark, NULL,
5769 ofproto_trace(ofproto, &flow, packet, &result);
5770 unixctl_command_reply(conn, ds_cstr(&result));
5773 ds_destroy(&result);
5774 ofpbuf_delete(packet);
5775 ofpbuf_uninit(&odp_key);
5776 ofpbuf_uninit(&odp_mask);
5780 ofproto_trace(struct ofproto_dpif *ofproto, const struct flow *flow,
5781 const struct ofpbuf *packet, struct ds *ds)
5783 struct rule_dpif *rule;
5785 ds_put_cstr(ds, "Flow: ");
5786 flow_format(ds, flow);
5787 ds_put_char(ds, '\n');
5789 rule = rule_dpif_lookup(ofproto, flow, NULL);
5791 trace_format_rule(ds, 0, rule);
5792 if (rule == ofproto->miss_rule) {
5793 ds_put_cstr(ds, "\nNo match, flow generates \"packet in\"s.\n");
5794 } else if (rule == ofproto->no_packet_in_rule) {
5795 ds_put_cstr(ds, "\nNo match, packets dropped because "
5796 "OFPPC_NO_PACKET_IN is set on in_port.\n");
5797 } else if (rule == ofproto->drop_frags_rule) {
5798 ds_put_cstr(ds, "\nPackets dropped because they are IP fragments "
5799 "and the fragment handling mode is \"drop\".\n");
5803 uint64_t odp_actions_stub[1024 / 8];
5804 struct ofpbuf odp_actions;
5805 struct trace_ctx trace;
5809 tcp_flags = packet ? packet_get_tcp_flags(packet, flow) : 0;
5812 ofpbuf_use_stub(&odp_actions,
5813 odp_actions_stub, sizeof odp_actions_stub);
5814 xlate_in_init(&trace.xin, ofproto, flow, rule, tcp_flags, packet);
5815 trace.xin.resubmit_hook = trace_resubmit;
5816 trace.xin.report_hook = trace_report;
5818 xlate_actions(&trace.xin, &trace.xout);
5820 ds_put_char(ds, '\n');
5821 trace_format_flow(ds, 0, "Final flow", &trace);
5823 match_init(&match, flow, &trace.xout.wc);
5824 ds_put_cstr(ds, "Relevant fields: ");
5825 match_format(&match, ds, OFP_DEFAULT_PRIORITY);
5826 ds_put_char(ds, '\n');
5828 ds_put_cstr(ds, "Datapath actions: ");
5829 format_odp_actions(ds, trace.xout.odp_actions.data,
5830 trace.xout.odp_actions.size);
5832 if (trace.xout.slow) {
5833 ds_put_cstr(ds, "\nThis flow is handled by the userspace "
5834 "slow path because it:");
5835 switch (trace.xout.slow) {
5837 ds_put_cstr(ds, "\n\t- Consists of CFM packets.");
5840 ds_put_cstr(ds, "\n\t- Consists of LACP packets.");
5843 ds_put_cstr(ds, "\n\t- Consists of STP packets.");
5846 ds_put_cstr(ds, "\n\t- Consists of BFD packets.");
5848 case SLOW_CONTROLLER:
5849 ds_put_cstr(ds, "\n\t- Sends \"packet-in\" messages "
5850 "to the OpenFlow controller.");
5857 xlate_out_uninit(&trace.xout);
5862 ofproto_dpif_clog(struct unixctl_conn *conn OVS_UNUSED, int argc OVS_UNUSED,
5863 const char *argv[] OVS_UNUSED, void *aux OVS_UNUSED)
5866 unixctl_command_reply(conn, NULL);
5870 ofproto_dpif_unclog(struct unixctl_conn *conn OVS_UNUSED, int argc OVS_UNUSED,
5871 const char *argv[] OVS_UNUSED, void *aux OVS_UNUSED)
5874 unixctl_command_reply(conn, NULL);
5877 /* Runs a self-check of flow translations in 'ofproto'. Appends a message to
5878 * 'reply' describing the results. */
5880 ofproto_dpif_self_check__(struct ofproto_dpif *ofproto, struct ds *reply)
5882 struct cls_cursor cursor;
5883 struct facet *facet;
5887 cls_cursor_init(&cursor, &ofproto->facets, NULL);
5888 CLS_CURSOR_FOR_EACH (facet, cr, &cursor) {
5889 if (!facet_check_consistency(facet)) {
5894 ofproto->backer->need_revalidate = REV_INCONSISTENCY;
5898 ds_put_format(reply, "%s: self-check failed (%d errors)\n",
5899 ofproto->up.name, errors);
5901 ds_put_format(reply, "%s: self-check passed\n", ofproto->up.name);
5906 ofproto_dpif_self_check(struct unixctl_conn *conn,
5907 int argc, const char *argv[], void *aux OVS_UNUSED)
5909 struct ds reply = DS_EMPTY_INITIALIZER;
5910 struct ofproto_dpif *ofproto;
5913 ofproto = ofproto_dpif_lookup(argv[1]);
5915 unixctl_command_reply_error(conn, "Unknown ofproto (use "
5916 "ofproto/list for help)");
5919 ofproto_dpif_self_check__(ofproto, &reply);
5921 HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
5922 ofproto_dpif_self_check__(ofproto, &reply);
5926 unixctl_command_reply(conn, ds_cstr(&reply));
5930 /* Store the current ofprotos in 'ofproto_shash'. Returns a sorted list
5931 * of the 'ofproto_shash' nodes. It is the responsibility of the caller
5932 * to destroy 'ofproto_shash' and free the returned value. */
5933 static const struct shash_node **
5934 get_ofprotos(struct shash *ofproto_shash)
5936 const struct ofproto_dpif *ofproto;
5938 HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
5939 char *name = xasprintf("%s@%s", ofproto->up.type, ofproto->up.name);
5940 shash_add_nocopy(ofproto_shash, name, ofproto);
5943 return shash_sort(ofproto_shash);
5947 ofproto_unixctl_dpif_dump_dps(struct unixctl_conn *conn, int argc OVS_UNUSED,
5948 const char *argv[] OVS_UNUSED,
5949 void *aux OVS_UNUSED)
5951 struct ds ds = DS_EMPTY_INITIALIZER;
5952 struct shash ofproto_shash;
5953 const struct shash_node **sorted_ofprotos;
5956 shash_init(&ofproto_shash);
5957 sorted_ofprotos = get_ofprotos(&ofproto_shash);
5958 for (i = 0; i < shash_count(&ofproto_shash); i++) {
5959 const struct shash_node *node = sorted_ofprotos[i];
5960 ds_put_format(&ds, "%s\n", node->name);
5963 shash_destroy(&ofproto_shash);
5964 free(sorted_ofprotos);
5966 unixctl_command_reply(conn, ds_cstr(&ds));
5971 show_dp_rates(struct ds *ds, const char *heading,
5972 const struct avg_subfacet_rates *rates)
5974 ds_put_format(ds, "%s add rate: %5.3f/min, del rate: %5.3f/min\n",
5975 heading, rates->add_rate, rates->del_rate);
5979 dpif_show_backer(const struct dpif_backer *backer, struct ds *ds)
5981 const struct shash_node **ofprotos;
5982 struct ofproto_dpif *ofproto;
5983 struct shash ofproto_shash;
5984 uint64_t n_hit, n_missed;
5985 long long int minutes;
5988 n_hit = n_missed = 0;
5989 HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
5990 if (ofproto->backer == backer) {
5991 n_missed += ofproto->n_missed;
5992 n_hit += ofproto->n_hit;
5996 ds_put_format(ds, "%s: hit:%"PRIu64" missed:%"PRIu64"\n",
5997 dpif_name(backer->dpif), n_hit, n_missed);
5998 ds_put_format(ds, "\tflows: cur: %zu, avg: %u, max: %u,"
5999 " life span: %lldms\n", hmap_count(&backer->subfacets),
6000 backer->avg_n_subfacet, backer->max_n_subfacet,
6001 backer->avg_subfacet_life);
6003 minutes = (time_msec() - backer->created) / (1000 * 60);
6004 if (minutes >= 60) {
6005 show_dp_rates(ds, "\thourly avg:", &backer->hourly);
6007 if (minutes >= 60 * 24) {
6008 show_dp_rates(ds, "\tdaily avg:", &backer->daily);
6010 show_dp_rates(ds, "\toverall avg:", &backer->lifetime);
6012 shash_init(&ofproto_shash);
6013 ofprotos = get_ofprotos(&ofproto_shash);
6014 for (i = 0; i < shash_count(&ofproto_shash); i++) {
6015 struct ofproto_dpif *ofproto = ofprotos[i]->data;
6016 const struct shash_node **ports;
6019 if (ofproto->backer != backer) {
6023 ds_put_format(ds, "\t%s: hit:%"PRIu64" missed:%"PRIu64"\n",
6024 ofproto->up.name, ofproto->n_hit, ofproto->n_missed);
6026 ports = shash_sort(&ofproto->up.port_by_name);
6027 for (j = 0; j < shash_count(&ofproto->up.port_by_name); j++) {
6028 const struct shash_node *node = ports[j];
6029 struct ofport *ofport = node->data;
6031 odp_port_t odp_port;
6033 ds_put_format(ds, "\t\t%s %u/", netdev_get_name(ofport->netdev),
6036 odp_port = ofp_port_to_odp_port(ofproto, ofport->ofp_port);
6037 if (odp_port != ODPP_NONE) {
6038 ds_put_format(ds, "%"PRIu32":", odp_port);
6040 ds_put_cstr(ds, "none:");
6043 ds_put_format(ds, " (%s", netdev_get_type(ofport->netdev));
6046 if (!netdev_get_config(ofport->netdev, &config)) {
6047 const struct smap_node **nodes;
6050 nodes = smap_sort(&config);
6051 for (i = 0; i < smap_count(&config); i++) {
6052 const struct smap_node *node = nodes[i];
6053 ds_put_format(ds, "%c %s=%s", i ? ',' : ':',
6054 node->key, node->value);
6058 smap_destroy(&config);
6060 ds_put_char(ds, ')');
6061 ds_put_char(ds, '\n');
6065 shash_destroy(&ofproto_shash);
6070 ofproto_unixctl_dpif_show(struct unixctl_conn *conn, int argc OVS_UNUSED,
6071 const char *argv[] OVS_UNUSED, void *aux OVS_UNUSED)
6073 struct ds ds = DS_EMPTY_INITIALIZER;
6074 const struct shash_node **backers;
6077 backers = shash_sort(&all_dpif_backers);
6078 for (i = 0; i < shash_count(&all_dpif_backers); i++) {
6079 dpif_show_backer(backers[i]->data, &ds);
6083 unixctl_command_reply(conn, ds_cstr(&ds));
6087 /* Dump the megaflow (facet) cache. This is useful to check the
6088 * correctness of flow wildcarding, since the same mechanism is used for
6089 * both xlate caching and kernel wildcarding.
6091 * It's important to note that in the output the flow description uses
6092 * OpenFlow (OFP) ports, but the actions use datapath (ODP) ports.
6094 * This command is only needed for advanced debugging, so it's not
6095 * documented in the man page. */
6097 ofproto_unixctl_dpif_dump_megaflows(struct unixctl_conn *conn,
6098 int argc OVS_UNUSED, const char *argv[],
6099 void *aux OVS_UNUSED)
6101 struct ds ds = DS_EMPTY_INITIALIZER;
6102 const struct ofproto_dpif *ofproto;
6103 long long int now = time_msec();
6104 struct cls_cursor cursor;
6105 struct facet *facet;
6107 ofproto = ofproto_dpif_lookup(argv[1]);
6109 unixctl_command_reply_error(conn, "no such bridge");
6113 cls_cursor_init(&cursor, &ofproto->facets, NULL);
6114 CLS_CURSOR_FOR_EACH (facet, cr, &cursor) {
6115 cls_rule_format(&facet->cr, &ds);
6116 ds_put_cstr(&ds, ", ");
6117 ds_put_format(&ds, "n_subfacets:%zu, ", list_size(&facet->subfacets));
6118 ds_put_format(&ds, "used:%.3fs, ", (now - facet->used) / 1000.0);
6119 ds_put_cstr(&ds, "Datapath actions: ");
6120 if (facet->xout.slow) {
6121 uint64_t slow_path_stub[128 / 8];
6122 const struct nlattr *actions;
6125 compose_slow_path(ofproto, &facet->flow, facet->xout.slow,
6126 slow_path_stub, sizeof slow_path_stub,
6127 &actions, &actions_len);
6128 format_odp_actions(&ds, actions, actions_len);
6130 format_odp_actions(&ds, facet->xout.odp_actions.data,
6131 facet->xout.odp_actions.size);
6133 ds_put_cstr(&ds, "\n");
6136 ds_chomp(&ds, '\n');
6137 unixctl_command_reply(conn, ds_cstr(&ds));
6141 /* Disable using the megaflows.
6143 * This command is only needed for advanced debugging, so it's not
6144 * documented in the man page. */
6146 ofproto_unixctl_dpif_disable_megaflows(struct unixctl_conn *conn,
6147 int argc OVS_UNUSED,
6148 const char *argv[] OVS_UNUSED,
6149 void *aux OVS_UNUSED)
6151 struct ofproto_dpif *ofproto;
6153 enable_megaflows = false;
6155 HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
6156 flush(&ofproto->up);
6159 unixctl_command_reply(conn, "megaflows disabled");
6162 /* Re-enable using megaflows.
6164 * This command is only needed for advanced debugging, so it's not
6165 * documented in the man page. */
6167 ofproto_unixctl_dpif_enable_megaflows(struct unixctl_conn *conn,
6168 int argc OVS_UNUSED,
6169 const char *argv[] OVS_UNUSED,
6170 void *aux OVS_UNUSED)
6172 struct ofproto_dpif *ofproto;
6174 enable_megaflows = true;
6176 HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
6177 flush(&ofproto->up);
6180 unixctl_command_reply(conn, "megaflows enabled");
6184 ofproto_unixctl_dpif_dump_flows(struct unixctl_conn *conn,
6185 int argc OVS_UNUSED, const char *argv[],
6186 void *aux OVS_UNUSED)
6188 struct ds ds = DS_EMPTY_INITIALIZER;
6189 const struct ofproto_dpif *ofproto;
6190 struct subfacet *subfacet;
6192 ofproto = ofproto_dpif_lookup(argv[1]);
6194 unixctl_command_reply_error(conn, "no such bridge");
6198 update_stats(ofproto->backer);
6200 HMAP_FOR_EACH (subfacet, hmap_node, &ofproto->backer->subfacets) {
6201 struct facet *facet = subfacet->facet;
6203 if (ofproto_dpif_cast(facet->rule->up.ofproto) != ofproto) {
6207 odp_flow_key_format(subfacet->key, subfacet->key_len, &ds);
6209 ds_put_format(&ds, ", packets:%"PRIu64", bytes:%"PRIu64", used:",
6210 subfacet->dp_packet_count, subfacet->dp_byte_count);
6211 if (subfacet->used) {
6212 ds_put_format(&ds, "%.3fs",
6213 (time_msec() - subfacet->used) / 1000.0);
6215 ds_put_format(&ds, "never");
6217 if (subfacet->facet->tcp_flags) {
6218 ds_put_cstr(&ds, ", flags:");
6219 packet_format_tcp_flags(&ds, subfacet->facet->tcp_flags);
6222 ds_put_cstr(&ds, ", actions:");
6223 if (facet->xout.slow) {
6224 uint64_t slow_path_stub[128 / 8];
6225 const struct nlattr *actions;
6228 compose_slow_path(ofproto, &facet->flow, facet->xout.slow,
6229 slow_path_stub, sizeof slow_path_stub,
6230 &actions, &actions_len);
6231 format_odp_actions(&ds, actions, actions_len);
6233 format_odp_actions(&ds, facet->xout.odp_actions.data,
6234 facet->xout.odp_actions.size);
6236 ds_put_char(&ds, '\n');
6239 unixctl_command_reply(conn, ds_cstr(&ds));
6244 ofproto_unixctl_dpif_del_flows(struct unixctl_conn *conn,
6245 int argc OVS_UNUSED, const char *argv[],
6246 void *aux OVS_UNUSED)
6248 struct ds ds = DS_EMPTY_INITIALIZER;
6249 struct ofproto_dpif *ofproto;
6251 ofproto = ofproto_dpif_lookup(argv[1]);
6253 unixctl_command_reply_error(conn, "no such bridge");
6257 flush(&ofproto->up);
6259 unixctl_command_reply(conn, ds_cstr(&ds));
6264 ofproto_dpif_unixctl_init(void)
6266 static bool registered;
6272 unixctl_command_register(
6274 "[dp_name]|bridge odp_flow|br_flow [-generate|packet]",
6275 1, 3, ofproto_unixctl_trace, NULL);
6276 unixctl_command_register("fdb/flush", "[bridge]", 0, 1,
6277 ofproto_unixctl_fdb_flush, NULL);
6278 unixctl_command_register("fdb/show", "bridge", 1, 1,
6279 ofproto_unixctl_fdb_show, NULL);
6280 unixctl_command_register("ofproto/clog", "", 0, 0,
6281 ofproto_dpif_clog, NULL);
6282 unixctl_command_register("ofproto/unclog", "", 0, 0,
6283 ofproto_dpif_unclog, NULL);
6284 unixctl_command_register("ofproto/self-check", "[bridge]", 0, 1,
6285 ofproto_dpif_self_check, NULL);
6286 unixctl_command_register("dpif/dump-dps", "", 0, 0,
6287 ofproto_unixctl_dpif_dump_dps, NULL);
6288 unixctl_command_register("dpif/show", "", 0, 0, ofproto_unixctl_dpif_show,
6290 unixctl_command_register("dpif/dump-flows", "bridge", 1, 1,
6291 ofproto_unixctl_dpif_dump_flows, NULL);
6292 unixctl_command_register("dpif/del-flows", "bridge", 1, 1,
6293 ofproto_unixctl_dpif_del_flows, NULL);
6294 unixctl_command_register("dpif/dump-megaflows", "bridge", 1, 1,
6295 ofproto_unixctl_dpif_dump_megaflows, NULL);
6296 unixctl_command_register("dpif/disable-megaflows", "", 0, 0,
6297 ofproto_unixctl_dpif_disable_megaflows, NULL);
6298 unixctl_command_register("dpif/enable-megaflows", "", 0, 0,
6299 ofproto_unixctl_dpif_enable_megaflows, NULL);
6302 /* Linux VLAN device support (e.g. "eth0.10" for VLAN 10.)
6304 * This is deprecated. It is only for compatibility with broken device drivers
6305 * in old versions of Linux that do not properly support VLANs when VLAN
6306 * devices are not used. When broken device drivers are no longer in
6307 * widespread use, we will delete these interfaces. */
6310 set_realdev(struct ofport *ofport_, ofp_port_t realdev_ofp_port, int vid)
6312 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport_->ofproto);
6313 struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
6315 if (realdev_ofp_port == ofport->realdev_ofp_port
6316 && vid == ofport->vlandev_vid) {
6320 ofproto->backer->need_revalidate = REV_RECONFIGURE;
6322 if (ofport->realdev_ofp_port) {
6325 if (realdev_ofp_port && ofport->bundle) {
6326 /* vlandevs are enslaved to their realdevs, so they are not allowed to
6327 * themselves be part of a bundle. */
6328 bundle_set(ofport->up.ofproto, ofport->bundle, NULL);
6331 ofport->realdev_ofp_port = realdev_ofp_port;
6332 ofport->vlandev_vid = vid;
6334 if (realdev_ofp_port) {
6335 vsp_add(ofport, realdev_ofp_port, vid);
6342 hash_realdev_vid(ofp_port_t realdev_ofp_port, int vid)
6344 return hash_2words(ofp_to_u16(realdev_ofp_port), vid);
6347 /* Returns the OFP port number of the Linux VLAN device that corresponds to
6348 * 'vlan_tci' on the network device with port number 'realdev_ofp_port' in
6349 * 'struct ofport_dpif'. For example, given 'realdev_ofp_port' of eth0 and
6350 * 'vlan_tci' 9, it would return the port number of eth0.9.
6352 * Unless VLAN splinters are enabled for port 'realdev_ofp_port', this
6353 * function just returns its 'realdev_ofp_port' argument. */
6355 vsp_realdev_to_vlandev(const struct ofproto_dpif *ofproto,
6356 ofp_port_t realdev_ofp_port, ovs_be16 vlan_tci)
6358 if (!hmap_is_empty(&ofproto->realdev_vid_map)) {
6359 int vid = vlan_tci_to_vid(vlan_tci);
6360 const struct vlan_splinter *vsp;
6362 HMAP_FOR_EACH_WITH_HASH (vsp, realdev_vid_node,
6363 hash_realdev_vid(realdev_ofp_port, vid),
6364 &ofproto->realdev_vid_map) {
6365 if (vsp->realdev_ofp_port == realdev_ofp_port
6366 && vsp->vid == vid) {
6367 return vsp->vlandev_ofp_port;
6371 return realdev_ofp_port;
6374 static struct vlan_splinter *
6375 vlandev_find(const struct ofproto_dpif *ofproto, ofp_port_t vlandev_ofp_port)
6377 struct vlan_splinter *vsp;
6379 HMAP_FOR_EACH_WITH_HASH (vsp, vlandev_node,
6380 hash_ofp_port(vlandev_ofp_port),
6381 &ofproto->vlandev_map) {
6382 if (vsp->vlandev_ofp_port == vlandev_ofp_port) {
6390 /* Returns the OpenFlow port number of the "real" device underlying the Linux
6391 * VLAN device with OpenFlow port number 'vlandev_ofp_port' and stores the
6392 * VLAN VID of the Linux VLAN device in '*vid'. For example, given
6393 * 'vlandev_ofp_port' of eth0.9, it would return the OpenFlow port number of
6394 * eth0 and store 9 in '*vid'.
6396 * Returns 0 and does not modify '*vid' if 'vlandev_ofp_port' is not a Linux
6397 * VLAN device. Unless VLAN splinters are enabled, this is what this function
6400 vsp_vlandev_to_realdev(const struct ofproto_dpif *ofproto,
6401 ofp_port_t vlandev_ofp_port, int *vid)
6403 if (!hmap_is_empty(&ofproto->vlandev_map)) {
6404 const struct vlan_splinter *vsp;
6406 vsp = vlandev_find(ofproto, vlandev_ofp_port);
6411 return vsp->realdev_ofp_port;
6417 /* Given 'flow', a flow representing a packet received on 'ofproto', checks
6418 * whether 'flow->in_port' represents a Linux VLAN device. If so, changes
6419 * 'flow->in_port' to the "real" device backing the VLAN device, sets
6420 * 'flow->vlan_tci' to the VLAN VID, and returns true. Otherwise (which is
6421 * always the case unless VLAN splinters are enabled), returns false without
6422 * making any changes. */
6424 vsp_adjust_flow(const struct ofproto_dpif *ofproto, struct flow *flow)
6429 realdev = vsp_vlandev_to_realdev(ofproto, flow->in_port.ofp_port, &vid);
6434 /* Cause the flow to be processed as if it came in on the real device with
6435 * the VLAN device's VLAN ID. */
6436 flow->in_port.ofp_port = realdev;
6437 flow->vlan_tci = htons((vid & VLAN_VID_MASK) | VLAN_CFI);
6442 vsp_remove(struct ofport_dpif *port)
6444 struct ofproto_dpif *ofproto = ofproto_dpif_cast(port->up.ofproto);
6445 struct vlan_splinter *vsp;
6447 vsp = vlandev_find(ofproto, port->up.ofp_port);
6449 hmap_remove(&ofproto->vlandev_map, &vsp->vlandev_node);
6450 hmap_remove(&ofproto->realdev_vid_map, &vsp->realdev_vid_node);
6453 port->realdev_ofp_port = 0;
6455 VLOG_ERR("missing vlan device record");
6460 vsp_add(struct ofport_dpif *port, ofp_port_t realdev_ofp_port, int vid)
6462 struct ofproto_dpif *ofproto = ofproto_dpif_cast(port->up.ofproto);
6464 if (!vsp_vlandev_to_realdev(ofproto, port->up.ofp_port, NULL)
6465 && (vsp_realdev_to_vlandev(ofproto, realdev_ofp_port, htons(vid))
6466 == realdev_ofp_port)) {
6467 struct vlan_splinter *vsp;
6469 vsp = xmalloc(sizeof *vsp);
6470 hmap_insert(&ofproto->vlandev_map, &vsp->vlandev_node,
6471 hash_ofp_port(port->up.ofp_port));
6472 hmap_insert(&ofproto->realdev_vid_map, &vsp->realdev_vid_node,
6473 hash_realdev_vid(realdev_ofp_port, vid));
6474 vsp->realdev_ofp_port = realdev_ofp_port;
6475 vsp->vlandev_ofp_port = port->up.ofp_port;
6478 port->realdev_ofp_port = realdev_ofp_port;
6480 VLOG_ERR("duplicate vlan device record");
6485 ofp_port_to_odp_port(const struct ofproto_dpif *ofproto, ofp_port_t ofp_port)
6487 const struct ofport_dpif *ofport = get_ofp_port(ofproto, ofp_port);
6488 return ofport ? ofport->odp_port : ODPP_NONE;
6491 static struct ofport_dpif *
6492 odp_port_to_ofport(const struct dpif_backer *backer, odp_port_t odp_port)
6494 struct ofport_dpif *port;
6496 HMAP_FOR_EACH_IN_BUCKET (port, odp_port_node, hash_odp_port(odp_port),
6497 &backer->odp_to_ofport_map) {
6498 if (port->odp_port == odp_port) {
6507 odp_port_to_ofp_port(const struct ofproto_dpif *ofproto, odp_port_t odp_port)
6509 struct ofport_dpif *port;
6511 port = odp_port_to_ofport(ofproto->backer, odp_port);
6512 if (port && &ofproto->up == port->up.ofproto) {
6513 return port->up.ofp_port;
6519 /* Compute exponentially weighted moving average, adding 'new' as the newest,
6520 * most heavily weighted element. 'base' designates the rate of decay: after
6521 * 'base' further updates, 'new''s weight in the EWMA decays to about 1/e
6524 exp_mavg(double *avg, int base, double new)
6526 *avg = (*avg * (base - 1) + new) / base;
6530 update_moving_averages(struct dpif_backer *backer)
6532 const int min_ms = 60 * 1000; /* milliseconds in one minute. */
6533 long long int minutes = (time_msec() - backer->created) / min_ms;
6536 backer->lifetime.add_rate = (double) backer->total_subfacet_add_count
6538 backer->lifetime.del_rate = (double) backer->total_subfacet_del_count
6541 backer->lifetime.add_rate = 0.0;
6542 backer->lifetime.del_rate = 0.0;
6545 /* Update hourly averages on the minute boundaries. */
6546 if (time_msec() - backer->last_minute >= min_ms) {
6547 exp_mavg(&backer->hourly.add_rate, 60, backer->subfacet_add_count);
6548 exp_mavg(&backer->hourly.del_rate, 60, backer->subfacet_del_count);
6550 /* Update daily averages on the hour boundaries. */
6551 if ((backer->last_minute - backer->created) / min_ms % 60 == 59) {
6552 exp_mavg(&backer->daily.add_rate, 24, backer->hourly.add_rate);
6553 exp_mavg(&backer->daily.del_rate, 24, backer->hourly.del_rate);
6556 backer->total_subfacet_add_count += backer->subfacet_add_count;
6557 backer->total_subfacet_del_count += backer->subfacet_del_count;
6558 backer->subfacet_add_count = 0;
6559 backer->subfacet_del_count = 0;
6560 backer->last_minute += min_ms;
6564 const struct ofproto_class ofproto_dpif_class = {
6599 port_is_lacp_current,
6600 NULL, /* rule_choose_table */
6607 rule_modify_actions,
6621 get_stp_port_status,
6628 is_mirror_output_bundle,
6629 forward_bpdu_changed,
6630 set_mac_table_config,
6632 NULL, /* meter_get_features */
6633 NULL, /* meter_set */
6634 NULL, /* meter_get */
6635 NULL, /* meter_del */