2 * Copyright (c) 2009, 2010, 2011, 2012, 2013 Nicira, Inc.
4 * Licensed under the Apache License, Version 2.0 (the "License");
5 * you may not use this file except in compliance with the License.
6 * You may obtain a copy of the License at:
8 * http://www.apache.org/licenses/LICENSE-2.0
10 * Unless required by applicable law or agreed to in writing, software
11 * distributed under the License is distributed on an "AS IS" BASIS,
12 * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
13 * See the License for the specific language governing permissions and
14 * limitations under the License.
19 #include "ofproto/ofproto-dpif.h"
20 #include "ofproto/ofproto-provider.h"
27 #include "byte-order.h"
32 #include "dynamic-string.h"
33 #include "fail-open.h"
37 #include "mac-learning.h"
38 #include "meta-flow.h"
39 #include "multipath.h"
40 #include "netdev-vport.h"
45 #include "odp-execute.h"
48 #include "ofp-actions.h"
49 #include "ofp-parse.h"
50 #include "ofp-print.h"
51 #include "ofproto-dpif-governor.h"
52 #include "ofproto-dpif-ipfix.h"
53 #include "ofproto-dpif-sflow.h"
54 #include "ofproto-dpif-xlate.h"
55 #include "poll-loop.h"
60 #include "unaligned.h"
62 #include "vlan-bitmap.h"
65 VLOG_DEFINE_THIS_MODULE(ofproto_dpif);
67 COVERAGE_DEFINE(ofproto_dpif_expired);
68 COVERAGE_DEFINE(facet_changed_rule);
69 COVERAGE_DEFINE(facet_revalidate);
70 COVERAGE_DEFINE(facet_unexpected);
71 COVERAGE_DEFINE(facet_suppress);
76 static struct rule_dpif *rule_dpif_lookup(struct ofproto_dpif *,
78 struct flow_wildcards *wc);
80 static void rule_get_stats(struct rule *, uint64_t *packets, uint64_t *bytes);
81 static void rule_invalidate(const struct rule_dpif *);
83 static void mirror_destroy(struct ofmirror *);
84 static void update_mirror_stats(struct ofproto_dpif *ofproto,
85 mirror_mask_t mirrors,
86 uint64_t packets, uint64_t bytes);
88 static void bundle_remove(struct ofport *);
89 static void bundle_update(struct ofbundle *);
90 static void bundle_destroy(struct ofbundle *);
91 static void bundle_del_port(struct ofport_dpif *);
92 static void bundle_run(struct ofbundle *);
93 static void bundle_wait(struct ofbundle *);
95 static void stp_run(struct ofproto_dpif *ofproto);
96 static void stp_wait(struct ofproto_dpif *ofproto);
97 static int set_stp_port(struct ofport *,
98 const struct ofproto_port_stp_settings *);
100 static void compose_slow_path(const struct ofproto_dpif *, const struct flow *,
101 enum slow_path_reason,
102 uint64_t *stub, size_t stub_size,
103 const struct nlattr **actionsp,
104 size_t *actions_lenp);
106 /* A subfacet (see "struct subfacet" below) has three possible installation
109 * - SF_NOT_INSTALLED: Not installed in the datapath. This will only be the
110 * case just after the subfacet is created, just before the subfacet is
111 * destroyed, or if the datapath returns an error when we try to install a
114 * - SF_FAST_PATH: The subfacet's actions are installed in the datapath.
116 * - SF_SLOW_PATH: An action that sends every packet for the subfacet through
117 * ofproto_dpif is installed in the datapath.
120 SF_NOT_INSTALLED, /* No datapath flow for this subfacet. */
121 SF_FAST_PATH, /* Full actions are installed. */
122 SF_SLOW_PATH, /* Send-to-userspace action is installed. */
125 /* A dpif flow and actions associated with a facet.
127 * See also the large comment on struct facet. */
130 struct hmap_node hmap_node; /* In struct ofproto_dpif 'subfacets' list. */
131 struct list list_node; /* In struct facet's 'facets' list. */
132 struct facet *facet; /* Owning facet. */
133 struct dpif_backer *backer; /* Owning backer. */
135 enum odp_key_fitness key_fitness;
139 long long int used; /* Time last used; time created if not used. */
140 long long int created; /* Time created. */
142 uint64_t dp_packet_count; /* Last known packet count in the datapath. */
143 uint64_t dp_byte_count; /* Last known byte count in the datapath. */
145 enum subfacet_path path; /* Installed in datapath? */
148 #define SUBFACET_DESTROY_MAX_BATCH 50
150 static struct subfacet *subfacet_create(struct facet *, struct flow_miss *miss,
152 static struct subfacet *subfacet_find(struct dpif_backer *,
153 const struct nlattr *key, size_t key_len,
155 static void subfacet_destroy(struct subfacet *);
156 static void subfacet_destroy__(struct subfacet *);
157 static void subfacet_destroy_batch(struct dpif_backer *,
158 struct subfacet **, int n);
159 static void subfacet_reset_dp_stats(struct subfacet *,
160 struct dpif_flow_stats *);
161 static void subfacet_update_stats(struct subfacet *,
162 const struct dpif_flow_stats *);
163 static int subfacet_install(struct subfacet *,
164 const struct ofpbuf *odp_actions,
165 struct dpif_flow_stats *);
166 static void subfacet_uninstall(struct subfacet *);
168 /* A unique, non-overlapping instantiation of an OpenFlow flow.
170 * A facet associates a "struct flow", which represents the Open vSwitch
171 * userspace idea of an exact-match flow, with one or more subfacets.
172 * While the facet is created based on an exact-match flow, it is stored
173 * within the ofproto based on the wildcards that could be expressed
174 * based on the flow table and other configuration. (See the 'wc'
175 * description in "struct xlate_out" for more details.)
177 * Each subfacet tracks the datapath's idea of the flow equivalent to
178 * the facet. When the kernel module (or other dpif implementation) and
179 * Open vSwitch userspace agree on the definition of a flow key, there
180 * is exactly one subfacet per facet. If the dpif implementation
181 * supports more-specific flow matching than userspace, however, a facet
182 * can have more than one subfacet. Examples include the dpif
183 * implementation not supporting the same wildcards as userspace or some
184 * distinction in flow that userspace simply doesn't understand.
186 * Flow expiration works in terms of subfacets, so a facet must have at
187 * least one subfacet or it will never expire, leaking memory. */
190 struct hmap_node hmap_node; /* In owning ofproto's 'facets' hmap. */
191 struct list list_node; /* In owning rule's 'facets' list. */
192 struct rule_dpif *rule; /* Owning rule. */
195 struct list subfacets;
196 long long int used; /* Time last used; time created if not used. */
199 struct flow flow; /* Flow of the creating subfacet. */
200 struct cls_rule cr; /* In 'ofproto_dpif's facets classifier. */
204 * - Do include packets and bytes sent "by hand", e.g. with
207 * - Do include packets and bytes that were obtained from the datapath
208 * when a subfacet's statistics were reset (e.g. dpif_flow_put() with
209 * DPIF_FP_ZERO_STATS).
211 * - Do not include packets or bytes that can be obtained from the
212 * datapath for any existing subfacet.
214 uint64_t packet_count; /* Number of packets received. */
215 uint64_t byte_count; /* Number of bytes received. */
217 /* Resubmit statistics. */
218 uint64_t prev_packet_count; /* Number of packets from last stats push. */
219 uint64_t prev_byte_count; /* Number of bytes from last stats push. */
220 long long int prev_used; /* Used time from last stats push. */
223 uint64_t accounted_bytes; /* Bytes processed by facet_account(). */
224 struct netflow_flow nf_flow; /* Per-flow NetFlow tracking data. */
225 uint8_t tcp_flags; /* TCP flags seen for this 'rule'. */
227 struct xlate_out xout;
229 /* Storage for a single subfacet, to reduce malloc() time and space
230 * overhead. (A facet always has at least one subfacet and in the common
231 * case has exactly one subfacet. However, 'one_subfacet' may not
232 * always be valid, since it could have been removed after newer
233 * subfacets were pushed onto the 'subfacets' list.) */
234 struct subfacet one_subfacet;
236 long long int learn_rl; /* Rate limiter for facet_learn(). */
239 static struct facet *facet_create(const struct flow_miss *, struct rule_dpif *,
241 struct dpif_flow_stats *);
242 static void facet_remove(struct facet *);
243 static void facet_free(struct facet *);
245 static struct facet *facet_find(struct ofproto_dpif *, const struct flow *);
246 static struct facet *facet_lookup_valid(struct ofproto_dpif *,
247 const struct flow *);
248 static bool facet_revalidate(struct facet *);
249 static bool facet_check_consistency(struct facet *);
251 static void facet_flush_stats(struct facet *);
253 static void facet_reset_counters(struct facet *);
254 static void facet_push_stats(struct facet *, bool may_learn);
255 static void facet_learn(struct facet *);
256 static void facet_account(struct facet *);
257 static void push_all_stats(void);
259 static bool facet_is_controller_flow(struct facet *);
261 /* Node in 'ofport_dpif''s 'priorities' map. Used to maintain a map from
262 * 'priority' (the datapath's term for QoS queue) to the dscp bits which all
263 * traffic egressing the 'ofport' with that priority should be marked with. */
264 struct priority_to_dscp {
265 struct hmap_node hmap_node; /* Node in 'ofport_dpif''s 'priorities' map. */
266 uint32_t priority; /* Priority of this queue (see struct flow). */
268 uint8_t dscp; /* DSCP bits to mark outgoing traffic with. */
271 /* Linux VLAN device support (e.g. "eth0.10" for VLAN 10.)
273 * This is deprecated. It is only for compatibility with broken device drivers
274 * in old versions of Linux that do not properly support VLANs when VLAN
275 * devices are not used. When broken device drivers are no longer in
276 * widespread use, we will delete these interfaces. */
277 struct vlan_splinter {
278 struct hmap_node realdev_vid_node;
279 struct hmap_node vlandev_node;
280 ofp_port_t realdev_ofp_port;
281 ofp_port_t vlandev_ofp_port;
285 static bool vsp_adjust_flow(const struct ofproto_dpif *, struct flow *);
286 static void vsp_remove(struct ofport_dpif *);
287 static void vsp_add(struct ofport_dpif *, ofp_port_t realdev_ofp_port, int vid);
289 static ofp_port_t odp_port_to_ofp_port(const struct ofproto_dpif *,
290 odp_port_t odp_port);
292 static struct ofport_dpif *
293 ofport_dpif_cast(const struct ofport *ofport)
295 return ofport ? CONTAINER_OF(ofport, struct ofport_dpif, up) : NULL;
298 static void port_run(struct ofport_dpif *);
299 static void port_run_fast(struct ofport_dpif *);
300 static void port_wait(struct ofport_dpif *);
301 static int set_bfd(struct ofport *, const struct smap *);
302 static int set_cfm(struct ofport *, const struct cfm_settings *);
303 static void ofport_clear_priorities(struct ofport_dpif *);
304 static void ofport_update_peer(struct ofport_dpif *);
305 static void run_fast_rl(void);
307 struct dpif_completion {
308 struct list list_node;
309 struct ofoperation *op;
312 /* Reasons that we might need to revalidate every facet, and corresponding
315 * A value of 0 means that there is no need to revalidate.
317 * It would be nice to have some cleaner way to integrate with coverage
318 * counters, but with only a few reasons I guess this is good enough for
320 enum revalidate_reason {
321 REV_RECONFIGURE = 1, /* Switch configuration changed. */
322 REV_STP, /* Spanning tree protocol port status change. */
323 REV_PORT_TOGGLED, /* Port enabled or disabled by CFM, LACP, ...*/
324 REV_FLOW_TABLE, /* Flow table changed. */
325 REV_INCONSISTENCY /* Facet self-check failed. */
327 COVERAGE_DEFINE(rev_reconfigure);
328 COVERAGE_DEFINE(rev_stp);
329 COVERAGE_DEFINE(rev_port_toggled);
330 COVERAGE_DEFINE(rev_flow_table);
331 COVERAGE_DEFINE(rev_inconsistency);
333 /* Drop keys are odp flow keys which have drop flows installed in the kernel.
334 * These are datapath flows which have no associated ofproto, if they did we
335 * would use facets. */
337 struct hmap_node hmap_node;
342 struct avg_subfacet_rates {
343 double add_rate; /* Moving average of new flows created per minute. */
344 double del_rate; /* Moving average of flows deleted per minute. */
347 /* All datapaths of a given type share a single dpif backer instance. */
352 struct timer next_expiration;
353 struct hmap odp_to_ofport_map; /* ODP port to ofport mapping. */
355 struct simap tnl_backers; /* Set of dpif ports backing tunnels. */
357 /* Facet revalidation flags applying to facets which use this backer. */
358 enum revalidate_reason need_revalidate; /* Revalidate every facet. */
359 struct tag_set revalidate_set; /* Revalidate only matching facets. */
361 struct hmap drop_keys; /* Set of dropped odp keys. */
362 bool recv_set_enable; /* Enables or disables receiving packets. */
364 struct hmap subfacets;
365 struct governor *governor;
367 /* Subfacet statistics.
369 * These keep track of the total number of subfacets added and deleted and
370 * flow life span. They are useful for computing the flow rates stats
371 * exposed via "ovs-appctl dpif/show". The goal is to learn about
372 * traffic patterns in ways that we can use later to improve Open vSwitch
373 * performance in new situations. */
374 long long int created; /* Time when it is created. */
375 unsigned max_n_subfacet; /* Maximum number of flows */
376 unsigned avg_n_subfacet; /* Average number of flows. */
377 long long int avg_subfacet_life; /* Average life span of subfacets. */
379 /* The average number of subfacets... */
380 struct avg_subfacet_rates hourly; /* ...over the last hour. */
381 struct avg_subfacet_rates daily; /* ...over the last day. */
382 struct avg_subfacet_rates lifetime; /* ...over the switch lifetime. */
383 long long int last_minute; /* Last time 'hourly' was updated. */
385 /* Number of subfacets added or deleted since 'last_minute'. */
386 unsigned subfacet_add_count;
387 unsigned subfacet_del_count;
389 /* Number of subfacets added or deleted from 'created' to 'last_minute.' */
390 unsigned long long int total_subfacet_add_count;
391 unsigned long long int total_subfacet_del_count;
394 /* All existing ofproto_backer instances, indexed by ofproto->up.type. */
395 static struct shash all_dpif_backers = SHASH_INITIALIZER(&all_dpif_backers);
397 static void drop_key_clear(struct dpif_backer *);
398 static struct ofport_dpif *
399 odp_port_to_ofport(const struct dpif_backer *, odp_port_t odp_port);
400 static void update_moving_averages(struct dpif_backer *backer);
402 /* Defer flow mod completion until "ovs-appctl ofproto/unclog"? (Useful only
403 * for debugging the asynchronous flow_mod implementation.) */
406 /* All existing ofproto_dpif instances, indexed by ->up.name. */
407 static struct hmap all_ofproto_dpifs = HMAP_INITIALIZER(&all_ofproto_dpifs);
409 static void ofproto_dpif_unixctl_init(void);
412 #define FLOW_MISS_MAX_BATCH 50
413 static int handle_upcalls(struct dpif_backer *, unsigned int max_batch);
415 /* Flow expiration. */
416 static int expire(struct dpif_backer *);
419 static void send_netflow_active_timeouts(struct ofproto_dpif *);
422 static int send_packet(const struct ofport_dpif *, struct ofpbuf *packet);
424 /* Global variables. */
425 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
427 /* Initial mappings of port to bridge mappings. */
428 static struct shash init_ofp_ports = SHASH_INITIALIZER(&init_ofp_ports);
430 /* Factory functions. */
433 init(const struct shash *iface_hints)
435 struct shash_node *node;
437 /* Make a local copy, since we don't own 'iface_hints' elements. */
438 SHASH_FOR_EACH(node, iface_hints) {
439 const struct iface_hint *orig_hint = node->data;
440 struct iface_hint *new_hint = xmalloc(sizeof *new_hint);
442 new_hint->br_name = xstrdup(orig_hint->br_name);
443 new_hint->br_type = xstrdup(orig_hint->br_type);
444 new_hint->ofp_port = orig_hint->ofp_port;
446 shash_add(&init_ofp_ports, node->name, new_hint);
451 enumerate_types(struct sset *types)
453 dp_enumerate_types(types);
457 enumerate_names(const char *type, struct sset *names)
459 struct ofproto_dpif *ofproto;
462 HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
463 if (strcmp(type, ofproto->up.type)) {
466 sset_add(names, ofproto->up.name);
473 del(const char *type, const char *name)
478 error = dpif_open(name, type, &dpif);
480 error = dpif_delete(dpif);
487 port_open_type(const char *datapath_type, const char *port_type)
489 return dpif_port_open_type(datapath_type, port_type);
492 /* Type functions. */
494 static struct ofproto_dpif *
495 lookup_ofproto_dpif_by_port_name(const char *name)
497 struct ofproto_dpif *ofproto;
499 HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
500 if (sset_contains(&ofproto->ports, name)) {
509 type_run(const char *type)
511 static long long int push_timer = LLONG_MIN;
512 struct dpif_backer *backer;
516 backer = shash_find_data(&all_dpif_backers, type);
518 /* This is not necessarily a problem, since backers are only
519 * created on demand. */
523 dpif_run(backer->dpif);
525 /* The most natural place to push facet statistics is when they're pulled
526 * from the datapath. However, when there are many flows in the datapath,
527 * this expensive operation can occur so frequently, that it reduces our
528 * ability to quickly set up flows. To reduce the cost, we push statistics
530 if (time_msec() > push_timer) {
531 push_timer = time_msec() + 2000;
535 /* If vswitchd started with other_config:flow_restore_wait set as "true",
536 * and the configuration has now changed to "false", enable receiving
537 * packets from the datapath. */
538 if (!backer->recv_set_enable && !ofproto_get_flow_restore_wait()) {
539 backer->recv_set_enable = true;
541 error = dpif_recv_set(backer->dpif, backer->recv_set_enable);
543 VLOG_ERR("Failed to enable receiving packets in dpif.");
546 dpif_flow_flush(backer->dpif);
547 backer->need_revalidate = REV_RECONFIGURE;
550 if (backer->need_revalidate
551 || !tag_set_is_empty(&backer->revalidate_set)) {
552 struct tag_set revalidate_set = backer->revalidate_set;
553 bool need_revalidate = backer->need_revalidate;
554 struct ofproto_dpif *ofproto;
555 struct simap_node *node;
556 struct simap tmp_backers;
558 /* Handle tunnel garbage collection. */
559 simap_init(&tmp_backers);
560 simap_swap(&backer->tnl_backers, &tmp_backers);
562 HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
563 struct ofport_dpif *iter;
565 if (backer != ofproto->backer) {
569 HMAP_FOR_EACH (iter, up.hmap_node, &ofproto->up.ports) {
570 char namebuf[NETDEV_VPORT_NAME_BUFSIZE];
573 if (!iter->tnl_port) {
577 dp_port = netdev_vport_get_dpif_port(iter->up.netdev,
578 namebuf, sizeof namebuf);
579 node = simap_find(&tmp_backers, dp_port);
581 simap_put(&backer->tnl_backers, dp_port, node->data);
582 simap_delete(&tmp_backers, node);
583 node = simap_find(&backer->tnl_backers, dp_port);
585 node = simap_find(&backer->tnl_backers, dp_port);
587 odp_port_t odp_port = ODPP_NONE;
589 if (!dpif_port_add(backer->dpif, iter->up.netdev,
591 simap_put(&backer->tnl_backers, dp_port,
592 odp_to_u32(odp_port));
593 node = simap_find(&backer->tnl_backers, dp_port);
598 iter->odp_port = node ? u32_to_odp(node->data) : ODPP_NONE;
599 if (tnl_port_reconfigure(&iter->up, iter->odp_port,
601 backer->need_revalidate = REV_RECONFIGURE;
606 SIMAP_FOR_EACH (node, &tmp_backers) {
607 dpif_port_del(backer->dpif, u32_to_odp(node->data));
609 simap_destroy(&tmp_backers);
611 switch (backer->need_revalidate) {
612 case REV_RECONFIGURE: COVERAGE_INC(rev_reconfigure); break;
613 case REV_STP: COVERAGE_INC(rev_stp); break;
614 case REV_PORT_TOGGLED: COVERAGE_INC(rev_port_toggled); break;
615 case REV_FLOW_TABLE: COVERAGE_INC(rev_flow_table); break;
616 case REV_INCONSISTENCY: COVERAGE_INC(rev_inconsistency); break;
619 if (backer->need_revalidate) {
620 /* Clear the drop_keys in case we should now be accepting some
621 * formerly dropped flows. */
622 drop_key_clear(backer);
625 /* Clear the revalidation flags. */
626 tag_set_init(&backer->revalidate_set);
627 backer->need_revalidate = 0;
629 HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
630 struct facet *facet, *next;
631 struct cls_cursor cursor;
633 if (ofproto->backer != backer) {
637 cls_cursor_init(&cursor, &ofproto->facets, NULL);
638 CLS_CURSOR_FOR_EACH_SAFE (facet, next, cr, &cursor) {
640 || tag_set_intersects(&revalidate_set, facet->xout.tags)) {
641 facet_revalidate(facet);
648 if (!backer->recv_set_enable) {
649 /* Wake up before a max of 1000ms. */
650 timer_set_duration(&backer->next_expiration, 1000);
651 } else if (timer_expired(&backer->next_expiration)) {
652 int delay = expire(backer);
653 timer_set_duration(&backer->next_expiration, delay);
656 /* Check for port changes in the dpif. */
657 while ((error = dpif_port_poll(backer->dpif, &devname)) == 0) {
658 struct ofproto_dpif *ofproto;
659 struct dpif_port port;
661 /* Don't report on the datapath's device. */
662 if (!strcmp(devname, dpif_base_name(backer->dpif))) {
666 HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node,
667 &all_ofproto_dpifs) {
668 if (simap_contains(&ofproto->backer->tnl_backers, devname)) {
673 ofproto = lookup_ofproto_dpif_by_port_name(devname);
674 if (dpif_port_query_by_name(backer->dpif, devname, &port)) {
675 /* The port was removed. If we know the datapath,
676 * report it through poll_set(). If we don't, it may be
677 * notifying us of a removal we initiated, so ignore it.
678 * If there's a pending ENOBUFS, let it stand, since
679 * everything will be reevaluated. */
680 if (ofproto && ofproto->port_poll_errno != ENOBUFS) {
681 sset_add(&ofproto->port_poll_set, devname);
682 ofproto->port_poll_errno = 0;
684 } else if (!ofproto) {
685 /* The port was added, but we don't know with which
686 * ofproto we should associate it. Delete it. */
687 dpif_port_del(backer->dpif, port.port_no);
689 dpif_port_destroy(&port);
695 if (error != EAGAIN) {
696 struct ofproto_dpif *ofproto;
698 /* There was some sort of error, so propagate it to all
699 * ofprotos that use this backer. */
700 HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node,
701 &all_ofproto_dpifs) {
702 if (ofproto->backer == backer) {
703 sset_clear(&ofproto->port_poll_set);
704 ofproto->port_poll_errno = error;
709 if (backer->governor) {
712 governor_run(backer->governor);
714 /* If the governor has shrunk to its minimum size and the number of
715 * subfacets has dwindled, then drop the governor entirely.
717 * For hysteresis, the number of subfacets to drop the governor is
718 * smaller than the number needed to trigger its creation. */
719 n_subfacets = hmap_count(&backer->subfacets);
720 if (n_subfacets * 4 < flow_eviction_threshold
721 && governor_is_idle(backer->governor)) {
722 governor_destroy(backer->governor);
723 backer->governor = NULL;
731 dpif_backer_run_fast(struct dpif_backer *backer, int max_batch)
735 /* If recv_set_enable is false, we should not handle upcalls. */
736 if (!backer->recv_set_enable) {
740 /* Handle one or more batches of upcalls, until there's nothing left to do
741 * or until we do a fixed total amount of work.
743 * We do work in batches because it can be much cheaper to set up a number
744 * of flows and fire off their patches all at once. We do multiple batches
745 * because in some cases handling a packet can cause another packet to be
746 * queued almost immediately as part of the return flow. Both
747 * optimizations can make major improvements on some benchmarks and
748 * presumably for real traffic as well. */
750 while (work < max_batch) {
751 int retval = handle_upcalls(backer, max_batch - work);
762 type_run_fast(const char *type)
764 struct dpif_backer *backer;
766 backer = shash_find_data(&all_dpif_backers, type);
768 /* This is not necessarily a problem, since backers are only
769 * created on demand. */
773 return dpif_backer_run_fast(backer, FLOW_MISS_MAX_BATCH);
779 static long long int port_rl = LLONG_MIN;
780 static unsigned int backer_rl = 0;
782 if (time_msec() >= port_rl) {
783 struct ofproto_dpif *ofproto;
784 struct ofport_dpif *ofport;
786 HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
788 HMAP_FOR_EACH (ofport, up.hmap_node, &ofproto->up.ports) {
789 port_run_fast(ofport);
792 port_rl = time_msec() + 200;
795 /* XXX: We have to be careful not to do too much work in this function. If
796 * we call dpif_backer_run_fast() too often, or with too large a batch,
797 * performance improves signifcantly, but at a cost. It's possible for the
798 * number of flows in the datapath to increase without bound, and for poll
799 * loops to take 10s of seconds. The correct solution to this problem,
800 * long term, is to separate flow miss handling into it's own thread so it
801 * isn't affected by revalidations, and expirations. Until then, this is
802 * the best we can do. */
803 if (++backer_rl >= 10) {
804 struct shash_node *node;
807 SHASH_FOR_EACH (node, &all_dpif_backers) {
808 dpif_backer_run_fast(node->data, 1);
814 type_wait(const char *type)
816 struct dpif_backer *backer;
818 backer = shash_find_data(&all_dpif_backers, type);
820 /* This is not necessarily a problem, since backers are only
821 * created on demand. */
825 if (backer->governor) {
826 governor_wait(backer->governor);
829 timer_wait(&backer->next_expiration);
832 /* Basic life-cycle. */
834 static int add_internal_flows(struct ofproto_dpif *);
836 static struct ofproto *
839 struct ofproto_dpif *ofproto = xmalloc(sizeof *ofproto);
844 dealloc(struct ofproto *ofproto_)
846 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
851 close_dpif_backer(struct dpif_backer *backer)
853 struct shash_node *node;
855 ovs_assert(backer->refcount > 0);
857 if (--backer->refcount) {
861 drop_key_clear(backer);
862 hmap_destroy(&backer->drop_keys);
864 simap_destroy(&backer->tnl_backers);
865 hmap_destroy(&backer->odp_to_ofport_map);
866 node = shash_find(&all_dpif_backers, backer->type);
868 shash_delete(&all_dpif_backers, node);
869 dpif_close(backer->dpif);
871 ovs_assert(hmap_is_empty(&backer->subfacets));
872 hmap_destroy(&backer->subfacets);
873 governor_destroy(backer->governor);
878 /* Datapath port slated for removal from datapath. */
880 struct list list_node;
885 open_dpif_backer(const char *type, struct dpif_backer **backerp)
887 struct dpif_backer *backer;
888 struct dpif_port_dump port_dump;
889 struct dpif_port port;
890 struct shash_node *node;
891 struct list garbage_list;
892 struct odp_garbage *garbage, *next;
898 backer = shash_find_data(&all_dpif_backers, type);
905 backer_name = xasprintf("ovs-%s", type);
907 /* Remove any existing datapaths, since we assume we're the only
908 * userspace controlling the datapath. */
910 dp_enumerate_names(type, &names);
911 SSET_FOR_EACH(name, &names) {
912 struct dpif *old_dpif;
914 /* Don't remove our backer if it exists. */
915 if (!strcmp(name, backer_name)) {
919 if (dpif_open(name, type, &old_dpif)) {
920 VLOG_WARN("couldn't open old datapath %s to remove it", name);
922 dpif_delete(old_dpif);
923 dpif_close(old_dpif);
926 sset_destroy(&names);
928 backer = xmalloc(sizeof *backer);
930 error = dpif_create_and_open(backer_name, type, &backer->dpif);
933 VLOG_ERR("failed to open datapath of type %s: %s", type,
939 backer->type = xstrdup(type);
940 backer->governor = NULL;
941 backer->refcount = 1;
942 hmap_init(&backer->odp_to_ofport_map);
943 hmap_init(&backer->drop_keys);
944 hmap_init(&backer->subfacets);
945 timer_set_duration(&backer->next_expiration, 1000);
946 backer->need_revalidate = 0;
947 simap_init(&backer->tnl_backers);
948 tag_set_init(&backer->revalidate_set);
949 backer->recv_set_enable = !ofproto_get_flow_restore_wait();
952 if (backer->recv_set_enable) {
953 dpif_flow_flush(backer->dpif);
956 /* Loop through the ports already on the datapath and remove any
957 * that we don't need anymore. */
958 list_init(&garbage_list);
959 dpif_port_dump_start(&port_dump, backer->dpif);
960 while (dpif_port_dump_next(&port_dump, &port)) {
961 node = shash_find(&init_ofp_ports, port.name);
962 if (!node && strcmp(port.name, dpif_base_name(backer->dpif))) {
963 garbage = xmalloc(sizeof *garbage);
964 garbage->odp_port = port.port_no;
965 list_push_front(&garbage_list, &garbage->list_node);
968 dpif_port_dump_done(&port_dump);
970 LIST_FOR_EACH_SAFE (garbage, next, list_node, &garbage_list) {
971 dpif_port_del(backer->dpif, garbage->odp_port);
972 list_remove(&garbage->list_node);
976 shash_add(&all_dpif_backers, type, backer);
978 error = dpif_recv_set(backer->dpif, backer->recv_set_enable);
980 VLOG_ERR("failed to listen on datapath of type %s: %s",
981 type, strerror(error));
982 close_dpif_backer(backer);
986 backer->max_n_subfacet = 0;
987 backer->created = time_msec();
988 backer->last_minute = backer->created;
989 memset(&backer->hourly, 0, sizeof backer->hourly);
990 memset(&backer->daily, 0, sizeof backer->daily);
991 memset(&backer->lifetime, 0, sizeof backer->lifetime);
992 backer->subfacet_add_count = 0;
993 backer->subfacet_del_count = 0;
994 backer->total_subfacet_add_count = 0;
995 backer->total_subfacet_del_count = 0;
996 backer->avg_n_subfacet = 0;
997 backer->avg_subfacet_life = 0;
1003 construct(struct ofproto *ofproto_)
1005 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1006 struct shash_node *node, *next;
1007 odp_port_t max_ports;
1011 error = open_dpif_backer(ofproto->up.type, &ofproto->backer);
1016 max_ports = dpif_get_max_ports(ofproto->backer->dpif);
1017 ofproto_init_max_ports(ofproto_, u16_to_ofp(MIN(odp_to_u32(max_ports),
1018 ofp_to_u16(OFPP_MAX))));
1020 ofproto->netflow = NULL;
1021 ofproto->sflow = NULL;
1022 ofproto->ipfix = NULL;
1023 ofproto->stp = NULL;
1024 hmap_init(&ofproto->bundles);
1025 ofproto->ml = mac_learning_create(MAC_ENTRY_DEFAULT_IDLE_TIME);
1026 for (i = 0; i < MAX_MIRRORS; i++) {
1027 ofproto->mirrors[i] = NULL;
1029 ofproto->has_bonded_bundles = false;
1031 classifier_init(&ofproto->facets);
1032 ofproto->consistency_rl = LLONG_MIN;
1034 for (i = 0; i < N_TABLES; i++) {
1035 struct table_dpif *table = &ofproto->tables[i];
1037 table->catchall_table = NULL;
1038 table->other_table = NULL;
1039 table->basis = random_uint32();
1042 list_init(&ofproto->completions);
1044 ofproto_dpif_unixctl_init();
1046 ofproto->has_mirrors = false;
1047 ofproto->has_bundle_action = false;
1049 hmap_init(&ofproto->vlandev_map);
1050 hmap_init(&ofproto->realdev_vid_map);
1052 sset_init(&ofproto->ports);
1053 sset_init(&ofproto->ghost_ports);
1054 sset_init(&ofproto->port_poll_set);
1055 ofproto->port_poll_errno = 0;
1057 SHASH_FOR_EACH_SAFE (node, next, &init_ofp_ports) {
1058 struct iface_hint *iface_hint = node->data;
1060 if (!strcmp(iface_hint->br_name, ofproto->up.name)) {
1061 /* Check if the datapath already has this port. */
1062 if (dpif_port_exists(ofproto->backer->dpif, node->name)) {
1063 sset_add(&ofproto->ports, node->name);
1066 free(iface_hint->br_name);
1067 free(iface_hint->br_type);
1069 shash_delete(&init_ofp_ports, node);
1073 hmap_insert(&all_ofproto_dpifs, &ofproto->all_ofproto_dpifs_node,
1074 hash_string(ofproto->up.name, 0));
1075 memset(&ofproto->stats, 0, sizeof ofproto->stats);
1077 ofproto_init_tables(ofproto_, N_TABLES);
1078 error = add_internal_flows(ofproto);
1079 ofproto->up.tables[TBL_INTERNAL].flags = OFTABLE_HIDDEN | OFTABLE_READONLY;
1082 ofproto->n_missed = 0;
1088 add_internal_flow(struct ofproto_dpif *ofproto, int id,
1089 const struct ofpbuf *ofpacts, struct rule_dpif **rulep)
1091 struct ofputil_flow_mod fm;
1094 match_init_catchall(&fm.match);
1096 match_set_reg(&fm.match, 0, id);
1097 fm.new_cookie = htonll(0);
1098 fm.cookie = htonll(0);
1099 fm.cookie_mask = htonll(0);
1100 fm.table_id = TBL_INTERNAL;
1101 fm.command = OFPFC_ADD;
1102 fm.idle_timeout = 0;
1103 fm.hard_timeout = 0;
1107 fm.ofpacts = ofpacts->data;
1108 fm.ofpacts_len = ofpacts->size;
1110 error = ofproto_flow_mod(&ofproto->up, &fm);
1112 VLOG_ERR_RL(&rl, "failed to add internal flow %d (%s)",
1113 id, ofperr_to_string(error));
1117 *rulep = rule_dpif_lookup_in_table(ofproto, &fm.match.flow, NULL,
1119 ovs_assert(*rulep != NULL);
1125 add_internal_flows(struct ofproto_dpif *ofproto)
1127 struct ofpact_controller *controller;
1128 uint64_t ofpacts_stub[128 / 8];
1129 struct ofpbuf ofpacts;
1133 ofpbuf_use_stack(&ofpacts, ofpacts_stub, sizeof ofpacts_stub);
1136 controller = ofpact_put_CONTROLLER(&ofpacts);
1137 controller->max_len = UINT16_MAX;
1138 controller->controller_id = 0;
1139 controller->reason = OFPR_NO_MATCH;
1140 ofpact_pad(&ofpacts);
1142 error = add_internal_flow(ofproto, id++, &ofpacts, &ofproto->miss_rule);
1147 ofpbuf_clear(&ofpacts);
1148 error = add_internal_flow(ofproto, id++, &ofpacts,
1149 &ofproto->no_packet_in_rule);
1154 error = add_internal_flow(ofproto, id++, &ofpacts,
1155 &ofproto->drop_frags_rule);
1160 complete_operations(struct ofproto_dpif *ofproto)
1162 struct dpif_completion *c, *next;
1164 LIST_FOR_EACH_SAFE (c, next, list_node, &ofproto->completions) {
1165 ofoperation_complete(c->op, 0);
1166 list_remove(&c->list_node);
1172 destruct(struct ofproto *ofproto_)
1174 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1175 struct rule_dpif *rule, *next_rule;
1176 struct oftable *table;
1179 ofproto->backer->need_revalidate = REV_RECONFIGURE;
1180 hmap_remove(&all_ofproto_dpifs, &ofproto->all_ofproto_dpifs_node);
1181 complete_operations(ofproto);
1183 OFPROTO_FOR_EACH_TABLE (table, &ofproto->up) {
1184 struct cls_cursor cursor;
1186 cls_cursor_init(&cursor, &table->cls, NULL);
1187 CLS_CURSOR_FOR_EACH_SAFE (rule, next_rule, up.cr, &cursor) {
1188 ofproto_rule_destroy(&rule->up);
1192 for (i = 0; i < MAX_MIRRORS; i++) {
1193 mirror_destroy(ofproto->mirrors[i]);
1196 netflow_destroy(ofproto->netflow);
1197 dpif_sflow_destroy(ofproto->sflow);
1198 hmap_destroy(&ofproto->bundles);
1199 mac_learning_destroy(ofproto->ml);
1201 classifier_destroy(&ofproto->facets);
1203 hmap_destroy(&ofproto->vlandev_map);
1204 hmap_destroy(&ofproto->realdev_vid_map);
1206 sset_destroy(&ofproto->ports);
1207 sset_destroy(&ofproto->ghost_ports);
1208 sset_destroy(&ofproto->port_poll_set);
1210 close_dpif_backer(ofproto->backer);
1214 run_fast(struct ofproto *ofproto_)
1216 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1217 struct ofport_dpif *ofport;
1219 /* Do not perform any periodic activity required by 'ofproto' while
1220 * waiting for flow restore to complete. */
1221 if (ofproto_get_flow_restore_wait()) {
1225 HMAP_FOR_EACH (ofport, up.hmap_node, &ofproto->up.ports) {
1226 port_run_fast(ofport);
1233 run(struct ofproto *ofproto_)
1235 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1236 struct ofport_dpif *ofport;
1237 struct ofbundle *bundle;
1241 complete_operations(ofproto);
1244 /* Do not perform any periodic activity below required by 'ofproto' while
1245 * waiting for flow restore to complete. */
1246 if (ofproto_get_flow_restore_wait()) {
1250 error = run_fast(ofproto_);
1255 if (ofproto->netflow) {
1256 if (netflow_run(ofproto->netflow)) {
1257 send_netflow_active_timeouts(ofproto);
1260 if (ofproto->sflow) {
1261 dpif_sflow_run(ofproto->sflow);
1264 HMAP_FOR_EACH (ofport, up.hmap_node, &ofproto->up.ports) {
1267 HMAP_FOR_EACH (bundle, hmap_node, &ofproto->bundles) {
1272 mac_learning_run(ofproto->ml, &ofproto->backer->revalidate_set);
1274 /* Check the consistency of a random facet, to aid debugging. */
1275 if (time_msec() >= ofproto->consistency_rl
1276 && !classifier_is_empty(&ofproto->facets)
1277 && !ofproto->backer->need_revalidate) {
1278 struct cls_table *table;
1279 struct cls_rule *cr;
1280 struct facet *facet;
1282 ofproto->consistency_rl = time_msec() + 250;
1284 table = CONTAINER_OF(hmap_random_node(&ofproto->facets.tables),
1285 struct cls_table, hmap_node);
1286 cr = CONTAINER_OF(hmap_random_node(&table->rules), struct cls_rule,
1288 facet = CONTAINER_OF(cr, struct facet, cr);
1290 if (!tag_set_intersects(&ofproto->backer->revalidate_set,
1291 facet->xout.tags)) {
1292 if (!facet_check_consistency(facet)) {
1293 ofproto->backer->need_revalidate = REV_INCONSISTENCY;
1302 wait(struct ofproto *ofproto_)
1304 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1305 struct ofport_dpif *ofport;
1306 struct ofbundle *bundle;
1308 if (!clogged && !list_is_empty(&ofproto->completions)) {
1309 poll_immediate_wake();
1312 if (ofproto_get_flow_restore_wait()) {
1316 dpif_wait(ofproto->backer->dpif);
1317 dpif_recv_wait(ofproto->backer->dpif);
1318 if (ofproto->sflow) {
1319 dpif_sflow_wait(ofproto->sflow);
1321 if (!tag_set_is_empty(&ofproto->backer->revalidate_set)) {
1322 poll_immediate_wake();
1324 HMAP_FOR_EACH (ofport, up.hmap_node, &ofproto->up.ports) {
1327 HMAP_FOR_EACH (bundle, hmap_node, &ofproto->bundles) {
1328 bundle_wait(bundle);
1330 if (ofproto->netflow) {
1331 netflow_wait(ofproto->netflow);
1333 mac_learning_wait(ofproto->ml);
1335 if (ofproto->backer->need_revalidate) {
1336 /* Shouldn't happen, but if it does just go around again. */
1337 VLOG_DBG_RL(&rl, "need revalidate in ofproto_wait_cb()");
1338 poll_immediate_wake();
1343 get_memory_usage(const struct ofproto *ofproto_, struct simap *usage)
1345 const struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1346 struct cls_cursor cursor;
1347 size_t n_subfacets = 0;
1348 struct facet *facet;
1350 simap_increase(usage, "facets", classifier_count(&ofproto->facets));
1352 cls_cursor_init(&cursor, &ofproto->facets, NULL);
1353 CLS_CURSOR_FOR_EACH (facet, cr, &cursor) {
1354 n_subfacets += list_size(&facet->subfacets);
1356 simap_increase(usage, "subfacets", n_subfacets);
1360 flush(struct ofproto *ofproto_)
1362 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1363 struct subfacet *subfacet, *next_subfacet;
1364 struct subfacet *batch[SUBFACET_DESTROY_MAX_BATCH];
1368 HMAP_FOR_EACH_SAFE (subfacet, next_subfacet, hmap_node,
1369 &ofproto->backer->subfacets) {
1370 if (ofproto_dpif_cast(subfacet->facet->rule->up.ofproto) != ofproto) {
1374 if (subfacet->path != SF_NOT_INSTALLED) {
1375 batch[n_batch++] = subfacet;
1376 if (n_batch >= SUBFACET_DESTROY_MAX_BATCH) {
1377 subfacet_destroy_batch(ofproto->backer, batch, n_batch);
1381 subfacet_destroy(subfacet);
1386 subfacet_destroy_batch(ofproto->backer, batch, n_batch);
1391 get_features(struct ofproto *ofproto_ OVS_UNUSED,
1392 bool *arp_match_ip, enum ofputil_action_bitmap *actions)
1394 *arp_match_ip = true;
1395 *actions = (OFPUTIL_A_OUTPUT |
1396 OFPUTIL_A_SET_VLAN_VID |
1397 OFPUTIL_A_SET_VLAN_PCP |
1398 OFPUTIL_A_STRIP_VLAN |
1399 OFPUTIL_A_SET_DL_SRC |
1400 OFPUTIL_A_SET_DL_DST |
1401 OFPUTIL_A_SET_NW_SRC |
1402 OFPUTIL_A_SET_NW_DST |
1403 OFPUTIL_A_SET_NW_TOS |
1404 OFPUTIL_A_SET_TP_SRC |
1405 OFPUTIL_A_SET_TP_DST |
1410 get_tables(struct ofproto *ofproto_, struct ofp12_table_stats *ots)
1412 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1413 struct dpif_dp_stats s;
1414 uint64_t n_miss, n_no_pkt_in, n_bytes, n_dropped_frags;
1417 strcpy(ots->name, "classifier");
1419 dpif_get_dp_stats(ofproto->backer->dpif, &s);
1420 rule_get_stats(&ofproto->miss_rule->up, &n_miss, &n_bytes);
1421 rule_get_stats(&ofproto->no_packet_in_rule->up, &n_no_pkt_in, &n_bytes);
1422 rule_get_stats(&ofproto->drop_frags_rule->up, &n_dropped_frags, &n_bytes);
1424 n_lookup = s.n_hit + s.n_missed - n_dropped_frags;
1425 ots->lookup_count = htonll(n_lookup);
1426 ots->matched_count = htonll(n_lookup - n_miss - n_no_pkt_in);
1429 static struct ofport *
1432 struct ofport_dpif *port = xmalloc(sizeof *port);
1437 port_dealloc(struct ofport *port_)
1439 struct ofport_dpif *port = ofport_dpif_cast(port_);
1444 port_construct(struct ofport *port_)
1446 struct ofport_dpif *port = ofport_dpif_cast(port_);
1447 struct ofproto_dpif *ofproto = ofproto_dpif_cast(port->up.ofproto);
1448 const struct netdev *netdev = port->up.netdev;
1449 char namebuf[NETDEV_VPORT_NAME_BUFSIZE];
1450 struct dpif_port dpif_port;
1453 ofproto->backer->need_revalidate = REV_RECONFIGURE;
1454 port->bundle = NULL;
1457 port->tag = tag_create_random();
1458 port->may_enable = true;
1459 port->stp_port = NULL;
1460 port->stp_state = STP_DISABLED;
1461 port->tnl_port = NULL;
1463 hmap_init(&port->priorities);
1464 port->realdev_ofp_port = 0;
1465 port->vlandev_vid = 0;
1466 port->carrier_seq = netdev_get_carrier_resets(netdev);
1468 if (netdev_vport_is_patch(netdev)) {
1469 /* By bailing out here, we don't submit the port to the sFlow module
1470 * to be considered for counter polling export. This is correct
1471 * because the patch port represents an interface that sFlow considers
1472 * to be "internal" to the switch as a whole, and therefore not an
1473 * candidate for counter polling. */
1474 port->odp_port = ODPP_NONE;
1475 ofport_update_peer(port);
1479 error = dpif_port_query_by_name(ofproto->backer->dpif,
1480 netdev_vport_get_dpif_port(netdev, namebuf,
1487 port->odp_port = dpif_port.port_no;
1489 if (netdev_get_tunnel_config(netdev)) {
1490 port->tnl_port = tnl_port_add(&port->up, port->odp_port);
1492 /* Sanity-check that a mapping doesn't already exist. This
1493 * shouldn't happen for non-tunnel ports. */
1494 if (odp_port_to_ofp_port(ofproto, port->odp_port) != OFPP_NONE) {
1495 VLOG_ERR("port %s already has an OpenFlow port number",
1497 dpif_port_destroy(&dpif_port);
1501 hmap_insert(&ofproto->backer->odp_to_ofport_map, &port->odp_port_node,
1502 hash_int(odp_to_u32(port->odp_port), 0));
1504 dpif_port_destroy(&dpif_port);
1506 if (ofproto->sflow) {
1507 dpif_sflow_add_port(ofproto->sflow, port_, port->odp_port);
1514 port_destruct(struct ofport *port_)
1516 struct ofport_dpif *port = ofport_dpif_cast(port_);
1517 struct ofproto_dpif *ofproto = ofproto_dpif_cast(port->up.ofproto);
1518 const char *devname = netdev_get_name(port->up.netdev);
1519 char namebuf[NETDEV_VPORT_NAME_BUFSIZE];
1520 const char *dp_port_name;
1522 dp_port_name = netdev_vport_get_dpif_port(port->up.netdev, namebuf,
1524 if (dpif_port_exists(ofproto->backer->dpif, dp_port_name)) {
1525 /* The underlying device is still there, so delete it. This
1526 * happens when the ofproto is being destroyed, since the caller
1527 * assumes that removal of attached ports will happen as part of
1529 if (!port->tnl_port) {
1530 dpif_port_del(ofproto->backer->dpif, port->odp_port);
1532 ofproto->backer->need_revalidate = REV_RECONFIGURE;
1536 port->peer->peer = NULL;
1540 if (port->odp_port != ODPP_NONE && !port->tnl_port) {
1541 hmap_remove(&ofproto->backer->odp_to_ofport_map, &port->odp_port_node);
1544 tnl_port_del(port->tnl_port);
1545 sset_find_and_delete(&ofproto->ports, devname);
1546 sset_find_and_delete(&ofproto->ghost_ports, devname);
1547 ofproto->backer->need_revalidate = REV_RECONFIGURE;
1548 bundle_remove(port_);
1549 set_cfm(port_, NULL);
1550 set_bfd(port_, NULL);
1551 if (ofproto->sflow) {
1552 dpif_sflow_del_port(ofproto->sflow, port->odp_port);
1555 ofport_clear_priorities(port);
1556 hmap_destroy(&port->priorities);
1560 port_modified(struct ofport *port_)
1562 struct ofport_dpif *port = ofport_dpif_cast(port_);
1564 if (port->bundle && port->bundle->bond) {
1565 bond_slave_set_netdev(port->bundle->bond, port, port->up.netdev);
1569 cfm_set_netdev(port->cfm, port->up.netdev);
1572 if (port->tnl_port && tnl_port_reconfigure(&port->up, port->odp_port,
1574 ofproto_dpif_cast(port->up.ofproto)->backer->need_revalidate = true;
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_destroy(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_destroy(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_destroy(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) {
2129 /* Looks up each of the 'n_auxes' pointers in 'auxes' as bundles and adds the
2130 * ones that are found to 'bundles'. */
2132 bundle_lookup_multiple(struct ofproto_dpif *ofproto,
2133 void **auxes, size_t n_auxes,
2134 struct hmapx *bundles)
2138 hmapx_init(bundles);
2139 for (i = 0; i < n_auxes; i++) {
2140 struct ofbundle *bundle = bundle_lookup(ofproto, auxes[i]);
2142 hmapx_add(bundles, bundle);
2148 bundle_update(struct ofbundle *bundle)
2150 struct ofport_dpif *port;
2152 bundle->floodable = true;
2153 LIST_FOR_EACH (port, bundle_node, &bundle->ports) {
2154 if (port->up.pp.config & OFPUTIL_PC_NO_FLOOD
2155 || !stp_forward_in_state(port->stp_state)) {
2156 bundle->floodable = false;
2163 bundle_del_port(struct ofport_dpif *port)
2165 struct ofbundle *bundle = port->bundle;
2167 bundle->ofproto->backer->need_revalidate = REV_RECONFIGURE;
2169 list_remove(&port->bundle_node);
2170 port->bundle = NULL;
2173 lacp_slave_unregister(bundle->lacp, port);
2176 bond_slave_unregister(bundle->bond, port);
2179 bundle_update(bundle);
2183 bundle_add_port(struct ofbundle *bundle, ofp_port_t ofp_port,
2184 struct lacp_slave_settings *lacp)
2186 struct ofport_dpif *port;
2188 port = get_ofp_port(bundle->ofproto, ofp_port);
2193 if (port->bundle != bundle) {
2194 bundle->ofproto->backer->need_revalidate = REV_RECONFIGURE;
2196 bundle_del_port(port);
2199 port->bundle = bundle;
2200 list_push_back(&bundle->ports, &port->bundle_node);
2201 if (port->up.pp.config & OFPUTIL_PC_NO_FLOOD
2202 || !stp_forward_in_state(port->stp_state)) {
2203 bundle->floodable = false;
2207 bundle->ofproto->backer->need_revalidate = REV_RECONFIGURE;
2208 lacp_slave_register(bundle->lacp, port, lacp);
2215 bundle_destroy(struct ofbundle *bundle)
2217 struct ofproto_dpif *ofproto;
2218 struct ofport_dpif *port, *next_port;
2225 ofproto = bundle->ofproto;
2226 for (i = 0; i < MAX_MIRRORS; i++) {
2227 struct ofmirror *m = ofproto->mirrors[i];
2229 if (m->out == bundle) {
2231 } else if (hmapx_find_and_delete(&m->srcs, bundle)
2232 || hmapx_find_and_delete(&m->dsts, bundle)) {
2233 ofproto->backer->need_revalidate = REV_RECONFIGURE;
2238 LIST_FOR_EACH_SAFE (port, next_port, bundle_node, &bundle->ports) {
2239 bundle_del_port(port);
2242 bundle_flush_macs(bundle, true);
2243 hmap_remove(&ofproto->bundles, &bundle->hmap_node);
2245 free(bundle->trunks);
2246 lacp_destroy(bundle->lacp);
2247 bond_destroy(bundle->bond);
2252 bundle_set(struct ofproto *ofproto_, void *aux,
2253 const struct ofproto_bundle_settings *s)
2255 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
2256 bool need_flush = false;
2257 struct ofport_dpif *port;
2258 struct ofbundle *bundle;
2259 unsigned long *trunks;
2265 bundle_destroy(bundle_lookup(ofproto, aux));
2269 ovs_assert(s->n_slaves == 1 || s->bond != NULL);
2270 ovs_assert((s->lacp != NULL) == (s->lacp_slaves != NULL));
2272 bundle = bundle_lookup(ofproto, aux);
2274 bundle = xmalloc(sizeof *bundle);
2276 bundle->ofproto = ofproto;
2277 hmap_insert(&ofproto->bundles, &bundle->hmap_node,
2278 hash_pointer(aux, 0));
2280 bundle->name = NULL;
2282 list_init(&bundle->ports);
2283 bundle->vlan_mode = PORT_VLAN_TRUNK;
2285 bundle->trunks = NULL;
2286 bundle->use_priority_tags = s->use_priority_tags;
2287 bundle->lacp = NULL;
2288 bundle->bond = NULL;
2290 bundle->floodable = true;
2292 bundle->src_mirrors = 0;
2293 bundle->dst_mirrors = 0;
2294 bundle->mirror_out = 0;
2297 if (!bundle->name || strcmp(s->name, bundle->name)) {
2299 bundle->name = xstrdup(s->name);
2304 if (!bundle->lacp) {
2305 ofproto->backer->need_revalidate = REV_RECONFIGURE;
2306 bundle->lacp = lacp_create();
2308 lacp_configure(bundle->lacp, s->lacp);
2310 lacp_destroy(bundle->lacp);
2311 bundle->lacp = NULL;
2314 /* Update set of ports. */
2316 for (i = 0; i < s->n_slaves; i++) {
2317 if (!bundle_add_port(bundle, s->slaves[i],
2318 s->lacp ? &s->lacp_slaves[i] : NULL)) {
2322 if (!ok || list_size(&bundle->ports) != s->n_slaves) {
2323 struct ofport_dpif *next_port;
2325 LIST_FOR_EACH_SAFE (port, next_port, bundle_node, &bundle->ports) {
2326 for (i = 0; i < s->n_slaves; i++) {
2327 if (s->slaves[i] == port->up.ofp_port) {
2332 bundle_del_port(port);
2336 ovs_assert(list_size(&bundle->ports) <= s->n_slaves);
2338 if (list_is_empty(&bundle->ports)) {
2339 bundle_destroy(bundle);
2343 /* Set VLAN tagging mode */
2344 if (s->vlan_mode != bundle->vlan_mode
2345 || s->use_priority_tags != bundle->use_priority_tags) {
2346 bundle->vlan_mode = s->vlan_mode;
2347 bundle->use_priority_tags = s->use_priority_tags;
2352 vlan = (s->vlan_mode == PORT_VLAN_TRUNK ? -1
2353 : s->vlan >= 0 && s->vlan <= 4095 ? s->vlan
2355 if (vlan != bundle->vlan) {
2356 bundle->vlan = vlan;
2360 /* Get trunked VLANs. */
2361 switch (s->vlan_mode) {
2362 case PORT_VLAN_ACCESS:
2366 case PORT_VLAN_TRUNK:
2367 trunks = CONST_CAST(unsigned long *, s->trunks);
2370 case PORT_VLAN_NATIVE_UNTAGGED:
2371 case PORT_VLAN_NATIVE_TAGGED:
2372 if (vlan != 0 && (!s->trunks
2373 || !bitmap_is_set(s->trunks, vlan)
2374 || bitmap_is_set(s->trunks, 0))) {
2375 /* Force trunking the native VLAN and prohibit trunking VLAN 0. */
2377 trunks = bitmap_clone(s->trunks, 4096);
2379 trunks = bitmap_allocate1(4096);
2381 bitmap_set1(trunks, vlan);
2382 bitmap_set0(trunks, 0);
2384 trunks = CONST_CAST(unsigned long *, s->trunks);
2391 if (!vlan_bitmap_equal(trunks, bundle->trunks)) {
2392 free(bundle->trunks);
2393 if (trunks == s->trunks) {
2394 bundle->trunks = vlan_bitmap_clone(trunks);
2396 bundle->trunks = trunks;
2401 if (trunks != s->trunks) {
2406 if (!list_is_short(&bundle->ports)) {
2407 bundle->ofproto->has_bonded_bundles = true;
2409 if (bond_reconfigure(bundle->bond, s->bond)) {
2410 ofproto->backer->need_revalidate = REV_RECONFIGURE;
2413 bundle->bond = bond_create(s->bond);
2414 ofproto->backer->need_revalidate = REV_RECONFIGURE;
2417 LIST_FOR_EACH (port, bundle_node, &bundle->ports) {
2418 bond_slave_register(bundle->bond, port, port->up.netdev);
2421 bond_destroy(bundle->bond);
2422 bundle->bond = NULL;
2425 /* If we changed something that would affect MAC learning, un-learn
2426 * everything on this port and force flow revalidation. */
2428 bundle_flush_macs(bundle, false);
2435 bundle_remove(struct ofport *port_)
2437 struct ofport_dpif *port = ofport_dpif_cast(port_);
2438 struct ofbundle *bundle = port->bundle;
2441 bundle_del_port(port);
2442 if (list_is_empty(&bundle->ports)) {
2443 bundle_destroy(bundle);
2444 } else if (list_is_short(&bundle->ports)) {
2445 bond_destroy(bundle->bond);
2446 bundle->bond = NULL;
2452 send_pdu_cb(void *port_, const void *pdu, size_t pdu_size)
2454 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 10);
2455 struct ofport_dpif *port = port_;
2456 uint8_t ea[ETH_ADDR_LEN];
2459 error = netdev_get_etheraddr(port->up.netdev, ea);
2461 struct ofpbuf packet;
2464 ofpbuf_init(&packet, 0);
2465 packet_pdu = eth_compose(&packet, eth_addr_lacp, ea, ETH_TYPE_LACP,
2467 memcpy(packet_pdu, pdu, pdu_size);
2469 send_packet(port, &packet);
2470 ofpbuf_uninit(&packet);
2472 VLOG_ERR_RL(&rl, "port %s: cannot obtain Ethernet address of iface "
2473 "%s (%s)", port->bundle->name,
2474 netdev_get_name(port->up.netdev), strerror(error));
2479 bundle_send_learning_packets(struct ofbundle *bundle)
2481 struct ofproto_dpif *ofproto = bundle->ofproto;
2482 int error, n_packets, n_errors;
2483 struct mac_entry *e;
2485 error = n_packets = n_errors = 0;
2486 LIST_FOR_EACH (e, lru_node, &ofproto->ml->lrus) {
2487 if (e->port.p != bundle) {
2488 struct ofpbuf *learning_packet;
2489 struct ofport_dpif *port;
2493 /* The assignment to "port" is unnecessary but makes "grep"ing for
2494 * struct ofport_dpif more effective. */
2495 learning_packet = bond_compose_learning_packet(bundle->bond,
2499 ret = send_packet(port, learning_packet);
2500 ofpbuf_delete(learning_packet);
2510 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
2511 VLOG_WARN_RL(&rl, "bond %s: %d errors sending %d gratuitous learning "
2512 "packets, last error was: %s",
2513 bundle->name, n_errors, n_packets, strerror(error));
2515 VLOG_DBG("bond %s: sent %d gratuitous learning packets",
2516 bundle->name, n_packets);
2521 bundle_run(struct ofbundle *bundle)
2524 lacp_run(bundle->lacp, send_pdu_cb);
2527 struct ofport_dpif *port;
2529 LIST_FOR_EACH (port, bundle_node, &bundle->ports) {
2530 bond_slave_set_may_enable(bundle->bond, port, port->may_enable);
2533 bond_run(bundle->bond, &bundle->ofproto->backer->revalidate_set,
2534 lacp_status(bundle->lacp));
2535 if (bond_should_send_learning_packets(bundle->bond)) {
2536 bundle_send_learning_packets(bundle);
2542 bundle_wait(struct ofbundle *bundle)
2545 lacp_wait(bundle->lacp);
2548 bond_wait(bundle->bond);
2555 mirror_scan(struct ofproto_dpif *ofproto)
2559 for (idx = 0; idx < MAX_MIRRORS; idx++) {
2560 if (!ofproto->mirrors[idx]) {
2567 static struct ofmirror *
2568 mirror_lookup(struct ofproto_dpif *ofproto, void *aux)
2572 for (i = 0; i < MAX_MIRRORS; i++) {
2573 struct ofmirror *mirror = ofproto->mirrors[i];
2574 if (mirror && mirror->aux == aux) {
2582 /* Update the 'dup_mirrors' member of each of the ofmirrors in 'ofproto'. */
2584 mirror_update_dups(struct ofproto_dpif *ofproto)
2588 for (i = 0; i < MAX_MIRRORS; i++) {
2589 struct ofmirror *m = ofproto->mirrors[i];
2592 m->dup_mirrors = MIRROR_MASK_C(1) << i;
2596 for (i = 0; i < MAX_MIRRORS; i++) {
2597 struct ofmirror *m1 = ofproto->mirrors[i];
2604 for (j = i + 1; j < MAX_MIRRORS; j++) {
2605 struct ofmirror *m2 = ofproto->mirrors[j];
2607 if (m2 && m1->out == m2->out && m1->out_vlan == m2->out_vlan) {
2608 m1->dup_mirrors |= MIRROR_MASK_C(1) << j;
2609 m2->dup_mirrors |= m1->dup_mirrors;
2616 mirror_set(struct ofproto *ofproto_, void *aux,
2617 const struct ofproto_mirror_settings *s)
2619 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
2620 mirror_mask_t mirror_bit;
2621 struct ofbundle *bundle;
2622 struct ofmirror *mirror;
2623 struct ofbundle *out;
2624 struct hmapx srcs; /* Contains "struct ofbundle *"s. */
2625 struct hmapx dsts; /* Contains "struct ofbundle *"s. */
2628 mirror = mirror_lookup(ofproto, aux);
2630 mirror_destroy(mirror);
2636 idx = mirror_scan(ofproto);
2638 VLOG_WARN("bridge %s: maximum of %d port mirrors reached, "
2640 ofproto->up.name, MAX_MIRRORS, s->name);
2644 mirror = ofproto->mirrors[idx] = xzalloc(sizeof *mirror);
2645 mirror->ofproto = ofproto;
2648 mirror->out_vlan = -1;
2649 mirror->name = NULL;
2652 if (!mirror->name || strcmp(s->name, mirror->name)) {
2654 mirror->name = xstrdup(s->name);
2657 /* Get the new configuration. */
2658 if (s->out_bundle) {
2659 out = bundle_lookup(ofproto, s->out_bundle);
2661 mirror_destroy(mirror);
2667 out_vlan = s->out_vlan;
2669 bundle_lookup_multiple(ofproto, s->srcs, s->n_srcs, &srcs);
2670 bundle_lookup_multiple(ofproto, s->dsts, s->n_dsts, &dsts);
2672 /* If the configuration has not changed, do nothing. */
2673 if (hmapx_equals(&srcs, &mirror->srcs)
2674 && hmapx_equals(&dsts, &mirror->dsts)
2675 && vlan_bitmap_equal(mirror->vlans, s->src_vlans)
2676 && mirror->out == out
2677 && mirror->out_vlan == out_vlan)
2679 hmapx_destroy(&srcs);
2680 hmapx_destroy(&dsts);
2684 hmapx_swap(&srcs, &mirror->srcs);
2685 hmapx_destroy(&srcs);
2687 hmapx_swap(&dsts, &mirror->dsts);
2688 hmapx_destroy(&dsts);
2690 free(mirror->vlans);
2691 mirror->vlans = vlan_bitmap_clone(s->src_vlans);
2694 mirror->out_vlan = out_vlan;
2696 /* Update bundles. */
2697 mirror_bit = MIRROR_MASK_C(1) << mirror->idx;
2698 HMAP_FOR_EACH (bundle, hmap_node, &mirror->ofproto->bundles) {
2699 if (hmapx_contains(&mirror->srcs, bundle)) {
2700 bundle->src_mirrors |= mirror_bit;
2702 bundle->src_mirrors &= ~mirror_bit;
2705 if (hmapx_contains(&mirror->dsts, bundle)) {
2706 bundle->dst_mirrors |= mirror_bit;
2708 bundle->dst_mirrors &= ~mirror_bit;
2711 if (mirror->out == bundle) {
2712 bundle->mirror_out |= mirror_bit;
2714 bundle->mirror_out &= ~mirror_bit;
2718 ofproto->backer->need_revalidate = REV_RECONFIGURE;
2719 ofproto->has_mirrors = true;
2720 mac_learning_flush(ofproto->ml,
2721 &ofproto->backer->revalidate_set);
2722 mirror_update_dups(ofproto);
2728 mirror_destroy(struct ofmirror *mirror)
2730 struct ofproto_dpif *ofproto;
2731 mirror_mask_t mirror_bit;
2732 struct ofbundle *bundle;
2739 ofproto = mirror->ofproto;
2740 ofproto->backer->need_revalidate = REV_RECONFIGURE;
2741 mac_learning_flush(ofproto->ml, &ofproto->backer->revalidate_set);
2743 mirror_bit = MIRROR_MASK_C(1) << mirror->idx;
2744 HMAP_FOR_EACH (bundle, hmap_node, &ofproto->bundles) {
2745 bundle->src_mirrors &= ~mirror_bit;
2746 bundle->dst_mirrors &= ~mirror_bit;
2747 bundle->mirror_out &= ~mirror_bit;
2750 hmapx_destroy(&mirror->srcs);
2751 hmapx_destroy(&mirror->dsts);
2752 free(mirror->vlans);
2754 ofproto->mirrors[mirror->idx] = NULL;
2758 mirror_update_dups(ofproto);
2760 ofproto->has_mirrors = false;
2761 for (i = 0; i < MAX_MIRRORS; i++) {
2762 if (ofproto->mirrors[i]) {
2763 ofproto->has_mirrors = true;
2770 mirror_get_stats(struct ofproto *ofproto_, void *aux,
2771 uint64_t *packets, uint64_t *bytes)
2773 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
2774 struct ofmirror *mirror = mirror_lookup(ofproto, aux);
2777 *packets = *bytes = UINT64_MAX;
2783 *packets = mirror->packet_count;
2784 *bytes = mirror->byte_count;
2790 set_flood_vlans(struct ofproto *ofproto_, unsigned long *flood_vlans)
2792 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
2793 if (mac_learning_set_flood_vlans(ofproto->ml, flood_vlans)) {
2794 mac_learning_flush(ofproto->ml, &ofproto->backer->revalidate_set);
2800 is_mirror_output_bundle(const struct ofproto *ofproto_, void *aux)
2802 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
2803 struct ofbundle *bundle = bundle_lookup(ofproto, aux);
2804 return bundle && bundle->mirror_out != 0;
2808 forward_bpdu_changed(struct ofproto *ofproto_)
2810 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
2811 ofproto->backer->need_revalidate = REV_RECONFIGURE;
2815 set_mac_table_config(struct ofproto *ofproto_, unsigned int idle_time,
2818 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
2819 mac_learning_set_idle_time(ofproto->ml, idle_time);
2820 mac_learning_set_max_entries(ofproto->ml, max_entries);
2825 struct ofport_dpif *
2826 get_ofp_port(const struct ofproto_dpif *ofproto, ofp_port_t ofp_port)
2828 struct ofport *ofport = ofproto_get_port(&ofproto->up, ofp_port);
2829 return ofport ? ofport_dpif_cast(ofport) : NULL;
2832 struct ofport_dpif *
2833 get_odp_port(const struct ofproto_dpif *ofproto, odp_port_t odp_port)
2835 struct ofport_dpif *port = odp_port_to_ofport(ofproto->backer, odp_port);
2836 return port && &ofproto->up == port->up.ofproto ? port : NULL;
2840 ofproto_port_from_dpif_port(struct ofproto_dpif *ofproto,
2841 struct ofproto_port *ofproto_port,
2842 struct dpif_port *dpif_port)
2844 ofproto_port->name = dpif_port->name;
2845 ofproto_port->type = dpif_port->type;
2846 ofproto_port->ofp_port = odp_port_to_ofp_port(ofproto, dpif_port->port_no);
2850 ofport_update_peer(struct ofport_dpif *ofport)
2852 const struct ofproto_dpif *ofproto;
2853 struct dpif_backer *backer;
2854 const char *peer_name;
2856 if (!netdev_vport_is_patch(ofport->up.netdev)) {
2860 backer = ofproto_dpif_cast(ofport->up.ofproto)->backer;
2861 backer->need_revalidate = true;
2864 ofport->peer->peer = NULL;
2865 ofport->peer = NULL;
2868 peer_name = netdev_vport_patch_peer(ofport->up.netdev);
2873 HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
2874 struct ofport *peer_ofport;
2875 struct ofport_dpif *peer;
2876 const char *peer_peer;
2878 if (ofproto->backer != backer) {
2882 peer_ofport = shash_find_data(&ofproto->up.port_by_name, peer_name);
2887 peer = ofport_dpif_cast(peer_ofport);
2888 peer_peer = netdev_vport_patch_peer(peer->up.netdev);
2889 if (peer_peer && !strcmp(netdev_get_name(ofport->up.netdev),
2891 ofport->peer = peer;
2892 ofport->peer->peer = ofport;
2900 port_run_fast(struct ofport_dpif *ofport)
2902 if (ofport->cfm && cfm_should_send_ccm(ofport->cfm)) {
2903 struct ofpbuf packet;
2905 ofpbuf_init(&packet, 0);
2906 cfm_compose_ccm(ofport->cfm, &packet, ofport->up.pp.hw_addr);
2907 send_packet(ofport, &packet);
2908 ofpbuf_uninit(&packet);
2911 if (ofport->bfd && bfd_should_send_packet(ofport->bfd)) {
2912 struct ofpbuf packet;
2914 ofpbuf_init(&packet, 0);
2915 bfd_put_packet(ofport->bfd, &packet, ofport->up.pp.hw_addr);
2916 send_packet(ofport, &packet);
2917 ofpbuf_uninit(&packet);
2922 port_run(struct ofport_dpif *ofport)
2924 long long int carrier_seq = netdev_get_carrier_resets(ofport->up.netdev);
2925 bool carrier_changed = carrier_seq != ofport->carrier_seq;
2926 bool enable = netdev_get_carrier(ofport->up.netdev);
2928 ofport->carrier_seq = carrier_seq;
2930 port_run_fast(ofport);
2933 int cfm_opup = cfm_get_opup(ofport->cfm);
2935 cfm_run(ofport->cfm);
2936 enable = enable && !cfm_get_fault(ofport->cfm);
2938 if (cfm_opup >= 0) {
2939 enable = enable && cfm_opup;
2944 bfd_run(ofport->bfd);
2945 enable = enable && bfd_forwarding(ofport->bfd);
2948 if (ofport->bundle) {
2949 enable = enable && lacp_slave_may_enable(ofport->bundle->lacp, ofport);
2950 if (carrier_changed) {
2951 lacp_slave_carrier_changed(ofport->bundle->lacp, ofport);
2955 if (ofport->may_enable != enable) {
2956 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport->up.ofproto);
2958 if (ofproto->has_bundle_action) {
2959 ofproto->backer->need_revalidate = REV_PORT_TOGGLED;
2963 ofport->may_enable = enable;
2967 port_wait(struct ofport_dpif *ofport)
2970 cfm_wait(ofport->cfm);
2974 bfd_wait(ofport->bfd);
2979 port_query_by_name(const struct ofproto *ofproto_, const char *devname,
2980 struct ofproto_port *ofproto_port)
2982 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
2983 struct dpif_port dpif_port;
2986 if (sset_contains(&ofproto->ghost_ports, devname)) {
2987 const char *type = netdev_get_type_from_name(devname);
2989 /* We may be called before ofproto->up.port_by_name is populated with
2990 * the appropriate ofport. For this reason, we must get the name and
2991 * type from the netdev layer directly. */
2993 const struct ofport *ofport;
2995 ofport = shash_find_data(&ofproto->up.port_by_name, devname);
2996 ofproto_port->ofp_port = ofport ? ofport->ofp_port : OFPP_NONE;
2997 ofproto_port->name = xstrdup(devname);
2998 ofproto_port->type = xstrdup(type);
3004 if (!sset_contains(&ofproto->ports, devname)) {
3007 error = dpif_port_query_by_name(ofproto->backer->dpif,
3008 devname, &dpif_port);
3010 ofproto_port_from_dpif_port(ofproto, ofproto_port, &dpif_port);
3016 port_add(struct ofproto *ofproto_, struct netdev *netdev)
3018 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
3019 const char *devname = netdev_get_name(netdev);
3020 char namebuf[NETDEV_VPORT_NAME_BUFSIZE];
3021 const char *dp_port_name;
3023 if (netdev_vport_is_patch(netdev)) {
3024 sset_add(&ofproto->ghost_ports, netdev_get_name(netdev));
3028 dp_port_name = netdev_vport_get_dpif_port(netdev, namebuf, sizeof namebuf);
3029 if (!dpif_port_exists(ofproto->backer->dpif, dp_port_name)) {
3030 odp_port_t port_no = ODPP_NONE;
3033 error = dpif_port_add(ofproto->backer->dpif, netdev, &port_no);
3037 if (netdev_get_tunnel_config(netdev)) {
3038 simap_put(&ofproto->backer->tnl_backers,
3039 dp_port_name, odp_to_u32(port_no));
3043 if (netdev_get_tunnel_config(netdev)) {
3044 sset_add(&ofproto->ghost_ports, devname);
3046 sset_add(&ofproto->ports, devname);
3052 port_del(struct ofproto *ofproto_, ofp_port_t ofp_port)
3054 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
3055 struct ofport_dpif *ofport = get_ofp_port(ofproto, ofp_port);
3062 sset_find_and_delete(&ofproto->ghost_ports,
3063 netdev_get_name(ofport->up.netdev));
3064 ofproto->backer->need_revalidate = REV_RECONFIGURE;
3065 if (!ofport->tnl_port) {
3066 error = dpif_port_del(ofproto->backer->dpif, ofport->odp_port);
3068 /* The caller is going to close ofport->up.netdev. If this is a
3069 * bonded port, then the bond is using that netdev, so remove it
3070 * from the bond. The client will need to reconfigure everything
3071 * after deleting ports, so then the slave will get re-added. */
3072 bundle_remove(&ofport->up);
3079 port_get_stats(const struct ofport *ofport_, struct netdev_stats *stats)
3081 struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
3086 error = netdev_get_stats(ofport->up.netdev, stats);
3088 if (!error && ofport_->ofp_port == OFPP_LOCAL) {
3089 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport->up.ofproto);
3091 /* ofproto->stats.tx_packets represents packets that we created
3092 * internally and sent to some port (e.g. packets sent with
3093 * send_packet()). Account for them as if they had come from
3094 * OFPP_LOCAL and got forwarded. */
3096 if (stats->rx_packets != UINT64_MAX) {
3097 stats->rx_packets += ofproto->stats.tx_packets;
3100 if (stats->rx_bytes != UINT64_MAX) {
3101 stats->rx_bytes += ofproto->stats.tx_bytes;
3104 /* ofproto->stats.rx_packets represents packets that were received on
3105 * some port and we processed internally and dropped (e.g. STP).
3106 * Account for them as if they had been forwarded to OFPP_LOCAL. */
3108 if (stats->tx_packets != UINT64_MAX) {
3109 stats->tx_packets += ofproto->stats.rx_packets;
3112 if (stats->tx_bytes != UINT64_MAX) {
3113 stats->tx_bytes += ofproto->stats.rx_bytes;
3120 struct port_dump_state {
3125 struct ofproto_port port;
3130 port_dump_start(const struct ofproto *ofproto_ OVS_UNUSED, void **statep)
3132 *statep = xzalloc(sizeof(struct port_dump_state));
3137 port_dump_next(const struct ofproto *ofproto_, void *state_,
3138 struct ofproto_port *port)
3140 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
3141 struct port_dump_state *state = state_;
3142 const struct sset *sset;
3143 struct sset_node *node;
3145 if (state->has_port) {
3146 ofproto_port_destroy(&state->port);
3147 state->has_port = false;
3149 sset = state->ghost ? &ofproto->ghost_ports : &ofproto->ports;
3150 while ((node = sset_at_position(sset, &state->bucket, &state->offset))) {
3153 error = port_query_by_name(ofproto_, node->name, &state->port);
3155 *port = state->port;
3156 state->has_port = true;
3158 } else if (error != ENODEV) {
3163 if (!state->ghost) {
3164 state->ghost = true;
3167 return port_dump_next(ofproto_, state_, port);
3174 port_dump_done(const struct ofproto *ofproto_ OVS_UNUSED, void *state_)
3176 struct port_dump_state *state = state_;
3178 if (state->has_port) {
3179 ofproto_port_destroy(&state->port);
3186 port_poll(const struct ofproto *ofproto_, char **devnamep)
3188 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
3190 if (ofproto->port_poll_errno) {
3191 int error = ofproto->port_poll_errno;
3192 ofproto->port_poll_errno = 0;
3196 if (sset_is_empty(&ofproto->port_poll_set)) {
3200 *devnamep = sset_pop(&ofproto->port_poll_set);
3205 port_poll_wait(const struct ofproto *ofproto_)
3207 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
3208 dpif_port_poll_wait(ofproto->backer->dpif);
3212 port_is_lacp_current(const struct ofport *ofport_)
3214 const struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
3215 return (ofport->bundle && ofport->bundle->lacp
3216 ? lacp_slave_is_current(ofport->bundle->lacp, ofport)
3220 /* Upcall handling. */
3222 /* Flow miss batching.
3224 * Some dpifs implement operations faster when you hand them off in a batch.
3225 * To allow batching, "struct flow_miss" queues the dpif-related work needed
3226 * for a given flow. Each "struct flow_miss" corresponds to sending one or
3227 * more packets, plus possibly installing the flow in the dpif.
3229 * So far we only batch the operations that affect flow setup time the most.
3230 * It's possible to batch more than that, but the benefit might be minimal. */
3232 struct hmap_node hmap_node;
3233 struct ofproto_dpif *ofproto;
3235 enum odp_key_fitness key_fitness;
3236 const struct nlattr *key;
3238 struct list packets;
3239 enum dpif_upcall_type upcall_type;
3242 struct flow_miss_op {
3243 struct dpif_op dpif_op;
3245 uint64_t slow_stub[128 / 8]; /* Buffer for compose_slow_path() */
3246 struct xlate_out xout;
3247 bool xout_garbage; /* 'xout' needs to be uninitialized? */
3250 /* Sends an OFPT_PACKET_IN message for 'packet' of type OFPR_NO_MATCH to each
3251 * OpenFlow controller as necessary according to their individual
3252 * configurations. */
3254 send_packet_in_miss(struct ofproto_dpif *ofproto, const struct ofpbuf *packet,
3255 const struct flow *flow)
3257 struct ofputil_packet_in pin;
3259 pin.packet = packet->data;
3260 pin.packet_len = packet->size;
3261 pin.reason = OFPR_NO_MATCH;
3262 pin.controller_id = 0;
3267 pin.send_len = 0; /* not used for flow table misses */
3269 flow_get_metadata(flow, &pin.fmd);
3271 connmgr_send_packet_in(ofproto->up.connmgr, &pin);
3274 static struct flow_miss *
3275 flow_miss_find(struct hmap *todo, const struct ofproto_dpif *ofproto,
3276 const struct flow *flow, uint32_t hash)
3278 struct flow_miss *miss;
3280 HMAP_FOR_EACH_WITH_HASH (miss, hmap_node, hash, todo) {
3281 if (miss->ofproto == ofproto && flow_equal(&miss->flow, flow)) {
3289 /* Partially Initializes 'op' as an "execute" operation for 'miss' and
3290 * 'packet'. The caller must initialize op->actions and op->actions_len. If
3291 * 'miss' is associated with a subfacet the caller must also initialize the
3292 * returned op->subfacet, and if anything needs to be freed after processing
3293 * the op, the caller must initialize op->garbage also. */
3295 init_flow_miss_execute_op(struct flow_miss *miss, struct ofpbuf *packet,
3296 struct flow_miss_op *op)
3298 if (miss->flow.in_port.ofp_port
3299 != vsp_realdev_to_vlandev(miss->ofproto, miss->flow.in_port.ofp_port,
3300 miss->flow.vlan_tci)) {
3301 /* This packet was received on a VLAN splinter port. We
3302 * added a VLAN to the packet to make the packet resemble
3303 * the flow, but the actions were composed assuming that
3304 * the packet contained no VLAN. So, we must remove the
3305 * VLAN header from the packet before trying to execute the
3307 eth_pop_vlan(packet);
3310 op->xout_garbage = false;
3311 op->dpif_op.type = DPIF_OP_EXECUTE;
3312 op->dpif_op.u.execute.key = miss->key;
3313 op->dpif_op.u.execute.key_len = miss->key_len;
3314 op->dpif_op.u.execute.packet = packet;
3317 /* Helper for handle_flow_miss_without_facet() and
3318 * handle_flow_miss_with_facet(). */
3320 handle_flow_miss_common(struct rule_dpif *rule,
3321 struct ofpbuf *packet, const struct flow *flow)
3323 struct ofproto_dpif *ofproto = ofproto_dpif_cast(rule->up.ofproto);
3325 if (rule->up.cr.priority == FAIL_OPEN_PRIORITY) {
3327 * Extra-special case for fail-open mode.
3329 * We are in fail-open mode and the packet matched the fail-open
3330 * rule, but we are connected to a controller too. We should send
3331 * the packet up to the controller in the hope that it will try to
3332 * set up a flow and thereby allow us to exit fail-open.
3334 * See the top-level comment in fail-open.c for more information.
3336 send_packet_in_miss(ofproto, packet, flow);
3340 /* Figures out whether a flow that missed in 'ofproto', whose details are in
3341 * 'miss' masked by 'wc', is likely to be worth tracking in detail in userspace
3342 * and (usually) installing a datapath flow. The answer is usually "yes" (a
3343 * return value of true). However, for short flows the cost of bookkeeping is
3344 * much higher than the benefits, so when the datapath holds a large number of
3345 * flows we impose some heuristics to decide which flows are likely to be worth
3348 flow_miss_should_make_facet(struct flow_miss *miss, struct flow_wildcards *wc)
3350 struct dpif_backer *backer = miss->ofproto->backer;
3353 if (!backer->governor) {
3356 n_subfacets = hmap_count(&backer->subfacets);
3357 if (n_subfacets * 2 <= flow_eviction_threshold) {
3361 backer->governor = governor_create();
3364 hash = flow_hash_in_wildcards(&miss->flow, wc, 0);
3365 return governor_should_install_flow(backer->governor, hash,
3366 list_size(&miss->packets));
3369 /* Handles 'miss' without creating a facet or subfacet or creating any datapath
3370 * flow. 'miss->flow' must have matched 'rule' and been xlated into 'xout'.
3371 * May add an "execute" operation to 'ops' and increment '*n_ops'. */
3373 handle_flow_miss_without_facet(struct rule_dpif *rule, struct xlate_out *xout,
3374 struct flow_miss *miss,
3375 struct flow_miss_op *ops, size_t *n_ops)
3377 struct ofpbuf *packet;
3379 LIST_FOR_EACH (packet, list_node, &miss->packets) {
3381 COVERAGE_INC(facet_suppress);
3383 handle_flow_miss_common(rule, packet, &miss->flow);
3386 struct xlate_in xin;
3388 xlate_in_init(&xin, miss->ofproto, &miss->flow, rule, 0, packet);
3389 xlate_actions_for_side_effects(&xin);
3392 if (xout->odp_actions.size) {
3393 struct flow_miss_op *op = &ops[*n_ops];
3394 struct dpif_execute *execute = &op->dpif_op.u.execute;
3396 init_flow_miss_execute_op(miss, packet, op);
3397 xlate_out_copy(&op->xout, xout);
3398 execute->actions = op->xout.odp_actions.data;
3399 execute->actions_len = op->xout.odp_actions.size;
3400 op->xout_garbage = true;
3407 /* Handles 'miss', which matches 'facet'. May add any required datapath
3408 * operations to 'ops', incrementing '*n_ops' for each new op.
3410 * All of the packets in 'miss' are considered to have arrived at time 'now'.
3411 * This is really important only for new facets: if we just called time_msec()
3412 * here, then the new subfacet or its packets could look (occasionally) as
3413 * though it was used some time after the facet was used. That can make a
3414 * one-packet flow look like it has a nonzero duration, which looks odd in
3415 * e.g. NetFlow statistics.
3417 * If non-null, 'stats' will be folded into 'facet'. */
3419 handle_flow_miss_with_facet(struct flow_miss *miss, struct facet *facet,
3420 long long int now, struct dpif_flow_stats *stats,
3421 struct flow_miss_op *ops, size_t *n_ops)
3423 struct ofproto_dpif *ofproto = ofproto_dpif_cast(facet->rule->up.ofproto);
3424 enum subfacet_path want_path;
3425 struct subfacet *subfacet;
3426 struct ofpbuf *packet;
3428 subfacet = subfacet_create(facet, miss, now);
3429 want_path = facet->xout.slow ? SF_SLOW_PATH : SF_FAST_PATH;
3431 subfacet_update_stats(subfacet, stats);
3434 LIST_FOR_EACH (packet, list_node, &miss->packets) {
3435 struct flow_miss_op *op = &ops[*n_ops];
3437 handle_flow_miss_common(facet->rule, packet, &miss->flow);
3439 if (want_path != SF_FAST_PATH) {
3440 struct xlate_in xin;
3442 xlate_in_init(&xin, ofproto, &miss->flow, facet->rule, 0, packet);
3443 xlate_actions_for_side_effects(&xin);
3446 if (facet->xout.odp_actions.size) {
3447 struct dpif_execute *execute = &op->dpif_op.u.execute;
3449 init_flow_miss_execute_op(miss, packet, op);
3450 execute->actions = facet->xout.odp_actions.data,
3451 execute->actions_len = facet->xout.odp_actions.size;
3456 if (miss->upcall_type == DPIF_UC_MISS || subfacet->path != want_path) {
3457 struct flow_miss_op *op = &ops[(*n_ops)++];
3458 struct dpif_flow_put *put = &op->dpif_op.u.flow_put;
3460 subfacet->path = want_path;
3462 op->xout_garbage = false;
3463 op->dpif_op.type = DPIF_OP_FLOW_PUT;
3464 put->flags = DPIF_FP_CREATE | DPIF_FP_MODIFY;
3465 put->key = miss->key;
3466 put->key_len = miss->key_len;
3469 if (want_path == SF_FAST_PATH) {
3470 put->actions = facet->xout.odp_actions.data;
3471 put->actions_len = facet->xout.odp_actions.size;
3473 compose_slow_path(ofproto, &miss->flow, facet->xout.slow,
3474 op->slow_stub, sizeof op->slow_stub,
3475 &put->actions, &put->actions_len);
3481 /* Handles flow miss 'miss'. May add any required datapath operations
3482 * to 'ops', incrementing '*n_ops' for each new op. */
3484 handle_flow_miss(struct flow_miss *miss, struct flow_miss_op *ops,
3487 struct ofproto_dpif *ofproto = miss->ofproto;
3488 struct dpif_flow_stats stats__;
3489 struct dpif_flow_stats *stats = &stats__;
3490 struct ofpbuf *packet;
3491 struct facet *facet;
3495 memset(stats, 0, sizeof *stats);
3497 LIST_FOR_EACH (packet, list_node, &miss->packets) {
3498 stats->tcp_flags |= packet_get_tcp_flags(packet, &miss->flow);
3499 stats->n_bytes += packet->size;
3503 facet = facet_lookup_valid(ofproto, &miss->flow);
3505 struct flow_wildcards wc;
3506 struct rule_dpif *rule;
3507 struct xlate_out xout;
3508 struct xlate_in xin;
3510 flow_wildcards_init_catchall(&wc);
3511 rule = rule_dpif_lookup(ofproto, &miss->flow, &wc);
3512 rule_credit_stats(rule, stats);
3514 xlate_in_init(&xin, ofproto, &miss->flow, rule, stats->tcp_flags,
3516 xin.resubmit_stats = stats;
3517 xin.may_learn = true;
3518 xlate_actions(&xin, &xout);
3519 flow_wildcards_or(&xout.wc, &xout.wc, &wc);
3521 /* There does not exist a bijection between 'struct flow' and datapath
3522 * flow keys with fitness ODP_FIT_TO_LITTLE. This breaks a fundamental
3523 * assumption used throughout the facet and subfacet handling code.
3524 * Since we have to handle these misses in userspace anyway, we simply
3525 * skip facet creation, avoiding the problem altogether. */
3526 if (miss->key_fitness == ODP_FIT_TOO_LITTLE
3527 || !flow_miss_should_make_facet(miss, &xout.wc)) {
3528 handle_flow_miss_without_facet(rule, &xout, miss, ops, n_ops);
3532 facet = facet_create(miss, rule, &xout, stats);
3535 handle_flow_miss_with_facet(miss, facet, now, stats, ops, n_ops);
3538 static struct drop_key *
3539 drop_key_lookup(const struct dpif_backer *backer, const struct nlattr *key,
3542 struct drop_key *drop_key;
3544 HMAP_FOR_EACH_WITH_HASH (drop_key, hmap_node, hash_bytes(key, key_len, 0),
3545 &backer->drop_keys) {
3546 if (drop_key->key_len == key_len
3547 && !memcmp(drop_key->key, key, key_len)) {
3555 drop_key_clear(struct dpif_backer *backer)
3557 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 15);
3558 struct drop_key *drop_key, *next;
3560 HMAP_FOR_EACH_SAFE (drop_key, next, hmap_node, &backer->drop_keys) {
3563 error = dpif_flow_del(backer->dpif, drop_key->key, drop_key->key_len,
3565 if (error && !VLOG_DROP_WARN(&rl)) {
3566 struct ds ds = DS_EMPTY_INITIALIZER;
3567 odp_flow_key_format(drop_key->key, drop_key->key_len, &ds);
3568 VLOG_WARN("Failed to delete drop key (%s) (%s)", strerror(error),
3573 hmap_remove(&backer->drop_keys, &drop_key->hmap_node);
3574 free(drop_key->key);
3579 /* Given a datpath, packet, and flow metadata ('backer', 'packet', and 'key'
3580 * respectively), populates 'flow' with the result of odp_flow_key_to_flow().
3581 * Optionally, if nonnull, populates 'fitnessp' with the fitness of 'flow' as
3582 * returned by odp_flow_key_to_flow(). Also, optionally populates 'ofproto'
3583 * with the ofproto_dpif, and 'odp_in_port' with the datapath in_port, that
3584 * 'packet' ingressed.
3586 * If 'ofproto' is nonnull, requires 'flow''s in_port to exist. Otherwise sets
3587 * 'flow''s in_port to OFPP_NONE.
3589 * This function does post-processing on data returned from
3590 * odp_flow_key_to_flow() to help make VLAN splinters transparent to the rest
3591 * of the upcall processing logic. In particular, if the extracted in_port is
3592 * a VLAN splinter port, it replaces flow->in_port by the "real" port, sets
3593 * flow->vlan_tci correctly for the VLAN of the VLAN splinter port, and pushes
3594 * a VLAN header onto 'packet' (if it is nonnull).
3596 * Similarly, this function also includes some logic to help with tunnels. It
3597 * may modify 'flow' as necessary to make the tunneling implementation
3598 * transparent to the upcall processing logic.
3600 * Returns 0 if successful, ENODEV if the parsed flow has no associated ofport,
3601 * or some other positive errno if there are other problems. */
3603 ofproto_receive(const struct dpif_backer *backer, struct ofpbuf *packet,
3604 const struct nlattr *key, size_t key_len,
3605 struct flow *flow, enum odp_key_fitness *fitnessp,
3606 struct ofproto_dpif **ofproto, odp_port_t *odp_in_port)
3608 const struct ofport_dpif *port;
3609 enum odp_key_fitness fitness;
3612 fitness = odp_flow_key_to_flow(key, key_len, flow);
3613 if (fitness == ODP_FIT_ERROR) {
3619 *odp_in_port = flow->in_port.odp_port;
3622 port = (tnl_port_should_receive(flow)
3623 ? ofport_dpif_cast(tnl_port_receive(flow))
3624 : odp_port_to_ofport(backer, flow->in_port.odp_port));
3625 flow->in_port.ofp_port = port ? port->up.ofp_port : OFPP_NONE;
3630 /* XXX: Since the tunnel module is not scoped per backer, for a tunnel port
3631 * it's theoretically possible that we'll receive an ofport belonging to an
3632 * entirely different datapath. In practice, this can't happen because no
3633 * platforms has two separate datapaths which each support tunneling. */
3634 ovs_assert(ofproto_dpif_cast(port->up.ofproto)->backer == backer);
3636 if (vsp_adjust_flow(ofproto_dpif_cast(port->up.ofproto), flow)) {
3638 /* Make the packet resemble the flow, so that it gets sent to
3639 * an OpenFlow controller properly, so that it looks correct
3640 * for sFlow, and so that flow_extract() will get the correct
3641 * vlan_tci if it is called on 'packet'.
3643 * The allocated space inside 'packet' probably also contains
3644 * 'key', that is, both 'packet' and 'key' are probably part of
3645 * a struct dpif_upcall (see the large comment on that
3646 * structure definition), so pushing data on 'packet' is in
3647 * general not a good idea since it could overwrite 'key' or
3648 * free it as a side effect. However, it's OK in this special
3649 * case because we know that 'packet' is inside a Netlink
3650 * attribute: pushing 4 bytes will just overwrite the 4-byte
3651 * "struct nlattr", which is fine since we don't need that
3652 * header anymore. */
3653 eth_push_vlan(packet, flow->vlan_tci);
3655 /* We can't reproduce 'key' from 'flow'. */
3656 fitness = fitness == ODP_FIT_PERFECT ? ODP_FIT_TOO_MUCH : fitness;
3661 *ofproto = ofproto_dpif_cast(port->up.ofproto);
3666 *fitnessp = fitness;
3672 handle_miss_upcalls(struct dpif_backer *backer, struct dpif_upcall *upcalls,
3675 struct dpif_upcall *upcall;
3676 struct flow_miss *miss;
3677 struct flow_miss misses[FLOW_MISS_MAX_BATCH];
3678 struct flow_miss_op flow_miss_ops[FLOW_MISS_MAX_BATCH * 2];
3679 struct dpif_op *dpif_ops[FLOW_MISS_MAX_BATCH * 2];
3689 /* Construct the to-do list.
3691 * This just amounts to extracting the flow from each packet and sticking
3692 * the packets that have the same flow in the same "flow_miss" structure so
3693 * that we can process them together. */
3696 for (upcall = upcalls; upcall < &upcalls[n_upcalls]; upcall++) {
3697 struct flow_miss *miss = &misses[n_misses];
3698 struct flow_miss *existing_miss;
3699 struct ofproto_dpif *ofproto;
3700 odp_port_t odp_in_port;
3705 error = ofproto_receive(backer, upcall->packet, upcall->key,
3706 upcall->key_len, &flow, &miss->key_fitness,
3707 &ofproto, &odp_in_port);
3708 if (error == ENODEV) {
3709 struct drop_key *drop_key;
3711 /* Received packet on datapath port for which we couldn't
3712 * associate an ofproto. This can happen if a port is removed
3713 * while traffic is being received. Print a rate-limited message
3714 * in case it happens frequently. Install a drop flow so
3715 * that future packets of the flow are inexpensively dropped
3717 VLOG_INFO_RL(&rl, "received packet on unassociated datapath port "
3718 "%"PRIu32, odp_in_port);
3720 drop_key = drop_key_lookup(backer, upcall->key, upcall->key_len);
3722 drop_key = xmalloc(sizeof *drop_key);
3723 drop_key->key = xmemdup(upcall->key, upcall->key_len);
3724 drop_key->key_len = upcall->key_len;
3726 hmap_insert(&backer->drop_keys, &drop_key->hmap_node,
3727 hash_bytes(drop_key->key, drop_key->key_len, 0));
3728 dpif_flow_put(backer->dpif, DPIF_FP_CREATE | DPIF_FP_MODIFY,
3729 drop_key->key, drop_key->key_len,
3730 NULL, 0, NULL, 0, NULL);
3738 ofproto->n_missed++;
3739 flow_extract(upcall->packet, flow.skb_priority, flow.skb_mark,
3740 &flow.tunnel, &flow.in_port, &miss->flow);
3742 /* Add other packets to a to-do list. */
3743 hash = flow_hash(&miss->flow, 0);
3744 existing_miss = flow_miss_find(&todo, ofproto, &miss->flow, hash);
3745 if (!existing_miss) {
3746 hmap_insert(&todo, &miss->hmap_node, hash);
3747 miss->ofproto = ofproto;
3748 miss->key = upcall->key;
3749 miss->key_len = upcall->key_len;
3750 miss->upcall_type = upcall->type;
3751 list_init(&miss->packets);
3755 miss = existing_miss;
3757 list_push_back(&miss->packets, &upcall->packet->list_node);
3760 /* Process each element in the to-do list, constructing the set of
3761 * operations to batch. */
3763 HMAP_FOR_EACH (miss, hmap_node, &todo) {
3764 handle_flow_miss(miss, flow_miss_ops, &n_ops);
3766 ovs_assert(n_ops <= ARRAY_SIZE(flow_miss_ops));
3768 /* Execute batch. */
3769 for (i = 0; i < n_ops; i++) {
3770 dpif_ops[i] = &flow_miss_ops[i].dpif_op;
3772 dpif_operate(backer->dpif, dpif_ops, n_ops);
3775 for (i = 0; i < n_ops; i++) {
3776 if (flow_miss_ops[i].xout_garbage) {
3777 xlate_out_uninit(&flow_miss_ops[i].xout);
3780 hmap_destroy(&todo);
3783 static enum { SFLOW_UPCALL, MISS_UPCALL, BAD_UPCALL, FLOW_SAMPLE_UPCALL,
3785 classify_upcall(const struct dpif_upcall *upcall)
3787 size_t userdata_len;
3788 union user_action_cookie cookie;
3790 /* First look at the upcall type. */
3791 switch (upcall->type) {
3792 case DPIF_UC_ACTION:
3798 case DPIF_N_UC_TYPES:
3800 VLOG_WARN_RL(&rl, "upcall has unexpected type %"PRIu32, upcall->type);
3804 /* "action" upcalls need a closer look. */
3805 if (!upcall->userdata) {
3806 VLOG_WARN_RL(&rl, "action upcall missing cookie");
3809 userdata_len = nl_attr_get_size(upcall->userdata);
3810 if (userdata_len < sizeof cookie.type
3811 || userdata_len > sizeof cookie) {
3812 VLOG_WARN_RL(&rl, "action upcall cookie has unexpected size %zu",
3816 memset(&cookie, 0, sizeof cookie);
3817 memcpy(&cookie, nl_attr_get(upcall->userdata), userdata_len);
3818 if (userdata_len == sizeof cookie.sflow
3819 && cookie.type == USER_ACTION_COOKIE_SFLOW) {
3820 return SFLOW_UPCALL;
3821 } else if (userdata_len == sizeof cookie.slow_path
3822 && cookie.type == USER_ACTION_COOKIE_SLOW_PATH) {
3824 } else if (userdata_len == sizeof cookie.flow_sample
3825 && cookie.type == USER_ACTION_COOKIE_FLOW_SAMPLE) {
3826 return FLOW_SAMPLE_UPCALL;
3827 } else if (userdata_len == sizeof cookie.ipfix
3828 && cookie.type == USER_ACTION_COOKIE_IPFIX) {
3829 return IPFIX_UPCALL;
3831 VLOG_WARN_RL(&rl, "invalid user cookie of type %"PRIu16
3832 " and size %zu", cookie.type, userdata_len);
3838 handle_sflow_upcall(struct dpif_backer *backer,
3839 const struct dpif_upcall *upcall)
3841 struct ofproto_dpif *ofproto;
3842 union user_action_cookie cookie;
3844 odp_port_t odp_in_port;
3846 if (ofproto_receive(backer, upcall->packet, upcall->key, upcall->key_len,
3847 &flow, NULL, &ofproto, &odp_in_port)
3848 || !ofproto->sflow) {
3852 memset(&cookie, 0, sizeof cookie);
3853 memcpy(&cookie, nl_attr_get(upcall->userdata), sizeof cookie.sflow);
3854 dpif_sflow_received(ofproto->sflow, upcall->packet, &flow,
3855 odp_in_port, &cookie);
3859 handle_flow_sample_upcall(struct dpif_backer *backer,
3860 const struct dpif_upcall *upcall)
3862 struct ofproto_dpif *ofproto;
3863 union user_action_cookie cookie;
3866 if (ofproto_receive(backer, upcall->packet, upcall->key, upcall->key_len,
3867 &flow, NULL, &ofproto, NULL)
3868 || !ofproto->ipfix) {
3872 memset(&cookie, 0, sizeof cookie);
3873 memcpy(&cookie, nl_attr_get(upcall->userdata), sizeof cookie.flow_sample);
3875 /* The flow reflects exactly the contents of the packet. Sample
3876 * the packet using it. */
3877 dpif_ipfix_flow_sample(ofproto->ipfix, upcall->packet, &flow,
3878 cookie.flow_sample.collector_set_id,
3879 cookie.flow_sample.probability,
3880 cookie.flow_sample.obs_domain_id,
3881 cookie.flow_sample.obs_point_id);
3885 handle_ipfix_upcall(struct dpif_backer *backer,
3886 const struct dpif_upcall *upcall)
3888 struct ofproto_dpif *ofproto;
3891 if (ofproto_receive(backer, upcall->packet, upcall->key, upcall->key_len,
3892 &flow, NULL, &ofproto, NULL)
3893 || !ofproto->ipfix) {
3897 /* The flow reflects exactly the contents of the packet. Sample
3898 * the packet using it. */
3899 dpif_ipfix_bridge_sample(ofproto->ipfix, upcall->packet, &flow);
3903 handle_upcalls(struct dpif_backer *backer, unsigned int max_batch)
3905 struct dpif_upcall misses[FLOW_MISS_MAX_BATCH];
3906 struct ofpbuf miss_bufs[FLOW_MISS_MAX_BATCH];
3907 uint64_t miss_buf_stubs[FLOW_MISS_MAX_BATCH][4096 / 8];
3912 ovs_assert(max_batch <= FLOW_MISS_MAX_BATCH);
3915 for (n_processed = 0; n_processed < max_batch; n_processed++) {
3916 struct dpif_upcall *upcall = &misses[n_misses];
3917 struct ofpbuf *buf = &miss_bufs[n_misses];
3920 ofpbuf_use_stub(buf, miss_buf_stubs[n_misses],
3921 sizeof miss_buf_stubs[n_misses]);
3922 error = dpif_recv(backer->dpif, upcall, buf);
3928 switch (classify_upcall(upcall)) {
3930 /* Handle it later. */
3935 handle_sflow_upcall(backer, upcall);
3939 case FLOW_SAMPLE_UPCALL:
3940 handle_flow_sample_upcall(backer, upcall);
3945 handle_ipfix_upcall(backer, upcall);
3955 /* Handle deferred MISS_UPCALL processing. */
3956 handle_miss_upcalls(backer, misses, n_misses);
3957 for (i = 0; i < n_misses; i++) {
3958 ofpbuf_uninit(&miss_bufs[i]);
3964 /* Flow expiration. */
3966 static int subfacet_max_idle(const struct dpif_backer *);
3967 static void update_stats(struct dpif_backer *);
3968 static void rule_expire(struct rule_dpif *);
3969 static void expire_subfacets(struct dpif_backer *, int dp_max_idle);
3971 /* This function is called periodically by run(). Its job is to collect
3972 * updates for the flows that have been installed into the datapath, most
3973 * importantly when they last were used, and then use that information to
3974 * expire flows that have not been used recently.
3976 * Returns the number of milliseconds after which it should be called again. */
3978 expire(struct dpif_backer *backer)
3980 struct ofproto_dpif *ofproto;
3984 /* Periodically clear out the drop keys in an effort to keep them
3985 * relatively few. */
3986 drop_key_clear(backer);
3988 /* Update stats for each flow in the backer. */
3989 update_stats(backer);
3991 n_subfacets = hmap_count(&backer->subfacets);
3993 struct subfacet *subfacet;
3994 long long int total, now;
3998 HMAP_FOR_EACH (subfacet, hmap_node, &backer->subfacets) {
3999 total += now - subfacet->created;
4001 backer->avg_subfacet_life += total / n_subfacets;
4003 backer->avg_subfacet_life /= 2;
4005 backer->avg_n_subfacet += n_subfacets;
4006 backer->avg_n_subfacet /= 2;
4008 backer->max_n_subfacet = MAX(backer->max_n_subfacet, n_subfacets);
4010 max_idle = subfacet_max_idle(backer);
4011 expire_subfacets(backer, max_idle);
4013 HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
4014 struct rule *rule, *next_rule;
4016 if (ofproto->backer != backer) {
4020 /* Expire OpenFlow flows whose idle_timeout or hard_timeout
4022 LIST_FOR_EACH_SAFE (rule, next_rule, expirable,
4023 &ofproto->up.expirable) {
4024 rule_expire(rule_dpif_cast(rule));
4027 /* All outstanding data in existing flows has been accounted, so it's a
4028 * good time to do bond rebalancing. */
4029 if (ofproto->has_bonded_bundles) {
4030 struct ofbundle *bundle;
4032 HMAP_FOR_EACH (bundle, hmap_node, &ofproto->bundles) {
4034 bond_rebalance(bundle->bond, &backer->revalidate_set);
4040 return MIN(max_idle, 1000);
4043 /* Updates flow table statistics given that the datapath just reported 'stats'
4044 * as 'subfacet''s statistics. */
4046 update_subfacet_stats(struct subfacet *subfacet,
4047 const struct dpif_flow_stats *stats)
4049 struct facet *facet = subfacet->facet;
4050 struct ofproto_dpif *ofproto = ofproto_dpif_cast(facet->rule->up.ofproto);
4051 struct dpif_flow_stats diff;
4053 diff.tcp_flags = stats->tcp_flags;
4054 diff.used = stats->used;
4056 if (stats->n_packets >= subfacet->dp_packet_count) {
4057 diff.n_packets = stats->n_packets - subfacet->dp_packet_count;
4059 VLOG_WARN_RL(&rl, "unexpected packet count from the datapath");
4063 if (stats->n_bytes >= subfacet->dp_byte_count) {
4064 diff.n_bytes = stats->n_bytes - subfacet->dp_byte_count;
4066 VLOG_WARN_RL(&rl, "unexpected byte count from datapath");
4070 ofproto->n_hit += diff.n_packets;
4071 subfacet->dp_packet_count = stats->n_packets;
4072 subfacet->dp_byte_count = stats->n_bytes;
4073 subfacet_update_stats(subfacet, &diff);
4075 if (facet->accounted_bytes < facet->byte_count) {
4077 facet_account(facet);
4078 facet->accounted_bytes = facet->byte_count;
4082 /* 'key' with length 'key_len' bytes is a flow in 'dpif' that we know nothing
4083 * about, or a flow that shouldn't be installed but was anyway. Delete it. */
4085 delete_unexpected_flow(struct dpif_backer *backer,
4086 const struct nlattr *key, size_t key_len)
4088 if (!VLOG_DROP_WARN(&rl)) {
4092 odp_flow_key_format(key, key_len, &s);
4093 VLOG_WARN("unexpected flow: %s", ds_cstr(&s));
4097 COVERAGE_INC(facet_unexpected);
4098 dpif_flow_del(backer->dpif, key, key_len, NULL);
4101 /* Update 'packet_count', 'byte_count', and 'used' members of installed facets.
4103 * This function also pushes statistics updates to rules which each facet
4104 * resubmits into. Generally these statistics will be accurate. However, if a
4105 * facet changes the rule it resubmits into at some time in between
4106 * update_stats() runs, it is possible that statistics accrued to the
4107 * old rule will be incorrectly attributed to the new rule. This could be
4108 * avoided by calling update_stats() whenever rules are created or
4109 * deleted. However, the performance impact of making so many calls to the
4110 * datapath do not justify the benefit of having perfectly accurate statistics.
4112 * In addition, this function maintains per ofproto flow hit counts. The patch
4113 * port is not treated specially. e.g. A packet ingress from br0 patched into
4114 * br1 will increase the hit count of br0 by 1, however, does not affect
4115 * the hit or miss counts of br1.
4118 update_stats(struct dpif_backer *backer)
4120 const struct dpif_flow_stats *stats;
4121 struct dpif_flow_dump dump;
4122 const struct nlattr *key;
4125 dpif_flow_dump_start(&dump, backer->dpif);
4126 while (dpif_flow_dump_next(&dump, &key, &key_len,
4127 NULL, NULL, NULL, NULL, &stats)) {
4128 struct subfacet *subfacet;
4131 key_hash = odp_flow_key_hash(key, key_len);
4132 subfacet = subfacet_find(backer, key, key_len, key_hash);
4133 switch (subfacet ? subfacet->path : SF_NOT_INSTALLED) {
4135 update_subfacet_stats(subfacet, stats);
4139 /* Stats are updated per-packet. */
4142 case SF_NOT_INSTALLED:
4144 delete_unexpected_flow(backer, key, key_len);
4149 dpif_flow_dump_done(&dump);
4151 update_moving_averages(backer);
4154 /* Calculates and returns the number of milliseconds of idle time after which
4155 * subfacets should expire from the datapath. When a subfacet expires, we fold
4156 * its statistics into its facet, and when a facet's last subfacet expires, we
4157 * fold its statistic into its rule. */
4159 subfacet_max_idle(const struct dpif_backer *backer)
4162 * Idle time histogram.
4164 * Most of the time a switch has a relatively small number of subfacets.
4165 * When this is the case we might as well keep statistics for all of them
4166 * in userspace and to cache them in the kernel datapath for performance as
4169 * As the number of subfacets increases, the memory required to maintain
4170 * statistics about them in userspace and in the kernel becomes
4171 * significant. However, with a large number of subfacets it is likely
4172 * that only a few of them are "heavy hitters" that consume a large amount
4173 * of bandwidth. At this point, only heavy hitters are worth caching in
4174 * the kernel and maintaining in userspaces; other subfacets we can
4177 * The technique used to compute the idle time is to build a histogram with
4178 * N_BUCKETS buckets whose width is BUCKET_WIDTH msecs each. Each subfacet
4179 * that is installed in the kernel gets dropped in the appropriate bucket.
4180 * After the histogram has been built, we compute the cutoff so that only
4181 * the most-recently-used 1% of subfacets (but at least
4182 * flow_eviction_threshold flows) are kept cached. At least
4183 * the most-recently-used bucket of subfacets is kept, so actually an
4184 * arbitrary number of subfacets can be kept in any given expiration run
4185 * (though the next run will delete most of those unless they receive
4188 * This requires a second pass through the subfacets, in addition to the
4189 * pass made by update_stats(), because the former function never looks at
4190 * uninstallable subfacets.
4192 enum { BUCKET_WIDTH = ROUND_UP(100, TIME_UPDATE_INTERVAL) };
4193 enum { N_BUCKETS = 5000 / BUCKET_WIDTH };
4194 int buckets[N_BUCKETS] = { 0 };
4195 int total, subtotal, bucket;
4196 struct subfacet *subfacet;
4200 total = hmap_count(&backer->subfacets);
4201 if (total <= flow_eviction_threshold) {
4202 return N_BUCKETS * BUCKET_WIDTH;
4205 /* Build histogram. */
4207 HMAP_FOR_EACH (subfacet, hmap_node, &backer->subfacets) {
4208 long long int idle = now - subfacet->used;
4209 int bucket = (idle <= 0 ? 0
4210 : idle >= BUCKET_WIDTH * N_BUCKETS ? N_BUCKETS - 1
4211 : (unsigned int) idle / BUCKET_WIDTH);
4215 /* Find the first bucket whose flows should be expired. */
4216 subtotal = bucket = 0;
4218 subtotal += buckets[bucket++];
4219 } while (bucket < N_BUCKETS &&
4220 subtotal < MAX(flow_eviction_threshold, total / 100));
4222 if (VLOG_IS_DBG_ENABLED()) {
4226 ds_put_cstr(&s, "keep");
4227 for (i = 0; i < N_BUCKETS; i++) {
4229 ds_put_cstr(&s, ", drop");
4232 ds_put_format(&s, " %d:%d", i * BUCKET_WIDTH, buckets[i]);
4235 VLOG_INFO("%s (msec:count)", ds_cstr(&s));
4239 return bucket * BUCKET_WIDTH;
4243 expire_subfacets(struct dpif_backer *backer, int dp_max_idle)
4245 /* Cutoff time for most flows. */
4246 long long int normal_cutoff = time_msec() - dp_max_idle;
4248 /* We really want to keep flows for special protocols around, so use a more
4249 * conservative cutoff. */
4250 long long int special_cutoff = time_msec() - 10000;
4252 struct subfacet *subfacet, *next_subfacet;
4253 struct subfacet *batch[SUBFACET_DESTROY_MAX_BATCH];
4257 HMAP_FOR_EACH_SAFE (subfacet, next_subfacet, hmap_node,
4258 &backer->subfacets) {
4259 long long int cutoff;
4261 cutoff = (subfacet->facet->xout.slow & (SLOW_CFM | SLOW_BFD | SLOW_LACP
4265 if (subfacet->used < cutoff) {
4266 if (subfacet->path != SF_NOT_INSTALLED) {
4267 batch[n_batch++] = subfacet;
4268 if (n_batch >= SUBFACET_DESTROY_MAX_BATCH) {
4269 subfacet_destroy_batch(backer, batch, n_batch);
4273 subfacet_destroy(subfacet);
4279 subfacet_destroy_batch(backer, batch, n_batch);
4283 /* If 'rule' is an OpenFlow rule, that has expired according to OpenFlow rules,
4284 * then delete it entirely. */
4286 rule_expire(struct rule_dpif *rule)
4288 struct facet *facet, *next_facet;
4292 if (rule->up.pending) {
4293 /* We'll have to expire it later. */
4297 /* Has 'rule' expired? */
4299 if (rule->up.hard_timeout
4300 && now > rule->up.modified + rule->up.hard_timeout * 1000) {
4301 reason = OFPRR_HARD_TIMEOUT;
4302 } else if (rule->up.idle_timeout
4303 && now > rule->up.used + rule->up.idle_timeout * 1000) {
4304 reason = OFPRR_IDLE_TIMEOUT;
4309 COVERAGE_INC(ofproto_dpif_expired);
4311 /* Update stats. (This is a no-op if the rule expired due to an idle
4312 * timeout, because that only happens when the rule has no facets left.) */
4313 LIST_FOR_EACH_SAFE (facet, next_facet, list_node, &rule->facets) {
4314 facet_remove(facet);
4317 /* Get rid of the rule. */
4318 ofproto_rule_expire(&rule->up, reason);
4323 /* Creates and returns a new facet based on 'miss'.
4325 * The caller must already have determined that no facet with an identical
4326 * 'miss->flow' exists in 'miss->ofproto'.
4328 * 'rule' and 'xout' must have been created based on 'miss'.
4330 * 'facet'' statistics are initialized based on 'stats'.
4332 * The facet will initially have no subfacets. The caller should create (at
4333 * least) one subfacet with subfacet_create(). */
4334 static struct facet *
4335 facet_create(const struct flow_miss *miss, struct rule_dpif *rule,
4336 struct xlate_out *xout, struct dpif_flow_stats *stats)
4338 struct ofproto_dpif *ofproto = miss->ofproto;
4339 struct facet *facet;
4342 facet = xzalloc(sizeof *facet);
4343 facet->packet_count = facet->prev_packet_count = stats->n_packets;
4344 facet->byte_count = facet->prev_byte_count = stats->n_bytes;
4345 facet->tcp_flags = stats->tcp_flags;
4346 facet->used = stats->used;
4347 facet->flow = miss->flow;
4348 facet->learn_rl = time_msec() + 500;
4351 list_push_back(&facet->rule->facets, &facet->list_node);
4352 list_init(&facet->subfacets);
4353 netflow_flow_init(&facet->nf_flow);
4354 netflow_flow_update_time(ofproto->netflow, &facet->nf_flow, facet->used);
4356 xlate_out_copy(&facet->xout, xout);
4358 match_init(&match, &facet->flow, &facet->xout.wc);
4359 cls_rule_init(&facet->cr, &match, OFP_DEFAULT_PRIORITY);
4360 classifier_insert(&ofproto->facets, &facet->cr);
4362 facet->nf_flow.output_iface = facet->xout.nf_output_iface;
4368 facet_free(struct facet *facet)
4371 xlate_out_uninit(&facet->xout);
4376 /* Executes, within 'ofproto', the 'n_actions' actions in 'actions' on
4377 * 'packet', which arrived on 'in_port'. */
4379 execute_odp_actions(struct ofproto_dpif *ofproto, const struct flow *flow,
4380 const struct nlattr *odp_actions, size_t actions_len,
4381 struct ofpbuf *packet)
4383 struct odputil_keybuf keybuf;
4387 ofpbuf_use_stack(&key, &keybuf, sizeof keybuf);
4388 odp_flow_key_from_flow(&key, flow,
4389 ofp_port_to_odp_port(ofproto, flow->in_port.ofp_port));
4391 error = dpif_execute(ofproto->backer->dpif, key.data, key.size,
4392 odp_actions, actions_len, packet);
4396 /* Remove 'facet' from its ofproto and free up the associated memory:
4398 * - If 'facet' was installed in the datapath, uninstalls it and updates its
4399 * rule's statistics, via subfacet_uninstall().
4401 * - Removes 'facet' from its rule and from ofproto->facets.
4404 facet_remove(struct facet *facet)
4406 struct ofproto_dpif *ofproto = ofproto_dpif_cast(facet->rule->up.ofproto);
4407 struct subfacet *subfacet, *next_subfacet;
4409 ovs_assert(!list_is_empty(&facet->subfacets));
4411 /* First uninstall all of the subfacets to get final statistics. */
4412 LIST_FOR_EACH (subfacet, list_node, &facet->subfacets) {
4413 subfacet_uninstall(subfacet);
4416 /* Flush the final stats to the rule.
4418 * This might require us to have at least one subfacet around so that we
4419 * can use its actions for accounting in facet_account(), which is why we
4420 * have uninstalled but not yet destroyed the subfacets. */
4421 facet_flush_stats(facet);
4423 /* Now we're really all done so destroy everything. */
4424 LIST_FOR_EACH_SAFE (subfacet, next_subfacet, list_node,
4425 &facet->subfacets) {
4426 subfacet_destroy__(subfacet);
4428 classifier_remove(&ofproto->facets, &facet->cr);
4429 cls_rule_destroy(&facet->cr);
4430 list_remove(&facet->list_node);
4434 /* Feed information from 'facet' back into the learning table to keep it in
4435 * sync with what is actually flowing through the datapath. */
4437 facet_learn(struct facet *facet)
4439 long long int now = time_msec();
4441 if (!facet->xout.has_fin_timeout && now < facet->learn_rl) {
4445 facet->learn_rl = now + 500;
4447 if (!facet->xout.has_learn
4448 && !facet->xout.has_normal
4449 && (!facet->xout.has_fin_timeout
4450 || !(facet->tcp_flags & (TCP_FIN | TCP_RST)))) {
4454 facet_push_stats(facet, true);
4458 facet_account(struct facet *facet)
4460 struct ofproto_dpif *ofproto = ofproto_dpif_cast(facet->rule->up.ofproto);
4461 const struct nlattr *a;
4466 if (!facet->xout.has_normal || !ofproto->has_bonded_bundles) {
4469 n_bytes = facet->byte_count - facet->accounted_bytes;
4471 /* This loop feeds byte counters to bond_account() for rebalancing to use
4472 * as a basis. We also need to track the actual VLAN on which the packet
4473 * is going to be sent to ensure that it matches the one passed to
4474 * bond_choose_output_slave(). (Otherwise, we will account to the wrong
4477 * We use the actions from an arbitrary subfacet because they should all
4478 * be equally valid for our purpose. */
4479 vlan_tci = facet->flow.vlan_tci;
4480 NL_ATTR_FOR_EACH_UNSAFE (a, left, facet->xout.odp_actions.data,
4481 facet->xout.odp_actions.size) {
4482 const struct ovs_action_push_vlan *vlan;
4483 struct ofport_dpif *port;
4485 switch (nl_attr_type(a)) {
4486 case OVS_ACTION_ATTR_OUTPUT:
4487 port = get_odp_port(ofproto, nl_attr_get_odp_port(a));
4488 if (port && port->bundle && port->bundle->bond) {
4489 bond_account(port->bundle->bond, &facet->flow,
4490 vlan_tci_to_vid(vlan_tci), n_bytes);
4494 case OVS_ACTION_ATTR_POP_VLAN:
4495 vlan_tci = htons(0);
4498 case OVS_ACTION_ATTR_PUSH_VLAN:
4499 vlan = nl_attr_get(a);
4500 vlan_tci = vlan->vlan_tci;
4506 /* Returns true if the only action for 'facet' is to send to the controller.
4507 * (We don't report NetFlow expiration messages for such facets because they
4508 * are just part of the control logic for the network, not real traffic). */
4510 facet_is_controller_flow(struct facet *facet)
4513 const struct rule *rule = &facet->rule->up;
4514 const struct ofpact *ofpacts = rule->ofpacts;
4515 size_t ofpacts_len = rule->ofpacts_len;
4517 if (ofpacts_len > 0 &&
4518 ofpacts->type == OFPACT_CONTROLLER &&
4519 ofpact_next(ofpacts) >= ofpact_end(ofpacts, ofpacts_len)) {
4526 /* Folds all of 'facet''s statistics into its rule. Also updates the
4527 * accounting ofhook and emits a NetFlow expiration if appropriate. All of
4528 * 'facet''s statistics in the datapath should have been zeroed and folded into
4529 * its packet and byte counts before this function is called. */
4531 facet_flush_stats(struct facet *facet)
4533 struct ofproto_dpif *ofproto = ofproto_dpif_cast(facet->rule->up.ofproto);
4534 struct subfacet *subfacet;
4536 LIST_FOR_EACH (subfacet, list_node, &facet->subfacets) {
4537 ovs_assert(!subfacet->dp_byte_count);
4538 ovs_assert(!subfacet->dp_packet_count);
4541 facet_push_stats(facet, false);
4542 if (facet->accounted_bytes < facet->byte_count) {
4543 facet_account(facet);
4544 facet->accounted_bytes = facet->byte_count;
4547 if (ofproto->netflow && !facet_is_controller_flow(facet)) {
4548 struct ofexpired expired;
4549 expired.flow = facet->flow;
4550 expired.packet_count = facet->packet_count;
4551 expired.byte_count = facet->byte_count;
4552 expired.used = facet->used;
4553 netflow_expire(ofproto->netflow, &facet->nf_flow, &expired);
4556 /* Reset counters to prevent double counting if 'facet' ever gets
4558 facet_reset_counters(facet);
4560 netflow_flow_clear(&facet->nf_flow);
4561 facet->tcp_flags = 0;
4564 /* Searches 'ofproto''s table of facets for one which would be responsible for
4565 * 'flow'. Returns it if found, otherwise a null pointer.
4567 * The returned facet might need revalidation; use facet_lookup_valid()
4568 * instead if that is important. */
4569 static struct facet *
4570 facet_find(struct ofproto_dpif *ofproto, const struct flow *flow)
4572 struct cls_rule *cr = classifier_lookup(&ofproto->facets, flow, NULL);
4573 return cr ? CONTAINER_OF(cr, struct facet, cr) : NULL;
4576 /* Searches 'ofproto''s table of facets for one capable that covers
4577 * 'flow'. Returns it if found, otherwise a null pointer.
4579 * The returned facet is guaranteed to be valid. */
4580 static struct facet *
4581 facet_lookup_valid(struct ofproto_dpif *ofproto, const struct flow *flow)
4583 struct facet *facet;
4585 facet = facet_find(ofproto, flow);
4587 && (ofproto->backer->need_revalidate
4588 || tag_set_intersects(&ofproto->backer->revalidate_set,
4590 && !facet_revalidate(facet)) {
4598 facet_check_consistency(struct facet *facet)
4600 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 15);
4602 struct ofproto_dpif *ofproto = ofproto_dpif_cast(facet->rule->up.ofproto);
4604 struct xlate_out xout;
4605 struct xlate_in xin;
4607 struct rule_dpif *rule;
4610 /* Check the rule for consistency. */
4611 rule = rule_dpif_lookup(ofproto, &facet->flow, NULL);
4612 if (rule != facet->rule) {
4613 if (!VLOG_DROP_WARN(&rl)) {
4614 struct ds s = DS_EMPTY_INITIALIZER;
4616 flow_format(&s, &facet->flow);
4617 ds_put_format(&s, ": facet associated with wrong rule (was "
4618 "table=%"PRIu8",", facet->rule->up.table_id);
4619 cls_rule_format(&facet->rule->up.cr, &s);
4620 ds_put_format(&s, ") (should have been table=%"PRIu8",",
4622 cls_rule_format(&rule->up.cr, &s);
4623 ds_put_char(&s, ')');
4625 VLOG_WARN("%s", ds_cstr(&s));
4631 /* Check the datapath actions for consistency. */
4632 xlate_in_init(&xin, ofproto, &facet->flow, rule, 0, NULL);
4633 xlate_actions(&xin, &xout);
4635 ok = ofpbuf_equal(&facet->xout.odp_actions, &xout.odp_actions)
4636 && facet->xout.slow == xout.slow;
4637 if (!ok && !VLOG_DROP_WARN(&rl)) {
4638 struct ds s = DS_EMPTY_INITIALIZER;
4640 flow_format(&s, &facet->flow);
4641 ds_put_cstr(&s, ": inconsistency in facet");
4643 if (!ofpbuf_equal(&facet->xout.odp_actions, &xout.odp_actions)) {
4644 ds_put_cstr(&s, " (actions were: ");
4645 format_odp_actions(&s, facet->xout.odp_actions.data,
4646 facet->xout.odp_actions.size);
4647 ds_put_cstr(&s, ") (correct actions: ");
4648 format_odp_actions(&s, xout.odp_actions.data,
4649 xout.odp_actions.size);
4650 ds_put_char(&s, ')');
4653 if (facet->xout.slow != xout.slow) {
4654 ds_put_format(&s, " slow path incorrect. should be %d", xout.slow);
4657 VLOG_WARN("%s", ds_cstr(&s));
4660 xlate_out_uninit(&xout);
4665 /* Re-searches the classifier for 'facet':
4667 * - If the rule found is different from 'facet''s current rule, moves
4668 * 'facet' to the new rule and recompiles its actions.
4670 * - If the rule found is the same as 'facet''s current rule, leaves 'facet'
4671 * where it is and recompiles its actions anyway.
4673 * - If any of 'facet''s subfacets correspond to a new flow according to
4674 * ofproto_receive(), 'facet' is removed.
4676 * Returns true if 'facet' is still valid. False if 'facet' was removed. */
4678 facet_revalidate(struct facet *facet)
4680 struct ofproto_dpif *ofproto = ofproto_dpif_cast(facet->rule->up.ofproto);
4681 struct rule_dpif *new_rule;
4682 struct subfacet *subfacet;
4683 struct flow_wildcards wc;
4684 struct xlate_out xout;
4685 struct xlate_in xin;
4687 COVERAGE_INC(facet_revalidate);
4689 /* Check that child subfacets still correspond to this facet. Tunnel
4690 * configuration changes could cause a subfacet's OpenFlow in_port to
4692 LIST_FOR_EACH (subfacet, list_node, &facet->subfacets) {
4693 struct ofproto_dpif *recv_ofproto;
4694 struct flow recv_flow;
4697 error = ofproto_receive(ofproto->backer, NULL, subfacet->key,
4698 subfacet->key_len, &recv_flow, NULL,
4699 &recv_ofproto, NULL);
4701 || recv_ofproto != ofproto
4702 || facet != facet_find(ofproto, &recv_flow)) {
4703 facet_remove(facet);
4708 flow_wildcards_init_catchall(&wc);
4709 new_rule = rule_dpif_lookup(ofproto, &facet->flow, &wc);
4711 /* Calculate new datapath actions.
4713 * We do not modify any 'facet' state yet, because we might need to, e.g.,
4714 * emit a NetFlow expiration and, if so, we need to have the old state
4715 * around to properly compose it. */
4716 xlate_in_init(&xin, ofproto, &facet->flow, new_rule, 0, NULL);
4717 xlate_actions(&xin, &xout);
4718 flow_wildcards_or(&xout.wc, &xout.wc, &wc);
4720 /* A facet's slow path reason should only change under dramatic
4721 * circumstances. Rather than try to update everything, it's simpler to
4722 * remove the facet and start over.
4724 * More importantly, if a facet's wildcards change, it will be relatively
4725 * difficult to figure out if its subfacets still belong to it, and if not
4726 * which facet they may belong to. Again, to avoid the complexity, we
4727 * simply give up instead. */
4728 if (facet->xout.slow != xout.slow
4729 || memcmp(&facet->xout.wc, &xout.wc, sizeof xout.wc)) {
4730 facet_remove(facet);
4731 xlate_out_uninit(&xout);
4735 if (!ofpbuf_equal(&facet->xout.odp_actions, &xout.odp_actions)) {
4736 LIST_FOR_EACH(subfacet, list_node, &facet->subfacets) {
4737 if (subfacet->path == SF_FAST_PATH) {
4738 struct dpif_flow_stats stats;
4740 subfacet_install(subfacet, &xout.odp_actions, &stats);
4741 subfacet_update_stats(subfacet, &stats);
4745 facet_flush_stats(facet);
4747 ofpbuf_clear(&facet->xout.odp_actions);
4748 ofpbuf_put(&facet->xout.odp_actions, xout.odp_actions.data,
4749 xout.odp_actions.size);
4752 /* Update 'facet' now that we've taken care of all the old state. */
4753 facet->xout.tags = xout.tags;
4754 facet->xout.slow = xout.slow;
4755 facet->xout.has_learn = xout.has_learn;
4756 facet->xout.has_normal = xout.has_normal;
4757 facet->xout.has_fin_timeout = xout.has_fin_timeout;
4758 facet->xout.nf_output_iface = xout.nf_output_iface;
4759 facet->xout.mirrors = xout.mirrors;
4760 facet->nf_flow.output_iface = facet->xout.nf_output_iface;
4762 if (facet->rule != new_rule) {
4763 COVERAGE_INC(facet_changed_rule);
4764 list_remove(&facet->list_node);
4765 list_push_back(&new_rule->facets, &facet->list_node);
4766 facet->rule = new_rule;
4767 facet->used = new_rule->up.created;
4768 facet->prev_used = facet->used;
4771 xlate_out_uninit(&xout);
4776 facet_reset_counters(struct facet *facet)
4778 facet->packet_count = 0;
4779 facet->byte_count = 0;
4780 facet->prev_packet_count = 0;
4781 facet->prev_byte_count = 0;
4782 facet->accounted_bytes = 0;
4786 facet_push_stats(struct facet *facet, bool may_learn)
4788 struct dpif_flow_stats stats;
4790 ovs_assert(facet->packet_count >= facet->prev_packet_count);
4791 ovs_assert(facet->byte_count >= facet->prev_byte_count);
4792 ovs_assert(facet->used >= facet->prev_used);
4794 stats.n_packets = facet->packet_count - facet->prev_packet_count;
4795 stats.n_bytes = facet->byte_count - facet->prev_byte_count;
4796 stats.used = facet->used;
4797 stats.tcp_flags = facet->tcp_flags;
4799 if (may_learn || stats.n_packets || facet->used > facet->prev_used) {
4800 struct ofproto_dpif *ofproto =
4801 ofproto_dpif_cast(facet->rule->up.ofproto);
4803 struct ofport_dpif *in_port;
4804 struct xlate_in xin;
4806 facet->prev_packet_count = facet->packet_count;
4807 facet->prev_byte_count = facet->byte_count;
4808 facet->prev_used = facet->used;
4810 in_port = get_ofp_port(ofproto, facet->flow.in_port.ofp_port);
4811 if (in_port && in_port->tnl_port) {
4812 netdev_vport_inc_rx(in_port->up.netdev, &stats);
4815 rule_credit_stats(facet->rule, &stats);
4816 netflow_flow_update_time(ofproto->netflow, &facet->nf_flow,
4818 netflow_flow_update_flags(&facet->nf_flow, facet->tcp_flags);
4819 update_mirror_stats(ofproto, facet->xout.mirrors, stats.n_packets,
4822 xlate_in_init(&xin, ofproto, &facet->flow, facet->rule,
4823 stats.tcp_flags, NULL);
4824 xin.resubmit_stats = &stats;
4825 xin.may_learn = may_learn;
4826 xlate_actions_for_side_effects(&xin);
4831 push_all_stats__(bool run_fast)
4833 static long long int rl = LLONG_MIN;
4834 struct ofproto_dpif *ofproto;
4836 if (time_msec() < rl) {
4840 HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
4841 struct cls_cursor cursor;
4842 struct facet *facet;
4844 cls_cursor_init(&cursor, &ofproto->facets, NULL);
4845 CLS_CURSOR_FOR_EACH (facet, cr, &cursor) {
4846 facet_push_stats(facet, false);
4853 rl = time_msec() + 100;
4857 push_all_stats(void)
4859 push_all_stats__(true);
4863 rule_credit_stats(struct rule_dpif *rule, const struct dpif_flow_stats *stats)
4865 rule->packet_count += stats->n_packets;
4866 rule->byte_count += stats->n_bytes;
4867 ofproto_rule_update_used(&rule->up, stats->used);
4872 static struct subfacet *
4873 subfacet_find(struct dpif_backer *backer, const struct nlattr *key,
4874 size_t key_len, uint32_t key_hash)
4876 struct subfacet *subfacet;
4878 HMAP_FOR_EACH_WITH_HASH (subfacet, hmap_node, key_hash,
4879 &backer->subfacets) {
4880 if (subfacet->key_len == key_len
4881 && !memcmp(key, subfacet->key, key_len)) {
4889 /* Searches 'facet' (within 'ofproto') for a subfacet with the specified
4890 * 'key_fitness', 'key', and 'key_len' members in 'miss'. Returns the
4891 * existing subfacet if there is one, otherwise creates and returns a
4893 static struct subfacet *
4894 subfacet_create(struct facet *facet, struct flow_miss *miss,
4897 struct dpif_backer *backer = miss->ofproto->backer;
4898 enum odp_key_fitness key_fitness = miss->key_fitness;
4899 const struct nlattr *key = miss->key;
4900 size_t key_len = miss->key_len;
4902 struct subfacet *subfacet;
4904 key_hash = odp_flow_key_hash(key, key_len);
4906 if (list_is_empty(&facet->subfacets)) {
4907 subfacet = &facet->one_subfacet;
4909 subfacet = subfacet_find(backer, key, key_len, key_hash);
4911 if (subfacet->facet == facet) {
4915 /* This shouldn't happen. */
4916 VLOG_ERR_RL(&rl, "subfacet with wrong facet");
4917 subfacet_destroy(subfacet);
4920 subfacet = xmalloc(sizeof *subfacet);
4923 hmap_insert(&backer->subfacets, &subfacet->hmap_node, key_hash);
4924 list_push_back(&facet->subfacets, &subfacet->list_node);
4925 subfacet->facet = facet;
4926 subfacet->key_fitness = key_fitness;
4927 subfacet->key = xmemdup(key, key_len);
4928 subfacet->key_len = key_len;
4929 subfacet->used = now;
4930 subfacet->created = now;
4931 subfacet->dp_packet_count = 0;
4932 subfacet->dp_byte_count = 0;
4933 subfacet->path = SF_NOT_INSTALLED;
4934 subfacet->backer = backer;
4936 backer->subfacet_add_count++;
4940 /* Uninstalls 'subfacet' from the datapath, if it is installed, removes it from
4941 * its facet within 'ofproto', and frees it. */
4943 subfacet_destroy__(struct subfacet *subfacet)
4945 struct facet *facet = subfacet->facet;
4946 struct ofproto_dpif *ofproto = ofproto_dpif_cast(facet->rule->up.ofproto);
4948 /* Update ofproto stats before uninstall the subfacet. */
4949 ofproto->backer->subfacet_del_count++;
4951 subfacet_uninstall(subfacet);
4952 hmap_remove(&subfacet->backer->subfacets, &subfacet->hmap_node);
4953 list_remove(&subfacet->list_node);
4954 free(subfacet->key);
4955 if (subfacet != &facet->one_subfacet) {
4960 /* Destroys 'subfacet', as with subfacet_destroy__(), and then if this was the
4961 * last remaining subfacet in its facet destroys the facet too. */
4963 subfacet_destroy(struct subfacet *subfacet)
4965 struct facet *facet = subfacet->facet;
4967 if (list_is_singleton(&facet->subfacets)) {
4968 /* facet_remove() needs at least one subfacet (it will remove it). */
4969 facet_remove(facet);
4971 subfacet_destroy__(subfacet);
4976 subfacet_destroy_batch(struct dpif_backer *backer,
4977 struct subfacet **subfacets, int n)
4979 struct dpif_op ops[SUBFACET_DESTROY_MAX_BATCH];
4980 struct dpif_op *opsp[SUBFACET_DESTROY_MAX_BATCH];
4981 struct dpif_flow_stats stats[SUBFACET_DESTROY_MAX_BATCH];
4984 for (i = 0; i < n; i++) {
4985 ops[i].type = DPIF_OP_FLOW_DEL;
4986 ops[i].u.flow_del.key = subfacets[i]->key;
4987 ops[i].u.flow_del.key_len = subfacets[i]->key_len;
4988 ops[i].u.flow_del.stats = &stats[i];
4992 dpif_operate(backer->dpif, opsp, n);
4993 for (i = 0; i < n; i++) {
4994 subfacet_reset_dp_stats(subfacets[i], &stats[i]);
4995 subfacets[i]->path = SF_NOT_INSTALLED;
4996 subfacet_destroy(subfacets[i]);
5001 /* Updates 'subfacet''s datapath flow, setting its actions to 'actions_len'
5002 * bytes of actions in 'actions'. If 'stats' is non-null, statistics counters
5003 * in the datapath will be zeroed and 'stats' will be updated with traffic new
5004 * since 'subfacet' was last updated.
5006 * Returns 0 if successful, otherwise a positive errno value. */
5008 subfacet_install(struct subfacet *subfacet, const struct ofpbuf *odp_actions,
5009 struct dpif_flow_stats *stats)
5011 struct facet *facet = subfacet->facet;
5012 struct ofproto_dpif *ofproto = ofproto_dpif_cast(facet->rule->up.ofproto);
5013 enum subfacet_path path = facet->xout.slow ? SF_SLOW_PATH : SF_FAST_PATH;
5014 const struct nlattr *actions = odp_actions->data;
5015 size_t actions_len = odp_actions->size;
5017 uint64_t slow_path_stub[128 / 8];
5018 enum dpif_flow_put_flags flags;
5021 flags = DPIF_FP_CREATE | DPIF_FP_MODIFY;
5023 flags |= DPIF_FP_ZERO_STATS;
5026 if (path == SF_SLOW_PATH) {
5027 compose_slow_path(ofproto, &facet->flow, facet->xout.slow,
5028 slow_path_stub, sizeof slow_path_stub,
5029 &actions, &actions_len);
5032 ret = dpif_flow_put(ofproto->backer->dpif, flags, subfacet->key,
5033 subfacet->key_len, NULL, 0,
5034 actions, actions_len, stats);
5037 subfacet_reset_dp_stats(subfacet, stats);
5041 subfacet->path = path;
5046 /* If 'subfacet' is installed in the datapath, uninstalls it. */
5048 subfacet_uninstall(struct subfacet *subfacet)
5050 if (subfacet->path != SF_NOT_INSTALLED) {
5051 struct rule_dpif *rule = subfacet->facet->rule;
5052 struct ofproto_dpif *ofproto = ofproto_dpif_cast(rule->up.ofproto);
5053 struct dpif_flow_stats stats;
5056 error = dpif_flow_del(ofproto->backer->dpif, subfacet->key,
5057 subfacet->key_len, &stats);
5058 subfacet_reset_dp_stats(subfacet, &stats);
5060 subfacet_update_stats(subfacet, &stats);
5062 subfacet->path = SF_NOT_INSTALLED;
5064 ovs_assert(subfacet->dp_packet_count == 0);
5065 ovs_assert(subfacet->dp_byte_count == 0);
5069 /* Resets 'subfacet''s datapath statistics counters. This should be called
5070 * when 'subfacet''s statistics are cleared in the datapath. If 'stats' is
5071 * non-null, it should contain the statistics returned by dpif when 'subfacet'
5072 * was reset in the datapath. 'stats' will be modified to include only
5073 * statistics new since 'subfacet' was last updated. */
5075 subfacet_reset_dp_stats(struct subfacet *subfacet,
5076 struct dpif_flow_stats *stats)
5079 && subfacet->dp_packet_count <= stats->n_packets
5080 && subfacet->dp_byte_count <= stats->n_bytes) {
5081 stats->n_packets -= subfacet->dp_packet_count;
5082 stats->n_bytes -= subfacet->dp_byte_count;
5085 subfacet->dp_packet_count = 0;
5086 subfacet->dp_byte_count = 0;
5089 /* Folds the statistics from 'stats' into the counters in 'subfacet'.
5091 * Because of the meaning of a subfacet's counters, it only makes sense to do
5092 * this if 'stats' are not tracked in the datapath, that is, if 'stats'
5093 * represents a packet that was sent by hand or if it represents statistics
5094 * that have been cleared out of the datapath. */
5096 subfacet_update_stats(struct subfacet *subfacet,
5097 const struct dpif_flow_stats *stats)
5099 if (stats->n_packets || stats->used > subfacet->used) {
5100 struct facet *facet = subfacet->facet;
5102 subfacet->used = MAX(subfacet->used, stats->used);
5103 facet->used = MAX(facet->used, stats->used);
5104 facet->packet_count += stats->n_packets;
5105 facet->byte_count += stats->n_bytes;
5106 facet->tcp_flags |= stats->tcp_flags;
5112 /* Lookup 'flow' in 'ofproto''s classifier. If 'wc' is non-null, sets
5113 * the fields that were relevant as part of the lookup. */
5114 static struct rule_dpif *
5115 rule_dpif_lookup(struct ofproto_dpif *ofproto, const struct flow *flow,
5116 struct flow_wildcards *wc)
5118 struct rule_dpif *rule;
5120 rule = rule_dpif_lookup_in_table(ofproto, flow, wc, 0);
5125 return rule_dpif_miss_rule(ofproto, flow);
5129 rule_dpif_lookup_in_table(struct ofproto_dpif *ofproto,
5130 const struct flow *flow, struct flow_wildcards *wc,
5133 struct cls_rule *cls_rule;
5134 struct classifier *cls;
5137 if (table_id >= N_TABLES) {
5142 wc->masks.nw_frag |= FLOW_NW_FRAG_MASK;
5145 cls = &ofproto->up.tables[table_id].cls;
5146 frag = (flow->nw_frag & FLOW_NW_FRAG_ANY) != 0;
5147 if (frag && ofproto->up.frag_handling == OFPC_FRAG_NORMAL) {
5148 /* We must pretend that transport ports are unavailable. */
5149 struct flow ofpc_normal_flow = *flow;
5150 ofpc_normal_flow.tp_src = htons(0);
5151 ofpc_normal_flow.tp_dst = htons(0);
5152 cls_rule = classifier_lookup(cls, &ofpc_normal_flow, wc);
5153 } else if (frag && ofproto->up.frag_handling == OFPC_FRAG_DROP) {
5154 cls_rule = &ofproto->drop_frags_rule->up.cr;
5156 flow_wildcards_init_exact(wc);
5159 cls_rule = classifier_lookup(cls, flow, wc);
5161 return rule_dpif_cast(rule_from_cls_rule(cls_rule));
5165 rule_dpif_miss_rule(struct ofproto_dpif *ofproto, const struct flow *flow)
5167 struct ofport_dpif *port;
5169 port = get_ofp_port(ofproto, flow->in_port.ofp_port);
5171 VLOG_WARN_RL(&rl, "packet-in on unknown OpenFlow port %"PRIu16,
5172 flow->in_port.ofp_port);
5173 return ofproto->miss_rule;
5176 if (port->up.pp.config & OFPUTIL_PC_NO_PACKET_IN) {
5177 return ofproto->no_packet_in_rule;
5179 return ofproto->miss_rule;
5183 complete_operation(struct rule_dpif *rule)
5185 struct ofproto_dpif *ofproto = ofproto_dpif_cast(rule->up.ofproto);
5187 rule_invalidate(rule);
5189 struct dpif_completion *c = xmalloc(sizeof *c);
5190 c->op = rule->up.pending;
5191 list_push_back(&ofproto->completions, &c->list_node);
5193 ofoperation_complete(rule->up.pending, 0);
5197 static struct rule *
5200 struct rule_dpif *rule = xmalloc(sizeof *rule);
5205 rule_dealloc(struct rule *rule_)
5207 struct rule_dpif *rule = rule_dpif_cast(rule_);
5212 rule_construct(struct rule *rule_)
5214 struct rule_dpif *rule = rule_dpif_cast(rule_);
5215 struct ofproto_dpif *ofproto = ofproto_dpif_cast(rule->up.ofproto);
5216 struct rule_dpif *victim;
5219 rule->packet_count = 0;
5220 rule->byte_count = 0;
5222 victim = rule_dpif_cast(ofoperation_get_victim(rule->up.pending));
5223 if (victim && !list_is_empty(&victim->facets)) {
5224 struct facet *facet;
5226 rule->facets = victim->facets;
5227 list_moved(&rule->facets);
5228 LIST_FOR_EACH (facet, list_node, &rule->facets) {
5229 /* XXX: We're only clearing our local counters here. It's possible
5230 * that quite a few packets are unaccounted for in the datapath
5231 * statistics. These will be accounted to the new rule instead of
5232 * cleared as required. This could be fixed by clearing out the
5233 * datapath statistics for this facet, but currently it doesn't
5235 facet_reset_counters(facet);
5239 /* Must avoid list_moved() in this case. */
5240 list_init(&rule->facets);
5243 table_id = rule->up.table_id;
5245 rule->tag = victim->tag;
5246 } else if (table_id == 0) {
5251 miniflow_expand(&rule->up.cr.match.flow, &flow);
5252 rule->tag = rule_calculate_tag(&flow, &rule->up.cr.match.mask,
5253 ofproto->tables[table_id].basis);
5256 complete_operation(rule);
5261 rule_destruct(struct rule *rule_)
5263 struct rule_dpif *rule = rule_dpif_cast(rule_);
5264 struct facet *facet, *next_facet;
5266 LIST_FOR_EACH_SAFE (facet, next_facet, list_node, &rule->facets) {
5267 facet_revalidate(facet);
5270 complete_operation(rule);
5274 rule_get_stats(struct rule *rule_, uint64_t *packets, uint64_t *bytes)
5276 struct rule_dpif *rule = rule_dpif_cast(rule_);
5278 /* push_all_stats() can handle flow misses which, when using the learn
5279 * action, can cause rules to be added and deleted. This can corrupt our
5280 * caller's datastructures which assume that rule_get_stats() doesn't have
5281 * an impact on the flow table. To be safe, we disable miss handling. */
5282 push_all_stats__(false);
5284 /* Start from historical data for 'rule' itself that are no longer tracked
5285 * in facets. This counts, for example, facets that have expired. */
5286 *packets = rule->packet_count;
5287 *bytes = rule->byte_count;
5291 rule_dpif_execute(struct rule_dpif *rule, const struct flow *flow,
5292 struct ofpbuf *packet)
5294 struct ofproto_dpif *ofproto = ofproto_dpif_cast(rule->up.ofproto);
5295 struct dpif_flow_stats stats;
5296 struct xlate_out xout;
5297 struct xlate_in xin;
5299 dpif_flow_stats_extract(flow, packet, time_msec(), &stats);
5300 rule_credit_stats(rule, &stats);
5302 xlate_in_init(&xin, ofproto, flow, rule, stats.tcp_flags, packet);
5303 xin.resubmit_stats = &stats;
5304 xlate_actions(&xin, &xout);
5306 execute_odp_actions(ofproto, flow, xout.odp_actions.data,
5307 xout.odp_actions.size, packet);
5309 xlate_out_uninit(&xout);
5313 rule_execute(struct rule *rule, const struct flow *flow,
5314 struct ofpbuf *packet)
5316 rule_dpif_execute(rule_dpif_cast(rule), flow, packet);
5317 ofpbuf_delete(packet);
5322 rule_modify_actions(struct rule *rule_)
5324 struct rule_dpif *rule = rule_dpif_cast(rule_);
5326 complete_operation(rule);
5329 /* Sends 'packet' out 'ofport'.
5330 * May modify 'packet'.
5331 * Returns 0 if successful, otherwise a positive errno value. */
5333 send_packet(const struct ofport_dpif *ofport, struct ofpbuf *packet)
5335 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport->up.ofproto);
5336 uint64_t odp_actions_stub[1024 / 8];
5337 struct ofpbuf key, odp_actions;
5338 struct dpif_flow_stats stats;
5339 struct odputil_keybuf keybuf;
5340 struct ofpact_output output;
5341 struct xlate_out xout;
5342 struct xlate_in xin;
5344 union flow_in_port in_port_;
5347 ofpbuf_use_stub(&odp_actions, odp_actions_stub, sizeof odp_actions_stub);
5348 ofpbuf_use_stack(&key, &keybuf, sizeof keybuf);
5350 /* Use OFPP_NONE as the in_port to avoid special packet processing. */
5351 in_port_.ofp_port = OFPP_NONE;
5352 flow_extract(packet, 0, 0, NULL, &in_port_, &flow);
5353 odp_flow_key_from_flow(&key, &flow, ofp_port_to_odp_port(ofproto,
5355 dpif_flow_stats_extract(&flow, packet, time_msec(), &stats);
5357 ofpact_init(&output.ofpact, OFPACT_OUTPUT, sizeof output);
5358 output.port = ofport->up.ofp_port;
5361 xlate_in_init(&xin, ofproto, &flow, NULL, 0, packet);
5362 xin.ofpacts_len = sizeof output;
5363 xin.ofpacts = &output.ofpact;
5364 xin.resubmit_stats = &stats;
5365 xlate_actions(&xin, &xout);
5367 error = dpif_execute(ofproto->backer->dpif,
5369 xout.odp_actions.data, xout.odp_actions.size,
5371 xlate_out_uninit(&xout);
5374 VLOG_WARN_RL(&rl, "%s: failed to send packet on port %s (%s)",
5375 ofproto->up.name, netdev_get_name(ofport->up.netdev),
5379 ofproto->stats.tx_packets++;
5380 ofproto->stats.tx_bytes += packet->size;
5384 /* Composes an ODP action for a "slow path" action for 'flow' within 'ofproto'.
5385 * The action will state 'slow' as the reason that the action is in the slow
5386 * path. (This is purely informational: it allows a human viewing "ovs-dpctl
5387 * dump-flows" output to see why a flow is in the slow path.)
5389 * The 'stub_size' bytes in 'stub' will be used to store the action.
5390 * 'stub_size' must be large enough for the action.
5392 * The action and its size will be stored in '*actionsp' and '*actions_lenp',
5395 compose_slow_path(const struct ofproto_dpif *ofproto, const struct flow *flow,
5396 enum slow_path_reason slow,
5397 uint64_t *stub, size_t stub_size,
5398 const struct nlattr **actionsp, size_t *actions_lenp)
5400 union user_action_cookie cookie;
5403 cookie.type = USER_ACTION_COOKIE_SLOW_PATH;
5404 cookie.slow_path.unused = 0;
5405 cookie.slow_path.reason = slow;
5407 ofpbuf_use_stack(&buf, stub, stub_size);
5408 if (slow & (SLOW_CFM | SLOW_BFD | SLOW_LACP | SLOW_STP)) {
5409 uint32_t pid = dpif_port_get_pid(ofproto->backer->dpif,
5411 odp_put_userspace_action(pid, &cookie, sizeof cookie.slow_path, &buf);
5413 put_userspace_action(ofproto, &buf, flow, &cookie,
5414 sizeof cookie.slow_path);
5416 *actionsp = buf.data;
5417 *actions_lenp = buf.size;
5421 put_userspace_action(const struct ofproto_dpif *ofproto,
5422 struct ofpbuf *odp_actions,
5423 const struct flow *flow,
5424 const union user_action_cookie *cookie,
5425 const size_t cookie_size)
5429 pid = dpif_port_get_pid(ofproto->backer->dpif,
5430 ofp_port_to_odp_port(ofproto,
5431 flow->in_port.ofp_port));
5433 return odp_put_userspace_action(pid, cookie, cookie_size, odp_actions);
5438 update_mirror_stats(struct ofproto_dpif *ofproto, mirror_mask_t mirrors,
5439 uint64_t packets, uint64_t bytes)
5445 for (; mirrors; mirrors = zero_rightmost_1bit(mirrors)) {
5448 m = ofproto->mirrors[mirror_mask_ffs(mirrors) - 1];
5451 /* In normal circumstances 'm' will not be NULL. However,
5452 * if mirrors are reconfigured, we can temporarily get out
5453 * of sync in facet_revalidate(). We could "correct" the
5454 * mirror list before reaching here, but doing that would
5455 * not properly account the traffic stats we've currently
5456 * accumulated for previous mirror configuration. */
5460 m->packet_count += packets;
5461 m->byte_count += bytes;
5466 /* Optimized flow revalidation.
5468 * It's a difficult problem, in general, to tell which facets need to have
5469 * their actions recalculated whenever the OpenFlow flow table changes. We
5470 * don't try to solve that general problem: for most kinds of OpenFlow flow
5471 * table changes, we recalculate the actions for every facet. This is
5472 * relatively expensive, but it's good enough if the OpenFlow flow table
5473 * doesn't change very often.
5475 * However, we can expect one particular kind of OpenFlow flow table change to
5476 * happen frequently: changes caused by MAC learning. To avoid wasting a lot
5477 * of CPU on revalidating every facet whenever MAC learning modifies the flow
5478 * table, we add a special case that applies to flow tables in which every rule
5479 * has the same form (that is, the same wildcards), except that the table is
5480 * also allowed to have a single "catch-all" flow that matches all packets. We
5481 * optimize this case by tagging all of the facets that resubmit into the table
5482 * and invalidating the same tag whenever a flow changes in that table. The
5483 * end result is that we revalidate just the facets that need it (and sometimes
5484 * a few more, but not all of the facets or even all of the facets that
5485 * resubmit to the table modified by MAC learning). */
5487 /* Calculates the tag to use for 'flow' and mask 'mask' when it is inserted
5488 * into an OpenFlow table with the given 'basis'. */
5490 rule_calculate_tag(const struct flow *flow, const struct minimask *mask,
5493 if (minimask_is_catchall(mask)) {
5496 uint32_t hash = flow_hash_in_minimask(flow, mask, secret);
5497 return tag_create_deterministic(hash);
5501 /* Following a change to OpenFlow table 'table_id' in 'ofproto', update the
5502 * taggability of that table.
5504 * This function must be called after *each* change to a flow table. If you
5505 * skip calling it on some changes then the pointer comparisons at the end can
5506 * be invalid if you get unlucky. For example, if a flow removal causes a
5507 * cls_table to be destroyed and then a flow insertion causes a cls_table with
5508 * different wildcards to be created with the same address, then this function
5509 * will incorrectly skip revalidation. */
5511 table_update_taggable(struct ofproto_dpif *ofproto, uint8_t table_id)
5513 struct table_dpif *table = &ofproto->tables[table_id];
5514 const struct oftable *oftable = &ofproto->up.tables[table_id];
5515 struct cls_table *catchall, *other;
5516 struct cls_table *t;
5518 catchall = other = NULL;
5520 switch (hmap_count(&oftable->cls.tables)) {
5522 /* We could tag this OpenFlow table but it would make the logic a
5523 * little harder and it's a corner case that doesn't seem worth it
5529 HMAP_FOR_EACH (t, hmap_node, &oftable->cls.tables) {
5530 if (cls_table_is_catchall(t)) {
5532 } else if (!other) {
5535 /* Indicate that we can't tag this by setting both tables to
5536 * NULL. (We know that 'catchall' is already NULL.) */
5543 /* Can't tag this table. */
5547 if (table->catchall_table != catchall || table->other_table != other) {
5548 table->catchall_table = catchall;
5549 table->other_table = other;
5550 ofproto->backer->need_revalidate = REV_FLOW_TABLE;
5554 /* Given 'rule' that has changed in some way (either it is a rule being
5555 * inserted, a rule being deleted, or a rule whose actions are being
5556 * modified), marks facets for revalidation to ensure that packets will be
5557 * forwarded correctly according to the new state of the flow table.
5559 * This function must be called after *each* change to a flow table. See
5560 * the comment on table_update_taggable() for more information. */
5562 rule_invalidate(const struct rule_dpif *rule)
5564 struct ofproto_dpif *ofproto = ofproto_dpif_cast(rule->up.ofproto);
5566 table_update_taggable(ofproto, rule->up.table_id);
5568 if (!ofproto->backer->need_revalidate) {
5569 struct table_dpif *table = &ofproto->tables[rule->up.table_id];
5571 if (table->other_table && rule->tag) {
5572 tag_set_add(&ofproto->backer->revalidate_set, rule->tag);
5574 ofproto->backer->need_revalidate = REV_FLOW_TABLE;
5580 set_frag_handling(struct ofproto *ofproto_,
5581 enum ofp_config_flags frag_handling)
5583 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
5584 if (frag_handling != OFPC_FRAG_REASM) {
5585 ofproto->backer->need_revalidate = REV_RECONFIGURE;
5593 packet_out(struct ofproto *ofproto_, struct ofpbuf *packet,
5594 const struct flow *flow,
5595 const struct ofpact *ofpacts, size_t ofpacts_len)
5597 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
5598 struct odputil_keybuf keybuf;
5599 struct dpif_flow_stats stats;
5600 struct xlate_out xout;
5601 struct xlate_in xin;
5605 ofpbuf_use_stack(&key, &keybuf, sizeof keybuf);
5606 odp_flow_key_from_flow(&key, flow,
5607 ofp_port_to_odp_port(ofproto,
5608 flow->in_port.ofp_port));
5610 dpif_flow_stats_extract(flow, packet, time_msec(), &stats);
5612 xlate_in_init(&xin, ofproto, flow, NULL, stats.tcp_flags, packet);
5613 xin.resubmit_stats = &stats;
5614 xin.ofpacts_len = ofpacts_len;
5615 xin.ofpacts = ofpacts;
5617 xlate_actions(&xin, &xout);
5618 dpif_execute(ofproto->backer->dpif, key.data, key.size,
5619 xout.odp_actions.data, xout.odp_actions.size, packet);
5620 xlate_out_uninit(&xout);
5628 set_netflow(struct ofproto *ofproto_,
5629 const struct netflow_options *netflow_options)
5631 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
5633 if (netflow_options) {
5634 if (!ofproto->netflow) {
5635 ofproto->netflow = netflow_create();
5636 ofproto->backer->need_revalidate = REV_RECONFIGURE;
5638 return netflow_set_options(ofproto->netflow, netflow_options);
5639 } else if (ofproto->netflow) {
5640 ofproto->backer->need_revalidate = REV_RECONFIGURE;
5641 netflow_destroy(ofproto->netflow);
5642 ofproto->netflow = NULL;
5649 get_netflow_ids(const struct ofproto *ofproto_,
5650 uint8_t *engine_type, uint8_t *engine_id)
5652 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
5654 dpif_get_netflow_ids(ofproto->backer->dpif, engine_type, engine_id);
5658 send_active_timeout(struct ofproto_dpif *ofproto, struct facet *facet)
5660 if (!facet_is_controller_flow(facet) &&
5661 netflow_active_timeout_expired(ofproto->netflow, &facet->nf_flow)) {
5662 struct subfacet *subfacet;
5663 struct ofexpired expired;
5665 LIST_FOR_EACH (subfacet, list_node, &facet->subfacets) {
5666 if (subfacet->path == SF_FAST_PATH) {
5667 struct dpif_flow_stats stats;
5669 subfacet_install(subfacet, &facet->xout.odp_actions,
5671 subfacet_update_stats(subfacet, &stats);
5675 expired.flow = facet->flow;
5676 expired.packet_count = facet->packet_count;
5677 expired.byte_count = facet->byte_count;
5678 expired.used = facet->used;
5679 netflow_expire(ofproto->netflow, &facet->nf_flow, &expired);
5684 send_netflow_active_timeouts(struct ofproto_dpif *ofproto)
5686 struct cls_cursor cursor;
5687 struct facet *facet;
5689 cls_cursor_init(&cursor, &ofproto->facets, NULL);
5690 CLS_CURSOR_FOR_EACH (facet, cr, &cursor) {
5691 send_active_timeout(ofproto, facet);
5695 static struct ofproto_dpif *
5696 ofproto_dpif_lookup(const char *name)
5698 struct ofproto_dpif *ofproto;
5700 HMAP_FOR_EACH_WITH_HASH (ofproto, all_ofproto_dpifs_node,
5701 hash_string(name, 0), &all_ofproto_dpifs) {
5702 if (!strcmp(ofproto->up.name, name)) {
5710 ofproto_unixctl_fdb_flush(struct unixctl_conn *conn, int argc,
5711 const char *argv[], void *aux OVS_UNUSED)
5713 struct ofproto_dpif *ofproto;
5716 ofproto = ofproto_dpif_lookup(argv[1]);
5718 unixctl_command_reply_error(conn, "no such bridge");
5721 mac_learning_flush(ofproto->ml, &ofproto->backer->revalidate_set);
5723 HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
5724 mac_learning_flush(ofproto->ml, &ofproto->backer->revalidate_set);
5728 unixctl_command_reply(conn, "table successfully flushed");
5732 ofproto_unixctl_fdb_show(struct unixctl_conn *conn, int argc OVS_UNUSED,
5733 const char *argv[], void *aux OVS_UNUSED)
5735 struct ds ds = DS_EMPTY_INITIALIZER;
5736 const struct ofproto_dpif *ofproto;
5737 const struct mac_entry *e;
5739 ofproto = ofproto_dpif_lookup(argv[1]);
5741 unixctl_command_reply_error(conn, "no such bridge");
5745 ds_put_cstr(&ds, " port VLAN MAC Age\n");
5746 LIST_FOR_EACH (e, lru_node, &ofproto->ml->lrus) {
5747 struct ofbundle *bundle = e->port.p;
5748 ds_put_format(&ds, "%5d %4d "ETH_ADDR_FMT" %3d\n",
5749 ofbundle_get_a_port(bundle)->odp_port,
5750 e->vlan, ETH_ADDR_ARGS(e->mac),
5751 mac_entry_age(ofproto->ml, e));
5753 unixctl_command_reply(conn, ds_cstr(&ds));
5758 struct xlate_out xout;
5759 struct xlate_in xin;
5765 trace_format_rule(struct ds *result, int level, const struct rule_dpif *rule)
5767 ds_put_char_multiple(result, '\t', level);
5769 ds_put_cstr(result, "No match\n");
5773 ds_put_format(result, "Rule: table=%"PRIu8" cookie=%#"PRIx64" ",
5774 rule ? rule->up.table_id : 0, ntohll(rule->up.flow_cookie));
5775 cls_rule_format(&rule->up.cr, result);
5776 ds_put_char(result, '\n');
5778 ds_put_char_multiple(result, '\t', level);
5779 ds_put_cstr(result, "OpenFlow ");
5780 ofpacts_format(rule->up.ofpacts, rule->up.ofpacts_len, result);
5781 ds_put_char(result, '\n');
5785 trace_format_flow(struct ds *result, int level, const char *title,
5786 struct trace_ctx *trace)
5788 ds_put_char_multiple(result, '\t', level);
5789 ds_put_format(result, "%s: ", title);
5790 if (flow_equal(&trace->xin.flow, &trace->flow)) {
5791 ds_put_cstr(result, "unchanged");
5793 flow_format(result, &trace->xin.flow);
5794 trace->flow = trace->xin.flow;
5796 ds_put_char(result, '\n');
5800 trace_format_regs(struct ds *result, int level, const char *title,
5801 struct trace_ctx *trace)
5805 ds_put_char_multiple(result, '\t', level);
5806 ds_put_format(result, "%s:", title);
5807 for (i = 0; i < FLOW_N_REGS; i++) {
5808 ds_put_format(result, " reg%zu=0x%"PRIx32, i, trace->flow.regs[i]);
5810 ds_put_char(result, '\n');
5814 trace_format_odp(struct ds *result, int level, const char *title,
5815 struct trace_ctx *trace)
5817 struct ofpbuf *odp_actions = &trace->xout.odp_actions;
5819 ds_put_char_multiple(result, '\t', level);
5820 ds_put_format(result, "%s: ", title);
5821 format_odp_actions(result, odp_actions->data, odp_actions->size);
5822 ds_put_char(result, '\n');
5826 trace_resubmit(struct xlate_in *xin, struct rule_dpif *rule, int recurse)
5828 struct trace_ctx *trace = CONTAINER_OF(xin, struct trace_ctx, xin);
5829 struct ds *result = trace->result;
5831 ds_put_char(result, '\n');
5832 trace_format_flow(result, recurse + 1, "Resubmitted flow", trace);
5833 trace_format_regs(result, recurse + 1, "Resubmitted regs", trace);
5834 trace_format_odp(result, recurse + 1, "Resubmitted odp", trace);
5835 trace_format_rule(result, recurse + 1, rule);
5839 trace_report(struct xlate_in *xin, const char *s, int recurse)
5841 struct trace_ctx *trace = CONTAINER_OF(xin, struct trace_ctx, xin);
5842 struct ds *result = trace->result;
5844 ds_put_char_multiple(result, '\t', recurse);
5845 ds_put_cstr(result, s);
5846 ds_put_char(result, '\n');
5850 ofproto_unixctl_trace(struct unixctl_conn *conn, int argc, const char *argv[],
5851 void *aux OVS_UNUSED)
5853 const struct dpif_backer *backer;
5854 struct ofproto_dpif *ofproto;
5855 struct ofpbuf odp_key;
5856 struct ofpbuf *packet;
5864 ofpbuf_init(&odp_key, 0);
5866 /* Handle "-generate" or a hex string as the last argument. */
5867 if (!strcmp(argv[argc - 1], "-generate")) {
5868 packet = ofpbuf_new(0);
5871 const char *error = eth_from_hex(argv[argc - 1], &packet);
5874 } else if (argc == 4) {
5875 /* The 3-argument form must end in "-generate' or a hex string. */
5876 unixctl_command_reply_error(conn, error);
5881 /* Parse the flow and determine whether a datapath or
5882 * bridge is specified. If function odp_flow_key_from_string()
5883 * returns 0, the flow is a odp_flow. If function
5884 * parse_ofp_exact_flow() returns 0, the flow is a br_flow. */
5885 if (!odp_flow_from_string(argv[argc - 1], NULL, &odp_key, NULL)) {
5886 /* If the odp_flow is the second argument,
5887 * the datapath name is the first argument. */
5889 const char *dp_type;
5890 if (!strncmp(argv[1], "ovs-", 4)) {
5891 dp_type = argv[1] + 4;
5895 backer = shash_find_data(&all_dpif_backers, dp_type);
5897 unixctl_command_reply_error(conn, "Cannot find datapath "
5902 /* No datapath name specified, so there should be only one
5904 struct shash_node *node;
5905 if (shash_count(&all_dpif_backers) != 1) {
5906 unixctl_command_reply_error(conn, "Must specify datapath "
5907 "name, there is more than one type of datapath");
5910 node = shash_first(&all_dpif_backers);
5911 backer = node->data;
5914 /* Extract the ofproto_dpif object from the ofproto_receive()
5916 if (ofproto_receive(backer, NULL, odp_key.data,
5917 odp_key.size, &flow, NULL, &ofproto, NULL)) {
5918 unixctl_command_reply_error(conn, "Invalid datapath flow");
5921 ds_put_format(&result, "Bridge: %s\n", ofproto->up.name);
5922 } else if (!parse_ofp_exact_flow(&flow, argv[argc - 1])) {
5924 unixctl_command_reply_error(conn, "Must specify bridge name");
5928 ofproto = ofproto_dpif_lookup(argv[1]);
5930 unixctl_command_reply_error(conn, "Unknown bridge name");
5934 unixctl_command_reply_error(conn, "Bad flow syntax");
5938 /* Generate a packet, if requested. */
5940 if (!packet->size) {
5941 flow_compose(packet, &flow);
5943 union flow_in_port in_port_;
5945 in_port_ = flow.in_port;
5946 ds_put_cstr(&result, "Packet: ");
5947 s = ofp_packet_to_string(packet->data, packet->size);
5948 ds_put_cstr(&result, s);
5951 /* Use the metadata from the flow and the packet argument
5952 * to reconstruct the flow. */
5953 flow_extract(packet, flow.skb_priority, flow.skb_mark, NULL,
5958 ofproto_trace(ofproto, &flow, packet, &result);
5959 unixctl_command_reply(conn, ds_cstr(&result));
5962 ds_destroy(&result);
5963 ofpbuf_delete(packet);
5964 ofpbuf_uninit(&odp_key);
5968 ofproto_trace(struct ofproto_dpif *ofproto, const struct flow *flow,
5969 const struct ofpbuf *packet, struct ds *ds)
5971 struct rule_dpif *rule;
5973 ds_put_cstr(ds, "Flow: ");
5974 flow_format(ds, flow);
5975 ds_put_char(ds, '\n');
5977 rule = rule_dpif_lookup(ofproto, flow, NULL);
5979 trace_format_rule(ds, 0, rule);
5980 if (rule == ofproto->miss_rule) {
5981 ds_put_cstr(ds, "\nNo match, flow generates \"packet in\"s.\n");
5982 } else if (rule == ofproto->no_packet_in_rule) {
5983 ds_put_cstr(ds, "\nNo match, packets dropped because "
5984 "OFPPC_NO_PACKET_IN is set on in_port.\n");
5985 } else if (rule == ofproto->drop_frags_rule) {
5986 ds_put_cstr(ds, "\nPackets dropped because they are IP fragments "
5987 "and the fragment handling mode is \"drop\".\n");
5991 uint64_t odp_actions_stub[1024 / 8];
5992 struct ofpbuf odp_actions;
5993 struct trace_ctx trace;
5997 tcp_flags = packet ? packet_get_tcp_flags(packet, flow) : 0;
6000 ofpbuf_use_stub(&odp_actions,
6001 odp_actions_stub, sizeof odp_actions_stub);
6002 xlate_in_init(&trace.xin, ofproto, flow, rule, tcp_flags, packet);
6003 trace.xin.resubmit_hook = trace_resubmit;
6004 trace.xin.report_hook = trace_report;
6006 xlate_actions(&trace.xin, &trace.xout);
6008 ds_put_char(ds, '\n');
6009 trace_format_flow(ds, 0, "Final flow", &trace);
6011 match_init(&match, flow, &trace.xout.wc);
6012 ds_put_cstr(ds, "Relevant fields: ");
6013 match_format(&match, ds, OFP_DEFAULT_PRIORITY);
6014 ds_put_char(ds, '\n');
6016 ds_put_cstr(ds, "Datapath actions: ");
6017 format_odp_actions(ds, trace.xout.odp_actions.data,
6018 trace.xout.odp_actions.size);
6020 if (trace.xout.slow) {
6021 ds_put_cstr(ds, "\nThis flow is handled by the userspace "
6022 "slow path because it:");
6023 switch (trace.xout.slow) {
6025 ds_put_cstr(ds, "\n\t- Consists of CFM packets.");
6028 ds_put_cstr(ds, "\n\t- Consists of LACP packets.");
6031 ds_put_cstr(ds, "\n\t- Consists of STP packets.");
6034 ds_put_cstr(ds, "\n\t- Consists of BFD packets.");
6036 case SLOW_CONTROLLER:
6037 ds_put_cstr(ds, "\n\t- Sends \"packet-in\" messages "
6038 "to the OpenFlow controller.");
6045 xlate_out_uninit(&trace.xout);
6050 ofproto_dpif_clog(struct unixctl_conn *conn OVS_UNUSED, int argc OVS_UNUSED,
6051 const char *argv[] OVS_UNUSED, void *aux OVS_UNUSED)
6054 unixctl_command_reply(conn, NULL);
6058 ofproto_dpif_unclog(struct unixctl_conn *conn OVS_UNUSED, int argc OVS_UNUSED,
6059 const char *argv[] OVS_UNUSED, void *aux OVS_UNUSED)
6062 unixctl_command_reply(conn, NULL);
6065 /* Runs a self-check of flow translations in 'ofproto'. Appends a message to
6066 * 'reply' describing the results. */
6068 ofproto_dpif_self_check__(struct ofproto_dpif *ofproto, struct ds *reply)
6070 struct cls_cursor cursor;
6071 struct facet *facet;
6075 cls_cursor_init(&cursor, &ofproto->facets, NULL);
6076 CLS_CURSOR_FOR_EACH (facet, cr, &cursor) {
6077 if (!facet_check_consistency(facet)) {
6082 ofproto->backer->need_revalidate = REV_INCONSISTENCY;
6086 ds_put_format(reply, "%s: self-check failed (%d errors)\n",
6087 ofproto->up.name, errors);
6089 ds_put_format(reply, "%s: self-check passed\n", ofproto->up.name);
6094 ofproto_dpif_self_check(struct unixctl_conn *conn,
6095 int argc, const char *argv[], void *aux OVS_UNUSED)
6097 struct ds reply = DS_EMPTY_INITIALIZER;
6098 struct ofproto_dpif *ofproto;
6101 ofproto = ofproto_dpif_lookup(argv[1]);
6103 unixctl_command_reply_error(conn, "Unknown ofproto (use "
6104 "ofproto/list for help)");
6107 ofproto_dpif_self_check__(ofproto, &reply);
6109 HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
6110 ofproto_dpif_self_check__(ofproto, &reply);
6114 unixctl_command_reply(conn, ds_cstr(&reply));
6118 /* Store the current ofprotos in 'ofproto_shash'. Returns a sorted list
6119 * of the 'ofproto_shash' nodes. It is the responsibility of the caller
6120 * to destroy 'ofproto_shash' and free the returned value. */
6121 static const struct shash_node **
6122 get_ofprotos(struct shash *ofproto_shash)
6124 const struct ofproto_dpif *ofproto;
6126 HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
6127 char *name = xasprintf("%s@%s", ofproto->up.type, ofproto->up.name);
6128 shash_add_nocopy(ofproto_shash, name, ofproto);
6131 return shash_sort(ofproto_shash);
6135 ofproto_unixctl_dpif_dump_dps(struct unixctl_conn *conn, int argc OVS_UNUSED,
6136 const char *argv[] OVS_UNUSED,
6137 void *aux OVS_UNUSED)
6139 struct ds ds = DS_EMPTY_INITIALIZER;
6140 struct shash ofproto_shash;
6141 const struct shash_node **sorted_ofprotos;
6144 shash_init(&ofproto_shash);
6145 sorted_ofprotos = get_ofprotos(&ofproto_shash);
6146 for (i = 0; i < shash_count(&ofproto_shash); i++) {
6147 const struct shash_node *node = sorted_ofprotos[i];
6148 ds_put_format(&ds, "%s\n", node->name);
6151 shash_destroy(&ofproto_shash);
6152 free(sorted_ofprotos);
6154 unixctl_command_reply(conn, ds_cstr(&ds));
6159 show_dp_rates(struct ds *ds, const char *heading,
6160 const struct avg_subfacet_rates *rates)
6162 ds_put_format(ds, "%s add rate: %5.3f/min, del rate: %5.3f/min\n",
6163 heading, rates->add_rate, rates->del_rate);
6167 dpif_show_backer(const struct dpif_backer *backer, struct ds *ds)
6169 const struct shash_node **ofprotos;
6170 struct ofproto_dpif *ofproto;
6171 struct shash ofproto_shash;
6172 uint64_t n_hit, n_missed;
6173 long long int minutes;
6176 n_hit = n_missed = 0;
6177 HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
6178 if (ofproto->backer == backer) {
6179 n_missed += ofproto->n_missed;
6180 n_hit += ofproto->n_hit;
6184 ds_put_format(ds, "%s: hit:%"PRIu64" missed:%"PRIu64"\n",
6185 dpif_name(backer->dpif), n_hit, n_missed);
6186 ds_put_format(ds, "\tflows: cur: %zu, avg: %u, max: %u,"
6187 " life span: %lldms\n", hmap_count(&backer->subfacets),
6188 backer->avg_n_subfacet, backer->max_n_subfacet,
6189 backer->avg_subfacet_life);
6191 minutes = (time_msec() - backer->created) / (1000 * 60);
6192 if (minutes >= 60) {
6193 show_dp_rates(ds, "\thourly avg:", &backer->hourly);
6195 if (minutes >= 60 * 24) {
6196 show_dp_rates(ds, "\tdaily avg:", &backer->daily);
6198 show_dp_rates(ds, "\toverall avg:", &backer->lifetime);
6200 shash_init(&ofproto_shash);
6201 ofprotos = get_ofprotos(&ofproto_shash);
6202 for (i = 0; i < shash_count(&ofproto_shash); i++) {
6203 struct ofproto_dpif *ofproto = ofprotos[i]->data;
6204 const struct shash_node **ports;
6207 if (ofproto->backer != backer) {
6211 ds_put_format(ds, "\t%s: hit:%"PRIu64" missed:%"PRIu64"\n",
6212 ofproto->up.name, ofproto->n_hit, ofproto->n_missed);
6214 ports = shash_sort(&ofproto->up.port_by_name);
6215 for (j = 0; j < shash_count(&ofproto->up.port_by_name); j++) {
6216 const struct shash_node *node = ports[j];
6217 struct ofport *ofport = node->data;
6219 odp_port_t odp_port;
6221 ds_put_format(ds, "\t\t%s %u/", netdev_get_name(ofport->netdev),
6224 odp_port = ofp_port_to_odp_port(ofproto, ofport->ofp_port);
6225 if (odp_port != ODPP_NONE) {
6226 ds_put_format(ds, "%"PRIu32":", odp_port);
6228 ds_put_cstr(ds, "none:");
6231 ds_put_format(ds, " (%s", netdev_get_type(ofport->netdev));
6234 if (!netdev_get_config(ofport->netdev, &config)) {
6235 const struct smap_node **nodes;
6238 nodes = smap_sort(&config);
6239 for (i = 0; i < smap_count(&config); i++) {
6240 const struct smap_node *node = nodes[i];
6241 ds_put_format(ds, "%c %s=%s", i ? ',' : ':',
6242 node->key, node->value);
6246 smap_destroy(&config);
6248 ds_put_char(ds, ')');
6249 ds_put_char(ds, '\n');
6253 shash_destroy(&ofproto_shash);
6258 ofproto_unixctl_dpif_show(struct unixctl_conn *conn, int argc OVS_UNUSED,
6259 const char *argv[] OVS_UNUSED, void *aux OVS_UNUSED)
6261 struct ds ds = DS_EMPTY_INITIALIZER;
6262 const struct shash_node **backers;
6265 backers = shash_sort(&all_dpif_backers);
6266 for (i = 0; i < shash_count(&all_dpif_backers); i++) {
6267 dpif_show_backer(backers[i]->data, &ds);
6271 unixctl_command_reply(conn, ds_cstr(&ds));
6275 /* Dump the megaflow (facet) cache. This is useful to check the
6276 * correctness of flow wildcarding, since the same mechanism is used for
6277 * both xlate caching and kernel wildcarding.
6279 * It's important to note that in the output the flow description uses
6280 * OpenFlow (OFP) ports, but the actions use datapath (ODP) ports.
6282 * This command is only needed for advanced debugging, so it's not
6283 * documented in the man page. */
6285 ofproto_unixctl_dpif_dump_megaflows(struct unixctl_conn *conn,
6286 int argc OVS_UNUSED, const char *argv[],
6287 void *aux OVS_UNUSED)
6289 struct ds ds = DS_EMPTY_INITIALIZER;
6290 const struct ofproto_dpif *ofproto;
6291 long long int now = time_msec();
6292 struct cls_cursor cursor;
6293 struct facet *facet;
6295 ofproto = ofproto_dpif_lookup(argv[1]);
6297 unixctl_command_reply_error(conn, "no such bridge");
6301 cls_cursor_init(&cursor, &ofproto->facets, NULL);
6302 CLS_CURSOR_FOR_EACH (facet, cr, &cursor) {
6303 cls_rule_format(&facet->cr, &ds);
6304 ds_put_cstr(&ds, ", ");
6305 ds_put_format(&ds, "n_subfacets:%zu, ", list_size(&facet->subfacets));
6306 ds_put_format(&ds, "used:%.3fs, ", (now - facet->used) / 1000.0);
6307 ds_put_cstr(&ds, "Datapath actions: ");
6308 if (facet->xout.slow) {
6309 uint64_t slow_path_stub[128 / 8];
6310 const struct nlattr *actions;
6313 compose_slow_path(ofproto, &facet->flow, facet->xout.slow,
6314 slow_path_stub, sizeof slow_path_stub,
6315 &actions, &actions_len);
6316 format_odp_actions(&ds, actions, actions_len);
6318 format_odp_actions(&ds, facet->xout.odp_actions.data,
6319 facet->xout.odp_actions.size);
6321 ds_put_cstr(&ds, "\n");
6324 ds_chomp(&ds, '\n');
6325 unixctl_command_reply(conn, ds_cstr(&ds));
6330 ofproto_unixctl_dpif_dump_flows(struct unixctl_conn *conn,
6331 int argc OVS_UNUSED, const char *argv[],
6332 void *aux OVS_UNUSED)
6334 struct ds ds = DS_EMPTY_INITIALIZER;
6335 const struct ofproto_dpif *ofproto;
6336 struct subfacet *subfacet;
6338 ofproto = ofproto_dpif_lookup(argv[1]);
6340 unixctl_command_reply_error(conn, "no such bridge");
6344 update_stats(ofproto->backer);
6346 HMAP_FOR_EACH (subfacet, hmap_node, &ofproto->backer->subfacets) {
6347 struct facet *facet = subfacet->facet;
6349 if (ofproto_dpif_cast(facet->rule->up.ofproto) != ofproto) {
6353 odp_flow_key_format(subfacet->key, subfacet->key_len, &ds);
6355 ds_put_format(&ds, ", packets:%"PRIu64", bytes:%"PRIu64", used:",
6356 subfacet->dp_packet_count, subfacet->dp_byte_count);
6357 if (subfacet->used) {
6358 ds_put_format(&ds, "%.3fs",
6359 (time_msec() - subfacet->used) / 1000.0);
6361 ds_put_format(&ds, "never");
6363 if (subfacet->facet->tcp_flags) {
6364 ds_put_cstr(&ds, ", flags:");
6365 packet_format_tcp_flags(&ds, subfacet->facet->tcp_flags);
6368 ds_put_cstr(&ds, ", actions:");
6369 if (facet->xout.slow) {
6370 uint64_t slow_path_stub[128 / 8];
6371 const struct nlattr *actions;
6374 compose_slow_path(ofproto, &facet->flow, facet->xout.slow,
6375 slow_path_stub, sizeof slow_path_stub,
6376 &actions, &actions_len);
6377 format_odp_actions(&ds, actions, actions_len);
6379 format_odp_actions(&ds, facet->xout.odp_actions.data,
6380 facet->xout.odp_actions.size);
6382 ds_put_char(&ds, '\n');
6385 unixctl_command_reply(conn, ds_cstr(&ds));
6390 ofproto_unixctl_dpif_del_flows(struct unixctl_conn *conn,
6391 int argc OVS_UNUSED, const char *argv[],
6392 void *aux OVS_UNUSED)
6394 struct ds ds = DS_EMPTY_INITIALIZER;
6395 struct ofproto_dpif *ofproto;
6397 ofproto = ofproto_dpif_lookup(argv[1]);
6399 unixctl_command_reply_error(conn, "no such bridge");
6403 flush(&ofproto->up);
6405 unixctl_command_reply(conn, ds_cstr(&ds));
6410 ofproto_dpif_unixctl_init(void)
6412 static bool registered;
6418 unixctl_command_register(
6420 "[dp_name]|bridge odp_flow|br_flow [-generate|packet]",
6421 1, 3, ofproto_unixctl_trace, NULL);
6422 unixctl_command_register("fdb/flush", "[bridge]", 0, 1,
6423 ofproto_unixctl_fdb_flush, NULL);
6424 unixctl_command_register("fdb/show", "bridge", 1, 1,
6425 ofproto_unixctl_fdb_show, NULL);
6426 unixctl_command_register("ofproto/clog", "", 0, 0,
6427 ofproto_dpif_clog, NULL);
6428 unixctl_command_register("ofproto/unclog", "", 0, 0,
6429 ofproto_dpif_unclog, NULL);
6430 unixctl_command_register("ofproto/self-check", "[bridge]", 0, 1,
6431 ofproto_dpif_self_check, NULL);
6432 unixctl_command_register("dpif/dump-dps", "", 0, 0,
6433 ofproto_unixctl_dpif_dump_dps, NULL);
6434 unixctl_command_register("dpif/show", "", 0, 0, ofproto_unixctl_dpif_show,
6436 unixctl_command_register("dpif/dump-flows", "bridge", 1, 1,
6437 ofproto_unixctl_dpif_dump_flows, NULL);
6438 unixctl_command_register("dpif/del-flows", "bridge", 1, 1,
6439 ofproto_unixctl_dpif_del_flows, NULL);
6440 unixctl_command_register("dpif/dump-megaflows", "bridge", 1, 1,
6441 ofproto_unixctl_dpif_dump_megaflows, NULL);
6444 /* Linux VLAN device support (e.g. "eth0.10" for VLAN 10.)
6446 * This is deprecated. It is only for compatibility with broken device drivers
6447 * in old versions of Linux that do not properly support VLANs when VLAN
6448 * devices are not used. When broken device drivers are no longer in
6449 * widespread use, we will delete these interfaces. */
6452 set_realdev(struct ofport *ofport_, ofp_port_t realdev_ofp_port, int vid)
6454 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport_->ofproto);
6455 struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
6457 if (realdev_ofp_port == ofport->realdev_ofp_port
6458 && vid == ofport->vlandev_vid) {
6462 ofproto->backer->need_revalidate = REV_RECONFIGURE;
6464 if (ofport->realdev_ofp_port) {
6467 if (realdev_ofp_port && ofport->bundle) {
6468 /* vlandevs are enslaved to their realdevs, so they are not allowed to
6469 * themselves be part of a bundle. */
6470 bundle_set(ofport->up.ofproto, ofport->bundle, NULL);
6473 ofport->realdev_ofp_port = realdev_ofp_port;
6474 ofport->vlandev_vid = vid;
6476 if (realdev_ofp_port) {
6477 vsp_add(ofport, realdev_ofp_port, vid);
6484 hash_realdev_vid(ofp_port_t realdev_ofp_port, int vid)
6486 return hash_2words(ofp_to_u16(realdev_ofp_port), vid);
6489 /* Returns the OFP port number of the Linux VLAN device that corresponds to
6490 * 'vlan_tci' on the network device with port number 'realdev_ofp_port' in
6491 * 'struct ofport_dpif'. For example, given 'realdev_ofp_port' of eth0 and
6492 * 'vlan_tci' 9, it would return the port number of eth0.9.
6494 * Unless VLAN splinters are enabled for port 'realdev_ofp_port', this
6495 * function just returns its 'realdev_ofp_port' argument. */
6497 vsp_realdev_to_vlandev(const struct ofproto_dpif *ofproto,
6498 ofp_port_t realdev_ofp_port, ovs_be16 vlan_tci)
6500 if (!hmap_is_empty(&ofproto->realdev_vid_map)) {
6501 int vid = vlan_tci_to_vid(vlan_tci);
6502 const struct vlan_splinter *vsp;
6504 HMAP_FOR_EACH_WITH_HASH (vsp, realdev_vid_node,
6505 hash_realdev_vid(realdev_ofp_port, vid),
6506 &ofproto->realdev_vid_map) {
6507 if (vsp->realdev_ofp_port == realdev_ofp_port
6508 && vsp->vid == vid) {
6509 return vsp->vlandev_ofp_port;
6513 return realdev_ofp_port;
6516 static struct vlan_splinter *
6517 vlandev_find(const struct ofproto_dpif *ofproto, ofp_port_t vlandev_ofp_port)
6519 struct vlan_splinter *vsp;
6521 HMAP_FOR_EACH_WITH_HASH (vsp, vlandev_node,
6522 hash_int(ofp_to_u16(vlandev_ofp_port), 0),
6523 &ofproto->vlandev_map) {
6524 if (vsp->vlandev_ofp_port == vlandev_ofp_port) {
6532 /* Returns the OpenFlow port number of the "real" device underlying the Linux
6533 * VLAN device with OpenFlow port number 'vlandev_ofp_port' and stores the
6534 * VLAN VID of the Linux VLAN device in '*vid'. For example, given
6535 * 'vlandev_ofp_port' of eth0.9, it would return the OpenFlow port number of
6536 * eth0 and store 9 in '*vid'.
6538 * Returns 0 and does not modify '*vid' if 'vlandev_ofp_port' is not a Linux
6539 * VLAN device. Unless VLAN splinters are enabled, this is what this function
6542 vsp_vlandev_to_realdev(const struct ofproto_dpif *ofproto,
6543 ofp_port_t vlandev_ofp_port, int *vid)
6545 if (!hmap_is_empty(&ofproto->vlandev_map)) {
6546 const struct vlan_splinter *vsp;
6548 vsp = vlandev_find(ofproto, vlandev_ofp_port);
6553 return vsp->realdev_ofp_port;
6559 /* Given 'flow', a flow representing a packet received on 'ofproto', checks
6560 * whether 'flow->in_port' represents a Linux VLAN device. If so, changes
6561 * 'flow->in_port' to the "real" device backing the VLAN device, sets
6562 * 'flow->vlan_tci' to the VLAN VID, and returns true. Otherwise (which is
6563 * always the case unless VLAN splinters are enabled), returns false without
6564 * making any changes. */
6566 vsp_adjust_flow(const struct ofproto_dpif *ofproto, struct flow *flow)
6571 realdev = vsp_vlandev_to_realdev(ofproto, flow->in_port.ofp_port, &vid);
6576 /* Cause the flow to be processed as if it came in on the real device with
6577 * the VLAN device's VLAN ID. */
6578 flow->in_port.ofp_port = realdev;
6579 flow->vlan_tci = htons((vid & VLAN_VID_MASK) | VLAN_CFI);
6584 vsp_remove(struct ofport_dpif *port)
6586 struct ofproto_dpif *ofproto = ofproto_dpif_cast(port->up.ofproto);
6587 struct vlan_splinter *vsp;
6589 vsp = vlandev_find(ofproto, port->up.ofp_port);
6591 hmap_remove(&ofproto->vlandev_map, &vsp->vlandev_node);
6592 hmap_remove(&ofproto->realdev_vid_map, &vsp->realdev_vid_node);
6595 port->realdev_ofp_port = 0;
6597 VLOG_ERR("missing vlan device record");
6602 vsp_add(struct ofport_dpif *port, ofp_port_t realdev_ofp_port, int vid)
6604 struct ofproto_dpif *ofproto = ofproto_dpif_cast(port->up.ofproto);
6606 if (!vsp_vlandev_to_realdev(ofproto, port->up.ofp_port, NULL)
6607 && (vsp_realdev_to_vlandev(ofproto, realdev_ofp_port, htons(vid))
6608 == realdev_ofp_port)) {
6609 struct vlan_splinter *vsp;
6611 vsp = xmalloc(sizeof *vsp);
6612 hmap_insert(&ofproto->vlandev_map, &vsp->vlandev_node,
6613 hash_int(ofp_to_u16(port->up.ofp_port), 0));
6614 hmap_insert(&ofproto->realdev_vid_map, &vsp->realdev_vid_node,
6615 hash_realdev_vid(realdev_ofp_port, vid));
6616 vsp->realdev_ofp_port = realdev_ofp_port;
6617 vsp->vlandev_ofp_port = port->up.ofp_port;
6620 port->realdev_ofp_port = realdev_ofp_port;
6622 VLOG_ERR("duplicate vlan device record");
6627 ofp_port_to_odp_port(const struct ofproto_dpif *ofproto, ofp_port_t ofp_port)
6629 const struct ofport_dpif *ofport = get_ofp_port(ofproto, ofp_port);
6630 return ofport ? ofport->odp_port : ODPP_NONE;
6633 static struct ofport_dpif *
6634 odp_port_to_ofport(const struct dpif_backer *backer, odp_port_t odp_port)
6636 struct ofport_dpif *port;
6638 HMAP_FOR_EACH_IN_BUCKET (port, odp_port_node,
6639 hash_int(odp_to_u32(odp_port), 0),
6640 &backer->odp_to_ofport_map) {
6641 if (port->odp_port == odp_port) {
6650 odp_port_to_ofp_port(const struct ofproto_dpif *ofproto, odp_port_t odp_port)
6652 struct ofport_dpif *port;
6654 port = odp_port_to_ofport(ofproto->backer, odp_port);
6655 if (port && &ofproto->up == port->up.ofproto) {
6656 return port->up.ofp_port;
6662 /* Compute exponentially weighted moving average, adding 'new' as the newest,
6663 * most heavily weighted element. 'base' designates the rate of decay: after
6664 * 'base' further updates, 'new''s weight in the EWMA decays to about 1/e
6667 exp_mavg(double *avg, int base, double new)
6669 *avg = (*avg * (base - 1) + new) / base;
6673 update_moving_averages(struct dpif_backer *backer)
6675 const int min_ms = 60 * 1000; /* milliseconds in one minute. */
6676 long long int minutes = (time_msec() - backer->created) / min_ms;
6679 backer->lifetime.add_rate = (double) backer->total_subfacet_add_count
6681 backer->lifetime.del_rate = (double) backer->total_subfacet_del_count
6684 backer->lifetime.add_rate = 0.0;
6685 backer->lifetime.del_rate = 0.0;
6688 /* Update hourly averages on the minute boundaries. */
6689 if (time_msec() - backer->last_minute >= min_ms) {
6690 exp_mavg(&backer->hourly.add_rate, 60, backer->subfacet_add_count);
6691 exp_mavg(&backer->hourly.del_rate, 60, backer->subfacet_del_count);
6693 /* Update daily averages on the hour boundaries. */
6694 if ((backer->last_minute - backer->created) / min_ms % 60 == 59) {
6695 exp_mavg(&backer->daily.add_rate, 24, backer->hourly.add_rate);
6696 exp_mavg(&backer->daily.del_rate, 24, backer->hourly.del_rate);
6699 backer->total_subfacet_add_count += backer->subfacet_add_count;
6700 backer->total_subfacet_del_count += backer->subfacet_del_count;
6701 backer->subfacet_add_count = 0;
6702 backer->subfacet_del_count = 0;
6703 backer->last_minute += min_ms;
6707 const struct ofproto_class ofproto_dpif_class = {
6742 port_is_lacp_current,
6743 NULL, /* rule_choose_table */
6750 rule_modify_actions,
6764 get_stp_port_status,
6771 is_mirror_output_bundle,
6772 forward_bpdu_changed,
6773 set_mac_table_config,