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 "multipath.h"
44 #include "ofp-print.h"
45 #include "ofproto-dpif-sflow.h"
46 #include "poll-loop.h"
48 #include "unaligned.h"
50 #include "vlan-bitmap.h"
53 VLOG_DEFINE_THIS_MODULE(ofproto_dpif);
55 COVERAGE_DEFINE(ofproto_dpif_ctlr_action);
56 COVERAGE_DEFINE(ofproto_dpif_expired);
57 COVERAGE_DEFINE(ofproto_dpif_no_packet_in);
58 COVERAGE_DEFINE(ofproto_dpif_xlate);
59 COVERAGE_DEFINE(facet_changed_rule);
60 COVERAGE_DEFINE(facet_invalidated);
61 COVERAGE_DEFINE(facet_revalidate);
62 COVERAGE_DEFINE(facet_unexpected);
64 /* Maximum depth of flow table recursion (due to resubmit actions) in a
65 * flow translation. */
66 #define MAX_RESUBMIT_RECURSION 32
68 /* Number of implemented OpenFlow tables. */
69 enum { N_TABLES = 255 };
70 BUILD_ASSERT_DECL(N_TABLES >= 1 && N_TABLES <= 255);
78 long long int used; /* Time last used; time created if not used. */
82 * - Do include packets and bytes from facets that have been deleted or
83 * whose own statistics have been folded into the rule.
85 * - Do include packets and bytes sent "by hand" that were accounted to
86 * the rule without any facet being involved (this is a rare corner
87 * case in rule_execute()).
89 * - Do not include packet or bytes that can be obtained from any facet's
90 * packet_count or byte_count member or that can be obtained from the
91 * datapath by, e.g., dpif_flow_get() for any subfacet.
93 uint64_t packet_count; /* Number of packets received. */
94 uint64_t byte_count; /* Number of bytes received. */
96 tag_type tag; /* Caches rule_calculate_tag() result. */
98 struct list facets; /* List of "struct facet"s. */
101 static struct rule_dpif *rule_dpif_cast(const struct rule *rule)
103 return rule ? CONTAINER_OF(rule, struct rule_dpif, up) : NULL;
106 static struct rule_dpif *rule_dpif_lookup(struct ofproto_dpif *,
107 const struct flow *, uint8_t table);
109 static void flow_push_stats(const struct rule_dpif *, const struct flow *,
110 uint64_t packets, uint64_t bytes,
113 static uint32_t rule_calculate_tag(const struct flow *,
114 const struct flow_wildcards *,
116 static void rule_invalidate(const struct rule_dpif *);
118 #define MAX_MIRRORS 32
119 typedef uint32_t mirror_mask_t;
120 #define MIRROR_MASK_C(X) UINT32_C(X)
121 BUILD_ASSERT_DECL(sizeof(mirror_mask_t) * CHAR_BIT >= MAX_MIRRORS);
123 struct ofproto_dpif *ofproto; /* Owning ofproto. */
124 size_t idx; /* In ofproto's "mirrors" array. */
125 void *aux; /* Key supplied by ofproto's client. */
126 char *name; /* Identifier for log messages. */
128 /* Selection criteria. */
129 struct hmapx srcs; /* Contains "struct ofbundle *"s. */
130 struct hmapx dsts; /* Contains "struct ofbundle *"s. */
131 unsigned long *vlans; /* Bitmap of chosen VLANs, NULL selects all. */
133 /* Output (exactly one of out == NULL and out_vlan == -1 is true). */
134 struct ofbundle *out; /* Output port or NULL. */
135 int out_vlan; /* Output VLAN or -1. */
136 mirror_mask_t dup_mirrors; /* Bitmap of mirrors with the same output. */
139 int64_t packet_count; /* Number of packets sent. */
140 int64_t byte_count; /* Number of bytes sent. */
143 static void mirror_destroy(struct ofmirror *);
144 static void update_mirror_stats(struct ofproto_dpif *ofproto,
145 mirror_mask_t mirrors,
146 uint64_t packets, uint64_t bytes);
149 struct ofproto_dpif *ofproto; /* Owning ofproto. */
150 struct hmap_node hmap_node; /* In struct ofproto's "bundles" hmap. */
151 void *aux; /* Key supplied by ofproto's client. */
152 char *name; /* Identifier for log messages. */
155 struct list ports; /* Contains "struct ofport"s. */
156 enum port_vlan_mode vlan_mode; /* VLAN mode */
157 int vlan; /* -1=trunk port, else a 12-bit VLAN ID. */
158 unsigned long *trunks; /* Bitmap of trunked VLANs, if 'vlan' == -1.
159 * NULL if all VLANs are trunked. */
160 struct lacp *lacp; /* LACP if LACP is enabled, otherwise NULL. */
161 struct bond *bond; /* Nonnull iff more than one port. */
162 bool use_priority_tags; /* Use 802.1p tag for frames in VLAN 0? */
165 bool floodable; /* True if no port has OFPPC_NO_FLOOD set. */
167 /* Port mirroring info. */
168 mirror_mask_t src_mirrors; /* Mirrors triggered when packet received. */
169 mirror_mask_t dst_mirrors; /* Mirrors triggered when packet sent. */
170 mirror_mask_t mirror_out; /* Mirrors that output to this bundle. */
173 static void bundle_remove(struct ofport *);
174 static void bundle_update(struct ofbundle *);
175 static void bundle_destroy(struct ofbundle *);
176 static void bundle_del_port(struct ofport_dpif *);
177 static void bundle_run(struct ofbundle *);
178 static void bundle_wait(struct ofbundle *);
179 static struct ofbundle *lookup_input_bundle(struct ofproto_dpif *,
180 uint16_t in_port, bool warn);
182 /* A controller may use OFPP_NONE as the ingress port to indicate that
183 * it did not arrive on a "real" port. 'ofpp_none_bundle' exists for
184 * when an input bundle is needed for validation (e.g., mirroring or
185 * OFPP_NORMAL processing). It is not connected to an 'ofproto' or have
186 * any 'port' structs, so care must be taken when dealing with it. */
187 static struct ofbundle ofpp_none_bundle = {
189 .vlan_mode = PORT_VLAN_TRUNK
192 static void stp_run(struct ofproto_dpif *ofproto);
193 static void stp_wait(struct ofproto_dpif *ofproto);
195 static bool ofbundle_includes_vlan(const struct ofbundle *, uint16_t vlan);
197 struct action_xlate_ctx {
198 /* action_xlate_ctx_init() initializes these members. */
201 struct ofproto_dpif *ofproto;
203 /* Flow to which the OpenFlow actions apply. xlate_actions() will modify
204 * this flow when actions change header fields. */
207 /* The packet corresponding to 'flow', or a null pointer if we are
208 * revalidating without a packet to refer to. */
209 const struct ofpbuf *packet;
211 /* Should OFPP_NORMAL MAC learning and NXAST_LEARN actions execute? We
212 * want to execute them if we are actually processing a packet, or if we
213 * are accounting for packets that the datapath has processed, but not if
214 * we are just revalidating. */
217 /* Cookie of the currently matching rule, or 0. */
220 /* If nonnull, called just before executing a resubmit action.
222 * This is normally null so the client has to set it manually after
223 * calling action_xlate_ctx_init(). */
224 void (*resubmit_hook)(struct action_xlate_ctx *, struct rule_dpif *);
226 /* xlate_actions() initializes and uses these members. The client might want
227 * to look at them after it returns. */
229 struct ofpbuf *odp_actions; /* Datapath actions. */
230 tag_type tags; /* Tags associated with actions. */
231 bool may_set_up_flow; /* True ordinarily; false if the actions must
232 * be reassessed for every packet. */
233 bool has_learn; /* Actions include NXAST_LEARN? */
234 bool has_normal; /* Actions output to OFPP_NORMAL? */
235 uint16_t nf_output_iface; /* Output interface index for NetFlow. */
236 mirror_mask_t mirrors; /* Bitmap of associated mirrors. */
238 /* xlate_actions() initializes and uses these members, but the client has no
239 * reason to look at them. */
241 int recurse; /* Recursion level, via xlate_table_action. */
242 struct flow base_flow; /* Flow at the last commit. */
243 uint32_t orig_skb_priority; /* Priority when packet arrived. */
244 uint8_t table_id; /* OpenFlow table ID where flow was found. */
245 uint32_t sflow_n_outputs; /* Number of output ports. */
246 uint16_t sflow_odp_port; /* Output port for composing sFlow action. */
247 uint16_t user_cookie_offset;/* Used for user_action_cookie fixup. */
248 bool exit; /* No further actions should be processed. */
251 static void action_xlate_ctx_init(struct action_xlate_ctx *,
252 struct ofproto_dpif *, const struct flow *,
253 ovs_be16 initial_tci, ovs_be64 cookie,
254 const struct ofpbuf *);
255 static struct ofpbuf *xlate_actions(struct action_xlate_ctx *,
256 const union ofp_action *in, size_t n_in);
258 /* An exact-match instantiation of an OpenFlow flow.
260 * A facet associates a "struct flow", which represents the Open vSwitch
261 * userspace idea of an exact-match flow, with one or more subfacets. Each
262 * subfacet tracks the datapath's idea of the exact-match flow equivalent to
263 * the facet. When the kernel module (or other dpif implementation) and Open
264 * vSwitch userspace agree on the definition of a flow key, there is exactly
265 * one subfacet per facet. If the dpif implementation supports more-specific
266 * flow matching than userspace, however, a facet can have more than one
267 * subfacet, each of which corresponds to some distinction in flow that
268 * userspace simply doesn't understand.
270 * Flow expiration works in terms of subfacets, so a facet must have at least
271 * one subfacet or it will never expire, leaking memory. */
274 struct hmap_node hmap_node; /* In owning ofproto's 'facets' hmap. */
275 struct list list_node; /* In owning rule's 'facets' list. */
276 struct rule_dpif *rule; /* Owning rule. */
279 struct list subfacets;
280 long long int used; /* Time last used; time created if not used. */
287 * - Do include packets and bytes sent "by hand", e.g. with
290 * - Do include packets and bytes that were obtained from the datapath
291 * when a subfacet's statistics were reset (e.g. dpif_flow_put() with
292 * DPIF_FP_ZERO_STATS).
294 * - Do not include packets or bytes that can be obtained from the
295 * datapath for any existing subfacet.
297 uint64_t packet_count; /* Number of packets received. */
298 uint64_t byte_count; /* Number of bytes received. */
300 /* Resubmit statistics. */
301 uint64_t prev_packet_count; /* Number of packets from last stats push. */
302 uint64_t prev_byte_count; /* Number of bytes from last stats push. */
303 long long int prev_used; /* Used time from last stats push. */
306 uint64_t accounted_bytes; /* Bytes processed by facet_account(). */
307 struct netflow_flow nf_flow; /* Per-flow NetFlow tracking data. */
309 /* Properties of datapath actions.
311 * Every subfacet has its own actions because actions can differ slightly
312 * between splintered and non-splintered subfacets due to the VLAN tag
313 * being initially different (present vs. absent). All of them have these
314 * properties in common so we just store one copy of them here. */
315 bool may_install; /* Reassess actions for every packet? */
316 bool has_learn; /* Actions include NXAST_LEARN? */
317 bool has_normal; /* Actions output to OFPP_NORMAL? */
318 tag_type tags; /* Tags that would require revalidation. */
319 mirror_mask_t mirrors; /* Bitmap of dependent mirrors. */
322 static struct facet *facet_create(struct rule_dpif *, const struct flow *);
323 static void facet_remove(struct facet *);
324 static void facet_free(struct facet *);
326 static struct facet *facet_find(struct ofproto_dpif *, const struct flow *);
327 static struct facet *facet_lookup_valid(struct ofproto_dpif *,
328 const struct flow *);
329 static bool facet_revalidate(struct facet *);
330 static bool facet_check_consistency(struct facet *);
332 static void facet_flush_stats(struct facet *);
334 static void facet_update_time(struct facet *, long long int used);
335 static void facet_reset_counters(struct facet *);
336 static void facet_push_stats(struct facet *);
337 static void facet_account(struct facet *);
339 static bool facet_is_controller_flow(struct facet *);
341 /* A dpif flow and actions associated with a facet.
343 * See also the large comment on struct facet. */
346 struct hmap_node hmap_node; /* In struct ofproto_dpif 'subfacets' list. */
347 struct list list_node; /* In struct facet's 'facets' list. */
348 struct facet *facet; /* Owning facet. */
352 * To save memory in the common case, 'key' is NULL if 'key_fitness' is
353 * ODP_FIT_PERFECT, that is, odp_flow_key_from_flow() can accurately
354 * regenerate the ODP flow key from ->facet->flow. */
355 enum odp_key_fitness key_fitness;
359 long long int used; /* Time last used; time created if not used. */
361 uint64_t dp_packet_count; /* Last known packet count in the datapath. */
362 uint64_t dp_byte_count; /* Last known byte count in the datapath. */
366 * These should be essentially identical for every subfacet in a facet, but
367 * may differ in trivial ways due to VLAN splinters. */
368 size_t actions_len; /* Number of bytes in actions[]. */
369 struct nlattr *actions; /* Datapath actions. */
371 bool installed; /* Installed in datapath? */
373 /* This value is normally the same as ->facet->flow.vlan_tci. Only VLAN
374 * splinters can cause it to differ. This value should be removed when
375 * the VLAN splinters feature is no longer needed. */
376 ovs_be16 initial_tci; /* Initial VLAN TCI value. */
379 static struct subfacet *subfacet_create(struct facet *, enum odp_key_fitness,
380 const struct nlattr *key,
381 size_t key_len, ovs_be16 initial_tci);
382 static struct subfacet *subfacet_find(struct ofproto_dpif *,
383 const struct nlattr *key, size_t key_len);
384 static void subfacet_destroy(struct subfacet *);
385 static void subfacet_destroy__(struct subfacet *);
386 static void subfacet_get_key(struct subfacet *, struct odputil_keybuf *,
388 static void subfacet_reset_dp_stats(struct subfacet *,
389 struct dpif_flow_stats *);
390 static void subfacet_update_time(struct subfacet *, long long int used);
391 static void subfacet_update_stats(struct subfacet *,
392 const struct dpif_flow_stats *);
393 static void subfacet_make_actions(struct subfacet *,
394 const struct ofpbuf *packet);
395 static int subfacet_install(struct subfacet *,
396 const struct nlattr *actions, size_t actions_len,
397 struct dpif_flow_stats *);
398 static void subfacet_uninstall(struct subfacet *);
404 struct ofbundle *bundle; /* Bundle that contains this port, if any. */
405 struct list bundle_node; /* In struct ofbundle's "ports" list. */
406 struct cfm *cfm; /* Connectivity Fault Management, if any. */
407 tag_type tag; /* Tag associated with this port. */
408 uint32_t bond_stable_id; /* stable_id to use as bond slave, or 0. */
409 bool may_enable; /* May be enabled in bonds. */
412 struct stp_port *stp_port; /* Spanning Tree Protocol, if any. */
413 enum stp_state stp_state; /* Always STP_DISABLED if STP not in use. */
414 long long int stp_state_entered;
416 struct hmap priorities; /* Map of attached 'priority_to_dscp's. */
418 /* Linux VLAN device support (e.g. "eth0.10" for VLAN 10.)
420 * This is deprecated. It is only for compatibility with broken device
421 * drivers in old versions of Linux that do not properly support VLANs when
422 * VLAN devices are not used. When broken device drivers are no longer in
423 * widespread use, we will delete these interfaces. */
424 uint16_t realdev_ofp_port;
428 /* Node in 'ofport_dpif''s 'priorities' map. Used to maintain a map from
429 * 'priority' (the datapath's term for QoS queue) to the dscp bits which all
430 * traffic egressing the 'ofport' with that priority should be marked with. */
431 struct priority_to_dscp {
432 struct hmap_node hmap_node; /* Node in 'ofport_dpif''s 'priorities' map. */
433 uint32_t priority; /* Priority of this queue (see struct flow). */
435 uint8_t dscp; /* DSCP bits to mark outgoing traffic with. */
438 /* Linux VLAN device support (e.g. "eth0.10" for VLAN 10.)
440 * This is deprecated. It is only for compatibility with broken device drivers
441 * in old versions of Linux that do not properly support VLANs when VLAN
442 * devices are not used. When broken device drivers are no longer in
443 * widespread use, we will delete these interfaces. */
444 struct vlan_splinter {
445 struct hmap_node realdev_vid_node;
446 struct hmap_node vlandev_node;
447 uint16_t realdev_ofp_port;
448 uint16_t vlandev_ofp_port;
452 static uint32_t vsp_realdev_to_vlandev(const struct ofproto_dpif *,
453 uint32_t realdev, ovs_be16 vlan_tci);
454 static uint16_t vsp_vlandev_to_realdev(const struct ofproto_dpif *,
455 uint16_t vlandev, int *vid);
456 static void vsp_remove(struct ofport_dpif *);
457 static void vsp_add(struct ofport_dpif *, uint16_t realdev_ofp_port, int vid);
459 static struct ofport_dpif *
460 ofport_dpif_cast(const struct ofport *ofport)
462 assert(ofport->ofproto->ofproto_class == &ofproto_dpif_class);
463 return ofport ? CONTAINER_OF(ofport, struct ofport_dpif, up) : NULL;
466 static void port_run(struct ofport_dpif *);
467 static void port_wait(struct ofport_dpif *);
468 static int set_cfm(struct ofport *, const struct cfm_settings *);
469 static void ofport_clear_priorities(struct ofport_dpif *);
471 struct dpif_completion {
472 struct list list_node;
473 struct ofoperation *op;
476 /* Extra information about a classifier table.
477 * Currently used just for optimized flow revalidation. */
479 /* If either of these is nonnull, then this table has a form that allows
480 * flows to be tagged to avoid revalidating most flows for the most common
481 * kinds of flow table changes. */
482 struct cls_table *catchall_table; /* Table that wildcards all fields. */
483 struct cls_table *other_table; /* Table with any other wildcard set. */
484 uint32_t basis; /* Keeps each table's tags separate. */
487 struct ofproto_dpif {
488 struct hmap_node all_ofproto_dpifs_node; /* In 'all_ofproto_dpifs'. */
497 struct netflow *netflow;
498 struct dpif_sflow *sflow;
499 struct hmap bundles; /* Contains "struct ofbundle"s. */
500 struct mac_learning *ml;
501 struct ofmirror *mirrors[MAX_MIRRORS];
502 bool has_bonded_bundles;
505 struct timer next_expiration;
509 struct hmap subfacets;
512 struct table_dpif tables[N_TABLES];
513 bool need_revalidate;
514 struct tag_set revalidate_set;
516 /* Support for debugging async flow mods. */
517 struct list completions;
519 bool has_bundle_action; /* True when the first bundle action appears. */
520 struct netdev_stats stats; /* To account packets generated and consumed in
525 long long int stp_last_tick;
527 /* VLAN splinters. */
528 struct hmap realdev_vid_map; /* (realdev,vid) -> vlandev. */
529 struct hmap vlandev_map; /* vlandev -> (realdev,vid). */
532 /* Defer flow mod completion until "ovs-appctl ofproto/unclog"? (Useful only
533 * for debugging the asynchronous flow_mod implementation.) */
536 /* All existing ofproto_dpif instances, indexed by ->up.name. */
537 static struct hmap all_ofproto_dpifs = HMAP_INITIALIZER(&all_ofproto_dpifs);
539 static void ofproto_dpif_unixctl_init(void);
541 static struct ofproto_dpif *
542 ofproto_dpif_cast(const struct ofproto *ofproto)
544 assert(ofproto->ofproto_class == &ofproto_dpif_class);
545 return CONTAINER_OF(ofproto, struct ofproto_dpif, up);
548 static struct ofport_dpif *get_ofp_port(struct ofproto_dpif *,
550 static struct ofport_dpif *get_odp_port(struct ofproto_dpif *,
553 /* Packet processing. */
554 static void update_learning_table(struct ofproto_dpif *,
555 const struct flow *, int vlan,
558 #define FLOW_MISS_MAX_BATCH 50
559 static int handle_upcalls(struct ofproto_dpif *, unsigned int max_batch);
561 /* Flow expiration. */
562 static int expire(struct ofproto_dpif *);
565 static void send_netflow_active_timeouts(struct ofproto_dpif *);
568 static int send_packet(const struct ofport_dpif *, struct ofpbuf *packet);
570 compose_sflow_action(const struct ofproto_dpif *, struct ofpbuf *odp_actions,
571 const struct flow *, uint32_t odp_port);
572 static void add_mirror_actions(struct action_xlate_ctx *ctx,
573 const struct flow *flow);
574 /* Global variables. */
575 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
577 /* Factory functions. */
580 enumerate_types(struct sset *types)
582 dp_enumerate_types(types);
586 enumerate_names(const char *type, struct sset *names)
588 return dp_enumerate_names(type, names);
592 del(const char *type, const char *name)
597 error = dpif_open(name, type, &dpif);
599 error = dpif_delete(dpif);
605 /* Basic life-cycle. */
607 static struct ofproto *
610 struct ofproto_dpif *ofproto = xmalloc(sizeof *ofproto);
615 dealloc(struct ofproto *ofproto_)
617 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
622 construct(struct ofproto *ofproto_, int *n_tablesp)
624 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
625 const char *name = ofproto->up.name;
629 error = dpif_create_and_open(name, ofproto->up.type, &ofproto->dpif);
631 VLOG_ERR("failed to open datapath %s: %s", name, strerror(error));
635 ofproto->max_ports = dpif_get_max_ports(ofproto->dpif);
636 ofproto->n_matches = 0;
638 dpif_flow_flush(ofproto->dpif);
639 dpif_recv_purge(ofproto->dpif);
641 error = dpif_recv_set(ofproto->dpif, true);
643 VLOG_ERR("failed to listen on datapath %s: %s", name, strerror(error));
644 dpif_close(ofproto->dpif);
648 ofproto->netflow = NULL;
649 ofproto->sflow = NULL;
651 hmap_init(&ofproto->bundles);
652 ofproto->ml = mac_learning_create();
653 for (i = 0; i < MAX_MIRRORS; i++) {
654 ofproto->mirrors[i] = NULL;
656 ofproto->has_bonded_bundles = false;
658 timer_set_duration(&ofproto->next_expiration, 1000);
660 hmap_init(&ofproto->facets);
661 hmap_init(&ofproto->subfacets);
663 for (i = 0; i < N_TABLES; i++) {
664 struct table_dpif *table = &ofproto->tables[i];
666 table->catchall_table = NULL;
667 table->other_table = NULL;
668 table->basis = random_uint32();
670 ofproto->need_revalidate = false;
671 tag_set_init(&ofproto->revalidate_set);
673 list_init(&ofproto->completions);
675 ofproto_dpif_unixctl_init();
677 ofproto->has_bundle_action = false;
679 hmap_init(&ofproto->vlandev_map);
680 hmap_init(&ofproto->realdev_vid_map);
682 hmap_insert(&all_ofproto_dpifs, &ofproto->all_ofproto_dpifs_node,
683 hash_string(ofproto->up.name, 0));
685 *n_tablesp = N_TABLES;
686 memset(&ofproto->stats, 0, sizeof ofproto->stats);
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 classifier *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, 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);
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);
1303 stp_wait(struct ofproto_dpif *ofproto)
1306 poll_timer_wait(1000);
1310 /* Returns true if STP should process 'flow'. */
1312 stp_should_process_flow(const struct flow *flow)
1314 return eth_addr_equals(flow->dl_dst, eth_addr_stp);
1318 stp_process_packet(const struct ofport_dpif *ofport,
1319 const struct ofpbuf *packet)
1321 struct ofpbuf payload = *packet;
1322 struct eth_header *eth = payload.data;
1323 struct stp_port *sp = ofport->stp_port;
1325 /* Sink packets on ports that have STP disabled when the bridge has
1327 if (!sp || stp_port_get_state(sp) == STP_DISABLED) {
1331 /* Trim off padding on payload. */
1332 if (payload.size > ntohs(eth->eth_type) + ETH_HEADER_LEN) {
1333 payload.size = ntohs(eth->eth_type) + ETH_HEADER_LEN;
1336 if (ofpbuf_try_pull(&payload, ETH_HEADER_LEN + LLC_HEADER_LEN)) {
1337 stp_received_bpdu(sp, payload.data, payload.size);
1341 static struct priority_to_dscp *
1342 get_priority(const struct ofport_dpif *ofport, uint32_t priority)
1344 struct priority_to_dscp *pdscp;
1347 hash = hash_int(priority, 0);
1348 HMAP_FOR_EACH_IN_BUCKET (pdscp, hmap_node, hash, &ofport->priorities) {
1349 if (pdscp->priority == priority) {
1357 ofport_clear_priorities(struct ofport_dpif *ofport)
1359 struct priority_to_dscp *pdscp, *next;
1361 HMAP_FOR_EACH_SAFE (pdscp, next, hmap_node, &ofport->priorities) {
1362 hmap_remove(&ofport->priorities, &pdscp->hmap_node);
1368 set_queues(struct ofport *ofport_,
1369 const struct ofproto_port_queue *qdscp_list,
1372 struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
1373 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport->up.ofproto);
1374 struct hmap new = HMAP_INITIALIZER(&new);
1377 for (i = 0; i < n_qdscp; i++) {
1378 struct priority_to_dscp *pdscp;
1382 dscp = (qdscp_list[i].dscp << 2) & IP_DSCP_MASK;
1383 if (dpif_queue_to_priority(ofproto->dpif, qdscp_list[i].queue,
1388 pdscp = get_priority(ofport, priority);
1390 hmap_remove(&ofport->priorities, &pdscp->hmap_node);
1392 pdscp = xmalloc(sizeof *pdscp);
1393 pdscp->priority = priority;
1395 ofproto->need_revalidate = true;
1398 if (pdscp->dscp != dscp) {
1400 ofproto->need_revalidate = true;
1403 hmap_insert(&new, &pdscp->hmap_node, hash_int(pdscp->priority, 0));
1406 if (!hmap_is_empty(&ofport->priorities)) {
1407 ofport_clear_priorities(ofport);
1408 ofproto->need_revalidate = true;
1411 hmap_swap(&new, &ofport->priorities);
1419 /* Expires all MAC learning entries associated with 'bundle' and forces its
1420 * ofproto to revalidate every flow.
1422 * Normally MAC learning entries are removed only from the ofproto associated
1423 * with 'bundle', but if 'all_ofprotos' is true, then the MAC learning entries
1424 * are removed from every ofproto. When patch ports and SLB bonds are in use
1425 * and a VM migration happens and the gratuitous ARPs are somehow lost, this
1426 * avoids a MAC_ENTRY_IDLE_TIME delay before the migrated VM can communicate
1427 * with the host from which it migrated. */
1429 bundle_flush_macs(struct ofbundle *bundle, bool all_ofprotos)
1431 struct ofproto_dpif *ofproto = bundle->ofproto;
1432 struct mac_learning *ml = ofproto->ml;
1433 struct mac_entry *mac, *next_mac;
1435 ofproto->need_revalidate = true;
1436 LIST_FOR_EACH_SAFE (mac, next_mac, lru_node, &ml->lrus) {
1437 if (mac->port.p == bundle) {
1439 struct ofproto_dpif *o;
1441 HMAP_FOR_EACH (o, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
1443 struct mac_entry *e;
1445 e = mac_learning_lookup(o->ml, mac->mac, mac->vlan,
1448 tag_set_add(&o->revalidate_set, e->tag);
1449 mac_learning_expire(o->ml, e);
1455 mac_learning_expire(ml, mac);
1460 static struct ofbundle *
1461 bundle_lookup(const struct ofproto_dpif *ofproto, void *aux)
1463 struct ofbundle *bundle;
1465 HMAP_FOR_EACH_IN_BUCKET (bundle, hmap_node, hash_pointer(aux, 0),
1466 &ofproto->bundles) {
1467 if (bundle->aux == aux) {
1474 /* Looks up each of the 'n_auxes' pointers in 'auxes' as bundles and adds the
1475 * ones that are found to 'bundles'. */
1477 bundle_lookup_multiple(struct ofproto_dpif *ofproto,
1478 void **auxes, size_t n_auxes,
1479 struct hmapx *bundles)
1483 hmapx_init(bundles);
1484 for (i = 0; i < n_auxes; i++) {
1485 struct ofbundle *bundle = bundle_lookup(ofproto, auxes[i]);
1487 hmapx_add(bundles, bundle);
1493 bundle_update(struct ofbundle *bundle)
1495 struct ofport_dpif *port;
1497 bundle->floodable = true;
1498 LIST_FOR_EACH (port, bundle_node, &bundle->ports) {
1499 if (port->up.opp.config & htonl(OFPPC_NO_FLOOD)) {
1500 bundle->floodable = false;
1507 bundle_del_port(struct ofport_dpif *port)
1509 struct ofbundle *bundle = port->bundle;
1511 bundle->ofproto->need_revalidate = true;
1513 list_remove(&port->bundle_node);
1514 port->bundle = NULL;
1517 lacp_slave_unregister(bundle->lacp, port);
1520 bond_slave_unregister(bundle->bond, port);
1523 bundle_update(bundle);
1527 bundle_add_port(struct ofbundle *bundle, uint32_t ofp_port,
1528 struct lacp_slave_settings *lacp,
1529 uint32_t bond_stable_id)
1531 struct ofport_dpif *port;
1533 port = get_ofp_port(bundle->ofproto, ofp_port);
1538 if (port->bundle != bundle) {
1539 bundle->ofproto->need_revalidate = true;
1541 bundle_del_port(port);
1544 port->bundle = bundle;
1545 list_push_back(&bundle->ports, &port->bundle_node);
1546 if (port->up.opp.config & htonl(OFPPC_NO_FLOOD)) {
1547 bundle->floodable = false;
1551 port->bundle->ofproto->need_revalidate = true;
1552 lacp_slave_register(bundle->lacp, port, lacp);
1555 port->bond_stable_id = bond_stable_id;
1561 bundle_destroy(struct ofbundle *bundle)
1563 struct ofproto_dpif *ofproto;
1564 struct ofport_dpif *port, *next_port;
1571 ofproto = bundle->ofproto;
1572 for (i = 0; i < MAX_MIRRORS; i++) {
1573 struct ofmirror *m = ofproto->mirrors[i];
1575 if (m->out == bundle) {
1577 } else if (hmapx_find_and_delete(&m->srcs, bundle)
1578 || hmapx_find_and_delete(&m->dsts, bundle)) {
1579 ofproto->need_revalidate = true;
1584 LIST_FOR_EACH_SAFE (port, next_port, bundle_node, &bundle->ports) {
1585 bundle_del_port(port);
1588 bundle_flush_macs(bundle, true);
1589 hmap_remove(&ofproto->bundles, &bundle->hmap_node);
1591 free(bundle->trunks);
1592 lacp_destroy(bundle->lacp);
1593 bond_destroy(bundle->bond);
1598 bundle_set(struct ofproto *ofproto_, void *aux,
1599 const struct ofproto_bundle_settings *s)
1601 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1602 bool need_flush = false;
1603 struct ofport_dpif *port;
1604 struct ofbundle *bundle;
1605 unsigned long *trunks;
1611 bundle_destroy(bundle_lookup(ofproto, aux));
1615 assert(s->n_slaves == 1 || s->bond != NULL);
1616 assert((s->lacp != NULL) == (s->lacp_slaves != NULL));
1618 bundle = bundle_lookup(ofproto, aux);
1620 bundle = xmalloc(sizeof *bundle);
1622 bundle->ofproto = ofproto;
1623 hmap_insert(&ofproto->bundles, &bundle->hmap_node,
1624 hash_pointer(aux, 0));
1626 bundle->name = NULL;
1628 list_init(&bundle->ports);
1629 bundle->vlan_mode = PORT_VLAN_TRUNK;
1631 bundle->trunks = NULL;
1632 bundle->use_priority_tags = s->use_priority_tags;
1633 bundle->lacp = NULL;
1634 bundle->bond = NULL;
1636 bundle->floodable = true;
1638 bundle->src_mirrors = 0;
1639 bundle->dst_mirrors = 0;
1640 bundle->mirror_out = 0;
1643 if (!bundle->name || strcmp(s->name, bundle->name)) {
1645 bundle->name = xstrdup(s->name);
1650 if (!bundle->lacp) {
1651 ofproto->need_revalidate = true;
1652 bundle->lacp = lacp_create();
1654 lacp_configure(bundle->lacp, s->lacp);
1656 lacp_destroy(bundle->lacp);
1657 bundle->lacp = NULL;
1660 /* Update set of ports. */
1662 for (i = 0; i < s->n_slaves; i++) {
1663 if (!bundle_add_port(bundle, s->slaves[i],
1664 s->lacp ? &s->lacp_slaves[i] : NULL,
1665 s->bond_stable_ids ? s->bond_stable_ids[i] : 0)) {
1669 if (!ok || list_size(&bundle->ports) != s->n_slaves) {
1670 struct ofport_dpif *next_port;
1672 LIST_FOR_EACH_SAFE (port, next_port, bundle_node, &bundle->ports) {
1673 for (i = 0; i < s->n_slaves; i++) {
1674 if (s->slaves[i] == port->up.ofp_port) {
1679 bundle_del_port(port);
1683 assert(list_size(&bundle->ports) <= s->n_slaves);
1685 if (list_is_empty(&bundle->ports)) {
1686 bundle_destroy(bundle);
1690 /* Set VLAN tagging mode */
1691 if (s->vlan_mode != bundle->vlan_mode
1692 || s->use_priority_tags != bundle->use_priority_tags) {
1693 bundle->vlan_mode = s->vlan_mode;
1694 bundle->use_priority_tags = s->use_priority_tags;
1699 vlan = (s->vlan_mode == PORT_VLAN_TRUNK ? -1
1700 : s->vlan >= 0 && s->vlan <= 4095 ? s->vlan
1702 if (vlan != bundle->vlan) {
1703 bundle->vlan = vlan;
1707 /* Get trunked VLANs. */
1708 switch (s->vlan_mode) {
1709 case PORT_VLAN_ACCESS:
1713 case PORT_VLAN_TRUNK:
1714 trunks = (unsigned long *) s->trunks;
1717 case PORT_VLAN_NATIVE_UNTAGGED:
1718 case PORT_VLAN_NATIVE_TAGGED:
1719 if (vlan != 0 && (!s->trunks
1720 || !bitmap_is_set(s->trunks, vlan)
1721 || bitmap_is_set(s->trunks, 0))) {
1722 /* Force trunking the native VLAN and prohibit trunking VLAN 0. */
1724 trunks = bitmap_clone(s->trunks, 4096);
1726 trunks = bitmap_allocate1(4096);
1728 bitmap_set1(trunks, vlan);
1729 bitmap_set0(trunks, 0);
1731 trunks = (unsigned long *) s->trunks;
1738 if (!vlan_bitmap_equal(trunks, bundle->trunks)) {
1739 free(bundle->trunks);
1740 if (trunks == s->trunks) {
1741 bundle->trunks = vlan_bitmap_clone(trunks);
1743 bundle->trunks = trunks;
1748 if (trunks != s->trunks) {
1753 if (!list_is_short(&bundle->ports)) {
1754 bundle->ofproto->has_bonded_bundles = true;
1756 if (bond_reconfigure(bundle->bond, s->bond)) {
1757 ofproto->need_revalidate = true;
1760 bundle->bond = bond_create(s->bond);
1761 ofproto->need_revalidate = true;
1764 LIST_FOR_EACH (port, bundle_node, &bundle->ports) {
1765 bond_slave_register(bundle->bond, port, port->bond_stable_id,
1769 bond_destroy(bundle->bond);
1770 bundle->bond = NULL;
1773 /* If we changed something that would affect MAC learning, un-learn
1774 * everything on this port and force flow revalidation. */
1776 bundle_flush_macs(bundle, false);
1783 bundle_remove(struct ofport *port_)
1785 struct ofport_dpif *port = ofport_dpif_cast(port_);
1786 struct ofbundle *bundle = port->bundle;
1789 bundle_del_port(port);
1790 if (list_is_empty(&bundle->ports)) {
1791 bundle_destroy(bundle);
1792 } else if (list_is_short(&bundle->ports)) {
1793 bond_destroy(bundle->bond);
1794 bundle->bond = NULL;
1800 send_pdu_cb(void *port_, const void *pdu, size_t pdu_size)
1802 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 10);
1803 struct ofport_dpif *port = port_;
1804 uint8_t ea[ETH_ADDR_LEN];
1807 error = netdev_get_etheraddr(port->up.netdev, ea);
1809 struct ofpbuf packet;
1812 ofpbuf_init(&packet, 0);
1813 packet_pdu = eth_compose(&packet, eth_addr_lacp, ea, ETH_TYPE_LACP,
1815 memcpy(packet_pdu, pdu, pdu_size);
1817 send_packet(port, &packet);
1818 ofpbuf_uninit(&packet);
1820 VLOG_ERR_RL(&rl, "port %s: cannot obtain Ethernet address of iface "
1821 "%s (%s)", port->bundle->name,
1822 netdev_get_name(port->up.netdev), strerror(error));
1827 bundle_send_learning_packets(struct ofbundle *bundle)
1829 struct ofproto_dpif *ofproto = bundle->ofproto;
1830 int error, n_packets, n_errors;
1831 struct mac_entry *e;
1833 error = n_packets = n_errors = 0;
1834 LIST_FOR_EACH (e, lru_node, &ofproto->ml->lrus) {
1835 if (e->port.p != bundle) {
1836 struct ofpbuf *learning_packet;
1837 struct ofport_dpif *port;
1841 /* The assignment to "port" is unnecessary but makes "grep"ing for
1842 * struct ofport_dpif more effective. */
1843 learning_packet = bond_compose_learning_packet(bundle->bond,
1847 ret = send_packet(port, learning_packet);
1848 ofpbuf_delete(learning_packet);
1858 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
1859 VLOG_WARN_RL(&rl, "bond %s: %d errors sending %d gratuitous learning "
1860 "packets, last error was: %s",
1861 bundle->name, n_errors, n_packets, strerror(error));
1863 VLOG_DBG("bond %s: sent %d gratuitous learning packets",
1864 bundle->name, n_packets);
1869 bundle_run(struct ofbundle *bundle)
1872 lacp_run(bundle->lacp, send_pdu_cb);
1875 struct ofport_dpif *port;
1877 LIST_FOR_EACH (port, bundle_node, &bundle->ports) {
1878 bond_slave_set_may_enable(bundle->bond, port, port->may_enable);
1881 bond_run(bundle->bond, &bundle->ofproto->revalidate_set,
1882 lacp_negotiated(bundle->lacp));
1883 if (bond_should_send_learning_packets(bundle->bond)) {
1884 bundle_send_learning_packets(bundle);
1890 bundle_wait(struct ofbundle *bundle)
1893 lacp_wait(bundle->lacp);
1896 bond_wait(bundle->bond);
1903 mirror_scan(struct ofproto_dpif *ofproto)
1907 for (idx = 0; idx < MAX_MIRRORS; idx++) {
1908 if (!ofproto->mirrors[idx]) {
1915 static struct ofmirror *
1916 mirror_lookup(struct ofproto_dpif *ofproto, void *aux)
1920 for (i = 0; i < MAX_MIRRORS; i++) {
1921 struct ofmirror *mirror = ofproto->mirrors[i];
1922 if (mirror && mirror->aux == aux) {
1930 /* Update the 'dup_mirrors' member of each of the ofmirrors in 'ofproto'. */
1932 mirror_update_dups(struct ofproto_dpif *ofproto)
1936 for (i = 0; i < MAX_MIRRORS; i++) {
1937 struct ofmirror *m = ofproto->mirrors[i];
1940 m->dup_mirrors = MIRROR_MASK_C(1) << i;
1944 for (i = 0; i < MAX_MIRRORS; i++) {
1945 struct ofmirror *m1 = ofproto->mirrors[i];
1952 for (j = i + 1; j < MAX_MIRRORS; j++) {
1953 struct ofmirror *m2 = ofproto->mirrors[j];
1955 if (m2 && m1->out == m2->out && m1->out_vlan == m2->out_vlan) {
1956 m1->dup_mirrors |= MIRROR_MASK_C(1) << j;
1957 m2->dup_mirrors |= m1->dup_mirrors;
1964 mirror_set(struct ofproto *ofproto_, void *aux,
1965 const struct ofproto_mirror_settings *s)
1967 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1968 mirror_mask_t mirror_bit;
1969 struct ofbundle *bundle;
1970 struct ofmirror *mirror;
1971 struct ofbundle *out;
1972 struct hmapx srcs; /* Contains "struct ofbundle *"s. */
1973 struct hmapx dsts; /* Contains "struct ofbundle *"s. */
1976 mirror = mirror_lookup(ofproto, aux);
1978 mirror_destroy(mirror);
1984 idx = mirror_scan(ofproto);
1986 VLOG_WARN("bridge %s: maximum of %d port mirrors reached, "
1988 ofproto->up.name, MAX_MIRRORS, s->name);
1992 mirror = ofproto->mirrors[idx] = xzalloc(sizeof *mirror);
1993 mirror->ofproto = ofproto;
1996 mirror->out_vlan = -1;
1997 mirror->name = NULL;
2000 if (!mirror->name || strcmp(s->name, mirror->name)) {
2002 mirror->name = xstrdup(s->name);
2005 /* Get the new configuration. */
2006 if (s->out_bundle) {
2007 out = bundle_lookup(ofproto, s->out_bundle);
2009 mirror_destroy(mirror);
2015 out_vlan = s->out_vlan;
2017 bundle_lookup_multiple(ofproto, s->srcs, s->n_srcs, &srcs);
2018 bundle_lookup_multiple(ofproto, s->dsts, s->n_dsts, &dsts);
2020 /* If the configuration has not changed, do nothing. */
2021 if (hmapx_equals(&srcs, &mirror->srcs)
2022 && hmapx_equals(&dsts, &mirror->dsts)
2023 && vlan_bitmap_equal(mirror->vlans, s->src_vlans)
2024 && mirror->out == out
2025 && mirror->out_vlan == out_vlan)
2027 hmapx_destroy(&srcs);
2028 hmapx_destroy(&dsts);
2032 hmapx_swap(&srcs, &mirror->srcs);
2033 hmapx_destroy(&srcs);
2035 hmapx_swap(&dsts, &mirror->dsts);
2036 hmapx_destroy(&dsts);
2038 free(mirror->vlans);
2039 mirror->vlans = vlan_bitmap_clone(s->src_vlans);
2042 mirror->out_vlan = out_vlan;
2044 /* Update bundles. */
2045 mirror_bit = MIRROR_MASK_C(1) << mirror->idx;
2046 HMAP_FOR_EACH (bundle, hmap_node, &mirror->ofproto->bundles) {
2047 if (hmapx_contains(&mirror->srcs, bundle)) {
2048 bundle->src_mirrors |= mirror_bit;
2050 bundle->src_mirrors &= ~mirror_bit;
2053 if (hmapx_contains(&mirror->dsts, bundle)) {
2054 bundle->dst_mirrors |= mirror_bit;
2056 bundle->dst_mirrors &= ~mirror_bit;
2059 if (mirror->out == bundle) {
2060 bundle->mirror_out |= mirror_bit;
2062 bundle->mirror_out &= ~mirror_bit;
2066 ofproto->need_revalidate = true;
2067 mac_learning_flush(ofproto->ml);
2068 mirror_update_dups(ofproto);
2074 mirror_destroy(struct ofmirror *mirror)
2076 struct ofproto_dpif *ofproto;
2077 mirror_mask_t mirror_bit;
2078 struct ofbundle *bundle;
2084 ofproto = mirror->ofproto;
2085 ofproto->need_revalidate = true;
2086 mac_learning_flush(ofproto->ml);
2088 mirror_bit = MIRROR_MASK_C(1) << mirror->idx;
2089 HMAP_FOR_EACH (bundle, hmap_node, &ofproto->bundles) {
2090 bundle->src_mirrors &= ~mirror_bit;
2091 bundle->dst_mirrors &= ~mirror_bit;
2092 bundle->mirror_out &= ~mirror_bit;
2095 hmapx_destroy(&mirror->srcs);
2096 hmapx_destroy(&mirror->dsts);
2097 free(mirror->vlans);
2099 ofproto->mirrors[mirror->idx] = NULL;
2103 mirror_update_dups(ofproto);
2107 mirror_get_stats(struct ofproto *ofproto_, void *aux,
2108 uint64_t *packets, uint64_t *bytes)
2110 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
2111 struct ofmirror *mirror = mirror_lookup(ofproto, aux);
2114 *packets = *bytes = UINT64_MAX;
2118 *packets = mirror->packet_count;
2119 *bytes = mirror->byte_count;
2125 set_flood_vlans(struct ofproto *ofproto_, unsigned long *flood_vlans)
2127 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
2128 if (mac_learning_set_flood_vlans(ofproto->ml, flood_vlans)) {
2129 ofproto->need_revalidate = true;
2130 mac_learning_flush(ofproto->ml);
2136 is_mirror_output_bundle(const struct ofproto *ofproto_, void *aux)
2138 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
2139 struct ofbundle *bundle = bundle_lookup(ofproto, aux);
2140 return bundle && bundle->mirror_out != 0;
2144 forward_bpdu_changed(struct ofproto *ofproto_)
2146 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
2147 /* Revalidate cached flows whenever forward_bpdu option changes. */
2148 ofproto->need_revalidate = true;
2153 static struct ofport_dpif *
2154 get_ofp_port(struct ofproto_dpif *ofproto, uint16_t ofp_port)
2156 struct ofport *ofport = ofproto_get_port(&ofproto->up, ofp_port);
2157 return ofport ? ofport_dpif_cast(ofport) : NULL;
2160 static struct ofport_dpif *
2161 get_odp_port(struct ofproto_dpif *ofproto, uint32_t odp_port)
2163 return get_ofp_port(ofproto, odp_port_to_ofp_port(odp_port));
2167 ofproto_port_from_dpif_port(struct ofproto_port *ofproto_port,
2168 struct dpif_port *dpif_port)
2170 ofproto_port->name = dpif_port->name;
2171 ofproto_port->type = dpif_port->type;
2172 ofproto_port->ofp_port = odp_port_to_ofp_port(dpif_port->port_no);
2176 port_run(struct ofport_dpif *ofport)
2178 bool enable = netdev_get_carrier(ofport->up.netdev);
2181 cfm_run(ofport->cfm);
2183 if (cfm_should_send_ccm(ofport->cfm)) {
2184 struct ofpbuf packet;
2186 ofpbuf_init(&packet, 0);
2187 cfm_compose_ccm(ofport->cfm, &packet, ofport->up.opp.hw_addr);
2188 send_packet(ofport, &packet);
2189 ofpbuf_uninit(&packet);
2192 enable = enable && !cfm_get_fault(ofport->cfm)
2193 && cfm_get_opup(ofport->cfm);
2196 if (ofport->bundle) {
2197 enable = enable && lacp_slave_may_enable(ofport->bundle->lacp, ofport);
2200 if (ofport->may_enable != enable) {
2201 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport->up.ofproto);
2203 if (ofproto->has_bundle_action) {
2204 ofproto->need_revalidate = true;
2208 ofport->may_enable = enable;
2212 port_wait(struct ofport_dpif *ofport)
2215 cfm_wait(ofport->cfm);
2220 port_query_by_name(const struct ofproto *ofproto_, const char *devname,
2221 struct ofproto_port *ofproto_port)
2223 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
2224 struct dpif_port dpif_port;
2227 error = dpif_port_query_by_name(ofproto->dpif, devname, &dpif_port);
2229 ofproto_port_from_dpif_port(ofproto_port, &dpif_port);
2235 port_add(struct ofproto *ofproto_, struct netdev *netdev, uint16_t *ofp_portp)
2237 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
2241 error = dpif_port_add(ofproto->dpif, netdev, &odp_port);
2243 *ofp_portp = odp_port_to_ofp_port(odp_port);
2249 port_del(struct ofproto *ofproto_, uint16_t ofp_port)
2251 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
2254 error = dpif_port_del(ofproto->dpif, ofp_port_to_odp_port(ofp_port));
2256 struct ofport_dpif *ofport = get_ofp_port(ofproto, ofp_port);
2258 /* The caller is going to close ofport->up.netdev. If this is a
2259 * bonded port, then the bond is using that netdev, so remove it
2260 * from the bond. The client will need to reconfigure everything
2261 * after deleting ports, so then the slave will get re-added. */
2262 bundle_remove(&ofport->up);
2269 port_get_stats(const struct ofport *ofport_, struct netdev_stats *stats)
2271 struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
2274 error = netdev_get_stats(ofport->up.netdev, stats);
2276 if (!error && ofport->odp_port == OVSP_LOCAL) {
2277 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport->up.ofproto);
2279 /* ofproto->stats.tx_packets represents packets that we created
2280 * internally and sent to some port (e.g. packets sent with
2281 * send_packet()). Account for them as if they had come from
2282 * OFPP_LOCAL and got forwarded. */
2284 if (stats->rx_packets != UINT64_MAX) {
2285 stats->rx_packets += ofproto->stats.tx_packets;
2288 if (stats->rx_bytes != UINT64_MAX) {
2289 stats->rx_bytes += ofproto->stats.tx_bytes;
2292 /* ofproto->stats.rx_packets represents packets that were received on
2293 * some port and we processed internally and dropped (e.g. STP).
2294 * Account fro them as if they had been forwarded to OFPP_LOCAL. */
2296 if (stats->tx_packets != UINT64_MAX) {
2297 stats->tx_packets += ofproto->stats.rx_packets;
2300 if (stats->tx_bytes != UINT64_MAX) {
2301 stats->tx_bytes += ofproto->stats.rx_bytes;
2308 /* Account packets for LOCAL port. */
2310 ofproto_update_local_port_stats(const struct ofproto *ofproto_,
2311 size_t tx_size, size_t rx_size)
2313 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
2316 ofproto->stats.rx_packets++;
2317 ofproto->stats.rx_bytes += rx_size;
2320 ofproto->stats.tx_packets++;
2321 ofproto->stats.tx_bytes += tx_size;
2325 struct port_dump_state {
2326 struct dpif_port_dump dump;
2331 port_dump_start(const struct ofproto *ofproto_, void **statep)
2333 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
2334 struct port_dump_state *state;
2336 *statep = state = xmalloc(sizeof *state);
2337 dpif_port_dump_start(&state->dump, ofproto->dpif);
2338 state->done = false;
2343 port_dump_next(const struct ofproto *ofproto_ OVS_UNUSED, void *state_,
2344 struct ofproto_port *port)
2346 struct port_dump_state *state = state_;
2347 struct dpif_port dpif_port;
2349 if (dpif_port_dump_next(&state->dump, &dpif_port)) {
2350 ofproto_port_from_dpif_port(port, &dpif_port);
2353 int error = dpif_port_dump_done(&state->dump);
2355 return error ? error : EOF;
2360 port_dump_done(const struct ofproto *ofproto_ OVS_UNUSED, void *state_)
2362 struct port_dump_state *state = state_;
2365 dpif_port_dump_done(&state->dump);
2372 port_poll(const struct ofproto *ofproto_, char **devnamep)
2374 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
2375 return dpif_port_poll(ofproto->dpif, devnamep);
2379 port_poll_wait(const struct ofproto *ofproto_)
2381 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
2382 dpif_port_poll_wait(ofproto->dpif);
2386 port_is_lacp_current(const struct ofport *ofport_)
2388 const struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
2389 return (ofport->bundle && ofport->bundle->lacp
2390 ? lacp_slave_is_current(ofport->bundle->lacp, ofport)
2394 /* Upcall handling. */
2396 /* Flow miss batching.
2398 * Some dpifs implement operations faster when you hand them off in a batch.
2399 * To allow batching, "struct flow_miss" queues the dpif-related work needed
2400 * for a given flow. Each "struct flow_miss" corresponds to sending one or
2401 * more packets, plus possibly installing the flow in the dpif.
2403 * So far we only batch the operations that affect flow setup time the most.
2404 * It's possible to batch more than that, but the benefit might be minimal. */
2406 struct hmap_node hmap_node;
2408 enum odp_key_fitness key_fitness;
2409 const struct nlattr *key;
2411 ovs_be16 initial_tci;
2412 struct list packets;
2415 struct flow_miss_op {
2416 struct dpif_op dpif_op;
2417 struct subfacet *subfacet;
2420 /* Sends an OFPT_PACKET_IN message for 'packet' of type OFPR_NO_MATCH to each
2421 * OpenFlow controller as necessary according to their individual
2422 * configurations. */
2424 send_packet_in_miss(struct ofproto_dpif *ofproto, struct ofpbuf *packet,
2425 const struct flow *flow)
2427 struct ofputil_packet_in pin;
2429 pin.packet = packet->data;
2430 pin.packet_len = packet->size;
2431 pin.total_len = packet->size;
2432 pin.reason = OFPR_NO_MATCH;
2437 pin.buffer_id = 0; /* not yet known */
2438 pin.send_len = 0; /* not used for flow table misses */
2440 flow_get_metadata(flow, &pin.fmd);
2442 /* Registers aren't meaningful on a miss. */
2443 memset(pin.fmd.reg_masks, 0, sizeof pin.fmd.reg_masks);
2445 connmgr_send_packet_in(ofproto->up.connmgr, &pin, flow);
2449 process_special(struct ofproto_dpif *ofproto, const struct flow *flow,
2450 const struct ofpbuf *packet)
2452 struct ofport_dpif *ofport = get_ofp_port(ofproto, flow->in_port);
2458 if (ofport->cfm && cfm_should_process_flow(ofport->cfm, flow)) {
2460 cfm_process_heartbeat(ofport->cfm, packet);
2463 } else if (ofport->bundle && ofport->bundle->lacp
2464 && flow->dl_type == htons(ETH_TYPE_LACP)) {
2466 lacp_process_packet(ofport->bundle->lacp, ofport, packet);
2469 } else if (ofproto->stp && stp_should_process_flow(flow)) {
2471 stp_process_packet(ofport, packet);
2478 static struct flow_miss *
2479 flow_miss_create(struct hmap *todo, const struct flow *flow,
2480 enum odp_key_fitness key_fitness,
2481 const struct nlattr *key, size_t key_len,
2482 ovs_be16 initial_tci)
2484 uint32_t hash = flow_hash(flow, 0);
2485 struct flow_miss *miss;
2487 HMAP_FOR_EACH_WITH_HASH (miss, hmap_node, hash, todo) {
2488 if (flow_equal(&miss->flow, flow)) {
2493 miss = xmalloc(sizeof *miss);
2494 hmap_insert(todo, &miss->hmap_node, hash);
2496 miss->key_fitness = key_fitness;
2498 miss->key_len = key_len;
2499 miss->initial_tci = initial_tci;
2500 list_init(&miss->packets);
2505 handle_flow_miss(struct ofproto_dpif *ofproto, struct flow_miss *miss,
2506 struct flow_miss_op *ops, size_t *n_ops)
2508 const struct flow *flow = &miss->flow;
2509 struct ofpbuf *packet, *next_packet;
2510 struct subfacet *subfacet;
2511 struct facet *facet;
2513 facet = facet_lookup_valid(ofproto, flow);
2515 struct rule_dpif *rule;
2517 rule = rule_dpif_lookup(ofproto, flow, 0);
2519 /* Don't send a packet-in if OFPPC_NO_PACKET_IN asserted. */
2520 struct ofport_dpif *port = get_ofp_port(ofproto, flow->in_port);
2522 if (port->up.opp.config & htonl(OFPPC_NO_PACKET_IN)) {
2523 COVERAGE_INC(ofproto_dpif_no_packet_in);
2524 /* XXX install 'drop' flow entry */
2528 VLOG_WARN_RL(&rl, "packet-in on unknown port %"PRIu16,
2532 LIST_FOR_EACH (packet, list_node, &miss->packets) {
2533 send_packet_in_miss(ofproto, packet, flow);
2539 facet = facet_create(rule, flow);
2542 subfacet = subfacet_create(facet,
2543 miss->key_fitness, miss->key, miss->key_len,
2546 LIST_FOR_EACH_SAFE (packet, next_packet, list_node, &miss->packets) {
2547 struct dpif_flow_stats stats;
2548 struct flow_miss_op *op;
2549 struct dpif_execute *execute;
2551 ofproto->n_matches++;
2553 if (facet->rule->up.cr.priority == FAIL_OPEN_PRIORITY) {
2555 * Extra-special case for fail-open mode.
2557 * We are in fail-open mode and the packet matched the fail-open
2558 * rule, but we are connected to a controller too. We should send
2559 * the packet up to the controller in the hope that it will try to
2560 * set up a flow and thereby allow us to exit fail-open.
2562 * See the top-level comment in fail-open.c for more information.
2564 send_packet_in_miss(ofproto, packet, flow);
2567 if (!facet->may_install || !subfacet->actions) {
2568 subfacet_make_actions(subfacet, packet);
2571 dpif_flow_stats_extract(&facet->flow, packet, &stats);
2572 subfacet_update_stats(subfacet, &stats);
2574 if (!subfacet->actions_len) {
2575 /* No actions to execute, so skip talking to the dpif. */
2579 list_remove(&packet->list_node);
2580 if (flow->vlan_tci != subfacet->initial_tci) {
2581 /* This packet was received on a VLAN splinter port. We added
2582 * a VLAN to the packet to make the packet resemble the flow,
2583 * but the actions were composed assuming that the packet
2584 * contained no VLAN. So, we must remove the VLAN header from
2585 * the packet before trying to execute the actions. */
2586 eth_pop_vlan(packet);
2589 op = &ops[(*n_ops)++];
2590 execute = &op->dpif_op.u.execute;
2591 op->subfacet = subfacet;
2592 op->dpif_op.type = DPIF_OP_EXECUTE;
2593 execute->key = miss->key;
2594 execute->key_len = miss->key_len;
2595 execute->actions = (facet->may_install
2597 : xmemdup(subfacet->actions,
2598 subfacet->actions_len));
2599 execute->actions_len = subfacet->actions_len;
2600 execute->packet = packet;
2603 if (facet->may_install && subfacet->key_fitness != ODP_FIT_TOO_LITTLE) {
2604 struct flow_miss_op *op = &ops[(*n_ops)++];
2605 struct dpif_flow_put *put = &op->dpif_op.u.flow_put;
2607 op->subfacet = subfacet;
2608 op->dpif_op.type = DPIF_OP_FLOW_PUT;
2609 put->flags = DPIF_FP_CREATE | DPIF_FP_MODIFY;
2610 put->key = miss->key;
2611 put->key_len = miss->key_len;
2612 put->actions = subfacet->actions;
2613 put->actions_len = subfacet->actions_len;
2618 /* Like odp_flow_key_to_flow(), this function converts the 'key_len' bytes of
2619 * OVS_KEY_ATTR_* attributes in 'key' to a flow structure in 'flow' and returns
2620 * an ODP_FIT_* value that indicates how well 'key' fits our expectations for
2621 * what a flow key should contain.
2623 * This function also includes some logic to help make VLAN splinters
2624 * transparent to the rest of the upcall processing logic. In particular, if
2625 * the extracted in_port is a VLAN splinter port, it replaces flow->in_port by
2626 * the "real" port, sets flow->vlan_tci correctly for the VLAN of the VLAN
2627 * splinter port, and pushes a VLAN header onto 'packet' (if it is nonnull).
2629 * Sets '*initial_tci' to the VLAN TCI with which the packet was really
2630 * received, that is, the actual VLAN TCI extracted by odp_flow_key_to_flow().
2631 * (This differs from the value returned in flow->vlan_tci only for packets
2632 * received on VLAN splinters.)
2634 static enum odp_key_fitness
2635 ofproto_dpif_extract_flow_key(const struct ofproto_dpif *ofproto,
2636 const struct nlattr *key, size_t key_len,
2637 struct flow *flow, ovs_be16 *initial_tci,
2638 struct ofpbuf *packet)
2640 enum odp_key_fitness fitness;
2644 fitness = odp_flow_key_to_flow(key, key_len, flow);
2645 if (fitness == ODP_FIT_ERROR) {
2648 *initial_tci = flow->vlan_tci;
2650 realdev = vsp_vlandev_to_realdev(ofproto, flow->in_port, &vid);
2652 /* Cause the flow to be processed as if it came in on the real device
2653 * with the VLAN device's VLAN ID. */
2654 flow->in_port = realdev;
2655 flow->vlan_tci = htons((vid & VLAN_VID_MASK) | VLAN_CFI);
2657 /* Make the packet resemble the flow, so that it gets sent to an
2658 * OpenFlow controller properly, so that it looks correct for
2659 * sFlow, and so that flow_extract() will get the correct vlan_tci
2660 * if it is called on 'packet'.
2662 * The allocated space inside 'packet' probably also contains
2663 * 'key', that is, both 'packet' and 'key' are probably part of a
2664 * struct dpif_upcall (see the large comment on that structure
2665 * definition), so pushing data on 'packet' is in general not a
2666 * good idea since it could overwrite 'key' or free it as a side
2667 * effect. However, it's OK in this special case because we know
2668 * that 'packet' is inside a Netlink attribute: pushing 4 bytes
2669 * will just overwrite the 4-byte "struct nlattr", which is fine
2670 * since we don't need that header anymore. */
2671 eth_push_vlan(packet, flow->vlan_tci);
2674 /* Let the caller know that we can't reproduce 'key' from 'flow'. */
2675 if (fitness == ODP_FIT_PERFECT) {
2676 fitness = ODP_FIT_TOO_MUCH;
2684 handle_miss_upcalls(struct ofproto_dpif *ofproto, struct dpif_upcall *upcalls,
2687 struct dpif_upcall *upcall;
2688 struct flow_miss *miss, *next_miss;
2689 struct flow_miss_op flow_miss_ops[FLOW_MISS_MAX_BATCH * 2];
2690 struct dpif_op *dpif_ops[FLOW_MISS_MAX_BATCH * 2];
2699 /* Construct the to-do list.
2701 * This just amounts to extracting the flow from each packet and sticking
2702 * the packets that have the same flow in the same "flow_miss" structure so
2703 * that we can process them together. */
2705 for (upcall = upcalls; upcall < &upcalls[n_upcalls]; upcall++) {
2706 enum odp_key_fitness fitness;
2707 struct flow_miss *miss;
2708 ovs_be16 initial_tci;
2711 /* Obtain metadata and check userspace/kernel agreement on flow match,
2712 * then set 'flow''s header pointers. */
2713 fitness = ofproto_dpif_extract_flow_key(ofproto,
2714 upcall->key, upcall->key_len,
2715 &flow, &initial_tci,
2717 if (fitness == ODP_FIT_ERROR) {
2718 ofpbuf_delete(upcall->packet);
2721 flow_extract(upcall->packet, flow.skb_priority, flow.tun_id,
2722 flow.in_port, &flow);
2724 /* Handle 802.1ag, LACP, and STP specially. */
2725 if (process_special(ofproto, &flow, upcall->packet)) {
2726 ofproto_update_local_port_stats(&ofproto->up,
2727 0, upcall->packet->size);
2728 ofpbuf_delete(upcall->packet);
2729 ofproto->n_matches++;
2733 /* Add other packets to a to-do list. */
2734 miss = flow_miss_create(&todo, &flow, fitness,
2735 upcall->key, upcall->key_len, initial_tci);
2736 list_push_back(&miss->packets, &upcall->packet->list_node);
2739 /* Process each element in the to-do list, constructing the set of
2740 * operations to batch. */
2742 HMAP_FOR_EACH_SAFE (miss, next_miss, hmap_node, &todo) {
2743 handle_flow_miss(ofproto, miss, flow_miss_ops, &n_ops);
2744 ofpbuf_list_delete(&miss->packets);
2745 hmap_remove(&todo, &miss->hmap_node);
2748 assert(n_ops <= ARRAY_SIZE(flow_miss_ops));
2749 hmap_destroy(&todo);
2751 /* Execute batch. */
2752 for (i = 0; i < n_ops; i++) {
2753 dpif_ops[i] = &flow_miss_ops[i].dpif_op;
2755 dpif_operate(ofproto->dpif, dpif_ops, n_ops);
2757 /* Free memory and update facets. */
2758 for (i = 0; i < n_ops; i++) {
2759 struct flow_miss_op *op = &flow_miss_ops[i];
2760 struct dpif_execute *execute;
2762 switch (op->dpif_op.type) {
2763 case DPIF_OP_EXECUTE:
2764 execute = &op->dpif_op.u.execute;
2765 if (op->subfacet->actions != execute->actions) {
2766 free((struct nlattr *) execute->actions);
2768 ofpbuf_delete((struct ofpbuf *) execute->packet);
2771 case DPIF_OP_FLOW_PUT:
2772 if (!op->dpif_op.error) {
2773 op->subfacet->installed = true;
2781 handle_userspace_upcall(struct ofproto_dpif *ofproto,
2782 struct dpif_upcall *upcall)
2784 struct user_action_cookie cookie;
2785 enum odp_key_fitness fitness;
2786 ovs_be16 initial_tci;
2789 memcpy(&cookie, &upcall->userdata, sizeof(cookie));
2791 fitness = ofproto_dpif_extract_flow_key(ofproto, upcall->key,
2792 upcall->key_len, &flow,
2793 &initial_tci, upcall->packet);
2794 if (fitness == ODP_FIT_ERROR) {
2795 ofpbuf_delete(upcall->packet);
2799 if (cookie.type == USER_ACTION_COOKIE_SFLOW) {
2800 if (ofproto->sflow) {
2801 dpif_sflow_received(ofproto->sflow, upcall->packet, &flow,
2805 VLOG_WARN_RL(&rl, "invalid user cookie : 0x%"PRIx64, upcall->userdata);
2807 ofpbuf_delete(upcall->packet);
2811 handle_upcalls(struct ofproto_dpif *ofproto, unsigned int max_batch)
2813 struct dpif_upcall misses[FLOW_MISS_MAX_BATCH];
2817 assert (max_batch <= FLOW_MISS_MAX_BATCH);
2820 for (i = 0; i < max_batch; i++) {
2821 struct dpif_upcall *upcall = &misses[n_misses];
2824 error = dpif_recv(ofproto->dpif, upcall);
2829 switch (upcall->type) {
2830 case DPIF_UC_ACTION:
2831 handle_userspace_upcall(ofproto, upcall);
2835 /* Handle it later. */
2839 case DPIF_N_UC_TYPES:
2841 VLOG_WARN_RL(&rl, "upcall has unexpected type %"PRIu32,
2847 handle_miss_upcalls(ofproto, misses, n_misses);
2852 /* Flow expiration. */
2854 static int subfacet_max_idle(const struct ofproto_dpif *);
2855 static void update_stats(struct ofproto_dpif *);
2856 static void rule_expire(struct rule_dpif *);
2857 static void expire_subfacets(struct ofproto_dpif *, int dp_max_idle);
2859 /* This function is called periodically by run(). Its job is to collect
2860 * updates for the flows that have been installed into the datapath, most
2861 * importantly when they last were used, and then use that information to
2862 * expire flows that have not been used recently.
2864 * Returns the number of milliseconds after which it should be called again. */
2866 expire(struct ofproto_dpif *ofproto)
2868 struct rule_dpif *rule, *next_rule;
2869 struct classifier *table;
2872 /* Update stats for each flow in the datapath. */
2873 update_stats(ofproto);
2875 /* Expire subfacets that have been idle too long. */
2876 dp_max_idle = subfacet_max_idle(ofproto);
2877 expire_subfacets(ofproto, dp_max_idle);
2879 /* Expire OpenFlow flows whose idle_timeout or hard_timeout has passed. */
2880 OFPROTO_FOR_EACH_TABLE (table, &ofproto->up) {
2881 struct cls_cursor cursor;
2883 cls_cursor_init(&cursor, table, NULL);
2884 CLS_CURSOR_FOR_EACH_SAFE (rule, next_rule, up.cr, &cursor) {
2889 /* All outstanding data in existing flows has been accounted, so it's a
2890 * good time to do bond rebalancing. */
2891 if (ofproto->has_bonded_bundles) {
2892 struct ofbundle *bundle;
2894 HMAP_FOR_EACH (bundle, hmap_node, &ofproto->bundles) {
2896 bond_rebalance(bundle->bond, &ofproto->revalidate_set);
2901 return MIN(dp_max_idle, 1000);
2904 /* Update 'packet_count', 'byte_count', and 'used' members of installed facets.
2906 * This function also pushes statistics updates to rules which each facet
2907 * resubmits into. Generally these statistics will be accurate. However, if a
2908 * facet changes the rule it resubmits into at some time in between
2909 * update_stats() runs, it is possible that statistics accrued to the
2910 * old rule will be incorrectly attributed to the new rule. This could be
2911 * avoided by calling update_stats() whenever rules are created or
2912 * deleted. However, the performance impact of making so many calls to the
2913 * datapath do not justify the benefit of having perfectly accurate statistics.
2916 update_stats(struct ofproto_dpif *p)
2918 const struct dpif_flow_stats *stats;
2919 struct dpif_flow_dump dump;
2920 const struct nlattr *key;
2923 dpif_flow_dump_start(&dump, p->dpif);
2924 while (dpif_flow_dump_next(&dump, &key, &key_len, NULL, NULL, &stats)) {
2925 struct subfacet *subfacet;
2927 subfacet = subfacet_find(p, key, key_len);
2928 if (subfacet && subfacet->installed) {
2929 struct facet *facet = subfacet->facet;
2931 if (stats->n_packets >= subfacet->dp_packet_count) {
2932 uint64_t extra = stats->n_packets - subfacet->dp_packet_count;
2933 facet->packet_count += extra;
2935 VLOG_WARN_RL(&rl, "unexpected packet count from the datapath");
2938 if (stats->n_bytes >= subfacet->dp_byte_count) {
2939 facet->byte_count += stats->n_bytes - subfacet->dp_byte_count;
2941 VLOG_WARN_RL(&rl, "unexpected byte count from datapath");
2944 subfacet->dp_packet_count = stats->n_packets;
2945 subfacet->dp_byte_count = stats->n_bytes;
2947 subfacet_update_time(subfacet, stats->used);
2948 facet_account(facet);
2949 facet_push_stats(facet);
2951 if (!VLOG_DROP_WARN(&rl)) {
2955 odp_flow_key_format(key, key_len, &s);
2956 VLOG_WARN("unexpected flow from datapath %s", ds_cstr(&s));
2960 COVERAGE_INC(facet_unexpected);
2961 /* There's a flow in the datapath that we know nothing about, or a
2962 * flow that shouldn't be installed but was anyway. Delete it. */
2963 dpif_flow_del(p->dpif, key, key_len, NULL);
2966 dpif_flow_dump_done(&dump);
2969 /* Calculates and returns the number of milliseconds of idle time after which
2970 * subfacets should expire from the datapath. When a subfacet expires, we fold
2971 * its statistics into its facet, and when a facet's last subfacet expires, we
2972 * fold its statistic into its rule. */
2974 subfacet_max_idle(const struct ofproto_dpif *ofproto)
2977 * Idle time histogram.
2979 * Most of the time a switch has a relatively small number of subfacets.
2980 * When this is the case we might as well keep statistics for all of them
2981 * in userspace and to cache them in the kernel datapath for performance as
2984 * As the number of subfacets increases, the memory required to maintain
2985 * statistics about them in userspace and in the kernel becomes
2986 * significant. However, with a large number of subfacets it is likely
2987 * that only a few of them are "heavy hitters" that consume a large amount
2988 * of bandwidth. At this point, only heavy hitters are worth caching in
2989 * the kernel and maintaining in userspaces; other subfacets we can
2992 * The technique used to compute the idle time is to build a histogram with
2993 * N_BUCKETS buckets whose width is BUCKET_WIDTH msecs each. Each subfacet
2994 * that is installed in the kernel gets dropped in the appropriate bucket.
2995 * After the histogram has been built, we compute the cutoff so that only
2996 * the most-recently-used 1% of subfacets (but at least
2997 * ofproto->up.flow_eviction_threshold flows) are kept cached. At least
2998 * the most-recently-used bucket of subfacets is kept, so actually an
2999 * arbitrary number of subfacets can be kept in any given expiration run
3000 * (though the next run will delete most of those unless they receive
3003 * This requires a second pass through the subfacets, in addition to the
3004 * pass made by update_stats(), because the former function never looks at
3005 * uninstallable subfacets.
3007 enum { BUCKET_WIDTH = ROUND_UP(100, TIME_UPDATE_INTERVAL) };
3008 enum { N_BUCKETS = 5000 / BUCKET_WIDTH };
3009 int buckets[N_BUCKETS] = { 0 };
3010 int total, subtotal, bucket;
3011 struct subfacet *subfacet;
3015 total = hmap_count(&ofproto->subfacets);
3016 if (total <= ofproto->up.flow_eviction_threshold) {
3017 return N_BUCKETS * BUCKET_WIDTH;
3020 /* Build histogram. */
3022 HMAP_FOR_EACH (subfacet, hmap_node, &ofproto->subfacets) {
3023 long long int idle = now - subfacet->used;
3024 int bucket = (idle <= 0 ? 0
3025 : idle >= BUCKET_WIDTH * N_BUCKETS ? N_BUCKETS - 1
3026 : (unsigned int) idle / BUCKET_WIDTH);
3030 /* Find the first bucket whose flows should be expired. */
3031 subtotal = bucket = 0;
3033 subtotal += buckets[bucket++];
3034 } while (bucket < N_BUCKETS &&
3035 subtotal < MAX(ofproto->up.flow_eviction_threshold, total / 100));
3037 if (VLOG_IS_DBG_ENABLED()) {
3041 ds_put_cstr(&s, "keep");
3042 for (i = 0; i < N_BUCKETS; i++) {
3044 ds_put_cstr(&s, ", drop");
3047 ds_put_format(&s, " %d:%d", i * BUCKET_WIDTH, buckets[i]);
3050 VLOG_INFO("%s: %s (msec:count)", ofproto->up.name, ds_cstr(&s));
3054 return bucket * BUCKET_WIDTH;
3058 expire_subfacets(struct ofproto_dpif *ofproto, int dp_max_idle)
3060 long long int cutoff = time_msec() - dp_max_idle;
3061 struct subfacet *subfacet, *next_subfacet;
3063 HMAP_FOR_EACH_SAFE (subfacet, next_subfacet, hmap_node,
3064 &ofproto->subfacets) {
3065 if (subfacet->used < cutoff) {
3066 subfacet_destroy(subfacet);
3071 /* If 'rule' is an OpenFlow rule, that has expired according to OpenFlow rules,
3072 * then delete it entirely. */
3074 rule_expire(struct rule_dpif *rule)
3076 struct facet *facet, *next_facet;
3080 /* Has 'rule' expired? */
3082 if (rule->up.hard_timeout
3083 && now > rule->up.modified + rule->up.hard_timeout * 1000) {
3084 reason = OFPRR_HARD_TIMEOUT;
3085 } else if (rule->up.idle_timeout && list_is_empty(&rule->facets)
3086 && now > rule->used + rule->up.idle_timeout * 1000) {
3087 reason = OFPRR_IDLE_TIMEOUT;
3092 COVERAGE_INC(ofproto_dpif_expired);
3094 /* Update stats. (This is a no-op if the rule expired due to an idle
3095 * timeout, because that only happens when the rule has no facets left.) */
3096 LIST_FOR_EACH_SAFE (facet, next_facet, list_node, &rule->facets) {
3097 facet_remove(facet);
3100 /* Get rid of the rule. */
3101 ofproto_rule_expire(&rule->up, reason);
3106 /* Creates and returns a new facet owned by 'rule', given a 'flow'.
3108 * The caller must already have determined that no facet with an identical
3109 * 'flow' exists in 'ofproto' and that 'flow' is the best match for 'rule' in
3110 * the ofproto's classifier table.
3112 * The facet will initially have no subfacets. The caller should create (at
3113 * least) one subfacet with subfacet_create(). */
3114 static struct facet *
3115 facet_create(struct rule_dpif *rule, const struct flow *flow)
3117 struct ofproto_dpif *ofproto = ofproto_dpif_cast(rule->up.ofproto);
3118 struct facet *facet;
3120 facet = xzalloc(sizeof *facet);
3121 facet->used = time_msec();
3122 hmap_insert(&ofproto->facets, &facet->hmap_node, flow_hash(flow, 0));
3123 list_push_back(&rule->facets, &facet->list_node);
3125 facet->flow = *flow;
3126 list_init(&facet->subfacets);
3127 netflow_flow_init(&facet->nf_flow);
3128 netflow_flow_update_time(ofproto->netflow, &facet->nf_flow, facet->used);
3134 facet_free(struct facet *facet)
3139 /* Executes, within 'ofproto', the 'n_actions' actions in 'actions' on
3140 * 'packet', which arrived on 'in_port'.
3142 * Takes ownership of 'packet'. */
3144 execute_odp_actions(struct ofproto_dpif *ofproto, const struct flow *flow,
3145 const struct nlattr *odp_actions, size_t actions_len,
3146 struct ofpbuf *packet)
3148 struct odputil_keybuf keybuf;
3152 ofpbuf_use_stack(&key, &keybuf, sizeof keybuf);
3153 odp_flow_key_from_flow(&key, flow);
3155 error = dpif_execute(ofproto->dpif, key.data, key.size,
3156 odp_actions, actions_len, packet);
3158 ofpbuf_delete(packet);
3162 /* Remove 'facet' from 'ofproto' and free up the associated memory:
3164 * - If 'facet' was installed in the datapath, uninstalls it and updates its
3165 * rule's statistics, via subfacet_uninstall().
3167 * - Removes 'facet' from its rule and from ofproto->facets.
3170 facet_remove(struct facet *facet)
3172 struct ofproto_dpif *ofproto = ofproto_dpif_cast(facet->rule->up.ofproto);
3173 struct subfacet *subfacet, *next_subfacet;
3175 assert(!list_is_empty(&facet->subfacets));
3177 /* First uninstall all of the subfacets to get final statistics. */
3178 LIST_FOR_EACH (subfacet, list_node, &facet->subfacets) {
3179 subfacet_uninstall(subfacet);
3182 /* Flush the final stats to the rule.
3184 * This might require us to have at least one subfacet around so that we
3185 * can use its actions for accounting in facet_account(), which is why we
3186 * have uninstalled but not yet destroyed the subfacets. */
3187 facet_flush_stats(facet);
3189 /* Now we're really all done so destroy everything. */
3190 LIST_FOR_EACH_SAFE (subfacet, next_subfacet, list_node,
3191 &facet->subfacets) {
3192 subfacet_destroy__(subfacet);
3194 hmap_remove(&ofproto->facets, &facet->hmap_node);
3195 list_remove(&facet->list_node);
3200 facet_account(struct facet *facet)
3202 struct ofproto_dpif *ofproto = ofproto_dpif_cast(facet->rule->up.ofproto);
3204 struct subfacet *subfacet;
3205 const struct nlattr *a;
3209 if (facet->byte_count <= facet->accounted_bytes) {
3212 n_bytes = facet->byte_count - facet->accounted_bytes;
3213 facet->accounted_bytes = facet->byte_count;
3215 /* Feed information from the active flows back into the learning table to
3216 * ensure that table is always in sync with what is actually flowing
3217 * through the datapath. */
3218 if (facet->has_learn || facet->has_normal) {
3219 struct action_xlate_ctx ctx;
3221 action_xlate_ctx_init(&ctx, ofproto, &facet->flow,
3222 facet->flow.vlan_tci,
3223 facet->rule->up.flow_cookie, NULL);
3224 ctx.may_learn = true;
3225 ofpbuf_delete(xlate_actions(&ctx, facet->rule->up.actions,
3226 facet->rule->up.n_actions));
3229 if (!facet->has_normal || !ofproto->has_bonded_bundles) {
3233 /* This loop feeds byte counters to bond_account() for rebalancing to use
3234 * as a basis. We also need to track the actual VLAN on which the packet
3235 * is going to be sent to ensure that it matches the one passed to
3236 * bond_choose_output_slave(). (Otherwise, we will account to the wrong
3239 * We use the actions from an arbitrary subfacet because they should all
3240 * be equally valid for our purpose. */
3241 subfacet = CONTAINER_OF(list_front(&facet->subfacets),
3242 struct subfacet, list_node);
3243 vlan_tci = facet->flow.vlan_tci;
3244 NL_ATTR_FOR_EACH_UNSAFE (a, left,
3245 subfacet->actions, subfacet->actions_len) {
3246 const struct ovs_action_push_vlan *vlan;
3247 struct ofport_dpif *port;
3249 switch (nl_attr_type(a)) {
3250 case OVS_ACTION_ATTR_OUTPUT:
3251 port = get_odp_port(ofproto, nl_attr_get_u32(a));
3252 if (port && port->bundle && port->bundle->bond) {
3253 bond_account(port->bundle->bond, &facet->flow,
3254 vlan_tci_to_vid(vlan_tci), n_bytes);
3258 case OVS_ACTION_ATTR_POP_VLAN:
3259 vlan_tci = htons(0);
3262 case OVS_ACTION_ATTR_PUSH_VLAN:
3263 vlan = nl_attr_get(a);
3264 vlan_tci = vlan->vlan_tci;
3270 /* Returns true if the only action for 'facet' is to send to the controller.
3271 * (We don't report NetFlow expiration messages for such facets because they
3272 * are just part of the control logic for the network, not real traffic). */
3274 facet_is_controller_flow(struct facet *facet)
3277 && facet->rule->up.n_actions == 1
3278 && action_outputs_to_port(&facet->rule->up.actions[0],
3279 htons(OFPP_CONTROLLER)));
3282 /* Folds all of 'facet''s statistics into its rule. Also updates the
3283 * accounting ofhook and emits a NetFlow expiration if appropriate. All of
3284 * 'facet''s statistics in the datapath should have been zeroed and folded into
3285 * its packet and byte counts before this function is called. */
3287 facet_flush_stats(struct facet *facet)
3289 struct ofproto_dpif *ofproto = ofproto_dpif_cast(facet->rule->up.ofproto);
3290 struct subfacet *subfacet;
3292 LIST_FOR_EACH (subfacet, list_node, &facet->subfacets) {
3293 assert(!subfacet->dp_byte_count);
3294 assert(!subfacet->dp_packet_count);
3297 facet_push_stats(facet);
3298 facet_account(facet);
3300 if (ofproto->netflow && !facet_is_controller_flow(facet)) {
3301 struct ofexpired expired;
3302 expired.flow = facet->flow;
3303 expired.packet_count = facet->packet_count;
3304 expired.byte_count = facet->byte_count;
3305 expired.used = facet->used;
3306 netflow_expire(ofproto->netflow, &facet->nf_flow, &expired);
3309 facet->rule->packet_count += facet->packet_count;
3310 facet->rule->byte_count += facet->byte_count;
3312 /* Reset counters to prevent double counting if 'facet' ever gets
3314 facet_reset_counters(facet);
3316 netflow_flow_clear(&facet->nf_flow);
3319 /* Searches 'ofproto''s table of facets for one exactly equal to 'flow'.
3320 * Returns it if found, otherwise a null pointer.
3322 * The returned facet might need revalidation; use facet_lookup_valid()
3323 * instead if that is important. */
3324 static struct facet *
3325 facet_find(struct ofproto_dpif *ofproto, const struct flow *flow)
3327 struct facet *facet;
3329 HMAP_FOR_EACH_WITH_HASH (facet, hmap_node, flow_hash(flow, 0),
3331 if (flow_equal(flow, &facet->flow)) {
3339 /* Searches 'ofproto''s table of facets for one exactly equal to 'flow'.
3340 * Returns it if found, otherwise a null pointer.
3342 * The returned facet is guaranteed to be valid. */
3343 static struct facet *
3344 facet_lookup_valid(struct ofproto_dpif *ofproto, const struct flow *flow)
3346 struct facet *facet = facet_find(ofproto, flow);
3348 /* The facet we found might not be valid, since we could be in need of
3349 * revalidation. If it is not valid, don't return it. */
3351 && (ofproto->need_revalidate
3352 || tag_set_intersects(&ofproto->revalidate_set, facet->tags))
3353 && !facet_revalidate(facet)) {
3354 COVERAGE_INC(facet_invalidated);
3362 facet_check_consistency(struct facet *facet)
3364 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 15);
3366 struct ofproto_dpif *ofproto = ofproto_dpif_cast(facet->rule->up.ofproto);
3368 struct rule_dpif *rule;
3369 struct subfacet *subfacet;
3372 /* Check the rule for consistency. */
3373 rule = rule_dpif_lookup(ofproto, &facet->flow, 0);
3375 if (!VLOG_DROP_WARN(&rl)) {
3376 char *s = flow_to_string(&facet->flow);
3377 VLOG_WARN("%s: facet should not exist", s);
3381 } else if (rule != facet->rule) {
3385 flow_format(&s, &facet->flow);
3386 ds_put_format(&s, ": facet associated with wrong rule (was "
3387 "table=%"PRIu8",", facet->rule->up.table_id);
3388 cls_rule_format(&facet->rule->up.cr, &s);
3389 ds_put_format(&s, ") (should have been table=%"PRIu8",",
3391 cls_rule_format(&rule->up.cr, &s);
3392 ds_put_char(&s, ')');
3394 VLOG_WARN("%s", ds_cstr(&s));
3402 /* Check the datapath actions for consistency. */
3403 LIST_FOR_EACH (subfacet, list_node, &facet->subfacets) {
3404 struct action_xlate_ctx ctx;
3405 struct ofpbuf *odp_actions;
3406 bool actions_changed;
3407 bool should_install;
3409 action_xlate_ctx_init(&ctx, ofproto, &facet->flow,
3410 subfacet->initial_tci, rule->up.flow_cookie,
3412 odp_actions = xlate_actions(&ctx, rule->up.actions,
3413 rule->up.n_actions);
3415 should_install = (ctx.may_set_up_flow
3416 && subfacet->key_fitness != ODP_FIT_TOO_LITTLE);
3417 if (!should_install && !subfacet->installed) {
3418 /* The actions for uninstallable flows may vary from one packet to
3419 * the next, so don't compare the actions. */
3423 actions_changed = (subfacet->actions_len != odp_actions->size
3424 || memcmp(subfacet->actions, odp_actions->data,
3425 subfacet->actions_len));
3426 if (should_install != subfacet->installed || actions_changed) {
3427 struct odputil_keybuf keybuf;
3434 subfacet_get_key(subfacet, &keybuf, &key);
3435 odp_flow_key_format(key.data, key.size, &s);
3437 ds_put_cstr(&s, ": inconsistency in subfacet");
3438 if (should_install != subfacet->installed) {
3439 enum odp_key_fitness fitness = subfacet->key_fitness;
3441 ds_put_format(&s, " (should%s have been installed)",
3442 should_install ? "" : " not");
3443 ds_put_format(&s, " (may_set_up_flow=%s, fitness=%s)",
3444 ctx.may_set_up_flow ? "true" : "false",
3445 odp_key_fitness_to_string(fitness));
3447 if (actions_changed) {
3448 ds_put_cstr(&s, " (actions were: ");
3449 format_odp_actions(&s, subfacet->actions,
3450 subfacet->actions_len);
3451 ds_put_cstr(&s, ") (correct actions: ");
3452 format_odp_actions(&s, odp_actions->data,
3454 ds_put_char(&s, ')');
3456 ds_put_cstr(&s, " (actions: ");
3457 format_odp_actions(&s, subfacet->actions,
3458 subfacet->actions_len);
3459 ds_put_char(&s, ')');
3461 VLOG_WARN("%s", ds_cstr(&s));
3466 ofpbuf_delete(odp_actions);
3472 /* Re-searches the classifier for 'facet':
3474 * - If the rule found is different from 'facet''s current rule, moves
3475 * 'facet' to the new rule and recompiles its actions.
3477 * - If the rule found is the same as 'facet''s current rule, leaves 'facet'
3478 * where it is and recompiles its actions anyway.
3480 * - If there is none, destroys 'facet'.
3482 * Returns true if 'facet' still exists, false if it has been destroyed. */
3484 facet_revalidate(struct facet *facet)
3486 struct ofproto_dpif *ofproto = ofproto_dpif_cast(facet->rule->up.ofproto);
3488 struct nlattr *odp_actions;
3491 struct actions *new_actions;
3493 struct action_xlate_ctx ctx;
3494 struct rule_dpif *new_rule;
3495 struct subfacet *subfacet;
3496 bool actions_changed;
3499 COVERAGE_INC(facet_revalidate);
3501 /* Determine the new rule. */
3502 new_rule = rule_dpif_lookup(ofproto, &facet->flow, 0);
3504 /* No new rule, so delete the facet. */
3505 facet_remove(facet);
3509 /* Calculate new datapath actions.
3511 * We do not modify any 'facet' state yet, because we might need to, e.g.,
3512 * emit a NetFlow expiration and, if so, we need to have the old state
3513 * around to properly compose it. */
3515 /* If the datapath actions changed or the installability changed,
3516 * then we need to talk to the datapath. */
3519 memset(&ctx, 0, sizeof ctx);
3520 LIST_FOR_EACH (subfacet, list_node, &facet->subfacets) {
3521 struct ofpbuf *odp_actions;
3522 bool should_install;
3524 action_xlate_ctx_init(&ctx, ofproto, &facet->flow,
3525 subfacet->initial_tci, new_rule->up.flow_cookie,
3527 odp_actions = xlate_actions(&ctx, new_rule->up.actions,
3528 new_rule->up.n_actions);
3529 actions_changed = (subfacet->actions_len != odp_actions->size
3530 || memcmp(subfacet->actions, odp_actions->data,
3531 subfacet->actions_len));
3533 should_install = (ctx.may_set_up_flow
3534 && subfacet->key_fitness != ODP_FIT_TOO_LITTLE);
3535 if (actions_changed || should_install != subfacet->installed) {
3536 if (should_install) {
3537 struct dpif_flow_stats stats;
3539 subfacet_install(subfacet,
3540 odp_actions->data, odp_actions->size, &stats);
3541 subfacet_update_stats(subfacet, &stats);
3543 subfacet_uninstall(subfacet);
3547 new_actions = xcalloc(list_size(&facet->subfacets),
3548 sizeof *new_actions);
3550 new_actions[i].odp_actions = xmemdup(odp_actions->data,
3552 new_actions[i].actions_len = odp_actions->size;
3555 ofpbuf_delete(odp_actions);
3559 facet_flush_stats(facet);
3562 /* Update 'facet' now that we've taken care of all the old state. */
3563 facet->tags = ctx.tags;
3564 facet->nf_flow.output_iface = ctx.nf_output_iface;
3565 facet->may_install = ctx.may_set_up_flow;
3566 facet->has_learn = ctx.has_learn;
3567 facet->has_normal = ctx.has_normal;
3568 facet->mirrors = ctx.mirrors;
3571 LIST_FOR_EACH (subfacet, list_node, &facet->subfacets) {
3572 if (new_actions[i].odp_actions) {
3573 free(subfacet->actions);
3574 subfacet->actions = new_actions[i].odp_actions;
3575 subfacet->actions_len = new_actions[i].actions_len;
3581 if (facet->rule != new_rule) {
3582 COVERAGE_INC(facet_changed_rule);
3583 list_remove(&facet->list_node);
3584 list_push_back(&new_rule->facets, &facet->list_node);
3585 facet->rule = new_rule;
3586 facet->used = new_rule->up.created;
3587 facet->prev_used = facet->used;
3593 /* Updates 'facet''s used time. Caller is responsible for calling
3594 * facet_push_stats() to update the flows which 'facet' resubmits into. */
3596 facet_update_time(struct facet *facet, long long int used)
3598 struct ofproto_dpif *ofproto = ofproto_dpif_cast(facet->rule->up.ofproto);
3599 if (used > facet->used) {
3601 if (used > facet->rule->used) {
3602 facet->rule->used = used;
3604 netflow_flow_update_time(ofproto->netflow, &facet->nf_flow, used);
3609 facet_reset_counters(struct facet *facet)
3611 facet->packet_count = 0;
3612 facet->byte_count = 0;
3613 facet->prev_packet_count = 0;
3614 facet->prev_byte_count = 0;
3615 facet->accounted_bytes = 0;
3619 facet_push_stats(struct facet *facet)
3621 uint64_t new_packets, new_bytes;
3623 assert(facet->packet_count >= facet->prev_packet_count);
3624 assert(facet->byte_count >= facet->prev_byte_count);
3625 assert(facet->used >= facet->prev_used);
3627 new_packets = facet->packet_count - facet->prev_packet_count;
3628 new_bytes = facet->byte_count - facet->prev_byte_count;
3630 if (new_packets || new_bytes || facet->used > facet->prev_used) {
3631 facet->prev_packet_count = facet->packet_count;
3632 facet->prev_byte_count = facet->byte_count;
3633 facet->prev_used = facet->used;
3635 flow_push_stats(facet->rule, &facet->flow,
3636 new_packets, new_bytes, facet->used);
3638 update_mirror_stats(ofproto_dpif_cast(facet->rule->up.ofproto),
3639 facet->mirrors, new_packets, new_bytes);
3643 struct ofproto_push {
3644 struct action_xlate_ctx ctx;
3651 push_resubmit(struct action_xlate_ctx *ctx, struct rule_dpif *rule)
3653 struct ofproto_push *push = CONTAINER_OF(ctx, struct ofproto_push, ctx);
3656 rule->packet_count += push->packets;
3657 rule->byte_count += push->bytes;
3658 rule->used = MAX(push->used, rule->used);
3662 /* Pushes flow statistics to the rules which 'flow' resubmits into given
3663 * 'rule''s actions and mirrors. */
3665 flow_push_stats(const struct rule_dpif *rule,
3666 const struct flow *flow, uint64_t packets, uint64_t bytes,
3669 struct ofproto_dpif *ofproto = ofproto_dpif_cast(rule->up.ofproto);
3670 struct ofproto_push push;
3672 push.packets = packets;
3676 action_xlate_ctx_init(&push.ctx, ofproto, flow, flow->vlan_tci,
3677 rule->up.flow_cookie, NULL);
3678 push.ctx.resubmit_hook = push_resubmit;
3679 ofpbuf_delete(xlate_actions(&push.ctx,
3680 rule->up.actions, rule->up.n_actions));
3685 static struct subfacet *
3686 subfacet_find__(struct ofproto_dpif *ofproto,
3687 const struct nlattr *key, size_t key_len, uint32_t key_hash,
3688 const struct flow *flow)
3690 struct subfacet *subfacet;
3692 HMAP_FOR_EACH_WITH_HASH (subfacet, hmap_node, key_hash,
3693 &ofproto->subfacets) {
3695 ? (subfacet->key_len == key_len
3696 && !memcmp(key, subfacet->key, key_len))
3697 : flow_equal(flow, &subfacet->facet->flow)) {
3705 /* Searches 'facet' (within 'ofproto') for a subfacet with the specified
3706 * 'key_fitness', 'key', and 'key_len'. Returns the existing subfacet if
3707 * there is one, otherwise creates and returns a new subfacet.
3709 * If the returned subfacet is new, then subfacet->actions will be NULL, in
3710 * which case the caller must populate the actions with
3711 * subfacet_make_actions(). */
3712 static struct subfacet *
3713 subfacet_create(struct facet *facet, enum odp_key_fitness key_fitness,
3714 const struct nlattr *key, size_t key_len, ovs_be16 initial_tci)
3716 struct ofproto_dpif *ofproto = ofproto_dpif_cast(facet->rule->up.ofproto);
3717 uint32_t key_hash = odp_flow_key_hash(key, key_len);
3718 struct subfacet *subfacet;
3720 subfacet = subfacet_find__(ofproto, key, key_len, key_hash, &facet->flow);
3722 if (subfacet->facet == facet) {
3726 /* This shouldn't happen. */
3727 VLOG_ERR_RL(&rl, "subfacet with wrong facet");
3728 subfacet_destroy(subfacet);
3731 subfacet = xzalloc(sizeof *subfacet);
3732 hmap_insert(&ofproto->subfacets, &subfacet->hmap_node, key_hash);
3733 list_push_back(&facet->subfacets, &subfacet->list_node);
3734 subfacet->facet = facet;
3735 subfacet->used = time_msec();
3736 subfacet->key_fitness = key_fitness;
3737 if (key_fitness != ODP_FIT_PERFECT) {
3738 subfacet->key = xmemdup(key, key_len);
3739 subfacet->key_len = key_len;
3741 subfacet->installed = false;
3742 subfacet->initial_tci = initial_tci;
3747 /* Searches 'ofproto' for a subfacet with the given 'key', 'key_len', and
3748 * 'flow'. Returns the subfacet if one exists, otherwise NULL. */
3749 static struct subfacet *
3750 subfacet_find(struct ofproto_dpif *ofproto,
3751 const struct nlattr *key, size_t key_len)
3753 uint32_t key_hash = odp_flow_key_hash(key, key_len);
3754 enum odp_key_fitness fitness;
3757 fitness = odp_flow_key_to_flow(key, key_len, &flow);
3758 if (fitness == ODP_FIT_ERROR) {
3762 return subfacet_find__(ofproto, key, key_len, key_hash, &flow);
3765 /* Uninstalls 'subfacet' from the datapath, if it is installed, removes it from
3766 * its facet within 'ofproto', and frees it. */
3768 subfacet_destroy__(struct subfacet *subfacet)
3770 struct facet *facet = subfacet->facet;
3771 struct ofproto_dpif *ofproto = ofproto_dpif_cast(facet->rule->up.ofproto);
3773 subfacet_uninstall(subfacet);
3774 hmap_remove(&ofproto->subfacets, &subfacet->hmap_node);
3775 list_remove(&subfacet->list_node);
3776 free(subfacet->key);
3777 free(subfacet->actions);
3781 /* Destroys 'subfacet', as with subfacet_destroy__(), and then if this was the
3782 * last remaining subfacet in its facet destroys the facet too. */
3784 subfacet_destroy(struct subfacet *subfacet)
3786 struct facet *facet = subfacet->facet;
3788 if (list_is_singleton(&facet->subfacets)) {
3789 /* facet_remove() needs at least one subfacet (it will remove it). */
3790 facet_remove(facet);
3792 subfacet_destroy__(subfacet);
3796 /* Initializes 'key' with the sequence of OVS_KEY_ATTR_* Netlink attributes
3797 * that can be used to refer to 'subfacet'. The caller must provide 'keybuf'
3798 * for use as temporary storage. */
3800 subfacet_get_key(struct subfacet *subfacet, struct odputil_keybuf *keybuf,
3803 if (!subfacet->key) {
3804 ofpbuf_use_stack(key, keybuf, sizeof *keybuf);
3805 odp_flow_key_from_flow(key, &subfacet->facet->flow);
3807 ofpbuf_use_const(key, subfacet->key, subfacet->key_len);
3811 /* Composes the datapath actions for 'subfacet' based on its rule's actions. */
3813 subfacet_make_actions(struct subfacet *subfacet, const struct ofpbuf *packet)
3815 struct facet *facet = subfacet->facet;
3816 const struct rule_dpif *rule = facet->rule;
3817 struct ofproto_dpif *ofproto = ofproto_dpif_cast(rule->up.ofproto);
3818 struct ofpbuf *odp_actions;
3819 struct action_xlate_ctx ctx;
3821 action_xlate_ctx_init(&ctx, ofproto, &facet->flow, subfacet->initial_tci,
3822 rule->up.flow_cookie, packet);
3823 odp_actions = xlate_actions(&ctx, rule->up.actions, rule->up.n_actions);
3824 facet->tags = ctx.tags;
3825 facet->may_install = ctx.may_set_up_flow;
3826 facet->has_learn = ctx.has_learn;
3827 facet->has_normal = ctx.has_normal;
3828 facet->nf_flow.output_iface = ctx.nf_output_iface;
3829 facet->mirrors = ctx.mirrors;
3831 if (subfacet->actions_len != odp_actions->size
3832 || memcmp(subfacet->actions, odp_actions->data, odp_actions->size)) {
3833 free(subfacet->actions);
3834 subfacet->actions_len = odp_actions->size;
3835 subfacet->actions = xmemdup(odp_actions->data, odp_actions->size);
3838 ofpbuf_delete(odp_actions);
3841 /* Updates 'subfacet''s datapath flow, setting its actions to 'actions_len'
3842 * bytes of actions in 'actions'. If 'stats' is non-null, statistics counters
3843 * in the datapath will be zeroed and 'stats' will be updated with traffic new
3844 * since 'subfacet' was last updated.
3846 * Returns 0 if successful, otherwise a positive errno value. */
3848 subfacet_install(struct subfacet *subfacet,
3849 const struct nlattr *actions, size_t actions_len,
3850 struct dpif_flow_stats *stats)
3852 struct facet *facet = subfacet->facet;
3853 struct ofproto_dpif *ofproto = ofproto_dpif_cast(facet->rule->up.ofproto);
3854 struct odputil_keybuf keybuf;
3855 enum dpif_flow_put_flags flags;
3859 flags = DPIF_FP_CREATE | DPIF_FP_MODIFY;
3861 flags |= DPIF_FP_ZERO_STATS;
3864 subfacet_get_key(subfacet, &keybuf, &key);
3865 ret = dpif_flow_put(ofproto->dpif, flags, key.data, key.size,
3866 actions, actions_len, stats);
3869 subfacet_reset_dp_stats(subfacet, stats);
3875 /* If 'subfacet' is installed in the datapath, uninstalls it. */
3877 subfacet_uninstall(struct subfacet *subfacet)
3879 if (subfacet->installed) {
3880 struct rule_dpif *rule = subfacet->facet->rule;
3881 struct ofproto_dpif *ofproto = ofproto_dpif_cast(rule->up.ofproto);
3882 struct odputil_keybuf keybuf;
3883 struct dpif_flow_stats stats;
3887 subfacet_get_key(subfacet, &keybuf, &key);
3888 error = dpif_flow_del(ofproto->dpif, key.data, key.size, &stats);
3889 subfacet_reset_dp_stats(subfacet, &stats);
3891 subfacet_update_stats(subfacet, &stats);
3893 subfacet->installed = false;
3895 assert(subfacet->dp_packet_count == 0);
3896 assert(subfacet->dp_byte_count == 0);
3900 /* Resets 'subfacet''s datapath statistics counters. This should be called
3901 * when 'subfacet''s statistics are cleared in the datapath. If 'stats' is
3902 * non-null, it should contain the statistics returned by dpif when 'subfacet'
3903 * was reset in the datapath. 'stats' will be modified to include only
3904 * statistics new since 'subfacet' was last updated. */
3906 subfacet_reset_dp_stats(struct subfacet *subfacet,
3907 struct dpif_flow_stats *stats)
3910 && subfacet->dp_packet_count <= stats->n_packets
3911 && subfacet->dp_byte_count <= stats->n_bytes) {
3912 stats->n_packets -= subfacet->dp_packet_count;
3913 stats->n_bytes -= subfacet->dp_byte_count;
3916 subfacet->dp_packet_count = 0;
3917 subfacet->dp_byte_count = 0;
3920 /* Updates 'subfacet''s used time. The caller is responsible for calling
3921 * facet_push_stats() to update the flows which 'subfacet' resubmits into. */
3923 subfacet_update_time(struct subfacet *subfacet, long long int used)
3925 if (used > subfacet->used) {
3926 subfacet->used = used;
3927 facet_update_time(subfacet->facet, used);
3931 /* Folds the statistics from 'stats' into the counters in 'subfacet'.
3933 * Because of the meaning of a subfacet's counters, it only makes sense to do
3934 * this if 'stats' are not tracked in the datapath, that is, if 'stats'
3935 * represents a packet that was sent by hand or if it represents statistics
3936 * that have been cleared out of the datapath. */
3938 subfacet_update_stats(struct subfacet *subfacet,
3939 const struct dpif_flow_stats *stats)
3941 if (stats->n_packets || stats->used > subfacet->used) {
3942 struct facet *facet = subfacet->facet;
3944 subfacet_update_time(subfacet, stats->used);
3945 facet->packet_count += stats->n_packets;
3946 facet->byte_count += stats->n_bytes;
3947 facet_push_stats(facet);
3948 netflow_flow_update_flags(&facet->nf_flow, stats->tcp_flags);
3954 static struct rule_dpif *
3955 rule_dpif_lookup(struct ofproto_dpif *ofproto, const struct flow *flow,
3958 struct cls_rule *cls_rule;
3959 struct classifier *cls;
3961 if (table_id >= N_TABLES) {
3965 cls = &ofproto->up.tables[table_id];
3966 if (flow->nw_frag & FLOW_NW_FRAG_ANY
3967 && ofproto->up.frag_handling == OFPC_FRAG_NORMAL) {
3968 /* For OFPC_NORMAL frag_handling, we must pretend that transport ports
3969 * are unavailable. */
3970 struct flow ofpc_normal_flow = *flow;
3971 ofpc_normal_flow.tp_src = htons(0);
3972 ofpc_normal_flow.tp_dst = htons(0);
3973 cls_rule = classifier_lookup(cls, &ofpc_normal_flow);
3975 cls_rule = classifier_lookup(cls, flow);
3977 return rule_dpif_cast(rule_from_cls_rule(cls_rule));
3981 complete_operation(struct rule_dpif *rule)
3983 struct ofproto_dpif *ofproto = ofproto_dpif_cast(rule->up.ofproto);
3985 rule_invalidate(rule);
3987 struct dpif_completion *c = xmalloc(sizeof *c);
3988 c->op = rule->up.pending;
3989 list_push_back(&ofproto->completions, &c->list_node);
3991 ofoperation_complete(rule->up.pending, 0);
3995 static struct rule *
3998 struct rule_dpif *rule = xmalloc(sizeof *rule);
4003 rule_dealloc(struct rule *rule_)
4005 struct rule_dpif *rule = rule_dpif_cast(rule_);
4010 rule_construct(struct rule *rule_)
4012 struct rule_dpif *rule = rule_dpif_cast(rule_);
4013 struct ofproto_dpif *ofproto = ofproto_dpif_cast(rule->up.ofproto);
4014 struct rule_dpif *victim;
4018 error = validate_actions(rule->up.actions, rule->up.n_actions,
4019 &rule->up.cr.flow, ofproto->max_ports);
4024 rule->used = rule->up.created;
4025 rule->packet_count = 0;
4026 rule->byte_count = 0;
4028 victim = rule_dpif_cast(ofoperation_get_victim(rule->up.pending));
4029 if (victim && !list_is_empty(&victim->facets)) {
4030 struct facet *facet;
4032 rule->facets = victim->facets;
4033 list_moved(&rule->facets);
4034 LIST_FOR_EACH (facet, list_node, &rule->facets) {
4035 /* XXX: We're only clearing our local counters here. It's possible
4036 * that quite a few packets are unaccounted for in the datapath
4037 * statistics. These will be accounted to the new rule instead of
4038 * cleared as required. This could be fixed by clearing out the
4039 * datapath statistics for this facet, but currently it doesn't
4041 facet_reset_counters(facet);
4045 /* Must avoid list_moved() in this case. */
4046 list_init(&rule->facets);
4049 table_id = rule->up.table_id;
4050 rule->tag = (victim ? victim->tag
4052 : rule_calculate_tag(&rule->up.cr.flow, &rule->up.cr.wc,
4053 ofproto->tables[table_id].basis));
4055 complete_operation(rule);
4060 rule_destruct(struct rule *rule_)
4062 struct rule_dpif *rule = rule_dpif_cast(rule_);
4063 struct facet *facet, *next_facet;
4065 LIST_FOR_EACH_SAFE (facet, next_facet, list_node, &rule->facets) {
4066 facet_revalidate(facet);
4069 complete_operation(rule);
4073 rule_get_stats(struct rule *rule_, uint64_t *packets, uint64_t *bytes)
4075 struct rule_dpif *rule = rule_dpif_cast(rule_);
4076 struct facet *facet;
4078 /* Start from historical data for 'rule' itself that are no longer tracked
4079 * in facets. This counts, for example, facets that have expired. */
4080 *packets = rule->packet_count;
4081 *bytes = rule->byte_count;
4083 /* Add any statistics that are tracked by facets. This includes
4084 * statistical data recently updated by ofproto_update_stats() as well as
4085 * stats for packets that were executed "by hand" via dpif_execute(). */
4086 LIST_FOR_EACH (facet, list_node, &rule->facets) {
4087 *packets += facet->packet_count;
4088 *bytes += facet->byte_count;
4093 rule_execute(struct rule *rule_, const struct flow *flow,
4094 struct ofpbuf *packet)
4096 struct rule_dpif *rule = rule_dpif_cast(rule_);
4097 struct ofproto_dpif *ofproto = ofproto_dpif_cast(rule->up.ofproto);
4098 struct action_xlate_ctx ctx;
4099 struct ofpbuf *odp_actions;
4102 action_xlate_ctx_init(&ctx, ofproto, flow, flow->vlan_tci,
4103 rule->up.flow_cookie, packet);
4104 odp_actions = xlate_actions(&ctx, rule->up.actions, rule->up.n_actions);
4105 size = packet->size;
4106 if (execute_odp_actions(ofproto, flow, odp_actions->data,
4107 odp_actions->size, packet)) {
4108 rule->used = time_msec();
4109 rule->packet_count++;
4110 rule->byte_count += size;
4111 flow_push_stats(rule, flow, 1, size, rule->used);
4113 ofpbuf_delete(odp_actions);
4119 rule_modify_actions(struct rule *rule_)
4121 struct rule_dpif *rule = rule_dpif_cast(rule_);
4122 struct ofproto_dpif *ofproto = ofproto_dpif_cast(rule->up.ofproto);
4125 error = validate_actions(rule->up.actions, rule->up.n_actions,
4126 &rule->up.cr.flow, ofproto->max_ports);
4128 ofoperation_complete(rule->up.pending, error);
4132 complete_operation(rule);
4135 /* Sends 'packet' out 'ofport'.
4136 * May modify 'packet'.
4137 * Returns 0 if successful, otherwise a positive errno value. */
4139 send_packet(const struct ofport_dpif *ofport, struct ofpbuf *packet)
4141 const struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport->up.ofproto);
4142 struct ofpbuf key, odp_actions;
4143 struct odputil_keybuf keybuf;
4148 flow_extract((struct ofpbuf *) packet, 0, 0, 0, &flow);
4149 odp_port = vsp_realdev_to_vlandev(ofproto, ofport->odp_port,
4151 if (odp_port != ofport->odp_port) {
4152 eth_pop_vlan(packet);
4153 flow.vlan_tci = htons(0);
4156 ofpbuf_use_stack(&key, &keybuf, sizeof keybuf);
4157 odp_flow_key_from_flow(&key, &flow);
4159 ofpbuf_init(&odp_actions, 32);
4160 compose_sflow_action(ofproto, &odp_actions, &flow, odp_port);
4162 nl_msg_put_u32(&odp_actions, OVS_ACTION_ATTR_OUTPUT, odp_port);
4163 error = dpif_execute(ofproto->dpif,
4165 odp_actions.data, odp_actions.size,
4167 ofpbuf_uninit(&odp_actions);
4170 VLOG_WARN_RL(&rl, "%s: failed to send packet on port %"PRIu32" (%s)",
4171 ofproto->up.name, odp_port, strerror(error));
4173 ofproto_update_local_port_stats(ofport->up.ofproto, packet->size, 0);
4177 /* OpenFlow to datapath action translation. */
4179 static void do_xlate_actions(const union ofp_action *in, size_t n_in,
4180 struct action_xlate_ctx *ctx);
4181 static void xlate_normal(struct action_xlate_ctx *);
4184 put_userspace_action(const struct ofproto_dpif *ofproto,
4185 struct ofpbuf *odp_actions,
4186 const struct flow *flow,
4187 const struct user_action_cookie *cookie)
4191 pid = dpif_port_get_pid(ofproto->dpif,
4192 ofp_port_to_odp_port(flow->in_port));
4194 return odp_put_userspace_action(pid, cookie, odp_actions);
4197 /* Compose SAMPLE action for sFlow. */
4199 compose_sflow_action(const struct ofproto_dpif *ofproto,
4200 struct ofpbuf *odp_actions,
4201 const struct flow *flow,
4204 uint32_t port_ifindex;
4205 uint32_t probability;
4206 struct user_action_cookie cookie;
4207 size_t sample_offset, actions_offset;
4208 int cookie_offset, n_output;
4210 if (!ofproto->sflow || flow->in_port == OFPP_NONE) {
4214 if (odp_port == OVSP_NONE) {
4218 port_ifindex = dpif_sflow_odp_port_to_ifindex(ofproto->sflow, odp_port);
4222 sample_offset = nl_msg_start_nested(odp_actions, OVS_ACTION_ATTR_SAMPLE);
4224 /* Number of packets out of UINT_MAX to sample. */
4225 probability = dpif_sflow_get_probability(ofproto->sflow);
4226 nl_msg_put_u32(odp_actions, OVS_SAMPLE_ATTR_PROBABILITY, probability);
4228 actions_offset = nl_msg_start_nested(odp_actions, OVS_SAMPLE_ATTR_ACTIONS);
4230 cookie.type = USER_ACTION_COOKIE_SFLOW;
4231 cookie.data = port_ifindex;
4232 cookie.n_output = n_output;
4233 cookie.vlan_tci = 0;
4234 cookie_offset = put_userspace_action(ofproto, odp_actions, flow, &cookie);
4236 nl_msg_end_nested(odp_actions, actions_offset);
4237 nl_msg_end_nested(odp_actions, sample_offset);
4238 return cookie_offset;
4241 /* SAMPLE action must be first action in any given list of actions.
4242 * At this point we do not have all information required to build it. So try to
4243 * build sample action as complete as possible. */
4245 add_sflow_action(struct action_xlate_ctx *ctx)
4247 ctx->user_cookie_offset = compose_sflow_action(ctx->ofproto,
4249 &ctx->flow, OVSP_NONE);
4250 ctx->sflow_odp_port = 0;
4251 ctx->sflow_n_outputs = 0;
4254 /* Fix SAMPLE action according to data collected while composing ODP actions.
4255 * We need to fix SAMPLE actions OVS_SAMPLE_ATTR_ACTIONS attribute, i.e. nested
4256 * USERSPACE action's user-cookie which is required for sflow. */
4258 fix_sflow_action(struct action_xlate_ctx *ctx)
4260 const struct flow *base = &ctx->base_flow;
4261 struct user_action_cookie *cookie;
4263 if (!ctx->user_cookie_offset) {
4267 cookie = ofpbuf_at(ctx->odp_actions, ctx->user_cookie_offset,
4269 assert(cookie != NULL);
4270 assert(cookie->type == USER_ACTION_COOKIE_SFLOW);
4272 if (ctx->sflow_n_outputs) {
4273 cookie->data = dpif_sflow_odp_port_to_ifindex(ctx->ofproto->sflow,
4274 ctx->sflow_odp_port);
4276 if (ctx->sflow_n_outputs >= 255) {
4277 cookie->n_output = 255;
4279 cookie->n_output = ctx->sflow_n_outputs;
4281 cookie->vlan_tci = base->vlan_tci;
4285 compose_output_action__(struct action_xlate_ctx *ctx, uint16_t ofp_port,
4288 const struct ofport_dpif *ofport = get_ofp_port(ctx->ofproto, ofp_port);
4289 uint16_t odp_port = ofp_port_to_odp_port(ofp_port);
4290 ovs_be16 flow_vlan_tci = ctx->flow.vlan_tci;
4291 uint8_t flow_nw_tos = ctx->flow.nw_tos;
4295 struct priority_to_dscp *pdscp;
4297 if (ofport->up.opp.config & htonl(OFPPC_NO_FWD)
4298 || (check_stp && !stp_forward_in_state(ofport->stp_state))) {
4302 pdscp = get_priority(ofport, ctx->flow.skb_priority);
4304 ctx->flow.nw_tos &= ~IP_DSCP_MASK;
4305 ctx->flow.nw_tos |= pdscp->dscp;
4308 /* We may not have an ofport record for this port, but it doesn't hurt
4309 * to allow forwarding to it anyhow. Maybe such a port will appear
4310 * later and we're pre-populating the flow table. */
4313 out_port = vsp_realdev_to_vlandev(ctx->ofproto, odp_port,
4314 ctx->flow.vlan_tci);
4315 if (out_port != odp_port) {
4316 ctx->flow.vlan_tci = htons(0);
4318 commit_odp_actions(&ctx->flow, &ctx->base_flow, ctx->odp_actions);
4319 nl_msg_put_u32(ctx->odp_actions, OVS_ACTION_ATTR_OUTPUT, out_port);
4321 ctx->sflow_odp_port = odp_port;
4322 ctx->sflow_n_outputs++;
4323 ctx->nf_output_iface = ofp_port;
4324 ctx->flow.vlan_tci = flow_vlan_tci;
4325 ctx->flow.nw_tos = flow_nw_tos;
4329 compose_output_action(struct action_xlate_ctx *ctx, uint16_t ofp_port)
4331 compose_output_action__(ctx, ofp_port, true);
4335 xlate_table_action(struct action_xlate_ctx *ctx,
4336 uint16_t in_port, uint8_t table_id)
4338 if (ctx->recurse < MAX_RESUBMIT_RECURSION) {
4339 struct ofproto_dpif *ofproto = ctx->ofproto;
4340 struct rule_dpif *rule;
4341 uint16_t old_in_port;
4342 uint8_t old_table_id;
4344 old_table_id = ctx->table_id;
4345 ctx->table_id = table_id;
4347 /* Look up a flow with 'in_port' as the input port. */
4348 old_in_port = ctx->flow.in_port;
4349 ctx->flow.in_port = in_port;
4350 rule = rule_dpif_lookup(ofproto, &ctx->flow, table_id);
4353 if (table_id > 0 && table_id < N_TABLES) {
4354 struct table_dpif *table = &ofproto->tables[table_id];
4355 if (table->other_table) {
4358 : rule_calculate_tag(&ctx->flow,
4359 &table->other_table->wc,
4364 /* Restore the original input port. Otherwise OFPP_NORMAL and
4365 * OFPP_IN_PORT will have surprising behavior. */
4366 ctx->flow.in_port = old_in_port;
4368 if (ctx->resubmit_hook) {
4369 ctx->resubmit_hook(ctx, rule);
4373 ovs_be64 old_cookie = ctx->cookie;
4376 ctx->cookie = rule->up.flow_cookie;
4377 do_xlate_actions(rule->up.actions, rule->up.n_actions, ctx);
4378 ctx->cookie = old_cookie;
4382 ctx->table_id = old_table_id;
4384 static struct vlog_rate_limit recurse_rl = VLOG_RATE_LIMIT_INIT(1, 1);
4386 VLOG_ERR_RL(&recurse_rl, "resubmit actions recursed over %d times",
4387 MAX_RESUBMIT_RECURSION);
4392 xlate_resubmit_table(struct action_xlate_ctx *ctx,
4393 const struct nx_action_resubmit *nar)
4398 in_port = (nar->in_port == htons(OFPP_IN_PORT)
4400 : ntohs(nar->in_port));
4401 table_id = nar->table == 255 ? ctx->table_id : nar->table;
4403 xlate_table_action(ctx, in_port, table_id);
4407 flood_packets(struct action_xlate_ctx *ctx, bool all)
4409 struct ofport_dpif *ofport;
4411 HMAP_FOR_EACH (ofport, up.hmap_node, &ctx->ofproto->up.ports) {
4412 uint16_t ofp_port = ofport->up.ofp_port;
4414 if (ofp_port == ctx->flow.in_port) {
4419 compose_output_action__(ctx, ofp_port, false);
4420 } else if (!(ofport->up.opp.config & htonl(OFPPC_NO_FLOOD))) {
4421 compose_output_action(ctx, ofp_port);
4425 ctx->nf_output_iface = NF_OUT_FLOOD;
4429 execute_controller_action(struct action_xlate_ctx *ctx, int len,
4430 enum ofp_packet_in_reason reason)
4432 struct ofputil_packet_in pin;
4433 struct ofpbuf *packet;
4435 ctx->may_set_up_flow = false;
4440 packet = ofpbuf_clone(ctx->packet);
4442 if (packet->l2 && packet->l3) {
4443 struct eth_header *eh;
4445 eth_pop_vlan(packet);
4447 assert(eh->eth_type == ctx->flow.dl_type);
4448 memcpy(eh->eth_src, ctx->flow.dl_src, sizeof eh->eth_src);
4449 memcpy(eh->eth_dst, ctx->flow.dl_dst, sizeof eh->eth_dst);
4451 if (ctx->flow.vlan_tci & htons(VLAN_CFI)) {
4452 eth_push_vlan(packet, ctx->flow.vlan_tci);
4456 if (ctx->flow.dl_type == htons(ETH_TYPE_IP)) {
4457 packet_set_ipv4(packet, ctx->flow.nw_src, ctx->flow.nw_dst,
4458 ctx->flow.nw_tos, ctx->flow.nw_ttl);
4462 if (ctx->flow.nw_proto == IPPROTO_TCP) {
4463 packet_set_tcp_port(packet, ctx->flow.tp_src,
4465 } else if (ctx->flow.nw_proto == IPPROTO_UDP) {
4466 packet_set_udp_port(packet, ctx->flow.tp_src,
4473 pin.packet = packet->data;
4474 pin.packet_len = packet->size;
4475 pin.reason = reason;
4476 pin.table_id = ctx->table_id;
4477 pin.cookie = ctx->cookie;
4481 pin.total_len = packet->size;
4482 flow_get_metadata(&ctx->flow, &pin.fmd);
4484 connmgr_send_packet_in(ctx->ofproto->up.connmgr, &pin, &ctx->flow);
4485 ofpbuf_delete(packet);
4489 compose_dec_ttl(struct action_xlate_ctx *ctx)
4491 if (ctx->flow.dl_type != htons(ETH_TYPE_IP) &&
4492 ctx->flow.dl_type != htons(ETH_TYPE_IPV6)) {
4496 if (ctx->flow.nw_ttl > 1) {
4500 execute_controller_action(ctx, UINT16_MAX, OFPR_INVALID_TTL);
4502 /* Stop processing for current table. */
4508 xlate_output_action__(struct action_xlate_ctx *ctx,
4509 uint16_t port, uint16_t max_len)
4511 uint16_t prev_nf_output_iface = ctx->nf_output_iface;
4513 ctx->nf_output_iface = NF_OUT_DROP;
4517 compose_output_action(ctx, ctx->flow.in_port);
4520 xlate_table_action(ctx, ctx->flow.in_port, ctx->table_id);
4526 flood_packets(ctx, false);
4529 flood_packets(ctx, true);
4531 case OFPP_CONTROLLER:
4532 execute_controller_action(ctx, max_len, OFPR_ACTION);
4535 compose_output_action(ctx, OFPP_LOCAL);
4540 if (port != ctx->flow.in_port) {
4541 compose_output_action(ctx, port);
4546 if (prev_nf_output_iface == NF_OUT_FLOOD) {
4547 ctx->nf_output_iface = NF_OUT_FLOOD;
4548 } else if (ctx->nf_output_iface == NF_OUT_DROP) {
4549 ctx->nf_output_iface = prev_nf_output_iface;
4550 } else if (prev_nf_output_iface != NF_OUT_DROP &&
4551 ctx->nf_output_iface != NF_OUT_FLOOD) {
4552 ctx->nf_output_iface = NF_OUT_MULTI;
4557 xlate_output_reg_action(struct action_xlate_ctx *ctx,
4558 const struct nx_action_output_reg *naor)
4562 ofp_port = nxm_read_field_bits(naor->src, naor->ofs_nbits, &ctx->flow);
4564 if (ofp_port <= UINT16_MAX) {
4565 xlate_output_action__(ctx, ofp_port, ntohs(naor->max_len));
4570 xlate_output_action(struct action_xlate_ctx *ctx,
4571 const struct ofp_action_output *oao)
4573 xlate_output_action__(ctx, ntohs(oao->port), ntohs(oao->max_len));
4577 xlate_enqueue_action(struct action_xlate_ctx *ctx,
4578 const struct ofp_action_enqueue *oae)
4581 uint32_t flow_priority, priority;
4584 error = dpif_queue_to_priority(ctx->ofproto->dpif, ntohl(oae->queue_id),
4587 /* Fall back to ordinary output action. */
4588 xlate_output_action__(ctx, ntohs(oae->port), 0);
4592 /* Figure out datapath output port. */
4593 ofp_port = ntohs(oae->port);
4594 if (ofp_port == OFPP_IN_PORT) {
4595 ofp_port = ctx->flow.in_port;
4596 } else if (ofp_port == ctx->flow.in_port) {
4600 /* Add datapath actions. */
4601 flow_priority = ctx->flow.skb_priority;
4602 ctx->flow.skb_priority = priority;
4603 compose_output_action(ctx, ofp_port);
4604 ctx->flow.skb_priority = flow_priority;
4606 /* Update NetFlow output port. */
4607 if (ctx->nf_output_iface == NF_OUT_DROP) {
4608 ctx->nf_output_iface = ofp_port;
4609 } else if (ctx->nf_output_iface != NF_OUT_FLOOD) {
4610 ctx->nf_output_iface = NF_OUT_MULTI;
4615 xlate_set_queue_action(struct action_xlate_ctx *ctx,
4616 const struct nx_action_set_queue *nasq)
4621 error = dpif_queue_to_priority(ctx->ofproto->dpif, ntohl(nasq->queue_id),
4624 /* Couldn't translate queue to a priority, so ignore. A warning
4625 * has already been logged. */
4629 ctx->flow.skb_priority = priority;
4632 struct xlate_reg_state {
4638 xlate_autopath(struct action_xlate_ctx *ctx,
4639 const struct nx_action_autopath *naa)
4641 uint16_t ofp_port = ntohl(naa->id);
4642 struct ofport_dpif *port = get_ofp_port(ctx->ofproto, ofp_port);
4644 if (!port || !port->bundle) {
4645 ofp_port = OFPP_NONE;
4646 } else if (port->bundle->bond) {
4647 /* Autopath does not support VLAN hashing. */
4648 struct ofport_dpif *slave = bond_choose_output_slave(
4649 port->bundle->bond, &ctx->flow, 0, &ctx->tags);
4651 ofp_port = slave->up.ofp_port;
4654 autopath_execute(naa, &ctx->flow, ofp_port);
4658 slave_enabled_cb(uint16_t ofp_port, void *ofproto_)
4660 struct ofproto_dpif *ofproto = ofproto_;
4661 struct ofport_dpif *port;
4671 case OFPP_CONTROLLER: /* Not supported by the bundle action. */
4674 port = get_ofp_port(ofproto, ofp_port);
4675 return port ? port->may_enable : false;
4680 xlate_learn_action(struct action_xlate_ctx *ctx,
4681 const struct nx_action_learn *learn)
4683 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(5, 1);
4684 struct ofputil_flow_mod fm;
4687 learn_execute(learn, &ctx->flow, &fm);
4689 error = ofproto_flow_mod(&ctx->ofproto->up, &fm);
4690 if (error && !VLOG_DROP_WARN(&rl)) {
4691 VLOG_WARN("learning action failed to modify flow table (%s)",
4692 ofperr_get_name(error));
4699 may_receive(const struct ofport_dpif *port, struct action_xlate_ctx *ctx)
4701 if (port->up.opp.config & (eth_addr_equals(ctx->flow.dl_dst, eth_addr_stp)
4702 ? htonl(OFPPC_NO_RECV_STP)
4703 : htonl(OFPPC_NO_RECV))) {
4707 /* Only drop packets here if both forwarding and learning are
4708 * disabled. If just learning is enabled, we need to have
4709 * OFPP_NORMAL and the learning action have a look at the packet
4710 * before we can drop it. */
4711 if (!stp_forward_in_state(port->stp_state)
4712 && !stp_learn_in_state(port->stp_state)) {
4720 do_xlate_actions(const union ofp_action *in, size_t n_in,
4721 struct action_xlate_ctx *ctx)
4723 const struct ofport_dpif *port;
4724 const union ofp_action *ia;
4727 port = get_ofp_port(ctx->ofproto, ctx->flow.in_port);
4728 if (port && !may_receive(port, ctx)) {
4729 /* Drop this flow. */
4733 OFPUTIL_ACTION_FOR_EACH_UNSAFE (ia, left, in, n_in) {
4734 const struct ofp_action_dl_addr *oada;
4735 const struct nx_action_resubmit *nar;
4736 const struct nx_action_set_tunnel *nast;
4737 const struct nx_action_set_queue *nasq;
4738 const struct nx_action_multipath *nam;
4739 const struct nx_action_autopath *naa;
4740 const struct nx_action_bundle *nab;
4741 const struct nx_action_output_reg *naor;
4742 enum ofputil_action_code code;
4749 code = ofputil_decode_action_unsafe(ia);
4751 case OFPUTIL_OFPAT_OUTPUT:
4752 xlate_output_action(ctx, &ia->output);
4755 case OFPUTIL_OFPAT_SET_VLAN_VID:
4756 ctx->flow.vlan_tci &= ~htons(VLAN_VID_MASK);
4757 ctx->flow.vlan_tci |= ia->vlan_vid.vlan_vid | htons(VLAN_CFI);
4760 case OFPUTIL_OFPAT_SET_VLAN_PCP:
4761 ctx->flow.vlan_tci &= ~htons(VLAN_PCP_MASK);
4762 ctx->flow.vlan_tci |= htons(
4763 (ia->vlan_pcp.vlan_pcp << VLAN_PCP_SHIFT) | VLAN_CFI);
4766 case OFPUTIL_OFPAT_STRIP_VLAN:
4767 ctx->flow.vlan_tci = htons(0);
4770 case OFPUTIL_OFPAT_SET_DL_SRC:
4771 oada = ((struct ofp_action_dl_addr *) ia);
4772 memcpy(ctx->flow.dl_src, oada->dl_addr, ETH_ADDR_LEN);
4775 case OFPUTIL_OFPAT_SET_DL_DST:
4776 oada = ((struct ofp_action_dl_addr *) ia);
4777 memcpy(ctx->flow.dl_dst, oada->dl_addr, ETH_ADDR_LEN);
4780 case OFPUTIL_OFPAT_SET_NW_SRC:
4781 ctx->flow.nw_src = ia->nw_addr.nw_addr;
4784 case OFPUTIL_OFPAT_SET_NW_DST:
4785 ctx->flow.nw_dst = ia->nw_addr.nw_addr;
4788 case OFPUTIL_OFPAT_SET_NW_TOS:
4789 /* OpenFlow 1.0 only supports IPv4. */
4790 if (ctx->flow.dl_type == htons(ETH_TYPE_IP)) {
4791 ctx->flow.nw_tos &= ~IP_DSCP_MASK;
4792 ctx->flow.nw_tos |= ia->nw_tos.nw_tos & IP_DSCP_MASK;
4796 case OFPUTIL_OFPAT_SET_TP_SRC:
4797 ctx->flow.tp_src = ia->tp_port.tp_port;
4800 case OFPUTIL_OFPAT_SET_TP_DST:
4801 ctx->flow.tp_dst = ia->tp_port.tp_port;
4804 case OFPUTIL_OFPAT_ENQUEUE:
4805 xlate_enqueue_action(ctx, (const struct ofp_action_enqueue *) ia);
4808 case OFPUTIL_NXAST_RESUBMIT:
4809 nar = (const struct nx_action_resubmit *) ia;
4810 xlate_table_action(ctx, ntohs(nar->in_port), ctx->table_id);
4813 case OFPUTIL_NXAST_RESUBMIT_TABLE:
4814 xlate_resubmit_table(ctx, (const struct nx_action_resubmit *) ia);
4817 case OFPUTIL_NXAST_SET_TUNNEL:
4818 nast = (const struct nx_action_set_tunnel *) ia;
4819 tun_id = htonll(ntohl(nast->tun_id));
4820 ctx->flow.tun_id = tun_id;
4823 case OFPUTIL_NXAST_SET_QUEUE:
4824 nasq = (const struct nx_action_set_queue *) ia;
4825 xlate_set_queue_action(ctx, nasq);
4828 case OFPUTIL_NXAST_POP_QUEUE:
4829 ctx->flow.skb_priority = ctx->orig_skb_priority;
4832 case OFPUTIL_NXAST_REG_MOVE:
4833 nxm_execute_reg_move((const struct nx_action_reg_move *) ia,
4837 case OFPUTIL_NXAST_REG_LOAD:
4838 nxm_execute_reg_load((const struct nx_action_reg_load *) ia,
4842 case OFPUTIL_NXAST_NOTE:
4843 /* Nothing to do. */
4846 case OFPUTIL_NXAST_SET_TUNNEL64:
4847 tun_id = ((const struct nx_action_set_tunnel64 *) ia)->tun_id;
4848 ctx->flow.tun_id = tun_id;
4851 case OFPUTIL_NXAST_MULTIPATH:
4852 nam = (const struct nx_action_multipath *) ia;
4853 multipath_execute(nam, &ctx->flow);
4856 case OFPUTIL_NXAST_AUTOPATH:
4857 naa = (const struct nx_action_autopath *) ia;
4858 xlate_autopath(ctx, naa);
4861 case OFPUTIL_NXAST_BUNDLE:
4862 ctx->ofproto->has_bundle_action = true;
4863 nab = (const struct nx_action_bundle *) ia;
4864 xlate_output_action__(ctx, bundle_execute(nab, &ctx->flow,
4869 case OFPUTIL_NXAST_BUNDLE_LOAD:
4870 ctx->ofproto->has_bundle_action = true;
4871 nab = (const struct nx_action_bundle *) ia;
4872 bundle_execute_load(nab, &ctx->flow, slave_enabled_cb,
4876 case OFPUTIL_NXAST_OUTPUT_REG:
4877 naor = (const struct nx_action_output_reg *) ia;
4878 xlate_output_reg_action(ctx, naor);
4881 case OFPUTIL_NXAST_LEARN:
4882 ctx->has_learn = true;
4883 if (ctx->may_learn) {
4884 xlate_learn_action(ctx, (const struct nx_action_learn *) ia);
4888 case OFPUTIL_NXAST_DEC_TTL:
4889 if (compose_dec_ttl(ctx)) {
4894 case OFPUTIL_NXAST_EXIT:
4901 /* We've let OFPP_NORMAL and the learning action look at the packet,
4902 * so drop it now if forwarding is disabled. */
4903 if (port && !stp_forward_in_state(port->stp_state)) {
4904 ofpbuf_clear(ctx->odp_actions);
4905 add_sflow_action(ctx);
4910 action_xlate_ctx_init(struct action_xlate_ctx *ctx,
4911 struct ofproto_dpif *ofproto, const struct flow *flow,
4912 ovs_be16 initial_tci, ovs_be64 cookie,
4913 const struct ofpbuf *packet)
4915 ctx->ofproto = ofproto;
4917 ctx->base_flow = ctx->flow;
4918 ctx->base_flow.tun_id = 0;
4919 ctx->base_flow.vlan_tci = initial_tci;
4920 ctx->cookie = cookie;
4921 ctx->packet = packet;
4922 ctx->may_learn = packet != NULL;
4923 ctx->resubmit_hook = NULL;
4926 static struct ofpbuf *
4927 xlate_actions(struct action_xlate_ctx *ctx,
4928 const union ofp_action *in, size_t n_in)
4930 struct flow orig_flow = ctx->flow;
4932 COVERAGE_INC(ofproto_dpif_xlate);
4934 ctx->odp_actions = ofpbuf_new(512);
4935 ofpbuf_reserve(ctx->odp_actions, NL_A_U32_SIZE);
4937 ctx->may_set_up_flow = true;
4938 ctx->has_learn = false;
4939 ctx->has_normal = false;
4940 ctx->nf_output_iface = NF_OUT_DROP;
4943 ctx->orig_skb_priority = ctx->flow.skb_priority;
4947 if (ctx->flow.nw_frag & FLOW_NW_FRAG_ANY) {
4948 switch (ctx->ofproto->up.frag_handling) {
4949 case OFPC_FRAG_NORMAL:
4950 /* We must pretend that transport ports are unavailable. */
4951 ctx->flow.tp_src = ctx->base_flow.tp_src = htons(0);
4952 ctx->flow.tp_dst = ctx->base_flow.tp_dst = htons(0);
4955 case OFPC_FRAG_DROP:
4956 return ctx->odp_actions;
4958 case OFPC_FRAG_REASM:
4961 case OFPC_FRAG_NX_MATCH:
4962 /* Nothing to do. */
4965 case OFPC_INVALID_TTL_TO_CONTROLLER:
4970 if (process_special(ctx->ofproto, &ctx->flow, ctx->packet)) {
4971 ctx->may_set_up_flow = false;
4972 return ctx->odp_actions;
4974 add_sflow_action(ctx);
4975 do_xlate_actions(in, n_in, ctx);
4977 if (!connmgr_may_set_up_flow(ctx->ofproto->up.connmgr, &ctx->flow,
4978 ctx->odp_actions->data,
4979 ctx->odp_actions->size)) {
4980 ctx->may_set_up_flow = false;
4982 && connmgr_msg_in_hook(ctx->ofproto->up.connmgr, &ctx->flow,
4984 compose_output_action(ctx, OFPP_LOCAL);
4987 add_mirror_actions(ctx, &orig_flow);
4988 fix_sflow_action(ctx);
4991 return ctx->odp_actions;
4994 /* OFPP_NORMAL implementation. */
4996 static struct ofport_dpif *ofbundle_get_a_port(const struct ofbundle *);
4998 /* Given 'vid', the VID obtained from the 802.1Q header that was received as
4999 * part of a packet (specify 0 if there was no 802.1Q header), and 'in_bundle',
5000 * the bundle on which the packet was received, returns the VLAN to which the
5003 * Both 'vid' and the return value are in the range 0...4095. */
5005 input_vid_to_vlan(const struct ofbundle *in_bundle, uint16_t vid)
5007 switch (in_bundle->vlan_mode) {
5008 case PORT_VLAN_ACCESS:
5009 return in_bundle->vlan;
5012 case PORT_VLAN_TRUNK:
5015 case PORT_VLAN_NATIVE_UNTAGGED:
5016 case PORT_VLAN_NATIVE_TAGGED:
5017 return vid ? vid : in_bundle->vlan;
5024 /* Checks whether a packet with the given 'vid' may ingress on 'in_bundle'.
5025 * If so, returns true. Otherwise, returns false and, if 'warn' is true, logs
5028 * 'vid' should be the VID obtained from the 802.1Q header that was received as
5029 * part of a packet (specify 0 if there was no 802.1Q header), in the range
5032 input_vid_is_valid(uint16_t vid, struct ofbundle *in_bundle, bool warn)
5034 /* Allow any VID on the OFPP_NONE port. */
5035 if (in_bundle == &ofpp_none_bundle) {
5039 switch (in_bundle->vlan_mode) {
5040 case PORT_VLAN_ACCESS:
5043 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
5044 VLOG_WARN_RL(&rl, "bridge %s: dropping VLAN %"PRIu16" tagged "
5045 "packet received on port %s configured as VLAN "
5046 "%"PRIu16" access port",
5047 in_bundle->ofproto->up.name, vid,
5048 in_bundle->name, in_bundle->vlan);
5054 case PORT_VLAN_NATIVE_UNTAGGED:
5055 case PORT_VLAN_NATIVE_TAGGED:
5057 /* Port must always carry its native VLAN. */
5061 case PORT_VLAN_TRUNK:
5062 if (!ofbundle_includes_vlan(in_bundle, vid)) {
5064 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
5065 VLOG_WARN_RL(&rl, "bridge %s: dropping VLAN %"PRIu16" packet "
5066 "received on port %s not configured for trunking "
5068 in_bundle->ofproto->up.name, vid,
5069 in_bundle->name, vid);
5081 /* Given 'vlan', the VLAN that a packet belongs to, and
5082 * 'out_bundle', a bundle on which the packet is to be output, returns the VID
5083 * that should be included in the 802.1Q header. (If the return value is 0,
5084 * then the 802.1Q header should only be included in the packet if there is a
5087 * Both 'vlan' and the return value are in the range 0...4095. */
5089 output_vlan_to_vid(const struct ofbundle *out_bundle, uint16_t vlan)
5091 switch (out_bundle->vlan_mode) {
5092 case PORT_VLAN_ACCESS:
5095 case PORT_VLAN_TRUNK:
5096 case PORT_VLAN_NATIVE_TAGGED:
5099 case PORT_VLAN_NATIVE_UNTAGGED:
5100 return vlan == out_bundle->vlan ? 0 : vlan;
5108 output_normal(struct action_xlate_ctx *ctx, const struct ofbundle *out_bundle,
5111 struct ofport_dpif *port;
5113 ovs_be16 tci, old_tci;
5115 vid = output_vlan_to_vid(out_bundle, vlan);
5116 if (!out_bundle->bond) {
5117 port = ofbundle_get_a_port(out_bundle);
5119 port = bond_choose_output_slave(out_bundle->bond, &ctx->flow,
5122 /* No slaves enabled, so drop packet. */
5127 old_tci = ctx->flow.vlan_tci;
5129 if (tci || out_bundle->use_priority_tags) {
5130 tci |= ctx->flow.vlan_tci & htons(VLAN_PCP_MASK);
5132 tci |= htons(VLAN_CFI);
5135 ctx->flow.vlan_tci = tci;
5137 compose_output_action(ctx, port->up.ofp_port);
5138 ctx->flow.vlan_tci = old_tci;
5142 mirror_mask_ffs(mirror_mask_t mask)
5144 BUILD_ASSERT_DECL(sizeof(unsigned int) >= sizeof(mask));
5149 ofbundle_trunks_vlan(const struct ofbundle *bundle, uint16_t vlan)
5151 return (bundle->vlan_mode != PORT_VLAN_ACCESS
5152 && (!bundle->trunks || bitmap_is_set(bundle->trunks, vlan)));
5156 ofbundle_includes_vlan(const struct ofbundle *bundle, uint16_t vlan)
5158 return vlan == bundle->vlan || ofbundle_trunks_vlan(bundle, vlan);
5161 /* Returns an arbitrary interface within 'bundle'. */
5162 static struct ofport_dpif *
5163 ofbundle_get_a_port(const struct ofbundle *bundle)
5165 return CONTAINER_OF(list_front(&bundle->ports),
5166 struct ofport_dpif, bundle_node);
5170 vlan_is_mirrored(const struct ofmirror *m, int vlan)
5172 return !m->vlans || bitmap_is_set(m->vlans, vlan);
5175 /* Returns true if a packet with Ethernet destination MAC 'dst' may be mirrored
5176 * to a VLAN. In general most packets may be mirrored but we want to drop
5177 * protocols that may confuse switches. */
5179 eth_dst_may_rspan(const uint8_t dst[ETH_ADDR_LEN])
5181 /* If you change this function's behavior, please update corresponding
5182 * documentation in vswitch.xml at the same time. */
5183 if (dst[0] != 0x01) {
5184 /* All the currently banned MACs happen to start with 01 currently, so
5185 * this is a quick way to eliminate most of the good ones. */
5187 if (eth_addr_is_reserved(dst)) {
5188 /* Drop STP, IEEE pause frames, and other reserved protocols
5189 * (01-80-c2-00-00-0x). */
5193 if (dst[0] == 0x01 && dst[1] == 0x00 && dst[2] == 0x0c) {
5195 if ((dst[3] & 0xfe) == 0xcc &&
5196 (dst[4] & 0xfe) == 0xcc &&
5197 (dst[5] & 0xfe) == 0xcc) {
5198 /* Drop the following protocols plus others following the same
5201 CDP, VTP, DTP, PAgP (01-00-0c-cc-cc-cc)
5202 Spanning Tree PVSTP+ (01-00-0c-cc-cc-cd)
5203 STP Uplink Fast (01-00-0c-cd-cd-cd) */
5207 if (!(dst[3] | dst[4] | dst[5])) {
5208 /* Drop Inter Switch Link packets (01-00-0c-00-00-00). */
5217 add_mirror_actions(struct action_xlate_ctx *ctx, const struct flow *orig_flow)
5219 struct ofproto_dpif *ofproto = ctx->ofproto;
5220 mirror_mask_t mirrors;
5221 struct ofbundle *in_bundle;
5224 const struct nlattr *a;
5227 in_bundle = lookup_input_bundle(ctx->ofproto, orig_flow->in_port,
5228 ctx->packet != NULL);
5232 mirrors = in_bundle->src_mirrors;
5234 /* Drop frames on bundles reserved for mirroring. */
5235 if (in_bundle->mirror_out) {
5236 if (ctx->packet != NULL) {
5237 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
5238 VLOG_WARN_RL(&rl, "bridge %s: dropping packet received on port "
5239 "%s, which is reserved exclusively for mirroring",
5240 ctx->ofproto->up.name, in_bundle->name);
5246 vid = vlan_tci_to_vid(orig_flow->vlan_tci);
5247 if (!input_vid_is_valid(vid, in_bundle, ctx->packet != NULL)) {
5250 vlan = input_vid_to_vlan(in_bundle, vid);
5252 /* Look at the output ports to check for destination selections. */
5254 NL_ATTR_FOR_EACH (a, left, ctx->odp_actions->data,
5255 ctx->odp_actions->size) {
5256 enum ovs_action_attr type = nl_attr_type(a);
5257 struct ofport_dpif *ofport;
5259 if (type != OVS_ACTION_ATTR_OUTPUT) {
5263 ofport = get_odp_port(ofproto, nl_attr_get_u32(a));
5264 if (ofport && ofport->bundle) {
5265 mirrors |= ofport->bundle->dst_mirrors;
5273 /* Restore the original packet before adding the mirror actions. */
5274 ctx->flow = *orig_flow;
5279 m = ofproto->mirrors[mirror_mask_ffs(mirrors) - 1];
5281 if (!vlan_is_mirrored(m, vlan)) {
5282 mirrors &= mirrors - 1;
5286 mirrors &= ~m->dup_mirrors;
5287 ctx->mirrors |= m->dup_mirrors;
5289 output_normal(ctx, m->out, vlan);
5290 } else if (eth_dst_may_rspan(orig_flow->dl_dst)
5291 && vlan != m->out_vlan) {
5292 struct ofbundle *bundle;
5294 HMAP_FOR_EACH (bundle, hmap_node, &ofproto->bundles) {
5295 if (ofbundle_includes_vlan(bundle, m->out_vlan)
5296 && !bundle->mirror_out) {
5297 output_normal(ctx, bundle, m->out_vlan);
5305 update_mirror_stats(struct ofproto_dpif *ofproto, mirror_mask_t mirrors,
5306 uint64_t packets, uint64_t bytes)
5312 for (; mirrors; mirrors &= mirrors - 1) {
5315 m = ofproto->mirrors[mirror_mask_ffs(mirrors) - 1];
5318 /* In normal circumstances 'm' will not be NULL. However,
5319 * if mirrors are reconfigured, we can temporarily get out
5320 * of sync in facet_revalidate(). We could "correct" the
5321 * mirror list before reaching here, but doing that would
5322 * not properly account the traffic stats we've currently
5323 * accumulated for previous mirror configuration. */
5327 m->packet_count += packets;
5328 m->byte_count += bytes;
5332 /* A VM broadcasts a gratuitous ARP to indicate that it has resumed after
5333 * migration. Older Citrix-patched Linux DomU used gratuitous ARP replies to
5334 * indicate this; newer upstream kernels use gratuitous ARP requests. */
5336 is_gratuitous_arp(const struct flow *flow)
5338 return (flow->dl_type == htons(ETH_TYPE_ARP)
5339 && eth_addr_is_broadcast(flow->dl_dst)
5340 && (flow->nw_proto == ARP_OP_REPLY
5341 || (flow->nw_proto == ARP_OP_REQUEST
5342 && flow->nw_src == flow->nw_dst)));
5346 update_learning_table(struct ofproto_dpif *ofproto,
5347 const struct flow *flow, int vlan,
5348 struct ofbundle *in_bundle)
5350 struct mac_entry *mac;
5352 /* Don't learn the OFPP_NONE port. */
5353 if (in_bundle == &ofpp_none_bundle) {
5357 if (!mac_learning_may_learn(ofproto->ml, flow->dl_src, vlan)) {
5361 mac = mac_learning_insert(ofproto->ml, flow->dl_src, vlan);
5362 if (is_gratuitous_arp(flow)) {
5363 /* We don't want to learn from gratuitous ARP packets that are
5364 * reflected back over bond slaves so we lock the learning table. */
5365 if (!in_bundle->bond) {
5366 mac_entry_set_grat_arp_lock(mac);
5367 } else if (mac_entry_is_grat_arp_locked(mac)) {
5372 if (mac_entry_is_new(mac) || mac->port.p != in_bundle) {
5373 /* The log messages here could actually be useful in debugging,
5374 * so keep the rate limit relatively high. */
5375 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(30, 300);
5376 VLOG_DBG_RL(&rl, "bridge %s: learned that "ETH_ADDR_FMT" is "
5377 "on port %s in VLAN %d",
5378 ofproto->up.name, ETH_ADDR_ARGS(flow->dl_src),
5379 in_bundle->name, vlan);
5381 mac->port.p = in_bundle;
5382 tag_set_add(&ofproto->revalidate_set,
5383 mac_learning_changed(ofproto->ml, mac));
5387 static struct ofbundle *
5388 lookup_input_bundle(struct ofproto_dpif *ofproto, uint16_t in_port, bool warn)
5390 struct ofport_dpif *ofport;
5392 /* Special-case OFPP_NONE, which a controller may use as the ingress
5393 * port for traffic that it is sourcing. */
5394 if (in_port == OFPP_NONE) {
5395 return &ofpp_none_bundle;
5398 /* Find the port and bundle for the received packet. */
5399 ofport = get_ofp_port(ofproto, in_port);
5400 if (ofport && ofport->bundle) {
5401 return ofport->bundle;
5404 /* Odd. A few possible reasons here:
5406 * - We deleted a port but there are still a few packets queued up
5409 * - Someone externally added a port (e.g. "ovs-dpctl add-if") that
5410 * we don't know about.
5412 * - The ofproto client didn't configure the port as part of a bundle.
5415 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
5417 VLOG_WARN_RL(&rl, "bridge %s: received packet on unknown "
5418 "port %"PRIu16, ofproto->up.name, in_port);
5423 /* Determines whether packets in 'flow' within 'ofproto' should be forwarded or
5424 * dropped. Returns true if they may be forwarded, false if they should be
5427 * 'in_port' must be the ofport_dpif that corresponds to flow->in_port.
5428 * 'in_port' must be part of a bundle (e.g. in_port->bundle must be nonnull).
5430 * 'vlan' must be the VLAN that corresponds to flow->vlan_tci on 'in_port', as
5431 * returned by input_vid_to_vlan(). It must be a valid VLAN for 'in_port', as
5432 * checked by input_vid_is_valid().
5434 * May also add tags to '*tags', although the current implementation only does
5435 * so in one special case.
5438 is_admissible(struct ofproto_dpif *ofproto, const struct flow *flow,
5439 struct ofport_dpif *in_port, uint16_t vlan, tag_type *tags)
5441 struct ofbundle *in_bundle = in_port->bundle;
5443 /* Drop frames for reserved multicast addresses
5444 * only if forward_bpdu option is absent. */
5445 if (eth_addr_is_reserved(flow->dl_dst) && !ofproto->up.forward_bpdu) {
5449 if (in_bundle->bond) {
5450 struct mac_entry *mac;
5452 switch (bond_check_admissibility(in_bundle->bond, in_port,
5453 flow->dl_dst, tags)) {
5460 case BV_DROP_IF_MOVED:
5461 mac = mac_learning_lookup(ofproto->ml, flow->dl_src, vlan, NULL);
5462 if (mac && mac->port.p != in_bundle &&
5463 (!is_gratuitous_arp(flow)
5464 || mac_entry_is_grat_arp_locked(mac))) {
5475 xlate_normal(struct action_xlate_ctx *ctx)
5477 struct ofport_dpif *in_port;
5478 struct ofbundle *in_bundle;
5479 struct mac_entry *mac;
5483 ctx->has_normal = true;
5485 in_bundle = lookup_input_bundle(ctx->ofproto, ctx->flow.in_port,
5486 ctx->packet != NULL);
5491 /* We know 'in_port' exists unless it is "ofpp_none_bundle",
5492 * since lookup_input_bundle() succeeded. */
5493 in_port = get_ofp_port(ctx->ofproto, ctx->flow.in_port);
5495 /* Drop malformed frames. */
5496 if (ctx->flow.dl_type == htons(ETH_TYPE_VLAN) &&
5497 !(ctx->flow.vlan_tci & htons(VLAN_CFI))) {
5498 if (ctx->packet != NULL) {
5499 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
5500 VLOG_WARN_RL(&rl, "bridge %s: dropping packet with partial "
5501 "VLAN tag received on port %s",
5502 ctx->ofproto->up.name, in_bundle->name);
5507 /* Drop frames on bundles reserved for mirroring. */
5508 if (in_bundle->mirror_out) {
5509 if (ctx->packet != NULL) {
5510 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
5511 VLOG_WARN_RL(&rl, "bridge %s: dropping packet received on port "
5512 "%s, which is reserved exclusively for mirroring",
5513 ctx->ofproto->up.name, in_bundle->name);
5519 vid = vlan_tci_to_vid(ctx->flow.vlan_tci);
5520 if (!input_vid_is_valid(vid, in_bundle, ctx->packet != NULL)) {
5523 vlan = input_vid_to_vlan(in_bundle, vid);
5525 /* Check other admissibility requirements. */
5527 !is_admissible(ctx->ofproto, &ctx->flow, in_port, vlan, &ctx->tags)) {
5531 /* Learn source MAC. */
5532 if (ctx->may_learn) {
5533 update_learning_table(ctx->ofproto, &ctx->flow, vlan, in_bundle);
5536 /* Determine output bundle. */
5537 mac = mac_learning_lookup(ctx->ofproto->ml, ctx->flow.dl_dst, vlan,
5540 if (mac->port.p != in_bundle) {
5541 output_normal(ctx, mac->port.p, vlan);
5544 struct ofbundle *bundle;
5546 HMAP_FOR_EACH (bundle, hmap_node, &ctx->ofproto->bundles) {
5547 if (bundle != in_bundle
5548 && ofbundle_includes_vlan(bundle, vlan)
5549 && bundle->floodable
5550 && !bundle->mirror_out) {
5551 output_normal(ctx, bundle, vlan);
5554 ctx->nf_output_iface = NF_OUT_FLOOD;
5558 /* Optimized flow revalidation.
5560 * It's a difficult problem, in general, to tell which facets need to have
5561 * their actions recalculated whenever the OpenFlow flow table changes. We
5562 * don't try to solve that general problem: for most kinds of OpenFlow flow
5563 * table changes, we recalculate the actions for every facet. This is
5564 * relatively expensive, but it's good enough if the OpenFlow flow table
5565 * doesn't change very often.
5567 * However, we can expect one particular kind of OpenFlow flow table change to
5568 * happen frequently: changes caused by MAC learning. To avoid wasting a lot
5569 * of CPU on revalidating every facet whenever MAC learning modifies the flow
5570 * table, we add a special case that applies to flow tables in which every rule
5571 * has the same form (that is, the same wildcards), except that the table is
5572 * also allowed to have a single "catch-all" flow that matches all packets. We
5573 * optimize this case by tagging all of the facets that resubmit into the table
5574 * and invalidating the same tag whenever a flow changes in that table. The
5575 * end result is that we revalidate just the facets that need it (and sometimes
5576 * a few more, but not all of the facets or even all of the facets that
5577 * resubmit to the table modified by MAC learning). */
5579 /* Calculates the tag to use for 'flow' and wildcards 'wc' when it is inserted
5580 * into an OpenFlow table with the given 'basis'. */
5582 rule_calculate_tag(const struct flow *flow, const struct flow_wildcards *wc,
5585 if (flow_wildcards_is_catchall(wc)) {
5588 struct flow tag_flow = *flow;
5589 flow_zero_wildcards(&tag_flow, wc);
5590 return tag_create_deterministic(flow_hash(&tag_flow, secret));
5594 /* Following a change to OpenFlow table 'table_id' in 'ofproto', update the
5595 * taggability of that table.
5597 * This function must be called after *each* change to a flow table. If you
5598 * skip calling it on some changes then the pointer comparisons at the end can
5599 * be invalid if you get unlucky. For example, if a flow removal causes a
5600 * cls_table to be destroyed and then a flow insertion causes a cls_table with
5601 * different wildcards to be created with the same address, then this function
5602 * will incorrectly skip revalidation. */
5604 table_update_taggable(struct ofproto_dpif *ofproto, uint8_t table_id)
5606 struct table_dpif *table = &ofproto->tables[table_id];
5607 const struct classifier *cls = &ofproto->up.tables[table_id];
5608 struct cls_table *catchall, *other;
5609 struct cls_table *t;
5611 catchall = other = NULL;
5613 switch (hmap_count(&cls->tables)) {
5615 /* We could tag this OpenFlow table but it would make the logic a
5616 * little harder and it's a corner case that doesn't seem worth it
5622 HMAP_FOR_EACH (t, hmap_node, &cls->tables) {
5623 if (cls_table_is_catchall(t)) {
5625 } else if (!other) {
5628 /* Indicate that we can't tag this by setting both tables to
5629 * NULL. (We know that 'catchall' is already NULL.) */
5636 /* Can't tag this table. */
5640 if (table->catchall_table != catchall || table->other_table != other) {
5641 table->catchall_table = catchall;
5642 table->other_table = other;
5643 ofproto->need_revalidate = true;
5647 /* Given 'rule' that has changed in some way (either it is a rule being
5648 * inserted, a rule being deleted, or a rule whose actions are being
5649 * modified), marks facets for revalidation to ensure that packets will be
5650 * forwarded correctly according to the new state of the flow table.
5652 * This function must be called after *each* change to a flow table. See
5653 * the comment on table_update_taggable() for more information. */
5655 rule_invalidate(const struct rule_dpif *rule)
5657 struct ofproto_dpif *ofproto = ofproto_dpif_cast(rule->up.ofproto);
5659 table_update_taggable(ofproto, rule->up.table_id);
5661 if (!ofproto->need_revalidate) {
5662 struct table_dpif *table = &ofproto->tables[rule->up.table_id];
5664 if (table->other_table && rule->tag) {
5665 tag_set_add(&ofproto->revalidate_set, rule->tag);
5667 ofproto->need_revalidate = true;
5673 set_frag_handling(struct ofproto *ofproto_,
5674 enum ofp_config_flags frag_handling)
5676 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
5678 if (frag_handling != OFPC_FRAG_REASM) {
5679 ofproto->need_revalidate = true;
5687 packet_out(struct ofproto *ofproto_, struct ofpbuf *packet,
5688 const struct flow *flow,
5689 const union ofp_action *ofp_actions, size_t n_ofp_actions)
5691 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
5694 if (flow->in_port >= ofproto->max_ports && flow->in_port < OFPP_MAX) {
5695 return OFPERR_NXBRC_BAD_IN_PORT;
5698 error = validate_actions(ofp_actions, n_ofp_actions, flow,
5699 ofproto->max_ports);
5701 struct odputil_keybuf keybuf;
5702 struct ofpbuf *odp_actions;
5703 struct ofproto_push push;
5706 ofpbuf_use_stack(&key, &keybuf, sizeof keybuf);
5707 odp_flow_key_from_flow(&key, flow);
5709 action_xlate_ctx_init(&push.ctx, ofproto, flow, flow->vlan_tci, 0,
5712 /* Ensure that resubmits in 'ofp_actions' get accounted to their
5713 * matching rules. */
5715 push.bytes = packet->size;
5716 push.used = time_msec();
5717 push.ctx.resubmit_hook = push_resubmit;
5719 odp_actions = xlate_actions(&push.ctx, ofp_actions, n_ofp_actions);
5720 dpif_execute(ofproto->dpif, key.data, key.size,
5721 odp_actions->data, odp_actions->size, packet);
5722 ofpbuf_delete(odp_actions);
5730 set_netflow(struct ofproto *ofproto_,
5731 const struct netflow_options *netflow_options)
5733 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
5735 if (netflow_options) {
5736 if (!ofproto->netflow) {
5737 ofproto->netflow = netflow_create();
5739 return netflow_set_options(ofproto->netflow, netflow_options);
5741 netflow_destroy(ofproto->netflow);
5742 ofproto->netflow = NULL;
5748 get_netflow_ids(const struct ofproto *ofproto_,
5749 uint8_t *engine_type, uint8_t *engine_id)
5751 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
5753 dpif_get_netflow_ids(ofproto->dpif, engine_type, engine_id);
5757 send_active_timeout(struct ofproto_dpif *ofproto, struct facet *facet)
5759 if (!facet_is_controller_flow(facet) &&
5760 netflow_active_timeout_expired(ofproto->netflow, &facet->nf_flow)) {
5761 struct subfacet *subfacet;
5762 struct ofexpired expired;
5764 LIST_FOR_EACH (subfacet, list_node, &facet->subfacets) {
5765 if (subfacet->installed) {
5766 struct dpif_flow_stats stats;
5768 subfacet_install(subfacet, subfacet->actions,
5769 subfacet->actions_len, &stats);
5770 subfacet_update_stats(subfacet, &stats);
5774 expired.flow = facet->flow;
5775 expired.packet_count = facet->packet_count;
5776 expired.byte_count = facet->byte_count;
5777 expired.used = facet->used;
5778 netflow_expire(ofproto->netflow, &facet->nf_flow, &expired);
5783 send_netflow_active_timeouts(struct ofproto_dpif *ofproto)
5785 struct facet *facet;
5787 HMAP_FOR_EACH (facet, hmap_node, &ofproto->facets) {
5788 send_active_timeout(ofproto, facet);
5792 static struct ofproto_dpif *
5793 ofproto_dpif_lookup(const char *name)
5795 struct ofproto_dpif *ofproto;
5797 HMAP_FOR_EACH_WITH_HASH (ofproto, all_ofproto_dpifs_node,
5798 hash_string(name, 0), &all_ofproto_dpifs) {
5799 if (!strcmp(ofproto->up.name, name)) {
5807 ofproto_unixctl_fdb_flush(struct unixctl_conn *conn, int argc OVS_UNUSED,
5808 const char *argv[], void *aux OVS_UNUSED)
5810 const struct ofproto_dpif *ofproto;
5812 ofproto = ofproto_dpif_lookup(argv[1]);
5814 unixctl_command_reply(conn, 501, "no such bridge");
5817 mac_learning_flush(ofproto->ml);
5819 unixctl_command_reply(conn, 200, "table successfully flushed");
5823 ofproto_unixctl_fdb_show(struct unixctl_conn *conn, int argc OVS_UNUSED,
5824 const char *argv[], void *aux OVS_UNUSED)
5826 struct ds ds = DS_EMPTY_INITIALIZER;
5827 const struct ofproto_dpif *ofproto;
5828 const struct mac_entry *e;
5830 ofproto = ofproto_dpif_lookup(argv[1]);
5832 unixctl_command_reply(conn, 501, "no such bridge");
5836 ds_put_cstr(&ds, " port VLAN MAC Age\n");
5837 LIST_FOR_EACH (e, lru_node, &ofproto->ml->lrus) {
5838 struct ofbundle *bundle = e->port.p;
5839 ds_put_format(&ds, "%5d %4d "ETH_ADDR_FMT" %3d\n",
5840 ofbundle_get_a_port(bundle)->odp_port,
5841 e->vlan, ETH_ADDR_ARGS(e->mac), mac_entry_age(e));
5843 unixctl_command_reply(conn, 200, ds_cstr(&ds));
5847 struct ofproto_trace {
5848 struct action_xlate_ctx ctx;
5854 trace_format_rule(struct ds *result, uint8_t table_id, int level,
5855 const struct rule_dpif *rule)
5857 ds_put_char_multiple(result, '\t', level);
5859 ds_put_cstr(result, "No match\n");
5863 ds_put_format(result, "Rule: table=%"PRIu8" cookie=%#"PRIx64" ",
5864 table_id, ntohll(rule->up.flow_cookie));
5865 cls_rule_format(&rule->up.cr, result);
5866 ds_put_char(result, '\n');
5868 ds_put_char_multiple(result, '\t', level);
5869 ds_put_cstr(result, "OpenFlow ");
5870 ofp_print_actions(result, rule->up.actions, rule->up.n_actions);
5871 ds_put_char(result, '\n');
5875 trace_format_flow(struct ds *result, int level, const char *title,
5876 struct ofproto_trace *trace)
5878 ds_put_char_multiple(result, '\t', level);
5879 ds_put_format(result, "%s: ", title);
5880 if (flow_equal(&trace->ctx.flow, &trace->flow)) {
5881 ds_put_cstr(result, "unchanged");
5883 flow_format(result, &trace->ctx.flow);
5884 trace->flow = trace->ctx.flow;
5886 ds_put_char(result, '\n');
5890 trace_format_regs(struct ds *result, int level, const char *title,
5891 struct ofproto_trace *trace)
5895 ds_put_char_multiple(result, '\t', level);
5896 ds_put_format(result, "%s:", title);
5897 for (i = 0; i < FLOW_N_REGS; i++) {
5898 ds_put_format(result, " reg%zu=0x%"PRIx32, i, trace->flow.regs[i]);
5900 ds_put_char(result, '\n');
5904 trace_resubmit(struct action_xlate_ctx *ctx, struct rule_dpif *rule)
5906 struct ofproto_trace *trace = CONTAINER_OF(ctx, struct ofproto_trace, ctx);
5907 struct ds *result = trace->result;
5909 ds_put_char(result, '\n');
5910 trace_format_flow(result, ctx->recurse + 1, "Resubmitted flow", trace);
5911 trace_format_regs(result, ctx->recurse + 1, "Resubmitted regs", trace);
5912 trace_format_rule(result, ctx->table_id, ctx->recurse + 1, rule);
5916 ofproto_unixctl_trace(struct unixctl_conn *conn, int argc, const char *argv[],
5917 void *aux OVS_UNUSED)
5919 const char *dpname = argv[1];
5920 struct ofproto_dpif *ofproto;
5921 struct ofpbuf odp_key;
5922 struct ofpbuf *packet;
5923 struct rule_dpif *rule;
5924 ovs_be16 initial_tci;
5930 ofpbuf_init(&odp_key, 0);
5933 ofproto = ofproto_dpif_lookup(dpname);
5935 unixctl_command_reply(conn, 501, "Unknown ofproto (use ofproto/list "
5939 if (argc == 3 || (argc == 4 && !strcmp(argv[3], "-generate"))) {
5940 /* ofproto/trace dpname flow [-generate] */
5941 const char *flow_s = argv[2];
5942 const char *generate_s = argv[3];
5945 /* Convert string to datapath key. */
5946 ofpbuf_init(&odp_key, 0);
5947 error = odp_flow_key_from_string(flow_s, NULL, &odp_key);
5949 unixctl_command_reply(conn, 501, "Bad flow syntax");
5953 /* Convert odp_key to flow. */
5954 error = ofproto_dpif_extract_flow_key(ofproto, odp_key.data,
5955 odp_key.size, &flow,
5956 &initial_tci, NULL);
5957 if (error == ODP_FIT_ERROR) {
5958 unixctl_command_reply(conn, 501, "Invalid flow");
5962 /* Generate a packet, if requested. */
5964 packet = ofpbuf_new(0);
5965 flow_compose(packet, &flow);
5967 } else if (argc == 6) {
5968 /* ofproto/trace dpname priority tun_id in_port packet */
5969 const char *priority_s = argv[2];
5970 const char *tun_id_s = argv[3];
5971 const char *in_port_s = argv[4];
5972 const char *packet_s = argv[5];
5973 uint16_t in_port = ofp_port_to_odp_port(atoi(in_port_s));
5974 ovs_be64 tun_id = htonll(strtoull(tun_id_s, NULL, 0));
5975 uint32_t priority = atoi(priority_s);
5978 msg = eth_from_hex(packet_s, &packet);
5980 unixctl_command_reply(conn, 501, msg);
5984 ds_put_cstr(&result, "Packet: ");
5985 s = ofp_packet_to_string(packet->data, packet->size);
5986 ds_put_cstr(&result, s);
5989 flow_extract(packet, priority, tun_id, in_port, &flow);
5990 initial_tci = flow.vlan_tci;
5992 unixctl_command_reply(conn, 501, "Bad command syntax");
5996 ds_put_cstr(&result, "Flow: ");
5997 flow_format(&result, &flow);
5998 ds_put_char(&result, '\n');
6000 rule = rule_dpif_lookup(ofproto, &flow, 0);
6001 trace_format_rule(&result, 0, 0, rule);
6003 struct ofproto_trace trace;
6004 struct ofpbuf *odp_actions;
6006 trace.result = &result;
6008 action_xlate_ctx_init(&trace.ctx, ofproto, &flow, initial_tci,
6009 rule->up.flow_cookie, packet);
6010 trace.ctx.resubmit_hook = trace_resubmit;
6011 odp_actions = xlate_actions(&trace.ctx,
6012 rule->up.actions, rule->up.n_actions);
6014 ds_put_char(&result, '\n');
6015 trace_format_flow(&result, 0, "Final flow", &trace);
6016 ds_put_cstr(&result, "Datapath actions: ");
6017 format_odp_actions(&result, odp_actions->data, odp_actions->size);
6018 ofpbuf_delete(odp_actions);
6020 if (!trace.ctx.may_set_up_flow) {
6022 ds_put_cstr(&result, "\nThis flow is not cachable.");
6024 ds_put_cstr(&result, "\nThe datapath actions are incomplete--"
6025 "for complete actions, please supply a packet.");
6030 unixctl_command_reply(conn, 200, ds_cstr(&result));
6033 ds_destroy(&result);
6034 ofpbuf_delete(packet);
6035 ofpbuf_uninit(&odp_key);
6039 ofproto_dpif_clog(struct unixctl_conn *conn OVS_UNUSED, int argc OVS_UNUSED,
6040 const char *argv[] OVS_UNUSED, void *aux OVS_UNUSED)
6043 unixctl_command_reply(conn, 200, NULL);
6047 ofproto_dpif_unclog(struct unixctl_conn *conn OVS_UNUSED, int argc OVS_UNUSED,
6048 const char *argv[] OVS_UNUSED, void *aux OVS_UNUSED)
6051 unixctl_command_reply(conn, 200, NULL);
6054 /* Runs a self-check of flow translations in 'ofproto'. Appends a message to
6055 * 'reply' describing the results. */
6057 ofproto_dpif_self_check__(struct ofproto_dpif *ofproto, struct ds *reply)
6059 struct facet *facet;
6063 HMAP_FOR_EACH (facet, hmap_node, &ofproto->facets) {
6064 if (!facet_check_consistency(facet)) {
6069 ofproto->need_revalidate = true;
6073 ds_put_format(reply, "%s: self-check failed (%d errors)\n",
6074 ofproto->up.name, errors);
6076 ds_put_format(reply, "%s: self-check passed\n", ofproto->up.name);
6081 ofproto_dpif_self_check(struct unixctl_conn *conn,
6082 int argc, const char *argv[], void *aux OVS_UNUSED)
6084 struct ds reply = DS_EMPTY_INITIALIZER;
6085 struct ofproto_dpif *ofproto;
6088 ofproto = ofproto_dpif_lookup(argv[1]);
6090 unixctl_command_reply(conn, 501, "Unknown ofproto (use "
6091 "ofproto/list for help)");
6094 ofproto_dpif_self_check__(ofproto, &reply);
6096 HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
6097 ofproto_dpif_self_check__(ofproto, &reply);
6101 unixctl_command_reply(conn, 200, ds_cstr(&reply));
6106 ofproto_dpif_unixctl_init(void)
6108 static bool registered;
6114 unixctl_command_register(
6116 "bridge {tun_id in_port packet | odp_flow [-generate]}",
6117 2, 4, ofproto_unixctl_trace, NULL);
6118 unixctl_command_register("fdb/flush", "bridge", 1, 1,
6119 ofproto_unixctl_fdb_flush, NULL);
6120 unixctl_command_register("fdb/show", "bridge", 1, 1,
6121 ofproto_unixctl_fdb_show, NULL);
6122 unixctl_command_register("ofproto/clog", "", 0, 0,
6123 ofproto_dpif_clog, NULL);
6124 unixctl_command_register("ofproto/unclog", "", 0, 0,
6125 ofproto_dpif_unclog, NULL);
6126 unixctl_command_register("ofproto/self-check", "[bridge]", 0, 1,
6127 ofproto_dpif_self_check, NULL);
6130 /* Linux VLAN device support (e.g. "eth0.10" for VLAN 10.)
6132 * This is deprecated. It is only for compatibility with broken device drivers
6133 * in old versions of Linux that do not properly support VLANs when VLAN
6134 * devices are not used. When broken device drivers are no longer in
6135 * widespread use, we will delete these interfaces. */
6138 set_realdev(struct ofport *ofport_, uint16_t realdev_ofp_port, int vid)
6140 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport_->ofproto);
6141 struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
6143 if (realdev_ofp_port == ofport->realdev_ofp_port
6144 && vid == ofport->vlandev_vid) {
6148 ofproto->need_revalidate = true;
6150 if (ofport->realdev_ofp_port) {
6153 if (realdev_ofp_port && ofport->bundle) {
6154 /* vlandevs are enslaved to their realdevs, so they are not allowed to
6155 * themselves be part of a bundle. */
6156 bundle_set(ofport->up.ofproto, ofport->bundle, NULL);
6159 ofport->realdev_ofp_port = realdev_ofp_port;
6160 ofport->vlandev_vid = vid;
6162 if (realdev_ofp_port) {
6163 vsp_add(ofport, realdev_ofp_port, vid);
6170 hash_realdev_vid(uint16_t realdev_ofp_port, int vid)
6172 return hash_2words(realdev_ofp_port, vid);
6176 vsp_realdev_to_vlandev(const struct ofproto_dpif *ofproto,
6177 uint32_t realdev_odp_port, ovs_be16 vlan_tci)
6179 if (!hmap_is_empty(&ofproto->realdev_vid_map)) {
6180 uint16_t realdev_ofp_port = odp_port_to_ofp_port(realdev_odp_port);
6181 int vid = vlan_tci_to_vid(vlan_tci);
6182 const struct vlan_splinter *vsp;
6184 HMAP_FOR_EACH_WITH_HASH (vsp, realdev_vid_node,
6185 hash_realdev_vid(realdev_ofp_port, vid),
6186 &ofproto->realdev_vid_map) {
6187 if (vsp->realdev_ofp_port == realdev_ofp_port
6188 && vsp->vid == vid) {
6189 return ofp_port_to_odp_port(vsp->vlandev_ofp_port);
6193 return realdev_odp_port;
6196 static struct vlan_splinter *
6197 vlandev_find(const struct ofproto_dpif *ofproto, uint16_t vlandev_ofp_port)
6199 struct vlan_splinter *vsp;
6201 HMAP_FOR_EACH_WITH_HASH (vsp, vlandev_node, hash_int(vlandev_ofp_port, 0),
6202 &ofproto->vlandev_map) {
6203 if (vsp->vlandev_ofp_port == vlandev_ofp_port) {
6212 vsp_vlandev_to_realdev(const struct ofproto_dpif *ofproto,
6213 uint16_t vlandev_ofp_port, int *vid)
6215 if (!hmap_is_empty(&ofproto->vlandev_map)) {
6216 const struct vlan_splinter *vsp;
6218 vsp = vlandev_find(ofproto, vlandev_ofp_port);
6223 return vsp->realdev_ofp_port;
6230 vsp_remove(struct ofport_dpif *port)
6232 struct ofproto_dpif *ofproto = ofproto_dpif_cast(port->up.ofproto);
6233 struct vlan_splinter *vsp;
6235 vsp = vlandev_find(ofproto, port->up.ofp_port);
6237 hmap_remove(&ofproto->vlandev_map, &vsp->vlandev_node);
6238 hmap_remove(&ofproto->realdev_vid_map, &vsp->realdev_vid_node);
6241 port->realdev_ofp_port = 0;
6243 VLOG_ERR("missing vlan device record");
6248 vsp_add(struct ofport_dpif *port, uint16_t realdev_ofp_port, int vid)
6250 struct ofproto_dpif *ofproto = ofproto_dpif_cast(port->up.ofproto);
6252 if (!vsp_vlandev_to_realdev(ofproto, port->up.ofp_port, NULL)
6253 && (vsp_realdev_to_vlandev(ofproto, realdev_ofp_port, htons(vid))
6254 == realdev_ofp_port)) {
6255 struct vlan_splinter *vsp;
6257 vsp = xmalloc(sizeof *vsp);
6258 hmap_insert(&ofproto->vlandev_map, &vsp->vlandev_node,
6259 hash_int(port->up.ofp_port, 0));
6260 hmap_insert(&ofproto->realdev_vid_map, &vsp->realdev_vid_node,
6261 hash_realdev_vid(realdev_ofp_port, vid));
6262 vsp->realdev_ofp_port = realdev_ofp_port;
6263 vsp->vlandev_ofp_port = port->up.ofp_port;
6266 port->realdev_ofp_port = realdev_ofp_port;
6268 VLOG_ERR("duplicate vlan device record");
6272 const struct ofproto_class ofproto_dpif_class = {
6301 port_is_lacp_current,
6302 NULL, /* rule_choose_table */
6309 rule_modify_actions,
6317 get_cfm_remote_mpids,
6321 get_stp_port_status,
6328 is_mirror_output_bundle,
6329 forward_bpdu_changed,