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 rule_credit_stats(struct rule_dpif *,
109 const struct dpif_flow_stats *);
110 static void flow_push_stats(struct rule_dpif *, const struct flow *,
111 const struct dpif_flow_stats *);
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? Should "learn"
213 * actions update the flow table?
215 * We want to update these tables if we are actually processing a packet,
216 * or if we are accounting for packets that the datapath has processed, but
217 * not if we are just revalidating. */
220 /* The rule that we are currently translating, or NULL. */
221 struct rule_dpif *rule;
223 /* Union of the set of TCP flags seen so far in this flow. (Used only by
224 * NXAST_FIN_TIMEOUT. Set to zero to avoid updating updating rules'
228 /* If nonnull, flow translation calls this function just before executing a
229 * resubmit or OFPP_TABLE action. In addition, disables logging of traces
230 * when the recursion depth is exceeded.
232 * 'rule' is the rule being submitted into. It will be null if the
233 * resubmit or OFPP_TABLE action didn't find a matching rule.
235 * This is normally null so the client has to set it manually after
236 * calling action_xlate_ctx_init(). */
237 void (*resubmit_hook)(struct action_xlate_ctx *, struct rule_dpif *rule);
239 /* If nonnull, flow translation credits the specified statistics to each
240 * rule reached through a resubmit or OFPP_TABLE action.
242 * This is normally null so the client has to set it manually after
243 * calling action_xlate_ctx_init(). */
244 const struct dpif_flow_stats *resubmit_stats;
246 /* xlate_actions() initializes and uses these members. The client might want
247 * to look at them after it returns. */
249 struct ofpbuf *odp_actions; /* Datapath actions. */
250 tag_type tags; /* Tags associated with actions. */
251 bool may_set_up_flow; /* True ordinarily; false if the actions must
252 * be reassessed for every packet. */
253 bool has_learn; /* Actions include NXAST_LEARN? */
254 bool has_normal; /* Actions output to OFPP_NORMAL? */
255 bool has_fin_timeout; /* Actions include NXAST_FIN_TIMEOUT? */
256 uint16_t nf_output_iface; /* Output interface index for NetFlow. */
257 mirror_mask_t mirrors; /* Bitmap of associated mirrors. */
259 /* xlate_actions() initializes and uses these members, but the client has no
260 * reason to look at them. */
262 int recurse; /* Recursion level, via xlate_table_action. */
263 bool max_resubmit_trigger; /* Recursed too deeply during translation. */
264 struct flow base_flow; /* Flow at the last commit. */
265 uint32_t orig_skb_priority; /* Priority when packet arrived. */
266 uint8_t table_id; /* OpenFlow table ID where flow was found. */
267 uint32_t sflow_n_outputs; /* Number of output ports. */
268 uint16_t sflow_odp_port; /* Output port for composing sFlow action. */
269 uint16_t user_cookie_offset;/* Used for user_action_cookie fixup. */
270 bool exit; /* No further actions should be processed. */
273 static void action_xlate_ctx_init(struct action_xlate_ctx *,
274 struct ofproto_dpif *, const struct flow *,
275 ovs_be16 initial_tci, struct rule_dpif *,
276 uint8_t tcp_flags, const struct ofpbuf *);
277 static void xlate_actions(struct action_xlate_ctx *,
278 const union ofp_action *in, size_t n_in,
279 struct ofpbuf *odp_actions);
280 static void xlate_actions_for_side_effects(struct action_xlate_ctx *,
281 const union ofp_action *in,
284 /* A dpif flow and actions associated with a facet.
286 * See also the large comment on struct facet. */
289 struct hmap_node hmap_node; /* In struct ofproto_dpif 'subfacets' list. */
290 struct list list_node; /* In struct facet's 'facets' list. */
291 struct facet *facet; /* Owning facet. */
295 * To save memory in the common case, 'key' is NULL if 'key_fitness' is
296 * ODP_FIT_PERFECT, that is, odp_flow_key_from_flow() can accurately
297 * regenerate the ODP flow key from ->facet->flow. */
298 enum odp_key_fitness key_fitness;
302 long long int used; /* Time last used; time created if not used. */
304 uint64_t dp_packet_count; /* Last known packet count in the datapath. */
305 uint64_t dp_byte_count; /* Last known byte count in the datapath. */
309 * These should be essentially identical for every subfacet in a facet, but
310 * may differ in trivial ways due to VLAN splinters. */
311 size_t actions_len; /* Number of bytes in actions[]. */
312 struct nlattr *actions; /* Datapath actions. */
314 bool installed; /* Installed in datapath? */
316 /* This value is normally the same as ->facet->flow.vlan_tci. Only VLAN
317 * splinters can cause it to differ. This value should be removed when
318 * the VLAN splinters feature is no longer needed. */
319 ovs_be16 initial_tci; /* Initial VLAN TCI value. */
322 static struct subfacet *subfacet_create(struct facet *, enum odp_key_fitness,
323 const struct nlattr *key,
324 size_t key_len, ovs_be16 initial_tci);
325 static struct subfacet *subfacet_find(struct ofproto_dpif *,
326 const struct nlattr *key, size_t key_len);
327 static void subfacet_destroy(struct subfacet *);
328 static void subfacet_destroy__(struct subfacet *);
329 static void subfacet_get_key(struct subfacet *, struct odputil_keybuf *,
331 static void subfacet_reset_dp_stats(struct subfacet *,
332 struct dpif_flow_stats *);
333 static void subfacet_update_time(struct subfacet *, long long int used);
334 static void subfacet_update_stats(struct subfacet *,
335 const struct dpif_flow_stats *);
336 static void subfacet_make_actions(struct subfacet *,
337 const struct ofpbuf *packet);
338 static int subfacet_install(struct subfacet *,
339 const struct nlattr *actions, size_t actions_len,
340 struct dpif_flow_stats *);
341 static void subfacet_uninstall(struct subfacet *);
343 /* An exact-match instantiation of an OpenFlow flow.
345 * A facet associates a "struct flow", which represents the Open vSwitch
346 * userspace idea of an exact-match flow, with one or more subfacets. Each
347 * subfacet tracks the datapath's idea of the exact-match flow equivalent to
348 * the facet. When the kernel module (or other dpif implementation) and Open
349 * vSwitch userspace agree on the definition of a flow key, there is exactly
350 * one subfacet per facet. If the dpif implementation supports more-specific
351 * flow matching than userspace, however, a facet can have more than one
352 * subfacet, each of which corresponds to some distinction in flow that
353 * userspace simply doesn't understand.
355 * Flow expiration works in terms of subfacets, so a facet must have at least
356 * one subfacet or it will never expire, leaking memory. */
359 struct hmap_node hmap_node; /* In owning ofproto's 'facets' hmap. */
360 struct list list_node; /* In owning rule's 'facets' list. */
361 struct rule_dpif *rule; /* Owning rule. */
364 struct list subfacets;
365 long long int used; /* Time last used; time created if not used. */
372 * - Do include packets and bytes sent "by hand", e.g. with
375 * - Do include packets and bytes that were obtained from the datapath
376 * when a subfacet's statistics were reset (e.g. dpif_flow_put() with
377 * DPIF_FP_ZERO_STATS).
379 * - Do not include packets or bytes that can be obtained from the
380 * datapath for any existing subfacet.
382 uint64_t packet_count; /* Number of packets received. */
383 uint64_t byte_count; /* Number of bytes received. */
385 /* Resubmit statistics. */
386 uint64_t prev_packet_count; /* Number of packets from last stats push. */
387 uint64_t prev_byte_count; /* Number of bytes from last stats push. */
388 long long int prev_used; /* Used time from last stats push. */
391 uint64_t accounted_bytes; /* Bytes processed by facet_account(). */
392 struct netflow_flow nf_flow; /* Per-flow NetFlow tracking data. */
393 uint8_t tcp_flags; /* TCP flags seen for this 'rule'. */
395 /* Properties of datapath actions.
397 * Every subfacet has its own actions because actions can differ slightly
398 * between splintered and non-splintered subfacets due to the VLAN tag
399 * being initially different (present vs. absent). All of them have these
400 * properties in common so we just store one copy of them here. */
401 bool may_install; /* Reassess actions for every packet? */
402 bool has_learn; /* Actions include NXAST_LEARN? */
403 bool has_normal; /* Actions output to OFPP_NORMAL? */
404 bool has_fin_timeout; /* Actions include NXAST_FIN_TIMEOUT? */
405 tag_type tags; /* Tags that would require revalidation. */
406 mirror_mask_t mirrors; /* Bitmap of dependent mirrors. */
408 /* Storage for a single subfacet, to reduce malloc() time and space
409 * overhead. (A facet always has at least one subfacet and in the common
410 * case has exactly one subfacet.) */
411 struct subfacet one_subfacet;
414 static struct facet *facet_create(struct rule_dpif *,
415 const struct flow *, uint32_t hash);
416 static void facet_remove(struct facet *);
417 static void facet_free(struct facet *);
419 static struct facet *facet_find(struct ofproto_dpif *,
420 const struct flow *, uint32_t hash);
421 static struct facet *facet_lookup_valid(struct ofproto_dpif *,
422 const struct flow *, uint32_t hash);
423 static bool facet_revalidate(struct facet *);
424 static bool facet_check_consistency(struct facet *);
426 static void facet_flush_stats(struct facet *);
428 static void facet_update_time(struct facet *, long long int used);
429 static void facet_reset_counters(struct facet *);
430 static void facet_push_stats(struct facet *);
431 static void facet_learn(struct facet *);
432 static void facet_account(struct facet *);
434 static bool facet_is_controller_flow(struct facet *);
440 struct ofbundle *bundle; /* Bundle that contains this port, if any. */
441 struct list bundle_node; /* In struct ofbundle's "ports" list. */
442 struct cfm *cfm; /* Connectivity Fault Management, if any. */
443 tag_type tag; /* Tag associated with this port. */
444 uint32_t bond_stable_id; /* stable_id to use as bond slave, or 0. */
445 bool may_enable; /* May be enabled in bonds. */
446 long long int carrier_seq; /* Carrier status changes. */
449 struct stp_port *stp_port; /* Spanning Tree Protocol, if any. */
450 enum stp_state stp_state; /* Always STP_DISABLED if STP not in use. */
451 long long int stp_state_entered;
453 struct hmap priorities; /* Map of attached 'priority_to_dscp's. */
455 /* Linux VLAN device support (e.g. "eth0.10" for VLAN 10.)
457 * This is deprecated. It is only for compatibility with broken device
458 * drivers in old versions of Linux that do not properly support VLANs when
459 * VLAN devices are not used. When broken device drivers are no longer in
460 * widespread use, we will delete these interfaces. */
461 uint16_t realdev_ofp_port;
465 /* Node in 'ofport_dpif''s 'priorities' map. Used to maintain a map from
466 * 'priority' (the datapath's term for QoS queue) to the dscp bits which all
467 * traffic egressing the 'ofport' with that priority should be marked with. */
468 struct priority_to_dscp {
469 struct hmap_node hmap_node; /* Node in 'ofport_dpif''s 'priorities' map. */
470 uint32_t priority; /* Priority of this queue (see struct flow). */
472 uint8_t dscp; /* DSCP bits to mark outgoing traffic with. */
475 /* Linux VLAN device support (e.g. "eth0.10" for VLAN 10.)
477 * This is deprecated. It is only for compatibility with broken device drivers
478 * in old versions of Linux that do not properly support VLANs when VLAN
479 * devices are not used. When broken device drivers are no longer in
480 * widespread use, we will delete these interfaces. */
481 struct vlan_splinter {
482 struct hmap_node realdev_vid_node;
483 struct hmap_node vlandev_node;
484 uint16_t realdev_ofp_port;
485 uint16_t vlandev_ofp_port;
489 static uint32_t vsp_realdev_to_vlandev(const struct ofproto_dpif *,
490 uint32_t realdev, ovs_be16 vlan_tci);
491 static uint16_t vsp_vlandev_to_realdev(const struct ofproto_dpif *,
492 uint16_t vlandev, int *vid);
493 static void vsp_remove(struct ofport_dpif *);
494 static void vsp_add(struct ofport_dpif *, uint16_t realdev_ofp_port, int vid);
496 static struct ofport_dpif *
497 ofport_dpif_cast(const struct ofport *ofport)
499 assert(ofport->ofproto->ofproto_class == &ofproto_dpif_class);
500 return ofport ? CONTAINER_OF(ofport, struct ofport_dpif, up) : NULL;
503 static void port_run(struct ofport_dpif *);
504 static void port_wait(struct ofport_dpif *);
505 static int set_cfm(struct ofport *, const struct cfm_settings *);
506 static void ofport_clear_priorities(struct ofport_dpif *);
508 struct dpif_completion {
509 struct list list_node;
510 struct ofoperation *op;
513 /* Extra information about a classifier table.
514 * Currently used just for optimized flow revalidation. */
516 /* If either of these is nonnull, then this table has a form that allows
517 * flows to be tagged to avoid revalidating most flows for the most common
518 * kinds of flow table changes. */
519 struct cls_table *catchall_table; /* Table that wildcards all fields. */
520 struct cls_table *other_table; /* Table with any other wildcard set. */
521 uint32_t basis; /* Keeps each table's tags separate. */
524 struct ofproto_dpif {
525 struct hmap_node all_ofproto_dpifs_node; /* In 'all_ofproto_dpifs'. */
534 struct netflow *netflow;
535 struct dpif_sflow *sflow;
536 struct hmap bundles; /* Contains "struct ofbundle"s. */
537 struct mac_learning *ml;
538 struct ofmirror *mirrors[MAX_MIRRORS];
539 bool has_bonded_bundles;
542 struct timer next_expiration;
546 struct hmap subfacets;
549 struct table_dpif tables[N_TABLES];
550 bool need_revalidate;
551 struct tag_set revalidate_set;
553 /* Support for debugging async flow mods. */
554 struct list completions;
556 bool has_bundle_action; /* True when the first bundle action appears. */
557 struct netdev_stats stats; /* To account packets generated and consumed in
562 long long int stp_last_tick;
564 /* VLAN splinters. */
565 struct hmap realdev_vid_map; /* (realdev,vid) -> vlandev. */
566 struct hmap vlandev_map; /* vlandev -> (realdev,vid). */
569 /* Defer flow mod completion until "ovs-appctl ofproto/unclog"? (Useful only
570 * for debugging the asynchronous flow_mod implementation.) */
573 /* All existing ofproto_dpif instances, indexed by ->up.name. */
574 static struct hmap all_ofproto_dpifs = HMAP_INITIALIZER(&all_ofproto_dpifs);
576 static void ofproto_dpif_unixctl_init(void);
578 static struct ofproto_dpif *
579 ofproto_dpif_cast(const struct ofproto *ofproto)
581 assert(ofproto->ofproto_class == &ofproto_dpif_class);
582 return CONTAINER_OF(ofproto, struct ofproto_dpif, up);
585 static struct ofport_dpif *get_ofp_port(struct ofproto_dpif *,
587 static struct ofport_dpif *get_odp_port(struct ofproto_dpif *,
589 static void ofproto_trace(struct ofproto_dpif *, const struct flow *,
590 const struct ofpbuf *, ovs_be16 initial_tci,
593 /* Packet processing. */
594 static void update_learning_table(struct ofproto_dpif *,
595 const struct flow *, int vlan,
598 #define FLOW_MISS_MAX_BATCH 50
599 static int handle_upcalls(struct ofproto_dpif *, unsigned int max_batch);
601 /* Flow expiration. */
602 static int expire(struct ofproto_dpif *);
605 static void send_netflow_active_timeouts(struct ofproto_dpif *);
608 static int send_packet(const struct ofport_dpif *, struct ofpbuf *packet);
610 compose_sflow_action(const struct ofproto_dpif *, struct ofpbuf *odp_actions,
611 const struct flow *, uint32_t odp_port);
612 static void add_mirror_actions(struct action_xlate_ctx *ctx,
613 const struct flow *flow);
614 /* Global variables. */
615 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
617 /* Factory functions. */
620 enumerate_types(struct sset *types)
622 dp_enumerate_types(types);
626 enumerate_names(const char *type, struct sset *names)
628 return dp_enumerate_names(type, names);
632 del(const char *type, const char *name)
637 error = dpif_open(name, type, &dpif);
639 error = dpif_delete(dpif);
645 /* Basic life-cycle. */
647 static struct ofproto *
650 struct ofproto_dpif *ofproto = xmalloc(sizeof *ofproto);
655 dealloc(struct ofproto *ofproto_)
657 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
662 construct(struct ofproto *ofproto_)
664 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
665 const char *name = ofproto->up.name;
669 error = dpif_create_and_open(name, ofproto->up.type, &ofproto->dpif);
671 VLOG_ERR("failed to open datapath %s: %s", name, strerror(error));
675 ofproto->max_ports = dpif_get_max_ports(ofproto->dpif);
676 ofproto->n_matches = 0;
678 dpif_flow_flush(ofproto->dpif);
679 dpif_recv_purge(ofproto->dpif);
681 error = dpif_recv_set(ofproto->dpif, true);
683 VLOG_ERR("failed to listen on datapath %s: %s", name, strerror(error));
684 dpif_close(ofproto->dpif);
688 ofproto->netflow = NULL;
689 ofproto->sflow = NULL;
691 hmap_init(&ofproto->bundles);
692 ofproto->ml = mac_learning_create(MAC_ENTRY_DEFAULT_IDLE_TIME);
693 for (i = 0; i < MAX_MIRRORS; i++) {
694 ofproto->mirrors[i] = NULL;
696 ofproto->has_bonded_bundles = false;
698 timer_set_duration(&ofproto->next_expiration, 1000);
700 hmap_init(&ofproto->facets);
701 hmap_init(&ofproto->subfacets);
703 for (i = 0; i < N_TABLES; i++) {
704 struct table_dpif *table = &ofproto->tables[i];
706 table->catchall_table = NULL;
707 table->other_table = NULL;
708 table->basis = random_uint32();
710 ofproto->need_revalidate = false;
711 tag_set_init(&ofproto->revalidate_set);
713 list_init(&ofproto->completions);
715 ofproto_dpif_unixctl_init();
717 ofproto->has_bundle_action = false;
719 hmap_init(&ofproto->vlandev_map);
720 hmap_init(&ofproto->realdev_vid_map);
722 hmap_insert(&all_ofproto_dpifs, &ofproto->all_ofproto_dpifs_node,
723 hash_string(ofproto->up.name, 0));
724 memset(&ofproto->stats, 0, sizeof ofproto->stats);
726 ofproto_init_tables(ofproto_, N_TABLES);
732 complete_operations(struct ofproto_dpif *ofproto)
734 struct dpif_completion *c, *next;
736 LIST_FOR_EACH_SAFE (c, next, list_node, &ofproto->completions) {
737 ofoperation_complete(c->op, 0);
738 list_remove(&c->list_node);
744 destruct(struct ofproto *ofproto_)
746 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
747 struct rule_dpif *rule, *next_rule;
748 struct oftable *table;
751 hmap_remove(&all_ofproto_dpifs, &ofproto->all_ofproto_dpifs_node);
752 complete_operations(ofproto);
754 OFPROTO_FOR_EACH_TABLE (table, &ofproto->up) {
755 struct cls_cursor cursor;
757 cls_cursor_init(&cursor, &table->cls, NULL);
758 CLS_CURSOR_FOR_EACH_SAFE (rule, next_rule, up.cr, &cursor) {
759 ofproto_rule_destroy(&rule->up);
763 for (i = 0; i < MAX_MIRRORS; i++) {
764 mirror_destroy(ofproto->mirrors[i]);
767 netflow_destroy(ofproto->netflow);
768 dpif_sflow_destroy(ofproto->sflow);
769 hmap_destroy(&ofproto->bundles);
770 mac_learning_destroy(ofproto->ml);
772 hmap_destroy(&ofproto->facets);
773 hmap_destroy(&ofproto->subfacets);
775 hmap_destroy(&ofproto->vlandev_map);
776 hmap_destroy(&ofproto->realdev_vid_map);
778 dpif_close(ofproto->dpif);
782 run_fast(struct ofproto *ofproto_)
784 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
787 /* Handle one or more batches of upcalls, until there's nothing left to do
788 * or until we do a fixed total amount of work.
790 * We do work in batches because it can be much cheaper to set up a number
791 * of flows and fire off their patches all at once. We do multiple batches
792 * because in some cases handling a packet can cause another packet to be
793 * queued almost immediately as part of the return flow. Both
794 * optimizations can make major improvements on some benchmarks and
795 * presumably for real traffic as well. */
797 while (work < FLOW_MISS_MAX_BATCH) {
798 int retval = handle_upcalls(ofproto, FLOW_MISS_MAX_BATCH - work);
808 run(struct ofproto *ofproto_)
810 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
811 struct ofport_dpif *ofport;
812 struct ofbundle *bundle;
816 complete_operations(ofproto);
818 dpif_run(ofproto->dpif);
820 error = run_fast(ofproto_);
825 if (timer_expired(&ofproto->next_expiration)) {
826 int delay = expire(ofproto);
827 timer_set_duration(&ofproto->next_expiration, delay);
830 if (ofproto->netflow) {
831 if (netflow_run(ofproto->netflow)) {
832 send_netflow_active_timeouts(ofproto);
835 if (ofproto->sflow) {
836 dpif_sflow_run(ofproto->sflow);
839 HMAP_FOR_EACH (ofport, up.hmap_node, &ofproto->up.ports) {
842 HMAP_FOR_EACH (bundle, hmap_node, &ofproto->bundles) {
847 mac_learning_run(ofproto->ml, &ofproto->revalidate_set);
849 /* Now revalidate if there's anything to do. */
850 if (ofproto->need_revalidate
851 || !tag_set_is_empty(&ofproto->revalidate_set)) {
852 struct tag_set revalidate_set = ofproto->revalidate_set;
853 bool revalidate_all = ofproto->need_revalidate;
854 struct facet *facet, *next;
856 /* Clear the revalidation flags. */
857 tag_set_init(&ofproto->revalidate_set);
858 ofproto->need_revalidate = false;
860 HMAP_FOR_EACH_SAFE (facet, next, hmap_node, &ofproto->facets) {
862 || tag_set_intersects(&revalidate_set, facet->tags)) {
863 facet_revalidate(facet);
868 /* Check the consistency of a random facet, to aid debugging. */
869 if (!hmap_is_empty(&ofproto->facets) && !ofproto->need_revalidate) {
872 facet = CONTAINER_OF(hmap_random_node(&ofproto->facets),
873 struct facet, hmap_node);
874 if (!tag_set_intersects(&ofproto->revalidate_set, facet->tags)) {
875 if (!facet_check_consistency(facet)) {
876 ofproto->need_revalidate = true;
885 wait(struct ofproto *ofproto_)
887 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
888 struct ofport_dpif *ofport;
889 struct ofbundle *bundle;
891 if (!clogged && !list_is_empty(&ofproto->completions)) {
892 poll_immediate_wake();
895 dpif_wait(ofproto->dpif);
896 dpif_recv_wait(ofproto->dpif);
897 if (ofproto->sflow) {
898 dpif_sflow_wait(ofproto->sflow);
900 if (!tag_set_is_empty(&ofproto->revalidate_set)) {
901 poll_immediate_wake();
903 HMAP_FOR_EACH (ofport, up.hmap_node, &ofproto->up.ports) {
906 HMAP_FOR_EACH (bundle, hmap_node, &ofproto->bundles) {
909 if (ofproto->netflow) {
910 netflow_wait(ofproto->netflow);
912 mac_learning_wait(ofproto->ml);
914 if (ofproto->need_revalidate) {
915 /* Shouldn't happen, but if it does just go around again. */
916 VLOG_DBG_RL(&rl, "need revalidate in ofproto_wait_cb()");
917 poll_immediate_wake();
919 timer_wait(&ofproto->next_expiration);
924 flush(struct ofproto *ofproto_)
926 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
927 struct facet *facet, *next_facet;
929 HMAP_FOR_EACH_SAFE (facet, next_facet, hmap_node, &ofproto->facets) {
930 /* Mark the facet as not installed so that facet_remove() doesn't
931 * bother trying to uninstall it. There is no point in uninstalling it
932 * individually since we are about to blow away all the facets with
933 * dpif_flow_flush(). */
934 struct subfacet *subfacet;
936 LIST_FOR_EACH (subfacet, list_node, &facet->subfacets) {
937 subfacet->installed = false;
938 subfacet->dp_packet_count = 0;
939 subfacet->dp_byte_count = 0;
943 dpif_flow_flush(ofproto->dpif);
947 get_features(struct ofproto *ofproto_ OVS_UNUSED,
948 bool *arp_match_ip, enum ofputil_action_bitmap *actions)
950 *arp_match_ip = true;
951 *actions = (OFPUTIL_A_OUTPUT |
952 OFPUTIL_A_SET_VLAN_VID |
953 OFPUTIL_A_SET_VLAN_PCP |
954 OFPUTIL_A_STRIP_VLAN |
955 OFPUTIL_A_SET_DL_SRC |
956 OFPUTIL_A_SET_DL_DST |
957 OFPUTIL_A_SET_NW_SRC |
958 OFPUTIL_A_SET_NW_DST |
959 OFPUTIL_A_SET_NW_TOS |
960 OFPUTIL_A_SET_TP_SRC |
961 OFPUTIL_A_SET_TP_DST |
966 get_tables(struct ofproto *ofproto_, struct ofp_table_stats *ots)
968 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
969 struct dpif_dp_stats s;
971 strcpy(ots->name, "classifier");
973 dpif_get_dp_stats(ofproto->dpif, &s);
974 put_32aligned_be64(&ots->lookup_count, htonll(s.n_hit + s.n_missed));
975 put_32aligned_be64(&ots->matched_count,
976 htonll(s.n_hit + ofproto->n_matches));
979 static struct ofport *
982 struct ofport_dpif *port = xmalloc(sizeof *port);
987 port_dealloc(struct ofport *port_)
989 struct ofport_dpif *port = ofport_dpif_cast(port_);
994 port_construct(struct ofport *port_)
996 struct ofport_dpif *port = ofport_dpif_cast(port_);
997 struct ofproto_dpif *ofproto = ofproto_dpif_cast(port->up.ofproto);
999 ofproto->need_revalidate = true;
1000 port->odp_port = ofp_port_to_odp_port(port->up.ofp_port);
1001 port->bundle = NULL;
1003 port->tag = tag_create_random();
1004 port->may_enable = true;
1005 port->stp_port = NULL;
1006 port->stp_state = STP_DISABLED;
1007 hmap_init(&port->priorities);
1008 port->realdev_ofp_port = 0;
1009 port->vlandev_vid = 0;
1010 port->carrier_seq = netdev_get_carrier_resets(port->up.netdev);
1012 if (ofproto->sflow) {
1013 dpif_sflow_add_port(ofproto->sflow, port_);
1020 port_destruct(struct ofport *port_)
1022 struct ofport_dpif *port = ofport_dpif_cast(port_);
1023 struct ofproto_dpif *ofproto = ofproto_dpif_cast(port->up.ofproto);
1025 ofproto->need_revalidate = true;
1026 bundle_remove(port_);
1027 set_cfm(port_, NULL);
1028 if (ofproto->sflow) {
1029 dpif_sflow_del_port(ofproto->sflow, port->odp_port);
1032 ofport_clear_priorities(port);
1033 hmap_destroy(&port->priorities);
1037 port_modified(struct ofport *port_)
1039 struct ofport_dpif *port = ofport_dpif_cast(port_);
1041 if (port->bundle && port->bundle->bond) {
1042 bond_slave_set_netdev(port->bundle->bond, port, port->up.netdev);
1047 port_reconfigured(struct ofport *port_, enum ofputil_port_config old_config)
1049 struct ofport_dpif *port = ofport_dpif_cast(port_);
1050 struct ofproto_dpif *ofproto = ofproto_dpif_cast(port->up.ofproto);
1051 enum ofputil_port_config changed = old_config ^ port->up.pp.config;
1053 if (changed & (OFPUTIL_PC_NO_RECV | OFPUTIL_PC_NO_RECV_STP |
1054 OFPUTIL_PC_NO_FWD | OFPUTIL_PC_NO_FLOOD)) {
1055 ofproto->need_revalidate = true;
1057 if (changed & OFPUTIL_PC_NO_FLOOD && port->bundle) {
1058 bundle_update(port->bundle);
1064 set_sflow(struct ofproto *ofproto_,
1065 const struct ofproto_sflow_options *sflow_options)
1067 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1068 struct dpif_sflow *ds = ofproto->sflow;
1070 if (sflow_options) {
1072 struct ofport_dpif *ofport;
1074 ds = ofproto->sflow = dpif_sflow_create(ofproto->dpif);
1075 HMAP_FOR_EACH (ofport, up.hmap_node, &ofproto->up.ports) {
1076 dpif_sflow_add_port(ds, &ofport->up);
1078 ofproto->need_revalidate = true;
1080 dpif_sflow_set_options(ds, sflow_options);
1083 dpif_sflow_destroy(ds);
1084 ofproto->need_revalidate = true;
1085 ofproto->sflow = NULL;
1092 set_cfm(struct ofport *ofport_, const struct cfm_settings *s)
1094 struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
1101 struct ofproto_dpif *ofproto;
1103 ofproto = ofproto_dpif_cast(ofport->up.ofproto);
1104 ofproto->need_revalidate = true;
1105 ofport->cfm = cfm_create(netdev_get_name(ofport->up.netdev));
1108 if (cfm_configure(ofport->cfm, s)) {
1114 cfm_destroy(ofport->cfm);
1120 get_cfm_fault(const struct ofport *ofport_)
1122 struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
1124 return ofport->cfm ? cfm_get_fault(ofport->cfm) : -1;
1128 get_cfm_remote_mpids(const struct ofport *ofport_, const uint64_t **rmps,
1131 struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
1134 cfm_get_remote_mpids(ofport->cfm, rmps, n_rmps);
1142 get_cfm_health(const struct ofport *ofport_)
1144 struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
1146 return ofport->cfm ? cfm_get_health(ofport->cfm) : -1;
1149 /* Spanning Tree. */
1152 send_bpdu_cb(struct ofpbuf *pkt, int port_num, void *ofproto_)
1154 struct ofproto_dpif *ofproto = ofproto_;
1155 struct stp_port *sp = stp_get_port(ofproto->stp, port_num);
1156 struct ofport_dpif *ofport;
1158 ofport = stp_port_get_aux(sp);
1160 VLOG_WARN_RL(&rl, "%s: cannot send BPDU on unknown port %d",
1161 ofproto->up.name, port_num);
1163 struct eth_header *eth = pkt->l2;
1165 netdev_get_etheraddr(ofport->up.netdev, eth->eth_src);
1166 if (eth_addr_is_zero(eth->eth_src)) {
1167 VLOG_WARN_RL(&rl, "%s: cannot send BPDU on port %d "
1168 "with unknown MAC", ofproto->up.name, port_num);
1170 send_packet(ofport, pkt);
1176 /* Configures STP on 'ofproto_' using the settings defined in 's'. */
1178 set_stp(struct ofproto *ofproto_, const struct ofproto_stp_settings *s)
1180 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1182 /* Only revalidate flows if the configuration changed. */
1183 if (!s != !ofproto->stp) {
1184 ofproto->need_revalidate = true;
1188 if (!ofproto->stp) {
1189 ofproto->stp = stp_create(ofproto_->name, s->system_id,
1190 send_bpdu_cb, ofproto);
1191 ofproto->stp_last_tick = time_msec();
1194 stp_set_bridge_id(ofproto->stp, s->system_id);
1195 stp_set_bridge_priority(ofproto->stp, s->priority);
1196 stp_set_hello_time(ofproto->stp, s->hello_time);
1197 stp_set_max_age(ofproto->stp, s->max_age);
1198 stp_set_forward_delay(ofproto->stp, s->fwd_delay);
1200 struct ofport *ofport;
1202 HMAP_FOR_EACH (ofport, hmap_node, &ofproto->up.ports) {
1203 set_stp_port(ofport, NULL);
1206 stp_destroy(ofproto->stp);
1207 ofproto->stp = NULL;
1214 get_stp_status(struct ofproto *ofproto_, struct ofproto_stp_status *s)
1216 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1220 s->bridge_id = stp_get_bridge_id(ofproto->stp);
1221 s->designated_root = stp_get_designated_root(ofproto->stp);
1222 s->root_path_cost = stp_get_root_path_cost(ofproto->stp);
1231 update_stp_port_state(struct ofport_dpif *ofport)
1233 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport->up.ofproto);
1234 enum stp_state state;
1236 /* Figure out new state. */
1237 state = ofport->stp_port ? stp_port_get_state(ofport->stp_port)
1241 if (ofport->stp_state != state) {
1242 enum ofputil_port_state of_state;
1245 VLOG_DBG_RL(&rl, "port %s: STP state changed from %s to %s",
1246 netdev_get_name(ofport->up.netdev),
1247 stp_state_name(ofport->stp_state),
1248 stp_state_name(state));
1249 if (stp_learn_in_state(ofport->stp_state)
1250 != stp_learn_in_state(state)) {
1251 /* xxx Learning action flows should also be flushed. */
1252 mac_learning_flush(ofproto->ml, &ofproto->revalidate_set);
1254 fwd_change = stp_forward_in_state(ofport->stp_state)
1255 != stp_forward_in_state(state);
1257 ofproto->need_revalidate = true;
1258 ofport->stp_state = state;
1259 ofport->stp_state_entered = time_msec();
1261 if (fwd_change && ofport->bundle) {
1262 bundle_update(ofport->bundle);
1265 /* Update the STP state bits in the OpenFlow port description. */
1266 of_state = ofport->up.pp.state & ~OFPUTIL_PS_STP_MASK;
1267 of_state |= (state == STP_LISTENING ? OFPUTIL_PS_STP_LISTEN
1268 : state == STP_LEARNING ? OFPUTIL_PS_STP_LEARN
1269 : state == STP_FORWARDING ? OFPUTIL_PS_STP_FORWARD
1270 : state == STP_BLOCKING ? OFPUTIL_PS_STP_BLOCK
1272 ofproto_port_set_state(&ofport->up, of_state);
1276 /* Configures STP on 'ofport_' using the settings defined in 's'. The
1277 * caller is responsible for assigning STP port numbers and ensuring
1278 * there are no duplicates. */
1280 set_stp_port(struct ofport *ofport_,
1281 const struct ofproto_port_stp_settings *s)
1283 struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
1284 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport->up.ofproto);
1285 struct stp_port *sp = ofport->stp_port;
1287 if (!s || !s->enable) {
1289 ofport->stp_port = NULL;
1290 stp_port_disable(sp);
1291 update_stp_port_state(ofport);
1294 } else if (sp && stp_port_no(sp) != s->port_num
1295 && ofport == stp_port_get_aux(sp)) {
1296 /* The port-id changed, so disable the old one if it's not
1297 * already in use by another port. */
1298 stp_port_disable(sp);
1301 sp = ofport->stp_port = stp_get_port(ofproto->stp, s->port_num);
1302 stp_port_enable(sp);
1304 stp_port_set_aux(sp, ofport);
1305 stp_port_set_priority(sp, s->priority);
1306 stp_port_set_path_cost(sp, s->path_cost);
1308 update_stp_port_state(ofport);
1314 get_stp_port_status(struct ofport *ofport_,
1315 struct ofproto_port_stp_status *s)
1317 struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
1318 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport->up.ofproto);
1319 struct stp_port *sp = ofport->stp_port;
1321 if (!ofproto->stp || !sp) {
1327 s->port_id = stp_port_get_id(sp);
1328 s->state = stp_port_get_state(sp);
1329 s->sec_in_state = (time_msec() - ofport->stp_state_entered) / 1000;
1330 s->role = stp_port_get_role(sp);
1331 stp_port_get_counts(sp, &s->tx_count, &s->rx_count, &s->error_count);
1337 stp_run(struct ofproto_dpif *ofproto)
1340 long long int now = time_msec();
1341 long long int elapsed = now - ofproto->stp_last_tick;
1342 struct stp_port *sp;
1345 stp_tick(ofproto->stp, MIN(INT_MAX, elapsed));
1346 ofproto->stp_last_tick = now;
1348 while (stp_get_changed_port(ofproto->stp, &sp)) {
1349 struct ofport_dpif *ofport = stp_port_get_aux(sp);
1352 update_stp_port_state(ofport);
1356 if (stp_check_and_reset_fdb_flush(ofproto->stp)) {
1357 mac_learning_flush(ofproto->ml, &ofproto->revalidate_set);
1363 stp_wait(struct ofproto_dpif *ofproto)
1366 poll_timer_wait(1000);
1370 /* Returns true if STP should process 'flow'. */
1372 stp_should_process_flow(const struct flow *flow)
1374 return eth_addr_equals(flow->dl_dst, eth_addr_stp);
1378 stp_process_packet(const struct ofport_dpif *ofport,
1379 const struct ofpbuf *packet)
1381 struct ofpbuf payload = *packet;
1382 struct eth_header *eth = payload.data;
1383 struct stp_port *sp = ofport->stp_port;
1385 /* Sink packets on ports that have STP disabled when the bridge has
1387 if (!sp || stp_port_get_state(sp) == STP_DISABLED) {
1391 /* Trim off padding on payload. */
1392 if (payload.size > ntohs(eth->eth_type) + ETH_HEADER_LEN) {
1393 payload.size = ntohs(eth->eth_type) + ETH_HEADER_LEN;
1396 if (ofpbuf_try_pull(&payload, ETH_HEADER_LEN + LLC_HEADER_LEN)) {
1397 stp_received_bpdu(sp, payload.data, payload.size);
1401 static struct priority_to_dscp *
1402 get_priority(const struct ofport_dpif *ofport, uint32_t priority)
1404 struct priority_to_dscp *pdscp;
1407 hash = hash_int(priority, 0);
1408 HMAP_FOR_EACH_IN_BUCKET (pdscp, hmap_node, hash, &ofport->priorities) {
1409 if (pdscp->priority == priority) {
1417 ofport_clear_priorities(struct ofport_dpif *ofport)
1419 struct priority_to_dscp *pdscp, *next;
1421 HMAP_FOR_EACH_SAFE (pdscp, next, hmap_node, &ofport->priorities) {
1422 hmap_remove(&ofport->priorities, &pdscp->hmap_node);
1428 set_queues(struct ofport *ofport_,
1429 const struct ofproto_port_queue *qdscp_list,
1432 struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
1433 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport->up.ofproto);
1434 struct hmap new = HMAP_INITIALIZER(&new);
1437 for (i = 0; i < n_qdscp; i++) {
1438 struct priority_to_dscp *pdscp;
1442 dscp = (qdscp_list[i].dscp << 2) & IP_DSCP_MASK;
1443 if (dpif_queue_to_priority(ofproto->dpif, qdscp_list[i].queue,
1448 pdscp = get_priority(ofport, priority);
1450 hmap_remove(&ofport->priorities, &pdscp->hmap_node);
1452 pdscp = xmalloc(sizeof *pdscp);
1453 pdscp->priority = priority;
1455 ofproto->need_revalidate = true;
1458 if (pdscp->dscp != dscp) {
1460 ofproto->need_revalidate = true;
1463 hmap_insert(&new, &pdscp->hmap_node, hash_int(pdscp->priority, 0));
1466 if (!hmap_is_empty(&ofport->priorities)) {
1467 ofport_clear_priorities(ofport);
1468 ofproto->need_revalidate = true;
1471 hmap_swap(&new, &ofport->priorities);
1479 /* Expires all MAC learning entries associated with 'bundle' and forces its
1480 * ofproto to revalidate every flow.
1482 * Normally MAC learning entries are removed only from the ofproto associated
1483 * with 'bundle', but if 'all_ofprotos' is true, then the MAC learning entries
1484 * are removed from every ofproto. When patch ports and SLB bonds are in use
1485 * and a VM migration happens and the gratuitous ARPs are somehow lost, this
1486 * avoids a MAC_ENTRY_IDLE_TIME delay before the migrated VM can communicate
1487 * with the host from which it migrated. */
1489 bundle_flush_macs(struct ofbundle *bundle, bool all_ofprotos)
1491 struct ofproto_dpif *ofproto = bundle->ofproto;
1492 struct mac_learning *ml = ofproto->ml;
1493 struct mac_entry *mac, *next_mac;
1495 ofproto->need_revalidate = true;
1496 LIST_FOR_EACH_SAFE (mac, next_mac, lru_node, &ml->lrus) {
1497 if (mac->port.p == bundle) {
1499 struct ofproto_dpif *o;
1501 HMAP_FOR_EACH (o, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
1503 struct mac_entry *e;
1505 e = mac_learning_lookup(o->ml, mac->mac, mac->vlan,
1508 tag_set_add(&o->revalidate_set, e->tag);
1509 mac_learning_expire(o->ml, e);
1515 mac_learning_expire(ml, mac);
1520 static struct ofbundle *
1521 bundle_lookup(const struct ofproto_dpif *ofproto, void *aux)
1523 struct ofbundle *bundle;
1525 HMAP_FOR_EACH_IN_BUCKET (bundle, hmap_node, hash_pointer(aux, 0),
1526 &ofproto->bundles) {
1527 if (bundle->aux == aux) {
1534 /* Looks up each of the 'n_auxes' pointers in 'auxes' as bundles and adds the
1535 * ones that are found to 'bundles'. */
1537 bundle_lookup_multiple(struct ofproto_dpif *ofproto,
1538 void **auxes, size_t n_auxes,
1539 struct hmapx *bundles)
1543 hmapx_init(bundles);
1544 for (i = 0; i < n_auxes; i++) {
1545 struct ofbundle *bundle = bundle_lookup(ofproto, auxes[i]);
1547 hmapx_add(bundles, bundle);
1553 bundle_update(struct ofbundle *bundle)
1555 struct ofport_dpif *port;
1557 bundle->floodable = true;
1558 LIST_FOR_EACH (port, bundle_node, &bundle->ports) {
1559 if (port->up.pp.config & OFPUTIL_PC_NO_FLOOD
1560 || !stp_forward_in_state(port->stp_state)) {
1561 bundle->floodable = false;
1568 bundle_del_port(struct ofport_dpif *port)
1570 struct ofbundle *bundle = port->bundle;
1572 bundle->ofproto->need_revalidate = true;
1574 list_remove(&port->bundle_node);
1575 port->bundle = NULL;
1578 lacp_slave_unregister(bundle->lacp, port);
1581 bond_slave_unregister(bundle->bond, port);
1584 bundle_update(bundle);
1588 bundle_add_port(struct ofbundle *bundle, uint32_t ofp_port,
1589 struct lacp_slave_settings *lacp,
1590 uint32_t bond_stable_id)
1592 struct ofport_dpif *port;
1594 port = get_ofp_port(bundle->ofproto, ofp_port);
1599 if (port->bundle != bundle) {
1600 bundle->ofproto->need_revalidate = true;
1602 bundle_del_port(port);
1605 port->bundle = bundle;
1606 list_push_back(&bundle->ports, &port->bundle_node);
1607 if (port->up.pp.config & OFPUTIL_PC_NO_FLOOD
1608 || !stp_forward_in_state(port->stp_state)) {
1609 bundle->floodable = false;
1613 port->bundle->ofproto->need_revalidate = true;
1614 lacp_slave_register(bundle->lacp, port, lacp);
1617 port->bond_stable_id = bond_stable_id;
1623 bundle_destroy(struct ofbundle *bundle)
1625 struct ofproto_dpif *ofproto;
1626 struct ofport_dpif *port, *next_port;
1633 ofproto = bundle->ofproto;
1634 for (i = 0; i < MAX_MIRRORS; i++) {
1635 struct ofmirror *m = ofproto->mirrors[i];
1637 if (m->out == bundle) {
1639 } else if (hmapx_find_and_delete(&m->srcs, bundle)
1640 || hmapx_find_and_delete(&m->dsts, bundle)) {
1641 ofproto->need_revalidate = true;
1646 LIST_FOR_EACH_SAFE (port, next_port, bundle_node, &bundle->ports) {
1647 bundle_del_port(port);
1650 bundle_flush_macs(bundle, true);
1651 hmap_remove(&ofproto->bundles, &bundle->hmap_node);
1653 free(bundle->trunks);
1654 lacp_destroy(bundle->lacp);
1655 bond_destroy(bundle->bond);
1660 bundle_set(struct ofproto *ofproto_, void *aux,
1661 const struct ofproto_bundle_settings *s)
1663 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1664 bool need_flush = false;
1665 struct ofport_dpif *port;
1666 struct ofbundle *bundle;
1667 unsigned long *trunks;
1673 bundle_destroy(bundle_lookup(ofproto, aux));
1677 assert(s->n_slaves == 1 || s->bond != NULL);
1678 assert((s->lacp != NULL) == (s->lacp_slaves != NULL));
1680 bundle = bundle_lookup(ofproto, aux);
1682 bundle = xmalloc(sizeof *bundle);
1684 bundle->ofproto = ofproto;
1685 hmap_insert(&ofproto->bundles, &bundle->hmap_node,
1686 hash_pointer(aux, 0));
1688 bundle->name = NULL;
1690 list_init(&bundle->ports);
1691 bundle->vlan_mode = PORT_VLAN_TRUNK;
1693 bundle->trunks = NULL;
1694 bundle->use_priority_tags = s->use_priority_tags;
1695 bundle->lacp = NULL;
1696 bundle->bond = NULL;
1698 bundle->floodable = true;
1700 bundle->src_mirrors = 0;
1701 bundle->dst_mirrors = 0;
1702 bundle->mirror_out = 0;
1705 if (!bundle->name || strcmp(s->name, bundle->name)) {
1707 bundle->name = xstrdup(s->name);
1712 if (!bundle->lacp) {
1713 ofproto->need_revalidate = true;
1714 bundle->lacp = lacp_create();
1716 lacp_configure(bundle->lacp, s->lacp);
1718 lacp_destroy(bundle->lacp);
1719 bundle->lacp = NULL;
1722 /* Update set of ports. */
1724 for (i = 0; i < s->n_slaves; i++) {
1725 if (!bundle_add_port(bundle, s->slaves[i],
1726 s->lacp ? &s->lacp_slaves[i] : NULL,
1727 s->bond_stable_ids ? s->bond_stable_ids[i] : 0)) {
1731 if (!ok || list_size(&bundle->ports) != s->n_slaves) {
1732 struct ofport_dpif *next_port;
1734 LIST_FOR_EACH_SAFE (port, next_port, bundle_node, &bundle->ports) {
1735 for (i = 0; i < s->n_slaves; i++) {
1736 if (s->slaves[i] == port->up.ofp_port) {
1741 bundle_del_port(port);
1745 assert(list_size(&bundle->ports) <= s->n_slaves);
1747 if (list_is_empty(&bundle->ports)) {
1748 bundle_destroy(bundle);
1752 /* Set VLAN tagging mode */
1753 if (s->vlan_mode != bundle->vlan_mode
1754 || s->use_priority_tags != bundle->use_priority_tags) {
1755 bundle->vlan_mode = s->vlan_mode;
1756 bundle->use_priority_tags = s->use_priority_tags;
1761 vlan = (s->vlan_mode == PORT_VLAN_TRUNK ? -1
1762 : s->vlan >= 0 && s->vlan <= 4095 ? s->vlan
1764 if (vlan != bundle->vlan) {
1765 bundle->vlan = vlan;
1769 /* Get trunked VLANs. */
1770 switch (s->vlan_mode) {
1771 case PORT_VLAN_ACCESS:
1775 case PORT_VLAN_TRUNK:
1776 trunks = (unsigned long *) s->trunks;
1779 case PORT_VLAN_NATIVE_UNTAGGED:
1780 case PORT_VLAN_NATIVE_TAGGED:
1781 if (vlan != 0 && (!s->trunks
1782 || !bitmap_is_set(s->trunks, vlan)
1783 || bitmap_is_set(s->trunks, 0))) {
1784 /* Force trunking the native VLAN and prohibit trunking VLAN 0. */
1786 trunks = bitmap_clone(s->trunks, 4096);
1788 trunks = bitmap_allocate1(4096);
1790 bitmap_set1(trunks, vlan);
1791 bitmap_set0(trunks, 0);
1793 trunks = (unsigned long *) s->trunks;
1800 if (!vlan_bitmap_equal(trunks, bundle->trunks)) {
1801 free(bundle->trunks);
1802 if (trunks == s->trunks) {
1803 bundle->trunks = vlan_bitmap_clone(trunks);
1805 bundle->trunks = trunks;
1810 if (trunks != s->trunks) {
1815 if (!list_is_short(&bundle->ports)) {
1816 bundle->ofproto->has_bonded_bundles = true;
1818 if (bond_reconfigure(bundle->bond, s->bond)) {
1819 ofproto->need_revalidate = true;
1822 bundle->bond = bond_create(s->bond);
1823 ofproto->need_revalidate = true;
1826 LIST_FOR_EACH (port, bundle_node, &bundle->ports) {
1827 bond_slave_register(bundle->bond, port, port->bond_stable_id,
1831 bond_destroy(bundle->bond);
1832 bundle->bond = NULL;
1835 /* If we changed something that would affect MAC learning, un-learn
1836 * everything on this port and force flow revalidation. */
1838 bundle_flush_macs(bundle, false);
1845 bundle_remove(struct ofport *port_)
1847 struct ofport_dpif *port = ofport_dpif_cast(port_);
1848 struct ofbundle *bundle = port->bundle;
1851 bundle_del_port(port);
1852 if (list_is_empty(&bundle->ports)) {
1853 bundle_destroy(bundle);
1854 } else if (list_is_short(&bundle->ports)) {
1855 bond_destroy(bundle->bond);
1856 bundle->bond = NULL;
1862 send_pdu_cb(void *port_, const void *pdu, size_t pdu_size)
1864 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 10);
1865 struct ofport_dpif *port = port_;
1866 uint8_t ea[ETH_ADDR_LEN];
1869 error = netdev_get_etheraddr(port->up.netdev, ea);
1871 struct ofpbuf packet;
1874 ofpbuf_init(&packet, 0);
1875 packet_pdu = eth_compose(&packet, eth_addr_lacp, ea, ETH_TYPE_LACP,
1877 memcpy(packet_pdu, pdu, pdu_size);
1879 send_packet(port, &packet);
1880 ofpbuf_uninit(&packet);
1882 VLOG_ERR_RL(&rl, "port %s: cannot obtain Ethernet address of iface "
1883 "%s (%s)", port->bundle->name,
1884 netdev_get_name(port->up.netdev), strerror(error));
1889 bundle_send_learning_packets(struct ofbundle *bundle)
1891 struct ofproto_dpif *ofproto = bundle->ofproto;
1892 int error, n_packets, n_errors;
1893 struct mac_entry *e;
1895 error = n_packets = n_errors = 0;
1896 LIST_FOR_EACH (e, lru_node, &ofproto->ml->lrus) {
1897 if (e->port.p != bundle) {
1898 struct ofpbuf *learning_packet;
1899 struct ofport_dpif *port;
1903 /* The assignment to "port" is unnecessary but makes "grep"ing for
1904 * struct ofport_dpif more effective. */
1905 learning_packet = bond_compose_learning_packet(bundle->bond,
1909 ret = send_packet(port, learning_packet);
1910 ofpbuf_delete(learning_packet);
1920 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
1921 VLOG_WARN_RL(&rl, "bond %s: %d errors sending %d gratuitous learning "
1922 "packets, last error was: %s",
1923 bundle->name, n_errors, n_packets, strerror(error));
1925 VLOG_DBG("bond %s: sent %d gratuitous learning packets",
1926 bundle->name, n_packets);
1931 bundle_run(struct ofbundle *bundle)
1934 lacp_run(bundle->lacp, send_pdu_cb);
1937 struct ofport_dpif *port;
1939 LIST_FOR_EACH (port, bundle_node, &bundle->ports) {
1940 bond_slave_set_may_enable(bundle->bond, port, port->may_enable);
1943 bond_run(bundle->bond, &bundle->ofproto->revalidate_set,
1944 lacp_status(bundle->lacp));
1945 if (bond_should_send_learning_packets(bundle->bond)) {
1946 bundle_send_learning_packets(bundle);
1952 bundle_wait(struct ofbundle *bundle)
1955 lacp_wait(bundle->lacp);
1958 bond_wait(bundle->bond);
1965 mirror_scan(struct ofproto_dpif *ofproto)
1969 for (idx = 0; idx < MAX_MIRRORS; idx++) {
1970 if (!ofproto->mirrors[idx]) {
1977 static struct ofmirror *
1978 mirror_lookup(struct ofproto_dpif *ofproto, void *aux)
1982 for (i = 0; i < MAX_MIRRORS; i++) {
1983 struct ofmirror *mirror = ofproto->mirrors[i];
1984 if (mirror && mirror->aux == aux) {
1992 /* Update the 'dup_mirrors' member of each of the ofmirrors in 'ofproto'. */
1994 mirror_update_dups(struct ofproto_dpif *ofproto)
1998 for (i = 0; i < MAX_MIRRORS; i++) {
1999 struct ofmirror *m = ofproto->mirrors[i];
2002 m->dup_mirrors = MIRROR_MASK_C(1) << i;
2006 for (i = 0; i < MAX_MIRRORS; i++) {
2007 struct ofmirror *m1 = ofproto->mirrors[i];
2014 for (j = i + 1; j < MAX_MIRRORS; j++) {
2015 struct ofmirror *m2 = ofproto->mirrors[j];
2017 if (m2 && m1->out == m2->out && m1->out_vlan == m2->out_vlan) {
2018 m1->dup_mirrors |= MIRROR_MASK_C(1) << j;
2019 m2->dup_mirrors |= m1->dup_mirrors;
2026 mirror_set(struct ofproto *ofproto_, void *aux,
2027 const struct ofproto_mirror_settings *s)
2029 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
2030 mirror_mask_t mirror_bit;
2031 struct ofbundle *bundle;
2032 struct ofmirror *mirror;
2033 struct ofbundle *out;
2034 struct hmapx srcs; /* Contains "struct ofbundle *"s. */
2035 struct hmapx dsts; /* Contains "struct ofbundle *"s. */
2038 mirror = mirror_lookup(ofproto, aux);
2040 mirror_destroy(mirror);
2046 idx = mirror_scan(ofproto);
2048 VLOG_WARN("bridge %s: maximum of %d port mirrors reached, "
2050 ofproto->up.name, MAX_MIRRORS, s->name);
2054 mirror = ofproto->mirrors[idx] = xzalloc(sizeof *mirror);
2055 mirror->ofproto = ofproto;
2058 mirror->out_vlan = -1;
2059 mirror->name = NULL;
2062 if (!mirror->name || strcmp(s->name, mirror->name)) {
2064 mirror->name = xstrdup(s->name);
2067 /* Get the new configuration. */
2068 if (s->out_bundle) {
2069 out = bundle_lookup(ofproto, s->out_bundle);
2071 mirror_destroy(mirror);
2077 out_vlan = s->out_vlan;
2079 bundle_lookup_multiple(ofproto, s->srcs, s->n_srcs, &srcs);
2080 bundle_lookup_multiple(ofproto, s->dsts, s->n_dsts, &dsts);
2082 /* If the configuration has not changed, do nothing. */
2083 if (hmapx_equals(&srcs, &mirror->srcs)
2084 && hmapx_equals(&dsts, &mirror->dsts)
2085 && vlan_bitmap_equal(mirror->vlans, s->src_vlans)
2086 && mirror->out == out
2087 && mirror->out_vlan == out_vlan)
2089 hmapx_destroy(&srcs);
2090 hmapx_destroy(&dsts);
2094 hmapx_swap(&srcs, &mirror->srcs);
2095 hmapx_destroy(&srcs);
2097 hmapx_swap(&dsts, &mirror->dsts);
2098 hmapx_destroy(&dsts);
2100 free(mirror->vlans);
2101 mirror->vlans = vlan_bitmap_clone(s->src_vlans);
2104 mirror->out_vlan = out_vlan;
2106 /* Update bundles. */
2107 mirror_bit = MIRROR_MASK_C(1) << mirror->idx;
2108 HMAP_FOR_EACH (bundle, hmap_node, &mirror->ofproto->bundles) {
2109 if (hmapx_contains(&mirror->srcs, bundle)) {
2110 bundle->src_mirrors |= mirror_bit;
2112 bundle->src_mirrors &= ~mirror_bit;
2115 if (hmapx_contains(&mirror->dsts, bundle)) {
2116 bundle->dst_mirrors |= mirror_bit;
2118 bundle->dst_mirrors &= ~mirror_bit;
2121 if (mirror->out == bundle) {
2122 bundle->mirror_out |= mirror_bit;
2124 bundle->mirror_out &= ~mirror_bit;
2128 ofproto->need_revalidate = true;
2129 mac_learning_flush(ofproto->ml, &ofproto->revalidate_set);
2130 mirror_update_dups(ofproto);
2136 mirror_destroy(struct ofmirror *mirror)
2138 struct ofproto_dpif *ofproto;
2139 mirror_mask_t mirror_bit;
2140 struct ofbundle *bundle;
2146 ofproto = mirror->ofproto;
2147 ofproto->need_revalidate = true;
2148 mac_learning_flush(ofproto->ml, &ofproto->revalidate_set);
2150 mirror_bit = MIRROR_MASK_C(1) << mirror->idx;
2151 HMAP_FOR_EACH (bundle, hmap_node, &ofproto->bundles) {
2152 bundle->src_mirrors &= ~mirror_bit;
2153 bundle->dst_mirrors &= ~mirror_bit;
2154 bundle->mirror_out &= ~mirror_bit;
2157 hmapx_destroy(&mirror->srcs);
2158 hmapx_destroy(&mirror->dsts);
2159 free(mirror->vlans);
2161 ofproto->mirrors[mirror->idx] = NULL;
2165 mirror_update_dups(ofproto);
2169 mirror_get_stats(struct ofproto *ofproto_, void *aux,
2170 uint64_t *packets, uint64_t *bytes)
2172 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
2173 struct ofmirror *mirror = mirror_lookup(ofproto, aux);
2176 *packets = *bytes = UINT64_MAX;
2180 *packets = mirror->packet_count;
2181 *bytes = mirror->byte_count;
2187 set_flood_vlans(struct ofproto *ofproto_, unsigned long *flood_vlans)
2189 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
2190 if (mac_learning_set_flood_vlans(ofproto->ml, flood_vlans)) {
2191 mac_learning_flush(ofproto->ml, &ofproto->revalidate_set);
2197 is_mirror_output_bundle(const struct ofproto *ofproto_, void *aux)
2199 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
2200 struct ofbundle *bundle = bundle_lookup(ofproto, aux);
2201 return bundle && bundle->mirror_out != 0;
2205 forward_bpdu_changed(struct ofproto *ofproto_)
2207 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
2208 /* Revalidate cached flows whenever forward_bpdu option changes. */
2209 ofproto->need_revalidate = true;
2213 set_mac_idle_time(struct ofproto *ofproto_, unsigned int idle_time)
2215 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
2216 mac_learning_set_idle_time(ofproto->ml, idle_time);
2221 static struct ofport_dpif *
2222 get_ofp_port(struct ofproto_dpif *ofproto, uint16_t ofp_port)
2224 struct ofport *ofport = ofproto_get_port(&ofproto->up, ofp_port);
2225 return ofport ? ofport_dpif_cast(ofport) : NULL;
2228 static struct ofport_dpif *
2229 get_odp_port(struct ofproto_dpif *ofproto, uint32_t odp_port)
2231 return get_ofp_port(ofproto, odp_port_to_ofp_port(odp_port));
2235 ofproto_port_from_dpif_port(struct ofproto_port *ofproto_port,
2236 struct dpif_port *dpif_port)
2238 ofproto_port->name = dpif_port->name;
2239 ofproto_port->type = dpif_port->type;
2240 ofproto_port->ofp_port = odp_port_to_ofp_port(dpif_port->port_no);
2244 port_run(struct ofport_dpif *ofport)
2246 long long int carrier_seq = netdev_get_carrier_resets(ofport->up.netdev);
2247 bool carrier_changed = carrier_seq != ofport->carrier_seq;
2248 bool enable = netdev_get_carrier(ofport->up.netdev);
2250 ofport->carrier_seq = carrier_seq;
2253 cfm_run(ofport->cfm);
2255 if (cfm_should_send_ccm(ofport->cfm)) {
2256 struct ofpbuf packet;
2258 ofpbuf_init(&packet, 0);
2259 cfm_compose_ccm(ofport->cfm, &packet, ofport->up.pp.hw_addr);
2260 send_packet(ofport, &packet);
2261 ofpbuf_uninit(&packet);
2264 enable = enable && !cfm_get_fault(ofport->cfm)
2265 && cfm_get_opup(ofport->cfm);
2268 if (ofport->bundle) {
2269 enable = enable && lacp_slave_may_enable(ofport->bundle->lacp, ofport);
2270 if (carrier_changed) {
2271 lacp_slave_carrier_changed(ofport->bundle->lacp, ofport);
2275 if (ofport->may_enable != enable) {
2276 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport->up.ofproto);
2278 if (ofproto->has_bundle_action) {
2279 ofproto->need_revalidate = true;
2283 ofport->may_enable = enable;
2287 port_wait(struct ofport_dpif *ofport)
2290 cfm_wait(ofport->cfm);
2295 port_query_by_name(const struct ofproto *ofproto_, const char *devname,
2296 struct ofproto_port *ofproto_port)
2298 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
2299 struct dpif_port dpif_port;
2302 error = dpif_port_query_by_name(ofproto->dpif, devname, &dpif_port);
2304 ofproto_port_from_dpif_port(ofproto_port, &dpif_port);
2310 port_add(struct ofproto *ofproto_, struct netdev *netdev, uint16_t *ofp_portp)
2312 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
2316 error = dpif_port_add(ofproto->dpif, netdev, &odp_port);
2318 *ofp_portp = odp_port_to_ofp_port(odp_port);
2324 port_del(struct ofproto *ofproto_, uint16_t ofp_port)
2326 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
2329 error = dpif_port_del(ofproto->dpif, ofp_port_to_odp_port(ofp_port));
2331 struct ofport_dpif *ofport = get_ofp_port(ofproto, ofp_port);
2333 /* The caller is going to close ofport->up.netdev. If this is a
2334 * bonded port, then the bond is using that netdev, so remove it
2335 * from the bond. The client will need to reconfigure everything
2336 * after deleting ports, so then the slave will get re-added. */
2337 bundle_remove(&ofport->up);
2344 port_get_stats(const struct ofport *ofport_, struct netdev_stats *stats)
2346 struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
2349 error = netdev_get_stats(ofport->up.netdev, stats);
2351 if (!error && ofport->odp_port == OVSP_LOCAL) {
2352 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport->up.ofproto);
2354 /* ofproto->stats.tx_packets represents packets that we created
2355 * internally and sent to some port (e.g. packets sent with
2356 * send_packet()). Account for them as if they had come from
2357 * OFPP_LOCAL and got forwarded. */
2359 if (stats->rx_packets != UINT64_MAX) {
2360 stats->rx_packets += ofproto->stats.tx_packets;
2363 if (stats->rx_bytes != UINT64_MAX) {
2364 stats->rx_bytes += ofproto->stats.tx_bytes;
2367 /* ofproto->stats.rx_packets represents packets that were received on
2368 * some port and we processed internally and dropped (e.g. STP).
2369 * Account fro them as if they had been forwarded to OFPP_LOCAL. */
2371 if (stats->tx_packets != UINT64_MAX) {
2372 stats->tx_packets += ofproto->stats.rx_packets;
2375 if (stats->tx_bytes != UINT64_MAX) {
2376 stats->tx_bytes += ofproto->stats.rx_bytes;
2383 /* Account packets for LOCAL port. */
2385 ofproto_update_local_port_stats(const struct ofproto *ofproto_,
2386 size_t tx_size, size_t rx_size)
2388 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
2391 ofproto->stats.rx_packets++;
2392 ofproto->stats.rx_bytes += rx_size;
2395 ofproto->stats.tx_packets++;
2396 ofproto->stats.tx_bytes += tx_size;
2400 struct port_dump_state {
2401 struct dpif_port_dump dump;
2406 port_dump_start(const struct ofproto *ofproto_, void **statep)
2408 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
2409 struct port_dump_state *state;
2411 *statep = state = xmalloc(sizeof *state);
2412 dpif_port_dump_start(&state->dump, ofproto->dpif);
2413 state->done = false;
2418 port_dump_next(const struct ofproto *ofproto_ OVS_UNUSED, void *state_,
2419 struct ofproto_port *port)
2421 struct port_dump_state *state = state_;
2422 struct dpif_port dpif_port;
2424 if (dpif_port_dump_next(&state->dump, &dpif_port)) {
2425 ofproto_port_from_dpif_port(port, &dpif_port);
2428 int error = dpif_port_dump_done(&state->dump);
2430 return error ? error : EOF;
2435 port_dump_done(const struct ofproto *ofproto_ OVS_UNUSED, void *state_)
2437 struct port_dump_state *state = state_;
2440 dpif_port_dump_done(&state->dump);
2447 port_poll(const struct ofproto *ofproto_, char **devnamep)
2449 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
2450 return dpif_port_poll(ofproto->dpif, devnamep);
2454 port_poll_wait(const struct ofproto *ofproto_)
2456 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
2457 dpif_port_poll_wait(ofproto->dpif);
2461 port_is_lacp_current(const struct ofport *ofport_)
2463 const struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
2464 return (ofport->bundle && ofport->bundle->lacp
2465 ? lacp_slave_is_current(ofport->bundle->lacp, ofport)
2469 /* Upcall handling. */
2471 /* Flow miss batching.
2473 * Some dpifs implement operations faster when you hand them off in a batch.
2474 * To allow batching, "struct flow_miss" queues the dpif-related work needed
2475 * for a given flow. Each "struct flow_miss" corresponds to sending one or
2476 * more packets, plus possibly installing the flow in the dpif.
2478 * So far we only batch the operations that affect flow setup time the most.
2479 * It's possible to batch more than that, but the benefit might be minimal. */
2481 struct hmap_node hmap_node;
2483 enum odp_key_fitness key_fitness;
2484 const struct nlattr *key;
2486 ovs_be16 initial_tci;
2487 struct list packets;
2490 struct flow_miss_op {
2491 struct dpif_op dpif_op;
2492 struct subfacet *subfacet;
2495 /* Sends an OFPT_PACKET_IN message for 'packet' of type OFPR_NO_MATCH to each
2496 * OpenFlow controller as necessary according to their individual
2497 * configurations. */
2499 send_packet_in_miss(struct ofproto_dpif *ofproto, const struct ofpbuf *packet,
2500 const struct flow *flow)
2502 struct ofputil_packet_in pin;
2504 pin.packet = packet->data;
2505 pin.packet_len = packet->size;
2506 pin.reason = OFPR_NO_MATCH;
2507 pin.controller_id = 0;
2512 pin.send_len = 0; /* not used for flow table misses */
2514 flow_get_metadata(flow, &pin.fmd);
2516 /* Registers aren't meaningful on a miss. */
2517 memset(pin.fmd.reg_masks, 0, sizeof pin.fmd.reg_masks);
2519 connmgr_send_packet_in(ofproto->up.connmgr, &pin);
2523 process_special(struct ofproto_dpif *ofproto, const struct flow *flow,
2524 const struct ofpbuf *packet)
2526 struct ofport_dpif *ofport = get_ofp_port(ofproto, flow->in_port);
2532 if (ofport->cfm && cfm_should_process_flow(ofport->cfm, flow)) {
2534 cfm_process_heartbeat(ofport->cfm, packet);
2537 } else if (ofport->bundle && ofport->bundle->lacp
2538 && flow->dl_type == htons(ETH_TYPE_LACP)) {
2540 lacp_process_packet(ofport->bundle->lacp, ofport, packet);
2543 } else if (ofproto->stp && stp_should_process_flow(flow)) {
2545 stp_process_packet(ofport, packet);
2552 static struct flow_miss *
2553 flow_miss_find(struct hmap *todo, const struct flow *flow, uint32_t hash)
2555 struct flow_miss *miss;
2557 HMAP_FOR_EACH_WITH_HASH (miss, hmap_node, hash, todo) {
2558 if (flow_equal(&miss->flow, flow)) {
2567 handle_flow_miss(struct ofproto_dpif *ofproto, struct flow_miss *miss,
2568 struct flow_miss_op *ops, size_t *n_ops)
2570 const struct flow *flow = &miss->flow;
2571 struct subfacet *subfacet;
2572 struct ofpbuf *packet;
2573 struct facet *facet;
2576 /* The caller must ensure that miss->hmap_node.hash contains
2577 * flow_hash(miss->flow, 0). */
2578 hash = miss->hmap_node.hash;
2580 facet = facet_lookup_valid(ofproto, flow, hash);
2582 struct rule_dpif *rule;
2584 rule = rule_dpif_lookup(ofproto, flow, 0);
2586 /* Don't send a packet-in if OFPUTIL_PC_NO_PACKET_IN asserted. */
2587 struct ofport_dpif *port = get_ofp_port(ofproto, flow->in_port);
2589 if (port->up.pp.config & OFPUTIL_PC_NO_PACKET_IN) {
2590 COVERAGE_INC(ofproto_dpif_no_packet_in);
2591 /* XXX install 'drop' flow entry */
2595 VLOG_WARN_RL(&rl, "packet-in on unknown port %"PRIu16,
2599 LIST_FOR_EACH (packet, list_node, &miss->packets) {
2600 send_packet_in_miss(ofproto, packet, flow);
2606 facet = facet_create(rule, flow, hash);
2609 subfacet = subfacet_create(facet,
2610 miss->key_fitness, miss->key, miss->key_len,
2613 LIST_FOR_EACH (packet, list_node, &miss->packets) {
2614 struct dpif_flow_stats stats;
2615 struct flow_miss_op *op;
2616 struct dpif_execute *execute;
2618 ofproto->n_matches++;
2620 if (facet->rule->up.cr.priority == FAIL_OPEN_PRIORITY) {
2622 * Extra-special case for fail-open mode.
2624 * We are in fail-open mode and the packet matched the fail-open
2625 * rule, but we are connected to a controller too. We should send
2626 * the packet up to the controller in the hope that it will try to
2627 * set up a flow and thereby allow us to exit fail-open.
2629 * See the top-level comment in fail-open.c for more information.
2631 send_packet_in_miss(ofproto, packet, flow);
2634 if (!facet->may_install || !subfacet->actions) {
2635 subfacet_make_actions(subfacet, packet);
2638 dpif_flow_stats_extract(&facet->flow, packet, &stats);
2639 subfacet_update_stats(subfacet, &stats);
2641 if (!subfacet->actions_len) {
2642 /* No actions to execute, so skip talking to the dpif. */
2646 if (flow->vlan_tci != subfacet->initial_tci) {
2647 /* This packet was received on a VLAN splinter port. We added
2648 * a VLAN to the packet to make the packet resemble the flow,
2649 * but the actions were composed assuming that the packet
2650 * contained no VLAN. So, we must remove the VLAN header from
2651 * the packet before trying to execute the actions. */
2652 eth_pop_vlan(packet);
2655 op = &ops[(*n_ops)++];
2656 execute = &op->dpif_op.u.execute;
2657 op->subfacet = subfacet;
2658 op->dpif_op.type = DPIF_OP_EXECUTE;
2659 execute->key = miss->key;
2660 execute->key_len = miss->key_len;
2661 execute->actions = (facet->may_install
2663 : xmemdup(subfacet->actions,
2664 subfacet->actions_len));
2665 execute->actions_len = subfacet->actions_len;
2666 execute->packet = packet;
2669 if (facet->may_install && subfacet->key_fitness != ODP_FIT_TOO_LITTLE) {
2670 struct flow_miss_op *op = &ops[(*n_ops)++];
2671 struct dpif_flow_put *put = &op->dpif_op.u.flow_put;
2673 op->subfacet = subfacet;
2674 op->dpif_op.type = DPIF_OP_FLOW_PUT;
2675 put->flags = DPIF_FP_CREATE | DPIF_FP_MODIFY;
2676 put->key = miss->key;
2677 put->key_len = miss->key_len;
2678 put->actions = subfacet->actions;
2679 put->actions_len = subfacet->actions_len;
2684 /* Like odp_flow_key_to_flow(), this function converts the 'key_len' bytes of
2685 * OVS_KEY_ATTR_* attributes in 'key' to a flow structure in 'flow' and returns
2686 * an ODP_FIT_* value that indicates how well 'key' fits our expectations for
2687 * what a flow key should contain.
2689 * This function also includes some logic to help make VLAN splinters
2690 * transparent to the rest of the upcall processing logic. In particular, if
2691 * the extracted in_port is a VLAN splinter port, it replaces flow->in_port by
2692 * the "real" port, sets flow->vlan_tci correctly for the VLAN of the VLAN
2693 * splinter port, and pushes a VLAN header onto 'packet' (if it is nonnull).
2695 * Sets '*initial_tci' to the VLAN TCI with which the packet was really
2696 * received, that is, the actual VLAN TCI extracted by odp_flow_key_to_flow().
2697 * (This differs from the value returned in flow->vlan_tci only for packets
2698 * received on VLAN splinters.)
2700 static enum odp_key_fitness
2701 ofproto_dpif_extract_flow_key(const struct ofproto_dpif *ofproto,
2702 const struct nlattr *key, size_t key_len,
2703 struct flow *flow, ovs_be16 *initial_tci,
2704 struct ofpbuf *packet)
2706 enum odp_key_fitness fitness;
2710 fitness = odp_flow_key_to_flow(key, key_len, flow);
2711 if (fitness == ODP_FIT_ERROR) {
2714 *initial_tci = flow->vlan_tci;
2716 realdev = vsp_vlandev_to_realdev(ofproto, flow->in_port, &vid);
2718 /* Cause the flow to be processed as if it came in on the real device
2719 * with the VLAN device's VLAN ID. */
2720 flow->in_port = realdev;
2721 flow->vlan_tci = htons((vid & VLAN_VID_MASK) | VLAN_CFI);
2723 /* Make the packet resemble the flow, so that it gets sent to an
2724 * OpenFlow controller properly, so that it looks correct for
2725 * sFlow, and so that flow_extract() will get the correct vlan_tci
2726 * if it is called on 'packet'.
2728 * The allocated space inside 'packet' probably also contains
2729 * 'key', that is, both 'packet' and 'key' are probably part of a
2730 * struct dpif_upcall (see the large comment on that structure
2731 * definition), so pushing data on 'packet' is in general not a
2732 * good idea since it could overwrite 'key' or free it as a side
2733 * effect. However, it's OK in this special case because we know
2734 * that 'packet' is inside a Netlink attribute: pushing 4 bytes
2735 * will just overwrite the 4-byte "struct nlattr", which is fine
2736 * since we don't need that header anymore. */
2737 eth_push_vlan(packet, flow->vlan_tci);
2740 /* Let the caller know that we can't reproduce 'key' from 'flow'. */
2741 if (fitness == ODP_FIT_PERFECT) {
2742 fitness = ODP_FIT_TOO_MUCH;
2750 handle_miss_upcalls(struct ofproto_dpif *ofproto, struct dpif_upcall *upcalls,
2753 struct dpif_upcall *upcall;
2754 struct flow_miss *miss;
2755 struct flow_miss misses[FLOW_MISS_MAX_BATCH];
2756 struct flow_miss_op flow_miss_ops[FLOW_MISS_MAX_BATCH * 2];
2757 struct dpif_op *dpif_ops[FLOW_MISS_MAX_BATCH * 2];
2767 /* Construct the to-do list.
2769 * This just amounts to extracting the flow from each packet and sticking
2770 * the packets that have the same flow in the same "flow_miss" structure so
2771 * that we can process them together. */
2774 for (upcall = upcalls; upcall < &upcalls[n_upcalls]; upcall++) {
2775 struct flow_miss *miss = &misses[n_misses];
2776 struct flow_miss *existing_miss;
2779 /* Obtain metadata and check userspace/kernel agreement on flow match,
2780 * then set 'flow''s header pointers. */
2781 miss->key_fitness = ofproto_dpif_extract_flow_key(
2782 ofproto, upcall->key, upcall->key_len,
2783 &miss->flow, &miss->initial_tci, upcall->packet);
2784 if (miss->key_fitness == ODP_FIT_ERROR) {
2787 flow_extract(upcall->packet, miss->flow.skb_priority,
2788 miss->flow.tun_id, miss->flow.in_port, &miss->flow);
2790 /* Handle 802.1ag, LACP, and STP specially. */
2791 if (process_special(ofproto, &miss->flow, upcall->packet)) {
2792 ofproto_update_local_port_stats(&ofproto->up,
2793 0, upcall->packet->size);
2794 ofproto->n_matches++;
2798 /* Add other packets to a to-do list. */
2799 hash = flow_hash(&miss->flow, 0);
2800 existing_miss = flow_miss_find(&todo, &miss->flow, hash);
2801 if (!existing_miss) {
2802 hmap_insert(&todo, &miss->hmap_node, hash);
2803 miss->key = upcall->key;
2804 miss->key_len = upcall->key_len;
2805 list_init(&miss->packets);
2809 miss = existing_miss;
2811 list_push_back(&miss->packets, &upcall->packet->list_node);
2814 /* Process each element in the to-do list, constructing the set of
2815 * operations to batch. */
2817 HMAP_FOR_EACH (miss, hmap_node, &todo) {
2818 handle_flow_miss(ofproto, miss, flow_miss_ops, &n_ops);
2820 assert(n_ops <= ARRAY_SIZE(flow_miss_ops));
2822 /* Execute batch. */
2823 for (i = 0; i < n_ops; i++) {
2824 dpif_ops[i] = &flow_miss_ops[i].dpif_op;
2826 dpif_operate(ofproto->dpif, dpif_ops, n_ops);
2828 /* Free memory and update facets. */
2829 for (i = 0; i < n_ops; i++) {
2830 struct flow_miss_op *op = &flow_miss_ops[i];
2831 struct dpif_execute *execute;
2833 switch (op->dpif_op.type) {
2834 case DPIF_OP_EXECUTE:
2835 execute = &op->dpif_op.u.execute;
2836 if (op->subfacet->actions != execute->actions) {
2837 free((struct nlattr *) execute->actions);
2841 case DPIF_OP_FLOW_PUT:
2842 if (!op->dpif_op.error) {
2843 op->subfacet->installed = true;
2847 case DPIF_OP_FLOW_DEL:
2851 hmap_destroy(&todo);
2855 handle_userspace_upcall(struct ofproto_dpif *ofproto,
2856 struct dpif_upcall *upcall)
2858 struct user_action_cookie cookie;
2859 enum odp_key_fitness fitness;
2860 ovs_be16 initial_tci;
2863 memcpy(&cookie, &upcall->userdata, sizeof(cookie));
2865 fitness = ofproto_dpif_extract_flow_key(ofproto, upcall->key,
2866 upcall->key_len, &flow,
2867 &initial_tci, upcall->packet);
2868 if (fitness == ODP_FIT_ERROR) {
2872 if (cookie.type == USER_ACTION_COOKIE_SFLOW) {
2873 if (ofproto->sflow) {
2874 dpif_sflow_received(ofproto->sflow, upcall->packet, &flow,
2878 VLOG_WARN_RL(&rl, "invalid user cookie : 0x%"PRIx64, upcall->userdata);
2883 handle_upcalls(struct ofproto_dpif *ofproto, unsigned int max_batch)
2885 struct dpif_upcall misses[FLOW_MISS_MAX_BATCH];
2886 struct ofpbuf miss_bufs[FLOW_MISS_MAX_BATCH];
2887 uint64_t miss_buf_stubs[FLOW_MISS_MAX_BATCH][4096 / 8];
2892 assert(max_batch <= FLOW_MISS_MAX_BATCH);
2896 for (n_processed = 0; n_processed < max_batch; n_processed++) {
2897 struct dpif_upcall *upcall = &misses[n_misses];
2898 struct ofpbuf *buf = &miss_bufs[n_misses];
2901 ofpbuf_use_stub(buf, miss_buf_stubs[n_misses],
2902 sizeof miss_buf_stubs[n_misses]);
2903 error = dpif_recv(ofproto->dpif, upcall, buf);
2909 switch (upcall->type) {
2910 case DPIF_UC_ACTION:
2911 handle_userspace_upcall(ofproto, upcall);
2916 /* Handle it later. */
2920 case DPIF_N_UC_TYPES:
2922 VLOG_WARN_RL(&rl, "upcall has unexpected type %"PRIu32,
2928 handle_miss_upcalls(ofproto, misses, n_misses);
2929 for (i = 0; i < n_misses; i++) {
2930 ofpbuf_uninit(&miss_bufs[i]);
2936 /* Flow expiration. */
2938 static int subfacet_max_idle(const struct ofproto_dpif *);
2939 static void update_stats(struct ofproto_dpif *);
2940 static void rule_expire(struct rule_dpif *);
2941 static void expire_subfacets(struct ofproto_dpif *, int dp_max_idle);
2943 /* This function is called periodically by run(). Its job is to collect
2944 * updates for the flows that have been installed into the datapath, most
2945 * importantly when they last were used, and then use that information to
2946 * expire flows that have not been used recently.
2948 * Returns the number of milliseconds after which it should be called again. */
2950 expire(struct ofproto_dpif *ofproto)
2952 struct rule_dpif *rule, *next_rule;
2953 struct oftable *table;
2956 /* Update stats for each flow in the datapath. */
2957 update_stats(ofproto);
2959 /* Expire subfacets that have been idle too long. */
2960 dp_max_idle = subfacet_max_idle(ofproto);
2961 expire_subfacets(ofproto, dp_max_idle);
2963 /* Expire OpenFlow flows whose idle_timeout or hard_timeout has passed. */
2964 OFPROTO_FOR_EACH_TABLE (table, &ofproto->up) {
2965 struct cls_cursor cursor;
2967 cls_cursor_init(&cursor, &table->cls, NULL);
2968 CLS_CURSOR_FOR_EACH_SAFE (rule, next_rule, up.cr, &cursor) {
2973 /* All outstanding data in existing flows has been accounted, so it's a
2974 * good time to do bond rebalancing. */
2975 if (ofproto->has_bonded_bundles) {
2976 struct ofbundle *bundle;
2978 HMAP_FOR_EACH (bundle, hmap_node, &ofproto->bundles) {
2980 bond_rebalance(bundle->bond, &ofproto->revalidate_set);
2985 return MIN(dp_max_idle, 1000);
2988 /* Update 'packet_count', 'byte_count', and 'used' members of installed facets.
2990 * This function also pushes statistics updates to rules which each facet
2991 * resubmits into. Generally these statistics will be accurate. However, if a
2992 * facet changes the rule it resubmits into at some time in between
2993 * update_stats() runs, it is possible that statistics accrued to the
2994 * old rule will be incorrectly attributed to the new rule. This could be
2995 * avoided by calling update_stats() whenever rules are created or
2996 * deleted. However, the performance impact of making so many calls to the
2997 * datapath do not justify the benefit of having perfectly accurate statistics.
3000 update_stats(struct ofproto_dpif *p)
3002 const struct dpif_flow_stats *stats;
3003 struct dpif_flow_dump dump;
3004 const struct nlattr *key;
3007 dpif_flow_dump_start(&dump, p->dpif);
3008 while (dpif_flow_dump_next(&dump, &key, &key_len, NULL, NULL, &stats)) {
3009 struct subfacet *subfacet;
3011 subfacet = subfacet_find(p, key, key_len);
3012 if (subfacet && subfacet->installed) {
3013 struct facet *facet = subfacet->facet;
3015 if (stats->n_packets >= subfacet->dp_packet_count) {
3016 uint64_t extra = stats->n_packets - subfacet->dp_packet_count;
3017 facet->packet_count += extra;
3019 VLOG_WARN_RL(&rl, "unexpected packet count from the datapath");
3022 if (stats->n_bytes >= subfacet->dp_byte_count) {
3023 facet->byte_count += stats->n_bytes - subfacet->dp_byte_count;
3025 VLOG_WARN_RL(&rl, "unexpected byte count from datapath");
3028 subfacet->dp_packet_count = stats->n_packets;
3029 subfacet->dp_byte_count = stats->n_bytes;
3031 facet->tcp_flags |= stats->tcp_flags;
3033 subfacet_update_time(subfacet, stats->used);
3034 if (facet->accounted_bytes < facet->byte_count) {
3036 facet_account(facet);
3037 facet->accounted_bytes = facet->byte_count;
3039 facet_push_stats(facet);
3041 if (!VLOG_DROP_WARN(&rl)) {
3045 odp_flow_key_format(key, key_len, &s);
3046 VLOG_WARN("unexpected flow from datapath %s", ds_cstr(&s));
3050 COVERAGE_INC(facet_unexpected);
3051 /* There's a flow in the datapath that we know nothing about, or a
3052 * flow that shouldn't be installed but was anyway. Delete it. */
3053 dpif_flow_del(p->dpif, key, key_len, NULL);
3056 dpif_flow_dump_done(&dump);
3059 /* Calculates and returns the number of milliseconds of idle time after which
3060 * subfacets should expire from the datapath. When a subfacet expires, we fold
3061 * its statistics into its facet, and when a facet's last subfacet expires, we
3062 * fold its statistic into its rule. */
3064 subfacet_max_idle(const struct ofproto_dpif *ofproto)
3067 * Idle time histogram.
3069 * Most of the time a switch has a relatively small number of subfacets.
3070 * When this is the case we might as well keep statistics for all of them
3071 * in userspace and to cache them in the kernel datapath for performance as
3074 * As the number of subfacets increases, the memory required to maintain
3075 * statistics about them in userspace and in the kernel becomes
3076 * significant. However, with a large number of subfacets it is likely
3077 * that only a few of them are "heavy hitters" that consume a large amount
3078 * of bandwidth. At this point, only heavy hitters are worth caching in
3079 * the kernel and maintaining in userspaces; other subfacets we can
3082 * The technique used to compute the idle time is to build a histogram with
3083 * N_BUCKETS buckets whose width is BUCKET_WIDTH msecs each. Each subfacet
3084 * that is installed in the kernel gets dropped in the appropriate bucket.
3085 * After the histogram has been built, we compute the cutoff so that only
3086 * the most-recently-used 1% of subfacets (but at least
3087 * ofproto->up.flow_eviction_threshold flows) are kept cached. At least
3088 * the most-recently-used bucket of subfacets is kept, so actually an
3089 * arbitrary number of subfacets can be kept in any given expiration run
3090 * (though the next run will delete most of those unless they receive
3093 * This requires a second pass through the subfacets, in addition to the
3094 * pass made by update_stats(), because the former function never looks at
3095 * uninstallable subfacets.
3097 enum { BUCKET_WIDTH = ROUND_UP(100, TIME_UPDATE_INTERVAL) };
3098 enum { N_BUCKETS = 5000 / BUCKET_WIDTH };
3099 int buckets[N_BUCKETS] = { 0 };
3100 int total, subtotal, bucket;
3101 struct subfacet *subfacet;
3105 total = hmap_count(&ofproto->subfacets);
3106 if (total <= ofproto->up.flow_eviction_threshold) {
3107 return N_BUCKETS * BUCKET_WIDTH;
3110 /* Build histogram. */
3112 HMAP_FOR_EACH (subfacet, hmap_node, &ofproto->subfacets) {
3113 long long int idle = now - subfacet->used;
3114 int bucket = (idle <= 0 ? 0
3115 : idle >= BUCKET_WIDTH * N_BUCKETS ? N_BUCKETS - 1
3116 : (unsigned int) idle / BUCKET_WIDTH);
3120 /* Find the first bucket whose flows should be expired. */
3121 subtotal = bucket = 0;
3123 subtotal += buckets[bucket++];
3124 } while (bucket < N_BUCKETS &&
3125 subtotal < MAX(ofproto->up.flow_eviction_threshold, total / 100));
3127 if (VLOG_IS_DBG_ENABLED()) {
3131 ds_put_cstr(&s, "keep");
3132 for (i = 0; i < N_BUCKETS; i++) {
3134 ds_put_cstr(&s, ", drop");
3137 ds_put_format(&s, " %d:%d", i * BUCKET_WIDTH, buckets[i]);
3140 VLOG_INFO("%s: %s (msec:count)", ofproto->up.name, ds_cstr(&s));
3144 return bucket * BUCKET_WIDTH;
3147 enum { EXPIRE_MAX_BATCH = 50 };
3150 expire_batch(struct ofproto_dpif *ofproto, struct subfacet **subfacets, int n)
3152 struct odputil_keybuf keybufs[EXPIRE_MAX_BATCH];
3153 struct dpif_op ops[EXPIRE_MAX_BATCH];
3154 struct dpif_op *opsp[EXPIRE_MAX_BATCH];
3155 struct ofpbuf keys[EXPIRE_MAX_BATCH];
3156 struct dpif_flow_stats stats[EXPIRE_MAX_BATCH];
3159 for (i = 0; i < n; i++) {
3160 ops[i].type = DPIF_OP_FLOW_DEL;
3161 subfacet_get_key(subfacets[i], &keybufs[i], &keys[i]);
3162 ops[i].u.flow_del.key = keys[i].data;
3163 ops[i].u.flow_del.key_len = keys[i].size;
3164 ops[i].u.flow_del.stats = &stats[i];
3168 dpif_operate(ofproto->dpif, opsp, n);
3169 for (i = 0; i < n; i++) {
3170 subfacet_reset_dp_stats(subfacets[i], &stats[i]);
3171 subfacets[i]->installed = false;
3172 subfacet_destroy(subfacets[i]);
3177 expire_subfacets(struct ofproto_dpif *ofproto, int dp_max_idle)
3179 long long int cutoff = time_msec() - dp_max_idle;
3181 struct subfacet *subfacet, *next_subfacet;
3182 struct subfacet *batch[EXPIRE_MAX_BATCH];
3186 HMAP_FOR_EACH_SAFE (subfacet, next_subfacet, hmap_node,
3187 &ofproto->subfacets) {
3188 if (subfacet->used < cutoff) {
3189 if (subfacet->installed) {
3190 batch[n_batch++] = subfacet;
3191 if (n_batch >= EXPIRE_MAX_BATCH) {
3192 expire_batch(ofproto, batch, n_batch);
3196 subfacet_destroy(subfacet);
3202 expire_batch(ofproto, batch, n_batch);
3206 /* If 'rule' is an OpenFlow rule, that has expired according to OpenFlow rules,
3207 * then delete it entirely. */
3209 rule_expire(struct rule_dpif *rule)
3211 struct facet *facet, *next_facet;
3215 /* Has 'rule' expired? */
3217 if (rule->up.hard_timeout
3218 && now > rule->up.modified + rule->up.hard_timeout * 1000) {
3219 reason = OFPRR_HARD_TIMEOUT;
3220 } else if (rule->up.idle_timeout
3221 && now > rule->up.used + rule->up.idle_timeout * 1000) {
3222 reason = OFPRR_IDLE_TIMEOUT;
3227 COVERAGE_INC(ofproto_dpif_expired);
3229 /* Update stats. (This is a no-op if the rule expired due to an idle
3230 * timeout, because that only happens when the rule has no facets left.) */
3231 LIST_FOR_EACH_SAFE (facet, next_facet, list_node, &rule->facets) {
3232 facet_remove(facet);
3235 /* Get rid of the rule. */
3236 ofproto_rule_expire(&rule->up, reason);
3241 /* Creates and returns a new facet owned by 'rule', given a 'flow'.
3243 * The caller must already have determined that no facet with an identical
3244 * 'flow' exists in 'ofproto' and that 'flow' is the best match for 'rule' in
3245 * the ofproto's classifier table.
3247 * 'hash' must be the return value of flow_hash(flow, 0).
3249 * The facet will initially have no subfacets. The caller should create (at
3250 * least) one subfacet with subfacet_create(). */
3251 static struct facet *
3252 facet_create(struct rule_dpif *rule, const struct flow *flow, uint32_t hash)
3254 struct ofproto_dpif *ofproto = ofproto_dpif_cast(rule->up.ofproto);
3255 struct facet *facet;
3257 facet = xzalloc(sizeof *facet);
3258 facet->used = time_msec();
3259 hmap_insert(&ofproto->facets, &facet->hmap_node, hash);
3260 list_push_back(&rule->facets, &facet->list_node);
3262 facet->flow = *flow;
3263 list_init(&facet->subfacets);
3264 netflow_flow_init(&facet->nf_flow);
3265 netflow_flow_update_time(ofproto->netflow, &facet->nf_flow, facet->used);
3271 facet_free(struct facet *facet)
3276 /* Executes, within 'ofproto', the 'n_actions' actions in 'actions' on
3277 * 'packet', which arrived on 'in_port'.
3279 * Takes ownership of 'packet'. */
3281 execute_odp_actions(struct ofproto_dpif *ofproto, const struct flow *flow,
3282 const struct nlattr *odp_actions, size_t actions_len,
3283 struct ofpbuf *packet)
3285 struct odputil_keybuf keybuf;
3289 ofpbuf_use_stack(&key, &keybuf, sizeof keybuf);
3290 odp_flow_key_from_flow(&key, flow);
3292 error = dpif_execute(ofproto->dpif, key.data, key.size,
3293 odp_actions, actions_len, packet);
3295 ofpbuf_delete(packet);
3299 /* Remove 'facet' from 'ofproto' and free up the associated memory:
3301 * - If 'facet' was installed in the datapath, uninstalls it and updates its
3302 * rule's statistics, via subfacet_uninstall().
3304 * - Removes 'facet' from its rule and from ofproto->facets.
3307 facet_remove(struct facet *facet)
3309 struct ofproto_dpif *ofproto = ofproto_dpif_cast(facet->rule->up.ofproto);
3310 struct subfacet *subfacet, *next_subfacet;
3312 assert(!list_is_empty(&facet->subfacets));
3314 /* First uninstall all of the subfacets to get final statistics. */
3315 LIST_FOR_EACH (subfacet, list_node, &facet->subfacets) {
3316 subfacet_uninstall(subfacet);
3319 /* Flush the final stats to the rule.
3321 * This might require us to have at least one subfacet around so that we
3322 * can use its actions for accounting in facet_account(), which is why we
3323 * have uninstalled but not yet destroyed the subfacets. */
3324 facet_flush_stats(facet);
3326 /* Now we're really all done so destroy everything. */
3327 LIST_FOR_EACH_SAFE (subfacet, next_subfacet, list_node,
3328 &facet->subfacets) {
3329 subfacet_destroy__(subfacet);
3331 hmap_remove(&ofproto->facets, &facet->hmap_node);
3332 list_remove(&facet->list_node);
3336 /* Feed information from 'facet' back into the learning table to keep it in
3337 * sync with what is actually flowing through the datapath. */
3339 facet_learn(struct facet *facet)
3341 struct ofproto_dpif *ofproto = ofproto_dpif_cast(facet->rule->up.ofproto);
3342 struct action_xlate_ctx ctx;
3344 if (!facet->has_learn
3345 && !facet->has_normal
3346 && (!facet->has_fin_timeout
3347 || !(facet->tcp_flags & (TCP_FIN | TCP_RST)))) {
3351 action_xlate_ctx_init(&ctx, ofproto, &facet->flow,
3352 facet->flow.vlan_tci,
3353 facet->rule, facet->tcp_flags, NULL);
3354 ctx.may_learn = true;
3355 xlate_actions_for_side_effects(&ctx, facet->rule->up.actions,
3356 facet->rule->up.n_actions);
3360 facet_account(struct facet *facet)
3362 struct ofproto_dpif *ofproto = ofproto_dpif_cast(facet->rule->up.ofproto);
3363 struct subfacet *subfacet;
3364 const struct nlattr *a;
3369 if (!facet->has_normal || !ofproto->has_bonded_bundles) {
3372 n_bytes = facet->byte_count - facet->accounted_bytes;
3374 /* This loop feeds byte counters to bond_account() for rebalancing to use
3375 * as a basis. We also need to track the actual VLAN on which the packet
3376 * is going to be sent to ensure that it matches the one passed to
3377 * bond_choose_output_slave(). (Otherwise, we will account to the wrong
3380 * We use the actions from an arbitrary subfacet because they should all
3381 * be equally valid for our purpose. */
3382 subfacet = CONTAINER_OF(list_front(&facet->subfacets),
3383 struct subfacet, list_node);
3384 vlan_tci = facet->flow.vlan_tci;
3385 NL_ATTR_FOR_EACH_UNSAFE (a, left,
3386 subfacet->actions, subfacet->actions_len) {
3387 const struct ovs_action_push_vlan *vlan;
3388 struct ofport_dpif *port;
3390 switch (nl_attr_type(a)) {
3391 case OVS_ACTION_ATTR_OUTPUT:
3392 port = get_odp_port(ofproto, nl_attr_get_u32(a));
3393 if (port && port->bundle && port->bundle->bond) {
3394 bond_account(port->bundle->bond, &facet->flow,
3395 vlan_tci_to_vid(vlan_tci), n_bytes);
3399 case OVS_ACTION_ATTR_POP_VLAN:
3400 vlan_tci = htons(0);
3403 case OVS_ACTION_ATTR_PUSH_VLAN:
3404 vlan = nl_attr_get(a);
3405 vlan_tci = vlan->vlan_tci;
3411 /* Returns true if the only action for 'facet' is to send to the controller.
3412 * (We don't report NetFlow expiration messages for such facets because they
3413 * are just part of the control logic for the network, not real traffic). */
3415 facet_is_controller_flow(struct facet *facet)
3418 && facet->rule->up.n_actions == 1
3419 && action_outputs_to_port(&facet->rule->up.actions[0],
3420 htons(OFPP_CONTROLLER)));
3423 /* Folds all of 'facet''s statistics into its rule. Also updates the
3424 * accounting ofhook and emits a NetFlow expiration if appropriate. All of
3425 * 'facet''s statistics in the datapath should have been zeroed and folded into
3426 * its packet and byte counts before this function is called. */
3428 facet_flush_stats(struct facet *facet)
3430 struct ofproto_dpif *ofproto = ofproto_dpif_cast(facet->rule->up.ofproto);
3431 struct subfacet *subfacet;
3433 LIST_FOR_EACH (subfacet, list_node, &facet->subfacets) {
3434 assert(!subfacet->dp_byte_count);
3435 assert(!subfacet->dp_packet_count);
3438 facet_push_stats(facet);
3439 if (facet->accounted_bytes < facet->byte_count) {
3440 facet_account(facet);
3441 facet->accounted_bytes = facet->byte_count;
3444 if (ofproto->netflow && !facet_is_controller_flow(facet)) {
3445 struct ofexpired expired;
3446 expired.flow = facet->flow;
3447 expired.packet_count = facet->packet_count;
3448 expired.byte_count = facet->byte_count;
3449 expired.used = facet->used;
3450 netflow_expire(ofproto->netflow, &facet->nf_flow, &expired);
3453 facet->rule->packet_count += facet->packet_count;
3454 facet->rule->byte_count += facet->byte_count;
3456 /* Reset counters to prevent double counting if 'facet' ever gets
3458 facet_reset_counters(facet);
3460 netflow_flow_clear(&facet->nf_flow);
3461 facet->tcp_flags = 0;
3464 /* Searches 'ofproto''s table of facets for one exactly equal to 'flow'.
3465 * Returns it if found, otherwise a null pointer.
3467 * 'hash' must be the return value of flow_hash(flow, 0).
3469 * The returned facet might need revalidation; use facet_lookup_valid()
3470 * instead if that is important. */
3471 static struct facet *
3472 facet_find(struct ofproto_dpif *ofproto,
3473 const struct flow *flow, uint32_t hash)
3475 struct facet *facet;
3477 HMAP_FOR_EACH_WITH_HASH (facet, hmap_node, hash, &ofproto->facets) {
3478 if (flow_equal(flow, &facet->flow)) {
3486 /* Searches 'ofproto''s table of facets for one exactly equal to 'flow'.
3487 * Returns it if found, otherwise a null pointer.
3489 * 'hash' must be the return value of flow_hash(flow, 0).
3491 * The returned facet is guaranteed to be valid. */
3492 static struct facet *
3493 facet_lookup_valid(struct ofproto_dpif *ofproto, const struct flow *flow,
3496 struct facet *facet = facet_find(ofproto, flow, hash);
3498 /* The facet we found might not be valid, since we could be in need of
3499 * revalidation. If it is not valid, don't return it. */
3501 && (ofproto->need_revalidate
3502 || tag_set_intersects(&ofproto->revalidate_set, facet->tags))
3503 && !facet_revalidate(facet)) {
3504 COVERAGE_INC(facet_invalidated);
3512 facet_check_consistency(struct facet *facet)
3514 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 15);
3516 struct ofproto_dpif *ofproto = ofproto_dpif_cast(facet->rule->up.ofproto);
3518 uint64_t odp_actions_stub[1024 / 8];
3519 struct ofpbuf odp_actions;
3521 struct rule_dpif *rule;
3522 struct subfacet *subfacet;
3523 bool may_log = false;
3526 /* Check the rule for consistency. */
3527 rule = rule_dpif_lookup(ofproto, &facet->flow, 0);
3529 if (!VLOG_DROP_WARN(&rl)) {
3530 char *s = flow_to_string(&facet->flow);
3531 VLOG_WARN("%s: facet should not exist", s);
3535 } else if (rule != facet->rule) {
3536 may_log = !VLOG_DROP_WARN(&rl);
3542 flow_format(&s, &facet->flow);
3543 ds_put_format(&s, ": facet associated with wrong rule (was "
3544 "table=%"PRIu8",", facet->rule->up.table_id);
3545 cls_rule_format(&facet->rule->up.cr, &s);
3546 ds_put_format(&s, ") (should have been table=%"PRIu8",",
3548 cls_rule_format(&rule->up.cr, &s);
3549 ds_put_char(&s, ')');
3551 VLOG_WARN("%s", ds_cstr(&s));
3558 /* Check the datapath actions for consistency. */
3559 ofpbuf_use_stub(&odp_actions, odp_actions_stub, sizeof odp_actions_stub);
3560 LIST_FOR_EACH (subfacet, list_node, &facet->subfacets) {
3561 struct action_xlate_ctx ctx;
3562 bool actions_changed;
3563 bool should_install;
3565 action_xlate_ctx_init(&ctx, ofproto, &facet->flow,
3566 subfacet->initial_tci, rule, 0, NULL);
3567 xlate_actions(&ctx, rule->up.actions, rule->up.n_actions,
3570 should_install = (ctx.may_set_up_flow
3571 && subfacet->key_fitness != ODP_FIT_TOO_LITTLE);
3572 if (!should_install && !subfacet->installed) {
3573 /* The actions for uninstallable flows may vary from one packet to
3574 * the next, so don't compare the actions. */
3578 actions_changed = (subfacet->actions_len != odp_actions.size
3579 || memcmp(subfacet->actions, odp_actions.data,
3580 subfacet->actions_len));
3581 if (should_install != subfacet->installed || actions_changed) {
3583 may_log = !VLOG_DROP_WARN(&rl);
3588 struct odputil_keybuf keybuf;
3593 subfacet_get_key(subfacet, &keybuf, &key);
3594 odp_flow_key_format(key.data, key.size, &s);
3596 ds_put_cstr(&s, ": inconsistency in subfacet");
3597 if (should_install != subfacet->installed) {
3598 enum odp_key_fitness fitness = subfacet->key_fitness;
3600 ds_put_format(&s, " (should%s have been installed)",
3601 should_install ? "" : " not");
3602 ds_put_format(&s, " (may_set_up_flow=%s, fitness=%s)",
3603 ctx.may_set_up_flow ? "true" : "false",
3604 odp_key_fitness_to_string(fitness));
3606 if (actions_changed) {
3607 ds_put_cstr(&s, " (actions were: ");
3608 format_odp_actions(&s, subfacet->actions,
3609 subfacet->actions_len);
3610 ds_put_cstr(&s, ") (correct actions: ");
3611 format_odp_actions(&s, odp_actions.data, odp_actions.size);
3612 ds_put_char(&s, ')');
3614 ds_put_cstr(&s, " (actions: ");
3615 format_odp_actions(&s, subfacet->actions,
3616 subfacet->actions_len);
3617 ds_put_char(&s, ')');
3619 VLOG_WARN("%s", ds_cstr(&s));
3624 ofpbuf_uninit(&odp_actions);
3629 /* Re-searches the classifier for 'facet':
3631 * - If the rule found is different from 'facet''s current rule, moves
3632 * 'facet' to the new rule and recompiles its actions.
3634 * - If the rule found is the same as 'facet''s current rule, leaves 'facet'
3635 * where it is and recompiles its actions anyway.
3637 * - If there is none, destroys 'facet'.
3639 * Returns true if 'facet' still exists, false if it has been destroyed. */
3641 facet_revalidate(struct facet *facet)
3643 struct ofproto_dpif *ofproto = ofproto_dpif_cast(facet->rule->up.ofproto);
3645 struct nlattr *odp_actions;
3648 struct actions *new_actions;
3650 struct action_xlate_ctx ctx;
3651 uint64_t odp_actions_stub[1024 / 8];
3652 struct ofpbuf odp_actions;
3654 struct rule_dpif *new_rule;
3655 struct subfacet *subfacet;
3656 bool actions_changed;
3659 COVERAGE_INC(facet_revalidate);
3661 /* Determine the new rule. */
3662 new_rule = rule_dpif_lookup(ofproto, &facet->flow, 0);
3664 /* No new rule, so delete the facet. */
3665 facet_remove(facet);
3669 /* Calculate new datapath actions.
3671 * We do not modify any 'facet' state yet, because we might need to, e.g.,
3672 * emit a NetFlow expiration and, if so, we need to have the old state
3673 * around to properly compose it. */
3675 /* If the datapath actions changed or the installability changed,
3676 * then we need to talk to the datapath. */
3679 memset(&ctx, 0, sizeof ctx);
3680 ofpbuf_use_stub(&odp_actions, odp_actions_stub, sizeof odp_actions_stub);
3681 LIST_FOR_EACH (subfacet, list_node, &facet->subfacets) {
3682 bool should_install;
3684 action_xlate_ctx_init(&ctx, ofproto, &facet->flow,
3685 subfacet->initial_tci, new_rule, 0, NULL);
3686 xlate_actions(&ctx, new_rule->up.actions, new_rule->up.n_actions,
3688 actions_changed = (subfacet->actions_len != odp_actions.size
3689 || memcmp(subfacet->actions, odp_actions.data,
3690 subfacet->actions_len));
3692 should_install = (ctx.may_set_up_flow
3693 && subfacet->key_fitness != ODP_FIT_TOO_LITTLE);
3694 if (actions_changed || should_install != subfacet->installed) {
3695 if (should_install) {
3696 struct dpif_flow_stats stats;
3698 subfacet_install(subfacet,
3699 odp_actions.data, odp_actions.size, &stats);
3700 subfacet_update_stats(subfacet, &stats);
3702 subfacet_uninstall(subfacet);
3706 new_actions = xcalloc(list_size(&facet->subfacets),
3707 sizeof *new_actions);
3709 new_actions[i].odp_actions = xmemdup(odp_actions.data,
3711 new_actions[i].actions_len = odp_actions.size;
3716 ofpbuf_uninit(&odp_actions);
3719 facet_flush_stats(facet);
3722 /* Update 'facet' now that we've taken care of all the old state. */
3723 facet->tags = ctx.tags;
3724 facet->nf_flow.output_iface = ctx.nf_output_iface;
3725 facet->may_install = ctx.may_set_up_flow;
3726 facet->has_learn = ctx.has_learn;
3727 facet->has_normal = ctx.has_normal;
3728 facet->has_fin_timeout = ctx.has_fin_timeout;
3729 facet->mirrors = ctx.mirrors;
3732 LIST_FOR_EACH (subfacet, list_node, &facet->subfacets) {
3733 if (new_actions[i].odp_actions) {
3734 free(subfacet->actions);
3735 subfacet->actions = new_actions[i].odp_actions;
3736 subfacet->actions_len = new_actions[i].actions_len;
3742 if (facet->rule != new_rule) {
3743 COVERAGE_INC(facet_changed_rule);
3744 list_remove(&facet->list_node);
3745 list_push_back(&new_rule->facets, &facet->list_node);
3746 facet->rule = new_rule;
3747 facet->used = new_rule->up.created;
3748 facet->prev_used = facet->used;
3754 /* Updates 'facet''s used time. Caller is responsible for calling
3755 * facet_push_stats() to update the flows which 'facet' resubmits into. */
3757 facet_update_time(struct facet *facet, long long int used)
3759 struct ofproto_dpif *ofproto = ofproto_dpif_cast(facet->rule->up.ofproto);
3760 if (used > facet->used) {
3762 ofproto_rule_update_used(&facet->rule->up, used);
3763 netflow_flow_update_time(ofproto->netflow, &facet->nf_flow, used);
3768 facet_reset_counters(struct facet *facet)
3770 facet->packet_count = 0;
3771 facet->byte_count = 0;
3772 facet->prev_packet_count = 0;
3773 facet->prev_byte_count = 0;
3774 facet->accounted_bytes = 0;
3778 facet_push_stats(struct facet *facet)
3780 struct dpif_flow_stats stats;
3782 assert(facet->packet_count >= facet->prev_packet_count);
3783 assert(facet->byte_count >= facet->prev_byte_count);
3784 assert(facet->used >= facet->prev_used);
3786 stats.n_packets = facet->packet_count - facet->prev_packet_count;
3787 stats.n_bytes = facet->byte_count - facet->prev_byte_count;
3788 stats.used = facet->used;
3789 stats.tcp_flags = 0;
3791 if (stats.n_packets || stats.n_bytes || facet->used > facet->prev_used) {
3792 facet->prev_packet_count = facet->packet_count;
3793 facet->prev_byte_count = facet->byte_count;
3794 facet->prev_used = facet->used;
3796 flow_push_stats(facet->rule, &facet->flow, &stats);
3798 update_mirror_stats(ofproto_dpif_cast(facet->rule->up.ofproto),
3799 facet->mirrors, stats.n_packets, stats.n_bytes);
3804 rule_credit_stats(struct rule_dpif *rule, const struct dpif_flow_stats *stats)
3806 rule->packet_count += stats->n_packets;
3807 rule->byte_count += stats->n_bytes;
3808 ofproto_rule_update_used(&rule->up, stats->used);
3811 /* Pushes flow statistics to the rules which 'flow' resubmits into given
3812 * 'rule''s actions and mirrors. */
3814 flow_push_stats(struct rule_dpif *rule,
3815 const struct flow *flow, const struct dpif_flow_stats *stats)
3817 struct ofproto_dpif *ofproto = ofproto_dpif_cast(rule->up.ofproto);
3818 struct action_xlate_ctx ctx;
3820 ofproto_rule_update_used(&rule->up, stats->used);
3822 action_xlate_ctx_init(&ctx, ofproto, flow, flow->vlan_tci, rule,
3824 ctx.resubmit_stats = stats;
3825 xlate_actions_for_side_effects(&ctx, rule->up.actions, rule->up.n_actions);
3830 static struct subfacet *
3831 subfacet_find__(struct ofproto_dpif *ofproto,
3832 const struct nlattr *key, size_t key_len, uint32_t key_hash,
3833 const struct flow *flow)
3835 struct subfacet *subfacet;
3837 HMAP_FOR_EACH_WITH_HASH (subfacet, hmap_node, key_hash,
3838 &ofproto->subfacets) {
3840 ? (subfacet->key_len == key_len
3841 && !memcmp(key, subfacet->key, key_len))
3842 : flow_equal(flow, &subfacet->facet->flow)) {
3850 /* Searches 'facet' (within 'ofproto') for a subfacet with the specified
3851 * 'key_fitness', 'key', and 'key_len'. Returns the existing subfacet if
3852 * there is one, otherwise creates and returns a new subfacet.
3854 * If the returned subfacet is new, then subfacet->actions will be NULL, in
3855 * which case the caller must populate the actions with
3856 * subfacet_make_actions(). */
3857 static struct subfacet *
3858 subfacet_create(struct facet *facet, enum odp_key_fitness key_fitness,
3859 const struct nlattr *key, size_t key_len, ovs_be16 initial_tci)
3861 struct ofproto_dpif *ofproto = ofproto_dpif_cast(facet->rule->up.ofproto);
3862 uint32_t key_hash = odp_flow_key_hash(key, key_len);
3863 struct subfacet *subfacet;
3865 subfacet = subfacet_find__(ofproto, key, key_len, key_hash, &facet->flow);
3867 if (subfacet->facet == facet) {
3871 /* This shouldn't happen. */
3872 VLOG_ERR_RL(&rl, "subfacet with wrong facet");
3873 subfacet_destroy(subfacet);
3876 subfacet = (list_is_empty(&facet->subfacets)
3877 ? &facet->one_subfacet
3878 : xmalloc(sizeof *subfacet));
3879 hmap_insert(&ofproto->subfacets, &subfacet->hmap_node, key_hash);
3880 list_push_back(&facet->subfacets, &subfacet->list_node);
3881 subfacet->facet = facet;
3882 subfacet->key_fitness = key_fitness;
3883 if (key_fitness != ODP_FIT_PERFECT) {
3884 subfacet->key = xmemdup(key, key_len);
3885 subfacet->key_len = key_len;
3887 subfacet->key = NULL;
3888 subfacet->key_len = 0;
3890 subfacet->used = time_msec();
3891 subfacet->dp_packet_count = 0;
3892 subfacet->dp_byte_count = 0;
3893 subfacet->actions_len = 0;
3894 subfacet->actions = NULL;
3895 subfacet->installed = false;
3896 subfacet->initial_tci = initial_tci;
3901 /* Searches 'ofproto' for a subfacet with the given 'key', 'key_len', and
3902 * 'flow'. Returns the subfacet if one exists, otherwise NULL. */
3903 static struct subfacet *
3904 subfacet_find(struct ofproto_dpif *ofproto,
3905 const struct nlattr *key, size_t key_len)
3907 uint32_t key_hash = odp_flow_key_hash(key, key_len);
3908 enum odp_key_fitness fitness;
3911 fitness = odp_flow_key_to_flow(key, key_len, &flow);
3912 if (fitness == ODP_FIT_ERROR) {
3916 return subfacet_find__(ofproto, key, key_len, key_hash, &flow);
3919 /* Uninstalls 'subfacet' from the datapath, if it is installed, removes it from
3920 * its facet within 'ofproto', and frees it. */
3922 subfacet_destroy__(struct subfacet *subfacet)
3924 struct facet *facet = subfacet->facet;
3925 struct ofproto_dpif *ofproto = ofproto_dpif_cast(facet->rule->up.ofproto);
3927 subfacet_uninstall(subfacet);
3928 hmap_remove(&ofproto->subfacets, &subfacet->hmap_node);
3929 list_remove(&subfacet->list_node);
3930 free(subfacet->key);
3931 free(subfacet->actions);
3932 if (subfacet != &facet->one_subfacet) {
3937 /* Destroys 'subfacet', as with subfacet_destroy__(), and then if this was the
3938 * last remaining subfacet in its facet destroys the facet too. */
3940 subfacet_destroy(struct subfacet *subfacet)
3942 struct facet *facet = subfacet->facet;
3944 if (list_is_singleton(&facet->subfacets)) {
3945 /* facet_remove() needs at least one subfacet (it will remove it). */
3946 facet_remove(facet);
3948 subfacet_destroy__(subfacet);
3952 /* Initializes 'key' with the sequence of OVS_KEY_ATTR_* Netlink attributes
3953 * that can be used to refer to 'subfacet'. The caller must provide 'keybuf'
3954 * for use as temporary storage. */
3956 subfacet_get_key(struct subfacet *subfacet, struct odputil_keybuf *keybuf,
3959 if (!subfacet->key) {
3960 ofpbuf_use_stack(key, keybuf, sizeof *keybuf);
3961 odp_flow_key_from_flow(key, &subfacet->facet->flow);
3963 ofpbuf_use_const(key, subfacet->key, subfacet->key_len);
3967 /* Composes the datapath actions for 'subfacet' based on its rule's actions. */
3969 subfacet_make_actions(struct subfacet *subfacet, const struct ofpbuf *packet)
3971 struct facet *facet = subfacet->facet;
3972 struct rule_dpif *rule = facet->rule;
3973 struct ofproto_dpif *ofproto = ofproto_dpif_cast(rule->up.ofproto);
3975 struct action_xlate_ctx ctx;
3976 uint64_t odp_actions_stub[1024 / 8];
3977 struct ofpbuf odp_actions;
3979 ofpbuf_use_stub(&odp_actions, odp_actions_stub, sizeof odp_actions_stub);
3980 action_xlate_ctx_init(&ctx, ofproto, &facet->flow, subfacet->initial_tci,
3982 xlate_actions(&ctx, rule->up.actions, rule->up.n_actions, &odp_actions);
3983 facet->tags = ctx.tags;
3984 facet->may_install = ctx.may_set_up_flow;
3985 facet->has_learn = ctx.has_learn;
3986 facet->has_normal = ctx.has_normal;
3987 facet->has_fin_timeout = ctx.has_fin_timeout;
3988 facet->nf_flow.output_iface = ctx.nf_output_iface;
3989 facet->mirrors = ctx.mirrors;
3991 if (subfacet->actions_len != odp_actions.size
3992 || memcmp(subfacet->actions, odp_actions.data, odp_actions.size)) {
3993 free(subfacet->actions);
3994 subfacet->actions_len = odp_actions.size;
3995 subfacet->actions = xmemdup(odp_actions.data, odp_actions.size);
3998 ofpbuf_uninit(&odp_actions);
4001 /* Updates 'subfacet''s datapath flow, setting its actions to 'actions_len'
4002 * bytes of actions in 'actions'. If 'stats' is non-null, statistics counters
4003 * in the datapath will be zeroed and 'stats' will be updated with traffic new
4004 * since 'subfacet' was last updated.
4006 * Returns 0 if successful, otherwise a positive errno value. */
4008 subfacet_install(struct subfacet *subfacet,
4009 const struct nlattr *actions, size_t actions_len,
4010 struct dpif_flow_stats *stats)
4012 struct facet *facet = subfacet->facet;
4013 struct ofproto_dpif *ofproto = ofproto_dpif_cast(facet->rule->up.ofproto);
4014 struct odputil_keybuf keybuf;
4015 enum dpif_flow_put_flags flags;
4019 flags = DPIF_FP_CREATE | DPIF_FP_MODIFY;
4021 flags |= DPIF_FP_ZERO_STATS;
4024 subfacet_get_key(subfacet, &keybuf, &key);
4025 ret = dpif_flow_put(ofproto->dpif, flags, key.data, key.size,
4026 actions, actions_len, stats);
4029 subfacet_reset_dp_stats(subfacet, stats);
4035 /* If 'subfacet' is installed in the datapath, uninstalls it. */
4037 subfacet_uninstall(struct subfacet *subfacet)
4039 if (subfacet->installed) {
4040 struct rule_dpif *rule = subfacet->facet->rule;
4041 struct ofproto_dpif *ofproto = ofproto_dpif_cast(rule->up.ofproto);
4042 struct odputil_keybuf keybuf;
4043 struct dpif_flow_stats stats;
4047 subfacet_get_key(subfacet, &keybuf, &key);
4048 error = dpif_flow_del(ofproto->dpif, key.data, key.size, &stats);
4049 subfacet_reset_dp_stats(subfacet, &stats);
4051 subfacet_update_stats(subfacet, &stats);
4053 subfacet->installed = false;
4055 assert(subfacet->dp_packet_count == 0);
4056 assert(subfacet->dp_byte_count == 0);
4060 /* Resets 'subfacet''s datapath statistics counters. This should be called
4061 * when 'subfacet''s statistics are cleared in the datapath. If 'stats' is
4062 * non-null, it should contain the statistics returned by dpif when 'subfacet'
4063 * was reset in the datapath. 'stats' will be modified to include only
4064 * statistics new since 'subfacet' was last updated. */
4066 subfacet_reset_dp_stats(struct subfacet *subfacet,
4067 struct dpif_flow_stats *stats)
4070 && subfacet->dp_packet_count <= stats->n_packets
4071 && subfacet->dp_byte_count <= stats->n_bytes) {
4072 stats->n_packets -= subfacet->dp_packet_count;
4073 stats->n_bytes -= subfacet->dp_byte_count;
4076 subfacet->dp_packet_count = 0;
4077 subfacet->dp_byte_count = 0;
4080 /* Updates 'subfacet''s used time. The caller is responsible for calling
4081 * facet_push_stats() to update the flows which 'subfacet' resubmits into. */
4083 subfacet_update_time(struct subfacet *subfacet, long long int used)
4085 if (used > subfacet->used) {
4086 subfacet->used = used;
4087 facet_update_time(subfacet->facet, used);
4091 /* Folds the statistics from 'stats' into the counters in 'subfacet'.
4093 * Because of the meaning of a subfacet's counters, it only makes sense to do
4094 * this if 'stats' are not tracked in the datapath, that is, if 'stats'
4095 * represents a packet that was sent by hand or if it represents statistics
4096 * that have been cleared out of the datapath. */
4098 subfacet_update_stats(struct subfacet *subfacet,
4099 const struct dpif_flow_stats *stats)
4101 if (stats->n_packets || stats->used > subfacet->used) {
4102 struct facet *facet = subfacet->facet;
4104 subfacet_update_time(subfacet, stats->used);
4105 facet->packet_count += stats->n_packets;
4106 facet->byte_count += stats->n_bytes;
4107 facet->tcp_flags |= stats->tcp_flags;
4108 facet_push_stats(facet);
4109 netflow_flow_update_flags(&facet->nf_flow, stats->tcp_flags);
4115 static struct rule_dpif *
4116 rule_dpif_lookup(struct ofproto_dpif *ofproto, const struct flow *flow,
4119 struct cls_rule *cls_rule;
4120 struct classifier *cls;
4122 if (table_id >= N_TABLES) {
4126 cls = &ofproto->up.tables[table_id].cls;
4127 if (flow->nw_frag & FLOW_NW_FRAG_ANY
4128 && ofproto->up.frag_handling == OFPC_FRAG_NORMAL) {
4129 /* For OFPC_NORMAL frag_handling, we must pretend that transport ports
4130 * are unavailable. */
4131 struct flow ofpc_normal_flow = *flow;
4132 ofpc_normal_flow.tp_src = htons(0);
4133 ofpc_normal_flow.tp_dst = htons(0);
4134 cls_rule = classifier_lookup(cls, &ofpc_normal_flow);
4136 cls_rule = classifier_lookup(cls, flow);
4138 return rule_dpif_cast(rule_from_cls_rule(cls_rule));
4142 complete_operation(struct rule_dpif *rule)
4144 struct ofproto_dpif *ofproto = ofproto_dpif_cast(rule->up.ofproto);
4146 rule_invalidate(rule);
4148 struct dpif_completion *c = xmalloc(sizeof *c);
4149 c->op = rule->up.pending;
4150 list_push_back(&ofproto->completions, &c->list_node);
4152 ofoperation_complete(rule->up.pending, 0);
4156 static struct rule *
4159 struct rule_dpif *rule = xmalloc(sizeof *rule);
4164 rule_dealloc(struct rule *rule_)
4166 struct rule_dpif *rule = rule_dpif_cast(rule_);
4171 rule_construct(struct rule *rule_)
4173 struct rule_dpif *rule = rule_dpif_cast(rule_);
4174 struct ofproto_dpif *ofproto = ofproto_dpif_cast(rule->up.ofproto);
4175 struct rule_dpif *victim;
4179 error = validate_actions(rule->up.actions, rule->up.n_actions,
4180 &rule->up.cr.flow, ofproto->max_ports);
4185 rule->packet_count = 0;
4186 rule->byte_count = 0;
4188 victim = rule_dpif_cast(ofoperation_get_victim(rule->up.pending));
4189 if (victim && !list_is_empty(&victim->facets)) {
4190 struct facet *facet;
4192 rule->facets = victim->facets;
4193 list_moved(&rule->facets);
4194 LIST_FOR_EACH (facet, list_node, &rule->facets) {
4195 /* XXX: We're only clearing our local counters here. It's possible
4196 * that quite a few packets are unaccounted for in the datapath
4197 * statistics. These will be accounted to the new rule instead of
4198 * cleared as required. This could be fixed by clearing out the
4199 * datapath statistics for this facet, but currently it doesn't
4201 facet_reset_counters(facet);
4205 /* Must avoid list_moved() in this case. */
4206 list_init(&rule->facets);
4209 table_id = rule->up.table_id;
4210 rule->tag = (victim ? victim->tag
4212 : rule_calculate_tag(&rule->up.cr.flow, &rule->up.cr.wc,
4213 ofproto->tables[table_id].basis));
4215 complete_operation(rule);
4220 rule_destruct(struct rule *rule_)
4222 struct rule_dpif *rule = rule_dpif_cast(rule_);
4223 struct facet *facet, *next_facet;
4225 LIST_FOR_EACH_SAFE (facet, next_facet, list_node, &rule->facets) {
4226 facet_revalidate(facet);
4229 complete_operation(rule);
4233 rule_get_stats(struct rule *rule_, uint64_t *packets, uint64_t *bytes)
4235 struct rule_dpif *rule = rule_dpif_cast(rule_);
4236 struct facet *facet;
4238 /* Start from historical data for 'rule' itself that are no longer tracked
4239 * in facets. This counts, for example, facets that have expired. */
4240 *packets = rule->packet_count;
4241 *bytes = rule->byte_count;
4243 /* Add any statistics that are tracked by facets. This includes
4244 * statistical data recently updated by ofproto_update_stats() as well as
4245 * stats for packets that were executed "by hand" via dpif_execute(). */
4246 LIST_FOR_EACH (facet, list_node, &rule->facets) {
4247 *packets += facet->packet_count;
4248 *bytes += facet->byte_count;
4253 rule_execute(struct rule *rule_, const struct flow *flow,
4254 struct ofpbuf *packet)
4256 struct rule_dpif *rule = rule_dpif_cast(rule_);
4257 struct ofproto_dpif *ofproto = ofproto_dpif_cast(rule->up.ofproto);
4259 struct dpif_flow_stats stats;
4261 struct action_xlate_ctx ctx;
4262 uint64_t odp_actions_stub[1024 / 8];
4263 struct ofpbuf odp_actions;
4265 dpif_flow_stats_extract(flow, packet, &stats);
4266 rule_credit_stats(rule, &stats);
4268 ofpbuf_use_stub(&odp_actions, odp_actions_stub, sizeof odp_actions_stub);
4269 action_xlate_ctx_init(&ctx, ofproto, flow, flow->vlan_tci,
4270 rule, stats.tcp_flags, packet);
4271 ctx.resubmit_stats = &stats;
4272 xlate_actions(&ctx, rule->up.actions, rule->up.n_actions, &odp_actions);
4274 execute_odp_actions(ofproto, flow, odp_actions.data,
4275 odp_actions.size, packet);
4277 ofpbuf_uninit(&odp_actions);
4283 rule_modify_actions(struct rule *rule_)
4285 struct rule_dpif *rule = rule_dpif_cast(rule_);
4286 struct ofproto_dpif *ofproto = ofproto_dpif_cast(rule->up.ofproto);
4289 error = validate_actions(rule->up.actions, rule->up.n_actions,
4290 &rule->up.cr.flow, ofproto->max_ports);
4292 ofoperation_complete(rule->up.pending, error);
4296 complete_operation(rule);
4299 /* Sends 'packet' out 'ofport'.
4300 * May modify 'packet'.
4301 * Returns 0 if successful, otherwise a positive errno value. */
4303 send_packet(const struct ofport_dpif *ofport, struct ofpbuf *packet)
4305 const struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport->up.ofproto);
4306 struct ofpbuf key, odp_actions;
4307 struct odputil_keybuf keybuf;
4312 flow_extract((struct ofpbuf *) packet, 0, 0, 0, &flow);
4313 odp_port = vsp_realdev_to_vlandev(ofproto, ofport->odp_port,
4315 if (odp_port != ofport->odp_port) {
4316 eth_pop_vlan(packet);
4317 flow.vlan_tci = htons(0);
4320 ofpbuf_use_stack(&key, &keybuf, sizeof keybuf);
4321 odp_flow_key_from_flow(&key, &flow);
4323 ofpbuf_init(&odp_actions, 32);
4324 compose_sflow_action(ofproto, &odp_actions, &flow, odp_port);
4326 nl_msg_put_u32(&odp_actions, OVS_ACTION_ATTR_OUTPUT, odp_port);
4327 error = dpif_execute(ofproto->dpif,
4329 odp_actions.data, odp_actions.size,
4331 ofpbuf_uninit(&odp_actions);
4334 VLOG_WARN_RL(&rl, "%s: failed to send packet on port %"PRIu32" (%s)",
4335 ofproto->up.name, odp_port, strerror(error));
4337 ofproto_update_local_port_stats(ofport->up.ofproto, packet->size, 0);
4341 /* OpenFlow to datapath action translation. */
4343 static void do_xlate_actions(const union ofp_action *in, size_t n_in,
4344 struct action_xlate_ctx *ctx);
4345 static void xlate_normal(struct action_xlate_ctx *);
4348 put_userspace_action(const struct ofproto_dpif *ofproto,
4349 struct ofpbuf *odp_actions,
4350 const struct flow *flow,
4351 const struct user_action_cookie *cookie)
4355 pid = dpif_port_get_pid(ofproto->dpif,
4356 ofp_port_to_odp_port(flow->in_port));
4358 return odp_put_userspace_action(pid, cookie, odp_actions);
4361 /* Compose SAMPLE action for sFlow. */
4363 compose_sflow_action(const struct ofproto_dpif *ofproto,
4364 struct ofpbuf *odp_actions,
4365 const struct flow *flow,
4368 uint32_t port_ifindex;
4369 uint32_t probability;
4370 struct user_action_cookie cookie;
4371 size_t sample_offset, actions_offset;
4372 int cookie_offset, n_output;
4374 if (!ofproto->sflow || flow->in_port == OFPP_NONE) {
4378 if (odp_port == OVSP_NONE) {
4382 port_ifindex = dpif_sflow_odp_port_to_ifindex(ofproto->sflow, odp_port);
4386 sample_offset = nl_msg_start_nested(odp_actions, OVS_ACTION_ATTR_SAMPLE);
4388 /* Number of packets out of UINT_MAX to sample. */
4389 probability = dpif_sflow_get_probability(ofproto->sflow);
4390 nl_msg_put_u32(odp_actions, OVS_SAMPLE_ATTR_PROBABILITY, probability);
4392 actions_offset = nl_msg_start_nested(odp_actions, OVS_SAMPLE_ATTR_ACTIONS);
4394 cookie.type = USER_ACTION_COOKIE_SFLOW;
4395 cookie.data = port_ifindex;
4396 cookie.n_output = n_output;
4397 cookie.vlan_tci = 0;
4398 cookie_offset = put_userspace_action(ofproto, odp_actions, flow, &cookie);
4400 nl_msg_end_nested(odp_actions, actions_offset);
4401 nl_msg_end_nested(odp_actions, sample_offset);
4402 return cookie_offset;
4405 /* SAMPLE action must be first action in any given list of actions.
4406 * At this point we do not have all information required to build it. So try to
4407 * build sample action as complete as possible. */
4409 add_sflow_action(struct action_xlate_ctx *ctx)
4411 ctx->user_cookie_offset = compose_sflow_action(ctx->ofproto,
4413 &ctx->flow, OVSP_NONE);
4414 ctx->sflow_odp_port = 0;
4415 ctx->sflow_n_outputs = 0;
4418 /* Fix SAMPLE action according to data collected while composing ODP actions.
4419 * We need to fix SAMPLE actions OVS_SAMPLE_ATTR_ACTIONS attribute, i.e. nested
4420 * USERSPACE action's user-cookie which is required for sflow. */
4422 fix_sflow_action(struct action_xlate_ctx *ctx)
4424 const struct flow *base = &ctx->base_flow;
4425 struct user_action_cookie *cookie;
4427 if (!ctx->user_cookie_offset) {
4431 cookie = ofpbuf_at(ctx->odp_actions, ctx->user_cookie_offset,
4433 assert(cookie != NULL);
4434 assert(cookie->type == USER_ACTION_COOKIE_SFLOW);
4436 if (ctx->sflow_n_outputs) {
4437 cookie->data = dpif_sflow_odp_port_to_ifindex(ctx->ofproto->sflow,
4438 ctx->sflow_odp_port);
4440 if (ctx->sflow_n_outputs >= 255) {
4441 cookie->n_output = 255;
4443 cookie->n_output = ctx->sflow_n_outputs;
4445 cookie->vlan_tci = base->vlan_tci;
4449 compose_output_action__(struct action_xlate_ctx *ctx, uint16_t ofp_port,
4452 const struct ofport_dpif *ofport = get_ofp_port(ctx->ofproto, ofp_port);
4453 uint16_t odp_port = ofp_port_to_odp_port(ofp_port);
4454 ovs_be16 flow_vlan_tci = ctx->flow.vlan_tci;
4455 uint8_t flow_nw_tos = ctx->flow.nw_tos;
4459 struct priority_to_dscp *pdscp;
4461 if (ofport->up.pp.config & OFPUTIL_PC_NO_FWD
4462 || (check_stp && !stp_forward_in_state(ofport->stp_state))) {
4466 pdscp = get_priority(ofport, ctx->flow.skb_priority);
4468 ctx->flow.nw_tos &= ~IP_DSCP_MASK;
4469 ctx->flow.nw_tos |= pdscp->dscp;
4472 /* We may not have an ofport record for this port, but it doesn't hurt
4473 * to allow forwarding to it anyhow. Maybe such a port will appear
4474 * later and we're pre-populating the flow table. */
4477 out_port = vsp_realdev_to_vlandev(ctx->ofproto, odp_port,
4478 ctx->flow.vlan_tci);
4479 if (out_port != odp_port) {
4480 ctx->flow.vlan_tci = htons(0);
4482 commit_odp_actions(&ctx->flow, &ctx->base_flow, ctx->odp_actions);
4483 nl_msg_put_u32(ctx->odp_actions, OVS_ACTION_ATTR_OUTPUT, out_port);
4485 ctx->sflow_odp_port = odp_port;
4486 ctx->sflow_n_outputs++;
4487 ctx->nf_output_iface = ofp_port;
4488 ctx->flow.vlan_tci = flow_vlan_tci;
4489 ctx->flow.nw_tos = flow_nw_tos;
4493 compose_output_action(struct action_xlate_ctx *ctx, uint16_t ofp_port)
4495 compose_output_action__(ctx, ofp_port, true);
4499 xlate_table_action(struct action_xlate_ctx *ctx,
4500 uint16_t in_port, uint8_t table_id)
4502 if (ctx->recurse < MAX_RESUBMIT_RECURSION) {
4503 struct ofproto_dpif *ofproto = ctx->ofproto;
4504 struct rule_dpif *rule;
4505 uint16_t old_in_port;
4506 uint8_t old_table_id;
4508 old_table_id = ctx->table_id;
4509 ctx->table_id = table_id;
4511 /* Look up a flow with 'in_port' as the input port. */
4512 old_in_port = ctx->flow.in_port;
4513 ctx->flow.in_port = in_port;
4514 rule = rule_dpif_lookup(ofproto, &ctx->flow, table_id);
4517 if (table_id > 0 && table_id < N_TABLES) {
4518 struct table_dpif *table = &ofproto->tables[table_id];
4519 if (table->other_table) {
4520 ctx->tags |= (rule && rule->tag
4522 : rule_calculate_tag(&ctx->flow,
4523 &table->other_table->wc,
4528 /* Restore the original input port. Otherwise OFPP_NORMAL and
4529 * OFPP_IN_PORT will have surprising behavior. */
4530 ctx->flow.in_port = old_in_port;
4532 if (ctx->resubmit_hook) {
4533 ctx->resubmit_hook(ctx, rule);
4537 struct rule_dpif *old_rule = ctx->rule;
4539 if (ctx->resubmit_stats) {
4540 rule_credit_stats(rule, ctx->resubmit_stats);
4545 do_xlate_actions(rule->up.actions, rule->up.n_actions, ctx);
4546 ctx->rule = old_rule;
4550 ctx->table_id = old_table_id;
4552 static struct vlog_rate_limit recurse_rl = VLOG_RATE_LIMIT_INIT(1, 1);
4554 VLOG_ERR_RL(&recurse_rl, "resubmit actions recursed over %d times",
4555 MAX_RESUBMIT_RECURSION);
4556 ctx->max_resubmit_trigger = true;
4561 xlate_resubmit_table(struct action_xlate_ctx *ctx,
4562 const struct nx_action_resubmit *nar)
4567 in_port = (nar->in_port == htons(OFPP_IN_PORT)
4569 : ntohs(nar->in_port));
4570 table_id = nar->table == 255 ? ctx->table_id : nar->table;
4572 xlate_table_action(ctx, in_port, table_id);
4576 flood_packets(struct action_xlate_ctx *ctx, bool all)
4578 struct ofport_dpif *ofport;
4580 HMAP_FOR_EACH (ofport, up.hmap_node, &ctx->ofproto->up.ports) {
4581 uint16_t ofp_port = ofport->up.ofp_port;
4583 if (ofp_port == ctx->flow.in_port) {
4588 compose_output_action__(ctx, ofp_port, false);
4589 } else if (!(ofport->up.pp.config & OFPUTIL_PC_NO_FLOOD)) {
4590 compose_output_action(ctx, ofp_port);
4594 ctx->nf_output_iface = NF_OUT_FLOOD;
4598 execute_controller_action(struct action_xlate_ctx *ctx, int len,
4599 enum ofp_packet_in_reason reason,
4600 uint16_t controller_id)
4602 struct ofputil_packet_in pin;
4603 struct ofpbuf *packet;
4605 ctx->may_set_up_flow = false;
4610 packet = ofpbuf_clone(ctx->packet);
4612 if (packet->l2 && packet->l3) {
4613 struct eth_header *eh;
4615 eth_pop_vlan(packet);
4618 /* If the Ethernet type is less than ETH_TYPE_MIN, it's likely an 802.2
4619 * LLC frame. Calculating the Ethernet type of these frames is more
4620 * trouble than seems appropriate for a simple assertion. */
4621 assert(ntohs(eh->eth_type) < ETH_TYPE_MIN
4622 || eh->eth_type == ctx->flow.dl_type);
4624 memcpy(eh->eth_src, ctx->flow.dl_src, sizeof eh->eth_src);
4625 memcpy(eh->eth_dst, ctx->flow.dl_dst, sizeof eh->eth_dst);
4627 if (ctx->flow.vlan_tci & htons(VLAN_CFI)) {
4628 eth_push_vlan(packet, ctx->flow.vlan_tci);
4632 if (ctx->flow.dl_type == htons(ETH_TYPE_IP)) {
4633 packet_set_ipv4(packet, ctx->flow.nw_src, ctx->flow.nw_dst,
4634 ctx->flow.nw_tos, ctx->flow.nw_ttl);
4638 if (ctx->flow.nw_proto == IPPROTO_TCP) {
4639 packet_set_tcp_port(packet, ctx->flow.tp_src,
4641 } else if (ctx->flow.nw_proto == IPPROTO_UDP) {
4642 packet_set_udp_port(packet, ctx->flow.tp_src,
4649 pin.packet = packet->data;
4650 pin.packet_len = packet->size;
4651 pin.reason = reason;
4652 pin.controller_id = controller_id;
4653 pin.table_id = ctx->table_id;
4654 pin.cookie = ctx->rule ? ctx->rule->up.flow_cookie : 0;
4657 flow_get_metadata(&ctx->flow, &pin.fmd);
4659 connmgr_send_packet_in(ctx->ofproto->up.connmgr, &pin);
4660 ofpbuf_delete(packet);
4664 compose_dec_ttl(struct action_xlate_ctx *ctx)
4666 if (ctx->flow.dl_type != htons(ETH_TYPE_IP) &&
4667 ctx->flow.dl_type != htons(ETH_TYPE_IPV6)) {
4671 if (ctx->flow.nw_ttl > 1) {
4675 execute_controller_action(ctx, UINT16_MAX, OFPR_INVALID_TTL, 0);
4677 /* Stop processing for current table. */
4683 xlate_output_action__(struct action_xlate_ctx *ctx,
4684 uint16_t port, uint16_t max_len)
4686 uint16_t prev_nf_output_iface = ctx->nf_output_iface;
4688 ctx->nf_output_iface = NF_OUT_DROP;
4692 compose_output_action(ctx, ctx->flow.in_port);
4695 xlate_table_action(ctx, ctx->flow.in_port, ctx->table_id);
4701 flood_packets(ctx, false);
4704 flood_packets(ctx, true);
4706 case OFPP_CONTROLLER:
4707 execute_controller_action(ctx, max_len, OFPR_ACTION, 0);
4713 if (port != ctx->flow.in_port) {
4714 compose_output_action(ctx, port);
4719 if (prev_nf_output_iface == NF_OUT_FLOOD) {
4720 ctx->nf_output_iface = NF_OUT_FLOOD;
4721 } else if (ctx->nf_output_iface == NF_OUT_DROP) {
4722 ctx->nf_output_iface = prev_nf_output_iface;
4723 } else if (prev_nf_output_iface != NF_OUT_DROP &&
4724 ctx->nf_output_iface != NF_OUT_FLOOD) {
4725 ctx->nf_output_iface = NF_OUT_MULTI;
4730 xlate_output_reg_action(struct action_xlate_ctx *ctx,
4731 const struct nx_action_output_reg *naor)
4733 struct mf_subfield src;
4736 nxm_decode(&src, naor->src, naor->ofs_nbits);
4737 ofp_port = mf_get_subfield(&src, &ctx->flow);
4739 if (ofp_port <= UINT16_MAX) {
4740 xlate_output_action__(ctx, ofp_port, ntohs(naor->max_len));
4745 xlate_output_action(struct action_xlate_ctx *ctx,
4746 const struct ofp_action_output *oao)
4748 xlate_output_action__(ctx, ntohs(oao->port), ntohs(oao->max_len));
4752 xlate_enqueue_action(struct action_xlate_ctx *ctx,
4753 const struct ofp_action_enqueue *oae)
4756 uint32_t flow_priority, priority;
4759 error = dpif_queue_to_priority(ctx->ofproto->dpif, ntohl(oae->queue_id),
4762 /* Fall back to ordinary output action. */
4763 xlate_output_action__(ctx, ntohs(oae->port), 0);
4767 /* Figure out datapath output port. */
4768 ofp_port = ntohs(oae->port);
4769 if (ofp_port == OFPP_IN_PORT) {
4770 ofp_port = ctx->flow.in_port;
4771 } else if (ofp_port == ctx->flow.in_port) {
4775 /* Add datapath actions. */
4776 flow_priority = ctx->flow.skb_priority;
4777 ctx->flow.skb_priority = priority;
4778 compose_output_action(ctx, ofp_port);
4779 ctx->flow.skb_priority = flow_priority;
4781 /* Update NetFlow output port. */
4782 if (ctx->nf_output_iface == NF_OUT_DROP) {
4783 ctx->nf_output_iface = ofp_port;
4784 } else if (ctx->nf_output_iface != NF_OUT_FLOOD) {
4785 ctx->nf_output_iface = NF_OUT_MULTI;
4790 xlate_set_queue_action(struct action_xlate_ctx *ctx,
4791 const struct nx_action_set_queue *nasq)
4796 error = dpif_queue_to_priority(ctx->ofproto->dpif, ntohl(nasq->queue_id),
4799 /* Couldn't translate queue to a priority, so ignore. A warning
4800 * has already been logged. */
4804 ctx->flow.skb_priority = priority;
4807 struct xlate_reg_state {
4813 xlate_autopath(struct action_xlate_ctx *ctx,
4814 const struct nx_action_autopath *naa)
4816 uint16_t ofp_port = ntohl(naa->id);
4817 struct ofport_dpif *port = get_ofp_port(ctx->ofproto, ofp_port);
4819 if (!port || !port->bundle) {
4820 ofp_port = OFPP_NONE;
4821 } else if (port->bundle->bond) {
4822 /* Autopath does not support VLAN hashing. */
4823 struct ofport_dpif *slave = bond_choose_output_slave(
4824 port->bundle->bond, &ctx->flow, 0, &ctx->tags);
4826 ofp_port = slave->up.ofp_port;
4829 autopath_execute(naa, &ctx->flow, ofp_port);
4833 slave_enabled_cb(uint16_t ofp_port, void *ofproto_)
4835 struct ofproto_dpif *ofproto = ofproto_;
4836 struct ofport_dpif *port;
4846 case OFPP_CONTROLLER: /* Not supported by the bundle action. */
4849 port = get_ofp_port(ofproto, ofp_port);
4850 return port ? port->may_enable : false;
4855 xlate_learn_action(struct action_xlate_ctx *ctx,
4856 const struct nx_action_learn *learn)
4858 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(5, 1);
4859 struct ofputil_flow_mod fm;
4862 learn_execute(learn, &ctx->flow, &fm);
4864 error = ofproto_flow_mod(&ctx->ofproto->up, &fm);
4865 if (error && !VLOG_DROP_WARN(&rl)) {
4866 VLOG_WARN("learning action failed to modify flow table (%s)",
4867 ofperr_get_name(error));
4873 /* Reduces '*timeout' to no more than 'max'. A value of zero in either case
4874 * means "infinite". */
4876 reduce_timeout(uint16_t max, uint16_t *timeout)
4878 if (max && (!*timeout || *timeout > max)) {
4884 xlate_fin_timeout(struct action_xlate_ctx *ctx,
4885 const struct nx_action_fin_timeout *naft)
4887 if (ctx->tcp_flags & (TCP_FIN | TCP_RST) && ctx->rule) {
4888 struct rule_dpif *rule = ctx->rule;
4890 reduce_timeout(ntohs(naft->fin_idle_timeout), &rule->up.idle_timeout);
4891 reduce_timeout(ntohs(naft->fin_hard_timeout), &rule->up.hard_timeout);
4896 may_receive(const struct ofport_dpif *port, struct action_xlate_ctx *ctx)
4898 if (port->up.pp.config & (eth_addr_equals(ctx->flow.dl_dst, eth_addr_stp)
4899 ? OFPUTIL_PC_NO_RECV_STP
4900 : OFPUTIL_PC_NO_RECV)) {
4904 /* Only drop packets here if both forwarding and learning are
4905 * disabled. If just learning is enabled, we need to have
4906 * OFPP_NORMAL and the learning action have a look at the packet
4907 * before we can drop it. */
4908 if (!stp_forward_in_state(port->stp_state)
4909 && !stp_learn_in_state(port->stp_state)) {
4917 do_xlate_actions(const union ofp_action *in, size_t n_in,
4918 struct action_xlate_ctx *ctx)
4920 const struct ofport_dpif *port;
4921 const union ofp_action *ia;
4922 bool was_evictable = true;
4925 port = get_ofp_port(ctx->ofproto, ctx->flow.in_port);
4926 if (port && !may_receive(port, ctx)) {
4927 /* Drop this flow. */
4932 /* Don't let the rule we're working on get evicted underneath us. */
4933 was_evictable = ctx->rule->up.evictable;
4934 ctx->rule->up.evictable = false;
4936 OFPUTIL_ACTION_FOR_EACH_UNSAFE (ia, left, in, n_in) {
4937 const struct ofp_action_dl_addr *oada;
4938 const struct nx_action_resubmit *nar;
4939 const struct nx_action_set_tunnel *nast;
4940 const struct nx_action_set_queue *nasq;
4941 const struct nx_action_multipath *nam;
4942 const struct nx_action_autopath *naa;
4943 const struct nx_action_bundle *nab;
4944 const struct nx_action_output_reg *naor;
4945 const struct nx_action_controller *nac;
4946 enum ofputil_action_code code;
4953 code = ofputil_decode_action_unsafe(ia);
4955 case OFPUTIL_OFPAT10_OUTPUT:
4956 xlate_output_action(ctx, &ia->output);
4959 case OFPUTIL_OFPAT10_SET_VLAN_VID:
4960 ctx->flow.vlan_tci &= ~htons(VLAN_VID_MASK);
4961 ctx->flow.vlan_tci |= ia->vlan_vid.vlan_vid | htons(VLAN_CFI);
4964 case OFPUTIL_OFPAT10_SET_VLAN_PCP:
4965 ctx->flow.vlan_tci &= ~htons(VLAN_PCP_MASK);
4966 ctx->flow.vlan_tci |= htons(
4967 (ia->vlan_pcp.vlan_pcp << VLAN_PCP_SHIFT) | VLAN_CFI);
4970 case OFPUTIL_OFPAT10_STRIP_VLAN:
4971 ctx->flow.vlan_tci = htons(0);
4974 case OFPUTIL_OFPAT10_SET_DL_SRC:
4975 oada = ((struct ofp_action_dl_addr *) ia);
4976 memcpy(ctx->flow.dl_src, oada->dl_addr, ETH_ADDR_LEN);
4979 case OFPUTIL_OFPAT10_SET_DL_DST:
4980 oada = ((struct ofp_action_dl_addr *) ia);
4981 memcpy(ctx->flow.dl_dst, oada->dl_addr, ETH_ADDR_LEN);
4984 case OFPUTIL_OFPAT10_SET_NW_SRC:
4985 ctx->flow.nw_src = ia->nw_addr.nw_addr;
4988 case OFPUTIL_OFPAT10_SET_NW_DST:
4989 ctx->flow.nw_dst = ia->nw_addr.nw_addr;
4992 case OFPUTIL_OFPAT10_SET_NW_TOS:
4993 /* OpenFlow 1.0 only supports IPv4. */
4994 if (ctx->flow.dl_type == htons(ETH_TYPE_IP)) {
4995 ctx->flow.nw_tos &= ~IP_DSCP_MASK;
4996 ctx->flow.nw_tos |= ia->nw_tos.nw_tos & IP_DSCP_MASK;
5000 case OFPUTIL_OFPAT10_SET_TP_SRC:
5001 ctx->flow.tp_src = ia->tp_port.tp_port;
5004 case OFPUTIL_OFPAT10_SET_TP_DST:
5005 ctx->flow.tp_dst = ia->tp_port.tp_port;
5008 case OFPUTIL_OFPAT10_ENQUEUE:
5009 xlate_enqueue_action(ctx, (const struct ofp_action_enqueue *) ia);
5012 case OFPUTIL_NXAST_RESUBMIT:
5013 nar = (const struct nx_action_resubmit *) ia;
5014 xlate_table_action(ctx, ntohs(nar->in_port), ctx->table_id);
5017 case OFPUTIL_NXAST_RESUBMIT_TABLE:
5018 xlate_resubmit_table(ctx, (const struct nx_action_resubmit *) ia);
5021 case OFPUTIL_NXAST_SET_TUNNEL:
5022 nast = (const struct nx_action_set_tunnel *) ia;
5023 tun_id = htonll(ntohl(nast->tun_id));
5024 ctx->flow.tun_id = tun_id;
5027 case OFPUTIL_NXAST_SET_QUEUE:
5028 nasq = (const struct nx_action_set_queue *) ia;
5029 xlate_set_queue_action(ctx, nasq);
5032 case OFPUTIL_NXAST_POP_QUEUE:
5033 ctx->flow.skb_priority = ctx->orig_skb_priority;
5036 case OFPUTIL_NXAST_REG_MOVE:
5037 nxm_execute_reg_move((const struct nx_action_reg_move *) ia,
5041 case OFPUTIL_NXAST_REG_LOAD:
5042 nxm_execute_reg_load((const struct nx_action_reg_load *) ia,
5046 case OFPUTIL_NXAST_NOTE:
5047 /* Nothing to do. */
5050 case OFPUTIL_NXAST_SET_TUNNEL64:
5051 tun_id = ((const struct nx_action_set_tunnel64 *) ia)->tun_id;
5052 ctx->flow.tun_id = tun_id;
5055 case OFPUTIL_NXAST_MULTIPATH:
5056 nam = (const struct nx_action_multipath *) ia;
5057 multipath_execute(nam, &ctx->flow);
5060 case OFPUTIL_NXAST_AUTOPATH:
5061 naa = (const struct nx_action_autopath *) ia;
5062 xlate_autopath(ctx, naa);
5065 case OFPUTIL_NXAST_BUNDLE:
5066 ctx->ofproto->has_bundle_action = true;
5067 nab = (const struct nx_action_bundle *) ia;
5068 xlate_output_action__(ctx, bundle_execute(nab, &ctx->flow,
5073 case OFPUTIL_NXAST_BUNDLE_LOAD:
5074 ctx->ofproto->has_bundle_action = true;
5075 nab = (const struct nx_action_bundle *) ia;
5076 bundle_execute_load(nab, &ctx->flow, slave_enabled_cb,
5080 case OFPUTIL_NXAST_OUTPUT_REG:
5081 naor = (const struct nx_action_output_reg *) ia;
5082 xlate_output_reg_action(ctx, naor);
5085 case OFPUTIL_NXAST_LEARN:
5086 ctx->has_learn = true;
5087 if (ctx->may_learn) {
5088 xlate_learn_action(ctx, (const struct nx_action_learn *) ia);
5092 case OFPUTIL_NXAST_DEC_TTL:
5093 if (compose_dec_ttl(ctx)) {
5098 case OFPUTIL_NXAST_EXIT:
5102 case OFPUTIL_NXAST_FIN_TIMEOUT:
5103 ctx->has_fin_timeout = true;
5104 xlate_fin_timeout(ctx, (const struct nx_action_fin_timeout *) ia);
5107 case OFPUTIL_NXAST_CONTROLLER:
5108 nac = (const struct nx_action_controller *) ia;
5109 execute_controller_action(ctx, ntohs(nac->max_len), nac->reason,
5110 ntohs(nac->controller_id));
5116 /* We've let OFPP_NORMAL and the learning action look at the packet,
5117 * so drop it now if forwarding is disabled. */
5118 if (port && !stp_forward_in_state(port->stp_state)) {
5119 ofpbuf_clear(ctx->odp_actions);
5120 add_sflow_action(ctx);
5123 ctx->rule->up.evictable = was_evictable;
5128 action_xlate_ctx_init(struct action_xlate_ctx *ctx,
5129 struct ofproto_dpif *ofproto, const struct flow *flow,
5130 ovs_be16 initial_tci, struct rule_dpif *rule,
5131 uint8_t tcp_flags, const struct ofpbuf *packet)
5133 ctx->ofproto = ofproto;
5135 ctx->base_flow = ctx->flow;
5136 ctx->base_flow.tun_id = 0;
5137 ctx->base_flow.vlan_tci = initial_tci;
5139 ctx->packet = packet;
5140 ctx->may_learn = packet != NULL;
5141 ctx->tcp_flags = tcp_flags;
5142 ctx->resubmit_hook = NULL;
5143 ctx->resubmit_stats = NULL;
5146 /* Translates the 'n_in' "union ofp_action"s in 'in' into datapath actions in
5147 * 'odp_actions', using 'ctx'. */
5149 xlate_actions(struct action_xlate_ctx *ctx,
5150 const union ofp_action *in, size_t n_in,
5151 struct ofpbuf *odp_actions)
5153 struct flow orig_flow = ctx->flow;
5155 COVERAGE_INC(ofproto_dpif_xlate);
5157 ofpbuf_clear(odp_actions);
5158 ofpbuf_reserve(odp_actions, NL_A_U32_SIZE);
5160 ctx->odp_actions = odp_actions;
5162 ctx->may_set_up_flow = true;
5163 ctx->has_learn = false;
5164 ctx->has_normal = false;
5165 ctx->has_fin_timeout = false;
5166 ctx->nf_output_iface = NF_OUT_DROP;
5169 ctx->max_resubmit_trigger = false;
5170 ctx->orig_skb_priority = ctx->flow.skb_priority;
5174 if (ctx->flow.nw_frag & FLOW_NW_FRAG_ANY) {
5175 switch (ctx->ofproto->up.frag_handling) {
5176 case OFPC_FRAG_NORMAL:
5177 /* We must pretend that transport ports are unavailable. */
5178 ctx->flow.tp_src = ctx->base_flow.tp_src = htons(0);
5179 ctx->flow.tp_dst = ctx->base_flow.tp_dst = htons(0);
5182 case OFPC_FRAG_DROP:
5185 case OFPC_FRAG_REASM:
5188 case OFPC_FRAG_NX_MATCH:
5189 /* Nothing to do. */
5192 case OFPC_INVALID_TTL_TO_CONTROLLER:
5197 if (process_special(ctx->ofproto, &ctx->flow, ctx->packet)) {
5198 ctx->may_set_up_flow = false;
5200 static struct vlog_rate_limit trace_rl = VLOG_RATE_LIMIT_INIT(1, 1);
5201 struct flow original_flow = ctx->flow;
5202 ovs_be16 initial_tci = ctx->base_flow.vlan_tci;
5204 add_sflow_action(ctx);
5205 do_xlate_actions(in, n_in, ctx);
5207 if (ctx->max_resubmit_trigger && !ctx->resubmit_hook
5208 && !VLOG_DROP_ERR(&trace_rl)) {
5209 struct ds ds = DS_EMPTY_INITIALIZER;
5211 ofproto_trace(ctx->ofproto, &original_flow, ctx->packet,
5213 VLOG_ERR("Trace triggered by excessive resubmit recursion:\n%s",
5218 if (!connmgr_may_set_up_flow(ctx->ofproto->up.connmgr, &ctx->flow,
5219 ctx->odp_actions->data,
5220 ctx->odp_actions->size)) {
5221 ctx->may_set_up_flow = false;
5223 && connmgr_msg_in_hook(ctx->ofproto->up.connmgr, &ctx->flow,
5225 compose_output_action(ctx, OFPP_LOCAL);
5228 add_mirror_actions(ctx, &orig_flow);
5229 fix_sflow_action(ctx);
5233 /* Translates the 'n_in' "union ofp_action"s in 'in' into datapath actions,
5234 * using 'ctx', and discards the datapath actions. */
5236 xlate_actions_for_side_effects(struct action_xlate_ctx *ctx,
5237 const union ofp_action *in, size_t n_in)
5239 uint64_t odp_actions_stub[1024 / 8];
5240 struct ofpbuf odp_actions;
5242 ofpbuf_use_stub(&odp_actions, odp_actions_stub, sizeof odp_actions_stub);
5243 xlate_actions(ctx, in, n_in, &odp_actions);
5244 ofpbuf_uninit(&odp_actions);
5247 /* OFPP_NORMAL implementation. */
5249 static struct ofport_dpif *ofbundle_get_a_port(const struct ofbundle *);
5251 /* Given 'vid', the VID obtained from the 802.1Q header that was received as
5252 * part of a packet (specify 0 if there was no 802.1Q header), and 'in_bundle',
5253 * the bundle on which the packet was received, returns the VLAN to which the
5256 * Both 'vid' and the return value are in the range 0...4095. */
5258 input_vid_to_vlan(const struct ofbundle *in_bundle, uint16_t vid)
5260 switch (in_bundle->vlan_mode) {
5261 case PORT_VLAN_ACCESS:
5262 return in_bundle->vlan;
5265 case PORT_VLAN_TRUNK:
5268 case PORT_VLAN_NATIVE_UNTAGGED:
5269 case PORT_VLAN_NATIVE_TAGGED:
5270 return vid ? vid : in_bundle->vlan;
5277 /* Checks whether a packet with the given 'vid' may ingress on 'in_bundle'.
5278 * If so, returns true. Otherwise, returns false and, if 'warn' is true, logs
5281 * 'vid' should be the VID obtained from the 802.1Q header that was received as
5282 * part of a packet (specify 0 if there was no 802.1Q header), in the range
5285 input_vid_is_valid(uint16_t vid, struct ofbundle *in_bundle, bool warn)
5287 /* Allow any VID on the OFPP_NONE port. */
5288 if (in_bundle == &ofpp_none_bundle) {
5292 switch (in_bundle->vlan_mode) {
5293 case PORT_VLAN_ACCESS:
5296 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
5297 VLOG_WARN_RL(&rl, "bridge %s: dropping VLAN %"PRIu16" tagged "
5298 "packet received on port %s configured as VLAN "
5299 "%"PRIu16" access port",
5300 in_bundle->ofproto->up.name, vid,
5301 in_bundle->name, in_bundle->vlan);
5307 case PORT_VLAN_NATIVE_UNTAGGED:
5308 case PORT_VLAN_NATIVE_TAGGED:
5310 /* Port must always carry its native VLAN. */
5314 case PORT_VLAN_TRUNK:
5315 if (!ofbundle_includes_vlan(in_bundle, vid)) {
5317 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
5318 VLOG_WARN_RL(&rl, "bridge %s: dropping VLAN %"PRIu16" packet "
5319 "received on port %s not configured for trunking "
5321 in_bundle->ofproto->up.name, vid,
5322 in_bundle->name, vid);
5334 /* Given 'vlan', the VLAN that a packet belongs to, and
5335 * 'out_bundle', a bundle on which the packet is to be output, returns the VID
5336 * that should be included in the 802.1Q header. (If the return value is 0,
5337 * then the 802.1Q header should only be included in the packet if there is a
5340 * Both 'vlan' and the return value are in the range 0...4095. */
5342 output_vlan_to_vid(const struct ofbundle *out_bundle, uint16_t vlan)
5344 switch (out_bundle->vlan_mode) {
5345 case PORT_VLAN_ACCESS:
5348 case PORT_VLAN_TRUNK:
5349 case PORT_VLAN_NATIVE_TAGGED:
5352 case PORT_VLAN_NATIVE_UNTAGGED:
5353 return vlan == out_bundle->vlan ? 0 : vlan;
5361 output_normal(struct action_xlate_ctx *ctx, const struct ofbundle *out_bundle,
5364 struct ofport_dpif *port;
5366 ovs_be16 tci, old_tci;
5368 vid = output_vlan_to_vid(out_bundle, vlan);
5369 if (!out_bundle->bond) {
5370 port = ofbundle_get_a_port(out_bundle);
5372 port = bond_choose_output_slave(out_bundle->bond, &ctx->flow,
5375 /* No slaves enabled, so drop packet. */
5380 old_tci = ctx->flow.vlan_tci;
5382 if (tci || out_bundle->use_priority_tags) {
5383 tci |= ctx->flow.vlan_tci & htons(VLAN_PCP_MASK);
5385 tci |= htons(VLAN_CFI);
5388 ctx->flow.vlan_tci = tci;
5390 compose_output_action(ctx, port->up.ofp_port);
5391 ctx->flow.vlan_tci = old_tci;
5395 mirror_mask_ffs(mirror_mask_t mask)
5397 BUILD_ASSERT_DECL(sizeof(unsigned int) >= sizeof(mask));
5402 ofbundle_trunks_vlan(const struct ofbundle *bundle, uint16_t vlan)
5404 return (bundle->vlan_mode != PORT_VLAN_ACCESS
5405 && (!bundle->trunks || bitmap_is_set(bundle->trunks, vlan)));
5409 ofbundle_includes_vlan(const struct ofbundle *bundle, uint16_t vlan)
5411 return vlan == bundle->vlan || ofbundle_trunks_vlan(bundle, vlan);
5414 /* Returns an arbitrary interface within 'bundle'. */
5415 static struct ofport_dpif *
5416 ofbundle_get_a_port(const struct ofbundle *bundle)
5418 return CONTAINER_OF(list_front(&bundle->ports),
5419 struct ofport_dpif, bundle_node);
5423 vlan_is_mirrored(const struct ofmirror *m, int vlan)
5425 return !m->vlans || bitmap_is_set(m->vlans, vlan);
5428 /* Returns true if a packet with Ethernet destination MAC 'dst' may be mirrored
5429 * to a VLAN. In general most packets may be mirrored but we want to drop
5430 * protocols that may confuse switches. */
5432 eth_dst_may_rspan(const uint8_t dst[ETH_ADDR_LEN])
5434 /* If you change this function's behavior, please update corresponding
5435 * documentation in vswitch.xml at the same time. */
5436 if (dst[0] != 0x01) {
5437 /* All the currently banned MACs happen to start with 01 currently, so
5438 * this is a quick way to eliminate most of the good ones. */
5440 if (eth_addr_is_reserved(dst)) {
5441 /* Drop STP, IEEE pause frames, and other reserved protocols
5442 * (01-80-c2-00-00-0x). */
5446 if (dst[0] == 0x01 && dst[1] == 0x00 && dst[2] == 0x0c) {
5448 if ((dst[3] & 0xfe) == 0xcc &&
5449 (dst[4] & 0xfe) == 0xcc &&
5450 (dst[5] & 0xfe) == 0xcc) {
5451 /* Drop the following protocols plus others following the same
5454 CDP, VTP, DTP, PAgP (01-00-0c-cc-cc-cc)
5455 Spanning Tree PVSTP+ (01-00-0c-cc-cc-cd)
5456 STP Uplink Fast (01-00-0c-cd-cd-cd) */
5460 if (!(dst[3] | dst[4] | dst[5])) {
5461 /* Drop Inter Switch Link packets (01-00-0c-00-00-00). */
5470 add_mirror_actions(struct action_xlate_ctx *ctx, const struct flow *orig_flow)
5472 struct ofproto_dpif *ofproto = ctx->ofproto;
5473 mirror_mask_t mirrors;
5474 struct ofbundle *in_bundle;
5477 const struct nlattr *a;
5480 in_bundle = lookup_input_bundle(ctx->ofproto, orig_flow->in_port,
5481 ctx->packet != NULL);
5485 mirrors = in_bundle->src_mirrors;
5487 /* Drop frames on bundles reserved for mirroring. */
5488 if (in_bundle->mirror_out) {
5489 if (ctx->packet != NULL) {
5490 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
5491 VLOG_WARN_RL(&rl, "bridge %s: dropping packet received on port "
5492 "%s, which is reserved exclusively for mirroring",
5493 ctx->ofproto->up.name, in_bundle->name);
5499 vid = vlan_tci_to_vid(orig_flow->vlan_tci);
5500 if (!input_vid_is_valid(vid, in_bundle, ctx->packet != NULL)) {
5503 vlan = input_vid_to_vlan(in_bundle, vid);
5505 /* Look at the output ports to check for destination selections. */
5507 NL_ATTR_FOR_EACH (a, left, ctx->odp_actions->data,
5508 ctx->odp_actions->size) {
5509 enum ovs_action_attr type = nl_attr_type(a);
5510 struct ofport_dpif *ofport;
5512 if (type != OVS_ACTION_ATTR_OUTPUT) {
5516 ofport = get_odp_port(ofproto, nl_attr_get_u32(a));
5517 if (ofport && ofport->bundle) {
5518 mirrors |= ofport->bundle->dst_mirrors;
5526 /* Restore the original packet before adding the mirror actions. */
5527 ctx->flow = *orig_flow;
5532 m = ofproto->mirrors[mirror_mask_ffs(mirrors) - 1];
5534 if (!vlan_is_mirrored(m, vlan)) {
5535 mirrors &= mirrors - 1;
5539 mirrors &= ~m->dup_mirrors;
5540 ctx->mirrors |= m->dup_mirrors;
5542 output_normal(ctx, m->out, vlan);
5543 } else if (eth_dst_may_rspan(orig_flow->dl_dst)
5544 && vlan != m->out_vlan) {
5545 struct ofbundle *bundle;
5547 HMAP_FOR_EACH (bundle, hmap_node, &ofproto->bundles) {
5548 if (ofbundle_includes_vlan(bundle, m->out_vlan)
5549 && !bundle->mirror_out) {
5550 output_normal(ctx, bundle, m->out_vlan);
5558 update_mirror_stats(struct ofproto_dpif *ofproto, mirror_mask_t mirrors,
5559 uint64_t packets, uint64_t bytes)
5565 for (; mirrors; mirrors &= mirrors - 1) {
5568 m = ofproto->mirrors[mirror_mask_ffs(mirrors) - 1];
5571 /* In normal circumstances 'm' will not be NULL. However,
5572 * if mirrors are reconfigured, we can temporarily get out
5573 * of sync in facet_revalidate(). We could "correct" the
5574 * mirror list before reaching here, but doing that would
5575 * not properly account the traffic stats we've currently
5576 * accumulated for previous mirror configuration. */
5580 m->packet_count += packets;
5581 m->byte_count += bytes;
5585 /* A VM broadcasts a gratuitous ARP to indicate that it has resumed after
5586 * migration. Older Citrix-patched Linux DomU used gratuitous ARP replies to
5587 * indicate this; newer upstream kernels use gratuitous ARP requests. */
5589 is_gratuitous_arp(const struct flow *flow)
5591 return (flow->dl_type == htons(ETH_TYPE_ARP)
5592 && eth_addr_is_broadcast(flow->dl_dst)
5593 && (flow->nw_proto == ARP_OP_REPLY
5594 || (flow->nw_proto == ARP_OP_REQUEST
5595 && flow->nw_src == flow->nw_dst)));
5599 update_learning_table(struct ofproto_dpif *ofproto,
5600 const struct flow *flow, int vlan,
5601 struct ofbundle *in_bundle)
5603 struct mac_entry *mac;
5605 /* Don't learn the OFPP_NONE port. */
5606 if (in_bundle == &ofpp_none_bundle) {
5610 if (!mac_learning_may_learn(ofproto->ml, flow->dl_src, vlan)) {
5614 mac = mac_learning_insert(ofproto->ml, flow->dl_src, vlan);
5615 if (is_gratuitous_arp(flow)) {
5616 /* We don't want to learn from gratuitous ARP packets that are
5617 * reflected back over bond slaves so we lock the learning table. */
5618 if (!in_bundle->bond) {
5619 mac_entry_set_grat_arp_lock(mac);
5620 } else if (mac_entry_is_grat_arp_locked(mac)) {
5625 if (mac_entry_is_new(mac) || mac->port.p != in_bundle) {
5626 /* The log messages here could actually be useful in debugging,
5627 * so keep the rate limit relatively high. */
5628 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(30, 300);
5629 VLOG_DBG_RL(&rl, "bridge %s: learned that "ETH_ADDR_FMT" is "
5630 "on port %s in VLAN %d",
5631 ofproto->up.name, ETH_ADDR_ARGS(flow->dl_src),
5632 in_bundle->name, vlan);
5634 mac->port.p = in_bundle;
5635 tag_set_add(&ofproto->revalidate_set,
5636 mac_learning_changed(ofproto->ml, mac));
5640 static struct ofbundle *
5641 lookup_input_bundle(struct ofproto_dpif *ofproto, uint16_t in_port, bool warn)
5643 struct ofport_dpif *ofport;
5645 /* Special-case OFPP_NONE, which a controller may use as the ingress
5646 * port for traffic that it is sourcing. */
5647 if (in_port == OFPP_NONE) {
5648 return &ofpp_none_bundle;
5651 /* Find the port and bundle for the received packet. */
5652 ofport = get_ofp_port(ofproto, in_port);
5653 if (ofport && ofport->bundle) {
5654 return ofport->bundle;
5657 /* Odd. A few possible reasons here:
5659 * - We deleted a port but there are still a few packets queued up
5662 * - Someone externally added a port (e.g. "ovs-dpctl add-if") that
5663 * we don't know about.
5665 * - The ofproto client didn't configure the port as part of a bundle.
5668 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
5670 VLOG_WARN_RL(&rl, "bridge %s: received packet on unknown "
5671 "port %"PRIu16, ofproto->up.name, in_port);
5676 /* Determines whether packets in 'flow' within 'ofproto' should be forwarded or
5677 * dropped. Returns true if they may be forwarded, false if they should be
5680 * 'in_port' must be the ofport_dpif that corresponds to flow->in_port.
5681 * 'in_port' must be part of a bundle (e.g. in_port->bundle must be nonnull).
5683 * 'vlan' must be the VLAN that corresponds to flow->vlan_tci on 'in_port', as
5684 * returned by input_vid_to_vlan(). It must be a valid VLAN for 'in_port', as
5685 * checked by input_vid_is_valid().
5687 * May also add tags to '*tags', although the current implementation only does
5688 * so in one special case.
5691 is_admissible(struct ofproto_dpif *ofproto, const struct flow *flow,
5692 struct ofport_dpif *in_port, uint16_t vlan, tag_type *tags)
5694 struct ofbundle *in_bundle = in_port->bundle;
5696 /* Drop frames for reserved multicast addresses
5697 * only if forward_bpdu option is absent. */
5698 if (eth_addr_is_reserved(flow->dl_dst) && !ofproto->up.forward_bpdu) {
5702 if (in_bundle->bond) {
5703 struct mac_entry *mac;
5705 switch (bond_check_admissibility(in_bundle->bond, in_port,
5706 flow->dl_dst, tags)) {
5713 case BV_DROP_IF_MOVED:
5714 mac = mac_learning_lookup(ofproto->ml, flow->dl_src, vlan, NULL);
5715 if (mac && mac->port.p != in_bundle &&
5716 (!is_gratuitous_arp(flow)
5717 || mac_entry_is_grat_arp_locked(mac))) {
5728 xlate_normal(struct action_xlate_ctx *ctx)
5730 struct ofport_dpif *in_port;
5731 struct ofbundle *in_bundle;
5732 struct mac_entry *mac;
5736 ctx->has_normal = true;
5738 in_bundle = lookup_input_bundle(ctx->ofproto, ctx->flow.in_port,
5739 ctx->packet != NULL);
5744 /* We know 'in_port' exists unless it is "ofpp_none_bundle",
5745 * since lookup_input_bundle() succeeded. */
5746 in_port = get_ofp_port(ctx->ofproto, ctx->flow.in_port);
5748 /* Drop malformed frames. */
5749 if (ctx->flow.dl_type == htons(ETH_TYPE_VLAN) &&
5750 !(ctx->flow.vlan_tci & htons(VLAN_CFI))) {
5751 if (ctx->packet != NULL) {
5752 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
5753 VLOG_WARN_RL(&rl, "bridge %s: dropping packet with partial "
5754 "VLAN tag received on port %s",
5755 ctx->ofproto->up.name, in_bundle->name);
5760 /* Drop frames on bundles reserved for mirroring. */
5761 if (in_bundle->mirror_out) {
5762 if (ctx->packet != NULL) {
5763 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
5764 VLOG_WARN_RL(&rl, "bridge %s: dropping packet received on port "
5765 "%s, which is reserved exclusively for mirroring",
5766 ctx->ofproto->up.name, in_bundle->name);
5772 vid = vlan_tci_to_vid(ctx->flow.vlan_tci);
5773 if (!input_vid_is_valid(vid, in_bundle, ctx->packet != NULL)) {
5776 vlan = input_vid_to_vlan(in_bundle, vid);
5778 /* Check other admissibility requirements. */
5780 !is_admissible(ctx->ofproto, &ctx->flow, in_port, vlan, &ctx->tags)) {
5784 /* Learn source MAC. */
5785 if (ctx->may_learn) {
5786 update_learning_table(ctx->ofproto, &ctx->flow, vlan, in_bundle);
5789 /* Determine output bundle. */
5790 mac = mac_learning_lookup(ctx->ofproto->ml, ctx->flow.dl_dst, vlan,
5793 if (mac->port.p != in_bundle) {
5794 output_normal(ctx, mac->port.p, vlan);
5797 struct ofbundle *bundle;
5799 HMAP_FOR_EACH (bundle, hmap_node, &ctx->ofproto->bundles) {
5800 if (bundle != in_bundle
5801 && ofbundle_includes_vlan(bundle, vlan)
5802 && bundle->floodable
5803 && !bundle->mirror_out) {
5804 output_normal(ctx, bundle, vlan);
5807 ctx->nf_output_iface = NF_OUT_FLOOD;
5811 /* Optimized flow revalidation.
5813 * It's a difficult problem, in general, to tell which facets need to have
5814 * their actions recalculated whenever the OpenFlow flow table changes. We
5815 * don't try to solve that general problem: for most kinds of OpenFlow flow
5816 * table changes, we recalculate the actions for every facet. This is
5817 * relatively expensive, but it's good enough if the OpenFlow flow table
5818 * doesn't change very often.
5820 * However, we can expect one particular kind of OpenFlow flow table change to
5821 * happen frequently: changes caused by MAC learning. To avoid wasting a lot
5822 * of CPU on revalidating every facet whenever MAC learning modifies the flow
5823 * table, we add a special case that applies to flow tables in which every rule
5824 * has the same form (that is, the same wildcards), except that the table is
5825 * also allowed to have a single "catch-all" flow that matches all packets. We
5826 * optimize this case by tagging all of the facets that resubmit into the table
5827 * and invalidating the same tag whenever a flow changes in that table. The
5828 * end result is that we revalidate just the facets that need it (and sometimes
5829 * a few more, but not all of the facets or even all of the facets that
5830 * resubmit to the table modified by MAC learning). */
5832 /* Calculates the tag to use for 'flow' and wildcards 'wc' when it is inserted
5833 * into an OpenFlow table with the given 'basis'. */
5835 rule_calculate_tag(const struct flow *flow, const struct flow_wildcards *wc,
5838 if (flow_wildcards_is_catchall(wc)) {
5841 struct flow tag_flow = *flow;
5842 flow_zero_wildcards(&tag_flow, wc);
5843 return tag_create_deterministic(flow_hash(&tag_flow, secret));
5847 /* Following a change to OpenFlow table 'table_id' in 'ofproto', update the
5848 * taggability of that table.
5850 * This function must be called after *each* change to a flow table. If you
5851 * skip calling it on some changes then the pointer comparisons at the end can
5852 * be invalid if you get unlucky. For example, if a flow removal causes a
5853 * cls_table to be destroyed and then a flow insertion causes a cls_table with
5854 * different wildcards to be created with the same address, then this function
5855 * will incorrectly skip revalidation. */
5857 table_update_taggable(struct ofproto_dpif *ofproto, uint8_t table_id)
5859 struct table_dpif *table = &ofproto->tables[table_id];
5860 const struct oftable *oftable = &ofproto->up.tables[table_id];
5861 struct cls_table *catchall, *other;
5862 struct cls_table *t;
5864 catchall = other = NULL;
5866 switch (hmap_count(&oftable->cls.tables)) {
5868 /* We could tag this OpenFlow table but it would make the logic a
5869 * little harder and it's a corner case that doesn't seem worth it
5875 HMAP_FOR_EACH (t, hmap_node, &oftable->cls.tables) {
5876 if (cls_table_is_catchall(t)) {
5878 } else if (!other) {
5881 /* Indicate that we can't tag this by setting both tables to
5882 * NULL. (We know that 'catchall' is already NULL.) */
5889 /* Can't tag this table. */
5893 if (table->catchall_table != catchall || table->other_table != other) {
5894 table->catchall_table = catchall;
5895 table->other_table = other;
5896 ofproto->need_revalidate = true;
5900 /* Given 'rule' that has changed in some way (either it is a rule being
5901 * inserted, a rule being deleted, or a rule whose actions are being
5902 * modified), marks facets for revalidation to ensure that packets will be
5903 * forwarded correctly according to the new state of the flow table.
5905 * This function must be called after *each* change to a flow table. See
5906 * the comment on table_update_taggable() for more information. */
5908 rule_invalidate(const struct rule_dpif *rule)
5910 struct ofproto_dpif *ofproto = ofproto_dpif_cast(rule->up.ofproto);
5912 table_update_taggable(ofproto, rule->up.table_id);
5914 if (!ofproto->need_revalidate) {
5915 struct table_dpif *table = &ofproto->tables[rule->up.table_id];
5917 if (table->other_table && rule->tag) {
5918 tag_set_add(&ofproto->revalidate_set, rule->tag);
5920 ofproto->need_revalidate = true;
5926 set_frag_handling(struct ofproto *ofproto_,
5927 enum ofp_config_flags frag_handling)
5929 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
5931 if (frag_handling != OFPC_FRAG_REASM) {
5932 ofproto->need_revalidate = true;
5940 packet_out(struct ofproto *ofproto_, struct ofpbuf *packet,
5941 const struct flow *flow,
5942 const union ofp_action *ofp_actions, size_t n_ofp_actions)
5944 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
5947 if (flow->in_port >= ofproto->max_ports && flow->in_port < OFPP_MAX) {
5948 return OFPERR_NXBRC_BAD_IN_PORT;
5951 error = validate_actions(ofp_actions, n_ofp_actions, flow,
5952 ofproto->max_ports);
5954 struct odputil_keybuf keybuf;
5955 struct dpif_flow_stats stats;
5959 struct action_xlate_ctx ctx;
5960 uint64_t odp_actions_stub[1024 / 8];
5961 struct ofpbuf odp_actions;
5963 ofpbuf_use_stack(&key, &keybuf, sizeof keybuf);
5964 odp_flow_key_from_flow(&key, flow);
5966 dpif_flow_stats_extract(flow, packet, &stats);
5968 action_xlate_ctx_init(&ctx, ofproto, flow, flow->vlan_tci, NULL,
5969 packet_get_tcp_flags(packet, flow), packet);
5970 ctx.resubmit_stats = &stats;
5972 ofpbuf_use_stub(&odp_actions,
5973 odp_actions_stub, sizeof odp_actions_stub);
5974 xlate_actions(&ctx, ofp_actions, n_ofp_actions, &odp_actions);
5975 dpif_execute(ofproto->dpif, key.data, key.size,
5976 odp_actions.data, odp_actions.size, packet);
5977 ofpbuf_uninit(&odp_actions);
5985 set_netflow(struct ofproto *ofproto_,
5986 const struct netflow_options *netflow_options)
5988 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
5990 if (netflow_options) {
5991 if (!ofproto->netflow) {
5992 ofproto->netflow = netflow_create();
5994 return netflow_set_options(ofproto->netflow, netflow_options);
5996 netflow_destroy(ofproto->netflow);
5997 ofproto->netflow = NULL;
6003 get_netflow_ids(const struct ofproto *ofproto_,
6004 uint8_t *engine_type, uint8_t *engine_id)
6006 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
6008 dpif_get_netflow_ids(ofproto->dpif, engine_type, engine_id);
6012 send_active_timeout(struct ofproto_dpif *ofproto, struct facet *facet)
6014 if (!facet_is_controller_flow(facet) &&
6015 netflow_active_timeout_expired(ofproto->netflow, &facet->nf_flow)) {
6016 struct subfacet *subfacet;
6017 struct ofexpired expired;
6019 LIST_FOR_EACH (subfacet, list_node, &facet->subfacets) {
6020 if (subfacet->installed) {
6021 struct dpif_flow_stats stats;
6023 subfacet_install(subfacet, subfacet->actions,
6024 subfacet->actions_len, &stats);
6025 subfacet_update_stats(subfacet, &stats);
6029 expired.flow = facet->flow;
6030 expired.packet_count = facet->packet_count;
6031 expired.byte_count = facet->byte_count;
6032 expired.used = facet->used;
6033 netflow_expire(ofproto->netflow, &facet->nf_flow, &expired);
6038 send_netflow_active_timeouts(struct ofproto_dpif *ofproto)
6040 struct facet *facet;
6042 HMAP_FOR_EACH (facet, hmap_node, &ofproto->facets) {
6043 send_active_timeout(ofproto, facet);
6047 static struct ofproto_dpif *
6048 ofproto_dpif_lookup(const char *name)
6050 struct ofproto_dpif *ofproto;
6052 HMAP_FOR_EACH_WITH_HASH (ofproto, all_ofproto_dpifs_node,
6053 hash_string(name, 0), &all_ofproto_dpifs) {
6054 if (!strcmp(ofproto->up.name, name)) {
6062 ofproto_unixctl_fdb_flush(struct unixctl_conn *conn, int argc,
6063 const char *argv[], void *aux OVS_UNUSED)
6065 struct ofproto_dpif *ofproto;
6068 ofproto = ofproto_dpif_lookup(argv[1]);
6070 unixctl_command_reply_error(conn, "no such bridge");
6073 mac_learning_flush(ofproto->ml, &ofproto->revalidate_set);
6075 HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
6076 mac_learning_flush(ofproto->ml, &ofproto->revalidate_set);
6080 unixctl_command_reply(conn, "table successfully flushed");
6084 ofproto_unixctl_fdb_show(struct unixctl_conn *conn, int argc OVS_UNUSED,
6085 const char *argv[], void *aux OVS_UNUSED)
6087 struct ds ds = DS_EMPTY_INITIALIZER;
6088 const struct ofproto_dpif *ofproto;
6089 const struct mac_entry *e;
6091 ofproto = ofproto_dpif_lookup(argv[1]);
6093 unixctl_command_reply_error(conn, "no such bridge");
6097 ds_put_cstr(&ds, " port VLAN MAC Age\n");
6098 LIST_FOR_EACH (e, lru_node, &ofproto->ml->lrus) {
6099 struct ofbundle *bundle = e->port.p;
6100 ds_put_format(&ds, "%5d %4d "ETH_ADDR_FMT" %3d\n",
6101 ofbundle_get_a_port(bundle)->odp_port,
6102 e->vlan, ETH_ADDR_ARGS(e->mac),
6103 mac_entry_age(ofproto->ml, e));
6105 unixctl_command_reply(conn, ds_cstr(&ds));
6110 struct action_xlate_ctx ctx;
6116 trace_format_rule(struct ds *result, uint8_t table_id, int level,
6117 const struct rule_dpif *rule)
6119 ds_put_char_multiple(result, '\t', level);
6121 ds_put_cstr(result, "No match\n");
6125 ds_put_format(result, "Rule: table=%"PRIu8" cookie=%#"PRIx64" ",
6126 table_id, ntohll(rule->up.flow_cookie));
6127 cls_rule_format(&rule->up.cr, result);
6128 ds_put_char(result, '\n');
6130 ds_put_char_multiple(result, '\t', level);
6131 ds_put_cstr(result, "OpenFlow ");
6132 ofp_print_actions(result, rule->up.actions, rule->up.n_actions);
6133 ds_put_char(result, '\n');
6137 trace_format_flow(struct ds *result, int level, const char *title,
6138 struct trace_ctx *trace)
6140 ds_put_char_multiple(result, '\t', level);
6141 ds_put_format(result, "%s: ", title);
6142 if (flow_equal(&trace->ctx.flow, &trace->flow)) {
6143 ds_put_cstr(result, "unchanged");
6145 flow_format(result, &trace->ctx.flow);
6146 trace->flow = trace->ctx.flow;
6148 ds_put_char(result, '\n');
6152 trace_format_regs(struct ds *result, int level, const char *title,
6153 struct trace_ctx *trace)
6157 ds_put_char_multiple(result, '\t', level);
6158 ds_put_format(result, "%s:", title);
6159 for (i = 0; i < FLOW_N_REGS; i++) {
6160 ds_put_format(result, " reg%zu=0x%"PRIx32, i, trace->flow.regs[i]);
6162 ds_put_char(result, '\n');
6166 trace_format_odp(struct ds *result, int level, const char *title,
6167 struct trace_ctx *trace)
6169 struct ofpbuf *odp_actions = trace->ctx.odp_actions;
6171 ds_put_char_multiple(result, '\t', level);
6172 ds_put_format(result, "%s: ", title);
6173 format_odp_actions(result, odp_actions->data, odp_actions->size);
6174 ds_put_char(result, '\n');
6178 trace_resubmit(struct action_xlate_ctx *ctx, struct rule_dpif *rule)
6180 struct trace_ctx *trace = CONTAINER_OF(ctx, struct trace_ctx, ctx);
6181 struct ds *result = trace->result;
6183 ds_put_char(result, '\n');
6184 trace_format_flow(result, ctx->recurse + 1, "Resubmitted flow", trace);
6185 trace_format_regs(result, ctx->recurse + 1, "Resubmitted regs", trace);
6186 trace_format_odp(result, ctx->recurse + 1, "Resubmitted odp", trace);
6187 trace_format_rule(result, ctx->table_id, ctx->recurse + 1, rule);
6191 ofproto_unixctl_trace(struct unixctl_conn *conn, int argc, const char *argv[],
6192 void *aux OVS_UNUSED)
6194 const char *dpname = argv[1];
6195 struct ofproto_dpif *ofproto;
6196 struct ofpbuf odp_key;
6197 struct ofpbuf *packet;
6198 ovs_be16 initial_tci;
6204 ofpbuf_init(&odp_key, 0);
6207 ofproto = ofproto_dpif_lookup(dpname);
6209 unixctl_command_reply_error(conn, "Unknown ofproto (use ofproto/list "
6213 if (argc == 3 || (argc == 4 && !strcmp(argv[3], "-generate"))) {
6214 /* ofproto/trace dpname flow [-generate] */
6215 const char *flow_s = argv[2];
6216 const char *generate_s = argv[3];
6219 /* Convert string to datapath key. */
6220 ofpbuf_init(&odp_key, 0);
6221 error = odp_flow_key_from_string(flow_s, NULL, &odp_key);
6223 unixctl_command_reply_error(conn, "Bad flow syntax");
6227 /* Convert odp_key to flow. */
6228 error = ofproto_dpif_extract_flow_key(ofproto, odp_key.data,
6229 odp_key.size, &flow,
6230 &initial_tci, NULL);
6231 if (error == ODP_FIT_ERROR) {
6232 unixctl_command_reply_error(conn, "Invalid flow");
6236 /* Generate a packet, if requested. */
6238 packet = ofpbuf_new(0);
6239 flow_compose(packet, &flow);
6241 } else if (argc == 6) {
6242 /* ofproto/trace dpname priority tun_id in_port packet */
6243 const char *priority_s = argv[2];
6244 const char *tun_id_s = argv[3];
6245 const char *in_port_s = argv[4];
6246 const char *packet_s = argv[5];
6247 uint16_t in_port = ofp_port_to_odp_port(atoi(in_port_s));
6248 ovs_be64 tun_id = htonll(strtoull(tun_id_s, NULL, 0));
6249 uint32_t priority = atoi(priority_s);
6252 msg = eth_from_hex(packet_s, &packet);
6254 unixctl_command_reply_error(conn, msg);
6258 ds_put_cstr(&result, "Packet: ");
6259 s = ofp_packet_to_string(packet->data, packet->size);
6260 ds_put_cstr(&result, s);
6263 flow_extract(packet, priority, tun_id, in_port, &flow);
6264 initial_tci = flow.vlan_tci;
6266 unixctl_command_reply_error(conn, "Bad command syntax");
6270 ofproto_trace(ofproto, &flow, packet, initial_tci, &result);
6271 unixctl_command_reply(conn, ds_cstr(&result));
6274 ds_destroy(&result);
6275 ofpbuf_delete(packet);
6276 ofpbuf_uninit(&odp_key);
6280 ofproto_trace(struct ofproto_dpif *ofproto, const struct flow *flow,
6281 const struct ofpbuf *packet, ovs_be16 initial_tci,
6284 struct rule_dpif *rule;
6286 ds_put_cstr(ds, "Flow: ");
6287 flow_format(ds, flow);
6288 ds_put_char(ds, '\n');
6290 rule = rule_dpif_lookup(ofproto, flow, 0);
6291 trace_format_rule(ds, 0, 0, rule);
6293 uint64_t odp_actions_stub[1024 / 8];
6294 struct ofpbuf odp_actions;
6296 struct trace_ctx trace;
6299 tcp_flags = packet ? packet_get_tcp_flags(packet, flow) : 0;
6302 ofpbuf_use_stub(&odp_actions,
6303 odp_actions_stub, sizeof odp_actions_stub);
6304 action_xlate_ctx_init(&trace.ctx, ofproto, flow, initial_tci,
6305 rule, tcp_flags, packet);
6306 trace.ctx.resubmit_hook = trace_resubmit;
6307 xlate_actions(&trace.ctx, rule->up.actions, rule->up.n_actions,
6310 ds_put_char(ds, '\n');
6311 trace_format_flow(ds, 0, "Final flow", &trace);
6312 ds_put_cstr(ds, "Datapath actions: ");
6313 format_odp_actions(ds, odp_actions.data, odp_actions.size);
6314 ofpbuf_uninit(&odp_actions);
6316 if (!trace.ctx.may_set_up_flow) {
6318 ds_put_cstr(ds, "\nThis flow is not cachable.");
6320 ds_put_cstr(ds, "\nThe datapath actions are incomplete--"
6321 "for complete actions, please supply a packet.");
6328 ofproto_dpif_clog(struct unixctl_conn *conn OVS_UNUSED, int argc OVS_UNUSED,
6329 const char *argv[] OVS_UNUSED, void *aux OVS_UNUSED)
6332 unixctl_command_reply(conn, NULL);
6336 ofproto_dpif_unclog(struct unixctl_conn *conn OVS_UNUSED, int argc OVS_UNUSED,
6337 const char *argv[] OVS_UNUSED, void *aux OVS_UNUSED)
6340 unixctl_command_reply(conn, NULL);
6343 /* Runs a self-check of flow translations in 'ofproto'. Appends a message to
6344 * 'reply' describing the results. */
6346 ofproto_dpif_self_check__(struct ofproto_dpif *ofproto, struct ds *reply)
6348 struct facet *facet;
6352 HMAP_FOR_EACH (facet, hmap_node, &ofproto->facets) {
6353 if (!facet_check_consistency(facet)) {
6358 ofproto->need_revalidate = true;
6362 ds_put_format(reply, "%s: self-check failed (%d errors)\n",
6363 ofproto->up.name, errors);
6365 ds_put_format(reply, "%s: self-check passed\n", ofproto->up.name);
6370 ofproto_dpif_self_check(struct unixctl_conn *conn,
6371 int argc, const char *argv[], void *aux OVS_UNUSED)
6373 struct ds reply = DS_EMPTY_INITIALIZER;
6374 struct ofproto_dpif *ofproto;
6377 ofproto = ofproto_dpif_lookup(argv[1]);
6379 unixctl_command_reply_error(conn, "Unknown ofproto (use "
6380 "ofproto/list for help)");
6383 ofproto_dpif_self_check__(ofproto, &reply);
6385 HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
6386 ofproto_dpif_self_check__(ofproto, &reply);
6390 unixctl_command_reply(conn, ds_cstr(&reply));
6395 ofproto_dpif_unixctl_init(void)
6397 static bool registered;
6403 unixctl_command_register(
6405 "bridge {tun_id in_port packet | odp_flow [-generate]}",
6406 2, 5, ofproto_unixctl_trace, NULL);
6407 unixctl_command_register("fdb/flush", "[bridge]", 0, 1,
6408 ofproto_unixctl_fdb_flush, NULL);
6409 unixctl_command_register("fdb/show", "bridge", 1, 1,
6410 ofproto_unixctl_fdb_show, NULL);
6411 unixctl_command_register("ofproto/clog", "", 0, 0,
6412 ofproto_dpif_clog, NULL);
6413 unixctl_command_register("ofproto/unclog", "", 0, 0,
6414 ofproto_dpif_unclog, NULL);
6415 unixctl_command_register("ofproto/self-check", "[bridge]", 0, 1,
6416 ofproto_dpif_self_check, NULL);
6419 /* Linux VLAN device support (e.g. "eth0.10" for VLAN 10.)
6421 * This is deprecated. It is only for compatibility with broken device drivers
6422 * in old versions of Linux that do not properly support VLANs when VLAN
6423 * devices are not used. When broken device drivers are no longer in
6424 * widespread use, we will delete these interfaces. */
6427 set_realdev(struct ofport *ofport_, uint16_t realdev_ofp_port, int vid)
6429 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport_->ofproto);
6430 struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
6432 if (realdev_ofp_port == ofport->realdev_ofp_port
6433 && vid == ofport->vlandev_vid) {
6437 ofproto->need_revalidate = true;
6439 if (ofport->realdev_ofp_port) {
6442 if (realdev_ofp_port && ofport->bundle) {
6443 /* vlandevs are enslaved to their realdevs, so they are not allowed to
6444 * themselves be part of a bundle. */
6445 bundle_set(ofport->up.ofproto, ofport->bundle, NULL);
6448 ofport->realdev_ofp_port = realdev_ofp_port;
6449 ofport->vlandev_vid = vid;
6451 if (realdev_ofp_port) {
6452 vsp_add(ofport, realdev_ofp_port, vid);
6459 hash_realdev_vid(uint16_t realdev_ofp_port, int vid)
6461 return hash_2words(realdev_ofp_port, vid);
6464 /* Returns the ODP port number of the Linux VLAN device that corresponds to
6465 * 'vlan_tci' on the network device with port number 'realdev_odp_port' in
6466 * 'ofproto'. For example, given 'realdev_odp_port' of eth0 and 'vlan_tci' 9,
6467 * it would return the port number of eth0.9.
6469 * Unless VLAN splinters are enabled for port 'realdev_odp_port', this
6470 * function just returns its 'realdev_odp_port' argument. */
6472 vsp_realdev_to_vlandev(const struct ofproto_dpif *ofproto,
6473 uint32_t realdev_odp_port, ovs_be16 vlan_tci)
6475 if (!hmap_is_empty(&ofproto->realdev_vid_map)) {
6476 uint16_t realdev_ofp_port = odp_port_to_ofp_port(realdev_odp_port);
6477 int vid = vlan_tci_to_vid(vlan_tci);
6478 const struct vlan_splinter *vsp;
6480 HMAP_FOR_EACH_WITH_HASH (vsp, realdev_vid_node,
6481 hash_realdev_vid(realdev_ofp_port, vid),
6482 &ofproto->realdev_vid_map) {
6483 if (vsp->realdev_ofp_port == realdev_ofp_port
6484 && vsp->vid == vid) {
6485 return ofp_port_to_odp_port(vsp->vlandev_ofp_port);
6489 return realdev_odp_port;
6492 static struct vlan_splinter *
6493 vlandev_find(const struct ofproto_dpif *ofproto, uint16_t vlandev_ofp_port)
6495 struct vlan_splinter *vsp;
6497 HMAP_FOR_EACH_WITH_HASH (vsp, vlandev_node, hash_int(vlandev_ofp_port, 0),
6498 &ofproto->vlandev_map) {
6499 if (vsp->vlandev_ofp_port == vlandev_ofp_port) {
6507 /* Returns the OpenFlow port number of the "real" device underlying the Linux
6508 * VLAN device with OpenFlow port number 'vlandev_ofp_port' and stores the
6509 * VLAN VID of the Linux VLAN device in '*vid'. For example, given
6510 * 'vlandev_ofp_port' of eth0.9, it would return the OpenFlow port number of
6511 * eth0 and store 9 in '*vid'.
6513 * Returns 0 and does not modify '*vid' if 'vlandev_ofp_port' is not a Linux
6514 * VLAN device. Unless VLAN splinters are enabled, this is what this function
6517 vsp_vlandev_to_realdev(const struct ofproto_dpif *ofproto,
6518 uint16_t vlandev_ofp_port, int *vid)
6520 if (!hmap_is_empty(&ofproto->vlandev_map)) {
6521 const struct vlan_splinter *vsp;
6523 vsp = vlandev_find(ofproto, vlandev_ofp_port);
6528 return vsp->realdev_ofp_port;
6535 vsp_remove(struct ofport_dpif *port)
6537 struct ofproto_dpif *ofproto = ofproto_dpif_cast(port->up.ofproto);
6538 struct vlan_splinter *vsp;
6540 vsp = vlandev_find(ofproto, port->up.ofp_port);
6542 hmap_remove(&ofproto->vlandev_map, &vsp->vlandev_node);
6543 hmap_remove(&ofproto->realdev_vid_map, &vsp->realdev_vid_node);
6546 port->realdev_ofp_port = 0;
6548 VLOG_ERR("missing vlan device record");
6553 vsp_add(struct ofport_dpif *port, uint16_t realdev_ofp_port, int vid)
6555 struct ofproto_dpif *ofproto = ofproto_dpif_cast(port->up.ofproto);
6557 if (!vsp_vlandev_to_realdev(ofproto, port->up.ofp_port, NULL)
6558 && (vsp_realdev_to_vlandev(ofproto, realdev_ofp_port, htons(vid))
6559 == realdev_ofp_port)) {
6560 struct vlan_splinter *vsp;
6562 vsp = xmalloc(sizeof *vsp);
6563 hmap_insert(&ofproto->vlandev_map, &vsp->vlandev_node,
6564 hash_int(port->up.ofp_port, 0));
6565 hmap_insert(&ofproto->realdev_vid_map, &vsp->realdev_vid_node,
6566 hash_realdev_vid(realdev_ofp_port, vid));
6567 vsp->realdev_ofp_port = realdev_ofp_port;
6568 vsp->vlandev_ofp_port = port->up.ofp_port;
6571 port->realdev_ofp_port = realdev_ofp_port;
6573 VLOG_ERR("duplicate vlan device record");
6577 const struct ofproto_class ofproto_dpif_class = {
6606 port_is_lacp_current,
6607 NULL, /* rule_choose_table */
6614 rule_modify_actions,
6622 get_cfm_remote_mpids,
6627 get_stp_port_status,
6634 is_mirror_output_bundle,
6635 forward_bpdu_changed,