2 * Copyright (c) 2009, 2010, 2011, 2012, 2013 Nicira, Inc.
4 * Licensed under the Apache License, Version 2.0 (the "License");
5 * you may not use this file except in compliance with the License.
6 * You may obtain a copy of the License at:
8 * http://www.apache.org/licenses/LICENSE-2.0
10 * Unless required by applicable law or agreed to in writing, software
11 * distributed under the License is distributed on an "AS IS" BASIS,
12 * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
13 * See the License for the specific language governing permissions and
14 * limitations under the License.
19 #include "ofproto/ofproto-dpif.h"
20 #include "ofproto/ofproto-provider.h"
27 #include "byte-order.h"
32 #include "dynamic-string.h"
33 #include "fail-open.h"
37 #include "mac-learning.h"
38 #include "meta-flow.h"
39 #include "multipath.h"
40 #include "netdev-vport.h"
45 #include "odp-execute.h"
48 #include "ofp-actions.h"
49 #include "ofp-parse.h"
50 #include "ofp-print.h"
51 #include "ofproto-dpif-governor.h"
52 #include "ofproto-dpif-ipfix.h"
53 #include "ofproto-dpif-sflow.h"
54 #include "ofproto-dpif-xlate.h"
55 #include "poll-loop.h"
60 #include "unaligned.h"
62 #include "vlan-bitmap.h"
65 VLOG_DEFINE_THIS_MODULE(ofproto_dpif);
67 COVERAGE_DEFINE(ofproto_dpif_expired);
68 COVERAGE_DEFINE(facet_changed_rule);
69 COVERAGE_DEFINE(facet_revalidate);
70 COVERAGE_DEFINE(facet_unexpected);
71 COVERAGE_DEFINE(facet_suppress);
72 COVERAGE_DEFINE(subfacet_install_fail);
77 static struct rule_dpif *rule_dpif_lookup(struct ofproto_dpif *,
79 struct flow_wildcards *wc);
81 static void rule_get_stats(struct rule *, uint64_t *packets, uint64_t *bytes);
82 static void rule_invalidate(const struct rule_dpif *);
84 static void mirror_destroy(struct ofmirror *);
85 static void update_mirror_stats(struct ofproto_dpif *ofproto,
86 mirror_mask_t mirrors,
87 uint64_t packets, uint64_t bytes);
89 static void bundle_remove(struct ofport *);
90 static void bundle_update(struct ofbundle *);
91 static void bundle_destroy(struct ofbundle *);
92 static void bundle_del_port(struct ofport_dpif *);
93 static void bundle_run(struct ofbundle *);
94 static void bundle_wait(struct ofbundle *);
96 static void stp_run(struct ofproto_dpif *ofproto);
97 static void stp_wait(struct ofproto_dpif *ofproto);
98 static int set_stp_port(struct ofport *,
99 const struct ofproto_port_stp_settings *);
101 static void compose_slow_path(const struct ofproto_dpif *, const struct flow *,
102 enum slow_path_reason,
103 uint64_t *stub, size_t stub_size,
104 const struct nlattr **actionsp,
105 size_t *actions_lenp);
107 /* A subfacet (see "struct subfacet" below) has three possible installation
110 * - SF_NOT_INSTALLED: Not installed in the datapath. This will only be the
111 * case just after the subfacet is created, just before the subfacet is
112 * destroyed, or if the datapath returns an error when we try to install a
115 * - SF_FAST_PATH: The subfacet's actions are installed in the datapath.
117 * - SF_SLOW_PATH: An action that sends every packet for the subfacet through
118 * ofproto_dpif is installed in the datapath.
121 SF_NOT_INSTALLED, /* No datapath flow for this subfacet. */
122 SF_FAST_PATH, /* Full actions are installed. */
123 SF_SLOW_PATH, /* Send-to-userspace action is installed. */
126 /* A dpif flow and actions associated with a facet.
128 * See also the large comment on struct facet. */
131 struct hmap_node hmap_node; /* In struct ofproto_dpif 'subfacets' list. */
132 struct list list_node; /* In struct facet's 'facets' list. */
133 struct facet *facet; /* Owning facet. */
134 struct dpif_backer *backer; /* Owning backer. */
136 enum odp_key_fitness key_fitness;
140 long long int used; /* Time last used; time created if not used. */
141 long long int created; /* Time created. */
143 uint64_t dp_packet_count; /* Last known packet count in the datapath. */
144 uint64_t dp_byte_count; /* Last known byte count in the datapath. */
146 enum subfacet_path path; /* Installed in datapath? */
149 #define SUBFACET_DESTROY_MAX_BATCH 50
151 static struct subfacet *subfacet_create(struct facet *, struct flow_miss *miss,
153 static struct subfacet *subfacet_find(struct dpif_backer *,
154 const struct nlattr *key, size_t key_len,
156 static void subfacet_destroy(struct subfacet *);
157 static void subfacet_destroy__(struct subfacet *);
158 static void subfacet_destroy_batch(struct dpif_backer *,
159 struct subfacet **, int n);
160 static void subfacet_reset_dp_stats(struct subfacet *,
161 struct dpif_flow_stats *);
162 static void subfacet_update_stats(struct subfacet *,
163 const struct dpif_flow_stats *);
164 static int subfacet_install(struct subfacet *,
165 const struct ofpbuf *odp_actions,
166 struct dpif_flow_stats *);
167 static void subfacet_uninstall(struct subfacet *);
169 /* A unique, non-overlapping instantiation of an OpenFlow flow.
171 * A facet associates a "struct flow", which represents the Open vSwitch
172 * userspace idea of an exact-match flow, with one or more subfacets.
173 * While the facet is created based on an exact-match flow, it is stored
174 * within the ofproto based on the wildcards that could be expressed
175 * based on the flow table and other configuration. (See the 'wc'
176 * description in "struct xlate_out" for more details.)
178 * Each subfacet tracks the datapath's idea of the flow equivalent to
179 * the facet. When the kernel module (or other dpif implementation) and
180 * Open vSwitch userspace agree on the definition of a flow key, there
181 * is exactly one subfacet per facet. If the dpif implementation
182 * supports more-specific flow matching than userspace, however, a facet
183 * can have more than one subfacet. Examples include the dpif
184 * implementation not supporting the same wildcards as userspace or some
185 * distinction in flow that userspace simply doesn't understand.
187 * Flow expiration works in terms of subfacets, so a facet must have at
188 * least one subfacet or it will never expire, leaking memory. */
191 struct hmap_node hmap_node; /* In owning ofproto's 'facets' hmap. */
192 struct list list_node; /* In owning rule's 'facets' list. */
193 struct rule_dpif *rule; /* Owning rule. */
196 struct list subfacets;
197 long long int used; /* Time last used; time created if not used. */
200 struct flow flow; /* Flow of the creating subfacet. */
201 struct cls_rule cr; /* In 'ofproto_dpif's facets classifier. */
205 * - Do include packets and bytes sent "by hand", e.g. with
208 * - Do include packets and bytes that were obtained from the datapath
209 * when a subfacet's statistics were reset (e.g. dpif_flow_put() with
210 * DPIF_FP_ZERO_STATS).
212 * - Do not include packets or bytes that can be obtained from the
213 * datapath for any existing subfacet.
215 uint64_t packet_count; /* Number of packets received. */
216 uint64_t byte_count; /* Number of bytes received. */
218 /* Resubmit statistics. */
219 uint64_t prev_packet_count; /* Number of packets from last stats push. */
220 uint64_t prev_byte_count; /* Number of bytes from last stats push. */
221 long long int prev_used; /* Used time from last stats push. */
224 uint64_t accounted_bytes; /* Bytes processed by facet_account(). */
225 struct netflow_flow nf_flow; /* Per-flow NetFlow tracking data. */
226 uint8_t tcp_flags; /* TCP flags seen for this 'rule'. */
228 struct xlate_out xout;
230 /* Storage for a single subfacet, to reduce malloc() time and space
231 * overhead. (A facet always has at least one subfacet and in the common
232 * case has exactly one subfacet. However, 'one_subfacet' may not
233 * always be valid, since it could have been removed after newer
234 * subfacets were pushed onto the 'subfacets' list.) */
235 struct subfacet one_subfacet;
237 long long int learn_rl; /* Rate limiter for facet_learn(). */
240 static struct facet *facet_create(const struct flow_miss *, struct rule_dpif *,
242 struct dpif_flow_stats *);
243 static void facet_remove(struct facet *);
244 static void facet_free(struct facet *);
246 static struct facet *facet_find(struct ofproto_dpif *, const struct flow *);
247 static struct facet *facet_lookup_valid(struct ofproto_dpif *,
248 const struct flow *);
249 static bool facet_revalidate(struct facet *);
250 static bool facet_check_consistency(struct facet *);
252 static void facet_flush_stats(struct facet *);
254 static void facet_reset_counters(struct facet *);
255 static void facet_push_stats(struct facet *, bool may_learn);
256 static void facet_learn(struct facet *);
257 static void facet_account(struct facet *);
258 static void push_all_stats(void);
260 static bool facet_is_controller_flow(struct facet *);
262 /* Node in 'ofport_dpif''s 'priorities' map. Used to maintain a map from
263 * 'priority' (the datapath's term for QoS queue) to the dscp bits which all
264 * traffic egressing the 'ofport' with that priority should be marked with. */
265 struct priority_to_dscp {
266 struct hmap_node hmap_node; /* Node in 'ofport_dpif''s 'priorities' map. */
267 uint32_t priority; /* Priority of this queue (see struct flow). */
269 uint8_t dscp; /* DSCP bits to mark outgoing traffic with. */
272 /* Linux VLAN device support (e.g. "eth0.10" for VLAN 10.)
274 * This is deprecated. It is only for compatibility with broken device drivers
275 * in old versions of Linux that do not properly support VLANs when VLAN
276 * devices are not used. When broken device drivers are no longer in
277 * widespread use, we will delete these interfaces. */
278 struct vlan_splinter {
279 struct hmap_node realdev_vid_node;
280 struct hmap_node vlandev_node;
281 ofp_port_t realdev_ofp_port;
282 ofp_port_t vlandev_ofp_port;
286 static bool vsp_adjust_flow(const struct ofproto_dpif *, struct flow *);
287 static void vsp_remove(struct ofport_dpif *);
288 static void vsp_add(struct ofport_dpif *, ofp_port_t realdev_ofp_port, int vid);
290 static ofp_port_t odp_port_to_ofp_port(const struct ofproto_dpif *,
291 odp_port_t odp_port);
293 static struct ofport_dpif *
294 ofport_dpif_cast(const struct ofport *ofport)
296 return ofport ? CONTAINER_OF(ofport, struct ofport_dpif, up) : NULL;
299 static void port_run(struct ofport_dpif *);
300 static void port_run_fast(struct ofport_dpif *);
301 static void port_wait(struct ofport_dpif *);
302 static int set_bfd(struct ofport *, const struct smap *);
303 static int set_cfm(struct ofport *, const struct cfm_settings *);
304 static void ofport_clear_priorities(struct ofport_dpif *);
305 static void ofport_update_peer(struct ofport_dpif *);
306 static void run_fast_rl(void);
308 struct dpif_completion {
309 struct list list_node;
310 struct ofoperation *op;
313 /* Reasons that we might need to revalidate every facet, and corresponding
316 * A value of 0 means that there is no need to revalidate.
318 * It would be nice to have some cleaner way to integrate with coverage
319 * counters, but with only a few reasons I guess this is good enough for
321 enum revalidate_reason {
322 REV_RECONFIGURE = 1, /* Switch configuration changed. */
323 REV_STP, /* Spanning tree protocol port status change. */
324 REV_PORT_TOGGLED, /* Port enabled or disabled by CFM, LACP, ...*/
325 REV_FLOW_TABLE, /* Flow table changed. */
326 REV_INCONSISTENCY /* Facet self-check failed. */
328 COVERAGE_DEFINE(rev_reconfigure);
329 COVERAGE_DEFINE(rev_stp);
330 COVERAGE_DEFINE(rev_port_toggled);
331 COVERAGE_DEFINE(rev_flow_table);
332 COVERAGE_DEFINE(rev_inconsistency);
334 /* Drop keys are odp flow keys which have drop flows installed in the kernel.
335 * These are datapath flows which have no associated ofproto, if they did we
336 * would use facets. */
338 struct hmap_node hmap_node;
343 struct avg_subfacet_rates {
344 double add_rate; /* Moving average of new flows created per minute. */
345 double del_rate; /* Moving average of flows deleted per minute. */
348 /* All datapaths of a given type share a single dpif backer instance. */
353 struct timer next_expiration;
354 struct hmap odp_to_ofport_map; /* ODP port to ofport mapping. */
356 struct simap tnl_backers; /* Set of dpif ports backing tunnels. */
358 /* Facet revalidation flags applying to facets which use this backer. */
359 enum revalidate_reason need_revalidate; /* Revalidate every facet. */
360 struct tag_set revalidate_set; /* Revalidate only matching facets. */
362 struct hmap drop_keys; /* Set of dropped odp keys. */
363 bool recv_set_enable; /* Enables or disables receiving packets. */
365 struct hmap subfacets;
366 struct governor *governor;
368 /* Subfacet statistics.
370 * These keep track of the total number of subfacets added and deleted and
371 * flow life span. They are useful for computing the flow rates stats
372 * exposed via "ovs-appctl dpif/show". The goal is to learn about
373 * traffic patterns in ways that we can use later to improve Open vSwitch
374 * performance in new situations. */
375 long long int created; /* Time when it is created. */
376 unsigned max_n_subfacet; /* Maximum number of flows */
377 unsigned avg_n_subfacet; /* Average number of flows. */
378 long long int avg_subfacet_life; /* Average life span of subfacets. */
380 /* The average number of subfacets... */
381 struct avg_subfacet_rates hourly; /* ...over the last hour. */
382 struct avg_subfacet_rates daily; /* ...over the last day. */
383 struct avg_subfacet_rates lifetime; /* ...over the switch lifetime. */
384 long long int last_minute; /* Last time 'hourly' was updated. */
386 /* Number of subfacets added or deleted since 'last_minute'. */
387 unsigned subfacet_add_count;
388 unsigned subfacet_del_count;
390 /* Number of subfacets added or deleted from 'created' to 'last_minute.' */
391 unsigned long long int total_subfacet_add_count;
392 unsigned long long int total_subfacet_del_count;
395 /* All existing ofproto_backer instances, indexed by ofproto->up.type. */
396 static struct shash all_dpif_backers = SHASH_INITIALIZER(&all_dpif_backers);
398 static void drop_key_clear(struct dpif_backer *);
399 static struct ofport_dpif *
400 odp_port_to_ofport(const struct dpif_backer *, odp_port_t odp_port);
401 static void update_moving_averages(struct dpif_backer *backer);
403 /* Defer flow mod completion until "ovs-appctl ofproto/unclog"? (Useful only
404 * for debugging the asynchronous flow_mod implementation.) */
407 /* By default, flows in the datapath are wildcarded (megaflows). They
408 * may be disabled with the "ovs-appctl dpif/disable-megaflows" command. */
409 static bool enable_megaflows = true;
411 /* All existing ofproto_dpif instances, indexed by ->up.name. */
412 static struct hmap all_ofproto_dpifs = HMAP_INITIALIZER(&all_ofproto_dpifs);
414 static void ofproto_dpif_unixctl_init(void);
417 #define FLOW_MISS_MAX_BATCH 50
418 static int handle_upcalls(struct dpif_backer *, unsigned int max_batch);
420 /* Flow expiration. */
421 static int expire(struct dpif_backer *);
424 static void send_netflow_active_timeouts(struct ofproto_dpif *);
427 static int send_packet(const struct ofport_dpif *, struct ofpbuf *packet);
429 /* Global variables. */
430 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
432 /* Initial mappings of port to bridge mappings. */
433 static struct shash init_ofp_ports = SHASH_INITIALIZER(&init_ofp_ports);
435 /* Factory functions. */
438 init(const struct shash *iface_hints)
440 struct shash_node *node;
442 /* Make a local copy, since we don't own 'iface_hints' elements. */
443 SHASH_FOR_EACH(node, iface_hints) {
444 const struct iface_hint *orig_hint = node->data;
445 struct iface_hint *new_hint = xmalloc(sizeof *new_hint);
447 new_hint->br_name = xstrdup(orig_hint->br_name);
448 new_hint->br_type = xstrdup(orig_hint->br_type);
449 new_hint->ofp_port = orig_hint->ofp_port;
451 shash_add(&init_ofp_ports, node->name, new_hint);
456 enumerate_types(struct sset *types)
458 dp_enumerate_types(types);
462 enumerate_names(const char *type, struct sset *names)
464 struct ofproto_dpif *ofproto;
467 HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
468 if (strcmp(type, ofproto->up.type)) {
471 sset_add(names, ofproto->up.name);
478 del(const char *type, const char *name)
483 error = dpif_open(name, type, &dpif);
485 error = dpif_delete(dpif);
492 port_open_type(const char *datapath_type, const char *port_type)
494 return dpif_port_open_type(datapath_type, port_type);
497 /* Type functions. */
499 static struct ofproto_dpif *
500 lookup_ofproto_dpif_by_port_name(const char *name)
502 struct ofproto_dpif *ofproto;
504 HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
505 if (sset_contains(&ofproto->ports, name)) {
514 type_run(const char *type)
516 static long long int push_timer = LLONG_MIN;
517 struct dpif_backer *backer;
521 backer = shash_find_data(&all_dpif_backers, type);
523 /* This is not necessarily a problem, since backers are only
524 * created on demand. */
528 dpif_run(backer->dpif);
530 /* The most natural place to push facet statistics is when they're pulled
531 * from the datapath. However, when there are many flows in the datapath,
532 * this expensive operation can occur so frequently, that it reduces our
533 * ability to quickly set up flows. To reduce the cost, we push statistics
535 if (time_msec() > push_timer) {
536 push_timer = time_msec() + 2000;
540 /* If vswitchd started with other_config:flow_restore_wait set as "true",
541 * and the configuration has now changed to "false", enable receiving
542 * packets from the datapath. */
543 if (!backer->recv_set_enable && !ofproto_get_flow_restore_wait()) {
544 backer->recv_set_enable = true;
546 error = dpif_recv_set(backer->dpif, backer->recv_set_enable);
548 VLOG_ERR("Failed to enable receiving packets in dpif.");
551 dpif_flow_flush(backer->dpif);
552 backer->need_revalidate = REV_RECONFIGURE;
555 if (backer->need_revalidate
556 || !tag_set_is_empty(&backer->revalidate_set)) {
557 struct tag_set revalidate_set = backer->revalidate_set;
558 bool need_revalidate = backer->need_revalidate;
559 struct ofproto_dpif *ofproto;
560 struct simap_node *node;
561 struct simap tmp_backers;
563 /* Handle tunnel garbage collection. */
564 simap_init(&tmp_backers);
565 simap_swap(&backer->tnl_backers, &tmp_backers);
567 HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
568 struct ofport_dpif *iter;
570 if (backer != ofproto->backer) {
574 HMAP_FOR_EACH (iter, up.hmap_node, &ofproto->up.ports) {
575 char namebuf[NETDEV_VPORT_NAME_BUFSIZE];
578 if (!iter->is_tunnel) {
582 dp_port = netdev_vport_get_dpif_port(iter->up.netdev,
583 namebuf, sizeof namebuf);
584 node = simap_find(&tmp_backers, dp_port);
586 simap_put(&backer->tnl_backers, dp_port, node->data);
587 simap_delete(&tmp_backers, node);
588 node = simap_find(&backer->tnl_backers, dp_port);
590 node = simap_find(&backer->tnl_backers, dp_port);
592 odp_port_t odp_port = ODPP_NONE;
594 if (!dpif_port_add(backer->dpif, iter->up.netdev,
596 simap_put(&backer->tnl_backers, dp_port,
597 odp_to_u32(odp_port));
598 node = simap_find(&backer->tnl_backers, dp_port);
603 iter->odp_port = node ? u32_to_odp(node->data) : ODPP_NONE;
604 if (tnl_port_reconfigure(iter, iter->up.netdev,
606 backer->need_revalidate = REV_RECONFIGURE;
611 SIMAP_FOR_EACH (node, &tmp_backers) {
612 dpif_port_del(backer->dpif, u32_to_odp(node->data));
614 simap_destroy(&tmp_backers);
616 switch (backer->need_revalidate) {
617 case REV_RECONFIGURE: COVERAGE_INC(rev_reconfigure); break;
618 case REV_STP: COVERAGE_INC(rev_stp); break;
619 case REV_PORT_TOGGLED: COVERAGE_INC(rev_port_toggled); break;
620 case REV_FLOW_TABLE: COVERAGE_INC(rev_flow_table); break;
621 case REV_INCONSISTENCY: COVERAGE_INC(rev_inconsistency); break;
624 if (backer->need_revalidate) {
625 /* Clear the drop_keys in case we should now be accepting some
626 * formerly dropped flows. */
627 drop_key_clear(backer);
630 /* Clear the revalidation flags. */
631 tag_set_init(&backer->revalidate_set);
632 backer->need_revalidate = 0;
634 HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
635 struct facet *facet, *next;
636 struct cls_cursor cursor;
638 if (ofproto->backer != backer) {
642 cls_cursor_init(&cursor, &ofproto->facets, NULL);
643 CLS_CURSOR_FOR_EACH_SAFE (facet, next, cr, &cursor) {
645 || tag_set_intersects(&revalidate_set, facet->xout.tags)) {
646 facet_revalidate(facet);
653 if (!backer->recv_set_enable) {
654 /* Wake up before a max of 1000ms. */
655 timer_set_duration(&backer->next_expiration, 1000);
656 } else if (timer_expired(&backer->next_expiration)) {
657 int delay = expire(backer);
658 timer_set_duration(&backer->next_expiration, delay);
661 /* Check for port changes in the dpif. */
662 while ((error = dpif_port_poll(backer->dpif, &devname)) == 0) {
663 struct ofproto_dpif *ofproto;
664 struct dpif_port port;
666 /* Don't report on the datapath's device. */
667 if (!strcmp(devname, dpif_base_name(backer->dpif))) {
671 HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node,
672 &all_ofproto_dpifs) {
673 if (simap_contains(&ofproto->backer->tnl_backers, devname)) {
678 ofproto = lookup_ofproto_dpif_by_port_name(devname);
679 if (dpif_port_query_by_name(backer->dpif, devname, &port)) {
680 /* The port was removed. If we know the datapath,
681 * report it through poll_set(). If we don't, it may be
682 * notifying us of a removal we initiated, so ignore it.
683 * If there's a pending ENOBUFS, let it stand, since
684 * everything will be reevaluated. */
685 if (ofproto && ofproto->port_poll_errno != ENOBUFS) {
686 sset_add(&ofproto->port_poll_set, devname);
687 ofproto->port_poll_errno = 0;
689 } else if (!ofproto) {
690 /* The port was added, but we don't know with which
691 * ofproto we should associate it. Delete it. */
692 dpif_port_del(backer->dpif, port.port_no);
694 dpif_port_destroy(&port);
700 if (error != EAGAIN) {
701 struct ofproto_dpif *ofproto;
703 /* There was some sort of error, so propagate it to all
704 * ofprotos that use this backer. */
705 HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node,
706 &all_ofproto_dpifs) {
707 if (ofproto->backer == backer) {
708 sset_clear(&ofproto->port_poll_set);
709 ofproto->port_poll_errno = error;
714 if (backer->governor) {
717 governor_run(backer->governor);
719 /* If the governor has shrunk to its minimum size and the number of
720 * subfacets has dwindled, then drop the governor entirely.
722 * For hysteresis, the number of subfacets to drop the governor is
723 * smaller than the number needed to trigger its creation. */
724 n_subfacets = hmap_count(&backer->subfacets);
725 if (n_subfacets * 4 < flow_eviction_threshold
726 && governor_is_idle(backer->governor)) {
727 governor_destroy(backer->governor);
728 backer->governor = NULL;
736 dpif_backer_run_fast(struct dpif_backer *backer, int max_batch)
740 /* If recv_set_enable is false, we should not handle upcalls. */
741 if (!backer->recv_set_enable) {
745 /* Handle one or more batches of upcalls, until there's nothing left to do
746 * or until we do a fixed total amount of work.
748 * We do work in batches because it can be much cheaper to set up a number
749 * of flows and fire off their patches all at once. We do multiple batches
750 * because in some cases handling a packet can cause another packet to be
751 * queued almost immediately as part of the return flow. Both
752 * optimizations can make major improvements on some benchmarks and
753 * presumably for real traffic as well. */
755 while (work < max_batch) {
756 int retval = handle_upcalls(backer, max_batch - work);
767 type_run_fast(const char *type)
769 struct dpif_backer *backer;
771 backer = shash_find_data(&all_dpif_backers, type);
773 /* This is not necessarily a problem, since backers are only
774 * created on demand. */
778 return dpif_backer_run_fast(backer, FLOW_MISS_MAX_BATCH);
784 static long long int port_rl = LLONG_MIN;
785 static unsigned int backer_rl = 0;
787 if (time_msec() >= port_rl) {
788 struct ofproto_dpif *ofproto;
789 struct ofport_dpif *ofport;
791 HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
793 HMAP_FOR_EACH (ofport, up.hmap_node, &ofproto->up.ports) {
794 port_run_fast(ofport);
797 port_rl = time_msec() + 200;
800 /* XXX: We have to be careful not to do too much work in this function. If
801 * we call dpif_backer_run_fast() too often, or with too large a batch,
802 * performance improves signifcantly, but at a cost. It's possible for the
803 * number of flows in the datapath to increase without bound, and for poll
804 * loops to take 10s of seconds. The correct solution to this problem,
805 * long term, is to separate flow miss handling into it's own thread so it
806 * isn't affected by revalidations, and expirations. Until then, this is
807 * the best we can do. */
808 if (++backer_rl >= 10) {
809 struct shash_node *node;
812 SHASH_FOR_EACH (node, &all_dpif_backers) {
813 dpif_backer_run_fast(node->data, 1);
819 type_wait(const char *type)
821 struct dpif_backer *backer;
823 backer = shash_find_data(&all_dpif_backers, type);
825 /* This is not necessarily a problem, since backers are only
826 * created on demand. */
830 if (backer->governor) {
831 governor_wait(backer->governor);
834 timer_wait(&backer->next_expiration);
837 /* Basic life-cycle. */
839 static int add_internal_flows(struct ofproto_dpif *);
841 static struct ofproto *
844 struct ofproto_dpif *ofproto = xmalloc(sizeof *ofproto);
849 dealloc(struct ofproto *ofproto_)
851 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
856 close_dpif_backer(struct dpif_backer *backer)
858 struct shash_node *node;
860 ovs_assert(backer->refcount > 0);
862 if (--backer->refcount) {
866 drop_key_clear(backer);
867 hmap_destroy(&backer->drop_keys);
869 simap_destroy(&backer->tnl_backers);
870 hmap_destroy(&backer->odp_to_ofport_map);
871 node = shash_find(&all_dpif_backers, backer->type);
873 shash_delete(&all_dpif_backers, node);
874 dpif_close(backer->dpif);
876 ovs_assert(hmap_is_empty(&backer->subfacets));
877 hmap_destroy(&backer->subfacets);
878 governor_destroy(backer->governor);
883 /* Datapath port slated for removal from datapath. */
885 struct list list_node;
890 open_dpif_backer(const char *type, struct dpif_backer **backerp)
892 struct dpif_backer *backer;
893 struct dpif_port_dump port_dump;
894 struct dpif_port port;
895 struct shash_node *node;
896 struct list garbage_list;
897 struct odp_garbage *garbage, *next;
903 backer = shash_find_data(&all_dpif_backers, type);
910 backer_name = xasprintf("ovs-%s", type);
912 /* Remove any existing datapaths, since we assume we're the only
913 * userspace controlling the datapath. */
915 dp_enumerate_names(type, &names);
916 SSET_FOR_EACH(name, &names) {
917 struct dpif *old_dpif;
919 /* Don't remove our backer if it exists. */
920 if (!strcmp(name, backer_name)) {
924 if (dpif_open(name, type, &old_dpif)) {
925 VLOG_WARN("couldn't open old datapath %s to remove it", name);
927 dpif_delete(old_dpif);
928 dpif_close(old_dpif);
931 sset_destroy(&names);
933 backer = xmalloc(sizeof *backer);
935 error = dpif_create_and_open(backer_name, type, &backer->dpif);
938 VLOG_ERR("failed to open datapath of type %s: %s", type,
939 ovs_strerror(error));
944 backer->type = xstrdup(type);
945 backer->governor = NULL;
946 backer->refcount = 1;
947 hmap_init(&backer->odp_to_ofport_map);
948 hmap_init(&backer->drop_keys);
949 hmap_init(&backer->subfacets);
950 timer_set_duration(&backer->next_expiration, 1000);
951 backer->need_revalidate = 0;
952 simap_init(&backer->tnl_backers);
953 tag_set_init(&backer->revalidate_set);
954 backer->recv_set_enable = !ofproto_get_flow_restore_wait();
957 if (backer->recv_set_enable) {
958 dpif_flow_flush(backer->dpif);
961 /* Loop through the ports already on the datapath and remove any
962 * that we don't need anymore. */
963 list_init(&garbage_list);
964 dpif_port_dump_start(&port_dump, backer->dpif);
965 while (dpif_port_dump_next(&port_dump, &port)) {
966 node = shash_find(&init_ofp_ports, port.name);
967 if (!node && strcmp(port.name, dpif_base_name(backer->dpif))) {
968 garbage = xmalloc(sizeof *garbage);
969 garbage->odp_port = port.port_no;
970 list_push_front(&garbage_list, &garbage->list_node);
973 dpif_port_dump_done(&port_dump);
975 LIST_FOR_EACH_SAFE (garbage, next, list_node, &garbage_list) {
976 dpif_port_del(backer->dpif, garbage->odp_port);
977 list_remove(&garbage->list_node);
981 shash_add(&all_dpif_backers, type, backer);
983 error = dpif_recv_set(backer->dpif, backer->recv_set_enable);
985 VLOG_ERR("failed to listen on datapath of type %s: %s",
986 type, ovs_strerror(error));
987 close_dpif_backer(backer);
991 backer->max_n_subfacet = 0;
992 backer->created = time_msec();
993 backer->last_minute = backer->created;
994 memset(&backer->hourly, 0, sizeof backer->hourly);
995 memset(&backer->daily, 0, sizeof backer->daily);
996 memset(&backer->lifetime, 0, sizeof backer->lifetime);
997 backer->subfacet_add_count = 0;
998 backer->subfacet_del_count = 0;
999 backer->total_subfacet_add_count = 0;
1000 backer->total_subfacet_del_count = 0;
1001 backer->avg_n_subfacet = 0;
1002 backer->avg_subfacet_life = 0;
1008 construct(struct ofproto *ofproto_)
1010 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1011 struct shash_node *node, *next;
1012 odp_port_t max_ports;
1016 error = open_dpif_backer(ofproto->up.type, &ofproto->backer);
1021 max_ports = dpif_get_max_ports(ofproto->backer->dpif);
1022 ofproto_init_max_ports(ofproto_, u16_to_ofp(MIN(odp_to_u32(max_ports),
1023 ofp_to_u16(OFPP_MAX))));
1025 ofproto->netflow = NULL;
1026 ofproto->sflow = NULL;
1027 ofproto->ipfix = NULL;
1028 ofproto->stp = NULL;
1029 hmap_init(&ofproto->bundles);
1030 ofproto->ml = mac_learning_create(MAC_ENTRY_DEFAULT_IDLE_TIME);
1031 for (i = 0; i < MAX_MIRRORS; i++) {
1032 ofproto->mirrors[i] = NULL;
1034 ofproto->has_bonded_bundles = false;
1036 classifier_init(&ofproto->facets);
1037 ofproto->consistency_rl = LLONG_MIN;
1039 for (i = 0; i < N_TABLES; i++) {
1040 struct table_dpif *table = &ofproto->tables[i];
1042 table->catchall_table = NULL;
1043 table->other_table = NULL;
1044 table->basis = random_uint32();
1047 list_init(&ofproto->completions);
1049 ofproto_dpif_unixctl_init();
1051 ofproto->has_mirrors = false;
1052 hmap_init(&ofproto->vlandev_map);
1053 hmap_init(&ofproto->realdev_vid_map);
1055 sset_init(&ofproto->ports);
1056 sset_init(&ofproto->ghost_ports);
1057 sset_init(&ofproto->port_poll_set);
1058 ofproto->port_poll_errno = 0;
1060 SHASH_FOR_EACH_SAFE (node, next, &init_ofp_ports) {
1061 struct iface_hint *iface_hint = node->data;
1063 if (!strcmp(iface_hint->br_name, ofproto->up.name)) {
1064 /* Check if the datapath already has this port. */
1065 if (dpif_port_exists(ofproto->backer->dpif, node->name)) {
1066 sset_add(&ofproto->ports, node->name);
1069 free(iface_hint->br_name);
1070 free(iface_hint->br_type);
1072 shash_delete(&init_ofp_ports, node);
1076 hmap_insert(&all_ofproto_dpifs, &ofproto->all_ofproto_dpifs_node,
1077 hash_string(ofproto->up.name, 0));
1078 memset(&ofproto->stats, 0, sizeof ofproto->stats);
1080 ofproto_init_tables(ofproto_, N_TABLES);
1081 error = add_internal_flows(ofproto);
1082 ofproto->up.tables[TBL_INTERNAL].flags = OFTABLE_HIDDEN | OFTABLE_READONLY;
1085 ofproto->n_missed = 0;
1091 add_internal_flow(struct ofproto_dpif *ofproto, int id,
1092 const struct ofpbuf *ofpacts, struct rule_dpif **rulep)
1094 struct ofputil_flow_mod fm;
1097 match_init_catchall(&fm.match);
1099 match_set_reg(&fm.match, 0, id);
1100 fm.new_cookie = htonll(0);
1101 fm.cookie = htonll(0);
1102 fm.cookie_mask = htonll(0);
1103 fm.table_id = TBL_INTERNAL;
1104 fm.command = OFPFC_ADD;
1105 fm.idle_timeout = 0;
1106 fm.hard_timeout = 0;
1110 fm.ofpacts = ofpacts->data;
1111 fm.ofpacts_len = ofpacts->size;
1113 error = ofproto_flow_mod(&ofproto->up, &fm);
1115 VLOG_ERR_RL(&rl, "failed to add internal flow %d (%s)",
1116 id, ofperr_to_string(error));
1120 *rulep = rule_dpif_lookup_in_table(ofproto, &fm.match.flow, NULL,
1122 ovs_assert(*rulep != NULL);
1128 add_internal_flows(struct ofproto_dpif *ofproto)
1130 struct ofpact_controller *controller;
1131 uint64_t ofpacts_stub[128 / 8];
1132 struct ofpbuf ofpacts;
1136 ofpbuf_use_stack(&ofpacts, ofpacts_stub, sizeof ofpacts_stub);
1139 controller = ofpact_put_CONTROLLER(&ofpacts);
1140 controller->max_len = UINT16_MAX;
1141 controller->controller_id = 0;
1142 controller->reason = OFPR_NO_MATCH;
1143 ofpact_pad(&ofpacts);
1145 error = add_internal_flow(ofproto, id++, &ofpacts, &ofproto->miss_rule);
1150 ofpbuf_clear(&ofpacts);
1151 error = add_internal_flow(ofproto, id++, &ofpacts,
1152 &ofproto->no_packet_in_rule);
1157 error = add_internal_flow(ofproto, id++, &ofpacts,
1158 &ofproto->drop_frags_rule);
1163 complete_operations(struct ofproto_dpif *ofproto)
1165 struct dpif_completion *c, *next;
1167 LIST_FOR_EACH_SAFE (c, next, list_node, &ofproto->completions) {
1168 ofoperation_complete(c->op, 0);
1169 list_remove(&c->list_node);
1175 destruct(struct ofproto *ofproto_)
1177 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1178 struct rule_dpif *rule, *next_rule;
1179 struct oftable *table;
1182 ofproto->backer->need_revalidate = REV_RECONFIGURE;
1183 hmap_remove(&all_ofproto_dpifs, &ofproto->all_ofproto_dpifs_node);
1184 complete_operations(ofproto);
1186 OFPROTO_FOR_EACH_TABLE (table, &ofproto->up) {
1187 struct cls_cursor cursor;
1189 cls_cursor_init(&cursor, &table->cls, NULL);
1190 CLS_CURSOR_FOR_EACH_SAFE (rule, next_rule, up.cr, &cursor) {
1191 ofproto_rule_destroy(&rule->up);
1195 for (i = 0; i < MAX_MIRRORS; i++) {
1196 mirror_destroy(ofproto->mirrors[i]);
1199 netflow_destroy(ofproto->netflow);
1200 dpif_sflow_unref(ofproto->sflow);
1201 hmap_destroy(&ofproto->bundles);
1202 mac_learning_unref(ofproto->ml);
1204 classifier_destroy(&ofproto->facets);
1206 hmap_destroy(&ofproto->vlandev_map);
1207 hmap_destroy(&ofproto->realdev_vid_map);
1209 sset_destroy(&ofproto->ports);
1210 sset_destroy(&ofproto->ghost_ports);
1211 sset_destroy(&ofproto->port_poll_set);
1213 close_dpif_backer(ofproto->backer);
1217 run_fast(struct ofproto *ofproto_)
1219 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1220 struct ofport_dpif *ofport;
1222 /* Do not perform any periodic activity required by 'ofproto' while
1223 * waiting for flow restore to complete. */
1224 if (ofproto_get_flow_restore_wait()) {
1228 HMAP_FOR_EACH (ofport, up.hmap_node, &ofproto->up.ports) {
1229 port_run_fast(ofport);
1236 run(struct ofproto *ofproto_)
1238 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1239 struct ofport_dpif *ofport;
1240 struct ofbundle *bundle;
1244 complete_operations(ofproto);
1247 /* Do not perform any periodic activity below required by 'ofproto' while
1248 * waiting for flow restore to complete. */
1249 if (ofproto_get_flow_restore_wait()) {
1253 error = run_fast(ofproto_);
1258 if (ofproto->netflow) {
1259 if (netflow_run(ofproto->netflow)) {
1260 send_netflow_active_timeouts(ofproto);
1263 if (ofproto->sflow) {
1264 dpif_sflow_run(ofproto->sflow);
1267 HMAP_FOR_EACH (ofport, up.hmap_node, &ofproto->up.ports) {
1270 HMAP_FOR_EACH (bundle, hmap_node, &ofproto->bundles) {
1275 mac_learning_run(ofproto->ml, &ofproto->backer->revalidate_set);
1277 /* Check the consistency of a random facet, to aid debugging. */
1278 if (time_msec() >= ofproto->consistency_rl
1279 && !classifier_is_empty(&ofproto->facets)
1280 && !ofproto->backer->need_revalidate) {
1281 struct cls_table *table;
1282 struct cls_rule *cr;
1283 struct facet *facet;
1285 ofproto->consistency_rl = time_msec() + 250;
1287 table = CONTAINER_OF(hmap_random_node(&ofproto->facets.tables),
1288 struct cls_table, hmap_node);
1289 cr = CONTAINER_OF(hmap_random_node(&table->rules), struct cls_rule,
1291 facet = CONTAINER_OF(cr, struct facet, cr);
1293 if (!tag_set_intersects(&ofproto->backer->revalidate_set,
1294 facet->xout.tags)) {
1295 if (!facet_check_consistency(facet)) {
1296 ofproto->backer->need_revalidate = REV_INCONSISTENCY;
1305 wait(struct ofproto *ofproto_)
1307 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1308 struct ofport_dpif *ofport;
1309 struct ofbundle *bundle;
1311 if (!clogged && !list_is_empty(&ofproto->completions)) {
1312 poll_immediate_wake();
1315 if (ofproto_get_flow_restore_wait()) {
1319 dpif_wait(ofproto->backer->dpif);
1320 dpif_recv_wait(ofproto->backer->dpif);
1321 if (ofproto->sflow) {
1322 dpif_sflow_wait(ofproto->sflow);
1324 if (!tag_set_is_empty(&ofproto->backer->revalidate_set)) {
1325 poll_immediate_wake();
1327 HMAP_FOR_EACH (ofport, up.hmap_node, &ofproto->up.ports) {
1330 HMAP_FOR_EACH (bundle, hmap_node, &ofproto->bundles) {
1331 bundle_wait(bundle);
1333 if (ofproto->netflow) {
1334 netflow_wait(ofproto->netflow);
1336 mac_learning_wait(ofproto->ml);
1338 if (ofproto->backer->need_revalidate) {
1339 /* Shouldn't happen, but if it does just go around again. */
1340 VLOG_DBG_RL(&rl, "need revalidate in ofproto_wait_cb()");
1341 poll_immediate_wake();
1346 get_memory_usage(const struct ofproto *ofproto_, struct simap *usage)
1348 const struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1349 struct cls_cursor cursor;
1350 size_t n_subfacets = 0;
1351 struct facet *facet;
1353 simap_increase(usage, "facets", classifier_count(&ofproto->facets));
1355 cls_cursor_init(&cursor, &ofproto->facets, NULL);
1356 CLS_CURSOR_FOR_EACH (facet, cr, &cursor) {
1357 n_subfacets += list_size(&facet->subfacets);
1359 simap_increase(usage, "subfacets", n_subfacets);
1363 flush(struct ofproto *ofproto_)
1365 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1366 struct subfacet *subfacet, *next_subfacet;
1367 struct subfacet *batch[SUBFACET_DESTROY_MAX_BATCH];
1371 HMAP_FOR_EACH_SAFE (subfacet, next_subfacet, hmap_node,
1372 &ofproto->backer->subfacets) {
1373 if (ofproto_dpif_cast(subfacet->facet->rule->up.ofproto) != ofproto) {
1377 if (subfacet->path != SF_NOT_INSTALLED) {
1378 batch[n_batch++] = subfacet;
1379 if (n_batch >= SUBFACET_DESTROY_MAX_BATCH) {
1380 subfacet_destroy_batch(ofproto->backer, batch, n_batch);
1384 subfacet_destroy(subfacet);
1389 subfacet_destroy_batch(ofproto->backer, batch, n_batch);
1394 get_features(struct ofproto *ofproto_ OVS_UNUSED,
1395 bool *arp_match_ip, enum ofputil_action_bitmap *actions)
1397 *arp_match_ip = true;
1398 *actions = (OFPUTIL_A_OUTPUT |
1399 OFPUTIL_A_SET_VLAN_VID |
1400 OFPUTIL_A_SET_VLAN_PCP |
1401 OFPUTIL_A_STRIP_VLAN |
1402 OFPUTIL_A_SET_DL_SRC |
1403 OFPUTIL_A_SET_DL_DST |
1404 OFPUTIL_A_SET_NW_SRC |
1405 OFPUTIL_A_SET_NW_DST |
1406 OFPUTIL_A_SET_NW_TOS |
1407 OFPUTIL_A_SET_TP_SRC |
1408 OFPUTIL_A_SET_TP_DST |
1413 get_tables(struct ofproto *ofproto_, struct ofp12_table_stats *ots)
1415 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1416 struct dpif_dp_stats s;
1417 uint64_t n_miss, n_no_pkt_in, n_bytes, n_dropped_frags;
1420 strcpy(ots->name, "classifier");
1422 dpif_get_dp_stats(ofproto->backer->dpif, &s);
1423 rule_get_stats(&ofproto->miss_rule->up, &n_miss, &n_bytes);
1424 rule_get_stats(&ofproto->no_packet_in_rule->up, &n_no_pkt_in, &n_bytes);
1425 rule_get_stats(&ofproto->drop_frags_rule->up, &n_dropped_frags, &n_bytes);
1427 n_lookup = s.n_hit + s.n_missed - n_dropped_frags;
1428 ots->lookup_count = htonll(n_lookup);
1429 ots->matched_count = htonll(n_lookup - n_miss - n_no_pkt_in);
1432 static struct ofport *
1435 struct ofport_dpif *port = xmalloc(sizeof *port);
1440 port_dealloc(struct ofport *port_)
1442 struct ofport_dpif *port = ofport_dpif_cast(port_);
1447 port_construct(struct ofport *port_)
1449 struct ofport_dpif *port = ofport_dpif_cast(port_);
1450 struct ofproto_dpif *ofproto = ofproto_dpif_cast(port->up.ofproto);
1451 const struct netdev *netdev = port->up.netdev;
1452 char namebuf[NETDEV_VPORT_NAME_BUFSIZE];
1453 struct dpif_port dpif_port;
1456 ofproto->backer->need_revalidate = REV_RECONFIGURE;
1457 port->bundle = NULL;
1460 port->tag = tag_create_random();
1461 port->may_enable = true;
1462 port->stp_port = NULL;
1463 port->stp_state = STP_DISABLED;
1464 port->is_tunnel = false;
1466 hmap_init(&port->priorities);
1467 port->realdev_ofp_port = 0;
1468 port->vlandev_vid = 0;
1469 port->carrier_seq = netdev_get_carrier_resets(netdev);
1471 if (netdev_vport_is_patch(netdev)) {
1472 /* By bailing out here, we don't submit the port to the sFlow module
1473 * to be considered for counter polling export. This is correct
1474 * because the patch port represents an interface that sFlow considers
1475 * to be "internal" to the switch as a whole, and therefore not an
1476 * candidate for counter polling. */
1477 port->odp_port = ODPP_NONE;
1478 ofport_update_peer(port);
1482 error = dpif_port_query_by_name(ofproto->backer->dpif,
1483 netdev_vport_get_dpif_port(netdev, namebuf,
1490 port->odp_port = dpif_port.port_no;
1492 if (netdev_get_tunnel_config(netdev)) {
1493 tnl_port_add(port, port->up.netdev, port->odp_port);
1494 port->is_tunnel = true;
1496 /* Sanity-check that a mapping doesn't already exist. This
1497 * shouldn't happen for non-tunnel ports. */
1498 if (odp_port_to_ofp_port(ofproto, port->odp_port) != OFPP_NONE) {
1499 VLOG_ERR("port %s already has an OpenFlow port number",
1501 dpif_port_destroy(&dpif_port);
1505 hmap_insert(&ofproto->backer->odp_to_ofport_map, &port->odp_port_node,
1506 hash_odp_port(port->odp_port));
1508 dpif_port_destroy(&dpif_port);
1510 if (ofproto->sflow) {
1511 dpif_sflow_add_port(ofproto->sflow, port_, port->odp_port);
1518 port_destruct(struct ofport *port_)
1520 struct ofport_dpif *port = ofport_dpif_cast(port_);
1521 struct ofproto_dpif *ofproto = ofproto_dpif_cast(port->up.ofproto);
1522 const char *devname = netdev_get_name(port->up.netdev);
1523 char namebuf[NETDEV_VPORT_NAME_BUFSIZE];
1524 const char *dp_port_name;
1526 ofproto->backer->need_revalidate = REV_RECONFIGURE;
1528 dp_port_name = netdev_vport_get_dpif_port(port->up.netdev, namebuf,
1530 if (dpif_port_exists(ofproto->backer->dpif, dp_port_name)) {
1531 /* The underlying device is still there, so delete it. This
1532 * happens when the ofproto is being destroyed, since the caller
1533 * assumes that removal of attached ports will happen as part of
1535 if (!port->is_tunnel) {
1536 dpif_port_del(ofproto->backer->dpif, port->odp_port);
1541 port->peer->peer = NULL;
1545 if (port->odp_port != ODPP_NONE && !port->is_tunnel) {
1546 hmap_remove(&ofproto->backer->odp_to_ofport_map, &port->odp_port_node);
1550 sset_find_and_delete(&ofproto->ports, devname);
1551 sset_find_and_delete(&ofproto->ghost_ports, devname);
1552 bundle_remove(port_);
1553 set_cfm(port_, NULL);
1554 set_bfd(port_, NULL);
1555 if (ofproto->sflow) {
1556 dpif_sflow_del_port(ofproto->sflow, port->odp_port);
1559 ofport_clear_priorities(port);
1560 hmap_destroy(&port->priorities);
1564 port_modified(struct ofport *port_)
1566 struct ofport_dpif *port = ofport_dpif_cast(port_);
1568 if (port->bundle && port->bundle->bond) {
1569 bond_slave_set_netdev(port->bundle->bond, port, port->up.netdev);
1573 cfm_set_netdev(port->cfm, port->up.netdev);
1576 if (port->is_tunnel && tnl_port_reconfigure(port, port->up.netdev,
1578 ofproto_dpif_cast(port->up.ofproto)->backer->need_revalidate =
1582 ofport_update_peer(port);
1586 port_reconfigured(struct ofport *port_, enum ofputil_port_config old_config)
1588 struct ofport_dpif *port = ofport_dpif_cast(port_);
1589 struct ofproto_dpif *ofproto = ofproto_dpif_cast(port->up.ofproto);
1590 enum ofputil_port_config changed = old_config ^ port->up.pp.config;
1592 if (changed & (OFPUTIL_PC_NO_RECV | OFPUTIL_PC_NO_RECV_STP |
1593 OFPUTIL_PC_NO_FWD | OFPUTIL_PC_NO_FLOOD |
1594 OFPUTIL_PC_NO_PACKET_IN)) {
1595 ofproto->backer->need_revalidate = REV_RECONFIGURE;
1597 if (changed & OFPUTIL_PC_NO_FLOOD && port->bundle) {
1598 bundle_update(port->bundle);
1604 set_sflow(struct ofproto *ofproto_,
1605 const struct ofproto_sflow_options *sflow_options)
1607 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1608 struct dpif_sflow *ds = ofproto->sflow;
1610 if (sflow_options) {
1612 struct ofport_dpif *ofport;
1614 ds = ofproto->sflow = dpif_sflow_create();
1615 HMAP_FOR_EACH (ofport, up.hmap_node, &ofproto->up.ports) {
1616 dpif_sflow_add_port(ds, &ofport->up, ofport->odp_port);
1618 ofproto->backer->need_revalidate = REV_RECONFIGURE;
1620 dpif_sflow_set_options(ds, sflow_options);
1623 dpif_sflow_unref(ds);
1624 ofproto->backer->need_revalidate = REV_RECONFIGURE;
1625 ofproto->sflow = NULL;
1633 struct ofproto *ofproto_,
1634 const struct ofproto_ipfix_bridge_exporter_options *bridge_exporter_options,
1635 const struct ofproto_ipfix_flow_exporter_options *flow_exporters_options,
1636 size_t n_flow_exporters_options)
1638 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1639 struct dpif_ipfix *di = ofproto->ipfix;
1641 if (bridge_exporter_options || flow_exporters_options) {
1643 di = ofproto->ipfix = dpif_ipfix_create();
1645 dpif_ipfix_set_options(
1646 di, bridge_exporter_options, flow_exporters_options,
1647 n_flow_exporters_options);
1650 dpif_ipfix_unref(di);
1651 ofproto->ipfix = NULL;
1658 set_cfm(struct ofport *ofport_, const struct cfm_settings *s)
1660 struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
1667 struct ofproto_dpif *ofproto;
1669 ofproto = ofproto_dpif_cast(ofport->up.ofproto);
1670 ofproto->backer->need_revalidate = REV_RECONFIGURE;
1671 ofport->cfm = cfm_create(ofport->up.netdev);
1674 if (cfm_configure(ofport->cfm, s)) {
1680 cfm_unref(ofport->cfm);
1686 get_cfm_status(const struct ofport *ofport_,
1687 struct ofproto_cfm_status *status)
1689 struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
1692 status->faults = cfm_get_fault(ofport->cfm);
1693 status->remote_opstate = cfm_get_opup(ofport->cfm);
1694 status->health = cfm_get_health(ofport->cfm);
1695 cfm_get_remote_mpids(ofport->cfm, &status->rmps, &status->n_rmps);
1703 set_bfd(struct ofport *ofport_, const struct smap *cfg)
1705 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport_->ofproto);
1706 struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
1710 ofport->bfd = bfd_configure(old, netdev_get_name(ofport->up.netdev), cfg);
1711 if (ofport->bfd != old) {
1712 ofproto->backer->need_revalidate = REV_RECONFIGURE;
1719 get_bfd_status(struct ofport *ofport_, struct smap *smap)
1721 struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
1724 bfd_get_status(ofport->bfd, smap);
1731 /* Spanning Tree. */
1734 send_bpdu_cb(struct ofpbuf *pkt, int port_num, void *ofproto_)
1736 struct ofproto_dpif *ofproto = ofproto_;
1737 struct stp_port *sp = stp_get_port(ofproto->stp, port_num);
1738 struct ofport_dpif *ofport;
1740 ofport = stp_port_get_aux(sp);
1742 VLOG_WARN_RL(&rl, "%s: cannot send BPDU on unknown port %d",
1743 ofproto->up.name, port_num);
1745 struct eth_header *eth = pkt->l2;
1747 netdev_get_etheraddr(ofport->up.netdev, eth->eth_src);
1748 if (eth_addr_is_zero(eth->eth_src)) {
1749 VLOG_WARN_RL(&rl, "%s: cannot send BPDU on port %d "
1750 "with unknown MAC", ofproto->up.name, port_num);
1752 send_packet(ofport, pkt);
1758 /* Configures STP on 'ofproto_' using the settings defined in 's'. */
1760 set_stp(struct ofproto *ofproto_, const struct ofproto_stp_settings *s)
1762 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1764 /* Only revalidate flows if the configuration changed. */
1765 if (!s != !ofproto->stp) {
1766 ofproto->backer->need_revalidate = REV_RECONFIGURE;
1770 if (!ofproto->stp) {
1771 ofproto->stp = stp_create(ofproto_->name, s->system_id,
1772 send_bpdu_cb, ofproto);
1773 ofproto->stp_last_tick = time_msec();
1776 stp_set_bridge_id(ofproto->stp, s->system_id);
1777 stp_set_bridge_priority(ofproto->stp, s->priority);
1778 stp_set_hello_time(ofproto->stp, s->hello_time);
1779 stp_set_max_age(ofproto->stp, s->max_age);
1780 stp_set_forward_delay(ofproto->stp, s->fwd_delay);
1782 struct ofport *ofport;
1784 HMAP_FOR_EACH (ofport, hmap_node, &ofproto->up.ports) {
1785 set_stp_port(ofport, NULL);
1788 stp_destroy(ofproto->stp);
1789 ofproto->stp = NULL;
1796 get_stp_status(struct ofproto *ofproto_, struct ofproto_stp_status *s)
1798 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1802 s->bridge_id = stp_get_bridge_id(ofproto->stp);
1803 s->designated_root = stp_get_designated_root(ofproto->stp);
1804 s->root_path_cost = stp_get_root_path_cost(ofproto->stp);
1813 update_stp_port_state(struct ofport_dpif *ofport)
1815 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport->up.ofproto);
1816 enum stp_state state;
1818 /* Figure out new state. */
1819 state = ofport->stp_port ? stp_port_get_state(ofport->stp_port)
1823 if (ofport->stp_state != state) {
1824 enum ofputil_port_state of_state;
1827 VLOG_DBG_RL(&rl, "port %s: STP state changed from %s to %s",
1828 netdev_get_name(ofport->up.netdev),
1829 stp_state_name(ofport->stp_state),
1830 stp_state_name(state));
1831 if (stp_learn_in_state(ofport->stp_state)
1832 != stp_learn_in_state(state)) {
1833 /* xxx Learning action flows should also be flushed. */
1834 mac_learning_flush(ofproto->ml,
1835 &ofproto->backer->revalidate_set);
1837 fwd_change = stp_forward_in_state(ofport->stp_state)
1838 != stp_forward_in_state(state);
1840 ofproto->backer->need_revalidate = REV_STP;
1841 ofport->stp_state = state;
1842 ofport->stp_state_entered = time_msec();
1844 if (fwd_change && ofport->bundle) {
1845 bundle_update(ofport->bundle);
1848 /* Update the STP state bits in the OpenFlow port description. */
1849 of_state = ofport->up.pp.state & ~OFPUTIL_PS_STP_MASK;
1850 of_state |= (state == STP_LISTENING ? OFPUTIL_PS_STP_LISTEN
1851 : state == STP_LEARNING ? OFPUTIL_PS_STP_LEARN
1852 : state == STP_FORWARDING ? OFPUTIL_PS_STP_FORWARD
1853 : state == STP_BLOCKING ? OFPUTIL_PS_STP_BLOCK
1855 ofproto_port_set_state(&ofport->up, of_state);
1859 /* Configures STP on 'ofport_' using the settings defined in 's'. The
1860 * caller is responsible for assigning STP port numbers and ensuring
1861 * there are no duplicates. */
1863 set_stp_port(struct ofport *ofport_,
1864 const struct ofproto_port_stp_settings *s)
1866 struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
1867 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport->up.ofproto);
1868 struct stp_port *sp = ofport->stp_port;
1870 if (!s || !s->enable) {
1872 ofport->stp_port = NULL;
1873 stp_port_disable(sp);
1874 update_stp_port_state(ofport);
1877 } else if (sp && stp_port_no(sp) != s->port_num
1878 && ofport == stp_port_get_aux(sp)) {
1879 /* The port-id changed, so disable the old one if it's not
1880 * already in use by another port. */
1881 stp_port_disable(sp);
1884 sp = ofport->stp_port = stp_get_port(ofproto->stp, s->port_num);
1885 stp_port_enable(sp);
1887 stp_port_set_aux(sp, ofport);
1888 stp_port_set_priority(sp, s->priority);
1889 stp_port_set_path_cost(sp, s->path_cost);
1891 update_stp_port_state(ofport);
1897 get_stp_port_status(struct ofport *ofport_,
1898 struct ofproto_port_stp_status *s)
1900 struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
1901 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport->up.ofproto);
1902 struct stp_port *sp = ofport->stp_port;
1904 if (!ofproto->stp || !sp) {
1910 s->port_id = stp_port_get_id(sp);
1911 s->state = stp_port_get_state(sp);
1912 s->sec_in_state = (time_msec() - ofport->stp_state_entered) / 1000;
1913 s->role = stp_port_get_role(sp);
1914 stp_port_get_counts(sp, &s->tx_count, &s->rx_count, &s->error_count);
1920 stp_run(struct ofproto_dpif *ofproto)
1923 long long int now = time_msec();
1924 long long int elapsed = now - ofproto->stp_last_tick;
1925 struct stp_port *sp;
1928 stp_tick(ofproto->stp, MIN(INT_MAX, elapsed));
1929 ofproto->stp_last_tick = now;
1931 while (stp_get_changed_port(ofproto->stp, &sp)) {
1932 struct ofport_dpif *ofport = stp_port_get_aux(sp);
1935 update_stp_port_state(ofport);
1939 if (stp_check_and_reset_fdb_flush(ofproto->stp)) {
1940 mac_learning_flush(ofproto->ml, &ofproto->backer->revalidate_set);
1946 stp_wait(struct ofproto_dpif *ofproto)
1949 poll_timer_wait(1000);
1953 /* Returns true if STP should process 'flow'. Sets fields in 'wc' that
1954 * were used to make the determination.*/
1956 stp_should_process_flow(const struct flow *flow, struct flow_wildcards *wc)
1958 memset(&wc->masks.dl_dst, 0xff, sizeof wc->masks.dl_dst);
1959 return eth_addr_equals(flow->dl_dst, eth_addr_stp);
1963 stp_process_packet(const struct ofport_dpif *ofport,
1964 const struct ofpbuf *packet)
1966 struct ofpbuf payload = *packet;
1967 struct eth_header *eth = payload.data;
1968 struct stp_port *sp = ofport->stp_port;
1970 /* Sink packets on ports that have STP disabled when the bridge has
1972 if (!sp || stp_port_get_state(sp) == STP_DISABLED) {
1976 /* Trim off padding on payload. */
1977 if (payload.size > ntohs(eth->eth_type) + ETH_HEADER_LEN) {
1978 payload.size = ntohs(eth->eth_type) + ETH_HEADER_LEN;
1981 if (ofpbuf_try_pull(&payload, ETH_HEADER_LEN + LLC_HEADER_LEN)) {
1982 stp_received_bpdu(sp, payload.data, payload.size);
1987 ofproto_dpif_queue_to_priority(const struct ofproto_dpif *ofproto,
1988 uint32_t queue_id, uint32_t *priority)
1990 return dpif_queue_to_priority(ofproto->backer->dpif, queue_id, priority);
1993 static struct priority_to_dscp *
1994 get_priority(const struct ofport_dpif *ofport, uint32_t priority)
1996 struct priority_to_dscp *pdscp;
1999 hash = hash_int(priority, 0);
2000 HMAP_FOR_EACH_IN_BUCKET (pdscp, hmap_node, hash, &ofport->priorities) {
2001 if (pdscp->priority == priority) {
2009 ofproto_dpif_dscp_from_priority(const struct ofport_dpif *ofport,
2010 uint32_t priority, uint8_t *dscp)
2012 struct priority_to_dscp *pdscp = get_priority(ofport, priority);
2013 *dscp = pdscp ? pdscp->dscp : 0;
2014 return pdscp != NULL;
2018 ofport_clear_priorities(struct ofport_dpif *ofport)
2020 struct priority_to_dscp *pdscp, *next;
2022 HMAP_FOR_EACH_SAFE (pdscp, next, hmap_node, &ofport->priorities) {
2023 hmap_remove(&ofport->priorities, &pdscp->hmap_node);
2029 set_queues(struct ofport *ofport_,
2030 const struct ofproto_port_queue *qdscp_list,
2033 struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
2034 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport->up.ofproto);
2035 struct hmap new = HMAP_INITIALIZER(&new);
2038 for (i = 0; i < n_qdscp; i++) {
2039 struct priority_to_dscp *pdscp;
2043 dscp = (qdscp_list[i].dscp << 2) & IP_DSCP_MASK;
2044 if (dpif_queue_to_priority(ofproto->backer->dpif, qdscp_list[i].queue,
2049 pdscp = get_priority(ofport, priority);
2051 hmap_remove(&ofport->priorities, &pdscp->hmap_node);
2053 pdscp = xmalloc(sizeof *pdscp);
2054 pdscp->priority = priority;
2056 ofproto->backer->need_revalidate = REV_RECONFIGURE;
2059 if (pdscp->dscp != dscp) {
2061 ofproto->backer->need_revalidate = REV_RECONFIGURE;
2064 hmap_insert(&new, &pdscp->hmap_node, hash_int(pdscp->priority, 0));
2067 if (!hmap_is_empty(&ofport->priorities)) {
2068 ofport_clear_priorities(ofport);
2069 ofproto->backer->need_revalidate = REV_RECONFIGURE;
2072 hmap_swap(&new, &ofport->priorities);
2080 /* Expires all MAC learning entries associated with 'bundle' and forces its
2081 * ofproto to revalidate every flow.
2083 * Normally MAC learning entries are removed only from the ofproto associated
2084 * with 'bundle', but if 'all_ofprotos' is true, then the MAC learning entries
2085 * are removed from every ofproto. When patch ports and SLB bonds are in use
2086 * and a VM migration happens and the gratuitous ARPs are somehow lost, this
2087 * avoids a MAC_ENTRY_IDLE_TIME delay before the migrated VM can communicate
2088 * with the host from which it migrated. */
2090 bundle_flush_macs(struct ofbundle *bundle, bool all_ofprotos)
2092 struct ofproto_dpif *ofproto = bundle->ofproto;
2093 struct mac_learning *ml = ofproto->ml;
2094 struct mac_entry *mac, *next_mac;
2096 ofproto->backer->need_revalidate = REV_RECONFIGURE;
2097 LIST_FOR_EACH_SAFE (mac, next_mac, lru_node, &ml->lrus) {
2098 if (mac->port.p == bundle) {
2100 struct ofproto_dpif *o;
2102 HMAP_FOR_EACH (o, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
2104 struct mac_entry *e;
2106 e = mac_learning_lookup(o->ml, mac->mac, mac->vlan,
2109 mac_learning_expire(o->ml, e);
2115 mac_learning_expire(ml, mac);
2120 static struct ofbundle *
2121 bundle_lookup(const struct ofproto_dpif *ofproto, void *aux)
2123 struct ofbundle *bundle;
2125 HMAP_FOR_EACH_IN_BUCKET (bundle, hmap_node, hash_pointer(aux, 0),
2126 &ofproto->bundles) {
2127 if (bundle->aux == aux) {
2134 /* Looks up each of the 'n_auxes' pointers in 'auxes' as bundles and adds the
2135 * ones that are found to 'bundles'. */
2137 bundle_lookup_multiple(struct ofproto_dpif *ofproto,
2138 void **auxes, size_t n_auxes,
2139 struct hmapx *bundles)
2143 hmapx_init(bundles);
2144 for (i = 0; i < n_auxes; i++) {
2145 struct ofbundle *bundle = bundle_lookup(ofproto, auxes[i]);
2147 hmapx_add(bundles, bundle);
2153 bundle_update(struct ofbundle *bundle)
2155 struct ofport_dpif *port;
2157 bundle->floodable = true;
2158 LIST_FOR_EACH (port, bundle_node, &bundle->ports) {
2159 if (port->up.pp.config & OFPUTIL_PC_NO_FLOOD
2160 || !stp_forward_in_state(port->stp_state)) {
2161 bundle->floodable = false;
2168 bundle_del_port(struct ofport_dpif *port)
2170 struct ofbundle *bundle = port->bundle;
2172 bundle->ofproto->backer->need_revalidate = REV_RECONFIGURE;
2174 list_remove(&port->bundle_node);
2175 port->bundle = NULL;
2178 lacp_slave_unregister(bundle->lacp, port);
2181 bond_slave_unregister(bundle->bond, port);
2184 bundle_update(bundle);
2188 bundle_add_port(struct ofbundle *bundle, ofp_port_t ofp_port,
2189 struct lacp_slave_settings *lacp)
2191 struct ofport_dpif *port;
2193 port = get_ofp_port(bundle->ofproto, ofp_port);
2198 if (port->bundle != bundle) {
2199 bundle->ofproto->backer->need_revalidate = REV_RECONFIGURE;
2201 bundle_del_port(port);
2204 port->bundle = bundle;
2205 list_push_back(&bundle->ports, &port->bundle_node);
2206 if (port->up.pp.config & OFPUTIL_PC_NO_FLOOD
2207 || !stp_forward_in_state(port->stp_state)) {
2208 bundle->floodable = false;
2212 bundle->ofproto->backer->need_revalidate = REV_RECONFIGURE;
2213 lacp_slave_register(bundle->lacp, port, lacp);
2220 bundle_destroy(struct ofbundle *bundle)
2222 struct ofproto_dpif *ofproto;
2223 struct ofport_dpif *port, *next_port;
2230 ofproto = bundle->ofproto;
2231 for (i = 0; i < MAX_MIRRORS; i++) {
2232 struct ofmirror *m = ofproto->mirrors[i];
2234 if (m->out == bundle) {
2236 } else if (hmapx_find_and_delete(&m->srcs, bundle)
2237 || hmapx_find_and_delete(&m->dsts, bundle)) {
2238 ofproto->backer->need_revalidate = REV_RECONFIGURE;
2243 LIST_FOR_EACH_SAFE (port, next_port, bundle_node, &bundle->ports) {
2244 bundle_del_port(port);
2247 bundle_flush_macs(bundle, true);
2248 hmap_remove(&ofproto->bundles, &bundle->hmap_node);
2250 free(bundle->trunks);
2251 lacp_unref(bundle->lacp);
2252 bond_unref(bundle->bond);
2257 bundle_set(struct ofproto *ofproto_, void *aux,
2258 const struct ofproto_bundle_settings *s)
2260 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
2261 bool need_flush = false;
2262 struct ofport_dpif *port;
2263 struct ofbundle *bundle;
2264 unsigned long *trunks;
2270 bundle_destroy(bundle_lookup(ofproto, aux));
2274 ovs_assert(s->n_slaves == 1 || s->bond != NULL);
2275 ovs_assert((s->lacp != NULL) == (s->lacp_slaves != NULL));
2277 bundle = bundle_lookup(ofproto, aux);
2279 bundle = xmalloc(sizeof *bundle);
2281 bundle->ofproto = ofproto;
2282 hmap_insert(&ofproto->bundles, &bundle->hmap_node,
2283 hash_pointer(aux, 0));
2285 bundle->name = NULL;
2287 list_init(&bundle->ports);
2288 bundle->vlan_mode = PORT_VLAN_TRUNK;
2290 bundle->trunks = NULL;
2291 bundle->use_priority_tags = s->use_priority_tags;
2292 bundle->lacp = NULL;
2293 bundle->bond = NULL;
2295 bundle->floodable = true;
2297 bundle->src_mirrors = 0;
2298 bundle->dst_mirrors = 0;
2299 bundle->mirror_out = 0;
2302 if (!bundle->name || strcmp(s->name, bundle->name)) {
2304 bundle->name = xstrdup(s->name);
2309 if (!bundle->lacp) {
2310 ofproto->backer->need_revalidate = REV_RECONFIGURE;
2311 bundle->lacp = lacp_create();
2313 lacp_configure(bundle->lacp, s->lacp);
2315 lacp_unref(bundle->lacp);
2316 bundle->lacp = NULL;
2319 /* Update set of ports. */
2321 for (i = 0; i < s->n_slaves; i++) {
2322 if (!bundle_add_port(bundle, s->slaves[i],
2323 s->lacp ? &s->lacp_slaves[i] : NULL)) {
2327 if (!ok || list_size(&bundle->ports) != s->n_slaves) {
2328 struct ofport_dpif *next_port;
2330 LIST_FOR_EACH_SAFE (port, next_port, bundle_node, &bundle->ports) {
2331 for (i = 0; i < s->n_slaves; i++) {
2332 if (s->slaves[i] == port->up.ofp_port) {
2337 bundle_del_port(port);
2341 ovs_assert(list_size(&bundle->ports) <= s->n_slaves);
2343 if (list_is_empty(&bundle->ports)) {
2344 bundle_destroy(bundle);
2348 /* Set VLAN tagging mode */
2349 if (s->vlan_mode != bundle->vlan_mode
2350 || s->use_priority_tags != bundle->use_priority_tags) {
2351 bundle->vlan_mode = s->vlan_mode;
2352 bundle->use_priority_tags = s->use_priority_tags;
2357 vlan = (s->vlan_mode == PORT_VLAN_TRUNK ? -1
2358 : s->vlan >= 0 && s->vlan <= 4095 ? s->vlan
2360 if (vlan != bundle->vlan) {
2361 bundle->vlan = vlan;
2365 /* Get trunked VLANs. */
2366 switch (s->vlan_mode) {
2367 case PORT_VLAN_ACCESS:
2371 case PORT_VLAN_TRUNK:
2372 trunks = CONST_CAST(unsigned long *, s->trunks);
2375 case PORT_VLAN_NATIVE_UNTAGGED:
2376 case PORT_VLAN_NATIVE_TAGGED:
2377 if (vlan != 0 && (!s->trunks
2378 || !bitmap_is_set(s->trunks, vlan)
2379 || bitmap_is_set(s->trunks, 0))) {
2380 /* Force trunking the native VLAN and prohibit trunking VLAN 0. */
2382 trunks = bitmap_clone(s->trunks, 4096);
2384 trunks = bitmap_allocate1(4096);
2386 bitmap_set1(trunks, vlan);
2387 bitmap_set0(trunks, 0);
2389 trunks = CONST_CAST(unsigned long *, s->trunks);
2396 if (!vlan_bitmap_equal(trunks, bundle->trunks)) {
2397 free(bundle->trunks);
2398 if (trunks == s->trunks) {
2399 bundle->trunks = vlan_bitmap_clone(trunks);
2401 bundle->trunks = trunks;
2406 if (trunks != s->trunks) {
2411 if (!list_is_short(&bundle->ports)) {
2412 bundle->ofproto->has_bonded_bundles = true;
2414 if (bond_reconfigure(bundle->bond, s->bond)) {
2415 ofproto->backer->need_revalidate = REV_RECONFIGURE;
2418 bundle->bond = bond_create(s->bond);
2419 ofproto->backer->need_revalidate = REV_RECONFIGURE;
2422 LIST_FOR_EACH (port, bundle_node, &bundle->ports) {
2423 bond_slave_register(bundle->bond, port, port->up.netdev);
2426 bond_unref(bundle->bond);
2427 bundle->bond = NULL;
2430 /* If we changed something that would affect MAC learning, un-learn
2431 * everything on this port and force flow revalidation. */
2433 bundle_flush_macs(bundle, false);
2440 bundle_remove(struct ofport *port_)
2442 struct ofport_dpif *port = ofport_dpif_cast(port_);
2443 struct ofbundle *bundle = port->bundle;
2446 bundle_del_port(port);
2447 if (list_is_empty(&bundle->ports)) {
2448 bundle_destroy(bundle);
2449 } else if (list_is_short(&bundle->ports)) {
2450 bond_unref(bundle->bond);
2451 bundle->bond = NULL;
2457 send_pdu_cb(void *port_, const void *pdu, size_t pdu_size)
2459 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 10);
2460 struct ofport_dpif *port = port_;
2461 uint8_t ea[ETH_ADDR_LEN];
2464 error = netdev_get_etheraddr(port->up.netdev, ea);
2466 struct ofpbuf packet;
2469 ofpbuf_init(&packet, 0);
2470 packet_pdu = eth_compose(&packet, eth_addr_lacp, ea, ETH_TYPE_LACP,
2472 memcpy(packet_pdu, pdu, pdu_size);
2474 send_packet(port, &packet);
2475 ofpbuf_uninit(&packet);
2477 VLOG_ERR_RL(&rl, "port %s: cannot obtain Ethernet address of iface "
2478 "%s (%s)", port->bundle->name,
2479 netdev_get_name(port->up.netdev), ovs_strerror(error));
2484 bundle_send_learning_packets(struct ofbundle *bundle)
2486 struct ofproto_dpif *ofproto = bundle->ofproto;
2487 int error, n_packets, n_errors;
2488 struct mac_entry *e;
2490 error = n_packets = n_errors = 0;
2491 LIST_FOR_EACH (e, lru_node, &ofproto->ml->lrus) {
2492 if (e->port.p != bundle) {
2493 struct ofpbuf *learning_packet;
2494 struct ofport_dpif *port;
2498 /* The assignment to "port" is unnecessary but makes "grep"ing for
2499 * struct ofport_dpif more effective. */
2500 learning_packet = bond_compose_learning_packet(bundle->bond,
2504 ret = send_packet(port, learning_packet);
2505 ofpbuf_delete(learning_packet);
2515 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
2516 VLOG_WARN_RL(&rl, "bond %s: %d errors sending %d gratuitous learning "
2517 "packets, last error was: %s",
2518 bundle->name, n_errors, n_packets, ovs_strerror(error));
2520 VLOG_DBG("bond %s: sent %d gratuitous learning packets",
2521 bundle->name, n_packets);
2526 bundle_run(struct ofbundle *bundle)
2529 lacp_run(bundle->lacp, send_pdu_cb);
2532 struct ofport_dpif *port;
2534 LIST_FOR_EACH (port, bundle_node, &bundle->ports) {
2535 bond_slave_set_may_enable(bundle->bond, port, port->may_enable);
2538 bond_run(bundle->bond, &bundle->ofproto->backer->revalidate_set,
2539 lacp_status(bundle->lacp));
2540 if (bond_should_send_learning_packets(bundle->bond)) {
2541 bundle_send_learning_packets(bundle);
2547 bundle_wait(struct ofbundle *bundle)
2550 lacp_wait(bundle->lacp);
2553 bond_wait(bundle->bond);
2560 mirror_scan(struct ofproto_dpif *ofproto)
2564 for (idx = 0; idx < MAX_MIRRORS; idx++) {
2565 if (!ofproto->mirrors[idx]) {
2572 static struct ofmirror *
2573 mirror_lookup(struct ofproto_dpif *ofproto, void *aux)
2577 for (i = 0; i < MAX_MIRRORS; i++) {
2578 struct ofmirror *mirror = ofproto->mirrors[i];
2579 if (mirror && mirror->aux == aux) {
2587 /* Update the 'dup_mirrors' member of each of the ofmirrors in 'ofproto'. */
2589 mirror_update_dups(struct ofproto_dpif *ofproto)
2593 for (i = 0; i < MAX_MIRRORS; i++) {
2594 struct ofmirror *m = ofproto->mirrors[i];
2597 m->dup_mirrors = MIRROR_MASK_C(1) << i;
2601 for (i = 0; i < MAX_MIRRORS; i++) {
2602 struct ofmirror *m1 = ofproto->mirrors[i];
2609 for (j = i + 1; j < MAX_MIRRORS; j++) {
2610 struct ofmirror *m2 = ofproto->mirrors[j];
2612 if (m2 && m1->out == m2->out && m1->out_vlan == m2->out_vlan) {
2613 m1->dup_mirrors |= MIRROR_MASK_C(1) << j;
2614 m2->dup_mirrors |= m1->dup_mirrors;
2621 mirror_set(struct ofproto *ofproto_, void *aux,
2622 const struct ofproto_mirror_settings *s)
2624 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
2625 mirror_mask_t mirror_bit;
2626 struct ofbundle *bundle;
2627 struct ofmirror *mirror;
2628 struct ofbundle *out;
2629 struct hmapx srcs; /* Contains "struct ofbundle *"s. */
2630 struct hmapx dsts; /* Contains "struct ofbundle *"s. */
2633 mirror = mirror_lookup(ofproto, aux);
2635 mirror_destroy(mirror);
2641 idx = mirror_scan(ofproto);
2643 VLOG_WARN("bridge %s: maximum of %d port mirrors reached, "
2645 ofproto->up.name, MAX_MIRRORS, s->name);
2649 mirror = ofproto->mirrors[idx] = xzalloc(sizeof *mirror);
2650 mirror->ofproto = ofproto;
2653 mirror->out_vlan = -1;
2654 mirror->name = NULL;
2657 if (!mirror->name || strcmp(s->name, mirror->name)) {
2659 mirror->name = xstrdup(s->name);
2662 /* Get the new configuration. */
2663 if (s->out_bundle) {
2664 out = bundle_lookup(ofproto, s->out_bundle);
2666 mirror_destroy(mirror);
2672 out_vlan = s->out_vlan;
2674 bundle_lookup_multiple(ofproto, s->srcs, s->n_srcs, &srcs);
2675 bundle_lookup_multiple(ofproto, s->dsts, s->n_dsts, &dsts);
2677 /* If the configuration has not changed, do nothing. */
2678 if (hmapx_equals(&srcs, &mirror->srcs)
2679 && hmapx_equals(&dsts, &mirror->dsts)
2680 && vlan_bitmap_equal(mirror->vlans, s->src_vlans)
2681 && mirror->out == out
2682 && mirror->out_vlan == out_vlan)
2684 hmapx_destroy(&srcs);
2685 hmapx_destroy(&dsts);
2689 hmapx_swap(&srcs, &mirror->srcs);
2690 hmapx_destroy(&srcs);
2692 hmapx_swap(&dsts, &mirror->dsts);
2693 hmapx_destroy(&dsts);
2695 free(mirror->vlans);
2696 mirror->vlans = vlan_bitmap_clone(s->src_vlans);
2699 mirror->out_vlan = out_vlan;
2701 /* Update bundles. */
2702 mirror_bit = MIRROR_MASK_C(1) << mirror->idx;
2703 HMAP_FOR_EACH (bundle, hmap_node, &mirror->ofproto->bundles) {
2704 if (hmapx_contains(&mirror->srcs, bundle)) {
2705 bundle->src_mirrors |= mirror_bit;
2707 bundle->src_mirrors &= ~mirror_bit;
2710 if (hmapx_contains(&mirror->dsts, bundle)) {
2711 bundle->dst_mirrors |= mirror_bit;
2713 bundle->dst_mirrors &= ~mirror_bit;
2716 if (mirror->out == bundle) {
2717 bundle->mirror_out |= mirror_bit;
2719 bundle->mirror_out &= ~mirror_bit;
2723 ofproto->backer->need_revalidate = REV_RECONFIGURE;
2724 ofproto->has_mirrors = true;
2725 mac_learning_flush(ofproto->ml,
2726 &ofproto->backer->revalidate_set);
2727 mirror_update_dups(ofproto);
2733 mirror_destroy(struct ofmirror *mirror)
2735 struct ofproto_dpif *ofproto;
2736 mirror_mask_t mirror_bit;
2737 struct ofbundle *bundle;
2744 ofproto = mirror->ofproto;
2745 ofproto->backer->need_revalidate = REV_RECONFIGURE;
2746 mac_learning_flush(ofproto->ml, &ofproto->backer->revalidate_set);
2748 mirror_bit = MIRROR_MASK_C(1) << mirror->idx;
2749 HMAP_FOR_EACH (bundle, hmap_node, &ofproto->bundles) {
2750 bundle->src_mirrors &= ~mirror_bit;
2751 bundle->dst_mirrors &= ~mirror_bit;
2752 bundle->mirror_out &= ~mirror_bit;
2755 hmapx_destroy(&mirror->srcs);
2756 hmapx_destroy(&mirror->dsts);
2757 free(mirror->vlans);
2759 ofproto->mirrors[mirror->idx] = NULL;
2763 mirror_update_dups(ofproto);
2765 ofproto->has_mirrors = false;
2766 for (i = 0; i < MAX_MIRRORS; i++) {
2767 if (ofproto->mirrors[i]) {
2768 ofproto->has_mirrors = true;
2775 mirror_get_stats(struct ofproto *ofproto_, void *aux,
2776 uint64_t *packets, uint64_t *bytes)
2778 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
2779 struct ofmirror *mirror = mirror_lookup(ofproto, aux);
2782 *packets = *bytes = UINT64_MAX;
2788 *packets = mirror->packet_count;
2789 *bytes = mirror->byte_count;
2795 set_flood_vlans(struct ofproto *ofproto_, unsigned long *flood_vlans)
2797 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
2798 if (mac_learning_set_flood_vlans(ofproto->ml, flood_vlans)) {
2799 mac_learning_flush(ofproto->ml, &ofproto->backer->revalidate_set);
2805 is_mirror_output_bundle(const struct ofproto *ofproto_, void *aux)
2807 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
2808 struct ofbundle *bundle = bundle_lookup(ofproto, aux);
2809 return bundle && bundle->mirror_out != 0;
2813 forward_bpdu_changed(struct ofproto *ofproto_)
2815 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
2816 ofproto->backer->need_revalidate = REV_RECONFIGURE;
2820 set_mac_table_config(struct ofproto *ofproto_, unsigned int idle_time,
2823 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
2824 mac_learning_set_idle_time(ofproto->ml, idle_time);
2825 mac_learning_set_max_entries(ofproto->ml, max_entries);
2830 struct ofport_dpif *
2831 get_ofp_port(const struct ofproto_dpif *ofproto, ofp_port_t ofp_port)
2833 struct ofport *ofport = ofproto_get_port(&ofproto->up, ofp_port);
2834 return ofport ? ofport_dpif_cast(ofport) : NULL;
2837 struct ofport_dpif *
2838 get_odp_port(const struct ofproto_dpif *ofproto, odp_port_t odp_port)
2840 struct ofport_dpif *port = odp_port_to_ofport(ofproto->backer, odp_port);
2841 return port && &ofproto->up == port->up.ofproto ? port : NULL;
2845 ofproto_port_from_dpif_port(struct ofproto_dpif *ofproto,
2846 struct ofproto_port *ofproto_port,
2847 struct dpif_port *dpif_port)
2849 ofproto_port->name = dpif_port->name;
2850 ofproto_port->type = dpif_port->type;
2851 ofproto_port->ofp_port = odp_port_to_ofp_port(ofproto, dpif_port->port_no);
2855 ofport_update_peer(struct ofport_dpif *ofport)
2857 const struct ofproto_dpif *ofproto;
2858 struct dpif_backer *backer;
2859 const char *peer_name;
2861 if (!netdev_vport_is_patch(ofport->up.netdev)) {
2865 backer = ofproto_dpif_cast(ofport->up.ofproto)->backer;
2866 backer->need_revalidate = REV_RECONFIGURE;
2869 ofport->peer->peer = NULL;
2870 ofport->peer = NULL;
2873 peer_name = netdev_vport_patch_peer(ofport->up.netdev);
2878 HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
2879 struct ofport *peer_ofport;
2880 struct ofport_dpif *peer;
2881 const char *peer_peer;
2883 if (ofproto->backer != backer) {
2887 peer_ofport = shash_find_data(&ofproto->up.port_by_name, peer_name);
2892 peer = ofport_dpif_cast(peer_ofport);
2893 peer_peer = netdev_vport_patch_peer(peer->up.netdev);
2894 if (peer_peer && !strcmp(netdev_get_name(ofport->up.netdev),
2896 ofport->peer = peer;
2897 ofport->peer->peer = ofport;
2905 port_run_fast(struct ofport_dpif *ofport)
2907 if (ofport->cfm && cfm_should_send_ccm(ofport->cfm)) {
2908 struct ofpbuf packet;
2910 ofpbuf_init(&packet, 0);
2911 cfm_compose_ccm(ofport->cfm, &packet, ofport->up.pp.hw_addr);
2912 send_packet(ofport, &packet);
2913 ofpbuf_uninit(&packet);
2916 if (ofport->bfd && bfd_should_send_packet(ofport->bfd)) {
2917 struct ofpbuf packet;
2919 ofpbuf_init(&packet, 0);
2920 bfd_put_packet(ofport->bfd, &packet, ofport->up.pp.hw_addr);
2921 send_packet(ofport, &packet);
2922 ofpbuf_uninit(&packet);
2927 port_run(struct ofport_dpif *ofport)
2929 long long int carrier_seq = netdev_get_carrier_resets(ofport->up.netdev);
2930 bool carrier_changed = carrier_seq != ofport->carrier_seq;
2931 bool enable = netdev_get_carrier(ofport->up.netdev);
2933 ofport->carrier_seq = carrier_seq;
2935 port_run_fast(ofport);
2938 int cfm_opup = cfm_get_opup(ofport->cfm);
2940 cfm_run(ofport->cfm);
2941 enable = enable && !cfm_get_fault(ofport->cfm);
2943 if (cfm_opup >= 0) {
2944 enable = enable && cfm_opup;
2949 bfd_run(ofport->bfd);
2950 enable = enable && bfd_forwarding(ofport->bfd);
2953 if (ofport->bundle) {
2954 enable = enable && lacp_slave_may_enable(ofport->bundle->lacp, ofport);
2955 if (carrier_changed) {
2956 lacp_slave_carrier_changed(ofport->bundle->lacp, ofport);
2960 if (ofport->may_enable != enable) {
2961 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport->up.ofproto);
2962 ofproto->backer->need_revalidate = REV_PORT_TOGGLED;
2965 ofport->may_enable = enable;
2969 port_wait(struct ofport_dpif *ofport)
2972 cfm_wait(ofport->cfm);
2976 bfd_wait(ofport->bfd);
2981 port_query_by_name(const struct ofproto *ofproto_, const char *devname,
2982 struct ofproto_port *ofproto_port)
2984 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
2985 struct dpif_port dpif_port;
2988 if (sset_contains(&ofproto->ghost_ports, devname)) {
2989 const char *type = netdev_get_type_from_name(devname);
2991 /* We may be called before ofproto->up.port_by_name is populated with
2992 * the appropriate ofport. For this reason, we must get the name and
2993 * type from the netdev layer directly. */
2995 const struct ofport *ofport;
2997 ofport = shash_find_data(&ofproto->up.port_by_name, devname);
2998 ofproto_port->ofp_port = ofport ? ofport->ofp_port : OFPP_NONE;
2999 ofproto_port->name = xstrdup(devname);
3000 ofproto_port->type = xstrdup(type);
3006 if (!sset_contains(&ofproto->ports, devname)) {
3009 error = dpif_port_query_by_name(ofproto->backer->dpif,
3010 devname, &dpif_port);
3012 ofproto_port_from_dpif_port(ofproto, ofproto_port, &dpif_port);
3018 port_add(struct ofproto *ofproto_, struct netdev *netdev)
3020 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
3021 const char *devname = netdev_get_name(netdev);
3022 char namebuf[NETDEV_VPORT_NAME_BUFSIZE];
3023 const char *dp_port_name;
3025 if (netdev_vport_is_patch(netdev)) {
3026 sset_add(&ofproto->ghost_ports, netdev_get_name(netdev));
3030 dp_port_name = netdev_vport_get_dpif_port(netdev, namebuf, sizeof namebuf);
3031 if (!dpif_port_exists(ofproto->backer->dpif, dp_port_name)) {
3032 odp_port_t port_no = ODPP_NONE;
3035 error = dpif_port_add(ofproto->backer->dpif, netdev, &port_no);
3039 if (netdev_get_tunnel_config(netdev)) {
3040 simap_put(&ofproto->backer->tnl_backers,
3041 dp_port_name, odp_to_u32(port_no));
3045 if (netdev_get_tunnel_config(netdev)) {
3046 sset_add(&ofproto->ghost_ports, devname);
3048 sset_add(&ofproto->ports, devname);
3054 port_del(struct ofproto *ofproto_, ofp_port_t ofp_port)
3056 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
3057 struct ofport_dpif *ofport = get_ofp_port(ofproto, ofp_port);
3064 sset_find_and_delete(&ofproto->ghost_ports,
3065 netdev_get_name(ofport->up.netdev));
3066 ofproto->backer->need_revalidate = REV_RECONFIGURE;
3067 if (!ofport->is_tunnel) {
3068 error = dpif_port_del(ofproto->backer->dpif, ofport->odp_port);
3070 /* The caller is going to close ofport->up.netdev. If this is a
3071 * bonded port, then the bond is using that netdev, so remove it
3072 * from the bond. The client will need to reconfigure everything
3073 * after deleting ports, so then the slave will get re-added. */
3074 bundle_remove(&ofport->up);
3081 port_get_stats(const struct ofport *ofport_, struct netdev_stats *stats)
3083 struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
3088 error = netdev_get_stats(ofport->up.netdev, stats);
3090 if (!error && ofport_->ofp_port == OFPP_LOCAL) {
3091 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport->up.ofproto);
3093 /* ofproto->stats.tx_packets represents packets that we created
3094 * internally and sent to some port (e.g. packets sent with
3095 * send_packet()). Account for them as if they had come from
3096 * OFPP_LOCAL and got forwarded. */
3098 if (stats->rx_packets != UINT64_MAX) {
3099 stats->rx_packets += ofproto->stats.tx_packets;
3102 if (stats->rx_bytes != UINT64_MAX) {
3103 stats->rx_bytes += ofproto->stats.tx_bytes;
3106 /* ofproto->stats.rx_packets represents packets that were received on
3107 * some port and we processed internally and dropped (e.g. STP).
3108 * Account for them as if they had been forwarded to OFPP_LOCAL. */
3110 if (stats->tx_packets != UINT64_MAX) {
3111 stats->tx_packets += ofproto->stats.rx_packets;
3114 if (stats->tx_bytes != UINT64_MAX) {
3115 stats->tx_bytes += ofproto->stats.rx_bytes;
3122 struct port_dump_state {
3127 struct ofproto_port port;
3132 port_dump_start(const struct ofproto *ofproto_ OVS_UNUSED, void **statep)
3134 *statep = xzalloc(sizeof(struct port_dump_state));
3139 port_dump_next(const struct ofproto *ofproto_, void *state_,
3140 struct ofproto_port *port)
3142 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
3143 struct port_dump_state *state = state_;
3144 const struct sset *sset;
3145 struct sset_node *node;
3147 if (state->has_port) {
3148 ofproto_port_destroy(&state->port);
3149 state->has_port = false;
3151 sset = state->ghost ? &ofproto->ghost_ports : &ofproto->ports;
3152 while ((node = sset_at_position(sset, &state->bucket, &state->offset))) {
3155 error = port_query_by_name(ofproto_, node->name, &state->port);
3157 *port = state->port;
3158 state->has_port = true;
3160 } else if (error != ENODEV) {
3165 if (!state->ghost) {
3166 state->ghost = true;
3169 return port_dump_next(ofproto_, state_, port);
3176 port_dump_done(const struct ofproto *ofproto_ OVS_UNUSED, void *state_)
3178 struct port_dump_state *state = state_;
3180 if (state->has_port) {
3181 ofproto_port_destroy(&state->port);
3188 port_poll(const struct ofproto *ofproto_, char **devnamep)
3190 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
3192 if (ofproto->port_poll_errno) {
3193 int error = ofproto->port_poll_errno;
3194 ofproto->port_poll_errno = 0;
3198 if (sset_is_empty(&ofproto->port_poll_set)) {
3202 *devnamep = sset_pop(&ofproto->port_poll_set);
3207 port_poll_wait(const struct ofproto *ofproto_)
3209 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
3210 dpif_port_poll_wait(ofproto->backer->dpif);
3214 port_is_lacp_current(const struct ofport *ofport_)
3216 const struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
3217 return (ofport->bundle && ofport->bundle->lacp
3218 ? lacp_slave_is_current(ofport->bundle->lacp, ofport)
3222 /* Upcall handling. */
3224 /* Flow miss batching.
3226 * Some dpifs implement operations faster when you hand them off in a batch.
3227 * To allow batching, "struct flow_miss" queues the dpif-related work needed
3228 * for a given flow. Each "struct flow_miss" corresponds to sending one or
3229 * more packets, plus possibly installing the flow in the dpif.
3231 * So far we only batch the operations that affect flow setup time the most.
3232 * It's possible to batch more than that, but the benefit might be minimal. */
3234 struct hmap_node hmap_node;
3235 struct ofproto_dpif *ofproto;
3237 enum odp_key_fitness key_fitness;
3238 const struct nlattr *key;
3240 struct list packets;
3241 enum dpif_upcall_type upcall_type;
3244 struct flow_miss_op {
3245 struct dpif_op dpif_op;
3247 uint64_t slow_stub[128 / 8]; /* Buffer for compose_slow_path() */
3248 struct xlate_out xout;
3249 bool xout_garbage; /* 'xout' needs to be uninitialized? */
3251 struct ofpbuf mask; /* Flow mask for "put" ops. */
3252 struct odputil_keybuf maskbuf;
3254 /* If this is a "put" op, then a pointer to the subfacet that should
3255 * be marked as uninstalled if the operation fails. */
3256 struct subfacet *subfacet;
3259 /* Sends an OFPT_PACKET_IN message for 'packet' of type OFPR_NO_MATCH to each
3260 * OpenFlow controller as necessary according to their individual
3261 * configurations. */
3263 send_packet_in_miss(struct ofproto_dpif *ofproto, const struct ofpbuf *packet,
3264 const struct flow *flow)
3266 struct ofputil_packet_in pin;
3268 pin.packet = packet->data;
3269 pin.packet_len = packet->size;
3270 pin.reason = OFPR_NO_MATCH;
3271 pin.controller_id = 0;
3276 pin.send_len = 0; /* not used for flow table misses */
3278 flow_get_metadata(flow, &pin.fmd);
3280 connmgr_send_packet_in(ofproto->up.connmgr, &pin);
3283 static struct flow_miss *
3284 flow_miss_find(struct hmap *todo, const struct ofproto_dpif *ofproto,
3285 const struct flow *flow, uint32_t hash)
3287 struct flow_miss *miss;
3289 HMAP_FOR_EACH_WITH_HASH (miss, hmap_node, hash, todo) {
3290 if (miss->ofproto == ofproto && flow_equal(&miss->flow, flow)) {
3298 /* Partially Initializes 'op' as an "execute" operation for 'miss' and
3299 * 'packet'. The caller must initialize op->actions and op->actions_len. If
3300 * 'miss' is associated with a subfacet the caller must also initialize the
3301 * returned op->subfacet, and if anything needs to be freed after processing
3302 * the op, the caller must initialize op->garbage also. */
3304 init_flow_miss_execute_op(struct flow_miss *miss, struct ofpbuf *packet,
3305 struct flow_miss_op *op)
3307 if (miss->flow.in_port.ofp_port
3308 != vsp_realdev_to_vlandev(miss->ofproto, miss->flow.in_port.ofp_port,
3309 miss->flow.vlan_tci)) {
3310 /* This packet was received on a VLAN splinter port. We
3311 * added a VLAN to the packet to make the packet resemble
3312 * the flow, but the actions were composed assuming that
3313 * the packet contained no VLAN. So, we must remove the
3314 * VLAN header from the packet before trying to execute the
3316 eth_pop_vlan(packet);
3319 op->subfacet = NULL;
3320 op->xout_garbage = false;
3321 op->dpif_op.type = DPIF_OP_EXECUTE;
3322 op->dpif_op.u.execute.key = miss->key;
3323 op->dpif_op.u.execute.key_len = miss->key_len;
3324 op->dpif_op.u.execute.packet = packet;
3325 ofpbuf_use_stack(&op->mask, &op->maskbuf, sizeof op->maskbuf);
3328 /* Helper for handle_flow_miss_without_facet() and
3329 * handle_flow_miss_with_facet(). */
3331 handle_flow_miss_common(struct rule_dpif *rule,
3332 struct ofpbuf *packet, const struct flow *flow)
3334 struct ofproto_dpif *ofproto = ofproto_dpif_cast(rule->up.ofproto);
3336 if (rule->up.cr.priority == FAIL_OPEN_PRIORITY) {
3338 * Extra-special case for fail-open mode.
3340 * We are in fail-open mode and the packet matched the fail-open
3341 * rule, but we are connected to a controller too. We should send
3342 * the packet up to the controller in the hope that it will try to
3343 * set up a flow and thereby allow us to exit fail-open.
3345 * See the top-level comment in fail-open.c for more information.
3347 send_packet_in_miss(ofproto, packet, flow);
3351 /* Figures out whether a flow that missed in 'ofproto', whose details are in
3352 * 'miss' masked by 'wc', is likely to be worth tracking in detail in userspace
3353 * and (usually) installing a datapath flow. The answer is usually "yes" (a
3354 * return value of true). However, for short flows the cost of bookkeeping is
3355 * much higher than the benefits, so when the datapath holds a large number of
3356 * flows we impose some heuristics to decide which flows are likely to be worth
3359 flow_miss_should_make_facet(struct flow_miss *miss, struct flow_wildcards *wc)
3361 struct dpif_backer *backer = miss->ofproto->backer;
3364 switch (flow_miss_model) {
3365 case OFPROTO_HANDLE_MISS_AUTO:
3367 case OFPROTO_HANDLE_MISS_WITH_FACETS:
3369 case OFPROTO_HANDLE_MISS_WITHOUT_FACETS:
3373 if (!backer->governor) {
3376 n_subfacets = hmap_count(&backer->subfacets);
3377 if (n_subfacets * 2 <= flow_eviction_threshold) {
3381 backer->governor = governor_create();
3384 hash = flow_hash_in_wildcards(&miss->flow, wc, 0);
3385 return governor_should_install_flow(backer->governor, hash,
3386 list_size(&miss->packets));
3389 /* Handles 'miss' without creating a facet or subfacet or creating any datapath
3390 * flow. 'miss->flow' must have matched 'rule' and been xlated into 'xout'.
3391 * May add an "execute" operation to 'ops' and increment '*n_ops'. */
3393 handle_flow_miss_without_facet(struct rule_dpif *rule, struct xlate_out *xout,
3394 struct flow_miss *miss,
3395 struct flow_miss_op *ops, size_t *n_ops)
3397 struct ofpbuf *packet;
3399 LIST_FOR_EACH (packet, list_node, &miss->packets) {
3401 COVERAGE_INC(facet_suppress);
3403 handle_flow_miss_common(rule, packet, &miss->flow);
3406 struct xlate_in xin;
3408 xlate_in_init(&xin, miss->ofproto, &miss->flow, rule, 0, packet);
3409 xlate_actions_for_side_effects(&xin);
3412 if (xout->odp_actions.size) {
3413 struct flow_miss_op *op = &ops[*n_ops];
3414 struct dpif_execute *execute = &op->dpif_op.u.execute;
3416 init_flow_miss_execute_op(miss, packet, op);
3417 xlate_out_copy(&op->xout, xout);
3418 execute->actions = op->xout.odp_actions.data;
3419 execute->actions_len = op->xout.odp_actions.size;
3420 op->xout_garbage = true;
3427 /* Handles 'miss', which matches 'facet'. May add any required datapath
3428 * operations to 'ops', incrementing '*n_ops' for each new op.
3430 * All of the packets in 'miss' are considered to have arrived at time 'now'.
3431 * This is really important only for new facets: if we just called time_msec()
3432 * here, then the new subfacet or its packets could look (occasionally) as
3433 * though it was used some time after the facet was used. That can make a
3434 * one-packet flow look like it has a nonzero duration, which looks odd in
3435 * e.g. NetFlow statistics.
3437 * If non-null, 'stats' will be folded into 'facet'. */
3439 handle_flow_miss_with_facet(struct flow_miss *miss, struct facet *facet,
3440 long long int now, struct dpif_flow_stats *stats,
3441 struct flow_miss_op *ops, size_t *n_ops)
3443 struct ofproto_dpif *ofproto = ofproto_dpif_cast(facet->rule->up.ofproto);
3444 enum subfacet_path want_path;
3445 struct subfacet *subfacet;
3446 struct ofpbuf *packet;
3448 subfacet = subfacet_create(facet, miss, now);
3449 want_path = facet->xout.slow ? SF_SLOW_PATH : SF_FAST_PATH;
3451 subfacet_update_stats(subfacet, stats);
3454 LIST_FOR_EACH (packet, list_node, &miss->packets) {
3455 struct flow_miss_op *op = &ops[*n_ops];
3457 handle_flow_miss_common(facet->rule, packet, &miss->flow);
3459 if (want_path != SF_FAST_PATH) {
3460 struct xlate_in xin;
3462 xlate_in_init(&xin, ofproto, &miss->flow, facet->rule, 0, packet);
3463 xlate_actions_for_side_effects(&xin);
3466 if (facet->xout.odp_actions.size) {
3467 struct dpif_execute *execute = &op->dpif_op.u.execute;
3469 init_flow_miss_execute_op(miss, packet, op);
3470 execute->actions = facet->xout.odp_actions.data,
3471 execute->actions_len = facet->xout.odp_actions.size;
3476 if (miss->upcall_type == DPIF_UC_MISS || subfacet->path != want_path) {
3477 struct flow_miss_op *op = &ops[(*n_ops)++];
3478 struct dpif_flow_put *put = &op->dpif_op.u.flow_put;
3480 subfacet->path = want_path;
3482 ofpbuf_use_stack(&op->mask, &op->maskbuf, sizeof op->maskbuf);
3483 if (enable_megaflows) {
3484 odp_flow_key_from_mask(&op->mask, &facet->xout.wc.masks,
3485 &miss->flow, UINT32_MAX);
3488 op->xout_garbage = false;
3489 op->dpif_op.type = DPIF_OP_FLOW_PUT;
3490 op->subfacet = subfacet;
3491 put->flags = DPIF_FP_CREATE | DPIF_FP_MODIFY;
3492 put->key = miss->key;
3493 put->key_len = miss->key_len;
3494 put->mask = op->mask.data;
3495 put->mask_len = op->mask.size;
3497 if (want_path == SF_FAST_PATH) {
3498 put->actions = facet->xout.odp_actions.data;
3499 put->actions_len = facet->xout.odp_actions.size;
3501 compose_slow_path(ofproto, &miss->flow, facet->xout.slow,
3502 op->slow_stub, sizeof op->slow_stub,
3503 &put->actions, &put->actions_len);
3509 /* Handles flow miss 'miss'. May add any required datapath operations
3510 * to 'ops', incrementing '*n_ops' for each new op. */
3512 handle_flow_miss(struct flow_miss *miss, struct flow_miss_op *ops,
3515 struct ofproto_dpif *ofproto = miss->ofproto;
3516 struct dpif_flow_stats stats__;
3517 struct dpif_flow_stats *stats = &stats__;
3518 struct ofpbuf *packet;
3519 struct facet *facet;
3523 memset(stats, 0, sizeof *stats);
3525 LIST_FOR_EACH (packet, list_node, &miss->packets) {
3526 stats->tcp_flags |= packet_get_tcp_flags(packet, &miss->flow);
3527 stats->n_bytes += packet->size;
3531 facet = facet_lookup_valid(ofproto, &miss->flow);
3533 struct flow_wildcards wc;
3534 struct rule_dpif *rule;
3535 struct xlate_out xout;
3536 struct xlate_in xin;
3538 flow_wildcards_init_catchall(&wc);
3539 rule = rule_dpif_lookup(ofproto, &miss->flow, &wc);
3540 rule_credit_stats(rule, stats);
3542 xlate_in_init(&xin, ofproto, &miss->flow, rule, stats->tcp_flags,
3544 xin.resubmit_stats = stats;
3545 xin.may_learn = true;
3546 xlate_actions(&xin, &xout);
3547 flow_wildcards_or(&xout.wc, &xout.wc, &wc);
3549 /* There does not exist a bijection between 'struct flow' and datapath
3550 * flow keys with fitness ODP_FIT_TO_LITTLE. This breaks a fundamental
3551 * assumption used throughout the facet and subfacet handling code.
3552 * Since we have to handle these misses in userspace anyway, we simply
3553 * skip facet creation, avoiding the problem altogether. */
3554 if (miss->key_fitness == ODP_FIT_TOO_LITTLE
3555 || !flow_miss_should_make_facet(miss, &xout.wc)) {
3556 handle_flow_miss_without_facet(rule, &xout, miss, ops, n_ops);
3560 facet = facet_create(miss, rule, &xout, stats);
3563 handle_flow_miss_with_facet(miss, facet, now, stats, ops, n_ops);
3566 static struct drop_key *
3567 drop_key_lookup(const struct dpif_backer *backer, const struct nlattr *key,
3570 struct drop_key *drop_key;
3572 HMAP_FOR_EACH_WITH_HASH (drop_key, hmap_node, hash_bytes(key, key_len, 0),
3573 &backer->drop_keys) {
3574 if (drop_key->key_len == key_len
3575 && !memcmp(drop_key->key, key, key_len)) {
3583 drop_key_clear(struct dpif_backer *backer)
3585 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 15);
3586 struct drop_key *drop_key, *next;
3588 HMAP_FOR_EACH_SAFE (drop_key, next, hmap_node, &backer->drop_keys) {
3591 error = dpif_flow_del(backer->dpif, drop_key->key, drop_key->key_len,
3593 if (error && !VLOG_DROP_WARN(&rl)) {
3594 struct ds ds = DS_EMPTY_INITIALIZER;
3595 odp_flow_key_format(drop_key->key, drop_key->key_len, &ds);
3596 VLOG_WARN("Failed to delete drop key (%s) (%s)",
3597 ovs_strerror(error), ds_cstr(&ds));
3601 hmap_remove(&backer->drop_keys, &drop_key->hmap_node);
3602 free(drop_key->key);
3607 /* Given a datpath, packet, and flow metadata ('backer', 'packet', and 'key'
3608 * respectively), populates 'flow' with the result of odp_flow_key_to_flow().
3609 * Optionally, if nonnull, populates 'fitnessp' with the fitness of 'flow' as
3610 * returned by odp_flow_key_to_flow(). Also, optionally populates 'ofproto'
3611 * with the ofproto_dpif, and 'odp_in_port' with the datapath in_port, that
3612 * 'packet' ingressed.
3614 * If 'ofproto' is nonnull, requires 'flow''s in_port to exist. Otherwise sets
3615 * 'flow''s in_port to OFPP_NONE.
3617 * This function does post-processing on data returned from
3618 * odp_flow_key_to_flow() to help make VLAN splinters transparent to the rest
3619 * of the upcall processing logic. In particular, if the extracted in_port is
3620 * a VLAN splinter port, it replaces flow->in_port by the "real" port, sets
3621 * flow->vlan_tci correctly for the VLAN of the VLAN splinter port, and pushes
3622 * a VLAN header onto 'packet' (if it is nonnull).
3624 * Similarly, this function also includes some logic to help with tunnels. It
3625 * may modify 'flow' as necessary to make the tunneling implementation
3626 * transparent to the upcall processing logic.
3628 * Returns 0 if successful, ENODEV if the parsed flow has no associated ofport,
3629 * or some other positive errno if there are other problems. */
3631 ofproto_receive(const struct dpif_backer *backer, struct ofpbuf *packet,
3632 const struct nlattr *key, size_t key_len,
3633 struct flow *flow, enum odp_key_fitness *fitnessp,
3634 struct ofproto_dpif **ofproto, odp_port_t *odp_in_port)
3636 const struct ofport_dpif *port;
3637 enum odp_key_fitness fitness;
3640 fitness = odp_flow_key_to_flow(key, key_len, flow);
3641 if (fitness == ODP_FIT_ERROR) {
3647 *odp_in_port = flow->in_port.odp_port;
3650 port = (tnl_port_should_receive(flow)
3651 ? tnl_port_receive(flow)
3652 : odp_port_to_ofport(backer, flow->in_port.odp_port));
3653 flow->in_port.ofp_port = port ? port->up.ofp_port : OFPP_NONE;
3658 /* XXX: Since the tunnel module is not scoped per backer, for a tunnel port
3659 * it's theoretically possible that we'll receive an ofport belonging to an
3660 * entirely different datapath. In practice, this can't happen because no
3661 * platforms has two separate datapaths which each support tunneling. */
3662 ovs_assert(ofproto_dpif_cast(port->up.ofproto)->backer == backer);
3664 if (vsp_adjust_flow(ofproto_dpif_cast(port->up.ofproto), flow)) {
3666 /* Make the packet resemble the flow, so that it gets sent to
3667 * an OpenFlow controller properly, so that it looks correct
3668 * for sFlow, and so that flow_extract() will get the correct
3669 * vlan_tci if it is called on 'packet'.
3671 * The allocated space inside 'packet' probably also contains
3672 * 'key', that is, both 'packet' and 'key' are probably part of
3673 * a struct dpif_upcall (see the large comment on that
3674 * structure definition), so pushing data on 'packet' is in
3675 * general not a good idea since it could overwrite 'key' or
3676 * free it as a side effect. However, it's OK in this special
3677 * case because we know that 'packet' is inside a Netlink
3678 * attribute: pushing 4 bytes will just overwrite the 4-byte
3679 * "struct nlattr", which is fine since we don't need that
3680 * header anymore. */
3681 eth_push_vlan(packet, flow->vlan_tci);
3683 /* We can't reproduce 'key' from 'flow'. */
3684 fitness = fitness == ODP_FIT_PERFECT ? ODP_FIT_TOO_MUCH : fitness;
3689 *ofproto = ofproto_dpif_cast(port->up.ofproto);
3694 *fitnessp = fitness;
3700 handle_miss_upcalls(struct dpif_backer *backer, struct dpif_upcall *upcalls,
3703 struct dpif_upcall *upcall;
3704 struct flow_miss *miss;
3705 struct flow_miss misses[FLOW_MISS_MAX_BATCH];
3706 struct flow_miss_op flow_miss_ops[FLOW_MISS_MAX_BATCH * 2];
3707 struct dpif_op *dpif_ops[FLOW_MISS_MAX_BATCH * 2];
3717 /* Construct the to-do list.
3719 * This just amounts to extracting the flow from each packet and sticking
3720 * the packets that have the same flow in the same "flow_miss" structure so
3721 * that we can process them together. */
3724 for (upcall = upcalls; upcall < &upcalls[n_upcalls]; upcall++) {
3725 struct flow_miss *miss = &misses[n_misses];
3726 struct flow_miss *existing_miss;
3727 struct ofproto_dpif *ofproto;
3728 odp_port_t odp_in_port;
3733 error = ofproto_receive(backer, upcall->packet, upcall->key,
3734 upcall->key_len, &flow, &miss->key_fitness,
3735 &ofproto, &odp_in_port);
3736 if (error == ENODEV) {
3737 struct drop_key *drop_key;
3739 /* Received packet on datapath port for which we couldn't
3740 * associate an ofproto. This can happen if a port is removed
3741 * while traffic is being received. Print a rate-limited message
3742 * in case it happens frequently. Install a drop flow so
3743 * that future packets of the flow are inexpensively dropped
3745 VLOG_INFO_RL(&rl, "received packet on unassociated datapath port "
3746 "%"PRIu32, odp_in_port);
3748 drop_key = drop_key_lookup(backer, upcall->key, upcall->key_len);
3750 drop_key = xmalloc(sizeof *drop_key);
3751 drop_key->key = xmemdup(upcall->key, upcall->key_len);
3752 drop_key->key_len = upcall->key_len;
3754 hmap_insert(&backer->drop_keys, &drop_key->hmap_node,
3755 hash_bytes(drop_key->key, drop_key->key_len, 0));
3756 dpif_flow_put(backer->dpif, DPIF_FP_CREATE | DPIF_FP_MODIFY,
3757 drop_key->key, drop_key->key_len,
3758 NULL, 0, NULL, 0, NULL);
3766 ofproto->n_missed++;
3767 flow_extract(upcall->packet, flow.skb_priority, flow.skb_mark,
3768 &flow.tunnel, &flow.in_port, &miss->flow);
3770 /* Add other packets to a to-do list. */
3771 hash = flow_hash(&miss->flow, 0);
3772 existing_miss = flow_miss_find(&todo, ofproto, &miss->flow, hash);
3773 if (!existing_miss) {
3774 hmap_insert(&todo, &miss->hmap_node, hash);
3775 miss->ofproto = ofproto;
3776 miss->key = upcall->key;
3777 miss->key_len = upcall->key_len;
3778 miss->upcall_type = upcall->type;
3779 list_init(&miss->packets);
3783 miss = existing_miss;
3785 list_push_back(&miss->packets, &upcall->packet->list_node);
3788 /* Process each element in the to-do list, constructing the set of
3789 * operations to batch. */
3791 HMAP_FOR_EACH (miss, hmap_node, &todo) {
3792 handle_flow_miss(miss, flow_miss_ops, &n_ops);
3794 ovs_assert(n_ops <= ARRAY_SIZE(flow_miss_ops));
3796 /* Execute batch. */
3797 for (i = 0; i < n_ops; i++) {
3798 dpif_ops[i] = &flow_miss_ops[i].dpif_op;
3800 dpif_operate(backer->dpif, dpif_ops, n_ops);
3802 for (i = 0; i < n_ops; i++) {
3803 if (dpif_ops[i]->error != 0
3804 && flow_miss_ops[i].dpif_op.type == DPIF_OP_FLOW_PUT
3805 && flow_miss_ops[i].subfacet) {
3806 struct subfacet *subfacet = flow_miss_ops[i].subfacet;
3808 COVERAGE_INC(subfacet_install_fail);
3810 subfacet->path = SF_NOT_INSTALLED;
3814 if (flow_miss_ops[i].xout_garbage) {
3815 xlate_out_uninit(&flow_miss_ops[i].xout);
3818 hmap_destroy(&todo);
3821 static enum { SFLOW_UPCALL, MISS_UPCALL, BAD_UPCALL, FLOW_SAMPLE_UPCALL,
3823 classify_upcall(const struct dpif_upcall *upcall)
3825 size_t userdata_len;
3826 union user_action_cookie cookie;
3828 /* First look at the upcall type. */
3829 switch (upcall->type) {
3830 case DPIF_UC_ACTION:
3836 case DPIF_N_UC_TYPES:
3838 VLOG_WARN_RL(&rl, "upcall has unexpected type %"PRIu32, upcall->type);
3842 /* "action" upcalls need a closer look. */
3843 if (!upcall->userdata) {
3844 VLOG_WARN_RL(&rl, "action upcall missing cookie");
3847 userdata_len = nl_attr_get_size(upcall->userdata);
3848 if (userdata_len < sizeof cookie.type
3849 || userdata_len > sizeof cookie) {
3850 VLOG_WARN_RL(&rl, "action upcall cookie has unexpected size %zu",
3854 memset(&cookie, 0, sizeof cookie);
3855 memcpy(&cookie, nl_attr_get(upcall->userdata), userdata_len);
3856 if (userdata_len == sizeof cookie.sflow
3857 && cookie.type == USER_ACTION_COOKIE_SFLOW) {
3858 return SFLOW_UPCALL;
3859 } else if (userdata_len == sizeof cookie.slow_path
3860 && cookie.type == USER_ACTION_COOKIE_SLOW_PATH) {
3862 } else if (userdata_len == sizeof cookie.flow_sample
3863 && cookie.type == USER_ACTION_COOKIE_FLOW_SAMPLE) {
3864 return FLOW_SAMPLE_UPCALL;
3865 } else if (userdata_len == sizeof cookie.ipfix
3866 && cookie.type == USER_ACTION_COOKIE_IPFIX) {
3867 return IPFIX_UPCALL;
3869 VLOG_WARN_RL(&rl, "invalid user cookie of type %"PRIu16
3870 " and size %zu", cookie.type, userdata_len);
3876 handle_sflow_upcall(struct dpif_backer *backer,
3877 const struct dpif_upcall *upcall)
3879 struct ofproto_dpif *ofproto;
3880 union user_action_cookie cookie;
3882 odp_port_t odp_in_port;
3884 if (ofproto_receive(backer, upcall->packet, upcall->key, upcall->key_len,
3885 &flow, NULL, &ofproto, &odp_in_port)
3886 || !ofproto->sflow) {
3890 memset(&cookie, 0, sizeof cookie);
3891 memcpy(&cookie, nl_attr_get(upcall->userdata), sizeof cookie.sflow);
3892 dpif_sflow_received(ofproto->sflow, upcall->packet, &flow,
3893 odp_in_port, &cookie);
3897 handle_flow_sample_upcall(struct dpif_backer *backer,
3898 const struct dpif_upcall *upcall)
3900 struct ofproto_dpif *ofproto;
3901 union user_action_cookie cookie;
3904 if (ofproto_receive(backer, upcall->packet, upcall->key, upcall->key_len,
3905 &flow, NULL, &ofproto, NULL)
3906 || !ofproto->ipfix) {
3910 memset(&cookie, 0, sizeof cookie);
3911 memcpy(&cookie, nl_attr_get(upcall->userdata), sizeof cookie.flow_sample);
3913 /* The flow reflects exactly the contents of the packet. Sample
3914 * the packet using it. */
3915 dpif_ipfix_flow_sample(ofproto->ipfix, upcall->packet, &flow,
3916 cookie.flow_sample.collector_set_id,
3917 cookie.flow_sample.probability,
3918 cookie.flow_sample.obs_domain_id,
3919 cookie.flow_sample.obs_point_id);
3923 handle_ipfix_upcall(struct dpif_backer *backer,
3924 const struct dpif_upcall *upcall)
3926 struct ofproto_dpif *ofproto;
3929 if (ofproto_receive(backer, upcall->packet, upcall->key, upcall->key_len,
3930 &flow, NULL, &ofproto, NULL)
3931 || !ofproto->ipfix) {
3935 /* The flow reflects exactly the contents of the packet. Sample
3936 * the packet using it. */
3937 dpif_ipfix_bridge_sample(ofproto->ipfix, upcall->packet, &flow);
3941 handle_upcalls(struct dpif_backer *backer, unsigned int max_batch)
3943 struct dpif_upcall misses[FLOW_MISS_MAX_BATCH];
3944 struct ofpbuf miss_bufs[FLOW_MISS_MAX_BATCH];
3945 uint64_t miss_buf_stubs[FLOW_MISS_MAX_BATCH][4096 / 8];
3950 ovs_assert(max_batch <= FLOW_MISS_MAX_BATCH);
3953 for (n_processed = 0; n_processed < max_batch; n_processed++) {
3954 struct dpif_upcall *upcall = &misses[n_misses];
3955 struct ofpbuf *buf = &miss_bufs[n_misses];
3958 ofpbuf_use_stub(buf, miss_buf_stubs[n_misses],
3959 sizeof miss_buf_stubs[n_misses]);
3960 error = dpif_recv(backer->dpif, upcall, buf);
3966 switch (classify_upcall(upcall)) {
3968 /* Handle it later. */
3973 handle_sflow_upcall(backer, upcall);
3977 case FLOW_SAMPLE_UPCALL:
3978 handle_flow_sample_upcall(backer, upcall);
3983 handle_ipfix_upcall(backer, upcall);
3993 /* Handle deferred MISS_UPCALL processing. */
3994 handle_miss_upcalls(backer, misses, n_misses);
3995 for (i = 0; i < n_misses; i++) {
3996 ofpbuf_uninit(&miss_bufs[i]);
4002 /* Flow expiration. */
4004 static int subfacet_max_idle(const struct dpif_backer *);
4005 static void update_stats(struct dpif_backer *);
4006 static void rule_expire(struct rule_dpif *);
4007 static void expire_subfacets(struct dpif_backer *, int dp_max_idle);
4009 /* This function is called periodically by run(). Its job is to collect
4010 * updates for the flows that have been installed into the datapath, most
4011 * importantly when they last were used, and then use that information to
4012 * expire flows that have not been used recently.
4014 * Returns the number of milliseconds after which it should be called again. */
4016 expire(struct dpif_backer *backer)
4018 struct ofproto_dpif *ofproto;
4022 /* Periodically clear out the drop keys in an effort to keep them
4023 * relatively few. */
4024 drop_key_clear(backer);
4026 /* Update stats for each flow in the backer. */
4027 update_stats(backer);
4029 n_subfacets = hmap_count(&backer->subfacets);
4031 struct subfacet *subfacet;
4032 long long int total, now;
4036 HMAP_FOR_EACH (subfacet, hmap_node, &backer->subfacets) {
4037 total += now - subfacet->created;
4039 backer->avg_subfacet_life += total / n_subfacets;
4041 backer->avg_subfacet_life /= 2;
4043 backer->avg_n_subfacet += n_subfacets;
4044 backer->avg_n_subfacet /= 2;
4046 backer->max_n_subfacet = MAX(backer->max_n_subfacet, n_subfacets);
4048 max_idle = subfacet_max_idle(backer);
4049 expire_subfacets(backer, max_idle);
4051 HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
4052 struct rule *rule, *next_rule;
4054 if (ofproto->backer != backer) {
4058 /* Expire OpenFlow flows whose idle_timeout or hard_timeout
4060 LIST_FOR_EACH_SAFE (rule, next_rule, expirable,
4061 &ofproto->up.expirable) {
4062 rule_expire(rule_dpif_cast(rule));
4065 /* All outstanding data in existing flows has been accounted, so it's a
4066 * good time to do bond rebalancing. */
4067 if (ofproto->has_bonded_bundles) {
4068 struct ofbundle *bundle;
4070 HMAP_FOR_EACH (bundle, hmap_node, &ofproto->bundles) {
4072 bond_rebalance(bundle->bond, &backer->revalidate_set);
4078 return MIN(max_idle, 1000);
4081 /* Updates flow table statistics given that the datapath just reported 'stats'
4082 * as 'subfacet''s statistics. */
4084 update_subfacet_stats(struct subfacet *subfacet,
4085 const struct dpif_flow_stats *stats)
4087 struct facet *facet = subfacet->facet;
4088 struct ofproto_dpif *ofproto = ofproto_dpif_cast(facet->rule->up.ofproto);
4089 struct dpif_flow_stats diff;
4091 diff.tcp_flags = stats->tcp_flags;
4092 diff.used = stats->used;
4094 if (stats->n_packets >= subfacet->dp_packet_count) {
4095 diff.n_packets = stats->n_packets - subfacet->dp_packet_count;
4097 VLOG_WARN_RL(&rl, "unexpected packet count from the datapath");
4101 if (stats->n_bytes >= subfacet->dp_byte_count) {
4102 diff.n_bytes = stats->n_bytes - subfacet->dp_byte_count;
4104 VLOG_WARN_RL(&rl, "unexpected byte count from datapath");
4108 ofproto->n_hit += diff.n_packets;
4109 subfacet->dp_packet_count = stats->n_packets;
4110 subfacet->dp_byte_count = stats->n_bytes;
4111 subfacet_update_stats(subfacet, &diff);
4113 if (facet->accounted_bytes < facet->byte_count) {
4115 facet_account(facet);
4116 facet->accounted_bytes = facet->byte_count;
4120 /* 'key' with length 'key_len' bytes is a flow in 'dpif' that we know nothing
4121 * about, or a flow that shouldn't be installed but was anyway. Delete it. */
4123 delete_unexpected_flow(struct dpif_backer *backer,
4124 const struct nlattr *key, size_t key_len)
4126 if (!VLOG_DROP_WARN(&rl)) {
4130 odp_flow_key_format(key, key_len, &s);
4131 VLOG_WARN("unexpected flow: %s", ds_cstr(&s));
4135 COVERAGE_INC(facet_unexpected);
4136 dpif_flow_del(backer->dpif, key, key_len, NULL);
4139 /* Update 'packet_count', 'byte_count', and 'used' members of installed facets.
4141 * This function also pushes statistics updates to rules which each facet
4142 * resubmits into. Generally these statistics will be accurate. However, if a
4143 * facet changes the rule it resubmits into at some time in between
4144 * update_stats() runs, it is possible that statistics accrued to the
4145 * old rule will be incorrectly attributed to the new rule. This could be
4146 * avoided by calling update_stats() whenever rules are created or
4147 * deleted. However, the performance impact of making so many calls to the
4148 * datapath do not justify the benefit of having perfectly accurate statistics.
4150 * In addition, this function maintains per ofproto flow hit counts. The patch
4151 * port is not treated specially. e.g. A packet ingress from br0 patched into
4152 * br1 will increase the hit count of br0 by 1, however, does not affect
4153 * the hit or miss counts of br1.
4156 update_stats(struct dpif_backer *backer)
4158 const struct dpif_flow_stats *stats;
4159 struct dpif_flow_dump dump;
4160 const struct nlattr *key, *mask;
4161 size_t key_len, mask_len;
4163 dpif_flow_dump_start(&dump, backer->dpif);
4164 while (dpif_flow_dump_next(&dump, &key, &key_len,
4165 &mask, &mask_len, NULL, NULL, &stats)) {
4166 struct subfacet *subfacet;
4169 key_hash = odp_flow_key_hash(key, key_len);
4170 subfacet = subfacet_find(backer, key, key_len, key_hash);
4171 switch (subfacet ? subfacet->path : SF_NOT_INSTALLED) {
4173 update_subfacet_stats(subfacet, stats);
4177 /* Stats are updated per-packet. */
4180 case SF_NOT_INSTALLED:
4182 delete_unexpected_flow(backer, key, key_len);
4187 dpif_flow_dump_done(&dump);
4189 update_moving_averages(backer);
4192 /* Calculates and returns the number of milliseconds of idle time after which
4193 * subfacets should expire from the datapath. When a subfacet expires, we fold
4194 * its statistics into its facet, and when a facet's last subfacet expires, we
4195 * fold its statistic into its rule. */
4197 subfacet_max_idle(const struct dpif_backer *backer)
4200 * Idle time histogram.
4202 * Most of the time a switch has a relatively small number of subfacets.
4203 * When this is the case we might as well keep statistics for all of them
4204 * in userspace and to cache them in the kernel datapath for performance as
4207 * As the number of subfacets increases, the memory required to maintain
4208 * statistics about them in userspace and in the kernel becomes
4209 * significant. However, with a large number of subfacets it is likely
4210 * that only a few of them are "heavy hitters" that consume a large amount
4211 * of bandwidth. At this point, only heavy hitters are worth caching in
4212 * the kernel and maintaining in userspaces; other subfacets we can
4215 * The technique used to compute the idle time is to build a histogram with
4216 * N_BUCKETS buckets whose width is BUCKET_WIDTH msecs each. Each subfacet
4217 * that is installed in the kernel gets dropped in the appropriate bucket.
4218 * After the histogram has been built, we compute the cutoff so that only
4219 * the most-recently-used 1% of subfacets (but at least
4220 * flow_eviction_threshold flows) are kept cached. At least
4221 * the most-recently-used bucket of subfacets is kept, so actually an
4222 * arbitrary number of subfacets can be kept in any given expiration run
4223 * (though the next run will delete most of those unless they receive
4226 * This requires a second pass through the subfacets, in addition to the
4227 * pass made by update_stats(), because the former function never looks at
4228 * uninstallable subfacets.
4230 enum { BUCKET_WIDTH = ROUND_UP(100, TIME_UPDATE_INTERVAL) };
4231 enum { N_BUCKETS = 5000 / BUCKET_WIDTH };
4232 int buckets[N_BUCKETS] = { 0 };
4233 int total, subtotal, bucket;
4234 struct subfacet *subfacet;
4238 total = hmap_count(&backer->subfacets);
4239 if (total <= flow_eviction_threshold) {
4240 return N_BUCKETS * BUCKET_WIDTH;
4243 /* Build histogram. */
4245 HMAP_FOR_EACH (subfacet, hmap_node, &backer->subfacets) {
4246 long long int idle = now - subfacet->used;
4247 int bucket = (idle <= 0 ? 0
4248 : idle >= BUCKET_WIDTH * N_BUCKETS ? N_BUCKETS - 1
4249 : (unsigned int) idle / BUCKET_WIDTH);
4253 /* Find the first bucket whose flows should be expired. */
4254 subtotal = bucket = 0;
4256 subtotal += buckets[bucket++];
4257 } while (bucket < N_BUCKETS &&
4258 subtotal < MAX(flow_eviction_threshold, total / 100));
4260 if (VLOG_IS_DBG_ENABLED()) {
4264 ds_put_cstr(&s, "keep");
4265 for (i = 0; i < N_BUCKETS; i++) {
4267 ds_put_cstr(&s, ", drop");
4270 ds_put_format(&s, " %d:%d", i * BUCKET_WIDTH, buckets[i]);
4273 VLOG_INFO("%s (msec:count)", ds_cstr(&s));
4277 return bucket * BUCKET_WIDTH;
4281 expire_subfacets(struct dpif_backer *backer, int dp_max_idle)
4283 /* Cutoff time for most flows. */
4284 long long int normal_cutoff = time_msec() - dp_max_idle;
4286 /* We really want to keep flows for special protocols around, so use a more
4287 * conservative cutoff. */
4288 long long int special_cutoff = time_msec() - 10000;
4290 struct subfacet *subfacet, *next_subfacet;
4291 struct subfacet *batch[SUBFACET_DESTROY_MAX_BATCH];
4295 HMAP_FOR_EACH_SAFE (subfacet, next_subfacet, hmap_node,
4296 &backer->subfacets) {
4297 long long int cutoff;
4299 cutoff = (subfacet->facet->xout.slow & (SLOW_CFM | SLOW_BFD | SLOW_LACP
4303 if (subfacet->used < cutoff) {
4304 if (subfacet->path != SF_NOT_INSTALLED) {
4305 batch[n_batch++] = subfacet;
4306 if (n_batch >= SUBFACET_DESTROY_MAX_BATCH) {
4307 subfacet_destroy_batch(backer, batch, n_batch);
4311 subfacet_destroy(subfacet);
4317 subfacet_destroy_batch(backer, batch, n_batch);
4321 /* If 'rule' is an OpenFlow rule, that has expired according to OpenFlow rules,
4322 * then delete it entirely. */
4324 rule_expire(struct rule_dpif *rule)
4326 struct facet *facet, *next_facet;
4330 if (rule->up.pending) {
4331 /* We'll have to expire it later. */
4335 /* Has 'rule' expired? */
4337 if (rule->up.hard_timeout
4338 && now > rule->up.modified + rule->up.hard_timeout * 1000) {
4339 reason = OFPRR_HARD_TIMEOUT;
4340 } else if (rule->up.idle_timeout
4341 && now > rule->up.used + rule->up.idle_timeout * 1000) {
4342 reason = OFPRR_IDLE_TIMEOUT;
4347 COVERAGE_INC(ofproto_dpif_expired);
4349 /* Update stats. (This is a no-op if the rule expired due to an idle
4350 * timeout, because that only happens when the rule has no facets left.) */
4351 LIST_FOR_EACH_SAFE (facet, next_facet, list_node, &rule->facets) {
4352 facet_remove(facet);
4355 /* Get rid of the rule. */
4356 ofproto_rule_expire(&rule->up, reason);
4361 /* Creates and returns a new facet based on 'miss'.
4363 * The caller must already have determined that no facet with an identical
4364 * 'miss->flow' exists in 'miss->ofproto'.
4366 * 'rule' and 'xout' must have been created based on 'miss'.
4368 * 'facet'' statistics are initialized based on 'stats'.
4370 * The facet will initially have no subfacets. The caller should create (at
4371 * least) one subfacet with subfacet_create(). */
4372 static struct facet *
4373 facet_create(const struct flow_miss *miss, struct rule_dpif *rule,
4374 struct xlate_out *xout, struct dpif_flow_stats *stats)
4376 struct ofproto_dpif *ofproto = miss->ofproto;
4377 struct facet *facet;
4380 facet = xzalloc(sizeof *facet);
4381 facet->packet_count = facet->prev_packet_count = stats->n_packets;
4382 facet->byte_count = facet->prev_byte_count = stats->n_bytes;
4383 facet->tcp_flags = stats->tcp_flags;
4384 facet->used = stats->used;
4385 facet->flow = miss->flow;
4386 facet->learn_rl = time_msec() + 500;
4389 list_push_back(&facet->rule->facets, &facet->list_node);
4390 list_init(&facet->subfacets);
4391 netflow_flow_init(&facet->nf_flow);
4392 netflow_flow_update_time(ofproto->netflow, &facet->nf_flow, facet->used);
4394 xlate_out_copy(&facet->xout, xout);
4396 match_init(&match, &facet->flow, &facet->xout.wc);
4397 cls_rule_init(&facet->cr, &match, OFP_DEFAULT_PRIORITY);
4398 classifier_insert(&ofproto->facets, &facet->cr);
4400 facet->nf_flow.output_iface = facet->xout.nf_output_iface;
4406 facet_free(struct facet *facet)
4409 xlate_out_uninit(&facet->xout);
4414 /* Executes, within 'ofproto', the 'n_actions' actions in 'actions' on
4415 * 'packet', which arrived on 'in_port'. */
4417 execute_odp_actions(struct ofproto_dpif *ofproto, const struct flow *flow,
4418 const struct nlattr *odp_actions, size_t actions_len,
4419 struct ofpbuf *packet)
4421 struct odputil_keybuf keybuf;
4425 ofpbuf_use_stack(&key, &keybuf, sizeof keybuf);
4426 odp_flow_key_from_flow(&key, flow,
4427 ofp_port_to_odp_port(ofproto, flow->in_port.ofp_port));
4429 error = dpif_execute(ofproto->backer->dpif, key.data, key.size,
4430 odp_actions, actions_len, packet);
4434 /* Remove 'facet' from its ofproto and free up the associated memory:
4436 * - If 'facet' was installed in the datapath, uninstalls it and updates its
4437 * rule's statistics, via subfacet_uninstall().
4439 * - Removes 'facet' from its rule and from ofproto->facets.
4442 facet_remove(struct facet *facet)
4444 struct ofproto_dpif *ofproto = ofproto_dpif_cast(facet->rule->up.ofproto);
4445 struct subfacet *subfacet, *next_subfacet;
4447 ovs_assert(!list_is_empty(&facet->subfacets));
4449 /* First uninstall all of the subfacets to get final statistics. */
4450 LIST_FOR_EACH (subfacet, list_node, &facet->subfacets) {
4451 subfacet_uninstall(subfacet);
4454 /* Flush the final stats to the rule.
4456 * This might require us to have at least one subfacet around so that we
4457 * can use its actions for accounting in facet_account(), which is why we
4458 * have uninstalled but not yet destroyed the subfacets. */
4459 facet_flush_stats(facet);
4461 /* Now we're really all done so destroy everything. */
4462 LIST_FOR_EACH_SAFE (subfacet, next_subfacet, list_node,
4463 &facet->subfacets) {
4464 subfacet_destroy__(subfacet);
4466 classifier_remove(&ofproto->facets, &facet->cr);
4467 cls_rule_destroy(&facet->cr);
4468 list_remove(&facet->list_node);
4472 /* Feed information from 'facet' back into the learning table to keep it in
4473 * sync with what is actually flowing through the datapath. */
4475 facet_learn(struct facet *facet)
4477 long long int now = time_msec();
4479 if (!facet->xout.has_fin_timeout && now < facet->learn_rl) {
4483 facet->learn_rl = now + 500;
4485 if (!facet->xout.has_learn
4486 && !facet->xout.has_normal
4487 && (!facet->xout.has_fin_timeout
4488 || !(facet->tcp_flags & (TCP_FIN | TCP_RST)))) {
4492 facet_push_stats(facet, true);
4496 facet_account(struct facet *facet)
4498 struct ofproto_dpif *ofproto = ofproto_dpif_cast(facet->rule->up.ofproto);
4499 const struct nlattr *a;
4504 if (!facet->xout.has_normal || !ofproto->has_bonded_bundles) {
4507 n_bytes = facet->byte_count - facet->accounted_bytes;
4509 /* This loop feeds byte counters to bond_account() for rebalancing to use
4510 * as a basis. We also need to track the actual VLAN on which the packet
4511 * is going to be sent to ensure that it matches the one passed to
4512 * bond_choose_output_slave(). (Otherwise, we will account to the wrong
4515 * We use the actions from an arbitrary subfacet because they should all
4516 * be equally valid for our purpose. */
4517 vlan_tci = facet->flow.vlan_tci;
4518 NL_ATTR_FOR_EACH_UNSAFE (a, left, facet->xout.odp_actions.data,
4519 facet->xout.odp_actions.size) {
4520 const struct ovs_action_push_vlan *vlan;
4521 struct ofport_dpif *port;
4523 switch (nl_attr_type(a)) {
4524 case OVS_ACTION_ATTR_OUTPUT:
4525 port = get_odp_port(ofproto, nl_attr_get_odp_port(a));
4526 if (port && port->bundle && port->bundle->bond) {
4527 bond_account(port->bundle->bond, &facet->flow,
4528 vlan_tci_to_vid(vlan_tci), n_bytes);
4532 case OVS_ACTION_ATTR_POP_VLAN:
4533 vlan_tci = htons(0);
4536 case OVS_ACTION_ATTR_PUSH_VLAN:
4537 vlan = nl_attr_get(a);
4538 vlan_tci = vlan->vlan_tci;
4544 /* Returns true if the only action for 'facet' is to send to the controller.
4545 * (We don't report NetFlow expiration messages for such facets because they
4546 * are just part of the control logic for the network, not real traffic). */
4548 facet_is_controller_flow(struct facet *facet)
4551 const struct rule *rule = &facet->rule->up;
4552 const struct ofpact *ofpacts = rule->ofpacts;
4553 size_t ofpacts_len = rule->ofpacts_len;
4555 if (ofpacts_len > 0 &&
4556 ofpacts->type == OFPACT_CONTROLLER &&
4557 ofpact_next(ofpacts) >= ofpact_end(ofpacts, ofpacts_len)) {
4564 /* Folds all of 'facet''s statistics into its rule. Also updates the
4565 * accounting ofhook and emits a NetFlow expiration if appropriate. All of
4566 * 'facet''s statistics in the datapath should have been zeroed and folded into
4567 * its packet and byte counts before this function is called. */
4569 facet_flush_stats(struct facet *facet)
4571 struct ofproto_dpif *ofproto = ofproto_dpif_cast(facet->rule->up.ofproto);
4572 struct subfacet *subfacet;
4574 LIST_FOR_EACH (subfacet, list_node, &facet->subfacets) {
4575 ovs_assert(!subfacet->dp_byte_count);
4576 ovs_assert(!subfacet->dp_packet_count);
4579 facet_push_stats(facet, false);
4580 if (facet->accounted_bytes < facet->byte_count) {
4581 facet_account(facet);
4582 facet->accounted_bytes = facet->byte_count;
4585 if (ofproto->netflow && !facet_is_controller_flow(facet)) {
4586 struct ofexpired expired;
4587 expired.flow = facet->flow;
4588 expired.packet_count = facet->packet_count;
4589 expired.byte_count = facet->byte_count;
4590 expired.used = facet->used;
4591 netflow_expire(ofproto->netflow, &facet->nf_flow, &expired);
4594 /* Reset counters to prevent double counting if 'facet' ever gets
4596 facet_reset_counters(facet);
4598 netflow_flow_clear(&facet->nf_flow);
4599 facet->tcp_flags = 0;
4602 /* Searches 'ofproto''s table of facets for one which would be responsible for
4603 * 'flow'. Returns it if found, otherwise a null pointer.
4605 * The returned facet might need revalidation; use facet_lookup_valid()
4606 * instead if that is important. */
4607 static struct facet *
4608 facet_find(struct ofproto_dpif *ofproto, const struct flow *flow)
4610 struct cls_rule *cr = classifier_lookup(&ofproto->facets, flow, NULL);
4611 return cr ? CONTAINER_OF(cr, struct facet, cr) : NULL;
4614 /* Searches 'ofproto''s table of facets for one capable that covers
4615 * 'flow'. Returns it if found, otherwise a null pointer.
4617 * The returned facet is guaranteed to be valid. */
4618 static struct facet *
4619 facet_lookup_valid(struct ofproto_dpif *ofproto, const struct flow *flow)
4621 struct facet *facet;
4623 facet = facet_find(ofproto, flow);
4625 && (ofproto->backer->need_revalidate
4626 || tag_set_intersects(&ofproto->backer->revalidate_set,
4628 && !facet_revalidate(facet)) {
4636 facet_check_consistency(struct facet *facet)
4638 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 15);
4640 struct ofproto_dpif *ofproto = ofproto_dpif_cast(facet->rule->up.ofproto);
4642 struct xlate_out xout;
4643 struct xlate_in xin;
4645 struct rule_dpif *rule;
4648 /* Check the rule for consistency. */
4649 rule = rule_dpif_lookup(ofproto, &facet->flow, NULL);
4650 if (rule != facet->rule) {
4651 if (!VLOG_DROP_WARN(&rl)) {
4652 struct ds s = DS_EMPTY_INITIALIZER;
4654 flow_format(&s, &facet->flow);
4655 ds_put_format(&s, ": facet associated with wrong rule (was "
4656 "table=%"PRIu8",", facet->rule->up.table_id);
4657 cls_rule_format(&facet->rule->up.cr, &s);
4658 ds_put_format(&s, ") (should have been table=%"PRIu8",",
4660 cls_rule_format(&rule->up.cr, &s);
4661 ds_put_char(&s, ')');
4663 VLOG_WARN("%s", ds_cstr(&s));
4669 /* Check the datapath actions for consistency. */
4670 xlate_in_init(&xin, ofproto, &facet->flow, rule, 0, NULL);
4671 xlate_actions(&xin, &xout);
4673 ok = ofpbuf_equal(&facet->xout.odp_actions, &xout.odp_actions)
4674 && facet->xout.slow == xout.slow;
4675 if (!ok && !VLOG_DROP_WARN(&rl)) {
4676 struct ds s = DS_EMPTY_INITIALIZER;
4678 flow_format(&s, &facet->flow);
4679 ds_put_cstr(&s, ": inconsistency in facet");
4681 if (!ofpbuf_equal(&facet->xout.odp_actions, &xout.odp_actions)) {
4682 ds_put_cstr(&s, " (actions were: ");
4683 format_odp_actions(&s, facet->xout.odp_actions.data,
4684 facet->xout.odp_actions.size);
4685 ds_put_cstr(&s, ") (correct actions: ");
4686 format_odp_actions(&s, xout.odp_actions.data,
4687 xout.odp_actions.size);
4688 ds_put_char(&s, ')');
4691 if (facet->xout.slow != xout.slow) {
4692 ds_put_format(&s, " slow path incorrect. should be %d", xout.slow);
4695 VLOG_WARN("%s", ds_cstr(&s));
4698 xlate_out_uninit(&xout);
4703 /* Re-searches the classifier for 'facet':
4705 * - If the rule found is different from 'facet''s current rule, moves
4706 * 'facet' to the new rule and recompiles its actions.
4708 * - If the rule found is the same as 'facet''s current rule, leaves 'facet'
4709 * where it is and recompiles its actions anyway.
4711 * - If any of 'facet''s subfacets correspond to a new flow according to
4712 * ofproto_receive(), 'facet' is removed.
4714 * Returns true if 'facet' is still valid. False if 'facet' was removed. */
4716 facet_revalidate(struct facet *facet)
4718 struct ofproto_dpif *ofproto = ofproto_dpif_cast(facet->rule->up.ofproto);
4719 struct rule_dpif *new_rule;
4720 struct subfacet *subfacet;
4721 struct flow_wildcards wc;
4722 struct xlate_out xout;
4723 struct xlate_in xin;
4725 COVERAGE_INC(facet_revalidate);
4727 /* Check that child subfacets still correspond to this facet. Tunnel
4728 * configuration changes could cause a subfacet's OpenFlow in_port to
4730 LIST_FOR_EACH (subfacet, list_node, &facet->subfacets) {
4731 struct ofproto_dpif *recv_ofproto;
4732 struct flow recv_flow;
4735 error = ofproto_receive(ofproto->backer, NULL, subfacet->key,
4736 subfacet->key_len, &recv_flow, NULL,
4737 &recv_ofproto, NULL);
4739 || recv_ofproto != ofproto
4740 || facet != facet_find(ofproto, &recv_flow)) {
4741 facet_remove(facet);
4746 flow_wildcards_init_catchall(&wc);
4747 new_rule = rule_dpif_lookup(ofproto, &facet->flow, &wc);
4749 /* Calculate new datapath actions.
4751 * We do not modify any 'facet' state yet, because we might need to, e.g.,
4752 * emit a NetFlow expiration and, if so, we need to have the old state
4753 * around to properly compose it. */
4754 xlate_in_init(&xin, ofproto, &facet->flow, new_rule, 0, NULL);
4755 xlate_actions(&xin, &xout);
4756 flow_wildcards_or(&xout.wc, &xout.wc, &wc);
4758 /* A facet's slow path reason should only change under dramatic
4759 * circumstances. Rather than try to update everything, it's simpler to
4760 * remove the facet and start over.
4762 * More importantly, if a facet's wildcards change, it will be relatively
4763 * difficult to figure out if its subfacets still belong to it, and if not
4764 * which facet they may belong to. Again, to avoid the complexity, we
4765 * simply give up instead. */
4766 if (facet->xout.slow != xout.slow
4767 || memcmp(&facet->xout.wc, &xout.wc, sizeof xout.wc)) {
4768 facet_remove(facet);
4769 xlate_out_uninit(&xout);
4773 if (!ofpbuf_equal(&facet->xout.odp_actions, &xout.odp_actions)) {
4774 LIST_FOR_EACH(subfacet, list_node, &facet->subfacets) {
4775 if (subfacet->path == SF_FAST_PATH) {
4776 struct dpif_flow_stats stats;
4778 subfacet_install(subfacet, &xout.odp_actions, &stats);
4779 subfacet_update_stats(subfacet, &stats);
4783 facet_flush_stats(facet);
4785 ofpbuf_clear(&facet->xout.odp_actions);
4786 ofpbuf_put(&facet->xout.odp_actions, xout.odp_actions.data,
4787 xout.odp_actions.size);
4790 /* Update 'facet' now that we've taken care of all the old state. */
4791 facet->xout.tags = xout.tags;
4792 facet->xout.slow = xout.slow;
4793 facet->xout.has_learn = xout.has_learn;
4794 facet->xout.has_normal = xout.has_normal;
4795 facet->xout.has_fin_timeout = xout.has_fin_timeout;
4796 facet->xout.nf_output_iface = xout.nf_output_iface;
4797 facet->xout.mirrors = xout.mirrors;
4798 facet->nf_flow.output_iface = facet->xout.nf_output_iface;
4800 if (facet->rule != new_rule) {
4801 COVERAGE_INC(facet_changed_rule);
4802 list_remove(&facet->list_node);
4803 list_push_back(&new_rule->facets, &facet->list_node);
4804 facet->rule = new_rule;
4805 facet->used = new_rule->up.created;
4806 facet->prev_used = facet->used;
4809 xlate_out_uninit(&xout);
4814 facet_reset_counters(struct facet *facet)
4816 facet->packet_count = 0;
4817 facet->byte_count = 0;
4818 facet->prev_packet_count = 0;
4819 facet->prev_byte_count = 0;
4820 facet->accounted_bytes = 0;
4824 facet_push_stats(struct facet *facet, bool may_learn)
4826 struct dpif_flow_stats stats;
4828 ovs_assert(facet->packet_count >= facet->prev_packet_count);
4829 ovs_assert(facet->byte_count >= facet->prev_byte_count);
4830 ovs_assert(facet->used >= facet->prev_used);
4832 stats.n_packets = facet->packet_count - facet->prev_packet_count;
4833 stats.n_bytes = facet->byte_count - facet->prev_byte_count;
4834 stats.used = facet->used;
4835 stats.tcp_flags = facet->tcp_flags;
4837 if (may_learn || stats.n_packets || facet->used > facet->prev_used) {
4838 struct ofproto_dpif *ofproto =
4839 ofproto_dpif_cast(facet->rule->up.ofproto);
4841 struct ofport_dpif *in_port;
4842 struct xlate_in xin;
4844 facet->prev_packet_count = facet->packet_count;
4845 facet->prev_byte_count = facet->byte_count;
4846 facet->prev_used = facet->used;
4848 in_port = get_ofp_port(ofproto, facet->flow.in_port.ofp_port);
4849 if (in_port && in_port->is_tunnel) {
4850 netdev_vport_inc_rx(in_port->up.netdev, &stats);
4853 rule_credit_stats(facet->rule, &stats);
4854 netflow_flow_update_time(ofproto->netflow, &facet->nf_flow,
4856 netflow_flow_update_flags(&facet->nf_flow, facet->tcp_flags);
4857 update_mirror_stats(ofproto, facet->xout.mirrors, stats.n_packets,
4860 xlate_in_init(&xin, ofproto, &facet->flow, facet->rule,
4861 stats.tcp_flags, NULL);
4862 xin.resubmit_stats = &stats;
4863 xin.may_learn = may_learn;
4864 xlate_actions_for_side_effects(&xin);
4869 push_all_stats__(bool run_fast)
4871 static long long int rl = LLONG_MIN;
4872 struct ofproto_dpif *ofproto;
4874 if (time_msec() < rl) {
4878 HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
4879 struct cls_cursor cursor;
4880 struct facet *facet;
4882 cls_cursor_init(&cursor, &ofproto->facets, NULL);
4883 CLS_CURSOR_FOR_EACH (facet, cr, &cursor) {
4884 facet_push_stats(facet, false);
4891 rl = time_msec() + 100;
4895 push_all_stats(void)
4897 push_all_stats__(true);
4901 rule_credit_stats(struct rule_dpif *rule, const struct dpif_flow_stats *stats)
4903 rule->packet_count += stats->n_packets;
4904 rule->byte_count += stats->n_bytes;
4905 ofproto_rule_update_used(&rule->up, stats->used);
4910 static struct subfacet *
4911 subfacet_find(struct dpif_backer *backer, const struct nlattr *key,
4912 size_t key_len, uint32_t key_hash)
4914 struct subfacet *subfacet;
4916 HMAP_FOR_EACH_WITH_HASH (subfacet, hmap_node, key_hash,
4917 &backer->subfacets) {
4918 if (subfacet->key_len == key_len
4919 && !memcmp(key, subfacet->key, key_len)) {
4927 /* Searches 'facet' (within 'ofproto') for a subfacet with the specified
4928 * 'key_fitness', 'key', and 'key_len' members in 'miss'. Returns the
4929 * existing subfacet if there is one, otherwise creates and returns a
4931 static struct subfacet *
4932 subfacet_create(struct facet *facet, struct flow_miss *miss,
4935 struct dpif_backer *backer = miss->ofproto->backer;
4936 enum odp_key_fitness key_fitness = miss->key_fitness;
4937 const struct nlattr *key = miss->key;
4938 size_t key_len = miss->key_len;
4940 struct subfacet *subfacet;
4942 key_hash = odp_flow_key_hash(key, key_len);
4944 if (list_is_empty(&facet->subfacets)) {
4945 subfacet = &facet->one_subfacet;
4947 subfacet = subfacet_find(backer, key, key_len, key_hash);
4949 if (subfacet->facet == facet) {
4953 /* This shouldn't happen. */
4954 VLOG_ERR_RL(&rl, "subfacet with wrong facet");
4955 subfacet_destroy(subfacet);
4958 subfacet = xmalloc(sizeof *subfacet);
4961 hmap_insert(&backer->subfacets, &subfacet->hmap_node, key_hash);
4962 list_push_back(&facet->subfacets, &subfacet->list_node);
4963 subfacet->facet = facet;
4964 subfacet->key_fitness = key_fitness;
4965 subfacet->key = xmemdup(key, key_len);
4966 subfacet->key_len = key_len;
4967 subfacet->used = now;
4968 subfacet->created = now;
4969 subfacet->dp_packet_count = 0;
4970 subfacet->dp_byte_count = 0;
4971 subfacet->path = SF_NOT_INSTALLED;
4972 subfacet->backer = backer;
4974 backer->subfacet_add_count++;
4978 /* Uninstalls 'subfacet' from the datapath, if it is installed, removes it from
4979 * its facet within 'ofproto', and frees it. */
4981 subfacet_destroy__(struct subfacet *subfacet)
4983 struct facet *facet = subfacet->facet;
4984 struct ofproto_dpif *ofproto = ofproto_dpif_cast(facet->rule->up.ofproto);
4986 /* Update ofproto stats before uninstall the subfacet. */
4987 ofproto->backer->subfacet_del_count++;
4989 subfacet_uninstall(subfacet);
4990 hmap_remove(&subfacet->backer->subfacets, &subfacet->hmap_node);
4991 list_remove(&subfacet->list_node);
4992 free(subfacet->key);
4993 if (subfacet != &facet->one_subfacet) {
4998 /* Destroys 'subfacet', as with subfacet_destroy__(), and then if this was the
4999 * last remaining subfacet in its facet destroys the facet too. */
5001 subfacet_destroy(struct subfacet *subfacet)
5003 struct facet *facet = subfacet->facet;
5005 if (list_is_singleton(&facet->subfacets)) {
5006 /* facet_remove() needs at least one subfacet (it will remove it). */
5007 facet_remove(facet);
5009 subfacet_destroy__(subfacet);
5014 subfacet_destroy_batch(struct dpif_backer *backer,
5015 struct subfacet **subfacets, int n)
5017 struct dpif_op ops[SUBFACET_DESTROY_MAX_BATCH];
5018 struct dpif_op *opsp[SUBFACET_DESTROY_MAX_BATCH];
5019 struct dpif_flow_stats stats[SUBFACET_DESTROY_MAX_BATCH];
5022 for (i = 0; i < n; i++) {
5023 ops[i].type = DPIF_OP_FLOW_DEL;
5024 ops[i].u.flow_del.key = subfacets[i]->key;
5025 ops[i].u.flow_del.key_len = subfacets[i]->key_len;
5026 ops[i].u.flow_del.stats = &stats[i];
5030 dpif_operate(backer->dpif, opsp, n);
5031 for (i = 0; i < n; i++) {
5032 subfacet_reset_dp_stats(subfacets[i], &stats[i]);
5033 subfacets[i]->path = SF_NOT_INSTALLED;
5034 subfacet_destroy(subfacets[i]);
5039 /* Updates 'subfacet''s datapath flow, setting its actions to 'actions_len'
5040 * bytes of actions in 'actions'. If 'stats' is non-null, statistics counters
5041 * in the datapath will be zeroed and 'stats' will be updated with traffic new
5042 * since 'subfacet' was last updated.
5044 * Returns 0 if successful, otherwise a positive errno value. */
5046 subfacet_install(struct subfacet *subfacet, const struct ofpbuf *odp_actions,
5047 struct dpif_flow_stats *stats)
5049 struct facet *facet = subfacet->facet;
5050 struct ofproto_dpif *ofproto = ofproto_dpif_cast(facet->rule->up.ofproto);
5051 enum subfacet_path path = facet->xout.slow ? SF_SLOW_PATH : SF_FAST_PATH;
5052 const struct nlattr *actions = odp_actions->data;
5053 size_t actions_len = odp_actions->size;
5054 struct odputil_keybuf maskbuf;
5057 uint64_t slow_path_stub[128 / 8];
5058 enum dpif_flow_put_flags flags;
5061 flags = DPIF_FP_CREATE | DPIF_FP_MODIFY;
5063 flags |= DPIF_FP_ZERO_STATS;
5066 if (path == SF_SLOW_PATH) {
5067 compose_slow_path(ofproto, &facet->flow, facet->xout.slow,
5068 slow_path_stub, sizeof slow_path_stub,
5069 &actions, &actions_len);
5072 ofpbuf_use_stack(&mask, &maskbuf, sizeof maskbuf);
5073 if (enable_megaflows) {
5074 odp_flow_key_from_mask(&mask, &facet->xout.wc.masks,
5075 &facet->flow, UINT32_MAX);
5078 ret = dpif_flow_put(subfacet->backer->dpif, flags, subfacet->key,
5079 subfacet->key_len, mask.data, mask.size,
5080 actions, actions_len, stats);
5083 subfacet_reset_dp_stats(subfacet, stats);
5087 COVERAGE_INC(subfacet_install_fail);
5089 subfacet->path = path;
5094 /* If 'subfacet' is installed in the datapath, uninstalls it. */
5096 subfacet_uninstall(struct subfacet *subfacet)
5098 if (subfacet->path != SF_NOT_INSTALLED) {
5099 struct rule_dpif *rule = subfacet->facet->rule;
5100 struct ofproto_dpif *ofproto = ofproto_dpif_cast(rule->up.ofproto);
5101 struct dpif_flow_stats stats;
5104 error = dpif_flow_del(ofproto->backer->dpif, subfacet->key,
5105 subfacet->key_len, &stats);
5106 subfacet_reset_dp_stats(subfacet, &stats);
5108 subfacet_update_stats(subfacet, &stats);
5110 subfacet->path = SF_NOT_INSTALLED;
5112 ovs_assert(subfacet->dp_packet_count == 0);
5113 ovs_assert(subfacet->dp_byte_count == 0);
5117 /* Resets 'subfacet''s datapath statistics counters. This should be called
5118 * when 'subfacet''s statistics are cleared in the datapath. If 'stats' is
5119 * non-null, it should contain the statistics returned by dpif when 'subfacet'
5120 * was reset in the datapath. 'stats' will be modified to include only
5121 * statistics new since 'subfacet' was last updated. */
5123 subfacet_reset_dp_stats(struct subfacet *subfacet,
5124 struct dpif_flow_stats *stats)
5127 && subfacet->dp_packet_count <= stats->n_packets
5128 && subfacet->dp_byte_count <= stats->n_bytes) {
5129 stats->n_packets -= subfacet->dp_packet_count;
5130 stats->n_bytes -= subfacet->dp_byte_count;
5133 subfacet->dp_packet_count = 0;
5134 subfacet->dp_byte_count = 0;
5137 /* Folds the statistics from 'stats' into the counters in 'subfacet'.
5139 * Because of the meaning of a subfacet's counters, it only makes sense to do
5140 * this if 'stats' are not tracked in the datapath, that is, if 'stats'
5141 * represents a packet that was sent by hand or if it represents statistics
5142 * that have been cleared out of the datapath. */
5144 subfacet_update_stats(struct subfacet *subfacet,
5145 const struct dpif_flow_stats *stats)
5147 if (stats->n_packets || stats->used > subfacet->used) {
5148 struct facet *facet = subfacet->facet;
5150 subfacet->used = MAX(subfacet->used, stats->used);
5151 facet->used = MAX(facet->used, stats->used);
5152 facet->packet_count += stats->n_packets;
5153 facet->byte_count += stats->n_bytes;
5154 facet->tcp_flags |= stats->tcp_flags;
5160 /* Lookup 'flow' in 'ofproto''s classifier. If 'wc' is non-null, sets
5161 * the fields that were relevant as part of the lookup. */
5162 static struct rule_dpif *
5163 rule_dpif_lookup(struct ofproto_dpif *ofproto, const struct flow *flow,
5164 struct flow_wildcards *wc)
5166 struct rule_dpif *rule;
5168 rule = rule_dpif_lookup_in_table(ofproto, flow, wc, 0);
5173 return rule_dpif_miss_rule(ofproto, flow);
5177 rule_dpif_lookup_in_table(struct ofproto_dpif *ofproto,
5178 const struct flow *flow, struct flow_wildcards *wc,
5181 struct cls_rule *cls_rule;
5182 struct classifier *cls;
5185 if (table_id >= N_TABLES) {
5190 memset(&wc->masks.dl_type, 0xff, sizeof wc->masks.dl_type);
5191 wc->masks.nw_frag |= FLOW_NW_FRAG_MASK;
5194 cls = &ofproto->up.tables[table_id].cls;
5195 frag = (flow->nw_frag & FLOW_NW_FRAG_ANY) != 0;
5196 if (frag && ofproto->up.frag_handling == OFPC_FRAG_NORMAL) {
5197 /* We must pretend that transport ports are unavailable. */
5198 struct flow ofpc_normal_flow = *flow;
5199 ofpc_normal_flow.tp_src = htons(0);
5200 ofpc_normal_flow.tp_dst = htons(0);
5201 cls_rule = classifier_lookup(cls, &ofpc_normal_flow, wc);
5202 } else if (frag && ofproto->up.frag_handling == OFPC_FRAG_DROP) {
5203 cls_rule = &ofproto->drop_frags_rule->up.cr;
5205 flow_wildcards_init_exact(wc);
5208 cls_rule = classifier_lookup(cls, flow, wc);
5210 return rule_dpif_cast(rule_from_cls_rule(cls_rule));
5214 rule_dpif_miss_rule(struct ofproto_dpif *ofproto, const struct flow *flow)
5216 struct ofport_dpif *port;
5218 port = get_ofp_port(ofproto, flow->in_port.ofp_port);
5220 VLOG_WARN_RL(&rl, "packet-in on unknown OpenFlow port %"PRIu16,
5221 flow->in_port.ofp_port);
5222 return ofproto->miss_rule;
5225 if (port->up.pp.config & OFPUTIL_PC_NO_PACKET_IN) {
5226 return ofproto->no_packet_in_rule;
5228 return ofproto->miss_rule;
5232 complete_operation(struct rule_dpif *rule)
5234 struct ofproto_dpif *ofproto = ofproto_dpif_cast(rule->up.ofproto);
5236 rule_invalidate(rule);
5238 struct dpif_completion *c = xmalloc(sizeof *c);
5239 c->op = rule->up.pending;
5240 list_push_back(&ofproto->completions, &c->list_node);
5242 ofoperation_complete(rule->up.pending, 0);
5246 static struct rule *
5249 struct rule_dpif *rule = xmalloc(sizeof *rule);
5254 rule_dealloc(struct rule *rule_)
5256 struct rule_dpif *rule = rule_dpif_cast(rule_);
5261 rule_construct(struct rule *rule_)
5263 struct rule_dpif *rule = rule_dpif_cast(rule_);
5264 struct ofproto_dpif *ofproto = ofproto_dpif_cast(rule->up.ofproto);
5265 struct rule_dpif *victim;
5268 rule->packet_count = 0;
5269 rule->byte_count = 0;
5271 victim = rule_dpif_cast(ofoperation_get_victim(rule->up.pending));
5272 if (victim && !list_is_empty(&victim->facets)) {
5273 struct facet *facet;
5275 rule->facets = victim->facets;
5276 list_moved(&rule->facets);
5277 LIST_FOR_EACH (facet, list_node, &rule->facets) {
5278 /* XXX: We're only clearing our local counters here. It's possible
5279 * that quite a few packets are unaccounted for in the datapath
5280 * statistics. These will be accounted to the new rule instead of
5281 * cleared as required. This could be fixed by clearing out the
5282 * datapath statistics for this facet, but currently it doesn't
5284 facet_reset_counters(facet);
5288 /* Must avoid list_moved() in this case. */
5289 list_init(&rule->facets);
5292 table_id = rule->up.table_id;
5294 rule->tag = victim->tag;
5295 } else if (table_id == 0) {
5300 miniflow_expand(&rule->up.cr.match.flow, &flow);
5301 rule->tag = rule_calculate_tag(&flow, &rule->up.cr.match.mask,
5302 ofproto->tables[table_id].basis);
5305 complete_operation(rule);
5310 rule_destruct(struct rule *rule_)
5312 struct rule_dpif *rule = rule_dpif_cast(rule_);
5313 struct facet *facet, *next_facet;
5315 LIST_FOR_EACH_SAFE (facet, next_facet, list_node, &rule->facets) {
5316 facet_revalidate(facet);
5319 complete_operation(rule);
5323 rule_get_stats(struct rule *rule_, uint64_t *packets, uint64_t *bytes)
5325 struct rule_dpif *rule = rule_dpif_cast(rule_);
5327 /* push_all_stats() can handle flow misses which, when using the learn
5328 * action, can cause rules to be added and deleted. This can corrupt our
5329 * caller's datastructures which assume that rule_get_stats() doesn't have
5330 * an impact on the flow table. To be safe, we disable miss handling. */
5331 push_all_stats__(false);
5333 /* Start from historical data for 'rule' itself that are no longer tracked
5334 * in facets. This counts, for example, facets that have expired. */
5335 *packets = rule->packet_count;
5336 *bytes = rule->byte_count;
5340 rule_dpif_execute(struct rule_dpif *rule, const struct flow *flow,
5341 struct ofpbuf *packet)
5343 struct ofproto_dpif *ofproto = ofproto_dpif_cast(rule->up.ofproto);
5344 struct dpif_flow_stats stats;
5345 struct xlate_out xout;
5346 struct xlate_in xin;
5348 dpif_flow_stats_extract(flow, packet, time_msec(), &stats);
5349 rule_credit_stats(rule, &stats);
5351 xlate_in_init(&xin, ofproto, flow, rule, stats.tcp_flags, packet);
5352 xin.resubmit_stats = &stats;
5353 xlate_actions(&xin, &xout);
5355 execute_odp_actions(ofproto, flow, xout.odp_actions.data,
5356 xout.odp_actions.size, packet);
5358 xlate_out_uninit(&xout);
5362 rule_execute(struct rule *rule, const struct flow *flow,
5363 struct ofpbuf *packet)
5365 rule_dpif_execute(rule_dpif_cast(rule), flow, packet);
5366 ofpbuf_delete(packet);
5371 rule_modify_actions(struct rule *rule_)
5373 struct rule_dpif *rule = rule_dpif_cast(rule_);
5375 complete_operation(rule);
5378 /* Sends 'packet' out 'ofport'.
5379 * May modify 'packet'.
5380 * Returns 0 if successful, otherwise a positive errno value. */
5382 send_packet(const struct ofport_dpif *ofport, struct ofpbuf *packet)
5384 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport->up.ofproto);
5385 uint64_t odp_actions_stub[1024 / 8];
5386 struct ofpbuf key, odp_actions;
5387 struct dpif_flow_stats stats;
5388 struct odputil_keybuf keybuf;
5389 struct ofpact_output output;
5390 struct xlate_out xout;
5391 struct xlate_in xin;
5393 union flow_in_port in_port_;
5396 ofpbuf_use_stub(&odp_actions, odp_actions_stub, sizeof odp_actions_stub);
5397 ofpbuf_use_stack(&key, &keybuf, sizeof keybuf);
5399 /* Use OFPP_NONE as the in_port to avoid special packet processing. */
5400 in_port_.ofp_port = OFPP_NONE;
5401 flow_extract(packet, 0, 0, NULL, &in_port_, &flow);
5402 odp_flow_key_from_flow(&key, &flow, ofp_port_to_odp_port(ofproto,
5404 dpif_flow_stats_extract(&flow, packet, time_msec(), &stats);
5406 ofpact_init(&output.ofpact, OFPACT_OUTPUT, sizeof output);
5407 output.port = ofport->up.ofp_port;
5410 xlate_in_init(&xin, ofproto, &flow, NULL, 0, packet);
5411 xin.ofpacts_len = sizeof output;
5412 xin.ofpacts = &output.ofpact;
5413 xin.resubmit_stats = &stats;
5414 xlate_actions(&xin, &xout);
5416 error = dpif_execute(ofproto->backer->dpif,
5418 xout.odp_actions.data, xout.odp_actions.size,
5420 xlate_out_uninit(&xout);
5423 VLOG_WARN_RL(&rl, "%s: failed to send packet on port %s (%s)",
5424 ofproto->up.name, netdev_get_name(ofport->up.netdev),
5425 ovs_strerror(error));
5428 ofproto->stats.tx_packets++;
5429 ofproto->stats.tx_bytes += packet->size;
5433 /* Composes an ODP action for a "slow path" action for 'flow' within 'ofproto'.
5434 * The action will state 'slow' as the reason that the action is in the slow
5435 * path. (This is purely informational: it allows a human viewing "ovs-dpctl
5436 * dump-flows" output to see why a flow is in the slow path.)
5438 * The 'stub_size' bytes in 'stub' will be used to store the action.
5439 * 'stub_size' must be large enough for the action.
5441 * The action and its size will be stored in '*actionsp' and '*actions_lenp',
5444 compose_slow_path(const struct ofproto_dpif *ofproto, const struct flow *flow,
5445 enum slow_path_reason slow,
5446 uint64_t *stub, size_t stub_size,
5447 const struct nlattr **actionsp, size_t *actions_lenp)
5449 union user_action_cookie cookie;
5452 cookie.type = USER_ACTION_COOKIE_SLOW_PATH;
5453 cookie.slow_path.unused = 0;
5454 cookie.slow_path.reason = slow;
5456 ofpbuf_use_stack(&buf, stub, stub_size);
5457 if (slow & (SLOW_CFM | SLOW_BFD | SLOW_LACP | SLOW_STP)) {
5458 uint32_t pid = dpif_port_get_pid(ofproto->backer->dpif,
5460 odp_put_userspace_action(pid, &cookie, sizeof cookie.slow_path, &buf);
5462 put_userspace_action(ofproto, &buf, flow, &cookie,
5463 sizeof cookie.slow_path);
5465 *actionsp = buf.data;
5466 *actions_lenp = buf.size;
5470 put_userspace_action(const struct ofproto_dpif *ofproto,
5471 struct ofpbuf *odp_actions,
5472 const struct flow *flow,
5473 const union user_action_cookie *cookie,
5474 const size_t cookie_size)
5478 pid = dpif_port_get_pid(ofproto->backer->dpif,
5479 ofp_port_to_odp_port(ofproto,
5480 flow->in_port.ofp_port));
5482 return odp_put_userspace_action(pid, cookie, cookie_size, odp_actions);
5487 update_mirror_stats(struct ofproto_dpif *ofproto, mirror_mask_t mirrors,
5488 uint64_t packets, uint64_t bytes)
5494 for (; mirrors; mirrors = zero_rightmost_1bit(mirrors)) {
5497 m = ofproto->mirrors[mirror_mask_ffs(mirrors) - 1];
5500 /* In normal circumstances 'm' will not be NULL. However,
5501 * if mirrors are reconfigured, we can temporarily get out
5502 * of sync in facet_revalidate(). We could "correct" the
5503 * mirror list before reaching here, but doing that would
5504 * not properly account the traffic stats we've currently
5505 * accumulated for previous mirror configuration. */
5509 m->packet_count += packets;
5510 m->byte_count += bytes;
5515 calculate_flow_tag(struct ofproto_dpif *ofproto, const struct flow *flow,
5516 uint8_t table_id, struct rule_dpif *rule)
5518 if (table_id > 0 && table_id < N_TABLES) {
5519 struct table_dpif *table = &ofproto->tables[table_id];
5520 if (table->other_table) {
5521 return (rule && rule->tag
5523 : rule_calculate_tag(flow, &table->other_table->mask,
5531 /* Optimized flow revalidation.
5533 * It's a difficult problem, in general, to tell which facets need to have
5534 * their actions recalculated whenever the OpenFlow flow table changes. We
5535 * don't try to solve that general problem: for most kinds of OpenFlow flow
5536 * table changes, we recalculate the actions for every facet. This is
5537 * relatively expensive, but it's good enough if the OpenFlow flow table
5538 * doesn't change very often.
5540 * However, we can expect one particular kind of OpenFlow flow table change to
5541 * happen frequently: changes caused by MAC learning. To avoid wasting a lot
5542 * of CPU on revalidating every facet whenever MAC learning modifies the flow
5543 * table, we add a special case that applies to flow tables in which every rule
5544 * has the same form (that is, the same wildcards), except that the table is
5545 * also allowed to have a single "catch-all" flow that matches all packets. We
5546 * optimize this case by tagging all of the facets that resubmit into the table
5547 * and invalidating the same tag whenever a flow changes in that table. The
5548 * end result is that we revalidate just the facets that need it (and sometimes
5549 * a few more, but not all of the facets or even all of the facets that
5550 * resubmit to the table modified by MAC learning). */
5552 /* Calculates the tag to use for 'flow' and mask 'mask' when it is inserted
5553 * into an OpenFlow table with the given 'basis'. */
5555 rule_calculate_tag(const struct flow *flow, const struct minimask *mask,
5558 if (minimask_is_catchall(mask)) {
5561 uint32_t hash = flow_hash_in_minimask(flow, mask, secret);
5562 return tag_create_deterministic(hash);
5566 /* Following a change to OpenFlow table 'table_id' in 'ofproto', update the
5567 * taggability of that table.
5569 * This function must be called after *each* change to a flow table. If you
5570 * skip calling it on some changes then the pointer comparisons at the end can
5571 * be invalid if you get unlucky. For example, if a flow removal causes a
5572 * cls_table to be destroyed and then a flow insertion causes a cls_table with
5573 * different wildcards to be created with the same address, then this function
5574 * will incorrectly skip revalidation. */
5576 table_update_taggable(struct ofproto_dpif *ofproto, uint8_t table_id)
5578 struct table_dpif *table = &ofproto->tables[table_id];
5579 const struct oftable *oftable = &ofproto->up.tables[table_id];
5580 struct cls_table *catchall, *other;
5581 struct cls_table *t;
5583 catchall = other = NULL;
5585 switch (hmap_count(&oftable->cls.tables)) {
5587 /* We could tag this OpenFlow table but it would make the logic a
5588 * little harder and it's a corner case that doesn't seem worth it
5594 HMAP_FOR_EACH (t, hmap_node, &oftable->cls.tables) {
5595 if (cls_table_is_catchall(t)) {
5597 } else if (!other) {
5600 /* Indicate that we can't tag this by setting both tables to
5601 * NULL. (We know that 'catchall' is already NULL.) */
5608 /* Can't tag this table. */
5612 if (table->catchall_table != catchall || table->other_table != other) {
5613 table->catchall_table = catchall;
5614 table->other_table = other;
5615 ofproto->backer->need_revalidate = REV_FLOW_TABLE;
5619 /* Given 'rule' that has changed in some way (either it is a rule being
5620 * inserted, a rule being deleted, or a rule whose actions are being
5621 * modified), marks facets for revalidation to ensure that packets will be
5622 * forwarded correctly according to the new state of the flow table.
5624 * This function must be called after *each* change to a flow table. See
5625 * the comment on table_update_taggable() for more information. */
5627 rule_invalidate(const struct rule_dpif *rule)
5629 struct ofproto_dpif *ofproto = ofproto_dpif_cast(rule->up.ofproto);
5631 table_update_taggable(ofproto, rule->up.table_id);
5633 if (!ofproto->backer->need_revalidate) {
5634 struct table_dpif *table = &ofproto->tables[rule->up.table_id];
5636 if (table->other_table && rule->tag) {
5637 tag_set_add(&ofproto->backer->revalidate_set, rule->tag);
5639 ofproto->backer->need_revalidate = REV_FLOW_TABLE;
5645 set_frag_handling(struct ofproto *ofproto_,
5646 enum ofp_config_flags frag_handling)
5648 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
5649 if (frag_handling != OFPC_FRAG_REASM) {
5650 ofproto->backer->need_revalidate = REV_RECONFIGURE;
5658 packet_out(struct ofproto *ofproto_, struct ofpbuf *packet,
5659 const struct flow *flow,
5660 const struct ofpact *ofpacts, size_t ofpacts_len)
5662 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
5663 struct odputil_keybuf keybuf;
5664 struct dpif_flow_stats stats;
5665 struct xlate_out xout;
5666 struct xlate_in xin;
5670 ofpbuf_use_stack(&key, &keybuf, sizeof keybuf);
5671 odp_flow_key_from_flow(&key, flow,
5672 ofp_port_to_odp_port(ofproto,
5673 flow->in_port.ofp_port));
5675 dpif_flow_stats_extract(flow, packet, time_msec(), &stats);
5677 xlate_in_init(&xin, ofproto, flow, NULL, stats.tcp_flags, packet);
5678 xin.resubmit_stats = &stats;
5679 xin.ofpacts_len = ofpacts_len;
5680 xin.ofpacts = ofpacts;
5682 xlate_actions(&xin, &xout);
5683 dpif_execute(ofproto->backer->dpif, key.data, key.size,
5684 xout.odp_actions.data, xout.odp_actions.size, packet);
5685 xlate_out_uninit(&xout);
5693 set_netflow(struct ofproto *ofproto_,
5694 const struct netflow_options *netflow_options)
5696 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
5698 if (netflow_options) {
5699 if (!ofproto->netflow) {
5700 ofproto->netflow = netflow_create();
5701 ofproto->backer->need_revalidate = REV_RECONFIGURE;
5703 return netflow_set_options(ofproto->netflow, netflow_options);
5704 } else if (ofproto->netflow) {
5705 ofproto->backer->need_revalidate = REV_RECONFIGURE;
5706 netflow_destroy(ofproto->netflow);
5707 ofproto->netflow = NULL;
5714 get_netflow_ids(const struct ofproto *ofproto_,
5715 uint8_t *engine_type, uint8_t *engine_id)
5717 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
5719 dpif_get_netflow_ids(ofproto->backer->dpif, engine_type, engine_id);
5723 send_active_timeout(struct ofproto_dpif *ofproto, struct facet *facet)
5725 if (!facet_is_controller_flow(facet) &&
5726 netflow_active_timeout_expired(ofproto->netflow, &facet->nf_flow)) {
5727 struct subfacet *subfacet;
5728 struct ofexpired expired;
5730 LIST_FOR_EACH (subfacet, list_node, &facet->subfacets) {
5731 if (subfacet->path == SF_FAST_PATH) {
5732 struct dpif_flow_stats stats;
5734 subfacet_install(subfacet, &facet->xout.odp_actions,
5736 subfacet_update_stats(subfacet, &stats);
5740 expired.flow = facet->flow;
5741 expired.packet_count = facet->packet_count;
5742 expired.byte_count = facet->byte_count;
5743 expired.used = facet->used;
5744 netflow_expire(ofproto->netflow, &facet->nf_flow, &expired);
5749 send_netflow_active_timeouts(struct ofproto_dpif *ofproto)
5751 struct cls_cursor cursor;
5752 struct facet *facet;
5754 cls_cursor_init(&cursor, &ofproto->facets, NULL);
5755 CLS_CURSOR_FOR_EACH (facet, cr, &cursor) {
5756 send_active_timeout(ofproto, facet);
5760 static struct ofproto_dpif *
5761 ofproto_dpif_lookup(const char *name)
5763 struct ofproto_dpif *ofproto;
5765 HMAP_FOR_EACH_WITH_HASH (ofproto, all_ofproto_dpifs_node,
5766 hash_string(name, 0), &all_ofproto_dpifs) {
5767 if (!strcmp(ofproto->up.name, name)) {
5775 ofproto_unixctl_fdb_flush(struct unixctl_conn *conn, int argc,
5776 const char *argv[], void *aux OVS_UNUSED)
5778 struct ofproto_dpif *ofproto;
5781 ofproto = ofproto_dpif_lookup(argv[1]);
5783 unixctl_command_reply_error(conn, "no such bridge");
5786 mac_learning_flush(ofproto->ml, &ofproto->backer->revalidate_set);
5788 HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
5789 mac_learning_flush(ofproto->ml, &ofproto->backer->revalidate_set);
5793 unixctl_command_reply(conn, "table successfully flushed");
5797 ofproto_unixctl_fdb_show(struct unixctl_conn *conn, int argc OVS_UNUSED,
5798 const char *argv[], void *aux OVS_UNUSED)
5800 struct ds ds = DS_EMPTY_INITIALIZER;
5801 const struct ofproto_dpif *ofproto;
5802 const struct mac_entry *e;
5804 ofproto = ofproto_dpif_lookup(argv[1]);
5806 unixctl_command_reply_error(conn, "no such bridge");
5810 ds_put_cstr(&ds, " port VLAN MAC Age\n");
5811 LIST_FOR_EACH (e, lru_node, &ofproto->ml->lrus) {
5812 struct ofbundle *bundle = e->port.p;
5813 char name[OFP_MAX_PORT_NAME_LEN];
5815 ofputil_port_to_string(ofbundle_get_a_port(bundle)->up.ofp_port,
5817 ds_put_format(&ds, "%5s %4d "ETH_ADDR_FMT" %3d\n",
5818 name, e->vlan, ETH_ADDR_ARGS(e->mac),
5819 mac_entry_age(ofproto->ml, e));
5821 unixctl_command_reply(conn, ds_cstr(&ds));
5826 struct xlate_out xout;
5827 struct xlate_in xin;
5833 trace_format_rule(struct ds *result, int level, const struct rule_dpif *rule)
5835 ds_put_char_multiple(result, '\t', level);
5837 ds_put_cstr(result, "No match\n");
5841 ds_put_format(result, "Rule: table=%"PRIu8" cookie=%#"PRIx64" ",
5842 rule ? rule->up.table_id : 0, ntohll(rule->up.flow_cookie));
5843 cls_rule_format(&rule->up.cr, result);
5844 ds_put_char(result, '\n');
5846 ds_put_char_multiple(result, '\t', level);
5847 ds_put_cstr(result, "OpenFlow ");
5848 ofpacts_format(rule->up.ofpacts, rule->up.ofpacts_len, result);
5849 ds_put_char(result, '\n');
5853 trace_format_flow(struct ds *result, int level, const char *title,
5854 struct trace_ctx *trace)
5856 ds_put_char_multiple(result, '\t', level);
5857 ds_put_format(result, "%s: ", title);
5858 if (flow_equal(&trace->xin.flow, &trace->flow)) {
5859 ds_put_cstr(result, "unchanged");
5861 flow_format(result, &trace->xin.flow);
5862 trace->flow = trace->xin.flow;
5864 ds_put_char(result, '\n');
5868 trace_format_regs(struct ds *result, int level, const char *title,
5869 struct trace_ctx *trace)
5873 ds_put_char_multiple(result, '\t', level);
5874 ds_put_format(result, "%s:", title);
5875 for (i = 0; i < FLOW_N_REGS; i++) {
5876 ds_put_format(result, " reg%zu=0x%"PRIx32, i, trace->flow.regs[i]);
5878 ds_put_char(result, '\n');
5882 trace_format_odp(struct ds *result, int level, const char *title,
5883 struct trace_ctx *trace)
5885 struct ofpbuf *odp_actions = &trace->xout.odp_actions;
5887 ds_put_char_multiple(result, '\t', level);
5888 ds_put_format(result, "%s: ", title);
5889 format_odp_actions(result, odp_actions->data, odp_actions->size);
5890 ds_put_char(result, '\n');
5894 trace_resubmit(struct xlate_in *xin, struct rule_dpif *rule, int recurse)
5896 struct trace_ctx *trace = CONTAINER_OF(xin, struct trace_ctx, xin);
5897 struct ds *result = trace->result;
5899 ds_put_char(result, '\n');
5900 trace_format_flow(result, recurse + 1, "Resubmitted flow", trace);
5901 trace_format_regs(result, recurse + 1, "Resubmitted regs", trace);
5902 trace_format_odp(result, recurse + 1, "Resubmitted odp", trace);
5903 trace_format_rule(result, recurse + 1, rule);
5907 trace_report(struct xlate_in *xin, const char *s, int recurse)
5909 struct trace_ctx *trace = CONTAINER_OF(xin, struct trace_ctx, xin);
5910 struct ds *result = trace->result;
5912 ds_put_char_multiple(result, '\t', recurse);
5913 ds_put_cstr(result, s);
5914 ds_put_char(result, '\n');
5918 ofproto_unixctl_trace(struct unixctl_conn *conn, int argc, const char *argv[],
5919 void *aux OVS_UNUSED)
5921 const struct dpif_backer *backer;
5922 struct ofproto_dpif *ofproto;
5923 struct ofpbuf odp_key, odp_mask;
5924 struct ofpbuf *packet;
5932 ofpbuf_init(&odp_key, 0);
5933 ofpbuf_init(&odp_mask, 0);
5935 /* Handle "-generate" or a hex string as the last argument. */
5936 if (!strcmp(argv[argc - 1], "-generate")) {
5937 packet = ofpbuf_new(0);
5940 const char *error = eth_from_hex(argv[argc - 1], &packet);
5943 } else if (argc == 4) {
5944 /* The 3-argument form must end in "-generate' or a hex string. */
5945 unixctl_command_reply_error(conn, error);
5950 /* Parse the flow and determine whether a datapath or
5951 * bridge is specified. If function odp_flow_key_from_string()
5952 * returns 0, the flow is a odp_flow. If function
5953 * parse_ofp_exact_flow() returns 0, the flow is a br_flow. */
5954 if (!odp_flow_from_string(argv[argc - 1], NULL, &odp_key, &odp_mask)) {
5955 /* If the odp_flow is the second argument,
5956 * the datapath name is the first argument. */
5958 const char *dp_type;
5959 if (!strncmp(argv[1], "ovs-", 4)) {
5960 dp_type = argv[1] + 4;
5964 backer = shash_find_data(&all_dpif_backers, dp_type);
5966 unixctl_command_reply_error(conn, "Cannot find datapath "
5971 /* No datapath name specified, so there should be only one
5973 struct shash_node *node;
5974 if (shash_count(&all_dpif_backers) != 1) {
5975 unixctl_command_reply_error(conn, "Must specify datapath "
5976 "name, there is more than one type of datapath");
5979 node = shash_first(&all_dpif_backers);
5980 backer = node->data;
5983 /* Extract the ofproto_dpif object from the ofproto_receive()
5985 if (ofproto_receive(backer, NULL, odp_key.data,
5986 odp_key.size, &flow, NULL, &ofproto, NULL)) {
5987 unixctl_command_reply_error(conn, "Invalid datapath flow");
5990 ds_put_format(&result, "Bridge: %s\n", ofproto->up.name);
5991 } else if (!parse_ofp_exact_flow(&flow, argv[argc - 1])) {
5993 unixctl_command_reply_error(conn, "Must specify bridge name");
5997 ofproto = ofproto_dpif_lookup(argv[1]);
5999 unixctl_command_reply_error(conn, "Unknown bridge name");
6003 unixctl_command_reply_error(conn, "Bad flow syntax");
6007 /* Generate a packet, if requested. */
6009 if (!packet->size) {
6010 flow_compose(packet, &flow);
6012 union flow_in_port in_port_;
6014 in_port_ = flow.in_port;
6015 ds_put_cstr(&result, "Packet: ");
6016 s = ofp_packet_to_string(packet->data, packet->size);
6017 ds_put_cstr(&result, s);
6020 /* Use the metadata from the flow and the packet argument
6021 * to reconstruct the flow. */
6022 flow_extract(packet, flow.skb_priority, flow.skb_mark, NULL,
6027 ofproto_trace(ofproto, &flow, packet, &result);
6028 unixctl_command_reply(conn, ds_cstr(&result));
6031 ds_destroy(&result);
6032 ofpbuf_delete(packet);
6033 ofpbuf_uninit(&odp_key);
6034 ofpbuf_uninit(&odp_mask);
6038 ofproto_trace(struct ofproto_dpif *ofproto, const struct flow *flow,
6039 const struct ofpbuf *packet, struct ds *ds)
6041 struct rule_dpif *rule;
6043 ds_put_cstr(ds, "Flow: ");
6044 flow_format(ds, flow);
6045 ds_put_char(ds, '\n');
6047 rule = rule_dpif_lookup(ofproto, flow, NULL);
6049 trace_format_rule(ds, 0, rule);
6050 if (rule == ofproto->miss_rule) {
6051 ds_put_cstr(ds, "\nNo match, flow generates \"packet in\"s.\n");
6052 } else if (rule == ofproto->no_packet_in_rule) {
6053 ds_put_cstr(ds, "\nNo match, packets dropped because "
6054 "OFPPC_NO_PACKET_IN is set on in_port.\n");
6055 } else if (rule == ofproto->drop_frags_rule) {
6056 ds_put_cstr(ds, "\nPackets dropped because they are IP fragments "
6057 "and the fragment handling mode is \"drop\".\n");
6061 uint64_t odp_actions_stub[1024 / 8];
6062 struct ofpbuf odp_actions;
6063 struct trace_ctx trace;
6067 tcp_flags = packet ? packet_get_tcp_flags(packet, flow) : 0;
6070 ofpbuf_use_stub(&odp_actions,
6071 odp_actions_stub, sizeof odp_actions_stub);
6072 xlate_in_init(&trace.xin, ofproto, flow, rule, tcp_flags, packet);
6073 trace.xin.resubmit_hook = trace_resubmit;
6074 trace.xin.report_hook = trace_report;
6076 xlate_actions(&trace.xin, &trace.xout);
6078 ds_put_char(ds, '\n');
6079 trace_format_flow(ds, 0, "Final flow", &trace);
6081 match_init(&match, flow, &trace.xout.wc);
6082 ds_put_cstr(ds, "Relevant fields: ");
6083 match_format(&match, ds, OFP_DEFAULT_PRIORITY);
6084 ds_put_char(ds, '\n');
6086 ds_put_cstr(ds, "Datapath actions: ");
6087 format_odp_actions(ds, trace.xout.odp_actions.data,
6088 trace.xout.odp_actions.size);
6090 if (trace.xout.slow) {
6091 ds_put_cstr(ds, "\nThis flow is handled by the userspace "
6092 "slow path because it:");
6093 switch (trace.xout.slow) {
6095 ds_put_cstr(ds, "\n\t- Consists of CFM packets.");
6098 ds_put_cstr(ds, "\n\t- Consists of LACP packets.");
6101 ds_put_cstr(ds, "\n\t- Consists of STP packets.");
6104 ds_put_cstr(ds, "\n\t- Consists of BFD packets.");
6106 case SLOW_CONTROLLER:
6107 ds_put_cstr(ds, "\n\t- Sends \"packet-in\" messages "
6108 "to the OpenFlow controller.");
6115 xlate_out_uninit(&trace.xout);
6120 ofproto_dpif_clog(struct unixctl_conn *conn OVS_UNUSED, int argc OVS_UNUSED,
6121 const char *argv[] OVS_UNUSED, void *aux OVS_UNUSED)
6124 unixctl_command_reply(conn, NULL);
6128 ofproto_dpif_unclog(struct unixctl_conn *conn OVS_UNUSED, int argc OVS_UNUSED,
6129 const char *argv[] OVS_UNUSED, void *aux OVS_UNUSED)
6132 unixctl_command_reply(conn, NULL);
6135 /* Runs a self-check of flow translations in 'ofproto'. Appends a message to
6136 * 'reply' describing the results. */
6138 ofproto_dpif_self_check__(struct ofproto_dpif *ofproto, struct ds *reply)
6140 struct cls_cursor cursor;
6141 struct facet *facet;
6145 cls_cursor_init(&cursor, &ofproto->facets, NULL);
6146 CLS_CURSOR_FOR_EACH (facet, cr, &cursor) {
6147 if (!facet_check_consistency(facet)) {
6152 ofproto->backer->need_revalidate = REV_INCONSISTENCY;
6156 ds_put_format(reply, "%s: self-check failed (%d errors)\n",
6157 ofproto->up.name, errors);
6159 ds_put_format(reply, "%s: self-check passed\n", ofproto->up.name);
6164 ofproto_dpif_self_check(struct unixctl_conn *conn,
6165 int argc, const char *argv[], void *aux OVS_UNUSED)
6167 struct ds reply = DS_EMPTY_INITIALIZER;
6168 struct ofproto_dpif *ofproto;
6171 ofproto = ofproto_dpif_lookup(argv[1]);
6173 unixctl_command_reply_error(conn, "Unknown ofproto (use "
6174 "ofproto/list for help)");
6177 ofproto_dpif_self_check__(ofproto, &reply);
6179 HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
6180 ofproto_dpif_self_check__(ofproto, &reply);
6184 unixctl_command_reply(conn, ds_cstr(&reply));
6188 /* Store the current ofprotos in 'ofproto_shash'. Returns a sorted list
6189 * of the 'ofproto_shash' nodes. It is the responsibility of the caller
6190 * to destroy 'ofproto_shash' and free the returned value. */
6191 static const struct shash_node **
6192 get_ofprotos(struct shash *ofproto_shash)
6194 const struct ofproto_dpif *ofproto;
6196 HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
6197 char *name = xasprintf("%s@%s", ofproto->up.type, ofproto->up.name);
6198 shash_add_nocopy(ofproto_shash, name, ofproto);
6201 return shash_sort(ofproto_shash);
6205 ofproto_unixctl_dpif_dump_dps(struct unixctl_conn *conn, int argc OVS_UNUSED,
6206 const char *argv[] OVS_UNUSED,
6207 void *aux OVS_UNUSED)
6209 struct ds ds = DS_EMPTY_INITIALIZER;
6210 struct shash ofproto_shash;
6211 const struct shash_node **sorted_ofprotos;
6214 shash_init(&ofproto_shash);
6215 sorted_ofprotos = get_ofprotos(&ofproto_shash);
6216 for (i = 0; i < shash_count(&ofproto_shash); i++) {
6217 const struct shash_node *node = sorted_ofprotos[i];
6218 ds_put_format(&ds, "%s\n", node->name);
6221 shash_destroy(&ofproto_shash);
6222 free(sorted_ofprotos);
6224 unixctl_command_reply(conn, ds_cstr(&ds));
6229 show_dp_rates(struct ds *ds, const char *heading,
6230 const struct avg_subfacet_rates *rates)
6232 ds_put_format(ds, "%s add rate: %5.3f/min, del rate: %5.3f/min\n",
6233 heading, rates->add_rate, rates->del_rate);
6237 dpif_show_backer(const struct dpif_backer *backer, struct ds *ds)
6239 const struct shash_node **ofprotos;
6240 struct ofproto_dpif *ofproto;
6241 struct shash ofproto_shash;
6242 uint64_t n_hit, n_missed;
6243 long long int minutes;
6246 n_hit = n_missed = 0;
6247 HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
6248 if (ofproto->backer == backer) {
6249 n_missed += ofproto->n_missed;
6250 n_hit += ofproto->n_hit;
6254 ds_put_format(ds, "%s: hit:%"PRIu64" missed:%"PRIu64"\n",
6255 dpif_name(backer->dpif), n_hit, n_missed);
6256 ds_put_format(ds, "\tflows: cur: %zu, avg: %u, max: %u,"
6257 " life span: %lldms\n", hmap_count(&backer->subfacets),
6258 backer->avg_n_subfacet, backer->max_n_subfacet,
6259 backer->avg_subfacet_life);
6261 minutes = (time_msec() - backer->created) / (1000 * 60);
6262 if (minutes >= 60) {
6263 show_dp_rates(ds, "\thourly avg:", &backer->hourly);
6265 if (minutes >= 60 * 24) {
6266 show_dp_rates(ds, "\tdaily avg:", &backer->daily);
6268 show_dp_rates(ds, "\toverall avg:", &backer->lifetime);
6270 shash_init(&ofproto_shash);
6271 ofprotos = get_ofprotos(&ofproto_shash);
6272 for (i = 0; i < shash_count(&ofproto_shash); i++) {
6273 struct ofproto_dpif *ofproto = ofprotos[i]->data;
6274 const struct shash_node **ports;
6277 if (ofproto->backer != backer) {
6281 ds_put_format(ds, "\t%s: hit:%"PRIu64" missed:%"PRIu64"\n",
6282 ofproto->up.name, ofproto->n_hit, ofproto->n_missed);
6284 ports = shash_sort(&ofproto->up.port_by_name);
6285 for (j = 0; j < shash_count(&ofproto->up.port_by_name); j++) {
6286 const struct shash_node *node = ports[j];
6287 struct ofport *ofport = node->data;
6289 odp_port_t odp_port;
6291 ds_put_format(ds, "\t\t%s %u/", netdev_get_name(ofport->netdev),
6294 odp_port = ofp_port_to_odp_port(ofproto, ofport->ofp_port);
6295 if (odp_port != ODPP_NONE) {
6296 ds_put_format(ds, "%"PRIu32":", odp_port);
6298 ds_put_cstr(ds, "none:");
6301 ds_put_format(ds, " (%s", netdev_get_type(ofport->netdev));
6304 if (!netdev_get_config(ofport->netdev, &config)) {
6305 const struct smap_node **nodes;
6308 nodes = smap_sort(&config);
6309 for (i = 0; i < smap_count(&config); i++) {
6310 const struct smap_node *node = nodes[i];
6311 ds_put_format(ds, "%c %s=%s", i ? ',' : ':',
6312 node->key, node->value);
6316 smap_destroy(&config);
6318 ds_put_char(ds, ')');
6319 ds_put_char(ds, '\n');
6323 shash_destroy(&ofproto_shash);
6328 ofproto_unixctl_dpif_show(struct unixctl_conn *conn, int argc OVS_UNUSED,
6329 const char *argv[] OVS_UNUSED, void *aux OVS_UNUSED)
6331 struct ds ds = DS_EMPTY_INITIALIZER;
6332 const struct shash_node **backers;
6335 backers = shash_sort(&all_dpif_backers);
6336 for (i = 0; i < shash_count(&all_dpif_backers); i++) {
6337 dpif_show_backer(backers[i]->data, &ds);
6341 unixctl_command_reply(conn, ds_cstr(&ds));
6345 /* Dump the megaflow (facet) cache. This is useful to check the
6346 * correctness of flow wildcarding, since the same mechanism is used for
6347 * both xlate caching and kernel wildcarding.
6349 * It's important to note that in the output the flow description uses
6350 * OpenFlow (OFP) ports, but the actions use datapath (ODP) ports.
6352 * This command is only needed for advanced debugging, so it's not
6353 * documented in the man page. */
6355 ofproto_unixctl_dpif_dump_megaflows(struct unixctl_conn *conn,
6356 int argc OVS_UNUSED, const char *argv[],
6357 void *aux OVS_UNUSED)
6359 struct ds ds = DS_EMPTY_INITIALIZER;
6360 const struct ofproto_dpif *ofproto;
6361 long long int now = time_msec();
6362 struct cls_cursor cursor;
6363 struct facet *facet;
6365 ofproto = ofproto_dpif_lookup(argv[1]);
6367 unixctl_command_reply_error(conn, "no such bridge");
6371 cls_cursor_init(&cursor, &ofproto->facets, NULL);
6372 CLS_CURSOR_FOR_EACH (facet, cr, &cursor) {
6373 cls_rule_format(&facet->cr, &ds);
6374 ds_put_cstr(&ds, ", ");
6375 ds_put_format(&ds, "n_subfacets:%zu, ", list_size(&facet->subfacets));
6376 ds_put_format(&ds, "used:%.3fs, ", (now - facet->used) / 1000.0);
6377 ds_put_cstr(&ds, "Datapath actions: ");
6378 if (facet->xout.slow) {
6379 uint64_t slow_path_stub[128 / 8];
6380 const struct nlattr *actions;
6383 compose_slow_path(ofproto, &facet->flow, facet->xout.slow,
6384 slow_path_stub, sizeof slow_path_stub,
6385 &actions, &actions_len);
6386 format_odp_actions(&ds, actions, actions_len);
6388 format_odp_actions(&ds, facet->xout.odp_actions.data,
6389 facet->xout.odp_actions.size);
6391 ds_put_cstr(&ds, "\n");
6394 ds_chomp(&ds, '\n');
6395 unixctl_command_reply(conn, ds_cstr(&ds));
6399 /* Disable using the megaflows.
6401 * This command is only needed for advanced debugging, so it's not
6402 * documented in the man page. */
6404 ofproto_unixctl_dpif_disable_megaflows(struct unixctl_conn *conn,
6405 int argc OVS_UNUSED,
6406 const char *argv[] OVS_UNUSED,
6407 void *aux OVS_UNUSED)
6409 struct ofproto_dpif *ofproto;
6411 enable_megaflows = false;
6413 HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
6414 flush(&ofproto->up);
6417 unixctl_command_reply(conn, "megaflows disabled");
6420 /* Re-enable using megaflows.
6422 * This command is only needed for advanced debugging, so it's not
6423 * documented in the man page. */
6425 ofproto_unixctl_dpif_enable_megaflows(struct unixctl_conn *conn,
6426 int argc OVS_UNUSED,
6427 const char *argv[] OVS_UNUSED,
6428 void *aux OVS_UNUSED)
6430 struct ofproto_dpif *ofproto;
6432 enable_megaflows = true;
6434 HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
6435 flush(&ofproto->up);
6438 unixctl_command_reply(conn, "megaflows enabled");
6442 ofproto_unixctl_dpif_dump_flows(struct unixctl_conn *conn,
6443 int argc OVS_UNUSED, const char *argv[],
6444 void *aux OVS_UNUSED)
6446 struct ds ds = DS_EMPTY_INITIALIZER;
6447 const struct ofproto_dpif *ofproto;
6448 struct subfacet *subfacet;
6450 ofproto = ofproto_dpif_lookup(argv[1]);
6452 unixctl_command_reply_error(conn, "no such bridge");
6456 update_stats(ofproto->backer);
6458 HMAP_FOR_EACH (subfacet, hmap_node, &ofproto->backer->subfacets) {
6459 struct facet *facet = subfacet->facet;
6461 if (ofproto_dpif_cast(facet->rule->up.ofproto) != ofproto) {
6465 odp_flow_key_format(subfacet->key, subfacet->key_len, &ds);
6467 ds_put_format(&ds, ", packets:%"PRIu64", bytes:%"PRIu64", used:",
6468 subfacet->dp_packet_count, subfacet->dp_byte_count);
6469 if (subfacet->used) {
6470 ds_put_format(&ds, "%.3fs",
6471 (time_msec() - subfacet->used) / 1000.0);
6473 ds_put_format(&ds, "never");
6475 if (subfacet->facet->tcp_flags) {
6476 ds_put_cstr(&ds, ", flags:");
6477 packet_format_tcp_flags(&ds, subfacet->facet->tcp_flags);
6480 ds_put_cstr(&ds, ", actions:");
6481 if (facet->xout.slow) {
6482 uint64_t slow_path_stub[128 / 8];
6483 const struct nlattr *actions;
6486 compose_slow_path(ofproto, &facet->flow, facet->xout.slow,
6487 slow_path_stub, sizeof slow_path_stub,
6488 &actions, &actions_len);
6489 format_odp_actions(&ds, actions, actions_len);
6491 format_odp_actions(&ds, facet->xout.odp_actions.data,
6492 facet->xout.odp_actions.size);
6494 ds_put_char(&ds, '\n');
6497 unixctl_command_reply(conn, ds_cstr(&ds));
6502 ofproto_unixctl_dpif_del_flows(struct unixctl_conn *conn,
6503 int argc OVS_UNUSED, const char *argv[],
6504 void *aux OVS_UNUSED)
6506 struct ds ds = DS_EMPTY_INITIALIZER;
6507 struct ofproto_dpif *ofproto;
6509 ofproto = ofproto_dpif_lookup(argv[1]);
6511 unixctl_command_reply_error(conn, "no such bridge");
6515 flush(&ofproto->up);
6517 unixctl_command_reply(conn, ds_cstr(&ds));
6522 ofproto_dpif_unixctl_init(void)
6524 static bool registered;
6530 unixctl_command_register(
6532 "[dp_name]|bridge odp_flow|br_flow [-generate|packet]",
6533 1, 3, ofproto_unixctl_trace, NULL);
6534 unixctl_command_register("fdb/flush", "[bridge]", 0, 1,
6535 ofproto_unixctl_fdb_flush, NULL);
6536 unixctl_command_register("fdb/show", "bridge", 1, 1,
6537 ofproto_unixctl_fdb_show, NULL);
6538 unixctl_command_register("ofproto/clog", "", 0, 0,
6539 ofproto_dpif_clog, NULL);
6540 unixctl_command_register("ofproto/unclog", "", 0, 0,
6541 ofproto_dpif_unclog, NULL);
6542 unixctl_command_register("ofproto/self-check", "[bridge]", 0, 1,
6543 ofproto_dpif_self_check, NULL);
6544 unixctl_command_register("dpif/dump-dps", "", 0, 0,
6545 ofproto_unixctl_dpif_dump_dps, NULL);
6546 unixctl_command_register("dpif/show", "", 0, 0, ofproto_unixctl_dpif_show,
6548 unixctl_command_register("dpif/dump-flows", "bridge", 1, 1,
6549 ofproto_unixctl_dpif_dump_flows, NULL);
6550 unixctl_command_register("dpif/del-flows", "bridge", 1, 1,
6551 ofproto_unixctl_dpif_del_flows, NULL);
6552 unixctl_command_register("dpif/dump-megaflows", "bridge", 1, 1,
6553 ofproto_unixctl_dpif_dump_megaflows, NULL);
6554 unixctl_command_register("dpif/disable-megaflows", "", 0, 0,
6555 ofproto_unixctl_dpif_disable_megaflows, NULL);
6556 unixctl_command_register("dpif/enable-megaflows", "", 0, 0,
6557 ofproto_unixctl_dpif_enable_megaflows, NULL);
6560 /* Linux VLAN device support (e.g. "eth0.10" for VLAN 10.)
6562 * This is deprecated. It is only for compatibility with broken device drivers
6563 * in old versions of Linux that do not properly support VLANs when VLAN
6564 * devices are not used. When broken device drivers are no longer in
6565 * widespread use, we will delete these interfaces. */
6568 set_realdev(struct ofport *ofport_, ofp_port_t realdev_ofp_port, int vid)
6570 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport_->ofproto);
6571 struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
6573 if (realdev_ofp_port == ofport->realdev_ofp_port
6574 && vid == ofport->vlandev_vid) {
6578 ofproto->backer->need_revalidate = REV_RECONFIGURE;
6580 if (ofport->realdev_ofp_port) {
6583 if (realdev_ofp_port && ofport->bundle) {
6584 /* vlandevs are enslaved to their realdevs, so they are not allowed to
6585 * themselves be part of a bundle. */
6586 bundle_set(ofport->up.ofproto, ofport->bundle, NULL);
6589 ofport->realdev_ofp_port = realdev_ofp_port;
6590 ofport->vlandev_vid = vid;
6592 if (realdev_ofp_port) {
6593 vsp_add(ofport, realdev_ofp_port, vid);
6600 hash_realdev_vid(ofp_port_t realdev_ofp_port, int vid)
6602 return hash_2words(ofp_to_u16(realdev_ofp_port), vid);
6605 /* Returns the OFP port number of the Linux VLAN device that corresponds to
6606 * 'vlan_tci' on the network device with port number 'realdev_ofp_port' in
6607 * 'struct ofport_dpif'. For example, given 'realdev_ofp_port' of eth0 and
6608 * 'vlan_tci' 9, it would return the port number of eth0.9.
6610 * Unless VLAN splinters are enabled for port 'realdev_ofp_port', this
6611 * function just returns its 'realdev_ofp_port' argument. */
6613 vsp_realdev_to_vlandev(const struct ofproto_dpif *ofproto,
6614 ofp_port_t realdev_ofp_port, ovs_be16 vlan_tci)
6616 if (!hmap_is_empty(&ofproto->realdev_vid_map)) {
6617 int vid = vlan_tci_to_vid(vlan_tci);
6618 const struct vlan_splinter *vsp;
6620 HMAP_FOR_EACH_WITH_HASH (vsp, realdev_vid_node,
6621 hash_realdev_vid(realdev_ofp_port, vid),
6622 &ofproto->realdev_vid_map) {
6623 if (vsp->realdev_ofp_port == realdev_ofp_port
6624 && vsp->vid == vid) {
6625 return vsp->vlandev_ofp_port;
6629 return realdev_ofp_port;
6632 static struct vlan_splinter *
6633 vlandev_find(const struct ofproto_dpif *ofproto, ofp_port_t vlandev_ofp_port)
6635 struct vlan_splinter *vsp;
6637 HMAP_FOR_EACH_WITH_HASH (vsp, vlandev_node,
6638 hash_ofp_port(vlandev_ofp_port),
6639 &ofproto->vlandev_map) {
6640 if (vsp->vlandev_ofp_port == vlandev_ofp_port) {
6648 /* Returns the OpenFlow port number of the "real" device underlying the Linux
6649 * VLAN device with OpenFlow port number 'vlandev_ofp_port' and stores the
6650 * VLAN VID of the Linux VLAN device in '*vid'. For example, given
6651 * 'vlandev_ofp_port' of eth0.9, it would return the OpenFlow port number of
6652 * eth0 and store 9 in '*vid'.
6654 * Returns 0 and does not modify '*vid' if 'vlandev_ofp_port' is not a Linux
6655 * VLAN device. Unless VLAN splinters are enabled, this is what this function
6658 vsp_vlandev_to_realdev(const struct ofproto_dpif *ofproto,
6659 ofp_port_t vlandev_ofp_port, int *vid)
6661 if (!hmap_is_empty(&ofproto->vlandev_map)) {
6662 const struct vlan_splinter *vsp;
6664 vsp = vlandev_find(ofproto, vlandev_ofp_port);
6669 return vsp->realdev_ofp_port;
6675 /* Given 'flow', a flow representing a packet received on 'ofproto', checks
6676 * whether 'flow->in_port' represents a Linux VLAN device. If so, changes
6677 * 'flow->in_port' to the "real" device backing the VLAN device, sets
6678 * 'flow->vlan_tci' to the VLAN VID, and returns true. Otherwise (which is
6679 * always the case unless VLAN splinters are enabled), returns false without
6680 * making any changes. */
6682 vsp_adjust_flow(const struct ofproto_dpif *ofproto, struct flow *flow)
6687 realdev = vsp_vlandev_to_realdev(ofproto, flow->in_port.ofp_port, &vid);
6692 /* Cause the flow to be processed as if it came in on the real device with
6693 * the VLAN device's VLAN ID. */
6694 flow->in_port.ofp_port = realdev;
6695 flow->vlan_tci = htons((vid & VLAN_VID_MASK) | VLAN_CFI);
6700 vsp_remove(struct ofport_dpif *port)
6702 struct ofproto_dpif *ofproto = ofproto_dpif_cast(port->up.ofproto);
6703 struct vlan_splinter *vsp;
6705 vsp = vlandev_find(ofproto, port->up.ofp_port);
6707 hmap_remove(&ofproto->vlandev_map, &vsp->vlandev_node);
6708 hmap_remove(&ofproto->realdev_vid_map, &vsp->realdev_vid_node);
6711 port->realdev_ofp_port = 0;
6713 VLOG_ERR("missing vlan device record");
6718 vsp_add(struct ofport_dpif *port, ofp_port_t realdev_ofp_port, int vid)
6720 struct ofproto_dpif *ofproto = ofproto_dpif_cast(port->up.ofproto);
6722 if (!vsp_vlandev_to_realdev(ofproto, port->up.ofp_port, NULL)
6723 && (vsp_realdev_to_vlandev(ofproto, realdev_ofp_port, htons(vid))
6724 == realdev_ofp_port)) {
6725 struct vlan_splinter *vsp;
6727 vsp = xmalloc(sizeof *vsp);
6728 hmap_insert(&ofproto->vlandev_map, &vsp->vlandev_node,
6729 hash_ofp_port(port->up.ofp_port));
6730 hmap_insert(&ofproto->realdev_vid_map, &vsp->realdev_vid_node,
6731 hash_realdev_vid(realdev_ofp_port, vid));
6732 vsp->realdev_ofp_port = realdev_ofp_port;
6733 vsp->vlandev_ofp_port = port->up.ofp_port;
6736 port->realdev_ofp_port = realdev_ofp_port;
6738 VLOG_ERR("duplicate vlan device record");
6743 ofp_port_to_odp_port(const struct ofproto_dpif *ofproto, ofp_port_t ofp_port)
6745 const struct ofport_dpif *ofport = get_ofp_port(ofproto, ofp_port);
6746 return ofport ? ofport->odp_port : ODPP_NONE;
6749 static struct ofport_dpif *
6750 odp_port_to_ofport(const struct dpif_backer *backer, odp_port_t odp_port)
6752 struct ofport_dpif *port;
6754 HMAP_FOR_EACH_IN_BUCKET (port, odp_port_node, hash_odp_port(odp_port),
6755 &backer->odp_to_ofport_map) {
6756 if (port->odp_port == odp_port) {
6765 odp_port_to_ofp_port(const struct ofproto_dpif *ofproto, odp_port_t odp_port)
6767 struct ofport_dpif *port;
6769 port = odp_port_to_ofport(ofproto->backer, odp_port);
6770 if (port && &ofproto->up == port->up.ofproto) {
6771 return port->up.ofp_port;
6777 /* Compute exponentially weighted moving average, adding 'new' as the newest,
6778 * most heavily weighted element. 'base' designates the rate of decay: after
6779 * 'base' further updates, 'new''s weight in the EWMA decays to about 1/e
6782 exp_mavg(double *avg, int base, double new)
6784 *avg = (*avg * (base - 1) + new) / base;
6788 update_moving_averages(struct dpif_backer *backer)
6790 const int min_ms = 60 * 1000; /* milliseconds in one minute. */
6791 long long int minutes = (time_msec() - backer->created) / min_ms;
6794 backer->lifetime.add_rate = (double) backer->total_subfacet_add_count
6796 backer->lifetime.del_rate = (double) backer->total_subfacet_del_count
6799 backer->lifetime.add_rate = 0.0;
6800 backer->lifetime.del_rate = 0.0;
6803 /* Update hourly averages on the minute boundaries. */
6804 if (time_msec() - backer->last_minute >= min_ms) {
6805 exp_mavg(&backer->hourly.add_rate, 60, backer->subfacet_add_count);
6806 exp_mavg(&backer->hourly.del_rate, 60, backer->subfacet_del_count);
6808 /* Update daily averages on the hour boundaries. */
6809 if ((backer->last_minute - backer->created) / min_ms % 60 == 59) {
6810 exp_mavg(&backer->daily.add_rate, 24, backer->hourly.add_rate);
6811 exp_mavg(&backer->daily.del_rate, 24, backer->hourly.del_rate);
6814 backer->total_subfacet_add_count += backer->subfacet_add_count;
6815 backer->total_subfacet_del_count += backer->subfacet_del_count;
6816 backer->subfacet_add_count = 0;
6817 backer->subfacet_del_count = 0;
6818 backer->last_minute += min_ms;
6822 const struct ofproto_class ofproto_dpif_class = {
6857 port_is_lacp_current,
6858 NULL, /* rule_choose_table */
6865 rule_modify_actions,
6879 get_stp_port_status,
6886 is_mirror_output_bundle,
6887 forward_bpdu_changed,
6888 set_mac_table_config,
6890 NULL, /* meter_get_features */
6891 NULL, /* meter_set */
6892 NULL, /* meter_get */
6893 NULL, /* meter_del */