2 * Copyright (c) 2009, 2010, 2011, 2012 Nicira Networks.
4 * Licensed under the Apache License, Version 2.0 (the "License");
5 * you may not use this file except in compliance with the License.
6 * You may obtain a copy of the License at:
8 * http://www.apache.org/licenses/LICENSE-2.0
10 * Unless required by applicable law or agreed to in writing, software
11 * distributed under the License is distributed on an "AS IS" BASIS,
12 * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
13 * See the License for the specific language governing permissions and
14 * limitations under the License.
19 #include "ofproto/ofproto-provider.h"
26 #include "byte-order.h"
31 #include "dynamic-string.h"
32 #include "fail-open.h"
36 #include "mac-learning.h"
37 #include "meta-flow.h"
38 #include "multipath.h"
45 #include "ofp-print.h"
46 #include "ofproto-dpif-sflow.h"
47 #include "poll-loop.h"
49 #include "unaligned.h"
51 #include "vlan-bitmap.h"
54 VLOG_DEFINE_THIS_MODULE(ofproto_dpif);
56 COVERAGE_DEFINE(ofproto_dpif_ctlr_action);
57 COVERAGE_DEFINE(ofproto_dpif_expired);
58 COVERAGE_DEFINE(ofproto_dpif_no_packet_in);
59 COVERAGE_DEFINE(ofproto_dpif_xlate);
60 COVERAGE_DEFINE(facet_changed_rule);
61 COVERAGE_DEFINE(facet_invalidated);
62 COVERAGE_DEFINE(facet_revalidate);
63 COVERAGE_DEFINE(facet_unexpected);
65 /* Maximum depth of flow table recursion (due to resubmit actions) in a
66 * flow translation. */
67 #define MAX_RESUBMIT_RECURSION 32
69 /* Number of implemented OpenFlow tables. */
70 enum { N_TABLES = 255 };
71 BUILD_ASSERT_DECL(N_TABLES >= 1 && N_TABLES <= 255);
81 * - Do include packets and bytes from facets that have been deleted or
82 * whose own statistics have been folded into the rule.
84 * - Do include packets and bytes sent "by hand" that were accounted to
85 * the rule without any facet being involved (this is a rare corner
86 * case in rule_execute()).
88 * - Do not include packet or bytes that can be obtained from any facet's
89 * packet_count or byte_count member or that can be obtained from the
90 * datapath by, e.g., dpif_flow_get() for any subfacet.
92 uint64_t packet_count; /* Number of packets received. */
93 uint64_t byte_count; /* Number of bytes received. */
95 tag_type tag; /* Caches rule_calculate_tag() result. */
97 struct list facets; /* List of "struct facet"s. */
100 static struct rule_dpif *rule_dpif_cast(const struct rule *rule)
102 return rule ? CONTAINER_OF(rule, struct rule_dpif, up) : NULL;
105 static struct rule_dpif *rule_dpif_lookup(struct ofproto_dpif *,
106 const struct flow *, uint8_t table);
108 static void flow_push_stats(struct rule_dpif *, const struct flow *,
109 uint64_t packets, uint64_t bytes,
112 static uint32_t rule_calculate_tag(const struct flow *,
113 const struct flow_wildcards *,
115 static void rule_invalidate(const struct rule_dpif *);
117 #define MAX_MIRRORS 32
118 typedef uint32_t mirror_mask_t;
119 #define MIRROR_MASK_C(X) UINT32_C(X)
120 BUILD_ASSERT_DECL(sizeof(mirror_mask_t) * CHAR_BIT >= MAX_MIRRORS);
122 struct ofproto_dpif *ofproto; /* Owning ofproto. */
123 size_t idx; /* In ofproto's "mirrors" array. */
124 void *aux; /* Key supplied by ofproto's client. */
125 char *name; /* Identifier for log messages. */
127 /* Selection criteria. */
128 struct hmapx srcs; /* Contains "struct ofbundle *"s. */
129 struct hmapx dsts; /* Contains "struct ofbundle *"s. */
130 unsigned long *vlans; /* Bitmap of chosen VLANs, NULL selects all. */
132 /* Output (exactly one of out == NULL and out_vlan == -1 is true). */
133 struct ofbundle *out; /* Output port or NULL. */
134 int out_vlan; /* Output VLAN or -1. */
135 mirror_mask_t dup_mirrors; /* Bitmap of mirrors with the same output. */
138 int64_t packet_count; /* Number of packets sent. */
139 int64_t byte_count; /* Number of bytes sent. */
142 static void mirror_destroy(struct ofmirror *);
143 static void update_mirror_stats(struct ofproto_dpif *ofproto,
144 mirror_mask_t mirrors,
145 uint64_t packets, uint64_t bytes);
148 struct ofproto_dpif *ofproto; /* Owning ofproto. */
149 struct hmap_node hmap_node; /* In struct ofproto's "bundles" hmap. */
150 void *aux; /* Key supplied by ofproto's client. */
151 char *name; /* Identifier for log messages. */
154 struct list ports; /* Contains "struct ofport"s. */
155 enum port_vlan_mode vlan_mode; /* VLAN mode */
156 int vlan; /* -1=trunk port, else a 12-bit VLAN ID. */
157 unsigned long *trunks; /* Bitmap of trunked VLANs, if 'vlan' == -1.
158 * NULL if all VLANs are trunked. */
159 struct lacp *lacp; /* LACP if LACP is enabled, otherwise NULL. */
160 struct bond *bond; /* Nonnull iff more than one port. */
161 bool use_priority_tags; /* Use 802.1p tag for frames in VLAN 0? */
164 bool floodable; /* True if no port has OFPUTIL_PC_NO_FLOOD set. */
166 /* Port mirroring info. */
167 mirror_mask_t src_mirrors; /* Mirrors triggered when packet received. */
168 mirror_mask_t dst_mirrors; /* Mirrors triggered when packet sent. */
169 mirror_mask_t mirror_out; /* Mirrors that output to this bundle. */
172 static void bundle_remove(struct ofport *);
173 static void bundle_update(struct ofbundle *);
174 static void bundle_destroy(struct ofbundle *);
175 static void bundle_del_port(struct ofport_dpif *);
176 static void bundle_run(struct ofbundle *);
177 static void bundle_wait(struct ofbundle *);
178 static struct ofbundle *lookup_input_bundle(struct ofproto_dpif *,
179 uint16_t in_port, bool warn);
181 /* A controller may use OFPP_NONE as the ingress port to indicate that
182 * it did not arrive on a "real" port. 'ofpp_none_bundle' exists for
183 * when an input bundle is needed for validation (e.g., mirroring or
184 * OFPP_NORMAL processing). It is not connected to an 'ofproto' or have
185 * any 'port' structs, so care must be taken when dealing with it. */
186 static struct ofbundle ofpp_none_bundle = {
188 .vlan_mode = PORT_VLAN_TRUNK
191 static void stp_run(struct ofproto_dpif *ofproto);
192 static void stp_wait(struct ofproto_dpif *ofproto);
193 static int set_stp_port(struct ofport *,
194 const struct ofproto_port_stp_settings *);
196 static bool ofbundle_includes_vlan(const struct ofbundle *, uint16_t vlan);
198 struct action_xlate_ctx {
199 /* action_xlate_ctx_init() initializes these members. */
202 struct ofproto_dpif *ofproto;
204 /* Flow to which the OpenFlow actions apply. xlate_actions() will modify
205 * this flow when actions change header fields. */
208 /* The packet corresponding to 'flow', or a null pointer if we are
209 * revalidating without a packet to refer to. */
210 const struct ofpbuf *packet;
212 /* Should OFPP_NORMAL MAC learning and NXAST_LEARN actions execute? We
213 * want to execute them if we are actually processing a packet, or if we
214 * are accounting for packets that the datapath has processed, but not if
215 * we are just revalidating. */
218 /* The rule that we are currently translating, or NULL. */
219 struct rule_dpif *rule;
221 /* Union of the set of TCP flags seen so far in this flow. (Used only by
222 * NXAST_FIN_TIMEOUT. Set to zero to avoid updating updating rules'
226 /* If nonnull, called just before executing a resubmit action. In
227 * addition, disables logging of traces when the recursion depth is
230 * This is normally null so the client has to set it manually after
231 * calling action_xlate_ctx_init(). */
232 void (*resubmit_hook)(struct action_xlate_ctx *, struct rule_dpif *);
234 /* xlate_actions() initializes and uses these members. The client might want
235 * to look at them after it returns. */
237 struct ofpbuf *odp_actions; /* Datapath actions. */
238 tag_type tags; /* Tags associated with actions. */
239 bool may_set_up_flow; /* True ordinarily; false if the actions must
240 * be reassessed for every packet. */
241 bool has_learn; /* Actions include NXAST_LEARN? */
242 bool has_normal; /* Actions output to OFPP_NORMAL? */
243 bool has_fin_timeout; /* Actions include NXAST_FIN_TIMEOUT? */
244 uint16_t nf_output_iface; /* Output interface index for NetFlow. */
245 mirror_mask_t mirrors; /* Bitmap of associated mirrors. */
247 /* xlate_actions() initializes and uses these members, but the client has no
248 * reason to look at them. */
250 int recurse; /* Recursion level, via xlate_table_action. */
251 bool max_resubmit_trigger; /* Recursed too deeply during translation. */
252 struct flow base_flow; /* Flow at the last commit. */
253 uint32_t orig_skb_priority; /* Priority when packet arrived. */
254 uint8_t table_id; /* OpenFlow table ID where flow was found. */
255 uint32_t sflow_n_outputs; /* Number of output ports. */
256 uint16_t sflow_odp_port; /* Output port for composing sFlow action. */
257 uint16_t user_cookie_offset;/* Used for user_action_cookie fixup. */
258 bool exit; /* No further actions should be processed. */
261 static void action_xlate_ctx_init(struct action_xlate_ctx *,
262 struct ofproto_dpif *, const struct flow *,
263 ovs_be16 initial_tci, struct rule_dpif *,
264 uint8_t tcp_flags, const struct ofpbuf *);
265 static struct ofpbuf *xlate_actions(struct action_xlate_ctx *,
266 const union ofp_action *in, size_t n_in);
268 /* An exact-match instantiation of an OpenFlow flow.
270 * A facet associates a "struct flow", which represents the Open vSwitch
271 * userspace idea of an exact-match flow, with one or more subfacets. Each
272 * subfacet tracks the datapath's idea of the exact-match flow equivalent to
273 * the facet. When the kernel module (or other dpif implementation) and Open
274 * vSwitch userspace agree on the definition of a flow key, there is exactly
275 * one subfacet per facet. If the dpif implementation supports more-specific
276 * flow matching than userspace, however, a facet can have more than one
277 * subfacet, each of which corresponds to some distinction in flow that
278 * userspace simply doesn't understand.
280 * Flow expiration works in terms of subfacets, so a facet must have at least
281 * one subfacet or it will never expire, leaking memory. */
284 struct hmap_node hmap_node; /* In owning ofproto's 'facets' hmap. */
285 struct list list_node; /* In owning rule's 'facets' list. */
286 struct rule_dpif *rule; /* Owning rule. */
289 struct list subfacets;
290 long long int used; /* Time last used; time created if not used. */
297 * - Do include packets and bytes sent "by hand", e.g. with
300 * - Do include packets and bytes that were obtained from the datapath
301 * when a subfacet's statistics were reset (e.g. dpif_flow_put() with
302 * DPIF_FP_ZERO_STATS).
304 * - Do not include packets or bytes that can be obtained from the
305 * datapath for any existing subfacet.
307 uint64_t packet_count; /* Number of packets received. */
308 uint64_t byte_count; /* Number of bytes received. */
310 /* Resubmit statistics. */
311 uint64_t prev_packet_count; /* Number of packets from last stats push. */
312 uint64_t prev_byte_count; /* Number of bytes from last stats push. */
313 long long int prev_used; /* Used time from last stats push. */
316 uint64_t accounted_bytes; /* Bytes processed by facet_account(). */
317 struct netflow_flow nf_flow; /* Per-flow NetFlow tracking data. */
318 uint8_t tcp_flags; /* TCP flags seen for this 'rule'. */
320 /* Properties of datapath actions.
322 * Every subfacet has its own actions because actions can differ slightly
323 * between splintered and non-splintered subfacets due to the VLAN tag
324 * being initially different (present vs. absent). All of them have these
325 * properties in common so we just store one copy of them here. */
326 bool may_install; /* Reassess actions for every packet? */
327 bool has_learn; /* Actions include NXAST_LEARN? */
328 bool has_normal; /* Actions output to OFPP_NORMAL? */
329 bool has_fin_timeout; /* Actions include NXAST_FIN_TIMEOUT? */
330 tag_type tags; /* Tags that would require revalidation. */
331 mirror_mask_t mirrors; /* Bitmap of dependent mirrors. */
334 static struct facet *facet_create(struct rule_dpif *, const struct flow *);
335 static void facet_remove(struct facet *);
336 static void facet_free(struct facet *);
338 static struct facet *facet_find(struct ofproto_dpif *, const struct flow *);
339 static struct facet *facet_lookup_valid(struct ofproto_dpif *,
340 const struct flow *);
341 static bool facet_revalidate(struct facet *);
342 static bool facet_check_consistency(struct facet *);
344 static void facet_flush_stats(struct facet *);
346 static void facet_update_time(struct facet *, long long int used);
347 static void facet_reset_counters(struct facet *);
348 static void facet_push_stats(struct facet *);
349 static void facet_account(struct facet *);
351 static bool facet_is_controller_flow(struct facet *);
353 /* A dpif flow and actions associated with a facet.
355 * See also the large comment on struct facet. */
358 struct hmap_node hmap_node; /* In struct ofproto_dpif 'subfacets' list. */
359 struct list list_node; /* In struct facet's 'facets' list. */
360 struct facet *facet; /* Owning facet. */
364 * To save memory in the common case, 'key' is NULL if 'key_fitness' is
365 * ODP_FIT_PERFECT, that is, odp_flow_key_from_flow() can accurately
366 * regenerate the ODP flow key from ->facet->flow. */
367 enum odp_key_fitness key_fitness;
371 long long int used; /* Time last used; time created if not used. */
373 uint64_t dp_packet_count; /* Last known packet count in the datapath. */
374 uint64_t dp_byte_count; /* Last known byte count in the datapath. */
378 * These should be essentially identical for every subfacet in a facet, but
379 * may differ in trivial ways due to VLAN splinters. */
380 size_t actions_len; /* Number of bytes in actions[]. */
381 struct nlattr *actions; /* Datapath actions. */
383 bool installed; /* Installed in datapath? */
385 /* This value is normally the same as ->facet->flow.vlan_tci. Only VLAN
386 * splinters can cause it to differ. This value should be removed when
387 * the VLAN splinters feature is no longer needed. */
388 ovs_be16 initial_tci; /* Initial VLAN TCI value. */
391 static struct subfacet *subfacet_create(struct facet *, enum odp_key_fitness,
392 const struct nlattr *key,
393 size_t key_len, ovs_be16 initial_tci);
394 static struct subfacet *subfacet_find(struct ofproto_dpif *,
395 const struct nlattr *key, size_t key_len);
396 static void subfacet_destroy(struct subfacet *);
397 static void subfacet_destroy__(struct subfacet *);
398 static void subfacet_get_key(struct subfacet *, struct odputil_keybuf *,
400 static void subfacet_reset_dp_stats(struct subfacet *,
401 struct dpif_flow_stats *);
402 static void subfacet_update_time(struct subfacet *, long long int used);
403 static void subfacet_update_stats(struct subfacet *,
404 const struct dpif_flow_stats *);
405 static void subfacet_make_actions(struct subfacet *,
406 const struct ofpbuf *packet);
407 static int subfacet_install(struct subfacet *,
408 const struct nlattr *actions, size_t actions_len,
409 struct dpif_flow_stats *);
410 static void subfacet_uninstall(struct subfacet *);
416 struct ofbundle *bundle; /* Bundle that contains this port, if any. */
417 struct list bundle_node; /* In struct ofbundle's "ports" list. */
418 struct cfm *cfm; /* Connectivity Fault Management, if any. */
419 tag_type tag; /* Tag associated with this port. */
420 uint32_t bond_stable_id; /* stable_id to use as bond slave, or 0. */
421 bool may_enable; /* May be enabled in bonds. */
422 long long int carrier_seq; /* Carrier status changes. */
425 struct stp_port *stp_port; /* Spanning Tree Protocol, if any. */
426 enum stp_state stp_state; /* Always STP_DISABLED if STP not in use. */
427 long long int stp_state_entered;
429 struct hmap priorities; /* Map of attached 'priority_to_dscp's. */
431 /* Linux VLAN device support (e.g. "eth0.10" for VLAN 10.)
433 * This is deprecated. It is only for compatibility with broken device
434 * drivers in old versions of Linux that do not properly support VLANs when
435 * VLAN devices are not used. When broken device drivers are no longer in
436 * widespread use, we will delete these interfaces. */
437 uint16_t realdev_ofp_port;
441 /* Node in 'ofport_dpif''s 'priorities' map. Used to maintain a map from
442 * 'priority' (the datapath's term for QoS queue) to the dscp bits which all
443 * traffic egressing the 'ofport' with that priority should be marked with. */
444 struct priority_to_dscp {
445 struct hmap_node hmap_node; /* Node in 'ofport_dpif''s 'priorities' map. */
446 uint32_t priority; /* Priority of this queue (see struct flow). */
448 uint8_t dscp; /* DSCP bits to mark outgoing traffic with. */
451 /* Linux VLAN device support (e.g. "eth0.10" for VLAN 10.)
453 * This is deprecated. It is only for compatibility with broken device drivers
454 * in old versions of Linux that do not properly support VLANs when VLAN
455 * devices are not used. When broken device drivers are no longer in
456 * widespread use, we will delete these interfaces. */
457 struct vlan_splinter {
458 struct hmap_node realdev_vid_node;
459 struct hmap_node vlandev_node;
460 uint16_t realdev_ofp_port;
461 uint16_t vlandev_ofp_port;
465 static uint32_t vsp_realdev_to_vlandev(const struct ofproto_dpif *,
466 uint32_t realdev, ovs_be16 vlan_tci);
467 static uint16_t vsp_vlandev_to_realdev(const struct ofproto_dpif *,
468 uint16_t vlandev, int *vid);
469 static void vsp_remove(struct ofport_dpif *);
470 static void vsp_add(struct ofport_dpif *, uint16_t realdev_ofp_port, int vid);
472 static struct ofport_dpif *
473 ofport_dpif_cast(const struct ofport *ofport)
475 assert(ofport->ofproto->ofproto_class == &ofproto_dpif_class);
476 return ofport ? CONTAINER_OF(ofport, struct ofport_dpif, up) : NULL;
479 static void port_run(struct ofport_dpif *);
480 static void port_wait(struct ofport_dpif *);
481 static int set_cfm(struct ofport *, const struct cfm_settings *);
482 static void ofport_clear_priorities(struct ofport_dpif *);
484 struct dpif_completion {
485 struct list list_node;
486 struct ofoperation *op;
489 /* Extra information about a classifier table.
490 * Currently used just for optimized flow revalidation. */
492 /* If either of these is nonnull, then this table has a form that allows
493 * flows to be tagged to avoid revalidating most flows for the most common
494 * kinds of flow table changes. */
495 struct cls_table *catchall_table; /* Table that wildcards all fields. */
496 struct cls_table *other_table; /* Table with any other wildcard set. */
497 uint32_t basis; /* Keeps each table's tags separate. */
500 struct ofproto_dpif {
501 struct hmap_node all_ofproto_dpifs_node; /* In 'all_ofproto_dpifs'. */
510 struct netflow *netflow;
511 struct dpif_sflow *sflow;
512 struct hmap bundles; /* Contains "struct ofbundle"s. */
513 struct mac_learning *ml;
514 struct ofmirror *mirrors[MAX_MIRRORS];
515 bool has_bonded_bundles;
518 struct timer next_expiration;
522 struct hmap subfacets;
525 struct table_dpif tables[N_TABLES];
526 bool need_revalidate;
527 struct tag_set revalidate_set;
529 /* Support for debugging async flow mods. */
530 struct list completions;
532 bool has_bundle_action; /* True when the first bundle action appears. */
533 struct netdev_stats stats; /* To account packets generated and consumed in
538 long long int stp_last_tick;
540 /* VLAN splinters. */
541 struct hmap realdev_vid_map; /* (realdev,vid) -> vlandev. */
542 struct hmap vlandev_map; /* vlandev -> (realdev,vid). */
545 /* Defer flow mod completion until "ovs-appctl ofproto/unclog"? (Useful only
546 * for debugging the asynchronous flow_mod implementation.) */
549 /* All existing ofproto_dpif instances, indexed by ->up.name. */
550 static struct hmap all_ofproto_dpifs = HMAP_INITIALIZER(&all_ofproto_dpifs);
552 static void ofproto_dpif_unixctl_init(void);
554 static struct ofproto_dpif *
555 ofproto_dpif_cast(const struct ofproto *ofproto)
557 assert(ofproto->ofproto_class == &ofproto_dpif_class);
558 return CONTAINER_OF(ofproto, struct ofproto_dpif, up);
561 static struct ofport_dpif *get_ofp_port(struct ofproto_dpif *,
563 static struct ofport_dpif *get_odp_port(struct ofproto_dpif *,
565 static void ofproto_trace(struct ofproto_dpif *, const struct flow *,
566 const struct ofpbuf *, ovs_be16 initial_tci,
569 /* Packet processing. */
570 static void update_learning_table(struct ofproto_dpif *,
571 const struct flow *, int vlan,
574 #define FLOW_MISS_MAX_BATCH 50
575 static int handle_upcalls(struct ofproto_dpif *, unsigned int max_batch);
577 /* Flow expiration. */
578 static int expire(struct ofproto_dpif *);
581 static void send_netflow_active_timeouts(struct ofproto_dpif *);
584 static int send_packet(const struct ofport_dpif *, struct ofpbuf *packet);
586 compose_sflow_action(const struct ofproto_dpif *, struct ofpbuf *odp_actions,
587 const struct flow *, uint32_t odp_port);
588 static void add_mirror_actions(struct action_xlate_ctx *ctx,
589 const struct flow *flow);
590 /* Global variables. */
591 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
593 /* Factory functions. */
596 enumerate_types(struct sset *types)
598 dp_enumerate_types(types);
602 enumerate_names(const char *type, struct sset *names)
604 return dp_enumerate_names(type, names);
608 del(const char *type, const char *name)
613 error = dpif_open(name, type, &dpif);
615 error = dpif_delete(dpif);
621 /* Basic life-cycle. */
623 static struct ofproto *
626 struct ofproto_dpif *ofproto = xmalloc(sizeof *ofproto);
631 dealloc(struct ofproto *ofproto_)
633 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
638 construct(struct ofproto *ofproto_)
640 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
641 const char *name = ofproto->up.name;
645 error = dpif_create_and_open(name, ofproto->up.type, &ofproto->dpif);
647 VLOG_ERR("failed to open datapath %s: %s", name, strerror(error));
651 ofproto->max_ports = dpif_get_max_ports(ofproto->dpif);
652 ofproto->n_matches = 0;
654 dpif_flow_flush(ofproto->dpif);
655 dpif_recv_purge(ofproto->dpif);
657 error = dpif_recv_set(ofproto->dpif, true);
659 VLOG_ERR("failed to listen on datapath %s: %s", name, strerror(error));
660 dpif_close(ofproto->dpif);
664 ofproto->netflow = NULL;
665 ofproto->sflow = NULL;
667 hmap_init(&ofproto->bundles);
668 ofproto->ml = mac_learning_create(MAC_ENTRY_DEFAULT_IDLE_TIME);
669 for (i = 0; i < MAX_MIRRORS; i++) {
670 ofproto->mirrors[i] = NULL;
672 ofproto->has_bonded_bundles = false;
674 timer_set_duration(&ofproto->next_expiration, 1000);
676 hmap_init(&ofproto->facets);
677 hmap_init(&ofproto->subfacets);
679 for (i = 0; i < N_TABLES; i++) {
680 struct table_dpif *table = &ofproto->tables[i];
682 table->catchall_table = NULL;
683 table->other_table = NULL;
684 table->basis = random_uint32();
686 ofproto->need_revalidate = false;
687 tag_set_init(&ofproto->revalidate_set);
689 list_init(&ofproto->completions);
691 ofproto_dpif_unixctl_init();
693 ofproto->has_bundle_action = false;
695 hmap_init(&ofproto->vlandev_map);
696 hmap_init(&ofproto->realdev_vid_map);
698 hmap_insert(&all_ofproto_dpifs, &ofproto->all_ofproto_dpifs_node,
699 hash_string(ofproto->up.name, 0));
700 memset(&ofproto->stats, 0, sizeof ofproto->stats);
702 ofproto_init_tables(ofproto_, N_TABLES);
708 complete_operations(struct ofproto_dpif *ofproto)
710 struct dpif_completion *c, *next;
712 LIST_FOR_EACH_SAFE (c, next, list_node, &ofproto->completions) {
713 ofoperation_complete(c->op, 0);
714 list_remove(&c->list_node);
720 destruct(struct ofproto *ofproto_)
722 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
723 struct rule_dpif *rule, *next_rule;
724 struct oftable *table;
727 hmap_remove(&all_ofproto_dpifs, &ofproto->all_ofproto_dpifs_node);
728 complete_operations(ofproto);
730 OFPROTO_FOR_EACH_TABLE (table, &ofproto->up) {
731 struct cls_cursor cursor;
733 cls_cursor_init(&cursor, &table->cls, NULL);
734 CLS_CURSOR_FOR_EACH_SAFE (rule, next_rule, up.cr, &cursor) {
735 ofproto_rule_destroy(&rule->up);
739 for (i = 0; i < MAX_MIRRORS; i++) {
740 mirror_destroy(ofproto->mirrors[i]);
743 netflow_destroy(ofproto->netflow);
744 dpif_sflow_destroy(ofproto->sflow);
745 hmap_destroy(&ofproto->bundles);
746 mac_learning_destroy(ofproto->ml);
748 hmap_destroy(&ofproto->facets);
749 hmap_destroy(&ofproto->subfacets);
751 hmap_destroy(&ofproto->vlandev_map);
752 hmap_destroy(&ofproto->realdev_vid_map);
754 dpif_close(ofproto->dpif);
758 run_fast(struct ofproto *ofproto_)
760 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
763 /* Handle one or more batches of upcalls, until there's nothing left to do
764 * or until we do a fixed total amount of work.
766 * We do work in batches because it can be much cheaper to set up a number
767 * of flows and fire off their patches all at once. We do multiple batches
768 * because in some cases handling a packet can cause another packet to be
769 * queued almost immediately as part of the return flow. Both
770 * optimizations can make major improvements on some benchmarks and
771 * presumably for real traffic as well. */
773 while (work < FLOW_MISS_MAX_BATCH) {
774 int retval = handle_upcalls(ofproto, FLOW_MISS_MAX_BATCH - work);
784 run(struct ofproto *ofproto_)
786 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
787 struct ofport_dpif *ofport;
788 struct ofbundle *bundle;
792 complete_operations(ofproto);
794 dpif_run(ofproto->dpif);
796 error = run_fast(ofproto_);
801 if (timer_expired(&ofproto->next_expiration)) {
802 int delay = expire(ofproto);
803 timer_set_duration(&ofproto->next_expiration, delay);
806 if (ofproto->netflow) {
807 if (netflow_run(ofproto->netflow)) {
808 send_netflow_active_timeouts(ofproto);
811 if (ofproto->sflow) {
812 dpif_sflow_run(ofproto->sflow);
815 HMAP_FOR_EACH (ofport, up.hmap_node, &ofproto->up.ports) {
818 HMAP_FOR_EACH (bundle, hmap_node, &ofproto->bundles) {
823 mac_learning_run(ofproto->ml, &ofproto->revalidate_set);
825 /* Now revalidate if there's anything to do. */
826 if (ofproto->need_revalidate
827 || !tag_set_is_empty(&ofproto->revalidate_set)) {
828 struct tag_set revalidate_set = ofproto->revalidate_set;
829 bool revalidate_all = ofproto->need_revalidate;
830 struct facet *facet, *next;
832 /* Clear the revalidation flags. */
833 tag_set_init(&ofproto->revalidate_set);
834 ofproto->need_revalidate = false;
836 HMAP_FOR_EACH_SAFE (facet, next, hmap_node, &ofproto->facets) {
838 || tag_set_intersects(&revalidate_set, facet->tags)) {
839 facet_revalidate(facet);
844 /* Check the consistency of a random facet, to aid debugging. */
845 if (!hmap_is_empty(&ofproto->facets) && !ofproto->need_revalidate) {
848 facet = CONTAINER_OF(hmap_random_node(&ofproto->facets),
849 struct facet, hmap_node);
850 if (!tag_set_intersects(&ofproto->revalidate_set, facet->tags)) {
851 if (!facet_check_consistency(facet)) {
852 ofproto->need_revalidate = true;
861 wait(struct ofproto *ofproto_)
863 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
864 struct ofport_dpif *ofport;
865 struct ofbundle *bundle;
867 if (!clogged && !list_is_empty(&ofproto->completions)) {
868 poll_immediate_wake();
871 dpif_wait(ofproto->dpif);
872 dpif_recv_wait(ofproto->dpif);
873 if (ofproto->sflow) {
874 dpif_sflow_wait(ofproto->sflow);
876 if (!tag_set_is_empty(&ofproto->revalidate_set)) {
877 poll_immediate_wake();
879 HMAP_FOR_EACH (ofport, up.hmap_node, &ofproto->up.ports) {
882 HMAP_FOR_EACH (bundle, hmap_node, &ofproto->bundles) {
885 if (ofproto->netflow) {
886 netflow_wait(ofproto->netflow);
888 mac_learning_wait(ofproto->ml);
890 if (ofproto->need_revalidate) {
891 /* Shouldn't happen, but if it does just go around again. */
892 VLOG_DBG_RL(&rl, "need revalidate in ofproto_wait_cb()");
893 poll_immediate_wake();
895 timer_wait(&ofproto->next_expiration);
900 flush(struct ofproto *ofproto_)
902 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
903 struct facet *facet, *next_facet;
905 HMAP_FOR_EACH_SAFE (facet, next_facet, hmap_node, &ofproto->facets) {
906 /* Mark the facet as not installed so that facet_remove() doesn't
907 * bother trying to uninstall it. There is no point in uninstalling it
908 * individually since we are about to blow away all the facets with
909 * dpif_flow_flush(). */
910 struct subfacet *subfacet;
912 LIST_FOR_EACH (subfacet, list_node, &facet->subfacets) {
913 subfacet->installed = false;
914 subfacet->dp_packet_count = 0;
915 subfacet->dp_byte_count = 0;
919 dpif_flow_flush(ofproto->dpif);
923 get_features(struct ofproto *ofproto_ OVS_UNUSED,
924 bool *arp_match_ip, enum ofputil_action_bitmap *actions)
926 *arp_match_ip = true;
927 *actions = (OFPUTIL_A_OUTPUT |
928 OFPUTIL_A_SET_VLAN_VID |
929 OFPUTIL_A_SET_VLAN_PCP |
930 OFPUTIL_A_STRIP_VLAN |
931 OFPUTIL_A_SET_DL_SRC |
932 OFPUTIL_A_SET_DL_DST |
933 OFPUTIL_A_SET_NW_SRC |
934 OFPUTIL_A_SET_NW_DST |
935 OFPUTIL_A_SET_NW_TOS |
936 OFPUTIL_A_SET_TP_SRC |
937 OFPUTIL_A_SET_TP_DST |
942 get_tables(struct ofproto *ofproto_, struct ofp_table_stats *ots)
944 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
945 struct dpif_dp_stats s;
947 strcpy(ots->name, "classifier");
949 dpif_get_dp_stats(ofproto->dpif, &s);
950 put_32aligned_be64(&ots->lookup_count, htonll(s.n_hit + s.n_missed));
951 put_32aligned_be64(&ots->matched_count,
952 htonll(s.n_hit + ofproto->n_matches));
955 static struct ofport *
958 struct ofport_dpif *port = xmalloc(sizeof *port);
963 port_dealloc(struct ofport *port_)
965 struct ofport_dpif *port = ofport_dpif_cast(port_);
970 port_construct(struct ofport *port_)
972 struct ofport_dpif *port = ofport_dpif_cast(port_);
973 struct ofproto_dpif *ofproto = ofproto_dpif_cast(port->up.ofproto);
975 ofproto->need_revalidate = true;
976 port->odp_port = ofp_port_to_odp_port(port->up.ofp_port);
979 port->tag = tag_create_random();
980 port->may_enable = true;
981 port->stp_port = NULL;
982 port->stp_state = STP_DISABLED;
983 hmap_init(&port->priorities);
984 port->realdev_ofp_port = 0;
985 port->vlandev_vid = 0;
986 port->carrier_seq = netdev_get_carrier_resets(port->up.netdev);
988 if (ofproto->sflow) {
989 dpif_sflow_add_port(ofproto->sflow, port_);
996 port_destruct(struct ofport *port_)
998 struct ofport_dpif *port = ofport_dpif_cast(port_);
999 struct ofproto_dpif *ofproto = ofproto_dpif_cast(port->up.ofproto);
1001 ofproto->need_revalidate = true;
1002 bundle_remove(port_);
1003 set_cfm(port_, NULL);
1004 if (ofproto->sflow) {
1005 dpif_sflow_del_port(ofproto->sflow, port->odp_port);
1008 ofport_clear_priorities(port);
1009 hmap_destroy(&port->priorities);
1013 port_modified(struct ofport *port_)
1015 struct ofport_dpif *port = ofport_dpif_cast(port_);
1017 if (port->bundle && port->bundle->bond) {
1018 bond_slave_set_netdev(port->bundle->bond, port, port->up.netdev);
1023 port_reconfigured(struct ofport *port_, enum ofputil_port_config old_config)
1025 struct ofport_dpif *port = ofport_dpif_cast(port_);
1026 struct ofproto_dpif *ofproto = ofproto_dpif_cast(port->up.ofproto);
1027 enum ofputil_port_config changed = old_config ^ port->up.pp.config;
1029 if (changed & (OFPUTIL_PC_NO_RECV | OFPUTIL_PC_NO_RECV_STP |
1030 OFPUTIL_PC_NO_FWD | OFPUTIL_PC_NO_FLOOD)) {
1031 ofproto->need_revalidate = true;
1033 if (changed & OFPUTIL_PC_NO_FLOOD && port->bundle) {
1034 bundle_update(port->bundle);
1040 set_sflow(struct ofproto *ofproto_,
1041 const struct ofproto_sflow_options *sflow_options)
1043 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1044 struct dpif_sflow *ds = ofproto->sflow;
1046 if (sflow_options) {
1048 struct ofport_dpif *ofport;
1050 ds = ofproto->sflow = dpif_sflow_create(ofproto->dpif);
1051 HMAP_FOR_EACH (ofport, up.hmap_node, &ofproto->up.ports) {
1052 dpif_sflow_add_port(ds, &ofport->up);
1054 ofproto->need_revalidate = true;
1056 dpif_sflow_set_options(ds, sflow_options);
1059 dpif_sflow_destroy(ds);
1060 ofproto->need_revalidate = true;
1061 ofproto->sflow = NULL;
1068 set_cfm(struct ofport *ofport_, const struct cfm_settings *s)
1070 struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
1077 struct ofproto_dpif *ofproto;
1079 ofproto = ofproto_dpif_cast(ofport->up.ofproto);
1080 ofproto->need_revalidate = true;
1081 ofport->cfm = cfm_create(netdev_get_name(ofport->up.netdev));
1084 if (cfm_configure(ofport->cfm, s)) {
1090 cfm_destroy(ofport->cfm);
1096 get_cfm_fault(const struct ofport *ofport_)
1098 struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
1100 return ofport->cfm ? cfm_get_fault(ofport->cfm) : -1;
1104 get_cfm_remote_mpids(const struct ofport *ofport_, const uint64_t **rmps,
1107 struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
1110 cfm_get_remote_mpids(ofport->cfm, rmps, n_rmps);
1117 /* Spanning Tree. */
1120 send_bpdu_cb(struct ofpbuf *pkt, int port_num, void *ofproto_)
1122 struct ofproto_dpif *ofproto = ofproto_;
1123 struct stp_port *sp = stp_get_port(ofproto->stp, port_num);
1124 struct ofport_dpif *ofport;
1126 ofport = stp_port_get_aux(sp);
1128 VLOG_WARN_RL(&rl, "%s: cannot send BPDU on unknown port %d",
1129 ofproto->up.name, port_num);
1131 struct eth_header *eth = pkt->l2;
1133 netdev_get_etheraddr(ofport->up.netdev, eth->eth_src);
1134 if (eth_addr_is_zero(eth->eth_src)) {
1135 VLOG_WARN_RL(&rl, "%s: cannot send BPDU on port %d "
1136 "with unknown MAC", ofproto->up.name, port_num);
1138 send_packet(ofport, pkt);
1144 /* Configures STP on 'ofproto_' using the settings defined in 's'. */
1146 set_stp(struct ofproto *ofproto_, const struct ofproto_stp_settings *s)
1148 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1150 /* Only revalidate flows if the configuration changed. */
1151 if (!s != !ofproto->stp) {
1152 ofproto->need_revalidate = true;
1156 if (!ofproto->stp) {
1157 ofproto->stp = stp_create(ofproto_->name, s->system_id,
1158 send_bpdu_cb, ofproto);
1159 ofproto->stp_last_tick = time_msec();
1162 stp_set_bridge_id(ofproto->stp, s->system_id);
1163 stp_set_bridge_priority(ofproto->stp, s->priority);
1164 stp_set_hello_time(ofproto->stp, s->hello_time);
1165 stp_set_max_age(ofproto->stp, s->max_age);
1166 stp_set_forward_delay(ofproto->stp, s->fwd_delay);
1168 struct ofport *ofport;
1170 HMAP_FOR_EACH (ofport, hmap_node, &ofproto->up.ports) {
1171 set_stp_port(ofport, NULL);
1174 stp_destroy(ofproto->stp);
1175 ofproto->stp = NULL;
1182 get_stp_status(struct ofproto *ofproto_, struct ofproto_stp_status *s)
1184 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1188 s->bridge_id = stp_get_bridge_id(ofproto->stp);
1189 s->designated_root = stp_get_designated_root(ofproto->stp);
1190 s->root_path_cost = stp_get_root_path_cost(ofproto->stp);
1199 update_stp_port_state(struct ofport_dpif *ofport)
1201 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport->up.ofproto);
1202 enum stp_state state;
1204 /* Figure out new state. */
1205 state = ofport->stp_port ? stp_port_get_state(ofport->stp_port)
1209 if (ofport->stp_state != state) {
1210 enum ofputil_port_state of_state;
1213 VLOG_DBG_RL(&rl, "port %s: STP state changed from %s to %s",
1214 netdev_get_name(ofport->up.netdev),
1215 stp_state_name(ofport->stp_state),
1216 stp_state_name(state));
1217 if (stp_learn_in_state(ofport->stp_state)
1218 != stp_learn_in_state(state)) {
1219 /* xxx Learning action flows should also be flushed. */
1220 mac_learning_flush(ofproto->ml, &ofproto->revalidate_set);
1222 fwd_change = stp_forward_in_state(ofport->stp_state)
1223 != stp_forward_in_state(state);
1225 ofproto->need_revalidate = true;
1226 ofport->stp_state = state;
1227 ofport->stp_state_entered = time_msec();
1229 if (fwd_change && ofport->bundle) {
1230 bundle_update(ofport->bundle);
1233 /* Update the STP state bits in the OpenFlow port description. */
1234 of_state = ofport->up.pp.state & ~OFPUTIL_PS_STP_MASK;
1235 of_state |= (state == STP_LISTENING ? OFPUTIL_PS_STP_LISTEN
1236 : state == STP_LEARNING ? OFPUTIL_PS_STP_LEARN
1237 : state == STP_FORWARDING ? OFPUTIL_PS_STP_FORWARD
1238 : state == STP_BLOCKING ? OFPUTIL_PS_STP_BLOCK
1240 ofproto_port_set_state(&ofport->up, of_state);
1244 /* Configures STP on 'ofport_' using the settings defined in 's'. The
1245 * caller is responsible for assigning STP port numbers and ensuring
1246 * there are no duplicates. */
1248 set_stp_port(struct ofport *ofport_,
1249 const struct ofproto_port_stp_settings *s)
1251 struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
1252 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport->up.ofproto);
1253 struct stp_port *sp = ofport->stp_port;
1255 if (!s || !s->enable) {
1257 ofport->stp_port = NULL;
1258 stp_port_disable(sp);
1259 update_stp_port_state(ofport);
1262 } else if (sp && stp_port_no(sp) != s->port_num
1263 && ofport == stp_port_get_aux(sp)) {
1264 /* The port-id changed, so disable the old one if it's not
1265 * already in use by another port. */
1266 stp_port_disable(sp);
1269 sp = ofport->stp_port = stp_get_port(ofproto->stp, s->port_num);
1270 stp_port_enable(sp);
1272 stp_port_set_aux(sp, ofport);
1273 stp_port_set_priority(sp, s->priority);
1274 stp_port_set_path_cost(sp, s->path_cost);
1276 update_stp_port_state(ofport);
1282 get_stp_port_status(struct ofport *ofport_,
1283 struct ofproto_port_stp_status *s)
1285 struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
1286 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport->up.ofproto);
1287 struct stp_port *sp = ofport->stp_port;
1289 if (!ofproto->stp || !sp) {
1295 s->port_id = stp_port_get_id(sp);
1296 s->state = stp_port_get_state(sp);
1297 s->sec_in_state = (time_msec() - ofport->stp_state_entered) / 1000;
1298 s->role = stp_port_get_role(sp);
1299 stp_port_get_counts(sp, &s->tx_count, &s->rx_count, &s->error_count);
1305 stp_run(struct ofproto_dpif *ofproto)
1308 long long int now = time_msec();
1309 long long int elapsed = now - ofproto->stp_last_tick;
1310 struct stp_port *sp;
1313 stp_tick(ofproto->stp, MIN(INT_MAX, elapsed));
1314 ofproto->stp_last_tick = now;
1316 while (stp_get_changed_port(ofproto->stp, &sp)) {
1317 struct ofport_dpif *ofport = stp_port_get_aux(sp);
1320 update_stp_port_state(ofport);
1324 if (stp_check_and_reset_fdb_flush(ofproto->stp)) {
1325 mac_learning_flush(ofproto->ml, &ofproto->revalidate_set);
1331 stp_wait(struct ofproto_dpif *ofproto)
1334 poll_timer_wait(1000);
1338 /* Returns true if STP should process 'flow'. */
1340 stp_should_process_flow(const struct flow *flow)
1342 return eth_addr_equals(flow->dl_dst, eth_addr_stp);
1346 stp_process_packet(const struct ofport_dpif *ofport,
1347 const struct ofpbuf *packet)
1349 struct ofpbuf payload = *packet;
1350 struct eth_header *eth = payload.data;
1351 struct stp_port *sp = ofport->stp_port;
1353 /* Sink packets on ports that have STP disabled when the bridge has
1355 if (!sp || stp_port_get_state(sp) == STP_DISABLED) {
1359 /* Trim off padding on payload. */
1360 if (payload.size > ntohs(eth->eth_type) + ETH_HEADER_LEN) {
1361 payload.size = ntohs(eth->eth_type) + ETH_HEADER_LEN;
1364 if (ofpbuf_try_pull(&payload, ETH_HEADER_LEN + LLC_HEADER_LEN)) {
1365 stp_received_bpdu(sp, payload.data, payload.size);
1369 static struct priority_to_dscp *
1370 get_priority(const struct ofport_dpif *ofport, uint32_t priority)
1372 struct priority_to_dscp *pdscp;
1375 hash = hash_int(priority, 0);
1376 HMAP_FOR_EACH_IN_BUCKET (pdscp, hmap_node, hash, &ofport->priorities) {
1377 if (pdscp->priority == priority) {
1385 ofport_clear_priorities(struct ofport_dpif *ofport)
1387 struct priority_to_dscp *pdscp, *next;
1389 HMAP_FOR_EACH_SAFE (pdscp, next, hmap_node, &ofport->priorities) {
1390 hmap_remove(&ofport->priorities, &pdscp->hmap_node);
1396 set_queues(struct ofport *ofport_,
1397 const struct ofproto_port_queue *qdscp_list,
1400 struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
1401 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport->up.ofproto);
1402 struct hmap new = HMAP_INITIALIZER(&new);
1405 for (i = 0; i < n_qdscp; i++) {
1406 struct priority_to_dscp *pdscp;
1410 dscp = (qdscp_list[i].dscp << 2) & IP_DSCP_MASK;
1411 if (dpif_queue_to_priority(ofproto->dpif, qdscp_list[i].queue,
1416 pdscp = get_priority(ofport, priority);
1418 hmap_remove(&ofport->priorities, &pdscp->hmap_node);
1420 pdscp = xmalloc(sizeof *pdscp);
1421 pdscp->priority = priority;
1423 ofproto->need_revalidate = true;
1426 if (pdscp->dscp != dscp) {
1428 ofproto->need_revalidate = true;
1431 hmap_insert(&new, &pdscp->hmap_node, hash_int(pdscp->priority, 0));
1434 if (!hmap_is_empty(&ofport->priorities)) {
1435 ofport_clear_priorities(ofport);
1436 ofproto->need_revalidate = true;
1439 hmap_swap(&new, &ofport->priorities);
1447 /* Expires all MAC learning entries associated with 'bundle' and forces its
1448 * ofproto to revalidate every flow.
1450 * Normally MAC learning entries are removed only from the ofproto associated
1451 * with 'bundle', but if 'all_ofprotos' is true, then the MAC learning entries
1452 * are removed from every ofproto. When patch ports and SLB bonds are in use
1453 * and a VM migration happens and the gratuitous ARPs are somehow lost, this
1454 * avoids a MAC_ENTRY_IDLE_TIME delay before the migrated VM can communicate
1455 * with the host from which it migrated. */
1457 bundle_flush_macs(struct ofbundle *bundle, bool all_ofprotos)
1459 struct ofproto_dpif *ofproto = bundle->ofproto;
1460 struct mac_learning *ml = ofproto->ml;
1461 struct mac_entry *mac, *next_mac;
1463 ofproto->need_revalidate = true;
1464 LIST_FOR_EACH_SAFE (mac, next_mac, lru_node, &ml->lrus) {
1465 if (mac->port.p == bundle) {
1467 struct ofproto_dpif *o;
1469 HMAP_FOR_EACH (o, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
1471 struct mac_entry *e;
1473 e = mac_learning_lookup(o->ml, mac->mac, mac->vlan,
1476 tag_set_add(&o->revalidate_set, e->tag);
1477 mac_learning_expire(o->ml, e);
1483 mac_learning_expire(ml, mac);
1488 static struct ofbundle *
1489 bundle_lookup(const struct ofproto_dpif *ofproto, void *aux)
1491 struct ofbundle *bundle;
1493 HMAP_FOR_EACH_IN_BUCKET (bundle, hmap_node, hash_pointer(aux, 0),
1494 &ofproto->bundles) {
1495 if (bundle->aux == aux) {
1502 /* Looks up each of the 'n_auxes' pointers in 'auxes' as bundles and adds the
1503 * ones that are found to 'bundles'. */
1505 bundle_lookup_multiple(struct ofproto_dpif *ofproto,
1506 void **auxes, size_t n_auxes,
1507 struct hmapx *bundles)
1511 hmapx_init(bundles);
1512 for (i = 0; i < n_auxes; i++) {
1513 struct ofbundle *bundle = bundle_lookup(ofproto, auxes[i]);
1515 hmapx_add(bundles, bundle);
1521 bundle_update(struct ofbundle *bundle)
1523 struct ofport_dpif *port;
1525 bundle->floodable = true;
1526 LIST_FOR_EACH (port, bundle_node, &bundle->ports) {
1527 if (port->up.pp.config & OFPUTIL_PC_NO_FLOOD
1528 || !stp_forward_in_state(port->stp_state)) {
1529 bundle->floodable = false;
1536 bundle_del_port(struct ofport_dpif *port)
1538 struct ofbundle *bundle = port->bundle;
1540 bundle->ofproto->need_revalidate = true;
1542 list_remove(&port->bundle_node);
1543 port->bundle = NULL;
1546 lacp_slave_unregister(bundle->lacp, port);
1549 bond_slave_unregister(bundle->bond, port);
1552 bundle_update(bundle);
1556 bundle_add_port(struct ofbundle *bundle, uint32_t ofp_port,
1557 struct lacp_slave_settings *lacp,
1558 uint32_t bond_stable_id)
1560 struct ofport_dpif *port;
1562 port = get_ofp_port(bundle->ofproto, ofp_port);
1567 if (port->bundle != bundle) {
1568 bundle->ofproto->need_revalidate = true;
1570 bundle_del_port(port);
1573 port->bundle = bundle;
1574 list_push_back(&bundle->ports, &port->bundle_node);
1575 if (port->up.pp.config & OFPUTIL_PC_NO_FLOOD
1576 || !stp_forward_in_state(port->stp_state)) {
1577 bundle->floodable = false;
1581 port->bundle->ofproto->need_revalidate = true;
1582 lacp_slave_register(bundle->lacp, port, lacp);
1585 port->bond_stable_id = bond_stable_id;
1591 bundle_destroy(struct ofbundle *bundle)
1593 struct ofproto_dpif *ofproto;
1594 struct ofport_dpif *port, *next_port;
1601 ofproto = bundle->ofproto;
1602 for (i = 0; i < MAX_MIRRORS; i++) {
1603 struct ofmirror *m = ofproto->mirrors[i];
1605 if (m->out == bundle) {
1607 } else if (hmapx_find_and_delete(&m->srcs, bundle)
1608 || hmapx_find_and_delete(&m->dsts, bundle)) {
1609 ofproto->need_revalidate = true;
1614 LIST_FOR_EACH_SAFE (port, next_port, bundle_node, &bundle->ports) {
1615 bundle_del_port(port);
1618 bundle_flush_macs(bundle, true);
1619 hmap_remove(&ofproto->bundles, &bundle->hmap_node);
1621 free(bundle->trunks);
1622 lacp_destroy(bundle->lacp);
1623 bond_destroy(bundle->bond);
1628 bundle_set(struct ofproto *ofproto_, void *aux,
1629 const struct ofproto_bundle_settings *s)
1631 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1632 bool need_flush = false;
1633 struct ofport_dpif *port;
1634 struct ofbundle *bundle;
1635 unsigned long *trunks;
1641 bundle_destroy(bundle_lookup(ofproto, aux));
1645 assert(s->n_slaves == 1 || s->bond != NULL);
1646 assert((s->lacp != NULL) == (s->lacp_slaves != NULL));
1648 bundle = bundle_lookup(ofproto, aux);
1650 bundle = xmalloc(sizeof *bundle);
1652 bundle->ofproto = ofproto;
1653 hmap_insert(&ofproto->bundles, &bundle->hmap_node,
1654 hash_pointer(aux, 0));
1656 bundle->name = NULL;
1658 list_init(&bundle->ports);
1659 bundle->vlan_mode = PORT_VLAN_TRUNK;
1661 bundle->trunks = NULL;
1662 bundle->use_priority_tags = s->use_priority_tags;
1663 bundle->lacp = NULL;
1664 bundle->bond = NULL;
1666 bundle->floodable = true;
1668 bundle->src_mirrors = 0;
1669 bundle->dst_mirrors = 0;
1670 bundle->mirror_out = 0;
1673 if (!bundle->name || strcmp(s->name, bundle->name)) {
1675 bundle->name = xstrdup(s->name);
1680 if (!bundle->lacp) {
1681 ofproto->need_revalidate = true;
1682 bundle->lacp = lacp_create();
1684 lacp_configure(bundle->lacp, s->lacp);
1686 lacp_destroy(bundle->lacp);
1687 bundle->lacp = NULL;
1690 /* Update set of ports. */
1692 for (i = 0; i < s->n_slaves; i++) {
1693 if (!bundle_add_port(bundle, s->slaves[i],
1694 s->lacp ? &s->lacp_slaves[i] : NULL,
1695 s->bond_stable_ids ? s->bond_stable_ids[i] : 0)) {
1699 if (!ok || list_size(&bundle->ports) != s->n_slaves) {
1700 struct ofport_dpif *next_port;
1702 LIST_FOR_EACH_SAFE (port, next_port, bundle_node, &bundle->ports) {
1703 for (i = 0; i < s->n_slaves; i++) {
1704 if (s->slaves[i] == port->up.ofp_port) {
1709 bundle_del_port(port);
1713 assert(list_size(&bundle->ports) <= s->n_slaves);
1715 if (list_is_empty(&bundle->ports)) {
1716 bundle_destroy(bundle);
1720 /* Set VLAN tagging mode */
1721 if (s->vlan_mode != bundle->vlan_mode
1722 || s->use_priority_tags != bundle->use_priority_tags) {
1723 bundle->vlan_mode = s->vlan_mode;
1724 bundle->use_priority_tags = s->use_priority_tags;
1729 vlan = (s->vlan_mode == PORT_VLAN_TRUNK ? -1
1730 : s->vlan >= 0 && s->vlan <= 4095 ? s->vlan
1732 if (vlan != bundle->vlan) {
1733 bundle->vlan = vlan;
1737 /* Get trunked VLANs. */
1738 switch (s->vlan_mode) {
1739 case PORT_VLAN_ACCESS:
1743 case PORT_VLAN_TRUNK:
1744 trunks = (unsigned long *) s->trunks;
1747 case PORT_VLAN_NATIVE_UNTAGGED:
1748 case PORT_VLAN_NATIVE_TAGGED:
1749 if (vlan != 0 && (!s->trunks
1750 || !bitmap_is_set(s->trunks, vlan)
1751 || bitmap_is_set(s->trunks, 0))) {
1752 /* Force trunking the native VLAN and prohibit trunking VLAN 0. */
1754 trunks = bitmap_clone(s->trunks, 4096);
1756 trunks = bitmap_allocate1(4096);
1758 bitmap_set1(trunks, vlan);
1759 bitmap_set0(trunks, 0);
1761 trunks = (unsigned long *) s->trunks;
1768 if (!vlan_bitmap_equal(trunks, bundle->trunks)) {
1769 free(bundle->trunks);
1770 if (trunks == s->trunks) {
1771 bundle->trunks = vlan_bitmap_clone(trunks);
1773 bundle->trunks = trunks;
1778 if (trunks != s->trunks) {
1783 if (!list_is_short(&bundle->ports)) {
1784 bundle->ofproto->has_bonded_bundles = true;
1786 if (bond_reconfigure(bundle->bond, s->bond)) {
1787 ofproto->need_revalidate = true;
1790 bundle->bond = bond_create(s->bond);
1791 ofproto->need_revalidate = true;
1794 LIST_FOR_EACH (port, bundle_node, &bundle->ports) {
1795 bond_slave_register(bundle->bond, port, port->bond_stable_id,
1799 bond_destroy(bundle->bond);
1800 bundle->bond = NULL;
1803 /* If we changed something that would affect MAC learning, un-learn
1804 * everything on this port and force flow revalidation. */
1806 bundle_flush_macs(bundle, false);
1813 bundle_remove(struct ofport *port_)
1815 struct ofport_dpif *port = ofport_dpif_cast(port_);
1816 struct ofbundle *bundle = port->bundle;
1819 bundle_del_port(port);
1820 if (list_is_empty(&bundle->ports)) {
1821 bundle_destroy(bundle);
1822 } else if (list_is_short(&bundle->ports)) {
1823 bond_destroy(bundle->bond);
1824 bundle->bond = NULL;
1830 send_pdu_cb(void *port_, const void *pdu, size_t pdu_size)
1832 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 10);
1833 struct ofport_dpif *port = port_;
1834 uint8_t ea[ETH_ADDR_LEN];
1837 error = netdev_get_etheraddr(port->up.netdev, ea);
1839 struct ofpbuf packet;
1842 ofpbuf_init(&packet, 0);
1843 packet_pdu = eth_compose(&packet, eth_addr_lacp, ea, ETH_TYPE_LACP,
1845 memcpy(packet_pdu, pdu, pdu_size);
1847 send_packet(port, &packet);
1848 ofpbuf_uninit(&packet);
1850 VLOG_ERR_RL(&rl, "port %s: cannot obtain Ethernet address of iface "
1851 "%s (%s)", port->bundle->name,
1852 netdev_get_name(port->up.netdev), strerror(error));
1857 bundle_send_learning_packets(struct ofbundle *bundle)
1859 struct ofproto_dpif *ofproto = bundle->ofproto;
1860 int error, n_packets, n_errors;
1861 struct mac_entry *e;
1863 error = n_packets = n_errors = 0;
1864 LIST_FOR_EACH (e, lru_node, &ofproto->ml->lrus) {
1865 if (e->port.p != bundle) {
1866 struct ofpbuf *learning_packet;
1867 struct ofport_dpif *port;
1871 /* The assignment to "port" is unnecessary but makes "grep"ing for
1872 * struct ofport_dpif more effective. */
1873 learning_packet = bond_compose_learning_packet(bundle->bond,
1877 ret = send_packet(port, learning_packet);
1878 ofpbuf_delete(learning_packet);
1888 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
1889 VLOG_WARN_RL(&rl, "bond %s: %d errors sending %d gratuitous learning "
1890 "packets, last error was: %s",
1891 bundle->name, n_errors, n_packets, strerror(error));
1893 VLOG_DBG("bond %s: sent %d gratuitous learning packets",
1894 bundle->name, n_packets);
1899 bundle_run(struct ofbundle *bundle)
1902 lacp_run(bundle->lacp, send_pdu_cb);
1905 struct ofport_dpif *port;
1907 LIST_FOR_EACH (port, bundle_node, &bundle->ports) {
1908 bond_slave_set_may_enable(bundle->bond, port, port->may_enable);
1911 bond_run(bundle->bond, &bundle->ofproto->revalidate_set,
1912 lacp_status(bundle->lacp));
1913 if (bond_should_send_learning_packets(bundle->bond)) {
1914 bundle_send_learning_packets(bundle);
1920 bundle_wait(struct ofbundle *bundle)
1923 lacp_wait(bundle->lacp);
1926 bond_wait(bundle->bond);
1933 mirror_scan(struct ofproto_dpif *ofproto)
1937 for (idx = 0; idx < MAX_MIRRORS; idx++) {
1938 if (!ofproto->mirrors[idx]) {
1945 static struct ofmirror *
1946 mirror_lookup(struct ofproto_dpif *ofproto, void *aux)
1950 for (i = 0; i < MAX_MIRRORS; i++) {
1951 struct ofmirror *mirror = ofproto->mirrors[i];
1952 if (mirror && mirror->aux == aux) {
1960 /* Update the 'dup_mirrors' member of each of the ofmirrors in 'ofproto'. */
1962 mirror_update_dups(struct ofproto_dpif *ofproto)
1966 for (i = 0; i < MAX_MIRRORS; i++) {
1967 struct ofmirror *m = ofproto->mirrors[i];
1970 m->dup_mirrors = MIRROR_MASK_C(1) << i;
1974 for (i = 0; i < MAX_MIRRORS; i++) {
1975 struct ofmirror *m1 = ofproto->mirrors[i];
1982 for (j = i + 1; j < MAX_MIRRORS; j++) {
1983 struct ofmirror *m2 = ofproto->mirrors[j];
1985 if (m2 && m1->out == m2->out && m1->out_vlan == m2->out_vlan) {
1986 m1->dup_mirrors |= MIRROR_MASK_C(1) << j;
1987 m2->dup_mirrors |= m1->dup_mirrors;
1994 mirror_set(struct ofproto *ofproto_, void *aux,
1995 const struct ofproto_mirror_settings *s)
1997 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1998 mirror_mask_t mirror_bit;
1999 struct ofbundle *bundle;
2000 struct ofmirror *mirror;
2001 struct ofbundle *out;
2002 struct hmapx srcs; /* Contains "struct ofbundle *"s. */
2003 struct hmapx dsts; /* Contains "struct ofbundle *"s. */
2006 mirror = mirror_lookup(ofproto, aux);
2008 mirror_destroy(mirror);
2014 idx = mirror_scan(ofproto);
2016 VLOG_WARN("bridge %s: maximum of %d port mirrors reached, "
2018 ofproto->up.name, MAX_MIRRORS, s->name);
2022 mirror = ofproto->mirrors[idx] = xzalloc(sizeof *mirror);
2023 mirror->ofproto = ofproto;
2026 mirror->out_vlan = -1;
2027 mirror->name = NULL;
2030 if (!mirror->name || strcmp(s->name, mirror->name)) {
2032 mirror->name = xstrdup(s->name);
2035 /* Get the new configuration. */
2036 if (s->out_bundle) {
2037 out = bundle_lookup(ofproto, s->out_bundle);
2039 mirror_destroy(mirror);
2045 out_vlan = s->out_vlan;
2047 bundle_lookup_multiple(ofproto, s->srcs, s->n_srcs, &srcs);
2048 bundle_lookup_multiple(ofproto, s->dsts, s->n_dsts, &dsts);
2050 /* If the configuration has not changed, do nothing. */
2051 if (hmapx_equals(&srcs, &mirror->srcs)
2052 && hmapx_equals(&dsts, &mirror->dsts)
2053 && vlan_bitmap_equal(mirror->vlans, s->src_vlans)
2054 && mirror->out == out
2055 && mirror->out_vlan == out_vlan)
2057 hmapx_destroy(&srcs);
2058 hmapx_destroy(&dsts);
2062 hmapx_swap(&srcs, &mirror->srcs);
2063 hmapx_destroy(&srcs);
2065 hmapx_swap(&dsts, &mirror->dsts);
2066 hmapx_destroy(&dsts);
2068 free(mirror->vlans);
2069 mirror->vlans = vlan_bitmap_clone(s->src_vlans);
2072 mirror->out_vlan = out_vlan;
2074 /* Update bundles. */
2075 mirror_bit = MIRROR_MASK_C(1) << mirror->idx;
2076 HMAP_FOR_EACH (bundle, hmap_node, &mirror->ofproto->bundles) {
2077 if (hmapx_contains(&mirror->srcs, bundle)) {
2078 bundle->src_mirrors |= mirror_bit;
2080 bundle->src_mirrors &= ~mirror_bit;
2083 if (hmapx_contains(&mirror->dsts, bundle)) {
2084 bundle->dst_mirrors |= mirror_bit;
2086 bundle->dst_mirrors &= ~mirror_bit;
2089 if (mirror->out == bundle) {
2090 bundle->mirror_out |= mirror_bit;
2092 bundle->mirror_out &= ~mirror_bit;
2096 ofproto->need_revalidate = true;
2097 mac_learning_flush(ofproto->ml, &ofproto->revalidate_set);
2098 mirror_update_dups(ofproto);
2104 mirror_destroy(struct ofmirror *mirror)
2106 struct ofproto_dpif *ofproto;
2107 mirror_mask_t mirror_bit;
2108 struct ofbundle *bundle;
2114 ofproto = mirror->ofproto;
2115 ofproto->need_revalidate = true;
2116 mac_learning_flush(ofproto->ml, &ofproto->revalidate_set);
2118 mirror_bit = MIRROR_MASK_C(1) << mirror->idx;
2119 HMAP_FOR_EACH (bundle, hmap_node, &ofproto->bundles) {
2120 bundle->src_mirrors &= ~mirror_bit;
2121 bundle->dst_mirrors &= ~mirror_bit;
2122 bundle->mirror_out &= ~mirror_bit;
2125 hmapx_destroy(&mirror->srcs);
2126 hmapx_destroy(&mirror->dsts);
2127 free(mirror->vlans);
2129 ofproto->mirrors[mirror->idx] = NULL;
2133 mirror_update_dups(ofproto);
2137 mirror_get_stats(struct ofproto *ofproto_, void *aux,
2138 uint64_t *packets, uint64_t *bytes)
2140 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
2141 struct ofmirror *mirror = mirror_lookup(ofproto, aux);
2144 *packets = *bytes = UINT64_MAX;
2148 *packets = mirror->packet_count;
2149 *bytes = mirror->byte_count;
2155 set_flood_vlans(struct ofproto *ofproto_, unsigned long *flood_vlans)
2157 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
2158 if (mac_learning_set_flood_vlans(ofproto->ml, flood_vlans)) {
2159 mac_learning_flush(ofproto->ml, &ofproto->revalidate_set);
2165 is_mirror_output_bundle(const struct ofproto *ofproto_, void *aux)
2167 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
2168 struct ofbundle *bundle = bundle_lookup(ofproto, aux);
2169 return bundle && bundle->mirror_out != 0;
2173 forward_bpdu_changed(struct ofproto *ofproto_)
2175 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
2176 /* Revalidate cached flows whenever forward_bpdu option changes. */
2177 ofproto->need_revalidate = true;
2181 set_mac_idle_time(struct ofproto *ofproto_, unsigned int idle_time)
2183 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
2184 mac_learning_set_idle_time(ofproto->ml, idle_time);
2189 static struct ofport_dpif *
2190 get_ofp_port(struct ofproto_dpif *ofproto, uint16_t ofp_port)
2192 struct ofport *ofport = ofproto_get_port(&ofproto->up, ofp_port);
2193 return ofport ? ofport_dpif_cast(ofport) : NULL;
2196 static struct ofport_dpif *
2197 get_odp_port(struct ofproto_dpif *ofproto, uint32_t odp_port)
2199 return get_ofp_port(ofproto, odp_port_to_ofp_port(odp_port));
2203 ofproto_port_from_dpif_port(struct ofproto_port *ofproto_port,
2204 struct dpif_port *dpif_port)
2206 ofproto_port->name = dpif_port->name;
2207 ofproto_port->type = dpif_port->type;
2208 ofproto_port->ofp_port = odp_port_to_ofp_port(dpif_port->port_no);
2212 port_run(struct ofport_dpif *ofport)
2214 long long int carrier_seq = netdev_get_carrier_resets(ofport->up.netdev);
2215 bool carrier_changed = carrier_seq != ofport->carrier_seq;
2216 bool enable = netdev_get_carrier(ofport->up.netdev);
2218 ofport->carrier_seq = carrier_seq;
2221 cfm_run(ofport->cfm);
2223 if (cfm_should_send_ccm(ofport->cfm)) {
2224 struct ofpbuf packet;
2226 ofpbuf_init(&packet, 0);
2227 cfm_compose_ccm(ofport->cfm, &packet, ofport->up.pp.hw_addr);
2228 send_packet(ofport, &packet);
2229 ofpbuf_uninit(&packet);
2232 enable = enable && !cfm_get_fault(ofport->cfm)
2233 && cfm_get_opup(ofport->cfm);
2236 if (ofport->bundle) {
2237 enable = enable && lacp_slave_may_enable(ofport->bundle->lacp, ofport);
2238 if (carrier_changed) {
2239 lacp_slave_carrier_changed(ofport->bundle->lacp, ofport);
2243 if (ofport->may_enable != enable) {
2244 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport->up.ofproto);
2246 if (ofproto->has_bundle_action) {
2247 ofproto->need_revalidate = true;
2251 ofport->may_enable = enable;
2255 port_wait(struct ofport_dpif *ofport)
2258 cfm_wait(ofport->cfm);
2263 port_query_by_name(const struct ofproto *ofproto_, const char *devname,
2264 struct ofproto_port *ofproto_port)
2266 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
2267 struct dpif_port dpif_port;
2270 error = dpif_port_query_by_name(ofproto->dpif, devname, &dpif_port);
2272 ofproto_port_from_dpif_port(ofproto_port, &dpif_port);
2278 port_add(struct ofproto *ofproto_, struct netdev *netdev, uint16_t *ofp_portp)
2280 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
2284 error = dpif_port_add(ofproto->dpif, netdev, &odp_port);
2286 *ofp_portp = odp_port_to_ofp_port(odp_port);
2292 port_del(struct ofproto *ofproto_, uint16_t ofp_port)
2294 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
2297 error = dpif_port_del(ofproto->dpif, ofp_port_to_odp_port(ofp_port));
2299 struct ofport_dpif *ofport = get_ofp_port(ofproto, ofp_port);
2301 /* The caller is going to close ofport->up.netdev. If this is a
2302 * bonded port, then the bond is using that netdev, so remove it
2303 * from the bond. The client will need to reconfigure everything
2304 * after deleting ports, so then the slave will get re-added. */
2305 bundle_remove(&ofport->up);
2312 port_get_stats(const struct ofport *ofport_, struct netdev_stats *stats)
2314 struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
2317 error = netdev_get_stats(ofport->up.netdev, stats);
2319 if (!error && ofport->odp_port == OVSP_LOCAL) {
2320 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport->up.ofproto);
2322 /* ofproto->stats.tx_packets represents packets that we created
2323 * internally and sent to some port (e.g. packets sent with
2324 * send_packet()). Account for them as if they had come from
2325 * OFPP_LOCAL and got forwarded. */
2327 if (stats->rx_packets != UINT64_MAX) {
2328 stats->rx_packets += ofproto->stats.tx_packets;
2331 if (stats->rx_bytes != UINT64_MAX) {
2332 stats->rx_bytes += ofproto->stats.tx_bytes;
2335 /* ofproto->stats.rx_packets represents packets that were received on
2336 * some port and we processed internally and dropped (e.g. STP).
2337 * Account fro them as if they had been forwarded to OFPP_LOCAL. */
2339 if (stats->tx_packets != UINT64_MAX) {
2340 stats->tx_packets += ofproto->stats.rx_packets;
2343 if (stats->tx_bytes != UINT64_MAX) {
2344 stats->tx_bytes += ofproto->stats.rx_bytes;
2351 /* Account packets for LOCAL port. */
2353 ofproto_update_local_port_stats(const struct ofproto *ofproto_,
2354 size_t tx_size, size_t rx_size)
2356 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
2359 ofproto->stats.rx_packets++;
2360 ofproto->stats.rx_bytes += rx_size;
2363 ofproto->stats.tx_packets++;
2364 ofproto->stats.tx_bytes += tx_size;
2368 struct port_dump_state {
2369 struct dpif_port_dump dump;
2374 port_dump_start(const struct ofproto *ofproto_, void **statep)
2376 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
2377 struct port_dump_state *state;
2379 *statep = state = xmalloc(sizeof *state);
2380 dpif_port_dump_start(&state->dump, ofproto->dpif);
2381 state->done = false;
2386 port_dump_next(const struct ofproto *ofproto_ OVS_UNUSED, void *state_,
2387 struct ofproto_port *port)
2389 struct port_dump_state *state = state_;
2390 struct dpif_port dpif_port;
2392 if (dpif_port_dump_next(&state->dump, &dpif_port)) {
2393 ofproto_port_from_dpif_port(port, &dpif_port);
2396 int error = dpif_port_dump_done(&state->dump);
2398 return error ? error : EOF;
2403 port_dump_done(const struct ofproto *ofproto_ OVS_UNUSED, void *state_)
2405 struct port_dump_state *state = state_;
2408 dpif_port_dump_done(&state->dump);
2415 port_poll(const struct ofproto *ofproto_, char **devnamep)
2417 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
2418 return dpif_port_poll(ofproto->dpif, devnamep);
2422 port_poll_wait(const struct ofproto *ofproto_)
2424 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
2425 dpif_port_poll_wait(ofproto->dpif);
2429 port_is_lacp_current(const struct ofport *ofport_)
2431 const struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
2432 return (ofport->bundle && ofport->bundle->lacp
2433 ? lacp_slave_is_current(ofport->bundle->lacp, ofport)
2437 /* Upcall handling. */
2439 /* Flow miss batching.
2441 * Some dpifs implement operations faster when you hand them off in a batch.
2442 * To allow batching, "struct flow_miss" queues the dpif-related work needed
2443 * for a given flow. Each "struct flow_miss" corresponds to sending one or
2444 * more packets, plus possibly installing the flow in the dpif.
2446 * So far we only batch the operations that affect flow setup time the most.
2447 * It's possible to batch more than that, but the benefit might be minimal. */
2449 struct hmap_node hmap_node;
2451 enum odp_key_fitness key_fitness;
2452 const struct nlattr *key;
2454 ovs_be16 initial_tci;
2455 struct list packets;
2458 struct flow_miss_op {
2459 struct dpif_op dpif_op;
2460 struct subfacet *subfacet;
2463 /* Sends an OFPT_PACKET_IN message for 'packet' of type OFPR_NO_MATCH to each
2464 * OpenFlow controller as necessary according to their individual
2465 * configurations. */
2467 send_packet_in_miss(struct ofproto_dpif *ofproto, const struct ofpbuf *packet,
2468 const struct flow *flow)
2470 struct ofputil_packet_in pin;
2472 pin.packet = packet->data;
2473 pin.packet_len = packet->size;
2474 pin.total_len = packet->size;
2475 pin.reason = OFPR_NO_MATCH;
2476 pin.controller_id = 0;
2481 pin.buffer_id = 0; /* not yet known */
2482 pin.send_len = 0; /* not used for flow table misses */
2484 flow_get_metadata(flow, &pin.fmd);
2486 /* Registers aren't meaningful on a miss. */
2487 memset(pin.fmd.reg_masks, 0, sizeof pin.fmd.reg_masks);
2489 connmgr_send_packet_in(ofproto->up.connmgr, &pin, flow);
2493 process_special(struct ofproto_dpif *ofproto, const struct flow *flow,
2494 const struct ofpbuf *packet)
2496 struct ofport_dpif *ofport = get_ofp_port(ofproto, flow->in_port);
2502 if (ofport->cfm && cfm_should_process_flow(ofport->cfm, flow)) {
2504 cfm_process_heartbeat(ofport->cfm, packet);
2507 } else if (ofport->bundle && ofport->bundle->lacp
2508 && flow->dl_type == htons(ETH_TYPE_LACP)) {
2510 lacp_process_packet(ofport->bundle->lacp, ofport, packet);
2513 } else if (ofproto->stp && stp_should_process_flow(flow)) {
2515 stp_process_packet(ofport, packet);
2522 static struct flow_miss *
2523 flow_miss_create(struct hmap *todo, const struct flow *flow,
2524 enum odp_key_fitness key_fitness,
2525 const struct nlattr *key, size_t key_len,
2526 ovs_be16 initial_tci)
2528 uint32_t hash = flow_hash(flow, 0);
2529 struct flow_miss *miss;
2531 HMAP_FOR_EACH_WITH_HASH (miss, hmap_node, hash, todo) {
2532 if (flow_equal(&miss->flow, flow)) {
2537 miss = xmalloc(sizeof *miss);
2538 hmap_insert(todo, &miss->hmap_node, hash);
2540 miss->key_fitness = key_fitness;
2542 miss->key_len = key_len;
2543 miss->initial_tci = initial_tci;
2544 list_init(&miss->packets);
2549 handle_flow_miss(struct ofproto_dpif *ofproto, struct flow_miss *miss,
2550 struct flow_miss_op *ops, size_t *n_ops)
2552 const struct flow *flow = &miss->flow;
2553 struct ofpbuf *packet, *next_packet;
2554 struct subfacet *subfacet;
2555 struct facet *facet;
2557 facet = facet_lookup_valid(ofproto, flow);
2559 struct rule_dpif *rule;
2561 rule = rule_dpif_lookup(ofproto, flow, 0);
2563 /* Don't send a packet-in if OFPUTIL_PC_NO_PACKET_IN asserted. */
2564 struct ofport_dpif *port = get_ofp_port(ofproto, flow->in_port);
2566 if (port->up.pp.config & OFPUTIL_PC_NO_PACKET_IN) {
2567 COVERAGE_INC(ofproto_dpif_no_packet_in);
2568 /* XXX install 'drop' flow entry */
2572 VLOG_WARN_RL(&rl, "packet-in on unknown port %"PRIu16,
2576 LIST_FOR_EACH (packet, list_node, &miss->packets) {
2577 send_packet_in_miss(ofproto, packet, flow);
2583 facet = facet_create(rule, flow);
2586 subfacet = subfacet_create(facet,
2587 miss->key_fitness, miss->key, miss->key_len,
2590 LIST_FOR_EACH_SAFE (packet, next_packet, list_node, &miss->packets) {
2591 struct dpif_flow_stats stats;
2592 struct flow_miss_op *op;
2593 struct dpif_execute *execute;
2595 ofproto->n_matches++;
2597 if (facet->rule->up.cr.priority == FAIL_OPEN_PRIORITY) {
2599 * Extra-special case for fail-open mode.
2601 * We are in fail-open mode and the packet matched the fail-open
2602 * rule, but we are connected to a controller too. We should send
2603 * the packet up to the controller in the hope that it will try to
2604 * set up a flow and thereby allow us to exit fail-open.
2606 * See the top-level comment in fail-open.c for more information.
2608 send_packet_in_miss(ofproto, packet, flow);
2611 if (!facet->may_install || !subfacet->actions) {
2612 subfacet_make_actions(subfacet, packet);
2615 dpif_flow_stats_extract(&facet->flow, packet, &stats);
2616 subfacet_update_stats(subfacet, &stats);
2618 if (!subfacet->actions_len) {
2619 /* No actions to execute, so skip talking to the dpif. */
2623 if (flow->vlan_tci != subfacet->initial_tci) {
2624 /* This packet was received on a VLAN splinter port. We added
2625 * a VLAN to the packet to make the packet resemble the flow,
2626 * but the actions were composed assuming that the packet
2627 * contained no VLAN. So, we must remove the VLAN header from
2628 * the packet before trying to execute the actions. */
2629 eth_pop_vlan(packet);
2632 op = &ops[(*n_ops)++];
2633 execute = &op->dpif_op.u.execute;
2634 op->subfacet = subfacet;
2635 op->dpif_op.type = DPIF_OP_EXECUTE;
2636 execute->key = miss->key;
2637 execute->key_len = miss->key_len;
2638 execute->actions = (facet->may_install
2640 : xmemdup(subfacet->actions,
2641 subfacet->actions_len));
2642 execute->actions_len = subfacet->actions_len;
2643 execute->packet = packet;
2646 if (facet->may_install && subfacet->key_fitness != ODP_FIT_TOO_LITTLE) {
2647 struct flow_miss_op *op = &ops[(*n_ops)++];
2648 struct dpif_flow_put *put = &op->dpif_op.u.flow_put;
2650 op->subfacet = subfacet;
2651 op->dpif_op.type = DPIF_OP_FLOW_PUT;
2652 put->flags = DPIF_FP_CREATE | DPIF_FP_MODIFY;
2653 put->key = miss->key;
2654 put->key_len = miss->key_len;
2655 put->actions = subfacet->actions;
2656 put->actions_len = subfacet->actions_len;
2661 /* Like odp_flow_key_to_flow(), this function converts the 'key_len' bytes of
2662 * OVS_KEY_ATTR_* attributes in 'key' to a flow structure in 'flow' and returns
2663 * an ODP_FIT_* value that indicates how well 'key' fits our expectations for
2664 * what a flow key should contain.
2666 * This function also includes some logic to help make VLAN splinters
2667 * transparent to the rest of the upcall processing logic. In particular, if
2668 * the extracted in_port is a VLAN splinter port, it replaces flow->in_port by
2669 * the "real" port, sets flow->vlan_tci correctly for the VLAN of the VLAN
2670 * splinter port, and pushes a VLAN header onto 'packet' (if it is nonnull).
2672 * Sets '*initial_tci' to the VLAN TCI with which the packet was really
2673 * received, that is, the actual VLAN TCI extracted by odp_flow_key_to_flow().
2674 * (This differs from the value returned in flow->vlan_tci only for packets
2675 * received on VLAN splinters.)
2677 static enum odp_key_fitness
2678 ofproto_dpif_extract_flow_key(const struct ofproto_dpif *ofproto,
2679 const struct nlattr *key, size_t key_len,
2680 struct flow *flow, ovs_be16 *initial_tci,
2681 struct ofpbuf *packet)
2683 enum odp_key_fitness fitness;
2687 fitness = odp_flow_key_to_flow(key, key_len, flow);
2688 if (fitness == ODP_FIT_ERROR) {
2691 *initial_tci = flow->vlan_tci;
2693 realdev = vsp_vlandev_to_realdev(ofproto, flow->in_port, &vid);
2695 /* Cause the flow to be processed as if it came in on the real device
2696 * with the VLAN device's VLAN ID. */
2697 flow->in_port = realdev;
2698 flow->vlan_tci = htons((vid & VLAN_VID_MASK) | VLAN_CFI);
2700 /* Make the packet resemble the flow, so that it gets sent to an
2701 * OpenFlow controller properly, so that it looks correct for
2702 * sFlow, and so that flow_extract() will get the correct vlan_tci
2703 * if it is called on 'packet'.
2705 * The allocated space inside 'packet' probably also contains
2706 * 'key', that is, both 'packet' and 'key' are probably part of a
2707 * struct dpif_upcall (see the large comment on that structure
2708 * definition), so pushing data on 'packet' is in general not a
2709 * good idea since it could overwrite 'key' or free it as a side
2710 * effect. However, it's OK in this special case because we know
2711 * that 'packet' is inside a Netlink attribute: pushing 4 bytes
2712 * will just overwrite the 4-byte "struct nlattr", which is fine
2713 * since we don't need that header anymore. */
2714 eth_push_vlan(packet, flow->vlan_tci);
2717 /* Let the caller know that we can't reproduce 'key' from 'flow'. */
2718 if (fitness == ODP_FIT_PERFECT) {
2719 fitness = ODP_FIT_TOO_MUCH;
2727 handle_miss_upcalls(struct ofproto_dpif *ofproto, struct dpif_upcall *upcalls,
2730 struct dpif_upcall *upcall;
2731 struct flow_miss *miss, *next_miss;
2732 struct flow_miss_op flow_miss_ops[FLOW_MISS_MAX_BATCH * 2];
2733 struct dpif_op *dpif_ops[FLOW_MISS_MAX_BATCH * 2];
2742 /* Construct the to-do list.
2744 * This just amounts to extracting the flow from each packet and sticking
2745 * the packets that have the same flow in the same "flow_miss" structure so
2746 * that we can process them together. */
2748 for (upcall = upcalls; upcall < &upcalls[n_upcalls]; upcall++) {
2749 enum odp_key_fitness fitness;
2750 struct flow_miss *miss;
2751 ovs_be16 initial_tci;
2754 /* Obtain metadata and check userspace/kernel agreement on flow match,
2755 * then set 'flow''s header pointers. */
2756 fitness = ofproto_dpif_extract_flow_key(ofproto,
2757 upcall->key, upcall->key_len,
2758 &flow, &initial_tci,
2760 if (fitness == ODP_FIT_ERROR) {
2761 ofpbuf_delete(upcall->packet);
2764 flow_extract(upcall->packet, flow.skb_priority, flow.tun_id,
2765 flow.in_port, &flow);
2767 /* Handle 802.1ag, LACP, and STP specially. */
2768 if (process_special(ofproto, &flow, upcall->packet)) {
2769 ofproto_update_local_port_stats(&ofproto->up,
2770 0, upcall->packet->size);
2771 ofpbuf_delete(upcall->packet);
2772 ofproto->n_matches++;
2776 /* Add other packets to a to-do list. */
2777 miss = flow_miss_create(&todo, &flow, fitness,
2778 upcall->key, upcall->key_len, initial_tci);
2779 list_push_back(&miss->packets, &upcall->packet->list_node);
2782 /* Process each element in the to-do list, constructing the set of
2783 * operations to batch. */
2785 HMAP_FOR_EACH (miss, hmap_node, &todo) {
2786 handle_flow_miss(ofproto, miss, flow_miss_ops, &n_ops);
2788 assert(n_ops <= ARRAY_SIZE(flow_miss_ops));
2790 /* Execute batch. */
2791 for (i = 0; i < n_ops; i++) {
2792 dpif_ops[i] = &flow_miss_ops[i].dpif_op;
2794 dpif_operate(ofproto->dpif, dpif_ops, n_ops);
2796 /* Free memory and update facets. */
2797 for (i = 0; i < n_ops; i++) {
2798 struct flow_miss_op *op = &flow_miss_ops[i];
2799 struct dpif_execute *execute;
2801 switch (op->dpif_op.type) {
2802 case DPIF_OP_EXECUTE:
2803 execute = &op->dpif_op.u.execute;
2804 if (op->subfacet->actions != execute->actions) {
2805 free((struct nlattr *) execute->actions);
2809 case DPIF_OP_FLOW_PUT:
2810 if (!op->dpif_op.error) {
2811 op->subfacet->installed = true;
2816 HMAP_FOR_EACH_SAFE (miss, next_miss, hmap_node, &todo) {
2817 ofpbuf_list_delete(&miss->packets);
2818 hmap_remove(&todo, &miss->hmap_node);
2821 hmap_destroy(&todo);
2825 handle_userspace_upcall(struct ofproto_dpif *ofproto,
2826 struct dpif_upcall *upcall)
2828 struct user_action_cookie cookie;
2829 enum odp_key_fitness fitness;
2830 ovs_be16 initial_tci;
2833 memcpy(&cookie, &upcall->userdata, sizeof(cookie));
2835 fitness = ofproto_dpif_extract_flow_key(ofproto, upcall->key,
2836 upcall->key_len, &flow,
2837 &initial_tci, upcall->packet);
2838 if (fitness == ODP_FIT_ERROR) {
2839 ofpbuf_delete(upcall->packet);
2843 if (cookie.type == USER_ACTION_COOKIE_SFLOW) {
2844 if (ofproto->sflow) {
2845 dpif_sflow_received(ofproto->sflow, upcall->packet, &flow,
2849 VLOG_WARN_RL(&rl, "invalid user cookie : 0x%"PRIx64, upcall->userdata);
2851 ofpbuf_delete(upcall->packet);
2855 handle_upcalls(struct ofproto_dpif *ofproto, unsigned int max_batch)
2857 struct dpif_upcall misses[FLOW_MISS_MAX_BATCH];
2861 assert (max_batch <= FLOW_MISS_MAX_BATCH);
2864 for (i = 0; i < max_batch; i++) {
2865 struct dpif_upcall *upcall = &misses[n_misses];
2868 error = dpif_recv(ofproto->dpif, upcall);
2873 switch (upcall->type) {
2874 case DPIF_UC_ACTION:
2875 handle_userspace_upcall(ofproto, upcall);
2879 /* Handle it later. */
2883 case DPIF_N_UC_TYPES:
2885 VLOG_WARN_RL(&rl, "upcall has unexpected type %"PRIu32,
2891 handle_miss_upcalls(ofproto, misses, n_misses);
2896 /* Flow expiration. */
2898 static int subfacet_max_idle(const struct ofproto_dpif *);
2899 static void update_stats(struct ofproto_dpif *);
2900 static void rule_expire(struct rule_dpif *);
2901 static void expire_subfacets(struct ofproto_dpif *, int dp_max_idle);
2903 /* This function is called periodically by run(). Its job is to collect
2904 * updates for the flows that have been installed into the datapath, most
2905 * importantly when they last were used, and then use that information to
2906 * expire flows that have not been used recently.
2908 * Returns the number of milliseconds after which it should be called again. */
2910 expire(struct ofproto_dpif *ofproto)
2912 struct rule_dpif *rule, *next_rule;
2913 struct oftable *table;
2916 /* Update stats for each flow in the datapath. */
2917 update_stats(ofproto);
2919 /* Expire subfacets that have been idle too long. */
2920 dp_max_idle = subfacet_max_idle(ofproto);
2921 expire_subfacets(ofproto, dp_max_idle);
2923 /* Expire OpenFlow flows whose idle_timeout or hard_timeout has passed. */
2924 OFPROTO_FOR_EACH_TABLE (table, &ofproto->up) {
2925 struct cls_cursor cursor;
2927 cls_cursor_init(&cursor, &table->cls, NULL);
2928 CLS_CURSOR_FOR_EACH_SAFE (rule, next_rule, up.cr, &cursor) {
2933 /* All outstanding data in existing flows has been accounted, so it's a
2934 * good time to do bond rebalancing. */
2935 if (ofproto->has_bonded_bundles) {
2936 struct ofbundle *bundle;
2938 HMAP_FOR_EACH (bundle, hmap_node, &ofproto->bundles) {
2940 bond_rebalance(bundle->bond, &ofproto->revalidate_set);
2945 return MIN(dp_max_idle, 1000);
2948 /* Update 'packet_count', 'byte_count', and 'used' members of installed facets.
2950 * This function also pushes statistics updates to rules which each facet
2951 * resubmits into. Generally these statistics will be accurate. However, if a
2952 * facet changes the rule it resubmits into at some time in between
2953 * update_stats() runs, it is possible that statistics accrued to the
2954 * old rule will be incorrectly attributed to the new rule. This could be
2955 * avoided by calling update_stats() whenever rules are created or
2956 * deleted. However, the performance impact of making so many calls to the
2957 * datapath do not justify the benefit of having perfectly accurate statistics.
2960 update_stats(struct ofproto_dpif *p)
2962 const struct dpif_flow_stats *stats;
2963 struct dpif_flow_dump dump;
2964 const struct nlattr *key;
2967 dpif_flow_dump_start(&dump, p->dpif);
2968 while (dpif_flow_dump_next(&dump, &key, &key_len, NULL, NULL, &stats)) {
2969 struct subfacet *subfacet;
2971 subfacet = subfacet_find(p, key, key_len);
2972 if (subfacet && subfacet->installed) {
2973 struct facet *facet = subfacet->facet;
2975 if (stats->n_packets >= subfacet->dp_packet_count) {
2976 uint64_t extra = stats->n_packets - subfacet->dp_packet_count;
2977 facet->packet_count += extra;
2979 VLOG_WARN_RL(&rl, "unexpected packet count from the datapath");
2982 if (stats->n_bytes >= subfacet->dp_byte_count) {
2983 facet->byte_count += stats->n_bytes - subfacet->dp_byte_count;
2985 VLOG_WARN_RL(&rl, "unexpected byte count from datapath");
2988 subfacet->dp_packet_count = stats->n_packets;
2989 subfacet->dp_byte_count = stats->n_bytes;
2991 facet->tcp_flags |= stats->tcp_flags;
2993 subfacet_update_time(subfacet, stats->used);
2994 facet_account(facet);
2995 facet_push_stats(facet);
2997 if (!VLOG_DROP_WARN(&rl)) {
3001 odp_flow_key_format(key, key_len, &s);
3002 VLOG_WARN("unexpected flow from datapath %s", ds_cstr(&s));
3006 COVERAGE_INC(facet_unexpected);
3007 /* There's a flow in the datapath that we know nothing about, or a
3008 * flow that shouldn't be installed but was anyway. Delete it. */
3009 dpif_flow_del(p->dpif, key, key_len, NULL);
3012 dpif_flow_dump_done(&dump);
3015 /* Calculates and returns the number of milliseconds of idle time after which
3016 * subfacets should expire from the datapath. When a subfacet expires, we fold
3017 * its statistics into its facet, and when a facet's last subfacet expires, we
3018 * fold its statistic into its rule. */
3020 subfacet_max_idle(const struct ofproto_dpif *ofproto)
3023 * Idle time histogram.
3025 * Most of the time a switch has a relatively small number of subfacets.
3026 * When this is the case we might as well keep statistics for all of them
3027 * in userspace and to cache them in the kernel datapath for performance as
3030 * As the number of subfacets increases, the memory required to maintain
3031 * statistics about them in userspace and in the kernel becomes
3032 * significant. However, with a large number of subfacets it is likely
3033 * that only a few of them are "heavy hitters" that consume a large amount
3034 * of bandwidth. At this point, only heavy hitters are worth caching in
3035 * the kernel and maintaining in userspaces; other subfacets we can
3038 * The technique used to compute the idle time is to build a histogram with
3039 * N_BUCKETS buckets whose width is BUCKET_WIDTH msecs each. Each subfacet
3040 * that is installed in the kernel gets dropped in the appropriate bucket.
3041 * After the histogram has been built, we compute the cutoff so that only
3042 * the most-recently-used 1% of subfacets (but at least
3043 * ofproto->up.flow_eviction_threshold flows) are kept cached. At least
3044 * the most-recently-used bucket of subfacets is kept, so actually an
3045 * arbitrary number of subfacets can be kept in any given expiration run
3046 * (though the next run will delete most of those unless they receive
3049 * This requires a second pass through the subfacets, in addition to the
3050 * pass made by update_stats(), because the former function never looks at
3051 * uninstallable subfacets.
3053 enum { BUCKET_WIDTH = ROUND_UP(100, TIME_UPDATE_INTERVAL) };
3054 enum { N_BUCKETS = 5000 / BUCKET_WIDTH };
3055 int buckets[N_BUCKETS] = { 0 };
3056 int total, subtotal, bucket;
3057 struct subfacet *subfacet;
3061 total = hmap_count(&ofproto->subfacets);
3062 if (total <= ofproto->up.flow_eviction_threshold) {
3063 return N_BUCKETS * BUCKET_WIDTH;
3066 /* Build histogram. */
3068 HMAP_FOR_EACH (subfacet, hmap_node, &ofproto->subfacets) {
3069 long long int idle = now - subfacet->used;
3070 int bucket = (idle <= 0 ? 0
3071 : idle >= BUCKET_WIDTH * N_BUCKETS ? N_BUCKETS - 1
3072 : (unsigned int) idle / BUCKET_WIDTH);
3076 /* Find the first bucket whose flows should be expired. */
3077 subtotal = bucket = 0;
3079 subtotal += buckets[bucket++];
3080 } while (bucket < N_BUCKETS &&
3081 subtotal < MAX(ofproto->up.flow_eviction_threshold, total / 100));
3083 if (VLOG_IS_DBG_ENABLED()) {
3087 ds_put_cstr(&s, "keep");
3088 for (i = 0; i < N_BUCKETS; i++) {
3090 ds_put_cstr(&s, ", drop");
3093 ds_put_format(&s, " %d:%d", i * BUCKET_WIDTH, buckets[i]);
3096 VLOG_INFO("%s: %s (msec:count)", ofproto->up.name, ds_cstr(&s));
3100 return bucket * BUCKET_WIDTH;
3104 expire_subfacets(struct ofproto_dpif *ofproto, int dp_max_idle)
3106 long long int cutoff = time_msec() - dp_max_idle;
3107 struct subfacet *subfacet, *next_subfacet;
3109 HMAP_FOR_EACH_SAFE (subfacet, next_subfacet, hmap_node,
3110 &ofproto->subfacets) {
3111 if (subfacet->used < cutoff) {
3112 subfacet_destroy(subfacet);
3117 /* If 'rule' is an OpenFlow rule, that has expired according to OpenFlow rules,
3118 * then delete it entirely. */
3120 rule_expire(struct rule_dpif *rule)
3122 struct facet *facet, *next_facet;
3126 /* Has 'rule' expired? */
3128 if (rule->up.hard_timeout
3129 && now > rule->up.modified + rule->up.hard_timeout * 1000) {
3130 reason = OFPRR_HARD_TIMEOUT;
3131 } else if (rule->up.idle_timeout
3132 && now > rule->up.used + rule->up.idle_timeout * 1000) {
3133 reason = OFPRR_IDLE_TIMEOUT;
3138 COVERAGE_INC(ofproto_dpif_expired);
3140 /* Update stats. (This is a no-op if the rule expired due to an idle
3141 * timeout, because that only happens when the rule has no facets left.) */
3142 LIST_FOR_EACH_SAFE (facet, next_facet, list_node, &rule->facets) {
3143 facet_remove(facet);
3146 /* Get rid of the rule. */
3147 ofproto_rule_expire(&rule->up, reason);
3152 /* Creates and returns a new facet owned by 'rule', given a 'flow'.
3154 * The caller must already have determined that no facet with an identical
3155 * 'flow' exists in 'ofproto' and that 'flow' is the best match for 'rule' in
3156 * the ofproto's classifier table.
3158 * The facet will initially have no subfacets. The caller should create (at
3159 * least) one subfacet with subfacet_create(). */
3160 static struct facet *
3161 facet_create(struct rule_dpif *rule, const struct flow *flow)
3163 struct ofproto_dpif *ofproto = ofproto_dpif_cast(rule->up.ofproto);
3164 struct facet *facet;
3166 facet = xzalloc(sizeof *facet);
3167 facet->used = time_msec();
3168 hmap_insert(&ofproto->facets, &facet->hmap_node, flow_hash(flow, 0));
3169 list_push_back(&rule->facets, &facet->list_node);
3171 facet->flow = *flow;
3172 list_init(&facet->subfacets);
3173 netflow_flow_init(&facet->nf_flow);
3174 netflow_flow_update_time(ofproto->netflow, &facet->nf_flow, facet->used);
3180 facet_free(struct facet *facet)
3185 /* Executes, within 'ofproto', the 'n_actions' actions in 'actions' on
3186 * 'packet', which arrived on 'in_port'.
3188 * Takes ownership of 'packet'. */
3190 execute_odp_actions(struct ofproto_dpif *ofproto, const struct flow *flow,
3191 const struct nlattr *odp_actions, size_t actions_len,
3192 struct ofpbuf *packet)
3194 struct odputil_keybuf keybuf;
3198 ofpbuf_use_stack(&key, &keybuf, sizeof keybuf);
3199 odp_flow_key_from_flow(&key, flow);
3201 error = dpif_execute(ofproto->dpif, key.data, key.size,
3202 odp_actions, actions_len, packet);
3204 ofpbuf_delete(packet);
3208 /* Remove 'facet' from 'ofproto' and free up the associated memory:
3210 * - If 'facet' was installed in the datapath, uninstalls it and updates its
3211 * rule's statistics, via subfacet_uninstall().
3213 * - Removes 'facet' from its rule and from ofproto->facets.
3216 facet_remove(struct facet *facet)
3218 struct ofproto_dpif *ofproto = ofproto_dpif_cast(facet->rule->up.ofproto);
3219 struct subfacet *subfacet, *next_subfacet;
3221 assert(!list_is_empty(&facet->subfacets));
3223 /* First uninstall all of the subfacets to get final statistics. */
3224 LIST_FOR_EACH (subfacet, list_node, &facet->subfacets) {
3225 subfacet_uninstall(subfacet);
3228 /* Flush the final stats to the rule.
3230 * This might require us to have at least one subfacet around so that we
3231 * can use its actions for accounting in facet_account(), which is why we
3232 * have uninstalled but not yet destroyed the subfacets. */
3233 facet_flush_stats(facet);
3235 /* Now we're really all done so destroy everything. */
3236 LIST_FOR_EACH_SAFE (subfacet, next_subfacet, list_node,
3237 &facet->subfacets) {
3238 subfacet_destroy__(subfacet);
3240 hmap_remove(&ofproto->facets, &facet->hmap_node);
3241 list_remove(&facet->list_node);
3246 facet_account(struct facet *facet)
3248 struct ofproto_dpif *ofproto = ofproto_dpif_cast(facet->rule->up.ofproto);
3250 struct subfacet *subfacet;
3251 const struct nlattr *a;
3255 if (facet->byte_count <= facet->accounted_bytes) {
3258 n_bytes = facet->byte_count - facet->accounted_bytes;
3259 facet->accounted_bytes = facet->byte_count;
3261 /* Feed information from the active flows back into the learning table to
3262 * ensure that table is always in sync with what is actually flowing
3263 * through the datapath. */
3264 if (facet->has_learn || facet->has_normal
3265 || (facet->has_fin_timeout
3266 && facet->tcp_flags & (TCP_FIN | TCP_RST))) {
3267 struct action_xlate_ctx ctx;
3269 action_xlate_ctx_init(&ctx, ofproto, &facet->flow,
3270 facet->flow.vlan_tci,
3271 facet->rule, facet->tcp_flags, NULL);
3272 ctx.may_learn = true;
3273 ofpbuf_delete(xlate_actions(&ctx, facet->rule->up.actions,
3274 facet->rule->up.n_actions));
3277 if (!facet->has_normal || !ofproto->has_bonded_bundles) {
3281 /* This loop feeds byte counters to bond_account() for rebalancing to use
3282 * as a basis. We also need to track the actual VLAN on which the packet
3283 * is going to be sent to ensure that it matches the one passed to
3284 * bond_choose_output_slave(). (Otherwise, we will account to the wrong
3287 * We use the actions from an arbitrary subfacet because they should all
3288 * be equally valid for our purpose. */
3289 subfacet = CONTAINER_OF(list_front(&facet->subfacets),
3290 struct subfacet, list_node);
3291 vlan_tci = facet->flow.vlan_tci;
3292 NL_ATTR_FOR_EACH_UNSAFE (a, left,
3293 subfacet->actions, subfacet->actions_len) {
3294 const struct ovs_action_push_vlan *vlan;
3295 struct ofport_dpif *port;
3297 switch (nl_attr_type(a)) {
3298 case OVS_ACTION_ATTR_OUTPUT:
3299 port = get_odp_port(ofproto, nl_attr_get_u32(a));
3300 if (port && port->bundle && port->bundle->bond) {
3301 bond_account(port->bundle->bond, &facet->flow,
3302 vlan_tci_to_vid(vlan_tci), n_bytes);
3306 case OVS_ACTION_ATTR_POP_VLAN:
3307 vlan_tci = htons(0);
3310 case OVS_ACTION_ATTR_PUSH_VLAN:
3311 vlan = nl_attr_get(a);
3312 vlan_tci = vlan->vlan_tci;
3318 /* Returns true if the only action for 'facet' is to send to the controller.
3319 * (We don't report NetFlow expiration messages for such facets because they
3320 * are just part of the control logic for the network, not real traffic). */
3322 facet_is_controller_flow(struct facet *facet)
3325 && facet->rule->up.n_actions == 1
3326 && action_outputs_to_port(&facet->rule->up.actions[0],
3327 htons(OFPP_CONTROLLER)));
3330 /* Folds all of 'facet''s statistics into its rule. Also updates the
3331 * accounting ofhook and emits a NetFlow expiration if appropriate. All of
3332 * 'facet''s statistics in the datapath should have been zeroed and folded into
3333 * its packet and byte counts before this function is called. */
3335 facet_flush_stats(struct facet *facet)
3337 struct ofproto_dpif *ofproto = ofproto_dpif_cast(facet->rule->up.ofproto);
3338 struct subfacet *subfacet;
3340 LIST_FOR_EACH (subfacet, list_node, &facet->subfacets) {
3341 assert(!subfacet->dp_byte_count);
3342 assert(!subfacet->dp_packet_count);
3345 facet_push_stats(facet);
3346 facet_account(facet);
3348 if (ofproto->netflow && !facet_is_controller_flow(facet)) {
3349 struct ofexpired expired;
3350 expired.flow = facet->flow;
3351 expired.packet_count = facet->packet_count;
3352 expired.byte_count = facet->byte_count;
3353 expired.used = facet->used;
3354 netflow_expire(ofproto->netflow, &facet->nf_flow, &expired);
3357 facet->rule->packet_count += facet->packet_count;
3358 facet->rule->byte_count += facet->byte_count;
3360 /* Reset counters to prevent double counting if 'facet' ever gets
3362 facet_reset_counters(facet);
3364 netflow_flow_clear(&facet->nf_flow);
3365 facet->tcp_flags = 0;
3368 /* Searches 'ofproto''s table of facets for one exactly equal to 'flow'.
3369 * Returns it if found, otherwise a null pointer.
3371 * The returned facet might need revalidation; use facet_lookup_valid()
3372 * instead if that is important. */
3373 static struct facet *
3374 facet_find(struct ofproto_dpif *ofproto, const struct flow *flow)
3376 struct facet *facet;
3378 HMAP_FOR_EACH_WITH_HASH (facet, hmap_node, flow_hash(flow, 0),
3380 if (flow_equal(flow, &facet->flow)) {
3388 /* Searches 'ofproto''s table of facets for one exactly equal to 'flow'.
3389 * Returns it if found, otherwise a null pointer.
3391 * The returned facet is guaranteed to be valid. */
3392 static struct facet *
3393 facet_lookup_valid(struct ofproto_dpif *ofproto, const struct flow *flow)
3395 struct facet *facet = facet_find(ofproto, flow);
3397 /* The facet we found might not be valid, since we could be in need of
3398 * revalidation. If it is not valid, don't return it. */
3400 && (ofproto->need_revalidate
3401 || tag_set_intersects(&ofproto->revalidate_set, facet->tags))
3402 && !facet_revalidate(facet)) {
3403 COVERAGE_INC(facet_invalidated);
3411 facet_check_consistency(struct facet *facet)
3413 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 15);
3415 struct ofproto_dpif *ofproto = ofproto_dpif_cast(facet->rule->up.ofproto);
3417 struct rule_dpif *rule;
3418 struct subfacet *subfacet;
3419 bool may_log = false;
3422 /* Check the rule for consistency. */
3423 rule = rule_dpif_lookup(ofproto, &facet->flow, 0);
3425 if (!VLOG_DROP_WARN(&rl)) {
3426 char *s = flow_to_string(&facet->flow);
3427 VLOG_WARN("%s: facet should not exist", s);
3431 } else if (rule != facet->rule) {
3432 may_log = !VLOG_DROP_WARN(&rl);
3438 flow_format(&s, &facet->flow);
3439 ds_put_format(&s, ": facet associated with wrong rule (was "
3440 "table=%"PRIu8",", facet->rule->up.table_id);
3441 cls_rule_format(&facet->rule->up.cr, &s);
3442 ds_put_format(&s, ") (should have been table=%"PRIu8",",
3444 cls_rule_format(&rule->up.cr, &s);
3445 ds_put_char(&s, ')');
3447 VLOG_WARN("%s", ds_cstr(&s));
3454 /* Check the datapath actions for consistency. */
3455 LIST_FOR_EACH (subfacet, list_node, &facet->subfacets) {
3456 struct action_xlate_ctx ctx;
3457 struct ofpbuf *odp_actions;
3458 bool actions_changed;
3459 bool should_install;
3461 action_xlate_ctx_init(&ctx, ofproto, &facet->flow,
3462 subfacet->initial_tci, rule, 0, NULL);
3463 odp_actions = xlate_actions(&ctx, rule->up.actions,
3464 rule->up.n_actions);
3466 should_install = (ctx.may_set_up_flow
3467 && subfacet->key_fitness != ODP_FIT_TOO_LITTLE);
3468 if (!should_install && !subfacet->installed) {
3469 /* The actions for uninstallable flows may vary from one packet to
3470 * the next, so don't compare the actions. */
3474 actions_changed = (subfacet->actions_len != odp_actions->size
3475 || memcmp(subfacet->actions, odp_actions->data,
3476 subfacet->actions_len));
3477 if (should_install != subfacet->installed || actions_changed) {
3479 may_log = !VLOG_DROP_WARN(&rl);
3484 struct odputil_keybuf keybuf;
3489 subfacet_get_key(subfacet, &keybuf, &key);
3490 odp_flow_key_format(key.data, key.size, &s);
3492 ds_put_cstr(&s, ": inconsistency in subfacet");
3493 if (should_install != subfacet->installed) {
3494 enum odp_key_fitness fitness = subfacet->key_fitness;
3496 ds_put_format(&s, " (should%s have been installed)",
3497 should_install ? "" : " not");
3498 ds_put_format(&s, " (may_set_up_flow=%s, fitness=%s)",
3499 ctx.may_set_up_flow ? "true" : "false",
3500 odp_key_fitness_to_string(fitness));
3502 if (actions_changed) {
3503 ds_put_cstr(&s, " (actions were: ");
3504 format_odp_actions(&s, subfacet->actions,
3505 subfacet->actions_len);
3506 ds_put_cstr(&s, ") (correct actions: ");
3507 format_odp_actions(&s, odp_actions->data,
3509 ds_put_char(&s, ')');
3511 ds_put_cstr(&s, " (actions: ");
3512 format_odp_actions(&s, subfacet->actions,
3513 subfacet->actions_len);
3514 ds_put_char(&s, ')');
3516 VLOG_WARN("%s", ds_cstr(&s));
3522 ofpbuf_delete(odp_actions);
3528 /* Re-searches the classifier for 'facet':
3530 * - If the rule found is different from 'facet''s current rule, moves
3531 * 'facet' to the new rule and recompiles its actions.
3533 * - If the rule found is the same as 'facet''s current rule, leaves 'facet'
3534 * where it is and recompiles its actions anyway.
3536 * - If there is none, destroys 'facet'.
3538 * Returns true if 'facet' still exists, false if it has been destroyed. */
3540 facet_revalidate(struct facet *facet)
3542 struct ofproto_dpif *ofproto = ofproto_dpif_cast(facet->rule->up.ofproto);
3544 struct nlattr *odp_actions;
3547 struct actions *new_actions;
3549 struct action_xlate_ctx ctx;
3550 struct rule_dpif *new_rule;
3551 struct subfacet *subfacet;
3552 bool actions_changed;
3555 COVERAGE_INC(facet_revalidate);
3557 /* Determine the new rule. */
3558 new_rule = rule_dpif_lookup(ofproto, &facet->flow, 0);
3560 /* No new rule, so delete the facet. */
3561 facet_remove(facet);
3565 /* Calculate new datapath actions.
3567 * We do not modify any 'facet' state yet, because we might need to, e.g.,
3568 * emit a NetFlow expiration and, if so, we need to have the old state
3569 * around to properly compose it. */
3571 /* If the datapath actions changed or the installability changed,
3572 * then we need to talk to the datapath. */
3575 memset(&ctx, 0, sizeof ctx);
3576 LIST_FOR_EACH (subfacet, list_node, &facet->subfacets) {
3577 struct ofpbuf *odp_actions;
3578 bool should_install;
3580 action_xlate_ctx_init(&ctx, ofproto, &facet->flow,
3581 subfacet->initial_tci, new_rule, 0, NULL);
3582 odp_actions = xlate_actions(&ctx, new_rule->up.actions,
3583 new_rule->up.n_actions);
3584 actions_changed = (subfacet->actions_len != odp_actions->size
3585 || memcmp(subfacet->actions, odp_actions->data,
3586 subfacet->actions_len));
3588 should_install = (ctx.may_set_up_flow
3589 && subfacet->key_fitness != ODP_FIT_TOO_LITTLE);
3590 if (actions_changed || should_install != subfacet->installed) {
3591 if (should_install) {
3592 struct dpif_flow_stats stats;
3594 subfacet_install(subfacet,
3595 odp_actions->data, odp_actions->size, &stats);
3596 subfacet_update_stats(subfacet, &stats);
3598 subfacet_uninstall(subfacet);
3602 new_actions = xcalloc(list_size(&facet->subfacets),
3603 sizeof *new_actions);
3605 new_actions[i].odp_actions = xmemdup(odp_actions->data,
3607 new_actions[i].actions_len = odp_actions->size;
3610 ofpbuf_delete(odp_actions);
3614 facet_flush_stats(facet);
3617 /* Update 'facet' now that we've taken care of all the old state. */
3618 facet->tags = ctx.tags;
3619 facet->nf_flow.output_iface = ctx.nf_output_iface;
3620 facet->may_install = ctx.may_set_up_flow;
3621 facet->has_learn = ctx.has_learn;
3622 facet->has_normal = ctx.has_normal;
3623 facet->has_fin_timeout = ctx.has_fin_timeout;
3624 facet->mirrors = ctx.mirrors;
3627 LIST_FOR_EACH (subfacet, list_node, &facet->subfacets) {
3628 if (new_actions[i].odp_actions) {
3629 free(subfacet->actions);
3630 subfacet->actions = new_actions[i].odp_actions;
3631 subfacet->actions_len = new_actions[i].actions_len;
3637 if (facet->rule != new_rule) {
3638 COVERAGE_INC(facet_changed_rule);
3639 list_remove(&facet->list_node);
3640 list_push_back(&new_rule->facets, &facet->list_node);
3641 facet->rule = new_rule;
3642 facet->used = new_rule->up.created;
3643 facet->prev_used = facet->used;
3649 /* Updates 'facet''s used time. Caller is responsible for calling
3650 * facet_push_stats() to update the flows which 'facet' resubmits into. */
3652 facet_update_time(struct facet *facet, long long int used)
3654 struct ofproto_dpif *ofproto = ofproto_dpif_cast(facet->rule->up.ofproto);
3655 if (used > facet->used) {
3657 ofproto_rule_update_used(&facet->rule->up, used);
3658 netflow_flow_update_time(ofproto->netflow, &facet->nf_flow, used);
3663 facet_reset_counters(struct facet *facet)
3665 facet->packet_count = 0;
3666 facet->byte_count = 0;
3667 facet->prev_packet_count = 0;
3668 facet->prev_byte_count = 0;
3669 facet->accounted_bytes = 0;
3673 facet_push_stats(struct facet *facet)
3675 uint64_t new_packets, new_bytes;
3677 assert(facet->packet_count >= facet->prev_packet_count);
3678 assert(facet->byte_count >= facet->prev_byte_count);
3679 assert(facet->used >= facet->prev_used);
3681 new_packets = facet->packet_count - facet->prev_packet_count;
3682 new_bytes = facet->byte_count - facet->prev_byte_count;
3684 if (new_packets || new_bytes || facet->used > facet->prev_used) {
3685 facet->prev_packet_count = facet->packet_count;
3686 facet->prev_byte_count = facet->byte_count;
3687 facet->prev_used = facet->used;
3689 flow_push_stats(facet->rule, &facet->flow,
3690 new_packets, new_bytes, facet->used);
3692 update_mirror_stats(ofproto_dpif_cast(facet->rule->up.ofproto),
3693 facet->mirrors, new_packets, new_bytes);
3697 struct ofproto_push {
3698 struct action_xlate_ctx ctx;
3705 push_resubmit(struct action_xlate_ctx *ctx, struct rule_dpif *rule)
3707 struct ofproto_push *push = CONTAINER_OF(ctx, struct ofproto_push, ctx);
3710 rule->packet_count += push->packets;
3711 rule->byte_count += push->bytes;
3712 ofproto_rule_update_used(&rule->up, push->used);
3716 /* Pushes flow statistics to the rules which 'flow' resubmits into given
3717 * 'rule''s actions and mirrors. */
3719 flow_push_stats(struct rule_dpif *rule,
3720 const struct flow *flow, uint64_t packets, uint64_t bytes,
3723 struct ofproto_dpif *ofproto = ofproto_dpif_cast(rule->up.ofproto);
3724 struct ofproto_push push;
3726 push.packets = packets;
3730 ofproto_rule_update_used(&rule->up, used);
3732 action_xlate_ctx_init(&push.ctx, ofproto, flow, flow->vlan_tci, rule,
3734 push.ctx.resubmit_hook = push_resubmit;
3735 ofpbuf_delete(xlate_actions(&push.ctx,
3736 rule->up.actions, rule->up.n_actions));
3741 static struct subfacet *
3742 subfacet_find__(struct ofproto_dpif *ofproto,
3743 const struct nlattr *key, size_t key_len, uint32_t key_hash,
3744 const struct flow *flow)
3746 struct subfacet *subfacet;
3748 HMAP_FOR_EACH_WITH_HASH (subfacet, hmap_node, key_hash,
3749 &ofproto->subfacets) {
3751 ? (subfacet->key_len == key_len
3752 && !memcmp(key, subfacet->key, key_len))
3753 : flow_equal(flow, &subfacet->facet->flow)) {
3761 /* Searches 'facet' (within 'ofproto') for a subfacet with the specified
3762 * 'key_fitness', 'key', and 'key_len'. Returns the existing subfacet if
3763 * there is one, otherwise creates and returns a new subfacet.
3765 * If the returned subfacet is new, then subfacet->actions will be NULL, in
3766 * which case the caller must populate the actions with
3767 * subfacet_make_actions(). */
3768 static struct subfacet *
3769 subfacet_create(struct facet *facet, enum odp_key_fitness key_fitness,
3770 const struct nlattr *key, size_t key_len, ovs_be16 initial_tci)
3772 struct ofproto_dpif *ofproto = ofproto_dpif_cast(facet->rule->up.ofproto);
3773 uint32_t key_hash = odp_flow_key_hash(key, key_len);
3774 struct subfacet *subfacet;
3776 subfacet = subfacet_find__(ofproto, key, key_len, key_hash, &facet->flow);
3778 if (subfacet->facet == facet) {
3782 /* This shouldn't happen. */
3783 VLOG_ERR_RL(&rl, "subfacet with wrong facet");
3784 subfacet_destroy(subfacet);
3787 subfacet = xzalloc(sizeof *subfacet);
3788 hmap_insert(&ofproto->subfacets, &subfacet->hmap_node, key_hash);
3789 list_push_back(&facet->subfacets, &subfacet->list_node);
3790 subfacet->facet = facet;
3791 subfacet->used = time_msec();
3792 subfacet->key_fitness = key_fitness;
3793 if (key_fitness != ODP_FIT_PERFECT) {
3794 subfacet->key = xmemdup(key, key_len);
3795 subfacet->key_len = key_len;
3797 subfacet->installed = false;
3798 subfacet->initial_tci = initial_tci;
3803 /* Searches 'ofproto' for a subfacet with the given 'key', 'key_len', and
3804 * 'flow'. Returns the subfacet if one exists, otherwise NULL. */
3805 static struct subfacet *
3806 subfacet_find(struct ofproto_dpif *ofproto,
3807 const struct nlattr *key, size_t key_len)
3809 uint32_t key_hash = odp_flow_key_hash(key, key_len);
3810 enum odp_key_fitness fitness;
3813 fitness = odp_flow_key_to_flow(key, key_len, &flow);
3814 if (fitness == ODP_FIT_ERROR) {
3818 return subfacet_find__(ofproto, key, key_len, key_hash, &flow);
3821 /* Uninstalls 'subfacet' from the datapath, if it is installed, removes it from
3822 * its facet within 'ofproto', and frees it. */
3824 subfacet_destroy__(struct subfacet *subfacet)
3826 struct facet *facet = subfacet->facet;
3827 struct ofproto_dpif *ofproto = ofproto_dpif_cast(facet->rule->up.ofproto);
3829 subfacet_uninstall(subfacet);
3830 hmap_remove(&ofproto->subfacets, &subfacet->hmap_node);
3831 list_remove(&subfacet->list_node);
3832 free(subfacet->key);
3833 free(subfacet->actions);
3837 /* Destroys 'subfacet', as with subfacet_destroy__(), and then if this was the
3838 * last remaining subfacet in its facet destroys the facet too. */
3840 subfacet_destroy(struct subfacet *subfacet)
3842 struct facet *facet = subfacet->facet;
3844 if (list_is_singleton(&facet->subfacets)) {
3845 /* facet_remove() needs at least one subfacet (it will remove it). */
3846 facet_remove(facet);
3848 subfacet_destroy__(subfacet);
3852 /* Initializes 'key' with the sequence of OVS_KEY_ATTR_* Netlink attributes
3853 * that can be used to refer to 'subfacet'. The caller must provide 'keybuf'
3854 * for use as temporary storage. */
3856 subfacet_get_key(struct subfacet *subfacet, struct odputil_keybuf *keybuf,
3859 if (!subfacet->key) {
3860 ofpbuf_use_stack(key, keybuf, sizeof *keybuf);
3861 odp_flow_key_from_flow(key, &subfacet->facet->flow);
3863 ofpbuf_use_const(key, subfacet->key, subfacet->key_len);
3867 /* Composes the datapath actions for 'subfacet' based on its rule's actions. */
3869 subfacet_make_actions(struct subfacet *subfacet, const struct ofpbuf *packet)
3871 struct facet *facet = subfacet->facet;
3872 struct rule_dpif *rule = facet->rule;
3873 struct ofproto_dpif *ofproto = ofproto_dpif_cast(rule->up.ofproto);
3874 struct ofpbuf *odp_actions;
3875 struct action_xlate_ctx ctx;
3877 action_xlate_ctx_init(&ctx, ofproto, &facet->flow, subfacet->initial_tci,
3879 odp_actions = xlate_actions(&ctx, rule->up.actions, rule->up.n_actions);
3880 facet->tags = ctx.tags;
3881 facet->may_install = ctx.may_set_up_flow;
3882 facet->has_learn = ctx.has_learn;
3883 facet->has_normal = ctx.has_normal;
3884 facet->has_fin_timeout = ctx.has_fin_timeout;
3885 facet->nf_flow.output_iface = ctx.nf_output_iface;
3886 facet->mirrors = ctx.mirrors;
3888 if (subfacet->actions_len != odp_actions->size
3889 || memcmp(subfacet->actions, odp_actions->data, odp_actions->size)) {
3890 free(subfacet->actions);
3891 subfacet->actions_len = odp_actions->size;
3892 subfacet->actions = xmemdup(odp_actions->data, odp_actions->size);
3895 ofpbuf_delete(odp_actions);
3898 /* Updates 'subfacet''s datapath flow, setting its actions to 'actions_len'
3899 * bytes of actions in 'actions'. If 'stats' is non-null, statistics counters
3900 * in the datapath will be zeroed and 'stats' will be updated with traffic new
3901 * since 'subfacet' was last updated.
3903 * Returns 0 if successful, otherwise a positive errno value. */
3905 subfacet_install(struct subfacet *subfacet,
3906 const struct nlattr *actions, size_t actions_len,
3907 struct dpif_flow_stats *stats)
3909 struct facet *facet = subfacet->facet;
3910 struct ofproto_dpif *ofproto = ofproto_dpif_cast(facet->rule->up.ofproto);
3911 struct odputil_keybuf keybuf;
3912 enum dpif_flow_put_flags flags;
3916 flags = DPIF_FP_CREATE | DPIF_FP_MODIFY;
3918 flags |= DPIF_FP_ZERO_STATS;
3921 subfacet_get_key(subfacet, &keybuf, &key);
3922 ret = dpif_flow_put(ofproto->dpif, flags, key.data, key.size,
3923 actions, actions_len, stats);
3926 subfacet_reset_dp_stats(subfacet, stats);
3932 /* If 'subfacet' is installed in the datapath, uninstalls it. */
3934 subfacet_uninstall(struct subfacet *subfacet)
3936 if (subfacet->installed) {
3937 struct rule_dpif *rule = subfacet->facet->rule;
3938 struct ofproto_dpif *ofproto = ofproto_dpif_cast(rule->up.ofproto);
3939 struct odputil_keybuf keybuf;
3940 struct dpif_flow_stats stats;
3944 subfacet_get_key(subfacet, &keybuf, &key);
3945 error = dpif_flow_del(ofproto->dpif, key.data, key.size, &stats);
3946 subfacet_reset_dp_stats(subfacet, &stats);
3948 subfacet_update_stats(subfacet, &stats);
3950 subfacet->installed = false;
3952 assert(subfacet->dp_packet_count == 0);
3953 assert(subfacet->dp_byte_count == 0);
3957 /* Resets 'subfacet''s datapath statistics counters. This should be called
3958 * when 'subfacet''s statistics are cleared in the datapath. If 'stats' is
3959 * non-null, it should contain the statistics returned by dpif when 'subfacet'
3960 * was reset in the datapath. 'stats' will be modified to include only
3961 * statistics new since 'subfacet' was last updated. */
3963 subfacet_reset_dp_stats(struct subfacet *subfacet,
3964 struct dpif_flow_stats *stats)
3967 && subfacet->dp_packet_count <= stats->n_packets
3968 && subfacet->dp_byte_count <= stats->n_bytes) {
3969 stats->n_packets -= subfacet->dp_packet_count;
3970 stats->n_bytes -= subfacet->dp_byte_count;
3973 subfacet->dp_packet_count = 0;
3974 subfacet->dp_byte_count = 0;
3977 /* Updates 'subfacet''s used time. The caller is responsible for calling
3978 * facet_push_stats() to update the flows which 'subfacet' resubmits into. */
3980 subfacet_update_time(struct subfacet *subfacet, long long int used)
3982 if (used > subfacet->used) {
3983 subfacet->used = used;
3984 facet_update_time(subfacet->facet, used);
3988 /* Folds the statistics from 'stats' into the counters in 'subfacet'.
3990 * Because of the meaning of a subfacet's counters, it only makes sense to do
3991 * this if 'stats' are not tracked in the datapath, that is, if 'stats'
3992 * represents a packet that was sent by hand or if it represents statistics
3993 * that have been cleared out of the datapath. */
3995 subfacet_update_stats(struct subfacet *subfacet,
3996 const struct dpif_flow_stats *stats)
3998 if (stats->n_packets || stats->used > subfacet->used) {
3999 struct facet *facet = subfacet->facet;
4001 subfacet_update_time(subfacet, stats->used);
4002 facet->packet_count += stats->n_packets;
4003 facet->byte_count += stats->n_bytes;
4004 facet->tcp_flags |= stats->tcp_flags;
4005 facet_push_stats(facet);
4006 netflow_flow_update_flags(&facet->nf_flow, stats->tcp_flags);
4012 static struct rule_dpif *
4013 rule_dpif_lookup(struct ofproto_dpif *ofproto, const struct flow *flow,
4016 struct cls_rule *cls_rule;
4017 struct classifier *cls;
4019 if (table_id >= N_TABLES) {
4023 cls = &ofproto->up.tables[table_id].cls;
4024 if (flow->nw_frag & FLOW_NW_FRAG_ANY
4025 && ofproto->up.frag_handling == OFPC_FRAG_NORMAL) {
4026 /* For OFPC_NORMAL frag_handling, we must pretend that transport ports
4027 * are unavailable. */
4028 struct flow ofpc_normal_flow = *flow;
4029 ofpc_normal_flow.tp_src = htons(0);
4030 ofpc_normal_flow.tp_dst = htons(0);
4031 cls_rule = classifier_lookup(cls, &ofpc_normal_flow);
4033 cls_rule = classifier_lookup(cls, flow);
4035 return rule_dpif_cast(rule_from_cls_rule(cls_rule));
4039 complete_operation(struct rule_dpif *rule)
4041 struct ofproto_dpif *ofproto = ofproto_dpif_cast(rule->up.ofproto);
4043 rule_invalidate(rule);
4045 struct dpif_completion *c = xmalloc(sizeof *c);
4046 c->op = rule->up.pending;
4047 list_push_back(&ofproto->completions, &c->list_node);
4049 ofoperation_complete(rule->up.pending, 0);
4053 static struct rule *
4056 struct rule_dpif *rule = xmalloc(sizeof *rule);
4061 rule_dealloc(struct rule *rule_)
4063 struct rule_dpif *rule = rule_dpif_cast(rule_);
4068 rule_construct(struct rule *rule_)
4070 struct rule_dpif *rule = rule_dpif_cast(rule_);
4071 struct ofproto_dpif *ofproto = ofproto_dpif_cast(rule->up.ofproto);
4072 struct rule_dpif *victim;
4076 error = validate_actions(rule->up.actions, rule->up.n_actions,
4077 &rule->up.cr.flow, ofproto->max_ports);
4082 rule->packet_count = 0;
4083 rule->byte_count = 0;
4085 victim = rule_dpif_cast(ofoperation_get_victim(rule->up.pending));
4086 if (victim && !list_is_empty(&victim->facets)) {
4087 struct facet *facet;
4089 rule->facets = victim->facets;
4090 list_moved(&rule->facets);
4091 LIST_FOR_EACH (facet, list_node, &rule->facets) {
4092 /* XXX: We're only clearing our local counters here. It's possible
4093 * that quite a few packets are unaccounted for in the datapath
4094 * statistics. These will be accounted to the new rule instead of
4095 * cleared as required. This could be fixed by clearing out the
4096 * datapath statistics for this facet, but currently it doesn't
4098 facet_reset_counters(facet);
4102 /* Must avoid list_moved() in this case. */
4103 list_init(&rule->facets);
4106 table_id = rule->up.table_id;
4107 rule->tag = (victim ? victim->tag
4109 : rule_calculate_tag(&rule->up.cr.flow, &rule->up.cr.wc,
4110 ofproto->tables[table_id].basis));
4112 complete_operation(rule);
4117 rule_destruct(struct rule *rule_)
4119 struct rule_dpif *rule = rule_dpif_cast(rule_);
4120 struct facet *facet, *next_facet;
4122 LIST_FOR_EACH_SAFE (facet, next_facet, list_node, &rule->facets) {
4123 facet_revalidate(facet);
4126 complete_operation(rule);
4130 rule_get_stats(struct rule *rule_, uint64_t *packets, uint64_t *bytes)
4132 struct rule_dpif *rule = rule_dpif_cast(rule_);
4133 struct facet *facet;
4135 /* Start from historical data for 'rule' itself that are no longer tracked
4136 * in facets. This counts, for example, facets that have expired. */
4137 *packets = rule->packet_count;
4138 *bytes = rule->byte_count;
4140 /* Add any statistics that are tracked by facets. This includes
4141 * statistical data recently updated by ofproto_update_stats() as well as
4142 * stats for packets that were executed "by hand" via dpif_execute(). */
4143 LIST_FOR_EACH (facet, list_node, &rule->facets) {
4144 *packets += facet->packet_count;
4145 *bytes += facet->byte_count;
4150 rule_execute(struct rule *rule_, const struct flow *flow,
4151 struct ofpbuf *packet)
4153 struct rule_dpif *rule = rule_dpif_cast(rule_);
4154 struct ofproto_dpif *ofproto = ofproto_dpif_cast(rule->up.ofproto);
4155 struct action_xlate_ctx ctx;
4156 struct ofpbuf *odp_actions;
4159 action_xlate_ctx_init(&ctx, ofproto, flow, flow->vlan_tci,
4160 rule, packet_get_tcp_flags(packet, flow), packet);
4161 odp_actions = xlate_actions(&ctx, rule->up.actions, rule->up.n_actions);
4162 size = packet->size;
4163 if (execute_odp_actions(ofproto, flow, odp_actions->data,
4164 odp_actions->size, packet)) {
4165 rule->packet_count++;
4166 rule->byte_count += size;
4167 flow_push_stats(rule, flow, 1, size, time_msec());
4169 ofpbuf_delete(odp_actions);
4175 rule_modify_actions(struct rule *rule_)
4177 struct rule_dpif *rule = rule_dpif_cast(rule_);
4178 struct ofproto_dpif *ofproto = ofproto_dpif_cast(rule->up.ofproto);
4181 error = validate_actions(rule->up.actions, rule->up.n_actions,
4182 &rule->up.cr.flow, ofproto->max_ports);
4184 ofoperation_complete(rule->up.pending, error);
4188 complete_operation(rule);
4191 /* Sends 'packet' out 'ofport'.
4192 * May modify 'packet'.
4193 * Returns 0 if successful, otherwise a positive errno value. */
4195 send_packet(const struct ofport_dpif *ofport, struct ofpbuf *packet)
4197 const struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport->up.ofproto);
4198 struct ofpbuf key, odp_actions;
4199 struct odputil_keybuf keybuf;
4204 flow_extract((struct ofpbuf *) packet, 0, 0, 0, &flow);
4205 odp_port = vsp_realdev_to_vlandev(ofproto, ofport->odp_port,
4207 if (odp_port != ofport->odp_port) {
4208 eth_pop_vlan(packet);
4209 flow.vlan_tci = htons(0);
4212 ofpbuf_use_stack(&key, &keybuf, sizeof keybuf);
4213 odp_flow_key_from_flow(&key, &flow);
4215 ofpbuf_init(&odp_actions, 32);
4216 compose_sflow_action(ofproto, &odp_actions, &flow, odp_port);
4218 nl_msg_put_u32(&odp_actions, OVS_ACTION_ATTR_OUTPUT, odp_port);
4219 error = dpif_execute(ofproto->dpif,
4221 odp_actions.data, odp_actions.size,
4223 ofpbuf_uninit(&odp_actions);
4226 VLOG_WARN_RL(&rl, "%s: failed to send packet on port %"PRIu32" (%s)",
4227 ofproto->up.name, odp_port, strerror(error));
4229 ofproto_update_local_port_stats(ofport->up.ofproto, packet->size, 0);
4233 /* OpenFlow to datapath action translation. */
4235 static void do_xlate_actions(const union ofp_action *in, size_t n_in,
4236 struct action_xlate_ctx *ctx);
4237 static void xlate_normal(struct action_xlate_ctx *);
4240 put_userspace_action(const struct ofproto_dpif *ofproto,
4241 struct ofpbuf *odp_actions,
4242 const struct flow *flow,
4243 const struct user_action_cookie *cookie)
4247 pid = dpif_port_get_pid(ofproto->dpif,
4248 ofp_port_to_odp_port(flow->in_port));
4250 return odp_put_userspace_action(pid, cookie, odp_actions);
4253 /* Compose SAMPLE action for sFlow. */
4255 compose_sflow_action(const struct ofproto_dpif *ofproto,
4256 struct ofpbuf *odp_actions,
4257 const struct flow *flow,
4260 uint32_t port_ifindex;
4261 uint32_t probability;
4262 struct user_action_cookie cookie;
4263 size_t sample_offset, actions_offset;
4264 int cookie_offset, n_output;
4266 if (!ofproto->sflow || flow->in_port == OFPP_NONE) {
4270 if (odp_port == OVSP_NONE) {
4274 port_ifindex = dpif_sflow_odp_port_to_ifindex(ofproto->sflow, odp_port);
4278 sample_offset = nl_msg_start_nested(odp_actions, OVS_ACTION_ATTR_SAMPLE);
4280 /* Number of packets out of UINT_MAX to sample. */
4281 probability = dpif_sflow_get_probability(ofproto->sflow);
4282 nl_msg_put_u32(odp_actions, OVS_SAMPLE_ATTR_PROBABILITY, probability);
4284 actions_offset = nl_msg_start_nested(odp_actions, OVS_SAMPLE_ATTR_ACTIONS);
4286 cookie.type = USER_ACTION_COOKIE_SFLOW;
4287 cookie.data = port_ifindex;
4288 cookie.n_output = n_output;
4289 cookie.vlan_tci = 0;
4290 cookie_offset = put_userspace_action(ofproto, odp_actions, flow, &cookie);
4292 nl_msg_end_nested(odp_actions, actions_offset);
4293 nl_msg_end_nested(odp_actions, sample_offset);
4294 return cookie_offset;
4297 /* SAMPLE action must be first action in any given list of actions.
4298 * At this point we do not have all information required to build it. So try to
4299 * build sample action as complete as possible. */
4301 add_sflow_action(struct action_xlate_ctx *ctx)
4303 ctx->user_cookie_offset = compose_sflow_action(ctx->ofproto,
4305 &ctx->flow, OVSP_NONE);
4306 ctx->sflow_odp_port = 0;
4307 ctx->sflow_n_outputs = 0;
4310 /* Fix SAMPLE action according to data collected while composing ODP actions.
4311 * We need to fix SAMPLE actions OVS_SAMPLE_ATTR_ACTIONS attribute, i.e. nested
4312 * USERSPACE action's user-cookie which is required for sflow. */
4314 fix_sflow_action(struct action_xlate_ctx *ctx)
4316 const struct flow *base = &ctx->base_flow;
4317 struct user_action_cookie *cookie;
4319 if (!ctx->user_cookie_offset) {
4323 cookie = ofpbuf_at(ctx->odp_actions, ctx->user_cookie_offset,
4325 assert(cookie != NULL);
4326 assert(cookie->type == USER_ACTION_COOKIE_SFLOW);
4328 if (ctx->sflow_n_outputs) {
4329 cookie->data = dpif_sflow_odp_port_to_ifindex(ctx->ofproto->sflow,
4330 ctx->sflow_odp_port);
4332 if (ctx->sflow_n_outputs >= 255) {
4333 cookie->n_output = 255;
4335 cookie->n_output = ctx->sflow_n_outputs;
4337 cookie->vlan_tci = base->vlan_tci;
4341 compose_output_action__(struct action_xlate_ctx *ctx, uint16_t ofp_port,
4344 const struct ofport_dpif *ofport = get_ofp_port(ctx->ofproto, ofp_port);
4345 uint16_t odp_port = ofp_port_to_odp_port(ofp_port);
4346 ovs_be16 flow_vlan_tci = ctx->flow.vlan_tci;
4347 uint8_t flow_nw_tos = ctx->flow.nw_tos;
4351 struct priority_to_dscp *pdscp;
4353 if (ofport->up.pp.config & OFPUTIL_PC_NO_FWD
4354 || (check_stp && !stp_forward_in_state(ofport->stp_state))) {
4358 pdscp = get_priority(ofport, ctx->flow.skb_priority);
4360 ctx->flow.nw_tos &= ~IP_DSCP_MASK;
4361 ctx->flow.nw_tos |= pdscp->dscp;
4364 /* We may not have an ofport record for this port, but it doesn't hurt
4365 * to allow forwarding to it anyhow. Maybe such a port will appear
4366 * later and we're pre-populating the flow table. */
4369 out_port = vsp_realdev_to_vlandev(ctx->ofproto, odp_port,
4370 ctx->flow.vlan_tci);
4371 if (out_port != odp_port) {
4372 ctx->flow.vlan_tci = htons(0);
4374 commit_odp_actions(&ctx->flow, &ctx->base_flow, ctx->odp_actions);
4375 nl_msg_put_u32(ctx->odp_actions, OVS_ACTION_ATTR_OUTPUT, out_port);
4377 ctx->sflow_odp_port = odp_port;
4378 ctx->sflow_n_outputs++;
4379 ctx->nf_output_iface = ofp_port;
4380 ctx->flow.vlan_tci = flow_vlan_tci;
4381 ctx->flow.nw_tos = flow_nw_tos;
4385 compose_output_action(struct action_xlate_ctx *ctx, uint16_t ofp_port)
4387 compose_output_action__(ctx, ofp_port, true);
4391 xlate_table_action(struct action_xlate_ctx *ctx,
4392 uint16_t in_port, uint8_t table_id)
4394 if (ctx->recurse < MAX_RESUBMIT_RECURSION) {
4395 struct ofproto_dpif *ofproto = ctx->ofproto;
4396 struct rule_dpif *rule;
4397 uint16_t old_in_port;
4398 uint8_t old_table_id;
4400 old_table_id = ctx->table_id;
4401 ctx->table_id = table_id;
4403 /* Look up a flow with 'in_port' as the input port. */
4404 old_in_port = ctx->flow.in_port;
4405 ctx->flow.in_port = in_port;
4406 rule = rule_dpif_lookup(ofproto, &ctx->flow, table_id);
4409 if (table_id > 0 && table_id < N_TABLES) {
4410 struct table_dpif *table = &ofproto->tables[table_id];
4411 if (table->other_table) {
4414 : rule_calculate_tag(&ctx->flow,
4415 &table->other_table->wc,
4420 /* Restore the original input port. Otherwise OFPP_NORMAL and
4421 * OFPP_IN_PORT will have surprising behavior. */
4422 ctx->flow.in_port = old_in_port;
4424 if (ctx->resubmit_hook) {
4425 ctx->resubmit_hook(ctx, rule);
4429 struct rule_dpif *old_rule = ctx->rule;
4433 do_xlate_actions(rule->up.actions, rule->up.n_actions, ctx);
4434 ctx->rule = old_rule;
4438 ctx->table_id = old_table_id;
4440 static struct vlog_rate_limit recurse_rl = VLOG_RATE_LIMIT_INIT(1, 1);
4442 VLOG_ERR_RL(&recurse_rl, "resubmit actions recursed over %d times",
4443 MAX_RESUBMIT_RECURSION);
4444 ctx->max_resubmit_trigger = true;
4449 xlate_resubmit_table(struct action_xlate_ctx *ctx,
4450 const struct nx_action_resubmit *nar)
4455 in_port = (nar->in_port == htons(OFPP_IN_PORT)
4457 : ntohs(nar->in_port));
4458 table_id = nar->table == 255 ? ctx->table_id : nar->table;
4460 xlate_table_action(ctx, in_port, table_id);
4464 flood_packets(struct action_xlate_ctx *ctx, bool all)
4466 struct ofport_dpif *ofport;
4468 HMAP_FOR_EACH (ofport, up.hmap_node, &ctx->ofproto->up.ports) {
4469 uint16_t ofp_port = ofport->up.ofp_port;
4471 if (ofp_port == ctx->flow.in_port) {
4476 compose_output_action__(ctx, ofp_port, false);
4477 } else if (!(ofport->up.pp.config & OFPUTIL_PC_NO_FLOOD)) {
4478 compose_output_action(ctx, ofp_port);
4482 ctx->nf_output_iface = NF_OUT_FLOOD;
4486 execute_controller_action(struct action_xlate_ctx *ctx, int len,
4487 enum ofp_packet_in_reason reason,
4488 uint16_t controller_id)
4490 struct ofputil_packet_in pin;
4491 struct ofpbuf *packet;
4493 ctx->may_set_up_flow = false;
4498 packet = ofpbuf_clone(ctx->packet);
4500 if (packet->l2 && packet->l3) {
4501 struct eth_header *eh;
4503 eth_pop_vlan(packet);
4505 assert(eh->eth_type == ctx->flow.dl_type);
4506 memcpy(eh->eth_src, ctx->flow.dl_src, sizeof eh->eth_src);
4507 memcpy(eh->eth_dst, ctx->flow.dl_dst, sizeof eh->eth_dst);
4509 if (ctx->flow.vlan_tci & htons(VLAN_CFI)) {
4510 eth_push_vlan(packet, ctx->flow.vlan_tci);
4514 if (ctx->flow.dl_type == htons(ETH_TYPE_IP)) {
4515 packet_set_ipv4(packet, ctx->flow.nw_src, ctx->flow.nw_dst,
4516 ctx->flow.nw_tos, ctx->flow.nw_ttl);
4520 if (ctx->flow.nw_proto == IPPROTO_TCP) {
4521 packet_set_tcp_port(packet, ctx->flow.tp_src,
4523 } else if (ctx->flow.nw_proto == IPPROTO_UDP) {
4524 packet_set_udp_port(packet, ctx->flow.tp_src,
4531 pin.packet = packet->data;
4532 pin.packet_len = packet->size;
4533 pin.reason = reason;
4534 pin.controller_id = controller_id;
4535 pin.table_id = ctx->table_id;
4536 pin.cookie = ctx->rule ? ctx->rule->up.flow_cookie : 0;
4540 pin.total_len = packet->size;
4541 flow_get_metadata(&ctx->flow, &pin.fmd);
4543 connmgr_send_packet_in(ctx->ofproto->up.connmgr, &pin, &ctx->flow);
4544 ofpbuf_delete(packet);
4548 compose_dec_ttl(struct action_xlate_ctx *ctx)
4550 if (ctx->flow.dl_type != htons(ETH_TYPE_IP) &&
4551 ctx->flow.dl_type != htons(ETH_TYPE_IPV6)) {
4555 if (ctx->flow.nw_ttl > 1) {
4559 execute_controller_action(ctx, UINT16_MAX, OFPR_INVALID_TTL, 0);
4561 /* Stop processing for current table. */
4567 xlate_output_action__(struct action_xlate_ctx *ctx,
4568 uint16_t port, uint16_t max_len)
4570 uint16_t prev_nf_output_iface = ctx->nf_output_iface;
4572 ctx->nf_output_iface = NF_OUT_DROP;
4576 compose_output_action(ctx, ctx->flow.in_port);
4579 xlate_table_action(ctx, ctx->flow.in_port, ctx->table_id);
4585 flood_packets(ctx, false);
4588 flood_packets(ctx, true);
4590 case OFPP_CONTROLLER:
4591 execute_controller_action(ctx, max_len, OFPR_ACTION, 0);
4597 if (port != ctx->flow.in_port) {
4598 compose_output_action(ctx, port);
4603 if (prev_nf_output_iface == NF_OUT_FLOOD) {
4604 ctx->nf_output_iface = NF_OUT_FLOOD;
4605 } else if (ctx->nf_output_iface == NF_OUT_DROP) {
4606 ctx->nf_output_iface = prev_nf_output_iface;
4607 } else if (prev_nf_output_iface != NF_OUT_DROP &&
4608 ctx->nf_output_iface != NF_OUT_FLOOD) {
4609 ctx->nf_output_iface = NF_OUT_MULTI;
4614 xlate_output_reg_action(struct action_xlate_ctx *ctx,
4615 const struct nx_action_output_reg *naor)
4617 struct mf_subfield src;
4620 nxm_decode(&src, naor->src, naor->ofs_nbits);
4621 ofp_port = mf_get_subfield(&src, &ctx->flow);
4623 if (ofp_port <= UINT16_MAX) {
4624 xlate_output_action__(ctx, ofp_port, ntohs(naor->max_len));
4629 xlate_output_action(struct action_xlate_ctx *ctx,
4630 const struct ofp_action_output *oao)
4632 xlate_output_action__(ctx, ntohs(oao->port), ntohs(oao->max_len));
4636 xlate_enqueue_action(struct action_xlate_ctx *ctx,
4637 const struct ofp_action_enqueue *oae)
4640 uint32_t flow_priority, priority;
4643 error = dpif_queue_to_priority(ctx->ofproto->dpif, ntohl(oae->queue_id),
4646 /* Fall back to ordinary output action. */
4647 xlate_output_action__(ctx, ntohs(oae->port), 0);
4651 /* Figure out datapath output port. */
4652 ofp_port = ntohs(oae->port);
4653 if (ofp_port == OFPP_IN_PORT) {
4654 ofp_port = ctx->flow.in_port;
4655 } else if (ofp_port == ctx->flow.in_port) {
4659 /* Add datapath actions. */
4660 flow_priority = ctx->flow.skb_priority;
4661 ctx->flow.skb_priority = priority;
4662 compose_output_action(ctx, ofp_port);
4663 ctx->flow.skb_priority = flow_priority;
4665 /* Update NetFlow output port. */
4666 if (ctx->nf_output_iface == NF_OUT_DROP) {
4667 ctx->nf_output_iface = ofp_port;
4668 } else if (ctx->nf_output_iface != NF_OUT_FLOOD) {
4669 ctx->nf_output_iface = NF_OUT_MULTI;
4674 xlate_set_queue_action(struct action_xlate_ctx *ctx,
4675 const struct nx_action_set_queue *nasq)
4680 error = dpif_queue_to_priority(ctx->ofproto->dpif, ntohl(nasq->queue_id),
4683 /* Couldn't translate queue to a priority, so ignore. A warning
4684 * has already been logged. */
4688 ctx->flow.skb_priority = priority;
4691 struct xlate_reg_state {
4697 xlate_autopath(struct action_xlate_ctx *ctx,
4698 const struct nx_action_autopath *naa)
4700 uint16_t ofp_port = ntohl(naa->id);
4701 struct ofport_dpif *port = get_ofp_port(ctx->ofproto, ofp_port);
4703 if (!port || !port->bundle) {
4704 ofp_port = OFPP_NONE;
4705 } else if (port->bundle->bond) {
4706 /* Autopath does not support VLAN hashing. */
4707 struct ofport_dpif *slave = bond_choose_output_slave(
4708 port->bundle->bond, &ctx->flow, 0, &ctx->tags);
4710 ofp_port = slave->up.ofp_port;
4713 autopath_execute(naa, &ctx->flow, ofp_port);
4717 slave_enabled_cb(uint16_t ofp_port, void *ofproto_)
4719 struct ofproto_dpif *ofproto = ofproto_;
4720 struct ofport_dpif *port;
4730 case OFPP_CONTROLLER: /* Not supported by the bundle action. */
4733 port = get_ofp_port(ofproto, ofp_port);
4734 return port ? port->may_enable : false;
4739 xlate_learn_action(struct action_xlate_ctx *ctx,
4740 const struct nx_action_learn *learn)
4742 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(5, 1);
4743 struct ofputil_flow_mod fm;
4746 learn_execute(learn, &ctx->flow, &fm);
4748 error = ofproto_flow_mod(&ctx->ofproto->up, &fm);
4749 if (error && !VLOG_DROP_WARN(&rl)) {
4750 VLOG_WARN("learning action failed to modify flow table (%s)",
4751 ofperr_get_name(error));
4757 /* Reduces '*timeout' to no more than 'max'. A value of zero in either case
4758 * means "infinite". */
4760 reduce_timeout(uint16_t max, uint16_t *timeout)
4762 if (max && (!*timeout || *timeout > max)) {
4768 xlate_fin_timeout(struct action_xlate_ctx *ctx,
4769 const struct nx_action_fin_timeout *naft)
4771 if (ctx->tcp_flags & (TCP_FIN | TCP_RST) && ctx->rule) {
4772 struct rule_dpif *rule = ctx->rule;
4774 reduce_timeout(ntohs(naft->fin_idle_timeout), &rule->up.idle_timeout);
4775 reduce_timeout(ntohs(naft->fin_hard_timeout), &rule->up.hard_timeout);
4780 may_receive(const struct ofport_dpif *port, struct action_xlate_ctx *ctx)
4782 if (port->up.pp.config & (eth_addr_equals(ctx->flow.dl_dst, eth_addr_stp)
4783 ? OFPUTIL_PC_NO_RECV_STP
4784 : OFPUTIL_PC_NO_RECV)) {
4788 /* Only drop packets here if both forwarding and learning are
4789 * disabled. If just learning is enabled, we need to have
4790 * OFPP_NORMAL and the learning action have a look at the packet
4791 * before we can drop it. */
4792 if (!stp_forward_in_state(port->stp_state)
4793 && !stp_learn_in_state(port->stp_state)) {
4801 do_xlate_actions(const union ofp_action *in, size_t n_in,
4802 struct action_xlate_ctx *ctx)
4804 const struct ofport_dpif *port;
4805 const union ofp_action *ia;
4806 bool was_evictable = true;
4809 port = get_ofp_port(ctx->ofproto, ctx->flow.in_port);
4810 if (port && !may_receive(port, ctx)) {
4811 /* Drop this flow. */
4816 /* Don't let the rule we're working on get evicted underneath us. */
4817 was_evictable = ctx->rule->up.evictable;
4818 ctx->rule->up.evictable = false;
4820 OFPUTIL_ACTION_FOR_EACH_UNSAFE (ia, left, in, n_in) {
4821 const struct ofp_action_dl_addr *oada;
4822 const struct nx_action_resubmit *nar;
4823 const struct nx_action_set_tunnel *nast;
4824 const struct nx_action_set_queue *nasq;
4825 const struct nx_action_multipath *nam;
4826 const struct nx_action_autopath *naa;
4827 const struct nx_action_bundle *nab;
4828 const struct nx_action_output_reg *naor;
4829 const struct nx_action_controller *nac;
4830 enum ofputil_action_code code;
4837 code = ofputil_decode_action_unsafe(ia);
4839 case OFPUTIL_OFPAT10_OUTPUT:
4840 xlate_output_action(ctx, &ia->output);
4843 case OFPUTIL_OFPAT10_SET_VLAN_VID:
4844 ctx->flow.vlan_tci &= ~htons(VLAN_VID_MASK);
4845 ctx->flow.vlan_tci |= ia->vlan_vid.vlan_vid | htons(VLAN_CFI);
4848 case OFPUTIL_OFPAT10_SET_VLAN_PCP:
4849 ctx->flow.vlan_tci &= ~htons(VLAN_PCP_MASK);
4850 ctx->flow.vlan_tci |= htons(
4851 (ia->vlan_pcp.vlan_pcp << VLAN_PCP_SHIFT) | VLAN_CFI);
4854 case OFPUTIL_OFPAT10_STRIP_VLAN:
4855 ctx->flow.vlan_tci = htons(0);
4858 case OFPUTIL_OFPAT10_SET_DL_SRC:
4859 oada = ((struct ofp_action_dl_addr *) ia);
4860 memcpy(ctx->flow.dl_src, oada->dl_addr, ETH_ADDR_LEN);
4863 case OFPUTIL_OFPAT10_SET_DL_DST:
4864 oada = ((struct ofp_action_dl_addr *) ia);
4865 memcpy(ctx->flow.dl_dst, oada->dl_addr, ETH_ADDR_LEN);
4868 case OFPUTIL_OFPAT10_SET_NW_SRC:
4869 ctx->flow.nw_src = ia->nw_addr.nw_addr;
4872 case OFPUTIL_OFPAT10_SET_NW_DST:
4873 ctx->flow.nw_dst = ia->nw_addr.nw_addr;
4876 case OFPUTIL_OFPAT10_SET_NW_TOS:
4877 /* OpenFlow 1.0 only supports IPv4. */
4878 if (ctx->flow.dl_type == htons(ETH_TYPE_IP)) {
4879 ctx->flow.nw_tos &= ~IP_DSCP_MASK;
4880 ctx->flow.nw_tos |= ia->nw_tos.nw_tos & IP_DSCP_MASK;
4884 case OFPUTIL_OFPAT10_SET_TP_SRC:
4885 ctx->flow.tp_src = ia->tp_port.tp_port;
4888 case OFPUTIL_OFPAT10_SET_TP_DST:
4889 ctx->flow.tp_dst = ia->tp_port.tp_port;
4892 case OFPUTIL_OFPAT10_ENQUEUE:
4893 xlate_enqueue_action(ctx, (const struct ofp_action_enqueue *) ia);
4896 case OFPUTIL_NXAST_RESUBMIT:
4897 nar = (const struct nx_action_resubmit *) ia;
4898 xlate_table_action(ctx, ntohs(nar->in_port), ctx->table_id);
4901 case OFPUTIL_NXAST_RESUBMIT_TABLE:
4902 xlate_resubmit_table(ctx, (const struct nx_action_resubmit *) ia);
4905 case OFPUTIL_NXAST_SET_TUNNEL:
4906 nast = (const struct nx_action_set_tunnel *) ia;
4907 tun_id = htonll(ntohl(nast->tun_id));
4908 ctx->flow.tun_id = tun_id;
4911 case OFPUTIL_NXAST_SET_QUEUE:
4912 nasq = (const struct nx_action_set_queue *) ia;
4913 xlate_set_queue_action(ctx, nasq);
4916 case OFPUTIL_NXAST_POP_QUEUE:
4917 ctx->flow.skb_priority = ctx->orig_skb_priority;
4920 case OFPUTIL_NXAST_REG_MOVE:
4921 nxm_execute_reg_move((const struct nx_action_reg_move *) ia,
4925 case OFPUTIL_NXAST_REG_LOAD:
4926 nxm_execute_reg_load((const struct nx_action_reg_load *) ia,
4930 case OFPUTIL_NXAST_NOTE:
4931 /* Nothing to do. */
4934 case OFPUTIL_NXAST_SET_TUNNEL64:
4935 tun_id = ((const struct nx_action_set_tunnel64 *) ia)->tun_id;
4936 ctx->flow.tun_id = tun_id;
4939 case OFPUTIL_NXAST_MULTIPATH:
4940 nam = (const struct nx_action_multipath *) ia;
4941 multipath_execute(nam, &ctx->flow);
4944 case OFPUTIL_NXAST_AUTOPATH:
4945 naa = (const struct nx_action_autopath *) ia;
4946 xlate_autopath(ctx, naa);
4949 case OFPUTIL_NXAST_BUNDLE:
4950 ctx->ofproto->has_bundle_action = true;
4951 nab = (const struct nx_action_bundle *) ia;
4952 xlate_output_action__(ctx, bundle_execute(nab, &ctx->flow,
4957 case OFPUTIL_NXAST_BUNDLE_LOAD:
4958 ctx->ofproto->has_bundle_action = true;
4959 nab = (const struct nx_action_bundle *) ia;
4960 bundle_execute_load(nab, &ctx->flow, slave_enabled_cb,
4964 case OFPUTIL_NXAST_OUTPUT_REG:
4965 naor = (const struct nx_action_output_reg *) ia;
4966 xlate_output_reg_action(ctx, naor);
4969 case OFPUTIL_NXAST_LEARN:
4970 ctx->has_learn = true;
4971 if (ctx->may_learn) {
4972 xlate_learn_action(ctx, (const struct nx_action_learn *) ia);
4976 case OFPUTIL_NXAST_DEC_TTL:
4977 if (compose_dec_ttl(ctx)) {
4982 case OFPUTIL_NXAST_EXIT:
4986 case OFPUTIL_NXAST_FIN_TIMEOUT:
4987 ctx->has_fin_timeout = true;
4988 xlate_fin_timeout(ctx, (const struct nx_action_fin_timeout *) ia);
4991 case OFPUTIL_NXAST_CONTROLLER:
4992 nac = (const struct nx_action_controller *) ia;
4993 execute_controller_action(ctx, ntohs(nac->max_len), nac->reason,
4994 ntohs(nac->controller_id));
5000 /* We've let OFPP_NORMAL and the learning action look at the packet,
5001 * so drop it now if forwarding is disabled. */
5002 if (port && !stp_forward_in_state(port->stp_state)) {
5003 ofpbuf_clear(ctx->odp_actions);
5004 add_sflow_action(ctx);
5007 ctx->rule->up.evictable = was_evictable;
5012 action_xlate_ctx_init(struct action_xlate_ctx *ctx,
5013 struct ofproto_dpif *ofproto, const struct flow *flow,
5014 ovs_be16 initial_tci, struct rule_dpif *rule,
5015 uint8_t tcp_flags, const struct ofpbuf *packet)
5017 ctx->ofproto = ofproto;
5019 ctx->base_flow = ctx->flow;
5020 ctx->base_flow.tun_id = 0;
5021 ctx->base_flow.vlan_tci = initial_tci;
5023 ctx->packet = packet;
5024 ctx->may_learn = packet != NULL;
5025 ctx->tcp_flags = tcp_flags;
5026 ctx->resubmit_hook = NULL;
5029 static struct ofpbuf *
5030 xlate_actions(struct action_xlate_ctx *ctx,
5031 const union ofp_action *in, size_t n_in)
5033 struct flow orig_flow = ctx->flow;
5035 COVERAGE_INC(ofproto_dpif_xlate);
5037 ctx->odp_actions = ofpbuf_new(512);
5038 ofpbuf_reserve(ctx->odp_actions, NL_A_U32_SIZE);
5040 ctx->may_set_up_flow = true;
5041 ctx->has_learn = false;
5042 ctx->has_normal = false;
5043 ctx->has_fin_timeout = false;
5044 ctx->nf_output_iface = NF_OUT_DROP;
5047 ctx->max_resubmit_trigger = false;
5048 ctx->orig_skb_priority = ctx->flow.skb_priority;
5052 if (ctx->flow.nw_frag & FLOW_NW_FRAG_ANY) {
5053 switch (ctx->ofproto->up.frag_handling) {
5054 case OFPC_FRAG_NORMAL:
5055 /* We must pretend that transport ports are unavailable. */
5056 ctx->flow.tp_src = ctx->base_flow.tp_src = htons(0);
5057 ctx->flow.tp_dst = ctx->base_flow.tp_dst = htons(0);
5060 case OFPC_FRAG_DROP:
5061 return ctx->odp_actions;
5063 case OFPC_FRAG_REASM:
5066 case OFPC_FRAG_NX_MATCH:
5067 /* Nothing to do. */
5070 case OFPC_INVALID_TTL_TO_CONTROLLER:
5075 if (process_special(ctx->ofproto, &ctx->flow, ctx->packet)) {
5076 ctx->may_set_up_flow = false;
5077 return ctx->odp_actions;
5079 static struct vlog_rate_limit trace_rl = VLOG_RATE_LIMIT_INIT(1, 1);
5080 struct flow original_flow = ctx->flow;
5081 ovs_be16 initial_tci = ctx->base_flow.vlan_tci;
5083 add_sflow_action(ctx);
5084 do_xlate_actions(in, n_in, ctx);
5086 if (ctx->max_resubmit_trigger && !ctx->resubmit_hook
5087 && !VLOG_DROP_ERR(&trace_rl)) {
5088 struct ds ds = DS_EMPTY_INITIALIZER;
5090 ofproto_trace(ctx->ofproto, &original_flow, ctx->packet,
5092 VLOG_ERR("Trace triggered by excessive resubmit recursion:\n%s",
5097 if (!connmgr_may_set_up_flow(ctx->ofproto->up.connmgr, &ctx->flow,
5098 ctx->odp_actions->data,
5099 ctx->odp_actions->size)) {
5100 ctx->may_set_up_flow = false;
5102 && connmgr_msg_in_hook(ctx->ofproto->up.connmgr, &ctx->flow,
5104 compose_output_action(ctx, OFPP_LOCAL);
5107 add_mirror_actions(ctx, &orig_flow);
5108 fix_sflow_action(ctx);
5111 return ctx->odp_actions;
5114 /* OFPP_NORMAL implementation. */
5116 static struct ofport_dpif *ofbundle_get_a_port(const struct ofbundle *);
5118 /* Given 'vid', the VID obtained from the 802.1Q header that was received as
5119 * part of a packet (specify 0 if there was no 802.1Q header), and 'in_bundle',
5120 * the bundle on which the packet was received, returns the VLAN to which the
5123 * Both 'vid' and the return value are in the range 0...4095. */
5125 input_vid_to_vlan(const struct ofbundle *in_bundle, uint16_t vid)
5127 switch (in_bundle->vlan_mode) {
5128 case PORT_VLAN_ACCESS:
5129 return in_bundle->vlan;
5132 case PORT_VLAN_TRUNK:
5135 case PORT_VLAN_NATIVE_UNTAGGED:
5136 case PORT_VLAN_NATIVE_TAGGED:
5137 return vid ? vid : in_bundle->vlan;
5144 /* Checks whether a packet with the given 'vid' may ingress on 'in_bundle'.
5145 * If so, returns true. Otherwise, returns false and, if 'warn' is true, logs
5148 * 'vid' should be the VID obtained from the 802.1Q header that was received as
5149 * part of a packet (specify 0 if there was no 802.1Q header), in the range
5152 input_vid_is_valid(uint16_t vid, struct ofbundle *in_bundle, bool warn)
5154 /* Allow any VID on the OFPP_NONE port. */
5155 if (in_bundle == &ofpp_none_bundle) {
5159 switch (in_bundle->vlan_mode) {
5160 case PORT_VLAN_ACCESS:
5163 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
5164 VLOG_WARN_RL(&rl, "bridge %s: dropping VLAN %"PRIu16" tagged "
5165 "packet received on port %s configured as VLAN "
5166 "%"PRIu16" access port",
5167 in_bundle->ofproto->up.name, vid,
5168 in_bundle->name, in_bundle->vlan);
5174 case PORT_VLAN_NATIVE_UNTAGGED:
5175 case PORT_VLAN_NATIVE_TAGGED:
5177 /* Port must always carry its native VLAN. */
5181 case PORT_VLAN_TRUNK:
5182 if (!ofbundle_includes_vlan(in_bundle, vid)) {
5184 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
5185 VLOG_WARN_RL(&rl, "bridge %s: dropping VLAN %"PRIu16" packet "
5186 "received on port %s not configured for trunking "
5188 in_bundle->ofproto->up.name, vid,
5189 in_bundle->name, vid);
5201 /* Given 'vlan', the VLAN that a packet belongs to, and
5202 * 'out_bundle', a bundle on which the packet is to be output, returns the VID
5203 * that should be included in the 802.1Q header. (If the return value is 0,
5204 * then the 802.1Q header should only be included in the packet if there is a
5207 * Both 'vlan' and the return value are in the range 0...4095. */
5209 output_vlan_to_vid(const struct ofbundle *out_bundle, uint16_t vlan)
5211 switch (out_bundle->vlan_mode) {
5212 case PORT_VLAN_ACCESS:
5215 case PORT_VLAN_TRUNK:
5216 case PORT_VLAN_NATIVE_TAGGED:
5219 case PORT_VLAN_NATIVE_UNTAGGED:
5220 return vlan == out_bundle->vlan ? 0 : vlan;
5228 output_normal(struct action_xlate_ctx *ctx, const struct ofbundle *out_bundle,
5231 struct ofport_dpif *port;
5233 ovs_be16 tci, old_tci;
5235 vid = output_vlan_to_vid(out_bundle, vlan);
5236 if (!out_bundle->bond) {
5237 port = ofbundle_get_a_port(out_bundle);
5239 port = bond_choose_output_slave(out_bundle->bond, &ctx->flow,
5242 /* No slaves enabled, so drop packet. */
5247 old_tci = ctx->flow.vlan_tci;
5249 if (tci || out_bundle->use_priority_tags) {
5250 tci |= ctx->flow.vlan_tci & htons(VLAN_PCP_MASK);
5252 tci |= htons(VLAN_CFI);
5255 ctx->flow.vlan_tci = tci;
5257 compose_output_action(ctx, port->up.ofp_port);
5258 ctx->flow.vlan_tci = old_tci;
5262 mirror_mask_ffs(mirror_mask_t mask)
5264 BUILD_ASSERT_DECL(sizeof(unsigned int) >= sizeof(mask));
5269 ofbundle_trunks_vlan(const struct ofbundle *bundle, uint16_t vlan)
5271 return (bundle->vlan_mode != PORT_VLAN_ACCESS
5272 && (!bundle->trunks || bitmap_is_set(bundle->trunks, vlan)));
5276 ofbundle_includes_vlan(const struct ofbundle *bundle, uint16_t vlan)
5278 return vlan == bundle->vlan || ofbundle_trunks_vlan(bundle, vlan);
5281 /* Returns an arbitrary interface within 'bundle'. */
5282 static struct ofport_dpif *
5283 ofbundle_get_a_port(const struct ofbundle *bundle)
5285 return CONTAINER_OF(list_front(&bundle->ports),
5286 struct ofport_dpif, bundle_node);
5290 vlan_is_mirrored(const struct ofmirror *m, int vlan)
5292 return !m->vlans || bitmap_is_set(m->vlans, vlan);
5295 /* Returns true if a packet with Ethernet destination MAC 'dst' may be mirrored
5296 * to a VLAN. In general most packets may be mirrored but we want to drop
5297 * protocols that may confuse switches. */
5299 eth_dst_may_rspan(const uint8_t dst[ETH_ADDR_LEN])
5301 /* If you change this function's behavior, please update corresponding
5302 * documentation in vswitch.xml at the same time. */
5303 if (dst[0] != 0x01) {
5304 /* All the currently banned MACs happen to start with 01 currently, so
5305 * this is a quick way to eliminate most of the good ones. */
5307 if (eth_addr_is_reserved(dst)) {
5308 /* Drop STP, IEEE pause frames, and other reserved protocols
5309 * (01-80-c2-00-00-0x). */
5313 if (dst[0] == 0x01 && dst[1] == 0x00 && dst[2] == 0x0c) {
5315 if ((dst[3] & 0xfe) == 0xcc &&
5316 (dst[4] & 0xfe) == 0xcc &&
5317 (dst[5] & 0xfe) == 0xcc) {
5318 /* Drop the following protocols plus others following the same
5321 CDP, VTP, DTP, PAgP (01-00-0c-cc-cc-cc)
5322 Spanning Tree PVSTP+ (01-00-0c-cc-cc-cd)
5323 STP Uplink Fast (01-00-0c-cd-cd-cd) */
5327 if (!(dst[3] | dst[4] | dst[5])) {
5328 /* Drop Inter Switch Link packets (01-00-0c-00-00-00). */
5337 add_mirror_actions(struct action_xlate_ctx *ctx, const struct flow *orig_flow)
5339 struct ofproto_dpif *ofproto = ctx->ofproto;
5340 mirror_mask_t mirrors;
5341 struct ofbundle *in_bundle;
5344 const struct nlattr *a;
5347 in_bundle = lookup_input_bundle(ctx->ofproto, orig_flow->in_port,
5348 ctx->packet != NULL);
5352 mirrors = in_bundle->src_mirrors;
5354 /* Drop frames on bundles reserved for mirroring. */
5355 if (in_bundle->mirror_out) {
5356 if (ctx->packet != NULL) {
5357 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
5358 VLOG_WARN_RL(&rl, "bridge %s: dropping packet received on port "
5359 "%s, which is reserved exclusively for mirroring",
5360 ctx->ofproto->up.name, in_bundle->name);
5366 vid = vlan_tci_to_vid(orig_flow->vlan_tci);
5367 if (!input_vid_is_valid(vid, in_bundle, ctx->packet != NULL)) {
5370 vlan = input_vid_to_vlan(in_bundle, vid);
5372 /* Look at the output ports to check for destination selections. */
5374 NL_ATTR_FOR_EACH (a, left, ctx->odp_actions->data,
5375 ctx->odp_actions->size) {
5376 enum ovs_action_attr type = nl_attr_type(a);
5377 struct ofport_dpif *ofport;
5379 if (type != OVS_ACTION_ATTR_OUTPUT) {
5383 ofport = get_odp_port(ofproto, nl_attr_get_u32(a));
5384 if (ofport && ofport->bundle) {
5385 mirrors |= ofport->bundle->dst_mirrors;
5393 /* Restore the original packet before adding the mirror actions. */
5394 ctx->flow = *orig_flow;
5399 m = ofproto->mirrors[mirror_mask_ffs(mirrors) - 1];
5401 if (!vlan_is_mirrored(m, vlan)) {
5402 mirrors &= mirrors - 1;
5406 mirrors &= ~m->dup_mirrors;
5407 ctx->mirrors |= m->dup_mirrors;
5409 output_normal(ctx, m->out, vlan);
5410 } else if (eth_dst_may_rspan(orig_flow->dl_dst)
5411 && vlan != m->out_vlan) {
5412 struct ofbundle *bundle;
5414 HMAP_FOR_EACH (bundle, hmap_node, &ofproto->bundles) {
5415 if (ofbundle_includes_vlan(bundle, m->out_vlan)
5416 && !bundle->mirror_out) {
5417 output_normal(ctx, bundle, m->out_vlan);
5425 update_mirror_stats(struct ofproto_dpif *ofproto, mirror_mask_t mirrors,
5426 uint64_t packets, uint64_t bytes)
5432 for (; mirrors; mirrors &= mirrors - 1) {
5435 m = ofproto->mirrors[mirror_mask_ffs(mirrors) - 1];
5438 /* In normal circumstances 'm' will not be NULL. However,
5439 * if mirrors are reconfigured, we can temporarily get out
5440 * of sync in facet_revalidate(). We could "correct" the
5441 * mirror list before reaching here, but doing that would
5442 * not properly account the traffic stats we've currently
5443 * accumulated for previous mirror configuration. */
5447 m->packet_count += packets;
5448 m->byte_count += bytes;
5452 /* A VM broadcasts a gratuitous ARP to indicate that it has resumed after
5453 * migration. Older Citrix-patched Linux DomU used gratuitous ARP replies to
5454 * indicate this; newer upstream kernels use gratuitous ARP requests. */
5456 is_gratuitous_arp(const struct flow *flow)
5458 return (flow->dl_type == htons(ETH_TYPE_ARP)
5459 && eth_addr_is_broadcast(flow->dl_dst)
5460 && (flow->nw_proto == ARP_OP_REPLY
5461 || (flow->nw_proto == ARP_OP_REQUEST
5462 && flow->nw_src == flow->nw_dst)));
5466 update_learning_table(struct ofproto_dpif *ofproto,
5467 const struct flow *flow, int vlan,
5468 struct ofbundle *in_bundle)
5470 struct mac_entry *mac;
5472 /* Don't learn the OFPP_NONE port. */
5473 if (in_bundle == &ofpp_none_bundle) {
5477 if (!mac_learning_may_learn(ofproto->ml, flow->dl_src, vlan)) {
5481 mac = mac_learning_insert(ofproto->ml, flow->dl_src, vlan);
5482 if (is_gratuitous_arp(flow)) {
5483 /* We don't want to learn from gratuitous ARP packets that are
5484 * reflected back over bond slaves so we lock the learning table. */
5485 if (!in_bundle->bond) {
5486 mac_entry_set_grat_arp_lock(mac);
5487 } else if (mac_entry_is_grat_arp_locked(mac)) {
5492 if (mac_entry_is_new(mac) || mac->port.p != in_bundle) {
5493 /* The log messages here could actually be useful in debugging,
5494 * so keep the rate limit relatively high. */
5495 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(30, 300);
5496 VLOG_DBG_RL(&rl, "bridge %s: learned that "ETH_ADDR_FMT" is "
5497 "on port %s in VLAN %d",
5498 ofproto->up.name, ETH_ADDR_ARGS(flow->dl_src),
5499 in_bundle->name, vlan);
5501 mac->port.p = in_bundle;
5502 tag_set_add(&ofproto->revalidate_set,
5503 mac_learning_changed(ofproto->ml, mac));
5507 static struct ofbundle *
5508 lookup_input_bundle(struct ofproto_dpif *ofproto, uint16_t in_port, bool warn)
5510 struct ofport_dpif *ofport;
5512 /* Special-case OFPP_NONE, which a controller may use as the ingress
5513 * port for traffic that it is sourcing. */
5514 if (in_port == OFPP_NONE) {
5515 return &ofpp_none_bundle;
5518 /* Find the port and bundle for the received packet. */
5519 ofport = get_ofp_port(ofproto, in_port);
5520 if (ofport && ofport->bundle) {
5521 return ofport->bundle;
5524 /* Odd. A few possible reasons here:
5526 * - We deleted a port but there are still a few packets queued up
5529 * - Someone externally added a port (e.g. "ovs-dpctl add-if") that
5530 * we don't know about.
5532 * - The ofproto client didn't configure the port as part of a bundle.
5535 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
5537 VLOG_WARN_RL(&rl, "bridge %s: received packet on unknown "
5538 "port %"PRIu16, ofproto->up.name, in_port);
5543 /* Determines whether packets in 'flow' within 'ofproto' should be forwarded or
5544 * dropped. Returns true if they may be forwarded, false if they should be
5547 * 'in_port' must be the ofport_dpif that corresponds to flow->in_port.
5548 * 'in_port' must be part of a bundle (e.g. in_port->bundle must be nonnull).
5550 * 'vlan' must be the VLAN that corresponds to flow->vlan_tci on 'in_port', as
5551 * returned by input_vid_to_vlan(). It must be a valid VLAN for 'in_port', as
5552 * checked by input_vid_is_valid().
5554 * May also add tags to '*tags', although the current implementation only does
5555 * so in one special case.
5558 is_admissible(struct ofproto_dpif *ofproto, const struct flow *flow,
5559 struct ofport_dpif *in_port, uint16_t vlan, tag_type *tags)
5561 struct ofbundle *in_bundle = in_port->bundle;
5563 /* Drop frames for reserved multicast addresses
5564 * only if forward_bpdu option is absent. */
5565 if (eth_addr_is_reserved(flow->dl_dst) && !ofproto->up.forward_bpdu) {
5569 if (in_bundle->bond) {
5570 struct mac_entry *mac;
5572 switch (bond_check_admissibility(in_bundle->bond, in_port,
5573 flow->dl_dst, tags)) {
5580 case BV_DROP_IF_MOVED:
5581 mac = mac_learning_lookup(ofproto->ml, flow->dl_src, vlan, NULL);
5582 if (mac && mac->port.p != in_bundle &&
5583 (!is_gratuitous_arp(flow)
5584 || mac_entry_is_grat_arp_locked(mac))) {
5595 xlate_normal(struct action_xlate_ctx *ctx)
5597 struct ofport_dpif *in_port;
5598 struct ofbundle *in_bundle;
5599 struct mac_entry *mac;
5603 ctx->has_normal = true;
5605 in_bundle = lookup_input_bundle(ctx->ofproto, ctx->flow.in_port,
5606 ctx->packet != NULL);
5611 /* We know 'in_port' exists unless it is "ofpp_none_bundle",
5612 * since lookup_input_bundle() succeeded. */
5613 in_port = get_ofp_port(ctx->ofproto, ctx->flow.in_port);
5615 /* Drop malformed frames. */
5616 if (ctx->flow.dl_type == htons(ETH_TYPE_VLAN) &&
5617 !(ctx->flow.vlan_tci & htons(VLAN_CFI))) {
5618 if (ctx->packet != NULL) {
5619 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
5620 VLOG_WARN_RL(&rl, "bridge %s: dropping packet with partial "
5621 "VLAN tag received on port %s",
5622 ctx->ofproto->up.name, in_bundle->name);
5627 /* Drop frames on bundles reserved for mirroring. */
5628 if (in_bundle->mirror_out) {
5629 if (ctx->packet != NULL) {
5630 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
5631 VLOG_WARN_RL(&rl, "bridge %s: dropping packet received on port "
5632 "%s, which is reserved exclusively for mirroring",
5633 ctx->ofproto->up.name, in_bundle->name);
5639 vid = vlan_tci_to_vid(ctx->flow.vlan_tci);
5640 if (!input_vid_is_valid(vid, in_bundle, ctx->packet != NULL)) {
5643 vlan = input_vid_to_vlan(in_bundle, vid);
5645 /* Check other admissibility requirements. */
5647 !is_admissible(ctx->ofproto, &ctx->flow, in_port, vlan, &ctx->tags)) {
5651 /* Learn source MAC. */
5652 if (ctx->may_learn) {
5653 update_learning_table(ctx->ofproto, &ctx->flow, vlan, in_bundle);
5656 /* Determine output bundle. */
5657 mac = mac_learning_lookup(ctx->ofproto->ml, ctx->flow.dl_dst, vlan,
5660 if (mac->port.p != in_bundle) {
5661 output_normal(ctx, mac->port.p, vlan);
5664 struct ofbundle *bundle;
5666 HMAP_FOR_EACH (bundle, hmap_node, &ctx->ofproto->bundles) {
5667 if (bundle != in_bundle
5668 && ofbundle_includes_vlan(bundle, vlan)
5669 && bundle->floodable
5670 && !bundle->mirror_out) {
5671 output_normal(ctx, bundle, vlan);
5674 ctx->nf_output_iface = NF_OUT_FLOOD;
5678 /* Optimized flow revalidation.
5680 * It's a difficult problem, in general, to tell which facets need to have
5681 * their actions recalculated whenever the OpenFlow flow table changes. We
5682 * don't try to solve that general problem: for most kinds of OpenFlow flow
5683 * table changes, we recalculate the actions for every facet. This is
5684 * relatively expensive, but it's good enough if the OpenFlow flow table
5685 * doesn't change very often.
5687 * However, we can expect one particular kind of OpenFlow flow table change to
5688 * happen frequently: changes caused by MAC learning. To avoid wasting a lot
5689 * of CPU on revalidating every facet whenever MAC learning modifies the flow
5690 * table, we add a special case that applies to flow tables in which every rule
5691 * has the same form (that is, the same wildcards), except that the table is
5692 * also allowed to have a single "catch-all" flow that matches all packets. We
5693 * optimize this case by tagging all of the facets that resubmit into the table
5694 * and invalidating the same tag whenever a flow changes in that table. The
5695 * end result is that we revalidate just the facets that need it (and sometimes
5696 * a few more, but not all of the facets or even all of the facets that
5697 * resubmit to the table modified by MAC learning). */
5699 /* Calculates the tag to use for 'flow' and wildcards 'wc' when it is inserted
5700 * into an OpenFlow table with the given 'basis'. */
5702 rule_calculate_tag(const struct flow *flow, const struct flow_wildcards *wc,
5705 if (flow_wildcards_is_catchall(wc)) {
5708 struct flow tag_flow = *flow;
5709 flow_zero_wildcards(&tag_flow, wc);
5710 return tag_create_deterministic(flow_hash(&tag_flow, secret));
5714 /* Following a change to OpenFlow table 'table_id' in 'ofproto', update the
5715 * taggability of that table.
5717 * This function must be called after *each* change to a flow table. If you
5718 * skip calling it on some changes then the pointer comparisons at the end can
5719 * be invalid if you get unlucky. For example, if a flow removal causes a
5720 * cls_table to be destroyed and then a flow insertion causes a cls_table with
5721 * different wildcards to be created with the same address, then this function
5722 * will incorrectly skip revalidation. */
5724 table_update_taggable(struct ofproto_dpif *ofproto, uint8_t table_id)
5726 struct table_dpif *table = &ofproto->tables[table_id];
5727 const struct oftable *oftable = &ofproto->up.tables[table_id];
5728 struct cls_table *catchall, *other;
5729 struct cls_table *t;
5731 catchall = other = NULL;
5733 switch (hmap_count(&oftable->cls.tables)) {
5735 /* We could tag this OpenFlow table but it would make the logic a
5736 * little harder and it's a corner case that doesn't seem worth it
5742 HMAP_FOR_EACH (t, hmap_node, &oftable->cls.tables) {
5743 if (cls_table_is_catchall(t)) {
5745 } else if (!other) {
5748 /* Indicate that we can't tag this by setting both tables to
5749 * NULL. (We know that 'catchall' is already NULL.) */
5756 /* Can't tag this table. */
5760 if (table->catchall_table != catchall || table->other_table != other) {
5761 table->catchall_table = catchall;
5762 table->other_table = other;
5763 ofproto->need_revalidate = true;
5767 /* Given 'rule' that has changed in some way (either it is a rule being
5768 * inserted, a rule being deleted, or a rule whose actions are being
5769 * modified), marks facets for revalidation to ensure that packets will be
5770 * forwarded correctly according to the new state of the flow table.
5772 * This function must be called after *each* change to a flow table. See
5773 * the comment on table_update_taggable() for more information. */
5775 rule_invalidate(const struct rule_dpif *rule)
5777 struct ofproto_dpif *ofproto = ofproto_dpif_cast(rule->up.ofproto);
5779 table_update_taggable(ofproto, rule->up.table_id);
5781 if (!ofproto->need_revalidate) {
5782 struct table_dpif *table = &ofproto->tables[rule->up.table_id];
5784 if (table->other_table && rule->tag) {
5785 tag_set_add(&ofproto->revalidate_set, rule->tag);
5787 ofproto->need_revalidate = true;
5793 set_frag_handling(struct ofproto *ofproto_,
5794 enum ofp_config_flags frag_handling)
5796 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
5798 if (frag_handling != OFPC_FRAG_REASM) {
5799 ofproto->need_revalidate = true;
5807 packet_out(struct ofproto *ofproto_, struct ofpbuf *packet,
5808 const struct flow *flow,
5809 const union ofp_action *ofp_actions, size_t n_ofp_actions)
5811 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
5814 if (flow->in_port >= ofproto->max_ports && flow->in_port < OFPP_MAX) {
5815 return OFPERR_NXBRC_BAD_IN_PORT;
5818 error = validate_actions(ofp_actions, n_ofp_actions, flow,
5819 ofproto->max_ports);
5821 struct odputil_keybuf keybuf;
5822 struct ofpbuf *odp_actions;
5823 struct ofproto_push push;
5826 ofpbuf_use_stack(&key, &keybuf, sizeof keybuf);
5827 odp_flow_key_from_flow(&key, flow);
5829 action_xlate_ctx_init(&push.ctx, ofproto, flow, flow->vlan_tci, NULL,
5830 packet_get_tcp_flags(packet, flow), packet);
5832 /* Ensure that resubmits in 'ofp_actions' get accounted to their
5833 * matching rules. */
5835 push.bytes = packet->size;
5836 push.used = time_msec();
5837 push.ctx.resubmit_hook = push_resubmit;
5839 odp_actions = xlate_actions(&push.ctx, ofp_actions, n_ofp_actions);
5840 dpif_execute(ofproto->dpif, key.data, key.size,
5841 odp_actions->data, odp_actions->size, packet);
5842 ofpbuf_delete(odp_actions);
5850 set_netflow(struct ofproto *ofproto_,
5851 const struct netflow_options *netflow_options)
5853 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
5855 if (netflow_options) {
5856 if (!ofproto->netflow) {
5857 ofproto->netflow = netflow_create();
5859 return netflow_set_options(ofproto->netflow, netflow_options);
5861 netflow_destroy(ofproto->netflow);
5862 ofproto->netflow = NULL;
5868 get_netflow_ids(const struct ofproto *ofproto_,
5869 uint8_t *engine_type, uint8_t *engine_id)
5871 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
5873 dpif_get_netflow_ids(ofproto->dpif, engine_type, engine_id);
5877 send_active_timeout(struct ofproto_dpif *ofproto, struct facet *facet)
5879 if (!facet_is_controller_flow(facet) &&
5880 netflow_active_timeout_expired(ofproto->netflow, &facet->nf_flow)) {
5881 struct subfacet *subfacet;
5882 struct ofexpired expired;
5884 LIST_FOR_EACH (subfacet, list_node, &facet->subfacets) {
5885 if (subfacet->installed) {
5886 struct dpif_flow_stats stats;
5888 subfacet_install(subfacet, subfacet->actions,
5889 subfacet->actions_len, &stats);
5890 subfacet_update_stats(subfacet, &stats);
5894 expired.flow = facet->flow;
5895 expired.packet_count = facet->packet_count;
5896 expired.byte_count = facet->byte_count;
5897 expired.used = facet->used;
5898 netflow_expire(ofproto->netflow, &facet->nf_flow, &expired);
5903 send_netflow_active_timeouts(struct ofproto_dpif *ofproto)
5905 struct facet *facet;
5907 HMAP_FOR_EACH (facet, hmap_node, &ofproto->facets) {
5908 send_active_timeout(ofproto, facet);
5912 static struct ofproto_dpif *
5913 ofproto_dpif_lookup(const char *name)
5915 struct ofproto_dpif *ofproto;
5917 HMAP_FOR_EACH_WITH_HASH (ofproto, all_ofproto_dpifs_node,
5918 hash_string(name, 0), &all_ofproto_dpifs) {
5919 if (!strcmp(ofproto->up.name, name)) {
5927 ofproto_unixctl_fdb_flush(struct unixctl_conn *conn, int argc,
5928 const char *argv[], void *aux OVS_UNUSED)
5930 struct ofproto_dpif *ofproto;
5933 ofproto = ofproto_dpif_lookup(argv[1]);
5935 unixctl_command_reply_error(conn, "no such bridge");
5938 mac_learning_flush(ofproto->ml, &ofproto->revalidate_set);
5940 HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
5941 mac_learning_flush(ofproto->ml, &ofproto->revalidate_set);
5945 unixctl_command_reply(conn, "table successfully flushed");
5949 ofproto_unixctl_fdb_show(struct unixctl_conn *conn, int argc OVS_UNUSED,
5950 const char *argv[], void *aux OVS_UNUSED)
5952 struct ds ds = DS_EMPTY_INITIALIZER;
5953 const struct ofproto_dpif *ofproto;
5954 const struct mac_entry *e;
5956 ofproto = ofproto_dpif_lookup(argv[1]);
5958 unixctl_command_reply_error(conn, "no such bridge");
5962 ds_put_cstr(&ds, " port VLAN MAC Age\n");
5963 LIST_FOR_EACH (e, lru_node, &ofproto->ml->lrus) {
5964 struct ofbundle *bundle = e->port.p;
5965 ds_put_format(&ds, "%5d %4d "ETH_ADDR_FMT" %3d\n",
5966 ofbundle_get_a_port(bundle)->odp_port,
5967 e->vlan, ETH_ADDR_ARGS(e->mac),
5968 mac_entry_age(ofproto->ml, e));
5970 unixctl_command_reply(conn, ds_cstr(&ds));
5975 struct action_xlate_ctx ctx;
5981 trace_format_rule(struct ds *result, uint8_t table_id, int level,
5982 const struct rule_dpif *rule)
5984 ds_put_char_multiple(result, '\t', level);
5986 ds_put_cstr(result, "No match\n");
5990 ds_put_format(result, "Rule: table=%"PRIu8" cookie=%#"PRIx64" ",
5991 table_id, ntohll(rule->up.flow_cookie));
5992 cls_rule_format(&rule->up.cr, result);
5993 ds_put_char(result, '\n');
5995 ds_put_char_multiple(result, '\t', level);
5996 ds_put_cstr(result, "OpenFlow ");
5997 ofp_print_actions(result, rule->up.actions, rule->up.n_actions);
5998 ds_put_char(result, '\n');
6002 trace_format_flow(struct ds *result, int level, const char *title,
6003 struct trace_ctx *trace)
6005 ds_put_char_multiple(result, '\t', level);
6006 ds_put_format(result, "%s: ", title);
6007 if (flow_equal(&trace->ctx.flow, &trace->flow)) {
6008 ds_put_cstr(result, "unchanged");
6010 flow_format(result, &trace->ctx.flow);
6011 trace->flow = trace->ctx.flow;
6013 ds_put_char(result, '\n');
6017 trace_format_regs(struct ds *result, int level, const char *title,
6018 struct trace_ctx *trace)
6022 ds_put_char_multiple(result, '\t', level);
6023 ds_put_format(result, "%s:", title);
6024 for (i = 0; i < FLOW_N_REGS; i++) {
6025 ds_put_format(result, " reg%zu=0x%"PRIx32, i, trace->flow.regs[i]);
6027 ds_put_char(result, '\n');
6031 trace_format_odp(struct ds *result, int level, const char *title,
6032 struct trace_ctx *trace)
6034 struct ofpbuf *odp_actions = trace->ctx.odp_actions;
6036 ds_put_char_multiple(result, '\t', level);
6037 ds_put_format(result, "%s: ", title);
6038 format_odp_actions(result, odp_actions->data, odp_actions->size);
6039 ds_put_char(result, '\n');
6043 trace_resubmit(struct action_xlate_ctx *ctx, struct rule_dpif *rule)
6045 struct trace_ctx *trace = CONTAINER_OF(ctx, struct trace_ctx, ctx);
6046 struct ds *result = trace->result;
6048 ds_put_char(result, '\n');
6049 trace_format_flow(result, ctx->recurse + 1, "Resubmitted flow", trace);
6050 trace_format_regs(result, ctx->recurse + 1, "Resubmitted regs", trace);
6051 trace_format_odp(result, ctx->recurse + 1, "Resubmitted odp", trace);
6052 trace_format_rule(result, ctx->table_id, ctx->recurse + 1, rule);
6056 ofproto_unixctl_trace(struct unixctl_conn *conn, int argc, const char *argv[],
6057 void *aux OVS_UNUSED)
6059 const char *dpname = argv[1];
6060 struct ofproto_dpif *ofproto;
6061 struct ofpbuf odp_key;
6062 struct ofpbuf *packet;
6063 ovs_be16 initial_tci;
6069 ofpbuf_init(&odp_key, 0);
6072 ofproto = ofproto_dpif_lookup(dpname);
6074 unixctl_command_reply_error(conn, "Unknown ofproto (use ofproto/list "
6078 if (argc == 3 || (argc == 4 && !strcmp(argv[3], "-generate"))) {
6079 /* ofproto/trace dpname flow [-generate] */
6080 const char *flow_s = argv[2];
6081 const char *generate_s = argv[3];
6084 /* Convert string to datapath key. */
6085 ofpbuf_init(&odp_key, 0);
6086 error = odp_flow_key_from_string(flow_s, NULL, &odp_key);
6088 unixctl_command_reply_error(conn, "Bad flow syntax");
6092 /* Convert odp_key to flow. */
6093 error = ofproto_dpif_extract_flow_key(ofproto, odp_key.data,
6094 odp_key.size, &flow,
6095 &initial_tci, NULL);
6096 if (error == ODP_FIT_ERROR) {
6097 unixctl_command_reply_error(conn, "Invalid flow");
6101 /* Generate a packet, if requested. */
6103 packet = ofpbuf_new(0);
6104 flow_compose(packet, &flow);
6106 } else if (argc == 6) {
6107 /* ofproto/trace dpname priority tun_id in_port packet */
6108 const char *priority_s = argv[2];
6109 const char *tun_id_s = argv[3];
6110 const char *in_port_s = argv[4];
6111 const char *packet_s = argv[5];
6112 uint16_t in_port = ofp_port_to_odp_port(atoi(in_port_s));
6113 ovs_be64 tun_id = htonll(strtoull(tun_id_s, NULL, 0));
6114 uint32_t priority = atoi(priority_s);
6117 msg = eth_from_hex(packet_s, &packet);
6119 unixctl_command_reply_error(conn, msg);
6123 ds_put_cstr(&result, "Packet: ");
6124 s = ofp_packet_to_string(packet->data, packet->size);
6125 ds_put_cstr(&result, s);
6128 flow_extract(packet, priority, tun_id, in_port, &flow);
6129 initial_tci = flow.vlan_tci;
6131 unixctl_command_reply_error(conn, "Bad command syntax");
6135 ofproto_trace(ofproto, &flow, packet, initial_tci, &result);
6136 unixctl_command_reply(conn, ds_cstr(&result));
6139 ds_destroy(&result);
6140 ofpbuf_delete(packet);
6141 ofpbuf_uninit(&odp_key);
6145 ofproto_trace(struct ofproto_dpif *ofproto, const struct flow *flow,
6146 const struct ofpbuf *packet, ovs_be16 initial_tci,
6149 struct rule_dpif *rule;
6151 ds_put_cstr(ds, "Flow: ");
6152 flow_format(ds, flow);
6153 ds_put_char(ds, '\n');
6155 rule = rule_dpif_lookup(ofproto, flow, 0);
6156 trace_format_rule(ds, 0, 0, rule);
6158 struct trace_ctx trace;
6159 struct ofpbuf *odp_actions;
6162 tcp_flags = packet ? packet_get_tcp_flags(packet, flow) : 0;
6165 action_xlate_ctx_init(&trace.ctx, ofproto, flow, initial_tci,
6166 rule, tcp_flags, packet);
6167 trace.ctx.resubmit_hook = trace_resubmit;
6168 odp_actions = xlate_actions(&trace.ctx,
6169 rule->up.actions, rule->up.n_actions);
6171 ds_put_char(ds, '\n');
6172 trace_format_flow(ds, 0, "Final flow", &trace);
6173 ds_put_cstr(ds, "Datapath actions: ");
6174 format_odp_actions(ds, odp_actions->data, odp_actions->size);
6175 ofpbuf_delete(odp_actions);
6177 if (!trace.ctx.may_set_up_flow) {
6179 ds_put_cstr(ds, "\nThis flow is not cachable.");
6181 ds_put_cstr(ds, "\nThe datapath actions are incomplete--"
6182 "for complete actions, please supply a packet.");
6189 ofproto_dpif_clog(struct unixctl_conn *conn OVS_UNUSED, int argc OVS_UNUSED,
6190 const char *argv[] OVS_UNUSED, void *aux OVS_UNUSED)
6193 unixctl_command_reply(conn, NULL);
6197 ofproto_dpif_unclog(struct unixctl_conn *conn OVS_UNUSED, int argc OVS_UNUSED,
6198 const char *argv[] OVS_UNUSED, void *aux OVS_UNUSED)
6201 unixctl_command_reply(conn, NULL);
6204 /* Runs a self-check of flow translations in 'ofproto'. Appends a message to
6205 * 'reply' describing the results. */
6207 ofproto_dpif_self_check__(struct ofproto_dpif *ofproto, struct ds *reply)
6209 struct facet *facet;
6213 HMAP_FOR_EACH (facet, hmap_node, &ofproto->facets) {
6214 if (!facet_check_consistency(facet)) {
6219 ofproto->need_revalidate = true;
6223 ds_put_format(reply, "%s: self-check failed (%d errors)\n",
6224 ofproto->up.name, errors);
6226 ds_put_format(reply, "%s: self-check passed\n", ofproto->up.name);
6231 ofproto_dpif_self_check(struct unixctl_conn *conn,
6232 int argc, const char *argv[], void *aux OVS_UNUSED)
6234 struct ds reply = DS_EMPTY_INITIALIZER;
6235 struct ofproto_dpif *ofproto;
6238 ofproto = ofproto_dpif_lookup(argv[1]);
6240 unixctl_command_reply_error(conn, "Unknown ofproto (use "
6241 "ofproto/list for help)");
6244 ofproto_dpif_self_check__(ofproto, &reply);
6246 HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
6247 ofproto_dpif_self_check__(ofproto, &reply);
6251 unixctl_command_reply(conn, ds_cstr(&reply));
6256 ofproto_dpif_unixctl_init(void)
6258 static bool registered;
6264 unixctl_command_register(
6266 "bridge {tun_id in_port packet | odp_flow [-generate]}",
6267 2, 5, ofproto_unixctl_trace, NULL);
6268 unixctl_command_register("fdb/flush", "[bridge]", 0, 1,
6269 ofproto_unixctl_fdb_flush, NULL);
6270 unixctl_command_register("fdb/show", "bridge", 1, 1,
6271 ofproto_unixctl_fdb_show, NULL);
6272 unixctl_command_register("ofproto/clog", "", 0, 0,
6273 ofproto_dpif_clog, NULL);
6274 unixctl_command_register("ofproto/unclog", "", 0, 0,
6275 ofproto_dpif_unclog, NULL);
6276 unixctl_command_register("ofproto/self-check", "[bridge]", 0, 1,
6277 ofproto_dpif_self_check, NULL);
6280 /* Linux VLAN device support (e.g. "eth0.10" for VLAN 10.)
6282 * This is deprecated. It is only for compatibility with broken device drivers
6283 * in old versions of Linux that do not properly support VLANs when VLAN
6284 * devices are not used. When broken device drivers are no longer in
6285 * widespread use, we will delete these interfaces. */
6288 set_realdev(struct ofport *ofport_, uint16_t realdev_ofp_port, int vid)
6290 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport_->ofproto);
6291 struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
6293 if (realdev_ofp_port == ofport->realdev_ofp_port
6294 && vid == ofport->vlandev_vid) {
6298 ofproto->need_revalidate = true;
6300 if (ofport->realdev_ofp_port) {
6303 if (realdev_ofp_port && ofport->bundle) {
6304 /* vlandevs are enslaved to their realdevs, so they are not allowed to
6305 * themselves be part of a bundle. */
6306 bundle_set(ofport->up.ofproto, ofport->bundle, NULL);
6309 ofport->realdev_ofp_port = realdev_ofp_port;
6310 ofport->vlandev_vid = vid;
6312 if (realdev_ofp_port) {
6313 vsp_add(ofport, realdev_ofp_port, vid);
6320 hash_realdev_vid(uint16_t realdev_ofp_port, int vid)
6322 return hash_2words(realdev_ofp_port, vid);
6326 vsp_realdev_to_vlandev(const struct ofproto_dpif *ofproto,
6327 uint32_t realdev_odp_port, ovs_be16 vlan_tci)
6329 if (!hmap_is_empty(&ofproto->realdev_vid_map)) {
6330 uint16_t realdev_ofp_port = odp_port_to_ofp_port(realdev_odp_port);
6331 int vid = vlan_tci_to_vid(vlan_tci);
6332 const struct vlan_splinter *vsp;
6334 HMAP_FOR_EACH_WITH_HASH (vsp, realdev_vid_node,
6335 hash_realdev_vid(realdev_ofp_port, vid),
6336 &ofproto->realdev_vid_map) {
6337 if (vsp->realdev_ofp_port == realdev_ofp_port
6338 && vsp->vid == vid) {
6339 return ofp_port_to_odp_port(vsp->vlandev_ofp_port);
6343 return realdev_odp_port;
6346 static struct vlan_splinter *
6347 vlandev_find(const struct ofproto_dpif *ofproto, uint16_t vlandev_ofp_port)
6349 struct vlan_splinter *vsp;
6351 HMAP_FOR_EACH_WITH_HASH (vsp, vlandev_node, hash_int(vlandev_ofp_port, 0),
6352 &ofproto->vlandev_map) {
6353 if (vsp->vlandev_ofp_port == vlandev_ofp_port) {
6362 vsp_vlandev_to_realdev(const struct ofproto_dpif *ofproto,
6363 uint16_t vlandev_ofp_port, int *vid)
6365 if (!hmap_is_empty(&ofproto->vlandev_map)) {
6366 const struct vlan_splinter *vsp;
6368 vsp = vlandev_find(ofproto, vlandev_ofp_port);
6373 return vsp->realdev_ofp_port;
6380 vsp_remove(struct ofport_dpif *port)
6382 struct ofproto_dpif *ofproto = ofproto_dpif_cast(port->up.ofproto);
6383 struct vlan_splinter *vsp;
6385 vsp = vlandev_find(ofproto, port->up.ofp_port);
6387 hmap_remove(&ofproto->vlandev_map, &vsp->vlandev_node);
6388 hmap_remove(&ofproto->realdev_vid_map, &vsp->realdev_vid_node);
6391 port->realdev_ofp_port = 0;
6393 VLOG_ERR("missing vlan device record");
6398 vsp_add(struct ofport_dpif *port, uint16_t realdev_ofp_port, int vid)
6400 struct ofproto_dpif *ofproto = ofproto_dpif_cast(port->up.ofproto);
6402 if (!vsp_vlandev_to_realdev(ofproto, port->up.ofp_port, NULL)
6403 && (vsp_realdev_to_vlandev(ofproto, realdev_ofp_port, htons(vid))
6404 == realdev_ofp_port)) {
6405 struct vlan_splinter *vsp;
6407 vsp = xmalloc(sizeof *vsp);
6408 hmap_insert(&ofproto->vlandev_map, &vsp->vlandev_node,
6409 hash_int(port->up.ofp_port, 0));
6410 hmap_insert(&ofproto->realdev_vid_map, &vsp->realdev_vid_node,
6411 hash_realdev_vid(realdev_ofp_port, vid));
6412 vsp->realdev_ofp_port = realdev_ofp_port;
6413 vsp->vlandev_ofp_port = port->up.ofp_port;
6416 port->realdev_ofp_port = realdev_ofp_port;
6418 VLOG_ERR("duplicate vlan device record");
6422 const struct ofproto_class ofproto_dpif_class = {
6451 port_is_lacp_current,
6452 NULL, /* rule_choose_table */
6459 rule_modify_actions,
6467 get_cfm_remote_mpids,
6471 get_stp_port_status,
6478 is_mirror_output_bundle,
6479 forward_bpdu_changed,