2 * Copyright (c) 2009, 2010, 2011, 2012 Nicira Networks.
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-provider.h"
26 #include "byte-order.h"
31 #include "dynamic-string.h"
32 #include "fail-open.h"
36 #include "mac-learning.h"
37 #include "meta-flow.h"
38 #include "multipath.h"
45 #include "ofp-print.h"
46 #include "ofproto-dpif-sflow.h"
47 #include "poll-loop.h"
49 #include "unaligned.h"
51 #include "vlan-bitmap.h"
54 VLOG_DEFINE_THIS_MODULE(ofproto_dpif);
56 COVERAGE_DEFINE(ofproto_dpif_ctlr_action);
57 COVERAGE_DEFINE(ofproto_dpif_expired);
58 COVERAGE_DEFINE(ofproto_dpif_no_packet_in);
59 COVERAGE_DEFINE(ofproto_dpif_xlate);
60 COVERAGE_DEFINE(facet_changed_rule);
61 COVERAGE_DEFINE(facet_invalidated);
62 COVERAGE_DEFINE(facet_revalidate);
63 COVERAGE_DEFINE(facet_unexpected);
65 /* Maximum depth of flow table recursion (due to resubmit actions) in a
66 * flow translation. */
67 #define MAX_RESUBMIT_RECURSION 32
69 /* Number of implemented OpenFlow tables. */
70 enum { N_TABLES = 255 };
71 BUILD_ASSERT_DECL(N_TABLES >= 1 && N_TABLES <= 255);
81 * - Do include packets and bytes from facets that have been deleted or
82 * whose own statistics have been folded into the rule.
84 * - Do include packets and bytes sent "by hand" that were accounted to
85 * the rule without any facet being involved (this is a rare corner
86 * case in rule_execute()).
88 * - Do not include packet or bytes that can be obtained from any facet's
89 * packet_count or byte_count member or that can be obtained from the
90 * datapath by, e.g., dpif_flow_get() for any subfacet.
92 uint64_t packet_count; /* Number of packets received. */
93 uint64_t byte_count; /* Number of bytes received. */
95 tag_type tag; /* Caches rule_calculate_tag() result. */
97 struct list facets; /* List of "struct facet"s. */
100 static struct rule_dpif *rule_dpif_cast(const struct rule *rule)
102 return rule ? CONTAINER_OF(rule, struct rule_dpif, up) : NULL;
105 static struct rule_dpif *rule_dpif_lookup(struct ofproto_dpif *,
106 const struct flow *, uint8_t table);
108 static void flow_push_stats(const struct rule_dpif *, const struct flow *,
109 uint64_t packets, uint64_t bytes,
112 static uint32_t rule_calculate_tag(const struct flow *,
113 const struct flow_wildcards *,
115 static void rule_invalidate(const struct rule_dpif *);
117 #define MAX_MIRRORS 32
118 typedef uint32_t mirror_mask_t;
119 #define MIRROR_MASK_C(X) UINT32_C(X)
120 BUILD_ASSERT_DECL(sizeof(mirror_mask_t) * CHAR_BIT >= MAX_MIRRORS);
122 struct ofproto_dpif *ofproto; /* Owning ofproto. */
123 size_t idx; /* In ofproto's "mirrors" array. */
124 void *aux; /* Key supplied by ofproto's client. */
125 char *name; /* Identifier for log messages. */
127 /* Selection criteria. */
128 struct hmapx srcs; /* Contains "struct ofbundle *"s. */
129 struct hmapx dsts; /* Contains "struct ofbundle *"s. */
130 unsigned long *vlans; /* Bitmap of chosen VLANs, NULL selects all. */
132 /* Output (exactly one of out == NULL and out_vlan == -1 is true). */
133 struct ofbundle *out; /* Output port or NULL. */
134 int out_vlan; /* Output VLAN or -1. */
135 mirror_mask_t dup_mirrors; /* Bitmap of mirrors with the same output. */
138 int64_t packet_count; /* Number of packets sent. */
139 int64_t byte_count; /* Number of bytes sent. */
142 static void mirror_destroy(struct ofmirror *);
143 static void update_mirror_stats(struct ofproto_dpif *ofproto,
144 mirror_mask_t mirrors,
145 uint64_t packets, uint64_t bytes);
148 struct ofproto_dpif *ofproto; /* Owning ofproto. */
149 struct hmap_node hmap_node; /* In struct ofproto's "bundles" hmap. */
150 void *aux; /* Key supplied by ofproto's client. */
151 char *name; /* Identifier for log messages. */
154 struct list ports; /* Contains "struct ofport"s. */
155 enum port_vlan_mode vlan_mode; /* VLAN mode */
156 int vlan; /* -1=trunk port, else a 12-bit VLAN ID. */
157 unsigned long *trunks; /* Bitmap of trunked VLANs, if 'vlan' == -1.
158 * NULL if all VLANs are trunked. */
159 struct lacp *lacp; /* LACP if LACP is enabled, otherwise NULL. */
160 struct bond *bond; /* Nonnull iff more than one port. */
161 bool use_priority_tags; /* Use 802.1p tag for frames in VLAN 0? */
164 bool floodable; /* True if no port has OFPPC_NO_FLOOD set. */
166 /* Port mirroring info. */
167 mirror_mask_t src_mirrors; /* Mirrors triggered when packet received. */
168 mirror_mask_t dst_mirrors; /* Mirrors triggered when packet sent. */
169 mirror_mask_t mirror_out; /* Mirrors that output to this bundle. */
172 static void bundle_remove(struct ofport *);
173 static void bundle_update(struct ofbundle *);
174 static void bundle_destroy(struct ofbundle *);
175 static void bundle_del_port(struct ofport_dpif *);
176 static void bundle_run(struct ofbundle *);
177 static void bundle_wait(struct ofbundle *);
178 static struct ofbundle *lookup_input_bundle(struct ofproto_dpif *,
179 uint16_t in_port, bool warn);
181 /* A controller may use OFPP_NONE as the ingress port to indicate that
182 * it did not arrive on a "real" port. 'ofpp_none_bundle' exists for
183 * when an input bundle is needed for validation (e.g., mirroring or
184 * OFPP_NORMAL processing). It is not connected to an 'ofproto' or have
185 * any 'port' structs, so care must be taken when dealing with it. */
186 static struct ofbundle ofpp_none_bundle = {
188 .vlan_mode = PORT_VLAN_TRUNK
191 static void stp_run(struct ofproto_dpif *ofproto);
192 static void stp_wait(struct ofproto_dpif *ofproto);
194 static bool ofbundle_includes_vlan(const struct ofbundle *, uint16_t vlan);
196 struct action_xlate_ctx {
197 /* action_xlate_ctx_init() initializes these members. */
200 struct ofproto_dpif *ofproto;
202 /* Flow to which the OpenFlow actions apply. xlate_actions() will modify
203 * this flow when actions change header fields. */
206 /* The packet corresponding to 'flow', or a null pointer if we are
207 * revalidating without a packet to refer to. */
208 const struct ofpbuf *packet;
210 /* Should OFPP_NORMAL MAC learning and NXAST_LEARN actions execute? We
211 * want to execute them if we are actually processing a packet, or if we
212 * are accounting for packets that the datapath has processed, but not if
213 * we are just revalidating. */
216 /* Cookie of the currently matching rule, or 0. */
219 /* If nonnull, called just before executing a resubmit action.
221 * This is normally null so the client has to set it manually after
222 * calling action_xlate_ctx_init(). */
223 void (*resubmit_hook)(struct action_xlate_ctx *, struct rule_dpif *);
225 /* xlate_actions() initializes and uses these members. The client might want
226 * to look at them after it returns. */
228 struct ofpbuf *odp_actions; /* Datapath actions. */
229 tag_type tags; /* Tags associated with actions. */
230 bool may_set_up_flow; /* True ordinarily; false if the actions must
231 * be reassessed for every packet. */
232 bool has_learn; /* Actions include NXAST_LEARN? */
233 bool has_normal; /* Actions output to OFPP_NORMAL? */
234 uint16_t nf_output_iface; /* Output interface index for NetFlow. */
235 mirror_mask_t mirrors; /* Bitmap of associated mirrors. */
237 /* xlate_actions() initializes and uses these members, but the client has no
238 * reason to look at them. */
240 int recurse; /* Recursion level, via xlate_table_action. */
241 struct flow base_flow; /* Flow at the last commit. */
242 uint32_t orig_skb_priority; /* Priority when packet arrived. */
243 uint8_t table_id; /* OpenFlow table ID where flow was found. */
244 uint32_t sflow_n_outputs; /* Number of output ports. */
245 uint16_t sflow_odp_port; /* Output port for composing sFlow action. */
246 uint16_t user_cookie_offset;/* Used for user_action_cookie fixup. */
247 bool exit; /* No further actions should be processed. */
250 static void action_xlate_ctx_init(struct action_xlate_ctx *,
251 struct ofproto_dpif *, const struct flow *,
252 ovs_be16 initial_tci, ovs_be64 cookie,
253 const struct ofpbuf *);
254 static struct ofpbuf *xlate_actions(struct action_xlate_ctx *,
255 const union ofp_action *in, size_t n_in);
257 /* An exact-match instantiation of an OpenFlow flow.
259 * A facet associates a "struct flow", which represents the Open vSwitch
260 * userspace idea of an exact-match flow, with one or more subfacets. Each
261 * subfacet tracks the datapath's idea of the exact-match flow equivalent to
262 * the facet. When the kernel module (or other dpif implementation) and Open
263 * vSwitch userspace agree on the definition of a flow key, there is exactly
264 * one subfacet per facet. If the dpif implementation supports more-specific
265 * flow matching than userspace, however, a facet can have more than one
266 * subfacet, each of which corresponds to some distinction in flow that
267 * userspace simply doesn't understand.
269 * Flow expiration works in terms of subfacets, so a facet must have at least
270 * one subfacet or it will never expire, leaking memory. */
273 struct hmap_node hmap_node; /* In owning ofproto's 'facets' hmap. */
274 struct list list_node; /* In owning rule's 'facets' list. */
275 struct rule_dpif *rule; /* Owning rule. */
278 struct list subfacets;
279 long long int used; /* Time last used; time created if not used. */
286 * - Do include packets and bytes sent "by hand", e.g. with
289 * - Do include packets and bytes that were obtained from the datapath
290 * when a subfacet's statistics were reset (e.g. dpif_flow_put() with
291 * DPIF_FP_ZERO_STATS).
293 * - Do not include packets or bytes that can be obtained from the
294 * datapath for any existing subfacet.
296 uint64_t packet_count; /* Number of packets received. */
297 uint64_t byte_count; /* Number of bytes received. */
299 /* Resubmit statistics. */
300 uint64_t prev_packet_count; /* Number of packets from last stats push. */
301 uint64_t prev_byte_count; /* Number of bytes from last stats push. */
302 long long int prev_used; /* Used time from last stats push. */
305 uint64_t accounted_bytes; /* Bytes processed by facet_account(). */
306 struct netflow_flow nf_flow; /* Per-flow NetFlow tracking data. */
308 /* Properties of datapath actions.
310 * Every subfacet has its own actions because actions can differ slightly
311 * between splintered and non-splintered subfacets due to the VLAN tag
312 * being initially different (present vs. absent). All of them have these
313 * properties in common so we just store one copy of them here. */
314 bool may_install; /* Reassess actions for every packet? */
315 bool has_learn; /* Actions include NXAST_LEARN? */
316 bool has_normal; /* Actions output to OFPP_NORMAL? */
317 tag_type tags; /* Tags that would require revalidation. */
318 mirror_mask_t mirrors; /* Bitmap of dependent mirrors. */
321 static struct facet *facet_create(struct rule_dpif *, const struct flow *);
322 static void facet_remove(struct facet *);
323 static void facet_free(struct facet *);
325 static struct facet *facet_find(struct ofproto_dpif *, const struct flow *);
326 static struct facet *facet_lookup_valid(struct ofproto_dpif *,
327 const struct flow *);
328 static bool facet_revalidate(struct facet *);
329 static bool facet_check_consistency(struct facet *);
331 static void facet_flush_stats(struct facet *);
333 static void facet_update_time(struct facet *, long long int used);
334 static void facet_reset_counters(struct facet *);
335 static void facet_push_stats(struct facet *);
336 static void facet_account(struct facet *);
338 static bool facet_is_controller_flow(struct facet *);
340 /* A dpif flow and actions associated with a facet.
342 * See also the large comment on struct facet. */
345 struct hmap_node hmap_node; /* In struct ofproto_dpif 'subfacets' list. */
346 struct list list_node; /* In struct facet's 'facets' list. */
347 struct facet *facet; /* Owning facet. */
351 * To save memory in the common case, 'key' is NULL if 'key_fitness' is
352 * ODP_FIT_PERFECT, that is, odp_flow_key_from_flow() can accurately
353 * regenerate the ODP flow key from ->facet->flow. */
354 enum odp_key_fitness key_fitness;
358 long long int used; /* Time last used; time created if not used. */
360 uint64_t dp_packet_count; /* Last known packet count in the datapath. */
361 uint64_t dp_byte_count; /* Last known byte count in the datapath. */
365 * These should be essentially identical for every subfacet in a facet, but
366 * may differ in trivial ways due to VLAN splinters. */
367 size_t actions_len; /* Number of bytes in actions[]. */
368 struct nlattr *actions; /* Datapath actions. */
370 bool installed; /* Installed in datapath? */
372 /* This value is normally the same as ->facet->flow.vlan_tci. Only VLAN
373 * splinters can cause it to differ. This value should be removed when
374 * the VLAN splinters feature is no longer needed. */
375 ovs_be16 initial_tci; /* Initial VLAN TCI value. */
378 static struct subfacet *subfacet_create(struct facet *, enum odp_key_fitness,
379 const struct nlattr *key,
380 size_t key_len, ovs_be16 initial_tci);
381 static struct subfacet *subfacet_find(struct ofproto_dpif *,
382 const struct nlattr *key, size_t key_len);
383 static void subfacet_destroy(struct subfacet *);
384 static void subfacet_destroy__(struct subfacet *);
385 static void subfacet_get_key(struct subfacet *, struct odputil_keybuf *,
387 static void subfacet_reset_dp_stats(struct subfacet *,
388 struct dpif_flow_stats *);
389 static void subfacet_update_time(struct subfacet *, long long int used);
390 static void subfacet_update_stats(struct subfacet *,
391 const struct dpif_flow_stats *);
392 static void subfacet_make_actions(struct subfacet *,
393 const struct ofpbuf *packet);
394 static int subfacet_install(struct subfacet *,
395 const struct nlattr *actions, size_t actions_len,
396 struct dpif_flow_stats *);
397 static void subfacet_uninstall(struct subfacet *);
403 struct ofbundle *bundle; /* Bundle that contains this port, if any. */
404 struct list bundle_node; /* In struct ofbundle's "ports" list. */
405 struct cfm *cfm; /* Connectivity Fault Management, if any. */
406 tag_type tag; /* Tag associated with this port. */
407 uint32_t bond_stable_id; /* stable_id to use as bond slave, or 0. */
408 bool may_enable; /* May be enabled in bonds. */
411 struct stp_port *stp_port; /* Spanning Tree Protocol, if any. */
412 enum stp_state stp_state; /* Always STP_DISABLED if STP not in use. */
413 long long int stp_state_entered;
415 struct hmap priorities; /* Map of attached 'priority_to_dscp's. */
417 /* Linux VLAN device support (e.g. "eth0.10" for VLAN 10.)
419 * This is deprecated. It is only for compatibility with broken device
420 * drivers in old versions of Linux that do not properly support VLANs when
421 * VLAN devices are not used. When broken device drivers are no longer in
422 * widespread use, we will delete these interfaces. */
423 uint16_t realdev_ofp_port;
427 /* Node in 'ofport_dpif''s 'priorities' map. Used to maintain a map from
428 * 'priority' (the datapath's term for QoS queue) to the dscp bits which all
429 * traffic egressing the 'ofport' with that priority should be marked with. */
430 struct priority_to_dscp {
431 struct hmap_node hmap_node; /* Node in 'ofport_dpif''s 'priorities' map. */
432 uint32_t priority; /* Priority of this queue (see struct flow). */
434 uint8_t dscp; /* DSCP bits to mark outgoing traffic with. */
437 /* Linux VLAN device support (e.g. "eth0.10" for VLAN 10.)
439 * This is deprecated. It is only for compatibility with broken device drivers
440 * in old versions of Linux that do not properly support VLANs when VLAN
441 * devices are not used. When broken device drivers are no longer in
442 * widespread use, we will delete these interfaces. */
443 struct vlan_splinter {
444 struct hmap_node realdev_vid_node;
445 struct hmap_node vlandev_node;
446 uint16_t realdev_ofp_port;
447 uint16_t vlandev_ofp_port;
451 static uint32_t vsp_realdev_to_vlandev(const struct ofproto_dpif *,
452 uint32_t realdev, ovs_be16 vlan_tci);
453 static uint16_t vsp_vlandev_to_realdev(const struct ofproto_dpif *,
454 uint16_t vlandev, int *vid);
455 static void vsp_remove(struct ofport_dpif *);
456 static void vsp_add(struct ofport_dpif *, uint16_t realdev_ofp_port, int vid);
458 static struct ofport_dpif *
459 ofport_dpif_cast(const struct ofport *ofport)
461 assert(ofport->ofproto->ofproto_class == &ofproto_dpif_class);
462 return ofport ? CONTAINER_OF(ofport, struct ofport_dpif, up) : NULL;
465 static void port_run(struct ofport_dpif *);
466 static void port_wait(struct ofport_dpif *);
467 static int set_cfm(struct ofport *, const struct cfm_settings *);
468 static void ofport_clear_priorities(struct ofport_dpif *);
470 struct dpif_completion {
471 struct list list_node;
472 struct ofoperation *op;
475 /* Extra information about a classifier table.
476 * Currently used just for optimized flow revalidation. */
478 /* If either of these is nonnull, then this table has a form that allows
479 * flows to be tagged to avoid revalidating most flows for the most common
480 * kinds of flow table changes. */
481 struct cls_table *catchall_table; /* Table that wildcards all fields. */
482 struct cls_table *other_table; /* Table with any other wildcard set. */
483 uint32_t basis; /* Keeps each table's tags separate. */
486 struct ofproto_dpif {
487 struct hmap_node all_ofproto_dpifs_node; /* In 'all_ofproto_dpifs'. */
496 struct netflow *netflow;
497 struct dpif_sflow *sflow;
498 struct hmap bundles; /* Contains "struct ofbundle"s. */
499 struct mac_learning *ml;
500 struct ofmirror *mirrors[MAX_MIRRORS];
501 bool has_bonded_bundles;
504 struct timer next_expiration;
508 struct hmap subfacets;
511 struct table_dpif tables[N_TABLES];
512 bool need_revalidate;
513 struct tag_set revalidate_set;
515 /* Support for debugging async flow mods. */
516 struct list completions;
518 bool has_bundle_action; /* True when the first bundle action appears. */
519 struct netdev_stats stats; /* To account packets generated and consumed in
524 long long int stp_last_tick;
526 /* VLAN splinters. */
527 struct hmap realdev_vid_map; /* (realdev,vid) -> vlandev. */
528 struct hmap vlandev_map; /* vlandev -> (realdev,vid). */
531 /* Defer flow mod completion until "ovs-appctl ofproto/unclog"? (Useful only
532 * for debugging the asynchronous flow_mod implementation.) */
535 /* All existing ofproto_dpif instances, indexed by ->up.name. */
536 static struct hmap all_ofproto_dpifs = HMAP_INITIALIZER(&all_ofproto_dpifs);
538 static void ofproto_dpif_unixctl_init(void);
540 static struct ofproto_dpif *
541 ofproto_dpif_cast(const struct ofproto *ofproto)
543 assert(ofproto->ofproto_class == &ofproto_dpif_class);
544 return CONTAINER_OF(ofproto, struct ofproto_dpif, up);
547 static struct ofport_dpif *get_ofp_port(struct ofproto_dpif *,
549 static struct ofport_dpif *get_odp_port(struct ofproto_dpif *,
552 /* Packet processing. */
553 static void update_learning_table(struct ofproto_dpif *,
554 const struct flow *, int vlan,
557 #define FLOW_MISS_MAX_BATCH 50
558 static int handle_upcalls(struct ofproto_dpif *, unsigned int max_batch);
560 /* Flow expiration. */
561 static int expire(struct ofproto_dpif *);
564 static void send_netflow_active_timeouts(struct ofproto_dpif *);
567 static int send_packet(const struct ofport_dpif *, struct ofpbuf *packet);
569 compose_sflow_action(const struct ofproto_dpif *, struct ofpbuf *odp_actions,
570 const struct flow *, uint32_t odp_port);
571 static void add_mirror_actions(struct action_xlate_ctx *ctx,
572 const struct flow *flow);
573 /* Global variables. */
574 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
576 /* Factory functions. */
579 enumerate_types(struct sset *types)
581 dp_enumerate_types(types);
585 enumerate_names(const char *type, struct sset *names)
587 return dp_enumerate_names(type, names);
591 del(const char *type, const char *name)
596 error = dpif_open(name, type, &dpif);
598 error = dpif_delete(dpif);
604 /* Basic life-cycle. */
606 static struct ofproto *
609 struct ofproto_dpif *ofproto = xmalloc(sizeof *ofproto);
614 dealloc(struct ofproto *ofproto_)
616 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
621 construct(struct ofproto *ofproto_)
623 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
624 const char *name = ofproto->up.name;
628 error = dpif_create_and_open(name, ofproto->up.type, &ofproto->dpif);
630 VLOG_ERR("failed to open datapath %s: %s", name, strerror(error));
634 ofproto->max_ports = dpif_get_max_ports(ofproto->dpif);
635 ofproto->n_matches = 0;
637 dpif_flow_flush(ofproto->dpif);
638 dpif_recv_purge(ofproto->dpif);
640 error = dpif_recv_set(ofproto->dpif, true);
642 VLOG_ERR("failed to listen on datapath %s: %s", name, strerror(error));
643 dpif_close(ofproto->dpif);
647 ofproto->netflow = NULL;
648 ofproto->sflow = NULL;
650 hmap_init(&ofproto->bundles);
651 ofproto->ml = mac_learning_create();
652 for (i = 0; i < MAX_MIRRORS; i++) {
653 ofproto->mirrors[i] = NULL;
655 ofproto->has_bonded_bundles = false;
657 timer_set_duration(&ofproto->next_expiration, 1000);
659 hmap_init(&ofproto->facets);
660 hmap_init(&ofproto->subfacets);
662 for (i = 0; i < N_TABLES; i++) {
663 struct table_dpif *table = &ofproto->tables[i];
665 table->catchall_table = NULL;
666 table->other_table = NULL;
667 table->basis = random_uint32();
669 ofproto->need_revalidate = false;
670 tag_set_init(&ofproto->revalidate_set);
672 list_init(&ofproto->completions);
674 ofproto_dpif_unixctl_init();
676 ofproto->has_bundle_action = false;
678 hmap_init(&ofproto->vlandev_map);
679 hmap_init(&ofproto->realdev_vid_map);
681 hmap_insert(&all_ofproto_dpifs, &ofproto->all_ofproto_dpifs_node,
682 hash_string(ofproto->up.name, 0));
683 memset(&ofproto->stats, 0, sizeof ofproto->stats);
685 ofproto_init_tables(ofproto_, N_TABLES);
691 complete_operations(struct ofproto_dpif *ofproto)
693 struct dpif_completion *c, *next;
695 LIST_FOR_EACH_SAFE (c, next, list_node, &ofproto->completions) {
696 ofoperation_complete(c->op, 0);
697 list_remove(&c->list_node);
703 destruct(struct ofproto *ofproto_)
705 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
706 struct rule_dpif *rule, *next_rule;
707 struct oftable *table;
710 hmap_remove(&all_ofproto_dpifs, &ofproto->all_ofproto_dpifs_node);
711 complete_operations(ofproto);
713 OFPROTO_FOR_EACH_TABLE (table, &ofproto->up) {
714 struct cls_cursor cursor;
716 cls_cursor_init(&cursor, &table->cls, NULL);
717 CLS_CURSOR_FOR_EACH_SAFE (rule, next_rule, up.cr, &cursor) {
718 ofproto_rule_destroy(&rule->up);
722 for (i = 0; i < MAX_MIRRORS; i++) {
723 mirror_destroy(ofproto->mirrors[i]);
726 netflow_destroy(ofproto->netflow);
727 dpif_sflow_destroy(ofproto->sflow);
728 hmap_destroy(&ofproto->bundles);
729 mac_learning_destroy(ofproto->ml);
731 hmap_destroy(&ofproto->facets);
732 hmap_destroy(&ofproto->subfacets);
734 hmap_destroy(&ofproto->vlandev_map);
735 hmap_destroy(&ofproto->realdev_vid_map);
737 dpif_close(ofproto->dpif);
741 run_fast(struct ofproto *ofproto_)
743 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
746 /* Handle one or more batches of upcalls, until there's nothing left to do
747 * or until we do a fixed total amount of work.
749 * We do work in batches because it can be much cheaper to set up a number
750 * of flows and fire off their patches all at once. We do multiple batches
751 * because in some cases handling a packet can cause another packet to be
752 * queued almost immediately as part of the return flow. Both
753 * optimizations can make major improvements on some benchmarks and
754 * presumably for real traffic as well. */
756 while (work < FLOW_MISS_MAX_BATCH) {
757 int retval = handle_upcalls(ofproto, FLOW_MISS_MAX_BATCH - work);
767 run(struct ofproto *ofproto_)
769 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
770 struct ofport_dpif *ofport;
771 struct ofbundle *bundle;
775 complete_operations(ofproto);
777 dpif_run(ofproto->dpif);
779 error = run_fast(ofproto_);
784 if (timer_expired(&ofproto->next_expiration)) {
785 int delay = expire(ofproto);
786 timer_set_duration(&ofproto->next_expiration, delay);
789 if (ofproto->netflow) {
790 if (netflow_run(ofproto->netflow)) {
791 send_netflow_active_timeouts(ofproto);
794 if (ofproto->sflow) {
795 dpif_sflow_run(ofproto->sflow);
798 HMAP_FOR_EACH (ofport, up.hmap_node, &ofproto->up.ports) {
801 HMAP_FOR_EACH (bundle, hmap_node, &ofproto->bundles) {
806 mac_learning_run(ofproto->ml, &ofproto->revalidate_set);
808 /* Now revalidate if there's anything to do. */
809 if (ofproto->need_revalidate
810 || !tag_set_is_empty(&ofproto->revalidate_set)) {
811 struct tag_set revalidate_set = ofproto->revalidate_set;
812 bool revalidate_all = ofproto->need_revalidate;
813 struct facet *facet, *next;
815 /* Clear the revalidation flags. */
816 tag_set_init(&ofproto->revalidate_set);
817 ofproto->need_revalidate = false;
819 HMAP_FOR_EACH_SAFE (facet, next, hmap_node, &ofproto->facets) {
821 || tag_set_intersects(&revalidate_set, facet->tags)) {
822 facet_revalidate(facet);
827 /* Check the consistency of a random facet, to aid debugging. */
828 if (!hmap_is_empty(&ofproto->facets) && !ofproto->need_revalidate) {
831 facet = CONTAINER_OF(hmap_random_node(&ofproto->facets),
832 struct facet, hmap_node);
833 if (!tag_set_intersects(&ofproto->revalidate_set, facet->tags)) {
834 if (!facet_check_consistency(facet)) {
835 ofproto->need_revalidate = true;
844 wait(struct ofproto *ofproto_)
846 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
847 struct ofport_dpif *ofport;
848 struct ofbundle *bundle;
850 if (!clogged && !list_is_empty(&ofproto->completions)) {
851 poll_immediate_wake();
854 dpif_wait(ofproto->dpif);
855 dpif_recv_wait(ofproto->dpif);
856 if (ofproto->sflow) {
857 dpif_sflow_wait(ofproto->sflow);
859 if (!tag_set_is_empty(&ofproto->revalidate_set)) {
860 poll_immediate_wake();
862 HMAP_FOR_EACH (ofport, up.hmap_node, &ofproto->up.ports) {
865 HMAP_FOR_EACH (bundle, hmap_node, &ofproto->bundles) {
868 if (ofproto->netflow) {
869 netflow_wait(ofproto->netflow);
871 mac_learning_wait(ofproto->ml);
873 if (ofproto->need_revalidate) {
874 /* Shouldn't happen, but if it does just go around again. */
875 VLOG_DBG_RL(&rl, "need revalidate in ofproto_wait_cb()");
876 poll_immediate_wake();
878 timer_wait(&ofproto->next_expiration);
883 flush(struct ofproto *ofproto_)
885 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
886 struct facet *facet, *next_facet;
888 HMAP_FOR_EACH_SAFE (facet, next_facet, hmap_node, &ofproto->facets) {
889 /* Mark the facet as not installed so that facet_remove() doesn't
890 * bother trying to uninstall it. There is no point in uninstalling it
891 * individually since we are about to blow away all the facets with
892 * dpif_flow_flush(). */
893 struct subfacet *subfacet;
895 LIST_FOR_EACH (subfacet, list_node, &facet->subfacets) {
896 subfacet->installed = false;
897 subfacet->dp_packet_count = 0;
898 subfacet->dp_byte_count = 0;
902 dpif_flow_flush(ofproto->dpif);
906 get_features(struct ofproto *ofproto_ OVS_UNUSED,
907 bool *arp_match_ip, uint32_t *actions)
909 *arp_match_ip = true;
910 *actions = ((1u << OFPAT_OUTPUT) |
911 (1u << OFPAT_SET_VLAN_VID) |
912 (1u << OFPAT_SET_VLAN_PCP) |
913 (1u << OFPAT_STRIP_VLAN) |
914 (1u << OFPAT_SET_DL_SRC) |
915 (1u << OFPAT_SET_DL_DST) |
916 (1u << OFPAT_SET_NW_SRC) |
917 (1u << OFPAT_SET_NW_DST) |
918 (1u << OFPAT_SET_NW_TOS) |
919 (1u << OFPAT_SET_TP_SRC) |
920 (1u << OFPAT_SET_TP_DST) |
921 (1u << OFPAT_ENQUEUE));
925 get_tables(struct ofproto *ofproto_, struct ofp_table_stats *ots)
927 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
928 struct dpif_dp_stats s;
930 strcpy(ots->name, "classifier");
932 dpif_get_dp_stats(ofproto->dpif, &s);
933 put_32aligned_be64(&ots->lookup_count, htonll(s.n_hit + s.n_missed));
934 put_32aligned_be64(&ots->matched_count,
935 htonll(s.n_hit + ofproto->n_matches));
938 static struct ofport *
941 struct ofport_dpif *port = xmalloc(sizeof *port);
946 port_dealloc(struct ofport *port_)
948 struct ofport_dpif *port = ofport_dpif_cast(port_);
953 port_construct(struct ofport *port_)
955 struct ofport_dpif *port = ofport_dpif_cast(port_);
956 struct ofproto_dpif *ofproto = ofproto_dpif_cast(port->up.ofproto);
958 ofproto->need_revalidate = true;
959 port->odp_port = ofp_port_to_odp_port(port->up.ofp_port);
962 port->tag = tag_create_random();
963 port->may_enable = true;
964 port->stp_port = NULL;
965 port->stp_state = STP_DISABLED;
966 hmap_init(&port->priorities);
967 port->realdev_ofp_port = 0;
968 port->vlandev_vid = 0;
970 if (ofproto->sflow) {
971 dpif_sflow_add_port(ofproto->sflow, port_);
978 port_destruct(struct ofport *port_)
980 struct ofport_dpif *port = ofport_dpif_cast(port_);
981 struct ofproto_dpif *ofproto = ofproto_dpif_cast(port->up.ofproto);
983 ofproto->need_revalidate = true;
984 bundle_remove(port_);
985 set_cfm(port_, NULL);
986 if (ofproto->sflow) {
987 dpif_sflow_del_port(ofproto->sflow, port->odp_port);
990 ofport_clear_priorities(port);
991 hmap_destroy(&port->priorities);
995 port_modified(struct ofport *port_)
997 struct ofport_dpif *port = ofport_dpif_cast(port_);
999 if (port->bundle && port->bundle->bond) {
1000 bond_slave_set_netdev(port->bundle->bond, port, port->up.netdev);
1005 port_reconfigured(struct ofport *port_, ovs_be32 old_config)
1007 struct ofport_dpif *port = ofport_dpif_cast(port_);
1008 struct ofproto_dpif *ofproto = ofproto_dpif_cast(port->up.ofproto);
1009 ovs_be32 changed = old_config ^ port->up.opp.config;
1011 if (changed & htonl(OFPPC_NO_RECV | OFPPC_NO_RECV_STP |
1012 OFPPC_NO_FWD | OFPPC_NO_FLOOD)) {
1013 ofproto->need_revalidate = true;
1015 if (changed & htonl(OFPPC_NO_FLOOD) && port->bundle) {
1016 bundle_update(port->bundle);
1022 set_sflow(struct ofproto *ofproto_,
1023 const struct ofproto_sflow_options *sflow_options)
1025 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1026 struct dpif_sflow *ds = ofproto->sflow;
1028 if (sflow_options) {
1030 struct ofport_dpif *ofport;
1032 ds = ofproto->sflow = dpif_sflow_create(ofproto->dpif);
1033 HMAP_FOR_EACH (ofport, up.hmap_node, &ofproto->up.ports) {
1034 dpif_sflow_add_port(ds, &ofport->up);
1036 ofproto->need_revalidate = true;
1038 dpif_sflow_set_options(ds, sflow_options);
1041 dpif_sflow_destroy(ds);
1042 ofproto->need_revalidate = true;
1043 ofproto->sflow = NULL;
1050 set_cfm(struct ofport *ofport_, const struct cfm_settings *s)
1052 struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
1059 struct ofproto_dpif *ofproto;
1061 ofproto = ofproto_dpif_cast(ofport->up.ofproto);
1062 ofproto->need_revalidate = true;
1063 ofport->cfm = cfm_create(netdev_get_name(ofport->up.netdev));
1066 if (cfm_configure(ofport->cfm, s)) {
1072 cfm_destroy(ofport->cfm);
1078 get_cfm_fault(const struct ofport *ofport_)
1080 struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
1082 return ofport->cfm ? cfm_get_fault(ofport->cfm) : -1;
1086 get_cfm_remote_mpids(const struct ofport *ofport_, const uint64_t **rmps,
1089 struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
1092 cfm_get_remote_mpids(ofport->cfm, rmps, n_rmps);
1099 /* Spanning Tree. */
1102 send_bpdu_cb(struct ofpbuf *pkt, int port_num, void *ofproto_)
1104 struct ofproto_dpif *ofproto = ofproto_;
1105 struct stp_port *sp = stp_get_port(ofproto->stp, port_num);
1106 struct ofport_dpif *ofport;
1108 ofport = stp_port_get_aux(sp);
1110 VLOG_WARN_RL(&rl, "%s: cannot send BPDU on unknown port %d",
1111 ofproto->up.name, port_num);
1113 struct eth_header *eth = pkt->l2;
1115 netdev_get_etheraddr(ofport->up.netdev, eth->eth_src);
1116 if (eth_addr_is_zero(eth->eth_src)) {
1117 VLOG_WARN_RL(&rl, "%s: cannot send BPDU on port %d "
1118 "with unknown MAC", ofproto->up.name, port_num);
1120 send_packet(ofport, pkt);
1126 /* Configures STP on 'ofproto_' using the settings defined in 's'. */
1128 set_stp(struct ofproto *ofproto_, const struct ofproto_stp_settings *s)
1130 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1132 /* Only revalidate flows if the configuration changed. */
1133 if (!s != !ofproto->stp) {
1134 ofproto->need_revalidate = true;
1138 if (!ofproto->stp) {
1139 ofproto->stp = stp_create(ofproto_->name, s->system_id,
1140 send_bpdu_cb, ofproto);
1141 ofproto->stp_last_tick = time_msec();
1144 stp_set_bridge_id(ofproto->stp, s->system_id);
1145 stp_set_bridge_priority(ofproto->stp, s->priority);
1146 stp_set_hello_time(ofproto->stp, s->hello_time);
1147 stp_set_max_age(ofproto->stp, s->max_age);
1148 stp_set_forward_delay(ofproto->stp, s->fwd_delay);
1150 stp_destroy(ofproto->stp);
1151 ofproto->stp = NULL;
1158 get_stp_status(struct ofproto *ofproto_, struct ofproto_stp_status *s)
1160 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1164 s->bridge_id = stp_get_bridge_id(ofproto->stp);
1165 s->designated_root = stp_get_designated_root(ofproto->stp);
1166 s->root_path_cost = stp_get_root_path_cost(ofproto->stp);
1175 update_stp_port_state(struct ofport_dpif *ofport)
1177 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport->up.ofproto);
1178 enum stp_state state;
1180 /* Figure out new state. */
1181 state = ofport->stp_port ? stp_port_get_state(ofport->stp_port)
1185 if (ofport->stp_state != state) {
1189 VLOG_DBG_RL(&rl, "port %s: STP state changed from %s to %s",
1190 netdev_get_name(ofport->up.netdev),
1191 stp_state_name(ofport->stp_state),
1192 stp_state_name(state));
1193 if (stp_learn_in_state(ofport->stp_state)
1194 != stp_learn_in_state(state)) {
1195 /* xxx Learning action flows should also be flushed. */
1196 mac_learning_flush(ofproto->ml, &ofproto->revalidate_set);
1198 fwd_change = stp_forward_in_state(ofport->stp_state)
1199 != stp_forward_in_state(state);
1201 ofproto->need_revalidate = true;
1202 ofport->stp_state = state;
1203 ofport->stp_state_entered = time_msec();
1205 if (fwd_change && ofport->bundle) {
1206 bundle_update(ofport->bundle);
1209 /* Update the STP state bits in the OpenFlow port description. */
1210 of_state = (ofport->up.opp.state & htonl(~OFPPS_STP_MASK))
1211 | htonl(state == STP_LISTENING ? OFPPS_STP_LISTEN
1212 : state == STP_LEARNING ? OFPPS_STP_LEARN
1213 : state == STP_FORWARDING ? OFPPS_STP_FORWARD
1214 : state == STP_BLOCKING ? OFPPS_STP_BLOCK
1216 ofproto_port_set_state(&ofport->up, of_state);
1220 /* Configures STP on 'ofport_' using the settings defined in 's'. The
1221 * caller is responsible for assigning STP port numbers and ensuring
1222 * there are no duplicates. */
1224 set_stp_port(struct ofport *ofport_,
1225 const struct ofproto_port_stp_settings *s)
1227 struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
1228 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport->up.ofproto);
1229 struct stp_port *sp = ofport->stp_port;
1231 if (!s || !s->enable) {
1233 ofport->stp_port = NULL;
1234 stp_port_disable(sp);
1235 update_stp_port_state(ofport);
1238 } else if (sp && stp_port_no(sp) != s->port_num
1239 && ofport == stp_port_get_aux(sp)) {
1240 /* The port-id changed, so disable the old one if it's not
1241 * already in use by another port. */
1242 stp_port_disable(sp);
1245 sp = ofport->stp_port = stp_get_port(ofproto->stp, s->port_num);
1246 stp_port_enable(sp);
1248 stp_port_set_aux(sp, ofport);
1249 stp_port_set_priority(sp, s->priority);
1250 stp_port_set_path_cost(sp, s->path_cost);
1252 update_stp_port_state(ofport);
1258 get_stp_port_status(struct ofport *ofport_,
1259 struct ofproto_port_stp_status *s)
1261 struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
1262 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport->up.ofproto);
1263 struct stp_port *sp = ofport->stp_port;
1265 if (!ofproto->stp || !sp) {
1271 s->port_id = stp_port_get_id(sp);
1272 s->state = stp_port_get_state(sp);
1273 s->sec_in_state = (time_msec() - ofport->stp_state_entered) / 1000;
1274 s->role = stp_port_get_role(sp);
1275 stp_port_get_counts(sp, &s->tx_count, &s->rx_count, &s->error_count);
1281 stp_run(struct ofproto_dpif *ofproto)
1284 long long int now = time_msec();
1285 long long int elapsed = now - ofproto->stp_last_tick;
1286 struct stp_port *sp;
1289 stp_tick(ofproto->stp, MIN(INT_MAX, elapsed));
1290 ofproto->stp_last_tick = now;
1292 while (stp_get_changed_port(ofproto->stp, &sp)) {
1293 struct ofport_dpif *ofport = stp_port_get_aux(sp);
1296 update_stp_port_state(ofport);
1300 if (stp_check_and_reset_fdb_flush(ofproto->stp)) {
1301 mac_learning_flush(ofproto->ml, &ofproto->revalidate_set);
1307 stp_wait(struct ofproto_dpif *ofproto)
1310 poll_timer_wait(1000);
1314 /* Returns true if STP should process 'flow'. */
1316 stp_should_process_flow(const struct flow *flow)
1318 return eth_addr_equals(flow->dl_dst, eth_addr_stp);
1322 stp_process_packet(const struct ofport_dpif *ofport,
1323 const struct ofpbuf *packet)
1325 struct ofpbuf payload = *packet;
1326 struct eth_header *eth = payload.data;
1327 struct stp_port *sp = ofport->stp_port;
1329 /* Sink packets on ports that have STP disabled when the bridge has
1331 if (!sp || stp_port_get_state(sp) == STP_DISABLED) {
1335 /* Trim off padding on payload. */
1336 if (payload.size > ntohs(eth->eth_type) + ETH_HEADER_LEN) {
1337 payload.size = ntohs(eth->eth_type) + ETH_HEADER_LEN;
1340 if (ofpbuf_try_pull(&payload, ETH_HEADER_LEN + LLC_HEADER_LEN)) {
1341 stp_received_bpdu(sp, payload.data, payload.size);
1345 static struct priority_to_dscp *
1346 get_priority(const struct ofport_dpif *ofport, uint32_t priority)
1348 struct priority_to_dscp *pdscp;
1351 hash = hash_int(priority, 0);
1352 HMAP_FOR_EACH_IN_BUCKET (pdscp, hmap_node, hash, &ofport->priorities) {
1353 if (pdscp->priority == priority) {
1361 ofport_clear_priorities(struct ofport_dpif *ofport)
1363 struct priority_to_dscp *pdscp, *next;
1365 HMAP_FOR_EACH_SAFE (pdscp, next, hmap_node, &ofport->priorities) {
1366 hmap_remove(&ofport->priorities, &pdscp->hmap_node);
1372 set_queues(struct ofport *ofport_,
1373 const struct ofproto_port_queue *qdscp_list,
1376 struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
1377 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport->up.ofproto);
1378 struct hmap new = HMAP_INITIALIZER(&new);
1381 for (i = 0; i < n_qdscp; i++) {
1382 struct priority_to_dscp *pdscp;
1386 dscp = (qdscp_list[i].dscp << 2) & IP_DSCP_MASK;
1387 if (dpif_queue_to_priority(ofproto->dpif, qdscp_list[i].queue,
1392 pdscp = get_priority(ofport, priority);
1394 hmap_remove(&ofport->priorities, &pdscp->hmap_node);
1396 pdscp = xmalloc(sizeof *pdscp);
1397 pdscp->priority = priority;
1399 ofproto->need_revalidate = true;
1402 if (pdscp->dscp != dscp) {
1404 ofproto->need_revalidate = true;
1407 hmap_insert(&new, &pdscp->hmap_node, hash_int(pdscp->priority, 0));
1410 if (!hmap_is_empty(&ofport->priorities)) {
1411 ofport_clear_priorities(ofport);
1412 ofproto->need_revalidate = true;
1415 hmap_swap(&new, &ofport->priorities);
1423 /* Expires all MAC learning entries associated with 'bundle' and forces its
1424 * ofproto to revalidate every flow.
1426 * Normally MAC learning entries are removed only from the ofproto associated
1427 * with 'bundle', but if 'all_ofprotos' is true, then the MAC learning entries
1428 * are removed from every ofproto. When patch ports and SLB bonds are in use
1429 * and a VM migration happens and the gratuitous ARPs are somehow lost, this
1430 * avoids a MAC_ENTRY_IDLE_TIME delay before the migrated VM can communicate
1431 * with the host from which it migrated. */
1433 bundle_flush_macs(struct ofbundle *bundle, bool all_ofprotos)
1435 struct ofproto_dpif *ofproto = bundle->ofproto;
1436 struct mac_learning *ml = ofproto->ml;
1437 struct mac_entry *mac, *next_mac;
1439 ofproto->need_revalidate = true;
1440 LIST_FOR_EACH_SAFE (mac, next_mac, lru_node, &ml->lrus) {
1441 if (mac->port.p == bundle) {
1443 struct ofproto_dpif *o;
1445 HMAP_FOR_EACH (o, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
1447 struct mac_entry *e;
1449 e = mac_learning_lookup(o->ml, mac->mac, mac->vlan,
1452 tag_set_add(&o->revalidate_set, e->tag);
1453 mac_learning_expire(o->ml, e);
1459 mac_learning_expire(ml, mac);
1464 static struct ofbundle *
1465 bundle_lookup(const struct ofproto_dpif *ofproto, void *aux)
1467 struct ofbundle *bundle;
1469 HMAP_FOR_EACH_IN_BUCKET (bundle, hmap_node, hash_pointer(aux, 0),
1470 &ofproto->bundles) {
1471 if (bundle->aux == aux) {
1478 /* Looks up each of the 'n_auxes' pointers in 'auxes' as bundles and adds the
1479 * ones that are found to 'bundles'. */
1481 bundle_lookup_multiple(struct ofproto_dpif *ofproto,
1482 void **auxes, size_t n_auxes,
1483 struct hmapx *bundles)
1487 hmapx_init(bundles);
1488 for (i = 0; i < n_auxes; i++) {
1489 struct ofbundle *bundle = bundle_lookup(ofproto, auxes[i]);
1491 hmapx_add(bundles, bundle);
1497 bundle_update(struct ofbundle *bundle)
1499 struct ofport_dpif *port;
1501 bundle->floodable = true;
1502 LIST_FOR_EACH (port, bundle_node, &bundle->ports) {
1503 if (port->up.opp.config & htonl(OFPPC_NO_FLOOD)) {
1504 bundle->floodable = false;
1511 bundle_del_port(struct ofport_dpif *port)
1513 struct ofbundle *bundle = port->bundle;
1515 bundle->ofproto->need_revalidate = true;
1517 list_remove(&port->bundle_node);
1518 port->bundle = NULL;
1521 lacp_slave_unregister(bundle->lacp, port);
1524 bond_slave_unregister(bundle->bond, port);
1527 bundle_update(bundle);
1531 bundle_add_port(struct ofbundle *bundle, uint32_t ofp_port,
1532 struct lacp_slave_settings *lacp,
1533 uint32_t bond_stable_id)
1535 struct ofport_dpif *port;
1537 port = get_ofp_port(bundle->ofproto, ofp_port);
1542 if (port->bundle != bundle) {
1543 bundle->ofproto->need_revalidate = true;
1545 bundle_del_port(port);
1548 port->bundle = bundle;
1549 list_push_back(&bundle->ports, &port->bundle_node);
1550 if (port->up.opp.config & htonl(OFPPC_NO_FLOOD)) {
1551 bundle->floodable = false;
1555 port->bundle->ofproto->need_revalidate = true;
1556 lacp_slave_register(bundle->lacp, port, lacp);
1559 port->bond_stable_id = bond_stable_id;
1565 bundle_destroy(struct ofbundle *bundle)
1567 struct ofproto_dpif *ofproto;
1568 struct ofport_dpif *port, *next_port;
1575 ofproto = bundle->ofproto;
1576 for (i = 0; i < MAX_MIRRORS; i++) {
1577 struct ofmirror *m = ofproto->mirrors[i];
1579 if (m->out == bundle) {
1581 } else if (hmapx_find_and_delete(&m->srcs, bundle)
1582 || hmapx_find_and_delete(&m->dsts, bundle)) {
1583 ofproto->need_revalidate = true;
1588 LIST_FOR_EACH_SAFE (port, next_port, bundle_node, &bundle->ports) {
1589 bundle_del_port(port);
1592 bundle_flush_macs(bundle, true);
1593 hmap_remove(&ofproto->bundles, &bundle->hmap_node);
1595 free(bundle->trunks);
1596 lacp_destroy(bundle->lacp);
1597 bond_destroy(bundle->bond);
1602 bundle_set(struct ofproto *ofproto_, void *aux,
1603 const struct ofproto_bundle_settings *s)
1605 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1606 bool need_flush = false;
1607 struct ofport_dpif *port;
1608 struct ofbundle *bundle;
1609 unsigned long *trunks;
1615 bundle_destroy(bundle_lookup(ofproto, aux));
1619 assert(s->n_slaves == 1 || s->bond != NULL);
1620 assert((s->lacp != NULL) == (s->lacp_slaves != NULL));
1622 bundle = bundle_lookup(ofproto, aux);
1624 bundle = xmalloc(sizeof *bundle);
1626 bundle->ofproto = ofproto;
1627 hmap_insert(&ofproto->bundles, &bundle->hmap_node,
1628 hash_pointer(aux, 0));
1630 bundle->name = NULL;
1632 list_init(&bundle->ports);
1633 bundle->vlan_mode = PORT_VLAN_TRUNK;
1635 bundle->trunks = NULL;
1636 bundle->use_priority_tags = s->use_priority_tags;
1637 bundle->lacp = NULL;
1638 bundle->bond = NULL;
1640 bundle->floodable = true;
1642 bundle->src_mirrors = 0;
1643 bundle->dst_mirrors = 0;
1644 bundle->mirror_out = 0;
1647 if (!bundle->name || strcmp(s->name, bundle->name)) {
1649 bundle->name = xstrdup(s->name);
1654 if (!bundle->lacp) {
1655 ofproto->need_revalidate = true;
1656 bundle->lacp = lacp_create();
1658 lacp_configure(bundle->lacp, s->lacp);
1660 lacp_destroy(bundle->lacp);
1661 bundle->lacp = NULL;
1664 /* Update set of ports. */
1666 for (i = 0; i < s->n_slaves; i++) {
1667 if (!bundle_add_port(bundle, s->slaves[i],
1668 s->lacp ? &s->lacp_slaves[i] : NULL,
1669 s->bond_stable_ids ? s->bond_stable_ids[i] : 0)) {
1673 if (!ok || list_size(&bundle->ports) != s->n_slaves) {
1674 struct ofport_dpif *next_port;
1676 LIST_FOR_EACH_SAFE (port, next_port, bundle_node, &bundle->ports) {
1677 for (i = 0; i < s->n_slaves; i++) {
1678 if (s->slaves[i] == port->up.ofp_port) {
1683 bundle_del_port(port);
1687 assert(list_size(&bundle->ports) <= s->n_slaves);
1689 if (list_is_empty(&bundle->ports)) {
1690 bundle_destroy(bundle);
1694 /* Set VLAN tagging mode */
1695 if (s->vlan_mode != bundle->vlan_mode
1696 || s->use_priority_tags != bundle->use_priority_tags) {
1697 bundle->vlan_mode = s->vlan_mode;
1698 bundle->use_priority_tags = s->use_priority_tags;
1703 vlan = (s->vlan_mode == PORT_VLAN_TRUNK ? -1
1704 : s->vlan >= 0 && s->vlan <= 4095 ? s->vlan
1706 if (vlan != bundle->vlan) {
1707 bundle->vlan = vlan;
1711 /* Get trunked VLANs. */
1712 switch (s->vlan_mode) {
1713 case PORT_VLAN_ACCESS:
1717 case PORT_VLAN_TRUNK:
1718 trunks = (unsigned long *) s->trunks;
1721 case PORT_VLAN_NATIVE_UNTAGGED:
1722 case PORT_VLAN_NATIVE_TAGGED:
1723 if (vlan != 0 && (!s->trunks
1724 || !bitmap_is_set(s->trunks, vlan)
1725 || bitmap_is_set(s->trunks, 0))) {
1726 /* Force trunking the native VLAN and prohibit trunking VLAN 0. */
1728 trunks = bitmap_clone(s->trunks, 4096);
1730 trunks = bitmap_allocate1(4096);
1732 bitmap_set1(trunks, vlan);
1733 bitmap_set0(trunks, 0);
1735 trunks = (unsigned long *) s->trunks;
1742 if (!vlan_bitmap_equal(trunks, bundle->trunks)) {
1743 free(bundle->trunks);
1744 if (trunks == s->trunks) {
1745 bundle->trunks = vlan_bitmap_clone(trunks);
1747 bundle->trunks = trunks;
1752 if (trunks != s->trunks) {
1757 if (!list_is_short(&bundle->ports)) {
1758 bundle->ofproto->has_bonded_bundles = true;
1760 if (bond_reconfigure(bundle->bond, s->bond)) {
1761 ofproto->need_revalidate = true;
1764 bundle->bond = bond_create(s->bond);
1765 ofproto->need_revalidate = true;
1768 LIST_FOR_EACH (port, bundle_node, &bundle->ports) {
1769 bond_slave_register(bundle->bond, port, port->bond_stable_id,
1773 bond_destroy(bundle->bond);
1774 bundle->bond = NULL;
1777 /* If we changed something that would affect MAC learning, un-learn
1778 * everything on this port and force flow revalidation. */
1780 bundle_flush_macs(bundle, false);
1787 bundle_remove(struct ofport *port_)
1789 struct ofport_dpif *port = ofport_dpif_cast(port_);
1790 struct ofbundle *bundle = port->bundle;
1793 bundle_del_port(port);
1794 if (list_is_empty(&bundle->ports)) {
1795 bundle_destroy(bundle);
1796 } else if (list_is_short(&bundle->ports)) {
1797 bond_destroy(bundle->bond);
1798 bundle->bond = NULL;
1804 send_pdu_cb(void *port_, const void *pdu, size_t pdu_size)
1806 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 10);
1807 struct ofport_dpif *port = port_;
1808 uint8_t ea[ETH_ADDR_LEN];
1811 error = netdev_get_etheraddr(port->up.netdev, ea);
1813 struct ofpbuf packet;
1816 ofpbuf_init(&packet, 0);
1817 packet_pdu = eth_compose(&packet, eth_addr_lacp, ea, ETH_TYPE_LACP,
1819 memcpy(packet_pdu, pdu, pdu_size);
1821 send_packet(port, &packet);
1822 ofpbuf_uninit(&packet);
1824 VLOG_ERR_RL(&rl, "port %s: cannot obtain Ethernet address of iface "
1825 "%s (%s)", port->bundle->name,
1826 netdev_get_name(port->up.netdev), strerror(error));
1831 bundle_send_learning_packets(struct ofbundle *bundle)
1833 struct ofproto_dpif *ofproto = bundle->ofproto;
1834 int error, n_packets, n_errors;
1835 struct mac_entry *e;
1837 error = n_packets = n_errors = 0;
1838 LIST_FOR_EACH (e, lru_node, &ofproto->ml->lrus) {
1839 if (e->port.p != bundle) {
1840 struct ofpbuf *learning_packet;
1841 struct ofport_dpif *port;
1845 /* The assignment to "port" is unnecessary but makes "grep"ing for
1846 * struct ofport_dpif more effective. */
1847 learning_packet = bond_compose_learning_packet(bundle->bond,
1851 ret = send_packet(port, learning_packet);
1852 ofpbuf_delete(learning_packet);
1862 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
1863 VLOG_WARN_RL(&rl, "bond %s: %d errors sending %d gratuitous learning "
1864 "packets, last error was: %s",
1865 bundle->name, n_errors, n_packets, strerror(error));
1867 VLOG_DBG("bond %s: sent %d gratuitous learning packets",
1868 bundle->name, n_packets);
1873 bundle_run(struct ofbundle *bundle)
1876 lacp_run(bundle->lacp, send_pdu_cb);
1879 struct ofport_dpif *port;
1881 LIST_FOR_EACH (port, bundle_node, &bundle->ports) {
1882 bond_slave_set_may_enable(bundle->bond, port, port->may_enable);
1885 bond_run(bundle->bond, &bundle->ofproto->revalidate_set,
1886 lacp_status(bundle->lacp));
1887 if (bond_should_send_learning_packets(bundle->bond)) {
1888 bundle_send_learning_packets(bundle);
1894 bundle_wait(struct ofbundle *bundle)
1897 lacp_wait(bundle->lacp);
1900 bond_wait(bundle->bond);
1907 mirror_scan(struct ofproto_dpif *ofproto)
1911 for (idx = 0; idx < MAX_MIRRORS; idx++) {
1912 if (!ofproto->mirrors[idx]) {
1919 static struct ofmirror *
1920 mirror_lookup(struct ofproto_dpif *ofproto, void *aux)
1924 for (i = 0; i < MAX_MIRRORS; i++) {
1925 struct ofmirror *mirror = ofproto->mirrors[i];
1926 if (mirror && mirror->aux == aux) {
1934 /* Update the 'dup_mirrors' member of each of the ofmirrors in 'ofproto'. */
1936 mirror_update_dups(struct ofproto_dpif *ofproto)
1940 for (i = 0; i < MAX_MIRRORS; i++) {
1941 struct ofmirror *m = ofproto->mirrors[i];
1944 m->dup_mirrors = MIRROR_MASK_C(1) << i;
1948 for (i = 0; i < MAX_MIRRORS; i++) {
1949 struct ofmirror *m1 = ofproto->mirrors[i];
1956 for (j = i + 1; j < MAX_MIRRORS; j++) {
1957 struct ofmirror *m2 = ofproto->mirrors[j];
1959 if (m2 && m1->out == m2->out && m1->out_vlan == m2->out_vlan) {
1960 m1->dup_mirrors |= MIRROR_MASK_C(1) << j;
1961 m2->dup_mirrors |= m1->dup_mirrors;
1968 mirror_set(struct ofproto *ofproto_, void *aux,
1969 const struct ofproto_mirror_settings *s)
1971 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1972 mirror_mask_t mirror_bit;
1973 struct ofbundle *bundle;
1974 struct ofmirror *mirror;
1975 struct ofbundle *out;
1976 struct hmapx srcs; /* Contains "struct ofbundle *"s. */
1977 struct hmapx dsts; /* Contains "struct ofbundle *"s. */
1980 mirror = mirror_lookup(ofproto, aux);
1982 mirror_destroy(mirror);
1988 idx = mirror_scan(ofproto);
1990 VLOG_WARN("bridge %s: maximum of %d port mirrors reached, "
1992 ofproto->up.name, MAX_MIRRORS, s->name);
1996 mirror = ofproto->mirrors[idx] = xzalloc(sizeof *mirror);
1997 mirror->ofproto = ofproto;
2000 mirror->out_vlan = -1;
2001 mirror->name = NULL;
2004 if (!mirror->name || strcmp(s->name, mirror->name)) {
2006 mirror->name = xstrdup(s->name);
2009 /* Get the new configuration. */
2010 if (s->out_bundle) {
2011 out = bundle_lookup(ofproto, s->out_bundle);
2013 mirror_destroy(mirror);
2019 out_vlan = s->out_vlan;
2021 bundle_lookup_multiple(ofproto, s->srcs, s->n_srcs, &srcs);
2022 bundle_lookup_multiple(ofproto, s->dsts, s->n_dsts, &dsts);
2024 /* If the configuration has not changed, do nothing. */
2025 if (hmapx_equals(&srcs, &mirror->srcs)
2026 && hmapx_equals(&dsts, &mirror->dsts)
2027 && vlan_bitmap_equal(mirror->vlans, s->src_vlans)
2028 && mirror->out == out
2029 && mirror->out_vlan == out_vlan)
2031 hmapx_destroy(&srcs);
2032 hmapx_destroy(&dsts);
2036 hmapx_swap(&srcs, &mirror->srcs);
2037 hmapx_destroy(&srcs);
2039 hmapx_swap(&dsts, &mirror->dsts);
2040 hmapx_destroy(&dsts);
2042 free(mirror->vlans);
2043 mirror->vlans = vlan_bitmap_clone(s->src_vlans);
2046 mirror->out_vlan = out_vlan;
2048 /* Update bundles. */
2049 mirror_bit = MIRROR_MASK_C(1) << mirror->idx;
2050 HMAP_FOR_EACH (bundle, hmap_node, &mirror->ofproto->bundles) {
2051 if (hmapx_contains(&mirror->srcs, bundle)) {
2052 bundle->src_mirrors |= mirror_bit;
2054 bundle->src_mirrors &= ~mirror_bit;
2057 if (hmapx_contains(&mirror->dsts, bundle)) {
2058 bundle->dst_mirrors |= mirror_bit;
2060 bundle->dst_mirrors &= ~mirror_bit;
2063 if (mirror->out == bundle) {
2064 bundle->mirror_out |= mirror_bit;
2066 bundle->mirror_out &= ~mirror_bit;
2070 ofproto->need_revalidate = true;
2071 mac_learning_flush(ofproto->ml, &ofproto->revalidate_set);
2072 mirror_update_dups(ofproto);
2078 mirror_destroy(struct ofmirror *mirror)
2080 struct ofproto_dpif *ofproto;
2081 mirror_mask_t mirror_bit;
2082 struct ofbundle *bundle;
2088 ofproto = mirror->ofproto;
2089 ofproto->need_revalidate = true;
2090 mac_learning_flush(ofproto->ml, &ofproto->revalidate_set);
2092 mirror_bit = MIRROR_MASK_C(1) << mirror->idx;
2093 HMAP_FOR_EACH (bundle, hmap_node, &ofproto->bundles) {
2094 bundle->src_mirrors &= ~mirror_bit;
2095 bundle->dst_mirrors &= ~mirror_bit;
2096 bundle->mirror_out &= ~mirror_bit;
2099 hmapx_destroy(&mirror->srcs);
2100 hmapx_destroy(&mirror->dsts);
2101 free(mirror->vlans);
2103 ofproto->mirrors[mirror->idx] = NULL;
2107 mirror_update_dups(ofproto);
2111 mirror_get_stats(struct ofproto *ofproto_, void *aux,
2112 uint64_t *packets, uint64_t *bytes)
2114 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
2115 struct ofmirror *mirror = mirror_lookup(ofproto, aux);
2118 *packets = *bytes = UINT64_MAX;
2122 *packets = mirror->packet_count;
2123 *bytes = mirror->byte_count;
2129 set_flood_vlans(struct ofproto *ofproto_, unsigned long *flood_vlans)
2131 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
2132 if (mac_learning_set_flood_vlans(ofproto->ml, flood_vlans)) {
2133 mac_learning_flush(ofproto->ml, &ofproto->revalidate_set);
2139 is_mirror_output_bundle(const struct ofproto *ofproto_, void *aux)
2141 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
2142 struct ofbundle *bundle = bundle_lookup(ofproto, aux);
2143 return bundle && bundle->mirror_out != 0;
2147 forward_bpdu_changed(struct ofproto *ofproto_)
2149 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
2150 /* Revalidate cached flows whenever forward_bpdu option changes. */
2151 ofproto->need_revalidate = true;
2156 static struct ofport_dpif *
2157 get_ofp_port(struct ofproto_dpif *ofproto, uint16_t ofp_port)
2159 struct ofport *ofport = ofproto_get_port(&ofproto->up, ofp_port);
2160 return ofport ? ofport_dpif_cast(ofport) : NULL;
2163 static struct ofport_dpif *
2164 get_odp_port(struct ofproto_dpif *ofproto, uint32_t odp_port)
2166 return get_ofp_port(ofproto, odp_port_to_ofp_port(odp_port));
2170 ofproto_port_from_dpif_port(struct ofproto_port *ofproto_port,
2171 struct dpif_port *dpif_port)
2173 ofproto_port->name = dpif_port->name;
2174 ofproto_port->type = dpif_port->type;
2175 ofproto_port->ofp_port = odp_port_to_ofp_port(dpif_port->port_no);
2179 port_run(struct ofport_dpif *ofport)
2181 bool enable = netdev_get_carrier(ofport->up.netdev);
2184 cfm_run(ofport->cfm);
2186 if (cfm_should_send_ccm(ofport->cfm)) {
2187 struct ofpbuf packet;
2189 ofpbuf_init(&packet, 0);
2190 cfm_compose_ccm(ofport->cfm, &packet, ofport->up.opp.hw_addr);
2191 send_packet(ofport, &packet);
2192 ofpbuf_uninit(&packet);
2195 enable = enable && !cfm_get_fault(ofport->cfm)
2196 && cfm_get_opup(ofport->cfm);
2199 if (ofport->bundle) {
2200 enable = enable && lacp_slave_may_enable(ofport->bundle->lacp, ofport);
2203 if (ofport->may_enable != enable) {
2204 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport->up.ofproto);
2206 if (ofproto->has_bundle_action) {
2207 ofproto->need_revalidate = true;
2211 ofport->may_enable = enable;
2215 port_wait(struct ofport_dpif *ofport)
2218 cfm_wait(ofport->cfm);
2223 port_query_by_name(const struct ofproto *ofproto_, const char *devname,
2224 struct ofproto_port *ofproto_port)
2226 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
2227 struct dpif_port dpif_port;
2230 error = dpif_port_query_by_name(ofproto->dpif, devname, &dpif_port);
2232 ofproto_port_from_dpif_port(ofproto_port, &dpif_port);
2238 port_add(struct ofproto *ofproto_, struct netdev *netdev, uint16_t *ofp_portp)
2240 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
2244 error = dpif_port_add(ofproto->dpif, netdev, &odp_port);
2246 *ofp_portp = odp_port_to_ofp_port(odp_port);
2252 port_del(struct ofproto *ofproto_, uint16_t ofp_port)
2254 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
2257 error = dpif_port_del(ofproto->dpif, ofp_port_to_odp_port(ofp_port));
2259 struct ofport_dpif *ofport = get_ofp_port(ofproto, ofp_port);
2261 /* The caller is going to close ofport->up.netdev. If this is a
2262 * bonded port, then the bond is using that netdev, so remove it
2263 * from the bond. The client will need to reconfigure everything
2264 * after deleting ports, so then the slave will get re-added. */
2265 bundle_remove(&ofport->up);
2272 port_get_stats(const struct ofport *ofport_, struct netdev_stats *stats)
2274 struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
2277 error = netdev_get_stats(ofport->up.netdev, stats);
2279 if (!error && ofport->odp_port == OVSP_LOCAL) {
2280 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport->up.ofproto);
2282 /* ofproto->stats.tx_packets represents packets that we created
2283 * internally and sent to some port (e.g. packets sent with
2284 * send_packet()). Account for them as if they had come from
2285 * OFPP_LOCAL and got forwarded. */
2287 if (stats->rx_packets != UINT64_MAX) {
2288 stats->rx_packets += ofproto->stats.tx_packets;
2291 if (stats->rx_bytes != UINT64_MAX) {
2292 stats->rx_bytes += ofproto->stats.tx_bytes;
2295 /* ofproto->stats.rx_packets represents packets that were received on
2296 * some port and we processed internally and dropped (e.g. STP).
2297 * Account fro them as if they had been forwarded to OFPP_LOCAL. */
2299 if (stats->tx_packets != UINT64_MAX) {
2300 stats->tx_packets += ofproto->stats.rx_packets;
2303 if (stats->tx_bytes != UINT64_MAX) {
2304 stats->tx_bytes += ofproto->stats.rx_bytes;
2311 /* Account packets for LOCAL port. */
2313 ofproto_update_local_port_stats(const struct ofproto *ofproto_,
2314 size_t tx_size, size_t rx_size)
2316 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
2319 ofproto->stats.rx_packets++;
2320 ofproto->stats.rx_bytes += rx_size;
2323 ofproto->stats.tx_packets++;
2324 ofproto->stats.tx_bytes += tx_size;
2328 struct port_dump_state {
2329 struct dpif_port_dump dump;
2334 port_dump_start(const struct ofproto *ofproto_, void **statep)
2336 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
2337 struct port_dump_state *state;
2339 *statep = state = xmalloc(sizeof *state);
2340 dpif_port_dump_start(&state->dump, ofproto->dpif);
2341 state->done = false;
2346 port_dump_next(const struct ofproto *ofproto_ OVS_UNUSED, void *state_,
2347 struct ofproto_port *port)
2349 struct port_dump_state *state = state_;
2350 struct dpif_port dpif_port;
2352 if (dpif_port_dump_next(&state->dump, &dpif_port)) {
2353 ofproto_port_from_dpif_port(port, &dpif_port);
2356 int error = dpif_port_dump_done(&state->dump);
2358 return error ? error : EOF;
2363 port_dump_done(const struct ofproto *ofproto_ OVS_UNUSED, void *state_)
2365 struct port_dump_state *state = state_;
2368 dpif_port_dump_done(&state->dump);
2375 port_poll(const struct ofproto *ofproto_, char **devnamep)
2377 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
2378 return dpif_port_poll(ofproto->dpif, devnamep);
2382 port_poll_wait(const struct ofproto *ofproto_)
2384 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
2385 dpif_port_poll_wait(ofproto->dpif);
2389 port_is_lacp_current(const struct ofport *ofport_)
2391 const struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
2392 return (ofport->bundle && ofport->bundle->lacp
2393 ? lacp_slave_is_current(ofport->bundle->lacp, ofport)
2397 /* Upcall handling. */
2399 /* Flow miss batching.
2401 * Some dpifs implement operations faster when you hand them off in a batch.
2402 * To allow batching, "struct flow_miss" queues the dpif-related work needed
2403 * for a given flow. Each "struct flow_miss" corresponds to sending one or
2404 * more packets, plus possibly installing the flow in the dpif.
2406 * So far we only batch the operations that affect flow setup time the most.
2407 * It's possible to batch more than that, but the benefit might be minimal. */
2409 struct hmap_node hmap_node;
2411 enum odp_key_fitness key_fitness;
2412 const struct nlattr *key;
2414 ovs_be16 initial_tci;
2415 struct list packets;
2418 struct flow_miss_op {
2419 struct dpif_op dpif_op;
2420 struct subfacet *subfacet;
2423 /* Sends an OFPT_PACKET_IN message for 'packet' of type OFPR_NO_MATCH to each
2424 * OpenFlow controller as necessary according to their individual
2425 * configurations. */
2427 send_packet_in_miss(struct ofproto_dpif *ofproto, const struct ofpbuf *packet,
2428 const struct flow *flow)
2430 struct ofputil_packet_in pin;
2432 pin.packet = packet->data;
2433 pin.packet_len = packet->size;
2434 pin.total_len = packet->size;
2435 pin.reason = OFPR_NO_MATCH;
2440 pin.buffer_id = 0; /* not yet known */
2441 pin.send_len = 0; /* not used for flow table misses */
2443 flow_get_metadata(flow, &pin.fmd);
2445 /* Registers aren't meaningful on a miss. */
2446 memset(pin.fmd.reg_masks, 0, sizeof pin.fmd.reg_masks);
2448 connmgr_send_packet_in(ofproto->up.connmgr, &pin, flow);
2452 process_special(struct ofproto_dpif *ofproto, const struct flow *flow,
2453 const struct ofpbuf *packet)
2455 struct ofport_dpif *ofport = get_ofp_port(ofproto, flow->in_port);
2461 if (ofport->cfm && cfm_should_process_flow(ofport->cfm, flow)) {
2463 cfm_process_heartbeat(ofport->cfm, packet);
2466 } else if (ofport->bundle && ofport->bundle->lacp
2467 && flow->dl_type == htons(ETH_TYPE_LACP)) {
2469 lacp_process_packet(ofport->bundle->lacp, ofport, packet);
2472 } else if (ofproto->stp && stp_should_process_flow(flow)) {
2474 stp_process_packet(ofport, packet);
2481 static struct flow_miss *
2482 flow_miss_create(struct hmap *todo, const struct flow *flow,
2483 enum odp_key_fitness key_fitness,
2484 const struct nlattr *key, size_t key_len,
2485 ovs_be16 initial_tci)
2487 uint32_t hash = flow_hash(flow, 0);
2488 struct flow_miss *miss;
2490 HMAP_FOR_EACH_WITH_HASH (miss, hmap_node, hash, todo) {
2491 if (flow_equal(&miss->flow, flow)) {
2496 miss = xmalloc(sizeof *miss);
2497 hmap_insert(todo, &miss->hmap_node, hash);
2499 miss->key_fitness = key_fitness;
2501 miss->key_len = key_len;
2502 miss->initial_tci = initial_tci;
2503 list_init(&miss->packets);
2508 handle_flow_miss(struct ofproto_dpif *ofproto, struct flow_miss *miss,
2509 struct flow_miss_op *ops, size_t *n_ops)
2511 const struct flow *flow = &miss->flow;
2512 struct ofpbuf *packet, *next_packet;
2513 struct subfacet *subfacet;
2514 struct facet *facet;
2516 facet = facet_lookup_valid(ofproto, flow);
2518 struct rule_dpif *rule;
2520 rule = rule_dpif_lookup(ofproto, flow, 0);
2522 /* Don't send a packet-in if OFPPC_NO_PACKET_IN asserted. */
2523 struct ofport_dpif *port = get_ofp_port(ofproto, flow->in_port);
2525 if (port->up.opp.config & htonl(OFPPC_NO_PACKET_IN)) {
2526 COVERAGE_INC(ofproto_dpif_no_packet_in);
2527 /* XXX install 'drop' flow entry */
2531 VLOG_WARN_RL(&rl, "packet-in on unknown port %"PRIu16,
2535 LIST_FOR_EACH (packet, list_node, &miss->packets) {
2536 send_packet_in_miss(ofproto, packet, flow);
2542 facet = facet_create(rule, flow);
2545 subfacet = subfacet_create(facet,
2546 miss->key_fitness, miss->key, miss->key_len,
2549 LIST_FOR_EACH_SAFE (packet, next_packet, list_node, &miss->packets) {
2550 struct dpif_flow_stats stats;
2551 struct flow_miss_op *op;
2552 struct dpif_execute *execute;
2554 ofproto->n_matches++;
2556 if (facet->rule->up.cr.priority == FAIL_OPEN_PRIORITY) {
2558 * Extra-special case for fail-open mode.
2560 * We are in fail-open mode and the packet matched the fail-open
2561 * rule, but we are connected to a controller too. We should send
2562 * the packet up to the controller in the hope that it will try to
2563 * set up a flow and thereby allow us to exit fail-open.
2565 * See the top-level comment in fail-open.c for more information.
2567 send_packet_in_miss(ofproto, packet, flow);
2570 if (!facet->may_install || !subfacet->actions) {
2571 subfacet_make_actions(subfacet, packet);
2574 dpif_flow_stats_extract(&facet->flow, packet, &stats);
2575 subfacet_update_stats(subfacet, &stats);
2577 if (!subfacet->actions_len) {
2578 /* No actions to execute, so skip talking to the dpif. */
2582 if (flow->vlan_tci != subfacet->initial_tci) {
2583 /* This packet was received on a VLAN splinter port. We added
2584 * a VLAN to the packet to make the packet resemble the flow,
2585 * but the actions were composed assuming that the packet
2586 * contained no VLAN. So, we must remove the VLAN header from
2587 * the packet before trying to execute the actions. */
2588 eth_pop_vlan(packet);
2591 op = &ops[(*n_ops)++];
2592 execute = &op->dpif_op.u.execute;
2593 op->subfacet = subfacet;
2594 op->dpif_op.type = DPIF_OP_EXECUTE;
2595 execute->key = miss->key;
2596 execute->key_len = miss->key_len;
2597 execute->actions = (facet->may_install
2599 : xmemdup(subfacet->actions,
2600 subfacet->actions_len));
2601 execute->actions_len = subfacet->actions_len;
2602 execute->packet = packet;
2605 if (facet->may_install && subfacet->key_fitness != ODP_FIT_TOO_LITTLE) {
2606 struct flow_miss_op *op = &ops[(*n_ops)++];
2607 struct dpif_flow_put *put = &op->dpif_op.u.flow_put;
2609 op->subfacet = subfacet;
2610 op->dpif_op.type = DPIF_OP_FLOW_PUT;
2611 put->flags = DPIF_FP_CREATE | DPIF_FP_MODIFY;
2612 put->key = miss->key;
2613 put->key_len = miss->key_len;
2614 put->actions = subfacet->actions;
2615 put->actions_len = subfacet->actions_len;
2620 /* Like odp_flow_key_to_flow(), this function converts the 'key_len' bytes of
2621 * OVS_KEY_ATTR_* attributes in 'key' to a flow structure in 'flow' and returns
2622 * an ODP_FIT_* value that indicates how well 'key' fits our expectations for
2623 * what a flow key should contain.
2625 * This function also includes some logic to help make VLAN splinters
2626 * transparent to the rest of the upcall processing logic. In particular, if
2627 * the extracted in_port is a VLAN splinter port, it replaces flow->in_port by
2628 * the "real" port, sets flow->vlan_tci correctly for the VLAN of the VLAN
2629 * splinter port, and pushes a VLAN header onto 'packet' (if it is nonnull).
2631 * Sets '*initial_tci' to the VLAN TCI with which the packet was really
2632 * received, that is, the actual VLAN TCI extracted by odp_flow_key_to_flow().
2633 * (This differs from the value returned in flow->vlan_tci only for packets
2634 * received on VLAN splinters.)
2636 static enum odp_key_fitness
2637 ofproto_dpif_extract_flow_key(const struct ofproto_dpif *ofproto,
2638 const struct nlattr *key, size_t key_len,
2639 struct flow *flow, ovs_be16 *initial_tci,
2640 struct ofpbuf *packet)
2642 enum odp_key_fitness fitness;
2646 fitness = odp_flow_key_to_flow(key, key_len, flow);
2647 if (fitness == ODP_FIT_ERROR) {
2650 *initial_tci = flow->vlan_tci;
2652 realdev = vsp_vlandev_to_realdev(ofproto, flow->in_port, &vid);
2654 /* Cause the flow to be processed as if it came in on the real device
2655 * with the VLAN device's VLAN ID. */
2656 flow->in_port = realdev;
2657 flow->vlan_tci = htons((vid & VLAN_VID_MASK) | VLAN_CFI);
2659 /* Make the packet resemble the flow, so that it gets sent to an
2660 * OpenFlow controller properly, so that it looks correct for
2661 * sFlow, and so that flow_extract() will get the correct vlan_tci
2662 * if it is called on 'packet'.
2664 * The allocated space inside 'packet' probably also contains
2665 * 'key', that is, both 'packet' and 'key' are probably part of a
2666 * struct dpif_upcall (see the large comment on that structure
2667 * definition), so pushing data on 'packet' is in general not a
2668 * good idea since it could overwrite 'key' or free it as a side
2669 * effect. However, it's OK in this special case because we know
2670 * that 'packet' is inside a Netlink attribute: pushing 4 bytes
2671 * will just overwrite the 4-byte "struct nlattr", which is fine
2672 * since we don't need that header anymore. */
2673 eth_push_vlan(packet, flow->vlan_tci);
2676 /* Let the caller know that we can't reproduce 'key' from 'flow'. */
2677 if (fitness == ODP_FIT_PERFECT) {
2678 fitness = ODP_FIT_TOO_MUCH;
2686 handle_miss_upcalls(struct ofproto_dpif *ofproto, struct dpif_upcall *upcalls,
2689 struct dpif_upcall *upcall;
2690 struct flow_miss *miss, *next_miss;
2691 struct flow_miss_op flow_miss_ops[FLOW_MISS_MAX_BATCH * 2];
2692 struct dpif_op *dpif_ops[FLOW_MISS_MAX_BATCH * 2];
2701 /* Construct the to-do list.
2703 * This just amounts to extracting the flow from each packet and sticking
2704 * the packets that have the same flow in the same "flow_miss" structure so
2705 * that we can process them together. */
2707 for (upcall = upcalls; upcall < &upcalls[n_upcalls]; upcall++) {
2708 enum odp_key_fitness fitness;
2709 struct flow_miss *miss;
2710 ovs_be16 initial_tci;
2713 /* Obtain metadata and check userspace/kernel agreement on flow match,
2714 * then set 'flow''s header pointers. */
2715 fitness = ofproto_dpif_extract_flow_key(ofproto,
2716 upcall->key, upcall->key_len,
2717 &flow, &initial_tci,
2719 if (fitness == ODP_FIT_ERROR) {
2720 ofpbuf_delete(upcall->packet);
2723 flow_extract(upcall->packet, flow.skb_priority, flow.tun_id,
2724 flow.in_port, &flow);
2726 /* Handle 802.1ag, LACP, and STP specially. */
2727 if (process_special(ofproto, &flow, upcall->packet)) {
2728 ofproto_update_local_port_stats(&ofproto->up,
2729 0, upcall->packet->size);
2730 ofpbuf_delete(upcall->packet);
2731 ofproto->n_matches++;
2735 /* Add other packets to a to-do list. */
2736 miss = flow_miss_create(&todo, &flow, fitness,
2737 upcall->key, upcall->key_len, initial_tci);
2738 list_push_back(&miss->packets, &upcall->packet->list_node);
2741 /* Process each element in the to-do list, constructing the set of
2742 * operations to batch. */
2744 HMAP_FOR_EACH (miss, hmap_node, &todo) {
2745 handle_flow_miss(ofproto, miss, flow_miss_ops, &n_ops);
2747 assert(n_ops <= ARRAY_SIZE(flow_miss_ops));
2749 /* Execute batch. */
2750 for (i = 0; i < n_ops; i++) {
2751 dpif_ops[i] = &flow_miss_ops[i].dpif_op;
2753 dpif_operate(ofproto->dpif, dpif_ops, n_ops);
2755 /* Free memory and update facets. */
2756 for (i = 0; i < n_ops; i++) {
2757 struct flow_miss_op *op = &flow_miss_ops[i];
2758 struct dpif_execute *execute;
2760 switch (op->dpif_op.type) {
2761 case DPIF_OP_EXECUTE:
2762 execute = &op->dpif_op.u.execute;
2763 if (op->subfacet->actions != execute->actions) {
2764 free((struct nlattr *) execute->actions);
2768 case DPIF_OP_FLOW_PUT:
2769 if (!op->dpif_op.error) {
2770 op->subfacet->installed = true;
2775 HMAP_FOR_EACH_SAFE (miss, next_miss, hmap_node, &todo) {
2776 ofpbuf_list_delete(&miss->packets);
2777 hmap_remove(&todo, &miss->hmap_node);
2780 hmap_destroy(&todo);
2784 handle_userspace_upcall(struct ofproto_dpif *ofproto,
2785 struct dpif_upcall *upcall)
2787 struct user_action_cookie cookie;
2788 enum odp_key_fitness fitness;
2789 ovs_be16 initial_tci;
2792 memcpy(&cookie, &upcall->userdata, sizeof(cookie));
2794 fitness = ofproto_dpif_extract_flow_key(ofproto, upcall->key,
2795 upcall->key_len, &flow,
2796 &initial_tci, upcall->packet);
2797 if (fitness == ODP_FIT_ERROR) {
2798 ofpbuf_delete(upcall->packet);
2802 if (cookie.type == USER_ACTION_COOKIE_SFLOW) {
2803 if (ofproto->sflow) {
2804 dpif_sflow_received(ofproto->sflow, upcall->packet, &flow,
2808 VLOG_WARN_RL(&rl, "invalid user cookie : 0x%"PRIx64, upcall->userdata);
2810 ofpbuf_delete(upcall->packet);
2814 handle_upcalls(struct ofproto_dpif *ofproto, unsigned int max_batch)
2816 struct dpif_upcall misses[FLOW_MISS_MAX_BATCH];
2820 assert (max_batch <= FLOW_MISS_MAX_BATCH);
2823 for (i = 0; i < max_batch; i++) {
2824 struct dpif_upcall *upcall = &misses[n_misses];
2827 error = dpif_recv(ofproto->dpif, upcall);
2832 switch (upcall->type) {
2833 case DPIF_UC_ACTION:
2834 handle_userspace_upcall(ofproto, upcall);
2838 /* Handle it later. */
2842 case DPIF_N_UC_TYPES:
2844 VLOG_WARN_RL(&rl, "upcall has unexpected type %"PRIu32,
2850 handle_miss_upcalls(ofproto, misses, n_misses);
2855 /* Flow expiration. */
2857 static int subfacet_max_idle(const struct ofproto_dpif *);
2858 static void update_stats(struct ofproto_dpif *);
2859 static void rule_expire(struct rule_dpif *);
2860 static void expire_subfacets(struct ofproto_dpif *, int dp_max_idle);
2862 /* This function is called periodically by run(). Its job is to collect
2863 * updates for the flows that have been installed into the datapath, most
2864 * importantly when they last were used, and then use that information to
2865 * expire flows that have not been used recently.
2867 * Returns the number of milliseconds after which it should be called again. */
2869 expire(struct ofproto_dpif *ofproto)
2871 struct rule_dpif *rule, *next_rule;
2872 struct oftable *table;
2875 /* Update stats for each flow in the datapath. */
2876 update_stats(ofproto);
2878 /* Expire subfacets that have been idle too long. */
2879 dp_max_idle = subfacet_max_idle(ofproto);
2880 expire_subfacets(ofproto, dp_max_idle);
2882 /* Expire OpenFlow flows whose idle_timeout or hard_timeout has passed. */
2883 OFPROTO_FOR_EACH_TABLE (table, &ofproto->up) {
2884 struct cls_cursor cursor;
2886 cls_cursor_init(&cursor, &table->cls, NULL);
2887 CLS_CURSOR_FOR_EACH_SAFE (rule, next_rule, up.cr, &cursor) {
2892 /* All outstanding data in existing flows has been accounted, so it's a
2893 * good time to do bond rebalancing. */
2894 if (ofproto->has_bonded_bundles) {
2895 struct ofbundle *bundle;
2897 HMAP_FOR_EACH (bundle, hmap_node, &ofproto->bundles) {
2899 bond_rebalance(bundle->bond, &ofproto->revalidate_set);
2904 return MIN(dp_max_idle, 1000);
2907 /* Update 'packet_count', 'byte_count', and 'used' members of installed facets.
2909 * This function also pushes statistics updates to rules which each facet
2910 * resubmits into. Generally these statistics will be accurate. However, if a
2911 * facet changes the rule it resubmits into at some time in between
2912 * update_stats() runs, it is possible that statistics accrued to the
2913 * old rule will be incorrectly attributed to the new rule. This could be
2914 * avoided by calling update_stats() whenever rules are created or
2915 * deleted. However, the performance impact of making so many calls to the
2916 * datapath do not justify the benefit of having perfectly accurate statistics.
2919 update_stats(struct ofproto_dpif *p)
2921 const struct dpif_flow_stats *stats;
2922 struct dpif_flow_dump dump;
2923 const struct nlattr *key;
2926 dpif_flow_dump_start(&dump, p->dpif);
2927 while (dpif_flow_dump_next(&dump, &key, &key_len, NULL, NULL, &stats)) {
2928 struct subfacet *subfacet;
2930 subfacet = subfacet_find(p, key, key_len);
2931 if (subfacet && subfacet->installed) {
2932 struct facet *facet = subfacet->facet;
2934 if (stats->n_packets >= subfacet->dp_packet_count) {
2935 uint64_t extra = stats->n_packets - subfacet->dp_packet_count;
2936 facet->packet_count += extra;
2938 VLOG_WARN_RL(&rl, "unexpected packet count from the datapath");
2941 if (stats->n_bytes >= subfacet->dp_byte_count) {
2942 facet->byte_count += stats->n_bytes - subfacet->dp_byte_count;
2944 VLOG_WARN_RL(&rl, "unexpected byte count from datapath");
2947 subfacet->dp_packet_count = stats->n_packets;
2948 subfacet->dp_byte_count = stats->n_bytes;
2950 subfacet_update_time(subfacet, stats->used);
2951 facet_account(facet);
2952 facet_push_stats(facet);
2954 if (!VLOG_DROP_WARN(&rl)) {
2958 odp_flow_key_format(key, key_len, &s);
2959 VLOG_WARN("unexpected flow from datapath %s", ds_cstr(&s));
2963 COVERAGE_INC(facet_unexpected);
2964 /* There's a flow in the datapath that we know nothing about, or a
2965 * flow that shouldn't be installed but was anyway. Delete it. */
2966 dpif_flow_del(p->dpif, key, key_len, NULL);
2969 dpif_flow_dump_done(&dump);
2972 /* Calculates and returns the number of milliseconds of idle time after which
2973 * subfacets should expire from the datapath. When a subfacet expires, we fold
2974 * its statistics into its facet, and when a facet's last subfacet expires, we
2975 * fold its statistic into its rule. */
2977 subfacet_max_idle(const struct ofproto_dpif *ofproto)
2980 * Idle time histogram.
2982 * Most of the time a switch has a relatively small number of subfacets.
2983 * When this is the case we might as well keep statistics for all of them
2984 * in userspace and to cache them in the kernel datapath for performance as
2987 * As the number of subfacets increases, the memory required to maintain
2988 * statistics about them in userspace and in the kernel becomes
2989 * significant. However, with a large number of subfacets it is likely
2990 * that only a few of them are "heavy hitters" that consume a large amount
2991 * of bandwidth. At this point, only heavy hitters are worth caching in
2992 * the kernel and maintaining in userspaces; other subfacets we can
2995 * The technique used to compute the idle time is to build a histogram with
2996 * N_BUCKETS buckets whose width is BUCKET_WIDTH msecs each. Each subfacet
2997 * that is installed in the kernel gets dropped in the appropriate bucket.
2998 * After the histogram has been built, we compute the cutoff so that only
2999 * the most-recently-used 1% of subfacets (but at least
3000 * ofproto->up.flow_eviction_threshold flows) are kept cached. At least
3001 * the most-recently-used bucket of subfacets is kept, so actually an
3002 * arbitrary number of subfacets can be kept in any given expiration run
3003 * (though the next run will delete most of those unless they receive
3006 * This requires a second pass through the subfacets, in addition to the
3007 * pass made by update_stats(), because the former function never looks at
3008 * uninstallable subfacets.
3010 enum { BUCKET_WIDTH = ROUND_UP(100, TIME_UPDATE_INTERVAL) };
3011 enum { N_BUCKETS = 5000 / BUCKET_WIDTH };
3012 int buckets[N_BUCKETS] = { 0 };
3013 int total, subtotal, bucket;
3014 struct subfacet *subfacet;
3018 total = hmap_count(&ofproto->subfacets);
3019 if (total <= ofproto->up.flow_eviction_threshold) {
3020 return N_BUCKETS * BUCKET_WIDTH;
3023 /* Build histogram. */
3025 HMAP_FOR_EACH (subfacet, hmap_node, &ofproto->subfacets) {
3026 long long int idle = now - subfacet->used;
3027 int bucket = (idle <= 0 ? 0
3028 : idle >= BUCKET_WIDTH * N_BUCKETS ? N_BUCKETS - 1
3029 : (unsigned int) idle / BUCKET_WIDTH);
3033 /* Find the first bucket whose flows should be expired. */
3034 subtotal = bucket = 0;
3036 subtotal += buckets[bucket++];
3037 } while (bucket < N_BUCKETS &&
3038 subtotal < MAX(ofproto->up.flow_eviction_threshold, total / 100));
3040 if (VLOG_IS_DBG_ENABLED()) {
3044 ds_put_cstr(&s, "keep");
3045 for (i = 0; i < N_BUCKETS; i++) {
3047 ds_put_cstr(&s, ", drop");
3050 ds_put_format(&s, " %d:%d", i * BUCKET_WIDTH, buckets[i]);
3053 VLOG_INFO("%s: %s (msec:count)", ofproto->up.name, ds_cstr(&s));
3057 return bucket * BUCKET_WIDTH;
3061 expire_subfacets(struct ofproto_dpif *ofproto, int dp_max_idle)
3063 long long int cutoff = time_msec() - dp_max_idle;
3064 struct subfacet *subfacet, *next_subfacet;
3066 HMAP_FOR_EACH_SAFE (subfacet, next_subfacet, hmap_node,
3067 &ofproto->subfacets) {
3068 if (subfacet->used < cutoff) {
3069 subfacet_destroy(subfacet);
3074 /* If 'rule' is an OpenFlow rule, that has expired according to OpenFlow rules,
3075 * then delete it entirely. */
3077 rule_expire(struct rule_dpif *rule)
3079 struct facet *facet, *next_facet;
3083 /* Has 'rule' expired? */
3085 if (rule->up.hard_timeout
3086 && now > rule->up.modified + rule->up.hard_timeout * 1000) {
3087 reason = OFPRR_HARD_TIMEOUT;
3088 } else if (rule->up.idle_timeout && list_is_empty(&rule->facets)
3089 && now > rule->up.used + rule->up.idle_timeout * 1000) {
3090 reason = OFPRR_IDLE_TIMEOUT;
3095 COVERAGE_INC(ofproto_dpif_expired);
3097 /* Update stats. (This is a no-op if the rule expired due to an idle
3098 * timeout, because that only happens when the rule has no facets left.) */
3099 LIST_FOR_EACH_SAFE (facet, next_facet, list_node, &rule->facets) {
3100 facet_remove(facet);
3103 /* Get rid of the rule. */
3104 ofproto_rule_expire(&rule->up, reason);
3109 /* Creates and returns a new facet owned by 'rule', given a 'flow'.
3111 * The caller must already have determined that no facet with an identical
3112 * 'flow' exists in 'ofproto' and that 'flow' is the best match for 'rule' in
3113 * the ofproto's classifier table.
3115 * The facet will initially have no subfacets. The caller should create (at
3116 * least) one subfacet with subfacet_create(). */
3117 static struct facet *
3118 facet_create(struct rule_dpif *rule, const struct flow *flow)
3120 struct ofproto_dpif *ofproto = ofproto_dpif_cast(rule->up.ofproto);
3121 struct facet *facet;
3123 facet = xzalloc(sizeof *facet);
3124 facet->used = time_msec();
3125 hmap_insert(&ofproto->facets, &facet->hmap_node, flow_hash(flow, 0));
3126 list_push_back(&rule->facets, &facet->list_node);
3128 facet->flow = *flow;
3129 list_init(&facet->subfacets);
3130 netflow_flow_init(&facet->nf_flow);
3131 netflow_flow_update_time(ofproto->netflow, &facet->nf_flow, facet->used);
3137 facet_free(struct facet *facet)
3142 /* Executes, within 'ofproto', the 'n_actions' actions in 'actions' on
3143 * 'packet', which arrived on 'in_port'.
3145 * Takes ownership of 'packet'. */
3147 execute_odp_actions(struct ofproto_dpif *ofproto, const struct flow *flow,
3148 const struct nlattr *odp_actions, size_t actions_len,
3149 struct ofpbuf *packet)
3151 struct odputil_keybuf keybuf;
3155 ofpbuf_use_stack(&key, &keybuf, sizeof keybuf);
3156 odp_flow_key_from_flow(&key, flow);
3158 error = dpif_execute(ofproto->dpif, key.data, key.size,
3159 odp_actions, actions_len, packet);
3161 ofpbuf_delete(packet);
3165 /* Remove 'facet' from 'ofproto' and free up the associated memory:
3167 * - If 'facet' was installed in the datapath, uninstalls it and updates its
3168 * rule's statistics, via subfacet_uninstall().
3170 * - Removes 'facet' from its rule and from ofproto->facets.
3173 facet_remove(struct facet *facet)
3175 struct ofproto_dpif *ofproto = ofproto_dpif_cast(facet->rule->up.ofproto);
3176 struct subfacet *subfacet, *next_subfacet;
3178 assert(!list_is_empty(&facet->subfacets));
3180 /* First uninstall all of the subfacets to get final statistics. */
3181 LIST_FOR_EACH (subfacet, list_node, &facet->subfacets) {
3182 subfacet_uninstall(subfacet);
3185 /* Flush the final stats to the rule.
3187 * This might require us to have at least one subfacet around so that we
3188 * can use its actions for accounting in facet_account(), which is why we
3189 * have uninstalled but not yet destroyed the subfacets. */
3190 facet_flush_stats(facet);
3192 /* Now we're really all done so destroy everything. */
3193 LIST_FOR_EACH_SAFE (subfacet, next_subfacet, list_node,
3194 &facet->subfacets) {
3195 subfacet_destroy__(subfacet);
3197 hmap_remove(&ofproto->facets, &facet->hmap_node);
3198 list_remove(&facet->list_node);
3203 facet_account(struct facet *facet)
3205 struct ofproto_dpif *ofproto = ofproto_dpif_cast(facet->rule->up.ofproto);
3207 struct subfacet *subfacet;
3208 const struct nlattr *a;
3212 if (facet->byte_count <= facet->accounted_bytes) {
3215 n_bytes = facet->byte_count - facet->accounted_bytes;
3216 facet->accounted_bytes = facet->byte_count;
3218 /* Feed information from the active flows back into the learning table to
3219 * ensure that table is always in sync with what is actually flowing
3220 * through the datapath. */
3221 if (facet->has_learn || facet->has_normal) {
3222 struct action_xlate_ctx ctx;
3224 action_xlate_ctx_init(&ctx, ofproto, &facet->flow,
3225 facet->flow.vlan_tci,
3226 facet->rule->up.flow_cookie, NULL);
3227 ctx.may_learn = true;
3228 ofpbuf_delete(xlate_actions(&ctx, facet->rule->up.actions,
3229 facet->rule->up.n_actions));
3232 if (!facet->has_normal || !ofproto->has_bonded_bundles) {
3236 /* This loop feeds byte counters to bond_account() for rebalancing to use
3237 * as a basis. We also need to track the actual VLAN on which the packet
3238 * is going to be sent to ensure that it matches the one passed to
3239 * bond_choose_output_slave(). (Otherwise, we will account to the wrong
3242 * We use the actions from an arbitrary subfacet because they should all
3243 * be equally valid for our purpose. */
3244 subfacet = CONTAINER_OF(list_front(&facet->subfacets),
3245 struct subfacet, list_node);
3246 vlan_tci = facet->flow.vlan_tci;
3247 NL_ATTR_FOR_EACH_UNSAFE (a, left,
3248 subfacet->actions, subfacet->actions_len) {
3249 const struct ovs_action_push_vlan *vlan;
3250 struct ofport_dpif *port;
3252 switch (nl_attr_type(a)) {
3253 case OVS_ACTION_ATTR_OUTPUT:
3254 port = get_odp_port(ofproto, nl_attr_get_u32(a));
3255 if (port && port->bundle && port->bundle->bond) {
3256 bond_account(port->bundle->bond, &facet->flow,
3257 vlan_tci_to_vid(vlan_tci), n_bytes);
3261 case OVS_ACTION_ATTR_POP_VLAN:
3262 vlan_tci = htons(0);
3265 case OVS_ACTION_ATTR_PUSH_VLAN:
3266 vlan = nl_attr_get(a);
3267 vlan_tci = vlan->vlan_tci;
3273 /* Returns true if the only action for 'facet' is to send to the controller.
3274 * (We don't report NetFlow expiration messages for such facets because they
3275 * are just part of the control logic for the network, not real traffic). */
3277 facet_is_controller_flow(struct facet *facet)
3280 && facet->rule->up.n_actions == 1
3281 && action_outputs_to_port(&facet->rule->up.actions[0],
3282 htons(OFPP_CONTROLLER)));
3285 /* Folds all of 'facet''s statistics into its rule. Also updates the
3286 * accounting ofhook and emits a NetFlow expiration if appropriate. All of
3287 * 'facet''s statistics in the datapath should have been zeroed and folded into
3288 * its packet and byte counts before this function is called. */
3290 facet_flush_stats(struct facet *facet)
3292 struct ofproto_dpif *ofproto = ofproto_dpif_cast(facet->rule->up.ofproto);
3293 struct subfacet *subfacet;
3295 LIST_FOR_EACH (subfacet, list_node, &facet->subfacets) {
3296 assert(!subfacet->dp_byte_count);
3297 assert(!subfacet->dp_packet_count);
3300 facet_push_stats(facet);
3301 facet_account(facet);
3303 if (ofproto->netflow && !facet_is_controller_flow(facet)) {
3304 struct ofexpired expired;
3305 expired.flow = facet->flow;
3306 expired.packet_count = facet->packet_count;
3307 expired.byte_count = facet->byte_count;
3308 expired.used = facet->used;
3309 netflow_expire(ofproto->netflow, &facet->nf_flow, &expired);
3312 facet->rule->packet_count += facet->packet_count;
3313 facet->rule->byte_count += facet->byte_count;
3315 /* Reset counters to prevent double counting if 'facet' ever gets
3317 facet_reset_counters(facet);
3319 netflow_flow_clear(&facet->nf_flow);
3322 /* Searches 'ofproto''s table of facets for one exactly equal to 'flow'.
3323 * Returns it if found, otherwise a null pointer.
3325 * The returned facet might need revalidation; use facet_lookup_valid()
3326 * instead if that is important. */
3327 static struct facet *
3328 facet_find(struct ofproto_dpif *ofproto, const struct flow *flow)
3330 struct facet *facet;
3332 HMAP_FOR_EACH_WITH_HASH (facet, hmap_node, flow_hash(flow, 0),
3334 if (flow_equal(flow, &facet->flow)) {
3342 /* Searches 'ofproto''s table of facets for one exactly equal to 'flow'.
3343 * Returns it if found, otherwise a null pointer.
3345 * The returned facet is guaranteed to be valid. */
3346 static struct facet *
3347 facet_lookup_valid(struct ofproto_dpif *ofproto, const struct flow *flow)
3349 struct facet *facet = facet_find(ofproto, flow);
3351 /* The facet we found might not be valid, since we could be in need of
3352 * revalidation. If it is not valid, don't return it. */
3354 && (ofproto->need_revalidate
3355 || tag_set_intersects(&ofproto->revalidate_set, facet->tags))
3356 && !facet_revalidate(facet)) {
3357 COVERAGE_INC(facet_invalidated);
3365 facet_check_consistency(struct facet *facet)
3367 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 15);
3369 struct ofproto_dpif *ofproto = ofproto_dpif_cast(facet->rule->up.ofproto);
3371 struct rule_dpif *rule;
3372 struct subfacet *subfacet;
3373 bool may_log = false;
3376 /* Check the rule for consistency. */
3377 rule = rule_dpif_lookup(ofproto, &facet->flow, 0);
3379 if (!VLOG_DROP_WARN(&rl)) {
3380 char *s = flow_to_string(&facet->flow);
3381 VLOG_WARN("%s: facet should not exist", s);
3385 } else if (rule != facet->rule) {
3386 may_log = !VLOG_DROP_WARN(&rl);
3392 flow_format(&s, &facet->flow);
3393 ds_put_format(&s, ": facet associated with wrong rule (was "
3394 "table=%"PRIu8",", facet->rule->up.table_id);
3395 cls_rule_format(&facet->rule->up.cr, &s);
3396 ds_put_format(&s, ") (should have been table=%"PRIu8",",
3398 cls_rule_format(&rule->up.cr, &s);
3399 ds_put_char(&s, ')');
3401 VLOG_WARN("%s", ds_cstr(&s));
3408 /* Check the datapath actions for consistency. */
3409 LIST_FOR_EACH (subfacet, list_node, &facet->subfacets) {
3410 struct action_xlate_ctx ctx;
3411 struct ofpbuf *odp_actions;
3412 bool actions_changed;
3413 bool should_install;
3415 action_xlate_ctx_init(&ctx, ofproto, &facet->flow,
3416 subfacet->initial_tci, rule->up.flow_cookie,
3418 odp_actions = xlate_actions(&ctx, rule->up.actions,
3419 rule->up.n_actions);
3421 should_install = (ctx.may_set_up_flow
3422 && subfacet->key_fitness != ODP_FIT_TOO_LITTLE);
3423 if (!should_install && !subfacet->installed) {
3424 /* The actions for uninstallable flows may vary from one packet to
3425 * the next, so don't compare the actions. */
3429 actions_changed = (subfacet->actions_len != odp_actions->size
3430 || memcmp(subfacet->actions, odp_actions->data,
3431 subfacet->actions_len));
3432 if (should_install != subfacet->installed || actions_changed) {
3434 may_log = !VLOG_DROP_WARN(&rl);
3439 struct odputil_keybuf keybuf;
3444 subfacet_get_key(subfacet, &keybuf, &key);
3445 odp_flow_key_format(key.data, key.size, &s);
3447 ds_put_cstr(&s, ": inconsistency in subfacet");
3448 if (should_install != subfacet->installed) {
3449 enum odp_key_fitness fitness = subfacet->key_fitness;
3451 ds_put_format(&s, " (should%s have been installed)",
3452 should_install ? "" : " not");
3453 ds_put_format(&s, " (may_set_up_flow=%s, fitness=%s)",
3454 ctx.may_set_up_flow ? "true" : "false",
3455 odp_key_fitness_to_string(fitness));
3457 if (actions_changed) {
3458 ds_put_cstr(&s, " (actions were: ");
3459 format_odp_actions(&s, subfacet->actions,
3460 subfacet->actions_len);
3461 ds_put_cstr(&s, ") (correct actions: ");
3462 format_odp_actions(&s, odp_actions->data,
3464 ds_put_char(&s, ')');
3466 ds_put_cstr(&s, " (actions: ");
3467 format_odp_actions(&s, subfacet->actions,
3468 subfacet->actions_len);
3469 ds_put_char(&s, ')');
3471 VLOG_WARN("%s", ds_cstr(&s));
3477 ofpbuf_delete(odp_actions);
3483 /* Re-searches the classifier for 'facet':
3485 * - If the rule found is different from 'facet''s current rule, moves
3486 * 'facet' to the new rule and recompiles its actions.
3488 * - If the rule found is the same as 'facet''s current rule, leaves 'facet'
3489 * where it is and recompiles its actions anyway.
3491 * - If there is none, destroys 'facet'.
3493 * Returns true if 'facet' still exists, false if it has been destroyed. */
3495 facet_revalidate(struct facet *facet)
3497 struct ofproto_dpif *ofproto = ofproto_dpif_cast(facet->rule->up.ofproto);
3499 struct nlattr *odp_actions;
3502 struct actions *new_actions;
3504 struct action_xlate_ctx ctx;
3505 struct rule_dpif *new_rule;
3506 struct subfacet *subfacet;
3507 bool actions_changed;
3510 COVERAGE_INC(facet_revalidate);
3512 /* Determine the new rule. */
3513 new_rule = rule_dpif_lookup(ofproto, &facet->flow, 0);
3515 /* No new rule, so delete the facet. */
3516 facet_remove(facet);
3520 /* Calculate new datapath actions.
3522 * We do not modify any 'facet' state yet, because we might need to, e.g.,
3523 * emit a NetFlow expiration and, if so, we need to have the old state
3524 * around to properly compose it. */
3526 /* If the datapath actions changed or the installability changed,
3527 * then we need to talk to the datapath. */
3530 memset(&ctx, 0, sizeof ctx);
3531 LIST_FOR_EACH (subfacet, list_node, &facet->subfacets) {
3532 struct ofpbuf *odp_actions;
3533 bool should_install;
3535 action_xlate_ctx_init(&ctx, ofproto, &facet->flow,
3536 subfacet->initial_tci, new_rule->up.flow_cookie,
3538 odp_actions = xlate_actions(&ctx, new_rule->up.actions,
3539 new_rule->up.n_actions);
3540 actions_changed = (subfacet->actions_len != odp_actions->size
3541 || memcmp(subfacet->actions, odp_actions->data,
3542 subfacet->actions_len));
3544 should_install = (ctx.may_set_up_flow
3545 && subfacet->key_fitness != ODP_FIT_TOO_LITTLE);
3546 if (actions_changed || should_install != subfacet->installed) {
3547 if (should_install) {
3548 struct dpif_flow_stats stats;
3550 subfacet_install(subfacet,
3551 odp_actions->data, odp_actions->size, &stats);
3552 subfacet_update_stats(subfacet, &stats);
3554 subfacet_uninstall(subfacet);
3558 new_actions = xcalloc(list_size(&facet->subfacets),
3559 sizeof *new_actions);
3561 new_actions[i].odp_actions = xmemdup(odp_actions->data,
3563 new_actions[i].actions_len = odp_actions->size;
3566 ofpbuf_delete(odp_actions);
3570 facet_flush_stats(facet);
3573 /* Update 'facet' now that we've taken care of all the old state. */
3574 facet->tags = ctx.tags;
3575 facet->nf_flow.output_iface = ctx.nf_output_iface;
3576 facet->may_install = ctx.may_set_up_flow;
3577 facet->has_learn = ctx.has_learn;
3578 facet->has_normal = ctx.has_normal;
3579 facet->mirrors = ctx.mirrors;
3582 LIST_FOR_EACH (subfacet, list_node, &facet->subfacets) {
3583 if (new_actions[i].odp_actions) {
3584 free(subfacet->actions);
3585 subfacet->actions = new_actions[i].odp_actions;
3586 subfacet->actions_len = new_actions[i].actions_len;
3592 if (facet->rule != new_rule) {
3593 COVERAGE_INC(facet_changed_rule);
3594 list_remove(&facet->list_node);
3595 list_push_back(&new_rule->facets, &facet->list_node);
3596 facet->rule = new_rule;
3597 facet->used = new_rule->up.created;
3598 facet->prev_used = facet->used;
3604 /* Updates 'facet''s used time. Caller is responsible for calling
3605 * facet_push_stats() to update the flows which 'facet' resubmits into. */
3607 facet_update_time(struct facet *facet, long long int used)
3609 struct ofproto_dpif *ofproto = ofproto_dpif_cast(facet->rule->up.ofproto);
3610 if (used > facet->used) {
3612 ofproto_rule_update_used(&facet->rule->up, used);
3613 netflow_flow_update_time(ofproto->netflow, &facet->nf_flow, used);
3618 facet_reset_counters(struct facet *facet)
3620 facet->packet_count = 0;
3621 facet->byte_count = 0;
3622 facet->prev_packet_count = 0;
3623 facet->prev_byte_count = 0;
3624 facet->accounted_bytes = 0;
3628 facet_push_stats(struct facet *facet)
3630 uint64_t new_packets, new_bytes;
3632 assert(facet->packet_count >= facet->prev_packet_count);
3633 assert(facet->byte_count >= facet->prev_byte_count);
3634 assert(facet->used >= facet->prev_used);
3636 new_packets = facet->packet_count - facet->prev_packet_count;
3637 new_bytes = facet->byte_count - facet->prev_byte_count;
3639 if (new_packets || new_bytes || facet->used > facet->prev_used) {
3640 facet->prev_packet_count = facet->packet_count;
3641 facet->prev_byte_count = facet->byte_count;
3642 facet->prev_used = facet->used;
3644 flow_push_stats(facet->rule, &facet->flow,
3645 new_packets, new_bytes, facet->used);
3647 update_mirror_stats(ofproto_dpif_cast(facet->rule->up.ofproto),
3648 facet->mirrors, new_packets, new_bytes);
3652 struct ofproto_push {
3653 struct action_xlate_ctx ctx;
3660 push_resubmit(struct action_xlate_ctx *ctx, struct rule_dpif *rule)
3662 struct ofproto_push *push = CONTAINER_OF(ctx, struct ofproto_push, ctx);
3665 rule->packet_count += push->packets;
3666 rule->byte_count += push->bytes;
3667 ofproto_rule_update_used(&rule->up, push->used);
3671 /* Pushes flow statistics to the rules which 'flow' resubmits into given
3672 * 'rule''s actions and mirrors. */
3674 flow_push_stats(const struct rule_dpif *rule,
3675 const struct flow *flow, uint64_t packets, uint64_t bytes,
3678 struct ofproto_dpif *ofproto = ofproto_dpif_cast(rule->up.ofproto);
3679 struct ofproto_push push;
3681 push.packets = packets;
3685 action_xlate_ctx_init(&push.ctx, ofproto, flow, flow->vlan_tci,
3686 rule->up.flow_cookie, NULL);
3687 push.ctx.resubmit_hook = push_resubmit;
3688 ofpbuf_delete(xlate_actions(&push.ctx,
3689 rule->up.actions, rule->up.n_actions));
3694 static struct subfacet *
3695 subfacet_find__(struct ofproto_dpif *ofproto,
3696 const struct nlattr *key, size_t key_len, uint32_t key_hash,
3697 const struct flow *flow)
3699 struct subfacet *subfacet;
3701 HMAP_FOR_EACH_WITH_HASH (subfacet, hmap_node, key_hash,
3702 &ofproto->subfacets) {
3704 ? (subfacet->key_len == key_len
3705 && !memcmp(key, subfacet->key, key_len))
3706 : flow_equal(flow, &subfacet->facet->flow)) {
3714 /* Searches 'facet' (within 'ofproto') for a subfacet with the specified
3715 * 'key_fitness', 'key', and 'key_len'. Returns the existing subfacet if
3716 * there is one, otherwise creates and returns a new subfacet.
3718 * If the returned subfacet is new, then subfacet->actions will be NULL, in
3719 * which case the caller must populate the actions with
3720 * subfacet_make_actions(). */
3721 static struct subfacet *
3722 subfacet_create(struct facet *facet, enum odp_key_fitness key_fitness,
3723 const struct nlattr *key, size_t key_len, ovs_be16 initial_tci)
3725 struct ofproto_dpif *ofproto = ofproto_dpif_cast(facet->rule->up.ofproto);
3726 uint32_t key_hash = odp_flow_key_hash(key, key_len);
3727 struct subfacet *subfacet;
3729 subfacet = subfacet_find__(ofproto, key, key_len, key_hash, &facet->flow);
3731 if (subfacet->facet == facet) {
3735 /* This shouldn't happen. */
3736 VLOG_ERR_RL(&rl, "subfacet with wrong facet");
3737 subfacet_destroy(subfacet);
3740 subfacet = xzalloc(sizeof *subfacet);
3741 hmap_insert(&ofproto->subfacets, &subfacet->hmap_node, key_hash);
3742 list_push_back(&facet->subfacets, &subfacet->list_node);
3743 subfacet->facet = facet;
3744 subfacet->used = time_msec();
3745 subfacet->key_fitness = key_fitness;
3746 if (key_fitness != ODP_FIT_PERFECT) {
3747 subfacet->key = xmemdup(key, key_len);
3748 subfacet->key_len = key_len;
3750 subfacet->installed = false;
3751 subfacet->initial_tci = initial_tci;
3756 /* Searches 'ofproto' for a subfacet with the given 'key', 'key_len', and
3757 * 'flow'. Returns the subfacet if one exists, otherwise NULL. */
3758 static struct subfacet *
3759 subfacet_find(struct ofproto_dpif *ofproto,
3760 const struct nlattr *key, size_t key_len)
3762 uint32_t key_hash = odp_flow_key_hash(key, key_len);
3763 enum odp_key_fitness fitness;
3766 fitness = odp_flow_key_to_flow(key, key_len, &flow);
3767 if (fitness == ODP_FIT_ERROR) {
3771 return subfacet_find__(ofproto, key, key_len, key_hash, &flow);
3774 /* Uninstalls 'subfacet' from the datapath, if it is installed, removes it from
3775 * its facet within 'ofproto', and frees it. */
3777 subfacet_destroy__(struct subfacet *subfacet)
3779 struct facet *facet = subfacet->facet;
3780 struct ofproto_dpif *ofproto = ofproto_dpif_cast(facet->rule->up.ofproto);
3782 subfacet_uninstall(subfacet);
3783 hmap_remove(&ofproto->subfacets, &subfacet->hmap_node);
3784 list_remove(&subfacet->list_node);
3785 free(subfacet->key);
3786 free(subfacet->actions);
3790 /* Destroys 'subfacet', as with subfacet_destroy__(), and then if this was the
3791 * last remaining subfacet in its facet destroys the facet too. */
3793 subfacet_destroy(struct subfacet *subfacet)
3795 struct facet *facet = subfacet->facet;
3797 if (list_is_singleton(&facet->subfacets)) {
3798 /* facet_remove() needs at least one subfacet (it will remove it). */
3799 facet_remove(facet);
3801 subfacet_destroy__(subfacet);
3805 /* Initializes 'key' with the sequence of OVS_KEY_ATTR_* Netlink attributes
3806 * that can be used to refer to 'subfacet'. The caller must provide 'keybuf'
3807 * for use as temporary storage. */
3809 subfacet_get_key(struct subfacet *subfacet, struct odputil_keybuf *keybuf,
3812 if (!subfacet->key) {
3813 ofpbuf_use_stack(key, keybuf, sizeof *keybuf);
3814 odp_flow_key_from_flow(key, &subfacet->facet->flow);
3816 ofpbuf_use_const(key, subfacet->key, subfacet->key_len);
3820 /* Composes the datapath actions for 'subfacet' based on its rule's actions. */
3822 subfacet_make_actions(struct subfacet *subfacet, const struct ofpbuf *packet)
3824 struct facet *facet = subfacet->facet;
3825 const struct rule_dpif *rule = facet->rule;
3826 struct ofproto_dpif *ofproto = ofproto_dpif_cast(rule->up.ofproto);
3827 struct ofpbuf *odp_actions;
3828 struct action_xlate_ctx ctx;
3830 action_xlate_ctx_init(&ctx, ofproto, &facet->flow, subfacet->initial_tci,
3831 rule->up.flow_cookie, packet);
3832 odp_actions = xlate_actions(&ctx, rule->up.actions, rule->up.n_actions);
3833 facet->tags = ctx.tags;
3834 facet->may_install = ctx.may_set_up_flow;
3835 facet->has_learn = ctx.has_learn;
3836 facet->has_normal = ctx.has_normal;
3837 facet->nf_flow.output_iface = ctx.nf_output_iface;
3838 facet->mirrors = ctx.mirrors;
3840 if (subfacet->actions_len != odp_actions->size
3841 || memcmp(subfacet->actions, odp_actions->data, odp_actions->size)) {
3842 free(subfacet->actions);
3843 subfacet->actions_len = odp_actions->size;
3844 subfacet->actions = xmemdup(odp_actions->data, odp_actions->size);
3847 ofpbuf_delete(odp_actions);
3850 /* Updates 'subfacet''s datapath flow, setting its actions to 'actions_len'
3851 * bytes of actions in 'actions'. If 'stats' is non-null, statistics counters
3852 * in the datapath will be zeroed and 'stats' will be updated with traffic new
3853 * since 'subfacet' was last updated.
3855 * Returns 0 if successful, otherwise a positive errno value. */
3857 subfacet_install(struct subfacet *subfacet,
3858 const struct nlattr *actions, size_t actions_len,
3859 struct dpif_flow_stats *stats)
3861 struct facet *facet = subfacet->facet;
3862 struct ofproto_dpif *ofproto = ofproto_dpif_cast(facet->rule->up.ofproto);
3863 struct odputil_keybuf keybuf;
3864 enum dpif_flow_put_flags flags;
3868 flags = DPIF_FP_CREATE | DPIF_FP_MODIFY;
3870 flags |= DPIF_FP_ZERO_STATS;
3873 subfacet_get_key(subfacet, &keybuf, &key);
3874 ret = dpif_flow_put(ofproto->dpif, flags, key.data, key.size,
3875 actions, actions_len, stats);
3878 subfacet_reset_dp_stats(subfacet, stats);
3884 /* If 'subfacet' is installed in the datapath, uninstalls it. */
3886 subfacet_uninstall(struct subfacet *subfacet)
3888 if (subfacet->installed) {
3889 struct rule_dpif *rule = subfacet->facet->rule;
3890 struct ofproto_dpif *ofproto = ofproto_dpif_cast(rule->up.ofproto);
3891 struct odputil_keybuf keybuf;
3892 struct dpif_flow_stats stats;
3896 subfacet_get_key(subfacet, &keybuf, &key);
3897 error = dpif_flow_del(ofproto->dpif, key.data, key.size, &stats);
3898 subfacet_reset_dp_stats(subfacet, &stats);
3900 subfacet_update_stats(subfacet, &stats);
3902 subfacet->installed = false;
3904 assert(subfacet->dp_packet_count == 0);
3905 assert(subfacet->dp_byte_count == 0);
3909 /* Resets 'subfacet''s datapath statistics counters. This should be called
3910 * when 'subfacet''s statistics are cleared in the datapath. If 'stats' is
3911 * non-null, it should contain the statistics returned by dpif when 'subfacet'
3912 * was reset in the datapath. 'stats' will be modified to include only
3913 * statistics new since 'subfacet' was last updated. */
3915 subfacet_reset_dp_stats(struct subfacet *subfacet,
3916 struct dpif_flow_stats *stats)
3919 && subfacet->dp_packet_count <= stats->n_packets
3920 && subfacet->dp_byte_count <= stats->n_bytes) {
3921 stats->n_packets -= subfacet->dp_packet_count;
3922 stats->n_bytes -= subfacet->dp_byte_count;
3925 subfacet->dp_packet_count = 0;
3926 subfacet->dp_byte_count = 0;
3929 /* Updates 'subfacet''s used time. The caller is responsible for calling
3930 * facet_push_stats() to update the flows which 'subfacet' resubmits into. */
3932 subfacet_update_time(struct subfacet *subfacet, long long int used)
3934 if (used > subfacet->used) {
3935 subfacet->used = used;
3936 facet_update_time(subfacet->facet, used);
3940 /* Folds the statistics from 'stats' into the counters in 'subfacet'.
3942 * Because of the meaning of a subfacet's counters, it only makes sense to do
3943 * this if 'stats' are not tracked in the datapath, that is, if 'stats'
3944 * represents a packet that was sent by hand or if it represents statistics
3945 * that have been cleared out of the datapath. */
3947 subfacet_update_stats(struct subfacet *subfacet,
3948 const struct dpif_flow_stats *stats)
3950 if (stats->n_packets || stats->used > subfacet->used) {
3951 struct facet *facet = subfacet->facet;
3953 subfacet_update_time(subfacet, stats->used);
3954 facet->packet_count += stats->n_packets;
3955 facet->byte_count += stats->n_bytes;
3956 facet_push_stats(facet);
3957 netflow_flow_update_flags(&facet->nf_flow, stats->tcp_flags);
3963 static struct rule_dpif *
3964 rule_dpif_lookup(struct ofproto_dpif *ofproto, const struct flow *flow,
3967 struct cls_rule *cls_rule;
3968 struct classifier *cls;
3970 if (table_id >= N_TABLES) {
3974 cls = &ofproto->up.tables[table_id].cls;
3975 if (flow->nw_frag & FLOW_NW_FRAG_ANY
3976 && ofproto->up.frag_handling == OFPC_FRAG_NORMAL) {
3977 /* For OFPC_NORMAL frag_handling, we must pretend that transport ports
3978 * are unavailable. */
3979 struct flow ofpc_normal_flow = *flow;
3980 ofpc_normal_flow.tp_src = htons(0);
3981 ofpc_normal_flow.tp_dst = htons(0);
3982 cls_rule = classifier_lookup(cls, &ofpc_normal_flow);
3984 cls_rule = classifier_lookup(cls, flow);
3986 return rule_dpif_cast(rule_from_cls_rule(cls_rule));
3990 complete_operation(struct rule_dpif *rule)
3992 struct ofproto_dpif *ofproto = ofproto_dpif_cast(rule->up.ofproto);
3994 rule_invalidate(rule);
3996 struct dpif_completion *c = xmalloc(sizeof *c);
3997 c->op = rule->up.pending;
3998 list_push_back(&ofproto->completions, &c->list_node);
4000 ofoperation_complete(rule->up.pending, 0);
4004 static struct rule *
4007 struct rule_dpif *rule = xmalloc(sizeof *rule);
4012 rule_dealloc(struct rule *rule_)
4014 struct rule_dpif *rule = rule_dpif_cast(rule_);
4019 rule_construct(struct rule *rule_)
4021 struct rule_dpif *rule = rule_dpif_cast(rule_);
4022 struct ofproto_dpif *ofproto = ofproto_dpif_cast(rule->up.ofproto);
4023 struct rule_dpif *victim;
4027 error = validate_actions(rule->up.actions, rule->up.n_actions,
4028 &rule->up.cr.flow, ofproto->max_ports);
4033 rule->packet_count = 0;
4034 rule->byte_count = 0;
4036 victim = rule_dpif_cast(ofoperation_get_victim(rule->up.pending));
4037 if (victim && !list_is_empty(&victim->facets)) {
4038 struct facet *facet;
4040 rule->facets = victim->facets;
4041 list_moved(&rule->facets);
4042 LIST_FOR_EACH (facet, list_node, &rule->facets) {
4043 /* XXX: We're only clearing our local counters here. It's possible
4044 * that quite a few packets are unaccounted for in the datapath
4045 * statistics. These will be accounted to the new rule instead of
4046 * cleared as required. This could be fixed by clearing out the
4047 * datapath statistics for this facet, but currently it doesn't
4049 facet_reset_counters(facet);
4053 /* Must avoid list_moved() in this case. */
4054 list_init(&rule->facets);
4057 table_id = rule->up.table_id;
4058 rule->tag = (victim ? victim->tag
4060 : rule_calculate_tag(&rule->up.cr.flow, &rule->up.cr.wc,
4061 ofproto->tables[table_id].basis));
4063 complete_operation(rule);
4068 rule_destruct(struct rule *rule_)
4070 struct rule_dpif *rule = rule_dpif_cast(rule_);
4071 struct facet *facet, *next_facet;
4073 LIST_FOR_EACH_SAFE (facet, next_facet, list_node, &rule->facets) {
4074 facet_revalidate(facet);
4077 complete_operation(rule);
4081 rule_get_stats(struct rule *rule_, uint64_t *packets, uint64_t *bytes)
4083 struct rule_dpif *rule = rule_dpif_cast(rule_);
4084 struct facet *facet;
4086 /* Start from historical data for 'rule' itself that are no longer tracked
4087 * in facets. This counts, for example, facets that have expired. */
4088 *packets = rule->packet_count;
4089 *bytes = rule->byte_count;
4091 /* Add any statistics that are tracked by facets. This includes
4092 * statistical data recently updated by ofproto_update_stats() as well as
4093 * stats for packets that were executed "by hand" via dpif_execute(). */
4094 LIST_FOR_EACH (facet, list_node, &rule->facets) {
4095 *packets += facet->packet_count;
4096 *bytes += facet->byte_count;
4101 rule_execute(struct rule *rule_, const struct flow *flow,
4102 struct ofpbuf *packet)
4104 struct rule_dpif *rule = rule_dpif_cast(rule_);
4105 struct ofproto_dpif *ofproto = ofproto_dpif_cast(rule->up.ofproto);
4106 struct action_xlate_ctx ctx;
4107 struct ofpbuf *odp_actions;
4110 action_xlate_ctx_init(&ctx, ofproto, flow, flow->vlan_tci,
4111 rule->up.flow_cookie, packet);
4112 odp_actions = xlate_actions(&ctx, rule->up.actions, rule->up.n_actions);
4113 size = packet->size;
4114 if (execute_odp_actions(ofproto, flow, odp_actions->data,
4115 odp_actions->size, packet)) {
4116 ofproto_rule_update_used(&rule->up, time_msec());
4117 rule->packet_count++;
4118 rule->byte_count += size;
4119 flow_push_stats(rule, flow, 1, size, rule->up.used);
4121 ofpbuf_delete(odp_actions);
4127 rule_modify_actions(struct rule *rule_)
4129 struct rule_dpif *rule = rule_dpif_cast(rule_);
4130 struct ofproto_dpif *ofproto = ofproto_dpif_cast(rule->up.ofproto);
4133 error = validate_actions(rule->up.actions, rule->up.n_actions,
4134 &rule->up.cr.flow, ofproto->max_ports);
4136 ofoperation_complete(rule->up.pending, error);
4140 complete_operation(rule);
4143 /* Sends 'packet' out 'ofport'.
4144 * May modify 'packet'.
4145 * Returns 0 if successful, otherwise a positive errno value. */
4147 send_packet(const struct ofport_dpif *ofport, struct ofpbuf *packet)
4149 const struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport->up.ofproto);
4150 struct ofpbuf key, odp_actions;
4151 struct odputil_keybuf keybuf;
4156 flow_extract((struct ofpbuf *) packet, 0, 0, 0, &flow);
4157 odp_port = vsp_realdev_to_vlandev(ofproto, ofport->odp_port,
4159 if (odp_port != ofport->odp_port) {
4160 eth_pop_vlan(packet);
4161 flow.vlan_tci = htons(0);
4164 ofpbuf_use_stack(&key, &keybuf, sizeof keybuf);
4165 odp_flow_key_from_flow(&key, &flow);
4167 ofpbuf_init(&odp_actions, 32);
4168 compose_sflow_action(ofproto, &odp_actions, &flow, odp_port);
4170 nl_msg_put_u32(&odp_actions, OVS_ACTION_ATTR_OUTPUT, odp_port);
4171 error = dpif_execute(ofproto->dpif,
4173 odp_actions.data, odp_actions.size,
4175 ofpbuf_uninit(&odp_actions);
4178 VLOG_WARN_RL(&rl, "%s: failed to send packet on port %"PRIu32" (%s)",
4179 ofproto->up.name, odp_port, strerror(error));
4181 ofproto_update_local_port_stats(ofport->up.ofproto, packet->size, 0);
4185 /* OpenFlow to datapath action translation. */
4187 static void do_xlate_actions(const union ofp_action *in, size_t n_in,
4188 struct action_xlate_ctx *ctx);
4189 static void xlate_normal(struct action_xlate_ctx *);
4192 put_userspace_action(const struct ofproto_dpif *ofproto,
4193 struct ofpbuf *odp_actions,
4194 const struct flow *flow,
4195 const struct user_action_cookie *cookie)
4199 pid = dpif_port_get_pid(ofproto->dpif,
4200 ofp_port_to_odp_port(flow->in_port));
4202 return odp_put_userspace_action(pid, cookie, odp_actions);
4205 /* Compose SAMPLE action for sFlow. */
4207 compose_sflow_action(const struct ofproto_dpif *ofproto,
4208 struct ofpbuf *odp_actions,
4209 const struct flow *flow,
4212 uint32_t port_ifindex;
4213 uint32_t probability;
4214 struct user_action_cookie cookie;
4215 size_t sample_offset, actions_offset;
4216 int cookie_offset, n_output;
4218 if (!ofproto->sflow || flow->in_port == OFPP_NONE) {
4222 if (odp_port == OVSP_NONE) {
4226 port_ifindex = dpif_sflow_odp_port_to_ifindex(ofproto->sflow, odp_port);
4230 sample_offset = nl_msg_start_nested(odp_actions, OVS_ACTION_ATTR_SAMPLE);
4232 /* Number of packets out of UINT_MAX to sample. */
4233 probability = dpif_sflow_get_probability(ofproto->sflow);
4234 nl_msg_put_u32(odp_actions, OVS_SAMPLE_ATTR_PROBABILITY, probability);
4236 actions_offset = nl_msg_start_nested(odp_actions, OVS_SAMPLE_ATTR_ACTIONS);
4238 cookie.type = USER_ACTION_COOKIE_SFLOW;
4239 cookie.data = port_ifindex;
4240 cookie.n_output = n_output;
4241 cookie.vlan_tci = 0;
4242 cookie_offset = put_userspace_action(ofproto, odp_actions, flow, &cookie);
4244 nl_msg_end_nested(odp_actions, actions_offset);
4245 nl_msg_end_nested(odp_actions, sample_offset);
4246 return cookie_offset;
4249 /* SAMPLE action must be first action in any given list of actions.
4250 * At this point we do not have all information required to build it. So try to
4251 * build sample action as complete as possible. */
4253 add_sflow_action(struct action_xlate_ctx *ctx)
4255 ctx->user_cookie_offset = compose_sflow_action(ctx->ofproto,
4257 &ctx->flow, OVSP_NONE);
4258 ctx->sflow_odp_port = 0;
4259 ctx->sflow_n_outputs = 0;
4262 /* Fix SAMPLE action according to data collected while composing ODP actions.
4263 * We need to fix SAMPLE actions OVS_SAMPLE_ATTR_ACTIONS attribute, i.e. nested
4264 * USERSPACE action's user-cookie which is required for sflow. */
4266 fix_sflow_action(struct action_xlate_ctx *ctx)
4268 const struct flow *base = &ctx->base_flow;
4269 struct user_action_cookie *cookie;
4271 if (!ctx->user_cookie_offset) {
4275 cookie = ofpbuf_at(ctx->odp_actions, ctx->user_cookie_offset,
4277 assert(cookie != NULL);
4278 assert(cookie->type == USER_ACTION_COOKIE_SFLOW);
4280 if (ctx->sflow_n_outputs) {
4281 cookie->data = dpif_sflow_odp_port_to_ifindex(ctx->ofproto->sflow,
4282 ctx->sflow_odp_port);
4284 if (ctx->sflow_n_outputs >= 255) {
4285 cookie->n_output = 255;
4287 cookie->n_output = ctx->sflow_n_outputs;
4289 cookie->vlan_tci = base->vlan_tci;
4293 compose_output_action__(struct action_xlate_ctx *ctx, uint16_t ofp_port,
4296 const struct ofport_dpif *ofport = get_ofp_port(ctx->ofproto, ofp_port);
4297 uint16_t odp_port = ofp_port_to_odp_port(ofp_port);
4298 ovs_be16 flow_vlan_tci = ctx->flow.vlan_tci;
4299 uint8_t flow_nw_tos = ctx->flow.nw_tos;
4303 struct priority_to_dscp *pdscp;
4305 if (ofport->up.opp.config & htonl(OFPPC_NO_FWD)
4306 || (check_stp && !stp_forward_in_state(ofport->stp_state))) {
4310 pdscp = get_priority(ofport, ctx->flow.skb_priority);
4312 ctx->flow.nw_tos &= ~IP_DSCP_MASK;
4313 ctx->flow.nw_tos |= pdscp->dscp;
4316 /* We may not have an ofport record for this port, but it doesn't hurt
4317 * to allow forwarding to it anyhow. Maybe such a port will appear
4318 * later and we're pre-populating the flow table. */
4321 out_port = vsp_realdev_to_vlandev(ctx->ofproto, odp_port,
4322 ctx->flow.vlan_tci);
4323 if (out_port != odp_port) {
4324 ctx->flow.vlan_tci = htons(0);
4326 commit_odp_actions(&ctx->flow, &ctx->base_flow, ctx->odp_actions);
4327 nl_msg_put_u32(ctx->odp_actions, OVS_ACTION_ATTR_OUTPUT, out_port);
4329 ctx->sflow_odp_port = odp_port;
4330 ctx->sflow_n_outputs++;
4331 ctx->nf_output_iface = ofp_port;
4332 ctx->flow.vlan_tci = flow_vlan_tci;
4333 ctx->flow.nw_tos = flow_nw_tos;
4337 compose_output_action(struct action_xlate_ctx *ctx, uint16_t ofp_port)
4339 compose_output_action__(ctx, ofp_port, true);
4343 xlate_table_action(struct action_xlate_ctx *ctx,
4344 uint16_t in_port, uint8_t table_id)
4346 if (ctx->recurse < MAX_RESUBMIT_RECURSION) {
4347 struct ofproto_dpif *ofproto = ctx->ofproto;
4348 struct rule_dpif *rule;
4349 uint16_t old_in_port;
4350 uint8_t old_table_id;
4352 old_table_id = ctx->table_id;
4353 ctx->table_id = table_id;
4355 /* Look up a flow with 'in_port' as the input port. */
4356 old_in_port = ctx->flow.in_port;
4357 ctx->flow.in_port = in_port;
4358 rule = rule_dpif_lookup(ofproto, &ctx->flow, table_id);
4361 if (table_id > 0 && table_id < N_TABLES) {
4362 struct table_dpif *table = &ofproto->tables[table_id];
4363 if (table->other_table) {
4366 : rule_calculate_tag(&ctx->flow,
4367 &table->other_table->wc,
4372 /* Restore the original input port. Otherwise OFPP_NORMAL and
4373 * OFPP_IN_PORT will have surprising behavior. */
4374 ctx->flow.in_port = old_in_port;
4376 if (ctx->resubmit_hook) {
4377 ctx->resubmit_hook(ctx, rule);
4381 ovs_be64 old_cookie = ctx->cookie;
4384 ctx->cookie = rule->up.flow_cookie;
4385 do_xlate_actions(rule->up.actions, rule->up.n_actions, ctx);
4386 ctx->cookie = old_cookie;
4390 ctx->table_id = old_table_id;
4392 static struct vlog_rate_limit recurse_rl = VLOG_RATE_LIMIT_INIT(1, 1);
4394 VLOG_ERR_RL(&recurse_rl, "resubmit actions recursed over %d times",
4395 MAX_RESUBMIT_RECURSION);
4400 xlate_resubmit_table(struct action_xlate_ctx *ctx,
4401 const struct nx_action_resubmit *nar)
4406 in_port = (nar->in_port == htons(OFPP_IN_PORT)
4408 : ntohs(nar->in_port));
4409 table_id = nar->table == 255 ? ctx->table_id : nar->table;
4411 xlate_table_action(ctx, in_port, table_id);
4415 flood_packets(struct action_xlate_ctx *ctx, bool all)
4417 struct ofport_dpif *ofport;
4419 HMAP_FOR_EACH (ofport, up.hmap_node, &ctx->ofproto->up.ports) {
4420 uint16_t ofp_port = ofport->up.ofp_port;
4422 if (ofp_port == ctx->flow.in_port) {
4427 compose_output_action__(ctx, ofp_port, false);
4428 } else if (!(ofport->up.opp.config & htonl(OFPPC_NO_FLOOD))) {
4429 compose_output_action(ctx, ofp_port);
4433 ctx->nf_output_iface = NF_OUT_FLOOD;
4437 execute_controller_action(struct action_xlate_ctx *ctx, int len,
4438 enum ofp_packet_in_reason reason)
4440 struct ofputil_packet_in pin;
4441 struct ofpbuf *packet;
4443 ctx->may_set_up_flow = false;
4448 packet = ofpbuf_clone(ctx->packet);
4450 if (packet->l2 && packet->l3) {
4451 struct eth_header *eh;
4453 eth_pop_vlan(packet);
4455 assert(eh->eth_type == ctx->flow.dl_type);
4456 memcpy(eh->eth_src, ctx->flow.dl_src, sizeof eh->eth_src);
4457 memcpy(eh->eth_dst, ctx->flow.dl_dst, sizeof eh->eth_dst);
4459 if (ctx->flow.vlan_tci & htons(VLAN_CFI)) {
4460 eth_push_vlan(packet, ctx->flow.vlan_tci);
4464 if (ctx->flow.dl_type == htons(ETH_TYPE_IP)) {
4465 packet_set_ipv4(packet, ctx->flow.nw_src, ctx->flow.nw_dst,
4466 ctx->flow.nw_tos, ctx->flow.nw_ttl);
4470 if (ctx->flow.nw_proto == IPPROTO_TCP) {
4471 packet_set_tcp_port(packet, ctx->flow.tp_src,
4473 } else if (ctx->flow.nw_proto == IPPROTO_UDP) {
4474 packet_set_udp_port(packet, ctx->flow.tp_src,
4481 pin.packet = packet->data;
4482 pin.packet_len = packet->size;
4483 pin.reason = reason;
4484 pin.table_id = ctx->table_id;
4485 pin.cookie = ctx->cookie;
4489 pin.total_len = packet->size;
4490 flow_get_metadata(&ctx->flow, &pin.fmd);
4492 connmgr_send_packet_in(ctx->ofproto->up.connmgr, &pin, &ctx->flow);
4493 ofpbuf_delete(packet);
4497 compose_dec_ttl(struct action_xlate_ctx *ctx)
4499 if (ctx->flow.dl_type != htons(ETH_TYPE_IP) &&
4500 ctx->flow.dl_type != htons(ETH_TYPE_IPV6)) {
4504 if (ctx->flow.nw_ttl > 1) {
4508 execute_controller_action(ctx, UINT16_MAX, OFPR_INVALID_TTL);
4510 /* Stop processing for current table. */
4516 xlate_output_action__(struct action_xlate_ctx *ctx,
4517 uint16_t port, uint16_t max_len)
4519 uint16_t prev_nf_output_iface = ctx->nf_output_iface;
4521 ctx->nf_output_iface = NF_OUT_DROP;
4525 compose_output_action(ctx, ctx->flow.in_port);
4528 xlate_table_action(ctx, ctx->flow.in_port, ctx->table_id);
4534 flood_packets(ctx, false);
4537 flood_packets(ctx, true);
4539 case OFPP_CONTROLLER:
4540 execute_controller_action(ctx, max_len, OFPR_ACTION);
4543 compose_output_action(ctx, OFPP_LOCAL);
4548 if (port != ctx->flow.in_port) {
4549 compose_output_action(ctx, port);
4554 if (prev_nf_output_iface == NF_OUT_FLOOD) {
4555 ctx->nf_output_iface = NF_OUT_FLOOD;
4556 } else if (ctx->nf_output_iface == NF_OUT_DROP) {
4557 ctx->nf_output_iface = prev_nf_output_iface;
4558 } else if (prev_nf_output_iface != NF_OUT_DROP &&
4559 ctx->nf_output_iface != NF_OUT_FLOOD) {
4560 ctx->nf_output_iface = NF_OUT_MULTI;
4565 xlate_output_reg_action(struct action_xlate_ctx *ctx,
4566 const struct nx_action_output_reg *naor)
4568 struct mf_subfield src;
4571 nxm_decode(&src, naor->src, naor->ofs_nbits);
4572 ofp_port = mf_get_subfield(&src, &ctx->flow);
4574 if (ofp_port <= UINT16_MAX) {
4575 xlate_output_action__(ctx, ofp_port, ntohs(naor->max_len));
4580 xlate_output_action(struct action_xlate_ctx *ctx,
4581 const struct ofp_action_output *oao)
4583 xlate_output_action__(ctx, ntohs(oao->port), ntohs(oao->max_len));
4587 xlate_enqueue_action(struct action_xlate_ctx *ctx,
4588 const struct ofp_action_enqueue *oae)
4591 uint32_t flow_priority, priority;
4594 error = dpif_queue_to_priority(ctx->ofproto->dpif, ntohl(oae->queue_id),
4597 /* Fall back to ordinary output action. */
4598 xlate_output_action__(ctx, ntohs(oae->port), 0);
4602 /* Figure out datapath output port. */
4603 ofp_port = ntohs(oae->port);
4604 if (ofp_port == OFPP_IN_PORT) {
4605 ofp_port = ctx->flow.in_port;
4606 } else if (ofp_port == ctx->flow.in_port) {
4610 /* Add datapath actions. */
4611 flow_priority = ctx->flow.skb_priority;
4612 ctx->flow.skb_priority = priority;
4613 compose_output_action(ctx, ofp_port);
4614 ctx->flow.skb_priority = flow_priority;
4616 /* Update NetFlow output port. */
4617 if (ctx->nf_output_iface == NF_OUT_DROP) {
4618 ctx->nf_output_iface = ofp_port;
4619 } else if (ctx->nf_output_iface != NF_OUT_FLOOD) {
4620 ctx->nf_output_iface = NF_OUT_MULTI;
4625 xlate_set_queue_action(struct action_xlate_ctx *ctx,
4626 const struct nx_action_set_queue *nasq)
4631 error = dpif_queue_to_priority(ctx->ofproto->dpif, ntohl(nasq->queue_id),
4634 /* Couldn't translate queue to a priority, so ignore. A warning
4635 * has already been logged. */
4639 ctx->flow.skb_priority = priority;
4642 struct xlate_reg_state {
4648 xlate_autopath(struct action_xlate_ctx *ctx,
4649 const struct nx_action_autopath *naa)
4651 uint16_t ofp_port = ntohl(naa->id);
4652 struct ofport_dpif *port = get_ofp_port(ctx->ofproto, ofp_port);
4654 if (!port || !port->bundle) {
4655 ofp_port = OFPP_NONE;
4656 } else if (port->bundle->bond) {
4657 /* Autopath does not support VLAN hashing. */
4658 struct ofport_dpif *slave = bond_choose_output_slave(
4659 port->bundle->bond, &ctx->flow, 0, &ctx->tags);
4661 ofp_port = slave->up.ofp_port;
4664 autopath_execute(naa, &ctx->flow, ofp_port);
4668 slave_enabled_cb(uint16_t ofp_port, void *ofproto_)
4670 struct ofproto_dpif *ofproto = ofproto_;
4671 struct ofport_dpif *port;
4681 case OFPP_CONTROLLER: /* Not supported by the bundle action. */
4684 port = get_ofp_port(ofproto, ofp_port);
4685 return port ? port->may_enable : false;
4690 xlate_learn_action(struct action_xlate_ctx *ctx,
4691 const struct nx_action_learn *learn)
4693 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(5, 1);
4694 struct ofputil_flow_mod fm;
4697 learn_execute(learn, &ctx->flow, &fm);
4699 error = ofproto_flow_mod(&ctx->ofproto->up, &fm);
4700 if (error && !VLOG_DROP_WARN(&rl)) {
4701 VLOG_WARN("learning action failed to modify flow table (%s)",
4702 ofperr_get_name(error));
4709 may_receive(const struct ofport_dpif *port, struct action_xlate_ctx *ctx)
4711 if (port->up.opp.config & (eth_addr_equals(ctx->flow.dl_dst, eth_addr_stp)
4712 ? htonl(OFPPC_NO_RECV_STP)
4713 : htonl(OFPPC_NO_RECV))) {
4717 /* Only drop packets here if both forwarding and learning are
4718 * disabled. If just learning is enabled, we need to have
4719 * OFPP_NORMAL and the learning action have a look at the packet
4720 * before we can drop it. */
4721 if (!stp_forward_in_state(port->stp_state)
4722 && !stp_learn_in_state(port->stp_state)) {
4730 do_xlate_actions(const union ofp_action *in, size_t n_in,
4731 struct action_xlate_ctx *ctx)
4733 const struct ofport_dpif *port;
4734 const union ofp_action *ia;
4737 port = get_ofp_port(ctx->ofproto, ctx->flow.in_port);
4738 if (port && !may_receive(port, ctx)) {
4739 /* Drop this flow. */
4743 OFPUTIL_ACTION_FOR_EACH_UNSAFE (ia, left, in, n_in) {
4744 const struct ofp_action_dl_addr *oada;
4745 const struct nx_action_resubmit *nar;
4746 const struct nx_action_set_tunnel *nast;
4747 const struct nx_action_set_queue *nasq;
4748 const struct nx_action_multipath *nam;
4749 const struct nx_action_autopath *naa;
4750 const struct nx_action_bundle *nab;
4751 const struct nx_action_output_reg *naor;
4752 enum ofputil_action_code code;
4759 code = ofputil_decode_action_unsafe(ia);
4761 case OFPUTIL_OFPAT_OUTPUT:
4762 xlate_output_action(ctx, &ia->output);
4765 case OFPUTIL_OFPAT_SET_VLAN_VID:
4766 ctx->flow.vlan_tci &= ~htons(VLAN_VID_MASK);
4767 ctx->flow.vlan_tci |= ia->vlan_vid.vlan_vid | htons(VLAN_CFI);
4770 case OFPUTIL_OFPAT_SET_VLAN_PCP:
4771 ctx->flow.vlan_tci &= ~htons(VLAN_PCP_MASK);
4772 ctx->flow.vlan_tci |= htons(
4773 (ia->vlan_pcp.vlan_pcp << VLAN_PCP_SHIFT) | VLAN_CFI);
4776 case OFPUTIL_OFPAT_STRIP_VLAN:
4777 ctx->flow.vlan_tci = htons(0);
4780 case OFPUTIL_OFPAT_SET_DL_SRC:
4781 oada = ((struct ofp_action_dl_addr *) ia);
4782 memcpy(ctx->flow.dl_src, oada->dl_addr, ETH_ADDR_LEN);
4785 case OFPUTIL_OFPAT_SET_DL_DST:
4786 oada = ((struct ofp_action_dl_addr *) ia);
4787 memcpy(ctx->flow.dl_dst, oada->dl_addr, ETH_ADDR_LEN);
4790 case OFPUTIL_OFPAT_SET_NW_SRC:
4791 ctx->flow.nw_src = ia->nw_addr.nw_addr;
4794 case OFPUTIL_OFPAT_SET_NW_DST:
4795 ctx->flow.nw_dst = ia->nw_addr.nw_addr;
4798 case OFPUTIL_OFPAT_SET_NW_TOS:
4799 /* OpenFlow 1.0 only supports IPv4. */
4800 if (ctx->flow.dl_type == htons(ETH_TYPE_IP)) {
4801 ctx->flow.nw_tos &= ~IP_DSCP_MASK;
4802 ctx->flow.nw_tos |= ia->nw_tos.nw_tos & IP_DSCP_MASK;
4806 case OFPUTIL_OFPAT_SET_TP_SRC:
4807 ctx->flow.tp_src = ia->tp_port.tp_port;
4810 case OFPUTIL_OFPAT_SET_TP_DST:
4811 ctx->flow.tp_dst = ia->tp_port.tp_port;
4814 case OFPUTIL_OFPAT_ENQUEUE:
4815 xlate_enqueue_action(ctx, (const struct ofp_action_enqueue *) ia);
4818 case OFPUTIL_NXAST_RESUBMIT:
4819 nar = (const struct nx_action_resubmit *) ia;
4820 xlate_table_action(ctx, ntohs(nar->in_port), ctx->table_id);
4823 case OFPUTIL_NXAST_RESUBMIT_TABLE:
4824 xlate_resubmit_table(ctx, (const struct nx_action_resubmit *) ia);
4827 case OFPUTIL_NXAST_SET_TUNNEL:
4828 nast = (const struct nx_action_set_tunnel *) ia;
4829 tun_id = htonll(ntohl(nast->tun_id));
4830 ctx->flow.tun_id = tun_id;
4833 case OFPUTIL_NXAST_SET_QUEUE:
4834 nasq = (const struct nx_action_set_queue *) ia;
4835 xlate_set_queue_action(ctx, nasq);
4838 case OFPUTIL_NXAST_POP_QUEUE:
4839 ctx->flow.skb_priority = ctx->orig_skb_priority;
4842 case OFPUTIL_NXAST_REG_MOVE:
4843 nxm_execute_reg_move((const struct nx_action_reg_move *) ia,
4847 case OFPUTIL_NXAST_REG_LOAD:
4848 nxm_execute_reg_load((const struct nx_action_reg_load *) ia,
4852 case OFPUTIL_NXAST_NOTE:
4853 /* Nothing to do. */
4856 case OFPUTIL_NXAST_SET_TUNNEL64:
4857 tun_id = ((const struct nx_action_set_tunnel64 *) ia)->tun_id;
4858 ctx->flow.tun_id = tun_id;
4861 case OFPUTIL_NXAST_MULTIPATH:
4862 nam = (const struct nx_action_multipath *) ia;
4863 multipath_execute(nam, &ctx->flow);
4866 case OFPUTIL_NXAST_AUTOPATH:
4867 naa = (const struct nx_action_autopath *) ia;
4868 xlate_autopath(ctx, naa);
4871 case OFPUTIL_NXAST_BUNDLE:
4872 ctx->ofproto->has_bundle_action = true;
4873 nab = (const struct nx_action_bundle *) ia;
4874 xlate_output_action__(ctx, bundle_execute(nab, &ctx->flow,
4879 case OFPUTIL_NXAST_BUNDLE_LOAD:
4880 ctx->ofproto->has_bundle_action = true;
4881 nab = (const struct nx_action_bundle *) ia;
4882 bundle_execute_load(nab, &ctx->flow, slave_enabled_cb,
4886 case OFPUTIL_NXAST_OUTPUT_REG:
4887 naor = (const struct nx_action_output_reg *) ia;
4888 xlate_output_reg_action(ctx, naor);
4891 case OFPUTIL_NXAST_LEARN:
4892 ctx->has_learn = true;
4893 if (ctx->may_learn) {
4894 xlate_learn_action(ctx, (const struct nx_action_learn *) ia);
4898 case OFPUTIL_NXAST_DEC_TTL:
4899 if (compose_dec_ttl(ctx)) {
4904 case OFPUTIL_NXAST_EXIT:
4911 /* We've let OFPP_NORMAL and the learning action look at the packet,
4912 * so drop it now if forwarding is disabled. */
4913 if (port && !stp_forward_in_state(port->stp_state)) {
4914 ofpbuf_clear(ctx->odp_actions);
4915 add_sflow_action(ctx);
4920 action_xlate_ctx_init(struct action_xlate_ctx *ctx,
4921 struct ofproto_dpif *ofproto, const struct flow *flow,
4922 ovs_be16 initial_tci, ovs_be64 cookie,
4923 const struct ofpbuf *packet)
4925 ctx->ofproto = ofproto;
4927 ctx->base_flow = ctx->flow;
4928 ctx->base_flow.tun_id = 0;
4929 ctx->base_flow.vlan_tci = initial_tci;
4930 ctx->cookie = cookie;
4931 ctx->packet = packet;
4932 ctx->may_learn = packet != NULL;
4933 ctx->resubmit_hook = NULL;
4936 static struct ofpbuf *
4937 xlate_actions(struct action_xlate_ctx *ctx,
4938 const union ofp_action *in, size_t n_in)
4940 struct flow orig_flow = ctx->flow;
4942 COVERAGE_INC(ofproto_dpif_xlate);
4944 ctx->odp_actions = ofpbuf_new(512);
4945 ofpbuf_reserve(ctx->odp_actions, NL_A_U32_SIZE);
4947 ctx->may_set_up_flow = true;
4948 ctx->has_learn = false;
4949 ctx->has_normal = false;
4950 ctx->nf_output_iface = NF_OUT_DROP;
4953 ctx->orig_skb_priority = ctx->flow.skb_priority;
4957 if (ctx->flow.nw_frag & FLOW_NW_FRAG_ANY) {
4958 switch (ctx->ofproto->up.frag_handling) {
4959 case OFPC_FRAG_NORMAL:
4960 /* We must pretend that transport ports are unavailable. */
4961 ctx->flow.tp_src = ctx->base_flow.tp_src = htons(0);
4962 ctx->flow.tp_dst = ctx->base_flow.tp_dst = htons(0);
4965 case OFPC_FRAG_DROP:
4966 return ctx->odp_actions;
4968 case OFPC_FRAG_REASM:
4971 case OFPC_FRAG_NX_MATCH:
4972 /* Nothing to do. */
4975 case OFPC_INVALID_TTL_TO_CONTROLLER:
4980 if (process_special(ctx->ofproto, &ctx->flow, ctx->packet)) {
4981 ctx->may_set_up_flow = false;
4982 return ctx->odp_actions;
4984 add_sflow_action(ctx);
4985 do_xlate_actions(in, n_in, ctx);
4987 if (!connmgr_may_set_up_flow(ctx->ofproto->up.connmgr, &ctx->flow,
4988 ctx->odp_actions->data,
4989 ctx->odp_actions->size)) {
4990 ctx->may_set_up_flow = false;
4992 && connmgr_msg_in_hook(ctx->ofproto->up.connmgr, &ctx->flow,
4994 compose_output_action(ctx, OFPP_LOCAL);
4997 add_mirror_actions(ctx, &orig_flow);
4998 fix_sflow_action(ctx);
5001 return ctx->odp_actions;
5004 /* OFPP_NORMAL implementation. */
5006 static struct ofport_dpif *ofbundle_get_a_port(const struct ofbundle *);
5008 /* Given 'vid', the VID obtained from the 802.1Q header that was received as
5009 * part of a packet (specify 0 if there was no 802.1Q header), and 'in_bundle',
5010 * the bundle on which the packet was received, returns the VLAN to which the
5013 * Both 'vid' and the return value are in the range 0...4095. */
5015 input_vid_to_vlan(const struct ofbundle *in_bundle, uint16_t vid)
5017 switch (in_bundle->vlan_mode) {
5018 case PORT_VLAN_ACCESS:
5019 return in_bundle->vlan;
5022 case PORT_VLAN_TRUNK:
5025 case PORT_VLAN_NATIVE_UNTAGGED:
5026 case PORT_VLAN_NATIVE_TAGGED:
5027 return vid ? vid : in_bundle->vlan;
5034 /* Checks whether a packet with the given 'vid' may ingress on 'in_bundle'.
5035 * If so, returns true. Otherwise, returns false and, if 'warn' is true, logs
5038 * 'vid' should be the VID obtained from the 802.1Q header that was received as
5039 * part of a packet (specify 0 if there was no 802.1Q header), in the range
5042 input_vid_is_valid(uint16_t vid, struct ofbundle *in_bundle, bool warn)
5044 /* Allow any VID on the OFPP_NONE port. */
5045 if (in_bundle == &ofpp_none_bundle) {
5049 switch (in_bundle->vlan_mode) {
5050 case PORT_VLAN_ACCESS:
5053 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
5054 VLOG_WARN_RL(&rl, "bridge %s: dropping VLAN %"PRIu16" tagged "
5055 "packet received on port %s configured as VLAN "
5056 "%"PRIu16" access port",
5057 in_bundle->ofproto->up.name, vid,
5058 in_bundle->name, in_bundle->vlan);
5064 case PORT_VLAN_NATIVE_UNTAGGED:
5065 case PORT_VLAN_NATIVE_TAGGED:
5067 /* Port must always carry its native VLAN. */
5071 case PORT_VLAN_TRUNK:
5072 if (!ofbundle_includes_vlan(in_bundle, vid)) {
5074 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
5075 VLOG_WARN_RL(&rl, "bridge %s: dropping VLAN %"PRIu16" packet "
5076 "received on port %s not configured for trunking "
5078 in_bundle->ofproto->up.name, vid,
5079 in_bundle->name, vid);
5091 /* Given 'vlan', the VLAN that a packet belongs to, and
5092 * 'out_bundle', a bundle on which the packet is to be output, returns the VID
5093 * that should be included in the 802.1Q header. (If the return value is 0,
5094 * then the 802.1Q header should only be included in the packet if there is a
5097 * Both 'vlan' and the return value are in the range 0...4095. */
5099 output_vlan_to_vid(const struct ofbundle *out_bundle, uint16_t vlan)
5101 switch (out_bundle->vlan_mode) {
5102 case PORT_VLAN_ACCESS:
5105 case PORT_VLAN_TRUNK:
5106 case PORT_VLAN_NATIVE_TAGGED:
5109 case PORT_VLAN_NATIVE_UNTAGGED:
5110 return vlan == out_bundle->vlan ? 0 : vlan;
5118 output_normal(struct action_xlate_ctx *ctx, const struct ofbundle *out_bundle,
5121 struct ofport_dpif *port;
5123 ovs_be16 tci, old_tci;
5125 vid = output_vlan_to_vid(out_bundle, vlan);
5126 if (!out_bundle->bond) {
5127 port = ofbundle_get_a_port(out_bundle);
5129 port = bond_choose_output_slave(out_bundle->bond, &ctx->flow,
5132 /* No slaves enabled, so drop packet. */
5137 old_tci = ctx->flow.vlan_tci;
5139 if (tci || out_bundle->use_priority_tags) {
5140 tci |= ctx->flow.vlan_tci & htons(VLAN_PCP_MASK);
5142 tci |= htons(VLAN_CFI);
5145 ctx->flow.vlan_tci = tci;
5147 compose_output_action(ctx, port->up.ofp_port);
5148 ctx->flow.vlan_tci = old_tci;
5152 mirror_mask_ffs(mirror_mask_t mask)
5154 BUILD_ASSERT_DECL(sizeof(unsigned int) >= sizeof(mask));
5159 ofbundle_trunks_vlan(const struct ofbundle *bundle, uint16_t vlan)
5161 return (bundle->vlan_mode != PORT_VLAN_ACCESS
5162 && (!bundle->trunks || bitmap_is_set(bundle->trunks, vlan)));
5166 ofbundle_includes_vlan(const struct ofbundle *bundle, uint16_t vlan)
5168 return vlan == bundle->vlan || ofbundle_trunks_vlan(bundle, vlan);
5171 /* Returns an arbitrary interface within 'bundle'. */
5172 static struct ofport_dpif *
5173 ofbundle_get_a_port(const struct ofbundle *bundle)
5175 return CONTAINER_OF(list_front(&bundle->ports),
5176 struct ofport_dpif, bundle_node);
5180 vlan_is_mirrored(const struct ofmirror *m, int vlan)
5182 return !m->vlans || bitmap_is_set(m->vlans, vlan);
5185 /* Returns true if a packet with Ethernet destination MAC 'dst' may be mirrored
5186 * to a VLAN. In general most packets may be mirrored but we want to drop
5187 * protocols that may confuse switches. */
5189 eth_dst_may_rspan(const uint8_t dst[ETH_ADDR_LEN])
5191 /* If you change this function's behavior, please update corresponding
5192 * documentation in vswitch.xml at the same time. */
5193 if (dst[0] != 0x01) {
5194 /* All the currently banned MACs happen to start with 01 currently, so
5195 * this is a quick way to eliminate most of the good ones. */
5197 if (eth_addr_is_reserved(dst)) {
5198 /* Drop STP, IEEE pause frames, and other reserved protocols
5199 * (01-80-c2-00-00-0x). */
5203 if (dst[0] == 0x01 && dst[1] == 0x00 && dst[2] == 0x0c) {
5205 if ((dst[3] & 0xfe) == 0xcc &&
5206 (dst[4] & 0xfe) == 0xcc &&
5207 (dst[5] & 0xfe) == 0xcc) {
5208 /* Drop the following protocols plus others following the same
5211 CDP, VTP, DTP, PAgP (01-00-0c-cc-cc-cc)
5212 Spanning Tree PVSTP+ (01-00-0c-cc-cc-cd)
5213 STP Uplink Fast (01-00-0c-cd-cd-cd) */
5217 if (!(dst[3] | dst[4] | dst[5])) {
5218 /* Drop Inter Switch Link packets (01-00-0c-00-00-00). */
5227 add_mirror_actions(struct action_xlate_ctx *ctx, const struct flow *orig_flow)
5229 struct ofproto_dpif *ofproto = ctx->ofproto;
5230 mirror_mask_t mirrors;
5231 struct ofbundle *in_bundle;
5234 const struct nlattr *a;
5237 in_bundle = lookup_input_bundle(ctx->ofproto, orig_flow->in_port,
5238 ctx->packet != NULL);
5242 mirrors = in_bundle->src_mirrors;
5244 /* Drop frames on bundles reserved for mirroring. */
5245 if (in_bundle->mirror_out) {
5246 if (ctx->packet != NULL) {
5247 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
5248 VLOG_WARN_RL(&rl, "bridge %s: dropping packet received on port "
5249 "%s, which is reserved exclusively for mirroring",
5250 ctx->ofproto->up.name, in_bundle->name);
5256 vid = vlan_tci_to_vid(orig_flow->vlan_tci);
5257 if (!input_vid_is_valid(vid, in_bundle, ctx->packet != NULL)) {
5260 vlan = input_vid_to_vlan(in_bundle, vid);
5262 /* Look at the output ports to check for destination selections. */
5264 NL_ATTR_FOR_EACH (a, left, ctx->odp_actions->data,
5265 ctx->odp_actions->size) {
5266 enum ovs_action_attr type = nl_attr_type(a);
5267 struct ofport_dpif *ofport;
5269 if (type != OVS_ACTION_ATTR_OUTPUT) {
5273 ofport = get_odp_port(ofproto, nl_attr_get_u32(a));
5274 if (ofport && ofport->bundle) {
5275 mirrors |= ofport->bundle->dst_mirrors;
5283 /* Restore the original packet before adding the mirror actions. */
5284 ctx->flow = *orig_flow;
5289 m = ofproto->mirrors[mirror_mask_ffs(mirrors) - 1];
5291 if (!vlan_is_mirrored(m, vlan)) {
5292 mirrors &= mirrors - 1;
5296 mirrors &= ~m->dup_mirrors;
5297 ctx->mirrors |= m->dup_mirrors;
5299 output_normal(ctx, m->out, vlan);
5300 } else if (eth_dst_may_rspan(orig_flow->dl_dst)
5301 && vlan != m->out_vlan) {
5302 struct ofbundle *bundle;
5304 HMAP_FOR_EACH (bundle, hmap_node, &ofproto->bundles) {
5305 if (ofbundle_includes_vlan(bundle, m->out_vlan)
5306 && !bundle->mirror_out) {
5307 output_normal(ctx, bundle, m->out_vlan);
5315 update_mirror_stats(struct ofproto_dpif *ofproto, mirror_mask_t mirrors,
5316 uint64_t packets, uint64_t bytes)
5322 for (; mirrors; mirrors &= mirrors - 1) {
5325 m = ofproto->mirrors[mirror_mask_ffs(mirrors) - 1];
5328 /* In normal circumstances 'm' will not be NULL. However,
5329 * if mirrors are reconfigured, we can temporarily get out
5330 * of sync in facet_revalidate(). We could "correct" the
5331 * mirror list before reaching here, but doing that would
5332 * not properly account the traffic stats we've currently
5333 * accumulated for previous mirror configuration. */
5337 m->packet_count += packets;
5338 m->byte_count += bytes;
5342 /* A VM broadcasts a gratuitous ARP to indicate that it has resumed after
5343 * migration. Older Citrix-patched Linux DomU used gratuitous ARP replies to
5344 * indicate this; newer upstream kernels use gratuitous ARP requests. */
5346 is_gratuitous_arp(const struct flow *flow)
5348 return (flow->dl_type == htons(ETH_TYPE_ARP)
5349 && eth_addr_is_broadcast(flow->dl_dst)
5350 && (flow->nw_proto == ARP_OP_REPLY
5351 || (flow->nw_proto == ARP_OP_REQUEST
5352 && flow->nw_src == flow->nw_dst)));
5356 update_learning_table(struct ofproto_dpif *ofproto,
5357 const struct flow *flow, int vlan,
5358 struct ofbundle *in_bundle)
5360 struct mac_entry *mac;
5362 /* Don't learn the OFPP_NONE port. */
5363 if (in_bundle == &ofpp_none_bundle) {
5367 if (!mac_learning_may_learn(ofproto->ml, flow->dl_src, vlan)) {
5371 mac = mac_learning_insert(ofproto->ml, flow->dl_src, vlan);
5372 if (is_gratuitous_arp(flow)) {
5373 /* We don't want to learn from gratuitous ARP packets that are
5374 * reflected back over bond slaves so we lock the learning table. */
5375 if (!in_bundle->bond) {
5376 mac_entry_set_grat_arp_lock(mac);
5377 } else if (mac_entry_is_grat_arp_locked(mac)) {
5382 if (mac_entry_is_new(mac) || mac->port.p != in_bundle) {
5383 /* The log messages here could actually be useful in debugging,
5384 * so keep the rate limit relatively high. */
5385 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(30, 300);
5386 VLOG_DBG_RL(&rl, "bridge %s: learned that "ETH_ADDR_FMT" is "
5387 "on port %s in VLAN %d",
5388 ofproto->up.name, ETH_ADDR_ARGS(flow->dl_src),
5389 in_bundle->name, vlan);
5391 mac->port.p = in_bundle;
5392 tag_set_add(&ofproto->revalidate_set,
5393 mac_learning_changed(ofproto->ml, mac));
5397 static struct ofbundle *
5398 lookup_input_bundle(struct ofproto_dpif *ofproto, uint16_t in_port, bool warn)
5400 struct ofport_dpif *ofport;
5402 /* Special-case OFPP_NONE, which a controller may use as the ingress
5403 * port for traffic that it is sourcing. */
5404 if (in_port == OFPP_NONE) {
5405 return &ofpp_none_bundle;
5408 /* Find the port and bundle for the received packet. */
5409 ofport = get_ofp_port(ofproto, in_port);
5410 if (ofport && ofport->bundle) {
5411 return ofport->bundle;
5414 /* Odd. A few possible reasons here:
5416 * - We deleted a port but there are still a few packets queued up
5419 * - Someone externally added a port (e.g. "ovs-dpctl add-if") that
5420 * we don't know about.
5422 * - The ofproto client didn't configure the port as part of a bundle.
5425 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
5427 VLOG_WARN_RL(&rl, "bridge %s: received packet on unknown "
5428 "port %"PRIu16, ofproto->up.name, in_port);
5433 /* Determines whether packets in 'flow' within 'ofproto' should be forwarded or
5434 * dropped. Returns true if they may be forwarded, false if they should be
5437 * 'in_port' must be the ofport_dpif that corresponds to flow->in_port.
5438 * 'in_port' must be part of a bundle (e.g. in_port->bundle must be nonnull).
5440 * 'vlan' must be the VLAN that corresponds to flow->vlan_tci on 'in_port', as
5441 * returned by input_vid_to_vlan(). It must be a valid VLAN for 'in_port', as
5442 * checked by input_vid_is_valid().
5444 * May also add tags to '*tags', although the current implementation only does
5445 * so in one special case.
5448 is_admissible(struct ofproto_dpif *ofproto, const struct flow *flow,
5449 struct ofport_dpif *in_port, uint16_t vlan, tag_type *tags)
5451 struct ofbundle *in_bundle = in_port->bundle;
5453 /* Drop frames for reserved multicast addresses
5454 * only if forward_bpdu option is absent. */
5455 if (eth_addr_is_reserved(flow->dl_dst) && !ofproto->up.forward_bpdu) {
5459 if (in_bundle->bond) {
5460 struct mac_entry *mac;
5462 switch (bond_check_admissibility(in_bundle->bond, in_port,
5463 flow->dl_dst, tags)) {
5470 case BV_DROP_IF_MOVED:
5471 mac = mac_learning_lookup(ofproto->ml, flow->dl_src, vlan, NULL);
5472 if (mac && mac->port.p != in_bundle &&
5473 (!is_gratuitous_arp(flow)
5474 || mac_entry_is_grat_arp_locked(mac))) {
5485 xlate_normal(struct action_xlate_ctx *ctx)
5487 struct ofport_dpif *in_port;
5488 struct ofbundle *in_bundle;
5489 struct mac_entry *mac;
5493 ctx->has_normal = true;
5495 in_bundle = lookup_input_bundle(ctx->ofproto, ctx->flow.in_port,
5496 ctx->packet != NULL);
5501 /* We know 'in_port' exists unless it is "ofpp_none_bundle",
5502 * since lookup_input_bundle() succeeded. */
5503 in_port = get_ofp_port(ctx->ofproto, ctx->flow.in_port);
5505 /* Drop malformed frames. */
5506 if (ctx->flow.dl_type == htons(ETH_TYPE_VLAN) &&
5507 !(ctx->flow.vlan_tci & htons(VLAN_CFI))) {
5508 if (ctx->packet != NULL) {
5509 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
5510 VLOG_WARN_RL(&rl, "bridge %s: dropping packet with partial "
5511 "VLAN tag received on port %s",
5512 ctx->ofproto->up.name, in_bundle->name);
5517 /* Drop frames on bundles reserved for mirroring. */
5518 if (in_bundle->mirror_out) {
5519 if (ctx->packet != NULL) {
5520 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
5521 VLOG_WARN_RL(&rl, "bridge %s: dropping packet received on port "
5522 "%s, which is reserved exclusively for mirroring",
5523 ctx->ofproto->up.name, in_bundle->name);
5529 vid = vlan_tci_to_vid(ctx->flow.vlan_tci);
5530 if (!input_vid_is_valid(vid, in_bundle, ctx->packet != NULL)) {
5533 vlan = input_vid_to_vlan(in_bundle, vid);
5535 /* Check other admissibility requirements. */
5537 !is_admissible(ctx->ofproto, &ctx->flow, in_port, vlan, &ctx->tags)) {
5541 /* Learn source MAC. */
5542 if (ctx->may_learn) {
5543 update_learning_table(ctx->ofproto, &ctx->flow, vlan, in_bundle);
5546 /* Determine output bundle. */
5547 mac = mac_learning_lookup(ctx->ofproto->ml, ctx->flow.dl_dst, vlan,
5550 if (mac->port.p != in_bundle) {
5551 output_normal(ctx, mac->port.p, vlan);
5554 struct ofbundle *bundle;
5556 HMAP_FOR_EACH (bundle, hmap_node, &ctx->ofproto->bundles) {
5557 if (bundle != in_bundle
5558 && ofbundle_includes_vlan(bundle, vlan)
5559 && bundle->floodable
5560 && !bundle->mirror_out) {
5561 output_normal(ctx, bundle, vlan);
5564 ctx->nf_output_iface = NF_OUT_FLOOD;
5568 /* Optimized flow revalidation.
5570 * It's a difficult problem, in general, to tell which facets need to have
5571 * their actions recalculated whenever the OpenFlow flow table changes. We
5572 * don't try to solve that general problem: for most kinds of OpenFlow flow
5573 * table changes, we recalculate the actions for every facet. This is
5574 * relatively expensive, but it's good enough if the OpenFlow flow table
5575 * doesn't change very often.
5577 * However, we can expect one particular kind of OpenFlow flow table change to
5578 * happen frequently: changes caused by MAC learning. To avoid wasting a lot
5579 * of CPU on revalidating every facet whenever MAC learning modifies the flow
5580 * table, we add a special case that applies to flow tables in which every rule
5581 * has the same form (that is, the same wildcards), except that the table is
5582 * also allowed to have a single "catch-all" flow that matches all packets. We
5583 * optimize this case by tagging all of the facets that resubmit into the table
5584 * and invalidating the same tag whenever a flow changes in that table. The
5585 * end result is that we revalidate just the facets that need it (and sometimes
5586 * a few more, but not all of the facets or even all of the facets that
5587 * resubmit to the table modified by MAC learning). */
5589 /* Calculates the tag to use for 'flow' and wildcards 'wc' when it is inserted
5590 * into an OpenFlow table with the given 'basis'. */
5592 rule_calculate_tag(const struct flow *flow, const struct flow_wildcards *wc,
5595 if (flow_wildcards_is_catchall(wc)) {
5598 struct flow tag_flow = *flow;
5599 flow_zero_wildcards(&tag_flow, wc);
5600 return tag_create_deterministic(flow_hash(&tag_flow, secret));
5604 /* Following a change to OpenFlow table 'table_id' in 'ofproto', update the
5605 * taggability of that table.
5607 * This function must be called after *each* change to a flow table. If you
5608 * skip calling it on some changes then the pointer comparisons at the end can
5609 * be invalid if you get unlucky. For example, if a flow removal causes a
5610 * cls_table to be destroyed and then a flow insertion causes a cls_table with
5611 * different wildcards to be created with the same address, then this function
5612 * will incorrectly skip revalidation. */
5614 table_update_taggable(struct ofproto_dpif *ofproto, uint8_t table_id)
5616 struct table_dpif *table = &ofproto->tables[table_id];
5617 const struct oftable *oftable = &ofproto->up.tables[table_id];
5618 struct cls_table *catchall, *other;
5619 struct cls_table *t;
5621 catchall = other = NULL;
5623 switch (hmap_count(&oftable->cls.tables)) {
5625 /* We could tag this OpenFlow table but it would make the logic a
5626 * little harder and it's a corner case that doesn't seem worth it
5632 HMAP_FOR_EACH (t, hmap_node, &oftable->cls.tables) {
5633 if (cls_table_is_catchall(t)) {
5635 } else if (!other) {
5638 /* Indicate that we can't tag this by setting both tables to
5639 * NULL. (We know that 'catchall' is already NULL.) */
5646 /* Can't tag this table. */
5650 if (table->catchall_table != catchall || table->other_table != other) {
5651 table->catchall_table = catchall;
5652 table->other_table = other;
5653 ofproto->need_revalidate = true;
5657 /* Given 'rule' that has changed in some way (either it is a rule being
5658 * inserted, a rule being deleted, or a rule whose actions are being
5659 * modified), marks facets for revalidation to ensure that packets will be
5660 * forwarded correctly according to the new state of the flow table.
5662 * This function must be called after *each* change to a flow table. See
5663 * the comment on table_update_taggable() for more information. */
5665 rule_invalidate(const struct rule_dpif *rule)
5667 struct ofproto_dpif *ofproto = ofproto_dpif_cast(rule->up.ofproto);
5669 table_update_taggable(ofproto, rule->up.table_id);
5671 if (!ofproto->need_revalidate) {
5672 struct table_dpif *table = &ofproto->tables[rule->up.table_id];
5674 if (table->other_table && rule->tag) {
5675 tag_set_add(&ofproto->revalidate_set, rule->tag);
5677 ofproto->need_revalidate = true;
5683 set_frag_handling(struct ofproto *ofproto_,
5684 enum ofp_config_flags frag_handling)
5686 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
5688 if (frag_handling != OFPC_FRAG_REASM) {
5689 ofproto->need_revalidate = true;
5697 packet_out(struct ofproto *ofproto_, struct ofpbuf *packet,
5698 const struct flow *flow,
5699 const union ofp_action *ofp_actions, size_t n_ofp_actions)
5701 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
5704 if (flow->in_port >= ofproto->max_ports && flow->in_port < OFPP_MAX) {
5705 return OFPERR_NXBRC_BAD_IN_PORT;
5708 error = validate_actions(ofp_actions, n_ofp_actions, flow,
5709 ofproto->max_ports);
5711 struct odputil_keybuf keybuf;
5712 struct ofpbuf *odp_actions;
5713 struct ofproto_push push;
5716 ofpbuf_use_stack(&key, &keybuf, sizeof keybuf);
5717 odp_flow_key_from_flow(&key, flow);
5719 action_xlate_ctx_init(&push.ctx, ofproto, flow, flow->vlan_tci, 0,
5722 /* Ensure that resubmits in 'ofp_actions' get accounted to their
5723 * matching rules. */
5725 push.bytes = packet->size;
5726 push.used = time_msec();
5727 push.ctx.resubmit_hook = push_resubmit;
5729 odp_actions = xlate_actions(&push.ctx, ofp_actions, n_ofp_actions);
5730 dpif_execute(ofproto->dpif, key.data, key.size,
5731 odp_actions->data, odp_actions->size, packet);
5732 ofpbuf_delete(odp_actions);
5740 set_netflow(struct ofproto *ofproto_,
5741 const struct netflow_options *netflow_options)
5743 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
5745 if (netflow_options) {
5746 if (!ofproto->netflow) {
5747 ofproto->netflow = netflow_create();
5749 return netflow_set_options(ofproto->netflow, netflow_options);
5751 netflow_destroy(ofproto->netflow);
5752 ofproto->netflow = NULL;
5758 get_netflow_ids(const struct ofproto *ofproto_,
5759 uint8_t *engine_type, uint8_t *engine_id)
5761 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
5763 dpif_get_netflow_ids(ofproto->dpif, engine_type, engine_id);
5767 send_active_timeout(struct ofproto_dpif *ofproto, struct facet *facet)
5769 if (!facet_is_controller_flow(facet) &&
5770 netflow_active_timeout_expired(ofproto->netflow, &facet->nf_flow)) {
5771 struct subfacet *subfacet;
5772 struct ofexpired expired;
5774 LIST_FOR_EACH (subfacet, list_node, &facet->subfacets) {
5775 if (subfacet->installed) {
5776 struct dpif_flow_stats stats;
5778 subfacet_install(subfacet, subfacet->actions,
5779 subfacet->actions_len, &stats);
5780 subfacet_update_stats(subfacet, &stats);
5784 expired.flow = facet->flow;
5785 expired.packet_count = facet->packet_count;
5786 expired.byte_count = facet->byte_count;
5787 expired.used = facet->used;
5788 netflow_expire(ofproto->netflow, &facet->nf_flow, &expired);
5793 send_netflow_active_timeouts(struct ofproto_dpif *ofproto)
5795 struct facet *facet;
5797 HMAP_FOR_EACH (facet, hmap_node, &ofproto->facets) {
5798 send_active_timeout(ofproto, facet);
5802 static struct ofproto_dpif *
5803 ofproto_dpif_lookup(const char *name)
5805 struct ofproto_dpif *ofproto;
5807 HMAP_FOR_EACH_WITH_HASH (ofproto, all_ofproto_dpifs_node,
5808 hash_string(name, 0), &all_ofproto_dpifs) {
5809 if (!strcmp(ofproto->up.name, name)) {
5817 ofproto_unixctl_fdb_flush(struct unixctl_conn *conn, int argc,
5818 const char *argv[], void *aux OVS_UNUSED)
5820 struct ofproto_dpif *ofproto;
5823 ofproto = ofproto_dpif_lookup(argv[1]);
5825 unixctl_command_reply(conn, 501, "no such bridge");
5828 mac_learning_flush(ofproto->ml, &ofproto->revalidate_set);
5830 HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
5831 mac_learning_flush(ofproto->ml, &ofproto->revalidate_set);
5835 unixctl_command_reply(conn, 200, "table successfully flushed");
5839 ofproto_unixctl_fdb_show(struct unixctl_conn *conn, int argc OVS_UNUSED,
5840 const char *argv[], void *aux OVS_UNUSED)
5842 struct ds ds = DS_EMPTY_INITIALIZER;
5843 const struct ofproto_dpif *ofproto;
5844 const struct mac_entry *e;
5846 ofproto = ofproto_dpif_lookup(argv[1]);
5848 unixctl_command_reply(conn, 501, "no such bridge");
5852 ds_put_cstr(&ds, " port VLAN MAC Age\n");
5853 LIST_FOR_EACH (e, lru_node, &ofproto->ml->lrus) {
5854 struct ofbundle *bundle = e->port.p;
5855 ds_put_format(&ds, "%5d %4d "ETH_ADDR_FMT" %3d\n",
5856 ofbundle_get_a_port(bundle)->odp_port,
5857 e->vlan, ETH_ADDR_ARGS(e->mac), mac_entry_age(e));
5859 unixctl_command_reply(conn, 200, ds_cstr(&ds));
5863 struct ofproto_trace {
5864 struct action_xlate_ctx ctx;
5870 trace_format_rule(struct ds *result, uint8_t table_id, int level,
5871 const struct rule_dpif *rule)
5873 ds_put_char_multiple(result, '\t', level);
5875 ds_put_cstr(result, "No match\n");
5879 ds_put_format(result, "Rule: table=%"PRIu8" cookie=%#"PRIx64" ",
5880 table_id, ntohll(rule->up.flow_cookie));
5881 cls_rule_format(&rule->up.cr, result);
5882 ds_put_char(result, '\n');
5884 ds_put_char_multiple(result, '\t', level);
5885 ds_put_cstr(result, "OpenFlow ");
5886 ofp_print_actions(result, rule->up.actions, rule->up.n_actions);
5887 ds_put_char(result, '\n');
5891 trace_format_flow(struct ds *result, int level, const char *title,
5892 struct ofproto_trace *trace)
5894 ds_put_char_multiple(result, '\t', level);
5895 ds_put_format(result, "%s: ", title);
5896 if (flow_equal(&trace->ctx.flow, &trace->flow)) {
5897 ds_put_cstr(result, "unchanged");
5899 flow_format(result, &trace->ctx.flow);
5900 trace->flow = trace->ctx.flow;
5902 ds_put_char(result, '\n');
5906 trace_format_regs(struct ds *result, int level, const char *title,
5907 struct ofproto_trace *trace)
5911 ds_put_char_multiple(result, '\t', level);
5912 ds_put_format(result, "%s:", title);
5913 for (i = 0; i < FLOW_N_REGS; i++) {
5914 ds_put_format(result, " reg%zu=0x%"PRIx32, i, trace->flow.regs[i]);
5916 ds_put_char(result, '\n');
5920 trace_resubmit(struct action_xlate_ctx *ctx, struct rule_dpif *rule)
5922 struct ofproto_trace *trace = CONTAINER_OF(ctx, struct ofproto_trace, ctx);
5923 struct ds *result = trace->result;
5925 ds_put_char(result, '\n');
5926 trace_format_flow(result, ctx->recurse + 1, "Resubmitted flow", trace);
5927 trace_format_regs(result, ctx->recurse + 1, "Resubmitted regs", trace);
5928 trace_format_rule(result, ctx->table_id, ctx->recurse + 1, rule);
5932 ofproto_unixctl_trace(struct unixctl_conn *conn, int argc, const char *argv[],
5933 void *aux OVS_UNUSED)
5935 const char *dpname = argv[1];
5936 struct ofproto_dpif *ofproto;
5937 struct ofpbuf odp_key;
5938 struct ofpbuf *packet;
5939 struct rule_dpif *rule;
5940 ovs_be16 initial_tci;
5946 ofpbuf_init(&odp_key, 0);
5949 ofproto = ofproto_dpif_lookup(dpname);
5951 unixctl_command_reply(conn, 501, "Unknown ofproto (use ofproto/list "
5955 if (argc == 3 || (argc == 4 && !strcmp(argv[3], "-generate"))) {
5956 /* ofproto/trace dpname flow [-generate] */
5957 const char *flow_s = argv[2];
5958 const char *generate_s = argv[3];
5961 /* Convert string to datapath key. */
5962 ofpbuf_init(&odp_key, 0);
5963 error = odp_flow_key_from_string(flow_s, NULL, &odp_key);
5965 unixctl_command_reply(conn, 501, "Bad flow syntax");
5969 /* Convert odp_key to flow. */
5970 error = ofproto_dpif_extract_flow_key(ofproto, odp_key.data,
5971 odp_key.size, &flow,
5972 &initial_tci, NULL);
5973 if (error == ODP_FIT_ERROR) {
5974 unixctl_command_reply(conn, 501, "Invalid flow");
5978 /* Generate a packet, if requested. */
5980 packet = ofpbuf_new(0);
5981 flow_compose(packet, &flow);
5983 } else if (argc == 6) {
5984 /* ofproto/trace dpname priority tun_id in_port packet */
5985 const char *priority_s = argv[2];
5986 const char *tun_id_s = argv[3];
5987 const char *in_port_s = argv[4];
5988 const char *packet_s = argv[5];
5989 uint16_t in_port = ofp_port_to_odp_port(atoi(in_port_s));
5990 ovs_be64 tun_id = htonll(strtoull(tun_id_s, NULL, 0));
5991 uint32_t priority = atoi(priority_s);
5994 msg = eth_from_hex(packet_s, &packet);
5996 unixctl_command_reply(conn, 501, msg);
6000 ds_put_cstr(&result, "Packet: ");
6001 s = ofp_packet_to_string(packet->data, packet->size);
6002 ds_put_cstr(&result, s);
6005 flow_extract(packet, priority, tun_id, in_port, &flow);
6006 initial_tci = flow.vlan_tci;
6008 unixctl_command_reply(conn, 501, "Bad command syntax");
6012 ds_put_cstr(&result, "Flow: ");
6013 flow_format(&result, &flow);
6014 ds_put_char(&result, '\n');
6016 rule = rule_dpif_lookup(ofproto, &flow, 0);
6017 trace_format_rule(&result, 0, 0, rule);
6019 struct ofproto_trace trace;
6020 struct ofpbuf *odp_actions;
6022 trace.result = &result;
6024 action_xlate_ctx_init(&trace.ctx, ofproto, &flow, initial_tci,
6025 rule->up.flow_cookie, packet);
6026 trace.ctx.resubmit_hook = trace_resubmit;
6027 odp_actions = xlate_actions(&trace.ctx,
6028 rule->up.actions, rule->up.n_actions);
6030 ds_put_char(&result, '\n');
6031 trace_format_flow(&result, 0, "Final flow", &trace);
6032 ds_put_cstr(&result, "Datapath actions: ");
6033 format_odp_actions(&result, odp_actions->data, odp_actions->size);
6034 ofpbuf_delete(odp_actions);
6036 if (!trace.ctx.may_set_up_flow) {
6038 ds_put_cstr(&result, "\nThis flow is not cachable.");
6040 ds_put_cstr(&result, "\nThe datapath actions are incomplete--"
6041 "for complete actions, please supply a packet.");
6046 unixctl_command_reply(conn, 200, ds_cstr(&result));
6049 ds_destroy(&result);
6050 ofpbuf_delete(packet);
6051 ofpbuf_uninit(&odp_key);
6055 ofproto_dpif_clog(struct unixctl_conn *conn OVS_UNUSED, int argc OVS_UNUSED,
6056 const char *argv[] OVS_UNUSED, void *aux OVS_UNUSED)
6059 unixctl_command_reply(conn, 200, NULL);
6063 ofproto_dpif_unclog(struct unixctl_conn *conn OVS_UNUSED, int argc OVS_UNUSED,
6064 const char *argv[] OVS_UNUSED, void *aux OVS_UNUSED)
6067 unixctl_command_reply(conn, 200, NULL);
6070 /* Runs a self-check of flow translations in 'ofproto'. Appends a message to
6071 * 'reply' describing the results. */
6073 ofproto_dpif_self_check__(struct ofproto_dpif *ofproto, struct ds *reply)
6075 struct facet *facet;
6079 HMAP_FOR_EACH (facet, hmap_node, &ofproto->facets) {
6080 if (!facet_check_consistency(facet)) {
6085 ofproto->need_revalidate = true;
6089 ds_put_format(reply, "%s: self-check failed (%d errors)\n",
6090 ofproto->up.name, errors);
6092 ds_put_format(reply, "%s: self-check passed\n", ofproto->up.name);
6097 ofproto_dpif_self_check(struct unixctl_conn *conn,
6098 int argc, const char *argv[], void *aux OVS_UNUSED)
6100 struct ds reply = DS_EMPTY_INITIALIZER;
6101 struct ofproto_dpif *ofproto;
6104 ofproto = ofproto_dpif_lookup(argv[1]);
6106 unixctl_command_reply(conn, 501, "Unknown ofproto (use "
6107 "ofproto/list for help)");
6110 ofproto_dpif_self_check__(ofproto, &reply);
6112 HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
6113 ofproto_dpif_self_check__(ofproto, &reply);
6117 unixctl_command_reply(conn, 200, ds_cstr(&reply));
6122 ofproto_dpif_unixctl_init(void)
6124 static bool registered;
6130 unixctl_command_register(
6132 "bridge {tun_id in_port packet | odp_flow [-generate]}",
6133 2, 5, ofproto_unixctl_trace, NULL);
6134 unixctl_command_register("fdb/flush", "[bridge]", 0, 1,
6135 ofproto_unixctl_fdb_flush, NULL);
6136 unixctl_command_register("fdb/show", "bridge", 1, 1,
6137 ofproto_unixctl_fdb_show, NULL);
6138 unixctl_command_register("ofproto/clog", "", 0, 0,
6139 ofproto_dpif_clog, NULL);
6140 unixctl_command_register("ofproto/unclog", "", 0, 0,
6141 ofproto_dpif_unclog, NULL);
6142 unixctl_command_register("ofproto/self-check", "[bridge]", 0, 1,
6143 ofproto_dpif_self_check, NULL);
6146 /* Linux VLAN device support (e.g. "eth0.10" for VLAN 10.)
6148 * This is deprecated. It is only for compatibility with broken device drivers
6149 * in old versions of Linux that do not properly support VLANs when VLAN
6150 * devices are not used. When broken device drivers are no longer in
6151 * widespread use, we will delete these interfaces. */
6154 set_realdev(struct ofport *ofport_, uint16_t realdev_ofp_port, int vid)
6156 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport_->ofproto);
6157 struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
6159 if (realdev_ofp_port == ofport->realdev_ofp_port
6160 && vid == ofport->vlandev_vid) {
6164 ofproto->need_revalidate = true;
6166 if (ofport->realdev_ofp_port) {
6169 if (realdev_ofp_port && ofport->bundle) {
6170 /* vlandevs are enslaved to their realdevs, so they are not allowed to
6171 * themselves be part of a bundle. */
6172 bundle_set(ofport->up.ofproto, ofport->bundle, NULL);
6175 ofport->realdev_ofp_port = realdev_ofp_port;
6176 ofport->vlandev_vid = vid;
6178 if (realdev_ofp_port) {
6179 vsp_add(ofport, realdev_ofp_port, vid);
6186 hash_realdev_vid(uint16_t realdev_ofp_port, int vid)
6188 return hash_2words(realdev_ofp_port, vid);
6192 vsp_realdev_to_vlandev(const struct ofproto_dpif *ofproto,
6193 uint32_t realdev_odp_port, ovs_be16 vlan_tci)
6195 if (!hmap_is_empty(&ofproto->realdev_vid_map)) {
6196 uint16_t realdev_ofp_port = odp_port_to_ofp_port(realdev_odp_port);
6197 int vid = vlan_tci_to_vid(vlan_tci);
6198 const struct vlan_splinter *vsp;
6200 HMAP_FOR_EACH_WITH_HASH (vsp, realdev_vid_node,
6201 hash_realdev_vid(realdev_ofp_port, vid),
6202 &ofproto->realdev_vid_map) {
6203 if (vsp->realdev_ofp_port == realdev_ofp_port
6204 && vsp->vid == vid) {
6205 return ofp_port_to_odp_port(vsp->vlandev_ofp_port);
6209 return realdev_odp_port;
6212 static struct vlan_splinter *
6213 vlandev_find(const struct ofproto_dpif *ofproto, uint16_t vlandev_ofp_port)
6215 struct vlan_splinter *vsp;
6217 HMAP_FOR_EACH_WITH_HASH (vsp, vlandev_node, hash_int(vlandev_ofp_port, 0),
6218 &ofproto->vlandev_map) {
6219 if (vsp->vlandev_ofp_port == vlandev_ofp_port) {
6228 vsp_vlandev_to_realdev(const struct ofproto_dpif *ofproto,
6229 uint16_t vlandev_ofp_port, int *vid)
6231 if (!hmap_is_empty(&ofproto->vlandev_map)) {
6232 const struct vlan_splinter *vsp;
6234 vsp = vlandev_find(ofproto, vlandev_ofp_port);
6239 return vsp->realdev_ofp_port;
6246 vsp_remove(struct ofport_dpif *port)
6248 struct ofproto_dpif *ofproto = ofproto_dpif_cast(port->up.ofproto);
6249 struct vlan_splinter *vsp;
6251 vsp = vlandev_find(ofproto, port->up.ofp_port);
6253 hmap_remove(&ofproto->vlandev_map, &vsp->vlandev_node);
6254 hmap_remove(&ofproto->realdev_vid_map, &vsp->realdev_vid_node);
6257 port->realdev_ofp_port = 0;
6259 VLOG_ERR("missing vlan device record");
6264 vsp_add(struct ofport_dpif *port, uint16_t realdev_ofp_port, int vid)
6266 struct ofproto_dpif *ofproto = ofproto_dpif_cast(port->up.ofproto);
6268 if (!vsp_vlandev_to_realdev(ofproto, port->up.ofp_port, NULL)
6269 && (vsp_realdev_to_vlandev(ofproto, realdev_ofp_port, htons(vid))
6270 == realdev_ofp_port)) {
6271 struct vlan_splinter *vsp;
6273 vsp = xmalloc(sizeof *vsp);
6274 hmap_insert(&ofproto->vlandev_map, &vsp->vlandev_node,
6275 hash_int(port->up.ofp_port, 0));
6276 hmap_insert(&ofproto->realdev_vid_map, &vsp->realdev_vid_node,
6277 hash_realdev_vid(realdev_ofp_port, vid));
6278 vsp->realdev_ofp_port = realdev_ofp_port;
6279 vsp->vlandev_ofp_port = port->up.ofp_port;
6282 port->realdev_ofp_port = realdev_ofp_port;
6284 VLOG_ERR("duplicate vlan device record");
6288 const struct ofproto_class ofproto_dpif_class = {
6317 port_is_lacp_current,
6318 NULL, /* rule_choose_table */
6325 rule_modify_actions,
6333 get_cfm_remote_mpids,
6337 get_stp_port_status,
6344 is_mirror_output_bundle,
6345 forward_bpdu_changed,