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 uint16_t realdev_ofp_port;
281 uint16_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 *, uint16_t realdev_ofp_port, int vid);
289 static uint16_t odp_port_to_ofp_port(const struct ofproto_dpif *,
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 *, uint32_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 uint32_t odp_port = UINT32_MAX;
589 if (!dpif_port_add(backer->dpif, iter->up.netdev,
591 simap_put(&backer->tnl_backers, dp_port, odp_port);
592 node = simap_find(&backer->tnl_backers, dp_port);
597 iter->odp_port = node ? node->data : OVSP_NONE;
598 if (tnl_port_reconfigure(&iter->up, iter->odp_port,
600 backer->need_revalidate = REV_RECONFIGURE;
605 SIMAP_FOR_EACH (node, &tmp_backers) {
606 dpif_port_del(backer->dpif, node->data);
608 simap_destroy(&tmp_backers);
610 switch (backer->need_revalidate) {
611 case REV_RECONFIGURE: COVERAGE_INC(rev_reconfigure); break;
612 case REV_STP: COVERAGE_INC(rev_stp); break;
613 case REV_PORT_TOGGLED: COVERAGE_INC(rev_port_toggled); break;
614 case REV_FLOW_TABLE: COVERAGE_INC(rev_flow_table); break;
615 case REV_INCONSISTENCY: COVERAGE_INC(rev_inconsistency); break;
618 if (backer->need_revalidate) {
619 /* Clear the drop_keys in case we should now be accepting some
620 * formerly dropped flows. */
621 drop_key_clear(backer);
624 /* Clear the revalidation flags. */
625 tag_set_init(&backer->revalidate_set);
626 backer->need_revalidate = 0;
628 HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
629 struct facet *facet, *next;
630 struct cls_cursor cursor;
632 if (ofproto->backer != backer) {
636 cls_cursor_init(&cursor, &ofproto->facets, NULL);
637 CLS_CURSOR_FOR_EACH_SAFE (facet, next, cr, &cursor) {
639 || tag_set_intersects(&revalidate_set, facet->xout.tags)) {
640 facet_revalidate(facet);
647 if (!backer->recv_set_enable) {
648 /* Wake up before a max of 1000ms. */
649 timer_set_duration(&backer->next_expiration, 1000);
650 } else if (timer_expired(&backer->next_expiration)) {
651 int delay = expire(backer);
652 timer_set_duration(&backer->next_expiration, delay);
655 /* Check for port changes in the dpif. */
656 while ((error = dpif_port_poll(backer->dpif, &devname)) == 0) {
657 struct ofproto_dpif *ofproto;
658 struct dpif_port port;
660 /* Don't report on the datapath's device. */
661 if (!strcmp(devname, dpif_base_name(backer->dpif))) {
665 HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node,
666 &all_ofproto_dpifs) {
667 if (simap_contains(&ofproto->backer->tnl_backers, devname)) {
672 ofproto = lookup_ofproto_dpif_by_port_name(devname);
673 if (dpif_port_query_by_name(backer->dpif, devname, &port)) {
674 /* The port was removed. If we know the datapath,
675 * report it through poll_set(). If we don't, it may be
676 * notifying us of a removal we initiated, so ignore it.
677 * If there's a pending ENOBUFS, let it stand, since
678 * everything will be reevaluated. */
679 if (ofproto && ofproto->port_poll_errno != ENOBUFS) {
680 sset_add(&ofproto->port_poll_set, devname);
681 ofproto->port_poll_errno = 0;
683 } else if (!ofproto) {
684 /* The port was added, but we don't know with which
685 * ofproto we should associate it. Delete it. */
686 dpif_port_del(backer->dpif, port.port_no);
688 dpif_port_destroy(&port);
694 if (error != EAGAIN) {
695 struct ofproto_dpif *ofproto;
697 /* There was some sort of error, so propagate it to all
698 * ofprotos that use this backer. */
699 HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node,
700 &all_ofproto_dpifs) {
701 if (ofproto->backer == backer) {
702 sset_clear(&ofproto->port_poll_set);
703 ofproto->port_poll_errno = error;
708 if (backer->governor) {
711 governor_run(backer->governor);
713 /* If the governor has shrunk to its minimum size and the number of
714 * subfacets has dwindled, then drop the governor entirely.
716 * For hysteresis, the number of subfacets to drop the governor is
717 * smaller than the number needed to trigger its creation. */
718 n_subfacets = hmap_count(&backer->subfacets);
719 if (n_subfacets * 4 < flow_eviction_threshold
720 && governor_is_idle(backer->governor)) {
721 governor_destroy(backer->governor);
722 backer->governor = NULL;
730 dpif_backer_run_fast(struct dpif_backer *backer, int max_batch)
734 /* If recv_set_enable is false, we should not handle upcalls. */
735 if (!backer->recv_set_enable) {
739 /* Handle one or more batches of upcalls, until there's nothing left to do
740 * or until we do a fixed total amount of work.
742 * We do work in batches because it can be much cheaper to set up a number
743 * of flows and fire off their patches all at once. We do multiple batches
744 * because in some cases handling a packet can cause another packet to be
745 * queued almost immediately as part of the return flow. Both
746 * optimizations can make major improvements on some benchmarks and
747 * presumably for real traffic as well. */
749 while (work < max_batch) {
750 int retval = handle_upcalls(backer, max_batch - work);
761 type_run_fast(const char *type)
763 struct dpif_backer *backer;
765 backer = shash_find_data(&all_dpif_backers, type);
767 /* This is not necessarily a problem, since backers are only
768 * created on demand. */
772 return dpif_backer_run_fast(backer, FLOW_MISS_MAX_BATCH);
778 static long long int port_rl = LLONG_MIN;
779 static unsigned int backer_rl = 0;
781 if (time_msec() >= port_rl) {
782 struct ofproto_dpif *ofproto;
783 struct ofport_dpif *ofport;
785 HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
787 HMAP_FOR_EACH (ofport, up.hmap_node, &ofproto->up.ports) {
788 port_run_fast(ofport);
791 port_rl = time_msec() + 200;
794 /* XXX: We have to be careful not to do too much work in this function. If
795 * we call dpif_backer_run_fast() too often, or with too large a batch,
796 * performance improves signifcantly, but at a cost. It's possible for the
797 * number of flows in the datapath to increase without bound, and for poll
798 * loops to take 10s of seconds. The correct solution to this problem,
799 * long term, is to separate flow miss handling into it's own thread so it
800 * isn't affected by revalidations, and expirations. Until then, this is
801 * the best we can do. */
802 if (++backer_rl >= 10) {
803 struct shash_node *node;
806 SHASH_FOR_EACH (node, &all_dpif_backers) {
807 dpif_backer_run_fast(node->data, 1);
813 type_wait(const char *type)
815 struct dpif_backer *backer;
817 backer = shash_find_data(&all_dpif_backers, type);
819 /* This is not necessarily a problem, since backers are only
820 * created on demand. */
824 if (backer->governor) {
825 governor_wait(backer->governor);
828 timer_wait(&backer->next_expiration);
831 /* Basic life-cycle. */
833 static int add_internal_flows(struct ofproto_dpif *);
835 static struct ofproto *
838 struct ofproto_dpif *ofproto = xmalloc(sizeof *ofproto);
843 dealloc(struct ofproto *ofproto_)
845 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
850 close_dpif_backer(struct dpif_backer *backer)
852 struct shash_node *node;
854 ovs_assert(backer->refcount > 0);
856 if (--backer->refcount) {
860 drop_key_clear(backer);
861 hmap_destroy(&backer->drop_keys);
863 simap_destroy(&backer->tnl_backers);
864 hmap_destroy(&backer->odp_to_ofport_map);
865 node = shash_find(&all_dpif_backers, backer->type);
867 shash_delete(&all_dpif_backers, node);
868 dpif_close(backer->dpif);
870 ovs_assert(hmap_is_empty(&backer->subfacets));
871 hmap_destroy(&backer->subfacets);
872 governor_destroy(backer->governor);
877 /* Datapath port slated for removal from datapath. */
879 struct list list_node;
884 open_dpif_backer(const char *type, struct dpif_backer **backerp)
886 struct dpif_backer *backer;
887 struct dpif_port_dump port_dump;
888 struct dpif_port port;
889 struct shash_node *node;
890 struct list garbage_list;
891 struct odp_garbage *garbage, *next;
897 backer = shash_find_data(&all_dpif_backers, type);
904 backer_name = xasprintf("ovs-%s", type);
906 /* Remove any existing datapaths, since we assume we're the only
907 * userspace controlling the datapath. */
909 dp_enumerate_names(type, &names);
910 SSET_FOR_EACH(name, &names) {
911 struct dpif *old_dpif;
913 /* Don't remove our backer if it exists. */
914 if (!strcmp(name, backer_name)) {
918 if (dpif_open(name, type, &old_dpif)) {
919 VLOG_WARN("couldn't open old datapath %s to remove it", name);
921 dpif_delete(old_dpif);
922 dpif_close(old_dpif);
925 sset_destroy(&names);
927 backer = xmalloc(sizeof *backer);
929 error = dpif_create_and_open(backer_name, type, &backer->dpif);
932 VLOG_ERR("failed to open datapath of type %s: %s", type,
938 backer->type = xstrdup(type);
939 backer->governor = NULL;
940 backer->refcount = 1;
941 hmap_init(&backer->odp_to_ofport_map);
942 hmap_init(&backer->drop_keys);
943 hmap_init(&backer->subfacets);
944 timer_set_duration(&backer->next_expiration, 1000);
945 backer->need_revalidate = 0;
946 simap_init(&backer->tnl_backers);
947 tag_set_init(&backer->revalidate_set);
948 backer->recv_set_enable = !ofproto_get_flow_restore_wait();
951 if (backer->recv_set_enable) {
952 dpif_flow_flush(backer->dpif);
955 /* Loop through the ports already on the datapath and remove any
956 * that we don't need anymore. */
957 list_init(&garbage_list);
958 dpif_port_dump_start(&port_dump, backer->dpif);
959 while (dpif_port_dump_next(&port_dump, &port)) {
960 node = shash_find(&init_ofp_ports, port.name);
961 if (!node && strcmp(port.name, dpif_base_name(backer->dpif))) {
962 garbage = xmalloc(sizeof *garbage);
963 garbage->odp_port = port.port_no;
964 list_push_front(&garbage_list, &garbage->list_node);
967 dpif_port_dump_done(&port_dump);
969 LIST_FOR_EACH_SAFE (garbage, next, list_node, &garbage_list) {
970 dpif_port_del(backer->dpif, garbage->odp_port);
971 list_remove(&garbage->list_node);
975 shash_add(&all_dpif_backers, type, backer);
977 error = dpif_recv_set(backer->dpif, backer->recv_set_enable);
979 VLOG_ERR("failed to listen on datapath of type %s: %s",
980 type, strerror(error));
981 close_dpif_backer(backer);
985 backer->max_n_subfacet = 0;
986 backer->created = time_msec();
987 backer->last_minute = backer->created;
988 memset(&backer->hourly, 0, sizeof backer->hourly);
989 memset(&backer->daily, 0, sizeof backer->daily);
990 memset(&backer->lifetime, 0, sizeof backer->lifetime);
991 backer->subfacet_add_count = 0;
992 backer->subfacet_del_count = 0;
993 backer->total_subfacet_add_count = 0;
994 backer->total_subfacet_del_count = 0;
995 backer->avg_n_subfacet = 0;
996 backer->avg_subfacet_life = 0;
1002 construct(struct ofproto *ofproto_)
1004 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1005 struct shash_node *node, *next;
1010 error = open_dpif_backer(ofproto->up.type, &ofproto->backer);
1015 max_ports = dpif_get_max_ports(ofproto->backer->dpif);
1016 ofproto_init_max_ports(ofproto_, MIN(max_ports, OFPP_MAX));
1018 ofproto->netflow = NULL;
1019 ofproto->sflow = NULL;
1020 ofproto->ipfix = NULL;
1021 ofproto->stp = NULL;
1022 hmap_init(&ofproto->bundles);
1023 ofproto->ml = mac_learning_create(MAC_ENTRY_DEFAULT_IDLE_TIME);
1024 for (i = 0; i < MAX_MIRRORS; i++) {
1025 ofproto->mirrors[i] = NULL;
1027 ofproto->has_bonded_bundles = false;
1029 classifier_init(&ofproto->facets);
1030 ofproto->consistency_rl = LLONG_MIN;
1032 for (i = 0; i < N_TABLES; i++) {
1033 struct table_dpif *table = &ofproto->tables[i];
1035 table->catchall_table = NULL;
1036 table->other_table = NULL;
1037 table->basis = random_uint32();
1040 list_init(&ofproto->completions);
1042 ofproto_dpif_unixctl_init();
1044 ofproto->has_mirrors = false;
1045 ofproto->has_bundle_action = false;
1047 hmap_init(&ofproto->vlandev_map);
1048 hmap_init(&ofproto->realdev_vid_map);
1050 sset_init(&ofproto->ports);
1051 sset_init(&ofproto->ghost_ports);
1052 sset_init(&ofproto->port_poll_set);
1053 ofproto->port_poll_errno = 0;
1055 SHASH_FOR_EACH_SAFE (node, next, &init_ofp_ports) {
1056 struct iface_hint *iface_hint = node->data;
1058 if (!strcmp(iface_hint->br_name, ofproto->up.name)) {
1059 /* Check if the datapath already has this port. */
1060 if (dpif_port_exists(ofproto->backer->dpif, node->name)) {
1061 sset_add(&ofproto->ports, node->name);
1064 free(iface_hint->br_name);
1065 free(iface_hint->br_type);
1067 shash_delete(&init_ofp_ports, node);
1071 hmap_insert(&all_ofproto_dpifs, &ofproto->all_ofproto_dpifs_node,
1072 hash_string(ofproto->up.name, 0));
1073 memset(&ofproto->stats, 0, sizeof ofproto->stats);
1075 ofproto_init_tables(ofproto_, N_TABLES);
1076 error = add_internal_flows(ofproto);
1077 ofproto->up.tables[TBL_INTERNAL].flags = OFTABLE_HIDDEN | OFTABLE_READONLY;
1080 ofproto->n_missed = 0;
1086 add_internal_flow(struct ofproto_dpif *ofproto, int id,
1087 const struct ofpbuf *ofpacts, struct rule_dpif **rulep)
1089 struct ofputil_flow_mod fm;
1092 match_init_catchall(&fm.match);
1094 match_set_reg(&fm.match, 0, id);
1095 fm.new_cookie = htonll(0);
1096 fm.cookie = htonll(0);
1097 fm.cookie_mask = htonll(0);
1098 fm.table_id = TBL_INTERNAL;
1099 fm.command = OFPFC_ADD;
1100 fm.idle_timeout = 0;
1101 fm.hard_timeout = 0;
1105 fm.ofpacts = ofpacts->data;
1106 fm.ofpacts_len = ofpacts->size;
1108 error = ofproto_flow_mod(&ofproto->up, &fm);
1110 VLOG_ERR_RL(&rl, "failed to add internal flow %d (%s)",
1111 id, ofperr_to_string(error));
1115 *rulep = rule_dpif_lookup_in_table(ofproto, &fm.match.flow, NULL,
1117 ovs_assert(*rulep != NULL);
1123 add_internal_flows(struct ofproto_dpif *ofproto)
1125 struct ofpact_controller *controller;
1126 uint64_t ofpacts_stub[128 / 8];
1127 struct ofpbuf ofpacts;
1131 ofpbuf_use_stack(&ofpacts, ofpacts_stub, sizeof ofpacts_stub);
1134 controller = ofpact_put_CONTROLLER(&ofpacts);
1135 controller->max_len = UINT16_MAX;
1136 controller->controller_id = 0;
1137 controller->reason = OFPR_NO_MATCH;
1138 ofpact_pad(&ofpacts);
1140 error = add_internal_flow(ofproto, id++, &ofpacts, &ofproto->miss_rule);
1145 ofpbuf_clear(&ofpacts);
1146 error = add_internal_flow(ofproto, id++, &ofpacts,
1147 &ofproto->no_packet_in_rule);
1152 error = add_internal_flow(ofproto, id++, &ofpacts,
1153 &ofproto->drop_frags_rule);
1158 complete_operations(struct ofproto_dpif *ofproto)
1160 struct dpif_completion *c, *next;
1162 LIST_FOR_EACH_SAFE (c, next, list_node, &ofproto->completions) {
1163 ofoperation_complete(c->op, 0);
1164 list_remove(&c->list_node);
1170 destruct(struct ofproto *ofproto_)
1172 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1173 struct rule_dpif *rule, *next_rule;
1174 struct oftable *table;
1177 ofproto->backer->need_revalidate = REV_RECONFIGURE;
1178 hmap_remove(&all_ofproto_dpifs, &ofproto->all_ofproto_dpifs_node);
1179 complete_operations(ofproto);
1181 OFPROTO_FOR_EACH_TABLE (table, &ofproto->up) {
1182 struct cls_cursor cursor;
1184 cls_cursor_init(&cursor, &table->cls, NULL);
1185 CLS_CURSOR_FOR_EACH_SAFE (rule, next_rule, up.cr, &cursor) {
1186 ofproto_rule_destroy(&rule->up);
1190 for (i = 0; i < MAX_MIRRORS; i++) {
1191 mirror_destroy(ofproto->mirrors[i]);
1194 netflow_destroy(ofproto->netflow);
1195 dpif_sflow_destroy(ofproto->sflow);
1196 hmap_destroy(&ofproto->bundles);
1197 mac_learning_destroy(ofproto->ml);
1199 classifier_destroy(&ofproto->facets);
1201 hmap_destroy(&ofproto->vlandev_map);
1202 hmap_destroy(&ofproto->realdev_vid_map);
1204 sset_destroy(&ofproto->ports);
1205 sset_destroy(&ofproto->ghost_ports);
1206 sset_destroy(&ofproto->port_poll_set);
1208 close_dpif_backer(ofproto->backer);
1212 run_fast(struct ofproto *ofproto_)
1214 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1215 struct ofport_dpif *ofport;
1217 /* Do not perform any periodic activity required by 'ofproto' while
1218 * waiting for flow restore to complete. */
1219 if (ofproto_get_flow_restore_wait()) {
1223 HMAP_FOR_EACH (ofport, up.hmap_node, &ofproto->up.ports) {
1224 port_run_fast(ofport);
1231 run(struct ofproto *ofproto_)
1233 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1234 struct ofport_dpif *ofport;
1235 struct ofbundle *bundle;
1239 complete_operations(ofproto);
1242 /* Do not perform any periodic activity below required by 'ofproto' while
1243 * waiting for flow restore to complete. */
1244 if (ofproto_get_flow_restore_wait()) {
1248 error = run_fast(ofproto_);
1253 if (ofproto->netflow) {
1254 if (netflow_run(ofproto->netflow)) {
1255 send_netflow_active_timeouts(ofproto);
1258 if (ofproto->sflow) {
1259 dpif_sflow_run(ofproto->sflow);
1262 HMAP_FOR_EACH (ofport, up.hmap_node, &ofproto->up.ports) {
1265 HMAP_FOR_EACH (bundle, hmap_node, &ofproto->bundles) {
1270 mac_learning_run(ofproto->ml, &ofproto->backer->revalidate_set);
1272 /* Check the consistency of a random facet, to aid debugging. */
1273 if (time_msec() >= ofproto->consistency_rl
1274 && !classifier_is_empty(&ofproto->facets)
1275 && !ofproto->backer->need_revalidate) {
1276 struct cls_table *table;
1277 struct cls_rule *cr;
1278 struct facet *facet;
1280 ofproto->consistency_rl = time_msec() + 250;
1282 table = CONTAINER_OF(hmap_random_node(&ofproto->facets.tables),
1283 struct cls_table, hmap_node);
1284 cr = CONTAINER_OF(hmap_random_node(&table->rules), struct cls_rule,
1286 facet = CONTAINER_OF(cr, struct facet, cr);
1288 if (!tag_set_intersects(&ofproto->backer->revalidate_set,
1289 facet->xout.tags)) {
1290 if (!facet_check_consistency(facet)) {
1291 ofproto->backer->need_revalidate = REV_INCONSISTENCY;
1300 wait(struct ofproto *ofproto_)
1302 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1303 struct ofport_dpif *ofport;
1304 struct ofbundle *bundle;
1306 if (!clogged && !list_is_empty(&ofproto->completions)) {
1307 poll_immediate_wake();
1310 if (ofproto_get_flow_restore_wait()) {
1314 dpif_wait(ofproto->backer->dpif);
1315 dpif_recv_wait(ofproto->backer->dpif);
1316 if (ofproto->sflow) {
1317 dpif_sflow_wait(ofproto->sflow);
1319 if (!tag_set_is_empty(&ofproto->backer->revalidate_set)) {
1320 poll_immediate_wake();
1322 HMAP_FOR_EACH (ofport, up.hmap_node, &ofproto->up.ports) {
1325 HMAP_FOR_EACH (bundle, hmap_node, &ofproto->bundles) {
1326 bundle_wait(bundle);
1328 if (ofproto->netflow) {
1329 netflow_wait(ofproto->netflow);
1331 mac_learning_wait(ofproto->ml);
1333 if (ofproto->backer->need_revalidate) {
1334 /* Shouldn't happen, but if it does just go around again. */
1335 VLOG_DBG_RL(&rl, "need revalidate in ofproto_wait_cb()");
1336 poll_immediate_wake();
1341 get_memory_usage(const struct ofproto *ofproto_, struct simap *usage)
1343 const struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1344 struct cls_cursor cursor;
1345 size_t n_subfacets = 0;
1346 struct facet *facet;
1348 simap_increase(usage, "facets", classifier_count(&ofproto->facets));
1350 cls_cursor_init(&cursor, &ofproto->facets, NULL);
1351 CLS_CURSOR_FOR_EACH (facet, cr, &cursor) {
1352 n_subfacets += list_size(&facet->subfacets);
1354 simap_increase(usage, "subfacets", n_subfacets);
1358 flush(struct ofproto *ofproto_)
1360 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1361 struct subfacet *subfacet, *next_subfacet;
1362 struct subfacet *batch[SUBFACET_DESTROY_MAX_BATCH];
1366 HMAP_FOR_EACH_SAFE (subfacet, next_subfacet, hmap_node,
1367 &ofproto->backer->subfacets) {
1368 if (ofproto_dpif_cast(subfacet->facet->rule->up.ofproto) != ofproto) {
1372 if (subfacet->path != SF_NOT_INSTALLED) {
1373 batch[n_batch++] = subfacet;
1374 if (n_batch >= SUBFACET_DESTROY_MAX_BATCH) {
1375 subfacet_destroy_batch(ofproto->backer, batch, n_batch);
1379 subfacet_destroy(subfacet);
1384 subfacet_destroy_batch(ofproto->backer, batch, n_batch);
1389 get_features(struct ofproto *ofproto_ OVS_UNUSED,
1390 bool *arp_match_ip, enum ofputil_action_bitmap *actions)
1392 *arp_match_ip = true;
1393 *actions = (OFPUTIL_A_OUTPUT |
1394 OFPUTIL_A_SET_VLAN_VID |
1395 OFPUTIL_A_SET_VLAN_PCP |
1396 OFPUTIL_A_STRIP_VLAN |
1397 OFPUTIL_A_SET_DL_SRC |
1398 OFPUTIL_A_SET_DL_DST |
1399 OFPUTIL_A_SET_NW_SRC |
1400 OFPUTIL_A_SET_NW_DST |
1401 OFPUTIL_A_SET_NW_TOS |
1402 OFPUTIL_A_SET_TP_SRC |
1403 OFPUTIL_A_SET_TP_DST |
1408 get_tables(struct ofproto *ofproto_, struct ofp12_table_stats *ots)
1410 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1411 struct dpif_dp_stats s;
1412 uint64_t n_miss, n_no_pkt_in, n_bytes, n_dropped_frags;
1415 strcpy(ots->name, "classifier");
1417 dpif_get_dp_stats(ofproto->backer->dpif, &s);
1418 rule_get_stats(&ofproto->miss_rule->up, &n_miss, &n_bytes);
1419 rule_get_stats(&ofproto->no_packet_in_rule->up, &n_no_pkt_in, &n_bytes);
1420 rule_get_stats(&ofproto->drop_frags_rule->up, &n_dropped_frags, &n_bytes);
1422 n_lookup = s.n_hit + s.n_missed - n_dropped_frags;
1423 ots->lookup_count = htonll(n_lookup);
1424 ots->matched_count = htonll(n_lookup - n_miss - n_no_pkt_in);
1427 static struct ofport *
1430 struct ofport_dpif *port = xmalloc(sizeof *port);
1435 port_dealloc(struct ofport *port_)
1437 struct ofport_dpif *port = ofport_dpif_cast(port_);
1442 port_construct(struct ofport *port_)
1444 struct ofport_dpif *port = ofport_dpif_cast(port_);
1445 struct ofproto_dpif *ofproto = ofproto_dpif_cast(port->up.ofproto);
1446 const struct netdev *netdev = port->up.netdev;
1447 char namebuf[NETDEV_VPORT_NAME_BUFSIZE];
1448 struct dpif_port dpif_port;
1451 ofproto->backer->need_revalidate = REV_RECONFIGURE;
1452 port->bundle = NULL;
1455 port->tag = tag_create_random();
1456 port->may_enable = true;
1457 port->stp_port = NULL;
1458 port->stp_state = STP_DISABLED;
1459 port->tnl_port = NULL;
1461 hmap_init(&port->priorities);
1462 port->realdev_ofp_port = 0;
1463 port->vlandev_vid = 0;
1464 port->carrier_seq = netdev_get_carrier_resets(netdev);
1466 if (netdev_vport_is_patch(netdev)) {
1467 /* By bailing out here, we don't submit the port to the sFlow module
1468 * to be considered for counter polling export. This is correct
1469 * because the patch port represents an interface that sFlow considers
1470 * to be "internal" to the switch as a whole, and therefore not an
1471 * candidate for counter polling. */
1472 port->odp_port = OVSP_NONE;
1473 ofport_update_peer(port);
1477 error = dpif_port_query_by_name(ofproto->backer->dpif,
1478 netdev_vport_get_dpif_port(netdev, namebuf,
1485 port->odp_port = dpif_port.port_no;
1487 if (netdev_get_tunnel_config(netdev)) {
1488 port->tnl_port = tnl_port_add(&port->up, port->odp_port);
1490 /* Sanity-check that a mapping doesn't already exist. This
1491 * shouldn't happen for non-tunnel ports. */
1492 if (odp_port_to_ofp_port(ofproto, port->odp_port) != OFPP_NONE) {
1493 VLOG_ERR("port %s already has an OpenFlow port number",
1495 dpif_port_destroy(&dpif_port);
1499 hmap_insert(&ofproto->backer->odp_to_ofport_map, &port->odp_port_node,
1500 hash_int(port->odp_port, 0));
1502 dpif_port_destroy(&dpif_port);
1504 if (ofproto->sflow) {
1505 dpif_sflow_add_port(ofproto->sflow, port_, port->odp_port);
1512 port_destruct(struct ofport *port_)
1514 struct ofport_dpif *port = ofport_dpif_cast(port_);
1515 struct ofproto_dpif *ofproto = ofproto_dpif_cast(port->up.ofproto);
1516 const char *devname = netdev_get_name(port->up.netdev);
1517 char namebuf[NETDEV_VPORT_NAME_BUFSIZE];
1518 const char *dp_port_name;
1520 dp_port_name = netdev_vport_get_dpif_port(port->up.netdev, namebuf,
1522 if (dpif_port_exists(ofproto->backer->dpif, dp_port_name)) {
1523 /* The underlying device is still there, so delete it. This
1524 * happens when the ofproto is being destroyed, since the caller
1525 * assumes that removal of attached ports will happen as part of
1527 if (!port->tnl_port) {
1528 dpif_port_del(ofproto->backer->dpif, port->odp_port);
1530 ofproto->backer->need_revalidate = REV_RECONFIGURE;
1534 port->peer->peer = NULL;
1538 if (port->odp_port != OVSP_NONE && !port->tnl_port) {
1539 hmap_remove(&ofproto->backer->odp_to_ofport_map, &port->odp_port_node);
1542 tnl_port_del(port->tnl_port);
1543 sset_find_and_delete(&ofproto->ports, devname);
1544 sset_find_and_delete(&ofproto->ghost_ports, devname);
1545 ofproto->backer->need_revalidate = REV_RECONFIGURE;
1546 bundle_remove(port_);
1547 set_cfm(port_, NULL);
1548 set_bfd(port_, NULL);
1549 if (ofproto->sflow) {
1550 dpif_sflow_del_port(ofproto->sflow, port->odp_port);
1553 ofport_clear_priorities(port);
1554 hmap_destroy(&port->priorities);
1558 port_modified(struct ofport *port_)
1560 struct ofport_dpif *port = ofport_dpif_cast(port_);
1562 if (port->bundle && port->bundle->bond) {
1563 bond_slave_set_netdev(port->bundle->bond, port, port->up.netdev);
1567 cfm_set_netdev(port->cfm, port->up.netdev);
1570 if (port->tnl_port && tnl_port_reconfigure(&port->up, port->odp_port,
1572 ofproto_dpif_cast(port->up.ofproto)->backer->need_revalidate = true;
1575 ofport_update_peer(port);
1579 port_reconfigured(struct ofport *port_, enum ofputil_port_config old_config)
1581 struct ofport_dpif *port = ofport_dpif_cast(port_);
1582 struct ofproto_dpif *ofproto = ofproto_dpif_cast(port->up.ofproto);
1583 enum ofputil_port_config changed = old_config ^ port->up.pp.config;
1585 if (changed & (OFPUTIL_PC_NO_RECV | OFPUTIL_PC_NO_RECV_STP |
1586 OFPUTIL_PC_NO_FWD | OFPUTIL_PC_NO_FLOOD |
1587 OFPUTIL_PC_NO_PACKET_IN)) {
1588 ofproto->backer->need_revalidate = REV_RECONFIGURE;
1590 if (changed & OFPUTIL_PC_NO_FLOOD && port->bundle) {
1591 bundle_update(port->bundle);
1597 set_sflow(struct ofproto *ofproto_,
1598 const struct ofproto_sflow_options *sflow_options)
1600 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1601 struct dpif_sflow *ds = ofproto->sflow;
1603 if (sflow_options) {
1605 struct ofport_dpif *ofport;
1607 ds = ofproto->sflow = dpif_sflow_create();
1608 HMAP_FOR_EACH (ofport, up.hmap_node, &ofproto->up.ports) {
1609 dpif_sflow_add_port(ds, &ofport->up, ofport->odp_port);
1611 ofproto->backer->need_revalidate = REV_RECONFIGURE;
1613 dpif_sflow_set_options(ds, sflow_options);
1616 dpif_sflow_destroy(ds);
1617 ofproto->backer->need_revalidate = REV_RECONFIGURE;
1618 ofproto->sflow = NULL;
1626 struct ofproto *ofproto_,
1627 const struct ofproto_ipfix_bridge_exporter_options *bridge_exporter_options,
1628 const struct ofproto_ipfix_flow_exporter_options *flow_exporters_options,
1629 size_t n_flow_exporters_options)
1631 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1632 struct dpif_ipfix *di = ofproto->ipfix;
1634 if (bridge_exporter_options || flow_exporters_options) {
1636 di = ofproto->ipfix = dpif_ipfix_create();
1638 dpif_ipfix_set_options(
1639 di, bridge_exporter_options, flow_exporters_options,
1640 n_flow_exporters_options);
1643 dpif_ipfix_destroy(di);
1644 ofproto->ipfix = NULL;
1651 set_cfm(struct ofport *ofport_, const struct cfm_settings *s)
1653 struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
1660 struct ofproto_dpif *ofproto;
1662 ofproto = ofproto_dpif_cast(ofport->up.ofproto);
1663 ofproto->backer->need_revalidate = REV_RECONFIGURE;
1664 ofport->cfm = cfm_create(ofport->up.netdev);
1667 if (cfm_configure(ofport->cfm, s)) {
1673 cfm_destroy(ofport->cfm);
1679 get_cfm_status(const struct ofport *ofport_,
1680 struct ofproto_cfm_status *status)
1682 struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
1685 status->faults = cfm_get_fault(ofport->cfm);
1686 status->remote_opstate = cfm_get_opup(ofport->cfm);
1687 status->health = cfm_get_health(ofport->cfm);
1688 cfm_get_remote_mpids(ofport->cfm, &status->rmps, &status->n_rmps);
1696 set_bfd(struct ofport *ofport_, const struct smap *cfg)
1698 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport_->ofproto);
1699 struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
1703 ofport->bfd = bfd_configure(old, netdev_get_name(ofport->up.netdev), cfg);
1704 if (ofport->bfd != old) {
1705 ofproto->backer->need_revalidate = REV_RECONFIGURE;
1712 get_bfd_status(struct ofport *ofport_, struct smap *smap)
1714 struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
1717 bfd_get_status(ofport->bfd, smap);
1724 /* Spanning Tree. */
1727 send_bpdu_cb(struct ofpbuf *pkt, int port_num, void *ofproto_)
1729 struct ofproto_dpif *ofproto = ofproto_;
1730 struct stp_port *sp = stp_get_port(ofproto->stp, port_num);
1731 struct ofport_dpif *ofport;
1733 ofport = stp_port_get_aux(sp);
1735 VLOG_WARN_RL(&rl, "%s: cannot send BPDU on unknown port %d",
1736 ofproto->up.name, port_num);
1738 struct eth_header *eth = pkt->l2;
1740 netdev_get_etheraddr(ofport->up.netdev, eth->eth_src);
1741 if (eth_addr_is_zero(eth->eth_src)) {
1742 VLOG_WARN_RL(&rl, "%s: cannot send BPDU on port %d "
1743 "with unknown MAC", ofproto->up.name, port_num);
1745 send_packet(ofport, pkt);
1751 /* Configures STP on 'ofproto_' using the settings defined in 's'. */
1753 set_stp(struct ofproto *ofproto_, const struct ofproto_stp_settings *s)
1755 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1757 /* Only revalidate flows if the configuration changed. */
1758 if (!s != !ofproto->stp) {
1759 ofproto->backer->need_revalidate = REV_RECONFIGURE;
1763 if (!ofproto->stp) {
1764 ofproto->stp = stp_create(ofproto_->name, s->system_id,
1765 send_bpdu_cb, ofproto);
1766 ofproto->stp_last_tick = time_msec();
1769 stp_set_bridge_id(ofproto->stp, s->system_id);
1770 stp_set_bridge_priority(ofproto->stp, s->priority);
1771 stp_set_hello_time(ofproto->stp, s->hello_time);
1772 stp_set_max_age(ofproto->stp, s->max_age);
1773 stp_set_forward_delay(ofproto->stp, s->fwd_delay);
1775 struct ofport *ofport;
1777 HMAP_FOR_EACH (ofport, hmap_node, &ofproto->up.ports) {
1778 set_stp_port(ofport, NULL);
1781 stp_destroy(ofproto->stp);
1782 ofproto->stp = NULL;
1789 get_stp_status(struct ofproto *ofproto_, struct ofproto_stp_status *s)
1791 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1795 s->bridge_id = stp_get_bridge_id(ofproto->stp);
1796 s->designated_root = stp_get_designated_root(ofproto->stp);
1797 s->root_path_cost = stp_get_root_path_cost(ofproto->stp);
1806 update_stp_port_state(struct ofport_dpif *ofport)
1808 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport->up.ofproto);
1809 enum stp_state state;
1811 /* Figure out new state. */
1812 state = ofport->stp_port ? stp_port_get_state(ofport->stp_port)
1816 if (ofport->stp_state != state) {
1817 enum ofputil_port_state of_state;
1820 VLOG_DBG_RL(&rl, "port %s: STP state changed from %s to %s",
1821 netdev_get_name(ofport->up.netdev),
1822 stp_state_name(ofport->stp_state),
1823 stp_state_name(state));
1824 if (stp_learn_in_state(ofport->stp_state)
1825 != stp_learn_in_state(state)) {
1826 /* xxx Learning action flows should also be flushed. */
1827 mac_learning_flush(ofproto->ml,
1828 &ofproto->backer->revalidate_set);
1830 fwd_change = stp_forward_in_state(ofport->stp_state)
1831 != stp_forward_in_state(state);
1833 ofproto->backer->need_revalidate = REV_STP;
1834 ofport->stp_state = state;
1835 ofport->stp_state_entered = time_msec();
1837 if (fwd_change && ofport->bundle) {
1838 bundle_update(ofport->bundle);
1841 /* Update the STP state bits in the OpenFlow port description. */
1842 of_state = ofport->up.pp.state & ~OFPUTIL_PS_STP_MASK;
1843 of_state |= (state == STP_LISTENING ? OFPUTIL_PS_STP_LISTEN
1844 : state == STP_LEARNING ? OFPUTIL_PS_STP_LEARN
1845 : state == STP_FORWARDING ? OFPUTIL_PS_STP_FORWARD
1846 : state == STP_BLOCKING ? OFPUTIL_PS_STP_BLOCK
1848 ofproto_port_set_state(&ofport->up, of_state);
1852 /* Configures STP on 'ofport_' using the settings defined in 's'. The
1853 * caller is responsible for assigning STP port numbers and ensuring
1854 * there are no duplicates. */
1856 set_stp_port(struct ofport *ofport_,
1857 const struct ofproto_port_stp_settings *s)
1859 struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
1860 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport->up.ofproto);
1861 struct stp_port *sp = ofport->stp_port;
1863 if (!s || !s->enable) {
1865 ofport->stp_port = NULL;
1866 stp_port_disable(sp);
1867 update_stp_port_state(ofport);
1870 } else if (sp && stp_port_no(sp) != s->port_num
1871 && ofport == stp_port_get_aux(sp)) {
1872 /* The port-id changed, so disable the old one if it's not
1873 * already in use by another port. */
1874 stp_port_disable(sp);
1877 sp = ofport->stp_port = stp_get_port(ofproto->stp, s->port_num);
1878 stp_port_enable(sp);
1880 stp_port_set_aux(sp, ofport);
1881 stp_port_set_priority(sp, s->priority);
1882 stp_port_set_path_cost(sp, s->path_cost);
1884 update_stp_port_state(ofport);
1890 get_stp_port_status(struct ofport *ofport_,
1891 struct ofproto_port_stp_status *s)
1893 struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
1894 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport->up.ofproto);
1895 struct stp_port *sp = ofport->stp_port;
1897 if (!ofproto->stp || !sp) {
1903 s->port_id = stp_port_get_id(sp);
1904 s->state = stp_port_get_state(sp);
1905 s->sec_in_state = (time_msec() - ofport->stp_state_entered) / 1000;
1906 s->role = stp_port_get_role(sp);
1907 stp_port_get_counts(sp, &s->tx_count, &s->rx_count, &s->error_count);
1913 stp_run(struct ofproto_dpif *ofproto)
1916 long long int now = time_msec();
1917 long long int elapsed = now - ofproto->stp_last_tick;
1918 struct stp_port *sp;
1921 stp_tick(ofproto->stp, MIN(INT_MAX, elapsed));
1922 ofproto->stp_last_tick = now;
1924 while (stp_get_changed_port(ofproto->stp, &sp)) {
1925 struct ofport_dpif *ofport = stp_port_get_aux(sp);
1928 update_stp_port_state(ofport);
1932 if (stp_check_and_reset_fdb_flush(ofproto->stp)) {
1933 mac_learning_flush(ofproto->ml, &ofproto->backer->revalidate_set);
1939 stp_wait(struct ofproto_dpif *ofproto)
1942 poll_timer_wait(1000);
1946 /* Returns true if STP should process 'flow'. Sets fields in 'wc' that
1947 * were used to make the determination.*/
1949 stp_should_process_flow(const struct flow *flow, struct flow_wildcards *wc)
1951 memset(&wc->masks.dl_dst, 0xff, sizeof wc->masks.dl_dst);
1952 return eth_addr_equals(flow->dl_dst, eth_addr_stp);
1956 stp_process_packet(const struct ofport_dpif *ofport,
1957 const struct ofpbuf *packet)
1959 struct ofpbuf payload = *packet;
1960 struct eth_header *eth = payload.data;
1961 struct stp_port *sp = ofport->stp_port;
1963 /* Sink packets on ports that have STP disabled when the bridge has
1965 if (!sp || stp_port_get_state(sp) == STP_DISABLED) {
1969 /* Trim off padding on payload. */
1970 if (payload.size > ntohs(eth->eth_type) + ETH_HEADER_LEN) {
1971 payload.size = ntohs(eth->eth_type) + ETH_HEADER_LEN;
1974 if (ofpbuf_try_pull(&payload, ETH_HEADER_LEN + LLC_HEADER_LEN)) {
1975 stp_received_bpdu(sp, payload.data, payload.size);
1980 ofproto_dpif_queue_to_priority(const struct ofproto_dpif *ofproto,
1981 uint32_t queue_id, uint32_t *priority)
1983 return dpif_queue_to_priority(ofproto->backer->dpif, queue_id, priority);
1986 static struct priority_to_dscp *
1987 get_priority(const struct ofport_dpif *ofport, uint32_t priority)
1989 struct priority_to_dscp *pdscp;
1992 hash = hash_int(priority, 0);
1993 HMAP_FOR_EACH_IN_BUCKET (pdscp, hmap_node, hash, &ofport->priorities) {
1994 if (pdscp->priority == priority) {
2002 ofproto_dpif_dscp_from_priority(const struct ofport_dpif *ofport,
2003 uint32_t priority, uint8_t *dscp)
2005 struct priority_to_dscp *pdscp = get_priority(ofport, priority);
2006 *dscp = pdscp ? pdscp->dscp : 0;
2007 return pdscp != NULL;
2011 ofport_clear_priorities(struct ofport_dpif *ofport)
2013 struct priority_to_dscp *pdscp, *next;
2015 HMAP_FOR_EACH_SAFE (pdscp, next, hmap_node, &ofport->priorities) {
2016 hmap_remove(&ofport->priorities, &pdscp->hmap_node);
2022 set_queues(struct ofport *ofport_,
2023 const struct ofproto_port_queue *qdscp_list,
2026 struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
2027 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport->up.ofproto);
2028 struct hmap new = HMAP_INITIALIZER(&new);
2031 for (i = 0; i < n_qdscp; i++) {
2032 struct priority_to_dscp *pdscp;
2036 dscp = (qdscp_list[i].dscp << 2) & IP_DSCP_MASK;
2037 if (dpif_queue_to_priority(ofproto->backer->dpif, qdscp_list[i].queue,
2042 pdscp = get_priority(ofport, priority);
2044 hmap_remove(&ofport->priorities, &pdscp->hmap_node);
2046 pdscp = xmalloc(sizeof *pdscp);
2047 pdscp->priority = priority;
2049 ofproto->backer->need_revalidate = REV_RECONFIGURE;
2052 if (pdscp->dscp != dscp) {
2054 ofproto->backer->need_revalidate = REV_RECONFIGURE;
2057 hmap_insert(&new, &pdscp->hmap_node, hash_int(pdscp->priority, 0));
2060 if (!hmap_is_empty(&ofport->priorities)) {
2061 ofport_clear_priorities(ofport);
2062 ofproto->backer->need_revalidate = REV_RECONFIGURE;
2065 hmap_swap(&new, &ofport->priorities);
2073 /* Expires all MAC learning entries associated with 'bundle' and forces its
2074 * ofproto to revalidate every flow.
2076 * Normally MAC learning entries are removed only from the ofproto associated
2077 * with 'bundle', but if 'all_ofprotos' is true, then the MAC learning entries
2078 * are removed from every ofproto. When patch ports and SLB bonds are in use
2079 * and a VM migration happens and the gratuitous ARPs are somehow lost, this
2080 * avoids a MAC_ENTRY_IDLE_TIME delay before the migrated VM can communicate
2081 * with the host from which it migrated. */
2083 bundle_flush_macs(struct ofbundle *bundle, bool all_ofprotos)
2085 struct ofproto_dpif *ofproto = bundle->ofproto;
2086 struct mac_learning *ml = ofproto->ml;
2087 struct mac_entry *mac, *next_mac;
2089 ofproto->backer->need_revalidate = REV_RECONFIGURE;
2090 LIST_FOR_EACH_SAFE (mac, next_mac, lru_node, &ml->lrus) {
2091 if (mac->port.p == bundle) {
2093 struct ofproto_dpif *o;
2095 HMAP_FOR_EACH (o, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
2097 struct mac_entry *e;
2099 e = mac_learning_lookup(o->ml, mac->mac, mac->vlan,
2102 mac_learning_expire(o->ml, e);
2108 mac_learning_expire(ml, mac);
2113 static struct ofbundle *
2114 bundle_lookup(const struct ofproto_dpif *ofproto, void *aux)
2116 struct ofbundle *bundle;
2118 HMAP_FOR_EACH_IN_BUCKET (bundle, hmap_node, hash_pointer(aux, 0),
2119 &ofproto->bundles) {
2120 if (bundle->aux == aux) {
2127 /* Looks up each of the 'n_auxes' pointers in 'auxes' as bundles and adds the
2128 * ones that are found to 'bundles'. */
2130 bundle_lookup_multiple(struct ofproto_dpif *ofproto,
2131 void **auxes, size_t n_auxes,
2132 struct hmapx *bundles)
2136 hmapx_init(bundles);
2137 for (i = 0; i < n_auxes; i++) {
2138 struct ofbundle *bundle = bundle_lookup(ofproto, auxes[i]);
2140 hmapx_add(bundles, bundle);
2146 bundle_update(struct ofbundle *bundle)
2148 struct ofport_dpif *port;
2150 bundle->floodable = true;
2151 LIST_FOR_EACH (port, bundle_node, &bundle->ports) {
2152 if (port->up.pp.config & OFPUTIL_PC_NO_FLOOD
2153 || !stp_forward_in_state(port->stp_state)) {
2154 bundle->floodable = false;
2161 bundle_del_port(struct ofport_dpif *port)
2163 struct ofbundle *bundle = port->bundle;
2165 bundle->ofproto->backer->need_revalidate = REV_RECONFIGURE;
2167 list_remove(&port->bundle_node);
2168 port->bundle = NULL;
2171 lacp_slave_unregister(bundle->lacp, port);
2174 bond_slave_unregister(bundle->bond, port);
2177 bundle_update(bundle);
2181 bundle_add_port(struct ofbundle *bundle, uint16_t ofp_port,
2182 struct lacp_slave_settings *lacp)
2184 struct ofport_dpif *port;
2186 port = get_ofp_port(bundle->ofproto, ofp_port);
2191 if (port->bundle != bundle) {
2192 bundle->ofproto->backer->need_revalidate = REV_RECONFIGURE;
2194 bundle_del_port(port);
2197 port->bundle = bundle;
2198 list_push_back(&bundle->ports, &port->bundle_node);
2199 if (port->up.pp.config & OFPUTIL_PC_NO_FLOOD
2200 || !stp_forward_in_state(port->stp_state)) {
2201 bundle->floodable = false;
2205 bundle->ofproto->backer->need_revalidate = REV_RECONFIGURE;
2206 lacp_slave_register(bundle->lacp, port, lacp);
2213 bundle_destroy(struct ofbundle *bundle)
2215 struct ofproto_dpif *ofproto;
2216 struct ofport_dpif *port, *next_port;
2223 ofproto = bundle->ofproto;
2224 for (i = 0; i < MAX_MIRRORS; i++) {
2225 struct ofmirror *m = ofproto->mirrors[i];
2227 if (m->out == bundle) {
2229 } else if (hmapx_find_and_delete(&m->srcs, bundle)
2230 || hmapx_find_and_delete(&m->dsts, bundle)) {
2231 ofproto->backer->need_revalidate = REV_RECONFIGURE;
2236 LIST_FOR_EACH_SAFE (port, next_port, bundle_node, &bundle->ports) {
2237 bundle_del_port(port);
2240 bundle_flush_macs(bundle, true);
2241 hmap_remove(&ofproto->bundles, &bundle->hmap_node);
2243 free(bundle->trunks);
2244 lacp_destroy(bundle->lacp);
2245 bond_destroy(bundle->bond);
2250 bundle_set(struct ofproto *ofproto_, void *aux,
2251 const struct ofproto_bundle_settings *s)
2253 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
2254 bool need_flush = false;
2255 struct ofport_dpif *port;
2256 struct ofbundle *bundle;
2257 unsigned long *trunks;
2263 bundle_destroy(bundle_lookup(ofproto, aux));
2267 ovs_assert(s->n_slaves == 1 || s->bond != NULL);
2268 ovs_assert((s->lacp != NULL) == (s->lacp_slaves != NULL));
2270 bundle = bundle_lookup(ofproto, aux);
2272 bundle = xmalloc(sizeof *bundle);
2274 bundle->ofproto = ofproto;
2275 hmap_insert(&ofproto->bundles, &bundle->hmap_node,
2276 hash_pointer(aux, 0));
2278 bundle->name = NULL;
2280 list_init(&bundle->ports);
2281 bundle->vlan_mode = PORT_VLAN_TRUNK;
2283 bundle->trunks = NULL;
2284 bundle->use_priority_tags = s->use_priority_tags;
2285 bundle->lacp = NULL;
2286 bundle->bond = NULL;
2288 bundle->floodable = true;
2290 bundle->src_mirrors = 0;
2291 bundle->dst_mirrors = 0;
2292 bundle->mirror_out = 0;
2295 if (!bundle->name || strcmp(s->name, bundle->name)) {
2297 bundle->name = xstrdup(s->name);
2302 if (!bundle->lacp) {
2303 ofproto->backer->need_revalidate = REV_RECONFIGURE;
2304 bundle->lacp = lacp_create();
2306 lacp_configure(bundle->lacp, s->lacp);
2308 lacp_destroy(bundle->lacp);
2309 bundle->lacp = NULL;
2312 /* Update set of ports. */
2314 for (i = 0; i < s->n_slaves; i++) {
2315 if (!bundle_add_port(bundle, s->slaves[i],
2316 s->lacp ? &s->lacp_slaves[i] : NULL)) {
2320 if (!ok || list_size(&bundle->ports) != s->n_slaves) {
2321 struct ofport_dpif *next_port;
2323 LIST_FOR_EACH_SAFE (port, next_port, bundle_node, &bundle->ports) {
2324 for (i = 0; i < s->n_slaves; i++) {
2325 if (s->slaves[i] == port->up.ofp_port) {
2330 bundle_del_port(port);
2334 ovs_assert(list_size(&bundle->ports) <= s->n_slaves);
2336 if (list_is_empty(&bundle->ports)) {
2337 bundle_destroy(bundle);
2341 /* Set VLAN tagging mode */
2342 if (s->vlan_mode != bundle->vlan_mode
2343 || s->use_priority_tags != bundle->use_priority_tags) {
2344 bundle->vlan_mode = s->vlan_mode;
2345 bundle->use_priority_tags = s->use_priority_tags;
2350 vlan = (s->vlan_mode == PORT_VLAN_TRUNK ? -1
2351 : s->vlan >= 0 && s->vlan <= 4095 ? s->vlan
2353 if (vlan != bundle->vlan) {
2354 bundle->vlan = vlan;
2358 /* Get trunked VLANs. */
2359 switch (s->vlan_mode) {
2360 case PORT_VLAN_ACCESS:
2364 case PORT_VLAN_TRUNK:
2365 trunks = CONST_CAST(unsigned long *, s->trunks);
2368 case PORT_VLAN_NATIVE_UNTAGGED:
2369 case PORT_VLAN_NATIVE_TAGGED:
2370 if (vlan != 0 && (!s->trunks
2371 || !bitmap_is_set(s->trunks, vlan)
2372 || bitmap_is_set(s->trunks, 0))) {
2373 /* Force trunking the native VLAN and prohibit trunking VLAN 0. */
2375 trunks = bitmap_clone(s->trunks, 4096);
2377 trunks = bitmap_allocate1(4096);
2379 bitmap_set1(trunks, vlan);
2380 bitmap_set0(trunks, 0);
2382 trunks = CONST_CAST(unsigned long *, s->trunks);
2389 if (!vlan_bitmap_equal(trunks, bundle->trunks)) {
2390 free(bundle->trunks);
2391 if (trunks == s->trunks) {
2392 bundle->trunks = vlan_bitmap_clone(trunks);
2394 bundle->trunks = trunks;
2399 if (trunks != s->trunks) {
2404 if (!list_is_short(&bundle->ports)) {
2405 bundle->ofproto->has_bonded_bundles = true;
2407 if (bond_reconfigure(bundle->bond, s->bond)) {
2408 ofproto->backer->need_revalidate = REV_RECONFIGURE;
2411 bundle->bond = bond_create(s->bond);
2412 ofproto->backer->need_revalidate = REV_RECONFIGURE;
2415 LIST_FOR_EACH (port, bundle_node, &bundle->ports) {
2416 bond_slave_register(bundle->bond, port, port->up.netdev);
2419 bond_destroy(bundle->bond);
2420 bundle->bond = NULL;
2423 /* If we changed something that would affect MAC learning, un-learn
2424 * everything on this port and force flow revalidation. */
2426 bundle_flush_macs(bundle, false);
2433 bundle_remove(struct ofport *port_)
2435 struct ofport_dpif *port = ofport_dpif_cast(port_);
2436 struct ofbundle *bundle = port->bundle;
2439 bundle_del_port(port);
2440 if (list_is_empty(&bundle->ports)) {
2441 bundle_destroy(bundle);
2442 } else if (list_is_short(&bundle->ports)) {
2443 bond_destroy(bundle->bond);
2444 bundle->bond = NULL;
2450 send_pdu_cb(void *port_, const void *pdu, size_t pdu_size)
2452 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 10);
2453 struct ofport_dpif *port = port_;
2454 uint8_t ea[ETH_ADDR_LEN];
2457 error = netdev_get_etheraddr(port->up.netdev, ea);
2459 struct ofpbuf packet;
2462 ofpbuf_init(&packet, 0);
2463 packet_pdu = eth_compose(&packet, eth_addr_lacp, ea, ETH_TYPE_LACP,
2465 memcpy(packet_pdu, pdu, pdu_size);
2467 send_packet(port, &packet);
2468 ofpbuf_uninit(&packet);
2470 VLOG_ERR_RL(&rl, "port %s: cannot obtain Ethernet address of iface "
2471 "%s (%s)", port->bundle->name,
2472 netdev_get_name(port->up.netdev), strerror(error));
2477 bundle_send_learning_packets(struct ofbundle *bundle)
2479 struct ofproto_dpif *ofproto = bundle->ofproto;
2480 int error, n_packets, n_errors;
2481 struct mac_entry *e;
2483 error = n_packets = n_errors = 0;
2484 LIST_FOR_EACH (e, lru_node, &ofproto->ml->lrus) {
2485 if (e->port.p != bundle) {
2486 struct ofpbuf *learning_packet;
2487 struct ofport_dpif *port;
2491 /* The assignment to "port" is unnecessary but makes "grep"ing for
2492 * struct ofport_dpif more effective. */
2493 learning_packet = bond_compose_learning_packet(bundle->bond,
2497 ret = send_packet(port, learning_packet);
2498 ofpbuf_delete(learning_packet);
2508 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
2509 VLOG_WARN_RL(&rl, "bond %s: %d errors sending %d gratuitous learning "
2510 "packets, last error was: %s",
2511 bundle->name, n_errors, n_packets, strerror(error));
2513 VLOG_DBG("bond %s: sent %d gratuitous learning packets",
2514 bundle->name, n_packets);
2519 bundle_run(struct ofbundle *bundle)
2522 lacp_run(bundle->lacp, send_pdu_cb);
2525 struct ofport_dpif *port;
2527 LIST_FOR_EACH (port, bundle_node, &bundle->ports) {
2528 bond_slave_set_may_enable(bundle->bond, port, port->may_enable);
2531 bond_run(bundle->bond, &bundle->ofproto->backer->revalidate_set,
2532 lacp_status(bundle->lacp));
2533 if (bond_should_send_learning_packets(bundle->bond)) {
2534 bundle_send_learning_packets(bundle);
2540 bundle_wait(struct ofbundle *bundle)
2543 lacp_wait(bundle->lacp);
2546 bond_wait(bundle->bond);
2553 mirror_scan(struct ofproto_dpif *ofproto)
2557 for (idx = 0; idx < MAX_MIRRORS; idx++) {
2558 if (!ofproto->mirrors[idx]) {
2565 static struct ofmirror *
2566 mirror_lookup(struct ofproto_dpif *ofproto, void *aux)
2570 for (i = 0; i < MAX_MIRRORS; i++) {
2571 struct ofmirror *mirror = ofproto->mirrors[i];
2572 if (mirror && mirror->aux == aux) {
2580 /* Update the 'dup_mirrors' member of each of the ofmirrors in 'ofproto'. */
2582 mirror_update_dups(struct ofproto_dpif *ofproto)
2586 for (i = 0; i < MAX_MIRRORS; i++) {
2587 struct ofmirror *m = ofproto->mirrors[i];
2590 m->dup_mirrors = MIRROR_MASK_C(1) << i;
2594 for (i = 0; i < MAX_MIRRORS; i++) {
2595 struct ofmirror *m1 = ofproto->mirrors[i];
2602 for (j = i + 1; j < MAX_MIRRORS; j++) {
2603 struct ofmirror *m2 = ofproto->mirrors[j];
2605 if (m2 && m1->out == m2->out && m1->out_vlan == m2->out_vlan) {
2606 m1->dup_mirrors |= MIRROR_MASK_C(1) << j;
2607 m2->dup_mirrors |= m1->dup_mirrors;
2614 mirror_set(struct ofproto *ofproto_, void *aux,
2615 const struct ofproto_mirror_settings *s)
2617 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
2618 mirror_mask_t mirror_bit;
2619 struct ofbundle *bundle;
2620 struct ofmirror *mirror;
2621 struct ofbundle *out;
2622 struct hmapx srcs; /* Contains "struct ofbundle *"s. */
2623 struct hmapx dsts; /* Contains "struct ofbundle *"s. */
2626 mirror = mirror_lookup(ofproto, aux);
2628 mirror_destroy(mirror);
2634 idx = mirror_scan(ofproto);
2636 VLOG_WARN("bridge %s: maximum of %d port mirrors reached, "
2638 ofproto->up.name, MAX_MIRRORS, s->name);
2642 mirror = ofproto->mirrors[idx] = xzalloc(sizeof *mirror);
2643 mirror->ofproto = ofproto;
2646 mirror->out_vlan = -1;
2647 mirror->name = NULL;
2650 if (!mirror->name || strcmp(s->name, mirror->name)) {
2652 mirror->name = xstrdup(s->name);
2655 /* Get the new configuration. */
2656 if (s->out_bundle) {
2657 out = bundle_lookup(ofproto, s->out_bundle);
2659 mirror_destroy(mirror);
2665 out_vlan = s->out_vlan;
2667 bundle_lookup_multiple(ofproto, s->srcs, s->n_srcs, &srcs);
2668 bundle_lookup_multiple(ofproto, s->dsts, s->n_dsts, &dsts);
2670 /* If the configuration has not changed, do nothing. */
2671 if (hmapx_equals(&srcs, &mirror->srcs)
2672 && hmapx_equals(&dsts, &mirror->dsts)
2673 && vlan_bitmap_equal(mirror->vlans, s->src_vlans)
2674 && mirror->out == out
2675 && mirror->out_vlan == out_vlan)
2677 hmapx_destroy(&srcs);
2678 hmapx_destroy(&dsts);
2682 hmapx_swap(&srcs, &mirror->srcs);
2683 hmapx_destroy(&srcs);
2685 hmapx_swap(&dsts, &mirror->dsts);
2686 hmapx_destroy(&dsts);
2688 free(mirror->vlans);
2689 mirror->vlans = vlan_bitmap_clone(s->src_vlans);
2692 mirror->out_vlan = out_vlan;
2694 /* Update bundles. */
2695 mirror_bit = MIRROR_MASK_C(1) << mirror->idx;
2696 HMAP_FOR_EACH (bundle, hmap_node, &mirror->ofproto->bundles) {
2697 if (hmapx_contains(&mirror->srcs, bundle)) {
2698 bundle->src_mirrors |= mirror_bit;
2700 bundle->src_mirrors &= ~mirror_bit;
2703 if (hmapx_contains(&mirror->dsts, bundle)) {
2704 bundle->dst_mirrors |= mirror_bit;
2706 bundle->dst_mirrors &= ~mirror_bit;
2709 if (mirror->out == bundle) {
2710 bundle->mirror_out |= mirror_bit;
2712 bundle->mirror_out &= ~mirror_bit;
2716 ofproto->backer->need_revalidate = REV_RECONFIGURE;
2717 ofproto->has_mirrors = true;
2718 mac_learning_flush(ofproto->ml,
2719 &ofproto->backer->revalidate_set);
2720 mirror_update_dups(ofproto);
2726 mirror_destroy(struct ofmirror *mirror)
2728 struct ofproto_dpif *ofproto;
2729 mirror_mask_t mirror_bit;
2730 struct ofbundle *bundle;
2737 ofproto = mirror->ofproto;
2738 ofproto->backer->need_revalidate = REV_RECONFIGURE;
2739 mac_learning_flush(ofproto->ml, &ofproto->backer->revalidate_set);
2741 mirror_bit = MIRROR_MASK_C(1) << mirror->idx;
2742 HMAP_FOR_EACH (bundle, hmap_node, &ofproto->bundles) {
2743 bundle->src_mirrors &= ~mirror_bit;
2744 bundle->dst_mirrors &= ~mirror_bit;
2745 bundle->mirror_out &= ~mirror_bit;
2748 hmapx_destroy(&mirror->srcs);
2749 hmapx_destroy(&mirror->dsts);
2750 free(mirror->vlans);
2752 ofproto->mirrors[mirror->idx] = NULL;
2756 mirror_update_dups(ofproto);
2758 ofproto->has_mirrors = false;
2759 for (i = 0; i < MAX_MIRRORS; i++) {
2760 if (ofproto->mirrors[i]) {
2761 ofproto->has_mirrors = true;
2768 mirror_get_stats(struct ofproto *ofproto_, void *aux,
2769 uint64_t *packets, uint64_t *bytes)
2771 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
2772 struct ofmirror *mirror = mirror_lookup(ofproto, aux);
2775 *packets = *bytes = UINT64_MAX;
2781 *packets = mirror->packet_count;
2782 *bytes = mirror->byte_count;
2788 set_flood_vlans(struct ofproto *ofproto_, unsigned long *flood_vlans)
2790 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
2791 if (mac_learning_set_flood_vlans(ofproto->ml, flood_vlans)) {
2792 mac_learning_flush(ofproto->ml, &ofproto->backer->revalidate_set);
2798 is_mirror_output_bundle(const struct ofproto *ofproto_, void *aux)
2800 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
2801 struct ofbundle *bundle = bundle_lookup(ofproto, aux);
2802 return bundle && bundle->mirror_out != 0;
2806 forward_bpdu_changed(struct ofproto *ofproto_)
2808 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
2809 ofproto->backer->need_revalidate = REV_RECONFIGURE;
2813 set_mac_table_config(struct ofproto *ofproto_, unsigned int idle_time,
2816 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
2817 mac_learning_set_idle_time(ofproto->ml, idle_time);
2818 mac_learning_set_max_entries(ofproto->ml, max_entries);
2823 struct ofport_dpif *
2824 get_ofp_port(const struct ofproto_dpif *ofproto, uint16_t ofp_port)
2826 struct ofport *ofport = ofproto_get_port(&ofproto->up, ofp_port);
2827 return ofport ? ofport_dpif_cast(ofport) : NULL;
2830 struct ofport_dpif *
2831 get_odp_port(const struct ofproto_dpif *ofproto, uint32_t odp_port)
2833 struct ofport_dpif *port = odp_port_to_ofport(ofproto->backer, odp_port);
2834 return port && &ofproto->up == port->up.ofproto ? port : NULL;
2838 ofproto_port_from_dpif_port(struct ofproto_dpif *ofproto,
2839 struct ofproto_port *ofproto_port,
2840 struct dpif_port *dpif_port)
2842 ofproto_port->name = dpif_port->name;
2843 ofproto_port->type = dpif_port->type;
2844 ofproto_port->ofp_port = odp_port_to_ofp_port(ofproto, dpif_port->port_no);
2848 ofport_update_peer(struct ofport_dpif *ofport)
2850 const struct ofproto_dpif *ofproto;
2851 struct dpif_backer *backer;
2852 const char *peer_name;
2854 if (!netdev_vport_is_patch(ofport->up.netdev)) {
2858 backer = ofproto_dpif_cast(ofport->up.ofproto)->backer;
2859 backer->need_revalidate = true;
2862 ofport->peer->peer = NULL;
2863 ofport->peer = NULL;
2866 peer_name = netdev_vport_patch_peer(ofport->up.netdev);
2871 HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
2872 struct ofport *peer_ofport;
2873 struct ofport_dpif *peer;
2874 const char *peer_peer;
2876 if (ofproto->backer != backer) {
2880 peer_ofport = shash_find_data(&ofproto->up.port_by_name, peer_name);
2885 peer = ofport_dpif_cast(peer_ofport);
2886 peer_peer = netdev_vport_patch_peer(peer->up.netdev);
2887 if (peer_peer && !strcmp(netdev_get_name(ofport->up.netdev),
2889 ofport->peer = peer;
2890 ofport->peer->peer = ofport;
2898 port_run_fast(struct ofport_dpif *ofport)
2900 if (ofport->cfm && cfm_should_send_ccm(ofport->cfm)) {
2901 struct ofpbuf packet;
2903 ofpbuf_init(&packet, 0);
2904 cfm_compose_ccm(ofport->cfm, &packet, ofport->up.pp.hw_addr);
2905 send_packet(ofport, &packet);
2906 ofpbuf_uninit(&packet);
2909 if (ofport->bfd && bfd_should_send_packet(ofport->bfd)) {
2910 struct ofpbuf packet;
2912 ofpbuf_init(&packet, 0);
2913 bfd_put_packet(ofport->bfd, &packet, ofport->up.pp.hw_addr);
2914 send_packet(ofport, &packet);
2915 ofpbuf_uninit(&packet);
2920 port_run(struct ofport_dpif *ofport)
2922 long long int carrier_seq = netdev_get_carrier_resets(ofport->up.netdev);
2923 bool carrier_changed = carrier_seq != ofport->carrier_seq;
2924 bool enable = netdev_get_carrier(ofport->up.netdev);
2926 ofport->carrier_seq = carrier_seq;
2928 port_run_fast(ofport);
2931 int cfm_opup = cfm_get_opup(ofport->cfm);
2933 cfm_run(ofport->cfm);
2934 enable = enable && !cfm_get_fault(ofport->cfm);
2936 if (cfm_opup >= 0) {
2937 enable = enable && cfm_opup;
2942 bfd_run(ofport->bfd);
2943 enable = enable && bfd_forwarding(ofport->bfd);
2946 if (ofport->bundle) {
2947 enable = enable && lacp_slave_may_enable(ofport->bundle->lacp, ofport);
2948 if (carrier_changed) {
2949 lacp_slave_carrier_changed(ofport->bundle->lacp, ofport);
2953 if (ofport->may_enable != enable) {
2954 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport->up.ofproto);
2956 if (ofproto->has_bundle_action) {
2957 ofproto->backer->need_revalidate = REV_PORT_TOGGLED;
2961 ofport->may_enable = enable;
2965 port_wait(struct ofport_dpif *ofport)
2968 cfm_wait(ofport->cfm);
2972 bfd_wait(ofport->bfd);
2977 port_query_by_name(const struct ofproto *ofproto_, const char *devname,
2978 struct ofproto_port *ofproto_port)
2980 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
2981 struct dpif_port dpif_port;
2984 if (sset_contains(&ofproto->ghost_ports, devname)) {
2985 const char *type = netdev_get_type_from_name(devname);
2987 /* We may be called before ofproto->up.port_by_name is populated with
2988 * the appropriate ofport. For this reason, we must get the name and
2989 * type from the netdev layer directly. */
2991 const struct ofport *ofport;
2993 ofport = shash_find_data(&ofproto->up.port_by_name, devname);
2994 ofproto_port->ofp_port = ofport ? ofport->ofp_port : OFPP_NONE;
2995 ofproto_port->name = xstrdup(devname);
2996 ofproto_port->type = xstrdup(type);
3002 if (!sset_contains(&ofproto->ports, devname)) {
3005 error = dpif_port_query_by_name(ofproto->backer->dpif,
3006 devname, &dpif_port);
3008 ofproto_port_from_dpif_port(ofproto, ofproto_port, &dpif_port);
3014 port_add(struct ofproto *ofproto_, struct netdev *netdev)
3016 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
3017 const char *devname = netdev_get_name(netdev);
3018 char namebuf[NETDEV_VPORT_NAME_BUFSIZE];
3019 const char *dp_port_name;
3021 if (netdev_vport_is_patch(netdev)) {
3022 sset_add(&ofproto->ghost_ports, netdev_get_name(netdev));
3026 dp_port_name = netdev_vport_get_dpif_port(netdev, namebuf, sizeof namebuf);
3027 if (!dpif_port_exists(ofproto->backer->dpif, dp_port_name)) {
3028 uint32_t port_no = UINT32_MAX;
3031 error = dpif_port_add(ofproto->backer->dpif, netdev, &port_no);
3035 if (netdev_get_tunnel_config(netdev)) {
3036 simap_put(&ofproto->backer->tnl_backers, dp_port_name, port_no);
3040 if (netdev_get_tunnel_config(netdev)) {
3041 sset_add(&ofproto->ghost_ports, devname);
3043 sset_add(&ofproto->ports, devname);
3049 port_del(struct ofproto *ofproto_, uint16_t ofp_port)
3051 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
3052 struct ofport_dpif *ofport = get_ofp_port(ofproto, ofp_port);
3059 sset_find_and_delete(&ofproto->ghost_ports,
3060 netdev_get_name(ofport->up.netdev));
3061 ofproto->backer->need_revalidate = REV_RECONFIGURE;
3062 if (!ofport->tnl_port) {
3063 error = dpif_port_del(ofproto->backer->dpif, ofport->odp_port);
3065 /* The caller is going to close ofport->up.netdev. If this is a
3066 * bonded port, then the bond is using that netdev, so remove it
3067 * from the bond. The client will need to reconfigure everything
3068 * after deleting ports, so then the slave will get re-added. */
3069 bundle_remove(&ofport->up);
3076 port_get_stats(const struct ofport *ofport_, struct netdev_stats *stats)
3078 struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
3083 error = netdev_get_stats(ofport->up.netdev, stats);
3085 if (!error && ofport_->ofp_port == OFPP_LOCAL) {
3086 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport->up.ofproto);
3088 /* ofproto->stats.tx_packets represents packets that we created
3089 * internally and sent to some port (e.g. packets sent with
3090 * send_packet()). Account for them as if they had come from
3091 * OFPP_LOCAL and got forwarded. */
3093 if (stats->rx_packets != UINT64_MAX) {
3094 stats->rx_packets += ofproto->stats.tx_packets;
3097 if (stats->rx_bytes != UINT64_MAX) {
3098 stats->rx_bytes += ofproto->stats.tx_bytes;
3101 /* ofproto->stats.rx_packets represents packets that were received on
3102 * some port and we processed internally and dropped (e.g. STP).
3103 * Account for them as if they had been forwarded to OFPP_LOCAL. */
3105 if (stats->tx_packets != UINT64_MAX) {
3106 stats->tx_packets += ofproto->stats.rx_packets;
3109 if (stats->tx_bytes != UINT64_MAX) {
3110 stats->tx_bytes += ofproto->stats.rx_bytes;
3117 struct port_dump_state {
3122 struct ofproto_port port;
3127 port_dump_start(const struct ofproto *ofproto_ OVS_UNUSED, void **statep)
3129 *statep = xzalloc(sizeof(struct port_dump_state));
3134 port_dump_next(const struct ofproto *ofproto_, void *state_,
3135 struct ofproto_port *port)
3137 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
3138 struct port_dump_state *state = state_;
3139 const struct sset *sset;
3140 struct sset_node *node;
3142 if (state->has_port) {
3143 ofproto_port_destroy(&state->port);
3144 state->has_port = false;
3146 sset = state->ghost ? &ofproto->ghost_ports : &ofproto->ports;
3147 while ((node = sset_at_position(sset, &state->bucket, &state->offset))) {
3150 error = port_query_by_name(ofproto_, node->name, &state->port);
3152 *port = state->port;
3153 state->has_port = true;
3155 } else if (error != ENODEV) {
3160 if (!state->ghost) {
3161 state->ghost = true;
3164 return port_dump_next(ofproto_, state_, port);
3171 port_dump_done(const struct ofproto *ofproto_ OVS_UNUSED, void *state_)
3173 struct port_dump_state *state = state_;
3175 if (state->has_port) {
3176 ofproto_port_destroy(&state->port);
3183 port_poll(const struct ofproto *ofproto_, char **devnamep)
3185 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
3187 if (ofproto->port_poll_errno) {
3188 int error = ofproto->port_poll_errno;
3189 ofproto->port_poll_errno = 0;
3193 if (sset_is_empty(&ofproto->port_poll_set)) {
3197 *devnamep = sset_pop(&ofproto->port_poll_set);
3202 port_poll_wait(const struct ofproto *ofproto_)
3204 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
3205 dpif_port_poll_wait(ofproto->backer->dpif);
3209 port_is_lacp_current(const struct ofport *ofport_)
3211 const struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
3212 return (ofport->bundle && ofport->bundle->lacp
3213 ? lacp_slave_is_current(ofport->bundle->lacp, ofport)
3217 /* Upcall handling. */
3219 /* Flow miss batching.
3221 * Some dpifs implement operations faster when you hand them off in a batch.
3222 * To allow batching, "struct flow_miss" queues the dpif-related work needed
3223 * for a given flow. Each "struct flow_miss" corresponds to sending one or
3224 * more packets, plus possibly installing the flow in the dpif.
3226 * So far we only batch the operations that affect flow setup time the most.
3227 * It's possible to batch more than that, but the benefit might be minimal. */
3229 struct hmap_node hmap_node;
3230 struct ofproto_dpif *ofproto;
3232 enum odp_key_fitness key_fitness;
3233 const struct nlattr *key;
3235 struct list packets;
3236 enum dpif_upcall_type upcall_type;
3239 struct flow_miss_op {
3240 struct dpif_op dpif_op;
3242 uint64_t slow_stub[128 / 8]; /* Buffer for compose_slow_path() */
3243 struct xlate_out xout;
3244 bool xout_garbage; /* 'xout' needs to be uninitialized? */
3247 /* Sends an OFPT_PACKET_IN message for 'packet' of type OFPR_NO_MATCH to each
3248 * OpenFlow controller as necessary according to their individual
3249 * configurations. */
3251 send_packet_in_miss(struct ofproto_dpif *ofproto, const struct ofpbuf *packet,
3252 const struct flow *flow)
3254 struct ofputil_packet_in pin;
3256 pin.packet = packet->data;
3257 pin.packet_len = packet->size;
3258 pin.reason = OFPR_NO_MATCH;
3259 pin.controller_id = 0;
3264 pin.send_len = 0; /* not used for flow table misses */
3266 flow_get_metadata(flow, &pin.fmd);
3268 connmgr_send_packet_in(ofproto->up.connmgr, &pin);
3271 static struct flow_miss *
3272 flow_miss_find(struct hmap *todo, const struct ofproto_dpif *ofproto,
3273 const struct flow *flow, uint32_t hash)
3275 struct flow_miss *miss;
3277 HMAP_FOR_EACH_WITH_HASH (miss, hmap_node, hash, todo) {
3278 if (miss->ofproto == ofproto && flow_equal(&miss->flow, flow)) {
3286 /* Partially Initializes 'op' as an "execute" operation for 'miss' and
3287 * 'packet'. The caller must initialize op->actions and op->actions_len. If
3288 * 'miss' is associated with a subfacet the caller must also initialize the
3289 * returned op->subfacet, and if anything needs to be freed after processing
3290 * the op, the caller must initialize op->garbage also. */
3292 init_flow_miss_execute_op(struct flow_miss *miss, struct ofpbuf *packet,
3293 struct flow_miss_op *op)
3295 if (miss->flow.in_port
3296 != vsp_realdev_to_vlandev(miss->ofproto, miss->flow.in_port,
3297 miss->flow.vlan_tci)) {
3298 /* This packet was received on a VLAN splinter port. We
3299 * added a VLAN to the packet to make the packet resemble
3300 * the flow, but the actions were composed assuming that
3301 * the packet contained no VLAN. So, we must remove the
3302 * VLAN header from the packet before trying to execute the
3304 eth_pop_vlan(packet);
3307 op->xout_garbage = false;
3308 op->dpif_op.type = DPIF_OP_EXECUTE;
3309 op->dpif_op.u.execute.key = miss->key;
3310 op->dpif_op.u.execute.key_len = miss->key_len;
3311 op->dpif_op.u.execute.packet = packet;
3314 /* Helper for handle_flow_miss_without_facet() and
3315 * handle_flow_miss_with_facet(). */
3317 handle_flow_miss_common(struct rule_dpif *rule,
3318 struct ofpbuf *packet, const struct flow *flow)
3320 struct ofproto_dpif *ofproto = ofproto_dpif_cast(rule->up.ofproto);
3322 if (rule->up.cr.priority == FAIL_OPEN_PRIORITY) {
3324 * Extra-special case for fail-open mode.
3326 * We are in fail-open mode and the packet matched the fail-open
3327 * rule, but we are connected to a controller too. We should send
3328 * the packet up to the controller in the hope that it will try to
3329 * set up a flow and thereby allow us to exit fail-open.
3331 * See the top-level comment in fail-open.c for more information.
3333 send_packet_in_miss(ofproto, packet, flow);
3337 /* Figures out whether a flow that missed in 'ofproto', whose details are in
3338 * 'miss' masked by 'wc', is likely to be worth tracking in detail in userspace
3339 * and (usually) installing a datapath flow. The answer is usually "yes" (a
3340 * return value of true). However, for short flows the cost of bookkeeping is
3341 * much higher than the benefits, so when the datapath holds a large number of
3342 * flows we impose some heuristics to decide which flows are likely to be worth
3345 flow_miss_should_make_facet(struct flow_miss *miss, struct flow_wildcards *wc)
3347 struct dpif_backer *backer = miss->ofproto->backer;
3350 if (!backer->governor) {
3353 n_subfacets = hmap_count(&backer->subfacets);
3354 if (n_subfacets * 2 <= flow_eviction_threshold) {
3358 backer->governor = governor_create();
3361 hash = flow_hash_in_wildcards(&miss->flow, wc, 0);
3362 return governor_should_install_flow(backer->governor, hash,
3363 list_size(&miss->packets));
3366 /* Handles 'miss' without creating a facet or subfacet or creating any datapath
3367 * flow. 'miss->flow' must have matched 'rule' and been xlated into 'xout'.
3368 * May add an "execute" operation to 'ops' and increment '*n_ops'. */
3370 handle_flow_miss_without_facet(struct rule_dpif *rule, struct xlate_out *xout,
3371 struct flow_miss *miss,
3372 struct flow_miss_op *ops, size_t *n_ops)
3374 struct ofpbuf *packet;
3376 LIST_FOR_EACH (packet, list_node, &miss->packets) {
3378 COVERAGE_INC(facet_suppress);
3380 handle_flow_miss_common(rule, packet, &miss->flow);
3383 struct xlate_in xin;
3385 xlate_in_init(&xin, miss->ofproto, &miss->flow, rule, 0, packet);
3386 xlate_actions_for_side_effects(&xin);
3389 if (xout->odp_actions.size) {
3390 struct flow_miss_op *op = &ops[*n_ops];
3391 struct dpif_execute *execute = &op->dpif_op.u.execute;
3393 init_flow_miss_execute_op(miss, packet, op);
3394 xlate_out_copy(&op->xout, xout);
3395 execute->actions = op->xout.odp_actions.data;
3396 execute->actions_len = op->xout.odp_actions.size;
3397 op->xout_garbage = true;
3404 /* Handles 'miss', which matches 'facet'. May add any required datapath
3405 * operations to 'ops', incrementing '*n_ops' for each new op.
3407 * All of the packets in 'miss' are considered to have arrived at time 'now'.
3408 * This is really important only for new facets: if we just called time_msec()
3409 * here, then the new subfacet or its packets could look (occasionally) as
3410 * though it was used some time after the facet was used. That can make a
3411 * one-packet flow look like it has a nonzero duration, which looks odd in
3412 * e.g. NetFlow statistics.
3414 * If non-null, 'stats' will be folded into 'facet'. */
3416 handle_flow_miss_with_facet(struct flow_miss *miss, struct facet *facet,
3417 long long int now, struct dpif_flow_stats *stats,
3418 struct flow_miss_op *ops, size_t *n_ops)
3420 struct ofproto_dpif *ofproto = ofproto_dpif_cast(facet->rule->up.ofproto);
3421 enum subfacet_path want_path;
3422 struct subfacet *subfacet;
3423 struct ofpbuf *packet;
3425 subfacet = subfacet_create(facet, miss, now);
3426 want_path = facet->xout.slow ? SF_SLOW_PATH : SF_FAST_PATH;
3428 subfacet_update_stats(subfacet, stats);
3431 LIST_FOR_EACH (packet, list_node, &miss->packets) {
3432 struct flow_miss_op *op = &ops[*n_ops];
3434 handle_flow_miss_common(facet->rule, packet, &miss->flow);
3436 if (want_path != SF_FAST_PATH) {
3437 struct xlate_in xin;
3439 xlate_in_init(&xin, ofproto, &miss->flow, facet->rule, 0, packet);
3440 xlate_actions_for_side_effects(&xin);
3443 if (facet->xout.odp_actions.size) {
3444 struct dpif_execute *execute = &op->dpif_op.u.execute;
3446 init_flow_miss_execute_op(miss, packet, op);
3447 execute->actions = facet->xout.odp_actions.data,
3448 execute->actions_len = facet->xout.odp_actions.size;
3453 if (miss->upcall_type == DPIF_UC_MISS || subfacet->path != want_path) {
3454 struct flow_miss_op *op = &ops[(*n_ops)++];
3455 struct dpif_flow_put *put = &op->dpif_op.u.flow_put;
3457 subfacet->path = want_path;
3459 op->xout_garbage = false;
3460 op->dpif_op.type = DPIF_OP_FLOW_PUT;
3461 put->flags = DPIF_FP_CREATE | DPIF_FP_MODIFY;
3462 put->key = miss->key;
3463 put->key_len = miss->key_len;
3466 if (want_path == SF_FAST_PATH) {
3467 put->actions = facet->xout.odp_actions.data;
3468 put->actions_len = facet->xout.odp_actions.size;
3470 compose_slow_path(ofproto, &miss->flow, facet->xout.slow,
3471 op->slow_stub, sizeof op->slow_stub,
3472 &put->actions, &put->actions_len);
3478 /* Handles flow miss 'miss'. May add any required datapath operations
3479 * to 'ops', incrementing '*n_ops' for each new op. */
3481 handle_flow_miss(struct flow_miss *miss, struct flow_miss_op *ops,
3484 struct ofproto_dpif *ofproto = miss->ofproto;
3485 struct dpif_flow_stats stats__;
3486 struct dpif_flow_stats *stats = &stats__;
3487 struct ofpbuf *packet;
3488 struct facet *facet;
3492 memset(stats, 0, sizeof *stats);
3494 LIST_FOR_EACH (packet, list_node, &miss->packets) {
3495 stats->tcp_flags |= packet_get_tcp_flags(packet, &miss->flow);
3496 stats->n_bytes += packet->size;
3500 facet = facet_lookup_valid(ofproto, &miss->flow);
3502 struct flow_wildcards wc;
3503 struct rule_dpif *rule;
3504 struct xlate_out xout;
3505 struct xlate_in xin;
3507 flow_wildcards_init_catchall(&wc);
3508 rule = rule_dpif_lookup(ofproto, &miss->flow, &wc);
3509 rule_credit_stats(rule, stats);
3511 xlate_in_init(&xin, ofproto, &miss->flow, rule, stats->tcp_flags,
3513 xin.resubmit_stats = stats;
3514 xin.may_learn = true;
3515 xlate_actions(&xin, &xout);
3516 flow_wildcards_or(&xout.wc, &xout.wc, &wc);
3518 /* There does not exist a bijection between 'struct flow' and datapath
3519 * flow keys with fitness ODP_FIT_TO_LITTLE. This breaks a fundamental
3520 * assumption used throughout the facet and subfacet handling code.
3521 * Since we have to handle these misses in userspace anyway, we simply
3522 * skip facet creation, avoiding the problem altogether. */
3523 if (miss->key_fitness == ODP_FIT_TOO_LITTLE
3524 || !flow_miss_should_make_facet(miss, &xout.wc)) {
3525 handle_flow_miss_without_facet(rule, &xout, miss, ops, n_ops);
3529 facet = facet_create(miss, rule, &xout, stats);
3532 handle_flow_miss_with_facet(miss, facet, now, stats, ops, n_ops);
3535 static struct drop_key *
3536 drop_key_lookup(const struct dpif_backer *backer, const struct nlattr *key,
3539 struct drop_key *drop_key;
3541 HMAP_FOR_EACH_WITH_HASH (drop_key, hmap_node, hash_bytes(key, key_len, 0),
3542 &backer->drop_keys) {
3543 if (drop_key->key_len == key_len
3544 && !memcmp(drop_key->key, key, key_len)) {
3552 drop_key_clear(struct dpif_backer *backer)
3554 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 15);
3555 struct drop_key *drop_key, *next;
3557 HMAP_FOR_EACH_SAFE (drop_key, next, hmap_node, &backer->drop_keys) {
3560 error = dpif_flow_del(backer->dpif, drop_key->key, drop_key->key_len,
3562 if (error && !VLOG_DROP_WARN(&rl)) {
3563 struct ds ds = DS_EMPTY_INITIALIZER;
3564 odp_flow_key_format(drop_key->key, drop_key->key_len, &ds);
3565 VLOG_WARN("Failed to delete drop key (%s) (%s)", strerror(error),
3570 hmap_remove(&backer->drop_keys, &drop_key->hmap_node);
3571 free(drop_key->key);
3576 /* Given a datpath, packet, and flow metadata ('backer', 'packet', and 'key'
3577 * respectively), populates 'flow' with the result of odp_flow_key_to_flow().
3578 * Optionally, if nonnull, populates 'fitnessp' with the fitness of 'flow' as
3579 * returned by odp_flow_key_to_flow(). Also, optionally populates 'ofproto'
3580 * with the ofproto_dpif, and 'odp_in_port' with the datapath in_port, that
3581 * 'packet' ingressed.
3583 * If 'ofproto' is nonnull, requires 'flow''s in_port to exist. Otherwise sets
3584 * 'flow''s in_port to OFPP_NONE.
3586 * This function does post-processing on data returned from
3587 * odp_flow_key_to_flow() to help make VLAN splinters transparent to the rest
3588 * of the upcall processing logic. In particular, if the extracted in_port is
3589 * a VLAN splinter port, it replaces flow->in_port by the "real" port, sets
3590 * flow->vlan_tci correctly for the VLAN of the VLAN splinter port, and pushes
3591 * a VLAN header onto 'packet' (if it is nonnull).
3593 * Similarly, this function also includes some logic to help with tunnels. It
3594 * may modify 'flow' as necessary to make the tunneling implementation
3595 * transparent to the upcall processing logic.
3597 * Returns 0 if successful, ENODEV if the parsed flow has no associated ofport,
3598 * or some other positive errno if there are other problems. */
3600 ofproto_receive(const struct dpif_backer *backer, struct ofpbuf *packet,
3601 const struct nlattr *key, size_t key_len,
3602 struct flow *flow, enum odp_key_fitness *fitnessp,
3603 struct ofproto_dpif **ofproto, uint32_t *odp_in_port)
3605 const struct ofport_dpif *port;
3606 enum odp_key_fitness fitness;
3609 fitness = odp_flow_key_to_flow(key, key_len, flow);
3610 if (fitness == ODP_FIT_ERROR) {
3616 *odp_in_port = flow->in_port;
3619 port = (tnl_port_should_receive(flow)
3620 ? ofport_dpif_cast(tnl_port_receive(flow))
3621 : odp_port_to_ofport(backer, flow->in_port));
3622 flow->in_port = port ? port->up.ofp_port : OFPP_NONE;
3627 /* XXX: Since the tunnel module is not scoped per backer, for a tunnel port
3628 * it's theoretically possible that we'll receive an ofport belonging to an
3629 * entirely different datapath. In practice, this can't happen because no
3630 * platforms has two separate datapaths which each support tunneling. */
3631 ovs_assert(ofproto_dpif_cast(port->up.ofproto)->backer == backer);
3633 if (vsp_adjust_flow(ofproto_dpif_cast(port->up.ofproto), flow)) {
3635 /* Make the packet resemble the flow, so that it gets sent to
3636 * an OpenFlow controller properly, so that it looks correct
3637 * for sFlow, and so that flow_extract() will get the correct
3638 * vlan_tci if it is called on 'packet'.
3640 * The allocated space inside 'packet' probably also contains
3641 * 'key', that is, both 'packet' and 'key' are probably part of
3642 * a struct dpif_upcall (see the large comment on that
3643 * structure definition), so pushing data on 'packet' is in
3644 * general not a good idea since it could overwrite 'key' or
3645 * free it as a side effect. However, it's OK in this special
3646 * case because we know that 'packet' is inside a Netlink
3647 * attribute: pushing 4 bytes will just overwrite the 4-byte
3648 * "struct nlattr", which is fine since we don't need that
3649 * header anymore. */
3650 eth_push_vlan(packet, flow->vlan_tci);
3652 /* We can't reproduce 'key' from 'flow'. */
3653 fitness = fitness == ODP_FIT_PERFECT ? ODP_FIT_TOO_MUCH : fitness;
3658 *ofproto = ofproto_dpif_cast(port->up.ofproto);
3663 *fitnessp = fitness;
3669 handle_miss_upcalls(struct dpif_backer *backer, struct dpif_upcall *upcalls,
3672 struct dpif_upcall *upcall;
3673 struct flow_miss *miss;
3674 struct flow_miss misses[FLOW_MISS_MAX_BATCH];
3675 struct flow_miss_op flow_miss_ops[FLOW_MISS_MAX_BATCH * 2];
3676 struct dpif_op *dpif_ops[FLOW_MISS_MAX_BATCH * 2];
3686 /* Construct the to-do list.
3688 * This just amounts to extracting the flow from each packet and sticking
3689 * the packets that have the same flow in the same "flow_miss" structure so
3690 * that we can process them together. */
3693 for (upcall = upcalls; upcall < &upcalls[n_upcalls]; upcall++) {
3694 struct flow_miss *miss = &misses[n_misses];
3695 struct flow_miss *existing_miss;
3696 struct ofproto_dpif *ofproto;
3697 uint32_t odp_in_port;
3702 error = ofproto_receive(backer, upcall->packet, upcall->key,
3703 upcall->key_len, &flow, &miss->key_fitness,
3704 &ofproto, &odp_in_port);
3705 if (error == ENODEV) {
3706 struct drop_key *drop_key;
3708 /* Received packet on datapath port for which we couldn't
3709 * associate an ofproto. This can happen if a port is removed
3710 * while traffic is being received. Print a rate-limited message
3711 * in case it happens frequently. Install a drop flow so
3712 * that future packets of the flow are inexpensively dropped
3714 VLOG_INFO_RL(&rl, "received packet on unassociated datapath port "
3715 "%"PRIu32, odp_in_port);
3717 drop_key = drop_key_lookup(backer, upcall->key, upcall->key_len);
3719 drop_key = xmalloc(sizeof *drop_key);
3720 drop_key->key = xmemdup(upcall->key, upcall->key_len);
3721 drop_key->key_len = upcall->key_len;
3723 hmap_insert(&backer->drop_keys, &drop_key->hmap_node,
3724 hash_bytes(drop_key->key, drop_key->key_len, 0));
3725 dpif_flow_put(backer->dpif, DPIF_FP_CREATE | DPIF_FP_MODIFY,
3726 drop_key->key, drop_key->key_len,
3727 NULL, 0, NULL, 0, NULL);
3735 ofproto->n_missed++;
3736 flow_extract(upcall->packet, flow.skb_priority, flow.skb_mark,
3737 &flow.tunnel, flow.in_port, &miss->flow);
3739 /* Add other packets to a to-do list. */
3740 hash = flow_hash(&miss->flow, 0);
3741 existing_miss = flow_miss_find(&todo, ofproto, &miss->flow, hash);
3742 if (!existing_miss) {
3743 hmap_insert(&todo, &miss->hmap_node, hash);
3744 miss->ofproto = ofproto;
3745 miss->key = upcall->key;
3746 miss->key_len = upcall->key_len;
3747 miss->upcall_type = upcall->type;
3748 list_init(&miss->packets);
3752 miss = existing_miss;
3754 list_push_back(&miss->packets, &upcall->packet->list_node);
3757 /* Process each element in the to-do list, constructing the set of
3758 * operations to batch. */
3760 HMAP_FOR_EACH (miss, hmap_node, &todo) {
3761 handle_flow_miss(miss, flow_miss_ops, &n_ops);
3763 ovs_assert(n_ops <= ARRAY_SIZE(flow_miss_ops));
3765 /* Execute batch. */
3766 for (i = 0; i < n_ops; i++) {
3767 dpif_ops[i] = &flow_miss_ops[i].dpif_op;
3769 dpif_operate(backer->dpif, dpif_ops, n_ops);
3772 for (i = 0; i < n_ops; i++) {
3773 if (flow_miss_ops[i].xout_garbage) {
3774 xlate_out_uninit(&flow_miss_ops[i].xout);
3777 hmap_destroy(&todo);
3780 static enum { SFLOW_UPCALL, MISS_UPCALL, BAD_UPCALL, FLOW_SAMPLE_UPCALL,
3782 classify_upcall(const struct dpif_upcall *upcall)
3784 size_t userdata_len;
3785 union user_action_cookie cookie;
3787 /* First look at the upcall type. */
3788 switch (upcall->type) {
3789 case DPIF_UC_ACTION:
3795 case DPIF_N_UC_TYPES:
3797 VLOG_WARN_RL(&rl, "upcall has unexpected type %"PRIu32, upcall->type);
3801 /* "action" upcalls need a closer look. */
3802 if (!upcall->userdata) {
3803 VLOG_WARN_RL(&rl, "action upcall missing cookie");
3806 userdata_len = nl_attr_get_size(upcall->userdata);
3807 if (userdata_len < sizeof cookie.type
3808 || userdata_len > sizeof cookie) {
3809 VLOG_WARN_RL(&rl, "action upcall cookie has unexpected size %zu",
3813 memset(&cookie, 0, sizeof cookie);
3814 memcpy(&cookie, nl_attr_get(upcall->userdata), userdata_len);
3815 if (userdata_len == sizeof cookie.sflow
3816 && cookie.type == USER_ACTION_COOKIE_SFLOW) {
3817 return SFLOW_UPCALL;
3818 } else if (userdata_len == sizeof cookie.slow_path
3819 && cookie.type == USER_ACTION_COOKIE_SLOW_PATH) {
3821 } else if (userdata_len == sizeof cookie.flow_sample
3822 && cookie.type == USER_ACTION_COOKIE_FLOW_SAMPLE) {
3823 return FLOW_SAMPLE_UPCALL;
3824 } else if (userdata_len == sizeof cookie.ipfix
3825 && cookie.type == USER_ACTION_COOKIE_IPFIX) {
3826 return IPFIX_UPCALL;
3828 VLOG_WARN_RL(&rl, "invalid user cookie of type %"PRIu16
3829 " and size %zu", cookie.type, userdata_len);
3835 handle_sflow_upcall(struct dpif_backer *backer,
3836 const struct dpif_upcall *upcall)
3838 struct ofproto_dpif *ofproto;
3839 union user_action_cookie cookie;
3841 uint32_t odp_in_port;
3843 if (ofproto_receive(backer, upcall->packet, upcall->key, upcall->key_len,
3844 &flow, NULL, &ofproto, &odp_in_port)
3845 || !ofproto->sflow) {
3849 memset(&cookie, 0, sizeof cookie);
3850 memcpy(&cookie, nl_attr_get(upcall->userdata), sizeof cookie.sflow);
3851 dpif_sflow_received(ofproto->sflow, upcall->packet, &flow,
3852 odp_in_port, &cookie);
3856 handle_flow_sample_upcall(struct dpif_backer *backer,
3857 const struct dpif_upcall *upcall)
3859 struct ofproto_dpif *ofproto;
3860 union user_action_cookie cookie;
3863 if (ofproto_receive(backer, upcall->packet, upcall->key, upcall->key_len,
3864 &flow, NULL, &ofproto, NULL)
3865 || !ofproto->ipfix) {
3869 memset(&cookie, 0, sizeof cookie);
3870 memcpy(&cookie, nl_attr_get(upcall->userdata), sizeof cookie.flow_sample);
3872 /* The flow reflects exactly the contents of the packet. Sample
3873 * the packet using it. */
3874 dpif_ipfix_flow_sample(ofproto->ipfix, upcall->packet, &flow,
3875 cookie.flow_sample.collector_set_id,
3876 cookie.flow_sample.probability,
3877 cookie.flow_sample.obs_domain_id,
3878 cookie.flow_sample.obs_point_id);
3882 handle_ipfix_upcall(struct dpif_backer *backer,
3883 const struct dpif_upcall *upcall)
3885 struct ofproto_dpif *ofproto;
3888 if (ofproto_receive(backer, upcall->packet, upcall->key, upcall->key_len,
3889 &flow, NULL, &ofproto, NULL)
3890 || !ofproto->ipfix) {
3894 /* The flow reflects exactly the contents of the packet. Sample
3895 * the packet using it. */
3896 dpif_ipfix_bridge_sample(ofproto->ipfix, upcall->packet, &flow);
3900 handle_upcalls(struct dpif_backer *backer, unsigned int max_batch)
3902 struct dpif_upcall misses[FLOW_MISS_MAX_BATCH];
3903 struct ofpbuf miss_bufs[FLOW_MISS_MAX_BATCH];
3904 uint64_t miss_buf_stubs[FLOW_MISS_MAX_BATCH][4096 / 8];
3909 ovs_assert(max_batch <= FLOW_MISS_MAX_BATCH);
3912 for (n_processed = 0; n_processed < max_batch; n_processed++) {
3913 struct dpif_upcall *upcall = &misses[n_misses];
3914 struct ofpbuf *buf = &miss_bufs[n_misses];
3917 ofpbuf_use_stub(buf, miss_buf_stubs[n_misses],
3918 sizeof miss_buf_stubs[n_misses]);
3919 error = dpif_recv(backer->dpif, upcall, buf);
3925 switch (classify_upcall(upcall)) {
3927 /* Handle it later. */
3932 handle_sflow_upcall(backer, upcall);
3936 case FLOW_SAMPLE_UPCALL:
3937 handle_flow_sample_upcall(backer, upcall);
3942 handle_ipfix_upcall(backer, upcall);
3952 /* Handle deferred MISS_UPCALL processing. */
3953 handle_miss_upcalls(backer, misses, n_misses);
3954 for (i = 0; i < n_misses; i++) {
3955 ofpbuf_uninit(&miss_bufs[i]);
3961 /* Flow expiration. */
3963 static int subfacet_max_idle(const struct dpif_backer *);
3964 static void update_stats(struct dpif_backer *);
3965 static void rule_expire(struct rule_dpif *);
3966 static void expire_subfacets(struct dpif_backer *, int dp_max_idle);
3968 /* This function is called periodically by run(). Its job is to collect
3969 * updates for the flows that have been installed into the datapath, most
3970 * importantly when they last were used, and then use that information to
3971 * expire flows that have not been used recently.
3973 * Returns the number of milliseconds after which it should be called again. */
3975 expire(struct dpif_backer *backer)
3977 struct ofproto_dpif *ofproto;
3981 /* Periodically clear out the drop keys in an effort to keep them
3982 * relatively few. */
3983 drop_key_clear(backer);
3985 /* Update stats for each flow in the backer. */
3986 update_stats(backer);
3988 n_subfacets = hmap_count(&backer->subfacets);
3990 struct subfacet *subfacet;
3991 long long int total, now;
3995 HMAP_FOR_EACH (subfacet, hmap_node, &backer->subfacets) {
3996 total += now - subfacet->created;
3998 backer->avg_subfacet_life += total / n_subfacets;
4000 backer->avg_subfacet_life /= 2;
4002 backer->avg_n_subfacet += n_subfacets;
4003 backer->avg_n_subfacet /= 2;
4005 backer->max_n_subfacet = MAX(backer->max_n_subfacet, n_subfacets);
4007 max_idle = subfacet_max_idle(backer);
4008 expire_subfacets(backer, max_idle);
4010 HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
4011 struct rule *rule, *next_rule;
4013 if (ofproto->backer != backer) {
4017 /* Expire OpenFlow flows whose idle_timeout or hard_timeout
4019 LIST_FOR_EACH_SAFE (rule, next_rule, expirable,
4020 &ofproto->up.expirable) {
4021 rule_expire(rule_dpif_cast(rule));
4024 /* All outstanding data in existing flows has been accounted, so it's a
4025 * good time to do bond rebalancing. */
4026 if (ofproto->has_bonded_bundles) {
4027 struct ofbundle *bundle;
4029 HMAP_FOR_EACH (bundle, hmap_node, &ofproto->bundles) {
4031 bond_rebalance(bundle->bond, &backer->revalidate_set);
4037 return MIN(max_idle, 1000);
4040 /* Updates flow table statistics given that the datapath just reported 'stats'
4041 * as 'subfacet''s statistics. */
4043 update_subfacet_stats(struct subfacet *subfacet,
4044 const struct dpif_flow_stats *stats)
4046 struct facet *facet = subfacet->facet;
4047 struct ofproto_dpif *ofproto = ofproto_dpif_cast(facet->rule->up.ofproto);
4048 struct dpif_flow_stats diff;
4050 diff.tcp_flags = stats->tcp_flags;
4051 diff.used = stats->used;
4053 if (stats->n_packets >= subfacet->dp_packet_count) {
4054 diff.n_packets = stats->n_packets - subfacet->dp_packet_count;
4056 VLOG_WARN_RL(&rl, "unexpected packet count from the datapath");
4060 if (stats->n_bytes >= subfacet->dp_byte_count) {
4061 diff.n_bytes = stats->n_bytes - subfacet->dp_byte_count;
4063 VLOG_WARN_RL(&rl, "unexpected byte count from datapath");
4067 ofproto->n_hit += diff.n_packets;
4068 subfacet->dp_packet_count = stats->n_packets;
4069 subfacet->dp_byte_count = stats->n_bytes;
4070 subfacet_update_stats(subfacet, &diff);
4072 if (facet->accounted_bytes < facet->byte_count) {
4074 facet_account(facet);
4075 facet->accounted_bytes = facet->byte_count;
4079 /* 'key' with length 'key_len' bytes is a flow in 'dpif' that we know nothing
4080 * about, or a flow that shouldn't be installed but was anyway. Delete it. */
4082 delete_unexpected_flow(struct dpif_backer *backer,
4083 const struct nlattr *key, size_t key_len)
4085 if (!VLOG_DROP_WARN(&rl)) {
4089 odp_flow_key_format(key, key_len, &s);
4090 VLOG_WARN("unexpected flow: %s", ds_cstr(&s));
4094 COVERAGE_INC(facet_unexpected);
4095 dpif_flow_del(backer->dpif, key, key_len, NULL);
4098 /* Update 'packet_count', 'byte_count', and 'used' members of installed facets.
4100 * This function also pushes statistics updates to rules which each facet
4101 * resubmits into. Generally these statistics will be accurate. However, if a
4102 * facet changes the rule it resubmits into at some time in between
4103 * update_stats() runs, it is possible that statistics accrued to the
4104 * old rule will be incorrectly attributed to the new rule. This could be
4105 * avoided by calling update_stats() whenever rules are created or
4106 * deleted. However, the performance impact of making so many calls to the
4107 * datapath do not justify the benefit of having perfectly accurate statistics.
4109 * In addition, this function maintains per ofproto flow hit counts. The patch
4110 * port is not treated specially. e.g. A packet ingress from br0 patched into
4111 * br1 will increase the hit count of br0 by 1, however, does not affect
4112 * the hit or miss counts of br1.
4115 update_stats(struct dpif_backer *backer)
4117 const struct dpif_flow_stats *stats;
4118 struct dpif_flow_dump dump;
4119 const struct nlattr *key;
4122 dpif_flow_dump_start(&dump, backer->dpif);
4123 while (dpif_flow_dump_next(&dump, &key, &key_len,
4124 NULL, NULL, NULL, NULL, &stats)) {
4125 struct subfacet *subfacet;
4128 key_hash = odp_flow_key_hash(key, key_len);
4129 subfacet = subfacet_find(backer, key, key_len, key_hash);
4130 switch (subfacet ? subfacet->path : SF_NOT_INSTALLED) {
4132 update_subfacet_stats(subfacet, stats);
4136 /* Stats are updated per-packet. */
4139 case SF_NOT_INSTALLED:
4141 delete_unexpected_flow(backer, key, key_len);
4146 dpif_flow_dump_done(&dump);
4148 update_moving_averages(backer);
4151 /* Calculates and returns the number of milliseconds of idle time after which
4152 * subfacets should expire from the datapath. When a subfacet expires, we fold
4153 * its statistics into its facet, and when a facet's last subfacet expires, we
4154 * fold its statistic into its rule. */
4156 subfacet_max_idle(const struct dpif_backer *backer)
4159 * Idle time histogram.
4161 * Most of the time a switch has a relatively small number of subfacets.
4162 * When this is the case we might as well keep statistics for all of them
4163 * in userspace and to cache them in the kernel datapath for performance as
4166 * As the number of subfacets increases, the memory required to maintain
4167 * statistics about them in userspace and in the kernel becomes
4168 * significant. However, with a large number of subfacets it is likely
4169 * that only a few of them are "heavy hitters" that consume a large amount
4170 * of bandwidth. At this point, only heavy hitters are worth caching in
4171 * the kernel and maintaining in userspaces; other subfacets we can
4174 * The technique used to compute the idle time is to build a histogram with
4175 * N_BUCKETS buckets whose width is BUCKET_WIDTH msecs each. Each subfacet
4176 * that is installed in the kernel gets dropped in the appropriate bucket.
4177 * After the histogram has been built, we compute the cutoff so that only
4178 * the most-recently-used 1% of subfacets (but at least
4179 * flow_eviction_threshold flows) are kept cached. At least
4180 * the most-recently-used bucket of subfacets is kept, so actually an
4181 * arbitrary number of subfacets can be kept in any given expiration run
4182 * (though the next run will delete most of those unless they receive
4185 * This requires a second pass through the subfacets, in addition to the
4186 * pass made by update_stats(), because the former function never looks at
4187 * uninstallable subfacets.
4189 enum { BUCKET_WIDTH = ROUND_UP(100, TIME_UPDATE_INTERVAL) };
4190 enum { N_BUCKETS = 5000 / BUCKET_WIDTH };
4191 int buckets[N_BUCKETS] = { 0 };
4192 int total, subtotal, bucket;
4193 struct subfacet *subfacet;
4197 total = hmap_count(&backer->subfacets);
4198 if (total <= flow_eviction_threshold) {
4199 return N_BUCKETS * BUCKET_WIDTH;
4202 /* Build histogram. */
4204 HMAP_FOR_EACH (subfacet, hmap_node, &backer->subfacets) {
4205 long long int idle = now - subfacet->used;
4206 int bucket = (idle <= 0 ? 0
4207 : idle >= BUCKET_WIDTH * N_BUCKETS ? N_BUCKETS - 1
4208 : (unsigned int) idle / BUCKET_WIDTH);
4212 /* Find the first bucket whose flows should be expired. */
4213 subtotal = bucket = 0;
4215 subtotal += buckets[bucket++];
4216 } while (bucket < N_BUCKETS &&
4217 subtotal < MAX(flow_eviction_threshold, total / 100));
4219 if (VLOG_IS_DBG_ENABLED()) {
4223 ds_put_cstr(&s, "keep");
4224 for (i = 0; i < N_BUCKETS; i++) {
4226 ds_put_cstr(&s, ", drop");
4229 ds_put_format(&s, " %d:%d", i * BUCKET_WIDTH, buckets[i]);
4232 VLOG_INFO("%s (msec:count)", ds_cstr(&s));
4236 return bucket * BUCKET_WIDTH;
4240 expire_subfacets(struct dpif_backer *backer, int dp_max_idle)
4242 /* Cutoff time for most flows. */
4243 long long int normal_cutoff = time_msec() - dp_max_idle;
4245 /* We really want to keep flows for special protocols around, so use a more
4246 * conservative cutoff. */
4247 long long int special_cutoff = time_msec() - 10000;
4249 struct subfacet *subfacet, *next_subfacet;
4250 struct subfacet *batch[SUBFACET_DESTROY_MAX_BATCH];
4254 HMAP_FOR_EACH_SAFE (subfacet, next_subfacet, hmap_node,
4255 &backer->subfacets) {
4256 long long int cutoff;
4258 cutoff = (subfacet->facet->xout.slow & (SLOW_CFM | SLOW_BFD | SLOW_LACP
4262 if (subfacet->used < cutoff) {
4263 if (subfacet->path != SF_NOT_INSTALLED) {
4264 batch[n_batch++] = subfacet;
4265 if (n_batch >= SUBFACET_DESTROY_MAX_BATCH) {
4266 subfacet_destroy_batch(backer, batch, n_batch);
4270 subfacet_destroy(subfacet);
4276 subfacet_destroy_batch(backer, batch, n_batch);
4280 /* If 'rule' is an OpenFlow rule, that has expired according to OpenFlow rules,
4281 * then delete it entirely. */
4283 rule_expire(struct rule_dpif *rule)
4285 struct facet *facet, *next_facet;
4289 if (rule->up.pending) {
4290 /* We'll have to expire it later. */
4294 /* Has 'rule' expired? */
4296 if (rule->up.hard_timeout
4297 && now > rule->up.modified + rule->up.hard_timeout * 1000) {
4298 reason = OFPRR_HARD_TIMEOUT;
4299 } else if (rule->up.idle_timeout
4300 && now > rule->up.used + rule->up.idle_timeout * 1000) {
4301 reason = OFPRR_IDLE_TIMEOUT;
4306 COVERAGE_INC(ofproto_dpif_expired);
4308 /* Update stats. (This is a no-op if the rule expired due to an idle
4309 * timeout, because that only happens when the rule has no facets left.) */
4310 LIST_FOR_EACH_SAFE (facet, next_facet, list_node, &rule->facets) {
4311 facet_remove(facet);
4314 /* Get rid of the rule. */
4315 ofproto_rule_expire(&rule->up, reason);
4320 /* Creates and returns a new facet based on 'miss'.
4322 * The caller must already have determined that no facet with an identical
4323 * 'miss->flow' exists in 'miss->ofproto'.
4325 * 'rule' and 'xout' must have been created based on 'miss'.
4327 * 'facet'' statistics are initialized based on 'stats'.
4329 * The facet will initially have no subfacets. The caller should create (at
4330 * least) one subfacet with subfacet_create(). */
4331 static struct facet *
4332 facet_create(const struct flow_miss *miss, struct rule_dpif *rule,
4333 struct xlate_out *xout, struct dpif_flow_stats *stats)
4335 struct ofproto_dpif *ofproto = miss->ofproto;
4336 struct facet *facet;
4339 facet = xzalloc(sizeof *facet);
4340 facet->packet_count = facet->prev_packet_count = stats->n_packets;
4341 facet->byte_count = facet->prev_byte_count = stats->n_bytes;
4342 facet->tcp_flags = stats->tcp_flags;
4343 facet->used = stats->used;
4344 facet->flow = miss->flow;
4345 facet->learn_rl = time_msec() + 500;
4348 list_push_back(&facet->rule->facets, &facet->list_node);
4349 list_init(&facet->subfacets);
4350 netflow_flow_init(&facet->nf_flow);
4351 netflow_flow_update_time(ofproto->netflow, &facet->nf_flow, facet->used);
4353 xlate_out_copy(&facet->xout, xout);
4355 match_init(&match, &facet->flow, &facet->xout.wc);
4356 cls_rule_init(&facet->cr, &match, OFP_DEFAULT_PRIORITY);
4357 classifier_insert(&ofproto->facets, &facet->cr);
4359 facet->nf_flow.output_iface = facet->xout.nf_output_iface;
4365 facet_free(struct facet *facet)
4368 xlate_out_uninit(&facet->xout);
4373 /* Executes, within 'ofproto', the 'n_actions' actions in 'actions' on
4374 * 'packet', which arrived on 'in_port'. */
4376 execute_odp_actions(struct ofproto_dpif *ofproto, const struct flow *flow,
4377 const struct nlattr *odp_actions, size_t actions_len,
4378 struct ofpbuf *packet)
4380 struct odputil_keybuf keybuf;
4384 ofpbuf_use_stack(&key, &keybuf, sizeof keybuf);
4385 odp_flow_key_from_flow(&key, flow,
4386 ofp_port_to_odp_port(ofproto, flow->in_port));
4388 error = dpif_execute(ofproto->backer->dpif, key.data, key.size,
4389 odp_actions, actions_len, packet);
4393 /* Remove 'facet' from its ofproto and free up the associated memory:
4395 * - If 'facet' was installed in the datapath, uninstalls it and updates its
4396 * rule's statistics, via subfacet_uninstall().
4398 * - Removes 'facet' from its rule and from ofproto->facets.
4401 facet_remove(struct facet *facet)
4403 struct ofproto_dpif *ofproto = ofproto_dpif_cast(facet->rule->up.ofproto);
4404 struct subfacet *subfacet, *next_subfacet;
4406 ovs_assert(!list_is_empty(&facet->subfacets));
4408 /* First uninstall all of the subfacets to get final statistics. */
4409 LIST_FOR_EACH (subfacet, list_node, &facet->subfacets) {
4410 subfacet_uninstall(subfacet);
4413 /* Flush the final stats to the rule.
4415 * This might require us to have at least one subfacet around so that we
4416 * can use its actions for accounting in facet_account(), which is why we
4417 * have uninstalled but not yet destroyed the subfacets. */
4418 facet_flush_stats(facet);
4420 /* Now we're really all done so destroy everything. */
4421 LIST_FOR_EACH_SAFE (subfacet, next_subfacet, list_node,
4422 &facet->subfacets) {
4423 subfacet_destroy__(subfacet);
4425 classifier_remove(&ofproto->facets, &facet->cr);
4426 cls_rule_destroy(&facet->cr);
4427 list_remove(&facet->list_node);
4431 /* Feed information from 'facet' back into the learning table to keep it in
4432 * sync with what is actually flowing through the datapath. */
4434 facet_learn(struct facet *facet)
4436 long long int now = time_msec();
4438 if (!facet->xout.has_fin_timeout && now < facet->learn_rl) {
4442 facet->learn_rl = now + 500;
4444 if (!facet->xout.has_learn
4445 && !facet->xout.has_normal
4446 && (!facet->xout.has_fin_timeout
4447 || !(facet->tcp_flags & (TCP_FIN | TCP_RST)))) {
4451 facet_push_stats(facet, true);
4455 facet_account(struct facet *facet)
4457 struct ofproto_dpif *ofproto = ofproto_dpif_cast(facet->rule->up.ofproto);
4458 const struct nlattr *a;
4463 if (!facet->xout.has_normal || !ofproto->has_bonded_bundles) {
4466 n_bytes = facet->byte_count - facet->accounted_bytes;
4468 /* This loop feeds byte counters to bond_account() for rebalancing to use
4469 * as a basis. We also need to track the actual VLAN on which the packet
4470 * is going to be sent to ensure that it matches the one passed to
4471 * bond_choose_output_slave(). (Otherwise, we will account to the wrong
4474 * We use the actions from an arbitrary subfacet because they should all
4475 * be equally valid for our purpose. */
4476 vlan_tci = facet->flow.vlan_tci;
4477 NL_ATTR_FOR_EACH_UNSAFE (a, left, facet->xout.odp_actions.data,
4478 facet->xout.odp_actions.size) {
4479 const struct ovs_action_push_vlan *vlan;
4480 struct ofport_dpif *port;
4482 switch (nl_attr_type(a)) {
4483 case OVS_ACTION_ATTR_OUTPUT:
4484 port = get_odp_port(ofproto, nl_attr_get_u32(a));
4485 if (port && port->bundle && port->bundle->bond) {
4486 bond_account(port->bundle->bond, &facet->flow,
4487 vlan_tci_to_vid(vlan_tci), n_bytes);
4491 case OVS_ACTION_ATTR_POP_VLAN:
4492 vlan_tci = htons(0);
4495 case OVS_ACTION_ATTR_PUSH_VLAN:
4496 vlan = nl_attr_get(a);
4497 vlan_tci = vlan->vlan_tci;
4503 /* Returns true if the only action for 'facet' is to send to the controller.
4504 * (We don't report NetFlow expiration messages for such facets because they
4505 * are just part of the control logic for the network, not real traffic). */
4507 facet_is_controller_flow(struct facet *facet)
4510 const struct rule *rule = &facet->rule->up;
4511 const struct ofpact *ofpacts = rule->ofpacts;
4512 size_t ofpacts_len = rule->ofpacts_len;
4514 if (ofpacts_len > 0 &&
4515 ofpacts->type == OFPACT_CONTROLLER &&
4516 ofpact_next(ofpacts) >= ofpact_end(ofpacts, ofpacts_len)) {
4523 /* Folds all of 'facet''s statistics into its rule. Also updates the
4524 * accounting ofhook and emits a NetFlow expiration if appropriate. All of
4525 * 'facet''s statistics in the datapath should have been zeroed and folded into
4526 * its packet and byte counts before this function is called. */
4528 facet_flush_stats(struct facet *facet)
4530 struct ofproto_dpif *ofproto = ofproto_dpif_cast(facet->rule->up.ofproto);
4531 struct subfacet *subfacet;
4533 LIST_FOR_EACH (subfacet, list_node, &facet->subfacets) {
4534 ovs_assert(!subfacet->dp_byte_count);
4535 ovs_assert(!subfacet->dp_packet_count);
4538 facet_push_stats(facet, false);
4539 if (facet->accounted_bytes < facet->byte_count) {
4540 facet_account(facet);
4541 facet->accounted_bytes = facet->byte_count;
4544 if (ofproto->netflow && !facet_is_controller_flow(facet)) {
4545 struct ofexpired expired;
4546 expired.flow = facet->flow;
4547 expired.packet_count = facet->packet_count;
4548 expired.byte_count = facet->byte_count;
4549 expired.used = facet->used;
4550 netflow_expire(ofproto->netflow, &facet->nf_flow, &expired);
4553 /* Reset counters to prevent double counting if 'facet' ever gets
4555 facet_reset_counters(facet);
4557 netflow_flow_clear(&facet->nf_flow);
4558 facet->tcp_flags = 0;
4561 /* Searches 'ofproto''s table of facets for one which would be responsible for
4562 * 'flow'. Returns it if found, otherwise a null pointer.
4564 * The returned facet might need revalidation; use facet_lookup_valid()
4565 * instead if that is important. */
4566 static struct facet *
4567 facet_find(struct ofproto_dpif *ofproto, const struct flow *flow)
4569 struct cls_rule *cr = classifier_lookup(&ofproto->facets, flow, NULL);
4570 return cr ? CONTAINER_OF(cr, struct facet, cr) : NULL;
4573 /* Searches 'ofproto''s table of facets for one capable that covers
4574 * 'flow'. Returns it if found, otherwise a null pointer.
4576 * The returned facet is guaranteed to be valid. */
4577 static struct facet *
4578 facet_lookup_valid(struct ofproto_dpif *ofproto, const struct flow *flow)
4580 struct facet *facet;
4582 facet = facet_find(ofproto, flow);
4584 && (ofproto->backer->need_revalidate
4585 || tag_set_intersects(&ofproto->backer->revalidate_set,
4587 && !facet_revalidate(facet)) {
4595 facet_check_consistency(struct facet *facet)
4597 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 15);
4599 struct ofproto_dpif *ofproto = ofproto_dpif_cast(facet->rule->up.ofproto);
4601 struct xlate_out xout;
4602 struct xlate_in xin;
4604 struct rule_dpif *rule;
4607 /* Check the rule for consistency. */
4608 rule = rule_dpif_lookup(ofproto, &facet->flow, NULL);
4609 if (rule != facet->rule) {
4610 if (!VLOG_DROP_WARN(&rl)) {
4611 struct ds s = DS_EMPTY_INITIALIZER;
4613 flow_format(&s, &facet->flow);
4614 ds_put_format(&s, ": facet associated with wrong rule (was "
4615 "table=%"PRIu8",", facet->rule->up.table_id);
4616 cls_rule_format(&facet->rule->up.cr, &s);
4617 ds_put_format(&s, ") (should have been table=%"PRIu8",",
4619 cls_rule_format(&rule->up.cr, &s);
4620 ds_put_char(&s, ')');
4622 VLOG_WARN("%s", ds_cstr(&s));
4628 /* Check the datapath actions for consistency. */
4629 xlate_in_init(&xin, ofproto, &facet->flow, rule, 0, NULL);
4630 xlate_actions(&xin, &xout);
4632 ok = ofpbuf_equal(&facet->xout.odp_actions, &xout.odp_actions)
4633 && facet->xout.slow == xout.slow;
4634 if (!ok && !VLOG_DROP_WARN(&rl)) {
4635 struct ds s = DS_EMPTY_INITIALIZER;
4637 flow_format(&s, &facet->flow);
4638 ds_put_cstr(&s, ": inconsistency in facet");
4640 if (!ofpbuf_equal(&facet->xout.odp_actions, &xout.odp_actions)) {
4641 ds_put_cstr(&s, " (actions were: ");
4642 format_odp_actions(&s, facet->xout.odp_actions.data,
4643 facet->xout.odp_actions.size);
4644 ds_put_cstr(&s, ") (correct actions: ");
4645 format_odp_actions(&s, xout.odp_actions.data,
4646 xout.odp_actions.size);
4647 ds_put_char(&s, ')');
4650 if (facet->xout.slow != xout.slow) {
4651 ds_put_format(&s, " slow path incorrect. should be %d", xout.slow);
4654 VLOG_WARN("%s", ds_cstr(&s));
4657 xlate_out_uninit(&xout);
4662 /* Re-searches the classifier for 'facet':
4664 * - If the rule found is different from 'facet''s current rule, moves
4665 * 'facet' to the new rule and recompiles its actions.
4667 * - If the rule found is the same as 'facet''s current rule, leaves 'facet'
4668 * where it is and recompiles its actions anyway.
4670 * - If any of 'facet''s subfacets correspond to a new flow according to
4671 * ofproto_receive(), 'facet' is removed.
4673 * Returns true if 'facet' is still valid. False if 'facet' was removed. */
4675 facet_revalidate(struct facet *facet)
4677 struct ofproto_dpif *ofproto = ofproto_dpif_cast(facet->rule->up.ofproto);
4678 struct rule_dpif *new_rule;
4679 struct subfacet *subfacet;
4680 struct flow_wildcards wc;
4681 struct xlate_out xout;
4682 struct xlate_in xin;
4684 COVERAGE_INC(facet_revalidate);
4686 /* Check that child subfacets still correspond to this facet. Tunnel
4687 * configuration changes could cause a subfacet's OpenFlow in_port to
4689 LIST_FOR_EACH (subfacet, list_node, &facet->subfacets) {
4690 struct ofproto_dpif *recv_ofproto;
4691 struct flow recv_flow;
4694 error = ofproto_receive(ofproto->backer, NULL, subfacet->key,
4695 subfacet->key_len, &recv_flow, NULL,
4696 &recv_ofproto, NULL);
4698 || recv_ofproto != ofproto
4699 || facet != facet_find(ofproto, &recv_flow)) {
4700 facet_remove(facet);
4705 flow_wildcards_init_catchall(&wc);
4706 new_rule = rule_dpif_lookup(ofproto, &facet->flow, &wc);
4708 /* Calculate new datapath actions.
4710 * We do not modify any 'facet' state yet, because we might need to, e.g.,
4711 * emit a NetFlow expiration and, if so, we need to have the old state
4712 * around to properly compose it. */
4713 xlate_in_init(&xin, ofproto, &facet->flow, new_rule, 0, NULL);
4714 xlate_actions(&xin, &xout);
4715 flow_wildcards_or(&xout.wc, &xout.wc, &wc);
4717 /* A facet's slow path reason should only change under dramatic
4718 * circumstances. Rather than try to update everything, it's simpler to
4719 * remove the facet and start over.
4721 * More importantly, if a facet's wildcards change, it will be relatively
4722 * difficult to figure out if its subfacets still belong to it, and if not
4723 * which facet they may belong to. Again, to avoid the complexity, we
4724 * simply give up instead. */
4725 if (facet->xout.slow != xout.slow
4726 || memcmp(&facet->xout.wc, &xout.wc, sizeof xout.wc)) {
4727 facet_remove(facet);
4728 xlate_out_uninit(&xout);
4732 if (!ofpbuf_equal(&facet->xout.odp_actions, &xout.odp_actions)) {
4733 LIST_FOR_EACH(subfacet, list_node, &facet->subfacets) {
4734 if (subfacet->path == SF_FAST_PATH) {
4735 struct dpif_flow_stats stats;
4737 subfacet_install(subfacet, &xout.odp_actions, &stats);
4738 subfacet_update_stats(subfacet, &stats);
4742 facet_flush_stats(facet);
4744 ofpbuf_clear(&facet->xout.odp_actions);
4745 ofpbuf_put(&facet->xout.odp_actions, xout.odp_actions.data,
4746 xout.odp_actions.size);
4749 /* Update 'facet' now that we've taken care of all the old state. */
4750 facet->xout.tags = xout.tags;
4751 facet->xout.slow = xout.slow;
4752 facet->xout.has_learn = xout.has_learn;
4753 facet->xout.has_normal = xout.has_normal;
4754 facet->xout.has_fin_timeout = xout.has_fin_timeout;
4755 facet->xout.nf_output_iface = xout.nf_output_iface;
4756 facet->xout.mirrors = xout.mirrors;
4757 facet->nf_flow.output_iface = facet->xout.nf_output_iface;
4759 if (facet->rule != new_rule) {
4760 COVERAGE_INC(facet_changed_rule);
4761 list_remove(&facet->list_node);
4762 list_push_back(&new_rule->facets, &facet->list_node);
4763 facet->rule = new_rule;
4764 facet->used = new_rule->up.created;
4765 facet->prev_used = facet->used;
4768 xlate_out_uninit(&xout);
4773 facet_reset_counters(struct facet *facet)
4775 facet->packet_count = 0;
4776 facet->byte_count = 0;
4777 facet->prev_packet_count = 0;
4778 facet->prev_byte_count = 0;
4779 facet->accounted_bytes = 0;
4783 facet_push_stats(struct facet *facet, bool may_learn)
4785 struct dpif_flow_stats stats;
4787 ovs_assert(facet->packet_count >= facet->prev_packet_count);
4788 ovs_assert(facet->byte_count >= facet->prev_byte_count);
4789 ovs_assert(facet->used >= facet->prev_used);
4791 stats.n_packets = facet->packet_count - facet->prev_packet_count;
4792 stats.n_bytes = facet->byte_count - facet->prev_byte_count;
4793 stats.used = facet->used;
4794 stats.tcp_flags = facet->tcp_flags;
4796 if (may_learn || stats.n_packets || facet->used > facet->prev_used) {
4797 struct ofproto_dpif *ofproto =
4798 ofproto_dpif_cast(facet->rule->up.ofproto);
4800 struct ofport_dpif *in_port;
4801 struct xlate_in xin;
4803 facet->prev_packet_count = facet->packet_count;
4804 facet->prev_byte_count = facet->byte_count;
4805 facet->prev_used = facet->used;
4807 in_port = get_ofp_port(ofproto, facet->flow.in_port);
4808 if (in_port && in_port->tnl_port) {
4809 netdev_vport_inc_rx(in_port->up.netdev, &stats);
4812 rule_credit_stats(facet->rule, &stats);
4813 netflow_flow_update_time(ofproto->netflow, &facet->nf_flow,
4815 netflow_flow_update_flags(&facet->nf_flow, facet->tcp_flags);
4816 update_mirror_stats(ofproto, facet->xout.mirrors, stats.n_packets,
4819 xlate_in_init(&xin, ofproto, &facet->flow, facet->rule,
4820 stats.tcp_flags, NULL);
4821 xin.resubmit_stats = &stats;
4822 xin.may_learn = may_learn;
4823 xlate_actions_for_side_effects(&xin);
4828 push_all_stats__(bool run_fast)
4830 static long long int rl = LLONG_MIN;
4831 struct ofproto_dpif *ofproto;
4833 if (time_msec() < rl) {
4837 HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
4838 struct cls_cursor cursor;
4839 struct facet *facet;
4841 cls_cursor_init(&cursor, &ofproto->facets, NULL);
4842 CLS_CURSOR_FOR_EACH (facet, cr, &cursor) {
4843 facet_push_stats(facet, false);
4850 rl = time_msec() + 100;
4854 push_all_stats(void)
4856 push_all_stats__(true);
4860 rule_credit_stats(struct rule_dpif *rule, const struct dpif_flow_stats *stats)
4862 rule->packet_count += stats->n_packets;
4863 rule->byte_count += stats->n_bytes;
4864 ofproto_rule_update_used(&rule->up, stats->used);
4869 static struct subfacet *
4870 subfacet_find(struct dpif_backer *backer, const struct nlattr *key,
4871 size_t key_len, uint32_t key_hash)
4873 struct subfacet *subfacet;
4875 HMAP_FOR_EACH_WITH_HASH (subfacet, hmap_node, key_hash,
4876 &backer->subfacets) {
4877 if (subfacet->key_len == key_len
4878 && !memcmp(key, subfacet->key, key_len)) {
4886 /* Searches 'facet' (within 'ofproto') for a subfacet with the specified
4887 * 'key_fitness', 'key', and 'key_len' members in 'miss'. Returns the
4888 * existing subfacet if there is one, otherwise creates and returns a
4890 static struct subfacet *
4891 subfacet_create(struct facet *facet, struct flow_miss *miss,
4894 struct dpif_backer *backer = miss->ofproto->backer;
4895 enum odp_key_fitness key_fitness = miss->key_fitness;
4896 const struct nlattr *key = miss->key;
4897 size_t key_len = miss->key_len;
4899 struct subfacet *subfacet;
4901 key_hash = odp_flow_key_hash(key, key_len);
4903 if (list_is_empty(&facet->subfacets)) {
4904 subfacet = &facet->one_subfacet;
4906 subfacet = subfacet_find(backer, key, key_len, key_hash);
4908 if (subfacet->facet == facet) {
4912 /* This shouldn't happen. */
4913 VLOG_ERR_RL(&rl, "subfacet with wrong facet");
4914 subfacet_destroy(subfacet);
4917 subfacet = xmalloc(sizeof *subfacet);
4920 hmap_insert(&backer->subfacets, &subfacet->hmap_node, key_hash);
4921 list_push_back(&facet->subfacets, &subfacet->list_node);
4922 subfacet->facet = facet;
4923 subfacet->key_fitness = key_fitness;
4924 subfacet->key = xmemdup(key, key_len);
4925 subfacet->key_len = key_len;
4926 subfacet->used = now;
4927 subfacet->created = now;
4928 subfacet->dp_packet_count = 0;
4929 subfacet->dp_byte_count = 0;
4930 subfacet->path = SF_NOT_INSTALLED;
4931 subfacet->backer = backer;
4933 backer->subfacet_add_count++;
4937 /* Uninstalls 'subfacet' from the datapath, if it is installed, removes it from
4938 * its facet within 'ofproto', and frees it. */
4940 subfacet_destroy__(struct subfacet *subfacet)
4942 struct facet *facet = subfacet->facet;
4943 struct ofproto_dpif *ofproto = ofproto_dpif_cast(facet->rule->up.ofproto);
4945 /* Update ofproto stats before uninstall the subfacet. */
4946 ofproto->backer->subfacet_del_count++;
4948 subfacet_uninstall(subfacet);
4949 hmap_remove(&subfacet->backer->subfacets, &subfacet->hmap_node);
4950 list_remove(&subfacet->list_node);
4951 free(subfacet->key);
4952 if (subfacet != &facet->one_subfacet) {
4957 /* Destroys 'subfacet', as with subfacet_destroy__(), and then if this was the
4958 * last remaining subfacet in its facet destroys the facet too. */
4960 subfacet_destroy(struct subfacet *subfacet)
4962 struct facet *facet = subfacet->facet;
4964 if (list_is_singleton(&facet->subfacets)) {
4965 /* facet_remove() needs at least one subfacet (it will remove it). */
4966 facet_remove(facet);
4968 subfacet_destroy__(subfacet);
4973 subfacet_destroy_batch(struct dpif_backer *backer,
4974 struct subfacet **subfacets, int n)
4976 struct dpif_op ops[SUBFACET_DESTROY_MAX_BATCH];
4977 struct dpif_op *opsp[SUBFACET_DESTROY_MAX_BATCH];
4978 struct dpif_flow_stats stats[SUBFACET_DESTROY_MAX_BATCH];
4981 for (i = 0; i < n; i++) {
4982 ops[i].type = DPIF_OP_FLOW_DEL;
4983 ops[i].u.flow_del.key = subfacets[i]->key;
4984 ops[i].u.flow_del.key_len = subfacets[i]->key_len;
4985 ops[i].u.flow_del.stats = &stats[i];
4989 dpif_operate(backer->dpif, opsp, n);
4990 for (i = 0; i < n; i++) {
4991 subfacet_reset_dp_stats(subfacets[i], &stats[i]);
4992 subfacets[i]->path = SF_NOT_INSTALLED;
4993 subfacet_destroy(subfacets[i]);
4998 /* Updates 'subfacet''s datapath flow, setting its actions to 'actions_len'
4999 * bytes of actions in 'actions'. If 'stats' is non-null, statistics counters
5000 * in the datapath will be zeroed and 'stats' will be updated with traffic new
5001 * since 'subfacet' was last updated.
5003 * Returns 0 if successful, otherwise a positive errno value. */
5005 subfacet_install(struct subfacet *subfacet, const struct ofpbuf *odp_actions,
5006 struct dpif_flow_stats *stats)
5008 struct facet *facet = subfacet->facet;
5009 struct ofproto_dpif *ofproto = ofproto_dpif_cast(facet->rule->up.ofproto);
5010 enum subfacet_path path = facet->xout.slow ? SF_SLOW_PATH : SF_FAST_PATH;
5011 const struct nlattr *actions = odp_actions->data;
5012 size_t actions_len = odp_actions->size;
5014 uint64_t slow_path_stub[128 / 8];
5015 enum dpif_flow_put_flags flags;
5018 flags = DPIF_FP_CREATE | DPIF_FP_MODIFY;
5020 flags |= DPIF_FP_ZERO_STATS;
5023 if (path == SF_SLOW_PATH) {
5024 compose_slow_path(ofproto, &facet->flow, facet->xout.slow,
5025 slow_path_stub, sizeof slow_path_stub,
5026 &actions, &actions_len);
5029 ret = dpif_flow_put(ofproto->backer->dpif, flags, subfacet->key,
5030 subfacet->key_len, NULL, 0,
5031 actions, actions_len, stats);
5034 subfacet_reset_dp_stats(subfacet, stats);
5038 subfacet->path = path;
5043 /* If 'subfacet' is installed in the datapath, uninstalls it. */
5045 subfacet_uninstall(struct subfacet *subfacet)
5047 if (subfacet->path != SF_NOT_INSTALLED) {
5048 struct rule_dpif *rule = subfacet->facet->rule;
5049 struct ofproto_dpif *ofproto = ofproto_dpif_cast(rule->up.ofproto);
5050 struct dpif_flow_stats stats;
5053 error = dpif_flow_del(ofproto->backer->dpif, subfacet->key,
5054 subfacet->key_len, &stats);
5055 subfacet_reset_dp_stats(subfacet, &stats);
5057 subfacet_update_stats(subfacet, &stats);
5059 subfacet->path = SF_NOT_INSTALLED;
5061 ovs_assert(subfacet->dp_packet_count == 0);
5062 ovs_assert(subfacet->dp_byte_count == 0);
5066 /* Resets 'subfacet''s datapath statistics counters. This should be called
5067 * when 'subfacet''s statistics are cleared in the datapath. If 'stats' is
5068 * non-null, it should contain the statistics returned by dpif when 'subfacet'
5069 * was reset in the datapath. 'stats' will be modified to include only
5070 * statistics new since 'subfacet' was last updated. */
5072 subfacet_reset_dp_stats(struct subfacet *subfacet,
5073 struct dpif_flow_stats *stats)
5076 && subfacet->dp_packet_count <= stats->n_packets
5077 && subfacet->dp_byte_count <= stats->n_bytes) {
5078 stats->n_packets -= subfacet->dp_packet_count;
5079 stats->n_bytes -= subfacet->dp_byte_count;
5082 subfacet->dp_packet_count = 0;
5083 subfacet->dp_byte_count = 0;
5086 /* Folds the statistics from 'stats' into the counters in 'subfacet'.
5088 * Because of the meaning of a subfacet's counters, it only makes sense to do
5089 * this if 'stats' are not tracked in the datapath, that is, if 'stats'
5090 * represents a packet that was sent by hand or if it represents statistics
5091 * that have been cleared out of the datapath. */
5093 subfacet_update_stats(struct subfacet *subfacet,
5094 const struct dpif_flow_stats *stats)
5096 if (stats->n_packets || stats->used > subfacet->used) {
5097 struct facet *facet = subfacet->facet;
5099 subfacet->used = MAX(subfacet->used, stats->used);
5100 facet->used = MAX(facet->used, stats->used);
5101 facet->packet_count += stats->n_packets;
5102 facet->byte_count += stats->n_bytes;
5103 facet->tcp_flags |= stats->tcp_flags;
5109 /* Lookup 'flow' in 'ofproto''s classifier. If 'wc' is non-null, sets
5110 * the fields that were relevant as part of the lookup. */
5111 static struct rule_dpif *
5112 rule_dpif_lookup(struct ofproto_dpif *ofproto, const struct flow *flow,
5113 struct flow_wildcards *wc)
5115 struct rule_dpif *rule;
5117 rule = rule_dpif_lookup_in_table(ofproto, flow, wc, 0);
5122 return rule_dpif_miss_rule(ofproto, flow);
5126 rule_dpif_lookup_in_table(struct ofproto_dpif *ofproto,
5127 const struct flow *flow, struct flow_wildcards *wc,
5130 struct cls_rule *cls_rule;
5131 struct classifier *cls;
5134 if (table_id >= N_TABLES) {
5139 wc->masks.nw_frag |= FLOW_NW_FRAG_MASK;
5142 cls = &ofproto->up.tables[table_id].cls;
5143 frag = (flow->nw_frag & FLOW_NW_FRAG_ANY) != 0;
5144 if (frag && ofproto->up.frag_handling == OFPC_FRAG_NORMAL) {
5145 /* We must pretend that transport ports are unavailable. */
5146 struct flow ofpc_normal_flow = *flow;
5147 ofpc_normal_flow.tp_src = htons(0);
5148 ofpc_normal_flow.tp_dst = htons(0);
5149 cls_rule = classifier_lookup(cls, &ofpc_normal_flow, wc);
5150 } else if (frag && ofproto->up.frag_handling == OFPC_FRAG_DROP) {
5151 cls_rule = &ofproto->drop_frags_rule->up.cr;
5153 flow_wildcards_init_exact(wc);
5156 cls_rule = classifier_lookup(cls, flow, wc);
5158 return rule_dpif_cast(rule_from_cls_rule(cls_rule));
5162 rule_dpif_miss_rule(struct ofproto_dpif *ofproto, const struct flow *flow)
5164 struct ofport_dpif *port;
5166 port = get_ofp_port(ofproto, flow->in_port);
5168 VLOG_WARN_RL(&rl, "packet-in on unknown port %"PRIu16, flow->in_port);
5169 return ofproto->miss_rule;
5172 if (port->up.pp.config & OFPUTIL_PC_NO_PACKET_IN) {
5173 return ofproto->no_packet_in_rule;
5175 return ofproto->miss_rule;
5179 complete_operation(struct rule_dpif *rule)
5181 struct ofproto_dpif *ofproto = ofproto_dpif_cast(rule->up.ofproto);
5183 rule_invalidate(rule);
5185 struct dpif_completion *c = xmalloc(sizeof *c);
5186 c->op = rule->up.pending;
5187 list_push_back(&ofproto->completions, &c->list_node);
5189 ofoperation_complete(rule->up.pending, 0);
5193 static struct rule *
5196 struct rule_dpif *rule = xmalloc(sizeof *rule);
5201 rule_dealloc(struct rule *rule_)
5203 struct rule_dpif *rule = rule_dpif_cast(rule_);
5208 rule_construct(struct rule *rule_)
5210 struct rule_dpif *rule = rule_dpif_cast(rule_);
5211 struct ofproto_dpif *ofproto = ofproto_dpif_cast(rule->up.ofproto);
5212 struct rule_dpif *victim;
5215 rule->packet_count = 0;
5216 rule->byte_count = 0;
5218 victim = rule_dpif_cast(ofoperation_get_victim(rule->up.pending));
5219 if (victim && !list_is_empty(&victim->facets)) {
5220 struct facet *facet;
5222 rule->facets = victim->facets;
5223 list_moved(&rule->facets);
5224 LIST_FOR_EACH (facet, list_node, &rule->facets) {
5225 /* XXX: We're only clearing our local counters here. It's possible
5226 * that quite a few packets are unaccounted for in the datapath
5227 * statistics. These will be accounted to the new rule instead of
5228 * cleared as required. This could be fixed by clearing out the
5229 * datapath statistics for this facet, but currently it doesn't
5231 facet_reset_counters(facet);
5235 /* Must avoid list_moved() in this case. */
5236 list_init(&rule->facets);
5239 table_id = rule->up.table_id;
5241 rule->tag = victim->tag;
5242 } else if (table_id == 0) {
5247 miniflow_expand(&rule->up.cr.match.flow, &flow);
5248 rule->tag = rule_calculate_tag(&flow, &rule->up.cr.match.mask,
5249 ofproto->tables[table_id].basis);
5252 complete_operation(rule);
5257 rule_destruct(struct rule *rule_)
5259 struct rule_dpif *rule = rule_dpif_cast(rule_);
5260 struct facet *facet, *next_facet;
5262 LIST_FOR_EACH_SAFE (facet, next_facet, list_node, &rule->facets) {
5263 facet_revalidate(facet);
5266 complete_operation(rule);
5270 rule_get_stats(struct rule *rule_, uint64_t *packets, uint64_t *bytes)
5272 struct rule_dpif *rule = rule_dpif_cast(rule_);
5274 /* push_all_stats() can handle flow misses which, when using the learn
5275 * action, can cause rules to be added and deleted. This can corrupt our
5276 * caller's datastructures which assume that rule_get_stats() doesn't have
5277 * an impact on the flow table. To be safe, we disable miss handling. */
5278 push_all_stats__(false);
5280 /* Start from historical data for 'rule' itself that are no longer tracked
5281 * in facets. This counts, for example, facets that have expired. */
5282 *packets = rule->packet_count;
5283 *bytes = rule->byte_count;
5287 rule_dpif_execute(struct rule_dpif *rule, const struct flow *flow,
5288 struct ofpbuf *packet)
5290 struct ofproto_dpif *ofproto = ofproto_dpif_cast(rule->up.ofproto);
5291 struct dpif_flow_stats stats;
5292 struct xlate_out xout;
5293 struct xlate_in xin;
5295 dpif_flow_stats_extract(flow, packet, time_msec(), &stats);
5296 rule_credit_stats(rule, &stats);
5298 xlate_in_init(&xin, ofproto, flow, rule, stats.tcp_flags, packet);
5299 xin.resubmit_stats = &stats;
5300 xlate_actions(&xin, &xout);
5302 execute_odp_actions(ofproto, flow, xout.odp_actions.data,
5303 xout.odp_actions.size, packet);
5305 xlate_out_uninit(&xout);
5309 rule_execute(struct rule *rule, const struct flow *flow,
5310 struct ofpbuf *packet)
5312 rule_dpif_execute(rule_dpif_cast(rule), flow, packet);
5313 ofpbuf_delete(packet);
5318 rule_modify_actions(struct rule *rule_)
5320 struct rule_dpif *rule = rule_dpif_cast(rule_);
5322 complete_operation(rule);
5325 /* Sends 'packet' out 'ofport'.
5326 * May modify 'packet'.
5327 * Returns 0 if successful, otherwise a positive errno value. */
5329 send_packet(const struct ofport_dpif *ofport, struct ofpbuf *packet)
5331 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport->up.ofproto);
5332 uint64_t odp_actions_stub[1024 / 8];
5333 struct ofpbuf key, odp_actions;
5334 struct dpif_flow_stats stats;
5335 struct odputil_keybuf keybuf;
5336 struct ofpact_output output;
5337 struct xlate_out xout;
5338 struct xlate_in xin;
5342 ofpbuf_use_stub(&odp_actions, odp_actions_stub, sizeof odp_actions_stub);
5343 ofpbuf_use_stack(&key, &keybuf, sizeof keybuf);
5345 /* Use OFPP_NONE as the in_port to avoid special packet processing. */
5346 flow_extract(packet, 0, 0, NULL, OFPP_NONE, &flow);
5347 odp_flow_key_from_flow(&key, &flow, ofp_port_to_odp_port(ofproto,
5349 dpif_flow_stats_extract(&flow, packet, time_msec(), &stats);
5351 ofpact_init(&output.ofpact, OFPACT_OUTPUT, sizeof output);
5352 output.port = ofport->up.ofp_port;
5355 xlate_in_init(&xin, ofproto, &flow, NULL, 0, packet);
5356 xin.ofpacts_len = sizeof output;
5357 xin.ofpacts = &output.ofpact;
5358 xin.resubmit_stats = &stats;
5359 xlate_actions(&xin, &xout);
5361 error = dpif_execute(ofproto->backer->dpif,
5363 xout.odp_actions.data, xout.odp_actions.size,
5365 xlate_out_uninit(&xout);
5368 VLOG_WARN_RL(&rl, "%s: failed to send packet on port %s (%s)",
5369 ofproto->up.name, netdev_get_name(ofport->up.netdev),
5373 ofproto->stats.tx_packets++;
5374 ofproto->stats.tx_bytes += packet->size;
5378 /* Composes an ODP action for a "slow path" action for 'flow' within 'ofproto'.
5379 * The action will state 'slow' as the reason that the action is in the slow
5380 * path. (This is purely informational: it allows a human viewing "ovs-dpctl
5381 * dump-flows" output to see why a flow is in the slow path.)
5383 * The 'stub_size' bytes in 'stub' will be used to store the action.
5384 * 'stub_size' must be large enough for the action.
5386 * The action and its size will be stored in '*actionsp' and '*actions_lenp',
5389 compose_slow_path(const struct ofproto_dpif *ofproto, const struct flow *flow,
5390 enum slow_path_reason slow,
5391 uint64_t *stub, size_t stub_size,
5392 const struct nlattr **actionsp, size_t *actions_lenp)
5394 union user_action_cookie cookie;
5397 cookie.type = USER_ACTION_COOKIE_SLOW_PATH;
5398 cookie.slow_path.unused = 0;
5399 cookie.slow_path.reason = slow;
5401 ofpbuf_use_stack(&buf, stub, stub_size);
5402 if (slow & (SLOW_CFM | SLOW_BFD | SLOW_LACP | SLOW_STP)) {
5403 uint32_t pid = dpif_port_get_pid(ofproto->backer->dpif, UINT32_MAX);
5404 odp_put_userspace_action(pid, &cookie, sizeof cookie.slow_path, &buf);
5406 put_userspace_action(ofproto, &buf, flow, &cookie,
5407 sizeof cookie.slow_path);
5409 *actionsp = buf.data;
5410 *actions_lenp = buf.size;
5414 put_userspace_action(const struct ofproto_dpif *ofproto,
5415 struct ofpbuf *odp_actions,
5416 const struct flow *flow,
5417 const union user_action_cookie *cookie,
5418 const size_t cookie_size)
5422 pid = dpif_port_get_pid(ofproto->backer->dpif,
5423 ofp_port_to_odp_port(ofproto, flow->in_port));
5425 return odp_put_userspace_action(pid, cookie, cookie_size, odp_actions);
5430 update_mirror_stats(struct ofproto_dpif *ofproto, mirror_mask_t mirrors,
5431 uint64_t packets, uint64_t bytes)
5437 for (; mirrors; mirrors = zero_rightmost_1bit(mirrors)) {
5440 m = ofproto->mirrors[mirror_mask_ffs(mirrors) - 1];
5443 /* In normal circumstances 'm' will not be NULL. However,
5444 * if mirrors are reconfigured, we can temporarily get out
5445 * of sync in facet_revalidate(). We could "correct" the
5446 * mirror list before reaching here, but doing that would
5447 * not properly account the traffic stats we've currently
5448 * accumulated for previous mirror configuration. */
5452 m->packet_count += packets;
5453 m->byte_count += bytes;
5458 /* Optimized flow revalidation.
5460 * It's a difficult problem, in general, to tell which facets need to have
5461 * their actions recalculated whenever the OpenFlow flow table changes. We
5462 * don't try to solve that general problem: for most kinds of OpenFlow flow
5463 * table changes, we recalculate the actions for every facet. This is
5464 * relatively expensive, but it's good enough if the OpenFlow flow table
5465 * doesn't change very often.
5467 * However, we can expect one particular kind of OpenFlow flow table change to
5468 * happen frequently: changes caused by MAC learning. To avoid wasting a lot
5469 * of CPU on revalidating every facet whenever MAC learning modifies the flow
5470 * table, we add a special case that applies to flow tables in which every rule
5471 * has the same form (that is, the same wildcards), except that the table is
5472 * also allowed to have a single "catch-all" flow that matches all packets. We
5473 * optimize this case by tagging all of the facets that resubmit into the table
5474 * and invalidating the same tag whenever a flow changes in that table. The
5475 * end result is that we revalidate just the facets that need it (and sometimes
5476 * a few more, but not all of the facets or even all of the facets that
5477 * resubmit to the table modified by MAC learning). */
5479 /* Calculates the tag to use for 'flow' and mask 'mask' when it is inserted
5480 * into an OpenFlow table with the given 'basis'. */
5482 rule_calculate_tag(const struct flow *flow, const struct minimask *mask,
5485 if (minimask_is_catchall(mask)) {
5488 uint32_t hash = flow_hash_in_minimask(flow, mask, secret);
5489 return tag_create_deterministic(hash);
5493 /* Following a change to OpenFlow table 'table_id' in 'ofproto', update the
5494 * taggability of that table.
5496 * This function must be called after *each* change to a flow table. If you
5497 * skip calling it on some changes then the pointer comparisons at the end can
5498 * be invalid if you get unlucky. For example, if a flow removal causes a
5499 * cls_table to be destroyed and then a flow insertion causes a cls_table with
5500 * different wildcards to be created with the same address, then this function
5501 * will incorrectly skip revalidation. */
5503 table_update_taggable(struct ofproto_dpif *ofproto, uint8_t table_id)
5505 struct table_dpif *table = &ofproto->tables[table_id];
5506 const struct oftable *oftable = &ofproto->up.tables[table_id];
5507 struct cls_table *catchall, *other;
5508 struct cls_table *t;
5510 catchall = other = NULL;
5512 switch (hmap_count(&oftable->cls.tables)) {
5514 /* We could tag this OpenFlow table but it would make the logic a
5515 * little harder and it's a corner case that doesn't seem worth it
5521 HMAP_FOR_EACH (t, hmap_node, &oftable->cls.tables) {
5522 if (cls_table_is_catchall(t)) {
5524 } else if (!other) {
5527 /* Indicate that we can't tag this by setting both tables to
5528 * NULL. (We know that 'catchall' is already NULL.) */
5535 /* Can't tag this table. */
5539 if (table->catchall_table != catchall || table->other_table != other) {
5540 table->catchall_table = catchall;
5541 table->other_table = other;
5542 ofproto->backer->need_revalidate = REV_FLOW_TABLE;
5546 /* Given 'rule' that has changed in some way (either it is a rule being
5547 * inserted, a rule being deleted, or a rule whose actions are being
5548 * modified), marks facets for revalidation to ensure that packets will be
5549 * forwarded correctly according to the new state of the flow table.
5551 * This function must be called after *each* change to a flow table. See
5552 * the comment on table_update_taggable() for more information. */
5554 rule_invalidate(const struct rule_dpif *rule)
5556 struct ofproto_dpif *ofproto = ofproto_dpif_cast(rule->up.ofproto);
5558 table_update_taggable(ofproto, rule->up.table_id);
5560 if (!ofproto->backer->need_revalidate) {
5561 struct table_dpif *table = &ofproto->tables[rule->up.table_id];
5563 if (table->other_table && rule->tag) {
5564 tag_set_add(&ofproto->backer->revalidate_set, rule->tag);
5566 ofproto->backer->need_revalidate = REV_FLOW_TABLE;
5572 set_frag_handling(struct ofproto *ofproto_,
5573 enum ofp_config_flags frag_handling)
5575 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
5576 if (frag_handling != OFPC_FRAG_REASM) {
5577 ofproto->backer->need_revalidate = REV_RECONFIGURE;
5585 packet_out(struct ofproto *ofproto_, struct ofpbuf *packet,
5586 const struct flow *flow,
5587 const struct ofpact *ofpacts, size_t ofpacts_len)
5589 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
5590 struct odputil_keybuf keybuf;
5591 struct dpif_flow_stats stats;
5592 struct xlate_out xout;
5593 struct xlate_in xin;
5597 ofpbuf_use_stack(&key, &keybuf, sizeof keybuf);
5598 odp_flow_key_from_flow(&key, flow,
5599 ofp_port_to_odp_port(ofproto, flow->in_port));
5601 dpif_flow_stats_extract(flow, packet, time_msec(), &stats);
5603 xlate_in_init(&xin, ofproto, flow, NULL, stats.tcp_flags, packet);
5604 xin.resubmit_stats = &stats;
5605 xin.ofpacts_len = ofpacts_len;
5606 xin.ofpacts = ofpacts;
5608 xlate_actions(&xin, &xout);
5609 dpif_execute(ofproto->backer->dpif, key.data, key.size,
5610 xout.odp_actions.data, xout.odp_actions.size, packet);
5611 xlate_out_uninit(&xout);
5619 set_netflow(struct ofproto *ofproto_,
5620 const struct netflow_options *netflow_options)
5622 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
5624 if (netflow_options) {
5625 if (!ofproto->netflow) {
5626 ofproto->netflow = netflow_create();
5627 ofproto->backer->need_revalidate = REV_RECONFIGURE;
5629 return netflow_set_options(ofproto->netflow, netflow_options);
5630 } else if (ofproto->netflow) {
5631 ofproto->backer->need_revalidate = REV_RECONFIGURE;
5632 netflow_destroy(ofproto->netflow);
5633 ofproto->netflow = NULL;
5640 get_netflow_ids(const struct ofproto *ofproto_,
5641 uint8_t *engine_type, uint8_t *engine_id)
5643 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
5645 dpif_get_netflow_ids(ofproto->backer->dpif, engine_type, engine_id);
5649 send_active_timeout(struct ofproto_dpif *ofproto, struct facet *facet)
5651 if (!facet_is_controller_flow(facet) &&
5652 netflow_active_timeout_expired(ofproto->netflow, &facet->nf_flow)) {
5653 struct subfacet *subfacet;
5654 struct ofexpired expired;
5656 LIST_FOR_EACH (subfacet, list_node, &facet->subfacets) {
5657 if (subfacet->path == SF_FAST_PATH) {
5658 struct dpif_flow_stats stats;
5660 subfacet_install(subfacet, &facet->xout.odp_actions,
5662 subfacet_update_stats(subfacet, &stats);
5666 expired.flow = facet->flow;
5667 expired.packet_count = facet->packet_count;
5668 expired.byte_count = facet->byte_count;
5669 expired.used = facet->used;
5670 netflow_expire(ofproto->netflow, &facet->nf_flow, &expired);
5675 send_netflow_active_timeouts(struct ofproto_dpif *ofproto)
5677 struct cls_cursor cursor;
5678 struct facet *facet;
5680 cls_cursor_init(&cursor, &ofproto->facets, NULL);
5681 CLS_CURSOR_FOR_EACH (facet, cr, &cursor) {
5682 send_active_timeout(ofproto, facet);
5686 static struct ofproto_dpif *
5687 ofproto_dpif_lookup(const char *name)
5689 struct ofproto_dpif *ofproto;
5691 HMAP_FOR_EACH_WITH_HASH (ofproto, all_ofproto_dpifs_node,
5692 hash_string(name, 0), &all_ofproto_dpifs) {
5693 if (!strcmp(ofproto->up.name, name)) {
5701 ofproto_unixctl_fdb_flush(struct unixctl_conn *conn, int argc,
5702 const char *argv[], void *aux OVS_UNUSED)
5704 struct ofproto_dpif *ofproto;
5707 ofproto = ofproto_dpif_lookup(argv[1]);
5709 unixctl_command_reply_error(conn, "no such bridge");
5712 mac_learning_flush(ofproto->ml, &ofproto->backer->revalidate_set);
5714 HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
5715 mac_learning_flush(ofproto->ml, &ofproto->backer->revalidate_set);
5719 unixctl_command_reply(conn, "table successfully flushed");
5723 ofproto_unixctl_fdb_show(struct unixctl_conn *conn, int argc OVS_UNUSED,
5724 const char *argv[], void *aux OVS_UNUSED)
5726 struct ds ds = DS_EMPTY_INITIALIZER;
5727 const struct ofproto_dpif *ofproto;
5728 const struct mac_entry *e;
5730 ofproto = ofproto_dpif_lookup(argv[1]);
5732 unixctl_command_reply_error(conn, "no such bridge");
5736 ds_put_cstr(&ds, " port VLAN MAC Age\n");
5737 LIST_FOR_EACH (e, lru_node, &ofproto->ml->lrus) {
5738 struct ofbundle *bundle = e->port.p;
5739 ds_put_format(&ds, "%5d %4d "ETH_ADDR_FMT" %3d\n",
5740 ofbundle_get_a_port(bundle)->odp_port,
5741 e->vlan, ETH_ADDR_ARGS(e->mac),
5742 mac_entry_age(ofproto->ml, e));
5744 unixctl_command_reply(conn, ds_cstr(&ds));
5749 struct xlate_out xout;
5750 struct xlate_in xin;
5756 trace_format_rule(struct ds *result, int level, const struct rule_dpif *rule)
5758 ds_put_char_multiple(result, '\t', level);
5760 ds_put_cstr(result, "No match\n");
5764 ds_put_format(result, "Rule: table=%"PRIu8" cookie=%#"PRIx64" ",
5765 rule ? rule->up.table_id : 0, ntohll(rule->up.flow_cookie));
5766 cls_rule_format(&rule->up.cr, result);
5767 ds_put_char(result, '\n');
5769 ds_put_char_multiple(result, '\t', level);
5770 ds_put_cstr(result, "OpenFlow ");
5771 ofpacts_format(rule->up.ofpacts, rule->up.ofpacts_len, result);
5772 ds_put_char(result, '\n');
5776 trace_format_flow(struct ds *result, int level, const char *title,
5777 struct trace_ctx *trace)
5779 ds_put_char_multiple(result, '\t', level);
5780 ds_put_format(result, "%s: ", title);
5781 if (flow_equal(&trace->xin.flow, &trace->flow)) {
5782 ds_put_cstr(result, "unchanged");
5784 flow_format(result, &trace->xin.flow);
5785 trace->flow = trace->xin.flow;
5787 ds_put_char(result, '\n');
5791 trace_format_regs(struct ds *result, int level, const char *title,
5792 struct trace_ctx *trace)
5796 ds_put_char_multiple(result, '\t', level);
5797 ds_put_format(result, "%s:", title);
5798 for (i = 0; i < FLOW_N_REGS; i++) {
5799 ds_put_format(result, " reg%zu=0x%"PRIx32, i, trace->flow.regs[i]);
5801 ds_put_char(result, '\n');
5805 trace_format_odp(struct ds *result, int level, const char *title,
5806 struct trace_ctx *trace)
5808 struct ofpbuf *odp_actions = &trace->xout.odp_actions;
5810 ds_put_char_multiple(result, '\t', level);
5811 ds_put_format(result, "%s: ", title);
5812 format_odp_actions(result, odp_actions->data, odp_actions->size);
5813 ds_put_char(result, '\n');
5817 trace_resubmit(struct xlate_in *xin, struct rule_dpif *rule, int recurse)
5819 struct trace_ctx *trace = CONTAINER_OF(xin, struct trace_ctx, xin);
5820 struct ds *result = trace->result;
5822 ds_put_char(result, '\n');
5823 trace_format_flow(result, recurse + 1, "Resubmitted flow", trace);
5824 trace_format_regs(result, recurse + 1, "Resubmitted regs", trace);
5825 trace_format_odp(result, recurse + 1, "Resubmitted odp", trace);
5826 trace_format_rule(result, recurse + 1, rule);
5830 trace_report(struct xlate_in *xin, const char *s, int recurse)
5832 struct trace_ctx *trace = CONTAINER_OF(xin, struct trace_ctx, xin);
5833 struct ds *result = trace->result;
5835 ds_put_char_multiple(result, '\t', recurse);
5836 ds_put_cstr(result, s);
5837 ds_put_char(result, '\n');
5841 ofproto_unixctl_trace(struct unixctl_conn *conn, int argc, const char *argv[],
5842 void *aux OVS_UNUSED)
5844 const struct dpif_backer *backer;
5845 struct ofproto_dpif *ofproto;
5846 struct ofpbuf odp_key;
5847 struct ofpbuf *packet;
5855 ofpbuf_init(&odp_key, 0);
5857 /* Handle "-generate" or a hex string as the last argument. */
5858 if (!strcmp(argv[argc - 1], "-generate")) {
5859 packet = ofpbuf_new(0);
5862 const char *error = eth_from_hex(argv[argc - 1], &packet);
5865 } else if (argc == 4) {
5866 /* The 3-argument form must end in "-generate' or a hex string. */
5867 unixctl_command_reply_error(conn, error);
5872 /* Parse the flow and determine whether a datapath or
5873 * bridge is specified. If function odp_flow_key_from_string()
5874 * returns 0, the flow is a odp_flow. If function
5875 * parse_ofp_exact_flow() returns 0, the flow is a br_flow. */
5876 if (!odp_flow_from_string(argv[argc - 1], NULL, &odp_key, NULL)) {
5877 /* If the odp_flow is the second argument,
5878 * the datapath name is the first argument. */
5880 const char *dp_type;
5881 if (!strncmp(argv[1], "ovs-", 4)) {
5882 dp_type = argv[1] + 4;
5886 backer = shash_find_data(&all_dpif_backers, dp_type);
5888 unixctl_command_reply_error(conn, "Cannot find datapath "
5893 /* No datapath name specified, so there should be only one
5895 struct shash_node *node;
5896 if (shash_count(&all_dpif_backers) != 1) {
5897 unixctl_command_reply_error(conn, "Must specify datapath "
5898 "name, there is more than one type of datapath");
5901 node = shash_first(&all_dpif_backers);
5902 backer = node->data;
5905 /* Extract the ofproto_dpif object from the ofproto_receive()
5907 if (ofproto_receive(backer, NULL, odp_key.data,
5908 odp_key.size, &flow, NULL, &ofproto, NULL)) {
5909 unixctl_command_reply_error(conn, "Invalid datapath flow");
5912 ds_put_format(&result, "Bridge: %s\n", ofproto->up.name);
5913 } else if (!parse_ofp_exact_flow(&flow, argv[argc - 1])) {
5915 unixctl_command_reply_error(conn, "Must specify bridge name");
5919 ofproto = ofproto_dpif_lookup(argv[1]);
5921 unixctl_command_reply_error(conn, "Unknown bridge name");
5925 unixctl_command_reply_error(conn, "Bad flow syntax");
5929 /* Generate a packet, if requested. */
5931 if (!packet->size) {
5932 flow_compose(packet, &flow);
5934 ds_put_cstr(&result, "Packet: ");
5935 s = ofp_packet_to_string(packet->data, packet->size);
5936 ds_put_cstr(&result, s);
5939 /* Use the metadata from the flow and the packet argument
5940 * to reconstruct the flow. */
5941 flow_extract(packet, flow.skb_priority, flow.skb_mark, NULL,
5942 flow.in_port, &flow);
5946 ofproto_trace(ofproto, &flow, packet, &result);
5947 unixctl_command_reply(conn, ds_cstr(&result));
5950 ds_destroy(&result);
5951 ofpbuf_delete(packet);
5952 ofpbuf_uninit(&odp_key);
5956 ofproto_trace(struct ofproto_dpif *ofproto, const struct flow *flow,
5957 const struct ofpbuf *packet, struct ds *ds)
5959 struct rule_dpif *rule;
5961 ds_put_cstr(ds, "Flow: ");
5962 flow_format(ds, flow);
5963 ds_put_char(ds, '\n');
5965 rule = rule_dpif_lookup(ofproto, flow, NULL);
5967 trace_format_rule(ds, 0, rule);
5968 if (rule == ofproto->miss_rule) {
5969 ds_put_cstr(ds, "\nNo match, flow generates \"packet in\"s.\n");
5970 } else if (rule == ofproto->no_packet_in_rule) {
5971 ds_put_cstr(ds, "\nNo match, packets dropped because "
5972 "OFPPC_NO_PACKET_IN is set on in_port.\n");
5973 } else if (rule == ofproto->drop_frags_rule) {
5974 ds_put_cstr(ds, "\nPackets dropped because they are IP fragments "
5975 "and the fragment handling mode is \"drop\".\n");
5979 uint64_t odp_actions_stub[1024 / 8];
5980 struct ofpbuf odp_actions;
5981 struct trace_ctx trace;
5985 tcp_flags = packet ? packet_get_tcp_flags(packet, flow) : 0;
5988 ofpbuf_use_stub(&odp_actions,
5989 odp_actions_stub, sizeof odp_actions_stub);
5990 xlate_in_init(&trace.xin, ofproto, flow, rule, tcp_flags, packet);
5991 trace.xin.resubmit_hook = trace_resubmit;
5992 trace.xin.report_hook = trace_report;
5994 xlate_actions(&trace.xin, &trace.xout);
5996 ds_put_char(ds, '\n');
5997 trace_format_flow(ds, 0, "Final flow", &trace);
5999 match_init(&match, flow, &trace.xout.wc);
6000 ds_put_cstr(ds, "Relevant fields: ");
6001 match_format(&match, ds, OFP_DEFAULT_PRIORITY);
6002 ds_put_char(ds, '\n');
6004 ds_put_cstr(ds, "Datapath actions: ");
6005 format_odp_actions(ds, trace.xout.odp_actions.data,
6006 trace.xout.odp_actions.size);
6008 if (trace.xout.slow) {
6009 ds_put_cstr(ds, "\nThis flow is handled by the userspace "
6010 "slow path because it:");
6011 switch (trace.xout.slow) {
6013 ds_put_cstr(ds, "\n\t- Consists of CFM packets.");
6016 ds_put_cstr(ds, "\n\t- Consists of LACP packets.");
6019 ds_put_cstr(ds, "\n\t- Consists of STP packets.");
6022 ds_put_cstr(ds, "\n\t- Consists of BFD packets.");
6024 case SLOW_CONTROLLER:
6025 ds_put_cstr(ds, "\n\t- Sends \"packet-in\" messages "
6026 "to the OpenFlow controller.");
6033 xlate_out_uninit(&trace.xout);
6038 ofproto_dpif_clog(struct unixctl_conn *conn OVS_UNUSED, int argc OVS_UNUSED,
6039 const char *argv[] OVS_UNUSED, void *aux OVS_UNUSED)
6042 unixctl_command_reply(conn, NULL);
6046 ofproto_dpif_unclog(struct unixctl_conn *conn OVS_UNUSED, int argc OVS_UNUSED,
6047 const char *argv[] OVS_UNUSED, void *aux OVS_UNUSED)
6050 unixctl_command_reply(conn, NULL);
6053 /* Runs a self-check of flow translations in 'ofproto'. Appends a message to
6054 * 'reply' describing the results. */
6056 ofproto_dpif_self_check__(struct ofproto_dpif *ofproto, struct ds *reply)
6058 struct cls_cursor cursor;
6059 struct facet *facet;
6063 cls_cursor_init(&cursor, &ofproto->facets, NULL);
6064 CLS_CURSOR_FOR_EACH (facet, cr, &cursor) {
6065 if (!facet_check_consistency(facet)) {
6070 ofproto->backer->need_revalidate = REV_INCONSISTENCY;
6074 ds_put_format(reply, "%s: self-check failed (%d errors)\n",
6075 ofproto->up.name, errors);
6077 ds_put_format(reply, "%s: self-check passed\n", ofproto->up.name);
6082 ofproto_dpif_self_check(struct unixctl_conn *conn,
6083 int argc, const char *argv[], void *aux OVS_UNUSED)
6085 struct ds reply = DS_EMPTY_INITIALIZER;
6086 struct ofproto_dpif *ofproto;
6089 ofproto = ofproto_dpif_lookup(argv[1]);
6091 unixctl_command_reply_error(conn, "Unknown ofproto (use "
6092 "ofproto/list for help)");
6095 ofproto_dpif_self_check__(ofproto, &reply);
6097 HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
6098 ofproto_dpif_self_check__(ofproto, &reply);
6102 unixctl_command_reply(conn, ds_cstr(&reply));
6106 /* Store the current ofprotos in 'ofproto_shash'. Returns a sorted list
6107 * of the 'ofproto_shash' nodes. It is the responsibility of the caller
6108 * to destroy 'ofproto_shash' and free the returned value. */
6109 static const struct shash_node **
6110 get_ofprotos(struct shash *ofproto_shash)
6112 const struct ofproto_dpif *ofproto;
6114 HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
6115 char *name = xasprintf("%s@%s", ofproto->up.type, ofproto->up.name);
6116 shash_add_nocopy(ofproto_shash, name, ofproto);
6119 return shash_sort(ofproto_shash);
6123 ofproto_unixctl_dpif_dump_dps(struct unixctl_conn *conn, int argc OVS_UNUSED,
6124 const char *argv[] OVS_UNUSED,
6125 void *aux OVS_UNUSED)
6127 struct ds ds = DS_EMPTY_INITIALIZER;
6128 struct shash ofproto_shash;
6129 const struct shash_node **sorted_ofprotos;
6132 shash_init(&ofproto_shash);
6133 sorted_ofprotos = get_ofprotos(&ofproto_shash);
6134 for (i = 0; i < shash_count(&ofproto_shash); i++) {
6135 const struct shash_node *node = sorted_ofprotos[i];
6136 ds_put_format(&ds, "%s\n", node->name);
6139 shash_destroy(&ofproto_shash);
6140 free(sorted_ofprotos);
6142 unixctl_command_reply(conn, ds_cstr(&ds));
6147 show_dp_rates(struct ds *ds, const char *heading,
6148 const struct avg_subfacet_rates *rates)
6150 ds_put_format(ds, "%s add rate: %5.3f/min, del rate: %5.3f/min\n",
6151 heading, rates->add_rate, rates->del_rate);
6155 dpif_show_backer(const struct dpif_backer *backer, struct ds *ds)
6157 const struct shash_node **ofprotos;
6158 struct ofproto_dpif *ofproto;
6159 struct shash ofproto_shash;
6160 uint64_t n_hit, n_missed;
6161 long long int minutes;
6164 n_hit = n_missed = 0;
6165 HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
6166 if (ofproto->backer == backer) {
6167 n_missed += ofproto->n_missed;
6168 n_hit += ofproto->n_hit;
6172 ds_put_format(ds, "%s: hit:%"PRIu64" missed:%"PRIu64"\n",
6173 dpif_name(backer->dpif), n_hit, n_missed);
6174 ds_put_format(ds, "\tflows: cur: %zu, avg: %u, max: %u,"
6175 " life span: %lldms\n", hmap_count(&backer->subfacets),
6176 backer->avg_n_subfacet, backer->max_n_subfacet,
6177 backer->avg_subfacet_life);
6179 minutes = (time_msec() - backer->created) / (1000 * 60);
6180 if (minutes >= 60) {
6181 show_dp_rates(ds, "\thourly avg:", &backer->hourly);
6183 if (minutes >= 60 * 24) {
6184 show_dp_rates(ds, "\tdaily avg:", &backer->daily);
6186 show_dp_rates(ds, "\toverall avg:", &backer->lifetime);
6188 shash_init(&ofproto_shash);
6189 ofprotos = get_ofprotos(&ofproto_shash);
6190 for (i = 0; i < shash_count(&ofproto_shash); i++) {
6191 struct ofproto_dpif *ofproto = ofprotos[i]->data;
6192 const struct shash_node **ports;
6195 if (ofproto->backer != backer) {
6199 ds_put_format(ds, "\t%s: hit:%"PRIu64" missed:%"PRIu64"\n",
6200 ofproto->up.name, ofproto->n_hit, ofproto->n_missed);
6202 ports = shash_sort(&ofproto->up.port_by_name);
6203 for (j = 0; j < shash_count(&ofproto->up.port_by_name); j++) {
6204 const struct shash_node *node = ports[j];
6205 struct ofport *ofport = node->data;
6209 ds_put_format(ds, "\t\t%s %u/", netdev_get_name(ofport->netdev),
6212 odp_port = ofp_port_to_odp_port(ofproto, ofport->ofp_port);
6213 if (odp_port != OVSP_NONE) {
6214 ds_put_format(ds, "%"PRIu32":", odp_port);
6216 ds_put_cstr(ds, "none:");
6219 ds_put_format(ds, " (%s", netdev_get_type(ofport->netdev));
6222 if (!netdev_get_config(ofport->netdev, &config)) {
6223 const struct smap_node **nodes;
6226 nodes = smap_sort(&config);
6227 for (i = 0; i < smap_count(&config); i++) {
6228 const struct smap_node *node = nodes[i];
6229 ds_put_format(ds, "%c %s=%s", i ? ',' : ':',
6230 node->key, node->value);
6234 smap_destroy(&config);
6236 ds_put_char(ds, ')');
6237 ds_put_char(ds, '\n');
6241 shash_destroy(&ofproto_shash);
6246 ofproto_unixctl_dpif_show(struct unixctl_conn *conn, int argc OVS_UNUSED,
6247 const char *argv[] OVS_UNUSED, void *aux OVS_UNUSED)
6249 struct ds ds = DS_EMPTY_INITIALIZER;
6250 const struct shash_node **backers;
6253 backers = shash_sort(&all_dpif_backers);
6254 for (i = 0; i < shash_count(&all_dpif_backers); i++) {
6255 dpif_show_backer(backers[i]->data, &ds);
6259 unixctl_command_reply(conn, ds_cstr(&ds));
6263 /* Dump the megaflow (facet) cache. This is useful to check the
6264 * correctness of flow wildcarding, since the same mechanism is used for
6265 * both xlate caching and kernel wildcarding.
6267 * It's important to note that in the output the flow description uses
6268 * OpenFlow (OFP) ports, but the actions use datapath (ODP) ports.
6270 * This command is only needed for advanced debugging, so it's not
6271 * documented in the man page. */
6273 ofproto_unixctl_dpif_dump_megaflows(struct unixctl_conn *conn,
6274 int argc OVS_UNUSED, const char *argv[],
6275 void *aux OVS_UNUSED)
6277 struct ds ds = DS_EMPTY_INITIALIZER;
6278 const struct ofproto_dpif *ofproto;
6279 long long int now = time_msec();
6280 struct cls_cursor cursor;
6281 struct facet *facet;
6283 ofproto = ofproto_dpif_lookup(argv[1]);
6285 unixctl_command_reply_error(conn, "no such bridge");
6289 cls_cursor_init(&cursor, &ofproto->facets, NULL);
6290 CLS_CURSOR_FOR_EACH (facet, cr, &cursor) {
6291 cls_rule_format(&facet->cr, &ds);
6292 ds_put_cstr(&ds, ", ");
6293 ds_put_format(&ds, "n_subfacets:%zu, ", list_size(&facet->subfacets));
6294 ds_put_format(&ds, "used:%.3fs, ", (now - facet->used) / 1000.0);
6295 ds_put_cstr(&ds, "Datapath actions: ");
6296 if (facet->xout.slow) {
6297 uint64_t slow_path_stub[128 / 8];
6298 const struct nlattr *actions;
6301 compose_slow_path(ofproto, &facet->flow, facet->xout.slow,
6302 slow_path_stub, sizeof slow_path_stub,
6303 &actions, &actions_len);
6304 format_odp_actions(&ds, actions, actions_len);
6306 format_odp_actions(&ds, facet->xout.odp_actions.data,
6307 facet->xout.odp_actions.size);
6309 ds_put_cstr(&ds, "\n");
6312 ds_chomp(&ds, '\n');
6313 unixctl_command_reply(conn, ds_cstr(&ds));
6318 ofproto_unixctl_dpif_dump_flows(struct unixctl_conn *conn,
6319 int argc OVS_UNUSED, const char *argv[],
6320 void *aux OVS_UNUSED)
6322 struct ds ds = DS_EMPTY_INITIALIZER;
6323 const struct ofproto_dpif *ofproto;
6324 struct subfacet *subfacet;
6326 ofproto = ofproto_dpif_lookup(argv[1]);
6328 unixctl_command_reply_error(conn, "no such bridge");
6332 update_stats(ofproto->backer);
6334 HMAP_FOR_EACH (subfacet, hmap_node, &ofproto->backer->subfacets) {
6335 struct facet *facet = subfacet->facet;
6337 if (ofproto_dpif_cast(facet->rule->up.ofproto) != ofproto) {
6341 odp_flow_key_format(subfacet->key, subfacet->key_len, &ds);
6343 ds_put_format(&ds, ", packets:%"PRIu64", bytes:%"PRIu64", used:",
6344 subfacet->dp_packet_count, subfacet->dp_byte_count);
6345 if (subfacet->used) {
6346 ds_put_format(&ds, "%.3fs",
6347 (time_msec() - subfacet->used) / 1000.0);
6349 ds_put_format(&ds, "never");
6351 if (subfacet->facet->tcp_flags) {
6352 ds_put_cstr(&ds, ", flags:");
6353 packet_format_tcp_flags(&ds, subfacet->facet->tcp_flags);
6356 ds_put_cstr(&ds, ", actions:");
6357 if (facet->xout.slow) {
6358 uint64_t slow_path_stub[128 / 8];
6359 const struct nlattr *actions;
6362 compose_slow_path(ofproto, &facet->flow, facet->xout.slow,
6363 slow_path_stub, sizeof slow_path_stub,
6364 &actions, &actions_len);
6365 format_odp_actions(&ds, actions, actions_len);
6367 format_odp_actions(&ds, facet->xout.odp_actions.data,
6368 facet->xout.odp_actions.size);
6370 ds_put_char(&ds, '\n');
6373 unixctl_command_reply(conn, ds_cstr(&ds));
6378 ofproto_unixctl_dpif_del_flows(struct unixctl_conn *conn,
6379 int argc OVS_UNUSED, const char *argv[],
6380 void *aux OVS_UNUSED)
6382 struct ds ds = DS_EMPTY_INITIALIZER;
6383 struct ofproto_dpif *ofproto;
6385 ofproto = ofproto_dpif_lookup(argv[1]);
6387 unixctl_command_reply_error(conn, "no such bridge");
6391 flush(&ofproto->up);
6393 unixctl_command_reply(conn, ds_cstr(&ds));
6398 ofproto_dpif_unixctl_init(void)
6400 static bool registered;
6406 unixctl_command_register(
6408 "[dp_name]|bridge odp_flow|br_flow [-generate|packet]",
6409 1, 3, ofproto_unixctl_trace, NULL);
6410 unixctl_command_register("fdb/flush", "[bridge]", 0, 1,
6411 ofproto_unixctl_fdb_flush, NULL);
6412 unixctl_command_register("fdb/show", "bridge", 1, 1,
6413 ofproto_unixctl_fdb_show, NULL);
6414 unixctl_command_register("ofproto/clog", "", 0, 0,
6415 ofproto_dpif_clog, NULL);
6416 unixctl_command_register("ofproto/unclog", "", 0, 0,
6417 ofproto_dpif_unclog, NULL);
6418 unixctl_command_register("ofproto/self-check", "[bridge]", 0, 1,
6419 ofproto_dpif_self_check, NULL);
6420 unixctl_command_register("dpif/dump-dps", "", 0, 0,
6421 ofproto_unixctl_dpif_dump_dps, NULL);
6422 unixctl_command_register("dpif/show", "", 0, 0, ofproto_unixctl_dpif_show,
6424 unixctl_command_register("dpif/dump-flows", "bridge", 1, 1,
6425 ofproto_unixctl_dpif_dump_flows, NULL);
6426 unixctl_command_register("dpif/del-flows", "bridge", 1, 1,
6427 ofproto_unixctl_dpif_del_flows, NULL);
6428 unixctl_command_register("dpif/dump-megaflows", "bridge", 1, 1,
6429 ofproto_unixctl_dpif_dump_megaflows, NULL);
6432 /* Linux VLAN device support (e.g. "eth0.10" for VLAN 10.)
6434 * This is deprecated. It is only for compatibility with broken device drivers
6435 * in old versions of Linux that do not properly support VLANs when VLAN
6436 * devices are not used. When broken device drivers are no longer in
6437 * widespread use, we will delete these interfaces. */
6440 set_realdev(struct ofport *ofport_, uint16_t realdev_ofp_port, int vid)
6442 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport_->ofproto);
6443 struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
6445 if (realdev_ofp_port == ofport->realdev_ofp_port
6446 && vid == ofport->vlandev_vid) {
6450 ofproto->backer->need_revalidate = REV_RECONFIGURE;
6452 if (ofport->realdev_ofp_port) {
6455 if (realdev_ofp_port && ofport->bundle) {
6456 /* vlandevs are enslaved to their realdevs, so they are not allowed to
6457 * themselves be part of a bundle. */
6458 bundle_set(ofport->up.ofproto, ofport->bundle, NULL);
6461 ofport->realdev_ofp_port = realdev_ofp_port;
6462 ofport->vlandev_vid = vid;
6464 if (realdev_ofp_port) {
6465 vsp_add(ofport, realdev_ofp_port, vid);
6472 hash_realdev_vid(uint16_t realdev_ofp_port, int vid)
6474 return hash_2words(realdev_ofp_port, vid);
6477 /* Returns the OFP port number of the Linux VLAN device that corresponds to
6478 * 'vlan_tci' on the network device with port number 'realdev_ofp_port' in
6479 * 'struct ofport_dpif'. For example, given 'realdev_ofp_port' of eth0 and
6480 * 'vlan_tci' 9, it would return the port number of eth0.9.
6482 * Unless VLAN splinters are enabled for port 'realdev_ofp_port', this
6483 * function just returns its 'realdev_ofp_port' argument. */
6485 vsp_realdev_to_vlandev(const struct ofproto_dpif *ofproto,
6486 uint16_t realdev_ofp_port, ovs_be16 vlan_tci)
6488 if (!hmap_is_empty(&ofproto->realdev_vid_map)) {
6489 int vid = vlan_tci_to_vid(vlan_tci);
6490 const struct vlan_splinter *vsp;
6492 HMAP_FOR_EACH_WITH_HASH (vsp, realdev_vid_node,
6493 hash_realdev_vid(realdev_ofp_port, vid),
6494 &ofproto->realdev_vid_map) {
6495 if (vsp->realdev_ofp_port == realdev_ofp_port
6496 && vsp->vid == vid) {
6497 return vsp->vlandev_ofp_port;
6501 return realdev_ofp_port;
6504 static struct vlan_splinter *
6505 vlandev_find(const struct ofproto_dpif *ofproto, uint16_t vlandev_ofp_port)
6507 struct vlan_splinter *vsp;
6509 HMAP_FOR_EACH_WITH_HASH (vsp, vlandev_node, hash_int(vlandev_ofp_port, 0),
6510 &ofproto->vlandev_map) {
6511 if (vsp->vlandev_ofp_port == vlandev_ofp_port) {
6519 /* Returns the OpenFlow port number of the "real" device underlying the Linux
6520 * VLAN device with OpenFlow port number 'vlandev_ofp_port' and stores the
6521 * VLAN VID of the Linux VLAN device in '*vid'. For example, given
6522 * 'vlandev_ofp_port' of eth0.9, it would return the OpenFlow port number of
6523 * eth0 and store 9 in '*vid'.
6525 * Returns 0 and does not modify '*vid' if 'vlandev_ofp_port' is not a Linux
6526 * VLAN device. Unless VLAN splinters are enabled, this is what this function
6529 vsp_vlandev_to_realdev(const struct ofproto_dpif *ofproto,
6530 uint16_t vlandev_ofp_port, int *vid)
6532 if (!hmap_is_empty(&ofproto->vlandev_map)) {
6533 const struct vlan_splinter *vsp;
6535 vsp = vlandev_find(ofproto, vlandev_ofp_port);
6540 return vsp->realdev_ofp_port;
6546 /* Given 'flow', a flow representing a packet received on 'ofproto', checks
6547 * whether 'flow->in_port' represents a Linux VLAN device. If so, changes
6548 * 'flow->in_port' to the "real" device backing the VLAN device, sets
6549 * 'flow->vlan_tci' to the VLAN VID, and returns true. Otherwise (which is
6550 * always the case unless VLAN splinters are enabled), returns false without
6551 * making any changes. */
6553 vsp_adjust_flow(const struct ofproto_dpif *ofproto, struct flow *flow)
6558 realdev = vsp_vlandev_to_realdev(ofproto, flow->in_port, &vid);
6563 /* Cause the flow to be processed as if it came in on the real device with
6564 * the VLAN device's VLAN ID. */
6565 flow->in_port = realdev;
6566 flow->vlan_tci = htons((vid & VLAN_VID_MASK) | VLAN_CFI);
6571 vsp_remove(struct ofport_dpif *port)
6573 struct ofproto_dpif *ofproto = ofproto_dpif_cast(port->up.ofproto);
6574 struct vlan_splinter *vsp;
6576 vsp = vlandev_find(ofproto, port->up.ofp_port);
6578 hmap_remove(&ofproto->vlandev_map, &vsp->vlandev_node);
6579 hmap_remove(&ofproto->realdev_vid_map, &vsp->realdev_vid_node);
6582 port->realdev_ofp_port = 0;
6584 VLOG_ERR("missing vlan device record");
6589 vsp_add(struct ofport_dpif *port, uint16_t realdev_ofp_port, int vid)
6591 struct ofproto_dpif *ofproto = ofproto_dpif_cast(port->up.ofproto);
6593 if (!vsp_vlandev_to_realdev(ofproto, port->up.ofp_port, NULL)
6594 && (vsp_realdev_to_vlandev(ofproto, realdev_ofp_port, htons(vid))
6595 == realdev_ofp_port)) {
6596 struct vlan_splinter *vsp;
6598 vsp = xmalloc(sizeof *vsp);
6599 hmap_insert(&ofproto->vlandev_map, &vsp->vlandev_node,
6600 hash_int(port->up.ofp_port, 0));
6601 hmap_insert(&ofproto->realdev_vid_map, &vsp->realdev_vid_node,
6602 hash_realdev_vid(realdev_ofp_port, vid));
6603 vsp->realdev_ofp_port = realdev_ofp_port;
6604 vsp->vlandev_ofp_port = port->up.ofp_port;
6607 port->realdev_ofp_port = realdev_ofp_port;
6609 VLOG_ERR("duplicate vlan device record");
6614 ofp_port_to_odp_port(const struct ofproto_dpif *ofproto, uint16_t ofp_port)
6616 const struct ofport_dpif *ofport = get_ofp_port(ofproto, ofp_port);
6617 return ofport ? ofport->odp_port : OVSP_NONE;
6620 static struct ofport_dpif *
6621 odp_port_to_ofport(const struct dpif_backer *backer, uint32_t odp_port)
6623 struct ofport_dpif *port;
6625 HMAP_FOR_EACH_IN_BUCKET (port, odp_port_node,
6626 hash_int(odp_port, 0),
6627 &backer->odp_to_ofport_map) {
6628 if (port->odp_port == odp_port) {
6637 odp_port_to_ofp_port(const struct ofproto_dpif *ofproto, uint32_t odp_port)
6639 struct ofport_dpif *port;
6641 port = odp_port_to_ofport(ofproto->backer, odp_port);
6642 if (port && &ofproto->up == port->up.ofproto) {
6643 return port->up.ofp_port;
6649 /* Compute exponentially weighted moving average, adding 'new' as the newest,
6650 * most heavily weighted element. 'base' designates the rate of decay: after
6651 * 'base' further updates, 'new''s weight in the EWMA decays to about 1/e
6654 exp_mavg(double *avg, int base, double new)
6656 *avg = (*avg * (base - 1) + new) / base;
6660 update_moving_averages(struct dpif_backer *backer)
6662 const int min_ms = 60 * 1000; /* milliseconds in one minute. */
6663 long long int minutes = (time_msec() - backer->created) / min_ms;
6666 backer->lifetime.add_rate = (double) backer->total_subfacet_add_count
6668 backer->lifetime.del_rate = (double) backer->total_subfacet_del_count
6671 backer->lifetime.add_rate = 0.0;
6672 backer->lifetime.del_rate = 0.0;
6675 /* Update hourly averages on the minute boundaries. */
6676 if (time_msec() - backer->last_minute >= min_ms) {
6677 exp_mavg(&backer->hourly.add_rate, 60, backer->subfacet_add_count);
6678 exp_mavg(&backer->hourly.del_rate, 60, backer->subfacet_del_count);
6680 /* Update daily averages on the hour boundaries. */
6681 if ((backer->last_minute - backer->created) / min_ms % 60 == 59) {
6682 exp_mavg(&backer->daily.add_rate, 24, backer->hourly.add_rate);
6683 exp_mavg(&backer->daily.del_rate, 24, backer->hourly.del_rate);
6686 backer->total_subfacet_add_count += backer->subfacet_add_count;
6687 backer->total_subfacet_del_count += backer->subfacet_del_count;
6688 backer->subfacet_add_count = 0;
6689 backer->subfacet_del_count = 0;
6690 backer->last_minute += min_ms;
6694 const struct ofproto_class ofproto_dpif_class = {
6729 port_is_lacp_current,
6730 NULL, /* rule_choose_table */
6737 rule_modify_actions,
6751 get_stp_port_status,
6758 is_mirror_output_bundle,
6759 forward_bpdu_changed,
6760 set_mac_table_config,