2 * Copyright (c) 2009, 2010, 2011, 2012 Nicira Networks.
4 * Licensed under the Apache License, Version 2.0 (the "License");
5 * you may not use this file except in compliance with the License.
6 * You may obtain a copy of the License at:
8 * http://www.apache.org/licenses/LICENSE-2.0
10 * Unless required by applicable law or agreed to in writing, software
11 * distributed under the License is distributed on an "AS IS" BASIS,
12 * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
13 * See the License for the specific language governing permissions and
14 * limitations under the License.
19 #include "ofproto/ofproto-provider.h"
26 #include "byte-order.h"
31 #include "dynamic-string.h"
32 #include "fail-open.h"
36 #include "mac-learning.h"
37 #include "meta-flow.h"
38 #include "multipath.h"
45 #include "ofp-print.h"
46 #include "ofproto-dpif-sflow.h"
47 #include "poll-loop.h"
49 #include "unaligned.h"
51 #include "vlan-bitmap.h"
54 VLOG_DEFINE_THIS_MODULE(ofproto_dpif);
56 COVERAGE_DEFINE(ofproto_dpif_ctlr_action);
57 COVERAGE_DEFINE(ofproto_dpif_expired);
58 COVERAGE_DEFINE(ofproto_dpif_no_packet_in);
59 COVERAGE_DEFINE(ofproto_dpif_xlate);
60 COVERAGE_DEFINE(facet_changed_rule);
61 COVERAGE_DEFINE(facet_invalidated);
62 COVERAGE_DEFINE(facet_revalidate);
63 COVERAGE_DEFINE(facet_unexpected);
65 /* Maximum depth of flow table recursion (due to resubmit actions) in a
66 * flow translation. */
67 #define MAX_RESUBMIT_RECURSION 32
69 /* Number of implemented OpenFlow tables. */
70 enum { N_TABLES = 255 };
71 BUILD_ASSERT_DECL(N_TABLES >= 1 && N_TABLES <= 255);
81 * - Do include packets and bytes from facets that have been deleted or
82 * whose own statistics have been folded into the rule.
84 * - Do include packets and bytes sent "by hand" that were accounted to
85 * the rule without any facet being involved (this is a rare corner
86 * case in rule_execute()).
88 * - Do not include packet or bytes that can be obtained from any facet's
89 * packet_count or byte_count member or that can be obtained from the
90 * datapath by, e.g., dpif_flow_get() for any subfacet.
92 uint64_t packet_count; /* Number of packets received. */
93 uint64_t byte_count; /* Number of bytes received. */
95 tag_type tag; /* Caches rule_calculate_tag() result. */
97 struct list facets; /* List of "struct facet"s. */
100 static struct rule_dpif *rule_dpif_cast(const struct rule *rule)
102 return rule ? CONTAINER_OF(rule, struct rule_dpif, up) : NULL;
105 static struct rule_dpif *rule_dpif_lookup(struct ofproto_dpif *,
106 const struct flow *, uint8_t table);
108 static void flow_push_stats(struct rule_dpif *, const struct flow *,
109 uint64_t packets, uint64_t bytes,
112 static tag_type rule_calculate_tag(const struct flow *,
113 const struct flow_wildcards *,
115 static void rule_invalidate(const struct rule_dpif *);
117 #define MAX_MIRRORS 32
118 typedef uint32_t mirror_mask_t;
119 #define MIRROR_MASK_C(X) UINT32_C(X)
120 BUILD_ASSERT_DECL(sizeof(mirror_mask_t) * CHAR_BIT >= MAX_MIRRORS);
122 struct ofproto_dpif *ofproto; /* Owning ofproto. */
123 size_t idx; /* In ofproto's "mirrors" array. */
124 void *aux; /* Key supplied by ofproto's client. */
125 char *name; /* Identifier for log messages. */
127 /* Selection criteria. */
128 struct hmapx srcs; /* Contains "struct ofbundle *"s. */
129 struct hmapx dsts; /* Contains "struct ofbundle *"s. */
130 unsigned long *vlans; /* Bitmap of chosen VLANs, NULL selects all. */
132 /* Output (exactly one of out == NULL and out_vlan == -1 is true). */
133 struct ofbundle *out; /* Output port or NULL. */
134 int out_vlan; /* Output VLAN or -1. */
135 mirror_mask_t dup_mirrors; /* Bitmap of mirrors with the same output. */
138 int64_t packet_count; /* Number of packets sent. */
139 int64_t byte_count; /* Number of bytes sent. */
142 static void mirror_destroy(struct ofmirror *);
143 static void update_mirror_stats(struct ofproto_dpif *ofproto,
144 mirror_mask_t mirrors,
145 uint64_t packets, uint64_t bytes);
148 struct hmap_node hmap_node; /* In struct ofproto's "bundles" hmap. */
149 struct ofproto_dpif *ofproto; /* Owning ofproto. */
150 void *aux; /* Key supplied by ofproto's client. */
151 char *name; /* Identifier for log messages. */
154 struct list ports; /* Contains "struct ofport"s. */
155 enum port_vlan_mode vlan_mode; /* VLAN mode */
156 int vlan; /* -1=trunk port, else a 12-bit VLAN ID. */
157 unsigned long *trunks; /* Bitmap of trunked VLANs, if 'vlan' == -1.
158 * NULL if all VLANs are trunked. */
159 struct lacp *lacp; /* LACP if LACP is enabled, otherwise NULL. */
160 struct bond *bond; /* Nonnull iff more than one port. */
161 bool use_priority_tags; /* Use 802.1p tag for frames in VLAN 0? */
164 bool floodable; /* True if no port has OFPUTIL_PC_NO_FLOOD set. */
166 /* Port mirroring info. */
167 mirror_mask_t src_mirrors; /* Mirrors triggered when packet received. */
168 mirror_mask_t dst_mirrors; /* Mirrors triggered when packet sent. */
169 mirror_mask_t mirror_out; /* Mirrors that output to this bundle. */
172 static void bundle_remove(struct ofport *);
173 static void bundle_update(struct ofbundle *);
174 static void bundle_destroy(struct ofbundle *);
175 static void bundle_del_port(struct ofport_dpif *);
176 static void bundle_run(struct ofbundle *);
177 static void bundle_wait(struct ofbundle *);
178 static struct ofbundle *lookup_input_bundle(struct ofproto_dpif *,
179 uint16_t in_port, bool warn);
181 /* A controller may use OFPP_NONE as the ingress port to indicate that
182 * it did not arrive on a "real" port. 'ofpp_none_bundle' exists for
183 * when an input bundle is needed for validation (e.g., mirroring or
184 * OFPP_NORMAL processing). It is not connected to an 'ofproto' or have
185 * any 'port' structs, so care must be taken when dealing with it. */
186 static struct ofbundle ofpp_none_bundle = {
188 .vlan_mode = PORT_VLAN_TRUNK
191 static void stp_run(struct ofproto_dpif *ofproto);
192 static void stp_wait(struct ofproto_dpif *ofproto);
193 static int set_stp_port(struct ofport *,
194 const struct ofproto_port_stp_settings *);
196 static bool ofbundle_includes_vlan(const struct ofbundle *, uint16_t vlan);
198 struct action_xlate_ctx {
199 /* action_xlate_ctx_init() initializes these members. */
202 struct ofproto_dpif *ofproto;
204 /* Flow to which the OpenFlow actions apply. xlate_actions() will modify
205 * this flow when actions change header fields. */
208 /* The packet corresponding to 'flow', or a null pointer if we are
209 * revalidating without a packet to refer to. */
210 const struct ofpbuf *packet;
212 /* Should OFPP_NORMAL update the MAC learning table? We want to update it
213 * if we are actually processing a packet, or if we are accounting for
214 * packets that the datapath has processed, but not if we are just
218 /* Should "learn" actions update the flow table? We want to update it if
219 * we are actually processing a packet, or in most cases if we are
220 * accounting for packets that the datapath has processed, but not if we
221 * are just revalidating. */
224 /* The rule that we are currently translating, or NULL. */
225 struct rule_dpif *rule;
227 /* Union of the set of TCP flags seen so far in this flow. (Used only by
228 * NXAST_FIN_TIMEOUT. Set to zero to avoid updating updating rules'
232 /* If nonnull, called just before executing a resubmit action. In
233 * addition, disables logging of traces when the recursion depth is
236 * This is normally null so the client has to set it manually after
237 * calling action_xlate_ctx_init(). */
238 void (*resubmit_hook)(struct action_xlate_ctx *, struct rule_dpif *);
240 /* xlate_actions() initializes and uses these members. The client might want
241 * to look at them after it returns. */
243 struct ofpbuf *odp_actions; /* Datapath actions. */
244 tag_type tags; /* Tags associated with actions. */
245 bool may_set_up_flow; /* True ordinarily; false if the actions must
246 * be reassessed for every packet. */
247 bool has_learn; /* Actions include NXAST_LEARN? */
248 bool has_normal; /* Actions output to OFPP_NORMAL? */
249 bool has_fin_timeout; /* Actions include NXAST_FIN_TIMEOUT? */
250 uint16_t nf_output_iface; /* Output interface index for NetFlow. */
251 mirror_mask_t mirrors; /* Bitmap of associated mirrors. */
253 /* xlate_actions() initializes and uses these members, but the client has no
254 * reason to look at them. */
256 int recurse; /* Recursion level, via xlate_table_action. */
257 bool max_resubmit_trigger; /* Recursed too deeply during translation. */
258 struct flow base_flow; /* Flow at the last commit. */
259 uint32_t orig_skb_priority; /* Priority when packet arrived. */
260 uint8_t table_id; /* OpenFlow table ID where flow was found. */
261 uint32_t sflow_n_outputs; /* Number of output ports. */
262 uint16_t sflow_odp_port; /* Output port for composing sFlow action. */
263 uint16_t user_cookie_offset;/* Used for user_action_cookie fixup. */
264 bool exit; /* No further actions should be processed. */
267 static void action_xlate_ctx_init(struct action_xlate_ctx *,
268 struct ofproto_dpif *, const struct flow *,
269 ovs_be16 initial_tci, struct rule_dpif *,
270 uint8_t tcp_flags, const struct ofpbuf *);
271 static struct ofpbuf *xlate_actions(struct action_xlate_ctx *,
272 const union ofp_action *in, size_t n_in);
274 /* An exact-match instantiation of an OpenFlow flow.
276 * A facet associates a "struct flow", which represents the Open vSwitch
277 * userspace idea of an exact-match flow, with one or more subfacets. Each
278 * subfacet tracks the datapath's idea of the exact-match flow equivalent to
279 * the facet. When the kernel module (or other dpif implementation) and Open
280 * vSwitch userspace agree on the definition of a flow key, there is exactly
281 * one subfacet per facet. If the dpif implementation supports more-specific
282 * flow matching than userspace, however, a facet can have more than one
283 * subfacet, each of which corresponds to some distinction in flow that
284 * userspace simply doesn't understand.
286 * Flow expiration works in terms of subfacets, so a facet must have at least
287 * one subfacet or it will never expire, leaking memory. */
290 struct hmap_node hmap_node; /* In owning ofproto's 'facets' hmap. */
291 struct list list_node; /* In owning rule's 'facets' list. */
292 struct rule_dpif *rule; /* Owning rule. */
295 struct list subfacets;
296 long long int used; /* Time last used; time created if not used. */
303 * - Do include packets and bytes sent "by hand", e.g. with
306 * - Do include packets and bytes that were obtained from the datapath
307 * when a subfacet's statistics were reset (e.g. dpif_flow_put() with
308 * DPIF_FP_ZERO_STATS).
310 * - Do not include packets or bytes that can be obtained from the
311 * datapath for any existing subfacet.
313 uint64_t packet_count; /* Number of packets received. */
314 uint64_t byte_count; /* Number of bytes received. */
316 /* Resubmit statistics. */
317 uint64_t prev_packet_count; /* Number of packets from last stats push. */
318 uint64_t prev_byte_count; /* Number of bytes from last stats push. */
319 long long int prev_used; /* Used time from last stats push. */
322 uint64_t accounted_bytes; /* Bytes processed by facet_account(). */
323 struct netflow_flow nf_flow; /* Per-flow NetFlow tracking data. */
324 uint8_t tcp_flags; /* TCP flags seen for this 'rule'. */
326 /* Properties of datapath actions.
328 * Every subfacet has its own actions because actions can differ slightly
329 * between splintered and non-splintered subfacets due to the VLAN tag
330 * being initially different (present vs. absent). All of them have these
331 * properties in common so we just store one copy of them here. */
332 bool may_install; /* Reassess actions for every packet? */
333 bool has_learn; /* Actions include NXAST_LEARN? */
334 bool has_normal; /* Actions output to OFPP_NORMAL? */
335 bool has_fin_timeout; /* Actions include NXAST_FIN_TIMEOUT? */
336 tag_type tags; /* Tags that would require revalidation. */
337 mirror_mask_t mirrors; /* Bitmap of dependent mirrors. */
340 static struct facet *facet_create(struct rule_dpif *, const struct flow *);
341 static void facet_remove(struct facet *);
342 static void facet_free(struct facet *);
344 static struct facet *facet_find(struct ofproto_dpif *, const struct flow *);
345 static struct facet *facet_lookup_valid(struct ofproto_dpif *,
346 const struct flow *);
347 static bool facet_revalidate(struct facet *);
348 static bool facet_check_consistency(struct facet *);
350 static void facet_flush_stats(struct facet *);
352 static void facet_update_time(struct facet *, long long int used);
353 static void facet_reset_counters(struct facet *);
354 static void facet_push_stats(struct facet *);
355 static void facet_account(struct facet *, bool may_flow_mod);
357 static bool facet_is_controller_flow(struct facet *);
359 /* A dpif flow and actions associated with a facet.
361 * See also the large comment on struct facet. */
364 struct hmap_node hmap_node; /* In struct ofproto_dpif 'subfacets' list. */
365 struct list list_node; /* In struct facet's 'facets' list. */
366 struct facet *facet; /* Owning facet. */
370 * To save memory in the common case, 'key' is NULL if 'key_fitness' is
371 * ODP_FIT_PERFECT, that is, odp_flow_key_from_flow() can accurately
372 * regenerate the ODP flow key from ->facet->flow. */
373 enum odp_key_fitness key_fitness;
377 long long int used; /* Time last used; time created if not used. */
379 uint64_t dp_packet_count; /* Last known packet count in the datapath. */
380 uint64_t dp_byte_count; /* Last known byte count in the datapath. */
384 * These should be essentially identical for every subfacet in a facet, but
385 * may differ in trivial ways due to VLAN splinters. */
386 size_t actions_len; /* Number of bytes in actions[]. */
387 struct nlattr *actions; /* Datapath actions. */
389 bool installed; /* Installed in datapath? */
391 /* This value is normally the same as ->facet->flow.vlan_tci. Only VLAN
392 * splinters can cause it to differ. This value should be removed when
393 * the VLAN splinters feature is no longer needed. */
394 ovs_be16 initial_tci; /* Initial VLAN TCI value. */
397 static struct subfacet *subfacet_create(struct facet *, enum odp_key_fitness,
398 const struct nlattr *key,
399 size_t key_len, ovs_be16 initial_tci);
400 static struct subfacet *subfacet_find(struct ofproto_dpif *,
401 const struct nlattr *key, size_t key_len);
402 static void subfacet_destroy(struct subfacet *);
403 static void subfacet_destroy__(struct subfacet *);
404 static void subfacet_get_key(struct subfacet *, struct odputil_keybuf *,
406 static void subfacet_reset_dp_stats(struct subfacet *,
407 struct dpif_flow_stats *);
408 static void subfacet_update_time(struct subfacet *, long long int used);
409 static void subfacet_update_stats(struct subfacet *,
410 const struct dpif_flow_stats *);
411 static void subfacet_make_actions(struct subfacet *,
412 const struct ofpbuf *packet);
413 static int subfacet_install(struct subfacet *,
414 const struct nlattr *actions, size_t actions_len,
415 struct dpif_flow_stats *);
416 static void subfacet_uninstall(struct subfacet *);
422 struct ofbundle *bundle; /* Bundle that contains this port, if any. */
423 struct list bundle_node; /* In struct ofbundle's "ports" list. */
424 struct cfm *cfm; /* Connectivity Fault Management, if any. */
425 tag_type tag; /* Tag associated with this port. */
426 uint32_t bond_stable_id; /* stable_id to use as bond slave, or 0. */
427 bool may_enable; /* May be enabled in bonds. */
428 long long int carrier_seq; /* Carrier status changes. */
431 struct stp_port *stp_port; /* Spanning Tree Protocol, if any. */
432 enum stp_state stp_state; /* Always STP_DISABLED if STP not in use. */
433 long long int stp_state_entered;
435 struct hmap priorities; /* Map of attached 'priority_to_dscp's. */
437 /* Linux VLAN device support (e.g. "eth0.10" for VLAN 10.)
439 * This is deprecated. It is only for compatibility with broken device
440 * drivers in old versions of Linux that do not properly support VLANs when
441 * VLAN devices are not used. When broken device drivers are no longer in
442 * widespread use, we will delete these interfaces. */
443 uint16_t realdev_ofp_port;
447 /* Node in 'ofport_dpif''s 'priorities' map. Used to maintain a map from
448 * 'priority' (the datapath's term for QoS queue) to the dscp bits which all
449 * traffic egressing the 'ofport' with that priority should be marked with. */
450 struct priority_to_dscp {
451 struct hmap_node hmap_node; /* Node in 'ofport_dpif''s 'priorities' map. */
452 uint32_t priority; /* Priority of this queue (see struct flow). */
454 uint8_t dscp; /* DSCP bits to mark outgoing traffic with. */
457 /* Linux VLAN device support (e.g. "eth0.10" for VLAN 10.)
459 * This is deprecated. It is only for compatibility with broken device drivers
460 * in old versions of Linux that do not properly support VLANs when VLAN
461 * devices are not used. When broken device drivers are no longer in
462 * widespread use, we will delete these interfaces. */
463 struct vlan_splinter {
464 struct hmap_node realdev_vid_node;
465 struct hmap_node vlandev_node;
466 uint16_t realdev_ofp_port;
467 uint16_t vlandev_ofp_port;
471 static uint32_t vsp_realdev_to_vlandev(const struct ofproto_dpif *,
472 uint32_t realdev, ovs_be16 vlan_tci);
473 static uint16_t vsp_vlandev_to_realdev(const struct ofproto_dpif *,
474 uint16_t vlandev, int *vid);
475 static void vsp_remove(struct ofport_dpif *);
476 static void vsp_add(struct ofport_dpif *, uint16_t realdev_ofp_port, int vid);
478 static struct ofport_dpif *
479 ofport_dpif_cast(const struct ofport *ofport)
481 assert(ofport->ofproto->ofproto_class == &ofproto_dpif_class);
482 return ofport ? CONTAINER_OF(ofport, struct ofport_dpif, up) : NULL;
485 static void port_run(struct ofport_dpif *);
486 static void port_wait(struct ofport_dpif *);
487 static int set_cfm(struct ofport *, const struct cfm_settings *);
488 static void ofport_clear_priorities(struct ofport_dpif *);
490 struct dpif_completion {
491 struct list list_node;
492 struct ofoperation *op;
495 /* Extra information about a classifier table.
496 * Currently used just for optimized flow revalidation. */
498 /* If either of these is nonnull, then this table has a form that allows
499 * flows to be tagged to avoid revalidating most flows for the most common
500 * kinds of flow table changes. */
501 struct cls_table *catchall_table; /* Table that wildcards all fields. */
502 struct cls_table *other_table; /* Table with any other wildcard set. */
503 uint32_t basis; /* Keeps each table's tags separate. */
506 struct ofproto_dpif {
507 struct hmap_node all_ofproto_dpifs_node; /* In 'all_ofproto_dpifs'. */
516 struct netflow *netflow;
517 struct dpif_sflow *sflow;
518 struct hmap bundles; /* Contains "struct ofbundle"s. */
519 struct mac_learning *ml;
520 struct ofmirror *mirrors[MAX_MIRRORS];
521 bool has_bonded_bundles;
524 struct timer next_expiration;
528 struct hmap subfacets;
531 struct table_dpif tables[N_TABLES];
532 bool need_revalidate;
533 struct tag_set revalidate_set;
535 /* Support for debugging async flow mods. */
536 struct list completions;
538 bool has_bundle_action; /* True when the first bundle action appears. */
539 struct netdev_stats stats; /* To account packets generated and consumed in
544 long long int stp_last_tick;
546 /* VLAN splinters. */
547 struct hmap realdev_vid_map; /* (realdev,vid) -> vlandev. */
548 struct hmap vlandev_map; /* vlandev -> (realdev,vid). */
551 /* Defer flow mod completion until "ovs-appctl ofproto/unclog"? (Useful only
552 * for debugging the asynchronous flow_mod implementation.) */
555 /* All existing ofproto_dpif instances, indexed by ->up.name. */
556 static struct hmap all_ofproto_dpifs = HMAP_INITIALIZER(&all_ofproto_dpifs);
558 static void ofproto_dpif_unixctl_init(void);
560 static struct ofproto_dpif *
561 ofproto_dpif_cast(const struct ofproto *ofproto)
563 assert(ofproto->ofproto_class == &ofproto_dpif_class);
564 return CONTAINER_OF(ofproto, struct ofproto_dpif, up);
567 static struct ofport_dpif *get_ofp_port(struct ofproto_dpif *,
569 static struct ofport_dpif *get_odp_port(struct ofproto_dpif *,
571 static void ofproto_trace(struct ofproto_dpif *, const struct flow *,
572 const struct ofpbuf *, ovs_be16 initial_tci,
575 /* Packet processing. */
576 static void update_learning_table(struct ofproto_dpif *,
577 const struct flow *, int vlan,
580 #define FLOW_MISS_MAX_BATCH 50
581 static int handle_upcalls(struct ofproto_dpif *, unsigned int max_batch);
583 /* Flow expiration. */
584 static int expire(struct ofproto_dpif *);
587 static void send_netflow_active_timeouts(struct ofproto_dpif *);
590 static int send_packet(const struct ofport_dpif *, struct ofpbuf *packet);
592 compose_sflow_action(const struct ofproto_dpif *, struct ofpbuf *odp_actions,
593 const struct flow *, uint32_t odp_port);
594 static void add_mirror_actions(struct action_xlate_ctx *ctx,
595 const struct flow *flow);
596 /* Global variables. */
597 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
599 /* Factory functions. */
602 enumerate_types(struct sset *types)
604 dp_enumerate_types(types);
608 enumerate_names(const char *type, struct sset *names)
610 return dp_enumerate_names(type, names);
614 del(const char *type, const char *name)
619 error = dpif_open(name, type, &dpif);
621 error = dpif_delete(dpif);
627 /* Basic life-cycle. */
629 static struct ofproto *
632 struct ofproto_dpif *ofproto = xmalloc(sizeof *ofproto);
637 dealloc(struct ofproto *ofproto_)
639 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
644 construct(struct ofproto *ofproto_)
646 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
647 const char *name = ofproto->up.name;
651 error = dpif_create_and_open(name, ofproto->up.type, &ofproto->dpif);
653 VLOG_ERR("failed to open datapath %s: %s", name, strerror(error));
657 ofproto->max_ports = dpif_get_max_ports(ofproto->dpif);
658 ofproto->n_matches = 0;
660 dpif_flow_flush(ofproto->dpif);
661 dpif_recv_purge(ofproto->dpif);
663 error = dpif_recv_set(ofproto->dpif, true);
665 VLOG_ERR("failed to listen on datapath %s: %s", name, strerror(error));
666 dpif_close(ofproto->dpif);
670 ofproto->netflow = NULL;
671 ofproto->sflow = NULL;
673 hmap_init(&ofproto->bundles);
674 ofproto->ml = mac_learning_create(MAC_ENTRY_DEFAULT_IDLE_TIME);
675 for (i = 0; i < MAX_MIRRORS; i++) {
676 ofproto->mirrors[i] = NULL;
678 ofproto->has_bonded_bundles = false;
680 timer_set_duration(&ofproto->next_expiration, 1000);
682 hmap_init(&ofproto->facets);
683 hmap_init(&ofproto->subfacets);
685 for (i = 0; i < N_TABLES; i++) {
686 struct table_dpif *table = &ofproto->tables[i];
688 table->catchall_table = NULL;
689 table->other_table = NULL;
690 table->basis = random_uint32();
692 ofproto->need_revalidate = false;
693 tag_set_init(&ofproto->revalidate_set);
695 list_init(&ofproto->completions);
697 ofproto_dpif_unixctl_init();
699 ofproto->has_bundle_action = false;
701 hmap_init(&ofproto->vlandev_map);
702 hmap_init(&ofproto->realdev_vid_map);
704 hmap_insert(&all_ofproto_dpifs, &ofproto->all_ofproto_dpifs_node,
705 hash_string(ofproto->up.name, 0));
706 memset(&ofproto->stats, 0, sizeof ofproto->stats);
708 ofproto_init_tables(ofproto_, N_TABLES);
714 complete_operations(struct ofproto_dpif *ofproto)
716 struct dpif_completion *c, *next;
718 LIST_FOR_EACH_SAFE (c, next, list_node, &ofproto->completions) {
719 ofoperation_complete(c->op, 0);
720 list_remove(&c->list_node);
726 destruct(struct ofproto *ofproto_)
728 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
729 struct rule_dpif *rule, *next_rule;
730 struct oftable *table;
733 hmap_remove(&all_ofproto_dpifs, &ofproto->all_ofproto_dpifs_node);
734 complete_operations(ofproto);
736 OFPROTO_FOR_EACH_TABLE (table, &ofproto->up) {
737 struct cls_cursor cursor;
739 cls_cursor_init(&cursor, &table->cls, NULL);
740 CLS_CURSOR_FOR_EACH_SAFE (rule, next_rule, up.cr, &cursor) {
741 ofproto_rule_destroy(&rule->up);
745 for (i = 0; i < MAX_MIRRORS; i++) {
746 mirror_destroy(ofproto->mirrors[i]);
749 netflow_destroy(ofproto->netflow);
750 dpif_sflow_destroy(ofproto->sflow);
751 hmap_destroy(&ofproto->bundles);
752 mac_learning_destroy(ofproto->ml);
754 hmap_destroy(&ofproto->facets);
755 hmap_destroy(&ofproto->subfacets);
757 hmap_destroy(&ofproto->vlandev_map);
758 hmap_destroy(&ofproto->realdev_vid_map);
760 dpif_close(ofproto->dpif);
764 run_fast(struct ofproto *ofproto_)
766 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
769 /* Handle one or more batches of upcalls, until there's nothing left to do
770 * or until we do a fixed total amount of work.
772 * We do work in batches because it can be much cheaper to set up a number
773 * of flows and fire off their patches all at once. We do multiple batches
774 * because in some cases handling a packet can cause another packet to be
775 * queued almost immediately as part of the return flow. Both
776 * optimizations can make major improvements on some benchmarks and
777 * presumably for real traffic as well. */
779 while (work < FLOW_MISS_MAX_BATCH) {
780 int retval = handle_upcalls(ofproto, FLOW_MISS_MAX_BATCH - work);
790 run(struct ofproto *ofproto_)
792 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
793 struct ofport_dpif *ofport;
794 struct ofbundle *bundle;
798 complete_operations(ofproto);
800 dpif_run(ofproto->dpif);
802 error = run_fast(ofproto_);
807 if (timer_expired(&ofproto->next_expiration)) {
808 int delay = expire(ofproto);
809 timer_set_duration(&ofproto->next_expiration, delay);
812 if (ofproto->netflow) {
813 if (netflow_run(ofproto->netflow)) {
814 send_netflow_active_timeouts(ofproto);
817 if (ofproto->sflow) {
818 dpif_sflow_run(ofproto->sflow);
821 HMAP_FOR_EACH (ofport, up.hmap_node, &ofproto->up.ports) {
824 HMAP_FOR_EACH (bundle, hmap_node, &ofproto->bundles) {
829 mac_learning_run(ofproto->ml, &ofproto->revalidate_set);
831 /* Now revalidate if there's anything to do. */
832 if (ofproto->need_revalidate
833 || !tag_set_is_empty(&ofproto->revalidate_set)) {
834 struct tag_set revalidate_set = ofproto->revalidate_set;
835 bool revalidate_all = ofproto->need_revalidate;
836 struct facet *facet, *next;
838 /* Clear the revalidation flags. */
839 tag_set_init(&ofproto->revalidate_set);
840 ofproto->need_revalidate = false;
842 HMAP_FOR_EACH_SAFE (facet, next, hmap_node, &ofproto->facets) {
844 || tag_set_intersects(&revalidate_set, facet->tags)) {
845 facet_revalidate(facet);
850 /* Check the consistency of a random facet, to aid debugging. */
851 if (!hmap_is_empty(&ofproto->facets) && !ofproto->need_revalidate) {
854 facet = CONTAINER_OF(hmap_random_node(&ofproto->facets),
855 struct facet, hmap_node);
856 if (!tag_set_intersects(&ofproto->revalidate_set, facet->tags)) {
857 if (!facet_check_consistency(facet)) {
858 ofproto->need_revalidate = true;
867 wait(struct ofproto *ofproto_)
869 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
870 struct ofport_dpif *ofport;
871 struct ofbundle *bundle;
873 if (!clogged && !list_is_empty(&ofproto->completions)) {
874 poll_immediate_wake();
877 dpif_wait(ofproto->dpif);
878 dpif_recv_wait(ofproto->dpif);
879 if (ofproto->sflow) {
880 dpif_sflow_wait(ofproto->sflow);
882 if (!tag_set_is_empty(&ofproto->revalidate_set)) {
883 poll_immediate_wake();
885 HMAP_FOR_EACH (ofport, up.hmap_node, &ofproto->up.ports) {
888 HMAP_FOR_EACH (bundle, hmap_node, &ofproto->bundles) {
891 if (ofproto->netflow) {
892 netflow_wait(ofproto->netflow);
894 mac_learning_wait(ofproto->ml);
896 if (ofproto->need_revalidate) {
897 /* Shouldn't happen, but if it does just go around again. */
898 VLOG_DBG_RL(&rl, "need revalidate in ofproto_wait_cb()");
899 poll_immediate_wake();
901 timer_wait(&ofproto->next_expiration);
906 flush(struct ofproto *ofproto_)
908 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
909 struct facet *facet, *next_facet;
911 HMAP_FOR_EACH_SAFE (facet, next_facet, hmap_node, &ofproto->facets) {
912 /* Mark the facet as not installed so that facet_remove() doesn't
913 * bother trying to uninstall it. There is no point in uninstalling it
914 * individually since we are about to blow away all the facets with
915 * dpif_flow_flush(). */
916 struct subfacet *subfacet;
918 LIST_FOR_EACH (subfacet, list_node, &facet->subfacets) {
919 subfacet->installed = false;
920 subfacet->dp_packet_count = 0;
921 subfacet->dp_byte_count = 0;
925 dpif_flow_flush(ofproto->dpif);
929 get_features(struct ofproto *ofproto_ OVS_UNUSED,
930 bool *arp_match_ip, enum ofputil_action_bitmap *actions)
932 *arp_match_ip = true;
933 *actions = (OFPUTIL_A_OUTPUT |
934 OFPUTIL_A_SET_VLAN_VID |
935 OFPUTIL_A_SET_VLAN_PCP |
936 OFPUTIL_A_STRIP_VLAN |
937 OFPUTIL_A_SET_DL_SRC |
938 OFPUTIL_A_SET_DL_DST |
939 OFPUTIL_A_SET_NW_SRC |
940 OFPUTIL_A_SET_NW_DST |
941 OFPUTIL_A_SET_NW_TOS |
942 OFPUTIL_A_SET_TP_SRC |
943 OFPUTIL_A_SET_TP_DST |
948 get_tables(struct ofproto *ofproto_, struct ofp_table_stats *ots)
950 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
951 struct dpif_dp_stats s;
953 strcpy(ots->name, "classifier");
955 dpif_get_dp_stats(ofproto->dpif, &s);
956 put_32aligned_be64(&ots->lookup_count, htonll(s.n_hit + s.n_missed));
957 put_32aligned_be64(&ots->matched_count,
958 htonll(s.n_hit + ofproto->n_matches));
961 static struct ofport *
964 struct ofport_dpif *port = xmalloc(sizeof *port);
969 port_dealloc(struct ofport *port_)
971 struct ofport_dpif *port = ofport_dpif_cast(port_);
976 port_construct(struct ofport *port_)
978 struct ofport_dpif *port = ofport_dpif_cast(port_);
979 struct ofproto_dpif *ofproto = ofproto_dpif_cast(port->up.ofproto);
981 ofproto->need_revalidate = true;
982 port->odp_port = ofp_port_to_odp_port(port->up.ofp_port);
985 port->tag = tag_create_random();
986 port->may_enable = true;
987 port->stp_port = NULL;
988 port->stp_state = STP_DISABLED;
989 hmap_init(&port->priorities);
990 port->realdev_ofp_port = 0;
991 port->vlandev_vid = 0;
992 port->carrier_seq = netdev_get_carrier_resets(port->up.netdev);
994 if (ofproto->sflow) {
995 dpif_sflow_add_port(ofproto->sflow, port_);
1002 port_destruct(struct ofport *port_)
1004 struct ofport_dpif *port = ofport_dpif_cast(port_);
1005 struct ofproto_dpif *ofproto = ofproto_dpif_cast(port->up.ofproto);
1007 ofproto->need_revalidate = true;
1008 bundle_remove(port_);
1009 set_cfm(port_, NULL);
1010 if (ofproto->sflow) {
1011 dpif_sflow_del_port(ofproto->sflow, port->odp_port);
1014 ofport_clear_priorities(port);
1015 hmap_destroy(&port->priorities);
1019 port_modified(struct ofport *port_)
1021 struct ofport_dpif *port = ofport_dpif_cast(port_);
1023 if (port->bundle && port->bundle->bond) {
1024 bond_slave_set_netdev(port->bundle->bond, port, port->up.netdev);
1029 port_reconfigured(struct ofport *port_, enum ofputil_port_config old_config)
1031 struct ofport_dpif *port = ofport_dpif_cast(port_);
1032 struct ofproto_dpif *ofproto = ofproto_dpif_cast(port->up.ofproto);
1033 enum ofputil_port_config changed = old_config ^ port->up.pp.config;
1035 if (changed & (OFPUTIL_PC_NO_RECV | OFPUTIL_PC_NO_RECV_STP |
1036 OFPUTIL_PC_NO_FWD | OFPUTIL_PC_NO_FLOOD)) {
1037 ofproto->need_revalidate = true;
1039 if (changed & OFPUTIL_PC_NO_FLOOD && port->bundle) {
1040 bundle_update(port->bundle);
1046 set_sflow(struct ofproto *ofproto_,
1047 const struct ofproto_sflow_options *sflow_options)
1049 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1050 struct dpif_sflow *ds = ofproto->sflow;
1052 if (sflow_options) {
1054 struct ofport_dpif *ofport;
1056 ds = ofproto->sflow = dpif_sflow_create(ofproto->dpif);
1057 HMAP_FOR_EACH (ofport, up.hmap_node, &ofproto->up.ports) {
1058 dpif_sflow_add_port(ds, &ofport->up);
1060 ofproto->need_revalidate = true;
1062 dpif_sflow_set_options(ds, sflow_options);
1065 dpif_sflow_destroy(ds);
1066 ofproto->need_revalidate = true;
1067 ofproto->sflow = NULL;
1074 set_cfm(struct ofport *ofport_, const struct cfm_settings *s)
1076 struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
1083 struct ofproto_dpif *ofproto;
1085 ofproto = ofproto_dpif_cast(ofport->up.ofproto);
1086 ofproto->need_revalidate = true;
1087 ofport->cfm = cfm_create(netdev_get_name(ofport->up.netdev));
1090 if (cfm_configure(ofport->cfm, s)) {
1096 cfm_destroy(ofport->cfm);
1102 get_cfm_fault(const struct ofport *ofport_)
1104 struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
1106 return ofport->cfm ? cfm_get_fault(ofport->cfm) : -1;
1110 get_cfm_remote_mpids(const struct ofport *ofport_, const uint64_t **rmps,
1113 struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
1116 cfm_get_remote_mpids(ofport->cfm, rmps, n_rmps);
1124 get_cfm_health(const struct ofport *ofport_)
1126 struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
1128 return ofport->cfm ? cfm_get_health(ofport->cfm) : -1;
1131 /* Spanning Tree. */
1134 send_bpdu_cb(struct ofpbuf *pkt, int port_num, void *ofproto_)
1136 struct ofproto_dpif *ofproto = ofproto_;
1137 struct stp_port *sp = stp_get_port(ofproto->stp, port_num);
1138 struct ofport_dpif *ofport;
1140 ofport = stp_port_get_aux(sp);
1142 VLOG_WARN_RL(&rl, "%s: cannot send BPDU on unknown port %d",
1143 ofproto->up.name, port_num);
1145 struct eth_header *eth = pkt->l2;
1147 netdev_get_etheraddr(ofport->up.netdev, eth->eth_src);
1148 if (eth_addr_is_zero(eth->eth_src)) {
1149 VLOG_WARN_RL(&rl, "%s: cannot send BPDU on port %d "
1150 "with unknown MAC", ofproto->up.name, port_num);
1152 send_packet(ofport, pkt);
1158 /* Configures STP on 'ofproto_' using the settings defined in 's'. */
1160 set_stp(struct ofproto *ofproto_, const struct ofproto_stp_settings *s)
1162 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1164 /* Only revalidate flows if the configuration changed. */
1165 if (!s != !ofproto->stp) {
1166 ofproto->need_revalidate = true;
1170 if (!ofproto->stp) {
1171 ofproto->stp = stp_create(ofproto_->name, s->system_id,
1172 send_bpdu_cb, ofproto);
1173 ofproto->stp_last_tick = time_msec();
1176 stp_set_bridge_id(ofproto->stp, s->system_id);
1177 stp_set_bridge_priority(ofproto->stp, s->priority);
1178 stp_set_hello_time(ofproto->stp, s->hello_time);
1179 stp_set_max_age(ofproto->stp, s->max_age);
1180 stp_set_forward_delay(ofproto->stp, s->fwd_delay);
1182 struct ofport *ofport;
1184 HMAP_FOR_EACH (ofport, hmap_node, &ofproto->up.ports) {
1185 set_stp_port(ofport, NULL);
1188 stp_destroy(ofproto->stp);
1189 ofproto->stp = NULL;
1196 get_stp_status(struct ofproto *ofproto_, struct ofproto_stp_status *s)
1198 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1202 s->bridge_id = stp_get_bridge_id(ofproto->stp);
1203 s->designated_root = stp_get_designated_root(ofproto->stp);
1204 s->root_path_cost = stp_get_root_path_cost(ofproto->stp);
1213 update_stp_port_state(struct ofport_dpif *ofport)
1215 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport->up.ofproto);
1216 enum stp_state state;
1218 /* Figure out new state. */
1219 state = ofport->stp_port ? stp_port_get_state(ofport->stp_port)
1223 if (ofport->stp_state != state) {
1224 enum ofputil_port_state of_state;
1227 VLOG_DBG_RL(&rl, "port %s: STP state changed from %s to %s",
1228 netdev_get_name(ofport->up.netdev),
1229 stp_state_name(ofport->stp_state),
1230 stp_state_name(state));
1231 if (stp_learn_in_state(ofport->stp_state)
1232 != stp_learn_in_state(state)) {
1233 /* xxx Learning action flows should also be flushed. */
1234 mac_learning_flush(ofproto->ml, &ofproto->revalidate_set);
1236 fwd_change = stp_forward_in_state(ofport->stp_state)
1237 != stp_forward_in_state(state);
1239 ofproto->need_revalidate = true;
1240 ofport->stp_state = state;
1241 ofport->stp_state_entered = time_msec();
1243 if (fwd_change && ofport->bundle) {
1244 bundle_update(ofport->bundle);
1247 /* Update the STP state bits in the OpenFlow port description. */
1248 of_state = ofport->up.pp.state & ~OFPUTIL_PS_STP_MASK;
1249 of_state |= (state == STP_LISTENING ? OFPUTIL_PS_STP_LISTEN
1250 : state == STP_LEARNING ? OFPUTIL_PS_STP_LEARN
1251 : state == STP_FORWARDING ? OFPUTIL_PS_STP_FORWARD
1252 : state == STP_BLOCKING ? OFPUTIL_PS_STP_BLOCK
1254 ofproto_port_set_state(&ofport->up, of_state);
1258 /* Configures STP on 'ofport_' using the settings defined in 's'. The
1259 * caller is responsible for assigning STP port numbers and ensuring
1260 * there are no duplicates. */
1262 set_stp_port(struct ofport *ofport_,
1263 const struct ofproto_port_stp_settings *s)
1265 struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
1266 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport->up.ofproto);
1267 struct stp_port *sp = ofport->stp_port;
1269 if (!s || !s->enable) {
1271 ofport->stp_port = NULL;
1272 stp_port_disable(sp);
1273 update_stp_port_state(ofport);
1276 } else if (sp && stp_port_no(sp) != s->port_num
1277 && ofport == stp_port_get_aux(sp)) {
1278 /* The port-id changed, so disable the old one if it's not
1279 * already in use by another port. */
1280 stp_port_disable(sp);
1283 sp = ofport->stp_port = stp_get_port(ofproto->stp, s->port_num);
1284 stp_port_enable(sp);
1286 stp_port_set_aux(sp, ofport);
1287 stp_port_set_priority(sp, s->priority);
1288 stp_port_set_path_cost(sp, s->path_cost);
1290 update_stp_port_state(ofport);
1296 get_stp_port_status(struct ofport *ofport_,
1297 struct ofproto_port_stp_status *s)
1299 struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
1300 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport->up.ofproto);
1301 struct stp_port *sp = ofport->stp_port;
1303 if (!ofproto->stp || !sp) {
1309 s->port_id = stp_port_get_id(sp);
1310 s->state = stp_port_get_state(sp);
1311 s->sec_in_state = (time_msec() - ofport->stp_state_entered) / 1000;
1312 s->role = stp_port_get_role(sp);
1313 stp_port_get_counts(sp, &s->tx_count, &s->rx_count, &s->error_count);
1319 stp_run(struct ofproto_dpif *ofproto)
1322 long long int now = time_msec();
1323 long long int elapsed = now - ofproto->stp_last_tick;
1324 struct stp_port *sp;
1327 stp_tick(ofproto->stp, MIN(INT_MAX, elapsed));
1328 ofproto->stp_last_tick = now;
1330 while (stp_get_changed_port(ofproto->stp, &sp)) {
1331 struct ofport_dpif *ofport = stp_port_get_aux(sp);
1334 update_stp_port_state(ofport);
1338 if (stp_check_and_reset_fdb_flush(ofproto->stp)) {
1339 mac_learning_flush(ofproto->ml, &ofproto->revalidate_set);
1345 stp_wait(struct ofproto_dpif *ofproto)
1348 poll_timer_wait(1000);
1352 /* Returns true if STP should process 'flow'. */
1354 stp_should_process_flow(const struct flow *flow)
1356 return eth_addr_equals(flow->dl_dst, eth_addr_stp);
1360 stp_process_packet(const struct ofport_dpif *ofport,
1361 const struct ofpbuf *packet)
1363 struct ofpbuf payload = *packet;
1364 struct eth_header *eth = payload.data;
1365 struct stp_port *sp = ofport->stp_port;
1367 /* Sink packets on ports that have STP disabled when the bridge has
1369 if (!sp || stp_port_get_state(sp) == STP_DISABLED) {
1373 /* Trim off padding on payload. */
1374 if (payload.size > ntohs(eth->eth_type) + ETH_HEADER_LEN) {
1375 payload.size = ntohs(eth->eth_type) + ETH_HEADER_LEN;
1378 if (ofpbuf_try_pull(&payload, ETH_HEADER_LEN + LLC_HEADER_LEN)) {
1379 stp_received_bpdu(sp, payload.data, payload.size);
1383 static struct priority_to_dscp *
1384 get_priority(const struct ofport_dpif *ofport, uint32_t priority)
1386 struct priority_to_dscp *pdscp;
1389 hash = hash_int(priority, 0);
1390 HMAP_FOR_EACH_IN_BUCKET (pdscp, hmap_node, hash, &ofport->priorities) {
1391 if (pdscp->priority == priority) {
1399 ofport_clear_priorities(struct ofport_dpif *ofport)
1401 struct priority_to_dscp *pdscp, *next;
1403 HMAP_FOR_EACH_SAFE (pdscp, next, hmap_node, &ofport->priorities) {
1404 hmap_remove(&ofport->priorities, &pdscp->hmap_node);
1410 set_queues(struct ofport *ofport_,
1411 const struct ofproto_port_queue *qdscp_list,
1414 struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
1415 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport->up.ofproto);
1416 struct hmap new = HMAP_INITIALIZER(&new);
1419 for (i = 0; i < n_qdscp; i++) {
1420 struct priority_to_dscp *pdscp;
1424 dscp = (qdscp_list[i].dscp << 2) & IP_DSCP_MASK;
1425 if (dpif_queue_to_priority(ofproto->dpif, qdscp_list[i].queue,
1430 pdscp = get_priority(ofport, priority);
1432 hmap_remove(&ofport->priorities, &pdscp->hmap_node);
1434 pdscp = xmalloc(sizeof *pdscp);
1435 pdscp->priority = priority;
1437 ofproto->need_revalidate = true;
1440 if (pdscp->dscp != dscp) {
1442 ofproto->need_revalidate = true;
1445 hmap_insert(&new, &pdscp->hmap_node, hash_int(pdscp->priority, 0));
1448 if (!hmap_is_empty(&ofport->priorities)) {
1449 ofport_clear_priorities(ofport);
1450 ofproto->need_revalidate = true;
1453 hmap_swap(&new, &ofport->priorities);
1461 /* Expires all MAC learning entries associated with 'bundle' and forces its
1462 * ofproto to revalidate every flow.
1464 * Normally MAC learning entries are removed only from the ofproto associated
1465 * with 'bundle', but if 'all_ofprotos' is true, then the MAC learning entries
1466 * are removed from every ofproto. When patch ports and SLB bonds are in use
1467 * and a VM migration happens and the gratuitous ARPs are somehow lost, this
1468 * avoids a MAC_ENTRY_IDLE_TIME delay before the migrated VM can communicate
1469 * with the host from which it migrated. */
1471 bundle_flush_macs(struct ofbundle *bundle, bool all_ofprotos)
1473 struct ofproto_dpif *ofproto = bundle->ofproto;
1474 struct mac_learning *ml = ofproto->ml;
1475 struct mac_entry *mac, *next_mac;
1477 ofproto->need_revalidate = true;
1478 LIST_FOR_EACH_SAFE (mac, next_mac, lru_node, &ml->lrus) {
1479 if (mac->port.p == bundle) {
1481 struct ofproto_dpif *o;
1483 HMAP_FOR_EACH (o, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
1485 struct mac_entry *e;
1487 e = mac_learning_lookup(o->ml, mac->mac, mac->vlan,
1490 tag_set_add(&o->revalidate_set, e->tag);
1491 mac_learning_expire(o->ml, e);
1497 mac_learning_expire(ml, mac);
1502 static struct ofbundle *
1503 bundle_lookup(const struct ofproto_dpif *ofproto, void *aux)
1505 struct ofbundle *bundle;
1507 HMAP_FOR_EACH_IN_BUCKET (bundle, hmap_node, hash_pointer(aux, 0),
1508 &ofproto->bundles) {
1509 if (bundle->aux == aux) {
1516 /* Looks up each of the 'n_auxes' pointers in 'auxes' as bundles and adds the
1517 * ones that are found to 'bundles'. */
1519 bundle_lookup_multiple(struct ofproto_dpif *ofproto,
1520 void **auxes, size_t n_auxes,
1521 struct hmapx *bundles)
1525 hmapx_init(bundles);
1526 for (i = 0; i < n_auxes; i++) {
1527 struct ofbundle *bundle = bundle_lookup(ofproto, auxes[i]);
1529 hmapx_add(bundles, bundle);
1535 bundle_update(struct ofbundle *bundle)
1537 struct ofport_dpif *port;
1539 bundle->floodable = true;
1540 LIST_FOR_EACH (port, bundle_node, &bundle->ports) {
1541 if (port->up.pp.config & OFPUTIL_PC_NO_FLOOD
1542 || !stp_forward_in_state(port->stp_state)) {
1543 bundle->floodable = false;
1550 bundle_del_port(struct ofport_dpif *port)
1552 struct ofbundle *bundle = port->bundle;
1554 bundle->ofproto->need_revalidate = true;
1556 list_remove(&port->bundle_node);
1557 port->bundle = NULL;
1560 lacp_slave_unregister(bundle->lacp, port);
1563 bond_slave_unregister(bundle->bond, port);
1566 bundle_update(bundle);
1570 bundle_add_port(struct ofbundle *bundle, uint32_t ofp_port,
1571 struct lacp_slave_settings *lacp,
1572 uint32_t bond_stable_id)
1574 struct ofport_dpif *port;
1576 port = get_ofp_port(bundle->ofproto, ofp_port);
1581 if (port->bundle != bundle) {
1582 bundle->ofproto->need_revalidate = true;
1584 bundle_del_port(port);
1587 port->bundle = bundle;
1588 list_push_back(&bundle->ports, &port->bundle_node);
1589 if (port->up.pp.config & OFPUTIL_PC_NO_FLOOD
1590 || !stp_forward_in_state(port->stp_state)) {
1591 bundle->floodable = false;
1595 port->bundle->ofproto->need_revalidate = true;
1596 lacp_slave_register(bundle->lacp, port, lacp);
1599 port->bond_stable_id = bond_stable_id;
1605 bundle_destroy(struct ofbundle *bundle)
1607 struct ofproto_dpif *ofproto;
1608 struct ofport_dpif *port, *next_port;
1615 ofproto = bundle->ofproto;
1616 for (i = 0; i < MAX_MIRRORS; i++) {
1617 struct ofmirror *m = ofproto->mirrors[i];
1619 if (m->out == bundle) {
1621 } else if (hmapx_find_and_delete(&m->srcs, bundle)
1622 || hmapx_find_and_delete(&m->dsts, bundle)) {
1623 ofproto->need_revalidate = true;
1628 LIST_FOR_EACH_SAFE (port, next_port, bundle_node, &bundle->ports) {
1629 bundle_del_port(port);
1632 bundle_flush_macs(bundle, true);
1633 hmap_remove(&ofproto->bundles, &bundle->hmap_node);
1635 free(bundle->trunks);
1636 lacp_destroy(bundle->lacp);
1637 bond_destroy(bundle->bond);
1642 bundle_set(struct ofproto *ofproto_, void *aux,
1643 const struct ofproto_bundle_settings *s)
1645 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1646 bool need_flush = false;
1647 struct ofport_dpif *port;
1648 struct ofbundle *bundle;
1649 unsigned long *trunks;
1655 bundle_destroy(bundle_lookup(ofproto, aux));
1659 assert(s->n_slaves == 1 || s->bond != NULL);
1660 assert((s->lacp != NULL) == (s->lacp_slaves != NULL));
1662 bundle = bundle_lookup(ofproto, aux);
1664 bundle = xmalloc(sizeof *bundle);
1666 bundle->ofproto = ofproto;
1667 hmap_insert(&ofproto->bundles, &bundle->hmap_node,
1668 hash_pointer(aux, 0));
1670 bundle->name = NULL;
1672 list_init(&bundle->ports);
1673 bundle->vlan_mode = PORT_VLAN_TRUNK;
1675 bundle->trunks = NULL;
1676 bundle->use_priority_tags = s->use_priority_tags;
1677 bundle->lacp = NULL;
1678 bundle->bond = NULL;
1680 bundle->floodable = true;
1682 bundle->src_mirrors = 0;
1683 bundle->dst_mirrors = 0;
1684 bundle->mirror_out = 0;
1687 if (!bundle->name || strcmp(s->name, bundle->name)) {
1689 bundle->name = xstrdup(s->name);
1694 if (!bundle->lacp) {
1695 ofproto->need_revalidate = true;
1696 bundle->lacp = lacp_create();
1698 lacp_configure(bundle->lacp, s->lacp);
1700 lacp_destroy(bundle->lacp);
1701 bundle->lacp = NULL;
1704 /* Update set of ports. */
1706 for (i = 0; i < s->n_slaves; i++) {
1707 if (!bundle_add_port(bundle, s->slaves[i],
1708 s->lacp ? &s->lacp_slaves[i] : NULL,
1709 s->bond_stable_ids ? s->bond_stable_ids[i] : 0)) {
1713 if (!ok || list_size(&bundle->ports) != s->n_slaves) {
1714 struct ofport_dpif *next_port;
1716 LIST_FOR_EACH_SAFE (port, next_port, bundle_node, &bundle->ports) {
1717 for (i = 0; i < s->n_slaves; i++) {
1718 if (s->slaves[i] == port->up.ofp_port) {
1723 bundle_del_port(port);
1727 assert(list_size(&bundle->ports) <= s->n_slaves);
1729 if (list_is_empty(&bundle->ports)) {
1730 bundle_destroy(bundle);
1734 /* Set VLAN tagging mode */
1735 if (s->vlan_mode != bundle->vlan_mode
1736 || s->use_priority_tags != bundle->use_priority_tags) {
1737 bundle->vlan_mode = s->vlan_mode;
1738 bundle->use_priority_tags = s->use_priority_tags;
1743 vlan = (s->vlan_mode == PORT_VLAN_TRUNK ? -1
1744 : s->vlan >= 0 && s->vlan <= 4095 ? s->vlan
1746 if (vlan != bundle->vlan) {
1747 bundle->vlan = vlan;
1751 /* Get trunked VLANs. */
1752 switch (s->vlan_mode) {
1753 case PORT_VLAN_ACCESS:
1757 case PORT_VLAN_TRUNK:
1758 trunks = (unsigned long *) s->trunks;
1761 case PORT_VLAN_NATIVE_UNTAGGED:
1762 case PORT_VLAN_NATIVE_TAGGED:
1763 if (vlan != 0 && (!s->trunks
1764 || !bitmap_is_set(s->trunks, vlan)
1765 || bitmap_is_set(s->trunks, 0))) {
1766 /* Force trunking the native VLAN and prohibit trunking VLAN 0. */
1768 trunks = bitmap_clone(s->trunks, 4096);
1770 trunks = bitmap_allocate1(4096);
1772 bitmap_set1(trunks, vlan);
1773 bitmap_set0(trunks, 0);
1775 trunks = (unsigned long *) s->trunks;
1782 if (!vlan_bitmap_equal(trunks, bundle->trunks)) {
1783 free(bundle->trunks);
1784 if (trunks == s->trunks) {
1785 bundle->trunks = vlan_bitmap_clone(trunks);
1787 bundle->trunks = trunks;
1792 if (trunks != s->trunks) {
1797 if (!list_is_short(&bundle->ports)) {
1798 bundle->ofproto->has_bonded_bundles = true;
1800 if (bond_reconfigure(bundle->bond, s->bond)) {
1801 ofproto->need_revalidate = true;
1804 bundle->bond = bond_create(s->bond);
1805 ofproto->need_revalidate = true;
1808 LIST_FOR_EACH (port, bundle_node, &bundle->ports) {
1809 bond_slave_register(bundle->bond, port, port->bond_stable_id,
1813 bond_destroy(bundle->bond);
1814 bundle->bond = NULL;
1817 /* If we changed something that would affect MAC learning, un-learn
1818 * everything on this port and force flow revalidation. */
1820 bundle_flush_macs(bundle, false);
1827 bundle_remove(struct ofport *port_)
1829 struct ofport_dpif *port = ofport_dpif_cast(port_);
1830 struct ofbundle *bundle = port->bundle;
1833 bundle_del_port(port);
1834 if (list_is_empty(&bundle->ports)) {
1835 bundle_destroy(bundle);
1836 } else if (list_is_short(&bundle->ports)) {
1837 bond_destroy(bundle->bond);
1838 bundle->bond = NULL;
1844 send_pdu_cb(void *port_, const void *pdu, size_t pdu_size)
1846 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 10);
1847 struct ofport_dpif *port = port_;
1848 uint8_t ea[ETH_ADDR_LEN];
1851 error = netdev_get_etheraddr(port->up.netdev, ea);
1853 struct ofpbuf packet;
1856 ofpbuf_init(&packet, 0);
1857 packet_pdu = eth_compose(&packet, eth_addr_lacp, ea, ETH_TYPE_LACP,
1859 memcpy(packet_pdu, pdu, pdu_size);
1861 send_packet(port, &packet);
1862 ofpbuf_uninit(&packet);
1864 VLOG_ERR_RL(&rl, "port %s: cannot obtain Ethernet address of iface "
1865 "%s (%s)", port->bundle->name,
1866 netdev_get_name(port->up.netdev), strerror(error));
1871 bundle_send_learning_packets(struct ofbundle *bundle)
1873 struct ofproto_dpif *ofproto = bundle->ofproto;
1874 int error, n_packets, n_errors;
1875 struct mac_entry *e;
1877 error = n_packets = n_errors = 0;
1878 LIST_FOR_EACH (e, lru_node, &ofproto->ml->lrus) {
1879 if (e->port.p != bundle) {
1880 struct ofpbuf *learning_packet;
1881 struct ofport_dpif *port;
1885 /* The assignment to "port" is unnecessary but makes "grep"ing for
1886 * struct ofport_dpif more effective. */
1887 learning_packet = bond_compose_learning_packet(bundle->bond,
1891 ret = send_packet(port, learning_packet);
1892 ofpbuf_delete(learning_packet);
1902 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
1903 VLOG_WARN_RL(&rl, "bond %s: %d errors sending %d gratuitous learning "
1904 "packets, last error was: %s",
1905 bundle->name, n_errors, n_packets, strerror(error));
1907 VLOG_DBG("bond %s: sent %d gratuitous learning packets",
1908 bundle->name, n_packets);
1913 bundle_run(struct ofbundle *bundle)
1916 lacp_run(bundle->lacp, send_pdu_cb);
1919 struct ofport_dpif *port;
1921 LIST_FOR_EACH (port, bundle_node, &bundle->ports) {
1922 bond_slave_set_may_enable(bundle->bond, port, port->may_enable);
1925 bond_run(bundle->bond, &bundle->ofproto->revalidate_set,
1926 lacp_status(bundle->lacp));
1927 if (bond_should_send_learning_packets(bundle->bond)) {
1928 bundle_send_learning_packets(bundle);
1934 bundle_wait(struct ofbundle *bundle)
1937 lacp_wait(bundle->lacp);
1940 bond_wait(bundle->bond);
1947 mirror_scan(struct ofproto_dpif *ofproto)
1951 for (idx = 0; idx < MAX_MIRRORS; idx++) {
1952 if (!ofproto->mirrors[idx]) {
1959 static struct ofmirror *
1960 mirror_lookup(struct ofproto_dpif *ofproto, void *aux)
1964 for (i = 0; i < MAX_MIRRORS; i++) {
1965 struct ofmirror *mirror = ofproto->mirrors[i];
1966 if (mirror && mirror->aux == aux) {
1974 /* Update the 'dup_mirrors' member of each of the ofmirrors in 'ofproto'. */
1976 mirror_update_dups(struct ofproto_dpif *ofproto)
1980 for (i = 0; i < MAX_MIRRORS; i++) {
1981 struct ofmirror *m = ofproto->mirrors[i];
1984 m->dup_mirrors = MIRROR_MASK_C(1) << i;
1988 for (i = 0; i < MAX_MIRRORS; i++) {
1989 struct ofmirror *m1 = ofproto->mirrors[i];
1996 for (j = i + 1; j < MAX_MIRRORS; j++) {
1997 struct ofmirror *m2 = ofproto->mirrors[j];
1999 if (m2 && m1->out == m2->out && m1->out_vlan == m2->out_vlan) {
2000 m1->dup_mirrors |= MIRROR_MASK_C(1) << j;
2001 m2->dup_mirrors |= m1->dup_mirrors;
2008 mirror_set(struct ofproto *ofproto_, void *aux,
2009 const struct ofproto_mirror_settings *s)
2011 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
2012 mirror_mask_t mirror_bit;
2013 struct ofbundle *bundle;
2014 struct ofmirror *mirror;
2015 struct ofbundle *out;
2016 struct hmapx srcs; /* Contains "struct ofbundle *"s. */
2017 struct hmapx dsts; /* Contains "struct ofbundle *"s. */
2020 mirror = mirror_lookup(ofproto, aux);
2022 mirror_destroy(mirror);
2028 idx = mirror_scan(ofproto);
2030 VLOG_WARN("bridge %s: maximum of %d port mirrors reached, "
2032 ofproto->up.name, MAX_MIRRORS, s->name);
2036 mirror = ofproto->mirrors[idx] = xzalloc(sizeof *mirror);
2037 mirror->ofproto = ofproto;
2040 mirror->out_vlan = -1;
2041 mirror->name = NULL;
2044 if (!mirror->name || strcmp(s->name, mirror->name)) {
2046 mirror->name = xstrdup(s->name);
2049 /* Get the new configuration. */
2050 if (s->out_bundle) {
2051 out = bundle_lookup(ofproto, s->out_bundle);
2053 mirror_destroy(mirror);
2059 out_vlan = s->out_vlan;
2061 bundle_lookup_multiple(ofproto, s->srcs, s->n_srcs, &srcs);
2062 bundle_lookup_multiple(ofproto, s->dsts, s->n_dsts, &dsts);
2064 /* If the configuration has not changed, do nothing. */
2065 if (hmapx_equals(&srcs, &mirror->srcs)
2066 && hmapx_equals(&dsts, &mirror->dsts)
2067 && vlan_bitmap_equal(mirror->vlans, s->src_vlans)
2068 && mirror->out == out
2069 && mirror->out_vlan == out_vlan)
2071 hmapx_destroy(&srcs);
2072 hmapx_destroy(&dsts);
2076 hmapx_swap(&srcs, &mirror->srcs);
2077 hmapx_destroy(&srcs);
2079 hmapx_swap(&dsts, &mirror->dsts);
2080 hmapx_destroy(&dsts);
2082 free(mirror->vlans);
2083 mirror->vlans = vlan_bitmap_clone(s->src_vlans);
2086 mirror->out_vlan = out_vlan;
2088 /* Update bundles. */
2089 mirror_bit = MIRROR_MASK_C(1) << mirror->idx;
2090 HMAP_FOR_EACH (bundle, hmap_node, &mirror->ofproto->bundles) {
2091 if (hmapx_contains(&mirror->srcs, bundle)) {
2092 bundle->src_mirrors |= mirror_bit;
2094 bundle->src_mirrors &= ~mirror_bit;
2097 if (hmapx_contains(&mirror->dsts, bundle)) {
2098 bundle->dst_mirrors |= mirror_bit;
2100 bundle->dst_mirrors &= ~mirror_bit;
2103 if (mirror->out == bundle) {
2104 bundle->mirror_out |= mirror_bit;
2106 bundle->mirror_out &= ~mirror_bit;
2110 ofproto->need_revalidate = true;
2111 mac_learning_flush(ofproto->ml, &ofproto->revalidate_set);
2112 mirror_update_dups(ofproto);
2118 mirror_destroy(struct ofmirror *mirror)
2120 struct ofproto_dpif *ofproto;
2121 mirror_mask_t mirror_bit;
2122 struct ofbundle *bundle;
2128 ofproto = mirror->ofproto;
2129 ofproto->need_revalidate = true;
2130 mac_learning_flush(ofproto->ml, &ofproto->revalidate_set);
2132 mirror_bit = MIRROR_MASK_C(1) << mirror->idx;
2133 HMAP_FOR_EACH (bundle, hmap_node, &ofproto->bundles) {
2134 bundle->src_mirrors &= ~mirror_bit;
2135 bundle->dst_mirrors &= ~mirror_bit;
2136 bundle->mirror_out &= ~mirror_bit;
2139 hmapx_destroy(&mirror->srcs);
2140 hmapx_destroy(&mirror->dsts);
2141 free(mirror->vlans);
2143 ofproto->mirrors[mirror->idx] = NULL;
2147 mirror_update_dups(ofproto);
2151 mirror_get_stats(struct ofproto *ofproto_, void *aux,
2152 uint64_t *packets, uint64_t *bytes)
2154 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
2155 struct ofmirror *mirror = mirror_lookup(ofproto, aux);
2158 *packets = *bytes = UINT64_MAX;
2162 *packets = mirror->packet_count;
2163 *bytes = mirror->byte_count;
2169 set_flood_vlans(struct ofproto *ofproto_, unsigned long *flood_vlans)
2171 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
2172 if (mac_learning_set_flood_vlans(ofproto->ml, flood_vlans)) {
2173 mac_learning_flush(ofproto->ml, &ofproto->revalidate_set);
2179 is_mirror_output_bundle(const struct ofproto *ofproto_, void *aux)
2181 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
2182 struct ofbundle *bundle = bundle_lookup(ofproto, aux);
2183 return bundle && bundle->mirror_out != 0;
2187 forward_bpdu_changed(struct ofproto *ofproto_)
2189 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
2190 /* Revalidate cached flows whenever forward_bpdu option changes. */
2191 ofproto->need_revalidate = true;
2195 set_mac_idle_time(struct ofproto *ofproto_, unsigned int idle_time)
2197 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
2198 mac_learning_set_idle_time(ofproto->ml, idle_time);
2203 static struct ofport_dpif *
2204 get_ofp_port(struct ofproto_dpif *ofproto, uint16_t ofp_port)
2206 struct ofport *ofport = ofproto_get_port(&ofproto->up, ofp_port);
2207 return ofport ? ofport_dpif_cast(ofport) : NULL;
2210 static struct ofport_dpif *
2211 get_odp_port(struct ofproto_dpif *ofproto, uint32_t odp_port)
2213 return get_ofp_port(ofproto, odp_port_to_ofp_port(odp_port));
2217 ofproto_port_from_dpif_port(struct ofproto_port *ofproto_port,
2218 struct dpif_port *dpif_port)
2220 ofproto_port->name = dpif_port->name;
2221 ofproto_port->type = dpif_port->type;
2222 ofproto_port->ofp_port = odp_port_to_ofp_port(dpif_port->port_no);
2226 port_run(struct ofport_dpif *ofport)
2228 long long int carrier_seq = netdev_get_carrier_resets(ofport->up.netdev);
2229 bool carrier_changed = carrier_seq != ofport->carrier_seq;
2230 bool enable = netdev_get_carrier(ofport->up.netdev);
2232 ofport->carrier_seq = carrier_seq;
2235 cfm_run(ofport->cfm);
2237 if (cfm_should_send_ccm(ofport->cfm)) {
2238 struct ofpbuf packet;
2240 ofpbuf_init(&packet, 0);
2241 cfm_compose_ccm(ofport->cfm, &packet, ofport->up.pp.hw_addr);
2242 send_packet(ofport, &packet);
2243 ofpbuf_uninit(&packet);
2246 enable = enable && !cfm_get_fault(ofport->cfm)
2247 && cfm_get_opup(ofport->cfm);
2250 if (ofport->bundle) {
2251 enable = enable && lacp_slave_may_enable(ofport->bundle->lacp, ofport);
2252 if (carrier_changed) {
2253 lacp_slave_carrier_changed(ofport->bundle->lacp, ofport);
2257 if (ofport->may_enable != enable) {
2258 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport->up.ofproto);
2260 if (ofproto->has_bundle_action) {
2261 ofproto->need_revalidate = true;
2265 ofport->may_enable = enable;
2269 port_wait(struct ofport_dpif *ofport)
2272 cfm_wait(ofport->cfm);
2277 port_query_by_name(const struct ofproto *ofproto_, const char *devname,
2278 struct ofproto_port *ofproto_port)
2280 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
2281 struct dpif_port dpif_port;
2284 error = dpif_port_query_by_name(ofproto->dpif, devname, &dpif_port);
2286 ofproto_port_from_dpif_port(ofproto_port, &dpif_port);
2292 port_add(struct ofproto *ofproto_, struct netdev *netdev, uint16_t *ofp_portp)
2294 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
2298 error = dpif_port_add(ofproto->dpif, netdev, &odp_port);
2300 *ofp_portp = odp_port_to_ofp_port(odp_port);
2306 port_del(struct ofproto *ofproto_, uint16_t ofp_port)
2308 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
2311 error = dpif_port_del(ofproto->dpif, ofp_port_to_odp_port(ofp_port));
2313 struct ofport_dpif *ofport = get_ofp_port(ofproto, ofp_port);
2315 /* The caller is going to close ofport->up.netdev. If this is a
2316 * bonded port, then the bond is using that netdev, so remove it
2317 * from the bond. The client will need to reconfigure everything
2318 * after deleting ports, so then the slave will get re-added. */
2319 bundle_remove(&ofport->up);
2326 port_get_stats(const struct ofport *ofport_, struct netdev_stats *stats)
2328 struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
2331 error = netdev_get_stats(ofport->up.netdev, stats);
2333 if (!error && ofport->odp_port == OVSP_LOCAL) {
2334 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport->up.ofproto);
2336 /* ofproto->stats.tx_packets represents packets that we created
2337 * internally and sent to some port (e.g. packets sent with
2338 * send_packet()). Account for them as if they had come from
2339 * OFPP_LOCAL and got forwarded. */
2341 if (stats->rx_packets != UINT64_MAX) {
2342 stats->rx_packets += ofproto->stats.tx_packets;
2345 if (stats->rx_bytes != UINT64_MAX) {
2346 stats->rx_bytes += ofproto->stats.tx_bytes;
2349 /* ofproto->stats.rx_packets represents packets that were received on
2350 * some port and we processed internally and dropped (e.g. STP).
2351 * Account fro them as if they had been forwarded to OFPP_LOCAL. */
2353 if (stats->tx_packets != UINT64_MAX) {
2354 stats->tx_packets += ofproto->stats.rx_packets;
2357 if (stats->tx_bytes != UINT64_MAX) {
2358 stats->tx_bytes += ofproto->stats.rx_bytes;
2365 /* Account packets for LOCAL port. */
2367 ofproto_update_local_port_stats(const struct ofproto *ofproto_,
2368 size_t tx_size, size_t rx_size)
2370 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
2373 ofproto->stats.rx_packets++;
2374 ofproto->stats.rx_bytes += rx_size;
2377 ofproto->stats.tx_packets++;
2378 ofproto->stats.tx_bytes += tx_size;
2382 struct port_dump_state {
2383 struct dpif_port_dump dump;
2388 port_dump_start(const struct ofproto *ofproto_, void **statep)
2390 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
2391 struct port_dump_state *state;
2393 *statep = state = xmalloc(sizeof *state);
2394 dpif_port_dump_start(&state->dump, ofproto->dpif);
2395 state->done = false;
2400 port_dump_next(const struct ofproto *ofproto_ OVS_UNUSED, void *state_,
2401 struct ofproto_port *port)
2403 struct port_dump_state *state = state_;
2404 struct dpif_port dpif_port;
2406 if (dpif_port_dump_next(&state->dump, &dpif_port)) {
2407 ofproto_port_from_dpif_port(port, &dpif_port);
2410 int error = dpif_port_dump_done(&state->dump);
2412 return error ? error : EOF;
2417 port_dump_done(const struct ofproto *ofproto_ OVS_UNUSED, void *state_)
2419 struct port_dump_state *state = state_;
2422 dpif_port_dump_done(&state->dump);
2429 port_poll(const struct ofproto *ofproto_, char **devnamep)
2431 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
2432 return dpif_port_poll(ofproto->dpif, devnamep);
2436 port_poll_wait(const struct ofproto *ofproto_)
2438 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
2439 dpif_port_poll_wait(ofproto->dpif);
2443 port_is_lacp_current(const struct ofport *ofport_)
2445 const struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
2446 return (ofport->bundle && ofport->bundle->lacp
2447 ? lacp_slave_is_current(ofport->bundle->lacp, ofport)
2451 /* Upcall handling. */
2453 /* Flow miss batching.
2455 * Some dpifs implement operations faster when you hand them off in a batch.
2456 * To allow batching, "struct flow_miss" queues the dpif-related work needed
2457 * for a given flow. Each "struct flow_miss" corresponds to sending one or
2458 * more packets, plus possibly installing the flow in the dpif.
2460 * So far we only batch the operations that affect flow setup time the most.
2461 * It's possible to batch more than that, but the benefit might be minimal. */
2463 struct hmap_node hmap_node;
2465 enum odp_key_fitness key_fitness;
2466 const struct nlattr *key;
2468 ovs_be16 initial_tci;
2469 struct list packets;
2472 struct flow_miss_op {
2473 struct dpif_op dpif_op;
2474 struct subfacet *subfacet;
2477 /* Sends an OFPT_PACKET_IN message for 'packet' of type OFPR_NO_MATCH to each
2478 * OpenFlow controller as necessary according to their individual
2479 * configurations. */
2481 send_packet_in_miss(struct ofproto_dpif *ofproto, const struct ofpbuf *packet,
2482 const struct flow *flow)
2484 struct ofputil_packet_in pin;
2486 pin.packet = packet->data;
2487 pin.packet_len = packet->size;
2488 pin.reason = OFPR_NO_MATCH;
2489 pin.controller_id = 0;
2494 pin.send_len = 0; /* not used for flow table misses */
2496 flow_get_metadata(flow, &pin.fmd);
2498 /* Registers aren't meaningful on a miss. */
2499 memset(pin.fmd.reg_masks, 0, sizeof pin.fmd.reg_masks);
2501 connmgr_send_packet_in(ofproto->up.connmgr, &pin);
2505 process_special(struct ofproto_dpif *ofproto, const struct flow *flow,
2506 const struct ofpbuf *packet)
2508 struct ofport_dpif *ofport = get_ofp_port(ofproto, flow->in_port);
2514 if (ofport->cfm && cfm_should_process_flow(ofport->cfm, flow)) {
2516 cfm_process_heartbeat(ofport->cfm, packet);
2519 } else if (ofport->bundle && ofport->bundle->lacp
2520 && flow->dl_type == htons(ETH_TYPE_LACP)) {
2522 lacp_process_packet(ofport->bundle->lacp, ofport, packet);
2525 } else if (ofproto->stp && stp_should_process_flow(flow)) {
2527 stp_process_packet(ofport, packet);
2534 static struct flow_miss *
2535 flow_miss_create(struct hmap *todo, const struct flow *flow,
2536 enum odp_key_fitness key_fitness,
2537 const struct nlattr *key, size_t key_len,
2538 ovs_be16 initial_tci)
2540 uint32_t hash = flow_hash(flow, 0);
2541 struct flow_miss *miss;
2543 HMAP_FOR_EACH_WITH_HASH (miss, hmap_node, hash, todo) {
2544 if (flow_equal(&miss->flow, flow)) {
2549 miss = xmalloc(sizeof *miss);
2550 hmap_insert(todo, &miss->hmap_node, hash);
2552 miss->key_fitness = key_fitness;
2554 miss->key_len = key_len;
2555 miss->initial_tci = initial_tci;
2556 list_init(&miss->packets);
2561 handle_flow_miss(struct ofproto_dpif *ofproto, struct flow_miss *miss,
2562 struct flow_miss_op *ops, size_t *n_ops)
2564 const struct flow *flow = &miss->flow;
2565 struct subfacet *subfacet;
2566 struct ofpbuf *packet;
2567 struct facet *facet;
2569 facet = facet_lookup_valid(ofproto, flow);
2571 struct rule_dpif *rule;
2573 rule = rule_dpif_lookup(ofproto, flow, 0);
2575 /* Don't send a packet-in if OFPUTIL_PC_NO_PACKET_IN asserted. */
2576 struct ofport_dpif *port = get_ofp_port(ofproto, flow->in_port);
2578 if (port->up.pp.config & OFPUTIL_PC_NO_PACKET_IN) {
2579 COVERAGE_INC(ofproto_dpif_no_packet_in);
2580 /* XXX install 'drop' flow entry */
2584 VLOG_WARN_RL(&rl, "packet-in on unknown port %"PRIu16,
2588 LIST_FOR_EACH (packet, list_node, &miss->packets) {
2589 send_packet_in_miss(ofproto, packet, flow);
2595 facet = facet_create(rule, flow);
2598 subfacet = subfacet_create(facet,
2599 miss->key_fitness, miss->key, miss->key_len,
2602 LIST_FOR_EACH (packet, list_node, &miss->packets) {
2603 struct dpif_flow_stats stats;
2604 struct flow_miss_op *op;
2605 struct dpif_execute *execute;
2607 ofproto->n_matches++;
2609 if (facet->rule->up.cr.priority == FAIL_OPEN_PRIORITY) {
2611 * Extra-special case for fail-open mode.
2613 * We are in fail-open mode and the packet matched the fail-open
2614 * rule, but we are connected to a controller too. We should send
2615 * the packet up to the controller in the hope that it will try to
2616 * set up a flow and thereby allow us to exit fail-open.
2618 * See the top-level comment in fail-open.c for more information.
2620 send_packet_in_miss(ofproto, packet, flow);
2623 if (!facet->may_install || !subfacet->actions) {
2624 subfacet_make_actions(subfacet, packet);
2627 dpif_flow_stats_extract(&facet->flow, packet, &stats);
2628 subfacet_update_stats(subfacet, &stats);
2630 if (!subfacet->actions_len) {
2631 /* No actions to execute, so skip talking to the dpif. */
2635 if (flow->vlan_tci != subfacet->initial_tci) {
2636 /* This packet was received on a VLAN splinter port. We added
2637 * a VLAN to the packet to make the packet resemble the flow,
2638 * but the actions were composed assuming that the packet
2639 * contained no VLAN. So, we must remove the VLAN header from
2640 * the packet before trying to execute the actions. */
2641 eth_pop_vlan(packet);
2644 op = &ops[(*n_ops)++];
2645 execute = &op->dpif_op.u.execute;
2646 op->subfacet = subfacet;
2647 op->dpif_op.type = DPIF_OP_EXECUTE;
2648 execute->key = miss->key;
2649 execute->key_len = miss->key_len;
2650 execute->actions = (facet->may_install
2652 : xmemdup(subfacet->actions,
2653 subfacet->actions_len));
2654 execute->actions_len = subfacet->actions_len;
2655 execute->packet = packet;
2658 if (facet->may_install && subfacet->key_fitness != ODP_FIT_TOO_LITTLE) {
2659 struct flow_miss_op *op = &ops[(*n_ops)++];
2660 struct dpif_flow_put *put = &op->dpif_op.u.flow_put;
2662 op->subfacet = subfacet;
2663 op->dpif_op.type = DPIF_OP_FLOW_PUT;
2664 put->flags = DPIF_FP_CREATE | DPIF_FP_MODIFY;
2665 put->key = miss->key;
2666 put->key_len = miss->key_len;
2667 put->actions = subfacet->actions;
2668 put->actions_len = subfacet->actions_len;
2673 /* Like odp_flow_key_to_flow(), this function converts the 'key_len' bytes of
2674 * OVS_KEY_ATTR_* attributes in 'key' to a flow structure in 'flow' and returns
2675 * an ODP_FIT_* value that indicates how well 'key' fits our expectations for
2676 * what a flow key should contain.
2678 * This function also includes some logic to help make VLAN splinters
2679 * transparent to the rest of the upcall processing logic. In particular, if
2680 * the extracted in_port is a VLAN splinter port, it replaces flow->in_port by
2681 * the "real" port, sets flow->vlan_tci correctly for the VLAN of the VLAN
2682 * splinter port, and pushes a VLAN header onto 'packet' (if it is nonnull).
2684 * Sets '*initial_tci' to the VLAN TCI with which the packet was really
2685 * received, that is, the actual VLAN TCI extracted by odp_flow_key_to_flow().
2686 * (This differs from the value returned in flow->vlan_tci only for packets
2687 * received on VLAN splinters.)
2689 static enum odp_key_fitness
2690 ofproto_dpif_extract_flow_key(const struct ofproto_dpif *ofproto,
2691 const struct nlattr *key, size_t key_len,
2692 struct flow *flow, ovs_be16 *initial_tci,
2693 struct ofpbuf *packet)
2695 enum odp_key_fitness fitness;
2699 fitness = odp_flow_key_to_flow(key, key_len, flow);
2700 if (fitness == ODP_FIT_ERROR) {
2703 *initial_tci = flow->vlan_tci;
2705 realdev = vsp_vlandev_to_realdev(ofproto, flow->in_port, &vid);
2707 /* Cause the flow to be processed as if it came in on the real device
2708 * with the VLAN device's VLAN ID. */
2709 flow->in_port = realdev;
2710 flow->vlan_tci = htons((vid & VLAN_VID_MASK) | VLAN_CFI);
2712 /* Make the packet resemble the flow, so that it gets sent to an
2713 * OpenFlow controller properly, so that it looks correct for
2714 * sFlow, and so that flow_extract() will get the correct vlan_tci
2715 * if it is called on 'packet'.
2717 * The allocated space inside 'packet' probably also contains
2718 * 'key', that is, both 'packet' and 'key' are probably part of a
2719 * struct dpif_upcall (see the large comment on that structure
2720 * definition), so pushing data on 'packet' is in general not a
2721 * good idea since it could overwrite 'key' or free it as a side
2722 * effect. However, it's OK in this special case because we know
2723 * that 'packet' is inside a Netlink attribute: pushing 4 bytes
2724 * will just overwrite the 4-byte "struct nlattr", which is fine
2725 * since we don't need that header anymore. */
2726 eth_push_vlan(packet, flow->vlan_tci);
2729 /* Let the caller know that we can't reproduce 'key' from 'flow'. */
2730 if (fitness == ODP_FIT_PERFECT) {
2731 fitness = ODP_FIT_TOO_MUCH;
2739 handle_miss_upcalls(struct ofproto_dpif *ofproto, struct dpif_upcall *upcalls,
2742 struct dpif_upcall *upcall;
2743 struct flow_miss *miss, *next_miss;
2744 struct flow_miss_op flow_miss_ops[FLOW_MISS_MAX_BATCH * 2];
2745 struct dpif_op *dpif_ops[FLOW_MISS_MAX_BATCH * 2];
2754 /* Construct the to-do list.
2756 * This just amounts to extracting the flow from each packet and sticking
2757 * the packets that have the same flow in the same "flow_miss" structure so
2758 * that we can process them together. */
2760 for (upcall = upcalls; upcall < &upcalls[n_upcalls]; upcall++) {
2761 enum odp_key_fitness fitness;
2762 struct flow_miss *miss;
2763 ovs_be16 initial_tci;
2766 /* Obtain metadata and check userspace/kernel agreement on flow match,
2767 * then set 'flow''s header pointers. */
2768 fitness = ofproto_dpif_extract_flow_key(ofproto,
2769 upcall->key, upcall->key_len,
2770 &flow, &initial_tci,
2772 if (fitness == ODP_FIT_ERROR) {
2773 ofpbuf_delete(upcall->packet);
2776 flow_extract(upcall->packet, flow.skb_priority, flow.tun_id,
2777 flow.in_port, &flow);
2779 /* Handle 802.1ag, LACP, and STP specially. */
2780 if (process_special(ofproto, &flow, upcall->packet)) {
2781 ofproto_update_local_port_stats(&ofproto->up,
2782 0, upcall->packet->size);
2783 ofpbuf_delete(upcall->packet);
2784 ofproto->n_matches++;
2788 /* Add other packets to a to-do list. */
2789 miss = flow_miss_create(&todo, &flow, fitness,
2790 upcall->key, upcall->key_len, initial_tci);
2791 list_push_back(&miss->packets, &upcall->packet->list_node);
2794 /* Process each element in the to-do list, constructing the set of
2795 * operations to batch. */
2797 HMAP_FOR_EACH (miss, hmap_node, &todo) {
2798 handle_flow_miss(ofproto, miss, flow_miss_ops, &n_ops);
2800 assert(n_ops <= ARRAY_SIZE(flow_miss_ops));
2802 /* Execute batch. */
2803 for (i = 0; i < n_ops; i++) {
2804 dpif_ops[i] = &flow_miss_ops[i].dpif_op;
2806 dpif_operate(ofproto->dpif, dpif_ops, n_ops);
2808 /* Free memory and update facets. */
2809 for (i = 0; i < n_ops; i++) {
2810 struct flow_miss_op *op = &flow_miss_ops[i];
2811 struct dpif_execute *execute;
2813 switch (op->dpif_op.type) {
2814 case DPIF_OP_EXECUTE:
2815 execute = &op->dpif_op.u.execute;
2816 if (op->subfacet->actions != execute->actions) {
2817 free((struct nlattr *) execute->actions);
2821 case DPIF_OP_FLOW_PUT:
2822 if (!op->dpif_op.error) {
2823 op->subfacet->installed = true;
2827 case DPIF_OP_FLOW_DEL:
2831 HMAP_FOR_EACH_SAFE (miss, next_miss, hmap_node, &todo) {
2832 ofpbuf_list_delete(&miss->packets);
2833 hmap_remove(&todo, &miss->hmap_node);
2836 hmap_destroy(&todo);
2840 handle_userspace_upcall(struct ofproto_dpif *ofproto,
2841 struct dpif_upcall *upcall)
2843 struct user_action_cookie cookie;
2844 enum odp_key_fitness fitness;
2845 ovs_be16 initial_tci;
2848 memcpy(&cookie, &upcall->userdata, sizeof(cookie));
2850 fitness = ofproto_dpif_extract_flow_key(ofproto, upcall->key,
2851 upcall->key_len, &flow,
2852 &initial_tci, upcall->packet);
2853 if (fitness == ODP_FIT_ERROR) {
2854 ofpbuf_delete(upcall->packet);
2858 if (cookie.type == USER_ACTION_COOKIE_SFLOW) {
2859 if (ofproto->sflow) {
2860 dpif_sflow_received(ofproto->sflow, upcall->packet, &flow,
2864 VLOG_WARN_RL(&rl, "invalid user cookie : 0x%"PRIx64, upcall->userdata);
2866 ofpbuf_delete(upcall->packet);
2870 handle_upcalls(struct ofproto_dpif *ofproto, unsigned int max_batch)
2872 struct dpif_upcall misses[FLOW_MISS_MAX_BATCH];
2876 assert (max_batch <= FLOW_MISS_MAX_BATCH);
2879 for (i = 0; i < max_batch; i++) {
2880 struct dpif_upcall *upcall = &misses[n_misses];
2883 error = dpif_recv(ofproto->dpif, upcall);
2888 switch (upcall->type) {
2889 case DPIF_UC_ACTION:
2890 handle_userspace_upcall(ofproto, upcall);
2894 /* Handle it later. */
2898 case DPIF_N_UC_TYPES:
2900 VLOG_WARN_RL(&rl, "upcall has unexpected type %"PRIu32,
2906 handle_miss_upcalls(ofproto, misses, n_misses);
2911 /* Flow expiration. */
2913 static int subfacet_max_idle(const struct ofproto_dpif *);
2914 static void update_stats(struct ofproto_dpif *);
2915 static void rule_expire(struct rule_dpif *);
2916 static void expire_subfacets(struct ofproto_dpif *, int dp_max_idle);
2918 /* This function is called periodically by run(). Its job is to collect
2919 * updates for the flows that have been installed into the datapath, most
2920 * importantly when they last were used, and then use that information to
2921 * expire flows that have not been used recently.
2923 * Returns the number of milliseconds after which it should be called again. */
2925 expire(struct ofproto_dpif *ofproto)
2927 struct rule_dpif *rule, *next_rule;
2928 struct oftable *table;
2931 /* Update stats for each flow in the datapath. */
2932 update_stats(ofproto);
2934 /* Expire subfacets that have been idle too long. */
2935 dp_max_idle = subfacet_max_idle(ofproto);
2936 expire_subfacets(ofproto, dp_max_idle);
2938 /* Expire OpenFlow flows whose idle_timeout or hard_timeout has passed. */
2939 OFPROTO_FOR_EACH_TABLE (table, &ofproto->up) {
2940 struct cls_cursor cursor;
2942 cls_cursor_init(&cursor, &table->cls, NULL);
2943 CLS_CURSOR_FOR_EACH_SAFE (rule, next_rule, up.cr, &cursor) {
2948 /* All outstanding data in existing flows has been accounted, so it's a
2949 * good time to do bond rebalancing. */
2950 if (ofproto->has_bonded_bundles) {
2951 struct ofbundle *bundle;
2953 HMAP_FOR_EACH (bundle, hmap_node, &ofproto->bundles) {
2955 bond_rebalance(bundle->bond, &ofproto->revalidate_set);
2960 return MIN(dp_max_idle, 1000);
2963 /* Update 'packet_count', 'byte_count', and 'used' members of installed facets.
2965 * This function also pushes statistics updates to rules which each facet
2966 * resubmits into. Generally these statistics will be accurate. However, if a
2967 * facet changes the rule it resubmits into at some time in between
2968 * update_stats() runs, it is possible that statistics accrued to the
2969 * old rule will be incorrectly attributed to the new rule. This could be
2970 * avoided by calling update_stats() whenever rules are created or
2971 * deleted. However, the performance impact of making so many calls to the
2972 * datapath do not justify the benefit of having perfectly accurate statistics.
2975 update_stats(struct ofproto_dpif *p)
2977 const struct dpif_flow_stats *stats;
2978 struct dpif_flow_dump dump;
2979 const struct nlattr *key;
2982 dpif_flow_dump_start(&dump, p->dpif);
2983 while (dpif_flow_dump_next(&dump, &key, &key_len, NULL, NULL, &stats)) {
2984 struct subfacet *subfacet;
2986 subfacet = subfacet_find(p, key, key_len);
2987 if (subfacet && subfacet->installed) {
2988 struct facet *facet = subfacet->facet;
2990 if (stats->n_packets >= subfacet->dp_packet_count) {
2991 uint64_t extra = stats->n_packets - subfacet->dp_packet_count;
2992 facet->packet_count += extra;
2994 VLOG_WARN_RL(&rl, "unexpected packet count from the datapath");
2997 if (stats->n_bytes >= subfacet->dp_byte_count) {
2998 facet->byte_count += stats->n_bytes - subfacet->dp_byte_count;
3000 VLOG_WARN_RL(&rl, "unexpected byte count from datapath");
3003 subfacet->dp_packet_count = stats->n_packets;
3004 subfacet->dp_byte_count = stats->n_bytes;
3006 facet->tcp_flags |= stats->tcp_flags;
3008 subfacet_update_time(subfacet, stats->used);
3009 facet_account(facet, true);
3010 facet_push_stats(facet);
3012 if (!VLOG_DROP_WARN(&rl)) {
3016 odp_flow_key_format(key, key_len, &s);
3017 VLOG_WARN("unexpected flow from datapath %s", ds_cstr(&s));
3021 COVERAGE_INC(facet_unexpected);
3022 /* There's a flow in the datapath that we know nothing about, or a
3023 * flow that shouldn't be installed but was anyway. Delete it. */
3024 dpif_flow_del(p->dpif, key, key_len, NULL);
3027 dpif_flow_dump_done(&dump);
3030 /* Calculates and returns the number of milliseconds of idle time after which
3031 * subfacets should expire from the datapath. When a subfacet expires, we fold
3032 * its statistics into its facet, and when a facet's last subfacet expires, we
3033 * fold its statistic into its rule. */
3035 subfacet_max_idle(const struct ofproto_dpif *ofproto)
3038 * Idle time histogram.
3040 * Most of the time a switch has a relatively small number of subfacets.
3041 * When this is the case we might as well keep statistics for all of them
3042 * in userspace and to cache them in the kernel datapath for performance as
3045 * As the number of subfacets increases, the memory required to maintain
3046 * statistics about them in userspace and in the kernel becomes
3047 * significant. However, with a large number of subfacets it is likely
3048 * that only a few of them are "heavy hitters" that consume a large amount
3049 * of bandwidth. At this point, only heavy hitters are worth caching in
3050 * the kernel and maintaining in userspaces; other subfacets we can
3053 * The technique used to compute the idle time is to build a histogram with
3054 * N_BUCKETS buckets whose width is BUCKET_WIDTH msecs each. Each subfacet
3055 * that is installed in the kernel gets dropped in the appropriate bucket.
3056 * After the histogram has been built, we compute the cutoff so that only
3057 * the most-recently-used 1% of subfacets (but at least
3058 * ofproto->up.flow_eviction_threshold flows) are kept cached. At least
3059 * the most-recently-used bucket of subfacets is kept, so actually an
3060 * arbitrary number of subfacets can be kept in any given expiration run
3061 * (though the next run will delete most of those unless they receive
3064 * This requires a second pass through the subfacets, in addition to the
3065 * pass made by update_stats(), because the former function never looks at
3066 * uninstallable subfacets.
3068 enum { BUCKET_WIDTH = ROUND_UP(100, TIME_UPDATE_INTERVAL) };
3069 enum { N_BUCKETS = 5000 / BUCKET_WIDTH };
3070 int buckets[N_BUCKETS] = { 0 };
3071 int total, subtotal, bucket;
3072 struct subfacet *subfacet;
3076 total = hmap_count(&ofproto->subfacets);
3077 if (total <= ofproto->up.flow_eviction_threshold) {
3078 return N_BUCKETS * BUCKET_WIDTH;
3081 /* Build histogram. */
3083 HMAP_FOR_EACH (subfacet, hmap_node, &ofproto->subfacets) {
3084 long long int idle = now - subfacet->used;
3085 int bucket = (idle <= 0 ? 0
3086 : idle >= BUCKET_WIDTH * N_BUCKETS ? N_BUCKETS - 1
3087 : (unsigned int) idle / BUCKET_WIDTH);
3091 /* Find the first bucket whose flows should be expired. */
3092 subtotal = bucket = 0;
3094 subtotal += buckets[bucket++];
3095 } while (bucket < N_BUCKETS &&
3096 subtotal < MAX(ofproto->up.flow_eviction_threshold, total / 100));
3098 if (VLOG_IS_DBG_ENABLED()) {
3102 ds_put_cstr(&s, "keep");
3103 for (i = 0; i < N_BUCKETS; i++) {
3105 ds_put_cstr(&s, ", drop");
3108 ds_put_format(&s, " %d:%d", i * BUCKET_WIDTH, buckets[i]);
3111 VLOG_INFO("%s: %s (msec:count)", ofproto->up.name, ds_cstr(&s));
3115 return bucket * BUCKET_WIDTH;
3118 enum { EXPIRE_MAX_BATCH = 50 };
3121 expire_batch(struct ofproto_dpif *ofproto, struct subfacet **subfacets, int n)
3123 struct odputil_keybuf keybufs[EXPIRE_MAX_BATCH];
3124 struct dpif_op ops[EXPIRE_MAX_BATCH];
3125 struct dpif_op *opsp[EXPIRE_MAX_BATCH];
3126 struct ofpbuf keys[EXPIRE_MAX_BATCH];
3127 struct dpif_flow_stats stats[EXPIRE_MAX_BATCH];
3130 for (i = 0; i < n; i++) {
3131 ops[i].type = DPIF_OP_FLOW_DEL;
3132 subfacet_get_key(subfacets[i], &keybufs[i], &keys[i]);
3133 ops[i].u.flow_del.key = keys[i].data;
3134 ops[i].u.flow_del.key_len = keys[i].size;
3135 ops[i].u.flow_del.stats = &stats[i];
3139 dpif_operate(ofproto->dpif, opsp, n);
3140 for (i = 0; i < n; i++) {
3141 subfacet_reset_dp_stats(subfacets[i], &stats[i]);
3142 subfacets[i]->installed = false;
3143 subfacet_destroy(subfacets[i]);
3148 expire_subfacets(struct ofproto_dpif *ofproto, int dp_max_idle)
3150 long long int cutoff = time_msec() - dp_max_idle;
3152 struct subfacet *subfacet, *next_subfacet;
3153 struct subfacet *batch[EXPIRE_MAX_BATCH];
3157 HMAP_FOR_EACH_SAFE (subfacet, next_subfacet, hmap_node,
3158 &ofproto->subfacets) {
3159 if (subfacet->used < cutoff) {
3160 if (subfacet->installed) {
3161 batch[n_batch++] = subfacet;
3162 if (n_batch >= EXPIRE_MAX_BATCH) {
3163 expire_batch(ofproto, batch, n_batch);
3167 subfacet_destroy(subfacet);
3173 expire_batch(ofproto, batch, n_batch);
3177 /* If 'rule' is an OpenFlow rule, that has expired according to OpenFlow rules,
3178 * then delete it entirely. */
3180 rule_expire(struct rule_dpif *rule)
3182 struct facet *facet, *next_facet;
3186 /* Has 'rule' expired? */
3188 if (rule->up.hard_timeout
3189 && now > rule->up.modified + rule->up.hard_timeout * 1000) {
3190 reason = OFPRR_HARD_TIMEOUT;
3191 } else if (rule->up.idle_timeout
3192 && now > rule->up.used + rule->up.idle_timeout * 1000) {
3193 reason = OFPRR_IDLE_TIMEOUT;
3198 COVERAGE_INC(ofproto_dpif_expired);
3200 /* Update stats. (This is a no-op if the rule expired due to an idle
3201 * timeout, because that only happens when the rule has no facets left.) */
3202 LIST_FOR_EACH_SAFE (facet, next_facet, list_node, &rule->facets) {
3203 facet_remove(facet);
3206 /* Get rid of the rule. */
3207 ofproto_rule_expire(&rule->up, reason);
3212 /* Creates and returns a new facet owned by 'rule', given a 'flow'.
3214 * The caller must already have determined that no facet with an identical
3215 * 'flow' exists in 'ofproto' and that 'flow' is the best match for 'rule' in
3216 * the ofproto's classifier table.
3218 * The facet will initially have no subfacets. The caller should create (at
3219 * least) one subfacet with subfacet_create(). */
3220 static struct facet *
3221 facet_create(struct rule_dpif *rule, const struct flow *flow)
3223 struct ofproto_dpif *ofproto = ofproto_dpif_cast(rule->up.ofproto);
3224 struct facet *facet;
3226 facet = xzalloc(sizeof *facet);
3227 facet->used = time_msec();
3228 hmap_insert(&ofproto->facets, &facet->hmap_node, flow_hash(flow, 0));
3229 list_push_back(&rule->facets, &facet->list_node);
3231 facet->flow = *flow;
3232 list_init(&facet->subfacets);
3233 netflow_flow_init(&facet->nf_flow);
3234 netflow_flow_update_time(ofproto->netflow, &facet->nf_flow, facet->used);
3240 facet_free(struct facet *facet)
3245 /* Executes, within 'ofproto', the 'n_actions' actions in 'actions' on
3246 * 'packet', which arrived on 'in_port'.
3248 * Takes ownership of 'packet'. */
3250 execute_odp_actions(struct ofproto_dpif *ofproto, const struct flow *flow,
3251 const struct nlattr *odp_actions, size_t actions_len,
3252 struct ofpbuf *packet)
3254 struct odputil_keybuf keybuf;
3258 ofpbuf_use_stack(&key, &keybuf, sizeof keybuf);
3259 odp_flow_key_from_flow(&key, flow);
3261 error = dpif_execute(ofproto->dpif, key.data, key.size,
3262 odp_actions, actions_len, packet);
3264 ofpbuf_delete(packet);
3268 /* Remove 'facet' from 'ofproto' and free up the associated memory:
3270 * - If 'facet' was installed in the datapath, uninstalls it and updates its
3271 * rule's statistics, via subfacet_uninstall().
3273 * - Removes 'facet' from its rule and from ofproto->facets.
3276 facet_remove(struct facet *facet)
3278 struct ofproto_dpif *ofproto = ofproto_dpif_cast(facet->rule->up.ofproto);
3279 struct subfacet *subfacet, *next_subfacet;
3281 assert(!list_is_empty(&facet->subfacets));
3283 /* First uninstall all of the subfacets to get final statistics. */
3284 LIST_FOR_EACH (subfacet, list_node, &facet->subfacets) {
3285 subfacet_uninstall(subfacet);
3288 /* Flush the final stats to the rule.
3290 * This might require us to have at least one subfacet around so that we
3291 * can use its actions for accounting in facet_account(), which is why we
3292 * have uninstalled but not yet destroyed the subfacets. */
3293 facet_flush_stats(facet);
3295 /* Now we're really all done so destroy everything. */
3296 LIST_FOR_EACH_SAFE (subfacet, next_subfacet, list_node,
3297 &facet->subfacets) {
3298 subfacet_destroy__(subfacet);
3300 hmap_remove(&ofproto->facets, &facet->hmap_node);
3301 list_remove(&facet->list_node);
3306 facet_account(struct facet *facet, bool may_flow_mod)
3308 struct ofproto_dpif *ofproto = ofproto_dpif_cast(facet->rule->up.ofproto);
3310 struct subfacet *subfacet;
3311 const struct nlattr *a;
3315 if (facet->byte_count <= facet->accounted_bytes) {
3318 n_bytes = facet->byte_count - facet->accounted_bytes;
3319 facet->accounted_bytes = facet->byte_count;
3321 /* Feed information from the active flows back into the learning table to
3322 * ensure that table is always in sync with what is actually flowing
3323 * through the datapath. */
3324 if (facet->has_learn || facet->has_normal
3325 || (facet->has_fin_timeout
3326 && facet->tcp_flags & (TCP_FIN | TCP_RST))) {
3327 struct action_xlate_ctx ctx;
3329 action_xlate_ctx_init(&ctx, ofproto, &facet->flow,
3330 facet->flow.vlan_tci,
3331 facet->rule, facet->tcp_flags, NULL);
3332 ctx.may_learn_macs = true;
3333 ctx.may_flow_mod = may_flow_mod;
3334 ofpbuf_delete(xlate_actions(&ctx, facet->rule->up.actions,
3335 facet->rule->up.n_actions));
3338 if (!facet->has_normal || !ofproto->has_bonded_bundles) {
3342 /* This loop feeds byte counters to bond_account() for rebalancing to use
3343 * as a basis. We also need to track the actual VLAN on which the packet
3344 * is going to be sent to ensure that it matches the one passed to
3345 * bond_choose_output_slave(). (Otherwise, we will account to the wrong
3348 * We use the actions from an arbitrary subfacet because they should all
3349 * be equally valid for our purpose. */
3350 subfacet = CONTAINER_OF(list_front(&facet->subfacets),
3351 struct subfacet, list_node);
3352 vlan_tci = facet->flow.vlan_tci;
3353 NL_ATTR_FOR_EACH_UNSAFE (a, left,
3354 subfacet->actions, subfacet->actions_len) {
3355 const struct ovs_action_push_vlan *vlan;
3356 struct ofport_dpif *port;
3358 switch (nl_attr_type(a)) {
3359 case OVS_ACTION_ATTR_OUTPUT:
3360 port = get_odp_port(ofproto, nl_attr_get_u32(a));
3361 if (port && port->bundle && port->bundle->bond) {
3362 bond_account(port->bundle->bond, &facet->flow,
3363 vlan_tci_to_vid(vlan_tci), n_bytes);
3367 case OVS_ACTION_ATTR_POP_VLAN:
3368 vlan_tci = htons(0);
3371 case OVS_ACTION_ATTR_PUSH_VLAN:
3372 vlan = nl_attr_get(a);
3373 vlan_tci = vlan->vlan_tci;
3379 /* Returns true if the only action for 'facet' is to send to the controller.
3380 * (We don't report NetFlow expiration messages for such facets because they
3381 * are just part of the control logic for the network, not real traffic). */
3383 facet_is_controller_flow(struct facet *facet)
3386 && facet->rule->up.n_actions == 1
3387 && action_outputs_to_port(&facet->rule->up.actions[0],
3388 htons(OFPP_CONTROLLER)));
3391 /* Folds all of 'facet''s statistics into its rule. Also updates the
3392 * accounting ofhook and emits a NetFlow expiration if appropriate. All of
3393 * 'facet''s statistics in the datapath should have been zeroed and folded into
3394 * its packet and byte counts before this function is called. */
3396 facet_flush_stats(struct facet *facet)
3398 struct ofproto_dpif *ofproto = ofproto_dpif_cast(facet->rule->up.ofproto);
3399 struct subfacet *subfacet;
3401 LIST_FOR_EACH (subfacet, list_node, &facet->subfacets) {
3402 assert(!subfacet->dp_byte_count);
3403 assert(!subfacet->dp_packet_count);
3406 facet_push_stats(facet);
3407 facet_account(facet, false);
3409 if (ofproto->netflow && !facet_is_controller_flow(facet)) {
3410 struct ofexpired expired;
3411 expired.flow = facet->flow;
3412 expired.packet_count = facet->packet_count;
3413 expired.byte_count = facet->byte_count;
3414 expired.used = facet->used;
3415 netflow_expire(ofproto->netflow, &facet->nf_flow, &expired);
3418 facet->rule->packet_count += facet->packet_count;
3419 facet->rule->byte_count += facet->byte_count;
3421 /* Reset counters to prevent double counting if 'facet' ever gets
3423 facet_reset_counters(facet);
3425 netflow_flow_clear(&facet->nf_flow);
3426 facet->tcp_flags = 0;
3429 /* Searches 'ofproto''s table of facets for one exactly equal to 'flow'.
3430 * Returns it if found, otherwise a null pointer.
3432 * The returned facet might need revalidation; use facet_lookup_valid()
3433 * instead if that is important. */
3434 static struct facet *
3435 facet_find(struct ofproto_dpif *ofproto, const struct flow *flow)
3437 struct facet *facet;
3439 HMAP_FOR_EACH_WITH_HASH (facet, hmap_node, flow_hash(flow, 0),
3441 if (flow_equal(flow, &facet->flow)) {
3449 /* Searches 'ofproto''s table of facets for one exactly equal to 'flow'.
3450 * Returns it if found, otherwise a null pointer.
3452 * The returned facet is guaranteed to be valid. */
3453 static struct facet *
3454 facet_lookup_valid(struct ofproto_dpif *ofproto, const struct flow *flow)
3456 struct facet *facet = facet_find(ofproto, flow);
3458 /* The facet we found might not be valid, since we could be in need of
3459 * revalidation. If it is not valid, don't return it. */
3461 && (ofproto->need_revalidate
3462 || tag_set_intersects(&ofproto->revalidate_set, facet->tags))
3463 && !facet_revalidate(facet)) {
3464 COVERAGE_INC(facet_invalidated);
3472 facet_check_consistency(struct facet *facet)
3474 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 15);
3476 struct ofproto_dpif *ofproto = ofproto_dpif_cast(facet->rule->up.ofproto);
3478 struct rule_dpif *rule;
3479 struct subfacet *subfacet;
3480 bool may_log = false;
3483 /* Check the rule for consistency. */
3484 rule = rule_dpif_lookup(ofproto, &facet->flow, 0);
3486 if (!VLOG_DROP_WARN(&rl)) {
3487 char *s = flow_to_string(&facet->flow);
3488 VLOG_WARN("%s: facet should not exist", s);
3492 } else if (rule != facet->rule) {
3493 may_log = !VLOG_DROP_WARN(&rl);
3499 flow_format(&s, &facet->flow);
3500 ds_put_format(&s, ": facet associated with wrong rule (was "
3501 "table=%"PRIu8",", facet->rule->up.table_id);
3502 cls_rule_format(&facet->rule->up.cr, &s);
3503 ds_put_format(&s, ") (should have been table=%"PRIu8",",
3505 cls_rule_format(&rule->up.cr, &s);
3506 ds_put_char(&s, ')');
3508 VLOG_WARN("%s", ds_cstr(&s));
3515 /* Check the datapath actions for consistency. */
3516 LIST_FOR_EACH (subfacet, list_node, &facet->subfacets) {
3517 struct action_xlate_ctx ctx;
3518 struct ofpbuf *odp_actions;
3519 bool actions_changed;
3520 bool should_install;
3522 action_xlate_ctx_init(&ctx, ofproto, &facet->flow,
3523 subfacet->initial_tci, rule, 0, NULL);
3524 odp_actions = xlate_actions(&ctx, rule->up.actions,
3525 rule->up.n_actions);
3527 should_install = (ctx.may_set_up_flow
3528 && subfacet->key_fitness != ODP_FIT_TOO_LITTLE);
3529 if (!should_install && !subfacet->installed) {
3530 /* The actions for uninstallable flows may vary from one packet to
3531 * the next, so don't compare the actions. */
3535 actions_changed = (subfacet->actions_len != odp_actions->size
3536 || memcmp(subfacet->actions, odp_actions->data,
3537 subfacet->actions_len));
3538 if (should_install != subfacet->installed || actions_changed) {
3540 may_log = !VLOG_DROP_WARN(&rl);
3545 struct odputil_keybuf keybuf;
3550 subfacet_get_key(subfacet, &keybuf, &key);
3551 odp_flow_key_format(key.data, key.size, &s);
3553 ds_put_cstr(&s, ": inconsistency in subfacet");
3554 if (should_install != subfacet->installed) {
3555 enum odp_key_fitness fitness = subfacet->key_fitness;
3557 ds_put_format(&s, " (should%s have been installed)",
3558 should_install ? "" : " not");
3559 ds_put_format(&s, " (may_set_up_flow=%s, fitness=%s)",
3560 ctx.may_set_up_flow ? "true" : "false",
3561 odp_key_fitness_to_string(fitness));
3563 if (actions_changed) {
3564 ds_put_cstr(&s, " (actions were: ");
3565 format_odp_actions(&s, subfacet->actions,
3566 subfacet->actions_len);
3567 ds_put_cstr(&s, ") (correct actions: ");
3568 format_odp_actions(&s, odp_actions->data,
3570 ds_put_char(&s, ')');
3572 ds_put_cstr(&s, " (actions: ");
3573 format_odp_actions(&s, subfacet->actions,
3574 subfacet->actions_len);
3575 ds_put_char(&s, ')');
3577 VLOG_WARN("%s", ds_cstr(&s));
3583 ofpbuf_delete(odp_actions);
3589 /* Re-searches the classifier for 'facet':
3591 * - If the rule found is different from 'facet''s current rule, moves
3592 * 'facet' to the new rule and recompiles its actions.
3594 * - If the rule found is the same as 'facet''s current rule, leaves 'facet'
3595 * where it is and recompiles its actions anyway.
3597 * - If there is none, destroys 'facet'.
3599 * Returns true if 'facet' still exists, false if it has been destroyed. */
3601 facet_revalidate(struct facet *facet)
3603 struct ofproto_dpif *ofproto = ofproto_dpif_cast(facet->rule->up.ofproto);
3605 struct nlattr *odp_actions;
3608 struct actions *new_actions;
3610 struct action_xlate_ctx ctx;
3611 struct rule_dpif *new_rule;
3612 struct subfacet *subfacet;
3613 bool actions_changed;
3616 COVERAGE_INC(facet_revalidate);
3618 /* Determine the new rule. */
3619 new_rule = rule_dpif_lookup(ofproto, &facet->flow, 0);
3621 /* No new rule, so delete the facet. */
3622 facet_remove(facet);
3626 /* Calculate new datapath actions.
3628 * We do not modify any 'facet' state yet, because we might need to, e.g.,
3629 * emit a NetFlow expiration and, if so, we need to have the old state
3630 * around to properly compose it. */
3632 /* If the datapath actions changed or the installability changed,
3633 * then we need to talk to the datapath. */
3636 memset(&ctx, 0, sizeof ctx);
3637 LIST_FOR_EACH (subfacet, list_node, &facet->subfacets) {
3638 struct ofpbuf *odp_actions;
3639 bool should_install;
3641 action_xlate_ctx_init(&ctx, ofproto, &facet->flow,
3642 subfacet->initial_tci, new_rule, 0, NULL);
3643 odp_actions = xlate_actions(&ctx, new_rule->up.actions,
3644 new_rule->up.n_actions);
3645 actions_changed = (subfacet->actions_len != odp_actions->size
3646 || memcmp(subfacet->actions, odp_actions->data,
3647 subfacet->actions_len));
3649 should_install = (ctx.may_set_up_flow
3650 && subfacet->key_fitness != ODP_FIT_TOO_LITTLE);
3651 if (actions_changed || should_install != subfacet->installed) {
3652 if (should_install) {
3653 struct dpif_flow_stats stats;
3655 subfacet_install(subfacet,
3656 odp_actions->data, odp_actions->size, &stats);
3657 subfacet_update_stats(subfacet, &stats);
3659 subfacet_uninstall(subfacet);
3663 new_actions = xcalloc(list_size(&facet->subfacets),
3664 sizeof *new_actions);
3666 new_actions[i].odp_actions = xmemdup(odp_actions->data,
3668 new_actions[i].actions_len = odp_actions->size;
3671 ofpbuf_delete(odp_actions);
3675 facet_flush_stats(facet);
3678 /* Update 'facet' now that we've taken care of all the old state. */
3679 facet->tags = ctx.tags;
3680 facet->nf_flow.output_iface = ctx.nf_output_iface;
3681 facet->may_install = ctx.may_set_up_flow;
3682 facet->has_learn = ctx.has_learn;
3683 facet->has_normal = ctx.has_normal;
3684 facet->has_fin_timeout = ctx.has_fin_timeout;
3685 facet->mirrors = ctx.mirrors;
3688 LIST_FOR_EACH (subfacet, list_node, &facet->subfacets) {
3689 if (new_actions[i].odp_actions) {
3690 free(subfacet->actions);
3691 subfacet->actions = new_actions[i].odp_actions;
3692 subfacet->actions_len = new_actions[i].actions_len;
3698 if (facet->rule != new_rule) {
3699 COVERAGE_INC(facet_changed_rule);
3700 list_remove(&facet->list_node);
3701 list_push_back(&new_rule->facets, &facet->list_node);
3702 facet->rule = new_rule;
3703 facet->used = new_rule->up.created;
3704 facet->prev_used = facet->used;
3710 /* Updates 'facet''s used time. Caller is responsible for calling
3711 * facet_push_stats() to update the flows which 'facet' resubmits into. */
3713 facet_update_time(struct facet *facet, long long int used)
3715 struct ofproto_dpif *ofproto = ofproto_dpif_cast(facet->rule->up.ofproto);
3716 if (used > facet->used) {
3718 ofproto_rule_update_used(&facet->rule->up, used);
3719 netflow_flow_update_time(ofproto->netflow, &facet->nf_flow, used);
3724 facet_reset_counters(struct facet *facet)
3726 facet->packet_count = 0;
3727 facet->byte_count = 0;
3728 facet->prev_packet_count = 0;
3729 facet->prev_byte_count = 0;
3730 facet->accounted_bytes = 0;
3734 facet_push_stats(struct facet *facet)
3736 uint64_t new_packets, new_bytes;
3738 assert(facet->packet_count >= facet->prev_packet_count);
3739 assert(facet->byte_count >= facet->prev_byte_count);
3740 assert(facet->used >= facet->prev_used);
3742 new_packets = facet->packet_count - facet->prev_packet_count;
3743 new_bytes = facet->byte_count - facet->prev_byte_count;
3745 if (new_packets || new_bytes || facet->used > facet->prev_used) {
3746 facet->prev_packet_count = facet->packet_count;
3747 facet->prev_byte_count = facet->byte_count;
3748 facet->prev_used = facet->used;
3750 flow_push_stats(facet->rule, &facet->flow,
3751 new_packets, new_bytes, facet->used);
3753 update_mirror_stats(ofproto_dpif_cast(facet->rule->up.ofproto),
3754 facet->mirrors, new_packets, new_bytes);
3758 struct ofproto_push {
3759 struct action_xlate_ctx ctx;
3766 push_resubmit(struct action_xlate_ctx *ctx, struct rule_dpif *rule)
3768 struct ofproto_push *push = CONTAINER_OF(ctx, struct ofproto_push, ctx);
3771 rule->packet_count += push->packets;
3772 rule->byte_count += push->bytes;
3773 ofproto_rule_update_used(&rule->up, push->used);
3777 /* Pushes flow statistics to the rules which 'flow' resubmits into given
3778 * 'rule''s actions and mirrors. */
3780 flow_push_stats(struct rule_dpif *rule,
3781 const struct flow *flow, uint64_t packets, uint64_t bytes,
3784 struct ofproto_dpif *ofproto = ofproto_dpif_cast(rule->up.ofproto);
3785 struct ofproto_push push;
3787 push.packets = packets;
3791 ofproto_rule_update_used(&rule->up, used);
3793 action_xlate_ctx_init(&push.ctx, ofproto, flow, flow->vlan_tci, rule,
3795 push.ctx.resubmit_hook = push_resubmit;
3796 ofpbuf_delete(xlate_actions(&push.ctx,
3797 rule->up.actions, rule->up.n_actions));
3802 static struct subfacet *
3803 subfacet_find__(struct ofproto_dpif *ofproto,
3804 const struct nlattr *key, size_t key_len, uint32_t key_hash,
3805 const struct flow *flow)
3807 struct subfacet *subfacet;
3809 HMAP_FOR_EACH_WITH_HASH (subfacet, hmap_node, key_hash,
3810 &ofproto->subfacets) {
3812 ? (subfacet->key_len == key_len
3813 && !memcmp(key, subfacet->key, key_len))
3814 : flow_equal(flow, &subfacet->facet->flow)) {
3822 /* Searches 'facet' (within 'ofproto') for a subfacet with the specified
3823 * 'key_fitness', 'key', and 'key_len'. Returns the existing subfacet if
3824 * there is one, otherwise creates and returns a new subfacet.
3826 * If the returned subfacet is new, then subfacet->actions will be NULL, in
3827 * which case the caller must populate the actions with
3828 * subfacet_make_actions(). */
3829 static struct subfacet *
3830 subfacet_create(struct facet *facet, enum odp_key_fitness key_fitness,
3831 const struct nlattr *key, size_t key_len, ovs_be16 initial_tci)
3833 struct ofproto_dpif *ofproto = ofproto_dpif_cast(facet->rule->up.ofproto);
3834 uint32_t key_hash = odp_flow_key_hash(key, key_len);
3835 struct subfacet *subfacet;
3837 subfacet = subfacet_find__(ofproto, key, key_len, key_hash, &facet->flow);
3839 if (subfacet->facet == facet) {
3843 /* This shouldn't happen. */
3844 VLOG_ERR_RL(&rl, "subfacet with wrong facet");
3845 subfacet_destroy(subfacet);
3848 subfacet = xzalloc(sizeof *subfacet);
3849 hmap_insert(&ofproto->subfacets, &subfacet->hmap_node, key_hash);
3850 list_push_back(&facet->subfacets, &subfacet->list_node);
3851 subfacet->facet = facet;
3852 subfacet->used = time_msec();
3853 subfacet->key_fitness = key_fitness;
3854 if (key_fitness != ODP_FIT_PERFECT) {
3855 subfacet->key = xmemdup(key, key_len);
3856 subfacet->key_len = key_len;
3858 subfacet->installed = false;
3859 subfacet->initial_tci = initial_tci;
3864 /* Searches 'ofproto' for a subfacet with the given 'key', 'key_len', and
3865 * 'flow'. Returns the subfacet if one exists, otherwise NULL. */
3866 static struct subfacet *
3867 subfacet_find(struct ofproto_dpif *ofproto,
3868 const struct nlattr *key, size_t key_len)
3870 uint32_t key_hash = odp_flow_key_hash(key, key_len);
3871 enum odp_key_fitness fitness;
3874 fitness = odp_flow_key_to_flow(key, key_len, &flow);
3875 if (fitness == ODP_FIT_ERROR) {
3879 return subfacet_find__(ofproto, key, key_len, key_hash, &flow);
3882 /* Uninstalls 'subfacet' from the datapath, if it is installed, removes it from
3883 * its facet within 'ofproto', and frees it. */
3885 subfacet_destroy__(struct subfacet *subfacet)
3887 struct facet *facet = subfacet->facet;
3888 struct ofproto_dpif *ofproto = ofproto_dpif_cast(facet->rule->up.ofproto);
3890 subfacet_uninstall(subfacet);
3891 hmap_remove(&ofproto->subfacets, &subfacet->hmap_node);
3892 list_remove(&subfacet->list_node);
3893 free(subfacet->key);
3894 free(subfacet->actions);
3898 /* Destroys 'subfacet', as with subfacet_destroy__(), and then if this was the
3899 * last remaining subfacet in its facet destroys the facet too. */
3901 subfacet_destroy(struct subfacet *subfacet)
3903 struct facet *facet = subfacet->facet;
3905 if (list_is_singleton(&facet->subfacets)) {
3906 /* facet_remove() needs at least one subfacet (it will remove it). */
3907 facet_remove(facet);
3909 subfacet_destroy__(subfacet);
3913 /* Initializes 'key' with the sequence of OVS_KEY_ATTR_* Netlink attributes
3914 * that can be used to refer to 'subfacet'. The caller must provide 'keybuf'
3915 * for use as temporary storage. */
3917 subfacet_get_key(struct subfacet *subfacet, struct odputil_keybuf *keybuf,
3920 if (!subfacet->key) {
3921 ofpbuf_use_stack(key, keybuf, sizeof *keybuf);
3922 odp_flow_key_from_flow(key, &subfacet->facet->flow);
3924 ofpbuf_use_const(key, subfacet->key, subfacet->key_len);
3928 /* Composes the datapath actions for 'subfacet' based on its rule's actions. */
3930 subfacet_make_actions(struct subfacet *subfacet, const struct ofpbuf *packet)
3932 struct facet *facet = subfacet->facet;
3933 struct rule_dpif *rule = facet->rule;
3934 struct ofproto_dpif *ofproto = ofproto_dpif_cast(rule->up.ofproto);
3935 struct ofpbuf *odp_actions;
3936 struct action_xlate_ctx ctx;
3938 action_xlate_ctx_init(&ctx, ofproto, &facet->flow, subfacet->initial_tci,
3940 odp_actions = xlate_actions(&ctx, rule->up.actions, rule->up.n_actions);
3941 facet->tags = ctx.tags;
3942 facet->may_install = ctx.may_set_up_flow;
3943 facet->has_learn = ctx.has_learn;
3944 facet->has_normal = ctx.has_normal;
3945 facet->has_fin_timeout = ctx.has_fin_timeout;
3946 facet->nf_flow.output_iface = ctx.nf_output_iface;
3947 facet->mirrors = ctx.mirrors;
3949 if (subfacet->actions_len != odp_actions->size
3950 || memcmp(subfacet->actions, odp_actions->data, odp_actions->size)) {
3951 free(subfacet->actions);
3952 subfacet->actions_len = odp_actions->size;
3953 subfacet->actions = xmemdup(odp_actions->data, odp_actions->size);
3956 ofpbuf_delete(odp_actions);
3959 /* Updates 'subfacet''s datapath flow, setting its actions to 'actions_len'
3960 * bytes of actions in 'actions'. If 'stats' is non-null, statistics counters
3961 * in the datapath will be zeroed and 'stats' will be updated with traffic new
3962 * since 'subfacet' was last updated.
3964 * Returns 0 if successful, otherwise a positive errno value. */
3966 subfacet_install(struct subfacet *subfacet,
3967 const struct nlattr *actions, size_t actions_len,
3968 struct dpif_flow_stats *stats)
3970 struct facet *facet = subfacet->facet;
3971 struct ofproto_dpif *ofproto = ofproto_dpif_cast(facet->rule->up.ofproto);
3972 struct odputil_keybuf keybuf;
3973 enum dpif_flow_put_flags flags;
3977 flags = DPIF_FP_CREATE | DPIF_FP_MODIFY;
3979 flags |= DPIF_FP_ZERO_STATS;
3982 subfacet_get_key(subfacet, &keybuf, &key);
3983 ret = dpif_flow_put(ofproto->dpif, flags, key.data, key.size,
3984 actions, actions_len, stats);
3987 subfacet_reset_dp_stats(subfacet, stats);
3993 /* If 'subfacet' is installed in the datapath, uninstalls it. */
3995 subfacet_uninstall(struct subfacet *subfacet)
3997 if (subfacet->installed) {
3998 struct rule_dpif *rule = subfacet->facet->rule;
3999 struct ofproto_dpif *ofproto = ofproto_dpif_cast(rule->up.ofproto);
4000 struct odputil_keybuf keybuf;
4001 struct dpif_flow_stats stats;
4005 subfacet_get_key(subfacet, &keybuf, &key);
4006 error = dpif_flow_del(ofproto->dpif, key.data, key.size, &stats);
4007 subfacet_reset_dp_stats(subfacet, &stats);
4009 subfacet_update_stats(subfacet, &stats);
4011 subfacet->installed = false;
4013 assert(subfacet->dp_packet_count == 0);
4014 assert(subfacet->dp_byte_count == 0);
4018 /* Resets 'subfacet''s datapath statistics counters. This should be called
4019 * when 'subfacet''s statistics are cleared in the datapath. If 'stats' is
4020 * non-null, it should contain the statistics returned by dpif when 'subfacet'
4021 * was reset in the datapath. 'stats' will be modified to include only
4022 * statistics new since 'subfacet' was last updated. */
4024 subfacet_reset_dp_stats(struct subfacet *subfacet,
4025 struct dpif_flow_stats *stats)
4028 && subfacet->dp_packet_count <= stats->n_packets
4029 && subfacet->dp_byte_count <= stats->n_bytes) {
4030 stats->n_packets -= subfacet->dp_packet_count;
4031 stats->n_bytes -= subfacet->dp_byte_count;
4034 subfacet->dp_packet_count = 0;
4035 subfacet->dp_byte_count = 0;
4038 /* Updates 'subfacet''s used time. The caller is responsible for calling
4039 * facet_push_stats() to update the flows which 'subfacet' resubmits into. */
4041 subfacet_update_time(struct subfacet *subfacet, long long int used)
4043 if (used > subfacet->used) {
4044 subfacet->used = used;
4045 facet_update_time(subfacet->facet, used);
4049 /* Folds the statistics from 'stats' into the counters in 'subfacet'.
4051 * Because of the meaning of a subfacet's counters, it only makes sense to do
4052 * this if 'stats' are not tracked in the datapath, that is, if 'stats'
4053 * represents a packet that was sent by hand or if it represents statistics
4054 * that have been cleared out of the datapath. */
4056 subfacet_update_stats(struct subfacet *subfacet,
4057 const struct dpif_flow_stats *stats)
4059 if (stats->n_packets || stats->used > subfacet->used) {
4060 struct facet *facet = subfacet->facet;
4062 subfacet_update_time(subfacet, stats->used);
4063 facet->packet_count += stats->n_packets;
4064 facet->byte_count += stats->n_bytes;
4065 facet->tcp_flags |= stats->tcp_flags;
4066 facet_push_stats(facet);
4067 netflow_flow_update_flags(&facet->nf_flow, stats->tcp_flags);
4073 static struct rule_dpif *
4074 rule_dpif_lookup(struct ofproto_dpif *ofproto, const struct flow *flow,
4077 struct cls_rule *cls_rule;
4078 struct classifier *cls;
4080 if (table_id >= N_TABLES) {
4084 cls = &ofproto->up.tables[table_id].cls;
4085 if (flow->nw_frag & FLOW_NW_FRAG_ANY
4086 && ofproto->up.frag_handling == OFPC_FRAG_NORMAL) {
4087 /* For OFPC_NORMAL frag_handling, we must pretend that transport ports
4088 * are unavailable. */
4089 struct flow ofpc_normal_flow = *flow;
4090 ofpc_normal_flow.tp_src = htons(0);
4091 ofpc_normal_flow.tp_dst = htons(0);
4092 cls_rule = classifier_lookup(cls, &ofpc_normal_flow);
4094 cls_rule = classifier_lookup(cls, flow);
4096 return rule_dpif_cast(rule_from_cls_rule(cls_rule));
4100 complete_operation(struct rule_dpif *rule)
4102 struct ofproto_dpif *ofproto = ofproto_dpif_cast(rule->up.ofproto);
4104 rule_invalidate(rule);
4106 struct dpif_completion *c = xmalloc(sizeof *c);
4107 c->op = rule->up.pending;
4108 list_push_back(&ofproto->completions, &c->list_node);
4110 ofoperation_complete(rule->up.pending, 0);
4114 static struct rule *
4117 struct rule_dpif *rule = xmalloc(sizeof *rule);
4122 rule_dealloc(struct rule *rule_)
4124 struct rule_dpif *rule = rule_dpif_cast(rule_);
4129 rule_construct(struct rule *rule_)
4131 struct rule_dpif *rule = rule_dpif_cast(rule_);
4132 struct ofproto_dpif *ofproto = ofproto_dpif_cast(rule->up.ofproto);
4133 struct rule_dpif *victim;
4137 error = validate_actions(rule->up.actions, rule->up.n_actions,
4138 &rule->up.cr.flow, ofproto->max_ports);
4143 rule->packet_count = 0;
4144 rule->byte_count = 0;
4146 victim = rule_dpif_cast(ofoperation_get_victim(rule->up.pending));
4147 if (victim && !list_is_empty(&victim->facets)) {
4148 struct facet *facet;
4150 rule->facets = victim->facets;
4151 list_moved(&rule->facets);
4152 LIST_FOR_EACH (facet, list_node, &rule->facets) {
4153 /* XXX: We're only clearing our local counters here. It's possible
4154 * that quite a few packets are unaccounted for in the datapath
4155 * statistics. These will be accounted to the new rule instead of
4156 * cleared as required. This could be fixed by clearing out the
4157 * datapath statistics for this facet, but currently it doesn't
4159 facet_reset_counters(facet);
4163 /* Must avoid list_moved() in this case. */
4164 list_init(&rule->facets);
4167 table_id = rule->up.table_id;
4168 rule->tag = (victim ? victim->tag
4170 : rule_calculate_tag(&rule->up.cr.flow, &rule->up.cr.wc,
4171 ofproto->tables[table_id].basis));
4173 complete_operation(rule);
4178 rule_destruct(struct rule *rule_)
4180 struct rule_dpif *rule = rule_dpif_cast(rule_);
4181 struct facet *facet, *next_facet;
4183 LIST_FOR_EACH_SAFE (facet, next_facet, list_node, &rule->facets) {
4184 facet_revalidate(facet);
4187 complete_operation(rule);
4191 rule_get_stats(struct rule *rule_, uint64_t *packets, uint64_t *bytes)
4193 struct rule_dpif *rule = rule_dpif_cast(rule_);
4194 struct facet *facet;
4196 /* Start from historical data for 'rule' itself that are no longer tracked
4197 * in facets. This counts, for example, facets that have expired. */
4198 *packets = rule->packet_count;
4199 *bytes = rule->byte_count;
4201 /* Add any statistics that are tracked by facets. This includes
4202 * statistical data recently updated by ofproto_update_stats() as well as
4203 * stats for packets that were executed "by hand" via dpif_execute(). */
4204 LIST_FOR_EACH (facet, list_node, &rule->facets) {
4205 *packets += facet->packet_count;
4206 *bytes += facet->byte_count;
4211 rule_execute(struct rule *rule_, const struct flow *flow,
4212 struct ofpbuf *packet)
4214 struct rule_dpif *rule = rule_dpif_cast(rule_);
4215 struct ofproto_dpif *ofproto = ofproto_dpif_cast(rule->up.ofproto);
4216 struct action_xlate_ctx ctx;
4217 struct ofpbuf *odp_actions;
4220 action_xlate_ctx_init(&ctx, ofproto, flow, flow->vlan_tci,
4221 rule, packet_get_tcp_flags(packet, flow), packet);
4222 odp_actions = xlate_actions(&ctx, rule->up.actions, rule->up.n_actions);
4223 size = packet->size;
4224 if (execute_odp_actions(ofproto, flow, odp_actions->data,
4225 odp_actions->size, packet)) {
4226 rule->packet_count++;
4227 rule->byte_count += size;
4228 flow_push_stats(rule, flow, 1, size, time_msec());
4230 ofpbuf_delete(odp_actions);
4236 rule_modify_actions(struct rule *rule_)
4238 struct rule_dpif *rule = rule_dpif_cast(rule_);
4239 struct ofproto_dpif *ofproto = ofproto_dpif_cast(rule->up.ofproto);
4242 error = validate_actions(rule->up.actions, rule->up.n_actions,
4243 &rule->up.cr.flow, ofproto->max_ports);
4245 ofoperation_complete(rule->up.pending, error);
4249 complete_operation(rule);
4252 /* Sends 'packet' out 'ofport'.
4253 * May modify 'packet'.
4254 * Returns 0 if successful, otherwise a positive errno value. */
4256 send_packet(const struct ofport_dpif *ofport, struct ofpbuf *packet)
4258 const struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport->up.ofproto);
4259 struct ofpbuf key, odp_actions;
4260 struct odputil_keybuf keybuf;
4265 flow_extract((struct ofpbuf *) packet, 0, 0, 0, &flow);
4266 odp_port = vsp_realdev_to_vlandev(ofproto, ofport->odp_port,
4268 if (odp_port != ofport->odp_port) {
4269 eth_pop_vlan(packet);
4270 flow.vlan_tci = htons(0);
4273 ofpbuf_use_stack(&key, &keybuf, sizeof keybuf);
4274 odp_flow_key_from_flow(&key, &flow);
4276 ofpbuf_init(&odp_actions, 32);
4277 compose_sflow_action(ofproto, &odp_actions, &flow, odp_port);
4279 nl_msg_put_u32(&odp_actions, OVS_ACTION_ATTR_OUTPUT, odp_port);
4280 error = dpif_execute(ofproto->dpif,
4282 odp_actions.data, odp_actions.size,
4284 ofpbuf_uninit(&odp_actions);
4287 VLOG_WARN_RL(&rl, "%s: failed to send packet on port %"PRIu32" (%s)",
4288 ofproto->up.name, odp_port, strerror(error));
4290 ofproto_update_local_port_stats(ofport->up.ofproto, packet->size, 0);
4294 /* OpenFlow to datapath action translation. */
4296 static void do_xlate_actions(const union ofp_action *in, size_t n_in,
4297 struct action_xlate_ctx *ctx);
4298 static void xlate_normal(struct action_xlate_ctx *);
4301 put_userspace_action(const struct ofproto_dpif *ofproto,
4302 struct ofpbuf *odp_actions,
4303 const struct flow *flow,
4304 const struct user_action_cookie *cookie)
4308 pid = dpif_port_get_pid(ofproto->dpif,
4309 ofp_port_to_odp_port(flow->in_port));
4311 return odp_put_userspace_action(pid, cookie, odp_actions);
4314 /* Compose SAMPLE action for sFlow. */
4316 compose_sflow_action(const struct ofproto_dpif *ofproto,
4317 struct ofpbuf *odp_actions,
4318 const struct flow *flow,
4321 uint32_t port_ifindex;
4322 uint32_t probability;
4323 struct user_action_cookie cookie;
4324 size_t sample_offset, actions_offset;
4325 int cookie_offset, n_output;
4327 if (!ofproto->sflow || flow->in_port == OFPP_NONE) {
4331 if (odp_port == OVSP_NONE) {
4335 port_ifindex = dpif_sflow_odp_port_to_ifindex(ofproto->sflow, odp_port);
4339 sample_offset = nl_msg_start_nested(odp_actions, OVS_ACTION_ATTR_SAMPLE);
4341 /* Number of packets out of UINT_MAX to sample. */
4342 probability = dpif_sflow_get_probability(ofproto->sflow);
4343 nl_msg_put_u32(odp_actions, OVS_SAMPLE_ATTR_PROBABILITY, probability);
4345 actions_offset = nl_msg_start_nested(odp_actions, OVS_SAMPLE_ATTR_ACTIONS);
4347 cookie.type = USER_ACTION_COOKIE_SFLOW;
4348 cookie.data = port_ifindex;
4349 cookie.n_output = n_output;
4350 cookie.vlan_tci = 0;
4351 cookie_offset = put_userspace_action(ofproto, odp_actions, flow, &cookie);
4353 nl_msg_end_nested(odp_actions, actions_offset);
4354 nl_msg_end_nested(odp_actions, sample_offset);
4355 return cookie_offset;
4358 /* SAMPLE action must be first action in any given list of actions.
4359 * At this point we do not have all information required to build it. So try to
4360 * build sample action as complete as possible. */
4362 add_sflow_action(struct action_xlate_ctx *ctx)
4364 ctx->user_cookie_offset = compose_sflow_action(ctx->ofproto,
4366 &ctx->flow, OVSP_NONE);
4367 ctx->sflow_odp_port = 0;
4368 ctx->sflow_n_outputs = 0;
4371 /* Fix SAMPLE action according to data collected while composing ODP actions.
4372 * We need to fix SAMPLE actions OVS_SAMPLE_ATTR_ACTIONS attribute, i.e. nested
4373 * USERSPACE action's user-cookie which is required for sflow. */
4375 fix_sflow_action(struct action_xlate_ctx *ctx)
4377 const struct flow *base = &ctx->base_flow;
4378 struct user_action_cookie *cookie;
4380 if (!ctx->user_cookie_offset) {
4384 cookie = ofpbuf_at(ctx->odp_actions, ctx->user_cookie_offset,
4386 assert(cookie != NULL);
4387 assert(cookie->type == USER_ACTION_COOKIE_SFLOW);
4389 if (ctx->sflow_n_outputs) {
4390 cookie->data = dpif_sflow_odp_port_to_ifindex(ctx->ofproto->sflow,
4391 ctx->sflow_odp_port);
4393 if (ctx->sflow_n_outputs >= 255) {
4394 cookie->n_output = 255;
4396 cookie->n_output = ctx->sflow_n_outputs;
4398 cookie->vlan_tci = base->vlan_tci;
4402 compose_output_action__(struct action_xlate_ctx *ctx, uint16_t ofp_port,
4405 const struct ofport_dpif *ofport = get_ofp_port(ctx->ofproto, ofp_port);
4406 uint16_t odp_port = ofp_port_to_odp_port(ofp_port);
4407 ovs_be16 flow_vlan_tci = ctx->flow.vlan_tci;
4408 uint8_t flow_nw_tos = ctx->flow.nw_tos;
4412 struct priority_to_dscp *pdscp;
4414 if (ofport->up.pp.config & OFPUTIL_PC_NO_FWD
4415 || (check_stp && !stp_forward_in_state(ofport->stp_state))) {
4419 pdscp = get_priority(ofport, ctx->flow.skb_priority);
4421 ctx->flow.nw_tos &= ~IP_DSCP_MASK;
4422 ctx->flow.nw_tos |= pdscp->dscp;
4425 /* We may not have an ofport record for this port, but it doesn't hurt
4426 * to allow forwarding to it anyhow. Maybe such a port will appear
4427 * later and we're pre-populating the flow table. */
4430 out_port = vsp_realdev_to_vlandev(ctx->ofproto, odp_port,
4431 ctx->flow.vlan_tci);
4432 if (out_port != odp_port) {
4433 ctx->flow.vlan_tci = htons(0);
4435 commit_odp_actions(&ctx->flow, &ctx->base_flow, ctx->odp_actions);
4436 nl_msg_put_u32(ctx->odp_actions, OVS_ACTION_ATTR_OUTPUT, out_port);
4438 ctx->sflow_odp_port = odp_port;
4439 ctx->sflow_n_outputs++;
4440 ctx->nf_output_iface = ofp_port;
4441 ctx->flow.vlan_tci = flow_vlan_tci;
4442 ctx->flow.nw_tos = flow_nw_tos;
4446 compose_output_action(struct action_xlate_ctx *ctx, uint16_t ofp_port)
4448 compose_output_action__(ctx, ofp_port, true);
4452 xlate_table_action(struct action_xlate_ctx *ctx,
4453 uint16_t in_port, uint8_t table_id)
4455 if (ctx->recurse < MAX_RESUBMIT_RECURSION) {
4456 struct ofproto_dpif *ofproto = ctx->ofproto;
4457 struct rule_dpif *rule;
4458 uint16_t old_in_port;
4459 uint8_t old_table_id;
4461 old_table_id = ctx->table_id;
4462 ctx->table_id = table_id;
4464 /* Look up a flow with 'in_port' as the input port. */
4465 old_in_port = ctx->flow.in_port;
4466 ctx->flow.in_port = in_port;
4467 rule = rule_dpif_lookup(ofproto, &ctx->flow, table_id);
4470 if (table_id > 0 && table_id < N_TABLES) {
4471 struct table_dpif *table = &ofproto->tables[table_id];
4472 if (table->other_table) {
4473 ctx->tags |= (rule && rule->tag
4475 : rule_calculate_tag(&ctx->flow,
4476 &table->other_table->wc,
4481 /* Restore the original input port. Otherwise OFPP_NORMAL and
4482 * OFPP_IN_PORT will have surprising behavior. */
4483 ctx->flow.in_port = old_in_port;
4485 if (ctx->resubmit_hook) {
4486 ctx->resubmit_hook(ctx, rule);
4490 struct rule_dpif *old_rule = ctx->rule;
4494 do_xlate_actions(rule->up.actions, rule->up.n_actions, ctx);
4495 ctx->rule = old_rule;
4499 ctx->table_id = old_table_id;
4501 static struct vlog_rate_limit recurse_rl = VLOG_RATE_LIMIT_INIT(1, 1);
4503 VLOG_ERR_RL(&recurse_rl, "resubmit actions recursed over %d times",
4504 MAX_RESUBMIT_RECURSION);
4505 ctx->max_resubmit_trigger = true;
4510 xlate_resubmit_table(struct action_xlate_ctx *ctx,
4511 const struct nx_action_resubmit *nar)
4516 in_port = (nar->in_port == htons(OFPP_IN_PORT)
4518 : ntohs(nar->in_port));
4519 table_id = nar->table == 255 ? ctx->table_id : nar->table;
4521 xlate_table_action(ctx, in_port, table_id);
4525 flood_packets(struct action_xlate_ctx *ctx, bool all)
4527 struct ofport_dpif *ofport;
4529 HMAP_FOR_EACH (ofport, up.hmap_node, &ctx->ofproto->up.ports) {
4530 uint16_t ofp_port = ofport->up.ofp_port;
4532 if (ofp_port == ctx->flow.in_port) {
4537 compose_output_action__(ctx, ofp_port, false);
4538 } else if (!(ofport->up.pp.config & OFPUTIL_PC_NO_FLOOD)) {
4539 compose_output_action(ctx, ofp_port);
4543 ctx->nf_output_iface = NF_OUT_FLOOD;
4547 execute_controller_action(struct action_xlate_ctx *ctx, int len,
4548 enum ofp_packet_in_reason reason,
4549 uint16_t controller_id)
4551 struct ofputil_packet_in pin;
4552 struct ofpbuf *packet;
4554 ctx->may_set_up_flow = false;
4559 packet = ofpbuf_clone(ctx->packet);
4561 if (packet->l2 && packet->l3) {
4562 struct eth_header *eh;
4564 eth_pop_vlan(packet);
4567 /* If the Ethernet type is less than ETH_TYPE_MIN, it's likely an 802.2
4568 * LLC frame. Calculating the Ethernet type of these frames is more
4569 * trouble than seems appropriate for a simple assertion. */
4570 assert(ntohs(eh->eth_type) < ETH_TYPE_MIN
4571 || eh->eth_type == ctx->flow.dl_type);
4573 memcpy(eh->eth_src, ctx->flow.dl_src, sizeof eh->eth_src);
4574 memcpy(eh->eth_dst, ctx->flow.dl_dst, sizeof eh->eth_dst);
4576 if (ctx->flow.vlan_tci & htons(VLAN_CFI)) {
4577 eth_push_vlan(packet, ctx->flow.vlan_tci);
4581 if (ctx->flow.dl_type == htons(ETH_TYPE_IP)) {
4582 packet_set_ipv4(packet, ctx->flow.nw_src, ctx->flow.nw_dst,
4583 ctx->flow.nw_tos, ctx->flow.nw_ttl);
4587 if (ctx->flow.nw_proto == IPPROTO_TCP) {
4588 packet_set_tcp_port(packet, ctx->flow.tp_src,
4590 } else if (ctx->flow.nw_proto == IPPROTO_UDP) {
4591 packet_set_udp_port(packet, ctx->flow.tp_src,
4598 pin.packet = packet->data;
4599 pin.packet_len = packet->size;
4600 pin.reason = reason;
4601 pin.controller_id = controller_id;
4602 pin.table_id = ctx->table_id;
4603 pin.cookie = ctx->rule ? ctx->rule->up.flow_cookie : 0;
4606 flow_get_metadata(&ctx->flow, &pin.fmd);
4608 connmgr_send_packet_in(ctx->ofproto->up.connmgr, &pin);
4609 ofpbuf_delete(packet);
4613 compose_dec_ttl(struct action_xlate_ctx *ctx)
4615 if (ctx->flow.dl_type != htons(ETH_TYPE_IP) &&
4616 ctx->flow.dl_type != htons(ETH_TYPE_IPV6)) {
4620 if (ctx->flow.nw_ttl > 1) {
4624 execute_controller_action(ctx, UINT16_MAX, OFPR_INVALID_TTL, 0);
4626 /* Stop processing for current table. */
4632 xlate_output_action__(struct action_xlate_ctx *ctx,
4633 uint16_t port, uint16_t max_len)
4635 uint16_t prev_nf_output_iface = ctx->nf_output_iface;
4637 ctx->nf_output_iface = NF_OUT_DROP;
4641 compose_output_action(ctx, ctx->flow.in_port);
4644 xlate_table_action(ctx, ctx->flow.in_port, ctx->table_id);
4650 flood_packets(ctx, false);
4653 flood_packets(ctx, true);
4655 case OFPP_CONTROLLER:
4656 execute_controller_action(ctx, max_len, OFPR_ACTION, 0);
4662 if (port != ctx->flow.in_port) {
4663 compose_output_action(ctx, port);
4668 if (prev_nf_output_iface == NF_OUT_FLOOD) {
4669 ctx->nf_output_iface = NF_OUT_FLOOD;
4670 } else if (ctx->nf_output_iface == NF_OUT_DROP) {
4671 ctx->nf_output_iface = prev_nf_output_iface;
4672 } else if (prev_nf_output_iface != NF_OUT_DROP &&
4673 ctx->nf_output_iface != NF_OUT_FLOOD) {
4674 ctx->nf_output_iface = NF_OUT_MULTI;
4679 xlate_output_reg_action(struct action_xlate_ctx *ctx,
4680 const struct nx_action_output_reg *naor)
4682 struct mf_subfield src;
4685 nxm_decode(&src, naor->src, naor->ofs_nbits);
4686 ofp_port = mf_get_subfield(&src, &ctx->flow);
4688 if (ofp_port <= UINT16_MAX) {
4689 xlate_output_action__(ctx, ofp_port, ntohs(naor->max_len));
4694 xlate_output_action(struct action_xlate_ctx *ctx,
4695 const struct ofp_action_output *oao)
4697 xlate_output_action__(ctx, ntohs(oao->port), ntohs(oao->max_len));
4701 xlate_enqueue_action(struct action_xlate_ctx *ctx,
4702 const struct ofp_action_enqueue *oae)
4705 uint32_t flow_priority, priority;
4708 error = dpif_queue_to_priority(ctx->ofproto->dpif, ntohl(oae->queue_id),
4711 /* Fall back to ordinary output action. */
4712 xlate_output_action__(ctx, ntohs(oae->port), 0);
4716 /* Figure out datapath output port. */
4717 ofp_port = ntohs(oae->port);
4718 if (ofp_port == OFPP_IN_PORT) {
4719 ofp_port = ctx->flow.in_port;
4720 } else if (ofp_port == ctx->flow.in_port) {
4724 /* Add datapath actions. */
4725 flow_priority = ctx->flow.skb_priority;
4726 ctx->flow.skb_priority = priority;
4727 compose_output_action(ctx, ofp_port);
4728 ctx->flow.skb_priority = flow_priority;
4730 /* Update NetFlow output port. */
4731 if (ctx->nf_output_iface == NF_OUT_DROP) {
4732 ctx->nf_output_iface = ofp_port;
4733 } else if (ctx->nf_output_iface != NF_OUT_FLOOD) {
4734 ctx->nf_output_iface = NF_OUT_MULTI;
4739 xlate_set_queue_action(struct action_xlate_ctx *ctx,
4740 const struct nx_action_set_queue *nasq)
4745 error = dpif_queue_to_priority(ctx->ofproto->dpif, ntohl(nasq->queue_id),
4748 /* Couldn't translate queue to a priority, so ignore. A warning
4749 * has already been logged. */
4753 ctx->flow.skb_priority = priority;
4756 struct xlate_reg_state {
4762 xlate_autopath(struct action_xlate_ctx *ctx,
4763 const struct nx_action_autopath *naa)
4765 uint16_t ofp_port = ntohl(naa->id);
4766 struct ofport_dpif *port = get_ofp_port(ctx->ofproto, ofp_port);
4768 if (!port || !port->bundle) {
4769 ofp_port = OFPP_NONE;
4770 } else if (port->bundle->bond) {
4771 /* Autopath does not support VLAN hashing. */
4772 struct ofport_dpif *slave = bond_choose_output_slave(
4773 port->bundle->bond, &ctx->flow, 0, &ctx->tags);
4775 ofp_port = slave->up.ofp_port;
4778 autopath_execute(naa, &ctx->flow, ofp_port);
4782 slave_enabled_cb(uint16_t ofp_port, void *ofproto_)
4784 struct ofproto_dpif *ofproto = ofproto_;
4785 struct ofport_dpif *port;
4795 case OFPP_CONTROLLER: /* Not supported by the bundle action. */
4798 port = get_ofp_port(ofproto, ofp_port);
4799 return port ? port->may_enable : false;
4804 xlate_learn_action(struct action_xlate_ctx *ctx,
4805 const struct nx_action_learn *learn)
4807 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(5, 1);
4808 struct ofputil_flow_mod fm;
4811 learn_execute(learn, &ctx->flow, &fm);
4813 error = ofproto_flow_mod(&ctx->ofproto->up, &fm);
4814 if (error && !VLOG_DROP_WARN(&rl)) {
4815 VLOG_WARN("learning action failed to modify flow table (%s)",
4816 ofperr_get_name(error));
4822 /* Reduces '*timeout' to no more than 'max'. A value of zero in either case
4823 * means "infinite". */
4825 reduce_timeout(uint16_t max, uint16_t *timeout)
4827 if (max && (!*timeout || *timeout > max)) {
4833 xlate_fin_timeout(struct action_xlate_ctx *ctx,
4834 const struct nx_action_fin_timeout *naft)
4836 if (ctx->tcp_flags & (TCP_FIN | TCP_RST) && ctx->rule) {
4837 struct rule_dpif *rule = ctx->rule;
4839 reduce_timeout(ntohs(naft->fin_idle_timeout), &rule->up.idle_timeout);
4840 reduce_timeout(ntohs(naft->fin_hard_timeout), &rule->up.hard_timeout);
4845 may_receive(const struct ofport_dpif *port, struct action_xlate_ctx *ctx)
4847 if (port->up.pp.config & (eth_addr_equals(ctx->flow.dl_dst, eth_addr_stp)
4848 ? OFPUTIL_PC_NO_RECV_STP
4849 : OFPUTIL_PC_NO_RECV)) {
4853 /* Only drop packets here if both forwarding and learning are
4854 * disabled. If just learning is enabled, we need to have
4855 * OFPP_NORMAL and the learning action have a look at the packet
4856 * before we can drop it. */
4857 if (!stp_forward_in_state(port->stp_state)
4858 && !stp_learn_in_state(port->stp_state)) {
4866 do_xlate_actions(const union ofp_action *in, size_t n_in,
4867 struct action_xlate_ctx *ctx)
4869 const struct ofport_dpif *port;
4870 const union ofp_action *ia;
4871 bool was_evictable = true;
4874 port = get_ofp_port(ctx->ofproto, ctx->flow.in_port);
4875 if (port && !may_receive(port, ctx)) {
4876 /* Drop this flow. */
4881 /* Don't let the rule we're working on get evicted underneath us. */
4882 was_evictable = ctx->rule->up.evictable;
4883 ctx->rule->up.evictable = false;
4885 OFPUTIL_ACTION_FOR_EACH_UNSAFE (ia, left, in, n_in) {
4886 const struct ofp_action_dl_addr *oada;
4887 const struct nx_action_resubmit *nar;
4888 const struct nx_action_set_tunnel *nast;
4889 const struct nx_action_set_queue *nasq;
4890 const struct nx_action_multipath *nam;
4891 const struct nx_action_autopath *naa;
4892 const struct nx_action_bundle *nab;
4893 const struct nx_action_output_reg *naor;
4894 const struct nx_action_controller *nac;
4895 enum ofputil_action_code code;
4902 code = ofputil_decode_action_unsafe(ia);
4904 case OFPUTIL_OFPAT10_OUTPUT:
4905 xlate_output_action(ctx, &ia->output);
4908 case OFPUTIL_OFPAT10_SET_VLAN_VID:
4909 ctx->flow.vlan_tci &= ~htons(VLAN_VID_MASK);
4910 ctx->flow.vlan_tci |= ia->vlan_vid.vlan_vid | htons(VLAN_CFI);
4913 case OFPUTIL_OFPAT10_SET_VLAN_PCP:
4914 ctx->flow.vlan_tci &= ~htons(VLAN_PCP_MASK);
4915 ctx->flow.vlan_tci |= htons(
4916 (ia->vlan_pcp.vlan_pcp << VLAN_PCP_SHIFT) | VLAN_CFI);
4919 case OFPUTIL_OFPAT10_STRIP_VLAN:
4920 ctx->flow.vlan_tci = htons(0);
4923 case OFPUTIL_OFPAT10_SET_DL_SRC:
4924 oada = ((struct ofp_action_dl_addr *) ia);
4925 memcpy(ctx->flow.dl_src, oada->dl_addr, ETH_ADDR_LEN);
4928 case OFPUTIL_OFPAT10_SET_DL_DST:
4929 oada = ((struct ofp_action_dl_addr *) ia);
4930 memcpy(ctx->flow.dl_dst, oada->dl_addr, ETH_ADDR_LEN);
4933 case OFPUTIL_OFPAT10_SET_NW_SRC:
4934 ctx->flow.nw_src = ia->nw_addr.nw_addr;
4937 case OFPUTIL_OFPAT10_SET_NW_DST:
4938 ctx->flow.nw_dst = ia->nw_addr.nw_addr;
4941 case OFPUTIL_OFPAT10_SET_NW_TOS:
4942 /* OpenFlow 1.0 only supports IPv4. */
4943 if (ctx->flow.dl_type == htons(ETH_TYPE_IP)) {
4944 ctx->flow.nw_tos &= ~IP_DSCP_MASK;
4945 ctx->flow.nw_tos |= ia->nw_tos.nw_tos & IP_DSCP_MASK;
4949 case OFPUTIL_OFPAT10_SET_TP_SRC:
4950 ctx->flow.tp_src = ia->tp_port.tp_port;
4953 case OFPUTIL_OFPAT10_SET_TP_DST:
4954 ctx->flow.tp_dst = ia->tp_port.tp_port;
4957 case OFPUTIL_OFPAT10_ENQUEUE:
4958 xlate_enqueue_action(ctx, (const struct ofp_action_enqueue *) ia);
4961 case OFPUTIL_NXAST_RESUBMIT:
4962 nar = (const struct nx_action_resubmit *) ia;
4963 xlate_table_action(ctx, ntohs(nar->in_port), ctx->table_id);
4966 case OFPUTIL_NXAST_RESUBMIT_TABLE:
4967 xlate_resubmit_table(ctx, (const struct nx_action_resubmit *) ia);
4970 case OFPUTIL_NXAST_SET_TUNNEL:
4971 nast = (const struct nx_action_set_tunnel *) ia;
4972 tun_id = htonll(ntohl(nast->tun_id));
4973 ctx->flow.tun_id = tun_id;
4976 case OFPUTIL_NXAST_SET_QUEUE:
4977 nasq = (const struct nx_action_set_queue *) ia;
4978 xlate_set_queue_action(ctx, nasq);
4981 case OFPUTIL_NXAST_POP_QUEUE:
4982 ctx->flow.skb_priority = ctx->orig_skb_priority;
4985 case OFPUTIL_NXAST_REG_MOVE:
4986 nxm_execute_reg_move((const struct nx_action_reg_move *) ia,
4990 case OFPUTIL_NXAST_REG_LOAD:
4991 nxm_execute_reg_load((const struct nx_action_reg_load *) ia,
4995 case OFPUTIL_NXAST_NOTE:
4996 /* Nothing to do. */
4999 case OFPUTIL_NXAST_SET_TUNNEL64:
5000 tun_id = ((const struct nx_action_set_tunnel64 *) ia)->tun_id;
5001 ctx->flow.tun_id = tun_id;
5004 case OFPUTIL_NXAST_MULTIPATH:
5005 nam = (const struct nx_action_multipath *) ia;
5006 multipath_execute(nam, &ctx->flow);
5009 case OFPUTIL_NXAST_AUTOPATH:
5010 naa = (const struct nx_action_autopath *) ia;
5011 xlate_autopath(ctx, naa);
5014 case OFPUTIL_NXAST_BUNDLE:
5015 ctx->ofproto->has_bundle_action = true;
5016 nab = (const struct nx_action_bundle *) ia;
5017 xlate_output_action__(ctx, bundle_execute(nab, &ctx->flow,
5022 case OFPUTIL_NXAST_BUNDLE_LOAD:
5023 ctx->ofproto->has_bundle_action = true;
5024 nab = (const struct nx_action_bundle *) ia;
5025 bundle_execute_load(nab, &ctx->flow, slave_enabled_cb,
5029 case OFPUTIL_NXAST_OUTPUT_REG:
5030 naor = (const struct nx_action_output_reg *) ia;
5031 xlate_output_reg_action(ctx, naor);
5034 case OFPUTIL_NXAST_LEARN:
5035 ctx->has_learn = true;
5036 if (ctx->may_flow_mod) {
5037 xlate_learn_action(ctx, (const struct nx_action_learn *) ia);
5041 case OFPUTIL_NXAST_DEC_TTL:
5042 if (compose_dec_ttl(ctx)) {
5047 case OFPUTIL_NXAST_EXIT:
5051 case OFPUTIL_NXAST_FIN_TIMEOUT:
5052 ctx->has_fin_timeout = true;
5053 xlate_fin_timeout(ctx, (const struct nx_action_fin_timeout *) ia);
5056 case OFPUTIL_NXAST_CONTROLLER:
5057 nac = (const struct nx_action_controller *) ia;
5058 execute_controller_action(ctx, ntohs(nac->max_len), nac->reason,
5059 ntohs(nac->controller_id));
5065 /* We've let OFPP_NORMAL and the learning action look at the packet,
5066 * so drop it now if forwarding is disabled. */
5067 if (port && !stp_forward_in_state(port->stp_state)) {
5068 ofpbuf_clear(ctx->odp_actions);
5069 add_sflow_action(ctx);
5072 ctx->rule->up.evictable = was_evictable;
5077 action_xlate_ctx_init(struct action_xlate_ctx *ctx,
5078 struct ofproto_dpif *ofproto, const struct flow *flow,
5079 ovs_be16 initial_tci, struct rule_dpif *rule,
5080 uint8_t tcp_flags, const struct ofpbuf *packet)
5082 ctx->ofproto = ofproto;
5084 ctx->base_flow = ctx->flow;
5085 ctx->base_flow.tun_id = 0;
5086 ctx->base_flow.vlan_tci = initial_tci;
5088 ctx->packet = packet;
5089 ctx->may_learn_macs = packet != NULL;
5090 ctx->may_flow_mod = packet != NULL;
5091 ctx->tcp_flags = tcp_flags;
5092 ctx->resubmit_hook = NULL;
5095 static struct ofpbuf *
5096 xlate_actions(struct action_xlate_ctx *ctx,
5097 const union ofp_action *in, size_t n_in)
5099 struct flow orig_flow = ctx->flow;
5101 COVERAGE_INC(ofproto_dpif_xlate);
5103 ctx->odp_actions = ofpbuf_new(512);
5104 ofpbuf_reserve(ctx->odp_actions, NL_A_U32_SIZE);
5106 ctx->may_set_up_flow = true;
5107 ctx->has_learn = false;
5108 ctx->has_normal = false;
5109 ctx->has_fin_timeout = false;
5110 ctx->nf_output_iface = NF_OUT_DROP;
5113 ctx->max_resubmit_trigger = false;
5114 ctx->orig_skb_priority = ctx->flow.skb_priority;
5118 if (ctx->flow.nw_frag & FLOW_NW_FRAG_ANY) {
5119 switch (ctx->ofproto->up.frag_handling) {
5120 case OFPC_FRAG_NORMAL:
5121 /* We must pretend that transport ports are unavailable. */
5122 ctx->flow.tp_src = ctx->base_flow.tp_src = htons(0);
5123 ctx->flow.tp_dst = ctx->base_flow.tp_dst = htons(0);
5126 case OFPC_FRAG_DROP:
5127 return ctx->odp_actions;
5129 case OFPC_FRAG_REASM:
5132 case OFPC_FRAG_NX_MATCH:
5133 /* Nothing to do. */
5136 case OFPC_INVALID_TTL_TO_CONTROLLER:
5141 if (process_special(ctx->ofproto, &ctx->flow, ctx->packet)) {
5142 ctx->may_set_up_flow = false;
5143 return ctx->odp_actions;
5145 static struct vlog_rate_limit trace_rl = VLOG_RATE_LIMIT_INIT(1, 1);
5146 struct flow original_flow = ctx->flow;
5147 ovs_be16 initial_tci = ctx->base_flow.vlan_tci;
5149 add_sflow_action(ctx);
5150 do_xlate_actions(in, n_in, ctx);
5152 if (ctx->max_resubmit_trigger && !ctx->resubmit_hook
5153 && !VLOG_DROP_ERR(&trace_rl)) {
5154 struct ds ds = DS_EMPTY_INITIALIZER;
5156 ofproto_trace(ctx->ofproto, &original_flow, ctx->packet,
5158 VLOG_ERR("Trace triggered by excessive resubmit recursion:\n%s",
5163 if (!connmgr_may_set_up_flow(ctx->ofproto->up.connmgr, &ctx->flow,
5164 ctx->odp_actions->data,
5165 ctx->odp_actions->size)) {
5166 ctx->may_set_up_flow = false;
5168 && connmgr_msg_in_hook(ctx->ofproto->up.connmgr, &ctx->flow,
5170 compose_output_action(ctx, OFPP_LOCAL);
5173 add_mirror_actions(ctx, &orig_flow);
5174 fix_sflow_action(ctx);
5177 return ctx->odp_actions;
5180 /* OFPP_NORMAL implementation. */
5182 static struct ofport_dpif *ofbundle_get_a_port(const struct ofbundle *);
5184 /* Given 'vid', the VID obtained from the 802.1Q header that was received as
5185 * part of a packet (specify 0 if there was no 802.1Q header), and 'in_bundle',
5186 * the bundle on which the packet was received, returns the VLAN to which the
5189 * Both 'vid' and the return value are in the range 0...4095. */
5191 input_vid_to_vlan(const struct ofbundle *in_bundle, uint16_t vid)
5193 switch (in_bundle->vlan_mode) {
5194 case PORT_VLAN_ACCESS:
5195 return in_bundle->vlan;
5198 case PORT_VLAN_TRUNK:
5201 case PORT_VLAN_NATIVE_UNTAGGED:
5202 case PORT_VLAN_NATIVE_TAGGED:
5203 return vid ? vid : in_bundle->vlan;
5210 /* Checks whether a packet with the given 'vid' may ingress on 'in_bundle'.
5211 * If so, returns true. Otherwise, returns false and, if 'warn' is true, logs
5214 * 'vid' should be the VID obtained from the 802.1Q header that was received as
5215 * part of a packet (specify 0 if there was no 802.1Q header), in the range
5218 input_vid_is_valid(uint16_t vid, struct ofbundle *in_bundle, bool warn)
5220 /* Allow any VID on the OFPP_NONE port. */
5221 if (in_bundle == &ofpp_none_bundle) {
5225 switch (in_bundle->vlan_mode) {
5226 case PORT_VLAN_ACCESS:
5229 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
5230 VLOG_WARN_RL(&rl, "bridge %s: dropping VLAN %"PRIu16" tagged "
5231 "packet received on port %s configured as VLAN "
5232 "%"PRIu16" access port",
5233 in_bundle->ofproto->up.name, vid,
5234 in_bundle->name, in_bundle->vlan);
5240 case PORT_VLAN_NATIVE_UNTAGGED:
5241 case PORT_VLAN_NATIVE_TAGGED:
5243 /* Port must always carry its native VLAN. */
5247 case PORT_VLAN_TRUNK:
5248 if (!ofbundle_includes_vlan(in_bundle, vid)) {
5250 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
5251 VLOG_WARN_RL(&rl, "bridge %s: dropping VLAN %"PRIu16" packet "
5252 "received on port %s not configured for trunking "
5254 in_bundle->ofproto->up.name, vid,
5255 in_bundle->name, vid);
5267 /* Given 'vlan', the VLAN that a packet belongs to, and
5268 * 'out_bundle', a bundle on which the packet is to be output, returns the VID
5269 * that should be included in the 802.1Q header. (If the return value is 0,
5270 * then the 802.1Q header should only be included in the packet if there is a
5273 * Both 'vlan' and the return value are in the range 0...4095. */
5275 output_vlan_to_vid(const struct ofbundle *out_bundle, uint16_t vlan)
5277 switch (out_bundle->vlan_mode) {
5278 case PORT_VLAN_ACCESS:
5281 case PORT_VLAN_TRUNK:
5282 case PORT_VLAN_NATIVE_TAGGED:
5285 case PORT_VLAN_NATIVE_UNTAGGED:
5286 return vlan == out_bundle->vlan ? 0 : vlan;
5294 output_normal(struct action_xlate_ctx *ctx, const struct ofbundle *out_bundle,
5297 struct ofport_dpif *port;
5299 ovs_be16 tci, old_tci;
5301 vid = output_vlan_to_vid(out_bundle, vlan);
5302 if (!out_bundle->bond) {
5303 port = ofbundle_get_a_port(out_bundle);
5305 port = bond_choose_output_slave(out_bundle->bond, &ctx->flow,
5308 /* No slaves enabled, so drop packet. */
5313 old_tci = ctx->flow.vlan_tci;
5315 if (tci || out_bundle->use_priority_tags) {
5316 tci |= ctx->flow.vlan_tci & htons(VLAN_PCP_MASK);
5318 tci |= htons(VLAN_CFI);
5321 ctx->flow.vlan_tci = tci;
5323 compose_output_action(ctx, port->up.ofp_port);
5324 ctx->flow.vlan_tci = old_tci;
5328 mirror_mask_ffs(mirror_mask_t mask)
5330 BUILD_ASSERT_DECL(sizeof(unsigned int) >= sizeof(mask));
5335 ofbundle_trunks_vlan(const struct ofbundle *bundle, uint16_t vlan)
5337 return (bundle->vlan_mode != PORT_VLAN_ACCESS
5338 && (!bundle->trunks || bitmap_is_set(bundle->trunks, vlan)));
5342 ofbundle_includes_vlan(const struct ofbundle *bundle, uint16_t vlan)
5344 return vlan == bundle->vlan || ofbundle_trunks_vlan(bundle, vlan);
5347 /* Returns an arbitrary interface within 'bundle'. */
5348 static struct ofport_dpif *
5349 ofbundle_get_a_port(const struct ofbundle *bundle)
5351 return CONTAINER_OF(list_front(&bundle->ports),
5352 struct ofport_dpif, bundle_node);
5356 vlan_is_mirrored(const struct ofmirror *m, int vlan)
5358 return !m->vlans || bitmap_is_set(m->vlans, vlan);
5361 /* Returns true if a packet with Ethernet destination MAC 'dst' may be mirrored
5362 * to a VLAN. In general most packets may be mirrored but we want to drop
5363 * protocols that may confuse switches. */
5365 eth_dst_may_rspan(const uint8_t dst[ETH_ADDR_LEN])
5367 /* If you change this function's behavior, please update corresponding
5368 * documentation in vswitch.xml at the same time. */
5369 if (dst[0] != 0x01) {
5370 /* All the currently banned MACs happen to start with 01 currently, so
5371 * this is a quick way to eliminate most of the good ones. */
5373 if (eth_addr_is_reserved(dst)) {
5374 /* Drop STP, IEEE pause frames, and other reserved protocols
5375 * (01-80-c2-00-00-0x). */
5379 if (dst[0] == 0x01 && dst[1] == 0x00 && dst[2] == 0x0c) {
5381 if ((dst[3] & 0xfe) == 0xcc &&
5382 (dst[4] & 0xfe) == 0xcc &&
5383 (dst[5] & 0xfe) == 0xcc) {
5384 /* Drop the following protocols plus others following the same
5387 CDP, VTP, DTP, PAgP (01-00-0c-cc-cc-cc)
5388 Spanning Tree PVSTP+ (01-00-0c-cc-cc-cd)
5389 STP Uplink Fast (01-00-0c-cd-cd-cd) */
5393 if (!(dst[3] | dst[4] | dst[5])) {
5394 /* Drop Inter Switch Link packets (01-00-0c-00-00-00). */
5403 add_mirror_actions(struct action_xlate_ctx *ctx, const struct flow *orig_flow)
5405 struct ofproto_dpif *ofproto = ctx->ofproto;
5406 mirror_mask_t mirrors;
5407 struct ofbundle *in_bundle;
5410 const struct nlattr *a;
5413 in_bundle = lookup_input_bundle(ctx->ofproto, orig_flow->in_port,
5414 ctx->packet != NULL);
5418 mirrors = in_bundle->src_mirrors;
5420 /* Drop frames on bundles reserved for mirroring. */
5421 if (in_bundle->mirror_out) {
5422 if (ctx->packet != NULL) {
5423 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
5424 VLOG_WARN_RL(&rl, "bridge %s: dropping packet received on port "
5425 "%s, which is reserved exclusively for mirroring",
5426 ctx->ofproto->up.name, in_bundle->name);
5432 vid = vlan_tci_to_vid(orig_flow->vlan_tci);
5433 if (!input_vid_is_valid(vid, in_bundle, ctx->packet != NULL)) {
5436 vlan = input_vid_to_vlan(in_bundle, vid);
5438 /* Look at the output ports to check for destination selections. */
5440 NL_ATTR_FOR_EACH (a, left, ctx->odp_actions->data,
5441 ctx->odp_actions->size) {
5442 enum ovs_action_attr type = nl_attr_type(a);
5443 struct ofport_dpif *ofport;
5445 if (type != OVS_ACTION_ATTR_OUTPUT) {
5449 ofport = get_odp_port(ofproto, nl_attr_get_u32(a));
5450 if (ofport && ofport->bundle) {
5451 mirrors |= ofport->bundle->dst_mirrors;
5459 /* Restore the original packet before adding the mirror actions. */
5460 ctx->flow = *orig_flow;
5465 m = ofproto->mirrors[mirror_mask_ffs(mirrors) - 1];
5467 if (!vlan_is_mirrored(m, vlan)) {
5468 mirrors &= mirrors - 1;
5472 mirrors &= ~m->dup_mirrors;
5473 ctx->mirrors |= m->dup_mirrors;
5475 output_normal(ctx, m->out, vlan);
5476 } else if (eth_dst_may_rspan(orig_flow->dl_dst)
5477 && vlan != m->out_vlan) {
5478 struct ofbundle *bundle;
5480 HMAP_FOR_EACH (bundle, hmap_node, &ofproto->bundles) {
5481 if (ofbundle_includes_vlan(bundle, m->out_vlan)
5482 && !bundle->mirror_out) {
5483 output_normal(ctx, bundle, m->out_vlan);
5491 update_mirror_stats(struct ofproto_dpif *ofproto, mirror_mask_t mirrors,
5492 uint64_t packets, uint64_t bytes)
5498 for (; mirrors; mirrors &= mirrors - 1) {
5501 m = ofproto->mirrors[mirror_mask_ffs(mirrors) - 1];
5504 /* In normal circumstances 'm' will not be NULL. However,
5505 * if mirrors are reconfigured, we can temporarily get out
5506 * of sync in facet_revalidate(). We could "correct" the
5507 * mirror list before reaching here, but doing that would
5508 * not properly account the traffic stats we've currently
5509 * accumulated for previous mirror configuration. */
5513 m->packet_count += packets;
5514 m->byte_count += bytes;
5518 /* A VM broadcasts a gratuitous ARP to indicate that it has resumed after
5519 * migration. Older Citrix-patched Linux DomU used gratuitous ARP replies to
5520 * indicate this; newer upstream kernels use gratuitous ARP requests. */
5522 is_gratuitous_arp(const struct flow *flow)
5524 return (flow->dl_type == htons(ETH_TYPE_ARP)
5525 && eth_addr_is_broadcast(flow->dl_dst)
5526 && (flow->nw_proto == ARP_OP_REPLY
5527 || (flow->nw_proto == ARP_OP_REQUEST
5528 && flow->nw_src == flow->nw_dst)));
5532 update_learning_table(struct ofproto_dpif *ofproto,
5533 const struct flow *flow, int vlan,
5534 struct ofbundle *in_bundle)
5536 struct mac_entry *mac;
5538 /* Don't learn the OFPP_NONE port. */
5539 if (in_bundle == &ofpp_none_bundle) {
5543 if (!mac_learning_may_learn(ofproto->ml, flow->dl_src, vlan)) {
5547 mac = mac_learning_insert(ofproto->ml, flow->dl_src, vlan);
5548 if (is_gratuitous_arp(flow)) {
5549 /* We don't want to learn from gratuitous ARP packets that are
5550 * reflected back over bond slaves so we lock the learning table. */
5551 if (!in_bundle->bond) {
5552 mac_entry_set_grat_arp_lock(mac);
5553 } else if (mac_entry_is_grat_arp_locked(mac)) {
5558 if (mac_entry_is_new(mac) || mac->port.p != in_bundle) {
5559 /* The log messages here could actually be useful in debugging,
5560 * so keep the rate limit relatively high. */
5561 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(30, 300);
5562 VLOG_DBG_RL(&rl, "bridge %s: learned that "ETH_ADDR_FMT" is "
5563 "on port %s in VLAN %d",
5564 ofproto->up.name, ETH_ADDR_ARGS(flow->dl_src),
5565 in_bundle->name, vlan);
5567 mac->port.p = in_bundle;
5568 tag_set_add(&ofproto->revalidate_set,
5569 mac_learning_changed(ofproto->ml, mac));
5573 static struct ofbundle *
5574 lookup_input_bundle(struct ofproto_dpif *ofproto, uint16_t in_port, bool warn)
5576 struct ofport_dpif *ofport;
5578 /* Special-case OFPP_NONE, which a controller may use as the ingress
5579 * port for traffic that it is sourcing. */
5580 if (in_port == OFPP_NONE) {
5581 return &ofpp_none_bundle;
5584 /* Find the port and bundle for the received packet. */
5585 ofport = get_ofp_port(ofproto, in_port);
5586 if (ofport && ofport->bundle) {
5587 return ofport->bundle;
5590 /* Odd. A few possible reasons here:
5592 * - We deleted a port but there are still a few packets queued up
5595 * - Someone externally added a port (e.g. "ovs-dpctl add-if") that
5596 * we don't know about.
5598 * - The ofproto client didn't configure the port as part of a bundle.
5601 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
5603 VLOG_WARN_RL(&rl, "bridge %s: received packet on unknown "
5604 "port %"PRIu16, ofproto->up.name, in_port);
5609 /* Determines whether packets in 'flow' within 'ofproto' should be forwarded or
5610 * dropped. Returns true if they may be forwarded, false if they should be
5613 * 'in_port' must be the ofport_dpif that corresponds to flow->in_port.
5614 * 'in_port' must be part of a bundle (e.g. in_port->bundle must be nonnull).
5616 * 'vlan' must be the VLAN that corresponds to flow->vlan_tci on 'in_port', as
5617 * returned by input_vid_to_vlan(). It must be a valid VLAN for 'in_port', as
5618 * checked by input_vid_is_valid().
5620 * May also add tags to '*tags', although the current implementation only does
5621 * so in one special case.
5624 is_admissible(struct ofproto_dpif *ofproto, const struct flow *flow,
5625 struct ofport_dpif *in_port, uint16_t vlan, tag_type *tags)
5627 struct ofbundle *in_bundle = in_port->bundle;
5629 /* Drop frames for reserved multicast addresses
5630 * only if forward_bpdu option is absent. */
5631 if (eth_addr_is_reserved(flow->dl_dst) && !ofproto->up.forward_bpdu) {
5635 if (in_bundle->bond) {
5636 struct mac_entry *mac;
5638 switch (bond_check_admissibility(in_bundle->bond, in_port,
5639 flow->dl_dst, tags)) {
5646 case BV_DROP_IF_MOVED:
5647 mac = mac_learning_lookup(ofproto->ml, flow->dl_src, vlan, NULL);
5648 if (mac && mac->port.p != in_bundle &&
5649 (!is_gratuitous_arp(flow)
5650 || mac_entry_is_grat_arp_locked(mac))) {
5661 xlate_normal(struct action_xlate_ctx *ctx)
5663 struct ofport_dpif *in_port;
5664 struct ofbundle *in_bundle;
5665 struct mac_entry *mac;
5669 ctx->has_normal = true;
5671 in_bundle = lookup_input_bundle(ctx->ofproto, ctx->flow.in_port,
5672 ctx->packet != NULL);
5677 /* We know 'in_port' exists unless it is "ofpp_none_bundle",
5678 * since lookup_input_bundle() succeeded. */
5679 in_port = get_ofp_port(ctx->ofproto, ctx->flow.in_port);
5681 /* Drop malformed frames. */
5682 if (ctx->flow.dl_type == htons(ETH_TYPE_VLAN) &&
5683 !(ctx->flow.vlan_tci & htons(VLAN_CFI))) {
5684 if (ctx->packet != NULL) {
5685 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
5686 VLOG_WARN_RL(&rl, "bridge %s: dropping packet with partial "
5687 "VLAN tag received on port %s",
5688 ctx->ofproto->up.name, in_bundle->name);
5693 /* Drop frames on bundles reserved for mirroring. */
5694 if (in_bundle->mirror_out) {
5695 if (ctx->packet != NULL) {
5696 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
5697 VLOG_WARN_RL(&rl, "bridge %s: dropping packet received on port "
5698 "%s, which is reserved exclusively for mirroring",
5699 ctx->ofproto->up.name, in_bundle->name);
5705 vid = vlan_tci_to_vid(ctx->flow.vlan_tci);
5706 if (!input_vid_is_valid(vid, in_bundle, ctx->packet != NULL)) {
5709 vlan = input_vid_to_vlan(in_bundle, vid);
5711 /* Check other admissibility requirements. */
5713 !is_admissible(ctx->ofproto, &ctx->flow, in_port, vlan, &ctx->tags)) {
5717 /* Learn source MAC. */
5718 if (ctx->may_learn_macs) {
5719 update_learning_table(ctx->ofproto, &ctx->flow, vlan, in_bundle);
5722 /* Determine output bundle. */
5723 mac = mac_learning_lookup(ctx->ofproto->ml, ctx->flow.dl_dst, vlan,
5726 if (mac->port.p != in_bundle) {
5727 output_normal(ctx, mac->port.p, vlan);
5730 struct ofbundle *bundle;
5732 HMAP_FOR_EACH (bundle, hmap_node, &ctx->ofproto->bundles) {
5733 if (bundle != in_bundle
5734 && ofbundle_includes_vlan(bundle, vlan)
5735 && bundle->floodable
5736 && !bundle->mirror_out) {
5737 output_normal(ctx, bundle, vlan);
5740 ctx->nf_output_iface = NF_OUT_FLOOD;
5744 /* Optimized flow revalidation.
5746 * It's a difficult problem, in general, to tell which facets need to have
5747 * their actions recalculated whenever the OpenFlow flow table changes. We
5748 * don't try to solve that general problem: for most kinds of OpenFlow flow
5749 * table changes, we recalculate the actions for every facet. This is
5750 * relatively expensive, but it's good enough if the OpenFlow flow table
5751 * doesn't change very often.
5753 * However, we can expect one particular kind of OpenFlow flow table change to
5754 * happen frequently: changes caused by MAC learning. To avoid wasting a lot
5755 * of CPU on revalidating every facet whenever MAC learning modifies the flow
5756 * table, we add a special case that applies to flow tables in which every rule
5757 * has the same form (that is, the same wildcards), except that the table is
5758 * also allowed to have a single "catch-all" flow that matches all packets. We
5759 * optimize this case by tagging all of the facets that resubmit into the table
5760 * and invalidating the same tag whenever a flow changes in that table. The
5761 * end result is that we revalidate just the facets that need it (and sometimes
5762 * a few more, but not all of the facets or even all of the facets that
5763 * resubmit to the table modified by MAC learning). */
5765 /* Calculates the tag to use for 'flow' and wildcards 'wc' when it is inserted
5766 * into an OpenFlow table with the given 'basis'. */
5768 rule_calculate_tag(const struct flow *flow, const struct flow_wildcards *wc,
5771 if (flow_wildcards_is_catchall(wc)) {
5774 struct flow tag_flow = *flow;
5775 flow_zero_wildcards(&tag_flow, wc);
5776 return tag_create_deterministic(flow_hash(&tag_flow, secret));
5780 /* Following a change to OpenFlow table 'table_id' in 'ofproto', update the
5781 * taggability of that table.
5783 * This function must be called after *each* change to a flow table. If you
5784 * skip calling it on some changes then the pointer comparisons at the end can
5785 * be invalid if you get unlucky. For example, if a flow removal causes a
5786 * cls_table to be destroyed and then a flow insertion causes a cls_table with
5787 * different wildcards to be created with the same address, then this function
5788 * will incorrectly skip revalidation. */
5790 table_update_taggable(struct ofproto_dpif *ofproto, uint8_t table_id)
5792 struct table_dpif *table = &ofproto->tables[table_id];
5793 const struct oftable *oftable = &ofproto->up.tables[table_id];
5794 struct cls_table *catchall, *other;
5795 struct cls_table *t;
5797 catchall = other = NULL;
5799 switch (hmap_count(&oftable->cls.tables)) {
5801 /* We could tag this OpenFlow table but it would make the logic a
5802 * little harder and it's a corner case that doesn't seem worth it
5808 HMAP_FOR_EACH (t, hmap_node, &oftable->cls.tables) {
5809 if (cls_table_is_catchall(t)) {
5811 } else if (!other) {
5814 /* Indicate that we can't tag this by setting both tables to
5815 * NULL. (We know that 'catchall' is already NULL.) */
5822 /* Can't tag this table. */
5826 if (table->catchall_table != catchall || table->other_table != other) {
5827 table->catchall_table = catchall;
5828 table->other_table = other;
5829 ofproto->need_revalidate = true;
5833 /* Given 'rule' that has changed in some way (either it is a rule being
5834 * inserted, a rule being deleted, or a rule whose actions are being
5835 * modified), marks facets for revalidation to ensure that packets will be
5836 * forwarded correctly according to the new state of the flow table.
5838 * This function must be called after *each* change to a flow table. See
5839 * the comment on table_update_taggable() for more information. */
5841 rule_invalidate(const struct rule_dpif *rule)
5843 struct ofproto_dpif *ofproto = ofproto_dpif_cast(rule->up.ofproto);
5845 table_update_taggable(ofproto, rule->up.table_id);
5847 if (!ofproto->need_revalidate) {
5848 struct table_dpif *table = &ofproto->tables[rule->up.table_id];
5850 if (table->other_table && rule->tag) {
5851 tag_set_add(&ofproto->revalidate_set, rule->tag);
5853 ofproto->need_revalidate = true;
5859 set_frag_handling(struct ofproto *ofproto_,
5860 enum ofp_config_flags frag_handling)
5862 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
5864 if (frag_handling != OFPC_FRAG_REASM) {
5865 ofproto->need_revalidate = true;
5873 packet_out(struct ofproto *ofproto_, struct ofpbuf *packet,
5874 const struct flow *flow,
5875 const union ofp_action *ofp_actions, size_t n_ofp_actions)
5877 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
5880 if (flow->in_port >= ofproto->max_ports && flow->in_port < OFPP_MAX) {
5881 return OFPERR_NXBRC_BAD_IN_PORT;
5884 error = validate_actions(ofp_actions, n_ofp_actions, flow,
5885 ofproto->max_ports);
5887 struct odputil_keybuf keybuf;
5888 struct ofpbuf *odp_actions;
5889 struct ofproto_push push;
5892 ofpbuf_use_stack(&key, &keybuf, sizeof keybuf);
5893 odp_flow_key_from_flow(&key, flow);
5895 action_xlate_ctx_init(&push.ctx, ofproto, flow, flow->vlan_tci, NULL,
5896 packet_get_tcp_flags(packet, flow), packet);
5898 /* Ensure that resubmits in 'ofp_actions' get accounted to their
5899 * matching rules. */
5901 push.bytes = packet->size;
5902 push.used = time_msec();
5903 push.ctx.resubmit_hook = push_resubmit;
5905 odp_actions = xlate_actions(&push.ctx, ofp_actions, n_ofp_actions);
5906 dpif_execute(ofproto->dpif, key.data, key.size,
5907 odp_actions->data, odp_actions->size, packet);
5908 ofpbuf_delete(odp_actions);
5916 set_netflow(struct ofproto *ofproto_,
5917 const struct netflow_options *netflow_options)
5919 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
5921 if (netflow_options) {
5922 if (!ofproto->netflow) {
5923 ofproto->netflow = netflow_create();
5925 return netflow_set_options(ofproto->netflow, netflow_options);
5927 netflow_destroy(ofproto->netflow);
5928 ofproto->netflow = NULL;
5934 get_netflow_ids(const struct ofproto *ofproto_,
5935 uint8_t *engine_type, uint8_t *engine_id)
5937 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
5939 dpif_get_netflow_ids(ofproto->dpif, engine_type, engine_id);
5943 send_active_timeout(struct ofproto_dpif *ofproto, struct facet *facet)
5945 if (!facet_is_controller_flow(facet) &&
5946 netflow_active_timeout_expired(ofproto->netflow, &facet->nf_flow)) {
5947 struct subfacet *subfacet;
5948 struct ofexpired expired;
5950 LIST_FOR_EACH (subfacet, list_node, &facet->subfacets) {
5951 if (subfacet->installed) {
5952 struct dpif_flow_stats stats;
5954 subfacet_install(subfacet, subfacet->actions,
5955 subfacet->actions_len, &stats);
5956 subfacet_update_stats(subfacet, &stats);
5960 expired.flow = facet->flow;
5961 expired.packet_count = facet->packet_count;
5962 expired.byte_count = facet->byte_count;
5963 expired.used = facet->used;
5964 netflow_expire(ofproto->netflow, &facet->nf_flow, &expired);
5969 send_netflow_active_timeouts(struct ofproto_dpif *ofproto)
5971 struct facet *facet;
5973 HMAP_FOR_EACH (facet, hmap_node, &ofproto->facets) {
5974 send_active_timeout(ofproto, facet);
5978 static struct ofproto_dpif *
5979 ofproto_dpif_lookup(const char *name)
5981 struct ofproto_dpif *ofproto;
5983 HMAP_FOR_EACH_WITH_HASH (ofproto, all_ofproto_dpifs_node,
5984 hash_string(name, 0), &all_ofproto_dpifs) {
5985 if (!strcmp(ofproto->up.name, name)) {
5993 ofproto_unixctl_fdb_flush(struct unixctl_conn *conn, int argc,
5994 const char *argv[], void *aux OVS_UNUSED)
5996 struct ofproto_dpif *ofproto;
5999 ofproto = ofproto_dpif_lookup(argv[1]);
6001 unixctl_command_reply_error(conn, "no such bridge");
6004 mac_learning_flush(ofproto->ml, &ofproto->revalidate_set);
6006 HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
6007 mac_learning_flush(ofproto->ml, &ofproto->revalidate_set);
6011 unixctl_command_reply(conn, "table successfully flushed");
6015 ofproto_unixctl_fdb_show(struct unixctl_conn *conn, int argc OVS_UNUSED,
6016 const char *argv[], void *aux OVS_UNUSED)
6018 struct ds ds = DS_EMPTY_INITIALIZER;
6019 const struct ofproto_dpif *ofproto;
6020 const struct mac_entry *e;
6022 ofproto = ofproto_dpif_lookup(argv[1]);
6024 unixctl_command_reply_error(conn, "no such bridge");
6028 ds_put_cstr(&ds, " port VLAN MAC Age\n");
6029 LIST_FOR_EACH (e, lru_node, &ofproto->ml->lrus) {
6030 struct ofbundle *bundle = e->port.p;
6031 ds_put_format(&ds, "%5d %4d "ETH_ADDR_FMT" %3d\n",
6032 ofbundle_get_a_port(bundle)->odp_port,
6033 e->vlan, ETH_ADDR_ARGS(e->mac),
6034 mac_entry_age(ofproto->ml, e));
6036 unixctl_command_reply(conn, ds_cstr(&ds));
6041 struct action_xlate_ctx ctx;
6047 trace_format_rule(struct ds *result, uint8_t table_id, int level,
6048 const struct rule_dpif *rule)
6050 ds_put_char_multiple(result, '\t', level);
6052 ds_put_cstr(result, "No match\n");
6056 ds_put_format(result, "Rule: table=%"PRIu8" cookie=%#"PRIx64" ",
6057 table_id, ntohll(rule->up.flow_cookie));
6058 cls_rule_format(&rule->up.cr, result);
6059 ds_put_char(result, '\n');
6061 ds_put_char_multiple(result, '\t', level);
6062 ds_put_cstr(result, "OpenFlow ");
6063 ofp_print_actions(result, rule->up.actions, rule->up.n_actions);
6064 ds_put_char(result, '\n');
6068 trace_format_flow(struct ds *result, int level, const char *title,
6069 struct trace_ctx *trace)
6071 ds_put_char_multiple(result, '\t', level);
6072 ds_put_format(result, "%s: ", title);
6073 if (flow_equal(&trace->ctx.flow, &trace->flow)) {
6074 ds_put_cstr(result, "unchanged");
6076 flow_format(result, &trace->ctx.flow);
6077 trace->flow = trace->ctx.flow;
6079 ds_put_char(result, '\n');
6083 trace_format_regs(struct ds *result, int level, const char *title,
6084 struct trace_ctx *trace)
6088 ds_put_char_multiple(result, '\t', level);
6089 ds_put_format(result, "%s:", title);
6090 for (i = 0; i < FLOW_N_REGS; i++) {
6091 ds_put_format(result, " reg%zu=0x%"PRIx32, i, trace->flow.regs[i]);
6093 ds_put_char(result, '\n');
6097 trace_format_odp(struct ds *result, int level, const char *title,
6098 struct trace_ctx *trace)
6100 struct ofpbuf *odp_actions = trace->ctx.odp_actions;
6102 ds_put_char_multiple(result, '\t', level);
6103 ds_put_format(result, "%s: ", title);
6104 format_odp_actions(result, odp_actions->data, odp_actions->size);
6105 ds_put_char(result, '\n');
6109 trace_resubmit(struct action_xlate_ctx *ctx, struct rule_dpif *rule)
6111 struct trace_ctx *trace = CONTAINER_OF(ctx, struct trace_ctx, ctx);
6112 struct ds *result = trace->result;
6114 ds_put_char(result, '\n');
6115 trace_format_flow(result, ctx->recurse + 1, "Resubmitted flow", trace);
6116 trace_format_regs(result, ctx->recurse + 1, "Resubmitted regs", trace);
6117 trace_format_odp(result, ctx->recurse + 1, "Resubmitted odp", trace);
6118 trace_format_rule(result, ctx->table_id, ctx->recurse + 1, rule);
6122 ofproto_unixctl_trace(struct unixctl_conn *conn, int argc, const char *argv[],
6123 void *aux OVS_UNUSED)
6125 const char *dpname = argv[1];
6126 struct ofproto_dpif *ofproto;
6127 struct ofpbuf odp_key;
6128 struct ofpbuf *packet;
6129 ovs_be16 initial_tci;
6135 ofpbuf_init(&odp_key, 0);
6138 ofproto = ofproto_dpif_lookup(dpname);
6140 unixctl_command_reply_error(conn, "Unknown ofproto (use ofproto/list "
6144 if (argc == 3 || (argc == 4 && !strcmp(argv[3], "-generate"))) {
6145 /* ofproto/trace dpname flow [-generate] */
6146 const char *flow_s = argv[2];
6147 const char *generate_s = argv[3];
6150 /* Convert string to datapath key. */
6151 ofpbuf_init(&odp_key, 0);
6152 error = odp_flow_key_from_string(flow_s, NULL, &odp_key);
6154 unixctl_command_reply_error(conn, "Bad flow syntax");
6158 /* Convert odp_key to flow. */
6159 error = ofproto_dpif_extract_flow_key(ofproto, odp_key.data,
6160 odp_key.size, &flow,
6161 &initial_tci, NULL);
6162 if (error == ODP_FIT_ERROR) {
6163 unixctl_command_reply_error(conn, "Invalid flow");
6167 /* Generate a packet, if requested. */
6169 packet = ofpbuf_new(0);
6170 flow_compose(packet, &flow);
6172 } else if (argc == 6) {
6173 /* ofproto/trace dpname priority tun_id in_port packet */
6174 const char *priority_s = argv[2];
6175 const char *tun_id_s = argv[3];
6176 const char *in_port_s = argv[4];
6177 const char *packet_s = argv[5];
6178 uint16_t in_port = ofp_port_to_odp_port(atoi(in_port_s));
6179 ovs_be64 tun_id = htonll(strtoull(tun_id_s, NULL, 0));
6180 uint32_t priority = atoi(priority_s);
6183 msg = eth_from_hex(packet_s, &packet);
6185 unixctl_command_reply_error(conn, msg);
6189 ds_put_cstr(&result, "Packet: ");
6190 s = ofp_packet_to_string(packet->data, packet->size);
6191 ds_put_cstr(&result, s);
6194 flow_extract(packet, priority, tun_id, in_port, &flow);
6195 initial_tci = flow.vlan_tci;
6197 unixctl_command_reply_error(conn, "Bad command syntax");
6201 ofproto_trace(ofproto, &flow, packet, initial_tci, &result);
6202 unixctl_command_reply(conn, ds_cstr(&result));
6205 ds_destroy(&result);
6206 ofpbuf_delete(packet);
6207 ofpbuf_uninit(&odp_key);
6211 ofproto_trace(struct ofproto_dpif *ofproto, const struct flow *flow,
6212 const struct ofpbuf *packet, ovs_be16 initial_tci,
6215 struct rule_dpif *rule;
6217 ds_put_cstr(ds, "Flow: ");
6218 flow_format(ds, flow);
6219 ds_put_char(ds, '\n');
6221 rule = rule_dpif_lookup(ofproto, flow, 0);
6222 trace_format_rule(ds, 0, 0, rule);
6224 struct trace_ctx trace;
6225 struct ofpbuf *odp_actions;
6228 tcp_flags = packet ? packet_get_tcp_flags(packet, flow) : 0;
6231 action_xlate_ctx_init(&trace.ctx, ofproto, flow, initial_tci,
6232 rule, tcp_flags, packet);
6233 trace.ctx.resubmit_hook = trace_resubmit;
6234 odp_actions = xlate_actions(&trace.ctx,
6235 rule->up.actions, rule->up.n_actions);
6237 ds_put_char(ds, '\n');
6238 trace_format_flow(ds, 0, "Final flow", &trace);
6239 ds_put_cstr(ds, "Datapath actions: ");
6240 format_odp_actions(ds, odp_actions->data, odp_actions->size);
6241 ofpbuf_delete(odp_actions);
6243 if (!trace.ctx.may_set_up_flow) {
6245 ds_put_cstr(ds, "\nThis flow is not cachable.");
6247 ds_put_cstr(ds, "\nThe datapath actions are incomplete--"
6248 "for complete actions, please supply a packet.");
6255 ofproto_dpif_clog(struct unixctl_conn *conn OVS_UNUSED, int argc OVS_UNUSED,
6256 const char *argv[] OVS_UNUSED, void *aux OVS_UNUSED)
6259 unixctl_command_reply(conn, NULL);
6263 ofproto_dpif_unclog(struct unixctl_conn *conn OVS_UNUSED, int argc OVS_UNUSED,
6264 const char *argv[] OVS_UNUSED, void *aux OVS_UNUSED)
6267 unixctl_command_reply(conn, NULL);
6270 /* Runs a self-check of flow translations in 'ofproto'. Appends a message to
6271 * 'reply' describing the results. */
6273 ofproto_dpif_self_check__(struct ofproto_dpif *ofproto, struct ds *reply)
6275 struct facet *facet;
6279 HMAP_FOR_EACH (facet, hmap_node, &ofproto->facets) {
6280 if (!facet_check_consistency(facet)) {
6285 ofproto->need_revalidate = true;
6289 ds_put_format(reply, "%s: self-check failed (%d errors)\n",
6290 ofproto->up.name, errors);
6292 ds_put_format(reply, "%s: self-check passed\n", ofproto->up.name);
6297 ofproto_dpif_self_check(struct unixctl_conn *conn,
6298 int argc, const char *argv[], void *aux OVS_UNUSED)
6300 struct ds reply = DS_EMPTY_INITIALIZER;
6301 struct ofproto_dpif *ofproto;
6304 ofproto = ofproto_dpif_lookup(argv[1]);
6306 unixctl_command_reply_error(conn, "Unknown ofproto (use "
6307 "ofproto/list for help)");
6310 ofproto_dpif_self_check__(ofproto, &reply);
6312 HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
6313 ofproto_dpif_self_check__(ofproto, &reply);
6317 unixctl_command_reply(conn, ds_cstr(&reply));
6322 ofproto_dpif_unixctl_init(void)
6324 static bool registered;
6330 unixctl_command_register(
6332 "bridge {tun_id in_port packet | odp_flow [-generate]}",
6333 2, 5, ofproto_unixctl_trace, NULL);
6334 unixctl_command_register("fdb/flush", "[bridge]", 0, 1,
6335 ofproto_unixctl_fdb_flush, NULL);
6336 unixctl_command_register("fdb/show", "bridge", 1, 1,
6337 ofproto_unixctl_fdb_show, NULL);
6338 unixctl_command_register("ofproto/clog", "", 0, 0,
6339 ofproto_dpif_clog, NULL);
6340 unixctl_command_register("ofproto/unclog", "", 0, 0,
6341 ofproto_dpif_unclog, NULL);
6342 unixctl_command_register("ofproto/self-check", "[bridge]", 0, 1,
6343 ofproto_dpif_self_check, NULL);
6346 /* Linux VLAN device support (e.g. "eth0.10" for VLAN 10.)
6348 * This is deprecated. It is only for compatibility with broken device drivers
6349 * in old versions of Linux that do not properly support VLANs when VLAN
6350 * devices are not used. When broken device drivers are no longer in
6351 * widespread use, we will delete these interfaces. */
6354 set_realdev(struct ofport *ofport_, uint16_t realdev_ofp_port, int vid)
6356 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport_->ofproto);
6357 struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
6359 if (realdev_ofp_port == ofport->realdev_ofp_port
6360 && vid == ofport->vlandev_vid) {
6364 ofproto->need_revalidate = true;
6366 if (ofport->realdev_ofp_port) {
6369 if (realdev_ofp_port && ofport->bundle) {
6370 /* vlandevs are enslaved to their realdevs, so they are not allowed to
6371 * themselves be part of a bundle. */
6372 bundle_set(ofport->up.ofproto, ofport->bundle, NULL);
6375 ofport->realdev_ofp_port = realdev_ofp_port;
6376 ofport->vlandev_vid = vid;
6378 if (realdev_ofp_port) {
6379 vsp_add(ofport, realdev_ofp_port, vid);
6386 hash_realdev_vid(uint16_t realdev_ofp_port, int vid)
6388 return hash_2words(realdev_ofp_port, vid);
6391 /* Returns the ODP port number of the Linux VLAN device that corresponds to
6392 * 'vlan_tci' on the network device with port number 'realdev_odp_port' in
6393 * 'ofproto'. For example, given 'realdev_odp_port' of eth0 and 'vlan_tci' 9,
6394 * it would return the port number of eth0.9.
6396 * Unless VLAN splinters are enabled for port 'realdev_odp_port', this
6397 * function just returns its 'realdev_odp_port' argument. */
6399 vsp_realdev_to_vlandev(const struct ofproto_dpif *ofproto,
6400 uint32_t realdev_odp_port, ovs_be16 vlan_tci)
6402 if (!hmap_is_empty(&ofproto->realdev_vid_map)) {
6403 uint16_t realdev_ofp_port = odp_port_to_ofp_port(realdev_odp_port);
6404 int vid = vlan_tci_to_vid(vlan_tci);
6405 const struct vlan_splinter *vsp;
6407 HMAP_FOR_EACH_WITH_HASH (vsp, realdev_vid_node,
6408 hash_realdev_vid(realdev_ofp_port, vid),
6409 &ofproto->realdev_vid_map) {
6410 if (vsp->realdev_ofp_port == realdev_ofp_port
6411 && vsp->vid == vid) {
6412 return ofp_port_to_odp_port(vsp->vlandev_ofp_port);
6416 return realdev_odp_port;
6419 static struct vlan_splinter *
6420 vlandev_find(const struct ofproto_dpif *ofproto, uint16_t vlandev_ofp_port)
6422 struct vlan_splinter *vsp;
6424 HMAP_FOR_EACH_WITH_HASH (vsp, vlandev_node, hash_int(vlandev_ofp_port, 0),
6425 &ofproto->vlandev_map) {
6426 if (vsp->vlandev_ofp_port == vlandev_ofp_port) {
6434 /* Returns the OpenFlow port number of the "real" device underlying the Linux
6435 * VLAN device with OpenFlow port number 'vlandev_ofp_port' and stores the
6436 * VLAN VID of the Linux VLAN device in '*vid'. For example, given
6437 * 'vlandev_ofp_port' of eth0.9, it would return the OpenFlow port number of
6438 * eth0 and store 9 in '*vid'.
6440 * Returns 0 and does not modify '*vid' if 'vlandev_ofp_port' is not a Linux
6441 * VLAN device. Unless VLAN splinters are enabled, this is what this function
6444 vsp_vlandev_to_realdev(const struct ofproto_dpif *ofproto,
6445 uint16_t vlandev_ofp_port, int *vid)
6447 if (!hmap_is_empty(&ofproto->vlandev_map)) {
6448 const struct vlan_splinter *vsp;
6450 vsp = vlandev_find(ofproto, vlandev_ofp_port);
6455 return vsp->realdev_ofp_port;
6462 vsp_remove(struct ofport_dpif *port)
6464 struct ofproto_dpif *ofproto = ofproto_dpif_cast(port->up.ofproto);
6465 struct vlan_splinter *vsp;
6467 vsp = vlandev_find(ofproto, port->up.ofp_port);
6469 hmap_remove(&ofproto->vlandev_map, &vsp->vlandev_node);
6470 hmap_remove(&ofproto->realdev_vid_map, &vsp->realdev_vid_node);
6473 port->realdev_ofp_port = 0;
6475 VLOG_ERR("missing vlan device record");
6480 vsp_add(struct ofport_dpif *port, uint16_t realdev_ofp_port, int vid)
6482 struct ofproto_dpif *ofproto = ofproto_dpif_cast(port->up.ofproto);
6484 if (!vsp_vlandev_to_realdev(ofproto, port->up.ofp_port, NULL)
6485 && (vsp_realdev_to_vlandev(ofproto, realdev_ofp_port, htons(vid))
6486 == realdev_ofp_port)) {
6487 struct vlan_splinter *vsp;
6489 vsp = xmalloc(sizeof *vsp);
6490 hmap_insert(&ofproto->vlandev_map, &vsp->vlandev_node,
6491 hash_int(port->up.ofp_port, 0));
6492 hmap_insert(&ofproto->realdev_vid_map, &vsp->realdev_vid_node,
6493 hash_realdev_vid(realdev_ofp_port, vid));
6494 vsp->realdev_ofp_port = realdev_ofp_port;
6495 vsp->vlandev_ofp_port = port->up.ofp_port;
6498 port->realdev_ofp_port = realdev_ofp_port;
6500 VLOG_ERR("duplicate vlan device record");
6504 const struct ofproto_class ofproto_dpif_class = {
6533 port_is_lacp_current,
6534 NULL, /* rule_choose_table */
6541 rule_modify_actions,
6549 get_cfm_remote_mpids,
6554 get_stp_port_status,
6561 is_mirror_output_bundle,
6562 forward_bpdu_changed,