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 OFPUTIL_PC_NO_FLOOD set. */
166 /* Port mirroring info. */
167 mirror_mask_t src_mirrors; /* Mirrors triggered when packet received. */
168 mirror_mask_t dst_mirrors; /* Mirrors triggered when packet sent. */
169 mirror_mask_t mirror_out; /* Mirrors that output to this bundle. */
172 static void bundle_remove(struct ofport *);
173 static void bundle_update(struct ofbundle *);
174 static void bundle_destroy(struct ofbundle *);
175 static void bundle_del_port(struct ofport_dpif *);
176 static void bundle_run(struct ofbundle *);
177 static void bundle_wait(struct ofbundle *);
178 static struct ofbundle *lookup_input_bundle(struct ofproto_dpif *,
179 uint16_t in_port, bool warn);
181 /* A controller may use OFPP_NONE as the ingress port to indicate that
182 * it did not arrive on a "real" port. 'ofpp_none_bundle' exists for
183 * when an input bundle is needed for validation (e.g., mirroring or
184 * OFPP_NORMAL processing). It is not connected to an 'ofproto' or have
185 * any 'port' structs, so care must be taken when dealing with it. */
186 static struct ofbundle ofpp_none_bundle = {
188 .vlan_mode = PORT_VLAN_TRUNK
191 static void stp_run(struct ofproto_dpif *ofproto);
192 static void stp_wait(struct ofproto_dpif *ofproto);
193 static int set_stp_port(struct ofport *,
194 const struct ofproto_port_stp_settings *);
196 static bool ofbundle_includes_vlan(const struct ofbundle *, uint16_t vlan);
198 struct action_xlate_ctx {
199 /* action_xlate_ctx_init() initializes these members. */
202 struct ofproto_dpif *ofproto;
204 /* Flow to which the OpenFlow actions apply. xlate_actions() will modify
205 * this flow when actions change header fields. */
208 /* The packet corresponding to 'flow', or a null pointer if we are
209 * revalidating without a packet to refer to. */
210 const struct ofpbuf *packet;
212 /* Should OFPP_NORMAL 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. */
419 long long int carrier_seq; /* Carrier status changes. */
422 struct stp_port *stp_port; /* Spanning Tree Protocol, if any. */
423 enum stp_state stp_state; /* Always STP_DISABLED if STP not in use. */
424 long long int stp_state_entered;
426 struct hmap priorities; /* Map of attached 'priority_to_dscp's. */
428 /* Linux VLAN device support (e.g. "eth0.10" for VLAN 10.)
430 * This is deprecated. It is only for compatibility with broken device
431 * drivers in old versions of Linux that do not properly support VLANs when
432 * VLAN devices are not used. When broken device drivers are no longer in
433 * widespread use, we will delete these interfaces. */
434 uint16_t realdev_ofp_port;
438 /* Node in 'ofport_dpif''s 'priorities' map. Used to maintain a map from
439 * 'priority' (the datapath's term for QoS queue) to the dscp bits which all
440 * traffic egressing the 'ofport' with that priority should be marked with. */
441 struct priority_to_dscp {
442 struct hmap_node hmap_node; /* Node in 'ofport_dpif''s 'priorities' map. */
443 uint32_t priority; /* Priority of this queue (see struct flow). */
445 uint8_t dscp; /* DSCP bits to mark outgoing traffic with. */
448 /* Linux VLAN device support (e.g. "eth0.10" for VLAN 10.)
450 * This is deprecated. It is only for compatibility with broken device drivers
451 * in old versions of Linux that do not properly support VLANs when VLAN
452 * devices are not used. When broken device drivers are no longer in
453 * widespread use, we will delete these interfaces. */
454 struct vlan_splinter {
455 struct hmap_node realdev_vid_node;
456 struct hmap_node vlandev_node;
457 uint16_t realdev_ofp_port;
458 uint16_t vlandev_ofp_port;
462 static uint32_t vsp_realdev_to_vlandev(const struct ofproto_dpif *,
463 uint32_t realdev, ovs_be16 vlan_tci);
464 static uint16_t vsp_vlandev_to_realdev(const struct ofproto_dpif *,
465 uint16_t vlandev, int *vid);
466 static void vsp_remove(struct ofport_dpif *);
467 static void vsp_add(struct ofport_dpif *, uint16_t realdev_ofp_port, int vid);
469 static struct ofport_dpif *
470 ofport_dpif_cast(const struct ofport *ofport)
472 assert(ofport->ofproto->ofproto_class == &ofproto_dpif_class);
473 return ofport ? CONTAINER_OF(ofport, struct ofport_dpif, up) : NULL;
476 static void port_run(struct ofport_dpif *);
477 static void port_wait(struct ofport_dpif *);
478 static int set_cfm(struct ofport *, const struct cfm_settings *);
479 static void ofport_clear_priorities(struct ofport_dpif *);
481 struct dpif_completion {
482 struct list list_node;
483 struct ofoperation *op;
486 /* Extra information about a classifier table.
487 * Currently used just for optimized flow revalidation. */
489 /* If either of these is nonnull, then this table has a form that allows
490 * flows to be tagged to avoid revalidating most flows for the most common
491 * kinds of flow table changes. */
492 struct cls_table *catchall_table; /* Table that wildcards all fields. */
493 struct cls_table *other_table; /* Table with any other wildcard set. */
494 uint32_t basis; /* Keeps each table's tags separate. */
497 struct ofproto_dpif {
498 struct hmap_node all_ofproto_dpifs_node; /* In 'all_ofproto_dpifs'. */
507 struct netflow *netflow;
508 struct dpif_sflow *sflow;
509 struct hmap bundles; /* Contains "struct ofbundle"s. */
510 struct mac_learning *ml;
511 struct ofmirror *mirrors[MAX_MIRRORS];
512 bool has_bonded_bundles;
515 struct timer next_expiration;
519 struct hmap subfacets;
522 struct table_dpif tables[N_TABLES];
523 bool need_revalidate;
524 struct tag_set revalidate_set;
526 /* Support for debugging async flow mods. */
527 struct list completions;
529 bool has_bundle_action; /* True when the first bundle action appears. */
530 struct netdev_stats stats; /* To account packets generated and consumed in
535 long long int stp_last_tick;
537 /* VLAN splinters. */
538 struct hmap realdev_vid_map; /* (realdev,vid) -> vlandev. */
539 struct hmap vlandev_map; /* vlandev -> (realdev,vid). */
542 /* Defer flow mod completion until "ovs-appctl ofproto/unclog"? (Useful only
543 * for debugging the asynchronous flow_mod implementation.) */
546 /* All existing ofproto_dpif instances, indexed by ->up.name. */
547 static struct hmap all_ofproto_dpifs = HMAP_INITIALIZER(&all_ofproto_dpifs);
549 static void ofproto_dpif_unixctl_init(void);
551 static struct ofproto_dpif *
552 ofproto_dpif_cast(const struct ofproto *ofproto)
554 assert(ofproto->ofproto_class == &ofproto_dpif_class);
555 return CONTAINER_OF(ofproto, struct ofproto_dpif, up);
558 static struct ofport_dpif *get_ofp_port(struct ofproto_dpif *,
560 static struct ofport_dpif *get_odp_port(struct ofproto_dpif *,
563 /* Packet processing. */
564 static void update_learning_table(struct ofproto_dpif *,
565 const struct flow *, int vlan,
568 #define FLOW_MISS_MAX_BATCH 50
569 static int handle_upcalls(struct ofproto_dpif *, unsigned int max_batch);
571 /* Flow expiration. */
572 static int expire(struct ofproto_dpif *);
575 static void send_netflow_active_timeouts(struct ofproto_dpif *);
578 static int send_packet(const struct ofport_dpif *, struct ofpbuf *packet);
580 compose_sflow_action(const struct ofproto_dpif *, struct ofpbuf *odp_actions,
581 const struct flow *, uint32_t odp_port);
582 static void add_mirror_actions(struct action_xlate_ctx *ctx,
583 const struct flow *flow);
584 /* Global variables. */
585 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
587 /* Factory functions. */
590 enumerate_types(struct sset *types)
592 dp_enumerate_types(types);
596 enumerate_names(const char *type, struct sset *names)
598 return dp_enumerate_names(type, names);
602 del(const char *type, const char *name)
607 error = dpif_open(name, type, &dpif);
609 error = dpif_delete(dpif);
615 /* Basic life-cycle. */
617 static struct ofproto *
620 struct ofproto_dpif *ofproto = xmalloc(sizeof *ofproto);
625 dealloc(struct ofproto *ofproto_)
627 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
632 construct(struct ofproto *ofproto_)
634 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
635 const char *name = ofproto->up.name;
639 error = dpif_create_and_open(name, ofproto->up.type, &ofproto->dpif);
641 VLOG_ERR("failed to open datapath %s: %s", name, strerror(error));
645 ofproto->max_ports = dpif_get_max_ports(ofproto->dpif);
646 ofproto->n_matches = 0;
648 dpif_flow_flush(ofproto->dpif);
649 dpif_recv_purge(ofproto->dpif);
651 error = dpif_recv_set(ofproto->dpif, true);
653 VLOG_ERR("failed to listen on datapath %s: %s", name, strerror(error));
654 dpif_close(ofproto->dpif);
658 ofproto->netflow = NULL;
659 ofproto->sflow = NULL;
661 hmap_init(&ofproto->bundles);
662 ofproto->ml = mac_learning_create(MAC_ENTRY_DEFAULT_IDLE_TIME);
663 for (i = 0; i < MAX_MIRRORS; i++) {
664 ofproto->mirrors[i] = NULL;
666 ofproto->has_bonded_bundles = false;
668 timer_set_duration(&ofproto->next_expiration, 1000);
670 hmap_init(&ofproto->facets);
671 hmap_init(&ofproto->subfacets);
673 for (i = 0; i < N_TABLES; i++) {
674 struct table_dpif *table = &ofproto->tables[i];
676 table->catchall_table = NULL;
677 table->other_table = NULL;
678 table->basis = random_uint32();
680 ofproto->need_revalidate = false;
681 tag_set_init(&ofproto->revalidate_set);
683 list_init(&ofproto->completions);
685 ofproto_dpif_unixctl_init();
687 ofproto->has_bundle_action = false;
689 hmap_init(&ofproto->vlandev_map);
690 hmap_init(&ofproto->realdev_vid_map);
692 hmap_insert(&all_ofproto_dpifs, &ofproto->all_ofproto_dpifs_node,
693 hash_string(ofproto->up.name, 0));
694 memset(&ofproto->stats, 0, sizeof ofproto->stats);
696 ofproto_init_tables(ofproto_, N_TABLES);
702 complete_operations(struct ofproto_dpif *ofproto)
704 struct dpif_completion *c, *next;
706 LIST_FOR_EACH_SAFE (c, next, list_node, &ofproto->completions) {
707 ofoperation_complete(c->op, 0);
708 list_remove(&c->list_node);
714 destruct(struct ofproto *ofproto_)
716 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
717 struct rule_dpif *rule, *next_rule;
718 struct oftable *table;
721 hmap_remove(&all_ofproto_dpifs, &ofproto->all_ofproto_dpifs_node);
722 complete_operations(ofproto);
724 OFPROTO_FOR_EACH_TABLE (table, &ofproto->up) {
725 struct cls_cursor cursor;
727 cls_cursor_init(&cursor, &table->cls, NULL);
728 CLS_CURSOR_FOR_EACH_SAFE (rule, next_rule, up.cr, &cursor) {
729 ofproto_rule_destroy(&rule->up);
733 for (i = 0; i < MAX_MIRRORS; i++) {
734 mirror_destroy(ofproto->mirrors[i]);
737 netflow_destroy(ofproto->netflow);
738 dpif_sflow_destroy(ofproto->sflow);
739 hmap_destroy(&ofproto->bundles);
740 mac_learning_destroy(ofproto->ml);
742 hmap_destroy(&ofproto->facets);
743 hmap_destroy(&ofproto->subfacets);
745 hmap_destroy(&ofproto->vlandev_map);
746 hmap_destroy(&ofproto->realdev_vid_map);
748 dpif_close(ofproto->dpif);
752 run_fast(struct ofproto *ofproto_)
754 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
757 /* Handle one or more batches of upcalls, until there's nothing left to do
758 * or until we do a fixed total amount of work.
760 * We do work in batches because it can be much cheaper to set up a number
761 * of flows and fire off their patches all at once. We do multiple batches
762 * because in some cases handling a packet can cause another packet to be
763 * queued almost immediately as part of the return flow. Both
764 * optimizations can make major improvements on some benchmarks and
765 * presumably for real traffic as well. */
767 while (work < FLOW_MISS_MAX_BATCH) {
768 int retval = handle_upcalls(ofproto, FLOW_MISS_MAX_BATCH - work);
778 run(struct ofproto *ofproto_)
780 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
781 struct ofport_dpif *ofport;
782 struct ofbundle *bundle;
786 complete_operations(ofproto);
788 dpif_run(ofproto->dpif);
790 error = run_fast(ofproto_);
795 if (timer_expired(&ofproto->next_expiration)) {
796 int delay = expire(ofproto);
797 timer_set_duration(&ofproto->next_expiration, delay);
800 if (ofproto->netflow) {
801 if (netflow_run(ofproto->netflow)) {
802 send_netflow_active_timeouts(ofproto);
805 if (ofproto->sflow) {
806 dpif_sflow_run(ofproto->sflow);
809 HMAP_FOR_EACH (ofport, up.hmap_node, &ofproto->up.ports) {
812 HMAP_FOR_EACH (bundle, hmap_node, &ofproto->bundles) {
817 mac_learning_run(ofproto->ml, &ofproto->revalidate_set);
819 /* Now revalidate if there's anything to do. */
820 if (ofproto->need_revalidate
821 || !tag_set_is_empty(&ofproto->revalidate_set)) {
822 struct tag_set revalidate_set = ofproto->revalidate_set;
823 bool revalidate_all = ofproto->need_revalidate;
824 struct facet *facet, *next;
826 /* Clear the revalidation flags. */
827 tag_set_init(&ofproto->revalidate_set);
828 ofproto->need_revalidate = false;
830 HMAP_FOR_EACH_SAFE (facet, next, hmap_node, &ofproto->facets) {
832 || tag_set_intersects(&revalidate_set, facet->tags)) {
833 facet_revalidate(facet);
838 /* Check the consistency of a random facet, to aid debugging. */
839 if (!hmap_is_empty(&ofproto->facets) && !ofproto->need_revalidate) {
842 facet = CONTAINER_OF(hmap_random_node(&ofproto->facets),
843 struct facet, hmap_node);
844 if (!tag_set_intersects(&ofproto->revalidate_set, facet->tags)) {
845 if (!facet_check_consistency(facet)) {
846 ofproto->need_revalidate = true;
855 wait(struct ofproto *ofproto_)
857 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
858 struct ofport_dpif *ofport;
859 struct ofbundle *bundle;
861 if (!clogged && !list_is_empty(&ofproto->completions)) {
862 poll_immediate_wake();
865 dpif_wait(ofproto->dpif);
866 dpif_recv_wait(ofproto->dpif);
867 if (ofproto->sflow) {
868 dpif_sflow_wait(ofproto->sflow);
870 if (!tag_set_is_empty(&ofproto->revalidate_set)) {
871 poll_immediate_wake();
873 HMAP_FOR_EACH (ofport, up.hmap_node, &ofproto->up.ports) {
876 HMAP_FOR_EACH (bundle, hmap_node, &ofproto->bundles) {
879 if (ofproto->netflow) {
880 netflow_wait(ofproto->netflow);
882 mac_learning_wait(ofproto->ml);
884 if (ofproto->need_revalidate) {
885 /* Shouldn't happen, but if it does just go around again. */
886 VLOG_DBG_RL(&rl, "need revalidate in ofproto_wait_cb()");
887 poll_immediate_wake();
889 timer_wait(&ofproto->next_expiration);
894 flush(struct ofproto *ofproto_)
896 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
897 struct facet *facet, *next_facet;
899 HMAP_FOR_EACH_SAFE (facet, next_facet, hmap_node, &ofproto->facets) {
900 /* Mark the facet as not installed so that facet_remove() doesn't
901 * bother trying to uninstall it. There is no point in uninstalling it
902 * individually since we are about to blow away all the facets with
903 * dpif_flow_flush(). */
904 struct subfacet *subfacet;
906 LIST_FOR_EACH (subfacet, list_node, &facet->subfacets) {
907 subfacet->installed = false;
908 subfacet->dp_packet_count = 0;
909 subfacet->dp_byte_count = 0;
913 dpif_flow_flush(ofproto->dpif);
917 get_features(struct ofproto *ofproto_ OVS_UNUSED,
918 bool *arp_match_ip, enum ofputil_action_bitmap *actions)
920 *arp_match_ip = true;
921 *actions = (OFPUTIL_A_OUTPUT |
922 OFPUTIL_A_SET_VLAN_VID |
923 OFPUTIL_A_SET_VLAN_PCP |
924 OFPUTIL_A_STRIP_VLAN |
925 OFPUTIL_A_SET_DL_SRC |
926 OFPUTIL_A_SET_DL_DST |
927 OFPUTIL_A_SET_NW_SRC |
928 OFPUTIL_A_SET_NW_DST |
929 OFPUTIL_A_SET_NW_TOS |
930 OFPUTIL_A_SET_TP_SRC |
931 OFPUTIL_A_SET_TP_DST |
936 get_tables(struct ofproto *ofproto_, struct ofp_table_stats *ots)
938 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
939 struct dpif_dp_stats s;
941 strcpy(ots->name, "classifier");
943 dpif_get_dp_stats(ofproto->dpif, &s);
944 put_32aligned_be64(&ots->lookup_count, htonll(s.n_hit + s.n_missed));
945 put_32aligned_be64(&ots->matched_count,
946 htonll(s.n_hit + ofproto->n_matches));
949 static struct ofport *
952 struct ofport_dpif *port = xmalloc(sizeof *port);
957 port_dealloc(struct ofport *port_)
959 struct ofport_dpif *port = ofport_dpif_cast(port_);
964 port_construct(struct ofport *port_)
966 struct ofport_dpif *port = ofport_dpif_cast(port_);
967 struct ofproto_dpif *ofproto = ofproto_dpif_cast(port->up.ofproto);
969 ofproto->need_revalidate = true;
970 port->odp_port = ofp_port_to_odp_port(port->up.ofp_port);
973 port->tag = tag_create_random();
974 port->may_enable = true;
975 port->stp_port = NULL;
976 port->stp_state = STP_DISABLED;
977 hmap_init(&port->priorities);
978 port->realdev_ofp_port = 0;
979 port->vlandev_vid = 0;
980 port->carrier_seq = netdev_get_carrier_resets(port->up.netdev);
982 if (ofproto->sflow) {
983 dpif_sflow_add_port(ofproto->sflow, port_);
990 port_destruct(struct ofport *port_)
992 struct ofport_dpif *port = ofport_dpif_cast(port_);
993 struct ofproto_dpif *ofproto = ofproto_dpif_cast(port->up.ofproto);
995 ofproto->need_revalidate = true;
996 bundle_remove(port_);
997 set_cfm(port_, NULL);
998 if (ofproto->sflow) {
999 dpif_sflow_del_port(ofproto->sflow, port->odp_port);
1002 ofport_clear_priorities(port);
1003 hmap_destroy(&port->priorities);
1007 port_modified(struct ofport *port_)
1009 struct ofport_dpif *port = ofport_dpif_cast(port_);
1011 if (port->bundle && port->bundle->bond) {
1012 bond_slave_set_netdev(port->bundle->bond, port, port->up.netdev);
1017 port_reconfigured(struct ofport *port_, enum ofputil_port_config old_config)
1019 struct ofport_dpif *port = ofport_dpif_cast(port_);
1020 struct ofproto_dpif *ofproto = ofproto_dpif_cast(port->up.ofproto);
1021 enum ofputil_port_config changed = old_config ^ port->up.pp.config;
1023 if (changed & (OFPUTIL_PC_NO_RECV | OFPUTIL_PC_NO_RECV_STP |
1024 OFPUTIL_PC_NO_FWD | OFPUTIL_PC_NO_FLOOD)) {
1025 ofproto->need_revalidate = true;
1027 if (changed & OFPUTIL_PC_NO_FLOOD && port->bundle) {
1028 bundle_update(port->bundle);
1034 set_sflow(struct ofproto *ofproto_,
1035 const struct ofproto_sflow_options *sflow_options)
1037 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1038 struct dpif_sflow *ds = ofproto->sflow;
1040 if (sflow_options) {
1042 struct ofport_dpif *ofport;
1044 ds = ofproto->sflow = dpif_sflow_create(ofproto->dpif);
1045 HMAP_FOR_EACH (ofport, up.hmap_node, &ofproto->up.ports) {
1046 dpif_sflow_add_port(ds, &ofport->up);
1048 ofproto->need_revalidate = true;
1050 dpif_sflow_set_options(ds, sflow_options);
1053 dpif_sflow_destroy(ds);
1054 ofproto->need_revalidate = true;
1055 ofproto->sflow = NULL;
1062 set_cfm(struct ofport *ofport_, const struct cfm_settings *s)
1064 struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
1071 struct ofproto_dpif *ofproto;
1073 ofproto = ofproto_dpif_cast(ofport->up.ofproto);
1074 ofproto->need_revalidate = true;
1075 ofport->cfm = cfm_create(netdev_get_name(ofport->up.netdev));
1078 if (cfm_configure(ofport->cfm, s)) {
1084 cfm_destroy(ofport->cfm);
1090 get_cfm_fault(const struct ofport *ofport_)
1092 struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
1094 return ofport->cfm ? cfm_get_fault(ofport->cfm) : -1;
1098 get_cfm_remote_mpids(const struct ofport *ofport_, const uint64_t **rmps,
1101 struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
1104 cfm_get_remote_mpids(ofport->cfm, rmps, n_rmps);
1111 /* Spanning Tree. */
1114 send_bpdu_cb(struct ofpbuf *pkt, int port_num, void *ofproto_)
1116 struct ofproto_dpif *ofproto = ofproto_;
1117 struct stp_port *sp = stp_get_port(ofproto->stp, port_num);
1118 struct ofport_dpif *ofport;
1120 ofport = stp_port_get_aux(sp);
1122 VLOG_WARN_RL(&rl, "%s: cannot send BPDU on unknown port %d",
1123 ofproto->up.name, port_num);
1125 struct eth_header *eth = pkt->l2;
1127 netdev_get_etheraddr(ofport->up.netdev, eth->eth_src);
1128 if (eth_addr_is_zero(eth->eth_src)) {
1129 VLOG_WARN_RL(&rl, "%s: cannot send BPDU on port %d "
1130 "with unknown MAC", ofproto->up.name, port_num);
1132 send_packet(ofport, pkt);
1138 /* Configures STP on 'ofproto_' using the settings defined in 's'. */
1140 set_stp(struct ofproto *ofproto_, const struct ofproto_stp_settings *s)
1142 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1144 /* Only revalidate flows if the configuration changed. */
1145 if (!s != !ofproto->stp) {
1146 ofproto->need_revalidate = true;
1150 if (!ofproto->stp) {
1151 ofproto->stp = stp_create(ofproto_->name, s->system_id,
1152 send_bpdu_cb, ofproto);
1153 ofproto->stp_last_tick = time_msec();
1156 stp_set_bridge_id(ofproto->stp, s->system_id);
1157 stp_set_bridge_priority(ofproto->stp, s->priority);
1158 stp_set_hello_time(ofproto->stp, s->hello_time);
1159 stp_set_max_age(ofproto->stp, s->max_age);
1160 stp_set_forward_delay(ofproto->stp, s->fwd_delay);
1162 struct ofport *ofport;
1164 HMAP_FOR_EACH (ofport, hmap_node, &ofproto->up.ports) {
1165 set_stp_port(ofport, NULL);
1168 stp_destroy(ofproto->stp);
1169 ofproto->stp = NULL;
1176 get_stp_status(struct ofproto *ofproto_, struct ofproto_stp_status *s)
1178 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1182 s->bridge_id = stp_get_bridge_id(ofproto->stp);
1183 s->designated_root = stp_get_designated_root(ofproto->stp);
1184 s->root_path_cost = stp_get_root_path_cost(ofproto->stp);
1193 update_stp_port_state(struct ofport_dpif *ofport)
1195 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport->up.ofproto);
1196 enum stp_state state;
1198 /* Figure out new state. */
1199 state = ofport->stp_port ? stp_port_get_state(ofport->stp_port)
1203 if (ofport->stp_state != state) {
1204 enum ofputil_port_state of_state;
1207 VLOG_DBG_RL(&rl, "port %s: STP state changed from %s to %s",
1208 netdev_get_name(ofport->up.netdev),
1209 stp_state_name(ofport->stp_state),
1210 stp_state_name(state));
1211 if (stp_learn_in_state(ofport->stp_state)
1212 != stp_learn_in_state(state)) {
1213 /* xxx Learning action flows should also be flushed. */
1214 mac_learning_flush(ofproto->ml, &ofproto->revalidate_set);
1216 fwd_change = stp_forward_in_state(ofport->stp_state)
1217 != stp_forward_in_state(state);
1219 ofproto->need_revalidate = true;
1220 ofport->stp_state = state;
1221 ofport->stp_state_entered = time_msec();
1223 if (fwd_change && ofport->bundle) {
1224 bundle_update(ofport->bundle);
1227 /* Update the STP state bits in the OpenFlow port description. */
1228 of_state = ofport->up.pp.state & ~OFPUTIL_PS_STP_MASK;
1229 of_state |= (state == STP_LISTENING ? OFPUTIL_PS_STP_LISTEN
1230 : state == STP_LEARNING ? OFPUTIL_PS_STP_LEARN
1231 : state == STP_FORWARDING ? OFPUTIL_PS_STP_FORWARD
1232 : state == STP_BLOCKING ? OFPUTIL_PS_STP_BLOCK
1234 ofproto_port_set_state(&ofport->up, of_state);
1238 /* Configures STP on 'ofport_' using the settings defined in 's'. The
1239 * caller is responsible for assigning STP port numbers and ensuring
1240 * there are no duplicates. */
1242 set_stp_port(struct ofport *ofport_,
1243 const struct ofproto_port_stp_settings *s)
1245 struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
1246 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport->up.ofproto);
1247 struct stp_port *sp = ofport->stp_port;
1249 if (!s || !s->enable) {
1251 ofport->stp_port = NULL;
1252 stp_port_disable(sp);
1253 update_stp_port_state(ofport);
1256 } else if (sp && stp_port_no(sp) != s->port_num
1257 && ofport == stp_port_get_aux(sp)) {
1258 /* The port-id changed, so disable the old one if it's not
1259 * already in use by another port. */
1260 stp_port_disable(sp);
1263 sp = ofport->stp_port = stp_get_port(ofproto->stp, s->port_num);
1264 stp_port_enable(sp);
1266 stp_port_set_aux(sp, ofport);
1267 stp_port_set_priority(sp, s->priority);
1268 stp_port_set_path_cost(sp, s->path_cost);
1270 update_stp_port_state(ofport);
1276 get_stp_port_status(struct ofport *ofport_,
1277 struct ofproto_port_stp_status *s)
1279 struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
1280 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport->up.ofproto);
1281 struct stp_port *sp = ofport->stp_port;
1283 if (!ofproto->stp || !sp) {
1289 s->port_id = stp_port_get_id(sp);
1290 s->state = stp_port_get_state(sp);
1291 s->sec_in_state = (time_msec() - ofport->stp_state_entered) / 1000;
1292 s->role = stp_port_get_role(sp);
1293 stp_port_get_counts(sp, &s->tx_count, &s->rx_count, &s->error_count);
1299 stp_run(struct ofproto_dpif *ofproto)
1302 long long int now = time_msec();
1303 long long int elapsed = now - ofproto->stp_last_tick;
1304 struct stp_port *sp;
1307 stp_tick(ofproto->stp, MIN(INT_MAX, elapsed));
1308 ofproto->stp_last_tick = now;
1310 while (stp_get_changed_port(ofproto->stp, &sp)) {
1311 struct ofport_dpif *ofport = stp_port_get_aux(sp);
1314 update_stp_port_state(ofport);
1318 if (stp_check_and_reset_fdb_flush(ofproto->stp)) {
1319 mac_learning_flush(ofproto->ml, &ofproto->revalidate_set);
1325 stp_wait(struct ofproto_dpif *ofproto)
1328 poll_timer_wait(1000);
1332 /* Returns true if STP should process 'flow'. */
1334 stp_should_process_flow(const struct flow *flow)
1336 return eth_addr_equals(flow->dl_dst, eth_addr_stp);
1340 stp_process_packet(const struct ofport_dpif *ofport,
1341 const struct ofpbuf *packet)
1343 struct ofpbuf payload = *packet;
1344 struct eth_header *eth = payload.data;
1345 struct stp_port *sp = ofport->stp_port;
1347 /* Sink packets on ports that have STP disabled when the bridge has
1349 if (!sp || stp_port_get_state(sp) == STP_DISABLED) {
1353 /* Trim off padding on payload. */
1354 if (payload.size > ntohs(eth->eth_type) + ETH_HEADER_LEN) {
1355 payload.size = ntohs(eth->eth_type) + ETH_HEADER_LEN;
1358 if (ofpbuf_try_pull(&payload, ETH_HEADER_LEN + LLC_HEADER_LEN)) {
1359 stp_received_bpdu(sp, payload.data, payload.size);
1363 static struct priority_to_dscp *
1364 get_priority(const struct ofport_dpif *ofport, uint32_t priority)
1366 struct priority_to_dscp *pdscp;
1369 hash = hash_int(priority, 0);
1370 HMAP_FOR_EACH_IN_BUCKET (pdscp, hmap_node, hash, &ofport->priorities) {
1371 if (pdscp->priority == priority) {
1379 ofport_clear_priorities(struct ofport_dpif *ofport)
1381 struct priority_to_dscp *pdscp, *next;
1383 HMAP_FOR_EACH_SAFE (pdscp, next, hmap_node, &ofport->priorities) {
1384 hmap_remove(&ofport->priorities, &pdscp->hmap_node);
1390 set_queues(struct ofport *ofport_,
1391 const struct ofproto_port_queue *qdscp_list,
1394 struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
1395 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport->up.ofproto);
1396 struct hmap new = HMAP_INITIALIZER(&new);
1399 for (i = 0; i < n_qdscp; i++) {
1400 struct priority_to_dscp *pdscp;
1404 dscp = (qdscp_list[i].dscp << 2) & IP_DSCP_MASK;
1405 if (dpif_queue_to_priority(ofproto->dpif, qdscp_list[i].queue,
1410 pdscp = get_priority(ofport, priority);
1412 hmap_remove(&ofport->priorities, &pdscp->hmap_node);
1414 pdscp = xmalloc(sizeof *pdscp);
1415 pdscp->priority = priority;
1417 ofproto->need_revalidate = true;
1420 if (pdscp->dscp != dscp) {
1422 ofproto->need_revalidate = true;
1425 hmap_insert(&new, &pdscp->hmap_node, hash_int(pdscp->priority, 0));
1428 if (!hmap_is_empty(&ofport->priorities)) {
1429 ofport_clear_priorities(ofport);
1430 ofproto->need_revalidate = true;
1433 hmap_swap(&new, &ofport->priorities);
1441 /* Expires all MAC learning entries associated with 'bundle' and forces its
1442 * ofproto to revalidate every flow.
1444 * Normally MAC learning entries are removed only from the ofproto associated
1445 * with 'bundle', but if 'all_ofprotos' is true, then the MAC learning entries
1446 * are removed from every ofproto. When patch ports and SLB bonds are in use
1447 * and a VM migration happens and the gratuitous ARPs are somehow lost, this
1448 * avoids a MAC_ENTRY_IDLE_TIME delay before the migrated VM can communicate
1449 * with the host from which it migrated. */
1451 bundle_flush_macs(struct ofbundle *bundle, bool all_ofprotos)
1453 struct ofproto_dpif *ofproto = bundle->ofproto;
1454 struct mac_learning *ml = ofproto->ml;
1455 struct mac_entry *mac, *next_mac;
1457 ofproto->need_revalidate = true;
1458 LIST_FOR_EACH_SAFE (mac, next_mac, lru_node, &ml->lrus) {
1459 if (mac->port.p == bundle) {
1461 struct ofproto_dpif *o;
1463 HMAP_FOR_EACH (o, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
1465 struct mac_entry *e;
1467 e = mac_learning_lookup(o->ml, mac->mac, mac->vlan,
1470 tag_set_add(&o->revalidate_set, e->tag);
1471 mac_learning_expire(o->ml, e);
1477 mac_learning_expire(ml, mac);
1482 static struct ofbundle *
1483 bundle_lookup(const struct ofproto_dpif *ofproto, void *aux)
1485 struct ofbundle *bundle;
1487 HMAP_FOR_EACH_IN_BUCKET (bundle, hmap_node, hash_pointer(aux, 0),
1488 &ofproto->bundles) {
1489 if (bundle->aux == aux) {
1496 /* Looks up each of the 'n_auxes' pointers in 'auxes' as bundles and adds the
1497 * ones that are found to 'bundles'. */
1499 bundle_lookup_multiple(struct ofproto_dpif *ofproto,
1500 void **auxes, size_t n_auxes,
1501 struct hmapx *bundles)
1505 hmapx_init(bundles);
1506 for (i = 0; i < n_auxes; i++) {
1507 struct ofbundle *bundle = bundle_lookup(ofproto, auxes[i]);
1509 hmapx_add(bundles, bundle);
1515 bundle_update(struct ofbundle *bundle)
1517 struct ofport_dpif *port;
1519 bundle->floodable = true;
1520 LIST_FOR_EACH (port, bundle_node, &bundle->ports) {
1521 if (port->up.pp.config & OFPUTIL_PC_NO_FLOOD
1522 || !stp_forward_in_state(port->stp_state)) {
1523 bundle->floodable = false;
1530 bundle_del_port(struct ofport_dpif *port)
1532 struct ofbundle *bundle = port->bundle;
1534 bundle->ofproto->need_revalidate = true;
1536 list_remove(&port->bundle_node);
1537 port->bundle = NULL;
1540 lacp_slave_unregister(bundle->lacp, port);
1543 bond_slave_unregister(bundle->bond, port);
1546 bundle_update(bundle);
1550 bundle_add_port(struct ofbundle *bundle, uint32_t ofp_port,
1551 struct lacp_slave_settings *lacp,
1552 uint32_t bond_stable_id)
1554 struct ofport_dpif *port;
1556 port = get_ofp_port(bundle->ofproto, ofp_port);
1561 if (port->bundle != bundle) {
1562 bundle->ofproto->need_revalidate = true;
1564 bundle_del_port(port);
1567 port->bundle = bundle;
1568 list_push_back(&bundle->ports, &port->bundle_node);
1569 if (port->up.pp.config & OFPUTIL_PC_NO_FLOOD
1570 || !stp_forward_in_state(port->stp_state)) {
1571 bundle->floodable = false;
1575 port->bundle->ofproto->need_revalidate = true;
1576 lacp_slave_register(bundle->lacp, port, lacp);
1579 port->bond_stable_id = bond_stable_id;
1585 bundle_destroy(struct ofbundle *bundle)
1587 struct ofproto_dpif *ofproto;
1588 struct ofport_dpif *port, *next_port;
1595 ofproto = bundle->ofproto;
1596 for (i = 0; i < MAX_MIRRORS; i++) {
1597 struct ofmirror *m = ofproto->mirrors[i];
1599 if (m->out == bundle) {
1601 } else if (hmapx_find_and_delete(&m->srcs, bundle)
1602 || hmapx_find_and_delete(&m->dsts, bundle)) {
1603 ofproto->need_revalidate = true;
1608 LIST_FOR_EACH_SAFE (port, next_port, bundle_node, &bundle->ports) {
1609 bundle_del_port(port);
1612 bundle_flush_macs(bundle, true);
1613 hmap_remove(&ofproto->bundles, &bundle->hmap_node);
1615 free(bundle->trunks);
1616 lacp_destroy(bundle->lacp);
1617 bond_destroy(bundle->bond);
1622 bundle_set(struct ofproto *ofproto_, void *aux,
1623 const struct ofproto_bundle_settings *s)
1625 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1626 bool need_flush = false;
1627 struct ofport_dpif *port;
1628 struct ofbundle *bundle;
1629 unsigned long *trunks;
1635 bundle_destroy(bundle_lookup(ofproto, aux));
1639 assert(s->n_slaves == 1 || s->bond != NULL);
1640 assert((s->lacp != NULL) == (s->lacp_slaves != NULL));
1642 bundle = bundle_lookup(ofproto, aux);
1644 bundle = xmalloc(sizeof *bundle);
1646 bundle->ofproto = ofproto;
1647 hmap_insert(&ofproto->bundles, &bundle->hmap_node,
1648 hash_pointer(aux, 0));
1650 bundle->name = NULL;
1652 list_init(&bundle->ports);
1653 bundle->vlan_mode = PORT_VLAN_TRUNK;
1655 bundle->trunks = NULL;
1656 bundle->use_priority_tags = s->use_priority_tags;
1657 bundle->lacp = NULL;
1658 bundle->bond = NULL;
1660 bundle->floodable = true;
1662 bundle->src_mirrors = 0;
1663 bundle->dst_mirrors = 0;
1664 bundle->mirror_out = 0;
1667 if (!bundle->name || strcmp(s->name, bundle->name)) {
1669 bundle->name = xstrdup(s->name);
1674 if (!bundle->lacp) {
1675 ofproto->need_revalidate = true;
1676 bundle->lacp = lacp_create();
1678 lacp_configure(bundle->lacp, s->lacp);
1680 lacp_destroy(bundle->lacp);
1681 bundle->lacp = NULL;
1684 /* Update set of ports. */
1686 for (i = 0; i < s->n_slaves; i++) {
1687 if (!bundle_add_port(bundle, s->slaves[i],
1688 s->lacp ? &s->lacp_slaves[i] : NULL,
1689 s->bond_stable_ids ? s->bond_stable_ids[i] : 0)) {
1693 if (!ok || list_size(&bundle->ports) != s->n_slaves) {
1694 struct ofport_dpif *next_port;
1696 LIST_FOR_EACH_SAFE (port, next_port, bundle_node, &bundle->ports) {
1697 for (i = 0; i < s->n_slaves; i++) {
1698 if (s->slaves[i] == port->up.ofp_port) {
1703 bundle_del_port(port);
1707 assert(list_size(&bundle->ports) <= s->n_slaves);
1709 if (list_is_empty(&bundle->ports)) {
1710 bundle_destroy(bundle);
1714 /* Set VLAN tagging mode */
1715 if (s->vlan_mode != bundle->vlan_mode
1716 || s->use_priority_tags != bundle->use_priority_tags) {
1717 bundle->vlan_mode = s->vlan_mode;
1718 bundle->use_priority_tags = s->use_priority_tags;
1723 vlan = (s->vlan_mode == PORT_VLAN_TRUNK ? -1
1724 : s->vlan >= 0 && s->vlan <= 4095 ? s->vlan
1726 if (vlan != bundle->vlan) {
1727 bundle->vlan = vlan;
1731 /* Get trunked VLANs. */
1732 switch (s->vlan_mode) {
1733 case PORT_VLAN_ACCESS:
1737 case PORT_VLAN_TRUNK:
1738 trunks = (unsigned long *) s->trunks;
1741 case PORT_VLAN_NATIVE_UNTAGGED:
1742 case PORT_VLAN_NATIVE_TAGGED:
1743 if (vlan != 0 && (!s->trunks
1744 || !bitmap_is_set(s->trunks, vlan)
1745 || bitmap_is_set(s->trunks, 0))) {
1746 /* Force trunking the native VLAN and prohibit trunking VLAN 0. */
1748 trunks = bitmap_clone(s->trunks, 4096);
1750 trunks = bitmap_allocate1(4096);
1752 bitmap_set1(trunks, vlan);
1753 bitmap_set0(trunks, 0);
1755 trunks = (unsigned long *) s->trunks;
1762 if (!vlan_bitmap_equal(trunks, bundle->trunks)) {
1763 free(bundle->trunks);
1764 if (trunks == s->trunks) {
1765 bundle->trunks = vlan_bitmap_clone(trunks);
1767 bundle->trunks = trunks;
1772 if (trunks != s->trunks) {
1777 if (!list_is_short(&bundle->ports)) {
1778 bundle->ofproto->has_bonded_bundles = true;
1780 if (bond_reconfigure(bundle->bond, s->bond)) {
1781 ofproto->need_revalidate = true;
1784 bundle->bond = bond_create(s->bond);
1785 ofproto->need_revalidate = true;
1788 LIST_FOR_EACH (port, bundle_node, &bundle->ports) {
1789 bond_slave_register(bundle->bond, port, port->bond_stable_id,
1793 bond_destroy(bundle->bond);
1794 bundle->bond = NULL;
1797 /* If we changed something that would affect MAC learning, un-learn
1798 * everything on this port and force flow revalidation. */
1800 bundle_flush_macs(bundle, false);
1807 bundle_remove(struct ofport *port_)
1809 struct ofport_dpif *port = ofport_dpif_cast(port_);
1810 struct ofbundle *bundle = port->bundle;
1813 bundle_del_port(port);
1814 if (list_is_empty(&bundle->ports)) {
1815 bundle_destroy(bundle);
1816 } else if (list_is_short(&bundle->ports)) {
1817 bond_destroy(bundle->bond);
1818 bundle->bond = NULL;
1824 send_pdu_cb(void *port_, const void *pdu, size_t pdu_size)
1826 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 10);
1827 struct ofport_dpif *port = port_;
1828 uint8_t ea[ETH_ADDR_LEN];
1831 error = netdev_get_etheraddr(port->up.netdev, ea);
1833 struct ofpbuf packet;
1836 ofpbuf_init(&packet, 0);
1837 packet_pdu = eth_compose(&packet, eth_addr_lacp, ea, ETH_TYPE_LACP,
1839 memcpy(packet_pdu, pdu, pdu_size);
1841 send_packet(port, &packet);
1842 ofpbuf_uninit(&packet);
1844 VLOG_ERR_RL(&rl, "port %s: cannot obtain Ethernet address of iface "
1845 "%s (%s)", port->bundle->name,
1846 netdev_get_name(port->up.netdev), strerror(error));
1851 bundle_send_learning_packets(struct ofbundle *bundle)
1853 struct ofproto_dpif *ofproto = bundle->ofproto;
1854 int error, n_packets, n_errors;
1855 struct mac_entry *e;
1857 error = n_packets = n_errors = 0;
1858 LIST_FOR_EACH (e, lru_node, &ofproto->ml->lrus) {
1859 if (e->port.p != bundle) {
1860 struct ofpbuf *learning_packet;
1861 struct ofport_dpif *port;
1865 /* The assignment to "port" is unnecessary but makes "grep"ing for
1866 * struct ofport_dpif more effective. */
1867 learning_packet = bond_compose_learning_packet(bundle->bond,
1871 ret = send_packet(port, learning_packet);
1872 ofpbuf_delete(learning_packet);
1882 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
1883 VLOG_WARN_RL(&rl, "bond %s: %d errors sending %d gratuitous learning "
1884 "packets, last error was: %s",
1885 bundle->name, n_errors, n_packets, strerror(error));
1887 VLOG_DBG("bond %s: sent %d gratuitous learning packets",
1888 bundle->name, n_packets);
1893 bundle_run(struct ofbundle *bundle)
1896 lacp_run(bundle->lacp, send_pdu_cb);
1899 struct ofport_dpif *port;
1901 LIST_FOR_EACH (port, bundle_node, &bundle->ports) {
1902 bond_slave_set_may_enable(bundle->bond, port, port->may_enable);
1905 bond_run(bundle->bond, &bundle->ofproto->revalidate_set,
1906 lacp_status(bundle->lacp));
1907 if (bond_should_send_learning_packets(bundle->bond)) {
1908 bundle_send_learning_packets(bundle);
1914 bundle_wait(struct ofbundle *bundle)
1917 lacp_wait(bundle->lacp);
1920 bond_wait(bundle->bond);
1927 mirror_scan(struct ofproto_dpif *ofproto)
1931 for (idx = 0; idx < MAX_MIRRORS; idx++) {
1932 if (!ofproto->mirrors[idx]) {
1939 static struct ofmirror *
1940 mirror_lookup(struct ofproto_dpif *ofproto, void *aux)
1944 for (i = 0; i < MAX_MIRRORS; i++) {
1945 struct ofmirror *mirror = ofproto->mirrors[i];
1946 if (mirror && mirror->aux == aux) {
1954 /* Update the 'dup_mirrors' member of each of the ofmirrors in 'ofproto'. */
1956 mirror_update_dups(struct ofproto_dpif *ofproto)
1960 for (i = 0; i < MAX_MIRRORS; i++) {
1961 struct ofmirror *m = ofproto->mirrors[i];
1964 m->dup_mirrors = MIRROR_MASK_C(1) << i;
1968 for (i = 0; i < MAX_MIRRORS; i++) {
1969 struct ofmirror *m1 = ofproto->mirrors[i];
1976 for (j = i + 1; j < MAX_MIRRORS; j++) {
1977 struct ofmirror *m2 = ofproto->mirrors[j];
1979 if (m2 && m1->out == m2->out && m1->out_vlan == m2->out_vlan) {
1980 m1->dup_mirrors |= MIRROR_MASK_C(1) << j;
1981 m2->dup_mirrors |= m1->dup_mirrors;
1988 mirror_set(struct ofproto *ofproto_, void *aux,
1989 const struct ofproto_mirror_settings *s)
1991 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1992 mirror_mask_t mirror_bit;
1993 struct ofbundle *bundle;
1994 struct ofmirror *mirror;
1995 struct ofbundle *out;
1996 struct hmapx srcs; /* Contains "struct ofbundle *"s. */
1997 struct hmapx dsts; /* Contains "struct ofbundle *"s. */
2000 mirror = mirror_lookup(ofproto, aux);
2002 mirror_destroy(mirror);
2008 idx = mirror_scan(ofproto);
2010 VLOG_WARN("bridge %s: maximum of %d port mirrors reached, "
2012 ofproto->up.name, MAX_MIRRORS, s->name);
2016 mirror = ofproto->mirrors[idx] = xzalloc(sizeof *mirror);
2017 mirror->ofproto = ofproto;
2020 mirror->out_vlan = -1;
2021 mirror->name = NULL;
2024 if (!mirror->name || strcmp(s->name, mirror->name)) {
2026 mirror->name = xstrdup(s->name);
2029 /* Get the new configuration. */
2030 if (s->out_bundle) {
2031 out = bundle_lookup(ofproto, s->out_bundle);
2033 mirror_destroy(mirror);
2039 out_vlan = s->out_vlan;
2041 bundle_lookup_multiple(ofproto, s->srcs, s->n_srcs, &srcs);
2042 bundle_lookup_multiple(ofproto, s->dsts, s->n_dsts, &dsts);
2044 /* If the configuration has not changed, do nothing. */
2045 if (hmapx_equals(&srcs, &mirror->srcs)
2046 && hmapx_equals(&dsts, &mirror->dsts)
2047 && vlan_bitmap_equal(mirror->vlans, s->src_vlans)
2048 && mirror->out == out
2049 && mirror->out_vlan == out_vlan)
2051 hmapx_destroy(&srcs);
2052 hmapx_destroy(&dsts);
2056 hmapx_swap(&srcs, &mirror->srcs);
2057 hmapx_destroy(&srcs);
2059 hmapx_swap(&dsts, &mirror->dsts);
2060 hmapx_destroy(&dsts);
2062 free(mirror->vlans);
2063 mirror->vlans = vlan_bitmap_clone(s->src_vlans);
2066 mirror->out_vlan = out_vlan;
2068 /* Update bundles. */
2069 mirror_bit = MIRROR_MASK_C(1) << mirror->idx;
2070 HMAP_FOR_EACH (bundle, hmap_node, &mirror->ofproto->bundles) {
2071 if (hmapx_contains(&mirror->srcs, bundle)) {
2072 bundle->src_mirrors |= mirror_bit;
2074 bundle->src_mirrors &= ~mirror_bit;
2077 if (hmapx_contains(&mirror->dsts, bundle)) {
2078 bundle->dst_mirrors |= mirror_bit;
2080 bundle->dst_mirrors &= ~mirror_bit;
2083 if (mirror->out == bundle) {
2084 bundle->mirror_out |= mirror_bit;
2086 bundle->mirror_out &= ~mirror_bit;
2090 ofproto->need_revalidate = true;
2091 mac_learning_flush(ofproto->ml, &ofproto->revalidate_set);
2092 mirror_update_dups(ofproto);
2098 mirror_destroy(struct ofmirror *mirror)
2100 struct ofproto_dpif *ofproto;
2101 mirror_mask_t mirror_bit;
2102 struct ofbundle *bundle;
2108 ofproto = mirror->ofproto;
2109 ofproto->need_revalidate = true;
2110 mac_learning_flush(ofproto->ml, &ofproto->revalidate_set);
2112 mirror_bit = MIRROR_MASK_C(1) << mirror->idx;
2113 HMAP_FOR_EACH (bundle, hmap_node, &ofproto->bundles) {
2114 bundle->src_mirrors &= ~mirror_bit;
2115 bundle->dst_mirrors &= ~mirror_bit;
2116 bundle->mirror_out &= ~mirror_bit;
2119 hmapx_destroy(&mirror->srcs);
2120 hmapx_destroy(&mirror->dsts);
2121 free(mirror->vlans);
2123 ofproto->mirrors[mirror->idx] = NULL;
2127 mirror_update_dups(ofproto);
2131 mirror_get_stats(struct ofproto *ofproto_, void *aux,
2132 uint64_t *packets, uint64_t *bytes)
2134 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
2135 struct ofmirror *mirror = mirror_lookup(ofproto, aux);
2138 *packets = *bytes = UINT64_MAX;
2142 *packets = mirror->packet_count;
2143 *bytes = mirror->byte_count;
2149 set_flood_vlans(struct ofproto *ofproto_, unsigned long *flood_vlans)
2151 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
2152 if (mac_learning_set_flood_vlans(ofproto->ml, flood_vlans)) {
2153 mac_learning_flush(ofproto->ml, &ofproto->revalidate_set);
2159 is_mirror_output_bundle(const struct ofproto *ofproto_, void *aux)
2161 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
2162 struct ofbundle *bundle = bundle_lookup(ofproto, aux);
2163 return bundle && bundle->mirror_out != 0;
2167 forward_bpdu_changed(struct ofproto *ofproto_)
2169 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
2170 /* Revalidate cached flows whenever forward_bpdu option changes. */
2171 ofproto->need_revalidate = true;
2175 set_mac_idle_time(struct ofproto *ofproto_, unsigned int idle_time)
2177 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
2178 mac_learning_set_idle_time(ofproto->ml, idle_time);
2183 static struct ofport_dpif *
2184 get_ofp_port(struct ofproto_dpif *ofproto, uint16_t ofp_port)
2186 struct ofport *ofport = ofproto_get_port(&ofproto->up, ofp_port);
2187 return ofport ? ofport_dpif_cast(ofport) : NULL;
2190 static struct ofport_dpif *
2191 get_odp_port(struct ofproto_dpif *ofproto, uint32_t odp_port)
2193 return get_ofp_port(ofproto, odp_port_to_ofp_port(odp_port));
2197 ofproto_port_from_dpif_port(struct ofproto_port *ofproto_port,
2198 struct dpif_port *dpif_port)
2200 ofproto_port->name = dpif_port->name;
2201 ofproto_port->type = dpif_port->type;
2202 ofproto_port->ofp_port = odp_port_to_ofp_port(dpif_port->port_no);
2206 port_run(struct ofport_dpif *ofport)
2208 long long int carrier_seq = netdev_get_carrier_resets(ofport->up.netdev);
2209 bool carrier_changed = carrier_seq != ofport->carrier_seq;
2210 bool enable = netdev_get_carrier(ofport->up.netdev);
2212 ofport->carrier_seq = carrier_seq;
2215 cfm_run(ofport->cfm);
2217 if (cfm_should_send_ccm(ofport->cfm)) {
2218 struct ofpbuf packet;
2220 ofpbuf_init(&packet, 0);
2221 cfm_compose_ccm(ofport->cfm, &packet, ofport->up.pp.hw_addr);
2222 send_packet(ofport, &packet);
2223 ofpbuf_uninit(&packet);
2226 enable = enable && !cfm_get_fault(ofport->cfm)
2227 && cfm_get_opup(ofport->cfm);
2230 if (ofport->bundle) {
2231 enable = enable && lacp_slave_may_enable(ofport->bundle->lacp, ofport);
2232 if (carrier_changed) {
2233 lacp_slave_carrier_changed(ofport->bundle->lacp, ofport);
2237 if (ofport->may_enable != enable) {
2238 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport->up.ofproto);
2240 if (ofproto->has_bundle_action) {
2241 ofproto->need_revalidate = true;
2245 ofport->may_enable = enable;
2249 port_wait(struct ofport_dpif *ofport)
2252 cfm_wait(ofport->cfm);
2257 port_query_by_name(const struct ofproto *ofproto_, const char *devname,
2258 struct ofproto_port *ofproto_port)
2260 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
2261 struct dpif_port dpif_port;
2264 error = dpif_port_query_by_name(ofproto->dpif, devname, &dpif_port);
2266 ofproto_port_from_dpif_port(ofproto_port, &dpif_port);
2272 port_add(struct ofproto *ofproto_, struct netdev *netdev, uint16_t *ofp_portp)
2274 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
2278 error = dpif_port_add(ofproto->dpif, netdev, &odp_port);
2280 *ofp_portp = odp_port_to_ofp_port(odp_port);
2286 port_del(struct ofproto *ofproto_, uint16_t ofp_port)
2288 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
2291 error = dpif_port_del(ofproto->dpif, ofp_port_to_odp_port(ofp_port));
2293 struct ofport_dpif *ofport = get_ofp_port(ofproto, ofp_port);
2295 /* The caller is going to close ofport->up.netdev. If this is a
2296 * bonded port, then the bond is using that netdev, so remove it
2297 * from the bond. The client will need to reconfigure everything
2298 * after deleting ports, so then the slave will get re-added. */
2299 bundle_remove(&ofport->up);
2306 port_get_stats(const struct ofport *ofport_, struct netdev_stats *stats)
2308 struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
2311 error = netdev_get_stats(ofport->up.netdev, stats);
2313 if (!error && ofport->odp_port == OVSP_LOCAL) {
2314 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport->up.ofproto);
2316 /* ofproto->stats.tx_packets represents packets that we created
2317 * internally and sent to some port (e.g. packets sent with
2318 * send_packet()). Account for them as if they had come from
2319 * OFPP_LOCAL and got forwarded. */
2321 if (stats->rx_packets != UINT64_MAX) {
2322 stats->rx_packets += ofproto->stats.tx_packets;
2325 if (stats->rx_bytes != UINT64_MAX) {
2326 stats->rx_bytes += ofproto->stats.tx_bytes;
2329 /* ofproto->stats.rx_packets represents packets that were received on
2330 * some port and we processed internally and dropped (e.g. STP).
2331 * Account fro them as if they had been forwarded to OFPP_LOCAL. */
2333 if (stats->tx_packets != UINT64_MAX) {
2334 stats->tx_packets += ofproto->stats.rx_packets;
2337 if (stats->tx_bytes != UINT64_MAX) {
2338 stats->tx_bytes += ofproto->stats.rx_bytes;
2345 /* Account packets for LOCAL port. */
2347 ofproto_update_local_port_stats(const struct ofproto *ofproto_,
2348 size_t tx_size, size_t rx_size)
2350 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
2353 ofproto->stats.rx_packets++;
2354 ofproto->stats.rx_bytes += rx_size;
2357 ofproto->stats.tx_packets++;
2358 ofproto->stats.tx_bytes += tx_size;
2362 struct port_dump_state {
2363 struct dpif_port_dump dump;
2368 port_dump_start(const struct ofproto *ofproto_, void **statep)
2370 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
2371 struct port_dump_state *state;
2373 *statep = state = xmalloc(sizeof *state);
2374 dpif_port_dump_start(&state->dump, ofproto->dpif);
2375 state->done = false;
2380 port_dump_next(const struct ofproto *ofproto_ OVS_UNUSED, void *state_,
2381 struct ofproto_port *port)
2383 struct port_dump_state *state = state_;
2384 struct dpif_port dpif_port;
2386 if (dpif_port_dump_next(&state->dump, &dpif_port)) {
2387 ofproto_port_from_dpif_port(port, &dpif_port);
2390 int error = dpif_port_dump_done(&state->dump);
2392 return error ? error : EOF;
2397 port_dump_done(const struct ofproto *ofproto_ OVS_UNUSED, void *state_)
2399 struct port_dump_state *state = state_;
2402 dpif_port_dump_done(&state->dump);
2409 port_poll(const struct ofproto *ofproto_, char **devnamep)
2411 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
2412 return dpif_port_poll(ofproto->dpif, devnamep);
2416 port_poll_wait(const struct ofproto *ofproto_)
2418 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
2419 dpif_port_poll_wait(ofproto->dpif);
2423 port_is_lacp_current(const struct ofport *ofport_)
2425 const struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
2426 return (ofport->bundle && ofport->bundle->lacp
2427 ? lacp_slave_is_current(ofport->bundle->lacp, ofport)
2431 /* Upcall handling. */
2433 /* Flow miss batching.
2435 * Some dpifs implement operations faster when you hand them off in a batch.
2436 * To allow batching, "struct flow_miss" queues the dpif-related work needed
2437 * for a given flow. Each "struct flow_miss" corresponds to sending one or
2438 * more packets, plus possibly installing the flow in the dpif.
2440 * So far we only batch the operations that affect flow setup time the most.
2441 * It's possible to batch more than that, but the benefit might be minimal. */
2443 struct hmap_node hmap_node;
2445 enum odp_key_fitness key_fitness;
2446 const struct nlattr *key;
2448 ovs_be16 initial_tci;
2449 struct list packets;
2452 struct flow_miss_op {
2453 struct dpif_op dpif_op;
2454 struct subfacet *subfacet;
2457 /* Sends an OFPT_PACKET_IN message for 'packet' of type OFPR_NO_MATCH to each
2458 * OpenFlow controller as necessary according to their individual
2459 * configurations. */
2461 send_packet_in_miss(struct ofproto_dpif *ofproto, const struct ofpbuf *packet,
2462 const struct flow *flow)
2464 struct ofputil_packet_in pin;
2466 pin.packet = packet->data;
2467 pin.packet_len = packet->size;
2468 pin.total_len = packet->size;
2469 pin.reason = OFPR_NO_MATCH;
2470 pin.controller_id = 0;
2475 pin.buffer_id = 0; /* not yet known */
2476 pin.send_len = 0; /* not used for flow table misses */
2478 flow_get_metadata(flow, &pin.fmd);
2480 /* Registers aren't meaningful on a miss. */
2481 memset(pin.fmd.reg_masks, 0, sizeof pin.fmd.reg_masks);
2483 connmgr_send_packet_in(ofproto->up.connmgr, &pin, flow);
2487 process_special(struct ofproto_dpif *ofproto, const struct flow *flow,
2488 const struct ofpbuf *packet)
2490 struct ofport_dpif *ofport = get_ofp_port(ofproto, flow->in_port);
2496 if (ofport->cfm && cfm_should_process_flow(ofport->cfm, flow)) {
2498 cfm_process_heartbeat(ofport->cfm, packet);
2501 } else if (ofport->bundle && ofport->bundle->lacp
2502 && flow->dl_type == htons(ETH_TYPE_LACP)) {
2504 lacp_process_packet(ofport->bundle->lacp, ofport, packet);
2507 } else if (ofproto->stp && stp_should_process_flow(flow)) {
2509 stp_process_packet(ofport, packet);
2516 static struct flow_miss *
2517 flow_miss_create(struct hmap *todo, const struct flow *flow,
2518 enum odp_key_fitness key_fitness,
2519 const struct nlattr *key, size_t key_len,
2520 ovs_be16 initial_tci)
2522 uint32_t hash = flow_hash(flow, 0);
2523 struct flow_miss *miss;
2525 HMAP_FOR_EACH_WITH_HASH (miss, hmap_node, hash, todo) {
2526 if (flow_equal(&miss->flow, flow)) {
2531 miss = xmalloc(sizeof *miss);
2532 hmap_insert(todo, &miss->hmap_node, hash);
2534 miss->key_fitness = key_fitness;
2536 miss->key_len = key_len;
2537 miss->initial_tci = initial_tci;
2538 list_init(&miss->packets);
2543 handle_flow_miss(struct ofproto_dpif *ofproto, struct flow_miss *miss,
2544 struct flow_miss_op *ops, size_t *n_ops)
2546 const struct flow *flow = &miss->flow;
2547 struct ofpbuf *packet, *next_packet;
2548 struct subfacet *subfacet;
2549 struct facet *facet;
2551 facet = facet_lookup_valid(ofproto, flow);
2553 struct rule_dpif *rule;
2555 rule = rule_dpif_lookup(ofproto, flow, 0);
2557 /* Don't send a packet-in if OFPUTIL_PC_NO_PACKET_IN asserted. */
2558 struct ofport_dpif *port = get_ofp_port(ofproto, flow->in_port);
2560 if (port->up.pp.config & OFPUTIL_PC_NO_PACKET_IN) {
2561 COVERAGE_INC(ofproto_dpif_no_packet_in);
2562 /* XXX install 'drop' flow entry */
2566 VLOG_WARN_RL(&rl, "packet-in on unknown port %"PRIu16,
2570 LIST_FOR_EACH (packet, list_node, &miss->packets) {
2571 send_packet_in_miss(ofproto, packet, flow);
2577 facet = facet_create(rule, flow);
2580 subfacet = subfacet_create(facet,
2581 miss->key_fitness, miss->key, miss->key_len,
2584 LIST_FOR_EACH_SAFE (packet, next_packet, list_node, &miss->packets) {
2585 struct dpif_flow_stats stats;
2586 struct flow_miss_op *op;
2587 struct dpif_execute *execute;
2589 ofproto->n_matches++;
2591 if (facet->rule->up.cr.priority == FAIL_OPEN_PRIORITY) {
2593 * Extra-special case for fail-open mode.
2595 * We are in fail-open mode and the packet matched the fail-open
2596 * rule, but we are connected to a controller too. We should send
2597 * the packet up to the controller in the hope that it will try to
2598 * set up a flow and thereby allow us to exit fail-open.
2600 * See the top-level comment in fail-open.c for more information.
2602 send_packet_in_miss(ofproto, packet, flow);
2605 if (!facet->may_install || !subfacet->actions) {
2606 subfacet_make_actions(subfacet, packet);
2609 dpif_flow_stats_extract(&facet->flow, packet, &stats);
2610 subfacet_update_stats(subfacet, &stats);
2612 if (!subfacet->actions_len) {
2613 /* No actions to execute, so skip talking to the dpif. */
2617 if (flow->vlan_tci != subfacet->initial_tci) {
2618 /* This packet was received on a VLAN splinter port. We added
2619 * a VLAN to the packet to make the packet resemble the flow,
2620 * but the actions were composed assuming that the packet
2621 * contained no VLAN. So, we must remove the VLAN header from
2622 * the packet before trying to execute the actions. */
2623 eth_pop_vlan(packet);
2626 op = &ops[(*n_ops)++];
2627 execute = &op->dpif_op.u.execute;
2628 op->subfacet = subfacet;
2629 op->dpif_op.type = DPIF_OP_EXECUTE;
2630 execute->key = miss->key;
2631 execute->key_len = miss->key_len;
2632 execute->actions = (facet->may_install
2634 : xmemdup(subfacet->actions,
2635 subfacet->actions_len));
2636 execute->actions_len = subfacet->actions_len;
2637 execute->packet = packet;
2640 if (facet->may_install && subfacet->key_fitness != ODP_FIT_TOO_LITTLE) {
2641 struct flow_miss_op *op = &ops[(*n_ops)++];
2642 struct dpif_flow_put *put = &op->dpif_op.u.flow_put;
2644 op->subfacet = subfacet;
2645 op->dpif_op.type = DPIF_OP_FLOW_PUT;
2646 put->flags = DPIF_FP_CREATE | DPIF_FP_MODIFY;
2647 put->key = miss->key;
2648 put->key_len = miss->key_len;
2649 put->actions = subfacet->actions;
2650 put->actions_len = subfacet->actions_len;
2655 /* Like odp_flow_key_to_flow(), this function converts the 'key_len' bytes of
2656 * OVS_KEY_ATTR_* attributes in 'key' to a flow structure in 'flow' and returns
2657 * an ODP_FIT_* value that indicates how well 'key' fits our expectations for
2658 * what a flow key should contain.
2660 * This function also includes some logic to help make VLAN splinters
2661 * transparent to the rest of the upcall processing logic. In particular, if
2662 * the extracted in_port is a VLAN splinter port, it replaces flow->in_port by
2663 * the "real" port, sets flow->vlan_tci correctly for the VLAN of the VLAN
2664 * splinter port, and pushes a VLAN header onto 'packet' (if it is nonnull).
2666 * Sets '*initial_tci' to the VLAN TCI with which the packet was really
2667 * received, that is, the actual VLAN TCI extracted by odp_flow_key_to_flow().
2668 * (This differs from the value returned in flow->vlan_tci only for packets
2669 * received on VLAN splinters.)
2671 static enum odp_key_fitness
2672 ofproto_dpif_extract_flow_key(const struct ofproto_dpif *ofproto,
2673 const struct nlattr *key, size_t key_len,
2674 struct flow *flow, ovs_be16 *initial_tci,
2675 struct ofpbuf *packet)
2677 enum odp_key_fitness fitness;
2681 fitness = odp_flow_key_to_flow(key, key_len, flow);
2682 if (fitness == ODP_FIT_ERROR) {
2685 *initial_tci = flow->vlan_tci;
2687 realdev = vsp_vlandev_to_realdev(ofproto, flow->in_port, &vid);
2689 /* Cause the flow to be processed as if it came in on the real device
2690 * with the VLAN device's VLAN ID. */
2691 flow->in_port = realdev;
2692 flow->vlan_tci = htons((vid & VLAN_VID_MASK) | VLAN_CFI);
2694 /* Make the packet resemble the flow, so that it gets sent to an
2695 * OpenFlow controller properly, so that it looks correct for
2696 * sFlow, and so that flow_extract() will get the correct vlan_tci
2697 * if it is called on 'packet'.
2699 * The allocated space inside 'packet' probably also contains
2700 * 'key', that is, both 'packet' and 'key' are probably part of a
2701 * struct dpif_upcall (see the large comment on that structure
2702 * definition), so pushing data on 'packet' is in general not a
2703 * good idea since it could overwrite 'key' or free it as a side
2704 * effect. However, it's OK in this special case because we know
2705 * that 'packet' is inside a Netlink attribute: pushing 4 bytes
2706 * will just overwrite the 4-byte "struct nlattr", which is fine
2707 * since we don't need that header anymore. */
2708 eth_push_vlan(packet, flow->vlan_tci);
2711 /* Let the caller know that we can't reproduce 'key' from 'flow'. */
2712 if (fitness == ODP_FIT_PERFECT) {
2713 fitness = ODP_FIT_TOO_MUCH;
2721 handle_miss_upcalls(struct ofproto_dpif *ofproto, struct dpif_upcall *upcalls,
2724 struct dpif_upcall *upcall;
2725 struct flow_miss *miss, *next_miss;
2726 struct flow_miss_op flow_miss_ops[FLOW_MISS_MAX_BATCH * 2];
2727 struct dpif_op *dpif_ops[FLOW_MISS_MAX_BATCH * 2];
2736 /* Construct the to-do list.
2738 * This just amounts to extracting the flow from each packet and sticking
2739 * the packets that have the same flow in the same "flow_miss" structure so
2740 * that we can process them together. */
2742 for (upcall = upcalls; upcall < &upcalls[n_upcalls]; upcall++) {
2743 enum odp_key_fitness fitness;
2744 struct flow_miss *miss;
2745 ovs_be16 initial_tci;
2748 /* Obtain metadata and check userspace/kernel agreement on flow match,
2749 * then set 'flow''s header pointers. */
2750 fitness = ofproto_dpif_extract_flow_key(ofproto,
2751 upcall->key, upcall->key_len,
2752 &flow, &initial_tci,
2754 if (fitness == ODP_FIT_ERROR) {
2755 ofpbuf_delete(upcall->packet);
2758 flow_extract(upcall->packet, flow.skb_priority, flow.tun_id,
2759 flow.in_port, &flow);
2761 /* Handle 802.1ag, LACP, and STP specially. */
2762 if (process_special(ofproto, &flow, upcall->packet)) {
2763 ofproto_update_local_port_stats(&ofproto->up,
2764 0, upcall->packet->size);
2765 ofpbuf_delete(upcall->packet);
2766 ofproto->n_matches++;
2770 /* Add other packets to a to-do list. */
2771 miss = flow_miss_create(&todo, &flow, fitness,
2772 upcall->key, upcall->key_len, initial_tci);
2773 list_push_back(&miss->packets, &upcall->packet->list_node);
2776 /* Process each element in the to-do list, constructing the set of
2777 * operations to batch. */
2779 HMAP_FOR_EACH (miss, hmap_node, &todo) {
2780 handle_flow_miss(ofproto, miss, flow_miss_ops, &n_ops);
2782 assert(n_ops <= ARRAY_SIZE(flow_miss_ops));
2784 /* Execute batch. */
2785 for (i = 0; i < n_ops; i++) {
2786 dpif_ops[i] = &flow_miss_ops[i].dpif_op;
2788 dpif_operate(ofproto->dpif, dpif_ops, n_ops);
2790 /* Free memory and update facets. */
2791 for (i = 0; i < n_ops; i++) {
2792 struct flow_miss_op *op = &flow_miss_ops[i];
2793 struct dpif_execute *execute;
2795 switch (op->dpif_op.type) {
2796 case DPIF_OP_EXECUTE:
2797 execute = &op->dpif_op.u.execute;
2798 if (op->subfacet->actions != execute->actions) {
2799 free((struct nlattr *) execute->actions);
2803 case DPIF_OP_FLOW_PUT:
2804 if (!op->dpif_op.error) {
2805 op->subfacet->installed = true;
2810 HMAP_FOR_EACH_SAFE (miss, next_miss, hmap_node, &todo) {
2811 ofpbuf_list_delete(&miss->packets);
2812 hmap_remove(&todo, &miss->hmap_node);
2815 hmap_destroy(&todo);
2819 handle_userspace_upcall(struct ofproto_dpif *ofproto,
2820 struct dpif_upcall *upcall)
2822 struct user_action_cookie cookie;
2823 enum odp_key_fitness fitness;
2824 ovs_be16 initial_tci;
2827 memcpy(&cookie, &upcall->userdata, sizeof(cookie));
2829 fitness = ofproto_dpif_extract_flow_key(ofproto, upcall->key,
2830 upcall->key_len, &flow,
2831 &initial_tci, upcall->packet);
2832 if (fitness == ODP_FIT_ERROR) {
2833 ofpbuf_delete(upcall->packet);
2837 if (cookie.type == USER_ACTION_COOKIE_SFLOW) {
2838 if (ofproto->sflow) {
2839 dpif_sflow_received(ofproto->sflow, upcall->packet, &flow,
2843 VLOG_WARN_RL(&rl, "invalid user cookie : 0x%"PRIx64, upcall->userdata);
2845 ofpbuf_delete(upcall->packet);
2849 handle_upcalls(struct ofproto_dpif *ofproto, unsigned int max_batch)
2851 struct dpif_upcall misses[FLOW_MISS_MAX_BATCH];
2855 assert (max_batch <= FLOW_MISS_MAX_BATCH);
2858 for (i = 0; i < max_batch; i++) {
2859 struct dpif_upcall *upcall = &misses[n_misses];
2862 error = dpif_recv(ofproto->dpif, upcall);
2867 switch (upcall->type) {
2868 case DPIF_UC_ACTION:
2869 handle_userspace_upcall(ofproto, upcall);
2873 /* Handle it later. */
2877 case DPIF_N_UC_TYPES:
2879 VLOG_WARN_RL(&rl, "upcall has unexpected type %"PRIu32,
2885 handle_miss_upcalls(ofproto, misses, n_misses);
2890 /* Flow expiration. */
2892 static int subfacet_max_idle(const struct ofproto_dpif *);
2893 static void update_stats(struct ofproto_dpif *);
2894 static void rule_expire(struct rule_dpif *);
2895 static void expire_subfacets(struct ofproto_dpif *, int dp_max_idle);
2897 /* This function is called periodically by run(). Its job is to collect
2898 * updates for the flows that have been installed into the datapath, most
2899 * importantly when they last were used, and then use that information to
2900 * expire flows that have not been used recently.
2902 * Returns the number of milliseconds after which it should be called again. */
2904 expire(struct ofproto_dpif *ofproto)
2906 struct rule_dpif *rule, *next_rule;
2907 struct oftable *table;
2910 /* Update stats for each flow in the datapath. */
2911 update_stats(ofproto);
2913 /* Expire subfacets that have been idle too long. */
2914 dp_max_idle = subfacet_max_idle(ofproto);
2915 expire_subfacets(ofproto, dp_max_idle);
2917 /* Expire OpenFlow flows whose idle_timeout or hard_timeout has passed. */
2918 OFPROTO_FOR_EACH_TABLE (table, &ofproto->up) {
2919 struct cls_cursor cursor;
2921 cls_cursor_init(&cursor, &table->cls, NULL);
2922 CLS_CURSOR_FOR_EACH_SAFE (rule, next_rule, up.cr, &cursor) {
2927 /* All outstanding data in existing flows has been accounted, so it's a
2928 * good time to do bond rebalancing. */
2929 if (ofproto->has_bonded_bundles) {
2930 struct ofbundle *bundle;
2932 HMAP_FOR_EACH (bundle, hmap_node, &ofproto->bundles) {
2934 bond_rebalance(bundle->bond, &ofproto->revalidate_set);
2939 return MIN(dp_max_idle, 1000);
2942 /* Update 'packet_count', 'byte_count', and 'used' members of installed facets.
2944 * This function also pushes statistics updates to rules which each facet
2945 * resubmits into. Generally these statistics will be accurate. However, if a
2946 * facet changes the rule it resubmits into at some time in between
2947 * update_stats() runs, it is possible that statistics accrued to the
2948 * old rule will be incorrectly attributed to the new rule. This could be
2949 * avoided by calling update_stats() whenever rules are created or
2950 * deleted. However, the performance impact of making so many calls to the
2951 * datapath do not justify the benefit of having perfectly accurate statistics.
2954 update_stats(struct ofproto_dpif *p)
2956 const struct dpif_flow_stats *stats;
2957 struct dpif_flow_dump dump;
2958 const struct nlattr *key;
2961 dpif_flow_dump_start(&dump, p->dpif);
2962 while (dpif_flow_dump_next(&dump, &key, &key_len, NULL, NULL, &stats)) {
2963 struct subfacet *subfacet;
2965 subfacet = subfacet_find(p, key, key_len);
2966 if (subfacet && subfacet->installed) {
2967 struct facet *facet = subfacet->facet;
2969 if (stats->n_packets >= subfacet->dp_packet_count) {
2970 uint64_t extra = stats->n_packets - subfacet->dp_packet_count;
2971 facet->packet_count += extra;
2973 VLOG_WARN_RL(&rl, "unexpected packet count from the datapath");
2976 if (stats->n_bytes >= subfacet->dp_byte_count) {
2977 facet->byte_count += stats->n_bytes - subfacet->dp_byte_count;
2979 VLOG_WARN_RL(&rl, "unexpected byte count from datapath");
2982 subfacet->dp_packet_count = stats->n_packets;
2983 subfacet->dp_byte_count = stats->n_bytes;
2985 facet->tcp_flags |= stats->tcp_flags;
2987 subfacet_update_time(subfacet, stats->used);
2988 facet_account(facet);
2989 facet_push_stats(facet);
2991 if (!VLOG_DROP_WARN(&rl)) {
2995 odp_flow_key_format(key, key_len, &s);
2996 VLOG_WARN("unexpected flow from datapath %s", ds_cstr(&s));
3000 COVERAGE_INC(facet_unexpected);
3001 /* There's a flow in the datapath that we know nothing about, or a
3002 * flow that shouldn't be installed but was anyway. Delete it. */
3003 dpif_flow_del(p->dpif, key, key_len, NULL);
3006 dpif_flow_dump_done(&dump);
3009 /* Calculates and returns the number of milliseconds of idle time after which
3010 * subfacets should expire from the datapath. When a subfacet expires, we fold
3011 * its statistics into its facet, and when a facet's last subfacet expires, we
3012 * fold its statistic into its rule. */
3014 subfacet_max_idle(const struct ofproto_dpif *ofproto)
3017 * Idle time histogram.
3019 * Most of the time a switch has a relatively small number of subfacets.
3020 * When this is the case we might as well keep statistics for all of them
3021 * in userspace and to cache them in the kernel datapath for performance as
3024 * As the number of subfacets increases, the memory required to maintain
3025 * statistics about them in userspace and in the kernel becomes
3026 * significant. However, with a large number of subfacets it is likely
3027 * that only a few of them are "heavy hitters" that consume a large amount
3028 * of bandwidth. At this point, only heavy hitters are worth caching in
3029 * the kernel and maintaining in userspaces; other subfacets we can
3032 * The technique used to compute the idle time is to build a histogram with
3033 * N_BUCKETS buckets whose width is BUCKET_WIDTH msecs each. Each subfacet
3034 * that is installed in the kernel gets dropped in the appropriate bucket.
3035 * After the histogram has been built, we compute the cutoff so that only
3036 * the most-recently-used 1% of subfacets (but at least
3037 * ofproto->up.flow_eviction_threshold flows) are kept cached. At least
3038 * the most-recently-used bucket of subfacets is kept, so actually an
3039 * arbitrary number of subfacets can be kept in any given expiration run
3040 * (though the next run will delete most of those unless they receive
3043 * This requires a second pass through the subfacets, in addition to the
3044 * pass made by update_stats(), because the former function never looks at
3045 * uninstallable subfacets.
3047 enum { BUCKET_WIDTH = ROUND_UP(100, TIME_UPDATE_INTERVAL) };
3048 enum { N_BUCKETS = 5000 / BUCKET_WIDTH };
3049 int buckets[N_BUCKETS] = { 0 };
3050 int total, subtotal, bucket;
3051 struct subfacet *subfacet;
3055 total = hmap_count(&ofproto->subfacets);
3056 if (total <= ofproto->up.flow_eviction_threshold) {
3057 return N_BUCKETS * BUCKET_WIDTH;
3060 /* Build histogram. */
3062 HMAP_FOR_EACH (subfacet, hmap_node, &ofproto->subfacets) {
3063 long long int idle = now - subfacet->used;
3064 int bucket = (idle <= 0 ? 0
3065 : idle >= BUCKET_WIDTH * N_BUCKETS ? N_BUCKETS - 1
3066 : (unsigned int) idle / BUCKET_WIDTH);
3070 /* Find the first bucket whose flows should be expired. */
3071 subtotal = bucket = 0;
3073 subtotal += buckets[bucket++];
3074 } while (bucket < N_BUCKETS &&
3075 subtotal < MAX(ofproto->up.flow_eviction_threshold, total / 100));
3077 if (VLOG_IS_DBG_ENABLED()) {
3081 ds_put_cstr(&s, "keep");
3082 for (i = 0; i < N_BUCKETS; i++) {
3084 ds_put_cstr(&s, ", drop");
3087 ds_put_format(&s, " %d:%d", i * BUCKET_WIDTH, buckets[i]);
3090 VLOG_INFO("%s: %s (msec:count)", ofproto->up.name, ds_cstr(&s));
3094 return bucket * BUCKET_WIDTH;
3098 expire_subfacets(struct ofproto_dpif *ofproto, int dp_max_idle)
3100 long long int cutoff = time_msec() - dp_max_idle;
3101 struct subfacet *subfacet, *next_subfacet;
3103 HMAP_FOR_EACH_SAFE (subfacet, next_subfacet, hmap_node,
3104 &ofproto->subfacets) {
3105 if (subfacet->used < cutoff) {
3106 subfacet_destroy(subfacet);
3111 /* If 'rule' is an OpenFlow rule, that has expired according to OpenFlow rules,
3112 * then delete it entirely. */
3114 rule_expire(struct rule_dpif *rule)
3116 struct facet *facet, *next_facet;
3120 /* Has 'rule' expired? */
3122 if (rule->up.hard_timeout
3123 && now > rule->up.modified + rule->up.hard_timeout * 1000) {
3124 reason = OFPRR_HARD_TIMEOUT;
3125 } else if (rule->up.idle_timeout
3126 && now > rule->up.used + rule->up.idle_timeout * 1000) {
3127 reason = OFPRR_IDLE_TIMEOUT;
3132 COVERAGE_INC(ofproto_dpif_expired);
3134 /* Update stats. (This is a no-op if the rule expired due to an idle
3135 * timeout, because that only happens when the rule has no facets left.) */
3136 LIST_FOR_EACH_SAFE (facet, next_facet, list_node, &rule->facets) {
3137 facet_remove(facet);
3140 /* Get rid of the rule. */
3141 ofproto_rule_expire(&rule->up, reason);
3146 /* Creates and returns a new facet owned by 'rule', given a 'flow'.
3148 * The caller must already have determined that no facet with an identical
3149 * 'flow' exists in 'ofproto' and that 'flow' is the best match for 'rule' in
3150 * the ofproto's classifier table.
3152 * The facet will initially have no subfacets. The caller should create (at
3153 * least) one subfacet with subfacet_create(). */
3154 static struct facet *
3155 facet_create(struct rule_dpif *rule, const struct flow *flow)
3157 struct ofproto_dpif *ofproto = ofproto_dpif_cast(rule->up.ofproto);
3158 struct facet *facet;
3160 facet = xzalloc(sizeof *facet);
3161 facet->used = time_msec();
3162 hmap_insert(&ofproto->facets, &facet->hmap_node, flow_hash(flow, 0));
3163 list_push_back(&rule->facets, &facet->list_node);
3165 facet->flow = *flow;
3166 list_init(&facet->subfacets);
3167 netflow_flow_init(&facet->nf_flow);
3168 netflow_flow_update_time(ofproto->netflow, &facet->nf_flow, facet->used);
3174 facet_free(struct facet *facet)
3179 /* Executes, within 'ofproto', the 'n_actions' actions in 'actions' on
3180 * 'packet', which arrived on 'in_port'.
3182 * Takes ownership of 'packet'. */
3184 execute_odp_actions(struct ofproto_dpif *ofproto, const struct flow *flow,
3185 const struct nlattr *odp_actions, size_t actions_len,
3186 struct ofpbuf *packet)
3188 struct odputil_keybuf keybuf;
3192 ofpbuf_use_stack(&key, &keybuf, sizeof keybuf);
3193 odp_flow_key_from_flow(&key, flow);
3195 error = dpif_execute(ofproto->dpif, key.data, key.size,
3196 odp_actions, actions_len, packet);
3198 ofpbuf_delete(packet);
3202 /* Remove 'facet' from 'ofproto' and free up the associated memory:
3204 * - If 'facet' was installed in the datapath, uninstalls it and updates its
3205 * rule's statistics, via subfacet_uninstall().
3207 * - Removes 'facet' from its rule and from ofproto->facets.
3210 facet_remove(struct facet *facet)
3212 struct ofproto_dpif *ofproto = ofproto_dpif_cast(facet->rule->up.ofproto);
3213 struct subfacet *subfacet, *next_subfacet;
3215 assert(!list_is_empty(&facet->subfacets));
3217 /* First uninstall all of the subfacets to get final statistics. */
3218 LIST_FOR_EACH (subfacet, list_node, &facet->subfacets) {
3219 subfacet_uninstall(subfacet);
3222 /* Flush the final stats to the rule.
3224 * This might require us to have at least one subfacet around so that we
3225 * can use its actions for accounting in facet_account(), which is why we
3226 * have uninstalled but not yet destroyed the subfacets. */
3227 facet_flush_stats(facet);
3229 /* Now we're really all done so destroy everything. */
3230 LIST_FOR_EACH_SAFE (subfacet, next_subfacet, list_node,
3231 &facet->subfacets) {
3232 subfacet_destroy__(subfacet);
3234 hmap_remove(&ofproto->facets, &facet->hmap_node);
3235 list_remove(&facet->list_node);
3240 facet_account(struct facet *facet)
3242 struct ofproto_dpif *ofproto = ofproto_dpif_cast(facet->rule->up.ofproto);
3244 struct subfacet *subfacet;
3245 const struct nlattr *a;
3249 if (facet->byte_count <= facet->accounted_bytes) {
3252 n_bytes = facet->byte_count - facet->accounted_bytes;
3253 facet->accounted_bytes = facet->byte_count;
3255 /* Feed information from the active flows back into the learning table to
3256 * ensure that table is always in sync with what is actually flowing
3257 * through the datapath. */
3258 if (facet->has_learn || facet->has_normal
3259 || (facet->has_fin_timeout
3260 && facet->tcp_flags & (TCP_FIN | TCP_RST))) {
3261 struct action_xlate_ctx ctx;
3263 action_xlate_ctx_init(&ctx, ofproto, &facet->flow,
3264 facet->flow.vlan_tci,
3265 facet->rule, facet->tcp_flags, NULL);
3266 ctx.may_learn = true;
3267 ofpbuf_delete(xlate_actions(&ctx, facet->rule->up.actions,
3268 facet->rule->up.n_actions));
3271 if (!facet->has_normal || !ofproto->has_bonded_bundles) {
3275 /* This loop feeds byte counters to bond_account() for rebalancing to use
3276 * as a basis. We also need to track the actual VLAN on which the packet
3277 * is going to be sent to ensure that it matches the one passed to
3278 * bond_choose_output_slave(). (Otherwise, we will account to the wrong
3281 * We use the actions from an arbitrary subfacet because they should all
3282 * be equally valid for our purpose. */
3283 subfacet = CONTAINER_OF(list_front(&facet->subfacets),
3284 struct subfacet, list_node);
3285 vlan_tci = facet->flow.vlan_tci;
3286 NL_ATTR_FOR_EACH_UNSAFE (a, left,
3287 subfacet->actions, subfacet->actions_len) {
3288 const struct ovs_action_push_vlan *vlan;
3289 struct ofport_dpif *port;
3291 switch (nl_attr_type(a)) {
3292 case OVS_ACTION_ATTR_OUTPUT:
3293 port = get_odp_port(ofproto, nl_attr_get_u32(a));
3294 if (port && port->bundle && port->bundle->bond) {
3295 bond_account(port->bundle->bond, &facet->flow,
3296 vlan_tci_to_vid(vlan_tci), n_bytes);
3300 case OVS_ACTION_ATTR_POP_VLAN:
3301 vlan_tci = htons(0);
3304 case OVS_ACTION_ATTR_PUSH_VLAN:
3305 vlan = nl_attr_get(a);
3306 vlan_tci = vlan->vlan_tci;
3312 /* Returns true if the only action for 'facet' is to send to the controller.
3313 * (We don't report NetFlow expiration messages for such facets because they
3314 * are just part of the control logic for the network, not real traffic). */
3316 facet_is_controller_flow(struct facet *facet)
3319 && facet->rule->up.n_actions == 1
3320 && action_outputs_to_port(&facet->rule->up.actions[0],
3321 htons(OFPP_CONTROLLER)));
3324 /* Folds all of 'facet''s statistics into its rule. Also updates the
3325 * accounting ofhook and emits a NetFlow expiration if appropriate. All of
3326 * 'facet''s statistics in the datapath should have been zeroed and folded into
3327 * its packet and byte counts before this function is called. */
3329 facet_flush_stats(struct facet *facet)
3331 struct ofproto_dpif *ofproto = ofproto_dpif_cast(facet->rule->up.ofproto);
3332 struct subfacet *subfacet;
3334 LIST_FOR_EACH (subfacet, list_node, &facet->subfacets) {
3335 assert(!subfacet->dp_byte_count);
3336 assert(!subfacet->dp_packet_count);
3339 facet_push_stats(facet);
3340 facet_account(facet);
3342 if (ofproto->netflow && !facet_is_controller_flow(facet)) {
3343 struct ofexpired expired;
3344 expired.flow = facet->flow;
3345 expired.packet_count = facet->packet_count;
3346 expired.byte_count = facet->byte_count;
3347 expired.used = facet->used;
3348 netflow_expire(ofproto->netflow, &facet->nf_flow, &expired);
3351 facet->rule->packet_count += facet->packet_count;
3352 facet->rule->byte_count += facet->byte_count;
3354 /* Reset counters to prevent double counting if 'facet' ever gets
3356 facet_reset_counters(facet);
3358 netflow_flow_clear(&facet->nf_flow);
3359 facet->tcp_flags = 0;
3362 /* Searches 'ofproto''s table of facets for one exactly equal to 'flow'.
3363 * Returns it if found, otherwise a null pointer.
3365 * The returned facet might need revalidation; use facet_lookup_valid()
3366 * instead if that is important. */
3367 static struct facet *
3368 facet_find(struct ofproto_dpif *ofproto, const struct flow *flow)
3370 struct facet *facet;
3372 HMAP_FOR_EACH_WITH_HASH (facet, hmap_node, flow_hash(flow, 0),
3374 if (flow_equal(flow, &facet->flow)) {
3382 /* Searches 'ofproto''s table of facets for one exactly equal to 'flow'.
3383 * Returns it if found, otherwise a null pointer.
3385 * The returned facet is guaranteed to be valid. */
3386 static struct facet *
3387 facet_lookup_valid(struct ofproto_dpif *ofproto, const struct flow *flow)
3389 struct facet *facet = facet_find(ofproto, flow);
3391 /* The facet we found might not be valid, since we could be in need of
3392 * revalidation. If it is not valid, don't return it. */
3394 && (ofproto->need_revalidate
3395 || tag_set_intersects(&ofproto->revalidate_set, facet->tags))
3396 && !facet_revalidate(facet)) {
3397 COVERAGE_INC(facet_invalidated);
3405 facet_check_consistency(struct facet *facet)
3407 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 15);
3409 struct ofproto_dpif *ofproto = ofproto_dpif_cast(facet->rule->up.ofproto);
3411 struct rule_dpif *rule;
3412 struct subfacet *subfacet;
3413 bool may_log = false;
3416 /* Check the rule for consistency. */
3417 rule = rule_dpif_lookup(ofproto, &facet->flow, 0);
3419 if (!VLOG_DROP_WARN(&rl)) {
3420 char *s = flow_to_string(&facet->flow);
3421 VLOG_WARN("%s: facet should not exist", s);
3425 } else if (rule != facet->rule) {
3426 may_log = !VLOG_DROP_WARN(&rl);
3432 flow_format(&s, &facet->flow);
3433 ds_put_format(&s, ": facet associated with wrong rule (was "
3434 "table=%"PRIu8",", facet->rule->up.table_id);
3435 cls_rule_format(&facet->rule->up.cr, &s);
3436 ds_put_format(&s, ") (should have been table=%"PRIu8",",
3438 cls_rule_format(&rule->up.cr, &s);
3439 ds_put_char(&s, ')');
3441 VLOG_WARN("%s", ds_cstr(&s));
3448 /* Check the datapath actions for consistency. */
3449 LIST_FOR_EACH (subfacet, list_node, &facet->subfacets) {
3450 struct action_xlate_ctx ctx;
3451 struct ofpbuf *odp_actions;
3452 bool actions_changed;
3453 bool should_install;
3455 action_xlate_ctx_init(&ctx, ofproto, &facet->flow,
3456 subfacet->initial_tci, rule, 0, NULL);
3457 odp_actions = xlate_actions(&ctx, rule->up.actions,
3458 rule->up.n_actions);
3460 should_install = (ctx.may_set_up_flow
3461 && subfacet->key_fitness != ODP_FIT_TOO_LITTLE);
3462 if (!should_install && !subfacet->installed) {
3463 /* The actions for uninstallable flows may vary from one packet to
3464 * the next, so don't compare the actions. */
3468 actions_changed = (subfacet->actions_len != odp_actions->size
3469 || memcmp(subfacet->actions, odp_actions->data,
3470 subfacet->actions_len));
3471 if (should_install != subfacet->installed || actions_changed) {
3473 may_log = !VLOG_DROP_WARN(&rl);
3478 struct odputil_keybuf keybuf;
3483 subfacet_get_key(subfacet, &keybuf, &key);
3484 odp_flow_key_format(key.data, key.size, &s);
3486 ds_put_cstr(&s, ": inconsistency in subfacet");
3487 if (should_install != subfacet->installed) {
3488 enum odp_key_fitness fitness = subfacet->key_fitness;
3490 ds_put_format(&s, " (should%s have been installed)",
3491 should_install ? "" : " not");
3492 ds_put_format(&s, " (may_set_up_flow=%s, fitness=%s)",
3493 ctx.may_set_up_flow ? "true" : "false",
3494 odp_key_fitness_to_string(fitness));
3496 if (actions_changed) {
3497 ds_put_cstr(&s, " (actions were: ");
3498 format_odp_actions(&s, subfacet->actions,
3499 subfacet->actions_len);
3500 ds_put_cstr(&s, ") (correct actions: ");
3501 format_odp_actions(&s, odp_actions->data,
3503 ds_put_char(&s, ')');
3505 ds_put_cstr(&s, " (actions: ");
3506 format_odp_actions(&s, subfacet->actions,
3507 subfacet->actions_len);
3508 ds_put_char(&s, ')');
3510 VLOG_WARN("%s", ds_cstr(&s));
3516 ofpbuf_delete(odp_actions);
3522 /* Re-searches the classifier for 'facet':
3524 * - If the rule found is different from 'facet''s current rule, moves
3525 * 'facet' to the new rule and recompiles its actions.
3527 * - If the rule found is the same as 'facet''s current rule, leaves 'facet'
3528 * where it is and recompiles its actions anyway.
3530 * - If there is none, destroys 'facet'.
3532 * Returns true if 'facet' still exists, false if it has been destroyed. */
3534 facet_revalidate(struct facet *facet)
3536 struct ofproto_dpif *ofproto = ofproto_dpif_cast(facet->rule->up.ofproto);
3538 struct nlattr *odp_actions;
3541 struct actions *new_actions;
3543 struct action_xlate_ctx ctx;
3544 struct rule_dpif *new_rule;
3545 struct subfacet *subfacet;
3546 bool actions_changed;
3549 COVERAGE_INC(facet_revalidate);
3551 /* Determine the new rule. */
3552 new_rule = rule_dpif_lookup(ofproto, &facet->flow, 0);
3554 /* No new rule, so delete the facet. */
3555 facet_remove(facet);
3559 /* Calculate new datapath actions.
3561 * We do not modify any 'facet' state yet, because we might need to, e.g.,
3562 * emit a NetFlow expiration and, if so, we need to have the old state
3563 * around to properly compose it. */
3565 /* If the datapath actions changed or the installability changed,
3566 * then we need to talk to the datapath. */
3569 memset(&ctx, 0, sizeof ctx);
3570 LIST_FOR_EACH (subfacet, list_node, &facet->subfacets) {
3571 struct ofpbuf *odp_actions;
3572 bool should_install;
3574 action_xlate_ctx_init(&ctx, ofproto, &facet->flow,
3575 subfacet->initial_tci, new_rule, 0, NULL);
3576 odp_actions = xlate_actions(&ctx, new_rule->up.actions,
3577 new_rule->up.n_actions);
3578 actions_changed = (subfacet->actions_len != odp_actions->size
3579 || memcmp(subfacet->actions, odp_actions->data,
3580 subfacet->actions_len));
3582 should_install = (ctx.may_set_up_flow
3583 && subfacet->key_fitness != ODP_FIT_TOO_LITTLE);
3584 if (actions_changed || should_install != subfacet->installed) {
3585 if (should_install) {
3586 struct dpif_flow_stats stats;
3588 subfacet_install(subfacet,
3589 odp_actions->data, odp_actions->size, &stats);
3590 subfacet_update_stats(subfacet, &stats);
3592 subfacet_uninstall(subfacet);
3596 new_actions = xcalloc(list_size(&facet->subfacets),
3597 sizeof *new_actions);
3599 new_actions[i].odp_actions = xmemdup(odp_actions->data,
3601 new_actions[i].actions_len = odp_actions->size;
3604 ofpbuf_delete(odp_actions);
3608 facet_flush_stats(facet);
3611 /* Update 'facet' now that we've taken care of all the old state. */
3612 facet->tags = ctx.tags;
3613 facet->nf_flow.output_iface = ctx.nf_output_iface;
3614 facet->may_install = ctx.may_set_up_flow;
3615 facet->has_learn = ctx.has_learn;
3616 facet->has_normal = ctx.has_normal;
3617 facet->has_fin_timeout = ctx.has_fin_timeout;
3618 facet->mirrors = ctx.mirrors;
3621 LIST_FOR_EACH (subfacet, list_node, &facet->subfacets) {
3622 if (new_actions[i].odp_actions) {
3623 free(subfacet->actions);
3624 subfacet->actions = new_actions[i].odp_actions;
3625 subfacet->actions_len = new_actions[i].actions_len;
3631 if (facet->rule != new_rule) {
3632 COVERAGE_INC(facet_changed_rule);
3633 list_remove(&facet->list_node);
3634 list_push_back(&new_rule->facets, &facet->list_node);
3635 facet->rule = new_rule;
3636 facet->used = new_rule->up.created;
3637 facet->prev_used = facet->used;
3643 /* Updates 'facet''s used time. Caller is responsible for calling
3644 * facet_push_stats() to update the flows which 'facet' resubmits into. */
3646 facet_update_time(struct facet *facet, long long int used)
3648 struct ofproto_dpif *ofproto = ofproto_dpif_cast(facet->rule->up.ofproto);
3649 if (used > facet->used) {
3651 ofproto_rule_update_used(&facet->rule->up, used);
3652 netflow_flow_update_time(ofproto->netflow, &facet->nf_flow, used);
3657 facet_reset_counters(struct facet *facet)
3659 facet->packet_count = 0;
3660 facet->byte_count = 0;
3661 facet->prev_packet_count = 0;
3662 facet->prev_byte_count = 0;
3663 facet->accounted_bytes = 0;
3667 facet_push_stats(struct facet *facet)
3669 uint64_t new_packets, new_bytes;
3671 assert(facet->packet_count >= facet->prev_packet_count);
3672 assert(facet->byte_count >= facet->prev_byte_count);
3673 assert(facet->used >= facet->prev_used);
3675 new_packets = facet->packet_count - facet->prev_packet_count;
3676 new_bytes = facet->byte_count - facet->prev_byte_count;
3678 if (new_packets || new_bytes || facet->used > facet->prev_used) {
3679 facet->prev_packet_count = facet->packet_count;
3680 facet->prev_byte_count = facet->byte_count;
3681 facet->prev_used = facet->used;
3683 flow_push_stats(facet->rule, &facet->flow,
3684 new_packets, new_bytes, facet->used);
3686 update_mirror_stats(ofproto_dpif_cast(facet->rule->up.ofproto),
3687 facet->mirrors, new_packets, new_bytes);
3691 struct ofproto_push {
3692 struct action_xlate_ctx ctx;
3699 push_resubmit(struct action_xlate_ctx *ctx, struct rule_dpif *rule)
3701 struct ofproto_push *push = CONTAINER_OF(ctx, struct ofproto_push, ctx);
3704 rule->packet_count += push->packets;
3705 rule->byte_count += push->bytes;
3706 ofproto_rule_update_used(&rule->up, push->used);
3710 /* Pushes flow statistics to the rules which 'flow' resubmits into given
3711 * 'rule''s actions and mirrors. */
3713 flow_push_stats(struct rule_dpif *rule,
3714 const struct flow *flow, uint64_t packets, uint64_t bytes,
3717 struct ofproto_dpif *ofproto = ofproto_dpif_cast(rule->up.ofproto);
3718 struct ofproto_push push;
3720 push.packets = packets;
3724 ofproto_rule_update_used(&rule->up, used);
3726 action_xlate_ctx_init(&push.ctx, ofproto, flow, flow->vlan_tci, rule,
3728 push.ctx.resubmit_hook = push_resubmit;
3729 ofpbuf_delete(xlate_actions(&push.ctx,
3730 rule->up.actions, rule->up.n_actions));
3735 static struct subfacet *
3736 subfacet_find__(struct ofproto_dpif *ofproto,
3737 const struct nlattr *key, size_t key_len, uint32_t key_hash,
3738 const struct flow *flow)
3740 struct subfacet *subfacet;
3742 HMAP_FOR_EACH_WITH_HASH (subfacet, hmap_node, key_hash,
3743 &ofproto->subfacets) {
3745 ? (subfacet->key_len == key_len
3746 && !memcmp(key, subfacet->key, key_len))
3747 : flow_equal(flow, &subfacet->facet->flow)) {
3755 /* Searches 'facet' (within 'ofproto') for a subfacet with the specified
3756 * 'key_fitness', 'key', and 'key_len'. Returns the existing subfacet if
3757 * there is one, otherwise creates and returns a new subfacet.
3759 * If the returned subfacet is new, then subfacet->actions will be NULL, in
3760 * which case the caller must populate the actions with
3761 * subfacet_make_actions(). */
3762 static struct subfacet *
3763 subfacet_create(struct facet *facet, enum odp_key_fitness key_fitness,
3764 const struct nlattr *key, size_t key_len, ovs_be16 initial_tci)
3766 struct ofproto_dpif *ofproto = ofproto_dpif_cast(facet->rule->up.ofproto);
3767 uint32_t key_hash = odp_flow_key_hash(key, key_len);
3768 struct subfacet *subfacet;
3770 subfacet = subfacet_find__(ofproto, key, key_len, key_hash, &facet->flow);
3772 if (subfacet->facet == facet) {
3776 /* This shouldn't happen. */
3777 VLOG_ERR_RL(&rl, "subfacet with wrong facet");
3778 subfacet_destroy(subfacet);
3781 subfacet = xzalloc(sizeof *subfacet);
3782 hmap_insert(&ofproto->subfacets, &subfacet->hmap_node, key_hash);
3783 list_push_back(&facet->subfacets, &subfacet->list_node);
3784 subfacet->facet = facet;
3785 subfacet->used = time_msec();
3786 subfacet->key_fitness = key_fitness;
3787 if (key_fitness != ODP_FIT_PERFECT) {
3788 subfacet->key = xmemdup(key, key_len);
3789 subfacet->key_len = key_len;
3791 subfacet->installed = false;
3792 subfacet->initial_tci = initial_tci;
3797 /* Searches 'ofproto' for a subfacet with the given 'key', 'key_len', and
3798 * 'flow'. Returns the subfacet if one exists, otherwise NULL. */
3799 static struct subfacet *
3800 subfacet_find(struct ofproto_dpif *ofproto,
3801 const struct nlattr *key, size_t key_len)
3803 uint32_t key_hash = odp_flow_key_hash(key, key_len);
3804 enum odp_key_fitness fitness;
3807 fitness = odp_flow_key_to_flow(key, key_len, &flow);
3808 if (fitness == ODP_FIT_ERROR) {
3812 return subfacet_find__(ofproto, key, key_len, key_hash, &flow);
3815 /* Uninstalls 'subfacet' from the datapath, if it is installed, removes it from
3816 * its facet within 'ofproto', and frees it. */
3818 subfacet_destroy__(struct subfacet *subfacet)
3820 struct facet *facet = subfacet->facet;
3821 struct ofproto_dpif *ofproto = ofproto_dpif_cast(facet->rule->up.ofproto);
3823 subfacet_uninstall(subfacet);
3824 hmap_remove(&ofproto->subfacets, &subfacet->hmap_node);
3825 list_remove(&subfacet->list_node);
3826 free(subfacet->key);
3827 free(subfacet->actions);
3831 /* Destroys 'subfacet', as with subfacet_destroy__(), and then if this was the
3832 * last remaining subfacet in its facet destroys the facet too. */
3834 subfacet_destroy(struct subfacet *subfacet)
3836 struct facet *facet = subfacet->facet;
3838 if (list_is_singleton(&facet->subfacets)) {
3839 /* facet_remove() needs at least one subfacet (it will remove it). */
3840 facet_remove(facet);
3842 subfacet_destroy__(subfacet);
3846 /* Initializes 'key' with the sequence of OVS_KEY_ATTR_* Netlink attributes
3847 * that can be used to refer to 'subfacet'. The caller must provide 'keybuf'
3848 * for use as temporary storage. */
3850 subfacet_get_key(struct subfacet *subfacet, struct odputil_keybuf *keybuf,
3853 if (!subfacet->key) {
3854 ofpbuf_use_stack(key, keybuf, sizeof *keybuf);
3855 odp_flow_key_from_flow(key, &subfacet->facet->flow);
3857 ofpbuf_use_const(key, subfacet->key, subfacet->key_len);
3861 /* Composes the datapath actions for 'subfacet' based on its rule's actions. */
3863 subfacet_make_actions(struct subfacet *subfacet, const struct ofpbuf *packet)
3865 struct facet *facet = subfacet->facet;
3866 struct rule_dpif *rule = facet->rule;
3867 struct ofproto_dpif *ofproto = ofproto_dpif_cast(rule->up.ofproto);
3868 struct ofpbuf *odp_actions;
3869 struct action_xlate_ctx ctx;
3871 action_xlate_ctx_init(&ctx, ofproto, &facet->flow, subfacet->initial_tci,
3873 odp_actions = xlate_actions(&ctx, rule->up.actions, rule->up.n_actions);
3874 facet->tags = ctx.tags;
3875 facet->may_install = ctx.may_set_up_flow;
3876 facet->has_learn = ctx.has_learn;
3877 facet->has_normal = ctx.has_normal;
3878 facet->has_fin_timeout = ctx.has_fin_timeout;
3879 facet->nf_flow.output_iface = ctx.nf_output_iface;
3880 facet->mirrors = ctx.mirrors;
3882 if (subfacet->actions_len != odp_actions->size
3883 || memcmp(subfacet->actions, odp_actions->data, odp_actions->size)) {
3884 free(subfacet->actions);
3885 subfacet->actions_len = odp_actions->size;
3886 subfacet->actions = xmemdup(odp_actions->data, odp_actions->size);
3889 ofpbuf_delete(odp_actions);
3892 /* Updates 'subfacet''s datapath flow, setting its actions to 'actions_len'
3893 * bytes of actions in 'actions'. If 'stats' is non-null, statistics counters
3894 * in the datapath will be zeroed and 'stats' will be updated with traffic new
3895 * since 'subfacet' was last updated.
3897 * Returns 0 if successful, otherwise a positive errno value. */
3899 subfacet_install(struct subfacet *subfacet,
3900 const struct nlattr *actions, size_t actions_len,
3901 struct dpif_flow_stats *stats)
3903 struct facet *facet = subfacet->facet;
3904 struct ofproto_dpif *ofproto = ofproto_dpif_cast(facet->rule->up.ofproto);
3905 struct odputil_keybuf keybuf;
3906 enum dpif_flow_put_flags flags;
3910 flags = DPIF_FP_CREATE | DPIF_FP_MODIFY;
3912 flags |= DPIF_FP_ZERO_STATS;
3915 subfacet_get_key(subfacet, &keybuf, &key);
3916 ret = dpif_flow_put(ofproto->dpif, flags, key.data, key.size,
3917 actions, actions_len, stats);
3920 subfacet_reset_dp_stats(subfacet, stats);
3926 /* If 'subfacet' is installed in the datapath, uninstalls it. */
3928 subfacet_uninstall(struct subfacet *subfacet)
3930 if (subfacet->installed) {
3931 struct rule_dpif *rule = subfacet->facet->rule;
3932 struct ofproto_dpif *ofproto = ofproto_dpif_cast(rule->up.ofproto);
3933 struct odputil_keybuf keybuf;
3934 struct dpif_flow_stats stats;
3938 subfacet_get_key(subfacet, &keybuf, &key);
3939 error = dpif_flow_del(ofproto->dpif, key.data, key.size, &stats);
3940 subfacet_reset_dp_stats(subfacet, &stats);
3942 subfacet_update_stats(subfacet, &stats);
3944 subfacet->installed = false;
3946 assert(subfacet->dp_packet_count == 0);
3947 assert(subfacet->dp_byte_count == 0);
3951 /* Resets 'subfacet''s datapath statistics counters. This should be called
3952 * when 'subfacet''s statistics are cleared in the datapath. If 'stats' is
3953 * non-null, it should contain the statistics returned by dpif when 'subfacet'
3954 * was reset in the datapath. 'stats' will be modified to include only
3955 * statistics new since 'subfacet' was last updated. */
3957 subfacet_reset_dp_stats(struct subfacet *subfacet,
3958 struct dpif_flow_stats *stats)
3961 && subfacet->dp_packet_count <= stats->n_packets
3962 && subfacet->dp_byte_count <= stats->n_bytes) {
3963 stats->n_packets -= subfacet->dp_packet_count;
3964 stats->n_bytes -= subfacet->dp_byte_count;
3967 subfacet->dp_packet_count = 0;
3968 subfacet->dp_byte_count = 0;
3971 /* Updates 'subfacet''s used time. The caller is responsible for calling
3972 * facet_push_stats() to update the flows which 'subfacet' resubmits into. */
3974 subfacet_update_time(struct subfacet *subfacet, long long int used)
3976 if (used > subfacet->used) {
3977 subfacet->used = used;
3978 facet_update_time(subfacet->facet, used);
3982 /* Folds the statistics from 'stats' into the counters in 'subfacet'.
3984 * Because of the meaning of a subfacet's counters, it only makes sense to do
3985 * this if 'stats' are not tracked in the datapath, that is, if 'stats'
3986 * represents a packet that was sent by hand or if it represents statistics
3987 * that have been cleared out of the datapath. */
3989 subfacet_update_stats(struct subfacet *subfacet,
3990 const struct dpif_flow_stats *stats)
3992 if (stats->n_packets || stats->used > subfacet->used) {
3993 struct facet *facet = subfacet->facet;
3995 subfacet_update_time(subfacet, stats->used);
3996 facet->packet_count += stats->n_packets;
3997 facet->byte_count += stats->n_bytes;
3998 facet->tcp_flags |= stats->tcp_flags;
3999 facet_push_stats(facet);
4000 netflow_flow_update_flags(&facet->nf_flow, stats->tcp_flags);
4006 static struct rule_dpif *
4007 rule_dpif_lookup(struct ofproto_dpif *ofproto, const struct flow *flow,
4010 struct cls_rule *cls_rule;
4011 struct classifier *cls;
4013 if (table_id >= N_TABLES) {
4017 cls = &ofproto->up.tables[table_id].cls;
4018 if (flow->nw_frag & FLOW_NW_FRAG_ANY
4019 && ofproto->up.frag_handling == OFPC_FRAG_NORMAL) {
4020 /* For OFPC_NORMAL frag_handling, we must pretend that transport ports
4021 * are unavailable. */
4022 struct flow ofpc_normal_flow = *flow;
4023 ofpc_normal_flow.tp_src = htons(0);
4024 ofpc_normal_flow.tp_dst = htons(0);
4025 cls_rule = classifier_lookup(cls, &ofpc_normal_flow);
4027 cls_rule = classifier_lookup(cls, flow);
4029 return rule_dpif_cast(rule_from_cls_rule(cls_rule));
4033 complete_operation(struct rule_dpif *rule)
4035 struct ofproto_dpif *ofproto = ofproto_dpif_cast(rule->up.ofproto);
4037 rule_invalidate(rule);
4039 struct dpif_completion *c = xmalloc(sizeof *c);
4040 c->op = rule->up.pending;
4041 list_push_back(&ofproto->completions, &c->list_node);
4043 ofoperation_complete(rule->up.pending, 0);
4047 static struct rule *
4050 struct rule_dpif *rule = xmalloc(sizeof *rule);
4055 rule_dealloc(struct rule *rule_)
4057 struct rule_dpif *rule = rule_dpif_cast(rule_);
4062 rule_construct(struct rule *rule_)
4064 struct rule_dpif *rule = rule_dpif_cast(rule_);
4065 struct ofproto_dpif *ofproto = ofproto_dpif_cast(rule->up.ofproto);
4066 struct rule_dpif *victim;
4070 error = validate_actions(rule->up.actions, rule->up.n_actions,
4071 &rule->up.cr.flow, ofproto->max_ports);
4076 rule->packet_count = 0;
4077 rule->byte_count = 0;
4079 victim = rule_dpif_cast(ofoperation_get_victim(rule->up.pending));
4080 if (victim && !list_is_empty(&victim->facets)) {
4081 struct facet *facet;
4083 rule->facets = victim->facets;
4084 list_moved(&rule->facets);
4085 LIST_FOR_EACH (facet, list_node, &rule->facets) {
4086 /* XXX: We're only clearing our local counters here. It's possible
4087 * that quite a few packets are unaccounted for in the datapath
4088 * statistics. These will be accounted to the new rule instead of
4089 * cleared as required. This could be fixed by clearing out the
4090 * datapath statistics for this facet, but currently it doesn't
4092 facet_reset_counters(facet);
4096 /* Must avoid list_moved() in this case. */
4097 list_init(&rule->facets);
4100 table_id = rule->up.table_id;
4101 rule->tag = (victim ? victim->tag
4103 : rule_calculate_tag(&rule->up.cr.flow, &rule->up.cr.wc,
4104 ofproto->tables[table_id].basis));
4106 complete_operation(rule);
4111 rule_destruct(struct rule *rule_)
4113 struct rule_dpif *rule = rule_dpif_cast(rule_);
4114 struct facet *facet, *next_facet;
4116 LIST_FOR_EACH_SAFE (facet, next_facet, list_node, &rule->facets) {
4117 facet_revalidate(facet);
4120 complete_operation(rule);
4124 rule_get_stats(struct rule *rule_, uint64_t *packets, uint64_t *bytes)
4126 struct rule_dpif *rule = rule_dpif_cast(rule_);
4127 struct facet *facet;
4129 /* Start from historical data for 'rule' itself that are no longer tracked
4130 * in facets. This counts, for example, facets that have expired. */
4131 *packets = rule->packet_count;
4132 *bytes = rule->byte_count;
4134 /* Add any statistics that are tracked by facets. This includes
4135 * statistical data recently updated by ofproto_update_stats() as well as
4136 * stats for packets that were executed "by hand" via dpif_execute(). */
4137 LIST_FOR_EACH (facet, list_node, &rule->facets) {
4138 *packets += facet->packet_count;
4139 *bytes += facet->byte_count;
4144 rule_execute(struct rule *rule_, const struct flow *flow,
4145 struct ofpbuf *packet)
4147 struct rule_dpif *rule = rule_dpif_cast(rule_);
4148 struct ofproto_dpif *ofproto = ofproto_dpif_cast(rule->up.ofproto);
4149 struct action_xlate_ctx ctx;
4150 struct ofpbuf *odp_actions;
4153 action_xlate_ctx_init(&ctx, ofproto, flow, flow->vlan_tci,
4154 rule, packet_get_tcp_flags(packet, flow), packet);
4155 odp_actions = xlate_actions(&ctx, rule->up.actions, rule->up.n_actions);
4156 size = packet->size;
4157 if (execute_odp_actions(ofproto, flow, odp_actions->data,
4158 odp_actions->size, packet)) {
4159 rule->packet_count++;
4160 rule->byte_count += size;
4161 flow_push_stats(rule, flow, 1, size, time_msec());
4163 ofpbuf_delete(odp_actions);
4169 rule_modify_actions(struct rule *rule_)
4171 struct rule_dpif *rule = rule_dpif_cast(rule_);
4172 struct ofproto_dpif *ofproto = ofproto_dpif_cast(rule->up.ofproto);
4175 error = validate_actions(rule->up.actions, rule->up.n_actions,
4176 &rule->up.cr.flow, ofproto->max_ports);
4178 ofoperation_complete(rule->up.pending, error);
4182 complete_operation(rule);
4185 /* Sends 'packet' out 'ofport'.
4186 * May modify 'packet'.
4187 * Returns 0 if successful, otherwise a positive errno value. */
4189 send_packet(const struct ofport_dpif *ofport, struct ofpbuf *packet)
4191 const struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport->up.ofproto);
4192 struct ofpbuf key, odp_actions;
4193 struct odputil_keybuf keybuf;
4198 flow_extract((struct ofpbuf *) packet, 0, 0, 0, &flow);
4199 odp_port = vsp_realdev_to_vlandev(ofproto, ofport->odp_port,
4201 if (odp_port != ofport->odp_port) {
4202 eth_pop_vlan(packet);
4203 flow.vlan_tci = htons(0);
4206 ofpbuf_use_stack(&key, &keybuf, sizeof keybuf);
4207 odp_flow_key_from_flow(&key, &flow);
4209 ofpbuf_init(&odp_actions, 32);
4210 compose_sflow_action(ofproto, &odp_actions, &flow, odp_port);
4212 nl_msg_put_u32(&odp_actions, OVS_ACTION_ATTR_OUTPUT, odp_port);
4213 error = dpif_execute(ofproto->dpif,
4215 odp_actions.data, odp_actions.size,
4217 ofpbuf_uninit(&odp_actions);
4220 VLOG_WARN_RL(&rl, "%s: failed to send packet on port %"PRIu32" (%s)",
4221 ofproto->up.name, odp_port, strerror(error));
4223 ofproto_update_local_port_stats(ofport->up.ofproto, packet->size, 0);
4227 /* OpenFlow to datapath action translation. */
4229 static void do_xlate_actions(const union ofp_action *in, size_t n_in,
4230 struct action_xlate_ctx *ctx);
4231 static void xlate_normal(struct action_xlate_ctx *);
4234 put_userspace_action(const struct ofproto_dpif *ofproto,
4235 struct ofpbuf *odp_actions,
4236 const struct flow *flow,
4237 const struct user_action_cookie *cookie)
4241 pid = dpif_port_get_pid(ofproto->dpif,
4242 ofp_port_to_odp_port(flow->in_port));
4244 return odp_put_userspace_action(pid, cookie, odp_actions);
4247 /* Compose SAMPLE action for sFlow. */
4249 compose_sflow_action(const struct ofproto_dpif *ofproto,
4250 struct ofpbuf *odp_actions,
4251 const struct flow *flow,
4254 uint32_t port_ifindex;
4255 uint32_t probability;
4256 struct user_action_cookie cookie;
4257 size_t sample_offset, actions_offset;
4258 int cookie_offset, n_output;
4260 if (!ofproto->sflow || flow->in_port == OFPP_NONE) {
4264 if (odp_port == OVSP_NONE) {
4268 port_ifindex = dpif_sflow_odp_port_to_ifindex(ofproto->sflow, odp_port);
4272 sample_offset = nl_msg_start_nested(odp_actions, OVS_ACTION_ATTR_SAMPLE);
4274 /* Number of packets out of UINT_MAX to sample. */
4275 probability = dpif_sflow_get_probability(ofproto->sflow);
4276 nl_msg_put_u32(odp_actions, OVS_SAMPLE_ATTR_PROBABILITY, probability);
4278 actions_offset = nl_msg_start_nested(odp_actions, OVS_SAMPLE_ATTR_ACTIONS);
4280 cookie.type = USER_ACTION_COOKIE_SFLOW;
4281 cookie.data = port_ifindex;
4282 cookie.n_output = n_output;
4283 cookie.vlan_tci = 0;
4284 cookie_offset = put_userspace_action(ofproto, odp_actions, flow, &cookie);
4286 nl_msg_end_nested(odp_actions, actions_offset);
4287 nl_msg_end_nested(odp_actions, sample_offset);
4288 return cookie_offset;
4291 /* SAMPLE action must be first action in any given list of actions.
4292 * At this point we do not have all information required to build it. So try to
4293 * build sample action as complete as possible. */
4295 add_sflow_action(struct action_xlate_ctx *ctx)
4297 ctx->user_cookie_offset = compose_sflow_action(ctx->ofproto,
4299 &ctx->flow, OVSP_NONE);
4300 ctx->sflow_odp_port = 0;
4301 ctx->sflow_n_outputs = 0;
4304 /* Fix SAMPLE action according to data collected while composing ODP actions.
4305 * We need to fix SAMPLE actions OVS_SAMPLE_ATTR_ACTIONS attribute, i.e. nested
4306 * USERSPACE action's user-cookie which is required for sflow. */
4308 fix_sflow_action(struct action_xlate_ctx *ctx)
4310 const struct flow *base = &ctx->base_flow;
4311 struct user_action_cookie *cookie;
4313 if (!ctx->user_cookie_offset) {
4317 cookie = ofpbuf_at(ctx->odp_actions, ctx->user_cookie_offset,
4319 assert(cookie != NULL);
4320 assert(cookie->type == USER_ACTION_COOKIE_SFLOW);
4322 if (ctx->sflow_n_outputs) {
4323 cookie->data = dpif_sflow_odp_port_to_ifindex(ctx->ofproto->sflow,
4324 ctx->sflow_odp_port);
4326 if (ctx->sflow_n_outputs >= 255) {
4327 cookie->n_output = 255;
4329 cookie->n_output = ctx->sflow_n_outputs;
4331 cookie->vlan_tci = base->vlan_tci;
4335 compose_output_action__(struct action_xlate_ctx *ctx, uint16_t ofp_port,
4338 const struct ofport_dpif *ofport = get_ofp_port(ctx->ofproto, ofp_port);
4339 uint16_t odp_port = ofp_port_to_odp_port(ofp_port);
4340 ovs_be16 flow_vlan_tci = ctx->flow.vlan_tci;
4341 uint8_t flow_nw_tos = ctx->flow.nw_tos;
4345 struct priority_to_dscp *pdscp;
4347 if (ofport->up.pp.config & OFPUTIL_PC_NO_FWD
4348 || (check_stp && !stp_forward_in_state(ofport->stp_state))) {
4352 pdscp = get_priority(ofport, ctx->flow.skb_priority);
4354 ctx->flow.nw_tos &= ~IP_DSCP_MASK;
4355 ctx->flow.nw_tos |= pdscp->dscp;
4358 /* We may not have an ofport record for this port, but it doesn't hurt
4359 * to allow forwarding to it anyhow. Maybe such a port will appear
4360 * later and we're pre-populating the flow table. */
4363 out_port = vsp_realdev_to_vlandev(ctx->ofproto, odp_port,
4364 ctx->flow.vlan_tci);
4365 if (out_port != odp_port) {
4366 ctx->flow.vlan_tci = htons(0);
4368 commit_odp_actions(&ctx->flow, &ctx->base_flow, ctx->odp_actions);
4369 nl_msg_put_u32(ctx->odp_actions, OVS_ACTION_ATTR_OUTPUT, out_port);
4371 ctx->sflow_odp_port = odp_port;
4372 ctx->sflow_n_outputs++;
4373 ctx->nf_output_iface = ofp_port;
4374 ctx->flow.vlan_tci = flow_vlan_tci;
4375 ctx->flow.nw_tos = flow_nw_tos;
4379 compose_output_action(struct action_xlate_ctx *ctx, uint16_t ofp_port)
4381 compose_output_action__(ctx, ofp_port, true);
4385 xlate_table_action(struct action_xlate_ctx *ctx,
4386 uint16_t in_port, uint8_t table_id)
4388 if (ctx->recurse < MAX_RESUBMIT_RECURSION) {
4389 struct ofproto_dpif *ofproto = ctx->ofproto;
4390 struct rule_dpif *rule;
4391 uint16_t old_in_port;
4392 uint8_t old_table_id;
4394 old_table_id = ctx->table_id;
4395 ctx->table_id = table_id;
4397 /* Look up a flow with 'in_port' as the input port. */
4398 old_in_port = ctx->flow.in_port;
4399 ctx->flow.in_port = in_port;
4400 rule = rule_dpif_lookup(ofproto, &ctx->flow, table_id);
4403 if (table_id > 0 && table_id < N_TABLES) {
4404 struct table_dpif *table = &ofproto->tables[table_id];
4405 if (table->other_table) {
4408 : rule_calculate_tag(&ctx->flow,
4409 &table->other_table->wc,
4414 /* Restore the original input port. Otherwise OFPP_NORMAL and
4415 * OFPP_IN_PORT will have surprising behavior. */
4416 ctx->flow.in_port = old_in_port;
4418 if (ctx->resubmit_hook) {
4419 ctx->resubmit_hook(ctx, rule);
4423 struct rule_dpif *old_rule = ctx->rule;
4427 do_xlate_actions(rule->up.actions, rule->up.n_actions, ctx);
4428 ctx->rule = old_rule;
4432 ctx->table_id = old_table_id;
4434 static struct vlog_rate_limit recurse_rl = VLOG_RATE_LIMIT_INIT(1, 1);
4436 VLOG_ERR_RL(&recurse_rl, "resubmit actions recursed over %d times",
4437 MAX_RESUBMIT_RECURSION);
4442 xlate_resubmit_table(struct action_xlate_ctx *ctx,
4443 const struct nx_action_resubmit *nar)
4448 in_port = (nar->in_port == htons(OFPP_IN_PORT)
4450 : ntohs(nar->in_port));
4451 table_id = nar->table == 255 ? ctx->table_id : nar->table;
4453 xlate_table_action(ctx, in_port, table_id);
4457 flood_packets(struct action_xlate_ctx *ctx, bool all)
4459 struct ofport_dpif *ofport;
4461 HMAP_FOR_EACH (ofport, up.hmap_node, &ctx->ofproto->up.ports) {
4462 uint16_t ofp_port = ofport->up.ofp_port;
4464 if (ofp_port == ctx->flow.in_port) {
4469 compose_output_action__(ctx, ofp_port, false);
4470 } else if (!(ofport->up.pp.config & OFPUTIL_PC_NO_FLOOD)) {
4471 compose_output_action(ctx, ofp_port);
4475 ctx->nf_output_iface = NF_OUT_FLOOD;
4479 execute_controller_action(struct action_xlate_ctx *ctx, int len,
4480 enum ofp_packet_in_reason reason,
4481 uint16_t controller_id)
4483 struct ofputil_packet_in pin;
4484 struct ofpbuf *packet;
4486 ctx->may_set_up_flow = false;
4491 packet = ofpbuf_clone(ctx->packet);
4493 if (packet->l2 && packet->l3) {
4494 struct eth_header *eh;
4496 eth_pop_vlan(packet);
4498 assert(eh->eth_type == ctx->flow.dl_type);
4499 memcpy(eh->eth_src, ctx->flow.dl_src, sizeof eh->eth_src);
4500 memcpy(eh->eth_dst, ctx->flow.dl_dst, sizeof eh->eth_dst);
4502 if (ctx->flow.vlan_tci & htons(VLAN_CFI)) {
4503 eth_push_vlan(packet, ctx->flow.vlan_tci);
4507 if (ctx->flow.dl_type == htons(ETH_TYPE_IP)) {
4508 packet_set_ipv4(packet, ctx->flow.nw_src, ctx->flow.nw_dst,
4509 ctx->flow.nw_tos, ctx->flow.nw_ttl);
4513 if (ctx->flow.nw_proto == IPPROTO_TCP) {
4514 packet_set_tcp_port(packet, ctx->flow.tp_src,
4516 } else if (ctx->flow.nw_proto == IPPROTO_UDP) {
4517 packet_set_udp_port(packet, ctx->flow.tp_src,
4524 pin.packet = packet->data;
4525 pin.packet_len = packet->size;
4526 pin.reason = reason;
4527 pin.controller_id = controller_id;
4528 pin.table_id = ctx->table_id;
4529 pin.cookie = ctx->rule ? ctx->rule->up.flow_cookie : 0;
4533 pin.total_len = packet->size;
4534 flow_get_metadata(&ctx->flow, &pin.fmd);
4536 connmgr_send_packet_in(ctx->ofproto->up.connmgr, &pin, &ctx->flow);
4537 ofpbuf_delete(packet);
4541 compose_dec_ttl(struct action_xlate_ctx *ctx)
4543 if (ctx->flow.dl_type != htons(ETH_TYPE_IP) &&
4544 ctx->flow.dl_type != htons(ETH_TYPE_IPV6)) {
4548 if (ctx->flow.nw_ttl > 1) {
4552 execute_controller_action(ctx, UINT16_MAX, OFPR_INVALID_TTL, 0);
4554 /* Stop processing for current table. */
4560 xlate_output_action__(struct action_xlate_ctx *ctx,
4561 uint16_t port, uint16_t max_len)
4563 uint16_t prev_nf_output_iface = ctx->nf_output_iface;
4565 ctx->nf_output_iface = NF_OUT_DROP;
4569 compose_output_action(ctx, ctx->flow.in_port);
4572 xlate_table_action(ctx, ctx->flow.in_port, ctx->table_id);
4578 flood_packets(ctx, false);
4581 flood_packets(ctx, true);
4583 case OFPP_CONTROLLER:
4584 execute_controller_action(ctx, max_len, OFPR_ACTION, 0);
4590 if (port != ctx->flow.in_port) {
4591 compose_output_action(ctx, port);
4596 if (prev_nf_output_iface == NF_OUT_FLOOD) {
4597 ctx->nf_output_iface = NF_OUT_FLOOD;
4598 } else if (ctx->nf_output_iface == NF_OUT_DROP) {
4599 ctx->nf_output_iface = prev_nf_output_iface;
4600 } else if (prev_nf_output_iface != NF_OUT_DROP &&
4601 ctx->nf_output_iface != NF_OUT_FLOOD) {
4602 ctx->nf_output_iface = NF_OUT_MULTI;
4607 xlate_output_reg_action(struct action_xlate_ctx *ctx,
4608 const struct nx_action_output_reg *naor)
4610 struct mf_subfield src;
4613 nxm_decode(&src, naor->src, naor->ofs_nbits);
4614 ofp_port = mf_get_subfield(&src, &ctx->flow);
4616 if (ofp_port <= UINT16_MAX) {
4617 xlate_output_action__(ctx, ofp_port, ntohs(naor->max_len));
4622 xlate_output_action(struct action_xlate_ctx *ctx,
4623 const struct ofp_action_output *oao)
4625 xlate_output_action__(ctx, ntohs(oao->port), ntohs(oao->max_len));
4629 xlate_enqueue_action(struct action_xlate_ctx *ctx,
4630 const struct ofp_action_enqueue *oae)
4633 uint32_t flow_priority, priority;
4636 error = dpif_queue_to_priority(ctx->ofproto->dpif, ntohl(oae->queue_id),
4639 /* Fall back to ordinary output action. */
4640 xlate_output_action__(ctx, ntohs(oae->port), 0);
4644 /* Figure out datapath output port. */
4645 ofp_port = ntohs(oae->port);
4646 if (ofp_port == OFPP_IN_PORT) {
4647 ofp_port = ctx->flow.in_port;
4648 } else if (ofp_port == ctx->flow.in_port) {
4652 /* Add datapath actions. */
4653 flow_priority = ctx->flow.skb_priority;
4654 ctx->flow.skb_priority = priority;
4655 compose_output_action(ctx, ofp_port);
4656 ctx->flow.skb_priority = flow_priority;
4658 /* Update NetFlow output port. */
4659 if (ctx->nf_output_iface == NF_OUT_DROP) {
4660 ctx->nf_output_iface = ofp_port;
4661 } else if (ctx->nf_output_iface != NF_OUT_FLOOD) {
4662 ctx->nf_output_iface = NF_OUT_MULTI;
4667 xlate_set_queue_action(struct action_xlate_ctx *ctx,
4668 const struct nx_action_set_queue *nasq)
4673 error = dpif_queue_to_priority(ctx->ofproto->dpif, ntohl(nasq->queue_id),
4676 /* Couldn't translate queue to a priority, so ignore. A warning
4677 * has already been logged. */
4681 ctx->flow.skb_priority = priority;
4684 struct xlate_reg_state {
4690 xlate_autopath(struct action_xlate_ctx *ctx,
4691 const struct nx_action_autopath *naa)
4693 uint16_t ofp_port = ntohl(naa->id);
4694 struct ofport_dpif *port = get_ofp_port(ctx->ofproto, ofp_port);
4696 if (!port || !port->bundle) {
4697 ofp_port = OFPP_NONE;
4698 } else if (port->bundle->bond) {
4699 /* Autopath does not support VLAN hashing. */
4700 struct ofport_dpif *slave = bond_choose_output_slave(
4701 port->bundle->bond, &ctx->flow, 0, &ctx->tags);
4703 ofp_port = slave->up.ofp_port;
4706 autopath_execute(naa, &ctx->flow, ofp_port);
4710 slave_enabled_cb(uint16_t ofp_port, void *ofproto_)
4712 struct ofproto_dpif *ofproto = ofproto_;
4713 struct ofport_dpif *port;
4723 case OFPP_CONTROLLER: /* Not supported by the bundle action. */
4726 port = get_ofp_port(ofproto, ofp_port);
4727 return port ? port->may_enable : false;
4732 xlate_learn_action(struct action_xlate_ctx *ctx,
4733 const struct nx_action_learn *learn)
4735 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(5, 1);
4736 struct ofputil_flow_mod fm;
4739 learn_execute(learn, &ctx->flow, &fm);
4741 error = ofproto_flow_mod(&ctx->ofproto->up, &fm);
4742 if (error && !VLOG_DROP_WARN(&rl)) {
4743 VLOG_WARN("learning action failed to modify flow table (%s)",
4744 ofperr_get_name(error));
4750 /* Reduces '*timeout' to no more than 'max'. A value of zero in either case
4751 * means "infinite". */
4753 reduce_timeout(uint16_t max, uint16_t *timeout)
4755 if (max && (!*timeout || *timeout > max)) {
4761 xlate_fin_timeout(struct action_xlate_ctx *ctx,
4762 const struct nx_action_fin_timeout *naft)
4764 if (ctx->tcp_flags & (TCP_FIN | TCP_RST) && ctx->rule) {
4765 struct rule_dpif *rule = ctx->rule;
4767 reduce_timeout(ntohs(naft->fin_idle_timeout), &rule->up.idle_timeout);
4768 reduce_timeout(ntohs(naft->fin_hard_timeout), &rule->up.hard_timeout);
4773 may_receive(const struct ofport_dpif *port, struct action_xlate_ctx *ctx)
4775 if (port->up.pp.config & (eth_addr_equals(ctx->flow.dl_dst, eth_addr_stp)
4776 ? OFPUTIL_PC_NO_RECV_STP
4777 : OFPUTIL_PC_NO_RECV)) {
4781 /* Only drop packets here if both forwarding and learning are
4782 * disabled. If just learning is enabled, we need to have
4783 * OFPP_NORMAL and the learning action have a look at the packet
4784 * before we can drop it. */
4785 if (!stp_forward_in_state(port->stp_state)
4786 && !stp_learn_in_state(port->stp_state)) {
4794 do_xlate_actions(const union ofp_action *in, size_t n_in,
4795 struct action_xlate_ctx *ctx)
4797 const struct ofport_dpif *port;
4798 const union ofp_action *ia;
4799 bool was_evictable = true;
4802 port = get_ofp_port(ctx->ofproto, ctx->flow.in_port);
4803 if (port && !may_receive(port, ctx)) {
4804 /* Drop this flow. */
4809 /* Don't let the rule we're working on get evicted underneath us. */
4810 was_evictable = ctx->rule->up.evictable;
4811 ctx->rule->up.evictable = false;
4813 OFPUTIL_ACTION_FOR_EACH_UNSAFE (ia, left, in, n_in) {
4814 const struct ofp_action_dl_addr *oada;
4815 const struct nx_action_resubmit *nar;
4816 const struct nx_action_set_tunnel *nast;
4817 const struct nx_action_set_queue *nasq;
4818 const struct nx_action_multipath *nam;
4819 const struct nx_action_autopath *naa;
4820 const struct nx_action_bundle *nab;
4821 const struct nx_action_output_reg *naor;
4822 const struct nx_action_controller *nac;
4823 enum ofputil_action_code code;
4830 code = ofputil_decode_action_unsafe(ia);
4832 case OFPUTIL_OFPAT10_OUTPUT:
4833 xlate_output_action(ctx, &ia->output);
4836 case OFPUTIL_OFPAT10_SET_VLAN_VID:
4837 ctx->flow.vlan_tci &= ~htons(VLAN_VID_MASK);
4838 ctx->flow.vlan_tci |= ia->vlan_vid.vlan_vid | htons(VLAN_CFI);
4841 case OFPUTIL_OFPAT10_SET_VLAN_PCP:
4842 ctx->flow.vlan_tci &= ~htons(VLAN_PCP_MASK);
4843 ctx->flow.vlan_tci |= htons(
4844 (ia->vlan_pcp.vlan_pcp << VLAN_PCP_SHIFT) | VLAN_CFI);
4847 case OFPUTIL_OFPAT10_STRIP_VLAN:
4848 ctx->flow.vlan_tci = htons(0);
4851 case OFPUTIL_OFPAT10_SET_DL_SRC:
4852 oada = ((struct ofp_action_dl_addr *) ia);
4853 memcpy(ctx->flow.dl_src, oada->dl_addr, ETH_ADDR_LEN);
4856 case OFPUTIL_OFPAT10_SET_DL_DST:
4857 oada = ((struct ofp_action_dl_addr *) ia);
4858 memcpy(ctx->flow.dl_dst, oada->dl_addr, ETH_ADDR_LEN);
4861 case OFPUTIL_OFPAT10_SET_NW_SRC:
4862 ctx->flow.nw_src = ia->nw_addr.nw_addr;
4865 case OFPUTIL_OFPAT10_SET_NW_DST:
4866 ctx->flow.nw_dst = ia->nw_addr.nw_addr;
4869 case OFPUTIL_OFPAT10_SET_NW_TOS:
4870 /* OpenFlow 1.0 only supports IPv4. */
4871 if (ctx->flow.dl_type == htons(ETH_TYPE_IP)) {
4872 ctx->flow.nw_tos &= ~IP_DSCP_MASK;
4873 ctx->flow.nw_tos |= ia->nw_tos.nw_tos & IP_DSCP_MASK;
4877 case OFPUTIL_OFPAT10_SET_TP_SRC:
4878 ctx->flow.tp_src = ia->tp_port.tp_port;
4881 case OFPUTIL_OFPAT10_SET_TP_DST:
4882 ctx->flow.tp_dst = ia->tp_port.tp_port;
4885 case OFPUTIL_OFPAT10_ENQUEUE:
4886 xlate_enqueue_action(ctx, (const struct ofp_action_enqueue *) ia);
4889 case OFPUTIL_NXAST_RESUBMIT:
4890 nar = (const struct nx_action_resubmit *) ia;
4891 xlate_table_action(ctx, ntohs(nar->in_port), ctx->table_id);
4894 case OFPUTIL_NXAST_RESUBMIT_TABLE:
4895 xlate_resubmit_table(ctx, (const struct nx_action_resubmit *) ia);
4898 case OFPUTIL_NXAST_SET_TUNNEL:
4899 nast = (const struct nx_action_set_tunnel *) ia;
4900 tun_id = htonll(ntohl(nast->tun_id));
4901 ctx->flow.tun_id = tun_id;
4904 case OFPUTIL_NXAST_SET_QUEUE:
4905 nasq = (const struct nx_action_set_queue *) ia;
4906 xlate_set_queue_action(ctx, nasq);
4909 case OFPUTIL_NXAST_POP_QUEUE:
4910 ctx->flow.skb_priority = ctx->orig_skb_priority;
4913 case OFPUTIL_NXAST_REG_MOVE:
4914 nxm_execute_reg_move((const struct nx_action_reg_move *) ia,
4918 case OFPUTIL_NXAST_REG_LOAD:
4919 nxm_execute_reg_load((const struct nx_action_reg_load *) ia,
4923 case OFPUTIL_NXAST_NOTE:
4924 /* Nothing to do. */
4927 case OFPUTIL_NXAST_SET_TUNNEL64:
4928 tun_id = ((const struct nx_action_set_tunnel64 *) ia)->tun_id;
4929 ctx->flow.tun_id = tun_id;
4932 case OFPUTIL_NXAST_MULTIPATH:
4933 nam = (const struct nx_action_multipath *) ia;
4934 multipath_execute(nam, &ctx->flow);
4937 case OFPUTIL_NXAST_AUTOPATH:
4938 naa = (const struct nx_action_autopath *) ia;
4939 xlate_autopath(ctx, naa);
4942 case OFPUTIL_NXAST_BUNDLE:
4943 ctx->ofproto->has_bundle_action = true;
4944 nab = (const struct nx_action_bundle *) ia;
4945 xlate_output_action__(ctx, bundle_execute(nab, &ctx->flow,
4950 case OFPUTIL_NXAST_BUNDLE_LOAD:
4951 ctx->ofproto->has_bundle_action = true;
4952 nab = (const struct nx_action_bundle *) ia;
4953 bundle_execute_load(nab, &ctx->flow, slave_enabled_cb,
4957 case OFPUTIL_NXAST_OUTPUT_REG:
4958 naor = (const struct nx_action_output_reg *) ia;
4959 xlate_output_reg_action(ctx, naor);
4962 case OFPUTIL_NXAST_LEARN:
4963 ctx->has_learn = true;
4964 if (ctx->may_learn) {
4965 xlate_learn_action(ctx, (const struct nx_action_learn *) ia);
4969 case OFPUTIL_NXAST_DEC_TTL:
4970 if (compose_dec_ttl(ctx)) {
4975 case OFPUTIL_NXAST_EXIT:
4979 case OFPUTIL_NXAST_FIN_TIMEOUT:
4980 ctx->has_fin_timeout = true;
4981 xlate_fin_timeout(ctx, (const struct nx_action_fin_timeout *) ia);
4984 case OFPUTIL_NXAST_CONTROLLER:
4985 nac = (const struct nx_action_controller *) ia;
4986 execute_controller_action(ctx, ntohs(nac->max_len), nac->reason,
4987 ntohs(nac->controller_id));
4993 /* We've let OFPP_NORMAL and the learning action look at the packet,
4994 * so drop it now if forwarding is disabled. */
4995 if (port && !stp_forward_in_state(port->stp_state)) {
4996 ofpbuf_clear(ctx->odp_actions);
4997 add_sflow_action(ctx);
5000 ctx->rule->up.evictable = was_evictable;
5005 action_xlate_ctx_init(struct action_xlate_ctx *ctx,
5006 struct ofproto_dpif *ofproto, const struct flow *flow,
5007 ovs_be16 initial_tci, struct rule_dpif *rule,
5008 uint8_t tcp_flags, const struct ofpbuf *packet)
5010 ctx->ofproto = ofproto;
5012 ctx->base_flow = ctx->flow;
5013 ctx->base_flow.tun_id = 0;
5014 ctx->base_flow.vlan_tci = initial_tci;
5016 ctx->packet = packet;
5017 ctx->may_learn = packet != NULL;
5018 ctx->tcp_flags = tcp_flags;
5019 ctx->resubmit_hook = NULL;
5022 static struct ofpbuf *
5023 xlate_actions(struct action_xlate_ctx *ctx,
5024 const union ofp_action *in, size_t n_in)
5026 struct flow orig_flow = ctx->flow;
5028 COVERAGE_INC(ofproto_dpif_xlate);
5030 ctx->odp_actions = ofpbuf_new(512);
5031 ofpbuf_reserve(ctx->odp_actions, NL_A_U32_SIZE);
5033 ctx->may_set_up_flow = true;
5034 ctx->has_learn = false;
5035 ctx->has_normal = false;
5036 ctx->has_fin_timeout = false;
5037 ctx->nf_output_iface = NF_OUT_DROP;
5040 ctx->orig_skb_priority = ctx->flow.skb_priority;
5044 if (ctx->flow.nw_frag & FLOW_NW_FRAG_ANY) {
5045 switch (ctx->ofproto->up.frag_handling) {
5046 case OFPC_FRAG_NORMAL:
5047 /* We must pretend that transport ports are unavailable. */
5048 ctx->flow.tp_src = ctx->base_flow.tp_src = htons(0);
5049 ctx->flow.tp_dst = ctx->base_flow.tp_dst = htons(0);
5052 case OFPC_FRAG_DROP:
5053 return ctx->odp_actions;
5055 case OFPC_FRAG_REASM:
5058 case OFPC_FRAG_NX_MATCH:
5059 /* Nothing to do. */
5062 case OFPC_INVALID_TTL_TO_CONTROLLER:
5067 if (process_special(ctx->ofproto, &ctx->flow, ctx->packet)) {
5068 ctx->may_set_up_flow = false;
5069 return ctx->odp_actions;
5071 add_sflow_action(ctx);
5072 do_xlate_actions(in, n_in, ctx);
5074 if (!connmgr_may_set_up_flow(ctx->ofproto->up.connmgr, &ctx->flow,
5075 ctx->odp_actions->data,
5076 ctx->odp_actions->size)) {
5077 ctx->may_set_up_flow = false;
5079 && connmgr_msg_in_hook(ctx->ofproto->up.connmgr, &ctx->flow,
5081 compose_output_action(ctx, OFPP_LOCAL);
5084 add_mirror_actions(ctx, &orig_flow);
5085 fix_sflow_action(ctx);
5088 return ctx->odp_actions;
5091 /* OFPP_NORMAL implementation. */
5093 static struct ofport_dpif *ofbundle_get_a_port(const struct ofbundle *);
5095 /* Given 'vid', the VID obtained from the 802.1Q header that was received as
5096 * part of a packet (specify 0 if there was no 802.1Q header), and 'in_bundle',
5097 * the bundle on which the packet was received, returns the VLAN to which the
5100 * Both 'vid' and the return value are in the range 0...4095. */
5102 input_vid_to_vlan(const struct ofbundle *in_bundle, uint16_t vid)
5104 switch (in_bundle->vlan_mode) {
5105 case PORT_VLAN_ACCESS:
5106 return in_bundle->vlan;
5109 case PORT_VLAN_TRUNK:
5112 case PORT_VLAN_NATIVE_UNTAGGED:
5113 case PORT_VLAN_NATIVE_TAGGED:
5114 return vid ? vid : in_bundle->vlan;
5121 /* Checks whether a packet with the given 'vid' may ingress on 'in_bundle'.
5122 * If so, returns true. Otherwise, returns false and, if 'warn' is true, logs
5125 * 'vid' should be the VID obtained from the 802.1Q header that was received as
5126 * part of a packet (specify 0 if there was no 802.1Q header), in the range
5129 input_vid_is_valid(uint16_t vid, struct ofbundle *in_bundle, bool warn)
5131 /* Allow any VID on the OFPP_NONE port. */
5132 if (in_bundle == &ofpp_none_bundle) {
5136 switch (in_bundle->vlan_mode) {
5137 case PORT_VLAN_ACCESS:
5140 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
5141 VLOG_WARN_RL(&rl, "bridge %s: dropping VLAN %"PRIu16" tagged "
5142 "packet received on port %s configured as VLAN "
5143 "%"PRIu16" access port",
5144 in_bundle->ofproto->up.name, vid,
5145 in_bundle->name, in_bundle->vlan);
5151 case PORT_VLAN_NATIVE_UNTAGGED:
5152 case PORT_VLAN_NATIVE_TAGGED:
5154 /* Port must always carry its native VLAN. */
5158 case PORT_VLAN_TRUNK:
5159 if (!ofbundle_includes_vlan(in_bundle, vid)) {
5161 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
5162 VLOG_WARN_RL(&rl, "bridge %s: dropping VLAN %"PRIu16" packet "
5163 "received on port %s not configured for trunking "
5165 in_bundle->ofproto->up.name, vid,
5166 in_bundle->name, vid);
5178 /* Given 'vlan', the VLAN that a packet belongs to, and
5179 * 'out_bundle', a bundle on which the packet is to be output, returns the VID
5180 * that should be included in the 802.1Q header. (If the return value is 0,
5181 * then the 802.1Q header should only be included in the packet if there is a
5184 * Both 'vlan' and the return value are in the range 0...4095. */
5186 output_vlan_to_vid(const struct ofbundle *out_bundle, uint16_t vlan)
5188 switch (out_bundle->vlan_mode) {
5189 case PORT_VLAN_ACCESS:
5192 case PORT_VLAN_TRUNK:
5193 case PORT_VLAN_NATIVE_TAGGED:
5196 case PORT_VLAN_NATIVE_UNTAGGED:
5197 return vlan == out_bundle->vlan ? 0 : vlan;
5205 output_normal(struct action_xlate_ctx *ctx, const struct ofbundle *out_bundle,
5208 struct ofport_dpif *port;
5210 ovs_be16 tci, old_tci;
5212 vid = output_vlan_to_vid(out_bundle, vlan);
5213 if (!out_bundle->bond) {
5214 port = ofbundle_get_a_port(out_bundle);
5216 port = bond_choose_output_slave(out_bundle->bond, &ctx->flow,
5219 /* No slaves enabled, so drop packet. */
5224 old_tci = ctx->flow.vlan_tci;
5226 if (tci || out_bundle->use_priority_tags) {
5227 tci |= ctx->flow.vlan_tci & htons(VLAN_PCP_MASK);
5229 tci |= htons(VLAN_CFI);
5232 ctx->flow.vlan_tci = tci;
5234 compose_output_action(ctx, port->up.ofp_port);
5235 ctx->flow.vlan_tci = old_tci;
5239 mirror_mask_ffs(mirror_mask_t mask)
5241 BUILD_ASSERT_DECL(sizeof(unsigned int) >= sizeof(mask));
5246 ofbundle_trunks_vlan(const struct ofbundle *bundle, uint16_t vlan)
5248 return (bundle->vlan_mode != PORT_VLAN_ACCESS
5249 && (!bundle->trunks || bitmap_is_set(bundle->trunks, vlan)));
5253 ofbundle_includes_vlan(const struct ofbundle *bundle, uint16_t vlan)
5255 return vlan == bundle->vlan || ofbundle_trunks_vlan(bundle, vlan);
5258 /* Returns an arbitrary interface within 'bundle'. */
5259 static struct ofport_dpif *
5260 ofbundle_get_a_port(const struct ofbundle *bundle)
5262 return CONTAINER_OF(list_front(&bundle->ports),
5263 struct ofport_dpif, bundle_node);
5267 vlan_is_mirrored(const struct ofmirror *m, int vlan)
5269 return !m->vlans || bitmap_is_set(m->vlans, vlan);
5272 /* Returns true if a packet with Ethernet destination MAC 'dst' may be mirrored
5273 * to a VLAN. In general most packets may be mirrored but we want to drop
5274 * protocols that may confuse switches. */
5276 eth_dst_may_rspan(const uint8_t dst[ETH_ADDR_LEN])
5278 /* If you change this function's behavior, please update corresponding
5279 * documentation in vswitch.xml at the same time. */
5280 if (dst[0] != 0x01) {
5281 /* All the currently banned MACs happen to start with 01 currently, so
5282 * this is a quick way to eliminate most of the good ones. */
5284 if (eth_addr_is_reserved(dst)) {
5285 /* Drop STP, IEEE pause frames, and other reserved protocols
5286 * (01-80-c2-00-00-0x). */
5290 if (dst[0] == 0x01 && dst[1] == 0x00 && dst[2] == 0x0c) {
5292 if ((dst[3] & 0xfe) == 0xcc &&
5293 (dst[4] & 0xfe) == 0xcc &&
5294 (dst[5] & 0xfe) == 0xcc) {
5295 /* Drop the following protocols plus others following the same
5298 CDP, VTP, DTP, PAgP (01-00-0c-cc-cc-cc)
5299 Spanning Tree PVSTP+ (01-00-0c-cc-cc-cd)
5300 STP Uplink Fast (01-00-0c-cd-cd-cd) */
5304 if (!(dst[3] | dst[4] | dst[5])) {
5305 /* Drop Inter Switch Link packets (01-00-0c-00-00-00). */
5314 add_mirror_actions(struct action_xlate_ctx *ctx, const struct flow *orig_flow)
5316 struct ofproto_dpif *ofproto = ctx->ofproto;
5317 mirror_mask_t mirrors;
5318 struct ofbundle *in_bundle;
5321 const struct nlattr *a;
5324 in_bundle = lookup_input_bundle(ctx->ofproto, orig_flow->in_port,
5325 ctx->packet != NULL);
5329 mirrors = in_bundle->src_mirrors;
5331 /* Drop frames on bundles reserved for mirroring. */
5332 if (in_bundle->mirror_out) {
5333 if (ctx->packet != NULL) {
5334 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
5335 VLOG_WARN_RL(&rl, "bridge %s: dropping packet received on port "
5336 "%s, which is reserved exclusively for mirroring",
5337 ctx->ofproto->up.name, in_bundle->name);
5343 vid = vlan_tci_to_vid(orig_flow->vlan_tci);
5344 if (!input_vid_is_valid(vid, in_bundle, ctx->packet != NULL)) {
5347 vlan = input_vid_to_vlan(in_bundle, vid);
5349 /* Look at the output ports to check for destination selections. */
5351 NL_ATTR_FOR_EACH (a, left, ctx->odp_actions->data,
5352 ctx->odp_actions->size) {
5353 enum ovs_action_attr type = nl_attr_type(a);
5354 struct ofport_dpif *ofport;
5356 if (type != OVS_ACTION_ATTR_OUTPUT) {
5360 ofport = get_odp_port(ofproto, nl_attr_get_u32(a));
5361 if (ofport && ofport->bundle) {
5362 mirrors |= ofport->bundle->dst_mirrors;
5370 /* Restore the original packet before adding the mirror actions. */
5371 ctx->flow = *orig_flow;
5376 m = ofproto->mirrors[mirror_mask_ffs(mirrors) - 1];
5378 if (!vlan_is_mirrored(m, vlan)) {
5379 mirrors &= mirrors - 1;
5383 mirrors &= ~m->dup_mirrors;
5384 ctx->mirrors |= m->dup_mirrors;
5386 output_normal(ctx, m->out, vlan);
5387 } else if (eth_dst_may_rspan(orig_flow->dl_dst)
5388 && vlan != m->out_vlan) {
5389 struct ofbundle *bundle;
5391 HMAP_FOR_EACH (bundle, hmap_node, &ofproto->bundles) {
5392 if (ofbundle_includes_vlan(bundle, m->out_vlan)
5393 && !bundle->mirror_out) {
5394 output_normal(ctx, bundle, m->out_vlan);
5402 update_mirror_stats(struct ofproto_dpif *ofproto, mirror_mask_t mirrors,
5403 uint64_t packets, uint64_t bytes)
5409 for (; mirrors; mirrors &= mirrors - 1) {
5412 m = ofproto->mirrors[mirror_mask_ffs(mirrors) - 1];
5415 /* In normal circumstances 'm' will not be NULL. However,
5416 * if mirrors are reconfigured, we can temporarily get out
5417 * of sync in facet_revalidate(). We could "correct" the
5418 * mirror list before reaching here, but doing that would
5419 * not properly account the traffic stats we've currently
5420 * accumulated for previous mirror configuration. */
5424 m->packet_count += packets;
5425 m->byte_count += bytes;
5429 /* A VM broadcasts a gratuitous ARP to indicate that it has resumed after
5430 * migration. Older Citrix-patched Linux DomU used gratuitous ARP replies to
5431 * indicate this; newer upstream kernels use gratuitous ARP requests. */
5433 is_gratuitous_arp(const struct flow *flow)
5435 return (flow->dl_type == htons(ETH_TYPE_ARP)
5436 && eth_addr_is_broadcast(flow->dl_dst)
5437 && (flow->nw_proto == ARP_OP_REPLY
5438 || (flow->nw_proto == ARP_OP_REQUEST
5439 && flow->nw_src == flow->nw_dst)));
5443 update_learning_table(struct ofproto_dpif *ofproto,
5444 const struct flow *flow, int vlan,
5445 struct ofbundle *in_bundle)
5447 struct mac_entry *mac;
5449 /* Don't learn the OFPP_NONE port. */
5450 if (in_bundle == &ofpp_none_bundle) {
5454 if (!mac_learning_may_learn(ofproto->ml, flow->dl_src, vlan)) {
5458 mac = mac_learning_insert(ofproto->ml, flow->dl_src, vlan);
5459 if (is_gratuitous_arp(flow)) {
5460 /* We don't want to learn from gratuitous ARP packets that are
5461 * reflected back over bond slaves so we lock the learning table. */
5462 if (!in_bundle->bond) {
5463 mac_entry_set_grat_arp_lock(mac);
5464 } else if (mac_entry_is_grat_arp_locked(mac)) {
5469 if (mac_entry_is_new(mac) || mac->port.p != in_bundle) {
5470 /* The log messages here could actually be useful in debugging,
5471 * so keep the rate limit relatively high. */
5472 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(30, 300);
5473 VLOG_DBG_RL(&rl, "bridge %s: learned that "ETH_ADDR_FMT" is "
5474 "on port %s in VLAN %d",
5475 ofproto->up.name, ETH_ADDR_ARGS(flow->dl_src),
5476 in_bundle->name, vlan);
5478 mac->port.p = in_bundle;
5479 tag_set_add(&ofproto->revalidate_set,
5480 mac_learning_changed(ofproto->ml, mac));
5484 static struct ofbundle *
5485 lookup_input_bundle(struct ofproto_dpif *ofproto, uint16_t in_port, bool warn)
5487 struct ofport_dpif *ofport;
5489 /* Special-case OFPP_NONE, which a controller may use as the ingress
5490 * port for traffic that it is sourcing. */
5491 if (in_port == OFPP_NONE) {
5492 return &ofpp_none_bundle;
5495 /* Find the port and bundle for the received packet. */
5496 ofport = get_ofp_port(ofproto, in_port);
5497 if (ofport && ofport->bundle) {
5498 return ofport->bundle;
5501 /* Odd. A few possible reasons here:
5503 * - We deleted a port but there are still a few packets queued up
5506 * - Someone externally added a port (e.g. "ovs-dpctl add-if") that
5507 * we don't know about.
5509 * - The ofproto client didn't configure the port as part of a bundle.
5512 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
5514 VLOG_WARN_RL(&rl, "bridge %s: received packet on unknown "
5515 "port %"PRIu16, ofproto->up.name, in_port);
5520 /* Determines whether packets in 'flow' within 'ofproto' should be forwarded or
5521 * dropped. Returns true if they may be forwarded, false if they should be
5524 * 'in_port' must be the ofport_dpif that corresponds to flow->in_port.
5525 * 'in_port' must be part of a bundle (e.g. in_port->bundle must be nonnull).
5527 * 'vlan' must be the VLAN that corresponds to flow->vlan_tci on 'in_port', as
5528 * returned by input_vid_to_vlan(). It must be a valid VLAN for 'in_port', as
5529 * checked by input_vid_is_valid().
5531 * May also add tags to '*tags', although the current implementation only does
5532 * so in one special case.
5535 is_admissible(struct ofproto_dpif *ofproto, const struct flow *flow,
5536 struct ofport_dpif *in_port, uint16_t vlan, tag_type *tags)
5538 struct ofbundle *in_bundle = in_port->bundle;
5540 /* Drop frames for reserved multicast addresses
5541 * only if forward_bpdu option is absent. */
5542 if (eth_addr_is_reserved(flow->dl_dst) && !ofproto->up.forward_bpdu) {
5546 if (in_bundle->bond) {
5547 struct mac_entry *mac;
5549 switch (bond_check_admissibility(in_bundle->bond, in_port,
5550 flow->dl_dst, tags)) {
5557 case BV_DROP_IF_MOVED:
5558 mac = mac_learning_lookup(ofproto->ml, flow->dl_src, vlan, NULL);
5559 if (mac && mac->port.p != in_bundle &&
5560 (!is_gratuitous_arp(flow)
5561 || mac_entry_is_grat_arp_locked(mac))) {
5572 xlate_normal(struct action_xlate_ctx *ctx)
5574 struct ofport_dpif *in_port;
5575 struct ofbundle *in_bundle;
5576 struct mac_entry *mac;
5580 ctx->has_normal = true;
5582 in_bundle = lookup_input_bundle(ctx->ofproto, ctx->flow.in_port,
5583 ctx->packet != NULL);
5588 /* We know 'in_port' exists unless it is "ofpp_none_bundle",
5589 * since lookup_input_bundle() succeeded. */
5590 in_port = get_ofp_port(ctx->ofproto, ctx->flow.in_port);
5592 /* Drop malformed frames. */
5593 if (ctx->flow.dl_type == htons(ETH_TYPE_VLAN) &&
5594 !(ctx->flow.vlan_tci & htons(VLAN_CFI))) {
5595 if (ctx->packet != NULL) {
5596 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
5597 VLOG_WARN_RL(&rl, "bridge %s: dropping packet with partial "
5598 "VLAN tag received on port %s",
5599 ctx->ofproto->up.name, in_bundle->name);
5604 /* Drop frames on bundles reserved for mirroring. */
5605 if (in_bundle->mirror_out) {
5606 if (ctx->packet != NULL) {
5607 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
5608 VLOG_WARN_RL(&rl, "bridge %s: dropping packet received on port "
5609 "%s, which is reserved exclusively for mirroring",
5610 ctx->ofproto->up.name, in_bundle->name);
5616 vid = vlan_tci_to_vid(ctx->flow.vlan_tci);
5617 if (!input_vid_is_valid(vid, in_bundle, ctx->packet != NULL)) {
5620 vlan = input_vid_to_vlan(in_bundle, vid);
5622 /* Check other admissibility requirements. */
5624 !is_admissible(ctx->ofproto, &ctx->flow, in_port, vlan, &ctx->tags)) {
5628 /* Learn source MAC. */
5629 if (ctx->may_learn) {
5630 update_learning_table(ctx->ofproto, &ctx->flow, vlan, in_bundle);
5633 /* Determine output bundle. */
5634 mac = mac_learning_lookup(ctx->ofproto->ml, ctx->flow.dl_dst, vlan,
5637 if (mac->port.p != in_bundle) {
5638 output_normal(ctx, mac->port.p, vlan);
5641 struct ofbundle *bundle;
5643 HMAP_FOR_EACH (bundle, hmap_node, &ctx->ofproto->bundles) {
5644 if (bundle != in_bundle
5645 && ofbundle_includes_vlan(bundle, vlan)
5646 && bundle->floodable
5647 && !bundle->mirror_out) {
5648 output_normal(ctx, bundle, vlan);
5651 ctx->nf_output_iface = NF_OUT_FLOOD;
5655 /* Optimized flow revalidation.
5657 * It's a difficult problem, in general, to tell which facets need to have
5658 * their actions recalculated whenever the OpenFlow flow table changes. We
5659 * don't try to solve that general problem: for most kinds of OpenFlow flow
5660 * table changes, we recalculate the actions for every facet. This is
5661 * relatively expensive, but it's good enough if the OpenFlow flow table
5662 * doesn't change very often.
5664 * However, we can expect one particular kind of OpenFlow flow table change to
5665 * happen frequently: changes caused by MAC learning. To avoid wasting a lot
5666 * of CPU on revalidating every facet whenever MAC learning modifies the flow
5667 * table, we add a special case that applies to flow tables in which every rule
5668 * has the same form (that is, the same wildcards), except that the table is
5669 * also allowed to have a single "catch-all" flow that matches all packets. We
5670 * optimize this case by tagging all of the facets that resubmit into the table
5671 * and invalidating the same tag whenever a flow changes in that table. The
5672 * end result is that we revalidate just the facets that need it (and sometimes
5673 * a few more, but not all of the facets or even all of the facets that
5674 * resubmit to the table modified by MAC learning). */
5676 /* Calculates the tag to use for 'flow' and wildcards 'wc' when it is inserted
5677 * into an OpenFlow table with the given 'basis'. */
5679 rule_calculate_tag(const struct flow *flow, const struct flow_wildcards *wc,
5682 if (flow_wildcards_is_catchall(wc)) {
5685 struct flow tag_flow = *flow;
5686 flow_zero_wildcards(&tag_flow, wc);
5687 return tag_create_deterministic(flow_hash(&tag_flow, secret));
5691 /* Following a change to OpenFlow table 'table_id' in 'ofproto', update the
5692 * taggability of that table.
5694 * This function must be called after *each* change to a flow table. If you
5695 * skip calling it on some changes then the pointer comparisons at the end can
5696 * be invalid if you get unlucky. For example, if a flow removal causes a
5697 * cls_table to be destroyed and then a flow insertion causes a cls_table with
5698 * different wildcards to be created with the same address, then this function
5699 * will incorrectly skip revalidation. */
5701 table_update_taggable(struct ofproto_dpif *ofproto, uint8_t table_id)
5703 struct table_dpif *table = &ofproto->tables[table_id];
5704 const struct oftable *oftable = &ofproto->up.tables[table_id];
5705 struct cls_table *catchall, *other;
5706 struct cls_table *t;
5708 catchall = other = NULL;
5710 switch (hmap_count(&oftable->cls.tables)) {
5712 /* We could tag this OpenFlow table but it would make the logic a
5713 * little harder and it's a corner case that doesn't seem worth it
5719 HMAP_FOR_EACH (t, hmap_node, &oftable->cls.tables) {
5720 if (cls_table_is_catchall(t)) {
5722 } else if (!other) {
5725 /* Indicate that we can't tag this by setting both tables to
5726 * NULL. (We know that 'catchall' is already NULL.) */
5733 /* Can't tag this table. */
5737 if (table->catchall_table != catchall || table->other_table != other) {
5738 table->catchall_table = catchall;
5739 table->other_table = other;
5740 ofproto->need_revalidate = true;
5744 /* Given 'rule' that has changed in some way (either it is a rule being
5745 * inserted, a rule being deleted, or a rule whose actions are being
5746 * modified), marks facets for revalidation to ensure that packets will be
5747 * forwarded correctly according to the new state of the flow table.
5749 * This function must be called after *each* change to a flow table. See
5750 * the comment on table_update_taggable() for more information. */
5752 rule_invalidate(const struct rule_dpif *rule)
5754 struct ofproto_dpif *ofproto = ofproto_dpif_cast(rule->up.ofproto);
5756 table_update_taggable(ofproto, rule->up.table_id);
5758 if (!ofproto->need_revalidate) {
5759 struct table_dpif *table = &ofproto->tables[rule->up.table_id];
5761 if (table->other_table && rule->tag) {
5762 tag_set_add(&ofproto->revalidate_set, rule->tag);
5764 ofproto->need_revalidate = true;
5770 set_frag_handling(struct ofproto *ofproto_,
5771 enum ofp_config_flags frag_handling)
5773 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
5775 if (frag_handling != OFPC_FRAG_REASM) {
5776 ofproto->need_revalidate = true;
5784 packet_out(struct ofproto *ofproto_, struct ofpbuf *packet,
5785 const struct flow *flow,
5786 const union ofp_action *ofp_actions, size_t n_ofp_actions)
5788 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
5791 if (flow->in_port >= ofproto->max_ports && flow->in_port < OFPP_MAX) {
5792 return OFPERR_NXBRC_BAD_IN_PORT;
5795 error = validate_actions(ofp_actions, n_ofp_actions, flow,
5796 ofproto->max_ports);
5798 struct odputil_keybuf keybuf;
5799 struct ofpbuf *odp_actions;
5800 struct ofproto_push push;
5803 ofpbuf_use_stack(&key, &keybuf, sizeof keybuf);
5804 odp_flow_key_from_flow(&key, flow);
5806 action_xlate_ctx_init(&push.ctx, ofproto, flow, flow->vlan_tci, NULL,
5807 packet_get_tcp_flags(packet, flow), packet);
5809 /* Ensure that resubmits in 'ofp_actions' get accounted to their
5810 * matching rules. */
5812 push.bytes = packet->size;
5813 push.used = time_msec();
5814 push.ctx.resubmit_hook = push_resubmit;
5816 odp_actions = xlate_actions(&push.ctx, ofp_actions, n_ofp_actions);
5817 dpif_execute(ofproto->dpif, key.data, key.size,
5818 odp_actions->data, odp_actions->size, packet);
5819 ofpbuf_delete(odp_actions);
5827 set_netflow(struct ofproto *ofproto_,
5828 const struct netflow_options *netflow_options)
5830 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
5832 if (netflow_options) {
5833 if (!ofproto->netflow) {
5834 ofproto->netflow = netflow_create();
5836 return netflow_set_options(ofproto->netflow, netflow_options);
5838 netflow_destroy(ofproto->netflow);
5839 ofproto->netflow = NULL;
5845 get_netflow_ids(const struct ofproto *ofproto_,
5846 uint8_t *engine_type, uint8_t *engine_id)
5848 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
5850 dpif_get_netflow_ids(ofproto->dpif, engine_type, engine_id);
5854 send_active_timeout(struct ofproto_dpif *ofproto, struct facet *facet)
5856 if (!facet_is_controller_flow(facet) &&
5857 netflow_active_timeout_expired(ofproto->netflow, &facet->nf_flow)) {
5858 struct subfacet *subfacet;
5859 struct ofexpired expired;
5861 LIST_FOR_EACH (subfacet, list_node, &facet->subfacets) {
5862 if (subfacet->installed) {
5863 struct dpif_flow_stats stats;
5865 subfacet_install(subfacet, subfacet->actions,
5866 subfacet->actions_len, &stats);
5867 subfacet_update_stats(subfacet, &stats);
5871 expired.flow = facet->flow;
5872 expired.packet_count = facet->packet_count;
5873 expired.byte_count = facet->byte_count;
5874 expired.used = facet->used;
5875 netflow_expire(ofproto->netflow, &facet->nf_flow, &expired);
5880 send_netflow_active_timeouts(struct ofproto_dpif *ofproto)
5882 struct facet *facet;
5884 HMAP_FOR_EACH (facet, hmap_node, &ofproto->facets) {
5885 send_active_timeout(ofproto, facet);
5889 static struct ofproto_dpif *
5890 ofproto_dpif_lookup(const char *name)
5892 struct ofproto_dpif *ofproto;
5894 HMAP_FOR_EACH_WITH_HASH (ofproto, all_ofproto_dpifs_node,
5895 hash_string(name, 0), &all_ofproto_dpifs) {
5896 if (!strcmp(ofproto->up.name, name)) {
5904 ofproto_unixctl_fdb_flush(struct unixctl_conn *conn, int argc,
5905 const char *argv[], void *aux OVS_UNUSED)
5907 struct ofproto_dpif *ofproto;
5910 ofproto = ofproto_dpif_lookup(argv[1]);
5912 unixctl_command_reply_error(conn, "no such bridge");
5915 mac_learning_flush(ofproto->ml, &ofproto->revalidate_set);
5917 HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
5918 mac_learning_flush(ofproto->ml, &ofproto->revalidate_set);
5922 unixctl_command_reply(conn, "table successfully flushed");
5926 ofproto_unixctl_fdb_show(struct unixctl_conn *conn, int argc OVS_UNUSED,
5927 const char *argv[], void *aux OVS_UNUSED)
5929 struct ds ds = DS_EMPTY_INITIALIZER;
5930 const struct ofproto_dpif *ofproto;
5931 const struct mac_entry *e;
5933 ofproto = ofproto_dpif_lookup(argv[1]);
5935 unixctl_command_reply_error(conn, "no such bridge");
5939 ds_put_cstr(&ds, " port VLAN MAC Age\n");
5940 LIST_FOR_EACH (e, lru_node, &ofproto->ml->lrus) {
5941 struct ofbundle *bundle = e->port.p;
5942 ds_put_format(&ds, "%5d %4d "ETH_ADDR_FMT" %3d\n",
5943 ofbundle_get_a_port(bundle)->odp_port,
5944 e->vlan, ETH_ADDR_ARGS(e->mac),
5945 mac_entry_age(ofproto->ml, e));
5947 unixctl_command_reply(conn, ds_cstr(&ds));
5951 struct ofproto_trace {
5952 struct action_xlate_ctx ctx;
5958 trace_format_rule(struct ds *result, uint8_t table_id, int level,
5959 const struct rule_dpif *rule)
5961 ds_put_char_multiple(result, '\t', level);
5963 ds_put_cstr(result, "No match\n");
5967 ds_put_format(result, "Rule: table=%"PRIu8" cookie=%#"PRIx64" ",
5968 table_id, ntohll(rule->up.flow_cookie));
5969 cls_rule_format(&rule->up.cr, result);
5970 ds_put_char(result, '\n');
5972 ds_put_char_multiple(result, '\t', level);
5973 ds_put_cstr(result, "OpenFlow ");
5974 ofp_print_actions(result, rule->up.actions, rule->up.n_actions);
5975 ds_put_char(result, '\n');
5979 trace_format_flow(struct ds *result, int level, const char *title,
5980 struct ofproto_trace *trace)
5982 ds_put_char_multiple(result, '\t', level);
5983 ds_put_format(result, "%s: ", title);
5984 if (flow_equal(&trace->ctx.flow, &trace->flow)) {
5985 ds_put_cstr(result, "unchanged");
5987 flow_format(result, &trace->ctx.flow);
5988 trace->flow = trace->ctx.flow;
5990 ds_put_char(result, '\n');
5994 trace_format_regs(struct ds *result, int level, const char *title,
5995 struct ofproto_trace *trace)
5999 ds_put_char_multiple(result, '\t', level);
6000 ds_put_format(result, "%s:", title);
6001 for (i = 0; i < FLOW_N_REGS; i++) {
6002 ds_put_format(result, " reg%zu=0x%"PRIx32, i, trace->flow.regs[i]);
6004 ds_put_char(result, '\n');
6008 trace_format_odp(struct ds *result, int level, const char *title,
6009 struct ofproto_trace *trace)
6011 struct ofpbuf *odp_actions = trace->ctx.odp_actions;
6013 ds_put_char_multiple(result, '\t', level);
6014 ds_put_format(result, "%s: ", title);
6015 format_odp_actions(result, odp_actions->data, odp_actions->size);
6016 ds_put_char(result, '\n');
6020 trace_resubmit(struct action_xlate_ctx *ctx, struct rule_dpif *rule)
6022 struct ofproto_trace *trace = CONTAINER_OF(ctx, struct ofproto_trace, ctx);
6023 struct ds *result = trace->result;
6025 ds_put_char(result, '\n');
6026 trace_format_flow(result, ctx->recurse + 1, "Resubmitted flow", trace);
6027 trace_format_regs(result, ctx->recurse + 1, "Resubmitted regs", trace);
6028 trace_format_odp(result, ctx->recurse + 1, "Resubmitted odp", trace);
6029 trace_format_rule(result, ctx->table_id, ctx->recurse + 1, rule);
6033 ofproto_unixctl_trace(struct unixctl_conn *conn, int argc, const char *argv[],
6034 void *aux OVS_UNUSED)
6036 const char *dpname = argv[1];
6037 struct ofproto_dpif *ofproto;
6038 struct ofpbuf odp_key;
6039 struct ofpbuf *packet;
6040 struct rule_dpif *rule;
6041 ovs_be16 initial_tci;
6047 ofpbuf_init(&odp_key, 0);
6050 ofproto = ofproto_dpif_lookup(dpname);
6052 unixctl_command_reply_error(conn, "Unknown ofproto (use ofproto/list "
6056 if (argc == 3 || (argc == 4 && !strcmp(argv[3], "-generate"))) {
6057 /* ofproto/trace dpname flow [-generate] */
6058 const char *flow_s = argv[2];
6059 const char *generate_s = argv[3];
6062 /* Convert string to datapath key. */
6063 ofpbuf_init(&odp_key, 0);
6064 error = odp_flow_key_from_string(flow_s, NULL, &odp_key);
6066 unixctl_command_reply_error(conn, "Bad flow syntax");
6070 /* Convert odp_key to flow. */
6071 error = ofproto_dpif_extract_flow_key(ofproto, odp_key.data,
6072 odp_key.size, &flow,
6073 &initial_tci, NULL);
6074 if (error == ODP_FIT_ERROR) {
6075 unixctl_command_reply_error(conn, "Invalid flow");
6079 /* Generate a packet, if requested. */
6081 packet = ofpbuf_new(0);
6082 flow_compose(packet, &flow);
6084 } else if (argc == 6) {
6085 /* ofproto/trace dpname priority tun_id in_port packet */
6086 const char *priority_s = argv[2];
6087 const char *tun_id_s = argv[3];
6088 const char *in_port_s = argv[4];
6089 const char *packet_s = argv[5];
6090 uint16_t in_port = ofp_port_to_odp_port(atoi(in_port_s));
6091 ovs_be64 tun_id = htonll(strtoull(tun_id_s, NULL, 0));
6092 uint32_t priority = atoi(priority_s);
6095 msg = eth_from_hex(packet_s, &packet);
6097 unixctl_command_reply_error(conn, msg);
6101 ds_put_cstr(&result, "Packet: ");
6102 s = ofp_packet_to_string(packet->data, packet->size);
6103 ds_put_cstr(&result, s);
6106 flow_extract(packet, priority, tun_id, in_port, &flow);
6107 initial_tci = flow.vlan_tci;
6109 unixctl_command_reply_error(conn, "Bad command syntax");
6113 ds_put_cstr(&result, "Flow: ");
6114 flow_format(&result, &flow);
6115 ds_put_char(&result, '\n');
6117 rule = rule_dpif_lookup(ofproto, &flow, 0);
6118 trace_format_rule(&result, 0, 0, rule);
6120 struct ofproto_trace trace;
6121 struct ofpbuf *odp_actions;
6124 tcp_flags = packet ? packet_get_tcp_flags(packet, &flow) : 0;
6125 trace.result = &result;
6127 action_xlate_ctx_init(&trace.ctx, ofproto, &flow, initial_tci,
6128 rule, tcp_flags, packet);
6129 trace.ctx.resubmit_hook = trace_resubmit;
6130 odp_actions = xlate_actions(&trace.ctx,
6131 rule->up.actions, rule->up.n_actions);
6133 ds_put_char(&result, '\n');
6134 trace_format_flow(&result, 0, "Final flow", &trace);
6135 ds_put_cstr(&result, "Datapath actions: ");
6136 format_odp_actions(&result, odp_actions->data, odp_actions->size);
6137 ofpbuf_delete(odp_actions);
6139 if (!trace.ctx.may_set_up_flow) {
6141 ds_put_cstr(&result, "\nThis flow is not cachable.");
6143 ds_put_cstr(&result, "\nThe datapath actions are incomplete--"
6144 "for complete actions, please supply a packet.");
6149 unixctl_command_reply(conn, ds_cstr(&result));
6152 ds_destroy(&result);
6153 ofpbuf_delete(packet);
6154 ofpbuf_uninit(&odp_key);
6158 ofproto_dpif_clog(struct unixctl_conn *conn OVS_UNUSED, int argc OVS_UNUSED,
6159 const char *argv[] OVS_UNUSED, void *aux OVS_UNUSED)
6162 unixctl_command_reply(conn, NULL);
6166 ofproto_dpif_unclog(struct unixctl_conn *conn OVS_UNUSED, int argc OVS_UNUSED,
6167 const char *argv[] OVS_UNUSED, void *aux OVS_UNUSED)
6170 unixctl_command_reply(conn, NULL);
6173 /* Runs a self-check of flow translations in 'ofproto'. Appends a message to
6174 * 'reply' describing the results. */
6176 ofproto_dpif_self_check__(struct ofproto_dpif *ofproto, struct ds *reply)
6178 struct facet *facet;
6182 HMAP_FOR_EACH (facet, hmap_node, &ofproto->facets) {
6183 if (!facet_check_consistency(facet)) {
6188 ofproto->need_revalidate = true;
6192 ds_put_format(reply, "%s: self-check failed (%d errors)\n",
6193 ofproto->up.name, errors);
6195 ds_put_format(reply, "%s: self-check passed\n", ofproto->up.name);
6200 ofproto_dpif_self_check(struct unixctl_conn *conn,
6201 int argc, const char *argv[], void *aux OVS_UNUSED)
6203 struct ds reply = DS_EMPTY_INITIALIZER;
6204 struct ofproto_dpif *ofproto;
6207 ofproto = ofproto_dpif_lookup(argv[1]);
6209 unixctl_command_reply_error(conn, "Unknown ofproto (use "
6210 "ofproto/list for help)");
6213 ofproto_dpif_self_check__(ofproto, &reply);
6215 HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
6216 ofproto_dpif_self_check__(ofproto, &reply);
6220 unixctl_command_reply(conn, ds_cstr(&reply));
6225 ofproto_dpif_unixctl_init(void)
6227 static bool registered;
6233 unixctl_command_register(
6235 "bridge {tun_id in_port packet | odp_flow [-generate]}",
6236 2, 5, ofproto_unixctl_trace, NULL);
6237 unixctl_command_register("fdb/flush", "[bridge]", 0, 1,
6238 ofproto_unixctl_fdb_flush, NULL);
6239 unixctl_command_register("fdb/show", "bridge", 1, 1,
6240 ofproto_unixctl_fdb_show, NULL);
6241 unixctl_command_register("ofproto/clog", "", 0, 0,
6242 ofproto_dpif_clog, NULL);
6243 unixctl_command_register("ofproto/unclog", "", 0, 0,
6244 ofproto_dpif_unclog, NULL);
6245 unixctl_command_register("ofproto/self-check", "[bridge]", 0, 1,
6246 ofproto_dpif_self_check, NULL);
6249 /* Linux VLAN device support (e.g. "eth0.10" for VLAN 10.)
6251 * This is deprecated. It is only for compatibility with broken device drivers
6252 * in old versions of Linux that do not properly support VLANs when VLAN
6253 * devices are not used. When broken device drivers are no longer in
6254 * widespread use, we will delete these interfaces. */
6257 set_realdev(struct ofport *ofport_, uint16_t realdev_ofp_port, int vid)
6259 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport_->ofproto);
6260 struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
6262 if (realdev_ofp_port == ofport->realdev_ofp_port
6263 && vid == ofport->vlandev_vid) {
6267 ofproto->need_revalidate = true;
6269 if (ofport->realdev_ofp_port) {
6272 if (realdev_ofp_port && ofport->bundle) {
6273 /* vlandevs are enslaved to their realdevs, so they are not allowed to
6274 * themselves be part of a bundle. */
6275 bundle_set(ofport->up.ofproto, ofport->bundle, NULL);
6278 ofport->realdev_ofp_port = realdev_ofp_port;
6279 ofport->vlandev_vid = vid;
6281 if (realdev_ofp_port) {
6282 vsp_add(ofport, realdev_ofp_port, vid);
6289 hash_realdev_vid(uint16_t realdev_ofp_port, int vid)
6291 return hash_2words(realdev_ofp_port, vid);
6295 vsp_realdev_to_vlandev(const struct ofproto_dpif *ofproto,
6296 uint32_t realdev_odp_port, ovs_be16 vlan_tci)
6298 if (!hmap_is_empty(&ofproto->realdev_vid_map)) {
6299 uint16_t realdev_ofp_port = odp_port_to_ofp_port(realdev_odp_port);
6300 int vid = vlan_tci_to_vid(vlan_tci);
6301 const struct vlan_splinter *vsp;
6303 HMAP_FOR_EACH_WITH_HASH (vsp, realdev_vid_node,
6304 hash_realdev_vid(realdev_ofp_port, vid),
6305 &ofproto->realdev_vid_map) {
6306 if (vsp->realdev_ofp_port == realdev_ofp_port
6307 && vsp->vid == vid) {
6308 return ofp_port_to_odp_port(vsp->vlandev_ofp_port);
6312 return realdev_odp_port;
6315 static struct vlan_splinter *
6316 vlandev_find(const struct ofproto_dpif *ofproto, uint16_t vlandev_ofp_port)
6318 struct vlan_splinter *vsp;
6320 HMAP_FOR_EACH_WITH_HASH (vsp, vlandev_node, hash_int(vlandev_ofp_port, 0),
6321 &ofproto->vlandev_map) {
6322 if (vsp->vlandev_ofp_port == vlandev_ofp_port) {
6331 vsp_vlandev_to_realdev(const struct ofproto_dpif *ofproto,
6332 uint16_t vlandev_ofp_port, int *vid)
6334 if (!hmap_is_empty(&ofproto->vlandev_map)) {
6335 const struct vlan_splinter *vsp;
6337 vsp = vlandev_find(ofproto, vlandev_ofp_port);
6342 return vsp->realdev_ofp_port;
6349 vsp_remove(struct ofport_dpif *port)
6351 struct ofproto_dpif *ofproto = ofproto_dpif_cast(port->up.ofproto);
6352 struct vlan_splinter *vsp;
6354 vsp = vlandev_find(ofproto, port->up.ofp_port);
6356 hmap_remove(&ofproto->vlandev_map, &vsp->vlandev_node);
6357 hmap_remove(&ofproto->realdev_vid_map, &vsp->realdev_vid_node);
6360 port->realdev_ofp_port = 0;
6362 VLOG_ERR("missing vlan device record");
6367 vsp_add(struct ofport_dpif *port, uint16_t realdev_ofp_port, int vid)
6369 struct ofproto_dpif *ofproto = ofproto_dpif_cast(port->up.ofproto);
6371 if (!vsp_vlandev_to_realdev(ofproto, port->up.ofp_port, NULL)
6372 && (vsp_realdev_to_vlandev(ofproto, realdev_ofp_port, htons(vid))
6373 == realdev_ofp_port)) {
6374 struct vlan_splinter *vsp;
6376 vsp = xmalloc(sizeof *vsp);
6377 hmap_insert(&ofproto->vlandev_map, &vsp->vlandev_node,
6378 hash_int(port->up.ofp_port, 0));
6379 hmap_insert(&ofproto->realdev_vid_map, &vsp->realdev_vid_node,
6380 hash_realdev_vid(realdev_ofp_port, vid));
6381 vsp->realdev_ofp_port = realdev_ofp_port;
6382 vsp->vlandev_ofp_port = port->up.ofp_port;
6385 port->realdev_ofp_port = realdev_ofp_port;
6387 VLOG_ERR("duplicate vlan device record");
6391 const struct ofproto_class ofproto_dpif_class = {
6420 port_is_lacp_current,
6421 NULL, /* rule_choose_table */
6428 rule_modify_actions,
6436 get_cfm_remote_mpids,
6440 get_stp_port_status,
6447 is_mirror_output_bundle,
6448 forward_bpdu_changed,