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.table_id = TBL_INTERNAL;
1097 fm.command = OFPFC_ADD;
1098 fm.idle_timeout = 0;
1099 fm.hard_timeout = 0;
1103 fm.ofpacts = ofpacts->data;
1104 fm.ofpacts_len = ofpacts->size;
1106 error = ofproto_flow_mod(&ofproto->up, &fm);
1108 VLOG_ERR_RL(&rl, "failed to add internal flow %d (%s)",
1109 id, ofperr_to_string(error));
1113 *rulep = rule_dpif_lookup_in_table(ofproto, &fm.match.flow, NULL,
1115 ovs_assert(*rulep != NULL);
1121 add_internal_flows(struct ofproto_dpif *ofproto)
1123 struct ofpact_controller *controller;
1124 uint64_t ofpacts_stub[128 / 8];
1125 struct ofpbuf ofpacts;
1129 ofpbuf_use_stack(&ofpacts, ofpacts_stub, sizeof ofpacts_stub);
1132 controller = ofpact_put_CONTROLLER(&ofpacts);
1133 controller->max_len = UINT16_MAX;
1134 controller->controller_id = 0;
1135 controller->reason = OFPR_NO_MATCH;
1136 ofpact_pad(&ofpacts);
1138 error = add_internal_flow(ofproto, id++, &ofpacts, &ofproto->miss_rule);
1143 ofpbuf_clear(&ofpacts);
1144 error = add_internal_flow(ofproto, id++, &ofpacts,
1145 &ofproto->no_packet_in_rule);
1150 error = add_internal_flow(ofproto, id++, &ofpacts,
1151 &ofproto->drop_frags_rule);
1156 complete_operations(struct ofproto_dpif *ofproto)
1158 struct dpif_completion *c, *next;
1160 LIST_FOR_EACH_SAFE (c, next, list_node, &ofproto->completions) {
1161 ofoperation_complete(c->op, 0);
1162 list_remove(&c->list_node);
1168 destruct(struct ofproto *ofproto_)
1170 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1171 struct rule_dpif *rule, *next_rule;
1172 struct oftable *table;
1174 ofproto->backer->need_revalidate = REV_RECONFIGURE;
1175 hmap_remove(&all_ofproto_dpifs, &ofproto->all_ofproto_dpifs_node);
1176 complete_operations(ofproto);
1178 OFPROTO_FOR_EACH_TABLE (table, &ofproto->up) {
1179 struct cls_cursor cursor;
1181 cls_cursor_init(&cursor, &table->cls, NULL);
1182 CLS_CURSOR_FOR_EACH_SAFE (rule, next_rule, up.cr, &cursor) {
1183 ofproto_rule_destroy(&rule->up);
1187 mbridge_unref(ofproto->mbridge);
1189 netflow_destroy(ofproto->netflow);
1190 dpif_sflow_unref(ofproto->sflow);
1191 hmap_destroy(&ofproto->bundles);
1192 mac_learning_unref(ofproto->ml);
1194 classifier_destroy(&ofproto->facets);
1196 hmap_destroy(&ofproto->vlandev_map);
1197 hmap_destroy(&ofproto->realdev_vid_map);
1199 sset_destroy(&ofproto->ports);
1200 sset_destroy(&ofproto->ghost_ports);
1201 sset_destroy(&ofproto->port_poll_set);
1203 close_dpif_backer(ofproto->backer);
1207 run_fast(struct ofproto *ofproto_)
1209 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1210 struct ofport_dpif *ofport;
1212 /* Do not perform any periodic activity required by 'ofproto' while
1213 * waiting for flow restore to complete. */
1214 if (ofproto_get_flow_restore_wait()) {
1218 HMAP_FOR_EACH (ofport, up.hmap_node, &ofproto->up.ports) {
1219 port_run_fast(ofport);
1226 run(struct ofproto *ofproto_)
1228 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1229 struct ofport_dpif *ofport;
1230 struct ofbundle *bundle;
1234 complete_operations(ofproto);
1237 if (mbridge_need_revalidate(ofproto->mbridge)) {
1238 ofproto->backer->need_revalidate = REV_RECONFIGURE;
1239 mac_learning_flush(ofproto->ml, NULL);
1242 /* Do not perform any periodic activity below required by 'ofproto' while
1243 * waiting for flow restore to complete. */
1244 if (ofproto_get_flow_restore_wait()) {
1248 error = run_fast(ofproto_);
1253 if (ofproto->netflow) {
1254 if (netflow_run(ofproto->netflow)) {
1255 send_netflow_active_timeouts(ofproto);
1258 if (ofproto->sflow) {
1259 dpif_sflow_run(ofproto->sflow);
1262 HMAP_FOR_EACH (ofport, up.hmap_node, &ofproto->up.ports) {
1265 HMAP_FOR_EACH (bundle, hmap_node, &ofproto->bundles) {
1270 mac_learning_run(ofproto->ml, &ofproto->backer->revalidate_set);
1272 /* Check the consistency of a random facet, to aid debugging. */
1273 if (time_msec() >= ofproto->consistency_rl
1274 && !classifier_is_empty(&ofproto->facets)
1275 && !ofproto->backer->need_revalidate) {
1276 struct cls_table *table;
1277 struct cls_rule *cr;
1278 struct facet *facet;
1280 ofproto->consistency_rl = time_msec() + 250;
1282 table = CONTAINER_OF(hmap_random_node(&ofproto->facets.tables),
1283 struct cls_table, hmap_node);
1284 cr = CONTAINER_OF(hmap_random_node(&table->rules), struct cls_rule,
1286 facet = CONTAINER_OF(cr, struct facet, cr);
1288 if (!tag_set_intersects(&ofproto->backer->revalidate_set,
1289 facet->xout.tags)) {
1290 if (!facet_check_consistency(facet)) {
1291 ofproto->backer->need_revalidate = REV_INCONSISTENCY;
1300 wait(struct ofproto *ofproto_)
1302 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1303 struct ofport_dpif *ofport;
1304 struct ofbundle *bundle;
1306 if (!clogged && !list_is_empty(&ofproto->completions)) {
1307 poll_immediate_wake();
1310 if (ofproto_get_flow_restore_wait()) {
1314 dpif_wait(ofproto->backer->dpif);
1315 dpif_recv_wait(ofproto->backer->dpif);
1316 if (ofproto->sflow) {
1317 dpif_sflow_wait(ofproto->sflow);
1319 if (!tag_set_is_empty(&ofproto->backer->revalidate_set)) {
1320 poll_immediate_wake();
1322 HMAP_FOR_EACH (ofport, up.hmap_node, &ofproto->up.ports) {
1325 HMAP_FOR_EACH (bundle, hmap_node, &ofproto->bundles) {
1326 bundle_wait(bundle);
1328 if (ofproto->netflow) {
1329 netflow_wait(ofproto->netflow);
1331 mac_learning_wait(ofproto->ml);
1333 if (ofproto->backer->need_revalidate) {
1334 /* Shouldn't happen, but if it does just go around again. */
1335 VLOG_DBG_RL(&rl, "need revalidate in ofproto_wait_cb()");
1336 poll_immediate_wake();
1341 get_memory_usage(const struct ofproto *ofproto_, struct simap *usage)
1343 const struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1344 struct cls_cursor cursor;
1345 size_t n_subfacets = 0;
1346 struct facet *facet;
1348 simap_increase(usage, "facets", classifier_count(&ofproto->facets));
1350 cls_cursor_init(&cursor, &ofproto->facets, NULL);
1351 CLS_CURSOR_FOR_EACH (facet, cr, &cursor) {
1352 n_subfacets += list_size(&facet->subfacets);
1354 simap_increase(usage, "subfacets", n_subfacets);
1358 flush(struct ofproto *ofproto_)
1360 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1361 struct subfacet *subfacet, *next_subfacet;
1362 struct subfacet *batch[SUBFACET_DESTROY_MAX_BATCH];
1366 HMAP_FOR_EACH_SAFE (subfacet, next_subfacet, hmap_node,
1367 &ofproto->backer->subfacets) {
1368 if (ofproto_dpif_cast(subfacet->facet->rule->up.ofproto) != ofproto) {
1372 if (subfacet->path != SF_NOT_INSTALLED) {
1373 batch[n_batch++] = subfacet;
1374 if (n_batch >= SUBFACET_DESTROY_MAX_BATCH) {
1375 subfacet_destroy_batch(ofproto->backer, batch, n_batch);
1379 subfacet_destroy(subfacet);
1384 subfacet_destroy_batch(ofproto->backer, batch, n_batch);
1389 get_features(struct ofproto *ofproto_ OVS_UNUSED,
1390 bool *arp_match_ip, enum ofputil_action_bitmap *actions)
1392 *arp_match_ip = true;
1393 *actions = (OFPUTIL_A_OUTPUT |
1394 OFPUTIL_A_SET_VLAN_VID |
1395 OFPUTIL_A_SET_VLAN_PCP |
1396 OFPUTIL_A_STRIP_VLAN |
1397 OFPUTIL_A_SET_DL_SRC |
1398 OFPUTIL_A_SET_DL_DST |
1399 OFPUTIL_A_SET_NW_SRC |
1400 OFPUTIL_A_SET_NW_DST |
1401 OFPUTIL_A_SET_NW_TOS |
1402 OFPUTIL_A_SET_TP_SRC |
1403 OFPUTIL_A_SET_TP_DST |
1408 get_tables(struct ofproto *ofproto_, struct ofp12_table_stats *ots)
1410 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1411 struct dpif_dp_stats s;
1412 uint64_t n_miss, n_no_pkt_in, n_bytes, n_dropped_frags;
1415 strcpy(ots->name, "classifier");
1417 dpif_get_dp_stats(ofproto->backer->dpif, &s);
1418 rule_get_stats(&ofproto->miss_rule->up, &n_miss, &n_bytes);
1419 rule_get_stats(&ofproto->no_packet_in_rule->up, &n_no_pkt_in, &n_bytes);
1420 rule_get_stats(&ofproto->drop_frags_rule->up, &n_dropped_frags, &n_bytes);
1422 n_lookup = s.n_hit + s.n_missed - n_dropped_frags;
1423 ots->lookup_count = htonll(n_lookup);
1424 ots->matched_count = htonll(n_lookup - n_miss - n_no_pkt_in);
1427 static struct ofport *
1430 struct ofport_dpif *port = xmalloc(sizeof *port);
1435 port_dealloc(struct ofport *port_)
1437 struct ofport_dpif *port = ofport_dpif_cast(port_);
1442 port_construct(struct ofport *port_)
1444 struct ofport_dpif *port = ofport_dpif_cast(port_);
1445 struct ofproto_dpif *ofproto = ofproto_dpif_cast(port->up.ofproto);
1446 const struct netdev *netdev = port->up.netdev;
1447 char namebuf[NETDEV_VPORT_NAME_BUFSIZE];
1448 struct dpif_port dpif_port;
1451 ofproto->backer->need_revalidate = REV_RECONFIGURE;
1452 port->bundle = NULL;
1455 port->tag = tag_create_random();
1456 port->may_enable = true;
1457 port->stp_port = NULL;
1458 port->stp_state = STP_DISABLED;
1459 port->is_tunnel = false;
1461 hmap_init(&port->priorities);
1462 port->realdev_ofp_port = 0;
1463 port->vlandev_vid = 0;
1464 port->carrier_seq = netdev_get_carrier_resets(netdev);
1466 if (netdev_vport_is_patch(netdev)) {
1467 /* By bailing out here, we don't submit the port to the sFlow module
1468 * to be considered for counter polling export. This is correct
1469 * because the patch port represents an interface that sFlow considers
1470 * to be "internal" to the switch as a whole, and therefore not an
1471 * candidate for counter polling. */
1472 port->odp_port = ODPP_NONE;
1473 ofport_update_peer(port);
1477 error = dpif_port_query_by_name(ofproto->backer->dpif,
1478 netdev_vport_get_dpif_port(netdev, namebuf,
1485 port->odp_port = dpif_port.port_no;
1487 if (netdev_get_tunnel_config(netdev)) {
1488 tnl_port_add(port, port->up.netdev, port->odp_port);
1489 port->is_tunnel = true;
1491 /* Sanity-check that a mapping doesn't already exist. This
1492 * shouldn't happen for non-tunnel ports. */
1493 if (odp_port_to_ofp_port(ofproto, port->odp_port) != OFPP_NONE) {
1494 VLOG_ERR("port %s already has an OpenFlow port number",
1496 dpif_port_destroy(&dpif_port);
1500 hmap_insert(&ofproto->backer->odp_to_ofport_map, &port->odp_port_node,
1501 hash_odp_port(port->odp_port));
1503 dpif_port_destroy(&dpif_port);
1505 if (ofproto->sflow) {
1506 dpif_sflow_add_port(ofproto->sflow, port_, port->odp_port);
1513 port_destruct(struct ofport *port_)
1515 struct ofport_dpif *port = ofport_dpif_cast(port_);
1516 struct ofproto_dpif *ofproto = ofproto_dpif_cast(port->up.ofproto);
1517 const char *devname = netdev_get_name(port->up.netdev);
1518 char namebuf[NETDEV_VPORT_NAME_BUFSIZE];
1519 const char *dp_port_name;
1521 ofproto->backer->need_revalidate = REV_RECONFIGURE;
1523 dp_port_name = netdev_vport_get_dpif_port(port->up.netdev, namebuf,
1525 if (dpif_port_exists(ofproto->backer->dpif, dp_port_name)) {
1526 /* The underlying device is still there, so delete it. This
1527 * happens when the ofproto is being destroyed, since the caller
1528 * assumes that removal of attached ports will happen as part of
1530 if (!port->is_tunnel) {
1531 dpif_port_del(ofproto->backer->dpif, port->odp_port);
1536 port->peer->peer = NULL;
1540 if (port->odp_port != ODPP_NONE && !port->is_tunnel) {
1541 hmap_remove(&ofproto->backer->odp_to_ofport_map, &port->odp_port_node);
1545 sset_find_and_delete(&ofproto->ports, devname);
1546 sset_find_and_delete(&ofproto->ghost_ports, devname);
1547 bundle_remove(port_);
1548 set_cfm(port_, NULL);
1549 set_bfd(port_, NULL);
1550 if (ofproto->sflow) {
1551 dpif_sflow_del_port(ofproto->sflow, port->odp_port);
1554 ofport_clear_priorities(port);
1555 hmap_destroy(&port->priorities);
1559 port_modified(struct ofport *port_)
1561 struct ofport_dpif *port = ofport_dpif_cast(port_);
1563 if (port->bundle && port->bundle->bond) {
1564 bond_slave_set_netdev(port->bundle->bond, port, port->up.netdev);
1568 cfm_set_netdev(port->cfm, port->up.netdev);
1571 if (port->is_tunnel && tnl_port_reconfigure(port, port->up.netdev,
1573 ofproto_dpif_cast(port->up.ofproto)->backer->need_revalidate =
1577 ofport_update_peer(port);
1581 port_reconfigured(struct ofport *port_, enum ofputil_port_config old_config)
1583 struct ofport_dpif *port = ofport_dpif_cast(port_);
1584 struct ofproto_dpif *ofproto = ofproto_dpif_cast(port->up.ofproto);
1585 enum ofputil_port_config changed = old_config ^ port->up.pp.config;
1587 if (changed & (OFPUTIL_PC_NO_RECV | OFPUTIL_PC_NO_RECV_STP |
1588 OFPUTIL_PC_NO_FWD | OFPUTIL_PC_NO_FLOOD |
1589 OFPUTIL_PC_NO_PACKET_IN)) {
1590 ofproto->backer->need_revalidate = REV_RECONFIGURE;
1592 if (changed & OFPUTIL_PC_NO_FLOOD && port->bundle) {
1593 bundle_update(port->bundle);
1599 set_sflow(struct ofproto *ofproto_,
1600 const struct ofproto_sflow_options *sflow_options)
1602 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1603 struct dpif_sflow *ds = ofproto->sflow;
1605 if (sflow_options) {
1607 struct ofport_dpif *ofport;
1609 ds = ofproto->sflow = dpif_sflow_create();
1610 HMAP_FOR_EACH (ofport, up.hmap_node, &ofproto->up.ports) {
1611 dpif_sflow_add_port(ds, &ofport->up, ofport->odp_port);
1613 ofproto->backer->need_revalidate = REV_RECONFIGURE;
1615 dpif_sflow_set_options(ds, sflow_options);
1618 dpif_sflow_unref(ds);
1619 ofproto->backer->need_revalidate = REV_RECONFIGURE;
1620 ofproto->sflow = NULL;
1628 struct ofproto *ofproto_,
1629 const struct ofproto_ipfix_bridge_exporter_options *bridge_exporter_options,
1630 const struct ofproto_ipfix_flow_exporter_options *flow_exporters_options,
1631 size_t n_flow_exporters_options)
1633 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1634 struct dpif_ipfix *di = ofproto->ipfix;
1636 if (bridge_exporter_options || flow_exporters_options) {
1638 di = ofproto->ipfix = dpif_ipfix_create();
1640 dpif_ipfix_set_options(
1641 di, bridge_exporter_options, flow_exporters_options,
1642 n_flow_exporters_options);
1645 dpif_ipfix_unref(di);
1646 ofproto->ipfix = NULL;
1653 set_cfm(struct ofport *ofport_, const struct cfm_settings *s)
1655 struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
1662 struct ofproto_dpif *ofproto;
1664 ofproto = ofproto_dpif_cast(ofport->up.ofproto);
1665 ofproto->backer->need_revalidate = REV_RECONFIGURE;
1666 ofport->cfm = cfm_create(ofport->up.netdev);
1669 if (cfm_configure(ofport->cfm, s)) {
1675 cfm_unref(ofport->cfm);
1681 get_cfm_status(const struct ofport *ofport_,
1682 struct ofproto_cfm_status *status)
1684 struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
1687 status->faults = cfm_get_fault(ofport->cfm);
1688 status->remote_opstate = cfm_get_opup(ofport->cfm);
1689 status->health = cfm_get_health(ofport->cfm);
1690 cfm_get_remote_mpids(ofport->cfm, &status->rmps, &status->n_rmps);
1698 set_bfd(struct ofport *ofport_, const struct smap *cfg)
1700 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport_->ofproto);
1701 struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
1705 ofport->bfd = bfd_configure(old, netdev_get_name(ofport->up.netdev), cfg);
1706 if (ofport->bfd != old) {
1707 ofproto->backer->need_revalidate = REV_RECONFIGURE;
1714 get_bfd_status(struct ofport *ofport_, struct smap *smap)
1716 struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
1719 bfd_get_status(ofport->bfd, smap);
1726 /* Spanning Tree. */
1729 send_bpdu_cb(struct ofpbuf *pkt, int port_num, void *ofproto_)
1731 struct ofproto_dpif *ofproto = ofproto_;
1732 struct stp_port *sp = stp_get_port(ofproto->stp, port_num);
1733 struct ofport_dpif *ofport;
1735 ofport = stp_port_get_aux(sp);
1737 VLOG_WARN_RL(&rl, "%s: cannot send BPDU on unknown port %d",
1738 ofproto->up.name, port_num);
1740 struct eth_header *eth = pkt->l2;
1742 netdev_get_etheraddr(ofport->up.netdev, eth->eth_src);
1743 if (eth_addr_is_zero(eth->eth_src)) {
1744 VLOG_WARN_RL(&rl, "%s: cannot send BPDU on port %d "
1745 "with unknown MAC", ofproto->up.name, port_num);
1747 send_packet(ofport, pkt);
1753 /* Configures STP on 'ofproto_' using the settings defined in 's'. */
1755 set_stp(struct ofproto *ofproto_, const struct ofproto_stp_settings *s)
1757 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1759 /* Only revalidate flows if the configuration changed. */
1760 if (!s != !ofproto->stp) {
1761 ofproto->backer->need_revalidate = REV_RECONFIGURE;
1765 if (!ofproto->stp) {
1766 ofproto->stp = stp_create(ofproto_->name, s->system_id,
1767 send_bpdu_cb, ofproto);
1768 ofproto->stp_last_tick = time_msec();
1771 stp_set_bridge_id(ofproto->stp, s->system_id);
1772 stp_set_bridge_priority(ofproto->stp, s->priority);
1773 stp_set_hello_time(ofproto->stp, s->hello_time);
1774 stp_set_max_age(ofproto->stp, s->max_age);
1775 stp_set_forward_delay(ofproto->stp, s->fwd_delay);
1777 struct ofport *ofport;
1779 HMAP_FOR_EACH (ofport, hmap_node, &ofproto->up.ports) {
1780 set_stp_port(ofport, NULL);
1783 stp_destroy(ofproto->stp);
1784 ofproto->stp = NULL;
1791 get_stp_status(struct ofproto *ofproto_, struct ofproto_stp_status *s)
1793 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1797 s->bridge_id = stp_get_bridge_id(ofproto->stp);
1798 s->designated_root = stp_get_designated_root(ofproto->stp);
1799 s->root_path_cost = stp_get_root_path_cost(ofproto->stp);
1808 update_stp_port_state(struct ofport_dpif *ofport)
1810 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport->up.ofproto);
1811 enum stp_state state;
1813 /* Figure out new state. */
1814 state = ofport->stp_port ? stp_port_get_state(ofport->stp_port)
1818 if (ofport->stp_state != state) {
1819 enum ofputil_port_state of_state;
1822 VLOG_DBG_RL(&rl, "port %s: STP state changed from %s to %s",
1823 netdev_get_name(ofport->up.netdev),
1824 stp_state_name(ofport->stp_state),
1825 stp_state_name(state));
1826 if (stp_learn_in_state(ofport->stp_state)
1827 != stp_learn_in_state(state)) {
1828 /* xxx Learning action flows should also be flushed. */
1829 mac_learning_flush(ofproto->ml,
1830 &ofproto->backer->revalidate_set);
1832 fwd_change = stp_forward_in_state(ofport->stp_state)
1833 != stp_forward_in_state(state);
1835 ofproto->backer->need_revalidate = REV_STP;
1836 ofport->stp_state = state;
1837 ofport->stp_state_entered = time_msec();
1839 if (fwd_change && ofport->bundle) {
1840 bundle_update(ofport->bundle);
1843 /* Update the STP state bits in the OpenFlow port description. */
1844 of_state = ofport->up.pp.state & ~OFPUTIL_PS_STP_MASK;
1845 of_state |= (state == STP_LISTENING ? OFPUTIL_PS_STP_LISTEN
1846 : state == STP_LEARNING ? OFPUTIL_PS_STP_LEARN
1847 : state == STP_FORWARDING ? OFPUTIL_PS_STP_FORWARD
1848 : state == STP_BLOCKING ? OFPUTIL_PS_STP_BLOCK
1850 ofproto_port_set_state(&ofport->up, of_state);
1854 /* Configures STP on 'ofport_' using the settings defined in 's'. The
1855 * caller is responsible for assigning STP port numbers and ensuring
1856 * there are no duplicates. */
1858 set_stp_port(struct ofport *ofport_,
1859 const struct ofproto_port_stp_settings *s)
1861 struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
1862 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport->up.ofproto);
1863 struct stp_port *sp = ofport->stp_port;
1865 if (!s || !s->enable) {
1867 ofport->stp_port = NULL;
1868 stp_port_disable(sp);
1869 update_stp_port_state(ofport);
1872 } else if (sp && stp_port_no(sp) != s->port_num
1873 && ofport == stp_port_get_aux(sp)) {
1874 /* The port-id changed, so disable the old one if it's not
1875 * already in use by another port. */
1876 stp_port_disable(sp);
1879 sp = ofport->stp_port = stp_get_port(ofproto->stp, s->port_num);
1880 stp_port_enable(sp);
1882 stp_port_set_aux(sp, ofport);
1883 stp_port_set_priority(sp, s->priority);
1884 stp_port_set_path_cost(sp, s->path_cost);
1886 update_stp_port_state(ofport);
1892 get_stp_port_status(struct ofport *ofport_,
1893 struct ofproto_port_stp_status *s)
1895 struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
1896 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport->up.ofproto);
1897 struct stp_port *sp = ofport->stp_port;
1899 if (!ofproto->stp || !sp) {
1905 s->port_id = stp_port_get_id(sp);
1906 s->state = stp_port_get_state(sp);
1907 s->sec_in_state = (time_msec() - ofport->stp_state_entered) / 1000;
1908 s->role = stp_port_get_role(sp);
1909 stp_port_get_counts(sp, &s->tx_count, &s->rx_count, &s->error_count);
1915 stp_run(struct ofproto_dpif *ofproto)
1918 long long int now = time_msec();
1919 long long int elapsed = now - ofproto->stp_last_tick;
1920 struct stp_port *sp;
1923 stp_tick(ofproto->stp, MIN(INT_MAX, elapsed));
1924 ofproto->stp_last_tick = now;
1926 while (stp_get_changed_port(ofproto->stp, &sp)) {
1927 struct ofport_dpif *ofport = stp_port_get_aux(sp);
1930 update_stp_port_state(ofport);
1934 if (stp_check_and_reset_fdb_flush(ofproto->stp)) {
1935 mac_learning_flush(ofproto->ml, &ofproto->backer->revalidate_set);
1941 stp_wait(struct ofproto_dpif *ofproto)
1944 poll_timer_wait(1000);
1948 /* Returns true if STP should process 'flow'. Sets fields in 'wc' that
1949 * were used to make the determination.*/
1951 stp_should_process_flow(const struct flow *flow, struct flow_wildcards *wc)
1953 memset(&wc->masks.dl_dst, 0xff, sizeof wc->masks.dl_dst);
1954 return eth_addr_equals(flow->dl_dst, eth_addr_stp);
1958 stp_process_packet(const struct ofport_dpif *ofport,
1959 const struct ofpbuf *packet)
1961 struct ofpbuf payload = *packet;
1962 struct eth_header *eth = payload.data;
1963 struct stp_port *sp = ofport->stp_port;
1965 /* Sink packets on ports that have STP disabled when the bridge has
1967 if (!sp || stp_port_get_state(sp) == STP_DISABLED) {
1971 /* Trim off padding on payload. */
1972 if (payload.size > ntohs(eth->eth_type) + ETH_HEADER_LEN) {
1973 payload.size = ntohs(eth->eth_type) + ETH_HEADER_LEN;
1976 if (ofpbuf_try_pull(&payload, ETH_HEADER_LEN + LLC_HEADER_LEN)) {
1977 stp_received_bpdu(sp, payload.data, payload.size);
1982 ofproto_dpif_queue_to_priority(const struct ofproto_dpif *ofproto,
1983 uint32_t queue_id, uint32_t *priority)
1985 return dpif_queue_to_priority(ofproto->backer->dpif, queue_id, priority);
1988 static struct priority_to_dscp *
1989 get_priority(const struct ofport_dpif *ofport, uint32_t priority)
1991 struct priority_to_dscp *pdscp;
1994 hash = hash_int(priority, 0);
1995 HMAP_FOR_EACH_IN_BUCKET (pdscp, hmap_node, hash, &ofport->priorities) {
1996 if (pdscp->priority == priority) {
2004 ofproto_dpif_dscp_from_priority(const struct ofport_dpif *ofport,
2005 uint32_t priority, uint8_t *dscp)
2007 struct priority_to_dscp *pdscp = get_priority(ofport, priority);
2008 *dscp = pdscp ? pdscp->dscp : 0;
2009 return pdscp != NULL;
2013 ofport_clear_priorities(struct ofport_dpif *ofport)
2015 struct priority_to_dscp *pdscp, *next;
2017 HMAP_FOR_EACH_SAFE (pdscp, next, hmap_node, &ofport->priorities) {
2018 hmap_remove(&ofport->priorities, &pdscp->hmap_node);
2024 set_queues(struct ofport *ofport_,
2025 const struct ofproto_port_queue *qdscp_list,
2028 struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
2029 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport->up.ofproto);
2030 struct hmap new = HMAP_INITIALIZER(&new);
2033 for (i = 0; i < n_qdscp; i++) {
2034 struct priority_to_dscp *pdscp;
2038 dscp = (qdscp_list[i].dscp << 2) & IP_DSCP_MASK;
2039 if (dpif_queue_to_priority(ofproto->backer->dpif, qdscp_list[i].queue,
2044 pdscp = get_priority(ofport, priority);
2046 hmap_remove(&ofport->priorities, &pdscp->hmap_node);
2048 pdscp = xmalloc(sizeof *pdscp);
2049 pdscp->priority = priority;
2051 ofproto->backer->need_revalidate = REV_RECONFIGURE;
2054 if (pdscp->dscp != dscp) {
2056 ofproto->backer->need_revalidate = REV_RECONFIGURE;
2059 hmap_insert(&new, &pdscp->hmap_node, hash_int(pdscp->priority, 0));
2062 if (!hmap_is_empty(&ofport->priorities)) {
2063 ofport_clear_priorities(ofport);
2064 ofproto->backer->need_revalidate = REV_RECONFIGURE;
2067 hmap_swap(&new, &ofport->priorities);
2075 /* Expires all MAC learning entries associated with 'bundle' and forces its
2076 * ofproto to revalidate every flow.
2078 * Normally MAC learning entries are removed only from the ofproto associated
2079 * with 'bundle', but if 'all_ofprotos' is true, then the MAC learning entries
2080 * are removed from every ofproto. When patch ports and SLB bonds are in use
2081 * and a VM migration happens and the gratuitous ARPs are somehow lost, this
2082 * avoids a MAC_ENTRY_IDLE_TIME delay before the migrated VM can communicate
2083 * with the host from which it migrated. */
2085 bundle_flush_macs(struct ofbundle *bundle, bool all_ofprotos)
2087 struct ofproto_dpif *ofproto = bundle->ofproto;
2088 struct mac_learning *ml = ofproto->ml;
2089 struct mac_entry *mac, *next_mac;
2091 ofproto->backer->need_revalidate = REV_RECONFIGURE;
2092 LIST_FOR_EACH_SAFE (mac, next_mac, lru_node, &ml->lrus) {
2093 if (mac->port.p == bundle) {
2095 struct ofproto_dpif *o;
2097 HMAP_FOR_EACH (o, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
2099 struct mac_entry *e;
2101 e = mac_learning_lookup(o->ml, mac->mac, mac->vlan,
2104 mac_learning_expire(o->ml, e);
2110 mac_learning_expire(ml, mac);
2115 static struct ofbundle *
2116 bundle_lookup(const struct ofproto_dpif *ofproto, void *aux)
2118 struct ofbundle *bundle;
2120 HMAP_FOR_EACH_IN_BUCKET (bundle, hmap_node, hash_pointer(aux, 0),
2121 &ofproto->bundles) {
2122 if (bundle->aux == aux) {
2130 bundle_update(struct ofbundle *bundle)
2132 struct ofport_dpif *port;
2134 bundle->floodable = true;
2135 LIST_FOR_EACH (port, bundle_node, &bundle->ports) {
2136 if (port->up.pp.config & OFPUTIL_PC_NO_FLOOD
2137 || !stp_forward_in_state(port->stp_state)) {
2138 bundle->floodable = false;
2145 bundle_del_port(struct ofport_dpif *port)
2147 struct ofbundle *bundle = port->bundle;
2149 bundle->ofproto->backer->need_revalidate = REV_RECONFIGURE;
2151 list_remove(&port->bundle_node);
2152 port->bundle = NULL;
2155 lacp_slave_unregister(bundle->lacp, port);
2158 bond_slave_unregister(bundle->bond, port);
2161 bundle_update(bundle);
2165 bundle_add_port(struct ofbundle *bundle, ofp_port_t ofp_port,
2166 struct lacp_slave_settings *lacp)
2168 struct ofport_dpif *port;
2170 port = get_ofp_port(bundle->ofproto, ofp_port);
2175 if (port->bundle != bundle) {
2176 bundle->ofproto->backer->need_revalidate = REV_RECONFIGURE;
2178 bundle_del_port(port);
2181 port->bundle = bundle;
2182 list_push_back(&bundle->ports, &port->bundle_node);
2183 if (port->up.pp.config & OFPUTIL_PC_NO_FLOOD
2184 || !stp_forward_in_state(port->stp_state)) {
2185 bundle->floodable = false;
2189 bundle->ofproto->backer->need_revalidate = REV_RECONFIGURE;
2190 lacp_slave_register(bundle->lacp, port, lacp);
2197 bundle_destroy(struct ofbundle *bundle)
2199 struct ofproto_dpif *ofproto;
2200 struct ofport_dpif *port, *next_port;
2206 ofproto = bundle->ofproto;
2207 mbridge_unregister_bundle(ofproto->mbridge, bundle->aux);
2209 LIST_FOR_EACH_SAFE (port, next_port, bundle_node, &bundle->ports) {
2210 bundle_del_port(port);
2213 bundle_flush_macs(bundle, true);
2214 hmap_remove(&ofproto->bundles, &bundle->hmap_node);
2216 free(bundle->trunks);
2217 lacp_unref(bundle->lacp);
2218 bond_unref(bundle->bond);
2223 bundle_set(struct ofproto *ofproto_, void *aux,
2224 const struct ofproto_bundle_settings *s)
2226 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
2227 bool need_flush = false;
2228 struct ofport_dpif *port;
2229 struct ofbundle *bundle;
2230 unsigned long *trunks;
2236 bundle_destroy(bundle_lookup(ofproto, aux));
2240 ovs_assert(s->n_slaves == 1 || s->bond != NULL);
2241 ovs_assert((s->lacp != NULL) == (s->lacp_slaves != NULL));
2243 bundle = bundle_lookup(ofproto, aux);
2245 bundle = xmalloc(sizeof *bundle);
2247 bundle->ofproto = ofproto;
2248 hmap_insert(&ofproto->bundles, &bundle->hmap_node,
2249 hash_pointer(aux, 0));
2251 bundle->name = NULL;
2253 list_init(&bundle->ports);
2254 bundle->vlan_mode = PORT_VLAN_TRUNK;
2256 bundle->trunks = NULL;
2257 bundle->use_priority_tags = s->use_priority_tags;
2258 bundle->lacp = NULL;
2259 bundle->bond = NULL;
2261 bundle->floodable = true;
2262 mbridge_register_bundle(ofproto->mbridge, bundle);
2265 if (!bundle->name || strcmp(s->name, bundle->name)) {
2267 bundle->name = xstrdup(s->name);
2272 if (!bundle->lacp) {
2273 ofproto->backer->need_revalidate = REV_RECONFIGURE;
2274 bundle->lacp = lacp_create();
2276 lacp_configure(bundle->lacp, s->lacp);
2278 lacp_unref(bundle->lacp);
2279 bundle->lacp = NULL;
2282 /* Update set of ports. */
2284 for (i = 0; i < s->n_slaves; i++) {
2285 if (!bundle_add_port(bundle, s->slaves[i],
2286 s->lacp ? &s->lacp_slaves[i] : NULL)) {
2290 if (!ok || list_size(&bundle->ports) != s->n_slaves) {
2291 struct ofport_dpif *next_port;
2293 LIST_FOR_EACH_SAFE (port, next_port, bundle_node, &bundle->ports) {
2294 for (i = 0; i < s->n_slaves; i++) {
2295 if (s->slaves[i] == port->up.ofp_port) {
2300 bundle_del_port(port);
2304 ovs_assert(list_size(&bundle->ports) <= s->n_slaves);
2306 if (list_is_empty(&bundle->ports)) {
2307 bundle_destroy(bundle);
2311 /* Set VLAN tagging mode */
2312 if (s->vlan_mode != bundle->vlan_mode
2313 || s->use_priority_tags != bundle->use_priority_tags) {
2314 bundle->vlan_mode = s->vlan_mode;
2315 bundle->use_priority_tags = s->use_priority_tags;
2320 vlan = (s->vlan_mode == PORT_VLAN_TRUNK ? -1
2321 : s->vlan >= 0 && s->vlan <= 4095 ? s->vlan
2323 if (vlan != bundle->vlan) {
2324 bundle->vlan = vlan;
2328 /* Get trunked VLANs. */
2329 switch (s->vlan_mode) {
2330 case PORT_VLAN_ACCESS:
2334 case PORT_VLAN_TRUNK:
2335 trunks = CONST_CAST(unsigned long *, s->trunks);
2338 case PORT_VLAN_NATIVE_UNTAGGED:
2339 case PORT_VLAN_NATIVE_TAGGED:
2340 if (vlan != 0 && (!s->trunks
2341 || !bitmap_is_set(s->trunks, vlan)
2342 || bitmap_is_set(s->trunks, 0))) {
2343 /* Force trunking the native VLAN and prohibit trunking VLAN 0. */
2345 trunks = bitmap_clone(s->trunks, 4096);
2347 trunks = bitmap_allocate1(4096);
2349 bitmap_set1(trunks, vlan);
2350 bitmap_set0(trunks, 0);
2352 trunks = CONST_CAST(unsigned long *, s->trunks);
2359 if (!vlan_bitmap_equal(trunks, bundle->trunks)) {
2360 free(bundle->trunks);
2361 if (trunks == s->trunks) {
2362 bundle->trunks = vlan_bitmap_clone(trunks);
2364 bundle->trunks = trunks;
2369 if (trunks != s->trunks) {
2374 if (!list_is_short(&bundle->ports)) {
2375 bundle->ofproto->has_bonded_bundles = true;
2377 if (bond_reconfigure(bundle->bond, s->bond)) {
2378 ofproto->backer->need_revalidate = REV_RECONFIGURE;
2381 bundle->bond = bond_create(s->bond);
2382 ofproto->backer->need_revalidate = REV_RECONFIGURE;
2385 LIST_FOR_EACH (port, bundle_node, &bundle->ports) {
2386 bond_slave_register(bundle->bond, port, port->up.netdev);
2389 bond_unref(bundle->bond);
2390 bundle->bond = NULL;
2393 /* If we changed something that would affect MAC learning, un-learn
2394 * everything on this port and force flow revalidation. */
2396 bundle_flush_macs(bundle, false);
2403 bundle_remove(struct ofport *port_)
2405 struct ofport_dpif *port = ofport_dpif_cast(port_);
2406 struct ofbundle *bundle = port->bundle;
2409 bundle_del_port(port);
2410 if (list_is_empty(&bundle->ports)) {
2411 bundle_destroy(bundle);
2412 } else if (list_is_short(&bundle->ports)) {
2413 bond_unref(bundle->bond);
2414 bundle->bond = NULL;
2420 send_pdu_cb(void *port_, const void *pdu, size_t pdu_size)
2422 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 10);
2423 struct ofport_dpif *port = port_;
2424 uint8_t ea[ETH_ADDR_LEN];
2427 error = netdev_get_etheraddr(port->up.netdev, ea);
2429 struct ofpbuf packet;
2432 ofpbuf_init(&packet, 0);
2433 packet_pdu = eth_compose(&packet, eth_addr_lacp, ea, ETH_TYPE_LACP,
2435 memcpy(packet_pdu, pdu, pdu_size);
2437 send_packet(port, &packet);
2438 ofpbuf_uninit(&packet);
2440 VLOG_ERR_RL(&rl, "port %s: cannot obtain Ethernet address of iface "
2441 "%s (%s)", port->bundle->name,
2442 netdev_get_name(port->up.netdev), ovs_strerror(error));
2447 bundle_send_learning_packets(struct ofbundle *bundle)
2449 struct ofproto_dpif *ofproto = bundle->ofproto;
2450 int error, n_packets, n_errors;
2451 struct mac_entry *e;
2453 error = n_packets = n_errors = 0;
2454 LIST_FOR_EACH (e, lru_node, &ofproto->ml->lrus) {
2455 if (e->port.p != bundle) {
2456 struct ofpbuf *learning_packet;
2457 struct ofport_dpif *port;
2461 /* The assignment to "port" is unnecessary but makes "grep"ing for
2462 * struct ofport_dpif more effective. */
2463 learning_packet = bond_compose_learning_packet(bundle->bond,
2467 ret = send_packet(port, learning_packet);
2468 ofpbuf_delete(learning_packet);
2478 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
2479 VLOG_WARN_RL(&rl, "bond %s: %d errors sending %d gratuitous learning "
2480 "packets, last error was: %s",
2481 bundle->name, n_errors, n_packets, ovs_strerror(error));
2483 VLOG_DBG("bond %s: sent %d gratuitous learning packets",
2484 bundle->name, n_packets);
2489 bundle_run(struct ofbundle *bundle)
2492 lacp_run(bundle->lacp, send_pdu_cb);
2495 struct ofport_dpif *port;
2497 LIST_FOR_EACH (port, bundle_node, &bundle->ports) {
2498 bond_slave_set_may_enable(bundle->bond, port, port->may_enable);
2501 bond_run(bundle->bond, &bundle->ofproto->backer->revalidate_set,
2502 lacp_status(bundle->lacp));
2503 if (bond_should_send_learning_packets(bundle->bond)) {
2504 bundle_send_learning_packets(bundle);
2510 bundle_wait(struct ofbundle *bundle)
2513 lacp_wait(bundle->lacp);
2516 bond_wait(bundle->bond);
2523 mirror_set__(struct ofproto *ofproto_, void *aux,
2524 const struct ofproto_mirror_settings *s)
2526 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
2527 struct ofbundle **srcs, **dsts;
2532 mirror_destroy(ofproto->mbridge, aux);
2536 srcs = xmalloc(s->n_srcs * sizeof *srcs);
2537 dsts = xmalloc(s->n_dsts * sizeof *dsts);
2539 for (i = 0; i < s->n_srcs; i++) {
2540 srcs[i] = bundle_lookup(ofproto, s->srcs[i]);
2543 for (i = 0; i < s->n_dsts; i++) {
2544 dsts[i] = bundle_lookup(ofproto, s->dsts[i]);
2547 error = mirror_set(ofproto->mbridge, aux, s->name, srcs, s->n_srcs, dsts,
2548 s->n_dsts, s->src_vlans,
2549 bundle_lookup(ofproto, s->out_bundle), s->out_vlan);
2556 mirror_get_stats__(struct ofproto *ofproto, void *aux,
2557 uint64_t *packets, uint64_t *bytes)
2560 return mirror_get_stats(ofproto_dpif_cast(ofproto)->mbridge, aux, packets,
2565 set_flood_vlans(struct ofproto *ofproto_, unsigned long *flood_vlans)
2567 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
2568 if (mac_learning_set_flood_vlans(ofproto->ml, flood_vlans)) {
2569 mac_learning_flush(ofproto->ml, &ofproto->backer->revalidate_set);
2575 is_mirror_output_bundle(const struct ofproto *ofproto_, void *aux)
2577 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
2578 struct ofbundle *bundle = bundle_lookup(ofproto, aux);
2579 return bundle && mirror_bundle_out(ofproto->mbridge, bundle) != 0;
2583 forward_bpdu_changed(struct ofproto *ofproto_)
2585 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
2586 ofproto->backer->need_revalidate = REV_RECONFIGURE;
2590 set_mac_table_config(struct ofproto *ofproto_, unsigned int idle_time,
2593 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
2594 mac_learning_set_idle_time(ofproto->ml, idle_time);
2595 mac_learning_set_max_entries(ofproto->ml, max_entries);
2600 struct ofport_dpif *
2601 get_ofp_port(const struct ofproto_dpif *ofproto, ofp_port_t ofp_port)
2603 struct ofport *ofport = ofproto_get_port(&ofproto->up, ofp_port);
2604 return ofport ? ofport_dpif_cast(ofport) : NULL;
2607 struct ofport_dpif *
2608 get_odp_port(const struct ofproto_dpif *ofproto, odp_port_t odp_port)
2610 struct ofport_dpif *port = odp_port_to_ofport(ofproto->backer, odp_port);
2611 return port && &ofproto->up == port->up.ofproto ? port : NULL;
2615 ofproto_port_from_dpif_port(struct ofproto_dpif *ofproto,
2616 struct ofproto_port *ofproto_port,
2617 struct dpif_port *dpif_port)
2619 ofproto_port->name = dpif_port->name;
2620 ofproto_port->type = dpif_port->type;
2621 ofproto_port->ofp_port = odp_port_to_ofp_port(ofproto, dpif_port->port_no);
2625 ofport_update_peer(struct ofport_dpif *ofport)
2627 const struct ofproto_dpif *ofproto;
2628 struct dpif_backer *backer;
2629 const char *peer_name;
2631 if (!netdev_vport_is_patch(ofport->up.netdev)) {
2635 backer = ofproto_dpif_cast(ofport->up.ofproto)->backer;
2636 backer->need_revalidate = REV_RECONFIGURE;
2639 ofport->peer->peer = NULL;
2640 ofport->peer = NULL;
2643 peer_name = netdev_vport_patch_peer(ofport->up.netdev);
2648 HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
2649 struct ofport *peer_ofport;
2650 struct ofport_dpif *peer;
2651 const char *peer_peer;
2653 if (ofproto->backer != backer) {
2657 peer_ofport = shash_find_data(&ofproto->up.port_by_name, peer_name);
2662 peer = ofport_dpif_cast(peer_ofport);
2663 peer_peer = netdev_vport_patch_peer(peer->up.netdev);
2664 if (peer_peer && !strcmp(netdev_get_name(ofport->up.netdev),
2666 ofport->peer = peer;
2667 ofport->peer->peer = ofport;
2675 port_run_fast(struct ofport_dpif *ofport)
2677 if (ofport->cfm && cfm_should_send_ccm(ofport->cfm)) {
2678 struct ofpbuf packet;
2680 ofpbuf_init(&packet, 0);
2681 cfm_compose_ccm(ofport->cfm, &packet, ofport->up.pp.hw_addr);
2682 send_packet(ofport, &packet);
2683 ofpbuf_uninit(&packet);
2686 if (ofport->bfd && bfd_should_send_packet(ofport->bfd)) {
2687 struct ofpbuf packet;
2689 ofpbuf_init(&packet, 0);
2690 bfd_put_packet(ofport->bfd, &packet, ofport->up.pp.hw_addr);
2691 send_packet(ofport, &packet);
2692 ofpbuf_uninit(&packet);
2697 port_run(struct ofport_dpif *ofport)
2699 long long int carrier_seq = netdev_get_carrier_resets(ofport->up.netdev);
2700 bool carrier_changed = carrier_seq != ofport->carrier_seq;
2701 bool enable = netdev_get_carrier(ofport->up.netdev);
2703 ofport->carrier_seq = carrier_seq;
2705 port_run_fast(ofport);
2708 int cfm_opup = cfm_get_opup(ofport->cfm);
2710 cfm_run(ofport->cfm);
2711 enable = enable && !cfm_get_fault(ofport->cfm);
2713 if (cfm_opup >= 0) {
2714 enable = enable && cfm_opup;
2719 bfd_run(ofport->bfd);
2720 enable = enable && bfd_forwarding(ofport->bfd);
2723 if (ofport->bundle) {
2724 enable = enable && lacp_slave_may_enable(ofport->bundle->lacp, ofport);
2725 if (carrier_changed) {
2726 lacp_slave_carrier_changed(ofport->bundle->lacp, ofport);
2730 if (ofport->may_enable != enable) {
2731 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport->up.ofproto);
2732 ofproto->backer->need_revalidate = REV_PORT_TOGGLED;
2735 ofport->may_enable = enable;
2739 port_wait(struct ofport_dpif *ofport)
2742 cfm_wait(ofport->cfm);
2746 bfd_wait(ofport->bfd);
2751 port_query_by_name(const struct ofproto *ofproto_, const char *devname,
2752 struct ofproto_port *ofproto_port)
2754 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
2755 struct dpif_port dpif_port;
2758 if (sset_contains(&ofproto->ghost_ports, devname)) {
2759 const char *type = netdev_get_type_from_name(devname);
2761 /* We may be called before ofproto->up.port_by_name is populated with
2762 * the appropriate ofport. For this reason, we must get the name and
2763 * type from the netdev layer directly. */
2765 const struct ofport *ofport;
2767 ofport = shash_find_data(&ofproto->up.port_by_name, devname);
2768 ofproto_port->ofp_port = ofport ? ofport->ofp_port : OFPP_NONE;
2769 ofproto_port->name = xstrdup(devname);
2770 ofproto_port->type = xstrdup(type);
2776 if (!sset_contains(&ofproto->ports, devname)) {
2779 error = dpif_port_query_by_name(ofproto->backer->dpif,
2780 devname, &dpif_port);
2782 ofproto_port_from_dpif_port(ofproto, ofproto_port, &dpif_port);
2788 port_add(struct ofproto *ofproto_, struct netdev *netdev)
2790 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
2791 const char *devname = netdev_get_name(netdev);
2792 char namebuf[NETDEV_VPORT_NAME_BUFSIZE];
2793 const char *dp_port_name;
2795 if (netdev_vport_is_patch(netdev)) {
2796 sset_add(&ofproto->ghost_ports, netdev_get_name(netdev));
2800 dp_port_name = netdev_vport_get_dpif_port(netdev, namebuf, sizeof namebuf);
2801 if (!dpif_port_exists(ofproto->backer->dpif, dp_port_name)) {
2802 odp_port_t port_no = ODPP_NONE;
2805 error = dpif_port_add(ofproto->backer->dpif, netdev, &port_no);
2809 if (netdev_get_tunnel_config(netdev)) {
2810 simap_put(&ofproto->backer->tnl_backers,
2811 dp_port_name, odp_to_u32(port_no));
2815 if (netdev_get_tunnel_config(netdev)) {
2816 sset_add(&ofproto->ghost_ports, devname);
2818 sset_add(&ofproto->ports, devname);
2824 port_del(struct ofproto *ofproto_, ofp_port_t ofp_port)
2826 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
2827 struct ofport_dpif *ofport = get_ofp_port(ofproto, ofp_port);
2834 sset_find_and_delete(&ofproto->ghost_ports,
2835 netdev_get_name(ofport->up.netdev));
2836 ofproto->backer->need_revalidate = REV_RECONFIGURE;
2837 if (!ofport->is_tunnel) {
2838 error = dpif_port_del(ofproto->backer->dpif, ofport->odp_port);
2840 /* The caller is going to close ofport->up.netdev. If this is a
2841 * bonded port, then the bond is using that netdev, so remove it
2842 * from the bond. The client will need to reconfigure everything
2843 * after deleting ports, so then the slave will get re-added. */
2844 bundle_remove(&ofport->up);
2851 port_get_stats(const struct ofport *ofport_, struct netdev_stats *stats)
2853 struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
2858 error = netdev_get_stats(ofport->up.netdev, stats);
2860 if (!error && ofport_->ofp_port == OFPP_LOCAL) {
2861 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport->up.ofproto);
2863 /* ofproto->stats.tx_packets represents packets that we created
2864 * internally and sent to some port (e.g. packets sent with
2865 * send_packet()). Account for them as if they had come from
2866 * OFPP_LOCAL and got forwarded. */
2868 if (stats->rx_packets != UINT64_MAX) {
2869 stats->rx_packets += ofproto->stats.tx_packets;
2872 if (stats->rx_bytes != UINT64_MAX) {
2873 stats->rx_bytes += ofproto->stats.tx_bytes;
2876 /* ofproto->stats.rx_packets represents packets that were received on
2877 * some port and we processed internally and dropped (e.g. STP).
2878 * Account for them as if they had been forwarded to OFPP_LOCAL. */
2880 if (stats->tx_packets != UINT64_MAX) {
2881 stats->tx_packets += ofproto->stats.rx_packets;
2884 if (stats->tx_bytes != UINT64_MAX) {
2885 stats->tx_bytes += ofproto->stats.rx_bytes;
2892 struct port_dump_state {
2897 struct ofproto_port port;
2902 port_dump_start(const struct ofproto *ofproto_ OVS_UNUSED, void **statep)
2904 *statep = xzalloc(sizeof(struct port_dump_state));
2909 port_dump_next(const struct ofproto *ofproto_, void *state_,
2910 struct ofproto_port *port)
2912 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
2913 struct port_dump_state *state = state_;
2914 const struct sset *sset;
2915 struct sset_node *node;
2917 if (state->has_port) {
2918 ofproto_port_destroy(&state->port);
2919 state->has_port = false;
2921 sset = state->ghost ? &ofproto->ghost_ports : &ofproto->ports;
2922 while ((node = sset_at_position(sset, &state->bucket, &state->offset))) {
2925 error = port_query_by_name(ofproto_, node->name, &state->port);
2927 *port = state->port;
2928 state->has_port = true;
2930 } else if (error != ENODEV) {
2935 if (!state->ghost) {
2936 state->ghost = true;
2939 return port_dump_next(ofproto_, state_, port);
2946 port_dump_done(const struct ofproto *ofproto_ OVS_UNUSED, void *state_)
2948 struct port_dump_state *state = state_;
2950 if (state->has_port) {
2951 ofproto_port_destroy(&state->port);
2958 port_poll(const struct ofproto *ofproto_, char **devnamep)
2960 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
2962 if (ofproto->port_poll_errno) {
2963 int error = ofproto->port_poll_errno;
2964 ofproto->port_poll_errno = 0;
2968 if (sset_is_empty(&ofproto->port_poll_set)) {
2972 *devnamep = sset_pop(&ofproto->port_poll_set);
2977 port_poll_wait(const struct ofproto *ofproto_)
2979 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
2980 dpif_port_poll_wait(ofproto->backer->dpif);
2984 port_is_lacp_current(const struct ofport *ofport_)
2986 const struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
2987 return (ofport->bundle && ofport->bundle->lacp
2988 ? lacp_slave_is_current(ofport->bundle->lacp, ofport)
2992 /* Upcall handling. */
2994 /* Flow miss batching.
2996 * Some dpifs implement operations faster when you hand them off in a batch.
2997 * To allow batching, "struct flow_miss" queues the dpif-related work needed
2998 * for a given flow. Each "struct flow_miss" corresponds to sending one or
2999 * more packets, plus possibly installing the flow in the dpif.
3001 * So far we only batch the operations that affect flow setup time the most.
3002 * It's possible to batch more than that, but the benefit might be minimal. */
3004 struct hmap_node hmap_node;
3005 struct ofproto_dpif *ofproto;
3007 enum odp_key_fitness key_fitness;
3008 const struct nlattr *key;
3010 struct list packets;
3011 enum dpif_upcall_type upcall_type;
3014 struct flow_miss_op {
3015 struct dpif_op dpif_op;
3017 uint64_t slow_stub[128 / 8]; /* Buffer for compose_slow_path() */
3018 struct xlate_out xout;
3019 bool xout_garbage; /* 'xout' needs to be uninitialized? */
3021 struct ofpbuf mask; /* Flow mask for "put" ops. */
3022 struct odputil_keybuf maskbuf;
3024 /* If this is a "put" op, then a pointer to the subfacet that should
3025 * be marked as uninstalled if the operation fails. */
3026 struct subfacet *subfacet;
3029 /* Sends an OFPT_PACKET_IN message for 'packet' of type OFPR_NO_MATCH to each
3030 * OpenFlow controller as necessary according to their individual
3031 * configurations. */
3033 send_packet_in_miss(struct ofproto_dpif *ofproto, const struct ofpbuf *packet,
3034 const struct flow *flow)
3036 struct ofputil_packet_in pin;
3038 pin.packet = packet->data;
3039 pin.packet_len = packet->size;
3040 pin.reason = OFPR_NO_MATCH;
3041 pin.controller_id = 0;
3046 pin.send_len = 0; /* not used for flow table misses */
3048 flow_get_metadata(flow, &pin.fmd);
3050 connmgr_send_packet_in(ofproto->up.connmgr, &pin);
3053 static struct flow_miss *
3054 flow_miss_find(struct hmap *todo, const struct ofproto_dpif *ofproto,
3055 const struct flow *flow, uint32_t hash)
3057 struct flow_miss *miss;
3059 HMAP_FOR_EACH_WITH_HASH (miss, hmap_node, hash, todo) {
3060 if (miss->ofproto == ofproto && flow_equal(&miss->flow, flow)) {
3068 /* Partially Initializes 'op' as an "execute" operation for 'miss' and
3069 * 'packet'. The caller must initialize op->actions and op->actions_len. If
3070 * 'miss' is associated with a subfacet the caller must also initialize the
3071 * returned op->subfacet, and if anything needs to be freed after processing
3072 * the op, the caller must initialize op->garbage also. */
3074 init_flow_miss_execute_op(struct flow_miss *miss, struct ofpbuf *packet,
3075 struct flow_miss_op *op)
3077 if (miss->flow.in_port.ofp_port
3078 != vsp_realdev_to_vlandev(miss->ofproto, miss->flow.in_port.ofp_port,
3079 miss->flow.vlan_tci)) {
3080 /* This packet was received on a VLAN splinter port. We
3081 * added a VLAN to the packet to make the packet resemble
3082 * the flow, but the actions were composed assuming that
3083 * the packet contained no VLAN. So, we must remove the
3084 * VLAN header from the packet before trying to execute the
3086 eth_pop_vlan(packet);
3089 op->subfacet = NULL;
3090 op->xout_garbage = false;
3091 op->dpif_op.type = DPIF_OP_EXECUTE;
3092 op->dpif_op.u.execute.key = miss->key;
3093 op->dpif_op.u.execute.key_len = miss->key_len;
3094 op->dpif_op.u.execute.packet = packet;
3095 ofpbuf_use_stack(&op->mask, &op->maskbuf, sizeof op->maskbuf);
3098 /* Helper for handle_flow_miss_without_facet() and
3099 * handle_flow_miss_with_facet(). */
3101 handle_flow_miss_common(struct rule_dpif *rule,
3102 struct ofpbuf *packet, const struct flow *flow)
3104 struct ofproto_dpif *ofproto = ofproto_dpif_cast(rule->up.ofproto);
3106 if (rule->up.cr.priority == FAIL_OPEN_PRIORITY) {
3108 * Extra-special case for fail-open mode.
3110 * We are in fail-open mode and the packet matched the fail-open
3111 * rule, but we are connected to a controller too. We should send
3112 * the packet up to the controller in the hope that it will try to
3113 * set up a flow and thereby allow us to exit fail-open.
3115 * See the top-level comment in fail-open.c for more information.
3117 send_packet_in_miss(ofproto, packet, flow);
3121 /* Figures out whether a flow that missed in 'ofproto', whose details are in
3122 * 'miss' masked by 'wc', is likely to be worth tracking in detail in userspace
3123 * and (usually) installing a datapath flow. The answer is usually "yes" (a
3124 * return value of true). However, for short flows the cost of bookkeeping is
3125 * much higher than the benefits, so when the datapath holds a large number of
3126 * flows we impose some heuristics to decide which flows are likely to be worth
3129 flow_miss_should_make_facet(struct flow_miss *miss, struct flow_wildcards *wc)
3131 struct dpif_backer *backer = miss->ofproto->backer;
3134 switch (flow_miss_model) {
3135 case OFPROTO_HANDLE_MISS_AUTO:
3137 case OFPROTO_HANDLE_MISS_WITH_FACETS:
3139 case OFPROTO_HANDLE_MISS_WITHOUT_FACETS:
3143 if (!backer->governor) {
3146 n_subfacets = hmap_count(&backer->subfacets);
3147 if (n_subfacets * 2 <= flow_eviction_threshold) {
3151 backer->governor = governor_create();
3154 hash = flow_hash_in_wildcards(&miss->flow, wc, 0);
3155 return governor_should_install_flow(backer->governor, hash,
3156 list_size(&miss->packets));
3159 /* Handles 'miss' without creating a facet or subfacet or creating any datapath
3160 * flow. 'miss->flow' must have matched 'rule' and been xlated into 'xout'.
3161 * May add an "execute" operation to 'ops' and increment '*n_ops'. */
3163 handle_flow_miss_without_facet(struct rule_dpif *rule, struct xlate_out *xout,
3164 struct flow_miss *miss,
3165 struct flow_miss_op *ops, size_t *n_ops)
3167 struct ofpbuf *packet;
3169 LIST_FOR_EACH (packet, list_node, &miss->packets) {
3171 COVERAGE_INC(facet_suppress);
3173 handle_flow_miss_common(rule, packet, &miss->flow);
3176 struct xlate_in xin;
3178 xlate_in_init(&xin, miss->ofproto, &miss->flow, rule, 0, packet);
3179 xlate_actions_for_side_effects(&xin);
3182 if (xout->odp_actions.size) {
3183 struct flow_miss_op *op = &ops[*n_ops];
3184 struct dpif_execute *execute = &op->dpif_op.u.execute;
3186 init_flow_miss_execute_op(miss, packet, op);
3187 xlate_out_copy(&op->xout, xout);
3188 execute->actions = op->xout.odp_actions.data;
3189 execute->actions_len = op->xout.odp_actions.size;
3190 op->xout_garbage = true;
3197 /* Handles 'miss', which matches 'facet'. May add any required datapath
3198 * operations to 'ops', incrementing '*n_ops' for each new op.
3200 * All of the packets in 'miss' are considered to have arrived at time 'now'.
3201 * This is really important only for new facets: if we just called time_msec()
3202 * here, then the new subfacet or its packets could look (occasionally) as
3203 * though it was used some time after the facet was used. That can make a
3204 * one-packet flow look like it has a nonzero duration, which looks odd in
3205 * e.g. NetFlow statistics.
3207 * If non-null, 'stats' will be folded into 'facet'. */
3209 handle_flow_miss_with_facet(struct flow_miss *miss, struct facet *facet,
3210 long long int now, struct dpif_flow_stats *stats,
3211 struct flow_miss_op *ops, size_t *n_ops)
3213 struct ofproto_dpif *ofproto = ofproto_dpif_cast(facet->rule->up.ofproto);
3214 enum subfacet_path want_path;
3215 struct subfacet *subfacet;
3216 struct ofpbuf *packet;
3218 subfacet = subfacet_create(facet, miss, now);
3219 want_path = facet->xout.slow ? SF_SLOW_PATH : SF_FAST_PATH;
3221 subfacet_update_stats(subfacet, stats);
3224 LIST_FOR_EACH (packet, list_node, &miss->packets) {
3225 struct flow_miss_op *op = &ops[*n_ops];
3227 handle_flow_miss_common(facet->rule, packet, &miss->flow);
3229 if (want_path != SF_FAST_PATH) {
3230 struct xlate_in xin;
3232 xlate_in_init(&xin, ofproto, &miss->flow, facet->rule, 0, packet);
3233 xlate_actions_for_side_effects(&xin);
3236 if (facet->xout.odp_actions.size) {
3237 struct dpif_execute *execute = &op->dpif_op.u.execute;
3239 init_flow_miss_execute_op(miss, packet, op);
3240 execute->actions = facet->xout.odp_actions.data,
3241 execute->actions_len = facet->xout.odp_actions.size;
3246 if (miss->upcall_type == DPIF_UC_MISS || subfacet->path != want_path) {
3247 struct flow_miss_op *op = &ops[(*n_ops)++];
3248 struct dpif_flow_put *put = &op->dpif_op.u.flow_put;
3250 subfacet->path = want_path;
3252 ofpbuf_use_stack(&op->mask, &op->maskbuf, sizeof op->maskbuf);
3253 if (enable_megaflows) {
3254 odp_flow_key_from_mask(&op->mask, &facet->xout.wc.masks,
3255 &miss->flow, UINT32_MAX);
3258 op->xout_garbage = false;
3259 op->dpif_op.type = DPIF_OP_FLOW_PUT;
3260 op->subfacet = subfacet;
3261 put->flags = DPIF_FP_CREATE | DPIF_FP_MODIFY;
3262 put->key = miss->key;
3263 put->key_len = miss->key_len;
3264 put->mask = op->mask.data;
3265 put->mask_len = op->mask.size;
3267 if (want_path == SF_FAST_PATH) {
3268 put->actions = facet->xout.odp_actions.data;
3269 put->actions_len = facet->xout.odp_actions.size;
3271 compose_slow_path(ofproto, &miss->flow, facet->xout.slow,
3272 op->slow_stub, sizeof op->slow_stub,
3273 &put->actions, &put->actions_len);
3279 /* Handles flow miss 'miss'. May add any required datapath operations
3280 * to 'ops', incrementing '*n_ops' for each new op. */
3282 handle_flow_miss(struct flow_miss *miss, struct flow_miss_op *ops,
3285 struct ofproto_dpif *ofproto = miss->ofproto;
3286 struct dpif_flow_stats stats__;
3287 struct dpif_flow_stats *stats = &stats__;
3288 struct ofpbuf *packet;
3289 struct facet *facet;
3293 memset(stats, 0, sizeof *stats);
3295 LIST_FOR_EACH (packet, list_node, &miss->packets) {
3296 stats->tcp_flags |= packet_get_tcp_flags(packet, &miss->flow);
3297 stats->n_bytes += packet->size;
3301 facet = facet_lookup_valid(ofproto, &miss->flow);
3303 struct flow_wildcards wc;
3304 struct rule_dpif *rule;
3305 struct xlate_out xout;
3306 struct xlate_in xin;
3308 flow_wildcards_init_catchall(&wc);
3309 rule = rule_dpif_lookup(ofproto, &miss->flow, &wc);
3310 rule_credit_stats(rule, stats);
3312 xlate_in_init(&xin, ofproto, &miss->flow, rule, stats->tcp_flags,
3314 xin.resubmit_stats = stats;
3315 xin.may_learn = true;
3316 xlate_actions(&xin, &xout);
3317 flow_wildcards_or(&xout.wc, &xout.wc, &wc);
3319 /* There does not exist a bijection between 'struct flow' and datapath
3320 * flow keys with fitness ODP_FIT_TO_LITTLE. This breaks a fundamental
3321 * assumption used throughout the facet and subfacet handling code.
3322 * Since we have to handle these misses in userspace anyway, we simply
3323 * skip facet creation, avoiding the problem altogether. */
3324 if (miss->key_fitness == ODP_FIT_TOO_LITTLE
3325 || !flow_miss_should_make_facet(miss, &xout.wc)) {
3326 handle_flow_miss_without_facet(rule, &xout, miss, ops, n_ops);
3330 facet = facet_create(miss, rule, &xout, stats);
3333 handle_flow_miss_with_facet(miss, facet, now, stats, ops, n_ops);
3336 static struct drop_key *
3337 drop_key_lookup(const struct dpif_backer *backer, const struct nlattr *key,
3340 struct drop_key *drop_key;
3342 HMAP_FOR_EACH_WITH_HASH (drop_key, hmap_node, hash_bytes(key, key_len, 0),
3343 &backer->drop_keys) {
3344 if (drop_key->key_len == key_len
3345 && !memcmp(drop_key->key, key, key_len)) {
3353 drop_key_clear(struct dpif_backer *backer)
3355 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 15);
3356 struct drop_key *drop_key, *next;
3358 HMAP_FOR_EACH_SAFE (drop_key, next, hmap_node, &backer->drop_keys) {
3361 error = dpif_flow_del(backer->dpif, drop_key->key, drop_key->key_len,
3363 if (error && !VLOG_DROP_WARN(&rl)) {
3364 struct ds ds = DS_EMPTY_INITIALIZER;
3365 odp_flow_key_format(drop_key->key, drop_key->key_len, &ds);
3366 VLOG_WARN("Failed to delete drop key (%s) (%s)",
3367 ovs_strerror(error), ds_cstr(&ds));
3371 hmap_remove(&backer->drop_keys, &drop_key->hmap_node);
3372 free(drop_key->key);
3377 /* Given a datpath, packet, and flow metadata ('backer', 'packet', and 'key'
3378 * respectively), populates 'flow' with the result of odp_flow_key_to_flow().
3379 * Optionally, if nonnull, populates 'fitnessp' with the fitness of 'flow' as
3380 * returned by odp_flow_key_to_flow(). Also, optionally populates 'ofproto'
3381 * with the ofproto_dpif, and 'odp_in_port' with the datapath in_port, that
3382 * 'packet' ingressed.
3384 * If 'ofproto' is nonnull, requires 'flow''s in_port to exist. Otherwise sets
3385 * 'flow''s in_port to OFPP_NONE.
3387 * This function does post-processing on data returned from
3388 * odp_flow_key_to_flow() to help make VLAN splinters transparent to the rest
3389 * of the upcall processing logic. In particular, if the extracted in_port is
3390 * a VLAN splinter port, it replaces flow->in_port by the "real" port, sets
3391 * flow->vlan_tci correctly for the VLAN of the VLAN splinter port, and pushes
3392 * a VLAN header onto 'packet' (if it is nonnull).
3394 * Similarly, this function also includes some logic to help with tunnels. It
3395 * may modify 'flow' as necessary to make the tunneling implementation
3396 * transparent to the upcall processing logic.
3398 * Returns 0 if successful, ENODEV if the parsed flow has no associated ofport,
3399 * or some other positive errno if there are other problems. */
3401 ofproto_receive(const struct dpif_backer *backer, struct ofpbuf *packet,
3402 const struct nlattr *key, size_t key_len,
3403 struct flow *flow, enum odp_key_fitness *fitnessp,
3404 struct ofproto_dpif **ofproto, odp_port_t *odp_in_port)
3406 const struct ofport_dpif *port;
3407 enum odp_key_fitness fitness;
3410 fitness = odp_flow_key_to_flow(key, key_len, flow);
3411 if (fitness == ODP_FIT_ERROR) {
3417 *odp_in_port = flow->in_port.odp_port;
3420 port = (tnl_port_should_receive(flow)
3421 ? tnl_port_receive(flow)
3422 : odp_port_to_ofport(backer, flow->in_port.odp_port));
3423 flow->in_port.ofp_port = port ? port->up.ofp_port : OFPP_NONE;
3428 /* XXX: Since the tunnel module is not scoped per backer, for a tunnel port
3429 * it's theoretically possible that we'll receive an ofport belonging to an
3430 * entirely different datapath. In practice, this can't happen because no
3431 * platforms has two separate datapaths which each support tunneling. */
3432 ovs_assert(ofproto_dpif_cast(port->up.ofproto)->backer == backer);
3434 if (vsp_adjust_flow(ofproto_dpif_cast(port->up.ofproto), flow)) {
3436 /* Make the packet resemble the flow, so that it gets sent to
3437 * an OpenFlow controller properly, so that it looks correct
3438 * for sFlow, and so that flow_extract() will get the correct
3439 * vlan_tci if it is called on 'packet'.
3441 * The allocated space inside 'packet' probably also contains
3442 * 'key', that is, both 'packet' and 'key' are probably part of
3443 * a struct dpif_upcall (see the large comment on that
3444 * structure definition), so pushing data on 'packet' is in
3445 * general not a good idea since it could overwrite 'key' or
3446 * free it as a side effect. However, it's OK in this special
3447 * case because we know that 'packet' is inside a Netlink
3448 * attribute: pushing 4 bytes will just overwrite the 4-byte
3449 * "struct nlattr", which is fine since we don't need that
3450 * header anymore. */
3451 eth_push_vlan(packet, flow->vlan_tci);
3453 /* We can't reproduce 'key' from 'flow'. */
3454 fitness = fitness == ODP_FIT_PERFECT ? ODP_FIT_TOO_MUCH : fitness;
3459 *ofproto = ofproto_dpif_cast(port->up.ofproto);
3464 *fitnessp = fitness;
3470 handle_miss_upcalls(struct dpif_backer *backer, struct dpif_upcall *upcalls,
3473 struct dpif_upcall *upcall;
3474 struct flow_miss *miss;
3475 struct flow_miss misses[FLOW_MISS_MAX_BATCH];
3476 struct flow_miss_op flow_miss_ops[FLOW_MISS_MAX_BATCH * 2];
3477 struct dpif_op *dpif_ops[FLOW_MISS_MAX_BATCH * 2];
3487 /* Construct the to-do list.
3489 * This just amounts to extracting the flow from each packet and sticking
3490 * the packets that have the same flow in the same "flow_miss" structure so
3491 * that we can process them together. */
3494 for (upcall = upcalls; upcall < &upcalls[n_upcalls]; upcall++) {
3495 struct flow_miss *miss = &misses[n_misses];
3496 struct flow_miss *existing_miss;
3497 struct ofproto_dpif *ofproto;
3498 odp_port_t odp_in_port;
3503 error = ofproto_receive(backer, upcall->packet, upcall->key,
3504 upcall->key_len, &flow, &miss->key_fitness,
3505 &ofproto, &odp_in_port);
3506 if (error == ENODEV) {
3507 struct drop_key *drop_key;
3509 /* Received packet on datapath port for which we couldn't
3510 * associate an ofproto. This can happen if a port is removed
3511 * while traffic is being received. Print a rate-limited message
3512 * in case it happens frequently. Install a drop flow so
3513 * that future packets of the flow are inexpensively dropped
3515 VLOG_INFO_RL(&rl, "received packet on unassociated datapath port "
3516 "%"PRIu32, odp_in_port);
3518 drop_key = drop_key_lookup(backer, upcall->key, upcall->key_len);
3520 drop_key = xmalloc(sizeof *drop_key);
3521 drop_key->key = xmemdup(upcall->key, upcall->key_len);
3522 drop_key->key_len = upcall->key_len;
3524 hmap_insert(&backer->drop_keys, &drop_key->hmap_node,
3525 hash_bytes(drop_key->key, drop_key->key_len, 0));
3526 dpif_flow_put(backer->dpif, DPIF_FP_CREATE | DPIF_FP_MODIFY,
3527 drop_key->key, drop_key->key_len,
3528 NULL, 0, NULL, 0, NULL);
3536 ofproto->n_missed++;
3537 flow_extract(upcall->packet, flow.skb_priority, flow.skb_mark,
3538 &flow.tunnel, &flow.in_port, &miss->flow);
3540 /* Add other packets to a to-do list. */
3541 hash = flow_hash(&miss->flow, 0);
3542 existing_miss = flow_miss_find(&todo, ofproto, &miss->flow, hash);
3543 if (!existing_miss) {
3544 hmap_insert(&todo, &miss->hmap_node, hash);
3545 miss->ofproto = ofproto;
3546 miss->key = upcall->key;
3547 miss->key_len = upcall->key_len;
3548 miss->upcall_type = upcall->type;
3549 list_init(&miss->packets);
3553 miss = existing_miss;
3555 list_push_back(&miss->packets, &upcall->packet->list_node);
3558 /* Process each element in the to-do list, constructing the set of
3559 * operations to batch. */
3561 HMAP_FOR_EACH (miss, hmap_node, &todo) {
3562 handle_flow_miss(miss, flow_miss_ops, &n_ops);
3564 ovs_assert(n_ops <= ARRAY_SIZE(flow_miss_ops));
3566 /* Execute batch. */
3567 for (i = 0; i < n_ops; i++) {
3568 dpif_ops[i] = &flow_miss_ops[i].dpif_op;
3570 dpif_operate(backer->dpif, dpif_ops, n_ops);
3572 for (i = 0; i < n_ops; i++) {
3573 if (dpif_ops[i]->error != 0
3574 && flow_miss_ops[i].dpif_op.type == DPIF_OP_FLOW_PUT
3575 && flow_miss_ops[i].subfacet) {
3576 struct subfacet *subfacet = flow_miss_ops[i].subfacet;
3578 COVERAGE_INC(subfacet_install_fail);
3580 subfacet->path = SF_NOT_INSTALLED;
3584 if (flow_miss_ops[i].xout_garbage) {
3585 xlate_out_uninit(&flow_miss_ops[i].xout);
3588 hmap_destroy(&todo);
3591 static enum { SFLOW_UPCALL, MISS_UPCALL, BAD_UPCALL, FLOW_SAMPLE_UPCALL,
3593 classify_upcall(const struct dpif_upcall *upcall)
3595 size_t userdata_len;
3596 union user_action_cookie cookie;
3598 /* First look at the upcall type. */
3599 switch (upcall->type) {
3600 case DPIF_UC_ACTION:
3606 case DPIF_N_UC_TYPES:
3608 VLOG_WARN_RL(&rl, "upcall has unexpected type %"PRIu32, upcall->type);
3612 /* "action" upcalls need a closer look. */
3613 if (!upcall->userdata) {
3614 VLOG_WARN_RL(&rl, "action upcall missing cookie");
3617 userdata_len = nl_attr_get_size(upcall->userdata);
3618 if (userdata_len < sizeof cookie.type
3619 || userdata_len > sizeof cookie) {
3620 VLOG_WARN_RL(&rl, "action upcall cookie has unexpected size %zu",
3624 memset(&cookie, 0, sizeof cookie);
3625 memcpy(&cookie, nl_attr_get(upcall->userdata), userdata_len);
3626 if (userdata_len == sizeof cookie.sflow
3627 && cookie.type == USER_ACTION_COOKIE_SFLOW) {
3628 return SFLOW_UPCALL;
3629 } else if (userdata_len == sizeof cookie.slow_path
3630 && cookie.type == USER_ACTION_COOKIE_SLOW_PATH) {
3632 } else if (userdata_len == sizeof cookie.flow_sample
3633 && cookie.type == USER_ACTION_COOKIE_FLOW_SAMPLE) {
3634 return FLOW_SAMPLE_UPCALL;
3635 } else if (userdata_len == sizeof cookie.ipfix
3636 && cookie.type == USER_ACTION_COOKIE_IPFIX) {
3637 return IPFIX_UPCALL;
3639 VLOG_WARN_RL(&rl, "invalid user cookie of type %"PRIu16
3640 " and size %zu", cookie.type, userdata_len);
3646 handle_sflow_upcall(struct dpif_backer *backer,
3647 const struct dpif_upcall *upcall)
3649 struct ofproto_dpif *ofproto;
3650 union user_action_cookie cookie;
3652 odp_port_t odp_in_port;
3654 if (ofproto_receive(backer, upcall->packet, upcall->key, upcall->key_len,
3655 &flow, NULL, &ofproto, &odp_in_port)
3656 || !ofproto->sflow) {
3660 memset(&cookie, 0, sizeof cookie);
3661 memcpy(&cookie, nl_attr_get(upcall->userdata), sizeof cookie.sflow);
3662 dpif_sflow_received(ofproto->sflow, upcall->packet, &flow,
3663 odp_in_port, &cookie);
3667 handle_flow_sample_upcall(struct dpif_backer *backer,
3668 const struct dpif_upcall *upcall)
3670 struct ofproto_dpif *ofproto;
3671 union user_action_cookie cookie;
3674 if (ofproto_receive(backer, upcall->packet, upcall->key, upcall->key_len,
3675 &flow, NULL, &ofproto, NULL)
3676 || !ofproto->ipfix) {
3680 memset(&cookie, 0, sizeof cookie);
3681 memcpy(&cookie, nl_attr_get(upcall->userdata), sizeof cookie.flow_sample);
3683 /* The flow reflects exactly the contents of the packet. Sample
3684 * the packet using it. */
3685 dpif_ipfix_flow_sample(ofproto->ipfix, upcall->packet, &flow,
3686 cookie.flow_sample.collector_set_id,
3687 cookie.flow_sample.probability,
3688 cookie.flow_sample.obs_domain_id,
3689 cookie.flow_sample.obs_point_id);
3693 handle_ipfix_upcall(struct dpif_backer *backer,
3694 const struct dpif_upcall *upcall)
3696 struct ofproto_dpif *ofproto;
3699 if (ofproto_receive(backer, upcall->packet, upcall->key, upcall->key_len,
3700 &flow, NULL, &ofproto, NULL)
3701 || !ofproto->ipfix) {
3705 /* The flow reflects exactly the contents of the packet. Sample
3706 * the packet using it. */
3707 dpif_ipfix_bridge_sample(ofproto->ipfix, upcall->packet, &flow);
3711 handle_upcalls(struct dpif_backer *backer, unsigned int max_batch)
3713 struct dpif_upcall misses[FLOW_MISS_MAX_BATCH];
3714 struct ofpbuf miss_bufs[FLOW_MISS_MAX_BATCH];
3715 uint64_t miss_buf_stubs[FLOW_MISS_MAX_BATCH][4096 / 8];
3720 ovs_assert(max_batch <= FLOW_MISS_MAX_BATCH);
3723 for (n_processed = 0; n_processed < max_batch; n_processed++) {
3724 struct dpif_upcall *upcall = &misses[n_misses];
3725 struct ofpbuf *buf = &miss_bufs[n_misses];
3728 ofpbuf_use_stub(buf, miss_buf_stubs[n_misses],
3729 sizeof miss_buf_stubs[n_misses]);
3730 error = dpif_recv(backer->dpif, upcall, buf);
3736 switch (classify_upcall(upcall)) {
3738 /* Handle it later. */
3743 handle_sflow_upcall(backer, upcall);
3747 case FLOW_SAMPLE_UPCALL:
3748 handle_flow_sample_upcall(backer, upcall);
3753 handle_ipfix_upcall(backer, upcall);
3763 /* Handle deferred MISS_UPCALL processing. */
3764 handle_miss_upcalls(backer, misses, n_misses);
3765 for (i = 0; i < n_misses; i++) {
3766 ofpbuf_uninit(&miss_bufs[i]);
3772 /* Flow expiration. */
3774 static int subfacet_max_idle(const struct dpif_backer *);
3775 static void update_stats(struct dpif_backer *);
3776 static void rule_expire(struct rule_dpif *);
3777 static void expire_subfacets(struct dpif_backer *, int dp_max_idle);
3779 /* This function is called periodically by run(). Its job is to collect
3780 * updates for the flows that have been installed into the datapath, most
3781 * importantly when they last were used, and then use that information to
3782 * expire flows that have not been used recently.
3784 * Returns the number of milliseconds after which it should be called again. */
3786 expire(struct dpif_backer *backer)
3788 struct ofproto_dpif *ofproto;
3792 /* Periodically clear out the drop keys in an effort to keep them
3793 * relatively few. */
3794 drop_key_clear(backer);
3796 /* Update stats for each flow in the backer. */
3797 update_stats(backer);
3799 n_subfacets = hmap_count(&backer->subfacets);
3801 struct subfacet *subfacet;
3802 long long int total, now;
3806 HMAP_FOR_EACH (subfacet, hmap_node, &backer->subfacets) {
3807 total += now - subfacet->created;
3809 backer->avg_subfacet_life += total / n_subfacets;
3811 backer->avg_subfacet_life /= 2;
3813 backer->avg_n_subfacet += n_subfacets;
3814 backer->avg_n_subfacet /= 2;
3816 backer->max_n_subfacet = MAX(backer->max_n_subfacet, n_subfacets);
3818 max_idle = subfacet_max_idle(backer);
3819 expire_subfacets(backer, max_idle);
3821 HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
3822 struct rule *rule, *next_rule;
3824 if (ofproto->backer != backer) {
3828 /* Expire OpenFlow flows whose idle_timeout or hard_timeout
3830 LIST_FOR_EACH_SAFE (rule, next_rule, expirable,
3831 &ofproto->up.expirable) {
3832 rule_expire(rule_dpif_cast(rule));
3835 /* All outstanding data in existing flows has been accounted, so it's a
3836 * good time to do bond rebalancing. */
3837 if (ofproto->has_bonded_bundles) {
3838 struct ofbundle *bundle;
3840 HMAP_FOR_EACH (bundle, hmap_node, &ofproto->bundles) {
3842 bond_rebalance(bundle->bond, &backer->revalidate_set);
3848 return MIN(max_idle, 1000);
3851 /* Updates flow table statistics given that the datapath just reported 'stats'
3852 * as 'subfacet''s statistics. */
3854 update_subfacet_stats(struct subfacet *subfacet,
3855 const struct dpif_flow_stats *stats)
3857 struct facet *facet = subfacet->facet;
3858 struct ofproto_dpif *ofproto = ofproto_dpif_cast(facet->rule->up.ofproto);
3859 struct dpif_flow_stats diff;
3861 diff.tcp_flags = stats->tcp_flags;
3862 diff.used = stats->used;
3864 if (stats->n_packets >= subfacet->dp_packet_count) {
3865 diff.n_packets = stats->n_packets - subfacet->dp_packet_count;
3867 VLOG_WARN_RL(&rl, "unexpected packet count from the datapath");
3871 if (stats->n_bytes >= subfacet->dp_byte_count) {
3872 diff.n_bytes = stats->n_bytes - subfacet->dp_byte_count;
3874 VLOG_WARN_RL(&rl, "unexpected byte count from datapath");
3878 ofproto->n_hit += diff.n_packets;
3879 subfacet->dp_packet_count = stats->n_packets;
3880 subfacet->dp_byte_count = stats->n_bytes;
3881 subfacet_update_stats(subfacet, &diff);
3883 if (facet->accounted_bytes < facet->byte_count) {
3885 facet_account(facet);
3886 facet->accounted_bytes = facet->byte_count;
3890 /* 'key' with length 'key_len' bytes is a flow in 'dpif' that we know nothing
3891 * about, or a flow that shouldn't be installed but was anyway. Delete it. */
3893 delete_unexpected_flow(struct dpif_backer *backer,
3894 const struct nlattr *key, size_t key_len)
3896 if (!VLOG_DROP_WARN(&rl)) {
3900 odp_flow_key_format(key, key_len, &s);
3901 VLOG_WARN("unexpected flow: %s", ds_cstr(&s));
3905 COVERAGE_INC(facet_unexpected);
3906 dpif_flow_del(backer->dpif, key, key_len, NULL);
3909 /* Update 'packet_count', 'byte_count', and 'used' members of installed facets.
3911 * This function also pushes statistics updates to rules which each facet
3912 * resubmits into. Generally these statistics will be accurate. However, if a
3913 * facet changes the rule it resubmits into at some time in between
3914 * update_stats() runs, it is possible that statistics accrued to the
3915 * old rule will be incorrectly attributed to the new rule. This could be
3916 * avoided by calling update_stats() whenever rules are created or
3917 * deleted. However, the performance impact of making so many calls to the
3918 * datapath do not justify the benefit of having perfectly accurate statistics.
3920 * In addition, this function maintains per ofproto flow hit counts. The patch
3921 * port is not treated specially. e.g. A packet ingress from br0 patched into
3922 * br1 will increase the hit count of br0 by 1, however, does not affect
3923 * the hit or miss counts of br1.
3926 update_stats(struct dpif_backer *backer)
3928 const struct dpif_flow_stats *stats;
3929 struct dpif_flow_dump dump;
3930 const struct nlattr *key, *mask;
3931 size_t key_len, mask_len;
3933 dpif_flow_dump_start(&dump, backer->dpif);
3934 while (dpif_flow_dump_next(&dump, &key, &key_len,
3935 &mask, &mask_len, NULL, NULL, &stats)) {
3936 struct subfacet *subfacet;
3939 key_hash = odp_flow_key_hash(key, key_len);
3940 subfacet = subfacet_find(backer, key, key_len, key_hash);
3941 switch (subfacet ? subfacet->path : SF_NOT_INSTALLED) {
3943 update_subfacet_stats(subfacet, stats);
3947 /* Stats are updated per-packet. */
3950 case SF_NOT_INSTALLED:
3952 delete_unexpected_flow(backer, key, key_len);
3957 dpif_flow_dump_done(&dump);
3959 update_moving_averages(backer);
3962 /* Calculates and returns the number of milliseconds of idle time after which
3963 * subfacets should expire from the datapath. When a subfacet expires, we fold
3964 * its statistics into its facet, and when a facet's last subfacet expires, we
3965 * fold its statistic into its rule. */
3967 subfacet_max_idle(const struct dpif_backer *backer)
3970 * Idle time histogram.
3972 * Most of the time a switch has a relatively small number of subfacets.
3973 * When this is the case we might as well keep statistics for all of them
3974 * in userspace and to cache them in the kernel datapath for performance as
3977 * As the number of subfacets increases, the memory required to maintain
3978 * statistics about them in userspace and in the kernel becomes
3979 * significant. However, with a large number of subfacets it is likely
3980 * that only a few of them are "heavy hitters" that consume a large amount
3981 * of bandwidth. At this point, only heavy hitters are worth caching in
3982 * the kernel and maintaining in userspaces; other subfacets we can
3985 * The technique used to compute the idle time is to build a histogram with
3986 * N_BUCKETS buckets whose width is BUCKET_WIDTH msecs each. Each subfacet
3987 * that is installed in the kernel gets dropped in the appropriate bucket.
3988 * After the histogram has been built, we compute the cutoff so that only
3989 * the most-recently-used 1% of subfacets (but at least
3990 * flow_eviction_threshold flows) are kept cached. At least
3991 * the most-recently-used bucket of subfacets is kept, so actually an
3992 * arbitrary number of subfacets can be kept in any given expiration run
3993 * (though the next run will delete most of those unless they receive
3996 * This requires a second pass through the subfacets, in addition to the
3997 * pass made by update_stats(), because the former function never looks at
3998 * uninstallable subfacets.
4000 enum { BUCKET_WIDTH = ROUND_UP(100, TIME_UPDATE_INTERVAL) };
4001 enum { N_BUCKETS = 5000 / BUCKET_WIDTH };
4002 int buckets[N_BUCKETS] = { 0 };
4003 int total, subtotal, bucket;
4004 struct subfacet *subfacet;
4008 total = hmap_count(&backer->subfacets);
4009 if (total <= flow_eviction_threshold) {
4010 return N_BUCKETS * BUCKET_WIDTH;
4013 /* Build histogram. */
4015 HMAP_FOR_EACH (subfacet, hmap_node, &backer->subfacets) {
4016 long long int idle = now - subfacet->used;
4017 int bucket = (idle <= 0 ? 0
4018 : idle >= BUCKET_WIDTH * N_BUCKETS ? N_BUCKETS - 1
4019 : (unsigned int) idle / BUCKET_WIDTH);
4023 /* Find the first bucket whose flows should be expired. */
4024 subtotal = bucket = 0;
4026 subtotal += buckets[bucket++];
4027 } while (bucket < N_BUCKETS &&
4028 subtotal < MAX(flow_eviction_threshold, total / 100));
4030 if (VLOG_IS_DBG_ENABLED()) {
4034 ds_put_cstr(&s, "keep");
4035 for (i = 0; i < N_BUCKETS; i++) {
4037 ds_put_cstr(&s, ", drop");
4040 ds_put_format(&s, " %d:%d", i * BUCKET_WIDTH, buckets[i]);
4043 VLOG_INFO("%s (msec:count)", ds_cstr(&s));
4047 return bucket * BUCKET_WIDTH;
4051 expire_subfacets(struct dpif_backer *backer, int dp_max_idle)
4053 /* Cutoff time for most flows. */
4054 long long int normal_cutoff = time_msec() - dp_max_idle;
4056 /* We really want to keep flows for special protocols around, so use a more
4057 * conservative cutoff. */
4058 long long int special_cutoff = time_msec() - 10000;
4060 struct subfacet *subfacet, *next_subfacet;
4061 struct subfacet *batch[SUBFACET_DESTROY_MAX_BATCH];
4065 HMAP_FOR_EACH_SAFE (subfacet, next_subfacet, hmap_node,
4066 &backer->subfacets) {
4067 long long int cutoff;
4069 cutoff = (subfacet->facet->xout.slow & (SLOW_CFM | SLOW_BFD | SLOW_LACP
4073 if (subfacet->used < cutoff) {
4074 if (subfacet->path != SF_NOT_INSTALLED) {
4075 batch[n_batch++] = subfacet;
4076 if (n_batch >= SUBFACET_DESTROY_MAX_BATCH) {
4077 subfacet_destroy_batch(backer, batch, n_batch);
4081 subfacet_destroy(subfacet);
4087 subfacet_destroy_batch(backer, batch, n_batch);
4091 /* If 'rule' is an OpenFlow rule, that has expired according to OpenFlow rules,
4092 * then delete it entirely. */
4094 rule_expire(struct rule_dpif *rule)
4096 struct facet *facet, *next_facet;
4100 if (rule->up.pending) {
4101 /* We'll have to expire it later. */
4105 /* Has 'rule' expired? */
4107 if (rule->up.hard_timeout
4108 && now > rule->up.modified + rule->up.hard_timeout * 1000) {
4109 reason = OFPRR_HARD_TIMEOUT;
4110 } else if (rule->up.idle_timeout
4111 && now > rule->up.used + rule->up.idle_timeout * 1000) {
4112 reason = OFPRR_IDLE_TIMEOUT;
4117 COVERAGE_INC(ofproto_dpif_expired);
4119 /* Update stats. (This is a no-op if the rule expired due to an idle
4120 * timeout, because that only happens when the rule has no facets left.) */
4121 LIST_FOR_EACH_SAFE (facet, next_facet, list_node, &rule->facets) {
4122 facet_remove(facet);
4125 /* Get rid of the rule. */
4126 ofproto_rule_expire(&rule->up, reason);
4131 /* Creates and returns a new facet based on 'miss'.
4133 * The caller must already have determined that no facet with an identical
4134 * 'miss->flow' exists in 'miss->ofproto'.
4136 * 'rule' and 'xout' must have been created based on 'miss'.
4138 * 'facet'' statistics are initialized based on 'stats'.
4140 * The facet will initially have no subfacets. The caller should create (at
4141 * least) one subfacet with subfacet_create(). */
4142 static struct facet *
4143 facet_create(const struct flow_miss *miss, struct rule_dpif *rule,
4144 struct xlate_out *xout, struct dpif_flow_stats *stats)
4146 struct ofproto_dpif *ofproto = miss->ofproto;
4147 struct facet *facet;
4150 facet = xzalloc(sizeof *facet);
4151 facet->packet_count = facet->prev_packet_count = stats->n_packets;
4152 facet->byte_count = facet->prev_byte_count = stats->n_bytes;
4153 facet->tcp_flags = stats->tcp_flags;
4154 facet->used = stats->used;
4155 facet->flow = miss->flow;
4156 facet->learn_rl = time_msec() + 500;
4159 list_push_back(&facet->rule->facets, &facet->list_node);
4160 list_init(&facet->subfacets);
4161 netflow_flow_init(&facet->nf_flow);
4162 netflow_flow_update_time(ofproto->netflow, &facet->nf_flow, facet->used);
4164 xlate_out_copy(&facet->xout, xout);
4166 match_init(&match, &facet->flow, &facet->xout.wc);
4167 cls_rule_init(&facet->cr, &match, OFP_DEFAULT_PRIORITY);
4168 classifier_insert(&ofproto->facets, &facet->cr);
4170 facet->nf_flow.output_iface = facet->xout.nf_output_iface;
4176 facet_free(struct facet *facet)
4179 xlate_out_uninit(&facet->xout);
4184 /* Executes, within 'ofproto', the 'n_actions' actions in 'actions' on
4185 * 'packet', which arrived on 'in_port'. */
4187 execute_odp_actions(struct ofproto_dpif *ofproto, const struct flow *flow,
4188 const struct nlattr *odp_actions, size_t actions_len,
4189 struct ofpbuf *packet)
4191 struct odputil_keybuf keybuf;
4195 ofpbuf_use_stack(&key, &keybuf, sizeof keybuf);
4196 odp_flow_key_from_flow(&key, flow,
4197 ofp_port_to_odp_port(ofproto, flow->in_port.ofp_port));
4199 error = dpif_execute(ofproto->backer->dpif, key.data, key.size,
4200 odp_actions, actions_len, packet);
4204 /* Remove 'facet' from its ofproto and free up the associated memory:
4206 * - If 'facet' was installed in the datapath, uninstalls it and updates its
4207 * rule's statistics, via subfacet_uninstall().
4209 * - Removes 'facet' from its rule and from ofproto->facets.
4212 facet_remove(struct facet *facet)
4214 struct ofproto_dpif *ofproto = ofproto_dpif_cast(facet->rule->up.ofproto);
4215 struct subfacet *subfacet, *next_subfacet;
4217 ovs_assert(!list_is_empty(&facet->subfacets));
4219 /* First uninstall all of the subfacets to get final statistics. */
4220 LIST_FOR_EACH (subfacet, list_node, &facet->subfacets) {
4221 subfacet_uninstall(subfacet);
4224 /* Flush the final stats to the rule.
4226 * This might require us to have at least one subfacet around so that we
4227 * can use its actions for accounting in facet_account(), which is why we
4228 * have uninstalled but not yet destroyed the subfacets. */
4229 facet_flush_stats(facet);
4231 /* Now we're really all done so destroy everything. */
4232 LIST_FOR_EACH_SAFE (subfacet, next_subfacet, list_node,
4233 &facet->subfacets) {
4234 subfacet_destroy__(subfacet);
4236 classifier_remove(&ofproto->facets, &facet->cr);
4237 cls_rule_destroy(&facet->cr);
4238 list_remove(&facet->list_node);
4242 /* Feed information from 'facet' back into the learning table to keep it in
4243 * sync with what is actually flowing through the datapath. */
4245 facet_learn(struct facet *facet)
4247 long long int now = time_msec();
4249 if (!facet->xout.has_fin_timeout && now < facet->learn_rl) {
4253 facet->learn_rl = now + 500;
4255 if (!facet->xout.has_learn
4256 && !facet->xout.has_normal
4257 && (!facet->xout.has_fin_timeout
4258 || !(facet->tcp_flags & (TCP_FIN | TCP_RST)))) {
4262 facet_push_stats(facet, true);
4266 facet_account(struct facet *facet)
4268 struct ofproto_dpif *ofproto = ofproto_dpif_cast(facet->rule->up.ofproto);
4269 const struct nlattr *a;
4274 if (!facet->xout.has_normal || !ofproto->has_bonded_bundles) {
4277 n_bytes = facet->byte_count - facet->accounted_bytes;
4279 /* This loop feeds byte counters to bond_account() for rebalancing to use
4280 * as a basis. We also need to track the actual VLAN on which the packet
4281 * is going to be sent to ensure that it matches the one passed to
4282 * bond_choose_output_slave(). (Otherwise, we will account to the wrong
4285 * We use the actions from an arbitrary subfacet because they should all
4286 * be equally valid for our purpose. */
4287 vlan_tci = facet->flow.vlan_tci;
4288 NL_ATTR_FOR_EACH_UNSAFE (a, left, facet->xout.odp_actions.data,
4289 facet->xout.odp_actions.size) {
4290 const struct ovs_action_push_vlan *vlan;
4291 struct ofport_dpif *port;
4293 switch (nl_attr_type(a)) {
4294 case OVS_ACTION_ATTR_OUTPUT:
4295 port = get_odp_port(ofproto, nl_attr_get_odp_port(a));
4296 if (port && port->bundle && port->bundle->bond) {
4297 bond_account(port->bundle->bond, &facet->flow,
4298 vlan_tci_to_vid(vlan_tci), n_bytes);
4302 case OVS_ACTION_ATTR_POP_VLAN:
4303 vlan_tci = htons(0);
4306 case OVS_ACTION_ATTR_PUSH_VLAN:
4307 vlan = nl_attr_get(a);
4308 vlan_tci = vlan->vlan_tci;
4314 /* Returns true if the only action for 'facet' is to send to the controller.
4315 * (We don't report NetFlow expiration messages for such facets because they
4316 * are just part of the control logic for the network, not real traffic). */
4318 facet_is_controller_flow(struct facet *facet)
4321 const struct rule *rule = &facet->rule->up;
4322 const struct ofpact *ofpacts = rule->ofpacts;
4323 size_t ofpacts_len = rule->ofpacts_len;
4325 if (ofpacts_len > 0 &&
4326 ofpacts->type == OFPACT_CONTROLLER &&
4327 ofpact_next(ofpacts) >= ofpact_end(ofpacts, ofpacts_len)) {
4334 /* Folds all of 'facet''s statistics into its rule. Also updates the
4335 * accounting ofhook and emits a NetFlow expiration if appropriate. All of
4336 * 'facet''s statistics in the datapath should have been zeroed and folded into
4337 * its packet and byte counts before this function is called. */
4339 facet_flush_stats(struct facet *facet)
4341 struct ofproto_dpif *ofproto = ofproto_dpif_cast(facet->rule->up.ofproto);
4342 struct subfacet *subfacet;
4344 LIST_FOR_EACH (subfacet, list_node, &facet->subfacets) {
4345 ovs_assert(!subfacet->dp_byte_count);
4346 ovs_assert(!subfacet->dp_packet_count);
4349 facet_push_stats(facet, false);
4350 if (facet->accounted_bytes < facet->byte_count) {
4351 facet_account(facet);
4352 facet->accounted_bytes = facet->byte_count;
4355 if (ofproto->netflow && !facet_is_controller_flow(facet)) {
4356 struct ofexpired expired;
4357 expired.flow = facet->flow;
4358 expired.packet_count = facet->packet_count;
4359 expired.byte_count = facet->byte_count;
4360 expired.used = facet->used;
4361 netflow_expire(ofproto->netflow, &facet->nf_flow, &expired);
4364 /* Reset counters to prevent double counting if 'facet' ever gets
4366 facet_reset_counters(facet);
4368 netflow_flow_clear(&facet->nf_flow);
4369 facet->tcp_flags = 0;
4372 /* Searches 'ofproto''s table of facets for one which would be responsible for
4373 * 'flow'. Returns it if found, otherwise a null pointer.
4375 * The returned facet might need revalidation; use facet_lookup_valid()
4376 * instead if that is important. */
4377 static struct facet *
4378 facet_find(struct ofproto_dpif *ofproto, const struct flow *flow)
4380 struct cls_rule *cr = classifier_lookup(&ofproto->facets, flow, NULL);
4381 return cr ? CONTAINER_OF(cr, struct facet, cr) : NULL;
4384 /* Searches 'ofproto''s table of facets for one capable that covers
4385 * 'flow'. Returns it if found, otherwise a null pointer.
4387 * The returned facet is guaranteed to be valid. */
4388 static struct facet *
4389 facet_lookup_valid(struct ofproto_dpif *ofproto, const struct flow *flow)
4391 struct facet *facet;
4393 facet = facet_find(ofproto, flow);
4395 && (ofproto->backer->need_revalidate
4396 || tag_set_intersects(&ofproto->backer->revalidate_set,
4398 && !facet_revalidate(facet)) {
4406 facet_check_consistency(struct facet *facet)
4408 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 15);
4410 struct ofproto_dpif *ofproto = ofproto_dpif_cast(facet->rule->up.ofproto);
4412 struct xlate_out xout;
4413 struct xlate_in xin;
4415 struct rule_dpif *rule;
4418 /* Check the rule for consistency. */
4419 rule = rule_dpif_lookup(ofproto, &facet->flow, NULL);
4420 if (rule != facet->rule) {
4421 if (!VLOG_DROP_WARN(&rl)) {
4422 struct ds s = DS_EMPTY_INITIALIZER;
4424 flow_format(&s, &facet->flow);
4425 ds_put_format(&s, ": facet associated with wrong rule (was "
4426 "table=%"PRIu8",", facet->rule->up.table_id);
4427 cls_rule_format(&facet->rule->up.cr, &s);
4428 ds_put_format(&s, ") (should have been table=%"PRIu8",",
4430 cls_rule_format(&rule->up.cr, &s);
4431 ds_put_char(&s, ')');
4433 VLOG_WARN("%s", ds_cstr(&s));
4439 /* Check the datapath actions for consistency. */
4440 xlate_in_init(&xin, ofproto, &facet->flow, rule, 0, NULL);
4441 xlate_actions(&xin, &xout);
4443 ok = ofpbuf_equal(&facet->xout.odp_actions, &xout.odp_actions)
4444 && facet->xout.slow == xout.slow;
4445 if (!ok && !VLOG_DROP_WARN(&rl)) {
4446 struct ds s = DS_EMPTY_INITIALIZER;
4448 flow_format(&s, &facet->flow);
4449 ds_put_cstr(&s, ": inconsistency in facet");
4451 if (!ofpbuf_equal(&facet->xout.odp_actions, &xout.odp_actions)) {
4452 ds_put_cstr(&s, " (actions were: ");
4453 format_odp_actions(&s, facet->xout.odp_actions.data,
4454 facet->xout.odp_actions.size);
4455 ds_put_cstr(&s, ") (correct actions: ");
4456 format_odp_actions(&s, xout.odp_actions.data,
4457 xout.odp_actions.size);
4458 ds_put_char(&s, ')');
4461 if (facet->xout.slow != xout.slow) {
4462 ds_put_format(&s, " slow path incorrect. should be %d", xout.slow);
4465 VLOG_WARN("%s", ds_cstr(&s));
4468 xlate_out_uninit(&xout);
4473 /* Re-searches the classifier for 'facet':
4475 * - If the rule found is different from 'facet''s current rule, moves
4476 * 'facet' to the new rule and recompiles its actions.
4478 * - If the rule found is the same as 'facet''s current rule, leaves 'facet'
4479 * where it is and recompiles its actions anyway.
4481 * - If any of 'facet''s subfacets correspond to a new flow according to
4482 * ofproto_receive(), 'facet' is removed.
4484 * Returns true if 'facet' is still valid. False if 'facet' was removed. */
4486 facet_revalidate(struct facet *facet)
4488 struct ofproto_dpif *ofproto = ofproto_dpif_cast(facet->rule->up.ofproto);
4489 struct rule_dpif *new_rule;
4490 struct subfacet *subfacet;
4491 struct flow_wildcards wc;
4492 struct xlate_out xout;
4493 struct xlate_in xin;
4495 COVERAGE_INC(facet_revalidate);
4497 /* Check that child subfacets still correspond to this facet. Tunnel
4498 * configuration changes could cause a subfacet's OpenFlow in_port to
4500 LIST_FOR_EACH (subfacet, list_node, &facet->subfacets) {
4501 struct ofproto_dpif *recv_ofproto;
4502 struct flow recv_flow;
4505 error = ofproto_receive(ofproto->backer, NULL, subfacet->key,
4506 subfacet->key_len, &recv_flow, NULL,
4507 &recv_ofproto, NULL);
4509 || recv_ofproto != ofproto
4510 || facet != facet_find(ofproto, &recv_flow)) {
4511 facet_remove(facet);
4516 flow_wildcards_init_catchall(&wc);
4517 new_rule = rule_dpif_lookup(ofproto, &facet->flow, &wc);
4519 /* Calculate new datapath actions.
4521 * We do not modify any 'facet' state yet, because we might need to, e.g.,
4522 * emit a NetFlow expiration and, if so, we need to have the old state
4523 * around to properly compose it. */
4524 xlate_in_init(&xin, ofproto, &facet->flow, new_rule, 0, NULL);
4525 xlate_actions(&xin, &xout);
4526 flow_wildcards_or(&xout.wc, &xout.wc, &wc);
4528 /* A facet's slow path reason should only change under dramatic
4529 * circumstances. Rather than try to update everything, it's simpler to
4530 * remove the facet and start over.
4532 * More importantly, if a facet's wildcards change, it will be relatively
4533 * difficult to figure out if its subfacets still belong to it, and if not
4534 * which facet they may belong to. Again, to avoid the complexity, we
4535 * simply give up instead. */
4536 if (facet->xout.slow != xout.slow
4537 || memcmp(&facet->xout.wc, &xout.wc, sizeof xout.wc)) {
4538 facet_remove(facet);
4539 xlate_out_uninit(&xout);
4543 if (!ofpbuf_equal(&facet->xout.odp_actions, &xout.odp_actions)) {
4544 LIST_FOR_EACH(subfacet, list_node, &facet->subfacets) {
4545 if (subfacet->path == SF_FAST_PATH) {
4546 struct dpif_flow_stats stats;
4548 subfacet_install(subfacet, &xout.odp_actions, &stats);
4549 subfacet_update_stats(subfacet, &stats);
4553 facet_flush_stats(facet);
4555 ofpbuf_clear(&facet->xout.odp_actions);
4556 ofpbuf_put(&facet->xout.odp_actions, xout.odp_actions.data,
4557 xout.odp_actions.size);
4560 /* Update 'facet' now that we've taken care of all the old state. */
4561 facet->xout.tags = xout.tags;
4562 facet->xout.slow = xout.slow;
4563 facet->xout.has_learn = xout.has_learn;
4564 facet->xout.has_normal = xout.has_normal;
4565 facet->xout.has_fin_timeout = xout.has_fin_timeout;
4566 facet->xout.nf_output_iface = xout.nf_output_iface;
4567 facet->xout.mirrors = xout.mirrors;
4568 facet->nf_flow.output_iface = facet->xout.nf_output_iface;
4570 if (facet->rule != new_rule) {
4571 COVERAGE_INC(facet_changed_rule);
4572 list_remove(&facet->list_node);
4573 list_push_back(&new_rule->facets, &facet->list_node);
4574 facet->rule = new_rule;
4575 facet->used = new_rule->up.created;
4576 facet->prev_used = facet->used;
4579 xlate_out_uninit(&xout);
4584 facet_reset_counters(struct facet *facet)
4586 facet->packet_count = 0;
4587 facet->byte_count = 0;
4588 facet->prev_packet_count = 0;
4589 facet->prev_byte_count = 0;
4590 facet->accounted_bytes = 0;
4594 facet_push_stats(struct facet *facet, bool may_learn)
4596 struct dpif_flow_stats stats;
4598 ovs_assert(facet->packet_count >= facet->prev_packet_count);
4599 ovs_assert(facet->byte_count >= facet->prev_byte_count);
4600 ovs_assert(facet->used >= facet->prev_used);
4602 stats.n_packets = facet->packet_count - facet->prev_packet_count;
4603 stats.n_bytes = facet->byte_count - facet->prev_byte_count;
4604 stats.used = facet->used;
4605 stats.tcp_flags = facet->tcp_flags;
4607 if (may_learn || stats.n_packets || facet->used > facet->prev_used) {
4608 struct ofproto_dpif *ofproto =
4609 ofproto_dpif_cast(facet->rule->up.ofproto);
4611 struct ofport_dpif *in_port;
4612 struct xlate_in xin;
4614 facet->prev_packet_count = facet->packet_count;
4615 facet->prev_byte_count = facet->byte_count;
4616 facet->prev_used = facet->used;
4618 in_port = get_ofp_port(ofproto, facet->flow.in_port.ofp_port);
4619 if (in_port && in_port->is_tunnel) {
4620 netdev_vport_inc_rx(in_port->up.netdev, &stats);
4623 rule_credit_stats(facet->rule, &stats);
4624 netflow_flow_update_time(ofproto->netflow, &facet->nf_flow,
4626 netflow_flow_update_flags(&facet->nf_flow, facet->tcp_flags);
4627 mirror_update_stats(ofproto->mbridge, facet->xout.mirrors,
4628 stats.n_packets, stats.n_bytes);
4630 xlate_in_init(&xin, ofproto, &facet->flow, facet->rule,
4631 stats.tcp_flags, NULL);
4632 xin.resubmit_stats = &stats;
4633 xin.may_learn = may_learn;
4634 xlate_actions_for_side_effects(&xin);
4639 push_all_stats__(bool run_fast)
4641 static long long int rl = LLONG_MIN;
4642 struct ofproto_dpif *ofproto;
4644 if (time_msec() < rl) {
4648 HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
4649 struct cls_cursor cursor;
4650 struct facet *facet;
4652 cls_cursor_init(&cursor, &ofproto->facets, NULL);
4653 CLS_CURSOR_FOR_EACH (facet, cr, &cursor) {
4654 facet_push_stats(facet, false);
4661 rl = time_msec() + 100;
4665 push_all_stats(void)
4667 push_all_stats__(true);
4671 rule_credit_stats(struct rule_dpif *rule, const struct dpif_flow_stats *stats)
4673 rule->packet_count += stats->n_packets;
4674 rule->byte_count += stats->n_bytes;
4675 ofproto_rule_update_used(&rule->up, stats->used);
4680 static struct subfacet *
4681 subfacet_find(struct dpif_backer *backer, const struct nlattr *key,
4682 size_t key_len, uint32_t key_hash)
4684 struct subfacet *subfacet;
4686 HMAP_FOR_EACH_WITH_HASH (subfacet, hmap_node, key_hash,
4687 &backer->subfacets) {
4688 if (subfacet->key_len == key_len
4689 && !memcmp(key, subfacet->key, key_len)) {
4697 /* Searches 'facet' (within 'ofproto') for a subfacet with the specified
4698 * 'key_fitness', 'key', and 'key_len' members in 'miss'. Returns the
4699 * existing subfacet if there is one, otherwise creates and returns a
4701 static struct subfacet *
4702 subfacet_create(struct facet *facet, struct flow_miss *miss,
4705 struct dpif_backer *backer = miss->ofproto->backer;
4706 enum odp_key_fitness key_fitness = miss->key_fitness;
4707 const struct nlattr *key = miss->key;
4708 size_t key_len = miss->key_len;
4710 struct subfacet *subfacet;
4712 key_hash = odp_flow_key_hash(key, key_len);
4714 if (list_is_empty(&facet->subfacets)) {
4715 subfacet = &facet->one_subfacet;
4717 subfacet = subfacet_find(backer, key, key_len, key_hash);
4719 if (subfacet->facet == facet) {
4723 /* This shouldn't happen. */
4724 VLOG_ERR_RL(&rl, "subfacet with wrong facet");
4725 subfacet_destroy(subfacet);
4728 subfacet = xmalloc(sizeof *subfacet);
4731 hmap_insert(&backer->subfacets, &subfacet->hmap_node, key_hash);
4732 list_push_back(&facet->subfacets, &subfacet->list_node);
4733 subfacet->facet = facet;
4734 subfacet->key_fitness = key_fitness;
4735 subfacet->key = xmemdup(key, key_len);
4736 subfacet->key_len = key_len;
4737 subfacet->used = now;
4738 subfacet->created = now;
4739 subfacet->dp_packet_count = 0;
4740 subfacet->dp_byte_count = 0;
4741 subfacet->path = SF_NOT_INSTALLED;
4742 subfacet->backer = backer;
4744 backer->subfacet_add_count++;
4748 /* Uninstalls 'subfacet' from the datapath, if it is installed, removes it from
4749 * its facet within 'ofproto', and frees it. */
4751 subfacet_destroy__(struct subfacet *subfacet)
4753 struct facet *facet = subfacet->facet;
4754 struct ofproto_dpif *ofproto = ofproto_dpif_cast(facet->rule->up.ofproto);
4756 /* Update ofproto stats before uninstall the subfacet. */
4757 ofproto->backer->subfacet_del_count++;
4759 subfacet_uninstall(subfacet);
4760 hmap_remove(&subfacet->backer->subfacets, &subfacet->hmap_node);
4761 list_remove(&subfacet->list_node);
4762 free(subfacet->key);
4763 if (subfacet != &facet->one_subfacet) {
4768 /* Destroys 'subfacet', as with subfacet_destroy__(), and then if this was the
4769 * last remaining subfacet in its facet destroys the facet too. */
4771 subfacet_destroy(struct subfacet *subfacet)
4773 struct facet *facet = subfacet->facet;
4775 if (list_is_singleton(&facet->subfacets)) {
4776 /* facet_remove() needs at least one subfacet (it will remove it). */
4777 facet_remove(facet);
4779 subfacet_destroy__(subfacet);
4784 subfacet_destroy_batch(struct dpif_backer *backer,
4785 struct subfacet **subfacets, int n)
4787 struct dpif_op ops[SUBFACET_DESTROY_MAX_BATCH];
4788 struct dpif_op *opsp[SUBFACET_DESTROY_MAX_BATCH];
4789 struct dpif_flow_stats stats[SUBFACET_DESTROY_MAX_BATCH];
4792 for (i = 0; i < n; i++) {
4793 ops[i].type = DPIF_OP_FLOW_DEL;
4794 ops[i].u.flow_del.key = subfacets[i]->key;
4795 ops[i].u.flow_del.key_len = subfacets[i]->key_len;
4796 ops[i].u.flow_del.stats = &stats[i];
4800 dpif_operate(backer->dpif, opsp, n);
4801 for (i = 0; i < n; i++) {
4802 subfacet_reset_dp_stats(subfacets[i], &stats[i]);
4803 subfacets[i]->path = SF_NOT_INSTALLED;
4804 subfacet_destroy(subfacets[i]);
4809 /* Updates 'subfacet''s datapath flow, setting its actions to 'actions_len'
4810 * bytes of actions in 'actions'. If 'stats' is non-null, statistics counters
4811 * in the datapath will be zeroed and 'stats' will be updated with traffic new
4812 * since 'subfacet' was last updated.
4814 * Returns 0 if successful, otherwise a positive errno value. */
4816 subfacet_install(struct subfacet *subfacet, const struct ofpbuf *odp_actions,
4817 struct dpif_flow_stats *stats)
4819 struct facet *facet = subfacet->facet;
4820 struct ofproto_dpif *ofproto = ofproto_dpif_cast(facet->rule->up.ofproto);
4821 enum subfacet_path path = facet->xout.slow ? SF_SLOW_PATH : SF_FAST_PATH;
4822 const struct nlattr *actions = odp_actions->data;
4823 size_t actions_len = odp_actions->size;
4824 struct odputil_keybuf maskbuf;
4827 uint64_t slow_path_stub[128 / 8];
4828 enum dpif_flow_put_flags flags;
4831 flags = DPIF_FP_CREATE | DPIF_FP_MODIFY;
4833 flags |= DPIF_FP_ZERO_STATS;
4836 if (path == SF_SLOW_PATH) {
4837 compose_slow_path(ofproto, &facet->flow, facet->xout.slow,
4838 slow_path_stub, sizeof slow_path_stub,
4839 &actions, &actions_len);
4842 ofpbuf_use_stack(&mask, &maskbuf, sizeof maskbuf);
4843 if (enable_megaflows) {
4844 odp_flow_key_from_mask(&mask, &facet->xout.wc.masks,
4845 &facet->flow, UINT32_MAX);
4848 ret = dpif_flow_put(subfacet->backer->dpif, flags, subfacet->key,
4849 subfacet->key_len, mask.data, mask.size,
4850 actions, actions_len, stats);
4853 subfacet_reset_dp_stats(subfacet, stats);
4857 COVERAGE_INC(subfacet_install_fail);
4859 subfacet->path = path;
4864 /* If 'subfacet' is installed in the datapath, uninstalls it. */
4866 subfacet_uninstall(struct subfacet *subfacet)
4868 if (subfacet->path != SF_NOT_INSTALLED) {
4869 struct rule_dpif *rule = subfacet->facet->rule;
4870 struct ofproto_dpif *ofproto = ofproto_dpif_cast(rule->up.ofproto);
4871 struct dpif_flow_stats stats;
4874 error = dpif_flow_del(ofproto->backer->dpif, subfacet->key,
4875 subfacet->key_len, &stats);
4876 subfacet_reset_dp_stats(subfacet, &stats);
4878 subfacet_update_stats(subfacet, &stats);
4880 subfacet->path = SF_NOT_INSTALLED;
4882 ovs_assert(subfacet->dp_packet_count == 0);
4883 ovs_assert(subfacet->dp_byte_count == 0);
4887 /* Resets 'subfacet''s datapath statistics counters. This should be called
4888 * when 'subfacet''s statistics are cleared in the datapath. If 'stats' is
4889 * non-null, it should contain the statistics returned by dpif when 'subfacet'
4890 * was reset in the datapath. 'stats' will be modified to include only
4891 * statistics new since 'subfacet' was last updated. */
4893 subfacet_reset_dp_stats(struct subfacet *subfacet,
4894 struct dpif_flow_stats *stats)
4897 && subfacet->dp_packet_count <= stats->n_packets
4898 && subfacet->dp_byte_count <= stats->n_bytes) {
4899 stats->n_packets -= subfacet->dp_packet_count;
4900 stats->n_bytes -= subfacet->dp_byte_count;
4903 subfacet->dp_packet_count = 0;
4904 subfacet->dp_byte_count = 0;
4907 /* Folds the statistics from 'stats' into the counters in 'subfacet'.
4909 * Because of the meaning of a subfacet's counters, it only makes sense to do
4910 * this if 'stats' are not tracked in the datapath, that is, if 'stats'
4911 * represents a packet that was sent by hand or if it represents statistics
4912 * that have been cleared out of the datapath. */
4914 subfacet_update_stats(struct subfacet *subfacet,
4915 const struct dpif_flow_stats *stats)
4917 if (stats->n_packets || stats->used > subfacet->used) {
4918 struct facet *facet = subfacet->facet;
4920 subfacet->used = MAX(subfacet->used, stats->used);
4921 facet->used = MAX(facet->used, stats->used);
4922 facet->packet_count += stats->n_packets;
4923 facet->byte_count += stats->n_bytes;
4924 facet->tcp_flags |= stats->tcp_flags;
4930 /* Lookup 'flow' in 'ofproto''s classifier. If 'wc' is non-null, sets
4931 * the fields that were relevant as part of the lookup. */
4932 static struct rule_dpif *
4933 rule_dpif_lookup(struct ofproto_dpif *ofproto, const struct flow *flow,
4934 struct flow_wildcards *wc)
4936 struct rule_dpif *rule;
4938 rule = rule_dpif_lookup_in_table(ofproto, flow, wc, 0);
4943 return rule_dpif_miss_rule(ofproto, flow);
4947 rule_dpif_lookup_in_table(struct ofproto_dpif *ofproto,
4948 const struct flow *flow, struct flow_wildcards *wc,
4951 struct cls_rule *cls_rule;
4952 struct classifier *cls;
4955 if (table_id >= N_TABLES) {
4960 memset(&wc->masks.dl_type, 0xff, sizeof wc->masks.dl_type);
4961 wc->masks.nw_frag |= FLOW_NW_FRAG_MASK;
4964 cls = &ofproto->up.tables[table_id].cls;
4965 frag = (flow->nw_frag & FLOW_NW_FRAG_ANY) != 0;
4966 if (frag && ofproto->up.frag_handling == OFPC_FRAG_NORMAL) {
4967 /* We must pretend that transport ports are unavailable. */
4968 struct flow ofpc_normal_flow = *flow;
4969 ofpc_normal_flow.tp_src = htons(0);
4970 ofpc_normal_flow.tp_dst = htons(0);
4971 cls_rule = classifier_lookup(cls, &ofpc_normal_flow, wc);
4972 } else if (frag && ofproto->up.frag_handling == OFPC_FRAG_DROP) {
4973 cls_rule = &ofproto->drop_frags_rule->up.cr;
4975 flow_wildcards_init_exact(wc);
4978 cls_rule = classifier_lookup(cls, flow, wc);
4980 return rule_dpif_cast(rule_from_cls_rule(cls_rule));
4984 rule_dpif_miss_rule(struct ofproto_dpif *ofproto, const struct flow *flow)
4986 struct ofport_dpif *port;
4988 port = get_ofp_port(ofproto, flow->in_port.ofp_port);
4990 VLOG_WARN_RL(&rl, "packet-in on unknown OpenFlow port %"PRIu16,
4991 flow->in_port.ofp_port);
4992 return ofproto->miss_rule;
4995 if (port->up.pp.config & OFPUTIL_PC_NO_PACKET_IN) {
4996 return ofproto->no_packet_in_rule;
4998 return ofproto->miss_rule;
5002 complete_operation(struct rule_dpif *rule)
5004 struct ofproto_dpif *ofproto = ofproto_dpif_cast(rule->up.ofproto);
5006 rule_invalidate(rule);
5008 struct dpif_completion *c = xmalloc(sizeof *c);
5009 c->op = rule->up.pending;
5010 list_push_back(&ofproto->completions, &c->list_node);
5012 ofoperation_complete(rule->up.pending, 0);
5016 static struct rule *
5019 struct rule_dpif *rule = xmalloc(sizeof *rule);
5024 rule_dealloc(struct rule *rule_)
5026 struct rule_dpif *rule = rule_dpif_cast(rule_);
5031 rule_construct(struct rule *rule_)
5033 struct rule_dpif *rule = rule_dpif_cast(rule_);
5034 struct ofproto_dpif *ofproto = ofproto_dpif_cast(rule->up.ofproto);
5035 struct rule_dpif *victim;
5038 rule->packet_count = 0;
5039 rule->byte_count = 0;
5041 victim = rule_dpif_cast(ofoperation_get_victim(rule->up.pending));
5042 if (victim && !list_is_empty(&victim->facets)) {
5043 struct facet *facet;
5045 rule->facets = victim->facets;
5046 list_moved(&rule->facets);
5047 LIST_FOR_EACH (facet, list_node, &rule->facets) {
5048 /* XXX: We're only clearing our local counters here. It's possible
5049 * that quite a few packets are unaccounted for in the datapath
5050 * statistics. These will be accounted to the new rule instead of
5051 * cleared as required. This could be fixed by clearing out the
5052 * datapath statistics for this facet, but currently it doesn't
5054 facet_reset_counters(facet);
5058 /* Must avoid list_moved() in this case. */
5059 list_init(&rule->facets);
5062 table_id = rule->up.table_id;
5064 rule->tag = victim->tag;
5065 } else if (table_id == 0) {
5070 miniflow_expand(&rule->up.cr.match.flow, &flow);
5071 rule->tag = rule_calculate_tag(&flow, &rule->up.cr.match.mask,
5072 ofproto->tables[table_id].basis);
5075 complete_operation(rule);
5080 rule_destruct(struct rule *rule_)
5082 struct rule_dpif *rule = rule_dpif_cast(rule_);
5083 struct facet *facet, *next_facet;
5085 LIST_FOR_EACH_SAFE (facet, next_facet, list_node, &rule->facets) {
5086 facet_revalidate(facet);
5089 complete_operation(rule);
5093 rule_get_stats(struct rule *rule_, uint64_t *packets, uint64_t *bytes)
5095 struct rule_dpif *rule = rule_dpif_cast(rule_);
5097 /* push_all_stats() can handle flow misses which, when using the learn
5098 * action, can cause rules to be added and deleted. This can corrupt our
5099 * caller's datastructures which assume that rule_get_stats() doesn't have
5100 * an impact on the flow table. To be safe, we disable miss handling. */
5101 push_all_stats__(false);
5103 /* Start from historical data for 'rule' itself that are no longer tracked
5104 * in facets. This counts, for example, facets that have expired. */
5105 *packets = rule->packet_count;
5106 *bytes = rule->byte_count;
5110 rule_dpif_execute(struct rule_dpif *rule, const struct flow *flow,
5111 struct ofpbuf *packet)
5113 struct ofproto_dpif *ofproto = ofproto_dpif_cast(rule->up.ofproto);
5114 struct dpif_flow_stats stats;
5115 struct xlate_out xout;
5116 struct xlate_in xin;
5118 dpif_flow_stats_extract(flow, packet, time_msec(), &stats);
5119 rule_credit_stats(rule, &stats);
5121 xlate_in_init(&xin, ofproto, flow, rule, stats.tcp_flags, packet);
5122 xin.resubmit_stats = &stats;
5123 xlate_actions(&xin, &xout);
5125 execute_odp_actions(ofproto, flow, xout.odp_actions.data,
5126 xout.odp_actions.size, packet);
5128 xlate_out_uninit(&xout);
5132 rule_execute(struct rule *rule, const struct flow *flow,
5133 struct ofpbuf *packet)
5135 rule_dpif_execute(rule_dpif_cast(rule), flow, packet);
5136 ofpbuf_delete(packet);
5141 rule_modify_actions(struct rule *rule_)
5143 struct rule_dpif *rule = rule_dpif_cast(rule_);
5145 complete_operation(rule);
5148 /* Sends 'packet' out 'ofport'.
5149 * May modify 'packet'.
5150 * Returns 0 if successful, otherwise a positive errno value. */
5152 send_packet(const struct ofport_dpif *ofport, struct ofpbuf *packet)
5154 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport->up.ofproto);
5155 uint64_t odp_actions_stub[1024 / 8];
5156 struct ofpbuf key, odp_actions;
5157 struct dpif_flow_stats stats;
5158 struct odputil_keybuf keybuf;
5159 struct ofpact_output output;
5160 struct xlate_out xout;
5161 struct xlate_in xin;
5163 union flow_in_port in_port_;
5166 ofpbuf_use_stub(&odp_actions, odp_actions_stub, sizeof odp_actions_stub);
5167 ofpbuf_use_stack(&key, &keybuf, sizeof keybuf);
5169 /* Use OFPP_NONE as the in_port to avoid special packet processing. */
5170 in_port_.ofp_port = OFPP_NONE;
5171 flow_extract(packet, 0, 0, NULL, &in_port_, &flow);
5172 odp_flow_key_from_flow(&key, &flow, ofp_port_to_odp_port(ofproto,
5174 dpif_flow_stats_extract(&flow, packet, time_msec(), &stats);
5176 ofpact_init(&output.ofpact, OFPACT_OUTPUT, sizeof output);
5177 output.port = ofport->up.ofp_port;
5180 xlate_in_init(&xin, ofproto, &flow, NULL, 0, packet);
5181 xin.ofpacts_len = sizeof output;
5182 xin.ofpacts = &output.ofpact;
5183 xin.resubmit_stats = &stats;
5184 xlate_actions(&xin, &xout);
5186 error = dpif_execute(ofproto->backer->dpif,
5188 xout.odp_actions.data, xout.odp_actions.size,
5190 xlate_out_uninit(&xout);
5193 VLOG_WARN_RL(&rl, "%s: failed to send packet on port %s (%s)",
5194 ofproto->up.name, netdev_get_name(ofport->up.netdev),
5195 ovs_strerror(error));
5198 ofproto->stats.tx_packets++;
5199 ofproto->stats.tx_bytes += packet->size;
5203 /* Composes an ODP action for a "slow path" action for 'flow' within 'ofproto'.
5204 * The action will state 'slow' as the reason that the action is in the slow
5205 * path. (This is purely informational: it allows a human viewing "ovs-dpctl
5206 * dump-flows" output to see why a flow is in the slow path.)
5208 * The 'stub_size' bytes in 'stub' will be used to store the action.
5209 * 'stub_size' must be large enough for the action.
5211 * The action and its size will be stored in '*actionsp' and '*actions_lenp',
5214 compose_slow_path(const struct ofproto_dpif *ofproto, const struct flow *flow,
5215 enum slow_path_reason slow,
5216 uint64_t *stub, size_t stub_size,
5217 const struct nlattr **actionsp, size_t *actions_lenp)
5219 union user_action_cookie cookie;
5222 cookie.type = USER_ACTION_COOKIE_SLOW_PATH;
5223 cookie.slow_path.unused = 0;
5224 cookie.slow_path.reason = slow;
5226 ofpbuf_use_stack(&buf, stub, stub_size);
5227 if (slow & (SLOW_CFM | SLOW_BFD | SLOW_LACP | SLOW_STP)) {
5228 uint32_t pid = dpif_port_get_pid(ofproto->backer->dpif,
5230 odp_put_userspace_action(pid, &cookie, sizeof cookie.slow_path, &buf);
5232 put_userspace_action(ofproto, &buf, flow, &cookie,
5233 sizeof cookie.slow_path);
5235 *actionsp = buf.data;
5236 *actions_lenp = buf.size;
5240 put_userspace_action(const struct ofproto_dpif *ofproto,
5241 struct ofpbuf *odp_actions,
5242 const struct flow *flow,
5243 const union user_action_cookie *cookie,
5244 const size_t cookie_size)
5248 pid = dpif_port_get_pid(ofproto->backer->dpif,
5249 ofp_port_to_odp_port(ofproto,
5250 flow->in_port.ofp_port));
5252 return odp_put_userspace_action(pid, cookie, cookie_size, odp_actions);
5256 calculate_flow_tag(struct ofproto_dpif *ofproto, const struct flow *flow,
5257 uint8_t table_id, struct rule_dpif *rule)
5259 if (table_id > 0 && table_id < N_TABLES) {
5260 struct table_dpif *table = &ofproto->tables[table_id];
5261 if (table->other_table) {
5262 return (rule && rule->tag
5264 : rule_calculate_tag(flow, &table->other_table->mask,
5272 /* Optimized flow revalidation.
5274 * It's a difficult problem, in general, to tell which facets need to have
5275 * their actions recalculated whenever the OpenFlow flow table changes. We
5276 * don't try to solve that general problem: for most kinds of OpenFlow flow
5277 * table changes, we recalculate the actions for every facet. This is
5278 * relatively expensive, but it's good enough if the OpenFlow flow table
5279 * doesn't change very often.
5281 * However, we can expect one particular kind of OpenFlow flow table change to
5282 * happen frequently: changes caused by MAC learning. To avoid wasting a lot
5283 * of CPU on revalidating every facet whenever MAC learning modifies the flow
5284 * table, we add a special case that applies to flow tables in which every rule
5285 * has the same form (that is, the same wildcards), except that the table is
5286 * also allowed to have a single "catch-all" flow that matches all packets. We
5287 * optimize this case by tagging all of the facets that resubmit into the table
5288 * and invalidating the same tag whenever a flow changes in that table. The
5289 * end result is that we revalidate just the facets that need it (and sometimes
5290 * a few more, but not all of the facets or even all of the facets that
5291 * resubmit to the table modified by MAC learning). */
5293 /* Calculates the tag to use for 'flow' and mask 'mask' when it is inserted
5294 * into an OpenFlow table with the given 'basis'. */
5296 rule_calculate_tag(const struct flow *flow, const struct minimask *mask,
5299 if (minimask_is_catchall(mask)) {
5302 uint32_t hash = flow_hash_in_minimask(flow, mask, secret);
5303 return tag_create_deterministic(hash);
5307 /* Following a change to OpenFlow table 'table_id' in 'ofproto', update the
5308 * taggability of that table.
5310 * This function must be called after *each* change to a flow table. If you
5311 * skip calling it on some changes then the pointer comparisons at the end can
5312 * be invalid if you get unlucky. For example, if a flow removal causes a
5313 * cls_table to be destroyed and then a flow insertion causes a cls_table with
5314 * different wildcards to be created with the same address, then this function
5315 * will incorrectly skip revalidation. */
5317 table_update_taggable(struct ofproto_dpif *ofproto, uint8_t table_id)
5319 struct table_dpif *table = &ofproto->tables[table_id];
5320 const struct oftable *oftable = &ofproto->up.tables[table_id];
5321 struct cls_table *catchall, *other;
5322 struct cls_table *t;
5324 catchall = other = NULL;
5326 switch (hmap_count(&oftable->cls.tables)) {
5328 /* We could tag this OpenFlow table but it would make the logic a
5329 * little harder and it's a corner case that doesn't seem worth it
5335 HMAP_FOR_EACH (t, hmap_node, &oftable->cls.tables) {
5336 if (cls_table_is_catchall(t)) {
5338 } else if (!other) {
5341 /* Indicate that we can't tag this by setting both tables to
5342 * NULL. (We know that 'catchall' is already NULL.) */
5349 /* Can't tag this table. */
5353 if (table->catchall_table != catchall || table->other_table != other) {
5354 table->catchall_table = catchall;
5355 table->other_table = other;
5356 ofproto->backer->need_revalidate = REV_FLOW_TABLE;
5360 /* Given 'rule' that has changed in some way (either it is a rule being
5361 * inserted, a rule being deleted, or a rule whose actions are being
5362 * modified), marks facets for revalidation to ensure that packets will be
5363 * forwarded correctly according to the new state of the flow table.
5365 * This function must be called after *each* change to a flow table. See
5366 * the comment on table_update_taggable() for more information. */
5368 rule_invalidate(const struct rule_dpif *rule)
5370 struct ofproto_dpif *ofproto = ofproto_dpif_cast(rule->up.ofproto);
5372 table_update_taggable(ofproto, rule->up.table_id);
5374 if (!ofproto->backer->need_revalidate) {
5375 struct table_dpif *table = &ofproto->tables[rule->up.table_id];
5377 if (table->other_table && rule->tag) {
5378 tag_set_add(&ofproto->backer->revalidate_set, rule->tag);
5380 ofproto->backer->need_revalidate = REV_FLOW_TABLE;
5386 set_frag_handling(struct ofproto *ofproto_,
5387 enum ofp_config_flags frag_handling)
5389 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
5390 if (frag_handling != OFPC_FRAG_REASM) {
5391 ofproto->backer->need_revalidate = REV_RECONFIGURE;
5399 packet_out(struct ofproto *ofproto_, struct ofpbuf *packet,
5400 const struct flow *flow,
5401 const struct ofpact *ofpacts, size_t ofpacts_len)
5403 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
5404 struct odputil_keybuf keybuf;
5405 struct dpif_flow_stats stats;
5406 struct xlate_out xout;
5407 struct xlate_in xin;
5411 ofpbuf_use_stack(&key, &keybuf, sizeof keybuf);
5412 odp_flow_key_from_flow(&key, flow,
5413 ofp_port_to_odp_port(ofproto,
5414 flow->in_port.ofp_port));
5416 dpif_flow_stats_extract(flow, packet, time_msec(), &stats);
5418 xlate_in_init(&xin, ofproto, flow, NULL, stats.tcp_flags, packet);
5419 xin.resubmit_stats = &stats;
5420 xin.ofpacts_len = ofpacts_len;
5421 xin.ofpacts = ofpacts;
5423 xlate_actions(&xin, &xout);
5424 dpif_execute(ofproto->backer->dpif, key.data, key.size,
5425 xout.odp_actions.data, xout.odp_actions.size, packet);
5426 xlate_out_uninit(&xout);
5434 set_netflow(struct ofproto *ofproto_,
5435 const struct netflow_options *netflow_options)
5437 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
5439 if (netflow_options) {
5440 if (!ofproto->netflow) {
5441 ofproto->netflow = netflow_create();
5442 ofproto->backer->need_revalidate = REV_RECONFIGURE;
5444 return netflow_set_options(ofproto->netflow, netflow_options);
5445 } else if (ofproto->netflow) {
5446 ofproto->backer->need_revalidate = REV_RECONFIGURE;
5447 netflow_destroy(ofproto->netflow);
5448 ofproto->netflow = NULL;
5455 get_netflow_ids(const struct ofproto *ofproto_,
5456 uint8_t *engine_type, uint8_t *engine_id)
5458 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
5460 dpif_get_netflow_ids(ofproto->backer->dpif, engine_type, engine_id);
5464 send_active_timeout(struct ofproto_dpif *ofproto, struct facet *facet)
5466 if (!facet_is_controller_flow(facet) &&
5467 netflow_active_timeout_expired(ofproto->netflow, &facet->nf_flow)) {
5468 struct subfacet *subfacet;
5469 struct ofexpired expired;
5471 LIST_FOR_EACH (subfacet, list_node, &facet->subfacets) {
5472 if (subfacet->path == SF_FAST_PATH) {
5473 struct dpif_flow_stats stats;
5475 subfacet_install(subfacet, &facet->xout.odp_actions,
5477 subfacet_update_stats(subfacet, &stats);
5481 expired.flow = facet->flow;
5482 expired.packet_count = facet->packet_count;
5483 expired.byte_count = facet->byte_count;
5484 expired.used = facet->used;
5485 netflow_expire(ofproto->netflow, &facet->nf_flow, &expired);
5490 send_netflow_active_timeouts(struct ofproto_dpif *ofproto)
5492 struct cls_cursor cursor;
5493 struct facet *facet;
5495 cls_cursor_init(&cursor, &ofproto->facets, NULL);
5496 CLS_CURSOR_FOR_EACH (facet, cr, &cursor) {
5497 send_active_timeout(ofproto, facet);
5501 static struct ofproto_dpif *
5502 ofproto_dpif_lookup(const char *name)
5504 struct ofproto_dpif *ofproto;
5506 HMAP_FOR_EACH_WITH_HASH (ofproto, all_ofproto_dpifs_node,
5507 hash_string(name, 0), &all_ofproto_dpifs) {
5508 if (!strcmp(ofproto->up.name, name)) {
5516 ofproto_unixctl_fdb_flush(struct unixctl_conn *conn, int argc,
5517 const char *argv[], void *aux OVS_UNUSED)
5519 struct ofproto_dpif *ofproto;
5522 ofproto = ofproto_dpif_lookup(argv[1]);
5524 unixctl_command_reply_error(conn, "no such bridge");
5527 mac_learning_flush(ofproto->ml, &ofproto->backer->revalidate_set);
5529 HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
5530 mac_learning_flush(ofproto->ml, &ofproto->backer->revalidate_set);
5534 unixctl_command_reply(conn, "table successfully flushed");
5538 ofproto_unixctl_fdb_show(struct unixctl_conn *conn, int argc OVS_UNUSED,
5539 const char *argv[], void *aux OVS_UNUSED)
5541 struct ds ds = DS_EMPTY_INITIALIZER;
5542 const struct ofproto_dpif *ofproto;
5543 const struct mac_entry *e;
5545 ofproto = ofproto_dpif_lookup(argv[1]);
5547 unixctl_command_reply_error(conn, "no such bridge");
5551 ds_put_cstr(&ds, " port VLAN MAC Age\n");
5552 LIST_FOR_EACH (e, lru_node, &ofproto->ml->lrus) {
5553 struct ofbundle *bundle = e->port.p;
5554 char name[OFP_MAX_PORT_NAME_LEN];
5556 ofputil_port_to_string(ofbundle_get_a_port(bundle)->up.ofp_port,
5558 ds_put_format(&ds, "%5s %4d "ETH_ADDR_FMT" %3d\n",
5559 name, e->vlan, ETH_ADDR_ARGS(e->mac),
5560 mac_entry_age(ofproto->ml, e));
5562 unixctl_command_reply(conn, ds_cstr(&ds));
5567 struct xlate_out xout;
5568 struct xlate_in xin;
5574 trace_format_rule(struct ds *result, int level, const struct rule_dpif *rule)
5576 ds_put_char_multiple(result, '\t', level);
5578 ds_put_cstr(result, "No match\n");
5582 ds_put_format(result, "Rule: table=%"PRIu8" cookie=%#"PRIx64" ",
5583 rule ? rule->up.table_id : 0, ntohll(rule->up.flow_cookie));
5584 cls_rule_format(&rule->up.cr, result);
5585 ds_put_char(result, '\n');
5587 ds_put_char_multiple(result, '\t', level);
5588 ds_put_cstr(result, "OpenFlow ");
5589 ofpacts_format(rule->up.ofpacts, rule->up.ofpacts_len, result);
5590 ds_put_char(result, '\n');
5594 trace_format_flow(struct ds *result, int level, const char *title,
5595 struct trace_ctx *trace)
5597 ds_put_char_multiple(result, '\t', level);
5598 ds_put_format(result, "%s: ", title);
5599 if (flow_equal(&trace->xin.flow, &trace->flow)) {
5600 ds_put_cstr(result, "unchanged");
5602 flow_format(result, &trace->xin.flow);
5603 trace->flow = trace->xin.flow;
5605 ds_put_char(result, '\n');
5609 trace_format_regs(struct ds *result, int level, const char *title,
5610 struct trace_ctx *trace)
5614 ds_put_char_multiple(result, '\t', level);
5615 ds_put_format(result, "%s:", title);
5616 for (i = 0; i < FLOW_N_REGS; i++) {
5617 ds_put_format(result, " reg%zu=0x%"PRIx32, i, trace->flow.regs[i]);
5619 ds_put_char(result, '\n');
5623 trace_format_odp(struct ds *result, int level, const char *title,
5624 struct trace_ctx *trace)
5626 struct ofpbuf *odp_actions = &trace->xout.odp_actions;
5628 ds_put_char_multiple(result, '\t', level);
5629 ds_put_format(result, "%s: ", title);
5630 format_odp_actions(result, odp_actions->data, odp_actions->size);
5631 ds_put_char(result, '\n');
5635 trace_resubmit(struct xlate_in *xin, struct rule_dpif *rule, int recurse)
5637 struct trace_ctx *trace = CONTAINER_OF(xin, struct trace_ctx, xin);
5638 struct ds *result = trace->result;
5640 ds_put_char(result, '\n');
5641 trace_format_flow(result, recurse + 1, "Resubmitted flow", trace);
5642 trace_format_regs(result, recurse + 1, "Resubmitted regs", trace);
5643 trace_format_odp(result, recurse + 1, "Resubmitted odp", trace);
5644 trace_format_rule(result, recurse + 1, rule);
5648 trace_report(struct xlate_in *xin, const char *s, int recurse)
5650 struct trace_ctx *trace = CONTAINER_OF(xin, struct trace_ctx, xin);
5651 struct ds *result = trace->result;
5653 ds_put_char_multiple(result, '\t', recurse);
5654 ds_put_cstr(result, s);
5655 ds_put_char(result, '\n');
5659 ofproto_unixctl_trace(struct unixctl_conn *conn, int argc, const char *argv[],
5660 void *aux OVS_UNUSED)
5662 const struct dpif_backer *backer;
5663 struct ofproto_dpif *ofproto;
5664 struct ofpbuf odp_key, odp_mask;
5665 struct ofpbuf *packet;
5673 ofpbuf_init(&odp_key, 0);
5674 ofpbuf_init(&odp_mask, 0);
5676 /* Handle "-generate" or a hex string as the last argument. */
5677 if (!strcmp(argv[argc - 1], "-generate")) {
5678 packet = ofpbuf_new(0);
5681 const char *error = eth_from_hex(argv[argc - 1], &packet);
5684 } else if (argc == 4) {
5685 /* The 3-argument form must end in "-generate' or a hex string. */
5686 unixctl_command_reply_error(conn, error);
5691 /* Parse the flow and determine whether a datapath or
5692 * bridge is specified. If function odp_flow_key_from_string()
5693 * returns 0, the flow is a odp_flow. If function
5694 * parse_ofp_exact_flow() returns 0, the flow is a br_flow. */
5695 if (!odp_flow_from_string(argv[argc - 1], NULL, &odp_key, &odp_mask)) {
5696 /* If the odp_flow is the second argument,
5697 * the datapath name is the first argument. */
5699 const char *dp_type;
5700 if (!strncmp(argv[1], "ovs-", 4)) {
5701 dp_type = argv[1] + 4;
5705 backer = shash_find_data(&all_dpif_backers, dp_type);
5707 unixctl_command_reply_error(conn, "Cannot find datapath "
5712 /* No datapath name specified, so there should be only one
5714 struct shash_node *node;
5715 if (shash_count(&all_dpif_backers) != 1) {
5716 unixctl_command_reply_error(conn, "Must specify datapath "
5717 "name, there is more than one type of datapath");
5720 node = shash_first(&all_dpif_backers);
5721 backer = node->data;
5724 /* Extract the ofproto_dpif object from the ofproto_receive()
5726 if (ofproto_receive(backer, NULL, odp_key.data,
5727 odp_key.size, &flow, NULL, &ofproto, NULL)) {
5728 unixctl_command_reply_error(conn, "Invalid datapath flow");
5731 ds_put_format(&result, "Bridge: %s\n", ofproto->up.name);
5732 } else if (!parse_ofp_exact_flow(&flow, argv[argc - 1])) {
5734 unixctl_command_reply_error(conn, "Must specify bridge name");
5738 ofproto = ofproto_dpif_lookup(argv[1]);
5740 unixctl_command_reply_error(conn, "Unknown bridge name");
5744 unixctl_command_reply_error(conn, "Bad flow syntax");
5748 /* Generate a packet, if requested. */
5750 if (!packet->size) {
5751 flow_compose(packet, &flow);
5753 union flow_in_port in_port_;
5755 in_port_ = flow.in_port;
5756 ds_put_cstr(&result, "Packet: ");
5757 s = ofp_packet_to_string(packet->data, packet->size);
5758 ds_put_cstr(&result, s);
5761 /* Use the metadata from the flow and the packet argument
5762 * to reconstruct the flow. */
5763 flow_extract(packet, flow.skb_priority, flow.skb_mark, NULL,
5768 ofproto_trace(ofproto, &flow, packet, &result);
5769 unixctl_command_reply(conn, ds_cstr(&result));
5772 ds_destroy(&result);
5773 ofpbuf_delete(packet);
5774 ofpbuf_uninit(&odp_key);
5775 ofpbuf_uninit(&odp_mask);
5779 ofproto_trace(struct ofproto_dpif *ofproto, const struct flow *flow,
5780 const struct ofpbuf *packet, struct ds *ds)
5782 struct rule_dpif *rule;
5784 ds_put_cstr(ds, "Flow: ");
5785 flow_format(ds, flow);
5786 ds_put_char(ds, '\n');
5788 rule = rule_dpif_lookup(ofproto, flow, NULL);
5790 trace_format_rule(ds, 0, rule);
5791 if (rule == ofproto->miss_rule) {
5792 ds_put_cstr(ds, "\nNo match, flow generates \"packet in\"s.\n");
5793 } else if (rule == ofproto->no_packet_in_rule) {
5794 ds_put_cstr(ds, "\nNo match, packets dropped because "
5795 "OFPPC_NO_PACKET_IN is set on in_port.\n");
5796 } else if (rule == ofproto->drop_frags_rule) {
5797 ds_put_cstr(ds, "\nPackets dropped because they are IP fragments "
5798 "and the fragment handling mode is \"drop\".\n");
5802 uint64_t odp_actions_stub[1024 / 8];
5803 struct ofpbuf odp_actions;
5804 struct trace_ctx trace;
5808 tcp_flags = packet ? packet_get_tcp_flags(packet, flow) : 0;
5811 ofpbuf_use_stub(&odp_actions,
5812 odp_actions_stub, sizeof odp_actions_stub);
5813 xlate_in_init(&trace.xin, ofproto, flow, rule, tcp_flags, packet);
5814 trace.xin.resubmit_hook = trace_resubmit;
5815 trace.xin.report_hook = trace_report;
5817 xlate_actions(&trace.xin, &trace.xout);
5819 ds_put_char(ds, '\n');
5820 trace_format_flow(ds, 0, "Final flow", &trace);
5822 match_init(&match, flow, &trace.xout.wc);
5823 ds_put_cstr(ds, "Relevant fields: ");
5824 match_format(&match, ds, OFP_DEFAULT_PRIORITY);
5825 ds_put_char(ds, '\n');
5827 ds_put_cstr(ds, "Datapath actions: ");
5828 format_odp_actions(ds, trace.xout.odp_actions.data,
5829 trace.xout.odp_actions.size);
5831 if (trace.xout.slow) {
5832 ds_put_cstr(ds, "\nThis flow is handled by the userspace "
5833 "slow path because it:");
5834 switch (trace.xout.slow) {
5836 ds_put_cstr(ds, "\n\t- Consists of CFM packets.");
5839 ds_put_cstr(ds, "\n\t- Consists of LACP packets.");
5842 ds_put_cstr(ds, "\n\t- Consists of STP packets.");
5845 ds_put_cstr(ds, "\n\t- Consists of BFD packets.");
5847 case SLOW_CONTROLLER:
5848 ds_put_cstr(ds, "\n\t- Sends \"packet-in\" messages "
5849 "to the OpenFlow controller.");
5856 xlate_out_uninit(&trace.xout);
5861 ofproto_dpif_clog(struct unixctl_conn *conn OVS_UNUSED, int argc OVS_UNUSED,
5862 const char *argv[] OVS_UNUSED, void *aux OVS_UNUSED)
5865 unixctl_command_reply(conn, NULL);
5869 ofproto_dpif_unclog(struct unixctl_conn *conn OVS_UNUSED, int argc OVS_UNUSED,
5870 const char *argv[] OVS_UNUSED, void *aux OVS_UNUSED)
5873 unixctl_command_reply(conn, NULL);
5876 /* Runs a self-check of flow translations in 'ofproto'. Appends a message to
5877 * 'reply' describing the results. */
5879 ofproto_dpif_self_check__(struct ofproto_dpif *ofproto, struct ds *reply)
5881 struct cls_cursor cursor;
5882 struct facet *facet;
5886 cls_cursor_init(&cursor, &ofproto->facets, NULL);
5887 CLS_CURSOR_FOR_EACH (facet, cr, &cursor) {
5888 if (!facet_check_consistency(facet)) {
5893 ofproto->backer->need_revalidate = REV_INCONSISTENCY;
5897 ds_put_format(reply, "%s: self-check failed (%d errors)\n",
5898 ofproto->up.name, errors);
5900 ds_put_format(reply, "%s: self-check passed\n", ofproto->up.name);
5905 ofproto_dpif_self_check(struct unixctl_conn *conn,
5906 int argc, const char *argv[], void *aux OVS_UNUSED)
5908 struct ds reply = DS_EMPTY_INITIALIZER;
5909 struct ofproto_dpif *ofproto;
5912 ofproto = ofproto_dpif_lookup(argv[1]);
5914 unixctl_command_reply_error(conn, "Unknown ofproto (use "
5915 "ofproto/list for help)");
5918 ofproto_dpif_self_check__(ofproto, &reply);
5920 HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
5921 ofproto_dpif_self_check__(ofproto, &reply);
5925 unixctl_command_reply(conn, ds_cstr(&reply));
5929 /* Store the current ofprotos in 'ofproto_shash'. Returns a sorted list
5930 * of the 'ofproto_shash' nodes. It is the responsibility of the caller
5931 * to destroy 'ofproto_shash' and free the returned value. */
5932 static const struct shash_node **
5933 get_ofprotos(struct shash *ofproto_shash)
5935 const struct ofproto_dpif *ofproto;
5937 HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
5938 char *name = xasprintf("%s@%s", ofproto->up.type, ofproto->up.name);
5939 shash_add_nocopy(ofproto_shash, name, ofproto);
5942 return shash_sort(ofproto_shash);
5946 ofproto_unixctl_dpif_dump_dps(struct unixctl_conn *conn, int argc OVS_UNUSED,
5947 const char *argv[] OVS_UNUSED,
5948 void *aux OVS_UNUSED)
5950 struct ds ds = DS_EMPTY_INITIALIZER;
5951 struct shash ofproto_shash;
5952 const struct shash_node **sorted_ofprotos;
5955 shash_init(&ofproto_shash);
5956 sorted_ofprotos = get_ofprotos(&ofproto_shash);
5957 for (i = 0; i < shash_count(&ofproto_shash); i++) {
5958 const struct shash_node *node = sorted_ofprotos[i];
5959 ds_put_format(&ds, "%s\n", node->name);
5962 shash_destroy(&ofproto_shash);
5963 free(sorted_ofprotos);
5965 unixctl_command_reply(conn, ds_cstr(&ds));
5970 show_dp_rates(struct ds *ds, const char *heading,
5971 const struct avg_subfacet_rates *rates)
5973 ds_put_format(ds, "%s add rate: %5.3f/min, del rate: %5.3f/min\n",
5974 heading, rates->add_rate, rates->del_rate);
5978 dpif_show_backer(const struct dpif_backer *backer, struct ds *ds)
5980 const struct shash_node **ofprotos;
5981 struct ofproto_dpif *ofproto;
5982 struct shash ofproto_shash;
5983 uint64_t n_hit, n_missed;
5984 long long int minutes;
5987 n_hit = n_missed = 0;
5988 HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
5989 if (ofproto->backer == backer) {
5990 n_missed += ofproto->n_missed;
5991 n_hit += ofproto->n_hit;
5995 ds_put_format(ds, "%s: hit:%"PRIu64" missed:%"PRIu64"\n",
5996 dpif_name(backer->dpif), n_hit, n_missed);
5997 ds_put_format(ds, "\tflows: cur: %zu, avg: %u, max: %u,"
5998 " life span: %lldms\n", hmap_count(&backer->subfacets),
5999 backer->avg_n_subfacet, backer->max_n_subfacet,
6000 backer->avg_subfacet_life);
6002 minutes = (time_msec() - backer->created) / (1000 * 60);
6003 if (minutes >= 60) {
6004 show_dp_rates(ds, "\thourly avg:", &backer->hourly);
6006 if (minutes >= 60 * 24) {
6007 show_dp_rates(ds, "\tdaily avg:", &backer->daily);
6009 show_dp_rates(ds, "\toverall avg:", &backer->lifetime);
6011 shash_init(&ofproto_shash);
6012 ofprotos = get_ofprotos(&ofproto_shash);
6013 for (i = 0; i < shash_count(&ofproto_shash); i++) {
6014 struct ofproto_dpif *ofproto = ofprotos[i]->data;
6015 const struct shash_node **ports;
6018 if (ofproto->backer != backer) {
6022 ds_put_format(ds, "\t%s: hit:%"PRIu64" missed:%"PRIu64"\n",
6023 ofproto->up.name, ofproto->n_hit, ofproto->n_missed);
6025 ports = shash_sort(&ofproto->up.port_by_name);
6026 for (j = 0; j < shash_count(&ofproto->up.port_by_name); j++) {
6027 const struct shash_node *node = ports[j];
6028 struct ofport *ofport = node->data;
6030 odp_port_t odp_port;
6032 ds_put_format(ds, "\t\t%s %u/", netdev_get_name(ofport->netdev),
6035 odp_port = ofp_port_to_odp_port(ofproto, ofport->ofp_port);
6036 if (odp_port != ODPP_NONE) {
6037 ds_put_format(ds, "%"PRIu32":", odp_port);
6039 ds_put_cstr(ds, "none:");
6042 ds_put_format(ds, " (%s", netdev_get_type(ofport->netdev));
6045 if (!netdev_get_config(ofport->netdev, &config)) {
6046 const struct smap_node **nodes;
6049 nodes = smap_sort(&config);
6050 for (i = 0; i < smap_count(&config); i++) {
6051 const struct smap_node *node = nodes[i];
6052 ds_put_format(ds, "%c %s=%s", i ? ',' : ':',
6053 node->key, node->value);
6057 smap_destroy(&config);
6059 ds_put_char(ds, ')');
6060 ds_put_char(ds, '\n');
6064 shash_destroy(&ofproto_shash);
6069 ofproto_unixctl_dpif_show(struct unixctl_conn *conn, int argc OVS_UNUSED,
6070 const char *argv[] OVS_UNUSED, void *aux OVS_UNUSED)
6072 struct ds ds = DS_EMPTY_INITIALIZER;
6073 const struct shash_node **backers;
6076 backers = shash_sort(&all_dpif_backers);
6077 for (i = 0; i < shash_count(&all_dpif_backers); i++) {
6078 dpif_show_backer(backers[i]->data, &ds);
6082 unixctl_command_reply(conn, ds_cstr(&ds));
6086 /* Dump the megaflow (facet) cache. This is useful to check the
6087 * correctness of flow wildcarding, since the same mechanism is used for
6088 * both xlate caching and kernel wildcarding.
6090 * It's important to note that in the output the flow description uses
6091 * OpenFlow (OFP) ports, but the actions use datapath (ODP) ports.
6093 * This command is only needed for advanced debugging, so it's not
6094 * documented in the man page. */
6096 ofproto_unixctl_dpif_dump_megaflows(struct unixctl_conn *conn,
6097 int argc OVS_UNUSED, const char *argv[],
6098 void *aux OVS_UNUSED)
6100 struct ds ds = DS_EMPTY_INITIALIZER;
6101 const struct ofproto_dpif *ofproto;
6102 long long int now = time_msec();
6103 struct cls_cursor cursor;
6104 struct facet *facet;
6106 ofproto = ofproto_dpif_lookup(argv[1]);
6108 unixctl_command_reply_error(conn, "no such bridge");
6112 cls_cursor_init(&cursor, &ofproto->facets, NULL);
6113 CLS_CURSOR_FOR_EACH (facet, cr, &cursor) {
6114 cls_rule_format(&facet->cr, &ds);
6115 ds_put_cstr(&ds, ", ");
6116 ds_put_format(&ds, "n_subfacets:%zu, ", list_size(&facet->subfacets));
6117 ds_put_format(&ds, "used:%.3fs, ", (now - facet->used) / 1000.0);
6118 ds_put_cstr(&ds, "Datapath actions: ");
6119 if (facet->xout.slow) {
6120 uint64_t slow_path_stub[128 / 8];
6121 const struct nlattr *actions;
6124 compose_slow_path(ofproto, &facet->flow, facet->xout.slow,
6125 slow_path_stub, sizeof slow_path_stub,
6126 &actions, &actions_len);
6127 format_odp_actions(&ds, actions, actions_len);
6129 format_odp_actions(&ds, facet->xout.odp_actions.data,
6130 facet->xout.odp_actions.size);
6132 ds_put_cstr(&ds, "\n");
6135 ds_chomp(&ds, '\n');
6136 unixctl_command_reply(conn, ds_cstr(&ds));
6140 /* Disable using the megaflows.
6142 * This command is only needed for advanced debugging, so it's not
6143 * documented in the man page. */
6145 ofproto_unixctl_dpif_disable_megaflows(struct unixctl_conn *conn,
6146 int argc OVS_UNUSED,
6147 const char *argv[] OVS_UNUSED,
6148 void *aux OVS_UNUSED)
6150 struct ofproto_dpif *ofproto;
6152 enable_megaflows = false;
6154 HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
6155 flush(&ofproto->up);
6158 unixctl_command_reply(conn, "megaflows disabled");
6161 /* Re-enable using megaflows.
6163 * This command is only needed for advanced debugging, so it's not
6164 * documented in the man page. */
6166 ofproto_unixctl_dpif_enable_megaflows(struct unixctl_conn *conn,
6167 int argc OVS_UNUSED,
6168 const char *argv[] OVS_UNUSED,
6169 void *aux OVS_UNUSED)
6171 struct ofproto_dpif *ofproto;
6173 enable_megaflows = true;
6175 HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
6176 flush(&ofproto->up);
6179 unixctl_command_reply(conn, "megaflows enabled");
6183 ofproto_unixctl_dpif_dump_flows(struct unixctl_conn *conn,
6184 int argc OVS_UNUSED, const char *argv[],
6185 void *aux OVS_UNUSED)
6187 struct ds ds = DS_EMPTY_INITIALIZER;
6188 const struct ofproto_dpif *ofproto;
6189 struct subfacet *subfacet;
6191 ofproto = ofproto_dpif_lookup(argv[1]);
6193 unixctl_command_reply_error(conn, "no such bridge");
6197 update_stats(ofproto->backer);
6199 HMAP_FOR_EACH (subfacet, hmap_node, &ofproto->backer->subfacets) {
6200 struct facet *facet = subfacet->facet;
6202 if (ofproto_dpif_cast(facet->rule->up.ofproto) != ofproto) {
6206 odp_flow_key_format(subfacet->key, subfacet->key_len, &ds);
6208 ds_put_format(&ds, ", packets:%"PRIu64", bytes:%"PRIu64", used:",
6209 subfacet->dp_packet_count, subfacet->dp_byte_count);
6210 if (subfacet->used) {
6211 ds_put_format(&ds, "%.3fs",
6212 (time_msec() - subfacet->used) / 1000.0);
6214 ds_put_format(&ds, "never");
6216 if (subfacet->facet->tcp_flags) {
6217 ds_put_cstr(&ds, ", flags:");
6218 packet_format_tcp_flags(&ds, subfacet->facet->tcp_flags);
6221 ds_put_cstr(&ds, ", actions:");
6222 if (facet->xout.slow) {
6223 uint64_t slow_path_stub[128 / 8];
6224 const struct nlattr *actions;
6227 compose_slow_path(ofproto, &facet->flow, facet->xout.slow,
6228 slow_path_stub, sizeof slow_path_stub,
6229 &actions, &actions_len);
6230 format_odp_actions(&ds, actions, actions_len);
6232 format_odp_actions(&ds, facet->xout.odp_actions.data,
6233 facet->xout.odp_actions.size);
6235 ds_put_char(&ds, '\n');
6238 unixctl_command_reply(conn, ds_cstr(&ds));
6243 ofproto_unixctl_dpif_del_flows(struct unixctl_conn *conn,
6244 int argc OVS_UNUSED, const char *argv[],
6245 void *aux OVS_UNUSED)
6247 struct ds ds = DS_EMPTY_INITIALIZER;
6248 struct ofproto_dpif *ofproto;
6250 ofproto = ofproto_dpif_lookup(argv[1]);
6252 unixctl_command_reply_error(conn, "no such bridge");
6256 flush(&ofproto->up);
6258 unixctl_command_reply(conn, ds_cstr(&ds));
6263 ofproto_dpif_unixctl_init(void)
6265 static bool registered;
6271 unixctl_command_register(
6273 "[dp_name]|bridge odp_flow|br_flow [-generate|packet]",
6274 1, 3, ofproto_unixctl_trace, NULL);
6275 unixctl_command_register("fdb/flush", "[bridge]", 0, 1,
6276 ofproto_unixctl_fdb_flush, NULL);
6277 unixctl_command_register("fdb/show", "bridge", 1, 1,
6278 ofproto_unixctl_fdb_show, NULL);
6279 unixctl_command_register("ofproto/clog", "", 0, 0,
6280 ofproto_dpif_clog, NULL);
6281 unixctl_command_register("ofproto/unclog", "", 0, 0,
6282 ofproto_dpif_unclog, NULL);
6283 unixctl_command_register("ofproto/self-check", "[bridge]", 0, 1,
6284 ofproto_dpif_self_check, NULL);
6285 unixctl_command_register("dpif/dump-dps", "", 0, 0,
6286 ofproto_unixctl_dpif_dump_dps, NULL);
6287 unixctl_command_register("dpif/show", "", 0, 0, ofproto_unixctl_dpif_show,
6289 unixctl_command_register("dpif/dump-flows", "bridge", 1, 1,
6290 ofproto_unixctl_dpif_dump_flows, NULL);
6291 unixctl_command_register("dpif/del-flows", "bridge", 1, 1,
6292 ofproto_unixctl_dpif_del_flows, NULL);
6293 unixctl_command_register("dpif/dump-megaflows", "bridge", 1, 1,
6294 ofproto_unixctl_dpif_dump_megaflows, NULL);
6295 unixctl_command_register("dpif/disable-megaflows", "", 0, 0,
6296 ofproto_unixctl_dpif_disable_megaflows, NULL);
6297 unixctl_command_register("dpif/enable-megaflows", "", 0, 0,
6298 ofproto_unixctl_dpif_enable_megaflows, NULL);
6301 /* Linux VLAN device support (e.g. "eth0.10" for VLAN 10.)
6303 * This is deprecated. It is only for compatibility with broken device drivers
6304 * in old versions of Linux that do not properly support VLANs when VLAN
6305 * devices are not used. When broken device drivers are no longer in
6306 * widespread use, we will delete these interfaces. */
6309 set_realdev(struct ofport *ofport_, ofp_port_t realdev_ofp_port, int vid)
6311 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport_->ofproto);
6312 struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
6314 if (realdev_ofp_port == ofport->realdev_ofp_port
6315 && vid == ofport->vlandev_vid) {
6319 ofproto->backer->need_revalidate = REV_RECONFIGURE;
6321 if (ofport->realdev_ofp_port) {
6324 if (realdev_ofp_port && ofport->bundle) {
6325 /* vlandevs are enslaved to their realdevs, so they are not allowed to
6326 * themselves be part of a bundle. */
6327 bundle_set(ofport->up.ofproto, ofport->bundle, NULL);
6330 ofport->realdev_ofp_port = realdev_ofp_port;
6331 ofport->vlandev_vid = vid;
6333 if (realdev_ofp_port) {
6334 vsp_add(ofport, realdev_ofp_port, vid);
6341 hash_realdev_vid(ofp_port_t realdev_ofp_port, int vid)
6343 return hash_2words(ofp_to_u16(realdev_ofp_port), vid);
6346 /* Returns the OFP port number of the Linux VLAN device that corresponds to
6347 * 'vlan_tci' on the network device with port number 'realdev_ofp_port' in
6348 * 'struct ofport_dpif'. For example, given 'realdev_ofp_port' of eth0 and
6349 * 'vlan_tci' 9, it would return the port number of eth0.9.
6351 * Unless VLAN splinters are enabled for port 'realdev_ofp_port', this
6352 * function just returns its 'realdev_ofp_port' argument. */
6354 vsp_realdev_to_vlandev(const struct ofproto_dpif *ofproto,
6355 ofp_port_t realdev_ofp_port, ovs_be16 vlan_tci)
6357 if (!hmap_is_empty(&ofproto->realdev_vid_map)) {
6358 int vid = vlan_tci_to_vid(vlan_tci);
6359 const struct vlan_splinter *vsp;
6361 HMAP_FOR_EACH_WITH_HASH (vsp, realdev_vid_node,
6362 hash_realdev_vid(realdev_ofp_port, vid),
6363 &ofproto->realdev_vid_map) {
6364 if (vsp->realdev_ofp_port == realdev_ofp_port
6365 && vsp->vid == vid) {
6366 return vsp->vlandev_ofp_port;
6370 return realdev_ofp_port;
6373 static struct vlan_splinter *
6374 vlandev_find(const struct ofproto_dpif *ofproto, ofp_port_t vlandev_ofp_port)
6376 struct vlan_splinter *vsp;
6378 HMAP_FOR_EACH_WITH_HASH (vsp, vlandev_node,
6379 hash_ofp_port(vlandev_ofp_port),
6380 &ofproto->vlandev_map) {
6381 if (vsp->vlandev_ofp_port == vlandev_ofp_port) {
6389 /* Returns the OpenFlow port number of the "real" device underlying the Linux
6390 * VLAN device with OpenFlow port number 'vlandev_ofp_port' and stores the
6391 * VLAN VID of the Linux VLAN device in '*vid'. For example, given
6392 * 'vlandev_ofp_port' of eth0.9, it would return the OpenFlow port number of
6393 * eth0 and store 9 in '*vid'.
6395 * Returns 0 and does not modify '*vid' if 'vlandev_ofp_port' is not a Linux
6396 * VLAN device. Unless VLAN splinters are enabled, this is what this function
6399 vsp_vlandev_to_realdev(const struct ofproto_dpif *ofproto,
6400 ofp_port_t vlandev_ofp_port, int *vid)
6402 if (!hmap_is_empty(&ofproto->vlandev_map)) {
6403 const struct vlan_splinter *vsp;
6405 vsp = vlandev_find(ofproto, vlandev_ofp_port);
6410 return vsp->realdev_ofp_port;
6416 /* Given 'flow', a flow representing a packet received on 'ofproto', checks
6417 * whether 'flow->in_port' represents a Linux VLAN device. If so, changes
6418 * 'flow->in_port' to the "real" device backing the VLAN device, sets
6419 * 'flow->vlan_tci' to the VLAN VID, and returns true. Otherwise (which is
6420 * always the case unless VLAN splinters are enabled), returns false without
6421 * making any changes. */
6423 vsp_adjust_flow(const struct ofproto_dpif *ofproto, struct flow *flow)
6428 realdev = vsp_vlandev_to_realdev(ofproto, flow->in_port.ofp_port, &vid);
6433 /* Cause the flow to be processed as if it came in on the real device with
6434 * the VLAN device's VLAN ID. */
6435 flow->in_port.ofp_port = realdev;
6436 flow->vlan_tci = htons((vid & VLAN_VID_MASK) | VLAN_CFI);
6441 vsp_remove(struct ofport_dpif *port)
6443 struct ofproto_dpif *ofproto = ofproto_dpif_cast(port->up.ofproto);
6444 struct vlan_splinter *vsp;
6446 vsp = vlandev_find(ofproto, port->up.ofp_port);
6448 hmap_remove(&ofproto->vlandev_map, &vsp->vlandev_node);
6449 hmap_remove(&ofproto->realdev_vid_map, &vsp->realdev_vid_node);
6452 port->realdev_ofp_port = 0;
6454 VLOG_ERR("missing vlan device record");
6459 vsp_add(struct ofport_dpif *port, ofp_port_t realdev_ofp_port, int vid)
6461 struct ofproto_dpif *ofproto = ofproto_dpif_cast(port->up.ofproto);
6463 if (!vsp_vlandev_to_realdev(ofproto, port->up.ofp_port, NULL)
6464 && (vsp_realdev_to_vlandev(ofproto, realdev_ofp_port, htons(vid))
6465 == realdev_ofp_port)) {
6466 struct vlan_splinter *vsp;
6468 vsp = xmalloc(sizeof *vsp);
6469 hmap_insert(&ofproto->vlandev_map, &vsp->vlandev_node,
6470 hash_ofp_port(port->up.ofp_port));
6471 hmap_insert(&ofproto->realdev_vid_map, &vsp->realdev_vid_node,
6472 hash_realdev_vid(realdev_ofp_port, vid));
6473 vsp->realdev_ofp_port = realdev_ofp_port;
6474 vsp->vlandev_ofp_port = port->up.ofp_port;
6477 port->realdev_ofp_port = realdev_ofp_port;
6479 VLOG_ERR("duplicate vlan device record");
6484 ofp_port_to_odp_port(const struct ofproto_dpif *ofproto, ofp_port_t ofp_port)
6486 const struct ofport_dpif *ofport = get_ofp_port(ofproto, ofp_port);
6487 return ofport ? ofport->odp_port : ODPP_NONE;
6490 static struct ofport_dpif *
6491 odp_port_to_ofport(const struct dpif_backer *backer, odp_port_t odp_port)
6493 struct ofport_dpif *port;
6495 HMAP_FOR_EACH_IN_BUCKET (port, odp_port_node, hash_odp_port(odp_port),
6496 &backer->odp_to_ofport_map) {
6497 if (port->odp_port == odp_port) {
6506 odp_port_to_ofp_port(const struct ofproto_dpif *ofproto, odp_port_t odp_port)
6508 struct ofport_dpif *port;
6510 port = odp_port_to_ofport(ofproto->backer, odp_port);
6511 if (port && &ofproto->up == port->up.ofproto) {
6512 return port->up.ofp_port;
6518 /* Compute exponentially weighted moving average, adding 'new' as the newest,
6519 * most heavily weighted element. 'base' designates the rate of decay: after
6520 * 'base' further updates, 'new''s weight in the EWMA decays to about 1/e
6523 exp_mavg(double *avg, int base, double new)
6525 *avg = (*avg * (base - 1) + new) / base;
6529 update_moving_averages(struct dpif_backer *backer)
6531 const int min_ms = 60 * 1000; /* milliseconds in one minute. */
6532 long long int minutes = (time_msec() - backer->created) / min_ms;
6535 backer->lifetime.add_rate = (double) backer->total_subfacet_add_count
6537 backer->lifetime.del_rate = (double) backer->total_subfacet_del_count
6540 backer->lifetime.add_rate = 0.0;
6541 backer->lifetime.del_rate = 0.0;
6544 /* Update hourly averages on the minute boundaries. */
6545 if (time_msec() - backer->last_minute >= min_ms) {
6546 exp_mavg(&backer->hourly.add_rate, 60, backer->subfacet_add_count);
6547 exp_mavg(&backer->hourly.del_rate, 60, backer->subfacet_del_count);
6549 /* Update daily averages on the hour boundaries. */
6550 if ((backer->last_minute - backer->created) / min_ms % 60 == 59) {
6551 exp_mavg(&backer->daily.add_rate, 24, backer->hourly.add_rate);
6552 exp_mavg(&backer->daily.del_rate, 24, backer->hourly.del_rate);
6555 backer->total_subfacet_add_count += backer->subfacet_add_count;
6556 backer->total_subfacet_del_count += backer->subfacet_del_count;
6557 backer->subfacet_add_count = 0;
6558 backer->subfacet_del_count = 0;
6559 backer->last_minute += min_ms;
6563 const struct ofproto_class ofproto_dpif_class = {
6598 port_is_lacp_current,
6599 NULL, /* rule_choose_table */
6606 rule_modify_actions,
6620 get_stp_port_status,
6627 is_mirror_output_bundle,
6628 forward_bpdu_changed,
6629 set_mac_table_config,
6631 NULL, /* meter_get_features */
6632 NULL, /* meter_set */
6633 NULL, /* meter_get */
6634 NULL, /* meter_del */