2 * Copyright (c) 2009, 2010, 2011, 2012 Nicira Networks.
4 * Licensed under the Apache License, Version 2.0 (the "License");
5 * you may not use this file except in compliance with the License.
6 * You may obtain a copy of the License at:
8 * http://www.apache.org/licenses/LICENSE-2.0
10 * Unless required by applicable law or agreed to in writing, software
11 * distributed under the License is distributed on an "AS IS" BASIS,
12 * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
13 * See the License for the specific language governing permissions and
14 * limitations under the License.
19 #include "ofproto/ofproto-provider.h"
26 #include "byte-order.h"
31 #include "dynamic-string.h"
32 #include "fail-open.h"
36 #include "mac-learning.h"
37 #include "meta-flow.h"
38 #include "multipath.h"
45 #include "ofp-print.h"
46 #include "ofproto-dpif-sflow.h"
47 #include "poll-loop.h"
49 #include "unaligned.h"
51 #include "vlan-bitmap.h"
54 VLOG_DEFINE_THIS_MODULE(ofproto_dpif);
56 COVERAGE_DEFINE(ofproto_dpif_ctlr_action);
57 COVERAGE_DEFINE(ofproto_dpif_expired);
58 COVERAGE_DEFINE(ofproto_dpif_no_packet_in);
59 COVERAGE_DEFINE(ofproto_dpif_xlate);
60 COVERAGE_DEFINE(facet_changed_rule);
61 COVERAGE_DEFINE(facet_invalidated);
62 COVERAGE_DEFINE(facet_revalidate);
63 COVERAGE_DEFINE(facet_unexpected);
65 /* Maximum depth of flow table recursion (due to resubmit actions) in a
66 * flow translation. */
67 #define MAX_RESUBMIT_RECURSION 32
69 /* Number of implemented OpenFlow tables. */
70 enum { N_TABLES = 255 };
71 BUILD_ASSERT_DECL(N_TABLES >= 1 && N_TABLES <= 255);
79 long long int used; /* Time last used; time created if not used. */
83 * - Do include packets and bytes from facets that have been deleted or
84 * whose own statistics have been folded into the rule.
86 * - Do include packets and bytes sent "by hand" that were accounted to
87 * the rule without any facet being involved (this is a rare corner
88 * case in rule_execute()).
90 * - Do not include packet or bytes that can be obtained from any facet's
91 * packet_count or byte_count member or that can be obtained from the
92 * datapath by, e.g., dpif_flow_get() for any subfacet.
94 uint64_t packet_count; /* Number of packets received. */
95 uint64_t byte_count; /* Number of bytes received. */
97 tag_type tag; /* Caches rule_calculate_tag() result. */
99 struct list facets; /* List of "struct facet"s. */
102 static struct rule_dpif *rule_dpif_cast(const struct rule *rule)
104 return rule ? CONTAINER_OF(rule, struct rule_dpif, up) : NULL;
107 static struct rule_dpif *rule_dpif_lookup(struct ofproto_dpif *,
108 const struct flow *, uint8_t table);
110 static void flow_push_stats(const struct rule_dpif *, const struct flow *,
111 uint64_t packets, uint64_t bytes,
114 static uint32_t rule_calculate_tag(const struct flow *,
115 const struct flow_wildcards *,
117 static void rule_invalidate(const struct rule_dpif *);
119 #define MAX_MIRRORS 32
120 typedef uint32_t mirror_mask_t;
121 #define MIRROR_MASK_C(X) UINT32_C(X)
122 BUILD_ASSERT_DECL(sizeof(mirror_mask_t) * CHAR_BIT >= MAX_MIRRORS);
124 struct ofproto_dpif *ofproto; /* Owning ofproto. */
125 size_t idx; /* In ofproto's "mirrors" array. */
126 void *aux; /* Key supplied by ofproto's client. */
127 char *name; /* Identifier for log messages. */
129 /* Selection criteria. */
130 struct hmapx srcs; /* Contains "struct ofbundle *"s. */
131 struct hmapx dsts; /* Contains "struct ofbundle *"s. */
132 unsigned long *vlans; /* Bitmap of chosen VLANs, NULL selects all. */
134 /* Output (exactly one of out == NULL and out_vlan == -1 is true). */
135 struct ofbundle *out; /* Output port or NULL. */
136 int out_vlan; /* Output VLAN or -1. */
137 mirror_mask_t dup_mirrors; /* Bitmap of mirrors with the same output. */
140 int64_t packet_count; /* Number of packets sent. */
141 int64_t byte_count; /* Number of bytes sent. */
144 static void mirror_destroy(struct ofmirror *);
145 static void update_mirror_stats(struct ofproto_dpif *ofproto,
146 mirror_mask_t mirrors,
147 uint64_t packets, uint64_t bytes);
150 struct ofproto_dpif *ofproto; /* Owning ofproto. */
151 struct hmap_node hmap_node; /* In struct ofproto's "bundles" hmap. */
152 void *aux; /* Key supplied by ofproto's client. */
153 char *name; /* Identifier for log messages. */
156 struct list ports; /* Contains "struct ofport"s. */
157 enum port_vlan_mode vlan_mode; /* VLAN mode */
158 int vlan; /* -1=trunk port, else a 12-bit VLAN ID. */
159 unsigned long *trunks; /* Bitmap of trunked VLANs, if 'vlan' == -1.
160 * NULL if all VLANs are trunked. */
161 struct lacp *lacp; /* LACP if LACP is enabled, otherwise NULL. */
162 struct bond *bond; /* Nonnull iff more than one port. */
163 bool use_priority_tags; /* Use 802.1p tag for frames in VLAN 0? */
166 bool floodable; /* True if no port has OFPPC_NO_FLOOD set. */
168 /* Port mirroring info. */
169 mirror_mask_t src_mirrors; /* Mirrors triggered when packet received. */
170 mirror_mask_t dst_mirrors; /* Mirrors triggered when packet sent. */
171 mirror_mask_t mirror_out; /* Mirrors that output to this bundle. */
174 static void bundle_remove(struct ofport *);
175 static void bundle_update(struct ofbundle *);
176 static void bundle_destroy(struct ofbundle *);
177 static void bundle_del_port(struct ofport_dpif *);
178 static void bundle_run(struct ofbundle *);
179 static void bundle_wait(struct ofbundle *);
180 static struct ofbundle *lookup_input_bundle(struct ofproto_dpif *,
181 uint16_t in_port, bool warn);
183 /* A controller may use OFPP_NONE as the ingress port to indicate that
184 * it did not arrive on a "real" port. 'ofpp_none_bundle' exists for
185 * when an input bundle is needed for validation (e.g., mirroring or
186 * OFPP_NORMAL processing). It is not connected to an 'ofproto' or have
187 * any 'port' structs, so care must be taken when dealing with it. */
188 static struct ofbundle ofpp_none_bundle = {
190 .vlan_mode = PORT_VLAN_TRUNK
193 static void stp_run(struct ofproto_dpif *ofproto);
194 static void stp_wait(struct ofproto_dpif *ofproto);
196 static bool ofbundle_includes_vlan(const struct ofbundle *, uint16_t vlan);
198 struct action_xlate_ctx {
199 /* action_xlate_ctx_init() initializes these members. */
202 struct ofproto_dpif *ofproto;
204 /* Flow to which the OpenFlow actions apply. xlate_actions() will modify
205 * this flow when actions change header fields. */
208 /* The packet corresponding to 'flow', or a null pointer if we are
209 * revalidating without a packet to refer to. */
210 const struct ofpbuf *packet;
212 /* Should OFPP_NORMAL MAC learning and NXAST_LEARN actions execute? We
213 * want to execute them if we are actually processing a packet, or if we
214 * are accounting for packets that the datapath has processed, but not if
215 * we are just revalidating. */
218 /* Cookie of the currently matching rule, or 0. */
221 /* If nonnull, called just before executing a resubmit action.
223 * This is normally null so the client has to set it manually after
224 * calling action_xlate_ctx_init(). */
225 void (*resubmit_hook)(struct action_xlate_ctx *, struct rule_dpif *);
227 /* xlate_actions() initializes and uses these members. The client might want
228 * to look at them after it returns. */
230 struct ofpbuf *odp_actions; /* Datapath actions. */
231 tag_type tags; /* Tags associated with actions. */
232 bool may_set_up_flow; /* True ordinarily; false if the actions must
233 * be reassessed for every packet. */
234 bool has_learn; /* Actions include NXAST_LEARN? */
235 bool has_normal; /* Actions output to OFPP_NORMAL? */
236 uint16_t nf_output_iface; /* Output interface index for NetFlow. */
237 mirror_mask_t mirrors; /* Bitmap of associated mirrors. */
239 /* xlate_actions() initializes and uses these members, but the client has no
240 * reason to look at them. */
242 int recurse; /* Recursion level, via xlate_table_action. */
243 struct flow base_flow; /* Flow at the last commit. */
244 uint32_t orig_skb_priority; /* Priority when packet arrived. */
245 uint8_t table_id; /* OpenFlow table ID where flow was found. */
246 uint32_t sflow_n_outputs; /* Number of output ports. */
247 uint16_t sflow_odp_port; /* Output port for composing sFlow action. */
248 uint16_t user_cookie_offset;/* Used for user_action_cookie fixup. */
249 bool exit; /* No further actions should be processed. */
252 static void action_xlate_ctx_init(struct action_xlate_ctx *,
253 struct ofproto_dpif *, const struct flow *,
254 ovs_be16 initial_tci, ovs_be64 cookie,
255 const struct ofpbuf *);
256 static struct ofpbuf *xlate_actions(struct action_xlate_ctx *,
257 const union ofp_action *in, size_t n_in);
259 /* An exact-match instantiation of an OpenFlow flow.
261 * A facet associates a "struct flow", which represents the Open vSwitch
262 * userspace idea of an exact-match flow, with one or more subfacets. Each
263 * subfacet tracks the datapath's idea of the exact-match flow equivalent to
264 * the facet. When the kernel module (or other dpif implementation) and Open
265 * vSwitch userspace agree on the definition of a flow key, there is exactly
266 * one subfacet per facet. If the dpif implementation supports more-specific
267 * flow matching than userspace, however, a facet can have more than one
268 * subfacet, each of which corresponds to some distinction in flow that
269 * userspace simply doesn't understand.
271 * Flow expiration works in terms of subfacets, so a facet must have at least
272 * one subfacet or it will never expire, leaking memory. */
275 struct hmap_node hmap_node; /* In owning ofproto's 'facets' hmap. */
276 struct list list_node; /* In owning rule's 'facets' list. */
277 struct rule_dpif *rule; /* Owning rule. */
280 struct list subfacets;
281 long long int used; /* Time last used; time created if not used. */
288 * - Do include packets and bytes sent "by hand", e.g. with
291 * - Do include packets and bytes that were obtained from the datapath
292 * when a subfacet's statistics were reset (e.g. dpif_flow_put() with
293 * DPIF_FP_ZERO_STATS).
295 * - Do not include packets or bytes that can be obtained from the
296 * datapath for any existing subfacet.
298 uint64_t packet_count; /* Number of packets received. */
299 uint64_t byte_count; /* Number of bytes received. */
301 /* Resubmit statistics. */
302 uint64_t prev_packet_count; /* Number of packets from last stats push. */
303 uint64_t prev_byte_count; /* Number of bytes from last stats push. */
304 long long int prev_used; /* Used time from last stats push. */
307 uint64_t accounted_bytes; /* Bytes processed by facet_account(). */
308 struct netflow_flow nf_flow; /* Per-flow NetFlow tracking data. */
310 /* Properties of datapath actions.
312 * Every subfacet has its own actions because actions can differ slightly
313 * between splintered and non-splintered subfacets due to the VLAN tag
314 * being initially different (present vs. absent). All of them have these
315 * properties in common so we just store one copy of them here. */
316 bool may_install; /* Reassess actions for every packet? */
317 bool has_learn; /* Actions include NXAST_LEARN? */
318 bool has_normal; /* Actions output to OFPP_NORMAL? */
319 tag_type tags; /* Tags that would require revalidation. */
320 mirror_mask_t mirrors; /* Bitmap of dependent mirrors. */
323 static struct facet *facet_create(struct rule_dpif *, const struct flow *);
324 static void facet_remove(struct facet *);
325 static void facet_free(struct facet *);
327 static struct facet *facet_find(struct ofproto_dpif *, const struct flow *);
328 static struct facet *facet_lookup_valid(struct ofproto_dpif *,
329 const struct flow *);
330 static bool facet_revalidate(struct facet *);
331 static bool facet_check_consistency(struct facet *);
333 static void facet_flush_stats(struct facet *);
335 static void facet_update_time(struct facet *, long long int used);
336 static void facet_reset_counters(struct facet *);
337 static void facet_push_stats(struct facet *);
338 static void facet_account(struct facet *);
340 static bool facet_is_controller_flow(struct facet *);
342 /* A dpif flow and actions associated with a facet.
344 * See also the large comment on struct facet. */
347 struct hmap_node hmap_node; /* In struct ofproto_dpif 'subfacets' list. */
348 struct list list_node; /* In struct facet's 'facets' list. */
349 struct facet *facet; /* Owning facet. */
353 * To save memory in the common case, 'key' is NULL if 'key_fitness' is
354 * ODP_FIT_PERFECT, that is, odp_flow_key_from_flow() can accurately
355 * regenerate the ODP flow key from ->facet->flow. */
356 enum odp_key_fitness key_fitness;
360 long long int used; /* Time last used; time created if not used. */
362 uint64_t dp_packet_count; /* Last known packet count in the datapath. */
363 uint64_t dp_byte_count; /* Last known byte count in the datapath. */
367 * These should be essentially identical for every subfacet in a facet, but
368 * may differ in trivial ways due to VLAN splinters. */
369 size_t actions_len; /* Number of bytes in actions[]. */
370 struct nlattr *actions; /* Datapath actions. */
372 bool installed; /* Installed in datapath? */
374 /* This value is normally the same as ->facet->flow.vlan_tci. Only VLAN
375 * splinters can cause it to differ. This value should be removed when
376 * the VLAN splinters feature is no longer needed. */
377 ovs_be16 initial_tci; /* Initial VLAN TCI value. */
380 static struct subfacet *subfacet_create(struct facet *, enum odp_key_fitness,
381 const struct nlattr *key,
382 size_t key_len, ovs_be16 initial_tci);
383 static struct subfacet *subfacet_find(struct ofproto_dpif *,
384 const struct nlattr *key, size_t key_len);
385 static void subfacet_destroy(struct subfacet *);
386 static void subfacet_destroy__(struct subfacet *);
387 static void subfacet_get_key(struct subfacet *, struct odputil_keybuf *,
389 static void subfacet_reset_dp_stats(struct subfacet *,
390 struct dpif_flow_stats *);
391 static void subfacet_update_time(struct subfacet *, long long int used);
392 static void subfacet_update_stats(struct subfacet *,
393 const struct dpif_flow_stats *);
394 static void subfacet_make_actions(struct subfacet *,
395 const struct ofpbuf *packet);
396 static int subfacet_install(struct subfacet *,
397 const struct nlattr *actions, size_t actions_len,
398 struct dpif_flow_stats *);
399 static void subfacet_uninstall(struct subfacet *);
405 struct ofbundle *bundle; /* Bundle that contains this port, if any. */
406 struct list bundle_node; /* In struct ofbundle's "ports" list. */
407 struct cfm *cfm; /* Connectivity Fault Management, if any. */
408 tag_type tag; /* Tag associated with this port. */
409 uint32_t bond_stable_id; /* stable_id to use as bond slave, or 0. */
410 bool may_enable; /* May be enabled in bonds. */
413 struct stp_port *stp_port; /* Spanning Tree Protocol, if any. */
414 enum stp_state stp_state; /* Always STP_DISABLED if STP not in use. */
415 long long int stp_state_entered;
417 struct hmap priorities; /* Map of attached 'priority_to_dscp's. */
419 /* Linux VLAN device support (e.g. "eth0.10" for VLAN 10.)
421 * This is deprecated. It is only for compatibility with broken device
422 * drivers in old versions of Linux that do not properly support VLANs when
423 * VLAN devices are not used. When broken device drivers are no longer in
424 * widespread use, we will delete these interfaces. */
425 uint16_t realdev_ofp_port;
429 /* Node in 'ofport_dpif''s 'priorities' map. Used to maintain a map from
430 * 'priority' (the datapath's term for QoS queue) to the dscp bits which all
431 * traffic egressing the 'ofport' with that priority should be marked with. */
432 struct priority_to_dscp {
433 struct hmap_node hmap_node; /* Node in 'ofport_dpif''s 'priorities' map. */
434 uint32_t priority; /* Priority of this queue (see struct flow). */
436 uint8_t dscp; /* DSCP bits to mark outgoing traffic with. */
439 /* Linux VLAN device support (e.g. "eth0.10" for VLAN 10.)
441 * This is deprecated. It is only for compatibility with broken device drivers
442 * in old versions of Linux that do not properly support VLANs when VLAN
443 * devices are not used. When broken device drivers are no longer in
444 * widespread use, we will delete these interfaces. */
445 struct vlan_splinter {
446 struct hmap_node realdev_vid_node;
447 struct hmap_node vlandev_node;
448 uint16_t realdev_ofp_port;
449 uint16_t vlandev_ofp_port;
453 static uint32_t vsp_realdev_to_vlandev(const struct ofproto_dpif *,
454 uint32_t realdev, ovs_be16 vlan_tci);
455 static uint16_t vsp_vlandev_to_realdev(const struct ofproto_dpif *,
456 uint16_t vlandev, int *vid);
457 static void vsp_remove(struct ofport_dpif *);
458 static void vsp_add(struct ofport_dpif *, uint16_t realdev_ofp_port, int vid);
460 static struct ofport_dpif *
461 ofport_dpif_cast(const struct ofport *ofport)
463 assert(ofport->ofproto->ofproto_class == &ofproto_dpif_class);
464 return ofport ? CONTAINER_OF(ofport, struct ofport_dpif, up) : NULL;
467 static void port_run(struct ofport_dpif *);
468 static void port_wait(struct ofport_dpif *);
469 static int set_cfm(struct ofport *, const struct cfm_settings *);
470 static void ofport_clear_priorities(struct ofport_dpif *);
472 struct dpif_completion {
473 struct list list_node;
474 struct ofoperation *op;
477 /* Extra information about a classifier table.
478 * Currently used just for optimized flow revalidation. */
480 /* If either of these is nonnull, then this table has a form that allows
481 * flows to be tagged to avoid revalidating most flows for the most common
482 * kinds of flow table changes. */
483 struct cls_table *catchall_table; /* Table that wildcards all fields. */
484 struct cls_table *other_table; /* Table with any other wildcard set. */
485 uint32_t basis; /* Keeps each table's tags separate. */
488 struct ofproto_dpif {
489 struct hmap_node all_ofproto_dpifs_node; /* In 'all_ofproto_dpifs'. */
498 struct netflow *netflow;
499 struct dpif_sflow *sflow;
500 struct hmap bundles; /* Contains "struct ofbundle"s. */
501 struct mac_learning *ml;
502 struct ofmirror *mirrors[MAX_MIRRORS];
503 bool has_bonded_bundles;
506 struct timer next_expiration;
510 struct hmap subfacets;
513 struct table_dpif tables[N_TABLES];
514 bool need_revalidate;
515 struct tag_set revalidate_set;
517 /* Support for debugging async flow mods. */
518 struct list completions;
520 bool has_bundle_action; /* True when the first bundle action appears. */
521 struct netdev_stats stats; /* To account packets generated and consumed in
526 long long int stp_last_tick;
528 /* VLAN splinters. */
529 struct hmap realdev_vid_map; /* (realdev,vid) -> vlandev. */
530 struct hmap vlandev_map; /* vlandev -> (realdev,vid). */
533 /* Defer flow mod completion until "ovs-appctl ofproto/unclog"? (Useful only
534 * for debugging the asynchronous flow_mod implementation.) */
537 /* All existing ofproto_dpif instances, indexed by ->up.name. */
538 static struct hmap all_ofproto_dpifs = HMAP_INITIALIZER(&all_ofproto_dpifs);
540 static void ofproto_dpif_unixctl_init(void);
542 static struct ofproto_dpif *
543 ofproto_dpif_cast(const struct ofproto *ofproto)
545 assert(ofproto->ofproto_class == &ofproto_dpif_class);
546 return CONTAINER_OF(ofproto, struct ofproto_dpif, up);
549 static struct ofport_dpif *get_ofp_port(struct ofproto_dpif *,
551 static struct ofport_dpif *get_odp_port(struct ofproto_dpif *,
554 /* Packet processing. */
555 static void update_learning_table(struct ofproto_dpif *,
556 const struct flow *, int vlan,
559 #define FLOW_MISS_MAX_BATCH 50
560 static int handle_upcalls(struct ofproto_dpif *, unsigned int max_batch);
562 /* Flow expiration. */
563 static int expire(struct ofproto_dpif *);
566 static void send_netflow_active_timeouts(struct ofproto_dpif *);
569 static int send_packet(const struct ofport_dpif *, struct ofpbuf *packet);
571 compose_sflow_action(const struct ofproto_dpif *, struct ofpbuf *odp_actions,
572 const struct flow *, uint32_t odp_port);
573 static void add_mirror_actions(struct action_xlate_ctx *ctx,
574 const struct flow *flow);
575 /* Global variables. */
576 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
578 /* Factory functions. */
581 enumerate_types(struct sset *types)
583 dp_enumerate_types(types);
587 enumerate_names(const char *type, struct sset *names)
589 return dp_enumerate_names(type, names);
593 del(const char *type, const char *name)
598 error = dpif_open(name, type, &dpif);
600 error = dpif_delete(dpif);
606 /* Basic life-cycle. */
608 static struct ofproto *
611 struct ofproto_dpif *ofproto = xmalloc(sizeof *ofproto);
616 dealloc(struct ofproto *ofproto_)
618 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
623 construct(struct ofproto *ofproto_, int *n_tablesp)
625 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
626 const char *name = ofproto->up.name;
630 error = dpif_create_and_open(name, ofproto->up.type, &ofproto->dpif);
632 VLOG_ERR("failed to open datapath %s: %s", name, strerror(error));
636 ofproto->max_ports = dpif_get_max_ports(ofproto->dpif);
637 ofproto->n_matches = 0;
639 dpif_flow_flush(ofproto->dpif);
640 dpif_recv_purge(ofproto->dpif);
642 error = dpif_recv_set(ofproto->dpif, true);
644 VLOG_ERR("failed to listen on datapath %s: %s", name, strerror(error));
645 dpif_close(ofproto->dpif);
649 ofproto->netflow = NULL;
650 ofproto->sflow = NULL;
652 hmap_init(&ofproto->bundles);
653 ofproto->ml = mac_learning_create();
654 for (i = 0; i < MAX_MIRRORS; i++) {
655 ofproto->mirrors[i] = NULL;
657 ofproto->has_bonded_bundles = false;
659 timer_set_duration(&ofproto->next_expiration, 1000);
661 hmap_init(&ofproto->facets);
662 hmap_init(&ofproto->subfacets);
664 for (i = 0; i < N_TABLES; i++) {
665 struct table_dpif *table = &ofproto->tables[i];
667 table->catchall_table = NULL;
668 table->other_table = NULL;
669 table->basis = random_uint32();
671 ofproto->need_revalidate = false;
672 tag_set_init(&ofproto->revalidate_set);
674 list_init(&ofproto->completions);
676 ofproto_dpif_unixctl_init();
678 ofproto->has_bundle_action = false;
680 hmap_init(&ofproto->vlandev_map);
681 hmap_init(&ofproto->realdev_vid_map);
683 hmap_insert(&all_ofproto_dpifs, &ofproto->all_ofproto_dpifs_node,
684 hash_string(ofproto->up.name, 0));
686 *n_tablesp = N_TABLES;
687 memset(&ofproto->stats, 0, sizeof ofproto->stats);
692 complete_operations(struct ofproto_dpif *ofproto)
694 struct dpif_completion *c, *next;
696 LIST_FOR_EACH_SAFE (c, next, list_node, &ofproto->completions) {
697 ofoperation_complete(c->op, 0);
698 list_remove(&c->list_node);
704 destruct(struct ofproto *ofproto_)
706 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
707 struct rule_dpif *rule, *next_rule;
708 struct oftable *table;
711 hmap_remove(&all_ofproto_dpifs, &ofproto->all_ofproto_dpifs_node);
712 complete_operations(ofproto);
714 OFPROTO_FOR_EACH_TABLE (table, &ofproto->up) {
715 struct cls_cursor cursor;
717 cls_cursor_init(&cursor, &table->cls, NULL);
718 CLS_CURSOR_FOR_EACH_SAFE (rule, next_rule, up.cr, &cursor) {
719 ofproto_rule_destroy(&rule->up);
723 for (i = 0; i < MAX_MIRRORS; i++) {
724 mirror_destroy(ofproto->mirrors[i]);
727 netflow_destroy(ofproto->netflow);
728 dpif_sflow_destroy(ofproto->sflow);
729 hmap_destroy(&ofproto->bundles);
730 mac_learning_destroy(ofproto->ml);
732 hmap_destroy(&ofproto->facets);
733 hmap_destroy(&ofproto->subfacets);
735 hmap_destroy(&ofproto->vlandev_map);
736 hmap_destroy(&ofproto->realdev_vid_map);
738 dpif_close(ofproto->dpif);
742 run_fast(struct ofproto *ofproto_)
744 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
747 /* Handle one or more batches of upcalls, until there's nothing left to do
748 * or until we do a fixed total amount of work.
750 * We do work in batches because it can be much cheaper to set up a number
751 * of flows and fire off their patches all at once. We do multiple batches
752 * because in some cases handling a packet can cause another packet to be
753 * queued almost immediately as part of the return flow. Both
754 * optimizations can make major improvements on some benchmarks and
755 * presumably for real traffic as well. */
757 while (work < FLOW_MISS_MAX_BATCH) {
758 int retval = handle_upcalls(ofproto, FLOW_MISS_MAX_BATCH - work);
768 run(struct ofproto *ofproto_)
770 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
771 struct ofport_dpif *ofport;
772 struct ofbundle *bundle;
776 complete_operations(ofproto);
778 dpif_run(ofproto->dpif);
780 error = run_fast(ofproto_);
785 if (timer_expired(&ofproto->next_expiration)) {
786 int delay = expire(ofproto);
787 timer_set_duration(&ofproto->next_expiration, delay);
790 if (ofproto->netflow) {
791 if (netflow_run(ofproto->netflow)) {
792 send_netflow_active_timeouts(ofproto);
795 if (ofproto->sflow) {
796 dpif_sflow_run(ofproto->sflow);
799 HMAP_FOR_EACH (ofport, up.hmap_node, &ofproto->up.ports) {
802 HMAP_FOR_EACH (bundle, hmap_node, &ofproto->bundles) {
807 mac_learning_run(ofproto->ml, &ofproto->revalidate_set);
809 /* Now revalidate if there's anything to do. */
810 if (ofproto->need_revalidate
811 || !tag_set_is_empty(&ofproto->revalidate_set)) {
812 struct tag_set revalidate_set = ofproto->revalidate_set;
813 bool revalidate_all = ofproto->need_revalidate;
814 struct facet *facet, *next;
816 /* Clear the revalidation flags. */
817 tag_set_init(&ofproto->revalidate_set);
818 ofproto->need_revalidate = false;
820 HMAP_FOR_EACH_SAFE (facet, next, hmap_node, &ofproto->facets) {
822 || tag_set_intersects(&revalidate_set, facet->tags)) {
823 facet_revalidate(facet);
828 /* Check the consistency of a random facet, to aid debugging. */
829 if (!hmap_is_empty(&ofproto->facets) && !ofproto->need_revalidate) {
832 facet = CONTAINER_OF(hmap_random_node(&ofproto->facets),
833 struct facet, hmap_node);
834 if (!tag_set_intersects(&ofproto->revalidate_set, facet->tags)) {
835 if (!facet_check_consistency(facet)) {
836 ofproto->need_revalidate = true;
845 wait(struct ofproto *ofproto_)
847 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
848 struct ofport_dpif *ofport;
849 struct ofbundle *bundle;
851 if (!clogged && !list_is_empty(&ofproto->completions)) {
852 poll_immediate_wake();
855 dpif_wait(ofproto->dpif);
856 dpif_recv_wait(ofproto->dpif);
857 if (ofproto->sflow) {
858 dpif_sflow_wait(ofproto->sflow);
860 if (!tag_set_is_empty(&ofproto->revalidate_set)) {
861 poll_immediate_wake();
863 HMAP_FOR_EACH (ofport, up.hmap_node, &ofproto->up.ports) {
866 HMAP_FOR_EACH (bundle, hmap_node, &ofproto->bundles) {
869 if (ofproto->netflow) {
870 netflow_wait(ofproto->netflow);
872 mac_learning_wait(ofproto->ml);
874 if (ofproto->need_revalidate) {
875 /* Shouldn't happen, but if it does just go around again. */
876 VLOG_DBG_RL(&rl, "need revalidate in ofproto_wait_cb()");
877 poll_immediate_wake();
879 timer_wait(&ofproto->next_expiration);
884 flush(struct ofproto *ofproto_)
886 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
887 struct facet *facet, *next_facet;
889 HMAP_FOR_EACH_SAFE (facet, next_facet, hmap_node, &ofproto->facets) {
890 /* Mark the facet as not installed so that facet_remove() doesn't
891 * bother trying to uninstall it. There is no point in uninstalling it
892 * individually since we are about to blow away all the facets with
893 * dpif_flow_flush(). */
894 struct subfacet *subfacet;
896 LIST_FOR_EACH (subfacet, list_node, &facet->subfacets) {
897 subfacet->installed = false;
898 subfacet->dp_packet_count = 0;
899 subfacet->dp_byte_count = 0;
903 dpif_flow_flush(ofproto->dpif);
907 get_features(struct ofproto *ofproto_ OVS_UNUSED,
908 bool *arp_match_ip, uint32_t *actions)
910 *arp_match_ip = true;
911 *actions = ((1u << OFPAT_OUTPUT) |
912 (1u << OFPAT_SET_VLAN_VID) |
913 (1u << OFPAT_SET_VLAN_PCP) |
914 (1u << OFPAT_STRIP_VLAN) |
915 (1u << OFPAT_SET_DL_SRC) |
916 (1u << OFPAT_SET_DL_DST) |
917 (1u << OFPAT_SET_NW_SRC) |
918 (1u << OFPAT_SET_NW_DST) |
919 (1u << OFPAT_SET_NW_TOS) |
920 (1u << OFPAT_SET_TP_SRC) |
921 (1u << OFPAT_SET_TP_DST) |
922 (1u << OFPAT_ENQUEUE));
926 get_tables(struct ofproto *ofproto_, struct ofp_table_stats *ots)
928 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
929 struct dpif_dp_stats s;
931 strcpy(ots->name, "classifier");
933 dpif_get_dp_stats(ofproto->dpif, &s);
934 put_32aligned_be64(&ots->lookup_count, htonll(s.n_hit + s.n_missed));
935 put_32aligned_be64(&ots->matched_count,
936 htonll(s.n_hit + ofproto->n_matches));
939 static struct ofport *
942 struct ofport_dpif *port = xmalloc(sizeof *port);
947 port_dealloc(struct ofport *port_)
949 struct ofport_dpif *port = ofport_dpif_cast(port_);
954 port_construct(struct ofport *port_)
956 struct ofport_dpif *port = ofport_dpif_cast(port_);
957 struct ofproto_dpif *ofproto = ofproto_dpif_cast(port->up.ofproto);
959 ofproto->need_revalidate = true;
960 port->odp_port = ofp_port_to_odp_port(port->up.ofp_port);
963 port->tag = tag_create_random();
964 port->may_enable = true;
965 port->stp_port = NULL;
966 port->stp_state = STP_DISABLED;
967 hmap_init(&port->priorities);
968 port->realdev_ofp_port = 0;
969 port->vlandev_vid = 0;
971 if (ofproto->sflow) {
972 dpif_sflow_add_port(ofproto->sflow, port_);
979 port_destruct(struct ofport *port_)
981 struct ofport_dpif *port = ofport_dpif_cast(port_);
982 struct ofproto_dpif *ofproto = ofproto_dpif_cast(port->up.ofproto);
984 ofproto->need_revalidate = true;
985 bundle_remove(port_);
986 set_cfm(port_, NULL);
987 if (ofproto->sflow) {
988 dpif_sflow_del_port(ofproto->sflow, port->odp_port);
991 ofport_clear_priorities(port);
992 hmap_destroy(&port->priorities);
996 port_modified(struct ofport *port_)
998 struct ofport_dpif *port = ofport_dpif_cast(port_);
1000 if (port->bundle && port->bundle->bond) {
1001 bond_slave_set_netdev(port->bundle->bond, port, port->up.netdev);
1006 port_reconfigured(struct ofport *port_, ovs_be32 old_config)
1008 struct ofport_dpif *port = ofport_dpif_cast(port_);
1009 struct ofproto_dpif *ofproto = ofproto_dpif_cast(port->up.ofproto);
1010 ovs_be32 changed = old_config ^ port->up.opp.config;
1012 if (changed & htonl(OFPPC_NO_RECV | OFPPC_NO_RECV_STP |
1013 OFPPC_NO_FWD | OFPPC_NO_FLOOD)) {
1014 ofproto->need_revalidate = true;
1016 if (changed & htonl(OFPPC_NO_FLOOD) && port->bundle) {
1017 bundle_update(port->bundle);
1023 set_sflow(struct ofproto *ofproto_,
1024 const struct ofproto_sflow_options *sflow_options)
1026 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1027 struct dpif_sflow *ds = ofproto->sflow;
1029 if (sflow_options) {
1031 struct ofport_dpif *ofport;
1033 ds = ofproto->sflow = dpif_sflow_create(ofproto->dpif);
1034 HMAP_FOR_EACH (ofport, up.hmap_node, &ofproto->up.ports) {
1035 dpif_sflow_add_port(ds, &ofport->up);
1037 ofproto->need_revalidate = true;
1039 dpif_sflow_set_options(ds, sflow_options);
1042 dpif_sflow_destroy(ds);
1043 ofproto->need_revalidate = true;
1044 ofproto->sflow = NULL;
1051 set_cfm(struct ofport *ofport_, const struct cfm_settings *s)
1053 struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
1060 struct ofproto_dpif *ofproto;
1062 ofproto = ofproto_dpif_cast(ofport->up.ofproto);
1063 ofproto->need_revalidate = true;
1064 ofport->cfm = cfm_create(netdev_get_name(ofport->up.netdev));
1067 if (cfm_configure(ofport->cfm, s)) {
1073 cfm_destroy(ofport->cfm);
1079 get_cfm_fault(const struct ofport *ofport_)
1081 struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
1083 return ofport->cfm ? cfm_get_fault(ofport->cfm) : -1;
1087 get_cfm_remote_mpids(const struct ofport *ofport_, const uint64_t **rmps,
1090 struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
1093 cfm_get_remote_mpids(ofport->cfm, rmps, n_rmps);
1100 /* Spanning Tree. */
1103 send_bpdu_cb(struct ofpbuf *pkt, int port_num, void *ofproto_)
1105 struct ofproto_dpif *ofproto = ofproto_;
1106 struct stp_port *sp = stp_get_port(ofproto->stp, port_num);
1107 struct ofport_dpif *ofport;
1109 ofport = stp_port_get_aux(sp);
1111 VLOG_WARN_RL(&rl, "%s: cannot send BPDU on unknown port %d",
1112 ofproto->up.name, port_num);
1114 struct eth_header *eth = pkt->l2;
1116 netdev_get_etheraddr(ofport->up.netdev, eth->eth_src);
1117 if (eth_addr_is_zero(eth->eth_src)) {
1118 VLOG_WARN_RL(&rl, "%s: cannot send BPDU on port %d "
1119 "with unknown MAC", ofproto->up.name, port_num);
1121 send_packet(ofport, pkt);
1127 /* Configures STP on 'ofproto_' using the settings defined in 's'. */
1129 set_stp(struct ofproto *ofproto_, const struct ofproto_stp_settings *s)
1131 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1133 /* Only revalidate flows if the configuration changed. */
1134 if (!s != !ofproto->stp) {
1135 ofproto->need_revalidate = true;
1139 if (!ofproto->stp) {
1140 ofproto->stp = stp_create(ofproto_->name, s->system_id,
1141 send_bpdu_cb, ofproto);
1142 ofproto->stp_last_tick = time_msec();
1145 stp_set_bridge_id(ofproto->stp, s->system_id);
1146 stp_set_bridge_priority(ofproto->stp, s->priority);
1147 stp_set_hello_time(ofproto->stp, s->hello_time);
1148 stp_set_max_age(ofproto->stp, s->max_age);
1149 stp_set_forward_delay(ofproto->stp, s->fwd_delay);
1151 stp_destroy(ofproto->stp);
1152 ofproto->stp = NULL;
1159 get_stp_status(struct ofproto *ofproto_, struct ofproto_stp_status *s)
1161 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1165 s->bridge_id = stp_get_bridge_id(ofproto->stp);
1166 s->designated_root = stp_get_designated_root(ofproto->stp);
1167 s->root_path_cost = stp_get_root_path_cost(ofproto->stp);
1176 update_stp_port_state(struct ofport_dpif *ofport)
1178 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport->up.ofproto);
1179 enum stp_state state;
1181 /* Figure out new state. */
1182 state = ofport->stp_port ? stp_port_get_state(ofport->stp_port)
1186 if (ofport->stp_state != state) {
1190 VLOG_DBG_RL(&rl, "port %s: STP state changed from %s to %s",
1191 netdev_get_name(ofport->up.netdev),
1192 stp_state_name(ofport->stp_state),
1193 stp_state_name(state));
1194 if (stp_learn_in_state(ofport->stp_state)
1195 != stp_learn_in_state(state)) {
1196 /* xxx Learning action flows should also be flushed. */
1197 mac_learning_flush(ofproto->ml, &ofproto->revalidate_set);
1199 fwd_change = stp_forward_in_state(ofport->stp_state)
1200 != stp_forward_in_state(state);
1202 ofproto->need_revalidate = true;
1203 ofport->stp_state = state;
1204 ofport->stp_state_entered = time_msec();
1206 if (fwd_change && ofport->bundle) {
1207 bundle_update(ofport->bundle);
1210 /* Update the STP state bits in the OpenFlow port description. */
1211 of_state = (ofport->up.opp.state & htonl(~OFPPS_STP_MASK))
1212 | htonl(state == STP_LISTENING ? OFPPS_STP_LISTEN
1213 : state == STP_LEARNING ? OFPPS_STP_LEARN
1214 : state == STP_FORWARDING ? OFPPS_STP_FORWARD
1215 : state == STP_BLOCKING ? OFPPS_STP_BLOCK
1217 ofproto_port_set_state(&ofport->up, of_state);
1221 /* Configures STP on 'ofport_' using the settings defined in 's'. The
1222 * caller is responsible for assigning STP port numbers and ensuring
1223 * there are no duplicates. */
1225 set_stp_port(struct ofport *ofport_,
1226 const struct ofproto_port_stp_settings *s)
1228 struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
1229 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport->up.ofproto);
1230 struct stp_port *sp = ofport->stp_port;
1232 if (!s || !s->enable) {
1234 ofport->stp_port = NULL;
1235 stp_port_disable(sp);
1236 update_stp_port_state(ofport);
1239 } else if (sp && stp_port_no(sp) != s->port_num
1240 && ofport == stp_port_get_aux(sp)) {
1241 /* The port-id changed, so disable the old one if it's not
1242 * already in use by another port. */
1243 stp_port_disable(sp);
1246 sp = ofport->stp_port = stp_get_port(ofproto->stp, s->port_num);
1247 stp_port_enable(sp);
1249 stp_port_set_aux(sp, ofport);
1250 stp_port_set_priority(sp, s->priority);
1251 stp_port_set_path_cost(sp, s->path_cost);
1253 update_stp_port_state(ofport);
1259 get_stp_port_status(struct ofport *ofport_,
1260 struct ofproto_port_stp_status *s)
1262 struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
1263 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport->up.ofproto);
1264 struct stp_port *sp = ofport->stp_port;
1266 if (!ofproto->stp || !sp) {
1272 s->port_id = stp_port_get_id(sp);
1273 s->state = stp_port_get_state(sp);
1274 s->sec_in_state = (time_msec() - ofport->stp_state_entered) / 1000;
1275 s->role = stp_port_get_role(sp);
1276 stp_port_get_counts(sp, &s->tx_count, &s->rx_count, &s->error_count);
1282 stp_run(struct ofproto_dpif *ofproto)
1285 long long int now = time_msec();
1286 long long int elapsed = now - ofproto->stp_last_tick;
1287 struct stp_port *sp;
1290 stp_tick(ofproto->stp, MIN(INT_MAX, elapsed));
1291 ofproto->stp_last_tick = now;
1293 while (stp_get_changed_port(ofproto->stp, &sp)) {
1294 struct ofport_dpif *ofport = stp_port_get_aux(sp);
1297 update_stp_port_state(ofport);
1301 if (stp_check_and_reset_fdb_flush(ofproto->stp)) {
1302 mac_learning_flush(ofproto->ml, &ofproto->revalidate_set);
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;
1846 /* The assignment to "port" is unnecessary but makes "grep"ing for
1847 * struct ofport_dpif more effective. */
1848 learning_packet = bond_compose_learning_packet(bundle->bond,
1852 ret = send_packet(port, learning_packet);
1853 ofpbuf_delete(learning_packet);
1863 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
1864 VLOG_WARN_RL(&rl, "bond %s: %d errors sending %d gratuitous learning "
1865 "packets, last error was: %s",
1866 bundle->name, n_errors, n_packets, strerror(error));
1868 VLOG_DBG("bond %s: sent %d gratuitous learning packets",
1869 bundle->name, n_packets);
1874 bundle_run(struct ofbundle *bundle)
1877 lacp_run(bundle->lacp, send_pdu_cb);
1880 struct ofport_dpif *port;
1882 LIST_FOR_EACH (port, bundle_node, &bundle->ports) {
1883 bond_slave_set_may_enable(bundle->bond, port, port->may_enable);
1886 bond_run(bundle->bond, &bundle->ofproto->revalidate_set,
1887 lacp_status(bundle->lacp));
1888 if (bond_should_send_learning_packets(bundle->bond)) {
1889 bundle_send_learning_packets(bundle);
1895 bundle_wait(struct ofbundle *bundle)
1898 lacp_wait(bundle->lacp);
1901 bond_wait(bundle->bond);
1908 mirror_scan(struct ofproto_dpif *ofproto)
1912 for (idx = 0; idx < MAX_MIRRORS; idx++) {
1913 if (!ofproto->mirrors[idx]) {
1920 static struct ofmirror *
1921 mirror_lookup(struct ofproto_dpif *ofproto, void *aux)
1925 for (i = 0; i < MAX_MIRRORS; i++) {
1926 struct ofmirror *mirror = ofproto->mirrors[i];
1927 if (mirror && mirror->aux == aux) {
1935 /* Update the 'dup_mirrors' member of each of the ofmirrors in 'ofproto'. */
1937 mirror_update_dups(struct ofproto_dpif *ofproto)
1941 for (i = 0; i < MAX_MIRRORS; i++) {
1942 struct ofmirror *m = ofproto->mirrors[i];
1945 m->dup_mirrors = MIRROR_MASK_C(1) << i;
1949 for (i = 0; i < MAX_MIRRORS; i++) {
1950 struct ofmirror *m1 = ofproto->mirrors[i];
1957 for (j = i + 1; j < MAX_MIRRORS; j++) {
1958 struct ofmirror *m2 = ofproto->mirrors[j];
1960 if (m2 && m1->out == m2->out && m1->out_vlan == m2->out_vlan) {
1961 m1->dup_mirrors |= MIRROR_MASK_C(1) << j;
1962 m2->dup_mirrors |= m1->dup_mirrors;
1969 mirror_set(struct ofproto *ofproto_, void *aux,
1970 const struct ofproto_mirror_settings *s)
1972 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1973 mirror_mask_t mirror_bit;
1974 struct ofbundle *bundle;
1975 struct ofmirror *mirror;
1976 struct ofbundle *out;
1977 struct hmapx srcs; /* Contains "struct ofbundle *"s. */
1978 struct hmapx dsts; /* Contains "struct ofbundle *"s. */
1981 mirror = mirror_lookup(ofproto, aux);
1983 mirror_destroy(mirror);
1989 idx = mirror_scan(ofproto);
1991 VLOG_WARN("bridge %s: maximum of %d port mirrors reached, "
1993 ofproto->up.name, MAX_MIRRORS, s->name);
1997 mirror = ofproto->mirrors[idx] = xzalloc(sizeof *mirror);
1998 mirror->ofproto = ofproto;
2001 mirror->out_vlan = -1;
2002 mirror->name = NULL;
2005 if (!mirror->name || strcmp(s->name, mirror->name)) {
2007 mirror->name = xstrdup(s->name);
2010 /* Get the new configuration. */
2011 if (s->out_bundle) {
2012 out = bundle_lookup(ofproto, s->out_bundle);
2014 mirror_destroy(mirror);
2020 out_vlan = s->out_vlan;
2022 bundle_lookup_multiple(ofproto, s->srcs, s->n_srcs, &srcs);
2023 bundle_lookup_multiple(ofproto, s->dsts, s->n_dsts, &dsts);
2025 /* If the configuration has not changed, do nothing. */
2026 if (hmapx_equals(&srcs, &mirror->srcs)
2027 && hmapx_equals(&dsts, &mirror->dsts)
2028 && vlan_bitmap_equal(mirror->vlans, s->src_vlans)
2029 && mirror->out == out
2030 && mirror->out_vlan == out_vlan)
2032 hmapx_destroy(&srcs);
2033 hmapx_destroy(&dsts);
2037 hmapx_swap(&srcs, &mirror->srcs);
2038 hmapx_destroy(&srcs);
2040 hmapx_swap(&dsts, &mirror->dsts);
2041 hmapx_destroy(&dsts);
2043 free(mirror->vlans);
2044 mirror->vlans = vlan_bitmap_clone(s->src_vlans);
2047 mirror->out_vlan = out_vlan;
2049 /* Update bundles. */
2050 mirror_bit = MIRROR_MASK_C(1) << mirror->idx;
2051 HMAP_FOR_EACH (bundle, hmap_node, &mirror->ofproto->bundles) {
2052 if (hmapx_contains(&mirror->srcs, bundle)) {
2053 bundle->src_mirrors |= mirror_bit;
2055 bundle->src_mirrors &= ~mirror_bit;
2058 if (hmapx_contains(&mirror->dsts, bundle)) {
2059 bundle->dst_mirrors |= mirror_bit;
2061 bundle->dst_mirrors &= ~mirror_bit;
2064 if (mirror->out == bundle) {
2065 bundle->mirror_out |= mirror_bit;
2067 bundle->mirror_out &= ~mirror_bit;
2071 ofproto->need_revalidate = true;
2072 mac_learning_flush(ofproto->ml, &ofproto->revalidate_set);
2073 mirror_update_dups(ofproto);
2079 mirror_destroy(struct ofmirror *mirror)
2081 struct ofproto_dpif *ofproto;
2082 mirror_mask_t mirror_bit;
2083 struct ofbundle *bundle;
2089 ofproto = mirror->ofproto;
2090 ofproto->need_revalidate = true;
2091 mac_learning_flush(ofproto->ml, &ofproto->revalidate_set);
2093 mirror_bit = MIRROR_MASK_C(1) << mirror->idx;
2094 HMAP_FOR_EACH (bundle, hmap_node, &ofproto->bundles) {
2095 bundle->src_mirrors &= ~mirror_bit;
2096 bundle->dst_mirrors &= ~mirror_bit;
2097 bundle->mirror_out &= ~mirror_bit;
2100 hmapx_destroy(&mirror->srcs);
2101 hmapx_destroy(&mirror->dsts);
2102 free(mirror->vlans);
2104 ofproto->mirrors[mirror->idx] = NULL;
2108 mirror_update_dups(ofproto);
2112 mirror_get_stats(struct ofproto *ofproto_, void *aux,
2113 uint64_t *packets, uint64_t *bytes)
2115 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
2116 struct ofmirror *mirror = mirror_lookup(ofproto, aux);
2119 *packets = *bytes = UINT64_MAX;
2123 *packets = mirror->packet_count;
2124 *bytes = mirror->byte_count;
2130 set_flood_vlans(struct ofproto *ofproto_, unsigned long *flood_vlans)
2132 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
2133 if (mac_learning_set_flood_vlans(ofproto->ml, flood_vlans)) {
2134 mac_learning_flush(ofproto->ml, &ofproto->revalidate_set);
2140 is_mirror_output_bundle(const struct ofproto *ofproto_, void *aux)
2142 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
2143 struct ofbundle *bundle = bundle_lookup(ofproto, aux);
2144 return bundle && bundle->mirror_out != 0;
2148 forward_bpdu_changed(struct ofproto *ofproto_)
2150 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
2151 /* Revalidate cached flows whenever forward_bpdu option changes. */
2152 ofproto->need_revalidate = true;
2157 static struct ofport_dpif *
2158 get_ofp_port(struct ofproto_dpif *ofproto, uint16_t ofp_port)
2160 struct ofport *ofport = ofproto_get_port(&ofproto->up, ofp_port);
2161 return ofport ? ofport_dpif_cast(ofport) : NULL;
2164 static struct ofport_dpif *
2165 get_odp_port(struct ofproto_dpif *ofproto, uint32_t odp_port)
2167 return get_ofp_port(ofproto, odp_port_to_ofp_port(odp_port));
2171 ofproto_port_from_dpif_port(struct ofproto_port *ofproto_port,
2172 struct dpif_port *dpif_port)
2174 ofproto_port->name = dpif_port->name;
2175 ofproto_port->type = dpif_port->type;
2176 ofproto_port->ofp_port = odp_port_to_ofp_port(dpif_port->port_no);
2180 port_run(struct ofport_dpif *ofport)
2182 bool enable = netdev_get_carrier(ofport->up.netdev);
2185 cfm_run(ofport->cfm);
2187 if (cfm_should_send_ccm(ofport->cfm)) {
2188 struct ofpbuf packet;
2190 ofpbuf_init(&packet, 0);
2191 cfm_compose_ccm(ofport->cfm, &packet, ofport->up.opp.hw_addr);
2192 send_packet(ofport, &packet);
2193 ofpbuf_uninit(&packet);
2196 enable = enable && !cfm_get_fault(ofport->cfm)
2197 && cfm_get_opup(ofport->cfm);
2200 if (ofport->bundle) {
2201 enable = enable && lacp_slave_may_enable(ofport->bundle->lacp, ofport);
2204 if (ofport->may_enable != enable) {
2205 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport->up.ofproto);
2207 if (ofproto->has_bundle_action) {
2208 ofproto->need_revalidate = true;
2212 ofport->may_enable = enable;
2216 port_wait(struct ofport_dpif *ofport)
2219 cfm_wait(ofport->cfm);
2224 port_query_by_name(const struct ofproto *ofproto_, const char *devname,
2225 struct ofproto_port *ofproto_port)
2227 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
2228 struct dpif_port dpif_port;
2231 error = dpif_port_query_by_name(ofproto->dpif, devname, &dpif_port);
2233 ofproto_port_from_dpif_port(ofproto_port, &dpif_port);
2239 port_add(struct ofproto *ofproto_, struct netdev *netdev, uint16_t *ofp_portp)
2241 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
2245 error = dpif_port_add(ofproto->dpif, netdev, &odp_port);
2247 *ofp_portp = odp_port_to_ofp_port(odp_port);
2253 port_del(struct ofproto *ofproto_, uint16_t ofp_port)
2255 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
2258 error = dpif_port_del(ofproto->dpif, ofp_port_to_odp_port(ofp_port));
2260 struct ofport_dpif *ofport = get_ofp_port(ofproto, ofp_port);
2262 /* The caller is going to close ofport->up.netdev. If this is a
2263 * bonded port, then the bond is using that netdev, so remove it
2264 * from the bond. The client will need to reconfigure everything
2265 * after deleting ports, so then the slave will get re-added. */
2266 bundle_remove(&ofport->up);
2273 port_get_stats(const struct ofport *ofport_, struct netdev_stats *stats)
2275 struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
2278 error = netdev_get_stats(ofport->up.netdev, stats);
2280 if (!error && ofport->odp_port == OVSP_LOCAL) {
2281 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport->up.ofproto);
2283 /* ofproto->stats.tx_packets represents packets that we created
2284 * internally and sent to some port (e.g. packets sent with
2285 * send_packet()). Account for them as if they had come from
2286 * OFPP_LOCAL and got forwarded. */
2288 if (stats->rx_packets != UINT64_MAX) {
2289 stats->rx_packets += ofproto->stats.tx_packets;
2292 if (stats->rx_bytes != UINT64_MAX) {
2293 stats->rx_bytes += ofproto->stats.tx_bytes;
2296 /* ofproto->stats.rx_packets represents packets that were received on
2297 * some port and we processed internally and dropped (e.g. STP).
2298 * Account fro them as if they had been forwarded to OFPP_LOCAL. */
2300 if (stats->tx_packets != UINT64_MAX) {
2301 stats->tx_packets += ofproto->stats.rx_packets;
2304 if (stats->tx_bytes != UINT64_MAX) {
2305 stats->tx_bytes += ofproto->stats.rx_bytes;
2312 /* Account packets for LOCAL port. */
2314 ofproto_update_local_port_stats(const struct ofproto *ofproto_,
2315 size_t tx_size, size_t rx_size)
2317 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
2320 ofproto->stats.rx_packets++;
2321 ofproto->stats.rx_bytes += rx_size;
2324 ofproto->stats.tx_packets++;
2325 ofproto->stats.tx_bytes += tx_size;
2329 struct port_dump_state {
2330 struct dpif_port_dump dump;
2335 port_dump_start(const struct ofproto *ofproto_, void **statep)
2337 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
2338 struct port_dump_state *state;
2340 *statep = state = xmalloc(sizeof *state);
2341 dpif_port_dump_start(&state->dump, ofproto->dpif);
2342 state->done = false;
2347 port_dump_next(const struct ofproto *ofproto_ OVS_UNUSED, void *state_,
2348 struct ofproto_port *port)
2350 struct port_dump_state *state = state_;
2351 struct dpif_port dpif_port;
2353 if (dpif_port_dump_next(&state->dump, &dpif_port)) {
2354 ofproto_port_from_dpif_port(port, &dpif_port);
2357 int error = dpif_port_dump_done(&state->dump);
2359 return error ? error : EOF;
2364 port_dump_done(const struct ofproto *ofproto_ OVS_UNUSED, void *state_)
2366 struct port_dump_state *state = state_;
2369 dpif_port_dump_done(&state->dump);
2376 port_poll(const struct ofproto *ofproto_, char **devnamep)
2378 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
2379 return dpif_port_poll(ofproto->dpif, devnamep);
2383 port_poll_wait(const struct ofproto *ofproto_)
2385 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
2386 dpif_port_poll_wait(ofproto->dpif);
2390 port_is_lacp_current(const struct ofport *ofport_)
2392 const struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
2393 return (ofport->bundle && ofport->bundle->lacp
2394 ? lacp_slave_is_current(ofport->bundle->lacp, ofport)
2398 /* Upcall handling. */
2400 /* Flow miss batching.
2402 * Some dpifs implement operations faster when you hand them off in a batch.
2403 * To allow batching, "struct flow_miss" queues the dpif-related work needed
2404 * for a given flow. Each "struct flow_miss" corresponds to sending one or
2405 * more packets, plus possibly installing the flow in the dpif.
2407 * So far we only batch the operations that affect flow setup time the most.
2408 * It's possible to batch more than that, but the benefit might be minimal. */
2410 struct hmap_node hmap_node;
2412 enum odp_key_fitness key_fitness;
2413 const struct nlattr *key;
2415 ovs_be16 initial_tci;
2416 struct list packets;
2419 struct flow_miss_op {
2420 struct dpif_op dpif_op;
2421 struct subfacet *subfacet;
2424 /* Sends an OFPT_PACKET_IN message for 'packet' of type OFPR_NO_MATCH to each
2425 * OpenFlow controller as necessary according to their individual
2426 * configurations. */
2428 send_packet_in_miss(struct ofproto_dpif *ofproto, const struct ofpbuf *packet,
2429 const struct flow *flow)
2431 struct ofputil_packet_in pin;
2433 pin.packet = packet->data;
2434 pin.packet_len = packet->size;
2435 pin.total_len = packet->size;
2436 pin.reason = OFPR_NO_MATCH;
2441 pin.buffer_id = 0; /* not yet known */
2442 pin.send_len = 0; /* not used for flow table misses */
2444 flow_get_metadata(flow, &pin.fmd);
2446 /* Registers aren't meaningful on a miss. */
2447 memset(pin.fmd.reg_masks, 0, sizeof pin.fmd.reg_masks);
2449 connmgr_send_packet_in(ofproto->up.connmgr, &pin, flow);
2453 process_special(struct ofproto_dpif *ofproto, const struct flow *flow,
2454 const struct ofpbuf *packet)
2456 struct ofport_dpif *ofport = get_ofp_port(ofproto, flow->in_port);
2462 if (ofport->cfm && cfm_should_process_flow(ofport->cfm, flow)) {
2464 cfm_process_heartbeat(ofport->cfm, packet);
2467 } else if (ofport->bundle && ofport->bundle->lacp
2468 && flow->dl_type == htons(ETH_TYPE_LACP)) {
2470 lacp_process_packet(ofport->bundle->lacp, ofport, packet);
2473 } else if (ofproto->stp && stp_should_process_flow(flow)) {
2475 stp_process_packet(ofport, packet);
2482 static struct flow_miss *
2483 flow_miss_create(struct hmap *todo, const struct flow *flow,
2484 enum odp_key_fitness key_fitness,
2485 const struct nlattr *key, size_t key_len,
2486 ovs_be16 initial_tci)
2488 uint32_t hash = flow_hash(flow, 0);
2489 struct flow_miss *miss;
2491 HMAP_FOR_EACH_WITH_HASH (miss, hmap_node, hash, todo) {
2492 if (flow_equal(&miss->flow, flow)) {
2497 miss = xmalloc(sizeof *miss);
2498 hmap_insert(todo, &miss->hmap_node, hash);
2500 miss->key_fitness = key_fitness;
2502 miss->key_len = key_len;
2503 miss->initial_tci = initial_tci;
2504 list_init(&miss->packets);
2509 handle_flow_miss(struct ofproto_dpif *ofproto, struct flow_miss *miss,
2510 struct flow_miss_op *ops, size_t *n_ops)
2512 const struct flow *flow = &miss->flow;
2513 struct ofpbuf *packet, *next_packet;
2514 struct subfacet *subfacet;
2515 struct facet *facet;
2517 facet = facet_lookup_valid(ofproto, flow);
2519 struct rule_dpif *rule;
2521 rule = rule_dpif_lookup(ofproto, flow, 0);
2523 /* Don't send a packet-in if OFPPC_NO_PACKET_IN asserted. */
2524 struct ofport_dpif *port = get_ofp_port(ofproto, flow->in_port);
2526 if (port->up.opp.config & htonl(OFPPC_NO_PACKET_IN)) {
2527 COVERAGE_INC(ofproto_dpif_no_packet_in);
2528 /* XXX install 'drop' flow entry */
2532 VLOG_WARN_RL(&rl, "packet-in on unknown port %"PRIu16,
2536 LIST_FOR_EACH (packet, list_node, &miss->packets) {
2537 send_packet_in_miss(ofproto, packet, flow);
2543 facet = facet_create(rule, flow);
2546 subfacet = subfacet_create(facet,
2547 miss->key_fitness, miss->key, miss->key_len,
2550 LIST_FOR_EACH_SAFE (packet, next_packet, list_node, &miss->packets) {
2551 struct dpif_flow_stats stats;
2552 struct flow_miss_op *op;
2553 struct dpif_execute *execute;
2555 ofproto->n_matches++;
2557 if (facet->rule->up.cr.priority == FAIL_OPEN_PRIORITY) {
2559 * Extra-special case for fail-open mode.
2561 * We are in fail-open mode and the packet matched the fail-open
2562 * rule, but we are connected to a controller too. We should send
2563 * the packet up to the controller in the hope that it will try to
2564 * set up a flow and thereby allow us to exit fail-open.
2566 * See the top-level comment in fail-open.c for more information.
2568 send_packet_in_miss(ofproto, packet, flow);
2571 if (!facet->may_install || !subfacet->actions) {
2572 subfacet_make_actions(subfacet, packet);
2575 dpif_flow_stats_extract(&facet->flow, packet, &stats);
2576 subfacet_update_stats(subfacet, &stats);
2578 if (!subfacet->actions_len) {
2579 /* No actions to execute, so skip talking to the dpif. */
2583 if (flow->vlan_tci != subfacet->initial_tci) {
2584 /* This packet was received on a VLAN splinter port. We added
2585 * a VLAN to the packet to make the packet resemble the flow,
2586 * but the actions were composed assuming that the packet
2587 * contained no VLAN. So, we must remove the VLAN header from
2588 * the packet before trying to execute the actions. */
2589 eth_pop_vlan(packet);
2592 op = &ops[(*n_ops)++];
2593 execute = &op->dpif_op.u.execute;
2594 op->subfacet = subfacet;
2595 op->dpif_op.type = DPIF_OP_EXECUTE;
2596 execute->key = miss->key;
2597 execute->key_len = miss->key_len;
2598 execute->actions = (facet->may_install
2600 : xmemdup(subfacet->actions,
2601 subfacet->actions_len));
2602 execute->actions_len = subfacet->actions_len;
2603 execute->packet = packet;
2606 if (facet->may_install && subfacet->key_fitness != ODP_FIT_TOO_LITTLE) {
2607 struct flow_miss_op *op = &ops[(*n_ops)++];
2608 struct dpif_flow_put *put = &op->dpif_op.u.flow_put;
2610 op->subfacet = subfacet;
2611 op->dpif_op.type = DPIF_OP_FLOW_PUT;
2612 put->flags = DPIF_FP_CREATE | DPIF_FP_MODIFY;
2613 put->key = miss->key;
2614 put->key_len = miss->key_len;
2615 put->actions = subfacet->actions;
2616 put->actions_len = subfacet->actions_len;
2621 /* Like odp_flow_key_to_flow(), this function converts the 'key_len' bytes of
2622 * OVS_KEY_ATTR_* attributes in 'key' to a flow structure in 'flow' and returns
2623 * an ODP_FIT_* value that indicates how well 'key' fits our expectations for
2624 * what a flow key should contain.
2626 * This function also includes some logic to help make VLAN splinters
2627 * transparent to the rest of the upcall processing logic. In particular, if
2628 * the extracted in_port is a VLAN splinter port, it replaces flow->in_port by
2629 * the "real" port, sets flow->vlan_tci correctly for the VLAN of the VLAN
2630 * splinter port, and pushes a VLAN header onto 'packet' (if it is nonnull).
2632 * Sets '*initial_tci' to the VLAN TCI with which the packet was really
2633 * received, that is, the actual VLAN TCI extracted by odp_flow_key_to_flow().
2634 * (This differs from the value returned in flow->vlan_tci only for packets
2635 * received on VLAN splinters.)
2637 static enum odp_key_fitness
2638 ofproto_dpif_extract_flow_key(const struct ofproto_dpif *ofproto,
2639 const struct nlattr *key, size_t key_len,
2640 struct flow *flow, ovs_be16 *initial_tci,
2641 struct ofpbuf *packet)
2643 enum odp_key_fitness fitness;
2647 fitness = odp_flow_key_to_flow(key, key_len, flow);
2648 if (fitness == ODP_FIT_ERROR) {
2651 *initial_tci = flow->vlan_tci;
2653 realdev = vsp_vlandev_to_realdev(ofproto, flow->in_port, &vid);
2655 /* Cause the flow to be processed as if it came in on the real device
2656 * with the VLAN device's VLAN ID. */
2657 flow->in_port = realdev;
2658 flow->vlan_tci = htons((vid & VLAN_VID_MASK) | VLAN_CFI);
2660 /* Make the packet resemble the flow, so that it gets sent to an
2661 * OpenFlow controller properly, so that it looks correct for
2662 * sFlow, and so that flow_extract() will get the correct vlan_tci
2663 * if it is called on 'packet'.
2665 * The allocated space inside 'packet' probably also contains
2666 * 'key', that is, both 'packet' and 'key' are probably part of a
2667 * struct dpif_upcall (see the large comment on that structure
2668 * definition), so pushing data on 'packet' is in general not a
2669 * good idea since it could overwrite 'key' or free it as a side
2670 * effect. However, it's OK in this special case because we know
2671 * that 'packet' is inside a Netlink attribute: pushing 4 bytes
2672 * will just overwrite the 4-byte "struct nlattr", which is fine
2673 * since we don't need that header anymore. */
2674 eth_push_vlan(packet, flow->vlan_tci);
2677 /* Let the caller know that we can't reproduce 'key' from 'flow'. */
2678 if (fitness == ODP_FIT_PERFECT) {
2679 fitness = ODP_FIT_TOO_MUCH;
2687 handle_miss_upcalls(struct ofproto_dpif *ofproto, struct dpif_upcall *upcalls,
2690 struct dpif_upcall *upcall;
2691 struct flow_miss *miss, *next_miss;
2692 struct flow_miss_op flow_miss_ops[FLOW_MISS_MAX_BATCH * 2];
2693 struct dpif_op *dpif_ops[FLOW_MISS_MAX_BATCH * 2];
2702 /* Construct the to-do list.
2704 * This just amounts to extracting the flow from each packet and sticking
2705 * the packets that have the same flow in the same "flow_miss" structure so
2706 * that we can process them together. */
2708 for (upcall = upcalls; upcall < &upcalls[n_upcalls]; upcall++) {
2709 enum odp_key_fitness fitness;
2710 struct flow_miss *miss;
2711 ovs_be16 initial_tci;
2714 /* Obtain metadata and check userspace/kernel agreement on flow match,
2715 * then set 'flow''s header pointers. */
2716 fitness = ofproto_dpif_extract_flow_key(ofproto,
2717 upcall->key, upcall->key_len,
2718 &flow, &initial_tci,
2720 if (fitness == ODP_FIT_ERROR) {
2721 ofpbuf_delete(upcall->packet);
2724 flow_extract(upcall->packet, flow.skb_priority, flow.tun_id,
2725 flow.in_port, &flow);
2727 /* Handle 802.1ag, LACP, and STP specially. */
2728 if (process_special(ofproto, &flow, upcall->packet)) {
2729 ofproto_update_local_port_stats(&ofproto->up,
2730 0, upcall->packet->size);
2731 ofpbuf_delete(upcall->packet);
2732 ofproto->n_matches++;
2736 /* Add other packets to a to-do list. */
2737 miss = flow_miss_create(&todo, &flow, fitness,
2738 upcall->key, upcall->key_len, initial_tci);
2739 list_push_back(&miss->packets, &upcall->packet->list_node);
2742 /* Process each element in the to-do list, constructing the set of
2743 * operations to batch. */
2745 HMAP_FOR_EACH (miss, hmap_node, &todo) {
2746 handle_flow_miss(ofproto, miss, flow_miss_ops, &n_ops);
2748 assert(n_ops <= ARRAY_SIZE(flow_miss_ops));
2750 /* Execute batch. */
2751 for (i = 0; i < n_ops; i++) {
2752 dpif_ops[i] = &flow_miss_ops[i].dpif_op;
2754 dpif_operate(ofproto->dpif, dpif_ops, n_ops);
2756 /* Free memory and update facets. */
2757 for (i = 0; i < n_ops; i++) {
2758 struct flow_miss_op *op = &flow_miss_ops[i];
2759 struct dpif_execute *execute;
2761 switch (op->dpif_op.type) {
2762 case DPIF_OP_EXECUTE:
2763 execute = &op->dpif_op.u.execute;
2764 if (op->subfacet->actions != execute->actions) {
2765 free((struct nlattr *) execute->actions);
2769 case DPIF_OP_FLOW_PUT:
2770 if (!op->dpif_op.error) {
2771 op->subfacet->installed = true;
2776 HMAP_FOR_EACH_SAFE (miss, next_miss, hmap_node, &todo) {
2777 ofpbuf_list_delete(&miss->packets);
2778 hmap_remove(&todo, &miss->hmap_node);
2781 hmap_destroy(&todo);
2785 handle_userspace_upcall(struct ofproto_dpif *ofproto,
2786 struct dpif_upcall *upcall)
2788 struct user_action_cookie cookie;
2789 enum odp_key_fitness fitness;
2790 ovs_be16 initial_tci;
2793 memcpy(&cookie, &upcall->userdata, sizeof(cookie));
2795 fitness = ofproto_dpif_extract_flow_key(ofproto, upcall->key,
2796 upcall->key_len, &flow,
2797 &initial_tci, upcall->packet);
2798 if (fitness == ODP_FIT_ERROR) {
2799 ofpbuf_delete(upcall->packet);
2803 if (cookie.type == USER_ACTION_COOKIE_SFLOW) {
2804 if (ofproto->sflow) {
2805 dpif_sflow_received(ofproto->sflow, upcall->packet, &flow,
2809 VLOG_WARN_RL(&rl, "invalid user cookie : 0x%"PRIx64, upcall->userdata);
2811 ofpbuf_delete(upcall->packet);
2815 handle_upcalls(struct ofproto_dpif *ofproto, unsigned int max_batch)
2817 struct dpif_upcall misses[FLOW_MISS_MAX_BATCH];
2821 assert (max_batch <= FLOW_MISS_MAX_BATCH);
2824 for (i = 0; i < max_batch; i++) {
2825 struct dpif_upcall *upcall = &misses[n_misses];
2828 error = dpif_recv(ofproto->dpif, upcall);
2833 switch (upcall->type) {
2834 case DPIF_UC_ACTION:
2835 handle_userspace_upcall(ofproto, upcall);
2839 /* Handle it later. */
2843 case DPIF_N_UC_TYPES:
2845 VLOG_WARN_RL(&rl, "upcall has unexpected type %"PRIu32,
2851 handle_miss_upcalls(ofproto, misses, n_misses);
2856 /* Flow expiration. */
2858 static int subfacet_max_idle(const struct ofproto_dpif *);
2859 static void update_stats(struct ofproto_dpif *);
2860 static void rule_expire(struct rule_dpif *);
2861 static void expire_subfacets(struct ofproto_dpif *, int dp_max_idle);
2863 /* This function is called periodically by run(). Its job is to collect
2864 * updates for the flows that have been installed into the datapath, most
2865 * importantly when they last were used, and then use that information to
2866 * expire flows that have not been used recently.
2868 * Returns the number of milliseconds after which it should be called again. */
2870 expire(struct ofproto_dpif *ofproto)
2872 struct rule_dpif *rule, *next_rule;
2873 struct oftable *table;
2876 /* Update stats for each flow in the datapath. */
2877 update_stats(ofproto);
2879 /* Expire subfacets that have been idle too long. */
2880 dp_max_idle = subfacet_max_idle(ofproto);
2881 expire_subfacets(ofproto, dp_max_idle);
2883 /* Expire OpenFlow flows whose idle_timeout or hard_timeout has passed. */
2884 OFPROTO_FOR_EACH_TABLE (table, &ofproto->up) {
2885 struct cls_cursor cursor;
2887 cls_cursor_init(&cursor, &table->cls, NULL);
2888 CLS_CURSOR_FOR_EACH_SAFE (rule, next_rule, up.cr, &cursor) {
2893 /* All outstanding data in existing flows has been accounted, so it's a
2894 * good time to do bond rebalancing. */
2895 if (ofproto->has_bonded_bundles) {
2896 struct ofbundle *bundle;
2898 HMAP_FOR_EACH (bundle, hmap_node, &ofproto->bundles) {
2900 bond_rebalance(bundle->bond, &ofproto->revalidate_set);
2905 return MIN(dp_max_idle, 1000);
2908 /* Update 'packet_count', 'byte_count', and 'used' members of installed facets.
2910 * This function also pushes statistics updates to rules which each facet
2911 * resubmits into. Generally these statistics will be accurate. However, if a
2912 * facet changes the rule it resubmits into at some time in between
2913 * update_stats() runs, it is possible that statistics accrued to the
2914 * old rule will be incorrectly attributed to the new rule. This could be
2915 * avoided by calling update_stats() whenever rules are created or
2916 * deleted. However, the performance impact of making so many calls to the
2917 * datapath do not justify the benefit of having perfectly accurate statistics.
2920 update_stats(struct ofproto_dpif *p)
2922 const struct dpif_flow_stats *stats;
2923 struct dpif_flow_dump dump;
2924 const struct nlattr *key;
2927 dpif_flow_dump_start(&dump, p->dpif);
2928 while (dpif_flow_dump_next(&dump, &key, &key_len, NULL, NULL, &stats)) {
2929 struct subfacet *subfacet;
2931 subfacet = subfacet_find(p, key, key_len);
2932 if (subfacet && subfacet->installed) {
2933 struct facet *facet = subfacet->facet;
2935 if (stats->n_packets >= subfacet->dp_packet_count) {
2936 uint64_t extra = stats->n_packets - subfacet->dp_packet_count;
2937 facet->packet_count += extra;
2939 VLOG_WARN_RL(&rl, "unexpected packet count from the datapath");
2942 if (stats->n_bytes >= subfacet->dp_byte_count) {
2943 facet->byte_count += stats->n_bytes - subfacet->dp_byte_count;
2945 VLOG_WARN_RL(&rl, "unexpected byte count from datapath");
2948 subfacet->dp_packet_count = stats->n_packets;
2949 subfacet->dp_byte_count = stats->n_bytes;
2951 subfacet_update_time(subfacet, stats->used);
2952 facet_account(facet);
2953 facet_push_stats(facet);
2955 if (!VLOG_DROP_WARN(&rl)) {
2959 odp_flow_key_format(key, key_len, &s);
2960 VLOG_WARN("unexpected flow from datapath %s", ds_cstr(&s));
2964 COVERAGE_INC(facet_unexpected);
2965 /* There's a flow in the datapath that we know nothing about, or a
2966 * flow that shouldn't be installed but was anyway. Delete it. */
2967 dpif_flow_del(p->dpif, key, key_len, NULL);
2970 dpif_flow_dump_done(&dump);
2973 /* Calculates and returns the number of milliseconds of idle time after which
2974 * subfacets should expire from the datapath. When a subfacet expires, we fold
2975 * its statistics into its facet, and when a facet's last subfacet expires, we
2976 * fold its statistic into its rule. */
2978 subfacet_max_idle(const struct ofproto_dpif *ofproto)
2981 * Idle time histogram.
2983 * Most of the time a switch has a relatively small number of subfacets.
2984 * When this is the case we might as well keep statistics for all of them
2985 * in userspace and to cache them in the kernel datapath for performance as
2988 * As the number of subfacets increases, the memory required to maintain
2989 * statistics about them in userspace and in the kernel becomes
2990 * significant. However, with a large number of subfacets it is likely
2991 * that only a few of them are "heavy hitters" that consume a large amount
2992 * of bandwidth. At this point, only heavy hitters are worth caching in
2993 * the kernel and maintaining in userspaces; other subfacets we can
2996 * The technique used to compute the idle time is to build a histogram with
2997 * N_BUCKETS buckets whose width is BUCKET_WIDTH msecs each. Each subfacet
2998 * that is installed in the kernel gets dropped in the appropriate bucket.
2999 * After the histogram has been built, we compute the cutoff so that only
3000 * the most-recently-used 1% of subfacets (but at least
3001 * ofproto->up.flow_eviction_threshold flows) are kept cached. At least
3002 * the most-recently-used bucket of subfacets is kept, so actually an
3003 * arbitrary number of subfacets can be kept in any given expiration run
3004 * (though the next run will delete most of those unless they receive
3007 * This requires a second pass through the subfacets, in addition to the
3008 * pass made by update_stats(), because the former function never looks at
3009 * uninstallable subfacets.
3011 enum { BUCKET_WIDTH = ROUND_UP(100, TIME_UPDATE_INTERVAL) };
3012 enum { N_BUCKETS = 5000 / BUCKET_WIDTH };
3013 int buckets[N_BUCKETS] = { 0 };
3014 int total, subtotal, bucket;
3015 struct subfacet *subfacet;
3019 total = hmap_count(&ofproto->subfacets);
3020 if (total <= ofproto->up.flow_eviction_threshold) {
3021 return N_BUCKETS * BUCKET_WIDTH;
3024 /* Build histogram. */
3026 HMAP_FOR_EACH (subfacet, hmap_node, &ofproto->subfacets) {
3027 long long int idle = now - subfacet->used;
3028 int bucket = (idle <= 0 ? 0
3029 : idle >= BUCKET_WIDTH * N_BUCKETS ? N_BUCKETS - 1
3030 : (unsigned int) idle / BUCKET_WIDTH);
3034 /* Find the first bucket whose flows should be expired. */
3035 subtotal = bucket = 0;
3037 subtotal += buckets[bucket++];
3038 } while (bucket < N_BUCKETS &&
3039 subtotal < MAX(ofproto->up.flow_eviction_threshold, total / 100));
3041 if (VLOG_IS_DBG_ENABLED()) {
3045 ds_put_cstr(&s, "keep");
3046 for (i = 0; i < N_BUCKETS; i++) {
3048 ds_put_cstr(&s, ", drop");
3051 ds_put_format(&s, " %d:%d", i * BUCKET_WIDTH, buckets[i]);
3054 VLOG_INFO("%s: %s (msec:count)", ofproto->up.name, ds_cstr(&s));
3058 return bucket * BUCKET_WIDTH;
3062 expire_subfacets(struct ofproto_dpif *ofproto, int dp_max_idle)
3064 long long int cutoff = time_msec() - dp_max_idle;
3065 struct subfacet *subfacet, *next_subfacet;
3067 HMAP_FOR_EACH_SAFE (subfacet, next_subfacet, hmap_node,
3068 &ofproto->subfacets) {
3069 if (subfacet->used < cutoff) {
3070 subfacet_destroy(subfacet);
3075 /* If 'rule' is an OpenFlow rule, that has expired according to OpenFlow rules,
3076 * then delete it entirely. */
3078 rule_expire(struct rule_dpif *rule)
3080 struct facet *facet, *next_facet;
3084 /* Has 'rule' expired? */
3086 if (rule->up.hard_timeout
3087 && now > rule->up.modified + rule->up.hard_timeout * 1000) {
3088 reason = OFPRR_HARD_TIMEOUT;
3089 } else if (rule->up.idle_timeout && list_is_empty(&rule->facets)
3090 && now > rule->used + rule->up.idle_timeout * 1000) {
3091 reason = OFPRR_IDLE_TIMEOUT;
3096 COVERAGE_INC(ofproto_dpif_expired);
3098 /* Update stats. (This is a no-op if the rule expired due to an idle
3099 * timeout, because that only happens when the rule has no facets left.) */
3100 LIST_FOR_EACH_SAFE (facet, next_facet, list_node, &rule->facets) {
3101 facet_remove(facet);
3104 /* Get rid of the rule. */
3105 ofproto_rule_expire(&rule->up, reason);
3110 /* Creates and returns a new facet owned by 'rule', given a 'flow'.
3112 * The caller must already have determined that no facet with an identical
3113 * 'flow' exists in 'ofproto' and that 'flow' is the best match for 'rule' in
3114 * the ofproto's classifier table.
3116 * The facet will initially have no subfacets. The caller should create (at
3117 * least) one subfacet with subfacet_create(). */
3118 static struct facet *
3119 facet_create(struct rule_dpif *rule, const struct flow *flow)
3121 struct ofproto_dpif *ofproto = ofproto_dpif_cast(rule->up.ofproto);
3122 struct facet *facet;
3124 facet = xzalloc(sizeof *facet);
3125 facet->used = time_msec();
3126 hmap_insert(&ofproto->facets, &facet->hmap_node, flow_hash(flow, 0));
3127 list_push_back(&rule->facets, &facet->list_node);
3129 facet->flow = *flow;
3130 list_init(&facet->subfacets);
3131 netflow_flow_init(&facet->nf_flow);
3132 netflow_flow_update_time(ofproto->netflow, &facet->nf_flow, facet->used);
3138 facet_free(struct facet *facet)
3143 /* Executes, within 'ofproto', the 'n_actions' actions in 'actions' on
3144 * 'packet', which arrived on 'in_port'.
3146 * Takes ownership of 'packet'. */
3148 execute_odp_actions(struct ofproto_dpif *ofproto, const struct flow *flow,
3149 const struct nlattr *odp_actions, size_t actions_len,
3150 struct ofpbuf *packet)
3152 struct odputil_keybuf keybuf;
3156 ofpbuf_use_stack(&key, &keybuf, sizeof keybuf);
3157 odp_flow_key_from_flow(&key, flow);
3159 error = dpif_execute(ofproto->dpif, key.data, key.size,
3160 odp_actions, actions_len, packet);
3162 ofpbuf_delete(packet);
3166 /* Remove 'facet' from 'ofproto' and free up the associated memory:
3168 * - If 'facet' was installed in the datapath, uninstalls it and updates its
3169 * rule's statistics, via subfacet_uninstall().
3171 * - Removes 'facet' from its rule and from ofproto->facets.
3174 facet_remove(struct facet *facet)
3176 struct ofproto_dpif *ofproto = ofproto_dpif_cast(facet->rule->up.ofproto);
3177 struct subfacet *subfacet, *next_subfacet;
3179 assert(!list_is_empty(&facet->subfacets));
3181 /* First uninstall all of the subfacets to get final statistics. */
3182 LIST_FOR_EACH (subfacet, list_node, &facet->subfacets) {
3183 subfacet_uninstall(subfacet);
3186 /* Flush the final stats to the rule.
3188 * This might require us to have at least one subfacet around so that we
3189 * can use its actions for accounting in facet_account(), which is why we
3190 * have uninstalled but not yet destroyed the subfacets. */
3191 facet_flush_stats(facet);
3193 /* Now we're really all done so destroy everything. */
3194 LIST_FOR_EACH_SAFE (subfacet, next_subfacet, list_node,
3195 &facet->subfacets) {
3196 subfacet_destroy__(subfacet);
3198 hmap_remove(&ofproto->facets, &facet->hmap_node);
3199 list_remove(&facet->list_node);
3204 facet_account(struct facet *facet)
3206 struct ofproto_dpif *ofproto = ofproto_dpif_cast(facet->rule->up.ofproto);
3208 struct subfacet *subfacet;
3209 const struct nlattr *a;
3213 if (facet->byte_count <= facet->accounted_bytes) {
3216 n_bytes = facet->byte_count - facet->accounted_bytes;
3217 facet->accounted_bytes = facet->byte_count;
3219 /* Feed information from the active flows back into the learning table to
3220 * ensure that table is always in sync with what is actually flowing
3221 * through the datapath. */
3222 if (facet->has_learn || facet->has_normal) {
3223 struct action_xlate_ctx ctx;
3225 action_xlate_ctx_init(&ctx, ofproto, &facet->flow,
3226 facet->flow.vlan_tci,
3227 facet->rule->up.flow_cookie, NULL);
3228 ctx.may_learn = true;
3229 ofpbuf_delete(xlate_actions(&ctx, facet->rule->up.actions,
3230 facet->rule->up.n_actions));
3233 if (!facet->has_normal || !ofproto->has_bonded_bundles) {
3237 /* This loop feeds byte counters to bond_account() for rebalancing to use
3238 * as a basis. We also need to track the actual VLAN on which the packet
3239 * is going to be sent to ensure that it matches the one passed to
3240 * bond_choose_output_slave(). (Otherwise, we will account to the wrong
3243 * We use the actions from an arbitrary subfacet because they should all
3244 * be equally valid for our purpose. */
3245 subfacet = CONTAINER_OF(list_front(&facet->subfacets),
3246 struct subfacet, list_node);
3247 vlan_tci = facet->flow.vlan_tci;
3248 NL_ATTR_FOR_EACH_UNSAFE (a, left,
3249 subfacet->actions, subfacet->actions_len) {
3250 const struct ovs_action_push_vlan *vlan;
3251 struct ofport_dpif *port;
3253 switch (nl_attr_type(a)) {
3254 case OVS_ACTION_ATTR_OUTPUT:
3255 port = get_odp_port(ofproto, nl_attr_get_u32(a));
3256 if (port && port->bundle && port->bundle->bond) {
3257 bond_account(port->bundle->bond, &facet->flow,
3258 vlan_tci_to_vid(vlan_tci), n_bytes);
3262 case OVS_ACTION_ATTR_POP_VLAN:
3263 vlan_tci = htons(0);
3266 case OVS_ACTION_ATTR_PUSH_VLAN:
3267 vlan = nl_attr_get(a);
3268 vlan_tci = vlan->vlan_tci;
3274 /* Returns true if the only action for 'facet' is to send to the controller.
3275 * (We don't report NetFlow expiration messages for such facets because they
3276 * are just part of the control logic for the network, not real traffic). */
3278 facet_is_controller_flow(struct facet *facet)
3281 && facet->rule->up.n_actions == 1
3282 && action_outputs_to_port(&facet->rule->up.actions[0],
3283 htons(OFPP_CONTROLLER)));
3286 /* Folds all of 'facet''s statistics into its rule. Also updates the
3287 * accounting ofhook and emits a NetFlow expiration if appropriate. All of
3288 * 'facet''s statistics in the datapath should have been zeroed and folded into
3289 * its packet and byte counts before this function is called. */
3291 facet_flush_stats(struct facet *facet)
3293 struct ofproto_dpif *ofproto = ofproto_dpif_cast(facet->rule->up.ofproto);
3294 struct subfacet *subfacet;
3296 LIST_FOR_EACH (subfacet, list_node, &facet->subfacets) {
3297 assert(!subfacet->dp_byte_count);
3298 assert(!subfacet->dp_packet_count);
3301 facet_push_stats(facet);
3302 facet_account(facet);
3304 if (ofproto->netflow && !facet_is_controller_flow(facet)) {
3305 struct ofexpired expired;
3306 expired.flow = facet->flow;
3307 expired.packet_count = facet->packet_count;
3308 expired.byte_count = facet->byte_count;
3309 expired.used = facet->used;
3310 netflow_expire(ofproto->netflow, &facet->nf_flow, &expired);
3313 facet->rule->packet_count += facet->packet_count;
3314 facet->rule->byte_count += facet->byte_count;
3316 /* Reset counters to prevent double counting if 'facet' ever gets
3318 facet_reset_counters(facet);
3320 netflow_flow_clear(&facet->nf_flow);
3323 /* Searches 'ofproto''s table of facets for one exactly equal to 'flow'.
3324 * Returns it if found, otherwise a null pointer.
3326 * The returned facet might need revalidation; use facet_lookup_valid()
3327 * instead if that is important. */
3328 static struct facet *
3329 facet_find(struct ofproto_dpif *ofproto, const struct flow *flow)
3331 struct facet *facet;
3333 HMAP_FOR_EACH_WITH_HASH (facet, hmap_node, flow_hash(flow, 0),
3335 if (flow_equal(flow, &facet->flow)) {
3343 /* Searches 'ofproto''s table of facets for one exactly equal to 'flow'.
3344 * Returns it if found, otherwise a null pointer.
3346 * The returned facet is guaranteed to be valid. */
3347 static struct facet *
3348 facet_lookup_valid(struct ofproto_dpif *ofproto, const struct flow *flow)
3350 struct facet *facet = facet_find(ofproto, flow);
3352 /* The facet we found might not be valid, since we could be in need of
3353 * revalidation. If it is not valid, don't return it. */
3355 && (ofproto->need_revalidate
3356 || tag_set_intersects(&ofproto->revalidate_set, facet->tags))
3357 && !facet_revalidate(facet)) {
3358 COVERAGE_INC(facet_invalidated);
3366 facet_check_consistency(struct facet *facet)
3368 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 15);
3370 struct ofproto_dpif *ofproto = ofproto_dpif_cast(facet->rule->up.ofproto);
3372 struct rule_dpif *rule;
3373 struct subfacet *subfacet;
3374 bool may_log = false;
3377 /* Check the rule for consistency. */
3378 rule = rule_dpif_lookup(ofproto, &facet->flow, 0);
3380 if (!VLOG_DROP_WARN(&rl)) {
3381 char *s = flow_to_string(&facet->flow);
3382 VLOG_WARN("%s: facet should not exist", s);
3386 } else if (rule != facet->rule) {
3387 may_log = !VLOG_DROP_WARN(&rl);
3393 flow_format(&s, &facet->flow);
3394 ds_put_format(&s, ": facet associated with wrong rule (was "
3395 "table=%"PRIu8",", facet->rule->up.table_id);
3396 cls_rule_format(&facet->rule->up.cr, &s);
3397 ds_put_format(&s, ") (should have been table=%"PRIu8",",
3399 cls_rule_format(&rule->up.cr, &s);
3400 ds_put_char(&s, ')');
3402 VLOG_WARN("%s", ds_cstr(&s));
3409 /* Check the datapath actions for consistency. */
3410 LIST_FOR_EACH (subfacet, list_node, &facet->subfacets) {
3411 struct action_xlate_ctx ctx;
3412 struct ofpbuf *odp_actions;
3413 bool actions_changed;
3414 bool should_install;
3416 action_xlate_ctx_init(&ctx, ofproto, &facet->flow,
3417 subfacet->initial_tci, rule->up.flow_cookie,
3419 odp_actions = xlate_actions(&ctx, rule->up.actions,
3420 rule->up.n_actions);
3422 should_install = (ctx.may_set_up_flow
3423 && subfacet->key_fitness != ODP_FIT_TOO_LITTLE);
3424 if (!should_install && !subfacet->installed) {
3425 /* The actions for uninstallable flows may vary from one packet to
3426 * the next, so don't compare the actions. */
3430 actions_changed = (subfacet->actions_len != odp_actions->size
3431 || memcmp(subfacet->actions, odp_actions->data,
3432 subfacet->actions_len));
3433 if (should_install != subfacet->installed || actions_changed) {
3435 may_log = !VLOG_DROP_WARN(&rl);
3440 struct odputil_keybuf keybuf;
3445 subfacet_get_key(subfacet, &keybuf, &key);
3446 odp_flow_key_format(key.data, key.size, &s);
3448 ds_put_cstr(&s, ": inconsistency in subfacet");
3449 if (should_install != subfacet->installed) {
3450 enum odp_key_fitness fitness = subfacet->key_fitness;
3452 ds_put_format(&s, " (should%s have been installed)",
3453 should_install ? "" : " not");
3454 ds_put_format(&s, " (may_set_up_flow=%s, fitness=%s)",
3455 ctx.may_set_up_flow ? "true" : "false",
3456 odp_key_fitness_to_string(fitness));
3458 if (actions_changed) {
3459 ds_put_cstr(&s, " (actions were: ");
3460 format_odp_actions(&s, subfacet->actions,
3461 subfacet->actions_len);
3462 ds_put_cstr(&s, ") (correct actions: ");
3463 format_odp_actions(&s, odp_actions->data,
3465 ds_put_char(&s, ')');
3467 ds_put_cstr(&s, " (actions: ");
3468 format_odp_actions(&s, subfacet->actions,
3469 subfacet->actions_len);
3470 ds_put_char(&s, ')');
3472 VLOG_WARN("%s", ds_cstr(&s));
3478 ofpbuf_delete(odp_actions);
3484 /* Re-searches the classifier for 'facet':
3486 * - If the rule found is different from 'facet''s current rule, moves
3487 * 'facet' to the new rule and recompiles its actions.
3489 * - If the rule found is the same as 'facet''s current rule, leaves 'facet'
3490 * where it is and recompiles its actions anyway.
3492 * - If there is none, destroys 'facet'.
3494 * Returns true if 'facet' still exists, false if it has been destroyed. */
3496 facet_revalidate(struct facet *facet)
3498 struct ofproto_dpif *ofproto = ofproto_dpif_cast(facet->rule->up.ofproto);
3500 struct nlattr *odp_actions;
3503 struct actions *new_actions;
3505 struct action_xlate_ctx ctx;
3506 struct rule_dpif *new_rule;
3507 struct subfacet *subfacet;
3508 bool actions_changed;
3511 COVERAGE_INC(facet_revalidate);
3513 /* Determine the new rule. */
3514 new_rule = rule_dpif_lookup(ofproto, &facet->flow, 0);
3516 /* No new rule, so delete the facet. */
3517 facet_remove(facet);
3521 /* Calculate new datapath actions.
3523 * We do not modify any 'facet' state yet, because we might need to, e.g.,
3524 * emit a NetFlow expiration and, if so, we need to have the old state
3525 * around to properly compose it. */
3527 /* If the datapath actions changed or the installability changed,
3528 * then we need to talk to the datapath. */
3531 memset(&ctx, 0, sizeof ctx);
3532 LIST_FOR_EACH (subfacet, list_node, &facet->subfacets) {
3533 struct ofpbuf *odp_actions;
3534 bool should_install;
3536 action_xlate_ctx_init(&ctx, ofproto, &facet->flow,
3537 subfacet->initial_tci, new_rule->up.flow_cookie,
3539 odp_actions = xlate_actions(&ctx, new_rule->up.actions,
3540 new_rule->up.n_actions);
3541 actions_changed = (subfacet->actions_len != odp_actions->size
3542 || memcmp(subfacet->actions, odp_actions->data,
3543 subfacet->actions_len));
3545 should_install = (ctx.may_set_up_flow
3546 && subfacet->key_fitness != ODP_FIT_TOO_LITTLE);
3547 if (actions_changed || should_install != subfacet->installed) {
3548 if (should_install) {
3549 struct dpif_flow_stats stats;
3551 subfacet_install(subfacet,
3552 odp_actions->data, odp_actions->size, &stats);
3553 subfacet_update_stats(subfacet, &stats);
3555 subfacet_uninstall(subfacet);
3559 new_actions = xcalloc(list_size(&facet->subfacets),
3560 sizeof *new_actions);
3562 new_actions[i].odp_actions = xmemdup(odp_actions->data,
3564 new_actions[i].actions_len = odp_actions->size;
3567 ofpbuf_delete(odp_actions);
3571 facet_flush_stats(facet);
3574 /* Update 'facet' now that we've taken care of all the old state. */
3575 facet->tags = ctx.tags;
3576 facet->nf_flow.output_iface = ctx.nf_output_iface;
3577 facet->may_install = ctx.may_set_up_flow;
3578 facet->has_learn = ctx.has_learn;
3579 facet->has_normal = ctx.has_normal;
3580 facet->mirrors = ctx.mirrors;
3583 LIST_FOR_EACH (subfacet, list_node, &facet->subfacets) {
3584 if (new_actions[i].odp_actions) {
3585 free(subfacet->actions);
3586 subfacet->actions = new_actions[i].odp_actions;
3587 subfacet->actions_len = new_actions[i].actions_len;
3593 if (facet->rule != new_rule) {
3594 COVERAGE_INC(facet_changed_rule);
3595 list_remove(&facet->list_node);
3596 list_push_back(&new_rule->facets, &facet->list_node);
3597 facet->rule = new_rule;
3598 facet->used = new_rule->up.created;
3599 facet->prev_used = facet->used;
3605 /* Updates 'facet''s used time. Caller is responsible for calling
3606 * facet_push_stats() to update the flows which 'facet' resubmits into. */
3608 facet_update_time(struct facet *facet, long long int used)
3610 struct ofproto_dpif *ofproto = ofproto_dpif_cast(facet->rule->up.ofproto);
3611 if (used > facet->used) {
3613 if (used > facet->rule->used) {
3614 facet->rule->used = used;
3616 netflow_flow_update_time(ofproto->netflow, &facet->nf_flow, used);
3621 facet_reset_counters(struct facet *facet)
3623 facet->packet_count = 0;
3624 facet->byte_count = 0;
3625 facet->prev_packet_count = 0;
3626 facet->prev_byte_count = 0;
3627 facet->accounted_bytes = 0;
3631 facet_push_stats(struct facet *facet)
3633 uint64_t new_packets, new_bytes;
3635 assert(facet->packet_count >= facet->prev_packet_count);
3636 assert(facet->byte_count >= facet->prev_byte_count);
3637 assert(facet->used >= facet->prev_used);
3639 new_packets = facet->packet_count - facet->prev_packet_count;
3640 new_bytes = facet->byte_count - facet->prev_byte_count;
3642 if (new_packets || new_bytes || facet->used > facet->prev_used) {
3643 facet->prev_packet_count = facet->packet_count;
3644 facet->prev_byte_count = facet->byte_count;
3645 facet->prev_used = facet->used;
3647 flow_push_stats(facet->rule, &facet->flow,
3648 new_packets, new_bytes, facet->used);
3650 update_mirror_stats(ofproto_dpif_cast(facet->rule->up.ofproto),
3651 facet->mirrors, new_packets, new_bytes);
3655 struct ofproto_push {
3656 struct action_xlate_ctx ctx;
3663 push_resubmit(struct action_xlate_ctx *ctx, struct rule_dpif *rule)
3665 struct ofproto_push *push = CONTAINER_OF(ctx, struct ofproto_push, ctx);
3668 rule->packet_count += push->packets;
3669 rule->byte_count += push->bytes;
3670 rule->used = MAX(push->used, rule->used);
3674 /* Pushes flow statistics to the rules which 'flow' resubmits into given
3675 * 'rule''s actions and mirrors. */
3677 flow_push_stats(const struct rule_dpif *rule,
3678 const struct flow *flow, uint64_t packets, uint64_t bytes,
3681 struct ofproto_dpif *ofproto = ofproto_dpif_cast(rule->up.ofproto);
3682 struct ofproto_push push;
3684 push.packets = packets;
3688 action_xlate_ctx_init(&push.ctx, ofproto, flow, flow->vlan_tci,
3689 rule->up.flow_cookie, NULL);
3690 push.ctx.resubmit_hook = push_resubmit;
3691 ofpbuf_delete(xlate_actions(&push.ctx,
3692 rule->up.actions, rule->up.n_actions));
3697 static struct subfacet *
3698 subfacet_find__(struct ofproto_dpif *ofproto,
3699 const struct nlattr *key, size_t key_len, uint32_t key_hash,
3700 const struct flow *flow)
3702 struct subfacet *subfacet;
3704 HMAP_FOR_EACH_WITH_HASH (subfacet, hmap_node, key_hash,
3705 &ofproto->subfacets) {
3707 ? (subfacet->key_len == key_len
3708 && !memcmp(key, subfacet->key, key_len))
3709 : flow_equal(flow, &subfacet->facet->flow)) {
3717 /* Searches 'facet' (within 'ofproto') for a subfacet with the specified
3718 * 'key_fitness', 'key', and 'key_len'. Returns the existing subfacet if
3719 * there is one, otherwise creates and returns a new subfacet.
3721 * If the returned subfacet is new, then subfacet->actions will be NULL, in
3722 * which case the caller must populate the actions with
3723 * subfacet_make_actions(). */
3724 static struct subfacet *
3725 subfacet_create(struct facet *facet, enum odp_key_fitness key_fitness,
3726 const struct nlattr *key, size_t key_len, ovs_be16 initial_tci)
3728 struct ofproto_dpif *ofproto = ofproto_dpif_cast(facet->rule->up.ofproto);
3729 uint32_t key_hash = odp_flow_key_hash(key, key_len);
3730 struct subfacet *subfacet;
3732 subfacet = subfacet_find__(ofproto, key, key_len, key_hash, &facet->flow);
3734 if (subfacet->facet == facet) {
3738 /* This shouldn't happen. */
3739 VLOG_ERR_RL(&rl, "subfacet with wrong facet");
3740 subfacet_destroy(subfacet);
3743 subfacet = xzalloc(sizeof *subfacet);
3744 hmap_insert(&ofproto->subfacets, &subfacet->hmap_node, key_hash);
3745 list_push_back(&facet->subfacets, &subfacet->list_node);
3746 subfacet->facet = facet;
3747 subfacet->used = time_msec();
3748 subfacet->key_fitness = key_fitness;
3749 if (key_fitness != ODP_FIT_PERFECT) {
3750 subfacet->key = xmemdup(key, key_len);
3751 subfacet->key_len = key_len;
3753 subfacet->installed = false;
3754 subfacet->initial_tci = initial_tci;
3759 /* Searches 'ofproto' for a subfacet with the given 'key', 'key_len', and
3760 * 'flow'. Returns the subfacet if one exists, otherwise NULL. */
3761 static struct subfacet *
3762 subfacet_find(struct ofproto_dpif *ofproto,
3763 const struct nlattr *key, size_t key_len)
3765 uint32_t key_hash = odp_flow_key_hash(key, key_len);
3766 enum odp_key_fitness fitness;
3769 fitness = odp_flow_key_to_flow(key, key_len, &flow);
3770 if (fitness == ODP_FIT_ERROR) {
3774 return subfacet_find__(ofproto, key, key_len, key_hash, &flow);
3777 /* Uninstalls 'subfacet' from the datapath, if it is installed, removes it from
3778 * its facet within 'ofproto', and frees it. */
3780 subfacet_destroy__(struct subfacet *subfacet)
3782 struct facet *facet = subfacet->facet;
3783 struct ofproto_dpif *ofproto = ofproto_dpif_cast(facet->rule->up.ofproto);
3785 subfacet_uninstall(subfacet);
3786 hmap_remove(&ofproto->subfacets, &subfacet->hmap_node);
3787 list_remove(&subfacet->list_node);
3788 free(subfacet->key);
3789 free(subfacet->actions);
3793 /* Destroys 'subfacet', as with subfacet_destroy__(), and then if this was the
3794 * last remaining subfacet in its facet destroys the facet too. */
3796 subfacet_destroy(struct subfacet *subfacet)
3798 struct facet *facet = subfacet->facet;
3800 if (list_is_singleton(&facet->subfacets)) {
3801 /* facet_remove() needs at least one subfacet (it will remove it). */
3802 facet_remove(facet);
3804 subfacet_destroy__(subfacet);
3808 /* Initializes 'key' with the sequence of OVS_KEY_ATTR_* Netlink attributes
3809 * that can be used to refer to 'subfacet'. The caller must provide 'keybuf'
3810 * for use as temporary storage. */
3812 subfacet_get_key(struct subfacet *subfacet, struct odputil_keybuf *keybuf,
3815 if (!subfacet->key) {
3816 ofpbuf_use_stack(key, keybuf, sizeof *keybuf);
3817 odp_flow_key_from_flow(key, &subfacet->facet->flow);
3819 ofpbuf_use_const(key, subfacet->key, subfacet->key_len);
3823 /* Composes the datapath actions for 'subfacet' based on its rule's actions. */
3825 subfacet_make_actions(struct subfacet *subfacet, const struct ofpbuf *packet)
3827 struct facet *facet = subfacet->facet;
3828 const struct rule_dpif *rule = facet->rule;
3829 struct ofproto_dpif *ofproto = ofproto_dpif_cast(rule->up.ofproto);
3830 struct ofpbuf *odp_actions;
3831 struct action_xlate_ctx ctx;
3833 action_xlate_ctx_init(&ctx, ofproto, &facet->flow, subfacet->initial_tci,
3834 rule->up.flow_cookie, packet);
3835 odp_actions = xlate_actions(&ctx, rule->up.actions, rule->up.n_actions);
3836 facet->tags = ctx.tags;
3837 facet->may_install = ctx.may_set_up_flow;
3838 facet->has_learn = ctx.has_learn;
3839 facet->has_normal = ctx.has_normal;
3840 facet->nf_flow.output_iface = ctx.nf_output_iface;
3841 facet->mirrors = ctx.mirrors;
3843 if (subfacet->actions_len != odp_actions->size
3844 || memcmp(subfacet->actions, odp_actions->data, odp_actions->size)) {
3845 free(subfacet->actions);
3846 subfacet->actions_len = odp_actions->size;
3847 subfacet->actions = xmemdup(odp_actions->data, odp_actions->size);
3850 ofpbuf_delete(odp_actions);
3853 /* Updates 'subfacet''s datapath flow, setting its actions to 'actions_len'
3854 * bytes of actions in 'actions'. If 'stats' is non-null, statistics counters
3855 * in the datapath will be zeroed and 'stats' will be updated with traffic new
3856 * since 'subfacet' was last updated.
3858 * Returns 0 if successful, otherwise a positive errno value. */
3860 subfacet_install(struct subfacet *subfacet,
3861 const struct nlattr *actions, size_t actions_len,
3862 struct dpif_flow_stats *stats)
3864 struct facet *facet = subfacet->facet;
3865 struct ofproto_dpif *ofproto = ofproto_dpif_cast(facet->rule->up.ofproto);
3866 struct odputil_keybuf keybuf;
3867 enum dpif_flow_put_flags flags;
3871 flags = DPIF_FP_CREATE | DPIF_FP_MODIFY;
3873 flags |= DPIF_FP_ZERO_STATS;
3876 subfacet_get_key(subfacet, &keybuf, &key);
3877 ret = dpif_flow_put(ofproto->dpif, flags, key.data, key.size,
3878 actions, actions_len, stats);
3881 subfacet_reset_dp_stats(subfacet, stats);
3887 /* If 'subfacet' is installed in the datapath, uninstalls it. */
3889 subfacet_uninstall(struct subfacet *subfacet)
3891 if (subfacet->installed) {
3892 struct rule_dpif *rule = subfacet->facet->rule;
3893 struct ofproto_dpif *ofproto = ofproto_dpif_cast(rule->up.ofproto);
3894 struct odputil_keybuf keybuf;
3895 struct dpif_flow_stats stats;
3899 subfacet_get_key(subfacet, &keybuf, &key);
3900 error = dpif_flow_del(ofproto->dpif, key.data, key.size, &stats);
3901 subfacet_reset_dp_stats(subfacet, &stats);
3903 subfacet_update_stats(subfacet, &stats);
3905 subfacet->installed = false;
3907 assert(subfacet->dp_packet_count == 0);
3908 assert(subfacet->dp_byte_count == 0);
3912 /* Resets 'subfacet''s datapath statistics counters. This should be called
3913 * when 'subfacet''s statistics are cleared in the datapath. If 'stats' is
3914 * non-null, it should contain the statistics returned by dpif when 'subfacet'
3915 * was reset in the datapath. 'stats' will be modified to include only
3916 * statistics new since 'subfacet' was last updated. */
3918 subfacet_reset_dp_stats(struct subfacet *subfacet,
3919 struct dpif_flow_stats *stats)
3922 && subfacet->dp_packet_count <= stats->n_packets
3923 && subfacet->dp_byte_count <= stats->n_bytes) {
3924 stats->n_packets -= subfacet->dp_packet_count;
3925 stats->n_bytes -= subfacet->dp_byte_count;
3928 subfacet->dp_packet_count = 0;
3929 subfacet->dp_byte_count = 0;
3932 /* Updates 'subfacet''s used time. The caller is responsible for calling
3933 * facet_push_stats() to update the flows which 'subfacet' resubmits into. */
3935 subfacet_update_time(struct subfacet *subfacet, long long int used)
3937 if (used > subfacet->used) {
3938 subfacet->used = used;
3939 facet_update_time(subfacet->facet, used);
3943 /* Folds the statistics from 'stats' into the counters in 'subfacet'.
3945 * Because of the meaning of a subfacet's counters, it only makes sense to do
3946 * this if 'stats' are not tracked in the datapath, that is, if 'stats'
3947 * represents a packet that was sent by hand or if it represents statistics
3948 * that have been cleared out of the datapath. */
3950 subfacet_update_stats(struct subfacet *subfacet,
3951 const struct dpif_flow_stats *stats)
3953 if (stats->n_packets || stats->used > subfacet->used) {
3954 struct facet *facet = subfacet->facet;
3956 subfacet_update_time(subfacet, stats->used);
3957 facet->packet_count += stats->n_packets;
3958 facet->byte_count += stats->n_bytes;
3959 facet_push_stats(facet);
3960 netflow_flow_update_flags(&facet->nf_flow, stats->tcp_flags);
3966 static struct rule_dpif *
3967 rule_dpif_lookup(struct ofproto_dpif *ofproto, const struct flow *flow,
3970 struct cls_rule *cls_rule;
3971 struct classifier *cls;
3973 if (table_id >= N_TABLES) {
3977 cls = &ofproto->up.tables[table_id].cls;
3978 if (flow->nw_frag & FLOW_NW_FRAG_ANY
3979 && ofproto->up.frag_handling == OFPC_FRAG_NORMAL) {
3980 /* For OFPC_NORMAL frag_handling, we must pretend that transport ports
3981 * are unavailable. */
3982 struct flow ofpc_normal_flow = *flow;
3983 ofpc_normal_flow.tp_src = htons(0);
3984 ofpc_normal_flow.tp_dst = htons(0);
3985 cls_rule = classifier_lookup(cls, &ofpc_normal_flow);
3987 cls_rule = classifier_lookup(cls, flow);
3989 return rule_dpif_cast(rule_from_cls_rule(cls_rule));
3993 complete_operation(struct rule_dpif *rule)
3995 struct ofproto_dpif *ofproto = ofproto_dpif_cast(rule->up.ofproto);
3997 rule_invalidate(rule);
3999 struct dpif_completion *c = xmalloc(sizeof *c);
4000 c->op = rule->up.pending;
4001 list_push_back(&ofproto->completions, &c->list_node);
4003 ofoperation_complete(rule->up.pending, 0);
4007 static struct rule *
4010 struct rule_dpif *rule = xmalloc(sizeof *rule);
4015 rule_dealloc(struct rule *rule_)
4017 struct rule_dpif *rule = rule_dpif_cast(rule_);
4022 rule_construct(struct rule *rule_)
4024 struct rule_dpif *rule = rule_dpif_cast(rule_);
4025 struct ofproto_dpif *ofproto = ofproto_dpif_cast(rule->up.ofproto);
4026 struct rule_dpif *victim;
4030 error = validate_actions(rule->up.actions, rule->up.n_actions,
4031 &rule->up.cr.flow, ofproto->max_ports);
4036 rule->used = rule->up.created;
4037 rule->packet_count = 0;
4038 rule->byte_count = 0;
4040 victim = rule_dpif_cast(ofoperation_get_victim(rule->up.pending));
4041 if (victim && !list_is_empty(&victim->facets)) {
4042 struct facet *facet;
4044 rule->facets = victim->facets;
4045 list_moved(&rule->facets);
4046 LIST_FOR_EACH (facet, list_node, &rule->facets) {
4047 /* XXX: We're only clearing our local counters here. It's possible
4048 * that quite a few packets are unaccounted for in the datapath
4049 * statistics. These will be accounted to the new rule instead of
4050 * cleared as required. This could be fixed by clearing out the
4051 * datapath statistics for this facet, but currently it doesn't
4053 facet_reset_counters(facet);
4057 /* Must avoid list_moved() in this case. */
4058 list_init(&rule->facets);
4061 table_id = rule->up.table_id;
4062 rule->tag = (victim ? victim->tag
4064 : rule_calculate_tag(&rule->up.cr.flow, &rule->up.cr.wc,
4065 ofproto->tables[table_id].basis));
4067 complete_operation(rule);
4072 rule_destruct(struct rule *rule_)
4074 struct rule_dpif *rule = rule_dpif_cast(rule_);
4075 struct facet *facet, *next_facet;
4077 LIST_FOR_EACH_SAFE (facet, next_facet, list_node, &rule->facets) {
4078 facet_revalidate(facet);
4081 complete_operation(rule);
4085 rule_get_stats(struct rule *rule_, uint64_t *packets, uint64_t *bytes)
4087 struct rule_dpif *rule = rule_dpif_cast(rule_);
4088 struct facet *facet;
4090 /* Start from historical data for 'rule' itself that are no longer tracked
4091 * in facets. This counts, for example, facets that have expired. */
4092 *packets = rule->packet_count;
4093 *bytes = rule->byte_count;
4095 /* Add any statistics that are tracked by facets. This includes
4096 * statistical data recently updated by ofproto_update_stats() as well as
4097 * stats for packets that were executed "by hand" via dpif_execute(). */
4098 LIST_FOR_EACH (facet, list_node, &rule->facets) {
4099 *packets += facet->packet_count;
4100 *bytes += facet->byte_count;
4105 rule_execute(struct rule *rule_, const struct flow *flow,
4106 struct ofpbuf *packet)
4108 struct rule_dpif *rule = rule_dpif_cast(rule_);
4109 struct ofproto_dpif *ofproto = ofproto_dpif_cast(rule->up.ofproto);
4110 struct action_xlate_ctx ctx;
4111 struct ofpbuf *odp_actions;
4114 action_xlate_ctx_init(&ctx, ofproto, flow, flow->vlan_tci,
4115 rule->up.flow_cookie, packet);
4116 odp_actions = xlate_actions(&ctx, rule->up.actions, rule->up.n_actions);
4117 size = packet->size;
4118 if (execute_odp_actions(ofproto, flow, odp_actions->data,
4119 odp_actions->size, packet)) {
4120 rule->used = time_msec();
4121 rule->packet_count++;
4122 rule->byte_count += size;
4123 flow_push_stats(rule, flow, 1, size, rule->used);
4125 ofpbuf_delete(odp_actions);
4131 rule_modify_actions(struct rule *rule_)
4133 struct rule_dpif *rule = rule_dpif_cast(rule_);
4134 struct ofproto_dpif *ofproto = ofproto_dpif_cast(rule->up.ofproto);
4137 error = validate_actions(rule->up.actions, rule->up.n_actions,
4138 &rule->up.cr.flow, ofproto->max_ports);
4140 ofoperation_complete(rule->up.pending, error);
4144 complete_operation(rule);
4147 /* Sends 'packet' out 'ofport'.
4148 * May modify 'packet'.
4149 * Returns 0 if successful, otherwise a positive errno value. */
4151 send_packet(const struct ofport_dpif *ofport, struct ofpbuf *packet)
4153 const struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport->up.ofproto);
4154 struct ofpbuf key, odp_actions;
4155 struct odputil_keybuf keybuf;
4160 flow_extract((struct ofpbuf *) packet, 0, 0, 0, &flow);
4161 odp_port = vsp_realdev_to_vlandev(ofproto, ofport->odp_port,
4163 if (odp_port != ofport->odp_port) {
4164 eth_pop_vlan(packet);
4165 flow.vlan_tci = htons(0);
4168 ofpbuf_use_stack(&key, &keybuf, sizeof keybuf);
4169 odp_flow_key_from_flow(&key, &flow);
4171 ofpbuf_init(&odp_actions, 32);
4172 compose_sflow_action(ofproto, &odp_actions, &flow, odp_port);
4174 nl_msg_put_u32(&odp_actions, OVS_ACTION_ATTR_OUTPUT, odp_port);
4175 error = dpif_execute(ofproto->dpif,
4177 odp_actions.data, odp_actions.size,
4179 ofpbuf_uninit(&odp_actions);
4182 VLOG_WARN_RL(&rl, "%s: failed to send packet on port %"PRIu32" (%s)",
4183 ofproto->up.name, odp_port, strerror(error));
4185 ofproto_update_local_port_stats(ofport->up.ofproto, packet->size, 0);
4189 /* OpenFlow to datapath action translation. */
4191 static void do_xlate_actions(const union ofp_action *in, size_t n_in,
4192 struct action_xlate_ctx *ctx);
4193 static void xlate_normal(struct action_xlate_ctx *);
4196 put_userspace_action(const struct ofproto_dpif *ofproto,
4197 struct ofpbuf *odp_actions,
4198 const struct flow *flow,
4199 const struct user_action_cookie *cookie)
4203 pid = dpif_port_get_pid(ofproto->dpif,
4204 ofp_port_to_odp_port(flow->in_port));
4206 return odp_put_userspace_action(pid, cookie, odp_actions);
4209 /* Compose SAMPLE action for sFlow. */
4211 compose_sflow_action(const struct ofproto_dpif *ofproto,
4212 struct ofpbuf *odp_actions,
4213 const struct flow *flow,
4216 uint32_t port_ifindex;
4217 uint32_t probability;
4218 struct user_action_cookie cookie;
4219 size_t sample_offset, actions_offset;
4220 int cookie_offset, n_output;
4222 if (!ofproto->sflow || flow->in_port == OFPP_NONE) {
4226 if (odp_port == OVSP_NONE) {
4230 port_ifindex = dpif_sflow_odp_port_to_ifindex(ofproto->sflow, odp_port);
4234 sample_offset = nl_msg_start_nested(odp_actions, OVS_ACTION_ATTR_SAMPLE);
4236 /* Number of packets out of UINT_MAX to sample. */
4237 probability = dpif_sflow_get_probability(ofproto->sflow);
4238 nl_msg_put_u32(odp_actions, OVS_SAMPLE_ATTR_PROBABILITY, probability);
4240 actions_offset = nl_msg_start_nested(odp_actions, OVS_SAMPLE_ATTR_ACTIONS);
4242 cookie.type = USER_ACTION_COOKIE_SFLOW;
4243 cookie.data = port_ifindex;
4244 cookie.n_output = n_output;
4245 cookie.vlan_tci = 0;
4246 cookie_offset = put_userspace_action(ofproto, odp_actions, flow, &cookie);
4248 nl_msg_end_nested(odp_actions, actions_offset);
4249 nl_msg_end_nested(odp_actions, sample_offset);
4250 return cookie_offset;
4253 /* SAMPLE action must be first action in any given list of actions.
4254 * At this point we do not have all information required to build it. So try to
4255 * build sample action as complete as possible. */
4257 add_sflow_action(struct action_xlate_ctx *ctx)
4259 ctx->user_cookie_offset = compose_sflow_action(ctx->ofproto,
4261 &ctx->flow, OVSP_NONE);
4262 ctx->sflow_odp_port = 0;
4263 ctx->sflow_n_outputs = 0;
4266 /* Fix SAMPLE action according to data collected while composing ODP actions.
4267 * We need to fix SAMPLE actions OVS_SAMPLE_ATTR_ACTIONS attribute, i.e. nested
4268 * USERSPACE action's user-cookie which is required for sflow. */
4270 fix_sflow_action(struct action_xlate_ctx *ctx)
4272 const struct flow *base = &ctx->base_flow;
4273 struct user_action_cookie *cookie;
4275 if (!ctx->user_cookie_offset) {
4279 cookie = ofpbuf_at(ctx->odp_actions, ctx->user_cookie_offset,
4281 assert(cookie != NULL);
4282 assert(cookie->type == USER_ACTION_COOKIE_SFLOW);
4284 if (ctx->sflow_n_outputs) {
4285 cookie->data = dpif_sflow_odp_port_to_ifindex(ctx->ofproto->sflow,
4286 ctx->sflow_odp_port);
4288 if (ctx->sflow_n_outputs >= 255) {
4289 cookie->n_output = 255;
4291 cookie->n_output = ctx->sflow_n_outputs;
4293 cookie->vlan_tci = base->vlan_tci;
4297 compose_output_action__(struct action_xlate_ctx *ctx, uint16_t ofp_port,
4300 const struct ofport_dpif *ofport = get_ofp_port(ctx->ofproto, ofp_port);
4301 uint16_t odp_port = ofp_port_to_odp_port(ofp_port);
4302 ovs_be16 flow_vlan_tci = ctx->flow.vlan_tci;
4303 uint8_t flow_nw_tos = ctx->flow.nw_tos;
4307 struct priority_to_dscp *pdscp;
4309 if (ofport->up.opp.config & htonl(OFPPC_NO_FWD)
4310 || (check_stp && !stp_forward_in_state(ofport->stp_state))) {
4314 pdscp = get_priority(ofport, ctx->flow.skb_priority);
4316 ctx->flow.nw_tos &= ~IP_DSCP_MASK;
4317 ctx->flow.nw_tos |= pdscp->dscp;
4320 /* We may not have an ofport record for this port, but it doesn't hurt
4321 * to allow forwarding to it anyhow. Maybe such a port will appear
4322 * later and we're pre-populating the flow table. */
4325 out_port = vsp_realdev_to_vlandev(ctx->ofproto, odp_port,
4326 ctx->flow.vlan_tci);
4327 if (out_port != odp_port) {
4328 ctx->flow.vlan_tci = htons(0);
4330 commit_odp_actions(&ctx->flow, &ctx->base_flow, ctx->odp_actions);
4331 nl_msg_put_u32(ctx->odp_actions, OVS_ACTION_ATTR_OUTPUT, out_port);
4333 ctx->sflow_odp_port = odp_port;
4334 ctx->sflow_n_outputs++;
4335 ctx->nf_output_iface = ofp_port;
4336 ctx->flow.vlan_tci = flow_vlan_tci;
4337 ctx->flow.nw_tos = flow_nw_tos;
4341 compose_output_action(struct action_xlate_ctx *ctx, uint16_t ofp_port)
4343 compose_output_action__(ctx, ofp_port, true);
4347 xlate_table_action(struct action_xlate_ctx *ctx,
4348 uint16_t in_port, uint8_t table_id)
4350 if (ctx->recurse < MAX_RESUBMIT_RECURSION) {
4351 struct ofproto_dpif *ofproto = ctx->ofproto;
4352 struct rule_dpif *rule;
4353 uint16_t old_in_port;
4354 uint8_t old_table_id;
4356 old_table_id = ctx->table_id;
4357 ctx->table_id = table_id;
4359 /* Look up a flow with 'in_port' as the input port. */
4360 old_in_port = ctx->flow.in_port;
4361 ctx->flow.in_port = in_port;
4362 rule = rule_dpif_lookup(ofproto, &ctx->flow, table_id);
4365 if (table_id > 0 && table_id < N_TABLES) {
4366 struct table_dpif *table = &ofproto->tables[table_id];
4367 if (table->other_table) {
4370 : rule_calculate_tag(&ctx->flow,
4371 &table->other_table->wc,
4376 /* Restore the original input port. Otherwise OFPP_NORMAL and
4377 * OFPP_IN_PORT will have surprising behavior. */
4378 ctx->flow.in_port = old_in_port;
4380 if (ctx->resubmit_hook) {
4381 ctx->resubmit_hook(ctx, rule);
4385 ovs_be64 old_cookie = ctx->cookie;
4388 ctx->cookie = rule->up.flow_cookie;
4389 do_xlate_actions(rule->up.actions, rule->up.n_actions, ctx);
4390 ctx->cookie = old_cookie;
4394 ctx->table_id = old_table_id;
4396 static struct vlog_rate_limit recurse_rl = VLOG_RATE_LIMIT_INIT(1, 1);
4398 VLOG_ERR_RL(&recurse_rl, "resubmit actions recursed over %d times",
4399 MAX_RESUBMIT_RECURSION);
4404 xlate_resubmit_table(struct action_xlate_ctx *ctx,
4405 const struct nx_action_resubmit *nar)
4410 in_port = (nar->in_port == htons(OFPP_IN_PORT)
4412 : ntohs(nar->in_port));
4413 table_id = nar->table == 255 ? ctx->table_id : nar->table;
4415 xlate_table_action(ctx, in_port, table_id);
4419 flood_packets(struct action_xlate_ctx *ctx, bool all)
4421 struct ofport_dpif *ofport;
4423 HMAP_FOR_EACH (ofport, up.hmap_node, &ctx->ofproto->up.ports) {
4424 uint16_t ofp_port = ofport->up.ofp_port;
4426 if (ofp_port == ctx->flow.in_port) {
4431 compose_output_action__(ctx, ofp_port, false);
4432 } else if (!(ofport->up.opp.config & htonl(OFPPC_NO_FLOOD))) {
4433 compose_output_action(ctx, ofp_port);
4437 ctx->nf_output_iface = NF_OUT_FLOOD;
4441 execute_controller_action(struct action_xlate_ctx *ctx, int len,
4442 enum ofp_packet_in_reason reason)
4444 struct ofputil_packet_in pin;
4445 struct ofpbuf *packet;
4447 ctx->may_set_up_flow = false;
4452 packet = ofpbuf_clone(ctx->packet);
4454 if (packet->l2 && packet->l3) {
4455 struct eth_header *eh;
4457 eth_pop_vlan(packet);
4459 assert(eh->eth_type == ctx->flow.dl_type);
4460 memcpy(eh->eth_src, ctx->flow.dl_src, sizeof eh->eth_src);
4461 memcpy(eh->eth_dst, ctx->flow.dl_dst, sizeof eh->eth_dst);
4463 if (ctx->flow.vlan_tci & htons(VLAN_CFI)) {
4464 eth_push_vlan(packet, ctx->flow.vlan_tci);
4468 if (ctx->flow.dl_type == htons(ETH_TYPE_IP)) {
4469 packet_set_ipv4(packet, ctx->flow.nw_src, ctx->flow.nw_dst,
4470 ctx->flow.nw_tos, ctx->flow.nw_ttl);
4474 if (ctx->flow.nw_proto == IPPROTO_TCP) {
4475 packet_set_tcp_port(packet, ctx->flow.tp_src,
4477 } else if (ctx->flow.nw_proto == IPPROTO_UDP) {
4478 packet_set_udp_port(packet, ctx->flow.tp_src,
4485 pin.packet = packet->data;
4486 pin.packet_len = packet->size;
4487 pin.reason = reason;
4488 pin.table_id = ctx->table_id;
4489 pin.cookie = ctx->cookie;
4493 pin.total_len = packet->size;
4494 flow_get_metadata(&ctx->flow, &pin.fmd);
4496 connmgr_send_packet_in(ctx->ofproto->up.connmgr, &pin, &ctx->flow);
4497 ofpbuf_delete(packet);
4501 compose_dec_ttl(struct action_xlate_ctx *ctx)
4503 if (ctx->flow.dl_type != htons(ETH_TYPE_IP) &&
4504 ctx->flow.dl_type != htons(ETH_TYPE_IPV6)) {
4508 if (ctx->flow.nw_ttl > 1) {
4512 execute_controller_action(ctx, UINT16_MAX, OFPR_INVALID_TTL);
4514 /* Stop processing for current table. */
4520 xlate_output_action__(struct action_xlate_ctx *ctx,
4521 uint16_t port, uint16_t max_len)
4523 uint16_t prev_nf_output_iface = ctx->nf_output_iface;
4525 ctx->nf_output_iface = NF_OUT_DROP;
4529 compose_output_action(ctx, ctx->flow.in_port);
4532 xlate_table_action(ctx, ctx->flow.in_port, ctx->table_id);
4538 flood_packets(ctx, false);
4541 flood_packets(ctx, true);
4543 case OFPP_CONTROLLER:
4544 execute_controller_action(ctx, max_len, OFPR_ACTION);
4547 compose_output_action(ctx, OFPP_LOCAL);
4552 if (port != ctx->flow.in_port) {
4553 compose_output_action(ctx, port);
4558 if (prev_nf_output_iface == NF_OUT_FLOOD) {
4559 ctx->nf_output_iface = NF_OUT_FLOOD;
4560 } else if (ctx->nf_output_iface == NF_OUT_DROP) {
4561 ctx->nf_output_iface = prev_nf_output_iface;
4562 } else if (prev_nf_output_iface != NF_OUT_DROP &&
4563 ctx->nf_output_iface != NF_OUT_FLOOD) {
4564 ctx->nf_output_iface = NF_OUT_MULTI;
4569 xlate_output_reg_action(struct action_xlate_ctx *ctx,
4570 const struct nx_action_output_reg *naor)
4572 struct mf_subfield src;
4575 nxm_decode(&src, naor->src, naor->ofs_nbits);
4576 ofp_port = mf_get_subfield(&src, &ctx->flow);
4578 if (ofp_port <= UINT16_MAX) {
4579 xlate_output_action__(ctx, ofp_port, ntohs(naor->max_len));
4584 xlate_output_action(struct action_xlate_ctx *ctx,
4585 const struct ofp_action_output *oao)
4587 xlate_output_action__(ctx, ntohs(oao->port), ntohs(oao->max_len));
4591 xlate_enqueue_action(struct action_xlate_ctx *ctx,
4592 const struct ofp_action_enqueue *oae)
4595 uint32_t flow_priority, priority;
4598 error = dpif_queue_to_priority(ctx->ofproto->dpif, ntohl(oae->queue_id),
4601 /* Fall back to ordinary output action. */
4602 xlate_output_action__(ctx, ntohs(oae->port), 0);
4606 /* Figure out datapath output port. */
4607 ofp_port = ntohs(oae->port);
4608 if (ofp_port == OFPP_IN_PORT) {
4609 ofp_port = ctx->flow.in_port;
4610 } else if (ofp_port == ctx->flow.in_port) {
4614 /* Add datapath actions. */
4615 flow_priority = ctx->flow.skb_priority;
4616 ctx->flow.skb_priority = priority;
4617 compose_output_action(ctx, ofp_port);
4618 ctx->flow.skb_priority = flow_priority;
4620 /* Update NetFlow output port. */
4621 if (ctx->nf_output_iface == NF_OUT_DROP) {
4622 ctx->nf_output_iface = ofp_port;
4623 } else if (ctx->nf_output_iface != NF_OUT_FLOOD) {
4624 ctx->nf_output_iface = NF_OUT_MULTI;
4629 xlate_set_queue_action(struct action_xlate_ctx *ctx,
4630 const struct nx_action_set_queue *nasq)
4635 error = dpif_queue_to_priority(ctx->ofproto->dpif, ntohl(nasq->queue_id),
4638 /* Couldn't translate queue to a priority, so ignore. A warning
4639 * has already been logged. */
4643 ctx->flow.skb_priority = priority;
4646 struct xlate_reg_state {
4652 xlate_autopath(struct action_xlate_ctx *ctx,
4653 const struct nx_action_autopath *naa)
4655 uint16_t ofp_port = ntohl(naa->id);
4656 struct ofport_dpif *port = get_ofp_port(ctx->ofproto, ofp_port);
4658 if (!port || !port->bundle) {
4659 ofp_port = OFPP_NONE;
4660 } else if (port->bundle->bond) {
4661 /* Autopath does not support VLAN hashing. */
4662 struct ofport_dpif *slave = bond_choose_output_slave(
4663 port->bundle->bond, &ctx->flow, 0, &ctx->tags);
4665 ofp_port = slave->up.ofp_port;
4668 autopath_execute(naa, &ctx->flow, ofp_port);
4672 slave_enabled_cb(uint16_t ofp_port, void *ofproto_)
4674 struct ofproto_dpif *ofproto = ofproto_;
4675 struct ofport_dpif *port;
4685 case OFPP_CONTROLLER: /* Not supported by the bundle action. */
4688 port = get_ofp_port(ofproto, ofp_port);
4689 return port ? port->may_enable : false;
4694 xlate_learn_action(struct action_xlate_ctx *ctx,
4695 const struct nx_action_learn *learn)
4697 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(5, 1);
4698 struct ofputil_flow_mod fm;
4701 learn_execute(learn, &ctx->flow, &fm);
4703 error = ofproto_flow_mod(&ctx->ofproto->up, &fm);
4704 if (error && !VLOG_DROP_WARN(&rl)) {
4705 VLOG_WARN("learning action failed to modify flow table (%s)",
4706 ofperr_get_name(error));
4713 may_receive(const struct ofport_dpif *port, struct action_xlate_ctx *ctx)
4715 if (port->up.opp.config & (eth_addr_equals(ctx->flow.dl_dst, eth_addr_stp)
4716 ? htonl(OFPPC_NO_RECV_STP)
4717 : htonl(OFPPC_NO_RECV))) {
4721 /* Only drop packets here if both forwarding and learning are
4722 * disabled. If just learning is enabled, we need to have
4723 * OFPP_NORMAL and the learning action have a look at the packet
4724 * before we can drop it. */
4725 if (!stp_forward_in_state(port->stp_state)
4726 && !stp_learn_in_state(port->stp_state)) {
4734 do_xlate_actions(const union ofp_action *in, size_t n_in,
4735 struct action_xlate_ctx *ctx)
4737 const struct ofport_dpif *port;
4738 const union ofp_action *ia;
4741 port = get_ofp_port(ctx->ofproto, ctx->flow.in_port);
4742 if (port && !may_receive(port, ctx)) {
4743 /* Drop this flow. */
4747 OFPUTIL_ACTION_FOR_EACH_UNSAFE (ia, left, in, n_in) {
4748 const struct ofp_action_dl_addr *oada;
4749 const struct nx_action_resubmit *nar;
4750 const struct nx_action_set_tunnel *nast;
4751 const struct nx_action_set_queue *nasq;
4752 const struct nx_action_multipath *nam;
4753 const struct nx_action_autopath *naa;
4754 const struct nx_action_bundle *nab;
4755 const struct nx_action_output_reg *naor;
4756 enum ofputil_action_code code;
4763 code = ofputil_decode_action_unsafe(ia);
4765 case OFPUTIL_OFPAT_OUTPUT:
4766 xlate_output_action(ctx, &ia->output);
4769 case OFPUTIL_OFPAT_SET_VLAN_VID:
4770 ctx->flow.vlan_tci &= ~htons(VLAN_VID_MASK);
4771 ctx->flow.vlan_tci |= ia->vlan_vid.vlan_vid | htons(VLAN_CFI);
4774 case OFPUTIL_OFPAT_SET_VLAN_PCP:
4775 ctx->flow.vlan_tci &= ~htons(VLAN_PCP_MASK);
4776 ctx->flow.vlan_tci |= htons(
4777 (ia->vlan_pcp.vlan_pcp << VLAN_PCP_SHIFT) | VLAN_CFI);
4780 case OFPUTIL_OFPAT_STRIP_VLAN:
4781 ctx->flow.vlan_tci = htons(0);
4784 case OFPUTIL_OFPAT_SET_DL_SRC:
4785 oada = ((struct ofp_action_dl_addr *) ia);
4786 memcpy(ctx->flow.dl_src, oada->dl_addr, ETH_ADDR_LEN);
4789 case OFPUTIL_OFPAT_SET_DL_DST:
4790 oada = ((struct ofp_action_dl_addr *) ia);
4791 memcpy(ctx->flow.dl_dst, oada->dl_addr, ETH_ADDR_LEN);
4794 case OFPUTIL_OFPAT_SET_NW_SRC:
4795 ctx->flow.nw_src = ia->nw_addr.nw_addr;
4798 case OFPUTIL_OFPAT_SET_NW_DST:
4799 ctx->flow.nw_dst = ia->nw_addr.nw_addr;
4802 case OFPUTIL_OFPAT_SET_NW_TOS:
4803 /* OpenFlow 1.0 only supports IPv4. */
4804 if (ctx->flow.dl_type == htons(ETH_TYPE_IP)) {
4805 ctx->flow.nw_tos &= ~IP_DSCP_MASK;
4806 ctx->flow.nw_tos |= ia->nw_tos.nw_tos & IP_DSCP_MASK;
4810 case OFPUTIL_OFPAT_SET_TP_SRC:
4811 ctx->flow.tp_src = ia->tp_port.tp_port;
4814 case OFPUTIL_OFPAT_SET_TP_DST:
4815 ctx->flow.tp_dst = ia->tp_port.tp_port;
4818 case OFPUTIL_OFPAT_ENQUEUE:
4819 xlate_enqueue_action(ctx, (const struct ofp_action_enqueue *) ia);
4822 case OFPUTIL_NXAST_RESUBMIT:
4823 nar = (const struct nx_action_resubmit *) ia;
4824 xlate_table_action(ctx, ntohs(nar->in_port), ctx->table_id);
4827 case OFPUTIL_NXAST_RESUBMIT_TABLE:
4828 xlate_resubmit_table(ctx, (const struct nx_action_resubmit *) ia);
4831 case OFPUTIL_NXAST_SET_TUNNEL:
4832 nast = (const struct nx_action_set_tunnel *) ia;
4833 tun_id = htonll(ntohl(nast->tun_id));
4834 ctx->flow.tun_id = tun_id;
4837 case OFPUTIL_NXAST_SET_QUEUE:
4838 nasq = (const struct nx_action_set_queue *) ia;
4839 xlate_set_queue_action(ctx, nasq);
4842 case OFPUTIL_NXAST_POP_QUEUE:
4843 ctx->flow.skb_priority = ctx->orig_skb_priority;
4846 case OFPUTIL_NXAST_REG_MOVE:
4847 nxm_execute_reg_move((const struct nx_action_reg_move *) ia,
4851 case OFPUTIL_NXAST_REG_LOAD:
4852 nxm_execute_reg_load((const struct nx_action_reg_load *) ia,
4856 case OFPUTIL_NXAST_NOTE:
4857 /* Nothing to do. */
4860 case OFPUTIL_NXAST_SET_TUNNEL64:
4861 tun_id = ((const struct nx_action_set_tunnel64 *) ia)->tun_id;
4862 ctx->flow.tun_id = tun_id;
4865 case OFPUTIL_NXAST_MULTIPATH:
4866 nam = (const struct nx_action_multipath *) ia;
4867 multipath_execute(nam, &ctx->flow);
4870 case OFPUTIL_NXAST_AUTOPATH:
4871 naa = (const struct nx_action_autopath *) ia;
4872 xlate_autopath(ctx, naa);
4875 case OFPUTIL_NXAST_BUNDLE:
4876 ctx->ofproto->has_bundle_action = true;
4877 nab = (const struct nx_action_bundle *) ia;
4878 xlate_output_action__(ctx, bundle_execute(nab, &ctx->flow,
4883 case OFPUTIL_NXAST_BUNDLE_LOAD:
4884 ctx->ofproto->has_bundle_action = true;
4885 nab = (const struct nx_action_bundle *) ia;
4886 bundle_execute_load(nab, &ctx->flow, slave_enabled_cb,
4890 case OFPUTIL_NXAST_OUTPUT_REG:
4891 naor = (const struct nx_action_output_reg *) ia;
4892 xlate_output_reg_action(ctx, naor);
4895 case OFPUTIL_NXAST_LEARN:
4896 ctx->has_learn = true;
4897 if (ctx->may_learn) {
4898 xlate_learn_action(ctx, (const struct nx_action_learn *) ia);
4902 case OFPUTIL_NXAST_DEC_TTL:
4903 if (compose_dec_ttl(ctx)) {
4908 case OFPUTIL_NXAST_EXIT:
4915 /* We've let OFPP_NORMAL and the learning action look at the packet,
4916 * so drop it now if forwarding is disabled. */
4917 if (port && !stp_forward_in_state(port->stp_state)) {
4918 ofpbuf_clear(ctx->odp_actions);
4919 add_sflow_action(ctx);
4924 action_xlate_ctx_init(struct action_xlate_ctx *ctx,
4925 struct ofproto_dpif *ofproto, const struct flow *flow,
4926 ovs_be16 initial_tci, ovs_be64 cookie,
4927 const struct ofpbuf *packet)
4929 ctx->ofproto = ofproto;
4931 ctx->base_flow = ctx->flow;
4932 ctx->base_flow.tun_id = 0;
4933 ctx->base_flow.vlan_tci = initial_tci;
4934 ctx->cookie = cookie;
4935 ctx->packet = packet;
4936 ctx->may_learn = packet != NULL;
4937 ctx->resubmit_hook = NULL;
4940 static struct ofpbuf *
4941 xlate_actions(struct action_xlate_ctx *ctx,
4942 const union ofp_action *in, size_t n_in)
4944 struct flow orig_flow = ctx->flow;
4946 COVERAGE_INC(ofproto_dpif_xlate);
4948 ctx->odp_actions = ofpbuf_new(512);
4949 ofpbuf_reserve(ctx->odp_actions, NL_A_U32_SIZE);
4951 ctx->may_set_up_flow = true;
4952 ctx->has_learn = false;
4953 ctx->has_normal = false;
4954 ctx->nf_output_iface = NF_OUT_DROP;
4957 ctx->orig_skb_priority = ctx->flow.skb_priority;
4961 if (ctx->flow.nw_frag & FLOW_NW_FRAG_ANY) {
4962 switch (ctx->ofproto->up.frag_handling) {
4963 case OFPC_FRAG_NORMAL:
4964 /* We must pretend that transport ports are unavailable. */
4965 ctx->flow.tp_src = ctx->base_flow.tp_src = htons(0);
4966 ctx->flow.tp_dst = ctx->base_flow.tp_dst = htons(0);
4969 case OFPC_FRAG_DROP:
4970 return ctx->odp_actions;
4972 case OFPC_FRAG_REASM:
4975 case OFPC_FRAG_NX_MATCH:
4976 /* Nothing to do. */
4979 case OFPC_INVALID_TTL_TO_CONTROLLER:
4984 if (process_special(ctx->ofproto, &ctx->flow, ctx->packet)) {
4985 ctx->may_set_up_flow = false;
4986 return ctx->odp_actions;
4988 add_sflow_action(ctx);
4989 do_xlate_actions(in, n_in, ctx);
4991 if (!connmgr_may_set_up_flow(ctx->ofproto->up.connmgr, &ctx->flow,
4992 ctx->odp_actions->data,
4993 ctx->odp_actions->size)) {
4994 ctx->may_set_up_flow = false;
4996 && connmgr_msg_in_hook(ctx->ofproto->up.connmgr, &ctx->flow,
4998 compose_output_action(ctx, OFPP_LOCAL);
5001 add_mirror_actions(ctx, &orig_flow);
5002 fix_sflow_action(ctx);
5005 return ctx->odp_actions;
5008 /* OFPP_NORMAL implementation. */
5010 static struct ofport_dpif *ofbundle_get_a_port(const struct ofbundle *);
5012 /* Given 'vid', the VID obtained from the 802.1Q header that was received as
5013 * part of a packet (specify 0 if there was no 802.1Q header), and 'in_bundle',
5014 * the bundle on which the packet was received, returns the VLAN to which the
5017 * Both 'vid' and the return value are in the range 0...4095. */
5019 input_vid_to_vlan(const struct ofbundle *in_bundle, uint16_t vid)
5021 switch (in_bundle->vlan_mode) {
5022 case PORT_VLAN_ACCESS:
5023 return in_bundle->vlan;
5026 case PORT_VLAN_TRUNK:
5029 case PORT_VLAN_NATIVE_UNTAGGED:
5030 case PORT_VLAN_NATIVE_TAGGED:
5031 return vid ? vid : in_bundle->vlan;
5038 /* Checks whether a packet with the given 'vid' may ingress on 'in_bundle'.
5039 * If so, returns true. Otherwise, returns false and, if 'warn' is true, logs
5042 * 'vid' should be the VID obtained from the 802.1Q header that was received as
5043 * part of a packet (specify 0 if there was no 802.1Q header), in the range
5046 input_vid_is_valid(uint16_t vid, struct ofbundle *in_bundle, bool warn)
5048 /* Allow any VID on the OFPP_NONE port. */
5049 if (in_bundle == &ofpp_none_bundle) {
5053 switch (in_bundle->vlan_mode) {
5054 case PORT_VLAN_ACCESS:
5057 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
5058 VLOG_WARN_RL(&rl, "bridge %s: dropping VLAN %"PRIu16" tagged "
5059 "packet received on port %s configured as VLAN "
5060 "%"PRIu16" access port",
5061 in_bundle->ofproto->up.name, vid,
5062 in_bundle->name, in_bundle->vlan);
5068 case PORT_VLAN_NATIVE_UNTAGGED:
5069 case PORT_VLAN_NATIVE_TAGGED:
5071 /* Port must always carry its native VLAN. */
5075 case PORT_VLAN_TRUNK:
5076 if (!ofbundle_includes_vlan(in_bundle, vid)) {
5078 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
5079 VLOG_WARN_RL(&rl, "bridge %s: dropping VLAN %"PRIu16" packet "
5080 "received on port %s not configured for trunking "
5082 in_bundle->ofproto->up.name, vid,
5083 in_bundle->name, vid);
5095 /* Given 'vlan', the VLAN that a packet belongs to, and
5096 * 'out_bundle', a bundle on which the packet is to be output, returns the VID
5097 * that should be included in the 802.1Q header. (If the return value is 0,
5098 * then the 802.1Q header should only be included in the packet if there is a
5101 * Both 'vlan' and the return value are in the range 0...4095. */
5103 output_vlan_to_vid(const struct ofbundle *out_bundle, uint16_t vlan)
5105 switch (out_bundle->vlan_mode) {
5106 case PORT_VLAN_ACCESS:
5109 case PORT_VLAN_TRUNK:
5110 case PORT_VLAN_NATIVE_TAGGED:
5113 case PORT_VLAN_NATIVE_UNTAGGED:
5114 return vlan == out_bundle->vlan ? 0 : vlan;
5122 output_normal(struct action_xlate_ctx *ctx, const struct ofbundle *out_bundle,
5125 struct ofport_dpif *port;
5127 ovs_be16 tci, old_tci;
5129 vid = output_vlan_to_vid(out_bundle, vlan);
5130 if (!out_bundle->bond) {
5131 port = ofbundle_get_a_port(out_bundle);
5133 port = bond_choose_output_slave(out_bundle->bond, &ctx->flow,
5136 /* No slaves enabled, so drop packet. */
5141 old_tci = ctx->flow.vlan_tci;
5143 if (tci || out_bundle->use_priority_tags) {
5144 tci |= ctx->flow.vlan_tci & htons(VLAN_PCP_MASK);
5146 tci |= htons(VLAN_CFI);
5149 ctx->flow.vlan_tci = tci;
5151 compose_output_action(ctx, port->up.ofp_port);
5152 ctx->flow.vlan_tci = old_tci;
5156 mirror_mask_ffs(mirror_mask_t mask)
5158 BUILD_ASSERT_DECL(sizeof(unsigned int) >= sizeof(mask));
5163 ofbundle_trunks_vlan(const struct ofbundle *bundle, uint16_t vlan)
5165 return (bundle->vlan_mode != PORT_VLAN_ACCESS
5166 && (!bundle->trunks || bitmap_is_set(bundle->trunks, vlan)));
5170 ofbundle_includes_vlan(const struct ofbundle *bundle, uint16_t vlan)
5172 return vlan == bundle->vlan || ofbundle_trunks_vlan(bundle, vlan);
5175 /* Returns an arbitrary interface within 'bundle'. */
5176 static struct ofport_dpif *
5177 ofbundle_get_a_port(const struct ofbundle *bundle)
5179 return CONTAINER_OF(list_front(&bundle->ports),
5180 struct ofport_dpif, bundle_node);
5184 vlan_is_mirrored(const struct ofmirror *m, int vlan)
5186 return !m->vlans || bitmap_is_set(m->vlans, vlan);
5189 /* Returns true if a packet with Ethernet destination MAC 'dst' may be mirrored
5190 * to a VLAN. In general most packets may be mirrored but we want to drop
5191 * protocols that may confuse switches. */
5193 eth_dst_may_rspan(const uint8_t dst[ETH_ADDR_LEN])
5195 /* If you change this function's behavior, please update corresponding
5196 * documentation in vswitch.xml at the same time. */
5197 if (dst[0] != 0x01) {
5198 /* All the currently banned MACs happen to start with 01 currently, so
5199 * this is a quick way to eliminate most of the good ones. */
5201 if (eth_addr_is_reserved(dst)) {
5202 /* Drop STP, IEEE pause frames, and other reserved protocols
5203 * (01-80-c2-00-00-0x). */
5207 if (dst[0] == 0x01 && dst[1] == 0x00 && dst[2] == 0x0c) {
5209 if ((dst[3] & 0xfe) == 0xcc &&
5210 (dst[4] & 0xfe) == 0xcc &&
5211 (dst[5] & 0xfe) == 0xcc) {
5212 /* Drop the following protocols plus others following the same
5215 CDP, VTP, DTP, PAgP (01-00-0c-cc-cc-cc)
5216 Spanning Tree PVSTP+ (01-00-0c-cc-cc-cd)
5217 STP Uplink Fast (01-00-0c-cd-cd-cd) */
5221 if (!(dst[3] | dst[4] | dst[5])) {
5222 /* Drop Inter Switch Link packets (01-00-0c-00-00-00). */
5231 add_mirror_actions(struct action_xlate_ctx *ctx, const struct flow *orig_flow)
5233 struct ofproto_dpif *ofproto = ctx->ofproto;
5234 mirror_mask_t mirrors;
5235 struct ofbundle *in_bundle;
5238 const struct nlattr *a;
5241 in_bundle = lookup_input_bundle(ctx->ofproto, orig_flow->in_port,
5242 ctx->packet != NULL);
5246 mirrors = in_bundle->src_mirrors;
5248 /* Drop frames on bundles reserved for mirroring. */
5249 if (in_bundle->mirror_out) {
5250 if (ctx->packet != NULL) {
5251 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
5252 VLOG_WARN_RL(&rl, "bridge %s: dropping packet received on port "
5253 "%s, which is reserved exclusively for mirroring",
5254 ctx->ofproto->up.name, in_bundle->name);
5260 vid = vlan_tci_to_vid(orig_flow->vlan_tci);
5261 if (!input_vid_is_valid(vid, in_bundle, ctx->packet != NULL)) {
5264 vlan = input_vid_to_vlan(in_bundle, vid);
5266 /* Look at the output ports to check for destination selections. */
5268 NL_ATTR_FOR_EACH (a, left, ctx->odp_actions->data,
5269 ctx->odp_actions->size) {
5270 enum ovs_action_attr type = nl_attr_type(a);
5271 struct ofport_dpif *ofport;
5273 if (type != OVS_ACTION_ATTR_OUTPUT) {
5277 ofport = get_odp_port(ofproto, nl_attr_get_u32(a));
5278 if (ofport && ofport->bundle) {
5279 mirrors |= ofport->bundle->dst_mirrors;
5287 /* Restore the original packet before adding the mirror actions. */
5288 ctx->flow = *orig_flow;
5293 m = ofproto->mirrors[mirror_mask_ffs(mirrors) - 1];
5295 if (!vlan_is_mirrored(m, vlan)) {
5296 mirrors &= mirrors - 1;
5300 mirrors &= ~m->dup_mirrors;
5301 ctx->mirrors |= m->dup_mirrors;
5303 output_normal(ctx, m->out, vlan);
5304 } else if (eth_dst_may_rspan(orig_flow->dl_dst)
5305 && vlan != m->out_vlan) {
5306 struct ofbundle *bundle;
5308 HMAP_FOR_EACH (bundle, hmap_node, &ofproto->bundles) {
5309 if (ofbundle_includes_vlan(bundle, m->out_vlan)
5310 && !bundle->mirror_out) {
5311 output_normal(ctx, bundle, m->out_vlan);
5319 update_mirror_stats(struct ofproto_dpif *ofproto, mirror_mask_t mirrors,
5320 uint64_t packets, uint64_t bytes)
5326 for (; mirrors; mirrors &= mirrors - 1) {
5329 m = ofproto->mirrors[mirror_mask_ffs(mirrors) - 1];
5332 /* In normal circumstances 'm' will not be NULL. However,
5333 * if mirrors are reconfigured, we can temporarily get out
5334 * of sync in facet_revalidate(). We could "correct" the
5335 * mirror list before reaching here, but doing that would
5336 * not properly account the traffic stats we've currently
5337 * accumulated for previous mirror configuration. */
5341 m->packet_count += packets;
5342 m->byte_count += bytes;
5346 /* A VM broadcasts a gratuitous ARP to indicate that it has resumed after
5347 * migration. Older Citrix-patched Linux DomU used gratuitous ARP replies to
5348 * indicate this; newer upstream kernels use gratuitous ARP requests. */
5350 is_gratuitous_arp(const struct flow *flow)
5352 return (flow->dl_type == htons(ETH_TYPE_ARP)
5353 && eth_addr_is_broadcast(flow->dl_dst)
5354 && (flow->nw_proto == ARP_OP_REPLY
5355 || (flow->nw_proto == ARP_OP_REQUEST
5356 && flow->nw_src == flow->nw_dst)));
5360 update_learning_table(struct ofproto_dpif *ofproto,
5361 const struct flow *flow, int vlan,
5362 struct ofbundle *in_bundle)
5364 struct mac_entry *mac;
5366 /* Don't learn the OFPP_NONE port. */
5367 if (in_bundle == &ofpp_none_bundle) {
5371 if (!mac_learning_may_learn(ofproto->ml, flow->dl_src, vlan)) {
5375 mac = mac_learning_insert(ofproto->ml, flow->dl_src, vlan);
5376 if (is_gratuitous_arp(flow)) {
5377 /* We don't want to learn from gratuitous ARP packets that are
5378 * reflected back over bond slaves so we lock the learning table. */
5379 if (!in_bundle->bond) {
5380 mac_entry_set_grat_arp_lock(mac);
5381 } else if (mac_entry_is_grat_arp_locked(mac)) {
5386 if (mac_entry_is_new(mac) || mac->port.p != in_bundle) {
5387 /* The log messages here could actually be useful in debugging,
5388 * so keep the rate limit relatively high. */
5389 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(30, 300);
5390 VLOG_DBG_RL(&rl, "bridge %s: learned that "ETH_ADDR_FMT" is "
5391 "on port %s in VLAN %d",
5392 ofproto->up.name, ETH_ADDR_ARGS(flow->dl_src),
5393 in_bundle->name, vlan);
5395 mac->port.p = in_bundle;
5396 tag_set_add(&ofproto->revalidate_set,
5397 mac_learning_changed(ofproto->ml, mac));
5401 static struct ofbundle *
5402 lookup_input_bundle(struct ofproto_dpif *ofproto, uint16_t in_port, bool warn)
5404 struct ofport_dpif *ofport;
5406 /* Special-case OFPP_NONE, which a controller may use as the ingress
5407 * port for traffic that it is sourcing. */
5408 if (in_port == OFPP_NONE) {
5409 return &ofpp_none_bundle;
5412 /* Find the port and bundle for the received packet. */
5413 ofport = get_ofp_port(ofproto, in_port);
5414 if (ofport && ofport->bundle) {
5415 return ofport->bundle;
5418 /* Odd. A few possible reasons here:
5420 * - We deleted a port but there are still a few packets queued up
5423 * - Someone externally added a port (e.g. "ovs-dpctl add-if") that
5424 * we don't know about.
5426 * - The ofproto client didn't configure the port as part of a bundle.
5429 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
5431 VLOG_WARN_RL(&rl, "bridge %s: received packet on unknown "
5432 "port %"PRIu16, ofproto->up.name, in_port);
5437 /* Determines whether packets in 'flow' within 'ofproto' should be forwarded or
5438 * dropped. Returns true if they may be forwarded, false if they should be
5441 * 'in_port' must be the ofport_dpif that corresponds to flow->in_port.
5442 * 'in_port' must be part of a bundle (e.g. in_port->bundle must be nonnull).
5444 * 'vlan' must be the VLAN that corresponds to flow->vlan_tci on 'in_port', as
5445 * returned by input_vid_to_vlan(). It must be a valid VLAN for 'in_port', as
5446 * checked by input_vid_is_valid().
5448 * May also add tags to '*tags', although the current implementation only does
5449 * so in one special case.
5452 is_admissible(struct ofproto_dpif *ofproto, const struct flow *flow,
5453 struct ofport_dpif *in_port, uint16_t vlan, tag_type *tags)
5455 struct ofbundle *in_bundle = in_port->bundle;
5457 /* Drop frames for reserved multicast addresses
5458 * only if forward_bpdu option is absent. */
5459 if (eth_addr_is_reserved(flow->dl_dst) && !ofproto->up.forward_bpdu) {
5463 if (in_bundle->bond) {
5464 struct mac_entry *mac;
5466 switch (bond_check_admissibility(in_bundle->bond, in_port,
5467 flow->dl_dst, tags)) {
5474 case BV_DROP_IF_MOVED:
5475 mac = mac_learning_lookup(ofproto->ml, flow->dl_src, vlan, NULL);
5476 if (mac && mac->port.p != in_bundle &&
5477 (!is_gratuitous_arp(flow)
5478 || mac_entry_is_grat_arp_locked(mac))) {
5489 xlate_normal(struct action_xlate_ctx *ctx)
5491 struct ofport_dpif *in_port;
5492 struct ofbundle *in_bundle;
5493 struct mac_entry *mac;
5497 ctx->has_normal = true;
5499 in_bundle = lookup_input_bundle(ctx->ofproto, ctx->flow.in_port,
5500 ctx->packet != NULL);
5505 /* We know 'in_port' exists unless it is "ofpp_none_bundle",
5506 * since lookup_input_bundle() succeeded. */
5507 in_port = get_ofp_port(ctx->ofproto, ctx->flow.in_port);
5509 /* Drop malformed frames. */
5510 if (ctx->flow.dl_type == htons(ETH_TYPE_VLAN) &&
5511 !(ctx->flow.vlan_tci & htons(VLAN_CFI))) {
5512 if (ctx->packet != NULL) {
5513 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
5514 VLOG_WARN_RL(&rl, "bridge %s: dropping packet with partial "
5515 "VLAN tag received on port %s",
5516 ctx->ofproto->up.name, in_bundle->name);
5521 /* Drop frames on bundles reserved for mirroring. */
5522 if (in_bundle->mirror_out) {
5523 if (ctx->packet != NULL) {
5524 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
5525 VLOG_WARN_RL(&rl, "bridge %s: dropping packet received on port "
5526 "%s, which is reserved exclusively for mirroring",
5527 ctx->ofproto->up.name, in_bundle->name);
5533 vid = vlan_tci_to_vid(ctx->flow.vlan_tci);
5534 if (!input_vid_is_valid(vid, in_bundle, ctx->packet != NULL)) {
5537 vlan = input_vid_to_vlan(in_bundle, vid);
5539 /* Check other admissibility requirements. */
5541 !is_admissible(ctx->ofproto, &ctx->flow, in_port, vlan, &ctx->tags)) {
5545 /* Learn source MAC. */
5546 if (ctx->may_learn) {
5547 update_learning_table(ctx->ofproto, &ctx->flow, vlan, in_bundle);
5550 /* Determine output bundle. */
5551 mac = mac_learning_lookup(ctx->ofproto->ml, ctx->flow.dl_dst, vlan,
5554 if (mac->port.p != in_bundle) {
5555 output_normal(ctx, mac->port.p, vlan);
5558 struct ofbundle *bundle;
5560 HMAP_FOR_EACH (bundle, hmap_node, &ctx->ofproto->bundles) {
5561 if (bundle != in_bundle
5562 && ofbundle_includes_vlan(bundle, vlan)
5563 && bundle->floodable
5564 && !bundle->mirror_out) {
5565 output_normal(ctx, bundle, vlan);
5568 ctx->nf_output_iface = NF_OUT_FLOOD;
5572 /* Optimized flow revalidation.
5574 * It's a difficult problem, in general, to tell which facets need to have
5575 * their actions recalculated whenever the OpenFlow flow table changes. We
5576 * don't try to solve that general problem: for most kinds of OpenFlow flow
5577 * table changes, we recalculate the actions for every facet. This is
5578 * relatively expensive, but it's good enough if the OpenFlow flow table
5579 * doesn't change very often.
5581 * However, we can expect one particular kind of OpenFlow flow table change to
5582 * happen frequently: changes caused by MAC learning. To avoid wasting a lot
5583 * of CPU on revalidating every facet whenever MAC learning modifies the flow
5584 * table, we add a special case that applies to flow tables in which every rule
5585 * has the same form (that is, the same wildcards), except that the table is
5586 * also allowed to have a single "catch-all" flow that matches all packets. We
5587 * optimize this case by tagging all of the facets that resubmit into the table
5588 * and invalidating the same tag whenever a flow changes in that table. The
5589 * end result is that we revalidate just the facets that need it (and sometimes
5590 * a few more, but not all of the facets or even all of the facets that
5591 * resubmit to the table modified by MAC learning). */
5593 /* Calculates the tag to use for 'flow' and wildcards 'wc' when it is inserted
5594 * into an OpenFlow table with the given 'basis'. */
5596 rule_calculate_tag(const struct flow *flow, const struct flow_wildcards *wc,
5599 if (flow_wildcards_is_catchall(wc)) {
5602 struct flow tag_flow = *flow;
5603 flow_zero_wildcards(&tag_flow, wc);
5604 return tag_create_deterministic(flow_hash(&tag_flow, secret));
5608 /* Following a change to OpenFlow table 'table_id' in 'ofproto', update the
5609 * taggability of that table.
5611 * This function must be called after *each* change to a flow table. If you
5612 * skip calling it on some changes then the pointer comparisons at the end can
5613 * be invalid if you get unlucky. For example, if a flow removal causes a
5614 * cls_table to be destroyed and then a flow insertion causes a cls_table with
5615 * different wildcards to be created with the same address, then this function
5616 * will incorrectly skip revalidation. */
5618 table_update_taggable(struct ofproto_dpif *ofproto, uint8_t table_id)
5620 struct table_dpif *table = &ofproto->tables[table_id];
5621 const struct oftable *oftable = &ofproto->up.tables[table_id];
5622 struct cls_table *catchall, *other;
5623 struct cls_table *t;
5625 catchall = other = NULL;
5627 switch (hmap_count(&oftable->cls.tables)) {
5629 /* We could tag this OpenFlow table but it would make the logic a
5630 * little harder and it's a corner case that doesn't seem worth it
5636 HMAP_FOR_EACH (t, hmap_node, &oftable->cls.tables) {
5637 if (cls_table_is_catchall(t)) {
5639 } else if (!other) {
5642 /* Indicate that we can't tag this by setting both tables to
5643 * NULL. (We know that 'catchall' is already NULL.) */
5650 /* Can't tag this table. */
5654 if (table->catchall_table != catchall || table->other_table != other) {
5655 table->catchall_table = catchall;
5656 table->other_table = other;
5657 ofproto->need_revalidate = true;
5661 /* Given 'rule' that has changed in some way (either it is a rule being
5662 * inserted, a rule being deleted, or a rule whose actions are being
5663 * modified), marks facets for revalidation to ensure that packets will be
5664 * forwarded correctly according to the new state of the flow table.
5666 * This function must be called after *each* change to a flow table. See
5667 * the comment on table_update_taggable() for more information. */
5669 rule_invalidate(const struct rule_dpif *rule)
5671 struct ofproto_dpif *ofproto = ofproto_dpif_cast(rule->up.ofproto);
5673 table_update_taggable(ofproto, rule->up.table_id);
5675 if (!ofproto->need_revalidate) {
5676 struct table_dpif *table = &ofproto->tables[rule->up.table_id];
5678 if (table->other_table && rule->tag) {
5679 tag_set_add(&ofproto->revalidate_set, rule->tag);
5681 ofproto->need_revalidate = true;
5687 set_frag_handling(struct ofproto *ofproto_,
5688 enum ofp_config_flags frag_handling)
5690 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
5692 if (frag_handling != OFPC_FRAG_REASM) {
5693 ofproto->need_revalidate = true;
5701 packet_out(struct ofproto *ofproto_, struct ofpbuf *packet,
5702 const struct flow *flow,
5703 const union ofp_action *ofp_actions, size_t n_ofp_actions)
5705 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
5708 if (flow->in_port >= ofproto->max_ports && flow->in_port < OFPP_MAX) {
5709 return OFPERR_NXBRC_BAD_IN_PORT;
5712 error = validate_actions(ofp_actions, n_ofp_actions, flow,
5713 ofproto->max_ports);
5715 struct odputil_keybuf keybuf;
5716 struct ofpbuf *odp_actions;
5717 struct ofproto_push push;
5720 ofpbuf_use_stack(&key, &keybuf, sizeof keybuf);
5721 odp_flow_key_from_flow(&key, flow);
5723 action_xlate_ctx_init(&push.ctx, ofproto, flow, flow->vlan_tci, 0,
5726 /* Ensure that resubmits in 'ofp_actions' get accounted to their
5727 * matching rules. */
5729 push.bytes = packet->size;
5730 push.used = time_msec();
5731 push.ctx.resubmit_hook = push_resubmit;
5733 odp_actions = xlate_actions(&push.ctx, ofp_actions, n_ofp_actions);
5734 dpif_execute(ofproto->dpif, key.data, key.size,
5735 odp_actions->data, odp_actions->size, packet);
5736 ofpbuf_delete(odp_actions);
5744 set_netflow(struct ofproto *ofproto_,
5745 const struct netflow_options *netflow_options)
5747 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
5749 if (netflow_options) {
5750 if (!ofproto->netflow) {
5751 ofproto->netflow = netflow_create();
5753 return netflow_set_options(ofproto->netflow, netflow_options);
5755 netflow_destroy(ofproto->netflow);
5756 ofproto->netflow = NULL;
5762 get_netflow_ids(const struct ofproto *ofproto_,
5763 uint8_t *engine_type, uint8_t *engine_id)
5765 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
5767 dpif_get_netflow_ids(ofproto->dpif, engine_type, engine_id);
5771 send_active_timeout(struct ofproto_dpif *ofproto, struct facet *facet)
5773 if (!facet_is_controller_flow(facet) &&
5774 netflow_active_timeout_expired(ofproto->netflow, &facet->nf_flow)) {
5775 struct subfacet *subfacet;
5776 struct ofexpired expired;
5778 LIST_FOR_EACH (subfacet, list_node, &facet->subfacets) {
5779 if (subfacet->installed) {
5780 struct dpif_flow_stats stats;
5782 subfacet_install(subfacet, subfacet->actions,
5783 subfacet->actions_len, &stats);
5784 subfacet_update_stats(subfacet, &stats);
5788 expired.flow = facet->flow;
5789 expired.packet_count = facet->packet_count;
5790 expired.byte_count = facet->byte_count;
5791 expired.used = facet->used;
5792 netflow_expire(ofproto->netflow, &facet->nf_flow, &expired);
5797 send_netflow_active_timeouts(struct ofproto_dpif *ofproto)
5799 struct facet *facet;
5801 HMAP_FOR_EACH (facet, hmap_node, &ofproto->facets) {
5802 send_active_timeout(ofproto, facet);
5806 static struct ofproto_dpif *
5807 ofproto_dpif_lookup(const char *name)
5809 struct ofproto_dpif *ofproto;
5811 HMAP_FOR_EACH_WITH_HASH (ofproto, all_ofproto_dpifs_node,
5812 hash_string(name, 0), &all_ofproto_dpifs) {
5813 if (!strcmp(ofproto->up.name, name)) {
5821 ofproto_unixctl_fdb_flush(struct unixctl_conn *conn, int argc,
5822 const char *argv[], void *aux OVS_UNUSED)
5824 struct ofproto_dpif *ofproto;
5827 ofproto = ofproto_dpif_lookup(argv[1]);
5829 unixctl_command_reply(conn, 501, "no such bridge");
5832 mac_learning_flush(ofproto->ml, &ofproto->revalidate_set);
5834 HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
5835 mac_learning_flush(ofproto->ml, &ofproto->revalidate_set);
5839 unixctl_command_reply(conn, 200, "table successfully flushed");
5843 ofproto_unixctl_fdb_show(struct unixctl_conn *conn, int argc OVS_UNUSED,
5844 const char *argv[], void *aux OVS_UNUSED)
5846 struct ds ds = DS_EMPTY_INITIALIZER;
5847 const struct ofproto_dpif *ofproto;
5848 const struct mac_entry *e;
5850 ofproto = ofproto_dpif_lookup(argv[1]);
5852 unixctl_command_reply(conn, 501, "no such bridge");
5856 ds_put_cstr(&ds, " port VLAN MAC Age\n");
5857 LIST_FOR_EACH (e, lru_node, &ofproto->ml->lrus) {
5858 struct ofbundle *bundle = e->port.p;
5859 ds_put_format(&ds, "%5d %4d "ETH_ADDR_FMT" %3d\n",
5860 ofbundle_get_a_port(bundle)->odp_port,
5861 e->vlan, ETH_ADDR_ARGS(e->mac), mac_entry_age(e));
5863 unixctl_command_reply(conn, 200, ds_cstr(&ds));
5867 struct ofproto_trace {
5868 struct action_xlate_ctx ctx;
5874 trace_format_rule(struct ds *result, uint8_t table_id, int level,
5875 const struct rule_dpif *rule)
5877 ds_put_char_multiple(result, '\t', level);
5879 ds_put_cstr(result, "No match\n");
5883 ds_put_format(result, "Rule: table=%"PRIu8" cookie=%#"PRIx64" ",
5884 table_id, ntohll(rule->up.flow_cookie));
5885 cls_rule_format(&rule->up.cr, result);
5886 ds_put_char(result, '\n');
5888 ds_put_char_multiple(result, '\t', level);
5889 ds_put_cstr(result, "OpenFlow ");
5890 ofp_print_actions(result, rule->up.actions, rule->up.n_actions);
5891 ds_put_char(result, '\n');
5895 trace_format_flow(struct ds *result, int level, const char *title,
5896 struct ofproto_trace *trace)
5898 ds_put_char_multiple(result, '\t', level);
5899 ds_put_format(result, "%s: ", title);
5900 if (flow_equal(&trace->ctx.flow, &trace->flow)) {
5901 ds_put_cstr(result, "unchanged");
5903 flow_format(result, &trace->ctx.flow);
5904 trace->flow = trace->ctx.flow;
5906 ds_put_char(result, '\n');
5910 trace_format_regs(struct ds *result, int level, const char *title,
5911 struct ofproto_trace *trace)
5915 ds_put_char_multiple(result, '\t', level);
5916 ds_put_format(result, "%s:", title);
5917 for (i = 0; i < FLOW_N_REGS; i++) {
5918 ds_put_format(result, " reg%zu=0x%"PRIx32, i, trace->flow.regs[i]);
5920 ds_put_char(result, '\n');
5924 trace_resubmit(struct action_xlate_ctx *ctx, struct rule_dpif *rule)
5926 struct ofproto_trace *trace = CONTAINER_OF(ctx, struct ofproto_trace, ctx);
5927 struct ds *result = trace->result;
5929 ds_put_char(result, '\n');
5930 trace_format_flow(result, ctx->recurse + 1, "Resubmitted flow", trace);
5931 trace_format_regs(result, ctx->recurse + 1, "Resubmitted regs", trace);
5932 trace_format_rule(result, ctx->table_id, ctx->recurse + 1, rule);
5936 ofproto_unixctl_trace(struct unixctl_conn *conn, int argc, const char *argv[],
5937 void *aux OVS_UNUSED)
5939 const char *dpname = argv[1];
5940 struct ofproto_dpif *ofproto;
5941 struct ofpbuf odp_key;
5942 struct ofpbuf *packet;
5943 struct rule_dpif *rule;
5944 ovs_be16 initial_tci;
5950 ofpbuf_init(&odp_key, 0);
5953 ofproto = ofproto_dpif_lookup(dpname);
5955 unixctl_command_reply(conn, 501, "Unknown ofproto (use ofproto/list "
5959 if (argc == 3 || (argc == 4 && !strcmp(argv[3], "-generate"))) {
5960 /* ofproto/trace dpname flow [-generate] */
5961 const char *flow_s = argv[2];
5962 const char *generate_s = argv[3];
5965 /* Convert string to datapath key. */
5966 ofpbuf_init(&odp_key, 0);
5967 error = odp_flow_key_from_string(flow_s, NULL, &odp_key);
5969 unixctl_command_reply(conn, 501, "Bad flow syntax");
5973 /* Convert odp_key to flow. */
5974 error = ofproto_dpif_extract_flow_key(ofproto, odp_key.data,
5975 odp_key.size, &flow,
5976 &initial_tci, NULL);
5977 if (error == ODP_FIT_ERROR) {
5978 unixctl_command_reply(conn, 501, "Invalid flow");
5982 /* Generate a packet, if requested. */
5984 packet = ofpbuf_new(0);
5985 flow_compose(packet, &flow);
5987 } else if (argc == 6) {
5988 /* ofproto/trace dpname priority tun_id in_port packet */
5989 const char *priority_s = argv[2];
5990 const char *tun_id_s = argv[3];
5991 const char *in_port_s = argv[4];
5992 const char *packet_s = argv[5];
5993 uint16_t in_port = ofp_port_to_odp_port(atoi(in_port_s));
5994 ovs_be64 tun_id = htonll(strtoull(tun_id_s, NULL, 0));
5995 uint32_t priority = atoi(priority_s);
5998 msg = eth_from_hex(packet_s, &packet);
6000 unixctl_command_reply(conn, 501, msg);
6004 ds_put_cstr(&result, "Packet: ");
6005 s = ofp_packet_to_string(packet->data, packet->size);
6006 ds_put_cstr(&result, s);
6009 flow_extract(packet, priority, tun_id, in_port, &flow);
6010 initial_tci = flow.vlan_tci;
6012 unixctl_command_reply(conn, 501, "Bad command syntax");
6016 ds_put_cstr(&result, "Flow: ");
6017 flow_format(&result, &flow);
6018 ds_put_char(&result, '\n');
6020 rule = rule_dpif_lookup(ofproto, &flow, 0);
6021 trace_format_rule(&result, 0, 0, rule);
6023 struct ofproto_trace trace;
6024 struct ofpbuf *odp_actions;
6026 trace.result = &result;
6028 action_xlate_ctx_init(&trace.ctx, ofproto, &flow, initial_tci,
6029 rule->up.flow_cookie, packet);
6030 trace.ctx.resubmit_hook = trace_resubmit;
6031 odp_actions = xlate_actions(&trace.ctx,
6032 rule->up.actions, rule->up.n_actions);
6034 ds_put_char(&result, '\n');
6035 trace_format_flow(&result, 0, "Final flow", &trace);
6036 ds_put_cstr(&result, "Datapath actions: ");
6037 format_odp_actions(&result, odp_actions->data, odp_actions->size);
6038 ofpbuf_delete(odp_actions);
6040 if (!trace.ctx.may_set_up_flow) {
6042 ds_put_cstr(&result, "\nThis flow is not cachable.");
6044 ds_put_cstr(&result, "\nThe datapath actions are incomplete--"
6045 "for complete actions, please supply a packet.");
6050 unixctl_command_reply(conn, 200, ds_cstr(&result));
6053 ds_destroy(&result);
6054 ofpbuf_delete(packet);
6055 ofpbuf_uninit(&odp_key);
6059 ofproto_dpif_clog(struct unixctl_conn *conn OVS_UNUSED, int argc OVS_UNUSED,
6060 const char *argv[] OVS_UNUSED, void *aux OVS_UNUSED)
6063 unixctl_command_reply(conn, 200, NULL);
6067 ofproto_dpif_unclog(struct unixctl_conn *conn OVS_UNUSED, int argc OVS_UNUSED,
6068 const char *argv[] OVS_UNUSED, void *aux OVS_UNUSED)
6071 unixctl_command_reply(conn, 200, NULL);
6074 /* Runs a self-check of flow translations in 'ofproto'. Appends a message to
6075 * 'reply' describing the results. */
6077 ofproto_dpif_self_check__(struct ofproto_dpif *ofproto, struct ds *reply)
6079 struct facet *facet;
6083 HMAP_FOR_EACH (facet, hmap_node, &ofproto->facets) {
6084 if (!facet_check_consistency(facet)) {
6089 ofproto->need_revalidate = true;
6093 ds_put_format(reply, "%s: self-check failed (%d errors)\n",
6094 ofproto->up.name, errors);
6096 ds_put_format(reply, "%s: self-check passed\n", ofproto->up.name);
6101 ofproto_dpif_self_check(struct unixctl_conn *conn,
6102 int argc, const char *argv[], void *aux OVS_UNUSED)
6104 struct ds reply = DS_EMPTY_INITIALIZER;
6105 struct ofproto_dpif *ofproto;
6108 ofproto = ofproto_dpif_lookup(argv[1]);
6110 unixctl_command_reply(conn, 501, "Unknown ofproto (use "
6111 "ofproto/list for help)");
6114 ofproto_dpif_self_check__(ofproto, &reply);
6116 HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
6117 ofproto_dpif_self_check__(ofproto, &reply);
6121 unixctl_command_reply(conn, 200, ds_cstr(&reply));
6126 ofproto_dpif_unixctl_init(void)
6128 static bool registered;
6134 unixctl_command_register(
6136 "bridge {tun_id in_port packet | odp_flow [-generate]}",
6137 2, 5, ofproto_unixctl_trace, NULL);
6138 unixctl_command_register("fdb/flush", "[bridge]", 0, 1,
6139 ofproto_unixctl_fdb_flush, NULL);
6140 unixctl_command_register("fdb/show", "bridge", 1, 1,
6141 ofproto_unixctl_fdb_show, NULL);
6142 unixctl_command_register("ofproto/clog", "", 0, 0,
6143 ofproto_dpif_clog, NULL);
6144 unixctl_command_register("ofproto/unclog", "", 0, 0,
6145 ofproto_dpif_unclog, NULL);
6146 unixctl_command_register("ofproto/self-check", "[bridge]", 0, 1,
6147 ofproto_dpif_self_check, NULL);
6150 /* Linux VLAN device support (e.g. "eth0.10" for VLAN 10.)
6152 * This is deprecated. It is only for compatibility with broken device drivers
6153 * in old versions of Linux that do not properly support VLANs when VLAN
6154 * devices are not used. When broken device drivers are no longer in
6155 * widespread use, we will delete these interfaces. */
6158 set_realdev(struct ofport *ofport_, uint16_t realdev_ofp_port, int vid)
6160 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport_->ofproto);
6161 struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
6163 if (realdev_ofp_port == ofport->realdev_ofp_port
6164 && vid == ofport->vlandev_vid) {
6168 ofproto->need_revalidate = true;
6170 if (ofport->realdev_ofp_port) {
6173 if (realdev_ofp_port && ofport->bundle) {
6174 /* vlandevs are enslaved to their realdevs, so they are not allowed to
6175 * themselves be part of a bundle. */
6176 bundle_set(ofport->up.ofproto, ofport->bundle, NULL);
6179 ofport->realdev_ofp_port = realdev_ofp_port;
6180 ofport->vlandev_vid = vid;
6182 if (realdev_ofp_port) {
6183 vsp_add(ofport, realdev_ofp_port, vid);
6190 hash_realdev_vid(uint16_t realdev_ofp_port, int vid)
6192 return hash_2words(realdev_ofp_port, vid);
6196 vsp_realdev_to_vlandev(const struct ofproto_dpif *ofproto,
6197 uint32_t realdev_odp_port, ovs_be16 vlan_tci)
6199 if (!hmap_is_empty(&ofproto->realdev_vid_map)) {
6200 uint16_t realdev_ofp_port = odp_port_to_ofp_port(realdev_odp_port);
6201 int vid = vlan_tci_to_vid(vlan_tci);
6202 const struct vlan_splinter *vsp;
6204 HMAP_FOR_EACH_WITH_HASH (vsp, realdev_vid_node,
6205 hash_realdev_vid(realdev_ofp_port, vid),
6206 &ofproto->realdev_vid_map) {
6207 if (vsp->realdev_ofp_port == realdev_ofp_port
6208 && vsp->vid == vid) {
6209 return ofp_port_to_odp_port(vsp->vlandev_ofp_port);
6213 return realdev_odp_port;
6216 static struct vlan_splinter *
6217 vlandev_find(const struct ofproto_dpif *ofproto, uint16_t vlandev_ofp_port)
6219 struct vlan_splinter *vsp;
6221 HMAP_FOR_EACH_WITH_HASH (vsp, vlandev_node, hash_int(vlandev_ofp_port, 0),
6222 &ofproto->vlandev_map) {
6223 if (vsp->vlandev_ofp_port == vlandev_ofp_port) {
6232 vsp_vlandev_to_realdev(const struct ofproto_dpif *ofproto,
6233 uint16_t vlandev_ofp_port, int *vid)
6235 if (!hmap_is_empty(&ofproto->vlandev_map)) {
6236 const struct vlan_splinter *vsp;
6238 vsp = vlandev_find(ofproto, vlandev_ofp_port);
6243 return vsp->realdev_ofp_port;
6250 vsp_remove(struct ofport_dpif *port)
6252 struct ofproto_dpif *ofproto = ofproto_dpif_cast(port->up.ofproto);
6253 struct vlan_splinter *vsp;
6255 vsp = vlandev_find(ofproto, port->up.ofp_port);
6257 hmap_remove(&ofproto->vlandev_map, &vsp->vlandev_node);
6258 hmap_remove(&ofproto->realdev_vid_map, &vsp->realdev_vid_node);
6261 port->realdev_ofp_port = 0;
6263 VLOG_ERR("missing vlan device record");
6268 vsp_add(struct ofport_dpif *port, uint16_t realdev_ofp_port, int vid)
6270 struct ofproto_dpif *ofproto = ofproto_dpif_cast(port->up.ofproto);
6272 if (!vsp_vlandev_to_realdev(ofproto, port->up.ofp_port, NULL)
6273 && (vsp_realdev_to_vlandev(ofproto, realdev_ofp_port, htons(vid))
6274 == realdev_ofp_port)) {
6275 struct vlan_splinter *vsp;
6277 vsp = xmalloc(sizeof *vsp);
6278 hmap_insert(&ofproto->vlandev_map, &vsp->vlandev_node,
6279 hash_int(port->up.ofp_port, 0));
6280 hmap_insert(&ofproto->realdev_vid_map, &vsp->realdev_vid_node,
6281 hash_realdev_vid(realdev_ofp_port, vid));
6282 vsp->realdev_ofp_port = realdev_ofp_port;
6283 vsp->vlandev_ofp_port = port->up.ofp_port;
6286 port->realdev_ofp_port = realdev_ofp_port;
6288 VLOG_ERR("duplicate vlan device record");
6292 const struct ofproto_class ofproto_dpif_class = {
6321 port_is_lacp_current,
6322 NULL, /* rule_choose_table */
6329 rule_modify_actions,
6337 get_cfm_remote_mpids,
6341 get_stp_port_status,
6348 is_mirror_output_bundle,
6349 forward_bpdu_changed,