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 "multipath.h"
44 #include "ofp-print.h"
45 #include "ofproto-dpif-sflow.h"
46 #include "poll-loop.h"
48 #include "unaligned.h"
50 #include "vlan-bitmap.h"
53 VLOG_DEFINE_THIS_MODULE(ofproto_dpif);
55 COVERAGE_DEFINE(ofproto_dpif_ctlr_action);
56 COVERAGE_DEFINE(ofproto_dpif_expired);
57 COVERAGE_DEFINE(ofproto_dpif_no_packet_in);
58 COVERAGE_DEFINE(ofproto_dpif_xlate);
59 COVERAGE_DEFINE(facet_changed_rule);
60 COVERAGE_DEFINE(facet_invalidated);
61 COVERAGE_DEFINE(facet_revalidate);
62 COVERAGE_DEFINE(facet_unexpected);
64 /* Maximum depth of flow table recursion (due to resubmit actions) in a
65 * flow translation. */
66 #define MAX_RESUBMIT_RECURSION 32
68 /* Number of implemented OpenFlow tables. */
69 enum { N_TABLES = 255 };
70 BUILD_ASSERT_DECL(N_TABLES >= 1 && N_TABLES <= 255);
78 long long int used; /* Time last used; time created if not used. */
82 * - Do include packets and bytes from facets that have been deleted or
83 * whose own statistics have been folded into the rule.
85 * - Do include packets and bytes sent "by hand" that were accounted to
86 * the rule without any facet being involved (this is a rare corner
87 * case in rule_execute()).
89 * - Do not include packet or bytes that can be obtained from any facet's
90 * packet_count or byte_count member or that can be obtained from the
91 * datapath by, e.g., dpif_flow_get() for any subfacet.
93 uint64_t packet_count; /* Number of packets received. */
94 uint64_t byte_count; /* Number of bytes received. */
96 tag_type tag; /* Caches rule_calculate_tag() result. */
98 struct list facets; /* List of "struct facet"s. */
101 static struct rule_dpif *rule_dpif_cast(const struct rule *rule)
103 return rule ? CONTAINER_OF(rule, struct rule_dpif, up) : NULL;
106 static struct rule_dpif *rule_dpif_lookup(struct ofproto_dpif *,
107 const struct flow *, uint8_t table);
109 static void flow_push_stats(const struct rule_dpif *, const struct flow *,
110 uint64_t packets, uint64_t bytes,
113 static tag_type rule_calculate_tag(const struct flow *,
114 const struct flow_wildcards *,
116 static void rule_invalidate(const struct rule_dpif *);
118 #define MAX_MIRRORS 32
119 typedef uint32_t mirror_mask_t;
120 #define MIRROR_MASK_C(X) UINT32_C(X)
121 BUILD_ASSERT_DECL(sizeof(mirror_mask_t) * CHAR_BIT >= MAX_MIRRORS);
123 struct ofproto_dpif *ofproto; /* Owning ofproto. */
124 size_t idx; /* In ofproto's "mirrors" array. */
125 void *aux; /* Key supplied by ofproto's client. */
126 char *name; /* Identifier for log messages. */
128 /* Selection criteria. */
129 struct hmapx srcs; /* Contains "struct ofbundle *"s. */
130 struct hmapx dsts; /* Contains "struct ofbundle *"s. */
131 unsigned long *vlans; /* Bitmap of chosen VLANs, NULL selects all. */
133 /* Output (exactly one of out == NULL and out_vlan == -1 is true). */
134 struct ofbundle *out; /* Output port or NULL. */
135 int out_vlan; /* Output VLAN or -1. */
136 mirror_mask_t dup_mirrors; /* Bitmap of mirrors with the same output. */
139 int64_t packet_count; /* Number of packets sent. */
140 int64_t byte_count; /* Number of bytes sent. */
143 static void mirror_destroy(struct ofmirror *);
144 static void update_mirror_stats(struct ofproto_dpif *ofproto,
145 mirror_mask_t mirrors,
146 uint64_t packets, uint64_t bytes);
149 struct ofproto_dpif *ofproto; /* Owning ofproto. */
150 struct hmap_node hmap_node; /* In struct ofproto's "bundles" hmap. */
151 void *aux; /* Key supplied by ofproto's client. */
152 char *name; /* Identifier for log messages. */
155 struct list ports; /* Contains "struct ofport"s. */
156 enum port_vlan_mode vlan_mode; /* VLAN mode */
157 int vlan; /* -1=trunk port, else a 12-bit VLAN ID. */
158 unsigned long *trunks; /* Bitmap of trunked VLANs, if 'vlan' == -1.
159 * NULL if all VLANs are trunked. */
160 struct lacp *lacp; /* LACP if LACP is enabled, otherwise NULL. */
161 struct bond *bond; /* Nonnull iff more than one port. */
162 bool use_priority_tags; /* Use 802.1p tag for frames in VLAN 0? */
165 bool floodable; /* True if no port has OFPPC_NO_FLOOD set. */
167 /* Port mirroring info. */
168 mirror_mask_t src_mirrors; /* Mirrors triggered when packet received. */
169 mirror_mask_t dst_mirrors; /* Mirrors triggered when packet sent. */
170 mirror_mask_t mirror_out; /* Mirrors that output to this bundle. */
173 static void bundle_remove(struct ofport *);
174 static void bundle_update(struct ofbundle *);
175 static void bundle_destroy(struct ofbundle *);
176 static void bundle_del_port(struct ofport_dpif *);
177 static void bundle_run(struct ofbundle *);
178 static void bundle_wait(struct ofbundle *);
179 static struct ofbundle *lookup_input_bundle(struct ofproto_dpif *,
180 uint16_t in_port, bool warn);
182 /* A controller may use OFPP_NONE as the ingress port to indicate that
183 * it did not arrive on a "real" port. 'ofpp_none_bundle' exists for
184 * when an input bundle is needed for validation (e.g., mirroring or
185 * OFPP_NORMAL processing). It is not connected to an 'ofproto' or have
186 * any 'port' structs, so care must be taken when dealing with it. */
187 static struct ofbundle ofpp_none_bundle = {
189 .vlan_mode = PORT_VLAN_TRUNK
192 static void stp_run(struct ofproto_dpif *ofproto);
193 static void stp_wait(struct ofproto_dpif *ofproto);
194 static int set_stp_port(struct ofport *,
195 const struct ofproto_port_stp_settings *);
197 static bool ofbundle_includes_vlan(const struct ofbundle *, uint16_t vlan);
199 struct action_xlate_ctx {
200 /* action_xlate_ctx_init() initializes these members. */
203 struct ofproto_dpif *ofproto;
205 /* Flow to which the OpenFlow actions apply. xlate_actions() will modify
206 * this flow when actions change header fields. */
209 /* The packet corresponding to 'flow', or a null pointer if we are
210 * revalidating without a packet to refer to. */
211 const struct ofpbuf *packet;
213 /* Should OFPP_NORMAL update the MAC learning table? We want to update it
214 * if we are actually processing a packet, or if we are accounting for
215 * packets that the datapath has processed, but not if we are just
219 /* Should "learn" actions update the flow table? We want to update it if
220 * we are actually processing a packet, or in most cases if we are
221 * accounting for packets that the datapath has processed, but not if we
222 * are just revalidating. */
225 /* If nonnull, called just before executing a resubmit action.
227 * This is normally null so the client has to set it manually after
228 * calling action_xlate_ctx_init(). */
229 void (*resubmit_hook)(struct action_xlate_ctx *, struct rule_dpif *);
231 /* xlate_actions() initializes and uses these members. The client might want
232 * to look at them after it returns. */
234 struct ofpbuf *odp_actions; /* Datapath actions. */
235 tag_type tags; /* Tags associated with actions. */
236 bool may_set_up_flow; /* True ordinarily; false if the actions must
237 * be reassessed for every packet. */
238 bool has_learn; /* Actions include NXAST_LEARN? */
239 bool has_normal; /* Actions output to OFPP_NORMAL? */
240 uint16_t nf_output_iface; /* Output interface index for NetFlow. */
241 mirror_mask_t mirrors; /* Bitmap of associated mirrors. */
243 /* xlate_actions() initializes and uses these members, but the client has no
244 * reason to look at them. */
246 int recurse; /* Recursion level, via xlate_table_action. */
247 struct flow base_flow; /* Flow at the last commit. */
248 uint32_t original_priority; /* Priority when packet arrived. */
249 uint8_t table_id; /* OpenFlow table ID where flow was found. */
250 uint32_t sflow_n_outputs; /* Number of output ports. */
251 uint16_t sflow_odp_port; /* Output port for composing sFlow action. */
252 uint16_t user_cookie_offset;/* Used for user_action_cookie fixup. */
253 bool exit; /* No further actions should be processed. */
256 static void action_xlate_ctx_init(struct action_xlate_ctx *,
257 struct ofproto_dpif *, const struct flow *,
258 ovs_be16 initial_tci, const struct ofpbuf *);
259 static struct ofpbuf *xlate_actions(struct action_xlate_ctx *,
260 const union ofp_action *in, size_t n_in);
262 /* An exact-match instantiation of an OpenFlow flow.
264 * A facet associates a "struct flow", which represents the Open vSwitch
265 * userspace idea of an exact-match flow, with one or more subfacets. Each
266 * subfacet tracks the datapath's idea of the exact-match flow equivalent to
267 * the facet. When the kernel module (or other dpif implementation) and Open
268 * vSwitch userspace agree on the definition of a flow key, there is exactly
269 * one subfacet per facet. If the dpif implementation supports more-specific
270 * flow matching than userspace, however, a facet can have more than one
271 * subfacet, each of which corresponds to some distinction in flow that
272 * userspace simply doesn't understand.
274 * Flow expiration works in terms of subfacets, so a facet must have at least
275 * one subfacet or it will never expire, leaking memory. */
278 struct hmap_node hmap_node; /* In owning ofproto's 'facets' hmap. */
279 struct list list_node; /* In owning rule's 'facets' list. */
280 struct rule_dpif *rule; /* Owning rule. */
283 struct list subfacets;
284 long long int used; /* Time last used; time created if not used. */
291 * - Do include packets and bytes sent "by hand", e.g. with
294 * - Do include packets and bytes that were obtained from the datapath
295 * when a subfacet's statistics were reset (e.g. dpif_flow_put() with
296 * DPIF_FP_ZERO_STATS).
298 * - Do not include packets or bytes that can be obtained from the
299 * datapath for any existing subfacet.
301 uint64_t packet_count; /* Number of packets received. */
302 uint64_t byte_count; /* Number of bytes received. */
304 /* Resubmit statistics. */
305 uint64_t prev_packet_count; /* Number of packets from last stats push. */
306 uint64_t prev_byte_count; /* Number of bytes from last stats push. */
307 long long int prev_used; /* Used time from last stats push. */
310 uint64_t accounted_bytes; /* Bytes processed by facet_account(). */
311 struct netflow_flow nf_flow; /* Per-flow NetFlow tracking data. */
313 /* Properties of datapath actions.
315 * Every subfacet has its own actions because actions can differ slightly
316 * between splintered and non-splintered subfacets due to the VLAN tag
317 * being initially different (present vs. absent). All of them have these
318 * properties in common so we just store one copy of them here. */
319 bool may_install; /* Reassess actions for every packet? */
320 bool has_learn; /* Actions include NXAST_LEARN? */
321 bool has_normal; /* Actions output to OFPP_NORMAL? */
322 tag_type tags; /* Tags that would require revalidation. */
323 mirror_mask_t mirrors; /* Bitmap of dependent mirrors. */
326 static struct facet *facet_create(struct rule_dpif *, const struct flow *);
327 static void facet_remove(struct ofproto_dpif *, struct facet *);
328 static void facet_free(struct facet *);
330 static struct facet *facet_find(struct ofproto_dpif *, const struct flow *);
331 static struct facet *facet_lookup_valid(struct ofproto_dpif *,
332 const struct flow *);
333 static bool facet_revalidate(struct ofproto_dpif *, struct facet *);
335 static bool execute_controller_action(struct ofproto_dpif *,
337 const struct nlattr *odp_actions,
339 struct ofpbuf *packet, bool clone);
341 static void facet_flush_stats(struct ofproto_dpif *, struct facet *);
343 static void facet_update_time(struct ofproto_dpif *, struct facet *,
345 static void facet_reset_counters(struct facet *);
346 static void facet_push_stats(struct facet *);
347 static void facet_account(struct ofproto_dpif *, struct facet *,
350 static bool facet_is_controller_flow(struct facet *);
352 /* A dpif flow and actions associated with a facet.
354 * See also the large comment on struct facet. */
357 struct hmap_node hmap_node; /* In struct ofproto_dpif 'subfacets' list. */
358 struct list list_node; /* In struct facet's 'facets' list. */
359 struct facet *facet; /* Owning facet. */
363 * To save memory in the common case, 'key' is NULL if 'key_fitness' is
364 * ODP_FIT_PERFECT, that is, odp_flow_key_from_flow() can accurately
365 * regenerate the ODP flow key from ->facet->flow. */
366 enum odp_key_fitness key_fitness;
370 long long int used; /* Time last used; time created if not used. */
372 uint64_t dp_packet_count; /* Last known packet count in the datapath. */
373 uint64_t dp_byte_count; /* Last known byte count in the datapath. */
377 * These should be essentially identical for every subfacet in a facet, but
378 * may differ in trivial ways due to VLAN splinters. */
379 size_t actions_len; /* Number of bytes in actions[]. */
380 struct nlattr *actions; /* Datapath actions. */
382 bool installed; /* Installed in datapath? */
384 /* This value is normally the same as ->facet->flow.vlan_tci. Only VLAN
385 * splinters can cause it to differ. This value should be removed when
386 * the VLAN splinters feature is no longer needed. */
387 ovs_be16 initial_tci; /* Initial VLAN TCI value. */
390 static struct subfacet *subfacet_create(struct ofproto_dpif *, struct facet *,
391 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 ofproto_dpif *, struct subfacet *);
397 static void subfacet_destroy__(struct ofproto_dpif *, struct subfacet *);
398 static void subfacet_reset_dp_stats(struct subfacet *,
399 struct dpif_flow_stats *);
400 static void subfacet_update_time(struct ofproto_dpif *, struct subfacet *,
402 static void subfacet_update_stats(struct ofproto_dpif *, struct subfacet *,
403 const struct dpif_flow_stats *);
404 static void subfacet_make_actions(struct ofproto_dpif *, struct subfacet *,
405 const struct ofpbuf *packet);
406 static int subfacet_install(struct ofproto_dpif *, struct subfacet *,
407 const struct nlattr *actions, size_t actions_len,
408 struct dpif_flow_stats *);
409 static void subfacet_uninstall(struct ofproto_dpif *, struct subfacet *);
415 struct ofbundle *bundle; /* Bundle that contains this port, if any. */
416 struct list bundle_node; /* In struct ofbundle's "ports" list. */
417 struct cfm *cfm; /* Connectivity Fault Management, if any. */
418 tag_type tag; /* Tag associated with this port. */
419 uint32_t bond_stable_id; /* stable_id to use as bond slave, or 0. */
420 bool may_enable; /* May be enabled in bonds. */
423 struct stp_port *stp_port; /* Spanning Tree Protocol, if any. */
424 enum stp_state stp_state; /* Always STP_DISABLED if STP not in use. */
425 long long int stp_state_entered;
427 struct hmap priorities; /* Map of attached 'priority_to_dscp's. */
429 /* Linux VLAN device support (e.g. "eth0.10" for VLAN 10.)
431 * This is deprecated. It is only for compatibility with broken device
432 * drivers in old versions of Linux that do not properly support VLANs when
433 * VLAN devices are not used. When broken device drivers are no longer in
434 * widespread use, we will delete these interfaces. */
435 uint16_t realdev_ofp_port;
439 /* Node in 'ofport_dpif''s 'priorities' map. Used to maintain a map from
440 * 'priority' (the datapath's term for QoS queue) to the dscp bits which all
441 * traffic egressing the 'ofport' with that priority should be marked with. */
442 struct priority_to_dscp {
443 struct hmap_node hmap_node; /* Node in 'ofport_dpif''s 'priorities' map. */
444 uint32_t priority; /* Priority of this queue (see struct flow). */
446 uint8_t dscp; /* DSCP bits to mark outgoing traffic with. */
449 /* Linux VLAN device support (e.g. "eth0.10" for VLAN 10.)
451 * This is deprecated. It is only for compatibility with broken device drivers
452 * in old versions of Linux that do not properly support VLANs when VLAN
453 * devices are not used. When broken device drivers are no longer in
454 * widespread use, we will delete these interfaces. */
455 struct vlan_splinter {
456 struct hmap_node realdev_vid_node;
457 struct hmap_node vlandev_node;
458 uint16_t realdev_ofp_port;
459 uint16_t vlandev_ofp_port;
463 static uint32_t vsp_realdev_to_vlandev(const struct ofproto_dpif *,
464 uint32_t realdev, ovs_be16 vlan_tci);
465 static uint16_t vsp_vlandev_to_realdev(const struct ofproto_dpif *,
466 uint16_t vlandev, int *vid);
467 static void vsp_remove(struct ofport_dpif *);
468 static void vsp_add(struct ofport_dpif *, uint16_t realdev_ofp_port, int vid);
470 static struct ofport_dpif *
471 ofport_dpif_cast(const struct ofport *ofport)
473 assert(ofport->ofproto->ofproto_class == &ofproto_dpif_class);
474 return ofport ? CONTAINER_OF(ofport, struct ofport_dpif, up) : NULL;
477 static void port_run(struct ofport_dpif *);
478 static void port_wait(struct ofport_dpif *);
479 static int set_cfm(struct ofport *, const struct cfm_settings *);
480 static void ofport_clear_priorities(struct ofport_dpif *);
482 struct dpif_completion {
483 struct list list_node;
484 struct ofoperation *op;
487 /* Extra information about a classifier table.
488 * Currently used just for optimized flow revalidation. */
490 /* If either of these is nonnull, then this table has a form that allows
491 * flows to be tagged to avoid revalidating most flows for the most common
492 * kinds of flow table changes. */
493 struct cls_table *catchall_table; /* Table that wildcards all fields. */
494 struct cls_table *other_table; /* Table with any other wildcard set. */
495 uint32_t basis; /* Keeps each table's tags separate. */
498 struct ofproto_dpif {
499 struct hmap_node all_ofproto_dpifs_node; /* In 'all_ofproto_dpifs'. */
508 struct netflow *netflow;
509 struct dpif_sflow *sflow;
510 struct hmap bundles; /* Contains "struct ofbundle"s. */
511 struct mac_learning *ml;
512 struct ofmirror *mirrors[MAX_MIRRORS];
513 bool has_bonded_bundles;
516 struct timer next_expiration;
520 struct hmap subfacets;
523 struct table_dpif tables[N_TABLES];
524 bool need_revalidate;
525 struct tag_set revalidate_set;
527 /* Support for debugging async flow mods. */
528 struct list completions;
530 bool has_bundle_action; /* True when the first bundle action appears. */
534 long long int stp_last_tick;
536 /* VLAN splinters. */
537 struct hmap realdev_vid_map; /* (realdev,vid) -> vlandev. */
538 struct hmap vlandev_map; /* vlandev -> (realdev,vid). */
541 /* Defer flow mod completion until "ovs-appctl ofproto/unclog"? (Useful only
542 * for debugging the asynchronous flow_mod implementation.) */
545 /* All existing ofproto_dpif instances, indexed by ->up.name. */
546 static struct hmap all_ofproto_dpifs = HMAP_INITIALIZER(&all_ofproto_dpifs);
548 static void ofproto_dpif_unixctl_init(void);
550 static struct ofproto_dpif *
551 ofproto_dpif_cast(const struct ofproto *ofproto)
553 assert(ofproto->ofproto_class == &ofproto_dpif_class);
554 return CONTAINER_OF(ofproto, struct ofproto_dpif, up);
557 static struct ofport_dpif *get_ofp_port(struct ofproto_dpif *,
559 static struct ofport_dpif *get_odp_port(struct ofproto_dpif *,
562 /* Packet processing. */
563 static void update_learning_table(struct ofproto_dpif *,
564 const struct flow *, int vlan,
567 #define FLOW_MISS_MAX_BATCH 50
568 static int handle_upcalls(struct ofproto_dpif *, unsigned int max_batch);
570 /* Flow expiration. */
571 static int expire(struct ofproto_dpif *);
574 static void send_netflow_active_timeouts(struct ofproto_dpif *);
577 static int send_packet(const struct ofport_dpif *, struct ofpbuf *packet);
579 compose_sflow_action(const struct ofproto_dpif *, struct ofpbuf *odp_actions,
580 const struct flow *, uint32_t odp_port);
581 static void add_mirror_actions(struct action_xlate_ctx *ctx,
582 const struct flow *flow);
583 /* Global variables. */
584 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
586 /* Factory functions. */
589 enumerate_types(struct sset *types)
591 dp_enumerate_types(types);
595 enumerate_names(const char *type, struct sset *names)
597 return dp_enumerate_names(type, names);
601 del(const char *type, const char *name)
606 error = dpif_open(name, type, &dpif);
608 error = dpif_delete(dpif);
614 /* Basic life-cycle. */
616 static struct ofproto *
619 struct ofproto_dpif *ofproto = xmalloc(sizeof *ofproto);
624 dealloc(struct ofproto *ofproto_)
626 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
631 construct(struct ofproto *ofproto_, int *n_tablesp)
633 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
634 const char *name = ofproto->up.name;
638 error = dpif_create_and_open(name, ofproto->up.type, &ofproto->dpif);
640 VLOG_ERR("failed to open datapath %s: %s", name, strerror(error));
644 ofproto->max_ports = dpif_get_max_ports(ofproto->dpif);
645 ofproto->n_matches = 0;
647 dpif_flow_flush(ofproto->dpif);
648 dpif_recv_purge(ofproto->dpif);
650 error = dpif_recv_set_mask(ofproto->dpif,
651 ((1u << DPIF_UC_MISS) |
652 (1u << DPIF_UC_ACTION)));
654 VLOG_ERR("failed to listen on datapath %s: %s", name, strerror(error));
655 dpif_close(ofproto->dpif);
659 ofproto->netflow = NULL;
660 ofproto->sflow = NULL;
662 hmap_init(&ofproto->bundles);
663 ofproto->ml = mac_learning_create(MAC_ENTRY_DEFAULT_IDLE_TIME);
664 for (i = 0; i < MAX_MIRRORS; i++) {
665 ofproto->mirrors[i] = NULL;
667 ofproto->has_bonded_bundles = false;
669 timer_set_duration(&ofproto->next_expiration, 1000);
671 hmap_init(&ofproto->facets);
672 hmap_init(&ofproto->subfacets);
674 for (i = 0; i < N_TABLES; i++) {
675 struct table_dpif *table = &ofproto->tables[i];
677 table->catchall_table = NULL;
678 table->other_table = NULL;
679 table->basis = random_uint32();
681 ofproto->need_revalidate = false;
682 tag_set_init(&ofproto->revalidate_set);
684 list_init(&ofproto->completions);
686 ofproto_dpif_unixctl_init();
688 ofproto->has_bundle_action = false;
690 hmap_init(&ofproto->vlandev_map);
691 hmap_init(&ofproto->realdev_vid_map);
693 hmap_insert(&all_ofproto_dpifs, &ofproto->all_ofproto_dpifs_node,
694 hash_string(ofproto->up.name, 0));
696 *n_tablesp = N_TABLES;
701 complete_operations(struct ofproto_dpif *ofproto)
703 struct dpif_completion *c, *next;
705 LIST_FOR_EACH_SAFE (c, next, list_node, &ofproto->completions) {
706 ofoperation_complete(c->op, 0);
707 list_remove(&c->list_node);
713 destruct(struct ofproto *ofproto_)
715 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
716 struct rule_dpif *rule, *next_rule;
717 struct classifier *table;
720 hmap_remove(&all_ofproto_dpifs, &ofproto->all_ofproto_dpifs_node);
721 complete_operations(ofproto);
723 OFPROTO_FOR_EACH_TABLE (table, &ofproto->up) {
724 struct cls_cursor cursor;
726 cls_cursor_init(&cursor, table, NULL);
727 CLS_CURSOR_FOR_EACH_SAFE (rule, next_rule, up.cr, &cursor) {
728 ofproto_rule_destroy(&rule->up);
732 for (i = 0; i < MAX_MIRRORS; i++) {
733 mirror_destroy(ofproto->mirrors[i]);
736 netflow_destroy(ofproto->netflow);
737 dpif_sflow_destroy(ofproto->sflow);
738 hmap_destroy(&ofproto->bundles);
739 mac_learning_destroy(ofproto->ml);
741 hmap_destroy(&ofproto->facets);
742 hmap_destroy(&ofproto->subfacets);
744 hmap_destroy(&ofproto->vlandev_map);
745 hmap_destroy(&ofproto->realdev_vid_map);
747 dpif_close(ofproto->dpif);
751 run_fast(struct ofproto *ofproto_)
753 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
756 /* Handle one or more batches of upcalls, until there's nothing left to do
757 * or until we do a fixed total amount of work.
759 * We do work in batches because it can be much cheaper to set up a number
760 * of flows and fire off their patches all at once. We do multiple batches
761 * because in some cases handling a packet can cause another packet to be
762 * queued almost immediately as part of the return flow. Both
763 * optimizations can make major improvements on some benchmarks and
764 * presumably for real traffic as well. */
766 while (work < FLOW_MISS_MAX_BATCH) {
767 int retval = handle_upcalls(ofproto, FLOW_MISS_MAX_BATCH - work);
777 run(struct ofproto *ofproto_)
779 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
780 struct ofport_dpif *ofport;
781 struct ofbundle *bundle;
785 complete_operations(ofproto);
787 dpif_run(ofproto->dpif);
789 error = run_fast(ofproto_);
794 if (timer_expired(&ofproto->next_expiration)) {
795 int delay = expire(ofproto);
796 timer_set_duration(&ofproto->next_expiration, delay);
799 if (ofproto->netflow) {
800 if (netflow_run(ofproto->netflow)) {
801 send_netflow_active_timeouts(ofproto);
804 if (ofproto->sflow) {
805 dpif_sflow_run(ofproto->sflow);
808 HMAP_FOR_EACH (ofport, up.hmap_node, &ofproto->up.ports) {
811 HMAP_FOR_EACH (bundle, hmap_node, &ofproto->bundles) {
816 mac_learning_run(ofproto->ml, &ofproto->revalidate_set);
818 /* Now revalidate if there's anything to do. */
819 if (ofproto->need_revalidate
820 || !tag_set_is_empty(&ofproto->revalidate_set)) {
821 struct tag_set revalidate_set = ofproto->revalidate_set;
822 bool revalidate_all = ofproto->need_revalidate;
823 struct facet *facet, *next;
825 /* Clear the revalidation flags. */
826 tag_set_init(&ofproto->revalidate_set);
827 ofproto->need_revalidate = false;
829 HMAP_FOR_EACH_SAFE (facet, next, hmap_node, &ofproto->facets) {
831 || tag_set_intersects(&revalidate_set, facet->tags)) {
832 facet_revalidate(ofproto, facet);
841 wait(struct ofproto *ofproto_)
843 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
844 struct ofport_dpif *ofport;
845 struct ofbundle *bundle;
847 if (!clogged && !list_is_empty(&ofproto->completions)) {
848 poll_immediate_wake();
851 dpif_wait(ofproto->dpif);
852 dpif_recv_wait(ofproto->dpif);
853 if (ofproto->sflow) {
854 dpif_sflow_wait(ofproto->sflow);
856 if (!tag_set_is_empty(&ofproto->revalidate_set)) {
857 poll_immediate_wake();
859 HMAP_FOR_EACH (ofport, up.hmap_node, &ofproto->up.ports) {
862 HMAP_FOR_EACH (bundle, hmap_node, &ofproto->bundles) {
865 if (ofproto->netflow) {
866 netflow_wait(ofproto->netflow);
868 mac_learning_wait(ofproto->ml);
870 if (ofproto->need_revalidate) {
871 /* Shouldn't happen, but if it does just go around again. */
872 VLOG_DBG_RL(&rl, "need revalidate in ofproto_wait_cb()");
873 poll_immediate_wake();
875 timer_wait(&ofproto->next_expiration);
880 flush(struct ofproto *ofproto_)
882 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
883 struct facet *facet, *next_facet;
885 HMAP_FOR_EACH_SAFE (facet, next_facet, hmap_node, &ofproto->facets) {
886 /* Mark the facet as not installed so that facet_remove() doesn't
887 * bother trying to uninstall it. There is no point in uninstalling it
888 * individually since we are about to blow away all the facets with
889 * dpif_flow_flush(). */
890 struct subfacet *subfacet;
892 LIST_FOR_EACH (subfacet, list_node, &facet->subfacets) {
893 subfacet->installed = false;
894 subfacet->dp_packet_count = 0;
895 subfacet->dp_byte_count = 0;
897 facet_remove(ofproto, facet);
899 dpif_flow_flush(ofproto->dpif);
903 get_features(struct ofproto *ofproto_ OVS_UNUSED,
904 bool *arp_match_ip, uint32_t *actions)
906 *arp_match_ip = true;
907 *actions = ((1u << OFPAT_OUTPUT) |
908 (1u << OFPAT_SET_VLAN_VID) |
909 (1u << OFPAT_SET_VLAN_PCP) |
910 (1u << OFPAT_STRIP_VLAN) |
911 (1u << OFPAT_SET_DL_SRC) |
912 (1u << OFPAT_SET_DL_DST) |
913 (1u << OFPAT_SET_NW_SRC) |
914 (1u << OFPAT_SET_NW_DST) |
915 (1u << OFPAT_SET_NW_TOS) |
916 (1u << OFPAT_SET_TP_SRC) |
917 (1u << OFPAT_SET_TP_DST) |
918 (1u << OFPAT_ENQUEUE));
922 get_tables(struct ofproto *ofproto_, struct ofp_table_stats *ots)
924 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
925 struct dpif_dp_stats s;
927 strcpy(ots->name, "classifier");
929 dpif_get_dp_stats(ofproto->dpif, &s);
930 put_32aligned_be64(&ots->lookup_count, htonll(s.n_hit + s.n_missed));
931 put_32aligned_be64(&ots->matched_count,
932 htonll(s.n_hit + ofproto->n_matches));
935 static struct ofport *
938 struct ofport_dpif *port = xmalloc(sizeof *port);
943 port_dealloc(struct ofport *port_)
945 struct ofport_dpif *port = ofport_dpif_cast(port_);
950 port_construct(struct ofport *port_)
952 struct ofport_dpif *port = ofport_dpif_cast(port_);
953 struct ofproto_dpif *ofproto = ofproto_dpif_cast(port->up.ofproto);
955 ofproto->need_revalidate = true;
956 port->odp_port = ofp_port_to_odp_port(port->up.ofp_port);
959 port->tag = tag_create_random();
960 port->may_enable = true;
961 port->stp_port = NULL;
962 port->stp_state = STP_DISABLED;
963 hmap_init(&port->priorities);
964 port->realdev_ofp_port = 0;
965 port->vlandev_vid = 0;
967 if (ofproto->sflow) {
968 dpif_sflow_add_port(ofproto->sflow, port->odp_port,
969 netdev_get_name(port->up.netdev));
976 port_destruct(struct ofport *port_)
978 struct ofport_dpif *port = ofport_dpif_cast(port_);
979 struct ofproto_dpif *ofproto = ofproto_dpif_cast(port->up.ofproto);
981 ofproto->need_revalidate = true;
982 bundle_remove(port_);
983 set_cfm(port_, NULL);
984 if (ofproto->sflow) {
985 dpif_sflow_del_port(ofproto->sflow, port->odp_port);
988 ofport_clear_priorities(port);
989 hmap_destroy(&port->priorities);
993 port_modified(struct ofport *port_)
995 struct ofport_dpif *port = ofport_dpif_cast(port_);
997 if (port->bundle && port->bundle->bond) {
998 bond_slave_set_netdev(port->bundle->bond, port, port->up.netdev);
1003 port_reconfigured(struct ofport *port_, ovs_be32 old_config)
1005 struct ofport_dpif *port = ofport_dpif_cast(port_);
1006 struct ofproto_dpif *ofproto = ofproto_dpif_cast(port->up.ofproto);
1007 ovs_be32 changed = old_config ^ port->up.opp.config;
1009 if (changed & htonl(OFPPC_NO_RECV | OFPPC_NO_RECV_STP |
1010 OFPPC_NO_FWD | OFPPC_NO_FLOOD)) {
1011 ofproto->need_revalidate = true;
1013 if (changed & htonl(OFPPC_NO_FLOOD) && port->bundle) {
1014 bundle_update(port->bundle);
1020 set_sflow(struct ofproto *ofproto_,
1021 const struct ofproto_sflow_options *sflow_options)
1023 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1024 struct dpif_sflow *ds = ofproto->sflow;
1026 if (sflow_options) {
1028 struct ofport_dpif *ofport;
1030 ds = ofproto->sflow = dpif_sflow_create(ofproto->dpif);
1031 HMAP_FOR_EACH (ofport, up.hmap_node, &ofproto->up.ports) {
1032 dpif_sflow_add_port(ds, ofport->odp_port,
1033 netdev_get_name(ofport->up.netdev));
1035 ofproto->need_revalidate = true;
1037 dpif_sflow_set_options(ds, sflow_options);
1040 dpif_sflow_destroy(ds);
1041 ofproto->need_revalidate = true;
1042 ofproto->sflow = NULL;
1049 set_cfm(struct ofport *ofport_, const struct cfm_settings *s)
1051 struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
1058 struct ofproto_dpif *ofproto;
1060 ofproto = ofproto_dpif_cast(ofport->up.ofproto);
1061 ofproto->need_revalidate = true;
1062 ofport->cfm = cfm_create(netdev_get_name(ofport->up.netdev));
1065 if (cfm_configure(ofport->cfm, s)) {
1071 cfm_destroy(ofport->cfm);
1077 get_cfm_fault(const struct ofport *ofport_)
1079 struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
1081 return ofport->cfm ? cfm_get_fault(ofport->cfm) : -1;
1085 get_cfm_remote_mpids(const struct ofport *ofport_, const uint64_t **rmps,
1088 struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
1091 cfm_get_remote_mpids(ofport->cfm, rmps, n_rmps);
1098 /* Spanning Tree. */
1101 send_bpdu_cb(struct ofpbuf *pkt, int port_num, void *ofproto_)
1103 struct ofproto_dpif *ofproto = ofproto_;
1104 struct stp_port *sp = stp_get_port(ofproto->stp, port_num);
1105 struct ofport_dpif *ofport;
1107 ofport = stp_port_get_aux(sp);
1109 VLOG_WARN_RL(&rl, "%s: cannot send BPDU on unknown port %d",
1110 ofproto->up.name, port_num);
1112 struct eth_header *eth = pkt->l2;
1114 netdev_get_etheraddr(ofport->up.netdev, eth->eth_src);
1115 if (eth_addr_is_zero(eth->eth_src)) {
1116 VLOG_WARN_RL(&rl, "%s: cannot send BPDU on port %d "
1117 "with unknown MAC", ofproto->up.name, port_num);
1119 send_packet(ofport, pkt);
1125 /* Configures STP on 'ofproto_' using the settings defined in 's'. */
1127 set_stp(struct ofproto *ofproto_, const struct ofproto_stp_settings *s)
1129 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1131 /* Only revalidate flows if the configuration changed. */
1132 if (!s != !ofproto->stp) {
1133 ofproto->need_revalidate = true;
1137 if (!ofproto->stp) {
1138 ofproto->stp = stp_create(ofproto_->name, s->system_id,
1139 send_bpdu_cb, ofproto);
1140 ofproto->stp_last_tick = time_msec();
1143 stp_set_bridge_id(ofproto->stp, s->system_id);
1144 stp_set_bridge_priority(ofproto->stp, s->priority);
1145 stp_set_hello_time(ofproto->stp, s->hello_time);
1146 stp_set_max_age(ofproto->stp, s->max_age);
1147 stp_set_forward_delay(ofproto->stp, s->fwd_delay);
1149 struct ofport *ofport;
1151 HMAP_FOR_EACH (ofport, hmap_node, &ofproto->up.ports) {
1152 set_stp_port(ofport, NULL);
1155 stp_destroy(ofproto->stp);
1156 ofproto->stp = NULL;
1163 get_stp_status(struct ofproto *ofproto_, struct ofproto_stp_status *s)
1165 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1169 s->bridge_id = stp_get_bridge_id(ofproto->stp);
1170 s->designated_root = stp_get_designated_root(ofproto->stp);
1171 s->root_path_cost = stp_get_root_path_cost(ofproto->stp);
1180 update_stp_port_state(struct ofport_dpif *ofport)
1182 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport->up.ofproto);
1183 enum stp_state state;
1185 /* Figure out new state. */
1186 state = ofport->stp_port ? stp_port_get_state(ofport->stp_port)
1190 if (ofport->stp_state != state) {
1194 VLOG_DBG_RL(&rl, "port %s: STP state changed from %s to %s",
1195 netdev_get_name(ofport->up.netdev),
1196 stp_state_name(ofport->stp_state),
1197 stp_state_name(state));
1198 if (stp_learn_in_state(ofport->stp_state)
1199 != stp_learn_in_state(state)) {
1200 /* xxx Learning action flows should also be flushed. */
1201 mac_learning_flush(ofproto->ml);
1203 fwd_change = stp_forward_in_state(ofport->stp_state)
1204 != stp_forward_in_state(state);
1206 ofproto->need_revalidate = true;
1207 ofport->stp_state = state;
1208 ofport->stp_state_entered = time_msec();
1210 if (fwd_change && ofport->bundle) {
1211 bundle_update(ofport->bundle);
1214 /* Update the STP state bits in the OpenFlow port description. */
1215 of_state = (ofport->up.opp.state & htonl(~OFPPS_STP_MASK))
1216 | htonl(state == STP_LISTENING ? OFPPS_STP_LISTEN
1217 : state == STP_LEARNING ? OFPPS_STP_LEARN
1218 : state == STP_FORWARDING ? OFPPS_STP_FORWARD
1219 : state == STP_BLOCKING ? OFPPS_STP_BLOCK
1221 ofproto_port_set_state(&ofport->up, of_state);
1225 /* Configures STP on 'ofport_' using the settings defined in 's'. The
1226 * caller is responsible for assigning STP port numbers and ensuring
1227 * there are no duplicates. */
1229 set_stp_port(struct ofport *ofport_,
1230 const struct ofproto_port_stp_settings *s)
1232 struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
1233 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport->up.ofproto);
1234 struct stp_port *sp = ofport->stp_port;
1236 if (!s || !s->enable) {
1238 ofport->stp_port = NULL;
1239 stp_port_disable(sp);
1240 update_stp_port_state(ofport);
1243 } else if (sp && stp_port_no(sp) != s->port_num
1244 && ofport == stp_port_get_aux(sp)) {
1245 /* The port-id changed, so disable the old one if it's not
1246 * already in use by another port. */
1247 stp_port_disable(sp);
1250 sp = ofport->stp_port = stp_get_port(ofproto->stp, s->port_num);
1251 stp_port_enable(sp);
1253 stp_port_set_aux(sp, ofport);
1254 stp_port_set_priority(sp, s->priority);
1255 stp_port_set_path_cost(sp, s->path_cost);
1257 update_stp_port_state(ofport);
1263 get_stp_port_status(struct ofport *ofport_,
1264 struct ofproto_port_stp_status *s)
1266 struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
1267 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport->up.ofproto);
1268 struct stp_port *sp = ofport->stp_port;
1270 if (!ofproto->stp || !sp) {
1276 s->port_id = stp_port_get_id(sp);
1277 s->state = stp_port_get_state(sp);
1278 s->sec_in_state = (time_msec() - ofport->stp_state_entered) / 1000;
1279 s->role = stp_port_get_role(sp);
1280 stp_port_get_counts(sp, &s->tx_count, &s->rx_count, &s->error_count);
1286 stp_run(struct ofproto_dpif *ofproto)
1289 long long int now = time_msec();
1290 long long int elapsed = now - ofproto->stp_last_tick;
1291 struct stp_port *sp;
1294 stp_tick(ofproto->stp, MIN(INT_MAX, elapsed));
1295 ofproto->stp_last_tick = now;
1297 while (stp_get_changed_port(ofproto->stp, &sp)) {
1298 struct ofport_dpif *ofport = stp_port_get_aux(sp);
1301 update_stp_port_state(ofport);
1308 stp_wait(struct ofproto_dpif *ofproto)
1311 poll_timer_wait(1000);
1315 /* Returns true if STP should process 'flow'. */
1317 stp_should_process_flow(const struct flow *flow)
1319 return eth_addr_equals(flow->dl_dst, eth_addr_stp);
1323 stp_process_packet(const struct ofport_dpif *ofport,
1324 const struct ofpbuf *packet)
1326 struct ofpbuf payload = *packet;
1327 struct eth_header *eth = payload.data;
1328 struct stp_port *sp = ofport->stp_port;
1330 /* Sink packets on ports that have STP disabled when the bridge has
1332 if (!sp || stp_port_get_state(sp) == STP_DISABLED) {
1336 /* Trim off padding on payload. */
1337 if (payload.size > ntohs(eth->eth_type) + ETH_HEADER_LEN) {
1338 payload.size = ntohs(eth->eth_type) + ETH_HEADER_LEN;
1341 if (ofpbuf_try_pull(&payload, ETH_HEADER_LEN + LLC_HEADER_LEN)) {
1342 stp_received_bpdu(sp, payload.data, payload.size);
1346 static struct priority_to_dscp *
1347 get_priority(const struct ofport_dpif *ofport, uint32_t priority)
1349 struct priority_to_dscp *pdscp;
1352 hash = hash_int(priority, 0);
1353 HMAP_FOR_EACH_IN_BUCKET (pdscp, hmap_node, hash, &ofport->priorities) {
1354 if (pdscp->priority == priority) {
1362 ofport_clear_priorities(struct ofport_dpif *ofport)
1364 struct priority_to_dscp *pdscp, *next;
1366 HMAP_FOR_EACH_SAFE (pdscp, next, hmap_node, &ofport->priorities) {
1367 hmap_remove(&ofport->priorities, &pdscp->hmap_node);
1373 set_queues(struct ofport *ofport_,
1374 const struct ofproto_port_queue *qdscp_list,
1377 struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
1378 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport->up.ofproto);
1379 struct hmap new = HMAP_INITIALIZER(&new);
1382 for (i = 0; i < n_qdscp; i++) {
1383 struct priority_to_dscp *pdscp;
1387 dscp = (qdscp_list[i].dscp << 2) & IP_DSCP_MASK;
1388 if (dpif_queue_to_priority(ofproto->dpif, qdscp_list[i].queue,
1393 pdscp = get_priority(ofport, priority);
1395 hmap_remove(&ofport->priorities, &pdscp->hmap_node);
1397 pdscp = xmalloc(sizeof *pdscp);
1398 pdscp->priority = priority;
1400 ofproto->need_revalidate = true;
1403 if (pdscp->dscp != dscp) {
1405 ofproto->need_revalidate = true;
1408 hmap_insert(&new, &pdscp->hmap_node, hash_int(pdscp->priority, 0));
1411 if (!hmap_is_empty(&ofport->priorities)) {
1412 ofport_clear_priorities(ofport);
1413 ofproto->need_revalidate = true;
1416 hmap_swap(&new, &ofport->priorities);
1424 /* Expires all MAC learning entries associated with 'bundle' and forces its
1425 * ofproto to revalidate every flow.
1427 * Normally MAC learning entries are removed only from the ofproto associated
1428 * with 'bundle', but if 'all_ofprotos' is true, then the MAC learning entries
1429 * are removed from every ofproto. When patch ports and SLB bonds are in use
1430 * and a VM migration happens and the gratuitous ARPs are somehow lost, this
1431 * avoids a MAC_ENTRY_IDLE_TIME delay before the migrated VM can communicate
1432 * with the host from which it migrated. */
1434 bundle_flush_macs(struct ofbundle *bundle, bool all_ofprotos)
1436 struct ofproto_dpif *ofproto = bundle->ofproto;
1437 struct mac_learning *ml = ofproto->ml;
1438 struct mac_entry *mac, *next_mac;
1440 ofproto->need_revalidate = true;
1441 LIST_FOR_EACH_SAFE (mac, next_mac, lru_node, &ml->lrus) {
1442 if (mac->port.p == bundle) {
1444 struct ofproto_dpif *o;
1446 HMAP_FOR_EACH (o, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
1448 struct mac_entry *e;
1450 e = mac_learning_lookup(o->ml, mac->mac, mac->vlan,
1453 tag_set_add(&o->revalidate_set, e->tag);
1454 mac_learning_expire(o->ml, e);
1460 mac_learning_expire(ml, mac);
1465 static struct ofbundle *
1466 bundle_lookup(const struct ofproto_dpif *ofproto, void *aux)
1468 struct ofbundle *bundle;
1470 HMAP_FOR_EACH_IN_BUCKET (bundle, hmap_node, hash_pointer(aux, 0),
1471 &ofproto->bundles) {
1472 if (bundle->aux == aux) {
1479 /* Looks up each of the 'n_auxes' pointers in 'auxes' as bundles and adds the
1480 * ones that are found to 'bundles'. */
1482 bundle_lookup_multiple(struct ofproto_dpif *ofproto,
1483 void **auxes, size_t n_auxes,
1484 struct hmapx *bundles)
1488 hmapx_init(bundles);
1489 for (i = 0; i < n_auxes; i++) {
1490 struct ofbundle *bundle = bundle_lookup(ofproto, auxes[i]);
1492 hmapx_add(bundles, bundle);
1498 bundle_update(struct ofbundle *bundle)
1500 struct ofport_dpif *port;
1502 bundle->floodable = true;
1503 LIST_FOR_EACH (port, bundle_node, &bundle->ports) {
1504 if (port->up.opp.config & htonl(OFPPC_NO_FLOOD)) {
1505 bundle->floodable = false;
1512 bundle_del_port(struct ofport_dpif *port)
1514 struct ofbundle *bundle = port->bundle;
1516 bundle->ofproto->need_revalidate = true;
1518 list_remove(&port->bundle_node);
1519 port->bundle = NULL;
1522 lacp_slave_unregister(bundle->lacp, port);
1525 bond_slave_unregister(bundle->bond, port);
1528 bundle_update(bundle);
1532 bundle_add_port(struct ofbundle *bundle, uint32_t ofp_port,
1533 struct lacp_slave_settings *lacp,
1534 uint32_t bond_stable_id)
1536 struct ofport_dpif *port;
1538 port = get_ofp_port(bundle->ofproto, ofp_port);
1543 if (port->bundle != bundle) {
1544 bundle->ofproto->need_revalidate = true;
1546 bundle_del_port(port);
1549 port->bundle = bundle;
1550 list_push_back(&bundle->ports, &port->bundle_node);
1551 if (port->up.opp.config & htonl(OFPPC_NO_FLOOD)) {
1552 bundle->floodable = false;
1556 port->bundle->ofproto->need_revalidate = true;
1557 lacp_slave_register(bundle->lacp, port, lacp);
1560 port->bond_stable_id = bond_stable_id;
1566 bundle_destroy(struct ofbundle *bundle)
1568 struct ofproto_dpif *ofproto;
1569 struct ofport_dpif *port, *next_port;
1576 ofproto = bundle->ofproto;
1577 for (i = 0; i < MAX_MIRRORS; i++) {
1578 struct ofmirror *m = ofproto->mirrors[i];
1580 if (m->out == bundle) {
1582 } else if (hmapx_find_and_delete(&m->srcs, bundle)
1583 || hmapx_find_and_delete(&m->dsts, bundle)) {
1584 ofproto->need_revalidate = true;
1589 LIST_FOR_EACH_SAFE (port, next_port, bundle_node, &bundle->ports) {
1590 bundle_del_port(port);
1593 bundle_flush_macs(bundle, true);
1594 hmap_remove(&ofproto->bundles, &bundle->hmap_node);
1596 free(bundle->trunks);
1597 lacp_destroy(bundle->lacp);
1598 bond_destroy(bundle->bond);
1603 bundle_set(struct ofproto *ofproto_, void *aux,
1604 const struct ofproto_bundle_settings *s)
1606 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1607 bool need_flush = false;
1608 struct ofport_dpif *port;
1609 struct ofbundle *bundle;
1610 unsigned long *trunks;
1616 bundle_destroy(bundle_lookup(ofproto, aux));
1620 assert(s->n_slaves == 1 || s->bond != NULL);
1621 assert((s->lacp != NULL) == (s->lacp_slaves != NULL));
1623 bundle = bundle_lookup(ofproto, aux);
1625 bundle = xmalloc(sizeof *bundle);
1627 bundle->ofproto = ofproto;
1628 hmap_insert(&ofproto->bundles, &bundle->hmap_node,
1629 hash_pointer(aux, 0));
1631 bundle->name = NULL;
1633 list_init(&bundle->ports);
1634 bundle->vlan_mode = PORT_VLAN_TRUNK;
1636 bundle->trunks = NULL;
1637 bundle->use_priority_tags = s->use_priority_tags;
1638 bundle->lacp = NULL;
1639 bundle->bond = NULL;
1641 bundle->floodable = true;
1643 bundle->src_mirrors = 0;
1644 bundle->dst_mirrors = 0;
1645 bundle->mirror_out = 0;
1648 if (!bundle->name || strcmp(s->name, bundle->name)) {
1650 bundle->name = xstrdup(s->name);
1655 if (!bundle->lacp) {
1656 ofproto->need_revalidate = true;
1657 bundle->lacp = lacp_create();
1659 lacp_configure(bundle->lacp, s->lacp);
1661 lacp_destroy(bundle->lacp);
1662 bundle->lacp = NULL;
1665 /* Update set of ports. */
1667 for (i = 0; i < s->n_slaves; i++) {
1668 if (!bundle_add_port(bundle, s->slaves[i],
1669 s->lacp ? &s->lacp_slaves[i] : NULL,
1670 s->bond_stable_ids ? s->bond_stable_ids[i] : 0)) {
1674 if (!ok || list_size(&bundle->ports) != s->n_slaves) {
1675 struct ofport_dpif *next_port;
1677 LIST_FOR_EACH_SAFE (port, next_port, bundle_node, &bundle->ports) {
1678 for (i = 0; i < s->n_slaves; i++) {
1679 if (s->slaves[i] == port->up.ofp_port) {
1684 bundle_del_port(port);
1688 assert(list_size(&bundle->ports) <= s->n_slaves);
1690 if (list_is_empty(&bundle->ports)) {
1691 bundle_destroy(bundle);
1695 /* Set VLAN tagging mode */
1696 if (s->vlan_mode != bundle->vlan_mode
1697 || s->use_priority_tags != bundle->use_priority_tags) {
1698 bundle->vlan_mode = s->vlan_mode;
1699 bundle->use_priority_tags = s->use_priority_tags;
1704 vlan = (s->vlan_mode == PORT_VLAN_TRUNK ? -1
1705 : s->vlan >= 0 && s->vlan <= 4095 ? s->vlan
1707 if (vlan != bundle->vlan) {
1708 bundle->vlan = vlan;
1712 /* Get trunked VLANs. */
1713 switch (s->vlan_mode) {
1714 case PORT_VLAN_ACCESS:
1718 case PORT_VLAN_TRUNK:
1719 trunks = (unsigned long *) s->trunks;
1722 case PORT_VLAN_NATIVE_UNTAGGED:
1723 case PORT_VLAN_NATIVE_TAGGED:
1724 if (vlan != 0 && (!s->trunks
1725 || !bitmap_is_set(s->trunks, vlan)
1726 || bitmap_is_set(s->trunks, 0))) {
1727 /* Force trunking the native VLAN and prohibit trunking VLAN 0. */
1729 trunks = bitmap_clone(s->trunks, 4096);
1731 trunks = bitmap_allocate1(4096);
1733 bitmap_set1(trunks, vlan);
1734 bitmap_set0(trunks, 0);
1736 trunks = (unsigned long *) s->trunks;
1743 if (!vlan_bitmap_equal(trunks, bundle->trunks)) {
1744 free(bundle->trunks);
1745 if (trunks == s->trunks) {
1746 bundle->trunks = vlan_bitmap_clone(trunks);
1748 bundle->trunks = trunks;
1753 if (trunks != s->trunks) {
1758 if (!list_is_short(&bundle->ports)) {
1759 bundle->ofproto->has_bonded_bundles = true;
1761 if (bond_reconfigure(bundle->bond, s->bond)) {
1762 ofproto->need_revalidate = true;
1765 bundle->bond = bond_create(s->bond);
1766 ofproto->need_revalidate = true;
1769 LIST_FOR_EACH (port, bundle_node, &bundle->ports) {
1770 bond_slave_register(bundle->bond, port, port->bond_stable_id,
1774 bond_destroy(bundle->bond);
1775 bundle->bond = NULL;
1778 /* If we changed something that would affect MAC learning, un-learn
1779 * everything on this port and force flow revalidation. */
1781 bundle_flush_macs(bundle, false);
1788 bundle_remove(struct ofport *port_)
1790 struct ofport_dpif *port = ofport_dpif_cast(port_);
1791 struct ofbundle *bundle = port->bundle;
1794 bundle_del_port(port);
1795 if (list_is_empty(&bundle->ports)) {
1796 bundle_destroy(bundle);
1797 } else if (list_is_short(&bundle->ports)) {
1798 bond_destroy(bundle->bond);
1799 bundle->bond = NULL;
1805 send_pdu_cb(void *port_, const void *pdu, size_t pdu_size)
1807 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 10);
1808 struct ofport_dpif *port = port_;
1809 uint8_t ea[ETH_ADDR_LEN];
1812 error = netdev_get_etheraddr(port->up.netdev, ea);
1814 struct ofpbuf packet;
1817 ofpbuf_init(&packet, 0);
1818 packet_pdu = eth_compose(&packet, eth_addr_lacp, ea, ETH_TYPE_LACP,
1820 memcpy(packet_pdu, pdu, pdu_size);
1822 send_packet(port, &packet);
1823 ofpbuf_uninit(&packet);
1825 VLOG_ERR_RL(&rl, "port %s: cannot obtain Ethernet address of iface "
1826 "%s (%s)", port->bundle->name,
1827 netdev_get_name(port->up.netdev), strerror(error));
1832 bundle_send_learning_packets(struct ofbundle *bundle)
1834 struct ofproto_dpif *ofproto = bundle->ofproto;
1835 int error, n_packets, n_errors;
1836 struct mac_entry *e;
1838 error = n_packets = n_errors = 0;
1839 LIST_FOR_EACH (e, lru_node, &ofproto->ml->lrus) {
1840 if (e->port.p != bundle) {
1841 struct ofpbuf *learning_packet;
1842 struct ofport_dpif *port;
1845 learning_packet = bond_compose_learning_packet(bundle->bond, e->mac,
1848 ret = send_packet(port, learning_packet);
1849 ofpbuf_delete(learning_packet);
1859 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
1860 VLOG_WARN_RL(&rl, "bond %s: %d errors sending %d gratuitous learning "
1861 "packets, last error was: %s",
1862 bundle->name, n_errors, n_packets, strerror(error));
1864 VLOG_DBG("bond %s: sent %d gratuitous learning packets",
1865 bundle->name, n_packets);
1870 bundle_run(struct ofbundle *bundle)
1873 lacp_run(bundle->lacp, send_pdu_cb);
1876 struct ofport_dpif *port;
1878 LIST_FOR_EACH (port, bundle_node, &bundle->ports) {
1879 bond_slave_set_may_enable(bundle->bond, port, port->may_enable);
1882 bond_run(bundle->bond, &bundle->ofproto->revalidate_set,
1883 lacp_negotiated(bundle->lacp));
1884 if (bond_should_send_learning_packets(bundle->bond)) {
1885 bundle_send_learning_packets(bundle);
1891 bundle_wait(struct ofbundle *bundle)
1894 lacp_wait(bundle->lacp);
1897 bond_wait(bundle->bond);
1904 mirror_scan(struct ofproto_dpif *ofproto)
1908 for (idx = 0; idx < MAX_MIRRORS; idx++) {
1909 if (!ofproto->mirrors[idx]) {
1916 static struct ofmirror *
1917 mirror_lookup(struct ofproto_dpif *ofproto, void *aux)
1921 for (i = 0; i < MAX_MIRRORS; i++) {
1922 struct ofmirror *mirror = ofproto->mirrors[i];
1923 if (mirror && mirror->aux == aux) {
1931 /* Update the 'dup_mirrors' member of each of the ofmirrors in 'ofproto'. */
1933 mirror_update_dups(struct ofproto_dpif *ofproto)
1937 for (i = 0; i < MAX_MIRRORS; i++) {
1938 struct ofmirror *m = ofproto->mirrors[i];
1941 m->dup_mirrors = MIRROR_MASK_C(1) << i;
1945 for (i = 0; i < MAX_MIRRORS; i++) {
1946 struct ofmirror *m1 = ofproto->mirrors[i];
1953 for (j = i + 1; j < MAX_MIRRORS; j++) {
1954 struct ofmirror *m2 = ofproto->mirrors[j];
1956 if (m2 && m1->out == m2->out && m1->out_vlan == m2->out_vlan) {
1957 m1->dup_mirrors |= MIRROR_MASK_C(1) << j;
1958 m2->dup_mirrors |= m1->dup_mirrors;
1965 mirror_set(struct ofproto *ofproto_, void *aux,
1966 const struct ofproto_mirror_settings *s)
1968 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1969 mirror_mask_t mirror_bit;
1970 struct ofbundle *bundle;
1971 struct ofmirror *mirror;
1972 struct ofbundle *out;
1973 struct hmapx srcs; /* Contains "struct ofbundle *"s. */
1974 struct hmapx dsts; /* Contains "struct ofbundle *"s. */
1977 mirror = mirror_lookup(ofproto, aux);
1979 mirror_destroy(mirror);
1985 idx = mirror_scan(ofproto);
1987 VLOG_WARN("bridge %s: maximum of %d port mirrors reached, "
1989 ofproto->up.name, MAX_MIRRORS, s->name);
1993 mirror = ofproto->mirrors[idx] = xzalloc(sizeof *mirror);
1994 mirror->ofproto = ofproto;
1997 mirror->out_vlan = -1;
1998 mirror->name = NULL;
2001 if (!mirror->name || strcmp(s->name, mirror->name)) {
2003 mirror->name = xstrdup(s->name);
2006 /* Get the new configuration. */
2007 if (s->out_bundle) {
2008 out = bundle_lookup(ofproto, s->out_bundle);
2010 mirror_destroy(mirror);
2016 out_vlan = s->out_vlan;
2018 bundle_lookup_multiple(ofproto, s->srcs, s->n_srcs, &srcs);
2019 bundle_lookup_multiple(ofproto, s->dsts, s->n_dsts, &dsts);
2021 /* If the configuration has not changed, do nothing. */
2022 if (hmapx_equals(&srcs, &mirror->srcs)
2023 && hmapx_equals(&dsts, &mirror->dsts)
2024 && vlan_bitmap_equal(mirror->vlans, s->src_vlans)
2025 && mirror->out == out
2026 && mirror->out_vlan == out_vlan)
2028 hmapx_destroy(&srcs);
2029 hmapx_destroy(&dsts);
2033 hmapx_swap(&srcs, &mirror->srcs);
2034 hmapx_destroy(&srcs);
2036 hmapx_swap(&dsts, &mirror->dsts);
2037 hmapx_destroy(&dsts);
2039 free(mirror->vlans);
2040 mirror->vlans = vlan_bitmap_clone(s->src_vlans);
2043 mirror->out_vlan = out_vlan;
2045 /* Update bundles. */
2046 mirror_bit = MIRROR_MASK_C(1) << mirror->idx;
2047 HMAP_FOR_EACH (bundle, hmap_node, &mirror->ofproto->bundles) {
2048 if (hmapx_contains(&mirror->srcs, bundle)) {
2049 bundle->src_mirrors |= mirror_bit;
2051 bundle->src_mirrors &= ~mirror_bit;
2054 if (hmapx_contains(&mirror->dsts, bundle)) {
2055 bundle->dst_mirrors |= mirror_bit;
2057 bundle->dst_mirrors &= ~mirror_bit;
2060 if (mirror->out == bundle) {
2061 bundle->mirror_out |= mirror_bit;
2063 bundle->mirror_out &= ~mirror_bit;
2067 ofproto->need_revalidate = true;
2068 mac_learning_flush(ofproto->ml);
2069 mirror_update_dups(ofproto);
2075 mirror_destroy(struct ofmirror *mirror)
2077 struct ofproto_dpif *ofproto;
2078 mirror_mask_t mirror_bit;
2079 struct ofbundle *bundle;
2085 ofproto = mirror->ofproto;
2086 ofproto->need_revalidate = true;
2087 mac_learning_flush(ofproto->ml);
2089 mirror_bit = MIRROR_MASK_C(1) << mirror->idx;
2090 HMAP_FOR_EACH (bundle, hmap_node, &ofproto->bundles) {
2091 bundle->src_mirrors &= ~mirror_bit;
2092 bundle->dst_mirrors &= ~mirror_bit;
2093 bundle->mirror_out &= ~mirror_bit;
2096 hmapx_destroy(&mirror->srcs);
2097 hmapx_destroy(&mirror->dsts);
2098 free(mirror->vlans);
2100 ofproto->mirrors[mirror->idx] = NULL;
2104 mirror_update_dups(ofproto);
2108 mirror_get_stats(struct ofproto *ofproto_, void *aux,
2109 uint64_t *packets, uint64_t *bytes)
2111 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
2112 struct ofmirror *mirror = mirror_lookup(ofproto, aux);
2115 *packets = *bytes = UINT64_MAX;
2119 *packets = mirror->packet_count;
2120 *bytes = mirror->byte_count;
2126 set_flood_vlans(struct ofproto *ofproto_, unsigned long *flood_vlans)
2128 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
2129 if (mac_learning_set_flood_vlans(ofproto->ml, flood_vlans)) {
2130 ofproto->need_revalidate = true;
2131 mac_learning_flush(ofproto->ml);
2137 is_mirror_output_bundle(const struct ofproto *ofproto_, void *aux)
2139 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
2140 struct ofbundle *bundle = bundle_lookup(ofproto, aux);
2141 return bundle && bundle->mirror_out != 0;
2145 forward_bpdu_changed(struct ofproto *ofproto_)
2147 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
2148 /* Revalidate cached flows whenever forward_bpdu option changes. */
2149 ofproto->need_revalidate = true;
2153 set_mac_idle_time(struct ofproto *ofproto_, unsigned int idle_time)
2155 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
2156 mac_learning_set_idle_time(ofproto->ml, idle_time);
2161 static struct ofport_dpif *
2162 get_ofp_port(struct ofproto_dpif *ofproto, uint16_t ofp_port)
2164 struct ofport *ofport = ofproto_get_port(&ofproto->up, ofp_port);
2165 return ofport ? ofport_dpif_cast(ofport) : NULL;
2168 static struct ofport_dpif *
2169 get_odp_port(struct ofproto_dpif *ofproto, uint32_t odp_port)
2171 return get_ofp_port(ofproto, odp_port_to_ofp_port(odp_port));
2175 ofproto_port_from_dpif_port(struct ofproto_port *ofproto_port,
2176 struct dpif_port *dpif_port)
2178 ofproto_port->name = dpif_port->name;
2179 ofproto_port->type = dpif_port->type;
2180 ofproto_port->ofp_port = odp_port_to_ofp_port(dpif_port->port_no);
2184 port_run(struct ofport_dpif *ofport)
2186 bool enable = netdev_get_carrier(ofport->up.netdev);
2189 cfm_run(ofport->cfm);
2191 if (cfm_should_send_ccm(ofport->cfm)) {
2192 struct ofpbuf packet;
2194 ofpbuf_init(&packet, 0);
2195 cfm_compose_ccm(ofport->cfm, &packet, ofport->up.opp.hw_addr);
2196 send_packet(ofport, &packet);
2197 ofpbuf_uninit(&packet);
2200 enable = enable && !cfm_get_fault(ofport->cfm)
2201 && cfm_get_opup(ofport->cfm);
2204 if (ofport->bundle) {
2205 enable = enable && lacp_slave_may_enable(ofport->bundle->lacp, ofport);
2208 if (ofport->may_enable != enable) {
2209 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport->up.ofproto);
2211 if (ofproto->has_bundle_action) {
2212 ofproto->need_revalidate = true;
2216 ofport->may_enable = enable;
2220 port_wait(struct ofport_dpif *ofport)
2223 cfm_wait(ofport->cfm);
2228 port_query_by_name(const struct ofproto *ofproto_, const char *devname,
2229 struct ofproto_port *ofproto_port)
2231 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
2232 struct dpif_port dpif_port;
2235 error = dpif_port_query_by_name(ofproto->dpif, devname, &dpif_port);
2237 ofproto_port_from_dpif_port(ofproto_port, &dpif_port);
2243 port_add(struct ofproto *ofproto_, struct netdev *netdev, uint16_t *ofp_portp)
2245 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
2249 error = dpif_port_add(ofproto->dpif, netdev, &odp_port);
2251 *ofp_portp = odp_port_to_ofp_port(odp_port);
2257 port_del(struct ofproto *ofproto_, uint16_t ofp_port)
2259 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
2262 error = dpif_port_del(ofproto->dpif, ofp_port_to_odp_port(ofp_port));
2264 struct ofport_dpif *ofport = get_ofp_port(ofproto, ofp_port);
2266 /* The caller is going to close ofport->up.netdev. If this is a
2267 * bonded port, then the bond is using that netdev, so remove it
2268 * from the bond. The client will need to reconfigure everything
2269 * after deleting ports, so then the slave will get re-added. */
2270 bundle_remove(&ofport->up);
2276 struct port_dump_state {
2277 struct dpif_port_dump dump;
2282 port_dump_start(const struct ofproto *ofproto_, void **statep)
2284 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
2285 struct port_dump_state *state;
2287 *statep = state = xmalloc(sizeof *state);
2288 dpif_port_dump_start(&state->dump, ofproto->dpif);
2289 state->done = false;
2294 port_dump_next(const struct ofproto *ofproto_ OVS_UNUSED, void *state_,
2295 struct ofproto_port *port)
2297 struct port_dump_state *state = state_;
2298 struct dpif_port dpif_port;
2300 if (dpif_port_dump_next(&state->dump, &dpif_port)) {
2301 ofproto_port_from_dpif_port(port, &dpif_port);
2304 int error = dpif_port_dump_done(&state->dump);
2306 return error ? error : EOF;
2311 port_dump_done(const struct ofproto *ofproto_ OVS_UNUSED, void *state_)
2313 struct port_dump_state *state = state_;
2316 dpif_port_dump_done(&state->dump);
2323 port_poll(const struct ofproto *ofproto_, char **devnamep)
2325 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
2326 return dpif_port_poll(ofproto->dpif, devnamep);
2330 port_poll_wait(const struct ofproto *ofproto_)
2332 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
2333 dpif_port_poll_wait(ofproto->dpif);
2337 port_is_lacp_current(const struct ofport *ofport_)
2339 const struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
2340 return (ofport->bundle && ofport->bundle->lacp
2341 ? lacp_slave_is_current(ofport->bundle->lacp, ofport)
2345 /* Upcall handling. */
2347 /* Flow miss batching.
2349 * Some dpifs implement operations faster when you hand them off in a batch.
2350 * To allow batching, "struct flow_miss" queues the dpif-related work needed
2351 * for a given flow. Each "struct flow_miss" corresponds to sending one or
2352 * more packets, plus possibly installing the flow in the dpif.
2354 * So far we only batch the operations that affect flow setup time the most.
2355 * It's possible to batch more than that, but the benefit might be minimal. */
2357 struct hmap_node hmap_node;
2359 enum odp_key_fitness key_fitness;
2360 const struct nlattr *key;
2362 ovs_be16 initial_tci;
2363 struct list packets;
2366 struct flow_miss_op {
2367 union dpif_op dpif_op;
2368 struct subfacet *subfacet;
2371 /* Sends an OFPT_PACKET_IN message for 'packet' of type OFPR_NO_MATCH to each
2372 * OpenFlow controller as necessary according to their individual
2375 * If 'clone' is true, the caller retains ownership of 'packet'. Otherwise,
2376 * ownership is transferred to this function. */
2378 send_packet_in_miss(struct ofproto_dpif *ofproto, struct ofpbuf *packet,
2379 const struct flow *flow, bool clone)
2381 struct ofputil_packet_in pin;
2383 pin.packet = packet;
2384 pin.in_port = flow->in_port;
2385 pin.reason = OFPR_NO_MATCH;
2386 pin.buffer_id = 0; /* not yet known */
2387 pin.send_len = 0; /* not used for flow table misses */
2388 connmgr_send_packet_in(ofproto->up.connmgr, &pin, flow,
2389 clone ? NULL : packet);
2392 /* Sends an OFPT_PACKET_IN message for 'packet' of type OFPR_ACTION to each
2393 * OpenFlow controller as necessary according to their individual
2396 * 'send_len' should be the number of bytes of 'packet' to send to the
2397 * controller, as specified in the action that caused the packet to be sent.
2399 * If 'clone' is true, the caller retains ownership of 'upcall->packet'.
2400 * Otherwise, ownership is transferred to this function. */
2402 send_packet_in_action(struct ofproto_dpif *ofproto, struct ofpbuf *packet,
2403 uint64_t userdata, const struct flow *flow, bool clone)
2405 struct ofputil_packet_in pin;
2406 struct user_action_cookie cookie;
2408 memcpy(&cookie, &userdata, sizeof(cookie));
2410 pin.packet = packet;
2411 pin.in_port = flow->in_port;
2412 pin.reason = OFPR_ACTION;
2413 pin.buffer_id = 0; /* not yet known */
2414 pin.send_len = cookie.data;
2415 connmgr_send_packet_in(ofproto->up.connmgr, &pin, flow,
2416 clone ? NULL : packet);
2420 process_special(struct ofproto_dpif *ofproto, const struct flow *flow,
2421 const struct ofpbuf *packet)
2423 struct ofport_dpif *ofport = get_ofp_port(ofproto, flow->in_port);
2429 if (ofport->cfm && cfm_should_process_flow(ofport->cfm, flow)) {
2431 cfm_process_heartbeat(ofport->cfm, packet);
2434 } else if (ofport->bundle && ofport->bundle->lacp
2435 && flow->dl_type == htons(ETH_TYPE_LACP)) {
2437 lacp_process_packet(ofport->bundle->lacp, ofport, packet);
2440 } else if (ofproto->stp && stp_should_process_flow(flow)) {
2442 stp_process_packet(ofport, packet);
2449 static struct flow_miss *
2450 flow_miss_create(struct hmap *todo, const struct flow *flow,
2451 enum odp_key_fitness key_fitness,
2452 const struct nlattr *key, size_t key_len,
2453 ovs_be16 initial_tci)
2455 uint32_t hash = flow_hash(flow, 0);
2456 struct flow_miss *miss;
2458 HMAP_FOR_EACH_WITH_HASH (miss, hmap_node, hash, todo) {
2459 if (flow_equal(&miss->flow, flow)) {
2464 miss = xmalloc(sizeof *miss);
2465 hmap_insert(todo, &miss->hmap_node, hash);
2467 miss->key_fitness = key_fitness;
2469 miss->key_len = key_len;
2470 miss->initial_tci = initial_tci;
2471 list_init(&miss->packets);
2476 handle_flow_miss(struct ofproto_dpif *ofproto, struct flow_miss *miss,
2477 struct flow_miss_op *ops, size_t *n_ops)
2479 const struct flow *flow = &miss->flow;
2480 struct ofpbuf *packet, *next_packet;
2481 struct subfacet *subfacet;
2482 struct facet *facet;
2484 facet = facet_lookup_valid(ofproto, flow);
2486 struct rule_dpif *rule;
2488 rule = rule_dpif_lookup(ofproto, flow, 0);
2490 /* Don't send a packet-in if OFPPC_NO_PACKET_IN asserted. */
2491 struct ofport_dpif *port = get_ofp_port(ofproto, flow->in_port);
2493 if (port->up.opp.config & htonl(OFPPC_NO_PACKET_IN)) {
2494 COVERAGE_INC(ofproto_dpif_no_packet_in);
2495 /* XXX install 'drop' flow entry */
2499 VLOG_WARN_RL(&rl, "packet-in on unknown port %"PRIu16,
2503 LIST_FOR_EACH_SAFE (packet, next_packet, list_node,
2505 list_remove(&packet->list_node);
2506 send_packet_in_miss(ofproto, packet, flow, false);
2512 facet = facet_create(rule, flow);
2515 subfacet = subfacet_create(ofproto, facet,
2516 miss->key_fitness, miss->key, miss->key_len,
2519 LIST_FOR_EACH_SAFE (packet, next_packet, list_node, &miss->packets) {
2520 struct dpif_flow_stats stats;
2522 list_remove(&packet->list_node);
2523 ofproto->n_matches++;
2525 if (facet->rule->up.cr.priority == FAIL_OPEN_PRIORITY) {
2527 * Extra-special case for fail-open mode.
2529 * We are in fail-open mode and the packet matched the fail-open
2530 * rule, but we are connected to a controller too. We should send
2531 * the packet up to the controller in the hope that it will try to
2532 * set up a flow and thereby allow us to exit fail-open.
2534 * See the top-level comment in fail-open.c for more information.
2536 send_packet_in_miss(ofproto, packet, flow, true);
2539 if (!facet->may_install || !subfacet->actions) {
2540 subfacet_make_actions(ofproto, subfacet, packet);
2543 /* Credit statistics to subfacet for this packet. We must do this now
2544 * because execute_controller_action() below may destroy 'packet'. */
2545 dpif_flow_stats_extract(&facet->flow, packet, &stats);
2546 subfacet_update_stats(ofproto, subfacet, &stats);
2548 if (!execute_controller_action(ofproto, &facet->flow,
2550 subfacet->actions_len, packet, true)) {
2551 struct flow_miss_op *op = &ops[(*n_ops)++];
2552 struct dpif_execute *execute = &op->dpif_op.execute;
2554 if (flow->vlan_tci != subfacet->initial_tci) {
2555 /* This packet was received on a VLAN splinter port. We added
2556 * a VLAN to the packet to make the packet resemble the flow,
2557 * but the actions were composed assuming that the packet
2558 * contained no VLAN. So, we must remove the VLAN header from
2559 * the packet before trying to execute the actions. */
2560 eth_pop_vlan(packet);
2563 op->subfacet = subfacet;
2564 execute->type = DPIF_OP_EXECUTE;
2565 execute->key = miss->key;
2566 execute->key_len = miss->key_len;
2568 = (facet->may_install
2570 : xmemdup(subfacet->actions, subfacet->actions_len));
2571 execute->actions_len = subfacet->actions_len;
2572 execute->packet = packet;
2576 if (facet->may_install && subfacet->key_fitness != ODP_FIT_TOO_LITTLE) {
2577 struct flow_miss_op *op = &ops[(*n_ops)++];
2578 struct dpif_flow_put *put = &op->dpif_op.flow_put;
2580 op->subfacet = subfacet;
2581 put->type = DPIF_OP_FLOW_PUT;
2582 put->flags = DPIF_FP_CREATE | DPIF_FP_MODIFY;
2583 put->key = miss->key;
2584 put->key_len = miss->key_len;
2585 put->actions = subfacet->actions;
2586 put->actions_len = subfacet->actions_len;
2591 /* Like odp_flow_key_to_flow(), this function converts the 'key_len' bytes of
2592 * OVS_KEY_ATTR_* attributes in 'key' to a flow structure in 'flow' and returns
2593 * an ODP_FIT_* value that indicates how well 'key' fits our expectations for
2594 * what a flow key should contain.
2596 * This function also includes some logic to help make VLAN splinters
2597 * transparent to the rest of the upcall processing logic. In particular, if
2598 * the extracted in_port is a VLAN splinter port, it replaces flow->in_port by
2599 * the "real" port, sets flow->vlan_tci correctly for the VLAN of the VLAN
2600 * splinter port, and pushes a VLAN header onto 'packet' (if it is nonnull).
2602 * Sets '*initial_tci' to the VLAN TCI with which the packet was really
2603 * received, that is, the actual VLAN TCI extracted by odp_flow_key_to_flow().
2604 * (This differs from the value returned in flow->vlan_tci only for packets
2605 * received on VLAN splinters.)
2607 static enum odp_key_fitness
2608 ofproto_dpif_extract_flow_key(const struct ofproto_dpif *ofproto,
2609 const struct nlattr *key, size_t key_len,
2610 struct flow *flow, ovs_be16 *initial_tci,
2611 struct ofpbuf *packet)
2613 enum odp_key_fitness fitness;
2617 fitness = odp_flow_key_to_flow(key, key_len, flow);
2618 if (fitness == ODP_FIT_ERROR) {
2621 *initial_tci = flow->vlan_tci;
2623 realdev = vsp_vlandev_to_realdev(ofproto, flow->in_port, &vid);
2625 /* Cause the flow to be processed as if it came in on the real device
2626 * with the VLAN device's VLAN ID. */
2627 flow->in_port = realdev;
2628 flow->vlan_tci = htons((vid & VLAN_VID_MASK) | VLAN_CFI);
2630 /* Make the packet resemble the flow, so that it gets sent to an
2631 * OpenFlow controller properly, so that it looks correct for
2632 * sFlow, and so that flow_extract() will get the correct vlan_tci
2633 * if it is called on 'packet'.
2635 * The allocated space inside 'packet' probably also contains
2636 * 'key', that is, both 'packet' and 'key' are probably part of a
2637 * struct dpif_upcall (see the large comment on that structure
2638 * definition), so pushing data on 'packet' is in general not a
2639 * good idea since it could overwrite 'key' or free it as a side
2640 * effect. However, it's OK in this special case because we know
2641 * that 'packet' is inside a Netlink attribute: pushing 4 bytes
2642 * will just overwrite the 4-byte "struct nlattr", which is fine
2643 * since we don't need that header anymore. */
2644 eth_push_vlan(packet, flow->vlan_tci);
2647 /* Let the caller know that we can't reproduce 'key' from 'flow'. */
2648 if (fitness == ODP_FIT_PERFECT) {
2649 fitness = ODP_FIT_TOO_MUCH;
2657 handle_miss_upcalls(struct ofproto_dpif *ofproto, struct dpif_upcall *upcalls,
2660 struct dpif_upcall *upcall;
2661 struct flow_miss *miss, *next_miss;
2662 struct flow_miss_op flow_miss_ops[FLOW_MISS_MAX_BATCH * 2];
2663 union dpif_op *dpif_ops[FLOW_MISS_MAX_BATCH * 2];
2672 /* Construct the to-do list.
2674 * This just amounts to extracting the flow from each packet and sticking
2675 * the packets that have the same flow in the same "flow_miss" structure so
2676 * that we can process them together. */
2678 for (upcall = upcalls; upcall < &upcalls[n_upcalls]; upcall++) {
2679 enum odp_key_fitness fitness;
2680 struct flow_miss *miss;
2681 ovs_be16 initial_tci;
2684 /* Obtain metadata and check userspace/kernel agreement on flow match,
2685 * then set 'flow''s header pointers. */
2686 fitness = ofproto_dpif_extract_flow_key(ofproto,
2687 upcall->key, upcall->key_len,
2688 &flow, &initial_tci,
2690 if (fitness == ODP_FIT_ERROR) {
2691 ofpbuf_delete(upcall->packet);
2694 flow_extract(upcall->packet, flow.priority, flow.tun_id,
2695 flow.in_port, &flow);
2697 /* Handle 802.1ag, LACP, and STP specially. */
2698 if (process_special(ofproto, &flow, upcall->packet)) {
2699 ofpbuf_delete(upcall->packet);
2700 ofproto->n_matches++;
2704 /* Add other packets to a to-do list. */
2705 miss = flow_miss_create(&todo, &flow, fitness,
2706 upcall->key, upcall->key_len, initial_tci);
2707 list_push_back(&miss->packets, &upcall->packet->list_node);
2710 /* Process each element in the to-do list, constructing the set of
2711 * operations to batch. */
2713 HMAP_FOR_EACH_SAFE (miss, next_miss, hmap_node, &todo) {
2714 handle_flow_miss(ofproto, miss, flow_miss_ops, &n_ops);
2715 ofpbuf_list_delete(&miss->packets);
2716 hmap_remove(&todo, &miss->hmap_node);
2719 assert(n_ops <= ARRAY_SIZE(flow_miss_ops));
2720 hmap_destroy(&todo);
2722 /* Execute batch. */
2723 for (i = 0; i < n_ops; i++) {
2724 dpif_ops[i] = &flow_miss_ops[i].dpif_op;
2726 dpif_operate(ofproto->dpif, dpif_ops, n_ops);
2728 /* Free memory and update facets. */
2729 for (i = 0; i < n_ops; i++) {
2730 struct flow_miss_op *op = &flow_miss_ops[i];
2731 struct dpif_execute *execute;
2732 struct dpif_flow_put *put;
2734 switch (op->dpif_op.type) {
2735 case DPIF_OP_EXECUTE:
2736 execute = &op->dpif_op.execute;
2737 if (op->subfacet->actions != execute->actions) {
2738 free((struct nlattr *) execute->actions);
2740 ofpbuf_delete((struct ofpbuf *) execute->packet);
2743 case DPIF_OP_FLOW_PUT:
2744 put = &op->dpif_op.flow_put;
2746 op->subfacet->installed = true;
2754 handle_userspace_upcall(struct ofproto_dpif *ofproto,
2755 struct dpif_upcall *upcall)
2757 struct user_action_cookie cookie;
2758 enum odp_key_fitness fitness;
2759 ovs_be16 initial_tci;
2762 memcpy(&cookie, &upcall->userdata, sizeof(cookie));
2764 fitness = ofproto_dpif_extract_flow_key(ofproto, upcall->key,
2765 upcall->key_len, &flow,
2766 &initial_tci, upcall->packet);
2767 if (fitness == ODP_FIT_ERROR) {
2768 ofpbuf_delete(upcall->packet);
2772 if (cookie.type == USER_ACTION_COOKIE_SFLOW) {
2773 if (ofproto->sflow) {
2774 dpif_sflow_received(ofproto->sflow, upcall->packet, &flow,
2777 ofpbuf_delete(upcall->packet);
2778 } else if (cookie.type == USER_ACTION_COOKIE_CONTROLLER) {
2779 COVERAGE_INC(ofproto_dpif_ctlr_action);
2780 send_packet_in_action(ofproto, upcall->packet, upcall->userdata,
2783 VLOG_WARN_RL(&rl, "invalid user cookie : 0x%"PRIx64, upcall->userdata);
2784 ofpbuf_delete(upcall->packet);
2789 handle_upcalls(struct ofproto_dpif *ofproto, unsigned int max_batch)
2791 struct dpif_upcall misses[FLOW_MISS_MAX_BATCH];
2795 assert (max_batch <= FLOW_MISS_MAX_BATCH);
2798 for (i = 0; i < max_batch; i++) {
2799 struct dpif_upcall *upcall = &misses[n_misses];
2802 error = dpif_recv(ofproto->dpif, upcall);
2807 switch (upcall->type) {
2808 case DPIF_UC_ACTION:
2809 handle_userspace_upcall(ofproto, upcall);
2813 /* Handle it later. */
2817 case DPIF_N_UC_TYPES:
2819 VLOG_WARN_RL(&rl, "upcall has unexpected type %"PRIu32,
2825 handle_miss_upcalls(ofproto, misses, n_misses);
2830 /* Flow expiration. */
2832 static int subfacet_max_idle(const struct ofproto_dpif *);
2833 static void update_stats(struct ofproto_dpif *);
2834 static void rule_expire(struct rule_dpif *);
2835 static void expire_subfacets(struct ofproto_dpif *, int dp_max_idle);
2837 /* This function is called periodically by run(). Its job is to collect
2838 * updates for the flows that have been installed into the datapath, most
2839 * importantly when they last were used, and then use that information to
2840 * expire flows that have not been used recently.
2842 * Returns the number of milliseconds after which it should be called again. */
2844 expire(struct ofproto_dpif *ofproto)
2846 struct rule_dpif *rule, *next_rule;
2847 struct classifier *table;
2850 /* Update stats for each flow in the datapath. */
2851 update_stats(ofproto);
2853 /* Expire subfacets that have been idle too long. */
2854 dp_max_idle = subfacet_max_idle(ofproto);
2855 expire_subfacets(ofproto, dp_max_idle);
2857 /* Expire OpenFlow flows whose idle_timeout or hard_timeout has passed. */
2858 OFPROTO_FOR_EACH_TABLE (table, &ofproto->up) {
2859 struct cls_cursor cursor;
2861 cls_cursor_init(&cursor, table, NULL);
2862 CLS_CURSOR_FOR_EACH_SAFE (rule, next_rule, up.cr, &cursor) {
2867 /* All outstanding data in existing flows has been accounted, so it's a
2868 * good time to do bond rebalancing. */
2869 if (ofproto->has_bonded_bundles) {
2870 struct ofbundle *bundle;
2872 HMAP_FOR_EACH (bundle, hmap_node, &ofproto->bundles) {
2874 bond_rebalance(bundle->bond, &ofproto->revalidate_set);
2879 return MIN(dp_max_idle, 1000);
2882 /* Update 'packet_count', 'byte_count', and 'used' members of installed facets.
2884 * This function also pushes statistics updates to rules which each facet
2885 * resubmits into. Generally these statistics will be accurate. However, if a
2886 * facet changes the rule it resubmits into at some time in between
2887 * update_stats() runs, it is possible that statistics accrued to the
2888 * old rule will be incorrectly attributed to the new rule. This could be
2889 * avoided by calling update_stats() whenever rules are created or
2890 * deleted. However, the performance impact of making so many calls to the
2891 * datapath do not justify the benefit of having perfectly accurate statistics.
2894 update_stats(struct ofproto_dpif *p)
2896 const struct dpif_flow_stats *stats;
2897 struct dpif_flow_dump dump;
2898 const struct nlattr *key;
2901 dpif_flow_dump_start(&dump, p->dpif);
2902 while (dpif_flow_dump_next(&dump, &key, &key_len, NULL, NULL, &stats)) {
2903 struct subfacet *subfacet;
2905 subfacet = subfacet_find(p, key, key_len);
2906 if (subfacet && subfacet->installed) {
2907 struct facet *facet = subfacet->facet;
2909 if (stats->n_packets >= subfacet->dp_packet_count) {
2910 uint64_t extra = stats->n_packets - subfacet->dp_packet_count;
2911 facet->packet_count += extra;
2913 VLOG_WARN_RL(&rl, "unexpected packet count from the datapath");
2916 if (stats->n_bytes >= subfacet->dp_byte_count) {
2917 facet->byte_count += stats->n_bytes - subfacet->dp_byte_count;
2919 VLOG_WARN_RL(&rl, "unexpected byte count from datapath");
2922 subfacet->dp_packet_count = stats->n_packets;
2923 subfacet->dp_byte_count = stats->n_bytes;
2925 subfacet_update_time(p, subfacet, stats->used);
2926 facet_account(p, facet, true);
2927 facet_push_stats(facet);
2929 if (!VLOG_DROP_WARN(&rl)) {
2933 odp_flow_key_format(key, key_len, &s);
2934 VLOG_WARN("unexpected flow from datapath %s", ds_cstr(&s));
2938 COVERAGE_INC(facet_unexpected);
2939 /* There's a flow in the datapath that we know nothing about, or a
2940 * flow that shouldn't be installed but was anyway. Delete it. */
2941 dpif_flow_del(p->dpif, key, key_len, NULL);
2944 dpif_flow_dump_done(&dump);
2947 /* Calculates and returns the number of milliseconds of idle time after which
2948 * subfacets should expire from the datapath. When a subfacet expires, we fold
2949 * its statistics into its facet, and when a facet's last subfacet expires, we
2950 * fold its statistic into its rule. */
2952 subfacet_max_idle(const struct ofproto_dpif *ofproto)
2955 * Idle time histogram.
2957 * Most of the time a switch has a relatively small number of subfacets.
2958 * When this is the case we might as well keep statistics for all of them
2959 * in userspace and to cache them in the kernel datapath for performance as
2962 * As the number of subfacets increases, the memory required to maintain
2963 * statistics about them in userspace and in the kernel becomes
2964 * significant. However, with a large number of subfacets it is likely
2965 * that only a few of them are "heavy hitters" that consume a large amount
2966 * of bandwidth. At this point, only heavy hitters are worth caching in
2967 * the kernel and maintaining in userspaces; other subfacets we can
2970 * The technique used to compute the idle time is to build a histogram with
2971 * N_BUCKETS buckets whose width is BUCKET_WIDTH msecs each. Each subfacet
2972 * that is installed in the kernel gets dropped in the appropriate bucket.
2973 * After the histogram has been built, we compute the cutoff so that only
2974 * the most-recently-used 1% of subfacets (but at least
2975 * ofproto->up.flow_eviction_threshold flows) are kept cached. At least
2976 * the most-recently-used bucket of subfacets is kept, so actually an
2977 * arbitrary number of subfacets can be kept in any given expiration run
2978 * (though the next run will delete most of those unless they receive
2981 * This requires a second pass through the subfacets, in addition to the
2982 * pass made by update_stats(), because the former function never looks at
2983 * uninstallable subfacets.
2985 enum { BUCKET_WIDTH = ROUND_UP(100, TIME_UPDATE_INTERVAL) };
2986 enum { N_BUCKETS = 5000 / BUCKET_WIDTH };
2987 int buckets[N_BUCKETS] = { 0 };
2988 int total, subtotal, bucket;
2989 struct subfacet *subfacet;
2993 total = hmap_count(&ofproto->subfacets);
2994 if (total <= ofproto->up.flow_eviction_threshold) {
2995 return N_BUCKETS * BUCKET_WIDTH;
2998 /* Build histogram. */
3000 HMAP_FOR_EACH (subfacet, hmap_node, &ofproto->subfacets) {
3001 long long int idle = now - subfacet->used;
3002 int bucket = (idle <= 0 ? 0
3003 : idle >= BUCKET_WIDTH * N_BUCKETS ? N_BUCKETS - 1
3004 : (unsigned int) idle / BUCKET_WIDTH);
3008 /* Find the first bucket whose flows should be expired. */
3009 subtotal = bucket = 0;
3011 subtotal += buckets[bucket++];
3012 } while (bucket < N_BUCKETS &&
3013 subtotal < MAX(ofproto->up.flow_eviction_threshold, total / 100));
3015 if (VLOG_IS_DBG_ENABLED()) {
3019 ds_put_cstr(&s, "keep");
3020 for (i = 0; i < N_BUCKETS; i++) {
3022 ds_put_cstr(&s, ", drop");
3025 ds_put_format(&s, " %d:%d", i * BUCKET_WIDTH, buckets[i]);
3028 VLOG_INFO("%s: %s (msec:count)", ofproto->up.name, ds_cstr(&s));
3032 return bucket * BUCKET_WIDTH;
3036 expire_subfacets(struct ofproto_dpif *ofproto, int dp_max_idle)
3038 long long int cutoff = time_msec() - dp_max_idle;
3039 struct subfacet *subfacet, *next_subfacet;
3041 HMAP_FOR_EACH_SAFE (subfacet, next_subfacet, hmap_node,
3042 &ofproto->subfacets) {
3043 if (subfacet->used < cutoff) {
3044 subfacet_destroy(ofproto, subfacet);
3049 /* If 'rule' is an OpenFlow rule, that has expired according to OpenFlow rules,
3050 * then delete it entirely. */
3052 rule_expire(struct rule_dpif *rule)
3054 struct ofproto_dpif *ofproto = ofproto_dpif_cast(rule->up.ofproto);
3055 struct facet *facet, *next_facet;
3059 /* Has 'rule' expired? */
3061 if (rule->up.hard_timeout
3062 && now > rule->up.modified + rule->up.hard_timeout * 1000) {
3063 reason = OFPRR_HARD_TIMEOUT;
3064 } else if (rule->up.idle_timeout && list_is_empty(&rule->facets)
3065 && now > rule->used + rule->up.idle_timeout * 1000) {
3066 reason = OFPRR_IDLE_TIMEOUT;
3071 COVERAGE_INC(ofproto_dpif_expired);
3073 /* Update stats. (This is a no-op if the rule expired due to an idle
3074 * timeout, because that only happens when the rule has no facets left.) */
3075 LIST_FOR_EACH_SAFE (facet, next_facet, list_node, &rule->facets) {
3076 facet_remove(ofproto, facet);
3079 /* Get rid of the rule. */
3080 ofproto_rule_expire(&rule->up, reason);
3085 /* Creates and returns a new facet owned by 'rule', given a 'flow'.
3087 * The caller must already have determined that no facet with an identical
3088 * 'flow' exists in 'ofproto' and that 'flow' is the best match for 'rule' in
3089 * the ofproto's classifier table.
3091 * The facet will initially have no subfacets. The caller should create (at
3092 * least) one subfacet with subfacet_create(). */
3093 static struct facet *
3094 facet_create(struct rule_dpif *rule, const struct flow *flow)
3096 struct ofproto_dpif *ofproto = ofproto_dpif_cast(rule->up.ofproto);
3097 struct facet *facet;
3099 facet = xzalloc(sizeof *facet);
3100 facet->used = time_msec();
3101 hmap_insert(&ofproto->facets, &facet->hmap_node, flow_hash(flow, 0));
3102 list_push_back(&rule->facets, &facet->list_node);
3104 facet->flow = *flow;
3105 list_init(&facet->subfacets);
3106 netflow_flow_init(&facet->nf_flow);
3107 netflow_flow_update_time(ofproto->netflow, &facet->nf_flow, facet->used);
3113 facet_free(struct facet *facet)
3118 /* If the 'actions_len' bytes of actions in 'odp_actions' are just a single
3119 * OVS_ACTION_ATTR_USERSPACE action, executes it internally and returns true.
3120 * Otherwise, returns false without doing anything.
3122 * If 'clone' is true, the caller always retains ownership of 'packet'.
3123 * Otherwise, ownership is transferred to this function if it returns true. */
3125 execute_controller_action(struct ofproto_dpif *ofproto,
3126 const struct flow *flow,
3127 const struct nlattr *odp_actions, size_t actions_len,
3128 struct ofpbuf *packet, bool clone)
3131 && odp_actions->nla_type == OVS_ACTION_ATTR_USERSPACE
3132 && NLA_ALIGN(odp_actions->nla_len) == actions_len) {
3133 /* As an optimization, avoid a round-trip from userspace to kernel to
3134 * userspace. This also avoids possibly filling up kernel packet
3135 * buffers along the way.
3137 * This optimization will not accidentally catch sFlow
3138 * OVS_ACTION_ATTR_USERSPACE actions, since those are encapsulated
3139 * inside OVS_ACTION_ATTR_SAMPLE. */
3140 const struct nlattr *nla;
3142 nla = nl_attr_find_nested(odp_actions, OVS_USERSPACE_ATTR_USERDATA);
3143 send_packet_in_action(ofproto, packet, nl_attr_get_u64(nla), flow,
3151 /* Executes, within 'ofproto', the 'n_actions' actions in 'actions' on
3152 * 'packet', which arrived on 'in_port'.
3154 * Takes ownership of 'packet'. */
3156 execute_odp_actions(struct ofproto_dpif *ofproto, const struct flow *flow,
3157 const struct nlattr *odp_actions, size_t actions_len,
3158 struct ofpbuf *packet)
3160 struct odputil_keybuf keybuf;
3164 if (execute_controller_action(ofproto, flow, odp_actions, actions_len,
3169 ofpbuf_use_stack(&key, &keybuf, sizeof keybuf);
3170 odp_flow_key_from_flow(&key, flow);
3172 error = dpif_execute(ofproto->dpif, key.data, key.size,
3173 odp_actions, actions_len, packet);
3175 ofpbuf_delete(packet);
3179 /* Remove 'facet' from 'ofproto' and free up the associated memory:
3181 * - If 'facet' was installed in the datapath, uninstalls it and updates its
3182 * rule's statistics, via subfacet_uninstall().
3184 * - Removes 'facet' from its rule and from ofproto->facets.
3187 facet_remove(struct ofproto_dpif *ofproto, struct facet *facet)
3189 struct subfacet *subfacet, *next_subfacet;
3191 assert(!list_is_empty(&facet->subfacets));
3193 /* First uninstall all of the subfacets to get final statistics. */
3194 LIST_FOR_EACH (subfacet, list_node, &facet->subfacets) {
3195 subfacet_uninstall(ofproto, subfacet);
3198 /* Flush the final stats to the rule.
3200 * This might require us to have at least one subfacet around so that we
3201 * can use its actions for accounting in facet_account(), which is why we
3202 * have uninstalled but not yet destroyed the subfacets. */
3203 facet_flush_stats(ofproto, facet);
3205 /* Now we're really all done so destroy everything. */
3206 LIST_FOR_EACH_SAFE (subfacet, next_subfacet, list_node,
3207 &facet->subfacets) {
3208 subfacet_destroy__(ofproto, subfacet);
3210 hmap_remove(&ofproto->facets, &facet->hmap_node);
3211 list_remove(&facet->list_node);
3216 facet_account(struct ofproto_dpif *ofproto, struct facet *facet,
3220 struct subfacet *subfacet;
3221 const struct nlattr *a;
3225 if (facet->byte_count <= facet->accounted_bytes) {
3228 n_bytes = facet->byte_count - facet->accounted_bytes;
3229 facet->accounted_bytes = facet->byte_count;
3231 /* Feed information from the active flows back into the learning table to
3232 * ensure that table is always in sync with what is actually flowing
3233 * through the datapath. */
3234 if (facet->has_learn || facet->has_normal) {
3235 struct action_xlate_ctx ctx;
3237 action_xlate_ctx_init(&ctx, ofproto, &facet->flow,
3238 facet->flow.vlan_tci, NULL);
3239 ctx.may_learn_macs = true;
3240 ctx.may_flow_mod = may_flow_mod;
3241 ofpbuf_delete(xlate_actions(&ctx, facet->rule->up.actions,
3242 facet->rule->up.n_actions));
3245 if (!facet->has_normal || !ofproto->has_bonded_bundles) {
3249 /* This loop feeds byte counters to bond_account() for rebalancing to use
3250 * as a basis. We also need to track the actual VLAN on which the packet
3251 * is going to be sent to ensure that it matches the one passed to
3252 * bond_choose_output_slave(). (Otherwise, we will account to the wrong
3255 * We use the actions from an arbitrary subfacet because they should all
3256 * be equally valid for our purpose. */
3257 subfacet = CONTAINER_OF(list_front(&facet->subfacets),
3258 struct subfacet, list_node);
3259 vlan_tci = facet->flow.vlan_tci;
3260 NL_ATTR_FOR_EACH_UNSAFE (a, left,
3261 subfacet->actions, subfacet->actions_len) {
3262 const struct ovs_action_push_vlan *vlan;
3263 struct ofport_dpif *port;
3265 switch (nl_attr_type(a)) {
3266 case OVS_ACTION_ATTR_OUTPUT:
3267 port = get_odp_port(ofproto, nl_attr_get_u32(a));
3268 if (port && port->bundle && port->bundle->bond) {
3269 bond_account(port->bundle->bond, &facet->flow,
3270 vlan_tci_to_vid(vlan_tci), n_bytes);
3274 case OVS_ACTION_ATTR_POP_VLAN:
3275 vlan_tci = htons(0);
3278 case OVS_ACTION_ATTR_PUSH_VLAN:
3279 vlan = nl_attr_get(a);
3280 vlan_tci = vlan->vlan_tci;
3286 /* Returns true if the only action for 'facet' is to send to the controller.
3287 * (We don't report NetFlow expiration messages for such facets because they
3288 * are just part of the control logic for the network, not real traffic). */
3290 facet_is_controller_flow(struct facet *facet)
3293 && facet->rule->up.n_actions == 1
3294 && action_outputs_to_port(&facet->rule->up.actions[0],
3295 htons(OFPP_CONTROLLER)));
3298 /* Folds all of 'facet''s statistics into its rule. Also updates the
3299 * accounting ofhook and emits a NetFlow expiration if appropriate. All of
3300 * 'facet''s statistics in the datapath should have been zeroed and folded into
3301 * its packet and byte counts before this function is called. */
3303 facet_flush_stats(struct ofproto_dpif *ofproto, struct facet *facet)
3305 struct subfacet *subfacet;
3307 LIST_FOR_EACH (subfacet, list_node, &facet->subfacets) {
3308 assert(!subfacet->dp_byte_count);
3309 assert(!subfacet->dp_packet_count);
3312 facet_push_stats(facet);
3313 facet_account(ofproto, facet, false);
3315 if (ofproto->netflow && !facet_is_controller_flow(facet)) {
3316 struct ofexpired expired;
3317 expired.flow = facet->flow;
3318 expired.packet_count = facet->packet_count;
3319 expired.byte_count = facet->byte_count;
3320 expired.used = facet->used;
3321 netflow_expire(ofproto->netflow, &facet->nf_flow, &expired);
3324 facet->rule->packet_count += facet->packet_count;
3325 facet->rule->byte_count += facet->byte_count;
3327 /* Reset counters to prevent double counting if 'facet' ever gets
3329 facet_reset_counters(facet);
3331 netflow_flow_clear(&facet->nf_flow);
3334 /* Searches 'ofproto''s table of facets for one exactly equal to 'flow'.
3335 * Returns it if found, otherwise a null pointer.
3337 * The returned facet might need revalidation; use facet_lookup_valid()
3338 * instead if that is important. */
3339 static struct facet *
3340 facet_find(struct ofproto_dpif *ofproto, const struct flow *flow)
3342 struct facet *facet;
3344 HMAP_FOR_EACH_WITH_HASH (facet, hmap_node, flow_hash(flow, 0),
3346 if (flow_equal(flow, &facet->flow)) {
3354 /* Searches 'ofproto''s table of facets for one exactly equal to 'flow'.
3355 * Returns it if found, otherwise a null pointer.
3357 * The returned facet is guaranteed to be valid. */
3358 static struct facet *
3359 facet_lookup_valid(struct ofproto_dpif *ofproto, const struct flow *flow)
3361 struct facet *facet = facet_find(ofproto, flow);
3363 /* The facet we found might not be valid, since we could be in need of
3364 * revalidation. If it is not valid, don't return it. */
3366 && (ofproto->need_revalidate
3367 || tag_set_intersects(&ofproto->revalidate_set, facet->tags))
3368 && !facet_revalidate(ofproto, facet)) {
3369 COVERAGE_INC(facet_invalidated);
3376 /* Re-searches 'ofproto''s classifier for a rule matching 'facet':
3378 * - If the rule found is different from 'facet''s current rule, moves
3379 * 'facet' to the new rule and recompiles its actions.
3381 * - If the rule found is the same as 'facet''s current rule, leaves 'facet'
3382 * where it is and recompiles its actions anyway.
3384 * - If there is none, destroys 'facet'.
3386 * Returns true if 'facet' still exists, false if it has been destroyed. */
3388 facet_revalidate(struct ofproto_dpif *ofproto, struct facet *facet)
3391 struct nlattr *odp_actions;
3394 struct actions *new_actions;
3396 struct action_xlate_ctx ctx;
3397 struct rule_dpif *new_rule;
3398 struct subfacet *subfacet;
3399 bool actions_changed;
3402 COVERAGE_INC(facet_revalidate);
3404 /* Determine the new rule. */
3405 new_rule = rule_dpif_lookup(ofproto, &facet->flow, 0);
3407 /* No new rule, so delete the facet. */
3408 facet_remove(ofproto, facet);
3412 /* Calculate new datapath actions.
3414 * We do not modify any 'facet' state yet, because we might need to, e.g.,
3415 * emit a NetFlow expiration and, if so, we need to have the old state
3416 * around to properly compose it. */
3418 /* If the datapath actions changed or the installability changed,
3419 * then we need to talk to the datapath. */
3422 memset(&ctx, 0, sizeof ctx);
3423 LIST_FOR_EACH (subfacet, list_node, &facet->subfacets) {
3424 struct ofpbuf *odp_actions;
3425 bool should_install;
3427 action_xlate_ctx_init(&ctx, ofproto, &facet->flow,
3428 subfacet->initial_tci, NULL);
3429 odp_actions = xlate_actions(&ctx, new_rule->up.actions,
3430 new_rule->up.n_actions);
3431 actions_changed = (subfacet->actions_len != odp_actions->size
3432 || memcmp(subfacet->actions, odp_actions->data,
3433 subfacet->actions_len));
3435 should_install = (ctx.may_set_up_flow
3436 && subfacet->key_fitness != ODP_FIT_TOO_LITTLE);
3437 if (actions_changed || should_install != subfacet->installed) {
3438 if (should_install) {
3439 struct dpif_flow_stats stats;
3441 subfacet_install(ofproto, subfacet,
3442 odp_actions->data, odp_actions->size, &stats);
3443 subfacet_update_stats(ofproto, subfacet, &stats);
3445 subfacet_uninstall(ofproto, subfacet);
3449 new_actions = xcalloc(list_size(&facet->subfacets),
3450 sizeof *new_actions);
3452 new_actions[i].odp_actions = xmemdup(odp_actions->data,
3454 new_actions[i].actions_len = odp_actions->size;
3457 ofpbuf_delete(odp_actions);
3461 facet_flush_stats(ofproto, facet);
3464 /* Update 'facet' now that we've taken care of all the old state. */
3465 facet->tags = ctx.tags;
3466 facet->nf_flow.output_iface = ctx.nf_output_iface;
3467 facet->may_install = ctx.may_set_up_flow;
3468 facet->has_learn = ctx.has_learn;
3469 facet->has_normal = ctx.has_normal;
3470 facet->mirrors = ctx.mirrors;
3473 LIST_FOR_EACH (subfacet, list_node, &facet->subfacets) {
3474 if (new_actions[i].odp_actions) {
3475 free(subfacet->actions);
3476 subfacet->actions = new_actions[i].odp_actions;
3477 subfacet->actions_len = new_actions[i].actions_len;
3483 if (facet->rule != new_rule) {
3484 COVERAGE_INC(facet_changed_rule);
3485 list_remove(&facet->list_node);
3486 list_push_back(&new_rule->facets, &facet->list_node);
3487 facet->rule = new_rule;
3488 facet->used = new_rule->up.created;
3489 facet->prev_used = facet->used;
3495 /* Updates 'facet''s used time. Caller is responsible for calling
3496 * facet_push_stats() to update the flows which 'facet' resubmits into. */
3498 facet_update_time(struct ofproto_dpif *ofproto, struct facet *facet,
3501 if (used > facet->used) {
3503 if (used > facet->rule->used) {
3504 facet->rule->used = used;
3506 netflow_flow_update_time(ofproto->netflow, &facet->nf_flow, used);
3511 facet_reset_counters(struct facet *facet)
3513 facet->packet_count = 0;
3514 facet->byte_count = 0;
3515 facet->prev_packet_count = 0;
3516 facet->prev_byte_count = 0;
3517 facet->accounted_bytes = 0;
3521 facet_push_stats(struct facet *facet)
3523 uint64_t new_packets, new_bytes;
3525 assert(facet->packet_count >= facet->prev_packet_count);
3526 assert(facet->byte_count >= facet->prev_byte_count);
3527 assert(facet->used >= facet->prev_used);
3529 new_packets = facet->packet_count - facet->prev_packet_count;
3530 new_bytes = facet->byte_count - facet->prev_byte_count;
3532 if (new_packets || new_bytes || facet->used > facet->prev_used) {
3533 facet->prev_packet_count = facet->packet_count;
3534 facet->prev_byte_count = facet->byte_count;
3535 facet->prev_used = facet->used;
3537 flow_push_stats(facet->rule, &facet->flow,
3538 new_packets, new_bytes, facet->used);
3540 update_mirror_stats(ofproto_dpif_cast(facet->rule->up.ofproto),
3541 facet->mirrors, new_packets, new_bytes);
3545 struct ofproto_push {
3546 struct action_xlate_ctx ctx;
3553 push_resubmit(struct action_xlate_ctx *ctx, struct rule_dpif *rule)
3555 struct ofproto_push *push = CONTAINER_OF(ctx, struct ofproto_push, ctx);
3558 rule->packet_count += push->packets;
3559 rule->byte_count += push->bytes;
3560 rule->used = MAX(push->used, rule->used);
3564 /* Pushes flow statistics to the rules which 'flow' resubmits into given
3565 * 'rule''s actions and mirrors. */
3567 flow_push_stats(const struct rule_dpif *rule,
3568 const struct flow *flow, uint64_t packets, uint64_t bytes,
3571 struct ofproto_dpif *ofproto = ofproto_dpif_cast(rule->up.ofproto);
3572 struct ofproto_push push;
3574 push.packets = packets;
3578 action_xlate_ctx_init(&push.ctx, ofproto, flow, flow->vlan_tci, NULL);
3579 push.ctx.resubmit_hook = push_resubmit;
3580 ofpbuf_delete(xlate_actions(&push.ctx,
3581 rule->up.actions, rule->up.n_actions));
3586 static struct subfacet *
3587 subfacet_find__(struct ofproto_dpif *ofproto,
3588 const struct nlattr *key, size_t key_len, uint32_t key_hash,
3589 const struct flow *flow)
3591 struct subfacet *subfacet;
3593 HMAP_FOR_EACH_WITH_HASH (subfacet, hmap_node, key_hash,
3594 &ofproto->subfacets) {
3596 ? (subfacet->key_len == key_len
3597 && !memcmp(key, subfacet->key, key_len))
3598 : flow_equal(flow, &subfacet->facet->flow)) {
3606 /* Searches 'facet' (within 'ofproto') for a subfacet with the specified
3607 * 'key_fitness', 'key', and 'key_len'. Returns the existing subfacet if
3608 * there is one, otherwise creates and returns a new subfacet.
3610 * If the returned subfacet is new, then subfacet->actions will be NULL, in
3611 * which case the caller must populate the actions with
3612 * subfacet_make_actions(). */
3613 static struct subfacet *
3614 subfacet_create(struct ofproto_dpif *ofproto, struct facet *facet,
3615 enum odp_key_fitness key_fitness,
3616 const struct nlattr *key, size_t key_len, ovs_be16 initial_tci)
3618 uint32_t key_hash = odp_flow_key_hash(key, key_len);
3619 struct subfacet *subfacet;
3621 subfacet = subfacet_find__(ofproto, key, key_len, key_hash, &facet->flow);
3623 if (subfacet->facet == facet) {
3627 /* This shouldn't happen. */
3628 VLOG_ERR_RL(&rl, "subfacet with wrong facet");
3629 subfacet_destroy(ofproto, subfacet);
3632 subfacet = xzalloc(sizeof *subfacet);
3633 hmap_insert(&ofproto->subfacets, &subfacet->hmap_node, key_hash);
3634 list_push_back(&facet->subfacets, &subfacet->list_node);
3635 subfacet->facet = facet;
3636 subfacet->used = time_msec();
3637 subfacet->key_fitness = key_fitness;
3638 if (key_fitness != ODP_FIT_PERFECT) {
3639 subfacet->key = xmemdup(key, key_len);
3640 subfacet->key_len = key_len;
3642 subfacet->installed = false;
3643 subfacet->initial_tci = initial_tci;
3648 /* Searches 'ofproto' for a subfacet with the given 'key', 'key_len', and
3649 * 'flow'. Returns the subfacet if one exists, otherwise NULL. */
3650 static struct subfacet *
3651 subfacet_find(struct ofproto_dpif *ofproto,
3652 const struct nlattr *key, size_t key_len)
3654 uint32_t key_hash = odp_flow_key_hash(key, key_len);
3655 enum odp_key_fitness fitness;
3658 fitness = odp_flow_key_to_flow(key, key_len, &flow);
3659 if (fitness == ODP_FIT_ERROR) {
3663 return subfacet_find__(ofproto, key, key_len, key_hash, &flow);
3666 /* Uninstalls 'subfacet' from the datapath, if it is installed, removes it from
3667 * its facet within 'ofproto', and frees it. */
3669 subfacet_destroy__(struct ofproto_dpif *ofproto, struct subfacet *subfacet)
3671 subfacet_uninstall(ofproto, subfacet);
3672 hmap_remove(&ofproto->subfacets, &subfacet->hmap_node);
3673 list_remove(&subfacet->list_node);
3674 free(subfacet->key);
3675 free(subfacet->actions);
3679 /* Destroys 'subfacet', as with subfacet_destroy__(), and then if this was the
3680 * last remaining subfacet in its facet destroys the facet too. */
3682 subfacet_destroy(struct ofproto_dpif *ofproto, struct subfacet *subfacet)
3684 struct facet *facet = subfacet->facet;
3686 if (list_is_singleton(&facet->subfacets)) {
3687 /* facet_remove() needs at least one subfacet (it will remove it). */
3688 facet_remove(ofproto, facet);
3690 subfacet_destroy__(ofproto, subfacet);
3694 /* Initializes 'key' with the sequence of OVS_KEY_ATTR_* Netlink attributes
3695 * that can be used to refer to 'subfacet'. The caller must provide 'keybuf'
3696 * for use as temporary storage. */
3698 subfacet_get_key(struct subfacet *subfacet, struct odputil_keybuf *keybuf,
3701 if (!subfacet->key) {
3702 ofpbuf_use_stack(key, keybuf, sizeof *keybuf);
3703 odp_flow_key_from_flow(key, &subfacet->facet->flow);
3705 ofpbuf_use_const(key, subfacet->key, subfacet->key_len);
3709 /* Composes the datapath actions for 'subfacet' based on its rule's actions. */
3711 subfacet_make_actions(struct ofproto_dpif *p, struct subfacet *subfacet,
3712 const struct ofpbuf *packet)
3714 struct facet *facet = subfacet->facet;
3715 const struct rule_dpif *rule = facet->rule;
3716 struct ofpbuf *odp_actions;
3717 struct action_xlate_ctx ctx;
3719 action_xlate_ctx_init(&ctx, p, &facet->flow, subfacet->initial_tci,
3721 odp_actions = xlate_actions(&ctx, rule->up.actions, rule->up.n_actions);
3722 facet->tags = ctx.tags;
3723 facet->may_install = ctx.may_set_up_flow;
3724 facet->has_learn = ctx.has_learn;
3725 facet->has_normal = ctx.has_normal;
3726 facet->nf_flow.output_iface = ctx.nf_output_iface;
3727 facet->mirrors = ctx.mirrors;
3729 if (subfacet->actions_len != odp_actions->size
3730 || memcmp(subfacet->actions, odp_actions->data, odp_actions->size)) {
3731 free(subfacet->actions);
3732 subfacet->actions_len = odp_actions->size;
3733 subfacet->actions = xmemdup(odp_actions->data, odp_actions->size);
3736 ofpbuf_delete(odp_actions);
3739 /* Updates 'subfacet''s datapath flow, setting its actions to 'actions_len'
3740 * bytes of actions in 'actions'. If 'stats' is non-null, statistics counters
3741 * in the datapath will be zeroed and 'stats' will be updated with traffic new
3742 * since 'subfacet' was last updated.
3744 * Returns 0 if successful, otherwise a positive errno value. */
3746 subfacet_install(struct ofproto_dpif *ofproto, struct subfacet *subfacet,
3747 const struct nlattr *actions, size_t actions_len,
3748 struct dpif_flow_stats *stats)
3750 struct odputil_keybuf keybuf;
3751 enum dpif_flow_put_flags flags;
3755 flags = DPIF_FP_CREATE | DPIF_FP_MODIFY;
3757 flags |= DPIF_FP_ZERO_STATS;
3760 subfacet_get_key(subfacet, &keybuf, &key);
3761 ret = dpif_flow_put(ofproto->dpif, flags, key.data, key.size,
3762 actions, actions_len, stats);
3765 subfacet_reset_dp_stats(subfacet, stats);
3771 /* If 'subfacet' is installed in the datapath, uninstalls it. */
3773 subfacet_uninstall(struct ofproto_dpif *p, struct subfacet *subfacet)
3775 if (subfacet->installed) {
3776 struct odputil_keybuf keybuf;
3777 struct dpif_flow_stats stats;
3781 subfacet_get_key(subfacet, &keybuf, &key);
3782 error = dpif_flow_del(p->dpif, key.data, key.size, &stats);
3783 subfacet_reset_dp_stats(subfacet, &stats);
3785 subfacet_update_stats(p, subfacet, &stats);
3787 subfacet->installed = false;
3789 assert(subfacet->dp_packet_count == 0);
3790 assert(subfacet->dp_byte_count == 0);
3794 /* Resets 'subfacet''s datapath statistics counters. This should be called
3795 * when 'subfacet''s statistics are cleared in the datapath. If 'stats' is
3796 * non-null, it should contain the statistics returned by dpif when 'subfacet'
3797 * was reset in the datapath. 'stats' will be modified to include only
3798 * statistics new since 'subfacet' was last updated. */
3800 subfacet_reset_dp_stats(struct subfacet *subfacet,
3801 struct dpif_flow_stats *stats)
3804 && subfacet->dp_packet_count <= stats->n_packets
3805 && subfacet->dp_byte_count <= stats->n_bytes) {
3806 stats->n_packets -= subfacet->dp_packet_count;
3807 stats->n_bytes -= subfacet->dp_byte_count;
3810 subfacet->dp_packet_count = 0;
3811 subfacet->dp_byte_count = 0;
3814 /* Updates 'subfacet''s used time. The caller is responsible for calling
3815 * facet_push_stats() to update the flows which 'subfacet' resubmits into. */
3817 subfacet_update_time(struct ofproto_dpif *ofproto, struct subfacet *subfacet,
3820 if (used > subfacet->used) {
3821 subfacet->used = used;
3822 facet_update_time(ofproto, subfacet->facet, used);
3826 /* Folds the statistics from 'stats' into the counters in 'subfacet'.
3828 * Because of the meaning of a subfacet's counters, it only makes sense to do
3829 * this if 'stats' are not tracked in the datapath, that is, if 'stats'
3830 * represents a packet that was sent by hand or if it represents statistics
3831 * that have been cleared out of the datapath. */
3833 subfacet_update_stats(struct ofproto_dpif *ofproto, struct subfacet *subfacet,
3834 const struct dpif_flow_stats *stats)
3836 if (stats->n_packets || stats->used > subfacet->used) {
3837 struct facet *facet = subfacet->facet;
3839 subfacet_update_time(ofproto, subfacet, stats->used);
3840 facet->packet_count += stats->n_packets;
3841 facet->byte_count += stats->n_bytes;
3842 facet_push_stats(facet);
3843 netflow_flow_update_flags(&facet->nf_flow, stats->tcp_flags);
3849 static struct rule_dpif *
3850 rule_dpif_lookup(struct ofproto_dpif *ofproto, const struct flow *flow,
3853 struct cls_rule *cls_rule;
3854 struct classifier *cls;
3856 if (table_id >= N_TABLES) {
3860 cls = &ofproto->up.tables[table_id];
3861 if (flow->nw_frag & FLOW_NW_FRAG_ANY
3862 && ofproto->up.frag_handling == OFPC_FRAG_NORMAL) {
3863 /* For OFPC_NORMAL frag_handling, we must pretend that transport ports
3864 * are unavailable. */
3865 struct flow ofpc_normal_flow = *flow;
3866 ofpc_normal_flow.tp_src = htons(0);
3867 ofpc_normal_flow.tp_dst = htons(0);
3868 cls_rule = classifier_lookup(cls, &ofpc_normal_flow);
3870 cls_rule = classifier_lookup(cls, flow);
3872 return rule_dpif_cast(rule_from_cls_rule(cls_rule));
3876 complete_operation(struct rule_dpif *rule)
3878 struct ofproto_dpif *ofproto = ofproto_dpif_cast(rule->up.ofproto);
3880 rule_invalidate(rule);
3882 struct dpif_completion *c = xmalloc(sizeof *c);
3883 c->op = rule->up.pending;
3884 list_push_back(&ofproto->completions, &c->list_node);
3886 ofoperation_complete(rule->up.pending, 0);
3890 static struct rule *
3893 struct rule_dpif *rule = xmalloc(sizeof *rule);
3898 rule_dealloc(struct rule *rule_)
3900 struct rule_dpif *rule = rule_dpif_cast(rule_);
3905 rule_construct(struct rule *rule_)
3907 struct rule_dpif *rule = rule_dpif_cast(rule_);
3908 struct ofproto_dpif *ofproto = ofproto_dpif_cast(rule->up.ofproto);
3909 struct rule_dpif *victim;
3913 error = validate_actions(rule->up.actions, rule->up.n_actions,
3914 &rule->up.cr.flow, ofproto->max_ports);
3919 rule->used = rule->up.created;
3920 rule->packet_count = 0;
3921 rule->byte_count = 0;
3923 victim = rule_dpif_cast(ofoperation_get_victim(rule->up.pending));
3924 if (victim && !list_is_empty(&victim->facets)) {
3925 struct facet *facet;
3927 rule->facets = victim->facets;
3928 list_moved(&rule->facets);
3929 LIST_FOR_EACH (facet, list_node, &rule->facets) {
3930 /* XXX: We're only clearing our local counters here. It's possible
3931 * that quite a few packets are unaccounted for in the datapath
3932 * statistics. These will be accounted to the new rule instead of
3933 * cleared as required. This could be fixed by clearing out the
3934 * datapath statistics for this facet, but currently it doesn't
3936 facet_reset_counters(facet);
3940 /* Must avoid list_moved() in this case. */
3941 list_init(&rule->facets);
3944 table_id = rule->up.table_id;
3945 rule->tag = (victim ? victim->tag
3947 : rule_calculate_tag(&rule->up.cr.flow, &rule->up.cr.wc,
3948 ofproto->tables[table_id].basis));
3950 complete_operation(rule);
3955 rule_destruct(struct rule *rule_)
3957 struct rule_dpif *rule = rule_dpif_cast(rule_);
3958 struct ofproto_dpif *ofproto = ofproto_dpif_cast(rule->up.ofproto);
3959 struct facet *facet, *next_facet;
3961 LIST_FOR_EACH_SAFE (facet, next_facet, list_node, &rule->facets) {
3962 facet_revalidate(ofproto, facet);
3965 complete_operation(rule);
3969 rule_get_stats(struct rule *rule_, uint64_t *packets, uint64_t *bytes)
3971 struct rule_dpif *rule = rule_dpif_cast(rule_);
3972 struct facet *facet;
3974 /* Start from historical data for 'rule' itself that are no longer tracked
3975 * in facets. This counts, for example, facets that have expired. */
3976 *packets = rule->packet_count;
3977 *bytes = rule->byte_count;
3979 /* Add any statistics that are tracked by facets. This includes
3980 * statistical data recently updated by ofproto_update_stats() as well as
3981 * stats for packets that were executed "by hand" via dpif_execute(). */
3982 LIST_FOR_EACH (facet, list_node, &rule->facets) {
3983 *packets += facet->packet_count;
3984 *bytes += facet->byte_count;
3989 rule_execute(struct rule *rule_, const struct flow *flow,
3990 struct ofpbuf *packet)
3992 struct rule_dpif *rule = rule_dpif_cast(rule_);
3993 struct ofproto_dpif *ofproto = ofproto_dpif_cast(rule->up.ofproto);
3994 struct action_xlate_ctx ctx;
3995 struct ofpbuf *odp_actions;
3998 action_xlate_ctx_init(&ctx, ofproto, flow, flow->vlan_tci, packet);
3999 odp_actions = xlate_actions(&ctx, rule->up.actions, rule->up.n_actions);
4000 size = packet->size;
4001 if (execute_odp_actions(ofproto, flow, odp_actions->data,
4002 odp_actions->size, packet)) {
4003 rule->used = time_msec();
4004 rule->packet_count++;
4005 rule->byte_count += size;
4006 flow_push_stats(rule, flow, 1, size, rule->used);
4008 ofpbuf_delete(odp_actions);
4014 rule_modify_actions(struct rule *rule_)
4016 struct rule_dpif *rule = rule_dpif_cast(rule_);
4017 struct ofproto_dpif *ofproto = ofproto_dpif_cast(rule->up.ofproto);
4020 error = validate_actions(rule->up.actions, rule->up.n_actions,
4021 &rule->up.cr.flow, ofproto->max_ports);
4023 ofoperation_complete(rule->up.pending, error);
4027 complete_operation(rule);
4030 /* Sends 'packet' out 'ofport'.
4031 * May modify 'packet'.
4032 * Returns 0 if successful, otherwise a positive errno value. */
4034 send_packet(const struct ofport_dpif *ofport, struct ofpbuf *packet)
4036 const struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport->up.ofproto);
4037 struct ofpbuf key, odp_actions;
4038 struct odputil_keybuf keybuf;
4043 flow_extract((struct ofpbuf *) packet, 0, 0, 0, &flow);
4044 odp_port = vsp_realdev_to_vlandev(ofproto, ofport->odp_port,
4046 if (odp_port != ofport->odp_port) {
4047 eth_pop_vlan(packet);
4048 flow.vlan_tci = htons(0);
4051 ofpbuf_use_stack(&key, &keybuf, sizeof keybuf);
4052 odp_flow_key_from_flow(&key, &flow);
4054 ofpbuf_init(&odp_actions, 32);
4055 compose_sflow_action(ofproto, &odp_actions, &flow, odp_port);
4057 nl_msg_put_u32(&odp_actions, OVS_ACTION_ATTR_OUTPUT, odp_port);
4058 error = dpif_execute(ofproto->dpif,
4060 odp_actions.data, odp_actions.size,
4062 ofpbuf_uninit(&odp_actions);
4065 VLOG_WARN_RL(&rl, "%s: failed to send packet on port %"PRIu32" (%s)",
4066 ofproto->up.name, odp_port, strerror(error));
4071 /* OpenFlow to datapath action translation. */
4073 static void do_xlate_actions(const union ofp_action *in, size_t n_in,
4074 struct action_xlate_ctx *ctx);
4075 static void xlate_normal(struct action_xlate_ctx *);
4078 put_userspace_action(const struct ofproto_dpif *ofproto,
4079 struct ofpbuf *odp_actions,
4080 const struct flow *flow,
4081 const struct user_action_cookie *cookie)
4085 pid = dpif_port_get_pid(ofproto->dpif,
4086 ofp_port_to_odp_port(flow->in_port));
4088 return odp_put_userspace_action(pid, cookie, odp_actions);
4091 /* Compose SAMPLE action for sFlow. */
4093 compose_sflow_action(const struct ofproto_dpif *ofproto,
4094 struct ofpbuf *odp_actions,
4095 const struct flow *flow,
4098 uint32_t port_ifindex;
4099 uint32_t probability;
4100 struct user_action_cookie cookie;
4101 size_t sample_offset, actions_offset;
4102 int cookie_offset, n_output;
4104 if (!ofproto->sflow || flow->in_port == OFPP_NONE) {
4108 if (odp_port == OVSP_NONE) {
4112 port_ifindex = dpif_sflow_odp_port_to_ifindex(ofproto->sflow, odp_port);
4116 sample_offset = nl_msg_start_nested(odp_actions, OVS_ACTION_ATTR_SAMPLE);
4118 /* Number of packets out of UINT_MAX to sample. */
4119 probability = dpif_sflow_get_probability(ofproto->sflow);
4120 nl_msg_put_u32(odp_actions, OVS_SAMPLE_ATTR_PROBABILITY, probability);
4122 actions_offset = nl_msg_start_nested(odp_actions, OVS_SAMPLE_ATTR_ACTIONS);
4124 cookie.type = USER_ACTION_COOKIE_SFLOW;
4125 cookie.data = port_ifindex;
4126 cookie.n_output = n_output;
4127 cookie.vlan_tci = 0;
4128 cookie_offset = put_userspace_action(ofproto, odp_actions, flow, &cookie);
4130 nl_msg_end_nested(odp_actions, actions_offset);
4131 nl_msg_end_nested(odp_actions, sample_offset);
4132 return cookie_offset;
4135 /* SAMPLE action must be first action in any given list of actions.
4136 * At this point we do not have all information required to build it. So try to
4137 * build sample action as complete as possible. */
4139 add_sflow_action(struct action_xlate_ctx *ctx)
4141 ctx->user_cookie_offset = compose_sflow_action(ctx->ofproto,
4143 &ctx->flow, OVSP_NONE);
4144 ctx->sflow_odp_port = 0;
4145 ctx->sflow_n_outputs = 0;
4148 /* Fix SAMPLE action according to data collected while composing ODP actions.
4149 * We need to fix SAMPLE actions OVS_SAMPLE_ATTR_ACTIONS attribute, i.e. nested
4150 * USERSPACE action's user-cookie which is required for sflow. */
4152 fix_sflow_action(struct action_xlate_ctx *ctx)
4154 const struct flow *base = &ctx->base_flow;
4155 struct user_action_cookie *cookie;
4157 if (!ctx->user_cookie_offset) {
4161 cookie = ofpbuf_at(ctx->odp_actions, ctx->user_cookie_offset,
4163 assert(cookie != NULL);
4164 assert(cookie->type == USER_ACTION_COOKIE_SFLOW);
4166 if (ctx->sflow_n_outputs) {
4167 cookie->data = dpif_sflow_odp_port_to_ifindex(ctx->ofproto->sflow,
4168 ctx->sflow_odp_port);
4170 if (ctx->sflow_n_outputs >= 255) {
4171 cookie->n_output = 255;
4173 cookie->n_output = ctx->sflow_n_outputs;
4175 cookie->vlan_tci = base->vlan_tci;
4179 compose_output_action__(struct action_xlate_ctx *ctx, uint16_t ofp_port,
4182 const struct ofport_dpif *ofport = get_ofp_port(ctx->ofproto, ofp_port);
4183 uint16_t odp_port = ofp_port_to_odp_port(ofp_port);
4184 ovs_be16 flow_vlan_tci = ctx->flow.vlan_tci;
4185 uint8_t flow_nw_tos = ctx->flow.nw_tos;
4189 struct priority_to_dscp *pdscp;
4191 if (ofport->up.opp.config & htonl(OFPPC_NO_FWD)
4192 || (check_stp && !stp_forward_in_state(ofport->stp_state))) {
4196 pdscp = get_priority(ofport, ctx->flow.priority);
4198 ctx->flow.nw_tos &= ~IP_DSCP_MASK;
4199 ctx->flow.nw_tos |= pdscp->dscp;
4202 /* We may not have an ofport record for this port, but it doesn't hurt
4203 * to allow forwarding to it anyhow. Maybe such a port will appear
4204 * later and we're pre-populating the flow table. */
4207 out_port = vsp_realdev_to_vlandev(ctx->ofproto, odp_port,
4208 ctx->flow.vlan_tci);
4209 if (out_port != odp_port) {
4210 ctx->flow.vlan_tci = htons(0);
4212 commit_odp_actions(&ctx->flow, &ctx->base_flow, ctx->odp_actions);
4213 nl_msg_put_u32(ctx->odp_actions, OVS_ACTION_ATTR_OUTPUT, out_port);
4215 ctx->sflow_odp_port = odp_port;
4216 ctx->sflow_n_outputs++;
4217 ctx->nf_output_iface = ofp_port;
4218 ctx->flow.vlan_tci = flow_vlan_tci;
4219 ctx->flow.nw_tos = flow_nw_tos;
4223 compose_output_action(struct action_xlate_ctx *ctx, uint16_t ofp_port)
4225 compose_output_action__(ctx, ofp_port, true);
4229 xlate_table_action(struct action_xlate_ctx *ctx,
4230 uint16_t in_port, uint8_t table_id)
4232 if (ctx->recurse < MAX_RESUBMIT_RECURSION) {
4233 struct ofproto_dpif *ofproto = ctx->ofproto;
4234 struct rule_dpif *rule;
4235 uint16_t old_in_port;
4236 uint8_t old_table_id;
4238 old_table_id = ctx->table_id;
4239 ctx->table_id = table_id;
4241 /* Look up a flow with 'in_port' as the input port. */
4242 old_in_port = ctx->flow.in_port;
4243 ctx->flow.in_port = in_port;
4244 rule = rule_dpif_lookup(ofproto, &ctx->flow, table_id);
4247 if (table_id > 0 && table_id < N_TABLES) {
4248 struct table_dpif *table = &ofproto->tables[table_id];
4249 if (table->other_table) {
4252 : rule_calculate_tag(&ctx->flow,
4253 &table->other_table->wc,
4258 /* Restore the original input port. Otherwise OFPP_NORMAL and
4259 * OFPP_IN_PORT will have surprising behavior. */
4260 ctx->flow.in_port = old_in_port;
4262 if (ctx->resubmit_hook) {
4263 ctx->resubmit_hook(ctx, rule);
4268 do_xlate_actions(rule->up.actions, rule->up.n_actions, ctx);
4272 ctx->table_id = old_table_id;
4274 static struct vlog_rate_limit recurse_rl = VLOG_RATE_LIMIT_INIT(1, 1);
4276 VLOG_ERR_RL(&recurse_rl, "resubmit actions recursed over %d times",
4277 MAX_RESUBMIT_RECURSION);
4282 xlate_resubmit_table(struct action_xlate_ctx *ctx,
4283 const struct nx_action_resubmit *nar)
4288 in_port = (nar->in_port == htons(OFPP_IN_PORT)
4290 : ntohs(nar->in_port));
4291 table_id = nar->table == 255 ? ctx->table_id : nar->table;
4293 xlate_table_action(ctx, in_port, table_id);
4297 flood_packets(struct action_xlate_ctx *ctx, bool all)
4299 struct ofport_dpif *ofport;
4301 HMAP_FOR_EACH (ofport, up.hmap_node, &ctx->ofproto->up.ports) {
4302 uint16_t ofp_port = ofport->up.ofp_port;
4304 if (ofp_port == ctx->flow.in_port) {
4309 compose_output_action__(ctx, ofp_port, false);
4310 } else if (!(ofport->up.opp.config & htonl(OFPPC_NO_FLOOD))) {
4311 compose_output_action(ctx, ofp_port);
4315 ctx->nf_output_iface = NF_OUT_FLOOD;
4319 compose_controller_action(struct action_xlate_ctx *ctx, int len)
4321 struct user_action_cookie cookie;
4323 commit_odp_actions(&ctx->flow, &ctx->base_flow, ctx->odp_actions);
4324 cookie.type = USER_ACTION_COOKIE_CONTROLLER;
4326 cookie.n_output = 0;
4327 cookie.vlan_tci = 0;
4328 put_userspace_action(ctx->ofproto, ctx->odp_actions, &ctx->flow, &cookie);
4332 xlate_output_action__(struct action_xlate_ctx *ctx,
4333 uint16_t port, uint16_t max_len)
4335 uint16_t prev_nf_output_iface = ctx->nf_output_iface;
4337 ctx->nf_output_iface = NF_OUT_DROP;
4341 compose_output_action(ctx, ctx->flow.in_port);
4344 xlate_table_action(ctx, ctx->flow.in_port, ctx->table_id);
4350 flood_packets(ctx, false);
4353 flood_packets(ctx, true);
4355 case OFPP_CONTROLLER:
4356 compose_controller_action(ctx, max_len);
4362 if (port != ctx->flow.in_port) {
4363 compose_output_action(ctx, port);
4368 if (prev_nf_output_iface == NF_OUT_FLOOD) {
4369 ctx->nf_output_iface = NF_OUT_FLOOD;
4370 } else if (ctx->nf_output_iface == NF_OUT_DROP) {
4371 ctx->nf_output_iface = prev_nf_output_iface;
4372 } else if (prev_nf_output_iface != NF_OUT_DROP &&
4373 ctx->nf_output_iface != NF_OUT_FLOOD) {
4374 ctx->nf_output_iface = NF_OUT_MULTI;
4379 xlate_output_reg_action(struct action_xlate_ctx *ctx,
4380 const struct nx_action_output_reg *naor)
4384 ofp_port = nxm_read_field_bits(naor->src, naor->ofs_nbits, &ctx->flow);
4386 if (ofp_port <= UINT16_MAX) {
4387 xlate_output_action__(ctx, ofp_port, ntohs(naor->max_len));
4392 xlate_output_action(struct action_xlate_ctx *ctx,
4393 const struct ofp_action_output *oao)
4395 xlate_output_action__(ctx, ntohs(oao->port), ntohs(oao->max_len));
4399 xlate_enqueue_action(struct action_xlate_ctx *ctx,
4400 const struct ofp_action_enqueue *oae)
4403 uint32_t flow_priority, priority;
4406 error = dpif_queue_to_priority(ctx->ofproto->dpif, ntohl(oae->queue_id),
4409 /* Fall back to ordinary output action. */
4410 xlate_output_action__(ctx, ntohs(oae->port), 0);
4414 /* Figure out datapath output port. */
4415 ofp_port = ntohs(oae->port);
4416 if (ofp_port == OFPP_IN_PORT) {
4417 ofp_port = ctx->flow.in_port;
4418 } else if (ofp_port == ctx->flow.in_port) {
4422 /* Add datapath actions. */
4423 flow_priority = ctx->flow.priority;
4424 ctx->flow.priority = priority;
4425 compose_output_action(ctx, ofp_port);
4426 ctx->flow.priority = flow_priority;
4428 /* Update NetFlow output port. */
4429 if (ctx->nf_output_iface == NF_OUT_DROP) {
4430 ctx->nf_output_iface = ofp_port;
4431 } else if (ctx->nf_output_iface != NF_OUT_FLOOD) {
4432 ctx->nf_output_iface = NF_OUT_MULTI;
4437 xlate_set_queue_action(struct action_xlate_ctx *ctx,
4438 const struct nx_action_set_queue *nasq)
4443 error = dpif_queue_to_priority(ctx->ofproto->dpif, ntohl(nasq->queue_id),
4446 /* Couldn't translate queue to a priority, so ignore. A warning
4447 * has already been logged. */
4451 ctx->flow.priority = priority;
4454 struct xlate_reg_state {
4460 xlate_autopath(struct action_xlate_ctx *ctx,
4461 const struct nx_action_autopath *naa)
4463 uint16_t ofp_port = ntohl(naa->id);
4464 struct ofport_dpif *port = get_ofp_port(ctx->ofproto, ofp_port);
4466 if (!port || !port->bundle) {
4467 ofp_port = OFPP_NONE;
4468 } else if (port->bundle->bond) {
4469 /* Autopath does not support VLAN hashing. */
4470 struct ofport_dpif *slave = bond_choose_output_slave(
4471 port->bundle->bond, &ctx->flow, 0, &ctx->tags);
4473 ofp_port = slave->up.ofp_port;
4476 autopath_execute(naa, &ctx->flow, ofp_port);
4480 slave_enabled_cb(uint16_t ofp_port, void *ofproto_)
4482 struct ofproto_dpif *ofproto = ofproto_;
4483 struct ofport_dpif *port;
4493 case OFPP_CONTROLLER: /* Not supported by the bundle action. */
4496 port = get_ofp_port(ofproto, ofp_port);
4497 return port ? port->may_enable : false;
4502 xlate_learn_action(struct action_xlate_ctx *ctx,
4503 const struct nx_action_learn *learn)
4505 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(5, 1);
4506 struct ofputil_flow_mod fm;
4509 learn_execute(learn, &ctx->flow, &fm);
4511 error = ofproto_flow_mod(&ctx->ofproto->up, &fm);
4512 if (error && !VLOG_DROP_WARN(&rl)) {
4513 char *msg = ofputil_error_to_string(error);
4514 VLOG_WARN("learning action failed to modify flow table (%s)", msg);
4522 may_receive(const struct ofport_dpif *port, struct action_xlate_ctx *ctx)
4524 if (port->up.opp.config & (eth_addr_equals(ctx->flow.dl_dst, eth_addr_stp)
4525 ? htonl(OFPPC_NO_RECV_STP)
4526 : htonl(OFPPC_NO_RECV))) {
4530 /* Only drop packets here if both forwarding and learning are
4531 * disabled. If just learning is enabled, we need to have
4532 * OFPP_NORMAL and the learning action have a look at the packet
4533 * before we can drop it. */
4534 if (!stp_forward_in_state(port->stp_state)
4535 && !stp_learn_in_state(port->stp_state)) {
4543 do_xlate_actions(const union ofp_action *in, size_t n_in,
4544 struct action_xlate_ctx *ctx)
4546 const struct ofport_dpif *port;
4547 const union ofp_action *ia;
4550 port = get_ofp_port(ctx->ofproto, ctx->flow.in_port);
4551 if (port && !may_receive(port, ctx)) {
4552 /* Drop this flow. */
4556 OFPUTIL_ACTION_FOR_EACH_UNSAFE (ia, left, in, n_in) {
4557 const struct ofp_action_dl_addr *oada;
4558 const struct nx_action_resubmit *nar;
4559 const struct nx_action_set_tunnel *nast;
4560 const struct nx_action_set_queue *nasq;
4561 const struct nx_action_multipath *nam;
4562 const struct nx_action_autopath *naa;
4563 const struct nx_action_bundle *nab;
4564 const struct nx_action_output_reg *naor;
4565 enum ofputil_action_code code;
4572 code = ofputil_decode_action_unsafe(ia);
4574 case OFPUTIL_OFPAT_OUTPUT:
4575 xlate_output_action(ctx, &ia->output);
4578 case OFPUTIL_OFPAT_SET_VLAN_VID:
4579 ctx->flow.vlan_tci &= ~htons(VLAN_VID_MASK);
4580 ctx->flow.vlan_tci |= ia->vlan_vid.vlan_vid | htons(VLAN_CFI);
4583 case OFPUTIL_OFPAT_SET_VLAN_PCP:
4584 ctx->flow.vlan_tci &= ~htons(VLAN_PCP_MASK);
4585 ctx->flow.vlan_tci |= htons(
4586 (ia->vlan_pcp.vlan_pcp << VLAN_PCP_SHIFT) | VLAN_CFI);
4589 case OFPUTIL_OFPAT_STRIP_VLAN:
4590 ctx->flow.vlan_tci = htons(0);
4593 case OFPUTIL_OFPAT_SET_DL_SRC:
4594 oada = ((struct ofp_action_dl_addr *) ia);
4595 memcpy(ctx->flow.dl_src, oada->dl_addr, ETH_ADDR_LEN);
4598 case OFPUTIL_OFPAT_SET_DL_DST:
4599 oada = ((struct ofp_action_dl_addr *) ia);
4600 memcpy(ctx->flow.dl_dst, oada->dl_addr, ETH_ADDR_LEN);
4603 case OFPUTIL_OFPAT_SET_NW_SRC:
4604 ctx->flow.nw_src = ia->nw_addr.nw_addr;
4607 case OFPUTIL_OFPAT_SET_NW_DST:
4608 ctx->flow.nw_dst = ia->nw_addr.nw_addr;
4611 case OFPUTIL_OFPAT_SET_NW_TOS:
4612 ctx->flow.nw_tos &= ~IP_DSCP_MASK;
4613 ctx->flow.nw_tos |= ia->nw_tos.nw_tos & IP_DSCP_MASK;
4616 case OFPUTIL_OFPAT_SET_TP_SRC:
4617 ctx->flow.tp_src = ia->tp_port.tp_port;
4620 case OFPUTIL_OFPAT_SET_TP_DST:
4621 ctx->flow.tp_dst = ia->tp_port.tp_port;
4624 case OFPUTIL_OFPAT_ENQUEUE:
4625 xlate_enqueue_action(ctx, (const struct ofp_action_enqueue *) ia);
4628 case OFPUTIL_NXAST_RESUBMIT:
4629 nar = (const struct nx_action_resubmit *) ia;
4630 xlate_table_action(ctx, ntohs(nar->in_port), ctx->table_id);
4633 case OFPUTIL_NXAST_RESUBMIT_TABLE:
4634 xlate_resubmit_table(ctx, (const struct nx_action_resubmit *) ia);
4637 case OFPUTIL_NXAST_SET_TUNNEL:
4638 nast = (const struct nx_action_set_tunnel *) ia;
4639 tun_id = htonll(ntohl(nast->tun_id));
4640 ctx->flow.tun_id = tun_id;
4643 case OFPUTIL_NXAST_SET_QUEUE:
4644 nasq = (const struct nx_action_set_queue *) ia;
4645 xlate_set_queue_action(ctx, nasq);
4648 case OFPUTIL_NXAST_POP_QUEUE:
4649 ctx->flow.priority = ctx->original_priority;
4652 case OFPUTIL_NXAST_REG_MOVE:
4653 nxm_execute_reg_move((const struct nx_action_reg_move *) ia,
4657 case OFPUTIL_NXAST_REG_LOAD:
4658 nxm_execute_reg_load((const struct nx_action_reg_load *) ia,
4662 case OFPUTIL_NXAST_NOTE:
4663 /* Nothing to do. */
4666 case OFPUTIL_NXAST_SET_TUNNEL64:
4667 tun_id = ((const struct nx_action_set_tunnel64 *) ia)->tun_id;
4668 ctx->flow.tun_id = tun_id;
4671 case OFPUTIL_NXAST_MULTIPATH:
4672 nam = (const struct nx_action_multipath *) ia;
4673 multipath_execute(nam, &ctx->flow);
4676 case OFPUTIL_NXAST_AUTOPATH:
4677 naa = (const struct nx_action_autopath *) ia;
4678 xlate_autopath(ctx, naa);
4681 case OFPUTIL_NXAST_BUNDLE:
4682 ctx->ofproto->has_bundle_action = true;
4683 nab = (const struct nx_action_bundle *) ia;
4684 xlate_output_action__(ctx, bundle_execute(nab, &ctx->flow,
4689 case OFPUTIL_NXAST_BUNDLE_LOAD:
4690 ctx->ofproto->has_bundle_action = true;
4691 nab = (const struct nx_action_bundle *) ia;
4692 bundle_execute_load(nab, &ctx->flow, slave_enabled_cb,
4696 case OFPUTIL_NXAST_OUTPUT_REG:
4697 naor = (const struct nx_action_output_reg *) ia;
4698 xlate_output_reg_action(ctx, naor);
4701 case OFPUTIL_NXAST_LEARN:
4702 ctx->has_learn = true;
4703 if (ctx->may_flow_mod) {
4704 xlate_learn_action(ctx, (const struct nx_action_learn *) ia);
4708 case OFPUTIL_NXAST_EXIT:
4714 /* We've let OFPP_NORMAL and the learning action look at the packet,
4715 * so drop it now if forwarding is disabled. */
4716 if (port && !stp_forward_in_state(port->stp_state)) {
4717 ofpbuf_clear(ctx->odp_actions);
4718 add_sflow_action(ctx);
4723 action_xlate_ctx_init(struct action_xlate_ctx *ctx,
4724 struct ofproto_dpif *ofproto, const struct flow *flow,
4725 ovs_be16 initial_tci, const struct ofpbuf *packet)
4727 ctx->ofproto = ofproto;
4729 ctx->base_flow = ctx->flow;
4730 ctx->base_flow.tun_id = 0;
4731 ctx->base_flow.vlan_tci = initial_tci;
4732 ctx->packet = packet;
4733 ctx->may_learn_macs = packet != NULL;
4734 ctx->may_flow_mod = packet != NULL;
4735 ctx->resubmit_hook = NULL;
4738 static struct ofpbuf *
4739 xlate_actions(struct action_xlate_ctx *ctx,
4740 const union ofp_action *in, size_t n_in)
4742 struct flow orig_flow = ctx->flow;
4744 COVERAGE_INC(ofproto_dpif_xlate);
4746 ctx->odp_actions = ofpbuf_new(512);
4747 ofpbuf_reserve(ctx->odp_actions, NL_A_U32_SIZE);
4749 ctx->may_set_up_flow = true;
4750 ctx->has_learn = false;
4751 ctx->has_normal = false;
4752 ctx->nf_output_iface = NF_OUT_DROP;
4755 ctx->original_priority = ctx->flow.priority;
4759 if (ctx->flow.nw_frag & FLOW_NW_FRAG_ANY) {
4760 switch (ctx->ofproto->up.frag_handling) {
4761 case OFPC_FRAG_NORMAL:
4762 /* We must pretend that transport ports are unavailable. */
4763 ctx->flow.tp_src = ctx->base_flow.tp_src = htons(0);
4764 ctx->flow.tp_dst = ctx->base_flow.tp_dst = htons(0);
4767 case OFPC_FRAG_DROP:
4768 return ctx->odp_actions;
4770 case OFPC_FRAG_REASM:
4773 case OFPC_FRAG_NX_MATCH:
4774 /* Nothing to do. */
4779 if (process_special(ctx->ofproto, &ctx->flow, ctx->packet)) {
4780 ctx->may_set_up_flow = false;
4781 return ctx->odp_actions;
4783 add_sflow_action(ctx);
4784 do_xlate_actions(in, n_in, ctx);
4786 if (!connmgr_may_set_up_flow(ctx->ofproto->up.connmgr, &ctx->flow,
4787 ctx->odp_actions->data,
4788 ctx->odp_actions->size)) {
4789 ctx->may_set_up_flow = false;
4791 && connmgr_msg_in_hook(ctx->ofproto->up.connmgr, &ctx->flow,
4793 compose_output_action(ctx, OFPP_LOCAL);
4796 add_mirror_actions(ctx, &orig_flow);
4797 fix_sflow_action(ctx);
4800 return ctx->odp_actions;
4803 /* OFPP_NORMAL implementation. */
4805 static struct ofport_dpif *ofbundle_get_a_port(const struct ofbundle *);
4807 /* Given 'vid', the VID obtained from the 802.1Q header that was received as
4808 * part of a packet (specify 0 if there was no 802.1Q header), and 'in_bundle',
4809 * the bundle on which the packet was received, returns the VLAN to which the
4812 * Both 'vid' and the return value are in the range 0...4095. */
4814 input_vid_to_vlan(const struct ofbundle *in_bundle, uint16_t vid)
4816 switch (in_bundle->vlan_mode) {
4817 case PORT_VLAN_ACCESS:
4818 return in_bundle->vlan;
4821 case PORT_VLAN_TRUNK:
4824 case PORT_VLAN_NATIVE_UNTAGGED:
4825 case PORT_VLAN_NATIVE_TAGGED:
4826 return vid ? vid : in_bundle->vlan;
4833 /* Checks whether a packet with the given 'vid' may ingress on 'in_bundle'.
4834 * If so, returns true. Otherwise, returns false and, if 'warn' is true, logs
4837 * 'vid' should be the VID obtained from the 802.1Q header that was received as
4838 * part of a packet (specify 0 if there was no 802.1Q header), in the range
4841 input_vid_is_valid(uint16_t vid, struct ofbundle *in_bundle, bool warn)
4843 /* Allow any VID on the OFPP_NONE port. */
4844 if (in_bundle == &ofpp_none_bundle) {
4848 switch (in_bundle->vlan_mode) {
4849 case PORT_VLAN_ACCESS:
4852 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
4853 VLOG_WARN_RL(&rl, "bridge %s: dropping VLAN %"PRIu16" tagged "
4854 "packet received on port %s configured as VLAN "
4855 "%"PRIu16" access port",
4856 in_bundle->ofproto->up.name, vid,
4857 in_bundle->name, in_bundle->vlan);
4863 case PORT_VLAN_NATIVE_UNTAGGED:
4864 case PORT_VLAN_NATIVE_TAGGED:
4866 /* Port must always carry its native VLAN. */
4870 case PORT_VLAN_TRUNK:
4871 if (!ofbundle_includes_vlan(in_bundle, vid)) {
4873 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
4874 VLOG_WARN_RL(&rl, "bridge %s: dropping VLAN %"PRIu16" packet "
4875 "received on port %s not configured for trunking "
4877 in_bundle->ofproto->up.name, vid,
4878 in_bundle->name, vid);
4890 /* Given 'vlan', the VLAN that a packet belongs to, and
4891 * 'out_bundle', a bundle on which the packet is to be output, returns the VID
4892 * that should be included in the 802.1Q header. (If the return value is 0,
4893 * then the 802.1Q header should only be included in the packet if there is a
4896 * Both 'vlan' and the return value are in the range 0...4095. */
4898 output_vlan_to_vid(const struct ofbundle *out_bundle, uint16_t vlan)
4900 switch (out_bundle->vlan_mode) {
4901 case PORT_VLAN_ACCESS:
4904 case PORT_VLAN_TRUNK:
4905 case PORT_VLAN_NATIVE_TAGGED:
4908 case PORT_VLAN_NATIVE_UNTAGGED:
4909 return vlan == out_bundle->vlan ? 0 : vlan;
4917 output_normal(struct action_xlate_ctx *ctx, const struct ofbundle *out_bundle,
4920 struct ofport_dpif *port;
4922 ovs_be16 tci, old_tci;
4924 vid = output_vlan_to_vid(out_bundle, vlan);
4925 if (!out_bundle->bond) {
4926 port = ofbundle_get_a_port(out_bundle);
4928 port = bond_choose_output_slave(out_bundle->bond, &ctx->flow,
4931 /* No slaves enabled, so drop packet. */
4936 old_tci = ctx->flow.vlan_tci;
4938 if (tci || out_bundle->use_priority_tags) {
4939 tci |= ctx->flow.vlan_tci & htons(VLAN_PCP_MASK);
4941 tci |= htons(VLAN_CFI);
4944 ctx->flow.vlan_tci = tci;
4946 compose_output_action(ctx, port->up.ofp_port);
4947 ctx->flow.vlan_tci = old_tci;
4951 mirror_mask_ffs(mirror_mask_t mask)
4953 BUILD_ASSERT_DECL(sizeof(unsigned int) >= sizeof(mask));
4958 ofbundle_trunks_vlan(const struct ofbundle *bundle, uint16_t vlan)
4960 return (bundle->vlan_mode != PORT_VLAN_ACCESS
4961 && (!bundle->trunks || bitmap_is_set(bundle->trunks, vlan)));
4965 ofbundle_includes_vlan(const struct ofbundle *bundle, uint16_t vlan)
4967 return vlan == bundle->vlan || ofbundle_trunks_vlan(bundle, vlan);
4970 /* Returns an arbitrary interface within 'bundle'. */
4971 static struct ofport_dpif *
4972 ofbundle_get_a_port(const struct ofbundle *bundle)
4974 return CONTAINER_OF(list_front(&bundle->ports),
4975 struct ofport_dpif, bundle_node);
4979 vlan_is_mirrored(const struct ofmirror *m, int vlan)
4981 return !m->vlans || bitmap_is_set(m->vlans, vlan);
4984 /* Returns true if a packet with Ethernet destination MAC 'dst' may be mirrored
4985 * to a VLAN. In general most packets may be mirrored but we want to drop
4986 * protocols that may confuse switches. */
4988 eth_dst_may_rspan(const uint8_t dst[ETH_ADDR_LEN])
4990 /* If you change this function's behavior, please update corresponding
4991 * documentation in vswitch.xml at the same time. */
4992 if (dst[0] != 0x01) {
4993 /* All the currently banned MACs happen to start with 01 currently, so
4994 * this is a quick way to eliminate most of the good ones. */
4996 if (eth_addr_is_reserved(dst)) {
4997 /* Drop STP, IEEE pause frames, and other reserved protocols
4998 * (01-80-c2-00-00-0x). */
5002 if (dst[0] == 0x01 && dst[1] == 0x00 && dst[2] == 0x0c) {
5004 if ((dst[3] & 0xfe) == 0xcc &&
5005 (dst[4] & 0xfe) == 0xcc &&
5006 (dst[5] & 0xfe) == 0xcc) {
5007 /* Drop the following protocols plus others following the same
5010 CDP, VTP, DTP, PAgP (01-00-0c-cc-cc-cc)
5011 Spanning Tree PVSTP+ (01-00-0c-cc-cc-cd)
5012 STP Uplink Fast (01-00-0c-cd-cd-cd) */
5016 if (!(dst[3] | dst[4] | dst[5])) {
5017 /* Drop Inter Switch Link packets (01-00-0c-00-00-00). */
5026 add_mirror_actions(struct action_xlate_ctx *ctx, const struct flow *orig_flow)
5028 struct ofproto_dpif *ofproto = ctx->ofproto;
5029 mirror_mask_t mirrors;
5030 struct ofbundle *in_bundle;
5033 const struct nlattr *a;
5036 in_bundle = lookup_input_bundle(ctx->ofproto, orig_flow->in_port,
5037 ctx->packet != NULL);
5041 mirrors = in_bundle->src_mirrors;
5043 /* Drop frames on bundles reserved for mirroring. */
5044 if (in_bundle->mirror_out) {
5045 if (ctx->packet != NULL) {
5046 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
5047 VLOG_WARN_RL(&rl, "bridge %s: dropping packet received on port "
5048 "%s, which is reserved exclusively for mirroring",
5049 ctx->ofproto->up.name, in_bundle->name);
5055 vid = vlan_tci_to_vid(orig_flow->vlan_tci);
5056 if (!input_vid_is_valid(vid, in_bundle, ctx->packet != NULL)) {
5059 vlan = input_vid_to_vlan(in_bundle, vid);
5061 /* Look at the output ports to check for destination selections. */
5063 NL_ATTR_FOR_EACH (a, left, ctx->odp_actions->data,
5064 ctx->odp_actions->size) {
5065 enum ovs_action_attr type = nl_attr_type(a);
5066 struct ofport_dpif *ofport;
5068 if (type != OVS_ACTION_ATTR_OUTPUT) {
5072 ofport = get_odp_port(ofproto, nl_attr_get_u32(a));
5073 if (ofport && ofport->bundle) {
5074 mirrors |= ofport->bundle->dst_mirrors;
5082 /* Restore the original packet before adding the mirror actions. */
5083 ctx->flow = *orig_flow;
5088 m = ofproto->mirrors[mirror_mask_ffs(mirrors) - 1];
5090 if (!vlan_is_mirrored(m, vlan)) {
5091 mirrors &= mirrors - 1;
5095 mirrors &= ~m->dup_mirrors;
5096 ctx->mirrors |= m->dup_mirrors;
5098 output_normal(ctx, m->out, vlan);
5099 } else if (eth_dst_may_rspan(orig_flow->dl_dst)
5100 && vlan != m->out_vlan) {
5101 struct ofbundle *bundle;
5103 HMAP_FOR_EACH (bundle, hmap_node, &ofproto->bundles) {
5104 if (ofbundle_includes_vlan(bundle, m->out_vlan)
5105 && !bundle->mirror_out) {
5106 output_normal(ctx, bundle, m->out_vlan);
5114 update_mirror_stats(struct ofproto_dpif *ofproto, mirror_mask_t mirrors,
5115 uint64_t packets, uint64_t bytes)
5121 for (; mirrors; mirrors &= mirrors - 1) {
5124 m = ofproto->mirrors[mirror_mask_ffs(mirrors) - 1];
5127 /* In normal circumstances 'm' will not be NULL. However,
5128 * if mirrors are reconfigured, we can temporarily get out
5129 * of sync in facet_revalidate(). We could "correct" the
5130 * mirror list before reaching here, but doing that would
5131 * not properly account the traffic stats we've currently
5132 * accumulated for previous mirror configuration. */
5136 m->packet_count += packets;
5137 m->byte_count += bytes;
5141 /* A VM broadcasts a gratuitous ARP to indicate that it has resumed after
5142 * migration. Older Citrix-patched Linux DomU used gratuitous ARP replies to
5143 * indicate this; newer upstream kernels use gratuitous ARP requests. */
5145 is_gratuitous_arp(const struct flow *flow)
5147 return (flow->dl_type == htons(ETH_TYPE_ARP)
5148 && eth_addr_is_broadcast(flow->dl_dst)
5149 && (flow->nw_proto == ARP_OP_REPLY
5150 || (flow->nw_proto == ARP_OP_REQUEST
5151 && flow->nw_src == flow->nw_dst)));
5155 update_learning_table(struct ofproto_dpif *ofproto,
5156 const struct flow *flow, int vlan,
5157 struct ofbundle *in_bundle)
5159 struct mac_entry *mac;
5161 /* Don't learn the OFPP_NONE port. */
5162 if (in_bundle == &ofpp_none_bundle) {
5166 if (!mac_learning_may_learn(ofproto->ml, flow->dl_src, vlan)) {
5170 mac = mac_learning_insert(ofproto->ml, flow->dl_src, vlan);
5171 if (is_gratuitous_arp(flow)) {
5172 /* We don't want to learn from gratuitous ARP packets that are
5173 * reflected back over bond slaves so we lock the learning table. */
5174 if (!in_bundle->bond) {
5175 mac_entry_set_grat_arp_lock(mac);
5176 } else if (mac_entry_is_grat_arp_locked(mac)) {
5181 if (mac_entry_is_new(mac) || mac->port.p != in_bundle) {
5182 /* The log messages here could actually be useful in debugging,
5183 * so keep the rate limit relatively high. */
5184 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(30, 300);
5185 VLOG_DBG_RL(&rl, "bridge %s: learned that "ETH_ADDR_FMT" is "
5186 "on port %s in VLAN %d",
5187 ofproto->up.name, ETH_ADDR_ARGS(flow->dl_src),
5188 in_bundle->name, vlan);
5190 mac->port.p = in_bundle;
5191 tag_set_add(&ofproto->revalidate_set,
5192 mac_learning_changed(ofproto->ml, mac));
5196 static struct ofbundle *
5197 lookup_input_bundle(struct ofproto_dpif *ofproto, uint16_t in_port, bool warn)
5199 struct ofport_dpif *ofport;
5201 /* Special-case OFPP_NONE, which a controller may use as the ingress
5202 * port for traffic that it is sourcing. */
5203 if (in_port == OFPP_NONE) {
5204 return &ofpp_none_bundle;
5207 /* Find the port and bundle for the received packet. */
5208 ofport = get_ofp_port(ofproto, in_port);
5209 if (ofport && ofport->bundle) {
5210 return ofport->bundle;
5213 /* Odd. A few possible reasons here:
5215 * - We deleted a port but there are still a few packets queued up
5218 * - Someone externally added a port (e.g. "ovs-dpctl add-if") that
5219 * we don't know about.
5221 * - The ofproto client didn't configure the port as part of a bundle.
5224 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
5226 VLOG_WARN_RL(&rl, "bridge %s: received packet on unknown "
5227 "port %"PRIu16, ofproto->up.name, in_port);
5232 /* Determines whether packets in 'flow' within 'ofproto' should be forwarded or
5233 * dropped. Returns true if they may be forwarded, false if they should be
5236 * 'in_port' must be the ofport_dpif that corresponds to flow->in_port.
5237 * 'in_port' must be part of a bundle (e.g. in_port->bundle must be nonnull).
5239 * 'vlan' must be the VLAN that corresponds to flow->vlan_tci on 'in_port', as
5240 * returned by input_vid_to_vlan(). It must be a valid VLAN for 'in_port', as
5241 * checked by input_vid_is_valid().
5243 * May also add tags to '*tags', although the current implementation only does
5244 * so in one special case.
5247 is_admissible(struct ofproto_dpif *ofproto, const struct flow *flow,
5248 struct ofport_dpif *in_port, uint16_t vlan, tag_type *tags)
5250 struct ofbundle *in_bundle = in_port->bundle;
5252 /* Drop frames for reserved multicast addresses
5253 * only if forward_bpdu option is absent. */
5254 if (eth_addr_is_reserved(flow->dl_dst) && !ofproto->up.forward_bpdu) {
5258 if (in_bundle->bond) {
5259 struct mac_entry *mac;
5261 switch (bond_check_admissibility(in_bundle->bond, in_port,
5262 flow->dl_dst, tags)) {
5269 case BV_DROP_IF_MOVED:
5270 mac = mac_learning_lookup(ofproto->ml, flow->dl_src, vlan, NULL);
5271 if (mac && mac->port.p != in_bundle &&
5272 (!is_gratuitous_arp(flow)
5273 || mac_entry_is_grat_arp_locked(mac))) {
5284 xlate_normal(struct action_xlate_ctx *ctx)
5286 struct ofport_dpif *in_port;
5287 struct ofbundle *in_bundle;
5288 struct mac_entry *mac;
5292 ctx->has_normal = true;
5294 in_bundle = lookup_input_bundle(ctx->ofproto, ctx->flow.in_port,
5295 ctx->packet != NULL);
5300 /* We know 'in_port' exists unless it is "ofpp_none_bundle",
5301 * since lookup_input_bundle() succeeded. */
5302 in_port = get_ofp_port(ctx->ofproto, ctx->flow.in_port);
5304 /* Drop malformed frames. */
5305 if (ctx->flow.dl_type == htons(ETH_TYPE_VLAN) &&
5306 !(ctx->flow.vlan_tci & htons(VLAN_CFI))) {
5307 if (ctx->packet != NULL) {
5308 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
5309 VLOG_WARN_RL(&rl, "bridge %s: dropping packet with partial "
5310 "VLAN tag received on port %s",
5311 ctx->ofproto->up.name, in_bundle->name);
5316 /* Drop frames on bundles reserved for mirroring. */
5317 if (in_bundle->mirror_out) {
5318 if (ctx->packet != NULL) {
5319 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
5320 VLOG_WARN_RL(&rl, "bridge %s: dropping packet received on port "
5321 "%s, which is reserved exclusively for mirroring",
5322 ctx->ofproto->up.name, in_bundle->name);
5328 vid = vlan_tci_to_vid(ctx->flow.vlan_tci);
5329 if (!input_vid_is_valid(vid, in_bundle, ctx->packet != NULL)) {
5332 vlan = input_vid_to_vlan(in_bundle, vid);
5334 /* Check other admissibility requirements. */
5336 !is_admissible(ctx->ofproto, &ctx->flow, in_port, vlan, &ctx->tags)) {
5340 /* Learn source MAC. */
5341 if (ctx->may_learn_macs) {
5342 update_learning_table(ctx->ofproto, &ctx->flow, vlan, in_bundle);
5345 /* Determine output bundle. */
5346 mac = mac_learning_lookup(ctx->ofproto->ml, ctx->flow.dl_dst, vlan,
5349 if (mac->port.p != in_bundle) {
5350 output_normal(ctx, mac->port.p, vlan);
5352 } else if (!ctx->packet && !eth_addr_is_multicast(ctx->flow.dl_dst)) {
5353 /* If we are revalidating but don't have a learning entry then eject
5354 * the flow. Installing a flow that floods packets opens up a window
5355 * of time where we could learn from a packet reflected on a bond and
5356 * blackhole packets before the learning table is updated to reflect
5357 * the correct port. */
5358 ctx->may_set_up_flow = false;
5361 struct ofbundle *bundle;
5363 HMAP_FOR_EACH (bundle, hmap_node, &ctx->ofproto->bundles) {
5364 if (bundle != in_bundle
5365 && ofbundle_includes_vlan(bundle, vlan)
5366 && bundle->floodable
5367 && !bundle->mirror_out) {
5368 output_normal(ctx, bundle, vlan);
5371 ctx->nf_output_iface = NF_OUT_FLOOD;
5375 /* Optimized flow revalidation.
5377 * It's a difficult problem, in general, to tell which facets need to have
5378 * their actions recalculated whenever the OpenFlow flow table changes. We
5379 * don't try to solve that general problem: for most kinds of OpenFlow flow
5380 * table changes, we recalculate the actions for every facet. This is
5381 * relatively expensive, but it's good enough if the OpenFlow flow table
5382 * doesn't change very often.
5384 * However, we can expect one particular kind of OpenFlow flow table change to
5385 * happen frequently: changes caused by MAC learning. To avoid wasting a lot
5386 * of CPU on revalidating every facet whenever MAC learning modifies the flow
5387 * table, we add a special case that applies to flow tables in which every rule
5388 * has the same form (that is, the same wildcards), except that the table is
5389 * also allowed to have a single "catch-all" flow that matches all packets. We
5390 * optimize this case by tagging all of the facets that resubmit into the table
5391 * and invalidating the same tag whenever a flow changes in that table. The
5392 * end result is that we revalidate just the facets that need it (and sometimes
5393 * a few more, but not all of the facets or even all of the facets that
5394 * resubmit to the table modified by MAC learning). */
5396 /* Calculates the tag to use for 'flow' and wildcards 'wc' when it is inserted
5397 * into an OpenFlow table with the given 'basis'. */
5399 rule_calculate_tag(const struct flow *flow, const struct flow_wildcards *wc,
5402 if (flow_wildcards_is_catchall(wc)) {
5405 struct flow tag_flow = *flow;
5406 flow_zero_wildcards(&tag_flow, wc);
5407 return tag_create_deterministic(flow_hash(&tag_flow, secret));
5411 /* Following a change to OpenFlow table 'table_id' in 'ofproto', update the
5412 * taggability of that table.
5414 * This function must be called after *each* change to a flow table. If you
5415 * skip calling it on some changes then the pointer comparisons at the end can
5416 * be invalid if you get unlucky. For example, if a flow removal causes a
5417 * cls_table to be destroyed and then a flow insertion causes a cls_table with
5418 * different wildcards to be created with the same address, then this function
5419 * will incorrectly skip revalidation. */
5421 table_update_taggable(struct ofproto_dpif *ofproto, uint8_t table_id)
5423 struct table_dpif *table = &ofproto->tables[table_id];
5424 const struct classifier *cls = &ofproto->up.tables[table_id];
5425 struct cls_table *catchall, *other;
5426 struct cls_table *t;
5428 catchall = other = NULL;
5430 switch (hmap_count(&cls->tables)) {
5432 /* We could tag this OpenFlow table but it would make the logic a
5433 * little harder and it's a corner case that doesn't seem worth it
5439 HMAP_FOR_EACH (t, hmap_node, &cls->tables) {
5440 if (cls_table_is_catchall(t)) {
5442 } else if (!other) {
5445 /* Indicate that we can't tag this by setting both tables to
5446 * NULL. (We know that 'catchall' is already NULL.) */
5453 /* Can't tag this table. */
5457 if (table->catchall_table != catchall || table->other_table != other) {
5458 table->catchall_table = catchall;
5459 table->other_table = other;
5460 ofproto->need_revalidate = true;
5464 /* Given 'rule' that has changed in some way (either it is a rule being
5465 * inserted, a rule being deleted, or a rule whose actions are being
5466 * modified), marks facets for revalidation to ensure that packets will be
5467 * forwarded correctly according to the new state of the flow table.
5469 * This function must be called after *each* change to a flow table. See
5470 * the comment on table_update_taggable() for more information. */
5472 rule_invalidate(const struct rule_dpif *rule)
5474 struct ofproto_dpif *ofproto = ofproto_dpif_cast(rule->up.ofproto);
5476 table_update_taggable(ofproto, rule->up.table_id);
5478 if (!ofproto->need_revalidate) {
5479 struct table_dpif *table = &ofproto->tables[rule->up.table_id];
5481 if (table->other_table && rule->tag) {
5482 tag_set_add(&ofproto->revalidate_set, rule->tag);
5484 ofproto->need_revalidate = true;
5490 set_frag_handling(struct ofproto *ofproto_,
5491 enum ofp_config_flags frag_handling)
5493 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
5495 if (frag_handling != OFPC_FRAG_REASM) {
5496 ofproto->need_revalidate = true;
5504 packet_out(struct ofproto *ofproto_, struct ofpbuf *packet,
5505 const struct flow *flow,
5506 const union ofp_action *ofp_actions, size_t n_ofp_actions)
5508 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
5511 if (flow->in_port >= ofproto->max_ports && flow->in_port < OFPP_MAX) {
5512 return ofp_mkerr_nicira(OFPET_BAD_REQUEST, NXBRC_BAD_IN_PORT);
5515 error = validate_actions(ofp_actions, n_ofp_actions, flow,
5516 ofproto->max_ports);
5518 struct odputil_keybuf keybuf;
5519 struct action_xlate_ctx ctx;
5520 struct ofpbuf *odp_actions;
5523 ofpbuf_use_stack(&key, &keybuf, sizeof keybuf);
5524 odp_flow_key_from_flow(&key, flow);
5526 action_xlate_ctx_init(&ctx, ofproto, flow, flow->vlan_tci, packet);
5527 odp_actions = xlate_actions(&ctx, ofp_actions, n_ofp_actions);
5528 dpif_execute(ofproto->dpif, key.data, key.size,
5529 odp_actions->data, odp_actions->size, packet);
5530 ofpbuf_delete(odp_actions);
5538 set_netflow(struct ofproto *ofproto_,
5539 const struct netflow_options *netflow_options)
5541 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
5543 if (netflow_options) {
5544 if (!ofproto->netflow) {
5545 ofproto->netflow = netflow_create();
5547 return netflow_set_options(ofproto->netflow, netflow_options);
5549 netflow_destroy(ofproto->netflow);
5550 ofproto->netflow = NULL;
5556 get_netflow_ids(const struct ofproto *ofproto_,
5557 uint8_t *engine_type, uint8_t *engine_id)
5559 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
5561 dpif_get_netflow_ids(ofproto->dpif, engine_type, engine_id);
5565 send_active_timeout(struct ofproto_dpif *ofproto, struct facet *facet)
5567 if (!facet_is_controller_flow(facet) &&
5568 netflow_active_timeout_expired(ofproto->netflow, &facet->nf_flow)) {
5569 struct subfacet *subfacet;
5570 struct ofexpired expired;
5572 LIST_FOR_EACH (subfacet, list_node, &facet->subfacets) {
5573 if (subfacet->installed) {
5574 struct dpif_flow_stats stats;
5576 subfacet_install(ofproto, subfacet, subfacet->actions,
5577 subfacet->actions_len, &stats);
5578 subfacet_update_stats(ofproto, subfacet, &stats);
5582 expired.flow = facet->flow;
5583 expired.packet_count = facet->packet_count;
5584 expired.byte_count = facet->byte_count;
5585 expired.used = facet->used;
5586 netflow_expire(ofproto->netflow, &facet->nf_flow, &expired);
5591 send_netflow_active_timeouts(struct ofproto_dpif *ofproto)
5593 struct facet *facet;
5595 HMAP_FOR_EACH (facet, hmap_node, &ofproto->facets) {
5596 send_active_timeout(ofproto, facet);
5600 static struct ofproto_dpif *
5601 ofproto_dpif_lookup(const char *name)
5603 struct ofproto_dpif *ofproto;
5605 HMAP_FOR_EACH_WITH_HASH (ofproto, all_ofproto_dpifs_node,
5606 hash_string(name, 0), &all_ofproto_dpifs) {
5607 if (!strcmp(ofproto->up.name, name)) {
5615 ofproto_unixctl_fdb_flush(struct unixctl_conn *conn,
5616 const char *args, void *aux OVS_UNUSED)
5618 struct ofproto_dpif *ofproto;
5620 if (args[0] != '\0') {
5621 ofproto = ofproto_dpif_lookup(args);
5623 unixctl_command_reply(conn, 501, "no such bridge");
5626 mac_learning_flush(ofproto->ml);
5627 ofproto->need_revalidate = true;
5629 HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
5630 mac_learning_flush(ofproto->ml);
5631 ofproto->need_revalidate = true;
5635 unixctl_command_reply(conn, 200, "table successfully flushed");
5639 ofproto_unixctl_fdb_show(struct unixctl_conn *conn,
5640 const char *args, void *aux OVS_UNUSED)
5642 struct ds ds = DS_EMPTY_INITIALIZER;
5643 const struct ofproto_dpif *ofproto;
5644 const struct mac_entry *e;
5646 ofproto = ofproto_dpif_lookup(args);
5648 unixctl_command_reply(conn, 501, "no such bridge");
5652 ds_put_cstr(&ds, " port VLAN MAC Age\n");
5653 LIST_FOR_EACH (e, lru_node, &ofproto->ml->lrus) {
5654 struct ofbundle *bundle = e->port.p;
5655 ds_put_format(&ds, "%5d %4d "ETH_ADDR_FMT" %3d\n",
5656 ofbundle_get_a_port(bundle)->odp_port,
5657 e->vlan, ETH_ADDR_ARGS(e->mac),
5658 mac_entry_age(ofproto->ml, e));
5660 unixctl_command_reply(conn, 200, ds_cstr(&ds));
5664 struct ofproto_trace {
5665 struct action_xlate_ctx ctx;
5671 trace_format_rule(struct ds *result, uint8_t table_id, int level,
5672 const struct rule_dpif *rule)
5674 ds_put_char_multiple(result, '\t', level);
5676 ds_put_cstr(result, "No match\n");
5680 ds_put_format(result, "Rule: table=%"PRIu8" cookie=%#"PRIx64" ",
5681 table_id, ntohll(rule->up.flow_cookie));
5682 cls_rule_format(&rule->up.cr, result);
5683 ds_put_char(result, '\n');
5685 ds_put_char_multiple(result, '\t', level);
5686 ds_put_cstr(result, "OpenFlow ");
5687 ofp_print_actions(result, rule->up.actions, rule->up.n_actions);
5688 ds_put_char(result, '\n');
5692 trace_format_flow(struct ds *result, int level, const char *title,
5693 struct ofproto_trace *trace)
5695 ds_put_char_multiple(result, '\t', level);
5696 ds_put_format(result, "%s: ", title);
5697 if (flow_equal(&trace->ctx.flow, &trace->flow)) {
5698 ds_put_cstr(result, "unchanged");
5700 flow_format(result, &trace->ctx.flow);
5701 trace->flow = trace->ctx.flow;
5703 ds_put_char(result, '\n');
5707 trace_format_regs(struct ds *result, int level, const char *title,
5708 struct ofproto_trace *trace)
5712 ds_put_char_multiple(result, '\t', level);
5713 ds_put_format(result, "%s:", title);
5714 for (i = 0; i < FLOW_N_REGS; i++) {
5715 ds_put_format(result, " reg%zu=0x%"PRIx32, i, trace->flow.regs[i]);
5717 ds_put_char(result, '\n');
5721 trace_resubmit(struct action_xlate_ctx *ctx, struct rule_dpif *rule)
5723 struct ofproto_trace *trace = CONTAINER_OF(ctx, struct ofproto_trace, ctx);
5724 struct ds *result = trace->result;
5726 ds_put_char(result, '\n');
5727 trace_format_flow(result, ctx->recurse + 1, "Resubmitted flow", trace);
5728 trace_format_regs(result, ctx->recurse + 1, "Resubmitted regs", trace);
5729 trace_format_rule(result, ctx->table_id, ctx->recurse + 1, rule);
5733 ofproto_unixctl_trace(struct unixctl_conn *conn, const char *args_,
5734 void *aux OVS_UNUSED)
5736 char *dpname, *arg1, *arg2, *arg3, *arg4;
5737 char *args = xstrdup(args_);
5738 char *save_ptr = NULL;
5739 struct ofproto_dpif *ofproto;
5740 struct ofpbuf odp_key;
5741 struct ofpbuf *packet;
5742 struct rule_dpif *rule;
5743 ovs_be16 initial_tci;
5749 ofpbuf_init(&odp_key, 0);
5752 dpname = strtok_r(args, " ", &save_ptr);
5754 unixctl_command_reply(conn, 501, "Bad command syntax");
5758 ofproto = ofproto_dpif_lookup(dpname);
5760 unixctl_command_reply(conn, 501, "Unknown ofproto (use ofproto/list "
5764 arg1 = strtok_r(NULL, " ", &save_ptr);
5765 arg2 = strtok_r(NULL, " ", &save_ptr);
5766 arg3 = strtok_r(NULL, " ", &save_ptr);
5767 arg4 = strtok_r(NULL, "", &save_ptr); /* Get entire rest of line. */
5768 if (dpname && arg1 && (!arg2 || !strcmp(arg2, "-generate")) && !arg3) {
5769 /* ofproto/trace dpname flow [-generate] */
5772 /* Convert string to datapath key. */
5773 ofpbuf_init(&odp_key, 0);
5774 error = odp_flow_key_from_string(arg1, NULL, &odp_key);
5776 unixctl_command_reply(conn, 501, "Bad flow syntax");
5780 /* Convert odp_key to flow. */
5781 error = ofproto_dpif_extract_flow_key(ofproto, odp_key.data,
5782 odp_key.size, &flow,
5783 &initial_tci, NULL);
5784 if (error == ODP_FIT_ERROR) {
5785 unixctl_command_reply(conn, 501, "Invalid flow");
5789 /* Generate a packet, if requested. */
5791 packet = ofpbuf_new(0);
5792 flow_compose(packet, &flow);
5794 } else if (dpname && arg1 && arg2 && arg3 && arg4) {
5795 /* ofproto/trace dpname priority tun_id in_port packet */
5800 priority = atoi(arg1);
5801 tun_id = htonll(strtoull(arg2, NULL, 0));
5802 in_port = ofp_port_to_odp_port(atoi(arg3));
5804 packet = ofpbuf_new(strlen(args) / 2);
5805 arg4 = ofpbuf_put_hex(packet, arg4, NULL);
5806 arg4 += strspn(arg4, " ");
5807 if (*arg4 != '\0') {
5808 unixctl_command_reply(conn, 501, "Trailing garbage in command");
5811 if (packet->size < ETH_HEADER_LEN) {
5812 unixctl_command_reply(conn, 501,
5813 "Packet data too short for Ethernet");
5817 ds_put_cstr(&result, "Packet: ");
5818 s = ofp_packet_to_string(packet->data, packet->size, packet->size);
5819 ds_put_cstr(&result, s);
5822 flow_extract(packet, priority, tun_id, in_port, &flow);
5823 initial_tci = flow.vlan_tci;
5825 unixctl_command_reply(conn, 501, "Bad command syntax");
5829 ds_put_cstr(&result, "Flow: ");
5830 flow_format(&result, &flow);
5831 ds_put_char(&result, '\n');
5833 rule = rule_dpif_lookup(ofproto, &flow, 0);
5834 trace_format_rule(&result, 0, 0, rule);
5836 struct ofproto_trace trace;
5837 struct ofpbuf *odp_actions;
5839 trace.result = &result;
5841 action_xlate_ctx_init(&trace.ctx, ofproto, &flow, initial_tci, packet);
5842 trace.ctx.resubmit_hook = trace_resubmit;
5843 odp_actions = xlate_actions(&trace.ctx,
5844 rule->up.actions, rule->up.n_actions);
5846 ds_put_char(&result, '\n');
5847 trace_format_flow(&result, 0, "Final flow", &trace);
5848 ds_put_cstr(&result, "Datapath actions: ");
5849 format_odp_actions(&result, odp_actions->data, odp_actions->size);
5850 ofpbuf_delete(odp_actions);
5852 if (!trace.ctx.may_set_up_flow) {
5854 ds_put_cstr(&result, "\nThis flow is not cachable.");
5856 ds_put_cstr(&result, "\nThe datapath actions are incomplete--"
5857 "for complete actions, please supply a packet.");
5862 unixctl_command_reply(conn, 200, ds_cstr(&result));
5865 ds_destroy(&result);
5866 ofpbuf_delete(packet);
5867 ofpbuf_uninit(&odp_key);
5872 ofproto_dpif_clog(struct unixctl_conn *conn OVS_UNUSED,
5873 const char *args_ OVS_UNUSED, void *aux OVS_UNUSED)
5876 unixctl_command_reply(conn, 200, NULL);
5880 ofproto_dpif_unclog(struct unixctl_conn *conn OVS_UNUSED,
5881 const char *args_ OVS_UNUSED, void *aux OVS_UNUSED)
5884 unixctl_command_reply(conn, 200, NULL);
5888 ofproto_dpif_unixctl_init(void)
5890 static bool registered;
5896 unixctl_command_register("ofproto/trace",
5897 "bridge {tun_id in_port packet | odp_flow [-generate]}",
5898 ofproto_unixctl_trace, NULL);
5899 unixctl_command_register("fdb/flush", "[bridge]",
5900 ofproto_unixctl_fdb_flush, NULL);
5901 unixctl_command_register("fdb/show", "bridge", ofproto_unixctl_fdb_show,
5903 unixctl_command_register("ofproto/clog", "", ofproto_dpif_clog, NULL);
5904 unixctl_command_register("ofproto/unclog", "", ofproto_dpif_unclog, NULL);
5907 /* Linux VLAN device support (e.g. "eth0.10" for VLAN 10.)
5909 * This is deprecated. It is only for compatibility with broken device drivers
5910 * in old versions of Linux that do not properly support VLANs when VLAN
5911 * devices are not used. When broken device drivers are no longer in
5912 * widespread use, we will delete these interfaces. */
5915 set_realdev(struct ofport *ofport_, uint16_t realdev_ofp_port, int vid)
5917 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport_->ofproto);
5918 struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
5920 if (realdev_ofp_port == ofport->realdev_ofp_port
5921 && vid == ofport->vlandev_vid) {
5925 ofproto->need_revalidate = true;
5927 if (ofport->realdev_ofp_port) {
5930 if (realdev_ofp_port && ofport->bundle) {
5931 /* vlandevs are enslaved to their realdevs, so they are not allowed to
5932 * themselves be part of a bundle. */
5933 bundle_set(ofport->up.ofproto, ofport->bundle, NULL);
5936 ofport->realdev_ofp_port = realdev_ofp_port;
5937 ofport->vlandev_vid = vid;
5939 if (realdev_ofp_port) {
5940 vsp_add(ofport, realdev_ofp_port, vid);
5947 hash_realdev_vid(uint16_t realdev_ofp_port, int vid)
5949 return hash_2words(realdev_ofp_port, vid);
5953 vsp_realdev_to_vlandev(const struct ofproto_dpif *ofproto,
5954 uint32_t realdev_odp_port, ovs_be16 vlan_tci)
5956 if (!hmap_is_empty(&ofproto->realdev_vid_map)) {
5957 uint16_t realdev_ofp_port = odp_port_to_ofp_port(realdev_odp_port);
5958 int vid = vlan_tci_to_vid(vlan_tci);
5959 const struct vlan_splinter *vsp;
5961 HMAP_FOR_EACH_WITH_HASH (vsp, realdev_vid_node,
5962 hash_realdev_vid(realdev_ofp_port, vid),
5963 &ofproto->realdev_vid_map) {
5964 if (vsp->realdev_ofp_port == realdev_ofp_port
5965 && vsp->vid == vid) {
5966 return ofp_port_to_odp_port(vsp->vlandev_ofp_port);
5970 return realdev_odp_port;
5973 static struct vlan_splinter *
5974 vlandev_find(const struct ofproto_dpif *ofproto, uint16_t vlandev_ofp_port)
5976 struct vlan_splinter *vsp;
5978 HMAP_FOR_EACH_WITH_HASH (vsp, vlandev_node, hash_int(vlandev_ofp_port, 0),
5979 &ofproto->vlandev_map) {
5980 if (vsp->vlandev_ofp_port == vlandev_ofp_port) {
5989 vsp_vlandev_to_realdev(const struct ofproto_dpif *ofproto,
5990 uint16_t vlandev_ofp_port, int *vid)
5992 if (!hmap_is_empty(&ofproto->vlandev_map)) {
5993 const struct vlan_splinter *vsp;
5995 vsp = vlandev_find(ofproto, vlandev_ofp_port);
6000 return vsp->realdev_ofp_port;
6007 vsp_remove(struct ofport_dpif *port)
6009 struct ofproto_dpif *ofproto = ofproto_dpif_cast(port->up.ofproto);
6010 struct vlan_splinter *vsp;
6012 vsp = vlandev_find(ofproto, port->up.ofp_port);
6014 hmap_remove(&ofproto->vlandev_map, &vsp->vlandev_node);
6015 hmap_remove(&ofproto->realdev_vid_map, &vsp->realdev_vid_node);
6018 port->realdev_ofp_port = 0;
6020 VLOG_ERR("missing vlan device record");
6025 vsp_add(struct ofport_dpif *port, uint16_t realdev_ofp_port, int vid)
6027 struct ofproto_dpif *ofproto = ofproto_dpif_cast(port->up.ofproto);
6029 if (!vsp_vlandev_to_realdev(ofproto, port->up.ofp_port, NULL)
6030 && (vsp_realdev_to_vlandev(ofproto, realdev_ofp_port, htons(vid))
6031 == realdev_ofp_port)) {
6032 struct vlan_splinter *vsp;
6034 vsp = xmalloc(sizeof *vsp);
6035 hmap_insert(&ofproto->vlandev_map, &vsp->vlandev_node,
6036 hash_int(port->up.ofp_port, 0));
6037 hmap_insert(&ofproto->realdev_vid_map, &vsp->realdev_vid_node,
6038 hash_realdev_vid(realdev_ofp_port, vid));
6039 vsp->realdev_ofp_port = realdev_ofp_port;
6040 vsp->vlandev_ofp_port = port->up.ofp_port;
6043 port->realdev_ofp_port = realdev_ofp_port;
6045 VLOG_ERR("duplicate vlan device record");
6049 const struct ofproto_class ofproto_dpif_class = {
6077 port_is_lacp_current,
6078 NULL, /* rule_choose_table */
6085 rule_modify_actions,
6093 get_cfm_remote_mpids,
6097 get_stp_port_status,
6104 is_mirror_output_bundle,
6105 forward_bpdu_changed,