2 * Copyright (c) 2009, 2010, 2011, 2012, 2013 Nicira, Inc.
4 * Licensed under the Apache License, Version 2.0 (the "License");
5 * you may not use this file except in compliance with the License.
6 * You may obtain a copy of the License at:
8 * http://www.apache.org/licenses/LICENSE-2.0
10 * Unless required by applicable law or agreed to in writing, software
11 * distributed under the License is distributed on an "AS IS" BASIS,
12 * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
13 * See the License for the specific language governing permissions and
14 * limitations under the License.
19 #include "ofproto/ofproto-dpif.h"
20 #include "ofproto/ofproto-provider.h"
27 #include "byte-order.h"
32 #include "dynamic-string.h"
33 #include "fail-open.h"
37 #include "mac-learning.h"
38 #include "meta-flow.h"
39 #include "multipath.h"
40 #include "netdev-vport.h"
45 #include "odp-execute.h"
48 #include "ofp-actions.h"
49 #include "ofp-parse.h"
50 #include "ofp-print.h"
51 #include "ofproto-dpif-governor.h"
52 #include "ofproto-dpif-ipfix.h"
53 #include "ofproto-dpif-sflow.h"
54 #include "ofproto-dpif-xlate.h"
55 #include "poll-loop.h"
60 #include "unaligned.h"
62 #include "vlan-bitmap.h"
65 VLOG_DEFINE_THIS_MODULE(ofproto_dpif);
67 COVERAGE_DEFINE(ofproto_dpif_expired);
68 COVERAGE_DEFINE(facet_changed_rule);
69 COVERAGE_DEFINE(facet_revalidate);
70 COVERAGE_DEFINE(facet_unexpected);
71 COVERAGE_DEFINE(facet_suppress);
76 static struct rule_dpif *rule_dpif_lookup(struct ofproto_dpif *,
78 struct flow_wildcards *wc);
80 static void rule_get_stats(struct rule *, uint64_t *packets, uint64_t *bytes);
81 static void rule_invalidate(const struct rule_dpif *);
83 static void mirror_destroy(struct ofmirror *);
84 static void update_mirror_stats(struct ofproto_dpif *ofproto,
85 mirror_mask_t mirrors,
86 uint64_t packets, uint64_t bytes);
88 static void bundle_remove(struct ofport *);
89 static void bundle_update(struct ofbundle *);
90 static void bundle_destroy(struct ofbundle *);
91 static void bundle_del_port(struct ofport_dpif *);
92 static void bundle_run(struct ofbundle *);
93 static void bundle_wait(struct ofbundle *);
95 static void stp_run(struct ofproto_dpif *ofproto);
96 static void stp_wait(struct ofproto_dpif *ofproto);
97 static int set_stp_port(struct ofport *,
98 const struct ofproto_port_stp_settings *);
100 static void compose_slow_path(const struct ofproto_dpif *, const struct flow *,
101 enum slow_path_reason,
102 uint64_t *stub, size_t stub_size,
103 const struct nlattr **actionsp,
104 size_t *actions_lenp);
106 /* A subfacet (see "struct subfacet" below) has three possible installation
109 * - SF_NOT_INSTALLED: Not installed in the datapath. This will only be the
110 * case just after the subfacet is created, just before the subfacet is
111 * destroyed, or if the datapath returns an error when we try to install a
114 * - SF_FAST_PATH: The subfacet's actions are installed in the datapath.
116 * - SF_SLOW_PATH: An action that sends every packet for the subfacet through
117 * ofproto_dpif is installed in the datapath.
120 SF_NOT_INSTALLED, /* No datapath flow for this subfacet. */
121 SF_FAST_PATH, /* Full actions are installed. */
122 SF_SLOW_PATH, /* Send-to-userspace action is installed. */
125 /* A dpif flow and actions associated with a facet.
127 * See also the large comment on struct facet. */
130 struct hmap_node hmap_node; /* In struct ofproto_dpif 'subfacets' list. */
131 struct list list_node; /* In struct facet's 'facets' list. */
132 struct facet *facet; /* Owning facet. */
133 struct dpif_backer *backer; /* Owning backer. */
135 enum odp_key_fitness key_fitness;
139 long long int used; /* Time last used; time created if not used. */
140 long long int created; /* Time created. */
142 uint64_t dp_packet_count; /* Last known packet count in the datapath. */
143 uint64_t dp_byte_count; /* Last known byte count in the datapath. */
145 enum subfacet_path path; /* Installed in datapath? */
148 #define SUBFACET_DESTROY_MAX_BATCH 50
150 static struct subfacet *subfacet_create(struct facet *, struct flow_miss *miss,
152 static struct subfacet *subfacet_find(struct dpif_backer *,
153 const struct nlattr *key, size_t key_len,
155 static void subfacet_destroy(struct subfacet *);
156 static void subfacet_destroy__(struct subfacet *);
157 static void subfacet_destroy_batch(struct dpif_backer *,
158 struct subfacet **, int n);
159 static void subfacet_reset_dp_stats(struct subfacet *,
160 struct dpif_flow_stats *);
161 static void subfacet_update_stats(struct subfacet *,
162 const struct dpif_flow_stats *);
163 static int subfacet_install(struct subfacet *,
164 const struct ofpbuf *odp_actions,
165 struct dpif_flow_stats *);
166 static void subfacet_uninstall(struct subfacet *);
168 /* A unique, non-overlapping instantiation of an OpenFlow flow.
170 * A facet associates a "struct flow", which represents the Open vSwitch
171 * userspace idea of an exact-match flow, with one or more subfacets.
172 * While the facet is created based on an exact-match flow, it is stored
173 * within the ofproto based on the wildcards that could be expressed
174 * based on the flow table and other configuration. (See the 'wc'
175 * description in "struct xlate_out" for more details.)
177 * Each subfacet tracks the datapath's idea of the flow equivalent to
178 * the facet. When the kernel module (or other dpif implementation) and
179 * Open vSwitch userspace agree on the definition of a flow key, there
180 * is exactly one subfacet per facet. If the dpif implementation
181 * supports more-specific flow matching than userspace, however, a facet
182 * can have more than one subfacet. Examples include the dpif
183 * implementation not supporting the same wildcards as userspace or some
184 * distinction in flow that userspace simply doesn't understand.
186 * Flow expiration works in terms of subfacets, so a facet must have at
187 * least one subfacet or it will never expire, leaking memory. */
190 struct hmap_node hmap_node; /* In owning ofproto's 'facets' hmap. */
191 struct list list_node; /* In owning rule's 'facets' list. */
192 struct rule_dpif *rule; /* Owning rule. */
195 struct list subfacets;
196 long long int used; /* Time last used; time created if not used. */
199 struct flow flow; /* Flow of the creating subfacet. */
200 struct cls_rule cr; /* In 'ofproto_dpif's facets classifier. */
204 * - Do include packets and bytes sent "by hand", e.g. with
207 * - Do include packets and bytes that were obtained from the datapath
208 * when a subfacet's statistics were reset (e.g. dpif_flow_put() with
209 * DPIF_FP_ZERO_STATS).
211 * - Do not include packets or bytes that can be obtained from the
212 * datapath for any existing subfacet.
214 uint64_t packet_count; /* Number of packets received. */
215 uint64_t byte_count; /* Number of bytes received. */
217 /* Resubmit statistics. */
218 uint64_t prev_packet_count; /* Number of packets from last stats push. */
219 uint64_t prev_byte_count; /* Number of bytes from last stats push. */
220 long long int prev_used; /* Used time from last stats push. */
223 uint64_t accounted_bytes; /* Bytes processed by facet_account(). */
224 struct netflow_flow nf_flow; /* Per-flow NetFlow tracking data. */
225 uint8_t tcp_flags; /* TCP flags seen for this 'rule'. */
227 struct xlate_out xout;
229 /* Storage for a single subfacet, to reduce malloc() time and space
230 * overhead. (A facet always has at least one subfacet and in the common
231 * case has exactly one subfacet. However, 'one_subfacet' may not
232 * always be valid, since it could have been removed after newer
233 * subfacets were pushed onto the 'subfacets' list.) */
234 struct subfacet one_subfacet;
236 long long int learn_rl; /* Rate limiter for facet_learn(). */
239 static struct facet *facet_create(const struct flow_miss *, struct rule_dpif *,
241 struct dpif_flow_stats *);
242 static void facet_remove(struct facet *);
243 static void facet_free(struct facet *);
245 static struct facet *facet_find(struct ofproto_dpif *, const struct flow *);
246 static struct facet *facet_lookup_valid(struct ofproto_dpif *,
247 const struct flow *);
248 static bool facet_revalidate(struct facet *);
249 static bool facet_check_consistency(struct facet *);
251 static void facet_flush_stats(struct facet *);
253 static void facet_reset_counters(struct facet *);
254 static void facet_push_stats(struct facet *, bool may_learn);
255 static void facet_learn(struct facet *);
256 static void facet_account(struct facet *);
257 static void push_all_stats(void);
259 static bool facet_is_controller_flow(struct facet *);
261 /* Node in 'ofport_dpif''s 'priorities' map. Used to maintain a map from
262 * 'priority' (the datapath's term for QoS queue) to the dscp bits which all
263 * traffic egressing the 'ofport' with that priority should be marked with. */
264 struct priority_to_dscp {
265 struct hmap_node hmap_node; /* Node in 'ofport_dpif''s 'priorities' map. */
266 uint32_t priority; /* Priority of this queue (see struct flow). */
268 uint8_t dscp; /* DSCP bits to mark outgoing traffic with. */
271 /* Linux VLAN device support (e.g. "eth0.10" for VLAN 10.)
273 * This is deprecated. It is only for compatibility with broken device drivers
274 * in old versions of Linux that do not properly support VLANs when VLAN
275 * devices are not used. When broken device drivers are no longer in
276 * widespread use, we will delete these interfaces. */
277 struct vlan_splinter {
278 struct hmap_node realdev_vid_node;
279 struct hmap_node vlandev_node;
280 uint16_t realdev_ofp_port;
281 uint16_t vlandev_ofp_port;
285 static bool vsp_adjust_flow(const struct ofproto_dpif *, struct flow *);
286 static void vsp_remove(struct ofport_dpif *);
287 static void vsp_add(struct ofport_dpif *, uint16_t realdev_ofp_port, int vid);
289 static uint16_t odp_port_to_ofp_port(const struct ofproto_dpif *,
292 static struct ofport_dpif *
293 ofport_dpif_cast(const struct ofport *ofport)
295 return ofport ? CONTAINER_OF(ofport, struct ofport_dpif, up) : NULL;
298 static void port_run(struct ofport_dpif *);
299 static void port_run_fast(struct ofport_dpif *);
300 static void port_wait(struct ofport_dpif *);
301 static int set_bfd(struct ofport *, const struct smap *);
302 static int set_cfm(struct ofport *, const struct cfm_settings *);
303 static void ofport_clear_priorities(struct ofport_dpif *);
304 static void run_fast_rl(void);
306 struct dpif_completion {
307 struct list list_node;
308 struct ofoperation *op;
311 /* Reasons that we might need to revalidate every facet, and corresponding
314 * A value of 0 means that there is no need to revalidate.
316 * It would be nice to have some cleaner way to integrate with coverage
317 * counters, but with only a few reasons I guess this is good enough for
319 enum revalidate_reason {
320 REV_RECONFIGURE = 1, /* Switch configuration changed. */
321 REV_STP, /* Spanning tree protocol port status change. */
322 REV_PORT_TOGGLED, /* Port enabled or disabled by CFM, LACP, ...*/
323 REV_FLOW_TABLE, /* Flow table changed. */
324 REV_INCONSISTENCY /* Facet self-check failed. */
326 COVERAGE_DEFINE(rev_reconfigure);
327 COVERAGE_DEFINE(rev_stp);
328 COVERAGE_DEFINE(rev_port_toggled);
329 COVERAGE_DEFINE(rev_flow_table);
330 COVERAGE_DEFINE(rev_inconsistency);
332 /* Drop keys are odp flow keys which have drop flows installed in the kernel.
333 * These are datapath flows which have no associated ofproto, if they did we
334 * would use facets. */
336 struct hmap_node hmap_node;
341 struct avg_subfacet_rates {
342 double add_rate; /* Moving average of new flows created per minute. */
343 double del_rate; /* Moving average of flows deleted per minute. */
346 /* All datapaths of a given type share a single dpif backer instance. */
351 struct timer next_expiration;
352 struct hmap odp_to_ofport_map; /* ODP port to ofport mapping. */
354 struct simap tnl_backers; /* Set of dpif ports backing tunnels. */
356 /* Facet revalidation flags applying to facets which use this backer. */
357 enum revalidate_reason need_revalidate; /* Revalidate every facet. */
358 struct tag_set revalidate_set; /* Revalidate only matching facets. */
360 struct hmap drop_keys; /* Set of dropped odp keys. */
361 bool recv_set_enable; /* Enables or disables receiving packets. */
363 struct hmap subfacets;
364 struct governor *governor;
366 /* Subfacet statistics.
368 * These keep track of the total number of subfacets added and deleted and
369 * flow life span. They are useful for computing the flow rates stats
370 * exposed via "ovs-appctl dpif/show". The goal is to learn about
371 * traffic patterns in ways that we can use later to improve Open vSwitch
372 * performance in new situations. */
373 long long int created; /* Time when it is created. */
374 unsigned max_n_subfacet; /* Maximum number of flows */
375 unsigned avg_n_subfacet; /* Average number of flows. */
376 long long int avg_subfacet_life; /* Average life span of subfacets. */
378 /* The average number of subfacets... */
379 struct avg_subfacet_rates hourly; /* ...over the last hour. */
380 struct avg_subfacet_rates daily; /* ...over the last day. */
381 struct avg_subfacet_rates lifetime; /* ...over the switch lifetime. */
382 long long int last_minute; /* Last time 'hourly' was updated. */
384 /* Number of subfacets added or deleted since 'last_minute'. */
385 unsigned subfacet_add_count;
386 unsigned subfacet_del_count;
388 /* Number of subfacets added or deleted from 'created' to 'last_minute.' */
389 unsigned long long int total_subfacet_add_count;
390 unsigned long long int total_subfacet_del_count;
393 /* All existing ofproto_backer instances, indexed by ofproto->up.type. */
394 static struct shash all_dpif_backers = SHASH_INITIALIZER(&all_dpif_backers);
396 static void drop_key_clear(struct dpif_backer *);
397 static struct ofport_dpif *
398 odp_port_to_ofport(const struct dpif_backer *, uint32_t odp_port);
399 static void update_moving_averages(struct dpif_backer *backer);
401 /* Defer flow mod completion until "ovs-appctl ofproto/unclog"? (Useful only
402 * for debugging the asynchronous flow_mod implementation.) */
405 /* All existing ofproto_dpif instances, indexed by ->up.name. */
406 static struct hmap all_ofproto_dpifs = HMAP_INITIALIZER(&all_ofproto_dpifs);
408 static void ofproto_dpif_unixctl_init(void);
411 #define FLOW_MISS_MAX_BATCH 50
412 static int handle_upcalls(struct dpif_backer *, unsigned int max_batch);
414 /* Flow expiration. */
415 static int expire(struct dpif_backer *);
418 static void send_netflow_active_timeouts(struct ofproto_dpif *);
421 static int send_packet(const struct ofport_dpif *, struct ofpbuf *packet);
423 /* Global variables. */
424 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
426 /* Initial mappings of port to bridge mappings. */
427 static struct shash init_ofp_ports = SHASH_INITIALIZER(&init_ofp_ports);
429 /* Factory functions. */
432 init(const struct shash *iface_hints)
434 struct shash_node *node;
436 /* Make a local copy, since we don't own 'iface_hints' elements. */
437 SHASH_FOR_EACH(node, iface_hints) {
438 const struct iface_hint *orig_hint = node->data;
439 struct iface_hint *new_hint = xmalloc(sizeof *new_hint);
441 new_hint->br_name = xstrdup(orig_hint->br_name);
442 new_hint->br_type = xstrdup(orig_hint->br_type);
443 new_hint->ofp_port = orig_hint->ofp_port;
445 shash_add(&init_ofp_ports, node->name, new_hint);
450 enumerate_types(struct sset *types)
452 dp_enumerate_types(types);
456 enumerate_names(const char *type, struct sset *names)
458 struct ofproto_dpif *ofproto;
461 HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
462 if (strcmp(type, ofproto->up.type)) {
465 sset_add(names, ofproto->up.name);
472 del(const char *type, const char *name)
477 error = dpif_open(name, type, &dpif);
479 error = dpif_delete(dpif);
486 port_open_type(const char *datapath_type, const char *port_type)
488 return dpif_port_open_type(datapath_type, port_type);
491 /* Type functions. */
493 static struct ofproto_dpif *
494 lookup_ofproto_dpif_by_port_name(const char *name)
496 struct ofproto_dpif *ofproto;
498 HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
499 if (sset_contains(&ofproto->ports, name)) {
508 type_run(const char *type)
510 static long long int push_timer = LLONG_MIN;
511 struct dpif_backer *backer;
515 backer = shash_find_data(&all_dpif_backers, type);
517 /* This is not necessarily a problem, since backers are only
518 * created on demand. */
522 dpif_run(backer->dpif);
524 /* The most natural place to push facet statistics is when they're pulled
525 * from the datapath. However, when there are many flows in the datapath,
526 * this expensive operation can occur so frequently, that it reduces our
527 * ability to quickly set up flows. To reduce the cost, we push statistics
529 if (time_msec() > push_timer) {
530 push_timer = time_msec() + 2000;
534 /* If vswitchd started with other_config:flow_restore_wait set as "true",
535 * and the configuration has now changed to "false", enable receiving
536 * packets from the datapath. */
537 if (!backer->recv_set_enable && !ofproto_get_flow_restore_wait()) {
538 backer->recv_set_enable = true;
540 error = dpif_recv_set(backer->dpif, backer->recv_set_enable);
542 VLOG_ERR("Failed to enable receiving packets in dpif.");
545 dpif_flow_flush(backer->dpif);
546 backer->need_revalidate = REV_RECONFIGURE;
549 if (backer->need_revalidate
550 || !tag_set_is_empty(&backer->revalidate_set)) {
551 struct tag_set revalidate_set = backer->revalidate_set;
552 bool need_revalidate = backer->need_revalidate;
553 struct ofproto_dpif *ofproto;
554 struct simap_node *node;
555 struct simap tmp_backers;
557 /* Handle tunnel garbage collection. */
558 simap_init(&tmp_backers);
559 simap_swap(&backer->tnl_backers, &tmp_backers);
561 HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
562 struct ofport_dpif *iter;
564 if (backer != ofproto->backer) {
568 HMAP_FOR_EACH (iter, up.hmap_node, &ofproto->up.ports) {
569 char namebuf[NETDEV_VPORT_NAME_BUFSIZE];
572 if (!iter->tnl_port) {
576 dp_port = netdev_vport_get_dpif_port(iter->up.netdev,
577 namebuf, sizeof namebuf);
578 node = simap_find(&tmp_backers, dp_port);
580 simap_put(&backer->tnl_backers, dp_port, node->data);
581 simap_delete(&tmp_backers, node);
582 node = simap_find(&backer->tnl_backers, dp_port);
584 node = simap_find(&backer->tnl_backers, dp_port);
586 uint32_t odp_port = UINT32_MAX;
588 if (!dpif_port_add(backer->dpif, iter->up.netdev,
590 simap_put(&backer->tnl_backers, dp_port, odp_port);
591 node = simap_find(&backer->tnl_backers, dp_port);
596 iter->odp_port = node ? node->data : OVSP_NONE;
597 if (tnl_port_reconfigure(&iter->up, iter->odp_port,
599 backer->need_revalidate = REV_RECONFIGURE;
604 SIMAP_FOR_EACH (node, &tmp_backers) {
605 dpif_port_del(backer->dpif, node->data);
607 simap_destroy(&tmp_backers);
609 switch (backer->need_revalidate) {
610 case REV_RECONFIGURE: COVERAGE_INC(rev_reconfigure); break;
611 case REV_STP: COVERAGE_INC(rev_stp); break;
612 case REV_PORT_TOGGLED: COVERAGE_INC(rev_port_toggled); break;
613 case REV_FLOW_TABLE: COVERAGE_INC(rev_flow_table); break;
614 case REV_INCONSISTENCY: COVERAGE_INC(rev_inconsistency); break;
617 if (backer->need_revalidate) {
618 /* Clear the drop_keys in case we should now be accepting some
619 * formerly dropped flows. */
620 drop_key_clear(backer);
623 /* Clear the revalidation flags. */
624 tag_set_init(&backer->revalidate_set);
625 backer->need_revalidate = 0;
627 HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
628 struct facet *facet, *next;
629 struct cls_cursor cursor;
631 if (ofproto->backer != backer) {
635 cls_cursor_init(&cursor, &ofproto->facets, NULL);
636 CLS_CURSOR_FOR_EACH_SAFE (facet, next, cr, &cursor) {
638 || tag_set_intersects(&revalidate_set, facet->xout.tags)) {
639 facet_revalidate(facet);
646 if (!backer->recv_set_enable) {
647 /* Wake up before a max of 1000ms. */
648 timer_set_duration(&backer->next_expiration, 1000);
649 } else if (timer_expired(&backer->next_expiration)) {
650 int delay = expire(backer);
651 timer_set_duration(&backer->next_expiration, delay);
654 /* Check for port changes in the dpif. */
655 while ((error = dpif_port_poll(backer->dpif, &devname)) == 0) {
656 struct ofproto_dpif *ofproto;
657 struct dpif_port port;
659 /* Don't report on the datapath's device. */
660 if (!strcmp(devname, dpif_base_name(backer->dpif))) {
664 HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node,
665 &all_ofproto_dpifs) {
666 if (simap_contains(&ofproto->backer->tnl_backers, devname)) {
671 ofproto = lookup_ofproto_dpif_by_port_name(devname);
672 if (dpif_port_query_by_name(backer->dpif, devname, &port)) {
673 /* The port was removed. If we know the datapath,
674 * report it through poll_set(). If we don't, it may be
675 * notifying us of a removal we initiated, so ignore it.
676 * If there's a pending ENOBUFS, let it stand, since
677 * everything will be reevaluated. */
678 if (ofproto && ofproto->port_poll_errno != ENOBUFS) {
679 sset_add(&ofproto->port_poll_set, devname);
680 ofproto->port_poll_errno = 0;
682 } else if (!ofproto) {
683 /* The port was added, but we don't know with which
684 * ofproto we should associate it. Delete it. */
685 dpif_port_del(backer->dpif, port.port_no);
687 dpif_port_destroy(&port);
693 if (error != EAGAIN) {
694 struct ofproto_dpif *ofproto;
696 /* There was some sort of error, so propagate it to all
697 * ofprotos that use this backer. */
698 HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node,
699 &all_ofproto_dpifs) {
700 if (ofproto->backer == backer) {
701 sset_clear(&ofproto->port_poll_set);
702 ofproto->port_poll_errno = error;
707 if (backer->governor) {
710 governor_run(backer->governor);
712 /* If the governor has shrunk to its minimum size and the number of
713 * subfacets has dwindled, then drop the governor entirely.
715 * For hysteresis, the number of subfacets to drop the governor is
716 * smaller than the number needed to trigger its creation. */
717 n_subfacets = hmap_count(&backer->subfacets);
718 if (n_subfacets * 4 < flow_eviction_threshold
719 && governor_is_idle(backer->governor)) {
720 governor_destroy(backer->governor);
721 backer->governor = NULL;
729 dpif_backer_run_fast(struct dpif_backer *backer, int max_batch)
733 /* If recv_set_enable is false, we should not handle upcalls. */
734 if (!backer->recv_set_enable) {
738 /* Handle one or more batches of upcalls, until there's nothing left to do
739 * or until we do a fixed total amount of work.
741 * We do work in batches because it can be much cheaper to set up a number
742 * of flows and fire off their patches all at once. We do multiple batches
743 * because in some cases handling a packet can cause another packet to be
744 * queued almost immediately as part of the return flow. Both
745 * optimizations can make major improvements on some benchmarks and
746 * presumably for real traffic as well. */
748 while (work < max_batch) {
749 int retval = handle_upcalls(backer, max_batch - work);
760 type_run_fast(const char *type)
762 struct dpif_backer *backer;
764 backer = shash_find_data(&all_dpif_backers, type);
766 /* This is not necessarily a problem, since backers are only
767 * created on demand. */
771 return dpif_backer_run_fast(backer, FLOW_MISS_MAX_BATCH);
777 static long long int port_rl = LLONG_MIN;
778 static unsigned int backer_rl = 0;
780 if (time_msec() >= port_rl) {
781 struct ofproto_dpif *ofproto;
782 struct ofport_dpif *ofport;
784 HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
786 HMAP_FOR_EACH (ofport, up.hmap_node, &ofproto->up.ports) {
787 port_run_fast(ofport);
790 port_rl = time_msec() + 200;
793 /* XXX: We have to be careful not to do too much work in this function. If
794 * we call dpif_backer_run_fast() too often, or with too large a batch,
795 * performance improves signifcantly, but at a cost. It's possible for the
796 * number of flows in the datapath to increase without bound, and for poll
797 * loops to take 10s of seconds. The correct solution to this problem,
798 * long term, is to separate flow miss handling into it's own thread so it
799 * isn't affected by revalidations, and expirations. Until then, this is
800 * the best we can do. */
801 if (++backer_rl >= 10) {
802 struct shash_node *node;
805 SHASH_FOR_EACH (node, &all_dpif_backers) {
806 dpif_backer_run_fast(node->data, 1);
812 type_wait(const char *type)
814 struct dpif_backer *backer;
816 backer = shash_find_data(&all_dpif_backers, type);
818 /* This is not necessarily a problem, since backers are only
819 * created on demand. */
823 if (backer->governor) {
824 governor_wait(backer->governor);
827 timer_wait(&backer->next_expiration);
830 /* Basic life-cycle. */
832 static int add_internal_flows(struct ofproto_dpif *);
834 static struct ofproto *
837 struct ofproto_dpif *ofproto = xmalloc(sizeof *ofproto);
842 dealloc(struct ofproto *ofproto_)
844 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
849 close_dpif_backer(struct dpif_backer *backer)
851 struct shash_node *node;
853 ovs_assert(backer->refcount > 0);
855 if (--backer->refcount) {
859 drop_key_clear(backer);
860 hmap_destroy(&backer->drop_keys);
862 simap_destroy(&backer->tnl_backers);
863 hmap_destroy(&backer->odp_to_ofport_map);
864 node = shash_find(&all_dpif_backers, backer->type);
866 shash_delete(&all_dpif_backers, node);
867 dpif_close(backer->dpif);
869 ovs_assert(hmap_is_empty(&backer->subfacets));
870 hmap_destroy(&backer->subfacets);
871 governor_destroy(backer->governor);
876 /* Datapath port slated for removal from datapath. */
878 struct list list_node;
883 open_dpif_backer(const char *type, struct dpif_backer **backerp)
885 struct dpif_backer *backer;
886 struct dpif_port_dump port_dump;
887 struct dpif_port port;
888 struct shash_node *node;
889 struct list garbage_list;
890 struct odp_garbage *garbage, *next;
896 backer = shash_find_data(&all_dpif_backers, type);
903 backer_name = xasprintf("ovs-%s", type);
905 /* Remove any existing datapaths, since we assume we're the only
906 * userspace controlling the datapath. */
908 dp_enumerate_names(type, &names);
909 SSET_FOR_EACH(name, &names) {
910 struct dpif *old_dpif;
912 /* Don't remove our backer if it exists. */
913 if (!strcmp(name, backer_name)) {
917 if (dpif_open(name, type, &old_dpif)) {
918 VLOG_WARN("couldn't open old datapath %s to remove it", name);
920 dpif_delete(old_dpif);
921 dpif_close(old_dpif);
924 sset_destroy(&names);
926 backer = xmalloc(sizeof *backer);
928 error = dpif_create_and_open(backer_name, type, &backer->dpif);
931 VLOG_ERR("failed to open datapath of type %s: %s", type,
937 backer->type = xstrdup(type);
938 backer->governor = NULL;
939 backer->refcount = 1;
940 hmap_init(&backer->odp_to_ofport_map);
941 hmap_init(&backer->drop_keys);
942 hmap_init(&backer->subfacets);
943 timer_set_duration(&backer->next_expiration, 1000);
944 backer->need_revalidate = 0;
945 simap_init(&backer->tnl_backers);
946 tag_set_init(&backer->revalidate_set);
947 backer->recv_set_enable = !ofproto_get_flow_restore_wait();
950 if (backer->recv_set_enable) {
951 dpif_flow_flush(backer->dpif);
954 /* Loop through the ports already on the datapath and remove any
955 * that we don't need anymore. */
956 list_init(&garbage_list);
957 dpif_port_dump_start(&port_dump, backer->dpif);
958 while (dpif_port_dump_next(&port_dump, &port)) {
959 node = shash_find(&init_ofp_ports, port.name);
960 if (!node && strcmp(port.name, dpif_base_name(backer->dpif))) {
961 garbage = xmalloc(sizeof *garbage);
962 garbage->odp_port = port.port_no;
963 list_push_front(&garbage_list, &garbage->list_node);
966 dpif_port_dump_done(&port_dump);
968 LIST_FOR_EACH_SAFE (garbage, next, list_node, &garbage_list) {
969 dpif_port_del(backer->dpif, garbage->odp_port);
970 list_remove(&garbage->list_node);
974 shash_add(&all_dpif_backers, type, backer);
976 error = dpif_recv_set(backer->dpif, backer->recv_set_enable);
978 VLOG_ERR("failed to listen on datapath of type %s: %s",
979 type, strerror(error));
980 close_dpif_backer(backer);
984 backer->max_n_subfacet = 0;
985 backer->created = time_msec();
986 backer->last_minute = backer->created;
987 memset(&backer->hourly, 0, sizeof backer->hourly);
988 memset(&backer->daily, 0, sizeof backer->daily);
989 memset(&backer->lifetime, 0, sizeof backer->lifetime);
990 backer->subfacet_add_count = 0;
991 backer->subfacet_del_count = 0;
992 backer->total_subfacet_add_count = 0;
993 backer->total_subfacet_del_count = 0;
994 backer->avg_n_subfacet = 0;
995 backer->avg_subfacet_life = 0;
1001 construct(struct ofproto *ofproto_)
1003 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1004 struct shash_node *node, *next;
1009 error = open_dpif_backer(ofproto->up.type, &ofproto->backer);
1014 max_ports = dpif_get_max_ports(ofproto->backer->dpif);
1015 ofproto_init_max_ports(ofproto_, MIN(max_ports, OFPP_MAX));
1017 ofproto->netflow = NULL;
1018 ofproto->sflow = NULL;
1019 ofproto->ipfix = NULL;
1020 ofproto->stp = NULL;
1021 hmap_init(&ofproto->bundles);
1022 ofproto->ml = mac_learning_create(MAC_ENTRY_DEFAULT_IDLE_TIME);
1023 for (i = 0; i < MAX_MIRRORS; i++) {
1024 ofproto->mirrors[i] = NULL;
1026 ofproto->has_bonded_bundles = false;
1028 classifier_init(&ofproto->facets);
1029 ofproto->consistency_rl = LLONG_MIN;
1031 for (i = 0; i < N_TABLES; i++) {
1032 struct table_dpif *table = &ofproto->tables[i];
1034 table->catchall_table = NULL;
1035 table->other_table = NULL;
1036 table->basis = random_uint32();
1039 list_init(&ofproto->completions);
1041 ofproto_dpif_unixctl_init();
1043 ofproto->has_mirrors = false;
1044 ofproto->has_bundle_action = false;
1046 hmap_init(&ofproto->vlandev_map);
1047 hmap_init(&ofproto->realdev_vid_map);
1049 sset_init(&ofproto->ports);
1050 sset_init(&ofproto->ghost_ports);
1051 sset_init(&ofproto->port_poll_set);
1052 ofproto->port_poll_errno = 0;
1054 SHASH_FOR_EACH_SAFE (node, next, &init_ofp_ports) {
1055 struct iface_hint *iface_hint = node->data;
1057 if (!strcmp(iface_hint->br_name, ofproto->up.name)) {
1058 /* Check if the datapath already has this port. */
1059 if (dpif_port_exists(ofproto->backer->dpif, node->name)) {
1060 sset_add(&ofproto->ports, node->name);
1063 free(iface_hint->br_name);
1064 free(iface_hint->br_type);
1066 shash_delete(&init_ofp_ports, node);
1070 hmap_insert(&all_ofproto_dpifs, &ofproto->all_ofproto_dpifs_node,
1071 hash_string(ofproto->up.name, 0));
1072 memset(&ofproto->stats, 0, sizeof ofproto->stats);
1074 ofproto_init_tables(ofproto_, N_TABLES);
1075 error = add_internal_flows(ofproto);
1076 ofproto->up.tables[TBL_INTERNAL].flags = OFTABLE_HIDDEN | OFTABLE_READONLY;
1079 ofproto->n_missed = 0;
1085 add_internal_flow(struct ofproto_dpif *ofproto, int id,
1086 const struct ofpbuf *ofpacts, struct rule_dpif **rulep)
1088 struct ofputil_flow_mod fm;
1091 match_init_catchall(&fm.match);
1093 match_set_reg(&fm.match, 0, id);
1094 fm.new_cookie = htonll(0);
1095 fm.cookie = htonll(0);
1096 fm.cookie_mask = htonll(0);
1097 fm.table_id = TBL_INTERNAL;
1098 fm.command = OFPFC_ADD;
1099 fm.idle_timeout = 0;
1100 fm.hard_timeout = 0;
1104 fm.ofpacts = ofpacts->data;
1105 fm.ofpacts_len = ofpacts->size;
1107 error = ofproto_flow_mod(&ofproto->up, &fm);
1109 VLOG_ERR_RL(&rl, "failed to add internal flow %d (%s)",
1110 id, ofperr_to_string(error));
1114 *rulep = rule_dpif_lookup_in_table(ofproto, &fm.match.flow, NULL,
1116 ovs_assert(*rulep != NULL);
1122 add_internal_flows(struct ofproto_dpif *ofproto)
1124 struct ofpact_controller *controller;
1125 uint64_t ofpacts_stub[128 / 8];
1126 struct ofpbuf ofpacts;
1130 ofpbuf_use_stack(&ofpacts, ofpacts_stub, sizeof ofpacts_stub);
1133 controller = ofpact_put_CONTROLLER(&ofpacts);
1134 controller->max_len = UINT16_MAX;
1135 controller->controller_id = 0;
1136 controller->reason = OFPR_NO_MATCH;
1137 ofpact_pad(&ofpacts);
1139 error = add_internal_flow(ofproto, id++, &ofpacts, &ofproto->miss_rule);
1144 ofpbuf_clear(&ofpacts);
1145 error = add_internal_flow(ofproto, id++, &ofpacts,
1146 &ofproto->no_packet_in_rule);
1151 error = add_internal_flow(ofproto, id++, &ofpacts,
1152 &ofproto->drop_frags_rule);
1157 complete_operations(struct ofproto_dpif *ofproto)
1159 struct dpif_completion *c, *next;
1161 LIST_FOR_EACH_SAFE (c, next, list_node, &ofproto->completions) {
1162 ofoperation_complete(c->op, 0);
1163 list_remove(&c->list_node);
1169 destruct(struct ofproto *ofproto_)
1171 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1172 struct rule_dpif *rule, *next_rule;
1173 struct oftable *table;
1176 hmap_remove(&all_ofproto_dpifs, &ofproto->all_ofproto_dpifs_node);
1177 complete_operations(ofproto);
1179 OFPROTO_FOR_EACH_TABLE (table, &ofproto->up) {
1180 struct cls_cursor cursor;
1182 cls_cursor_init(&cursor, &table->cls, NULL);
1183 CLS_CURSOR_FOR_EACH_SAFE (rule, next_rule, up.cr, &cursor) {
1184 ofproto_rule_destroy(&rule->up);
1188 for (i = 0; i < MAX_MIRRORS; i++) {
1189 mirror_destroy(ofproto->mirrors[i]);
1192 netflow_destroy(ofproto->netflow);
1193 dpif_sflow_destroy(ofproto->sflow);
1194 hmap_destroy(&ofproto->bundles);
1195 mac_learning_destroy(ofproto->ml);
1197 classifier_destroy(&ofproto->facets);
1199 hmap_destroy(&ofproto->vlandev_map);
1200 hmap_destroy(&ofproto->realdev_vid_map);
1202 sset_destroy(&ofproto->ports);
1203 sset_destroy(&ofproto->ghost_ports);
1204 sset_destroy(&ofproto->port_poll_set);
1206 close_dpif_backer(ofproto->backer);
1210 run_fast(struct ofproto *ofproto_)
1212 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1213 struct ofport_dpif *ofport;
1215 /* Do not perform any periodic activity required by 'ofproto' while
1216 * waiting for flow restore to complete. */
1217 if (ofproto_get_flow_restore_wait()) {
1221 HMAP_FOR_EACH (ofport, up.hmap_node, &ofproto->up.ports) {
1222 port_run_fast(ofport);
1229 run(struct ofproto *ofproto_)
1231 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1232 struct ofport_dpif *ofport;
1233 struct ofbundle *bundle;
1237 complete_operations(ofproto);
1240 /* Do not perform any periodic activity below required by 'ofproto' while
1241 * waiting for flow restore to complete. */
1242 if (ofproto_get_flow_restore_wait()) {
1246 error = run_fast(ofproto_);
1251 if (ofproto->netflow) {
1252 if (netflow_run(ofproto->netflow)) {
1253 send_netflow_active_timeouts(ofproto);
1256 if (ofproto->sflow) {
1257 dpif_sflow_run(ofproto->sflow);
1260 HMAP_FOR_EACH (ofport, up.hmap_node, &ofproto->up.ports) {
1263 HMAP_FOR_EACH (bundle, hmap_node, &ofproto->bundles) {
1268 mac_learning_run(ofproto->ml, &ofproto->backer->revalidate_set);
1270 /* Check the consistency of a random facet, to aid debugging. */
1271 if (time_msec() >= ofproto->consistency_rl
1272 && !classifier_is_empty(&ofproto->facets)
1273 && !ofproto->backer->need_revalidate) {
1274 struct cls_table *table;
1275 struct cls_rule *cr;
1276 struct facet *facet;
1278 ofproto->consistency_rl = time_msec() + 250;
1280 table = CONTAINER_OF(hmap_random_node(&ofproto->facets.tables),
1281 struct cls_table, hmap_node);
1282 cr = CONTAINER_OF(hmap_random_node(&table->rules), struct cls_rule,
1284 facet = CONTAINER_OF(cr, struct facet, cr);
1286 if (!tag_set_intersects(&ofproto->backer->revalidate_set,
1287 facet->xout.tags)) {
1288 if (!facet_check_consistency(facet)) {
1289 ofproto->backer->need_revalidate = REV_INCONSISTENCY;
1298 wait(struct ofproto *ofproto_)
1300 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1301 struct ofport_dpif *ofport;
1302 struct ofbundle *bundle;
1304 if (!clogged && !list_is_empty(&ofproto->completions)) {
1305 poll_immediate_wake();
1308 if (ofproto_get_flow_restore_wait()) {
1312 dpif_wait(ofproto->backer->dpif);
1313 dpif_recv_wait(ofproto->backer->dpif);
1314 if (ofproto->sflow) {
1315 dpif_sflow_wait(ofproto->sflow);
1317 if (!tag_set_is_empty(&ofproto->backer->revalidate_set)) {
1318 poll_immediate_wake();
1320 HMAP_FOR_EACH (ofport, up.hmap_node, &ofproto->up.ports) {
1323 HMAP_FOR_EACH (bundle, hmap_node, &ofproto->bundles) {
1324 bundle_wait(bundle);
1326 if (ofproto->netflow) {
1327 netflow_wait(ofproto->netflow);
1329 mac_learning_wait(ofproto->ml);
1331 if (ofproto->backer->need_revalidate) {
1332 /* Shouldn't happen, but if it does just go around again. */
1333 VLOG_DBG_RL(&rl, "need revalidate in ofproto_wait_cb()");
1334 poll_immediate_wake();
1339 get_memory_usage(const struct ofproto *ofproto_, struct simap *usage)
1341 const struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1342 struct cls_cursor cursor;
1343 size_t n_subfacets = 0;
1344 struct facet *facet;
1346 simap_increase(usage, "facets", classifier_count(&ofproto->facets));
1348 cls_cursor_init(&cursor, &ofproto->facets, NULL);
1349 CLS_CURSOR_FOR_EACH (facet, cr, &cursor) {
1350 n_subfacets += list_size(&facet->subfacets);
1352 simap_increase(usage, "subfacets", n_subfacets);
1356 flush(struct ofproto *ofproto_)
1358 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1359 struct subfacet *subfacet, *next_subfacet;
1360 struct subfacet *batch[SUBFACET_DESTROY_MAX_BATCH];
1364 HMAP_FOR_EACH_SAFE (subfacet, next_subfacet, hmap_node,
1365 &ofproto->backer->subfacets) {
1366 if (ofproto_dpif_cast(subfacet->facet->rule->up.ofproto) != ofproto) {
1370 if (subfacet->path != SF_NOT_INSTALLED) {
1371 batch[n_batch++] = subfacet;
1372 if (n_batch >= SUBFACET_DESTROY_MAX_BATCH) {
1373 subfacet_destroy_batch(ofproto->backer, batch, n_batch);
1377 subfacet_destroy(subfacet);
1382 subfacet_destroy_batch(ofproto->backer, batch, n_batch);
1387 get_features(struct ofproto *ofproto_ OVS_UNUSED,
1388 bool *arp_match_ip, enum ofputil_action_bitmap *actions)
1390 *arp_match_ip = true;
1391 *actions = (OFPUTIL_A_OUTPUT |
1392 OFPUTIL_A_SET_VLAN_VID |
1393 OFPUTIL_A_SET_VLAN_PCP |
1394 OFPUTIL_A_STRIP_VLAN |
1395 OFPUTIL_A_SET_DL_SRC |
1396 OFPUTIL_A_SET_DL_DST |
1397 OFPUTIL_A_SET_NW_SRC |
1398 OFPUTIL_A_SET_NW_DST |
1399 OFPUTIL_A_SET_NW_TOS |
1400 OFPUTIL_A_SET_TP_SRC |
1401 OFPUTIL_A_SET_TP_DST |
1406 get_tables(struct ofproto *ofproto_, struct ofp12_table_stats *ots)
1408 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1409 struct dpif_dp_stats s;
1410 uint64_t n_miss, n_no_pkt_in, n_bytes, n_dropped_frags;
1413 strcpy(ots->name, "classifier");
1415 dpif_get_dp_stats(ofproto->backer->dpif, &s);
1416 rule_get_stats(&ofproto->miss_rule->up, &n_miss, &n_bytes);
1417 rule_get_stats(&ofproto->no_packet_in_rule->up, &n_no_pkt_in, &n_bytes);
1418 rule_get_stats(&ofproto->drop_frags_rule->up, &n_dropped_frags, &n_bytes);
1420 n_lookup = s.n_hit + s.n_missed - n_dropped_frags;
1421 ots->lookup_count = htonll(n_lookup);
1422 ots->matched_count = htonll(n_lookup - n_miss - n_no_pkt_in);
1425 static struct ofport *
1428 struct ofport_dpif *port = xmalloc(sizeof *port);
1433 port_dealloc(struct ofport *port_)
1435 struct ofport_dpif *port = ofport_dpif_cast(port_);
1440 port_construct(struct ofport *port_)
1442 struct ofport_dpif *port = ofport_dpif_cast(port_);
1443 struct ofproto_dpif *ofproto = ofproto_dpif_cast(port->up.ofproto);
1444 const struct netdev *netdev = port->up.netdev;
1445 char namebuf[NETDEV_VPORT_NAME_BUFSIZE];
1446 struct dpif_port dpif_port;
1449 ofproto->backer->need_revalidate = REV_RECONFIGURE;
1450 port->bundle = NULL;
1453 port->tag = tag_create_random();
1454 port->may_enable = true;
1455 port->stp_port = NULL;
1456 port->stp_state = STP_DISABLED;
1457 port->tnl_port = NULL;
1458 hmap_init(&port->priorities);
1459 port->realdev_ofp_port = 0;
1460 port->vlandev_vid = 0;
1461 port->carrier_seq = netdev_get_carrier_resets(netdev);
1463 if (netdev_vport_is_patch(netdev)) {
1464 /* By bailing out here, we don't submit the port to the sFlow module
1465 * to be considered for counter polling export. This is correct
1466 * because the patch port represents an interface that sFlow considers
1467 * to be "internal" to the switch as a whole, and therefore not an
1468 * candidate for counter polling. */
1469 port->odp_port = OVSP_NONE;
1473 error = dpif_port_query_by_name(ofproto->backer->dpif,
1474 netdev_vport_get_dpif_port(netdev, namebuf,
1481 port->odp_port = dpif_port.port_no;
1483 if (netdev_get_tunnel_config(netdev)) {
1484 port->tnl_port = tnl_port_add(&port->up, port->odp_port);
1486 /* Sanity-check that a mapping doesn't already exist. This
1487 * shouldn't happen for non-tunnel ports. */
1488 if (odp_port_to_ofp_port(ofproto, port->odp_port) != OFPP_NONE) {
1489 VLOG_ERR("port %s already has an OpenFlow port number",
1491 dpif_port_destroy(&dpif_port);
1495 hmap_insert(&ofproto->backer->odp_to_ofport_map, &port->odp_port_node,
1496 hash_int(port->odp_port, 0));
1498 dpif_port_destroy(&dpif_port);
1500 if (ofproto->sflow) {
1501 dpif_sflow_add_port(ofproto->sflow, port_, port->odp_port);
1508 port_destruct(struct ofport *port_)
1510 struct ofport_dpif *port = ofport_dpif_cast(port_);
1511 struct ofproto_dpif *ofproto = ofproto_dpif_cast(port->up.ofproto);
1512 const char *devname = netdev_get_name(port->up.netdev);
1513 char namebuf[NETDEV_VPORT_NAME_BUFSIZE];
1514 const char *dp_port_name;
1516 dp_port_name = netdev_vport_get_dpif_port(port->up.netdev, namebuf,
1518 if (dpif_port_exists(ofproto->backer->dpif, dp_port_name)) {
1519 /* The underlying device is still there, so delete it. This
1520 * happens when the ofproto is being destroyed, since the caller
1521 * assumes that removal of attached ports will happen as part of
1523 if (!port->tnl_port) {
1524 dpif_port_del(ofproto->backer->dpif, port->odp_port);
1526 ofproto->backer->need_revalidate = REV_RECONFIGURE;
1529 if (port->odp_port != OVSP_NONE && !port->tnl_port) {
1530 hmap_remove(&ofproto->backer->odp_to_ofport_map, &port->odp_port_node);
1533 tnl_port_del(port->tnl_port);
1534 sset_find_and_delete(&ofproto->ports, devname);
1535 sset_find_and_delete(&ofproto->ghost_ports, devname);
1536 ofproto->backer->need_revalidate = REV_RECONFIGURE;
1537 bundle_remove(port_);
1538 set_cfm(port_, NULL);
1539 set_bfd(port_, NULL);
1540 if (ofproto->sflow) {
1541 dpif_sflow_del_port(ofproto->sflow, port->odp_port);
1544 ofport_clear_priorities(port);
1545 hmap_destroy(&port->priorities);
1549 port_modified(struct ofport *port_)
1551 struct ofport_dpif *port = ofport_dpif_cast(port_);
1553 if (port->bundle && port->bundle->bond) {
1554 bond_slave_set_netdev(port->bundle->bond, port, port->up.netdev);
1558 cfm_set_netdev(port->cfm, port->up.netdev);
1563 port_reconfigured(struct ofport *port_, enum ofputil_port_config old_config)
1565 struct ofport_dpif *port = ofport_dpif_cast(port_);
1566 struct ofproto_dpif *ofproto = ofproto_dpif_cast(port->up.ofproto);
1567 enum ofputil_port_config changed = old_config ^ port->up.pp.config;
1569 if (changed & (OFPUTIL_PC_NO_RECV | OFPUTIL_PC_NO_RECV_STP |
1570 OFPUTIL_PC_NO_FWD | OFPUTIL_PC_NO_FLOOD |
1571 OFPUTIL_PC_NO_PACKET_IN)) {
1572 ofproto->backer->need_revalidate = REV_RECONFIGURE;
1574 if (changed & OFPUTIL_PC_NO_FLOOD && port->bundle) {
1575 bundle_update(port->bundle);
1581 set_sflow(struct ofproto *ofproto_,
1582 const struct ofproto_sflow_options *sflow_options)
1584 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1585 struct dpif_sflow *ds = ofproto->sflow;
1587 if (sflow_options) {
1589 struct ofport_dpif *ofport;
1591 ds = ofproto->sflow = dpif_sflow_create();
1592 HMAP_FOR_EACH (ofport, up.hmap_node, &ofproto->up.ports) {
1593 dpif_sflow_add_port(ds, &ofport->up, ofport->odp_port);
1595 ofproto->backer->need_revalidate = REV_RECONFIGURE;
1597 dpif_sflow_set_options(ds, sflow_options);
1600 dpif_sflow_destroy(ds);
1601 ofproto->backer->need_revalidate = REV_RECONFIGURE;
1602 ofproto->sflow = NULL;
1610 struct ofproto *ofproto_,
1611 const struct ofproto_ipfix_bridge_exporter_options *bridge_exporter_options,
1612 const struct ofproto_ipfix_flow_exporter_options *flow_exporters_options,
1613 size_t n_flow_exporters_options)
1615 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1616 struct dpif_ipfix *di = ofproto->ipfix;
1618 if (bridge_exporter_options || flow_exporters_options) {
1620 di = ofproto->ipfix = dpif_ipfix_create();
1622 dpif_ipfix_set_options(
1623 di, bridge_exporter_options, flow_exporters_options,
1624 n_flow_exporters_options);
1627 dpif_ipfix_destroy(di);
1628 ofproto->ipfix = NULL;
1635 set_cfm(struct ofport *ofport_, const struct cfm_settings *s)
1637 struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
1644 struct ofproto_dpif *ofproto;
1646 ofproto = ofproto_dpif_cast(ofport->up.ofproto);
1647 ofproto->backer->need_revalidate = REV_RECONFIGURE;
1648 ofport->cfm = cfm_create(ofport->up.netdev);
1651 if (cfm_configure(ofport->cfm, s)) {
1657 cfm_destroy(ofport->cfm);
1663 get_cfm_status(const struct ofport *ofport_,
1664 struct ofproto_cfm_status *status)
1666 struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
1669 status->faults = cfm_get_fault(ofport->cfm);
1670 status->remote_opstate = cfm_get_opup(ofport->cfm);
1671 status->health = cfm_get_health(ofport->cfm);
1672 cfm_get_remote_mpids(ofport->cfm, &status->rmps, &status->n_rmps);
1680 set_bfd(struct ofport *ofport_, const struct smap *cfg)
1682 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport_->ofproto);
1683 struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
1687 ofport->bfd = bfd_configure(old, netdev_get_name(ofport->up.netdev), cfg);
1688 if (ofport->bfd != old) {
1689 ofproto->backer->need_revalidate = REV_RECONFIGURE;
1696 get_bfd_status(struct ofport *ofport_, struct smap *smap)
1698 struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
1701 bfd_get_status(ofport->bfd, smap);
1708 /* Spanning Tree. */
1711 send_bpdu_cb(struct ofpbuf *pkt, int port_num, void *ofproto_)
1713 struct ofproto_dpif *ofproto = ofproto_;
1714 struct stp_port *sp = stp_get_port(ofproto->stp, port_num);
1715 struct ofport_dpif *ofport;
1717 ofport = stp_port_get_aux(sp);
1719 VLOG_WARN_RL(&rl, "%s: cannot send BPDU on unknown port %d",
1720 ofproto->up.name, port_num);
1722 struct eth_header *eth = pkt->l2;
1724 netdev_get_etheraddr(ofport->up.netdev, eth->eth_src);
1725 if (eth_addr_is_zero(eth->eth_src)) {
1726 VLOG_WARN_RL(&rl, "%s: cannot send BPDU on port %d "
1727 "with unknown MAC", ofproto->up.name, port_num);
1729 send_packet(ofport, pkt);
1735 /* Configures STP on 'ofproto_' using the settings defined in 's'. */
1737 set_stp(struct ofproto *ofproto_, const struct ofproto_stp_settings *s)
1739 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1741 /* Only revalidate flows if the configuration changed. */
1742 if (!s != !ofproto->stp) {
1743 ofproto->backer->need_revalidate = REV_RECONFIGURE;
1747 if (!ofproto->stp) {
1748 ofproto->stp = stp_create(ofproto_->name, s->system_id,
1749 send_bpdu_cb, ofproto);
1750 ofproto->stp_last_tick = time_msec();
1753 stp_set_bridge_id(ofproto->stp, s->system_id);
1754 stp_set_bridge_priority(ofproto->stp, s->priority);
1755 stp_set_hello_time(ofproto->stp, s->hello_time);
1756 stp_set_max_age(ofproto->stp, s->max_age);
1757 stp_set_forward_delay(ofproto->stp, s->fwd_delay);
1759 struct ofport *ofport;
1761 HMAP_FOR_EACH (ofport, hmap_node, &ofproto->up.ports) {
1762 set_stp_port(ofport, NULL);
1765 stp_destroy(ofproto->stp);
1766 ofproto->stp = NULL;
1773 get_stp_status(struct ofproto *ofproto_, struct ofproto_stp_status *s)
1775 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1779 s->bridge_id = stp_get_bridge_id(ofproto->stp);
1780 s->designated_root = stp_get_designated_root(ofproto->stp);
1781 s->root_path_cost = stp_get_root_path_cost(ofproto->stp);
1790 update_stp_port_state(struct ofport_dpif *ofport)
1792 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport->up.ofproto);
1793 enum stp_state state;
1795 /* Figure out new state. */
1796 state = ofport->stp_port ? stp_port_get_state(ofport->stp_port)
1800 if (ofport->stp_state != state) {
1801 enum ofputil_port_state of_state;
1804 VLOG_DBG_RL(&rl, "port %s: STP state changed from %s to %s",
1805 netdev_get_name(ofport->up.netdev),
1806 stp_state_name(ofport->stp_state),
1807 stp_state_name(state));
1808 if (stp_learn_in_state(ofport->stp_state)
1809 != stp_learn_in_state(state)) {
1810 /* xxx Learning action flows should also be flushed. */
1811 mac_learning_flush(ofproto->ml,
1812 &ofproto->backer->revalidate_set);
1814 fwd_change = stp_forward_in_state(ofport->stp_state)
1815 != stp_forward_in_state(state);
1817 ofproto->backer->need_revalidate = REV_STP;
1818 ofport->stp_state = state;
1819 ofport->stp_state_entered = time_msec();
1821 if (fwd_change && ofport->bundle) {
1822 bundle_update(ofport->bundle);
1825 /* Update the STP state bits in the OpenFlow port description. */
1826 of_state = ofport->up.pp.state & ~OFPUTIL_PS_STP_MASK;
1827 of_state |= (state == STP_LISTENING ? OFPUTIL_PS_STP_LISTEN
1828 : state == STP_LEARNING ? OFPUTIL_PS_STP_LEARN
1829 : state == STP_FORWARDING ? OFPUTIL_PS_STP_FORWARD
1830 : state == STP_BLOCKING ? OFPUTIL_PS_STP_BLOCK
1832 ofproto_port_set_state(&ofport->up, of_state);
1836 /* Configures STP on 'ofport_' using the settings defined in 's'. The
1837 * caller is responsible for assigning STP port numbers and ensuring
1838 * there are no duplicates. */
1840 set_stp_port(struct ofport *ofport_,
1841 const struct ofproto_port_stp_settings *s)
1843 struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
1844 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport->up.ofproto);
1845 struct stp_port *sp = ofport->stp_port;
1847 if (!s || !s->enable) {
1849 ofport->stp_port = NULL;
1850 stp_port_disable(sp);
1851 update_stp_port_state(ofport);
1854 } else if (sp && stp_port_no(sp) != s->port_num
1855 && ofport == stp_port_get_aux(sp)) {
1856 /* The port-id changed, so disable the old one if it's not
1857 * already in use by another port. */
1858 stp_port_disable(sp);
1861 sp = ofport->stp_port = stp_get_port(ofproto->stp, s->port_num);
1862 stp_port_enable(sp);
1864 stp_port_set_aux(sp, ofport);
1865 stp_port_set_priority(sp, s->priority);
1866 stp_port_set_path_cost(sp, s->path_cost);
1868 update_stp_port_state(ofport);
1874 get_stp_port_status(struct ofport *ofport_,
1875 struct ofproto_port_stp_status *s)
1877 struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
1878 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport->up.ofproto);
1879 struct stp_port *sp = ofport->stp_port;
1881 if (!ofproto->stp || !sp) {
1887 s->port_id = stp_port_get_id(sp);
1888 s->state = stp_port_get_state(sp);
1889 s->sec_in_state = (time_msec() - ofport->stp_state_entered) / 1000;
1890 s->role = stp_port_get_role(sp);
1891 stp_port_get_counts(sp, &s->tx_count, &s->rx_count, &s->error_count);
1897 stp_run(struct ofproto_dpif *ofproto)
1900 long long int now = time_msec();
1901 long long int elapsed = now - ofproto->stp_last_tick;
1902 struct stp_port *sp;
1905 stp_tick(ofproto->stp, MIN(INT_MAX, elapsed));
1906 ofproto->stp_last_tick = now;
1908 while (stp_get_changed_port(ofproto->stp, &sp)) {
1909 struct ofport_dpif *ofport = stp_port_get_aux(sp);
1912 update_stp_port_state(ofport);
1916 if (stp_check_and_reset_fdb_flush(ofproto->stp)) {
1917 mac_learning_flush(ofproto->ml, &ofproto->backer->revalidate_set);
1923 stp_wait(struct ofproto_dpif *ofproto)
1926 poll_timer_wait(1000);
1930 /* Returns true if STP should process 'flow'. */
1932 stp_should_process_flow(const struct flow *flow)
1934 return eth_addr_equals(flow->dl_dst, eth_addr_stp);
1938 stp_process_packet(const struct ofport_dpif *ofport,
1939 const struct ofpbuf *packet)
1941 struct ofpbuf payload = *packet;
1942 struct eth_header *eth = payload.data;
1943 struct stp_port *sp = ofport->stp_port;
1945 /* Sink packets on ports that have STP disabled when the bridge has
1947 if (!sp || stp_port_get_state(sp) == STP_DISABLED) {
1951 /* Trim off padding on payload. */
1952 if (payload.size > ntohs(eth->eth_type) + ETH_HEADER_LEN) {
1953 payload.size = ntohs(eth->eth_type) + ETH_HEADER_LEN;
1956 if (ofpbuf_try_pull(&payload, ETH_HEADER_LEN + LLC_HEADER_LEN)) {
1957 stp_received_bpdu(sp, payload.data, payload.size);
1962 ofproto_dpif_queue_to_priority(const struct ofproto_dpif *ofproto,
1963 uint32_t queue_id, uint32_t *priority)
1965 return dpif_queue_to_priority(ofproto->backer->dpif, queue_id, priority);
1968 static struct priority_to_dscp *
1969 get_priority(const struct ofport_dpif *ofport, uint32_t priority)
1971 struct priority_to_dscp *pdscp;
1974 hash = hash_int(priority, 0);
1975 HMAP_FOR_EACH_IN_BUCKET (pdscp, hmap_node, hash, &ofport->priorities) {
1976 if (pdscp->priority == priority) {
1984 ofproto_dpif_dscp_from_priority(const struct ofport_dpif *ofport,
1985 uint32_t priority, uint8_t *dscp)
1987 struct priority_to_dscp *pdscp = get_priority(ofport, priority);
1988 *dscp = pdscp ? pdscp->dscp : 0;
1989 return pdscp != NULL;
1993 ofport_clear_priorities(struct ofport_dpif *ofport)
1995 struct priority_to_dscp *pdscp, *next;
1997 HMAP_FOR_EACH_SAFE (pdscp, next, hmap_node, &ofport->priorities) {
1998 hmap_remove(&ofport->priorities, &pdscp->hmap_node);
2004 set_queues(struct ofport *ofport_,
2005 const struct ofproto_port_queue *qdscp_list,
2008 struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
2009 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport->up.ofproto);
2010 struct hmap new = HMAP_INITIALIZER(&new);
2013 for (i = 0; i < n_qdscp; i++) {
2014 struct priority_to_dscp *pdscp;
2018 dscp = (qdscp_list[i].dscp << 2) & IP_DSCP_MASK;
2019 if (dpif_queue_to_priority(ofproto->backer->dpif, qdscp_list[i].queue,
2024 pdscp = get_priority(ofport, priority);
2026 hmap_remove(&ofport->priorities, &pdscp->hmap_node);
2028 pdscp = xmalloc(sizeof *pdscp);
2029 pdscp->priority = priority;
2031 ofproto->backer->need_revalidate = REV_RECONFIGURE;
2034 if (pdscp->dscp != dscp) {
2036 ofproto->backer->need_revalidate = REV_RECONFIGURE;
2039 hmap_insert(&new, &pdscp->hmap_node, hash_int(pdscp->priority, 0));
2042 if (!hmap_is_empty(&ofport->priorities)) {
2043 ofport_clear_priorities(ofport);
2044 ofproto->backer->need_revalidate = REV_RECONFIGURE;
2047 hmap_swap(&new, &ofport->priorities);
2055 /* Expires all MAC learning entries associated with 'bundle' and forces its
2056 * ofproto to revalidate every flow.
2058 * Normally MAC learning entries are removed only from the ofproto associated
2059 * with 'bundle', but if 'all_ofprotos' is true, then the MAC learning entries
2060 * are removed from every ofproto. When patch ports and SLB bonds are in use
2061 * and a VM migration happens and the gratuitous ARPs are somehow lost, this
2062 * avoids a MAC_ENTRY_IDLE_TIME delay before the migrated VM can communicate
2063 * with the host from which it migrated. */
2065 bundle_flush_macs(struct ofbundle *bundle, bool all_ofprotos)
2067 struct ofproto_dpif *ofproto = bundle->ofproto;
2068 struct mac_learning *ml = ofproto->ml;
2069 struct mac_entry *mac, *next_mac;
2071 ofproto->backer->need_revalidate = REV_RECONFIGURE;
2072 LIST_FOR_EACH_SAFE (mac, next_mac, lru_node, &ml->lrus) {
2073 if (mac->port.p == bundle) {
2075 struct ofproto_dpif *o;
2077 HMAP_FOR_EACH (o, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
2079 struct mac_entry *e;
2081 e = mac_learning_lookup(o->ml, mac->mac, mac->vlan,
2084 mac_learning_expire(o->ml, e);
2090 mac_learning_expire(ml, mac);
2095 static struct ofbundle *
2096 bundle_lookup(const struct ofproto_dpif *ofproto, void *aux)
2098 struct ofbundle *bundle;
2100 HMAP_FOR_EACH_IN_BUCKET (bundle, hmap_node, hash_pointer(aux, 0),
2101 &ofproto->bundles) {
2102 if (bundle->aux == aux) {
2109 /* Looks up each of the 'n_auxes' pointers in 'auxes' as bundles and adds the
2110 * ones that are found to 'bundles'. */
2112 bundle_lookup_multiple(struct ofproto_dpif *ofproto,
2113 void **auxes, size_t n_auxes,
2114 struct hmapx *bundles)
2118 hmapx_init(bundles);
2119 for (i = 0; i < n_auxes; i++) {
2120 struct ofbundle *bundle = bundle_lookup(ofproto, auxes[i]);
2122 hmapx_add(bundles, bundle);
2128 bundle_update(struct ofbundle *bundle)
2130 struct ofport_dpif *port;
2132 bundle->floodable = true;
2133 LIST_FOR_EACH (port, bundle_node, &bundle->ports) {
2134 if (port->up.pp.config & OFPUTIL_PC_NO_FLOOD
2135 || !stp_forward_in_state(port->stp_state)) {
2136 bundle->floodable = false;
2143 bundle_del_port(struct ofport_dpif *port)
2145 struct ofbundle *bundle = port->bundle;
2147 bundle->ofproto->backer->need_revalidate = REV_RECONFIGURE;
2149 list_remove(&port->bundle_node);
2150 port->bundle = NULL;
2153 lacp_slave_unregister(bundle->lacp, port);
2156 bond_slave_unregister(bundle->bond, port);
2159 bundle_update(bundle);
2163 bundle_add_port(struct ofbundle *bundle, uint16_t ofp_port,
2164 struct lacp_slave_settings *lacp)
2166 struct ofport_dpif *port;
2168 port = get_ofp_port(bundle->ofproto, ofp_port);
2173 if (port->bundle != bundle) {
2174 bundle->ofproto->backer->need_revalidate = REV_RECONFIGURE;
2176 bundle_del_port(port);
2179 port->bundle = bundle;
2180 list_push_back(&bundle->ports, &port->bundle_node);
2181 if (port->up.pp.config & OFPUTIL_PC_NO_FLOOD
2182 || !stp_forward_in_state(port->stp_state)) {
2183 bundle->floodable = false;
2187 bundle->ofproto->backer->need_revalidate = REV_RECONFIGURE;
2188 lacp_slave_register(bundle->lacp, port, lacp);
2195 bundle_destroy(struct ofbundle *bundle)
2197 struct ofproto_dpif *ofproto;
2198 struct ofport_dpif *port, *next_port;
2205 ofproto = bundle->ofproto;
2206 for (i = 0; i < MAX_MIRRORS; i++) {
2207 struct ofmirror *m = ofproto->mirrors[i];
2209 if (m->out == bundle) {
2211 } else if (hmapx_find_and_delete(&m->srcs, bundle)
2212 || hmapx_find_and_delete(&m->dsts, bundle)) {
2213 ofproto->backer->need_revalidate = REV_RECONFIGURE;
2218 LIST_FOR_EACH_SAFE (port, next_port, bundle_node, &bundle->ports) {
2219 bundle_del_port(port);
2222 bundle_flush_macs(bundle, true);
2223 hmap_remove(&ofproto->bundles, &bundle->hmap_node);
2225 free(bundle->trunks);
2226 lacp_destroy(bundle->lacp);
2227 bond_destroy(bundle->bond);
2232 bundle_set(struct ofproto *ofproto_, void *aux,
2233 const struct ofproto_bundle_settings *s)
2235 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
2236 bool need_flush = false;
2237 struct ofport_dpif *port;
2238 struct ofbundle *bundle;
2239 unsigned long *trunks;
2245 bundle_destroy(bundle_lookup(ofproto, aux));
2249 ovs_assert(s->n_slaves == 1 || s->bond != NULL);
2250 ovs_assert((s->lacp != NULL) == (s->lacp_slaves != NULL));
2252 bundle = bundle_lookup(ofproto, aux);
2254 bundle = xmalloc(sizeof *bundle);
2256 bundle->ofproto = ofproto;
2257 hmap_insert(&ofproto->bundles, &bundle->hmap_node,
2258 hash_pointer(aux, 0));
2260 bundle->name = NULL;
2262 list_init(&bundle->ports);
2263 bundle->vlan_mode = PORT_VLAN_TRUNK;
2265 bundle->trunks = NULL;
2266 bundle->use_priority_tags = s->use_priority_tags;
2267 bundle->lacp = NULL;
2268 bundle->bond = NULL;
2270 bundle->floodable = true;
2272 bundle->src_mirrors = 0;
2273 bundle->dst_mirrors = 0;
2274 bundle->mirror_out = 0;
2277 if (!bundle->name || strcmp(s->name, bundle->name)) {
2279 bundle->name = xstrdup(s->name);
2284 if (!bundle->lacp) {
2285 ofproto->backer->need_revalidate = REV_RECONFIGURE;
2286 bundle->lacp = lacp_create();
2288 lacp_configure(bundle->lacp, s->lacp);
2290 lacp_destroy(bundle->lacp);
2291 bundle->lacp = NULL;
2294 /* Update set of ports. */
2296 for (i = 0; i < s->n_slaves; i++) {
2297 if (!bundle_add_port(bundle, s->slaves[i],
2298 s->lacp ? &s->lacp_slaves[i] : NULL)) {
2302 if (!ok || list_size(&bundle->ports) != s->n_slaves) {
2303 struct ofport_dpif *next_port;
2305 LIST_FOR_EACH_SAFE (port, next_port, bundle_node, &bundle->ports) {
2306 for (i = 0; i < s->n_slaves; i++) {
2307 if (s->slaves[i] == port->up.ofp_port) {
2312 bundle_del_port(port);
2316 ovs_assert(list_size(&bundle->ports) <= s->n_slaves);
2318 if (list_is_empty(&bundle->ports)) {
2319 bundle_destroy(bundle);
2323 /* Set VLAN tagging mode */
2324 if (s->vlan_mode != bundle->vlan_mode
2325 || s->use_priority_tags != bundle->use_priority_tags) {
2326 bundle->vlan_mode = s->vlan_mode;
2327 bundle->use_priority_tags = s->use_priority_tags;
2332 vlan = (s->vlan_mode == PORT_VLAN_TRUNK ? -1
2333 : s->vlan >= 0 && s->vlan <= 4095 ? s->vlan
2335 if (vlan != bundle->vlan) {
2336 bundle->vlan = vlan;
2340 /* Get trunked VLANs. */
2341 switch (s->vlan_mode) {
2342 case PORT_VLAN_ACCESS:
2346 case PORT_VLAN_TRUNK:
2347 trunks = CONST_CAST(unsigned long *, s->trunks);
2350 case PORT_VLAN_NATIVE_UNTAGGED:
2351 case PORT_VLAN_NATIVE_TAGGED:
2352 if (vlan != 0 && (!s->trunks
2353 || !bitmap_is_set(s->trunks, vlan)
2354 || bitmap_is_set(s->trunks, 0))) {
2355 /* Force trunking the native VLAN and prohibit trunking VLAN 0. */
2357 trunks = bitmap_clone(s->trunks, 4096);
2359 trunks = bitmap_allocate1(4096);
2361 bitmap_set1(trunks, vlan);
2362 bitmap_set0(trunks, 0);
2364 trunks = CONST_CAST(unsigned long *, s->trunks);
2371 if (!vlan_bitmap_equal(trunks, bundle->trunks)) {
2372 free(bundle->trunks);
2373 if (trunks == s->trunks) {
2374 bundle->trunks = vlan_bitmap_clone(trunks);
2376 bundle->trunks = trunks;
2381 if (trunks != s->trunks) {
2386 if (!list_is_short(&bundle->ports)) {
2387 bundle->ofproto->has_bonded_bundles = true;
2389 if (bond_reconfigure(bundle->bond, s->bond)) {
2390 ofproto->backer->need_revalidate = REV_RECONFIGURE;
2393 bundle->bond = bond_create(s->bond);
2394 ofproto->backer->need_revalidate = REV_RECONFIGURE;
2397 LIST_FOR_EACH (port, bundle_node, &bundle->ports) {
2398 bond_slave_register(bundle->bond, port, port->up.netdev);
2401 bond_destroy(bundle->bond);
2402 bundle->bond = NULL;
2405 /* If we changed something that would affect MAC learning, un-learn
2406 * everything on this port and force flow revalidation. */
2408 bundle_flush_macs(bundle, false);
2415 bundle_remove(struct ofport *port_)
2417 struct ofport_dpif *port = ofport_dpif_cast(port_);
2418 struct ofbundle *bundle = port->bundle;
2421 bundle_del_port(port);
2422 if (list_is_empty(&bundle->ports)) {
2423 bundle_destroy(bundle);
2424 } else if (list_is_short(&bundle->ports)) {
2425 bond_destroy(bundle->bond);
2426 bundle->bond = NULL;
2432 send_pdu_cb(void *port_, const void *pdu, size_t pdu_size)
2434 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 10);
2435 struct ofport_dpif *port = port_;
2436 uint8_t ea[ETH_ADDR_LEN];
2439 error = netdev_get_etheraddr(port->up.netdev, ea);
2441 struct ofpbuf packet;
2444 ofpbuf_init(&packet, 0);
2445 packet_pdu = eth_compose(&packet, eth_addr_lacp, ea, ETH_TYPE_LACP,
2447 memcpy(packet_pdu, pdu, pdu_size);
2449 send_packet(port, &packet);
2450 ofpbuf_uninit(&packet);
2452 VLOG_ERR_RL(&rl, "port %s: cannot obtain Ethernet address of iface "
2453 "%s (%s)", port->bundle->name,
2454 netdev_get_name(port->up.netdev), strerror(error));
2459 bundle_send_learning_packets(struct ofbundle *bundle)
2461 struct ofproto_dpif *ofproto = bundle->ofproto;
2462 int error, n_packets, n_errors;
2463 struct mac_entry *e;
2465 error = n_packets = n_errors = 0;
2466 LIST_FOR_EACH (e, lru_node, &ofproto->ml->lrus) {
2467 if (e->port.p != bundle) {
2468 struct ofpbuf *learning_packet;
2469 struct ofport_dpif *port;
2473 /* The assignment to "port" is unnecessary but makes "grep"ing for
2474 * struct ofport_dpif more effective. */
2475 learning_packet = bond_compose_learning_packet(bundle->bond,
2479 ret = send_packet(port, learning_packet);
2480 ofpbuf_delete(learning_packet);
2490 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
2491 VLOG_WARN_RL(&rl, "bond %s: %d errors sending %d gratuitous learning "
2492 "packets, last error was: %s",
2493 bundle->name, n_errors, n_packets, strerror(error));
2495 VLOG_DBG("bond %s: sent %d gratuitous learning packets",
2496 bundle->name, n_packets);
2501 bundle_run(struct ofbundle *bundle)
2504 lacp_run(bundle->lacp, send_pdu_cb);
2507 struct ofport_dpif *port;
2509 LIST_FOR_EACH (port, bundle_node, &bundle->ports) {
2510 bond_slave_set_may_enable(bundle->bond, port, port->may_enable);
2513 bond_run(bundle->bond, &bundle->ofproto->backer->revalidate_set,
2514 lacp_status(bundle->lacp));
2515 if (bond_should_send_learning_packets(bundle->bond)) {
2516 bundle_send_learning_packets(bundle);
2522 bundle_wait(struct ofbundle *bundle)
2525 lacp_wait(bundle->lacp);
2528 bond_wait(bundle->bond);
2535 mirror_scan(struct ofproto_dpif *ofproto)
2539 for (idx = 0; idx < MAX_MIRRORS; idx++) {
2540 if (!ofproto->mirrors[idx]) {
2547 static struct ofmirror *
2548 mirror_lookup(struct ofproto_dpif *ofproto, void *aux)
2552 for (i = 0; i < MAX_MIRRORS; i++) {
2553 struct ofmirror *mirror = ofproto->mirrors[i];
2554 if (mirror && mirror->aux == aux) {
2562 /* Update the 'dup_mirrors' member of each of the ofmirrors in 'ofproto'. */
2564 mirror_update_dups(struct ofproto_dpif *ofproto)
2568 for (i = 0; i < MAX_MIRRORS; i++) {
2569 struct ofmirror *m = ofproto->mirrors[i];
2572 m->dup_mirrors = MIRROR_MASK_C(1) << i;
2576 for (i = 0; i < MAX_MIRRORS; i++) {
2577 struct ofmirror *m1 = ofproto->mirrors[i];
2584 for (j = i + 1; j < MAX_MIRRORS; j++) {
2585 struct ofmirror *m2 = ofproto->mirrors[j];
2587 if (m2 && m1->out == m2->out && m1->out_vlan == m2->out_vlan) {
2588 m1->dup_mirrors |= MIRROR_MASK_C(1) << j;
2589 m2->dup_mirrors |= m1->dup_mirrors;
2596 mirror_set(struct ofproto *ofproto_, void *aux,
2597 const struct ofproto_mirror_settings *s)
2599 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
2600 mirror_mask_t mirror_bit;
2601 struct ofbundle *bundle;
2602 struct ofmirror *mirror;
2603 struct ofbundle *out;
2604 struct hmapx srcs; /* Contains "struct ofbundle *"s. */
2605 struct hmapx dsts; /* Contains "struct ofbundle *"s. */
2608 mirror = mirror_lookup(ofproto, aux);
2610 mirror_destroy(mirror);
2616 idx = mirror_scan(ofproto);
2618 VLOG_WARN("bridge %s: maximum of %d port mirrors reached, "
2620 ofproto->up.name, MAX_MIRRORS, s->name);
2624 mirror = ofproto->mirrors[idx] = xzalloc(sizeof *mirror);
2625 mirror->ofproto = ofproto;
2628 mirror->out_vlan = -1;
2629 mirror->name = NULL;
2632 if (!mirror->name || strcmp(s->name, mirror->name)) {
2634 mirror->name = xstrdup(s->name);
2637 /* Get the new configuration. */
2638 if (s->out_bundle) {
2639 out = bundle_lookup(ofproto, s->out_bundle);
2641 mirror_destroy(mirror);
2647 out_vlan = s->out_vlan;
2649 bundle_lookup_multiple(ofproto, s->srcs, s->n_srcs, &srcs);
2650 bundle_lookup_multiple(ofproto, s->dsts, s->n_dsts, &dsts);
2652 /* If the configuration has not changed, do nothing. */
2653 if (hmapx_equals(&srcs, &mirror->srcs)
2654 && hmapx_equals(&dsts, &mirror->dsts)
2655 && vlan_bitmap_equal(mirror->vlans, s->src_vlans)
2656 && mirror->out == out
2657 && mirror->out_vlan == out_vlan)
2659 hmapx_destroy(&srcs);
2660 hmapx_destroy(&dsts);
2664 hmapx_swap(&srcs, &mirror->srcs);
2665 hmapx_destroy(&srcs);
2667 hmapx_swap(&dsts, &mirror->dsts);
2668 hmapx_destroy(&dsts);
2670 free(mirror->vlans);
2671 mirror->vlans = vlan_bitmap_clone(s->src_vlans);
2674 mirror->out_vlan = out_vlan;
2676 /* Update bundles. */
2677 mirror_bit = MIRROR_MASK_C(1) << mirror->idx;
2678 HMAP_FOR_EACH (bundle, hmap_node, &mirror->ofproto->bundles) {
2679 if (hmapx_contains(&mirror->srcs, bundle)) {
2680 bundle->src_mirrors |= mirror_bit;
2682 bundle->src_mirrors &= ~mirror_bit;
2685 if (hmapx_contains(&mirror->dsts, bundle)) {
2686 bundle->dst_mirrors |= mirror_bit;
2688 bundle->dst_mirrors &= ~mirror_bit;
2691 if (mirror->out == bundle) {
2692 bundle->mirror_out |= mirror_bit;
2694 bundle->mirror_out &= ~mirror_bit;
2698 ofproto->backer->need_revalidate = REV_RECONFIGURE;
2699 ofproto->has_mirrors = true;
2700 mac_learning_flush(ofproto->ml,
2701 &ofproto->backer->revalidate_set);
2702 mirror_update_dups(ofproto);
2708 mirror_destroy(struct ofmirror *mirror)
2710 struct ofproto_dpif *ofproto;
2711 mirror_mask_t mirror_bit;
2712 struct ofbundle *bundle;
2719 ofproto = mirror->ofproto;
2720 ofproto->backer->need_revalidate = REV_RECONFIGURE;
2721 mac_learning_flush(ofproto->ml, &ofproto->backer->revalidate_set);
2723 mirror_bit = MIRROR_MASK_C(1) << mirror->idx;
2724 HMAP_FOR_EACH (bundle, hmap_node, &ofproto->bundles) {
2725 bundle->src_mirrors &= ~mirror_bit;
2726 bundle->dst_mirrors &= ~mirror_bit;
2727 bundle->mirror_out &= ~mirror_bit;
2730 hmapx_destroy(&mirror->srcs);
2731 hmapx_destroy(&mirror->dsts);
2732 free(mirror->vlans);
2734 ofproto->mirrors[mirror->idx] = NULL;
2738 mirror_update_dups(ofproto);
2740 ofproto->has_mirrors = false;
2741 for (i = 0; i < MAX_MIRRORS; i++) {
2742 if (ofproto->mirrors[i]) {
2743 ofproto->has_mirrors = true;
2750 mirror_get_stats(struct ofproto *ofproto_, void *aux,
2751 uint64_t *packets, uint64_t *bytes)
2753 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
2754 struct ofmirror *mirror = mirror_lookup(ofproto, aux);
2757 *packets = *bytes = UINT64_MAX;
2763 *packets = mirror->packet_count;
2764 *bytes = mirror->byte_count;
2770 set_flood_vlans(struct ofproto *ofproto_, unsigned long *flood_vlans)
2772 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
2773 if (mac_learning_set_flood_vlans(ofproto->ml, flood_vlans)) {
2774 mac_learning_flush(ofproto->ml, &ofproto->backer->revalidate_set);
2780 is_mirror_output_bundle(const struct ofproto *ofproto_, void *aux)
2782 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
2783 struct ofbundle *bundle = bundle_lookup(ofproto, aux);
2784 return bundle && bundle->mirror_out != 0;
2788 forward_bpdu_changed(struct ofproto *ofproto_)
2790 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
2791 ofproto->backer->need_revalidate = REV_RECONFIGURE;
2795 set_mac_table_config(struct ofproto *ofproto_, unsigned int idle_time,
2798 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
2799 mac_learning_set_idle_time(ofproto->ml, idle_time);
2800 mac_learning_set_max_entries(ofproto->ml, max_entries);
2805 struct ofport_dpif *
2806 get_ofp_port(const struct ofproto_dpif *ofproto, uint16_t ofp_port)
2808 struct ofport *ofport = ofproto_get_port(&ofproto->up, ofp_port);
2809 return ofport ? ofport_dpif_cast(ofport) : NULL;
2812 struct ofport_dpif *
2813 get_odp_port(const struct ofproto_dpif *ofproto, uint32_t odp_port)
2815 struct ofport_dpif *port = odp_port_to_ofport(ofproto->backer, odp_port);
2816 return port && &ofproto->up == port->up.ofproto ? port : NULL;
2820 ofproto_port_from_dpif_port(struct ofproto_dpif *ofproto,
2821 struct ofproto_port *ofproto_port,
2822 struct dpif_port *dpif_port)
2824 ofproto_port->name = dpif_port->name;
2825 ofproto_port->type = dpif_port->type;
2826 ofproto_port->ofp_port = odp_port_to_ofp_port(ofproto, dpif_port->port_no);
2829 struct ofport_dpif *
2830 ofport_get_peer(const struct ofport_dpif *ofport_dpif)
2832 const struct ofproto_dpif *ofproto;
2833 const struct dpif_backer *backer;
2836 peer = netdev_vport_patch_peer(ofport_dpif->up.netdev);
2841 backer = ofproto_dpif_cast(ofport_dpif->up.ofproto)->backer;
2842 HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
2843 struct ofport *ofport;
2845 if (ofproto->backer != backer) {
2849 ofport = shash_find_data(&ofproto->up.port_by_name, peer);
2851 return ofport_dpif_cast(ofport);
2858 port_run_fast(struct ofport_dpif *ofport)
2860 if (ofport->cfm && cfm_should_send_ccm(ofport->cfm)) {
2861 struct ofpbuf packet;
2863 ofpbuf_init(&packet, 0);
2864 cfm_compose_ccm(ofport->cfm, &packet, ofport->up.pp.hw_addr);
2865 send_packet(ofport, &packet);
2866 ofpbuf_uninit(&packet);
2869 if (ofport->bfd && bfd_should_send_packet(ofport->bfd)) {
2870 struct ofpbuf packet;
2872 ofpbuf_init(&packet, 0);
2873 bfd_put_packet(ofport->bfd, &packet, ofport->up.pp.hw_addr);
2874 send_packet(ofport, &packet);
2875 ofpbuf_uninit(&packet);
2880 port_run(struct ofport_dpif *ofport)
2882 long long int carrier_seq = netdev_get_carrier_resets(ofport->up.netdev);
2883 bool carrier_changed = carrier_seq != ofport->carrier_seq;
2884 bool enable = netdev_get_carrier(ofport->up.netdev);
2886 ofport->carrier_seq = carrier_seq;
2888 port_run_fast(ofport);
2890 if (ofport->tnl_port
2891 && tnl_port_reconfigure(&ofport->up, ofport->odp_port,
2892 &ofport->tnl_port)) {
2893 ofproto_dpif_cast(ofport->up.ofproto)->backer->need_revalidate = true;
2897 int cfm_opup = cfm_get_opup(ofport->cfm);
2899 cfm_run(ofport->cfm);
2900 enable = enable && !cfm_get_fault(ofport->cfm);
2902 if (cfm_opup >= 0) {
2903 enable = enable && cfm_opup;
2908 bfd_run(ofport->bfd);
2909 enable = enable && bfd_forwarding(ofport->bfd);
2912 if (ofport->bundle) {
2913 enable = enable && lacp_slave_may_enable(ofport->bundle->lacp, ofport);
2914 if (carrier_changed) {
2915 lacp_slave_carrier_changed(ofport->bundle->lacp, ofport);
2919 if (ofport->may_enable != enable) {
2920 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport->up.ofproto);
2922 if (ofproto->has_bundle_action) {
2923 ofproto->backer->need_revalidate = REV_PORT_TOGGLED;
2927 ofport->may_enable = enable;
2931 port_wait(struct ofport_dpif *ofport)
2934 cfm_wait(ofport->cfm);
2938 bfd_wait(ofport->bfd);
2943 port_query_by_name(const struct ofproto *ofproto_, const char *devname,
2944 struct ofproto_port *ofproto_port)
2946 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
2947 struct dpif_port dpif_port;
2950 if (sset_contains(&ofproto->ghost_ports, devname)) {
2951 const char *type = netdev_get_type_from_name(devname);
2953 /* We may be called before ofproto->up.port_by_name is populated with
2954 * the appropriate ofport. For this reason, we must get the name and
2955 * type from the netdev layer directly. */
2957 const struct ofport *ofport;
2959 ofport = shash_find_data(&ofproto->up.port_by_name, devname);
2960 ofproto_port->ofp_port = ofport ? ofport->ofp_port : OFPP_NONE;
2961 ofproto_port->name = xstrdup(devname);
2962 ofproto_port->type = xstrdup(type);
2968 if (!sset_contains(&ofproto->ports, devname)) {
2971 error = dpif_port_query_by_name(ofproto->backer->dpif,
2972 devname, &dpif_port);
2974 ofproto_port_from_dpif_port(ofproto, ofproto_port, &dpif_port);
2980 port_add(struct ofproto *ofproto_, struct netdev *netdev)
2982 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
2983 const char *devname = netdev_get_name(netdev);
2984 char namebuf[NETDEV_VPORT_NAME_BUFSIZE];
2985 const char *dp_port_name;
2987 if (netdev_vport_is_patch(netdev)) {
2988 sset_add(&ofproto->ghost_ports, netdev_get_name(netdev));
2992 dp_port_name = netdev_vport_get_dpif_port(netdev, namebuf, sizeof namebuf);
2993 if (!dpif_port_exists(ofproto->backer->dpif, dp_port_name)) {
2994 uint32_t port_no = UINT32_MAX;
2997 error = dpif_port_add(ofproto->backer->dpif, netdev, &port_no);
3001 if (netdev_get_tunnel_config(netdev)) {
3002 simap_put(&ofproto->backer->tnl_backers, dp_port_name, port_no);
3006 if (netdev_get_tunnel_config(netdev)) {
3007 sset_add(&ofproto->ghost_ports, devname);
3009 sset_add(&ofproto->ports, devname);
3015 port_del(struct ofproto *ofproto_, uint16_t ofp_port)
3017 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
3018 struct ofport_dpif *ofport = get_ofp_port(ofproto, ofp_port);
3025 sset_find_and_delete(&ofproto->ghost_ports,
3026 netdev_get_name(ofport->up.netdev));
3027 ofproto->backer->need_revalidate = REV_RECONFIGURE;
3028 if (!ofport->tnl_port) {
3029 error = dpif_port_del(ofproto->backer->dpif, ofport->odp_port);
3031 /* The caller is going to close ofport->up.netdev. If this is a
3032 * bonded port, then the bond is using that netdev, so remove it
3033 * from the bond. The client will need to reconfigure everything
3034 * after deleting ports, so then the slave will get re-added. */
3035 bundle_remove(&ofport->up);
3042 port_get_stats(const struct ofport *ofport_, struct netdev_stats *stats)
3044 struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
3049 error = netdev_get_stats(ofport->up.netdev, stats);
3051 if (!error && ofport_->ofp_port == OFPP_LOCAL) {
3052 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport->up.ofproto);
3054 /* ofproto->stats.tx_packets represents packets that we created
3055 * internally and sent to some port (e.g. packets sent with
3056 * send_packet()). Account for them as if they had come from
3057 * OFPP_LOCAL and got forwarded. */
3059 if (stats->rx_packets != UINT64_MAX) {
3060 stats->rx_packets += ofproto->stats.tx_packets;
3063 if (stats->rx_bytes != UINT64_MAX) {
3064 stats->rx_bytes += ofproto->stats.tx_bytes;
3067 /* ofproto->stats.rx_packets represents packets that were received on
3068 * some port and we processed internally and dropped (e.g. STP).
3069 * Account for them as if they had been forwarded to OFPP_LOCAL. */
3071 if (stats->tx_packets != UINT64_MAX) {
3072 stats->tx_packets += ofproto->stats.rx_packets;
3075 if (stats->tx_bytes != UINT64_MAX) {
3076 stats->tx_bytes += ofproto->stats.rx_bytes;
3083 struct port_dump_state {
3088 struct ofproto_port port;
3093 port_dump_start(const struct ofproto *ofproto_ OVS_UNUSED, void **statep)
3095 *statep = xzalloc(sizeof(struct port_dump_state));
3100 port_dump_next(const struct ofproto *ofproto_, void *state_,
3101 struct ofproto_port *port)
3103 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
3104 struct port_dump_state *state = state_;
3105 const struct sset *sset;
3106 struct sset_node *node;
3108 if (state->has_port) {
3109 ofproto_port_destroy(&state->port);
3110 state->has_port = false;
3112 sset = state->ghost ? &ofproto->ghost_ports : &ofproto->ports;
3113 while ((node = sset_at_position(sset, &state->bucket, &state->offset))) {
3116 error = port_query_by_name(ofproto_, node->name, &state->port);
3118 *port = state->port;
3119 state->has_port = true;
3121 } else if (error != ENODEV) {
3126 if (!state->ghost) {
3127 state->ghost = true;
3130 return port_dump_next(ofproto_, state_, port);
3137 port_dump_done(const struct ofproto *ofproto_ OVS_UNUSED, void *state_)
3139 struct port_dump_state *state = state_;
3141 if (state->has_port) {
3142 ofproto_port_destroy(&state->port);
3149 port_poll(const struct ofproto *ofproto_, char **devnamep)
3151 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
3153 if (ofproto->port_poll_errno) {
3154 int error = ofproto->port_poll_errno;
3155 ofproto->port_poll_errno = 0;
3159 if (sset_is_empty(&ofproto->port_poll_set)) {
3163 *devnamep = sset_pop(&ofproto->port_poll_set);
3168 port_poll_wait(const struct ofproto *ofproto_)
3170 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
3171 dpif_port_poll_wait(ofproto->backer->dpif);
3175 port_is_lacp_current(const struct ofport *ofport_)
3177 const struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
3178 return (ofport->bundle && ofport->bundle->lacp
3179 ? lacp_slave_is_current(ofport->bundle->lacp, ofport)
3183 /* Upcall handling. */
3185 /* Flow miss batching.
3187 * Some dpifs implement operations faster when you hand them off in a batch.
3188 * To allow batching, "struct flow_miss" queues the dpif-related work needed
3189 * for a given flow. Each "struct flow_miss" corresponds to sending one or
3190 * more packets, plus possibly installing the flow in the dpif.
3192 * So far we only batch the operations that affect flow setup time the most.
3193 * It's possible to batch more than that, but the benefit might be minimal. */
3195 struct hmap_node hmap_node;
3196 struct ofproto_dpif *ofproto;
3198 enum odp_key_fitness key_fitness;
3199 const struct nlattr *key;
3201 struct list packets;
3202 enum dpif_upcall_type upcall_type;
3205 struct flow_miss_op {
3206 struct dpif_op dpif_op;
3208 uint64_t slow_stub[128 / 8]; /* Buffer for compose_slow_path() */
3209 struct xlate_out xout;
3210 bool xout_garbage; /* 'xout' needs to be uninitialized? */
3213 /* Sends an OFPT_PACKET_IN message for 'packet' of type OFPR_NO_MATCH to each
3214 * OpenFlow controller as necessary according to their individual
3215 * configurations. */
3217 send_packet_in_miss(struct ofproto_dpif *ofproto, const struct ofpbuf *packet,
3218 const struct flow *flow)
3220 struct ofputil_packet_in pin;
3222 pin.packet = packet->data;
3223 pin.packet_len = packet->size;
3224 pin.reason = OFPR_NO_MATCH;
3225 pin.controller_id = 0;
3230 pin.send_len = 0; /* not used for flow table misses */
3232 flow_get_metadata(flow, &pin.fmd);
3234 connmgr_send_packet_in(ofproto->up.connmgr, &pin);
3237 static struct flow_miss *
3238 flow_miss_find(struct hmap *todo, const struct ofproto_dpif *ofproto,
3239 const struct flow *flow, uint32_t hash)
3241 struct flow_miss *miss;
3243 HMAP_FOR_EACH_WITH_HASH (miss, hmap_node, hash, todo) {
3244 if (miss->ofproto == ofproto && flow_equal(&miss->flow, flow)) {
3252 /* Partially Initializes 'op' as an "execute" operation for 'miss' and
3253 * 'packet'. The caller must initialize op->actions and op->actions_len. If
3254 * 'miss' is associated with a subfacet the caller must also initialize the
3255 * returned op->subfacet, and if anything needs to be freed after processing
3256 * the op, the caller must initialize op->garbage also. */
3258 init_flow_miss_execute_op(struct flow_miss *miss, struct ofpbuf *packet,
3259 struct flow_miss_op *op)
3261 if (miss->flow.in_port
3262 != vsp_realdev_to_vlandev(miss->ofproto, miss->flow.in_port,
3263 miss->flow.vlan_tci)) {
3264 /* This packet was received on a VLAN splinter port. We
3265 * added a VLAN to the packet to make the packet resemble
3266 * the flow, but the actions were composed assuming that
3267 * the packet contained no VLAN. So, we must remove the
3268 * VLAN header from the packet before trying to execute the
3270 eth_pop_vlan(packet);
3273 op->xout_garbage = false;
3274 op->dpif_op.type = DPIF_OP_EXECUTE;
3275 op->dpif_op.u.execute.key = miss->key;
3276 op->dpif_op.u.execute.key_len = miss->key_len;
3277 op->dpif_op.u.execute.packet = packet;
3280 /* Helper for handle_flow_miss_without_facet() and
3281 * handle_flow_miss_with_facet(). */
3283 handle_flow_miss_common(struct rule_dpif *rule,
3284 struct ofpbuf *packet, const struct flow *flow)
3286 struct ofproto_dpif *ofproto = ofproto_dpif_cast(rule->up.ofproto);
3288 if (rule->up.cr.priority == FAIL_OPEN_PRIORITY) {
3290 * Extra-special case for fail-open mode.
3292 * We are in fail-open mode and the packet matched the fail-open
3293 * rule, but we are connected to a controller too. We should send
3294 * the packet up to the controller in the hope that it will try to
3295 * set up a flow and thereby allow us to exit fail-open.
3297 * See the top-level comment in fail-open.c for more information.
3299 send_packet_in_miss(ofproto, packet, flow);
3303 /* Figures out whether a flow that missed in 'ofproto', whose details are in
3304 * 'miss' masked by 'wc', is likely to be worth tracking in detail in userspace
3305 * and (usually) installing a datapath flow. The answer is usually "yes" (a
3306 * return value of true). However, for short flows the cost of bookkeeping is
3307 * much higher than the benefits, so when the datapath holds a large number of
3308 * flows we impose some heuristics to decide which flows are likely to be worth
3311 flow_miss_should_make_facet(struct flow_miss *miss, struct flow_wildcards *wc)
3313 struct dpif_backer *backer = miss->ofproto->backer;
3316 if (!backer->governor) {
3319 n_subfacets = hmap_count(&backer->subfacets);
3320 if (n_subfacets * 2 <= flow_eviction_threshold) {
3324 backer->governor = governor_create();
3327 hash = flow_hash_in_wildcards(&miss->flow, wc, 0);
3328 return governor_should_install_flow(backer->governor, hash,
3329 list_size(&miss->packets));
3332 /* Handles 'miss' without creating a facet or subfacet or creating any datapath
3333 * flow. 'miss->flow' must have matched 'rule' and been xlated into 'xout'.
3334 * May add an "execute" operation to 'ops' and increment '*n_ops'. */
3336 handle_flow_miss_without_facet(struct rule_dpif *rule, struct xlate_out *xout,
3337 struct flow_miss *miss,
3338 struct flow_miss_op *ops, size_t *n_ops)
3340 struct ofpbuf *packet;
3342 LIST_FOR_EACH (packet, list_node, &miss->packets) {
3344 COVERAGE_INC(facet_suppress);
3346 handle_flow_miss_common(rule, packet, &miss->flow);
3349 struct xlate_in xin;
3351 xlate_in_init(&xin, miss->ofproto, &miss->flow, rule, 0, packet);
3352 xlate_actions_for_side_effects(&xin);
3355 if (xout->odp_actions.size) {
3356 struct flow_miss_op *op = &ops[*n_ops];
3357 struct dpif_execute *execute = &op->dpif_op.u.execute;
3359 init_flow_miss_execute_op(miss, packet, op);
3360 xlate_out_copy(&op->xout, xout);
3361 execute->actions = op->xout.odp_actions.data;
3362 execute->actions_len = op->xout.odp_actions.size;
3363 op->xout_garbage = true;
3370 /* Handles 'miss', which matches 'facet'. May add any required datapath
3371 * operations to 'ops', incrementing '*n_ops' for each new op.
3373 * All of the packets in 'miss' are considered to have arrived at time 'now'.
3374 * This is really important only for new facets: if we just called time_msec()
3375 * here, then the new subfacet or its packets could look (occasionally) as
3376 * though it was used some time after the facet was used. That can make a
3377 * one-packet flow look like it has a nonzero duration, which looks odd in
3378 * e.g. NetFlow statistics.
3380 * If non-null, 'stats' will be folded into 'facet'. */
3382 handle_flow_miss_with_facet(struct flow_miss *miss, struct facet *facet,
3383 long long int now, struct dpif_flow_stats *stats,
3384 struct flow_miss_op *ops, size_t *n_ops)
3386 struct ofproto_dpif *ofproto = ofproto_dpif_cast(facet->rule->up.ofproto);
3387 enum subfacet_path want_path;
3388 struct subfacet *subfacet;
3389 struct ofpbuf *packet;
3391 subfacet = subfacet_create(facet, miss, now);
3392 want_path = facet->xout.slow ? SF_SLOW_PATH : SF_FAST_PATH;
3394 subfacet_update_stats(subfacet, stats);
3397 LIST_FOR_EACH (packet, list_node, &miss->packets) {
3398 struct flow_miss_op *op = &ops[*n_ops];
3400 handle_flow_miss_common(facet->rule, packet, &miss->flow);
3402 if (want_path != SF_FAST_PATH) {
3403 struct xlate_in xin;
3405 xlate_in_init(&xin, ofproto, &miss->flow, facet->rule, 0, packet);
3406 xlate_actions_for_side_effects(&xin);
3409 if (facet->xout.odp_actions.size) {
3410 struct dpif_execute *execute = &op->dpif_op.u.execute;
3412 init_flow_miss_execute_op(miss, packet, op);
3413 execute->actions = facet->xout.odp_actions.data,
3414 execute->actions_len = facet->xout.odp_actions.size;
3419 if (miss->upcall_type == DPIF_UC_MISS || subfacet->path != want_path) {
3420 struct flow_miss_op *op = &ops[(*n_ops)++];
3421 struct dpif_flow_put *put = &op->dpif_op.u.flow_put;
3423 subfacet->path = want_path;
3425 op->xout_garbage = false;
3426 op->dpif_op.type = DPIF_OP_FLOW_PUT;
3427 put->flags = DPIF_FP_CREATE | DPIF_FP_MODIFY;
3428 put->key = miss->key;
3429 put->key_len = miss->key_len;
3430 if (want_path == SF_FAST_PATH) {
3431 put->actions = facet->xout.odp_actions.data;
3432 put->actions_len = facet->xout.odp_actions.size;
3434 compose_slow_path(ofproto, &miss->flow, facet->xout.slow,
3435 op->slow_stub, sizeof op->slow_stub,
3436 &put->actions, &put->actions_len);
3442 /* Handles flow miss 'miss'. May add any required datapath operations
3443 * to 'ops', incrementing '*n_ops' for each new op. */
3445 handle_flow_miss(struct flow_miss *miss, struct flow_miss_op *ops,
3448 struct ofproto_dpif *ofproto = miss->ofproto;
3449 struct dpif_flow_stats stats__;
3450 struct dpif_flow_stats *stats = &stats__;
3451 struct ofpbuf *packet;
3452 struct facet *facet;
3456 memset(stats, 0, sizeof *stats);
3458 LIST_FOR_EACH (packet, list_node, &miss->packets) {
3459 stats->tcp_flags |= packet_get_tcp_flags(packet, &miss->flow);
3460 stats->n_bytes += packet->size;
3464 facet = facet_lookup_valid(ofproto, &miss->flow);
3466 struct flow_wildcards wc;
3467 struct rule_dpif *rule;
3468 struct xlate_out xout;
3469 struct xlate_in xin;
3471 flow_wildcards_init_catchall(&wc);
3472 rule = rule_dpif_lookup(ofproto, &miss->flow, &wc);
3473 rule_credit_stats(rule, stats);
3475 xlate_in_init(&xin, ofproto, &miss->flow, rule, stats->tcp_flags,
3477 xin.resubmit_stats = stats;
3478 xin.may_learn = true;
3479 xlate_actions(&xin, &xout);
3480 flow_wildcards_or(&xout.wc, &xout.wc, &wc);
3482 /* There does not exist a bijection between 'struct flow' and datapath
3483 * flow keys with fitness ODP_FIT_TO_LITTLE. This breaks a fundamental
3484 * assumption used throughout the facet and subfacet handling code.
3485 * Since we have to handle these misses in userspace anyway, we simply
3486 * skip facet creation, avoiding the problem altogether. */
3487 if (miss->key_fitness == ODP_FIT_TOO_LITTLE
3488 || !flow_miss_should_make_facet(miss, &xout.wc)) {
3489 handle_flow_miss_without_facet(rule, &xout, miss, ops, n_ops);
3493 facet = facet_create(miss, rule, &xout, stats);
3496 handle_flow_miss_with_facet(miss, facet, now, stats, ops, n_ops);
3499 static struct drop_key *
3500 drop_key_lookup(const struct dpif_backer *backer, const struct nlattr *key,
3503 struct drop_key *drop_key;
3505 HMAP_FOR_EACH_WITH_HASH (drop_key, hmap_node, hash_bytes(key, key_len, 0),
3506 &backer->drop_keys) {
3507 if (drop_key->key_len == key_len
3508 && !memcmp(drop_key->key, key, key_len)) {
3516 drop_key_clear(struct dpif_backer *backer)
3518 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 15);
3519 struct drop_key *drop_key, *next;
3521 HMAP_FOR_EACH_SAFE (drop_key, next, hmap_node, &backer->drop_keys) {
3524 error = dpif_flow_del(backer->dpif, drop_key->key, drop_key->key_len,
3526 if (error && !VLOG_DROP_WARN(&rl)) {
3527 struct ds ds = DS_EMPTY_INITIALIZER;
3528 odp_flow_key_format(drop_key->key, drop_key->key_len, &ds);
3529 VLOG_WARN("Failed to delete drop key (%s) (%s)", strerror(error),
3534 hmap_remove(&backer->drop_keys, &drop_key->hmap_node);
3535 free(drop_key->key);
3540 /* Given a datpath, packet, and flow metadata ('backer', 'packet', and 'key'
3541 * respectively), populates 'flow' with the result of odp_flow_key_to_flow().
3542 * Optionally, if nonnull, populates 'fitnessp' with the fitness of 'flow' as
3543 * returned by odp_flow_key_to_flow(). Also, optionally populates 'ofproto'
3544 * with the ofproto_dpif, and 'odp_in_port' with the datapath in_port, that
3545 * 'packet' ingressed.
3547 * If 'ofproto' is nonnull, requires 'flow''s in_port to exist. Otherwise sets
3548 * 'flow''s in_port to OFPP_NONE.
3550 * This function does post-processing on data returned from
3551 * odp_flow_key_to_flow() to help make VLAN splinters transparent to the rest
3552 * of the upcall processing logic. In particular, if the extracted in_port is
3553 * a VLAN splinter port, it replaces flow->in_port by the "real" port, sets
3554 * flow->vlan_tci correctly for the VLAN of the VLAN splinter port, and pushes
3555 * a VLAN header onto 'packet' (if it is nonnull).
3557 * Similarly, this function also includes some logic to help with tunnels. It
3558 * may modify 'flow' as necessary to make the tunneling implementation
3559 * transparent to the upcall processing logic.
3561 * Returns 0 if successful, ENODEV if the parsed flow has no associated ofport,
3562 * or some other positive errno if there are other problems. */
3564 ofproto_receive(const struct dpif_backer *backer, struct ofpbuf *packet,
3565 const struct nlattr *key, size_t key_len,
3566 struct flow *flow, enum odp_key_fitness *fitnessp,
3567 struct ofproto_dpif **ofproto, uint32_t *odp_in_port)
3569 const struct ofport_dpif *port;
3570 enum odp_key_fitness fitness;
3573 fitness = odp_flow_key_to_flow(key, key_len, flow);
3574 if (fitness == ODP_FIT_ERROR) {
3580 *odp_in_port = flow->in_port;
3583 port = (tnl_port_should_receive(flow)
3584 ? ofport_dpif_cast(tnl_port_receive(flow))
3585 : odp_port_to_ofport(backer, flow->in_port));
3586 flow->in_port = port ? port->up.ofp_port : OFPP_NONE;
3591 /* XXX: Since the tunnel module is not scoped per backer, for a tunnel port
3592 * it's theoretically possible that we'll receive an ofport belonging to an
3593 * entirely different datapath. In practice, this can't happen because no
3594 * platforms has two separate datapaths which each support tunneling. */
3595 ovs_assert(ofproto_dpif_cast(port->up.ofproto)->backer == backer);
3597 if (vsp_adjust_flow(ofproto_dpif_cast(port->up.ofproto), flow)) {
3599 /* Make the packet resemble the flow, so that it gets sent to
3600 * an OpenFlow controller properly, so that it looks correct
3601 * for sFlow, and so that flow_extract() will get the correct
3602 * vlan_tci if it is called on 'packet'.
3604 * The allocated space inside 'packet' probably also contains
3605 * 'key', that is, both 'packet' and 'key' are probably part of
3606 * a struct dpif_upcall (see the large comment on that
3607 * structure definition), so pushing data on 'packet' is in
3608 * general not a good idea since it could overwrite 'key' or
3609 * free it as a side effect. However, it's OK in this special
3610 * case because we know that 'packet' is inside a Netlink
3611 * attribute: pushing 4 bytes will just overwrite the 4-byte
3612 * "struct nlattr", which is fine since we don't need that
3613 * header anymore. */
3614 eth_push_vlan(packet, flow->vlan_tci);
3616 /* We can't reproduce 'key' from 'flow'. */
3617 fitness = fitness == ODP_FIT_PERFECT ? ODP_FIT_TOO_MUCH : fitness;
3622 *ofproto = ofproto_dpif_cast(port->up.ofproto);
3627 *fitnessp = fitness;
3633 handle_miss_upcalls(struct dpif_backer *backer, struct dpif_upcall *upcalls,
3636 struct dpif_upcall *upcall;
3637 struct flow_miss *miss;
3638 struct flow_miss misses[FLOW_MISS_MAX_BATCH];
3639 struct flow_miss_op flow_miss_ops[FLOW_MISS_MAX_BATCH * 2];
3640 struct dpif_op *dpif_ops[FLOW_MISS_MAX_BATCH * 2];
3650 /* Construct the to-do list.
3652 * This just amounts to extracting the flow from each packet and sticking
3653 * the packets that have the same flow in the same "flow_miss" structure so
3654 * that we can process them together. */
3657 for (upcall = upcalls; upcall < &upcalls[n_upcalls]; upcall++) {
3658 struct flow_miss *miss = &misses[n_misses];
3659 struct flow_miss *existing_miss;
3660 struct ofproto_dpif *ofproto;
3661 uint32_t odp_in_port;
3666 error = ofproto_receive(backer, upcall->packet, upcall->key,
3667 upcall->key_len, &flow, &miss->key_fitness,
3668 &ofproto, &odp_in_port);
3669 if (error == ENODEV) {
3670 struct drop_key *drop_key;
3672 /* Received packet on datapath port for which we couldn't
3673 * associate an ofproto. This can happen if a port is removed
3674 * while traffic is being received. Print a rate-limited message
3675 * in case it happens frequently. Install a drop flow so
3676 * that future packets of the flow are inexpensively dropped
3678 VLOG_INFO_RL(&rl, "received packet on unassociated datapath port "
3679 "%"PRIu32, odp_in_port);
3681 drop_key = drop_key_lookup(backer, upcall->key, upcall->key_len);
3683 drop_key = xmalloc(sizeof *drop_key);
3684 drop_key->key = xmemdup(upcall->key, upcall->key_len);
3685 drop_key->key_len = upcall->key_len;
3687 hmap_insert(&backer->drop_keys, &drop_key->hmap_node,
3688 hash_bytes(drop_key->key, drop_key->key_len, 0));
3689 dpif_flow_put(backer->dpif, DPIF_FP_CREATE | DPIF_FP_MODIFY,
3690 drop_key->key, drop_key->key_len, NULL, 0, NULL);
3698 ofproto->n_missed++;
3699 flow_extract(upcall->packet, flow.skb_priority, flow.skb_mark,
3700 &flow.tunnel, flow.in_port, &miss->flow);
3702 /* Add other packets to a to-do list. */
3703 hash = flow_hash(&miss->flow, 0);
3704 existing_miss = flow_miss_find(&todo, ofproto, &miss->flow, hash);
3705 if (!existing_miss) {
3706 hmap_insert(&todo, &miss->hmap_node, hash);
3707 miss->ofproto = ofproto;
3708 miss->key = upcall->key;
3709 miss->key_len = upcall->key_len;
3710 miss->upcall_type = upcall->type;
3711 list_init(&miss->packets);
3715 miss = existing_miss;
3717 list_push_back(&miss->packets, &upcall->packet->list_node);
3720 /* Process each element in the to-do list, constructing the set of
3721 * operations to batch. */
3723 HMAP_FOR_EACH (miss, hmap_node, &todo) {
3724 handle_flow_miss(miss, flow_miss_ops, &n_ops);
3726 ovs_assert(n_ops <= ARRAY_SIZE(flow_miss_ops));
3728 /* Execute batch. */
3729 for (i = 0; i < n_ops; i++) {
3730 dpif_ops[i] = &flow_miss_ops[i].dpif_op;
3732 dpif_operate(backer->dpif, dpif_ops, n_ops);
3735 for (i = 0; i < n_ops; i++) {
3736 if (flow_miss_ops[i].xout_garbage) {
3737 xlate_out_uninit(&flow_miss_ops[i].xout);
3740 hmap_destroy(&todo);
3743 static enum { SFLOW_UPCALL, MISS_UPCALL, BAD_UPCALL, FLOW_SAMPLE_UPCALL,
3745 classify_upcall(const struct dpif_upcall *upcall)
3747 size_t userdata_len;
3748 union user_action_cookie cookie;
3750 /* First look at the upcall type. */
3751 switch (upcall->type) {
3752 case DPIF_UC_ACTION:
3758 case DPIF_N_UC_TYPES:
3760 VLOG_WARN_RL(&rl, "upcall has unexpected type %"PRIu32, upcall->type);
3764 /* "action" upcalls need a closer look. */
3765 if (!upcall->userdata) {
3766 VLOG_WARN_RL(&rl, "action upcall missing cookie");
3769 userdata_len = nl_attr_get_size(upcall->userdata);
3770 if (userdata_len < sizeof cookie.type
3771 || userdata_len > sizeof cookie) {
3772 VLOG_WARN_RL(&rl, "action upcall cookie has unexpected size %zu",
3776 memset(&cookie, 0, sizeof cookie);
3777 memcpy(&cookie, nl_attr_get(upcall->userdata), userdata_len);
3778 if (userdata_len == sizeof cookie.sflow
3779 && cookie.type == USER_ACTION_COOKIE_SFLOW) {
3780 return SFLOW_UPCALL;
3781 } else if (userdata_len == sizeof cookie.slow_path
3782 && cookie.type == USER_ACTION_COOKIE_SLOW_PATH) {
3784 } else if (userdata_len == sizeof cookie.flow_sample
3785 && cookie.type == USER_ACTION_COOKIE_FLOW_SAMPLE) {
3786 return FLOW_SAMPLE_UPCALL;
3787 } else if (userdata_len == sizeof cookie.ipfix
3788 && cookie.type == USER_ACTION_COOKIE_IPFIX) {
3789 return IPFIX_UPCALL;
3791 VLOG_WARN_RL(&rl, "invalid user cookie of type %"PRIu16
3792 " and size %zu", cookie.type, userdata_len);
3798 handle_sflow_upcall(struct dpif_backer *backer,
3799 const struct dpif_upcall *upcall)
3801 struct ofproto_dpif *ofproto;
3802 union user_action_cookie cookie;
3804 uint32_t odp_in_port;
3806 if (ofproto_receive(backer, upcall->packet, upcall->key, upcall->key_len,
3807 &flow, NULL, &ofproto, &odp_in_port)
3808 || !ofproto->sflow) {
3812 memset(&cookie, 0, sizeof cookie);
3813 memcpy(&cookie, nl_attr_get(upcall->userdata), sizeof cookie.sflow);
3814 dpif_sflow_received(ofproto->sflow, upcall->packet, &flow,
3815 odp_in_port, &cookie);
3819 handle_flow_sample_upcall(struct dpif_backer *backer,
3820 const struct dpif_upcall *upcall)
3822 struct ofproto_dpif *ofproto;
3823 union user_action_cookie cookie;
3826 if (ofproto_receive(backer, upcall->packet, upcall->key, upcall->key_len,
3827 &flow, NULL, &ofproto, NULL)
3828 || !ofproto->ipfix) {
3832 memset(&cookie, 0, sizeof cookie);
3833 memcpy(&cookie, nl_attr_get(upcall->userdata), sizeof cookie.flow_sample);
3835 /* The flow reflects exactly the contents of the packet. Sample
3836 * the packet using it. */
3837 dpif_ipfix_flow_sample(ofproto->ipfix, upcall->packet, &flow,
3838 cookie.flow_sample.collector_set_id,
3839 cookie.flow_sample.probability,
3840 cookie.flow_sample.obs_domain_id,
3841 cookie.flow_sample.obs_point_id);
3845 handle_ipfix_upcall(struct dpif_backer *backer,
3846 const struct dpif_upcall *upcall)
3848 struct ofproto_dpif *ofproto;
3851 if (ofproto_receive(backer, upcall->packet, upcall->key, upcall->key_len,
3852 &flow, NULL, &ofproto, NULL)
3853 || !ofproto->ipfix) {
3857 /* The flow reflects exactly the contents of the packet. Sample
3858 * the packet using it. */
3859 dpif_ipfix_bridge_sample(ofproto->ipfix, upcall->packet, &flow);
3863 handle_upcalls(struct dpif_backer *backer, unsigned int max_batch)
3865 struct dpif_upcall misses[FLOW_MISS_MAX_BATCH];
3866 struct ofpbuf miss_bufs[FLOW_MISS_MAX_BATCH];
3867 uint64_t miss_buf_stubs[FLOW_MISS_MAX_BATCH][4096 / 8];
3872 ovs_assert(max_batch <= FLOW_MISS_MAX_BATCH);
3875 for (n_processed = 0; n_processed < max_batch; n_processed++) {
3876 struct dpif_upcall *upcall = &misses[n_misses];
3877 struct ofpbuf *buf = &miss_bufs[n_misses];
3880 ofpbuf_use_stub(buf, miss_buf_stubs[n_misses],
3881 sizeof miss_buf_stubs[n_misses]);
3882 error = dpif_recv(backer->dpif, upcall, buf);
3888 switch (classify_upcall(upcall)) {
3890 /* Handle it later. */
3895 handle_sflow_upcall(backer, upcall);
3899 case FLOW_SAMPLE_UPCALL:
3900 handle_flow_sample_upcall(backer, upcall);
3905 handle_ipfix_upcall(backer, upcall);
3915 /* Handle deferred MISS_UPCALL processing. */
3916 handle_miss_upcalls(backer, misses, n_misses);
3917 for (i = 0; i < n_misses; i++) {
3918 ofpbuf_uninit(&miss_bufs[i]);
3924 /* Flow expiration. */
3926 static int subfacet_max_idle(const struct dpif_backer *);
3927 static void update_stats(struct dpif_backer *);
3928 static void rule_expire(struct rule_dpif *);
3929 static void expire_subfacets(struct dpif_backer *, int dp_max_idle);
3931 /* This function is called periodically by run(). Its job is to collect
3932 * updates for the flows that have been installed into the datapath, most
3933 * importantly when they last were used, and then use that information to
3934 * expire flows that have not been used recently.
3936 * Returns the number of milliseconds after which it should be called again. */
3938 expire(struct dpif_backer *backer)
3940 struct ofproto_dpif *ofproto;
3944 /* Periodically clear out the drop keys in an effort to keep them
3945 * relatively few. */
3946 drop_key_clear(backer);
3948 /* Update stats for each flow in the backer. */
3949 update_stats(backer);
3951 n_subfacets = hmap_count(&backer->subfacets);
3953 struct subfacet *subfacet;
3954 long long int total, now;
3958 HMAP_FOR_EACH (subfacet, hmap_node, &backer->subfacets) {
3959 total += now - subfacet->created;
3961 backer->avg_subfacet_life += total / n_subfacets;
3963 backer->avg_subfacet_life /= 2;
3965 backer->avg_n_subfacet += n_subfacets;
3966 backer->avg_n_subfacet /= 2;
3968 backer->max_n_subfacet = MAX(backer->max_n_subfacet, n_subfacets);
3970 max_idle = subfacet_max_idle(backer);
3971 expire_subfacets(backer, max_idle);
3973 HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
3974 struct rule *rule, *next_rule;
3976 if (ofproto->backer != backer) {
3980 /* Expire OpenFlow flows whose idle_timeout or hard_timeout
3982 LIST_FOR_EACH_SAFE (rule, next_rule, expirable,
3983 &ofproto->up.expirable) {
3984 rule_expire(rule_dpif_cast(rule));
3987 /* All outstanding data in existing flows has been accounted, so it's a
3988 * good time to do bond rebalancing. */
3989 if (ofproto->has_bonded_bundles) {
3990 struct ofbundle *bundle;
3992 HMAP_FOR_EACH (bundle, hmap_node, &ofproto->bundles) {
3994 bond_rebalance(bundle->bond, &backer->revalidate_set);
4000 return MIN(max_idle, 1000);
4003 /* Updates flow table statistics given that the datapath just reported 'stats'
4004 * as 'subfacet''s statistics. */
4006 update_subfacet_stats(struct subfacet *subfacet,
4007 const struct dpif_flow_stats *stats)
4009 struct facet *facet = subfacet->facet;
4010 struct ofproto_dpif *ofproto = ofproto_dpif_cast(facet->rule->up.ofproto);
4011 struct dpif_flow_stats diff;
4013 diff.tcp_flags = stats->tcp_flags;
4014 diff.used = stats->used;
4016 if (stats->n_packets >= subfacet->dp_packet_count) {
4017 diff.n_packets = stats->n_packets - subfacet->dp_packet_count;
4019 VLOG_WARN_RL(&rl, "unexpected packet count from the datapath");
4023 if (stats->n_bytes >= subfacet->dp_byte_count) {
4024 diff.n_bytes = stats->n_bytes - subfacet->dp_byte_count;
4026 VLOG_WARN_RL(&rl, "unexpected byte count from datapath");
4030 ofproto->n_hit += diff.n_packets;
4031 subfacet->dp_packet_count = stats->n_packets;
4032 subfacet->dp_byte_count = stats->n_bytes;
4033 subfacet_update_stats(subfacet, &diff);
4035 if (facet->accounted_bytes < facet->byte_count) {
4037 facet_account(facet);
4038 facet->accounted_bytes = facet->byte_count;
4042 /* 'key' with length 'key_len' bytes is a flow in 'dpif' that we know nothing
4043 * about, or a flow that shouldn't be installed but was anyway. Delete it. */
4045 delete_unexpected_flow(struct dpif_backer *backer,
4046 const struct nlattr *key, size_t key_len)
4048 if (!VLOG_DROP_WARN(&rl)) {
4052 odp_flow_key_format(key, key_len, &s);
4053 VLOG_WARN("unexpected flow: %s", ds_cstr(&s));
4057 COVERAGE_INC(facet_unexpected);
4058 dpif_flow_del(backer->dpif, key, key_len, NULL);
4061 /* Update 'packet_count', 'byte_count', and 'used' members of installed facets.
4063 * This function also pushes statistics updates to rules which each facet
4064 * resubmits into. Generally these statistics will be accurate. However, if a
4065 * facet changes the rule it resubmits into at some time in between
4066 * update_stats() runs, it is possible that statistics accrued to the
4067 * old rule will be incorrectly attributed to the new rule. This could be
4068 * avoided by calling update_stats() whenever rules are created or
4069 * deleted. However, the performance impact of making so many calls to the
4070 * datapath do not justify the benefit of having perfectly accurate statistics.
4072 * In addition, this function maintains per ofproto flow hit counts. The patch
4073 * port is not treated specially. e.g. A packet ingress from br0 patched into
4074 * br1 will increase the hit count of br0 by 1, however, does not affect
4075 * the hit or miss counts of br1.
4078 update_stats(struct dpif_backer *backer)
4080 const struct dpif_flow_stats *stats;
4081 struct dpif_flow_dump dump;
4082 const struct nlattr *key;
4085 dpif_flow_dump_start(&dump, backer->dpif);
4086 while (dpif_flow_dump_next(&dump, &key, &key_len, NULL, NULL, &stats)) {
4087 struct subfacet *subfacet;
4090 key_hash = odp_flow_key_hash(key, key_len);
4091 subfacet = subfacet_find(backer, key, key_len, key_hash);
4092 switch (subfacet ? subfacet->path : SF_NOT_INSTALLED) {
4094 update_subfacet_stats(subfacet, stats);
4098 /* Stats are updated per-packet. */
4101 case SF_NOT_INSTALLED:
4103 delete_unexpected_flow(backer, key, key_len);
4108 dpif_flow_dump_done(&dump);
4110 update_moving_averages(backer);
4113 /* Calculates and returns the number of milliseconds of idle time after which
4114 * subfacets should expire from the datapath. When a subfacet expires, we fold
4115 * its statistics into its facet, and when a facet's last subfacet expires, we
4116 * fold its statistic into its rule. */
4118 subfacet_max_idle(const struct dpif_backer *backer)
4121 * Idle time histogram.
4123 * Most of the time a switch has a relatively small number of subfacets.
4124 * When this is the case we might as well keep statistics for all of them
4125 * in userspace and to cache them in the kernel datapath for performance as
4128 * As the number of subfacets increases, the memory required to maintain
4129 * statistics about them in userspace and in the kernel becomes
4130 * significant. However, with a large number of subfacets it is likely
4131 * that only a few of them are "heavy hitters" that consume a large amount
4132 * of bandwidth. At this point, only heavy hitters are worth caching in
4133 * the kernel and maintaining in userspaces; other subfacets we can
4136 * The technique used to compute the idle time is to build a histogram with
4137 * N_BUCKETS buckets whose width is BUCKET_WIDTH msecs each. Each subfacet
4138 * that is installed in the kernel gets dropped in the appropriate bucket.
4139 * After the histogram has been built, we compute the cutoff so that only
4140 * the most-recently-used 1% of subfacets (but at least
4141 * flow_eviction_threshold flows) are kept cached. At least
4142 * the most-recently-used bucket of subfacets is kept, so actually an
4143 * arbitrary number of subfacets can be kept in any given expiration run
4144 * (though the next run will delete most of those unless they receive
4147 * This requires a second pass through the subfacets, in addition to the
4148 * pass made by update_stats(), because the former function never looks at
4149 * uninstallable subfacets.
4151 enum { BUCKET_WIDTH = ROUND_UP(100, TIME_UPDATE_INTERVAL) };
4152 enum { N_BUCKETS = 5000 / BUCKET_WIDTH };
4153 int buckets[N_BUCKETS] = { 0 };
4154 int total, subtotal, bucket;
4155 struct subfacet *subfacet;
4159 total = hmap_count(&backer->subfacets);
4160 if (total <= flow_eviction_threshold) {
4161 return N_BUCKETS * BUCKET_WIDTH;
4164 /* Build histogram. */
4166 HMAP_FOR_EACH (subfacet, hmap_node, &backer->subfacets) {
4167 long long int idle = now - subfacet->used;
4168 int bucket = (idle <= 0 ? 0
4169 : idle >= BUCKET_WIDTH * N_BUCKETS ? N_BUCKETS - 1
4170 : (unsigned int) idle / BUCKET_WIDTH);
4174 /* Find the first bucket whose flows should be expired. */
4175 subtotal = bucket = 0;
4177 subtotal += buckets[bucket++];
4178 } while (bucket < N_BUCKETS &&
4179 subtotal < MAX(flow_eviction_threshold, total / 100));
4181 if (VLOG_IS_DBG_ENABLED()) {
4185 ds_put_cstr(&s, "keep");
4186 for (i = 0; i < N_BUCKETS; i++) {
4188 ds_put_cstr(&s, ", drop");
4191 ds_put_format(&s, " %d:%d", i * BUCKET_WIDTH, buckets[i]);
4194 VLOG_INFO("%s (msec:count)", ds_cstr(&s));
4198 return bucket * BUCKET_WIDTH;
4202 expire_subfacets(struct dpif_backer *backer, int dp_max_idle)
4204 /* Cutoff time for most flows. */
4205 long long int normal_cutoff = time_msec() - dp_max_idle;
4207 /* We really want to keep flows for special protocols around, so use a more
4208 * conservative cutoff. */
4209 long long int special_cutoff = time_msec() - 10000;
4211 struct subfacet *subfacet, *next_subfacet;
4212 struct subfacet *batch[SUBFACET_DESTROY_MAX_BATCH];
4216 HMAP_FOR_EACH_SAFE (subfacet, next_subfacet, hmap_node,
4217 &backer->subfacets) {
4218 long long int cutoff;
4220 cutoff = (subfacet->facet->xout.slow & (SLOW_CFM | SLOW_BFD | SLOW_LACP
4224 if (subfacet->used < cutoff) {
4225 if (subfacet->path != SF_NOT_INSTALLED) {
4226 batch[n_batch++] = subfacet;
4227 if (n_batch >= SUBFACET_DESTROY_MAX_BATCH) {
4228 subfacet_destroy_batch(backer, batch, n_batch);
4232 subfacet_destroy(subfacet);
4238 subfacet_destroy_batch(backer, batch, n_batch);
4242 /* If 'rule' is an OpenFlow rule, that has expired according to OpenFlow rules,
4243 * then delete it entirely. */
4245 rule_expire(struct rule_dpif *rule)
4247 struct facet *facet, *next_facet;
4251 if (rule->up.pending) {
4252 /* We'll have to expire it later. */
4256 /* Has 'rule' expired? */
4258 if (rule->up.hard_timeout
4259 && now > rule->up.modified + rule->up.hard_timeout * 1000) {
4260 reason = OFPRR_HARD_TIMEOUT;
4261 } else if (rule->up.idle_timeout
4262 && now > rule->up.used + rule->up.idle_timeout * 1000) {
4263 reason = OFPRR_IDLE_TIMEOUT;
4268 COVERAGE_INC(ofproto_dpif_expired);
4270 /* Update stats. (This is a no-op if the rule expired due to an idle
4271 * timeout, because that only happens when the rule has no facets left.) */
4272 LIST_FOR_EACH_SAFE (facet, next_facet, list_node, &rule->facets) {
4273 facet_remove(facet);
4276 /* Get rid of the rule. */
4277 ofproto_rule_expire(&rule->up, reason);
4282 /* Creates and returns a new facet based on 'miss'.
4284 * The caller must already have determined that no facet with an identical
4285 * 'miss->flow' exists in 'miss->ofproto'.
4287 * 'rule' and 'xout' must have been created based on 'miss'.
4289 * 'facet'' statistics are initialized based on 'stats'.
4291 * The facet will initially have no subfacets. The caller should create (at
4292 * least) one subfacet with subfacet_create(). */
4293 static struct facet *
4294 facet_create(const struct flow_miss *miss, struct rule_dpif *rule,
4295 struct xlate_out *xout, struct dpif_flow_stats *stats)
4297 struct ofproto_dpif *ofproto = miss->ofproto;
4298 struct facet *facet;
4301 facet = xzalloc(sizeof *facet);
4302 facet->packet_count = facet->prev_packet_count = stats->n_packets;
4303 facet->byte_count = facet->prev_byte_count = stats->n_bytes;
4304 facet->tcp_flags = stats->tcp_flags;
4305 facet->used = stats->used;
4306 facet->flow = miss->flow;
4307 facet->learn_rl = time_msec() + 500;
4310 list_push_back(&facet->rule->facets, &facet->list_node);
4311 list_init(&facet->subfacets);
4312 netflow_flow_init(&facet->nf_flow);
4313 netflow_flow_update_time(ofproto->netflow, &facet->nf_flow, facet->used);
4315 xlate_out_copy(&facet->xout, xout);
4317 match_init(&match, &facet->flow, &facet->xout.wc);
4318 cls_rule_init(&facet->cr, &match, OFP_DEFAULT_PRIORITY);
4319 classifier_insert(&ofproto->facets, &facet->cr);
4321 facet->nf_flow.output_iface = facet->xout.nf_output_iface;
4327 facet_free(struct facet *facet)
4330 xlate_out_uninit(&facet->xout);
4335 /* Executes, within 'ofproto', the 'n_actions' actions in 'actions' on
4336 * 'packet', which arrived on 'in_port'. */
4338 execute_odp_actions(struct ofproto_dpif *ofproto, const struct flow *flow,
4339 const struct nlattr *odp_actions, size_t actions_len,
4340 struct ofpbuf *packet)
4342 struct odputil_keybuf keybuf;
4346 ofpbuf_use_stack(&key, &keybuf, sizeof keybuf);
4347 odp_flow_key_from_flow(&key, flow,
4348 ofp_port_to_odp_port(ofproto, flow->in_port));
4350 error = dpif_execute(ofproto->backer->dpif, key.data, key.size,
4351 odp_actions, actions_len, packet);
4355 /* Remove 'facet' from its ofproto and free up the associated memory:
4357 * - If 'facet' was installed in the datapath, uninstalls it and updates its
4358 * rule's statistics, via subfacet_uninstall().
4360 * - Removes 'facet' from its rule and from ofproto->facets.
4363 facet_remove(struct facet *facet)
4365 struct ofproto_dpif *ofproto = ofproto_dpif_cast(facet->rule->up.ofproto);
4366 struct subfacet *subfacet, *next_subfacet;
4368 ovs_assert(!list_is_empty(&facet->subfacets));
4370 /* First uninstall all of the subfacets to get final statistics. */
4371 LIST_FOR_EACH (subfacet, list_node, &facet->subfacets) {
4372 subfacet_uninstall(subfacet);
4375 /* Flush the final stats to the rule.
4377 * This might require us to have at least one subfacet around so that we
4378 * can use its actions for accounting in facet_account(), which is why we
4379 * have uninstalled but not yet destroyed the subfacets. */
4380 facet_flush_stats(facet);
4382 /* Now we're really all done so destroy everything. */
4383 LIST_FOR_EACH_SAFE (subfacet, next_subfacet, list_node,
4384 &facet->subfacets) {
4385 subfacet_destroy__(subfacet);
4387 classifier_remove(&ofproto->facets, &facet->cr);
4388 cls_rule_destroy(&facet->cr);
4389 list_remove(&facet->list_node);
4393 /* Feed information from 'facet' back into the learning table to keep it in
4394 * sync with what is actually flowing through the datapath. */
4396 facet_learn(struct facet *facet)
4398 long long int now = time_msec();
4400 if (!facet->xout.has_fin_timeout && now < facet->learn_rl) {
4404 facet->learn_rl = now + 500;
4406 if (!facet->xout.has_learn
4407 && !facet->xout.has_normal
4408 && (!facet->xout.has_fin_timeout
4409 || !(facet->tcp_flags & (TCP_FIN | TCP_RST)))) {
4413 facet_push_stats(facet, true);
4417 facet_account(struct facet *facet)
4419 struct ofproto_dpif *ofproto = ofproto_dpif_cast(facet->rule->up.ofproto);
4420 const struct nlattr *a;
4425 if (!facet->xout.has_normal || !ofproto->has_bonded_bundles) {
4428 n_bytes = facet->byte_count - facet->accounted_bytes;
4430 /* This loop feeds byte counters to bond_account() for rebalancing to use
4431 * as a basis. We also need to track the actual VLAN on which the packet
4432 * is going to be sent to ensure that it matches the one passed to
4433 * bond_choose_output_slave(). (Otherwise, we will account to the wrong
4436 * We use the actions from an arbitrary subfacet because they should all
4437 * be equally valid for our purpose. */
4438 vlan_tci = facet->flow.vlan_tci;
4439 NL_ATTR_FOR_EACH_UNSAFE (a, left, facet->xout.odp_actions.data,
4440 facet->xout.odp_actions.size) {
4441 const struct ovs_action_push_vlan *vlan;
4442 struct ofport_dpif *port;
4444 switch (nl_attr_type(a)) {
4445 case OVS_ACTION_ATTR_OUTPUT:
4446 port = get_odp_port(ofproto, nl_attr_get_u32(a));
4447 if (port && port->bundle && port->bundle->bond) {
4448 bond_account(port->bundle->bond, &facet->flow,
4449 vlan_tci_to_vid(vlan_tci), n_bytes);
4453 case OVS_ACTION_ATTR_POP_VLAN:
4454 vlan_tci = htons(0);
4457 case OVS_ACTION_ATTR_PUSH_VLAN:
4458 vlan = nl_attr_get(a);
4459 vlan_tci = vlan->vlan_tci;
4465 /* Returns true if the only action for 'facet' is to send to the controller.
4466 * (We don't report NetFlow expiration messages for such facets because they
4467 * are just part of the control logic for the network, not real traffic). */
4469 facet_is_controller_flow(struct facet *facet)
4472 const struct rule *rule = &facet->rule->up;
4473 const struct ofpact *ofpacts = rule->ofpacts;
4474 size_t ofpacts_len = rule->ofpacts_len;
4476 if (ofpacts_len > 0 &&
4477 ofpacts->type == OFPACT_CONTROLLER &&
4478 ofpact_next(ofpacts) >= ofpact_end(ofpacts, ofpacts_len)) {
4485 /* Folds all of 'facet''s statistics into its rule. Also updates the
4486 * accounting ofhook and emits a NetFlow expiration if appropriate. All of
4487 * 'facet''s statistics in the datapath should have been zeroed and folded into
4488 * its packet and byte counts before this function is called. */
4490 facet_flush_stats(struct facet *facet)
4492 struct ofproto_dpif *ofproto = ofproto_dpif_cast(facet->rule->up.ofproto);
4493 struct subfacet *subfacet;
4495 LIST_FOR_EACH (subfacet, list_node, &facet->subfacets) {
4496 ovs_assert(!subfacet->dp_byte_count);
4497 ovs_assert(!subfacet->dp_packet_count);
4500 facet_push_stats(facet, false);
4501 if (facet->accounted_bytes < facet->byte_count) {
4502 facet_account(facet);
4503 facet->accounted_bytes = facet->byte_count;
4506 if (ofproto->netflow && !facet_is_controller_flow(facet)) {
4507 struct ofexpired expired;
4508 expired.flow = facet->flow;
4509 expired.packet_count = facet->packet_count;
4510 expired.byte_count = facet->byte_count;
4511 expired.used = facet->used;
4512 netflow_expire(ofproto->netflow, &facet->nf_flow, &expired);
4515 /* Reset counters to prevent double counting if 'facet' ever gets
4517 facet_reset_counters(facet);
4519 netflow_flow_clear(&facet->nf_flow);
4520 facet->tcp_flags = 0;
4523 /* Searches 'ofproto''s table of facets for one which would be responsible for
4524 * 'flow'. Returns it if found, otherwise a null pointer.
4526 * The returned facet might need revalidation; use facet_lookup_valid()
4527 * instead if that is important. */
4528 static struct facet *
4529 facet_find(struct ofproto_dpif *ofproto, const struct flow *flow)
4531 struct cls_rule *cr = classifier_lookup(&ofproto->facets, flow, NULL);
4532 return cr ? CONTAINER_OF(cr, struct facet, cr) : NULL;
4535 /* Searches 'ofproto''s table of facets for one capable that covers
4536 * 'flow'. Returns it if found, otherwise a null pointer.
4538 * The returned facet is guaranteed to be valid. */
4539 static struct facet *
4540 facet_lookup_valid(struct ofproto_dpif *ofproto, const struct flow *flow)
4542 struct facet *facet;
4544 facet = facet_find(ofproto, flow);
4546 && (ofproto->backer->need_revalidate
4547 || tag_set_intersects(&ofproto->backer->revalidate_set,
4549 && !facet_revalidate(facet)) {
4557 facet_check_consistency(struct facet *facet)
4559 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 15);
4561 struct ofproto_dpif *ofproto = ofproto_dpif_cast(facet->rule->up.ofproto);
4563 struct xlate_out xout;
4564 struct xlate_in xin;
4566 struct rule_dpif *rule;
4569 /* Check the rule for consistency. */
4570 rule = rule_dpif_lookup(ofproto, &facet->flow, NULL);
4571 if (rule != facet->rule) {
4572 if (!VLOG_DROP_WARN(&rl)) {
4573 struct ds s = DS_EMPTY_INITIALIZER;
4575 flow_format(&s, &facet->flow);
4576 ds_put_format(&s, ": facet associated with wrong rule (was "
4577 "table=%"PRIu8",", facet->rule->up.table_id);
4578 cls_rule_format(&facet->rule->up.cr, &s);
4579 ds_put_format(&s, ") (should have been table=%"PRIu8",",
4581 cls_rule_format(&rule->up.cr, &s);
4582 ds_put_char(&s, ')');
4584 VLOG_WARN("%s", ds_cstr(&s));
4590 /* Check the datapath actions for consistency. */
4591 xlate_in_init(&xin, ofproto, &facet->flow, rule, 0, NULL);
4592 xlate_actions(&xin, &xout);
4594 ok = ofpbuf_equal(&facet->xout.odp_actions, &xout.odp_actions)
4595 && facet->xout.slow == xout.slow;
4596 if (!ok && !VLOG_DROP_WARN(&rl)) {
4597 struct ds s = DS_EMPTY_INITIALIZER;
4599 flow_format(&s, &facet->flow);
4600 ds_put_cstr(&s, ": inconsistency in facet");
4602 if (!ofpbuf_equal(&facet->xout.odp_actions, &xout.odp_actions)) {
4603 ds_put_cstr(&s, " (actions were: ");
4604 format_odp_actions(&s, facet->xout.odp_actions.data,
4605 facet->xout.odp_actions.size);
4606 ds_put_cstr(&s, ") (correct actions: ");
4607 format_odp_actions(&s, xout.odp_actions.data,
4608 xout.odp_actions.size);
4609 ds_put_char(&s, ')');
4612 if (facet->xout.slow != xout.slow) {
4613 ds_put_format(&s, " slow path incorrect. should be %d", xout.slow);
4616 VLOG_WARN("%s", ds_cstr(&s));
4619 xlate_out_uninit(&xout);
4624 /* Re-searches the classifier for 'facet':
4626 * - If the rule found is different from 'facet''s current rule, moves
4627 * 'facet' to the new rule and recompiles its actions.
4629 * - If the rule found is the same as 'facet''s current rule, leaves 'facet'
4630 * where it is and recompiles its actions anyway.
4632 * - If any of 'facet''s subfacets correspond to a new flow according to
4633 * ofproto_receive(), 'facet' is removed.
4635 * Returns true if 'facet' is still valid. False if 'facet' was removed. */
4637 facet_revalidate(struct facet *facet)
4639 struct ofproto_dpif *ofproto = ofproto_dpif_cast(facet->rule->up.ofproto);
4640 struct rule_dpif *new_rule;
4641 struct subfacet *subfacet;
4642 struct flow_wildcards wc;
4643 struct xlate_out xout;
4644 struct xlate_in xin;
4646 COVERAGE_INC(facet_revalidate);
4648 /* Check that child subfacets still correspond to this facet. Tunnel
4649 * configuration changes could cause a subfacet's OpenFlow in_port to
4651 LIST_FOR_EACH (subfacet, list_node, &facet->subfacets) {
4652 struct ofproto_dpif *recv_ofproto;
4653 struct flow recv_flow;
4656 error = ofproto_receive(ofproto->backer, NULL, subfacet->key,
4657 subfacet->key_len, &recv_flow, NULL,
4658 &recv_ofproto, NULL);
4660 || recv_ofproto != ofproto
4661 || facet != facet_find(ofproto, &recv_flow)) {
4662 facet_remove(facet);
4667 flow_wildcards_init_catchall(&wc);
4668 new_rule = rule_dpif_lookup(ofproto, &facet->flow, &wc);
4670 /* Calculate new datapath actions.
4672 * We do not modify any 'facet' state yet, because we might need to, e.g.,
4673 * emit a NetFlow expiration and, if so, we need to have the old state
4674 * around to properly compose it. */
4675 xlate_in_init(&xin, ofproto, &facet->flow, new_rule, 0, NULL);
4676 xlate_actions(&xin, &xout);
4677 flow_wildcards_or(&xout.wc, &xout.wc, &wc);
4679 /* A facet's slow path reason should only change under dramatic
4680 * circumstances. Rather than try to update everything, it's simpler to
4681 * remove the facet and start over.
4683 * More importantly, if a facet's wildcards change, it will be relatively
4684 * difficult to figure out if its subfacets still belong to it, and if not
4685 * which facet they may belong to. Again, to avoid the complexity, we
4686 * simply give up instead. */
4687 if (facet->xout.slow != xout.slow
4688 || memcmp(&facet->xout.wc, &xout.wc, sizeof xout.wc)) {
4689 facet_remove(facet);
4690 xlate_out_uninit(&xout);
4694 if (!ofpbuf_equal(&facet->xout.odp_actions, &xout.odp_actions)) {
4695 LIST_FOR_EACH(subfacet, list_node, &facet->subfacets) {
4696 if (subfacet->path == SF_FAST_PATH) {
4697 struct dpif_flow_stats stats;
4699 subfacet_install(subfacet, &xout.odp_actions, &stats);
4700 subfacet_update_stats(subfacet, &stats);
4704 facet_flush_stats(facet);
4706 ofpbuf_clear(&facet->xout.odp_actions);
4707 ofpbuf_put(&facet->xout.odp_actions, xout.odp_actions.data,
4708 xout.odp_actions.size);
4711 /* Update 'facet' now that we've taken care of all the old state. */
4712 facet->xout.tags = xout.tags;
4713 facet->xout.slow = xout.slow;
4714 facet->xout.has_learn = xout.has_learn;
4715 facet->xout.has_normal = xout.has_normal;
4716 facet->xout.has_fin_timeout = xout.has_fin_timeout;
4717 facet->xout.nf_output_iface = xout.nf_output_iface;
4718 facet->xout.mirrors = xout.mirrors;
4719 facet->nf_flow.output_iface = facet->xout.nf_output_iface;
4721 if (facet->rule != new_rule) {
4722 COVERAGE_INC(facet_changed_rule);
4723 list_remove(&facet->list_node);
4724 list_push_back(&new_rule->facets, &facet->list_node);
4725 facet->rule = new_rule;
4726 facet->used = new_rule->up.created;
4727 facet->prev_used = facet->used;
4730 xlate_out_uninit(&xout);
4735 facet_reset_counters(struct facet *facet)
4737 facet->packet_count = 0;
4738 facet->byte_count = 0;
4739 facet->prev_packet_count = 0;
4740 facet->prev_byte_count = 0;
4741 facet->accounted_bytes = 0;
4745 facet_push_stats(struct facet *facet, bool may_learn)
4747 struct dpif_flow_stats stats;
4749 ovs_assert(facet->packet_count >= facet->prev_packet_count);
4750 ovs_assert(facet->byte_count >= facet->prev_byte_count);
4751 ovs_assert(facet->used >= facet->prev_used);
4753 stats.n_packets = facet->packet_count - facet->prev_packet_count;
4754 stats.n_bytes = facet->byte_count - facet->prev_byte_count;
4755 stats.used = facet->used;
4756 stats.tcp_flags = facet->tcp_flags;
4758 if (may_learn || stats.n_packets || facet->used > facet->prev_used) {
4759 struct ofproto_dpif *ofproto =
4760 ofproto_dpif_cast(facet->rule->up.ofproto);
4762 struct ofport_dpif *in_port;
4763 struct xlate_in xin;
4765 facet->prev_packet_count = facet->packet_count;
4766 facet->prev_byte_count = facet->byte_count;
4767 facet->prev_used = facet->used;
4769 in_port = get_ofp_port(ofproto, facet->flow.in_port);
4770 if (in_port && in_port->tnl_port) {
4771 netdev_vport_inc_rx(in_port->up.netdev, &stats);
4774 rule_credit_stats(facet->rule, &stats);
4775 netflow_flow_update_time(ofproto->netflow, &facet->nf_flow,
4777 netflow_flow_update_flags(&facet->nf_flow, facet->tcp_flags);
4778 update_mirror_stats(ofproto, facet->xout.mirrors, stats.n_packets,
4781 xlate_in_init(&xin, ofproto, &facet->flow, facet->rule,
4782 stats.tcp_flags, NULL);
4783 xin.resubmit_stats = &stats;
4784 xin.may_learn = may_learn;
4785 xlate_actions_for_side_effects(&xin);
4790 push_all_stats__(bool run_fast)
4792 static long long int rl = LLONG_MIN;
4793 struct ofproto_dpif *ofproto;
4795 if (time_msec() < rl) {
4799 HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
4800 struct cls_cursor cursor;
4801 struct facet *facet;
4803 cls_cursor_init(&cursor, &ofproto->facets, NULL);
4804 CLS_CURSOR_FOR_EACH (facet, cr, &cursor) {
4805 facet_push_stats(facet, false);
4812 rl = time_msec() + 100;
4816 push_all_stats(void)
4818 push_all_stats__(true);
4822 rule_credit_stats(struct rule_dpif *rule, const struct dpif_flow_stats *stats)
4824 rule->packet_count += stats->n_packets;
4825 rule->byte_count += stats->n_bytes;
4826 ofproto_rule_update_used(&rule->up, stats->used);
4831 static struct subfacet *
4832 subfacet_find(struct dpif_backer *backer, const struct nlattr *key,
4833 size_t key_len, uint32_t key_hash)
4835 struct subfacet *subfacet;
4837 HMAP_FOR_EACH_WITH_HASH (subfacet, hmap_node, key_hash,
4838 &backer->subfacets) {
4839 if (subfacet->key_len == key_len
4840 && !memcmp(key, subfacet->key, key_len)) {
4848 /* Searches 'facet' (within 'ofproto') for a subfacet with the specified
4849 * 'key_fitness', 'key', and 'key_len' members in 'miss'. Returns the
4850 * existing subfacet if there is one, otherwise creates and returns a
4852 static struct subfacet *
4853 subfacet_create(struct facet *facet, struct flow_miss *miss,
4856 struct dpif_backer *backer = miss->ofproto->backer;
4857 enum odp_key_fitness key_fitness = miss->key_fitness;
4858 const struct nlattr *key = miss->key;
4859 size_t key_len = miss->key_len;
4861 struct subfacet *subfacet;
4863 key_hash = odp_flow_key_hash(key, key_len);
4865 if (list_is_empty(&facet->subfacets)) {
4866 subfacet = &facet->one_subfacet;
4868 subfacet = subfacet_find(backer, key, key_len, key_hash);
4870 if (subfacet->facet == facet) {
4874 /* This shouldn't happen. */
4875 VLOG_ERR_RL(&rl, "subfacet with wrong facet");
4876 subfacet_destroy(subfacet);
4879 subfacet = xmalloc(sizeof *subfacet);
4882 hmap_insert(&backer->subfacets, &subfacet->hmap_node, key_hash);
4883 list_push_back(&facet->subfacets, &subfacet->list_node);
4884 subfacet->facet = facet;
4885 subfacet->key_fitness = key_fitness;
4886 subfacet->key = xmemdup(key, key_len);
4887 subfacet->key_len = key_len;
4888 subfacet->used = now;
4889 subfacet->created = now;
4890 subfacet->dp_packet_count = 0;
4891 subfacet->dp_byte_count = 0;
4892 subfacet->path = SF_NOT_INSTALLED;
4893 subfacet->backer = backer;
4895 backer->subfacet_add_count++;
4899 /* Uninstalls 'subfacet' from the datapath, if it is installed, removes it from
4900 * its facet within 'ofproto', and frees it. */
4902 subfacet_destroy__(struct subfacet *subfacet)
4904 struct facet *facet = subfacet->facet;
4905 struct ofproto_dpif *ofproto = ofproto_dpif_cast(facet->rule->up.ofproto);
4907 /* Update ofproto stats before uninstall the subfacet. */
4908 ofproto->backer->subfacet_del_count++;
4910 subfacet_uninstall(subfacet);
4911 hmap_remove(&subfacet->backer->subfacets, &subfacet->hmap_node);
4912 list_remove(&subfacet->list_node);
4913 free(subfacet->key);
4914 if (subfacet != &facet->one_subfacet) {
4919 /* Destroys 'subfacet', as with subfacet_destroy__(), and then if this was the
4920 * last remaining subfacet in its facet destroys the facet too. */
4922 subfacet_destroy(struct subfacet *subfacet)
4924 struct facet *facet = subfacet->facet;
4926 if (list_is_singleton(&facet->subfacets)) {
4927 /* facet_remove() needs at least one subfacet (it will remove it). */
4928 facet_remove(facet);
4930 subfacet_destroy__(subfacet);
4935 subfacet_destroy_batch(struct dpif_backer *backer,
4936 struct subfacet **subfacets, int n)
4938 struct dpif_op ops[SUBFACET_DESTROY_MAX_BATCH];
4939 struct dpif_op *opsp[SUBFACET_DESTROY_MAX_BATCH];
4940 struct dpif_flow_stats stats[SUBFACET_DESTROY_MAX_BATCH];
4943 for (i = 0; i < n; i++) {
4944 ops[i].type = DPIF_OP_FLOW_DEL;
4945 ops[i].u.flow_del.key = subfacets[i]->key;
4946 ops[i].u.flow_del.key_len = subfacets[i]->key_len;
4947 ops[i].u.flow_del.stats = &stats[i];
4951 dpif_operate(backer->dpif, opsp, n);
4952 for (i = 0; i < n; i++) {
4953 subfacet_reset_dp_stats(subfacets[i], &stats[i]);
4954 subfacets[i]->path = SF_NOT_INSTALLED;
4955 subfacet_destroy(subfacets[i]);
4960 /* Updates 'subfacet''s datapath flow, setting its actions to 'actions_len'
4961 * bytes of actions in 'actions'. If 'stats' is non-null, statistics counters
4962 * in the datapath will be zeroed and 'stats' will be updated with traffic new
4963 * since 'subfacet' was last updated.
4965 * Returns 0 if successful, otherwise a positive errno value. */
4967 subfacet_install(struct subfacet *subfacet, const struct ofpbuf *odp_actions,
4968 struct dpif_flow_stats *stats)
4970 struct facet *facet = subfacet->facet;
4971 struct ofproto_dpif *ofproto = ofproto_dpif_cast(facet->rule->up.ofproto);
4972 enum subfacet_path path = facet->xout.slow ? SF_SLOW_PATH : SF_FAST_PATH;
4973 const struct nlattr *actions = odp_actions->data;
4974 size_t actions_len = odp_actions->size;
4976 uint64_t slow_path_stub[128 / 8];
4977 enum dpif_flow_put_flags flags;
4980 flags = DPIF_FP_CREATE | DPIF_FP_MODIFY;
4982 flags |= DPIF_FP_ZERO_STATS;
4985 if (path == SF_SLOW_PATH) {
4986 compose_slow_path(ofproto, &facet->flow, facet->xout.slow,
4987 slow_path_stub, sizeof slow_path_stub,
4988 &actions, &actions_len);
4991 ret = dpif_flow_put(subfacet->backer->dpif, flags, subfacet->key,
4992 subfacet->key_len, actions, actions_len, stats);
4995 subfacet_reset_dp_stats(subfacet, stats);
4999 subfacet->path = path;
5004 /* If 'subfacet' is installed in the datapath, uninstalls it. */
5006 subfacet_uninstall(struct subfacet *subfacet)
5008 if (subfacet->path != SF_NOT_INSTALLED) {
5009 struct rule_dpif *rule = subfacet->facet->rule;
5010 struct ofproto_dpif *ofproto = ofproto_dpif_cast(rule->up.ofproto);
5011 struct dpif_flow_stats stats;
5014 error = dpif_flow_del(ofproto->backer->dpif, subfacet->key,
5015 subfacet->key_len, &stats);
5016 subfacet_reset_dp_stats(subfacet, &stats);
5018 subfacet_update_stats(subfacet, &stats);
5020 subfacet->path = SF_NOT_INSTALLED;
5022 ovs_assert(subfacet->dp_packet_count == 0);
5023 ovs_assert(subfacet->dp_byte_count == 0);
5027 /* Resets 'subfacet''s datapath statistics counters. This should be called
5028 * when 'subfacet''s statistics are cleared in the datapath. If 'stats' is
5029 * non-null, it should contain the statistics returned by dpif when 'subfacet'
5030 * was reset in the datapath. 'stats' will be modified to include only
5031 * statistics new since 'subfacet' was last updated. */
5033 subfacet_reset_dp_stats(struct subfacet *subfacet,
5034 struct dpif_flow_stats *stats)
5037 && subfacet->dp_packet_count <= stats->n_packets
5038 && subfacet->dp_byte_count <= stats->n_bytes) {
5039 stats->n_packets -= subfacet->dp_packet_count;
5040 stats->n_bytes -= subfacet->dp_byte_count;
5043 subfacet->dp_packet_count = 0;
5044 subfacet->dp_byte_count = 0;
5047 /* Folds the statistics from 'stats' into the counters in 'subfacet'.
5049 * Because of the meaning of a subfacet's counters, it only makes sense to do
5050 * this if 'stats' are not tracked in the datapath, that is, if 'stats'
5051 * represents a packet that was sent by hand or if it represents statistics
5052 * that have been cleared out of the datapath. */
5054 subfacet_update_stats(struct subfacet *subfacet,
5055 const struct dpif_flow_stats *stats)
5057 if (stats->n_packets || stats->used > subfacet->used) {
5058 struct facet *facet = subfacet->facet;
5060 subfacet->used = MAX(subfacet->used, stats->used);
5061 facet->used = MAX(facet->used, stats->used);
5062 facet->packet_count += stats->n_packets;
5063 facet->byte_count += stats->n_bytes;
5064 facet->tcp_flags |= stats->tcp_flags;
5070 /* Lookup 'flow' in 'ofproto''s classifier. If 'wc' is non-null, sets
5071 * the fields that were relevant as part of the lookup. */
5072 static struct rule_dpif *
5073 rule_dpif_lookup(struct ofproto_dpif *ofproto, const struct flow *flow,
5074 struct flow_wildcards *wc)
5076 struct rule_dpif *rule;
5078 rule = rule_dpif_lookup_in_table(ofproto, flow, wc, 0);
5083 return rule_dpif_miss_rule(ofproto, flow);
5087 rule_dpif_lookup_in_table(struct ofproto_dpif *ofproto,
5088 const struct flow *flow, struct flow_wildcards *wc,
5091 struct cls_rule *cls_rule;
5092 struct classifier *cls;
5095 if (table_id >= N_TABLES) {
5099 cls = &ofproto->up.tables[table_id].cls;
5100 frag = (flow->nw_frag & FLOW_NW_FRAG_ANY) != 0;
5101 if (frag && ofproto->up.frag_handling == OFPC_FRAG_NORMAL) {
5102 /* We must pretend that transport ports are unavailable. */
5103 struct flow ofpc_normal_flow = *flow;
5104 ofpc_normal_flow.tp_src = htons(0);
5105 ofpc_normal_flow.tp_dst = htons(0);
5106 cls_rule = classifier_lookup(cls, &ofpc_normal_flow, wc);
5107 } else if (frag && ofproto->up.frag_handling == OFPC_FRAG_DROP) {
5108 cls_rule = &ofproto->drop_frags_rule->up.cr;
5110 flow_wildcards_init_exact(wc);
5113 cls_rule = classifier_lookup(cls, flow, wc);
5115 return rule_dpif_cast(rule_from_cls_rule(cls_rule));
5119 rule_dpif_miss_rule(struct ofproto_dpif *ofproto, const struct flow *flow)
5121 struct ofport_dpif *port;
5123 port = get_ofp_port(ofproto, flow->in_port);
5125 VLOG_WARN_RL(&rl, "packet-in on unknown port %"PRIu16, flow->in_port);
5126 return ofproto->miss_rule;
5129 if (port->up.pp.config & OFPUTIL_PC_NO_PACKET_IN) {
5130 return ofproto->no_packet_in_rule;
5132 return ofproto->miss_rule;
5136 complete_operation(struct rule_dpif *rule)
5138 struct ofproto_dpif *ofproto = ofproto_dpif_cast(rule->up.ofproto);
5140 rule_invalidate(rule);
5142 struct dpif_completion *c = xmalloc(sizeof *c);
5143 c->op = rule->up.pending;
5144 list_push_back(&ofproto->completions, &c->list_node);
5146 ofoperation_complete(rule->up.pending, 0);
5150 static struct rule *
5153 struct rule_dpif *rule = xmalloc(sizeof *rule);
5158 rule_dealloc(struct rule *rule_)
5160 struct rule_dpif *rule = rule_dpif_cast(rule_);
5165 rule_construct(struct rule *rule_)
5167 struct rule_dpif *rule = rule_dpif_cast(rule_);
5168 struct ofproto_dpif *ofproto = ofproto_dpif_cast(rule->up.ofproto);
5169 struct rule_dpif *victim;
5172 rule->packet_count = 0;
5173 rule->byte_count = 0;
5175 victim = rule_dpif_cast(ofoperation_get_victim(rule->up.pending));
5176 if (victim && !list_is_empty(&victim->facets)) {
5177 struct facet *facet;
5179 rule->facets = victim->facets;
5180 list_moved(&rule->facets);
5181 LIST_FOR_EACH (facet, list_node, &rule->facets) {
5182 /* XXX: We're only clearing our local counters here. It's possible
5183 * that quite a few packets are unaccounted for in the datapath
5184 * statistics. These will be accounted to the new rule instead of
5185 * cleared as required. This could be fixed by clearing out the
5186 * datapath statistics for this facet, but currently it doesn't
5188 facet_reset_counters(facet);
5192 /* Must avoid list_moved() in this case. */
5193 list_init(&rule->facets);
5196 table_id = rule->up.table_id;
5198 rule->tag = victim->tag;
5199 } else if (table_id == 0) {
5204 miniflow_expand(&rule->up.cr.match.flow, &flow);
5205 rule->tag = rule_calculate_tag(&flow, &rule->up.cr.match.mask,
5206 ofproto->tables[table_id].basis);
5209 complete_operation(rule);
5214 rule_destruct(struct rule *rule_)
5216 struct rule_dpif *rule = rule_dpif_cast(rule_);
5217 struct facet *facet, *next_facet;
5219 LIST_FOR_EACH_SAFE (facet, next_facet, list_node, &rule->facets) {
5220 facet_revalidate(facet);
5223 complete_operation(rule);
5227 rule_get_stats(struct rule *rule_, uint64_t *packets, uint64_t *bytes)
5229 struct rule_dpif *rule = rule_dpif_cast(rule_);
5231 /* push_all_stats() can handle flow misses which, when using the learn
5232 * action, can cause rules to be added and deleted. This can corrupt our
5233 * caller's datastructures which assume that rule_get_stats() doesn't have
5234 * an impact on the flow table. To be safe, we disable miss handling. */
5235 push_all_stats__(false);
5237 /* Start from historical data for 'rule' itself that are no longer tracked
5238 * in facets. This counts, for example, facets that have expired. */
5239 *packets = rule->packet_count;
5240 *bytes = rule->byte_count;
5244 rule_dpif_execute(struct rule_dpif *rule, const struct flow *flow,
5245 struct ofpbuf *packet)
5247 struct ofproto_dpif *ofproto = ofproto_dpif_cast(rule->up.ofproto);
5248 struct dpif_flow_stats stats;
5249 struct xlate_out xout;
5250 struct xlate_in xin;
5252 dpif_flow_stats_extract(flow, packet, time_msec(), &stats);
5253 rule_credit_stats(rule, &stats);
5255 xlate_in_init(&xin, ofproto, flow, rule, stats.tcp_flags, packet);
5256 xin.resubmit_stats = &stats;
5257 xlate_actions(&xin, &xout);
5259 execute_odp_actions(ofproto, flow, xout.odp_actions.data,
5260 xout.odp_actions.size, packet);
5262 xlate_out_uninit(&xout);
5266 rule_execute(struct rule *rule, const struct flow *flow,
5267 struct ofpbuf *packet)
5269 rule_dpif_execute(rule_dpif_cast(rule), flow, packet);
5270 ofpbuf_delete(packet);
5275 rule_modify_actions(struct rule *rule_)
5277 struct rule_dpif *rule = rule_dpif_cast(rule_);
5279 complete_operation(rule);
5282 /* Sends 'packet' out 'ofport'.
5283 * May modify 'packet'.
5284 * Returns 0 if successful, otherwise a positive errno value. */
5286 send_packet(const struct ofport_dpif *ofport, struct ofpbuf *packet)
5288 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport->up.ofproto);
5289 uint64_t odp_actions_stub[1024 / 8];
5290 struct ofpbuf key, odp_actions;
5291 struct dpif_flow_stats stats;
5292 struct odputil_keybuf keybuf;
5293 struct ofpact_output output;
5294 struct xlate_out xout;
5295 struct xlate_in xin;
5299 ofpbuf_use_stub(&odp_actions, odp_actions_stub, sizeof odp_actions_stub);
5300 ofpbuf_use_stack(&key, &keybuf, sizeof keybuf);
5302 /* Use OFPP_NONE as the in_port to avoid special packet processing. */
5303 flow_extract(packet, 0, 0, NULL, OFPP_NONE, &flow);
5304 odp_flow_key_from_flow(&key, &flow, ofp_port_to_odp_port(ofproto,
5306 dpif_flow_stats_extract(&flow, packet, time_msec(), &stats);
5308 ofpact_init(&output.ofpact, OFPACT_OUTPUT, sizeof output);
5309 output.port = ofport->up.ofp_port;
5312 xlate_in_init(&xin, ofproto, &flow, NULL, 0, packet);
5313 xin.ofpacts_len = sizeof output;
5314 xin.ofpacts = &output.ofpact;
5315 xin.resubmit_stats = &stats;
5316 xlate_actions(&xin, &xout);
5318 error = dpif_execute(ofproto->backer->dpif,
5320 xout.odp_actions.data, xout.odp_actions.size,
5322 xlate_out_uninit(&xout);
5325 VLOG_WARN_RL(&rl, "%s: failed to send packet on port %s (%s)",
5326 ofproto->up.name, netdev_get_name(ofport->up.netdev),
5330 ofproto->stats.tx_packets++;
5331 ofproto->stats.tx_bytes += packet->size;
5335 /* Composes an ODP action for a "slow path" action for 'flow' within 'ofproto'.
5336 * The action will state 'slow' as the reason that the action is in the slow
5337 * path. (This is purely informational: it allows a human viewing "ovs-dpctl
5338 * dump-flows" output to see why a flow is in the slow path.)
5340 * The 'stub_size' bytes in 'stub' will be used to store the action.
5341 * 'stub_size' must be large enough for the action.
5343 * The action and its size will be stored in '*actionsp' and '*actions_lenp',
5346 compose_slow_path(const struct ofproto_dpif *ofproto, const struct flow *flow,
5347 enum slow_path_reason slow,
5348 uint64_t *stub, size_t stub_size,
5349 const struct nlattr **actionsp, size_t *actions_lenp)
5351 union user_action_cookie cookie;
5354 cookie.type = USER_ACTION_COOKIE_SLOW_PATH;
5355 cookie.slow_path.unused = 0;
5356 cookie.slow_path.reason = slow;
5358 ofpbuf_use_stack(&buf, stub, stub_size);
5359 if (slow & (SLOW_CFM | SLOW_BFD | SLOW_LACP | SLOW_STP)) {
5360 uint32_t pid = dpif_port_get_pid(ofproto->backer->dpif, UINT32_MAX);
5361 odp_put_userspace_action(pid, &cookie, sizeof cookie.slow_path, &buf);
5363 put_userspace_action(ofproto, &buf, flow, &cookie,
5364 sizeof cookie.slow_path);
5366 *actionsp = buf.data;
5367 *actions_lenp = buf.size;
5371 put_userspace_action(const struct ofproto_dpif *ofproto,
5372 struct ofpbuf *odp_actions,
5373 const struct flow *flow,
5374 const union user_action_cookie *cookie,
5375 const size_t cookie_size)
5379 pid = dpif_port_get_pid(ofproto->backer->dpif,
5380 ofp_port_to_odp_port(ofproto, flow->in_port));
5382 return odp_put_userspace_action(pid, cookie, cookie_size, odp_actions);
5387 update_mirror_stats(struct ofproto_dpif *ofproto, mirror_mask_t mirrors,
5388 uint64_t packets, uint64_t bytes)
5394 for (; mirrors; mirrors = zero_rightmost_1bit(mirrors)) {
5397 m = ofproto->mirrors[mirror_mask_ffs(mirrors) - 1];
5400 /* In normal circumstances 'm' will not be NULL. However,
5401 * if mirrors are reconfigured, we can temporarily get out
5402 * of sync in facet_revalidate(). We could "correct" the
5403 * mirror list before reaching here, but doing that would
5404 * not properly account the traffic stats we've currently
5405 * accumulated for previous mirror configuration. */
5409 m->packet_count += packets;
5410 m->byte_count += bytes;
5415 /* Optimized flow revalidation.
5417 * It's a difficult problem, in general, to tell which facets need to have
5418 * their actions recalculated whenever the OpenFlow flow table changes. We
5419 * don't try to solve that general problem: for most kinds of OpenFlow flow
5420 * table changes, we recalculate the actions for every facet. This is
5421 * relatively expensive, but it's good enough if the OpenFlow flow table
5422 * doesn't change very often.
5424 * However, we can expect one particular kind of OpenFlow flow table change to
5425 * happen frequently: changes caused by MAC learning. To avoid wasting a lot
5426 * of CPU on revalidating every facet whenever MAC learning modifies the flow
5427 * table, we add a special case that applies to flow tables in which every rule
5428 * has the same form (that is, the same wildcards), except that the table is
5429 * also allowed to have a single "catch-all" flow that matches all packets. We
5430 * optimize this case by tagging all of the facets that resubmit into the table
5431 * and invalidating the same tag whenever a flow changes in that table. The
5432 * end result is that we revalidate just the facets that need it (and sometimes
5433 * a few more, but not all of the facets or even all of the facets that
5434 * resubmit to the table modified by MAC learning). */
5436 /* Calculates the tag to use for 'flow' and mask 'mask' when it is inserted
5437 * into an OpenFlow table with the given 'basis'. */
5439 rule_calculate_tag(const struct flow *flow, const struct minimask *mask,
5442 if (minimask_is_catchall(mask)) {
5445 uint32_t hash = flow_hash_in_minimask(flow, mask, secret);
5446 return tag_create_deterministic(hash);
5450 /* Following a change to OpenFlow table 'table_id' in 'ofproto', update the
5451 * taggability of that table.
5453 * This function must be called after *each* change to a flow table. If you
5454 * skip calling it on some changes then the pointer comparisons at the end can
5455 * be invalid if you get unlucky. For example, if a flow removal causes a
5456 * cls_table to be destroyed and then a flow insertion causes a cls_table with
5457 * different wildcards to be created with the same address, then this function
5458 * will incorrectly skip revalidation. */
5460 table_update_taggable(struct ofproto_dpif *ofproto, uint8_t table_id)
5462 struct table_dpif *table = &ofproto->tables[table_id];
5463 const struct oftable *oftable = &ofproto->up.tables[table_id];
5464 struct cls_table *catchall, *other;
5465 struct cls_table *t;
5467 catchall = other = NULL;
5469 switch (hmap_count(&oftable->cls.tables)) {
5471 /* We could tag this OpenFlow table but it would make the logic a
5472 * little harder and it's a corner case that doesn't seem worth it
5478 HMAP_FOR_EACH (t, hmap_node, &oftable->cls.tables) {
5479 if (cls_table_is_catchall(t)) {
5481 } else if (!other) {
5484 /* Indicate that we can't tag this by setting both tables to
5485 * NULL. (We know that 'catchall' is already NULL.) */
5492 /* Can't tag this table. */
5496 if (table->catchall_table != catchall || table->other_table != other) {
5497 table->catchall_table = catchall;
5498 table->other_table = other;
5499 ofproto->backer->need_revalidate = REV_FLOW_TABLE;
5503 /* Given 'rule' that has changed in some way (either it is a rule being
5504 * inserted, a rule being deleted, or a rule whose actions are being
5505 * modified), marks facets for revalidation to ensure that packets will be
5506 * forwarded correctly according to the new state of the flow table.
5508 * This function must be called after *each* change to a flow table. See
5509 * the comment on table_update_taggable() for more information. */
5511 rule_invalidate(const struct rule_dpif *rule)
5513 struct ofproto_dpif *ofproto = ofproto_dpif_cast(rule->up.ofproto);
5515 table_update_taggable(ofproto, rule->up.table_id);
5517 if (!ofproto->backer->need_revalidate) {
5518 struct table_dpif *table = &ofproto->tables[rule->up.table_id];
5520 if (table->other_table && rule->tag) {
5521 tag_set_add(&ofproto->backer->revalidate_set, rule->tag);
5523 ofproto->backer->need_revalidate = REV_FLOW_TABLE;
5529 set_frag_handling(struct ofproto *ofproto_,
5530 enum ofp_config_flags frag_handling)
5532 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
5533 if (frag_handling != OFPC_FRAG_REASM) {
5534 ofproto->backer->need_revalidate = REV_RECONFIGURE;
5542 packet_out(struct ofproto *ofproto_, struct ofpbuf *packet,
5543 const struct flow *flow,
5544 const struct ofpact *ofpacts, size_t ofpacts_len)
5546 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
5547 struct odputil_keybuf keybuf;
5548 struct dpif_flow_stats stats;
5549 struct xlate_out xout;
5550 struct xlate_in xin;
5554 ofpbuf_use_stack(&key, &keybuf, sizeof keybuf);
5555 odp_flow_key_from_flow(&key, flow,
5556 ofp_port_to_odp_port(ofproto, flow->in_port));
5558 dpif_flow_stats_extract(flow, packet, time_msec(), &stats);
5560 xlate_in_init(&xin, ofproto, flow, NULL, stats.tcp_flags, packet);
5561 xin.resubmit_stats = &stats;
5562 xin.ofpacts_len = ofpacts_len;
5563 xin.ofpacts = ofpacts;
5565 xlate_actions(&xin, &xout);
5566 dpif_execute(ofproto->backer->dpif, key.data, key.size,
5567 xout.odp_actions.data, xout.odp_actions.size, packet);
5568 xlate_out_uninit(&xout);
5576 set_netflow(struct ofproto *ofproto_,
5577 const struct netflow_options *netflow_options)
5579 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
5581 if (netflow_options) {
5582 if (!ofproto->netflow) {
5583 ofproto->netflow = netflow_create();
5585 return netflow_set_options(ofproto->netflow, netflow_options);
5587 netflow_destroy(ofproto->netflow);
5588 ofproto->netflow = NULL;
5594 get_netflow_ids(const struct ofproto *ofproto_,
5595 uint8_t *engine_type, uint8_t *engine_id)
5597 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
5599 dpif_get_netflow_ids(ofproto->backer->dpif, engine_type, engine_id);
5603 send_active_timeout(struct ofproto_dpif *ofproto, struct facet *facet)
5605 if (!facet_is_controller_flow(facet) &&
5606 netflow_active_timeout_expired(ofproto->netflow, &facet->nf_flow)) {
5607 struct subfacet *subfacet;
5608 struct ofexpired expired;
5610 LIST_FOR_EACH (subfacet, list_node, &facet->subfacets) {
5611 if (subfacet->path == SF_FAST_PATH) {
5612 struct dpif_flow_stats stats;
5614 subfacet_install(subfacet, &facet->xout.odp_actions,
5616 subfacet_update_stats(subfacet, &stats);
5620 expired.flow = facet->flow;
5621 expired.packet_count = facet->packet_count;
5622 expired.byte_count = facet->byte_count;
5623 expired.used = facet->used;
5624 netflow_expire(ofproto->netflow, &facet->nf_flow, &expired);
5629 send_netflow_active_timeouts(struct ofproto_dpif *ofproto)
5631 struct cls_cursor cursor;
5632 struct facet *facet;
5634 cls_cursor_init(&cursor, &ofproto->facets, NULL);
5635 CLS_CURSOR_FOR_EACH (facet, cr, &cursor) {
5636 send_active_timeout(ofproto, facet);
5640 static struct ofproto_dpif *
5641 ofproto_dpif_lookup(const char *name)
5643 struct ofproto_dpif *ofproto;
5645 HMAP_FOR_EACH_WITH_HASH (ofproto, all_ofproto_dpifs_node,
5646 hash_string(name, 0), &all_ofproto_dpifs) {
5647 if (!strcmp(ofproto->up.name, name)) {
5655 ofproto_unixctl_fdb_flush(struct unixctl_conn *conn, int argc,
5656 const char *argv[], void *aux OVS_UNUSED)
5658 struct ofproto_dpif *ofproto;
5661 ofproto = ofproto_dpif_lookup(argv[1]);
5663 unixctl_command_reply_error(conn, "no such bridge");
5666 mac_learning_flush(ofproto->ml, &ofproto->backer->revalidate_set);
5668 HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
5669 mac_learning_flush(ofproto->ml, &ofproto->backer->revalidate_set);
5673 unixctl_command_reply(conn, "table successfully flushed");
5677 ofproto_unixctl_fdb_show(struct unixctl_conn *conn, int argc OVS_UNUSED,
5678 const char *argv[], void *aux OVS_UNUSED)
5680 struct ds ds = DS_EMPTY_INITIALIZER;
5681 const struct ofproto_dpif *ofproto;
5682 const struct mac_entry *e;
5684 ofproto = ofproto_dpif_lookup(argv[1]);
5686 unixctl_command_reply_error(conn, "no such bridge");
5690 ds_put_cstr(&ds, " port VLAN MAC Age\n");
5691 LIST_FOR_EACH (e, lru_node, &ofproto->ml->lrus) {
5692 struct ofbundle *bundle = e->port.p;
5693 ds_put_format(&ds, "%5d %4d "ETH_ADDR_FMT" %3d\n",
5694 ofbundle_get_a_port(bundle)->odp_port,
5695 e->vlan, ETH_ADDR_ARGS(e->mac),
5696 mac_entry_age(ofproto->ml, e));
5698 unixctl_command_reply(conn, ds_cstr(&ds));
5703 struct xlate_out xout;
5704 struct xlate_in xin;
5710 trace_format_rule(struct ds *result, int level, const struct rule_dpif *rule)
5712 ds_put_char_multiple(result, '\t', level);
5714 ds_put_cstr(result, "No match\n");
5718 ds_put_format(result, "Rule: table=%"PRIu8" cookie=%#"PRIx64" ",
5719 rule ? rule->up.table_id : 0, ntohll(rule->up.flow_cookie));
5720 cls_rule_format(&rule->up.cr, result);
5721 ds_put_char(result, '\n');
5723 ds_put_char_multiple(result, '\t', level);
5724 ds_put_cstr(result, "OpenFlow ");
5725 ofpacts_format(rule->up.ofpacts, rule->up.ofpacts_len, result);
5726 ds_put_char(result, '\n');
5730 trace_format_flow(struct ds *result, int level, const char *title,
5731 struct trace_ctx *trace)
5733 ds_put_char_multiple(result, '\t', level);
5734 ds_put_format(result, "%s: ", title);
5735 if (flow_equal(&trace->xin.flow, &trace->flow)) {
5736 ds_put_cstr(result, "unchanged");
5738 flow_format(result, &trace->xin.flow);
5739 trace->flow = trace->xin.flow;
5741 ds_put_char(result, '\n');
5745 trace_format_regs(struct ds *result, int level, const char *title,
5746 struct trace_ctx *trace)
5750 ds_put_char_multiple(result, '\t', level);
5751 ds_put_format(result, "%s:", title);
5752 for (i = 0; i < FLOW_N_REGS; i++) {
5753 ds_put_format(result, " reg%zu=0x%"PRIx32, i, trace->flow.regs[i]);
5755 ds_put_char(result, '\n');
5759 trace_format_odp(struct ds *result, int level, const char *title,
5760 struct trace_ctx *trace)
5762 struct ofpbuf *odp_actions = &trace->xout.odp_actions;
5764 ds_put_char_multiple(result, '\t', level);
5765 ds_put_format(result, "%s: ", title);
5766 format_odp_actions(result, odp_actions->data, odp_actions->size);
5767 ds_put_char(result, '\n');
5771 trace_resubmit(struct xlate_in *xin, struct rule_dpif *rule, int recurse)
5773 struct trace_ctx *trace = CONTAINER_OF(xin, struct trace_ctx, xin);
5774 struct ds *result = trace->result;
5776 ds_put_char(result, '\n');
5777 trace_format_flow(result, recurse + 1, "Resubmitted flow", trace);
5778 trace_format_regs(result, recurse + 1, "Resubmitted regs", trace);
5779 trace_format_odp(result, recurse + 1, "Resubmitted odp", trace);
5780 trace_format_rule(result, recurse + 1, rule);
5784 trace_report(struct xlate_in *xin, const char *s, int recurse)
5786 struct trace_ctx *trace = CONTAINER_OF(xin, struct trace_ctx, xin);
5787 struct ds *result = trace->result;
5789 ds_put_char_multiple(result, '\t', recurse);
5790 ds_put_cstr(result, s);
5791 ds_put_char(result, '\n');
5795 ofproto_unixctl_trace(struct unixctl_conn *conn, int argc, const char *argv[],
5796 void *aux OVS_UNUSED)
5798 const struct dpif_backer *backer;
5799 struct ofproto_dpif *ofproto;
5800 struct ofpbuf odp_key;
5801 struct ofpbuf *packet;
5809 ofpbuf_init(&odp_key, 0);
5811 /* Handle "-generate" or a hex string as the last argument. */
5812 if (!strcmp(argv[argc - 1], "-generate")) {
5813 packet = ofpbuf_new(0);
5816 const char *error = eth_from_hex(argv[argc - 1], &packet);
5819 } else if (argc == 4) {
5820 /* The 3-argument form must end in "-generate' or a hex string. */
5821 unixctl_command_reply_error(conn, error);
5826 /* Parse the flow and determine whether a datapath or
5827 * bridge is specified. If function odp_flow_key_from_string()
5828 * returns 0, the flow is a odp_flow. If function
5829 * parse_ofp_exact_flow() returns 0, the flow is a br_flow. */
5830 if (!odp_flow_key_from_string(argv[argc - 1], NULL, &odp_key)) {
5831 /* If the odp_flow is the second argument,
5832 * the datapath name is the first argument. */
5834 const char *dp_type;
5835 if (!strncmp(argv[1], "ovs-", 4)) {
5836 dp_type = argv[1] + 4;
5840 backer = shash_find_data(&all_dpif_backers, dp_type);
5842 unixctl_command_reply_error(conn, "Cannot find datapath "
5847 /* No datapath name specified, so there should be only one
5849 struct shash_node *node;
5850 if (shash_count(&all_dpif_backers) != 1) {
5851 unixctl_command_reply_error(conn, "Must specify datapath "
5852 "name, there is more than one type of datapath");
5855 node = shash_first(&all_dpif_backers);
5856 backer = node->data;
5859 /* Extract the ofproto_dpif object from the ofproto_receive()
5861 if (ofproto_receive(backer, NULL, odp_key.data,
5862 odp_key.size, &flow, NULL, &ofproto, NULL)) {
5863 unixctl_command_reply_error(conn, "Invalid datapath flow");
5866 ds_put_format(&result, "Bridge: %s\n", ofproto->up.name);
5867 } else if (!parse_ofp_exact_flow(&flow, argv[argc - 1])) {
5869 unixctl_command_reply_error(conn, "Must specify bridge name");
5873 ofproto = ofproto_dpif_lookup(argv[1]);
5875 unixctl_command_reply_error(conn, "Unknown bridge name");
5879 unixctl_command_reply_error(conn, "Bad flow syntax");
5883 /* Generate a packet, if requested. */
5885 if (!packet->size) {
5886 flow_compose(packet, &flow);
5888 ds_put_cstr(&result, "Packet: ");
5889 s = ofp_packet_to_string(packet->data, packet->size);
5890 ds_put_cstr(&result, s);
5893 /* Use the metadata from the flow and the packet argument
5894 * to reconstruct the flow. */
5895 flow_extract(packet, flow.skb_priority, flow.skb_mark, NULL,
5896 flow.in_port, &flow);
5900 ofproto_trace(ofproto, &flow, packet, &result);
5901 unixctl_command_reply(conn, ds_cstr(&result));
5904 ds_destroy(&result);
5905 ofpbuf_delete(packet);
5906 ofpbuf_uninit(&odp_key);
5910 ofproto_trace(struct ofproto_dpif *ofproto, const struct flow *flow,
5911 const struct ofpbuf *packet, struct ds *ds)
5913 struct rule_dpif *rule;
5915 ds_put_cstr(ds, "Flow: ");
5916 flow_format(ds, flow);
5917 ds_put_char(ds, '\n');
5919 rule = rule_dpif_lookup(ofproto, flow, NULL);
5921 trace_format_rule(ds, 0, rule);
5922 if (rule == ofproto->miss_rule) {
5923 ds_put_cstr(ds, "\nNo match, flow generates \"packet in\"s.\n");
5924 } else if (rule == ofproto->no_packet_in_rule) {
5925 ds_put_cstr(ds, "\nNo match, packets dropped because "
5926 "OFPPC_NO_PACKET_IN is set on in_port.\n");
5927 } else if (rule == ofproto->drop_frags_rule) {
5928 ds_put_cstr(ds, "\nPackets dropped because they are IP fragments "
5929 "and the fragment handling mode is \"drop\".\n");
5933 uint64_t odp_actions_stub[1024 / 8];
5934 struct ofpbuf odp_actions;
5935 struct trace_ctx trace;
5939 tcp_flags = packet ? packet_get_tcp_flags(packet, flow) : 0;
5942 ofpbuf_use_stub(&odp_actions,
5943 odp_actions_stub, sizeof odp_actions_stub);
5944 xlate_in_init(&trace.xin, ofproto, flow, rule, tcp_flags, packet);
5945 trace.xin.resubmit_hook = trace_resubmit;
5946 trace.xin.report_hook = trace_report;
5948 xlate_actions(&trace.xin, &trace.xout);
5950 ds_put_char(ds, '\n');
5951 trace_format_flow(ds, 0, "Final flow", &trace);
5953 match_init(&match, flow, &trace.xout.wc);
5954 ds_put_cstr(ds, "Relevant fields: ");
5955 match_format(&match, ds, OFP_DEFAULT_PRIORITY);
5956 ds_put_char(ds, '\n');
5958 ds_put_cstr(ds, "Datapath actions: ");
5959 format_odp_actions(ds, trace.xout.odp_actions.data,
5960 trace.xout.odp_actions.size);
5962 if (trace.xout.slow) {
5963 ds_put_cstr(ds, "\nThis flow is handled by the userspace "
5964 "slow path because it:");
5965 switch (trace.xout.slow) {
5967 ds_put_cstr(ds, "\n\t- Consists of CFM packets.");
5970 ds_put_cstr(ds, "\n\t- Consists of LACP packets.");
5973 ds_put_cstr(ds, "\n\t- Consists of STP packets.");
5976 ds_put_cstr(ds, "\n\t- Consists of BFD packets.");
5978 case SLOW_CONTROLLER:
5979 ds_put_cstr(ds, "\n\t- Sends \"packet-in\" messages "
5980 "to the OpenFlow controller.");
5987 xlate_out_uninit(&trace.xout);
5992 ofproto_dpif_clog(struct unixctl_conn *conn OVS_UNUSED, int argc OVS_UNUSED,
5993 const char *argv[] OVS_UNUSED, void *aux OVS_UNUSED)
5996 unixctl_command_reply(conn, NULL);
6000 ofproto_dpif_unclog(struct unixctl_conn *conn OVS_UNUSED, int argc OVS_UNUSED,
6001 const char *argv[] OVS_UNUSED, void *aux OVS_UNUSED)
6004 unixctl_command_reply(conn, NULL);
6007 /* Runs a self-check of flow translations in 'ofproto'. Appends a message to
6008 * 'reply' describing the results. */
6010 ofproto_dpif_self_check__(struct ofproto_dpif *ofproto, struct ds *reply)
6012 struct cls_cursor cursor;
6013 struct facet *facet;
6017 cls_cursor_init(&cursor, &ofproto->facets, NULL);
6018 CLS_CURSOR_FOR_EACH (facet, cr, &cursor) {
6019 if (!facet_check_consistency(facet)) {
6024 ofproto->backer->need_revalidate = REV_INCONSISTENCY;
6028 ds_put_format(reply, "%s: self-check failed (%d errors)\n",
6029 ofproto->up.name, errors);
6031 ds_put_format(reply, "%s: self-check passed\n", ofproto->up.name);
6036 ofproto_dpif_self_check(struct unixctl_conn *conn,
6037 int argc, const char *argv[], void *aux OVS_UNUSED)
6039 struct ds reply = DS_EMPTY_INITIALIZER;
6040 struct ofproto_dpif *ofproto;
6043 ofproto = ofproto_dpif_lookup(argv[1]);
6045 unixctl_command_reply_error(conn, "Unknown ofproto (use "
6046 "ofproto/list for help)");
6049 ofproto_dpif_self_check__(ofproto, &reply);
6051 HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
6052 ofproto_dpif_self_check__(ofproto, &reply);
6056 unixctl_command_reply(conn, ds_cstr(&reply));
6060 /* Store the current ofprotos in 'ofproto_shash'. Returns a sorted list
6061 * of the 'ofproto_shash' nodes. It is the responsibility of the caller
6062 * to destroy 'ofproto_shash' and free the returned value. */
6063 static const struct shash_node **
6064 get_ofprotos(struct shash *ofproto_shash)
6066 const struct ofproto_dpif *ofproto;
6068 HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
6069 char *name = xasprintf("%s@%s", ofproto->up.type, ofproto->up.name);
6070 shash_add_nocopy(ofproto_shash, name, ofproto);
6073 return shash_sort(ofproto_shash);
6077 ofproto_unixctl_dpif_dump_dps(struct unixctl_conn *conn, int argc OVS_UNUSED,
6078 const char *argv[] OVS_UNUSED,
6079 void *aux OVS_UNUSED)
6081 struct ds ds = DS_EMPTY_INITIALIZER;
6082 struct shash ofproto_shash;
6083 const struct shash_node **sorted_ofprotos;
6086 shash_init(&ofproto_shash);
6087 sorted_ofprotos = get_ofprotos(&ofproto_shash);
6088 for (i = 0; i < shash_count(&ofproto_shash); i++) {
6089 const struct shash_node *node = sorted_ofprotos[i];
6090 ds_put_format(&ds, "%s\n", node->name);
6093 shash_destroy(&ofproto_shash);
6094 free(sorted_ofprotos);
6096 unixctl_command_reply(conn, ds_cstr(&ds));
6101 show_dp_rates(struct ds *ds, const char *heading,
6102 const struct avg_subfacet_rates *rates)
6104 ds_put_format(ds, "%s add rate: %5.3f/min, del rate: %5.3f/min\n",
6105 heading, rates->add_rate, rates->del_rate);
6109 dpif_show_backer(const struct dpif_backer *backer, struct ds *ds)
6111 const struct shash_node **ofprotos;
6112 struct ofproto_dpif *ofproto;
6113 struct shash ofproto_shash;
6114 uint64_t n_hit, n_missed;
6115 long long int minutes;
6118 n_hit = n_missed = 0;
6119 HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
6120 if (ofproto->backer == backer) {
6121 n_missed += ofproto->n_missed;
6122 n_hit += ofproto->n_hit;
6126 ds_put_format(ds, "%s: hit:%"PRIu64" missed:%"PRIu64"\n",
6127 dpif_name(backer->dpif), n_hit, n_missed);
6128 ds_put_format(ds, "\tflows: cur: %zu, avg: %u, max: %u,"
6129 " life span: %lldms\n", hmap_count(&backer->subfacets),
6130 backer->avg_n_subfacet, backer->max_n_subfacet,
6131 backer->avg_subfacet_life);
6133 minutes = (time_msec() - backer->created) / (1000 * 60);
6134 if (minutes >= 60) {
6135 show_dp_rates(ds, "\thourly avg:", &backer->hourly);
6137 if (minutes >= 60 * 24) {
6138 show_dp_rates(ds, "\tdaily avg:", &backer->daily);
6140 show_dp_rates(ds, "\toverall avg:", &backer->lifetime);
6142 shash_init(&ofproto_shash);
6143 ofprotos = get_ofprotos(&ofproto_shash);
6144 for (i = 0; i < shash_count(&ofproto_shash); i++) {
6145 struct ofproto_dpif *ofproto = ofprotos[i]->data;
6146 const struct shash_node **ports;
6149 if (ofproto->backer != backer) {
6153 ds_put_format(ds, "\t%s: hit:%"PRIu64" missed:%"PRIu64"\n",
6154 ofproto->up.name, ofproto->n_hit, ofproto->n_missed);
6156 ports = shash_sort(&ofproto->up.port_by_name);
6157 for (j = 0; j < shash_count(&ofproto->up.port_by_name); j++) {
6158 const struct shash_node *node = ports[j];
6159 struct ofport *ofport = node->data;
6163 ds_put_format(ds, "\t\t%s %u/", netdev_get_name(ofport->netdev),
6166 odp_port = ofp_port_to_odp_port(ofproto, ofport->ofp_port);
6167 if (odp_port != OVSP_NONE) {
6168 ds_put_format(ds, "%"PRIu32":", odp_port);
6170 ds_put_cstr(ds, "none:");
6173 ds_put_format(ds, " (%s", netdev_get_type(ofport->netdev));
6176 if (!netdev_get_config(ofport->netdev, &config)) {
6177 const struct smap_node **nodes;
6180 nodes = smap_sort(&config);
6181 for (i = 0; i < smap_count(&config); i++) {
6182 const struct smap_node *node = nodes[i];
6183 ds_put_format(ds, "%c %s=%s", i ? ',' : ':',
6184 node->key, node->value);
6188 smap_destroy(&config);
6190 ds_put_char(ds, ')');
6191 ds_put_char(ds, '\n');
6195 shash_destroy(&ofproto_shash);
6200 ofproto_unixctl_dpif_show(struct unixctl_conn *conn, int argc OVS_UNUSED,
6201 const char *argv[] OVS_UNUSED, void *aux OVS_UNUSED)
6203 struct ds ds = DS_EMPTY_INITIALIZER;
6204 const struct shash_node **backers;
6207 backers = shash_sort(&all_dpif_backers);
6208 for (i = 0; i < shash_count(&all_dpif_backers); i++) {
6209 dpif_show_backer(backers[i]->data, &ds);
6213 unixctl_command_reply(conn, ds_cstr(&ds));
6217 /* Dump the megaflow (facet) cache. This is useful to check the
6218 * correctness of flow wildcarding, since the same mechanism is used for
6219 * both xlate caching and kernel wildcarding.
6221 * It's important to note that in the output the flow description uses
6222 * OpenFlow (OFP) ports, but the actions use datapath (ODP) ports.
6224 * This command is only needed for advanced debugging, so it's not
6225 * documented in the man page. */
6227 ofproto_unixctl_dpif_dump_megaflows(struct unixctl_conn *conn,
6228 int argc OVS_UNUSED, const char *argv[],
6229 void *aux OVS_UNUSED)
6231 struct ds ds = DS_EMPTY_INITIALIZER;
6232 const struct ofproto_dpif *ofproto;
6233 long long int now = time_msec();
6234 struct cls_cursor cursor;
6235 struct facet *facet;
6237 ofproto = ofproto_dpif_lookup(argv[1]);
6239 unixctl_command_reply_error(conn, "no such bridge");
6243 cls_cursor_init(&cursor, &ofproto->facets, NULL);
6244 CLS_CURSOR_FOR_EACH (facet, cr, &cursor) {
6245 cls_rule_format(&facet->cr, &ds);
6246 ds_put_cstr(&ds, ", ");
6247 ds_put_format(&ds, "n_subfacets:%zu, ", list_size(&facet->subfacets));
6248 ds_put_format(&ds, "used:%.3fs, ", (now - facet->used) / 1000.0);
6249 ds_put_cstr(&ds, "Datapath actions: ");
6250 if (facet->xout.slow) {
6251 uint64_t slow_path_stub[128 / 8];
6252 const struct nlattr *actions;
6255 compose_slow_path(ofproto, &facet->flow, facet->xout.slow,
6256 slow_path_stub, sizeof slow_path_stub,
6257 &actions, &actions_len);
6258 format_odp_actions(&ds, actions, actions_len);
6260 format_odp_actions(&ds, facet->xout.odp_actions.data,
6261 facet->xout.odp_actions.size);
6263 ds_put_cstr(&ds, "\n");
6266 ds_chomp(&ds, '\n');
6267 unixctl_command_reply(conn, ds_cstr(&ds));
6272 ofproto_unixctl_dpif_dump_flows(struct unixctl_conn *conn,
6273 int argc OVS_UNUSED, const char *argv[],
6274 void *aux OVS_UNUSED)
6276 struct ds ds = DS_EMPTY_INITIALIZER;
6277 const struct ofproto_dpif *ofproto;
6278 struct subfacet *subfacet;
6280 ofproto = ofproto_dpif_lookup(argv[1]);
6282 unixctl_command_reply_error(conn, "no such bridge");
6286 update_stats(ofproto->backer);
6288 HMAP_FOR_EACH (subfacet, hmap_node, &ofproto->backer->subfacets) {
6289 struct facet *facet = subfacet->facet;
6291 if (ofproto_dpif_cast(facet->rule->up.ofproto) != ofproto) {
6295 odp_flow_key_format(subfacet->key, subfacet->key_len, &ds);
6297 ds_put_format(&ds, ", packets:%"PRIu64", bytes:%"PRIu64", used:",
6298 subfacet->dp_packet_count, subfacet->dp_byte_count);
6299 if (subfacet->used) {
6300 ds_put_format(&ds, "%.3fs",
6301 (time_msec() - subfacet->used) / 1000.0);
6303 ds_put_format(&ds, "never");
6305 if (subfacet->facet->tcp_flags) {
6306 ds_put_cstr(&ds, ", flags:");
6307 packet_format_tcp_flags(&ds, subfacet->facet->tcp_flags);
6310 ds_put_cstr(&ds, ", actions:");
6311 if (facet->xout.slow) {
6312 uint64_t slow_path_stub[128 / 8];
6313 const struct nlattr *actions;
6316 compose_slow_path(ofproto, &facet->flow, facet->xout.slow,
6317 slow_path_stub, sizeof slow_path_stub,
6318 &actions, &actions_len);
6319 format_odp_actions(&ds, actions, actions_len);
6321 format_odp_actions(&ds, facet->xout.odp_actions.data,
6322 facet->xout.odp_actions.size);
6324 ds_put_char(&ds, '\n');
6327 unixctl_command_reply(conn, ds_cstr(&ds));
6332 ofproto_unixctl_dpif_del_flows(struct unixctl_conn *conn,
6333 int argc OVS_UNUSED, const char *argv[],
6334 void *aux OVS_UNUSED)
6336 struct ds ds = DS_EMPTY_INITIALIZER;
6337 struct ofproto_dpif *ofproto;
6339 ofproto = ofproto_dpif_lookup(argv[1]);
6341 unixctl_command_reply_error(conn, "no such bridge");
6345 flush(&ofproto->up);
6347 unixctl_command_reply(conn, ds_cstr(&ds));
6352 ofproto_dpif_unixctl_init(void)
6354 static bool registered;
6360 unixctl_command_register(
6362 "[dp_name]|bridge odp_flow|br_flow [-generate|packet]",
6363 1, 3, ofproto_unixctl_trace, NULL);
6364 unixctl_command_register("fdb/flush", "[bridge]", 0, 1,
6365 ofproto_unixctl_fdb_flush, NULL);
6366 unixctl_command_register("fdb/show", "bridge", 1, 1,
6367 ofproto_unixctl_fdb_show, NULL);
6368 unixctl_command_register("ofproto/clog", "", 0, 0,
6369 ofproto_dpif_clog, NULL);
6370 unixctl_command_register("ofproto/unclog", "", 0, 0,
6371 ofproto_dpif_unclog, NULL);
6372 unixctl_command_register("ofproto/self-check", "[bridge]", 0, 1,
6373 ofproto_dpif_self_check, NULL);
6374 unixctl_command_register("dpif/dump-dps", "", 0, 0,
6375 ofproto_unixctl_dpif_dump_dps, NULL);
6376 unixctl_command_register("dpif/show", "", 0, 0, ofproto_unixctl_dpif_show,
6378 unixctl_command_register("dpif/dump-flows", "bridge", 1, 1,
6379 ofproto_unixctl_dpif_dump_flows, NULL);
6380 unixctl_command_register("dpif/del-flows", "bridge", 1, 1,
6381 ofproto_unixctl_dpif_del_flows, NULL);
6382 unixctl_command_register("dpif/dump-megaflows", "bridge", 1, 1,
6383 ofproto_unixctl_dpif_dump_megaflows, NULL);
6386 /* Linux VLAN device support (e.g. "eth0.10" for VLAN 10.)
6388 * This is deprecated. It is only for compatibility with broken device drivers
6389 * in old versions of Linux that do not properly support VLANs when VLAN
6390 * devices are not used. When broken device drivers are no longer in
6391 * widespread use, we will delete these interfaces. */
6394 set_realdev(struct ofport *ofport_, uint16_t realdev_ofp_port, int vid)
6396 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport_->ofproto);
6397 struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
6399 if (realdev_ofp_port == ofport->realdev_ofp_port
6400 && vid == ofport->vlandev_vid) {
6404 ofproto->backer->need_revalidate = REV_RECONFIGURE;
6406 if (ofport->realdev_ofp_port) {
6409 if (realdev_ofp_port && ofport->bundle) {
6410 /* vlandevs are enslaved to their realdevs, so they are not allowed to
6411 * themselves be part of a bundle. */
6412 bundle_set(ofport->up.ofproto, ofport->bundle, NULL);
6415 ofport->realdev_ofp_port = realdev_ofp_port;
6416 ofport->vlandev_vid = vid;
6418 if (realdev_ofp_port) {
6419 vsp_add(ofport, realdev_ofp_port, vid);
6426 hash_realdev_vid(uint16_t realdev_ofp_port, int vid)
6428 return hash_2words(realdev_ofp_port, vid);
6431 /* Returns the OFP port number of the Linux VLAN device that corresponds to
6432 * 'vlan_tci' on the network device with port number 'realdev_ofp_port' in
6433 * 'struct ofport_dpif'. For example, given 'realdev_ofp_port' of eth0 and
6434 * 'vlan_tci' 9, it would return the port number of eth0.9.
6436 * Unless VLAN splinters are enabled for port 'realdev_ofp_port', this
6437 * function just returns its 'realdev_ofp_port' argument. */
6439 vsp_realdev_to_vlandev(const struct ofproto_dpif *ofproto,
6440 uint16_t realdev_ofp_port, ovs_be16 vlan_tci)
6442 if (!hmap_is_empty(&ofproto->realdev_vid_map)) {
6443 int vid = vlan_tci_to_vid(vlan_tci);
6444 const struct vlan_splinter *vsp;
6446 HMAP_FOR_EACH_WITH_HASH (vsp, realdev_vid_node,
6447 hash_realdev_vid(realdev_ofp_port, vid),
6448 &ofproto->realdev_vid_map) {
6449 if (vsp->realdev_ofp_port == realdev_ofp_port
6450 && vsp->vid == vid) {
6451 return vsp->vlandev_ofp_port;
6455 return realdev_ofp_port;
6458 static struct vlan_splinter *
6459 vlandev_find(const struct ofproto_dpif *ofproto, uint16_t vlandev_ofp_port)
6461 struct vlan_splinter *vsp;
6463 HMAP_FOR_EACH_WITH_HASH (vsp, vlandev_node, hash_int(vlandev_ofp_port, 0),
6464 &ofproto->vlandev_map) {
6465 if (vsp->vlandev_ofp_port == vlandev_ofp_port) {
6473 /* Returns the OpenFlow port number of the "real" device underlying the Linux
6474 * VLAN device with OpenFlow port number 'vlandev_ofp_port' and stores the
6475 * VLAN VID of the Linux VLAN device in '*vid'. For example, given
6476 * 'vlandev_ofp_port' of eth0.9, it would return the OpenFlow port number of
6477 * eth0 and store 9 in '*vid'.
6479 * Returns 0 and does not modify '*vid' if 'vlandev_ofp_port' is not a Linux
6480 * VLAN device. Unless VLAN splinters are enabled, this is what this function
6483 vsp_vlandev_to_realdev(const struct ofproto_dpif *ofproto,
6484 uint16_t vlandev_ofp_port, int *vid)
6486 if (!hmap_is_empty(&ofproto->vlandev_map)) {
6487 const struct vlan_splinter *vsp;
6489 vsp = vlandev_find(ofproto, vlandev_ofp_port);
6494 return vsp->realdev_ofp_port;
6500 /* Given 'flow', a flow representing a packet received on 'ofproto', checks
6501 * whether 'flow->in_port' represents a Linux VLAN device. If so, changes
6502 * 'flow->in_port' to the "real" device backing the VLAN device, sets
6503 * 'flow->vlan_tci' to the VLAN VID, and returns true. Otherwise (which is
6504 * always the case unless VLAN splinters are enabled), returns false without
6505 * making any changes. */
6507 vsp_adjust_flow(const struct ofproto_dpif *ofproto, struct flow *flow)
6512 realdev = vsp_vlandev_to_realdev(ofproto, flow->in_port, &vid);
6517 /* Cause the flow to be processed as if it came in on the real device with
6518 * the VLAN device's VLAN ID. */
6519 flow->in_port = realdev;
6520 flow->vlan_tci = htons((vid & VLAN_VID_MASK) | VLAN_CFI);
6525 vsp_remove(struct ofport_dpif *port)
6527 struct ofproto_dpif *ofproto = ofproto_dpif_cast(port->up.ofproto);
6528 struct vlan_splinter *vsp;
6530 vsp = vlandev_find(ofproto, port->up.ofp_port);
6532 hmap_remove(&ofproto->vlandev_map, &vsp->vlandev_node);
6533 hmap_remove(&ofproto->realdev_vid_map, &vsp->realdev_vid_node);
6536 port->realdev_ofp_port = 0;
6538 VLOG_ERR("missing vlan device record");
6543 vsp_add(struct ofport_dpif *port, uint16_t realdev_ofp_port, int vid)
6545 struct ofproto_dpif *ofproto = ofproto_dpif_cast(port->up.ofproto);
6547 if (!vsp_vlandev_to_realdev(ofproto, port->up.ofp_port, NULL)
6548 && (vsp_realdev_to_vlandev(ofproto, realdev_ofp_port, htons(vid))
6549 == realdev_ofp_port)) {
6550 struct vlan_splinter *vsp;
6552 vsp = xmalloc(sizeof *vsp);
6553 hmap_insert(&ofproto->vlandev_map, &vsp->vlandev_node,
6554 hash_int(port->up.ofp_port, 0));
6555 hmap_insert(&ofproto->realdev_vid_map, &vsp->realdev_vid_node,
6556 hash_realdev_vid(realdev_ofp_port, vid));
6557 vsp->realdev_ofp_port = realdev_ofp_port;
6558 vsp->vlandev_ofp_port = port->up.ofp_port;
6561 port->realdev_ofp_port = realdev_ofp_port;
6563 VLOG_ERR("duplicate vlan device record");
6568 ofp_port_to_odp_port(const struct ofproto_dpif *ofproto, uint16_t ofp_port)
6570 const struct ofport_dpif *ofport = get_ofp_port(ofproto, ofp_port);
6571 return ofport ? ofport->odp_port : OVSP_NONE;
6574 static struct ofport_dpif *
6575 odp_port_to_ofport(const struct dpif_backer *backer, uint32_t odp_port)
6577 struct ofport_dpif *port;
6579 HMAP_FOR_EACH_IN_BUCKET (port, odp_port_node,
6580 hash_int(odp_port, 0),
6581 &backer->odp_to_ofport_map) {
6582 if (port->odp_port == odp_port) {
6591 odp_port_to_ofp_port(const struct ofproto_dpif *ofproto, uint32_t odp_port)
6593 struct ofport_dpif *port;
6595 port = odp_port_to_ofport(ofproto->backer, odp_port);
6596 if (port && &ofproto->up == port->up.ofproto) {
6597 return port->up.ofp_port;
6603 /* Compute exponentially weighted moving average, adding 'new' as the newest,
6604 * most heavily weighted element. 'base' designates the rate of decay: after
6605 * 'base' further updates, 'new''s weight in the EWMA decays to about 1/e
6608 exp_mavg(double *avg, int base, double new)
6610 *avg = (*avg * (base - 1) + new) / base;
6614 update_moving_averages(struct dpif_backer *backer)
6616 const int min_ms = 60 * 1000; /* milliseconds in one minute. */
6617 long long int minutes = (time_msec() - backer->created) / min_ms;
6620 backer->lifetime.add_rate = (double) backer->total_subfacet_add_count
6622 backer->lifetime.del_rate = (double) backer->total_subfacet_del_count
6625 backer->lifetime.add_rate = 0.0;
6626 backer->lifetime.del_rate = 0.0;
6629 /* Update hourly averages on the minute boundaries. */
6630 if (time_msec() - backer->last_minute >= min_ms) {
6631 exp_mavg(&backer->hourly.add_rate, 60, backer->subfacet_add_count);
6632 exp_mavg(&backer->hourly.del_rate, 60, backer->subfacet_del_count);
6634 /* Update daily averages on the hour boundaries. */
6635 if ((backer->last_minute - backer->created) / min_ms % 60 == 59) {
6636 exp_mavg(&backer->daily.add_rate, 24, backer->hourly.add_rate);
6637 exp_mavg(&backer->daily.del_rate, 24, backer->hourly.del_rate);
6640 backer->total_subfacet_add_count += backer->subfacet_add_count;
6641 backer->total_subfacet_del_count += backer->subfacet_del_count;
6642 backer->subfacet_add_count = 0;
6643 backer->subfacet_del_count = 0;
6644 backer->last_minute += min_ms;
6648 const struct ofproto_class ofproto_dpif_class = {
6683 port_is_lacp_current,
6684 NULL, /* rule_choose_table */
6691 rule_modify_actions,
6705 get_stp_port_status,
6712 is_mirror_output_bundle,
6713 forward_bpdu_changed,
6714 set_mac_table_config,