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);
1561 if (port->tnl_port && tnl_port_reconfigure(&port->up, port->odp_port,
1563 ofproto_dpif_cast(port->up.ofproto)->backer->need_revalidate = true;
1568 port_reconfigured(struct ofport *port_, enum ofputil_port_config old_config)
1570 struct ofport_dpif *port = ofport_dpif_cast(port_);
1571 struct ofproto_dpif *ofproto = ofproto_dpif_cast(port->up.ofproto);
1572 enum ofputil_port_config changed = old_config ^ port->up.pp.config;
1574 if (changed & (OFPUTIL_PC_NO_RECV | OFPUTIL_PC_NO_RECV_STP |
1575 OFPUTIL_PC_NO_FWD | OFPUTIL_PC_NO_FLOOD |
1576 OFPUTIL_PC_NO_PACKET_IN)) {
1577 ofproto->backer->need_revalidate = REV_RECONFIGURE;
1579 if (changed & OFPUTIL_PC_NO_FLOOD && port->bundle) {
1580 bundle_update(port->bundle);
1586 set_sflow(struct ofproto *ofproto_,
1587 const struct ofproto_sflow_options *sflow_options)
1589 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1590 struct dpif_sflow *ds = ofproto->sflow;
1592 if (sflow_options) {
1594 struct ofport_dpif *ofport;
1596 ds = ofproto->sflow = dpif_sflow_create();
1597 HMAP_FOR_EACH (ofport, up.hmap_node, &ofproto->up.ports) {
1598 dpif_sflow_add_port(ds, &ofport->up, ofport->odp_port);
1600 ofproto->backer->need_revalidate = REV_RECONFIGURE;
1602 dpif_sflow_set_options(ds, sflow_options);
1605 dpif_sflow_destroy(ds);
1606 ofproto->backer->need_revalidate = REV_RECONFIGURE;
1607 ofproto->sflow = NULL;
1615 struct ofproto *ofproto_,
1616 const struct ofproto_ipfix_bridge_exporter_options *bridge_exporter_options,
1617 const struct ofproto_ipfix_flow_exporter_options *flow_exporters_options,
1618 size_t n_flow_exporters_options)
1620 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1621 struct dpif_ipfix *di = ofproto->ipfix;
1623 if (bridge_exporter_options || flow_exporters_options) {
1625 di = ofproto->ipfix = dpif_ipfix_create();
1627 dpif_ipfix_set_options(
1628 di, bridge_exporter_options, flow_exporters_options,
1629 n_flow_exporters_options);
1632 dpif_ipfix_destroy(di);
1633 ofproto->ipfix = NULL;
1640 set_cfm(struct ofport *ofport_, const struct cfm_settings *s)
1642 struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
1649 struct ofproto_dpif *ofproto;
1651 ofproto = ofproto_dpif_cast(ofport->up.ofproto);
1652 ofproto->backer->need_revalidate = REV_RECONFIGURE;
1653 ofport->cfm = cfm_create(ofport->up.netdev);
1656 if (cfm_configure(ofport->cfm, s)) {
1662 cfm_destroy(ofport->cfm);
1668 get_cfm_status(const struct ofport *ofport_,
1669 struct ofproto_cfm_status *status)
1671 struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
1674 status->faults = cfm_get_fault(ofport->cfm);
1675 status->remote_opstate = cfm_get_opup(ofport->cfm);
1676 status->health = cfm_get_health(ofport->cfm);
1677 cfm_get_remote_mpids(ofport->cfm, &status->rmps, &status->n_rmps);
1685 set_bfd(struct ofport *ofport_, const struct smap *cfg)
1687 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport_->ofproto);
1688 struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
1692 ofport->bfd = bfd_configure(old, netdev_get_name(ofport->up.netdev), cfg);
1693 if (ofport->bfd != old) {
1694 ofproto->backer->need_revalidate = REV_RECONFIGURE;
1701 get_bfd_status(struct ofport *ofport_, struct smap *smap)
1703 struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
1706 bfd_get_status(ofport->bfd, smap);
1713 /* Spanning Tree. */
1716 send_bpdu_cb(struct ofpbuf *pkt, int port_num, void *ofproto_)
1718 struct ofproto_dpif *ofproto = ofproto_;
1719 struct stp_port *sp = stp_get_port(ofproto->stp, port_num);
1720 struct ofport_dpif *ofport;
1722 ofport = stp_port_get_aux(sp);
1724 VLOG_WARN_RL(&rl, "%s: cannot send BPDU on unknown port %d",
1725 ofproto->up.name, port_num);
1727 struct eth_header *eth = pkt->l2;
1729 netdev_get_etheraddr(ofport->up.netdev, eth->eth_src);
1730 if (eth_addr_is_zero(eth->eth_src)) {
1731 VLOG_WARN_RL(&rl, "%s: cannot send BPDU on port %d "
1732 "with unknown MAC", ofproto->up.name, port_num);
1734 send_packet(ofport, pkt);
1740 /* Configures STP on 'ofproto_' using the settings defined in 's'. */
1742 set_stp(struct ofproto *ofproto_, const struct ofproto_stp_settings *s)
1744 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1746 /* Only revalidate flows if the configuration changed. */
1747 if (!s != !ofproto->stp) {
1748 ofproto->backer->need_revalidate = REV_RECONFIGURE;
1752 if (!ofproto->stp) {
1753 ofproto->stp = stp_create(ofproto_->name, s->system_id,
1754 send_bpdu_cb, ofproto);
1755 ofproto->stp_last_tick = time_msec();
1758 stp_set_bridge_id(ofproto->stp, s->system_id);
1759 stp_set_bridge_priority(ofproto->stp, s->priority);
1760 stp_set_hello_time(ofproto->stp, s->hello_time);
1761 stp_set_max_age(ofproto->stp, s->max_age);
1762 stp_set_forward_delay(ofproto->stp, s->fwd_delay);
1764 struct ofport *ofport;
1766 HMAP_FOR_EACH (ofport, hmap_node, &ofproto->up.ports) {
1767 set_stp_port(ofport, NULL);
1770 stp_destroy(ofproto->stp);
1771 ofproto->stp = NULL;
1778 get_stp_status(struct ofproto *ofproto_, struct ofproto_stp_status *s)
1780 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1784 s->bridge_id = stp_get_bridge_id(ofproto->stp);
1785 s->designated_root = stp_get_designated_root(ofproto->stp);
1786 s->root_path_cost = stp_get_root_path_cost(ofproto->stp);
1795 update_stp_port_state(struct ofport_dpif *ofport)
1797 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport->up.ofproto);
1798 enum stp_state state;
1800 /* Figure out new state. */
1801 state = ofport->stp_port ? stp_port_get_state(ofport->stp_port)
1805 if (ofport->stp_state != state) {
1806 enum ofputil_port_state of_state;
1809 VLOG_DBG_RL(&rl, "port %s: STP state changed from %s to %s",
1810 netdev_get_name(ofport->up.netdev),
1811 stp_state_name(ofport->stp_state),
1812 stp_state_name(state));
1813 if (stp_learn_in_state(ofport->stp_state)
1814 != stp_learn_in_state(state)) {
1815 /* xxx Learning action flows should also be flushed. */
1816 mac_learning_flush(ofproto->ml,
1817 &ofproto->backer->revalidate_set);
1819 fwd_change = stp_forward_in_state(ofport->stp_state)
1820 != stp_forward_in_state(state);
1822 ofproto->backer->need_revalidate = REV_STP;
1823 ofport->stp_state = state;
1824 ofport->stp_state_entered = time_msec();
1826 if (fwd_change && ofport->bundle) {
1827 bundle_update(ofport->bundle);
1830 /* Update the STP state bits in the OpenFlow port description. */
1831 of_state = ofport->up.pp.state & ~OFPUTIL_PS_STP_MASK;
1832 of_state |= (state == STP_LISTENING ? OFPUTIL_PS_STP_LISTEN
1833 : state == STP_LEARNING ? OFPUTIL_PS_STP_LEARN
1834 : state == STP_FORWARDING ? OFPUTIL_PS_STP_FORWARD
1835 : state == STP_BLOCKING ? OFPUTIL_PS_STP_BLOCK
1837 ofproto_port_set_state(&ofport->up, of_state);
1841 /* Configures STP on 'ofport_' using the settings defined in 's'. The
1842 * caller is responsible for assigning STP port numbers and ensuring
1843 * there are no duplicates. */
1845 set_stp_port(struct ofport *ofport_,
1846 const struct ofproto_port_stp_settings *s)
1848 struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
1849 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport->up.ofproto);
1850 struct stp_port *sp = ofport->stp_port;
1852 if (!s || !s->enable) {
1854 ofport->stp_port = NULL;
1855 stp_port_disable(sp);
1856 update_stp_port_state(ofport);
1859 } else if (sp && stp_port_no(sp) != s->port_num
1860 && ofport == stp_port_get_aux(sp)) {
1861 /* The port-id changed, so disable the old one if it's not
1862 * already in use by another port. */
1863 stp_port_disable(sp);
1866 sp = ofport->stp_port = stp_get_port(ofproto->stp, s->port_num);
1867 stp_port_enable(sp);
1869 stp_port_set_aux(sp, ofport);
1870 stp_port_set_priority(sp, s->priority);
1871 stp_port_set_path_cost(sp, s->path_cost);
1873 update_stp_port_state(ofport);
1879 get_stp_port_status(struct ofport *ofport_,
1880 struct ofproto_port_stp_status *s)
1882 struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
1883 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport->up.ofproto);
1884 struct stp_port *sp = ofport->stp_port;
1886 if (!ofproto->stp || !sp) {
1892 s->port_id = stp_port_get_id(sp);
1893 s->state = stp_port_get_state(sp);
1894 s->sec_in_state = (time_msec() - ofport->stp_state_entered) / 1000;
1895 s->role = stp_port_get_role(sp);
1896 stp_port_get_counts(sp, &s->tx_count, &s->rx_count, &s->error_count);
1902 stp_run(struct ofproto_dpif *ofproto)
1905 long long int now = time_msec();
1906 long long int elapsed = now - ofproto->stp_last_tick;
1907 struct stp_port *sp;
1910 stp_tick(ofproto->stp, MIN(INT_MAX, elapsed));
1911 ofproto->stp_last_tick = now;
1913 while (stp_get_changed_port(ofproto->stp, &sp)) {
1914 struct ofport_dpif *ofport = stp_port_get_aux(sp);
1917 update_stp_port_state(ofport);
1921 if (stp_check_and_reset_fdb_flush(ofproto->stp)) {
1922 mac_learning_flush(ofproto->ml, &ofproto->backer->revalidate_set);
1928 stp_wait(struct ofproto_dpif *ofproto)
1931 poll_timer_wait(1000);
1935 /* Returns true if STP should process 'flow'. Sets fields in 'wc' that
1936 * were used to make the determination.*/
1938 stp_should_process_flow(const struct flow *flow, struct flow_wildcards *wc)
1940 memset(&wc->masks.dl_dst, 0xff, sizeof wc->masks.dl_dst);
1941 return eth_addr_equals(flow->dl_dst, eth_addr_stp);
1945 stp_process_packet(const struct ofport_dpif *ofport,
1946 const struct ofpbuf *packet)
1948 struct ofpbuf payload = *packet;
1949 struct eth_header *eth = payload.data;
1950 struct stp_port *sp = ofport->stp_port;
1952 /* Sink packets on ports that have STP disabled when the bridge has
1954 if (!sp || stp_port_get_state(sp) == STP_DISABLED) {
1958 /* Trim off padding on payload. */
1959 if (payload.size > ntohs(eth->eth_type) + ETH_HEADER_LEN) {
1960 payload.size = ntohs(eth->eth_type) + ETH_HEADER_LEN;
1963 if (ofpbuf_try_pull(&payload, ETH_HEADER_LEN + LLC_HEADER_LEN)) {
1964 stp_received_bpdu(sp, payload.data, payload.size);
1969 ofproto_dpif_queue_to_priority(const struct ofproto_dpif *ofproto,
1970 uint32_t queue_id, uint32_t *priority)
1972 return dpif_queue_to_priority(ofproto->backer->dpif, queue_id, priority);
1975 static struct priority_to_dscp *
1976 get_priority(const struct ofport_dpif *ofport, uint32_t priority)
1978 struct priority_to_dscp *pdscp;
1981 hash = hash_int(priority, 0);
1982 HMAP_FOR_EACH_IN_BUCKET (pdscp, hmap_node, hash, &ofport->priorities) {
1983 if (pdscp->priority == priority) {
1991 ofproto_dpif_dscp_from_priority(const struct ofport_dpif *ofport,
1992 uint32_t priority, uint8_t *dscp)
1994 struct priority_to_dscp *pdscp = get_priority(ofport, priority);
1995 *dscp = pdscp ? pdscp->dscp : 0;
1996 return pdscp != NULL;
2000 ofport_clear_priorities(struct ofport_dpif *ofport)
2002 struct priority_to_dscp *pdscp, *next;
2004 HMAP_FOR_EACH_SAFE (pdscp, next, hmap_node, &ofport->priorities) {
2005 hmap_remove(&ofport->priorities, &pdscp->hmap_node);
2011 set_queues(struct ofport *ofport_,
2012 const struct ofproto_port_queue *qdscp_list,
2015 struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
2016 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport->up.ofproto);
2017 struct hmap new = HMAP_INITIALIZER(&new);
2020 for (i = 0; i < n_qdscp; i++) {
2021 struct priority_to_dscp *pdscp;
2025 dscp = (qdscp_list[i].dscp << 2) & IP_DSCP_MASK;
2026 if (dpif_queue_to_priority(ofproto->backer->dpif, qdscp_list[i].queue,
2031 pdscp = get_priority(ofport, priority);
2033 hmap_remove(&ofport->priorities, &pdscp->hmap_node);
2035 pdscp = xmalloc(sizeof *pdscp);
2036 pdscp->priority = priority;
2038 ofproto->backer->need_revalidate = REV_RECONFIGURE;
2041 if (pdscp->dscp != dscp) {
2043 ofproto->backer->need_revalidate = REV_RECONFIGURE;
2046 hmap_insert(&new, &pdscp->hmap_node, hash_int(pdscp->priority, 0));
2049 if (!hmap_is_empty(&ofport->priorities)) {
2050 ofport_clear_priorities(ofport);
2051 ofproto->backer->need_revalidate = REV_RECONFIGURE;
2054 hmap_swap(&new, &ofport->priorities);
2062 /* Expires all MAC learning entries associated with 'bundle' and forces its
2063 * ofproto to revalidate every flow.
2065 * Normally MAC learning entries are removed only from the ofproto associated
2066 * with 'bundle', but if 'all_ofprotos' is true, then the MAC learning entries
2067 * are removed from every ofproto. When patch ports and SLB bonds are in use
2068 * and a VM migration happens and the gratuitous ARPs are somehow lost, this
2069 * avoids a MAC_ENTRY_IDLE_TIME delay before the migrated VM can communicate
2070 * with the host from which it migrated. */
2072 bundle_flush_macs(struct ofbundle *bundle, bool all_ofprotos)
2074 struct ofproto_dpif *ofproto = bundle->ofproto;
2075 struct mac_learning *ml = ofproto->ml;
2076 struct mac_entry *mac, *next_mac;
2078 ofproto->backer->need_revalidate = REV_RECONFIGURE;
2079 LIST_FOR_EACH_SAFE (mac, next_mac, lru_node, &ml->lrus) {
2080 if (mac->port.p == bundle) {
2082 struct ofproto_dpif *o;
2084 HMAP_FOR_EACH (o, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
2086 struct mac_entry *e;
2088 e = mac_learning_lookup(o->ml, mac->mac, mac->vlan,
2091 mac_learning_expire(o->ml, e);
2097 mac_learning_expire(ml, mac);
2102 static struct ofbundle *
2103 bundle_lookup(const struct ofproto_dpif *ofproto, void *aux)
2105 struct ofbundle *bundle;
2107 HMAP_FOR_EACH_IN_BUCKET (bundle, hmap_node, hash_pointer(aux, 0),
2108 &ofproto->bundles) {
2109 if (bundle->aux == aux) {
2116 /* Looks up each of the 'n_auxes' pointers in 'auxes' as bundles and adds the
2117 * ones that are found to 'bundles'. */
2119 bundle_lookup_multiple(struct ofproto_dpif *ofproto,
2120 void **auxes, size_t n_auxes,
2121 struct hmapx *bundles)
2125 hmapx_init(bundles);
2126 for (i = 0; i < n_auxes; i++) {
2127 struct ofbundle *bundle = bundle_lookup(ofproto, auxes[i]);
2129 hmapx_add(bundles, bundle);
2135 bundle_update(struct ofbundle *bundle)
2137 struct ofport_dpif *port;
2139 bundle->floodable = true;
2140 LIST_FOR_EACH (port, bundle_node, &bundle->ports) {
2141 if (port->up.pp.config & OFPUTIL_PC_NO_FLOOD
2142 || !stp_forward_in_state(port->stp_state)) {
2143 bundle->floodable = false;
2150 bundle_del_port(struct ofport_dpif *port)
2152 struct ofbundle *bundle = port->bundle;
2154 bundle->ofproto->backer->need_revalidate = REV_RECONFIGURE;
2156 list_remove(&port->bundle_node);
2157 port->bundle = NULL;
2160 lacp_slave_unregister(bundle->lacp, port);
2163 bond_slave_unregister(bundle->bond, port);
2166 bundle_update(bundle);
2170 bundle_add_port(struct ofbundle *bundle, uint16_t ofp_port,
2171 struct lacp_slave_settings *lacp)
2173 struct ofport_dpif *port;
2175 port = get_ofp_port(bundle->ofproto, ofp_port);
2180 if (port->bundle != bundle) {
2181 bundle->ofproto->backer->need_revalidate = REV_RECONFIGURE;
2183 bundle_del_port(port);
2186 port->bundle = bundle;
2187 list_push_back(&bundle->ports, &port->bundle_node);
2188 if (port->up.pp.config & OFPUTIL_PC_NO_FLOOD
2189 || !stp_forward_in_state(port->stp_state)) {
2190 bundle->floodable = false;
2194 bundle->ofproto->backer->need_revalidate = REV_RECONFIGURE;
2195 lacp_slave_register(bundle->lacp, port, lacp);
2202 bundle_destroy(struct ofbundle *bundle)
2204 struct ofproto_dpif *ofproto;
2205 struct ofport_dpif *port, *next_port;
2212 ofproto = bundle->ofproto;
2213 for (i = 0; i < MAX_MIRRORS; i++) {
2214 struct ofmirror *m = ofproto->mirrors[i];
2216 if (m->out == bundle) {
2218 } else if (hmapx_find_and_delete(&m->srcs, bundle)
2219 || hmapx_find_and_delete(&m->dsts, bundle)) {
2220 ofproto->backer->need_revalidate = REV_RECONFIGURE;
2225 LIST_FOR_EACH_SAFE (port, next_port, bundle_node, &bundle->ports) {
2226 bundle_del_port(port);
2229 bundle_flush_macs(bundle, true);
2230 hmap_remove(&ofproto->bundles, &bundle->hmap_node);
2232 free(bundle->trunks);
2233 lacp_destroy(bundle->lacp);
2234 bond_destroy(bundle->bond);
2239 bundle_set(struct ofproto *ofproto_, void *aux,
2240 const struct ofproto_bundle_settings *s)
2242 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
2243 bool need_flush = false;
2244 struct ofport_dpif *port;
2245 struct ofbundle *bundle;
2246 unsigned long *trunks;
2252 bundle_destroy(bundle_lookup(ofproto, aux));
2256 ovs_assert(s->n_slaves == 1 || s->bond != NULL);
2257 ovs_assert((s->lacp != NULL) == (s->lacp_slaves != NULL));
2259 bundle = bundle_lookup(ofproto, aux);
2261 bundle = xmalloc(sizeof *bundle);
2263 bundle->ofproto = ofproto;
2264 hmap_insert(&ofproto->bundles, &bundle->hmap_node,
2265 hash_pointer(aux, 0));
2267 bundle->name = NULL;
2269 list_init(&bundle->ports);
2270 bundle->vlan_mode = PORT_VLAN_TRUNK;
2272 bundle->trunks = NULL;
2273 bundle->use_priority_tags = s->use_priority_tags;
2274 bundle->lacp = NULL;
2275 bundle->bond = NULL;
2277 bundle->floodable = true;
2279 bundle->src_mirrors = 0;
2280 bundle->dst_mirrors = 0;
2281 bundle->mirror_out = 0;
2284 if (!bundle->name || strcmp(s->name, bundle->name)) {
2286 bundle->name = xstrdup(s->name);
2291 if (!bundle->lacp) {
2292 ofproto->backer->need_revalidate = REV_RECONFIGURE;
2293 bundle->lacp = lacp_create();
2295 lacp_configure(bundle->lacp, s->lacp);
2297 lacp_destroy(bundle->lacp);
2298 bundle->lacp = NULL;
2301 /* Update set of ports. */
2303 for (i = 0; i < s->n_slaves; i++) {
2304 if (!bundle_add_port(bundle, s->slaves[i],
2305 s->lacp ? &s->lacp_slaves[i] : NULL)) {
2309 if (!ok || list_size(&bundle->ports) != s->n_slaves) {
2310 struct ofport_dpif *next_port;
2312 LIST_FOR_EACH_SAFE (port, next_port, bundle_node, &bundle->ports) {
2313 for (i = 0; i < s->n_slaves; i++) {
2314 if (s->slaves[i] == port->up.ofp_port) {
2319 bundle_del_port(port);
2323 ovs_assert(list_size(&bundle->ports) <= s->n_slaves);
2325 if (list_is_empty(&bundle->ports)) {
2326 bundle_destroy(bundle);
2330 /* Set VLAN tagging mode */
2331 if (s->vlan_mode != bundle->vlan_mode
2332 || s->use_priority_tags != bundle->use_priority_tags) {
2333 bundle->vlan_mode = s->vlan_mode;
2334 bundle->use_priority_tags = s->use_priority_tags;
2339 vlan = (s->vlan_mode == PORT_VLAN_TRUNK ? -1
2340 : s->vlan >= 0 && s->vlan <= 4095 ? s->vlan
2342 if (vlan != bundle->vlan) {
2343 bundle->vlan = vlan;
2347 /* Get trunked VLANs. */
2348 switch (s->vlan_mode) {
2349 case PORT_VLAN_ACCESS:
2353 case PORT_VLAN_TRUNK:
2354 trunks = CONST_CAST(unsigned long *, s->trunks);
2357 case PORT_VLAN_NATIVE_UNTAGGED:
2358 case PORT_VLAN_NATIVE_TAGGED:
2359 if (vlan != 0 && (!s->trunks
2360 || !bitmap_is_set(s->trunks, vlan)
2361 || bitmap_is_set(s->trunks, 0))) {
2362 /* Force trunking the native VLAN and prohibit trunking VLAN 0. */
2364 trunks = bitmap_clone(s->trunks, 4096);
2366 trunks = bitmap_allocate1(4096);
2368 bitmap_set1(trunks, vlan);
2369 bitmap_set0(trunks, 0);
2371 trunks = CONST_CAST(unsigned long *, s->trunks);
2378 if (!vlan_bitmap_equal(trunks, bundle->trunks)) {
2379 free(bundle->trunks);
2380 if (trunks == s->trunks) {
2381 bundle->trunks = vlan_bitmap_clone(trunks);
2383 bundle->trunks = trunks;
2388 if (trunks != s->trunks) {
2393 if (!list_is_short(&bundle->ports)) {
2394 bundle->ofproto->has_bonded_bundles = true;
2396 if (bond_reconfigure(bundle->bond, s->bond)) {
2397 ofproto->backer->need_revalidate = REV_RECONFIGURE;
2400 bundle->bond = bond_create(s->bond);
2401 ofproto->backer->need_revalidate = REV_RECONFIGURE;
2404 LIST_FOR_EACH (port, bundle_node, &bundle->ports) {
2405 bond_slave_register(bundle->bond, port, port->up.netdev);
2408 bond_destroy(bundle->bond);
2409 bundle->bond = NULL;
2412 /* If we changed something that would affect MAC learning, un-learn
2413 * everything on this port and force flow revalidation. */
2415 bundle_flush_macs(bundle, false);
2422 bundle_remove(struct ofport *port_)
2424 struct ofport_dpif *port = ofport_dpif_cast(port_);
2425 struct ofbundle *bundle = port->bundle;
2428 bundle_del_port(port);
2429 if (list_is_empty(&bundle->ports)) {
2430 bundle_destroy(bundle);
2431 } else if (list_is_short(&bundle->ports)) {
2432 bond_destroy(bundle->bond);
2433 bundle->bond = NULL;
2439 send_pdu_cb(void *port_, const void *pdu, size_t pdu_size)
2441 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 10);
2442 struct ofport_dpif *port = port_;
2443 uint8_t ea[ETH_ADDR_LEN];
2446 error = netdev_get_etheraddr(port->up.netdev, ea);
2448 struct ofpbuf packet;
2451 ofpbuf_init(&packet, 0);
2452 packet_pdu = eth_compose(&packet, eth_addr_lacp, ea, ETH_TYPE_LACP,
2454 memcpy(packet_pdu, pdu, pdu_size);
2456 send_packet(port, &packet);
2457 ofpbuf_uninit(&packet);
2459 VLOG_ERR_RL(&rl, "port %s: cannot obtain Ethernet address of iface "
2460 "%s (%s)", port->bundle->name,
2461 netdev_get_name(port->up.netdev), strerror(error));
2466 bundle_send_learning_packets(struct ofbundle *bundle)
2468 struct ofproto_dpif *ofproto = bundle->ofproto;
2469 int error, n_packets, n_errors;
2470 struct mac_entry *e;
2472 error = n_packets = n_errors = 0;
2473 LIST_FOR_EACH (e, lru_node, &ofproto->ml->lrus) {
2474 if (e->port.p != bundle) {
2475 struct ofpbuf *learning_packet;
2476 struct ofport_dpif *port;
2480 /* The assignment to "port" is unnecessary but makes "grep"ing for
2481 * struct ofport_dpif more effective. */
2482 learning_packet = bond_compose_learning_packet(bundle->bond,
2486 ret = send_packet(port, learning_packet);
2487 ofpbuf_delete(learning_packet);
2497 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
2498 VLOG_WARN_RL(&rl, "bond %s: %d errors sending %d gratuitous learning "
2499 "packets, last error was: %s",
2500 bundle->name, n_errors, n_packets, strerror(error));
2502 VLOG_DBG("bond %s: sent %d gratuitous learning packets",
2503 bundle->name, n_packets);
2508 bundle_run(struct ofbundle *bundle)
2511 lacp_run(bundle->lacp, send_pdu_cb);
2514 struct ofport_dpif *port;
2516 LIST_FOR_EACH (port, bundle_node, &bundle->ports) {
2517 bond_slave_set_may_enable(bundle->bond, port, port->may_enable);
2520 bond_run(bundle->bond, &bundle->ofproto->backer->revalidate_set,
2521 lacp_status(bundle->lacp));
2522 if (bond_should_send_learning_packets(bundle->bond)) {
2523 bundle_send_learning_packets(bundle);
2529 bundle_wait(struct ofbundle *bundle)
2532 lacp_wait(bundle->lacp);
2535 bond_wait(bundle->bond);
2542 mirror_scan(struct ofproto_dpif *ofproto)
2546 for (idx = 0; idx < MAX_MIRRORS; idx++) {
2547 if (!ofproto->mirrors[idx]) {
2554 static struct ofmirror *
2555 mirror_lookup(struct ofproto_dpif *ofproto, void *aux)
2559 for (i = 0; i < MAX_MIRRORS; i++) {
2560 struct ofmirror *mirror = ofproto->mirrors[i];
2561 if (mirror && mirror->aux == aux) {
2569 /* Update the 'dup_mirrors' member of each of the ofmirrors in 'ofproto'. */
2571 mirror_update_dups(struct ofproto_dpif *ofproto)
2575 for (i = 0; i < MAX_MIRRORS; i++) {
2576 struct ofmirror *m = ofproto->mirrors[i];
2579 m->dup_mirrors = MIRROR_MASK_C(1) << i;
2583 for (i = 0; i < MAX_MIRRORS; i++) {
2584 struct ofmirror *m1 = ofproto->mirrors[i];
2591 for (j = i + 1; j < MAX_MIRRORS; j++) {
2592 struct ofmirror *m2 = ofproto->mirrors[j];
2594 if (m2 && m1->out == m2->out && m1->out_vlan == m2->out_vlan) {
2595 m1->dup_mirrors |= MIRROR_MASK_C(1) << j;
2596 m2->dup_mirrors |= m1->dup_mirrors;
2603 mirror_set(struct ofproto *ofproto_, void *aux,
2604 const struct ofproto_mirror_settings *s)
2606 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
2607 mirror_mask_t mirror_bit;
2608 struct ofbundle *bundle;
2609 struct ofmirror *mirror;
2610 struct ofbundle *out;
2611 struct hmapx srcs; /* Contains "struct ofbundle *"s. */
2612 struct hmapx dsts; /* Contains "struct ofbundle *"s. */
2615 mirror = mirror_lookup(ofproto, aux);
2617 mirror_destroy(mirror);
2623 idx = mirror_scan(ofproto);
2625 VLOG_WARN("bridge %s: maximum of %d port mirrors reached, "
2627 ofproto->up.name, MAX_MIRRORS, s->name);
2631 mirror = ofproto->mirrors[idx] = xzalloc(sizeof *mirror);
2632 mirror->ofproto = ofproto;
2635 mirror->out_vlan = -1;
2636 mirror->name = NULL;
2639 if (!mirror->name || strcmp(s->name, mirror->name)) {
2641 mirror->name = xstrdup(s->name);
2644 /* Get the new configuration. */
2645 if (s->out_bundle) {
2646 out = bundle_lookup(ofproto, s->out_bundle);
2648 mirror_destroy(mirror);
2654 out_vlan = s->out_vlan;
2656 bundle_lookup_multiple(ofproto, s->srcs, s->n_srcs, &srcs);
2657 bundle_lookup_multiple(ofproto, s->dsts, s->n_dsts, &dsts);
2659 /* If the configuration has not changed, do nothing. */
2660 if (hmapx_equals(&srcs, &mirror->srcs)
2661 && hmapx_equals(&dsts, &mirror->dsts)
2662 && vlan_bitmap_equal(mirror->vlans, s->src_vlans)
2663 && mirror->out == out
2664 && mirror->out_vlan == out_vlan)
2666 hmapx_destroy(&srcs);
2667 hmapx_destroy(&dsts);
2671 hmapx_swap(&srcs, &mirror->srcs);
2672 hmapx_destroy(&srcs);
2674 hmapx_swap(&dsts, &mirror->dsts);
2675 hmapx_destroy(&dsts);
2677 free(mirror->vlans);
2678 mirror->vlans = vlan_bitmap_clone(s->src_vlans);
2681 mirror->out_vlan = out_vlan;
2683 /* Update bundles. */
2684 mirror_bit = MIRROR_MASK_C(1) << mirror->idx;
2685 HMAP_FOR_EACH (bundle, hmap_node, &mirror->ofproto->bundles) {
2686 if (hmapx_contains(&mirror->srcs, bundle)) {
2687 bundle->src_mirrors |= mirror_bit;
2689 bundle->src_mirrors &= ~mirror_bit;
2692 if (hmapx_contains(&mirror->dsts, bundle)) {
2693 bundle->dst_mirrors |= mirror_bit;
2695 bundle->dst_mirrors &= ~mirror_bit;
2698 if (mirror->out == bundle) {
2699 bundle->mirror_out |= mirror_bit;
2701 bundle->mirror_out &= ~mirror_bit;
2705 ofproto->backer->need_revalidate = REV_RECONFIGURE;
2706 ofproto->has_mirrors = true;
2707 mac_learning_flush(ofproto->ml,
2708 &ofproto->backer->revalidate_set);
2709 mirror_update_dups(ofproto);
2715 mirror_destroy(struct ofmirror *mirror)
2717 struct ofproto_dpif *ofproto;
2718 mirror_mask_t mirror_bit;
2719 struct ofbundle *bundle;
2726 ofproto = mirror->ofproto;
2727 ofproto->backer->need_revalidate = REV_RECONFIGURE;
2728 mac_learning_flush(ofproto->ml, &ofproto->backer->revalidate_set);
2730 mirror_bit = MIRROR_MASK_C(1) << mirror->idx;
2731 HMAP_FOR_EACH (bundle, hmap_node, &ofproto->bundles) {
2732 bundle->src_mirrors &= ~mirror_bit;
2733 bundle->dst_mirrors &= ~mirror_bit;
2734 bundle->mirror_out &= ~mirror_bit;
2737 hmapx_destroy(&mirror->srcs);
2738 hmapx_destroy(&mirror->dsts);
2739 free(mirror->vlans);
2741 ofproto->mirrors[mirror->idx] = NULL;
2745 mirror_update_dups(ofproto);
2747 ofproto->has_mirrors = false;
2748 for (i = 0; i < MAX_MIRRORS; i++) {
2749 if (ofproto->mirrors[i]) {
2750 ofproto->has_mirrors = true;
2757 mirror_get_stats(struct ofproto *ofproto_, void *aux,
2758 uint64_t *packets, uint64_t *bytes)
2760 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
2761 struct ofmirror *mirror = mirror_lookup(ofproto, aux);
2764 *packets = *bytes = UINT64_MAX;
2770 *packets = mirror->packet_count;
2771 *bytes = mirror->byte_count;
2777 set_flood_vlans(struct ofproto *ofproto_, unsigned long *flood_vlans)
2779 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
2780 if (mac_learning_set_flood_vlans(ofproto->ml, flood_vlans)) {
2781 mac_learning_flush(ofproto->ml, &ofproto->backer->revalidate_set);
2787 is_mirror_output_bundle(const struct ofproto *ofproto_, void *aux)
2789 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
2790 struct ofbundle *bundle = bundle_lookup(ofproto, aux);
2791 return bundle && bundle->mirror_out != 0;
2795 forward_bpdu_changed(struct ofproto *ofproto_)
2797 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
2798 ofproto->backer->need_revalidate = REV_RECONFIGURE;
2802 set_mac_table_config(struct ofproto *ofproto_, unsigned int idle_time,
2805 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
2806 mac_learning_set_idle_time(ofproto->ml, idle_time);
2807 mac_learning_set_max_entries(ofproto->ml, max_entries);
2812 struct ofport_dpif *
2813 get_ofp_port(const struct ofproto_dpif *ofproto, uint16_t ofp_port)
2815 struct ofport *ofport = ofproto_get_port(&ofproto->up, ofp_port);
2816 return ofport ? ofport_dpif_cast(ofport) : NULL;
2819 struct ofport_dpif *
2820 get_odp_port(const struct ofproto_dpif *ofproto, uint32_t odp_port)
2822 struct ofport_dpif *port = odp_port_to_ofport(ofproto->backer, odp_port);
2823 return port && &ofproto->up == port->up.ofproto ? port : NULL;
2827 ofproto_port_from_dpif_port(struct ofproto_dpif *ofproto,
2828 struct ofproto_port *ofproto_port,
2829 struct dpif_port *dpif_port)
2831 ofproto_port->name = dpif_port->name;
2832 ofproto_port->type = dpif_port->type;
2833 ofproto_port->ofp_port = odp_port_to_ofp_port(ofproto, dpif_port->port_no);
2836 struct ofport_dpif *
2837 ofport_get_peer(const struct ofport_dpif *ofport_dpif)
2839 const struct ofproto_dpif *ofproto;
2840 const struct dpif_backer *backer;
2843 peer = netdev_vport_patch_peer(ofport_dpif->up.netdev);
2848 backer = ofproto_dpif_cast(ofport_dpif->up.ofproto)->backer;
2849 HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
2850 struct ofport *ofport;
2852 if (ofproto->backer != backer) {
2856 ofport = shash_find_data(&ofproto->up.port_by_name, peer);
2858 return ofport_dpif_cast(ofport);
2865 port_run_fast(struct ofport_dpif *ofport)
2867 if (ofport->cfm && cfm_should_send_ccm(ofport->cfm)) {
2868 struct ofpbuf packet;
2870 ofpbuf_init(&packet, 0);
2871 cfm_compose_ccm(ofport->cfm, &packet, ofport->up.pp.hw_addr);
2872 send_packet(ofport, &packet);
2873 ofpbuf_uninit(&packet);
2876 if (ofport->bfd && bfd_should_send_packet(ofport->bfd)) {
2877 struct ofpbuf packet;
2879 ofpbuf_init(&packet, 0);
2880 bfd_put_packet(ofport->bfd, &packet, ofport->up.pp.hw_addr);
2881 send_packet(ofport, &packet);
2882 ofpbuf_uninit(&packet);
2887 port_run(struct ofport_dpif *ofport)
2889 long long int carrier_seq = netdev_get_carrier_resets(ofport->up.netdev);
2890 bool carrier_changed = carrier_seq != ofport->carrier_seq;
2891 bool enable = netdev_get_carrier(ofport->up.netdev);
2893 ofport->carrier_seq = carrier_seq;
2895 port_run_fast(ofport);
2898 int cfm_opup = cfm_get_opup(ofport->cfm);
2900 cfm_run(ofport->cfm);
2901 enable = enable && !cfm_get_fault(ofport->cfm);
2903 if (cfm_opup >= 0) {
2904 enable = enable && cfm_opup;
2909 bfd_run(ofport->bfd);
2910 enable = enable && bfd_forwarding(ofport->bfd);
2913 if (ofport->bundle) {
2914 enable = enable && lacp_slave_may_enable(ofport->bundle->lacp, ofport);
2915 if (carrier_changed) {
2916 lacp_slave_carrier_changed(ofport->bundle->lacp, ofport);
2920 if (ofport->may_enable != enable) {
2921 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport->up.ofproto);
2923 if (ofproto->has_bundle_action) {
2924 ofproto->backer->need_revalidate = REV_PORT_TOGGLED;
2928 ofport->may_enable = enable;
2932 port_wait(struct ofport_dpif *ofport)
2935 cfm_wait(ofport->cfm);
2939 bfd_wait(ofport->bfd);
2944 port_query_by_name(const struct ofproto *ofproto_, const char *devname,
2945 struct ofproto_port *ofproto_port)
2947 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
2948 struct dpif_port dpif_port;
2951 if (sset_contains(&ofproto->ghost_ports, devname)) {
2952 const char *type = netdev_get_type_from_name(devname);
2954 /* We may be called before ofproto->up.port_by_name is populated with
2955 * the appropriate ofport. For this reason, we must get the name and
2956 * type from the netdev layer directly. */
2958 const struct ofport *ofport;
2960 ofport = shash_find_data(&ofproto->up.port_by_name, devname);
2961 ofproto_port->ofp_port = ofport ? ofport->ofp_port : OFPP_NONE;
2962 ofproto_port->name = xstrdup(devname);
2963 ofproto_port->type = xstrdup(type);
2969 if (!sset_contains(&ofproto->ports, devname)) {
2972 error = dpif_port_query_by_name(ofproto->backer->dpif,
2973 devname, &dpif_port);
2975 ofproto_port_from_dpif_port(ofproto, ofproto_port, &dpif_port);
2981 port_add(struct ofproto *ofproto_, struct netdev *netdev)
2983 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
2984 const char *devname = netdev_get_name(netdev);
2985 char namebuf[NETDEV_VPORT_NAME_BUFSIZE];
2986 const char *dp_port_name;
2988 if (netdev_vport_is_patch(netdev)) {
2989 sset_add(&ofproto->ghost_ports, netdev_get_name(netdev));
2993 dp_port_name = netdev_vport_get_dpif_port(netdev, namebuf, sizeof namebuf);
2994 if (!dpif_port_exists(ofproto->backer->dpif, dp_port_name)) {
2995 uint32_t port_no = UINT32_MAX;
2998 error = dpif_port_add(ofproto->backer->dpif, netdev, &port_no);
3002 if (netdev_get_tunnel_config(netdev)) {
3003 simap_put(&ofproto->backer->tnl_backers, dp_port_name, port_no);
3007 if (netdev_get_tunnel_config(netdev)) {
3008 sset_add(&ofproto->ghost_ports, devname);
3010 sset_add(&ofproto->ports, devname);
3016 port_del(struct ofproto *ofproto_, uint16_t ofp_port)
3018 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
3019 struct ofport_dpif *ofport = get_ofp_port(ofproto, ofp_port);
3026 sset_find_and_delete(&ofproto->ghost_ports,
3027 netdev_get_name(ofport->up.netdev));
3028 ofproto->backer->need_revalidate = REV_RECONFIGURE;
3029 if (!ofport->tnl_port) {
3030 error = dpif_port_del(ofproto->backer->dpif, ofport->odp_port);
3032 /* The caller is going to close ofport->up.netdev. If this is a
3033 * bonded port, then the bond is using that netdev, so remove it
3034 * from the bond. The client will need to reconfigure everything
3035 * after deleting ports, so then the slave will get re-added. */
3036 bundle_remove(&ofport->up);
3043 port_get_stats(const struct ofport *ofport_, struct netdev_stats *stats)
3045 struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
3050 error = netdev_get_stats(ofport->up.netdev, stats);
3052 if (!error && ofport_->ofp_port == OFPP_LOCAL) {
3053 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport->up.ofproto);
3055 /* ofproto->stats.tx_packets represents packets that we created
3056 * internally and sent to some port (e.g. packets sent with
3057 * send_packet()). Account for them as if they had come from
3058 * OFPP_LOCAL and got forwarded. */
3060 if (stats->rx_packets != UINT64_MAX) {
3061 stats->rx_packets += ofproto->stats.tx_packets;
3064 if (stats->rx_bytes != UINT64_MAX) {
3065 stats->rx_bytes += ofproto->stats.tx_bytes;
3068 /* ofproto->stats.rx_packets represents packets that were received on
3069 * some port and we processed internally and dropped (e.g. STP).
3070 * Account for them as if they had been forwarded to OFPP_LOCAL. */
3072 if (stats->tx_packets != UINT64_MAX) {
3073 stats->tx_packets += ofproto->stats.rx_packets;
3076 if (stats->tx_bytes != UINT64_MAX) {
3077 stats->tx_bytes += ofproto->stats.rx_bytes;
3084 struct port_dump_state {
3089 struct ofproto_port port;
3094 port_dump_start(const struct ofproto *ofproto_ OVS_UNUSED, void **statep)
3096 *statep = xzalloc(sizeof(struct port_dump_state));
3101 port_dump_next(const struct ofproto *ofproto_, void *state_,
3102 struct ofproto_port *port)
3104 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
3105 struct port_dump_state *state = state_;
3106 const struct sset *sset;
3107 struct sset_node *node;
3109 if (state->has_port) {
3110 ofproto_port_destroy(&state->port);
3111 state->has_port = false;
3113 sset = state->ghost ? &ofproto->ghost_ports : &ofproto->ports;
3114 while ((node = sset_at_position(sset, &state->bucket, &state->offset))) {
3117 error = port_query_by_name(ofproto_, node->name, &state->port);
3119 *port = state->port;
3120 state->has_port = true;
3122 } else if (error != ENODEV) {
3127 if (!state->ghost) {
3128 state->ghost = true;
3131 return port_dump_next(ofproto_, state_, port);
3138 port_dump_done(const struct ofproto *ofproto_ OVS_UNUSED, void *state_)
3140 struct port_dump_state *state = state_;
3142 if (state->has_port) {
3143 ofproto_port_destroy(&state->port);
3150 port_poll(const struct ofproto *ofproto_, char **devnamep)
3152 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
3154 if (ofproto->port_poll_errno) {
3155 int error = ofproto->port_poll_errno;
3156 ofproto->port_poll_errno = 0;
3160 if (sset_is_empty(&ofproto->port_poll_set)) {
3164 *devnamep = sset_pop(&ofproto->port_poll_set);
3169 port_poll_wait(const struct ofproto *ofproto_)
3171 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
3172 dpif_port_poll_wait(ofproto->backer->dpif);
3176 port_is_lacp_current(const struct ofport *ofport_)
3178 const struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
3179 return (ofport->bundle && ofport->bundle->lacp
3180 ? lacp_slave_is_current(ofport->bundle->lacp, ofport)
3184 /* Upcall handling. */
3186 /* Flow miss batching.
3188 * Some dpifs implement operations faster when you hand them off in a batch.
3189 * To allow batching, "struct flow_miss" queues the dpif-related work needed
3190 * for a given flow. Each "struct flow_miss" corresponds to sending one or
3191 * more packets, plus possibly installing the flow in the dpif.
3193 * So far we only batch the operations that affect flow setup time the most.
3194 * It's possible to batch more than that, but the benefit might be minimal. */
3196 struct hmap_node hmap_node;
3197 struct ofproto_dpif *ofproto;
3199 enum odp_key_fitness key_fitness;
3200 const struct nlattr *key;
3202 struct list packets;
3203 enum dpif_upcall_type upcall_type;
3206 struct flow_miss_op {
3207 struct dpif_op dpif_op;
3209 uint64_t slow_stub[128 / 8]; /* Buffer for compose_slow_path() */
3210 struct xlate_out xout;
3211 bool xout_garbage; /* 'xout' needs to be uninitialized? */
3214 /* Sends an OFPT_PACKET_IN message for 'packet' of type OFPR_NO_MATCH to each
3215 * OpenFlow controller as necessary according to their individual
3216 * configurations. */
3218 send_packet_in_miss(struct ofproto_dpif *ofproto, const struct ofpbuf *packet,
3219 const struct flow *flow)
3221 struct ofputil_packet_in pin;
3223 pin.packet = packet->data;
3224 pin.packet_len = packet->size;
3225 pin.reason = OFPR_NO_MATCH;
3226 pin.controller_id = 0;
3231 pin.send_len = 0; /* not used for flow table misses */
3233 flow_get_metadata(flow, &pin.fmd);
3235 connmgr_send_packet_in(ofproto->up.connmgr, &pin);
3238 static struct flow_miss *
3239 flow_miss_find(struct hmap *todo, const struct ofproto_dpif *ofproto,
3240 const struct flow *flow, uint32_t hash)
3242 struct flow_miss *miss;
3244 HMAP_FOR_EACH_WITH_HASH (miss, hmap_node, hash, todo) {
3245 if (miss->ofproto == ofproto && flow_equal(&miss->flow, flow)) {
3253 /* Partially Initializes 'op' as an "execute" operation for 'miss' and
3254 * 'packet'. The caller must initialize op->actions and op->actions_len. If
3255 * 'miss' is associated with a subfacet the caller must also initialize the
3256 * returned op->subfacet, and if anything needs to be freed after processing
3257 * the op, the caller must initialize op->garbage also. */
3259 init_flow_miss_execute_op(struct flow_miss *miss, struct ofpbuf *packet,
3260 struct flow_miss_op *op)
3262 if (miss->flow.in_port
3263 != vsp_realdev_to_vlandev(miss->ofproto, miss->flow.in_port,
3264 miss->flow.vlan_tci)) {
3265 /* This packet was received on a VLAN splinter port. We
3266 * added a VLAN to the packet to make the packet resemble
3267 * the flow, but the actions were composed assuming that
3268 * the packet contained no VLAN. So, we must remove the
3269 * VLAN header from the packet before trying to execute the
3271 eth_pop_vlan(packet);
3274 op->xout_garbage = false;
3275 op->dpif_op.type = DPIF_OP_EXECUTE;
3276 op->dpif_op.u.execute.key = miss->key;
3277 op->dpif_op.u.execute.key_len = miss->key_len;
3278 op->dpif_op.u.execute.packet = packet;
3281 /* Helper for handle_flow_miss_without_facet() and
3282 * handle_flow_miss_with_facet(). */
3284 handle_flow_miss_common(struct rule_dpif *rule,
3285 struct ofpbuf *packet, const struct flow *flow)
3287 struct ofproto_dpif *ofproto = ofproto_dpif_cast(rule->up.ofproto);
3289 if (rule->up.cr.priority == FAIL_OPEN_PRIORITY) {
3291 * Extra-special case for fail-open mode.
3293 * We are in fail-open mode and the packet matched the fail-open
3294 * rule, but we are connected to a controller too. We should send
3295 * the packet up to the controller in the hope that it will try to
3296 * set up a flow and thereby allow us to exit fail-open.
3298 * See the top-level comment in fail-open.c for more information.
3300 send_packet_in_miss(ofproto, packet, flow);
3304 /* Figures out whether a flow that missed in 'ofproto', whose details are in
3305 * 'miss' masked by 'wc', is likely to be worth tracking in detail in userspace
3306 * and (usually) installing a datapath flow. The answer is usually "yes" (a
3307 * return value of true). However, for short flows the cost of bookkeeping is
3308 * much higher than the benefits, so when the datapath holds a large number of
3309 * flows we impose some heuristics to decide which flows are likely to be worth
3312 flow_miss_should_make_facet(struct flow_miss *miss, struct flow_wildcards *wc)
3314 struct dpif_backer *backer = miss->ofproto->backer;
3317 if (!backer->governor) {
3320 n_subfacets = hmap_count(&backer->subfacets);
3321 if (n_subfacets * 2 <= flow_eviction_threshold) {
3325 backer->governor = governor_create();
3328 hash = flow_hash_in_wildcards(&miss->flow, wc, 0);
3329 return governor_should_install_flow(backer->governor, hash,
3330 list_size(&miss->packets));
3333 /* Handles 'miss' without creating a facet or subfacet or creating any datapath
3334 * flow. 'miss->flow' must have matched 'rule' and been xlated into 'xout'.
3335 * May add an "execute" operation to 'ops' and increment '*n_ops'. */
3337 handle_flow_miss_without_facet(struct rule_dpif *rule, struct xlate_out *xout,
3338 struct flow_miss *miss,
3339 struct flow_miss_op *ops, size_t *n_ops)
3341 struct ofpbuf *packet;
3343 LIST_FOR_EACH (packet, list_node, &miss->packets) {
3345 COVERAGE_INC(facet_suppress);
3347 handle_flow_miss_common(rule, packet, &miss->flow);
3350 struct xlate_in xin;
3352 xlate_in_init(&xin, miss->ofproto, &miss->flow, rule, 0, packet);
3353 xlate_actions_for_side_effects(&xin);
3356 if (xout->odp_actions.size) {
3357 struct flow_miss_op *op = &ops[*n_ops];
3358 struct dpif_execute *execute = &op->dpif_op.u.execute;
3360 init_flow_miss_execute_op(miss, packet, op);
3361 xlate_out_copy(&op->xout, xout);
3362 execute->actions = op->xout.odp_actions.data;
3363 execute->actions_len = op->xout.odp_actions.size;
3364 op->xout_garbage = true;
3371 /* Handles 'miss', which matches 'facet'. May add any required datapath
3372 * operations to 'ops', incrementing '*n_ops' for each new op.
3374 * All of the packets in 'miss' are considered to have arrived at time 'now'.
3375 * This is really important only for new facets: if we just called time_msec()
3376 * here, then the new subfacet or its packets could look (occasionally) as
3377 * though it was used some time after the facet was used. That can make a
3378 * one-packet flow look like it has a nonzero duration, which looks odd in
3379 * e.g. NetFlow statistics.
3381 * If non-null, 'stats' will be folded into 'facet'. */
3383 handle_flow_miss_with_facet(struct flow_miss *miss, struct facet *facet,
3384 long long int now, struct dpif_flow_stats *stats,
3385 struct flow_miss_op *ops, size_t *n_ops)
3387 struct ofproto_dpif *ofproto = ofproto_dpif_cast(facet->rule->up.ofproto);
3388 enum subfacet_path want_path;
3389 struct subfacet *subfacet;
3390 struct ofpbuf *packet;
3392 subfacet = subfacet_create(facet, miss, now);
3393 want_path = facet->xout.slow ? SF_SLOW_PATH : SF_FAST_PATH;
3395 subfacet_update_stats(subfacet, stats);
3398 LIST_FOR_EACH (packet, list_node, &miss->packets) {
3399 struct flow_miss_op *op = &ops[*n_ops];
3401 handle_flow_miss_common(facet->rule, packet, &miss->flow);
3403 if (want_path != SF_FAST_PATH) {
3404 struct xlate_in xin;
3406 xlate_in_init(&xin, ofproto, &miss->flow, facet->rule, 0, packet);
3407 xlate_actions_for_side_effects(&xin);
3410 if (facet->xout.odp_actions.size) {
3411 struct dpif_execute *execute = &op->dpif_op.u.execute;
3413 init_flow_miss_execute_op(miss, packet, op);
3414 execute->actions = facet->xout.odp_actions.data,
3415 execute->actions_len = facet->xout.odp_actions.size;
3420 if (miss->upcall_type == DPIF_UC_MISS || subfacet->path != want_path) {
3421 struct flow_miss_op *op = &ops[(*n_ops)++];
3422 struct dpif_flow_put *put = &op->dpif_op.u.flow_put;
3424 subfacet->path = want_path;
3426 op->xout_garbage = false;
3427 op->dpif_op.type = DPIF_OP_FLOW_PUT;
3428 put->flags = DPIF_FP_CREATE | DPIF_FP_MODIFY;
3429 put->key = miss->key;
3430 put->key_len = miss->key_len;
3431 if (want_path == SF_FAST_PATH) {
3432 put->actions = facet->xout.odp_actions.data;
3433 put->actions_len = facet->xout.odp_actions.size;
3435 compose_slow_path(ofproto, &miss->flow, facet->xout.slow,
3436 op->slow_stub, sizeof op->slow_stub,
3437 &put->actions, &put->actions_len);
3443 /* Handles flow miss 'miss'. May add any required datapath operations
3444 * to 'ops', incrementing '*n_ops' for each new op. */
3446 handle_flow_miss(struct flow_miss *miss, struct flow_miss_op *ops,
3449 struct ofproto_dpif *ofproto = miss->ofproto;
3450 struct dpif_flow_stats stats__;
3451 struct dpif_flow_stats *stats = &stats__;
3452 struct ofpbuf *packet;
3453 struct facet *facet;
3457 memset(stats, 0, sizeof *stats);
3459 LIST_FOR_EACH (packet, list_node, &miss->packets) {
3460 stats->tcp_flags |= packet_get_tcp_flags(packet, &miss->flow);
3461 stats->n_bytes += packet->size;
3465 facet = facet_lookup_valid(ofproto, &miss->flow);
3467 struct flow_wildcards wc;
3468 struct rule_dpif *rule;
3469 struct xlate_out xout;
3470 struct xlate_in xin;
3472 flow_wildcards_init_catchall(&wc);
3473 rule = rule_dpif_lookup(ofproto, &miss->flow, &wc);
3474 rule_credit_stats(rule, stats);
3476 xlate_in_init(&xin, ofproto, &miss->flow, rule, stats->tcp_flags,
3478 xin.resubmit_stats = stats;
3479 xin.may_learn = true;
3480 xlate_actions(&xin, &xout);
3481 flow_wildcards_or(&xout.wc, &xout.wc, &wc);
3483 /* There does not exist a bijection between 'struct flow' and datapath
3484 * flow keys with fitness ODP_FIT_TO_LITTLE. This breaks a fundamental
3485 * assumption used throughout the facet and subfacet handling code.
3486 * Since we have to handle these misses in userspace anyway, we simply
3487 * skip facet creation, avoiding the problem altogether. */
3488 if (miss->key_fitness == ODP_FIT_TOO_LITTLE
3489 || !flow_miss_should_make_facet(miss, &xout.wc)) {
3490 handle_flow_miss_without_facet(rule, &xout, miss, ops, n_ops);
3494 facet = facet_create(miss, rule, &xout, stats);
3497 handle_flow_miss_with_facet(miss, facet, now, stats, ops, n_ops);
3500 static struct drop_key *
3501 drop_key_lookup(const struct dpif_backer *backer, const struct nlattr *key,
3504 struct drop_key *drop_key;
3506 HMAP_FOR_EACH_WITH_HASH (drop_key, hmap_node, hash_bytes(key, key_len, 0),
3507 &backer->drop_keys) {
3508 if (drop_key->key_len == key_len
3509 && !memcmp(drop_key->key, key, key_len)) {
3517 drop_key_clear(struct dpif_backer *backer)
3519 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 15);
3520 struct drop_key *drop_key, *next;
3522 HMAP_FOR_EACH_SAFE (drop_key, next, hmap_node, &backer->drop_keys) {
3525 error = dpif_flow_del(backer->dpif, drop_key->key, drop_key->key_len,
3527 if (error && !VLOG_DROP_WARN(&rl)) {
3528 struct ds ds = DS_EMPTY_INITIALIZER;
3529 odp_flow_key_format(drop_key->key, drop_key->key_len, &ds);
3530 VLOG_WARN("Failed to delete drop key (%s) (%s)", strerror(error),
3535 hmap_remove(&backer->drop_keys, &drop_key->hmap_node);
3536 free(drop_key->key);
3541 /* Given a datpath, packet, and flow metadata ('backer', 'packet', and 'key'
3542 * respectively), populates 'flow' with the result of odp_flow_key_to_flow().
3543 * Optionally, if nonnull, populates 'fitnessp' with the fitness of 'flow' as
3544 * returned by odp_flow_key_to_flow(). Also, optionally populates 'ofproto'
3545 * with the ofproto_dpif, and 'odp_in_port' with the datapath in_port, that
3546 * 'packet' ingressed.
3548 * If 'ofproto' is nonnull, requires 'flow''s in_port to exist. Otherwise sets
3549 * 'flow''s in_port to OFPP_NONE.
3551 * This function does post-processing on data returned from
3552 * odp_flow_key_to_flow() to help make VLAN splinters transparent to the rest
3553 * of the upcall processing logic. In particular, if the extracted in_port is
3554 * a VLAN splinter port, it replaces flow->in_port by the "real" port, sets
3555 * flow->vlan_tci correctly for the VLAN of the VLAN splinter port, and pushes
3556 * a VLAN header onto 'packet' (if it is nonnull).
3558 * Similarly, this function also includes some logic to help with tunnels. It
3559 * may modify 'flow' as necessary to make the tunneling implementation
3560 * transparent to the upcall processing logic.
3562 * Returns 0 if successful, ENODEV if the parsed flow has no associated ofport,
3563 * or some other positive errno if there are other problems. */
3565 ofproto_receive(const struct dpif_backer *backer, struct ofpbuf *packet,
3566 const struct nlattr *key, size_t key_len,
3567 struct flow *flow, enum odp_key_fitness *fitnessp,
3568 struct ofproto_dpif **ofproto, uint32_t *odp_in_port)
3570 const struct ofport_dpif *port;
3571 enum odp_key_fitness fitness;
3574 fitness = odp_flow_key_to_flow(key, key_len, flow);
3575 if (fitness == ODP_FIT_ERROR) {
3581 *odp_in_port = flow->in_port;
3584 port = (tnl_port_should_receive(flow)
3585 ? ofport_dpif_cast(tnl_port_receive(flow))
3586 : odp_port_to_ofport(backer, flow->in_port));
3587 flow->in_port = port ? port->up.ofp_port : OFPP_NONE;
3592 /* XXX: Since the tunnel module is not scoped per backer, for a tunnel port
3593 * it's theoretically possible that we'll receive an ofport belonging to an
3594 * entirely different datapath. In practice, this can't happen because no
3595 * platforms has two separate datapaths which each support tunneling. */
3596 ovs_assert(ofproto_dpif_cast(port->up.ofproto)->backer == backer);
3598 if (vsp_adjust_flow(ofproto_dpif_cast(port->up.ofproto), flow)) {
3600 /* Make the packet resemble the flow, so that it gets sent to
3601 * an OpenFlow controller properly, so that it looks correct
3602 * for sFlow, and so that flow_extract() will get the correct
3603 * vlan_tci if it is called on 'packet'.
3605 * The allocated space inside 'packet' probably also contains
3606 * 'key', that is, both 'packet' and 'key' are probably part of
3607 * a struct dpif_upcall (see the large comment on that
3608 * structure definition), so pushing data on 'packet' is in
3609 * general not a good idea since it could overwrite 'key' or
3610 * free it as a side effect. However, it's OK in this special
3611 * case because we know that 'packet' is inside a Netlink
3612 * attribute: pushing 4 bytes will just overwrite the 4-byte
3613 * "struct nlattr", which is fine since we don't need that
3614 * header anymore. */
3615 eth_push_vlan(packet, flow->vlan_tci);
3617 /* We can't reproduce 'key' from 'flow'. */
3618 fitness = fitness == ODP_FIT_PERFECT ? ODP_FIT_TOO_MUCH : fitness;
3623 *ofproto = ofproto_dpif_cast(port->up.ofproto);
3628 *fitnessp = fitness;
3634 handle_miss_upcalls(struct dpif_backer *backer, struct dpif_upcall *upcalls,
3637 struct dpif_upcall *upcall;
3638 struct flow_miss *miss;
3639 struct flow_miss misses[FLOW_MISS_MAX_BATCH];
3640 struct flow_miss_op flow_miss_ops[FLOW_MISS_MAX_BATCH * 2];
3641 struct dpif_op *dpif_ops[FLOW_MISS_MAX_BATCH * 2];
3651 /* Construct the to-do list.
3653 * This just amounts to extracting the flow from each packet and sticking
3654 * the packets that have the same flow in the same "flow_miss" structure so
3655 * that we can process them together. */
3658 for (upcall = upcalls; upcall < &upcalls[n_upcalls]; upcall++) {
3659 struct flow_miss *miss = &misses[n_misses];
3660 struct flow_miss *existing_miss;
3661 struct ofproto_dpif *ofproto;
3662 uint32_t odp_in_port;
3667 error = ofproto_receive(backer, upcall->packet, upcall->key,
3668 upcall->key_len, &flow, &miss->key_fitness,
3669 &ofproto, &odp_in_port);
3670 if (error == ENODEV) {
3671 struct drop_key *drop_key;
3673 /* Received packet on datapath port for which we couldn't
3674 * associate an ofproto. This can happen if a port is removed
3675 * while traffic is being received. Print a rate-limited message
3676 * in case it happens frequently. Install a drop flow so
3677 * that future packets of the flow are inexpensively dropped
3679 VLOG_INFO_RL(&rl, "received packet on unassociated datapath port "
3680 "%"PRIu32, odp_in_port);
3682 drop_key = drop_key_lookup(backer, upcall->key, upcall->key_len);
3684 drop_key = xmalloc(sizeof *drop_key);
3685 drop_key->key = xmemdup(upcall->key, upcall->key_len);
3686 drop_key->key_len = upcall->key_len;
3688 hmap_insert(&backer->drop_keys, &drop_key->hmap_node,
3689 hash_bytes(drop_key->key, drop_key->key_len, 0));
3690 dpif_flow_put(backer->dpif, DPIF_FP_CREATE | DPIF_FP_MODIFY,
3691 drop_key->key, drop_key->key_len, NULL, 0, NULL);
3699 ofproto->n_missed++;
3700 flow_extract(upcall->packet, flow.skb_priority, flow.skb_mark,
3701 &flow.tunnel, flow.in_port, &miss->flow);
3703 /* Add other packets to a to-do list. */
3704 hash = flow_hash(&miss->flow, 0);
3705 existing_miss = flow_miss_find(&todo, ofproto, &miss->flow, hash);
3706 if (!existing_miss) {
3707 hmap_insert(&todo, &miss->hmap_node, hash);
3708 miss->ofproto = ofproto;
3709 miss->key = upcall->key;
3710 miss->key_len = upcall->key_len;
3711 miss->upcall_type = upcall->type;
3712 list_init(&miss->packets);
3716 miss = existing_miss;
3718 list_push_back(&miss->packets, &upcall->packet->list_node);
3721 /* Process each element in the to-do list, constructing the set of
3722 * operations to batch. */
3724 HMAP_FOR_EACH (miss, hmap_node, &todo) {
3725 handle_flow_miss(miss, flow_miss_ops, &n_ops);
3727 ovs_assert(n_ops <= ARRAY_SIZE(flow_miss_ops));
3729 /* Execute batch. */
3730 for (i = 0; i < n_ops; i++) {
3731 dpif_ops[i] = &flow_miss_ops[i].dpif_op;
3733 dpif_operate(backer->dpif, dpif_ops, n_ops);
3736 for (i = 0; i < n_ops; i++) {
3737 if (flow_miss_ops[i].xout_garbage) {
3738 xlate_out_uninit(&flow_miss_ops[i].xout);
3741 hmap_destroy(&todo);
3744 static enum { SFLOW_UPCALL, MISS_UPCALL, BAD_UPCALL, FLOW_SAMPLE_UPCALL,
3746 classify_upcall(const struct dpif_upcall *upcall)
3748 size_t userdata_len;
3749 union user_action_cookie cookie;
3751 /* First look at the upcall type. */
3752 switch (upcall->type) {
3753 case DPIF_UC_ACTION:
3759 case DPIF_N_UC_TYPES:
3761 VLOG_WARN_RL(&rl, "upcall has unexpected type %"PRIu32, upcall->type);
3765 /* "action" upcalls need a closer look. */
3766 if (!upcall->userdata) {
3767 VLOG_WARN_RL(&rl, "action upcall missing cookie");
3770 userdata_len = nl_attr_get_size(upcall->userdata);
3771 if (userdata_len < sizeof cookie.type
3772 || userdata_len > sizeof cookie) {
3773 VLOG_WARN_RL(&rl, "action upcall cookie has unexpected size %zu",
3777 memset(&cookie, 0, sizeof cookie);
3778 memcpy(&cookie, nl_attr_get(upcall->userdata), userdata_len);
3779 if (userdata_len == sizeof cookie.sflow
3780 && cookie.type == USER_ACTION_COOKIE_SFLOW) {
3781 return SFLOW_UPCALL;
3782 } else if (userdata_len == sizeof cookie.slow_path
3783 && cookie.type == USER_ACTION_COOKIE_SLOW_PATH) {
3785 } else if (userdata_len == sizeof cookie.flow_sample
3786 && cookie.type == USER_ACTION_COOKIE_FLOW_SAMPLE) {
3787 return FLOW_SAMPLE_UPCALL;
3788 } else if (userdata_len == sizeof cookie.ipfix
3789 && cookie.type == USER_ACTION_COOKIE_IPFIX) {
3790 return IPFIX_UPCALL;
3792 VLOG_WARN_RL(&rl, "invalid user cookie of type %"PRIu16
3793 " and size %zu", cookie.type, userdata_len);
3799 handle_sflow_upcall(struct dpif_backer *backer,
3800 const struct dpif_upcall *upcall)
3802 struct ofproto_dpif *ofproto;
3803 union user_action_cookie cookie;
3805 uint32_t odp_in_port;
3807 if (ofproto_receive(backer, upcall->packet, upcall->key, upcall->key_len,
3808 &flow, NULL, &ofproto, &odp_in_port)
3809 || !ofproto->sflow) {
3813 memset(&cookie, 0, sizeof cookie);
3814 memcpy(&cookie, nl_attr_get(upcall->userdata), sizeof cookie.sflow);
3815 dpif_sflow_received(ofproto->sflow, upcall->packet, &flow,
3816 odp_in_port, &cookie);
3820 handle_flow_sample_upcall(struct dpif_backer *backer,
3821 const struct dpif_upcall *upcall)
3823 struct ofproto_dpif *ofproto;
3824 union user_action_cookie cookie;
3827 if (ofproto_receive(backer, upcall->packet, upcall->key, upcall->key_len,
3828 &flow, NULL, &ofproto, NULL)
3829 || !ofproto->ipfix) {
3833 memset(&cookie, 0, sizeof cookie);
3834 memcpy(&cookie, nl_attr_get(upcall->userdata), sizeof cookie.flow_sample);
3836 /* The flow reflects exactly the contents of the packet. Sample
3837 * the packet using it. */
3838 dpif_ipfix_flow_sample(ofproto->ipfix, upcall->packet, &flow,
3839 cookie.flow_sample.collector_set_id,
3840 cookie.flow_sample.probability,
3841 cookie.flow_sample.obs_domain_id,
3842 cookie.flow_sample.obs_point_id);
3846 handle_ipfix_upcall(struct dpif_backer *backer,
3847 const struct dpif_upcall *upcall)
3849 struct ofproto_dpif *ofproto;
3852 if (ofproto_receive(backer, upcall->packet, upcall->key, upcall->key_len,
3853 &flow, NULL, &ofproto, NULL)
3854 || !ofproto->ipfix) {
3858 /* The flow reflects exactly the contents of the packet. Sample
3859 * the packet using it. */
3860 dpif_ipfix_bridge_sample(ofproto->ipfix, upcall->packet, &flow);
3864 handle_upcalls(struct dpif_backer *backer, unsigned int max_batch)
3866 struct dpif_upcall misses[FLOW_MISS_MAX_BATCH];
3867 struct ofpbuf miss_bufs[FLOW_MISS_MAX_BATCH];
3868 uint64_t miss_buf_stubs[FLOW_MISS_MAX_BATCH][4096 / 8];
3873 ovs_assert(max_batch <= FLOW_MISS_MAX_BATCH);
3876 for (n_processed = 0; n_processed < max_batch; n_processed++) {
3877 struct dpif_upcall *upcall = &misses[n_misses];
3878 struct ofpbuf *buf = &miss_bufs[n_misses];
3881 ofpbuf_use_stub(buf, miss_buf_stubs[n_misses],
3882 sizeof miss_buf_stubs[n_misses]);
3883 error = dpif_recv(backer->dpif, upcall, buf);
3889 switch (classify_upcall(upcall)) {
3891 /* Handle it later. */
3896 handle_sflow_upcall(backer, upcall);
3900 case FLOW_SAMPLE_UPCALL:
3901 handle_flow_sample_upcall(backer, upcall);
3906 handle_ipfix_upcall(backer, upcall);
3916 /* Handle deferred MISS_UPCALL processing. */
3917 handle_miss_upcalls(backer, misses, n_misses);
3918 for (i = 0; i < n_misses; i++) {
3919 ofpbuf_uninit(&miss_bufs[i]);
3925 /* Flow expiration. */
3927 static int subfacet_max_idle(const struct dpif_backer *);
3928 static void update_stats(struct dpif_backer *);
3929 static void rule_expire(struct rule_dpif *);
3930 static void expire_subfacets(struct dpif_backer *, int dp_max_idle);
3932 /* This function is called periodically by run(). Its job is to collect
3933 * updates for the flows that have been installed into the datapath, most
3934 * importantly when they last were used, and then use that information to
3935 * expire flows that have not been used recently.
3937 * Returns the number of milliseconds after which it should be called again. */
3939 expire(struct dpif_backer *backer)
3941 struct ofproto_dpif *ofproto;
3945 /* Periodically clear out the drop keys in an effort to keep them
3946 * relatively few. */
3947 drop_key_clear(backer);
3949 /* Update stats for each flow in the backer. */
3950 update_stats(backer);
3952 n_subfacets = hmap_count(&backer->subfacets);
3954 struct subfacet *subfacet;
3955 long long int total, now;
3959 HMAP_FOR_EACH (subfacet, hmap_node, &backer->subfacets) {
3960 total += now - subfacet->created;
3962 backer->avg_subfacet_life += total / n_subfacets;
3964 backer->avg_subfacet_life /= 2;
3966 backer->avg_n_subfacet += n_subfacets;
3967 backer->avg_n_subfacet /= 2;
3969 backer->max_n_subfacet = MAX(backer->max_n_subfacet, n_subfacets);
3971 max_idle = subfacet_max_idle(backer);
3972 expire_subfacets(backer, max_idle);
3974 HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
3975 struct rule *rule, *next_rule;
3977 if (ofproto->backer != backer) {
3981 /* Expire OpenFlow flows whose idle_timeout or hard_timeout
3983 LIST_FOR_EACH_SAFE (rule, next_rule, expirable,
3984 &ofproto->up.expirable) {
3985 rule_expire(rule_dpif_cast(rule));
3988 /* All outstanding data in existing flows has been accounted, so it's a
3989 * good time to do bond rebalancing. */
3990 if (ofproto->has_bonded_bundles) {
3991 struct ofbundle *bundle;
3993 HMAP_FOR_EACH (bundle, hmap_node, &ofproto->bundles) {
3995 bond_rebalance(bundle->bond, &backer->revalidate_set);
4001 return MIN(max_idle, 1000);
4004 /* Updates flow table statistics given that the datapath just reported 'stats'
4005 * as 'subfacet''s statistics. */
4007 update_subfacet_stats(struct subfacet *subfacet,
4008 const struct dpif_flow_stats *stats)
4010 struct facet *facet = subfacet->facet;
4011 struct ofproto_dpif *ofproto = ofproto_dpif_cast(facet->rule->up.ofproto);
4012 struct dpif_flow_stats diff;
4014 diff.tcp_flags = stats->tcp_flags;
4015 diff.used = stats->used;
4017 if (stats->n_packets >= subfacet->dp_packet_count) {
4018 diff.n_packets = stats->n_packets - subfacet->dp_packet_count;
4020 VLOG_WARN_RL(&rl, "unexpected packet count from the datapath");
4024 if (stats->n_bytes >= subfacet->dp_byte_count) {
4025 diff.n_bytes = stats->n_bytes - subfacet->dp_byte_count;
4027 VLOG_WARN_RL(&rl, "unexpected byte count from datapath");
4031 ofproto->n_hit += diff.n_packets;
4032 subfacet->dp_packet_count = stats->n_packets;
4033 subfacet->dp_byte_count = stats->n_bytes;
4034 subfacet_update_stats(subfacet, &diff);
4036 if (facet->accounted_bytes < facet->byte_count) {
4038 facet_account(facet);
4039 facet->accounted_bytes = facet->byte_count;
4043 /* 'key' with length 'key_len' bytes is a flow in 'dpif' that we know nothing
4044 * about, or a flow that shouldn't be installed but was anyway. Delete it. */
4046 delete_unexpected_flow(struct dpif_backer *backer,
4047 const struct nlattr *key, size_t key_len)
4049 if (!VLOG_DROP_WARN(&rl)) {
4053 odp_flow_key_format(key, key_len, &s);
4054 VLOG_WARN("unexpected flow: %s", ds_cstr(&s));
4058 COVERAGE_INC(facet_unexpected);
4059 dpif_flow_del(backer->dpif, key, key_len, NULL);
4062 /* Update 'packet_count', 'byte_count', and 'used' members of installed facets.
4064 * This function also pushes statistics updates to rules which each facet
4065 * resubmits into. Generally these statistics will be accurate. However, if a
4066 * facet changes the rule it resubmits into at some time in between
4067 * update_stats() runs, it is possible that statistics accrued to the
4068 * old rule will be incorrectly attributed to the new rule. This could be
4069 * avoided by calling update_stats() whenever rules are created or
4070 * deleted. However, the performance impact of making so many calls to the
4071 * datapath do not justify the benefit of having perfectly accurate statistics.
4073 * In addition, this function maintains per ofproto flow hit counts. The patch
4074 * port is not treated specially. e.g. A packet ingress from br0 patched into
4075 * br1 will increase the hit count of br0 by 1, however, does not affect
4076 * the hit or miss counts of br1.
4079 update_stats(struct dpif_backer *backer)
4081 const struct dpif_flow_stats *stats;
4082 struct dpif_flow_dump dump;
4083 const struct nlattr *key;
4086 dpif_flow_dump_start(&dump, backer->dpif);
4087 while (dpif_flow_dump_next(&dump, &key, &key_len, NULL, NULL, &stats)) {
4088 struct subfacet *subfacet;
4091 key_hash = odp_flow_key_hash(key, key_len);
4092 subfacet = subfacet_find(backer, key, key_len, key_hash);
4093 switch (subfacet ? subfacet->path : SF_NOT_INSTALLED) {
4095 update_subfacet_stats(subfacet, stats);
4099 /* Stats are updated per-packet. */
4102 case SF_NOT_INSTALLED:
4104 delete_unexpected_flow(backer, key, key_len);
4109 dpif_flow_dump_done(&dump);
4111 update_moving_averages(backer);
4114 /* Calculates and returns the number of milliseconds of idle time after which
4115 * subfacets should expire from the datapath. When a subfacet expires, we fold
4116 * its statistics into its facet, and when a facet's last subfacet expires, we
4117 * fold its statistic into its rule. */
4119 subfacet_max_idle(const struct dpif_backer *backer)
4122 * Idle time histogram.
4124 * Most of the time a switch has a relatively small number of subfacets.
4125 * When this is the case we might as well keep statistics for all of them
4126 * in userspace and to cache them in the kernel datapath for performance as
4129 * As the number of subfacets increases, the memory required to maintain
4130 * statistics about them in userspace and in the kernel becomes
4131 * significant. However, with a large number of subfacets it is likely
4132 * that only a few of them are "heavy hitters" that consume a large amount
4133 * of bandwidth. At this point, only heavy hitters are worth caching in
4134 * the kernel and maintaining in userspaces; other subfacets we can
4137 * The technique used to compute the idle time is to build a histogram with
4138 * N_BUCKETS buckets whose width is BUCKET_WIDTH msecs each. Each subfacet
4139 * that is installed in the kernel gets dropped in the appropriate bucket.
4140 * After the histogram has been built, we compute the cutoff so that only
4141 * the most-recently-used 1% of subfacets (but at least
4142 * flow_eviction_threshold flows) are kept cached. At least
4143 * the most-recently-used bucket of subfacets is kept, so actually an
4144 * arbitrary number of subfacets can be kept in any given expiration run
4145 * (though the next run will delete most of those unless they receive
4148 * This requires a second pass through the subfacets, in addition to the
4149 * pass made by update_stats(), because the former function never looks at
4150 * uninstallable subfacets.
4152 enum { BUCKET_WIDTH = ROUND_UP(100, TIME_UPDATE_INTERVAL) };
4153 enum { N_BUCKETS = 5000 / BUCKET_WIDTH };
4154 int buckets[N_BUCKETS] = { 0 };
4155 int total, subtotal, bucket;
4156 struct subfacet *subfacet;
4160 total = hmap_count(&backer->subfacets);
4161 if (total <= flow_eviction_threshold) {
4162 return N_BUCKETS * BUCKET_WIDTH;
4165 /* Build histogram. */
4167 HMAP_FOR_EACH (subfacet, hmap_node, &backer->subfacets) {
4168 long long int idle = now - subfacet->used;
4169 int bucket = (idle <= 0 ? 0
4170 : idle >= BUCKET_WIDTH * N_BUCKETS ? N_BUCKETS - 1
4171 : (unsigned int) idle / BUCKET_WIDTH);
4175 /* Find the first bucket whose flows should be expired. */
4176 subtotal = bucket = 0;
4178 subtotal += buckets[bucket++];
4179 } while (bucket < N_BUCKETS &&
4180 subtotal < MAX(flow_eviction_threshold, total / 100));
4182 if (VLOG_IS_DBG_ENABLED()) {
4186 ds_put_cstr(&s, "keep");
4187 for (i = 0; i < N_BUCKETS; i++) {
4189 ds_put_cstr(&s, ", drop");
4192 ds_put_format(&s, " %d:%d", i * BUCKET_WIDTH, buckets[i]);
4195 VLOG_INFO("%s (msec:count)", ds_cstr(&s));
4199 return bucket * BUCKET_WIDTH;
4203 expire_subfacets(struct dpif_backer *backer, int dp_max_idle)
4205 /* Cutoff time for most flows. */
4206 long long int normal_cutoff = time_msec() - dp_max_idle;
4208 /* We really want to keep flows for special protocols around, so use a more
4209 * conservative cutoff. */
4210 long long int special_cutoff = time_msec() - 10000;
4212 struct subfacet *subfacet, *next_subfacet;
4213 struct subfacet *batch[SUBFACET_DESTROY_MAX_BATCH];
4217 HMAP_FOR_EACH_SAFE (subfacet, next_subfacet, hmap_node,
4218 &backer->subfacets) {
4219 long long int cutoff;
4221 cutoff = (subfacet->facet->xout.slow & (SLOW_CFM | SLOW_BFD | SLOW_LACP
4225 if (subfacet->used < cutoff) {
4226 if (subfacet->path != SF_NOT_INSTALLED) {
4227 batch[n_batch++] = subfacet;
4228 if (n_batch >= SUBFACET_DESTROY_MAX_BATCH) {
4229 subfacet_destroy_batch(backer, batch, n_batch);
4233 subfacet_destroy(subfacet);
4239 subfacet_destroy_batch(backer, batch, n_batch);
4243 /* If 'rule' is an OpenFlow rule, that has expired according to OpenFlow rules,
4244 * then delete it entirely. */
4246 rule_expire(struct rule_dpif *rule)
4248 struct facet *facet, *next_facet;
4252 if (rule->up.pending) {
4253 /* We'll have to expire it later. */
4257 /* Has 'rule' expired? */
4259 if (rule->up.hard_timeout
4260 && now > rule->up.modified + rule->up.hard_timeout * 1000) {
4261 reason = OFPRR_HARD_TIMEOUT;
4262 } else if (rule->up.idle_timeout
4263 && now > rule->up.used + rule->up.idle_timeout * 1000) {
4264 reason = OFPRR_IDLE_TIMEOUT;
4269 COVERAGE_INC(ofproto_dpif_expired);
4271 /* Update stats. (This is a no-op if the rule expired due to an idle
4272 * timeout, because that only happens when the rule has no facets left.) */
4273 LIST_FOR_EACH_SAFE (facet, next_facet, list_node, &rule->facets) {
4274 facet_remove(facet);
4277 /* Get rid of the rule. */
4278 ofproto_rule_expire(&rule->up, reason);
4283 /* Creates and returns a new facet based on 'miss'.
4285 * The caller must already have determined that no facet with an identical
4286 * 'miss->flow' exists in 'miss->ofproto'.
4288 * 'rule' and 'xout' must have been created based on 'miss'.
4290 * 'facet'' statistics are initialized based on 'stats'.
4292 * The facet will initially have no subfacets. The caller should create (at
4293 * least) one subfacet with subfacet_create(). */
4294 static struct facet *
4295 facet_create(const struct flow_miss *miss, struct rule_dpif *rule,
4296 struct xlate_out *xout, struct dpif_flow_stats *stats)
4298 struct ofproto_dpif *ofproto = miss->ofproto;
4299 struct facet *facet;
4302 facet = xzalloc(sizeof *facet);
4303 facet->packet_count = facet->prev_packet_count = stats->n_packets;
4304 facet->byte_count = facet->prev_byte_count = stats->n_bytes;
4305 facet->tcp_flags = stats->tcp_flags;
4306 facet->used = stats->used;
4307 facet->flow = miss->flow;
4308 facet->learn_rl = time_msec() + 500;
4311 list_push_back(&facet->rule->facets, &facet->list_node);
4312 list_init(&facet->subfacets);
4313 netflow_flow_init(&facet->nf_flow);
4314 netflow_flow_update_time(ofproto->netflow, &facet->nf_flow, facet->used);
4316 xlate_out_copy(&facet->xout, xout);
4318 match_init(&match, &facet->flow, &facet->xout.wc);
4319 cls_rule_init(&facet->cr, &match, OFP_DEFAULT_PRIORITY);
4320 classifier_insert(&ofproto->facets, &facet->cr);
4322 facet->nf_flow.output_iface = facet->xout.nf_output_iface;
4328 facet_free(struct facet *facet)
4331 xlate_out_uninit(&facet->xout);
4336 /* Executes, within 'ofproto', the 'n_actions' actions in 'actions' on
4337 * 'packet', which arrived on 'in_port'. */
4339 execute_odp_actions(struct ofproto_dpif *ofproto, const struct flow *flow,
4340 const struct nlattr *odp_actions, size_t actions_len,
4341 struct ofpbuf *packet)
4343 struct odputil_keybuf keybuf;
4347 ofpbuf_use_stack(&key, &keybuf, sizeof keybuf);
4348 odp_flow_key_from_flow(&key, flow,
4349 ofp_port_to_odp_port(ofproto, flow->in_port));
4351 error = dpif_execute(ofproto->backer->dpif, key.data, key.size,
4352 odp_actions, actions_len, packet);
4356 /* Remove 'facet' from its ofproto and free up the associated memory:
4358 * - If 'facet' was installed in the datapath, uninstalls it and updates its
4359 * rule's statistics, via subfacet_uninstall().
4361 * - Removes 'facet' from its rule and from ofproto->facets.
4364 facet_remove(struct facet *facet)
4366 struct ofproto_dpif *ofproto = ofproto_dpif_cast(facet->rule->up.ofproto);
4367 struct subfacet *subfacet, *next_subfacet;
4369 ovs_assert(!list_is_empty(&facet->subfacets));
4371 /* First uninstall all of the subfacets to get final statistics. */
4372 LIST_FOR_EACH (subfacet, list_node, &facet->subfacets) {
4373 subfacet_uninstall(subfacet);
4376 /* Flush the final stats to the rule.
4378 * This might require us to have at least one subfacet around so that we
4379 * can use its actions for accounting in facet_account(), which is why we
4380 * have uninstalled but not yet destroyed the subfacets. */
4381 facet_flush_stats(facet);
4383 /* Now we're really all done so destroy everything. */
4384 LIST_FOR_EACH_SAFE (subfacet, next_subfacet, list_node,
4385 &facet->subfacets) {
4386 subfacet_destroy__(subfacet);
4388 classifier_remove(&ofproto->facets, &facet->cr);
4389 cls_rule_destroy(&facet->cr);
4390 list_remove(&facet->list_node);
4394 /* Feed information from 'facet' back into the learning table to keep it in
4395 * sync with what is actually flowing through the datapath. */
4397 facet_learn(struct facet *facet)
4399 long long int now = time_msec();
4401 if (!facet->xout.has_fin_timeout && now < facet->learn_rl) {
4405 facet->learn_rl = now + 500;
4407 if (!facet->xout.has_learn
4408 && !facet->xout.has_normal
4409 && (!facet->xout.has_fin_timeout
4410 || !(facet->tcp_flags & (TCP_FIN | TCP_RST)))) {
4414 facet_push_stats(facet, true);
4418 facet_account(struct facet *facet)
4420 struct ofproto_dpif *ofproto = ofproto_dpif_cast(facet->rule->up.ofproto);
4421 const struct nlattr *a;
4426 if (!facet->xout.has_normal || !ofproto->has_bonded_bundles) {
4429 n_bytes = facet->byte_count - facet->accounted_bytes;
4431 /* This loop feeds byte counters to bond_account() for rebalancing to use
4432 * as a basis. We also need to track the actual VLAN on which the packet
4433 * is going to be sent to ensure that it matches the one passed to
4434 * bond_choose_output_slave(). (Otherwise, we will account to the wrong
4437 * We use the actions from an arbitrary subfacet because they should all
4438 * be equally valid for our purpose. */
4439 vlan_tci = facet->flow.vlan_tci;
4440 NL_ATTR_FOR_EACH_UNSAFE (a, left, facet->xout.odp_actions.data,
4441 facet->xout.odp_actions.size) {
4442 const struct ovs_action_push_vlan *vlan;
4443 struct ofport_dpif *port;
4445 switch (nl_attr_type(a)) {
4446 case OVS_ACTION_ATTR_OUTPUT:
4447 port = get_odp_port(ofproto, nl_attr_get_u32(a));
4448 if (port && port->bundle && port->bundle->bond) {
4449 bond_account(port->bundle->bond, &facet->flow,
4450 vlan_tci_to_vid(vlan_tci), n_bytes);
4454 case OVS_ACTION_ATTR_POP_VLAN:
4455 vlan_tci = htons(0);
4458 case OVS_ACTION_ATTR_PUSH_VLAN:
4459 vlan = nl_attr_get(a);
4460 vlan_tci = vlan->vlan_tci;
4466 /* Returns true if the only action for 'facet' is to send to the controller.
4467 * (We don't report NetFlow expiration messages for such facets because they
4468 * are just part of the control logic for the network, not real traffic). */
4470 facet_is_controller_flow(struct facet *facet)
4473 const struct rule *rule = &facet->rule->up;
4474 const struct ofpact *ofpacts = rule->ofpacts;
4475 size_t ofpacts_len = rule->ofpacts_len;
4477 if (ofpacts_len > 0 &&
4478 ofpacts->type == OFPACT_CONTROLLER &&
4479 ofpact_next(ofpacts) >= ofpact_end(ofpacts, ofpacts_len)) {
4486 /* Folds all of 'facet''s statistics into its rule. Also updates the
4487 * accounting ofhook and emits a NetFlow expiration if appropriate. All of
4488 * 'facet''s statistics in the datapath should have been zeroed and folded into
4489 * its packet and byte counts before this function is called. */
4491 facet_flush_stats(struct facet *facet)
4493 struct ofproto_dpif *ofproto = ofproto_dpif_cast(facet->rule->up.ofproto);
4494 struct subfacet *subfacet;
4496 LIST_FOR_EACH (subfacet, list_node, &facet->subfacets) {
4497 ovs_assert(!subfacet->dp_byte_count);
4498 ovs_assert(!subfacet->dp_packet_count);
4501 facet_push_stats(facet, false);
4502 if (facet->accounted_bytes < facet->byte_count) {
4503 facet_account(facet);
4504 facet->accounted_bytes = facet->byte_count;
4507 if (ofproto->netflow && !facet_is_controller_flow(facet)) {
4508 struct ofexpired expired;
4509 expired.flow = facet->flow;
4510 expired.packet_count = facet->packet_count;
4511 expired.byte_count = facet->byte_count;
4512 expired.used = facet->used;
4513 netflow_expire(ofproto->netflow, &facet->nf_flow, &expired);
4516 /* Reset counters to prevent double counting if 'facet' ever gets
4518 facet_reset_counters(facet);
4520 netflow_flow_clear(&facet->nf_flow);
4521 facet->tcp_flags = 0;
4524 /* Searches 'ofproto''s table of facets for one which would be responsible for
4525 * 'flow'. Returns it if found, otherwise a null pointer.
4527 * The returned facet might need revalidation; use facet_lookup_valid()
4528 * instead if that is important. */
4529 static struct facet *
4530 facet_find(struct ofproto_dpif *ofproto, const struct flow *flow)
4532 struct cls_rule *cr = classifier_lookup(&ofproto->facets, flow, NULL);
4533 return cr ? CONTAINER_OF(cr, struct facet, cr) : NULL;
4536 /* Searches 'ofproto''s table of facets for one capable that covers
4537 * 'flow'. Returns it if found, otherwise a null pointer.
4539 * The returned facet is guaranteed to be valid. */
4540 static struct facet *
4541 facet_lookup_valid(struct ofproto_dpif *ofproto, const struct flow *flow)
4543 struct facet *facet;
4545 facet = facet_find(ofproto, flow);
4547 && (ofproto->backer->need_revalidate
4548 || tag_set_intersects(&ofproto->backer->revalidate_set,
4550 && !facet_revalidate(facet)) {
4558 facet_check_consistency(struct facet *facet)
4560 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 15);
4562 struct ofproto_dpif *ofproto = ofproto_dpif_cast(facet->rule->up.ofproto);
4564 struct xlate_out xout;
4565 struct xlate_in xin;
4567 struct rule_dpif *rule;
4570 /* Check the rule for consistency. */
4571 rule = rule_dpif_lookup(ofproto, &facet->flow, NULL);
4572 if (rule != facet->rule) {
4573 if (!VLOG_DROP_WARN(&rl)) {
4574 struct ds s = DS_EMPTY_INITIALIZER;
4576 flow_format(&s, &facet->flow);
4577 ds_put_format(&s, ": facet associated with wrong rule (was "
4578 "table=%"PRIu8",", facet->rule->up.table_id);
4579 cls_rule_format(&facet->rule->up.cr, &s);
4580 ds_put_format(&s, ") (should have been table=%"PRIu8",",
4582 cls_rule_format(&rule->up.cr, &s);
4583 ds_put_char(&s, ')');
4585 VLOG_WARN("%s", ds_cstr(&s));
4591 /* Check the datapath actions for consistency. */
4592 xlate_in_init(&xin, ofproto, &facet->flow, rule, 0, NULL);
4593 xlate_actions(&xin, &xout);
4595 ok = ofpbuf_equal(&facet->xout.odp_actions, &xout.odp_actions)
4596 && facet->xout.slow == xout.slow;
4597 if (!ok && !VLOG_DROP_WARN(&rl)) {
4598 struct ds s = DS_EMPTY_INITIALIZER;
4600 flow_format(&s, &facet->flow);
4601 ds_put_cstr(&s, ": inconsistency in facet");
4603 if (!ofpbuf_equal(&facet->xout.odp_actions, &xout.odp_actions)) {
4604 ds_put_cstr(&s, " (actions were: ");
4605 format_odp_actions(&s, facet->xout.odp_actions.data,
4606 facet->xout.odp_actions.size);
4607 ds_put_cstr(&s, ") (correct actions: ");
4608 format_odp_actions(&s, xout.odp_actions.data,
4609 xout.odp_actions.size);
4610 ds_put_char(&s, ')');
4613 if (facet->xout.slow != xout.slow) {
4614 ds_put_format(&s, " slow path incorrect. should be %d", xout.slow);
4617 VLOG_WARN("%s", ds_cstr(&s));
4620 xlate_out_uninit(&xout);
4625 /* Re-searches the classifier for 'facet':
4627 * - If the rule found is different from 'facet''s current rule, moves
4628 * 'facet' to the new rule and recompiles its actions.
4630 * - If the rule found is the same as 'facet''s current rule, leaves 'facet'
4631 * where it is and recompiles its actions anyway.
4633 * - If any of 'facet''s subfacets correspond to a new flow according to
4634 * ofproto_receive(), 'facet' is removed.
4636 * Returns true if 'facet' is still valid. False if 'facet' was removed. */
4638 facet_revalidate(struct facet *facet)
4640 struct ofproto_dpif *ofproto = ofproto_dpif_cast(facet->rule->up.ofproto);
4641 struct rule_dpif *new_rule;
4642 struct subfacet *subfacet;
4643 struct flow_wildcards wc;
4644 struct xlate_out xout;
4645 struct xlate_in xin;
4647 COVERAGE_INC(facet_revalidate);
4649 /* Check that child subfacets still correspond to this facet. Tunnel
4650 * configuration changes could cause a subfacet's OpenFlow in_port to
4652 LIST_FOR_EACH (subfacet, list_node, &facet->subfacets) {
4653 struct ofproto_dpif *recv_ofproto;
4654 struct flow recv_flow;
4657 error = ofproto_receive(ofproto->backer, NULL, subfacet->key,
4658 subfacet->key_len, &recv_flow, NULL,
4659 &recv_ofproto, NULL);
4661 || recv_ofproto != ofproto
4662 || facet != facet_find(ofproto, &recv_flow)) {
4663 facet_remove(facet);
4668 flow_wildcards_init_catchall(&wc);
4669 new_rule = rule_dpif_lookup(ofproto, &facet->flow, &wc);
4671 /* Calculate new datapath actions.
4673 * We do not modify any 'facet' state yet, because we might need to, e.g.,
4674 * emit a NetFlow expiration and, if so, we need to have the old state
4675 * around to properly compose it. */
4676 xlate_in_init(&xin, ofproto, &facet->flow, new_rule, 0, NULL);
4677 xlate_actions(&xin, &xout);
4678 flow_wildcards_or(&xout.wc, &xout.wc, &wc);
4680 /* A facet's slow path reason should only change under dramatic
4681 * circumstances. Rather than try to update everything, it's simpler to
4682 * remove the facet and start over.
4684 * More importantly, if a facet's wildcards change, it will be relatively
4685 * difficult to figure out if its subfacets still belong to it, and if not
4686 * which facet they may belong to. Again, to avoid the complexity, we
4687 * simply give up instead. */
4688 if (facet->xout.slow != xout.slow
4689 || memcmp(&facet->xout.wc, &xout.wc, sizeof xout.wc)) {
4690 facet_remove(facet);
4691 xlate_out_uninit(&xout);
4695 if (!ofpbuf_equal(&facet->xout.odp_actions, &xout.odp_actions)) {
4696 LIST_FOR_EACH(subfacet, list_node, &facet->subfacets) {
4697 if (subfacet->path == SF_FAST_PATH) {
4698 struct dpif_flow_stats stats;
4700 subfacet_install(subfacet, &xout.odp_actions, &stats);
4701 subfacet_update_stats(subfacet, &stats);
4705 facet_flush_stats(facet);
4707 ofpbuf_clear(&facet->xout.odp_actions);
4708 ofpbuf_put(&facet->xout.odp_actions, xout.odp_actions.data,
4709 xout.odp_actions.size);
4712 /* Update 'facet' now that we've taken care of all the old state. */
4713 facet->xout.tags = xout.tags;
4714 facet->xout.slow = xout.slow;
4715 facet->xout.has_learn = xout.has_learn;
4716 facet->xout.has_normal = xout.has_normal;
4717 facet->xout.has_fin_timeout = xout.has_fin_timeout;
4718 facet->xout.nf_output_iface = xout.nf_output_iface;
4719 facet->xout.mirrors = xout.mirrors;
4720 facet->nf_flow.output_iface = facet->xout.nf_output_iface;
4722 if (facet->rule != new_rule) {
4723 COVERAGE_INC(facet_changed_rule);
4724 list_remove(&facet->list_node);
4725 list_push_back(&new_rule->facets, &facet->list_node);
4726 facet->rule = new_rule;
4727 facet->used = new_rule->up.created;
4728 facet->prev_used = facet->used;
4731 xlate_out_uninit(&xout);
4736 facet_reset_counters(struct facet *facet)
4738 facet->packet_count = 0;
4739 facet->byte_count = 0;
4740 facet->prev_packet_count = 0;
4741 facet->prev_byte_count = 0;
4742 facet->accounted_bytes = 0;
4746 facet_push_stats(struct facet *facet, bool may_learn)
4748 struct dpif_flow_stats stats;
4750 ovs_assert(facet->packet_count >= facet->prev_packet_count);
4751 ovs_assert(facet->byte_count >= facet->prev_byte_count);
4752 ovs_assert(facet->used >= facet->prev_used);
4754 stats.n_packets = facet->packet_count - facet->prev_packet_count;
4755 stats.n_bytes = facet->byte_count - facet->prev_byte_count;
4756 stats.used = facet->used;
4757 stats.tcp_flags = facet->tcp_flags;
4759 if (may_learn || stats.n_packets || facet->used > facet->prev_used) {
4760 struct ofproto_dpif *ofproto =
4761 ofproto_dpif_cast(facet->rule->up.ofproto);
4763 struct ofport_dpif *in_port;
4764 struct xlate_in xin;
4766 facet->prev_packet_count = facet->packet_count;
4767 facet->prev_byte_count = facet->byte_count;
4768 facet->prev_used = facet->used;
4770 in_port = get_ofp_port(ofproto, facet->flow.in_port);
4771 if (in_port && in_port->tnl_port) {
4772 netdev_vport_inc_rx(in_port->up.netdev, &stats);
4775 rule_credit_stats(facet->rule, &stats);
4776 netflow_flow_update_time(ofproto->netflow, &facet->nf_flow,
4778 netflow_flow_update_flags(&facet->nf_flow, facet->tcp_flags);
4779 update_mirror_stats(ofproto, facet->xout.mirrors, stats.n_packets,
4782 xlate_in_init(&xin, ofproto, &facet->flow, facet->rule,
4783 stats.tcp_flags, NULL);
4784 xin.resubmit_stats = &stats;
4785 xin.may_learn = may_learn;
4786 xlate_actions_for_side_effects(&xin);
4791 push_all_stats__(bool run_fast)
4793 static long long int rl = LLONG_MIN;
4794 struct ofproto_dpif *ofproto;
4796 if (time_msec() < rl) {
4800 HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
4801 struct cls_cursor cursor;
4802 struct facet *facet;
4804 cls_cursor_init(&cursor, &ofproto->facets, NULL);
4805 CLS_CURSOR_FOR_EACH (facet, cr, &cursor) {
4806 facet_push_stats(facet, false);
4813 rl = time_msec() + 100;
4817 push_all_stats(void)
4819 push_all_stats__(true);
4823 rule_credit_stats(struct rule_dpif *rule, const struct dpif_flow_stats *stats)
4825 rule->packet_count += stats->n_packets;
4826 rule->byte_count += stats->n_bytes;
4827 ofproto_rule_update_used(&rule->up, stats->used);
4832 static struct subfacet *
4833 subfacet_find(struct dpif_backer *backer, const struct nlattr *key,
4834 size_t key_len, uint32_t key_hash)
4836 struct subfacet *subfacet;
4838 HMAP_FOR_EACH_WITH_HASH (subfacet, hmap_node, key_hash,
4839 &backer->subfacets) {
4840 if (subfacet->key_len == key_len
4841 && !memcmp(key, subfacet->key, key_len)) {
4849 /* Searches 'facet' (within 'ofproto') for a subfacet with the specified
4850 * 'key_fitness', 'key', and 'key_len' members in 'miss'. Returns the
4851 * existing subfacet if there is one, otherwise creates and returns a
4853 static struct subfacet *
4854 subfacet_create(struct facet *facet, struct flow_miss *miss,
4857 struct dpif_backer *backer = miss->ofproto->backer;
4858 enum odp_key_fitness key_fitness = miss->key_fitness;
4859 const struct nlattr *key = miss->key;
4860 size_t key_len = miss->key_len;
4862 struct subfacet *subfacet;
4864 key_hash = odp_flow_key_hash(key, key_len);
4866 if (list_is_empty(&facet->subfacets)) {
4867 subfacet = &facet->one_subfacet;
4869 subfacet = subfacet_find(backer, key, key_len, key_hash);
4871 if (subfacet->facet == facet) {
4875 /* This shouldn't happen. */
4876 VLOG_ERR_RL(&rl, "subfacet with wrong facet");
4877 subfacet_destroy(subfacet);
4880 subfacet = xmalloc(sizeof *subfacet);
4883 hmap_insert(&backer->subfacets, &subfacet->hmap_node, key_hash);
4884 list_push_back(&facet->subfacets, &subfacet->list_node);
4885 subfacet->facet = facet;
4886 subfacet->key_fitness = key_fitness;
4887 subfacet->key = xmemdup(key, key_len);
4888 subfacet->key_len = key_len;
4889 subfacet->used = now;
4890 subfacet->created = now;
4891 subfacet->dp_packet_count = 0;
4892 subfacet->dp_byte_count = 0;
4893 subfacet->path = SF_NOT_INSTALLED;
4894 subfacet->backer = backer;
4896 backer->subfacet_add_count++;
4900 /* Uninstalls 'subfacet' from the datapath, if it is installed, removes it from
4901 * its facet within 'ofproto', and frees it. */
4903 subfacet_destroy__(struct subfacet *subfacet)
4905 struct facet *facet = subfacet->facet;
4906 struct ofproto_dpif *ofproto = ofproto_dpif_cast(facet->rule->up.ofproto);
4908 /* Update ofproto stats before uninstall the subfacet. */
4909 ofproto->backer->subfacet_del_count++;
4911 subfacet_uninstall(subfacet);
4912 hmap_remove(&subfacet->backer->subfacets, &subfacet->hmap_node);
4913 list_remove(&subfacet->list_node);
4914 free(subfacet->key);
4915 if (subfacet != &facet->one_subfacet) {
4920 /* Destroys 'subfacet', as with subfacet_destroy__(), and then if this was the
4921 * last remaining subfacet in its facet destroys the facet too. */
4923 subfacet_destroy(struct subfacet *subfacet)
4925 struct facet *facet = subfacet->facet;
4927 if (list_is_singleton(&facet->subfacets)) {
4928 /* facet_remove() needs at least one subfacet (it will remove it). */
4929 facet_remove(facet);
4931 subfacet_destroy__(subfacet);
4936 subfacet_destroy_batch(struct dpif_backer *backer,
4937 struct subfacet **subfacets, int n)
4939 struct dpif_op ops[SUBFACET_DESTROY_MAX_BATCH];
4940 struct dpif_op *opsp[SUBFACET_DESTROY_MAX_BATCH];
4941 struct dpif_flow_stats stats[SUBFACET_DESTROY_MAX_BATCH];
4944 for (i = 0; i < n; i++) {
4945 ops[i].type = DPIF_OP_FLOW_DEL;
4946 ops[i].u.flow_del.key = subfacets[i]->key;
4947 ops[i].u.flow_del.key_len = subfacets[i]->key_len;
4948 ops[i].u.flow_del.stats = &stats[i];
4952 dpif_operate(backer->dpif, opsp, n);
4953 for (i = 0; i < n; i++) {
4954 subfacet_reset_dp_stats(subfacets[i], &stats[i]);
4955 subfacets[i]->path = SF_NOT_INSTALLED;
4956 subfacet_destroy(subfacets[i]);
4961 /* Updates 'subfacet''s datapath flow, setting its actions to 'actions_len'
4962 * bytes of actions in 'actions'. If 'stats' is non-null, statistics counters
4963 * in the datapath will be zeroed and 'stats' will be updated with traffic new
4964 * since 'subfacet' was last updated.
4966 * Returns 0 if successful, otherwise a positive errno value. */
4968 subfacet_install(struct subfacet *subfacet, const struct ofpbuf *odp_actions,
4969 struct dpif_flow_stats *stats)
4971 struct facet *facet = subfacet->facet;
4972 struct ofproto_dpif *ofproto = ofproto_dpif_cast(facet->rule->up.ofproto);
4973 enum subfacet_path path = facet->xout.slow ? SF_SLOW_PATH : SF_FAST_PATH;
4974 const struct nlattr *actions = odp_actions->data;
4975 size_t actions_len = odp_actions->size;
4977 uint64_t slow_path_stub[128 / 8];
4978 enum dpif_flow_put_flags flags;
4981 flags = DPIF_FP_CREATE | DPIF_FP_MODIFY;
4983 flags |= DPIF_FP_ZERO_STATS;
4986 if (path == SF_SLOW_PATH) {
4987 compose_slow_path(ofproto, &facet->flow, facet->xout.slow,
4988 slow_path_stub, sizeof slow_path_stub,
4989 &actions, &actions_len);
4992 ret = dpif_flow_put(subfacet->backer->dpif, flags, subfacet->key,
4993 subfacet->key_len, actions, actions_len, stats);
4996 subfacet_reset_dp_stats(subfacet, stats);
5000 subfacet->path = path;
5005 /* If 'subfacet' is installed in the datapath, uninstalls it. */
5007 subfacet_uninstall(struct subfacet *subfacet)
5009 if (subfacet->path != SF_NOT_INSTALLED) {
5010 struct rule_dpif *rule = subfacet->facet->rule;
5011 struct ofproto_dpif *ofproto = ofproto_dpif_cast(rule->up.ofproto);
5012 struct dpif_flow_stats stats;
5015 error = dpif_flow_del(ofproto->backer->dpif, subfacet->key,
5016 subfacet->key_len, &stats);
5017 subfacet_reset_dp_stats(subfacet, &stats);
5019 subfacet_update_stats(subfacet, &stats);
5021 subfacet->path = SF_NOT_INSTALLED;
5023 ovs_assert(subfacet->dp_packet_count == 0);
5024 ovs_assert(subfacet->dp_byte_count == 0);
5028 /* Resets 'subfacet''s datapath statistics counters. This should be called
5029 * when 'subfacet''s statistics are cleared in the datapath. If 'stats' is
5030 * non-null, it should contain the statistics returned by dpif when 'subfacet'
5031 * was reset in the datapath. 'stats' will be modified to include only
5032 * statistics new since 'subfacet' was last updated. */
5034 subfacet_reset_dp_stats(struct subfacet *subfacet,
5035 struct dpif_flow_stats *stats)
5038 && subfacet->dp_packet_count <= stats->n_packets
5039 && subfacet->dp_byte_count <= stats->n_bytes) {
5040 stats->n_packets -= subfacet->dp_packet_count;
5041 stats->n_bytes -= subfacet->dp_byte_count;
5044 subfacet->dp_packet_count = 0;
5045 subfacet->dp_byte_count = 0;
5048 /* Folds the statistics from 'stats' into the counters in 'subfacet'.
5050 * Because of the meaning of a subfacet's counters, it only makes sense to do
5051 * this if 'stats' are not tracked in the datapath, that is, if 'stats'
5052 * represents a packet that was sent by hand or if it represents statistics
5053 * that have been cleared out of the datapath. */
5055 subfacet_update_stats(struct subfacet *subfacet,
5056 const struct dpif_flow_stats *stats)
5058 if (stats->n_packets || stats->used > subfacet->used) {
5059 struct facet *facet = subfacet->facet;
5061 subfacet->used = MAX(subfacet->used, stats->used);
5062 facet->used = MAX(facet->used, stats->used);
5063 facet->packet_count += stats->n_packets;
5064 facet->byte_count += stats->n_bytes;
5065 facet->tcp_flags |= stats->tcp_flags;
5071 /* Lookup 'flow' in 'ofproto''s classifier. If 'wc' is non-null, sets
5072 * the fields that were relevant as part of the lookup. */
5073 static struct rule_dpif *
5074 rule_dpif_lookup(struct ofproto_dpif *ofproto, const struct flow *flow,
5075 struct flow_wildcards *wc)
5077 struct rule_dpif *rule;
5079 rule = rule_dpif_lookup_in_table(ofproto, flow, wc, 0);
5084 return rule_dpif_miss_rule(ofproto, flow);
5088 rule_dpif_lookup_in_table(struct ofproto_dpif *ofproto,
5089 const struct flow *flow, struct flow_wildcards *wc,
5092 struct cls_rule *cls_rule;
5093 struct classifier *cls;
5096 if (table_id >= N_TABLES) {
5100 cls = &ofproto->up.tables[table_id].cls;
5101 frag = (flow->nw_frag & FLOW_NW_FRAG_ANY) != 0;
5102 if (frag && ofproto->up.frag_handling == OFPC_FRAG_NORMAL) {
5103 /* We must pretend that transport ports are unavailable. */
5104 struct flow ofpc_normal_flow = *flow;
5105 ofpc_normal_flow.tp_src = htons(0);
5106 ofpc_normal_flow.tp_dst = htons(0);
5107 cls_rule = classifier_lookup(cls, &ofpc_normal_flow, wc);
5108 } else if (frag && ofproto->up.frag_handling == OFPC_FRAG_DROP) {
5109 cls_rule = &ofproto->drop_frags_rule->up.cr;
5111 flow_wildcards_init_exact(wc);
5114 cls_rule = classifier_lookup(cls, flow, wc);
5116 return rule_dpif_cast(rule_from_cls_rule(cls_rule));
5120 rule_dpif_miss_rule(struct ofproto_dpif *ofproto, const struct flow *flow)
5122 struct ofport_dpif *port;
5124 port = get_ofp_port(ofproto, flow->in_port);
5126 VLOG_WARN_RL(&rl, "packet-in on unknown port %"PRIu16, flow->in_port);
5127 return ofproto->miss_rule;
5130 if (port->up.pp.config & OFPUTIL_PC_NO_PACKET_IN) {
5131 return ofproto->no_packet_in_rule;
5133 return ofproto->miss_rule;
5137 complete_operation(struct rule_dpif *rule)
5139 struct ofproto_dpif *ofproto = ofproto_dpif_cast(rule->up.ofproto);
5141 rule_invalidate(rule);
5143 struct dpif_completion *c = xmalloc(sizeof *c);
5144 c->op = rule->up.pending;
5145 list_push_back(&ofproto->completions, &c->list_node);
5147 ofoperation_complete(rule->up.pending, 0);
5151 static struct rule *
5154 struct rule_dpif *rule = xmalloc(sizeof *rule);
5159 rule_dealloc(struct rule *rule_)
5161 struct rule_dpif *rule = rule_dpif_cast(rule_);
5166 rule_construct(struct rule *rule_)
5168 struct rule_dpif *rule = rule_dpif_cast(rule_);
5169 struct ofproto_dpif *ofproto = ofproto_dpif_cast(rule->up.ofproto);
5170 struct rule_dpif *victim;
5173 rule->packet_count = 0;
5174 rule->byte_count = 0;
5176 victim = rule_dpif_cast(ofoperation_get_victim(rule->up.pending));
5177 if (victim && !list_is_empty(&victim->facets)) {
5178 struct facet *facet;
5180 rule->facets = victim->facets;
5181 list_moved(&rule->facets);
5182 LIST_FOR_EACH (facet, list_node, &rule->facets) {
5183 /* XXX: We're only clearing our local counters here. It's possible
5184 * that quite a few packets are unaccounted for in the datapath
5185 * statistics. These will be accounted to the new rule instead of
5186 * cleared as required. This could be fixed by clearing out the
5187 * datapath statistics for this facet, but currently it doesn't
5189 facet_reset_counters(facet);
5193 /* Must avoid list_moved() in this case. */
5194 list_init(&rule->facets);
5197 table_id = rule->up.table_id;
5199 rule->tag = victim->tag;
5200 } else if (table_id == 0) {
5205 miniflow_expand(&rule->up.cr.match.flow, &flow);
5206 rule->tag = rule_calculate_tag(&flow, &rule->up.cr.match.mask,
5207 ofproto->tables[table_id].basis);
5210 complete_operation(rule);
5215 rule_destruct(struct rule *rule_)
5217 struct rule_dpif *rule = rule_dpif_cast(rule_);
5218 struct facet *facet, *next_facet;
5220 LIST_FOR_EACH_SAFE (facet, next_facet, list_node, &rule->facets) {
5221 facet_revalidate(facet);
5224 complete_operation(rule);
5228 rule_get_stats(struct rule *rule_, uint64_t *packets, uint64_t *bytes)
5230 struct rule_dpif *rule = rule_dpif_cast(rule_);
5232 /* push_all_stats() can handle flow misses which, when using the learn
5233 * action, can cause rules to be added and deleted. This can corrupt our
5234 * caller's datastructures which assume that rule_get_stats() doesn't have
5235 * an impact on the flow table. To be safe, we disable miss handling. */
5236 push_all_stats__(false);
5238 /* Start from historical data for 'rule' itself that are no longer tracked
5239 * in facets. This counts, for example, facets that have expired. */
5240 *packets = rule->packet_count;
5241 *bytes = rule->byte_count;
5245 rule_dpif_execute(struct rule_dpif *rule, const struct flow *flow,
5246 struct ofpbuf *packet)
5248 struct ofproto_dpif *ofproto = ofproto_dpif_cast(rule->up.ofproto);
5249 struct dpif_flow_stats stats;
5250 struct xlate_out xout;
5251 struct xlate_in xin;
5253 dpif_flow_stats_extract(flow, packet, time_msec(), &stats);
5254 rule_credit_stats(rule, &stats);
5256 xlate_in_init(&xin, ofproto, flow, rule, stats.tcp_flags, packet);
5257 xin.resubmit_stats = &stats;
5258 xlate_actions(&xin, &xout);
5260 execute_odp_actions(ofproto, flow, xout.odp_actions.data,
5261 xout.odp_actions.size, packet);
5263 xlate_out_uninit(&xout);
5267 rule_execute(struct rule *rule, const struct flow *flow,
5268 struct ofpbuf *packet)
5270 rule_dpif_execute(rule_dpif_cast(rule), flow, packet);
5271 ofpbuf_delete(packet);
5276 rule_modify_actions(struct rule *rule_)
5278 struct rule_dpif *rule = rule_dpif_cast(rule_);
5280 complete_operation(rule);
5283 /* Sends 'packet' out 'ofport'.
5284 * May modify 'packet'.
5285 * Returns 0 if successful, otherwise a positive errno value. */
5287 send_packet(const struct ofport_dpif *ofport, struct ofpbuf *packet)
5289 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport->up.ofproto);
5290 uint64_t odp_actions_stub[1024 / 8];
5291 struct ofpbuf key, odp_actions;
5292 struct dpif_flow_stats stats;
5293 struct odputil_keybuf keybuf;
5294 struct ofpact_output output;
5295 struct xlate_out xout;
5296 struct xlate_in xin;
5300 ofpbuf_use_stub(&odp_actions, odp_actions_stub, sizeof odp_actions_stub);
5301 ofpbuf_use_stack(&key, &keybuf, sizeof keybuf);
5303 /* Use OFPP_NONE as the in_port to avoid special packet processing. */
5304 flow_extract(packet, 0, 0, NULL, OFPP_NONE, &flow);
5305 odp_flow_key_from_flow(&key, &flow, ofp_port_to_odp_port(ofproto,
5307 dpif_flow_stats_extract(&flow, packet, time_msec(), &stats);
5309 ofpact_init(&output.ofpact, OFPACT_OUTPUT, sizeof output);
5310 output.port = ofport->up.ofp_port;
5313 xlate_in_init(&xin, ofproto, &flow, NULL, 0, packet);
5314 xin.ofpacts_len = sizeof output;
5315 xin.ofpacts = &output.ofpact;
5316 xin.resubmit_stats = &stats;
5317 xlate_actions(&xin, &xout);
5319 error = dpif_execute(ofproto->backer->dpif,
5321 xout.odp_actions.data, xout.odp_actions.size,
5323 xlate_out_uninit(&xout);
5326 VLOG_WARN_RL(&rl, "%s: failed to send packet on port %s (%s)",
5327 ofproto->up.name, netdev_get_name(ofport->up.netdev),
5331 ofproto->stats.tx_packets++;
5332 ofproto->stats.tx_bytes += packet->size;
5336 /* Composes an ODP action for a "slow path" action for 'flow' within 'ofproto'.
5337 * The action will state 'slow' as the reason that the action is in the slow
5338 * path. (This is purely informational: it allows a human viewing "ovs-dpctl
5339 * dump-flows" output to see why a flow is in the slow path.)
5341 * The 'stub_size' bytes in 'stub' will be used to store the action.
5342 * 'stub_size' must be large enough for the action.
5344 * The action and its size will be stored in '*actionsp' and '*actions_lenp',
5347 compose_slow_path(const struct ofproto_dpif *ofproto, const struct flow *flow,
5348 enum slow_path_reason slow,
5349 uint64_t *stub, size_t stub_size,
5350 const struct nlattr **actionsp, size_t *actions_lenp)
5352 union user_action_cookie cookie;
5355 cookie.type = USER_ACTION_COOKIE_SLOW_PATH;
5356 cookie.slow_path.unused = 0;
5357 cookie.slow_path.reason = slow;
5359 ofpbuf_use_stack(&buf, stub, stub_size);
5360 if (slow & (SLOW_CFM | SLOW_BFD | SLOW_LACP | SLOW_STP)) {
5361 uint32_t pid = dpif_port_get_pid(ofproto->backer->dpif, UINT32_MAX);
5362 odp_put_userspace_action(pid, &cookie, sizeof cookie.slow_path, &buf);
5364 put_userspace_action(ofproto, &buf, flow, &cookie,
5365 sizeof cookie.slow_path);
5367 *actionsp = buf.data;
5368 *actions_lenp = buf.size;
5372 put_userspace_action(const struct ofproto_dpif *ofproto,
5373 struct ofpbuf *odp_actions,
5374 const struct flow *flow,
5375 const union user_action_cookie *cookie,
5376 const size_t cookie_size)
5380 pid = dpif_port_get_pid(ofproto->backer->dpif,
5381 ofp_port_to_odp_port(ofproto, flow->in_port));
5383 return odp_put_userspace_action(pid, cookie, cookie_size, odp_actions);
5388 update_mirror_stats(struct ofproto_dpif *ofproto, mirror_mask_t mirrors,
5389 uint64_t packets, uint64_t bytes)
5395 for (; mirrors; mirrors = zero_rightmost_1bit(mirrors)) {
5398 m = ofproto->mirrors[mirror_mask_ffs(mirrors) - 1];
5401 /* In normal circumstances 'm' will not be NULL. However,
5402 * if mirrors are reconfigured, we can temporarily get out
5403 * of sync in facet_revalidate(). We could "correct" the
5404 * mirror list before reaching here, but doing that would
5405 * not properly account the traffic stats we've currently
5406 * accumulated for previous mirror configuration. */
5410 m->packet_count += packets;
5411 m->byte_count += bytes;
5416 /* Optimized flow revalidation.
5418 * It's a difficult problem, in general, to tell which facets need to have
5419 * their actions recalculated whenever the OpenFlow flow table changes. We
5420 * don't try to solve that general problem: for most kinds of OpenFlow flow
5421 * table changes, we recalculate the actions for every facet. This is
5422 * relatively expensive, but it's good enough if the OpenFlow flow table
5423 * doesn't change very often.
5425 * However, we can expect one particular kind of OpenFlow flow table change to
5426 * happen frequently: changes caused by MAC learning. To avoid wasting a lot
5427 * of CPU on revalidating every facet whenever MAC learning modifies the flow
5428 * table, we add a special case that applies to flow tables in which every rule
5429 * has the same form (that is, the same wildcards), except that the table is
5430 * also allowed to have a single "catch-all" flow that matches all packets. We
5431 * optimize this case by tagging all of the facets that resubmit into the table
5432 * and invalidating the same tag whenever a flow changes in that table. The
5433 * end result is that we revalidate just the facets that need it (and sometimes
5434 * a few more, but not all of the facets or even all of the facets that
5435 * resubmit to the table modified by MAC learning). */
5437 /* Calculates the tag to use for 'flow' and mask 'mask' when it is inserted
5438 * into an OpenFlow table with the given 'basis'. */
5440 rule_calculate_tag(const struct flow *flow, const struct minimask *mask,
5443 if (minimask_is_catchall(mask)) {
5446 uint32_t hash = flow_hash_in_minimask(flow, mask, secret);
5447 return tag_create_deterministic(hash);
5451 /* Following a change to OpenFlow table 'table_id' in 'ofproto', update the
5452 * taggability of that table.
5454 * This function must be called after *each* change to a flow table. If you
5455 * skip calling it on some changes then the pointer comparisons at the end can
5456 * be invalid if you get unlucky. For example, if a flow removal causes a
5457 * cls_table to be destroyed and then a flow insertion causes a cls_table with
5458 * different wildcards to be created with the same address, then this function
5459 * will incorrectly skip revalidation. */
5461 table_update_taggable(struct ofproto_dpif *ofproto, uint8_t table_id)
5463 struct table_dpif *table = &ofproto->tables[table_id];
5464 const struct oftable *oftable = &ofproto->up.tables[table_id];
5465 struct cls_table *catchall, *other;
5466 struct cls_table *t;
5468 catchall = other = NULL;
5470 switch (hmap_count(&oftable->cls.tables)) {
5472 /* We could tag this OpenFlow table but it would make the logic a
5473 * little harder and it's a corner case that doesn't seem worth it
5479 HMAP_FOR_EACH (t, hmap_node, &oftable->cls.tables) {
5480 if (cls_table_is_catchall(t)) {
5482 } else if (!other) {
5485 /* Indicate that we can't tag this by setting both tables to
5486 * NULL. (We know that 'catchall' is already NULL.) */
5493 /* Can't tag this table. */
5497 if (table->catchall_table != catchall || table->other_table != other) {
5498 table->catchall_table = catchall;
5499 table->other_table = other;
5500 ofproto->backer->need_revalidate = REV_FLOW_TABLE;
5504 /* Given 'rule' that has changed in some way (either it is a rule being
5505 * inserted, a rule being deleted, or a rule whose actions are being
5506 * modified), marks facets for revalidation to ensure that packets will be
5507 * forwarded correctly according to the new state of the flow table.
5509 * This function must be called after *each* change to a flow table. See
5510 * the comment on table_update_taggable() for more information. */
5512 rule_invalidate(const struct rule_dpif *rule)
5514 struct ofproto_dpif *ofproto = ofproto_dpif_cast(rule->up.ofproto);
5516 table_update_taggable(ofproto, rule->up.table_id);
5518 if (!ofproto->backer->need_revalidate) {
5519 struct table_dpif *table = &ofproto->tables[rule->up.table_id];
5521 if (table->other_table && rule->tag) {
5522 tag_set_add(&ofproto->backer->revalidate_set, rule->tag);
5524 ofproto->backer->need_revalidate = REV_FLOW_TABLE;
5530 set_frag_handling(struct ofproto *ofproto_,
5531 enum ofp_config_flags frag_handling)
5533 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
5534 if (frag_handling != OFPC_FRAG_REASM) {
5535 ofproto->backer->need_revalidate = REV_RECONFIGURE;
5543 packet_out(struct ofproto *ofproto_, struct ofpbuf *packet,
5544 const struct flow *flow,
5545 const struct ofpact *ofpacts, size_t ofpacts_len)
5547 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
5548 struct odputil_keybuf keybuf;
5549 struct dpif_flow_stats stats;
5550 struct xlate_out xout;
5551 struct xlate_in xin;
5555 ofpbuf_use_stack(&key, &keybuf, sizeof keybuf);
5556 odp_flow_key_from_flow(&key, flow,
5557 ofp_port_to_odp_port(ofproto, flow->in_port));
5559 dpif_flow_stats_extract(flow, packet, time_msec(), &stats);
5561 xlate_in_init(&xin, ofproto, flow, NULL, stats.tcp_flags, packet);
5562 xin.resubmit_stats = &stats;
5563 xin.ofpacts_len = ofpacts_len;
5564 xin.ofpacts = ofpacts;
5566 xlate_actions(&xin, &xout);
5567 dpif_execute(ofproto->backer->dpif, key.data, key.size,
5568 xout.odp_actions.data, xout.odp_actions.size, packet);
5569 xlate_out_uninit(&xout);
5577 set_netflow(struct ofproto *ofproto_,
5578 const struct netflow_options *netflow_options)
5580 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
5582 if (netflow_options) {
5583 if (!ofproto->netflow) {
5584 ofproto->netflow = netflow_create();
5586 return netflow_set_options(ofproto->netflow, netflow_options);
5588 netflow_destroy(ofproto->netflow);
5589 ofproto->netflow = NULL;
5595 get_netflow_ids(const struct ofproto *ofproto_,
5596 uint8_t *engine_type, uint8_t *engine_id)
5598 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
5600 dpif_get_netflow_ids(ofproto->backer->dpif, engine_type, engine_id);
5604 send_active_timeout(struct ofproto_dpif *ofproto, struct facet *facet)
5606 if (!facet_is_controller_flow(facet) &&
5607 netflow_active_timeout_expired(ofproto->netflow, &facet->nf_flow)) {
5608 struct subfacet *subfacet;
5609 struct ofexpired expired;
5611 LIST_FOR_EACH (subfacet, list_node, &facet->subfacets) {
5612 if (subfacet->path == SF_FAST_PATH) {
5613 struct dpif_flow_stats stats;
5615 subfacet_install(subfacet, &facet->xout.odp_actions,
5617 subfacet_update_stats(subfacet, &stats);
5621 expired.flow = facet->flow;
5622 expired.packet_count = facet->packet_count;
5623 expired.byte_count = facet->byte_count;
5624 expired.used = facet->used;
5625 netflow_expire(ofproto->netflow, &facet->nf_flow, &expired);
5630 send_netflow_active_timeouts(struct ofproto_dpif *ofproto)
5632 struct cls_cursor cursor;
5633 struct facet *facet;
5635 cls_cursor_init(&cursor, &ofproto->facets, NULL);
5636 CLS_CURSOR_FOR_EACH (facet, cr, &cursor) {
5637 send_active_timeout(ofproto, facet);
5641 static struct ofproto_dpif *
5642 ofproto_dpif_lookup(const char *name)
5644 struct ofproto_dpif *ofproto;
5646 HMAP_FOR_EACH_WITH_HASH (ofproto, all_ofproto_dpifs_node,
5647 hash_string(name, 0), &all_ofproto_dpifs) {
5648 if (!strcmp(ofproto->up.name, name)) {
5656 ofproto_unixctl_fdb_flush(struct unixctl_conn *conn, int argc,
5657 const char *argv[], void *aux OVS_UNUSED)
5659 struct ofproto_dpif *ofproto;
5662 ofproto = ofproto_dpif_lookup(argv[1]);
5664 unixctl_command_reply_error(conn, "no such bridge");
5667 mac_learning_flush(ofproto->ml, &ofproto->backer->revalidate_set);
5669 HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
5670 mac_learning_flush(ofproto->ml, &ofproto->backer->revalidate_set);
5674 unixctl_command_reply(conn, "table successfully flushed");
5678 ofproto_unixctl_fdb_show(struct unixctl_conn *conn, int argc OVS_UNUSED,
5679 const char *argv[], void *aux OVS_UNUSED)
5681 struct ds ds = DS_EMPTY_INITIALIZER;
5682 const struct ofproto_dpif *ofproto;
5683 const struct mac_entry *e;
5685 ofproto = ofproto_dpif_lookup(argv[1]);
5687 unixctl_command_reply_error(conn, "no such bridge");
5691 ds_put_cstr(&ds, " port VLAN MAC Age\n");
5692 LIST_FOR_EACH (e, lru_node, &ofproto->ml->lrus) {
5693 struct ofbundle *bundle = e->port.p;
5694 ds_put_format(&ds, "%5d %4d "ETH_ADDR_FMT" %3d\n",
5695 ofbundle_get_a_port(bundle)->odp_port,
5696 e->vlan, ETH_ADDR_ARGS(e->mac),
5697 mac_entry_age(ofproto->ml, e));
5699 unixctl_command_reply(conn, ds_cstr(&ds));
5704 struct xlate_out xout;
5705 struct xlate_in xin;
5711 trace_format_rule(struct ds *result, int level, const struct rule_dpif *rule)
5713 ds_put_char_multiple(result, '\t', level);
5715 ds_put_cstr(result, "No match\n");
5719 ds_put_format(result, "Rule: table=%"PRIu8" cookie=%#"PRIx64" ",
5720 rule ? rule->up.table_id : 0, ntohll(rule->up.flow_cookie));
5721 cls_rule_format(&rule->up.cr, result);
5722 ds_put_char(result, '\n');
5724 ds_put_char_multiple(result, '\t', level);
5725 ds_put_cstr(result, "OpenFlow ");
5726 ofpacts_format(rule->up.ofpacts, rule->up.ofpacts_len, result);
5727 ds_put_char(result, '\n');
5731 trace_format_flow(struct ds *result, int level, const char *title,
5732 struct trace_ctx *trace)
5734 ds_put_char_multiple(result, '\t', level);
5735 ds_put_format(result, "%s: ", title);
5736 if (flow_equal(&trace->xin.flow, &trace->flow)) {
5737 ds_put_cstr(result, "unchanged");
5739 flow_format(result, &trace->xin.flow);
5740 trace->flow = trace->xin.flow;
5742 ds_put_char(result, '\n');
5746 trace_format_regs(struct ds *result, int level, const char *title,
5747 struct trace_ctx *trace)
5751 ds_put_char_multiple(result, '\t', level);
5752 ds_put_format(result, "%s:", title);
5753 for (i = 0; i < FLOW_N_REGS; i++) {
5754 ds_put_format(result, " reg%zu=0x%"PRIx32, i, trace->flow.regs[i]);
5756 ds_put_char(result, '\n');
5760 trace_format_odp(struct ds *result, int level, const char *title,
5761 struct trace_ctx *trace)
5763 struct ofpbuf *odp_actions = &trace->xout.odp_actions;
5765 ds_put_char_multiple(result, '\t', level);
5766 ds_put_format(result, "%s: ", title);
5767 format_odp_actions(result, odp_actions->data, odp_actions->size);
5768 ds_put_char(result, '\n');
5772 trace_resubmit(struct xlate_in *xin, struct rule_dpif *rule, int recurse)
5774 struct trace_ctx *trace = CONTAINER_OF(xin, struct trace_ctx, xin);
5775 struct ds *result = trace->result;
5777 ds_put_char(result, '\n');
5778 trace_format_flow(result, recurse + 1, "Resubmitted flow", trace);
5779 trace_format_regs(result, recurse + 1, "Resubmitted regs", trace);
5780 trace_format_odp(result, recurse + 1, "Resubmitted odp", trace);
5781 trace_format_rule(result, recurse + 1, rule);
5785 trace_report(struct xlate_in *xin, const char *s, int recurse)
5787 struct trace_ctx *trace = CONTAINER_OF(xin, struct trace_ctx, xin);
5788 struct ds *result = trace->result;
5790 ds_put_char_multiple(result, '\t', recurse);
5791 ds_put_cstr(result, s);
5792 ds_put_char(result, '\n');
5796 ofproto_unixctl_trace(struct unixctl_conn *conn, int argc, const char *argv[],
5797 void *aux OVS_UNUSED)
5799 const struct dpif_backer *backer;
5800 struct ofproto_dpif *ofproto;
5801 struct ofpbuf odp_key;
5802 struct ofpbuf *packet;
5810 ofpbuf_init(&odp_key, 0);
5812 /* Handle "-generate" or a hex string as the last argument. */
5813 if (!strcmp(argv[argc - 1], "-generate")) {
5814 packet = ofpbuf_new(0);
5817 const char *error = eth_from_hex(argv[argc - 1], &packet);
5820 } else if (argc == 4) {
5821 /* The 3-argument form must end in "-generate' or a hex string. */
5822 unixctl_command_reply_error(conn, error);
5827 /* Parse the flow and determine whether a datapath or
5828 * bridge is specified. If function odp_flow_key_from_string()
5829 * returns 0, the flow is a odp_flow. If function
5830 * parse_ofp_exact_flow() returns 0, the flow is a br_flow. */
5831 if (!odp_flow_key_from_string(argv[argc - 1], NULL, &odp_key)) {
5832 /* If the odp_flow is the second argument,
5833 * the datapath name is the first argument. */
5835 const char *dp_type;
5836 if (!strncmp(argv[1], "ovs-", 4)) {
5837 dp_type = argv[1] + 4;
5841 backer = shash_find_data(&all_dpif_backers, dp_type);
5843 unixctl_command_reply_error(conn, "Cannot find datapath "
5848 /* No datapath name specified, so there should be only one
5850 struct shash_node *node;
5851 if (shash_count(&all_dpif_backers) != 1) {
5852 unixctl_command_reply_error(conn, "Must specify datapath "
5853 "name, there is more than one type of datapath");
5856 node = shash_first(&all_dpif_backers);
5857 backer = node->data;
5860 /* Extract the ofproto_dpif object from the ofproto_receive()
5862 if (ofproto_receive(backer, NULL, odp_key.data,
5863 odp_key.size, &flow, NULL, &ofproto, NULL)) {
5864 unixctl_command_reply_error(conn, "Invalid datapath flow");
5867 ds_put_format(&result, "Bridge: %s\n", ofproto->up.name);
5868 } else if (!parse_ofp_exact_flow(&flow, argv[argc - 1])) {
5870 unixctl_command_reply_error(conn, "Must specify bridge name");
5874 ofproto = ofproto_dpif_lookup(argv[1]);
5876 unixctl_command_reply_error(conn, "Unknown bridge name");
5880 unixctl_command_reply_error(conn, "Bad flow syntax");
5884 /* Generate a packet, if requested. */
5886 if (!packet->size) {
5887 flow_compose(packet, &flow);
5889 ds_put_cstr(&result, "Packet: ");
5890 s = ofp_packet_to_string(packet->data, packet->size);
5891 ds_put_cstr(&result, s);
5894 /* Use the metadata from the flow and the packet argument
5895 * to reconstruct the flow. */
5896 flow_extract(packet, flow.skb_priority, flow.skb_mark, NULL,
5897 flow.in_port, &flow);
5901 ofproto_trace(ofproto, &flow, packet, &result);
5902 unixctl_command_reply(conn, ds_cstr(&result));
5905 ds_destroy(&result);
5906 ofpbuf_delete(packet);
5907 ofpbuf_uninit(&odp_key);
5911 ofproto_trace(struct ofproto_dpif *ofproto, const struct flow *flow,
5912 const struct ofpbuf *packet, struct ds *ds)
5914 struct rule_dpif *rule;
5916 ds_put_cstr(ds, "Flow: ");
5917 flow_format(ds, flow);
5918 ds_put_char(ds, '\n');
5920 rule = rule_dpif_lookup(ofproto, flow, NULL);
5922 trace_format_rule(ds, 0, rule);
5923 if (rule == ofproto->miss_rule) {
5924 ds_put_cstr(ds, "\nNo match, flow generates \"packet in\"s.\n");
5925 } else if (rule == ofproto->no_packet_in_rule) {
5926 ds_put_cstr(ds, "\nNo match, packets dropped because "
5927 "OFPPC_NO_PACKET_IN is set on in_port.\n");
5928 } else if (rule == ofproto->drop_frags_rule) {
5929 ds_put_cstr(ds, "\nPackets dropped because they are IP fragments "
5930 "and the fragment handling mode is \"drop\".\n");
5934 uint64_t odp_actions_stub[1024 / 8];
5935 struct ofpbuf odp_actions;
5936 struct trace_ctx trace;
5940 tcp_flags = packet ? packet_get_tcp_flags(packet, flow) : 0;
5943 ofpbuf_use_stub(&odp_actions,
5944 odp_actions_stub, sizeof odp_actions_stub);
5945 xlate_in_init(&trace.xin, ofproto, flow, rule, tcp_flags, packet);
5946 trace.xin.resubmit_hook = trace_resubmit;
5947 trace.xin.report_hook = trace_report;
5949 xlate_actions(&trace.xin, &trace.xout);
5951 ds_put_char(ds, '\n');
5952 trace_format_flow(ds, 0, "Final flow", &trace);
5954 match_init(&match, flow, &trace.xout.wc);
5955 ds_put_cstr(ds, "Relevant fields: ");
5956 match_format(&match, ds, OFP_DEFAULT_PRIORITY);
5957 ds_put_char(ds, '\n');
5959 ds_put_cstr(ds, "Datapath actions: ");
5960 format_odp_actions(ds, trace.xout.odp_actions.data,
5961 trace.xout.odp_actions.size);
5963 if (trace.xout.slow) {
5964 ds_put_cstr(ds, "\nThis flow is handled by the userspace "
5965 "slow path because it:");
5966 switch (trace.xout.slow) {
5968 ds_put_cstr(ds, "\n\t- Consists of CFM packets.");
5971 ds_put_cstr(ds, "\n\t- Consists of LACP packets.");
5974 ds_put_cstr(ds, "\n\t- Consists of STP packets.");
5977 ds_put_cstr(ds, "\n\t- Consists of BFD packets.");
5979 case SLOW_CONTROLLER:
5980 ds_put_cstr(ds, "\n\t- Sends \"packet-in\" messages "
5981 "to the OpenFlow controller.");
5988 xlate_out_uninit(&trace.xout);
5993 ofproto_dpif_clog(struct unixctl_conn *conn OVS_UNUSED, int argc OVS_UNUSED,
5994 const char *argv[] OVS_UNUSED, void *aux OVS_UNUSED)
5997 unixctl_command_reply(conn, NULL);
6001 ofproto_dpif_unclog(struct unixctl_conn *conn OVS_UNUSED, int argc OVS_UNUSED,
6002 const char *argv[] OVS_UNUSED, void *aux OVS_UNUSED)
6005 unixctl_command_reply(conn, NULL);
6008 /* Runs a self-check of flow translations in 'ofproto'. Appends a message to
6009 * 'reply' describing the results. */
6011 ofproto_dpif_self_check__(struct ofproto_dpif *ofproto, struct ds *reply)
6013 struct cls_cursor cursor;
6014 struct facet *facet;
6018 cls_cursor_init(&cursor, &ofproto->facets, NULL);
6019 CLS_CURSOR_FOR_EACH (facet, cr, &cursor) {
6020 if (!facet_check_consistency(facet)) {
6025 ofproto->backer->need_revalidate = REV_INCONSISTENCY;
6029 ds_put_format(reply, "%s: self-check failed (%d errors)\n",
6030 ofproto->up.name, errors);
6032 ds_put_format(reply, "%s: self-check passed\n", ofproto->up.name);
6037 ofproto_dpif_self_check(struct unixctl_conn *conn,
6038 int argc, const char *argv[], void *aux OVS_UNUSED)
6040 struct ds reply = DS_EMPTY_INITIALIZER;
6041 struct ofproto_dpif *ofproto;
6044 ofproto = ofproto_dpif_lookup(argv[1]);
6046 unixctl_command_reply_error(conn, "Unknown ofproto (use "
6047 "ofproto/list for help)");
6050 ofproto_dpif_self_check__(ofproto, &reply);
6052 HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
6053 ofproto_dpif_self_check__(ofproto, &reply);
6057 unixctl_command_reply(conn, ds_cstr(&reply));
6061 /* Store the current ofprotos in 'ofproto_shash'. Returns a sorted list
6062 * of the 'ofproto_shash' nodes. It is the responsibility of the caller
6063 * to destroy 'ofproto_shash' and free the returned value. */
6064 static const struct shash_node **
6065 get_ofprotos(struct shash *ofproto_shash)
6067 const struct ofproto_dpif *ofproto;
6069 HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
6070 char *name = xasprintf("%s@%s", ofproto->up.type, ofproto->up.name);
6071 shash_add_nocopy(ofproto_shash, name, ofproto);
6074 return shash_sort(ofproto_shash);
6078 ofproto_unixctl_dpif_dump_dps(struct unixctl_conn *conn, int argc OVS_UNUSED,
6079 const char *argv[] OVS_UNUSED,
6080 void *aux OVS_UNUSED)
6082 struct ds ds = DS_EMPTY_INITIALIZER;
6083 struct shash ofproto_shash;
6084 const struct shash_node **sorted_ofprotos;
6087 shash_init(&ofproto_shash);
6088 sorted_ofprotos = get_ofprotos(&ofproto_shash);
6089 for (i = 0; i < shash_count(&ofproto_shash); i++) {
6090 const struct shash_node *node = sorted_ofprotos[i];
6091 ds_put_format(&ds, "%s\n", node->name);
6094 shash_destroy(&ofproto_shash);
6095 free(sorted_ofprotos);
6097 unixctl_command_reply(conn, ds_cstr(&ds));
6102 show_dp_rates(struct ds *ds, const char *heading,
6103 const struct avg_subfacet_rates *rates)
6105 ds_put_format(ds, "%s add rate: %5.3f/min, del rate: %5.3f/min\n",
6106 heading, rates->add_rate, rates->del_rate);
6110 dpif_show_backer(const struct dpif_backer *backer, struct ds *ds)
6112 const struct shash_node **ofprotos;
6113 struct ofproto_dpif *ofproto;
6114 struct shash ofproto_shash;
6115 uint64_t n_hit, n_missed;
6116 long long int minutes;
6119 n_hit = n_missed = 0;
6120 HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
6121 if (ofproto->backer == backer) {
6122 n_missed += ofproto->n_missed;
6123 n_hit += ofproto->n_hit;
6127 ds_put_format(ds, "%s: hit:%"PRIu64" missed:%"PRIu64"\n",
6128 dpif_name(backer->dpif), n_hit, n_missed);
6129 ds_put_format(ds, "\tflows: cur: %zu, avg: %u, max: %u,"
6130 " life span: %lldms\n", hmap_count(&backer->subfacets),
6131 backer->avg_n_subfacet, backer->max_n_subfacet,
6132 backer->avg_subfacet_life);
6134 minutes = (time_msec() - backer->created) / (1000 * 60);
6135 if (minutes >= 60) {
6136 show_dp_rates(ds, "\thourly avg:", &backer->hourly);
6138 if (minutes >= 60 * 24) {
6139 show_dp_rates(ds, "\tdaily avg:", &backer->daily);
6141 show_dp_rates(ds, "\toverall avg:", &backer->lifetime);
6143 shash_init(&ofproto_shash);
6144 ofprotos = get_ofprotos(&ofproto_shash);
6145 for (i = 0; i < shash_count(&ofproto_shash); i++) {
6146 struct ofproto_dpif *ofproto = ofprotos[i]->data;
6147 const struct shash_node **ports;
6150 if (ofproto->backer != backer) {
6154 ds_put_format(ds, "\t%s: hit:%"PRIu64" missed:%"PRIu64"\n",
6155 ofproto->up.name, ofproto->n_hit, ofproto->n_missed);
6157 ports = shash_sort(&ofproto->up.port_by_name);
6158 for (j = 0; j < shash_count(&ofproto->up.port_by_name); j++) {
6159 const struct shash_node *node = ports[j];
6160 struct ofport *ofport = node->data;
6164 ds_put_format(ds, "\t\t%s %u/", netdev_get_name(ofport->netdev),
6167 odp_port = ofp_port_to_odp_port(ofproto, ofport->ofp_port);
6168 if (odp_port != OVSP_NONE) {
6169 ds_put_format(ds, "%"PRIu32":", odp_port);
6171 ds_put_cstr(ds, "none:");
6174 ds_put_format(ds, " (%s", netdev_get_type(ofport->netdev));
6177 if (!netdev_get_config(ofport->netdev, &config)) {
6178 const struct smap_node **nodes;
6181 nodes = smap_sort(&config);
6182 for (i = 0; i < smap_count(&config); i++) {
6183 const struct smap_node *node = nodes[i];
6184 ds_put_format(ds, "%c %s=%s", i ? ',' : ':',
6185 node->key, node->value);
6189 smap_destroy(&config);
6191 ds_put_char(ds, ')');
6192 ds_put_char(ds, '\n');
6196 shash_destroy(&ofproto_shash);
6201 ofproto_unixctl_dpif_show(struct unixctl_conn *conn, int argc OVS_UNUSED,
6202 const char *argv[] OVS_UNUSED, void *aux OVS_UNUSED)
6204 struct ds ds = DS_EMPTY_INITIALIZER;
6205 const struct shash_node **backers;
6208 backers = shash_sort(&all_dpif_backers);
6209 for (i = 0; i < shash_count(&all_dpif_backers); i++) {
6210 dpif_show_backer(backers[i]->data, &ds);
6214 unixctl_command_reply(conn, ds_cstr(&ds));
6218 /* Dump the megaflow (facet) cache. This is useful to check the
6219 * correctness of flow wildcarding, since the same mechanism is used for
6220 * both xlate caching and kernel wildcarding.
6222 * It's important to note that in the output the flow description uses
6223 * OpenFlow (OFP) ports, but the actions use datapath (ODP) ports.
6225 * This command is only needed for advanced debugging, so it's not
6226 * documented in the man page. */
6228 ofproto_unixctl_dpif_dump_megaflows(struct unixctl_conn *conn,
6229 int argc OVS_UNUSED, const char *argv[],
6230 void *aux OVS_UNUSED)
6232 struct ds ds = DS_EMPTY_INITIALIZER;
6233 const struct ofproto_dpif *ofproto;
6234 long long int now = time_msec();
6235 struct cls_cursor cursor;
6236 struct facet *facet;
6238 ofproto = ofproto_dpif_lookup(argv[1]);
6240 unixctl_command_reply_error(conn, "no such bridge");
6244 cls_cursor_init(&cursor, &ofproto->facets, NULL);
6245 CLS_CURSOR_FOR_EACH (facet, cr, &cursor) {
6246 cls_rule_format(&facet->cr, &ds);
6247 ds_put_cstr(&ds, ", ");
6248 ds_put_format(&ds, "n_subfacets:%zu, ", list_size(&facet->subfacets));
6249 ds_put_format(&ds, "used:%.3fs, ", (now - facet->used) / 1000.0);
6250 ds_put_cstr(&ds, "Datapath actions: ");
6251 if (facet->xout.slow) {
6252 uint64_t slow_path_stub[128 / 8];
6253 const struct nlattr *actions;
6256 compose_slow_path(ofproto, &facet->flow, facet->xout.slow,
6257 slow_path_stub, sizeof slow_path_stub,
6258 &actions, &actions_len);
6259 format_odp_actions(&ds, actions, actions_len);
6261 format_odp_actions(&ds, facet->xout.odp_actions.data,
6262 facet->xout.odp_actions.size);
6264 ds_put_cstr(&ds, "\n");
6267 ds_chomp(&ds, '\n');
6268 unixctl_command_reply(conn, ds_cstr(&ds));
6273 ofproto_unixctl_dpif_dump_flows(struct unixctl_conn *conn,
6274 int argc OVS_UNUSED, const char *argv[],
6275 void *aux OVS_UNUSED)
6277 struct ds ds = DS_EMPTY_INITIALIZER;
6278 const struct ofproto_dpif *ofproto;
6279 struct subfacet *subfacet;
6281 ofproto = ofproto_dpif_lookup(argv[1]);
6283 unixctl_command_reply_error(conn, "no such bridge");
6287 update_stats(ofproto->backer);
6289 HMAP_FOR_EACH (subfacet, hmap_node, &ofproto->backer->subfacets) {
6290 struct facet *facet = subfacet->facet;
6292 if (ofproto_dpif_cast(facet->rule->up.ofproto) != ofproto) {
6296 odp_flow_key_format(subfacet->key, subfacet->key_len, &ds);
6298 ds_put_format(&ds, ", packets:%"PRIu64", bytes:%"PRIu64", used:",
6299 subfacet->dp_packet_count, subfacet->dp_byte_count);
6300 if (subfacet->used) {
6301 ds_put_format(&ds, "%.3fs",
6302 (time_msec() - subfacet->used) / 1000.0);
6304 ds_put_format(&ds, "never");
6306 if (subfacet->facet->tcp_flags) {
6307 ds_put_cstr(&ds, ", flags:");
6308 packet_format_tcp_flags(&ds, subfacet->facet->tcp_flags);
6311 ds_put_cstr(&ds, ", actions:");
6312 if (facet->xout.slow) {
6313 uint64_t slow_path_stub[128 / 8];
6314 const struct nlattr *actions;
6317 compose_slow_path(ofproto, &facet->flow, facet->xout.slow,
6318 slow_path_stub, sizeof slow_path_stub,
6319 &actions, &actions_len);
6320 format_odp_actions(&ds, actions, actions_len);
6322 format_odp_actions(&ds, facet->xout.odp_actions.data,
6323 facet->xout.odp_actions.size);
6325 ds_put_char(&ds, '\n');
6328 unixctl_command_reply(conn, ds_cstr(&ds));
6333 ofproto_unixctl_dpif_del_flows(struct unixctl_conn *conn,
6334 int argc OVS_UNUSED, const char *argv[],
6335 void *aux OVS_UNUSED)
6337 struct ds ds = DS_EMPTY_INITIALIZER;
6338 struct ofproto_dpif *ofproto;
6340 ofproto = ofproto_dpif_lookup(argv[1]);
6342 unixctl_command_reply_error(conn, "no such bridge");
6346 flush(&ofproto->up);
6348 unixctl_command_reply(conn, ds_cstr(&ds));
6353 ofproto_dpif_unixctl_init(void)
6355 static bool registered;
6361 unixctl_command_register(
6363 "[dp_name]|bridge odp_flow|br_flow [-generate|packet]",
6364 1, 3, ofproto_unixctl_trace, NULL);
6365 unixctl_command_register("fdb/flush", "[bridge]", 0, 1,
6366 ofproto_unixctl_fdb_flush, NULL);
6367 unixctl_command_register("fdb/show", "bridge", 1, 1,
6368 ofproto_unixctl_fdb_show, NULL);
6369 unixctl_command_register("ofproto/clog", "", 0, 0,
6370 ofproto_dpif_clog, NULL);
6371 unixctl_command_register("ofproto/unclog", "", 0, 0,
6372 ofproto_dpif_unclog, NULL);
6373 unixctl_command_register("ofproto/self-check", "[bridge]", 0, 1,
6374 ofproto_dpif_self_check, NULL);
6375 unixctl_command_register("dpif/dump-dps", "", 0, 0,
6376 ofproto_unixctl_dpif_dump_dps, NULL);
6377 unixctl_command_register("dpif/show", "", 0, 0, ofproto_unixctl_dpif_show,
6379 unixctl_command_register("dpif/dump-flows", "bridge", 1, 1,
6380 ofproto_unixctl_dpif_dump_flows, NULL);
6381 unixctl_command_register("dpif/del-flows", "bridge", 1, 1,
6382 ofproto_unixctl_dpif_del_flows, NULL);
6383 unixctl_command_register("dpif/dump-megaflows", "bridge", 1, 1,
6384 ofproto_unixctl_dpif_dump_megaflows, NULL);
6387 /* Linux VLAN device support (e.g. "eth0.10" for VLAN 10.)
6389 * This is deprecated. It is only for compatibility with broken device drivers
6390 * in old versions of Linux that do not properly support VLANs when VLAN
6391 * devices are not used. When broken device drivers are no longer in
6392 * widespread use, we will delete these interfaces. */
6395 set_realdev(struct ofport *ofport_, uint16_t realdev_ofp_port, int vid)
6397 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport_->ofproto);
6398 struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
6400 if (realdev_ofp_port == ofport->realdev_ofp_port
6401 && vid == ofport->vlandev_vid) {
6405 ofproto->backer->need_revalidate = REV_RECONFIGURE;
6407 if (ofport->realdev_ofp_port) {
6410 if (realdev_ofp_port && ofport->bundle) {
6411 /* vlandevs are enslaved to their realdevs, so they are not allowed to
6412 * themselves be part of a bundle. */
6413 bundle_set(ofport->up.ofproto, ofport->bundle, NULL);
6416 ofport->realdev_ofp_port = realdev_ofp_port;
6417 ofport->vlandev_vid = vid;
6419 if (realdev_ofp_port) {
6420 vsp_add(ofport, realdev_ofp_port, vid);
6427 hash_realdev_vid(uint16_t realdev_ofp_port, int vid)
6429 return hash_2words(realdev_ofp_port, vid);
6432 /* Returns the OFP port number of the Linux VLAN device that corresponds to
6433 * 'vlan_tci' on the network device with port number 'realdev_ofp_port' in
6434 * 'struct ofport_dpif'. For example, given 'realdev_ofp_port' of eth0 and
6435 * 'vlan_tci' 9, it would return the port number of eth0.9.
6437 * Unless VLAN splinters are enabled for port 'realdev_ofp_port', this
6438 * function just returns its 'realdev_ofp_port' argument. */
6440 vsp_realdev_to_vlandev(const struct ofproto_dpif *ofproto,
6441 uint16_t realdev_ofp_port, ovs_be16 vlan_tci)
6443 if (!hmap_is_empty(&ofproto->realdev_vid_map)) {
6444 int vid = vlan_tci_to_vid(vlan_tci);
6445 const struct vlan_splinter *vsp;
6447 HMAP_FOR_EACH_WITH_HASH (vsp, realdev_vid_node,
6448 hash_realdev_vid(realdev_ofp_port, vid),
6449 &ofproto->realdev_vid_map) {
6450 if (vsp->realdev_ofp_port == realdev_ofp_port
6451 && vsp->vid == vid) {
6452 return vsp->vlandev_ofp_port;
6456 return realdev_ofp_port;
6459 static struct vlan_splinter *
6460 vlandev_find(const struct ofproto_dpif *ofproto, uint16_t vlandev_ofp_port)
6462 struct vlan_splinter *vsp;
6464 HMAP_FOR_EACH_WITH_HASH (vsp, vlandev_node, hash_int(vlandev_ofp_port, 0),
6465 &ofproto->vlandev_map) {
6466 if (vsp->vlandev_ofp_port == vlandev_ofp_port) {
6474 /* Returns the OpenFlow port number of the "real" device underlying the Linux
6475 * VLAN device with OpenFlow port number 'vlandev_ofp_port' and stores the
6476 * VLAN VID of the Linux VLAN device in '*vid'. For example, given
6477 * 'vlandev_ofp_port' of eth0.9, it would return the OpenFlow port number of
6478 * eth0 and store 9 in '*vid'.
6480 * Returns 0 and does not modify '*vid' if 'vlandev_ofp_port' is not a Linux
6481 * VLAN device. Unless VLAN splinters are enabled, this is what this function
6484 vsp_vlandev_to_realdev(const struct ofproto_dpif *ofproto,
6485 uint16_t vlandev_ofp_port, int *vid)
6487 if (!hmap_is_empty(&ofproto->vlandev_map)) {
6488 const struct vlan_splinter *vsp;
6490 vsp = vlandev_find(ofproto, vlandev_ofp_port);
6495 return vsp->realdev_ofp_port;
6501 /* Given 'flow', a flow representing a packet received on 'ofproto', checks
6502 * whether 'flow->in_port' represents a Linux VLAN device. If so, changes
6503 * 'flow->in_port' to the "real" device backing the VLAN device, sets
6504 * 'flow->vlan_tci' to the VLAN VID, and returns true. Otherwise (which is
6505 * always the case unless VLAN splinters are enabled), returns false without
6506 * making any changes. */
6508 vsp_adjust_flow(const struct ofproto_dpif *ofproto, struct flow *flow)
6513 realdev = vsp_vlandev_to_realdev(ofproto, flow->in_port, &vid);
6518 /* Cause the flow to be processed as if it came in on the real device with
6519 * the VLAN device's VLAN ID. */
6520 flow->in_port = realdev;
6521 flow->vlan_tci = htons((vid & VLAN_VID_MASK) | VLAN_CFI);
6526 vsp_remove(struct ofport_dpif *port)
6528 struct ofproto_dpif *ofproto = ofproto_dpif_cast(port->up.ofproto);
6529 struct vlan_splinter *vsp;
6531 vsp = vlandev_find(ofproto, port->up.ofp_port);
6533 hmap_remove(&ofproto->vlandev_map, &vsp->vlandev_node);
6534 hmap_remove(&ofproto->realdev_vid_map, &vsp->realdev_vid_node);
6537 port->realdev_ofp_port = 0;
6539 VLOG_ERR("missing vlan device record");
6544 vsp_add(struct ofport_dpif *port, uint16_t realdev_ofp_port, int vid)
6546 struct ofproto_dpif *ofproto = ofproto_dpif_cast(port->up.ofproto);
6548 if (!vsp_vlandev_to_realdev(ofproto, port->up.ofp_port, NULL)
6549 && (vsp_realdev_to_vlandev(ofproto, realdev_ofp_port, htons(vid))
6550 == realdev_ofp_port)) {
6551 struct vlan_splinter *vsp;
6553 vsp = xmalloc(sizeof *vsp);
6554 hmap_insert(&ofproto->vlandev_map, &vsp->vlandev_node,
6555 hash_int(port->up.ofp_port, 0));
6556 hmap_insert(&ofproto->realdev_vid_map, &vsp->realdev_vid_node,
6557 hash_realdev_vid(realdev_ofp_port, vid));
6558 vsp->realdev_ofp_port = realdev_ofp_port;
6559 vsp->vlandev_ofp_port = port->up.ofp_port;
6562 port->realdev_ofp_port = realdev_ofp_port;
6564 VLOG_ERR("duplicate vlan device record");
6569 ofp_port_to_odp_port(const struct ofproto_dpif *ofproto, uint16_t ofp_port)
6571 const struct ofport_dpif *ofport = get_ofp_port(ofproto, ofp_port);
6572 return ofport ? ofport->odp_port : OVSP_NONE;
6575 static struct ofport_dpif *
6576 odp_port_to_ofport(const struct dpif_backer *backer, uint32_t odp_port)
6578 struct ofport_dpif *port;
6580 HMAP_FOR_EACH_IN_BUCKET (port, odp_port_node,
6581 hash_int(odp_port, 0),
6582 &backer->odp_to_ofport_map) {
6583 if (port->odp_port == odp_port) {
6592 odp_port_to_ofp_port(const struct ofproto_dpif *ofproto, uint32_t odp_port)
6594 struct ofport_dpif *port;
6596 port = odp_port_to_ofport(ofproto->backer, odp_port);
6597 if (port && &ofproto->up == port->up.ofproto) {
6598 return port->up.ofp_port;
6604 /* Compute exponentially weighted moving average, adding 'new' as the newest,
6605 * most heavily weighted element. 'base' designates the rate of decay: after
6606 * 'base' further updates, 'new''s weight in the EWMA decays to about 1/e
6609 exp_mavg(double *avg, int base, double new)
6611 *avg = (*avg * (base - 1) + new) / base;
6615 update_moving_averages(struct dpif_backer *backer)
6617 const int min_ms = 60 * 1000; /* milliseconds in one minute. */
6618 long long int minutes = (time_msec() - backer->created) / min_ms;
6621 backer->lifetime.add_rate = (double) backer->total_subfacet_add_count
6623 backer->lifetime.del_rate = (double) backer->total_subfacet_del_count
6626 backer->lifetime.add_rate = 0.0;
6627 backer->lifetime.del_rate = 0.0;
6630 /* Update hourly averages on the minute boundaries. */
6631 if (time_msec() - backer->last_minute >= min_ms) {
6632 exp_mavg(&backer->hourly.add_rate, 60, backer->subfacet_add_count);
6633 exp_mavg(&backer->hourly.del_rate, 60, backer->subfacet_del_count);
6635 /* Update daily averages on the hour boundaries. */
6636 if ((backer->last_minute - backer->created) / min_ms % 60 == 59) {
6637 exp_mavg(&backer->daily.add_rate, 24, backer->hourly.add_rate);
6638 exp_mavg(&backer->daily.del_rate, 24, backer->hourly.del_rate);
6641 backer->total_subfacet_add_count += backer->subfacet_add_count;
6642 backer->total_subfacet_del_count += backer->subfacet_del_count;
6643 backer->subfacet_add_count = 0;
6644 backer->subfacet_del_count = 0;
6645 backer->last_minute += min_ms;
6649 const struct ofproto_class ofproto_dpif_class = {
6684 port_is_lacp_current,
6685 NULL, /* rule_choose_table */
6692 rule_modify_actions,
6706 get_stp_port_status,
6713 is_mirror_output_bundle,
6714 forward_bpdu_changed,
6715 set_mac_table_config,