2 * Copyright (c) 2009, 2010, 2011, 2012 Nicira Networks.
4 * Licensed under the Apache License, Version 2.0 (the "License");
5 * you may not use this file except in compliance with the License.
6 * You may obtain a copy of the License at:
8 * http://www.apache.org/licenses/LICENSE-2.0
10 * Unless required by applicable law or agreed to in writing, software
11 * distributed under the License is distributed on an "AS IS" BASIS,
12 * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
13 * See the License for the specific language governing permissions and
14 * limitations under the License.
19 #include "ofproto/ofproto-provider.h"
26 #include "byte-order.h"
31 #include "dynamic-string.h"
32 #include "fail-open.h"
36 #include "mac-learning.h"
37 #include "meta-flow.h"
38 #include "multipath.h"
45 #include "ofp-print.h"
46 #include "ofproto-dpif-sflow.h"
47 #include "poll-loop.h"
49 #include "unaligned.h"
51 #include "vlan-bitmap.h"
54 VLOG_DEFINE_THIS_MODULE(ofproto_dpif);
56 COVERAGE_DEFINE(ofproto_dpif_ctlr_action);
57 COVERAGE_DEFINE(ofproto_dpif_expired);
58 COVERAGE_DEFINE(ofproto_dpif_no_packet_in);
59 COVERAGE_DEFINE(ofproto_dpif_xlate);
60 COVERAGE_DEFINE(facet_changed_rule);
61 COVERAGE_DEFINE(facet_invalidated);
62 COVERAGE_DEFINE(facet_revalidate);
63 COVERAGE_DEFINE(facet_unexpected);
65 /* Maximum depth of flow table recursion (due to resubmit actions) in a
66 * flow translation. */
67 #define MAX_RESUBMIT_RECURSION 32
69 /* Number of implemented OpenFlow tables. */
70 enum { N_TABLES = 255 };
71 BUILD_ASSERT_DECL(N_TABLES >= 1 && N_TABLES <= 255);
81 * - Do include packets and bytes from facets that have been deleted or
82 * whose own statistics have been folded into the rule.
84 * - Do include packets and bytes sent "by hand" that were accounted to
85 * the rule without any facet being involved (this is a rare corner
86 * case in rule_execute()).
88 * - Do not include packet or bytes that can be obtained from any facet's
89 * packet_count or byte_count member or that can be obtained from the
90 * datapath by, e.g., dpif_flow_get() for any subfacet.
92 uint64_t packet_count; /* Number of packets received. */
93 uint64_t byte_count; /* Number of bytes received. */
95 tag_type tag; /* Caches rule_calculate_tag() result. */
97 struct list facets; /* List of "struct facet"s. */
100 static struct rule_dpif *rule_dpif_cast(const struct rule *rule)
102 return rule ? CONTAINER_OF(rule, struct rule_dpif, up) : NULL;
105 static struct rule_dpif *rule_dpif_lookup(struct ofproto_dpif *,
106 const struct flow *, uint8_t table);
108 static void flow_push_stats(struct rule_dpif *, const struct flow *,
109 uint64_t packets, uint64_t bytes,
112 static tag_type rule_calculate_tag(const struct flow *,
113 const struct flow_wildcards *,
115 static void rule_invalidate(const struct rule_dpif *);
117 #define MAX_MIRRORS 32
118 typedef uint32_t mirror_mask_t;
119 #define MIRROR_MASK_C(X) UINT32_C(X)
120 BUILD_ASSERT_DECL(sizeof(mirror_mask_t) * CHAR_BIT >= MAX_MIRRORS);
122 struct ofproto_dpif *ofproto; /* Owning ofproto. */
123 size_t idx; /* In ofproto's "mirrors" array. */
124 void *aux; /* Key supplied by ofproto's client. */
125 char *name; /* Identifier for log messages. */
127 /* Selection criteria. */
128 struct hmapx srcs; /* Contains "struct ofbundle *"s. */
129 struct hmapx dsts; /* Contains "struct ofbundle *"s. */
130 unsigned long *vlans; /* Bitmap of chosen VLANs, NULL selects all. */
132 /* Output (exactly one of out == NULL and out_vlan == -1 is true). */
133 struct ofbundle *out; /* Output port or NULL. */
134 int out_vlan; /* Output VLAN or -1. */
135 mirror_mask_t dup_mirrors; /* Bitmap of mirrors with the same output. */
138 int64_t packet_count; /* Number of packets sent. */
139 int64_t byte_count; /* Number of bytes sent. */
142 static void mirror_destroy(struct ofmirror *);
143 static void update_mirror_stats(struct ofproto_dpif *ofproto,
144 mirror_mask_t mirrors,
145 uint64_t packets, uint64_t bytes);
148 struct ofproto_dpif *ofproto; /* Owning ofproto. */
149 struct hmap_node hmap_node; /* In struct ofproto's "bundles" hmap. */
150 void *aux; /* Key supplied by ofproto's client. */
151 char *name; /* Identifier for log messages. */
154 struct list ports; /* Contains "struct ofport"s. */
155 enum port_vlan_mode vlan_mode; /* VLAN mode */
156 int vlan; /* -1=trunk port, else a 12-bit VLAN ID. */
157 unsigned long *trunks; /* Bitmap of trunked VLANs, if 'vlan' == -1.
158 * NULL if all VLANs are trunked. */
159 struct lacp *lacp; /* LACP if LACP is enabled, otherwise NULL. */
160 struct bond *bond; /* Nonnull iff more than one port. */
161 bool use_priority_tags; /* Use 802.1p tag for frames in VLAN 0? */
164 bool floodable; /* True if no port has OFPUTIL_PC_NO_FLOOD set. */
166 /* Port mirroring info. */
167 mirror_mask_t src_mirrors; /* Mirrors triggered when packet received. */
168 mirror_mask_t dst_mirrors; /* Mirrors triggered when packet sent. */
169 mirror_mask_t mirror_out; /* Mirrors that output to this bundle. */
172 static void bundle_remove(struct ofport *);
173 static void bundle_update(struct ofbundle *);
174 static void bundle_destroy(struct ofbundle *);
175 static void bundle_del_port(struct ofport_dpif *);
176 static void bundle_run(struct ofbundle *);
177 static void bundle_wait(struct ofbundle *);
178 static struct ofbundle *lookup_input_bundle(struct ofproto_dpif *,
179 uint16_t in_port, bool warn);
181 /* A controller may use OFPP_NONE as the ingress port to indicate that
182 * it did not arrive on a "real" port. 'ofpp_none_bundle' exists for
183 * when an input bundle is needed for validation (e.g., mirroring or
184 * OFPP_NORMAL processing). It is not connected to an 'ofproto' or have
185 * any 'port' structs, so care must be taken when dealing with it. */
186 static struct ofbundle ofpp_none_bundle = {
188 .vlan_mode = PORT_VLAN_TRUNK
191 static void stp_run(struct ofproto_dpif *ofproto);
192 static void stp_wait(struct ofproto_dpif *ofproto);
193 static int set_stp_port(struct ofport *,
194 const struct ofproto_port_stp_settings *);
196 static bool ofbundle_includes_vlan(const struct ofbundle *, uint16_t vlan);
198 struct action_xlate_ctx {
199 /* action_xlate_ctx_init() initializes these members. */
202 struct ofproto_dpif *ofproto;
204 /* Flow to which the OpenFlow actions apply. xlate_actions() will modify
205 * this flow when actions change header fields. */
208 /* The packet corresponding to 'flow', or a null pointer if we are
209 * revalidating without a packet to refer to. */
210 const struct ofpbuf *packet;
212 /* Should OFPP_NORMAL update the MAC learning table? We want to update it
213 * if we are actually processing a packet, or if we are accounting for
214 * packets that the datapath has processed, but not if we are just
218 /* Should "learn" actions update the flow table? We want to update it if
219 * we are actually processing a packet, or in most cases if we are
220 * accounting for packets that the datapath has processed, but not if we
221 * are just revalidating. */
224 /* The rule that we are currently translating, or NULL. */
225 struct rule_dpif *rule;
227 /* Union of the set of TCP flags seen so far in this flow. (Used only by
228 * NXAST_FIN_TIMEOUT. Set to zero to avoid updating updating rules'
232 /* If nonnull, called just before executing a resubmit action. In
233 * addition, disables logging of traces when the recursion depth is
236 * This is normally null so the client has to set it manually after
237 * calling action_xlate_ctx_init(). */
238 void (*resubmit_hook)(struct action_xlate_ctx *, struct rule_dpif *);
240 /* xlate_actions() initializes and uses these members. The client might want
241 * to look at them after it returns. */
243 struct ofpbuf *odp_actions; /* Datapath actions. */
244 tag_type tags; /* Tags associated with actions. */
245 bool may_set_up_flow; /* True ordinarily; false if the actions must
246 * be reassessed for every packet. */
247 bool has_learn; /* Actions include NXAST_LEARN? */
248 bool has_normal; /* Actions output to OFPP_NORMAL? */
249 bool has_fin_timeout; /* Actions include NXAST_FIN_TIMEOUT? */
250 uint16_t nf_output_iface; /* Output interface index for NetFlow. */
251 mirror_mask_t mirrors; /* Bitmap of associated mirrors. */
253 /* xlate_actions() initializes and uses these members, but the client has no
254 * reason to look at them. */
256 int recurse; /* Recursion level, via xlate_table_action. */
257 bool max_resubmit_trigger; /* Recursed too deeply during translation. */
258 struct flow base_flow; /* Flow at the last commit. */
259 uint32_t orig_skb_priority; /* Priority when packet arrived. */
260 uint8_t table_id; /* OpenFlow table ID where flow was found. */
261 uint32_t sflow_n_outputs; /* Number of output ports. */
262 uint16_t sflow_odp_port; /* Output port for composing sFlow action. */
263 uint16_t user_cookie_offset;/* Used for user_action_cookie fixup. */
264 bool exit; /* No further actions should be processed. */
267 static void action_xlate_ctx_init(struct action_xlate_ctx *,
268 struct ofproto_dpif *, const struct flow *,
269 ovs_be16 initial_tci, struct rule_dpif *,
270 uint8_t tcp_flags, const struct ofpbuf *);
271 static struct ofpbuf *xlate_actions(struct action_xlate_ctx *,
272 const union ofp_action *in, size_t n_in);
274 /* An exact-match instantiation of an OpenFlow flow.
276 * A facet associates a "struct flow", which represents the Open vSwitch
277 * userspace idea of an exact-match flow, with one or more subfacets. Each
278 * subfacet tracks the datapath's idea of the exact-match flow equivalent to
279 * the facet. When the kernel module (or other dpif implementation) and Open
280 * vSwitch userspace agree on the definition of a flow key, there is exactly
281 * one subfacet per facet. If the dpif implementation supports more-specific
282 * flow matching than userspace, however, a facet can have more than one
283 * subfacet, each of which corresponds to some distinction in flow that
284 * userspace simply doesn't understand.
286 * Flow expiration works in terms of subfacets, so a facet must have at least
287 * one subfacet or it will never expire, leaking memory. */
290 struct hmap_node hmap_node; /* In owning ofproto's 'facets' hmap. */
291 struct list list_node; /* In owning rule's 'facets' list. */
292 struct rule_dpif *rule; /* Owning rule. */
295 struct list subfacets;
296 long long int used; /* Time last used; time created if not used. */
303 * - Do include packets and bytes sent "by hand", e.g. with
306 * - Do include packets and bytes that were obtained from the datapath
307 * when a subfacet's statistics were reset (e.g. dpif_flow_put() with
308 * DPIF_FP_ZERO_STATS).
310 * - Do not include packets or bytes that can be obtained from the
311 * datapath for any existing subfacet.
313 uint64_t packet_count; /* Number of packets received. */
314 uint64_t byte_count; /* Number of bytes received. */
316 /* Resubmit statistics. */
317 uint64_t prev_packet_count; /* Number of packets from last stats push. */
318 uint64_t prev_byte_count; /* Number of bytes from last stats push. */
319 long long int prev_used; /* Used time from last stats push. */
322 uint64_t accounted_bytes; /* Bytes processed by facet_account(). */
323 struct netflow_flow nf_flow; /* Per-flow NetFlow tracking data. */
324 uint8_t tcp_flags; /* TCP flags seen for this 'rule'. */
326 /* Properties of datapath actions.
328 * Every subfacet has its own actions because actions can differ slightly
329 * between splintered and non-splintered subfacets due to the VLAN tag
330 * being initially different (present vs. absent). All of them have these
331 * properties in common so we just store one copy of them here. */
332 bool may_install; /* Reassess actions for every packet? */
333 bool has_learn; /* Actions include NXAST_LEARN? */
334 bool has_normal; /* Actions output to OFPP_NORMAL? */
335 bool has_fin_timeout; /* Actions include NXAST_FIN_TIMEOUT? */
336 tag_type tags; /* Tags that would require revalidation. */
337 mirror_mask_t mirrors; /* Bitmap of dependent mirrors. */
340 static struct facet *facet_create(struct rule_dpif *, const struct flow *);
341 static void facet_remove(struct facet *);
342 static void facet_free(struct facet *);
344 static struct facet *facet_find(struct ofproto_dpif *, const struct flow *);
345 static struct facet *facet_lookup_valid(struct ofproto_dpif *,
346 const struct flow *);
347 static bool facet_revalidate(struct facet *);
348 static bool facet_check_consistency(struct facet *);
350 static void facet_flush_stats(struct facet *);
352 static void facet_update_time(struct facet *, long long int used);
353 static void facet_reset_counters(struct facet *);
354 static void facet_push_stats(struct facet *);
355 static void facet_account(struct facet *, bool may_flow_mod);
357 static bool facet_is_controller_flow(struct facet *);
359 /* A dpif flow and actions associated with a facet.
361 * See also the large comment on struct facet. */
364 struct hmap_node hmap_node; /* In struct ofproto_dpif 'subfacets' list. */
365 struct list list_node; /* In struct facet's 'facets' list. */
366 struct facet *facet; /* Owning facet. */
370 * To save memory in the common case, 'key' is NULL if 'key_fitness' is
371 * ODP_FIT_PERFECT, that is, odp_flow_key_from_flow() can accurately
372 * regenerate the ODP flow key from ->facet->flow. */
373 enum odp_key_fitness key_fitness;
377 long long int used; /* Time last used; time created if not used. */
379 uint64_t dp_packet_count; /* Last known packet count in the datapath. */
380 uint64_t dp_byte_count; /* Last known byte count in the datapath. */
384 * These should be essentially identical for every subfacet in a facet, but
385 * may differ in trivial ways due to VLAN splinters. */
386 size_t actions_len; /* Number of bytes in actions[]. */
387 struct nlattr *actions; /* Datapath actions. */
389 bool installed; /* Installed in datapath? */
391 /* This value is normally the same as ->facet->flow.vlan_tci. Only VLAN
392 * splinters can cause it to differ. This value should be removed when
393 * the VLAN splinters feature is no longer needed. */
394 ovs_be16 initial_tci; /* Initial VLAN TCI value. */
397 static struct subfacet *subfacet_create(struct facet *, enum odp_key_fitness,
398 const struct nlattr *key,
399 size_t key_len, ovs_be16 initial_tci);
400 static struct subfacet *subfacet_find(struct ofproto_dpif *,
401 const struct nlattr *key, size_t key_len);
402 static void subfacet_destroy(struct subfacet *);
403 static void subfacet_destroy__(struct subfacet *);
404 static void subfacet_get_key(struct subfacet *, struct odputil_keybuf *,
406 static void subfacet_reset_dp_stats(struct subfacet *,
407 struct dpif_flow_stats *);
408 static void subfacet_update_time(struct subfacet *, long long int used);
409 static void subfacet_update_stats(struct subfacet *,
410 const struct dpif_flow_stats *);
411 static void subfacet_make_actions(struct subfacet *,
412 const struct ofpbuf *packet);
413 static int subfacet_install(struct subfacet *,
414 const struct nlattr *actions, size_t actions_len,
415 struct dpif_flow_stats *);
416 static void subfacet_uninstall(struct subfacet *);
422 struct ofbundle *bundle; /* Bundle that contains this port, if any. */
423 struct list bundle_node; /* In struct ofbundle's "ports" list. */
424 struct cfm *cfm; /* Connectivity Fault Management, if any. */
425 tag_type tag; /* Tag associated with this port. */
426 uint32_t bond_stable_id; /* stable_id to use as bond slave, or 0. */
427 bool may_enable; /* May be enabled in bonds. */
428 long long int carrier_seq; /* Carrier status changes. */
431 struct stp_port *stp_port; /* Spanning Tree Protocol, if any. */
432 enum stp_state stp_state; /* Always STP_DISABLED if STP not in use. */
433 long long int stp_state_entered;
435 struct hmap priorities; /* Map of attached 'priority_to_dscp's. */
437 /* Linux VLAN device support (e.g. "eth0.10" for VLAN 10.)
439 * This is deprecated. It is only for compatibility with broken device
440 * drivers in old versions of Linux that do not properly support VLANs when
441 * VLAN devices are not used. When broken device drivers are no longer in
442 * widespread use, we will delete these interfaces. */
443 uint16_t realdev_ofp_port;
447 /* Node in 'ofport_dpif''s 'priorities' map. Used to maintain a map from
448 * 'priority' (the datapath's term for QoS queue) to the dscp bits which all
449 * traffic egressing the 'ofport' with that priority should be marked with. */
450 struct priority_to_dscp {
451 struct hmap_node hmap_node; /* Node in 'ofport_dpif''s 'priorities' map. */
452 uint32_t priority; /* Priority of this queue (see struct flow). */
454 uint8_t dscp; /* DSCP bits to mark outgoing traffic with. */
457 /* Linux VLAN device support (e.g. "eth0.10" for VLAN 10.)
459 * This is deprecated. It is only for compatibility with broken device drivers
460 * in old versions of Linux that do not properly support VLANs when VLAN
461 * devices are not used. When broken device drivers are no longer in
462 * widespread use, we will delete these interfaces. */
463 struct vlan_splinter {
464 struct hmap_node realdev_vid_node;
465 struct hmap_node vlandev_node;
466 uint16_t realdev_ofp_port;
467 uint16_t vlandev_ofp_port;
471 static uint32_t vsp_realdev_to_vlandev(const struct ofproto_dpif *,
472 uint32_t realdev, ovs_be16 vlan_tci);
473 static uint16_t vsp_vlandev_to_realdev(const struct ofproto_dpif *,
474 uint16_t vlandev, int *vid);
475 static void vsp_remove(struct ofport_dpif *);
476 static void vsp_add(struct ofport_dpif *, uint16_t realdev_ofp_port, int vid);
478 static struct ofport_dpif *
479 ofport_dpif_cast(const struct ofport *ofport)
481 assert(ofport->ofproto->ofproto_class == &ofproto_dpif_class);
482 return ofport ? CONTAINER_OF(ofport, struct ofport_dpif, up) : NULL;
485 static void port_run(struct ofport_dpif *);
486 static void port_wait(struct ofport_dpif *);
487 static int set_cfm(struct ofport *, const struct cfm_settings *);
488 static void ofport_clear_priorities(struct ofport_dpif *);
490 struct dpif_completion {
491 struct list list_node;
492 struct ofoperation *op;
495 /* Extra information about a classifier table.
496 * Currently used just for optimized flow revalidation. */
498 /* If either of these is nonnull, then this table has a form that allows
499 * flows to be tagged to avoid revalidating most flows for the most common
500 * kinds of flow table changes. */
501 struct cls_table *catchall_table; /* Table that wildcards all fields. */
502 struct cls_table *other_table; /* Table with any other wildcard set. */
503 uint32_t basis; /* Keeps each table's tags separate. */
506 struct ofproto_dpif {
507 struct hmap_node all_ofproto_dpifs_node; /* In 'all_ofproto_dpifs'. */
516 struct netflow *netflow;
517 struct dpif_sflow *sflow;
518 struct hmap bundles; /* Contains "struct ofbundle"s. */
519 struct mac_learning *ml;
520 struct ofmirror *mirrors[MAX_MIRRORS];
521 bool has_bonded_bundles;
524 struct timer next_expiration;
528 struct hmap subfacets;
531 struct table_dpif tables[N_TABLES];
532 bool need_revalidate;
533 struct tag_set revalidate_set;
535 /* Support for debugging async flow mods. */
536 struct list completions;
538 bool has_bundle_action; /* True when the first bundle action appears. */
539 struct netdev_stats stats; /* To account packets generated and consumed in
544 long long int stp_last_tick;
546 /* VLAN splinters. */
547 struct hmap realdev_vid_map; /* (realdev,vid) -> vlandev. */
548 struct hmap vlandev_map; /* vlandev -> (realdev,vid). */
551 /* Defer flow mod completion until "ovs-appctl ofproto/unclog"? (Useful only
552 * for debugging the asynchronous flow_mod implementation.) */
555 /* All existing ofproto_dpif instances, indexed by ->up.name. */
556 static struct hmap all_ofproto_dpifs = HMAP_INITIALIZER(&all_ofproto_dpifs);
558 static void ofproto_dpif_unixctl_init(void);
560 static struct ofproto_dpif *
561 ofproto_dpif_cast(const struct ofproto *ofproto)
563 assert(ofproto->ofproto_class == &ofproto_dpif_class);
564 return CONTAINER_OF(ofproto, struct ofproto_dpif, up);
567 static struct ofport_dpif *get_ofp_port(struct ofproto_dpif *,
569 static struct ofport_dpif *get_odp_port(struct ofproto_dpif *,
571 static void ofproto_trace(struct ofproto_dpif *, const struct flow *,
572 const struct ofpbuf *, ovs_be16 initial_tci,
575 /* Packet processing. */
576 static void update_learning_table(struct ofproto_dpif *,
577 const struct flow *, int vlan,
580 #define FLOW_MISS_MAX_BATCH 50
581 static int handle_upcalls(struct ofproto_dpif *, unsigned int max_batch);
583 /* Flow expiration. */
584 static int expire(struct ofproto_dpif *);
587 static void send_netflow_active_timeouts(struct ofproto_dpif *);
590 static int send_packet(const struct ofport_dpif *, struct ofpbuf *packet);
592 compose_sflow_action(const struct ofproto_dpif *, struct ofpbuf *odp_actions,
593 const struct flow *, uint32_t odp_port);
594 static void add_mirror_actions(struct action_xlate_ctx *ctx,
595 const struct flow *flow);
596 /* Global variables. */
597 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
599 /* Factory functions. */
602 enumerate_types(struct sset *types)
604 dp_enumerate_types(types);
608 enumerate_names(const char *type, struct sset *names)
610 return dp_enumerate_names(type, names);
614 del(const char *type, const char *name)
619 error = dpif_open(name, type, &dpif);
621 error = dpif_delete(dpif);
627 /* Basic life-cycle. */
629 static struct ofproto *
632 struct ofproto_dpif *ofproto = xmalloc(sizeof *ofproto);
637 dealloc(struct ofproto *ofproto_)
639 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
644 construct(struct ofproto *ofproto_)
646 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
647 const char *name = ofproto->up.name;
651 error = dpif_create_and_open(name, ofproto->up.type, &ofproto->dpif);
653 VLOG_ERR("failed to open datapath %s: %s", name, strerror(error));
657 ofproto->max_ports = dpif_get_max_ports(ofproto->dpif);
658 ofproto->n_matches = 0;
660 dpif_flow_flush(ofproto->dpif);
661 dpif_recv_purge(ofproto->dpif);
663 error = dpif_recv_set(ofproto->dpif, true);
665 VLOG_ERR("failed to listen on datapath %s: %s", name, strerror(error));
666 dpif_close(ofproto->dpif);
670 ofproto->netflow = NULL;
671 ofproto->sflow = NULL;
673 hmap_init(&ofproto->bundles);
674 ofproto->ml = mac_learning_create(MAC_ENTRY_DEFAULT_IDLE_TIME);
675 for (i = 0; i < MAX_MIRRORS; i++) {
676 ofproto->mirrors[i] = NULL;
678 ofproto->has_bonded_bundles = false;
680 timer_set_duration(&ofproto->next_expiration, 1000);
682 hmap_init(&ofproto->facets);
683 hmap_init(&ofproto->subfacets);
685 for (i = 0; i < N_TABLES; i++) {
686 struct table_dpif *table = &ofproto->tables[i];
688 table->catchall_table = NULL;
689 table->other_table = NULL;
690 table->basis = random_uint32();
692 ofproto->need_revalidate = false;
693 tag_set_init(&ofproto->revalidate_set);
695 list_init(&ofproto->completions);
697 ofproto_dpif_unixctl_init();
699 ofproto->has_bundle_action = false;
701 hmap_init(&ofproto->vlandev_map);
702 hmap_init(&ofproto->realdev_vid_map);
704 hmap_insert(&all_ofproto_dpifs, &ofproto->all_ofproto_dpifs_node,
705 hash_string(ofproto->up.name, 0));
706 memset(&ofproto->stats, 0, sizeof ofproto->stats);
708 ofproto_init_tables(ofproto_, N_TABLES);
714 complete_operations(struct ofproto_dpif *ofproto)
716 struct dpif_completion *c, *next;
718 LIST_FOR_EACH_SAFE (c, next, list_node, &ofproto->completions) {
719 ofoperation_complete(c->op, 0);
720 list_remove(&c->list_node);
726 destruct(struct ofproto *ofproto_)
728 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
729 struct rule_dpif *rule, *next_rule;
730 struct oftable *table;
733 hmap_remove(&all_ofproto_dpifs, &ofproto->all_ofproto_dpifs_node);
734 complete_operations(ofproto);
736 OFPROTO_FOR_EACH_TABLE (table, &ofproto->up) {
737 struct cls_cursor cursor;
739 cls_cursor_init(&cursor, &table->cls, NULL);
740 CLS_CURSOR_FOR_EACH_SAFE (rule, next_rule, up.cr, &cursor) {
741 ofproto_rule_destroy(&rule->up);
745 for (i = 0; i < MAX_MIRRORS; i++) {
746 mirror_destroy(ofproto->mirrors[i]);
749 netflow_destroy(ofproto->netflow);
750 dpif_sflow_destroy(ofproto->sflow);
751 hmap_destroy(&ofproto->bundles);
752 mac_learning_destroy(ofproto->ml);
754 hmap_destroy(&ofproto->facets);
755 hmap_destroy(&ofproto->subfacets);
757 hmap_destroy(&ofproto->vlandev_map);
758 hmap_destroy(&ofproto->realdev_vid_map);
760 dpif_close(ofproto->dpif);
764 run_fast(struct ofproto *ofproto_)
766 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
769 /* Handle one or more batches of upcalls, until there's nothing left to do
770 * or until we do a fixed total amount of work.
772 * We do work in batches because it can be much cheaper to set up a number
773 * of flows and fire off their patches all at once. We do multiple batches
774 * because in some cases handling a packet can cause another packet to be
775 * queued almost immediately as part of the return flow. Both
776 * optimizations can make major improvements on some benchmarks and
777 * presumably for real traffic as well. */
779 while (work < FLOW_MISS_MAX_BATCH) {
780 int retval = handle_upcalls(ofproto, FLOW_MISS_MAX_BATCH - work);
790 run(struct ofproto *ofproto_)
792 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
793 struct ofport_dpif *ofport;
794 struct ofbundle *bundle;
798 complete_operations(ofproto);
800 dpif_run(ofproto->dpif);
802 error = run_fast(ofproto_);
807 if (timer_expired(&ofproto->next_expiration)) {
808 int delay = expire(ofproto);
809 timer_set_duration(&ofproto->next_expiration, delay);
812 if (ofproto->netflow) {
813 if (netflow_run(ofproto->netflow)) {
814 send_netflow_active_timeouts(ofproto);
817 if (ofproto->sflow) {
818 dpif_sflow_run(ofproto->sflow);
821 HMAP_FOR_EACH (ofport, up.hmap_node, &ofproto->up.ports) {
824 HMAP_FOR_EACH (bundle, hmap_node, &ofproto->bundles) {
829 mac_learning_run(ofproto->ml, &ofproto->revalidate_set);
831 /* Now revalidate if there's anything to do. */
832 if (ofproto->need_revalidate
833 || !tag_set_is_empty(&ofproto->revalidate_set)) {
834 struct tag_set revalidate_set = ofproto->revalidate_set;
835 bool revalidate_all = ofproto->need_revalidate;
836 struct facet *facet, *next;
838 /* Clear the revalidation flags. */
839 tag_set_init(&ofproto->revalidate_set);
840 ofproto->need_revalidate = false;
842 HMAP_FOR_EACH_SAFE (facet, next, hmap_node, &ofproto->facets) {
844 || tag_set_intersects(&revalidate_set, facet->tags)) {
845 facet_revalidate(facet);
850 /* Check the consistency of a random facet, to aid debugging. */
851 if (!hmap_is_empty(&ofproto->facets) && !ofproto->need_revalidate) {
854 facet = CONTAINER_OF(hmap_random_node(&ofproto->facets),
855 struct facet, hmap_node);
856 if (!tag_set_intersects(&ofproto->revalidate_set, facet->tags)) {
857 if (!facet_check_consistency(facet)) {
858 ofproto->need_revalidate = true;
867 wait(struct ofproto *ofproto_)
869 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
870 struct ofport_dpif *ofport;
871 struct ofbundle *bundle;
873 if (!clogged && !list_is_empty(&ofproto->completions)) {
874 poll_immediate_wake();
877 dpif_wait(ofproto->dpif);
878 dpif_recv_wait(ofproto->dpif);
879 if (ofproto->sflow) {
880 dpif_sflow_wait(ofproto->sflow);
882 if (!tag_set_is_empty(&ofproto->revalidate_set)) {
883 poll_immediate_wake();
885 HMAP_FOR_EACH (ofport, up.hmap_node, &ofproto->up.ports) {
888 HMAP_FOR_EACH (bundle, hmap_node, &ofproto->bundles) {
891 if (ofproto->netflow) {
892 netflow_wait(ofproto->netflow);
894 mac_learning_wait(ofproto->ml);
896 if (ofproto->need_revalidate) {
897 /* Shouldn't happen, but if it does just go around again. */
898 VLOG_DBG_RL(&rl, "need revalidate in ofproto_wait_cb()");
899 poll_immediate_wake();
901 timer_wait(&ofproto->next_expiration);
906 flush(struct ofproto *ofproto_)
908 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
909 struct facet *facet, *next_facet;
911 HMAP_FOR_EACH_SAFE (facet, next_facet, hmap_node, &ofproto->facets) {
912 /* Mark the facet as not installed so that facet_remove() doesn't
913 * bother trying to uninstall it. There is no point in uninstalling it
914 * individually since we are about to blow away all the facets with
915 * dpif_flow_flush(). */
916 struct subfacet *subfacet;
918 LIST_FOR_EACH (subfacet, list_node, &facet->subfacets) {
919 subfacet->installed = false;
920 subfacet->dp_packet_count = 0;
921 subfacet->dp_byte_count = 0;
925 dpif_flow_flush(ofproto->dpif);
929 get_features(struct ofproto *ofproto_ OVS_UNUSED,
930 bool *arp_match_ip, enum ofputil_action_bitmap *actions)
932 *arp_match_ip = true;
933 *actions = (OFPUTIL_A_OUTPUT |
934 OFPUTIL_A_SET_VLAN_VID |
935 OFPUTIL_A_SET_VLAN_PCP |
936 OFPUTIL_A_STRIP_VLAN |
937 OFPUTIL_A_SET_DL_SRC |
938 OFPUTIL_A_SET_DL_DST |
939 OFPUTIL_A_SET_NW_SRC |
940 OFPUTIL_A_SET_NW_DST |
941 OFPUTIL_A_SET_NW_TOS |
942 OFPUTIL_A_SET_TP_SRC |
943 OFPUTIL_A_SET_TP_DST |
948 get_tables(struct ofproto *ofproto_, struct ofp_table_stats *ots)
950 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
951 struct dpif_dp_stats s;
953 strcpy(ots->name, "classifier");
955 dpif_get_dp_stats(ofproto->dpif, &s);
956 put_32aligned_be64(&ots->lookup_count, htonll(s.n_hit + s.n_missed));
957 put_32aligned_be64(&ots->matched_count,
958 htonll(s.n_hit + ofproto->n_matches));
961 static struct ofport *
964 struct ofport_dpif *port = xmalloc(sizeof *port);
969 port_dealloc(struct ofport *port_)
971 struct ofport_dpif *port = ofport_dpif_cast(port_);
976 port_construct(struct ofport *port_)
978 struct ofport_dpif *port = ofport_dpif_cast(port_);
979 struct ofproto_dpif *ofproto = ofproto_dpif_cast(port->up.ofproto);
981 ofproto->need_revalidate = true;
982 port->odp_port = ofp_port_to_odp_port(port->up.ofp_port);
985 port->tag = tag_create_random();
986 port->may_enable = true;
987 port->stp_port = NULL;
988 port->stp_state = STP_DISABLED;
989 hmap_init(&port->priorities);
990 port->realdev_ofp_port = 0;
991 port->vlandev_vid = 0;
992 port->carrier_seq = netdev_get_carrier_resets(port->up.netdev);
994 if (ofproto->sflow) {
995 dpif_sflow_add_port(ofproto->sflow, port_);
1002 port_destruct(struct ofport *port_)
1004 struct ofport_dpif *port = ofport_dpif_cast(port_);
1005 struct ofproto_dpif *ofproto = ofproto_dpif_cast(port->up.ofproto);
1007 ofproto->need_revalidate = true;
1008 bundle_remove(port_);
1009 set_cfm(port_, NULL);
1010 if (ofproto->sflow) {
1011 dpif_sflow_del_port(ofproto->sflow, port->odp_port);
1014 ofport_clear_priorities(port);
1015 hmap_destroy(&port->priorities);
1019 port_modified(struct ofport *port_)
1021 struct ofport_dpif *port = ofport_dpif_cast(port_);
1023 if (port->bundle && port->bundle->bond) {
1024 bond_slave_set_netdev(port->bundle->bond, port, port->up.netdev);
1029 port_reconfigured(struct ofport *port_, enum ofputil_port_config old_config)
1031 struct ofport_dpif *port = ofport_dpif_cast(port_);
1032 struct ofproto_dpif *ofproto = ofproto_dpif_cast(port->up.ofproto);
1033 enum ofputil_port_config changed = old_config ^ port->up.pp.config;
1035 if (changed & (OFPUTIL_PC_NO_RECV | OFPUTIL_PC_NO_RECV_STP |
1036 OFPUTIL_PC_NO_FWD | OFPUTIL_PC_NO_FLOOD)) {
1037 ofproto->need_revalidate = true;
1039 if (changed & OFPUTIL_PC_NO_FLOOD && port->bundle) {
1040 bundle_update(port->bundle);
1046 set_sflow(struct ofproto *ofproto_,
1047 const struct ofproto_sflow_options *sflow_options)
1049 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1050 struct dpif_sflow *ds = ofproto->sflow;
1052 if (sflow_options) {
1054 struct ofport_dpif *ofport;
1056 ds = ofproto->sflow = dpif_sflow_create(ofproto->dpif);
1057 HMAP_FOR_EACH (ofport, up.hmap_node, &ofproto->up.ports) {
1058 dpif_sflow_add_port(ds, &ofport->up);
1060 ofproto->need_revalidate = true;
1062 dpif_sflow_set_options(ds, sflow_options);
1065 dpif_sflow_destroy(ds);
1066 ofproto->need_revalidate = true;
1067 ofproto->sflow = NULL;
1074 set_cfm(struct ofport *ofport_, const struct cfm_settings *s)
1076 struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
1083 struct ofproto_dpif *ofproto;
1085 ofproto = ofproto_dpif_cast(ofport->up.ofproto);
1086 ofproto->need_revalidate = true;
1087 ofport->cfm = cfm_create(netdev_get_name(ofport->up.netdev));
1090 if (cfm_configure(ofport->cfm, s)) {
1096 cfm_destroy(ofport->cfm);
1102 get_cfm_fault(const struct ofport *ofport_)
1104 struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
1106 return ofport->cfm ? cfm_get_fault(ofport->cfm) : -1;
1110 get_cfm_remote_mpids(const struct ofport *ofport_, const uint64_t **rmps,
1113 struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
1116 cfm_get_remote_mpids(ofport->cfm, rmps, n_rmps);
1123 /* Spanning Tree. */
1126 send_bpdu_cb(struct ofpbuf *pkt, int port_num, void *ofproto_)
1128 struct ofproto_dpif *ofproto = ofproto_;
1129 struct stp_port *sp = stp_get_port(ofproto->stp, port_num);
1130 struct ofport_dpif *ofport;
1132 ofport = stp_port_get_aux(sp);
1134 VLOG_WARN_RL(&rl, "%s: cannot send BPDU on unknown port %d",
1135 ofproto->up.name, port_num);
1137 struct eth_header *eth = pkt->l2;
1139 netdev_get_etheraddr(ofport->up.netdev, eth->eth_src);
1140 if (eth_addr_is_zero(eth->eth_src)) {
1141 VLOG_WARN_RL(&rl, "%s: cannot send BPDU on port %d "
1142 "with unknown MAC", ofproto->up.name, port_num);
1144 send_packet(ofport, pkt);
1150 /* Configures STP on 'ofproto_' using the settings defined in 's'. */
1152 set_stp(struct ofproto *ofproto_, const struct ofproto_stp_settings *s)
1154 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1156 /* Only revalidate flows if the configuration changed. */
1157 if (!s != !ofproto->stp) {
1158 ofproto->need_revalidate = true;
1162 if (!ofproto->stp) {
1163 ofproto->stp = stp_create(ofproto_->name, s->system_id,
1164 send_bpdu_cb, ofproto);
1165 ofproto->stp_last_tick = time_msec();
1168 stp_set_bridge_id(ofproto->stp, s->system_id);
1169 stp_set_bridge_priority(ofproto->stp, s->priority);
1170 stp_set_hello_time(ofproto->stp, s->hello_time);
1171 stp_set_max_age(ofproto->stp, s->max_age);
1172 stp_set_forward_delay(ofproto->stp, s->fwd_delay);
1174 struct ofport *ofport;
1176 HMAP_FOR_EACH (ofport, hmap_node, &ofproto->up.ports) {
1177 set_stp_port(ofport, NULL);
1180 stp_destroy(ofproto->stp);
1181 ofproto->stp = NULL;
1188 get_stp_status(struct ofproto *ofproto_, struct ofproto_stp_status *s)
1190 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1194 s->bridge_id = stp_get_bridge_id(ofproto->stp);
1195 s->designated_root = stp_get_designated_root(ofproto->stp);
1196 s->root_path_cost = stp_get_root_path_cost(ofproto->stp);
1205 update_stp_port_state(struct ofport_dpif *ofport)
1207 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport->up.ofproto);
1208 enum stp_state state;
1210 /* Figure out new state. */
1211 state = ofport->stp_port ? stp_port_get_state(ofport->stp_port)
1215 if (ofport->stp_state != state) {
1216 enum ofputil_port_state of_state;
1219 VLOG_DBG_RL(&rl, "port %s: STP state changed from %s to %s",
1220 netdev_get_name(ofport->up.netdev),
1221 stp_state_name(ofport->stp_state),
1222 stp_state_name(state));
1223 if (stp_learn_in_state(ofport->stp_state)
1224 != stp_learn_in_state(state)) {
1225 /* xxx Learning action flows should also be flushed. */
1226 mac_learning_flush(ofproto->ml, &ofproto->revalidate_set);
1228 fwd_change = stp_forward_in_state(ofport->stp_state)
1229 != stp_forward_in_state(state);
1231 ofproto->need_revalidate = true;
1232 ofport->stp_state = state;
1233 ofport->stp_state_entered = time_msec();
1235 if (fwd_change && ofport->bundle) {
1236 bundle_update(ofport->bundle);
1239 /* Update the STP state bits in the OpenFlow port description. */
1240 of_state = ofport->up.pp.state & ~OFPUTIL_PS_STP_MASK;
1241 of_state |= (state == STP_LISTENING ? OFPUTIL_PS_STP_LISTEN
1242 : state == STP_LEARNING ? OFPUTIL_PS_STP_LEARN
1243 : state == STP_FORWARDING ? OFPUTIL_PS_STP_FORWARD
1244 : state == STP_BLOCKING ? OFPUTIL_PS_STP_BLOCK
1246 ofproto_port_set_state(&ofport->up, of_state);
1250 /* Configures STP on 'ofport_' using the settings defined in 's'. The
1251 * caller is responsible for assigning STP port numbers and ensuring
1252 * there are no duplicates. */
1254 set_stp_port(struct ofport *ofport_,
1255 const struct ofproto_port_stp_settings *s)
1257 struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
1258 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport->up.ofproto);
1259 struct stp_port *sp = ofport->stp_port;
1261 if (!s || !s->enable) {
1263 ofport->stp_port = NULL;
1264 stp_port_disable(sp);
1265 update_stp_port_state(ofport);
1268 } else if (sp && stp_port_no(sp) != s->port_num
1269 && ofport == stp_port_get_aux(sp)) {
1270 /* The port-id changed, so disable the old one if it's not
1271 * already in use by another port. */
1272 stp_port_disable(sp);
1275 sp = ofport->stp_port = stp_get_port(ofproto->stp, s->port_num);
1276 stp_port_enable(sp);
1278 stp_port_set_aux(sp, ofport);
1279 stp_port_set_priority(sp, s->priority);
1280 stp_port_set_path_cost(sp, s->path_cost);
1282 update_stp_port_state(ofport);
1288 get_stp_port_status(struct ofport *ofport_,
1289 struct ofproto_port_stp_status *s)
1291 struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
1292 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport->up.ofproto);
1293 struct stp_port *sp = ofport->stp_port;
1295 if (!ofproto->stp || !sp) {
1301 s->port_id = stp_port_get_id(sp);
1302 s->state = stp_port_get_state(sp);
1303 s->sec_in_state = (time_msec() - ofport->stp_state_entered) / 1000;
1304 s->role = stp_port_get_role(sp);
1305 stp_port_get_counts(sp, &s->tx_count, &s->rx_count, &s->error_count);
1311 stp_run(struct ofproto_dpif *ofproto)
1314 long long int now = time_msec();
1315 long long int elapsed = now - ofproto->stp_last_tick;
1316 struct stp_port *sp;
1319 stp_tick(ofproto->stp, MIN(INT_MAX, elapsed));
1320 ofproto->stp_last_tick = now;
1322 while (stp_get_changed_port(ofproto->stp, &sp)) {
1323 struct ofport_dpif *ofport = stp_port_get_aux(sp);
1326 update_stp_port_state(ofport);
1330 if (stp_check_and_reset_fdb_flush(ofproto->stp)) {
1331 mac_learning_flush(ofproto->ml, &ofproto->revalidate_set);
1337 stp_wait(struct ofproto_dpif *ofproto)
1340 poll_timer_wait(1000);
1344 /* Returns true if STP should process 'flow'. */
1346 stp_should_process_flow(const struct flow *flow)
1348 return eth_addr_equals(flow->dl_dst, eth_addr_stp);
1352 stp_process_packet(const struct ofport_dpif *ofport,
1353 const struct ofpbuf *packet)
1355 struct ofpbuf payload = *packet;
1356 struct eth_header *eth = payload.data;
1357 struct stp_port *sp = ofport->stp_port;
1359 /* Sink packets on ports that have STP disabled when the bridge has
1361 if (!sp || stp_port_get_state(sp) == STP_DISABLED) {
1365 /* Trim off padding on payload. */
1366 if (payload.size > ntohs(eth->eth_type) + ETH_HEADER_LEN) {
1367 payload.size = ntohs(eth->eth_type) + ETH_HEADER_LEN;
1370 if (ofpbuf_try_pull(&payload, ETH_HEADER_LEN + LLC_HEADER_LEN)) {
1371 stp_received_bpdu(sp, payload.data, payload.size);
1375 static struct priority_to_dscp *
1376 get_priority(const struct ofport_dpif *ofport, uint32_t priority)
1378 struct priority_to_dscp *pdscp;
1381 hash = hash_int(priority, 0);
1382 HMAP_FOR_EACH_IN_BUCKET (pdscp, hmap_node, hash, &ofport->priorities) {
1383 if (pdscp->priority == priority) {
1391 ofport_clear_priorities(struct ofport_dpif *ofport)
1393 struct priority_to_dscp *pdscp, *next;
1395 HMAP_FOR_EACH_SAFE (pdscp, next, hmap_node, &ofport->priorities) {
1396 hmap_remove(&ofport->priorities, &pdscp->hmap_node);
1402 set_queues(struct ofport *ofport_,
1403 const struct ofproto_port_queue *qdscp_list,
1406 struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
1407 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport->up.ofproto);
1408 struct hmap new = HMAP_INITIALIZER(&new);
1411 for (i = 0; i < n_qdscp; i++) {
1412 struct priority_to_dscp *pdscp;
1416 dscp = (qdscp_list[i].dscp << 2) & IP_DSCP_MASK;
1417 if (dpif_queue_to_priority(ofproto->dpif, qdscp_list[i].queue,
1422 pdscp = get_priority(ofport, priority);
1424 hmap_remove(&ofport->priorities, &pdscp->hmap_node);
1426 pdscp = xmalloc(sizeof *pdscp);
1427 pdscp->priority = priority;
1429 ofproto->need_revalidate = true;
1432 if (pdscp->dscp != dscp) {
1434 ofproto->need_revalidate = true;
1437 hmap_insert(&new, &pdscp->hmap_node, hash_int(pdscp->priority, 0));
1440 if (!hmap_is_empty(&ofport->priorities)) {
1441 ofport_clear_priorities(ofport);
1442 ofproto->need_revalidate = true;
1445 hmap_swap(&new, &ofport->priorities);
1453 /* Expires all MAC learning entries associated with 'bundle' and forces its
1454 * ofproto to revalidate every flow.
1456 * Normally MAC learning entries are removed only from the ofproto associated
1457 * with 'bundle', but if 'all_ofprotos' is true, then the MAC learning entries
1458 * are removed from every ofproto. When patch ports and SLB bonds are in use
1459 * and a VM migration happens and the gratuitous ARPs are somehow lost, this
1460 * avoids a MAC_ENTRY_IDLE_TIME delay before the migrated VM can communicate
1461 * with the host from which it migrated. */
1463 bundle_flush_macs(struct ofbundle *bundle, bool all_ofprotos)
1465 struct ofproto_dpif *ofproto = bundle->ofproto;
1466 struct mac_learning *ml = ofproto->ml;
1467 struct mac_entry *mac, *next_mac;
1469 ofproto->need_revalidate = true;
1470 LIST_FOR_EACH_SAFE (mac, next_mac, lru_node, &ml->lrus) {
1471 if (mac->port.p == bundle) {
1473 struct ofproto_dpif *o;
1475 HMAP_FOR_EACH (o, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
1477 struct mac_entry *e;
1479 e = mac_learning_lookup(o->ml, mac->mac, mac->vlan,
1482 tag_set_add(&o->revalidate_set, e->tag);
1483 mac_learning_expire(o->ml, e);
1489 mac_learning_expire(ml, mac);
1494 static struct ofbundle *
1495 bundle_lookup(const struct ofproto_dpif *ofproto, void *aux)
1497 struct ofbundle *bundle;
1499 HMAP_FOR_EACH_IN_BUCKET (bundle, hmap_node, hash_pointer(aux, 0),
1500 &ofproto->bundles) {
1501 if (bundle->aux == aux) {
1508 /* Looks up each of the 'n_auxes' pointers in 'auxes' as bundles and adds the
1509 * ones that are found to 'bundles'. */
1511 bundle_lookup_multiple(struct ofproto_dpif *ofproto,
1512 void **auxes, size_t n_auxes,
1513 struct hmapx *bundles)
1517 hmapx_init(bundles);
1518 for (i = 0; i < n_auxes; i++) {
1519 struct ofbundle *bundle = bundle_lookup(ofproto, auxes[i]);
1521 hmapx_add(bundles, bundle);
1527 bundle_update(struct ofbundle *bundle)
1529 struct ofport_dpif *port;
1531 bundle->floodable = true;
1532 LIST_FOR_EACH (port, bundle_node, &bundle->ports) {
1533 if (port->up.pp.config & OFPUTIL_PC_NO_FLOOD
1534 || !stp_forward_in_state(port->stp_state)) {
1535 bundle->floodable = false;
1542 bundle_del_port(struct ofport_dpif *port)
1544 struct ofbundle *bundle = port->bundle;
1546 bundle->ofproto->need_revalidate = true;
1548 list_remove(&port->bundle_node);
1549 port->bundle = NULL;
1552 lacp_slave_unregister(bundle->lacp, port);
1555 bond_slave_unregister(bundle->bond, port);
1558 bundle_update(bundle);
1562 bundle_add_port(struct ofbundle *bundle, uint32_t ofp_port,
1563 struct lacp_slave_settings *lacp,
1564 uint32_t bond_stable_id)
1566 struct ofport_dpif *port;
1568 port = get_ofp_port(bundle->ofproto, ofp_port);
1573 if (port->bundle != bundle) {
1574 bundle->ofproto->need_revalidate = true;
1576 bundle_del_port(port);
1579 port->bundle = bundle;
1580 list_push_back(&bundle->ports, &port->bundle_node);
1581 if (port->up.pp.config & OFPUTIL_PC_NO_FLOOD
1582 || !stp_forward_in_state(port->stp_state)) {
1583 bundle->floodable = false;
1587 port->bundle->ofproto->need_revalidate = true;
1588 lacp_slave_register(bundle->lacp, port, lacp);
1591 port->bond_stable_id = bond_stable_id;
1597 bundle_destroy(struct ofbundle *bundle)
1599 struct ofproto_dpif *ofproto;
1600 struct ofport_dpif *port, *next_port;
1607 ofproto = bundle->ofproto;
1608 for (i = 0; i < MAX_MIRRORS; i++) {
1609 struct ofmirror *m = ofproto->mirrors[i];
1611 if (m->out == bundle) {
1613 } else if (hmapx_find_and_delete(&m->srcs, bundle)
1614 || hmapx_find_and_delete(&m->dsts, bundle)) {
1615 ofproto->need_revalidate = true;
1620 LIST_FOR_EACH_SAFE (port, next_port, bundle_node, &bundle->ports) {
1621 bundle_del_port(port);
1624 bundle_flush_macs(bundle, true);
1625 hmap_remove(&ofproto->bundles, &bundle->hmap_node);
1627 free(bundle->trunks);
1628 lacp_destroy(bundle->lacp);
1629 bond_destroy(bundle->bond);
1634 bundle_set(struct ofproto *ofproto_, void *aux,
1635 const struct ofproto_bundle_settings *s)
1637 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1638 bool need_flush = false;
1639 struct ofport_dpif *port;
1640 struct ofbundle *bundle;
1641 unsigned long *trunks;
1647 bundle_destroy(bundle_lookup(ofproto, aux));
1651 assert(s->n_slaves == 1 || s->bond != NULL);
1652 assert((s->lacp != NULL) == (s->lacp_slaves != NULL));
1654 bundle = bundle_lookup(ofproto, aux);
1656 bundle = xmalloc(sizeof *bundle);
1658 bundle->ofproto = ofproto;
1659 hmap_insert(&ofproto->bundles, &bundle->hmap_node,
1660 hash_pointer(aux, 0));
1662 bundle->name = NULL;
1664 list_init(&bundle->ports);
1665 bundle->vlan_mode = PORT_VLAN_TRUNK;
1667 bundle->trunks = NULL;
1668 bundle->use_priority_tags = s->use_priority_tags;
1669 bundle->lacp = NULL;
1670 bundle->bond = NULL;
1672 bundle->floodable = true;
1674 bundle->src_mirrors = 0;
1675 bundle->dst_mirrors = 0;
1676 bundle->mirror_out = 0;
1679 if (!bundle->name || strcmp(s->name, bundle->name)) {
1681 bundle->name = xstrdup(s->name);
1686 if (!bundle->lacp) {
1687 ofproto->need_revalidate = true;
1688 bundle->lacp = lacp_create();
1690 lacp_configure(bundle->lacp, s->lacp);
1692 lacp_destroy(bundle->lacp);
1693 bundle->lacp = NULL;
1696 /* Update set of ports. */
1698 for (i = 0; i < s->n_slaves; i++) {
1699 if (!bundle_add_port(bundle, s->slaves[i],
1700 s->lacp ? &s->lacp_slaves[i] : NULL,
1701 s->bond_stable_ids ? s->bond_stable_ids[i] : 0)) {
1705 if (!ok || list_size(&bundle->ports) != s->n_slaves) {
1706 struct ofport_dpif *next_port;
1708 LIST_FOR_EACH_SAFE (port, next_port, bundle_node, &bundle->ports) {
1709 for (i = 0; i < s->n_slaves; i++) {
1710 if (s->slaves[i] == port->up.ofp_port) {
1715 bundle_del_port(port);
1719 assert(list_size(&bundle->ports) <= s->n_slaves);
1721 if (list_is_empty(&bundle->ports)) {
1722 bundle_destroy(bundle);
1726 /* Set VLAN tagging mode */
1727 if (s->vlan_mode != bundle->vlan_mode
1728 || s->use_priority_tags != bundle->use_priority_tags) {
1729 bundle->vlan_mode = s->vlan_mode;
1730 bundle->use_priority_tags = s->use_priority_tags;
1735 vlan = (s->vlan_mode == PORT_VLAN_TRUNK ? -1
1736 : s->vlan >= 0 && s->vlan <= 4095 ? s->vlan
1738 if (vlan != bundle->vlan) {
1739 bundle->vlan = vlan;
1743 /* Get trunked VLANs. */
1744 switch (s->vlan_mode) {
1745 case PORT_VLAN_ACCESS:
1749 case PORT_VLAN_TRUNK:
1750 trunks = (unsigned long *) s->trunks;
1753 case PORT_VLAN_NATIVE_UNTAGGED:
1754 case PORT_VLAN_NATIVE_TAGGED:
1755 if (vlan != 0 && (!s->trunks
1756 || !bitmap_is_set(s->trunks, vlan)
1757 || bitmap_is_set(s->trunks, 0))) {
1758 /* Force trunking the native VLAN and prohibit trunking VLAN 0. */
1760 trunks = bitmap_clone(s->trunks, 4096);
1762 trunks = bitmap_allocate1(4096);
1764 bitmap_set1(trunks, vlan);
1765 bitmap_set0(trunks, 0);
1767 trunks = (unsigned long *) s->trunks;
1774 if (!vlan_bitmap_equal(trunks, bundle->trunks)) {
1775 free(bundle->trunks);
1776 if (trunks == s->trunks) {
1777 bundle->trunks = vlan_bitmap_clone(trunks);
1779 bundle->trunks = trunks;
1784 if (trunks != s->trunks) {
1789 if (!list_is_short(&bundle->ports)) {
1790 bundle->ofproto->has_bonded_bundles = true;
1792 if (bond_reconfigure(bundle->bond, s->bond)) {
1793 ofproto->need_revalidate = true;
1796 bundle->bond = bond_create(s->bond);
1797 ofproto->need_revalidate = true;
1800 LIST_FOR_EACH (port, bundle_node, &bundle->ports) {
1801 bond_slave_register(bundle->bond, port, port->bond_stable_id,
1805 bond_destroy(bundle->bond);
1806 bundle->bond = NULL;
1809 /* If we changed something that would affect MAC learning, un-learn
1810 * everything on this port and force flow revalidation. */
1812 bundle_flush_macs(bundle, false);
1819 bundle_remove(struct ofport *port_)
1821 struct ofport_dpif *port = ofport_dpif_cast(port_);
1822 struct ofbundle *bundle = port->bundle;
1825 bundle_del_port(port);
1826 if (list_is_empty(&bundle->ports)) {
1827 bundle_destroy(bundle);
1828 } else if (list_is_short(&bundle->ports)) {
1829 bond_destroy(bundle->bond);
1830 bundle->bond = NULL;
1836 send_pdu_cb(void *port_, const void *pdu, size_t pdu_size)
1838 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 10);
1839 struct ofport_dpif *port = port_;
1840 uint8_t ea[ETH_ADDR_LEN];
1843 error = netdev_get_etheraddr(port->up.netdev, ea);
1845 struct ofpbuf packet;
1848 ofpbuf_init(&packet, 0);
1849 packet_pdu = eth_compose(&packet, eth_addr_lacp, ea, ETH_TYPE_LACP,
1851 memcpy(packet_pdu, pdu, pdu_size);
1853 send_packet(port, &packet);
1854 ofpbuf_uninit(&packet);
1856 VLOG_ERR_RL(&rl, "port %s: cannot obtain Ethernet address of iface "
1857 "%s (%s)", port->bundle->name,
1858 netdev_get_name(port->up.netdev), strerror(error));
1863 bundle_send_learning_packets(struct ofbundle *bundle)
1865 struct ofproto_dpif *ofproto = bundle->ofproto;
1866 int error, n_packets, n_errors;
1867 struct mac_entry *e;
1869 error = n_packets = n_errors = 0;
1870 LIST_FOR_EACH (e, lru_node, &ofproto->ml->lrus) {
1871 if (e->port.p != bundle) {
1872 struct ofpbuf *learning_packet;
1873 struct ofport_dpif *port;
1877 /* The assignment to "port" is unnecessary but makes "grep"ing for
1878 * struct ofport_dpif more effective. */
1879 learning_packet = bond_compose_learning_packet(bundle->bond,
1883 ret = send_packet(port, learning_packet);
1884 ofpbuf_delete(learning_packet);
1894 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
1895 VLOG_WARN_RL(&rl, "bond %s: %d errors sending %d gratuitous learning "
1896 "packets, last error was: %s",
1897 bundle->name, n_errors, n_packets, strerror(error));
1899 VLOG_DBG("bond %s: sent %d gratuitous learning packets",
1900 bundle->name, n_packets);
1905 bundle_run(struct ofbundle *bundle)
1908 lacp_run(bundle->lacp, send_pdu_cb);
1911 struct ofport_dpif *port;
1913 LIST_FOR_EACH (port, bundle_node, &bundle->ports) {
1914 bond_slave_set_may_enable(bundle->bond, port, port->may_enable);
1917 bond_run(bundle->bond, &bundle->ofproto->revalidate_set,
1918 lacp_status(bundle->lacp));
1919 if (bond_should_send_learning_packets(bundle->bond)) {
1920 bundle_send_learning_packets(bundle);
1926 bundle_wait(struct ofbundle *bundle)
1929 lacp_wait(bundle->lacp);
1932 bond_wait(bundle->bond);
1939 mirror_scan(struct ofproto_dpif *ofproto)
1943 for (idx = 0; idx < MAX_MIRRORS; idx++) {
1944 if (!ofproto->mirrors[idx]) {
1951 static struct ofmirror *
1952 mirror_lookup(struct ofproto_dpif *ofproto, void *aux)
1956 for (i = 0; i < MAX_MIRRORS; i++) {
1957 struct ofmirror *mirror = ofproto->mirrors[i];
1958 if (mirror && mirror->aux == aux) {
1966 /* Update the 'dup_mirrors' member of each of the ofmirrors in 'ofproto'. */
1968 mirror_update_dups(struct ofproto_dpif *ofproto)
1972 for (i = 0; i < MAX_MIRRORS; i++) {
1973 struct ofmirror *m = ofproto->mirrors[i];
1976 m->dup_mirrors = MIRROR_MASK_C(1) << i;
1980 for (i = 0; i < MAX_MIRRORS; i++) {
1981 struct ofmirror *m1 = ofproto->mirrors[i];
1988 for (j = i + 1; j < MAX_MIRRORS; j++) {
1989 struct ofmirror *m2 = ofproto->mirrors[j];
1991 if (m2 && m1->out == m2->out && m1->out_vlan == m2->out_vlan) {
1992 m1->dup_mirrors |= MIRROR_MASK_C(1) << j;
1993 m2->dup_mirrors |= m1->dup_mirrors;
2000 mirror_set(struct ofproto *ofproto_, void *aux,
2001 const struct ofproto_mirror_settings *s)
2003 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
2004 mirror_mask_t mirror_bit;
2005 struct ofbundle *bundle;
2006 struct ofmirror *mirror;
2007 struct ofbundle *out;
2008 struct hmapx srcs; /* Contains "struct ofbundle *"s. */
2009 struct hmapx dsts; /* Contains "struct ofbundle *"s. */
2012 mirror = mirror_lookup(ofproto, aux);
2014 mirror_destroy(mirror);
2020 idx = mirror_scan(ofproto);
2022 VLOG_WARN("bridge %s: maximum of %d port mirrors reached, "
2024 ofproto->up.name, MAX_MIRRORS, s->name);
2028 mirror = ofproto->mirrors[idx] = xzalloc(sizeof *mirror);
2029 mirror->ofproto = ofproto;
2032 mirror->out_vlan = -1;
2033 mirror->name = NULL;
2036 if (!mirror->name || strcmp(s->name, mirror->name)) {
2038 mirror->name = xstrdup(s->name);
2041 /* Get the new configuration. */
2042 if (s->out_bundle) {
2043 out = bundle_lookup(ofproto, s->out_bundle);
2045 mirror_destroy(mirror);
2051 out_vlan = s->out_vlan;
2053 bundle_lookup_multiple(ofproto, s->srcs, s->n_srcs, &srcs);
2054 bundle_lookup_multiple(ofproto, s->dsts, s->n_dsts, &dsts);
2056 /* If the configuration has not changed, do nothing. */
2057 if (hmapx_equals(&srcs, &mirror->srcs)
2058 && hmapx_equals(&dsts, &mirror->dsts)
2059 && vlan_bitmap_equal(mirror->vlans, s->src_vlans)
2060 && mirror->out == out
2061 && mirror->out_vlan == out_vlan)
2063 hmapx_destroy(&srcs);
2064 hmapx_destroy(&dsts);
2068 hmapx_swap(&srcs, &mirror->srcs);
2069 hmapx_destroy(&srcs);
2071 hmapx_swap(&dsts, &mirror->dsts);
2072 hmapx_destroy(&dsts);
2074 free(mirror->vlans);
2075 mirror->vlans = vlan_bitmap_clone(s->src_vlans);
2078 mirror->out_vlan = out_vlan;
2080 /* Update bundles. */
2081 mirror_bit = MIRROR_MASK_C(1) << mirror->idx;
2082 HMAP_FOR_EACH (bundle, hmap_node, &mirror->ofproto->bundles) {
2083 if (hmapx_contains(&mirror->srcs, bundle)) {
2084 bundle->src_mirrors |= mirror_bit;
2086 bundle->src_mirrors &= ~mirror_bit;
2089 if (hmapx_contains(&mirror->dsts, bundle)) {
2090 bundle->dst_mirrors |= mirror_bit;
2092 bundle->dst_mirrors &= ~mirror_bit;
2095 if (mirror->out == bundle) {
2096 bundle->mirror_out |= mirror_bit;
2098 bundle->mirror_out &= ~mirror_bit;
2102 ofproto->need_revalidate = true;
2103 mac_learning_flush(ofproto->ml, &ofproto->revalidate_set);
2104 mirror_update_dups(ofproto);
2110 mirror_destroy(struct ofmirror *mirror)
2112 struct ofproto_dpif *ofproto;
2113 mirror_mask_t mirror_bit;
2114 struct ofbundle *bundle;
2120 ofproto = mirror->ofproto;
2121 ofproto->need_revalidate = true;
2122 mac_learning_flush(ofproto->ml, &ofproto->revalidate_set);
2124 mirror_bit = MIRROR_MASK_C(1) << mirror->idx;
2125 HMAP_FOR_EACH (bundle, hmap_node, &ofproto->bundles) {
2126 bundle->src_mirrors &= ~mirror_bit;
2127 bundle->dst_mirrors &= ~mirror_bit;
2128 bundle->mirror_out &= ~mirror_bit;
2131 hmapx_destroy(&mirror->srcs);
2132 hmapx_destroy(&mirror->dsts);
2133 free(mirror->vlans);
2135 ofproto->mirrors[mirror->idx] = NULL;
2139 mirror_update_dups(ofproto);
2143 mirror_get_stats(struct ofproto *ofproto_, void *aux,
2144 uint64_t *packets, uint64_t *bytes)
2146 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
2147 struct ofmirror *mirror = mirror_lookup(ofproto, aux);
2150 *packets = *bytes = UINT64_MAX;
2154 *packets = mirror->packet_count;
2155 *bytes = mirror->byte_count;
2161 set_flood_vlans(struct ofproto *ofproto_, unsigned long *flood_vlans)
2163 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
2164 if (mac_learning_set_flood_vlans(ofproto->ml, flood_vlans)) {
2165 mac_learning_flush(ofproto->ml, &ofproto->revalidate_set);
2171 is_mirror_output_bundle(const struct ofproto *ofproto_, void *aux)
2173 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
2174 struct ofbundle *bundle = bundle_lookup(ofproto, aux);
2175 return bundle && bundle->mirror_out != 0;
2179 forward_bpdu_changed(struct ofproto *ofproto_)
2181 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
2182 /* Revalidate cached flows whenever forward_bpdu option changes. */
2183 ofproto->need_revalidate = true;
2187 set_mac_idle_time(struct ofproto *ofproto_, unsigned int idle_time)
2189 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
2190 mac_learning_set_idle_time(ofproto->ml, idle_time);
2195 static struct ofport_dpif *
2196 get_ofp_port(struct ofproto_dpif *ofproto, uint16_t ofp_port)
2198 struct ofport *ofport = ofproto_get_port(&ofproto->up, ofp_port);
2199 return ofport ? ofport_dpif_cast(ofport) : NULL;
2202 static struct ofport_dpif *
2203 get_odp_port(struct ofproto_dpif *ofproto, uint32_t odp_port)
2205 return get_ofp_port(ofproto, odp_port_to_ofp_port(odp_port));
2209 ofproto_port_from_dpif_port(struct ofproto_port *ofproto_port,
2210 struct dpif_port *dpif_port)
2212 ofproto_port->name = dpif_port->name;
2213 ofproto_port->type = dpif_port->type;
2214 ofproto_port->ofp_port = odp_port_to_ofp_port(dpif_port->port_no);
2218 port_run(struct ofport_dpif *ofport)
2220 long long int carrier_seq = netdev_get_carrier_resets(ofport->up.netdev);
2221 bool carrier_changed = carrier_seq != ofport->carrier_seq;
2222 bool enable = netdev_get_carrier(ofport->up.netdev);
2224 ofport->carrier_seq = carrier_seq;
2227 cfm_run(ofport->cfm);
2229 if (cfm_should_send_ccm(ofport->cfm)) {
2230 struct ofpbuf packet;
2232 ofpbuf_init(&packet, 0);
2233 cfm_compose_ccm(ofport->cfm, &packet, ofport->up.pp.hw_addr);
2234 send_packet(ofport, &packet);
2235 ofpbuf_uninit(&packet);
2238 enable = enable && !cfm_get_fault(ofport->cfm)
2239 && cfm_get_opup(ofport->cfm);
2242 if (ofport->bundle) {
2243 enable = enable && lacp_slave_may_enable(ofport->bundle->lacp, ofport);
2244 if (carrier_changed) {
2245 lacp_slave_carrier_changed(ofport->bundle->lacp, ofport);
2249 if (ofport->may_enable != enable) {
2250 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport->up.ofproto);
2252 if (ofproto->has_bundle_action) {
2253 ofproto->need_revalidate = true;
2257 ofport->may_enable = enable;
2261 port_wait(struct ofport_dpif *ofport)
2264 cfm_wait(ofport->cfm);
2269 port_query_by_name(const struct ofproto *ofproto_, const char *devname,
2270 struct ofproto_port *ofproto_port)
2272 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
2273 struct dpif_port dpif_port;
2276 error = dpif_port_query_by_name(ofproto->dpif, devname, &dpif_port);
2278 ofproto_port_from_dpif_port(ofproto_port, &dpif_port);
2284 port_add(struct ofproto *ofproto_, struct netdev *netdev, uint16_t *ofp_portp)
2286 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
2290 error = dpif_port_add(ofproto->dpif, netdev, &odp_port);
2292 *ofp_portp = odp_port_to_ofp_port(odp_port);
2298 port_del(struct ofproto *ofproto_, uint16_t ofp_port)
2300 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
2303 error = dpif_port_del(ofproto->dpif, ofp_port_to_odp_port(ofp_port));
2305 struct ofport_dpif *ofport = get_ofp_port(ofproto, ofp_port);
2307 /* The caller is going to close ofport->up.netdev. If this is a
2308 * bonded port, then the bond is using that netdev, so remove it
2309 * from the bond. The client will need to reconfigure everything
2310 * after deleting ports, so then the slave will get re-added. */
2311 bundle_remove(&ofport->up);
2318 port_get_stats(const struct ofport *ofport_, struct netdev_stats *stats)
2320 struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
2323 error = netdev_get_stats(ofport->up.netdev, stats);
2325 if (!error && ofport->odp_port == OVSP_LOCAL) {
2326 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport->up.ofproto);
2328 /* ofproto->stats.tx_packets represents packets that we created
2329 * internally and sent to some port (e.g. packets sent with
2330 * send_packet()). Account for them as if they had come from
2331 * OFPP_LOCAL and got forwarded. */
2333 if (stats->rx_packets != UINT64_MAX) {
2334 stats->rx_packets += ofproto->stats.tx_packets;
2337 if (stats->rx_bytes != UINT64_MAX) {
2338 stats->rx_bytes += ofproto->stats.tx_bytes;
2341 /* ofproto->stats.rx_packets represents packets that were received on
2342 * some port and we processed internally and dropped (e.g. STP).
2343 * Account fro them as if they had been forwarded to OFPP_LOCAL. */
2345 if (stats->tx_packets != UINT64_MAX) {
2346 stats->tx_packets += ofproto->stats.rx_packets;
2349 if (stats->tx_bytes != UINT64_MAX) {
2350 stats->tx_bytes += ofproto->stats.rx_bytes;
2357 /* Account packets for LOCAL port. */
2359 ofproto_update_local_port_stats(const struct ofproto *ofproto_,
2360 size_t tx_size, size_t rx_size)
2362 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
2365 ofproto->stats.rx_packets++;
2366 ofproto->stats.rx_bytes += rx_size;
2369 ofproto->stats.tx_packets++;
2370 ofproto->stats.tx_bytes += tx_size;
2374 struct port_dump_state {
2375 struct dpif_port_dump dump;
2380 port_dump_start(const struct ofproto *ofproto_, void **statep)
2382 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
2383 struct port_dump_state *state;
2385 *statep = state = xmalloc(sizeof *state);
2386 dpif_port_dump_start(&state->dump, ofproto->dpif);
2387 state->done = false;
2392 port_dump_next(const struct ofproto *ofproto_ OVS_UNUSED, void *state_,
2393 struct ofproto_port *port)
2395 struct port_dump_state *state = state_;
2396 struct dpif_port dpif_port;
2398 if (dpif_port_dump_next(&state->dump, &dpif_port)) {
2399 ofproto_port_from_dpif_port(port, &dpif_port);
2402 int error = dpif_port_dump_done(&state->dump);
2404 return error ? error : EOF;
2409 port_dump_done(const struct ofproto *ofproto_ OVS_UNUSED, void *state_)
2411 struct port_dump_state *state = state_;
2414 dpif_port_dump_done(&state->dump);
2421 port_poll(const struct ofproto *ofproto_, char **devnamep)
2423 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
2424 return dpif_port_poll(ofproto->dpif, devnamep);
2428 port_poll_wait(const struct ofproto *ofproto_)
2430 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
2431 dpif_port_poll_wait(ofproto->dpif);
2435 port_is_lacp_current(const struct ofport *ofport_)
2437 const struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
2438 return (ofport->bundle && ofport->bundle->lacp
2439 ? lacp_slave_is_current(ofport->bundle->lacp, ofport)
2443 /* Upcall handling. */
2445 /* Flow miss batching.
2447 * Some dpifs implement operations faster when you hand them off in a batch.
2448 * To allow batching, "struct flow_miss" queues the dpif-related work needed
2449 * for a given flow. Each "struct flow_miss" corresponds to sending one or
2450 * more packets, plus possibly installing the flow in the dpif.
2452 * So far we only batch the operations that affect flow setup time the most.
2453 * It's possible to batch more than that, but the benefit might be minimal. */
2455 struct hmap_node hmap_node;
2457 enum odp_key_fitness key_fitness;
2458 const struct nlattr *key;
2460 ovs_be16 initial_tci;
2461 struct list packets;
2464 struct flow_miss_op {
2465 struct dpif_op dpif_op;
2466 struct subfacet *subfacet;
2469 /* Sends an OFPT_PACKET_IN message for 'packet' of type OFPR_NO_MATCH to each
2470 * OpenFlow controller as necessary according to their individual
2471 * configurations. */
2473 send_packet_in_miss(struct ofproto_dpif *ofproto, const struct ofpbuf *packet,
2474 const struct flow *flow)
2476 struct ofputil_packet_in pin;
2478 pin.packet = packet->data;
2479 pin.packet_len = packet->size;
2480 pin.reason = OFPR_NO_MATCH;
2481 pin.controller_id = 0;
2486 pin.send_len = 0; /* not used for flow table misses */
2488 flow_get_metadata(flow, &pin.fmd);
2490 /* Registers aren't meaningful on a miss. */
2491 memset(pin.fmd.reg_masks, 0, sizeof pin.fmd.reg_masks);
2493 connmgr_send_packet_in(ofproto->up.connmgr, &pin);
2497 process_special(struct ofproto_dpif *ofproto, const struct flow *flow,
2498 const struct ofpbuf *packet)
2500 struct ofport_dpif *ofport = get_ofp_port(ofproto, flow->in_port);
2506 if (ofport->cfm && cfm_should_process_flow(ofport->cfm, flow)) {
2508 cfm_process_heartbeat(ofport->cfm, packet);
2511 } else if (ofport->bundle && ofport->bundle->lacp
2512 && flow->dl_type == htons(ETH_TYPE_LACP)) {
2514 lacp_process_packet(ofport->bundle->lacp, ofport, packet);
2517 } else if (ofproto->stp && stp_should_process_flow(flow)) {
2519 stp_process_packet(ofport, packet);
2526 static struct flow_miss *
2527 flow_miss_create(struct hmap *todo, const struct flow *flow,
2528 enum odp_key_fitness key_fitness,
2529 const struct nlattr *key, size_t key_len,
2530 ovs_be16 initial_tci)
2532 uint32_t hash = flow_hash(flow, 0);
2533 struct flow_miss *miss;
2535 HMAP_FOR_EACH_WITH_HASH (miss, hmap_node, hash, todo) {
2536 if (flow_equal(&miss->flow, flow)) {
2541 miss = xmalloc(sizeof *miss);
2542 hmap_insert(todo, &miss->hmap_node, hash);
2544 miss->key_fitness = key_fitness;
2546 miss->key_len = key_len;
2547 miss->initial_tci = initial_tci;
2548 list_init(&miss->packets);
2553 handle_flow_miss(struct ofproto_dpif *ofproto, struct flow_miss *miss,
2554 struct flow_miss_op *ops, size_t *n_ops)
2556 const struct flow *flow = &miss->flow;
2557 struct ofpbuf *packet, *next_packet;
2558 struct subfacet *subfacet;
2559 struct facet *facet;
2561 facet = facet_lookup_valid(ofproto, flow);
2563 struct rule_dpif *rule;
2565 rule = rule_dpif_lookup(ofproto, flow, 0);
2567 /* Don't send a packet-in if OFPUTIL_PC_NO_PACKET_IN asserted. */
2568 struct ofport_dpif *port = get_ofp_port(ofproto, flow->in_port);
2570 if (port->up.pp.config & OFPUTIL_PC_NO_PACKET_IN) {
2571 COVERAGE_INC(ofproto_dpif_no_packet_in);
2572 /* XXX install 'drop' flow entry */
2576 VLOG_WARN_RL(&rl, "packet-in on unknown port %"PRIu16,
2580 LIST_FOR_EACH (packet, list_node, &miss->packets) {
2581 send_packet_in_miss(ofproto, packet, flow);
2587 facet = facet_create(rule, flow);
2590 subfacet = subfacet_create(facet,
2591 miss->key_fitness, miss->key, miss->key_len,
2594 LIST_FOR_EACH_SAFE (packet, next_packet, list_node, &miss->packets) {
2595 struct dpif_flow_stats stats;
2596 struct flow_miss_op *op;
2597 struct dpif_execute *execute;
2599 ofproto->n_matches++;
2601 if (facet->rule->up.cr.priority == FAIL_OPEN_PRIORITY) {
2603 * Extra-special case for fail-open mode.
2605 * We are in fail-open mode and the packet matched the fail-open
2606 * rule, but we are connected to a controller too. We should send
2607 * the packet up to the controller in the hope that it will try to
2608 * set up a flow and thereby allow us to exit fail-open.
2610 * See the top-level comment in fail-open.c for more information.
2612 send_packet_in_miss(ofproto, packet, flow);
2615 if (!facet->may_install || !subfacet->actions) {
2616 subfacet_make_actions(subfacet, packet);
2619 dpif_flow_stats_extract(&facet->flow, packet, &stats);
2620 subfacet_update_stats(subfacet, &stats);
2622 if (!subfacet->actions_len) {
2623 /* No actions to execute, so skip talking to the dpif. */
2627 if (flow->vlan_tci != subfacet->initial_tci) {
2628 /* This packet was received on a VLAN splinter port. We added
2629 * a VLAN to the packet to make the packet resemble the flow,
2630 * but the actions were composed assuming that the packet
2631 * contained no VLAN. So, we must remove the VLAN header from
2632 * the packet before trying to execute the actions. */
2633 eth_pop_vlan(packet);
2636 op = &ops[(*n_ops)++];
2637 execute = &op->dpif_op.u.execute;
2638 op->subfacet = subfacet;
2639 op->dpif_op.type = DPIF_OP_EXECUTE;
2640 execute->key = miss->key;
2641 execute->key_len = miss->key_len;
2642 execute->actions = (facet->may_install
2644 : xmemdup(subfacet->actions,
2645 subfacet->actions_len));
2646 execute->actions_len = subfacet->actions_len;
2647 execute->packet = packet;
2650 if (facet->may_install && subfacet->key_fitness != ODP_FIT_TOO_LITTLE) {
2651 struct flow_miss_op *op = &ops[(*n_ops)++];
2652 struct dpif_flow_put *put = &op->dpif_op.u.flow_put;
2654 op->subfacet = subfacet;
2655 op->dpif_op.type = DPIF_OP_FLOW_PUT;
2656 put->flags = DPIF_FP_CREATE | DPIF_FP_MODIFY;
2657 put->key = miss->key;
2658 put->key_len = miss->key_len;
2659 put->actions = subfacet->actions;
2660 put->actions_len = subfacet->actions_len;
2665 /* Like odp_flow_key_to_flow(), this function converts the 'key_len' bytes of
2666 * OVS_KEY_ATTR_* attributes in 'key' to a flow structure in 'flow' and returns
2667 * an ODP_FIT_* value that indicates how well 'key' fits our expectations for
2668 * what a flow key should contain.
2670 * This function also includes some logic to help make VLAN splinters
2671 * transparent to the rest of the upcall processing logic. In particular, if
2672 * the extracted in_port is a VLAN splinter port, it replaces flow->in_port by
2673 * the "real" port, sets flow->vlan_tci correctly for the VLAN of the VLAN
2674 * splinter port, and pushes a VLAN header onto 'packet' (if it is nonnull).
2676 * Sets '*initial_tci' to the VLAN TCI with which the packet was really
2677 * received, that is, the actual VLAN TCI extracted by odp_flow_key_to_flow().
2678 * (This differs from the value returned in flow->vlan_tci only for packets
2679 * received on VLAN splinters.)
2681 static enum odp_key_fitness
2682 ofproto_dpif_extract_flow_key(const struct ofproto_dpif *ofproto,
2683 const struct nlattr *key, size_t key_len,
2684 struct flow *flow, ovs_be16 *initial_tci,
2685 struct ofpbuf *packet)
2687 enum odp_key_fitness fitness;
2691 fitness = odp_flow_key_to_flow(key, key_len, flow);
2692 if (fitness == ODP_FIT_ERROR) {
2695 *initial_tci = flow->vlan_tci;
2697 realdev = vsp_vlandev_to_realdev(ofproto, flow->in_port, &vid);
2699 /* Cause the flow to be processed as if it came in on the real device
2700 * with the VLAN device's VLAN ID. */
2701 flow->in_port = realdev;
2702 flow->vlan_tci = htons((vid & VLAN_VID_MASK) | VLAN_CFI);
2704 /* Make the packet resemble the flow, so that it gets sent to an
2705 * OpenFlow controller properly, so that it looks correct for
2706 * sFlow, and so that flow_extract() will get the correct vlan_tci
2707 * if it is called on 'packet'.
2709 * The allocated space inside 'packet' probably also contains
2710 * 'key', that is, both 'packet' and 'key' are probably part of a
2711 * struct dpif_upcall (see the large comment on that structure
2712 * definition), so pushing data on 'packet' is in general not a
2713 * good idea since it could overwrite 'key' or free it as a side
2714 * effect. However, it's OK in this special case because we know
2715 * that 'packet' is inside a Netlink attribute: pushing 4 bytes
2716 * will just overwrite the 4-byte "struct nlattr", which is fine
2717 * since we don't need that header anymore. */
2718 eth_push_vlan(packet, flow->vlan_tci);
2721 /* Let the caller know that we can't reproduce 'key' from 'flow'. */
2722 if (fitness == ODP_FIT_PERFECT) {
2723 fitness = ODP_FIT_TOO_MUCH;
2731 handle_miss_upcalls(struct ofproto_dpif *ofproto, struct dpif_upcall *upcalls,
2734 struct dpif_upcall *upcall;
2735 struct flow_miss *miss, *next_miss;
2736 struct flow_miss_op flow_miss_ops[FLOW_MISS_MAX_BATCH * 2];
2737 struct dpif_op *dpif_ops[FLOW_MISS_MAX_BATCH * 2];
2746 /* Construct the to-do list.
2748 * This just amounts to extracting the flow from each packet and sticking
2749 * the packets that have the same flow in the same "flow_miss" structure so
2750 * that we can process them together. */
2752 for (upcall = upcalls; upcall < &upcalls[n_upcalls]; upcall++) {
2753 enum odp_key_fitness fitness;
2754 struct flow_miss *miss;
2755 ovs_be16 initial_tci;
2758 /* Obtain metadata and check userspace/kernel agreement on flow match,
2759 * then set 'flow''s header pointers. */
2760 fitness = ofproto_dpif_extract_flow_key(ofproto,
2761 upcall->key, upcall->key_len,
2762 &flow, &initial_tci,
2764 if (fitness == ODP_FIT_ERROR) {
2765 ofpbuf_delete(upcall->packet);
2768 flow_extract(upcall->packet, flow.skb_priority, flow.tun_id,
2769 flow.in_port, &flow);
2771 /* Handle 802.1ag, LACP, and STP specially. */
2772 if (process_special(ofproto, &flow, upcall->packet)) {
2773 ofproto_update_local_port_stats(&ofproto->up,
2774 0, upcall->packet->size);
2775 ofpbuf_delete(upcall->packet);
2776 ofproto->n_matches++;
2780 /* Add other packets to a to-do list. */
2781 miss = flow_miss_create(&todo, &flow, fitness,
2782 upcall->key, upcall->key_len, initial_tci);
2783 list_push_back(&miss->packets, &upcall->packet->list_node);
2786 /* Process each element in the to-do list, constructing the set of
2787 * operations to batch. */
2789 HMAP_FOR_EACH (miss, hmap_node, &todo) {
2790 handle_flow_miss(ofproto, miss, flow_miss_ops, &n_ops);
2792 assert(n_ops <= ARRAY_SIZE(flow_miss_ops));
2794 /* Execute batch. */
2795 for (i = 0; i < n_ops; i++) {
2796 dpif_ops[i] = &flow_miss_ops[i].dpif_op;
2798 dpif_operate(ofproto->dpif, dpif_ops, n_ops);
2800 /* Free memory and update facets. */
2801 for (i = 0; i < n_ops; i++) {
2802 struct flow_miss_op *op = &flow_miss_ops[i];
2803 struct dpif_execute *execute;
2805 switch (op->dpif_op.type) {
2806 case DPIF_OP_EXECUTE:
2807 execute = &op->dpif_op.u.execute;
2808 if (op->subfacet->actions != execute->actions) {
2809 free((struct nlattr *) execute->actions);
2813 case DPIF_OP_FLOW_PUT:
2814 if (!op->dpif_op.error) {
2815 op->subfacet->installed = true;
2820 HMAP_FOR_EACH_SAFE (miss, next_miss, hmap_node, &todo) {
2821 ofpbuf_list_delete(&miss->packets);
2822 hmap_remove(&todo, &miss->hmap_node);
2825 hmap_destroy(&todo);
2829 handle_userspace_upcall(struct ofproto_dpif *ofproto,
2830 struct dpif_upcall *upcall)
2832 struct user_action_cookie cookie;
2833 enum odp_key_fitness fitness;
2834 ovs_be16 initial_tci;
2837 memcpy(&cookie, &upcall->userdata, sizeof(cookie));
2839 fitness = ofproto_dpif_extract_flow_key(ofproto, upcall->key,
2840 upcall->key_len, &flow,
2841 &initial_tci, upcall->packet);
2842 if (fitness == ODP_FIT_ERROR) {
2843 ofpbuf_delete(upcall->packet);
2847 if (cookie.type == USER_ACTION_COOKIE_SFLOW) {
2848 if (ofproto->sflow) {
2849 dpif_sflow_received(ofproto->sflow, upcall->packet, &flow,
2853 VLOG_WARN_RL(&rl, "invalid user cookie : 0x%"PRIx64, upcall->userdata);
2855 ofpbuf_delete(upcall->packet);
2859 handle_upcalls(struct ofproto_dpif *ofproto, unsigned int max_batch)
2861 struct dpif_upcall misses[FLOW_MISS_MAX_BATCH];
2865 assert (max_batch <= FLOW_MISS_MAX_BATCH);
2868 for (i = 0; i < max_batch; i++) {
2869 struct dpif_upcall *upcall = &misses[n_misses];
2872 error = dpif_recv(ofproto->dpif, upcall);
2877 switch (upcall->type) {
2878 case DPIF_UC_ACTION:
2879 handle_userspace_upcall(ofproto, upcall);
2883 /* Handle it later. */
2887 case DPIF_N_UC_TYPES:
2889 VLOG_WARN_RL(&rl, "upcall has unexpected type %"PRIu32,
2895 handle_miss_upcalls(ofproto, misses, n_misses);
2900 /* Flow expiration. */
2902 static int subfacet_max_idle(const struct ofproto_dpif *);
2903 static void update_stats(struct ofproto_dpif *);
2904 static void rule_expire(struct rule_dpif *);
2905 static void expire_subfacets(struct ofproto_dpif *, int dp_max_idle);
2907 /* This function is called periodically by run(). Its job is to collect
2908 * updates for the flows that have been installed into the datapath, most
2909 * importantly when they last were used, and then use that information to
2910 * expire flows that have not been used recently.
2912 * Returns the number of milliseconds after which it should be called again. */
2914 expire(struct ofproto_dpif *ofproto)
2916 struct rule_dpif *rule, *next_rule;
2917 struct oftable *table;
2920 /* Update stats for each flow in the datapath. */
2921 update_stats(ofproto);
2923 /* Expire subfacets that have been idle too long. */
2924 dp_max_idle = subfacet_max_idle(ofproto);
2925 expire_subfacets(ofproto, dp_max_idle);
2927 /* Expire OpenFlow flows whose idle_timeout or hard_timeout has passed. */
2928 OFPROTO_FOR_EACH_TABLE (table, &ofproto->up) {
2929 struct cls_cursor cursor;
2931 cls_cursor_init(&cursor, &table->cls, NULL);
2932 CLS_CURSOR_FOR_EACH_SAFE (rule, next_rule, up.cr, &cursor) {
2937 /* All outstanding data in existing flows has been accounted, so it's a
2938 * good time to do bond rebalancing. */
2939 if (ofproto->has_bonded_bundles) {
2940 struct ofbundle *bundle;
2942 HMAP_FOR_EACH (bundle, hmap_node, &ofproto->bundles) {
2944 bond_rebalance(bundle->bond, &ofproto->revalidate_set);
2949 return MIN(dp_max_idle, 1000);
2952 /* Update 'packet_count', 'byte_count', and 'used' members of installed facets.
2954 * This function also pushes statistics updates to rules which each facet
2955 * resubmits into. Generally these statistics will be accurate. However, if a
2956 * facet changes the rule it resubmits into at some time in between
2957 * update_stats() runs, it is possible that statistics accrued to the
2958 * old rule will be incorrectly attributed to the new rule. This could be
2959 * avoided by calling update_stats() whenever rules are created or
2960 * deleted. However, the performance impact of making so many calls to the
2961 * datapath do not justify the benefit of having perfectly accurate statistics.
2964 update_stats(struct ofproto_dpif *p)
2966 const struct dpif_flow_stats *stats;
2967 struct dpif_flow_dump dump;
2968 const struct nlattr *key;
2971 dpif_flow_dump_start(&dump, p->dpif);
2972 while (dpif_flow_dump_next(&dump, &key, &key_len, NULL, NULL, &stats)) {
2973 struct subfacet *subfacet;
2975 subfacet = subfacet_find(p, key, key_len);
2976 if (subfacet && subfacet->installed) {
2977 struct facet *facet = subfacet->facet;
2979 if (stats->n_packets >= subfacet->dp_packet_count) {
2980 uint64_t extra = stats->n_packets - subfacet->dp_packet_count;
2981 facet->packet_count += extra;
2983 VLOG_WARN_RL(&rl, "unexpected packet count from the datapath");
2986 if (stats->n_bytes >= subfacet->dp_byte_count) {
2987 facet->byte_count += stats->n_bytes - subfacet->dp_byte_count;
2989 VLOG_WARN_RL(&rl, "unexpected byte count from datapath");
2992 subfacet->dp_packet_count = stats->n_packets;
2993 subfacet->dp_byte_count = stats->n_bytes;
2995 facet->tcp_flags |= stats->tcp_flags;
2997 subfacet_update_time(subfacet, stats->used);
2998 facet_account(facet, true);
2999 facet_push_stats(facet);
3001 if (!VLOG_DROP_WARN(&rl)) {
3005 odp_flow_key_format(key, key_len, &s);
3006 VLOG_WARN("unexpected flow from datapath %s", ds_cstr(&s));
3010 COVERAGE_INC(facet_unexpected);
3011 /* There's a flow in the datapath that we know nothing about, or a
3012 * flow that shouldn't be installed but was anyway. Delete it. */
3013 dpif_flow_del(p->dpif, key, key_len, NULL);
3016 dpif_flow_dump_done(&dump);
3019 /* Calculates and returns the number of milliseconds of idle time after which
3020 * subfacets should expire from the datapath. When a subfacet expires, we fold
3021 * its statistics into its facet, and when a facet's last subfacet expires, we
3022 * fold its statistic into its rule. */
3024 subfacet_max_idle(const struct ofproto_dpif *ofproto)
3027 * Idle time histogram.
3029 * Most of the time a switch has a relatively small number of subfacets.
3030 * When this is the case we might as well keep statistics for all of them
3031 * in userspace and to cache them in the kernel datapath for performance as
3034 * As the number of subfacets increases, the memory required to maintain
3035 * statistics about them in userspace and in the kernel becomes
3036 * significant. However, with a large number of subfacets it is likely
3037 * that only a few of them are "heavy hitters" that consume a large amount
3038 * of bandwidth. At this point, only heavy hitters are worth caching in
3039 * the kernel and maintaining in userspaces; other subfacets we can
3042 * The technique used to compute the idle time is to build a histogram with
3043 * N_BUCKETS buckets whose width is BUCKET_WIDTH msecs each. Each subfacet
3044 * that is installed in the kernel gets dropped in the appropriate bucket.
3045 * After the histogram has been built, we compute the cutoff so that only
3046 * the most-recently-used 1% of subfacets (but at least
3047 * ofproto->up.flow_eviction_threshold flows) are kept cached. At least
3048 * the most-recently-used bucket of subfacets is kept, so actually an
3049 * arbitrary number of subfacets can be kept in any given expiration run
3050 * (though the next run will delete most of those unless they receive
3053 * This requires a second pass through the subfacets, in addition to the
3054 * pass made by update_stats(), because the former function never looks at
3055 * uninstallable subfacets.
3057 enum { BUCKET_WIDTH = ROUND_UP(100, TIME_UPDATE_INTERVAL) };
3058 enum { N_BUCKETS = 5000 / BUCKET_WIDTH };
3059 int buckets[N_BUCKETS] = { 0 };
3060 int total, subtotal, bucket;
3061 struct subfacet *subfacet;
3065 total = hmap_count(&ofproto->subfacets);
3066 if (total <= ofproto->up.flow_eviction_threshold) {
3067 return N_BUCKETS * BUCKET_WIDTH;
3070 /* Build histogram. */
3072 HMAP_FOR_EACH (subfacet, hmap_node, &ofproto->subfacets) {
3073 long long int idle = now - subfacet->used;
3074 int bucket = (idle <= 0 ? 0
3075 : idle >= BUCKET_WIDTH * N_BUCKETS ? N_BUCKETS - 1
3076 : (unsigned int) idle / BUCKET_WIDTH);
3080 /* Find the first bucket whose flows should be expired. */
3081 subtotal = bucket = 0;
3083 subtotal += buckets[bucket++];
3084 } while (bucket < N_BUCKETS &&
3085 subtotal < MAX(ofproto->up.flow_eviction_threshold, total / 100));
3087 if (VLOG_IS_DBG_ENABLED()) {
3091 ds_put_cstr(&s, "keep");
3092 for (i = 0; i < N_BUCKETS; i++) {
3094 ds_put_cstr(&s, ", drop");
3097 ds_put_format(&s, " %d:%d", i * BUCKET_WIDTH, buckets[i]);
3100 VLOG_INFO("%s: %s (msec:count)", ofproto->up.name, ds_cstr(&s));
3104 return bucket * BUCKET_WIDTH;
3108 expire_subfacets(struct ofproto_dpif *ofproto, int dp_max_idle)
3110 long long int cutoff = time_msec() - dp_max_idle;
3111 struct subfacet *subfacet, *next_subfacet;
3113 HMAP_FOR_EACH_SAFE (subfacet, next_subfacet, hmap_node,
3114 &ofproto->subfacets) {
3115 if (subfacet->used < cutoff) {
3116 subfacet_destroy(subfacet);
3121 /* If 'rule' is an OpenFlow rule, that has expired according to OpenFlow rules,
3122 * then delete it entirely. */
3124 rule_expire(struct rule_dpif *rule)
3126 struct facet *facet, *next_facet;
3130 /* Has 'rule' expired? */
3132 if (rule->up.hard_timeout
3133 && now > rule->up.modified + rule->up.hard_timeout * 1000) {
3134 reason = OFPRR_HARD_TIMEOUT;
3135 } else if (rule->up.idle_timeout
3136 && now > rule->up.used + rule->up.idle_timeout * 1000) {
3137 reason = OFPRR_IDLE_TIMEOUT;
3142 COVERAGE_INC(ofproto_dpif_expired);
3144 /* Update stats. (This is a no-op if the rule expired due to an idle
3145 * timeout, because that only happens when the rule has no facets left.) */
3146 LIST_FOR_EACH_SAFE (facet, next_facet, list_node, &rule->facets) {
3147 facet_remove(facet);
3150 /* Get rid of the rule. */
3151 ofproto_rule_expire(&rule->up, reason);
3156 /* Creates and returns a new facet owned by 'rule', given a 'flow'.
3158 * The caller must already have determined that no facet with an identical
3159 * 'flow' exists in 'ofproto' and that 'flow' is the best match for 'rule' in
3160 * the ofproto's classifier table.
3162 * The facet will initially have no subfacets. The caller should create (at
3163 * least) one subfacet with subfacet_create(). */
3164 static struct facet *
3165 facet_create(struct rule_dpif *rule, const struct flow *flow)
3167 struct ofproto_dpif *ofproto = ofproto_dpif_cast(rule->up.ofproto);
3168 struct facet *facet;
3170 facet = xzalloc(sizeof *facet);
3171 facet->used = time_msec();
3172 hmap_insert(&ofproto->facets, &facet->hmap_node, flow_hash(flow, 0));
3173 list_push_back(&rule->facets, &facet->list_node);
3175 facet->flow = *flow;
3176 list_init(&facet->subfacets);
3177 netflow_flow_init(&facet->nf_flow);
3178 netflow_flow_update_time(ofproto->netflow, &facet->nf_flow, facet->used);
3184 facet_free(struct facet *facet)
3189 /* Executes, within 'ofproto', the 'n_actions' actions in 'actions' on
3190 * 'packet', which arrived on 'in_port'.
3192 * Takes ownership of 'packet'. */
3194 execute_odp_actions(struct ofproto_dpif *ofproto, const struct flow *flow,
3195 const struct nlattr *odp_actions, size_t actions_len,
3196 struct ofpbuf *packet)
3198 struct odputil_keybuf keybuf;
3202 ofpbuf_use_stack(&key, &keybuf, sizeof keybuf);
3203 odp_flow_key_from_flow(&key, flow);
3205 error = dpif_execute(ofproto->dpif, key.data, key.size,
3206 odp_actions, actions_len, packet);
3208 ofpbuf_delete(packet);
3212 /* Remove 'facet' from 'ofproto' and free up the associated memory:
3214 * - If 'facet' was installed in the datapath, uninstalls it and updates its
3215 * rule's statistics, via subfacet_uninstall().
3217 * - Removes 'facet' from its rule and from ofproto->facets.
3220 facet_remove(struct facet *facet)
3222 struct ofproto_dpif *ofproto = ofproto_dpif_cast(facet->rule->up.ofproto);
3223 struct subfacet *subfacet, *next_subfacet;
3225 assert(!list_is_empty(&facet->subfacets));
3227 /* First uninstall all of the subfacets to get final statistics. */
3228 LIST_FOR_EACH (subfacet, list_node, &facet->subfacets) {
3229 subfacet_uninstall(subfacet);
3232 /* Flush the final stats to the rule.
3234 * This might require us to have at least one subfacet around so that we
3235 * can use its actions for accounting in facet_account(), which is why we
3236 * have uninstalled but not yet destroyed the subfacets. */
3237 facet_flush_stats(facet);
3239 /* Now we're really all done so destroy everything. */
3240 LIST_FOR_EACH_SAFE (subfacet, next_subfacet, list_node,
3241 &facet->subfacets) {
3242 subfacet_destroy__(subfacet);
3244 hmap_remove(&ofproto->facets, &facet->hmap_node);
3245 list_remove(&facet->list_node);
3250 facet_account(struct facet *facet, bool may_flow_mod)
3252 struct ofproto_dpif *ofproto = ofproto_dpif_cast(facet->rule->up.ofproto);
3254 struct subfacet *subfacet;
3255 const struct nlattr *a;
3259 if (facet->byte_count <= facet->accounted_bytes) {
3262 n_bytes = facet->byte_count - facet->accounted_bytes;
3263 facet->accounted_bytes = facet->byte_count;
3265 /* Feed information from the active flows back into the learning table to
3266 * ensure that table is always in sync with what is actually flowing
3267 * through the datapath. */
3268 if (facet->has_learn || facet->has_normal
3269 || (facet->has_fin_timeout
3270 && facet->tcp_flags & (TCP_FIN | TCP_RST))) {
3271 struct action_xlate_ctx ctx;
3273 action_xlate_ctx_init(&ctx, ofproto, &facet->flow,
3274 facet->flow.vlan_tci,
3275 facet->rule, facet->tcp_flags, NULL);
3276 ctx.may_learn_macs = true;
3277 ctx.may_flow_mod = may_flow_mod;
3278 ofpbuf_delete(xlate_actions(&ctx, facet->rule->up.actions,
3279 facet->rule->up.n_actions));
3282 if (!facet->has_normal || !ofproto->has_bonded_bundles) {
3286 /* This loop feeds byte counters to bond_account() for rebalancing to use
3287 * as a basis. We also need to track the actual VLAN on which the packet
3288 * is going to be sent to ensure that it matches the one passed to
3289 * bond_choose_output_slave(). (Otherwise, we will account to the wrong
3292 * We use the actions from an arbitrary subfacet because they should all
3293 * be equally valid for our purpose. */
3294 subfacet = CONTAINER_OF(list_front(&facet->subfacets),
3295 struct subfacet, list_node);
3296 vlan_tci = facet->flow.vlan_tci;
3297 NL_ATTR_FOR_EACH_UNSAFE (a, left,
3298 subfacet->actions, subfacet->actions_len) {
3299 const struct ovs_action_push_vlan *vlan;
3300 struct ofport_dpif *port;
3302 switch (nl_attr_type(a)) {
3303 case OVS_ACTION_ATTR_OUTPUT:
3304 port = get_odp_port(ofproto, nl_attr_get_u32(a));
3305 if (port && port->bundle && port->bundle->bond) {
3306 bond_account(port->bundle->bond, &facet->flow,
3307 vlan_tci_to_vid(vlan_tci), n_bytes);
3311 case OVS_ACTION_ATTR_POP_VLAN:
3312 vlan_tci = htons(0);
3315 case OVS_ACTION_ATTR_PUSH_VLAN:
3316 vlan = nl_attr_get(a);
3317 vlan_tci = vlan->vlan_tci;
3323 /* Returns true if the only action for 'facet' is to send to the controller.
3324 * (We don't report NetFlow expiration messages for such facets because they
3325 * are just part of the control logic for the network, not real traffic). */
3327 facet_is_controller_flow(struct facet *facet)
3330 && facet->rule->up.n_actions == 1
3331 && action_outputs_to_port(&facet->rule->up.actions[0],
3332 htons(OFPP_CONTROLLER)));
3335 /* Folds all of 'facet''s statistics into its rule. Also updates the
3336 * accounting ofhook and emits a NetFlow expiration if appropriate. All of
3337 * 'facet''s statistics in the datapath should have been zeroed and folded into
3338 * its packet and byte counts before this function is called. */
3340 facet_flush_stats(struct facet *facet)
3342 struct ofproto_dpif *ofproto = ofproto_dpif_cast(facet->rule->up.ofproto);
3343 struct subfacet *subfacet;
3345 LIST_FOR_EACH (subfacet, list_node, &facet->subfacets) {
3346 assert(!subfacet->dp_byte_count);
3347 assert(!subfacet->dp_packet_count);
3350 facet_push_stats(facet);
3351 facet_account(facet, false);
3353 if (ofproto->netflow && !facet_is_controller_flow(facet)) {
3354 struct ofexpired expired;
3355 expired.flow = facet->flow;
3356 expired.packet_count = facet->packet_count;
3357 expired.byte_count = facet->byte_count;
3358 expired.used = facet->used;
3359 netflow_expire(ofproto->netflow, &facet->nf_flow, &expired);
3362 facet->rule->packet_count += facet->packet_count;
3363 facet->rule->byte_count += facet->byte_count;
3365 /* Reset counters to prevent double counting if 'facet' ever gets
3367 facet_reset_counters(facet);
3369 netflow_flow_clear(&facet->nf_flow);
3370 facet->tcp_flags = 0;
3373 /* Searches 'ofproto''s table of facets for one exactly equal to 'flow'.
3374 * Returns it if found, otherwise a null pointer.
3376 * The returned facet might need revalidation; use facet_lookup_valid()
3377 * instead if that is important. */
3378 static struct facet *
3379 facet_find(struct ofproto_dpif *ofproto, const struct flow *flow)
3381 struct facet *facet;
3383 HMAP_FOR_EACH_WITH_HASH (facet, hmap_node, flow_hash(flow, 0),
3385 if (flow_equal(flow, &facet->flow)) {
3393 /* Searches 'ofproto''s table of facets for one exactly equal to 'flow'.
3394 * Returns it if found, otherwise a null pointer.
3396 * The returned facet is guaranteed to be valid. */
3397 static struct facet *
3398 facet_lookup_valid(struct ofproto_dpif *ofproto, const struct flow *flow)
3400 struct facet *facet = facet_find(ofproto, flow);
3402 /* The facet we found might not be valid, since we could be in need of
3403 * revalidation. If it is not valid, don't return it. */
3405 && (ofproto->need_revalidate
3406 || tag_set_intersects(&ofproto->revalidate_set, facet->tags))
3407 && !facet_revalidate(facet)) {
3408 COVERAGE_INC(facet_invalidated);
3416 facet_check_consistency(struct facet *facet)
3418 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 15);
3420 struct ofproto_dpif *ofproto = ofproto_dpif_cast(facet->rule->up.ofproto);
3422 struct rule_dpif *rule;
3423 struct subfacet *subfacet;
3424 bool may_log = false;
3427 /* Check the rule for consistency. */
3428 rule = rule_dpif_lookup(ofproto, &facet->flow, 0);
3430 if (!VLOG_DROP_WARN(&rl)) {
3431 char *s = flow_to_string(&facet->flow);
3432 VLOG_WARN("%s: facet should not exist", s);
3436 } else if (rule != facet->rule) {
3437 may_log = !VLOG_DROP_WARN(&rl);
3443 flow_format(&s, &facet->flow);
3444 ds_put_format(&s, ": facet associated with wrong rule (was "
3445 "table=%"PRIu8",", facet->rule->up.table_id);
3446 cls_rule_format(&facet->rule->up.cr, &s);
3447 ds_put_format(&s, ") (should have been table=%"PRIu8",",
3449 cls_rule_format(&rule->up.cr, &s);
3450 ds_put_char(&s, ')');
3452 VLOG_WARN("%s", ds_cstr(&s));
3459 /* Check the datapath actions for consistency. */
3460 LIST_FOR_EACH (subfacet, list_node, &facet->subfacets) {
3461 struct action_xlate_ctx ctx;
3462 struct ofpbuf *odp_actions;
3463 bool actions_changed;
3464 bool should_install;
3466 action_xlate_ctx_init(&ctx, ofproto, &facet->flow,
3467 subfacet->initial_tci, rule, 0, NULL);
3468 odp_actions = xlate_actions(&ctx, rule->up.actions,
3469 rule->up.n_actions);
3471 should_install = (ctx.may_set_up_flow
3472 && subfacet->key_fitness != ODP_FIT_TOO_LITTLE);
3473 if (!should_install && !subfacet->installed) {
3474 /* The actions for uninstallable flows may vary from one packet to
3475 * the next, so don't compare the actions. */
3479 actions_changed = (subfacet->actions_len != odp_actions->size
3480 || memcmp(subfacet->actions, odp_actions->data,
3481 subfacet->actions_len));
3482 if (should_install != subfacet->installed || actions_changed) {
3484 may_log = !VLOG_DROP_WARN(&rl);
3489 struct odputil_keybuf keybuf;
3494 subfacet_get_key(subfacet, &keybuf, &key);
3495 odp_flow_key_format(key.data, key.size, &s);
3497 ds_put_cstr(&s, ": inconsistency in subfacet");
3498 if (should_install != subfacet->installed) {
3499 enum odp_key_fitness fitness = subfacet->key_fitness;
3501 ds_put_format(&s, " (should%s have been installed)",
3502 should_install ? "" : " not");
3503 ds_put_format(&s, " (may_set_up_flow=%s, fitness=%s)",
3504 ctx.may_set_up_flow ? "true" : "false",
3505 odp_key_fitness_to_string(fitness));
3507 if (actions_changed) {
3508 ds_put_cstr(&s, " (actions were: ");
3509 format_odp_actions(&s, subfacet->actions,
3510 subfacet->actions_len);
3511 ds_put_cstr(&s, ") (correct actions: ");
3512 format_odp_actions(&s, odp_actions->data,
3514 ds_put_char(&s, ')');
3516 ds_put_cstr(&s, " (actions: ");
3517 format_odp_actions(&s, subfacet->actions,
3518 subfacet->actions_len);
3519 ds_put_char(&s, ')');
3521 VLOG_WARN("%s", ds_cstr(&s));
3527 ofpbuf_delete(odp_actions);
3533 /* Re-searches the classifier for 'facet':
3535 * - If the rule found is different from 'facet''s current rule, moves
3536 * 'facet' to the new rule and recompiles its actions.
3538 * - If the rule found is the same as 'facet''s current rule, leaves 'facet'
3539 * where it is and recompiles its actions anyway.
3541 * - If there is none, destroys 'facet'.
3543 * Returns true if 'facet' still exists, false if it has been destroyed. */
3545 facet_revalidate(struct facet *facet)
3547 struct ofproto_dpif *ofproto = ofproto_dpif_cast(facet->rule->up.ofproto);
3549 struct nlattr *odp_actions;
3552 struct actions *new_actions;
3554 struct action_xlate_ctx ctx;
3555 struct rule_dpif *new_rule;
3556 struct subfacet *subfacet;
3557 bool actions_changed;
3560 COVERAGE_INC(facet_revalidate);
3562 /* Determine the new rule. */
3563 new_rule = rule_dpif_lookup(ofproto, &facet->flow, 0);
3565 /* No new rule, so delete the facet. */
3566 facet_remove(facet);
3570 /* Calculate new datapath actions.
3572 * We do not modify any 'facet' state yet, because we might need to, e.g.,
3573 * emit a NetFlow expiration and, if so, we need to have the old state
3574 * around to properly compose it. */
3576 /* If the datapath actions changed or the installability changed,
3577 * then we need to talk to the datapath. */
3580 memset(&ctx, 0, sizeof ctx);
3581 LIST_FOR_EACH (subfacet, list_node, &facet->subfacets) {
3582 struct ofpbuf *odp_actions;
3583 bool should_install;
3585 action_xlate_ctx_init(&ctx, ofproto, &facet->flow,
3586 subfacet->initial_tci, new_rule, 0, NULL);
3587 odp_actions = xlate_actions(&ctx, new_rule->up.actions,
3588 new_rule->up.n_actions);
3589 actions_changed = (subfacet->actions_len != odp_actions->size
3590 || memcmp(subfacet->actions, odp_actions->data,
3591 subfacet->actions_len));
3593 should_install = (ctx.may_set_up_flow
3594 && subfacet->key_fitness != ODP_FIT_TOO_LITTLE);
3595 if (actions_changed || should_install != subfacet->installed) {
3596 if (should_install) {
3597 struct dpif_flow_stats stats;
3599 subfacet_install(subfacet,
3600 odp_actions->data, odp_actions->size, &stats);
3601 subfacet_update_stats(subfacet, &stats);
3603 subfacet_uninstall(subfacet);
3607 new_actions = xcalloc(list_size(&facet->subfacets),
3608 sizeof *new_actions);
3610 new_actions[i].odp_actions = xmemdup(odp_actions->data,
3612 new_actions[i].actions_len = odp_actions->size;
3615 ofpbuf_delete(odp_actions);
3619 facet_flush_stats(facet);
3622 /* Update 'facet' now that we've taken care of all the old state. */
3623 facet->tags = ctx.tags;
3624 facet->nf_flow.output_iface = ctx.nf_output_iface;
3625 facet->may_install = ctx.may_set_up_flow;
3626 facet->has_learn = ctx.has_learn;
3627 facet->has_normal = ctx.has_normal;
3628 facet->has_fin_timeout = ctx.has_fin_timeout;
3629 facet->mirrors = ctx.mirrors;
3632 LIST_FOR_EACH (subfacet, list_node, &facet->subfacets) {
3633 if (new_actions[i].odp_actions) {
3634 free(subfacet->actions);
3635 subfacet->actions = new_actions[i].odp_actions;
3636 subfacet->actions_len = new_actions[i].actions_len;
3642 if (facet->rule != new_rule) {
3643 COVERAGE_INC(facet_changed_rule);
3644 list_remove(&facet->list_node);
3645 list_push_back(&new_rule->facets, &facet->list_node);
3646 facet->rule = new_rule;
3647 facet->used = new_rule->up.created;
3648 facet->prev_used = facet->used;
3654 /* Updates 'facet''s used time. Caller is responsible for calling
3655 * facet_push_stats() to update the flows which 'facet' resubmits into. */
3657 facet_update_time(struct facet *facet, long long int used)
3659 struct ofproto_dpif *ofproto = ofproto_dpif_cast(facet->rule->up.ofproto);
3660 if (used > facet->used) {
3662 ofproto_rule_update_used(&facet->rule->up, used);
3663 netflow_flow_update_time(ofproto->netflow, &facet->nf_flow, used);
3668 facet_reset_counters(struct facet *facet)
3670 facet->packet_count = 0;
3671 facet->byte_count = 0;
3672 facet->prev_packet_count = 0;
3673 facet->prev_byte_count = 0;
3674 facet->accounted_bytes = 0;
3678 facet_push_stats(struct facet *facet)
3680 uint64_t new_packets, new_bytes;
3682 assert(facet->packet_count >= facet->prev_packet_count);
3683 assert(facet->byte_count >= facet->prev_byte_count);
3684 assert(facet->used >= facet->prev_used);
3686 new_packets = facet->packet_count - facet->prev_packet_count;
3687 new_bytes = facet->byte_count - facet->prev_byte_count;
3689 if (new_packets || new_bytes || facet->used > facet->prev_used) {
3690 facet->prev_packet_count = facet->packet_count;
3691 facet->prev_byte_count = facet->byte_count;
3692 facet->prev_used = facet->used;
3694 flow_push_stats(facet->rule, &facet->flow,
3695 new_packets, new_bytes, facet->used);
3697 update_mirror_stats(ofproto_dpif_cast(facet->rule->up.ofproto),
3698 facet->mirrors, new_packets, new_bytes);
3702 struct ofproto_push {
3703 struct action_xlate_ctx ctx;
3710 push_resubmit(struct action_xlate_ctx *ctx, struct rule_dpif *rule)
3712 struct ofproto_push *push = CONTAINER_OF(ctx, struct ofproto_push, ctx);
3715 rule->packet_count += push->packets;
3716 rule->byte_count += push->bytes;
3717 ofproto_rule_update_used(&rule->up, push->used);
3721 /* Pushes flow statistics to the rules which 'flow' resubmits into given
3722 * 'rule''s actions and mirrors. */
3724 flow_push_stats(struct rule_dpif *rule,
3725 const struct flow *flow, uint64_t packets, uint64_t bytes,
3728 struct ofproto_dpif *ofproto = ofproto_dpif_cast(rule->up.ofproto);
3729 struct ofproto_push push;
3731 push.packets = packets;
3735 ofproto_rule_update_used(&rule->up, used);
3737 action_xlate_ctx_init(&push.ctx, ofproto, flow, flow->vlan_tci, rule,
3739 push.ctx.resubmit_hook = push_resubmit;
3740 ofpbuf_delete(xlate_actions(&push.ctx,
3741 rule->up.actions, rule->up.n_actions));
3746 static struct subfacet *
3747 subfacet_find__(struct ofproto_dpif *ofproto,
3748 const struct nlattr *key, size_t key_len, uint32_t key_hash,
3749 const struct flow *flow)
3751 struct subfacet *subfacet;
3753 HMAP_FOR_EACH_WITH_HASH (subfacet, hmap_node, key_hash,
3754 &ofproto->subfacets) {
3756 ? (subfacet->key_len == key_len
3757 && !memcmp(key, subfacet->key, key_len))
3758 : flow_equal(flow, &subfacet->facet->flow)) {
3766 /* Searches 'facet' (within 'ofproto') for a subfacet with the specified
3767 * 'key_fitness', 'key', and 'key_len'. Returns the existing subfacet if
3768 * there is one, otherwise creates and returns a new subfacet.
3770 * If the returned subfacet is new, then subfacet->actions will be NULL, in
3771 * which case the caller must populate the actions with
3772 * subfacet_make_actions(). */
3773 static struct subfacet *
3774 subfacet_create(struct facet *facet, enum odp_key_fitness key_fitness,
3775 const struct nlattr *key, size_t key_len, ovs_be16 initial_tci)
3777 struct ofproto_dpif *ofproto = ofproto_dpif_cast(facet->rule->up.ofproto);
3778 uint32_t key_hash = odp_flow_key_hash(key, key_len);
3779 struct subfacet *subfacet;
3781 subfacet = subfacet_find__(ofproto, key, key_len, key_hash, &facet->flow);
3783 if (subfacet->facet == facet) {
3787 /* This shouldn't happen. */
3788 VLOG_ERR_RL(&rl, "subfacet with wrong facet");
3789 subfacet_destroy(subfacet);
3792 subfacet = xzalloc(sizeof *subfacet);
3793 hmap_insert(&ofproto->subfacets, &subfacet->hmap_node, key_hash);
3794 list_push_back(&facet->subfacets, &subfacet->list_node);
3795 subfacet->facet = facet;
3796 subfacet->used = time_msec();
3797 subfacet->key_fitness = key_fitness;
3798 if (key_fitness != ODP_FIT_PERFECT) {
3799 subfacet->key = xmemdup(key, key_len);
3800 subfacet->key_len = key_len;
3802 subfacet->installed = false;
3803 subfacet->initial_tci = initial_tci;
3808 /* Searches 'ofproto' for a subfacet with the given 'key', 'key_len', and
3809 * 'flow'. Returns the subfacet if one exists, otherwise NULL. */
3810 static struct subfacet *
3811 subfacet_find(struct ofproto_dpif *ofproto,
3812 const struct nlattr *key, size_t key_len)
3814 uint32_t key_hash = odp_flow_key_hash(key, key_len);
3815 enum odp_key_fitness fitness;
3818 fitness = odp_flow_key_to_flow(key, key_len, &flow);
3819 if (fitness == ODP_FIT_ERROR) {
3823 return subfacet_find__(ofproto, key, key_len, key_hash, &flow);
3826 /* Uninstalls 'subfacet' from the datapath, if it is installed, removes it from
3827 * its facet within 'ofproto', and frees it. */
3829 subfacet_destroy__(struct subfacet *subfacet)
3831 struct facet *facet = subfacet->facet;
3832 struct ofproto_dpif *ofproto = ofproto_dpif_cast(facet->rule->up.ofproto);
3834 subfacet_uninstall(subfacet);
3835 hmap_remove(&ofproto->subfacets, &subfacet->hmap_node);
3836 list_remove(&subfacet->list_node);
3837 free(subfacet->key);
3838 free(subfacet->actions);
3842 /* Destroys 'subfacet', as with subfacet_destroy__(), and then if this was the
3843 * last remaining subfacet in its facet destroys the facet too. */
3845 subfacet_destroy(struct subfacet *subfacet)
3847 struct facet *facet = subfacet->facet;
3849 if (list_is_singleton(&facet->subfacets)) {
3850 /* facet_remove() needs at least one subfacet (it will remove it). */
3851 facet_remove(facet);
3853 subfacet_destroy__(subfacet);
3857 /* Initializes 'key' with the sequence of OVS_KEY_ATTR_* Netlink attributes
3858 * that can be used to refer to 'subfacet'. The caller must provide 'keybuf'
3859 * for use as temporary storage. */
3861 subfacet_get_key(struct subfacet *subfacet, struct odputil_keybuf *keybuf,
3864 if (!subfacet->key) {
3865 ofpbuf_use_stack(key, keybuf, sizeof *keybuf);
3866 odp_flow_key_from_flow(key, &subfacet->facet->flow);
3868 ofpbuf_use_const(key, subfacet->key, subfacet->key_len);
3872 /* Composes the datapath actions for 'subfacet' based on its rule's actions. */
3874 subfacet_make_actions(struct subfacet *subfacet, const struct ofpbuf *packet)
3876 struct facet *facet = subfacet->facet;
3877 struct rule_dpif *rule = facet->rule;
3878 struct ofproto_dpif *ofproto = ofproto_dpif_cast(rule->up.ofproto);
3879 struct ofpbuf *odp_actions;
3880 struct action_xlate_ctx ctx;
3882 action_xlate_ctx_init(&ctx, ofproto, &facet->flow, subfacet->initial_tci,
3884 odp_actions = xlate_actions(&ctx, rule->up.actions, rule->up.n_actions);
3885 facet->tags = ctx.tags;
3886 facet->may_install = ctx.may_set_up_flow;
3887 facet->has_learn = ctx.has_learn;
3888 facet->has_normal = ctx.has_normal;
3889 facet->has_fin_timeout = ctx.has_fin_timeout;
3890 facet->nf_flow.output_iface = ctx.nf_output_iface;
3891 facet->mirrors = ctx.mirrors;
3893 if (subfacet->actions_len != odp_actions->size
3894 || memcmp(subfacet->actions, odp_actions->data, odp_actions->size)) {
3895 free(subfacet->actions);
3896 subfacet->actions_len = odp_actions->size;
3897 subfacet->actions = xmemdup(odp_actions->data, odp_actions->size);
3900 ofpbuf_delete(odp_actions);
3903 /* Updates 'subfacet''s datapath flow, setting its actions to 'actions_len'
3904 * bytes of actions in 'actions'. If 'stats' is non-null, statistics counters
3905 * in the datapath will be zeroed and 'stats' will be updated with traffic new
3906 * since 'subfacet' was last updated.
3908 * Returns 0 if successful, otherwise a positive errno value. */
3910 subfacet_install(struct subfacet *subfacet,
3911 const struct nlattr *actions, size_t actions_len,
3912 struct dpif_flow_stats *stats)
3914 struct facet *facet = subfacet->facet;
3915 struct ofproto_dpif *ofproto = ofproto_dpif_cast(facet->rule->up.ofproto);
3916 struct odputil_keybuf keybuf;
3917 enum dpif_flow_put_flags flags;
3921 flags = DPIF_FP_CREATE | DPIF_FP_MODIFY;
3923 flags |= DPIF_FP_ZERO_STATS;
3926 subfacet_get_key(subfacet, &keybuf, &key);
3927 ret = dpif_flow_put(ofproto->dpif, flags, key.data, key.size,
3928 actions, actions_len, stats);
3931 subfacet_reset_dp_stats(subfacet, stats);
3937 /* If 'subfacet' is installed in the datapath, uninstalls it. */
3939 subfacet_uninstall(struct subfacet *subfacet)
3941 if (subfacet->installed) {
3942 struct rule_dpif *rule = subfacet->facet->rule;
3943 struct ofproto_dpif *ofproto = ofproto_dpif_cast(rule->up.ofproto);
3944 struct odputil_keybuf keybuf;
3945 struct dpif_flow_stats stats;
3949 subfacet_get_key(subfacet, &keybuf, &key);
3950 error = dpif_flow_del(ofproto->dpif, key.data, key.size, &stats);
3951 subfacet_reset_dp_stats(subfacet, &stats);
3953 subfacet_update_stats(subfacet, &stats);
3955 subfacet->installed = false;
3957 assert(subfacet->dp_packet_count == 0);
3958 assert(subfacet->dp_byte_count == 0);
3962 /* Resets 'subfacet''s datapath statistics counters. This should be called
3963 * when 'subfacet''s statistics are cleared in the datapath. If 'stats' is
3964 * non-null, it should contain the statistics returned by dpif when 'subfacet'
3965 * was reset in the datapath. 'stats' will be modified to include only
3966 * statistics new since 'subfacet' was last updated. */
3968 subfacet_reset_dp_stats(struct subfacet *subfacet,
3969 struct dpif_flow_stats *stats)
3972 && subfacet->dp_packet_count <= stats->n_packets
3973 && subfacet->dp_byte_count <= stats->n_bytes) {
3974 stats->n_packets -= subfacet->dp_packet_count;
3975 stats->n_bytes -= subfacet->dp_byte_count;
3978 subfacet->dp_packet_count = 0;
3979 subfacet->dp_byte_count = 0;
3982 /* Updates 'subfacet''s used time. The caller is responsible for calling
3983 * facet_push_stats() to update the flows which 'subfacet' resubmits into. */
3985 subfacet_update_time(struct subfacet *subfacet, long long int used)
3987 if (used > subfacet->used) {
3988 subfacet->used = used;
3989 facet_update_time(subfacet->facet, used);
3993 /* Folds the statistics from 'stats' into the counters in 'subfacet'.
3995 * Because of the meaning of a subfacet's counters, it only makes sense to do
3996 * this if 'stats' are not tracked in the datapath, that is, if 'stats'
3997 * represents a packet that was sent by hand or if it represents statistics
3998 * that have been cleared out of the datapath. */
4000 subfacet_update_stats(struct subfacet *subfacet,
4001 const struct dpif_flow_stats *stats)
4003 if (stats->n_packets || stats->used > subfacet->used) {
4004 struct facet *facet = subfacet->facet;
4006 subfacet_update_time(subfacet, stats->used);
4007 facet->packet_count += stats->n_packets;
4008 facet->byte_count += stats->n_bytes;
4009 facet->tcp_flags |= stats->tcp_flags;
4010 facet_push_stats(facet);
4011 netflow_flow_update_flags(&facet->nf_flow, stats->tcp_flags);
4017 static struct rule_dpif *
4018 rule_dpif_lookup(struct ofproto_dpif *ofproto, const struct flow *flow,
4021 struct cls_rule *cls_rule;
4022 struct classifier *cls;
4024 if (table_id >= N_TABLES) {
4028 cls = &ofproto->up.tables[table_id].cls;
4029 if (flow->nw_frag & FLOW_NW_FRAG_ANY
4030 && ofproto->up.frag_handling == OFPC_FRAG_NORMAL) {
4031 /* For OFPC_NORMAL frag_handling, we must pretend that transport ports
4032 * are unavailable. */
4033 struct flow ofpc_normal_flow = *flow;
4034 ofpc_normal_flow.tp_src = htons(0);
4035 ofpc_normal_flow.tp_dst = htons(0);
4036 cls_rule = classifier_lookup(cls, &ofpc_normal_flow);
4038 cls_rule = classifier_lookup(cls, flow);
4040 return rule_dpif_cast(rule_from_cls_rule(cls_rule));
4044 complete_operation(struct rule_dpif *rule)
4046 struct ofproto_dpif *ofproto = ofproto_dpif_cast(rule->up.ofproto);
4048 rule_invalidate(rule);
4050 struct dpif_completion *c = xmalloc(sizeof *c);
4051 c->op = rule->up.pending;
4052 list_push_back(&ofproto->completions, &c->list_node);
4054 ofoperation_complete(rule->up.pending, 0);
4058 static struct rule *
4061 struct rule_dpif *rule = xmalloc(sizeof *rule);
4066 rule_dealloc(struct rule *rule_)
4068 struct rule_dpif *rule = rule_dpif_cast(rule_);
4073 rule_construct(struct rule *rule_)
4075 struct rule_dpif *rule = rule_dpif_cast(rule_);
4076 struct ofproto_dpif *ofproto = ofproto_dpif_cast(rule->up.ofproto);
4077 struct rule_dpif *victim;
4081 error = validate_actions(rule->up.actions, rule->up.n_actions,
4082 &rule->up.cr.flow, ofproto->max_ports);
4087 rule->packet_count = 0;
4088 rule->byte_count = 0;
4090 victim = rule_dpif_cast(ofoperation_get_victim(rule->up.pending));
4091 if (victim && !list_is_empty(&victim->facets)) {
4092 struct facet *facet;
4094 rule->facets = victim->facets;
4095 list_moved(&rule->facets);
4096 LIST_FOR_EACH (facet, list_node, &rule->facets) {
4097 /* XXX: We're only clearing our local counters here. It's possible
4098 * that quite a few packets are unaccounted for in the datapath
4099 * statistics. These will be accounted to the new rule instead of
4100 * cleared as required. This could be fixed by clearing out the
4101 * datapath statistics for this facet, but currently it doesn't
4103 facet_reset_counters(facet);
4107 /* Must avoid list_moved() in this case. */
4108 list_init(&rule->facets);
4111 table_id = rule->up.table_id;
4112 rule->tag = (victim ? victim->tag
4114 : rule_calculate_tag(&rule->up.cr.flow, &rule->up.cr.wc,
4115 ofproto->tables[table_id].basis));
4117 complete_operation(rule);
4122 rule_destruct(struct rule *rule_)
4124 struct rule_dpif *rule = rule_dpif_cast(rule_);
4125 struct facet *facet, *next_facet;
4127 LIST_FOR_EACH_SAFE (facet, next_facet, list_node, &rule->facets) {
4128 facet_revalidate(facet);
4131 complete_operation(rule);
4135 rule_get_stats(struct rule *rule_, uint64_t *packets, uint64_t *bytes)
4137 struct rule_dpif *rule = rule_dpif_cast(rule_);
4138 struct facet *facet;
4140 /* Start from historical data for 'rule' itself that are no longer tracked
4141 * in facets. This counts, for example, facets that have expired. */
4142 *packets = rule->packet_count;
4143 *bytes = rule->byte_count;
4145 /* Add any statistics that are tracked by facets. This includes
4146 * statistical data recently updated by ofproto_update_stats() as well as
4147 * stats for packets that were executed "by hand" via dpif_execute(). */
4148 LIST_FOR_EACH (facet, list_node, &rule->facets) {
4149 *packets += facet->packet_count;
4150 *bytes += facet->byte_count;
4155 rule_execute(struct rule *rule_, const struct flow *flow,
4156 struct ofpbuf *packet)
4158 struct rule_dpif *rule = rule_dpif_cast(rule_);
4159 struct ofproto_dpif *ofproto = ofproto_dpif_cast(rule->up.ofproto);
4160 struct action_xlate_ctx ctx;
4161 struct ofpbuf *odp_actions;
4164 action_xlate_ctx_init(&ctx, ofproto, flow, flow->vlan_tci,
4165 rule, packet_get_tcp_flags(packet, flow), packet);
4166 odp_actions = xlate_actions(&ctx, rule->up.actions, rule->up.n_actions);
4167 size = packet->size;
4168 if (execute_odp_actions(ofproto, flow, odp_actions->data,
4169 odp_actions->size, packet)) {
4170 rule->packet_count++;
4171 rule->byte_count += size;
4172 flow_push_stats(rule, flow, 1, size, time_msec());
4174 ofpbuf_delete(odp_actions);
4180 rule_modify_actions(struct rule *rule_)
4182 struct rule_dpif *rule = rule_dpif_cast(rule_);
4183 struct ofproto_dpif *ofproto = ofproto_dpif_cast(rule->up.ofproto);
4186 error = validate_actions(rule->up.actions, rule->up.n_actions,
4187 &rule->up.cr.flow, ofproto->max_ports);
4189 ofoperation_complete(rule->up.pending, error);
4193 complete_operation(rule);
4196 /* Sends 'packet' out 'ofport'.
4197 * May modify 'packet'.
4198 * Returns 0 if successful, otherwise a positive errno value. */
4200 send_packet(const struct ofport_dpif *ofport, struct ofpbuf *packet)
4202 const struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport->up.ofproto);
4203 struct ofpbuf key, odp_actions;
4204 struct odputil_keybuf keybuf;
4209 flow_extract((struct ofpbuf *) packet, 0, 0, 0, &flow);
4210 odp_port = vsp_realdev_to_vlandev(ofproto, ofport->odp_port,
4212 if (odp_port != ofport->odp_port) {
4213 eth_pop_vlan(packet);
4214 flow.vlan_tci = htons(0);
4217 ofpbuf_use_stack(&key, &keybuf, sizeof keybuf);
4218 odp_flow_key_from_flow(&key, &flow);
4220 ofpbuf_init(&odp_actions, 32);
4221 compose_sflow_action(ofproto, &odp_actions, &flow, odp_port);
4223 nl_msg_put_u32(&odp_actions, OVS_ACTION_ATTR_OUTPUT, odp_port);
4224 error = dpif_execute(ofproto->dpif,
4226 odp_actions.data, odp_actions.size,
4228 ofpbuf_uninit(&odp_actions);
4231 VLOG_WARN_RL(&rl, "%s: failed to send packet on port %"PRIu32" (%s)",
4232 ofproto->up.name, odp_port, strerror(error));
4234 ofproto_update_local_port_stats(ofport->up.ofproto, packet->size, 0);
4238 /* OpenFlow to datapath action translation. */
4240 static void do_xlate_actions(const union ofp_action *in, size_t n_in,
4241 struct action_xlate_ctx *ctx);
4242 static void xlate_normal(struct action_xlate_ctx *);
4245 put_userspace_action(const struct ofproto_dpif *ofproto,
4246 struct ofpbuf *odp_actions,
4247 const struct flow *flow,
4248 const struct user_action_cookie *cookie)
4252 pid = dpif_port_get_pid(ofproto->dpif,
4253 ofp_port_to_odp_port(flow->in_port));
4255 return odp_put_userspace_action(pid, cookie, odp_actions);
4258 /* Compose SAMPLE action for sFlow. */
4260 compose_sflow_action(const struct ofproto_dpif *ofproto,
4261 struct ofpbuf *odp_actions,
4262 const struct flow *flow,
4265 uint32_t port_ifindex;
4266 uint32_t probability;
4267 struct user_action_cookie cookie;
4268 size_t sample_offset, actions_offset;
4269 int cookie_offset, n_output;
4271 if (!ofproto->sflow || flow->in_port == OFPP_NONE) {
4275 if (odp_port == OVSP_NONE) {
4279 port_ifindex = dpif_sflow_odp_port_to_ifindex(ofproto->sflow, odp_port);
4283 sample_offset = nl_msg_start_nested(odp_actions, OVS_ACTION_ATTR_SAMPLE);
4285 /* Number of packets out of UINT_MAX to sample. */
4286 probability = dpif_sflow_get_probability(ofproto->sflow);
4287 nl_msg_put_u32(odp_actions, OVS_SAMPLE_ATTR_PROBABILITY, probability);
4289 actions_offset = nl_msg_start_nested(odp_actions, OVS_SAMPLE_ATTR_ACTIONS);
4291 cookie.type = USER_ACTION_COOKIE_SFLOW;
4292 cookie.data = port_ifindex;
4293 cookie.n_output = n_output;
4294 cookie.vlan_tci = 0;
4295 cookie_offset = put_userspace_action(ofproto, odp_actions, flow, &cookie);
4297 nl_msg_end_nested(odp_actions, actions_offset);
4298 nl_msg_end_nested(odp_actions, sample_offset);
4299 return cookie_offset;
4302 /* SAMPLE action must be first action in any given list of actions.
4303 * At this point we do not have all information required to build it. So try to
4304 * build sample action as complete as possible. */
4306 add_sflow_action(struct action_xlate_ctx *ctx)
4308 ctx->user_cookie_offset = compose_sflow_action(ctx->ofproto,
4310 &ctx->flow, OVSP_NONE);
4311 ctx->sflow_odp_port = 0;
4312 ctx->sflow_n_outputs = 0;
4315 /* Fix SAMPLE action according to data collected while composing ODP actions.
4316 * We need to fix SAMPLE actions OVS_SAMPLE_ATTR_ACTIONS attribute, i.e. nested
4317 * USERSPACE action's user-cookie which is required for sflow. */
4319 fix_sflow_action(struct action_xlate_ctx *ctx)
4321 const struct flow *base = &ctx->base_flow;
4322 struct user_action_cookie *cookie;
4324 if (!ctx->user_cookie_offset) {
4328 cookie = ofpbuf_at(ctx->odp_actions, ctx->user_cookie_offset,
4330 assert(cookie != NULL);
4331 assert(cookie->type == USER_ACTION_COOKIE_SFLOW);
4333 if (ctx->sflow_n_outputs) {
4334 cookie->data = dpif_sflow_odp_port_to_ifindex(ctx->ofproto->sflow,
4335 ctx->sflow_odp_port);
4337 if (ctx->sflow_n_outputs >= 255) {
4338 cookie->n_output = 255;
4340 cookie->n_output = ctx->sflow_n_outputs;
4342 cookie->vlan_tci = base->vlan_tci;
4346 compose_output_action__(struct action_xlate_ctx *ctx, uint16_t ofp_port,
4349 const struct ofport_dpif *ofport = get_ofp_port(ctx->ofproto, ofp_port);
4350 uint16_t odp_port = ofp_port_to_odp_port(ofp_port);
4351 ovs_be16 flow_vlan_tci = ctx->flow.vlan_tci;
4352 uint8_t flow_nw_tos = ctx->flow.nw_tos;
4356 struct priority_to_dscp *pdscp;
4358 if (ofport->up.pp.config & OFPUTIL_PC_NO_FWD
4359 || (check_stp && !stp_forward_in_state(ofport->stp_state))) {
4363 pdscp = get_priority(ofport, ctx->flow.skb_priority);
4365 ctx->flow.nw_tos &= ~IP_DSCP_MASK;
4366 ctx->flow.nw_tos |= pdscp->dscp;
4369 /* We may not have an ofport record for this port, but it doesn't hurt
4370 * to allow forwarding to it anyhow. Maybe such a port will appear
4371 * later and we're pre-populating the flow table. */
4374 out_port = vsp_realdev_to_vlandev(ctx->ofproto, odp_port,
4375 ctx->flow.vlan_tci);
4376 if (out_port != odp_port) {
4377 ctx->flow.vlan_tci = htons(0);
4379 commit_odp_actions(&ctx->flow, &ctx->base_flow, ctx->odp_actions);
4380 nl_msg_put_u32(ctx->odp_actions, OVS_ACTION_ATTR_OUTPUT, out_port);
4382 ctx->sflow_odp_port = odp_port;
4383 ctx->sflow_n_outputs++;
4384 ctx->nf_output_iface = ofp_port;
4385 ctx->flow.vlan_tci = flow_vlan_tci;
4386 ctx->flow.nw_tos = flow_nw_tos;
4390 compose_output_action(struct action_xlate_ctx *ctx, uint16_t ofp_port)
4392 compose_output_action__(ctx, ofp_port, true);
4396 xlate_table_action(struct action_xlate_ctx *ctx,
4397 uint16_t in_port, uint8_t table_id)
4399 if (ctx->recurse < MAX_RESUBMIT_RECURSION) {
4400 struct ofproto_dpif *ofproto = ctx->ofproto;
4401 struct rule_dpif *rule;
4402 uint16_t old_in_port;
4403 uint8_t old_table_id;
4405 old_table_id = ctx->table_id;
4406 ctx->table_id = table_id;
4408 /* Look up a flow with 'in_port' as the input port. */
4409 old_in_port = ctx->flow.in_port;
4410 ctx->flow.in_port = in_port;
4411 rule = rule_dpif_lookup(ofproto, &ctx->flow, table_id);
4414 if (table_id > 0 && table_id < N_TABLES) {
4415 struct table_dpif *table = &ofproto->tables[table_id];
4416 if (table->other_table) {
4417 ctx->tags |= (rule && rule->tag
4419 : rule_calculate_tag(&ctx->flow,
4420 &table->other_table->wc,
4425 /* Restore the original input port. Otherwise OFPP_NORMAL and
4426 * OFPP_IN_PORT will have surprising behavior. */
4427 ctx->flow.in_port = old_in_port;
4429 if (ctx->resubmit_hook) {
4430 ctx->resubmit_hook(ctx, rule);
4434 struct rule_dpif *old_rule = ctx->rule;
4438 do_xlate_actions(rule->up.actions, rule->up.n_actions, ctx);
4439 ctx->rule = old_rule;
4443 ctx->table_id = old_table_id;
4445 static struct vlog_rate_limit recurse_rl = VLOG_RATE_LIMIT_INIT(1, 1);
4447 VLOG_ERR_RL(&recurse_rl, "resubmit actions recursed over %d times",
4448 MAX_RESUBMIT_RECURSION);
4449 ctx->max_resubmit_trigger = true;
4454 xlate_resubmit_table(struct action_xlate_ctx *ctx,
4455 const struct nx_action_resubmit *nar)
4460 in_port = (nar->in_port == htons(OFPP_IN_PORT)
4462 : ntohs(nar->in_port));
4463 table_id = nar->table == 255 ? ctx->table_id : nar->table;
4465 xlate_table_action(ctx, in_port, table_id);
4469 flood_packets(struct action_xlate_ctx *ctx, bool all)
4471 struct ofport_dpif *ofport;
4473 HMAP_FOR_EACH (ofport, up.hmap_node, &ctx->ofproto->up.ports) {
4474 uint16_t ofp_port = ofport->up.ofp_port;
4476 if (ofp_port == ctx->flow.in_port) {
4481 compose_output_action__(ctx, ofp_port, false);
4482 } else if (!(ofport->up.pp.config & OFPUTIL_PC_NO_FLOOD)) {
4483 compose_output_action(ctx, ofp_port);
4487 ctx->nf_output_iface = NF_OUT_FLOOD;
4491 execute_controller_action(struct action_xlate_ctx *ctx, int len,
4492 enum ofp_packet_in_reason reason,
4493 uint16_t controller_id)
4495 struct ofputil_packet_in pin;
4496 struct ofpbuf *packet;
4498 ctx->may_set_up_flow = false;
4503 packet = ofpbuf_clone(ctx->packet);
4505 if (packet->l2 && packet->l3) {
4506 struct eth_header *eh;
4508 eth_pop_vlan(packet);
4511 /* If the Ethernet type is less than ETH_TYPE_MIN, it's likely an 802.2
4512 * LLC frame. Calculating the Ethernet type of these frames is more
4513 * trouble than seems appropriate for a simple assertion. */
4514 assert(ntohs(eh->eth_type) < ETH_TYPE_MIN
4515 || eh->eth_type == ctx->flow.dl_type);
4517 memcpy(eh->eth_src, ctx->flow.dl_src, sizeof eh->eth_src);
4518 memcpy(eh->eth_dst, ctx->flow.dl_dst, sizeof eh->eth_dst);
4520 if (ctx->flow.vlan_tci & htons(VLAN_CFI)) {
4521 eth_push_vlan(packet, ctx->flow.vlan_tci);
4525 if (ctx->flow.dl_type == htons(ETH_TYPE_IP)) {
4526 packet_set_ipv4(packet, ctx->flow.nw_src, ctx->flow.nw_dst,
4527 ctx->flow.nw_tos, ctx->flow.nw_ttl);
4531 if (ctx->flow.nw_proto == IPPROTO_TCP) {
4532 packet_set_tcp_port(packet, ctx->flow.tp_src,
4534 } else if (ctx->flow.nw_proto == IPPROTO_UDP) {
4535 packet_set_udp_port(packet, ctx->flow.tp_src,
4542 pin.packet = packet->data;
4543 pin.packet_len = packet->size;
4544 pin.reason = reason;
4545 pin.controller_id = controller_id;
4546 pin.table_id = ctx->table_id;
4547 pin.cookie = ctx->rule ? ctx->rule->up.flow_cookie : 0;
4550 flow_get_metadata(&ctx->flow, &pin.fmd);
4552 connmgr_send_packet_in(ctx->ofproto->up.connmgr, &pin);
4553 ofpbuf_delete(packet);
4557 compose_dec_ttl(struct action_xlate_ctx *ctx)
4559 if (ctx->flow.dl_type != htons(ETH_TYPE_IP) &&
4560 ctx->flow.dl_type != htons(ETH_TYPE_IPV6)) {
4564 if (ctx->flow.nw_ttl > 1) {
4568 execute_controller_action(ctx, UINT16_MAX, OFPR_INVALID_TTL, 0);
4570 /* Stop processing for current table. */
4576 xlate_output_action__(struct action_xlate_ctx *ctx,
4577 uint16_t port, uint16_t max_len)
4579 uint16_t prev_nf_output_iface = ctx->nf_output_iface;
4581 ctx->nf_output_iface = NF_OUT_DROP;
4585 compose_output_action(ctx, ctx->flow.in_port);
4588 xlate_table_action(ctx, ctx->flow.in_port, ctx->table_id);
4594 flood_packets(ctx, false);
4597 flood_packets(ctx, true);
4599 case OFPP_CONTROLLER:
4600 execute_controller_action(ctx, max_len, OFPR_ACTION, 0);
4606 if (port != ctx->flow.in_port) {
4607 compose_output_action(ctx, port);
4612 if (prev_nf_output_iface == NF_OUT_FLOOD) {
4613 ctx->nf_output_iface = NF_OUT_FLOOD;
4614 } else if (ctx->nf_output_iface == NF_OUT_DROP) {
4615 ctx->nf_output_iface = prev_nf_output_iface;
4616 } else if (prev_nf_output_iface != NF_OUT_DROP &&
4617 ctx->nf_output_iface != NF_OUT_FLOOD) {
4618 ctx->nf_output_iface = NF_OUT_MULTI;
4623 xlate_output_reg_action(struct action_xlate_ctx *ctx,
4624 const struct nx_action_output_reg *naor)
4626 struct mf_subfield src;
4629 nxm_decode(&src, naor->src, naor->ofs_nbits);
4630 ofp_port = mf_get_subfield(&src, &ctx->flow);
4632 if (ofp_port <= UINT16_MAX) {
4633 xlate_output_action__(ctx, ofp_port, ntohs(naor->max_len));
4638 xlate_output_action(struct action_xlate_ctx *ctx,
4639 const struct ofp_action_output *oao)
4641 xlate_output_action__(ctx, ntohs(oao->port), ntohs(oao->max_len));
4645 xlate_enqueue_action(struct action_xlate_ctx *ctx,
4646 const struct ofp_action_enqueue *oae)
4649 uint32_t flow_priority, priority;
4652 error = dpif_queue_to_priority(ctx->ofproto->dpif, ntohl(oae->queue_id),
4655 /* Fall back to ordinary output action. */
4656 xlate_output_action__(ctx, ntohs(oae->port), 0);
4660 /* Figure out datapath output port. */
4661 ofp_port = ntohs(oae->port);
4662 if (ofp_port == OFPP_IN_PORT) {
4663 ofp_port = ctx->flow.in_port;
4664 } else if (ofp_port == ctx->flow.in_port) {
4668 /* Add datapath actions. */
4669 flow_priority = ctx->flow.skb_priority;
4670 ctx->flow.skb_priority = priority;
4671 compose_output_action(ctx, ofp_port);
4672 ctx->flow.skb_priority = flow_priority;
4674 /* Update NetFlow output port. */
4675 if (ctx->nf_output_iface == NF_OUT_DROP) {
4676 ctx->nf_output_iface = ofp_port;
4677 } else if (ctx->nf_output_iface != NF_OUT_FLOOD) {
4678 ctx->nf_output_iface = NF_OUT_MULTI;
4683 xlate_set_queue_action(struct action_xlate_ctx *ctx,
4684 const struct nx_action_set_queue *nasq)
4689 error = dpif_queue_to_priority(ctx->ofproto->dpif, ntohl(nasq->queue_id),
4692 /* Couldn't translate queue to a priority, so ignore. A warning
4693 * has already been logged. */
4697 ctx->flow.skb_priority = priority;
4700 struct xlate_reg_state {
4706 xlate_autopath(struct action_xlate_ctx *ctx,
4707 const struct nx_action_autopath *naa)
4709 uint16_t ofp_port = ntohl(naa->id);
4710 struct ofport_dpif *port = get_ofp_port(ctx->ofproto, ofp_port);
4712 if (!port || !port->bundle) {
4713 ofp_port = OFPP_NONE;
4714 } else if (port->bundle->bond) {
4715 /* Autopath does not support VLAN hashing. */
4716 struct ofport_dpif *slave = bond_choose_output_slave(
4717 port->bundle->bond, &ctx->flow, 0, &ctx->tags);
4719 ofp_port = slave->up.ofp_port;
4722 autopath_execute(naa, &ctx->flow, ofp_port);
4726 slave_enabled_cb(uint16_t ofp_port, void *ofproto_)
4728 struct ofproto_dpif *ofproto = ofproto_;
4729 struct ofport_dpif *port;
4739 case OFPP_CONTROLLER: /* Not supported by the bundle action. */
4742 port = get_ofp_port(ofproto, ofp_port);
4743 return port ? port->may_enable : false;
4748 xlate_learn_action(struct action_xlate_ctx *ctx,
4749 const struct nx_action_learn *learn)
4751 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(5, 1);
4752 struct ofputil_flow_mod fm;
4755 learn_execute(learn, &ctx->flow, &fm);
4757 error = ofproto_flow_mod(&ctx->ofproto->up, &fm);
4758 if (error && !VLOG_DROP_WARN(&rl)) {
4759 VLOG_WARN("learning action failed to modify flow table (%s)",
4760 ofperr_get_name(error));
4766 /* Reduces '*timeout' to no more than 'max'. A value of zero in either case
4767 * means "infinite". */
4769 reduce_timeout(uint16_t max, uint16_t *timeout)
4771 if (max && (!*timeout || *timeout > max)) {
4777 xlate_fin_timeout(struct action_xlate_ctx *ctx,
4778 const struct nx_action_fin_timeout *naft)
4780 if (ctx->tcp_flags & (TCP_FIN | TCP_RST) && ctx->rule) {
4781 struct rule_dpif *rule = ctx->rule;
4783 reduce_timeout(ntohs(naft->fin_idle_timeout), &rule->up.idle_timeout);
4784 reduce_timeout(ntohs(naft->fin_hard_timeout), &rule->up.hard_timeout);
4789 may_receive(const struct ofport_dpif *port, struct action_xlate_ctx *ctx)
4791 if (port->up.pp.config & (eth_addr_equals(ctx->flow.dl_dst, eth_addr_stp)
4792 ? OFPUTIL_PC_NO_RECV_STP
4793 : OFPUTIL_PC_NO_RECV)) {
4797 /* Only drop packets here if both forwarding and learning are
4798 * disabled. If just learning is enabled, we need to have
4799 * OFPP_NORMAL and the learning action have a look at the packet
4800 * before we can drop it. */
4801 if (!stp_forward_in_state(port->stp_state)
4802 && !stp_learn_in_state(port->stp_state)) {
4810 do_xlate_actions(const union ofp_action *in, size_t n_in,
4811 struct action_xlate_ctx *ctx)
4813 const struct ofport_dpif *port;
4814 const union ofp_action *ia;
4815 bool was_evictable = true;
4818 port = get_ofp_port(ctx->ofproto, ctx->flow.in_port);
4819 if (port && !may_receive(port, ctx)) {
4820 /* Drop this flow. */
4825 /* Don't let the rule we're working on get evicted underneath us. */
4826 was_evictable = ctx->rule->up.evictable;
4827 ctx->rule->up.evictable = false;
4829 OFPUTIL_ACTION_FOR_EACH_UNSAFE (ia, left, in, n_in) {
4830 const struct ofp_action_dl_addr *oada;
4831 const struct nx_action_resubmit *nar;
4832 const struct nx_action_set_tunnel *nast;
4833 const struct nx_action_set_queue *nasq;
4834 const struct nx_action_multipath *nam;
4835 const struct nx_action_autopath *naa;
4836 const struct nx_action_bundle *nab;
4837 const struct nx_action_output_reg *naor;
4838 const struct nx_action_controller *nac;
4839 enum ofputil_action_code code;
4846 code = ofputil_decode_action_unsafe(ia);
4848 case OFPUTIL_OFPAT10_OUTPUT:
4849 xlate_output_action(ctx, &ia->output);
4852 case OFPUTIL_OFPAT10_SET_VLAN_VID:
4853 ctx->flow.vlan_tci &= ~htons(VLAN_VID_MASK);
4854 ctx->flow.vlan_tci |= ia->vlan_vid.vlan_vid | htons(VLAN_CFI);
4857 case OFPUTIL_OFPAT10_SET_VLAN_PCP:
4858 ctx->flow.vlan_tci &= ~htons(VLAN_PCP_MASK);
4859 ctx->flow.vlan_tci |= htons(
4860 (ia->vlan_pcp.vlan_pcp << VLAN_PCP_SHIFT) | VLAN_CFI);
4863 case OFPUTIL_OFPAT10_STRIP_VLAN:
4864 ctx->flow.vlan_tci = htons(0);
4867 case OFPUTIL_OFPAT10_SET_DL_SRC:
4868 oada = ((struct ofp_action_dl_addr *) ia);
4869 memcpy(ctx->flow.dl_src, oada->dl_addr, ETH_ADDR_LEN);
4872 case OFPUTIL_OFPAT10_SET_DL_DST:
4873 oada = ((struct ofp_action_dl_addr *) ia);
4874 memcpy(ctx->flow.dl_dst, oada->dl_addr, ETH_ADDR_LEN);
4877 case OFPUTIL_OFPAT10_SET_NW_SRC:
4878 ctx->flow.nw_src = ia->nw_addr.nw_addr;
4881 case OFPUTIL_OFPAT10_SET_NW_DST:
4882 ctx->flow.nw_dst = ia->nw_addr.nw_addr;
4885 case OFPUTIL_OFPAT10_SET_NW_TOS:
4886 /* OpenFlow 1.0 only supports IPv4. */
4887 if (ctx->flow.dl_type == htons(ETH_TYPE_IP)) {
4888 ctx->flow.nw_tos &= ~IP_DSCP_MASK;
4889 ctx->flow.nw_tos |= ia->nw_tos.nw_tos & IP_DSCP_MASK;
4893 case OFPUTIL_OFPAT10_SET_TP_SRC:
4894 ctx->flow.tp_src = ia->tp_port.tp_port;
4897 case OFPUTIL_OFPAT10_SET_TP_DST:
4898 ctx->flow.tp_dst = ia->tp_port.tp_port;
4901 case OFPUTIL_OFPAT10_ENQUEUE:
4902 xlate_enqueue_action(ctx, (const struct ofp_action_enqueue *) ia);
4905 case OFPUTIL_NXAST_RESUBMIT:
4906 nar = (const struct nx_action_resubmit *) ia;
4907 xlate_table_action(ctx, ntohs(nar->in_port), ctx->table_id);
4910 case OFPUTIL_NXAST_RESUBMIT_TABLE:
4911 xlate_resubmit_table(ctx, (const struct nx_action_resubmit *) ia);
4914 case OFPUTIL_NXAST_SET_TUNNEL:
4915 nast = (const struct nx_action_set_tunnel *) ia;
4916 tun_id = htonll(ntohl(nast->tun_id));
4917 ctx->flow.tun_id = tun_id;
4920 case OFPUTIL_NXAST_SET_QUEUE:
4921 nasq = (const struct nx_action_set_queue *) ia;
4922 xlate_set_queue_action(ctx, nasq);
4925 case OFPUTIL_NXAST_POP_QUEUE:
4926 ctx->flow.skb_priority = ctx->orig_skb_priority;
4929 case OFPUTIL_NXAST_REG_MOVE:
4930 nxm_execute_reg_move((const struct nx_action_reg_move *) ia,
4934 case OFPUTIL_NXAST_REG_LOAD:
4935 nxm_execute_reg_load((const struct nx_action_reg_load *) ia,
4939 case OFPUTIL_NXAST_NOTE:
4940 /* Nothing to do. */
4943 case OFPUTIL_NXAST_SET_TUNNEL64:
4944 tun_id = ((const struct nx_action_set_tunnel64 *) ia)->tun_id;
4945 ctx->flow.tun_id = tun_id;
4948 case OFPUTIL_NXAST_MULTIPATH:
4949 nam = (const struct nx_action_multipath *) ia;
4950 multipath_execute(nam, &ctx->flow);
4953 case OFPUTIL_NXAST_AUTOPATH:
4954 naa = (const struct nx_action_autopath *) ia;
4955 xlate_autopath(ctx, naa);
4958 case OFPUTIL_NXAST_BUNDLE:
4959 ctx->ofproto->has_bundle_action = true;
4960 nab = (const struct nx_action_bundle *) ia;
4961 xlate_output_action__(ctx, bundle_execute(nab, &ctx->flow,
4966 case OFPUTIL_NXAST_BUNDLE_LOAD:
4967 ctx->ofproto->has_bundle_action = true;
4968 nab = (const struct nx_action_bundle *) ia;
4969 bundle_execute_load(nab, &ctx->flow, slave_enabled_cb,
4973 case OFPUTIL_NXAST_OUTPUT_REG:
4974 naor = (const struct nx_action_output_reg *) ia;
4975 xlate_output_reg_action(ctx, naor);
4978 case OFPUTIL_NXAST_LEARN:
4979 ctx->has_learn = true;
4980 if (ctx->may_flow_mod) {
4981 xlate_learn_action(ctx, (const struct nx_action_learn *) ia);
4985 case OFPUTIL_NXAST_DEC_TTL:
4986 if (compose_dec_ttl(ctx)) {
4991 case OFPUTIL_NXAST_EXIT:
4995 case OFPUTIL_NXAST_FIN_TIMEOUT:
4996 ctx->has_fin_timeout = true;
4997 xlate_fin_timeout(ctx, (const struct nx_action_fin_timeout *) ia);
5000 case OFPUTIL_NXAST_CONTROLLER:
5001 nac = (const struct nx_action_controller *) ia;
5002 execute_controller_action(ctx, ntohs(nac->max_len), nac->reason,
5003 ntohs(nac->controller_id));
5009 /* We've let OFPP_NORMAL and the learning action look at the packet,
5010 * so drop it now if forwarding is disabled. */
5011 if (port && !stp_forward_in_state(port->stp_state)) {
5012 ofpbuf_clear(ctx->odp_actions);
5013 add_sflow_action(ctx);
5016 ctx->rule->up.evictable = was_evictable;
5021 action_xlate_ctx_init(struct action_xlate_ctx *ctx,
5022 struct ofproto_dpif *ofproto, const struct flow *flow,
5023 ovs_be16 initial_tci, struct rule_dpif *rule,
5024 uint8_t tcp_flags, const struct ofpbuf *packet)
5026 ctx->ofproto = ofproto;
5028 ctx->base_flow = ctx->flow;
5029 ctx->base_flow.tun_id = 0;
5030 ctx->base_flow.vlan_tci = initial_tci;
5032 ctx->packet = packet;
5033 ctx->may_learn_macs = packet != NULL;
5034 ctx->may_flow_mod = packet != NULL;
5035 ctx->tcp_flags = tcp_flags;
5036 ctx->resubmit_hook = NULL;
5039 static struct ofpbuf *
5040 xlate_actions(struct action_xlate_ctx *ctx,
5041 const union ofp_action *in, size_t n_in)
5043 struct flow orig_flow = ctx->flow;
5045 COVERAGE_INC(ofproto_dpif_xlate);
5047 ctx->odp_actions = ofpbuf_new(512);
5048 ofpbuf_reserve(ctx->odp_actions, NL_A_U32_SIZE);
5050 ctx->may_set_up_flow = true;
5051 ctx->has_learn = false;
5052 ctx->has_normal = false;
5053 ctx->has_fin_timeout = false;
5054 ctx->nf_output_iface = NF_OUT_DROP;
5057 ctx->max_resubmit_trigger = false;
5058 ctx->orig_skb_priority = ctx->flow.skb_priority;
5062 if (ctx->flow.nw_frag & FLOW_NW_FRAG_ANY) {
5063 switch (ctx->ofproto->up.frag_handling) {
5064 case OFPC_FRAG_NORMAL:
5065 /* We must pretend that transport ports are unavailable. */
5066 ctx->flow.tp_src = ctx->base_flow.tp_src = htons(0);
5067 ctx->flow.tp_dst = ctx->base_flow.tp_dst = htons(0);
5070 case OFPC_FRAG_DROP:
5071 return ctx->odp_actions;
5073 case OFPC_FRAG_REASM:
5076 case OFPC_FRAG_NX_MATCH:
5077 /* Nothing to do. */
5080 case OFPC_INVALID_TTL_TO_CONTROLLER:
5085 if (process_special(ctx->ofproto, &ctx->flow, ctx->packet)) {
5086 ctx->may_set_up_flow = false;
5087 return ctx->odp_actions;
5089 static struct vlog_rate_limit trace_rl = VLOG_RATE_LIMIT_INIT(1, 1);
5090 struct flow original_flow = ctx->flow;
5091 ovs_be16 initial_tci = ctx->base_flow.vlan_tci;
5093 add_sflow_action(ctx);
5094 do_xlate_actions(in, n_in, ctx);
5096 if (ctx->max_resubmit_trigger && !ctx->resubmit_hook
5097 && !VLOG_DROP_ERR(&trace_rl)) {
5098 struct ds ds = DS_EMPTY_INITIALIZER;
5100 ofproto_trace(ctx->ofproto, &original_flow, ctx->packet,
5102 VLOG_ERR("Trace triggered by excessive resubmit recursion:\n%s",
5107 if (!connmgr_may_set_up_flow(ctx->ofproto->up.connmgr, &ctx->flow,
5108 ctx->odp_actions->data,
5109 ctx->odp_actions->size)) {
5110 ctx->may_set_up_flow = false;
5112 && connmgr_msg_in_hook(ctx->ofproto->up.connmgr, &ctx->flow,
5114 compose_output_action(ctx, OFPP_LOCAL);
5117 add_mirror_actions(ctx, &orig_flow);
5118 fix_sflow_action(ctx);
5121 return ctx->odp_actions;
5124 /* OFPP_NORMAL implementation. */
5126 static struct ofport_dpif *ofbundle_get_a_port(const struct ofbundle *);
5128 /* Given 'vid', the VID obtained from the 802.1Q header that was received as
5129 * part of a packet (specify 0 if there was no 802.1Q header), and 'in_bundle',
5130 * the bundle on which the packet was received, returns the VLAN to which the
5133 * Both 'vid' and the return value are in the range 0...4095. */
5135 input_vid_to_vlan(const struct ofbundle *in_bundle, uint16_t vid)
5137 switch (in_bundle->vlan_mode) {
5138 case PORT_VLAN_ACCESS:
5139 return in_bundle->vlan;
5142 case PORT_VLAN_TRUNK:
5145 case PORT_VLAN_NATIVE_UNTAGGED:
5146 case PORT_VLAN_NATIVE_TAGGED:
5147 return vid ? vid : in_bundle->vlan;
5154 /* Checks whether a packet with the given 'vid' may ingress on 'in_bundle'.
5155 * If so, returns true. Otherwise, returns false and, if 'warn' is true, logs
5158 * 'vid' should be the VID obtained from the 802.1Q header that was received as
5159 * part of a packet (specify 0 if there was no 802.1Q header), in the range
5162 input_vid_is_valid(uint16_t vid, struct ofbundle *in_bundle, bool warn)
5164 /* Allow any VID on the OFPP_NONE port. */
5165 if (in_bundle == &ofpp_none_bundle) {
5169 switch (in_bundle->vlan_mode) {
5170 case PORT_VLAN_ACCESS:
5173 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
5174 VLOG_WARN_RL(&rl, "bridge %s: dropping VLAN %"PRIu16" tagged "
5175 "packet received on port %s configured as VLAN "
5176 "%"PRIu16" access port",
5177 in_bundle->ofproto->up.name, vid,
5178 in_bundle->name, in_bundle->vlan);
5184 case PORT_VLAN_NATIVE_UNTAGGED:
5185 case PORT_VLAN_NATIVE_TAGGED:
5187 /* Port must always carry its native VLAN. */
5191 case PORT_VLAN_TRUNK:
5192 if (!ofbundle_includes_vlan(in_bundle, vid)) {
5194 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
5195 VLOG_WARN_RL(&rl, "bridge %s: dropping VLAN %"PRIu16" packet "
5196 "received on port %s not configured for trunking "
5198 in_bundle->ofproto->up.name, vid,
5199 in_bundle->name, vid);
5211 /* Given 'vlan', the VLAN that a packet belongs to, and
5212 * 'out_bundle', a bundle on which the packet is to be output, returns the VID
5213 * that should be included in the 802.1Q header. (If the return value is 0,
5214 * then the 802.1Q header should only be included in the packet if there is a
5217 * Both 'vlan' and the return value are in the range 0...4095. */
5219 output_vlan_to_vid(const struct ofbundle *out_bundle, uint16_t vlan)
5221 switch (out_bundle->vlan_mode) {
5222 case PORT_VLAN_ACCESS:
5225 case PORT_VLAN_TRUNK:
5226 case PORT_VLAN_NATIVE_TAGGED:
5229 case PORT_VLAN_NATIVE_UNTAGGED:
5230 return vlan == out_bundle->vlan ? 0 : vlan;
5238 output_normal(struct action_xlate_ctx *ctx, const struct ofbundle *out_bundle,
5241 struct ofport_dpif *port;
5243 ovs_be16 tci, old_tci;
5245 vid = output_vlan_to_vid(out_bundle, vlan);
5246 if (!out_bundle->bond) {
5247 port = ofbundle_get_a_port(out_bundle);
5249 port = bond_choose_output_slave(out_bundle->bond, &ctx->flow,
5252 /* No slaves enabled, so drop packet. */
5257 old_tci = ctx->flow.vlan_tci;
5259 if (tci || out_bundle->use_priority_tags) {
5260 tci |= ctx->flow.vlan_tci & htons(VLAN_PCP_MASK);
5262 tci |= htons(VLAN_CFI);
5265 ctx->flow.vlan_tci = tci;
5267 compose_output_action(ctx, port->up.ofp_port);
5268 ctx->flow.vlan_tci = old_tci;
5272 mirror_mask_ffs(mirror_mask_t mask)
5274 BUILD_ASSERT_DECL(sizeof(unsigned int) >= sizeof(mask));
5279 ofbundle_trunks_vlan(const struct ofbundle *bundle, uint16_t vlan)
5281 return (bundle->vlan_mode != PORT_VLAN_ACCESS
5282 && (!bundle->trunks || bitmap_is_set(bundle->trunks, vlan)));
5286 ofbundle_includes_vlan(const struct ofbundle *bundle, uint16_t vlan)
5288 return vlan == bundle->vlan || ofbundle_trunks_vlan(bundle, vlan);
5291 /* Returns an arbitrary interface within 'bundle'. */
5292 static struct ofport_dpif *
5293 ofbundle_get_a_port(const struct ofbundle *bundle)
5295 return CONTAINER_OF(list_front(&bundle->ports),
5296 struct ofport_dpif, bundle_node);
5300 vlan_is_mirrored(const struct ofmirror *m, int vlan)
5302 return !m->vlans || bitmap_is_set(m->vlans, vlan);
5305 /* Returns true if a packet with Ethernet destination MAC 'dst' may be mirrored
5306 * to a VLAN. In general most packets may be mirrored but we want to drop
5307 * protocols that may confuse switches. */
5309 eth_dst_may_rspan(const uint8_t dst[ETH_ADDR_LEN])
5311 /* If you change this function's behavior, please update corresponding
5312 * documentation in vswitch.xml at the same time. */
5313 if (dst[0] != 0x01) {
5314 /* All the currently banned MACs happen to start with 01 currently, so
5315 * this is a quick way to eliminate most of the good ones. */
5317 if (eth_addr_is_reserved(dst)) {
5318 /* Drop STP, IEEE pause frames, and other reserved protocols
5319 * (01-80-c2-00-00-0x). */
5323 if (dst[0] == 0x01 && dst[1] == 0x00 && dst[2] == 0x0c) {
5325 if ((dst[3] & 0xfe) == 0xcc &&
5326 (dst[4] & 0xfe) == 0xcc &&
5327 (dst[5] & 0xfe) == 0xcc) {
5328 /* Drop the following protocols plus others following the same
5331 CDP, VTP, DTP, PAgP (01-00-0c-cc-cc-cc)
5332 Spanning Tree PVSTP+ (01-00-0c-cc-cc-cd)
5333 STP Uplink Fast (01-00-0c-cd-cd-cd) */
5337 if (!(dst[3] | dst[4] | dst[5])) {
5338 /* Drop Inter Switch Link packets (01-00-0c-00-00-00). */
5347 add_mirror_actions(struct action_xlate_ctx *ctx, const struct flow *orig_flow)
5349 struct ofproto_dpif *ofproto = ctx->ofproto;
5350 mirror_mask_t mirrors;
5351 struct ofbundle *in_bundle;
5354 const struct nlattr *a;
5357 in_bundle = lookup_input_bundle(ctx->ofproto, orig_flow->in_port,
5358 ctx->packet != NULL);
5362 mirrors = in_bundle->src_mirrors;
5364 /* Drop frames on bundles reserved for mirroring. */
5365 if (in_bundle->mirror_out) {
5366 if (ctx->packet != NULL) {
5367 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
5368 VLOG_WARN_RL(&rl, "bridge %s: dropping packet received on port "
5369 "%s, which is reserved exclusively for mirroring",
5370 ctx->ofproto->up.name, in_bundle->name);
5376 vid = vlan_tci_to_vid(orig_flow->vlan_tci);
5377 if (!input_vid_is_valid(vid, in_bundle, ctx->packet != NULL)) {
5380 vlan = input_vid_to_vlan(in_bundle, vid);
5382 /* Look at the output ports to check for destination selections. */
5384 NL_ATTR_FOR_EACH (a, left, ctx->odp_actions->data,
5385 ctx->odp_actions->size) {
5386 enum ovs_action_attr type = nl_attr_type(a);
5387 struct ofport_dpif *ofport;
5389 if (type != OVS_ACTION_ATTR_OUTPUT) {
5393 ofport = get_odp_port(ofproto, nl_attr_get_u32(a));
5394 if (ofport && ofport->bundle) {
5395 mirrors |= ofport->bundle->dst_mirrors;
5403 /* Restore the original packet before adding the mirror actions. */
5404 ctx->flow = *orig_flow;
5409 m = ofproto->mirrors[mirror_mask_ffs(mirrors) - 1];
5411 if (!vlan_is_mirrored(m, vlan)) {
5412 mirrors &= mirrors - 1;
5416 mirrors &= ~m->dup_mirrors;
5417 ctx->mirrors |= m->dup_mirrors;
5419 output_normal(ctx, m->out, vlan);
5420 } else if (eth_dst_may_rspan(orig_flow->dl_dst)
5421 && vlan != m->out_vlan) {
5422 struct ofbundle *bundle;
5424 HMAP_FOR_EACH (bundle, hmap_node, &ofproto->bundles) {
5425 if (ofbundle_includes_vlan(bundle, m->out_vlan)
5426 && !bundle->mirror_out) {
5427 output_normal(ctx, bundle, m->out_vlan);
5435 update_mirror_stats(struct ofproto_dpif *ofproto, mirror_mask_t mirrors,
5436 uint64_t packets, uint64_t bytes)
5442 for (; mirrors; mirrors &= mirrors - 1) {
5445 m = ofproto->mirrors[mirror_mask_ffs(mirrors) - 1];
5448 /* In normal circumstances 'm' will not be NULL. However,
5449 * if mirrors are reconfigured, we can temporarily get out
5450 * of sync in facet_revalidate(). We could "correct" the
5451 * mirror list before reaching here, but doing that would
5452 * not properly account the traffic stats we've currently
5453 * accumulated for previous mirror configuration. */
5457 m->packet_count += packets;
5458 m->byte_count += bytes;
5462 /* A VM broadcasts a gratuitous ARP to indicate that it has resumed after
5463 * migration. Older Citrix-patched Linux DomU used gratuitous ARP replies to
5464 * indicate this; newer upstream kernels use gratuitous ARP requests. */
5466 is_gratuitous_arp(const struct flow *flow)
5468 return (flow->dl_type == htons(ETH_TYPE_ARP)
5469 && eth_addr_is_broadcast(flow->dl_dst)
5470 && (flow->nw_proto == ARP_OP_REPLY
5471 || (flow->nw_proto == ARP_OP_REQUEST
5472 && flow->nw_src == flow->nw_dst)));
5476 update_learning_table(struct ofproto_dpif *ofproto,
5477 const struct flow *flow, int vlan,
5478 struct ofbundle *in_bundle)
5480 struct mac_entry *mac;
5482 /* Don't learn the OFPP_NONE port. */
5483 if (in_bundle == &ofpp_none_bundle) {
5487 if (!mac_learning_may_learn(ofproto->ml, flow->dl_src, vlan)) {
5491 mac = mac_learning_insert(ofproto->ml, flow->dl_src, vlan);
5492 if (is_gratuitous_arp(flow)) {
5493 /* We don't want to learn from gratuitous ARP packets that are
5494 * reflected back over bond slaves so we lock the learning table. */
5495 if (!in_bundle->bond) {
5496 mac_entry_set_grat_arp_lock(mac);
5497 } else if (mac_entry_is_grat_arp_locked(mac)) {
5502 if (mac_entry_is_new(mac) || mac->port.p != in_bundle) {
5503 /* The log messages here could actually be useful in debugging,
5504 * so keep the rate limit relatively high. */
5505 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(30, 300);
5506 VLOG_DBG_RL(&rl, "bridge %s: learned that "ETH_ADDR_FMT" is "
5507 "on port %s in VLAN %d",
5508 ofproto->up.name, ETH_ADDR_ARGS(flow->dl_src),
5509 in_bundle->name, vlan);
5511 mac->port.p = in_bundle;
5512 tag_set_add(&ofproto->revalidate_set,
5513 mac_learning_changed(ofproto->ml, mac));
5517 static struct ofbundle *
5518 lookup_input_bundle(struct ofproto_dpif *ofproto, uint16_t in_port, bool warn)
5520 struct ofport_dpif *ofport;
5522 /* Special-case OFPP_NONE, which a controller may use as the ingress
5523 * port for traffic that it is sourcing. */
5524 if (in_port == OFPP_NONE) {
5525 return &ofpp_none_bundle;
5528 /* Find the port and bundle for the received packet. */
5529 ofport = get_ofp_port(ofproto, in_port);
5530 if (ofport && ofport->bundle) {
5531 return ofport->bundle;
5534 /* Odd. A few possible reasons here:
5536 * - We deleted a port but there are still a few packets queued up
5539 * - Someone externally added a port (e.g. "ovs-dpctl add-if") that
5540 * we don't know about.
5542 * - The ofproto client didn't configure the port as part of a bundle.
5545 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
5547 VLOG_WARN_RL(&rl, "bridge %s: received packet on unknown "
5548 "port %"PRIu16, ofproto->up.name, in_port);
5553 /* Determines whether packets in 'flow' within 'ofproto' should be forwarded or
5554 * dropped. Returns true if they may be forwarded, false if they should be
5557 * 'in_port' must be the ofport_dpif that corresponds to flow->in_port.
5558 * 'in_port' must be part of a bundle (e.g. in_port->bundle must be nonnull).
5560 * 'vlan' must be the VLAN that corresponds to flow->vlan_tci on 'in_port', as
5561 * returned by input_vid_to_vlan(). It must be a valid VLAN for 'in_port', as
5562 * checked by input_vid_is_valid().
5564 * May also add tags to '*tags', although the current implementation only does
5565 * so in one special case.
5568 is_admissible(struct ofproto_dpif *ofproto, const struct flow *flow,
5569 struct ofport_dpif *in_port, uint16_t vlan, tag_type *tags)
5571 struct ofbundle *in_bundle = in_port->bundle;
5573 /* Drop frames for reserved multicast addresses
5574 * only if forward_bpdu option is absent. */
5575 if (eth_addr_is_reserved(flow->dl_dst) && !ofproto->up.forward_bpdu) {
5579 if (in_bundle->bond) {
5580 struct mac_entry *mac;
5582 switch (bond_check_admissibility(in_bundle->bond, in_port,
5583 flow->dl_dst, tags)) {
5590 case BV_DROP_IF_MOVED:
5591 mac = mac_learning_lookup(ofproto->ml, flow->dl_src, vlan, NULL);
5592 if (mac && mac->port.p != in_bundle &&
5593 (!is_gratuitous_arp(flow)
5594 || mac_entry_is_grat_arp_locked(mac))) {
5605 xlate_normal(struct action_xlate_ctx *ctx)
5607 struct ofport_dpif *in_port;
5608 struct ofbundle *in_bundle;
5609 struct mac_entry *mac;
5613 ctx->has_normal = true;
5615 in_bundle = lookup_input_bundle(ctx->ofproto, ctx->flow.in_port,
5616 ctx->packet != NULL);
5621 /* We know 'in_port' exists unless it is "ofpp_none_bundle",
5622 * since lookup_input_bundle() succeeded. */
5623 in_port = get_ofp_port(ctx->ofproto, ctx->flow.in_port);
5625 /* Drop malformed frames. */
5626 if (ctx->flow.dl_type == htons(ETH_TYPE_VLAN) &&
5627 !(ctx->flow.vlan_tci & htons(VLAN_CFI))) {
5628 if (ctx->packet != NULL) {
5629 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
5630 VLOG_WARN_RL(&rl, "bridge %s: dropping packet with partial "
5631 "VLAN tag received on port %s",
5632 ctx->ofproto->up.name, in_bundle->name);
5637 /* Drop frames on bundles reserved for mirroring. */
5638 if (in_bundle->mirror_out) {
5639 if (ctx->packet != NULL) {
5640 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
5641 VLOG_WARN_RL(&rl, "bridge %s: dropping packet received on port "
5642 "%s, which is reserved exclusively for mirroring",
5643 ctx->ofproto->up.name, in_bundle->name);
5649 vid = vlan_tci_to_vid(ctx->flow.vlan_tci);
5650 if (!input_vid_is_valid(vid, in_bundle, ctx->packet != NULL)) {
5653 vlan = input_vid_to_vlan(in_bundle, vid);
5655 /* Check other admissibility requirements. */
5657 !is_admissible(ctx->ofproto, &ctx->flow, in_port, vlan, &ctx->tags)) {
5661 /* Learn source MAC. */
5662 if (ctx->may_learn_macs) {
5663 update_learning_table(ctx->ofproto, &ctx->flow, vlan, in_bundle);
5666 /* Determine output bundle. */
5667 mac = mac_learning_lookup(ctx->ofproto->ml, ctx->flow.dl_dst, vlan,
5670 if (mac->port.p != in_bundle) {
5671 output_normal(ctx, mac->port.p, vlan);
5674 struct ofbundle *bundle;
5676 HMAP_FOR_EACH (bundle, hmap_node, &ctx->ofproto->bundles) {
5677 if (bundle != in_bundle
5678 && ofbundle_includes_vlan(bundle, vlan)
5679 && bundle->floodable
5680 && !bundle->mirror_out) {
5681 output_normal(ctx, bundle, vlan);
5684 ctx->nf_output_iface = NF_OUT_FLOOD;
5688 /* Optimized flow revalidation.
5690 * It's a difficult problem, in general, to tell which facets need to have
5691 * their actions recalculated whenever the OpenFlow flow table changes. We
5692 * don't try to solve that general problem: for most kinds of OpenFlow flow
5693 * table changes, we recalculate the actions for every facet. This is
5694 * relatively expensive, but it's good enough if the OpenFlow flow table
5695 * doesn't change very often.
5697 * However, we can expect one particular kind of OpenFlow flow table change to
5698 * happen frequently: changes caused by MAC learning. To avoid wasting a lot
5699 * of CPU on revalidating every facet whenever MAC learning modifies the flow
5700 * table, we add a special case that applies to flow tables in which every rule
5701 * has the same form (that is, the same wildcards), except that the table is
5702 * also allowed to have a single "catch-all" flow that matches all packets. We
5703 * optimize this case by tagging all of the facets that resubmit into the table
5704 * and invalidating the same tag whenever a flow changes in that table. The
5705 * end result is that we revalidate just the facets that need it (and sometimes
5706 * a few more, but not all of the facets or even all of the facets that
5707 * resubmit to the table modified by MAC learning). */
5709 /* Calculates the tag to use for 'flow' and wildcards 'wc' when it is inserted
5710 * into an OpenFlow table with the given 'basis'. */
5712 rule_calculate_tag(const struct flow *flow, const struct flow_wildcards *wc,
5715 if (flow_wildcards_is_catchall(wc)) {
5718 struct flow tag_flow = *flow;
5719 flow_zero_wildcards(&tag_flow, wc);
5720 return tag_create_deterministic(flow_hash(&tag_flow, secret));
5724 /* Following a change to OpenFlow table 'table_id' in 'ofproto', update the
5725 * taggability of that table.
5727 * This function must be called after *each* change to a flow table. If you
5728 * skip calling it on some changes then the pointer comparisons at the end can
5729 * be invalid if you get unlucky. For example, if a flow removal causes a
5730 * cls_table to be destroyed and then a flow insertion causes a cls_table with
5731 * different wildcards to be created with the same address, then this function
5732 * will incorrectly skip revalidation. */
5734 table_update_taggable(struct ofproto_dpif *ofproto, uint8_t table_id)
5736 struct table_dpif *table = &ofproto->tables[table_id];
5737 const struct oftable *oftable = &ofproto->up.tables[table_id];
5738 struct cls_table *catchall, *other;
5739 struct cls_table *t;
5741 catchall = other = NULL;
5743 switch (hmap_count(&oftable->cls.tables)) {
5745 /* We could tag this OpenFlow table but it would make the logic a
5746 * little harder and it's a corner case that doesn't seem worth it
5752 HMAP_FOR_EACH (t, hmap_node, &oftable->cls.tables) {
5753 if (cls_table_is_catchall(t)) {
5755 } else if (!other) {
5758 /* Indicate that we can't tag this by setting both tables to
5759 * NULL. (We know that 'catchall' is already NULL.) */
5766 /* Can't tag this table. */
5770 if (table->catchall_table != catchall || table->other_table != other) {
5771 table->catchall_table = catchall;
5772 table->other_table = other;
5773 ofproto->need_revalidate = true;
5777 /* Given 'rule' that has changed in some way (either it is a rule being
5778 * inserted, a rule being deleted, or a rule whose actions are being
5779 * modified), marks facets for revalidation to ensure that packets will be
5780 * forwarded correctly according to the new state of the flow table.
5782 * This function must be called after *each* change to a flow table. See
5783 * the comment on table_update_taggable() for more information. */
5785 rule_invalidate(const struct rule_dpif *rule)
5787 struct ofproto_dpif *ofproto = ofproto_dpif_cast(rule->up.ofproto);
5789 table_update_taggable(ofproto, rule->up.table_id);
5791 if (!ofproto->need_revalidate) {
5792 struct table_dpif *table = &ofproto->tables[rule->up.table_id];
5794 if (table->other_table && rule->tag) {
5795 tag_set_add(&ofproto->revalidate_set, rule->tag);
5797 ofproto->need_revalidate = true;
5803 set_frag_handling(struct ofproto *ofproto_,
5804 enum ofp_config_flags frag_handling)
5806 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
5808 if (frag_handling != OFPC_FRAG_REASM) {
5809 ofproto->need_revalidate = true;
5817 packet_out(struct ofproto *ofproto_, struct ofpbuf *packet,
5818 const struct flow *flow,
5819 const union ofp_action *ofp_actions, size_t n_ofp_actions)
5821 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
5824 if (flow->in_port >= ofproto->max_ports && flow->in_port < OFPP_MAX) {
5825 return OFPERR_NXBRC_BAD_IN_PORT;
5828 error = validate_actions(ofp_actions, n_ofp_actions, flow,
5829 ofproto->max_ports);
5831 struct odputil_keybuf keybuf;
5832 struct ofpbuf *odp_actions;
5833 struct ofproto_push push;
5836 ofpbuf_use_stack(&key, &keybuf, sizeof keybuf);
5837 odp_flow_key_from_flow(&key, flow);
5839 action_xlate_ctx_init(&push.ctx, ofproto, flow, flow->vlan_tci, NULL,
5840 packet_get_tcp_flags(packet, flow), packet);
5842 /* Ensure that resubmits in 'ofp_actions' get accounted to their
5843 * matching rules. */
5845 push.bytes = packet->size;
5846 push.used = time_msec();
5847 push.ctx.resubmit_hook = push_resubmit;
5849 odp_actions = xlate_actions(&push.ctx, ofp_actions, n_ofp_actions);
5850 dpif_execute(ofproto->dpif, key.data, key.size,
5851 odp_actions->data, odp_actions->size, packet);
5852 ofpbuf_delete(odp_actions);
5860 set_netflow(struct ofproto *ofproto_,
5861 const struct netflow_options *netflow_options)
5863 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
5865 if (netflow_options) {
5866 if (!ofproto->netflow) {
5867 ofproto->netflow = netflow_create();
5869 return netflow_set_options(ofproto->netflow, netflow_options);
5871 netflow_destroy(ofproto->netflow);
5872 ofproto->netflow = NULL;
5878 get_netflow_ids(const struct ofproto *ofproto_,
5879 uint8_t *engine_type, uint8_t *engine_id)
5881 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
5883 dpif_get_netflow_ids(ofproto->dpif, engine_type, engine_id);
5887 send_active_timeout(struct ofproto_dpif *ofproto, struct facet *facet)
5889 if (!facet_is_controller_flow(facet) &&
5890 netflow_active_timeout_expired(ofproto->netflow, &facet->nf_flow)) {
5891 struct subfacet *subfacet;
5892 struct ofexpired expired;
5894 LIST_FOR_EACH (subfacet, list_node, &facet->subfacets) {
5895 if (subfacet->installed) {
5896 struct dpif_flow_stats stats;
5898 subfacet_install(subfacet, subfacet->actions,
5899 subfacet->actions_len, &stats);
5900 subfacet_update_stats(subfacet, &stats);
5904 expired.flow = facet->flow;
5905 expired.packet_count = facet->packet_count;
5906 expired.byte_count = facet->byte_count;
5907 expired.used = facet->used;
5908 netflow_expire(ofproto->netflow, &facet->nf_flow, &expired);
5913 send_netflow_active_timeouts(struct ofproto_dpif *ofproto)
5915 struct facet *facet;
5917 HMAP_FOR_EACH (facet, hmap_node, &ofproto->facets) {
5918 send_active_timeout(ofproto, facet);
5922 static struct ofproto_dpif *
5923 ofproto_dpif_lookup(const char *name)
5925 struct ofproto_dpif *ofproto;
5927 HMAP_FOR_EACH_WITH_HASH (ofproto, all_ofproto_dpifs_node,
5928 hash_string(name, 0), &all_ofproto_dpifs) {
5929 if (!strcmp(ofproto->up.name, name)) {
5937 ofproto_unixctl_fdb_flush(struct unixctl_conn *conn, int argc,
5938 const char *argv[], void *aux OVS_UNUSED)
5940 struct ofproto_dpif *ofproto;
5943 ofproto = ofproto_dpif_lookup(argv[1]);
5945 unixctl_command_reply_error(conn, "no such bridge");
5948 mac_learning_flush(ofproto->ml, &ofproto->revalidate_set);
5950 HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
5951 mac_learning_flush(ofproto->ml, &ofproto->revalidate_set);
5955 unixctl_command_reply(conn, "table successfully flushed");
5959 ofproto_unixctl_fdb_show(struct unixctl_conn *conn, int argc OVS_UNUSED,
5960 const char *argv[], void *aux OVS_UNUSED)
5962 struct ds ds = DS_EMPTY_INITIALIZER;
5963 const struct ofproto_dpif *ofproto;
5964 const struct mac_entry *e;
5966 ofproto = ofproto_dpif_lookup(argv[1]);
5968 unixctl_command_reply_error(conn, "no such bridge");
5972 ds_put_cstr(&ds, " port VLAN MAC Age\n");
5973 LIST_FOR_EACH (e, lru_node, &ofproto->ml->lrus) {
5974 struct ofbundle *bundle = e->port.p;
5975 ds_put_format(&ds, "%5d %4d "ETH_ADDR_FMT" %3d\n",
5976 ofbundle_get_a_port(bundle)->odp_port,
5977 e->vlan, ETH_ADDR_ARGS(e->mac),
5978 mac_entry_age(ofproto->ml, e));
5980 unixctl_command_reply(conn, ds_cstr(&ds));
5985 struct action_xlate_ctx ctx;
5991 trace_format_rule(struct ds *result, uint8_t table_id, int level,
5992 const struct rule_dpif *rule)
5994 ds_put_char_multiple(result, '\t', level);
5996 ds_put_cstr(result, "No match\n");
6000 ds_put_format(result, "Rule: table=%"PRIu8" cookie=%#"PRIx64" ",
6001 table_id, ntohll(rule->up.flow_cookie));
6002 cls_rule_format(&rule->up.cr, result);
6003 ds_put_char(result, '\n');
6005 ds_put_char_multiple(result, '\t', level);
6006 ds_put_cstr(result, "OpenFlow ");
6007 ofp_print_actions(result, rule->up.actions, rule->up.n_actions);
6008 ds_put_char(result, '\n');
6012 trace_format_flow(struct ds *result, int level, const char *title,
6013 struct trace_ctx *trace)
6015 ds_put_char_multiple(result, '\t', level);
6016 ds_put_format(result, "%s: ", title);
6017 if (flow_equal(&trace->ctx.flow, &trace->flow)) {
6018 ds_put_cstr(result, "unchanged");
6020 flow_format(result, &trace->ctx.flow);
6021 trace->flow = trace->ctx.flow;
6023 ds_put_char(result, '\n');
6027 trace_format_regs(struct ds *result, int level, const char *title,
6028 struct trace_ctx *trace)
6032 ds_put_char_multiple(result, '\t', level);
6033 ds_put_format(result, "%s:", title);
6034 for (i = 0; i < FLOW_N_REGS; i++) {
6035 ds_put_format(result, " reg%zu=0x%"PRIx32, i, trace->flow.regs[i]);
6037 ds_put_char(result, '\n');
6041 trace_format_odp(struct ds *result, int level, const char *title,
6042 struct trace_ctx *trace)
6044 struct ofpbuf *odp_actions = trace->ctx.odp_actions;
6046 ds_put_char_multiple(result, '\t', level);
6047 ds_put_format(result, "%s: ", title);
6048 format_odp_actions(result, odp_actions->data, odp_actions->size);
6049 ds_put_char(result, '\n');
6053 trace_resubmit(struct action_xlate_ctx *ctx, struct rule_dpif *rule)
6055 struct trace_ctx *trace = CONTAINER_OF(ctx, struct trace_ctx, ctx);
6056 struct ds *result = trace->result;
6058 ds_put_char(result, '\n');
6059 trace_format_flow(result, ctx->recurse + 1, "Resubmitted flow", trace);
6060 trace_format_regs(result, ctx->recurse + 1, "Resubmitted regs", trace);
6061 trace_format_odp(result, ctx->recurse + 1, "Resubmitted odp", trace);
6062 trace_format_rule(result, ctx->table_id, ctx->recurse + 1, rule);
6066 ofproto_unixctl_trace(struct unixctl_conn *conn, int argc, const char *argv[],
6067 void *aux OVS_UNUSED)
6069 const char *dpname = argv[1];
6070 struct ofproto_dpif *ofproto;
6071 struct ofpbuf odp_key;
6072 struct ofpbuf *packet;
6073 ovs_be16 initial_tci;
6079 ofpbuf_init(&odp_key, 0);
6082 ofproto = ofproto_dpif_lookup(dpname);
6084 unixctl_command_reply_error(conn, "Unknown ofproto (use ofproto/list "
6088 if (argc == 3 || (argc == 4 && !strcmp(argv[3], "-generate"))) {
6089 /* ofproto/trace dpname flow [-generate] */
6090 const char *flow_s = argv[2];
6091 const char *generate_s = argv[3];
6094 /* Convert string to datapath key. */
6095 ofpbuf_init(&odp_key, 0);
6096 error = odp_flow_key_from_string(flow_s, NULL, &odp_key);
6098 unixctl_command_reply_error(conn, "Bad flow syntax");
6102 /* Convert odp_key to flow. */
6103 error = ofproto_dpif_extract_flow_key(ofproto, odp_key.data,
6104 odp_key.size, &flow,
6105 &initial_tci, NULL);
6106 if (error == ODP_FIT_ERROR) {
6107 unixctl_command_reply_error(conn, "Invalid flow");
6111 /* Generate a packet, if requested. */
6113 packet = ofpbuf_new(0);
6114 flow_compose(packet, &flow);
6116 } else if (argc == 6) {
6117 /* ofproto/trace dpname priority tun_id in_port packet */
6118 const char *priority_s = argv[2];
6119 const char *tun_id_s = argv[3];
6120 const char *in_port_s = argv[4];
6121 const char *packet_s = argv[5];
6122 uint16_t in_port = ofp_port_to_odp_port(atoi(in_port_s));
6123 ovs_be64 tun_id = htonll(strtoull(tun_id_s, NULL, 0));
6124 uint32_t priority = atoi(priority_s);
6127 msg = eth_from_hex(packet_s, &packet);
6129 unixctl_command_reply_error(conn, msg);
6133 ds_put_cstr(&result, "Packet: ");
6134 s = ofp_packet_to_string(packet->data, packet->size);
6135 ds_put_cstr(&result, s);
6138 flow_extract(packet, priority, tun_id, in_port, &flow);
6139 initial_tci = flow.vlan_tci;
6141 unixctl_command_reply_error(conn, "Bad command syntax");
6145 ofproto_trace(ofproto, &flow, packet, initial_tci, &result);
6146 unixctl_command_reply(conn, ds_cstr(&result));
6149 ds_destroy(&result);
6150 ofpbuf_delete(packet);
6151 ofpbuf_uninit(&odp_key);
6155 ofproto_trace(struct ofproto_dpif *ofproto, const struct flow *flow,
6156 const struct ofpbuf *packet, ovs_be16 initial_tci,
6159 struct rule_dpif *rule;
6161 ds_put_cstr(ds, "Flow: ");
6162 flow_format(ds, flow);
6163 ds_put_char(ds, '\n');
6165 rule = rule_dpif_lookup(ofproto, flow, 0);
6166 trace_format_rule(ds, 0, 0, rule);
6168 struct trace_ctx trace;
6169 struct ofpbuf *odp_actions;
6172 tcp_flags = packet ? packet_get_tcp_flags(packet, flow) : 0;
6175 action_xlate_ctx_init(&trace.ctx, ofproto, flow, initial_tci,
6176 rule, tcp_flags, packet);
6177 trace.ctx.resubmit_hook = trace_resubmit;
6178 odp_actions = xlate_actions(&trace.ctx,
6179 rule->up.actions, rule->up.n_actions);
6181 ds_put_char(ds, '\n');
6182 trace_format_flow(ds, 0, "Final flow", &trace);
6183 ds_put_cstr(ds, "Datapath actions: ");
6184 format_odp_actions(ds, odp_actions->data, odp_actions->size);
6185 ofpbuf_delete(odp_actions);
6187 if (!trace.ctx.may_set_up_flow) {
6189 ds_put_cstr(ds, "\nThis flow is not cachable.");
6191 ds_put_cstr(ds, "\nThe datapath actions are incomplete--"
6192 "for complete actions, please supply a packet.");
6199 ofproto_dpif_clog(struct unixctl_conn *conn OVS_UNUSED, int argc OVS_UNUSED,
6200 const char *argv[] OVS_UNUSED, void *aux OVS_UNUSED)
6203 unixctl_command_reply(conn, NULL);
6207 ofproto_dpif_unclog(struct unixctl_conn *conn OVS_UNUSED, int argc OVS_UNUSED,
6208 const char *argv[] OVS_UNUSED, void *aux OVS_UNUSED)
6211 unixctl_command_reply(conn, NULL);
6214 /* Runs a self-check of flow translations in 'ofproto'. Appends a message to
6215 * 'reply' describing the results. */
6217 ofproto_dpif_self_check__(struct ofproto_dpif *ofproto, struct ds *reply)
6219 struct facet *facet;
6223 HMAP_FOR_EACH (facet, hmap_node, &ofproto->facets) {
6224 if (!facet_check_consistency(facet)) {
6229 ofproto->need_revalidate = true;
6233 ds_put_format(reply, "%s: self-check failed (%d errors)\n",
6234 ofproto->up.name, errors);
6236 ds_put_format(reply, "%s: self-check passed\n", ofproto->up.name);
6241 ofproto_dpif_self_check(struct unixctl_conn *conn,
6242 int argc, const char *argv[], void *aux OVS_UNUSED)
6244 struct ds reply = DS_EMPTY_INITIALIZER;
6245 struct ofproto_dpif *ofproto;
6248 ofproto = ofproto_dpif_lookup(argv[1]);
6250 unixctl_command_reply_error(conn, "Unknown ofproto (use "
6251 "ofproto/list for help)");
6254 ofproto_dpif_self_check__(ofproto, &reply);
6256 HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
6257 ofproto_dpif_self_check__(ofproto, &reply);
6261 unixctl_command_reply(conn, ds_cstr(&reply));
6266 ofproto_dpif_unixctl_init(void)
6268 static bool registered;
6274 unixctl_command_register(
6276 "bridge {tun_id in_port packet | odp_flow [-generate]}",
6277 2, 5, ofproto_unixctl_trace, NULL);
6278 unixctl_command_register("fdb/flush", "[bridge]", 0, 1,
6279 ofproto_unixctl_fdb_flush, NULL);
6280 unixctl_command_register("fdb/show", "bridge", 1, 1,
6281 ofproto_unixctl_fdb_show, NULL);
6282 unixctl_command_register("ofproto/clog", "", 0, 0,
6283 ofproto_dpif_clog, NULL);
6284 unixctl_command_register("ofproto/unclog", "", 0, 0,
6285 ofproto_dpif_unclog, NULL);
6286 unixctl_command_register("ofproto/self-check", "[bridge]", 0, 1,
6287 ofproto_dpif_self_check, NULL);
6290 /* Linux VLAN device support (e.g. "eth0.10" for VLAN 10.)
6292 * This is deprecated. It is only for compatibility with broken device drivers
6293 * in old versions of Linux that do not properly support VLANs when VLAN
6294 * devices are not used. When broken device drivers are no longer in
6295 * widespread use, we will delete these interfaces. */
6298 set_realdev(struct ofport *ofport_, uint16_t realdev_ofp_port, int vid)
6300 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport_->ofproto);
6301 struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
6303 if (realdev_ofp_port == ofport->realdev_ofp_port
6304 && vid == ofport->vlandev_vid) {
6308 ofproto->need_revalidate = true;
6310 if (ofport->realdev_ofp_port) {
6313 if (realdev_ofp_port && ofport->bundle) {
6314 /* vlandevs are enslaved to their realdevs, so they are not allowed to
6315 * themselves be part of a bundle. */
6316 bundle_set(ofport->up.ofproto, ofport->bundle, NULL);
6319 ofport->realdev_ofp_port = realdev_ofp_port;
6320 ofport->vlandev_vid = vid;
6322 if (realdev_ofp_port) {
6323 vsp_add(ofport, realdev_ofp_port, vid);
6330 hash_realdev_vid(uint16_t realdev_ofp_port, int vid)
6332 return hash_2words(realdev_ofp_port, vid);
6335 /* Returns the ODP port number of the Linux VLAN device that corresponds to
6336 * 'vlan_tci' on the network device with port number 'realdev_odp_port' in
6337 * 'ofproto'. For example, given 'realdev_odp_port' of eth0 and 'vlan_tci' 9,
6338 * it would return the port number of eth0.9.
6340 * Unless VLAN splinters are enabled for port 'realdev_odp_port', this
6341 * function just returns its 'realdev_odp_port' argument. */
6343 vsp_realdev_to_vlandev(const struct ofproto_dpif *ofproto,
6344 uint32_t realdev_odp_port, ovs_be16 vlan_tci)
6346 if (!hmap_is_empty(&ofproto->realdev_vid_map)) {
6347 uint16_t realdev_ofp_port = odp_port_to_ofp_port(realdev_odp_port);
6348 int vid = vlan_tci_to_vid(vlan_tci);
6349 const struct vlan_splinter *vsp;
6351 HMAP_FOR_EACH_WITH_HASH (vsp, realdev_vid_node,
6352 hash_realdev_vid(realdev_ofp_port, vid),
6353 &ofproto->realdev_vid_map) {
6354 if (vsp->realdev_ofp_port == realdev_ofp_port
6355 && vsp->vid == vid) {
6356 return ofp_port_to_odp_port(vsp->vlandev_ofp_port);
6360 return realdev_odp_port;
6363 static struct vlan_splinter *
6364 vlandev_find(const struct ofproto_dpif *ofproto, uint16_t vlandev_ofp_port)
6366 struct vlan_splinter *vsp;
6368 HMAP_FOR_EACH_WITH_HASH (vsp, vlandev_node, hash_int(vlandev_ofp_port, 0),
6369 &ofproto->vlandev_map) {
6370 if (vsp->vlandev_ofp_port == vlandev_ofp_port) {
6378 /* Returns the OpenFlow port number of the "real" device underlying the Linux
6379 * VLAN device with OpenFlow port number 'vlandev_ofp_port' and stores the
6380 * VLAN VID of the Linux VLAN device in '*vid'. For example, given
6381 * 'vlandev_ofp_port' of eth0.9, it would return the OpenFlow port number of
6382 * eth0 and store 9 in '*vid'.
6384 * Returns 0 and does not modify '*vid' if 'vlandev_ofp_port' is not a Linux
6385 * VLAN device. Unless VLAN splinters are enabled, this is what this function
6388 vsp_vlandev_to_realdev(const struct ofproto_dpif *ofproto,
6389 uint16_t vlandev_ofp_port, int *vid)
6391 if (!hmap_is_empty(&ofproto->vlandev_map)) {
6392 const struct vlan_splinter *vsp;
6394 vsp = vlandev_find(ofproto, vlandev_ofp_port);
6399 return vsp->realdev_ofp_port;
6406 vsp_remove(struct ofport_dpif *port)
6408 struct ofproto_dpif *ofproto = ofproto_dpif_cast(port->up.ofproto);
6409 struct vlan_splinter *vsp;
6411 vsp = vlandev_find(ofproto, port->up.ofp_port);
6413 hmap_remove(&ofproto->vlandev_map, &vsp->vlandev_node);
6414 hmap_remove(&ofproto->realdev_vid_map, &vsp->realdev_vid_node);
6417 port->realdev_ofp_port = 0;
6419 VLOG_ERR("missing vlan device record");
6424 vsp_add(struct ofport_dpif *port, uint16_t realdev_ofp_port, int vid)
6426 struct ofproto_dpif *ofproto = ofproto_dpif_cast(port->up.ofproto);
6428 if (!vsp_vlandev_to_realdev(ofproto, port->up.ofp_port, NULL)
6429 && (vsp_realdev_to_vlandev(ofproto, realdev_ofp_port, htons(vid))
6430 == realdev_ofp_port)) {
6431 struct vlan_splinter *vsp;
6433 vsp = xmalloc(sizeof *vsp);
6434 hmap_insert(&ofproto->vlandev_map, &vsp->vlandev_node,
6435 hash_int(port->up.ofp_port, 0));
6436 hmap_insert(&ofproto->realdev_vid_map, &vsp->realdev_vid_node,
6437 hash_realdev_vid(realdev_ofp_port, vid));
6438 vsp->realdev_ofp_port = realdev_ofp_port;
6439 vsp->vlandev_ofp_port = port->up.ofp_port;
6442 port->realdev_ofp_port = realdev_ofp_port;
6444 VLOG_ERR("duplicate vlan device record");
6448 const struct ofproto_class ofproto_dpif_class = {
6477 port_is_lacp_current,
6478 NULL, /* rule_choose_table */
6485 rule_modify_actions,
6493 get_cfm_remote_mpids,
6497 get_stp_port_status,
6504 is_mirror_output_bundle,
6505 forward_bpdu_changed,