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 uint32_t rule_calculate_tag(const struct flow *,
113 const struct flow_wildcards *,
115 static void rule_invalidate(const struct rule_dpif *);
117 #define MAX_MIRRORS 32
118 typedef uint32_t mirror_mask_t;
119 #define MIRROR_MASK_C(X) UINT32_C(X)
120 BUILD_ASSERT_DECL(sizeof(mirror_mask_t) * CHAR_BIT >= MAX_MIRRORS);
122 struct ofproto_dpif *ofproto; /* Owning ofproto. */
123 size_t idx; /* In ofproto's "mirrors" array. */
124 void *aux; /* Key supplied by ofproto's client. */
125 char *name; /* Identifier for log messages. */
127 /* Selection criteria. */
128 struct hmapx srcs; /* Contains "struct ofbundle *"s. */
129 struct hmapx dsts; /* Contains "struct ofbundle *"s. */
130 unsigned long *vlans; /* Bitmap of chosen VLANs, NULL selects all. */
132 /* Output (exactly one of out == NULL and out_vlan == -1 is true). */
133 struct ofbundle *out; /* Output port or NULL. */
134 int out_vlan; /* Output VLAN or -1. */
135 mirror_mask_t dup_mirrors; /* Bitmap of mirrors with the same output. */
138 int64_t packet_count; /* Number of packets sent. */
139 int64_t byte_count; /* Number of bytes sent. */
142 static void mirror_destroy(struct ofmirror *);
143 static void update_mirror_stats(struct ofproto_dpif *ofproto,
144 mirror_mask_t mirrors,
145 uint64_t packets, uint64_t bytes);
148 struct ofproto_dpif *ofproto; /* Owning ofproto. */
149 struct hmap_node hmap_node; /* In struct ofproto's "bundles" hmap. */
150 void *aux; /* Key supplied by ofproto's client. */
151 char *name; /* Identifier for log messages. */
154 struct list ports; /* Contains "struct ofport"s. */
155 enum port_vlan_mode vlan_mode; /* VLAN mode */
156 int vlan; /* -1=trunk port, else a 12-bit VLAN ID. */
157 unsigned long *trunks; /* Bitmap of trunked VLANs, if 'vlan' == -1.
158 * NULL if all VLANs are trunked. */
159 struct lacp *lacp; /* LACP if LACP is enabled, otherwise NULL. */
160 struct bond *bond; /* Nonnull iff more than one port. */
161 bool use_priority_tags; /* Use 802.1p tag for frames in VLAN 0? */
164 bool floodable; /* True if no port has OFPPC_NO_FLOOD set. */
166 /* Port mirroring info. */
167 mirror_mask_t src_mirrors; /* Mirrors triggered when packet received. */
168 mirror_mask_t dst_mirrors; /* Mirrors triggered when packet sent. */
169 mirror_mask_t mirror_out; /* Mirrors that output to this bundle. */
172 static void bundle_remove(struct ofport *);
173 static void bundle_update(struct ofbundle *);
174 static void bundle_destroy(struct ofbundle *);
175 static void bundle_del_port(struct ofport_dpif *);
176 static void bundle_run(struct ofbundle *);
177 static void bundle_wait(struct ofbundle *);
178 static struct ofbundle *lookup_input_bundle(struct ofproto_dpif *,
179 uint16_t in_port, bool warn);
181 /* A controller may use OFPP_NONE as the ingress port to indicate that
182 * it did not arrive on a "real" port. 'ofpp_none_bundle' exists for
183 * when an input bundle is needed for validation (e.g., mirroring or
184 * OFPP_NORMAL processing). It is not connected to an 'ofproto' or have
185 * any 'port' structs, so care must be taken when dealing with it. */
186 static struct ofbundle ofpp_none_bundle = {
188 .vlan_mode = PORT_VLAN_TRUNK
191 static void stp_run(struct ofproto_dpif *ofproto);
192 static void stp_wait(struct ofproto_dpif *ofproto);
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 MAC learning and NXAST_LEARN actions execute? We
213 * want to execute them if we are actually processing a packet, or if we
214 * are accounting for packets that the datapath has processed, but not if
215 * we are just revalidating. */
218 /* The rule that we are currently translating, or NULL. */
219 struct rule_dpif *rule;
221 /* Union of the set of TCP flags seen so far in this flow. (Used only by
222 * NXAST_FIN_TIMEOUT. Set to zero to avoid updating updating rules'
226 /* If nonnull, called just before executing a resubmit action.
228 * This is normally null so the client has to set it manually after
229 * calling action_xlate_ctx_init(). */
230 void (*resubmit_hook)(struct action_xlate_ctx *, struct rule_dpif *);
232 /* xlate_actions() initializes and uses these members. The client might want
233 * to look at them after it returns. */
235 struct ofpbuf *odp_actions; /* Datapath actions. */
236 tag_type tags; /* Tags associated with actions. */
237 bool may_set_up_flow; /* True ordinarily; false if the actions must
238 * be reassessed for every packet. */
239 bool has_learn; /* Actions include NXAST_LEARN? */
240 bool has_normal; /* Actions output to OFPP_NORMAL? */
241 bool has_fin_timeout; /* Actions include NXAST_FIN_TIMEOUT? */
242 uint16_t nf_output_iface; /* Output interface index for NetFlow. */
243 mirror_mask_t mirrors; /* Bitmap of associated mirrors. */
245 /* xlate_actions() initializes and uses these members, but the client has no
246 * reason to look at them. */
248 int recurse; /* Recursion level, via xlate_table_action. */
249 struct flow base_flow; /* Flow at the last commit. */
250 uint32_t orig_skb_priority; /* Priority when packet arrived. */
251 uint8_t table_id; /* OpenFlow table ID where flow was found. */
252 uint32_t sflow_n_outputs; /* Number of output ports. */
253 uint16_t sflow_odp_port; /* Output port for composing sFlow action. */
254 uint16_t user_cookie_offset;/* Used for user_action_cookie fixup. */
255 bool exit; /* No further actions should be processed. */
258 static void action_xlate_ctx_init(struct action_xlate_ctx *,
259 struct ofproto_dpif *, const struct flow *,
260 ovs_be16 initial_tci, struct rule_dpif *,
261 uint8_t tcp_flags, const struct ofpbuf *);
262 static struct ofpbuf *xlate_actions(struct action_xlate_ctx *,
263 const union ofp_action *in, size_t n_in);
265 /* An exact-match instantiation of an OpenFlow flow.
267 * A facet associates a "struct flow", which represents the Open vSwitch
268 * userspace idea of an exact-match flow, with one or more subfacets. Each
269 * subfacet tracks the datapath's idea of the exact-match flow equivalent to
270 * the facet. When the kernel module (or other dpif implementation) and Open
271 * vSwitch userspace agree on the definition of a flow key, there is exactly
272 * one subfacet per facet. If the dpif implementation supports more-specific
273 * flow matching than userspace, however, a facet can have more than one
274 * subfacet, each of which corresponds to some distinction in flow that
275 * userspace simply doesn't understand.
277 * Flow expiration works in terms of subfacets, so a facet must have at least
278 * one subfacet or it will never expire, leaking memory. */
281 struct hmap_node hmap_node; /* In owning ofproto's 'facets' hmap. */
282 struct list list_node; /* In owning rule's 'facets' list. */
283 struct rule_dpif *rule; /* Owning rule. */
286 struct list subfacets;
287 long long int used; /* Time last used; time created if not used. */
294 * - Do include packets and bytes sent "by hand", e.g. with
297 * - Do include packets and bytes that were obtained from the datapath
298 * when a subfacet's statistics were reset (e.g. dpif_flow_put() with
299 * DPIF_FP_ZERO_STATS).
301 * - Do not include packets or bytes that can be obtained from the
302 * datapath for any existing subfacet.
304 uint64_t packet_count; /* Number of packets received. */
305 uint64_t byte_count; /* Number of bytes received. */
307 /* Resubmit statistics. */
308 uint64_t prev_packet_count; /* Number of packets from last stats push. */
309 uint64_t prev_byte_count; /* Number of bytes from last stats push. */
310 long long int prev_used; /* Used time from last stats push. */
313 uint64_t accounted_bytes; /* Bytes processed by facet_account(). */
314 struct netflow_flow nf_flow; /* Per-flow NetFlow tracking data. */
315 uint8_t tcp_flags; /* TCP flags seen for this 'rule'. */
317 /* Properties of datapath actions.
319 * Every subfacet has its own actions because actions can differ slightly
320 * between splintered and non-splintered subfacets due to the VLAN tag
321 * being initially different (present vs. absent). All of them have these
322 * properties in common so we just store one copy of them here. */
323 bool may_install; /* Reassess actions for every packet? */
324 bool has_learn; /* Actions include NXAST_LEARN? */
325 bool has_normal; /* Actions output to OFPP_NORMAL? */
326 bool has_fin_timeout; /* Actions include NXAST_FIN_TIMEOUT? */
327 tag_type tags; /* Tags that would require revalidation. */
328 mirror_mask_t mirrors; /* Bitmap of dependent mirrors. */
331 static struct facet *facet_create(struct rule_dpif *, const struct flow *);
332 static void facet_remove(struct facet *);
333 static void facet_free(struct facet *);
335 static struct facet *facet_find(struct ofproto_dpif *, const struct flow *);
336 static struct facet *facet_lookup_valid(struct ofproto_dpif *,
337 const struct flow *);
338 static bool facet_revalidate(struct facet *);
339 static bool facet_check_consistency(struct facet *);
341 static void facet_flush_stats(struct facet *);
343 static void facet_update_time(struct facet *, long long int used);
344 static void facet_reset_counters(struct facet *);
345 static void facet_push_stats(struct facet *);
346 static void facet_account(struct facet *);
348 static bool facet_is_controller_flow(struct facet *);
350 /* A dpif flow and actions associated with a facet.
352 * See also the large comment on struct facet. */
355 struct hmap_node hmap_node; /* In struct ofproto_dpif 'subfacets' list. */
356 struct list list_node; /* In struct facet's 'facets' list. */
357 struct facet *facet; /* Owning facet. */
361 * To save memory in the common case, 'key' is NULL if 'key_fitness' is
362 * ODP_FIT_PERFECT, that is, odp_flow_key_from_flow() can accurately
363 * regenerate the ODP flow key from ->facet->flow. */
364 enum odp_key_fitness key_fitness;
368 long long int used; /* Time last used; time created if not used. */
370 uint64_t dp_packet_count; /* Last known packet count in the datapath. */
371 uint64_t dp_byte_count; /* Last known byte count in the datapath. */
375 * These should be essentially identical for every subfacet in a facet, but
376 * may differ in trivial ways due to VLAN splinters. */
377 size_t actions_len; /* Number of bytes in actions[]. */
378 struct nlattr *actions; /* Datapath actions. */
380 bool installed; /* Installed in datapath? */
382 /* This value is normally the same as ->facet->flow.vlan_tci. Only VLAN
383 * splinters can cause it to differ. This value should be removed when
384 * the VLAN splinters feature is no longer needed. */
385 ovs_be16 initial_tci; /* Initial VLAN TCI value. */
388 static struct subfacet *subfacet_create(struct facet *, enum odp_key_fitness,
389 const struct nlattr *key,
390 size_t key_len, ovs_be16 initial_tci);
391 static struct subfacet *subfacet_find(struct ofproto_dpif *,
392 const struct nlattr *key, size_t key_len);
393 static void subfacet_destroy(struct subfacet *);
394 static void subfacet_destroy__(struct subfacet *);
395 static void subfacet_get_key(struct subfacet *, struct odputil_keybuf *,
397 static void subfacet_reset_dp_stats(struct subfacet *,
398 struct dpif_flow_stats *);
399 static void subfacet_update_time(struct subfacet *, long long int used);
400 static void subfacet_update_stats(struct subfacet *,
401 const struct dpif_flow_stats *);
402 static void subfacet_make_actions(struct subfacet *,
403 const struct ofpbuf *packet);
404 static int subfacet_install(struct subfacet *,
405 const struct nlattr *actions, size_t actions_len,
406 struct dpif_flow_stats *);
407 static void subfacet_uninstall(struct subfacet *);
413 struct ofbundle *bundle; /* Bundle that contains this port, if any. */
414 struct list bundle_node; /* In struct ofbundle's "ports" list. */
415 struct cfm *cfm; /* Connectivity Fault Management, if any. */
416 tag_type tag; /* Tag associated with this port. */
417 uint32_t bond_stable_id; /* stable_id to use as bond slave, or 0. */
418 bool may_enable; /* May be enabled in bonds. */
421 struct stp_port *stp_port; /* Spanning Tree Protocol, if any. */
422 enum stp_state stp_state; /* Always STP_DISABLED if STP not in use. */
423 long long int stp_state_entered;
425 struct hmap priorities; /* Map of attached 'priority_to_dscp's. */
427 /* Linux VLAN device support (e.g. "eth0.10" for VLAN 10.)
429 * This is deprecated. It is only for compatibility with broken device
430 * drivers in old versions of Linux that do not properly support VLANs when
431 * VLAN devices are not used. When broken device drivers are no longer in
432 * widespread use, we will delete these interfaces. */
433 uint16_t realdev_ofp_port;
437 /* Node in 'ofport_dpif''s 'priorities' map. Used to maintain a map from
438 * 'priority' (the datapath's term for QoS queue) to the dscp bits which all
439 * traffic egressing the 'ofport' with that priority should be marked with. */
440 struct priority_to_dscp {
441 struct hmap_node hmap_node; /* Node in 'ofport_dpif''s 'priorities' map. */
442 uint32_t priority; /* Priority of this queue (see struct flow). */
444 uint8_t dscp; /* DSCP bits to mark outgoing traffic with. */
447 /* Linux VLAN device support (e.g. "eth0.10" for VLAN 10.)
449 * This is deprecated. It is only for compatibility with broken device drivers
450 * in old versions of Linux that do not properly support VLANs when VLAN
451 * devices are not used. When broken device drivers are no longer in
452 * widespread use, we will delete these interfaces. */
453 struct vlan_splinter {
454 struct hmap_node realdev_vid_node;
455 struct hmap_node vlandev_node;
456 uint16_t realdev_ofp_port;
457 uint16_t vlandev_ofp_port;
461 static uint32_t vsp_realdev_to_vlandev(const struct ofproto_dpif *,
462 uint32_t realdev, ovs_be16 vlan_tci);
463 static uint16_t vsp_vlandev_to_realdev(const struct ofproto_dpif *,
464 uint16_t vlandev, int *vid);
465 static void vsp_remove(struct ofport_dpif *);
466 static void vsp_add(struct ofport_dpif *, uint16_t realdev_ofp_port, int vid);
468 static struct ofport_dpif *
469 ofport_dpif_cast(const struct ofport *ofport)
471 assert(ofport->ofproto->ofproto_class == &ofproto_dpif_class);
472 return ofport ? CONTAINER_OF(ofport, struct ofport_dpif, up) : NULL;
475 static void port_run(struct ofport_dpif *);
476 static void port_wait(struct ofport_dpif *);
477 static int set_cfm(struct ofport *, const struct cfm_settings *);
478 static void ofport_clear_priorities(struct ofport_dpif *);
480 struct dpif_completion {
481 struct list list_node;
482 struct ofoperation *op;
485 /* Extra information about a classifier table.
486 * Currently used just for optimized flow revalidation. */
488 /* If either of these is nonnull, then this table has a form that allows
489 * flows to be tagged to avoid revalidating most flows for the most common
490 * kinds of flow table changes. */
491 struct cls_table *catchall_table; /* Table that wildcards all fields. */
492 struct cls_table *other_table; /* Table with any other wildcard set. */
493 uint32_t basis; /* Keeps each table's tags separate. */
496 struct ofproto_dpif {
497 struct hmap_node all_ofproto_dpifs_node; /* In 'all_ofproto_dpifs'. */
506 struct netflow *netflow;
507 struct dpif_sflow *sflow;
508 struct hmap bundles; /* Contains "struct ofbundle"s. */
509 struct mac_learning *ml;
510 struct ofmirror *mirrors[MAX_MIRRORS];
511 bool has_bonded_bundles;
514 struct timer next_expiration;
518 struct hmap subfacets;
521 struct table_dpif tables[N_TABLES];
522 bool need_revalidate;
523 struct tag_set revalidate_set;
525 /* Support for debugging async flow mods. */
526 struct list completions;
528 bool has_bundle_action; /* True when the first bundle action appears. */
529 struct netdev_stats stats; /* To account packets generated and consumed in
534 long long int stp_last_tick;
536 /* VLAN splinters. */
537 struct hmap realdev_vid_map; /* (realdev,vid) -> vlandev. */
538 struct hmap vlandev_map; /* vlandev -> (realdev,vid). */
541 /* Defer flow mod completion until "ovs-appctl ofproto/unclog"? (Useful only
542 * for debugging the asynchronous flow_mod implementation.) */
545 /* All existing ofproto_dpif instances, indexed by ->up.name. */
546 static struct hmap all_ofproto_dpifs = HMAP_INITIALIZER(&all_ofproto_dpifs);
548 static void ofproto_dpif_unixctl_init(void);
550 static struct ofproto_dpif *
551 ofproto_dpif_cast(const struct ofproto *ofproto)
553 assert(ofproto->ofproto_class == &ofproto_dpif_class);
554 return CONTAINER_OF(ofproto, struct ofproto_dpif, up);
557 static struct ofport_dpif *get_ofp_port(struct ofproto_dpif *,
559 static struct ofport_dpif *get_odp_port(struct ofproto_dpif *,
562 /* Packet processing. */
563 static void update_learning_table(struct ofproto_dpif *,
564 const struct flow *, int vlan,
567 #define FLOW_MISS_MAX_BATCH 50
568 static int handle_upcalls(struct ofproto_dpif *, unsigned int max_batch);
570 /* Flow expiration. */
571 static int expire(struct ofproto_dpif *);
574 static void send_netflow_active_timeouts(struct ofproto_dpif *);
577 static int send_packet(const struct ofport_dpif *, struct ofpbuf *packet);
579 compose_sflow_action(const struct ofproto_dpif *, struct ofpbuf *odp_actions,
580 const struct flow *, uint32_t odp_port);
581 static void add_mirror_actions(struct action_xlate_ctx *ctx,
582 const struct flow *flow);
583 /* Global variables. */
584 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
586 /* Factory functions. */
589 enumerate_types(struct sset *types)
591 dp_enumerate_types(types);
595 enumerate_names(const char *type, struct sset *names)
597 return dp_enumerate_names(type, names);
601 del(const char *type, const char *name)
606 error = dpif_open(name, type, &dpif);
608 error = dpif_delete(dpif);
614 /* Basic life-cycle. */
616 static struct ofproto *
619 struct ofproto_dpif *ofproto = xmalloc(sizeof *ofproto);
624 dealloc(struct ofproto *ofproto_)
626 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
631 construct(struct ofproto *ofproto_)
633 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
634 const char *name = ofproto->up.name;
638 error = dpif_create_and_open(name, ofproto->up.type, &ofproto->dpif);
640 VLOG_ERR("failed to open datapath %s: %s", name, strerror(error));
644 ofproto->max_ports = dpif_get_max_ports(ofproto->dpif);
645 ofproto->n_matches = 0;
647 dpif_flow_flush(ofproto->dpif);
648 dpif_recv_purge(ofproto->dpif);
650 error = dpif_recv_set(ofproto->dpif, true);
652 VLOG_ERR("failed to listen on datapath %s: %s", name, strerror(error));
653 dpif_close(ofproto->dpif);
657 ofproto->netflow = NULL;
658 ofproto->sflow = NULL;
660 hmap_init(&ofproto->bundles);
661 ofproto->ml = mac_learning_create(MAC_ENTRY_DEFAULT_IDLE_TIME);
662 for (i = 0; i < MAX_MIRRORS; i++) {
663 ofproto->mirrors[i] = NULL;
665 ofproto->has_bonded_bundles = false;
667 timer_set_duration(&ofproto->next_expiration, 1000);
669 hmap_init(&ofproto->facets);
670 hmap_init(&ofproto->subfacets);
672 for (i = 0; i < N_TABLES; i++) {
673 struct table_dpif *table = &ofproto->tables[i];
675 table->catchall_table = NULL;
676 table->other_table = NULL;
677 table->basis = random_uint32();
679 ofproto->need_revalidate = false;
680 tag_set_init(&ofproto->revalidate_set);
682 list_init(&ofproto->completions);
684 ofproto_dpif_unixctl_init();
686 ofproto->has_bundle_action = false;
688 hmap_init(&ofproto->vlandev_map);
689 hmap_init(&ofproto->realdev_vid_map);
691 hmap_insert(&all_ofproto_dpifs, &ofproto->all_ofproto_dpifs_node,
692 hash_string(ofproto->up.name, 0));
693 memset(&ofproto->stats, 0, sizeof ofproto->stats);
695 ofproto_init_tables(ofproto_, N_TABLES);
701 complete_operations(struct ofproto_dpif *ofproto)
703 struct dpif_completion *c, *next;
705 LIST_FOR_EACH_SAFE (c, next, list_node, &ofproto->completions) {
706 ofoperation_complete(c->op, 0);
707 list_remove(&c->list_node);
713 destruct(struct ofproto *ofproto_)
715 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
716 struct rule_dpif *rule, *next_rule;
717 struct oftable *table;
720 hmap_remove(&all_ofproto_dpifs, &ofproto->all_ofproto_dpifs_node);
721 complete_operations(ofproto);
723 OFPROTO_FOR_EACH_TABLE (table, &ofproto->up) {
724 struct cls_cursor cursor;
726 cls_cursor_init(&cursor, &table->cls, NULL);
727 CLS_CURSOR_FOR_EACH_SAFE (rule, next_rule, up.cr, &cursor) {
728 ofproto_rule_destroy(&rule->up);
732 for (i = 0; i < MAX_MIRRORS; i++) {
733 mirror_destroy(ofproto->mirrors[i]);
736 netflow_destroy(ofproto->netflow);
737 dpif_sflow_destroy(ofproto->sflow);
738 hmap_destroy(&ofproto->bundles);
739 mac_learning_destroy(ofproto->ml);
741 hmap_destroy(&ofproto->facets);
742 hmap_destroy(&ofproto->subfacets);
744 hmap_destroy(&ofproto->vlandev_map);
745 hmap_destroy(&ofproto->realdev_vid_map);
747 dpif_close(ofproto->dpif);
751 run_fast(struct ofproto *ofproto_)
753 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
756 /* Handle one or more batches of upcalls, until there's nothing left to do
757 * or until we do a fixed total amount of work.
759 * We do work in batches because it can be much cheaper to set up a number
760 * of flows and fire off their patches all at once. We do multiple batches
761 * because in some cases handling a packet can cause another packet to be
762 * queued almost immediately as part of the return flow. Both
763 * optimizations can make major improvements on some benchmarks and
764 * presumably for real traffic as well. */
766 while (work < FLOW_MISS_MAX_BATCH) {
767 int retval = handle_upcalls(ofproto, FLOW_MISS_MAX_BATCH - work);
777 run(struct ofproto *ofproto_)
779 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
780 struct ofport_dpif *ofport;
781 struct ofbundle *bundle;
785 complete_operations(ofproto);
787 dpif_run(ofproto->dpif);
789 error = run_fast(ofproto_);
794 if (timer_expired(&ofproto->next_expiration)) {
795 int delay = expire(ofproto);
796 timer_set_duration(&ofproto->next_expiration, delay);
799 if (ofproto->netflow) {
800 if (netflow_run(ofproto->netflow)) {
801 send_netflow_active_timeouts(ofproto);
804 if (ofproto->sflow) {
805 dpif_sflow_run(ofproto->sflow);
808 HMAP_FOR_EACH (ofport, up.hmap_node, &ofproto->up.ports) {
811 HMAP_FOR_EACH (bundle, hmap_node, &ofproto->bundles) {
816 mac_learning_run(ofproto->ml, &ofproto->revalidate_set);
818 /* Now revalidate if there's anything to do. */
819 if (ofproto->need_revalidate
820 || !tag_set_is_empty(&ofproto->revalidate_set)) {
821 struct tag_set revalidate_set = ofproto->revalidate_set;
822 bool revalidate_all = ofproto->need_revalidate;
823 struct facet *facet, *next;
825 /* Clear the revalidation flags. */
826 tag_set_init(&ofproto->revalidate_set);
827 ofproto->need_revalidate = false;
829 HMAP_FOR_EACH_SAFE (facet, next, hmap_node, &ofproto->facets) {
831 || tag_set_intersects(&revalidate_set, facet->tags)) {
832 facet_revalidate(facet);
837 /* Check the consistency of a random facet, to aid debugging. */
838 if (!hmap_is_empty(&ofproto->facets) && !ofproto->need_revalidate) {
841 facet = CONTAINER_OF(hmap_random_node(&ofproto->facets),
842 struct facet, hmap_node);
843 if (!tag_set_intersects(&ofproto->revalidate_set, facet->tags)) {
844 if (!facet_check_consistency(facet)) {
845 ofproto->need_revalidate = true;
854 wait(struct ofproto *ofproto_)
856 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
857 struct ofport_dpif *ofport;
858 struct ofbundle *bundle;
860 if (!clogged && !list_is_empty(&ofproto->completions)) {
861 poll_immediate_wake();
864 dpif_wait(ofproto->dpif);
865 dpif_recv_wait(ofproto->dpif);
866 if (ofproto->sflow) {
867 dpif_sflow_wait(ofproto->sflow);
869 if (!tag_set_is_empty(&ofproto->revalidate_set)) {
870 poll_immediate_wake();
872 HMAP_FOR_EACH (ofport, up.hmap_node, &ofproto->up.ports) {
875 HMAP_FOR_EACH (bundle, hmap_node, &ofproto->bundles) {
878 if (ofproto->netflow) {
879 netflow_wait(ofproto->netflow);
881 mac_learning_wait(ofproto->ml);
883 if (ofproto->need_revalidate) {
884 /* Shouldn't happen, but if it does just go around again. */
885 VLOG_DBG_RL(&rl, "need revalidate in ofproto_wait_cb()");
886 poll_immediate_wake();
888 timer_wait(&ofproto->next_expiration);
893 flush(struct ofproto *ofproto_)
895 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
896 struct facet *facet, *next_facet;
898 HMAP_FOR_EACH_SAFE (facet, next_facet, hmap_node, &ofproto->facets) {
899 /* Mark the facet as not installed so that facet_remove() doesn't
900 * bother trying to uninstall it. There is no point in uninstalling it
901 * individually since we are about to blow away all the facets with
902 * dpif_flow_flush(). */
903 struct subfacet *subfacet;
905 LIST_FOR_EACH (subfacet, list_node, &facet->subfacets) {
906 subfacet->installed = false;
907 subfacet->dp_packet_count = 0;
908 subfacet->dp_byte_count = 0;
912 dpif_flow_flush(ofproto->dpif);
916 get_features(struct ofproto *ofproto_ OVS_UNUSED,
917 bool *arp_match_ip, uint32_t *actions)
919 *arp_match_ip = true;
920 *actions = ((1u << OFPAT_OUTPUT) |
921 (1u << OFPAT_SET_VLAN_VID) |
922 (1u << OFPAT_SET_VLAN_PCP) |
923 (1u << OFPAT_STRIP_VLAN) |
924 (1u << OFPAT_SET_DL_SRC) |
925 (1u << OFPAT_SET_DL_DST) |
926 (1u << OFPAT_SET_NW_SRC) |
927 (1u << OFPAT_SET_NW_DST) |
928 (1u << OFPAT_SET_NW_TOS) |
929 (1u << OFPAT_SET_TP_SRC) |
930 (1u << OFPAT_SET_TP_DST) |
931 (1u << OFPAT_ENQUEUE));
935 get_tables(struct ofproto *ofproto_, struct ofp_table_stats *ots)
937 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
938 struct dpif_dp_stats s;
940 strcpy(ots->name, "classifier");
942 dpif_get_dp_stats(ofproto->dpif, &s);
943 put_32aligned_be64(&ots->lookup_count, htonll(s.n_hit + s.n_missed));
944 put_32aligned_be64(&ots->matched_count,
945 htonll(s.n_hit + ofproto->n_matches));
948 static struct ofport *
951 struct ofport_dpif *port = xmalloc(sizeof *port);
956 port_dealloc(struct ofport *port_)
958 struct ofport_dpif *port = ofport_dpif_cast(port_);
963 port_construct(struct ofport *port_)
965 struct ofport_dpif *port = ofport_dpif_cast(port_);
966 struct ofproto_dpif *ofproto = ofproto_dpif_cast(port->up.ofproto);
968 ofproto->need_revalidate = true;
969 port->odp_port = ofp_port_to_odp_port(port->up.ofp_port);
972 port->tag = tag_create_random();
973 port->may_enable = true;
974 port->stp_port = NULL;
975 port->stp_state = STP_DISABLED;
976 hmap_init(&port->priorities);
977 port->realdev_ofp_port = 0;
978 port->vlandev_vid = 0;
980 if (ofproto->sflow) {
981 dpif_sflow_add_port(ofproto->sflow, port_);
988 port_destruct(struct ofport *port_)
990 struct ofport_dpif *port = ofport_dpif_cast(port_);
991 struct ofproto_dpif *ofproto = ofproto_dpif_cast(port->up.ofproto);
993 ofproto->need_revalidate = true;
994 bundle_remove(port_);
995 set_cfm(port_, NULL);
996 if (ofproto->sflow) {
997 dpif_sflow_del_port(ofproto->sflow, port->odp_port);
1000 ofport_clear_priorities(port);
1001 hmap_destroy(&port->priorities);
1005 port_modified(struct ofport *port_)
1007 struct ofport_dpif *port = ofport_dpif_cast(port_);
1009 if (port->bundle && port->bundle->bond) {
1010 bond_slave_set_netdev(port->bundle->bond, port, port->up.netdev);
1015 port_reconfigured(struct ofport *port_, ovs_be32 old_config)
1017 struct ofport_dpif *port = ofport_dpif_cast(port_);
1018 struct ofproto_dpif *ofproto = ofproto_dpif_cast(port->up.ofproto);
1019 ovs_be32 changed = old_config ^ port->up.opp.config;
1021 if (changed & htonl(OFPPC_NO_RECV | OFPPC_NO_RECV_STP |
1022 OFPPC_NO_FWD | OFPPC_NO_FLOOD)) {
1023 ofproto->need_revalidate = true;
1025 if (changed & htonl(OFPPC_NO_FLOOD) && port->bundle) {
1026 bundle_update(port->bundle);
1032 set_sflow(struct ofproto *ofproto_,
1033 const struct ofproto_sflow_options *sflow_options)
1035 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1036 struct dpif_sflow *ds = ofproto->sflow;
1038 if (sflow_options) {
1040 struct ofport_dpif *ofport;
1042 ds = ofproto->sflow = dpif_sflow_create(ofproto->dpif);
1043 HMAP_FOR_EACH (ofport, up.hmap_node, &ofproto->up.ports) {
1044 dpif_sflow_add_port(ds, &ofport->up);
1046 ofproto->need_revalidate = true;
1048 dpif_sflow_set_options(ds, sflow_options);
1051 dpif_sflow_destroy(ds);
1052 ofproto->need_revalidate = true;
1053 ofproto->sflow = NULL;
1060 set_cfm(struct ofport *ofport_, const struct cfm_settings *s)
1062 struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
1069 struct ofproto_dpif *ofproto;
1071 ofproto = ofproto_dpif_cast(ofport->up.ofproto);
1072 ofproto->need_revalidate = true;
1073 ofport->cfm = cfm_create(netdev_get_name(ofport->up.netdev));
1076 if (cfm_configure(ofport->cfm, s)) {
1082 cfm_destroy(ofport->cfm);
1088 get_cfm_fault(const struct ofport *ofport_)
1090 struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
1092 return ofport->cfm ? cfm_get_fault(ofport->cfm) : -1;
1096 get_cfm_remote_mpids(const struct ofport *ofport_, const uint64_t **rmps,
1099 struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
1102 cfm_get_remote_mpids(ofport->cfm, rmps, n_rmps);
1109 /* Spanning Tree. */
1112 send_bpdu_cb(struct ofpbuf *pkt, int port_num, void *ofproto_)
1114 struct ofproto_dpif *ofproto = ofproto_;
1115 struct stp_port *sp = stp_get_port(ofproto->stp, port_num);
1116 struct ofport_dpif *ofport;
1118 ofport = stp_port_get_aux(sp);
1120 VLOG_WARN_RL(&rl, "%s: cannot send BPDU on unknown port %d",
1121 ofproto->up.name, port_num);
1123 struct eth_header *eth = pkt->l2;
1125 netdev_get_etheraddr(ofport->up.netdev, eth->eth_src);
1126 if (eth_addr_is_zero(eth->eth_src)) {
1127 VLOG_WARN_RL(&rl, "%s: cannot send BPDU on port %d "
1128 "with unknown MAC", ofproto->up.name, port_num);
1130 send_packet(ofport, pkt);
1136 /* Configures STP on 'ofproto_' using the settings defined in 's'. */
1138 set_stp(struct ofproto *ofproto_, const struct ofproto_stp_settings *s)
1140 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1142 /* Only revalidate flows if the configuration changed. */
1143 if (!s != !ofproto->stp) {
1144 ofproto->need_revalidate = true;
1148 if (!ofproto->stp) {
1149 ofproto->stp = stp_create(ofproto_->name, s->system_id,
1150 send_bpdu_cb, ofproto);
1151 ofproto->stp_last_tick = time_msec();
1154 stp_set_bridge_id(ofproto->stp, s->system_id);
1155 stp_set_bridge_priority(ofproto->stp, s->priority);
1156 stp_set_hello_time(ofproto->stp, s->hello_time);
1157 stp_set_max_age(ofproto->stp, s->max_age);
1158 stp_set_forward_delay(ofproto->stp, s->fwd_delay);
1160 struct ofport *ofport;
1162 HMAP_FOR_EACH (ofport, hmap_node, &ofproto->up.ports) {
1163 set_stp_port(ofport, NULL);
1166 stp_destroy(ofproto->stp);
1167 ofproto->stp = NULL;
1174 get_stp_status(struct ofproto *ofproto_, struct ofproto_stp_status *s)
1176 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1180 s->bridge_id = stp_get_bridge_id(ofproto->stp);
1181 s->designated_root = stp_get_designated_root(ofproto->stp);
1182 s->root_path_cost = stp_get_root_path_cost(ofproto->stp);
1191 update_stp_port_state(struct ofport_dpif *ofport)
1193 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport->up.ofproto);
1194 enum stp_state state;
1196 /* Figure out new state. */
1197 state = ofport->stp_port ? stp_port_get_state(ofport->stp_port)
1201 if (ofport->stp_state != state) {
1205 VLOG_DBG_RL(&rl, "port %s: STP state changed from %s to %s",
1206 netdev_get_name(ofport->up.netdev),
1207 stp_state_name(ofport->stp_state),
1208 stp_state_name(state));
1209 if (stp_learn_in_state(ofport->stp_state)
1210 != stp_learn_in_state(state)) {
1211 /* xxx Learning action flows should also be flushed. */
1212 mac_learning_flush(ofproto->ml, &ofproto->revalidate_set);
1214 fwd_change = stp_forward_in_state(ofport->stp_state)
1215 != stp_forward_in_state(state);
1217 ofproto->need_revalidate = true;
1218 ofport->stp_state = state;
1219 ofport->stp_state_entered = time_msec();
1221 if (fwd_change && ofport->bundle) {
1222 bundle_update(ofport->bundle);
1225 /* Update the STP state bits in the OpenFlow port description. */
1226 of_state = (ofport->up.opp.state & htonl(~OFPPS_STP_MASK))
1227 | htonl(state == STP_LISTENING ? OFPPS_STP_LISTEN
1228 : state == STP_LEARNING ? OFPPS_STP_LEARN
1229 : state == STP_FORWARDING ? OFPPS_STP_FORWARD
1230 : state == STP_BLOCKING ? OFPPS_STP_BLOCK
1232 ofproto_port_set_state(&ofport->up, of_state);
1236 /* Configures STP on 'ofport_' using the settings defined in 's'. The
1237 * caller is responsible for assigning STP port numbers and ensuring
1238 * there are no duplicates. */
1240 set_stp_port(struct ofport *ofport_,
1241 const struct ofproto_port_stp_settings *s)
1243 struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
1244 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport->up.ofproto);
1245 struct stp_port *sp = ofport->stp_port;
1247 if (!s || !s->enable) {
1249 ofport->stp_port = NULL;
1250 stp_port_disable(sp);
1251 update_stp_port_state(ofport);
1254 } else if (sp && stp_port_no(sp) != s->port_num
1255 && ofport == stp_port_get_aux(sp)) {
1256 /* The port-id changed, so disable the old one if it's not
1257 * already in use by another port. */
1258 stp_port_disable(sp);
1261 sp = ofport->stp_port = stp_get_port(ofproto->stp, s->port_num);
1262 stp_port_enable(sp);
1264 stp_port_set_aux(sp, ofport);
1265 stp_port_set_priority(sp, s->priority);
1266 stp_port_set_path_cost(sp, s->path_cost);
1268 update_stp_port_state(ofport);
1274 get_stp_port_status(struct ofport *ofport_,
1275 struct ofproto_port_stp_status *s)
1277 struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
1278 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport->up.ofproto);
1279 struct stp_port *sp = ofport->stp_port;
1281 if (!ofproto->stp || !sp) {
1287 s->port_id = stp_port_get_id(sp);
1288 s->state = stp_port_get_state(sp);
1289 s->sec_in_state = (time_msec() - ofport->stp_state_entered) / 1000;
1290 s->role = stp_port_get_role(sp);
1291 stp_port_get_counts(sp, &s->tx_count, &s->rx_count, &s->error_count);
1297 stp_run(struct ofproto_dpif *ofproto)
1300 long long int now = time_msec();
1301 long long int elapsed = now - ofproto->stp_last_tick;
1302 struct stp_port *sp;
1305 stp_tick(ofproto->stp, MIN(INT_MAX, elapsed));
1306 ofproto->stp_last_tick = now;
1308 while (stp_get_changed_port(ofproto->stp, &sp)) {
1309 struct ofport_dpif *ofport = stp_port_get_aux(sp);
1312 update_stp_port_state(ofport);
1316 if (stp_check_and_reset_fdb_flush(ofproto->stp)) {
1317 mac_learning_flush(ofproto->ml, &ofproto->revalidate_set);
1323 stp_wait(struct ofproto_dpif *ofproto)
1326 poll_timer_wait(1000);
1330 /* Returns true if STP should process 'flow'. */
1332 stp_should_process_flow(const struct flow *flow)
1334 return eth_addr_equals(flow->dl_dst, eth_addr_stp);
1338 stp_process_packet(const struct ofport_dpif *ofport,
1339 const struct ofpbuf *packet)
1341 struct ofpbuf payload = *packet;
1342 struct eth_header *eth = payload.data;
1343 struct stp_port *sp = ofport->stp_port;
1345 /* Sink packets on ports that have STP disabled when the bridge has
1347 if (!sp || stp_port_get_state(sp) == STP_DISABLED) {
1351 /* Trim off padding on payload. */
1352 if (payload.size > ntohs(eth->eth_type) + ETH_HEADER_LEN) {
1353 payload.size = ntohs(eth->eth_type) + ETH_HEADER_LEN;
1356 if (ofpbuf_try_pull(&payload, ETH_HEADER_LEN + LLC_HEADER_LEN)) {
1357 stp_received_bpdu(sp, payload.data, payload.size);
1361 static struct priority_to_dscp *
1362 get_priority(const struct ofport_dpif *ofport, uint32_t priority)
1364 struct priority_to_dscp *pdscp;
1367 hash = hash_int(priority, 0);
1368 HMAP_FOR_EACH_IN_BUCKET (pdscp, hmap_node, hash, &ofport->priorities) {
1369 if (pdscp->priority == priority) {
1377 ofport_clear_priorities(struct ofport_dpif *ofport)
1379 struct priority_to_dscp *pdscp, *next;
1381 HMAP_FOR_EACH_SAFE (pdscp, next, hmap_node, &ofport->priorities) {
1382 hmap_remove(&ofport->priorities, &pdscp->hmap_node);
1388 set_queues(struct ofport *ofport_,
1389 const struct ofproto_port_queue *qdscp_list,
1392 struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
1393 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport->up.ofproto);
1394 struct hmap new = HMAP_INITIALIZER(&new);
1397 for (i = 0; i < n_qdscp; i++) {
1398 struct priority_to_dscp *pdscp;
1402 dscp = (qdscp_list[i].dscp << 2) & IP_DSCP_MASK;
1403 if (dpif_queue_to_priority(ofproto->dpif, qdscp_list[i].queue,
1408 pdscp = get_priority(ofport, priority);
1410 hmap_remove(&ofport->priorities, &pdscp->hmap_node);
1412 pdscp = xmalloc(sizeof *pdscp);
1413 pdscp->priority = priority;
1415 ofproto->need_revalidate = true;
1418 if (pdscp->dscp != dscp) {
1420 ofproto->need_revalidate = true;
1423 hmap_insert(&new, &pdscp->hmap_node, hash_int(pdscp->priority, 0));
1426 if (!hmap_is_empty(&ofport->priorities)) {
1427 ofport_clear_priorities(ofport);
1428 ofproto->need_revalidate = true;
1431 hmap_swap(&new, &ofport->priorities);
1439 /* Expires all MAC learning entries associated with 'bundle' and forces its
1440 * ofproto to revalidate every flow.
1442 * Normally MAC learning entries are removed only from the ofproto associated
1443 * with 'bundle', but if 'all_ofprotos' is true, then the MAC learning entries
1444 * are removed from every ofproto. When patch ports and SLB bonds are in use
1445 * and a VM migration happens and the gratuitous ARPs are somehow lost, this
1446 * avoids a MAC_ENTRY_IDLE_TIME delay before the migrated VM can communicate
1447 * with the host from which it migrated. */
1449 bundle_flush_macs(struct ofbundle *bundle, bool all_ofprotos)
1451 struct ofproto_dpif *ofproto = bundle->ofproto;
1452 struct mac_learning *ml = ofproto->ml;
1453 struct mac_entry *mac, *next_mac;
1455 ofproto->need_revalidate = true;
1456 LIST_FOR_EACH_SAFE (mac, next_mac, lru_node, &ml->lrus) {
1457 if (mac->port.p == bundle) {
1459 struct ofproto_dpif *o;
1461 HMAP_FOR_EACH (o, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
1463 struct mac_entry *e;
1465 e = mac_learning_lookup(o->ml, mac->mac, mac->vlan,
1468 tag_set_add(&o->revalidate_set, e->tag);
1469 mac_learning_expire(o->ml, e);
1475 mac_learning_expire(ml, mac);
1480 static struct ofbundle *
1481 bundle_lookup(const struct ofproto_dpif *ofproto, void *aux)
1483 struct ofbundle *bundle;
1485 HMAP_FOR_EACH_IN_BUCKET (bundle, hmap_node, hash_pointer(aux, 0),
1486 &ofproto->bundles) {
1487 if (bundle->aux == aux) {
1494 /* Looks up each of the 'n_auxes' pointers in 'auxes' as bundles and adds the
1495 * ones that are found to 'bundles'. */
1497 bundle_lookup_multiple(struct ofproto_dpif *ofproto,
1498 void **auxes, size_t n_auxes,
1499 struct hmapx *bundles)
1503 hmapx_init(bundles);
1504 for (i = 0; i < n_auxes; i++) {
1505 struct ofbundle *bundle = bundle_lookup(ofproto, auxes[i]);
1507 hmapx_add(bundles, bundle);
1513 bundle_update(struct ofbundle *bundle)
1515 struct ofport_dpif *port;
1517 bundle->floodable = true;
1518 LIST_FOR_EACH (port, bundle_node, &bundle->ports) {
1519 if (port->up.opp.config & htonl(OFPPC_NO_FLOOD)) {
1520 bundle->floodable = false;
1527 bundle_del_port(struct ofport_dpif *port)
1529 struct ofbundle *bundle = port->bundle;
1531 bundle->ofproto->need_revalidate = true;
1533 list_remove(&port->bundle_node);
1534 port->bundle = NULL;
1537 lacp_slave_unregister(bundle->lacp, port);
1540 bond_slave_unregister(bundle->bond, port);
1543 bundle_update(bundle);
1547 bundle_add_port(struct ofbundle *bundle, uint32_t ofp_port,
1548 struct lacp_slave_settings *lacp,
1549 uint32_t bond_stable_id)
1551 struct ofport_dpif *port;
1553 port = get_ofp_port(bundle->ofproto, ofp_port);
1558 if (port->bundle != bundle) {
1559 bundle->ofproto->need_revalidate = true;
1561 bundle_del_port(port);
1564 port->bundle = bundle;
1565 list_push_back(&bundle->ports, &port->bundle_node);
1566 if (port->up.opp.config & htonl(OFPPC_NO_FLOOD)) {
1567 bundle->floodable = false;
1571 port->bundle->ofproto->need_revalidate = true;
1572 lacp_slave_register(bundle->lacp, port, lacp);
1575 port->bond_stable_id = bond_stable_id;
1581 bundle_destroy(struct ofbundle *bundle)
1583 struct ofproto_dpif *ofproto;
1584 struct ofport_dpif *port, *next_port;
1591 ofproto = bundle->ofproto;
1592 for (i = 0; i < MAX_MIRRORS; i++) {
1593 struct ofmirror *m = ofproto->mirrors[i];
1595 if (m->out == bundle) {
1597 } else if (hmapx_find_and_delete(&m->srcs, bundle)
1598 || hmapx_find_and_delete(&m->dsts, bundle)) {
1599 ofproto->need_revalidate = true;
1604 LIST_FOR_EACH_SAFE (port, next_port, bundle_node, &bundle->ports) {
1605 bundle_del_port(port);
1608 bundle_flush_macs(bundle, true);
1609 hmap_remove(&ofproto->bundles, &bundle->hmap_node);
1611 free(bundle->trunks);
1612 lacp_destroy(bundle->lacp);
1613 bond_destroy(bundle->bond);
1618 bundle_set(struct ofproto *ofproto_, void *aux,
1619 const struct ofproto_bundle_settings *s)
1621 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1622 bool need_flush = false;
1623 struct ofport_dpif *port;
1624 struct ofbundle *bundle;
1625 unsigned long *trunks;
1631 bundle_destroy(bundle_lookup(ofproto, aux));
1635 assert(s->n_slaves == 1 || s->bond != NULL);
1636 assert((s->lacp != NULL) == (s->lacp_slaves != NULL));
1638 bundle = bundle_lookup(ofproto, aux);
1640 bundle = xmalloc(sizeof *bundle);
1642 bundle->ofproto = ofproto;
1643 hmap_insert(&ofproto->bundles, &bundle->hmap_node,
1644 hash_pointer(aux, 0));
1646 bundle->name = NULL;
1648 list_init(&bundle->ports);
1649 bundle->vlan_mode = PORT_VLAN_TRUNK;
1651 bundle->trunks = NULL;
1652 bundle->use_priority_tags = s->use_priority_tags;
1653 bundle->lacp = NULL;
1654 bundle->bond = NULL;
1656 bundle->floodable = true;
1658 bundle->src_mirrors = 0;
1659 bundle->dst_mirrors = 0;
1660 bundle->mirror_out = 0;
1663 if (!bundle->name || strcmp(s->name, bundle->name)) {
1665 bundle->name = xstrdup(s->name);
1670 if (!bundle->lacp) {
1671 ofproto->need_revalidate = true;
1672 bundle->lacp = lacp_create();
1674 lacp_configure(bundle->lacp, s->lacp);
1676 lacp_destroy(bundle->lacp);
1677 bundle->lacp = NULL;
1680 /* Update set of ports. */
1682 for (i = 0; i < s->n_slaves; i++) {
1683 if (!bundle_add_port(bundle, s->slaves[i],
1684 s->lacp ? &s->lacp_slaves[i] : NULL,
1685 s->bond_stable_ids ? s->bond_stable_ids[i] : 0)) {
1689 if (!ok || list_size(&bundle->ports) != s->n_slaves) {
1690 struct ofport_dpif *next_port;
1692 LIST_FOR_EACH_SAFE (port, next_port, bundle_node, &bundle->ports) {
1693 for (i = 0; i < s->n_slaves; i++) {
1694 if (s->slaves[i] == port->up.ofp_port) {
1699 bundle_del_port(port);
1703 assert(list_size(&bundle->ports) <= s->n_slaves);
1705 if (list_is_empty(&bundle->ports)) {
1706 bundle_destroy(bundle);
1710 /* Set VLAN tagging mode */
1711 if (s->vlan_mode != bundle->vlan_mode
1712 || s->use_priority_tags != bundle->use_priority_tags) {
1713 bundle->vlan_mode = s->vlan_mode;
1714 bundle->use_priority_tags = s->use_priority_tags;
1719 vlan = (s->vlan_mode == PORT_VLAN_TRUNK ? -1
1720 : s->vlan >= 0 && s->vlan <= 4095 ? s->vlan
1722 if (vlan != bundle->vlan) {
1723 bundle->vlan = vlan;
1727 /* Get trunked VLANs. */
1728 switch (s->vlan_mode) {
1729 case PORT_VLAN_ACCESS:
1733 case PORT_VLAN_TRUNK:
1734 trunks = (unsigned long *) s->trunks;
1737 case PORT_VLAN_NATIVE_UNTAGGED:
1738 case PORT_VLAN_NATIVE_TAGGED:
1739 if (vlan != 0 && (!s->trunks
1740 || !bitmap_is_set(s->trunks, vlan)
1741 || bitmap_is_set(s->trunks, 0))) {
1742 /* Force trunking the native VLAN and prohibit trunking VLAN 0. */
1744 trunks = bitmap_clone(s->trunks, 4096);
1746 trunks = bitmap_allocate1(4096);
1748 bitmap_set1(trunks, vlan);
1749 bitmap_set0(trunks, 0);
1751 trunks = (unsigned long *) s->trunks;
1758 if (!vlan_bitmap_equal(trunks, bundle->trunks)) {
1759 free(bundle->trunks);
1760 if (trunks == s->trunks) {
1761 bundle->trunks = vlan_bitmap_clone(trunks);
1763 bundle->trunks = trunks;
1768 if (trunks != s->trunks) {
1773 if (!list_is_short(&bundle->ports)) {
1774 bundle->ofproto->has_bonded_bundles = true;
1776 if (bond_reconfigure(bundle->bond, s->bond)) {
1777 ofproto->need_revalidate = true;
1780 bundle->bond = bond_create(s->bond);
1781 ofproto->need_revalidate = true;
1784 LIST_FOR_EACH (port, bundle_node, &bundle->ports) {
1785 bond_slave_register(bundle->bond, port, port->bond_stable_id,
1789 bond_destroy(bundle->bond);
1790 bundle->bond = NULL;
1793 /* If we changed something that would affect MAC learning, un-learn
1794 * everything on this port and force flow revalidation. */
1796 bundle_flush_macs(bundle, false);
1803 bundle_remove(struct ofport *port_)
1805 struct ofport_dpif *port = ofport_dpif_cast(port_);
1806 struct ofbundle *bundle = port->bundle;
1809 bundle_del_port(port);
1810 if (list_is_empty(&bundle->ports)) {
1811 bundle_destroy(bundle);
1812 } else if (list_is_short(&bundle->ports)) {
1813 bond_destroy(bundle->bond);
1814 bundle->bond = NULL;
1820 send_pdu_cb(void *port_, const void *pdu, size_t pdu_size)
1822 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 10);
1823 struct ofport_dpif *port = port_;
1824 uint8_t ea[ETH_ADDR_LEN];
1827 error = netdev_get_etheraddr(port->up.netdev, ea);
1829 struct ofpbuf packet;
1832 ofpbuf_init(&packet, 0);
1833 packet_pdu = eth_compose(&packet, eth_addr_lacp, ea, ETH_TYPE_LACP,
1835 memcpy(packet_pdu, pdu, pdu_size);
1837 send_packet(port, &packet);
1838 ofpbuf_uninit(&packet);
1840 VLOG_ERR_RL(&rl, "port %s: cannot obtain Ethernet address of iface "
1841 "%s (%s)", port->bundle->name,
1842 netdev_get_name(port->up.netdev), strerror(error));
1847 bundle_send_learning_packets(struct ofbundle *bundle)
1849 struct ofproto_dpif *ofproto = bundle->ofproto;
1850 int error, n_packets, n_errors;
1851 struct mac_entry *e;
1853 error = n_packets = n_errors = 0;
1854 LIST_FOR_EACH (e, lru_node, &ofproto->ml->lrus) {
1855 if (e->port.p != bundle) {
1856 struct ofpbuf *learning_packet;
1857 struct ofport_dpif *port;
1861 /* The assignment to "port" is unnecessary but makes "grep"ing for
1862 * struct ofport_dpif more effective. */
1863 learning_packet = bond_compose_learning_packet(bundle->bond,
1867 ret = send_packet(port, learning_packet);
1868 ofpbuf_delete(learning_packet);
1878 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
1879 VLOG_WARN_RL(&rl, "bond %s: %d errors sending %d gratuitous learning "
1880 "packets, last error was: %s",
1881 bundle->name, n_errors, n_packets, strerror(error));
1883 VLOG_DBG("bond %s: sent %d gratuitous learning packets",
1884 bundle->name, n_packets);
1889 bundle_run(struct ofbundle *bundle)
1892 lacp_run(bundle->lacp, send_pdu_cb);
1895 struct ofport_dpif *port;
1897 LIST_FOR_EACH (port, bundle_node, &bundle->ports) {
1898 bond_slave_set_may_enable(bundle->bond, port, port->may_enable);
1901 bond_run(bundle->bond, &bundle->ofproto->revalidate_set,
1902 lacp_status(bundle->lacp));
1903 if (bond_should_send_learning_packets(bundle->bond)) {
1904 bundle_send_learning_packets(bundle);
1910 bundle_wait(struct ofbundle *bundle)
1913 lacp_wait(bundle->lacp);
1916 bond_wait(bundle->bond);
1923 mirror_scan(struct ofproto_dpif *ofproto)
1927 for (idx = 0; idx < MAX_MIRRORS; idx++) {
1928 if (!ofproto->mirrors[idx]) {
1935 static struct ofmirror *
1936 mirror_lookup(struct ofproto_dpif *ofproto, void *aux)
1940 for (i = 0; i < MAX_MIRRORS; i++) {
1941 struct ofmirror *mirror = ofproto->mirrors[i];
1942 if (mirror && mirror->aux == aux) {
1950 /* Update the 'dup_mirrors' member of each of the ofmirrors in 'ofproto'. */
1952 mirror_update_dups(struct ofproto_dpif *ofproto)
1956 for (i = 0; i < MAX_MIRRORS; i++) {
1957 struct ofmirror *m = ofproto->mirrors[i];
1960 m->dup_mirrors = MIRROR_MASK_C(1) << i;
1964 for (i = 0; i < MAX_MIRRORS; i++) {
1965 struct ofmirror *m1 = ofproto->mirrors[i];
1972 for (j = i + 1; j < MAX_MIRRORS; j++) {
1973 struct ofmirror *m2 = ofproto->mirrors[j];
1975 if (m2 && m1->out == m2->out && m1->out_vlan == m2->out_vlan) {
1976 m1->dup_mirrors |= MIRROR_MASK_C(1) << j;
1977 m2->dup_mirrors |= m1->dup_mirrors;
1984 mirror_set(struct ofproto *ofproto_, void *aux,
1985 const struct ofproto_mirror_settings *s)
1987 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1988 mirror_mask_t mirror_bit;
1989 struct ofbundle *bundle;
1990 struct ofmirror *mirror;
1991 struct ofbundle *out;
1992 struct hmapx srcs; /* Contains "struct ofbundle *"s. */
1993 struct hmapx dsts; /* Contains "struct ofbundle *"s. */
1996 mirror = mirror_lookup(ofproto, aux);
1998 mirror_destroy(mirror);
2004 idx = mirror_scan(ofproto);
2006 VLOG_WARN("bridge %s: maximum of %d port mirrors reached, "
2008 ofproto->up.name, MAX_MIRRORS, s->name);
2012 mirror = ofproto->mirrors[idx] = xzalloc(sizeof *mirror);
2013 mirror->ofproto = ofproto;
2016 mirror->out_vlan = -1;
2017 mirror->name = NULL;
2020 if (!mirror->name || strcmp(s->name, mirror->name)) {
2022 mirror->name = xstrdup(s->name);
2025 /* Get the new configuration. */
2026 if (s->out_bundle) {
2027 out = bundle_lookup(ofproto, s->out_bundle);
2029 mirror_destroy(mirror);
2035 out_vlan = s->out_vlan;
2037 bundle_lookup_multiple(ofproto, s->srcs, s->n_srcs, &srcs);
2038 bundle_lookup_multiple(ofproto, s->dsts, s->n_dsts, &dsts);
2040 /* If the configuration has not changed, do nothing. */
2041 if (hmapx_equals(&srcs, &mirror->srcs)
2042 && hmapx_equals(&dsts, &mirror->dsts)
2043 && vlan_bitmap_equal(mirror->vlans, s->src_vlans)
2044 && mirror->out == out
2045 && mirror->out_vlan == out_vlan)
2047 hmapx_destroy(&srcs);
2048 hmapx_destroy(&dsts);
2052 hmapx_swap(&srcs, &mirror->srcs);
2053 hmapx_destroy(&srcs);
2055 hmapx_swap(&dsts, &mirror->dsts);
2056 hmapx_destroy(&dsts);
2058 free(mirror->vlans);
2059 mirror->vlans = vlan_bitmap_clone(s->src_vlans);
2062 mirror->out_vlan = out_vlan;
2064 /* Update bundles. */
2065 mirror_bit = MIRROR_MASK_C(1) << mirror->idx;
2066 HMAP_FOR_EACH (bundle, hmap_node, &mirror->ofproto->bundles) {
2067 if (hmapx_contains(&mirror->srcs, bundle)) {
2068 bundle->src_mirrors |= mirror_bit;
2070 bundle->src_mirrors &= ~mirror_bit;
2073 if (hmapx_contains(&mirror->dsts, bundle)) {
2074 bundle->dst_mirrors |= mirror_bit;
2076 bundle->dst_mirrors &= ~mirror_bit;
2079 if (mirror->out == bundle) {
2080 bundle->mirror_out |= mirror_bit;
2082 bundle->mirror_out &= ~mirror_bit;
2086 ofproto->need_revalidate = true;
2087 mac_learning_flush(ofproto->ml, &ofproto->revalidate_set);
2088 mirror_update_dups(ofproto);
2094 mirror_destroy(struct ofmirror *mirror)
2096 struct ofproto_dpif *ofproto;
2097 mirror_mask_t mirror_bit;
2098 struct ofbundle *bundle;
2104 ofproto = mirror->ofproto;
2105 ofproto->need_revalidate = true;
2106 mac_learning_flush(ofproto->ml, &ofproto->revalidate_set);
2108 mirror_bit = MIRROR_MASK_C(1) << mirror->idx;
2109 HMAP_FOR_EACH (bundle, hmap_node, &ofproto->bundles) {
2110 bundle->src_mirrors &= ~mirror_bit;
2111 bundle->dst_mirrors &= ~mirror_bit;
2112 bundle->mirror_out &= ~mirror_bit;
2115 hmapx_destroy(&mirror->srcs);
2116 hmapx_destroy(&mirror->dsts);
2117 free(mirror->vlans);
2119 ofproto->mirrors[mirror->idx] = NULL;
2123 mirror_update_dups(ofproto);
2127 mirror_get_stats(struct ofproto *ofproto_, void *aux,
2128 uint64_t *packets, uint64_t *bytes)
2130 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
2131 struct ofmirror *mirror = mirror_lookup(ofproto, aux);
2134 *packets = *bytes = UINT64_MAX;
2138 *packets = mirror->packet_count;
2139 *bytes = mirror->byte_count;
2145 set_flood_vlans(struct ofproto *ofproto_, unsigned long *flood_vlans)
2147 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
2148 if (mac_learning_set_flood_vlans(ofproto->ml, flood_vlans)) {
2149 mac_learning_flush(ofproto->ml, &ofproto->revalidate_set);
2155 is_mirror_output_bundle(const struct ofproto *ofproto_, void *aux)
2157 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
2158 struct ofbundle *bundle = bundle_lookup(ofproto, aux);
2159 return bundle && bundle->mirror_out != 0;
2163 forward_bpdu_changed(struct ofproto *ofproto_)
2165 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
2166 /* Revalidate cached flows whenever forward_bpdu option changes. */
2167 ofproto->need_revalidate = true;
2171 set_mac_idle_time(struct ofproto *ofproto_, unsigned int idle_time)
2173 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
2174 mac_learning_set_idle_time(ofproto->ml, idle_time);
2179 static struct ofport_dpif *
2180 get_ofp_port(struct ofproto_dpif *ofproto, uint16_t ofp_port)
2182 struct ofport *ofport = ofproto_get_port(&ofproto->up, ofp_port);
2183 return ofport ? ofport_dpif_cast(ofport) : NULL;
2186 static struct ofport_dpif *
2187 get_odp_port(struct ofproto_dpif *ofproto, uint32_t odp_port)
2189 return get_ofp_port(ofproto, odp_port_to_ofp_port(odp_port));
2193 ofproto_port_from_dpif_port(struct ofproto_port *ofproto_port,
2194 struct dpif_port *dpif_port)
2196 ofproto_port->name = dpif_port->name;
2197 ofproto_port->type = dpif_port->type;
2198 ofproto_port->ofp_port = odp_port_to_ofp_port(dpif_port->port_no);
2202 port_run(struct ofport_dpif *ofport)
2204 bool enable = netdev_get_carrier(ofport->up.netdev);
2207 cfm_run(ofport->cfm);
2209 if (cfm_should_send_ccm(ofport->cfm)) {
2210 struct ofpbuf packet;
2212 ofpbuf_init(&packet, 0);
2213 cfm_compose_ccm(ofport->cfm, &packet, ofport->up.opp.hw_addr);
2214 send_packet(ofport, &packet);
2215 ofpbuf_uninit(&packet);
2218 enable = enable && !cfm_get_fault(ofport->cfm)
2219 && cfm_get_opup(ofport->cfm);
2222 if (ofport->bundle) {
2223 enable = enable && lacp_slave_may_enable(ofport->bundle->lacp, ofport);
2226 if (ofport->may_enable != enable) {
2227 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport->up.ofproto);
2229 if (ofproto->has_bundle_action) {
2230 ofproto->need_revalidate = true;
2234 ofport->may_enable = enable;
2238 port_wait(struct ofport_dpif *ofport)
2241 cfm_wait(ofport->cfm);
2246 port_query_by_name(const struct ofproto *ofproto_, const char *devname,
2247 struct ofproto_port *ofproto_port)
2249 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
2250 struct dpif_port dpif_port;
2253 error = dpif_port_query_by_name(ofproto->dpif, devname, &dpif_port);
2255 ofproto_port_from_dpif_port(ofproto_port, &dpif_port);
2261 port_add(struct ofproto *ofproto_, struct netdev *netdev, uint16_t *ofp_portp)
2263 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
2267 error = dpif_port_add(ofproto->dpif, netdev, &odp_port);
2269 *ofp_portp = odp_port_to_ofp_port(odp_port);
2275 port_del(struct ofproto *ofproto_, uint16_t ofp_port)
2277 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
2280 error = dpif_port_del(ofproto->dpif, ofp_port_to_odp_port(ofp_port));
2282 struct ofport_dpif *ofport = get_ofp_port(ofproto, ofp_port);
2284 /* The caller is going to close ofport->up.netdev. If this is a
2285 * bonded port, then the bond is using that netdev, so remove it
2286 * from the bond. The client will need to reconfigure everything
2287 * after deleting ports, so then the slave will get re-added. */
2288 bundle_remove(&ofport->up);
2295 port_get_stats(const struct ofport *ofport_, struct netdev_stats *stats)
2297 struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
2300 error = netdev_get_stats(ofport->up.netdev, stats);
2302 if (!error && ofport->odp_port == OVSP_LOCAL) {
2303 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport->up.ofproto);
2305 /* ofproto->stats.tx_packets represents packets that we created
2306 * internally and sent to some port (e.g. packets sent with
2307 * send_packet()). Account for them as if they had come from
2308 * OFPP_LOCAL and got forwarded. */
2310 if (stats->rx_packets != UINT64_MAX) {
2311 stats->rx_packets += ofproto->stats.tx_packets;
2314 if (stats->rx_bytes != UINT64_MAX) {
2315 stats->rx_bytes += ofproto->stats.tx_bytes;
2318 /* ofproto->stats.rx_packets represents packets that were received on
2319 * some port and we processed internally and dropped (e.g. STP).
2320 * Account fro them as if they had been forwarded to OFPP_LOCAL. */
2322 if (stats->tx_packets != UINT64_MAX) {
2323 stats->tx_packets += ofproto->stats.rx_packets;
2326 if (stats->tx_bytes != UINT64_MAX) {
2327 stats->tx_bytes += ofproto->stats.rx_bytes;
2334 /* Account packets for LOCAL port. */
2336 ofproto_update_local_port_stats(const struct ofproto *ofproto_,
2337 size_t tx_size, size_t rx_size)
2339 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
2342 ofproto->stats.rx_packets++;
2343 ofproto->stats.rx_bytes += rx_size;
2346 ofproto->stats.tx_packets++;
2347 ofproto->stats.tx_bytes += tx_size;
2351 struct port_dump_state {
2352 struct dpif_port_dump dump;
2357 port_dump_start(const struct ofproto *ofproto_, void **statep)
2359 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
2360 struct port_dump_state *state;
2362 *statep = state = xmalloc(sizeof *state);
2363 dpif_port_dump_start(&state->dump, ofproto->dpif);
2364 state->done = false;
2369 port_dump_next(const struct ofproto *ofproto_ OVS_UNUSED, void *state_,
2370 struct ofproto_port *port)
2372 struct port_dump_state *state = state_;
2373 struct dpif_port dpif_port;
2375 if (dpif_port_dump_next(&state->dump, &dpif_port)) {
2376 ofproto_port_from_dpif_port(port, &dpif_port);
2379 int error = dpif_port_dump_done(&state->dump);
2381 return error ? error : EOF;
2386 port_dump_done(const struct ofproto *ofproto_ OVS_UNUSED, void *state_)
2388 struct port_dump_state *state = state_;
2391 dpif_port_dump_done(&state->dump);
2398 port_poll(const struct ofproto *ofproto_, char **devnamep)
2400 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
2401 return dpif_port_poll(ofproto->dpif, devnamep);
2405 port_poll_wait(const struct ofproto *ofproto_)
2407 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
2408 dpif_port_poll_wait(ofproto->dpif);
2412 port_is_lacp_current(const struct ofport *ofport_)
2414 const struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
2415 return (ofport->bundle && ofport->bundle->lacp
2416 ? lacp_slave_is_current(ofport->bundle->lacp, ofport)
2420 /* Upcall handling. */
2422 /* Flow miss batching.
2424 * Some dpifs implement operations faster when you hand them off in a batch.
2425 * To allow batching, "struct flow_miss" queues the dpif-related work needed
2426 * for a given flow. Each "struct flow_miss" corresponds to sending one or
2427 * more packets, plus possibly installing the flow in the dpif.
2429 * So far we only batch the operations that affect flow setup time the most.
2430 * It's possible to batch more than that, but the benefit might be minimal. */
2432 struct hmap_node hmap_node;
2434 enum odp_key_fitness key_fitness;
2435 const struct nlattr *key;
2437 ovs_be16 initial_tci;
2438 struct list packets;
2441 struct flow_miss_op {
2442 struct dpif_op dpif_op;
2443 struct subfacet *subfacet;
2446 /* Sends an OFPT_PACKET_IN message for 'packet' of type OFPR_NO_MATCH to each
2447 * OpenFlow controller as necessary according to their individual
2448 * configurations. */
2450 send_packet_in_miss(struct ofproto_dpif *ofproto, const struct ofpbuf *packet,
2451 const struct flow *flow)
2453 struct ofputil_packet_in pin;
2455 pin.packet = packet->data;
2456 pin.packet_len = packet->size;
2457 pin.total_len = packet->size;
2458 pin.reason = OFPR_NO_MATCH;
2463 pin.buffer_id = 0; /* not yet known */
2464 pin.send_len = 0; /* not used for flow table misses */
2466 flow_get_metadata(flow, &pin.fmd);
2468 /* Registers aren't meaningful on a miss. */
2469 memset(pin.fmd.reg_masks, 0, sizeof pin.fmd.reg_masks);
2471 connmgr_send_packet_in(ofproto->up.connmgr, &pin, flow);
2475 process_special(struct ofproto_dpif *ofproto, const struct flow *flow,
2476 const struct ofpbuf *packet)
2478 struct ofport_dpif *ofport = get_ofp_port(ofproto, flow->in_port);
2484 if (ofport->cfm && cfm_should_process_flow(ofport->cfm, flow)) {
2486 cfm_process_heartbeat(ofport->cfm, packet);
2489 } else if (ofport->bundle && ofport->bundle->lacp
2490 && flow->dl_type == htons(ETH_TYPE_LACP)) {
2492 lacp_process_packet(ofport->bundle->lacp, ofport, packet);
2495 } else if (ofproto->stp && stp_should_process_flow(flow)) {
2497 stp_process_packet(ofport, packet);
2504 static struct flow_miss *
2505 flow_miss_create(struct hmap *todo, const struct flow *flow,
2506 enum odp_key_fitness key_fitness,
2507 const struct nlattr *key, size_t key_len,
2508 ovs_be16 initial_tci)
2510 uint32_t hash = flow_hash(flow, 0);
2511 struct flow_miss *miss;
2513 HMAP_FOR_EACH_WITH_HASH (miss, hmap_node, hash, todo) {
2514 if (flow_equal(&miss->flow, flow)) {
2519 miss = xmalloc(sizeof *miss);
2520 hmap_insert(todo, &miss->hmap_node, hash);
2522 miss->key_fitness = key_fitness;
2524 miss->key_len = key_len;
2525 miss->initial_tci = initial_tci;
2526 list_init(&miss->packets);
2531 handle_flow_miss(struct ofproto_dpif *ofproto, struct flow_miss *miss,
2532 struct flow_miss_op *ops, size_t *n_ops)
2534 const struct flow *flow = &miss->flow;
2535 struct ofpbuf *packet, *next_packet;
2536 struct subfacet *subfacet;
2537 struct facet *facet;
2539 facet = facet_lookup_valid(ofproto, flow);
2541 struct rule_dpif *rule;
2543 rule = rule_dpif_lookup(ofproto, flow, 0);
2545 /* Don't send a packet-in if OFPPC_NO_PACKET_IN asserted. */
2546 struct ofport_dpif *port = get_ofp_port(ofproto, flow->in_port);
2548 if (port->up.opp.config & htonl(OFPPC_NO_PACKET_IN)) {
2549 COVERAGE_INC(ofproto_dpif_no_packet_in);
2550 /* XXX install 'drop' flow entry */
2554 VLOG_WARN_RL(&rl, "packet-in on unknown port %"PRIu16,
2558 LIST_FOR_EACH (packet, list_node, &miss->packets) {
2559 send_packet_in_miss(ofproto, packet, flow);
2565 facet = facet_create(rule, flow);
2568 subfacet = subfacet_create(facet,
2569 miss->key_fitness, miss->key, miss->key_len,
2572 LIST_FOR_EACH_SAFE (packet, next_packet, list_node, &miss->packets) {
2573 struct dpif_flow_stats stats;
2574 struct flow_miss_op *op;
2575 struct dpif_execute *execute;
2577 ofproto->n_matches++;
2579 if (facet->rule->up.cr.priority == FAIL_OPEN_PRIORITY) {
2581 * Extra-special case for fail-open mode.
2583 * We are in fail-open mode and the packet matched the fail-open
2584 * rule, but we are connected to a controller too. We should send
2585 * the packet up to the controller in the hope that it will try to
2586 * set up a flow and thereby allow us to exit fail-open.
2588 * See the top-level comment in fail-open.c for more information.
2590 send_packet_in_miss(ofproto, packet, flow);
2593 if (!facet->may_install || !subfacet->actions) {
2594 subfacet_make_actions(subfacet, packet);
2597 dpif_flow_stats_extract(&facet->flow, packet, &stats);
2598 subfacet_update_stats(subfacet, &stats);
2600 if (!subfacet->actions_len) {
2601 /* No actions to execute, so skip talking to the dpif. */
2605 if (flow->vlan_tci != subfacet->initial_tci) {
2606 /* This packet was received on a VLAN splinter port. We added
2607 * a VLAN to the packet to make the packet resemble the flow,
2608 * but the actions were composed assuming that the packet
2609 * contained no VLAN. So, we must remove the VLAN header from
2610 * the packet before trying to execute the actions. */
2611 eth_pop_vlan(packet);
2614 op = &ops[(*n_ops)++];
2615 execute = &op->dpif_op.u.execute;
2616 op->subfacet = subfacet;
2617 op->dpif_op.type = DPIF_OP_EXECUTE;
2618 execute->key = miss->key;
2619 execute->key_len = miss->key_len;
2620 execute->actions = (facet->may_install
2622 : xmemdup(subfacet->actions,
2623 subfacet->actions_len));
2624 execute->actions_len = subfacet->actions_len;
2625 execute->packet = packet;
2628 if (facet->may_install && subfacet->key_fitness != ODP_FIT_TOO_LITTLE) {
2629 struct flow_miss_op *op = &ops[(*n_ops)++];
2630 struct dpif_flow_put *put = &op->dpif_op.u.flow_put;
2632 op->subfacet = subfacet;
2633 op->dpif_op.type = DPIF_OP_FLOW_PUT;
2634 put->flags = DPIF_FP_CREATE | DPIF_FP_MODIFY;
2635 put->key = miss->key;
2636 put->key_len = miss->key_len;
2637 put->actions = subfacet->actions;
2638 put->actions_len = subfacet->actions_len;
2643 /* Like odp_flow_key_to_flow(), this function converts the 'key_len' bytes of
2644 * OVS_KEY_ATTR_* attributes in 'key' to a flow structure in 'flow' and returns
2645 * an ODP_FIT_* value that indicates how well 'key' fits our expectations for
2646 * what a flow key should contain.
2648 * This function also includes some logic to help make VLAN splinters
2649 * transparent to the rest of the upcall processing logic. In particular, if
2650 * the extracted in_port is a VLAN splinter port, it replaces flow->in_port by
2651 * the "real" port, sets flow->vlan_tci correctly for the VLAN of the VLAN
2652 * splinter port, and pushes a VLAN header onto 'packet' (if it is nonnull).
2654 * Sets '*initial_tci' to the VLAN TCI with which the packet was really
2655 * received, that is, the actual VLAN TCI extracted by odp_flow_key_to_flow().
2656 * (This differs from the value returned in flow->vlan_tci only for packets
2657 * received on VLAN splinters.)
2659 static enum odp_key_fitness
2660 ofproto_dpif_extract_flow_key(const struct ofproto_dpif *ofproto,
2661 const struct nlattr *key, size_t key_len,
2662 struct flow *flow, ovs_be16 *initial_tci,
2663 struct ofpbuf *packet)
2665 enum odp_key_fitness fitness;
2669 fitness = odp_flow_key_to_flow(key, key_len, flow);
2670 if (fitness == ODP_FIT_ERROR) {
2673 *initial_tci = flow->vlan_tci;
2675 realdev = vsp_vlandev_to_realdev(ofproto, flow->in_port, &vid);
2677 /* Cause the flow to be processed as if it came in on the real device
2678 * with the VLAN device's VLAN ID. */
2679 flow->in_port = realdev;
2680 flow->vlan_tci = htons((vid & VLAN_VID_MASK) | VLAN_CFI);
2682 /* Make the packet resemble the flow, so that it gets sent to an
2683 * OpenFlow controller properly, so that it looks correct for
2684 * sFlow, and so that flow_extract() will get the correct vlan_tci
2685 * if it is called on 'packet'.
2687 * The allocated space inside 'packet' probably also contains
2688 * 'key', that is, both 'packet' and 'key' are probably part of a
2689 * struct dpif_upcall (see the large comment on that structure
2690 * definition), so pushing data on 'packet' is in general not a
2691 * good idea since it could overwrite 'key' or free it as a side
2692 * effect. However, it's OK in this special case because we know
2693 * that 'packet' is inside a Netlink attribute: pushing 4 bytes
2694 * will just overwrite the 4-byte "struct nlattr", which is fine
2695 * since we don't need that header anymore. */
2696 eth_push_vlan(packet, flow->vlan_tci);
2699 /* Let the caller know that we can't reproduce 'key' from 'flow'. */
2700 if (fitness == ODP_FIT_PERFECT) {
2701 fitness = ODP_FIT_TOO_MUCH;
2709 handle_miss_upcalls(struct ofproto_dpif *ofproto, struct dpif_upcall *upcalls,
2712 struct dpif_upcall *upcall;
2713 struct flow_miss *miss, *next_miss;
2714 struct flow_miss_op flow_miss_ops[FLOW_MISS_MAX_BATCH * 2];
2715 struct dpif_op *dpif_ops[FLOW_MISS_MAX_BATCH * 2];
2724 /* Construct the to-do list.
2726 * This just amounts to extracting the flow from each packet and sticking
2727 * the packets that have the same flow in the same "flow_miss" structure so
2728 * that we can process them together. */
2730 for (upcall = upcalls; upcall < &upcalls[n_upcalls]; upcall++) {
2731 enum odp_key_fitness fitness;
2732 struct flow_miss *miss;
2733 ovs_be16 initial_tci;
2736 /* Obtain metadata and check userspace/kernel agreement on flow match,
2737 * then set 'flow''s header pointers. */
2738 fitness = ofproto_dpif_extract_flow_key(ofproto,
2739 upcall->key, upcall->key_len,
2740 &flow, &initial_tci,
2742 if (fitness == ODP_FIT_ERROR) {
2743 ofpbuf_delete(upcall->packet);
2746 flow_extract(upcall->packet, flow.skb_priority, flow.tun_id,
2747 flow.in_port, &flow);
2749 /* Handle 802.1ag, LACP, and STP specially. */
2750 if (process_special(ofproto, &flow, upcall->packet)) {
2751 ofproto_update_local_port_stats(&ofproto->up,
2752 0, upcall->packet->size);
2753 ofpbuf_delete(upcall->packet);
2754 ofproto->n_matches++;
2758 /* Add other packets to a to-do list. */
2759 miss = flow_miss_create(&todo, &flow, fitness,
2760 upcall->key, upcall->key_len, initial_tci);
2761 list_push_back(&miss->packets, &upcall->packet->list_node);
2764 /* Process each element in the to-do list, constructing the set of
2765 * operations to batch. */
2767 HMAP_FOR_EACH (miss, hmap_node, &todo) {
2768 handle_flow_miss(ofproto, miss, flow_miss_ops, &n_ops);
2770 assert(n_ops <= ARRAY_SIZE(flow_miss_ops));
2772 /* Execute batch. */
2773 for (i = 0; i < n_ops; i++) {
2774 dpif_ops[i] = &flow_miss_ops[i].dpif_op;
2776 dpif_operate(ofproto->dpif, dpif_ops, n_ops);
2778 /* Free memory and update facets. */
2779 for (i = 0; i < n_ops; i++) {
2780 struct flow_miss_op *op = &flow_miss_ops[i];
2781 struct dpif_execute *execute;
2783 switch (op->dpif_op.type) {
2784 case DPIF_OP_EXECUTE:
2785 execute = &op->dpif_op.u.execute;
2786 if (op->subfacet->actions != execute->actions) {
2787 free((struct nlattr *) execute->actions);
2791 case DPIF_OP_FLOW_PUT:
2792 if (!op->dpif_op.error) {
2793 op->subfacet->installed = true;
2798 HMAP_FOR_EACH_SAFE (miss, next_miss, hmap_node, &todo) {
2799 ofpbuf_list_delete(&miss->packets);
2800 hmap_remove(&todo, &miss->hmap_node);
2803 hmap_destroy(&todo);
2807 handle_userspace_upcall(struct ofproto_dpif *ofproto,
2808 struct dpif_upcall *upcall)
2810 struct user_action_cookie cookie;
2811 enum odp_key_fitness fitness;
2812 ovs_be16 initial_tci;
2815 memcpy(&cookie, &upcall->userdata, sizeof(cookie));
2817 fitness = ofproto_dpif_extract_flow_key(ofproto, upcall->key,
2818 upcall->key_len, &flow,
2819 &initial_tci, upcall->packet);
2820 if (fitness == ODP_FIT_ERROR) {
2821 ofpbuf_delete(upcall->packet);
2825 if (cookie.type == USER_ACTION_COOKIE_SFLOW) {
2826 if (ofproto->sflow) {
2827 dpif_sflow_received(ofproto->sflow, upcall->packet, &flow,
2831 VLOG_WARN_RL(&rl, "invalid user cookie : 0x%"PRIx64, upcall->userdata);
2833 ofpbuf_delete(upcall->packet);
2837 handle_upcalls(struct ofproto_dpif *ofproto, unsigned int max_batch)
2839 struct dpif_upcall misses[FLOW_MISS_MAX_BATCH];
2843 assert (max_batch <= FLOW_MISS_MAX_BATCH);
2846 for (i = 0; i < max_batch; i++) {
2847 struct dpif_upcall *upcall = &misses[n_misses];
2850 error = dpif_recv(ofproto->dpif, upcall);
2855 switch (upcall->type) {
2856 case DPIF_UC_ACTION:
2857 handle_userspace_upcall(ofproto, upcall);
2861 /* Handle it later. */
2865 case DPIF_N_UC_TYPES:
2867 VLOG_WARN_RL(&rl, "upcall has unexpected type %"PRIu32,
2873 handle_miss_upcalls(ofproto, misses, n_misses);
2878 /* Flow expiration. */
2880 static int subfacet_max_idle(const struct ofproto_dpif *);
2881 static void update_stats(struct ofproto_dpif *);
2882 static void rule_expire(struct rule_dpif *);
2883 static void expire_subfacets(struct ofproto_dpif *, int dp_max_idle);
2885 /* This function is called periodically by run(). Its job is to collect
2886 * updates for the flows that have been installed into the datapath, most
2887 * importantly when they last were used, and then use that information to
2888 * expire flows that have not been used recently.
2890 * Returns the number of milliseconds after which it should be called again. */
2892 expire(struct ofproto_dpif *ofproto)
2894 struct rule_dpif *rule, *next_rule;
2895 struct oftable *table;
2898 /* Update stats for each flow in the datapath. */
2899 update_stats(ofproto);
2901 /* Expire subfacets that have been idle too long. */
2902 dp_max_idle = subfacet_max_idle(ofproto);
2903 expire_subfacets(ofproto, dp_max_idle);
2905 /* Expire OpenFlow flows whose idle_timeout or hard_timeout has passed. */
2906 OFPROTO_FOR_EACH_TABLE (table, &ofproto->up) {
2907 struct cls_cursor cursor;
2909 cls_cursor_init(&cursor, &table->cls, NULL);
2910 CLS_CURSOR_FOR_EACH_SAFE (rule, next_rule, up.cr, &cursor) {
2915 /* All outstanding data in existing flows has been accounted, so it's a
2916 * good time to do bond rebalancing. */
2917 if (ofproto->has_bonded_bundles) {
2918 struct ofbundle *bundle;
2920 HMAP_FOR_EACH (bundle, hmap_node, &ofproto->bundles) {
2922 bond_rebalance(bundle->bond, &ofproto->revalidate_set);
2927 return MIN(dp_max_idle, 1000);
2930 /* Update 'packet_count', 'byte_count', and 'used' members of installed facets.
2932 * This function also pushes statistics updates to rules which each facet
2933 * resubmits into. Generally these statistics will be accurate. However, if a
2934 * facet changes the rule it resubmits into at some time in between
2935 * update_stats() runs, it is possible that statistics accrued to the
2936 * old rule will be incorrectly attributed to the new rule. This could be
2937 * avoided by calling update_stats() whenever rules are created or
2938 * deleted. However, the performance impact of making so many calls to the
2939 * datapath do not justify the benefit of having perfectly accurate statistics.
2942 update_stats(struct ofproto_dpif *p)
2944 const struct dpif_flow_stats *stats;
2945 struct dpif_flow_dump dump;
2946 const struct nlattr *key;
2949 dpif_flow_dump_start(&dump, p->dpif);
2950 while (dpif_flow_dump_next(&dump, &key, &key_len, NULL, NULL, &stats)) {
2951 struct subfacet *subfacet;
2953 subfacet = subfacet_find(p, key, key_len);
2954 if (subfacet && subfacet->installed) {
2955 struct facet *facet = subfacet->facet;
2957 if (stats->n_packets >= subfacet->dp_packet_count) {
2958 uint64_t extra = stats->n_packets - subfacet->dp_packet_count;
2959 facet->packet_count += extra;
2961 VLOG_WARN_RL(&rl, "unexpected packet count from the datapath");
2964 if (stats->n_bytes >= subfacet->dp_byte_count) {
2965 facet->byte_count += stats->n_bytes - subfacet->dp_byte_count;
2967 VLOG_WARN_RL(&rl, "unexpected byte count from datapath");
2970 subfacet->dp_packet_count = stats->n_packets;
2971 subfacet->dp_byte_count = stats->n_bytes;
2973 facet->tcp_flags |= stats->tcp_flags;
2975 subfacet_update_time(subfacet, stats->used);
2976 facet_account(facet);
2977 facet_push_stats(facet);
2979 if (!VLOG_DROP_WARN(&rl)) {
2983 odp_flow_key_format(key, key_len, &s);
2984 VLOG_WARN("unexpected flow from datapath %s", ds_cstr(&s));
2988 COVERAGE_INC(facet_unexpected);
2989 /* There's a flow in the datapath that we know nothing about, or a
2990 * flow that shouldn't be installed but was anyway. Delete it. */
2991 dpif_flow_del(p->dpif, key, key_len, NULL);
2994 dpif_flow_dump_done(&dump);
2997 /* Calculates and returns the number of milliseconds of idle time after which
2998 * subfacets should expire from the datapath. When a subfacet expires, we fold
2999 * its statistics into its facet, and when a facet's last subfacet expires, we
3000 * fold its statistic into its rule. */
3002 subfacet_max_idle(const struct ofproto_dpif *ofproto)
3005 * Idle time histogram.
3007 * Most of the time a switch has a relatively small number of subfacets.
3008 * When this is the case we might as well keep statistics for all of them
3009 * in userspace and to cache them in the kernel datapath for performance as
3012 * As the number of subfacets increases, the memory required to maintain
3013 * statistics about them in userspace and in the kernel becomes
3014 * significant. However, with a large number of subfacets it is likely
3015 * that only a few of them are "heavy hitters" that consume a large amount
3016 * of bandwidth. At this point, only heavy hitters are worth caching in
3017 * the kernel and maintaining in userspaces; other subfacets we can
3020 * The technique used to compute the idle time is to build a histogram with
3021 * N_BUCKETS buckets whose width is BUCKET_WIDTH msecs each. Each subfacet
3022 * that is installed in the kernel gets dropped in the appropriate bucket.
3023 * After the histogram has been built, we compute the cutoff so that only
3024 * the most-recently-used 1% of subfacets (but at least
3025 * ofproto->up.flow_eviction_threshold flows) are kept cached. At least
3026 * the most-recently-used bucket of subfacets is kept, so actually an
3027 * arbitrary number of subfacets can be kept in any given expiration run
3028 * (though the next run will delete most of those unless they receive
3031 * This requires a second pass through the subfacets, in addition to the
3032 * pass made by update_stats(), because the former function never looks at
3033 * uninstallable subfacets.
3035 enum { BUCKET_WIDTH = ROUND_UP(100, TIME_UPDATE_INTERVAL) };
3036 enum { N_BUCKETS = 5000 / BUCKET_WIDTH };
3037 int buckets[N_BUCKETS] = { 0 };
3038 int total, subtotal, bucket;
3039 struct subfacet *subfacet;
3043 total = hmap_count(&ofproto->subfacets);
3044 if (total <= ofproto->up.flow_eviction_threshold) {
3045 return N_BUCKETS * BUCKET_WIDTH;
3048 /* Build histogram. */
3050 HMAP_FOR_EACH (subfacet, hmap_node, &ofproto->subfacets) {
3051 long long int idle = now - subfacet->used;
3052 int bucket = (idle <= 0 ? 0
3053 : idle >= BUCKET_WIDTH * N_BUCKETS ? N_BUCKETS - 1
3054 : (unsigned int) idle / BUCKET_WIDTH);
3058 /* Find the first bucket whose flows should be expired. */
3059 subtotal = bucket = 0;
3061 subtotal += buckets[bucket++];
3062 } while (bucket < N_BUCKETS &&
3063 subtotal < MAX(ofproto->up.flow_eviction_threshold, total / 100));
3065 if (VLOG_IS_DBG_ENABLED()) {
3069 ds_put_cstr(&s, "keep");
3070 for (i = 0; i < N_BUCKETS; i++) {
3072 ds_put_cstr(&s, ", drop");
3075 ds_put_format(&s, " %d:%d", i * BUCKET_WIDTH, buckets[i]);
3078 VLOG_INFO("%s: %s (msec:count)", ofproto->up.name, ds_cstr(&s));
3082 return bucket * BUCKET_WIDTH;
3086 expire_subfacets(struct ofproto_dpif *ofproto, int dp_max_idle)
3088 long long int cutoff = time_msec() - dp_max_idle;
3089 struct subfacet *subfacet, *next_subfacet;
3091 HMAP_FOR_EACH_SAFE (subfacet, next_subfacet, hmap_node,
3092 &ofproto->subfacets) {
3093 if (subfacet->used < cutoff) {
3094 subfacet_destroy(subfacet);
3099 /* If 'rule' is an OpenFlow rule, that has expired according to OpenFlow rules,
3100 * then delete it entirely. */
3102 rule_expire(struct rule_dpif *rule)
3104 struct facet *facet, *next_facet;
3108 /* Has 'rule' expired? */
3110 if (rule->up.hard_timeout
3111 && now > rule->up.modified + rule->up.hard_timeout * 1000) {
3112 reason = OFPRR_HARD_TIMEOUT;
3113 } else if (rule->up.idle_timeout
3114 && now > rule->up.used + rule->up.idle_timeout * 1000) {
3115 reason = OFPRR_IDLE_TIMEOUT;
3120 COVERAGE_INC(ofproto_dpif_expired);
3122 /* Update stats. (This is a no-op if the rule expired due to an idle
3123 * timeout, because that only happens when the rule has no facets left.) */
3124 LIST_FOR_EACH_SAFE (facet, next_facet, list_node, &rule->facets) {
3125 facet_remove(facet);
3128 /* Get rid of the rule. */
3129 ofproto_rule_expire(&rule->up, reason);
3134 /* Creates and returns a new facet owned by 'rule', given a 'flow'.
3136 * The caller must already have determined that no facet with an identical
3137 * 'flow' exists in 'ofproto' and that 'flow' is the best match for 'rule' in
3138 * the ofproto's classifier table.
3140 * The facet will initially have no subfacets. The caller should create (at
3141 * least) one subfacet with subfacet_create(). */
3142 static struct facet *
3143 facet_create(struct rule_dpif *rule, const struct flow *flow)
3145 struct ofproto_dpif *ofproto = ofproto_dpif_cast(rule->up.ofproto);
3146 struct facet *facet;
3148 facet = xzalloc(sizeof *facet);
3149 facet->used = time_msec();
3150 hmap_insert(&ofproto->facets, &facet->hmap_node, flow_hash(flow, 0));
3151 list_push_back(&rule->facets, &facet->list_node);
3153 facet->flow = *flow;
3154 list_init(&facet->subfacets);
3155 netflow_flow_init(&facet->nf_flow);
3156 netflow_flow_update_time(ofproto->netflow, &facet->nf_flow, facet->used);
3162 facet_free(struct facet *facet)
3167 /* Executes, within 'ofproto', the 'n_actions' actions in 'actions' on
3168 * 'packet', which arrived on 'in_port'.
3170 * Takes ownership of 'packet'. */
3172 execute_odp_actions(struct ofproto_dpif *ofproto, const struct flow *flow,
3173 const struct nlattr *odp_actions, size_t actions_len,
3174 struct ofpbuf *packet)
3176 struct odputil_keybuf keybuf;
3180 ofpbuf_use_stack(&key, &keybuf, sizeof keybuf);
3181 odp_flow_key_from_flow(&key, flow);
3183 error = dpif_execute(ofproto->dpif, key.data, key.size,
3184 odp_actions, actions_len, packet);
3186 ofpbuf_delete(packet);
3190 /* Remove 'facet' from 'ofproto' and free up the associated memory:
3192 * - If 'facet' was installed in the datapath, uninstalls it and updates its
3193 * rule's statistics, via subfacet_uninstall().
3195 * - Removes 'facet' from its rule and from ofproto->facets.
3198 facet_remove(struct facet *facet)
3200 struct ofproto_dpif *ofproto = ofproto_dpif_cast(facet->rule->up.ofproto);
3201 struct subfacet *subfacet, *next_subfacet;
3203 assert(!list_is_empty(&facet->subfacets));
3205 /* First uninstall all of the subfacets to get final statistics. */
3206 LIST_FOR_EACH (subfacet, list_node, &facet->subfacets) {
3207 subfacet_uninstall(subfacet);
3210 /* Flush the final stats to the rule.
3212 * This might require us to have at least one subfacet around so that we
3213 * can use its actions for accounting in facet_account(), which is why we
3214 * have uninstalled but not yet destroyed the subfacets. */
3215 facet_flush_stats(facet);
3217 /* Now we're really all done so destroy everything. */
3218 LIST_FOR_EACH_SAFE (subfacet, next_subfacet, list_node,
3219 &facet->subfacets) {
3220 subfacet_destroy__(subfacet);
3222 hmap_remove(&ofproto->facets, &facet->hmap_node);
3223 list_remove(&facet->list_node);
3228 facet_account(struct facet *facet)
3230 struct ofproto_dpif *ofproto = ofproto_dpif_cast(facet->rule->up.ofproto);
3232 struct subfacet *subfacet;
3233 const struct nlattr *a;
3237 if (facet->byte_count <= facet->accounted_bytes) {
3240 n_bytes = facet->byte_count - facet->accounted_bytes;
3241 facet->accounted_bytes = facet->byte_count;
3243 /* Feed information from the active flows back into the learning table to
3244 * ensure that table is always in sync with what is actually flowing
3245 * through the datapath. */
3246 if (facet->has_learn || facet->has_normal
3247 || (facet->has_fin_timeout
3248 && facet->tcp_flags & (TCP_FIN | TCP_RST))) {
3249 struct action_xlate_ctx ctx;
3251 action_xlate_ctx_init(&ctx, ofproto, &facet->flow,
3252 facet->flow.vlan_tci,
3253 facet->rule, facet->tcp_flags, NULL);
3254 ctx.may_learn = true;
3255 ofpbuf_delete(xlate_actions(&ctx, facet->rule->up.actions,
3256 facet->rule->up.n_actions));
3259 if (!facet->has_normal || !ofproto->has_bonded_bundles) {
3263 /* This loop feeds byte counters to bond_account() for rebalancing to use
3264 * as a basis. We also need to track the actual VLAN on which the packet
3265 * is going to be sent to ensure that it matches the one passed to
3266 * bond_choose_output_slave(). (Otherwise, we will account to the wrong
3269 * We use the actions from an arbitrary subfacet because they should all
3270 * be equally valid for our purpose. */
3271 subfacet = CONTAINER_OF(list_front(&facet->subfacets),
3272 struct subfacet, list_node);
3273 vlan_tci = facet->flow.vlan_tci;
3274 NL_ATTR_FOR_EACH_UNSAFE (a, left,
3275 subfacet->actions, subfacet->actions_len) {
3276 const struct ovs_action_push_vlan *vlan;
3277 struct ofport_dpif *port;
3279 switch (nl_attr_type(a)) {
3280 case OVS_ACTION_ATTR_OUTPUT:
3281 port = get_odp_port(ofproto, nl_attr_get_u32(a));
3282 if (port && port->bundle && port->bundle->bond) {
3283 bond_account(port->bundle->bond, &facet->flow,
3284 vlan_tci_to_vid(vlan_tci), n_bytes);
3288 case OVS_ACTION_ATTR_POP_VLAN:
3289 vlan_tci = htons(0);
3292 case OVS_ACTION_ATTR_PUSH_VLAN:
3293 vlan = nl_attr_get(a);
3294 vlan_tci = vlan->vlan_tci;
3300 /* Returns true if the only action for 'facet' is to send to the controller.
3301 * (We don't report NetFlow expiration messages for such facets because they
3302 * are just part of the control logic for the network, not real traffic). */
3304 facet_is_controller_flow(struct facet *facet)
3307 && facet->rule->up.n_actions == 1
3308 && action_outputs_to_port(&facet->rule->up.actions[0],
3309 htons(OFPP_CONTROLLER)));
3312 /* Folds all of 'facet''s statistics into its rule. Also updates the
3313 * accounting ofhook and emits a NetFlow expiration if appropriate. All of
3314 * 'facet''s statistics in the datapath should have been zeroed and folded into
3315 * its packet and byte counts before this function is called. */
3317 facet_flush_stats(struct facet *facet)
3319 struct ofproto_dpif *ofproto = ofproto_dpif_cast(facet->rule->up.ofproto);
3320 struct subfacet *subfacet;
3322 LIST_FOR_EACH (subfacet, list_node, &facet->subfacets) {
3323 assert(!subfacet->dp_byte_count);
3324 assert(!subfacet->dp_packet_count);
3327 facet_push_stats(facet);
3328 facet_account(facet);
3330 if (ofproto->netflow && !facet_is_controller_flow(facet)) {
3331 struct ofexpired expired;
3332 expired.flow = facet->flow;
3333 expired.packet_count = facet->packet_count;
3334 expired.byte_count = facet->byte_count;
3335 expired.used = facet->used;
3336 netflow_expire(ofproto->netflow, &facet->nf_flow, &expired);
3339 facet->rule->packet_count += facet->packet_count;
3340 facet->rule->byte_count += facet->byte_count;
3342 /* Reset counters to prevent double counting if 'facet' ever gets
3344 facet_reset_counters(facet);
3346 netflow_flow_clear(&facet->nf_flow);
3347 facet->tcp_flags = 0;
3350 /* Searches 'ofproto''s table of facets for one exactly equal to 'flow'.
3351 * Returns it if found, otherwise a null pointer.
3353 * The returned facet might need revalidation; use facet_lookup_valid()
3354 * instead if that is important. */
3355 static struct facet *
3356 facet_find(struct ofproto_dpif *ofproto, const struct flow *flow)
3358 struct facet *facet;
3360 HMAP_FOR_EACH_WITH_HASH (facet, hmap_node, flow_hash(flow, 0),
3362 if (flow_equal(flow, &facet->flow)) {
3370 /* Searches 'ofproto''s table of facets for one exactly equal to 'flow'.
3371 * Returns it if found, otherwise a null pointer.
3373 * The returned facet is guaranteed to be valid. */
3374 static struct facet *
3375 facet_lookup_valid(struct ofproto_dpif *ofproto, const struct flow *flow)
3377 struct facet *facet = facet_find(ofproto, flow);
3379 /* The facet we found might not be valid, since we could be in need of
3380 * revalidation. If it is not valid, don't return it. */
3382 && (ofproto->need_revalidate
3383 || tag_set_intersects(&ofproto->revalidate_set, facet->tags))
3384 && !facet_revalidate(facet)) {
3385 COVERAGE_INC(facet_invalidated);
3393 facet_check_consistency(struct facet *facet)
3395 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 15);
3397 struct ofproto_dpif *ofproto = ofproto_dpif_cast(facet->rule->up.ofproto);
3399 struct rule_dpif *rule;
3400 struct subfacet *subfacet;
3401 bool may_log = false;
3404 /* Check the rule for consistency. */
3405 rule = rule_dpif_lookup(ofproto, &facet->flow, 0);
3407 if (!VLOG_DROP_WARN(&rl)) {
3408 char *s = flow_to_string(&facet->flow);
3409 VLOG_WARN("%s: facet should not exist", s);
3413 } else if (rule != facet->rule) {
3414 may_log = !VLOG_DROP_WARN(&rl);
3420 flow_format(&s, &facet->flow);
3421 ds_put_format(&s, ": facet associated with wrong rule (was "
3422 "table=%"PRIu8",", facet->rule->up.table_id);
3423 cls_rule_format(&facet->rule->up.cr, &s);
3424 ds_put_format(&s, ") (should have been table=%"PRIu8",",
3426 cls_rule_format(&rule->up.cr, &s);
3427 ds_put_char(&s, ')');
3429 VLOG_WARN("%s", ds_cstr(&s));
3436 /* Check the datapath actions for consistency. */
3437 LIST_FOR_EACH (subfacet, list_node, &facet->subfacets) {
3438 struct action_xlate_ctx ctx;
3439 struct ofpbuf *odp_actions;
3440 bool actions_changed;
3441 bool should_install;
3443 action_xlate_ctx_init(&ctx, ofproto, &facet->flow,
3444 subfacet->initial_tci, rule, 0, NULL);
3445 odp_actions = xlate_actions(&ctx, rule->up.actions,
3446 rule->up.n_actions);
3448 should_install = (ctx.may_set_up_flow
3449 && subfacet->key_fitness != ODP_FIT_TOO_LITTLE);
3450 if (!should_install && !subfacet->installed) {
3451 /* The actions for uninstallable flows may vary from one packet to
3452 * the next, so don't compare the actions. */
3456 actions_changed = (subfacet->actions_len != odp_actions->size
3457 || memcmp(subfacet->actions, odp_actions->data,
3458 subfacet->actions_len));
3459 if (should_install != subfacet->installed || actions_changed) {
3461 may_log = !VLOG_DROP_WARN(&rl);
3466 struct odputil_keybuf keybuf;
3471 subfacet_get_key(subfacet, &keybuf, &key);
3472 odp_flow_key_format(key.data, key.size, &s);
3474 ds_put_cstr(&s, ": inconsistency in subfacet");
3475 if (should_install != subfacet->installed) {
3476 enum odp_key_fitness fitness = subfacet->key_fitness;
3478 ds_put_format(&s, " (should%s have been installed)",
3479 should_install ? "" : " not");
3480 ds_put_format(&s, " (may_set_up_flow=%s, fitness=%s)",
3481 ctx.may_set_up_flow ? "true" : "false",
3482 odp_key_fitness_to_string(fitness));
3484 if (actions_changed) {
3485 ds_put_cstr(&s, " (actions were: ");
3486 format_odp_actions(&s, subfacet->actions,
3487 subfacet->actions_len);
3488 ds_put_cstr(&s, ") (correct actions: ");
3489 format_odp_actions(&s, odp_actions->data,
3491 ds_put_char(&s, ')');
3493 ds_put_cstr(&s, " (actions: ");
3494 format_odp_actions(&s, subfacet->actions,
3495 subfacet->actions_len);
3496 ds_put_char(&s, ')');
3498 VLOG_WARN("%s", ds_cstr(&s));
3504 ofpbuf_delete(odp_actions);
3510 /* Re-searches the classifier for 'facet':
3512 * - If the rule found is different from 'facet''s current rule, moves
3513 * 'facet' to the new rule and recompiles its actions.
3515 * - If the rule found is the same as 'facet''s current rule, leaves 'facet'
3516 * where it is and recompiles its actions anyway.
3518 * - If there is none, destroys 'facet'.
3520 * Returns true if 'facet' still exists, false if it has been destroyed. */
3522 facet_revalidate(struct facet *facet)
3524 struct ofproto_dpif *ofproto = ofproto_dpif_cast(facet->rule->up.ofproto);
3526 struct nlattr *odp_actions;
3529 struct actions *new_actions;
3531 struct action_xlate_ctx ctx;
3532 struct rule_dpif *new_rule;
3533 struct subfacet *subfacet;
3534 bool actions_changed;
3537 COVERAGE_INC(facet_revalidate);
3539 /* Determine the new rule. */
3540 new_rule = rule_dpif_lookup(ofproto, &facet->flow, 0);
3542 /* No new rule, so delete the facet. */
3543 facet_remove(facet);
3547 /* Calculate new datapath actions.
3549 * We do not modify any 'facet' state yet, because we might need to, e.g.,
3550 * emit a NetFlow expiration and, if so, we need to have the old state
3551 * around to properly compose it. */
3553 /* If the datapath actions changed or the installability changed,
3554 * then we need to talk to the datapath. */
3557 memset(&ctx, 0, sizeof ctx);
3558 LIST_FOR_EACH (subfacet, list_node, &facet->subfacets) {
3559 struct ofpbuf *odp_actions;
3560 bool should_install;
3562 action_xlate_ctx_init(&ctx, ofproto, &facet->flow,
3563 subfacet->initial_tci, new_rule, 0, NULL);
3564 odp_actions = xlate_actions(&ctx, new_rule->up.actions,
3565 new_rule->up.n_actions);
3566 actions_changed = (subfacet->actions_len != odp_actions->size
3567 || memcmp(subfacet->actions, odp_actions->data,
3568 subfacet->actions_len));
3570 should_install = (ctx.may_set_up_flow
3571 && subfacet->key_fitness != ODP_FIT_TOO_LITTLE);
3572 if (actions_changed || should_install != subfacet->installed) {
3573 if (should_install) {
3574 struct dpif_flow_stats stats;
3576 subfacet_install(subfacet,
3577 odp_actions->data, odp_actions->size, &stats);
3578 subfacet_update_stats(subfacet, &stats);
3580 subfacet_uninstall(subfacet);
3584 new_actions = xcalloc(list_size(&facet->subfacets),
3585 sizeof *new_actions);
3587 new_actions[i].odp_actions = xmemdup(odp_actions->data,
3589 new_actions[i].actions_len = odp_actions->size;
3592 ofpbuf_delete(odp_actions);
3596 facet_flush_stats(facet);
3599 /* Update 'facet' now that we've taken care of all the old state. */
3600 facet->tags = ctx.tags;
3601 facet->nf_flow.output_iface = ctx.nf_output_iface;
3602 facet->may_install = ctx.may_set_up_flow;
3603 facet->has_learn = ctx.has_learn;
3604 facet->has_normal = ctx.has_normal;
3605 facet->has_fin_timeout = ctx.has_fin_timeout;
3606 facet->mirrors = ctx.mirrors;
3609 LIST_FOR_EACH (subfacet, list_node, &facet->subfacets) {
3610 if (new_actions[i].odp_actions) {
3611 free(subfacet->actions);
3612 subfacet->actions = new_actions[i].odp_actions;
3613 subfacet->actions_len = new_actions[i].actions_len;
3619 if (facet->rule != new_rule) {
3620 COVERAGE_INC(facet_changed_rule);
3621 list_remove(&facet->list_node);
3622 list_push_back(&new_rule->facets, &facet->list_node);
3623 facet->rule = new_rule;
3624 facet->used = new_rule->up.created;
3625 facet->prev_used = facet->used;
3631 /* Updates 'facet''s used time. Caller is responsible for calling
3632 * facet_push_stats() to update the flows which 'facet' resubmits into. */
3634 facet_update_time(struct facet *facet, long long int used)
3636 struct ofproto_dpif *ofproto = ofproto_dpif_cast(facet->rule->up.ofproto);
3637 if (used > facet->used) {
3639 ofproto_rule_update_used(&facet->rule->up, used);
3640 netflow_flow_update_time(ofproto->netflow, &facet->nf_flow, used);
3645 facet_reset_counters(struct facet *facet)
3647 facet->packet_count = 0;
3648 facet->byte_count = 0;
3649 facet->prev_packet_count = 0;
3650 facet->prev_byte_count = 0;
3651 facet->accounted_bytes = 0;
3655 facet_push_stats(struct facet *facet)
3657 uint64_t new_packets, new_bytes;
3659 assert(facet->packet_count >= facet->prev_packet_count);
3660 assert(facet->byte_count >= facet->prev_byte_count);
3661 assert(facet->used >= facet->prev_used);
3663 new_packets = facet->packet_count - facet->prev_packet_count;
3664 new_bytes = facet->byte_count - facet->prev_byte_count;
3666 if (new_packets || new_bytes || facet->used > facet->prev_used) {
3667 facet->prev_packet_count = facet->packet_count;
3668 facet->prev_byte_count = facet->byte_count;
3669 facet->prev_used = facet->used;
3671 flow_push_stats(facet->rule, &facet->flow,
3672 new_packets, new_bytes, facet->used);
3674 update_mirror_stats(ofproto_dpif_cast(facet->rule->up.ofproto),
3675 facet->mirrors, new_packets, new_bytes);
3679 struct ofproto_push {
3680 struct action_xlate_ctx ctx;
3687 push_resubmit(struct action_xlate_ctx *ctx, struct rule_dpif *rule)
3689 struct ofproto_push *push = CONTAINER_OF(ctx, struct ofproto_push, ctx);
3692 rule->packet_count += push->packets;
3693 rule->byte_count += push->bytes;
3694 ofproto_rule_update_used(&rule->up, push->used);
3698 /* Pushes flow statistics to the rules which 'flow' resubmits into given
3699 * 'rule''s actions and mirrors. */
3701 flow_push_stats(struct rule_dpif *rule,
3702 const struct flow *flow, uint64_t packets, uint64_t bytes,
3705 struct ofproto_dpif *ofproto = ofproto_dpif_cast(rule->up.ofproto);
3706 struct ofproto_push push;
3708 push.packets = packets;
3712 ofproto_rule_update_used(&rule->up, used);
3714 action_xlate_ctx_init(&push.ctx, ofproto, flow, flow->vlan_tci, rule,
3716 push.ctx.resubmit_hook = push_resubmit;
3717 ofpbuf_delete(xlate_actions(&push.ctx,
3718 rule->up.actions, rule->up.n_actions));
3723 static struct subfacet *
3724 subfacet_find__(struct ofproto_dpif *ofproto,
3725 const struct nlattr *key, size_t key_len, uint32_t key_hash,
3726 const struct flow *flow)
3728 struct subfacet *subfacet;
3730 HMAP_FOR_EACH_WITH_HASH (subfacet, hmap_node, key_hash,
3731 &ofproto->subfacets) {
3733 ? (subfacet->key_len == key_len
3734 && !memcmp(key, subfacet->key, key_len))
3735 : flow_equal(flow, &subfacet->facet->flow)) {
3743 /* Searches 'facet' (within 'ofproto') for a subfacet with the specified
3744 * 'key_fitness', 'key', and 'key_len'. Returns the existing subfacet if
3745 * there is one, otherwise creates and returns a new subfacet.
3747 * If the returned subfacet is new, then subfacet->actions will be NULL, in
3748 * which case the caller must populate the actions with
3749 * subfacet_make_actions(). */
3750 static struct subfacet *
3751 subfacet_create(struct facet *facet, enum odp_key_fitness key_fitness,
3752 const struct nlattr *key, size_t key_len, ovs_be16 initial_tci)
3754 struct ofproto_dpif *ofproto = ofproto_dpif_cast(facet->rule->up.ofproto);
3755 uint32_t key_hash = odp_flow_key_hash(key, key_len);
3756 struct subfacet *subfacet;
3758 subfacet = subfacet_find__(ofproto, key, key_len, key_hash, &facet->flow);
3760 if (subfacet->facet == facet) {
3764 /* This shouldn't happen. */
3765 VLOG_ERR_RL(&rl, "subfacet with wrong facet");
3766 subfacet_destroy(subfacet);
3769 subfacet = xzalloc(sizeof *subfacet);
3770 hmap_insert(&ofproto->subfacets, &subfacet->hmap_node, key_hash);
3771 list_push_back(&facet->subfacets, &subfacet->list_node);
3772 subfacet->facet = facet;
3773 subfacet->used = time_msec();
3774 subfacet->key_fitness = key_fitness;
3775 if (key_fitness != ODP_FIT_PERFECT) {
3776 subfacet->key = xmemdup(key, key_len);
3777 subfacet->key_len = key_len;
3779 subfacet->installed = false;
3780 subfacet->initial_tci = initial_tci;
3785 /* Searches 'ofproto' for a subfacet with the given 'key', 'key_len', and
3786 * 'flow'. Returns the subfacet if one exists, otherwise NULL. */
3787 static struct subfacet *
3788 subfacet_find(struct ofproto_dpif *ofproto,
3789 const struct nlattr *key, size_t key_len)
3791 uint32_t key_hash = odp_flow_key_hash(key, key_len);
3792 enum odp_key_fitness fitness;
3795 fitness = odp_flow_key_to_flow(key, key_len, &flow);
3796 if (fitness == ODP_FIT_ERROR) {
3800 return subfacet_find__(ofproto, key, key_len, key_hash, &flow);
3803 /* Uninstalls 'subfacet' from the datapath, if it is installed, removes it from
3804 * its facet within 'ofproto', and frees it. */
3806 subfacet_destroy__(struct subfacet *subfacet)
3808 struct facet *facet = subfacet->facet;
3809 struct ofproto_dpif *ofproto = ofproto_dpif_cast(facet->rule->up.ofproto);
3811 subfacet_uninstall(subfacet);
3812 hmap_remove(&ofproto->subfacets, &subfacet->hmap_node);
3813 list_remove(&subfacet->list_node);
3814 free(subfacet->key);
3815 free(subfacet->actions);
3819 /* Destroys 'subfacet', as with subfacet_destroy__(), and then if this was the
3820 * last remaining subfacet in its facet destroys the facet too. */
3822 subfacet_destroy(struct subfacet *subfacet)
3824 struct facet *facet = subfacet->facet;
3826 if (list_is_singleton(&facet->subfacets)) {
3827 /* facet_remove() needs at least one subfacet (it will remove it). */
3828 facet_remove(facet);
3830 subfacet_destroy__(subfacet);
3834 /* Initializes 'key' with the sequence of OVS_KEY_ATTR_* Netlink attributes
3835 * that can be used to refer to 'subfacet'. The caller must provide 'keybuf'
3836 * for use as temporary storage. */
3838 subfacet_get_key(struct subfacet *subfacet, struct odputil_keybuf *keybuf,
3841 if (!subfacet->key) {
3842 ofpbuf_use_stack(key, keybuf, sizeof *keybuf);
3843 odp_flow_key_from_flow(key, &subfacet->facet->flow);
3845 ofpbuf_use_const(key, subfacet->key, subfacet->key_len);
3849 /* Composes the datapath actions for 'subfacet' based on its rule's actions. */
3851 subfacet_make_actions(struct subfacet *subfacet, const struct ofpbuf *packet)
3853 struct facet *facet = subfacet->facet;
3854 struct rule_dpif *rule = facet->rule;
3855 struct ofproto_dpif *ofproto = ofproto_dpif_cast(rule->up.ofproto);
3856 struct ofpbuf *odp_actions;
3857 struct action_xlate_ctx ctx;
3859 action_xlate_ctx_init(&ctx, ofproto, &facet->flow, subfacet->initial_tci,
3861 odp_actions = xlate_actions(&ctx, rule->up.actions, rule->up.n_actions);
3862 facet->tags = ctx.tags;
3863 facet->may_install = ctx.may_set_up_flow;
3864 facet->has_learn = ctx.has_learn;
3865 facet->has_normal = ctx.has_normal;
3866 facet->has_fin_timeout = ctx.has_fin_timeout;
3867 facet->nf_flow.output_iface = ctx.nf_output_iface;
3868 facet->mirrors = ctx.mirrors;
3870 if (subfacet->actions_len != odp_actions->size
3871 || memcmp(subfacet->actions, odp_actions->data, odp_actions->size)) {
3872 free(subfacet->actions);
3873 subfacet->actions_len = odp_actions->size;
3874 subfacet->actions = xmemdup(odp_actions->data, odp_actions->size);
3877 ofpbuf_delete(odp_actions);
3880 /* Updates 'subfacet''s datapath flow, setting its actions to 'actions_len'
3881 * bytes of actions in 'actions'. If 'stats' is non-null, statistics counters
3882 * in the datapath will be zeroed and 'stats' will be updated with traffic new
3883 * since 'subfacet' was last updated.
3885 * Returns 0 if successful, otherwise a positive errno value. */
3887 subfacet_install(struct subfacet *subfacet,
3888 const struct nlattr *actions, size_t actions_len,
3889 struct dpif_flow_stats *stats)
3891 struct facet *facet = subfacet->facet;
3892 struct ofproto_dpif *ofproto = ofproto_dpif_cast(facet->rule->up.ofproto);
3893 struct odputil_keybuf keybuf;
3894 enum dpif_flow_put_flags flags;
3898 flags = DPIF_FP_CREATE | DPIF_FP_MODIFY;
3900 flags |= DPIF_FP_ZERO_STATS;
3903 subfacet_get_key(subfacet, &keybuf, &key);
3904 ret = dpif_flow_put(ofproto->dpif, flags, key.data, key.size,
3905 actions, actions_len, stats);
3908 subfacet_reset_dp_stats(subfacet, stats);
3914 /* If 'subfacet' is installed in the datapath, uninstalls it. */
3916 subfacet_uninstall(struct subfacet *subfacet)
3918 if (subfacet->installed) {
3919 struct rule_dpif *rule = subfacet->facet->rule;
3920 struct ofproto_dpif *ofproto = ofproto_dpif_cast(rule->up.ofproto);
3921 struct odputil_keybuf keybuf;
3922 struct dpif_flow_stats stats;
3926 subfacet_get_key(subfacet, &keybuf, &key);
3927 error = dpif_flow_del(ofproto->dpif, key.data, key.size, &stats);
3928 subfacet_reset_dp_stats(subfacet, &stats);
3930 subfacet_update_stats(subfacet, &stats);
3932 subfacet->installed = false;
3934 assert(subfacet->dp_packet_count == 0);
3935 assert(subfacet->dp_byte_count == 0);
3939 /* Resets 'subfacet''s datapath statistics counters. This should be called
3940 * when 'subfacet''s statistics are cleared in the datapath. If 'stats' is
3941 * non-null, it should contain the statistics returned by dpif when 'subfacet'
3942 * was reset in the datapath. 'stats' will be modified to include only
3943 * statistics new since 'subfacet' was last updated. */
3945 subfacet_reset_dp_stats(struct subfacet *subfacet,
3946 struct dpif_flow_stats *stats)
3949 && subfacet->dp_packet_count <= stats->n_packets
3950 && subfacet->dp_byte_count <= stats->n_bytes) {
3951 stats->n_packets -= subfacet->dp_packet_count;
3952 stats->n_bytes -= subfacet->dp_byte_count;
3955 subfacet->dp_packet_count = 0;
3956 subfacet->dp_byte_count = 0;
3959 /* Updates 'subfacet''s used time. The caller is responsible for calling
3960 * facet_push_stats() to update the flows which 'subfacet' resubmits into. */
3962 subfacet_update_time(struct subfacet *subfacet, long long int used)
3964 if (used > subfacet->used) {
3965 subfacet->used = used;
3966 facet_update_time(subfacet->facet, used);
3970 /* Folds the statistics from 'stats' into the counters in 'subfacet'.
3972 * Because of the meaning of a subfacet's counters, it only makes sense to do
3973 * this if 'stats' are not tracked in the datapath, that is, if 'stats'
3974 * represents a packet that was sent by hand or if it represents statistics
3975 * that have been cleared out of the datapath. */
3977 subfacet_update_stats(struct subfacet *subfacet,
3978 const struct dpif_flow_stats *stats)
3980 if (stats->n_packets || stats->used > subfacet->used) {
3981 struct facet *facet = subfacet->facet;
3983 subfacet_update_time(subfacet, stats->used);
3984 facet->packet_count += stats->n_packets;
3985 facet->byte_count += stats->n_bytes;
3986 facet->tcp_flags |= stats->tcp_flags;
3987 facet_push_stats(facet);
3988 netflow_flow_update_flags(&facet->nf_flow, stats->tcp_flags);
3994 static struct rule_dpif *
3995 rule_dpif_lookup(struct ofproto_dpif *ofproto, const struct flow *flow,
3998 struct cls_rule *cls_rule;
3999 struct classifier *cls;
4001 if (table_id >= N_TABLES) {
4005 cls = &ofproto->up.tables[table_id].cls;
4006 if (flow->nw_frag & FLOW_NW_FRAG_ANY
4007 && ofproto->up.frag_handling == OFPC_FRAG_NORMAL) {
4008 /* For OFPC_NORMAL frag_handling, we must pretend that transport ports
4009 * are unavailable. */
4010 struct flow ofpc_normal_flow = *flow;
4011 ofpc_normal_flow.tp_src = htons(0);
4012 ofpc_normal_flow.tp_dst = htons(0);
4013 cls_rule = classifier_lookup(cls, &ofpc_normal_flow);
4015 cls_rule = classifier_lookup(cls, flow);
4017 return rule_dpif_cast(rule_from_cls_rule(cls_rule));
4021 complete_operation(struct rule_dpif *rule)
4023 struct ofproto_dpif *ofproto = ofproto_dpif_cast(rule->up.ofproto);
4025 rule_invalidate(rule);
4027 struct dpif_completion *c = xmalloc(sizeof *c);
4028 c->op = rule->up.pending;
4029 list_push_back(&ofproto->completions, &c->list_node);
4031 ofoperation_complete(rule->up.pending, 0);
4035 static struct rule *
4038 struct rule_dpif *rule = xmalloc(sizeof *rule);
4043 rule_dealloc(struct rule *rule_)
4045 struct rule_dpif *rule = rule_dpif_cast(rule_);
4050 rule_construct(struct rule *rule_)
4052 struct rule_dpif *rule = rule_dpif_cast(rule_);
4053 struct ofproto_dpif *ofproto = ofproto_dpif_cast(rule->up.ofproto);
4054 struct rule_dpif *victim;
4058 error = validate_actions(rule->up.actions, rule->up.n_actions,
4059 &rule->up.cr.flow, ofproto->max_ports);
4064 rule->packet_count = 0;
4065 rule->byte_count = 0;
4067 victim = rule_dpif_cast(ofoperation_get_victim(rule->up.pending));
4068 if (victim && !list_is_empty(&victim->facets)) {
4069 struct facet *facet;
4071 rule->facets = victim->facets;
4072 list_moved(&rule->facets);
4073 LIST_FOR_EACH (facet, list_node, &rule->facets) {
4074 /* XXX: We're only clearing our local counters here. It's possible
4075 * that quite a few packets are unaccounted for in the datapath
4076 * statistics. These will be accounted to the new rule instead of
4077 * cleared as required. This could be fixed by clearing out the
4078 * datapath statistics for this facet, but currently it doesn't
4080 facet_reset_counters(facet);
4084 /* Must avoid list_moved() in this case. */
4085 list_init(&rule->facets);
4088 table_id = rule->up.table_id;
4089 rule->tag = (victim ? victim->tag
4091 : rule_calculate_tag(&rule->up.cr.flow, &rule->up.cr.wc,
4092 ofproto->tables[table_id].basis));
4094 complete_operation(rule);
4099 rule_destruct(struct rule *rule_)
4101 struct rule_dpif *rule = rule_dpif_cast(rule_);
4102 struct facet *facet, *next_facet;
4104 LIST_FOR_EACH_SAFE (facet, next_facet, list_node, &rule->facets) {
4105 facet_revalidate(facet);
4108 complete_operation(rule);
4112 rule_get_stats(struct rule *rule_, uint64_t *packets, uint64_t *bytes)
4114 struct rule_dpif *rule = rule_dpif_cast(rule_);
4115 struct facet *facet;
4117 /* Start from historical data for 'rule' itself that are no longer tracked
4118 * in facets. This counts, for example, facets that have expired. */
4119 *packets = rule->packet_count;
4120 *bytes = rule->byte_count;
4122 /* Add any statistics that are tracked by facets. This includes
4123 * statistical data recently updated by ofproto_update_stats() as well as
4124 * stats for packets that were executed "by hand" via dpif_execute(). */
4125 LIST_FOR_EACH (facet, list_node, &rule->facets) {
4126 *packets += facet->packet_count;
4127 *bytes += facet->byte_count;
4132 rule_execute(struct rule *rule_, const struct flow *flow,
4133 struct ofpbuf *packet)
4135 struct rule_dpif *rule = rule_dpif_cast(rule_);
4136 struct ofproto_dpif *ofproto = ofproto_dpif_cast(rule->up.ofproto);
4137 struct action_xlate_ctx ctx;
4138 struct ofpbuf *odp_actions;
4141 action_xlate_ctx_init(&ctx, ofproto, flow, flow->vlan_tci,
4142 rule, packet_get_tcp_flags(packet, flow), packet);
4143 odp_actions = xlate_actions(&ctx, rule->up.actions, rule->up.n_actions);
4144 size = packet->size;
4145 if (execute_odp_actions(ofproto, flow, odp_actions->data,
4146 odp_actions->size, packet)) {
4147 rule->packet_count++;
4148 rule->byte_count += size;
4149 flow_push_stats(rule, flow, 1, size, time_msec());
4151 ofpbuf_delete(odp_actions);
4157 rule_modify_actions(struct rule *rule_)
4159 struct rule_dpif *rule = rule_dpif_cast(rule_);
4160 struct ofproto_dpif *ofproto = ofproto_dpif_cast(rule->up.ofproto);
4163 error = validate_actions(rule->up.actions, rule->up.n_actions,
4164 &rule->up.cr.flow, ofproto->max_ports);
4166 ofoperation_complete(rule->up.pending, error);
4170 complete_operation(rule);
4173 /* Sends 'packet' out 'ofport'.
4174 * May modify 'packet'.
4175 * Returns 0 if successful, otherwise a positive errno value. */
4177 send_packet(const struct ofport_dpif *ofport, struct ofpbuf *packet)
4179 const struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport->up.ofproto);
4180 struct ofpbuf key, odp_actions;
4181 struct odputil_keybuf keybuf;
4186 flow_extract((struct ofpbuf *) packet, 0, 0, 0, &flow);
4187 odp_port = vsp_realdev_to_vlandev(ofproto, ofport->odp_port,
4189 if (odp_port != ofport->odp_port) {
4190 eth_pop_vlan(packet);
4191 flow.vlan_tci = htons(0);
4194 ofpbuf_use_stack(&key, &keybuf, sizeof keybuf);
4195 odp_flow_key_from_flow(&key, &flow);
4197 ofpbuf_init(&odp_actions, 32);
4198 compose_sflow_action(ofproto, &odp_actions, &flow, odp_port);
4200 nl_msg_put_u32(&odp_actions, OVS_ACTION_ATTR_OUTPUT, odp_port);
4201 error = dpif_execute(ofproto->dpif,
4203 odp_actions.data, odp_actions.size,
4205 ofpbuf_uninit(&odp_actions);
4208 VLOG_WARN_RL(&rl, "%s: failed to send packet on port %"PRIu32" (%s)",
4209 ofproto->up.name, odp_port, strerror(error));
4211 ofproto_update_local_port_stats(ofport->up.ofproto, packet->size, 0);
4215 /* OpenFlow to datapath action translation. */
4217 static void do_xlate_actions(const union ofp_action *in, size_t n_in,
4218 struct action_xlate_ctx *ctx);
4219 static void xlate_normal(struct action_xlate_ctx *);
4222 put_userspace_action(const struct ofproto_dpif *ofproto,
4223 struct ofpbuf *odp_actions,
4224 const struct flow *flow,
4225 const struct user_action_cookie *cookie)
4229 pid = dpif_port_get_pid(ofproto->dpif,
4230 ofp_port_to_odp_port(flow->in_port));
4232 return odp_put_userspace_action(pid, cookie, odp_actions);
4235 /* Compose SAMPLE action for sFlow. */
4237 compose_sflow_action(const struct ofproto_dpif *ofproto,
4238 struct ofpbuf *odp_actions,
4239 const struct flow *flow,
4242 uint32_t port_ifindex;
4243 uint32_t probability;
4244 struct user_action_cookie cookie;
4245 size_t sample_offset, actions_offset;
4246 int cookie_offset, n_output;
4248 if (!ofproto->sflow || flow->in_port == OFPP_NONE) {
4252 if (odp_port == OVSP_NONE) {
4256 port_ifindex = dpif_sflow_odp_port_to_ifindex(ofproto->sflow, odp_port);
4260 sample_offset = nl_msg_start_nested(odp_actions, OVS_ACTION_ATTR_SAMPLE);
4262 /* Number of packets out of UINT_MAX to sample. */
4263 probability = dpif_sflow_get_probability(ofproto->sflow);
4264 nl_msg_put_u32(odp_actions, OVS_SAMPLE_ATTR_PROBABILITY, probability);
4266 actions_offset = nl_msg_start_nested(odp_actions, OVS_SAMPLE_ATTR_ACTIONS);
4268 cookie.type = USER_ACTION_COOKIE_SFLOW;
4269 cookie.data = port_ifindex;
4270 cookie.n_output = n_output;
4271 cookie.vlan_tci = 0;
4272 cookie_offset = put_userspace_action(ofproto, odp_actions, flow, &cookie);
4274 nl_msg_end_nested(odp_actions, actions_offset);
4275 nl_msg_end_nested(odp_actions, sample_offset);
4276 return cookie_offset;
4279 /* SAMPLE action must be first action in any given list of actions.
4280 * At this point we do not have all information required to build it. So try to
4281 * build sample action as complete as possible. */
4283 add_sflow_action(struct action_xlate_ctx *ctx)
4285 ctx->user_cookie_offset = compose_sflow_action(ctx->ofproto,
4287 &ctx->flow, OVSP_NONE);
4288 ctx->sflow_odp_port = 0;
4289 ctx->sflow_n_outputs = 0;
4292 /* Fix SAMPLE action according to data collected while composing ODP actions.
4293 * We need to fix SAMPLE actions OVS_SAMPLE_ATTR_ACTIONS attribute, i.e. nested
4294 * USERSPACE action's user-cookie which is required for sflow. */
4296 fix_sflow_action(struct action_xlate_ctx *ctx)
4298 const struct flow *base = &ctx->base_flow;
4299 struct user_action_cookie *cookie;
4301 if (!ctx->user_cookie_offset) {
4305 cookie = ofpbuf_at(ctx->odp_actions, ctx->user_cookie_offset,
4307 assert(cookie != NULL);
4308 assert(cookie->type == USER_ACTION_COOKIE_SFLOW);
4310 if (ctx->sflow_n_outputs) {
4311 cookie->data = dpif_sflow_odp_port_to_ifindex(ctx->ofproto->sflow,
4312 ctx->sflow_odp_port);
4314 if (ctx->sflow_n_outputs >= 255) {
4315 cookie->n_output = 255;
4317 cookie->n_output = ctx->sflow_n_outputs;
4319 cookie->vlan_tci = base->vlan_tci;
4323 compose_output_action__(struct action_xlate_ctx *ctx, uint16_t ofp_port,
4326 const struct ofport_dpif *ofport = get_ofp_port(ctx->ofproto, ofp_port);
4327 uint16_t odp_port = ofp_port_to_odp_port(ofp_port);
4328 ovs_be16 flow_vlan_tci = ctx->flow.vlan_tci;
4329 uint8_t flow_nw_tos = ctx->flow.nw_tos;
4333 struct priority_to_dscp *pdscp;
4335 if (ofport->up.opp.config & htonl(OFPPC_NO_FWD)
4336 || (check_stp && !stp_forward_in_state(ofport->stp_state))) {
4340 pdscp = get_priority(ofport, ctx->flow.skb_priority);
4342 ctx->flow.nw_tos &= ~IP_DSCP_MASK;
4343 ctx->flow.nw_tos |= pdscp->dscp;
4346 /* We may not have an ofport record for this port, but it doesn't hurt
4347 * to allow forwarding to it anyhow. Maybe such a port will appear
4348 * later and we're pre-populating the flow table. */
4351 out_port = vsp_realdev_to_vlandev(ctx->ofproto, odp_port,
4352 ctx->flow.vlan_tci);
4353 if (out_port != odp_port) {
4354 ctx->flow.vlan_tci = htons(0);
4356 commit_odp_actions(&ctx->flow, &ctx->base_flow, ctx->odp_actions);
4357 nl_msg_put_u32(ctx->odp_actions, OVS_ACTION_ATTR_OUTPUT, out_port);
4359 ctx->sflow_odp_port = odp_port;
4360 ctx->sflow_n_outputs++;
4361 ctx->nf_output_iface = ofp_port;
4362 ctx->flow.vlan_tci = flow_vlan_tci;
4363 ctx->flow.nw_tos = flow_nw_tos;
4367 compose_output_action(struct action_xlate_ctx *ctx, uint16_t ofp_port)
4369 compose_output_action__(ctx, ofp_port, true);
4373 xlate_table_action(struct action_xlate_ctx *ctx,
4374 uint16_t in_port, uint8_t table_id)
4376 if (ctx->recurse < MAX_RESUBMIT_RECURSION) {
4377 struct ofproto_dpif *ofproto = ctx->ofproto;
4378 struct rule_dpif *rule;
4379 uint16_t old_in_port;
4380 uint8_t old_table_id;
4382 old_table_id = ctx->table_id;
4383 ctx->table_id = table_id;
4385 /* Look up a flow with 'in_port' as the input port. */
4386 old_in_port = ctx->flow.in_port;
4387 ctx->flow.in_port = in_port;
4388 rule = rule_dpif_lookup(ofproto, &ctx->flow, table_id);
4391 if (table_id > 0 && table_id < N_TABLES) {
4392 struct table_dpif *table = &ofproto->tables[table_id];
4393 if (table->other_table) {
4396 : rule_calculate_tag(&ctx->flow,
4397 &table->other_table->wc,
4402 /* Restore the original input port. Otherwise OFPP_NORMAL and
4403 * OFPP_IN_PORT will have surprising behavior. */
4404 ctx->flow.in_port = old_in_port;
4406 if (ctx->resubmit_hook) {
4407 ctx->resubmit_hook(ctx, rule);
4411 struct rule_dpif *old_rule = ctx->rule;
4415 do_xlate_actions(rule->up.actions, rule->up.n_actions, ctx);
4416 ctx->rule = old_rule;
4420 ctx->table_id = old_table_id;
4422 static struct vlog_rate_limit recurse_rl = VLOG_RATE_LIMIT_INIT(1, 1);
4424 VLOG_ERR_RL(&recurse_rl, "resubmit actions recursed over %d times",
4425 MAX_RESUBMIT_RECURSION);
4430 xlate_resubmit_table(struct action_xlate_ctx *ctx,
4431 const struct nx_action_resubmit *nar)
4436 in_port = (nar->in_port == htons(OFPP_IN_PORT)
4438 : ntohs(nar->in_port));
4439 table_id = nar->table == 255 ? ctx->table_id : nar->table;
4441 xlate_table_action(ctx, in_port, table_id);
4445 flood_packets(struct action_xlate_ctx *ctx, bool all)
4447 struct ofport_dpif *ofport;
4449 HMAP_FOR_EACH (ofport, up.hmap_node, &ctx->ofproto->up.ports) {
4450 uint16_t ofp_port = ofport->up.ofp_port;
4452 if (ofp_port == ctx->flow.in_port) {
4457 compose_output_action__(ctx, ofp_port, false);
4458 } else if (!(ofport->up.opp.config & htonl(OFPPC_NO_FLOOD))) {
4459 compose_output_action(ctx, ofp_port);
4463 ctx->nf_output_iface = NF_OUT_FLOOD;
4467 execute_controller_action(struct action_xlate_ctx *ctx, int len,
4468 enum ofp_packet_in_reason reason)
4470 struct ofputil_packet_in pin;
4471 struct ofpbuf *packet;
4473 ctx->may_set_up_flow = false;
4478 packet = ofpbuf_clone(ctx->packet);
4480 if (packet->l2 && packet->l3) {
4481 struct eth_header *eh;
4483 eth_pop_vlan(packet);
4485 assert(eh->eth_type == ctx->flow.dl_type);
4486 memcpy(eh->eth_src, ctx->flow.dl_src, sizeof eh->eth_src);
4487 memcpy(eh->eth_dst, ctx->flow.dl_dst, sizeof eh->eth_dst);
4489 if (ctx->flow.vlan_tci & htons(VLAN_CFI)) {
4490 eth_push_vlan(packet, ctx->flow.vlan_tci);
4494 if (ctx->flow.dl_type == htons(ETH_TYPE_IP)) {
4495 packet_set_ipv4(packet, ctx->flow.nw_src, ctx->flow.nw_dst,
4496 ctx->flow.nw_tos, ctx->flow.nw_ttl);
4500 if (ctx->flow.nw_proto == IPPROTO_TCP) {
4501 packet_set_tcp_port(packet, ctx->flow.tp_src,
4503 } else if (ctx->flow.nw_proto == IPPROTO_UDP) {
4504 packet_set_udp_port(packet, ctx->flow.tp_src,
4511 pin.packet = packet->data;
4512 pin.packet_len = packet->size;
4513 pin.reason = reason;
4514 pin.table_id = ctx->table_id;
4515 pin.cookie = ctx->rule ? ctx->rule->up.flow_cookie : 0;
4519 pin.total_len = packet->size;
4520 flow_get_metadata(&ctx->flow, &pin.fmd);
4522 connmgr_send_packet_in(ctx->ofproto->up.connmgr, &pin, &ctx->flow);
4523 ofpbuf_delete(packet);
4527 compose_dec_ttl(struct action_xlate_ctx *ctx)
4529 if (ctx->flow.dl_type != htons(ETH_TYPE_IP) &&
4530 ctx->flow.dl_type != htons(ETH_TYPE_IPV6)) {
4534 if (ctx->flow.nw_ttl > 1) {
4538 execute_controller_action(ctx, UINT16_MAX, OFPR_INVALID_TTL);
4540 /* Stop processing for current table. */
4546 xlate_output_action__(struct action_xlate_ctx *ctx,
4547 uint16_t port, uint16_t max_len)
4549 uint16_t prev_nf_output_iface = ctx->nf_output_iface;
4551 ctx->nf_output_iface = NF_OUT_DROP;
4555 compose_output_action(ctx, ctx->flow.in_port);
4558 xlate_table_action(ctx, ctx->flow.in_port, ctx->table_id);
4564 flood_packets(ctx, false);
4567 flood_packets(ctx, true);
4569 case OFPP_CONTROLLER:
4570 execute_controller_action(ctx, max_len, OFPR_ACTION);
4576 if (port != ctx->flow.in_port) {
4577 compose_output_action(ctx, port);
4582 if (prev_nf_output_iface == NF_OUT_FLOOD) {
4583 ctx->nf_output_iface = NF_OUT_FLOOD;
4584 } else if (ctx->nf_output_iface == NF_OUT_DROP) {
4585 ctx->nf_output_iface = prev_nf_output_iface;
4586 } else if (prev_nf_output_iface != NF_OUT_DROP &&
4587 ctx->nf_output_iface != NF_OUT_FLOOD) {
4588 ctx->nf_output_iface = NF_OUT_MULTI;
4593 xlate_output_reg_action(struct action_xlate_ctx *ctx,
4594 const struct nx_action_output_reg *naor)
4596 struct mf_subfield src;
4599 nxm_decode(&src, naor->src, naor->ofs_nbits);
4600 ofp_port = mf_get_subfield(&src, &ctx->flow);
4602 if (ofp_port <= UINT16_MAX) {
4603 xlate_output_action__(ctx, ofp_port, ntohs(naor->max_len));
4608 xlate_output_action(struct action_xlate_ctx *ctx,
4609 const struct ofp_action_output *oao)
4611 xlate_output_action__(ctx, ntohs(oao->port), ntohs(oao->max_len));
4615 xlate_enqueue_action(struct action_xlate_ctx *ctx,
4616 const struct ofp_action_enqueue *oae)
4619 uint32_t flow_priority, priority;
4622 error = dpif_queue_to_priority(ctx->ofproto->dpif, ntohl(oae->queue_id),
4625 /* Fall back to ordinary output action. */
4626 xlate_output_action__(ctx, ntohs(oae->port), 0);
4630 /* Figure out datapath output port. */
4631 ofp_port = ntohs(oae->port);
4632 if (ofp_port == OFPP_IN_PORT) {
4633 ofp_port = ctx->flow.in_port;
4634 } else if (ofp_port == ctx->flow.in_port) {
4638 /* Add datapath actions. */
4639 flow_priority = ctx->flow.skb_priority;
4640 ctx->flow.skb_priority = priority;
4641 compose_output_action(ctx, ofp_port);
4642 ctx->flow.skb_priority = flow_priority;
4644 /* Update NetFlow output port. */
4645 if (ctx->nf_output_iface == NF_OUT_DROP) {
4646 ctx->nf_output_iface = ofp_port;
4647 } else if (ctx->nf_output_iface != NF_OUT_FLOOD) {
4648 ctx->nf_output_iface = NF_OUT_MULTI;
4653 xlate_set_queue_action(struct action_xlate_ctx *ctx,
4654 const struct nx_action_set_queue *nasq)
4659 error = dpif_queue_to_priority(ctx->ofproto->dpif, ntohl(nasq->queue_id),
4662 /* Couldn't translate queue to a priority, so ignore. A warning
4663 * has already been logged. */
4667 ctx->flow.skb_priority = priority;
4670 struct xlate_reg_state {
4676 xlate_autopath(struct action_xlate_ctx *ctx,
4677 const struct nx_action_autopath *naa)
4679 uint16_t ofp_port = ntohl(naa->id);
4680 struct ofport_dpif *port = get_ofp_port(ctx->ofproto, ofp_port);
4682 if (!port || !port->bundle) {
4683 ofp_port = OFPP_NONE;
4684 } else if (port->bundle->bond) {
4685 /* Autopath does not support VLAN hashing. */
4686 struct ofport_dpif *slave = bond_choose_output_slave(
4687 port->bundle->bond, &ctx->flow, 0, &ctx->tags);
4689 ofp_port = slave->up.ofp_port;
4692 autopath_execute(naa, &ctx->flow, ofp_port);
4696 slave_enabled_cb(uint16_t ofp_port, void *ofproto_)
4698 struct ofproto_dpif *ofproto = ofproto_;
4699 struct ofport_dpif *port;
4709 case OFPP_CONTROLLER: /* Not supported by the bundle action. */
4712 port = get_ofp_port(ofproto, ofp_port);
4713 return port ? port->may_enable : false;
4718 xlate_learn_action(struct action_xlate_ctx *ctx,
4719 const struct nx_action_learn *learn)
4721 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(5, 1);
4722 struct ofputil_flow_mod fm;
4725 learn_execute(learn, &ctx->flow, &fm);
4727 error = ofproto_flow_mod(&ctx->ofproto->up, &fm);
4728 if (error && !VLOG_DROP_WARN(&rl)) {
4729 VLOG_WARN("learning action failed to modify flow table (%s)",
4730 ofperr_get_name(error));
4736 /* Reduces '*timeout' to no more than 'max'. A value of zero in either case
4737 * means "infinite". */
4739 reduce_timeout(uint16_t max, uint16_t *timeout)
4741 if (max && (!*timeout || *timeout > max)) {
4747 xlate_fin_timeout(struct action_xlate_ctx *ctx,
4748 const struct nx_action_fin_timeout *naft)
4750 if (ctx->tcp_flags & (TCP_FIN | TCP_RST) && ctx->rule) {
4751 struct rule_dpif *rule = ctx->rule;
4753 reduce_timeout(ntohs(naft->fin_idle_timeout), &rule->up.idle_timeout);
4754 reduce_timeout(ntohs(naft->fin_hard_timeout), &rule->up.hard_timeout);
4759 may_receive(const struct ofport_dpif *port, struct action_xlate_ctx *ctx)
4761 if (port->up.opp.config & (eth_addr_equals(ctx->flow.dl_dst, eth_addr_stp)
4762 ? htonl(OFPPC_NO_RECV_STP)
4763 : htonl(OFPPC_NO_RECV))) {
4767 /* Only drop packets here if both forwarding and learning are
4768 * disabled. If just learning is enabled, we need to have
4769 * OFPP_NORMAL and the learning action have a look at the packet
4770 * before we can drop it. */
4771 if (!stp_forward_in_state(port->stp_state)
4772 && !stp_learn_in_state(port->stp_state)) {
4780 do_xlate_actions(const union ofp_action *in, size_t n_in,
4781 struct action_xlate_ctx *ctx)
4783 const struct ofport_dpif *port;
4784 const union ofp_action *ia;
4785 bool was_evictable = true;
4788 port = get_ofp_port(ctx->ofproto, ctx->flow.in_port);
4789 if (port && !may_receive(port, ctx)) {
4790 /* Drop this flow. */
4795 /* Don't let the rule we're working on get evicted underneath us. */
4796 was_evictable = ctx->rule->up.evictable;
4797 ctx->rule->up.evictable = false;
4799 OFPUTIL_ACTION_FOR_EACH_UNSAFE (ia, left, in, n_in) {
4800 const struct ofp_action_dl_addr *oada;
4801 const struct nx_action_resubmit *nar;
4802 const struct nx_action_set_tunnel *nast;
4803 const struct nx_action_set_queue *nasq;
4804 const struct nx_action_multipath *nam;
4805 const struct nx_action_autopath *naa;
4806 const struct nx_action_bundle *nab;
4807 const struct nx_action_output_reg *naor;
4808 enum ofputil_action_code code;
4815 code = ofputil_decode_action_unsafe(ia);
4817 case OFPUTIL_OFPAT_OUTPUT:
4818 xlate_output_action(ctx, &ia->output);
4821 case OFPUTIL_OFPAT_SET_VLAN_VID:
4822 ctx->flow.vlan_tci &= ~htons(VLAN_VID_MASK);
4823 ctx->flow.vlan_tci |= ia->vlan_vid.vlan_vid | htons(VLAN_CFI);
4826 case OFPUTIL_OFPAT_SET_VLAN_PCP:
4827 ctx->flow.vlan_tci &= ~htons(VLAN_PCP_MASK);
4828 ctx->flow.vlan_tci |= htons(
4829 (ia->vlan_pcp.vlan_pcp << VLAN_PCP_SHIFT) | VLAN_CFI);
4832 case OFPUTIL_OFPAT_STRIP_VLAN:
4833 ctx->flow.vlan_tci = htons(0);
4836 case OFPUTIL_OFPAT_SET_DL_SRC:
4837 oada = ((struct ofp_action_dl_addr *) ia);
4838 memcpy(ctx->flow.dl_src, oada->dl_addr, ETH_ADDR_LEN);
4841 case OFPUTIL_OFPAT_SET_DL_DST:
4842 oada = ((struct ofp_action_dl_addr *) ia);
4843 memcpy(ctx->flow.dl_dst, oada->dl_addr, ETH_ADDR_LEN);
4846 case OFPUTIL_OFPAT_SET_NW_SRC:
4847 ctx->flow.nw_src = ia->nw_addr.nw_addr;
4850 case OFPUTIL_OFPAT_SET_NW_DST:
4851 ctx->flow.nw_dst = ia->nw_addr.nw_addr;
4854 case OFPUTIL_OFPAT_SET_NW_TOS:
4855 /* OpenFlow 1.0 only supports IPv4. */
4856 if (ctx->flow.dl_type == htons(ETH_TYPE_IP)) {
4857 ctx->flow.nw_tos &= ~IP_DSCP_MASK;
4858 ctx->flow.nw_tos |= ia->nw_tos.nw_tos & IP_DSCP_MASK;
4862 case OFPUTIL_OFPAT_SET_TP_SRC:
4863 ctx->flow.tp_src = ia->tp_port.tp_port;
4866 case OFPUTIL_OFPAT_SET_TP_DST:
4867 ctx->flow.tp_dst = ia->tp_port.tp_port;
4870 case OFPUTIL_OFPAT_ENQUEUE:
4871 xlate_enqueue_action(ctx, (const struct ofp_action_enqueue *) ia);
4874 case OFPUTIL_NXAST_RESUBMIT:
4875 nar = (const struct nx_action_resubmit *) ia;
4876 xlate_table_action(ctx, ntohs(nar->in_port), ctx->table_id);
4879 case OFPUTIL_NXAST_RESUBMIT_TABLE:
4880 xlate_resubmit_table(ctx, (const struct nx_action_resubmit *) ia);
4883 case OFPUTIL_NXAST_SET_TUNNEL:
4884 nast = (const struct nx_action_set_tunnel *) ia;
4885 tun_id = htonll(ntohl(nast->tun_id));
4886 ctx->flow.tun_id = tun_id;
4889 case OFPUTIL_NXAST_SET_QUEUE:
4890 nasq = (const struct nx_action_set_queue *) ia;
4891 xlate_set_queue_action(ctx, nasq);
4894 case OFPUTIL_NXAST_POP_QUEUE:
4895 ctx->flow.skb_priority = ctx->orig_skb_priority;
4898 case OFPUTIL_NXAST_REG_MOVE:
4899 nxm_execute_reg_move((const struct nx_action_reg_move *) ia,
4903 case OFPUTIL_NXAST_REG_LOAD:
4904 nxm_execute_reg_load((const struct nx_action_reg_load *) ia,
4908 case OFPUTIL_NXAST_NOTE:
4909 /* Nothing to do. */
4912 case OFPUTIL_NXAST_SET_TUNNEL64:
4913 tun_id = ((const struct nx_action_set_tunnel64 *) ia)->tun_id;
4914 ctx->flow.tun_id = tun_id;
4917 case OFPUTIL_NXAST_MULTIPATH:
4918 nam = (const struct nx_action_multipath *) ia;
4919 multipath_execute(nam, &ctx->flow);
4922 case OFPUTIL_NXAST_AUTOPATH:
4923 naa = (const struct nx_action_autopath *) ia;
4924 xlate_autopath(ctx, naa);
4927 case OFPUTIL_NXAST_BUNDLE:
4928 ctx->ofproto->has_bundle_action = true;
4929 nab = (const struct nx_action_bundle *) ia;
4930 xlate_output_action__(ctx, bundle_execute(nab, &ctx->flow,
4935 case OFPUTIL_NXAST_BUNDLE_LOAD:
4936 ctx->ofproto->has_bundle_action = true;
4937 nab = (const struct nx_action_bundle *) ia;
4938 bundle_execute_load(nab, &ctx->flow, slave_enabled_cb,
4942 case OFPUTIL_NXAST_OUTPUT_REG:
4943 naor = (const struct nx_action_output_reg *) ia;
4944 xlate_output_reg_action(ctx, naor);
4947 case OFPUTIL_NXAST_LEARN:
4948 ctx->has_learn = true;
4949 if (ctx->may_learn) {
4950 xlate_learn_action(ctx, (const struct nx_action_learn *) ia);
4954 case OFPUTIL_NXAST_DEC_TTL:
4955 if (compose_dec_ttl(ctx)) {
4960 case OFPUTIL_NXAST_EXIT:
4964 case OFPUTIL_NXAST_FIN_TIMEOUT:
4965 ctx->has_fin_timeout = true;
4966 xlate_fin_timeout(ctx, (const struct nx_action_fin_timeout *) ia);
4972 /* We've let OFPP_NORMAL and the learning action look at the packet,
4973 * so drop it now if forwarding is disabled. */
4974 if (port && !stp_forward_in_state(port->stp_state)) {
4975 ofpbuf_clear(ctx->odp_actions);
4976 add_sflow_action(ctx);
4979 ctx->rule->up.evictable = was_evictable;
4984 action_xlate_ctx_init(struct action_xlate_ctx *ctx,
4985 struct ofproto_dpif *ofproto, const struct flow *flow,
4986 ovs_be16 initial_tci, struct rule_dpif *rule,
4987 uint8_t tcp_flags, const struct ofpbuf *packet)
4989 ctx->ofproto = ofproto;
4991 ctx->base_flow = ctx->flow;
4992 ctx->base_flow.tun_id = 0;
4993 ctx->base_flow.vlan_tci = initial_tci;
4995 ctx->packet = packet;
4996 ctx->may_learn = packet != NULL;
4997 ctx->tcp_flags = tcp_flags;
4998 ctx->resubmit_hook = NULL;
5001 static struct ofpbuf *
5002 xlate_actions(struct action_xlate_ctx *ctx,
5003 const union ofp_action *in, size_t n_in)
5005 struct flow orig_flow = ctx->flow;
5007 COVERAGE_INC(ofproto_dpif_xlate);
5009 ctx->odp_actions = ofpbuf_new(512);
5010 ofpbuf_reserve(ctx->odp_actions, NL_A_U32_SIZE);
5012 ctx->may_set_up_flow = true;
5013 ctx->has_learn = false;
5014 ctx->has_normal = false;
5015 ctx->has_fin_timeout = false;
5016 ctx->nf_output_iface = NF_OUT_DROP;
5019 ctx->orig_skb_priority = ctx->flow.skb_priority;
5023 if (ctx->flow.nw_frag & FLOW_NW_FRAG_ANY) {
5024 switch (ctx->ofproto->up.frag_handling) {
5025 case OFPC_FRAG_NORMAL:
5026 /* We must pretend that transport ports are unavailable. */
5027 ctx->flow.tp_src = ctx->base_flow.tp_src = htons(0);
5028 ctx->flow.tp_dst = ctx->base_flow.tp_dst = htons(0);
5031 case OFPC_FRAG_DROP:
5032 return ctx->odp_actions;
5034 case OFPC_FRAG_REASM:
5037 case OFPC_FRAG_NX_MATCH:
5038 /* Nothing to do. */
5041 case OFPC_INVALID_TTL_TO_CONTROLLER:
5046 if (process_special(ctx->ofproto, &ctx->flow, ctx->packet)) {
5047 ctx->may_set_up_flow = false;
5048 return ctx->odp_actions;
5050 add_sflow_action(ctx);
5051 do_xlate_actions(in, n_in, ctx);
5053 if (!connmgr_may_set_up_flow(ctx->ofproto->up.connmgr, &ctx->flow,
5054 ctx->odp_actions->data,
5055 ctx->odp_actions->size)) {
5056 ctx->may_set_up_flow = false;
5058 && connmgr_msg_in_hook(ctx->ofproto->up.connmgr, &ctx->flow,
5060 compose_output_action(ctx, OFPP_LOCAL);
5063 add_mirror_actions(ctx, &orig_flow);
5064 fix_sflow_action(ctx);
5067 return ctx->odp_actions;
5070 /* OFPP_NORMAL implementation. */
5072 static struct ofport_dpif *ofbundle_get_a_port(const struct ofbundle *);
5074 /* Given 'vid', the VID obtained from the 802.1Q header that was received as
5075 * part of a packet (specify 0 if there was no 802.1Q header), and 'in_bundle',
5076 * the bundle on which the packet was received, returns the VLAN to which the
5079 * Both 'vid' and the return value are in the range 0...4095. */
5081 input_vid_to_vlan(const struct ofbundle *in_bundle, uint16_t vid)
5083 switch (in_bundle->vlan_mode) {
5084 case PORT_VLAN_ACCESS:
5085 return in_bundle->vlan;
5088 case PORT_VLAN_TRUNK:
5091 case PORT_VLAN_NATIVE_UNTAGGED:
5092 case PORT_VLAN_NATIVE_TAGGED:
5093 return vid ? vid : in_bundle->vlan;
5100 /* Checks whether a packet with the given 'vid' may ingress on 'in_bundle'.
5101 * If so, returns true. Otherwise, returns false and, if 'warn' is true, logs
5104 * 'vid' should be the VID obtained from the 802.1Q header that was received as
5105 * part of a packet (specify 0 if there was no 802.1Q header), in the range
5108 input_vid_is_valid(uint16_t vid, struct ofbundle *in_bundle, bool warn)
5110 /* Allow any VID on the OFPP_NONE port. */
5111 if (in_bundle == &ofpp_none_bundle) {
5115 switch (in_bundle->vlan_mode) {
5116 case PORT_VLAN_ACCESS:
5119 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
5120 VLOG_WARN_RL(&rl, "bridge %s: dropping VLAN %"PRIu16" tagged "
5121 "packet received on port %s configured as VLAN "
5122 "%"PRIu16" access port",
5123 in_bundle->ofproto->up.name, vid,
5124 in_bundle->name, in_bundle->vlan);
5130 case PORT_VLAN_NATIVE_UNTAGGED:
5131 case PORT_VLAN_NATIVE_TAGGED:
5133 /* Port must always carry its native VLAN. */
5137 case PORT_VLAN_TRUNK:
5138 if (!ofbundle_includes_vlan(in_bundle, vid)) {
5140 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
5141 VLOG_WARN_RL(&rl, "bridge %s: dropping VLAN %"PRIu16" packet "
5142 "received on port %s not configured for trunking "
5144 in_bundle->ofproto->up.name, vid,
5145 in_bundle->name, vid);
5157 /* Given 'vlan', the VLAN that a packet belongs to, and
5158 * 'out_bundle', a bundle on which the packet is to be output, returns the VID
5159 * that should be included in the 802.1Q header. (If the return value is 0,
5160 * then the 802.1Q header should only be included in the packet if there is a
5163 * Both 'vlan' and the return value are in the range 0...4095. */
5165 output_vlan_to_vid(const struct ofbundle *out_bundle, uint16_t vlan)
5167 switch (out_bundle->vlan_mode) {
5168 case PORT_VLAN_ACCESS:
5171 case PORT_VLAN_TRUNK:
5172 case PORT_VLAN_NATIVE_TAGGED:
5175 case PORT_VLAN_NATIVE_UNTAGGED:
5176 return vlan == out_bundle->vlan ? 0 : vlan;
5184 output_normal(struct action_xlate_ctx *ctx, const struct ofbundle *out_bundle,
5187 struct ofport_dpif *port;
5189 ovs_be16 tci, old_tci;
5191 vid = output_vlan_to_vid(out_bundle, vlan);
5192 if (!out_bundle->bond) {
5193 port = ofbundle_get_a_port(out_bundle);
5195 port = bond_choose_output_slave(out_bundle->bond, &ctx->flow,
5198 /* No slaves enabled, so drop packet. */
5203 old_tci = ctx->flow.vlan_tci;
5205 if (tci || out_bundle->use_priority_tags) {
5206 tci |= ctx->flow.vlan_tci & htons(VLAN_PCP_MASK);
5208 tci |= htons(VLAN_CFI);
5211 ctx->flow.vlan_tci = tci;
5213 compose_output_action(ctx, port->up.ofp_port);
5214 ctx->flow.vlan_tci = old_tci;
5218 mirror_mask_ffs(mirror_mask_t mask)
5220 BUILD_ASSERT_DECL(sizeof(unsigned int) >= sizeof(mask));
5225 ofbundle_trunks_vlan(const struct ofbundle *bundle, uint16_t vlan)
5227 return (bundle->vlan_mode != PORT_VLAN_ACCESS
5228 && (!bundle->trunks || bitmap_is_set(bundle->trunks, vlan)));
5232 ofbundle_includes_vlan(const struct ofbundle *bundle, uint16_t vlan)
5234 return vlan == bundle->vlan || ofbundle_trunks_vlan(bundle, vlan);
5237 /* Returns an arbitrary interface within 'bundle'. */
5238 static struct ofport_dpif *
5239 ofbundle_get_a_port(const struct ofbundle *bundle)
5241 return CONTAINER_OF(list_front(&bundle->ports),
5242 struct ofport_dpif, bundle_node);
5246 vlan_is_mirrored(const struct ofmirror *m, int vlan)
5248 return !m->vlans || bitmap_is_set(m->vlans, vlan);
5251 /* Returns true if a packet with Ethernet destination MAC 'dst' may be mirrored
5252 * to a VLAN. In general most packets may be mirrored but we want to drop
5253 * protocols that may confuse switches. */
5255 eth_dst_may_rspan(const uint8_t dst[ETH_ADDR_LEN])
5257 /* If you change this function's behavior, please update corresponding
5258 * documentation in vswitch.xml at the same time. */
5259 if (dst[0] != 0x01) {
5260 /* All the currently banned MACs happen to start with 01 currently, so
5261 * this is a quick way to eliminate most of the good ones. */
5263 if (eth_addr_is_reserved(dst)) {
5264 /* Drop STP, IEEE pause frames, and other reserved protocols
5265 * (01-80-c2-00-00-0x). */
5269 if (dst[0] == 0x01 && dst[1] == 0x00 && dst[2] == 0x0c) {
5271 if ((dst[3] & 0xfe) == 0xcc &&
5272 (dst[4] & 0xfe) == 0xcc &&
5273 (dst[5] & 0xfe) == 0xcc) {
5274 /* Drop the following protocols plus others following the same
5277 CDP, VTP, DTP, PAgP (01-00-0c-cc-cc-cc)
5278 Spanning Tree PVSTP+ (01-00-0c-cc-cc-cd)
5279 STP Uplink Fast (01-00-0c-cd-cd-cd) */
5283 if (!(dst[3] | dst[4] | dst[5])) {
5284 /* Drop Inter Switch Link packets (01-00-0c-00-00-00). */
5293 add_mirror_actions(struct action_xlate_ctx *ctx, const struct flow *orig_flow)
5295 struct ofproto_dpif *ofproto = ctx->ofproto;
5296 mirror_mask_t mirrors;
5297 struct ofbundle *in_bundle;
5300 const struct nlattr *a;
5303 in_bundle = lookup_input_bundle(ctx->ofproto, orig_flow->in_port,
5304 ctx->packet != NULL);
5308 mirrors = in_bundle->src_mirrors;
5310 /* Drop frames on bundles reserved for mirroring. */
5311 if (in_bundle->mirror_out) {
5312 if (ctx->packet != NULL) {
5313 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
5314 VLOG_WARN_RL(&rl, "bridge %s: dropping packet received on port "
5315 "%s, which is reserved exclusively for mirroring",
5316 ctx->ofproto->up.name, in_bundle->name);
5322 vid = vlan_tci_to_vid(orig_flow->vlan_tci);
5323 if (!input_vid_is_valid(vid, in_bundle, ctx->packet != NULL)) {
5326 vlan = input_vid_to_vlan(in_bundle, vid);
5328 /* Look at the output ports to check for destination selections. */
5330 NL_ATTR_FOR_EACH (a, left, ctx->odp_actions->data,
5331 ctx->odp_actions->size) {
5332 enum ovs_action_attr type = nl_attr_type(a);
5333 struct ofport_dpif *ofport;
5335 if (type != OVS_ACTION_ATTR_OUTPUT) {
5339 ofport = get_odp_port(ofproto, nl_attr_get_u32(a));
5340 if (ofport && ofport->bundle) {
5341 mirrors |= ofport->bundle->dst_mirrors;
5349 /* Restore the original packet before adding the mirror actions. */
5350 ctx->flow = *orig_flow;
5355 m = ofproto->mirrors[mirror_mask_ffs(mirrors) - 1];
5357 if (!vlan_is_mirrored(m, vlan)) {
5358 mirrors &= mirrors - 1;
5362 mirrors &= ~m->dup_mirrors;
5363 ctx->mirrors |= m->dup_mirrors;
5365 output_normal(ctx, m->out, vlan);
5366 } else if (eth_dst_may_rspan(orig_flow->dl_dst)
5367 && vlan != m->out_vlan) {
5368 struct ofbundle *bundle;
5370 HMAP_FOR_EACH (bundle, hmap_node, &ofproto->bundles) {
5371 if (ofbundle_includes_vlan(bundle, m->out_vlan)
5372 && !bundle->mirror_out) {
5373 output_normal(ctx, bundle, m->out_vlan);
5381 update_mirror_stats(struct ofproto_dpif *ofproto, mirror_mask_t mirrors,
5382 uint64_t packets, uint64_t bytes)
5388 for (; mirrors; mirrors &= mirrors - 1) {
5391 m = ofproto->mirrors[mirror_mask_ffs(mirrors) - 1];
5394 /* In normal circumstances 'm' will not be NULL. However,
5395 * if mirrors are reconfigured, we can temporarily get out
5396 * of sync in facet_revalidate(). We could "correct" the
5397 * mirror list before reaching here, but doing that would
5398 * not properly account the traffic stats we've currently
5399 * accumulated for previous mirror configuration. */
5403 m->packet_count += packets;
5404 m->byte_count += bytes;
5408 /* A VM broadcasts a gratuitous ARP to indicate that it has resumed after
5409 * migration. Older Citrix-patched Linux DomU used gratuitous ARP replies to
5410 * indicate this; newer upstream kernels use gratuitous ARP requests. */
5412 is_gratuitous_arp(const struct flow *flow)
5414 return (flow->dl_type == htons(ETH_TYPE_ARP)
5415 && eth_addr_is_broadcast(flow->dl_dst)
5416 && (flow->nw_proto == ARP_OP_REPLY
5417 || (flow->nw_proto == ARP_OP_REQUEST
5418 && flow->nw_src == flow->nw_dst)));
5422 update_learning_table(struct ofproto_dpif *ofproto,
5423 const struct flow *flow, int vlan,
5424 struct ofbundle *in_bundle)
5426 struct mac_entry *mac;
5428 /* Don't learn the OFPP_NONE port. */
5429 if (in_bundle == &ofpp_none_bundle) {
5433 if (!mac_learning_may_learn(ofproto->ml, flow->dl_src, vlan)) {
5437 mac = mac_learning_insert(ofproto->ml, flow->dl_src, vlan);
5438 if (is_gratuitous_arp(flow)) {
5439 /* We don't want to learn from gratuitous ARP packets that are
5440 * reflected back over bond slaves so we lock the learning table. */
5441 if (!in_bundle->bond) {
5442 mac_entry_set_grat_arp_lock(mac);
5443 } else if (mac_entry_is_grat_arp_locked(mac)) {
5448 if (mac_entry_is_new(mac) || mac->port.p != in_bundle) {
5449 /* The log messages here could actually be useful in debugging,
5450 * so keep the rate limit relatively high. */
5451 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(30, 300);
5452 VLOG_DBG_RL(&rl, "bridge %s: learned that "ETH_ADDR_FMT" is "
5453 "on port %s in VLAN %d",
5454 ofproto->up.name, ETH_ADDR_ARGS(flow->dl_src),
5455 in_bundle->name, vlan);
5457 mac->port.p = in_bundle;
5458 tag_set_add(&ofproto->revalidate_set,
5459 mac_learning_changed(ofproto->ml, mac));
5463 static struct ofbundle *
5464 lookup_input_bundle(struct ofproto_dpif *ofproto, uint16_t in_port, bool warn)
5466 struct ofport_dpif *ofport;
5468 /* Special-case OFPP_NONE, which a controller may use as the ingress
5469 * port for traffic that it is sourcing. */
5470 if (in_port == OFPP_NONE) {
5471 return &ofpp_none_bundle;
5474 /* Find the port and bundle for the received packet. */
5475 ofport = get_ofp_port(ofproto, in_port);
5476 if (ofport && ofport->bundle) {
5477 return ofport->bundle;
5480 /* Odd. A few possible reasons here:
5482 * - We deleted a port but there are still a few packets queued up
5485 * - Someone externally added a port (e.g. "ovs-dpctl add-if") that
5486 * we don't know about.
5488 * - The ofproto client didn't configure the port as part of a bundle.
5491 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
5493 VLOG_WARN_RL(&rl, "bridge %s: received packet on unknown "
5494 "port %"PRIu16, ofproto->up.name, in_port);
5499 /* Determines whether packets in 'flow' within 'ofproto' should be forwarded or
5500 * dropped. Returns true if they may be forwarded, false if they should be
5503 * 'in_port' must be the ofport_dpif that corresponds to flow->in_port.
5504 * 'in_port' must be part of a bundle (e.g. in_port->bundle must be nonnull).
5506 * 'vlan' must be the VLAN that corresponds to flow->vlan_tci on 'in_port', as
5507 * returned by input_vid_to_vlan(). It must be a valid VLAN for 'in_port', as
5508 * checked by input_vid_is_valid().
5510 * May also add tags to '*tags', although the current implementation only does
5511 * so in one special case.
5514 is_admissible(struct ofproto_dpif *ofproto, const struct flow *flow,
5515 struct ofport_dpif *in_port, uint16_t vlan, tag_type *tags)
5517 struct ofbundle *in_bundle = in_port->bundle;
5519 /* Drop frames for reserved multicast addresses
5520 * only if forward_bpdu option is absent. */
5521 if (eth_addr_is_reserved(flow->dl_dst) && !ofproto->up.forward_bpdu) {
5525 if (in_bundle->bond) {
5526 struct mac_entry *mac;
5528 switch (bond_check_admissibility(in_bundle->bond, in_port,
5529 flow->dl_dst, tags)) {
5536 case BV_DROP_IF_MOVED:
5537 mac = mac_learning_lookup(ofproto->ml, flow->dl_src, vlan, NULL);
5538 if (mac && mac->port.p != in_bundle &&
5539 (!is_gratuitous_arp(flow)
5540 || mac_entry_is_grat_arp_locked(mac))) {
5551 xlate_normal(struct action_xlate_ctx *ctx)
5553 struct ofport_dpif *in_port;
5554 struct ofbundle *in_bundle;
5555 struct mac_entry *mac;
5559 ctx->has_normal = true;
5561 in_bundle = lookup_input_bundle(ctx->ofproto, ctx->flow.in_port,
5562 ctx->packet != NULL);
5567 /* We know 'in_port' exists unless it is "ofpp_none_bundle",
5568 * since lookup_input_bundle() succeeded. */
5569 in_port = get_ofp_port(ctx->ofproto, ctx->flow.in_port);
5571 /* Drop malformed frames. */
5572 if (ctx->flow.dl_type == htons(ETH_TYPE_VLAN) &&
5573 !(ctx->flow.vlan_tci & htons(VLAN_CFI))) {
5574 if (ctx->packet != NULL) {
5575 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
5576 VLOG_WARN_RL(&rl, "bridge %s: dropping packet with partial "
5577 "VLAN tag received on port %s",
5578 ctx->ofproto->up.name, in_bundle->name);
5583 /* Drop frames on bundles reserved for mirroring. */
5584 if (in_bundle->mirror_out) {
5585 if (ctx->packet != NULL) {
5586 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
5587 VLOG_WARN_RL(&rl, "bridge %s: dropping packet received on port "
5588 "%s, which is reserved exclusively for mirroring",
5589 ctx->ofproto->up.name, in_bundle->name);
5595 vid = vlan_tci_to_vid(ctx->flow.vlan_tci);
5596 if (!input_vid_is_valid(vid, in_bundle, ctx->packet != NULL)) {
5599 vlan = input_vid_to_vlan(in_bundle, vid);
5601 /* Check other admissibility requirements. */
5603 !is_admissible(ctx->ofproto, &ctx->flow, in_port, vlan, &ctx->tags)) {
5607 /* Learn source MAC. */
5608 if (ctx->may_learn) {
5609 update_learning_table(ctx->ofproto, &ctx->flow, vlan, in_bundle);
5612 /* Determine output bundle. */
5613 mac = mac_learning_lookup(ctx->ofproto->ml, ctx->flow.dl_dst, vlan,
5616 if (mac->port.p != in_bundle) {
5617 output_normal(ctx, mac->port.p, vlan);
5620 struct ofbundle *bundle;
5622 HMAP_FOR_EACH (bundle, hmap_node, &ctx->ofproto->bundles) {
5623 if (bundle != in_bundle
5624 && ofbundle_includes_vlan(bundle, vlan)
5625 && bundle->floodable
5626 && !bundle->mirror_out) {
5627 output_normal(ctx, bundle, vlan);
5630 ctx->nf_output_iface = NF_OUT_FLOOD;
5634 /* Optimized flow revalidation.
5636 * It's a difficult problem, in general, to tell which facets need to have
5637 * their actions recalculated whenever the OpenFlow flow table changes. We
5638 * don't try to solve that general problem: for most kinds of OpenFlow flow
5639 * table changes, we recalculate the actions for every facet. This is
5640 * relatively expensive, but it's good enough if the OpenFlow flow table
5641 * doesn't change very often.
5643 * However, we can expect one particular kind of OpenFlow flow table change to
5644 * happen frequently: changes caused by MAC learning. To avoid wasting a lot
5645 * of CPU on revalidating every facet whenever MAC learning modifies the flow
5646 * table, we add a special case that applies to flow tables in which every rule
5647 * has the same form (that is, the same wildcards), except that the table is
5648 * also allowed to have a single "catch-all" flow that matches all packets. We
5649 * optimize this case by tagging all of the facets that resubmit into the table
5650 * and invalidating the same tag whenever a flow changes in that table. The
5651 * end result is that we revalidate just the facets that need it (and sometimes
5652 * a few more, but not all of the facets or even all of the facets that
5653 * resubmit to the table modified by MAC learning). */
5655 /* Calculates the tag to use for 'flow' and wildcards 'wc' when it is inserted
5656 * into an OpenFlow table with the given 'basis'. */
5658 rule_calculate_tag(const struct flow *flow, const struct flow_wildcards *wc,
5661 if (flow_wildcards_is_catchall(wc)) {
5664 struct flow tag_flow = *flow;
5665 flow_zero_wildcards(&tag_flow, wc);
5666 return tag_create_deterministic(flow_hash(&tag_flow, secret));
5670 /* Following a change to OpenFlow table 'table_id' in 'ofproto', update the
5671 * taggability of that table.
5673 * This function must be called after *each* change to a flow table. If you
5674 * skip calling it on some changes then the pointer comparisons at the end can
5675 * be invalid if you get unlucky. For example, if a flow removal causes a
5676 * cls_table to be destroyed and then a flow insertion causes a cls_table with
5677 * different wildcards to be created with the same address, then this function
5678 * will incorrectly skip revalidation. */
5680 table_update_taggable(struct ofproto_dpif *ofproto, uint8_t table_id)
5682 struct table_dpif *table = &ofproto->tables[table_id];
5683 const struct oftable *oftable = &ofproto->up.tables[table_id];
5684 struct cls_table *catchall, *other;
5685 struct cls_table *t;
5687 catchall = other = NULL;
5689 switch (hmap_count(&oftable->cls.tables)) {
5691 /* We could tag this OpenFlow table but it would make the logic a
5692 * little harder and it's a corner case that doesn't seem worth it
5698 HMAP_FOR_EACH (t, hmap_node, &oftable->cls.tables) {
5699 if (cls_table_is_catchall(t)) {
5701 } else if (!other) {
5704 /* Indicate that we can't tag this by setting both tables to
5705 * NULL. (We know that 'catchall' is already NULL.) */
5712 /* Can't tag this table. */
5716 if (table->catchall_table != catchall || table->other_table != other) {
5717 table->catchall_table = catchall;
5718 table->other_table = other;
5719 ofproto->need_revalidate = true;
5723 /* Given 'rule' that has changed in some way (either it is a rule being
5724 * inserted, a rule being deleted, or a rule whose actions are being
5725 * modified), marks facets for revalidation to ensure that packets will be
5726 * forwarded correctly according to the new state of the flow table.
5728 * This function must be called after *each* change to a flow table. See
5729 * the comment on table_update_taggable() for more information. */
5731 rule_invalidate(const struct rule_dpif *rule)
5733 struct ofproto_dpif *ofproto = ofproto_dpif_cast(rule->up.ofproto);
5735 table_update_taggable(ofproto, rule->up.table_id);
5737 if (!ofproto->need_revalidate) {
5738 struct table_dpif *table = &ofproto->tables[rule->up.table_id];
5740 if (table->other_table && rule->tag) {
5741 tag_set_add(&ofproto->revalidate_set, rule->tag);
5743 ofproto->need_revalidate = true;
5749 set_frag_handling(struct ofproto *ofproto_,
5750 enum ofp_config_flags frag_handling)
5752 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
5754 if (frag_handling != OFPC_FRAG_REASM) {
5755 ofproto->need_revalidate = true;
5763 packet_out(struct ofproto *ofproto_, struct ofpbuf *packet,
5764 const struct flow *flow,
5765 const union ofp_action *ofp_actions, size_t n_ofp_actions)
5767 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
5770 if (flow->in_port >= ofproto->max_ports && flow->in_port < OFPP_MAX) {
5771 return OFPERR_NXBRC_BAD_IN_PORT;
5774 error = validate_actions(ofp_actions, n_ofp_actions, flow,
5775 ofproto->max_ports);
5777 struct odputil_keybuf keybuf;
5778 struct ofpbuf *odp_actions;
5779 struct ofproto_push push;
5782 ofpbuf_use_stack(&key, &keybuf, sizeof keybuf);
5783 odp_flow_key_from_flow(&key, flow);
5785 action_xlate_ctx_init(&push.ctx, ofproto, flow, flow->vlan_tci, NULL,
5786 packet_get_tcp_flags(packet, flow), packet);
5788 /* Ensure that resubmits in 'ofp_actions' get accounted to their
5789 * matching rules. */
5791 push.bytes = packet->size;
5792 push.used = time_msec();
5793 push.ctx.resubmit_hook = push_resubmit;
5795 odp_actions = xlate_actions(&push.ctx, ofp_actions, n_ofp_actions);
5796 dpif_execute(ofproto->dpif, key.data, key.size,
5797 odp_actions->data, odp_actions->size, packet);
5798 ofpbuf_delete(odp_actions);
5806 set_netflow(struct ofproto *ofproto_,
5807 const struct netflow_options *netflow_options)
5809 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
5811 if (netflow_options) {
5812 if (!ofproto->netflow) {
5813 ofproto->netflow = netflow_create();
5815 return netflow_set_options(ofproto->netflow, netflow_options);
5817 netflow_destroy(ofproto->netflow);
5818 ofproto->netflow = NULL;
5824 get_netflow_ids(const struct ofproto *ofproto_,
5825 uint8_t *engine_type, uint8_t *engine_id)
5827 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
5829 dpif_get_netflow_ids(ofproto->dpif, engine_type, engine_id);
5833 send_active_timeout(struct ofproto_dpif *ofproto, struct facet *facet)
5835 if (!facet_is_controller_flow(facet) &&
5836 netflow_active_timeout_expired(ofproto->netflow, &facet->nf_flow)) {
5837 struct subfacet *subfacet;
5838 struct ofexpired expired;
5840 LIST_FOR_EACH (subfacet, list_node, &facet->subfacets) {
5841 if (subfacet->installed) {
5842 struct dpif_flow_stats stats;
5844 subfacet_install(subfacet, subfacet->actions,
5845 subfacet->actions_len, &stats);
5846 subfacet_update_stats(subfacet, &stats);
5850 expired.flow = facet->flow;
5851 expired.packet_count = facet->packet_count;
5852 expired.byte_count = facet->byte_count;
5853 expired.used = facet->used;
5854 netflow_expire(ofproto->netflow, &facet->nf_flow, &expired);
5859 send_netflow_active_timeouts(struct ofproto_dpif *ofproto)
5861 struct facet *facet;
5863 HMAP_FOR_EACH (facet, hmap_node, &ofproto->facets) {
5864 send_active_timeout(ofproto, facet);
5868 static struct ofproto_dpif *
5869 ofproto_dpif_lookup(const char *name)
5871 struct ofproto_dpif *ofproto;
5873 HMAP_FOR_EACH_WITH_HASH (ofproto, all_ofproto_dpifs_node,
5874 hash_string(name, 0), &all_ofproto_dpifs) {
5875 if (!strcmp(ofproto->up.name, name)) {
5883 ofproto_unixctl_fdb_flush(struct unixctl_conn *conn, int argc,
5884 const char *argv[], void *aux OVS_UNUSED)
5886 struct ofproto_dpif *ofproto;
5889 ofproto = ofproto_dpif_lookup(argv[1]);
5891 unixctl_command_reply(conn, 501, "no such bridge");
5894 mac_learning_flush(ofproto->ml, &ofproto->revalidate_set);
5896 HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
5897 mac_learning_flush(ofproto->ml, &ofproto->revalidate_set);
5901 unixctl_command_reply(conn, 200, "table successfully flushed");
5905 ofproto_unixctl_fdb_show(struct unixctl_conn *conn, int argc OVS_UNUSED,
5906 const char *argv[], void *aux OVS_UNUSED)
5908 struct ds ds = DS_EMPTY_INITIALIZER;
5909 const struct ofproto_dpif *ofproto;
5910 const struct mac_entry *e;
5912 ofproto = ofproto_dpif_lookup(argv[1]);
5914 unixctl_command_reply(conn, 501, "no such bridge");
5918 ds_put_cstr(&ds, " port VLAN MAC Age\n");
5919 LIST_FOR_EACH (e, lru_node, &ofproto->ml->lrus) {
5920 struct ofbundle *bundle = e->port.p;
5921 ds_put_format(&ds, "%5d %4d "ETH_ADDR_FMT" %3d\n",
5922 ofbundle_get_a_port(bundle)->odp_port,
5923 e->vlan, ETH_ADDR_ARGS(e->mac),
5924 mac_entry_age(ofproto->ml, e));
5926 unixctl_command_reply(conn, 200, ds_cstr(&ds));
5930 struct ofproto_trace {
5931 struct action_xlate_ctx ctx;
5937 trace_format_rule(struct ds *result, uint8_t table_id, int level,
5938 const struct rule_dpif *rule)
5940 ds_put_char_multiple(result, '\t', level);
5942 ds_put_cstr(result, "No match\n");
5946 ds_put_format(result, "Rule: table=%"PRIu8" cookie=%#"PRIx64" ",
5947 table_id, ntohll(rule->up.flow_cookie));
5948 cls_rule_format(&rule->up.cr, result);
5949 ds_put_char(result, '\n');
5951 ds_put_char_multiple(result, '\t', level);
5952 ds_put_cstr(result, "OpenFlow ");
5953 ofp_print_actions(result, rule->up.actions, rule->up.n_actions);
5954 ds_put_char(result, '\n');
5958 trace_format_flow(struct ds *result, int level, const char *title,
5959 struct ofproto_trace *trace)
5961 ds_put_char_multiple(result, '\t', level);
5962 ds_put_format(result, "%s: ", title);
5963 if (flow_equal(&trace->ctx.flow, &trace->flow)) {
5964 ds_put_cstr(result, "unchanged");
5966 flow_format(result, &trace->ctx.flow);
5967 trace->flow = trace->ctx.flow;
5969 ds_put_char(result, '\n');
5973 trace_format_regs(struct ds *result, int level, const char *title,
5974 struct ofproto_trace *trace)
5978 ds_put_char_multiple(result, '\t', level);
5979 ds_put_format(result, "%s:", title);
5980 for (i = 0; i < FLOW_N_REGS; i++) {
5981 ds_put_format(result, " reg%zu=0x%"PRIx32, i, trace->flow.regs[i]);
5983 ds_put_char(result, '\n');
5987 trace_format_odp(struct ds *result, int level, const char *title,
5988 struct ofproto_trace *trace)
5990 struct ofpbuf *odp_actions = trace->ctx.odp_actions;
5992 ds_put_char_multiple(result, '\t', level);
5993 ds_put_format(result, "%s: ", title);
5994 format_odp_actions(result, odp_actions->data, odp_actions->size);
5995 ds_put_char(result, '\n');
5999 trace_resubmit(struct action_xlate_ctx *ctx, struct rule_dpif *rule)
6001 struct ofproto_trace *trace = CONTAINER_OF(ctx, struct ofproto_trace, ctx);
6002 struct ds *result = trace->result;
6004 ds_put_char(result, '\n');
6005 trace_format_flow(result, ctx->recurse + 1, "Resubmitted flow", trace);
6006 trace_format_regs(result, ctx->recurse + 1, "Resubmitted regs", trace);
6007 trace_format_odp(result, ctx->recurse + 1, "Resubmitted odp", trace);
6008 trace_format_rule(result, ctx->table_id, ctx->recurse + 1, rule);
6012 ofproto_unixctl_trace(struct unixctl_conn *conn, int argc, const char *argv[],
6013 void *aux OVS_UNUSED)
6015 const char *dpname = argv[1];
6016 struct ofproto_dpif *ofproto;
6017 struct ofpbuf odp_key;
6018 struct ofpbuf *packet;
6019 struct rule_dpif *rule;
6020 ovs_be16 initial_tci;
6026 ofpbuf_init(&odp_key, 0);
6029 ofproto = ofproto_dpif_lookup(dpname);
6031 unixctl_command_reply(conn, 501, "Unknown ofproto (use ofproto/list "
6035 if (argc == 3 || (argc == 4 && !strcmp(argv[3], "-generate"))) {
6036 /* ofproto/trace dpname flow [-generate] */
6037 const char *flow_s = argv[2];
6038 const char *generate_s = argv[3];
6041 /* Convert string to datapath key. */
6042 ofpbuf_init(&odp_key, 0);
6043 error = odp_flow_key_from_string(flow_s, NULL, &odp_key);
6045 unixctl_command_reply(conn, 501, "Bad flow syntax");
6049 /* Convert odp_key to flow. */
6050 error = ofproto_dpif_extract_flow_key(ofproto, odp_key.data,
6051 odp_key.size, &flow,
6052 &initial_tci, NULL);
6053 if (error == ODP_FIT_ERROR) {
6054 unixctl_command_reply(conn, 501, "Invalid flow");
6058 /* Generate a packet, if requested. */
6060 packet = ofpbuf_new(0);
6061 flow_compose(packet, &flow);
6063 } else if (argc == 6) {
6064 /* ofproto/trace dpname priority tun_id in_port packet */
6065 const char *priority_s = argv[2];
6066 const char *tun_id_s = argv[3];
6067 const char *in_port_s = argv[4];
6068 const char *packet_s = argv[5];
6069 uint16_t in_port = ofp_port_to_odp_port(atoi(in_port_s));
6070 ovs_be64 tun_id = htonll(strtoull(tun_id_s, NULL, 0));
6071 uint32_t priority = atoi(priority_s);
6074 msg = eth_from_hex(packet_s, &packet);
6076 unixctl_command_reply(conn, 501, msg);
6080 ds_put_cstr(&result, "Packet: ");
6081 s = ofp_packet_to_string(packet->data, packet->size);
6082 ds_put_cstr(&result, s);
6085 flow_extract(packet, priority, tun_id, in_port, &flow);
6086 initial_tci = flow.vlan_tci;
6088 unixctl_command_reply(conn, 501, "Bad command syntax");
6092 ds_put_cstr(&result, "Flow: ");
6093 flow_format(&result, &flow);
6094 ds_put_char(&result, '\n');
6096 rule = rule_dpif_lookup(ofproto, &flow, 0);
6097 trace_format_rule(&result, 0, 0, rule);
6099 struct ofproto_trace trace;
6100 struct ofpbuf *odp_actions;
6103 tcp_flags = packet ? packet_get_tcp_flags(packet, &flow) : 0;
6104 trace.result = &result;
6106 action_xlate_ctx_init(&trace.ctx, ofproto, &flow, initial_tci,
6107 rule, tcp_flags, packet);
6108 trace.ctx.resubmit_hook = trace_resubmit;
6109 odp_actions = xlate_actions(&trace.ctx,
6110 rule->up.actions, rule->up.n_actions);
6112 ds_put_char(&result, '\n');
6113 trace_format_flow(&result, 0, "Final flow", &trace);
6114 ds_put_cstr(&result, "Datapath actions: ");
6115 format_odp_actions(&result, odp_actions->data, odp_actions->size);
6116 ofpbuf_delete(odp_actions);
6118 if (!trace.ctx.may_set_up_flow) {
6120 ds_put_cstr(&result, "\nThis flow is not cachable.");
6122 ds_put_cstr(&result, "\nThe datapath actions are incomplete--"
6123 "for complete actions, please supply a packet.");
6128 unixctl_command_reply(conn, 200, ds_cstr(&result));
6131 ds_destroy(&result);
6132 ofpbuf_delete(packet);
6133 ofpbuf_uninit(&odp_key);
6137 ofproto_dpif_clog(struct unixctl_conn *conn OVS_UNUSED, int argc OVS_UNUSED,
6138 const char *argv[] OVS_UNUSED, void *aux OVS_UNUSED)
6141 unixctl_command_reply(conn, 200, NULL);
6145 ofproto_dpif_unclog(struct unixctl_conn *conn OVS_UNUSED, int argc OVS_UNUSED,
6146 const char *argv[] OVS_UNUSED, void *aux OVS_UNUSED)
6149 unixctl_command_reply(conn, 200, NULL);
6152 /* Runs a self-check of flow translations in 'ofproto'. Appends a message to
6153 * 'reply' describing the results. */
6155 ofproto_dpif_self_check__(struct ofproto_dpif *ofproto, struct ds *reply)
6157 struct facet *facet;
6161 HMAP_FOR_EACH (facet, hmap_node, &ofproto->facets) {
6162 if (!facet_check_consistency(facet)) {
6167 ofproto->need_revalidate = true;
6171 ds_put_format(reply, "%s: self-check failed (%d errors)\n",
6172 ofproto->up.name, errors);
6174 ds_put_format(reply, "%s: self-check passed\n", ofproto->up.name);
6179 ofproto_dpif_self_check(struct unixctl_conn *conn,
6180 int argc, const char *argv[], void *aux OVS_UNUSED)
6182 struct ds reply = DS_EMPTY_INITIALIZER;
6183 struct ofproto_dpif *ofproto;
6186 ofproto = ofproto_dpif_lookup(argv[1]);
6188 unixctl_command_reply(conn, 501, "Unknown ofproto (use "
6189 "ofproto/list for help)");
6192 ofproto_dpif_self_check__(ofproto, &reply);
6194 HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
6195 ofproto_dpif_self_check__(ofproto, &reply);
6199 unixctl_command_reply(conn, 200, ds_cstr(&reply));
6204 ofproto_dpif_unixctl_init(void)
6206 static bool registered;
6212 unixctl_command_register(
6214 "bridge {tun_id in_port packet | odp_flow [-generate]}",
6215 2, 5, ofproto_unixctl_trace, NULL);
6216 unixctl_command_register("fdb/flush", "[bridge]", 0, 1,
6217 ofproto_unixctl_fdb_flush, NULL);
6218 unixctl_command_register("fdb/show", "bridge", 1, 1,
6219 ofproto_unixctl_fdb_show, NULL);
6220 unixctl_command_register("ofproto/clog", "", 0, 0,
6221 ofproto_dpif_clog, NULL);
6222 unixctl_command_register("ofproto/unclog", "", 0, 0,
6223 ofproto_dpif_unclog, NULL);
6224 unixctl_command_register("ofproto/self-check", "[bridge]", 0, 1,
6225 ofproto_dpif_self_check, NULL);
6228 /* Linux VLAN device support (e.g. "eth0.10" for VLAN 10.)
6230 * This is deprecated. It is only for compatibility with broken device drivers
6231 * in old versions of Linux that do not properly support VLANs when VLAN
6232 * devices are not used. When broken device drivers are no longer in
6233 * widespread use, we will delete these interfaces. */
6236 set_realdev(struct ofport *ofport_, uint16_t realdev_ofp_port, int vid)
6238 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport_->ofproto);
6239 struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
6241 if (realdev_ofp_port == ofport->realdev_ofp_port
6242 && vid == ofport->vlandev_vid) {
6246 ofproto->need_revalidate = true;
6248 if (ofport->realdev_ofp_port) {
6251 if (realdev_ofp_port && ofport->bundle) {
6252 /* vlandevs are enslaved to their realdevs, so they are not allowed to
6253 * themselves be part of a bundle. */
6254 bundle_set(ofport->up.ofproto, ofport->bundle, NULL);
6257 ofport->realdev_ofp_port = realdev_ofp_port;
6258 ofport->vlandev_vid = vid;
6260 if (realdev_ofp_port) {
6261 vsp_add(ofport, realdev_ofp_port, vid);
6268 hash_realdev_vid(uint16_t realdev_ofp_port, int vid)
6270 return hash_2words(realdev_ofp_port, vid);
6274 vsp_realdev_to_vlandev(const struct ofproto_dpif *ofproto,
6275 uint32_t realdev_odp_port, ovs_be16 vlan_tci)
6277 if (!hmap_is_empty(&ofproto->realdev_vid_map)) {
6278 uint16_t realdev_ofp_port = odp_port_to_ofp_port(realdev_odp_port);
6279 int vid = vlan_tci_to_vid(vlan_tci);
6280 const struct vlan_splinter *vsp;
6282 HMAP_FOR_EACH_WITH_HASH (vsp, realdev_vid_node,
6283 hash_realdev_vid(realdev_ofp_port, vid),
6284 &ofproto->realdev_vid_map) {
6285 if (vsp->realdev_ofp_port == realdev_ofp_port
6286 && vsp->vid == vid) {
6287 return ofp_port_to_odp_port(vsp->vlandev_ofp_port);
6291 return realdev_odp_port;
6294 static struct vlan_splinter *
6295 vlandev_find(const struct ofproto_dpif *ofproto, uint16_t vlandev_ofp_port)
6297 struct vlan_splinter *vsp;
6299 HMAP_FOR_EACH_WITH_HASH (vsp, vlandev_node, hash_int(vlandev_ofp_port, 0),
6300 &ofproto->vlandev_map) {
6301 if (vsp->vlandev_ofp_port == vlandev_ofp_port) {
6310 vsp_vlandev_to_realdev(const struct ofproto_dpif *ofproto,
6311 uint16_t vlandev_ofp_port, int *vid)
6313 if (!hmap_is_empty(&ofproto->vlandev_map)) {
6314 const struct vlan_splinter *vsp;
6316 vsp = vlandev_find(ofproto, vlandev_ofp_port);
6321 return vsp->realdev_ofp_port;
6328 vsp_remove(struct ofport_dpif *port)
6330 struct ofproto_dpif *ofproto = ofproto_dpif_cast(port->up.ofproto);
6331 struct vlan_splinter *vsp;
6333 vsp = vlandev_find(ofproto, port->up.ofp_port);
6335 hmap_remove(&ofproto->vlandev_map, &vsp->vlandev_node);
6336 hmap_remove(&ofproto->realdev_vid_map, &vsp->realdev_vid_node);
6339 port->realdev_ofp_port = 0;
6341 VLOG_ERR("missing vlan device record");
6346 vsp_add(struct ofport_dpif *port, uint16_t realdev_ofp_port, int vid)
6348 struct ofproto_dpif *ofproto = ofproto_dpif_cast(port->up.ofproto);
6350 if (!vsp_vlandev_to_realdev(ofproto, port->up.ofp_port, NULL)
6351 && (vsp_realdev_to_vlandev(ofproto, realdev_ofp_port, htons(vid))
6352 == realdev_ofp_port)) {
6353 struct vlan_splinter *vsp;
6355 vsp = xmalloc(sizeof *vsp);
6356 hmap_insert(&ofproto->vlandev_map, &vsp->vlandev_node,
6357 hash_int(port->up.ofp_port, 0));
6358 hmap_insert(&ofproto->realdev_vid_map, &vsp->realdev_vid_node,
6359 hash_realdev_vid(realdev_ofp_port, vid));
6360 vsp->realdev_ofp_port = realdev_ofp_port;
6361 vsp->vlandev_ofp_port = port->up.ofp_port;
6364 port->realdev_ofp_port = realdev_ofp_port;
6366 VLOG_ERR("duplicate vlan device record");
6370 const struct ofproto_class ofproto_dpif_class = {
6399 port_is_lacp_current,
6400 NULL, /* rule_choose_table */
6407 rule_modify_actions,
6415 get_cfm_remote_mpids,
6419 get_stp_port_status,
6426 is_mirror_output_bundle,
6427 forward_bpdu_changed,